Doug / smarc-fsl-linux-kernel

Download as
Browse Code ยป

Commit ee79d1bdb6a10499e53f80b1e8d14110215178ba

Authored by Heiko Carstens
Committed by Ingo Molnar
1 parent c1dfdc7597
Exists in master and in 7 other branches imx_3.0.35_4.1.0, imx_3.10.17_1.0.1_ga, imx_3.10.53_1.1.0_ga, imx_3.14.28_1.0.0_ga, smarc-imx_3.10.53_1.1.0_ga, smarc-imx_3.14.28_1.0.0_ga, smarc-l5.0.0_1.0.0-ga

sched: let arch_update_cpu_topology indicate if topology changed 

Change arch_update_cpu_topology so it returns 1 if the cpu topology changed
and 0 if it didn't change. This will be useful for the next patch which adds
a call to this function in partition_sched_domains.

Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

Showing 3 changed files with 11 additions and 4 deletions Inline Diff

arch/s390/kernel/topology.c
Diff comments   View file @ ee79d1b
1 /* 1 /*
2 * Copyright IBM Corp. 2007 2 * Copyright IBM Corp. 2007
3 * Author(s): Heiko Carstens <heiko.carstens@de.ibm.com> 3 * Author(s): Heiko Carstens <heiko.carstens@de.ibm.com>
4 */ 4 */
5 5
6 #include <linux/kernel.h> 6 #include <linux/kernel.h>
7 #include <linux/mm.h> 7 #include <linux/mm.h>
8 #include <linux/init.h> 8 #include <linux/init.h>
9 #include <linux/device.h> 9 #include <linux/device.h>
10 #include <linux/bootmem.h> 10 #include <linux/bootmem.h>
11 #include <linux/sched.h> 11 #include <linux/sched.h>
12 #include <linux/workqueue.h> 12 #include <linux/workqueue.h>
13 #include <linux/cpu.h> 13 #include <linux/cpu.h>
14 #include <linux/smp.h> 14 #include <linux/smp.h>
15 #include <asm/delay.h> 15 #include <asm/delay.h>
16 #include <asm/s390_ext.h> 16 #include <asm/s390_ext.h>
17 #include <asm/sysinfo.h> 17 #include <asm/sysinfo.h>
18 18
19 #define CPU_BITS 64 19 #define CPU_BITS 64
20 #define NR_MAG 6 20 #define NR_MAG 6
21 21
22 #define PTF_HORIZONTAL (0UL) 22 #define PTF_HORIZONTAL (0UL)
23 #define PTF_VERTICAL (1UL) 23 #define PTF_VERTICAL (1UL)
24 #define PTF_CHECK (2UL) 24 #define PTF_CHECK (2UL)
25 25
26 struct tl_cpu { 26 struct tl_cpu {
27 unsigned char reserved0[4]; 27 unsigned char reserved0[4];
28 unsigned char :6; 28 unsigned char :6;
29 unsigned char pp:2; 29 unsigned char pp:2;
30 unsigned char reserved1; 30 unsigned char reserved1;
31 unsigned short origin; 31 unsigned short origin;
32 unsigned long mask[CPU_BITS / BITS_PER_LONG]; 32 unsigned long mask[CPU_BITS / BITS_PER_LONG];
33 }; 33 };
34 34
35 struct tl_container { 35 struct tl_container {
36 unsigned char reserved[8]; 36 unsigned char reserved[8];
37 }; 37 };
38 38
39 union tl_entry { 39 union tl_entry {
40 unsigned char nl; 40 unsigned char nl;
41 struct tl_cpu cpu; 41 struct tl_cpu cpu;
42 struct tl_container container; 42 struct tl_container container;
43 }; 43 };
44 44
45 struct tl_info { 45 struct tl_info {
46 unsigned char reserved0[2]; 46 unsigned char reserved0[2];
47 unsigned short length; 47 unsigned short length;
48 unsigned char mag[NR_MAG]; 48 unsigned char mag[NR_MAG];
49 unsigned char reserved1; 49 unsigned char reserved1;
50 unsigned char mnest; 50 unsigned char mnest;
51 unsigned char reserved2[4]; 51 unsigned char reserved2[4];
52 union tl_entry tle[0]; 52 union tl_entry tle[0];
53 }; 53 };
54 54
55 struct core_info { 55 struct core_info {
56 struct core_info *next; 56 struct core_info *next;
57 cpumask_t mask; 57 cpumask_t mask;
58 }; 58 };
59 59
60 static void topology_work_fn(struct work_struct *work); 60 static void topology_work_fn(struct work_struct *work);
61 static struct tl_info *tl_info; 61 static struct tl_info *tl_info;
62 static struct core_info core_info; 62 static struct core_info core_info;
63 static int machine_has_topology; 63 static int machine_has_topology;
64 static int machine_has_topology_irq; 64 static int machine_has_topology_irq;
65 static struct timer_list topology_timer; 65 static struct timer_list topology_timer;
66 static void set_topology_timer(void); 66 static void set_topology_timer(void);
67 static DECLARE_WORK(topology_work, topology_work_fn); 67 static DECLARE_WORK(topology_work, topology_work_fn);
68 /* topology_lock protects the core linked list */ 68 /* topology_lock protects the core linked list */
69 static DEFINE_SPINLOCK(topology_lock); 69 static DEFINE_SPINLOCK(topology_lock);
70 70
71 cpumask_t cpu_core_map[NR_CPUS]; 71 cpumask_t cpu_core_map[NR_CPUS];
72 72
73 cpumask_t cpu_coregroup_map(unsigned int cpu) 73 cpumask_t cpu_coregroup_map(unsigned int cpu)
74 { 74 {
75 struct core_info *core = &core_info; 75 struct core_info *core = &core_info;
76 unsigned long flags; 76 unsigned long flags;
77 cpumask_t mask; 77 cpumask_t mask;
78 78
79 cpus_clear(mask); 79 cpus_clear(mask);
80 if (!machine_has_topology) 80 if (!machine_has_topology)
81 return cpu_present_map; 81 return cpu_present_map;
82 spin_lock_irqsave(&topology_lock, flags); 82 spin_lock_irqsave(&topology_lock, flags);
83 while (core) { 83 while (core) {
84 if (cpu_isset(cpu, core->mask)) { 84 if (cpu_isset(cpu, core->mask)) {
85 mask = core->mask; 85 mask = core->mask;
86 break; 86 break;
87 } 87 }
88 core = core->next; 88 core = core->next;
89 } 89 }
90 spin_unlock_irqrestore(&topology_lock, flags); 90 spin_unlock_irqrestore(&topology_lock, flags);
91 if (cpus_empty(mask)) 91 if (cpus_empty(mask))
92 mask = cpumask_of_cpu(cpu); 92 mask = cpumask_of_cpu(cpu);
93 return mask; 93 return mask;
94 } 94 }
95 95
96 static void add_cpus_to_core(struct tl_cpu *tl_cpu, struct core_info *core) 96 static void add_cpus_to_core(struct tl_cpu *tl_cpu, struct core_info *core)
97 { 97 {
98 unsigned int cpu; 98 unsigned int cpu;
99 99
100 for (cpu = find_first_bit(&tl_cpu->mask[0], CPU_BITS); 100 for (cpu = find_first_bit(&tl_cpu->mask[0], CPU_BITS);
101 cpu < CPU_BITS; 101 cpu < CPU_BITS;
102 cpu = find_next_bit(&tl_cpu->mask[0], CPU_BITS, cpu + 1)) 102 cpu = find_next_bit(&tl_cpu->mask[0], CPU_BITS, cpu + 1))
103 { 103 {
104 unsigned int rcpu, lcpu; 104 unsigned int rcpu, lcpu;
105 105
106 rcpu = CPU_BITS - 1 - cpu + tl_cpu->origin; 106 rcpu = CPU_BITS - 1 - cpu + tl_cpu->origin;
107 for_each_present_cpu(lcpu) { 107 for_each_present_cpu(lcpu) {
108 if (__cpu_logical_map[lcpu] == rcpu) { 108 if (__cpu_logical_map[lcpu] == rcpu) {
109 cpu_set(lcpu, core->mask); 109 cpu_set(lcpu, core->mask);
110 smp_cpu_polarization[lcpu] = tl_cpu->pp; 110 smp_cpu_polarization[lcpu] = tl_cpu->pp;
111 } 111 }
112 } 112 }
113 } 113 }
114 } 114 }
115 115
116 static void clear_cores(void) 116 static void clear_cores(void)
117 { 117 {
118 struct core_info *core = &core_info; 118 struct core_info *core = &core_info;
119 119
120 while (core) { 120 while (core) {
121 cpus_clear(core->mask); 121 cpus_clear(core->mask);
122 core = core->next; 122 core = core->next;
123 } 123 }
124 } 124 }
125 125
126 static union tl_entry *next_tle(union tl_entry *tle) 126 static union tl_entry *next_tle(union tl_entry *tle)
127 { 127 {
128 if (tle->nl) 128 if (tle->nl)
129 return (union tl_entry *)((struct tl_container *)tle + 1); 129 return (union tl_entry *)((struct tl_container *)tle + 1);
130 else 130 else
131 return (union tl_entry *)((struct tl_cpu *)tle + 1); 131 return (union tl_entry *)((struct tl_cpu *)tle + 1);
132 } 132 }
133 133
134 static void tl_to_cores(struct tl_info *info) 134 static void tl_to_cores(struct tl_info *info)
135 { 135 {
136 union tl_entry *tle, *end; 136 union tl_entry *tle, *end;
137 struct core_info *core = &core_info; 137 struct core_info *core = &core_info;
138 138
139 spin_lock_irq(&topology_lock); 139 spin_lock_irq(&topology_lock);
140 clear_cores(); 140 clear_cores();
141 tle = info->tle; 141 tle = info->tle;
142 end = (union tl_entry *)((unsigned long)info + info->length); 142 end = (union tl_entry *)((unsigned long)info + info->length);
143 while (tle < end) { 143 while (tle < end) {
144 switch (tle->nl) { 144 switch (tle->nl) {
145 case 5: 145 case 5:
146 case 4: 146 case 4:
147 case 3: 147 case 3:
148 case 2: 148 case 2:
149 break; 149 break;
150 case 1: 150 case 1:
151 core = core->next; 151 core = core->next;
152 break; 152 break;
153 case 0: 153 case 0:
154 add_cpus_to_core(&tle->cpu, core); 154 add_cpus_to_core(&tle->cpu, core);
155 break; 155 break;
156 default: 156 default:
157 clear_cores(); 157 clear_cores();
158 machine_has_topology = 0; 158 machine_has_topology = 0;
159 return; 159 return;
160 } 160 }
161 tle = next_tle(tle); 161 tle = next_tle(tle);
162 } 162 }
163 spin_unlock_irq(&topology_lock); 163 spin_unlock_irq(&topology_lock);
164 } 164 }
165 165
166 static void topology_update_polarization_simple(void) 166 static void topology_update_polarization_simple(void)
167 { 167 {
168 int cpu; 168 int cpu;
169 169
170 mutex_lock(&smp_cpu_state_mutex); 170 mutex_lock(&smp_cpu_state_mutex);
171 for_each_present_cpu(cpu) 171 for_each_present_cpu(cpu)
172 smp_cpu_polarization[cpu] = POLARIZATION_HRZ; 172 smp_cpu_polarization[cpu] = POLARIZATION_HRZ;
173 mutex_unlock(&smp_cpu_state_mutex); 173 mutex_unlock(&smp_cpu_state_mutex);
174 } 174 }
175 175
176 static int ptf(unsigned long fc) 176 static int ptf(unsigned long fc)
177 { 177 {
178 int rc; 178 int rc;
179 179
180 asm volatile( 180 asm volatile(
181 " .insn rre,0xb9a20000,%1,%1\n" 181 " .insn rre,0xb9a20000,%1,%1\n"
182 " ipm %0\n" 182 " ipm %0\n"
183 " srl %0,28\n" 183 " srl %0,28\n"
184 : "=d" (rc) 184 : "=d" (rc)
185 : "d" (fc) : "cc"); 185 : "d" (fc) : "cc");
186 return rc; 186 return rc;
187 } 187 }
188 188
189 int topology_set_cpu_management(int fc) 189 int topology_set_cpu_management(int fc)
190 { 190 {
191 int cpu; 191 int cpu;
192 int rc; 192 int rc;
193 193
194 if (!machine_has_topology) 194 if (!machine_has_topology)
195 return -EOPNOTSUPP; 195 return -EOPNOTSUPP;
196 if (fc) 196 if (fc)
197 rc = ptf(PTF_VERTICAL); 197 rc = ptf(PTF_VERTICAL);
198 else 198 else
199 rc = ptf(PTF_HORIZONTAL); 199 rc = ptf(PTF_HORIZONTAL);
200 if (rc) 200 if (rc)
201 return -EBUSY; 201 return -EBUSY;
202 for_each_present_cpu(cpu) 202 for_each_present_cpu(cpu)
203 smp_cpu_polarization[cpu] = POLARIZATION_UNKNWN; 203 smp_cpu_polarization[cpu] = POLARIZATION_UNKNWN;
204 return rc; 204 return rc;
205 } 205 }
206 206
207 static void update_cpu_core_map(void) 207 static void update_cpu_core_map(void)
208 { 208 {
209 int cpu; 209 int cpu;
210 210
211 for_each_present_cpu(cpu) 211 for_each_present_cpu(cpu)
212 cpu_core_map[cpu] = cpu_coregroup_map(cpu); 212 cpu_core_map[cpu] = cpu_coregroup_map(cpu);
213 } 213 }
214 214
215 void arch_update_cpu_topology(void) 215 int arch_update_cpu_topology(void)
216 { 216 {
217 struct tl_info *info = tl_info; 217 struct tl_info *info = tl_info;
218 struct sys_device *sysdev; 218 struct sys_device *sysdev;
219 int cpu; 219 int cpu;
220 220
221 if (!machine_has_topology) { 221 if (!machine_has_topology) {
222 update_cpu_core_map(); 222 update_cpu_core_map();
223 topology_update_polarization_simple(); 223 topology_update_polarization_simple();
224 return; 224 return 0;
225 } 225 }
226 stsi(info, 15, 1, 2); 226 stsi(info, 15, 1, 2);
227 tl_to_cores(info); 227 tl_to_cores(info);
228 update_cpu_core_map(); 228 update_cpu_core_map();
229 for_each_online_cpu(cpu) { 229 for_each_online_cpu(cpu) {
230 sysdev = get_cpu_sysdev(cpu); 230 sysdev = get_cpu_sysdev(cpu);
231 kobject_uevent(&sysdev->kobj, KOBJ_CHANGE); 231 kobject_uevent(&sysdev->kobj, KOBJ_CHANGE);
232 } 232 }
233 return 1;
233 } 234 }
234 235
235 static void topology_work_fn(struct work_struct *work) 236 static void topology_work_fn(struct work_struct *work)
236 { 237 {
237 arch_reinit_sched_domains(); 238 arch_reinit_sched_domains();
238 } 239 }
239 240
240 void topology_schedule_update(void) 241 void topology_schedule_update(void)
241 { 242 {
242 schedule_work(&topology_work); 243 schedule_work(&topology_work);
243 } 244 }
244 245
245 static void topology_timer_fn(unsigned long ignored) 246 static void topology_timer_fn(unsigned long ignored)
246 { 247 {
247 if (ptf(PTF_CHECK)) 248 if (ptf(PTF_CHECK))
248 topology_schedule_update(); 249 topology_schedule_update();
249 set_topology_timer(); 250 set_topology_timer();
250 } 251 }
251 252
252 static void set_topology_timer(void) 253 static void set_topology_timer(void)
253 { 254 {
254 topology_timer.function = topology_timer_fn; 255 topology_timer.function = topology_timer_fn;
255 topology_timer.data = 0; 256 topology_timer.data = 0;
256 topology_timer.expires = jiffies + 60 * HZ; 257 topology_timer.expires = jiffies + 60 * HZ;
257 add_timer(&topology_timer); 258 add_timer(&topology_timer);
258 } 259 }
259 260
260 static void topology_interrupt(__u16 code) 261 static void topology_interrupt(__u16 code)
261 { 262 {
262 schedule_work(&topology_work); 263 schedule_work(&topology_work);
263 } 264 }
264 265
265 static int __init init_topology_update(void) 266 static int __init init_topology_update(void)
266 { 267 {
267 int rc; 268 int rc;
268 269
269 rc = 0; 270 rc = 0;
270 if (!machine_has_topology) { 271 if (!machine_has_topology) {
271 topology_update_polarization_simple(); 272 topology_update_polarization_simple();
272 goto out; 273 goto out;
273 } 274 }
274 init_timer_deferrable(&topology_timer); 275 init_timer_deferrable(&topology_timer);
275 if (machine_has_topology_irq) { 276 if (machine_has_topology_irq) {
276 rc = register_external_interrupt(0x2005, topology_interrupt); 277 rc = register_external_interrupt(0x2005, topology_interrupt);
277 if (rc) 278 if (rc)
278 goto out; 279 goto out;
279 ctl_set_bit(0, 8); 280 ctl_set_bit(0, 8);
280 } 281 }
281 else 282 else
282 set_topology_timer(); 283 set_topology_timer();
283 out: 284 out:
284 update_cpu_core_map(); 285 update_cpu_core_map();
285 return rc; 286 return rc;
286 } 287 }
287 __initcall(init_topology_update); 288 __initcall(init_topology_update);
288 289
289 void __init s390_init_cpu_topology(void) 290 void __init s390_init_cpu_topology(void)
290 { 291 {
291 unsigned long long facility_bits; 292 unsigned long long facility_bits;
292 struct tl_info *info; 293 struct tl_info *info;
293 struct core_info *core; 294 struct core_info *core;
294 int nr_cores; 295 int nr_cores;
295 int i; 296 int i;
296 297
297 if (stfle(&facility_bits, 1) <= 0) 298 if (stfle(&facility_bits, 1) <= 0)
298 return; 299 return;
299 if (!(facility_bits & (1ULL << 52)) || !(facility_bits & (1ULL << 61))) 300 if (!(facility_bits & (1ULL << 52)) || !(facility_bits & (1ULL << 61)))
300 return; 301 return;
301 machine_has_topology = 1; 302 machine_has_topology = 1;
302 303
303 if (facility_bits & (1ULL << 51)) 304 if (facility_bits & (1ULL << 51))
304 machine_has_topology_irq = 1; 305 machine_has_topology_irq = 1;
305 306
306 tl_info = alloc_bootmem_pages(PAGE_SIZE); 307 tl_info = alloc_bootmem_pages(PAGE_SIZE);
307 info = tl_info; 308 info = tl_info;
308 stsi(info, 15, 1, 2); 309 stsi(info, 15, 1, 2);
309 310
310 nr_cores = info->mag[NR_MAG - 2]; 311 nr_cores = info->mag[NR_MAG - 2];
311 for (i = 0; i < info->mnest - 2; i++) 312 for (i = 0; i < info->mnest - 2; i++)
312 nr_cores *= info->mag[NR_MAG - 3 - i]; 313 nr_cores *= info->mag[NR_MAG - 3 - i];
313 314
314 printk(KERN_INFO "CPU topology:"); 315 printk(KERN_INFO "CPU topology:");
315 for (i = 0; i < NR_MAG; i++) 316 for (i = 0; i < NR_MAG; i++)
316 printk(" %d", info->mag[i]); 317 printk(" %d", info->mag[i]);
317 printk(" / %d\n", info->mnest); 318 printk(" / %d\n", info->mnest);
318 319
319 core = &core_info; 320 core = &core_info;
320 for (i = 0; i < nr_cores; i++) { 321 for (i = 0; i < nr_cores; i++) {
321 core->next = alloc_bootmem(sizeof(struct core_info)); 322 core->next = alloc_bootmem(sizeof(struct core_info));
322 core = core->next; 323 core = core->next;
323 if (!core) 324 if (!core)
324 goto error; 325 goto error;
325 } 326 }
326 return; 327 return;
327 error: 328 error:
328 machine_has_topology = 0; 329 machine_has_topology = 0;
329 machine_has_topology_irq = 0; 330 machine_has_topology_irq = 0;
330 } 331 }
331 332
include/linux/topology.h
Diff comments   View file @ ee79d1b
1 /* 1 /*
2 * include/linux/topology.h 2 * include/linux/topology.h
3 * 3 *
4 * Written by: Matthew Dobson, IBM Corporation 4 * Written by: Matthew Dobson, IBM Corporation
5 * 5 *
6 * Copyright (C) 2002, IBM Corp. 6 * Copyright (C) 2002, IBM Corp.
7 * 7 *
8 * All rights reserved. 8 * All rights reserved.
9 * 9 *
10 * This program is free software; you can redistribute it and/or modify 10 * This program is free software; you can redistribute it and/or modify
11 * it under the terms of the GNU General Public License as published by 11 * it under the terms of the GNU General Public License as published by
12 * the Free Software Foundation; either version 2 of the License, or 12 * the Free Software Foundation; either version 2 of the License, or
13 * (at your option) any later version. 13 * (at your option) any later version.
14 * 14 *
15 * This program is distributed in the hope that it will be useful, but 15 * This program is distributed in the hope that it will be useful, but
16 * WITHOUT ANY WARRANTY; without even the implied warranty of 16 * WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or 17 * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
18 * NON INFRINGEMENT. See the GNU General Public License for more 18 * NON INFRINGEMENT. See the GNU General Public License for more
19 * details. 19 * details.
20 * 20 *
21 * You should have received a copy of the GNU General Public License 21 * You should have received a copy of the GNU General Public License
22 * along with this program; if not, write to the Free Software 22 * along with this program; if not, write to the Free Software
23 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. 23 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
24 * 24 *
25 * Send feedback to <colpatch@us.ibm.com> 25 * Send feedback to <colpatch@us.ibm.com>
26 */ 26 */
27 #ifndef _LINUX_TOPOLOGY_H 27 #ifndef _LINUX_TOPOLOGY_H
28 #define _LINUX_TOPOLOGY_H 28 #define _LINUX_TOPOLOGY_H
29 29
30 #include <linux/cpumask.h> 30 #include <linux/cpumask.h>
31 #include <linux/bitops.h> 31 #include <linux/bitops.h>
32 #include <linux/mmzone.h> 32 #include <linux/mmzone.h>
33 #include <linux/smp.h> 33 #include <linux/smp.h>
34 #include <asm/topology.h> 34 #include <asm/topology.h>
35 35
36 #ifndef node_has_online_mem 36 #ifndef node_has_online_mem
37 #define node_has_online_mem(nid) (1) 37 #define node_has_online_mem(nid) (1)
38 #endif 38 #endif
39 39
40 #ifndef nr_cpus_node 40 #ifndef nr_cpus_node
41 #define nr_cpus_node(node) \ 41 #define nr_cpus_node(node) \
42 ({ \ 42 ({ \
43 node_to_cpumask_ptr(__tmp__, node); \ 43 node_to_cpumask_ptr(__tmp__, node); \
44 cpus_weight(*__tmp__); \ 44 cpus_weight(*__tmp__); \
45 }) 45 })
46 #endif 46 #endif
47 47
48 #define for_each_node_with_cpus(node) \ 48 #define for_each_node_with_cpus(node) \
49 for_each_online_node(node) \ 49 for_each_online_node(node) \
50 if (nr_cpus_node(node)) 50 if (nr_cpus_node(node))
51 51
52 void arch_update_cpu_topology(void); 52 int arch_update_cpu_topology(void);
53 53
54 /* Conform to ACPI 2.0 SLIT distance definitions */ 54 /* Conform to ACPI 2.0 SLIT distance definitions */
55 #define LOCAL_DISTANCE 10 55 #define LOCAL_DISTANCE 10
56 #define REMOTE_DISTANCE 20 56 #define REMOTE_DISTANCE 20
57 #ifndef node_distance 57 #ifndef node_distance
58 #define node_distance(from,to) ((from) == (to) ? LOCAL_DISTANCE : REMOTE_DISTANCE) 58 #define node_distance(from,to) ((from) == (to) ? LOCAL_DISTANCE : REMOTE_DISTANCE)
59 #endif 59 #endif
60 #ifndef RECLAIM_DISTANCE 60 #ifndef RECLAIM_DISTANCE
61 /* 61 /*
62 * If the distance between nodes in a system is larger than RECLAIM_DISTANCE 62 * If the distance between nodes in a system is larger than RECLAIM_DISTANCE
63 * (in whatever arch specific measurement units returned by node_distance()) 63 * (in whatever arch specific measurement units returned by node_distance())
64 * then switch on zone reclaim on boot. 64 * then switch on zone reclaim on boot.
65 */ 65 */
66 #define RECLAIM_DISTANCE 20 66 #define RECLAIM_DISTANCE 20
67 #endif 67 #endif
68 #ifndef PENALTY_FOR_NODE_WITH_CPUS 68 #ifndef PENALTY_FOR_NODE_WITH_CPUS
69 #define PENALTY_FOR_NODE_WITH_CPUS (1) 69 #define PENALTY_FOR_NODE_WITH_CPUS (1)
70 #endif 70 #endif
71 71
72 /* 72 /*
73 * Below are the 3 major initializers used in building sched_domains: 73 * Below are the 3 major initializers used in building sched_domains:
74 * SD_SIBLING_INIT, for SMT domains 74 * SD_SIBLING_INIT, for SMT domains
75 * SD_CPU_INIT, for SMP domains 75 * SD_CPU_INIT, for SMP domains
76 * SD_NODE_INIT, for NUMA domains 76 * SD_NODE_INIT, for NUMA domains
77 * 77 *
78 * Any architecture that cares to do any tuning to these values should do so 78 * Any architecture that cares to do any tuning to these values should do so
79 * by defining their own arch-specific initializer in include/asm/topology.h. 79 * by defining their own arch-specific initializer in include/asm/topology.h.
80 * A definition there will automagically override these default initializers 80 * A definition there will automagically override these default initializers
81 * and allow arch-specific performance tuning of sched_domains. 81 * and allow arch-specific performance tuning of sched_domains.
82 * (Only non-zero and non-null fields need be specified.) 82 * (Only non-zero and non-null fields need be specified.)
83 */ 83 */
84 84
85 #ifdef CONFIG_SCHED_SMT 85 #ifdef CONFIG_SCHED_SMT
86 /* MCD - Do we really need this? It is always on if CONFIG_SCHED_SMT is, 86 /* MCD - Do we really need this? It is always on if CONFIG_SCHED_SMT is,
87 * so can't we drop this in favor of CONFIG_SCHED_SMT? 87 * so can't we drop this in favor of CONFIG_SCHED_SMT?
88 */ 88 */
89 #define ARCH_HAS_SCHED_WAKE_IDLE 89 #define ARCH_HAS_SCHED_WAKE_IDLE
90 /* Common values for SMT siblings */ 90 /* Common values for SMT siblings */
91 #ifndef SD_SIBLING_INIT 91 #ifndef SD_SIBLING_INIT
92 #define SD_SIBLING_INIT (struct sched_domain) { \ 92 #define SD_SIBLING_INIT (struct sched_domain) { \
93 .min_interval = 1, \ 93 .min_interval = 1, \
94 .max_interval = 2, \ 94 .max_interval = 2, \
95 .busy_factor = 64, \ 95 .busy_factor = 64, \
96 .imbalance_pct = 110, \ 96 .imbalance_pct = 110, \
97 .flags = SD_LOAD_BALANCE \ 97 .flags = SD_LOAD_BALANCE \
98 | SD_BALANCE_NEWIDLE \ 98 | SD_BALANCE_NEWIDLE \
99 | SD_BALANCE_FORK \ 99 | SD_BALANCE_FORK \
100 | SD_BALANCE_EXEC \ 100 | SD_BALANCE_EXEC \
101 | SD_WAKE_AFFINE \ 101 | SD_WAKE_AFFINE \
102 | SD_WAKE_BALANCE \ 102 | SD_WAKE_BALANCE \
103 | SD_SHARE_CPUPOWER, \ 103 | SD_SHARE_CPUPOWER, \
104 .last_balance = jiffies, \ 104 .last_balance = jiffies, \
105 .balance_interval = 1, \ 105 .balance_interval = 1, \
106 } 106 }
107 #endif 107 #endif
108 #endif /* CONFIG_SCHED_SMT */ 108 #endif /* CONFIG_SCHED_SMT */
109 109
110 #ifdef CONFIG_SCHED_MC 110 #ifdef CONFIG_SCHED_MC
111 /* Common values for MC siblings. for now mostly derived from SD_CPU_INIT */ 111 /* Common values for MC siblings. for now mostly derived from SD_CPU_INIT */
112 #ifndef SD_MC_INIT 112 #ifndef SD_MC_INIT
113 #define SD_MC_INIT (struct sched_domain) { \ 113 #define SD_MC_INIT (struct sched_domain) { \
114 .min_interval = 1, \ 114 .min_interval = 1, \
115 .max_interval = 4, \ 115 .max_interval = 4, \
116 .busy_factor = 64, \ 116 .busy_factor = 64, \
117 .imbalance_pct = 125, \ 117 .imbalance_pct = 125, \
118 .cache_nice_tries = 1, \ 118 .cache_nice_tries = 1, \
119 .busy_idx = 2, \ 119 .busy_idx = 2, \
120 .wake_idx = 1, \ 120 .wake_idx = 1, \
121 .forkexec_idx = 1, \ 121 .forkexec_idx = 1, \
122 .flags = SD_LOAD_BALANCE \ 122 .flags = SD_LOAD_BALANCE \
123 | SD_BALANCE_FORK \ 123 | SD_BALANCE_FORK \
124 | SD_BALANCE_EXEC \ 124 | SD_BALANCE_EXEC \
125 | SD_WAKE_AFFINE \ 125 | SD_WAKE_AFFINE \
126 | SD_WAKE_BALANCE \ 126 | SD_WAKE_BALANCE \
127 | SD_SHARE_PKG_RESOURCES\ 127 | SD_SHARE_PKG_RESOURCES\
128 | BALANCE_FOR_MC_POWER, \ 128 | BALANCE_FOR_MC_POWER, \
129 .last_balance = jiffies, \ 129 .last_balance = jiffies, \
130 .balance_interval = 1, \ 130 .balance_interval = 1, \
131 } 131 }
132 #endif 132 #endif
133 #endif /* CONFIG_SCHED_MC */ 133 #endif /* CONFIG_SCHED_MC */
134 134
135 /* Common values for CPUs */ 135 /* Common values for CPUs */
136 #ifndef SD_CPU_INIT 136 #ifndef SD_CPU_INIT
137 #define SD_CPU_INIT (struct sched_domain) { \ 137 #define SD_CPU_INIT (struct sched_domain) { \
138 .min_interval = 1, \ 138 .min_interval = 1, \
139 .max_interval = 4, \ 139 .max_interval = 4, \
140 .busy_factor = 64, \ 140 .busy_factor = 64, \
141 .imbalance_pct = 125, \ 141 .imbalance_pct = 125, \
142 .cache_nice_tries = 1, \ 142 .cache_nice_tries = 1, \
143 .busy_idx = 2, \ 143 .busy_idx = 2, \
144 .idle_idx = 1, \ 144 .idle_idx = 1, \
145 .newidle_idx = 2, \ 145 .newidle_idx = 2, \
146 .wake_idx = 1, \ 146 .wake_idx = 1, \
147 .forkexec_idx = 1, \ 147 .forkexec_idx = 1, \
148 .flags = SD_LOAD_BALANCE \ 148 .flags = SD_LOAD_BALANCE \
149 | SD_BALANCE_EXEC \ 149 | SD_BALANCE_EXEC \
150 | SD_BALANCE_FORK \ 150 | SD_BALANCE_FORK \
151 | SD_WAKE_AFFINE \ 151 | SD_WAKE_AFFINE \
152 | SD_WAKE_BALANCE \ 152 | SD_WAKE_BALANCE \
153 | BALANCE_FOR_PKG_POWER,\ 153 | BALANCE_FOR_PKG_POWER,\
154 .last_balance = jiffies, \ 154 .last_balance = jiffies, \
155 .balance_interval = 1, \ 155 .balance_interval = 1, \
156 } 156 }
157 #endif 157 #endif
158 158
159 /* sched_domains SD_ALLNODES_INIT for NUMA machines */ 159 /* sched_domains SD_ALLNODES_INIT for NUMA machines */
160 #define SD_ALLNODES_INIT (struct sched_domain) { \ 160 #define SD_ALLNODES_INIT (struct sched_domain) { \
161 .min_interval = 64, \ 161 .min_interval = 64, \
162 .max_interval = 64*num_online_cpus(), \ 162 .max_interval = 64*num_online_cpus(), \
163 .busy_factor = 128, \ 163 .busy_factor = 128, \
164 .imbalance_pct = 133, \ 164 .imbalance_pct = 133, \
165 .cache_nice_tries = 1, \ 165 .cache_nice_tries = 1, \
166 .busy_idx = 3, \ 166 .busy_idx = 3, \
167 .idle_idx = 3, \ 167 .idle_idx = 3, \
168 .flags = SD_LOAD_BALANCE \ 168 .flags = SD_LOAD_BALANCE \
169 | SD_BALANCE_NEWIDLE \ 169 | SD_BALANCE_NEWIDLE \
170 | SD_WAKE_AFFINE \ 170 | SD_WAKE_AFFINE \
171 | SD_SERIALIZE, \ 171 | SD_SERIALIZE, \
172 .last_balance = jiffies, \ 172 .last_balance = jiffies, \
173 .balance_interval = 64, \ 173 .balance_interval = 64, \
174 } 174 }
175 175
176 #ifdef CONFIG_NUMA 176 #ifdef CONFIG_NUMA
177 #ifndef SD_NODE_INIT 177 #ifndef SD_NODE_INIT
178 #error Please define an appropriate SD_NODE_INIT in include/asm/topology.h!!! 178 #error Please define an appropriate SD_NODE_INIT in include/asm/topology.h!!!
179 #endif 179 #endif
180 #endif /* CONFIG_NUMA */ 180 #endif /* CONFIG_NUMA */
181 181
182 #ifndef topology_physical_package_id 182 #ifndef topology_physical_package_id
183 #define topology_physical_package_id(cpu) ((void)(cpu), -1) 183 #define topology_physical_package_id(cpu) ((void)(cpu), -1)
184 #endif 184 #endif
185 #ifndef topology_core_id 185 #ifndef topology_core_id
186 #define topology_core_id(cpu) ((void)(cpu), 0) 186 #define topology_core_id(cpu) ((void)(cpu), 0)
187 #endif 187 #endif
188 #ifndef topology_thread_siblings 188 #ifndef topology_thread_siblings
189 #define topology_thread_siblings(cpu) cpumask_of_cpu(cpu) 189 #define topology_thread_siblings(cpu) cpumask_of_cpu(cpu)
190 #endif 190 #endif
191 #ifndef topology_core_siblings 191 #ifndef topology_core_siblings
192 #define topology_core_siblings(cpu) cpumask_of_cpu(cpu) 192 #define topology_core_siblings(cpu) cpumask_of_cpu(cpu)
193 #endif 193 #endif
194 194
195 #endif /* _LINUX_TOPOLOGY_H */ 195 #endif /* _LINUX_TOPOLOGY_H */
196 196
1 /* 1 /*
2 * kernel/sched.c 2 * kernel/sched.c
3 * 3 *
4 * Kernel scheduler and related syscalls 4 * Kernel scheduler and related syscalls
5 * 5 *
6 * Copyright (C) 1991-2002 Linus Torvalds 6 * Copyright (C) 1991-2002 Linus Torvalds
7 * 7 *
8 * 1996-12-23 Modified by Dave Grothe to fix bugs in semaphores and 8 * 1996-12-23 Modified by Dave Grothe to fix bugs in semaphores and
9 * make semaphores SMP safe 9 * make semaphores SMP safe
10 * 1998-11-19 Implemented schedule_timeout() and related stuff 10 * 1998-11-19 Implemented schedule_timeout() and related stuff
11 * by Andrea Arcangeli 11 * by Andrea Arcangeli
12 * 2002-01-04 New ultra-scalable O(1) scheduler by Ingo Molnar: 12 * 2002-01-04 New ultra-scalable O(1) scheduler by Ingo Molnar:
13 * hybrid priority-list and round-robin design with 13 * hybrid priority-list and round-robin design with
14 * an array-switch method of distributing timeslices 14 * an array-switch method of distributing timeslices
15 * and per-CPU runqueues. Cleanups and useful suggestions 15 * and per-CPU runqueues. Cleanups and useful suggestions
16 * by Davide Libenzi, preemptible kernel bits by Robert Love. 16 * by Davide Libenzi, preemptible kernel bits by Robert Love.
17 * 2003-09-03 Interactivity tuning by Con Kolivas. 17 * 2003-09-03 Interactivity tuning by Con Kolivas.
18 * 2004-04-02 Scheduler domains code by Nick Piggin 18 * 2004-04-02 Scheduler domains code by Nick Piggin
19 * 2007-04-15 Work begun on replacing all interactivity tuning with a 19 * 2007-04-15 Work begun on replacing all interactivity tuning with a
20 * fair scheduling design by Con Kolivas. 20 * fair scheduling design by Con Kolivas.
21 * 2007-05-05 Load balancing (smp-nice) and other improvements 21 * 2007-05-05 Load balancing (smp-nice) and other improvements
22 * by Peter Williams 22 * by Peter Williams
23 * 2007-05-06 Interactivity improvements to CFS by Mike Galbraith 23 * 2007-05-06 Interactivity improvements to CFS by Mike Galbraith
24 * 2007-07-01 Group scheduling enhancements by Srivatsa Vaddagiri 24 * 2007-07-01 Group scheduling enhancements by Srivatsa Vaddagiri
25 * 2007-11-29 RT balancing improvements by Steven Rostedt, Gregory Haskins, 25 * 2007-11-29 RT balancing improvements by Steven Rostedt, Gregory Haskins,
26 * Thomas Gleixner, Mike Kravetz 26 * Thomas Gleixner, Mike Kravetz
27 */ 27 */
28 28
29 #include <linux/mm.h> 29 #include <linux/mm.h>
30 #include <linux/module.h> 30 #include <linux/module.h>
31 #include <linux/nmi.h> 31 #include <linux/nmi.h>
32 #include <linux/init.h> 32 #include <linux/init.h>
33 #include <linux/uaccess.h> 33 #include <linux/uaccess.h>
34 #include <linux/highmem.h> 34 #include <linux/highmem.h>
35 #include <linux/smp_lock.h> 35 #include <linux/smp_lock.h>
36 #include <asm/mmu_context.h> 36 #include <asm/mmu_context.h>
37 #include <linux/interrupt.h> 37 #include <linux/interrupt.h>
38 #include <linux/capability.h> 38 #include <linux/capability.h>
39 #include <linux/completion.h> 39 #include <linux/completion.h>
40 #include <linux/kernel_stat.h> 40 #include <linux/kernel_stat.h>
41 #include <linux/debug_locks.h> 41 #include <linux/debug_locks.h>
42 #include <linux/security.h> 42 #include <linux/security.h>
43 #include <linux/notifier.h> 43 #include <linux/notifier.h>
44 #include <linux/profile.h> 44 #include <linux/profile.h>
45 #include <linux/freezer.h> 45 #include <linux/freezer.h>
46 #include <linux/vmalloc.h> 46 #include <linux/vmalloc.h>
47 #include <linux/blkdev.h> 47 #include <linux/blkdev.h>
48 #include <linux/delay.h> 48 #include <linux/delay.h>
49 #include <linux/pid_namespace.h> 49 #include <linux/pid_namespace.h>
50 #include <linux/smp.h> 50 #include <linux/smp.h>
51 #include <linux/threads.h> 51 #include <linux/threads.h>
52 #include <linux/timer.h> 52 #include <linux/timer.h>
53 #include <linux/rcupdate.h> 53 #include <linux/rcupdate.h>
54 #include <linux/cpu.h> 54 #include <linux/cpu.h>
55 #include <linux/cpuset.h> 55 #include <linux/cpuset.h>
56 #include <linux/percpu.h> 56 #include <linux/percpu.h>
57 #include <linux/kthread.h> 57 #include <linux/kthread.h>
58 #include <linux/proc_fs.h> 58 #include <linux/proc_fs.h>
59 #include <linux/seq_file.h> 59 #include <linux/seq_file.h>
60 #include <linux/sysctl.h> 60 #include <linux/sysctl.h>
61 #include <linux/syscalls.h> 61 #include <linux/syscalls.h>
62 #include <linux/times.h> 62 #include <linux/times.h>
63 #include <linux/tsacct_kern.h> 63 #include <linux/tsacct_kern.h>
64 #include <linux/kprobes.h> 64 #include <linux/kprobes.h>
65 #include <linux/delayacct.h> 65 #include <linux/delayacct.h>
66 #include <linux/reciprocal_div.h> 66 #include <linux/reciprocal_div.h>
67 #include <linux/unistd.h> 67 #include <linux/unistd.h>
68 #include <linux/pagemap.h> 68 #include <linux/pagemap.h>
69 #include <linux/hrtimer.h> 69 #include <linux/hrtimer.h>
70 #include <linux/tick.h> 70 #include <linux/tick.h>
71 #include <linux/bootmem.h> 71 #include <linux/bootmem.h>
72 #include <linux/debugfs.h> 72 #include <linux/debugfs.h>
73 #include <linux/ctype.h> 73 #include <linux/ctype.h>
74 #include <linux/ftrace.h> 74 #include <linux/ftrace.h>
75 #include <trace/sched.h> 75 #include <trace/sched.h>
76 76
77 #include <asm/tlb.h> 77 #include <asm/tlb.h>
78 #include <asm/irq_regs.h> 78 #include <asm/irq_regs.h>
79 79
80 #include "sched_cpupri.h" 80 #include "sched_cpupri.h"
81 81
82 /* 82 /*
83 * Convert user-nice values [ -20 ... 0 ... 19 ] 83 * Convert user-nice values [ -20 ... 0 ... 19 ]
84 * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ], 84 * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ],
85 * and back. 85 * and back.
86 */ 86 */
87 #define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20) 87 #define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20)
88 #define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20) 88 #define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20)
89 #define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio) 89 #define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio)
90 90
91 /* 91 /*
92 * 'User priority' is the nice value converted to something we 92 * 'User priority' is the nice value converted to something we
93 * can work with better when scaling various scheduler parameters, 93 * can work with better when scaling various scheduler parameters,
94 * it's a [ 0 ... 39 ] range. 94 * it's a [ 0 ... 39 ] range.
95 */ 95 */
96 #define USER_PRIO(p) ((p)-MAX_RT_PRIO) 96 #define USER_PRIO(p) ((p)-MAX_RT_PRIO)
97 #define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio) 97 #define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio)
98 #define MAX_USER_PRIO (USER_PRIO(MAX_PRIO)) 98 #define MAX_USER_PRIO (USER_PRIO(MAX_PRIO))
99 99
100 /* 100 /*
101 * Helpers for converting nanosecond timing to jiffy resolution 101 * Helpers for converting nanosecond timing to jiffy resolution
102 */ 102 */
103 #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ)) 103 #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
104 104
105 #define NICE_0_LOAD SCHED_LOAD_SCALE 105 #define NICE_0_LOAD SCHED_LOAD_SCALE
106 #define NICE_0_SHIFT SCHED_LOAD_SHIFT 106 #define NICE_0_SHIFT SCHED_LOAD_SHIFT
107 107
108 /* 108 /*
109 * These are the 'tuning knobs' of the scheduler: 109 * These are the 'tuning knobs' of the scheduler:
110 * 110 *
111 * default timeslice is 100 msecs (used only for SCHED_RR tasks). 111 * default timeslice is 100 msecs (used only for SCHED_RR tasks).
112 * Timeslices get refilled after they expire. 112 * Timeslices get refilled after they expire.
113 */ 113 */
114 #define DEF_TIMESLICE (100 * HZ / 1000) 114 #define DEF_TIMESLICE (100 * HZ / 1000)
115 115
116 /* 116 /*
117 * single value that denotes runtime == period, ie unlimited time. 117 * single value that denotes runtime == period, ie unlimited time.
118 */ 118 */
119 #define RUNTIME_INF ((u64)~0ULL) 119 #define RUNTIME_INF ((u64)~0ULL)
120 120
121 #ifdef CONFIG_SMP 121 #ifdef CONFIG_SMP
122 /* 122 /*
123 * Divide a load by a sched group cpu_power : (load / sg->__cpu_power) 123 * Divide a load by a sched group cpu_power : (load / sg->__cpu_power)
124 * Since cpu_power is a 'constant', we can use a reciprocal divide. 124 * Since cpu_power is a 'constant', we can use a reciprocal divide.
125 */ 125 */
126 static inline u32 sg_div_cpu_power(const struct sched_group *sg, u32 load) 126 static inline u32 sg_div_cpu_power(const struct sched_group *sg, u32 load)
127 { 127 {
128 return reciprocal_divide(load, sg->reciprocal_cpu_power); 128 return reciprocal_divide(load, sg->reciprocal_cpu_power);
129 } 129 }
130 130
131 /* 131 /*
132 * Each time a sched group cpu_power is changed, 132 * Each time a sched group cpu_power is changed,
133 * we must compute its reciprocal value 133 * we must compute its reciprocal value
134 */ 134 */
135 static inline void sg_inc_cpu_power(struct sched_group *sg, u32 val) 135 static inline void sg_inc_cpu_power(struct sched_group *sg, u32 val)
136 { 136 {
137 sg->__cpu_power += val; 137 sg->__cpu_power += val;
138 sg->reciprocal_cpu_power = reciprocal_value(sg->__cpu_power); 138 sg->reciprocal_cpu_power = reciprocal_value(sg->__cpu_power);
139 } 139 }
140 #endif 140 #endif
141 141
142 static inline int rt_policy(int policy) 142 static inline int rt_policy(int policy)
143 { 143 {
144 if (unlikely(policy == SCHED_FIFO || policy == SCHED_RR)) 144 if (unlikely(policy == SCHED_FIFO || policy == SCHED_RR))
145 return 1; 145 return 1;
146 return 0; 146 return 0;
147 } 147 }
148 148
149 static inline int task_has_rt_policy(struct task_struct *p) 149 static inline int task_has_rt_policy(struct task_struct *p)
150 { 150 {
151 return rt_policy(p->policy); 151 return rt_policy(p->policy);
152 } 152 }
153 153
154 /* 154 /*
155 * This is the priority-queue data structure of the RT scheduling class: 155 * This is the priority-queue data structure of the RT scheduling class:
156 */ 156 */
157 struct rt_prio_array { 157 struct rt_prio_array {
158 DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ 158 DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
159 struct list_head queue[MAX_RT_PRIO]; 159 struct list_head queue[MAX_RT_PRIO];
160 }; 160 };
161 161
162 struct rt_bandwidth { 162 struct rt_bandwidth {
163 /* nests inside the rq lock: */ 163 /* nests inside the rq lock: */
164 spinlock_t rt_runtime_lock; 164 spinlock_t rt_runtime_lock;
165 ktime_t rt_period; 165 ktime_t rt_period;
166 u64 rt_runtime; 166 u64 rt_runtime;
167 struct hrtimer rt_period_timer; 167 struct hrtimer rt_period_timer;
168 }; 168 };
169 169
170 static struct rt_bandwidth def_rt_bandwidth; 170 static struct rt_bandwidth def_rt_bandwidth;
171 171
172 static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun); 172 static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun);
173 173
174 static enum hrtimer_restart sched_rt_period_timer(struct hrtimer *timer) 174 static enum hrtimer_restart sched_rt_period_timer(struct hrtimer *timer)
175 { 175 {
176 struct rt_bandwidth *rt_b = 176 struct rt_bandwidth *rt_b =
177 container_of(timer, struct rt_bandwidth, rt_period_timer); 177 container_of(timer, struct rt_bandwidth, rt_period_timer);
178 ktime_t now; 178 ktime_t now;
179 int overrun; 179 int overrun;
180 int idle = 0; 180 int idle = 0;
181 181
182 for (;;) { 182 for (;;) {
183 now = hrtimer_cb_get_time(timer); 183 now = hrtimer_cb_get_time(timer);
184 overrun = hrtimer_forward(timer, now, rt_b->rt_period); 184 overrun = hrtimer_forward(timer, now, rt_b->rt_period);
185 185
186 if (!overrun) 186 if (!overrun)
187 break; 187 break;
188 188
189 idle = do_sched_rt_period_timer(rt_b, overrun); 189 idle = do_sched_rt_period_timer(rt_b, overrun);
190 } 190 }
191 191
192 return idle ? HRTIMER_NORESTART : HRTIMER_RESTART; 192 return idle ? HRTIMER_NORESTART : HRTIMER_RESTART;
193 } 193 }
194 194
195 static 195 static
196 void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime) 196 void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime)
197 { 197 {
198 rt_b->rt_period = ns_to_ktime(period); 198 rt_b->rt_period = ns_to_ktime(period);
199 rt_b->rt_runtime = runtime; 199 rt_b->rt_runtime = runtime;
200 200
201 spin_lock_init(&rt_b->rt_runtime_lock); 201 spin_lock_init(&rt_b->rt_runtime_lock);
202 202
203 hrtimer_init(&rt_b->rt_period_timer, 203 hrtimer_init(&rt_b->rt_period_timer,
204 CLOCK_MONOTONIC, HRTIMER_MODE_REL); 204 CLOCK_MONOTONIC, HRTIMER_MODE_REL);
205 rt_b->rt_period_timer.function = sched_rt_period_timer; 205 rt_b->rt_period_timer.function = sched_rt_period_timer;
206 rt_b->rt_period_timer.cb_mode = HRTIMER_CB_IRQSAFE_UNLOCKED; 206 rt_b->rt_period_timer.cb_mode = HRTIMER_CB_IRQSAFE_UNLOCKED;
207 } 207 }
208 208
209 static inline int rt_bandwidth_enabled(void) 209 static inline int rt_bandwidth_enabled(void)
210 { 210 {
211 return sysctl_sched_rt_runtime >= 0; 211 return sysctl_sched_rt_runtime >= 0;
212 } 212 }
213 213
214 static void start_rt_bandwidth(struct rt_bandwidth *rt_b) 214 static void start_rt_bandwidth(struct rt_bandwidth *rt_b)
215 { 215 {
216 ktime_t now; 216 ktime_t now;
217 217
218 if (rt_bandwidth_enabled() && rt_b->rt_runtime == RUNTIME_INF) 218 if (rt_bandwidth_enabled() && rt_b->rt_runtime == RUNTIME_INF)
219 return; 219 return;
220 220
221 if (hrtimer_active(&rt_b->rt_period_timer)) 221 if (hrtimer_active(&rt_b->rt_period_timer))
222 return; 222 return;
223 223
224 spin_lock(&rt_b->rt_runtime_lock); 224 spin_lock(&rt_b->rt_runtime_lock);
225 for (;;) { 225 for (;;) {
226 if (hrtimer_active(&rt_b->rt_period_timer)) 226 if (hrtimer_active(&rt_b->rt_period_timer))
227 break; 227 break;
228 228
229 now = hrtimer_cb_get_time(&rt_b->rt_period_timer); 229 now = hrtimer_cb_get_time(&rt_b->rt_period_timer);
230 hrtimer_forward(&rt_b->rt_period_timer, now, rt_b->rt_period); 230 hrtimer_forward(&rt_b->rt_period_timer, now, rt_b->rt_period);
231 hrtimer_start_expires(&rt_b->rt_period_timer, 231 hrtimer_start_expires(&rt_b->rt_period_timer,
232 HRTIMER_MODE_ABS); 232 HRTIMER_MODE_ABS);
233 } 233 }
234 spin_unlock(&rt_b->rt_runtime_lock); 234 spin_unlock(&rt_b->rt_runtime_lock);
235 } 235 }
236 236
237 #ifdef CONFIG_RT_GROUP_SCHED 237 #ifdef CONFIG_RT_GROUP_SCHED
238 static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b) 238 static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b)
239 { 239 {
240 hrtimer_cancel(&rt_b->rt_period_timer); 240 hrtimer_cancel(&rt_b->rt_period_timer);
241 } 241 }
242 #endif 242 #endif
243 243
244 /* 244 /*
245 * sched_domains_mutex serializes calls to arch_init_sched_domains, 245 * sched_domains_mutex serializes calls to arch_init_sched_domains,
246 * detach_destroy_domains and partition_sched_domains. 246 * detach_destroy_domains and partition_sched_domains.
247 */ 247 */
248 static DEFINE_MUTEX(sched_domains_mutex); 248 static DEFINE_MUTEX(sched_domains_mutex);
249 249
250 #ifdef CONFIG_GROUP_SCHED 250 #ifdef CONFIG_GROUP_SCHED
251 251
252 #include <linux/cgroup.h> 252 #include <linux/cgroup.h>
253 253
254 struct cfs_rq; 254 struct cfs_rq;
255 255
256 static LIST_HEAD(task_groups); 256 static LIST_HEAD(task_groups);
257 257
258 /* task group related information */ 258 /* task group related information */
259 struct task_group { 259 struct task_group {
260 #ifdef CONFIG_CGROUP_SCHED 260 #ifdef CONFIG_CGROUP_SCHED
261 struct cgroup_subsys_state css; 261 struct cgroup_subsys_state css;
262 #endif 262 #endif
263 263
264 #ifdef CONFIG_USER_SCHED 264 #ifdef CONFIG_USER_SCHED
265 uid_t uid; 265 uid_t uid;
266 #endif 266 #endif
267 267
268 #ifdef CONFIG_FAIR_GROUP_SCHED 268 #ifdef CONFIG_FAIR_GROUP_SCHED
269 /* schedulable entities of this group on each cpu */ 269 /* schedulable entities of this group on each cpu */
270 struct sched_entity **se; 270 struct sched_entity **se;
271 /* runqueue "owned" by this group on each cpu */ 271 /* runqueue "owned" by this group on each cpu */
272 struct cfs_rq **cfs_rq; 272 struct cfs_rq **cfs_rq;
273 unsigned long shares; 273 unsigned long shares;
274 #endif 274 #endif
275 275
276 #ifdef CONFIG_RT_GROUP_SCHED 276 #ifdef CONFIG_RT_GROUP_SCHED
277 struct sched_rt_entity **rt_se; 277 struct sched_rt_entity **rt_se;
278 struct rt_rq **rt_rq; 278 struct rt_rq **rt_rq;
279 279
280 struct rt_bandwidth rt_bandwidth; 280 struct rt_bandwidth rt_bandwidth;
281 #endif 281 #endif
282 282
283 struct rcu_head rcu; 283 struct rcu_head rcu;
284 struct list_head list; 284 struct list_head list;
285 285
286 struct task_group *parent; 286 struct task_group *parent;
287 struct list_head siblings; 287 struct list_head siblings;
288 struct list_head children; 288 struct list_head children;
289 }; 289 };
290 290
291 #ifdef CONFIG_USER_SCHED 291 #ifdef CONFIG_USER_SCHED
292 292
293 /* Helper function to pass uid information to create_sched_user() */ 293 /* Helper function to pass uid information to create_sched_user() */
294 void set_tg_uid(struct user_struct *user) 294 void set_tg_uid(struct user_struct *user)
295 { 295 {
296 user->tg->uid = user->uid; 296 user->tg->uid = user->uid;
297 } 297 }
298 298
299 /* 299 /*
300 * Root task group. 300 * Root task group.
301 * Every UID task group (including init_task_group aka UID-0) will 301 * Every UID task group (including init_task_group aka UID-0) will
302 * be a child to this group. 302 * be a child to this group.
303 */ 303 */
304 struct task_group root_task_group; 304 struct task_group root_task_group;
305 305
306 #ifdef CONFIG_FAIR_GROUP_SCHED 306 #ifdef CONFIG_FAIR_GROUP_SCHED
307 /* Default task group's sched entity on each cpu */ 307 /* Default task group's sched entity on each cpu */
308 static DEFINE_PER_CPU(struct sched_entity, init_sched_entity); 308 static DEFINE_PER_CPU(struct sched_entity, init_sched_entity);
309 /* Default task group's cfs_rq on each cpu */ 309 /* Default task group's cfs_rq on each cpu */
310 static DEFINE_PER_CPU(struct cfs_rq, init_cfs_rq) ____cacheline_aligned_in_smp; 310 static DEFINE_PER_CPU(struct cfs_rq, init_cfs_rq) ____cacheline_aligned_in_smp;
311 #endif /* CONFIG_FAIR_GROUP_SCHED */ 311 #endif /* CONFIG_FAIR_GROUP_SCHED */
312 312
313 #ifdef CONFIG_RT_GROUP_SCHED 313 #ifdef CONFIG_RT_GROUP_SCHED
314 static DEFINE_PER_CPU(struct sched_rt_entity, init_sched_rt_entity); 314 static DEFINE_PER_CPU(struct sched_rt_entity, init_sched_rt_entity);
315 static DEFINE_PER_CPU(struct rt_rq, init_rt_rq) ____cacheline_aligned_in_smp; 315 static DEFINE_PER_CPU(struct rt_rq, init_rt_rq) ____cacheline_aligned_in_smp;
316 #endif /* CONFIG_RT_GROUP_SCHED */ 316 #endif /* CONFIG_RT_GROUP_SCHED */
317 #else /* !CONFIG_USER_SCHED */ 317 #else /* !CONFIG_USER_SCHED */
318 #define root_task_group init_task_group 318 #define root_task_group init_task_group
319 #endif /* CONFIG_USER_SCHED */ 319 #endif /* CONFIG_USER_SCHED */
320 320
321 /* task_group_lock serializes add/remove of task groups and also changes to 321 /* task_group_lock serializes add/remove of task groups and also changes to
322 * a task group's cpu shares. 322 * a task group's cpu shares.
323 */ 323 */
324 static DEFINE_SPINLOCK(task_group_lock); 324 static DEFINE_SPINLOCK(task_group_lock);
325 325
326 #ifdef CONFIG_FAIR_GROUP_SCHED 326 #ifdef CONFIG_FAIR_GROUP_SCHED
327 #ifdef CONFIG_USER_SCHED 327 #ifdef CONFIG_USER_SCHED
328 # define INIT_TASK_GROUP_LOAD (2*NICE_0_LOAD) 328 # define INIT_TASK_GROUP_LOAD (2*NICE_0_LOAD)
329 #else /* !CONFIG_USER_SCHED */ 329 #else /* !CONFIG_USER_SCHED */
330 # define INIT_TASK_GROUP_LOAD NICE_0_LOAD 330 # define INIT_TASK_GROUP_LOAD NICE_0_LOAD
331 #endif /* CONFIG_USER_SCHED */ 331 #endif /* CONFIG_USER_SCHED */
332 332
333 /* 333 /*
334 * A weight of 0 or 1 can cause arithmetics problems. 334 * A weight of 0 or 1 can cause arithmetics problems.
335 * A weight of a cfs_rq is the sum of weights of which entities 335 * A weight of a cfs_rq is the sum of weights of which entities
336 * are queued on this cfs_rq, so a weight of a entity should not be 336 * are queued on this cfs_rq, so a weight of a entity should not be
337 * too large, so as the shares value of a task group. 337 * too large, so as the shares value of a task group.
338 * (The default weight is 1024 - so there's no practical 338 * (The default weight is 1024 - so there's no practical
339 * limitation from this.) 339 * limitation from this.)
340 */ 340 */
341 #define MIN_SHARES 2 341 #define MIN_SHARES 2
342 #define MAX_SHARES (1UL << 18) 342 #define MAX_SHARES (1UL << 18)
343 343
344 static int init_task_group_load = INIT_TASK_GROUP_LOAD; 344 static int init_task_group_load = INIT_TASK_GROUP_LOAD;
345 #endif 345 #endif
346 346
347 /* Default task group. 347 /* Default task group.
348 * Every task in system belong to this group at bootup. 348 * Every task in system belong to this group at bootup.
349 */ 349 */
350 struct task_group init_task_group; 350 struct task_group init_task_group;
351 351
352 /* return group to which a task belongs */ 352 /* return group to which a task belongs */
353 static inline struct task_group *task_group(struct task_struct *p) 353 static inline struct task_group *task_group(struct task_struct *p)
354 { 354 {
355 struct task_group *tg; 355 struct task_group *tg;
356 356
357 #ifdef CONFIG_USER_SCHED 357 #ifdef CONFIG_USER_SCHED
358 tg = p->user->tg; 358 tg = p->user->tg;
359 #elif defined(CONFIG_CGROUP_SCHED) 359 #elif defined(CONFIG_CGROUP_SCHED)
360 tg = container_of(task_subsys_state(p, cpu_cgroup_subsys_id), 360 tg = container_of(task_subsys_state(p, cpu_cgroup_subsys_id),
361 struct task_group, css); 361 struct task_group, css);
362 #else 362 #else
363 tg = &init_task_group; 363 tg = &init_task_group;
364 #endif 364 #endif
365 return tg; 365 return tg;
366 } 366 }
367 367
368 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ 368 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
369 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) 369 static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
370 { 370 {
371 #ifdef CONFIG_FAIR_GROUP_SCHED 371 #ifdef CONFIG_FAIR_GROUP_SCHED
372 p->se.cfs_rq = task_group(p)->cfs_rq[cpu]; 372 p->se.cfs_rq = task_group(p)->cfs_rq[cpu];
373 p->se.parent = task_group(p)->se[cpu]; 373 p->se.parent = task_group(p)->se[cpu];
374 #endif 374 #endif
375 375
376 #ifdef CONFIG_RT_GROUP_SCHED 376 #ifdef CONFIG_RT_GROUP_SCHED
377 p->rt.rt_rq = task_group(p)->rt_rq[cpu]; 377 p->rt.rt_rq = task_group(p)->rt_rq[cpu];
378 p->rt.parent = task_group(p)->rt_se[cpu]; 378 p->rt.parent = task_group(p)->rt_se[cpu];
379 #endif 379 #endif
380 } 380 }
381 381
382 #else 382 #else
383 383
384 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } 384 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
385 static inline struct task_group *task_group(struct task_struct *p) 385 static inline struct task_group *task_group(struct task_struct *p)
386 { 386 {
387 return NULL; 387 return NULL;
388 } 388 }
389 389
390 #endif /* CONFIG_GROUP_SCHED */ 390 #endif /* CONFIG_GROUP_SCHED */
391 391
392 /* CFS-related fields in a runqueue */ 392 /* CFS-related fields in a runqueue */
393 struct cfs_rq { 393 struct cfs_rq {
394 struct load_weight load; 394 struct load_weight load;
395 unsigned long nr_running; 395 unsigned long nr_running;
396 396
397 u64 exec_clock; 397 u64 exec_clock;
398 u64 min_vruntime; 398 u64 min_vruntime;
399 399
400 struct rb_root tasks_timeline; 400 struct rb_root tasks_timeline;
401 struct rb_node *rb_leftmost; 401 struct rb_node *rb_leftmost;
402 402
403 struct list_head tasks; 403 struct list_head tasks;
404 struct list_head *balance_iterator; 404 struct list_head *balance_iterator;
405 405
406 /* 406 /*
407 * 'curr' points to currently running entity on this cfs_rq. 407 * 'curr' points to currently running entity on this cfs_rq.
408 * It is set to NULL otherwise (i.e when none are currently running). 408 * It is set to NULL otherwise (i.e when none are currently running).
409 */ 409 */
410 struct sched_entity *curr, *next, *last; 410 struct sched_entity *curr, *next, *last;
411 411
412 unsigned int nr_spread_over; 412 unsigned int nr_spread_over;
413 413
414 #ifdef CONFIG_FAIR_GROUP_SCHED 414 #ifdef CONFIG_FAIR_GROUP_SCHED
415 struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ 415 struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */
416 416
417 /* 417 /*
418 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in 418 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
419 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities 419 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
420 * (like users, containers etc.) 420 * (like users, containers etc.)
421 * 421 *
422 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This 422 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This
423 * list is used during load balance. 423 * list is used during load balance.
424 */ 424 */
425 struct list_head leaf_cfs_rq_list; 425 struct list_head leaf_cfs_rq_list;
426 struct task_group *tg; /* group that "owns" this runqueue */ 426 struct task_group *tg; /* group that "owns" this runqueue */
427 427
428 #ifdef CONFIG_SMP 428 #ifdef CONFIG_SMP
429 /* 429 /*
430 * the part of load.weight contributed by tasks 430 * the part of load.weight contributed by tasks
431 */ 431 */
432 unsigned long task_weight; 432 unsigned long task_weight;
433 433
434 /* 434 /*
435 * h_load = weight * f(tg) 435 * h_load = weight * f(tg)
436 * 436 *
437 * Where f(tg) is the recursive weight fraction assigned to 437 * Where f(tg) is the recursive weight fraction assigned to
438 * this group. 438 * this group.
439 */ 439 */
440 unsigned long h_load; 440 unsigned long h_load;
441 441
442 /* 442 /*
443 * this cpu's part of tg->shares 443 * this cpu's part of tg->shares
444 */ 444 */
445 unsigned long shares; 445 unsigned long shares;
446 446
447 /* 447 /*
448 * load.weight at the time we set shares 448 * load.weight at the time we set shares
449 */ 449 */
450 unsigned long rq_weight; 450 unsigned long rq_weight;
451 #endif 451 #endif
452 #endif 452 #endif
453 }; 453 };
454 454
455 /* Real-Time classes' related field in a runqueue: */ 455 /* Real-Time classes' related field in a runqueue: */
456 struct rt_rq { 456 struct rt_rq {
457 struct rt_prio_array active; 457 struct rt_prio_array active;
458 unsigned long rt_nr_running; 458 unsigned long rt_nr_running;
459 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED 459 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
460 int highest_prio; /* highest queued rt task prio */ 460 int highest_prio; /* highest queued rt task prio */
461 #endif 461 #endif
462 #ifdef CONFIG_SMP 462 #ifdef CONFIG_SMP
463 unsigned long rt_nr_migratory; 463 unsigned long rt_nr_migratory;
464 int overloaded; 464 int overloaded;
465 #endif 465 #endif
466 int rt_throttled; 466 int rt_throttled;
467 u64 rt_time; 467 u64 rt_time;
468 u64 rt_runtime; 468 u64 rt_runtime;
469 /* Nests inside the rq lock: */ 469 /* Nests inside the rq lock: */
470 spinlock_t rt_runtime_lock; 470 spinlock_t rt_runtime_lock;
471 471
472 #ifdef CONFIG_RT_GROUP_SCHED 472 #ifdef CONFIG_RT_GROUP_SCHED
473 unsigned long rt_nr_boosted; 473 unsigned long rt_nr_boosted;
474 474
475 struct rq *rq; 475 struct rq *rq;
476 struct list_head leaf_rt_rq_list; 476 struct list_head leaf_rt_rq_list;
477 struct task_group *tg; 477 struct task_group *tg;
478 struct sched_rt_entity *rt_se; 478 struct sched_rt_entity *rt_se;
479 #endif 479 #endif
480 }; 480 };
481 481
482 #ifdef CONFIG_SMP 482 #ifdef CONFIG_SMP
483 483
484 /* 484 /*
485 * We add the notion of a root-domain which will be used to define per-domain 485 * We add the notion of a root-domain which will be used to define per-domain
486 * variables. Each exclusive cpuset essentially defines an island domain by 486 * variables. Each exclusive cpuset essentially defines an island domain by
487 * fully partitioning the member cpus from any other cpuset. Whenever a new 487 * fully partitioning the member cpus from any other cpuset. Whenever a new
488 * exclusive cpuset is created, we also create and attach a new root-domain 488 * exclusive cpuset is created, we also create and attach a new root-domain
489 * object. 489 * object.
490 * 490 *
491 */ 491 */
492 struct root_domain { 492 struct root_domain {
493 atomic_t refcount; 493 atomic_t refcount;
494 cpumask_t span; 494 cpumask_t span;
495 cpumask_t online; 495 cpumask_t online;
496 496
497 /* 497 /*
498 * The "RT overload" flag: it gets set if a CPU has more than 498 * The "RT overload" flag: it gets set if a CPU has more than
499 * one runnable RT task. 499 * one runnable RT task.
500 */ 500 */
501 cpumask_t rto_mask; 501 cpumask_t rto_mask;
502 atomic_t rto_count; 502 atomic_t rto_count;
503 #ifdef CONFIG_SMP 503 #ifdef CONFIG_SMP
504 struct cpupri cpupri; 504 struct cpupri cpupri;
505 #endif 505 #endif
506 }; 506 };
507 507
508 /* 508 /*
509 * By default the system creates a single root-domain with all cpus as 509 * By default the system creates a single root-domain with all cpus as
510 * members (mimicking the global state we have today). 510 * members (mimicking the global state we have today).
511 */ 511 */
512 static struct root_domain def_root_domain; 512 static struct root_domain def_root_domain;
513 513
514 #endif 514 #endif
515 515
516 /* 516 /*
517 * This is the main, per-CPU runqueue data structure. 517 * This is the main, per-CPU runqueue data structure.
518 * 518 *
519 * Locking rule: those places that want to lock multiple runqueues 519 * Locking rule: those places that want to lock multiple runqueues
520 * (such as the load balancing or the thread migration code), lock 520 * (such as the load balancing or the thread migration code), lock
521 * acquire operations must be ordered by ascending &runqueue. 521 * acquire operations must be ordered by ascending &runqueue.
522 */ 522 */
523 struct rq { 523 struct rq {
524 /* runqueue lock: */ 524 /* runqueue lock: */
525 spinlock_t lock; 525 spinlock_t lock;
526 526
527 /* 527 /*
528 * nr_running and cpu_load should be in the same cacheline because 528 * nr_running and cpu_load should be in the same cacheline because
529 * remote CPUs use both these fields when doing load calculation. 529 * remote CPUs use both these fields when doing load calculation.
530 */ 530 */
531 unsigned long nr_running; 531 unsigned long nr_running;
532 #define CPU_LOAD_IDX_MAX 5 532 #define CPU_LOAD_IDX_MAX 5
533 unsigned long cpu_load[CPU_LOAD_IDX_MAX]; 533 unsigned long cpu_load[CPU_LOAD_IDX_MAX];
534 unsigned char idle_at_tick; 534 unsigned char idle_at_tick;
535 #ifdef CONFIG_NO_HZ 535 #ifdef CONFIG_NO_HZ
536 unsigned long last_tick_seen; 536 unsigned long last_tick_seen;
537 unsigned char in_nohz_recently; 537 unsigned char in_nohz_recently;
538 #endif 538 #endif
539 /* capture load from *all* tasks on this cpu: */ 539 /* capture load from *all* tasks on this cpu: */
540 struct load_weight load; 540 struct load_weight load;
541 unsigned long nr_load_updates; 541 unsigned long nr_load_updates;
542 u64 nr_switches; 542 u64 nr_switches;
543 543
544 struct cfs_rq cfs; 544 struct cfs_rq cfs;
545 struct rt_rq rt; 545 struct rt_rq rt;
546 546
547 #ifdef CONFIG_FAIR_GROUP_SCHED 547 #ifdef CONFIG_FAIR_GROUP_SCHED
548 /* list of leaf cfs_rq on this cpu: */ 548 /* list of leaf cfs_rq on this cpu: */
549 struct list_head leaf_cfs_rq_list; 549 struct list_head leaf_cfs_rq_list;
550 #endif 550 #endif
551 #ifdef CONFIG_RT_GROUP_SCHED 551 #ifdef CONFIG_RT_GROUP_SCHED
552 struct list_head leaf_rt_rq_list; 552 struct list_head leaf_rt_rq_list;
553 #endif 553 #endif
554 554
555 /* 555 /*
556 * This is part of a global counter where only the total sum 556 * This is part of a global counter where only the total sum
557 * over all CPUs matters. A task can increase this counter on 557 * over all CPUs matters. A task can increase this counter on
558 * one CPU and if it got migrated afterwards it may decrease 558 * one CPU and if it got migrated afterwards it may decrease
559 * it on another CPU. Always updated under the runqueue lock: 559 * it on another CPU. Always updated under the runqueue lock:
560 */ 560 */
561 unsigned long nr_uninterruptible; 561 unsigned long nr_uninterruptible;
562 562
563 struct task_struct *curr, *idle; 563 struct task_struct *curr, *idle;
564 unsigned long next_balance; 564 unsigned long next_balance;
565 struct mm_struct *prev_mm; 565 struct mm_struct *prev_mm;
566 566
567 u64 clock; 567 u64 clock;
568 568
569 atomic_t nr_iowait; 569 atomic_t nr_iowait;
570 570
571 #ifdef CONFIG_SMP 571 #ifdef CONFIG_SMP
572 struct root_domain *rd; 572 struct root_domain *rd;
573 struct sched_domain *sd; 573 struct sched_domain *sd;
574 574
575 /* For active balancing */ 575 /* For active balancing */
576 int active_balance; 576 int active_balance;
577 int push_cpu; 577 int push_cpu;
578 /* cpu of this runqueue: */ 578 /* cpu of this runqueue: */
579 int cpu; 579 int cpu;
580 int online; 580 int online;
581 581
582 unsigned long avg_load_per_task; 582 unsigned long avg_load_per_task;
583 583
584 struct task_struct *migration_thread; 584 struct task_struct *migration_thread;
585 struct list_head migration_queue; 585 struct list_head migration_queue;
586 #endif 586 #endif
587 587
588 #ifdef CONFIG_SCHED_HRTICK 588 #ifdef CONFIG_SCHED_HRTICK
589 #ifdef CONFIG_SMP 589 #ifdef CONFIG_SMP
590 int hrtick_csd_pending; 590 int hrtick_csd_pending;
591 struct call_single_data hrtick_csd; 591 struct call_single_data hrtick_csd;
592 #endif 592 #endif
593 struct hrtimer hrtick_timer; 593 struct hrtimer hrtick_timer;
594 #endif 594 #endif
595 595
596 #ifdef CONFIG_SCHEDSTATS 596 #ifdef CONFIG_SCHEDSTATS
597 /* latency stats */ 597 /* latency stats */
598 struct sched_info rq_sched_info; 598 struct sched_info rq_sched_info;
599 599
600 /* sys_sched_yield() stats */ 600 /* sys_sched_yield() stats */
601 unsigned int yld_exp_empty; 601 unsigned int yld_exp_empty;
602 unsigned int yld_act_empty; 602 unsigned int yld_act_empty;
603 unsigned int yld_both_empty; 603 unsigned int yld_both_empty;
604 unsigned int yld_count; 604 unsigned int yld_count;
605 605
606 /* schedule() stats */ 606 /* schedule() stats */
607 unsigned int sched_switch; 607 unsigned int sched_switch;
608 unsigned int sched_count; 608 unsigned int sched_count;
609 unsigned int sched_goidle; 609 unsigned int sched_goidle;
610 610
611 /* try_to_wake_up() stats */ 611 /* try_to_wake_up() stats */
612 unsigned int ttwu_count; 612 unsigned int ttwu_count;
613 unsigned int ttwu_local; 613 unsigned int ttwu_local;
614 614
615 /* BKL stats */ 615 /* BKL stats */
616 unsigned int bkl_count; 616 unsigned int bkl_count;
617 #endif 617 #endif
618 }; 618 };
619 619
620 static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); 620 static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
621 621
622 static inline void check_preempt_curr(struct rq *rq, struct task_struct *p, int sync) 622 static inline void check_preempt_curr(struct rq *rq, struct task_struct *p, int sync)
623 { 623 {
624 rq->curr->sched_class->check_preempt_curr(rq, p, sync); 624 rq->curr->sched_class->check_preempt_curr(rq, p, sync);
625 } 625 }
626 626
627 static inline int cpu_of(struct rq *rq) 627 static inline int cpu_of(struct rq *rq)
628 { 628 {
629 #ifdef CONFIG_SMP 629 #ifdef CONFIG_SMP
630 return rq->cpu; 630 return rq->cpu;
631 #else 631 #else
632 return 0; 632 return 0;
633 #endif 633 #endif
634 } 634 }
635 635
636 /* 636 /*
637 * The domain tree (rq->sd) is protected by RCU's quiescent state transition. 637 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
638 * See detach_destroy_domains: synchronize_sched for details. 638 * See detach_destroy_domains: synchronize_sched for details.
639 * 639 *
640 * The domain tree of any CPU may only be accessed from within 640 * The domain tree of any CPU may only be accessed from within
641 * preempt-disabled sections. 641 * preempt-disabled sections.
642 */ 642 */
643 #define for_each_domain(cpu, __sd) \ 643 #define for_each_domain(cpu, __sd) \
644 for (__sd = rcu_dereference(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) 644 for (__sd = rcu_dereference(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent)
645 645
646 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) 646 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
647 #define this_rq() (&__get_cpu_var(runqueues)) 647 #define this_rq() (&__get_cpu_var(runqueues))
648 #define task_rq(p) cpu_rq(task_cpu(p)) 648 #define task_rq(p) cpu_rq(task_cpu(p))
649 #define cpu_curr(cpu) (cpu_rq(cpu)->curr) 649 #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
650 650
651 static inline void update_rq_clock(struct rq *rq) 651 static inline void update_rq_clock(struct rq *rq)
652 { 652 {
653 rq->clock = sched_clock_cpu(cpu_of(rq)); 653 rq->clock = sched_clock_cpu(cpu_of(rq));
654 } 654 }
655 655
656 /* 656 /*
657 * Tunables that become constants when CONFIG_SCHED_DEBUG is off: 657 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
658 */ 658 */
659 #ifdef CONFIG_SCHED_DEBUG 659 #ifdef CONFIG_SCHED_DEBUG
660 # define const_debug __read_mostly 660 # define const_debug __read_mostly
661 #else 661 #else
662 # define const_debug static const 662 # define const_debug static const
663 #endif 663 #endif
664 664
665 /** 665 /**
666 * runqueue_is_locked 666 * runqueue_is_locked
667 * 667 *
668 * Returns true if the current cpu runqueue is locked. 668 * Returns true if the current cpu runqueue is locked.
669 * This interface allows printk to be called with the runqueue lock 669 * This interface allows printk to be called with the runqueue lock
670 * held and know whether or not it is OK to wake up the klogd. 670 * held and know whether or not it is OK to wake up the klogd.
671 */ 671 */
672 int runqueue_is_locked(void) 672 int runqueue_is_locked(void)
673 { 673 {
674 int cpu = get_cpu(); 674 int cpu = get_cpu();
675 struct rq *rq = cpu_rq(cpu); 675 struct rq *rq = cpu_rq(cpu);
676 int ret; 676 int ret;
677 677
678 ret = spin_is_locked(&rq->lock); 678 ret = spin_is_locked(&rq->lock);
679 put_cpu(); 679 put_cpu();
680 return ret; 680 return ret;
681 } 681 }
682 682
683 /* 683 /*
684 * Debugging: various feature bits 684 * Debugging: various feature bits
685 */ 685 */
686 686
687 #define SCHED_FEAT(name, enabled) \ 687 #define SCHED_FEAT(name, enabled) \
688 __SCHED_FEAT_##name , 688 __SCHED_FEAT_##name ,
689 689
690 enum { 690 enum {
691 #include "sched_features.h" 691 #include "sched_features.h"
692 }; 692 };
693 693
694 #undef SCHED_FEAT 694 #undef SCHED_FEAT
695 695
696 #define SCHED_FEAT(name, enabled) \ 696 #define SCHED_FEAT(name, enabled) \
697 (1UL << __SCHED_FEAT_##name) * enabled | 697 (1UL << __SCHED_FEAT_##name) * enabled |
698 698
699 const_debug unsigned int sysctl_sched_features = 699 const_debug unsigned int sysctl_sched_features =
700 #include "sched_features.h" 700 #include "sched_features.h"
701 0; 701 0;
702 702
703 #undef SCHED_FEAT 703 #undef SCHED_FEAT
704 704
705 #ifdef CONFIG_SCHED_DEBUG 705 #ifdef CONFIG_SCHED_DEBUG
706 #define SCHED_FEAT(name, enabled) \ 706 #define SCHED_FEAT(name, enabled) \
707 #name , 707 #name ,
708 708
709 static __read_mostly char *sched_feat_names[] = { 709 static __read_mostly char *sched_feat_names[] = {
710 #include "sched_features.h" 710 #include "sched_features.h"
711 NULL 711 NULL
712 }; 712 };
713 713
714 #undef SCHED_FEAT 714 #undef SCHED_FEAT
715 715
716 static int sched_feat_show(struct seq_file *m, void *v) 716 static int sched_feat_show(struct seq_file *m, void *v)
717 { 717 {
718 int i; 718 int i;
719 719
720 for (i = 0; sched_feat_names[i]; i++) { 720 for (i = 0; sched_feat_names[i]; i++) {
721 if (!(sysctl_sched_features & (1UL << i))) 721 if (!(sysctl_sched_features & (1UL << i)))
722 seq_puts(m, "NO_"); 722 seq_puts(m, "NO_");
723 seq_printf(m, "%s ", sched_feat_names[i]); 723 seq_printf(m, "%s ", sched_feat_names[i]);
724 } 724 }
725 seq_puts(m, "\n"); 725 seq_puts(m, "\n");
726 726
727 return 0; 727 return 0;
728 } 728 }
729 729
730 static ssize_t 730 static ssize_t
731 sched_feat_write(struct file *filp, const char __user *ubuf, 731 sched_feat_write(struct file *filp, const char __user *ubuf,
732 size_t cnt, loff_t *ppos) 732 size_t cnt, loff_t *ppos)
733 { 733 {
734 char buf[64]; 734 char buf[64];
735 char *cmp = buf; 735 char *cmp = buf;
736 int neg = 0; 736 int neg = 0;
737 int i; 737 int i;
738 738
739 if (cnt > 63) 739 if (cnt > 63)
740 cnt = 63; 740 cnt = 63;
741 741
742 if (copy_from_user(&buf, ubuf, cnt)) 742 if (copy_from_user(&buf, ubuf, cnt))
743 return -EFAULT; 743 return -EFAULT;
744 744
745 buf[cnt] = 0; 745 buf[cnt] = 0;
746 746
747 if (strncmp(buf, "NO_", 3) == 0) { 747 if (strncmp(buf, "NO_", 3) == 0) {
748 neg = 1; 748 neg = 1;
749 cmp += 3; 749 cmp += 3;
750 } 750 }
751 751
752 for (i = 0; sched_feat_names[i]; i++) { 752 for (i = 0; sched_feat_names[i]; i++) {
753 int len = strlen(sched_feat_names[i]); 753 int len = strlen(sched_feat_names[i]);
754 754
755 if (strncmp(cmp, sched_feat_names[i], len) == 0) { 755 if (strncmp(cmp, sched_feat_names[i], len) == 0) {
756 if (neg) 756 if (neg)
757 sysctl_sched_features &= ~(1UL << i); 757 sysctl_sched_features &= ~(1UL << i);
758 else 758 else
759 sysctl_sched_features |= (1UL << i); 759 sysctl_sched_features |= (1UL << i);
760 break; 760 break;
761 } 761 }
762 } 762 }
763 763
764 if (!sched_feat_names[i]) 764 if (!sched_feat_names[i])
765 return -EINVAL; 765 return -EINVAL;
766 766
767 filp->f_pos += cnt; 767 filp->f_pos += cnt;
768 768
769 return cnt; 769 return cnt;
770 } 770 }
771 771
772 static int sched_feat_open(struct inode *inode, struct file *filp) 772 static int sched_feat_open(struct inode *inode, struct file *filp)
773 { 773 {
774 return single_open(filp, sched_feat_show, NULL); 774 return single_open(filp, sched_feat_show, NULL);
775 } 775 }
776 776
777 static struct file_operations sched_feat_fops = { 777 static struct file_operations sched_feat_fops = {
778 .open = sched_feat_open, 778 .open = sched_feat_open,
779 .write = sched_feat_write, 779 .write = sched_feat_write,
780 .read = seq_read, 780 .read = seq_read,
781 .llseek = seq_lseek, 781 .llseek = seq_lseek,
782 .release = single_release, 782 .release = single_release,
783 }; 783 };
784 784
785 static __init int sched_init_debug(void) 785 static __init int sched_init_debug(void)
786 { 786 {
787 debugfs_create_file("sched_features", 0644, NULL, NULL, 787 debugfs_create_file("sched_features", 0644, NULL, NULL,
788 &sched_feat_fops); 788 &sched_feat_fops);
789 789
790 return 0; 790 return 0;
791 } 791 }
792 late_initcall(sched_init_debug); 792 late_initcall(sched_init_debug);
793 793
794 #endif 794 #endif
795 795
796 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) 796 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
797 797
798 /* 798 /*
799 * Number of tasks to iterate in a single balance run. 799 * Number of tasks to iterate in a single balance run.
800 * Limited because this is done with IRQs disabled. 800 * Limited because this is done with IRQs disabled.
801 */ 801 */
802 const_debug unsigned int sysctl_sched_nr_migrate = 32; 802 const_debug unsigned int sysctl_sched_nr_migrate = 32;
803 803
804 /* 804 /*
805 * ratelimit for updating the group shares. 805 * ratelimit for updating the group shares.
806 * default: 0.25ms 806 * default: 0.25ms
807 */ 807 */
808 unsigned int sysctl_sched_shares_ratelimit = 250000; 808 unsigned int sysctl_sched_shares_ratelimit = 250000;
809 809
810 /* 810 /*
811 * Inject some fuzzyness into changing the per-cpu group shares 811 * Inject some fuzzyness into changing the per-cpu group shares
812 * this avoids remote rq-locks at the expense of fairness. 812 * this avoids remote rq-locks at the expense of fairness.
813 * default: 4 813 * default: 4
814 */ 814 */
815 unsigned int sysctl_sched_shares_thresh = 4; 815 unsigned int sysctl_sched_shares_thresh = 4;
816 816
817 /* 817 /*
818 * period over which we measure -rt task cpu usage in us. 818 * period over which we measure -rt task cpu usage in us.
819 * default: 1s 819 * default: 1s
820 */ 820 */
821 unsigned int sysctl_sched_rt_period = 1000000; 821 unsigned int sysctl_sched_rt_period = 1000000;
822 822
823 static __read_mostly int scheduler_running; 823 static __read_mostly int scheduler_running;
824 824
825 /* 825 /*
826 * part of the period that we allow rt tasks to run in us. 826 * part of the period that we allow rt tasks to run in us.
827 * default: 0.95s 827 * default: 0.95s
828 */ 828 */
829 int sysctl_sched_rt_runtime = 950000; 829 int sysctl_sched_rt_runtime = 950000;
830 830
831 static inline u64 global_rt_period(void) 831 static inline u64 global_rt_period(void)
832 { 832 {
833 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; 833 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
834 } 834 }
835 835
836 static inline u64 global_rt_runtime(void) 836 static inline u64 global_rt_runtime(void)
837 { 837 {
838 if (sysctl_sched_rt_runtime < 0) 838 if (sysctl_sched_rt_runtime < 0)
839 return RUNTIME_INF; 839 return RUNTIME_INF;
840 840
841 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; 841 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
842 } 842 }
843 843
844 #ifndef prepare_arch_switch 844 #ifndef prepare_arch_switch
845 # define prepare_arch_switch(next) do { } while (0) 845 # define prepare_arch_switch(next) do { } while (0)
846 #endif 846 #endif
847 #ifndef finish_arch_switch 847 #ifndef finish_arch_switch
848 # define finish_arch_switch(prev) do { } while (0) 848 # define finish_arch_switch(prev) do { } while (0)
849 #endif 849 #endif
850 850
851 static inline int task_current(struct rq *rq, struct task_struct *p) 851 static inline int task_current(struct rq *rq, struct task_struct *p)
852 { 852 {
853 return rq->curr == p; 853 return rq->curr == p;
854 } 854 }
855 855
856 #ifndef __ARCH_WANT_UNLOCKED_CTXSW 856 #ifndef __ARCH_WANT_UNLOCKED_CTXSW
857 static inline int task_running(struct rq *rq, struct task_struct *p) 857 static inline int task_running(struct rq *rq, struct task_struct *p)
858 { 858 {
859 return task_current(rq, p); 859 return task_current(rq, p);
860 } 860 }
861 861
862 static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) 862 static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next)
863 { 863 {
864 } 864 }
865 865
866 static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) 866 static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev)
867 { 867 {
868 #ifdef CONFIG_DEBUG_SPINLOCK 868 #ifdef CONFIG_DEBUG_SPINLOCK
869 /* this is a valid case when another task releases the spinlock */ 869 /* this is a valid case when another task releases the spinlock */
870 rq->lock.owner = current; 870 rq->lock.owner = current;
871 #endif 871 #endif
872 /* 872 /*
873 * If we are tracking spinlock dependencies then we have to 873 * If we are tracking spinlock dependencies then we have to
874 * fix up the runqueue lock - which gets 'carried over' from 874 * fix up the runqueue lock - which gets 'carried over' from
875 * prev into current: 875 * prev into current:
876 */ 876 */
877 spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); 877 spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_);
878 878
879 spin_unlock_irq(&rq->lock); 879 spin_unlock_irq(&rq->lock);
880 } 880 }
881 881
882 #else /* __ARCH_WANT_UNLOCKED_CTXSW */ 882 #else /* __ARCH_WANT_UNLOCKED_CTXSW */
883 static inline int task_running(struct rq *rq, struct task_struct *p) 883 static inline int task_running(struct rq *rq, struct task_struct *p)
884 { 884 {
885 #ifdef CONFIG_SMP 885 #ifdef CONFIG_SMP
886 return p->oncpu; 886 return p->oncpu;
887 #else 887 #else
888 return task_current(rq, p); 888 return task_current(rq, p);
889 #endif 889 #endif
890 } 890 }
891 891
892 static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) 892 static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next)
893 { 893 {
894 #ifdef CONFIG_SMP 894 #ifdef CONFIG_SMP
895 /* 895 /*
896 * We can optimise this out completely for !SMP, because the 896 * We can optimise this out completely for !SMP, because the
897 * SMP rebalancing from interrupt is the only thing that cares 897 * SMP rebalancing from interrupt is the only thing that cares
898 * here. 898 * here.
899 */ 899 */
900 next->oncpu = 1; 900 next->oncpu = 1;
901 #endif 901 #endif
902 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW 902 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
903 spin_unlock_irq(&rq->lock); 903 spin_unlock_irq(&rq->lock);
904 #else 904 #else
905 spin_unlock(&rq->lock); 905 spin_unlock(&rq->lock);
906 #endif 906 #endif
907 } 907 }
908 908
909 static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) 909 static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev)
910 { 910 {
911 #ifdef CONFIG_SMP 911 #ifdef CONFIG_SMP
912 /* 912 /*
913 * After ->oncpu is cleared, the task can be moved to a different CPU. 913 * After ->oncpu is cleared, the task can be moved to a different CPU.
914 * We must ensure this doesn't happen until the switch is completely 914 * We must ensure this doesn't happen until the switch is completely
915 * finished. 915 * finished.
916 */ 916 */
917 smp_wmb(); 917 smp_wmb();
918 prev->oncpu = 0; 918 prev->oncpu = 0;
919 #endif 919 #endif
920 #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW 920 #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW
921 local_irq_enable(); 921 local_irq_enable();
922 #endif 922 #endif
923 } 923 }
924 #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ 924 #endif /* __ARCH_WANT_UNLOCKED_CTXSW */
925 925
926 /* 926 /*
927 * __task_rq_lock - lock the runqueue a given task resides on. 927 * __task_rq_lock - lock the runqueue a given task resides on.
928 * Must be called interrupts disabled. 928 * Must be called interrupts disabled.
929 */ 929 */
930 static inline struct rq *__task_rq_lock(struct task_struct *p) 930 static inline struct rq *__task_rq_lock(struct task_struct *p)
931 __acquires(rq->lock) 931 __acquires(rq->lock)
932 { 932 {
933 for (;;) { 933 for (;;) {
934 struct rq *rq = task_rq(p); 934 struct rq *rq = task_rq(p);
935 spin_lock(&rq->lock); 935 spin_lock(&rq->lock);
936 if (likely(rq == task_rq(p))) 936 if (likely(rq == task_rq(p)))
937 return rq; 937 return rq;
938 spin_unlock(&rq->lock); 938 spin_unlock(&rq->lock);
939 } 939 }
940 } 940 }
941 941
942 /* 942 /*
943 * task_rq_lock - lock the runqueue a given task resides on and disable 943 * task_rq_lock - lock the runqueue a given task resides on and disable
944 * interrupts. Note the ordering: we can safely lookup the task_rq without 944 * interrupts. Note the ordering: we can safely lookup the task_rq without
945 * explicitly disabling preemption. 945 * explicitly disabling preemption.
946 */ 946 */
947 static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) 947 static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags)
948 __acquires(rq->lock) 948 __acquires(rq->lock)
949 { 949 {
950 struct rq *rq; 950 struct rq *rq;
951 951
952 for (;;) { 952 for (;;) {
953 local_irq_save(*flags); 953 local_irq_save(*flags);
954 rq = task_rq(p); 954 rq = task_rq(p);
955 spin_lock(&rq->lock); 955 spin_lock(&rq->lock);
956 if (likely(rq == task_rq(p))) 956 if (likely(rq == task_rq(p)))
957 return rq; 957 return rq;
958 spin_unlock_irqrestore(&rq->lock, *flags); 958 spin_unlock_irqrestore(&rq->lock, *flags);
959 } 959 }
960 } 960 }
961 961
962 void task_rq_unlock_wait(struct task_struct *p) 962 void task_rq_unlock_wait(struct task_struct *p)
963 { 963 {
964 struct rq *rq = task_rq(p); 964 struct rq *rq = task_rq(p);
965 965
966 smp_mb(); /* spin-unlock-wait is not a full memory barrier */ 966 smp_mb(); /* spin-unlock-wait is not a full memory barrier */
967 spin_unlock_wait(&rq->lock); 967 spin_unlock_wait(&rq->lock);
968 } 968 }
969 969
970 static void __task_rq_unlock(struct rq *rq) 970 static void __task_rq_unlock(struct rq *rq)
971 __releases(rq->lock) 971 __releases(rq->lock)
972 { 972 {
973 spin_unlock(&rq->lock); 973 spin_unlock(&rq->lock);
974 } 974 }
975 975
976 static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) 976 static inline void task_rq_unlock(struct rq *rq, unsigned long *flags)
977 __releases(rq->lock) 977 __releases(rq->lock)
978 { 978 {
979 spin_unlock_irqrestore(&rq->lock, *flags); 979 spin_unlock_irqrestore(&rq->lock, *flags);
980 } 980 }
981 981
982 /* 982 /*
983 * this_rq_lock - lock this runqueue and disable interrupts. 983 * this_rq_lock - lock this runqueue and disable interrupts.
984 */ 984 */
985 static struct rq *this_rq_lock(void) 985 static struct rq *this_rq_lock(void)
986 __acquires(rq->lock) 986 __acquires(rq->lock)
987 { 987 {
988 struct rq *rq; 988 struct rq *rq;
989 989
990 local_irq_disable(); 990 local_irq_disable();
991 rq = this_rq(); 991 rq = this_rq();
992 spin_lock(&rq->lock); 992 spin_lock(&rq->lock);
993 993
994 return rq; 994 return rq;
995 } 995 }
996 996
997 #ifdef CONFIG_SCHED_HRTICK 997 #ifdef CONFIG_SCHED_HRTICK
998 /* 998 /*
999 * Use HR-timers to deliver accurate preemption points. 999 * Use HR-timers to deliver accurate preemption points.
1000 * 1000 *
1001 * Its all a bit involved since we cannot program an hrt while holding the 1001 * Its all a bit involved since we cannot program an hrt while holding the
1002 * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a 1002 * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a
1003 * reschedule event. 1003 * reschedule event.
1004 * 1004 *
1005 * When we get rescheduled we reprogram the hrtick_timer outside of the 1005 * When we get rescheduled we reprogram the hrtick_timer outside of the
1006 * rq->lock. 1006 * rq->lock.
1007 */ 1007 */
1008 1008
1009 /* 1009 /*
1010 * Use hrtick when: 1010 * Use hrtick when:
1011 * - enabled by features 1011 * - enabled by features
1012 * - hrtimer is actually high res 1012 * - hrtimer is actually high res
1013 */ 1013 */
1014 static inline int hrtick_enabled(struct rq *rq) 1014 static inline int hrtick_enabled(struct rq *rq)
1015 { 1015 {
1016 if (!sched_feat(HRTICK)) 1016 if (!sched_feat(HRTICK))
1017 return 0; 1017 return 0;
1018 if (!cpu_active(cpu_of(rq))) 1018 if (!cpu_active(cpu_of(rq)))
1019 return 0; 1019 return 0;
1020 return hrtimer_is_hres_active(&rq->hrtick_timer); 1020 return hrtimer_is_hres_active(&rq->hrtick_timer);
1021 } 1021 }
1022 1022
1023 static void hrtick_clear(struct rq *rq) 1023 static void hrtick_clear(struct rq *rq)
1024 { 1024 {
1025 if (hrtimer_active(&rq->hrtick_timer)) 1025 if (hrtimer_active(&rq->hrtick_timer))
1026 hrtimer_cancel(&rq->hrtick_timer); 1026 hrtimer_cancel(&rq->hrtick_timer);
1027 } 1027 }
1028 1028
1029 /* 1029 /*
1030 * High-resolution timer tick. 1030 * High-resolution timer tick.
1031 * Runs from hardirq context with interrupts disabled. 1031 * Runs from hardirq context with interrupts disabled.
1032 */ 1032 */
1033 static enum hrtimer_restart hrtick(struct hrtimer *timer) 1033 static enum hrtimer_restart hrtick(struct hrtimer *timer)
1034 { 1034 {
1035 struct rq *rq = container_of(timer, struct rq, hrtick_timer); 1035 struct rq *rq = container_of(timer, struct rq, hrtick_timer);
1036 1036
1037 WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); 1037 WARN_ON_ONCE(cpu_of(rq) != smp_processor_id());
1038 1038
1039 spin_lock(&rq->lock); 1039 spin_lock(&rq->lock);
1040 update_rq_clock(rq); 1040 update_rq_clock(rq);
1041 rq->curr->sched_class->task_tick(rq, rq->curr, 1); 1041 rq->curr->sched_class->task_tick(rq, rq->curr, 1);
1042 spin_unlock(&rq->lock); 1042 spin_unlock(&rq->lock);
1043 1043
1044 return HRTIMER_NORESTART; 1044 return HRTIMER_NORESTART;
1045 } 1045 }
1046 1046
1047 #ifdef CONFIG_SMP 1047 #ifdef CONFIG_SMP
1048 /* 1048 /*
1049 * called from hardirq (IPI) context 1049 * called from hardirq (IPI) context
1050 */ 1050 */
1051 static void __hrtick_start(void *arg) 1051 static void __hrtick_start(void *arg)
1052 { 1052 {
1053 struct rq *rq = arg; 1053 struct rq *rq = arg;
1054 1054
1055 spin_lock(&rq->lock); 1055 spin_lock(&rq->lock);
1056 hrtimer_restart(&rq->hrtick_timer); 1056 hrtimer_restart(&rq->hrtick_timer);
1057 rq->hrtick_csd_pending = 0; 1057 rq->hrtick_csd_pending = 0;
1058 spin_unlock(&rq->lock); 1058 spin_unlock(&rq->lock);
1059 } 1059 }
1060 1060
1061 /* 1061 /*
1062 * Called to set the hrtick timer state. 1062 * Called to set the hrtick timer state.
1063 * 1063 *
1064 * called with rq->lock held and irqs disabled 1064 * called with rq->lock held and irqs disabled
1065 */ 1065 */
1066 static void hrtick_start(struct rq *rq, u64 delay) 1066 static void hrtick_start(struct rq *rq, u64 delay)
1067 { 1067 {
1068 struct hrtimer *timer = &rq->hrtick_timer; 1068 struct hrtimer *timer = &rq->hrtick_timer;
1069 ktime_t time = ktime_add_ns(timer->base->get_time(), delay); 1069 ktime_t time = ktime_add_ns(timer->base->get_time(), delay);
1070 1070
1071 hrtimer_set_expires(timer, time); 1071 hrtimer_set_expires(timer, time);
1072 1072
1073 if (rq == this_rq()) { 1073 if (rq == this_rq()) {
1074 hrtimer_restart(timer); 1074 hrtimer_restart(timer);
1075 } else if (!rq->hrtick_csd_pending) { 1075 } else if (!rq->hrtick_csd_pending) {
1076 __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd); 1076 __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd);
1077 rq->hrtick_csd_pending = 1; 1077 rq->hrtick_csd_pending = 1;
1078 } 1078 }
1079 } 1079 }
1080 1080
1081 static int 1081 static int
1082 hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) 1082 hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu)
1083 { 1083 {
1084 int cpu = (int)(long)hcpu; 1084 int cpu = (int)(long)hcpu;
1085 1085
1086 switch (action) { 1086 switch (action) {
1087 case CPU_UP_CANCELED: 1087 case CPU_UP_CANCELED:
1088 case CPU_UP_CANCELED_FROZEN: 1088 case CPU_UP_CANCELED_FROZEN:
1089 case CPU_DOWN_PREPARE: 1089 case CPU_DOWN_PREPARE:
1090 case CPU_DOWN_PREPARE_FROZEN: 1090 case CPU_DOWN_PREPARE_FROZEN:
1091 case CPU_DEAD: 1091 case CPU_DEAD:
1092 case CPU_DEAD_FROZEN: 1092 case CPU_DEAD_FROZEN:
1093 hrtick_clear(cpu_rq(cpu)); 1093 hrtick_clear(cpu_rq(cpu));
1094 return NOTIFY_OK; 1094 return NOTIFY_OK;
1095 } 1095 }
1096 1096
1097 return NOTIFY_DONE; 1097 return NOTIFY_DONE;
1098 } 1098 }
1099 1099
1100 static __init void init_hrtick(void) 1100 static __init void init_hrtick(void)
1101 { 1101 {
1102 hotcpu_notifier(hotplug_hrtick, 0); 1102 hotcpu_notifier(hotplug_hrtick, 0);
1103 } 1103 }
1104 #else 1104 #else
1105 /* 1105 /*
1106 * Called to set the hrtick timer state. 1106 * Called to set the hrtick timer state.
1107 * 1107 *
1108 * called with rq->lock held and irqs disabled 1108 * called with rq->lock held and irqs disabled
1109 */ 1109 */
1110 static void hrtick_start(struct rq *rq, u64 delay) 1110 static void hrtick_start(struct rq *rq, u64 delay)
1111 { 1111 {
1112 hrtimer_start(&rq->hrtick_timer, ns_to_ktime(delay), HRTIMER_MODE_REL); 1112 hrtimer_start(&rq->hrtick_timer, ns_to_ktime(delay), HRTIMER_MODE_REL);
1113 } 1113 }
1114 1114
1115 static inline void init_hrtick(void) 1115 static inline void init_hrtick(void)
1116 { 1116 {
1117 } 1117 }
1118 #endif /* CONFIG_SMP */ 1118 #endif /* CONFIG_SMP */
1119 1119
1120 static void init_rq_hrtick(struct rq *rq) 1120 static void init_rq_hrtick(struct rq *rq)
1121 { 1121 {
1122 #ifdef CONFIG_SMP 1122 #ifdef CONFIG_SMP
1123 rq->hrtick_csd_pending = 0; 1123 rq->hrtick_csd_pending = 0;
1124 1124
1125 rq->hrtick_csd.flags = 0; 1125 rq->hrtick_csd.flags = 0;
1126 rq->hrtick_csd.func = __hrtick_start; 1126 rq->hrtick_csd.func = __hrtick_start;
1127 rq->hrtick_csd.info = rq; 1127 rq->hrtick_csd.info = rq;
1128 #endif 1128 #endif
1129 1129
1130 hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); 1130 hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1131 rq->hrtick_timer.function = hrtick; 1131 rq->hrtick_timer.function = hrtick;
1132 rq->hrtick_timer.cb_mode = HRTIMER_CB_IRQSAFE_PERCPU; 1132 rq->hrtick_timer.cb_mode = HRTIMER_CB_IRQSAFE_PERCPU;
1133 } 1133 }
1134 #else /* CONFIG_SCHED_HRTICK */ 1134 #else /* CONFIG_SCHED_HRTICK */
1135 static inline void hrtick_clear(struct rq *rq) 1135 static inline void hrtick_clear(struct rq *rq)
1136 { 1136 {
1137 } 1137 }
1138 1138
1139 static inline void init_rq_hrtick(struct rq *rq) 1139 static inline void init_rq_hrtick(struct rq *rq)
1140 { 1140 {
1141 } 1141 }
1142 1142
1143 static inline void init_hrtick(void) 1143 static inline void init_hrtick(void)
1144 { 1144 {
1145 } 1145 }
1146 #endif /* CONFIG_SCHED_HRTICK */ 1146 #endif /* CONFIG_SCHED_HRTICK */
1147 1147
1148 /* 1148 /*
1149 * resched_task - mark a task 'to be rescheduled now'. 1149 * resched_task - mark a task 'to be rescheduled now'.
1150 * 1150 *
1151 * On UP this means the setting of the need_resched flag, on SMP it 1151 * On UP this means the setting of the need_resched flag, on SMP it
1152 * might also involve a cross-CPU call to trigger the scheduler on 1152 * might also involve a cross-CPU call to trigger the scheduler on
1153 * the target CPU. 1153 * the target CPU.
1154 */ 1154 */
1155 #ifdef CONFIG_SMP 1155 #ifdef CONFIG_SMP
1156 1156
1157 #ifndef tsk_is_polling 1157 #ifndef tsk_is_polling
1158 #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) 1158 #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG)
1159 #endif 1159 #endif
1160 1160
1161 static void resched_task(struct task_struct *p) 1161 static void resched_task(struct task_struct *p)
1162 { 1162 {
1163 int cpu; 1163 int cpu;
1164 1164
1165 assert_spin_locked(&task_rq(p)->lock); 1165 assert_spin_locked(&task_rq(p)->lock);
1166 1166
1167 if (unlikely(test_tsk_thread_flag(p, TIF_NEED_RESCHED))) 1167 if (unlikely(test_tsk_thread_flag(p, TIF_NEED_RESCHED)))
1168 return; 1168 return;
1169 1169
1170 set_tsk_thread_flag(p, TIF_NEED_RESCHED); 1170 set_tsk_thread_flag(p, TIF_NEED_RESCHED);
1171 1171
1172 cpu = task_cpu(p); 1172 cpu = task_cpu(p);
1173 if (cpu == smp_processor_id()) 1173 if (cpu == smp_processor_id())
1174 return; 1174 return;
1175 1175
1176 /* NEED_RESCHED must be visible before we test polling */ 1176 /* NEED_RESCHED must be visible before we test polling */
1177 smp_mb(); 1177 smp_mb();
1178 if (!tsk_is_polling(p)) 1178 if (!tsk_is_polling(p))
1179 smp_send_reschedule(cpu); 1179 smp_send_reschedule(cpu);
1180 } 1180 }
1181 1181
1182 static void resched_cpu(int cpu) 1182 static void resched_cpu(int cpu)
1183 { 1183 {
1184 struct rq *rq = cpu_rq(cpu); 1184 struct rq *rq = cpu_rq(cpu);
1185 unsigned long flags; 1185 unsigned long flags;
1186 1186
1187 if (!spin_trylock_irqsave(&rq->lock, flags)) 1187 if (!spin_trylock_irqsave(&rq->lock, flags))
1188 return; 1188 return;
1189 resched_task(cpu_curr(cpu)); 1189 resched_task(cpu_curr(cpu));
1190 spin_unlock_irqrestore(&rq->lock, flags); 1190 spin_unlock_irqrestore(&rq->lock, flags);
1191 } 1191 }
1192 1192
1193 #ifdef CONFIG_NO_HZ 1193 #ifdef CONFIG_NO_HZ
1194 /* 1194 /*
1195 * When add_timer_on() enqueues a timer into the timer wheel of an 1195 * When add_timer_on() enqueues a timer into the timer wheel of an
1196 * idle CPU then this timer might expire before the next timer event 1196 * idle CPU then this timer might expire before the next timer event
1197 * which is scheduled to wake up that CPU. In case of a completely 1197 * which is scheduled to wake up that CPU. In case of a completely
1198 * idle system the next event might even be infinite time into the 1198 * idle system the next event might even be infinite time into the
1199 * future. wake_up_idle_cpu() ensures that the CPU is woken up and 1199 * future. wake_up_idle_cpu() ensures that the CPU is woken up and
1200 * leaves the inner idle loop so the newly added timer is taken into 1200 * leaves the inner idle loop so the newly added timer is taken into
1201 * account when the CPU goes back to idle and evaluates the timer 1201 * account when the CPU goes back to idle and evaluates the timer
1202 * wheel for the next timer event. 1202 * wheel for the next timer event.
1203 */ 1203 */
1204 void wake_up_idle_cpu(int cpu) 1204 void wake_up_idle_cpu(int cpu)
1205 { 1205 {
1206 struct rq *rq = cpu_rq(cpu); 1206 struct rq *rq = cpu_rq(cpu);
1207 1207
1208 if (cpu == smp_processor_id()) 1208 if (cpu == smp_processor_id())
1209 return; 1209 return;
1210 1210
1211 /* 1211 /*
1212 * This is safe, as this function is called with the timer 1212 * This is safe, as this function is called with the timer
1213 * wheel base lock of (cpu) held. When the CPU is on the way 1213 * wheel base lock of (cpu) held. When the CPU is on the way
1214 * to idle and has not yet set rq->curr to idle then it will 1214 * to idle and has not yet set rq->curr to idle then it will
1215 * be serialized on the timer wheel base lock and take the new 1215 * be serialized on the timer wheel base lock and take the new
1216 * timer into account automatically. 1216 * timer into account automatically.
1217 */ 1217 */
1218 if (rq->curr != rq->idle) 1218 if (rq->curr != rq->idle)
1219 return; 1219 return;
1220 1220
1221 /* 1221 /*
1222 * We can set TIF_RESCHED on the idle task of the other CPU 1222 * We can set TIF_RESCHED on the idle task of the other CPU
1223 * lockless. The worst case is that the other CPU runs the 1223 * lockless. The worst case is that the other CPU runs the
1224 * idle task through an additional NOOP schedule() 1224 * idle task through an additional NOOP schedule()
1225 */ 1225 */
1226 set_tsk_thread_flag(rq->idle, TIF_NEED_RESCHED); 1226 set_tsk_thread_flag(rq->idle, TIF_NEED_RESCHED);
1227 1227
1228 /* NEED_RESCHED must be visible before we test polling */ 1228 /* NEED_RESCHED must be visible before we test polling */
1229 smp_mb(); 1229 smp_mb();
1230 if (!tsk_is_polling(rq->idle)) 1230 if (!tsk_is_polling(rq->idle))
1231 smp_send_reschedule(cpu); 1231 smp_send_reschedule(cpu);
1232 } 1232 }
1233 #endif /* CONFIG_NO_HZ */ 1233 #endif /* CONFIG_NO_HZ */
1234 1234
1235 #else /* !CONFIG_SMP */ 1235 #else /* !CONFIG_SMP */
1236 static void resched_task(struct task_struct *p) 1236 static void resched_task(struct task_struct *p)
1237 { 1237 {
1238 assert_spin_locked(&task_rq(p)->lock); 1238 assert_spin_locked(&task_rq(p)->lock);
1239 set_tsk_need_resched(p); 1239 set_tsk_need_resched(p);
1240 } 1240 }
1241 #endif /* CONFIG_SMP */ 1241 #endif /* CONFIG_SMP */
1242 1242
1243 #if BITS_PER_LONG == 32 1243 #if BITS_PER_LONG == 32
1244 # define WMULT_CONST (~0UL) 1244 # define WMULT_CONST (~0UL)
1245 #else 1245 #else
1246 # define WMULT_CONST (1UL << 32) 1246 # define WMULT_CONST (1UL << 32)
1247 #endif 1247 #endif
1248 1248
1249 #define WMULT_SHIFT 32 1249 #define WMULT_SHIFT 32
1250 1250
1251 /* 1251 /*
1252 * Shift right and round: 1252 * Shift right and round:
1253 */ 1253 */
1254 #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) 1254 #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y))
1255 1255
1256 /* 1256 /*
1257 * delta *= weight / lw 1257 * delta *= weight / lw
1258 */ 1258 */
1259 static unsigned long 1259 static unsigned long
1260 calc_delta_mine(unsigned long delta_exec, unsigned long weight, 1260 calc_delta_mine(unsigned long delta_exec, unsigned long weight,
1261 struct load_weight *lw) 1261 struct load_weight *lw)
1262 { 1262 {
1263 u64 tmp; 1263 u64 tmp;
1264 1264
1265 if (!lw->inv_weight) { 1265 if (!lw->inv_weight) {
1266 if (BITS_PER_LONG > 32 && unlikely(lw->weight >= WMULT_CONST)) 1266 if (BITS_PER_LONG > 32 && unlikely(lw->weight >= WMULT_CONST))
1267 lw->inv_weight = 1; 1267 lw->inv_weight = 1;
1268 else 1268 else
1269 lw->inv_weight = 1 + (WMULT_CONST-lw->weight/2) 1269 lw->inv_weight = 1 + (WMULT_CONST-lw->weight/2)
1270 / (lw->weight+1); 1270 / (lw->weight+1);
1271 } 1271 }
1272 1272
1273 tmp = (u64)delta_exec * weight; 1273 tmp = (u64)delta_exec * weight;
1274 /* 1274 /*
1275 * Check whether we'd overflow the 64-bit multiplication: 1275 * Check whether we'd overflow the 64-bit multiplication:
1276 */ 1276 */
1277 if (unlikely(tmp > WMULT_CONST)) 1277 if (unlikely(tmp > WMULT_CONST))
1278 tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, 1278 tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight,
1279 WMULT_SHIFT/2); 1279 WMULT_SHIFT/2);
1280 else 1280 else
1281 tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); 1281 tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT);
1282 1282
1283 return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); 1283 return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX);
1284 } 1284 }
1285 1285
1286 static inline void update_load_add(struct load_weight *lw, unsigned long inc) 1286 static inline void update_load_add(struct load_weight *lw, unsigned long inc)
1287 { 1287 {
1288 lw->weight += inc; 1288 lw->weight += inc;
1289 lw->inv_weight = 0; 1289 lw->inv_weight = 0;
1290 } 1290 }
1291 1291
1292 static inline void update_load_sub(struct load_weight *lw, unsigned long dec) 1292 static inline void update_load_sub(struct load_weight *lw, unsigned long dec)
1293 { 1293 {
1294 lw->weight -= dec; 1294 lw->weight -= dec;
1295 lw->inv_weight = 0; 1295 lw->inv_weight = 0;
1296 } 1296 }
1297 1297
1298 /* 1298 /*
1299 * To aid in avoiding the subversion of "niceness" due to uneven distribution 1299 * To aid in avoiding the subversion of "niceness" due to uneven distribution
1300 * of tasks with abnormal "nice" values across CPUs the contribution that 1300 * of tasks with abnormal "nice" values across CPUs the contribution that
1301 * each task makes to its run queue's load is weighted according to its 1301 * each task makes to its run queue's load is weighted according to its
1302 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a 1302 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
1303 * scaled version of the new time slice allocation that they receive on time 1303 * scaled version of the new time slice allocation that they receive on time
1304 * slice expiry etc. 1304 * slice expiry etc.
1305 */ 1305 */
1306 1306
1307 #define WEIGHT_IDLEPRIO 2 1307 #define WEIGHT_IDLEPRIO 2
1308 #define WMULT_IDLEPRIO (1 << 31) 1308 #define WMULT_IDLEPRIO (1 << 31)
1309 1309
1310 /* 1310 /*
1311 * Nice levels are multiplicative, with a gentle 10% change for every 1311 * Nice levels are multiplicative, with a gentle 10% change for every
1312 * nice level changed. I.e. when a CPU-bound task goes from nice 0 to 1312 * nice level changed. I.e. when a CPU-bound task goes from nice 0 to
1313 * nice 1, it will get ~10% less CPU time than another CPU-bound task 1313 * nice 1, it will get ~10% less CPU time than another CPU-bound task
1314 * that remained on nice 0. 1314 * that remained on nice 0.
1315 * 1315 *
1316 * The "10% effect" is relative and cumulative: from _any_ nice level, 1316 * The "10% effect" is relative and cumulative: from _any_ nice level,
1317 * if you go up 1 level, it's -10% CPU usage, if you go down 1 level 1317 * if you go up 1 level, it's -10% CPU usage, if you go down 1 level
1318 * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. 1318 * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25.
1319 * If a task goes up by ~10% and another task goes down by ~10% then 1319 * If a task goes up by ~10% and another task goes down by ~10% then
1320 * the relative distance between them is ~25%.) 1320 * the relative distance between them is ~25%.)
1321 */ 1321 */
1322 static const int prio_to_weight[40] = { 1322 static const int prio_to_weight[40] = {
1323 /* -20 */ 88761, 71755, 56483, 46273, 36291, 1323 /* -20 */ 88761, 71755, 56483, 46273, 36291,
1324 /* -15 */ 29154, 23254, 18705, 14949, 11916, 1324 /* -15 */ 29154, 23254, 18705, 14949, 11916,
1325 /* -10 */ 9548, 7620, 6100, 4904, 3906, 1325 /* -10 */ 9548, 7620, 6100, 4904, 3906,
1326 /* -5 */ 3121, 2501, 1991, 1586, 1277, 1326 /* -5 */ 3121, 2501, 1991, 1586, 1277,
1327 /* 0 */ 1024, 820, 655, 526, 423, 1327 /* 0 */ 1024, 820, 655, 526, 423,
1328 /* 5 */ 335, 272, 215, 172, 137, 1328 /* 5 */ 335, 272, 215, 172, 137,
1329 /* 10 */ 110, 87, 70, 56, 45, 1329 /* 10 */ 110, 87, 70, 56, 45,
1330 /* 15 */ 36, 29, 23, 18, 15, 1330 /* 15 */ 36, 29, 23, 18, 15,
1331 }; 1331 };
1332 1332
1333 /* 1333 /*
1334 * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. 1334 * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated.
1335 * 1335 *
1336 * In cases where the weight does not change often, we can use the 1336 * In cases where the weight does not change often, we can use the
1337 * precalculated inverse to speed up arithmetics by turning divisions 1337 * precalculated inverse to speed up arithmetics by turning divisions
1338 * into multiplications: 1338 * into multiplications:
1339 */ 1339 */
1340 static const u32 prio_to_wmult[40] = { 1340 static const u32 prio_to_wmult[40] = {
1341 /* -20 */ 48388, 59856, 76040, 92818, 118348, 1341 /* -20 */ 48388, 59856, 76040, 92818, 118348,
1342 /* -15 */ 147320, 184698, 229616, 287308, 360437, 1342 /* -15 */ 147320, 184698, 229616, 287308, 360437,
1343 /* -10 */ 449829, 563644, 704093, 875809, 1099582, 1343 /* -10 */ 449829, 563644, 704093, 875809, 1099582,
1344 /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, 1344 /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326,
1345 /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, 1345 /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587,
1346 /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, 1346 /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126,
1347 /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, 1347 /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717,
1348 /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, 1348 /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153,
1349 }; 1349 };
1350 1350
1351 static void activate_task(struct rq *rq, struct task_struct *p, int wakeup); 1351 static void activate_task(struct rq *rq, struct task_struct *p, int wakeup);
1352 1352
1353 /* 1353 /*
1354 * runqueue iterator, to support SMP load-balancing between different 1354 * runqueue iterator, to support SMP load-balancing between different
1355 * scheduling classes, without having to expose their internal data 1355 * scheduling classes, without having to expose their internal data
1356 * structures to the load-balancing proper: 1356 * structures to the load-balancing proper:
1357 */ 1357 */
1358 struct rq_iterator { 1358 struct rq_iterator {
1359 void *arg; 1359 void *arg;
1360 struct task_struct *(*start)(void *); 1360 struct task_struct *(*start)(void *);
1361 struct task_struct *(*next)(void *); 1361 struct task_struct *(*next)(void *);
1362 }; 1362 };
1363 1363
1364 #ifdef CONFIG_SMP 1364 #ifdef CONFIG_SMP
1365 static unsigned long 1365 static unsigned long
1366 balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, 1366 balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest,
1367 unsigned long max_load_move, struct sched_domain *sd, 1367 unsigned long max_load_move, struct sched_domain *sd,
1368 enum cpu_idle_type idle, int *all_pinned, 1368 enum cpu_idle_type idle, int *all_pinned,
1369 int *this_best_prio, struct rq_iterator *iterator); 1369 int *this_best_prio, struct rq_iterator *iterator);
1370 1370
1371 static int 1371 static int
1372 iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, 1372 iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest,
1373 struct sched_domain *sd, enum cpu_idle_type idle, 1373 struct sched_domain *sd, enum cpu_idle_type idle,
1374 struct rq_iterator *iterator); 1374 struct rq_iterator *iterator);
1375 #endif 1375 #endif
1376 1376
1377 #ifdef CONFIG_CGROUP_CPUACCT 1377 #ifdef CONFIG_CGROUP_CPUACCT
1378 static void cpuacct_charge(struct task_struct *tsk, u64 cputime); 1378 static void cpuacct_charge(struct task_struct *tsk, u64 cputime);
1379 #else 1379 #else
1380 static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} 1380 static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {}
1381 #endif 1381 #endif
1382 1382
1383 static inline void inc_cpu_load(struct rq *rq, unsigned long load) 1383 static inline void inc_cpu_load(struct rq *rq, unsigned long load)
1384 { 1384 {
1385 update_load_add(&rq->load, load); 1385 update_load_add(&rq->load, load);
1386 } 1386 }
1387 1387
1388 static inline void dec_cpu_load(struct rq *rq, unsigned long load) 1388 static inline void dec_cpu_load(struct rq *rq, unsigned long load)
1389 { 1389 {
1390 update_load_sub(&rq->load, load); 1390 update_load_sub(&rq->load, load);
1391 } 1391 }
1392 1392
1393 #if (defined(CONFIG_SMP) && defined(CONFIG_FAIR_GROUP_SCHED)) || defined(CONFIG_RT_GROUP_SCHED) 1393 #if (defined(CONFIG_SMP) && defined(CONFIG_FAIR_GROUP_SCHED)) || defined(CONFIG_RT_GROUP_SCHED)
1394 typedef int (*tg_visitor)(struct task_group *, void *); 1394 typedef int (*tg_visitor)(struct task_group *, void *);
1395 1395
1396 /* 1396 /*
1397 * Iterate the full tree, calling @down when first entering a node and @up when 1397 * Iterate the full tree, calling @down when first entering a node and @up when
1398 * leaving it for the final time. 1398 * leaving it for the final time.
1399 */ 1399 */
1400 static int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) 1400 static int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
1401 { 1401 {
1402 struct task_group *parent, *child; 1402 struct task_group *parent, *child;
1403 int ret; 1403 int ret;
1404 1404
1405 rcu_read_lock(); 1405 rcu_read_lock();
1406 parent = &root_task_group; 1406 parent = &root_task_group;
1407 down: 1407 down:
1408 ret = (*down)(parent, data); 1408 ret = (*down)(parent, data);
1409 if (ret) 1409 if (ret)
1410 goto out_unlock; 1410 goto out_unlock;
1411 list_for_each_entry_rcu(child, &parent->children, siblings) { 1411 list_for_each_entry_rcu(child, &parent->children, siblings) {
1412 parent = child; 1412 parent = child;
1413 goto down; 1413 goto down;
1414 1414
1415 up: 1415 up:
1416 continue; 1416 continue;
1417 } 1417 }
1418 ret = (*up)(parent, data); 1418 ret = (*up)(parent, data);
1419 if (ret) 1419 if (ret)
1420 goto out_unlock; 1420 goto out_unlock;
1421 1421
1422 child = parent; 1422 child = parent;
1423 parent = parent->parent; 1423 parent = parent->parent;
1424 if (parent) 1424 if (parent)
1425 goto up; 1425 goto up;
1426 out_unlock: 1426 out_unlock:
1427 rcu_read_unlock(); 1427 rcu_read_unlock();
1428 1428
1429 return ret; 1429 return ret;
1430 } 1430 }
1431 1431
1432 static int tg_nop(struct task_group *tg, void *data) 1432 static int tg_nop(struct task_group *tg, void *data)
1433 { 1433 {
1434 return 0; 1434 return 0;
1435 } 1435 }
1436 #endif 1436 #endif
1437 1437
1438 #ifdef CONFIG_SMP 1438 #ifdef CONFIG_SMP
1439 static unsigned long source_load(int cpu, int type); 1439 static unsigned long source_load(int cpu, int type);
1440 static unsigned long target_load(int cpu, int type); 1440 static unsigned long target_load(int cpu, int type);
1441 static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd); 1441 static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd);
1442 1442
1443 static unsigned long cpu_avg_load_per_task(int cpu) 1443 static unsigned long cpu_avg_load_per_task(int cpu)
1444 { 1444 {
1445 struct rq *rq = cpu_rq(cpu); 1445 struct rq *rq = cpu_rq(cpu);
1446 unsigned long nr_running = ACCESS_ONCE(rq->nr_running); 1446 unsigned long nr_running = ACCESS_ONCE(rq->nr_running);
1447 1447
1448 if (nr_running) 1448 if (nr_running)
1449 rq->avg_load_per_task = rq->load.weight / nr_running; 1449 rq->avg_load_per_task = rq->load.weight / nr_running;
1450 else 1450 else
1451 rq->avg_load_per_task = 0; 1451 rq->avg_load_per_task = 0;
1452 1452
1453 return rq->avg_load_per_task; 1453 return rq->avg_load_per_task;
1454 } 1454 }
1455 1455
1456 #ifdef CONFIG_FAIR_GROUP_SCHED 1456 #ifdef CONFIG_FAIR_GROUP_SCHED
1457 1457
1458 static void __set_se_shares(struct sched_entity *se, unsigned long shares); 1458 static void __set_se_shares(struct sched_entity *se, unsigned long shares);
1459 1459
1460 /* 1460 /*
1461 * Calculate and set the cpu's group shares. 1461 * Calculate and set the cpu's group shares.
1462 */ 1462 */
1463 static void 1463 static void
1464 update_group_shares_cpu(struct task_group *tg, int cpu, 1464 update_group_shares_cpu(struct task_group *tg, int cpu,
1465 unsigned long sd_shares, unsigned long sd_rq_weight) 1465 unsigned long sd_shares, unsigned long sd_rq_weight)
1466 { 1466 {
1467 unsigned long shares; 1467 unsigned long shares;
1468 unsigned long rq_weight; 1468 unsigned long rq_weight;
1469 1469
1470 if (!tg->se[cpu]) 1470 if (!tg->se[cpu])
1471 return; 1471 return;
1472 1472
1473 rq_weight = tg->cfs_rq[cpu]->rq_weight; 1473 rq_weight = tg->cfs_rq[cpu]->rq_weight;
1474 1474
1475 /* 1475 /*
1476 * \Sum shares * rq_weight 1476 * \Sum shares * rq_weight
1477 * shares = ----------------------- 1477 * shares = -----------------------
1478 * \Sum rq_weight 1478 * \Sum rq_weight
1479 * 1479 *
1480 */ 1480 */
1481 shares = (sd_shares * rq_weight) / sd_rq_weight; 1481 shares = (sd_shares * rq_weight) / sd_rq_weight;
1482 shares = clamp_t(unsigned long, shares, MIN_SHARES, MAX_SHARES); 1482 shares = clamp_t(unsigned long, shares, MIN_SHARES, MAX_SHARES);
1483 1483
1484 if (abs(shares - tg->se[cpu]->load.weight) > 1484 if (abs(shares - tg->se[cpu]->load.weight) >
1485 sysctl_sched_shares_thresh) { 1485 sysctl_sched_shares_thresh) {
1486 struct rq *rq = cpu_rq(cpu); 1486 struct rq *rq = cpu_rq(cpu);
1487 unsigned long flags; 1487 unsigned long flags;
1488 1488
1489 spin_lock_irqsave(&rq->lock, flags); 1489 spin_lock_irqsave(&rq->lock, flags);
1490 tg->cfs_rq[cpu]->shares = shares; 1490 tg->cfs_rq[cpu]->shares = shares;
1491 1491
1492 __set_se_shares(tg->se[cpu], shares); 1492 __set_se_shares(tg->se[cpu], shares);
1493 spin_unlock_irqrestore(&rq->lock, flags); 1493 spin_unlock_irqrestore(&rq->lock, flags);
1494 } 1494 }
1495 } 1495 }
1496 1496
1497 /* 1497 /*
1498 * Re-compute the task group their per cpu shares over the given domain. 1498 * Re-compute the task group their per cpu shares over the given domain.
1499 * This needs to be done in a bottom-up fashion because the rq weight of a 1499 * This needs to be done in a bottom-up fashion because the rq weight of a
1500 * parent group depends on the shares of its child groups. 1500 * parent group depends on the shares of its child groups.
1501 */ 1501 */
1502 static int tg_shares_up(struct task_group *tg, void *data) 1502 static int tg_shares_up(struct task_group *tg, void *data)
1503 { 1503 {
1504 unsigned long weight, rq_weight = 0; 1504 unsigned long weight, rq_weight = 0;
1505 unsigned long shares = 0; 1505 unsigned long shares = 0;
1506 struct sched_domain *sd = data; 1506 struct sched_domain *sd = data;
1507 int i; 1507 int i;
1508 1508
1509 for_each_cpu_mask(i, sd->span) { 1509 for_each_cpu_mask(i, sd->span) {
1510 /* 1510 /*
1511 * If there are currently no tasks on the cpu pretend there 1511 * If there are currently no tasks on the cpu pretend there
1512 * is one of average load so that when a new task gets to 1512 * is one of average load so that when a new task gets to
1513 * run here it will not get delayed by group starvation. 1513 * run here it will not get delayed by group starvation.
1514 */ 1514 */
1515 weight = tg->cfs_rq[i]->load.weight; 1515 weight = tg->cfs_rq[i]->load.weight;
1516 if (!weight) 1516 if (!weight)
1517 weight = NICE_0_LOAD; 1517 weight = NICE_0_LOAD;
1518 1518
1519 tg->cfs_rq[i]->rq_weight = weight; 1519 tg->cfs_rq[i]->rq_weight = weight;
1520 rq_weight += weight; 1520 rq_weight += weight;
1521 shares += tg->cfs_rq[i]->shares; 1521 shares += tg->cfs_rq[i]->shares;
1522 } 1522 }
1523 1523
1524 if ((!shares && rq_weight) || shares > tg->shares) 1524 if ((!shares && rq_weight) || shares > tg->shares)
1525 shares = tg->shares; 1525 shares = tg->shares;
1526 1526
1527 if (!sd->parent || !(sd->parent->flags & SD_LOAD_BALANCE)) 1527 if (!sd->parent || !(sd->parent->flags & SD_LOAD_BALANCE))
1528 shares = tg->shares; 1528 shares = tg->shares;
1529 1529
1530 for_each_cpu_mask(i, sd->span) 1530 for_each_cpu_mask(i, sd->span)
1531 update_group_shares_cpu(tg, i, shares, rq_weight); 1531 update_group_shares_cpu(tg, i, shares, rq_weight);
1532 1532
1533 return 0; 1533 return 0;
1534 } 1534 }
1535 1535
1536 /* 1536 /*
1537 * Compute the cpu's hierarchical load factor for each task group. 1537 * Compute the cpu's hierarchical load factor for each task group.
1538 * This needs to be done in a top-down fashion because the load of a child 1538 * This needs to be done in a top-down fashion because the load of a child
1539 * group is a fraction of its parents load. 1539 * group is a fraction of its parents load.
1540 */ 1540 */
1541 static int tg_load_down(struct task_group *tg, void *data) 1541 static int tg_load_down(struct task_group *tg, void *data)
1542 { 1542 {
1543 unsigned long load; 1543 unsigned long load;
1544 long cpu = (long)data; 1544 long cpu = (long)data;
1545 1545
1546 if (!tg->parent) { 1546 if (!tg->parent) {
1547 load = cpu_rq(cpu)->load.weight; 1547 load = cpu_rq(cpu)->load.weight;
1548 } else { 1548 } else {
1549 load = tg->parent->cfs_rq[cpu]->h_load; 1549 load = tg->parent->cfs_rq[cpu]->h_load;
1550 load *= tg->cfs_rq[cpu]->shares; 1550 load *= tg->cfs_rq[cpu]->shares;
1551 load /= tg->parent->cfs_rq[cpu]->load.weight + 1; 1551 load /= tg->parent->cfs_rq[cpu]->load.weight + 1;
1552 } 1552 }
1553 1553
1554 tg->cfs_rq[cpu]->h_load = load; 1554 tg->cfs_rq[cpu]->h_load = load;
1555 1555
1556 return 0; 1556 return 0;
1557 } 1557 }
1558 1558
1559 static void update_shares(struct sched_domain *sd) 1559 static void update_shares(struct sched_domain *sd)
1560 { 1560 {
1561 u64 now = cpu_clock(raw_smp_processor_id()); 1561 u64 now = cpu_clock(raw_smp_processor_id());
1562 s64 elapsed = now - sd->last_update; 1562 s64 elapsed = now - sd->last_update;
1563 1563
1564 if (elapsed >= (s64)(u64)sysctl_sched_shares_ratelimit) { 1564 if (elapsed >= (s64)(u64)sysctl_sched_shares_ratelimit) {
1565 sd->last_update = now; 1565 sd->last_update = now;
1566 walk_tg_tree(tg_nop, tg_shares_up, sd); 1566 walk_tg_tree(tg_nop, tg_shares_up, sd);
1567 } 1567 }
1568 } 1568 }
1569 1569
1570 static void update_shares_locked(struct rq *rq, struct sched_domain *sd) 1570 static void update_shares_locked(struct rq *rq, struct sched_domain *sd)
1571 { 1571 {
1572 spin_unlock(&rq->lock); 1572 spin_unlock(&rq->lock);
1573 update_shares(sd); 1573 update_shares(sd);
1574 spin_lock(&rq->lock); 1574 spin_lock(&rq->lock);
1575 } 1575 }
1576 1576
1577 static void update_h_load(long cpu) 1577 static void update_h_load(long cpu)
1578 { 1578 {
1579 walk_tg_tree(tg_load_down, tg_nop, (void *)cpu); 1579 walk_tg_tree(tg_load_down, tg_nop, (void *)cpu);
1580 } 1580 }
1581 1581
1582 #else 1582 #else
1583 1583
1584 static inline void update_shares(struct sched_domain *sd) 1584 static inline void update_shares(struct sched_domain *sd)
1585 { 1585 {
1586 } 1586 }
1587 1587
1588 static inline void update_shares_locked(struct rq *rq, struct sched_domain *sd) 1588 static inline void update_shares_locked(struct rq *rq, struct sched_domain *sd)
1589 { 1589 {
1590 } 1590 }
1591 1591
1592 #endif 1592 #endif
1593 1593
1594 /* 1594 /*
1595 * double_lock_balance - lock the busiest runqueue, this_rq is locked already. 1595 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
1596 */ 1596 */
1597 static int double_lock_balance(struct rq *this_rq, struct rq *busiest) 1597 static int double_lock_balance(struct rq *this_rq, struct rq *busiest)
1598 __releases(this_rq->lock) 1598 __releases(this_rq->lock)
1599 __acquires(busiest->lock) 1599 __acquires(busiest->lock)
1600 __acquires(this_rq->lock) 1600 __acquires(this_rq->lock)
1601 { 1601 {
1602 int ret = 0; 1602 int ret = 0;
1603 1603
1604 if (unlikely(!irqs_disabled())) { 1604 if (unlikely(!irqs_disabled())) {
1605 /* printk() doesn't work good under rq->lock */ 1605 /* printk() doesn't work good under rq->lock */
1606 spin_unlock(&this_rq->lock); 1606 spin_unlock(&this_rq->lock);
1607 BUG_ON(1); 1607 BUG_ON(1);
1608 } 1608 }
1609 if (unlikely(!spin_trylock(&busiest->lock))) { 1609 if (unlikely(!spin_trylock(&busiest->lock))) {
1610 if (busiest < this_rq) { 1610 if (busiest < this_rq) {
1611 spin_unlock(&this_rq->lock); 1611 spin_unlock(&this_rq->lock);
1612 spin_lock(&busiest->lock); 1612 spin_lock(&busiest->lock);
1613 spin_lock_nested(&this_rq->lock, SINGLE_DEPTH_NESTING); 1613 spin_lock_nested(&this_rq->lock, SINGLE_DEPTH_NESTING);
1614 ret = 1; 1614 ret = 1;
1615 } else 1615 } else
1616 spin_lock_nested(&busiest->lock, SINGLE_DEPTH_NESTING); 1616 spin_lock_nested(&busiest->lock, SINGLE_DEPTH_NESTING);
1617 } 1617 }
1618 return ret; 1618 return ret;
1619 } 1619 }
1620 1620
1621 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest) 1621 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
1622 __releases(busiest->lock) 1622 __releases(busiest->lock)
1623 { 1623 {
1624 spin_unlock(&busiest->lock); 1624 spin_unlock(&busiest->lock);
1625 lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); 1625 lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_);
1626 } 1626 }
1627 #endif 1627 #endif
1628 1628
1629 #ifdef CONFIG_FAIR_GROUP_SCHED 1629 #ifdef CONFIG_FAIR_GROUP_SCHED
1630 static void cfs_rq_set_shares(struct cfs_rq *cfs_rq, unsigned long shares) 1630 static void cfs_rq_set_shares(struct cfs_rq *cfs_rq, unsigned long shares)
1631 { 1631 {
1632 #ifdef CONFIG_SMP 1632 #ifdef CONFIG_SMP
1633 cfs_rq->shares = shares; 1633 cfs_rq->shares = shares;
1634 #endif 1634 #endif
1635 } 1635 }
1636 #endif 1636 #endif
1637 1637
1638 #include "sched_stats.h" 1638 #include "sched_stats.h"
1639 #include "sched_idletask.c" 1639 #include "sched_idletask.c"
1640 #include "sched_fair.c" 1640 #include "sched_fair.c"
1641 #include "sched_rt.c" 1641 #include "sched_rt.c"
1642 #ifdef CONFIG_SCHED_DEBUG 1642 #ifdef CONFIG_SCHED_DEBUG
1643 # include "sched_debug.c" 1643 # include "sched_debug.c"
1644 #endif 1644 #endif
1645 1645
1646 #define sched_class_highest (&rt_sched_class) 1646 #define sched_class_highest (&rt_sched_class)
1647 #define for_each_class(class) \ 1647 #define for_each_class(class) \
1648 for (class = sched_class_highest; class; class = class->next) 1648 for (class = sched_class_highest; class; class = class->next)
1649 1649
1650 static void inc_nr_running(struct rq *rq) 1650 static void inc_nr_running(struct rq *rq)
1651 { 1651 {
1652 rq->nr_running++; 1652 rq->nr_running++;
1653 } 1653 }
1654 1654
1655 static void dec_nr_running(struct rq *rq) 1655 static void dec_nr_running(struct rq *rq)
1656 { 1656 {
1657 rq->nr_running--; 1657 rq->nr_running--;
1658 } 1658 }
1659 1659
1660 static void set_load_weight(struct task_struct *p) 1660 static void set_load_weight(struct task_struct *p)
1661 { 1661 {
1662 if (task_has_rt_policy(p)) { 1662 if (task_has_rt_policy(p)) {
1663 p->se.load.weight = prio_to_weight[0] * 2; 1663 p->se.load.weight = prio_to_weight[0] * 2;
1664 p->se.load.inv_weight = prio_to_wmult[0] >> 1; 1664 p->se.load.inv_weight = prio_to_wmult[0] >> 1;
1665 return; 1665 return;
1666 } 1666 }
1667 1667
1668 /* 1668 /*
1669 * SCHED_IDLE tasks get minimal weight: 1669 * SCHED_IDLE tasks get minimal weight:
1670 */ 1670 */
1671 if (p->policy == SCHED_IDLE) { 1671 if (p->policy == SCHED_IDLE) {
1672 p->se.load.weight = WEIGHT_IDLEPRIO; 1672 p->se.load.weight = WEIGHT_IDLEPRIO;
1673 p->se.load.inv_weight = WMULT_IDLEPRIO; 1673 p->se.load.inv_weight = WMULT_IDLEPRIO;
1674 return; 1674 return;
1675 } 1675 }
1676 1676
1677 p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; 1677 p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO];
1678 p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; 1678 p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO];
1679 } 1679 }
1680 1680
1681 static void update_avg(u64 *avg, u64 sample) 1681 static void update_avg(u64 *avg, u64 sample)
1682 { 1682 {
1683 s64 diff = sample - *avg; 1683 s64 diff = sample - *avg;
1684 *avg += diff >> 3; 1684 *avg += diff >> 3;
1685 } 1685 }
1686 1686
1687 static void enqueue_task(struct rq *rq, struct task_struct *p, int wakeup) 1687 static void enqueue_task(struct rq *rq, struct task_struct *p, int wakeup)
1688 { 1688 {
1689 sched_info_queued(p); 1689 sched_info_queued(p);
1690 p->sched_class->enqueue_task(rq, p, wakeup); 1690 p->sched_class->enqueue_task(rq, p, wakeup);
1691 p->se.on_rq = 1; 1691 p->se.on_rq = 1;
1692 } 1692 }
1693 1693
1694 static void dequeue_task(struct rq *rq, struct task_struct *p, int sleep) 1694 static void dequeue_task(struct rq *rq, struct task_struct *p, int sleep)
1695 { 1695 {
1696 if (sleep && p->se.last_wakeup) { 1696 if (sleep && p->se.last_wakeup) {
1697 update_avg(&p->se.avg_overlap, 1697 update_avg(&p->se.avg_overlap,
1698 p->se.sum_exec_runtime - p->se.last_wakeup); 1698 p->se.sum_exec_runtime - p->se.last_wakeup);
1699 p->se.last_wakeup = 0; 1699 p->se.last_wakeup = 0;
1700 } 1700 }
1701 1701
1702 sched_info_dequeued(p); 1702 sched_info_dequeued(p);
1703 p->sched_class->dequeue_task(rq, p, sleep); 1703 p->sched_class->dequeue_task(rq, p, sleep);
1704 p->se.on_rq = 0; 1704 p->se.on_rq = 0;
1705 } 1705 }
1706 1706
1707 /* 1707 /*
1708 * __normal_prio - return the priority that is based on the static prio 1708 * __normal_prio - return the priority that is based on the static prio
1709 */ 1709 */
1710 static inline int __normal_prio(struct task_struct *p) 1710 static inline int __normal_prio(struct task_struct *p)
1711 { 1711 {
1712 return p->static_prio; 1712 return p->static_prio;
1713 } 1713 }
1714 1714
1715 /* 1715 /*
1716 * Calculate the expected normal priority: i.e. priority 1716 * Calculate the expected normal priority: i.e. priority
1717 * without taking RT-inheritance into account. Might be 1717 * without taking RT-inheritance into account. Might be
1718 * boosted by interactivity modifiers. Changes upon fork, 1718 * boosted by interactivity modifiers. Changes upon fork,
1719 * setprio syscalls, and whenever the interactivity 1719 * setprio syscalls, and whenever the interactivity
1720 * estimator recalculates. 1720 * estimator recalculates.
1721 */ 1721 */
1722 static inline int normal_prio(struct task_struct *p) 1722 static inline int normal_prio(struct task_struct *p)
1723 { 1723 {
1724 int prio; 1724 int prio;
1725 1725
1726 if (task_has_rt_policy(p)) 1726 if (task_has_rt_policy(p))
1727 prio = MAX_RT_PRIO-1 - p->rt_priority; 1727 prio = MAX_RT_PRIO-1 - p->rt_priority;
1728 else 1728 else
1729 prio = __normal_prio(p); 1729 prio = __normal_prio(p);
1730 return prio; 1730 return prio;
1731 } 1731 }
1732 1732
1733 /* 1733 /*
1734 * Calculate the current priority, i.e. the priority 1734 * Calculate the current priority, i.e. the priority
1735 * taken into account by the scheduler. This value might 1735 * taken into account by the scheduler. This value might
1736 * be boosted by RT tasks, or might be boosted by 1736 * be boosted by RT tasks, or might be boosted by
1737 * interactivity modifiers. Will be RT if the task got 1737 * interactivity modifiers. Will be RT if the task got
1738 * RT-boosted. If not then it returns p->normal_prio. 1738 * RT-boosted. If not then it returns p->normal_prio.
1739 */ 1739 */
1740 static int effective_prio(struct task_struct *p) 1740 static int effective_prio(struct task_struct *p)
1741 { 1741 {
1742 p->normal_prio = normal_prio(p); 1742 p->normal_prio = normal_prio(p);
1743 /* 1743 /*
1744 * If we are RT tasks or we were boosted to RT priority, 1744 * If we are RT tasks or we were boosted to RT priority,
1745 * keep the priority unchanged. Otherwise, update priority 1745 * keep the priority unchanged. Otherwise, update priority
1746 * to the normal priority: 1746 * to the normal priority:
1747 */ 1747 */
1748 if (!rt_prio(p->prio)) 1748 if (!rt_prio(p->prio))
1749 return p->normal_prio; 1749 return p->normal_prio;
1750 return p->prio; 1750 return p->prio;
1751 } 1751 }
1752 1752
1753 /* 1753 /*
1754 * activate_task - move a task to the runqueue. 1754 * activate_task - move a task to the runqueue.
1755 */ 1755 */
1756 static void activate_task(struct rq *rq, struct task_struct *p, int wakeup) 1756 static void activate_task(struct rq *rq, struct task_struct *p, int wakeup)
1757 { 1757 {
1758 if (task_contributes_to_load(p)) 1758 if (task_contributes_to_load(p))
1759 rq->nr_uninterruptible--; 1759 rq->nr_uninterruptible--;
1760 1760
1761 enqueue_task(rq, p, wakeup); 1761 enqueue_task(rq, p, wakeup);
1762 inc_nr_running(rq); 1762 inc_nr_running(rq);
1763 } 1763 }
1764 1764
1765 /* 1765 /*
1766 * deactivate_task - remove a task from the runqueue. 1766 * deactivate_task - remove a task from the runqueue.
1767 */ 1767 */
1768 static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep) 1768 static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep)
1769 { 1769 {
1770 if (task_contributes_to_load(p)) 1770 if (task_contributes_to_load(p))
1771 rq->nr_uninterruptible++; 1771 rq->nr_uninterruptible++;
1772 1772
1773 dequeue_task(rq, p, sleep); 1773 dequeue_task(rq, p, sleep);
1774 dec_nr_running(rq); 1774 dec_nr_running(rq);
1775 } 1775 }
1776 1776
1777 /** 1777 /**
1778 * task_curr - is this task currently executing on a CPU? 1778 * task_curr - is this task currently executing on a CPU?
1779 * @p: the task in question. 1779 * @p: the task in question.
1780 */ 1780 */
1781 inline int task_curr(const struct task_struct *p) 1781 inline int task_curr(const struct task_struct *p)
1782 { 1782 {
1783 return cpu_curr(task_cpu(p)) == p; 1783 return cpu_curr(task_cpu(p)) == p;
1784 } 1784 }
1785 1785
1786 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) 1786 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
1787 { 1787 {
1788 set_task_rq(p, cpu); 1788 set_task_rq(p, cpu);
1789 #ifdef CONFIG_SMP 1789 #ifdef CONFIG_SMP
1790 /* 1790 /*
1791 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be 1791 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1792 * successfuly executed on another CPU. We must ensure that updates of 1792 * successfuly executed on another CPU. We must ensure that updates of
1793 * per-task data have been completed by this moment. 1793 * per-task data have been completed by this moment.
1794 */ 1794 */
1795 smp_wmb(); 1795 smp_wmb();
1796 task_thread_info(p)->cpu = cpu; 1796 task_thread_info(p)->cpu = cpu;
1797 #endif 1797 #endif
1798 } 1798 }
1799 1799
1800 static inline void check_class_changed(struct rq *rq, struct task_struct *p, 1800 static inline void check_class_changed(struct rq *rq, struct task_struct *p,
1801 const struct sched_class *prev_class, 1801 const struct sched_class *prev_class,
1802 int oldprio, int running) 1802 int oldprio, int running)
1803 { 1803 {
1804 if (prev_class != p->sched_class) { 1804 if (prev_class != p->sched_class) {
1805 if (prev_class->switched_from) 1805 if (prev_class->switched_from)
1806 prev_class->switched_from(rq, p, running); 1806 prev_class->switched_from(rq, p, running);
1807 p->sched_class->switched_to(rq, p, running); 1807 p->sched_class->switched_to(rq, p, running);
1808 } else 1808 } else
1809 p->sched_class->prio_changed(rq, p, oldprio, running); 1809 p->sched_class->prio_changed(rq, p, oldprio, running);
1810 } 1810 }
1811 1811
1812 #ifdef CONFIG_SMP 1812 #ifdef CONFIG_SMP
1813 1813
1814 /* Used instead of source_load when we know the type == 0 */ 1814 /* Used instead of source_load when we know the type == 0 */
1815 static unsigned long weighted_cpuload(const int cpu) 1815 static unsigned long weighted_cpuload(const int cpu)
1816 { 1816 {
1817 return cpu_rq(cpu)->load.weight; 1817 return cpu_rq(cpu)->load.weight;
1818 } 1818 }
1819 1819
1820 /* 1820 /*
1821 * Is this task likely cache-hot: 1821 * Is this task likely cache-hot:
1822 */ 1822 */
1823 static int 1823 static int
1824 task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) 1824 task_hot(struct task_struct *p, u64 now, struct sched_domain *sd)
1825 { 1825 {
1826 s64 delta; 1826 s64 delta;
1827 1827
1828 /* 1828 /*
1829 * Buddy candidates are cache hot: 1829 * Buddy candidates are cache hot:
1830 */ 1830 */
1831 if (sched_feat(CACHE_HOT_BUDDY) && 1831 if (sched_feat(CACHE_HOT_BUDDY) &&
1832 (&p->se == cfs_rq_of(&p->se)->next || 1832 (&p->se == cfs_rq_of(&p->se)->next ||
1833 &p->se == cfs_rq_of(&p->se)->last)) 1833 &p->se == cfs_rq_of(&p->se)->last))
1834 return 1; 1834 return 1;
1835 1835
1836 if (p->sched_class != &fair_sched_class) 1836 if (p->sched_class != &fair_sched_class)
1837 return 0; 1837 return 0;
1838 1838
1839 if (sysctl_sched_migration_cost == -1) 1839 if (sysctl_sched_migration_cost == -1)
1840 return 1; 1840 return 1;
1841 if (sysctl_sched_migration_cost == 0) 1841 if (sysctl_sched_migration_cost == 0)
1842 return 0; 1842 return 0;
1843 1843
1844 delta = now - p->se.exec_start; 1844 delta = now - p->se.exec_start;
1845 1845
1846 return delta < (s64)sysctl_sched_migration_cost; 1846 return delta < (s64)sysctl_sched_migration_cost;
1847 } 1847 }
1848 1848
1849 1849
1850 void set_task_cpu(struct task_struct *p, unsigned int new_cpu) 1850 void set_task_cpu(struct task_struct *p, unsigned int new_cpu)
1851 { 1851 {
1852 int old_cpu = task_cpu(p); 1852 int old_cpu = task_cpu(p);
1853 struct rq *old_rq = cpu_rq(old_cpu), *new_rq = cpu_rq(new_cpu); 1853 struct rq *old_rq = cpu_rq(old_cpu), *new_rq = cpu_rq(new_cpu);
1854 struct cfs_rq *old_cfsrq = task_cfs_rq(p), 1854 struct cfs_rq *old_cfsrq = task_cfs_rq(p),
1855 *new_cfsrq = cpu_cfs_rq(old_cfsrq, new_cpu); 1855 *new_cfsrq = cpu_cfs_rq(old_cfsrq, new_cpu);
1856 u64 clock_offset; 1856 u64 clock_offset;
1857 1857
1858 clock_offset = old_rq->clock - new_rq->clock; 1858 clock_offset = old_rq->clock - new_rq->clock;
1859 1859
1860 #ifdef CONFIG_SCHEDSTATS 1860 #ifdef CONFIG_SCHEDSTATS
1861 if (p->se.wait_start) 1861 if (p->se.wait_start)
1862 p->se.wait_start -= clock_offset; 1862 p->se.wait_start -= clock_offset;
1863 if (p->se.sleep_start) 1863 if (p->se.sleep_start)
1864 p->se.sleep_start -= clock_offset; 1864 p->se.sleep_start -= clock_offset;
1865 if (p->se.block_start) 1865 if (p->se.block_start)
1866 p->se.block_start -= clock_offset; 1866 p->se.block_start -= clock_offset;
1867 if (old_cpu != new_cpu) { 1867 if (old_cpu != new_cpu) {
1868 schedstat_inc(p, se.nr_migrations); 1868 schedstat_inc(p, se.nr_migrations);
1869 if (task_hot(p, old_rq->clock, NULL)) 1869 if (task_hot(p, old_rq->clock, NULL))
1870 schedstat_inc(p, se.nr_forced2_migrations); 1870 schedstat_inc(p, se.nr_forced2_migrations);
1871 } 1871 }
1872 #endif 1872 #endif
1873 p->se.vruntime -= old_cfsrq->min_vruntime - 1873 p->se.vruntime -= old_cfsrq->min_vruntime -
1874 new_cfsrq->min_vruntime; 1874 new_cfsrq->min_vruntime;
1875 1875
1876 __set_task_cpu(p, new_cpu); 1876 __set_task_cpu(p, new_cpu);
1877 } 1877 }
1878 1878
1879 struct migration_req { 1879 struct migration_req {
1880 struct list_head list; 1880 struct list_head list;
1881 1881
1882 struct task_struct *task; 1882 struct task_struct *task;
1883 int dest_cpu; 1883 int dest_cpu;
1884 1884
1885 struct completion done; 1885 struct completion done;
1886 }; 1886 };
1887 1887
1888 /* 1888 /*
1889 * The task's runqueue lock must be held. 1889 * The task's runqueue lock must be held.
1890 * Returns true if you have to wait for migration thread. 1890 * Returns true if you have to wait for migration thread.
1891 */ 1891 */
1892 static int 1892 static int
1893 migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req) 1893 migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req)
1894 { 1894 {
1895 struct rq *rq = task_rq(p); 1895 struct rq *rq = task_rq(p);
1896 1896
1897 /* 1897 /*
1898 * If the task is not on a runqueue (and not running), then 1898 * If the task is not on a runqueue (and not running), then
1899 * it is sufficient to simply update the task's cpu field. 1899 * it is sufficient to simply update the task's cpu field.
1900 */ 1900 */
1901 if (!p->se.on_rq && !task_running(rq, p)) { 1901 if (!p->se.on_rq && !task_running(rq, p)) {
1902 set_task_cpu(p, dest_cpu); 1902 set_task_cpu(p, dest_cpu);
1903 return 0; 1903 return 0;
1904 } 1904 }
1905 1905
1906 init_completion(&req->done); 1906 init_completion(&req->done);
1907 req->task = p; 1907 req->task = p;
1908 req->dest_cpu = dest_cpu; 1908 req->dest_cpu = dest_cpu;
1909 list_add(&req->list, &rq->migration_queue); 1909 list_add(&req->list, &rq->migration_queue);
1910 1910
1911 return 1; 1911 return 1;
1912 } 1912 }
1913 1913
1914 /* 1914 /*
1915 * wait_task_inactive - wait for a thread to unschedule. 1915 * wait_task_inactive - wait for a thread to unschedule.
1916 * 1916 *
1917 * If @match_state is nonzero, it's the @p->state value just checked and 1917 * If @match_state is nonzero, it's the @p->state value just checked and
1918 * not expected to change. If it changes, i.e. @p might have woken up, 1918 * not expected to change. If it changes, i.e. @p might have woken up,
1919 * then return zero. When we succeed in waiting for @p to be off its CPU, 1919 * then return zero. When we succeed in waiting for @p to be off its CPU,
1920 * we return a positive number (its total switch count). If a second call 1920 * we return a positive number (its total switch count). If a second call
1921 * a short while later returns the same number, the caller can be sure that 1921 * a short while later returns the same number, the caller can be sure that
1922 * @p has remained unscheduled the whole time. 1922 * @p has remained unscheduled the whole time.
1923 * 1923 *
1924 * The caller must ensure that the task *will* unschedule sometime soon, 1924 * The caller must ensure that the task *will* unschedule sometime soon,
1925 * else this function might spin for a *long* time. This function can't 1925 * else this function might spin for a *long* time. This function can't
1926 * be called with interrupts off, or it may introduce deadlock with 1926 * be called with interrupts off, or it may introduce deadlock with
1927 * smp_call_function() if an IPI is sent by the same process we are 1927 * smp_call_function() if an IPI is sent by the same process we are
1928 * waiting to become inactive. 1928 * waiting to become inactive.
1929 */ 1929 */
1930 unsigned long wait_task_inactive(struct task_struct *p, long match_state) 1930 unsigned long wait_task_inactive(struct task_struct *p, long match_state)
1931 { 1931 {
1932 unsigned long flags; 1932 unsigned long flags;
1933 int running, on_rq; 1933 int running, on_rq;
1934 unsigned long ncsw; 1934 unsigned long ncsw;
1935 struct rq *rq; 1935 struct rq *rq;
1936 1936
1937 for (;;) { 1937 for (;;) {
1938 /* 1938 /*
1939 * We do the initial early heuristics without holding 1939 * We do the initial early heuristics without holding
1940 * any task-queue locks at all. We'll only try to get 1940 * any task-queue locks at all. We'll only try to get
1941 * the runqueue lock when things look like they will 1941 * the runqueue lock when things look like they will
1942 * work out! 1942 * work out!
1943 */ 1943 */
1944 rq = task_rq(p); 1944 rq = task_rq(p);
1945 1945
1946 /* 1946 /*
1947 * If the task is actively running on another CPU 1947 * If the task is actively running on another CPU
1948 * still, just relax and busy-wait without holding 1948 * still, just relax and busy-wait without holding
1949 * any locks. 1949 * any locks.
1950 * 1950 *
1951 * NOTE! Since we don't hold any locks, it's not 1951 * NOTE! Since we don't hold any locks, it's not
1952 * even sure that "rq" stays as the right runqueue! 1952 * even sure that "rq" stays as the right runqueue!
1953 * But we don't care, since "task_running()" will 1953 * But we don't care, since "task_running()" will
1954 * return false if the runqueue has changed and p 1954 * return false if the runqueue has changed and p
1955 * is actually now running somewhere else! 1955 * is actually now running somewhere else!
1956 */ 1956 */
1957 while (task_running(rq, p)) { 1957 while (task_running(rq, p)) {
1958 if (match_state && unlikely(p->state != match_state)) 1958 if (match_state && unlikely(p->state != match_state))
1959 return 0; 1959 return 0;
1960 cpu_relax(); 1960 cpu_relax();
1961 } 1961 }
1962 1962
1963 /* 1963 /*
1964 * Ok, time to look more closely! We need the rq 1964 * Ok, time to look more closely! We need the rq
1965 * lock now, to be *sure*. If we're wrong, we'll 1965 * lock now, to be *sure*. If we're wrong, we'll
1966 * just go back and repeat. 1966 * just go back and repeat.
1967 */ 1967 */
1968 rq = task_rq_lock(p, &flags); 1968 rq = task_rq_lock(p, &flags);
1969 trace_sched_wait_task(rq, p); 1969 trace_sched_wait_task(rq, p);
1970 running = task_running(rq, p); 1970 running = task_running(rq, p);
1971 on_rq = p->se.on_rq; 1971 on_rq = p->se.on_rq;
1972 ncsw = 0; 1972 ncsw = 0;
1973 if (!match_state || p->state == match_state) 1973 if (!match_state || p->state == match_state)
1974 ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ 1974 ncsw = p->nvcsw | LONG_MIN; /* sets MSB */
1975 task_rq_unlock(rq, &flags); 1975 task_rq_unlock(rq, &flags);
1976 1976
1977 /* 1977 /*
1978 * If it changed from the expected state, bail out now. 1978 * If it changed from the expected state, bail out now.
1979 */ 1979 */
1980 if (unlikely(!ncsw)) 1980 if (unlikely(!ncsw))
1981 break; 1981 break;
1982 1982
1983 /* 1983 /*
1984 * Was it really running after all now that we 1984 * Was it really running after all now that we
1985 * checked with the proper locks actually held? 1985 * checked with the proper locks actually held?
1986 * 1986 *
1987 * Oops. Go back and try again.. 1987 * Oops. Go back and try again..
1988 */ 1988 */
1989 if (unlikely(running)) { 1989 if (unlikely(running)) {
1990 cpu_relax(); 1990 cpu_relax();
1991 continue; 1991 continue;
1992 } 1992 }
1993 1993
1994 /* 1994 /*
1995 * It's not enough that it's not actively running, 1995 * It's not enough that it's not actively running,
1996 * it must be off the runqueue _entirely_, and not 1996 * it must be off the runqueue _entirely_, and not
1997 * preempted! 1997 * preempted!
1998 * 1998 *
1999 * So if it wa still runnable (but just not actively 1999 * So if it wa still runnable (but just not actively
2000 * running right now), it's preempted, and we should 2000 * running right now), it's preempted, and we should
2001 * yield - it could be a while. 2001 * yield - it could be a while.
2002 */ 2002 */
2003 if (unlikely(on_rq)) { 2003 if (unlikely(on_rq)) {
2004 schedule_timeout_uninterruptible(1); 2004 schedule_timeout_uninterruptible(1);
2005 continue; 2005 continue;
2006 } 2006 }
2007 2007
2008 /* 2008 /*
2009 * Ahh, all good. It wasn't running, and it wasn't 2009 * Ahh, all good. It wasn't running, and it wasn't
2010 * runnable, which means that it will never become 2010 * runnable, which means that it will never become
2011 * running in the future either. We're all done! 2011 * running in the future either. We're all done!
2012 */ 2012 */
2013 break; 2013 break;
2014 } 2014 }
2015 2015
2016 return ncsw; 2016 return ncsw;
2017 } 2017 }
2018 2018
2019 /*** 2019 /***
2020 * kick_process - kick a running thread to enter/exit the kernel 2020 * kick_process - kick a running thread to enter/exit the kernel
2021 * @p: the to-be-kicked thread 2021 * @p: the to-be-kicked thread
2022 * 2022 *
2023 * Cause a process which is running on another CPU to enter 2023 * Cause a process which is running on another CPU to enter
2024 * kernel-mode, without any delay. (to get signals handled.) 2024 * kernel-mode, without any delay. (to get signals handled.)
2025 * 2025 *
2026 * NOTE: this function doesnt have to take the runqueue lock, 2026 * NOTE: this function doesnt have to take the runqueue lock,
2027 * because all it wants to ensure is that the remote task enters 2027 * because all it wants to ensure is that the remote task enters
2028 * the kernel. If the IPI races and the task has been migrated 2028 * the kernel. If the IPI races and the task has been migrated
2029 * to another CPU then no harm is done and the purpose has been 2029 * to another CPU then no harm is done and the purpose has been
2030 * achieved as well. 2030 * achieved as well.
2031 */ 2031 */
2032 void kick_process(struct task_struct *p) 2032 void kick_process(struct task_struct *p)
2033 { 2033 {
2034 int cpu; 2034 int cpu;
2035 2035
2036 preempt_disable(); 2036 preempt_disable();
2037 cpu = task_cpu(p); 2037 cpu = task_cpu(p);
2038 if ((cpu != smp_processor_id()) && task_curr(p)) 2038 if ((cpu != smp_processor_id()) && task_curr(p))
2039 smp_send_reschedule(cpu); 2039 smp_send_reschedule(cpu);
2040 preempt_enable(); 2040 preempt_enable();
2041 } 2041 }
2042 2042
2043 /* 2043 /*
2044 * Return a low guess at the load of a migration-source cpu weighted 2044 * Return a low guess at the load of a migration-source cpu weighted
2045 * according to the scheduling class and "nice" value. 2045 * according to the scheduling class and "nice" value.
2046 * 2046 *
2047 * We want to under-estimate the load of migration sources, to 2047 * We want to under-estimate the load of migration sources, to
2048 * balance conservatively. 2048 * balance conservatively.
2049 */ 2049 */
2050 static unsigned long source_load(int cpu, int type) 2050 static unsigned long source_load(int cpu, int type)
2051 { 2051 {
2052 struct rq *rq = cpu_rq(cpu); 2052 struct rq *rq = cpu_rq(cpu);
2053 unsigned long total = weighted_cpuload(cpu); 2053 unsigned long total = weighted_cpuload(cpu);
2054 2054
2055 if (type == 0 || !sched_feat(LB_BIAS)) 2055 if (type == 0 || !sched_feat(LB_BIAS))
2056 return total; 2056 return total;
2057 2057
2058 return min(rq->cpu_load[type-1], total); 2058 return min(rq->cpu_load[type-1], total);
2059 } 2059 }
2060 2060
2061 /* 2061 /*
2062 * Return a high guess at the load of a migration-target cpu weighted 2062 * Return a high guess at the load of a migration-target cpu weighted
2063 * according to the scheduling class and "nice" value. 2063 * according to the scheduling class and "nice" value.
2064 */ 2064 */
2065 static unsigned long target_load(int cpu, int type) 2065 static unsigned long target_load(int cpu, int type)
2066 { 2066 {
2067 struct rq *rq = cpu_rq(cpu); 2067 struct rq *rq = cpu_rq(cpu);
2068 unsigned long total = weighted_cpuload(cpu); 2068 unsigned long total = weighted_cpuload(cpu);
2069 2069
2070 if (type == 0 || !sched_feat(LB_BIAS)) 2070 if (type == 0 || !sched_feat(LB_BIAS))
2071 return total; 2071 return total;
2072 2072
2073 return max(rq->cpu_load[type-1], total); 2073 return max(rq->cpu_load[type-1], total);
2074 } 2074 }
2075 2075
2076 /* 2076 /*
2077 * find_idlest_group finds and returns the least busy CPU group within the 2077 * find_idlest_group finds and returns the least busy CPU group within the
2078 * domain. 2078 * domain.
2079 */ 2079 */
2080 static struct sched_group * 2080 static struct sched_group *
2081 find_idlest_group(struct sched_domain *sd, struct task_struct *p, int this_cpu) 2081 find_idlest_group(struct sched_domain *sd, struct task_struct *p, int this_cpu)
2082 { 2082 {
2083 struct sched_group *idlest = NULL, *this = NULL, *group = sd->groups; 2083 struct sched_group *idlest = NULL, *this = NULL, *group = sd->groups;
2084 unsigned long min_load = ULONG_MAX, this_load = 0; 2084 unsigned long min_load = ULONG_MAX, this_load = 0;
2085 int load_idx = sd->forkexec_idx; 2085 int load_idx = sd->forkexec_idx;
2086 int imbalance = 100 + (sd->imbalance_pct-100)/2; 2086 int imbalance = 100 + (sd->imbalance_pct-100)/2;
2087 2087
2088 do { 2088 do {
2089 unsigned long load, avg_load; 2089 unsigned long load, avg_load;
2090 int local_group; 2090 int local_group;
2091 int i; 2091 int i;
2092 2092
2093 /* Skip over this group if it has no CPUs allowed */ 2093 /* Skip over this group if it has no CPUs allowed */
2094 if (!cpus_intersects(group->cpumask, p->cpus_allowed)) 2094 if (!cpus_intersects(group->cpumask, p->cpus_allowed))
2095 continue; 2095 continue;
2096 2096
2097 local_group = cpu_isset(this_cpu, group->cpumask); 2097 local_group = cpu_isset(this_cpu, group->cpumask);
2098 2098
2099 /* Tally up the load of all CPUs in the group */ 2099 /* Tally up the load of all CPUs in the group */
2100 avg_load = 0; 2100 avg_load = 0;
2101 2101
2102 for_each_cpu_mask_nr(i, group->cpumask) { 2102 for_each_cpu_mask_nr(i, group->cpumask) {
2103 /* Bias balancing toward cpus of our domain */ 2103 /* Bias balancing toward cpus of our domain */
2104 if (local_group) 2104 if (local_group)
2105 load = source_load(i, load_idx); 2105 load = source_load(i, load_idx);
2106 else 2106 else
2107 load = target_load(i, load_idx); 2107 load = target_load(i, load_idx);
2108 2108
2109 avg_load += load; 2109 avg_load += load;
2110 } 2110 }
2111 2111
2112 /* Adjust by relative CPU power of the group */ 2112 /* Adjust by relative CPU power of the group */
2113 avg_load = sg_div_cpu_power(group, 2113 avg_load = sg_div_cpu_power(group,
2114 avg_load * SCHED_LOAD_SCALE); 2114 avg_load * SCHED_LOAD_SCALE);
2115 2115
2116 if (local_group) { 2116 if (local_group) {
2117 this_load = avg_load; 2117 this_load = avg_load;
2118 this = group; 2118 this = group;
2119 } else if (avg_load < min_load) { 2119 } else if (avg_load < min_load) {
2120 min_load = avg_load; 2120 min_load = avg_load;
2121 idlest = group; 2121 idlest = group;
2122 } 2122 }
2123 } while (group = group->next, group != sd->groups); 2123 } while (group = group->next, group != sd->groups);
2124 2124
2125 if (!idlest || 100*this_load < imbalance*min_load) 2125 if (!idlest || 100*this_load < imbalance*min_load)
2126 return NULL; 2126 return NULL;
2127 return idlest; 2127 return idlest;
2128 } 2128 }
2129 2129
2130 /* 2130 /*
2131 * find_idlest_cpu - find the idlest cpu among the cpus in group. 2131 * find_idlest_cpu - find the idlest cpu among the cpus in group.
2132 */ 2132 */
2133 static int 2133 static int
2134 find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu, 2134 find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu,
2135 cpumask_t *tmp) 2135 cpumask_t *tmp)
2136 { 2136 {
2137 unsigned long load, min_load = ULONG_MAX; 2137 unsigned long load, min_load = ULONG_MAX;
2138 int idlest = -1; 2138 int idlest = -1;
2139 int i; 2139 int i;
2140 2140
2141 /* Traverse only the allowed CPUs */ 2141 /* Traverse only the allowed CPUs */
2142 cpus_and(*tmp, group->cpumask, p->cpus_allowed); 2142 cpus_and(*tmp, group->cpumask, p->cpus_allowed);
2143 2143
2144 for_each_cpu_mask_nr(i, *tmp) { 2144 for_each_cpu_mask_nr(i, *tmp) {
2145 load = weighted_cpuload(i); 2145 load = weighted_cpuload(i);
2146 2146
2147 if (load < min_load || (load == min_load && i == this_cpu)) { 2147 if (load < min_load || (load == min_load && i == this_cpu)) {
2148 min_load = load; 2148 min_load = load;
2149 idlest = i; 2149 idlest = i;
2150 } 2150 }
2151 } 2151 }
2152 2152
2153 return idlest; 2153 return idlest;
2154 } 2154 }
2155 2155
2156 /* 2156 /*
2157 * sched_balance_self: balance the current task (running on cpu) in domains 2157 * sched_balance_self: balance the current task (running on cpu) in domains
2158 * that have the 'flag' flag set. In practice, this is SD_BALANCE_FORK and 2158 * that have the 'flag' flag set. In practice, this is SD_BALANCE_FORK and
2159 * SD_BALANCE_EXEC. 2159 * SD_BALANCE_EXEC.
2160 * 2160 *
2161 * Balance, ie. select the least loaded group. 2161 * Balance, ie. select the least loaded group.
2162 * 2162 *
2163 * Returns the target CPU number, or the same CPU if no balancing is needed. 2163 * Returns the target CPU number, or the same CPU if no balancing is needed.
2164 * 2164 *
2165 * preempt must be disabled. 2165 * preempt must be disabled.
2166 */ 2166 */
2167 static int sched_balance_self(int cpu, int flag) 2167 static int sched_balance_self(int cpu, int flag)
2168 { 2168 {
2169 struct task_struct *t = current; 2169 struct task_struct *t = current;
2170 struct sched_domain *tmp, *sd = NULL; 2170 struct sched_domain *tmp, *sd = NULL;
2171 2171
2172 for_each_domain(cpu, tmp) { 2172 for_each_domain(cpu, tmp) {
2173 /* 2173 /*
2174 * If power savings logic is enabled for a domain, stop there. 2174 * If power savings logic is enabled for a domain, stop there.
2175 */ 2175 */
2176 if (tmp->flags & SD_POWERSAVINGS_BALANCE) 2176 if (tmp->flags & SD_POWERSAVINGS_BALANCE)
2177 break; 2177 break;
2178 if (tmp->flags & flag) 2178 if (tmp->flags & flag)
2179 sd = tmp; 2179 sd = tmp;
2180 } 2180 }
2181 2181
2182 if (sd) 2182 if (sd)
2183 update_shares(sd); 2183 update_shares(sd);
2184 2184
2185 while (sd) { 2185 while (sd) {
2186 cpumask_t span, tmpmask; 2186 cpumask_t span, tmpmask;
2187 struct sched_group *group; 2187 struct sched_group *group;
2188 int new_cpu, weight; 2188 int new_cpu, weight;
2189 2189
2190 if (!(sd->flags & flag)) { 2190 if (!(sd->flags & flag)) {
2191 sd = sd->child; 2191 sd = sd->child;
2192 continue; 2192 continue;
2193 } 2193 }
2194 2194
2195 span = sd->span; 2195 span = sd->span;
2196 group = find_idlest_group(sd, t, cpu); 2196 group = find_idlest_group(sd, t, cpu);
2197 if (!group) { 2197 if (!group) {
2198 sd = sd->child; 2198 sd = sd->child;
2199 continue; 2199 continue;
2200 } 2200 }
2201 2201
2202 new_cpu = find_idlest_cpu(group, t, cpu, &tmpmask); 2202 new_cpu = find_idlest_cpu(group, t, cpu, &tmpmask);
2203 if (new_cpu == -1 || new_cpu == cpu) { 2203 if (new_cpu == -1 || new_cpu == cpu) {
2204 /* Now try balancing at a lower domain level of cpu */ 2204 /* Now try balancing at a lower domain level of cpu */
2205 sd = sd->child; 2205 sd = sd->child;
2206 continue; 2206 continue;
2207 } 2207 }
2208 2208
2209 /* Now try balancing at a lower domain level of new_cpu */ 2209 /* Now try balancing at a lower domain level of new_cpu */
2210 cpu = new_cpu; 2210 cpu = new_cpu;
2211 sd = NULL; 2211 sd = NULL;
2212 weight = cpus_weight(span); 2212 weight = cpus_weight(span);
2213 for_each_domain(cpu, tmp) { 2213 for_each_domain(cpu, tmp) {
2214 if (weight <= cpus_weight(tmp->span)) 2214 if (weight <= cpus_weight(tmp->span))
2215 break; 2215 break;
2216 if (tmp->flags & flag) 2216 if (tmp->flags & flag)
2217 sd = tmp; 2217 sd = tmp;
2218 } 2218 }
2219 /* while loop will break here if sd == NULL */ 2219 /* while loop will break here if sd == NULL */
2220 } 2220 }
2221 2221
2222 return cpu; 2222 return cpu;
2223 } 2223 }
2224 2224
2225 #endif /* CONFIG_SMP */ 2225 #endif /* CONFIG_SMP */
2226 2226
2227 /*** 2227 /***
2228 * try_to_wake_up - wake up a thread 2228 * try_to_wake_up - wake up a thread
2229 * @p: the to-be-woken-up thread 2229 * @p: the to-be-woken-up thread
2230 * @state: the mask of task states that can be woken 2230 * @state: the mask of task states that can be woken
2231 * @sync: do a synchronous wakeup? 2231 * @sync: do a synchronous wakeup?
2232 * 2232 *
2233 * Put it on the run-queue if it's not already there. The "current" 2233 * Put it on the run-queue if it's not already there. The "current"
2234 * thread is always on the run-queue (except when the actual 2234 * thread is always on the run-queue (except when the actual
2235 * re-schedule is in progress), and as such you're allowed to do 2235 * re-schedule is in progress), and as such you're allowed to do
2236 * the simpler "current->state = TASK_RUNNING" to mark yourself 2236 * the simpler "current->state = TASK_RUNNING" to mark yourself
2237 * runnable without the overhead of this. 2237 * runnable without the overhead of this.
2238 * 2238 *
2239 * returns failure only if the task is already active. 2239 * returns failure only if the task is already active.
2240 */ 2240 */
2241 static int try_to_wake_up(struct task_struct *p, unsigned int state, int sync) 2241 static int try_to_wake_up(struct task_struct *p, unsigned int state, int sync)
2242 { 2242 {
2243 int cpu, orig_cpu, this_cpu, success = 0; 2243 int cpu, orig_cpu, this_cpu, success = 0;
2244 unsigned long flags; 2244 unsigned long flags;
2245 long old_state; 2245 long old_state;
2246 struct rq *rq; 2246 struct rq *rq;
2247 2247
2248 if (!sched_feat(SYNC_WAKEUPS)) 2248 if (!sched_feat(SYNC_WAKEUPS))
2249 sync = 0; 2249 sync = 0;
2250 2250
2251 #ifdef CONFIG_SMP 2251 #ifdef CONFIG_SMP
2252 if (sched_feat(LB_WAKEUP_UPDATE)) { 2252 if (sched_feat(LB_WAKEUP_UPDATE)) {
2253 struct sched_domain *sd; 2253 struct sched_domain *sd;
2254 2254
2255 this_cpu = raw_smp_processor_id(); 2255 this_cpu = raw_smp_processor_id();
2256 cpu = task_cpu(p); 2256 cpu = task_cpu(p);
2257 2257
2258 for_each_domain(this_cpu, sd) { 2258 for_each_domain(this_cpu, sd) {
2259 if (cpu_isset(cpu, sd->span)) { 2259 if (cpu_isset(cpu, sd->span)) {
2260 update_shares(sd); 2260 update_shares(sd);
2261 break; 2261 break;
2262 } 2262 }
2263 } 2263 }
2264 } 2264 }
2265 #endif 2265 #endif
2266 2266
2267 smp_wmb(); 2267 smp_wmb();
2268 rq = task_rq_lock(p, &flags); 2268 rq = task_rq_lock(p, &flags);
2269 old_state = p->state; 2269 old_state = p->state;
2270 if (!(old_state & state)) 2270 if (!(old_state & state))
2271 goto out; 2271 goto out;
2272 2272
2273 if (p->se.on_rq) 2273 if (p->se.on_rq)
2274 goto out_running; 2274 goto out_running;
2275 2275
2276 cpu = task_cpu(p); 2276 cpu = task_cpu(p);
2277 orig_cpu = cpu; 2277 orig_cpu = cpu;
2278 this_cpu = smp_processor_id(); 2278 this_cpu = smp_processor_id();
2279 2279
2280 #ifdef CONFIG_SMP 2280 #ifdef CONFIG_SMP
2281 if (unlikely(task_running(rq, p))) 2281 if (unlikely(task_running(rq, p)))
2282 goto out_activate; 2282 goto out_activate;
2283 2283
2284 cpu = p->sched_class->select_task_rq(p, sync); 2284 cpu = p->sched_class->select_task_rq(p, sync);
2285 if (cpu != orig_cpu) { 2285 if (cpu != orig_cpu) {
2286 set_task_cpu(p, cpu); 2286 set_task_cpu(p, cpu);
2287 task_rq_unlock(rq, &flags); 2287 task_rq_unlock(rq, &flags);
2288 /* might preempt at this point */ 2288 /* might preempt at this point */
2289 rq = task_rq_lock(p, &flags); 2289 rq = task_rq_lock(p, &flags);
2290 old_state = p->state; 2290 old_state = p->state;
2291 if (!(old_state & state)) 2291 if (!(old_state & state))
2292 goto out; 2292 goto out;
2293 if (p->se.on_rq) 2293 if (p->se.on_rq)
2294 goto out_running; 2294 goto out_running;
2295 2295
2296 this_cpu = smp_processor_id(); 2296 this_cpu = smp_processor_id();
2297 cpu = task_cpu(p); 2297 cpu = task_cpu(p);
2298 } 2298 }
2299 2299
2300 #ifdef CONFIG_SCHEDSTATS 2300 #ifdef CONFIG_SCHEDSTATS
2301 schedstat_inc(rq, ttwu_count); 2301 schedstat_inc(rq, ttwu_count);
2302 if (cpu == this_cpu) 2302 if (cpu == this_cpu)
2303 schedstat_inc(rq, ttwu_local); 2303 schedstat_inc(rq, ttwu_local);
2304 else { 2304 else {
2305 struct sched_domain *sd; 2305 struct sched_domain *sd;
2306 for_each_domain(this_cpu, sd) { 2306 for_each_domain(this_cpu, sd) {
2307 if (cpu_isset(cpu, sd->span)) { 2307 if (cpu_isset(cpu, sd->span)) {
2308 schedstat_inc(sd, ttwu_wake_remote); 2308 schedstat_inc(sd, ttwu_wake_remote);
2309 break; 2309 break;
2310 } 2310 }
2311 } 2311 }
2312 } 2312 }
2313 #endif /* CONFIG_SCHEDSTATS */ 2313 #endif /* CONFIG_SCHEDSTATS */
2314 2314
2315 out_activate: 2315 out_activate:
2316 #endif /* CONFIG_SMP */ 2316 #endif /* CONFIG_SMP */
2317 schedstat_inc(p, se.nr_wakeups); 2317 schedstat_inc(p, se.nr_wakeups);
2318 if (sync) 2318 if (sync)
2319 schedstat_inc(p, se.nr_wakeups_sync); 2319 schedstat_inc(p, se.nr_wakeups_sync);
2320 if (orig_cpu != cpu) 2320 if (orig_cpu != cpu)
2321 schedstat_inc(p, se.nr_wakeups_migrate); 2321 schedstat_inc(p, se.nr_wakeups_migrate);
2322 if (cpu == this_cpu) 2322 if (cpu == this_cpu)
2323 schedstat_inc(p, se.nr_wakeups_local); 2323 schedstat_inc(p, se.nr_wakeups_local);
2324 else 2324 else
2325 schedstat_inc(p, se.nr_wakeups_remote); 2325 schedstat_inc(p, se.nr_wakeups_remote);
2326 update_rq_clock(rq); 2326 update_rq_clock(rq);
2327 activate_task(rq, p, 1); 2327 activate_task(rq, p, 1);
2328 success = 1; 2328 success = 1;
2329 2329
2330 out_running: 2330 out_running:
2331 trace_sched_wakeup(rq, p); 2331 trace_sched_wakeup(rq, p);
2332 check_preempt_curr(rq, p, sync); 2332 check_preempt_curr(rq, p, sync);
2333 2333
2334 p->state = TASK_RUNNING; 2334 p->state = TASK_RUNNING;
2335 #ifdef CONFIG_SMP 2335 #ifdef CONFIG_SMP
2336 if (p->sched_class->task_wake_up) 2336 if (p->sched_class->task_wake_up)
2337 p->sched_class->task_wake_up(rq, p); 2337 p->sched_class->task_wake_up(rq, p);
2338 #endif 2338 #endif
2339 out: 2339 out:
2340 current->se.last_wakeup = current->se.sum_exec_runtime; 2340 current->se.last_wakeup = current->se.sum_exec_runtime;
2341 2341
2342 task_rq_unlock(rq, &flags); 2342 task_rq_unlock(rq, &flags);
2343 2343
2344 return success; 2344 return success;
2345 } 2345 }
2346 2346
2347 int wake_up_process(struct task_struct *p) 2347 int wake_up_process(struct task_struct *p)
2348 { 2348 {
2349 return try_to_wake_up(p, TASK_ALL, 0); 2349 return try_to_wake_up(p, TASK_ALL, 0);
2350 } 2350 }
2351 EXPORT_SYMBOL(wake_up_process); 2351 EXPORT_SYMBOL(wake_up_process);
2352 2352
2353 int wake_up_state(struct task_struct *p, unsigned int state) 2353 int wake_up_state(struct task_struct *p, unsigned int state)
2354 { 2354 {
2355 return try_to_wake_up(p, state, 0); 2355 return try_to_wake_up(p, state, 0);
2356 } 2356 }
2357 2357
2358 /* 2358 /*
2359 * Perform scheduler related setup for a newly forked process p. 2359 * Perform scheduler related setup for a newly forked process p.
2360 * p is forked by current. 2360 * p is forked by current.
2361 * 2361 *
2362 * __sched_fork() is basic setup used by init_idle() too: 2362 * __sched_fork() is basic setup used by init_idle() too:
2363 */ 2363 */
2364 static void __sched_fork(struct task_struct *p) 2364 static void __sched_fork(struct task_struct *p)
2365 { 2365 {
2366 p->se.exec_start = 0; 2366 p->se.exec_start = 0;
2367 p->se.sum_exec_runtime = 0; 2367 p->se.sum_exec_runtime = 0;
2368 p->se.prev_sum_exec_runtime = 0; 2368 p->se.prev_sum_exec_runtime = 0;
2369 p->se.last_wakeup = 0; 2369 p->se.last_wakeup = 0;
2370 p->se.avg_overlap = 0; 2370 p->se.avg_overlap = 0;
2371 2371
2372 #ifdef CONFIG_SCHEDSTATS 2372 #ifdef CONFIG_SCHEDSTATS
2373 p->se.wait_start = 0; 2373 p->se.wait_start = 0;
2374 p->se.sum_sleep_runtime = 0; 2374 p->se.sum_sleep_runtime = 0;
2375 p->se.sleep_start = 0; 2375 p->se.sleep_start = 0;
2376 p->se.block_start = 0; 2376 p->se.block_start = 0;
2377 p->se.sleep_max = 0; 2377 p->se.sleep_max = 0;
2378 p->se.block_max = 0; 2378 p->se.block_max = 0;
2379 p->se.exec_max = 0; 2379 p->se.exec_max = 0;
2380 p->se.slice_max = 0; 2380 p->se.slice_max = 0;
2381 p->se.wait_max = 0; 2381 p->se.wait_max = 0;
2382 #endif 2382 #endif
2383 2383
2384 INIT_LIST_HEAD(&p->rt.run_list); 2384 INIT_LIST_HEAD(&p->rt.run_list);
2385 p->se.on_rq = 0; 2385 p->se.on_rq = 0;
2386 INIT_LIST_HEAD(&p->se.group_node); 2386 INIT_LIST_HEAD(&p->se.group_node);
2387 2387
2388 #ifdef CONFIG_PREEMPT_NOTIFIERS 2388 #ifdef CONFIG_PREEMPT_NOTIFIERS
2389 INIT_HLIST_HEAD(&p->preempt_notifiers); 2389 INIT_HLIST_HEAD(&p->preempt_notifiers);
2390 #endif 2390 #endif
2391 2391
2392 /* 2392 /*
2393 * We mark the process as running here, but have not actually 2393 * We mark the process as running here, but have not actually
2394 * inserted it onto the runqueue yet. This guarantees that 2394 * inserted it onto the runqueue yet. This guarantees that
2395 * nobody will actually run it, and a signal or other external 2395 * nobody will actually run it, and a signal or other external
2396 * event cannot wake it up and insert it on the runqueue either. 2396 * event cannot wake it up and insert it on the runqueue either.
2397 */ 2397 */
2398 p->state = TASK_RUNNING; 2398 p->state = TASK_RUNNING;
2399 } 2399 }
2400 2400
2401 /* 2401 /*
2402 * fork()/clone()-time setup: 2402 * fork()/clone()-time setup:
2403 */ 2403 */
2404 void sched_fork(struct task_struct *p, int clone_flags) 2404 void sched_fork(struct task_struct *p, int clone_flags)
2405 { 2405 {
2406 int cpu = get_cpu(); 2406 int cpu = get_cpu();
2407 2407
2408 __sched_fork(p); 2408 __sched_fork(p);
2409 2409
2410 #ifdef CONFIG_SMP 2410 #ifdef CONFIG_SMP
2411 cpu = sched_balance_self(cpu, SD_BALANCE_FORK); 2411 cpu = sched_balance_self(cpu, SD_BALANCE_FORK);
2412 #endif 2412 #endif
2413 set_task_cpu(p, cpu); 2413 set_task_cpu(p, cpu);
2414 2414
2415 /* 2415 /*
2416 * Make sure we do not leak PI boosting priority to the child: 2416 * Make sure we do not leak PI boosting priority to the child:
2417 */ 2417 */
2418 p->prio = current->normal_prio; 2418 p->prio = current->normal_prio;
2419 if (!rt_prio(p->prio)) 2419 if (!rt_prio(p->prio))
2420 p->sched_class = &fair_sched_class; 2420 p->sched_class = &fair_sched_class;
2421 2421
2422 #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) 2422 #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT)
2423 if (likely(sched_info_on())) 2423 if (likely(sched_info_on()))
2424 memset(&p->sched_info, 0, sizeof(p->sched_info)); 2424 memset(&p->sched_info, 0, sizeof(p->sched_info));
2425 #endif 2425 #endif
2426 #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) 2426 #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW)
2427 p->oncpu = 0; 2427 p->oncpu = 0;
2428 #endif 2428 #endif
2429 #ifdef CONFIG_PREEMPT 2429 #ifdef CONFIG_PREEMPT
2430 /* Want to start with kernel preemption disabled. */ 2430 /* Want to start with kernel preemption disabled. */
2431 task_thread_info(p)->preempt_count = 1; 2431 task_thread_info(p)->preempt_count = 1;
2432 #endif 2432 #endif
2433 put_cpu(); 2433 put_cpu();
2434 } 2434 }
2435 2435
2436 /* 2436 /*
2437 * wake_up_new_task - wake up a newly created task for the first time. 2437 * wake_up_new_task - wake up a newly created task for the first time.
2438 * 2438 *
2439 * This function will do some initial scheduler statistics housekeeping 2439 * This function will do some initial scheduler statistics housekeeping
2440 * that must be done for every newly created context, then puts the task 2440 * that must be done for every newly created context, then puts the task
2441 * on the runqueue and wakes it. 2441 * on the runqueue and wakes it.
2442 */ 2442 */
2443 void wake_up_new_task(struct task_struct *p, unsigned long clone_flags) 2443 void wake_up_new_task(struct task_struct *p, unsigned long clone_flags)
2444 { 2444 {
2445 unsigned long flags; 2445 unsigned long flags;
2446 struct rq *rq; 2446 struct rq *rq;
2447 2447
2448 rq = task_rq_lock(p, &flags); 2448 rq = task_rq_lock(p, &flags);
2449 BUG_ON(p->state != TASK_RUNNING); 2449 BUG_ON(p->state != TASK_RUNNING);
2450 update_rq_clock(rq); 2450 update_rq_clock(rq);
2451 2451
2452 p->prio = effective_prio(p); 2452 p->prio = effective_prio(p);
2453 2453
2454 if (!p->sched_class->task_new || !current->se.on_rq) { 2454 if (!p->sched_class->task_new || !current->se.on_rq) {
2455 activate_task(rq, p, 0); 2455 activate_task(rq, p, 0);
2456 } else { 2456 } else {
2457 /* 2457 /*
2458 * Let the scheduling class do new task startup 2458 * Let the scheduling class do new task startup
2459 * management (if any): 2459 * management (if any):
2460 */ 2460 */
2461 p->sched_class->task_new(rq, p); 2461 p->sched_class->task_new(rq, p);
2462 inc_nr_running(rq); 2462 inc_nr_running(rq);
2463 } 2463 }
2464 trace_sched_wakeup_new(rq, p); 2464 trace_sched_wakeup_new(rq, p);
2465 check_preempt_curr(rq, p, 0); 2465 check_preempt_curr(rq, p, 0);
2466 #ifdef CONFIG_SMP 2466 #ifdef CONFIG_SMP
2467 if (p->sched_class->task_wake_up) 2467 if (p->sched_class->task_wake_up)
2468 p->sched_class->task_wake_up(rq, p); 2468 p->sched_class->task_wake_up(rq, p);
2469 #endif 2469 #endif
2470 task_rq_unlock(rq, &flags); 2470 task_rq_unlock(rq, &flags);
2471 } 2471 }
2472 2472
2473 #ifdef CONFIG_PREEMPT_NOTIFIERS 2473 #ifdef CONFIG_PREEMPT_NOTIFIERS
2474 2474
2475 /** 2475 /**
2476 * preempt_notifier_register - tell me when current is being being preempted & rescheduled 2476 * preempt_notifier_register - tell me when current is being being preempted & rescheduled
2477 * @notifier: notifier struct to register 2477 * @notifier: notifier struct to register
2478 */ 2478 */
2479 void preempt_notifier_register(struct preempt_notifier *notifier) 2479 void preempt_notifier_register(struct preempt_notifier *notifier)
2480 { 2480 {
2481 hlist_add_head(&notifier->link, &current->preempt_notifiers); 2481 hlist_add_head(&notifier->link, &current->preempt_notifiers);
2482 } 2482 }
2483 EXPORT_SYMBOL_GPL(preempt_notifier_register); 2483 EXPORT_SYMBOL_GPL(preempt_notifier_register);
2484 2484
2485 /** 2485 /**
2486 * preempt_notifier_unregister - no longer interested in preemption notifications 2486 * preempt_notifier_unregister - no longer interested in preemption notifications
2487 * @notifier: notifier struct to unregister 2487 * @notifier: notifier struct to unregister
2488 * 2488 *
2489 * This is safe to call from within a preemption notifier. 2489 * This is safe to call from within a preemption notifier.
2490 */ 2490 */
2491 void preempt_notifier_unregister(struct preempt_notifier *notifier) 2491 void preempt_notifier_unregister(struct preempt_notifier *notifier)
2492 { 2492 {
2493 hlist_del(&notifier->link); 2493 hlist_del(&notifier->link);
2494 } 2494 }
2495 EXPORT_SYMBOL_GPL(preempt_notifier_unregister); 2495 EXPORT_SYMBOL_GPL(preempt_notifier_unregister);
2496 2496
2497 static void fire_sched_in_preempt_notifiers(struct task_struct *curr) 2497 static void fire_sched_in_preempt_notifiers(struct task_struct *curr)
2498 { 2498 {
2499 struct preempt_notifier *notifier; 2499 struct preempt_notifier *notifier;
2500 struct hlist_node *node; 2500 struct hlist_node *node;
2501 2501
2502 hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) 2502 hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link)
2503 notifier->ops->sched_in(notifier, raw_smp_processor_id()); 2503 notifier->ops->sched_in(notifier, raw_smp_processor_id());
2504 } 2504 }
2505 2505
2506 static void 2506 static void
2507 fire_sched_out_preempt_notifiers(struct task_struct *curr, 2507 fire_sched_out_preempt_notifiers(struct task_struct *curr,
2508 struct task_struct *next) 2508 struct task_struct *next)
2509 { 2509 {
2510 struct preempt_notifier *notifier; 2510 struct preempt_notifier *notifier;
2511 struct hlist_node *node; 2511 struct hlist_node *node;
2512 2512
2513 hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) 2513 hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link)
2514 notifier->ops->sched_out(notifier, next); 2514 notifier->ops->sched_out(notifier, next);
2515 } 2515 }
2516 2516
2517 #else /* !CONFIG_PREEMPT_NOTIFIERS */ 2517 #else /* !CONFIG_PREEMPT_NOTIFIERS */
2518 2518
2519 static void fire_sched_in_preempt_notifiers(struct task_struct *curr) 2519 static void fire_sched_in_preempt_notifiers(struct task_struct *curr)
2520 { 2520 {
2521 } 2521 }
2522 2522
2523 static void 2523 static void
2524 fire_sched_out_preempt_notifiers(struct task_struct *curr, 2524 fire_sched_out_preempt_notifiers(struct task_struct *curr,
2525 struct task_struct *next) 2525 struct task_struct *next)
2526 { 2526 {
2527 } 2527 }
2528 2528
2529 #endif /* CONFIG_PREEMPT_NOTIFIERS */ 2529 #endif /* CONFIG_PREEMPT_NOTIFIERS */
2530 2530
2531 /** 2531 /**
2532 * prepare_task_switch - prepare to switch tasks 2532 * prepare_task_switch - prepare to switch tasks
2533 * @rq: the runqueue preparing to switch 2533 * @rq: the runqueue preparing to switch
2534 * @prev: the current task that is being switched out 2534 * @prev: the current task that is being switched out
2535 * @next: the task we are going to switch to. 2535 * @next: the task we are going to switch to.
2536 * 2536 *
2537 * This is called with the rq lock held and interrupts off. It must 2537 * This is called with the rq lock held and interrupts off. It must
2538 * be paired with a subsequent finish_task_switch after the context 2538 * be paired with a subsequent finish_task_switch after the context
2539 * switch. 2539 * switch.
2540 * 2540 *
2541 * prepare_task_switch sets up locking and calls architecture specific 2541 * prepare_task_switch sets up locking and calls architecture specific
2542 * hooks. 2542 * hooks.
2543 */ 2543 */
2544 static inline void 2544 static inline void
2545 prepare_task_switch(struct rq *rq, struct task_struct *prev, 2545 prepare_task_switch(struct rq *rq, struct task_struct *prev,
2546 struct task_struct *next) 2546 struct task_struct *next)
2547 { 2547 {
2548 fire_sched_out_preempt_notifiers(prev, next); 2548 fire_sched_out_preempt_notifiers(prev, next);
2549 prepare_lock_switch(rq, next); 2549 prepare_lock_switch(rq, next);
2550 prepare_arch_switch(next); 2550 prepare_arch_switch(next);
2551 } 2551 }
2552 2552
2553 /** 2553 /**
2554 * finish_task_switch - clean up after a task-switch 2554 * finish_task_switch - clean up after a task-switch
2555 * @rq: runqueue associated with task-switch 2555 * @rq: runqueue associated with task-switch
2556 * @prev: the thread we just switched away from. 2556 * @prev: the thread we just switched away from.
2557 * 2557 *
2558 * finish_task_switch must be called after the context switch, paired 2558 * finish_task_switch must be called after the context switch, paired
2559 * with a prepare_task_switch call before the context switch. 2559 * with a prepare_task_switch call before the context switch.
2560 * finish_task_switch will reconcile locking set up by prepare_task_switch, 2560 * finish_task_switch will reconcile locking set up by prepare_task_switch,
2561 * and do any other architecture-specific cleanup actions. 2561 * and do any other architecture-specific cleanup actions.
2562 * 2562 *
2563 * Note that we may have delayed dropping an mm in context_switch(). If 2563 * Note that we may have delayed dropping an mm in context_switch(). If
2564 * so, we finish that here outside of the runqueue lock. (Doing it 2564 * so, we finish that here outside of the runqueue lock. (Doing it
2565 * with the lock held can cause deadlocks; see schedule() for 2565 * with the lock held can cause deadlocks; see schedule() for
2566 * details.) 2566 * details.)
2567 */ 2567 */
2568 static void finish_task_switch(struct rq *rq, struct task_struct *prev) 2568 static void finish_task_switch(struct rq *rq, struct task_struct *prev)
2569 __releases(rq->lock) 2569 __releases(rq->lock)
2570 { 2570 {
2571 struct mm_struct *mm = rq->prev_mm; 2571 struct mm_struct *mm = rq->prev_mm;
2572 long prev_state; 2572 long prev_state;
2573 2573
2574 rq->prev_mm = NULL; 2574 rq->prev_mm = NULL;
2575 2575
2576 /* 2576 /*
2577 * A task struct has one reference for the use as "current". 2577 * A task struct has one reference for the use as "current".
2578 * If a task dies, then it sets TASK_DEAD in tsk->state and calls 2578 * If a task dies, then it sets TASK_DEAD in tsk->state and calls
2579 * schedule one last time. The schedule call will never return, and 2579 * schedule one last time. The schedule call will never return, and
2580 * the scheduled task must drop that reference. 2580 * the scheduled task must drop that reference.
2581 * The test for TASK_DEAD must occur while the runqueue locks are 2581 * The test for TASK_DEAD must occur while the runqueue locks are
2582 * still held, otherwise prev could be scheduled on another cpu, die 2582 * still held, otherwise prev could be scheduled on another cpu, die
2583 * there before we look at prev->state, and then the reference would 2583 * there before we look at prev->state, and then the reference would
2584 * be dropped twice. 2584 * be dropped twice.
2585 * Manfred Spraul <manfred@colorfullife.com> 2585 * Manfred Spraul <manfred@colorfullife.com>
2586 */ 2586 */
2587 prev_state = prev->state; 2587 prev_state = prev->state;
2588 finish_arch_switch(prev); 2588 finish_arch_switch(prev);
2589 finish_lock_switch(rq, prev); 2589 finish_lock_switch(rq, prev);
2590 #ifdef CONFIG_SMP 2590 #ifdef CONFIG_SMP
2591 if (current->sched_class->post_schedule) 2591 if (current->sched_class->post_schedule)
2592 current->sched_class->post_schedule(rq); 2592 current->sched_class->post_schedule(rq);
2593 #endif 2593 #endif
2594 2594
2595 fire_sched_in_preempt_notifiers(current); 2595 fire_sched_in_preempt_notifiers(current);
2596 if (mm) 2596 if (mm)
2597 mmdrop(mm); 2597 mmdrop(mm);
2598 if (unlikely(prev_state == TASK_DEAD)) { 2598 if (unlikely(prev_state == TASK_DEAD)) {
2599 /* 2599 /*
2600 * Remove function-return probe instances associated with this 2600 * Remove function-return probe instances associated with this
2601 * task and put them back on the free list. 2601 * task and put them back on the free list.
2602 */ 2602 */
2603 kprobe_flush_task(prev); 2603 kprobe_flush_task(prev);
2604 put_task_struct(prev); 2604 put_task_struct(prev);
2605 } 2605 }
2606 } 2606 }
2607 2607
2608 /** 2608 /**
2609 * schedule_tail - first thing a freshly forked thread must call. 2609 * schedule_tail - first thing a freshly forked thread must call.
2610 * @prev: the thread we just switched away from. 2610 * @prev: the thread we just switched away from.
2611 */ 2611 */
2612 asmlinkage void schedule_tail(struct task_struct *prev) 2612 asmlinkage void schedule_tail(struct task_struct *prev)
2613 __releases(rq->lock) 2613 __releases(rq->lock)
2614 { 2614 {
2615 struct rq *rq = this_rq(); 2615 struct rq *rq = this_rq();
2616 2616
2617 finish_task_switch(rq, prev); 2617 finish_task_switch(rq, prev);
2618 #ifdef __ARCH_WANT_UNLOCKED_CTXSW 2618 #ifdef __ARCH_WANT_UNLOCKED_CTXSW
2619 /* In this case, finish_task_switch does not reenable preemption */ 2619 /* In this case, finish_task_switch does not reenable preemption */
2620 preempt_enable(); 2620 preempt_enable();
2621 #endif 2621 #endif
2622 if (current->set_child_tid) 2622 if (current->set_child_tid)
2623 put_user(task_pid_vnr(current), current->set_child_tid); 2623 put_user(task_pid_vnr(current), current->set_child_tid);
2624 } 2624 }
2625 2625
2626 /* 2626 /*
2627 * context_switch - switch to the new MM and the new 2627 * context_switch - switch to the new MM and the new
2628 * thread's register state. 2628 * thread's register state.
2629 */ 2629 */
2630 static inline void 2630 static inline void
2631 context_switch(struct rq *rq, struct task_struct *prev, 2631 context_switch(struct rq *rq, struct task_struct *prev,
2632 struct task_struct *next) 2632 struct task_struct *next)
2633 { 2633 {
2634 struct mm_struct *mm, *oldmm; 2634 struct mm_struct *mm, *oldmm;
2635 2635
2636 prepare_task_switch(rq, prev, next); 2636 prepare_task_switch(rq, prev, next);
2637 trace_sched_switch(rq, prev, next); 2637 trace_sched_switch(rq, prev, next);
2638 mm = next->mm; 2638 mm = next->mm;
2639 oldmm = prev->active_mm; 2639 oldmm = prev->active_mm;
2640 /* 2640 /*
2641 * For paravirt, this is coupled with an exit in switch_to to 2641 * For paravirt, this is coupled with an exit in switch_to to
2642 * combine the page table reload and the switch backend into 2642 * combine the page table reload and the switch backend into
2643 * one hypercall. 2643 * one hypercall.
2644 */ 2644 */
2645 arch_enter_lazy_cpu_mode(); 2645 arch_enter_lazy_cpu_mode();
2646 2646
2647 if (unlikely(!mm)) { 2647 if (unlikely(!mm)) {
2648 next->active_mm = oldmm; 2648 next->active_mm = oldmm;
2649 atomic_inc(&oldmm->mm_count); 2649 atomic_inc(&oldmm->mm_count);
2650 enter_lazy_tlb(oldmm, next); 2650 enter_lazy_tlb(oldmm, next);
2651 } else 2651 } else
2652 switch_mm(oldmm, mm, next); 2652 switch_mm(oldmm, mm, next);
2653 2653
2654 if (unlikely(!prev->mm)) { 2654 if (unlikely(!prev->mm)) {
2655 prev->active_mm = NULL; 2655 prev->active_mm = NULL;
2656 rq->prev_mm = oldmm; 2656 rq->prev_mm = oldmm;
2657 } 2657 }
2658 /* 2658 /*
2659 * Since the runqueue lock will be released by the next 2659 * Since the runqueue lock will be released by the next
2660 * task (which is an invalid locking op but in the case 2660 * task (which is an invalid locking op but in the case
2661 * of the scheduler it's an obvious special-case), so we 2661 * of the scheduler it's an obvious special-case), so we
2662 * do an early lockdep release here: 2662 * do an early lockdep release here:
2663 */ 2663 */
2664 #ifndef __ARCH_WANT_UNLOCKED_CTXSW 2664 #ifndef __ARCH_WANT_UNLOCKED_CTXSW
2665 spin_release(&rq->lock.dep_map, 1, _THIS_IP_); 2665 spin_release(&rq->lock.dep_map, 1, _THIS_IP_);
2666 #endif 2666 #endif
2667 2667
2668 /* Here we just switch the register state and the stack. */ 2668 /* Here we just switch the register state and the stack. */
2669 switch_to(prev, next, prev); 2669 switch_to(prev, next, prev);
2670 2670
2671 barrier(); 2671 barrier();
2672 /* 2672 /*
2673 * this_rq must be evaluated again because prev may have moved 2673 * this_rq must be evaluated again because prev may have moved
2674 * CPUs since it called schedule(), thus the 'rq' on its stack 2674 * CPUs since it called schedule(), thus the 'rq' on its stack
2675 * frame will be invalid. 2675 * frame will be invalid.
2676 */ 2676 */
2677 finish_task_switch(this_rq(), prev); 2677 finish_task_switch(this_rq(), prev);
2678 } 2678 }
2679 2679
2680 /* 2680 /*
2681 * nr_running, nr_uninterruptible and nr_context_switches: 2681 * nr_running, nr_uninterruptible and nr_context_switches:
2682 * 2682 *
2683 * externally visible scheduler statistics: current number of runnable 2683 * externally visible scheduler statistics: current number of runnable
2684 * threads, current number of uninterruptible-sleeping threads, total 2684 * threads, current number of uninterruptible-sleeping threads, total
2685 * number of context switches performed since bootup. 2685 * number of context switches performed since bootup.
2686 */ 2686 */
2687 unsigned long nr_running(void) 2687 unsigned long nr_running(void)
2688 { 2688 {
2689 unsigned long i, sum = 0; 2689 unsigned long i, sum = 0;
2690 2690
2691 for_each_online_cpu(i) 2691 for_each_online_cpu(i)
2692 sum += cpu_rq(i)->nr_running; 2692 sum += cpu_rq(i)->nr_running;
2693 2693
2694 return sum; 2694 return sum;
2695 } 2695 }
2696 2696
2697 unsigned long nr_uninterruptible(void) 2697 unsigned long nr_uninterruptible(void)
2698 { 2698 {
2699 unsigned long i, sum = 0; 2699 unsigned long i, sum = 0;
2700 2700
2701 for_each_possible_cpu(i) 2701 for_each_possible_cpu(i)
2702 sum += cpu_rq(i)->nr_uninterruptible; 2702 sum += cpu_rq(i)->nr_uninterruptible;
2703 2703
2704 /* 2704 /*
2705 * Since we read the counters lockless, it might be slightly 2705 * Since we read the counters lockless, it might be slightly
2706 * inaccurate. Do not allow it to go below zero though: 2706 * inaccurate. Do not allow it to go below zero though:
2707 */ 2707 */
2708 if (unlikely((long)sum < 0)) 2708 if (unlikely((long)sum < 0))
2709 sum = 0; 2709 sum = 0;
2710 2710
2711 return sum; 2711 return sum;
2712 } 2712 }
2713 2713
2714 unsigned long long nr_context_switches(void) 2714 unsigned long long nr_context_switches(void)
2715 { 2715 {
2716 int i; 2716 int i;
2717 unsigned long long sum = 0; 2717 unsigned long long sum = 0;
2718 2718
2719 for_each_possible_cpu(i) 2719 for_each_possible_cpu(i)
2720 sum += cpu_rq(i)->nr_switches; 2720 sum += cpu_rq(i)->nr_switches;
2721 2721
2722 return sum; 2722 return sum;
2723 } 2723 }
2724 2724
2725 unsigned long nr_iowait(void) 2725 unsigned long nr_iowait(void)
2726 { 2726 {
2727 unsigned long i, sum = 0; 2727 unsigned long i, sum = 0;
2728 2728
2729 for_each_possible_cpu(i) 2729 for_each_possible_cpu(i)
2730 sum += atomic_read(&cpu_rq(i)->nr_iowait); 2730 sum += atomic_read(&cpu_rq(i)->nr_iowait);
2731 2731
2732 return sum; 2732 return sum;
2733 } 2733 }
2734 2734
2735 unsigned long nr_active(void) 2735 unsigned long nr_active(void)
2736 { 2736 {
2737 unsigned long i, running = 0, uninterruptible = 0; 2737 unsigned long i, running = 0, uninterruptible = 0;
2738 2738
2739 for_each_online_cpu(i) { 2739 for_each_online_cpu(i) {
2740 running += cpu_rq(i)->nr_running; 2740 running += cpu_rq(i)->nr_running;
2741 uninterruptible += cpu_rq(i)->nr_uninterruptible; 2741 uninterruptible += cpu_rq(i)->nr_uninterruptible;
2742 } 2742 }
2743 2743
2744 if (unlikely((long)uninterruptible < 0)) 2744 if (unlikely((long)uninterruptible < 0))
2745 uninterruptible = 0; 2745 uninterruptible = 0;
2746 2746
2747 return running + uninterruptible; 2747 return running + uninterruptible;
2748 } 2748 }
2749 2749
2750 /* 2750 /*
2751 * Update rq->cpu_load[] statistics. This function is usually called every 2751 * Update rq->cpu_load[] statistics. This function is usually called every
2752 * scheduler tick (TICK_NSEC). 2752 * scheduler tick (TICK_NSEC).
2753 */ 2753 */
2754 static void update_cpu_load(struct rq *this_rq) 2754 static void update_cpu_load(struct rq *this_rq)
2755 { 2755 {
2756 unsigned long this_load = this_rq->load.weight; 2756 unsigned long this_load = this_rq->load.weight;
2757 int i, scale; 2757 int i, scale;
2758 2758
2759 this_rq->nr_load_updates++; 2759 this_rq->nr_load_updates++;
2760 2760
2761 /* Update our load: */ 2761 /* Update our load: */
2762 for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { 2762 for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) {
2763 unsigned long old_load, new_load; 2763 unsigned long old_load, new_load;
2764 2764
2765 /* scale is effectively 1 << i now, and >> i divides by scale */ 2765 /* scale is effectively 1 << i now, and >> i divides by scale */
2766 2766
2767 old_load = this_rq->cpu_load[i]; 2767 old_load = this_rq->cpu_load[i];
2768 new_load = this_load; 2768 new_load = this_load;
2769 /* 2769 /*
2770 * Round up the averaging division if load is increasing. This 2770 * Round up the averaging division if load is increasing. This
2771 * prevents us from getting stuck on 9 if the load is 10, for 2771 * prevents us from getting stuck on 9 if the load is 10, for
2772 * example. 2772 * example.
2773 */ 2773 */
2774 if (new_load > old_load) 2774 if (new_load > old_load)
2775 new_load += scale-1; 2775 new_load += scale-1;
2776 this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; 2776 this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i;
2777 } 2777 }
2778 } 2778 }
2779 2779
2780 #ifdef CONFIG_SMP 2780 #ifdef CONFIG_SMP
2781 2781
2782 /* 2782 /*
2783 * double_rq_lock - safely lock two runqueues 2783 * double_rq_lock - safely lock two runqueues
2784 * 2784 *
2785 * Note this does not disable interrupts like task_rq_lock, 2785 * Note this does not disable interrupts like task_rq_lock,
2786 * you need to do so manually before calling. 2786 * you need to do so manually before calling.
2787 */ 2787 */
2788 static void double_rq_lock(struct rq *rq1, struct rq *rq2) 2788 static void double_rq_lock(struct rq *rq1, struct rq *rq2)
2789 __acquires(rq1->lock) 2789 __acquires(rq1->lock)
2790 __acquires(rq2->lock) 2790 __acquires(rq2->lock)
2791 { 2791 {
2792 BUG_ON(!irqs_disabled()); 2792 BUG_ON(!irqs_disabled());
2793 if (rq1 == rq2) { 2793 if (rq1 == rq2) {
2794 spin_lock(&rq1->lock); 2794 spin_lock(&rq1->lock);
2795 __acquire(rq2->lock); /* Fake it out ;) */ 2795 __acquire(rq2->lock); /* Fake it out ;) */
2796 } else { 2796 } else {
2797 if (rq1 < rq2) { 2797 if (rq1 < rq2) {
2798 spin_lock(&rq1->lock); 2798 spin_lock(&rq1->lock);
2799 spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); 2799 spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING);
2800 } else { 2800 } else {
2801 spin_lock(&rq2->lock); 2801 spin_lock(&rq2->lock);
2802 spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); 2802 spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING);
2803 } 2803 }
2804 } 2804 }
2805 update_rq_clock(rq1); 2805 update_rq_clock(rq1);
2806 update_rq_clock(rq2); 2806 update_rq_clock(rq2);
2807 } 2807 }
2808 2808
2809 /* 2809 /*
2810 * double_rq_unlock - safely unlock two runqueues 2810 * double_rq_unlock - safely unlock two runqueues
2811 * 2811 *
2812 * Note this does not restore interrupts like task_rq_unlock, 2812 * Note this does not restore interrupts like task_rq_unlock,
2813 * you need to do so manually after calling. 2813 * you need to do so manually after calling.
2814 */ 2814 */
2815 static void double_rq_unlock(struct rq *rq1, struct rq *rq2) 2815 static void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2816 __releases(rq1->lock) 2816 __releases(rq1->lock)
2817 __releases(rq2->lock) 2817 __releases(rq2->lock)
2818 { 2818 {
2819 spin_unlock(&rq1->lock); 2819 spin_unlock(&rq1->lock);
2820 if (rq1 != rq2) 2820 if (rq1 != rq2)
2821 spin_unlock(&rq2->lock); 2821 spin_unlock(&rq2->lock);
2822 else 2822 else
2823 __release(rq2->lock); 2823 __release(rq2->lock);
2824 } 2824 }
2825 2825
2826 /* 2826 /*
2827 * If dest_cpu is allowed for this process, migrate the task to it. 2827 * If dest_cpu is allowed for this process, migrate the task to it.
2828 * This is accomplished by forcing the cpu_allowed mask to only 2828 * This is accomplished by forcing the cpu_allowed mask to only
2829 * allow dest_cpu, which will force the cpu onto dest_cpu. Then 2829 * allow dest_cpu, which will force the cpu onto dest_cpu. Then
2830 * the cpu_allowed mask is restored. 2830 * the cpu_allowed mask is restored.
2831 */ 2831 */
2832 static void sched_migrate_task(struct task_struct *p, int dest_cpu) 2832 static void sched_migrate_task(struct task_struct *p, int dest_cpu)
2833 { 2833 {
2834 struct migration_req req; 2834 struct migration_req req;
2835 unsigned long flags; 2835 unsigned long flags;
2836 struct rq *rq; 2836 struct rq *rq;
2837 2837
2838 rq = task_rq_lock(p, &flags); 2838 rq = task_rq_lock(p, &flags);
2839 if (!cpu_isset(dest_cpu, p->cpus_allowed) 2839 if (!cpu_isset(dest_cpu, p->cpus_allowed)
2840 || unlikely(!cpu_active(dest_cpu))) 2840 || unlikely(!cpu_active(dest_cpu)))
2841 goto out; 2841 goto out;
2842 2842
2843 trace_sched_migrate_task(rq, p, dest_cpu); 2843 trace_sched_migrate_task(rq, p, dest_cpu);
2844 /* force the process onto the specified CPU */ 2844 /* force the process onto the specified CPU */
2845 if (migrate_task(p, dest_cpu, &req)) { 2845 if (migrate_task(p, dest_cpu, &req)) {
2846 /* Need to wait for migration thread (might exit: take ref). */ 2846 /* Need to wait for migration thread (might exit: take ref). */
2847 struct task_struct *mt = rq->migration_thread; 2847 struct task_struct *mt = rq->migration_thread;
2848 2848
2849 get_task_struct(mt); 2849 get_task_struct(mt);
2850 task_rq_unlock(rq, &flags); 2850 task_rq_unlock(rq, &flags);
2851 wake_up_process(mt); 2851 wake_up_process(mt);
2852 put_task_struct(mt); 2852 put_task_struct(mt);
2853 wait_for_completion(&req.done); 2853 wait_for_completion(&req.done);
2854 2854
2855 return; 2855 return;
2856 } 2856 }
2857 out: 2857 out:
2858 task_rq_unlock(rq, &flags); 2858 task_rq_unlock(rq, &flags);
2859 } 2859 }
2860 2860
2861 /* 2861 /*
2862 * sched_exec - execve() is a valuable balancing opportunity, because at 2862 * sched_exec - execve() is a valuable balancing opportunity, because at
2863 * this point the task has the smallest effective memory and cache footprint. 2863 * this point the task has the smallest effective memory and cache footprint.
2864 */ 2864 */
2865 void sched_exec(void) 2865 void sched_exec(void)
2866 { 2866 {
2867 int new_cpu, this_cpu = get_cpu(); 2867 int new_cpu, this_cpu = get_cpu();
2868 new_cpu = sched_balance_self(this_cpu, SD_BALANCE_EXEC); 2868 new_cpu = sched_balance_self(this_cpu, SD_BALANCE_EXEC);
2869 put_cpu(); 2869 put_cpu();
2870 if (new_cpu != this_cpu) 2870 if (new_cpu != this_cpu)
2871 sched_migrate_task(current, new_cpu); 2871 sched_migrate_task(current, new_cpu);
2872 } 2872 }
2873 2873
2874 /* 2874 /*
2875 * pull_task - move a task from a remote runqueue to the local runqueue. 2875 * pull_task - move a task from a remote runqueue to the local runqueue.
2876 * Both runqueues must be locked. 2876 * Both runqueues must be locked.
2877 */ 2877 */
2878 static void pull_task(struct rq *src_rq, struct task_struct *p, 2878 static void pull_task(struct rq *src_rq, struct task_struct *p,
2879 struct rq *this_rq, int this_cpu) 2879 struct rq *this_rq, int this_cpu)
2880 { 2880 {
2881 deactivate_task(src_rq, p, 0); 2881 deactivate_task(src_rq, p, 0);
2882 set_task_cpu(p, this_cpu); 2882 set_task_cpu(p, this_cpu);
2883 activate_task(this_rq, p, 0); 2883 activate_task(this_rq, p, 0);
2884 /* 2884 /*
2885 * Note that idle threads have a prio of MAX_PRIO, for this test 2885 * Note that idle threads have a prio of MAX_PRIO, for this test
2886 * to be always true for them. 2886 * to be always true for them.
2887 */ 2887 */
2888 check_preempt_curr(this_rq, p, 0); 2888 check_preempt_curr(this_rq, p, 0);
2889 } 2889 }
2890 2890
2891 /* 2891 /*
2892 * can_migrate_task - may task p from runqueue rq be migrated to this_cpu? 2892 * can_migrate_task - may task p from runqueue rq be migrated to this_cpu?
2893 */ 2893 */
2894 static 2894 static
2895 int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu, 2895 int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu,
2896 struct sched_domain *sd, enum cpu_idle_type idle, 2896 struct sched_domain *sd, enum cpu_idle_type idle,
2897 int *all_pinned) 2897 int *all_pinned)
2898 { 2898 {
2899 /* 2899 /*
2900 * We do not migrate tasks that are: 2900 * We do not migrate tasks that are:
2901 * 1) running (obviously), or 2901 * 1) running (obviously), or
2902 * 2) cannot be migrated to this CPU due to cpus_allowed, or 2902 * 2) cannot be migrated to this CPU due to cpus_allowed, or
2903 * 3) are cache-hot on their current CPU. 2903 * 3) are cache-hot on their current CPU.
2904 */ 2904 */
2905 if (!cpu_isset(this_cpu, p->cpus_allowed)) { 2905 if (!cpu_isset(this_cpu, p->cpus_allowed)) {
2906 schedstat_inc(p, se.nr_failed_migrations_affine); 2906 schedstat_inc(p, se.nr_failed_migrations_affine);
2907 return 0; 2907 return 0;
2908 } 2908 }
2909 *all_pinned = 0; 2909 *all_pinned = 0;
2910 2910
2911 if (task_running(rq, p)) { 2911 if (task_running(rq, p)) {
2912 schedstat_inc(p, se.nr_failed_migrations_running); 2912 schedstat_inc(p, se.nr_failed_migrations_running);
2913 return 0; 2913 return 0;
2914 } 2914 }
2915 2915
2916 /* 2916 /*
2917 * Aggressive migration if: 2917 * Aggressive migration if:
2918 * 1) task is cache cold, or 2918 * 1) task is cache cold, or
2919 * 2) too many balance attempts have failed. 2919 * 2) too many balance attempts have failed.
2920 */ 2920 */
2921 2921
2922 if (!task_hot(p, rq->clock, sd) || 2922 if (!task_hot(p, rq->clock, sd) ||
2923 sd->nr_balance_failed > sd->cache_nice_tries) { 2923 sd->nr_balance_failed > sd->cache_nice_tries) {
2924 #ifdef CONFIG_SCHEDSTATS 2924 #ifdef CONFIG_SCHEDSTATS
2925 if (task_hot(p, rq->clock, sd)) { 2925 if (task_hot(p, rq->clock, sd)) {
2926 schedstat_inc(sd, lb_hot_gained[idle]); 2926 schedstat_inc(sd, lb_hot_gained[idle]);
2927 schedstat_inc(p, se.nr_forced_migrations); 2927 schedstat_inc(p, se.nr_forced_migrations);
2928 } 2928 }
2929 #endif 2929 #endif
2930 return 1; 2930 return 1;
2931 } 2931 }
2932 2932
2933 if (task_hot(p, rq->clock, sd)) { 2933 if (task_hot(p, rq->clock, sd)) {
2934 schedstat_inc(p, se.nr_failed_migrations_hot); 2934 schedstat_inc(p, se.nr_failed_migrations_hot);
2935 return 0; 2935 return 0;
2936 } 2936 }
2937 return 1; 2937 return 1;
2938 } 2938 }
2939 2939
2940 static unsigned long 2940 static unsigned long
2941 balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, 2941 balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest,
2942 unsigned long max_load_move, struct sched_domain *sd, 2942 unsigned long max_load_move, struct sched_domain *sd,
2943 enum cpu_idle_type idle, int *all_pinned, 2943 enum cpu_idle_type idle, int *all_pinned,
2944 int *this_best_prio, struct rq_iterator *iterator) 2944 int *this_best_prio, struct rq_iterator *iterator)
2945 { 2945 {
2946 int loops = 0, pulled = 0, pinned = 0; 2946 int loops = 0, pulled = 0, pinned = 0;
2947 struct task_struct *p; 2947 struct task_struct *p;
2948 long rem_load_move = max_load_move; 2948 long rem_load_move = max_load_move;
2949 2949
2950 if (max_load_move == 0) 2950 if (max_load_move == 0)
2951 goto out; 2951 goto out;
2952 2952
2953 pinned = 1; 2953 pinned = 1;
2954 2954
2955 /* 2955 /*
2956 * Start the load-balancing iterator: 2956 * Start the load-balancing iterator:
2957 */ 2957 */
2958 p = iterator->start(iterator->arg); 2958 p = iterator->start(iterator->arg);
2959 next: 2959 next:
2960 if (!p || loops++ > sysctl_sched_nr_migrate) 2960 if (!p || loops++ > sysctl_sched_nr_migrate)
2961 goto out; 2961 goto out;
2962 2962
2963 if ((p->se.load.weight >> 1) > rem_load_move || 2963 if ((p->se.load.weight >> 1) > rem_load_move ||
2964 !can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { 2964 !can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) {
2965 p = iterator->next(iterator->arg); 2965 p = iterator->next(iterator->arg);
2966 goto next; 2966 goto next;
2967 } 2967 }
2968 2968
2969 pull_task(busiest, p, this_rq, this_cpu); 2969 pull_task(busiest, p, this_rq, this_cpu);
2970 pulled++; 2970 pulled++;
2971 rem_load_move -= p->se.load.weight; 2971 rem_load_move -= p->se.load.weight;
2972 2972
2973 /* 2973 /*
2974 * We only want to steal up to the prescribed amount of weighted load. 2974 * We only want to steal up to the prescribed amount of weighted load.
2975 */ 2975 */
2976 if (rem_load_move > 0) { 2976 if (rem_load_move > 0) {
2977 if (p->prio < *this_best_prio) 2977 if (p->prio < *this_best_prio)
2978 *this_best_prio = p->prio; 2978 *this_best_prio = p->prio;
2979 p = iterator->next(iterator->arg); 2979 p = iterator->next(iterator->arg);
2980 goto next; 2980 goto next;
2981 } 2981 }
2982 out: 2982 out:
2983 /* 2983 /*
2984 * Right now, this is one of only two places pull_task() is called, 2984 * Right now, this is one of only two places pull_task() is called,
2985 * so we can safely collect pull_task() stats here rather than 2985 * so we can safely collect pull_task() stats here rather than
2986 * inside pull_task(). 2986 * inside pull_task().
2987 */ 2987 */
2988 schedstat_add(sd, lb_gained[idle], pulled); 2988 schedstat_add(sd, lb_gained[idle], pulled);
2989 2989
2990 if (all_pinned) 2990 if (all_pinned)
2991 *all_pinned = pinned; 2991 *all_pinned = pinned;
2992 2992
2993 return max_load_move - rem_load_move; 2993 return max_load_move - rem_load_move;
2994 } 2994 }
2995 2995
2996 /* 2996 /*
2997 * move_tasks tries to move up to max_load_move weighted load from busiest to 2997 * move_tasks tries to move up to max_load_move weighted load from busiest to
2998 * this_rq, as part of a balancing operation within domain "sd". 2998 * this_rq, as part of a balancing operation within domain "sd".
2999 * Returns 1 if successful and 0 otherwise. 2999 * Returns 1 if successful and 0 otherwise.
3000 * 3000 *
3001 * Called with both runqueues locked. 3001 * Called with both runqueues locked.
3002 */ 3002 */
3003 static int move_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, 3003 static int move_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest,
3004 unsigned long max_load_move, 3004 unsigned long max_load_move,
3005 struct sched_domain *sd, enum cpu_idle_type idle, 3005 struct sched_domain *sd, enum cpu_idle_type idle,
3006 int *all_pinned) 3006 int *all_pinned)
3007 { 3007 {
3008 const struct sched_class *class = sched_class_highest; 3008 const struct sched_class *class = sched_class_highest;
3009 unsigned long total_load_moved = 0; 3009 unsigned long total_load_moved = 0;
3010 int this_best_prio = this_rq->curr->prio; 3010 int this_best_prio = this_rq->curr->prio;
3011 3011
3012 do { 3012 do {
3013 total_load_moved += 3013 total_load_moved +=
3014 class->load_balance(this_rq, this_cpu, busiest, 3014 class->load_balance(this_rq, this_cpu, busiest,
3015 max_load_move - total_load_moved, 3015 max_load_move - total_load_moved,
3016 sd, idle, all_pinned, &this_best_prio); 3016 sd, idle, all_pinned, &this_best_prio);
3017 class = class->next; 3017 class = class->next;
3018 3018
3019 if (idle == CPU_NEWLY_IDLE && this_rq->nr_running) 3019 if (idle == CPU_NEWLY_IDLE && this_rq->nr_running)
3020 break; 3020 break;
3021 3021
3022 } while (class && max_load_move > total_load_moved); 3022 } while (class && max_load_move > total_load_moved);
3023 3023
3024 return total_load_moved > 0; 3024 return total_load_moved > 0;
3025 } 3025 }
3026 3026
3027 static int 3027 static int
3028 iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, 3028 iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest,
3029 struct sched_domain *sd, enum cpu_idle_type idle, 3029 struct sched_domain *sd, enum cpu_idle_type idle,
3030 struct rq_iterator *iterator) 3030 struct rq_iterator *iterator)
3031 { 3031 {
3032 struct task_struct *p = iterator->start(iterator->arg); 3032 struct task_struct *p = iterator->start(iterator->arg);
3033 int pinned = 0; 3033 int pinned = 0;
3034 3034
3035 while (p) { 3035 while (p) {
3036 if (can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { 3036 if (can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) {
3037 pull_task(busiest, p, this_rq, this_cpu); 3037 pull_task(busiest, p, this_rq, this_cpu);
3038 /* 3038 /*
3039 * Right now, this is only the second place pull_task() 3039 * Right now, this is only the second place pull_task()
3040 * is called, so we can safely collect pull_task() 3040 * is called, so we can safely collect pull_task()
3041 * stats here rather than inside pull_task(). 3041 * stats here rather than inside pull_task().
3042 */ 3042 */
3043 schedstat_inc(sd, lb_gained[idle]); 3043 schedstat_inc(sd, lb_gained[idle]);
3044 3044
3045 return 1; 3045 return 1;
3046 } 3046 }
3047 p = iterator->next(iterator->arg); 3047 p = iterator->next(iterator->arg);
3048 } 3048 }
3049 3049
3050 return 0; 3050 return 0;
3051 } 3051 }
3052 3052
3053 /* 3053 /*
3054 * move_one_task tries to move exactly one task from busiest to this_rq, as 3054 * move_one_task tries to move exactly one task from busiest to this_rq, as
3055 * part of active balancing operations within "domain". 3055 * part of active balancing operations within "domain".
3056 * Returns 1 if successful and 0 otherwise. 3056 * Returns 1 if successful and 0 otherwise.
3057 * 3057 *
3058 * Called with both runqueues locked. 3058 * Called with both runqueues locked.
3059 */ 3059 */
3060 static int move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, 3060 static int move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest,
3061 struct sched_domain *sd, enum cpu_idle_type idle) 3061 struct sched_domain *sd, enum cpu_idle_type idle)
3062 { 3062 {
3063 const struct sched_class *class; 3063 const struct sched_class *class;
3064 3064
3065 for (class = sched_class_highest; class; class = class->next) 3065 for (class = sched_class_highest; class; class = class->next)
3066 if (class->move_one_task(this_rq, this_cpu, busiest, sd, idle)) 3066 if (class->move_one_task(this_rq, this_cpu, busiest, sd, idle))
3067 return 1; 3067 return 1;
3068 3068
3069 return 0; 3069 return 0;
3070 } 3070 }
3071 3071
3072 /* 3072 /*
3073 * find_busiest_group finds and returns the busiest CPU group within the 3073 * find_busiest_group finds and returns the busiest CPU group within the
3074 * domain. It calculates and returns the amount of weighted load which 3074 * domain. It calculates and returns the amount of weighted load which
3075 * should be moved to restore balance via the imbalance parameter. 3075 * should be moved to restore balance via the imbalance parameter.
3076 */ 3076 */
3077 static struct sched_group * 3077 static struct sched_group *
3078 find_busiest_group(struct sched_domain *sd, int this_cpu, 3078 find_busiest_group(struct sched_domain *sd, int this_cpu,
3079 unsigned long *imbalance, enum cpu_idle_type idle, 3079 unsigned long *imbalance, enum cpu_idle_type idle,
3080 int *sd_idle, const cpumask_t *cpus, int *balance) 3080 int *sd_idle, const cpumask_t *cpus, int *balance)
3081 { 3081 {
3082 struct sched_group *busiest = NULL, *this = NULL, *group = sd->groups; 3082 struct sched_group *busiest = NULL, *this = NULL, *group = sd->groups;
3083 unsigned long max_load, avg_load, total_load, this_load, total_pwr; 3083 unsigned long max_load, avg_load, total_load, this_load, total_pwr;
3084 unsigned long max_pull; 3084 unsigned long max_pull;
3085 unsigned long busiest_load_per_task, busiest_nr_running; 3085 unsigned long busiest_load_per_task, busiest_nr_running;
3086 unsigned long this_load_per_task, this_nr_running; 3086 unsigned long this_load_per_task, this_nr_running;
3087 int load_idx, group_imb = 0; 3087 int load_idx, group_imb = 0;
3088 #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) 3088 #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT)
3089 int power_savings_balance = 1; 3089 int power_savings_balance = 1;
3090 unsigned long leader_nr_running = 0, min_load_per_task = 0; 3090 unsigned long leader_nr_running = 0, min_load_per_task = 0;
3091 unsigned long min_nr_running = ULONG_MAX; 3091 unsigned long min_nr_running = ULONG_MAX;
3092 struct sched_group *group_min = NULL, *group_leader = NULL; 3092 struct sched_group *group_min = NULL, *group_leader = NULL;
3093 #endif 3093 #endif
3094 3094
3095 max_load = this_load = total_load = total_pwr = 0; 3095 max_load = this_load = total_load = total_pwr = 0;
3096 busiest_load_per_task = busiest_nr_running = 0; 3096 busiest_load_per_task = busiest_nr_running = 0;
3097 this_load_per_task = this_nr_running = 0; 3097 this_load_per_task = this_nr_running = 0;
3098 3098
3099 if (idle == CPU_NOT_IDLE) 3099 if (idle == CPU_NOT_IDLE)
3100 load_idx = sd->busy_idx; 3100 load_idx = sd->busy_idx;
3101 else if (idle == CPU_NEWLY_IDLE) 3101 else if (idle == CPU_NEWLY_IDLE)
3102 load_idx = sd->newidle_idx; 3102 load_idx = sd->newidle_idx;
3103 else 3103 else
3104 load_idx = sd->idle_idx; 3104 load_idx = sd->idle_idx;
3105 3105
3106 do { 3106 do {
3107 unsigned long load, group_capacity, max_cpu_load, min_cpu_load; 3107 unsigned long load, group_capacity, max_cpu_load, min_cpu_load;
3108 int local_group; 3108 int local_group;
3109 int i; 3109 int i;
3110 int __group_imb = 0; 3110 int __group_imb = 0;
3111 unsigned int balance_cpu = -1, first_idle_cpu = 0; 3111 unsigned int balance_cpu = -1, first_idle_cpu = 0;
3112 unsigned long sum_nr_running, sum_weighted_load; 3112 unsigned long sum_nr_running, sum_weighted_load;
3113 unsigned long sum_avg_load_per_task; 3113 unsigned long sum_avg_load_per_task;
3114 unsigned long avg_load_per_task; 3114 unsigned long avg_load_per_task;
3115 3115
3116 local_group = cpu_isset(this_cpu, group->cpumask); 3116 local_group = cpu_isset(this_cpu, group->cpumask);
3117 3117
3118 if (local_group) 3118 if (local_group)
3119 balance_cpu = first_cpu(group->cpumask); 3119 balance_cpu = first_cpu(group->cpumask);
3120 3120
3121 /* Tally up the load of all CPUs in the group */ 3121 /* Tally up the load of all CPUs in the group */
3122 sum_weighted_load = sum_nr_running = avg_load = 0; 3122 sum_weighted_load = sum_nr_running = avg_load = 0;
3123 sum_avg_load_per_task = avg_load_per_task = 0; 3123 sum_avg_load_per_task = avg_load_per_task = 0;
3124 3124
3125 max_cpu_load = 0; 3125 max_cpu_load = 0;
3126 min_cpu_load = ~0UL; 3126 min_cpu_load = ~0UL;
3127 3127
3128 for_each_cpu_mask_nr(i, group->cpumask) { 3128 for_each_cpu_mask_nr(i, group->cpumask) {
3129 struct rq *rq; 3129 struct rq *rq;
3130 3130
3131 if (!cpu_isset(i, *cpus)) 3131 if (!cpu_isset(i, *cpus))
3132 continue; 3132 continue;
3133 3133
3134 rq = cpu_rq(i); 3134 rq = cpu_rq(i);
3135 3135
3136 if (*sd_idle && rq->nr_running) 3136 if (*sd_idle && rq->nr_running)
3137 *sd_idle = 0; 3137 *sd_idle = 0;
3138 3138
3139 /* Bias balancing toward cpus of our domain */ 3139 /* Bias balancing toward cpus of our domain */
3140 if (local_group) { 3140 if (local_group) {
3141 if (idle_cpu(i) && !first_idle_cpu) { 3141 if (idle_cpu(i) && !first_idle_cpu) {
3142 first_idle_cpu = 1; 3142 first_idle_cpu = 1;
3143 balance_cpu = i; 3143 balance_cpu = i;
3144 } 3144 }
3145 3145
3146 load = target_load(i, load_idx); 3146 load = target_load(i, load_idx);
3147 } else { 3147 } else {
3148 load = source_load(i, load_idx); 3148 load = source_load(i, load_idx);
3149 if (load > max_cpu_load) 3149 if (load > max_cpu_load)
3150 max_cpu_load = load; 3150 max_cpu_load = load;
3151 if (min_cpu_load > load) 3151 if (min_cpu_load > load)
3152 min_cpu_load = load; 3152 min_cpu_load = load;
3153 } 3153 }
3154 3154
3155 avg_load += load; 3155 avg_load += load;
3156 sum_nr_running += rq->nr_running; 3156 sum_nr_running += rq->nr_running;
3157 sum_weighted_load += weighted_cpuload(i); 3157 sum_weighted_load += weighted_cpuload(i);
3158 3158
3159 sum_avg_load_per_task += cpu_avg_load_per_task(i); 3159 sum_avg_load_per_task += cpu_avg_load_per_task(i);
3160 } 3160 }
3161 3161
3162 /* 3162 /*
3163 * First idle cpu or the first cpu(busiest) in this sched group 3163 * First idle cpu or the first cpu(busiest) in this sched group
3164 * is eligible for doing load balancing at this and above 3164 * is eligible for doing load balancing at this and above
3165 * domains. In the newly idle case, we will allow all the cpu's 3165 * domains. In the newly idle case, we will allow all the cpu's
3166 * to do the newly idle load balance. 3166 * to do the newly idle load balance.
3167 */ 3167 */
3168 if (idle != CPU_NEWLY_IDLE && local_group && 3168 if (idle != CPU_NEWLY_IDLE && local_group &&
3169 balance_cpu != this_cpu && balance) { 3169 balance_cpu != this_cpu && balance) {
3170 *balance = 0; 3170 *balance = 0;
3171 goto ret; 3171 goto ret;
3172 } 3172 }
3173 3173
3174 total_load += avg_load; 3174 total_load += avg_load;
3175 total_pwr += group->__cpu_power; 3175 total_pwr += group->__cpu_power;
3176 3176
3177 /* Adjust by relative CPU power of the group */ 3177 /* Adjust by relative CPU power of the group */
3178 avg_load = sg_div_cpu_power(group, 3178 avg_load = sg_div_cpu_power(group,
3179 avg_load * SCHED_LOAD_SCALE); 3179 avg_load * SCHED_LOAD_SCALE);
3180 3180
3181 3181
3182 /* 3182 /*
3183 * Consider the group unbalanced when the imbalance is larger 3183 * Consider the group unbalanced when the imbalance is larger
3184 * than the average weight of two tasks. 3184 * than the average weight of two tasks.
3185 * 3185 *
3186 * APZ: with cgroup the avg task weight can vary wildly and 3186 * APZ: with cgroup the avg task weight can vary wildly and
3187 * might not be a suitable number - should we keep a 3187 * might not be a suitable number - should we keep a
3188 * normalized nr_running number somewhere that negates 3188 * normalized nr_running number somewhere that negates
3189 * the hierarchy? 3189 * the hierarchy?
3190 */ 3190 */
3191 avg_load_per_task = sg_div_cpu_power(group, 3191 avg_load_per_task = sg_div_cpu_power(group,
3192 sum_avg_load_per_task * SCHED_LOAD_SCALE); 3192 sum_avg_load_per_task * SCHED_LOAD_SCALE);
3193 3193
3194 if ((max_cpu_load - min_cpu_load) > 2*avg_load_per_task) 3194 if ((max_cpu_load - min_cpu_load) > 2*avg_load_per_task)
3195 __group_imb = 1; 3195 __group_imb = 1;
3196 3196
3197 group_capacity = group->__cpu_power / SCHED_LOAD_SCALE; 3197 group_capacity = group->__cpu_power / SCHED_LOAD_SCALE;
3198 3198
3199 if (local_group) { 3199 if (local_group) {
3200 this_load = avg_load; 3200 this_load = avg_load;
3201 this = group; 3201 this = group;
3202 this_nr_running = sum_nr_running; 3202 this_nr_running = sum_nr_running;
3203 this_load_per_task = sum_weighted_load; 3203 this_load_per_task = sum_weighted_load;
3204 } else if (avg_load > max_load && 3204 } else if (avg_load > max_load &&
3205 (sum_nr_running > group_capacity || __group_imb)) { 3205 (sum_nr_running > group_capacity || __group_imb)) {
3206 max_load = avg_load; 3206 max_load = avg_load;
3207 busiest = group; 3207 busiest = group;
3208 busiest_nr_running = sum_nr_running; 3208 busiest_nr_running = sum_nr_running;
3209 busiest_load_per_task = sum_weighted_load; 3209 busiest_load_per_task = sum_weighted_load;
3210 group_imb = __group_imb; 3210 group_imb = __group_imb;
3211 } 3211 }
3212 3212
3213 #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) 3213 #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT)
3214 /* 3214 /*
3215 * Busy processors will not participate in power savings 3215 * Busy processors will not participate in power savings
3216 * balance. 3216 * balance.
3217 */ 3217 */
3218 if (idle == CPU_NOT_IDLE || 3218 if (idle == CPU_NOT_IDLE ||
3219 !(sd->flags & SD_POWERSAVINGS_BALANCE)) 3219 !(sd->flags & SD_POWERSAVINGS_BALANCE))
3220 goto group_next; 3220 goto group_next;
3221 3221
3222 /* 3222 /*
3223 * If the local group is idle or completely loaded 3223 * If the local group is idle or completely loaded
3224 * no need to do power savings balance at this domain 3224 * no need to do power savings balance at this domain
3225 */ 3225 */
3226 if (local_group && (this_nr_running >= group_capacity || 3226 if (local_group && (this_nr_running >= group_capacity ||
3227 !this_nr_running)) 3227 !this_nr_running))
3228 power_savings_balance = 0; 3228 power_savings_balance = 0;
3229 3229
3230 /* 3230 /*
3231 * If a group is already running at full capacity or idle, 3231 * If a group is already running at full capacity or idle,
3232 * don't include that group in power savings calculations 3232 * don't include that group in power savings calculations
3233 */ 3233 */
3234 if (!power_savings_balance || sum_nr_running >= group_capacity 3234 if (!power_savings_balance || sum_nr_running >= group_capacity
3235 || !sum_nr_running) 3235 || !sum_nr_running)
3236 goto group_next; 3236 goto group_next;
3237 3237
3238 /* 3238 /*
3239 * Calculate the group which has the least non-idle load. 3239 * Calculate the group which has the least non-idle load.
3240 * This is the group from where we need to pick up the load 3240 * This is the group from where we need to pick up the load
3241 * for saving power 3241 * for saving power
3242 */ 3242 */
3243 if ((sum_nr_running < min_nr_running) || 3243 if ((sum_nr_running < min_nr_running) ||
3244 (sum_nr_running == min_nr_running && 3244 (sum_nr_running == min_nr_running &&
3245 first_cpu(group->cpumask) < 3245 first_cpu(group->cpumask) <
3246 first_cpu(group_min->cpumask))) { 3246 first_cpu(group_min->cpumask))) {
3247 group_min = group; 3247 group_min = group;
3248 min_nr_running = sum_nr_running; 3248 min_nr_running = sum_nr_running;
3249 min_load_per_task = sum_weighted_load / 3249 min_load_per_task = sum_weighted_load /
3250 sum_nr_running; 3250 sum_nr_running;
3251 } 3251 }
3252 3252
3253 /* 3253 /*
3254 * Calculate the group which is almost near its 3254 * Calculate the group which is almost near its
3255 * capacity but still has some space to pick up some load 3255 * capacity but still has some space to pick up some load
3256 * from other group and save more power 3256 * from other group and save more power
3257 */ 3257 */
3258 if (sum_nr_running <= group_capacity - 1) { 3258 if (sum_nr_running <= group_capacity - 1) {
3259 if (sum_nr_running > leader_nr_running || 3259 if (sum_nr_running > leader_nr_running ||
3260 (sum_nr_running == leader_nr_running && 3260 (sum_nr_running == leader_nr_running &&
3261 first_cpu(group->cpumask) > 3261 first_cpu(group->cpumask) >
3262 first_cpu(group_leader->cpumask))) { 3262 first_cpu(group_leader->cpumask))) {
3263 group_leader = group; 3263 group_leader = group;
3264 leader_nr_running = sum_nr_running; 3264 leader_nr_running = sum_nr_running;
3265 } 3265 }
3266 } 3266 }
3267 group_next: 3267 group_next:
3268 #endif 3268 #endif
3269 group = group->next; 3269 group = group->next;
3270 } while (group != sd->groups); 3270 } while (group != sd->groups);
3271 3271
3272 if (!busiest || this_load >= max_load || busiest_nr_running == 0) 3272 if (!busiest || this_load >= max_load || busiest_nr_running == 0)
3273 goto out_balanced; 3273 goto out_balanced;
3274 3274
3275 avg_load = (SCHED_LOAD_SCALE * total_load) / total_pwr; 3275 avg_load = (SCHED_LOAD_SCALE * total_load) / total_pwr;
3276 3276
3277 if (this_load >= avg_load || 3277 if (this_load >= avg_load ||
3278 100*max_load <= sd->imbalance_pct*this_load) 3278 100*max_load <= sd->imbalance_pct*this_load)
3279 goto out_balanced; 3279 goto out_balanced;
3280 3280
3281 busiest_load_per_task /= busiest_nr_running; 3281 busiest_load_per_task /= busiest_nr_running;
3282 if (group_imb) 3282 if (group_imb)
3283 busiest_load_per_task = min(busiest_load_per_task, avg_load); 3283 busiest_load_per_task = min(busiest_load_per_task, avg_load);
3284 3284
3285 /* 3285 /*
3286 * We're trying to get all the cpus to the average_load, so we don't 3286 * We're trying to get all the cpus to the average_load, so we don't
3287 * want to push ourselves above the average load, nor do we wish to 3287 * want to push ourselves above the average load, nor do we wish to
3288 * reduce the max loaded cpu below the average load, as either of these 3288 * reduce the max loaded cpu below the average load, as either of these
3289 * actions would just result in more rebalancing later, and ping-pong 3289 * actions would just result in more rebalancing later, and ping-pong
3290 * tasks around. Thus we look for the minimum possible imbalance. 3290 * tasks around. Thus we look for the minimum possible imbalance.
3291 * Negative imbalances (*we* are more loaded than anyone else) will 3291 * Negative imbalances (*we* are more loaded than anyone else) will
3292 * be counted as no imbalance for these purposes -- we can't fix that 3292 * be counted as no imbalance for these purposes -- we can't fix that
3293 * by pulling tasks to us. Be careful of negative numbers as they'll 3293 * by pulling tasks to us. Be careful of negative numbers as they'll
3294 * appear as very large values with unsigned longs. 3294 * appear as very large values with unsigned longs.
3295 */ 3295 */
3296 if (max_load <= busiest_load_per_task) 3296 if (max_load <= busiest_load_per_task)
3297 goto out_balanced; 3297 goto out_balanced;
3298 3298
3299 /* 3299 /*
3300 * In the presence of smp nice balancing, certain scenarios can have 3300 * In the presence of smp nice balancing, certain scenarios can have
3301 * max load less than avg load(as we skip the groups at or below 3301 * max load less than avg load(as we skip the groups at or below
3302 * its cpu_power, while calculating max_load..) 3302 * its cpu_power, while calculating max_load..)
3303 */ 3303 */
3304 if (max_load < avg_load) { 3304 if (max_load < avg_load) {
3305 *imbalance = 0; 3305 *imbalance = 0;
3306 goto small_imbalance; 3306 goto small_imbalance;
3307 } 3307 }
3308 3308
3309 /* Don't want to pull so many tasks that a group would go idle */ 3309 /* Don't want to pull so many tasks that a group would go idle */
3310 max_pull = min(max_load - avg_load, max_load - busiest_load_per_task); 3310 max_pull = min(max_load - avg_load, max_load - busiest_load_per_task);
3311 3311
3312 /* How much load to actually move to equalise the imbalance */ 3312 /* How much load to actually move to equalise the imbalance */
3313 *imbalance = min(max_pull * busiest->__cpu_power, 3313 *imbalance = min(max_pull * busiest->__cpu_power,
3314 (avg_load - this_load) * this->__cpu_power) 3314 (avg_load - this_load) * this->__cpu_power)
3315 / SCHED_LOAD_SCALE; 3315 / SCHED_LOAD_SCALE;
3316 3316
3317 /* 3317 /*
3318 * if *imbalance is less than the average load per runnable task 3318 * if *imbalance is less than the average load per runnable task
3319 * there is no gaurantee that any tasks will be moved so we'll have 3319 * there is no gaurantee that any tasks will be moved so we'll have
3320 * a think about bumping its value to force at least one task to be 3320 * a think about bumping its value to force at least one task to be
3321 * moved 3321 * moved
3322 */ 3322 */
3323 if (*imbalance < busiest_load_per_task) { 3323 if (*imbalance < busiest_load_per_task) {
3324 unsigned long tmp, pwr_now, pwr_move; 3324 unsigned long tmp, pwr_now, pwr_move;
3325 unsigned int imbn; 3325 unsigned int imbn;
3326 3326
3327 small_imbalance: 3327 small_imbalance:
3328 pwr_move = pwr_now = 0; 3328 pwr_move = pwr_now = 0;
3329 imbn = 2; 3329 imbn = 2;
3330 if (this_nr_running) { 3330 if (this_nr_running) {
3331 this_load_per_task /= this_nr_running; 3331 this_load_per_task /= this_nr_running;
3332 if (busiest_load_per_task > this_load_per_task) 3332 if (busiest_load_per_task > this_load_per_task)
3333 imbn = 1; 3333 imbn = 1;
3334 } else 3334 } else
3335 this_load_per_task = cpu_avg_load_per_task(this_cpu); 3335 this_load_per_task = cpu_avg_load_per_task(this_cpu);
3336 3336
3337 if (max_load - this_load + busiest_load_per_task >= 3337 if (max_load - this_load + busiest_load_per_task >=
3338 busiest_load_per_task * imbn) { 3338 busiest_load_per_task * imbn) {
3339 *imbalance = busiest_load_per_task; 3339 *imbalance = busiest_load_per_task;
3340 return busiest; 3340 return busiest;
3341 } 3341 }
3342 3342
3343 /* 3343 /*
3344 * OK, we don't have enough imbalance to justify moving tasks, 3344 * OK, we don't have enough imbalance to justify moving tasks,
3345 * however we may be able to increase total CPU power used by 3345 * however we may be able to increase total CPU power used by
3346 * moving them. 3346 * moving them.
3347 */ 3347 */
3348 3348
3349 pwr_now += busiest->__cpu_power * 3349 pwr_now += busiest->__cpu_power *
3350 min(busiest_load_per_task, max_load); 3350 min(busiest_load_per_task, max_load);
3351 pwr_now += this->__cpu_power * 3351 pwr_now += this->__cpu_power *
3352 min(this_load_per_task, this_load); 3352 min(this_load_per_task, this_load);
3353 pwr_now /= SCHED_LOAD_SCALE; 3353 pwr_now /= SCHED_LOAD_SCALE;
3354 3354
3355 /* Amount of load we'd subtract */ 3355 /* Amount of load we'd subtract */
3356 tmp = sg_div_cpu_power(busiest, 3356 tmp = sg_div_cpu_power(busiest,
3357 busiest_load_per_task * SCHED_LOAD_SCALE); 3357 busiest_load_per_task * SCHED_LOAD_SCALE);
3358 if (max_load > tmp) 3358 if (max_load > tmp)
3359 pwr_move += busiest->__cpu_power * 3359 pwr_move += busiest->__cpu_power *
3360 min(busiest_load_per_task, max_load - tmp); 3360 min(busiest_load_per_task, max_load - tmp);
3361 3361
3362 /* Amount of load we'd add */ 3362 /* Amount of load we'd add */
3363 if (max_load * busiest->__cpu_power < 3363 if (max_load * busiest->__cpu_power <
3364 busiest_load_per_task * SCHED_LOAD_SCALE) 3364 busiest_load_per_task * SCHED_LOAD_SCALE)
3365 tmp = sg_div_cpu_power(this, 3365 tmp = sg_div_cpu_power(this,
3366 max_load * busiest->__cpu_power); 3366 max_load * busiest->__cpu_power);
3367 else 3367 else
3368 tmp = sg_div_cpu_power(this, 3368 tmp = sg_div_cpu_power(this,
3369 busiest_load_per_task * SCHED_LOAD_SCALE); 3369 busiest_load_per_task * SCHED_LOAD_SCALE);
3370 pwr_move += this->__cpu_power * 3370 pwr_move += this->__cpu_power *
3371 min(this_load_per_task, this_load + tmp); 3371 min(this_load_per_task, this_load + tmp);
3372 pwr_move /= SCHED_LOAD_SCALE; 3372 pwr_move /= SCHED_LOAD_SCALE;
3373 3373
3374 /* Move if we gain throughput */ 3374 /* Move if we gain throughput */
3375 if (pwr_move > pwr_now) 3375 if (pwr_move > pwr_now)
3376 *imbalance = busiest_load_per_task; 3376 *imbalance = busiest_load_per_task;
3377 } 3377 }
3378 3378
3379 return busiest; 3379 return busiest;
3380 3380
3381 out_balanced: 3381 out_balanced:
3382 #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) 3382 #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT)
3383 if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE)) 3383 if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE))
3384 goto ret; 3384 goto ret;
3385 3385
3386 if (this == group_leader && group_leader != group_min) { 3386 if (this == group_leader && group_leader != group_min) {
3387 *imbalance = min_load_per_task; 3387 *imbalance = min_load_per_task;
3388 return group_min; 3388 return group_min;
3389 } 3389 }
3390 #endif 3390 #endif
3391 ret: 3391 ret:
3392 *imbalance = 0; 3392 *imbalance = 0;
3393 return NULL; 3393 return NULL;
3394 } 3394 }
3395 3395
3396 /* 3396 /*
3397 * find_busiest_queue - find the busiest runqueue among the cpus in group. 3397 * find_busiest_queue - find the busiest runqueue among the cpus in group.
3398 */ 3398 */
3399 static struct rq * 3399 static struct rq *
3400 find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle, 3400 find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle,
3401 unsigned long imbalance, const cpumask_t *cpus) 3401 unsigned long imbalance, const cpumask_t *cpus)
3402 { 3402 {
3403 struct rq *busiest = NULL, *rq; 3403 struct rq *busiest = NULL, *rq;
3404 unsigned long max_load = 0; 3404 unsigned long max_load = 0;
3405 int i; 3405 int i;
3406 3406
3407 for_each_cpu_mask_nr(i, group->cpumask) { 3407 for_each_cpu_mask_nr(i, group->cpumask) {
3408 unsigned long wl; 3408 unsigned long wl;
3409 3409
3410 if (!cpu_isset(i, *cpus)) 3410 if (!cpu_isset(i, *cpus))
3411 continue; 3411 continue;
3412 3412
3413 rq = cpu_rq(i); 3413 rq = cpu_rq(i);
3414 wl = weighted_cpuload(i); 3414 wl = weighted_cpuload(i);
3415 3415
3416 if (rq->nr_running == 1 && wl > imbalance) 3416 if (rq->nr_running == 1 && wl > imbalance)
3417 continue; 3417 continue;
3418 3418
3419 if (wl > max_load) { 3419 if (wl > max_load) {
3420 max_load = wl; 3420 max_load = wl;
3421 busiest = rq; 3421 busiest = rq;
3422 } 3422 }
3423 } 3423 }
3424 3424
3425 return busiest; 3425 return busiest;
3426 } 3426 }
3427 3427
3428 /* 3428 /*
3429 * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but 3429 * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but
3430 * so long as it is large enough. 3430 * so long as it is large enough.
3431 */ 3431 */
3432 #define MAX_PINNED_INTERVAL 512 3432 #define MAX_PINNED_INTERVAL 512
3433 3433
3434 /* 3434 /*
3435 * Check this_cpu to ensure it is balanced within domain. Attempt to move 3435 * Check this_cpu to ensure it is balanced within domain. Attempt to move
3436 * tasks if there is an imbalance. 3436 * tasks if there is an imbalance.
3437 */ 3437 */
3438 static int load_balance(int this_cpu, struct rq *this_rq, 3438 static int load_balance(int this_cpu, struct rq *this_rq,
3439 struct sched_domain *sd, enum cpu_idle_type idle, 3439 struct sched_domain *sd, enum cpu_idle_type idle,
3440 int *balance, cpumask_t *cpus) 3440 int *balance, cpumask_t *cpus)
3441 { 3441 {
3442 int ld_moved, all_pinned = 0, active_balance = 0, sd_idle = 0; 3442 int ld_moved, all_pinned = 0, active_balance = 0, sd_idle = 0;
3443 struct sched_group *group; 3443 struct sched_group *group;
3444 unsigned long imbalance; 3444 unsigned long imbalance;
3445 struct rq *busiest; 3445 struct rq *busiest;
3446 unsigned long flags; 3446 unsigned long flags;
3447 3447
3448 cpus_setall(*cpus); 3448 cpus_setall(*cpus);
3449 3449
3450 /* 3450 /*
3451 * When power savings policy is enabled for the parent domain, idle 3451 * When power savings policy is enabled for the parent domain, idle
3452 * sibling can pick up load irrespective of busy siblings. In this case, 3452 * sibling can pick up load irrespective of busy siblings. In this case,
3453 * let the state of idle sibling percolate up as CPU_IDLE, instead of 3453 * let the state of idle sibling percolate up as CPU_IDLE, instead of
3454 * portraying it as CPU_NOT_IDLE. 3454 * portraying it as CPU_NOT_IDLE.
3455 */ 3455 */
3456 if (idle != CPU_NOT_IDLE && sd->flags & SD_SHARE_CPUPOWER && 3456 if (idle != CPU_NOT_IDLE && sd->flags & SD_SHARE_CPUPOWER &&
3457 !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) 3457 !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE))
3458 sd_idle = 1; 3458 sd_idle = 1;
3459 3459
3460 schedstat_inc(sd, lb_count[idle]); 3460 schedstat_inc(sd, lb_count[idle]);
3461 3461
3462 redo: 3462 redo:
3463 update_shares(sd); 3463 update_shares(sd);
3464 group = find_busiest_group(sd, this_cpu, &imbalance, idle, &sd_idle, 3464 group = find_busiest_group(sd, this_cpu, &imbalance, idle, &sd_idle,
3465 cpus, balance); 3465 cpus, balance);
3466 3466
3467 if (*balance == 0) 3467 if (*balance == 0)
3468 goto out_balanced; 3468 goto out_balanced;
3469 3469
3470 if (!group) { 3470 if (!group) {
3471 schedstat_inc(sd, lb_nobusyg[idle]); 3471 schedstat_inc(sd, lb_nobusyg[idle]);
3472 goto out_balanced; 3472 goto out_balanced;
3473 } 3473 }
3474 3474
3475 busiest = find_busiest_queue(group, idle, imbalance, cpus); 3475 busiest = find_busiest_queue(group, idle, imbalance, cpus);
3476 if (!busiest) { 3476 if (!busiest) {
3477 schedstat_inc(sd, lb_nobusyq[idle]); 3477 schedstat_inc(sd, lb_nobusyq[idle]);
3478 goto out_balanced; 3478 goto out_balanced;
3479 } 3479 }
3480 3480
3481 BUG_ON(busiest == this_rq); 3481 BUG_ON(busiest == this_rq);
3482 3482
3483 schedstat_add(sd, lb_imbalance[idle], imbalance); 3483 schedstat_add(sd, lb_imbalance[idle], imbalance);
3484 3484
3485 ld_moved = 0; 3485 ld_moved = 0;
3486 if (busiest->nr_running > 1) { 3486 if (busiest->nr_running > 1) {
3487 /* 3487 /*
3488 * Attempt to move tasks. If find_busiest_group has found 3488 * Attempt to move tasks. If find_busiest_group has found
3489 * an imbalance but busiest->nr_running <= 1, the group is 3489 * an imbalance but busiest->nr_running <= 1, the group is
3490 * still unbalanced. ld_moved simply stays zero, so it is 3490 * still unbalanced. ld_moved simply stays zero, so it is
3491 * correctly treated as an imbalance. 3491 * correctly treated as an imbalance.
3492 */ 3492 */
3493 local_irq_save(flags); 3493 local_irq_save(flags);
3494 double_rq_lock(this_rq, busiest); 3494 double_rq_lock(this_rq, busiest);
3495 ld_moved = move_tasks(this_rq, this_cpu, busiest, 3495 ld_moved = move_tasks(this_rq, this_cpu, busiest,
3496 imbalance, sd, idle, &all_pinned); 3496 imbalance, sd, idle, &all_pinned);
3497 double_rq_unlock(this_rq, busiest); 3497 double_rq_unlock(this_rq, busiest);
3498 local_irq_restore(flags); 3498 local_irq_restore(flags);
3499 3499
3500 /* 3500 /*
3501 * some other cpu did the load balance for us. 3501 * some other cpu did the load balance for us.
3502 */ 3502 */
3503 if (ld_moved && this_cpu != smp_processor_id()) 3503 if (ld_moved && this_cpu != smp_processor_id())
3504 resched_cpu(this_cpu); 3504 resched_cpu(this_cpu);
3505 3505
3506 /* All tasks on this runqueue were pinned by CPU affinity */ 3506 /* All tasks on this runqueue were pinned by CPU affinity */
3507 if (unlikely(all_pinned)) { 3507 if (unlikely(all_pinned)) {
3508 cpu_clear(cpu_of(busiest), *cpus); 3508 cpu_clear(cpu_of(busiest), *cpus);
3509 if (!cpus_empty(*cpus)) 3509 if (!cpus_empty(*cpus))
3510 goto redo; 3510 goto redo;
3511 goto out_balanced; 3511 goto out_balanced;
3512 } 3512 }
3513 } 3513 }
3514 3514
3515 if (!ld_moved) { 3515 if (!ld_moved) {
3516 schedstat_inc(sd, lb_failed[idle]); 3516 schedstat_inc(sd, lb_failed[idle]);
3517 sd->nr_balance_failed++; 3517 sd->nr_balance_failed++;
3518 3518
3519 if (unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2)) { 3519 if (unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2)) {
3520 3520
3521 spin_lock_irqsave(&busiest->lock, flags); 3521 spin_lock_irqsave(&busiest->lock, flags);
3522 3522
3523 /* don't kick the migration_thread, if the curr 3523 /* don't kick the migration_thread, if the curr
3524 * task on busiest cpu can't be moved to this_cpu 3524 * task on busiest cpu can't be moved to this_cpu
3525 */ 3525 */
3526 if (!cpu_isset(this_cpu, busiest->curr->cpus_allowed)) { 3526 if (!cpu_isset(this_cpu, busiest->curr->cpus_allowed)) {
3527 spin_unlock_irqrestore(&busiest->lock, flags); 3527 spin_unlock_irqrestore(&busiest->lock, flags);
3528 all_pinned = 1; 3528 all_pinned = 1;
3529 goto out_one_pinned; 3529 goto out_one_pinned;
3530 } 3530 }
3531 3531
3532 if (!busiest->active_balance) { 3532 if (!busiest->active_balance) {
3533 busiest->active_balance = 1; 3533 busiest->active_balance = 1;
3534 busiest->push_cpu = this_cpu; 3534 busiest->push_cpu = this_cpu;
3535 active_balance = 1; 3535 active_balance = 1;
3536 } 3536 }
3537 spin_unlock_irqrestore(&busiest->lock, flags); 3537 spin_unlock_irqrestore(&busiest->lock, flags);
3538 if (active_balance) 3538 if (active_balance)
3539 wake_up_process(busiest->migration_thread); 3539 wake_up_process(busiest->migration_thread);
3540 3540
3541 /* 3541 /*
3542 * We've kicked active balancing, reset the failure 3542 * We've kicked active balancing, reset the failure
3543 * counter. 3543 * counter.
3544 */ 3544 */
3545 sd->nr_balance_failed = sd->cache_nice_tries+1; 3545 sd->nr_balance_failed = sd->cache_nice_tries+1;
3546 } 3546 }
3547 } else 3547 } else
3548 sd->nr_balance_failed = 0; 3548 sd->nr_balance_failed = 0;
3549 3549
3550 if (likely(!active_balance)) { 3550 if (likely(!active_balance)) {
3551 /* We were unbalanced, so reset the balancing interval */ 3551 /* We were unbalanced, so reset the balancing interval */
3552 sd->balance_interval = sd->min_interval; 3552 sd->balance_interval = sd->min_interval;
3553 } else { 3553 } else {
3554 /* 3554 /*
3555 * If we've begun active balancing, start to back off. This 3555 * If we've begun active balancing, start to back off. This
3556 * case may not be covered by the all_pinned logic if there 3556 * case may not be covered by the all_pinned logic if there
3557 * is only 1 task on the busy runqueue (because we don't call 3557 * is only 1 task on the busy runqueue (because we don't call
3558 * move_tasks). 3558 * move_tasks).
3559 */ 3559 */
3560 if (sd->balance_interval < sd->max_interval) 3560 if (sd->balance_interval < sd->max_interval)
3561 sd->balance_interval *= 2; 3561 sd->balance_interval *= 2;
3562 } 3562 }
3563 3563
3564 if (!ld_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER && 3564 if (!ld_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER &&
3565 !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) 3565 !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE))
3566 ld_moved = -1; 3566 ld_moved = -1;
3567 3567
3568 goto out; 3568 goto out;
3569 3569
3570 out_balanced: 3570 out_balanced:
3571 schedstat_inc(sd, lb_balanced[idle]); 3571 schedstat_inc(sd, lb_balanced[idle]);
3572 3572
3573 sd->nr_balance_failed = 0; 3573 sd->nr_balance_failed = 0;
3574 3574
3575 out_one_pinned: 3575 out_one_pinned:
3576 /* tune up the balancing interval */ 3576 /* tune up the balancing interval */
3577 if ((all_pinned && sd->balance_interval < MAX_PINNED_INTERVAL) || 3577 if ((all_pinned && sd->balance_interval < MAX_PINNED_INTERVAL) ||
3578 (sd->balance_interval < sd->max_interval)) 3578 (sd->balance_interval < sd->max_interval))
3579 sd->balance_interval *= 2; 3579 sd->balance_interval *= 2;
3580 3580
3581 if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && 3581 if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER &&
3582 !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) 3582 !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE))
3583 ld_moved = -1; 3583 ld_moved = -1;
3584 else 3584 else
3585 ld_moved = 0; 3585 ld_moved = 0;
3586 out: 3586 out:
3587 if (ld_moved) 3587 if (ld_moved)
3588 update_shares(sd); 3588 update_shares(sd);
3589 return ld_moved; 3589 return ld_moved;
3590 } 3590 }
3591 3591
3592 /* 3592 /*
3593 * Check this_cpu to ensure it is balanced within domain. Attempt to move 3593 * Check this_cpu to ensure it is balanced within domain. Attempt to move
3594 * tasks if there is an imbalance. 3594 * tasks if there is an imbalance.
3595 * 3595 *
3596 * Called from schedule when this_rq is about to become idle (CPU_NEWLY_IDLE). 3596 * Called from schedule when this_rq is about to become idle (CPU_NEWLY_IDLE).
3597 * this_rq is locked. 3597 * this_rq is locked.
3598 */ 3598 */
3599 static int 3599 static int
3600 load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd, 3600 load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd,
3601 cpumask_t *cpus) 3601 cpumask_t *cpus)
3602 { 3602 {
3603 struct sched_group *group; 3603 struct sched_group *group;
3604 struct rq *busiest = NULL; 3604 struct rq *busiest = NULL;
3605 unsigned long imbalance; 3605 unsigned long imbalance;
3606 int ld_moved = 0; 3606 int ld_moved = 0;
3607 int sd_idle = 0; 3607 int sd_idle = 0;
3608 int all_pinned = 0; 3608 int all_pinned = 0;
3609 3609
3610 cpus_setall(*cpus); 3610 cpus_setall(*cpus);
3611 3611
3612 /* 3612 /*
3613 * When power savings policy is enabled for the parent domain, idle 3613 * When power savings policy is enabled for the parent domain, idle
3614 * sibling can pick up load irrespective of busy siblings. In this case, 3614 * sibling can pick up load irrespective of busy siblings. In this case,
3615 * let the state of idle sibling percolate up as IDLE, instead of 3615 * let the state of idle sibling percolate up as IDLE, instead of
3616 * portraying it as CPU_NOT_IDLE. 3616 * portraying it as CPU_NOT_IDLE.
3617 */ 3617 */
3618 if (sd->flags & SD_SHARE_CPUPOWER && 3618 if (sd->flags & SD_SHARE_CPUPOWER &&
3619 !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) 3619 !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE))
3620 sd_idle = 1; 3620 sd_idle = 1;
3621 3621
3622 schedstat_inc(sd, lb_count[CPU_NEWLY_IDLE]); 3622 schedstat_inc(sd, lb_count[CPU_NEWLY_IDLE]);
3623 redo: 3623 redo:
3624 update_shares_locked(this_rq, sd); 3624 update_shares_locked(this_rq, sd);
3625 group = find_busiest_group(sd, this_cpu, &imbalance, CPU_NEWLY_IDLE, 3625 group = find_busiest_group(sd, this_cpu, &imbalance, CPU_NEWLY_IDLE,
3626 &sd_idle, cpus, NULL); 3626 &sd_idle, cpus, NULL);
3627 if (!group) { 3627 if (!group) {
3628 schedstat_inc(sd, lb_nobusyg[CPU_NEWLY_IDLE]); 3628 schedstat_inc(sd, lb_nobusyg[CPU_NEWLY_IDLE]);
3629 goto out_balanced; 3629 goto out_balanced;
3630 } 3630 }
3631 3631
3632 busiest = find_busiest_queue(group, CPU_NEWLY_IDLE, imbalance, cpus); 3632 busiest = find_busiest_queue(group, CPU_NEWLY_IDLE, imbalance, cpus);
3633 if (!busiest) { 3633 if (!busiest) {
3634 schedstat_inc(sd, lb_nobusyq[CPU_NEWLY_IDLE]); 3634 schedstat_inc(sd, lb_nobusyq[CPU_NEWLY_IDLE]);
3635 goto out_balanced; 3635 goto out_balanced;
3636 } 3636 }
3637 3637
3638 BUG_ON(busiest == this_rq); 3638 BUG_ON(busiest == this_rq);
3639 3639
3640 schedstat_add(sd, lb_imbalance[CPU_NEWLY_IDLE], imbalance); 3640 schedstat_add(sd, lb_imbalance[CPU_NEWLY_IDLE], imbalance);
3641 3641
3642 ld_moved = 0; 3642 ld_moved = 0;
3643 if (busiest->nr_running > 1) { 3643 if (busiest->nr_running > 1) {
3644 /* Attempt to move tasks */ 3644 /* Attempt to move tasks */
3645 double_lock_balance(this_rq, busiest); 3645 double_lock_balance(this_rq, busiest);
3646 /* this_rq->clock is already updated */ 3646 /* this_rq->clock is already updated */
3647 update_rq_clock(busiest); 3647 update_rq_clock(busiest);
3648 ld_moved = move_tasks(this_rq, this_cpu, busiest, 3648 ld_moved = move_tasks(this_rq, this_cpu, busiest,
3649 imbalance, sd, CPU_NEWLY_IDLE, 3649 imbalance, sd, CPU_NEWLY_IDLE,
3650 &all_pinned); 3650 &all_pinned);
3651 double_unlock_balance(this_rq, busiest); 3651 double_unlock_balance(this_rq, busiest);
3652 3652
3653 if (unlikely(all_pinned)) { 3653 if (unlikely(all_pinned)) {
3654 cpu_clear(cpu_of(busiest), *cpus); 3654 cpu_clear(cpu_of(busiest), *cpus);
3655 if (!cpus_empty(*cpus)) 3655 if (!cpus_empty(*cpus))
3656 goto redo; 3656 goto redo;
3657 } 3657 }
3658 } 3658 }
3659 3659
3660 if (!ld_moved) { 3660 if (!ld_moved) {
3661 schedstat_inc(sd, lb_failed[CPU_NEWLY_IDLE]); 3661 schedstat_inc(sd, lb_failed[CPU_NEWLY_IDLE]);
3662 if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && 3662 if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER &&
3663 !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) 3663 !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE))
3664 return -1; 3664 return -1;
3665 } else 3665 } else
3666 sd->nr_balance_failed = 0; 3666 sd->nr_balance_failed = 0;
3667 3667
3668 update_shares_locked(this_rq, sd); 3668 update_shares_locked(this_rq, sd);
3669 return ld_moved; 3669 return ld_moved;
3670 3670
3671 out_balanced: 3671 out_balanced:
3672 schedstat_inc(sd, lb_balanced[CPU_NEWLY_IDLE]); 3672 schedstat_inc(sd, lb_balanced[CPU_NEWLY_IDLE]);
3673 if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && 3673 if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER &&
3674 !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) 3674 !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE))
3675 return -1; 3675 return -1;
3676 sd->nr_balance_failed = 0; 3676 sd->nr_balance_failed = 0;
3677 3677
3678 return 0; 3678 return 0;
3679 } 3679 }
3680 3680
3681 /* 3681 /*
3682 * idle_balance is called by schedule() if this_cpu is about to become 3682 * idle_balance is called by schedule() if this_cpu is about to become
3683 * idle. Attempts to pull tasks from other CPUs. 3683 * idle. Attempts to pull tasks from other CPUs.
3684 */ 3684 */
3685 static void idle_balance(int this_cpu, struct rq *this_rq) 3685 static void idle_balance(int this_cpu, struct rq *this_rq)
3686 { 3686 {
3687 struct sched_domain *sd; 3687 struct sched_domain *sd;
3688 int pulled_task = 0; 3688 int pulled_task = 0;
3689 unsigned long next_balance = jiffies + HZ; 3689 unsigned long next_balance = jiffies + HZ;
3690 cpumask_t tmpmask; 3690 cpumask_t tmpmask;
3691 3691
3692 for_each_domain(this_cpu, sd) { 3692 for_each_domain(this_cpu, sd) {
3693 unsigned long interval; 3693 unsigned long interval;
3694 3694
3695 if (!(sd->flags & SD_LOAD_BALANCE)) 3695 if (!(sd->flags & SD_LOAD_BALANCE))
3696 continue; 3696 continue;
3697 3697
3698 if (sd->flags & SD_BALANCE_NEWIDLE) 3698 if (sd->flags & SD_BALANCE_NEWIDLE)
3699 /* If we've pulled tasks over stop searching: */ 3699 /* If we've pulled tasks over stop searching: */
3700 pulled_task = load_balance_newidle(this_cpu, this_rq, 3700 pulled_task = load_balance_newidle(this_cpu, this_rq,
3701 sd, &tmpmask); 3701 sd, &tmpmask);
3702 3702
3703 interval = msecs_to_jiffies(sd->balance_interval); 3703 interval = msecs_to_jiffies(sd->balance_interval);
3704 if (time_after(next_balance, sd->last_balance + interval)) 3704 if (time_after(next_balance, sd->last_balance + interval))
3705 next_balance = sd->last_balance + interval; 3705 next_balance = sd->last_balance + interval;
3706 if (pulled_task) 3706 if (pulled_task)
3707 break; 3707 break;
3708 } 3708 }
3709 if (pulled_task || time_after(jiffies, this_rq->next_balance)) { 3709 if (pulled_task || time_after(jiffies, this_rq->next_balance)) {
3710 /* 3710 /*
3711 * We are going idle. next_balance may be set based on 3711 * We are going idle. next_balance may be set based on
3712 * a busy processor. So reset next_balance. 3712 * a busy processor. So reset next_balance.
3713 */ 3713 */
3714 this_rq->next_balance = next_balance; 3714 this_rq->next_balance = next_balance;
3715 } 3715 }
3716 } 3716 }
3717 3717
3718 /* 3718 /*
3719 * active_load_balance is run by migration threads. It pushes running tasks 3719 * active_load_balance is run by migration threads. It pushes running tasks
3720 * off the busiest CPU onto idle CPUs. It requires at least 1 task to be 3720 * off the busiest CPU onto idle CPUs. It requires at least 1 task to be
3721 * running on each physical CPU where possible, and avoids physical / 3721 * running on each physical CPU where possible, and avoids physical /
3722 * logical imbalances. 3722 * logical imbalances.
3723 * 3723 *
3724 * Called with busiest_rq locked. 3724 * Called with busiest_rq locked.
3725 */ 3725 */
3726 static void active_load_balance(struct rq *busiest_rq, int busiest_cpu) 3726 static void active_load_balance(struct rq *busiest_rq, int busiest_cpu)
3727 { 3727 {
3728 int target_cpu = busiest_rq->push_cpu; 3728 int target_cpu = busiest_rq->push_cpu;
3729 struct sched_domain *sd; 3729 struct sched_domain *sd;
3730 struct rq *target_rq; 3730 struct rq *target_rq;
3731 3731
3732 /* Is there any task to move? */ 3732 /* Is there any task to move? */
3733 if (busiest_rq->nr_running <= 1) 3733 if (busiest_rq->nr_running <= 1)
3734 return; 3734 return;
3735 3735
3736 target_rq = cpu_rq(target_cpu); 3736 target_rq = cpu_rq(target_cpu);
3737 3737
3738 /* 3738 /*
3739 * This condition is "impossible", if it occurs 3739 * This condition is "impossible", if it occurs
3740 * we need to fix it. Originally reported by 3740 * we need to fix it. Originally reported by
3741 * Bjorn Helgaas on a 128-cpu setup. 3741 * Bjorn Helgaas on a 128-cpu setup.
3742 */ 3742 */
3743 BUG_ON(busiest_rq == target_rq); 3743 BUG_ON(busiest_rq == target_rq);
3744 3744
3745 /* move a task from busiest_rq to target_rq */ 3745 /* move a task from busiest_rq to target_rq */
3746 double_lock_balance(busiest_rq, target_rq); 3746 double_lock_balance(busiest_rq, target_rq);
3747 update_rq_clock(busiest_rq); 3747 update_rq_clock(busiest_rq);
3748 update_rq_clock(target_rq); 3748 update_rq_clock(target_rq);
3749 3749
3750 /* Search for an sd spanning us and the target CPU. */ 3750 /* Search for an sd spanning us and the target CPU. */
3751 for_each_domain(target_cpu, sd) { 3751 for_each_domain(target_cpu, sd) {
3752 if ((sd->flags & SD_LOAD_BALANCE) && 3752 if ((sd->flags & SD_LOAD_BALANCE) &&
3753 cpu_isset(busiest_cpu, sd->span)) 3753 cpu_isset(busiest_cpu, sd->span))
3754 break; 3754 break;
3755 } 3755 }
3756 3756
3757 if (likely(sd)) { 3757 if (likely(sd)) {
3758 schedstat_inc(sd, alb_count); 3758 schedstat_inc(sd, alb_count);
3759 3759
3760 if (move_one_task(target_rq, target_cpu, busiest_rq, 3760 if (move_one_task(target_rq, target_cpu, busiest_rq,
3761 sd, CPU_IDLE)) 3761 sd, CPU_IDLE))
3762 schedstat_inc(sd, alb_pushed); 3762 schedstat_inc(sd, alb_pushed);
3763 else 3763 else
3764 schedstat_inc(sd, alb_failed); 3764 schedstat_inc(sd, alb_failed);
3765 } 3765 }
3766 double_unlock_balance(busiest_rq, target_rq); 3766 double_unlock_balance(busiest_rq, target_rq);
3767 } 3767 }
3768 3768
3769 #ifdef CONFIG_NO_HZ 3769 #ifdef CONFIG_NO_HZ
3770 static struct { 3770 static struct {
3771 atomic_t load_balancer; 3771 atomic_t load_balancer;
3772 cpumask_t cpu_mask; 3772 cpumask_t cpu_mask;
3773 } nohz ____cacheline_aligned = { 3773 } nohz ____cacheline_aligned = {
3774 .load_balancer = ATOMIC_INIT(-1), 3774 .load_balancer = ATOMIC_INIT(-1),
3775 .cpu_mask = CPU_MASK_NONE, 3775 .cpu_mask = CPU_MASK_NONE,
3776 }; 3776 };
3777 3777
3778 /* 3778 /*
3779 * This routine will try to nominate the ilb (idle load balancing) 3779 * This routine will try to nominate the ilb (idle load balancing)
3780 * owner among the cpus whose ticks are stopped. ilb owner will do the idle 3780 * owner among the cpus whose ticks are stopped. ilb owner will do the idle
3781 * load balancing on behalf of all those cpus. If all the cpus in the system 3781 * load balancing on behalf of all those cpus. If all the cpus in the system
3782 * go into this tickless mode, then there will be no ilb owner (as there is 3782 * go into this tickless mode, then there will be no ilb owner (as there is
3783 * no need for one) and all the cpus will sleep till the next wakeup event 3783 * no need for one) and all the cpus will sleep till the next wakeup event
3784 * arrives... 3784 * arrives...
3785 * 3785 *
3786 * For the ilb owner, tick is not stopped. And this tick will be used 3786 * For the ilb owner, tick is not stopped. And this tick will be used
3787 * for idle load balancing. ilb owner will still be part of 3787 * for idle load balancing. ilb owner will still be part of
3788 * nohz.cpu_mask.. 3788 * nohz.cpu_mask..
3789 * 3789 *
3790 * While stopping the tick, this cpu will become the ilb owner if there 3790 * While stopping the tick, this cpu will become the ilb owner if there
3791 * is no other owner. And will be the owner till that cpu becomes busy 3791 * is no other owner. And will be the owner till that cpu becomes busy
3792 * or if all cpus in the system stop their ticks at which point 3792 * or if all cpus in the system stop their ticks at which point
3793 * there is no need for ilb owner. 3793 * there is no need for ilb owner.
3794 * 3794 *
3795 * When the ilb owner becomes busy, it nominates another owner, during the 3795 * When the ilb owner becomes busy, it nominates another owner, during the
3796 * next busy scheduler_tick() 3796 * next busy scheduler_tick()
3797 */ 3797 */
3798 int select_nohz_load_balancer(int stop_tick) 3798 int select_nohz_load_balancer(int stop_tick)
3799 { 3799 {
3800 int cpu = smp_processor_id(); 3800 int cpu = smp_processor_id();
3801 3801
3802 if (stop_tick) { 3802 if (stop_tick) {
3803 cpu_set(cpu, nohz.cpu_mask); 3803 cpu_set(cpu, nohz.cpu_mask);
3804 cpu_rq(cpu)->in_nohz_recently = 1; 3804 cpu_rq(cpu)->in_nohz_recently = 1;
3805 3805
3806 /* 3806 /*
3807 * If we are going offline and still the leader, give up! 3807 * If we are going offline and still the leader, give up!
3808 */ 3808 */
3809 if (!cpu_active(cpu) && 3809 if (!cpu_active(cpu) &&
3810 atomic_read(&nohz.load_balancer) == cpu) { 3810 atomic_read(&nohz.load_balancer) == cpu) {
3811 if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) 3811 if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu)
3812 BUG(); 3812 BUG();
3813 return 0; 3813 return 0;
3814 } 3814 }
3815 3815
3816 /* time for ilb owner also to sleep */ 3816 /* time for ilb owner also to sleep */
3817 if (cpus_weight(nohz.cpu_mask) == num_online_cpus()) { 3817 if (cpus_weight(nohz.cpu_mask) == num_online_cpus()) {
3818 if (atomic_read(&nohz.load_balancer) == cpu) 3818 if (atomic_read(&nohz.load_balancer) == cpu)
3819 atomic_set(&nohz.load_balancer, -1); 3819 atomic_set(&nohz.load_balancer, -1);
3820 return 0; 3820 return 0;
3821 } 3821 }
3822 3822
3823 if (atomic_read(&nohz.load_balancer) == -1) { 3823 if (atomic_read(&nohz.load_balancer) == -1) {
3824 /* make me the ilb owner */ 3824 /* make me the ilb owner */
3825 if (atomic_cmpxchg(&nohz.load_balancer, -1, cpu) == -1) 3825 if (atomic_cmpxchg(&nohz.load_balancer, -1, cpu) == -1)
3826 return 1; 3826 return 1;
3827 } else if (atomic_read(&nohz.load_balancer) == cpu) 3827 } else if (atomic_read(&nohz.load_balancer) == cpu)
3828 return 1; 3828 return 1;
3829 } else { 3829 } else {
3830 if (!cpu_isset(cpu, nohz.cpu_mask)) 3830 if (!cpu_isset(cpu, nohz.cpu_mask))
3831 return 0; 3831 return 0;
3832 3832
3833 cpu_clear(cpu, nohz.cpu_mask); 3833 cpu_clear(cpu, nohz.cpu_mask);
3834 3834
3835 if (atomic_read(&nohz.load_balancer) == cpu) 3835 if (atomic_read(&nohz.load_balancer) == cpu)
3836 if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) 3836 if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu)
3837 BUG(); 3837 BUG();
3838 } 3838 }
3839 return 0; 3839 return 0;
3840 } 3840 }
3841 #endif 3841 #endif
3842 3842
3843 static DEFINE_SPINLOCK(balancing); 3843 static DEFINE_SPINLOCK(balancing);
3844 3844
3845 /* 3845 /*
3846 * It checks each scheduling domain to see if it is due to be balanced, 3846 * It checks each scheduling domain to see if it is due to be balanced,
3847 * and initiates a balancing operation if so. 3847 * and initiates a balancing operation if so.
3848 * 3848 *
3849 * Balancing parameters are set up in arch_init_sched_domains. 3849 * Balancing parameters are set up in arch_init_sched_domains.
3850 */ 3850 */
3851 static void rebalance_domains(int cpu, enum cpu_idle_type idle) 3851 static void rebalance_domains(int cpu, enum cpu_idle_type idle)
3852 { 3852 {
3853 int balance = 1; 3853 int balance = 1;
3854 struct rq *rq = cpu_rq(cpu); 3854 struct rq *rq = cpu_rq(cpu);
3855 unsigned long interval; 3855 unsigned long interval;
3856 struct sched_domain *sd; 3856 struct sched_domain *sd;
3857 /* Earliest time when we have to do rebalance again */ 3857 /* Earliest time when we have to do rebalance again */
3858 unsigned long next_balance = jiffies + 60*HZ; 3858 unsigned long next_balance = jiffies + 60*HZ;
3859 int update_next_balance = 0; 3859 int update_next_balance = 0;
3860 int need_serialize; 3860 int need_serialize;
3861 cpumask_t tmp; 3861 cpumask_t tmp;
3862 3862
3863 for_each_domain(cpu, sd) { 3863 for_each_domain(cpu, sd) {
3864 if (!(sd->flags & SD_LOAD_BALANCE)) 3864 if (!(sd->flags & SD_LOAD_BALANCE))
3865 continue; 3865 continue;
3866 3866
3867 interval = sd->balance_interval; 3867 interval = sd->balance_interval;
3868 if (idle != CPU_IDLE) 3868 if (idle != CPU_IDLE)
3869 interval *= sd->busy_factor; 3869 interval *= sd->busy_factor;
3870 3870
3871 /* scale ms to jiffies */ 3871 /* scale ms to jiffies */
3872 interval = msecs_to_jiffies(interval); 3872 interval = msecs_to_jiffies(interval);
3873 if (unlikely(!interval)) 3873 if (unlikely(!interval))
3874 interval = 1; 3874 interval = 1;
3875 if (interval > HZ*NR_CPUS/10) 3875 if (interval > HZ*NR_CPUS/10)
3876 interval = HZ*NR_CPUS/10; 3876 interval = HZ*NR_CPUS/10;
3877 3877
3878 need_serialize = sd->flags & SD_SERIALIZE; 3878 need_serialize = sd->flags & SD_SERIALIZE;
3879 3879
3880 if (need_serialize) { 3880 if (need_serialize) {
3881 if (!spin_trylock(&balancing)) 3881 if (!spin_trylock(&balancing))
3882 goto out; 3882 goto out;
3883 } 3883 }
3884 3884
3885 if (time_after_eq(jiffies, sd->last_balance + interval)) { 3885 if (time_after_eq(jiffies, sd->last_balance + interval)) {
3886 if (load_balance(cpu, rq, sd, idle, &balance, &tmp)) { 3886 if (load_balance(cpu, rq, sd, idle, &balance, &tmp)) {
3887 /* 3887 /*
3888 * We've pulled tasks over so either we're no 3888 * We've pulled tasks over so either we're no
3889 * longer idle, or one of our SMT siblings is 3889 * longer idle, or one of our SMT siblings is
3890 * not idle. 3890 * not idle.
3891 */ 3891 */
3892 idle = CPU_NOT_IDLE; 3892 idle = CPU_NOT_IDLE;
3893 } 3893 }
3894 sd->last_balance = jiffies; 3894 sd->last_balance = jiffies;
3895 } 3895 }
3896 if (need_serialize) 3896 if (need_serialize)
3897 spin_unlock(&balancing); 3897 spin_unlock(&balancing);
3898 out: 3898 out:
3899 if (time_after(next_balance, sd->last_balance + interval)) { 3899 if (time_after(next_balance, sd->last_balance + interval)) {
3900 next_balance = sd->last_balance + interval; 3900 next_balance = sd->last_balance + interval;
3901 update_next_balance = 1; 3901 update_next_balance = 1;
3902 } 3902 }
3903 3903
3904 /* 3904 /*
3905 * Stop the load balance at this level. There is another 3905 * Stop the load balance at this level. There is another
3906 * CPU in our sched group which is doing load balancing more 3906 * CPU in our sched group which is doing load balancing more
3907 * actively. 3907 * actively.
3908 */ 3908 */
3909 if (!balance) 3909 if (!balance)
3910 break; 3910 break;
3911 } 3911 }
3912 3912
3913 /* 3913 /*
3914 * next_balance will be updated only when there is a need. 3914 * next_balance will be updated only when there is a need.
3915 * When the cpu is attached to null domain for ex, it will not be 3915 * When the cpu is attached to null domain for ex, it will not be
3916 * updated. 3916 * updated.
3917 */ 3917 */
3918 if (likely(update_next_balance)) 3918 if (likely(update_next_balance))
3919 rq->next_balance = next_balance; 3919 rq->next_balance = next_balance;
3920 } 3920 }
3921 3921
3922 /* 3922 /*
3923 * run_rebalance_domains is triggered when needed from the scheduler tick. 3923 * run_rebalance_domains is triggered when needed from the scheduler tick.
3924 * In CONFIG_NO_HZ case, the idle load balance owner will do the 3924 * In CONFIG_NO_HZ case, the idle load balance owner will do the
3925 * rebalancing for all the cpus for whom scheduler ticks are stopped. 3925 * rebalancing for all the cpus for whom scheduler ticks are stopped.
3926 */ 3926 */
3927 static void run_rebalance_domains(struct softirq_action *h) 3927 static void run_rebalance_domains(struct softirq_action *h)
3928 { 3928 {
3929 int this_cpu = smp_processor_id(); 3929 int this_cpu = smp_processor_id();
3930 struct rq *this_rq = cpu_rq(this_cpu); 3930 struct rq *this_rq = cpu_rq(this_cpu);
3931 enum cpu_idle_type idle = this_rq->idle_at_tick ? 3931 enum cpu_idle_type idle = this_rq->idle_at_tick ?
3932 CPU_IDLE : CPU_NOT_IDLE; 3932 CPU_IDLE : CPU_NOT_IDLE;
3933 3933
3934 rebalance_domains(this_cpu, idle); 3934 rebalance_domains(this_cpu, idle);
3935 3935
3936 #ifdef CONFIG_NO_HZ 3936 #ifdef CONFIG_NO_HZ
3937 /* 3937 /*
3938 * If this cpu is the owner for idle load balancing, then do the 3938 * If this cpu is the owner for idle load balancing, then do the
3939 * balancing on behalf of the other idle cpus whose ticks are 3939 * balancing on behalf of the other idle cpus whose ticks are
3940 * stopped. 3940 * stopped.
3941 */ 3941 */
3942 if (this_rq->idle_at_tick && 3942 if (this_rq->idle_at_tick &&
3943 atomic_read(&nohz.load_balancer) == this_cpu) { 3943 atomic_read(&nohz.load_balancer) == this_cpu) {
3944 cpumask_t cpus = nohz.cpu_mask; 3944 cpumask_t cpus = nohz.cpu_mask;
3945 struct rq *rq; 3945 struct rq *rq;
3946 int balance_cpu; 3946 int balance_cpu;
3947 3947
3948 cpu_clear(this_cpu, cpus); 3948 cpu_clear(this_cpu, cpus);
3949 for_each_cpu_mask_nr(balance_cpu, cpus) { 3949 for_each_cpu_mask_nr(balance_cpu, cpus) {
3950 /* 3950 /*
3951 * If this cpu gets work to do, stop the load balancing 3951 * If this cpu gets work to do, stop the load balancing
3952 * work being done for other cpus. Next load 3952 * work being done for other cpus. Next load
3953 * balancing owner will pick it up. 3953 * balancing owner will pick it up.
3954 */ 3954 */
3955 if (need_resched()) 3955 if (need_resched())
3956 break; 3956 break;
3957 3957
3958 rebalance_domains(balance_cpu, CPU_IDLE); 3958 rebalance_domains(balance_cpu, CPU_IDLE);
3959 3959
3960 rq = cpu_rq(balance_cpu); 3960 rq = cpu_rq(balance_cpu);
3961 if (time_after(this_rq->next_balance, rq->next_balance)) 3961 if (time_after(this_rq->next_balance, rq->next_balance))
3962 this_rq->next_balance = rq->next_balance; 3962 this_rq->next_balance = rq->next_balance;
3963 } 3963 }
3964 } 3964 }
3965 #endif 3965 #endif
3966 } 3966 }
3967 3967
3968 /* 3968 /*
3969 * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing. 3969 * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing.
3970 * 3970 *
3971 * In case of CONFIG_NO_HZ, this is the place where we nominate a new 3971 * In case of CONFIG_NO_HZ, this is the place where we nominate a new
3972 * idle load balancing owner or decide to stop the periodic load balancing, 3972 * idle load balancing owner or decide to stop the periodic load balancing,
3973 * if the whole system is idle. 3973 * if the whole system is idle.
3974 */ 3974 */
3975 static inline void trigger_load_balance(struct rq *rq, int cpu) 3975 static inline void trigger_load_balance(struct rq *rq, int cpu)
3976 { 3976 {
3977 #ifdef CONFIG_NO_HZ 3977 #ifdef CONFIG_NO_HZ
3978 /* 3978 /*
3979 * If we were in the nohz mode recently and busy at the current 3979 * If we were in the nohz mode recently and busy at the current
3980 * scheduler tick, then check if we need to nominate new idle 3980 * scheduler tick, then check if we need to nominate new idle
3981 * load balancer. 3981 * load balancer.
3982 */ 3982 */
3983 if (rq->in_nohz_recently && !rq->idle_at_tick) { 3983 if (rq->in_nohz_recently && !rq->idle_at_tick) {
3984 rq->in_nohz_recently = 0; 3984 rq->in_nohz_recently = 0;
3985 3985
3986 if (atomic_read(&nohz.load_balancer) == cpu) { 3986 if (atomic_read(&nohz.load_balancer) == cpu) {
3987 cpu_clear(cpu, nohz.cpu_mask); 3987 cpu_clear(cpu, nohz.cpu_mask);
3988 atomic_set(&nohz.load_balancer, -1); 3988 atomic_set(&nohz.load_balancer, -1);
3989 } 3989 }
3990 3990
3991 if (atomic_read(&nohz.load_balancer) == -1) { 3991 if (atomic_read(&nohz.load_balancer) == -1) {
3992 /* 3992 /*
3993 * simple selection for now: Nominate the 3993 * simple selection for now: Nominate the
3994 * first cpu in the nohz list to be the next 3994 * first cpu in the nohz list to be the next
3995 * ilb owner. 3995 * ilb owner.
3996 * 3996 *
3997 * TBD: Traverse the sched domains and nominate 3997 * TBD: Traverse the sched domains and nominate
3998 * the nearest cpu in the nohz.cpu_mask. 3998 * the nearest cpu in the nohz.cpu_mask.
3999 */ 3999 */
4000 int ilb = first_cpu(nohz.cpu_mask); 4000 int ilb = first_cpu(nohz.cpu_mask);
4001 4001
4002 if (ilb < nr_cpu_ids) 4002 if (ilb < nr_cpu_ids)
4003 resched_cpu(ilb); 4003 resched_cpu(ilb);
4004 } 4004 }
4005 } 4005 }
4006 4006
4007 /* 4007 /*
4008 * If this cpu is idle and doing idle load balancing for all the 4008 * If this cpu is idle and doing idle load balancing for all the
4009 * cpus with ticks stopped, is it time for that to stop? 4009 * cpus with ticks stopped, is it time for that to stop?
4010 */ 4010 */
4011 if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) == cpu && 4011 if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) == cpu &&
4012 cpus_weight(nohz.cpu_mask) == num_online_cpus()) { 4012 cpus_weight(nohz.cpu_mask) == num_online_cpus()) {
4013 resched_cpu(cpu); 4013 resched_cpu(cpu);
4014 return; 4014 return;
4015 } 4015 }
4016 4016
4017 /* 4017 /*
4018 * If this cpu is idle and the idle load balancing is done by 4018 * If this cpu is idle and the idle load balancing is done by
4019 * someone else, then no need raise the SCHED_SOFTIRQ 4019 * someone else, then no need raise the SCHED_SOFTIRQ
4020 */ 4020 */
4021 if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) != cpu && 4021 if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) != cpu &&
4022 cpu_isset(cpu, nohz.cpu_mask)) 4022 cpu_isset(cpu, nohz.cpu_mask))
4023 return; 4023 return;
4024 #endif 4024 #endif
4025 if (time_after_eq(jiffies, rq->next_balance)) 4025 if (time_after_eq(jiffies, rq->next_balance))
4026 raise_softirq(SCHED_SOFTIRQ); 4026 raise_softirq(SCHED_SOFTIRQ);
4027 } 4027 }
4028 4028
4029 #else /* CONFIG_SMP */ 4029 #else /* CONFIG_SMP */
4030 4030
4031 /* 4031 /*
4032 * on UP we do not need to balance between CPUs: 4032 * on UP we do not need to balance between CPUs:
4033 */ 4033 */
4034 static inline void idle_balance(int cpu, struct rq *rq) 4034 static inline void idle_balance(int cpu, struct rq *rq)
4035 { 4035 {
4036 } 4036 }
4037 4037
4038 #endif 4038 #endif
4039 4039
4040 DEFINE_PER_CPU(struct kernel_stat, kstat); 4040 DEFINE_PER_CPU(struct kernel_stat, kstat);
4041 4041
4042 EXPORT_PER_CPU_SYMBOL(kstat); 4042 EXPORT_PER_CPU_SYMBOL(kstat);
4043 4043
4044 /* 4044 /*
4045 * Return any ns on the sched_clock that have not yet been banked in 4045 * Return any ns on the sched_clock that have not yet been banked in
4046 * @p in case that task is currently running. 4046 * @p in case that task is currently running.
4047 */ 4047 */
4048 unsigned long long task_delta_exec(struct task_struct *p) 4048 unsigned long long task_delta_exec(struct task_struct *p)
4049 { 4049 {
4050 unsigned long flags; 4050 unsigned long flags;
4051 struct rq *rq; 4051 struct rq *rq;
4052 u64 ns = 0; 4052 u64 ns = 0;
4053 4053
4054 rq = task_rq_lock(p, &flags); 4054 rq = task_rq_lock(p, &flags);
4055 4055
4056 if (task_current(rq, p)) { 4056 if (task_current(rq, p)) {
4057 u64 delta_exec; 4057 u64 delta_exec;
4058 4058
4059 update_rq_clock(rq); 4059 update_rq_clock(rq);
4060 delta_exec = rq->clock - p->se.exec_start; 4060 delta_exec = rq->clock - p->se.exec_start;
4061 if ((s64)delta_exec > 0) 4061 if ((s64)delta_exec > 0)
4062 ns = delta_exec; 4062 ns = delta_exec;
4063 } 4063 }
4064 4064
4065 task_rq_unlock(rq, &flags); 4065 task_rq_unlock(rq, &flags);
4066 4066
4067 return ns; 4067 return ns;
4068 } 4068 }
4069 4069
4070 /* 4070 /*
4071 * Account user cpu time to a process. 4071 * Account user cpu time to a process.
4072 * @p: the process that the cpu time gets accounted to 4072 * @p: the process that the cpu time gets accounted to
4073 * @cputime: the cpu time spent in user space since the last update 4073 * @cputime: the cpu time spent in user space since the last update
4074 */ 4074 */
4075 void account_user_time(struct task_struct *p, cputime_t cputime) 4075 void account_user_time(struct task_struct *p, cputime_t cputime)
4076 { 4076 {
4077 struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; 4077 struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
4078 cputime64_t tmp; 4078 cputime64_t tmp;
4079 4079
4080 p->utime = cputime_add(p->utime, cputime); 4080 p->utime = cputime_add(p->utime, cputime);
4081 account_group_user_time(p, cputime); 4081 account_group_user_time(p, cputime);
4082 4082
4083 /* Add user time to cpustat. */ 4083 /* Add user time to cpustat. */
4084 tmp = cputime_to_cputime64(cputime); 4084 tmp = cputime_to_cputime64(cputime);
4085 if (TASK_NICE(p) > 0) 4085 if (TASK_NICE(p) > 0)
4086 cpustat->nice = cputime64_add(cpustat->nice, tmp); 4086 cpustat->nice = cputime64_add(cpustat->nice, tmp);
4087 else 4087 else
4088 cpustat->user = cputime64_add(cpustat->user, tmp); 4088 cpustat->user = cputime64_add(cpustat->user, tmp);
4089 /* Account for user time used */ 4089 /* Account for user time used */
4090 acct_update_integrals(p); 4090 acct_update_integrals(p);
4091 } 4091 }
4092 4092
4093 /* 4093 /*
4094 * Account guest cpu time to a process. 4094 * Account guest cpu time to a process.
4095 * @p: the process that the cpu time gets accounted to 4095 * @p: the process that the cpu time gets accounted to
4096 * @cputime: the cpu time spent in virtual machine since the last update 4096 * @cputime: the cpu time spent in virtual machine since the last update
4097 */ 4097 */
4098 static void account_guest_time(struct task_struct *p, cputime_t cputime) 4098 static void account_guest_time(struct task_struct *p, cputime_t cputime)
4099 { 4099 {
4100 cputime64_t tmp; 4100 cputime64_t tmp;
4101 struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; 4101 struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
4102 4102
4103 tmp = cputime_to_cputime64(cputime); 4103 tmp = cputime_to_cputime64(cputime);
4104 4104
4105 p->utime = cputime_add(p->utime, cputime); 4105 p->utime = cputime_add(p->utime, cputime);
4106 account_group_user_time(p, cputime); 4106 account_group_user_time(p, cputime);
4107 p->gtime = cputime_add(p->gtime, cputime); 4107 p->gtime = cputime_add(p->gtime, cputime);
4108 4108
4109 cpustat->user = cputime64_add(cpustat->user, tmp); 4109 cpustat->user = cputime64_add(cpustat->user, tmp);
4110 cpustat->guest = cputime64_add(cpustat->guest, tmp); 4110 cpustat->guest = cputime64_add(cpustat->guest, tmp);
4111 } 4111 }
4112 4112
4113 /* 4113 /*
4114 * Account scaled user cpu time to a process. 4114 * Account scaled user cpu time to a process.
4115 * @p: the process that the cpu time gets accounted to 4115 * @p: the process that the cpu time gets accounted to
4116 * @cputime: the cpu time spent in user space since the last update 4116 * @cputime: the cpu time spent in user space since the last update
4117 */ 4117 */
4118 void account_user_time_scaled(struct task_struct *p, cputime_t cputime) 4118 void account_user_time_scaled(struct task_struct *p, cputime_t cputime)
4119 { 4119 {
4120 p->utimescaled = cputime_add(p->utimescaled, cputime); 4120 p->utimescaled = cputime_add(p->utimescaled, cputime);
4121 } 4121 }
4122 4122
4123 /* 4123 /*
4124 * Account system cpu time to a process. 4124 * Account system cpu time to a process.
4125 * @p: the process that the cpu time gets accounted to 4125 * @p: the process that the cpu time gets accounted to
4126 * @hardirq_offset: the offset to subtract from hardirq_count() 4126 * @hardirq_offset: the offset to subtract from hardirq_count()
4127 * @cputime: the cpu time spent in kernel space since the last update 4127 * @cputime: the cpu time spent in kernel space since the last update
4128 */ 4128 */
4129 void account_system_time(struct task_struct *p, int hardirq_offset, 4129 void account_system_time(struct task_struct *p, int hardirq_offset,
4130 cputime_t cputime) 4130 cputime_t cputime)
4131 { 4131 {
4132 struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; 4132 struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
4133 struct rq *rq = this_rq(); 4133 struct rq *rq = this_rq();
4134 cputime64_t tmp; 4134 cputime64_t tmp;
4135 4135
4136 if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { 4136 if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) {
4137 account_guest_time(p, cputime); 4137 account_guest_time(p, cputime);
4138 return; 4138 return;
4139 } 4139 }
4140 4140
4141 p->stime = cputime_add(p->stime, cputime); 4141 p->stime = cputime_add(p->stime, cputime);
4142 account_group_system_time(p, cputime); 4142 account_group_system_time(p, cputime);
4143 4143
4144 /* Add system time to cpustat. */ 4144 /* Add system time to cpustat. */
4145 tmp = cputime_to_cputime64(cputime); 4145 tmp = cputime_to_cputime64(cputime);
4146 if (hardirq_count() - hardirq_offset) 4146 if (hardirq_count() - hardirq_offset)
4147 cpustat->irq = cputime64_add(cpustat->irq, tmp); 4147 cpustat->irq = cputime64_add(cpustat->irq, tmp);
4148 else if (softirq_count()) 4148 else if (softirq_count())
4149 cpustat->softirq = cputime64_add(cpustat->softirq, tmp); 4149 cpustat->softirq = cputime64_add(cpustat->softirq, tmp);
4150 else if (p != rq->idle) 4150 else if (p != rq->idle)
4151 cpustat->system = cputime64_add(cpustat->system, tmp); 4151 cpustat->system = cputime64_add(cpustat->system, tmp);
4152 else if (atomic_read(&rq->nr_iowait) > 0) 4152 else if (atomic_read(&rq->nr_iowait) > 0)
4153 cpustat->iowait = cputime64_add(cpustat->iowait, tmp); 4153 cpustat->iowait = cputime64_add(cpustat->iowait, tmp);
4154 else 4154 else
4155 cpustat->idle = cputime64_add(cpustat->idle, tmp); 4155 cpustat->idle = cputime64_add(cpustat->idle, tmp);
4156 /* Account for system time used */ 4156 /* Account for system time used */
4157 acct_update_integrals(p); 4157 acct_update_integrals(p);
4158 } 4158 }
4159 4159
4160 /* 4160 /*
4161 * Account scaled system cpu time to a process. 4161 * Account scaled system cpu time to a process.
4162 * @p: the process that the cpu time gets accounted to 4162 * @p: the process that the cpu time gets accounted to
4163 * @hardirq_offset: the offset to subtract from hardirq_count() 4163 * @hardirq_offset: the offset to subtract from hardirq_count()
4164 * @cputime: the cpu time spent in kernel space since the last update 4164 * @cputime: the cpu time spent in kernel space since the last update
4165 */ 4165 */
4166 void account_system_time_scaled(struct task_struct *p, cputime_t cputime) 4166 void account_system_time_scaled(struct task_struct *p, cputime_t cputime)
4167 { 4167 {
4168 p->stimescaled = cputime_add(p->stimescaled, cputime); 4168 p->stimescaled = cputime_add(p->stimescaled, cputime);
4169 } 4169 }
4170 4170
4171 /* 4171 /*
4172 * Account for involuntary wait time. 4172 * Account for involuntary wait time.
4173 * @p: the process from which the cpu time has been stolen 4173 * @p: the process from which the cpu time has been stolen
4174 * @steal: the cpu time spent in involuntary wait 4174 * @steal: the cpu time spent in involuntary wait
4175 */ 4175 */
4176 void account_steal_time(struct task_struct *p, cputime_t steal) 4176 void account_steal_time(struct task_struct *p, cputime_t steal)
4177 { 4177 {
4178 struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; 4178 struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
4179 cputime64_t tmp = cputime_to_cputime64(steal); 4179 cputime64_t tmp = cputime_to_cputime64(steal);
4180 struct rq *rq = this_rq(); 4180 struct rq *rq = this_rq();
4181 4181
4182 if (p == rq->idle) { 4182 if (p == rq->idle) {
4183 p->stime = cputime_add(p->stime, steal); 4183 p->stime = cputime_add(p->stime, steal);
4184 account_group_system_time(p, steal); 4184 account_group_system_time(p, steal);
4185 if (atomic_read(&rq->nr_iowait) > 0) 4185 if (atomic_read(&rq->nr_iowait) > 0)
4186 cpustat->iowait = cputime64_add(cpustat->iowait, tmp); 4186 cpustat->iowait = cputime64_add(cpustat->iowait, tmp);
4187 else 4187 else
4188 cpustat->idle = cputime64_add(cpustat->idle, tmp); 4188 cpustat->idle = cputime64_add(cpustat->idle, tmp);
4189 } else 4189 } else
4190 cpustat->steal = cputime64_add(cpustat->steal, tmp); 4190 cpustat->steal = cputime64_add(cpustat->steal, tmp);
4191 } 4191 }
4192 4192
4193 /* 4193 /*
4194 * Use precise platform statistics if available: 4194 * Use precise platform statistics if available:
4195 */ 4195 */
4196 #ifdef CONFIG_VIRT_CPU_ACCOUNTING 4196 #ifdef CONFIG_VIRT_CPU_ACCOUNTING
4197 cputime_t task_utime(struct task_struct *p) 4197 cputime_t task_utime(struct task_struct *p)
4198 { 4198 {
4199 return p->utime; 4199 return p->utime;
4200 } 4200 }
4201 4201
4202 cputime_t task_stime(struct task_struct *p) 4202 cputime_t task_stime(struct task_struct *p)
4203 { 4203 {
4204 return p->stime; 4204 return p->stime;
4205 } 4205 }
4206 #else 4206 #else
4207 cputime_t task_utime(struct task_struct *p) 4207 cputime_t task_utime(struct task_struct *p)
4208 { 4208 {
4209 clock_t utime = cputime_to_clock_t(p->utime), 4209 clock_t utime = cputime_to_clock_t(p->utime),
4210 total = utime + cputime_to_clock_t(p->stime); 4210 total = utime + cputime_to_clock_t(p->stime);
4211 u64 temp; 4211 u64 temp;
4212 4212
4213 /* 4213 /*
4214 * Use CFS's precise accounting: 4214 * Use CFS's precise accounting:
4215 */ 4215 */
4216 temp = (u64)nsec_to_clock_t(p->se.sum_exec_runtime); 4216 temp = (u64)nsec_to_clock_t(p->se.sum_exec_runtime);
4217 4217
4218 if (total) { 4218 if (total) {
4219 temp *= utime; 4219 temp *= utime;
4220 do_div(temp, total); 4220 do_div(temp, total);
4221 } 4221 }
4222 utime = (clock_t)temp; 4222 utime = (clock_t)temp;
4223 4223
4224 p->prev_utime = max(p->prev_utime, clock_t_to_cputime(utime)); 4224 p->prev_utime = max(p->prev_utime, clock_t_to_cputime(utime));
4225 return p->prev_utime; 4225 return p->prev_utime;
4226 } 4226 }
4227 4227
4228 cputime_t task_stime(struct task_struct *p) 4228 cputime_t task_stime(struct task_struct *p)
4229 { 4229 {
4230 clock_t stime; 4230 clock_t stime;
4231 4231
4232 /* 4232 /*
4233 * Use CFS's precise accounting. (we subtract utime from 4233 * Use CFS's precise accounting. (we subtract utime from
4234 * the total, to make sure the total observed by userspace 4234 * the total, to make sure the total observed by userspace
4235 * grows monotonically - apps rely on that): 4235 * grows monotonically - apps rely on that):
4236 */ 4236 */
4237 stime = nsec_to_clock_t(p->se.sum_exec_runtime) - 4237 stime = nsec_to_clock_t(p->se.sum_exec_runtime) -
4238 cputime_to_clock_t(task_utime(p)); 4238 cputime_to_clock_t(task_utime(p));
4239 4239
4240 if (stime >= 0) 4240 if (stime >= 0)
4241 p->prev_stime = max(p->prev_stime, clock_t_to_cputime(stime)); 4241 p->prev_stime = max(p->prev_stime, clock_t_to_cputime(stime));
4242 4242
4243 return p->prev_stime; 4243 return p->prev_stime;
4244 } 4244 }
4245 #endif 4245 #endif
4246 4246
4247 inline cputime_t task_gtime(struct task_struct *p) 4247 inline cputime_t task_gtime(struct task_struct *p)
4248 { 4248 {
4249 return p->gtime; 4249 return p->gtime;
4250 } 4250 }
4251 4251
4252 /* 4252 /*
4253 * This function gets called by the timer code, with HZ frequency. 4253 * This function gets called by the timer code, with HZ frequency.
4254 * We call it with interrupts disabled. 4254 * We call it with interrupts disabled.
4255 * 4255 *
4256 * It also gets called by the fork code, when changing the parent's 4256 * It also gets called by the fork code, when changing the parent's
4257 * timeslices. 4257 * timeslices.
4258 */ 4258 */
4259 void scheduler_tick(void) 4259 void scheduler_tick(void)
4260 { 4260 {
4261 int cpu = smp_processor_id(); 4261 int cpu = smp_processor_id();
4262 struct rq *rq = cpu_rq(cpu); 4262 struct rq *rq = cpu_rq(cpu);
4263 struct task_struct *curr = rq->curr; 4263 struct task_struct *curr = rq->curr;
4264 4264
4265 sched_clock_tick(); 4265 sched_clock_tick();
4266 4266
4267 spin_lock(&rq->lock); 4267 spin_lock(&rq->lock);
4268 update_rq_clock(rq); 4268 update_rq_clock(rq);
4269 update_cpu_load(rq); 4269 update_cpu_load(rq);
4270 curr->sched_class->task_tick(rq, curr, 0); 4270 curr->sched_class->task_tick(rq, curr, 0);
4271 spin_unlock(&rq->lock); 4271 spin_unlock(&rq->lock);
4272 4272
4273 #ifdef CONFIG_SMP 4273 #ifdef CONFIG_SMP
4274 rq->idle_at_tick = idle_cpu(cpu); 4274 rq->idle_at_tick = idle_cpu(cpu);
4275 trigger_load_balance(rq, cpu); 4275 trigger_load_balance(rq, cpu);
4276 #endif 4276 #endif
4277 } 4277 }
4278 4278
4279 #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ 4279 #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \
4280 defined(CONFIG_PREEMPT_TRACER)) 4280 defined(CONFIG_PREEMPT_TRACER))
4281 4281
4282 static inline unsigned long get_parent_ip(unsigned long addr) 4282 static inline unsigned long get_parent_ip(unsigned long addr)
4283 { 4283 {
4284 if (in_lock_functions(addr)) { 4284 if (in_lock_functions(addr)) {
4285 addr = CALLER_ADDR2; 4285 addr = CALLER_ADDR2;
4286 if (in_lock_functions(addr)) 4286 if (in_lock_functions(addr))
4287 addr = CALLER_ADDR3; 4287 addr = CALLER_ADDR3;
4288 } 4288 }
4289 return addr; 4289 return addr;
4290 } 4290 }
4291 4291
4292 void __kprobes add_preempt_count(int val) 4292 void __kprobes add_preempt_count(int val)
4293 { 4293 {
4294 #ifdef CONFIG_DEBUG_PREEMPT 4294 #ifdef CONFIG_DEBUG_PREEMPT
4295 /* 4295 /*
4296 * Underflow? 4296 * Underflow?
4297 */ 4297 */
4298 if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) 4298 if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0)))
4299 return; 4299 return;
4300 #endif 4300 #endif
4301 preempt_count() += val; 4301 preempt_count() += val;
4302 #ifdef CONFIG_DEBUG_PREEMPT 4302 #ifdef CONFIG_DEBUG_PREEMPT
4303 /* 4303 /*
4304 * Spinlock count overflowing soon? 4304 * Spinlock count overflowing soon?
4305 */ 4305 */
4306 DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= 4306 DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >=
4307 PREEMPT_MASK - 10); 4307 PREEMPT_MASK - 10);
4308 #endif 4308 #endif
4309 if (preempt_count() == val) 4309 if (preempt_count() == val)
4310 trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); 4310 trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1));
4311 } 4311 }
4312 EXPORT_SYMBOL(add_preempt_count); 4312 EXPORT_SYMBOL(add_preempt_count);
4313 4313
4314 void __kprobes sub_preempt_count(int val) 4314 void __kprobes sub_preempt_count(int val)
4315 { 4315 {
4316 #ifdef CONFIG_DEBUG_PREEMPT 4316 #ifdef CONFIG_DEBUG_PREEMPT
4317 /* 4317 /*
4318 * Underflow? 4318 * Underflow?
4319 */ 4319 */
4320 if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) 4320 if (DEBUG_LOCKS_WARN_ON(val > preempt_count()))
4321 return; 4321 return;
4322 /* 4322 /*
4323 * Is the spinlock portion underflowing? 4323 * Is the spinlock portion underflowing?
4324 */ 4324 */
4325 if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && 4325 if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) &&
4326 !(preempt_count() & PREEMPT_MASK))) 4326 !(preempt_count() & PREEMPT_MASK)))
4327 return; 4327 return;
4328 #endif 4328 #endif
4329 4329
4330 if (preempt_count() == val) 4330 if (preempt_count() == val)
4331 trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); 4331 trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1));
4332 preempt_count() -= val; 4332 preempt_count() -= val;
4333 } 4333 }
4334 EXPORT_SYMBOL(sub_preempt_count); 4334 EXPORT_SYMBOL(sub_preempt_count);
4335 4335
4336 #endif 4336 #endif
4337 4337
4338 /* 4338 /*
4339 * Print scheduling while atomic bug: 4339 * Print scheduling while atomic bug:
4340 */ 4340 */
4341 static noinline void __schedule_bug(struct task_struct *prev) 4341 static noinline void __schedule_bug(struct task_struct *prev)
4342 { 4342 {
4343 struct pt_regs *regs = get_irq_regs(); 4343 struct pt_regs *regs = get_irq_regs();
4344 4344
4345 printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", 4345 printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n",
4346 prev->comm, prev->pid, preempt_count()); 4346 prev->comm, prev->pid, preempt_count());
4347 4347
4348 debug_show_held_locks(prev); 4348 debug_show_held_locks(prev);
4349 print_modules(); 4349 print_modules();
4350 if (irqs_disabled()) 4350 if (irqs_disabled())
4351 print_irqtrace_events(prev); 4351 print_irqtrace_events(prev);
4352 4352
4353 if (regs) 4353 if (regs)
4354 show_regs(regs); 4354 show_regs(regs);
4355 else 4355 else
4356 dump_stack(); 4356 dump_stack();
4357 } 4357 }
4358 4358
4359 /* 4359 /*
4360 * Various schedule()-time debugging checks and statistics: 4360 * Various schedule()-time debugging checks and statistics:
4361 */ 4361 */
4362 static inline void schedule_debug(struct task_struct *prev) 4362 static inline void schedule_debug(struct task_struct *prev)
4363 { 4363 {
4364 /* 4364 /*
4365 * Test if we are atomic. Since do_exit() needs to call into 4365 * Test if we are atomic. Since do_exit() needs to call into
4366 * schedule() atomically, we ignore that path for now. 4366 * schedule() atomically, we ignore that path for now.
4367 * Otherwise, whine if we are scheduling when we should not be. 4367 * Otherwise, whine if we are scheduling when we should not be.
4368 */ 4368 */
4369 if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) 4369 if (unlikely(in_atomic_preempt_off() && !prev->exit_state))
4370 __schedule_bug(prev); 4370 __schedule_bug(prev);
4371 4371
4372 profile_hit(SCHED_PROFILING, __builtin_return_address(0)); 4372 profile_hit(SCHED_PROFILING, __builtin_return_address(0));
4373 4373
4374 schedstat_inc(this_rq(), sched_count); 4374 schedstat_inc(this_rq(), sched_count);
4375 #ifdef CONFIG_SCHEDSTATS 4375 #ifdef CONFIG_SCHEDSTATS
4376 if (unlikely(prev->lock_depth >= 0)) { 4376 if (unlikely(prev->lock_depth >= 0)) {
4377 schedstat_inc(this_rq(), bkl_count); 4377 schedstat_inc(this_rq(), bkl_count);
4378 schedstat_inc(prev, sched_info.bkl_count); 4378 schedstat_inc(prev, sched_info.bkl_count);
4379 } 4379 }
4380 #endif 4380 #endif
4381 } 4381 }
4382 4382
4383 /* 4383 /*
4384 * Pick up the highest-prio task: 4384 * Pick up the highest-prio task:
4385 */ 4385 */
4386 static inline struct task_struct * 4386 static inline struct task_struct *
4387 pick_next_task(struct rq *rq, struct task_struct *prev) 4387 pick_next_task(struct rq *rq, struct task_struct *prev)
4388 { 4388 {
4389 const struct sched_class *class; 4389 const struct sched_class *class;
4390 struct task_struct *p; 4390 struct task_struct *p;
4391 4391
4392 /* 4392 /*
4393 * Optimization: we know that if all tasks are in 4393 * Optimization: we know that if all tasks are in
4394 * the fair class we can call that function directly: 4394 * the fair class we can call that function directly:
4395 */ 4395 */
4396 if (likely(rq->nr_running == rq->cfs.nr_running)) { 4396 if (likely(rq->nr_running == rq->cfs.nr_running)) {
4397 p = fair_sched_class.pick_next_task(rq); 4397 p = fair_sched_class.pick_next_task(rq);
4398 if (likely(p)) 4398 if (likely(p))
4399 return p; 4399 return p;
4400 } 4400 }
4401 4401
4402 class = sched_class_highest; 4402 class = sched_class_highest;
4403 for ( ; ; ) { 4403 for ( ; ; ) {
4404 p = class->pick_next_task(rq); 4404 p = class->pick_next_task(rq);
4405 if (p) 4405 if (p)
4406 return p; 4406 return p;
4407 /* 4407 /*
4408 * Will never be NULL as the idle class always 4408 * Will never be NULL as the idle class always
4409 * returns a non-NULL p: 4409 * returns a non-NULL p:
4410 */ 4410 */
4411 class = class->next; 4411 class = class->next;
4412 } 4412 }
4413 } 4413 }
4414 4414
4415 /* 4415 /*
4416 * schedule() is the main scheduler function. 4416 * schedule() is the main scheduler function.
4417 */ 4417 */
4418 asmlinkage void __sched schedule(void) 4418 asmlinkage void __sched schedule(void)
4419 { 4419 {
4420 struct task_struct *prev, *next; 4420 struct task_struct *prev, *next;
4421 unsigned long *switch_count; 4421 unsigned long *switch_count;
4422 struct rq *rq; 4422 struct rq *rq;
4423 int cpu; 4423 int cpu;
4424 4424
4425 need_resched: 4425 need_resched:
4426 preempt_disable(); 4426 preempt_disable();
4427 cpu = smp_processor_id(); 4427 cpu = smp_processor_id();
4428 rq = cpu_rq(cpu); 4428 rq = cpu_rq(cpu);
4429 rcu_qsctr_inc(cpu); 4429 rcu_qsctr_inc(cpu);
4430 prev = rq->curr; 4430 prev = rq->curr;
4431 switch_count = &prev->nivcsw; 4431 switch_count = &prev->nivcsw;
4432 4432
4433 release_kernel_lock(prev); 4433 release_kernel_lock(prev);
4434 need_resched_nonpreemptible: 4434 need_resched_nonpreemptible:
4435 4435
4436 schedule_debug(prev); 4436 schedule_debug(prev);
4437 4437
4438 if (sched_feat(HRTICK)) 4438 if (sched_feat(HRTICK))
4439 hrtick_clear(rq); 4439 hrtick_clear(rq);
4440 4440
4441 spin_lock_irq(&rq->lock); 4441 spin_lock_irq(&rq->lock);
4442 update_rq_clock(rq); 4442 update_rq_clock(rq);
4443 clear_tsk_need_resched(prev); 4443 clear_tsk_need_resched(prev);
4444 4444
4445 if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { 4445 if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) {
4446 if (unlikely(signal_pending_state(prev->state, prev))) 4446 if (unlikely(signal_pending_state(prev->state, prev)))
4447 prev->state = TASK_RUNNING; 4447 prev->state = TASK_RUNNING;
4448 else 4448 else
4449 deactivate_task(rq, prev, 1); 4449 deactivate_task(rq, prev, 1);
4450 switch_count = &prev->nvcsw; 4450 switch_count = &prev->nvcsw;
4451 } 4451 }
4452 4452
4453 #ifdef CONFIG_SMP 4453 #ifdef CONFIG_SMP
4454 if (prev->sched_class->pre_schedule) 4454 if (prev->sched_class->pre_schedule)
4455 prev->sched_class->pre_schedule(rq, prev); 4455 prev->sched_class->pre_schedule(rq, prev);
4456 #endif 4456 #endif
4457 4457
4458 if (unlikely(!rq->nr_running)) 4458 if (unlikely(!rq->nr_running))
4459 idle_balance(cpu, rq); 4459 idle_balance(cpu, rq);
4460 4460
4461 prev->sched_class->put_prev_task(rq, prev); 4461 prev->sched_class->put_prev_task(rq, prev);
4462 next = pick_next_task(rq, prev); 4462 next = pick_next_task(rq, prev);
4463 4463
4464 if (likely(prev != next)) { 4464 if (likely(prev != next)) {
4465 sched_info_switch(prev, next); 4465 sched_info_switch(prev, next);
4466 4466
4467 rq->nr_switches++; 4467 rq->nr_switches++;
4468 rq->curr = next; 4468 rq->curr = next;
4469 ++*switch_count; 4469 ++*switch_count;
4470 4470
4471 context_switch(rq, prev, next); /* unlocks the rq */ 4471 context_switch(rq, prev, next); /* unlocks the rq */
4472 /* 4472 /*
4473 * the context switch might have flipped the stack from under 4473 * the context switch might have flipped the stack from under
4474 * us, hence refresh the local variables. 4474 * us, hence refresh the local variables.
4475 */ 4475 */
4476 cpu = smp_processor_id(); 4476 cpu = smp_processor_id();
4477 rq = cpu_rq(cpu); 4477 rq = cpu_rq(cpu);
4478 } else 4478 } else
4479 spin_unlock_irq(&rq->lock); 4479 spin_unlock_irq(&rq->lock);
4480 4480
4481 if (unlikely(reacquire_kernel_lock(current) < 0)) 4481 if (unlikely(reacquire_kernel_lock(current) < 0))
4482 goto need_resched_nonpreemptible; 4482 goto need_resched_nonpreemptible;
4483 4483
4484 preempt_enable_no_resched(); 4484 preempt_enable_no_resched();
4485 if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) 4485 if (unlikely(test_thread_flag(TIF_NEED_RESCHED)))
4486 goto need_resched; 4486 goto need_resched;
4487 } 4487 }
4488 EXPORT_SYMBOL(schedule); 4488 EXPORT_SYMBOL(schedule);
4489 4489
4490 #ifdef CONFIG_PREEMPT 4490 #ifdef CONFIG_PREEMPT
4491 /* 4491 /*
4492 * this is the entry point to schedule() from in-kernel preemption 4492 * this is the entry point to schedule() from in-kernel preemption
4493 * off of preempt_enable. Kernel preemptions off return from interrupt 4493 * off of preempt_enable. Kernel preemptions off return from interrupt
4494 * occur there and call schedule directly. 4494 * occur there and call schedule directly.
4495 */ 4495 */
4496 asmlinkage void __sched preempt_schedule(void) 4496 asmlinkage void __sched preempt_schedule(void)
4497 { 4497 {
4498 struct thread_info *ti = current_thread_info(); 4498 struct thread_info *ti = current_thread_info();
4499 4499
4500 /* 4500 /*
4501 * If there is a non-zero preempt_count or interrupts are disabled, 4501 * If there is a non-zero preempt_count or interrupts are disabled,
4502 * we do not want to preempt the current task. Just return.. 4502 * we do not want to preempt the current task. Just return..
4503 */ 4503 */
4504 if (likely(ti->preempt_count || irqs_disabled())) 4504 if (likely(ti->preempt_count || irqs_disabled()))
4505 return; 4505 return;
4506 4506
4507 do { 4507 do {
4508 add_preempt_count(PREEMPT_ACTIVE); 4508 add_preempt_count(PREEMPT_ACTIVE);
4509 schedule(); 4509 schedule();
4510 sub_preempt_count(PREEMPT_ACTIVE); 4510 sub_preempt_count(PREEMPT_ACTIVE);
4511 4511
4512 /* 4512 /*
4513 * Check again in case we missed a preemption opportunity 4513 * Check again in case we missed a preemption opportunity
4514 * between schedule and now. 4514 * between schedule and now.
4515 */ 4515 */
4516 barrier(); 4516 barrier();
4517 } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); 4517 } while (unlikely(test_thread_flag(TIF_NEED_RESCHED)));
4518 } 4518 }
4519 EXPORT_SYMBOL(preempt_schedule); 4519 EXPORT_SYMBOL(preempt_schedule);
4520 4520
4521 /* 4521 /*
4522 * this is the entry point to schedule() from kernel preemption 4522 * this is the entry point to schedule() from kernel preemption
4523 * off of irq context. 4523 * off of irq context.
4524 * Note, that this is called and return with irqs disabled. This will 4524 * Note, that this is called and return with irqs disabled. This will
4525 * protect us against recursive calling from irq. 4525 * protect us against recursive calling from irq.
4526 */ 4526 */
4527 asmlinkage void __sched preempt_schedule_irq(void) 4527 asmlinkage void __sched preempt_schedule_irq(void)
4528 { 4528 {
4529 struct thread_info *ti = current_thread_info(); 4529 struct thread_info *ti = current_thread_info();
4530 4530
4531 /* Catch callers which need to be fixed */ 4531 /* Catch callers which need to be fixed */
4532 BUG_ON(ti->preempt_count || !irqs_disabled()); 4532 BUG_ON(ti->preempt_count || !irqs_disabled());
4533 4533
4534 do { 4534 do {
4535 add_preempt_count(PREEMPT_ACTIVE); 4535 add_preempt_count(PREEMPT_ACTIVE);
4536 local_irq_enable(); 4536 local_irq_enable();
4537 schedule(); 4537 schedule();
4538 local_irq_disable(); 4538 local_irq_disable();
4539 sub_preempt_count(PREEMPT_ACTIVE); 4539 sub_preempt_count(PREEMPT_ACTIVE);
4540 4540
4541 /* 4541 /*
4542 * Check again in case we missed a preemption opportunity 4542 * Check again in case we missed a preemption opportunity
4543 * between schedule and now. 4543 * between schedule and now.
4544 */ 4544 */
4545 barrier(); 4545 barrier();
4546 } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); 4546 } while (unlikely(test_thread_flag(TIF_NEED_RESCHED)));
4547 } 4547 }
4548 4548
4549 #endif /* CONFIG_PREEMPT */ 4549 #endif /* CONFIG_PREEMPT */
4550 4550
4551 int default_wake_function(wait_queue_t *curr, unsigned mode, int sync, 4551 int default_wake_function(wait_queue_t *curr, unsigned mode, int sync,
4552 void *key) 4552 void *key)
4553 { 4553 {
4554 return try_to_wake_up(curr->private, mode, sync); 4554 return try_to_wake_up(curr->private, mode, sync);
4555 } 4555 }
4556 EXPORT_SYMBOL(default_wake_function); 4556 EXPORT_SYMBOL(default_wake_function);
4557 4557
4558 /* 4558 /*
4559 * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just 4559 * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just
4560 * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve 4560 * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve
4561 * number) then we wake all the non-exclusive tasks and one exclusive task. 4561 * number) then we wake all the non-exclusive tasks and one exclusive task.
4562 * 4562 *
4563 * There are circumstances in which we can try to wake a task which has already 4563 * There are circumstances in which we can try to wake a task which has already
4564 * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns 4564 * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns
4565 * zero in this (rare) case, and we handle it by continuing to scan the queue. 4565 * zero in this (rare) case, and we handle it by continuing to scan the queue.
4566 */ 4566 */
4567 static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, 4567 static void __wake_up_common(wait_queue_head_t *q, unsigned int mode,
4568 int nr_exclusive, int sync, void *key) 4568 int nr_exclusive, int sync, void *key)
4569 { 4569 {
4570 wait_queue_t *curr, *next; 4570 wait_queue_t *curr, *next;
4571 4571
4572 list_for_each_entry_safe(curr, next, &q->task_list, task_list) { 4572 list_for_each_entry_safe(curr, next, &q->task_list, task_list) {
4573 unsigned flags = curr->flags; 4573 unsigned flags = curr->flags;
4574 4574
4575 if (curr->func(curr, mode, sync, key) && 4575 if (curr->func(curr, mode, sync, key) &&
4576 (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) 4576 (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive)
4577 break; 4577 break;
4578 } 4578 }
4579 } 4579 }
4580 4580
4581 /** 4581 /**
4582 * __wake_up - wake up threads blocked on a waitqueue. 4582 * __wake_up - wake up threads blocked on a waitqueue.
4583 * @q: the waitqueue 4583 * @q: the waitqueue
4584 * @mode: which threads 4584 * @mode: which threads
4585 * @nr_exclusive: how many wake-one or wake-many threads to wake up 4585 * @nr_exclusive: how many wake-one or wake-many threads to wake up
4586 * @key: is directly passed to the wakeup function 4586 * @key: is directly passed to the wakeup function
4587 */ 4587 */
4588 void __wake_up(wait_queue_head_t *q, unsigned int mode, 4588 void __wake_up(wait_queue_head_t *q, unsigned int mode,
4589 int nr_exclusive, void *key) 4589 int nr_exclusive, void *key)
4590 { 4590 {
4591 unsigned long flags; 4591 unsigned long flags;
4592 4592
4593 spin_lock_irqsave(&q->lock, flags); 4593 spin_lock_irqsave(&q->lock, flags);
4594 __wake_up_common(q, mode, nr_exclusive, 0, key); 4594 __wake_up_common(q, mode, nr_exclusive, 0, key);
4595 spin_unlock_irqrestore(&q->lock, flags); 4595 spin_unlock_irqrestore(&q->lock, flags);
4596 } 4596 }
4597 EXPORT_SYMBOL(__wake_up); 4597 EXPORT_SYMBOL(__wake_up);
4598 4598
4599 /* 4599 /*
4600 * Same as __wake_up but called with the spinlock in wait_queue_head_t held. 4600 * Same as __wake_up but called with the spinlock in wait_queue_head_t held.
4601 */ 4601 */
4602 void __wake_up_locked(wait_queue_head_t *q, unsigned int mode) 4602 void __wake_up_locked(wait_queue_head_t *q, unsigned int mode)
4603 { 4603 {
4604 __wake_up_common(q, mode, 1, 0, NULL); 4604 __wake_up_common(q, mode, 1, 0, NULL);
4605 } 4605 }
4606 4606
4607 /** 4607 /**
4608 * __wake_up_sync - wake up threads blocked on a waitqueue. 4608 * __wake_up_sync - wake up threads blocked on a waitqueue.
4609 * @q: the waitqueue 4609 * @q: the waitqueue
4610 * @mode: which threads 4610 * @mode: which threads
4611 * @nr_exclusive: how many wake-one or wake-many threads to wake up 4611 * @nr_exclusive: how many wake-one or wake-many threads to wake up
4612 * 4612 *
4613 * The sync wakeup differs that the waker knows that it will schedule 4613 * The sync wakeup differs that the waker knows that it will schedule
4614 * away soon, so while the target thread will be woken up, it will not 4614 * away soon, so while the target thread will be woken up, it will not
4615 * be migrated to another CPU - ie. the two threads are 'synchronized' 4615 * be migrated to another CPU - ie. the two threads are 'synchronized'
4616 * with each other. This can prevent needless bouncing between CPUs. 4616 * with each other. This can prevent needless bouncing between CPUs.
4617 * 4617 *
4618 * On UP it can prevent extra preemption. 4618 * On UP it can prevent extra preemption.
4619 */ 4619 */
4620 void 4620 void
4621 __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) 4621 __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive)
4622 { 4622 {
4623 unsigned long flags; 4623 unsigned long flags;
4624 int sync = 1; 4624 int sync = 1;
4625 4625
4626 if (unlikely(!q)) 4626 if (unlikely(!q))
4627 return; 4627 return;
4628 4628
4629 if (unlikely(!nr_exclusive)) 4629 if (unlikely(!nr_exclusive))
4630 sync = 0; 4630 sync = 0;
4631 4631
4632 spin_lock_irqsave(&q->lock, flags); 4632 spin_lock_irqsave(&q->lock, flags);
4633 __wake_up_common(q, mode, nr_exclusive, sync, NULL); 4633 __wake_up_common(q, mode, nr_exclusive, sync, NULL);
4634 spin_unlock_irqrestore(&q->lock, flags); 4634 spin_unlock_irqrestore(&q->lock, flags);
4635 } 4635 }
4636 EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ 4636 EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */
4637 4637
4638 /** 4638 /**
4639 * complete: - signals a single thread waiting on this completion 4639 * complete: - signals a single thread waiting on this completion
4640 * @x: holds the state of this particular completion 4640 * @x: holds the state of this particular completion
4641 * 4641 *
4642 * This will wake up a single thread waiting on this completion. Threads will be 4642 * This will wake up a single thread waiting on this completion. Threads will be
4643 * awakened in the same order in which they were queued. 4643 * awakened in the same order in which they were queued.
4644 * 4644 *
4645 * See also complete_all(), wait_for_completion() and related routines. 4645 * See also complete_all(), wait_for_completion() and related routines.
4646 */ 4646 */
4647 void complete(struct completion *x) 4647 void complete(struct completion *x)
4648 { 4648 {
4649 unsigned long flags; 4649 unsigned long flags;
4650 4650
4651 spin_lock_irqsave(&x->wait.lock, flags); 4651 spin_lock_irqsave(&x->wait.lock, flags);
4652 x->done++; 4652 x->done++;
4653 __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); 4653 __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL);
4654 spin_unlock_irqrestore(&x->wait.lock, flags); 4654 spin_unlock_irqrestore(&x->wait.lock, flags);
4655 } 4655 }
4656 EXPORT_SYMBOL(complete); 4656 EXPORT_SYMBOL(complete);
4657 4657
4658 /** 4658 /**
4659 * complete_all: - signals all threads waiting on this completion 4659 * complete_all: - signals all threads waiting on this completion
4660 * @x: holds the state of this particular completion 4660 * @x: holds the state of this particular completion
4661 * 4661 *
4662 * This will wake up all threads waiting on this particular completion event. 4662 * This will wake up all threads waiting on this particular completion event.
4663 */ 4663 */
4664 void complete_all(struct completion *x) 4664 void complete_all(struct completion *x)
4665 { 4665 {
4666 unsigned long flags; 4666 unsigned long flags;
4667 4667
4668 spin_lock_irqsave(&x->wait.lock, flags); 4668 spin_lock_irqsave(&x->wait.lock, flags);
4669 x->done += UINT_MAX/2; 4669 x->done += UINT_MAX/2;
4670 __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); 4670 __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL);
4671 spin_unlock_irqrestore(&x->wait.lock, flags); 4671 spin_unlock_irqrestore(&x->wait.lock, flags);
4672 } 4672 }
4673 EXPORT_SYMBOL(complete_all); 4673 EXPORT_SYMBOL(complete_all);
4674 4674
4675 static inline long __sched 4675 static inline long __sched
4676 do_wait_for_common(struct completion *x, long timeout, int state) 4676 do_wait_for_common(struct completion *x, long timeout, int state)
4677 { 4677 {
4678 if (!x->done) { 4678 if (!x->done) {
4679 DECLARE_WAITQUEUE(wait, current); 4679 DECLARE_WAITQUEUE(wait, current);
4680 4680
4681 wait.flags |= WQ_FLAG_EXCLUSIVE; 4681 wait.flags |= WQ_FLAG_EXCLUSIVE;
4682 __add_wait_queue_tail(&x->wait, &wait); 4682 __add_wait_queue_tail(&x->wait, &wait);
4683 do { 4683 do {
4684 if (signal_pending_state(state, current)) { 4684 if (signal_pending_state(state, current)) {
4685 timeout = -ERESTARTSYS; 4685 timeout = -ERESTARTSYS;
4686 break; 4686 break;
4687 } 4687 }
4688 __set_current_state(state); 4688 __set_current_state(state);
4689 spin_unlock_irq(&x->wait.lock); 4689 spin_unlock_irq(&x->wait.lock);
4690 timeout = schedule_timeout(timeout); 4690 timeout = schedule_timeout(timeout);
4691 spin_lock_irq(&x->wait.lock); 4691 spin_lock_irq(&x->wait.lock);
4692 } while (!x->done && timeout); 4692 } while (!x->done && timeout);
4693 __remove_wait_queue(&x->wait, &wait); 4693 __remove_wait_queue(&x->wait, &wait);
4694 if (!x->done) 4694 if (!x->done)
4695 return timeout; 4695 return timeout;
4696 } 4696 }
4697 x->done--; 4697 x->done--;
4698 return timeout ?: 1; 4698 return timeout ?: 1;
4699 } 4699 }
4700 4700
4701 static long __sched 4701 static long __sched
4702 wait_for_common(struct completion *x, long timeout, int state) 4702 wait_for_common(struct completion *x, long timeout, int state)
4703 { 4703 {
4704 might_sleep(); 4704 might_sleep();
4705 4705
4706 spin_lock_irq(&x->wait.lock); 4706 spin_lock_irq(&x->wait.lock);
4707 timeout = do_wait_for_common(x, timeout, state); 4707 timeout = do_wait_for_common(x, timeout, state);
4708 spin_unlock_irq(&x->wait.lock); 4708 spin_unlock_irq(&x->wait.lock);
4709 return timeout; 4709 return timeout;
4710 } 4710 }
4711 4711
4712 /** 4712 /**
4713 * wait_for_completion: - waits for completion of a task 4713 * wait_for_completion: - waits for completion of a task
4714 * @x: holds the state of this particular completion 4714 * @x: holds the state of this particular completion
4715 * 4715 *
4716 * This waits to be signaled for completion of a specific task. It is NOT 4716 * This waits to be signaled for completion of a specific task. It is NOT
4717 * interruptible and there is no timeout. 4717 * interruptible and there is no timeout.
4718 * 4718 *
4719 * See also similar routines (i.e. wait_for_completion_timeout()) with timeout 4719 * See also similar routines (i.e. wait_for_completion_timeout()) with timeout
4720 * and interrupt capability. Also see complete(). 4720 * and interrupt capability. Also see complete().
4721 */ 4721 */
4722 void __sched wait_for_completion(struct completion *x) 4722 void __sched wait_for_completion(struct completion *x)
4723 { 4723 {
4724 wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); 4724 wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE);
4725 } 4725 }
4726 EXPORT_SYMBOL(wait_for_completion); 4726 EXPORT_SYMBOL(wait_for_completion);
4727 4727
4728 /** 4728 /**
4729 * wait_for_completion_timeout: - waits for completion of a task (w/timeout) 4729 * wait_for_completion_timeout: - waits for completion of a task (w/timeout)
4730 * @x: holds the state of this particular completion 4730 * @x: holds the state of this particular completion
4731 * @timeout: timeout value in jiffies 4731 * @timeout: timeout value in jiffies
4732 * 4732 *
4733 * This waits for either a completion of a specific task to be signaled or for a 4733 * This waits for either a completion of a specific task to be signaled or for a
4734 * specified timeout to expire. The timeout is in jiffies. It is not 4734 * specified timeout to expire. The timeout is in jiffies. It is not
4735 * interruptible. 4735 * interruptible.
4736 */ 4736 */
4737 unsigned long __sched 4737 unsigned long __sched
4738 wait_for_completion_timeout(struct completion *x, unsigned long timeout) 4738 wait_for_completion_timeout(struct completion *x, unsigned long timeout)
4739 { 4739 {
4740 return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); 4740 return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE);
4741 } 4741 }
4742 EXPORT_SYMBOL(wait_for_completion_timeout); 4742 EXPORT_SYMBOL(wait_for_completion_timeout);
4743 4743
4744 /** 4744 /**
4745 * wait_for_completion_interruptible: - waits for completion of a task (w/intr) 4745 * wait_for_completion_interruptible: - waits for completion of a task (w/intr)
4746 * @x: holds the state of this particular completion 4746 * @x: holds the state of this particular completion
4747 * 4747 *
4748 * This waits for completion of a specific task to be signaled. It is 4748 * This waits for completion of a specific task to be signaled. It is
4749 * interruptible. 4749 * interruptible.
4750 */ 4750 */
4751 int __sched wait_for_completion_interruptible(struct completion *x) 4751 int __sched wait_for_completion_interruptible(struct completion *x)
4752 { 4752 {
4753 long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); 4753 long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE);
4754 if (t == -ERESTARTSYS) 4754 if (t == -ERESTARTSYS)
4755 return t; 4755 return t;
4756 return 0; 4756 return 0;
4757 } 4757 }
4758 EXPORT_SYMBOL(wait_for_completion_interruptible); 4758 EXPORT_SYMBOL(wait_for_completion_interruptible);
4759 4759
4760 /** 4760 /**
4761 * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr)) 4761 * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr))
4762 * @x: holds the state of this particular completion 4762 * @x: holds the state of this particular completion
4763 * @timeout: timeout value in jiffies 4763 * @timeout: timeout value in jiffies
4764 * 4764 *
4765 * This waits for either a completion of a specific task to be signaled or for a 4765 * This waits for either a completion of a specific task to be signaled or for a
4766 * specified timeout to expire. It is interruptible. The timeout is in jiffies. 4766 * specified timeout to expire. It is interruptible. The timeout is in jiffies.
4767 */ 4767 */
4768 unsigned long __sched 4768 unsigned long __sched
4769 wait_for_completion_interruptible_timeout(struct completion *x, 4769 wait_for_completion_interruptible_timeout(struct completion *x,
4770 unsigned long timeout) 4770 unsigned long timeout)
4771 { 4771 {
4772 return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); 4772 return wait_for_common(x, timeout, TASK_INTERRUPTIBLE);
4773 } 4773 }
4774 EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); 4774 EXPORT_SYMBOL(wait_for_completion_interruptible_timeout);
4775 4775
4776 /** 4776 /**
4777 * wait_for_completion_killable: - waits for completion of a task (killable) 4777 * wait_for_completion_killable: - waits for completion of a task (killable)
4778 * @x: holds the state of this particular completion 4778 * @x: holds the state of this particular completion
4779 * 4779 *
4780 * This waits to be signaled for completion of a specific task. It can be 4780 * This waits to be signaled for completion of a specific task. It can be
4781 * interrupted by a kill signal. 4781 * interrupted by a kill signal.
4782 */ 4782 */
4783 int __sched wait_for_completion_killable(struct completion *x) 4783 int __sched wait_for_completion_killable(struct completion *x)
4784 { 4784 {
4785 long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); 4785 long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE);
4786 if (t == -ERESTARTSYS) 4786 if (t == -ERESTARTSYS)
4787 return t; 4787 return t;
4788 return 0; 4788 return 0;
4789 } 4789 }
4790 EXPORT_SYMBOL(wait_for_completion_killable); 4790 EXPORT_SYMBOL(wait_for_completion_killable);
4791 4791
4792 /** 4792 /**
4793 * try_wait_for_completion - try to decrement a completion without blocking 4793 * try_wait_for_completion - try to decrement a completion without blocking
4794 * @x: completion structure 4794 * @x: completion structure
4795 * 4795 *
4796 * Returns: 0 if a decrement cannot be done without blocking 4796 * Returns: 0 if a decrement cannot be done without blocking
4797 * 1 if a decrement succeeded. 4797 * 1 if a decrement succeeded.
4798 * 4798 *
4799 * If a completion is being used as a counting completion, 4799 * If a completion is being used as a counting completion,
4800 * attempt to decrement the counter without blocking. This 4800 * attempt to decrement the counter without blocking. This
4801 * enables us to avoid waiting if the resource the completion 4801 * enables us to avoid waiting if the resource the completion
4802 * is protecting is not available. 4802 * is protecting is not available.
4803 */ 4803 */
4804 bool try_wait_for_completion(struct completion *x) 4804 bool try_wait_for_completion(struct completion *x)
4805 { 4805 {
4806 int ret = 1; 4806 int ret = 1;
4807 4807
4808 spin_lock_irq(&x->wait.lock); 4808 spin_lock_irq(&x->wait.lock);
4809 if (!x->done) 4809 if (!x->done)
4810 ret = 0; 4810 ret = 0;
4811 else 4811 else
4812 x->done--; 4812 x->done--;
4813 spin_unlock_irq(&x->wait.lock); 4813 spin_unlock_irq(&x->wait.lock);
4814 return ret; 4814 return ret;
4815 } 4815 }
4816 EXPORT_SYMBOL(try_wait_for_completion); 4816 EXPORT_SYMBOL(try_wait_for_completion);
4817 4817
4818 /** 4818 /**
4819 * completion_done - Test to see if a completion has any waiters 4819 * completion_done - Test to see if a completion has any waiters
4820 * @x: completion structure 4820 * @x: completion structure
4821 * 4821 *
4822 * Returns: 0 if there are waiters (wait_for_completion() in progress) 4822 * Returns: 0 if there are waiters (wait_for_completion() in progress)
4823 * 1 if there are no waiters. 4823 * 1 if there are no waiters.
4824 * 4824 *
4825 */ 4825 */
4826 bool completion_done(struct completion *x) 4826 bool completion_done(struct completion *x)
4827 { 4827 {
4828 int ret = 1; 4828 int ret = 1;
4829 4829
4830 spin_lock_irq(&x->wait.lock); 4830 spin_lock_irq(&x->wait.lock);
4831 if (!x->done) 4831 if (!x->done)
4832 ret = 0; 4832 ret = 0;
4833 spin_unlock_irq(&x->wait.lock); 4833 spin_unlock_irq(&x->wait.lock);
4834 return ret; 4834 return ret;
4835 } 4835 }
4836 EXPORT_SYMBOL(completion_done); 4836 EXPORT_SYMBOL(completion_done);
4837 4837
4838 static long __sched 4838 static long __sched
4839 sleep_on_common(wait_queue_head_t *q, int state, long timeout) 4839 sleep_on_common(wait_queue_head_t *q, int state, long timeout)
4840 { 4840 {
4841 unsigned long flags; 4841 unsigned long flags;
4842 wait_queue_t wait; 4842 wait_queue_t wait;
4843 4843
4844 init_waitqueue_entry(&wait, current); 4844 init_waitqueue_entry(&wait, current);
4845 4845
4846 __set_current_state(state); 4846 __set_current_state(state);
4847 4847
4848 spin_lock_irqsave(&q->lock, flags); 4848 spin_lock_irqsave(&q->lock, flags);
4849 __add_wait_queue(q, &wait); 4849 __add_wait_queue(q, &wait);
4850 spin_unlock(&q->lock); 4850 spin_unlock(&q->lock);
4851 timeout = schedule_timeout(timeout); 4851 timeout = schedule_timeout(timeout);
4852 spin_lock_irq(&q->lock); 4852 spin_lock_irq(&q->lock);
4853 __remove_wait_queue(q, &wait); 4853 __remove_wait_queue(q, &wait);
4854 spin_unlock_irqrestore(&q->lock, flags); 4854 spin_unlock_irqrestore(&q->lock, flags);
4855 4855
4856 return timeout; 4856 return timeout;
4857 } 4857 }
4858 4858
4859 void __sched interruptible_sleep_on(wait_queue_head_t *q) 4859 void __sched interruptible_sleep_on(wait_queue_head_t *q)
4860 { 4860 {
4861 sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); 4861 sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
4862 } 4862 }
4863 EXPORT_SYMBOL(interruptible_sleep_on); 4863 EXPORT_SYMBOL(interruptible_sleep_on);
4864 4864
4865 long __sched 4865 long __sched
4866 interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) 4866 interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout)
4867 { 4867 {
4868 return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); 4868 return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout);
4869 } 4869 }
4870 EXPORT_SYMBOL(interruptible_sleep_on_timeout); 4870 EXPORT_SYMBOL(interruptible_sleep_on_timeout);
4871 4871
4872 void __sched sleep_on(wait_queue_head_t *q) 4872 void __sched sleep_on(wait_queue_head_t *q)
4873 { 4873 {
4874 sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); 4874 sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
4875 } 4875 }
4876 EXPORT_SYMBOL(sleep_on); 4876 EXPORT_SYMBOL(sleep_on);
4877 4877
4878 long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) 4878 long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout)
4879 { 4879 {
4880 return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); 4880 return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout);
4881 } 4881 }
4882 EXPORT_SYMBOL(sleep_on_timeout); 4882 EXPORT_SYMBOL(sleep_on_timeout);
4883 4883
4884 #ifdef CONFIG_RT_MUTEXES 4884 #ifdef CONFIG_RT_MUTEXES
4885 4885
4886 /* 4886 /*
4887 * rt_mutex_setprio - set the current priority of a task 4887 * rt_mutex_setprio - set the current priority of a task
4888 * @p: task 4888 * @p: task
4889 * @prio: prio value (kernel-internal form) 4889 * @prio: prio value (kernel-internal form)
4890 * 4890 *
4891 * This function changes the 'effective' priority of a task. It does 4891 * This function changes the 'effective' priority of a task. It does
4892 * not touch ->normal_prio like __setscheduler(). 4892 * not touch ->normal_prio like __setscheduler().
4893 * 4893 *
4894 * Used by the rt_mutex code to implement priority inheritance logic. 4894 * Used by the rt_mutex code to implement priority inheritance logic.
4895 */ 4895 */
4896 void rt_mutex_setprio(struct task_struct *p, int prio) 4896 void rt_mutex_setprio(struct task_struct *p, int prio)
4897 { 4897 {
4898 unsigned long flags; 4898 unsigned long flags;
4899 int oldprio, on_rq, running; 4899 int oldprio, on_rq, running;
4900 struct rq *rq; 4900 struct rq *rq;
4901 const struct sched_class *prev_class = p->sched_class; 4901 const struct sched_class *prev_class = p->sched_class;
4902 4902
4903 BUG_ON(prio < 0 || prio > MAX_PRIO); 4903 BUG_ON(prio < 0 || prio > MAX_PRIO);
4904 4904
4905 rq = task_rq_lock(p, &flags); 4905 rq = task_rq_lock(p, &flags);
4906 update_rq_clock(rq); 4906 update_rq_clock(rq);
4907 4907
4908 oldprio = p->prio; 4908 oldprio = p->prio;
4909 on_rq = p->se.on_rq; 4909 on_rq = p->se.on_rq;
4910 running = task_current(rq, p); 4910 running = task_current(rq, p);
4911 if (on_rq) 4911 if (on_rq)
4912 dequeue_task(rq, p, 0); 4912 dequeue_task(rq, p, 0);
4913 if (running) 4913 if (running)
4914 p->sched_class->put_prev_task(rq, p); 4914 p->sched_class->put_prev_task(rq, p);
4915 4915
4916 if (rt_prio(prio)) 4916 if (rt_prio(prio))
4917 p->sched_class = &rt_sched_class; 4917 p->sched_class = &rt_sched_class;
4918 else 4918 else
4919 p->sched_class = &fair_sched_class; 4919 p->sched_class = &fair_sched_class;
4920 4920
4921 p->prio = prio; 4921 p->prio = prio;
4922 4922
4923 if (running) 4923 if (running)
4924 p->sched_class->set_curr_task(rq); 4924 p->sched_class->set_curr_task(rq);
4925 if (on_rq) { 4925 if (on_rq) {
4926 enqueue_task(rq, p, 0); 4926 enqueue_task(rq, p, 0);
4927 4927
4928 check_class_changed(rq, p, prev_class, oldprio, running); 4928 check_class_changed(rq, p, prev_class, oldprio, running);
4929 } 4929 }
4930 task_rq_unlock(rq, &flags); 4930 task_rq_unlock(rq, &flags);
4931 } 4931 }
4932 4932
4933 #endif 4933 #endif
4934 4934
4935 void set_user_nice(struct task_struct *p, long nice) 4935 void set_user_nice(struct task_struct *p, long nice)
4936 { 4936 {
4937 int old_prio, delta, on_rq; 4937 int old_prio, delta, on_rq;
4938 unsigned long flags; 4938 unsigned long flags;
4939 struct rq *rq; 4939 struct rq *rq;
4940 4940
4941 if (TASK_NICE(p) == nice || nice < -20 || nice > 19) 4941 if (TASK_NICE(p) == nice || nice < -20 || nice > 19)
4942 return; 4942 return;
4943 /* 4943 /*
4944 * We have to be careful, if called from sys_setpriority(), 4944 * We have to be careful, if called from sys_setpriority(),
4945 * the task might be in the middle of scheduling on another CPU. 4945 * the task might be in the middle of scheduling on another CPU.
4946 */ 4946 */
4947 rq = task_rq_lock(p, &flags); 4947 rq = task_rq_lock(p, &flags);
4948 update_rq_clock(rq); 4948 update_rq_clock(rq);
4949 /* 4949 /*
4950 * The RT priorities are set via sched_setscheduler(), but we still 4950 * The RT priorities are set via sched_setscheduler(), but we still
4951 * allow the 'normal' nice value to be set - but as expected 4951 * allow the 'normal' nice value to be set - but as expected
4952 * it wont have any effect on scheduling until the task is 4952 * it wont have any effect on scheduling until the task is
4953 * SCHED_FIFO/SCHED_RR: 4953 * SCHED_FIFO/SCHED_RR:
4954 */ 4954 */
4955 if (task_has_rt_policy(p)) { 4955 if (task_has_rt_policy(p)) {
4956 p->static_prio = NICE_TO_PRIO(nice); 4956 p->static_prio = NICE_TO_PRIO(nice);
4957 goto out_unlock; 4957 goto out_unlock;
4958 } 4958 }
4959 on_rq = p->se.on_rq; 4959 on_rq = p->se.on_rq;
4960 if (on_rq) 4960 if (on_rq)
4961 dequeue_task(rq, p, 0); 4961 dequeue_task(rq, p, 0);
4962 4962
4963 p->static_prio = NICE_TO_PRIO(nice); 4963 p->static_prio = NICE_TO_PRIO(nice);
4964 set_load_weight(p); 4964 set_load_weight(p);
4965 old_prio = p->prio; 4965 old_prio = p->prio;
4966 p->prio = effective_prio(p); 4966 p->prio = effective_prio(p);
4967 delta = p->prio - old_prio; 4967 delta = p->prio - old_prio;
4968 4968
4969 if (on_rq) { 4969 if (on_rq) {
4970 enqueue_task(rq, p, 0); 4970 enqueue_task(rq, p, 0);
4971 /* 4971 /*
4972 * If the task increased its priority or is running and 4972 * If the task increased its priority or is running and
4973 * lowered its priority, then reschedule its CPU: 4973 * lowered its priority, then reschedule its CPU:
4974 */ 4974 */
4975 if (delta < 0 || (delta > 0 && task_running(rq, p))) 4975 if (delta < 0 || (delta > 0 && task_running(rq, p)))
4976 resched_task(rq->curr); 4976 resched_task(rq->curr);
4977 } 4977 }
4978 out_unlock: 4978 out_unlock:
4979 task_rq_unlock(rq, &flags); 4979 task_rq_unlock(rq, &flags);
4980 } 4980 }
4981 EXPORT_SYMBOL(set_user_nice); 4981 EXPORT_SYMBOL(set_user_nice);
4982 4982
4983 /* 4983 /*
4984 * can_nice - check if a task can reduce its nice value 4984 * can_nice - check if a task can reduce its nice value
4985 * @p: task 4985 * @p: task
4986 * @nice: nice value 4986 * @nice: nice value
4987 */ 4987 */
4988 int can_nice(const struct task_struct *p, const int nice) 4988 int can_nice(const struct task_struct *p, const int nice)
4989 { 4989 {
4990 /* convert nice value [19,-20] to rlimit style value [1,40] */ 4990 /* convert nice value [19,-20] to rlimit style value [1,40] */
4991 int nice_rlim = 20 - nice; 4991 int nice_rlim = 20 - nice;
4992 4992
4993 return (nice_rlim <= p->signal->rlim[RLIMIT_NICE].rlim_cur || 4993 return (nice_rlim <= p->signal->rlim[RLIMIT_NICE].rlim_cur ||
4994 capable(CAP_SYS_NICE)); 4994 capable(CAP_SYS_NICE));
4995 } 4995 }
4996 4996
4997 #ifdef __ARCH_WANT_SYS_NICE 4997 #ifdef __ARCH_WANT_SYS_NICE
4998 4998
4999 /* 4999 /*
5000 * sys_nice - change the priority of the current process. 5000 * sys_nice - change the priority of the current process.
5001 * @increment: priority increment 5001 * @increment: priority increment
5002 * 5002 *
5003 * sys_setpriority is a more generic, but much slower function that 5003 * sys_setpriority is a more generic, but much slower function that
5004 * does similar things. 5004 * does similar things.
5005 */ 5005 */
5006 asmlinkage long sys_nice(int increment) 5006 asmlinkage long sys_nice(int increment)
5007 { 5007 {
5008 long nice, retval; 5008 long nice, retval;
5009 5009
5010 /* 5010 /*
5011 * Setpriority might change our priority at the same moment. 5011 * Setpriority might change our priority at the same moment.
5012 * We don't have to worry. Conceptually one call occurs first 5012 * We don't have to worry. Conceptually one call occurs first
5013 * and we have a single winner. 5013 * and we have a single winner.
5014 */ 5014 */
5015 if (increment < -40) 5015 if (increment < -40)
5016 increment = -40; 5016 increment = -40;
5017 if (increment > 40) 5017 if (increment > 40)
5018 increment = 40; 5018 increment = 40;
5019 5019
5020 nice = PRIO_TO_NICE(current->static_prio) + increment; 5020 nice = PRIO_TO_NICE(current->static_prio) + increment;
5021 if (nice < -20) 5021 if (nice < -20)
5022 nice = -20; 5022 nice = -20;
5023 if (nice > 19) 5023 if (nice > 19)
5024 nice = 19; 5024 nice = 19;
5025 5025
5026 if (increment < 0 && !can_nice(current, nice)) 5026 if (increment < 0 && !can_nice(current, nice))
5027 return -EPERM; 5027 return -EPERM;
5028 5028
5029 retval = security_task_setnice(current, nice); 5029 retval = security_task_setnice(current, nice);
5030 if (retval) 5030 if (retval)
5031 return retval; 5031 return retval;
5032 5032
5033 set_user_nice(current, nice); 5033 set_user_nice(current, nice);
5034 return 0; 5034 return 0;
5035 } 5035 }
5036 5036
5037 #endif 5037 #endif
5038 5038
5039 /** 5039 /**
5040 * task_prio - return the priority value of a given task. 5040 * task_prio - return the priority value of a given task.
5041 * @p: the task in question. 5041 * @p: the task in question.
5042 * 5042 *
5043 * This is the priority value as seen by users in /proc. 5043 * This is the priority value as seen by users in /proc.
5044 * RT tasks are offset by -200. Normal tasks are centered 5044 * RT tasks are offset by -200. Normal tasks are centered
5045 * around 0, value goes from -16 to +15. 5045 * around 0, value goes from -16 to +15.
5046 */ 5046 */
5047 int task_prio(const struct task_struct *p) 5047 int task_prio(const struct task_struct *p)
5048 { 5048 {
5049 return p->prio - MAX_RT_PRIO; 5049 return p->prio - MAX_RT_PRIO;
5050 } 5050 }
5051 5051
5052 /** 5052 /**
5053 * task_nice - return the nice value of a given task. 5053 * task_nice - return the nice value of a given task.
5054 * @p: the task in question. 5054 * @p: the task in question.
5055 */ 5055 */
5056 int task_nice(const struct task_struct *p) 5056 int task_nice(const struct task_struct *p)
5057 { 5057 {
5058 return TASK_NICE(p); 5058 return TASK_NICE(p);
5059 } 5059 }
5060 EXPORT_SYMBOL(task_nice); 5060 EXPORT_SYMBOL(task_nice);
5061 5061
5062 /** 5062 /**
5063 * idle_cpu - is a given cpu idle currently? 5063 * idle_cpu - is a given cpu idle currently?
5064 * @cpu: the processor in question. 5064 * @cpu: the processor in question.
5065 */ 5065 */
5066 int idle_cpu(int cpu) 5066 int idle_cpu(int cpu)
5067 { 5067 {
5068 return cpu_curr(cpu) == cpu_rq(cpu)->idle; 5068 return cpu_curr(cpu) == cpu_rq(cpu)->idle;
5069 } 5069 }
5070 5070
5071 /** 5071 /**
5072 * idle_task - return the idle task for a given cpu. 5072 * idle_task - return the idle task for a given cpu.
5073 * @cpu: the processor in question. 5073 * @cpu: the processor in question.
5074 */ 5074 */
5075 struct task_struct *idle_task(int cpu) 5075 struct task_struct *idle_task(int cpu)
5076 { 5076 {
5077 return cpu_rq(cpu)->idle; 5077 return cpu_rq(cpu)->idle;
5078 } 5078 }
5079 5079
5080 /** 5080 /**
5081 * find_process_by_pid - find a process with a matching PID value. 5081 * find_process_by_pid - find a process with a matching PID value.
5082 * @pid: the pid in question. 5082 * @pid: the pid in question.
5083 */ 5083 */
5084 static struct task_struct *find_process_by_pid(pid_t pid) 5084 static struct task_struct *find_process_by_pid(pid_t pid)
5085 { 5085 {
5086 return pid ? find_task_by_vpid(pid) : current; 5086 return pid ? find_task_by_vpid(pid) : current;
5087 } 5087 }
5088 5088
5089 /* Actually do priority change: must hold rq lock. */ 5089 /* Actually do priority change: must hold rq lock. */
5090 static void 5090 static void
5091 __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) 5091 __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio)
5092 { 5092 {
5093 BUG_ON(p->se.on_rq); 5093 BUG_ON(p->se.on_rq);
5094 5094
5095 p->policy = policy; 5095 p->policy = policy;
5096 switch (p->policy) { 5096 switch (p->policy) {
5097 case SCHED_NORMAL: 5097 case SCHED_NORMAL:
5098 case SCHED_BATCH: 5098 case SCHED_BATCH:
5099 case SCHED_IDLE: 5099 case SCHED_IDLE:
5100 p->sched_class = &fair_sched_class; 5100 p->sched_class = &fair_sched_class;
5101 break; 5101 break;
5102 case SCHED_FIFO: 5102 case SCHED_FIFO:
5103 case SCHED_RR: 5103 case SCHED_RR:
5104 p->sched_class = &rt_sched_class; 5104 p->sched_class = &rt_sched_class;
5105 break; 5105 break;
5106 } 5106 }
5107 5107
5108 p->rt_priority = prio; 5108 p->rt_priority = prio;
5109 p->normal_prio = normal_prio(p); 5109 p->normal_prio = normal_prio(p);
5110 /* we are holding p->pi_lock already */ 5110 /* we are holding p->pi_lock already */
5111 p->prio = rt_mutex_getprio(p); 5111 p->prio = rt_mutex_getprio(p);
5112 set_load_weight(p); 5112 set_load_weight(p);
5113 } 5113 }
5114 5114
5115 static int __sched_setscheduler(struct task_struct *p, int policy, 5115 static int __sched_setscheduler(struct task_struct *p, int policy,
5116 struct sched_param *param, bool user) 5116 struct sched_param *param, bool user)
5117 { 5117 {
5118 int retval, oldprio, oldpolicy = -1, on_rq, running; 5118 int retval, oldprio, oldpolicy = -1, on_rq, running;
5119 unsigned long flags; 5119 unsigned long flags;
5120 const struct sched_class *prev_class = p->sched_class; 5120 const struct sched_class *prev_class = p->sched_class;
5121 struct rq *rq; 5121 struct rq *rq;
5122 5122
5123 /* may grab non-irq protected spin_locks */ 5123 /* may grab non-irq protected spin_locks */
5124 BUG_ON(in_interrupt()); 5124 BUG_ON(in_interrupt());
5125 recheck: 5125 recheck:
5126 /* double check policy once rq lock held */ 5126 /* double check policy once rq lock held */
5127 if (policy < 0) 5127 if (policy < 0)
5128 policy = oldpolicy = p->policy; 5128 policy = oldpolicy = p->policy;
5129 else if (policy != SCHED_FIFO && policy != SCHED_RR && 5129 else if (policy != SCHED_FIFO && policy != SCHED_RR &&
5130 policy != SCHED_NORMAL && policy != SCHED_BATCH && 5130 policy != SCHED_NORMAL && policy != SCHED_BATCH &&
5131 policy != SCHED_IDLE) 5131 policy != SCHED_IDLE)
5132 return -EINVAL; 5132 return -EINVAL;
5133 /* 5133 /*
5134 * Valid priorities for SCHED_FIFO and SCHED_RR are 5134 * Valid priorities for SCHED_FIFO and SCHED_RR are
5135 * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, 5135 * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL,
5136 * SCHED_BATCH and SCHED_IDLE is 0. 5136 * SCHED_BATCH and SCHED_IDLE is 0.
5137 */ 5137 */
5138 if (param->sched_priority < 0 || 5138 if (param->sched_priority < 0 ||
5139 (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || 5139 (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) ||
5140 (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) 5140 (!p->mm && param->sched_priority > MAX_RT_PRIO-1))
5141 return -EINVAL; 5141 return -EINVAL;
5142 if (rt_policy(policy) != (param->sched_priority != 0)) 5142 if (rt_policy(policy) != (param->sched_priority != 0))
5143 return -EINVAL; 5143 return -EINVAL;
5144 5144
5145 /* 5145 /*
5146 * Allow unprivileged RT tasks to decrease priority: 5146 * Allow unprivileged RT tasks to decrease priority:
5147 */ 5147 */
5148 if (user && !capable(CAP_SYS_NICE)) { 5148 if (user && !capable(CAP_SYS_NICE)) {
5149 if (rt_policy(policy)) { 5149 if (rt_policy(policy)) {
5150 unsigned long rlim_rtprio; 5150 unsigned long rlim_rtprio;
5151 5151
5152 if (!lock_task_sighand(p, &flags)) 5152 if (!lock_task_sighand(p, &flags))
5153 return -ESRCH; 5153 return -ESRCH;
5154 rlim_rtprio = p->signal->rlim[RLIMIT_RTPRIO].rlim_cur; 5154 rlim_rtprio = p->signal->rlim[RLIMIT_RTPRIO].rlim_cur;
5155 unlock_task_sighand(p, &flags); 5155 unlock_task_sighand(p, &flags);
5156 5156
5157 /* can't set/change the rt policy */ 5157 /* can't set/change the rt policy */
5158 if (policy != p->policy && !rlim_rtprio) 5158 if (policy != p->policy && !rlim_rtprio)
5159 return -EPERM; 5159 return -EPERM;
5160 5160
5161 /* can't increase priority */ 5161 /* can't increase priority */
5162 if (param->sched_priority > p->rt_priority && 5162 if (param->sched_priority > p->rt_priority &&
5163 param->sched_priority > rlim_rtprio) 5163 param->sched_priority > rlim_rtprio)
5164 return -EPERM; 5164 return -EPERM;
5165 } 5165 }
5166 /* 5166 /*
5167 * Like positive nice levels, dont allow tasks to 5167 * Like positive nice levels, dont allow tasks to
5168 * move out of SCHED_IDLE either: 5168 * move out of SCHED_IDLE either:
5169 */ 5169 */
5170 if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) 5170 if (p->policy == SCHED_IDLE && policy != SCHED_IDLE)
5171 return -EPERM; 5171 return -EPERM;
5172 5172
5173 /* can't change other user's priorities */ 5173 /* can't change other user's priorities */
5174 if ((current->euid != p->euid) && 5174 if ((current->euid != p->euid) &&
5175 (current->euid != p->uid)) 5175 (current->euid != p->uid))
5176 return -EPERM; 5176 return -EPERM;
5177 } 5177 }
5178 5178
5179 if (user) { 5179 if (user) {
5180 #ifdef CONFIG_RT_GROUP_SCHED 5180 #ifdef CONFIG_RT_GROUP_SCHED
5181 /* 5181 /*
5182 * Do not allow realtime tasks into groups that have no runtime 5182 * Do not allow realtime tasks into groups that have no runtime
5183 * assigned. 5183 * assigned.
5184 */ 5184 */
5185 if (rt_bandwidth_enabled() && rt_policy(policy) && 5185 if (rt_bandwidth_enabled() && rt_policy(policy) &&
5186 task_group(p)->rt_bandwidth.rt_runtime == 0) 5186 task_group(p)->rt_bandwidth.rt_runtime == 0)
5187 return -EPERM; 5187 return -EPERM;
5188 #endif 5188 #endif
5189 5189
5190 retval = security_task_setscheduler(p, policy, param); 5190 retval = security_task_setscheduler(p, policy, param);
5191 if (retval) 5191 if (retval)
5192 return retval; 5192 return retval;
5193 } 5193 }
5194 5194
5195 /* 5195 /*
5196 * make sure no PI-waiters arrive (or leave) while we are 5196 * make sure no PI-waiters arrive (or leave) while we are
5197 * changing the priority of the task: 5197 * changing the priority of the task:
5198 */ 5198 */
5199 spin_lock_irqsave(&p->pi_lock, flags); 5199 spin_lock_irqsave(&p->pi_lock, flags);
5200 /* 5200 /*
5201 * To be able to change p->policy safely, the apropriate 5201 * To be able to change p->policy safely, the apropriate
5202 * runqueue lock must be held. 5202 * runqueue lock must be held.
5203 */ 5203 */
5204 rq = __task_rq_lock(p); 5204 rq = __task_rq_lock(p);
5205 /* recheck policy now with rq lock held */ 5205 /* recheck policy now with rq lock held */
5206 if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { 5206 if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) {
5207 policy = oldpolicy = -1; 5207 policy = oldpolicy = -1;
5208 __task_rq_unlock(rq); 5208 __task_rq_unlock(rq);
5209 spin_unlock_irqrestore(&p->pi_lock, flags); 5209 spin_unlock_irqrestore(&p->pi_lock, flags);
5210 goto recheck; 5210 goto recheck;
5211 } 5211 }
5212 update_rq_clock(rq); 5212 update_rq_clock(rq);
5213 on_rq = p->se.on_rq; 5213 on_rq = p->se.on_rq;
5214 running = task_current(rq, p); 5214 running = task_current(rq, p);
5215 if (on_rq) 5215 if (on_rq)
5216 deactivate_task(rq, p, 0); 5216 deactivate_task(rq, p, 0);
5217 if (running) 5217 if (running)
5218 p->sched_class->put_prev_task(rq, p); 5218 p->sched_class->put_prev_task(rq, p);
5219 5219
5220 oldprio = p->prio; 5220 oldprio = p->prio;
5221 __setscheduler(rq, p, policy, param->sched_priority); 5221 __setscheduler(rq, p, policy, param->sched_priority);
5222 5222
5223 if (running) 5223 if (running)
5224 p->sched_class->set_curr_task(rq); 5224 p->sched_class->set_curr_task(rq);
5225 if (on_rq) { 5225 if (on_rq) {
5226 activate_task(rq, p, 0); 5226 activate_task(rq, p, 0);
5227 5227
5228 check_class_changed(rq, p, prev_class, oldprio, running); 5228 check_class_changed(rq, p, prev_class, oldprio, running);
5229 } 5229 }
5230 __task_rq_unlock(rq); 5230 __task_rq_unlock(rq);
5231 spin_unlock_irqrestore(&p->pi_lock, flags); 5231 spin_unlock_irqrestore(&p->pi_lock, flags);
5232 5232
5233 rt_mutex_adjust_pi(p); 5233 rt_mutex_adjust_pi(p);
5234 5234
5235 return 0; 5235 return 0;
5236 } 5236 }
5237 5237
5238 /** 5238 /**
5239 * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. 5239 * sched_setscheduler - change the scheduling policy and/or RT priority of a thread.
5240 * @p: the task in question. 5240 * @p: the task in question.
5241 * @policy: new policy. 5241 * @policy: new policy.
5242 * @param: structure containing the new RT priority. 5242 * @param: structure containing the new RT priority.
5243 * 5243 *
5244 * NOTE that the task may be already dead. 5244 * NOTE that the task may be already dead.
5245 */ 5245 */
5246 int sched_setscheduler(struct task_struct *p, int policy, 5246 int sched_setscheduler(struct task_struct *p, int policy,
5247 struct sched_param *param) 5247 struct sched_param *param)
5248 { 5248 {
5249 return __sched_setscheduler(p, policy, param, true); 5249 return __sched_setscheduler(p, policy, param, true);
5250 } 5250 }
5251 EXPORT_SYMBOL_GPL(sched_setscheduler); 5251 EXPORT_SYMBOL_GPL(sched_setscheduler);
5252 5252
5253 /** 5253 /**
5254 * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. 5254 * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace.
5255 * @p: the task in question. 5255 * @p: the task in question.
5256 * @policy: new policy. 5256 * @policy: new policy.
5257 * @param: structure containing the new RT priority. 5257 * @param: structure containing the new RT priority.
5258 * 5258 *
5259 * Just like sched_setscheduler, only don't bother checking if the 5259 * Just like sched_setscheduler, only don't bother checking if the
5260 * current context has permission. For example, this is needed in 5260 * current context has permission. For example, this is needed in
5261 * stop_machine(): we create temporary high priority worker threads, 5261 * stop_machine(): we create temporary high priority worker threads,
5262 * but our caller might not have that capability. 5262 * but our caller might not have that capability.
5263 */ 5263 */
5264 int sched_setscheduler_nocheck(struct task_struct *p, int policy, 5264 int sched_setscheduler_nocheck(struct task_struct *p, int policy,
5265 struct sched_param *param) 5265 struct sched_param *param)
5266 { 5266 {
5267 return __sched_setscheduler(p, policy, param, false); 5267 return __sched_setscheduler(p, policy, param, false);
5268 } 5268 }
5269 5269
5270 static int 5270 static int
5271 do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) 5271 do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param)
5272 { 5272 {
5273 struct sched_param lparam; 5273 struct sched_param lparam;
5274 struct task_struct *p; 5274 struct task_struct *p;
5275 int retval; 5275 int retval;
5276 5276
5277 if (!param || pid < 0) 5277 if (!param || pid < 0)
5278 return -EINVAL; 5278 return -EINVAL;
5279 if (copy_from_user(&lparam, param, sizeof(struct sched_param))) 5279 if (copy_from_user(&lparam, param, sizeof(struct sched_param)))
5280 return -EFAULT; 5280 return -EFAULT;
5281 5281
5282 rcu_read_lock(); 5282 rcu_read_lock();
5283 retval = -ESRCH; 5283 retval = -ESRCH;
5284 p = find_process_by_pid(pid); 5284 p = find_process_by_pid(pid);
5285 if (p != NULL) 5285 if (p != NULL)
5286 retval = sched_setscheduler(p, policy, &lparam); 5286 retval = sched_setscheduler(p, policy, &lparam);
5287 rcu_read_unlock(); 5287 rcu_read_unlock();
5288 5288
5289 return retval; 5289 return retval;
5290 } 5290 }
5291 5291
5292 /** 5292 /**
5293 * sys_sched_setscheduler - set/change the scheduler policy and RT priority 5293 * sys_sched_setscheduler - set/change the scheduler policy and RT priority
5294 * @pid: the pid in question. 5294 * @pid: the pid in question.
5295 * @policy: new policy. 5295 * @policy: new policy.
5296 * @param: structure containing the new RT priority. 5296 * @param: structure containing the new RT priority.
5297 */ 5297 */
5298 asmlinkage long 5298 asmlinkage long
5299 sys_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) 5299 sys_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param)
5300 { 5300 {
5301 /* negative values for policy are not valid */ 5301 /* negative values for policy are not valid */
5302 if (policy < 0) 5302 if (policy < 0)
5303 return -EINVAL; 5303 return -EINVAL;
5304 5304
5305 return do_sched_setscheduler(pid, policy, param); 5305 return do_sched_setscheduler(pid, policy, param);
5306 } 5306 }
5307 5307
5308 /** 5308 /**
5309 * sys_sched_setparam - set/change the RT priority of a thread 5309 * sys_sched_setparam - set/change the RT priority of a thread
5310 * @pid: the pid in question. 5310 * @pid: the pid in question.
5311 * @param: structure containing the new RT priority. 5311 * @param: structure containing the new RT priority.
5312 */ 5312 */
5313 asmlinkage long sys_sched_setparam(pid_t pid, struct sched_param __user *param) 5313 asmlinkage long sys_sched_setparam(pid_t pid, struct sched_param __user *param)
5314 { 5314 {
5315 return do_sched_setscheduler(pid, -1, param); 5315 return do_sched_setscheduler(pid, -1, param);
5316 } 5316 }
5317 5317
5318 /** 5318 /**
5319 * sys_sched_getscheduler - get the policy (scheduling class) of a thread 5319 * sys_sched_getscheduler - get the policy (scheduling class) of a thread
5320 * @pid: the pid in question. 5320 * @pid: the pid in question.
5321 */ 5321 */
5322 asmlinkage long sys_sched_getscheduler(pid_t pid) 5322 asmlinkage long sys_sched_getscheduler(pid_t pid)
5323 { 5323 {
5324 struct task_struct *p; 5324 struct task_struct *p;
5325 int retval; 5325 int retval;
5326 5326
5327 if (pid < 0) 5327 if (pid < 0)
5328 return -EINVAL; 5328 return -EINVAL;
5329 5329
5330 retval = -ESRCH; 5330 retval = -ESRCH;
5331 read_lock(&tasklist_lock); 5331 read_lock(&tasklist_lock);
5332 p = find_process_by_pid(pid); 5332 p = find_process_by_pid(pid);
5333 if (p) { 5333 if (p) {
5334 retval = security_task_getscheduler(p); 5334 retval = security_task_getscheduler(p);
5335 if (!retval) 5335 if (!retval)
5336 retval = p->policy; 5336 retval = p->policy;
5337 } 5337 }
5338 read_unlock(&tasklist_lock); 5338 read_unlock(&tasklist_lock);
5339 return retval; 5339 return retval;
5340 } 5340 }
5341 5341
5342 /** 5342 /**
5343 * sys_sched_getscheduler - get the RT priority of a thread 5343 * sys_sched_getscheduler - get the RT priority of a thread
5344 * @pid: the pid in question. 5344 * @pid: the pid in question.
5345 * @param: structure containing the RT priority. 5345 * @param: structure containing the RT priority.
5346 */ 5346 */
5347 asmlinkage long sys_sched_getparam(pid_t pid, struct sched_param __user *param) 5347 asmlinkage long sys_sched_getparam(pid_t pid, struct sched_param __user *param)
5348 { 5348 {
5349 struct sched_param lp; 5349 struct sched_param lp;
5350 struct task_struct *p; 5350 struct task_struct *p;
5351 int retval; 5351 int retval;
5352 5352
5353 if (!param || pid < 0) 5353 if (!param || pid < 0)
5354 return -EINVAL; 5354 return -EINVAL;
5355 5355
5356 read_lock(&tasklist_lock); 5356 read_lock(&tasklist_lock);
5357 p = find_process_by_pid(pid); 5357 p = find_process_by_pid(pid);
5358 retval = -ESRCH; 5358 retval = -ESRCH;
5359 if (!p) 5359 if (!p)
5360 goto out_unlock; 5360 goto out_unlock;
5361 5361
5362 retval = security_task_getscheduler(p); 5362 retval = security_task_getscheduler(p);
5363 if (retval) 5363 if (retval)
5364 goto out_unlock; 5364 goto out_unlock;
5365 5365
5366 lp.sched_priority = p->rt_priority; 5366 lp.sched_priority = p->rt_priority;
5367 read_unlock(&tasklist_lock); 5367 read_unlock(&tasklist_lock);
5368 5368
5369 /* 5369 /*
5370 * This one might sleep, we cannot do it with a spinlock held ... 5370 * This one might sleep, we cannot do it with a spinlock held ...
5371 */ 5371 */
5372 retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; 5372 retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0;
5373 5373
5374 return retval; 5374 return retval;
5375 5375
5376 out_unlock: 5376 out_unlock:
5377 read_unlock(&tasklist_lock); 5377 read_unlock(&tasklist_lock);
5378 return retval; 5378 return retval;
5379 } 5379 }
5380 5380
5381 long sched_setaffinity(pid_t pid, const cpumask_t *in_mask) 5381 long sched_setaffinity(pid_t pid, const cpumask_t *in_mask)
5382 { 5382 {
5383 cpumask_t cpus_allowed; 5383 cpumask_t cpus_allowed;
5384 cpumask_t new_mask = *in_mask; 5384 cpumask_t new_mask = *in_mask;
5385 struct task_struct *p; 5385 struct task_struct *p;
5386 int retval; 5386 int retval;
5387 5387
5388 get_online_cpus(); 5388 get_online_cpus();
5389 read_lock(&tasklist_lock); 5389 read_lock(&tasklist_lock);
5390 5390
5391 p = find_process_by_pid(pid); 5391 p = find_process_by_pid(pid);
5392 if (!p) { 5392 if (!p) {
5393 read_unlock(&tasklist_lock); 5393 read_unlock(&tasklist_lock);
5394 put_online_cpus(); 5394 put_online_cpus();
5395 return -ESRCH; 5395 return -ESRCH;
5396 } 5396 }
5397 5397
5398 /* 5398 /*
5399 * It is not safe to call set_cpus_allowed with the 5399 * It is not safe to call set_cpus_allowed with the
5400 * tasklist_lock held. We will bump the task_struct's 5400 * tasklist_lock held. We will bump the task_struct's
5401 * usage count and then drop tasklist_lock. 5401 * usage count and then drop tasklist_lock.
5402 */ 5402 */
5403 get_task_struct(p); 5403 get_task_struct(p);
5404 read_unlock(&tasklist_lock); 5404 read_unlock(&tasklist_lock);
5405 5405
5406 retval = -EPERM; 5406 retval = -EPERM;
5407 if ((current->euid != p->euid) && (current->euid != p->uid) && 5407 if ((current->euid != p->euid) && (current->euid != p->uid) &&
5408 !capable(CAP_SYS_NICE)) 5408 !capable(CAP_SYS_NICE))
5409 goto out_unlock; 5409 goto out_unlock;
5410 5410
5411 retval = security_task_setscheduler(p, 0, NULL); 5411 retval = security_task_setscheduler(p, 0, NULL);
5412 if (retval) 5412 if (retval)
5413 goto out_unlock; 5413 goto out_unlock;
5414 5414
5415 cpuset_cpus_allowed(p, &cpus_allowed); 5415 cpuset_cpus_allowed(p, &cpus_allowed);
5416 cpus_and(new_mask, new_mask, cpus_allowed); 5416 cpus_and(new_mask, new_mask, cpus_allowed);
5417 again: 5417 again:
5418 retval = set_cpus_allowed_ptr(p, &new_mask); 5418 retval = set_cpus_allowed_ptr(p, &new_mask);
5419 5419
5420 if (!retval) { 5420 if (!retval) {
5421 cpuset_cpus_allowed(p, &cpus_allowed); 5421 cpuset_cpus_allowed(p, &cpus_allowed);
5422 if (!cpus_subset(new_mask, cpus_allowed)) { 5422 if (!cpus_subset(new_mask, cpus_allowed)) {
5423 /* 5423 /*
5424 * We must have raced with a concurrent cpuset 5424 * We must have raced with a concurrent cpuset
5425 * update. Just reset the cpus_allowed to the 5425 * update. Just reset the cpus_allowed to the
5426 * cpuset's cpus_allowed 5426 * cpuset's cpus_allowed
5427 */ 5427 */
5428 new_mask = cpus_allowed; 5428 new_mask = cpus_allowed;
5429 goto again; 5429 goto again;
5430 } 5430 }
5431 } 5431 }
5432 out_unlock: 5432 out_unlock:
5433 put_task_struct(p); 5433 put_task_struct(p);
5434 put_online_cpus(); 5434 put_online_cpus();
5435 return retval; 5435 return retval;
5436 } 5436 }
5437 5437
5438 static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, 5438 static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len,
5439 cpumask_t *new_mask) 5439 cpumask_t *new_mask)
5440 { 5440 {
5441 if (len < sizeof(cpumask_t)) { 5441 if (len < sizeof(cpumask_t)) {
5442 memset(new_mask, 0, sizeof(cpumask_t)); 5442 memset(new_mask, 0, sizeof(cpumask_t));
5443 } else if (len > sizeof(cpumask_t)) { 5443 } else if (len > sizeof(cpumask_t)) {
5444 len = sizeof(cpumask_t); 5444 len = sizeof(cpumask_t);
5445 } 5445 }
5446 return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; 5446 return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0;
5447 } 5447 }
5448 5448
5449 /** 5449 /**
5450 * sys_sched_setaffinity - set the cpu affinity of a process 5450 * sys_sched_setaffinity - set the cpu affinity of a process
5451 * @pid: pid of the process 5451 * @pid: pid of the process
5452 * @len: length in bytes of the bitmask pointed to by user_mask_ptr 5452 * @len: length in bytes of the bitmask pointed to by user_mask_ptr
5453 * @user_mask_ptr: user-space pointer to the new cpu mask 5453 * @user_mask_ptr: user-space pointer to the new cpu mask
5454 */ 5454 */
5455 asmlinkage long sys_sched_setaffinity(pid_t pid, unsigned int len, 5455 asmlinkage long sys_sched_setaffinity(pid_t pid, unsigned int len,
5456 unsigned long __user *user_mask_ptr) 5456 unsigned long __user *user_mask_ptr)
5457 { 5457 {
5458 cpumask_t new_mask; 5458 cpumask_t new_mask;
5459 int retval; 5459 int retval;
5460 5460
5461 retval = get_user_cpu_mask(user_mask_ptr, len, &new_mask); 5461 retval = get_user_cpu_mask(user_mask_ptr, len, &new_mask);
5462 if (retval) 5462 if (retval)
5463 return retval; 5463 return retval;
5464 5464
5465 return sched_setaffinity(pid, &new_mask); 5465 return sched_setaffinity(pid, &new_mask);
5466 } 5466 }
5467 5467
5468 long sched_getaffinity(pid_t pid, cpumask_t *mask) 5468 long sched_getaffinity(pid_t pid, cpumask_t *mask)
5469 { 5469 {
5470 struct task_struct *p; 5470 struct task_struct *p;
5471 int retval; 5471 int retval;
5472 5472
5473 get_online_cpus(); 5473 get_online_cpus();
5474 read_lock(&tasklist_lock); 5474 read_lock(&tasklist_lock);
5475 5475
5476 retval = -ESRCH; 5476 retval = -ESRCH;
5477 p = find_process_by_pid(pid); 5477 p = find_process_by_pid(pid);
5478 if (!p) 5478 if (!p)
5479 goto out_unlock; 5479 goto out_unlock;
5480 5480
5481 retval = security_task_getscheduler(p); 5481 retval = security_task_getscheduler(p);
5482 if (retval) 5482 if (retval)
5483 goto out_unlock; 5483 goto out_unlock;
5484 5484
5485 cpus_and(*mask, p->cpus_allowed, cpu_online_map); 5485 cpus_and(*mask, p->cpus_allowed, cpu_online_map);
5486 5486
5487 out_unlock: 5487 out_unlock:
5488 read_unlock(&tasklist_lock); 5488 read_unlock(&tasklist_lock);
5489 put_online_cpus(); 5489 put_online_cpus();
5490 5490
5491 return retval; 5491 return retval;
5492 } 5492 }
5493 5493
5494 /** 5494 /**
5495 * sys_sched_getaffinity - get the cpu affinity of a process 5495 * sys_sched_getaffinity - get the cpu affinity of a process
5496 * @pid: pid of the process 5496 * @pid: pid of the process
5497 * @len: length in bytes of the bitmask pointed to by user_mask_ptr 5497 * @len: length in bytes of the bitmask pointed to by user_mask_ptr
5498 * @user_mask_ptr: user-space pointer to hold the current cpu mask 5498 * @user_mask_ptr: user-space pointer to hold the current cpu mask
5499 */ 5499 */
5500 asmlinkage long sys_sched_getaffinity(pid_t pid, unsigned int len, 5500 asmlinkage long sys_sched_getaffinity(pid_t pid, unsigned int len,
5501 unsigned long __user *user_mask_ptr) 5501 unsigned long __user *user_mask_ptr)
5502 { 5502 {
5503 int ret; 5503 int ret;
5504 cpumask_t mask; 5504 cpumask_t mask;
5505 5505
5506 if (len < sizeof(cpumask_t)) 5506 if (len < sizeof(cpumask_t))
5507 return -EINVAL; 5507 return -EINVAL;
5508 5508
5509 ret = sched_getaffinity(pid, &mask); 5509 ret = sched_getaffinity(pid, &mask);
5510 if (ret < 0) 5510 if (ret < 0)
5511 return ret; 5511 return ret;
5512 5512
5513 if (copy_to_user(user_mask_ptr, &mask, sizeof(cpumask_t))) 5513 if (copy_to_user(user_mask_ptr, &mask, sizeof(cpumask_t)))
5514 return -EFAULT; 5514 return -EFAULT;
5515 5515
5516 return sizeof(cpumask_t); 5516 return sizeof(cpumask_t);
5517 } 5517 }
5518 5518
5519 /** 5519 /**
5520 * sys_sched_yield - yield the current processor to other threads. 5520 * sys_sched_yield - yield the current processor to other threads.
5521 * 5521 *
5522 * This function yields the current CPU to other tasks. If there are no 5522 * This function yields the current CPU to other tasks. If there are no
5523 * other threads running on this CPU then this function will return. 5523 * other threads running on this CPU then this function will return.
5524 */ 5524 */
5525 asmlinkage long sys_sched_yield(void) 5525 asmlinkage long sys_sched_yield(void)
5526 { 5526 {
5527 struct rq *rq = this_rq_lock(); 5527 struct rq *rq = this_rq_lock();
5528 5528
5529 schedstat_inc(rq, yld_count); 5529 schedstat_inc(rq, yld_count);
5530 current->sched_class->yield_task(rq); 5530 current->sched_class->yield_task(rq);
5531 5531
5532 /* 5532 /*
5533 * Since we are going to call schedule() anyway, there's 5533 * Since we are going to call schedule() anyway, there's
5534 * no need to preempt or enable interrupts: 5534 * no need to preempt or enable interrupts:
5535 */ 5535 */
5536 __release(rq->lock); 5536 __release(rq->lock);
5537 spin_release(&rq->lock.dep_map, 1, _THIS_IP_); 5537 spin_release(&rq->lock.dep_map, 1, _THIS_IP_);
5538 _raw_spin_unlock(&rq->lock); 5538 _raw_spin_unlock(&rq->lock);
5539 preempt_enable_no_resched(); 5539 preempt_enable_no_resched();
5540 5540
5541 schedule(); 5541 schedule();
5542 5542
5543 return 0; 5543 return 0;
5544 } 5544 }
5545 5545
5546 static void __cond_resched(void) 5546 static void __cond_resched(void)
5547 { 5547 {
5548 #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP 5548 #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP
5549 __might_sleep(__FILE__, __LINE__); 5549 __might_sleep(__FILE__, __LINE__);
5550 #endif 5550 #endif
5551 /* 5551 /*
5552 * The BKS might be reacquired before we have dropped 5552 * The BKS might be reacquired before we have dropped
5553 * PREEMPT_ACTIVE, which could trigger a second 5553 * PREEMPT_ACTIVE, which could trigger a second
5554 * cond_resched() call. 5554 * cond_resched() call.
5555 */ 5555 */
5556 do { 5556 do {
5557 add_preempt_count(PREEMPT_ACTIVE); 5557 add_preempt_count(PREEMPT_ACTIVE);
5558 schedule(); 5558 schedule();
5559 sub_preempt_count(PREEMPT_ACTIVE); 5559 sub_preempt_count(PREEMPT_ACTIVE);
5560 } while (need_resched()); 5560 } while (need_resched());
5561 } 5561 }
5562 5562
5563 int __sched _cond_resched(void) 5563 int __sched _cond_resched(void)
5564 { 5564 {
5565 if (need_resched() && !(preempt_count() & PREEMPT_ACTIVE) && 5565 if (need_resched() && !(preempt_count() & PREEMPT_ACTIVE) &&
5566 system_state == SYSTEM_RUNNING) { 5566 system_state == SYSTEM_RUNNING) {
5567 __cond_resched(); 5567 __cond_resched();
5568 return 1; 5568 return 1;
5569 } 5569 }
5570 return 0; 5570 return 0;
5571 } 5571 }
5572 EXPORT_SYMBOL(_cond_resched); 5572 EXPORT_SYMBOL(_cond_resched);
5573 5573
5574 /* 5574 /*
5575 * cond_resched_lock() - if a reschedule is pending, drop the given lock, 5575 * cond_resched_lock() - if a reschedule is pending, drop the given lock,
5576 * call schedule, and on return reacquire the lock. 5576 * call schedule, and on return reacquire the lock.
5577 * 5577 *
5578 * This works OK both with and without CONFIG_PREEMPT. We do strange low-level 5578 * This works OK both with and without CONFIG_PREEMPT. We do strange low-level
5579 * operations here to prevent schedule() from being called twice (once via 5579 * operations here to prevent schedule() from being called twice (once via
5580 * spin_unlock(), once by hand). 5580 * spin_unlock(), once by hand).
5581 */ 5581 */
5582 int cond_resched_lock(spinlock_t *lock) 5582 int cond_resched_lock(spinlock_t *lock)
5583 { 5583 {
5584 int resched = need_resched() && system_state == SYSTEM_RUNNING; 5584 int resched = need_resched() && system_state == SYSTEM_RUNNING;
5585 int ret = 0; 5585 int ret = 0;
5586 5586
5587 if (spin_needbreak(lock) || resched) { 5587 if (spin_needbreak(lock) || resched) {
5588 spin_unlock(lock); 5588 spin_unlock(lock);
5589 if (resched && need_resched()) 5589 if (resched && need_resched())
5590 __cond_resched(); 5590 __cond_resched();
5591 else 5591 else
5592 cpu_relax(); 5592 cpu_relax();
5593 ret = 1; 5593 ret = 1;
5594 spin_lock(lock); 5594 spin_lock(lock);
5595 } 5595 }
5596 return ret; 5596 return ret;
5597 } 5597 }
5598 EXPORT_SYMBOL(cond_resched_lock); 5598 EXPORT_SYMBOL(cond_resched_lock);
5599 5599
5600 int __sched cond_resched_softirq(void) 5600 int __sched cond_resched_softirq(void)
5601 { 5601 {
5602 BUG_ON(!in_softirq()); 5602 BUG_ON(!in_softirq());
5603 5603
5604 if (need_resched() && system_state == SYSTEM_RUNNING) { 5604 if (need_resched() && system_state == SYSTEM_RUNNING) {
5605 local_bh_enable(); 5605 local_bh_enable();
5606 __cond_resched(); 5606 __cond_resched();
5607 local_bh_disable(); 5607 local_bh_disable();
5608 return 1; 5608 return 1;
5609 } 5609 }
5610 return 0; 5610 return 0;
5611 } 5611 }
5612 EXPORT_SYMBOL(cond_resched_softirq); 5612 EXPORT_SYMBOL(cond_resched_softirq);
5613 5613
5614 /** 5614 /**
5615 * yield - yield the current processor to other threads. 5615 * yield - yield the current processor to other threads.
5616 * 5616 *
5617 * This is a shortcut for kernel-space yielding - it marks the 5617 * This is a shortcut for kernel-space yielding - it marks the
5618 * thread runnable and calls sys_sched_yield(). 5618 * thread runnable and calls sys_sched_yield().
5619 */ 5619 */
5620 void __sched yield(void) 5620 void __sched yield(void)
5621 { 5621 {
5622 set_current_state(TASK_RUNNING); 5622 set_current_state(TASK_RUNNING);
5623 sys_sched_yield(); 5623 sys_sched_yield();
5624 } 5624 }
5625 EXPORT_SYMBOL(yield); 5625 EXPORT_SYMBOL(yield);
5626 5626
5627 /* 5627 /*
5628 * This task is about to go to sleep on IO. Increment rq->nr_iowait so 5628 * This task is about to go to sleep on IO. Increment rq->nr_iowait so
5629 * that process accounting knows that this is a task in IO wait state. 5629 * that process accounting knows that this is a task in IO wait state.
5630 * 5630 *
5631 * But don't do that if it is a deliberate, throttling IO wait (this task 5631 * But don't do that if it is a deliberate, throttling IO wait (this task
5632 * has set its backing_dev_info: the queue against which it should throttle) 5632 * has set its backing_dev_info: the queue against which it should throttle)
5633 */ 5633 */
5634 void __sched io_schedule(void) 5634 void __sched io_schedule(void)
5635 { 5635 {
5636 struct rq *rq = &__raw_get_cpu_var(runqueues); 5636 struct rq *rq = &__raw_get_cpu_var(runqueues);
5637 5637
5638 delayacct_blkio_start(); 5638 delayacct_blkio_start();
5639 atomic_inc(&rq->nr_iowait); 5639 atomic_inc(&rq->nr_iowait);
5640 schedule(); 5640 schedule();
5641 atomic_dec(&rq->nr_iowait); 5641 atomic_dec(&rq->nr_iowait);
5642 delayacct_blkio_end(); 5642 delayacct_blkio_end();
5643 } 5643 }
5644 EXPORT_SYMBOL(io_schedule); 5644 EXPORT_SYMBOL(io_schedule);
5645 5645
5646 long __sched io_schedule_timeout(long timeout) 5646 long __sched io_schedule_timeout(long timeout)
5647 { 5647 {
5648 struct rq *rq = &__raw_get_cpu_var(runqueues); 5648 struct rq *rq = &__raw_get_cpu_var(runqueues);
5649 long ret; 5649 long ret;
5650 5650
5651 delayacct_blkio_start(); 5651 delayacct_blkio_start();
5652 atomic_inc(&rq->nr_iowait); 5652 atomic_inc(&rq->nr_iowait);
5653 ret = schedule_timeout(timeout); 5653 ret = schedule_timeout(timeout);
5654 atomic_dec(&rq->nr_iowait); 5654 atomic_dec(&rq->nr_iowait);
5655 delayacct_blkio_end(); 5655 delayacct_blkio_end();
5656 return ret; 5656 return ret;
5657 } 5657 }
5658 5658
5659 /** 5659 /**
5660 * sys_sched_get_priority_max - return maximum RT priority. 5660 * sys_sched_get_priority_max - return maximum RT priority.
5661 * @policy: scheduling class. 5661 * @policy: scheduling class.
5662 * 5662 *
5663 * this syscall returns the maximum rt_priority that can be used 5663 * this syscall returns the maximum rt_priority that can be used
5664 * by a given scheduling class. 5664 * by a given scheduling class.
5665 */ 5665 */
5666 asmlinkage long sys_sched_get_priority_max(int policy) 5666 asmlinkage long sys_sched_get_priority_max(int policy)
5667 { 5667 {
5668 int ret = -EINVAL; 5668 int ret = -EINVAL;
5669 5669
5670 switch (policy) { 5670 switch (policy) {
5671 case SCHED_FIFO: 5671 case SCHED_FIFO:
5672 case SCHED_RR: 5672 case SCHED_RR:
5673 ret = MAX_USER_RT_PRIO-1; 5673 ret = MAX_USER_RT_PRIO-1;
5674 break; 5674 break;
5675 case SCHED_NORMAL: 5675 case SCHED_NORMAL:
5676 case SCHED_BATCH: 5676 case SCHED_BATCH:
5677 case SCHED_IDLE: 5677 case SCHED_IDLE:
5678 ret = 0; 5678 ret = 0;
5679 break; 5679 break;
5680 } 5680 }
5681 return ret; 5681 return ret;
5682 } 5682 }
5683 5683
5684 /** 5684 /**
5685 * sys_sched_get_priority_min - return minimum RT priority. 5685 * sys_sched_get_priority_min - return minimum RT priority.
5686 * @policy: scheduling class. 5686 * @policy: scheduling class.
5687 * 5687 *
5688 * this syscall returns the minimum rt_priority that can be used 5688 * this syscall returns the minimum rt_priority that can be used
5689 * by a given scheduling class. 5689 * by a given scheduling class.
5690 */ 5690 */
5691 asmlinkage long sys_sched_get_priority_min(int policy) 5691 asmlinkage long sys_sched_get_priority_min(int policy)
5692 { 5692 {
5693 int ret = -EINVAL; 5693 int ret = -EINVAL;
5694 5694
5695 switch (policy) { 5695 switch (policy) {
5696 case SCHED_FIFO: 5696 case SCHED_FIFO:
5697 case SCHED_RR: 5697 case SCHED_RR:
5698 ret = 1; 5698 ret = 1;
5699 break; 5699 break;
5700 case SCHED_NORMAL: 5700 case SCHED_NORMAL:
5701 case SCHED_BATCH: 5701 case SCHED_BATCH:
5702 case SCHED_IDLE: 5702 case SCHED_IDLE:
5703 ret = 0; 5703 ret = 0;
5704 } 5704 }
5705 return ret; 5705 return ret;
5706 } 5706 }
5707 5707
5708 /** 5708 /**
5709 * sys_sched_rr_get_interval - return the default timeslice of a process. 5709 * sys_sched_rr_get_interval - return the default timeslice of a process.
5710 * @pid: pid of the process. 5710 * @pid: pid of the process.
5711 * @interval: userspace pointer to the timeslice value. 5711 * @interval: userspace pointer to the timeslice value.
5712 * 5712 *
5713 * this syscall writes the default timeslice value of a given process 5713 * this syscall writes the default timeslice value of a given process
5714 * into the user-space timespec buffer. A value of '0' means infinity. 5714 * into the user-space timespec buffer. A value of '0' means infinity.
5715 */ 5715 */
5716 asmlinkage 5716 asmlinkage
5717 long sys_sched_rr_get_interval(pid_t pid, struct timespec __user *interval) 5717 long sys_sched_rr_get_interval(pid_t pid, struct timespec __user *interval)
5718 { 5718 {
5719 struct task_struct *p; 5719 struct task_struct *p;
5720 unsigned int time_slice; 5720 unsigned int time_slice;
5721 int retval; 5721 int retval;
5722 struct timespec t; 5722 struct timespec t;
5723 5723
5724 if (pid < 0) 5724 if (pid < 0)
5725 return -EINVAL; 5725 return -EINVAL;
5726 5726
5727 retval = -ESRCH; 5727 retval = -ESRCH;
5728 read_lock(&tasklist_lock); 5728 read_lock(&tasklist_lock);
5729 p = find_process_by_pid(pid); 5729 p = find_process_by_pid(pid);
5730 if (!p) 5730 if (!p)
5731 goto out_unlock; 5731 goto out_unlock;
5732 5732
5733 retval = security_task_getscheduler(p); 5733 retval = security_task_getscheduler(p);
5734 if (retval) 5734 if (retval)
5735 goto out_unlock; 5735 goto out_unlock;
5736 5736
5737 /* 5737 /*
5738 * Time slice is 0 for SCHED_FIFO tasks and for SCHED_OTHER 5738 * Time slice is 0 for SCHED_FIFO tasks and for SCHED_OTHER
5739 * tasks that are on an otherwise idle runqueue: 5739 * tasks that are on an otherwise idle runqueue:
5740 */ 5740 */
5741 time_slice = 0; 5741 time_slice = 0;
5742 if (p->policy == SCHED_RR) { 5742 if (p->policy == SCHED_RR) {
5743 time_slice = DEF_TIMESLICE; 5743 time_slice = DEF_TIMESLICE;
5744 } else if (p->policy != SCHED_FIFO) { 5744 } else if (p->policy != SCHED_FIFO) {
5745 struct sched_entity *se = &p->se; 5745 struct sched_entity *se = &p->se;
5746 unsigned long flags; 5746 unsigned long flags;
5747 struct rq *rq; 5747 struct rq *rq;
5748 5748
5749 rq = task_rq_lock(p, &flags); 5749 rq = task_rq_lock(p, &flags);
5750 if (rq->cfs.load.weight) 5750 if (rq->cfs.load.weight)
5751 time_slice = NS_TO_JIFFIES(sched_slice(&rq->cfs, se)); 5751 time_slice = NS_TO_JIFFIES(sched_slice(&rq->cfs, se));
5752 task_rq_unlock(rq, &flags); 5752 task_rq_unlock(rq, &flags);
5753 } 5753 }
5754 read_unlock(&tasklist_lock); 5754 read_unlock(&tasklist_lock);
5755 jiffies_to_timespec(time_slice, &t); 5755 jiffies_to_timespec(time_slice, &t);
5756 retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; 5756 retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0;
5757 return retval; 5757 return retval;
5758 5758
5759 out_unlock: 5759 out_unlock:
5760 read_unlock(&tasklist_lock); 5760 read_unlock(&tasklist_lock);
5761 return retval; 5761 return retval;
5762 } 5762 }
5763 5763
5764 static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; 5764 static const char stat_nam[] = TASK_STATE_TO_CHAR_STR;
5765 5765
5766 void sched_show_task(struct task_struct *p) 5766 void sched_show_task(struct task_struct *p)
5767 { 5767 {
5768 unsigned long free = 0; 5768 unsigned long free = 0;
5769 unsigned state; 5769 unsigned state;
5770 5770
5771 state = p->state ? __ffs(p->state) + 1 : 0; 5771 state = p->state ? __ffs(p->state) + 1 : 0;
5772 printk(KERN_INFO "%-13.13s %c", p->comm, 5772 printk(KERN_INFO "%-13.13s %c", p->comm,
5773 state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); 5773 state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?');
5774 #if BITS_PER_LONG == 32 5774 #if BITS_PER_LONG == 32
5775 if (state == TASK_RUNNING) 5775 if (state == TASK_RUNNING)
5776 printk(KERN_CONT " running "); 5776 printk(KERN_CONT " running ");
5777 else 5777 else
5778 printk(KERN_CONT " %08lx ", thread_saved_pc(p)); 5778 printk(KERN_CONT " %08lx ", thread_saved_pc(p));
5779 #else 5779 #else
5780 if (state == TASK_RUNNING) 5780 if (state == TASK_RUNNING)
5781 printk(KERN_CONT " running task "); 5781 printk(KERN_CONT " running task ");
5782 else 5782 else
5783 printk(KERN_CONT " %016lx ", thread_saved_pc(p)); 5783 printk(KERN_CONT " %016lx ", thread_saved_pc(p));
5784 #endif 5784 #endif
5785 #ifdef CONFIG_DEBUG_STACK_USAGE 5785 #ifdef CONFIG_DEBUG_STACK_USAGE
5786 { 5786 {
5787 unsigned long *n = end_of_stack(p); 5787 unsigned long *n = end_of_stack(p);
5788 while (!*n) 5788 while (!*n)
5789 n++; 5789 n++;
5790 free = (unsigned long)n - (unsigned long)end_of_stack(p); 5790 free = (unsigned long)n - (unsigned long)end_of_stack(p);
5791 } 5791 }
5792 #endif 5792 #endif
5793 printk(KERN_CONT "%5lu %5d %6d\n", free, 5793 printk(KERN_CONT "%5lu %5d %6d\n", free,
5794 task_pid_nr(p), task_pid_nr(p->real_parent)); 5794 task_pid_nr(p), task_pid_nr(p->real_parent));
5795 5795
5796 show_stack(p, NULL); 5796 show_stack(p, NULL);
5797 } 5797 }
5798 5798
5799 void show_state_filter(unsigned long state_filter) 5799 void show_state_filter(unsigned long state_filter)
5800 { 5800 {
5801 struct task_struct *g, *p; 5801 struct task_struct *g, *p;
5802 5802
5803 #if BITS_PER_LONG == 32 5803 #if BITS_PER_LONG == 32
5804 printk(KERN_INFO 5804 printk(KERN_INFO
5805 " task PC stack pid father\n"); 5805 " task PC stack pid father\n");
5806 #else 5806 #else
5807 printk(KERN_INFO 5807 printk(KERN_INFO
5808 " task PC stack pid father\n"); 5808 " task PC stack pid father\n");
5809 #endif 5809 #endif
5810 read_lock(&tasklist_lock); 5810 read_lock(&tasklist_lock);
5811 do_each_thread(g, p) { 5811 do_each_thread(g, p) {
5812 /* 5812 /*
5813 * reset the NMI-timeout, listing all files on a slow 5813 * reset the NMI-timeout, listing all files on a slow
5814 * console might take alot of time: 5814 * console might take alot of time:
5815 */ 5815 */
5816 touch_nmi_watchdog(); 5816 touch_nmi_watchdog();
5817 if (!state_filter || (p->state & state_filter)) 5817 if (!state_filter || (p->state & state_filter))
5818 sched_show_task(p); 5818 sched_show_task(p);
5819 } while_each_thread(g, p); 5819 } while_each_thread(g, p);
5820 5820
5821 touch_all_softlockup_watchdogs(); 5821 touch_all_softlockup_watchdogs();
5822 5822
5823 #ifdef CONFIG_SCHED_DEBUG 5823 #ifdef CONFIG_SCHED_DEBUG
5824 sysrq_sched_debug_show(); 5824 sysrq_sched_debug_show();
5825 #endif 5825 #endif
5826 read_unlock(&tasklist_lock); 5826 read_unlock(&tasklist_lock);
5827 /* 5827 /*
5828 * Only show locks if all tasks are dumped: 5828 * Only show locks if all tasks are dumped:
5829 */ 5829 */
5830 if (state_filter == -1) 5830 if (state_filter == -1)
5831 debug_show_all_locks(); 5831 debug_show_all_locks();
5832 } 5832 }
5833 5833
5834 void __cpuinit init_idle_bootup_task(struct task_struct *idle) 5834 void __cpuinit init_idle_bootup_task(struct task_struct *idle)
5835 { 5835 {
5836 idle->sched_class = &idle_sched_class; 5836 idle->sched_class = &idle_sched_class;
5837 } 5837 }
5838 5838
5839 /** 5839 /**
5840 * init_idle - set up an idle thread for a given CPU 5840 * init_idle - set up an idle thread for a given CPU
5841 * @idle: task in question 5841 * @idle: task in question
5842 * @cpu: cpu the idle task belongs to 5842 * @cpu: cpu the idle task belongs to
5843 * 5843 *
5844 * NOTE: this function does not set the idle thread's NEED_RESCHED 5844 * NOTE: this function does not set the idle thread's NEED_RESCHED
5845 * flag, to make booting more robust. 5845 * flag, to make booting more robust.
5846 */ 5846 */
5847 void __cpuinit init_idle(struct task_struct *idle, int cpu) 5847 void __cpuinit init_idle(struct task_struct *idle, int cpu)
5848 { 5848 {
5849 struct rq *rq = cpu_rq(cpu); 5849 struct rq *rq = cpu_rq(cpu);
5850 unsigned long flags; 5850 unsigned long flags;
5851 5851
5852 spin_lock_irqsave(&rq->lock, flags); 5852 spin_lock_irqsave(&rq->lock, flags);
5853 5853
5854 __sched_fork(idle); 5854 __sched_fork(idle);
5855 idle->se.exec_start = sched_clock(); 5855 idle->se.exec_start = sched_clock();
5856 5856
5857 idle->prio = idle->normal_prio = MAX_PRIO; 5857 idle->prio = idle->normal_prio = MAX_PRIO;
5858 idle->cpus_allowed = cpumask_of_cpu(cpu); 5858 idle->cpus_allowed = cpumask_of_cpu(cpu);
5859 __set_task_cpu(idle, cpu); 5859 __set_task_cpu(idle, cpu);
5860 5860
5861 rq->curr = rq->idle = idle; 5861 rq->curr = rq->idle = idle;
5862 #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) 5862 #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW)
5863 idle->oncpu = 1; 5863 idle->oncpu = 1;
5864 #endif 5864 #endif
5865 spin_unlock_irqrestore(&rq->lock, flags); 5865 spin_unlock_irqrestore(&rq->lock, flags);
5866 5866
5867 /* Set the preempt count _outside_ the spinlocks! */ 5867 /* Set the preempt count _outside_ the spinlocks! */
5868 #if defined(CONFIG_PREEMPT) 5868 #if defined(CONFIG_PREEMPT)
5869 task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); 5869 task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0);
5870 #else 5870 #else
5871 task_thread_info(idle)->preempt_count = 0; 5871 task_thread_info(idle)->preempt_count = 0;
5872 #endif 5872 #endif
5873 /* 5873 /*
5874 * The idle tasks have their own, simple scheduling class: 5874 * The idle tasks have their own, simple scheduling class:
5875 */ 5875 */
5876 idle->sched_class = &idle_sched_class; 5876 idle->sched_class = &idle_sched_class;
5877 } 5877 }
5878 5878
5879 /* 5879 /*
5880 * In a system that switches off the HZ timer nohz_cpu_mask 5880 * In a system that switches off the HZ timer nohz_cpu_mask
5881 * indicates which cpus entered this state. This is used 5881 * indicates which cpus entered this state. This is used
5882 * in the rcu update to wait only for active cpus. For system 5882 * in the rcu update to wait only for active cpus. For system
5883 * which do not switch off the HZ timer nohz_cpu_mask should 5883 * which do not switch off the HZ timer nohz_cpu_mask should
5884 * always be CPU_MASK_NONE. 5884 * always be CPU_MASK_NONE.
5885 */ 5885 */
5886 cpumask_t nohz_cpu_mask = CPU_MASK_NONE; 5886 cpumask_t nohz_cpu_mask = CPU_MASK_NONE;
5887 5887
5888 /* 5888 /*
5889 * Increase the granularity value when there are more CPUs, 5889 * Increase the granularity value when there are more CPUs,
5890 * because with more CPUs the 'effective latency' as visible 5890 * because with more CPUs the 'effective latency' as visible
5891 * to users decreases. But the relationship is not linear, 5891 * to users decreases. But the relationship is not linear,
5892 * so pick a second-best guess by going with the log2 of the 5892 * so pick a second-best guess by going with the log2 of the
5893 * number of CPUs. 5893 * number of CPUs.
5894 * 5894 *
5895 * This idea comes from the SD scheduler of Con Kolivas: 5895 * This idea comes from the SD scheduler of Con Kolivas:
5896 */ 5896 */
5897 static inline void sched_init_granularity(void) 5897 static inline void sched_init_granularity(void)
5898 { 5898 {
5899 unsigned int factor = 1 + ilog2(num_online_cpus()); 5899 unsigned int factor = 1 + ilog2(num_online_cpus());
5900 const unsigned long limit = 200000000; 5900 const unsigned long limit = 200000000;
5901 5901
5902 sysctl_sched_min_granularity *= factor; 5902 sysctl_sched_min_granularity *= factor;
5903 if (sysctl_sched_min_granularity > limit) 5903 if (sysctl_sched_min_granularity > limit)
5904 sysctl_sched_min_granularity = limit; 5904 sysctl_sched_min_granularity = limit;
5905 5905
5906 sysctl_sched_latency *= factor; 5906 sysctl_sched_latency *= factor;
5907 if (sysctl_sched_latency > limit) 5907 if (sysctl_sched_latency > limit)
5908 sysctl_sched_latency = limit; 5908 sysctl_sched_latency = limit;
5909 5909
5910 sysctl_sched_wakeup_granularity *= factor; 5910 sysctl_sched_wakeup_granularity *= factor;
5911 5911
5912 sysctl_sched_shares_ratelimit *= factor; 5912 sysctl_sched_shares_ratelimit *= factor;
5913 } 5913 }
5914 5914
5915 #ifdef CONFIG_SMP 5915 #ifdef CONFIG_SMP
5916 /* 5916 /*
5917 * This is how migration works: 5917 * This is how migration works:
5918 * 5918 *
5919 * 1) we queue a struct migration_req structure in the source CPU's 5919 * 1) we queue a struct migration_req structure in the source CPU's
5920 * runqueue and wake up that CPU's migration thread. 5920 * runqueue and wake up that CPU's migration thread.
5921 * 2) we down() the locked semaphore => thread blocks. 5921 * 2) we down() the locked semaphore => thread blocks.
5922 * 3) migration thread wakes up (implicitly it forces the migrated 5922 * 3) migration thread wakes up (implicitly it forces the migrated
5923 * thread off the CPU) 5923 * thread off the CPU)
5924 * 4) it gets the migration request and checks whether the migrated 5924 * 4) it gets the migration request and checks whether the migrated
5925 * task is still in the wrong runqueue. 5925 * task is still in the wrong runqueue.
5926 * 5) if it's in the wrong runqueue then the migration thread removes 5926 * 5) if it's in the wrong runqueue then the migration thread removes
5927 * it and puts it into the right queue. 5927 * it and puts it into the right queue.
5928 * 6) migration thread up()s the semaphore. 5928 * 6) migration thread up()s the semaphore.
5929 * 7) we wake up and the migration is done. 5929 * 7) we wake up and the migration is done.
5930 */ 5930 */
5931 5931
5932 /* 5932 /*
5933 * Change a given task's CPU affinity. Migrate the thread to a 5933 * Change a given task's CPU affinity. Migrate the thread to a
5934 * proper CPU and schedule it away if the CPU it's executing on 5934 * proper CPU and schedule it away if the CPU it's executing on
5935 * is removed from the allowed bitmask. 5935 * is removed from the allowed bitmask.
5936 * 5936 *
5937 * NOTE: the caller must have a valid reference to the task, the 5937 * NOTE: the caller must have a valid reference to the task, the
5938 * task must not exit() & deallocate itself prematurely. The 5938 * task must not exit() & deallocate itself prematurely. The
5939 * call is not atomic; no spinlocks may be held. 5939 * call is not atomic; no spinlocks may be held.
5940 */ 5940 */
5941 int set_cpus_allowed_ptr(struct task_struct *p, const cpumask_t *new_mask) 5941 int set_cpus_allowed_ptr(struct task_struct *p, const cpumask_t *new_mask)
5942 { 5942 {
5943 struct migration_req req; 5943 struct migration_req req;
5944 unsigned long flags; 5944 unsigned long flags;
5945 struct rq *rq; 5945 struct rq *rq;
5946 int ret = 0; 5946 int ret = 0;
5947 5947
5948 rq = task_rq_lock(p, &flags); 5948 rq = task_rq_lock(p, &flags);
5949 if (!cpus_intersects(*new_mask, cpu_online_map)) { 5949 if (!cpus_intersects(*new_mask, cpu_online_map)) {
5950 ret = -EINVAL; 5950 ret = -EINVAL;
5951 goto out; 5951 goto out;
5952 } 5952 }
5953 5953
5954 if (unlikely((p->flags & PF_THREAD_BOUND) && p != current && 5954 if (unlikely((p->flags & PF_THREAD_BOUND) && p != current &&
5955 !cpus_equal(p->cpus_allowed, *new_mask))) { 5955 !cpus_equal(p->cpus_allowed, *new_mask))) {
5956 ret = -EINVAL; 5956 ret = -EINVAL;
5957 goto out; 5957 goto out;
5958 } 5958 }
5959 5959
5960 if (p->sched_class->set_cpus_allowed) 5960 if (p->sched_class->set_cpus_allowed)
5961 p->sched_class->set_cpus_allowed(p, new_mask); 5961 p->sched_class->set_cpus_allowed(p, new_mask);
5962 else { 5962 else {
5963 p->cpus_allowed = *new_mask; 5963 p->cpus_allowed = *new_mask;
5964 p->rt.nr_cpus_allowed = cpus_weight(*new_mask); 5964 p->rt.nr_cpus_allowed = cpus_weight(*new_mask);
5965 } 5965 }
5966 5966
5967 /* Can the task run on the task's current CPU? If so, we're done */ 5967 /* Can the task run on the task's current CPU? If so, we're done */
5968 if (cpu_isset(task_cpu(p), *new_mask)) 5968 if (cpu_isset(task_cpu(p), *new_mask))
5969 goto out; 5969 goto out;
5970 5970
5971 if (migrate_task(p, any_online_cpu(*new_mask), &req)) { 5971 if (migrate_task(p, any_online_cpu(*new_mask), &req)) {
5972 /* Need help from migration thread: drop lock and wait. */ 5972 /* Need help from migration thread: drop lock and wait. */
5973 task_rq_unlock(rq, &flags); 5973 task_rq_unlock(rq, &flags);
5974 wake_up_process(rq->migration_thread); 5974 wake_up_process(rq->migration_thread);
5975 wait_for_completion(&req.done); 5975 wait_for_completion(&req.done);
5976 tlb_migrate_finish(p->mm); 5976 tlb_migrate_finish(p->mm);
5977 return 0; 5977 return 0;
5978 } 5978 }
5979 out: 5979 out:
5980 task_rq_unlock(rq, &flags); 5980 task_rq_unlock(rq, &flags);
5981 5981
5982 return ret; 5982 return ret;
5983 } 5983 }
5984 EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); 5984 EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr);
5985 5985
5986 /* 5986 /*
5987 * Move (not current) task off this cpu, onto dest cpu. We're doing 5987 * Move (not current) task off this cpu, onto dest cpu. We're doing
5988 * this because either it can't run here any more (set_cpus_allowed() 5988 * this because either it can't run here any more (set_cpus_allowed()
5989 * away from this CPU, or CPU going down), or because we're 5989 * away from this CPU, or CPU going down), or because we're
5990 * attempting to rebalance this task on exec (sched_exec). 5990 * attempting to rebalance this task on exec (sched_exec).
5991 * 5991 *
5992 * So we race with normal scheduler movements, but that's OK, as long 5992 * So we race with normal scheduler movements, but that's OK, as long
5993 * as the task is no longer on this CPU. 5993 * as the task is no longer on this CPU.
5994 * 5994 *
5995 * Returns non-zero if task was successfully migrated. 5995 * Returns non-zero if task was successfully migrated.
5996 */ 5996 */
5997 static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) 5997 static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu)
5998 { 5998 {
5999 struct rq *rq_dest, *rq_src; 5999 struct rq *rq_dest, *rq_src;
6000 int ret = 0, on_rq; 6000 int ret = 0, on_rq;
6001 6001
6002 if (unlikely(!cpu_active(dest_cpu))) 6002 if (unlikely(!cpu_active(dest_cpu)))
6003 return ret; 6003 return ret;
6004 6004
6005 rq_src = cpu_rq(src_cpu); 6005 rq_src = cpu_rq(src_cpu);
6006 rq_dest = cpu_rq(dest_cpu); 6006 rq_dest = cpu_rq(dest_cpu);
6007 6007
6008 double_rq_lock(rq_src, rq_dest); 6008 double_rq_lock(rq_src, rq_dest);
6009 /* Already moved. */ 6009 /* Already moved. */
6010 if (task_cpu(p) != src_cpu) 6010 if (task_cpu(p) != src_cpu)
6011 goto done; 6011 goto done;
6012 /* Affinity changed (again). */ 6012 /* Affinity changed (again). */
6013 if (!cpu_isset(dest_cpu, p->cpus_allowed)) 6013 if (!cpu_isset(dest_cpu, p->cpus_allowed))
6014 goto fail; 6014 goto fail;
6015 6015
6016 on_rq = p->se.on_rq; 6016 on_rq = p->se.on_rq;
6017 if (on_rq) 6017 if (on_rq)
6018 deactivate_task(rq_src, p, 0); 6018 deactivate_task(rq_src, p, 0);
6019 6019
6020 set_task_cpu(p, dest_cpu); 6020 set_task_cpu(p, dest_cpu);
6021 if (on_rq) { 6021 if (on_rq) {
6022 activate_task(rq_dest, p, 0); 6022 activate_task(rq_dest, p, 0);
6023 check_preempt_curr(rq_dest, p, 0); 6023 check_preempt_curr(rq_dest, p, 0);
6024 } 6024 }
6025 done: 6025 done:
6026 ret = 1; 6026 ret = 1;
6027 fail: 6027 fail:
6028 double_rq_unlock(rq_src, rq_dest); 6028 double_rq_unlock(rq_src, rq_dest);
6029 return ret; 6029 return ret;
6030 } 6030 }
6031 6031
6032 /* 6032 /*
6033 * migration_thread - this is a highprio system thread that performs 6033 * migration_thread - this is a highprio system thread that performs
6034 * thread migration by bumping thread off CPU then 'pushing' onto 6034 * thread migration by bumping thread off CPU then 'pushing' onto
6035 * another runqueue. 6035 * another runqueue.
6036 */ 6036 */
6037 static int migration_thread(void *data) 6037 static int migration_thread(void *data)
6038 { 6038 {
6039 int cpu = (long)data; 6039 int cpu = (long)data;
6040 struct rq *rq; 6040 struct rq *rq;
6041 6041
6042 rq = cpu_rq(cpu); 6042 rq = cpu_rq(cpu);
6043 BUG_ON(rq->migration_thread != current); 6043 BUG_ON(rq->migration_thread != current);
6044 6044
6045 set_current_state(TASK_INTERRUPTIBLE); 6045 set_current_state(TASK_INTERRUPTIBLE);
6046 while (!kthread_should_stop()) { 6046 while (!kthread_should_stop()) {
6047 struct migration_req *req; 6047 struct migration_req *req;
6048 struct list_head *head; 6048 struct list_head *head;
6049 6049
6050 spin_lock_irq(&rq->lock); 6050 spin_lock_irq(&rq->lock);
6051 6051
6052 if (cpu_is_offline(cpu)) { 6052 if (cpu_is_offline(cpu)) {
6053 spin_unlock_irq(&rq->lock); 6053 spin_unlock_irq(&rq->lock);
6054 goto wait_to_die; 6054 goto wait_to_die;
6055 } 6055 }
6056 6056
6057 if (rq->active_balance) { 6057 if (rq->active_balance) {
6058 active_load_balance(rq, cpu); 6058 active_load_balance(rq, cpu);
6059 rq->active_balance = 0; 6059 rq->active_balance = 0;
6060 } 6060 }
6061 6061
6062 head = &rq->migration_queue; 6062 head = &rq->migration_queue;
6063 6063
6064 if (list_empty(head)) { 6064 if (list_empty(head)) {
6065 spin_unlock_irq(&rq->lock); 6065 spin_unlock_irq(&rq->lock);
6066 schedule(); 6066 schedule();
6067 set_current_state(TASK_INTERRUPTIBLE); 6067 set_current_state(TASK_INTERRUPTIBLE);
6068 continue; 6068 continue;
6069 } 6069 }
6070 req = list_entry(head->next, struct migration_req, list); 6070 req = list_entry(head->next, struct migration_req, list);
6071 list_del_init(head->next); 6071 list_del_init(head->next);
6072 6072
6073 spin_unlock(&rq->lock); 6073 spin_unlock(&rq->lock);
6074 __migrate_task(req->task, cpu, req->dest_cpu); 6074 __migrate_task(req->task, cpu, req->dest_cpu);
6075 local_irq_enable(); 6075 local_irq_enable();
6076 6076
6077 complete(&req->done); 6077 complete(&req->done);
6078 } 6078 }
6079 __set_current_state(TASK_RUNNING); 6079 __set_current_state(TASK_RUNNING);
6080 return 0; 6080 return 0;
6081 6081
6082 wait_to_die: 6082 wait_to_die:
6083 /* Wait for kthread_stop */ 6083 /* Wait for kthread_stop */
6084 set_current_state(TASK_INTERRUPTIBLE); 6084 set_current_state(TASK_INTERRUPTIBLE);
6085 while (!kthread_should_stop()) { 6085 while (!kthread_should_stop()) {
6086 schedule(); 6086 schedule();
6087 set_current_state(TASK_INTERRUPTIBLE); 6087 set_current_state(TASK_INTERRUPTIBLE);
6088 } 6088 }
6089 __set_current_state(TASK_RUNNING); 6089 __set_current_state(TASK_RUNNING);
6090 return 0; 6090 return 0;
6091 } 6091 }
6092 6092
6093 #ifdef CONFIG_HOTPLUG_CPU 6093 #ifdef CONFIG_HOTPLUG_CPU
6094 6094
6095 static int __migrate_task_irq(struct task_struct *p, int src_cpu, int dest_cpu) 6095 static int __migrate_task_irq(struct task_struct *p, int src_cpu, int dest_cpu)
6096 { 6096 {
6097 int ret; 6097 int ret;
6098 6098
6099 local_irq_disable(); 6099 local_irq_disable();
6100 ret = __migrate_task(p, src_cpu, dest_cpu); 6100 ret = __migrate_task(p, src_cpu, dest_cpu);
6101 local_irq_enable(); 6101 local_irq_enable();
6102 return ret; 6102 return ret;
6103 } 6103 }
6104 6104
6105 /* 6105 /*
6106 * Figure out where task on dead CPU should go, use force if necessary. 6106 * Figure out where task on dead CPU should go, use force if necessary.
6107 */ 6107 */
6108 static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) 6108 static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p)
6109 { 6109 {
6110 unsigned long flags; 6110 unsigned long flags;
6111 cpumask_t mask; 6111 cpumask_t mask;
6112 struct rq *rq; 6112 struct rq *rq;
6113 int dest_cpu; 6113 int dest_cpu;
6114 6114
6115 do { 6115 do {
6116 /* On same node? */ 6116 /* On same node? */
6117 mask = node_to_cpumask(cpu_to_node(dead_cpu)); 6117 mask = node_to_cpumask(cpu_to_node(dead_cpu));
6118 cpus_and(mask, mask, p->cpus_allowed); 6118 cpus_and(mask, mask, p->cpus_allowed);
6119 dest_cpu = any_online_cpu(mask); 6119 dest_cpu = any_online_cpu(mask);
6120 6120
6121 /* On any allowed CPU? */ 6121 /* On any allowed CPU? */
6122 if (dest_cpu >= nr_cpu_ids) 6122 if (dest_cpu >= nr_cpu_ids)
6123 dest_cpu = any_online_cpu(p->cpus_allowed); 6123 dest_cpu = any_online_cpu(p->cpus_allowed);
6124 6124
6125 /* No more Mr. Nice Guy. */ 6125 /* No more Mr. Nice Guy. */
6126 if (dest_cpu >= nr_cpu_ids) { 6126 if (dest_cpu >= nr_cpu_ids) {
6127 cpumask_t cpus_allowed; 6127 cpumask_t cpus_allowed;
6128 6128
6129 cpuset_cpus_allowed_locked(p, &cpus_allowed); 6129 cpuset_cpus_allowed_locked(p, &cpus_allowed);
6130 /* 6130 /*
6131 * Try to stay on the same cpuset, where the 6131 * Try to stay on the same cpuset, where the
6132 * current cpuset may be a subset of all cpus. 6132 * current cpuset may be a subset of all cpus.
6133 * The cpuset_cpus_allowed_locked() variant of 6133 * The cpuset_cpus_allowed_locked() variant of
6134 * cpuset_cpus_allowed() will not block. It must be 6134 * cpuset_cpus_allowed() will not block. It must be
6135 * called within calls to cpuset_lock/cpuset_unlock. 6135 * called within calls to cpuset_lock/cpuset_unlock.
6136 */ 6136 */
6137 rq = task_rq_lock(p, &flags); 6137 rq = task_rq_lock(p, &flags);
6138 p->cpus_allowed = cpus_allowed; 6138 p->cpus_allowed = cpus_allowed;
6139 dest_cpu = any_online_cpu(p->cpus_allowed); 6139 dest_cpu = any_online_cpu(p->cpus_allowed);
6140 task_rq_unlock(rq, &flags); 6140 task_rq_unlock(rq, &flags);
6141 6141
6142 /* 6142 /*
6143 * Don't tell them about moving exiting tasks or 6143 * Don't tell them about moving exiting tasks or
6144 * kernel threads (both mm NULL), since they never 6144 * kernel threads (both mm NULL), since they never
6145 * leave kernel. 6145 * leave kernel.
6146 */ 6146 */
6147 if (p->mm && printk_ratelimit()) { 6147 if (p->mm && printk_ratelimit()) {
6148 printk(KERN_INFO "process %d (%s) no " 6148 printk(KERN_INFO "process %d (%s) no "
6149 "longer affine to cpu%d\n", 6149 "longer affine to cpu%d\n",
6150 task_pid_nr(p), p->comm, dead_cpu); 6150 task_pid_nr(p), p->comm, dead_cpu);
6151 } 6151 }
6152 } 6152 }
6153 } while (!__migrate_task_irq(p, dead_cpu, dest_cpu)); 6153 } while (!__migrate_task_irq(p, dead_cpu, dest_cpu));
6154 } 6154 }
6155 6155
6156 /* 6156 /*
6157 * While a dead CPU has no uninterruptible tasks queued at this point, 6157 * While a dead CPU has no uninterruptible tasks queued at this point,
6158 * it might still have a nonzero ->nr_uninterruptible counter, because 6158 * it might still have a nonzero ->nr_uninterruptible counter, because
6159 * for performance reasons the counter is not stricly tracking tasks to 6159 * for performance reasons the counter is not stricly tracking tasks to
6160 * their home CPUs. So we just add the counter to another CPU's counter, 6160 * their home CPUs. So we just add the counter to another CPU's counter,
6161 * to keep the global sum constant after CPU-down: 6161 * to keep the global sum constant after CPU-down:
6162 */ 6162 */
6163 static void migrate_nr_uninterruptible(struct rq *rq_src) 6163 static void migrate_nr_uninterruptible(struct rq *rq_src)
6164 { 6164 {
6165 struct rq *rq_dest = cpu_rq(any_online_cpu(*CPU_MASK_ALL_PTR)); 6165 struct rq *rq_dest = cpu_rq(any_online_cpu(*CPU_MASK_ALL_PTR));
6166 unsigned long flags; 6166 unsigned long flags;
6167 6167
6168 local_irq_save(flags); 6168 local_irq_save(flags);
6169 double_rq_lock(rq_src, rq_dest); 6169 double_rq_lock(rq_src, rq_dest);
6170 rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; 6170 rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible;
6171 rq_src->nr_uninterruptible = 0; 6171 rq_src->nr_uninterruptible = 0;
6172 double_rq_unlock(rq_src, rq_dest); 6172 double_rq_unlock(rq_src, rq_dest);
6173 local_irq_restore(flags); 6173 local_irq_restore(flags);
6174 } 6174 }
6175 6175
6176 /* Run through task list and migrate tasks from the dead cpu. */ 6176 /* Run through task list and migrate tasks from the dead cpu. */
6177 static void migrate_live_tasks(int src_cpu) 6177 static void migrate_live_tasks(int src_cpu)
6178 { 6178 {
6179 struct task_struct *p, *t; 6179 struct task_struct *p, *t;
6180 6180
6181 read_lock(&tasklist_lock); 6181 read_lock(&tasklist_lock);
6182 6182
6183 do_each_thread(t, p) { 6183 do_each_thread(t, p) {
6184 if (p == current) 6184 if (p == current)
6185 continue; 6185 continue;
6186 6186
6187 if (task_cpu(p) == src_cpu) 6187 if (task_cpu(p) == src_cpu)
6188 move_task_off_dead_cpu(src_cpu, p); 6188 move_task_off_dead_cpu(src_cpu, p);
6189 } while_each_thread(t, p); 6189 } while_each_thread(t, p);
6190 6190
6191 read_unlock(&tasklist_lock); 6191 read_unlock(&tasklist_lock);
6192 } 6192 }
6193 6193
6194 /* 6194 /*
6195 * Schedules idle task to be the next runnable task on current CPU. 6195 * Schedules idle task to be the next runnable task on current CPU.
6196 * It does so by boosting its priority to highest possible. 6196 * It does so by boosting its priority to highest possible.
6197 * Used by CPU offline code. 6197 * Used by CPU offline code.
6198 */ 6198 */
6199 void sched_idle_next(void) 6199 void sched_idle_next(void)
6200 { 6200 {
6201 int this_cpu = smp_processor_id(); 6201 int this_cpu = smp_processor_id();
6202 struct rq *rq = cpu_rq(this_cpu); 6202 struct rq *rq = cpu_rq(this_cpu);
6203 struct task_struct *p = rq->idle; 6203 struct task_struct *p = rq->idle;
6204 unsigned long flags; 6204 unsigned long flags;
6205 6205
6206 /* cpu has to be offline */ 6206 /* cpu has to be offline */
6207 BUG_ON(cpu_online(this_cpu)); 6207 BUG_ON(cpu_online(this_cpu));
6208 6208
6209 /* 6209 /*
6210 * Strictly not necessary since rest of the CPUs are stopped by now 6210 * Strictly not necessary since rest of the CPUs are stopped by now
6211 * and interrupts disabled on the current cpu. 6211 * and interrupts disabled on the current cpu.
6212 */ 6212 */
6213 spin_lock_irqsave(&rq->lock, flags); 6213 spin_lock_irqsave(&rq->lock, flags);
6214 6214
6215 __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); 6215 __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1);
6216 6216
6217 update_rq_clock(rq); 6217 update_rq_clock(rq);
6218 activate_task(rq, p, 0); 6218 activate_task(rq, p, 0);
6219 6219
6220 spin_unlock_irqrestore(&rq->lock, flags); 6220 spin_unlock_irqrestore(&rq->lock, flags);
6221 } 6221 }
6222 6222
6223 /* 6223 /*
6224 * Ensures that the idle task is using init_mm right before its cpu goes 6224 * Ensures that the idle task is using init_mm right before its cpu goes
6225 * offline. 6225 * offline.
6226 */ 6226 */
6227 void idle_task_exit(void) 6227 void idle_task_exit(void)
6228 { 6228 {
6229 struct mm_struct *mm = current->active_mm; 6229 struct mm_struct *mm = current->active_mm;
6230 6230
6231 BUG_ON(cpu_online(smp_processor_id())); 6231 BUG_ON(cpu_online(smp_processor_id()));
6232 6232
6233 if (mm != &init_mm) 6233 if (mm != &init_mm)
6234 switch_mm(mm, &init_mm, current); 6234 switch_mm(mm, &init_mm, current);
6235 mmdrop(mm); 6235 mmdrop(mm);
6236 } 6236 }
6237 6237
6238 /* called under rq->lock with disabled interrupts */ 6238 /* called under rq->lock with disabled interrupts */
6239 static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) 6239 static void migrate_dead(unsigned int dead_cpu, struct task_struct *p)
6240 { 6240 {
6241 struct rq *rq = cpu_rq(dead_cpu); 6241 struct rq *rq = cpu_rq(dead_cpu);
6242 6242
6243 /* Must be exiting, otherwise would be on tasklist. */ 6243 /* Must be exiting, otherwise would be on tasklist. */
6244 BUG_ON(!p->exit_state); 6244 BUG_ON(!p->exit_state);
6245 6245
6246 /* Cannot have done final schedule yet: would have vanished. */ 6246 /* Cannot have done final schedule yet: would have vanished. */
6247 BUG_ON(p->state == TASK_DEAD); 6247 BUG_ON(p->state == TASK_DEAD);
6248 6248
6249 get_task_struct(p); 6249 get_task_struct(p);
6250 6250
6251 /* 6251 /*
6252 * Drop lock around migration; if someone else moves it, 6252 * Drop lock around migration; if someone else moves it,
6253 * that's OK. No task can be added to this CPU, so iteration is 6253 * that's OK. No task can be added to this CPU, so iteration is
6254 * fine. 6254 * fine.
6255 */ 6255 */
6256 spin_unlock_irq(&rq->lock); 6256 spin_unlock_irq(&rq->lock);
6257 move_task_off_dead_cpu(dead_cpu, p); 6257 move_task_off_dead_cpu(dead_cpu, p);
6258 spin_lock_irq(&rq->lock); 6258 spin_lock_irq(&rq->lock);
6259 6259
6260 put_task_struct(p); 6260 put_task_struct(p);
6261 } 6261 }
6262 6262
6263 /* release_task() removes task from tasklist, so we won't find dead tasks. */ 6263 /* release_task() removes task from tasklist, so we won't find dead tasks. */
6264 static void migrate_dead_tasks(unsigned int dead_cpu) 6264 static void migrate_dead_tasks(unsigned int dead_cpu)
6265 { 6265 {
6266 struct rq *rq = cpu_rq(dead_cpu); 6266 struct rq *rq = cpu_rq(dead_cpu);
6267 struct task_struct *next; 6267 struct task_struct *next;
6268 6268
6269 for ( ; ; ) { 6269 for ( ; ; ) {
6270 if (!rq->nr_running) 6270 if (!rq->nr_running)
6271 break; 6271 break;
6272 update_rq_clock(rq); 6272 update_rq_clock(rq);
6273 next = pick_next_task(rq, rq->curr); 6273 next = pick_next_task(rq, rq->curr);
6274 if (!next) 6274 if (!next)
6275 break; 6275 break;
6276 next->sched_class->put_prev_task(rq, next); 6276 next->sched_class->put_prev_task(rq, next);
6277 migrate_dead(dead_cpu, next); 6277 migrate_dead(dead_cpu, next);
6278 6278
6279 } 6279 }
6280 } 6280 }
6281 #endif /* CONFIG_HOTPLUG_CPU */ 6281 #endif /* CONFIG_HOTPLUG_CPU */
6282 6282
6283 #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) 6283 #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL)
6284 6284
6285 static struct ctl_table sd_ctl_dir[] = { 6285 static struct ctl_table sd_ctl_dir[] = {
6286 { 6286 {
6287 .procname = "sched_domain", 6287 .procname = "sched_domain",
6288 .mode = 0555, 6288 .mode = 0555,
6289 }, 6289 },
6290 {0, }, 6290 {0, },
6291 }; 6291 };
6292 6292
6293 static struct ctl_table sd_ctl_root[] = { 6293 static struct ctl_table sd_ctl_root[] = {
6294 { 6294 {
6295 .ctl_name = CTL_KERN, 6295 .ctl_name = CTL_KERN,
6296 .procname = "kernel", 6296 .procname = "kernel",
6297 .mode = 0555, 6297 .mode = 0555,
6298 .child = sd_ctl_dir, 6298 .child = sd_ctl_dir,
6299 }, 6299 },
6300 {0, }, 6300 {0, },
6301 }; 6301 };
6302 6302
6303 static struct ctl_table *sd_alloc_ctl_entry(int n) 6303 static struct ctl_table *sd_alloc_ctl_entry(int n)
6304 { 6304 {
6305 struct ctl_table *entry = 6305 struct ctl_table *entry =
6306 kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); 6306 kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL);
6307 6307
6308 return entry; 6308 return entry;
6309 } 6309 }
6310 6310
6311 static void sd_free_ctl_entry(struct ctl_table **tablep) 6311 static void sd_free_ctl_entry(struct ctl_table **tablep)
6312 { 6312 {
6313 struct ctl_table *entry; 6313 struct ctl_table *entry;
6314 6314
6315 /* 6315 /*
6316 * In the intermediate directories, both the child directory and 6316 * In the intermediate directories, both the child directory and
6317 * procname are dynamically allocated and could fail but the mode 6317 * procname are dynamically allocated and could fail but the mode
6318 * will always be set. In the lowest directory the names are 6318 * will always be set. In the lowest directory the names are
6319 * static strings and all have proc handlers. 6319 * static strings and all have proc handlers.
6320 */ 6320 */
6321 for (entry = *tablep; entry->mode; entry++) { 6321 for (entry = *tablep; entry->mode; entry++) {
6322 if (entry->child) 6322 if (entry->child)
6323 sd_free_ctl_entry(&entry->child); 6323 sd_free_ctl_entry(&entry->child);
6324 if (entry->proc_handler == NULL) 6324 if (entry->proc_handler == NULL)
6325 kfree(entry->procname); 6325 kfree(entry->procname);
6326 } 6326 }
6327 6327
6328 kfree(*tablep); 6328 kfree(*tablep);
6329 *tablep = NULL; 6329 *tablep = NULL;
6330 } 6330 }
6331 6331
6332 static void 6332 static void
6333 set_table_entry(struct ctl_table *entry, 6333 set_table_entry(struct ctl_table *entry,
6334 const char *procname, void *data, int maxlen, 6334 const char *procname, void *data, int maxlen,
6335 mode_t mode, proc_handler *proc_handler) 6335 mode_t mode, proc_handler *proc_handler)
6336 { 6336 {
6337 entry->procname = procname; 6337 entry->procname = procname;
6338 entry->data = data; 6338 entry->data = data;
6339 entry->maxlen = maxlen; 6339 entry->maxlen = maxlen;
6340 entry->mode = mode; 6340 entry->mode = mode;
6341 entry->proc_handler = proc_handler; 6341 entry->proc_handler = proc_handler;
6342 } 6342 }
6343 6343
6344 static struct ctl_table * 6344 static struct ctl_table *
6345 sd_alloc_ctl_domain_table(struct sched_domain *sd) 6345 sd_alloc_ctl_domain_table(struct sched_domain *sd)
6346 { 6346 {
6347 struct ctl_table *table = sd_alloc_ctl_entry(13); 6347 struct ctl_table *table = sd_alloc_ctl_entry(13);
6348 6348
6349 if (table == NULL) 6349 if (table == NULL)
6350 return NULL; 6350 return NULL;
6351 6351
6352 set_table_entry(&table[0], "min_interval", &sd->min_interval, 6352 set_table_entry(&table[0], "min_interval", &sd->min_interval,
6353 sizeof(long), 0644, proc_doulongvec_minmax); 6353 sizeof(long), 0644, proc_doulongvec_minmax);
6354 set_table_entry(&table[1], "max_interval", &sd->max_interval, 6354 set_table_entry(&table[1], "max_interval", &sd->max_interval,
6355 sizeof(long), 0644, proc_doulongvec_minmax); 6355 sizeof(long), 0644, proc_doulongvec_minmax);
6356 set_table_entry(&table[2], "busy_idx", &sd->busy_idx, 6356 set_table_entry(&table[2], "busy_idx", &sd->busy_idx,
6357 sizeof(int), 0644, proc_dointvec_minmax); 6357 sizeof(int), 0644, proc_dointvec_minmax);
6358 set_table_entry(&table[3], "idle_idx", &sd->idle_idx, 6358 set_table_entry(&table[3], "idle_idx", &sd->idle_idx,
6359 sizeof(int), 0644, proc_dointvec_minmax); 6359 sizeof(int), 0644, proc_dointvec_minmax);
6360 set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, 6360 set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx,
6361 sizeof(int), 0644, proc_dointvec_minmax); 6361 sizeof(int), 0644, proc_dointvec_minmax);
6362 set_table_entry(&table[5], "wake_idx", &sd->wake_idx, 6362 set_table_entry(&table[5], "wake_idx", &sd->wake_idx,
6363 sizeof(int), 0644, proc_dointvec_minmax); 6363 sizeof(int), 0644, proc_dointvec_minmax);
6364 set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, 6364 set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx,
6365 sizeof(int), 0644, proc_dointvec_minmax); 6365 sizeof(int), 0644, proc_dointvec_minmax);
6366 set_table_entry(&table[7], "busy_factor", &sd->busy_factor, 6366 set_table_entry(&table[7], "busy_factor", &sd->busy_factor,
6367 sizeof(int), 0644, proc_dointvec_minmax); 6367 sizeof(int), 0644, proc_dointvec_minmax);
6368 set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, 6368 set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct,
6369 sizeof(int), 0644, proc_dointvec_minmax); 6369 sizeof(int), 0644, proc_dointvec_minmax);
6370 set_table_entry(&table[9], "cache_nice_tries", 6370 set_table_entry(&table[9], "cache_nice_tries",
6371 &sd->cache_nice_tries, 6371 &sd->cache_nice_tries,
6372 sizeof(int), 0644, proc_dointvec_minmax); 6372 sizeof(int), 0644, proc_dointvec_minmax);
6373 set_table_entry(&table[10], "flags", &sd->flags, 6373 set_table_entry(&table[10], "flags", &sd->flags,
6374 sizeof(int), 0644, proc_dointvec_minmax); 6374 sizeof(int), 0644, proc_dointvec_minmax);
6375 set_table_entry(&table[11], "name", sd->name, 6375 set_table_entry(&table[11], "name", sd->name,
6376 CORENAME_MAX_SIZE, 0444, proc_dostring); 6376 CORENAME_MAX_SIZE, 0444, proc_dostring);
6377 /* &table[12] is terminator */ 6377 /* &table[12] is terminator */
6378 6378
6379 return table; 6379 return table;
6380 } 6380 }
6381 6381
6382 static ctl_table *sd_alloc_ctl_cpu_table(int cpu) 6382 static ctl_table *sd_alloc_ctl_cpu_table(int cpu)
6383 { 6383 {
6384 struct ctl_table *entry, *table; 6384 struct ctl_table *entry, *table;
6385 struct sched_domain *sd; 6385 struct sched_domain *sd;
6386 int domain_num = 0, i; 6386 int domain_num = 0, i;
6387 char buf[32]; 6387 char buf[32];
6388 6388
6389 for_each_domain(cpu, sd) 6389 for_each_domain(cpu, sd)
6390 domain_num++; 6390 domain_num++;
6391 entry = table = sd_alloc_ctl_entry(domain_num + 1); 6391 entry = table = sd_alloc_ctl_entry(domain_num + 1);
6392 if (table == NULL) 6392 if (table == NULL)
6393 return NULL; 6393 return NULL;
6394 6394
6395 i = 0; 6395 i = 0;
6396 for_each_domain(cpu, sd) { 6396 for_each_domain(cpu, sd) {
6397 snprintf(buf, 32, "domain%d", i); 6397 snprintf(buf, 32, "domain%d", i);
6398 entry->procname = kstrdup(buf, GFP_KERNEL); 6398 entry->procname = kstrdup(buf, GFP_KERNEL);
6399 entry->mode = 0555; 6399 entry->mode = 0555;
6400 entry->child = sd_alloc_ctl_domain_table(sd); 6400 entry->child = sd_alloc_ctl_domain_table(sd);
6401 entry++; 6401 entry++;
6402 i++; 6402 i++;
6403 } 6403 }
6404 return table; 6404 return table;
6405 } 6405 }
6406 6406
6407 static struct ctl_table_header *sd_sysctl_header; 6407 static struct ctl_table_header *sd_sysctl_header;
6408 static void register_sched_domain_sysctl(void) 6408 static void register_sched_domain_sysctl(void)
6409 { 6409 {
6410 int i, cpu_num = num_online_cpus(); 6410 int i, cpu_num = num_online_cpus();
6411 struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); 6411 struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1);
6412 char buf[32]; 6412 char buf[32];
6413 6413
6414 WARN_ON(sd_ctl_dir[0].child); 6414 WARN_ON(sd_ctl_dir[0].child);
6415 sd_ctl_dir[0].child = entry; 6415 sd_ctl_dir[0].child = entry;
6416 6416
6417 if (entry == NULL) 6417 if (entry == NULL)
6418 return; 6418 return;
6419 6419
6420 for_each_online_cpu(i) { 6420 for_each_online_cpu(i) {
6421 snprintf(buf, 32, "cpu%d", i); 6421 snprintf(buf, 32, "cpu%d", i);
6422 entry->procname = kstrdup(buf, GFP_KERNEL); 6422 entry->procname = kstrdup(buf, GFP_KERNEL);
6423 entry->mode = 0555; 6423 entry->mode = 0555;
6424 entry->child = sd_alloc_ctl_cpu_table(i); 6424 entry->child = sd_alloc_ctl_cpu_table(i);
6425 entry++; 6425 entry++;
6426 } 6426 }
6427 6427
6428 WARN_ON(sd_sysctl_header); 6428 WARN_ON(sd_sysctl_header);
6429 sd_sysctl_header = register_sysctl_table(sd_ctl_root); 6429 sd_sysctl_header = register_sysctl_table(sd_ctl_root);
6430 } 6430 }
6431 6431
6432 /* may be called multiple times per register */ 6432 /* may be called multiple times per register */
6433 static void unregister_sched_domain_sysctl(void) 6433 static void unregister_sched_domain_sysctl(void)
6434 { 6434 {
6435 if (sd_sysctl_header) 6435 if (sd_sysctl_header)
6436 unregister_sysctl_table(sd_sysctl_header); 6436 unregister_sysctl_table(sd_sysctl_header);
6437 sd_sysctl_header = NULL; 6437 sd_sysctl_header = NULL;
6438 if (sd_ctl_dir[0].child) 6438 if (sd_ctl_dir[0].child)
6439 sd_free_ctl_entry(&sd_ctl_dir[0].child); 6439 sd_free_ctl_entry(&sd_ctl_dir[0].child);
6440 } 6440 }
6441 #else 6441 #else
6442 static void register_sched_domain_sysctl(void) 6442 static void register_sched_domain_sysctl(void)
6443 { 6443 {
6444 } 6444 }
6445 static void unregister_sched_domain_sysctl(void) 6445 static void unregister_sched_domain_sysctl(void)
6446 { 6446 {
6447 } 6447 }
6448 #endif 6448 #endif
6449 6449
6450 static void set_rq_online(struct rq *rq) 6450 static void set_rq_online(struct rq *rq)
6451 { 6451 {
6452 if (!rq->online) { 6452 if (!rq->online) {
6453 const struct sched_class *class; 6453 const struct sched_class *class;
6454 6454
6455 cpu_set(rq->cpu, rq->rd->online); 6455 cpu_set(rq->cpu, rq->rd->online);
6456 rq->online = 1; 6456 rq->online = 1;
6457 6457
6458 for_each_class(class) { 6458 for_each_class(class) {
6459 if (class->rq_online) 6459 if (class->rq_online)
6460 class->rq_online(rq); 6460 class->rq_online(rq);
6461 } 6461 }
6462 } 6462 }
6463 } 6463 }
6464 6464
6465 static void set_rq_offline(struct rq *rq) 6465 static void set_rq_offline(struct rq *rq)
6466 { 6466 {
6467 if (rq->online) { 6467 if (rq->online) {
6468 const struct sched_class *class; 6468 const struct sched_class *class;
6469 6469
6470 for_each_class(class) { 6470 for_each_class(class) {
6471 if (class->rq_offline) 6471 if (class->rq_offline)
6472 class->rq_offline(rq); 6472 class->rq_offline(rq);
6473 } 6473 }
6474 6474
6475 cpu_clear(rq->cpu, rq->rd->online); 6475 cpu_clear(rq->cpu, rq->rd->online);
6476 rq->online = 0; 6476 rq->online = 0;
6477 } 6477 }
6478 } 6478 }
6479 6479
6480 /* 6480 /*
6481 * migration_call - callback that gets triggered when a CPU is added. 6481 * migration_call - callback that gets triggered when a CPU is added.
6482 * Here we can start up the necessary migration thread for the new CPU. 6482 * Here we can start up the necessary migration thread for the new CPU.
6483 */ 6483 */
6484 static int __cpuinit 6484 static int __cpuinit
6485 migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) 6485 migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu)
6486 { 6486 {
6487 struct task_struct *p; 6487 struct task_struct *p;
6488 int cpu = (long)hcpu; 6488 int cpu = (long)hcpu;
6489 unsigned long flags; 6489 unsigned long flags;
6490 struct rq *rq; 6490 struct rq *rq;
6491 6491
6492 switch (action) { 6492 switch (action) {
6493 6493
6494 case CPU_UP_PREPARE: 6494 case CPU_UP_PREPARE:
6495 case CPU_UP_PREPARE_FROZEN: 6495 case CPU_UP_PREPARE_FROZEN:
6496 p = kthread_create(migration_thread, hcpu, "migration/%d", cpu); 6496 p = kthread_create(migration_thread, hcpu, "migration/%d", cpu);
6497 if (IS_ERR(p)) 6497 if (IS_ERR(p))
6498 return NOTIFY_BAD; 6498 return NOTIFY_BAD;
6499 kthread_bind(p, cpu); 6499 kthread_bind(p, cpu);
6500 /* Must be high prio: stop_machine expects to yield to it. */ 6500 /* Must be high prio: stop_machine expects to yield to it. */
6501 rq = task_rq_lock(p, &flags); 6501 rq = task_rq_lock(p, &flags);
6502 __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); 6502 __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1);
6503 task_rq_unlock(rq, &flags); 6503 task_rq_unlock(rq, &flags);
6504 cpu_rq(cpu)->migration_thread = p; 6504 cpu_rq(cpu)->migration_thread = p;
6505 break; 6505 break;
6506 6506
6507 case CPU_ONLINE: 6507 case CPU_ONLINE:
6508 case CPU_ONLINE_FROZEN: 6508 case CPU_ONLINE_FROZEN:
6509 /* Strictly unnecessary, as first user will wake it. */ 6509 /* Strictly unnecessary, as first user will wake it. */
6510 wake_up_process(cpu_rq(cpu)->migration_thread); 6510 wake_up_process(cpu_rq(cpu)->migration_thread);
6511 6511
6512 /* Update our root-domain */ 6512 /* Update our root-domain */
6513 rq = cpu_rq(cpu); 6513 rq = cpu_rq(cpu);
6514 spin_lock_irqsave(&rq->lock, flags); 6514 spin_lock_irqsave(&rq->lock, flags);
6515 if (rq->rd) { 6515 if (rq->rd) {
6516 BUG_ON(!cpu_isset(cpu, rq->rd->span)); 6516 BUG_ON(!cpu_isset(cpu, rq->rd->span));
6517 6517
6518 set_rq_online(rq); 6518 set_rq_online(rq);
6519 } 6519 }
6520 spin_unlock_irqrestore(&rq->lock, flags); 6520 spin_unlock_irqrestore(&rq->lock, flags);
6521 break; 6521 break;
6522 6522
6523 #ifdef CONFIG_HOTPLUG_CPU 6523 #ifdef CONFIG_HOTPLUG_CPU
6524 case CPU_UP_CANCELED: 6524 case CPU_UP_CANCELED:
6525 case CPU_UP_CANCELED_FROZEN: 6525 case CPU_UP_CANCELED_FROZEN:
6526 if (!cpu_rq(cpu)->migration_thread) 6526 if (!cpu_rq(cpu)->migration_thread)
6527 break; 6527 break;
6528 /* Unbind it from offline cpu so it can run. Fall thru. */ 6528 /* Unbind it from offline cpu so it can run. Fall thru. */
6529 kthread_bind(cpu_rq(cpu)->migration_thread, 6529 kthread_bind(cpu_rq(cpu)->migration_thread,
6530 any_online_cpu(cpu_online_map)); 6530 any_online_cpu(cpu_online_map));
6531 kthread_stop(cpu_rq(cpu)->migration_thread); 6531 kthread_stop(cpu_rq(cpu)->migration_thread);
6532 cpu_rq(cpu)->migration_thread = NULL; 6532 cpu_rq(cpu)->migration_thread = NULL;
6533 break; 6533 break;
6534 6534
6535 case CPU_DEAD: 6535 case CPU_DEAD:
6536 case CPU_DEAD_FROZEN: 6536 case CPU_DEAD_FROZEN:
6537 cpuset_lock(); /* around calls to cpuset_cpus_allowed_lock() */ 6537 cpuset_lock(); /* around calls to cpuset_cpus_allowed_lock() */
6538 migrate_live_tasks(cpu); 6538 migrate_live_tasks(cpu);
6539 rq = cpu_rq(cpu); 6539 rq = cpu_rq(cpu);
6540 kthread_stop(rq->migration_thread); 6540 kthread_stop(rq->migration_thread);
6541 rq->migration_thread = NULL; 6541 rq->migration_thread = NULL;
6542 /* Idle task back to normal (off runqueue, low prio) */ 6542 /* Idle task back to normal (off runqueue, low prio) */
6543 spin_lock_irq(&rq->lock); 6543 spin_lock_irq(&rq->lock);
6544 update_rq_clock(rq); 6544 update_rq_clock(rq);
6545 deactivate_task(rq, rq->idle, 0); 6545 deactivate_task(rq, rq->idle, 0);
6546 rq->idle->static_prio = MAX_PRIO; 6546 rq->idle->static_prio = MAX_PRIO;
6547 __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); 6547 __setscheduler(rq, rq->idle, SCHED_NORMAL, 0);
6548 rq->idle->sched_class = &idle_sched_class; 6548 rq->idle->sched_class = &idle_sched_class;
6549 migrate_dead_tasks(cpu); 6549 migrate_dead_tasks(cpu);
6550 spin_unlock_irq(&rq->lock); 6550 spin_unlock_irq(&rq->lock);
6551 cpuset_unlock(); 6551 cpuset_unlock();
6552 migrate_nr_uninterruptible(rq); 6552 migrate_nr_uninterruptible(rq);
6553 BUG_ON(rq->nr_running != 0); 6553 BUG_ON(rq->nr_running != 0);
6554 6554
6555 /* 6555 /*
6556 * No need to migrate the tasks: it was best-effort if 6556 * No need to migrate the tasks: it was best-effort if
6557 * they didn't take sched_hotcpu_mutex. Just wake up 6557 * they didn't take sched_hotcpu_mutex. Just wake up
6558 * the requestors. 6558 * the requestors.
6559 */ 6559 */
6560 spin_lock_irq(&rq->lock); 6560 spin_lock_irq(&rq->lock);
6561 while (!list_empty(&rq->migration_queue)) { 6561 while (!list_empty(&rq->migration_queue)) {
6562 struct migration_req *req; 6562 struct migration_req *req;
6563 6563
6564 req = list_entry(rq->migration_queue.next, 6564 req = list_entry(rq->migration_queue.next,
6565 struct migration_req, list); 6565 struct migration_req, list);
6566 list_del_init(&req->list); 6566 list_del_init(&req->list);
6567 spin_unlock_irq(&rq->lock); 6567 spin_unlock_irq(&rq->lock);
6568 complete(&req->done); 6568 complete(&req->done);
6569 spin_lock_irq(&rq->lock); 6569 spin_lock_irq(&rq->lock);
6570 } 6570 }
6571 spin_unlock_irq(&rq->lock); 6571 spin_unlock_irq(&rq->lock);
6572 break; 6572 break;
6573 6573
6574 case CPU_DYING: 6574 case CPU_DYING:
6575 case CPU_DYING_FROZEN: 6575 case CPU_DYING_FROZEN:
6576 /* Update our root-domain */ 6576 /* Update our root-domain */
6577 rq = cpu_rq(cpu); 6577 rq = cpu_rq(cpu);
6578 spin_lock_irqsave(&rq->lock, flags); 6578 spin_lock_irqsave(&rq->lock, flags);
6579 if (rq->rd) { 6579 if (rq->rd) {
6580 BUG_ON(!cpu_isset(cpu, rq->rd->span)); 6580 BUG_ON(!cpu_isset(cpu, rq->rd->span));
6581 set_rq_offline(rq); 6581 set_rq_offline(rq);
6582 } 6582 }
6583 spin_unlock_irqrestore(&rq->lock, flags); 6583 spin_unlock_irqrestore(&rq->lock, flags);
6584 break; 6584 break;
6585 #endif 6585 #endif
6586 } 6586 }
6587 return NOTIFY_OK; 6587 return NOTIFY_OK;
6588 } 6588 }
6589 6589
6590 /* Register at highest priority so that task migration (migrate_all_tasks) 6590 /* Register at highest priority so that task migration (migrate_all_tasks)
6591 * happens before everything else. 6591 * happens before everything else.
6592 */ 6592 */
6593 static struct notifier_block __cpuinitdata migration_notifier = { 6593 static struct notifier_block __cpuinitdata migration_notifier = {
6594 .notifier_call = migration_call, 6594 .notifier_call = migration_call,
6595 .priority = 10 6595 .priority = 10
6596 }; 6596 };
6597 6597
6598 static int __init migration_init(void) 6598 static int __init migration_init(void)
6599 { 6599 {
6600 void *cpu = (void *)(long)smp_processor_id(); 6600 void *cpu = (void *)(long)smp_processor_id();
6601 int err; 6601 int err;
6602 6602
6603 /* Start one for the boot CPU: */ 6603 /* Start one for the boot CPU: */
6604 err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); 6604 err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu);
6605 BUG_ON(err == NOTIFY_BAD); 6605 BUG_ON(err == NOTIFY_BAD);
6606 migration_call(&migration_notifier, CPU_ONLINE, cpu); 6606 migration_call(&migration_notifier, CPU_ONLINE, cpu);
6607 register_cpu_notifier(&migration_notifier); 6607 register_cpu_notifier(&migration_notifier);
6608 6608
6609 return err; 6609 return err;
6610 } 6610 }
6611 early_initcall(migration_init); 6611 early_initcall(migration_init);
6612 #endif 6612 #endif
6613 6613
6614 #ifdef CONFIG_SMP 6614 #ifdef CONFIG_SMP
6615 6615
6616 #ifdef CONFIG_SCHED_DEBUG 6616 #ifdef CONFIG_SCHED_DEBUG
6617 6617
6618 static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, 6618 static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level,
6619 cpumask_t *groupmask) 6619 cpumask_t *groupmask)
6620 { 6620 {
6621 struct sched_group *group = sd->groups; 6621 struct sched_group *group = sd->groups;
6622 char str[256]; 6622 char str[256];
6623 6623
6624 cpulist_scnprintf(str, sizeof(str), sd->span); 6624 cpulist_scnprintf(str, sizeof(str), sd->span);
6625 cpus_clear(*groupmask); 6625 cpus_clear(*groupmask);
6626 6626
6627 printk(KERN_DEBUG "%*s domain %d: ", level, "", level); 6627 printk(KERN_DEBUG "%*s domain %d: ", level, "", level);
6628 6628
6629 if (!(sd->flags & SD_LOAD_BALANCE)) { 6629 if (!(sd->flags & SD_LOAD_BALANCE)) {
6630 printk("does not load-balance\n"); 6630 printk("does not load-balance\n");
6631 if (sd->parent) 6631 if (sd->parent)
6632 printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" 6632 printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain"
6633 " has parent"); 6633 " has parent");
6634 return -1; 6634 return -1;
6635 } 6635 }
6636 6636
6637 printk(KERN_CONT "span %s level %s\n", str, sd->name); 6637 printk(KERN_CONT "span %s level %s\n", str, sd->name);
6638 6638
6639 if (!cpu_isset(cpu, sd->span)) { 6639 if (!cpu_isset(cpu, sd->span)) {
6640 printk(KERN_ERR "ERROR: domain->span does not contain " 6640 printk(KERN_ERR "ERROR: domain->span does not contain "
6641 "CPU%d\n", cpu); 6641 "CPU%d\n", cpu);
6642 } 6642 }
6643 if (!cpu_isset(cpu, group->cpumask)) { 6643 if (!cpu_isset(cpu, group->cpumask)) {
6644 printk(KERN_ERR "ERROR: domain->groups does not contain" 6644 printk(KERN_ERR "ERROR: domain->groups does not contain"
6645 " CPU%d\n", cpu); 6645 " CPU%d\n", cpu);
6646 } 6646 }
6647 6647
6648 printk(KERN_DEBUG "%*s groups:", level + 1, ""); 6648 printk(KERN_DEBUG "%*s groups:", level + 1, "");
6649 do { 6649 do {
6650 if (!group) { 6650 if (!group) {
6651 printk("\n"); 6651 printk("\n");
6652 printk(KERN_ERR "ERROR: group is NULL\n"); 6652 printk(KERN_ERR "ERROR: group is NULL\n");
6653 break; 6653 break;
6654 } 6654 }
6655 6655
6656 if (!group->__cpu_power) { 6656 if (!group->__cpu_power) {
6657 printk(KERN_CONT "\n"); 6657 printk(KERN_CONT "\n");
6658 printk(KERN_ERR "ERROR: domain->cpu_power not " 6658 printk(KERN_ERR "ERROR: domain->cpu_power not "
6659 "set\n"); 6659 "set\n");
6660 break; 6660 break;
6661 } 6661 }
6662 6662
6663 if (!cpus_weight(group->cpumask)) { 6663 if (!cpus_weight(group->cpumask)) {
6664 printk(KERN_CONT "\n"); 6664 printk(KERN_CONT "\n");
6665 printk(KERN_ERR "ERROR: empty group\n"); 6665 printk(KERN_ERR "ERROR: empty group\n");
6666 break; 6666 break;
6667 } 6667 }
6668 6668
6669 if (cpus_intersects(*groupmask, group->cpumask)) { 6669 if (cpus_intersects(*groupmask, group->cpumask)) {
6670 printk(KERN_CONT "\n"); 6670 printk(KERN_CONT "\n");
6671 printk(KERN_ERR "ERROR: repeated CPUs\n"); 6671 printk(KERN_ERR "ERROR: repeated CPUs\n");
6672 break; 6672 break;
6673 } 6673 }
6674 6674
6675 cpus_or(*groupmask, *groupmask, group->cpumask); 6675 cpus_or(*groupmask, *groupmask, group->cpumask);
6676 6676
6677 cpulist_scnprintf(str, sizeof(str), group->cpumask); 6677 cpulist_scnprintf(str, sizeof(str), group->cpumask);
6678 printk(KERN_CONT " %s", str); 6678 printk(KERN_CONT " %s", str);
6679 6679
6680 group = group->next; 6680 group = group->next;
6681 } while (group != sd->groups); 6681 } while (group != sd->groups);
6682 printk(KERN_CONT "\n"); 6682 printk(KERN_CONT "\n");
6683 6683
6684 if (!cpus_equal(sd->span, *groupmask)) 6684 if (!cpus_equal(sd->span, *groupmask))
6685 printk(KERN_ERR "ERROR: groups don't span domain->span\n"); 6685 printk(KERN_ERR "ERROR: groups don't span domain->span\n");
6686 6686
6687 if (sd->parent && !cpus_subset(*groupmask, sd->parent->span)) 6687 if (sd->parent && !cpus_subset(*groupmask, sd->parent->span))
6688 printk(KERN_ERR "ERROR: parent span is not a superset " 6688 printk(KERN_ERR "ERROR: parent span is not a superset "
6689 "of domain->span\n"); 6689 "of domain->span\n");
6690 return 0; 6690 return 0;
6691 } 6691 }
6692 6692
6693 static void sched_domain_debug(struct sched_domain *sd, int cpu) 6693 static void sched_domain_debug(struct sched_domain *sd, int cpu)
6694 { 6694 {
6695 cpumask_t *groupmask; 6695 cpumask_t *groupmask;
6696 int level = 0; 6696 int level = 0;
6697 6697
6698 if (!sd) { 6698 if (!sd) {
6699 printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); 6699 printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu);
6700 return; 6700 return;
6701 } 6701 }
6702 6702
6703 printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); 6703 printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu);
6704 6704
6705 groupmask = kmalloc(sizeof(cpumask_t), GFP_KERNEL); 6705 groupmask = kmalloc(sizeof(cpumask_t), GFP_KERNEL);
6706 if (!groupmask) { 6706 if (!groupmask) {
6707 printk(KERN_DEBUG "Cannot load-balance (out of memory)\n"); 6707 printk(KERN_DEBUG "Cannot load-balance (out of memory)\n");
6708 return; 6708 return;
6709 } 6709 }
6710 6710
6711 for (;;) { 6711 for (;;) {
6712 if (sched_domain_debug_one(sd, cpu, level, groupmask)) 6712 if (sched_domain_debug_one(sd, cpu, level, groupmask))
6713 break; 6713 break;
6714 level++; 6714 level++;
6715 sd = sd->parent; 6715 sd = sd->parent;
6716 if (!sd) 6716 if (!sd)
6717 break; 6717 break;
6718 } 6718 }
6719 kfree(groupmask); 6719 kfree(groupmask);
6720 } 6720 }
6721 #else /* !CONFIG_SCHED_DEBUG */ 6721 #else /* !CONFIG_SCHED_DEBUG */
6722 # define sched_domain_debug(sd, cpu) do { } while (0) 6722 # define sched_domain_debug(sd, cpu) do { } while (0)
6723 #endif /* CONFIG_SCHED_DEBUG */ 6723 #endif /* CONFIG_SCHED_DEBUG */
6724 6724
6725 static int sd_degenerate(struct sched_domain *sd) 6725 static int sd_degenerate(struct sched_domain *sd)
6726 { 6726 {
6727 if (cpus_weight(sd->span) == 1) 6727 if (cpus_weight(sd->span) == 1)
6728 return 1; 6728 return 1;
6729 6729
6730 /* Following flags need at least 2 groups */ 6730 /* Following flags need at least 2 groups */
6731 if (sd->flags & (SD_LOAD_BALANCE | 6731 if (sd->flags & (SD_LOAD_BALANCE |
6732 SD_BALANCE_NEWIDLE | 6732 SD_BALANCE_NEWIDLE |
6733 SD_BALANCE_FORK | 6733 SD_BALANCE_FORK |
6734 SD_BALANCE_EXEC | 6734 SD_BALANCE_EXEC |
6735 SD_SHARE_CPUPOWER | 6735 SD_SHARE_CPUPOWER |
6736 SD_SHARE_PKG_RESOURCES)) { 6736 SD_SHARE_PKG_RESOURCES)) {
6737 if (sd->groups != sd->groups->next) 6737 if (sd->groups != sd->groups->next)
6738 return 0; 6738 return 0;
6739 } 6739 }
6740 6740
6741 /* Following flags don't use groups */ 6741 /* Following flags don't use groups */
6742 if (sd->flags & (SD_WAKE_IDLE | 6742 if (sd->flags & (SD_WAKE_IDLE |
6743 SD_WAKE_AFFINE | 6743 SD_WAKE_AFFINE |
6744 SD_WAKE_BALANCE)) 6744 SD_WAKE_BALANCE))
6745 return 0; 6745 return 0;
6746 6746
6747 return 1; 6747 return 1;
6748 } 6748 }
6749 6749
6750 static int 6750 static int
6751 sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) 6751 sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent)
6752 { 6752 {
6753 unsigned long cflags = sd->flags, pflags = parent->flags; 6753 unsigned long cflags = sd->flags, pflags = parent->flags;
6754 6754
6755 if (sd_degenerate(parent)) 6755 if (sd_degenerate(parent))
6756 return 1; 6756 return 1;
6757 6757
6758 if (!cpus_equal(sd->span, parent->span)) 6758 if (!cpus_equal(sd->span, parent->span))
6759 return 0; 6759 return 0;
6760 6760
6761 /* Does parent contain flags not in child? */ 6761 /* Does parent contain flags not in child? */
6762 /* WAKE_BALANCE is a subset of WAKE_AFFINE */ 6762 /* WAKE_BALANCE is a subset of WAKE_AFFINE */
6763 if (cflags & SD_WAKE_AFFINE) 6763 if (cflags & SD_WAKE_AFFINE)
6764 pflags &= ~SD_WAKE_BALANCE; 6764 pflags &= ~SD_WAKE_BALANCE;
6765 /* Flags needing groups don't count if only 1 group in parent */ 6765 /* Flags needing groups don't count if only 1 group in parent */
6766 if (parent->groups == parent->groups->next) { 6766 if (parent->groups == parent->groups->next) {
6767 pflags &= ~(SD_LOAD_BALANCE | 6767 pflags &= ~(SD_LOAD_BALANCE |
6768 SD_BALANCE_NEWIDLE | 6768 SD_BALANCE_NEWIDLE |
6769 SD_BALANCE_FORK | 6769 SD_BALANCE_FORK |
6770 SD_BALANCE_EXEC | 6770 SD_BALANCE_EXEC |
6771 SD_SHARE_CPUPOWER | 6771 SD_SHARE_CPUPOWER |
6772 SD_SHARE_PKG_RESOURCES); 6772 SD_SHARE_PKG_RESOURCES);
6773 if (nr_node_ids == 1) 6773 if (nr_node_ids == 1)
6774 pflags &= ~SD_SERIALIZE; 6774 pflags &= ~SD_SERIALIZE;
6775 } 6775 }
6776 if (~cflags & pflags) 6776 if (~cflags & pflags)
6777 return 0; 6777 return 0;
6778 6778
6779 return 1; 6779 return 1;
6780 } 6780 }
6781 6781
6782 static void rq_attach_root(struct rq *rq, struct root_domain *rd) 6782 static void rq_attach_root(struct rq *rq, struct root_domain *rd)
6783 { 6783 {
6784 unsigned long flags; 6784 unsigned long flags;
6785 6785
6786 spin_lock_irqsave(&rq->lock, flags); 6786 spin_lock_irqsave(&rq->lock, flags);
6787 6787
6788 if (rq->rd) { 6788 if (rq->rd) {
6789 struct root_domain *old_rd = rq->rd; 6789 struct root_domain *old_rd = rq->rd;
6790 6790
6791 if (cpu_isset(rq->cpu, old_rd->online)) 6791 if (cpu_isset(rq->cpu, old_rd->online))
6792 set_rq_offline(rq); 6792 set_rq_offline(rq);
6793 6793
6794 cpu_clear(rq->cpu, old_rd->span); 6794 cpu_clear(rq->cpu, old_rd->span);
6795 6795
6796 if (atomic_dec_and_test(&old_rd->refcount)) 6796 if (atomic_dec_and_test(&old_rd->refcount))
6797 kfree(old_rd); 6797 kfree(old_rd);
6798 } 6798 }
6799 6799
6800 atomic_inc(&rd->refcount); 6800 atomic_inc(&rd->refcount);
6801 rq->rd = rd; 6801 rq->rd = rd;
6802 6802
6803 cpu_set(rq->cpu, rd->span); 6803 cpu_set(rq->cpu, rd->span);
6804 if (cpu_isset(rq->cpu, cpu_online_map)) 6804 if (cpu_isset(rq->cpu, cpu_online_map))
6805 set_rq_online(rq); 6805 set_rq_online(rq);
6806 6806
6807 spin_unlock_irqrestore(&rq->lock, flags); 6807 spin_unlock_irqrestore(&rq->lock, flags);
6808 } 6808 }
6809 6809
6810 static void init_rootdomain(struct root_domain *rd) 6810 static void init_rootdomain(struct root_domain *rd)
6811 { 6811 {
6812 memset(rd, 0, sizeof(*rd)); 6812 memset(rd, 0, sizeof(*rd));
6813 6813
6814 cpus_clear(rd->span); 6814 cpus_clear(rd->span);
6815 cpus_clear(rd->online); 6815 cpus_clear(rd->online);
6816 6816
6817 cpupri_init(&rd->cpupri); 6817 cpupri_init(&rd->cpupri);
6818 } 6818 }
6819 6819
6820 static void init_defrootdomain(void) 6820 static void init_defrootdomain(void)
6821 { 6821 {
6822 init_rootdomain(&def_root_domain); 6822 init_rootdomain(&def_root_domain);
6823 atomic_set(&def_root_domain.refcount, 1); 6823 atomic_set(&def_root_domain.refcount, 1);
6824 } 6824 }
6825 6825
6826 static struct root_domain *alloc_rootdomain(void) 6826 static struct root_domain *alloc_rootdomain(void)
6827 { 6827 {
6828 struct root_domain *rd; 6828 struct root_domain *rd;
6829 6829
6830 rd = kmalloc(sizeof(*rd), GFP_KERNEL); 6830 rd = kmalloc(sizeof(*rd), GFP_KERNEL);
6831 if (!rd) 6831 if (!rd)
6832 return NULL; 6832 return NULL;
6833 6833
6834 init_rootdomain(rd); 6834 init_rootdomain(rd);
6835 6835
6836 return rd; 6836 return rd;
6837 } 6837 }
6838 6838
6839 /* 6839 /*
6840 * Attach the domain 'sd' to 'cpu' as its base domain. Callers must 6840 * Attach the domain 'sd' to 'cpu' as its base domain. Callers must
6841 * hold the hotplug lock. 6841 * hold the hotplug lock.
6842 */ 6842 */
6843 static void 6843 static void
6844 cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) 6844 cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu)
6845 { 6845 {
6846 struct rq *rq = cpu_rq(cpu); 6846 struct rq *rq = cpu_rq(cpu);
6847 struct sched_domain *tmp; 6847 struct sched_domain *tmp;
6848 6848
6849 /* Remove the sched domains which do not contribute to scheduling. */ 6849 /* Remove the sched domains which do not contribute to scheduling. */
6850 for (tmp = sd; tmp; ) { 6850 for (tmp = sd; tmp; ) {
6851 struct sched_domain *parent = tmp->parent; 6851 struct sched_domain *parent = tmp->parent;
6852 if (!parent) 6852 if (!parent)
6853 break; 6853 break;
6854 6854
6855 if (sd_parent_degenerate(tmp, parent)) { 6855 if (sd_parent_degenerate(tmp, parent)) {
6856 tmp->parent = parent->parent; 6856 tmp->parent = parent->parent;
6857 if (parent->parent) 6857 if (parent->parent)
6858 parent->parent->child = tmp; 6858 parent->parent->child = tmp;
6859 } else 6859 } else
6860 tmp = tmp->parent; 6860 tmp = tmp->parent;
6861 } 6861 }
6862 6862
6863 if (sd && sd_degenerate(sd)) { 6863 if (sd && sd_degenerate(sd)) {
6864 sd = sd->parent; 6864 sd = sd->parent;
6865 if (sd) 6865 if (sd)
6866 sd->child = NULL; 6866 sd->child = NULL;
6867 } 6867 }
6868 6868
6869 sched_domain_debug(sd, cpu); 6869 sched_domain_debug(sd, cpu);
6870 6870
6871 rq_attach_root(rq, rd); 6871 rq_attach_root(rq, rd);
6872 rcu_assign_pointer(rq->sd, sd); 6872 rcu_assign_pointer(rq->sd, sd);
6873 } 6873 }
6874 6874
6875 /* cpus with isolated domains */ 6875 /* cpus with isolated domains */
6876 static cpumask_t cpu_isolated_map = CPU_MASK_NONE; 6876 static cpumask_t cpu_isolated_map = CPU_MASK_NONE;
6877 6877
6878 /* Setup the mask of cpus configured for isolated domains */ 6878 /* Setup the mask of cpus configured for isolated domains */
6879 static int __init isolated_cpu_setup(char *str) 6879 static int __init isolated_cpu_setup(char *str)
6880 { 6880 {
6881 static int __initdata ints[NR_CPUS]; 6881 static int __initdata ints[NR_CPUS];
6882 int i; 6882 int i;
6883 6883
6884 str = get_options(str, ARRAY_SIZE(ints), ints); 6884 str = get_options(str, ARRAY_SIZE(ints), ints);
6885 cpus_clear(cpu_isolated_map); 6885 cpus_clear(cpu_isolated_map);
6886 for (i = 1; i <= ints[0]; i++) 6886 for (i = 1; i <= ints[0]; i++)
6887 if (ints[i] < NR_CPUS) 6887 if (ints[i] < NR_CPUS)
6888 cpu_set(ints[i], cpu_isolated_map); 6888 cpu_set(ints[i], cpu_isolated_map);
6889 return 1; 6889 return 1;
6890 } 6890 }
6891 6891
6892 __setup("isolcpus=", isolated_cpu_setup); 6892 __setup("isolcpus=", isolated_cpu_setup);
6893 6893
6894 /* 6894 /*
6895 * init_sched_build_groups takes the cpumask we wish to span, and a pointer 6895 * init_sched_build_groups takes the cpumask we wish to span, and a pointer
6896 * to a function which identifies what group(along with sched group) a CPU 6896 * to a function which identifies what group(along with sched group) a CPU
6897 * belongs to. The return value of group_fn must be a >= 0 and < NR_CPUS 6897 * belongs to. The return value of group_fn must be a >= 0 and < NR_CPUS
6898 * (due to the fact that we keep track of groups covered with a cpumask_t). 6898 * (due to the fact that we keep track of groups covered with a cpumask_t).
6899 * 6899 *
6900 * init_sched_build_groups will build a circular linked list of the groups 6900 * init_sched_build_groups will build a circular linked list of the groups
6901 * covered by the given span, and will set each group's ->cpumask correctly, 6901 * covered by the given span, and will set each group's ->cpumask correctly,
6902 * and ->cpu_power to 0. 6902 * and ->cpu_power to 0.
6903 */ 6903 */
6904 static void 6904 static void
6905 init_sched_build_groups(const cpumask_t *span, const cpumask_t *cpu_map, 6905 init_sched_build_groups(const cpumask_t *span, const cpumask_t *cpu_map,
6906 int (*group_fn)(int cpu, const cpumask_t *cpu_map, 6906 int (*group_fn)(int cpu, const cpumask_t *cpu_map,
6907 struct sched_group **sg, 6907 struct sched_group **sg,
6908 cpumask_t *tmpmask), 6908 cpumask_t *tmpmask),
6909 cpumask_t *covered, cpumask_t *tmpmask) 6909 cpumask_t *covered, cpumask_t *tmpmask)
6910 { 6910 {
6911 struct sched_group *first = NULL, *last = NULL; 6911 struct sched_group *first = NULL, *last = NULL;
6912 int i; 6912 int i;
6913 6913
6914 cpus_clear(*covered); 6914 cpus_clear(*covered);
6915 6915
6916 for_each_cpu_mask_nr(i, *span) { 6916 for_each_cpu_mask_nr(i, *span) {
6917 struct sched_group *sg; 6917 struct sched_group *sg;
6918 int group = group_fn(i, cpu_map, &sg, tmpmask); 6918 int group = group_fn(i, cpu_map, &sg, tmpmask);
6919 int j; 6919 int j;
6920 6920
6921 if (cpu_isset(i, *covered)) 6921 if (cpu_isset(i, *covered))
6922 continue; 6922 continue;
6923 6923
6924 cpus_clear(sg->cpumask); 6924 cpus_clear(sg->cpumask);
6925 sg->__cpu_power = 0; 6925 sg->__cpu_power = 0;
6926 6926
6927 for_each_cpu_mask_nr(j, *span) { 6927 for_each_cpu_mask_nr(j, *span) {
6928 if (group_fn(j, cpu_map, NULL, tmpmask) != group) 6928 if (group_fn(j, cpu_map, NULL, tmpmask) != group)
6929 continue; 6929 continue;
6930 6930
6931 cpu_set(j, *covered); 6931 cpu_set(j, *covered);
6932 cpu_set(j, sg->cpumask); 6932 cpu_set(j, sg->cpumask);
6933 } 6933 }
6934 if (!first) 6934 if (!first)
6935 first = sg; 6935 first = sg;
6936 if (last) 6936 if (last)
6937 last->next = sg; 6937 last->next = sg;
6938 last = sg; 6938 last = sg;
6939 } 6939 }
6940 last->next = first; 6940 last->next = first;
6941 } 6941 }
6942 6942
6943 #define SD_NODES_PER_DOMAIN 16 6943 #define SD_NODES_PER_DOMAIN 16
6944 6944
6945 #ifdef CONFIG_NUMA 6945 #ifdef CONFIG_NUMA
6946 6946
6947 /** 6947 /**
6948 * find_next_best_node - find the next node to include in a sched_domain 6948 * find_next_best_node - find the next node to include in a sched_domain
6949 * @node: node whose sched_domain we're building 6949 * @node: node whose sched_domain we're building
6950 * @used_nodes: nodes already in the sched_domain 6950 * @used_nodes: nodes already in the sched_domain
6951 * 6951 *
6952 * Find the next node to include in a given scheduling domain. Simply 6952 * Find the next node to include in a given scheduling domain. Simply
6953 * finds the closest node not already in the @used_nodes map. 6953 * finds the closest node not already in the @used_nodes map.
6954 * 6954 *
6955 * Should use nodemask_t. 6955 * Should use nodemask_t.
6956 */ 6956 */
6957 static int find_next_best_node(int node, nodemask_t *used_nodes) 6957 static int find_next_best_node(int node, nodemask_t *used_nodes)
6958 { 6958 {
6959 int i, n, val, min_val, best_node = 0; 6959 int i, n, val, min_val, best_node = 0;
6960 6960
6961 min_val = INT_MAX; 6961 min_val = INT_MAX;
6962 6962
6963 for (i = 0; i < nr_node_ids; i++) { 6963 for (i = 0; i < nr_node_ids; i++) {
6964 /* Start at @node */ 6964 /* Start at @node */
6965 n = (node + i) % nr_node_ids; 6965 n = (node + i) % nr_node_ids;
6966 6966
6967 if (!nr_cpus_node(n)) 6967 if (!nr_cpus_node(n))
6968 continue; 6968 continue;
6969 6969
6970 /* Skip already used nodes */ 6970 /* Skip already used nodes */
6971 if (node_isset(n, *used_nodes)) 6971 if (node_isset(n, *used_nodes))
6972 continue; 6972 continue;
6973 6973
6974 /* Simple min distance search */ 6974 /* Simple min distance search */
6975 val = node_distance(node, n); 6975 val = node_distance(node, n);
6976 6976
6977 if (val < min_val) { 6977 if (val < min_val) {
6978 min_val = val; 6978 min_val = val;
6979 best_node = n; 6979 best_node = n;
6980 } 6980 }
6981 } 6981 }
6982 6982
6983 node_set(best_node, *used_nodes); 6983 node_set(best_node, *used_nodes);
6984 return best_node; 6984 return best_node;
6985 } 6985 }
6986 6986
6987 /** 6987 /**
6988 * sched_domain_node_span - get a cpumask for a node's sched_domain 6988 * sched_domain_node_span - get a cpumask for a node's sched_domain
6989 * @node: node whose cpumask we're constructing 6989 * @node: node whose cpumask we're constructing
6990 * @span: resulting cpumask 6990 * @span: resulting cpumask
6991 * 6991 *
6992 * Given a node, construct a good cpumask for its sched_domain to span. It 6992 * Given a node, construct a good cpumask for its sched_domain to span. It
6993 * should be one that prevents unnecessary balancing, but also spreads tasks 6993 * should be one that prevents unnecessary balancing, but also spreads tasks
6994 * out optimally. 6994 * out optimally.
6995 */ 6995 */
6996 static void sched_domain_node_span(int node, cpumask_t *span) 6996 static void sched_domain_node_span(int node, cpumask_t *span)
6997 { 6997 {
6998 nodemask_t used_nodes; 6998 nodemask_t used_nodes;
6999 node_to_cpumask_ptr(nodemask, node); 6999 node_to_cpumask_ptr(nodemask, node);
7000 int i; 7000 int i;
7001 7001
7002 cpus_clear(*span); 7002 cpus_clear(*span);
7003 nodes_clear(used_nodes); 7003 nodes_clear(used_nodes);
7004 7004
7005 cpus_or(*span, *span, *nodemask); 7005 cpus_or(*span, *span, *nodemask);
7006 node_set(node, used_nodes); 7006 node_set(node, used_nodes);
7007 7007
7008 for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { 7008 for (i = 1; i < SD_NODES_PER_DOMAIN; i++) {
7009 int next_node = find_next_best_node(node, &used_nodes); 7009 int next_node = find_next_best_node(node, &used_nodes);
7010 7010
7011 node_to_cpumask_ptr_next(nodemask, next_node); 7011 node_to_cpumask_ptr_next(nodemask, next_node);
7012 cpus_or(*span, *span, *nodemask); 7012 cpus_or(*span, *span, *nodemask);
7013 } 7013 }
7014 } 7014 }
7015 #endif /* CONFIG_NUMA */ 7015 #endif /* CONFIG_NUMA */
7016 7016
7017 int sched_smt_power_savings = 0, sched_mc_power_savings = 0; 7017 int sched_smt_power_savings = 0, sched_mc_power_savings = 0;
7018 7018
7019 /* 7019 /*
7020 * SMT sched-domains: 7020 * SMT sched-domains:
7021 */ 7021 */
7022 #ifdef CONFIG_SCHED_SMT 7022 #ifdef CONFIG_SCHED_SMT
7023 static DEFINE_PER_CPU(struct sched_domain, cpu_domains); 7023 static DEFINE_PER_CPU(struct sched_domain, cpu_domains);
7024 static DEFINE_PER_CPU(struct sched_group, sched_group_cpus); 7024 static DEFINE_PER_CPU(struct sched_group, sched_group_cpus);
7025 7025
7026 static int 7026 static int
7027 cpu_to_cpu_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, 7027 cpu_to_cpu_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg,
7028 cpumask_t *unused) 7028 cpumask_t *unused)
7029 { 7029 {
7030 if (sg) 7030 if (sg)
7031 *sg = &per_cpu(sched_group_cpus, cpu); 7031 *sg = &per_cpu(sched_group_cpus, cpu);
7032 return cpu; 7032 return cpu;
7033 } 7033 }
7034 #endif /* CONFIG_SCHED_SMT */ 7034 #endif /* CONFIG_SCHED_SMT */
7035 7035
7036 /* 7036 /*
7037 * multi-core sched-domains: 7037 * multi-core sched-domains:
7038 */ 7038 */
7039 #ifdef CONFIG_SCHED_MC 7039 #ifdef CONFIG_SCHED_MC
7040 static DEFINE_PER_CPU(struct sched_domain, core_domains); 7040 static DEFINE_PER_CPU(struct sched_domain, core_domains);
7041 static DEFINE_PER_CPU(struct sched_group, sched_group_core); 7041 static DEFINE_PER_CPU(struct sched_group, sched_group_core);
7042 #endif /* CONFIG_SCHED_MC */ 7042 #endif /* CONFIG_SCHED_MC */
7043 7043
7044 #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) 7044 #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT)
7045 static int 7045 static int
7046 cpu_to_core_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, 7046 cpu_to_core_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg,
7047 cpumask_t *mask) 7047 cpumask_t *mask)
7048 { 7048 {
7049 int group; 7049 int group;
7050 7050
7051 *mask = per_cpu(cpu_sibling_map, cpu); 7051 *mask = per_cpu(cpu_sibling_map, cpu);
7052 cpus_and(*mask, *mask, *cpu_map); 7052 cpus_and(*mask, *mask, *cpu_map);
7053 group = first_cpu(*mask); 7053 group = first_cpu(*mask);
7054 if (sg) 7054 if (sg)
7055 *sg = &per_cpu(sched_group_core, group); 7055 *sg = &per_cpu(sched_group_core, group);
7056 return group; 7056 return group;
7057 } 7057 }
7058 #elif defined(CONFIG_SCHED_MC) 7058 #elif defined(CONFIG_SCHED_MC)
7059 static int 7059 static int
7060 cpu_to_core_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, 7060 cpu_to_core_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg,
7061 cpumask_t *unused) 7061 cpumask_t *unused)
7062 { 7062 {
7063 if (sg) 7063 if (sg)
7064 *sg = &per_cpu(sched_group_core, cpu); 7064 *sg = &per_cpu(sched_group_core, cpu);
7065 return cpu; 7065 return cpu;
7066 } 7066 }
7067 #endif 7067 #endif
7068 7068
7069 static DEFINE_PER_CPU(struct sched_domain, phys_domains); 7069 static DEFINE_PER_CPU(struct sched_domain, phys_domains);
7070 static DEFINE_PER_CPU(struct sched_group, sched_group_phys); 7070 static DEFINE_PER_CPU(struct sched_group, sched_group_phys);
7071 7071
7072 static int 7072 static int
7073 cpu_to_phys_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, 7073 cpu_to_phys_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg,
7074 cpumask_t *mask) 7074 cpumask_t *mask)
7075 { 7075 {
7076 int group; 7076 int group;
7077 #ifdef CONFIG_SCHED_MC 7077 #ifdef CONFIG_SCHED_MC
7078 *mask = cpu_coregroup_map(cpu); 7078 *mask = cpu_coregroup_map(cpu);
7079 cpus_and(*mask, *mask, *cpu_map); 7079 cpus_and(*mask, *mask, *cpu_map);
7080 group = first_cpu(*mask); 7080 group = first_cpu(*mask);
7081 #elif defined(CONFIG_SCHED_SMT) 7081 #elif defined(CONFIG_SCHED_SMT)
7082 *mask = per_cpu(cpu_sibling_map, cpu); 7082 *mask = per_cpu(cpu_sibling_map, cpu);
7083 cpus_and(*mask, *mask, *cpu_map); 7083 cpus_and(*mask, *mask, *cpu_map);
7084 group = first_cpu(*mask); 7084 group = first_cpu(*mask);
7085 #else 7085 #else
7086 group = cpu; 7086 group = cpu;
7087 #endif 7087 #endif
7088 if (sg) 7088 if (sg)
7089 *sg = &per_cpu(sched_group_phys, group); 7089 *sg = &per_cpu(sched_group_phys, group);
7090 return group; 7090 return group;
7091 } 7091 }
7092 7092
7093 #ifdef CONFIG_NUMA 7093 #ifdef CONFIG_NUMA
7094 /* 7094 /*
7095 * The init_sched_build_groups can't handle what we want to do with node 7095 * The init_sched_build_groups can't handle what we want to do with node
7096 * groups, so roll our own. Now each node has its own list of groups which 7096 * groups, so roll our own. Now each node has its own list of groups which
7097 * gets dynamically allocated. 7097 * gets dynamically allocated.
7098 */ 7098 */
7099 static DEFINE_PER_CPU(struct sched_domain, node_domains); 7099 static DEFINE_PER_CPU(struct sched_domain, node_domains);
7100 static struct sched_group ***sched_group_nodes_bycpu; 7100 static struct sched_group ***sched_group_nodes_bycpu;
7101 7101
7102 static DEFINE_PER_CPU(struct sched_domain, allnodes_domains); 7102 static DEFINE_PER_CPU(struct sched_domain, allnodes_domains);
7103 static DEFINE_PER_CPU(struct sched_group, sched_group_allnodes); 7103 static DEFINE_PER_CPU(struct sched_group, sched_group_allnodes);
7104 7104
7105 static int cpu_to_allnodes_group(int cpu, const cpumask_t *cpu_map, 7105 static int cpu_to_allnodes_group(int cpu, const cpumask_t *cpu_map,
7106 struct sched_group **sg, cpumask_t *nodemask) 7106 struct sched_group **sg, cpumask_t *nodemask)
7107 { 7107 {
7108 int group; 7108 int group;
7109 7109
7110 *nodemask = node_to_cpumask(cpu_to_node(cpu)); 7110 *nodemask = node_to_cpumask(cpu_to_node(cpu));
7111 cpus_and(*nodemask, *nodemask, *cpu_map); 7111 cpus_and(*nodemask, *nodemask, *cpu_map);
7112 group = first_cpu(*nodemask); 7112 group = first_cpu(*nodemask);
7113 7113
7114 if (sg) 7114 if (sg)
7115 *sg = &per_cpu(sched_group_allnodes, group); 7115 *sg = &per_cpu(sched_group_allnodes, group);
7116 return group; 7116 return group;
7117 } 7117 }
7118 7118
7119 static void init_numa_sched_groups_power(struct sched_group *group_head) 7119 static void init_numa_sched_groups_power(struct sched_group *group_head)
7120 { 7120 {
7121 struct sched_group *sg = group_head; 7121 struct sched_group *sg = group_head;
7122 int j; 7122 int j;
7123 7123
7124 if (!sg) 7124 if (!sg)
7125 return; 7125 return;
7126 do { 7126 do {
7127 for_each_cpu_mask_nr(j, sg->cpumask) { 7127 for_each_cpu_mask_nr(j, sg->cpumask) {
7128 struct sched_domain *sd; 7128 struct sched_domain *sd;
7129 7129
7130 sd = &per_cpu(phys_domains, j); 7130 sd = &per_cpu(phys_domains, j);
7131 if (j != first_cpu(sd->groups->cpumask)) { 7131 if (j != first_cpu(sd->groups->cpumask)) {
7132 /* 7132 /*
7133 * Only add "power" once for each 7133 * Only add "power" once for each
7134 * physical package. 7134 * physical package.
7135 */ 7135 */
7136 continue; 7136 continue;
7137 } 7137 }
7138 7138
7139 sg_inc_cpu_power(sg, sd->groups->__cpu_power); 7139 sg_inc_cpu_power(sg, sd->groups->__cpu_power);
7140 } 7140 }
7141 sg = sg->next; 7141 sg = sg->next;
7142 } while (sg != group_head); 7142 } while (sg != group_head);
7143 } 7143 }
7144 #endif /* CONFIG_NUMA */ 7144 #endif /* CONFIG_NUMA */
7145 7145
7146 #ifdef CONFIG_NUMA 7146 #ifdef CONFIG_NUMA
7147 /* Free memory allocated for various sched_group structures */ 7147 /* Free memory allocated for various sched_group structures */
7148 static void free_sched_groups(const cpumask_t *cpu_map, cpumask_t *nodemask) 7148 static void free_sched_groups(const cpumask_t *cpu_map, cpumask_t *nodemask)
7149 { 7149 {
7150 int cpu, i; 7150 int cpu, i;
7151 7151
7152 for_each_cpu_mask_nr(cpu, *cpu_map) { 7152 for_each_cpu_mask_nr(cpu, *cpu_map) {
7153 struct sched_group **sched_group_nodes 7153 struct sched_group **sched_group_nodes
7154 = sched_group_nodes_bycpu[cpu]; 7154 = sched_group_nodes_bycpu[cpu];
7155 7155
7156 if (!sched_group_nodes) 7156 if (!sched_group_nodes)
7157 continue; 7157 continue;
7158 7158
7159 for (i = 0; i < nr_node_ids; i++) { 7159 for (i = 0; i < nr_node_ids; i++) {
7160 struct sched_group *oldsg, *sg = sched_group_nodes[i]; 7160 struct sched_group *oldsg, *sg = sched_group_nodes[i];
7161 7161
7162 *nodemask = node_to_cpumask(i); 7162 *nodemask = node_to_cpumask(i);
7163 cpus_and(*nodemask, *nodemask, *cpu_map); 7163 cpus_and(*nodemask, *nodemask, *cpu_map);
7164 if (cpus_empty(*nodemask)) 7164 if (cpus_empty(*nodemask))
7165 continue; 7165 continue;
7166 7166
7167 if (sg == NULL) 7167 if (sg == NULL)
7168 continue; 7168 continue;
7169 sg = sg->next; 7169 sg = sg->next;
7170 next_sg: 7170 next_sg:
7171 oldsg = sg; 7171 oldsg = sg;
7172 sg = sg->next; 7172 sg = sg->next;
7173 kfree(oldsg); 7173 kfree(oldsg);
7174 if (oldsg != sched_group_nodes[i]) 7174 if (oldsg != sched_group_nodes[i])
7175 goto next_sg; 7175 goto next_sg;
7176 } 7176 }
7177 kfree(sched_group_nodes); 7177 kfree(sched_group_nodes);
7178 sched_group_nodes_bycpu[cpu] = NULL; 7178 sched_group_nodes_bycpu[cpu] = NULL;
7179 } 7179 }
7180 } 7180 }
7181 #else /* !CONFIG_NUMA */ 7181 #else /* !CONFIG_NUMA */
7182 static void free_sched_groups(const cpumask_t *cpu_map, cpumask_t *nodemask) 7182 static void free_sched_groups(const cpumask_t *cpu_map, cpumask_t *nodemask)
7183 { 7183 {
7184 } 7184 }
7185 #endif /* CONFIG_NUMA */ 7185 #endif /* CONFIG_NUMA */
7186 7186
7187 /* 7187 /*
7188 * Initialize sched groups cpu_power. 7188 * Initialize sched groups cpu_power.
7189 * 7189 *
7190 * cpu_power indicates the capacity of sched group, which is used while 7190 * cpu_power indicates the capacity of sched group, which is used while
7191 * distributing the load between different sched groups in a sched domain. 7191 * distributing the load between different sched groups in a sched domain.
7192 * Typically cpu_power for all the groups in a sched domain will be same unless 7192 * Typically cpu_power for all the groups in a sched domain will be same unless
7193 * there are asymmetries in the topology. If there are asymmetries, group 7193 * there are asymmetries in the topology. If there are asymmetries, group
7194 * having more cpu_power will pickup more load compared to the group having 7194 * having more cpu_power will pickup more load compared to the group having
7195 * less cpu_power. 7195 * less cpu_power.
7196 * 7196 *
7197 * cpu_power will be a multiple of SCHED_LOAD_SCALE. This multiple represents 7197 * cpu_power will be a multiple of SCHED_LOAD_SCALE. This multiple represents
7198 * the maximum number of tasks a group can handle in the presence of other idle 7198 * the maximum number of tasks a group can handle in the presence of other idle
7199 * or lightly loaded groups in the same sched domain. 7199 * or lightly loaded groups in the same sched domain.
7200 */ 7200 */
7201 static void init_sched_groups_power(int cpu, struct sched_domain *sd) 7201 static void init_sched_groups_power(int cpu, struct sched_domain *sd)
7202 { 7202 {
7203 struct sched_domain *child; 7203 struct sched_domain *child;
7204 struct sched_group *group; 7204 struct sched_group *group;
7205 7205
7206 WARN_ON(!sd || !sd->groups); 7206 WARN_ON(!sd || !sd->groups);
7207 7207
7208 if (cpu != first_cpu(sd->groups->cpumask)) 7208 if (cpu != first_cpu(sd->groups->cpumask))
7209 return; 7209 return;
7210 7210
7211 child = sd->child; 7211 child = sd->child;
7212 7212
7213 sd->groups->__cpu_power = 0; 7213 sd->groups->__cpu_power = 0;
7214 7214
7215 /* 7215 /*
7216 * For perf policy, if the groups in child domain share resources 7216 * For perf policy, if the groups in child domain share resources
7217 * (for example cores sharing some portions of the cache hierarchy 7217 * (for example cores sharing some portions of the cache hierarchy
7218 * or SMT), then set this domain groups cpu_power such that each group 7218 * or SMT), then set this domain groups cpu_power such that each group
7219 * can handle only one task, when there are other idle groups in the 7219 * can handle only one task, when there are other idle groups in the
7220 * same sched domain. 7220 * same sched domain.
7221 */ 7221 */
7222 if (!child || (!(sd->flags & SD_POWERSAVINGS_BALANCE) && 7222 if (!child || (!(sd->flags & SD_POWERSAVINGS_BALANCE) &&
7223 (child->flags & 7223 (child->flags &
7224 (SD_SHARE_CPUPOWER | SD_SHARE_PKG_RESOURCES)))) { 7224 (SD_SHARE_CPUPOWER | SD_SHARE_PKG_RESOURCES)))) {
7225 sg_inc_cpu_power(sd->groups, SCHED_LOAD_SCALE); 7225 sg_inc_cpu_power(sd->groups, SCHED_LOAD_SCALE);
7226 return; 7226 return;
7227 } 7227 }
7228 7228
7229 /* 7229 /*
7230 * add cpu_power of each child group to this groups cpu_power 7230 * add cpu_power of each child group to this groups cpu_power
7231 */ 7231 */
7232 group = child->groups; 7232 group = child->groups;
7233 do { 7233 do {
7234 sg_inc_cpu_power(sd->groups, group->__cpu_power); 7234 sg_inc_cpu_power(sd->groups, group->__cpu_power);
7235 group = group->next; 7235 group = group->next;
7236 } while (group != child->groups); 7236 } while (group != child->groups);
7237 } 7237 }
7238 7238
7239 /* 7239 /*
7240 * Initializers for schedule domains 7240 * Initializers for schedule domains
7241 * Non-inlined to reduce accumulated stack pressure in build_sched_domains() 7241 * Non-inlined to reduce accumulated stack pressure in build_sched_domains()
7242 */ 7242 */
7243 7243
7244 #ifdef CONFIG_SCHED_DEBUG 7244 #ifdef CONFIG_SCHED_DEBUG
7245 # define SD_INIT_NAME(sd, type) sd->name = #type 7245 # define SD_INIT_NAME(sd, type) sd->name = #type
7246 #else 7246 #else
7247 # define SD_INIT_NAME(sd, type) do { } while (0) 7247 # define SD_INIT_NAME(sd, type) do { } while (0)
7248 #endif 7248 #endif
7249 7249
7250 #define SD_INIT(sd, type) sd_init_##type(sd) 7250 #define SD_INIT(sd, type) sd_init_##type(sd)
7251 7251
7252 #define SD_INIT_FUNC(type) \ 7252 #define SD_INIT_FUNC(type) \
7253 static noinline void sd_init_##type(struct sched_domain *sd) \ 7253 static noinline void sd_init_##type(struct sched_domain *sd) \
7254 { \ 7254 { \
7255 memset(sd, 0, sizeof(*sd)); \ 7255 memset(sd, 0, sizeof(*sd)); \
7256 *sd = SD_##type##_INIT; \ 7256 *sd = SD_##type##_INIT; \
7257 sd->level = SD_LV_##type; \ 7257 sd->level = SD_LV_##type; \
7258 SD_INIT_NAME(sd, type); \ 7258 SD_INIT_NAME(sd, type); \
7259 } 7259 }
7260 7260
7261 SD_INIT_FUNC(CPU) 7261 SD_INIT_FUNC(CPU)
7262 #ifdef CONFIG_NUMA 7262 #ifdef CONFIG_NUMA
7263 SD_INIT_FUNC(ALLNODES) 7263 SD_INIT_FUNC(ALLNODES)
7264 SD_INIT_FUNC(NODE) 7264 SD_INIT_FUNC(NODE)
7265 #endif 7265 #endif
7266 #ifdef CONFIG_SCHED_SMT 7266 #ifdef CONFIG_SCHED_SMT
7267 SD_INIT_FUNC(SIBLING) 7267 SD_INIT_FUNC(SIBLING)
7268 #endif 7268 #endif
7269 #ifdef CONFIG_SCHED_MC 7269 #ifdef CONFIG_SCHED_MC
7270 SD_INIT_FUNC(MC) 7270 SD_INIT_FUNC(MC)
7271 #endif 7271 #endif
7272 7272
7273 /* 7273 /*
7274 * To minimize stack usage kmalloc room for cpumasks and share the 7274 * To minimize stack usage kmalloc room for cpumasks and share the
7275 * space as the usage in build_sched_domains() dictates. Used only 7275 * space as the usage in build_sched_domains() dictates. Used only
7276 * if the amount of space is significant. 7276 * if the amount of space is significant.
7277 */ 7277 */
7278 struct allmasks { 7278 struct allmasks {
7279 cpumask_t tmpmask; /* make this one first */ 7279 cpumask_t tmpmask; /* make this one first */
7280 union { 7280 union {
7281 cpumask_t nodemask; 7281 cpumask_t nodemask;
7282 cpumask_t this_sibling_map; 7282 cpumask_t this_sibling_map;
7283 cpumask_t this_core_map; 7283 cpumask_t this_core_map;
7284 }; 7284 };
7285 cpumask_t send_covered; 7285 cpumask_t send_covered;
7286 7286
7287 #ifdef CONFIG_NUMA 7287 #ifdef CONFIG_NUMA
7288 cpumask_t domainspan; 7288 cpumask_t domainspan;
7289 cpumask_t covered; 7289 cpumask_t covered;
7290 cpumask_t notcovered; 7290 cpumask_t notcovered;
7291 #endif 7291 #endif
7292 }; 7292 };
7293 7293
7294 #if NR_CPUS > 128 7294 #if NR_CPUS > 128
7295 #define SCHED_CPUMASK_DECLARE(v) struct allmasks *v 7295 #define SCHED_CPUMASK_DECLARE(v) struct allmasks *v
7296 static inline void sched_cpumask_alloc(struct allmasks **masks) 7296 static inline void sched_cpumask_alloc(struct allmasks **masks)
7297 { 7297 {
7298 *masks = kmalloc(sizeof(**masks), GFP_KERNEL); 7298 *masks = kmalloc(sizeof(**masks), GFP_KERNEL);
7299 } 7299 }
7300 static inline void sched_cpumask_free(struct allmasks *masks) 7300 static inline void sched_cpumask_free(struct allmasks *masks)
7301 { 7301 {
7302 kfree(masks); 7302 kfree(masks);
7303 } 7303 }
7304 #else 7304 #else
7305 #define SCHED_CPUMASK_DECLARE(v) struct allmasks _v, *v = &_v 7305 #define SCHED_CPUMASK_DECLARE(v) struct allmasks _v, *v = &_v
7306 static inline void sched_cpumask_alloc(struct allmasks **masks) 7306 static inline void sched_cpumask_alloc(struct allmasks **masks)
7307 { } 7307 { }
7308 static inline void sched_cpumask_free(struct allmasks *masks) 7308 static inline void sched_cpumask_free(struct allmasks *masks)
7309 { } 7309 { }
7310 #endif 7310 #endif
7311 7311
7312 #define SCHED_CPUMASK_VAR(v, a) cpumask_t *v = (cpumask_t *) \ 7312 #define SCHED_CPUMASK_VAR(v, a) cpumask_t *v = (cpumask_t *) \
7313 ((unsigned long)(a) + offsetof(struct allmasks, v)) 7313 ((unsigned long)(a) + offsetof(struct allmasks, v))
7314 7314
7315 static int default_relax_domain_level = -1; 7315 static int default_relax_domain_level = -1;
7316 7316
7317 static int __init setup_relax_domain_level(char *str) 7317 static int __init setup_relax_domain_level(char *str)
7318 { 7318 {
7319 unsigned long val; 7319 unsigned long val;
7320 7320
7321 val = simple_strtoul(str, NULL, 0); 7321 val = simple_strtoul(str, NULL, 0);
7322 if (val < SD_LV_MAX) 7322 if (val < SD_LV_MAX)
7323 default_relax_domain_level = val; 7323 default_relax_domain_level = val;
7324 7324
7325 return 1; 7325 return 1;
7326 } 7326 }
7327 __setup("relax_domain_level=", setup_relax_domain_level); 7327 __setup("relax_domain_level=", setup_relax_domain_level);
7328 7328
7329 static void set_domain_attribute(struct sched_domain *sd, 7329 static void set_domain_attribute(struct sched_domain *sd,
7330 struct sched_domain_attr *attr) 7330 struct sched_domain_attr *attr)
7331 { 7331 {
7332 int request; 7332 int request;
7333 7333
7334 if (!attr || attr->relax_domain_level < 0) { 7334 if (!attr || attr->relax_domain_level < 0) {
7335 if (default_relax_domain_level < 0) 7335 if (default_relax_domain_level < 0)
7336 return; 7336 return;
7337 else 7337 else
7338 request = default_relax_domain_level; 7338 request = default_relax_domain_level;
7339 } else 7339 } else
7340 request = attr->relax_domain_level; 7340 request = attr->relax_domain_level;
7341 if (request < sd->level) { 7341 if (request < sd->level) {
7342 /* turn off idle balance on this domain */ 7342 /* turn off idle balance on this domain */
7343 sd->flags &= ~(SD_WAKE_IDLE|SD_BALANCE_NEWIDLE); 7343 sd->flags &= ~(SD_WAKE_IDLE|SD_BALANCE_NEWIDLE);
7344 } else { 7344 } else {
7345 /* turn on idle balance on this domain */ 7345 /* turn on idle balance on this domain */
7346 sd->flags |= (SD_WAKE_IDLE_FAR|SD_BALANCE_NEWIDLE); 7346 sd->flags |= (SD_WAKE_IDLE_FAR|SD_BALANCE_NEWIDLE);
7347 } 7347 }
7348 } 7348 }
7349 7349
7350 /* 7350 /*
7351 * Build sched domains for a given set of cpus and attach the sched domains 7351 * Build sched domains for a given set of cpus and attach the sched domains
7352 * to the individual cpus 7352 * to the individual cpus
7353 */ 7353 */
7354 static int __build_sched_domains(const cpumask_t *cpu_map, 7354 static int __build_sched_domains(const cpumask_t *cpu_map,
7355 struct sched_domain_attr *attr) 7355 struct sched_domain_attr *attr)
7356 { 7356 {
7357 int i; 7357 int i;
7358 struct root_domain *rd; 7358 struct root_domain *rd;
7359 SCHED_CPUMASK_DECLARE(allmasks); 7359 SCHED_CPUMASK_DECLARE(allmasks);
7360 cpumask_t *tmpmask; 7360 cpumask_t *tmpmask;
7361 #ifdef CONFIG_NUMA 7361 #ifdef CONFIG_NUMA
7362 struct sched_group **sched_group_nodes = NULL; 7362 struct sched_group **sched_group_nodes = NULL;
7363 int sd_allnodes = 0; 7363 int sd_allnodes = 0;
7364 7364
7365 /* 7365 /*
7366 * Allocate the per-node list of sched groups 7366 * Allocate the per-node list of sched groups
7367 */ 7367 */
7368 sched_group_nodes = kcalloc(nr_node_ids, sizeof(struct sched_group *), 7368 sched_group_nodes = kcalloc(nr_node_ids, sizeof(struct sched_group *),
7369 GFP_KERNEL); 7369 GFP_KERNEL);
7370 if (!sched_group_nodes) { 7370 if (!sched_group_nodes) {
7371 printk(KERN_WARNING "Can not alloc sched group node list\n"); 7371 printk(KERN_WARNING "Can not alloc sched group node list\n");
7372 return -ENOMEM; 7372 return -ENOMEM;
7373 } 7373 }
7374 #endif 7374 #endif
7375 7375
7376 rd = alloc_rootdomain(); 7376 rd = alloc_rootdomain();
7377 if (!rd) { 7377 if (!rd) {
7378 printk(KERN_WARNING "Cannot alloc root domain\n"); 7378 printk(KERN_WARNING "Cannot alloc root domain\n");
7379 #ifdef CONFIG_NUMA 7379 #ifdef CONFIG_NUMA
7380 kfree(sched_group_nodes); 7380 kfree(sched_group_nodes);
7381 #endif 7381 #endif
7382 return -ENOMEM; 7382 return -ENOMEM;
7383 } 7383 }
7384 7384
7385 /* get space for all scratch cpumask variables */ 7385 /* get space for all scratch cpumask variables */
7386 sched_cpumask_alloc(&allmasks); 7386 sched_cpumask_alloc(&allmasks);
7387 if (!allmasks) { 7387 if (!allmasks) {
7388 printk(KERN_WARNING "Cannot alloc cpumask array\n"); 7388 printk(KERN_WARNING "Cannot alloc cpumask array\n");
7389 kfree(rd); 7389 kfree(rd);
7390 #ifdef CONFIG_NUMA 7390 #ifdef CONFIG_NUMA
7391 kfree(sched_group_nodes); 7391 kfree(sched_group_nodes);
7392 #endif 7392 #endif
7393 return -ENOMEM; 7393 return -ENOMEM;
7394 } 7394 }
7395 7395
7396 tmpmask = (cpumask_t *)allmasks; 7396 tmpmask = (cpumask_t *)allmasks;
7397 7397
7398 7398
7399 #ifdef CONFIG_NUMA 7399 #ifdef CONFIG_NUMA
7400 sched_group_nodes_bycpu[first_cpu(*cpu_map)] = sched_group_nodes; 7400 sched_group_nodes_bycpu[first_cpu(*cpu_map)] = sched_group_nodes;
7401 #endif 7401 #endif
7402 7402
7403 /* 7403 /*
7404 * Set up domains for cpus specified by the cpu_map. 7404 * Set up domains for cpus specified by the cpu_map.
7405 */ 7405 */
7406 for_each_cpu_mask_nr(i, *cpu_map) { 7406 for_each_cpu_mask_nr(i, *cpu_map) {
7407 struct sched_domain *sd = NULL, *p; 7407 struct sched_domain *sd = NULL, *p;
7408 SCHED_CPUMASK_VAR(nodemask, allmasks); 7408 SCHED_CPUMASK_VAR(nodemask, allmasks);
7409 7409
7410 *nodemask = node_to_cpumask(cpu_to_node(i)); 7410 *nodemask = node_to_cpumask(cpu_to_node(i));
7411 cpus_and(*nodemask, *nodemask, *cpu_map); 7411 cpus_and(*nodemask, *nodemask, *cpu_map);
7412 7412
7413 #ifdef CONFIG_NUMA 7413 #ifdef CONFIG_NUMA
7414 if (cpus_weight(*cpu_map) > 7414 if (cpus_weight(*cpu_map) >
7415 SD_NODES_PER_DOMAIN*cpus_weight(*nodemask)) { 7415 SD_NODES_PER_DOMAIN*cpus_weight(*nodemask)) {
7416 sd = &per_cpu(allnodes_domains, i); 7416 sd = &per_cpu(allnodes_domains, i);
7417 SD_INIT(sd, ALLNODES); 7417 SD_INIT(sd, ALLNODES);
7418 set_domain_attribute(sd, attr); 7418 set_domain_attribute(sd, attr);
7419 sd->span = *cpu_map; 7419 sd->span = *cpu_map;
7420 cpu_to_allnodes_group(i, cpu_map, &sd->groups, tmpmask); 7420 cpu_to_allnodes_group(i, cpu_map, &sd->groups, tmpmask);
7421 p = sd; 7421 p = sd;
7422 sd_allnodes = 1; 7422 sd_allnodes = 1;
7423 } else 7423 } else
7424 p = NULL; 7424 p = NULL;
7425 7425
7426 sd = &per_cpu(node_domains, i); 7426 sd = &per_cpu(node_domains, i);
7427 SD_INIT(sd, NODE); 7427 SD_INIT(sd, NODE);
7428 set_domain_attribute(sd, attr); 7428 set_domain_attribute(sd, attr);
7429 sched_domain_node_span(cpu_to_node(i), &sd->span); 7429 sched_domain_node_span(cpu_to_node(i), &sd->span);
7430 sd->parent = p; 7430 sd->parent = p;
7431 if (p) 7431 if (p)
7432 p->child = sd; 7432 p->child = sd;
7433 cpus_and(sd->span, sd->span, *cpu_map); 7433 cpus_and(sd->span, sd->span, *cpu_map);
7434 #endif 7434 #endif
7435 7435
7436 p = sd; 7436 p = sd;
7437 sd = &per_cpu(phys_domains, i); 7437 sd = &per_cpu(phys_domains, i);
7438 SD_INIT(sd, CPU); 7438 SD_INIT(sd, CPU);
7439 set_domain_attribute(sd, attr); 7439 set_domain_attribute(sd, attr);
7440 sd->span = *nodemask; 7440 sd->span = *nodemask;
7441 sd->parent = p; 7441 sd->parent = p;
7442 if (p) 7442 if (p)
7443 p->child = sd; 7443 p->child = sd;
7444 cpu_to_phys_group(i, cpu_map, &sd->groups, tmpmask); 7444 cpu_to_phys_group(i, cpu_map, &sd->groups, tmpmask);
7445 7445
7446 #ifdef CONFIG_SCHED_MC 7446 #ifdef CONFIG_SCHED_MC
7447 p = sd; 7447 p = sd;
7448 sd = &per_cpu(core_domains, i); 7448 sd = &per_cpu(core_domains, i);
7449 SD_INIT(sd, MC); 7449 SD_INIT(sd, MC);
7450 set_domain_attribute(sd, attr); 7450 set_domain_attribute(sd, attr);
7451 sd->span = cpu_coregroup_map(i); 7451 sd->span = cpu_coregroup_map(i);
7452 cpus_and(sd->span, sd->span, *cpu_map); 7452 cpus_and(sd->span, sd->span, *cpu_map);
7453 sd->parent = p; 7453 sd->parent = p;
7454 p->child = sd; 7454 p->child = sd;
7455 cpu_to_core_group(i, cpu_map, &sd->groups, tmpmask); 7455 cpu_to_core_group(i, cpu_map, &sd->groups, tmpmask);
7456 #endif 7456 #endif
7457 7457
7458 #ifdef CONFIG_SCHED_SMT 7458 #ifdef CONFIG_SCHED_SMT
7459 p = sd; 7459 p = sd;
7460 sd = &per_cpu(cpu_domains, i); 7460 sd = &per_cpu(cpu_domains, i);
7461 SD_INIT(sd, SIBLING); 7461 SD_INIT(sd, SIBLING);
7462 set_domain_attribute(sd, attr); 7462 set_domain_attribute(sd, attr);
7463 sd->span = per_cpu(cpu_sibling_map, i); 7463 sd->span = per_cpu(cpu_sibling_map, i);
7464 cpus_and(sd->span, sd->span, *cpu_map); 7464 cpus_and(sd->span, sd->span, *cpu_map);
7465 sd->parent = p; 7465 sd->parent = p;
7466 p->child = sd; 7466 p->child = sd;
7467 cpu_to_cpu_group(i, cpu_map, &sd->groups, tmpmask); 7467 cpu_to_cpu_group(i, cpu_map, &sd->groups, tmpmask);
7468 #endif 7468 #endif
7469 } 7469 }
7470 7470
7471 #ifdef CONFIG_SCHED_SMT 7471 #ifdef CONFIG_SCHED_SMT
7472 /* Set up CPU (sibling) groups */ 7472 /* Set up CPU (sibling) groups */
7473 for_each_cpu_mask_nr(i, *cpu_map) { 7473 for_each_cpu_mask_nr(i, *cpu_map) {
7474 SCHED_CPUMASK_VAR(this_sibling_map, allmasks); 7474 SCHED_CPUMASK_VAR(this_sibling_map, allmasks);
7475 SCHED_CPUMASK_VAR(send_covered, allmasks); 7475 SCHED_CPUMASK_VAR(send_covered, allmasks);
7476 7476
7477 *this_sibling_map = per_cpu(cpu_sibling_map, i); 7477 *this_sibling_map = per_cpu(cpu_sibling_map, i);
7478 cpus_and(*this_sibling_map, *this_sibling_map, *cpu_map); 7478 cpus_and(*this_sibling_map, *this_sibling_map, *cpu_map);
7479 if (i != first_cpu(*this_sibling_map)) 7479 if (i != first_cpu(*this_sibling_map))
7480 continue; 7480 continue;
7481 7481
7482 init_sched_build_groups(this_sibling_map, cpu_map, 7482 init_sched_build_groups(this_sibling_map, cpu_map,
7483 &cpu_to_cpu_group, 7483 &cpu_to_cpu_group,
7484 send_covered, tmpmask); 7484 send_covered, tmpmask);
7485 } 7485 }
7486 #endif 7486 #endif
7487 7487
7488 #ifdef CONFIG_SCHED_MC 7488 #ifdef CONFIG_SCHED_MC
7489 /* Set up multi-core groups */ 7489 /* Set up multi-core groups */
7490 for_each_cpu_mask_nr(i, *cpu_map) { 7490 for_each_cpu_mask_nr(i, *cpu_map) {
7491 SCHED_CPUMASK_VAR(this_core_map, allmasks); 7491 SCHED_CPUMASK_VAR(this_core_map, allmasks);
7492 SCHED_CPUMASK_VAR(send_covered, allmasks); 7492 SCHED_CPUMASK_VAR(send_covered, allmasks);
7493 7493
7494 *this_core_map = cpu_coregroup_map(i); 7494 *this_core_map = cpu_coregroup_map(i);
7495 cpus_and(*this_core_map, *this_core_map, *cpu_map); 7495 cpus_and(*this_core_map, *this_core_map, *cpu_map);
7496 if (i != first_cpu(*this_core_map)) 7496 if (i != first_cpu(*this_core_map))
7497 continue; 7497 continue;
7498 7498
7499 init_sched_build_groups(this_core_map, cpu_map, 7499 init_sched_build_groups(this_core_map, cpu_map,
7500 &cpu_to_core_group, 7500 &cpu_to_core_group,
7501 send_covered, tmpmask); 7501 send_covered, tmpmask);
7502 } 7502 }
7503 #endif 7503 #endif
7504 7504
7505 /* Set up physical groups */ 7505 /* Set up physical groups */
7506 for (i = 0; i < nr_node_ids; i++) { 7506 for (i = 0; i < nr_node_ids; i++) {
7507 SCHED_CPUMASK_VAR(nodemask, allmasks); 7507 SCHED_CPUMASK_VAR(nodemask, allmasks);
7508 SCHED_CPUMASK_VAR(send_covered, allmasks); 7508 SCHED_CPUMASK_VAR(send_covered, allmasks);
7509 7509
7510 *nodemask = node_to_cpumask(i); 7510 *nodemask = node_to_cpumask(i);
7511 cpus_and(*nodemask, *nodemask, *cpu_map); 7511 cpus_and(*nodemask, *nodemask, *cpu_map);
7512 if (cpus_empty(*nodemask)) 7512 if (cpus_empty(*nodemask))
7513 continue; 7513 continue;
7514 7514
7515 init_sched_build_groups(nodemask, cpu_map, 7515 init_sched_build_groups(nodemask, cpu_map,
7516 &cpu_to_phys_group, 7516 &cpu_to_phys_group,
7517 send_covered, tmpmask); 7517 send_covered, tmpmask);
7518 } 7518 }
7519 7519
7520 #ifdef CONFIG_NUMA 7520 #ifdef CONFIG_NUMA
7521 /* Set up node groups */ 7521 /* Set up node groups */
7522 if (sd_allnodes) { 7522 if (sd_allnodes) {
7523 SCHED_CPUMASK_VAR(send_covered, allmasks); 7523 SCHED_CPUMASK_VAR(send_covered, allmasks);
7524 7524
7525 init_sched_build_groups(cpu_map, cpu_map, 7525 init_sched_build_groups(cpu_map, cpu_map,
7526 &cpu_to_allnodes_group, 7526 &cpu_to_allnodes_group,
7527 send_covered, tmpmask); 7527 send_covered, tmpmask);
7528 } 7528 }
7529 7529
7530 for (i = 0; i < nr_node_ids; i++) { 7530 for (i = 0; i < nr_node_ids; i++) {
7531 /* Set up node groups */ 7531 /* Set up node groups */
7532 struct sched_group *sg, *prev; 7532 struct sched_group *sg, *prev;
7533 SCHED_CPUMASK_VAR(nodemask, allmasks); 7533 SCHED_CPUMASK_VAR(nodemask, allmasks);
7534 SCHED_CPUMASK_VAR(domainspan, allmasks); 7534 SCHED_CPUMASK_VAR(domainspan, allmasks);
7535 SCHED_CPUMASK_VAR(covered, allmasks); 7535 SCHED_CPUMASK_VAR(covered, allmasks);
7536 int j; 7536 int j;
7537 7537
7538 *nodemask = node_to_cpumask(i); 7538 *nodemask = node_to_cpumask(i);
7539 cpus_clear(*covered); 7539 cpus_clear(*covered);
7540 7540
7541 cpus_and(*nodemask, *nodemask, *cpu_map); 7541 cpus_and(*nodemask, *nodemask, *cpu_map);
7542 if (cpus_empty(*nodemask)) { 7542 if (cpus_empty(*nodemask)) {
7543 sched_group_nodes[i] = NULL; 7543 sched_group_nodes[i] = NULL;
7544 continue; 7544 continue;
7545 } 7545 }
7546 7546
7547 sched_domain_node_span(i, domainspan); 7547 sched_domain_node_span(i, domainspan);
7548 cpus_and(*domainspan, *domainspan, *cpu_map); 7548 cpus_and(*domainspan, *domainspan, *cpu_map);
7549 7549
7550 sg = kmalloc_node(sizeof(struct sched_group), GFP_KERNEL, i); 7550 sg = kmalloc_node(sizeof(struct sched_group), GFP_KERNEL, i);
7551 if (!sg) { 7551 if (!sg) {
7552 printk(KERN_WARNING "Can not alloc domain group for " 7552 printk(KERN_WARNING "Can not alloc domain group for "
7553 "node %d\n", i); 7553 "node %d\n", i);
7554 goto error; 7554 goto error;
7555 } 7555 }
7556 sched_group_nodes[i] = sg; 7556 sched_group_nodes[i] = sg;
7557 for_each_cpu_mask_nr(j, *nodemask) { 7557 for_each_cpu_mask_nr(j, *nodemask) {
7558 struct sched_domain *sd; 7558 struct sched_domain *sd;
7559 7559
7560 sd = &per_cpu(node_domains, j); 7560 sd = &per_cpu(node_domains, j);
7561 sd->groups = sg; 7561 sd->groups = sg;
7562 } 7562 }
7563 sg->__cpu_power = 0; 7563 sg->__cpu_power = 0;
7564 sg->cpumask = *nodemask; 7564 sg->cpumask = *nodemask;
7565 sg->next = sg; 7565 sg->next = sg;
7566 cpus_or(*covered, *covered, *nodemask); 7566 cpus_or(*covered, *covered, *nodemask);
7567 prev = sg; 7567 prev = sg;
7568 7568
7569 for (j = 0; j < nr_node_ids; j++) { 7569 for (j = 0; j < nr_node_ids; j++) {
7570 SCHED_CPUMASK_VAR(notcovered, allmasks); 7570 SCHED_CPUMASK_VAR(notcovered, allmasks);
7571 int n = (i + j) % nr_node_ids; 7571 int n = (i + j) % nr_node_ids;
7572 node_to_cpumask_ptr(pnodemask, n); 7572 node_to_cpumask_ptr(pnodemask, n);
7573 7573
7574 cpus_complement(*notcovered, *covered); 7574 cpus_complement(*notcovered, *covered);
7575 cpus_and(*tmpmask, *notcovered, *cpu_map); 7575 cpus_and(*tmpmask, *notcovered, *cpu_map);
7576 cpus_and(*tmpmask, *tmpmask, *domainspan); 7576 cpus_and(*tmpmask, *tmpmask, *domainspan);
7577 if (cpus_empty(*tmpmask)) 7577 if (cpus_empty(*tmpmask))
7578 break; 7578 break;
7579 7579
7580 cpus_and(*tmpmask, *tmpmask, *pnodemask); 7580 cpus_and(*tmpmask, *tmpmask, *pnodemask);
7581 if (cpus_empty(*tmpmask)) 7581 if (cpus_empty(*tmpmask))
7582 continue; 7582 continue;
7583 7583
7584 sg = kmalloc_node(sizeof(struct sched_group), 7584 sg = kmalloc_node(sizeof(struct sched_group),
7585 GFP_KERNEL, i); 7585 GFP_KERNEL, i);
7586 if (!sg) { 7586 if (!sg) {
7587 printk(KERN_WARNING 7587 printk(KERN_WARNING
7588 "Can not alloc domain group for node %d\n", j); 7588 "Can not alloc domain group for node %d\n", j);
7589 goto error; 7589 goto error;
7590 } 7590 }
7591 sg->__cpu_power = 0; 7591 sg->__cpu_power = 0;
7592 sg->cpumask = *tmpmask; 7592 sg->cpumask = *tmpmask;
7593 sg->next = prev->next; 7593 sg->next = prev->next;
7594 cpus_or(*covered, *covered, *tmpmask); 7594 cpus_or(*covered, *covered, *tmpmask);
7595 prev->next = sg; 7595 prev->next = sg;
7596 prev = sg; 7596 prev = sg;
7597 } 7597 }
7598 } 7598 }
7599 #endif 7599 #endif
7600 7600
7601 /* Calculate CPU power for physical packages and nodes */ 7601 /* Calculate CPU power for physical packages and nodes */
7602 #ifdef CONFIG_SCHED_SMT 7602 #ifdef CONFIG_SCHED_SMT
7603 for_each_cpu_mask_nr(i, *cpu_map) { 7603 for_each_cpu_mask_nr(i, *cpu_map) {
7604 struct sched_domain *sd = &per_cpu(cpu_domains, i); 7604 struct sched_domain *sd = &per_cpu(cpu_domains, i);
7605 7605
7606 init_sched_groups_power(i, sd); 7606 init_sched_groups_power(i, sd);
7607 } 7607 }
7608 #endif 7608 #endif
7609 #ifdef CONFIG_SCHED_MC 7609 #ifdef CONFIG_SCHED_MC
7610 for_each_cpu_mask_nr(i, *cpu_map) { 7610 for_each_cpu_mask_nr(i, *cpu_map) {
7611 struct sched_domain *sd = &per_cpu(core_domains, i); 7611 struct sched_domain *sd = &per_cpu(core_domains, i);
7612 7612
7613 init_sched_groups_power(i, sd); 7613 init_sched_groups_power(i, sd);
7614 } 7614 }
7615 #endif 7615 #endif
7616 7616
7617 for_each_cpu_mask_nr(i, *cpu_map) { 7617 for_each_cpu_mask_nr(i, *cpu_map) {
7618 struct sched_domain *sd = &per_cpu(phys_domains, i); 7618 struct sched_domain *sd = &per_cpu(phys_domains, i);
7619 7619
7620 init_sched_groups_power(i, sd); 7620 init_sched_groups_power(i, sd);
7621 } 7621 }
7622 7622
7623 #ifdef CONFIG_NUMA 7623 #ifdef CONFIG_NUMA
7624 for (i = 0; i < nr_node_ids; i++) 7624 for (i = 0; i < nr_node_ids; i++)
7625 init_numa_sched_groups_power(sched_group_nodes[i]); 7625 init_numa_sched_groups_power(sched_group_nodes[i]);
7626 7626
7627 if (sd_allnodes) { 7627 if (sd_allnodes) {
7628 struct sched_group *sg; 7628 struct sched_group *sg;
7629 7629
7630 cpu_to_allnodes_group(first_cpu(*cpu_map), cpu_map, &sg, 7630 cpu_to_allnodes_group(first_cpu(*cpu_map), cpu_map, &sg,
7631 tmpmask); 7631 tmpmask);
7632 init_numa_sched_groups_power(sg); 7632 init_numa_sched_groups_power(sg);
7633 } 7633 }
7634 #endif 7634 #endif
7635 7635
7636 /* Attach the domains */ 7636 /* Attach the domains */
7637 for_each_cpu_mask_nr(i, *cpu_map) { 7637 for_each_cpu_mask_nr(i, *cpu_map) {
7638 struct sched_domain *sd; 7638 struct sched_domain *sd;
7639 #ifdef CONFIG_SCHED_SMT 7639 #ifdef CONFIG_SCHED_SMT
7640 sd = &per_cpu(cpu_domains, i); 7640 sd = &per_cpu(cpu_domains, i);
7641 #elif defined(CONFIG_SCHED_MC) 7641 #elif defined(CONFIG_SCHED_MC)
7642 sd = &per_cpu(core_domains, i); 7642 sd = &per_cpu(core_domains, i);
7643 #else 7643 #else
7644 sd = &per_cpu(phys_domains, i); 7644 sd = &per_cpu(phys_domains, i);
7645 #endif 7645 #endif
7646 cpu_attach_domain(sd, rd, i); 7646 cpu_attach_domain(sd, rd, i);
7647 } 7647 }
7648 7648
7649 sched_cpumask_free(allmasks); 7649 sched_cpumask_free(allmasks);
7650 return 0; 7650 return 0;
7651 7651
7652 #ifdef CONFIG_NUMA 7652 #ifdef CONFIG_NUMA
7653 error: 7653 error:
7654 free_sched_groups(cpu_map, tmpmask); 7654 free_sched_groups(cpu_map, tmpmask);
7655 sched_cpumask_free(allmasks); 7655 sched_cpumask_free(allmasks);
7656 kfree(rd); 7656 kfree(rd);
7657 return -ENOMEM; 7657 return -ENOMEM;
7658 #endif 7658 #endif
7659 } 7659 }
7660 7660
7661 static int build_sched_domains(const cpumask_t *cpu_map) 7661 static int build_sched_domains(const cpumask_t *cpu_map)
7662 { 7662 {
7663 return __build_sched_domains(cpu_map, NULL); 7663 return __build_sched_domains(cpu_map, NULL);
7664 } 7664 }
7665 7665
7666 static cpumask_t *doms_cur; /* current sched domains */ 7666 static cpumask_t *doms_cur; /* current sched domains */
7667 static int ndoms_cur; /* number of sched domains in 'doms_cur' */ 7667 static int ndoms_cur; /* number of sched domains in 'doms_cur' */
7668 static struct sched_domain_attr *dattr_cur; 7668 static struct sched_domain_attr *dattr_cur;
7669 /* attribues of custom domains in 'doms_cur' */ 7669 /* attribues of custom domains in 'doms_cur' */
7670 7670
7671 /* 7671 /*
7672 * Special case: If a kmalloc of a doms_cur partition (array of 7672 * Special case: If a kmalloc of a doms_cur partition (array of
7673 * cpumask_t) fails, then fallback to a single sched domain, 7673 * cpumask_t) fails, then fallback to a single sched domain,
7674 * as determined by the single cpumask_t fallback_doms. 7674 * as determined by the single cpumask_t fallback_doms.
7675 */ 7675 */
7676 static cpumask_t fallback_doms; 7676 static cpumask_t fallback_doms;
7677 7677
7678 void __attribute__((weak)) arch_update_cpu_topology(void) 7678 /*
7679 * arch_update_cpu_topology lets virtualized architectures update the
7680 * cpu core maps. It is supposed to return 1 if the topology changed
7681 * or 0 if it stayed the same.
7682 */
7683 int __attribute__((weak)) arch_update_cpu_topology(void)
7679 { 7684 {
7685 return 0;
7680 } 7686 }
7681 7687
7682 /* 7688 /*
7683 * Set up scheduler domains and groups. Callers must hold the hotplug lock. 7689 * Set up scheduler domains and groups. Callers must hold the hotplug lock.
7684 * For now this just excludes isolated cpus, but could be used to 7690 * For now this just excludes isolated cpus, but could be used to
7685 * exclude other special cases in the future. 7691 * exclude other special cases in the future.
7686 */ 7692 */
7687 static int arch_init_sched_domains(const cpumask_t *cpu_map) 7693 static int arch_init_sched_domains(const cpumask_t *cpu_map)
7688 { 7694 {
7689 int err; 7695 int err;
7690 7696
7691 arch_update_cpu_topology(); 7697 arch_update_cpu_topology();
7692 ndoms_cur = 1; 7698 ndoms_cur = 1;
7693 doms_cur = kmalloc(sizeof(cpumask_t), GFP_KERNEL); 7699 doms_cur = kmalloc(sizeof(cpumask_t), GFP_KERNEL);
7694 if (!doms_cur) 7700 if (!doms_cur)
7695 doms_cur = &fallback_doms; 7701 doms_cur = &fallback_doms;
7696 cpus_andnot(*doms_cur, *cpu_map, cpu_isolated_map); 7702 cpus_andnot(*doms_cur, *cpu_map, cpu_isolated_map);
7697 dattr_cur = NULL; 7703 dattr_cur = NULL;
7698 err = build_sched_domains(doms_cur); 7704 err = build_sched_domains(doms_cur);
7699 register_sched_domain_sysctl(); 7705 register_sched_domain_sysctl();
7700 7706
7701 return err; 7707 return err;
7702 } 7708 }
7703 7709
7704 static void arch_destroy_sched_domains(const cpumask_t *cpu_map, 7710 static void arch_destroy_sched_domains(const cpumask_t *cpu_map,
7705 cpumask_t *tmpmask) 7711 cpumask_t *tmpmask)
7706 { 7712 {
7707 free_sched_groups(cpu_map, tmpmask); 7713 free_sched_groups(cpu_map, tmpmask);
7708 } 7714 }
7709 7715
7710 /* 7716 /*
7711 * Detach sched domains from a group of cpus specified in cpu_map 7717 * Detach sched domains from a group of cpus specified in cpu_map
7712 * These cpus will now be attached to the NULL domain 7718 * These cpus will now be attached to the NULL domain
7713 */ 7719 */
7714 static void detach_destroy_domains(const cpumask_t *cpu_map) 7720 static void detach_destroy_domains(const cpumask_t *cpu_map)
7715 { 7721 {
7716 cpumask_t tmpmask; 7722 cpumask_t tmpmask;
7717 int i; 7723 int i;
7718 7724
7719 for_each_cpu_mask_nr(i, *cpu_map) 7725 for_each_cpu_mask_nr(i, *cpu_map)
7720 cpu_attach_domain(NULL, &def_root_domain, i); 7726 cpu_attach_domain(NULL, &def_root_domain, i);
7721 synchronize_sched(); 7727 synchronize_sched();
7722 arch_destroy_sched_domains(cpu_map, &tmpmask); 7728 arch_destroy_sched_domains(cpu_map, &tmpmask);
7723 } 7729 }
7724 7730
7725 /* handle null as "default" */ 7731 /* handle null as "default" */
7726 static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, 7732 static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur,
7727 struct sched_domain_attr *new, int idx_new) 7733 struct sched_domain_attr *new, int idx_new)
7728 { 7734 {
7729 struct sched_domain_attr tmp; 7735 struct sched_domain_attr tmp;
7730 7736
7731 /* fast path */ 7737 /* fast path */
7732 if (!new && !cur) 7738 if (!new && !cur)
7733 return 1; 7739 return 1;
7734 7740
7735 tmp = SD_ATTR_INIT; 7741 tmp = SD_ATTR_INIT;
7736 return !memcmp(cur ? (cur + idx_cur) : &tmp, 7742 return !memcmp(cur ? (cur + idx_cur) : &tmp,
7737 new ? (new + idx_new) : &tmp, 7743 new ? (new + idx_new) : &tmp,
7738 sizeof(struct sched_domain_attr)); 7744 sizeof(struct sched_domain_attr));
7739 } 7745 }
7740 7746
7741 /* 7747 /*
7742 * Partition sched domains as specified by the 'ndoms_new' 7748 * Partition sched domains as specified by the 'ndoms_new'
7743 * cpumasks in the array doms_new[] of cpumasks. This compares 7749 * cpumasks in the array doms_new[] of cpumasks. This compares
7744 * doms_new[] to the current sched domain partitioning, doms_cur[]. 7750 * doms_new[] to the current sched domain partitioning, doms_cur[].
7745 * It destroys each deleted domain and builds each new domain. 7751 * It destroys each deleted domain and builds each new domain.
7746 * 7752 *
7747 * 'doms_new' is an array of cpumask_t's of length 'ndoms_new'. 7753 * 'doms_new' is an array of cpumask_t's of length 'ndoms_new'.
7748 * The masks don't intersect (don't overlap.) We should setup one 7754 * The masks don't intersect (don't overlap.) We should setup one
7749 * sched domain for each mask. CPUs not in any of the cpumasks will 7755 * sched domain for each mask. CPUs not in any of the cpumasks will
7750 * not be load balanced. If the same cpumask appears both in the 7756 * not be load balanced. If the same cpumask appears both in the
7751 * current 'doms_cur' domains and in the new 'doms_new', we can leave 7757 * current 'doms_cur' domains and in the new 'doms_new', we can leave
7752 * it as it is. 7758 * it as it is.
7753 * 7759 *
7754 * The passed in 'doms_new' should be kmalloc'd. This routine takes 7760 * The passed in 'doms_new' should be kmalloc'd. This routine takes
7755 * ownership of it and will kfree it when done with it. If the caller 7761 * ownership of it and will kfree it when done with it. If the caller
7756 * failed the kmalloc call, then it can pass in doms_new == NULL && 7762 * failed the kmalloc call, then it can pass in doms_new == NULL &&
7757 * ndoms_new == 1, and partition_sched_domains() will fallback to 7763 * ndoms_new == 1, and partition_sched_domains() will fallback to
7758 * the single partition 'fallback_doms', it also forces the domains 7764 * the single partition 'fallback_doms', it also forces the domains
7759 * to be rebuilt. 7765 * to be rebuilt.
7760 * 7766 *
7761 * If doms_new == NULL it will be replaced with cpu_online_map. 7767 * If doms_new == NULL it will be replaced with cpu_online_map.
7762 * ndoms_new == 0 is a special case for destroying existing domains, 7768 * ndoms_new == 0 is a special case for destroying existing domains,
7763 * and it will not create the default domain. 7769 * and it will not create the default domain.
7764 * 7770 *
7765 * Call with hotplug lock held 7771 * Call with hotplug lock held
7766 */ 7772 */
7767 void partition_sched_domains(int ndoms_new, cpumask_t *doms_new, 7773 void partition_sched_domains(int ndoms_new, cpumask_t *doms_new,
7768 struct sched_domain_attr *dattr_new) 7774 struct sched_domain_attr *dattr_new)
7769 { 7775 {
7770 int i, j, n; 7776 int i, j, n;
7771 7777
7772 mutex_lock(&sched_domains_mutex); 7778 mutex_lock(&sched_domains_mutex);
7773 7779
7774 /* always unregister in case we don't destroy any domains */ 7780 /* always unregister in case we don't destroy any domains */
7775 unregister_sched_domain_sysctl(); 7781 unregister_sched_domain_sysctl();
7776 7782
7777 n = doms_new ? ndoms_new : 0; 7783 n = doms_new ? ndoms_new : 0;
7778 7784
7779 /* Destroy deleted domains */ 7785 /* Destroy deleted domains */
7780 for (i = 0; i < ndoms_cur; i++) { 7786 for (i = 0; i < ndoms_cur; i++) {
7781 for (j = 0; j < n; j++) { 7787 for (j = 0; j < n; j++) {
7782 if (cpus_equal(doms_cur[i], doms_new[j]) 7788 if (cpus_equal(doms_cur[i], doms_new[j])
7783 && dattrs_equal(dattr_cur, i, dattr_new, j)) 7789 && dattrs_equal(dattr_cur, i, dattr_new, j))
7784 goto match1; 7790 goto match1;
7785 } 7791 }
7786 /* no match - a current sched domain not in new doms_new[] */ 7792 /* no match - a current sched domain not in new doms_new[] */
7787 detach_destroy_domains(doms_cur + i); 7793 detach_destroy_domains(doms_cur + i);
7788 match1: 7794 match1:
7789 ; 7795 ;
7790 } 7796 }
7791 7797
7792 if (doms_new == NULL) { 7798 if (doms_new == NULL) {
7793 ndoms_cur = 0; 7799 ndoms_cur = 0;
7794 doms_new = &fallback_doms; 7800 doms_new = &fallback_doms;
7795 cpus_andnot(doms_new[0], cpu_online_map, cpu_isolated_map); 7801 cpus_andnot(doms_new[0], cpu_online_map, cpu_isolated_map);
7796 WARN_ON_ONCE(dattr_new); 7802 WARN_ON_ONCE(dattr_new);
7797 } 7803 }
7798 7804
7799 /* Build new domains */ 7805 /* Build new domains */
7800 for (i = 0; i < ndoms_new; i++) { 7806 for (i = 0; i < ndoms_new; i++) {
7801 for (j = 0; j < ndoms_cur; j++) { 7807 for (j = 0; j < ndoms_cur; j++) {
7802 if (cpus_equal(doms_new[i], doms_cur[j]) 7808 if (cpus_equal(doms_new[i], doms_cur[j])
7803 && dattrs_equal(dattr_new, i, dattr_cur, j)) 7809 && dattrs_equal(dattr_new, i, dattr_cur, j))
7804 goto match2; 7810 goto match2;
7805 } 7811 }
7806 /* no match - add a new doms_new */ 7812 /* no match - add a new doms_new */
7807 __build_sched_domains(doms_new + i, 7813 __build_sched_domains(doms_new + i,
7808 dattr_new ? dattr_new + i : NULL); 7814 dattr_new ? dattr_new + i : NULL);
7809 match2: 7815 match2:
7810 ; 7816 ;
7811 } 7817 }
7812 7818
7813 /* Remember the new sched domains */ 7819 /* Remember the new sched domains */
7814 if (doms_cur != &fallback_doms) 7820 if (doms_cur != &fallback_doms)
7815 kfree(doms_cur); 7821 kfree(doms_cur);
7816 kfree(dattr_cur); /* kfree(NULL) is safe */ 7822 kfree(dattr_cur); /* kfree(NULL) is safe */
7817 doms_cur = doms_new; 7823 doms_cur = doms_new;
7818 dattr_cur = dattr_new; 7824 dattr_cur = dattr_new;
7819 ndoms_cur = ndoms_new; 7825 ndoms_cur = ndoms_new;
7820 7826
7821 register_sched_domain_sysctl(); 7827 register_sched_domain_sysctl();
7822 7828
7823 mutex_unlock(&sched_domains_mutex); 7829 mutex_unlock(&sched_domains_mutex);
7824 } 7830 }
7825 7831
7826 #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) 7832 #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT)
7827 int arch_reinit_sched_domains(void) 7833 int arch_reinit_sched_domains(void)
7828 { 7834 {
7829 get_online_cpus(); 7835 get_online_cpus();
7830 7836
7831 /* Destroy domains first to force the rebuild */ 7837 /* Destroy domains first to force the rebuild */
7832 partition_sched_domains(0, NULL, NULL); 7838 partition_sched_domains(0, NULL, NULL);
7833 7839
7834 rebuild_sched_domains(); 7840 rebuild_sched_domains();
7835 put_online_cpus(); 7841 put_online_cpus();
7836 7842
7837 return 0; 7843 return 0;
7838 } 7844 }
7839 7845
7840 static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) 7846 static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt)
7841 { 7847 {
7842 int ret; 7848 int ret;
7843 7849
7844 if (buf[0] != '0' && buf[0] != '1') 7850 if (buf[0] != '0' && buf[0] != '1')
7845 return -EINVAL; 7851 return -EINVAL;
7846 7852
7847 if (smt) 7853 if (smt)
7848 sched_smt_power_savings = (buf[0] == '1'); 7854 sched_smt_power_savings = (buf[0] == '1');
7849 else 7855 else
7850 sched_mc_power_savings = (buf[0] == '1'); 7856 sched_mc_power_savings = (buf[0] == '1');
7851 7857
7852 ret = arch_reinit_sched_domains(); 7858 ret = arch_reinit_sched_domains();
7853 7859
7854 return ret ? ret : count; 7860 return ret ? ret : count;
7855 } 7861 }
7856 7862
7857 #ifdef CONFIG_SCHED_MC 7863 #ifdef CONFIG_SCHED_MC
7858 static ssize_t sched_mc_power_savings_show(struct sysdev_class *class, 7864 static ssize_t sched_mc_power_savings_show(struct sysdev_class *class,
7859 char *page) 7865 char *page)
7860 { 7866 {
7861 return sprintf(page, "%u\n", sched_mc_power_savings); 7867 return sprintf(page, "%u\n", sched_mc_power_savings);
7862 } 7868 }
7863 static ssize_t sched_mc_power_savings_store(struct sysdev_class *class, 7869 static ssize_t sched_mc_power_savings_store(struct sysdev_class *class,
7864 const char *buf, size_t count) 7870 const char *buf, size_t count)
7865 { 7871 {
7866 return sched_power_savings_store(buf, count, 0); 7872 return sched_power_savings_store(buf, count, 0);
7867 } 7873 }
7868 static SYSDEV_CLASS_ATTR(sched_mc_power_savings, 0644, 7874 static SYSDEV_CLASS_ATTR(sched_mc_power_savings, 0644,
7869 sched_mc_power_savings_show, 7875 sched_mc_power_savings_show,
7870 sched_mc_power_savings_store); 7876 sched_mc_power_savings_store);
7871 #endif 7877 #endif
7872 7878
7873 #ifdef CONFIG_SCHED_SMT 7879 #ifdef CONFIG_SCHED_SMT
7874 static ssize_t sched_smt_power_savings_show(struct sysdev_class *dev, 7880 static ssize_t sched_smt_power_savings_show(struct sysdev_class *dev,
7875 char *page) 7881 char *page)
7876 { 7882 {
7877 return sprintf(page, "%u\n", sched_smt_power_savings); 7883 return sprintf(page, "%u\n", sched_smt_power_savings);
7878 } 7884 }
7879 static ssize_t sched_smt_power_savings_store(struct sysdev_class *dev, 7885 static ssize_t sched_smt_power_savings_store(struct sysdev_class *dev,
7880 const char *buf, size_t count) 7886 const char *buf, size_t count)
7881 { 7887 {
7882 return sched_power_savings_store(buf, count, 1); 7888 return sched_power_savings_store(buf, count, 1);
7883 } 7889 }
7884 static SYSDEV_CLASS_ATTR(sched_smt_power_savings, 0644, 7890 static SYSDEV_CLASS_ATTR(sched_smt_power_savings, 0644,
7885 sched_smt_power_savings_show, 7891 sched_smt_power_savings_show,
7886 sched_smt_power_savings_store); 7892 sched_smt_power_savings_store);
7887 #endif 7893 #endif
7888 7894
7889 int sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) 7895 int sched_create_sysfs_power_savings_entries(struct sysdev_class *cls)
7890 { 7896 {
7891 int err = 0; 7897 int err = 0;
7892 7898
7893 #ifdef CONFIG_SCHED_SMT 7899 #ifdef CONFIG_SCHED_SMT
7894 if (smt_capable()) 7900 if (smt_capable())
7895 err = sysfs_create_file(&cls->kset.kobj, 7901 err = sysfs_create_file(&cls->kset.kobj,
7896 &attr_sched_smt_power_savings.attr); 7902 &attr_sched_smt_power_savings.attr);
7897 #endif 7903 #endif
7898 #ifdef CONFIG_SCHED_MC 7904 #ifdef CONFIG_SCHED_MC
7899 if (!err && mc_capable()) 7905 if (!err && mc_capable())
7900 err = sysfs_create_file(&cls->kset.kobj, 7906 err = sysfs_create_file(&cls->kset.kobj,
7901 &attr_sched_mc_power_savings.attr); 7907 &attr_sched_mc_power_savings.attr);
7902 #endif 7908 #endif
7903 return err; 7909 return err;
7904 } 7910 }
7905 #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ 7911 #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */
7906 7912
7907 #ifndef CONFIG_CPUSETS 7913 #ifndef CONFIG_CPUSETS
7908 /* 7914 /*
7909 * Add online and remove offline CPUs from the scheduler domains. 7915 * Add online and remove offline CPUs from the scheduler domains.
7910 * When cpusets are enabled they take over this function. 7916 * When cpusets are enabled they take over this function.
7911 */ 7917 */
7912 static int update_sched_domains(struct notifier_block *nfb, 7918 static int update_sched_domains(struct notifier_block *nfb,
7913 unsigned long action, void *hcpu) 7919 unsigned long action, void *hcpu)
7914 { 7920 {
7915 switch (action) { 7921 switch (action) {
7916 case CPU_ONLINE: 7922 case CPU_ONLINE:
7917 case CPU_ONLINE_FROZEN: 7923 case CPU_ONLINE_FROZEN:
7918 case CPU_DEAD: 7924 case CPU_DEAD:
7919 case CPU_DEAD_FROZEN: 7925 case CPU_DEAD_FROZEN:
7920 partition_sched_domains(1, NULL, NULL); 7926 partition_sched_domains(1, NULL, NULL);
7921 return NOTIFY_OK; 7927 return NOTIFY_OK;
7922 7928
7923 default: 7929 default:
7924 return NOTIFY_DONE; 7930 return NOTIFY_DONE;
7925 } 7931 }
7926 } 7932 }
7927 #endif 7933 #endif
7928 7934
7929 static int update_runtime(struct notifier_block *nfb, 7935 static int update_runtime(struct notifier_block *nfb,
7930 unsigned long action, void *hcpu) 7936 unsigned long action, void *hcpu)
7931 { 7937 {
7932 int cpu = (int)(long)hcpu; 7938 int cpu = (int)(long)hcpu;
7933 7939
7934 switch (action) { 7940 switch (action) {
7935 case CPU_DOWN_PREPARE: 7941 case CPU_DOWN_PREPARE:
7936 case CPU_DOWN_PREPARE_FROZEN: 7942 case CPU_DOWN_PREPARE_FROZEN:
7937 disable_runtime(cpu_rq(cpu)); 7943 disable_runtime(cpu_rq(cpu));
7938 return NOTIFY_OK; 7944 return NOTIFY_OK;
7939 7945
7940 case CPU_DOWN_FAILED: 7946 case CPU_DOWN_FAILED:
7941 case CPU_DOWN_FAILED_FROZEN: 7947 case CPU_DOWN_FAILED_FROZEN:
7942 case CPU_ONLINE: 7948 case CPU_ONLINE:
7943 case CPU_ONLINE_FROZEN: 7949 case CPU_ONLINE_FROZEN:
7944 enable_runtime(cpu_rq(cpu)); 7950 enable_runtime(cpu_rq(cpu));
7945 return NOTIFY_OK; 7951 return NOTIFY_OK;
7946 7952
7947 default: 7953 default:
7948 return NOTIFY_DONE; 7954 return NOTIFY_DONE;
7949 } 7955 }
7950 } 7956 }
7951 7957
7952 void __init sched_init_smp(void) 7958 void __init sched_init_smp(void)
7953 { 7959 {
7954 cpumask_t non_isolated_cpus; 7960 cpumask_t non_isolated_cpus;
7955 7961
7956 #if defined(CONFIG_NUMA) 7962 #if defined(CONFIG_NUMA)
7957 sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **), 7963 sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **),
7958 GFP_KERNEL); 7964 GFP_KERNEL);
7959 BUG_ON(sched_group_nodes_bycpu == NULL); 7965 BUG_ON(sched_group_nodes_bycpu == NULL);
7960 #endif 7966 #endif
7961 get_online_cpus(); 7967 get_online_cpus();
7962 mutex_lock(&sched_domains_mutex); 7968 mutex_lock(&sched_domains_mutex);
7963 arch_init_sched_domains(&cpu_online_map); 7969 arch_init_sched_domains(&cpu_online_map);
7964 cpus_andnot(non_isolated_cpus, cpu_possible_map, cpu_isolated_map); 7970 cpus_andnot(non_isolated_cpus, cpu_possible_map, cpu_isolated_map);
7965 if (cpus_empty(non_isolated_cpus)) 7971 if (cpus_empty(non_isolated_cpus))
7966 cpu_set(smp_processor_id(), non_isolated_cpus); 7972 cpu_set(smp_processor_id(), non_isolated_cpus);
7967 mutex_unlock(&sched_domains_mutex); 7973 mutex_unlock(&sched_domains_mutex);
7968 put_online_cpus(); 7974 put_online_cpus();
7969 7975
7970 #ifndef CONFIG_CPUSETS 7976 #ifndef CONFIG_CPUSETS
7971 /* XXX: Theoretical race here - CPU may be hotplugged now */ 7977 /* XXX: Theoretical race here - CPU may be hotplugged now */
7972 hotcpu_notifier(update_sched_domains, 0); 7978 hotcpu_notifier(update_sched_domains, 0);
7973 #endif 7979 #endif
7974 7980
7975 /* RT runtime code needs to handle some hotplug events */ 7981 /* RT runtime code needs to handle some hotplug events */
7976 hotcpu_notifier(update_runtime, 0); 7982 hotcpu_notifier(update_runtime, 0);
7977 7983
7978 init_hrtick(); 7984 init_hrtick();
7979 7985
7980 /* Move init over to a non-isolated CPU */ 7986 /* Move init over to a non-isolated CPU */
7981 if (set_cpus_allowed_ptr(current, &non_isolated_cpus) < 0) 7987 if (set_cpus_allowed_ptr(current, &non_isolated_cpus) < 0)
7982 BUG(); 7988 BUG();
7983 sched_init_granularity(); 7989 sched_init_granularity();
7984 } 7990 }
7985 #else 7991 #else
7986 void __init sched_init_smp(void) 7992 void __init sched_init_smp(void)
7987 { 7993 {
7988 sched_init_granularity(); 7994 sched_init_granularity();
7989 } 7995 }
7990 #endif /* CONFIG_SMP */ 7996 #endif /* CONFIG_SMP */
7991 7997
7992 int in_sched_functions(unsigned long addr) 7998 int in_sched_functions(unsigned long addr)
7993 { 7999 {
7994 return in_lock_functions(addr) || 8000 return in_lock_functions(addr) ||
7995 (addr >= (unsigned long)__sched_text_start 8001 (addr >= (unsigned long)__sched_text_start
7996 && addr < (unsigned long)__sched_text_end); 8002 && addr < (unsigned long)__sched_text_end);
7997 } 8003 }
7998 8004
7999 static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) 8005 static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq)
8000 { 8006 {
8001 cfs_rq->tasks_timeline = RB_ROOT; 8007 cfs_rq->tasks_timeline = RB_ROOT;
8002 INIT_LIST_HEAD(&cfs_rq->tasks); 8008 INIT_LIST_HEAD(&cfs_rq->tasks);
8003 #ifdef CONFIG_FAIR_GROUP_SCHED 8009 #ifdef CONFIG_FAIR_GROUP_SCHED
8004 cfs_rq->rq = rq; 8010 cfs_rq->rq = rq;
8005 #endif 8011 #endif
8006 cfs_rq->min_vruntime = (u64)(-(1LL << 20)); 8012 cfs_rq->min_vruntime = (u64)(-(1LL << 20));
8007 } 8013 }
8008 8014
8009 static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq) 8015 static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq)
8010 { 8016 {
8011 struct rt_prio_array *array; 8017 struct rt_prio_array *array;
8012 int i; 8018 int i;
8013 8019
8014 array = &rt_rq->active; 8020 array = &rt_rq->active;
8015 for (i = 0; i < MAX_RT_PRIO; i++) { 8021 for (i = 0; i < MAX_RT_PRIO; i++) {
8016 INIT_LIST_HEAD(array->queue + i); 8022 INIT_LIST_HEAD(array->queue + i);
8017 __clear_bit(i, array->bitmap); 8023 __clear_bit(i, array->bitmap);
8018 } 8024 }
8019 /* delimiter for bitsearch: */ 8025 /* delimiter for bitsearch: */
8020 __set_bit(MAX_RT_PRIO, array->bitmap); 8026 __set_bit(MAX_RT_PRIO, array->bitmap);
8021 8027
8022 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED 8028 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
8023 rt_rq->highest_prio = MAX_RT_PRIO; 8029 rt_rq->highest_prio = MAX_RT_PRIO;
8024 #endif 8030 #endif
8025 #ifdef CONFIG_SMP 8031 #ifdef CONFIG_SMP
8026 rt_rq->rt_nr_migratory = 0; 8032 rt_rq->rt_nr_migratory = 0;
8027 rt_rq->overloaded = 0; 8033 rt_rq->overloaded = 0;
8028 #endif 8034 #endif
8029 8035
8030 rt_rq->rt_time = 0; 8036 rt_rq->rt_time = 0;
8031 rt_rq->rt_throttled = 0; 8037 rt_rq->rt_throttled = 0;
8032 rt_rq->rt_runtime = 0; 8038 rt_rq->rt_runtime = 0;
8033 spin_lock_init(&rt_rq->rt_runtime_lock); 8039 spin_lock_init(&rt_rq->rt_runtime_lock);
8034 8040
8035 #ifdef CONFIG_RT_GROUP_SCHED 8041 #ifdef CONFIG_RT_GROUP_SCHED
8036 rt_rq->rt_nr_boosted = 0; 8042 rt_rq->rt_nr_boosted = 0;
8037 rt_rq->rq = rq; 8043 rt_rq->rq = rq;
8038 #endif 8044 #endif
8039 } 8045 }
8040 8046
8041 #ifdef CONFIG_FAIR_GROUP_SCHED 8047 #ifdef CONFIG_FAIR_GROUP_SCHED
8042 static void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, 8048 static void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
8043 struct sched_entity *se, int cpu, int add, 8049 struct sched_entity *se, int cpu, int add,
8044 struct sched_entity *parent) 8050 struct sched_entity *parent)
8045 { 8051 {
8046 struct rq *rq = cpu_rq(cpu); 8052 struct rq *rq = cpu_rq(cpu);
8047 tg->cfs_rq[cpu] = cfs_rq; 8053 tg->cfs_rq[cpu] = cfs_rq;
8048 init_cfs_rq(cfs_rq, rq); 8054 init_cfs_rq(cfs_rq, rq);
8049 cfs_rq->tg = tg; 8055 cfs_rq->tg = tg;
8050 if (add) 8056 if (add)
8051 list_add(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); 8057 list_add(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list);
8052 8058
8053 tg->se[cpu] = se; 8059 tg->se[cpu] = se;
8054 /* se could be NULL for init_task_group */ 8060 /* se could be NULL for init_task_group */
8055 if (!se) 8061 if (!se)
8056 return; 8062 return;
8057 8063
8058 if (!parent) 8064 if (!parent)
8059 se->cfs_rq = &rq->cfs; 8065 se->cfs_rq = &rq->cfs;
8060 else 8066 else
8061 se->cfs_rq = parent->my_q; 8067 se->cfs_rq = parent->my_q;
8062 8068
8063 se->my_q = cfs_rq; 8069 se->my_q = cfs_rq;
8064 se->load.weight = tg->shares; 8070 se->load.weight = tg->shares;
8065 se->load.inv_weight = 0; 8071 se->load.inv_weight = 0;
8066 se->parent = parent; 8072 se->parent = parent;
8067 } 8073 }
8068 #endif 8074 #endif
8069 8075
8070 #ifdef CONFIG_RT_GROUP_SCHED 8076 #ifdef CONFIG_RT_GROUP_SCHED
8071 static void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, 8077 static void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
8072 struct sched_rt_entity *rt_se, int cpu, int add, 8078 struct sched_rt_entity *rt_se, int cpu, int add,
8073 struct sched_rt_entity *parent) 8079 struct sched_rt_entity *parent)
8074 { 8080 {
8075 struct rq *rq = cpu_rq(cpu); 8081 struct rq *rq = cpu_rq(cpu);
8076 8082
8077 tg->rt_rq[cpu] = rt_rq; 8083 tg->rt_rq[cpu] = rt_rq;
8078 init_rt_rq(rt_rq, rq); 8084 init_rt_rq(rt_rq, rq);
8079 rt_rq->tg = tg; 8085 rt_rq->tg = tg;
8080 rt_rq->rt_se = rt_se; 8086 rt_rq->rt_se = rt_se;
8081 rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime; 8087 rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime;
8082 if (add) 8088 if (add)
8083 list_add(&rt_rq->leaf_rt_rq_list, &rq->leaf_rt_rq_list); 8089 list_add(&rt_rq->leaf_rt_rq_list, &rq->leaf_rt_rq_list);
8084 8090
8085 tg->rt_se[cpu] = rt_se; 8091 tg->rt_se[cpu] = rt_se;
8086 if (!rt_se) 8092 if (!rt_se)
8087 return; 8093 return;
8088 8094
8089 if (!parent) 8095 if (!parent)
8090 rt_se->rt_rq = &rq->rt; 8096 rt_se->rt_rq = &rq->rt;
8091 else 8097 else
8092 rt_se->rt_rq = parent->my_q; 8098 rt_se->rt_rq = parent->my_q;
8093 8099
8094 rt_se->my_q = rt_rq; 8100 rt_se->my_q = rt_rq;
8095 rt_se->parent = parent; 8101 rt_se->parent = parent;
8096 INIT_LIST_HEAD(&rt_se->run_list); 8102 INIT_LIST_HEAD(&rt_se->run_list);
8097 } 8103 }
8098 #endif 8104 #endif
8099 8105
8100 void __init sched_init(void) 8106 void __init sched_init(void)
8101 { 8107 {
8102 int i, j; 8108 int i, j;
8103 unsigned long alloc_size = 0, ptr; 8109 unsigned long alloc_size = 0, ptr;
8104 8110
8105 #ifdef CONFIG_FAIR_GROUP_SCHED 8111 #ifdef CONFIG_FAIR_GROUP_SCHED
8106 alloc_size += 2 * nr_cpu_ids * sizeof(void **); 8112 alloc_size += 2 * nr_cpu_ids * sizeof(void **);
8107 #endif 8113 #endif
8108 #ifdef CONFIG_RT_GROUP_SCHED 8114 #ifdef CONFIG_RT_GROUP_SCHED
8109 alloc_size += 2 * nr_cpu_ids * sizeof(void **); 8115 alloc_size += 2 * nr_cpu_ids * sizeof(void **);
8110 #endif 8116 #endif
8111 #ifdef CONFIG_USER_SCHED 8117 #ifdef CONFIG_USER_SCHED
8112 alloc_size *= 2; 8118 alloc_size *= 2;
8113 #endif 8119 #endif
8114 /* 8120 /*
8115 * As sched_init() is called before page_alloc is setup, 8121 * As sched_init() is called before page_alloc is setup,
8116 * we use alloc_bootmem(). 8122 * we use alloc_bootmem().
8117 */ 8123 */
8118 if (alloc_size) { 8124 if (alloc_size) {
8119 ptr = (unsigned long)alloc_bootmem(alloc_size); 8125 ptr = (unsigned long)alloc_bootmem(alloc_size);
8120 8126
8121 #ifdef CONFIG_FAIR_GROUP_SCHED 8127 #ifdef CONFIG_FAIR_GROUP_SCHED
8122 init_task_group.se = (struct sched_entity **)ptr; 8128 init_task_group.se = (struct sched_entity **)ptr;
8123 ptr += nr_cpu_ids * sizeof(void **); 8129 ptr += nr_cpu_ids * sizeof(void **);
8124 8130
8125 init_task_group.cfs_rq = (struct cfs_rq **)ptr; 8131 init_task_group.cfs_rq = (struct cfs_rq **)ptr;
8126 ptr += nr_cpu_ids * sizeof(void **); 8132 ptr += nr_cpu_ids * sizeof(void **);
8127 8133
8128 #ifdef CONFIG_USER_SCHED 8134 #ifdef CONFIG_USER_SCHED
8129 root_task_group.se = (struct sched_entity **)ptr; 8135 root_task_group.se = (struct sched_entity **)ptr;
8130 ptr += nr_cpu_ids * sizeof(void **); 8136 ptr += nr_cpu_ids * sizeof(void **);
8131 8137
8132 root_task_group.cfs_rq = (struct cfs_rq **)ptr; 8138 root_task_group.cfs_rq = (struct cfs_rq **)ptr;
8133 ptr += nr_cpu_ids * sizeof(void **); 8139 ptr += nr_cpu_ids * sizeof(void **);
8134 #endif /* CONFIG_USER_SCHED */ 8140 #endif /* CONFIG_USER_SCHED */
8135 #endif /* CONFIG_FAIR_GROUP_SCHED */ 8141 #endif /* CONFIG_FAIR_GROUP_SCHED */
8136 #ifdef CONFIG_RT_GROUP_SCHED 8142 #ifdef CONFIG_RT_GROUP_SCHED
8137 init_task_group.rt_se = (struct sched_rt_entity **)ptr; 8143 init_task_group.rt_se = (struct sched_rt_entity **)ptr;
8138 ptr += nr_cpu_ids * sizeof(void **); 8144 ptr += nr_cpu_ids * sizeof(void **);
8139 8145
8140 init_task_group.rt_rq = (struct rt_rq **)ptr; 8146 init_task_group.rt_rq = (struct rt_rq **)ptr;
8141 ptr += nr_cpu_ids * sizeof(void **); 8147 ptr += nr_cpu_ids * sizeof(void **);
8142 8148
8143 #ifdef CONFIG_USER_SCHED 8149 #ifdef CONFIG_USER_SCHED
8144 root_task_group.rt_se = (struct sched_rt_entity **)ptr; 8150 root_task_group.rt_se = (struct sched_rt_entity **)ptr;
8145 ptr += nr_cpu_ids * sizeof(void **); 8151 ptr += nr_cpu_ids * sizeof(void **);
8146 8152
8147 root_task_group.rt_rq = (struct rt_rq **)ptr; 8153 root_task_group.rt_rq = (struct rt_rq **)ptr;
8148 ptr += nr_cpu_ids * sizeof(void **); 8154 ptr += nr_cpu_ids * sizeof(void **);
8149 #endif /* CONFIG_USER_SCHED */ 8155 #endif /* CONFIG_USER_SCHED */
8150 #endif /* CONFIG_RT_GROUP_SCHED */ 8156 #endif /* CONFIG_RT_GROUP_SCHED */
8151 } 8157 }
8152 8158
8153 #ifdef CONFIG_SMP 8159 #ifdef CONFIG_SMP
8154 init_defrootdomain(); 8160 init_defrootdomain();
8155 #endif 8161 #endif
8156 8162
8157 init_rt_bandwidth(&def_rt_bandwidth, 8163 init_rt_bandwidth(&def_rt_bandwidth,
8158 global_rt_period(), global_rt_runtime()); 8164 global_rt_period(), global_rt_runtime());
8159 8165
8160 #ifdef CONFIG_RT_GROUP_SCHED 8166 #ifdef CONFIG_RT_GROUP_SCHED
8161 init_rt_bandwidth(&init_task_group.rt_bandwidth, 8167 init_rt_bandwidth(&init_task_group.rt_bandwidth,
8162 global_rt_period(), global_rt_runtime()); 8168 global_rt_period(), global_rt_runtime());
8163 #ifdef CONFIG_USER_SCHED 8169 #ifdef CONFIG_USER_SCHED
8164 init_rt_bandwidth(&root_task_group.rt_bandwidth, 8170 init_rt_bandwidth(&root_task_group.rt_bandwidth,
8165 global_rt_period(), RUNTIME_INF); 8171 global_rt_period(), RUNTIME_INF);
8166 #endif /* CONFIG_USER_SCHED */ 8172 #endif /* CONFIG_USER_SCHED */
8167 #endif /* CONFIG_RT_GROUP_SCHED */ 8173 #endif /* CONFIG_RT_GROUP_SCHED */
8168 8174
8169 #ifdef CONFIG_GROUP_SCHED 8175 #ifdef CONFIG_GROUP_SCHED
8170 list_add(&init_task_group.list, &task_groups); 8176 list_add(&init_task_group.list, &task_groups);
8171 INIT_LIST_HEAD(&init_task_group.children); 8177 INIT_LIST_HEAD(&init_task_group.children);
8172 8178
8173 #ifdef CONFIG_USER_SCHED 8179 #ifdef CONFIG_USER_SCHED
8174 INIT_LIST_HEAD(&root_task_group.children); 8180 INIT_LIST_HEAD(&root_task_group.children);
8175 init_task_group.parent = &root_task_group; 8181 init_task_group.parent = &root_task_group;
8176 list_add(&init_task_group.siblings, &root_task_group.children); 8182 list_add(&init_task_group.siblings, &root_task_group.children);
8177 #endif /* CONFIG_USER_SCHED */ 8183 #endif /* CONFIG_USER_SCHED */
8178 #endif /* CONFIG_GROUP_SCHED */ 8184 #endif /* CONFIG_GROUP_SCHED */
8179 8185
8180 for_each_possible_cpu(i) { 8186 for_each_possible_cpu(i) {
8181 struct rq *rq; 8187 struct rq *rq;
8182 8188
8183 rq = cpu_rq(i); 8189 rq = cpu_rq(i);
8184 spin_lock_init(&rq->lock); 8190 spin_lock_init(&rq->lock);
8185 rq->nr_running = 0; 8191 rq->nr_running = 0;
8186 init_cfs_rq(&rq->cfs, rq); 8192 init_cfs_rq(&rq->cfs, rq);
8187 init_rt_rq(&rq->rt, rq); 8193 init_rt_rq(&rq->rt, rq);
8188 #ifdef CONFIG_FAIR_GROUP_SCHED 8194 #ifdef CONFIG_FAIR_GROUP_SCHED
8189 init_task_group.shares = init_task_group_load; 8195 init_task_group.shares = init_task_group_load;
8190 INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); 8196 INIT_LIST_HEAD(&rq->leaf_cfs_rq_list);
8191 #ifdef CONFIG_CGROUP_SCHED 8197 #ifdef CONFIG_CGROUP_SCHED
8192 /* 8198 /*
8193 * How much cpu bandwidth does init_task_group get? 8199 * How much cpu bandwidth does init_task_group get?
8194 * 8200 *
8195 * In case of task-groups formed thr' the cgroup filesystem, it 8201 * In case of task-groups formed thr' the cgroup filesystem, it
8196 * gets 100% of the cpu resources in the system. This overall 8202 * gets 100% of the cpu resources in the system. This overall
8197 * system cpu resource is divided among the tasks of 8203 * system cpu resource is divided among the tasks of
8198 * init_task_group and its child task-groups in a fair manner, 8204 * init_task_group and its child task-groups in a fair manner,
8199 * based on each entity's (task or task-group's) weight 8205 * based on each entity's (task or task-group's) weight
8200 * (se->load.weight). 8206 * (se->load.weight).
8201 * 8207 *
8202 * In other words, if init_task_group has 10 tasks of weight 8208 * In other words, if init_task_group has 10 tasks of weight
8203 * 1024) and two child groups A0 and A1 (of weight 1024 each), 8209 * 1024) and two child groups A0 and A1 (of weight 1024 each),
8204 * then A0's share of the cpu resource is: 8210 * then A0's share of the cpu resource is:
8205 * 8211 *
8206 * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% 8212 * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33%
8207 * 8213 *
8208 * We achieve this by letting init_task_group's tasks sit 8214 * We achieve this by letting init_task_group's tasks sit
8209 * directly in rq->cfs (i.e init_task_group->se[] = NULL). 8215 * directly in rq->cfs (i.e init_task_group->se[] = NULL).
8210 */ 8216 */
8211 init_tg_cfs_entry(&init_task_group, &rq->cfs, NULL, i, 1, NULL); 8217 init_tg_cfs_entry(&init_task_group, &rq->cfs, NULL, i, 1, NULL);
8212 #elif defined CONFIG_USER_SCHED 8218 #elif defined CONFIG_USER_SCHED
8213 root_task_group.shares = NICE_0_LOAD; 8219 root_task_group.shares = NICE_0_LOAD;
8214 init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, 0, NULL); 8220 init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, 0, NULL);
8215 /* 8221 /*
8216 * In case of task-groups formed thr' the user id of tasks, 8222 * In case of task-groups formed thr' the user id of tasks,
8217 * init_task_group represents tasks belonging to root user. 8223 * init_task_group represents tasks belonging to root user.
8218 * Hence it forms a sibling of all subsequent groups formed. 8224 * Hence it forms a sibling of all subsequent groups formed.
8219 * In this case, init_task_group gets only a fraction of overall 8225 * In this case, init_task_group gets only a fraction of overall
8220 * system cpu resource, based on the weight assigned to root 8226 * system cpu resource, based on the weight assigned to root
8221 * user's cpu share (INIT_TASK_GROUP_LOAD). This is accomplished 8227 * user's cpu share (INIT_TASK_GROUP_LOAD). This is accomplished
8222 * by letting tasks of init_task_group sit in a separate cfs_rq 8228 * by letting tasks of init_task_group sit in a separate cfs_rq
8223 * (init_cfs_rq) and having one entity represent this group of 8229 * (init_cfs_rq) and having one entity represent this group of
8224 * tasks in rq->cfs (i.e init_task_group->se[] != NULL). 8230 * tasks in rq->cfs (i.e init_task_group->se[] != NULL).
8225 */ 8231 */
8226 init_tg_cfs_entry(&init_task_group, 8232 init_tg_cfs_entry(&init_task_group,
8227 &per_cpu(init_cfs_rq, i), 8233 &per_cpu(init_cfs_rq, i),
8228 &per_cpu(init_sched_entity, i), i, 1, 8234 &per_cpu(init_sched_entity, i), i, 1,
8229 root_task_group.se[i]); 8235 root_task_group.se[i]);
8230 8236
8231 #endif 8237 #endif
8232 #endif /* CONFIG_FAIR_GROUP_SCHED */ 8238 #endif /* CONFIG_FAIR_GROUP_SCHED */
8233 8239
8234 rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; 8240 rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime;
8235 #ifdef CONFIG_RT_GROUP_SCHED 8241 #ifdef CONFIG_RT_GROUP_SCHED
8236 INIT_LIST_HEAD(&rq->leaf_rt_rq_list); 8242 INIT_LIST_HEAD(&rq->leaf_rt_rq_list);
8237 #ifdef CONFIG_CGROUP_SCHED 8243 #ifdef CONFIG_CGROUP_SCHED
8238 init_tg_rt_entry(&init_task_group, &rq->rt, NULL, i, 1, NULL); 8244 init_tg_rt_entry(&init_task_group, &rq->rt, NULL, i, 1, NULL);
8239 #elif defined CONFIG_USER_SCHED 8245 #elif defined CONFIG_USER_SCHED
8240 init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, 0, NULL); 8246 init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, 0, NULL);
8241 init_tg_rt_entry(&init_task_group, 8247 init_tg_rt_entry(&init_task_group,
8242 &per_cpu(init_rt_rq, i), 8248 &per_cpu(init_rt_rq, i),
8243 &per_cpu(init_sched_rt_entity, i), i, 1, 8249 &per_cpu(init_sched_rt_entity, i), i, 1,
8244 root_task_group.rt_se[i]); 8250 root_task_group.rt_se[i]);
8245 #endif 8251 #endif
8246 #endif 8252 #endif
8247 8253
8248 for (j = 0; j < CPU_LOAD_IDX_MAX; j++) 8254 for (j = 0; j < CPU_LOAD_IDX_MAX; j++)
8249 rq->cpu_load[j] = 0; 8255 rq->cpu_load[j] = 0;
8250 #ifdef CONFIG_SMP 8256 #ifdef CONFIG_SMP
8251 rq->sd = NULL; 8257 rq->sd = NULL;
8252 rq->rd = NULL; 8258 rq->rd = NULL;
8253 rq->active_balance = 0; 8259 rq->active_balance = 0;
8254 rq->next_balance = jiffies; 8260 rq->next_balance = jiffies;
8255 rq->push_cpu = 0; 8261 rq->push_cpu = 0;
8256 rq->cpu = i; 8262 rq->cpu = i;
8257 rq->online = 0; 8263 rq->online = 0;
8258 rq->migration_thread = NULL; 8264 rq->migration_thread = NULL;
8259 INIT_LIST_HEAD(&rq->migration_queue); 8265 INIT_LIST_HEAD(&rq->migration_queue);
8260 rq_attach_root(rq, &def_root_domain); 8266 rq_attach_root(rq, &def_root_domain);
8261 #endif 8267 #endif
8262 init_rq_hrtick(rq); 8268 init_rq_hrtick(rq);
8263 atomic_set(&rq->nr_iowait, 0); 8269 atomic_set(&rq->nr_iowait, 0);
8264 } 8270 }
8265 8271
8266 set_load_weight(&init_task); 8272 set_load_weight(&init_task);
8267 8273
8268 #ifdef CONFIG_PREEMPT_NOTIFIERS 8274 #ifdef CONFIG_PREEMPT_NOTIFIERS
8269 INIT_HLIST_HEAD(&init_task.preempt_notifiers); 8275 INIT_HLIST_HEAD(&init_task.preempt_notifiers);
8270 #endif 8276 #endif
8271 8277
8272 #ifdef CONFIG_SMP 8278 #ifdef CONFIG_SMP
8273 open_softirq(SCHED_SOFTIRQ, run_rebalance_domains); 8279 open_softirq(SCHED_SOFTIRQ, run_rebalance_domains);
8274 #endif 8280 #endif
8275 8281
8276 #ifdef CONFIG_RT_MUTEXES 8282 #ifdef CONFIG_RT_MUTEXES
8277 plist_head_init(&init_task.pi_waiters, &init_task.pi_lock); 8283 plist_head_init(&init_task.pi_waiters, &init_task.pi_lock);
8278 #endif 8284 #endif
8279 8285
8280 /* 8286 /*
8281 * The boot idle thread does lazy MMU switching as well: 8287 * The boot idle thread does lazy MMU switching as well:
8282 */ 8288 */
8283 atomic_inc(&init_mm.mm_count); 8289 atomic_inc(&init_mm.mm_count);
8284 enter_lazy_tlb(&init_mm, current); 8290 enter_lazy_tlb(&init_mm, current);
8285 8291
8286 /* 8292 /*
8287 * Make us the idle thread. Technically, schedule() should not be 8293 * Make us the idle thread. Technically, schedule() should not be
8288 * called from this thread, however somewhere below it might be, 8294 * called from this thread, however somewhere below it might be,
8289 * but because we are the idle thread, we just pick up running again 8295 * but because we are the idle thread, we just pick up running again
8290 * when this runqueue becomes "idle". 8296 * when this runqueue becomes "idle".
8291 */ 8297 */
8292 init_idle(current, smp_processor_id()); 8298 init_idle(current, smp_processor_id());
8293 /* 8299 /*
8294 * During early bootup we pretend to be a normal task: 8300 * During early bootup we pretend to be a normal task:
8295 */ 8301 */
8296 current->sched_class = &fair_sched_class; 8302 current->sched_class = &fair_sched_class;
8297 8303
8298 scheduler_running = 1; 8304 scheduler_running = 1;
8299 } 8305 }
8300 8306
8301 #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP 8307 #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP
8302 void __might_sleep(char *file, int line) 8308 void __might_sleep(char *file, int line)
8303 { 8309 {
8304 #ifdef in_atomic 8310 #ifdef in_atomic
8305 static unsigned long prev_jiffy; /* ratelimiting */ 8311 static unsigned long prev_jiffy; /* ratelimiting */
8306 8312
8307 if ((!in_atomic() && !irqs_disabled()) || 8313 if ((!in_atomic() && !irqs_disabled()) ||
8308 system_state != SYSTEM_RUNNING || oops_in_progress) 8314 system_state != SYSTEM_RUNNING || oops_in_progress)
8309 return; 8315 return;
8310 if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) 8316 if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy)
8311 return; 8317 return;
8312 prev_jiffy = jiffies; 8318 prev_jiffy = jiffies;
8313 8319
8314 printk(KERN_ERR 8320 printk(KERN_ERR
8315 "BUG: sleeping function called from invalid context at %s:%d\n", 8321 "BUG: sleeping function called from invalid context at %s:%d\n",
8316 file, line); 8322 file, line);
8317 printk(KERN_ERR 8323 printk(KERN_ERR
8318 "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", 8324 "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n",
8319 in_atomic(), irqs_disabled(), 8325 in_atomic(), irqs_disabled(),
8320 current->pid, current->comm); 8326 current->pid, current->comm);
8321 8327
8322 debug_show_held_locks(current); 8328 debug_show_held_locks(current);
8323 if (irqs_disabled()) 8329 if (irqs_disabled())
8324 print_irqtrace_events(current); 8330 print_irqtrace_events(current);
8325 dump_stack(); 8331 dump_stack();
8326 #endif 8332 #endif
8327 } 8333 }
8328 EXPORT_SYMBOL(__might_sleep); 8334 EXPORT_SYMBOL(__might_sleep);
8329 #endif 8335 #endif
8330 8336
8331 #ifdef CONFIG_MAGIC_SYSRQ 8337 #ifdef CONFIG_MAGIC_SYSRQ
8332 static void normalize_task(struct rq *rq, struct task_struct *p) 8338 static void normalize_task(struct rq *rq, struct task_struct *p)
8333 { 8339 {
8334 int on_rq; 8340 int on_rq;
8335 8341
8336 update_rq_clock(rq); 8342 update_rq_clock(rq);
8337 on_rq = p->se.on_rq; 8343 on_rq = p->se.on_rq;
8338 if (on_rq) 8344 if (on_rq)
8339 deactivate_task(rq, p, 0); 8345 deactivate_task(rq, p, 0);
8340 __setscheduler(rq, p, SCHED_NORMAL, 0); 8346 __setscheduler(rq, p, SCHED_NORMAL, 0);
8341 if (on_rq) { 8347 if (on_rq) {
8342 activate_task(rq, p, 0); 8348 activate_task(rq, p, 0);
8343 resched_task(rq->curr); 8349 resched_task(rq->curr);
8344 } 8350 }
8345 } 8351 }
8346 8352
8347 void normalize_rt_tasks(void) 8353 void normalize_rt_tasks(void)
8348 { 8354 {
8349 struct task_struct *g, *p; 8355 struct task_struct *g, *p;
8350 unsigned long flags; 8356 unsigned long flags;
8351 struct rq *rq; 8357 struct rq *rq;
8352 8358
8353 read_lock_irqsave(&tasklist_lock, flags); 8359 read_lock_irqsave(&tasklist_lock, flags);
8354 do_each_thread(g, p) { 8360 do_each_thread(g, p) {
8355 /* 8361 /*
8356 * Only normalize user tasks: 8362 * Only normalize user tasks:
8357 */ 8363 */
8358 if (!p->mm) 8364 if (!p->mm)
8359 continue; 8365 continue;
8360 8366
8361 p->se.exec_start = 0; 8367 p->se.exec_start = 0;
8362 #ifdef CONFIG_SCHEDSTATS 8368 #ifdef CONFIG_SCHEDSTATS
8363 p->se.wait_start = 0; 8369 p->se.wait_start = 0;
8364 p->se.sleep_start = 0; 8370 p->se.sleep_start = 0;
8365 p->se.block_start = 0; 8371 p->se.block_start = 0;
8366 #endif 8372 #endif
8367 8373
8368 if (!rt_task(p)) { 8374 if (!rt_task(p)) {
8369 /* 8375 /*
8370 * Renice negative nice level userspace 8376 * Renice negative nice level userspace
8371 * tasks back to 0: 8377 * tasks back to 0:
8372 */ 8378 */
8373 if (TASK_NICE(p) < 0 && p->mm) 8379 if (TASK_NICE(p) < 0 && p->mm)
8374 set_user_nice(p, 0); 8380 set_user_nice(p, 0);
8375 continue; 8381 continue;
8376 } 8382 }
8377 8383
8378 spin_lock(&p->pi_lock); 8384 spin_lock(&p->pi_lock);
8379 rq = __task_rq_lock(p); 8385 rq = __task_rq_lock(p);
8380 8386
8381 normalize_task(rq, p); 8387 normalize_task(rq, p);
8382 8388
8383 __task_rq_unlock(rq); 8389 __task_rq_unlock(rq);
8384 spin_unlock(&p->pi_lock); 8390 spin_unlock(&p->pi_lock);
8385 } while_each_thread(g, p); 8391 } while_each_thread(g, p);
8386 8392
8387 read_unlock_irqrestore(&tasklist_lock, flags); 8393 read_unlock_irqrestore(&tasklist_lock, flags);
8388 } 8394 }
8389 8395
8390 #endif /* CONFIG_MAGIC_SYSRQ */ 8396 #endif /* CONFIG_MAGIC_SYSRQ */
8391 8397
8392 #ifdef CONFIG_IA64 8398 #ifdef CONFIG_IA64
8393 /* 8399 /*
8394 * These functions are only useful for the IA64 MCA handling. 8400 * These functions are only useful for the IA64 MCA handling.
8395 * 8401 *
8396 * They can only be called when the whole system has been 8402 * They can only be called when the whole system has been
8397 * stopped - every CPU needs to be quiescent, and no scheduling 8403 * stopped - every CPU needs to be quiescent, and no scheduling
8398 * activity can take place. Using them for anything else would 8404 * activity can take place. Using them for anything else would
8399 * be a serious bug, and as a result, they aren't even visible 8405 * be a serious bug, and as a result, they aren't even visible
8400 * under any other configuration. 8406 * under any other configuration.
8401 */ 8407 */
8402 8408
8403 /** 8409 /**
8404 * curr_task - return the current task for a given cpu. 8410 * curr_task - return the current task for a given cpu.
8405 * @cpu: the processor in question. 8411 * @cpu: the processor in question.
8406 * 8412 *
8407 * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! 8413 * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED!
8408 */ 8414 */
8409 struct task_struct *curr_task(int cpu) 8415 struct task_struct *curr_task(int cpu)
8410 { 8416 {
8411 return cpu_curr(cpu); 8417 return cpu_curr(cpu);
8412 } 8418 }
8413 8419
8414 /** 8420 /**
8415 * set_curr_task - set the current task for a given cpu. 8421 * set_curr_task - set the current task for a given cpu.
8416 * @cpu: the processor in question. 8422 * @cpu: the processor in question.
8417 * @p: the task pointer to set. 8423 * @p: the task pointer to set.
8418 * 8424 *
8419 * Description: This function must only be used when non-maskable interrupts 8425 * Description: This function must only be used when non-maskable interrupts
8420 * are serviced on a separate stack. It allows the architecture to switch the 8426 * are serviced on a separate stack. It allows the architecture to switch the
8421 * notion of the current task on a cpu in a non-blocking manner. This function 8427 * notion of the current task on a cpu in a non-blocking manner. This function
8422 * must be called with all CPU's synchronized, and interrupts disabled, the 8428 * must be called with all CPU's synchronized, and interrupts disabled, the
8423 * and caller must save the original value of the current task (see 8429 * and caller must save the original value of the current task (see
8424 * curr_task() above) and restore that value before reenabling interrupts and 8430 * curr_task() above) and restore that value before reenabling interrupts and
8425 * re-starting the system. 8431 * re-starting the system.
8426 * 8432 *
8427 * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! 8433 * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED!
8428 */ 8434 */
8429 void set_curr_task(int cpu, struct task_struct *p) 8435 void set_curr_task(int cpu, struct task_struct *p)
8430 { 8436 {
8431 cpu_curr(cpu) = p; 8437 cpu_curr(cpu) = p;
8432 } 8438 }
8433 8439
8434 #endif 8440 #endif
8435 8441
8436 #ifdef CONFIG_FAIR_GROUP_SCHED 8442 #ifdef CONFIG_FAIR_GROUP_SCHED
8437 static void free_fair_sched_group(struct task_group *tg) 8443 static void free_fair_sched_group(struct task_group *tg)
8438 { 8444 {
8439 int i; 8445 int i;
8440 8446
8441 for_each_possible_cpu(i) { 8447 for_each_possible_cpu(i) {
8442 if (tg->cfs_rq) 8448 if (tg->cfs_rq)
8443 kfree(tg->cfs_rq[i]); 8449 kfree(tg->cfs_rq[i]);
8444 if (tg->se) 8450 if (tg->se)
8445 kfree(tg->se[i]); 8451 kfree(tg->se[i]);
8446 } 8452 }
8447 8453
8448 kfree(tg->cfs_rq); 8454 kfree(tg->cfs_rq);
8449 kfree(tg->se); 8455 kfree(tg->se);
8450 } 8456 }
8451 8457
8452 static 8458 static
8453 int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) 8459 int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent)
8454 { 8460 {
8455 struct cfs_rq *cfs_rq; 8461 struct cfs_rq *cfs_rq;
8456 struct sched_entity *se; 8462 struct sched_entity *se;
8457 struct rq *rq; 8463 struct rq *rq;
8458 int i; 8464 int i;
8459 8465
8460 tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL); 8466 tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL);
8461 if (!tg->cfs_rq) 8467 if (!tg->cfs_rq)
8462 goto err; 8468 goto err;
8463 tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL); 8469 tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL);
8464 if (!tg->se) 8470 if (!tg->se)
8465 goto err; 8471 goto err;
8466 8472
8467 tg->shares = NICE_0_LOAD; 8473 tg->shares = NICE_0_LOAD;
8468 8474
8469 for_each_possible_cpu(i) { 8475 for_each_possible_cpu(i) {
8470 rq = cpu_rq(i); 8476 rq = cpu_rq(i);
8471 8477
8472 cfs_rq = kzalloc_node(sizeof(struct cfs_rq), 8478 cfs_rq = kzalloc_node(sizeof(struct cfs_rq),
8473 GFP_KERNEL, cpu_to_node(i)); 8479 GFP_KERNEL, cpu_to_node(i));
8474 if (!cfs_rq) 8480 if (!cfs_rq)
8475 goto err; 8481 goto err;
8476 8482
8477 se = kzalloc_node(sizeof(struct sched_entity), 8483 se = kzalloc_node(sizeof(struct sched_entity),
8478 GFP_KERNEL, cpu_to_node(i)); 8484 GFP_KERNEL, cpu_to_node(i));
8479 if (!se) 8485 if (!se)
8480 goto err; 8486 goto err;
8481 8487
8482 init_tg_cfs_entry(tg, cfs_rq, se, i, 0, parent->se[i]); 8488 init_tg_cfs_entry(tg, cfs_rq, se, i, 0, parent->se[i]);
8483 } 8489 }
8484 8490
8485 return 1; 8491 return 1;
8486 8492
8487 err: 8493 err:
8488 return 0; 8494 return 0;
8489 } 8495 }
8490 8496
8491 static inline void register_fair_sched_group(struct task_group *tg, int cpu) 8497 static inline void register_fair_sched_group(struct task_group *tg, int cpu)
8492 { 8498 {
8493 list_add_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list, 8499 list_add_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list,
8494 &cpu_rq(cpu)->leaf_cfs_rq_list); 8500 &cpu_rq(cpu)->leaf_cfs_rq_list);
8495 } 8501 }
8496 8502
8497 static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) 8503 static inline void unregister_fair_sched_group(struct task_group *tg, int cpu)
8498 { 8504 {
8499 list_del_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list); 8505 list_del_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list);
8500 } 8506 }
8501 #else /* !CONFG_FAIR_GROUP_SCHED */ 8507 #else /* !CONFG_FAIR_GROUP_SCHED */
8502 static inline void free_fair_sched_group(struct task_group *tg) 8508 static inline void free_fair_sched_group(struct task_group *tg)
8503 { 8509 {
8504 } 8510 }
8505 8511
8506 static inline 8512 static inline
8507 int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) 8513 int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent)
8508 { 8514 {
8509 return 1; 8515 return 1;
8510 } 8516 }
8511 8517
8512 static inline void register_fair_sched_group(struct task_group *tg, int cpu) 8518 static inline void register_fair_sched_group(struct task_group *tg, int cpu)
8513 { 8519 {
8514 } 8520 }
8515 8521
8516 static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) 8522 static inline void unregister_fair_sched_group(struct task_group *tg, int cpu)
8517 { 8523 {
8518 } 8524 }
8519 #endif /* CONFIG_FAIR_GROUP_SCHED */ 8525 #endif /* CONFIG_FAIR_GROUP_SCHED */
8520 8526
8521 #ifdef CONFIG_RT_GROUP_SCHED 8527 #ifdef CONFIG_RT_GROUP_SCHED
8522 static void free_rt_sched_group(struct task_group *tg) 8528 static void free_rt_sched_group(struct task_group *tg)
8523 { 8529 {
8524 int i; 8530 int i;
8525 8531
8526 destroy_rt_bandwidth(&tg->rt_bandwidth); 8532 destroy_rt_bandwidth(&tg->rt_bandwidth);
8527 8533
8528 for_each_possible_cpu(i) { 8534 for_each_possible_cpu(i) {
8529 if (tg->rt_rq) 8535 if (tg->rt_rq)
8530 kfree(tg->rt_rq[i]); 8536 kfree(tg->rt_rq[i]);
8531 if (tg->rt_se) 8537 if (tg->rt_se)
8532 kfree(tg->rt_se[i]); 8538 kfree(tg->rt_se[i]);
8533 } 8539 }
8534 8540
8535 kfree(tg->rt_rq); 8541 kfree(tg->rt_rq);
8536 kfree(tg->rt_se); 8542 kfree(tg->rt_se);
8537 } 8543 }
8538 8544
8539 static 8545 static
8540 int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) 8546 int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent)
8541 { 8547 {
8542 struct rt_rq *rt_rq; 8548 struct rt_rq *rt_rq;
8543 struct sched_rt_entity *rt_se; 8549 struct sched_rt_entity *rt_se;
8544 struct rq *rq; 8550 struct rq *rq;
8545 int i; 8551 int i;
8546 8552
8547 tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL); 8553 tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL);
8548 if (!tg->rt_rq) 8554 if (!tg->rt_rq)
8549 goto err; 8555 goto err;
8550 tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL); 8556 tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL);
8551 if (!tg->rt_se) 8557 if (!tg->rt_se)
8552 goto err; 8558 goto err;
8553 8559
8554 init_rt_bandwidth(&tg->rt_bandwidth, 8560 init_rt_bandwidth(&tg->rt_bandwidth,
8555 ktime_to_ns(def_rt_bandwidth.rt_period), 0); 8561 ktime_to_ns(def_rt_bandwidth.rt_period), 0);
8556 8562
8557 for_each_possible_cpu(i) { 8563 for_each_possible_cpu(i) {
8558 rq = cpu_rq(i); 8564 rq = cpu_rq(i);
8559 8565
8560 rt_rq = kzalloc_node(sizeof(struct rt_rq), 8566 rt_rq = kzalloc_node(sizeof(struct rt_rq),
8561 GFP_KERNEL, cpu_to_node(i)); 8567 GFP_KERNEL, cpu_to_node(i));
8562 if (!rt_rq) 8568 if (!rt_rq)
8563 goto err; 8569 goto err;
8564 8570
8565 rt_se = kzalloc_node(sizeof(struct sched_rt_entity), 8571 rt_se = kzalloc_node(sizeof(struct sched_rt_entity),
8566 GFP_KERNEL, cpu_to_node(i)); 8572 GFP_KERNEL, cpu_to_node(i));
8567 if (!rt_se) 8573 if (!rt_se)
8568 goto err; 8574 goto err;
8569 8575
8570 init_tg_rt_entry(tg, rt_rq, rt_se, i, 0, parent->rt_se[i]); 8576 init_tg_rt_entry(tg, rt_rq, rt_se, i, 0, parent->rt_se[i]);
8571 } 8577 }
8572 8578
8573 return 1; 8579 return 1;
8574 8580
8575 err: 8581 err:
8576 return 0; 8582 return 0;
8577 } 8583 }
8578 8584
8579 static inline void register_rt_sched_group(struct task_group *tg, int cpu) 8585 static inline void register_rt_sched_group(struct task_group *tg, int cpu)
8580 { 8586 {
8581 list_add_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list, 8587 list_add_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list,
8582 &cpu_rq(cpu)->leaf_rt_rq_list); 8588 &cpu_rq(cpu)->leaf_rt_rq_list);
8583 } 8589 }
8584 8590
8585 static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) 8591 static inline void unregister_rt_sched_group(struct task_group *tg, int cpu)
8586 { 8592 {
8587 list_del_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list); 8593 list_del_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list);
8588 } 8594 }
8589 #else /* !CONFIG_RT_GROUP_SCHED */ 8595 #else /* !CONFIG_RT_GROUP_SCHED */
8590 static inline void free_rt_sched_group(struct task_group *tg) 8596 static inline void free_rt_sched_group(struct task_group *tg)
8591 { 8597 {
8592 } 8598 }
8593 8599
8594 static inline 8600 static inline
8595 int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) 8601 int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent)
8596 { 8602 {
8597 return 1; 8603 return 1;
8598 } 8604 }
8599 8605
8600 static inline void register_rt_sched_group(struct task_group *tg, int cpu) 8606 static inline void register_rt_sched_group(struct task_group *tg, int cpu)
8601 { 8607 {
8602 } 8608 }
8603 8609
8604 static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) 8610 static inline void unregister_rt_sched_group(struct task_group *tg, int cpu)
8605 { 8611 {
8606 } 8612 }
8607 #endif /* CONFIG_RT_GROUP_SCHED */ 8613 #endif /* CONFIG_RT_GROUP_SCHED */
8608 8614
8609 #ifdef CONFIG_GROUP_SCHED 8615 #ifdef CONFIG_GROUP_SCHED
8610 static void free_sched_group(struct task_group *tg) 8616 static void free_sched_group(struct task_group *tg)
8611 { 8617 {
8612 free_fair_sched_group(tg); 8618 free_fair_sched_group(tg);
8613 free_rt_sched_group(tg); 8619 free_rt_sched_group(tg);
8614 kfree(tg); 8620 kfree(tg);
8615 } 8621 }
8616 8622
8617 /* allocate runqueue etc for a new task group */ 8623 /* allocate runqueue etc for a new task group */
8618 struct task_group *sched_create_group(struct task_group *parent) 8624 struct task_group *sched_create_group(struct task_group *parent)
8619 { 8625 {
8620 struct task_group *tg; 8626 struct task_group *tg;
8621 unsigned long flags; 8627 unsigned long flags;
8622 int i; 8628 int i;
8623 8629
8624 tg = kzalloc(sizeof(*tg), GFP_KERNEL); 8630 tg = kzalloc(sizeof(*tg), GFP_KERNEL);
8625 if (!tg) 8631 if (!tg)
8626 return ERR_PTR(-ENOMEM); 8632 return ERR_PTR(-ENOMEM);
8627 8633
8628 if (!alloc_fair_sched_group(tg, parent)) 8634 if (!alloc_fair_sched_group(tg, parent))
8629 goto err; 8635 goto err;
8630 8636
8631 if (!alloc_rt_sched_group(tg, parent)) 8637 if (!alloc_rt_sched_group(tg, parent))
8632 goto err; 8638 goto err;
8633 8639
8634 spin_lock_irqsave(&task_group_lock, flags); 8640 spin_lock_irqsave(&task_group_lock, flags);
8635 for_each_possible_cpu(i) { 8641 for_each_possible_cpu(i) {
8636 register_fair_sched_group(tg, i); 8642 register_fair_sched_group(tg, i);
8637 register_rt_sched_group(tg, i); 8643 register_rt_sched_group(tg, i);
8638 } 8644 }
8639 list_add_rcu(&tg->list, &task_groups); 8645 list_add_rcu(&tg->list, &task_groups);
8640 8646
8641 WARN_ON(!parent); /* root should already exist */ 8647 WARN_ON(!parent); /* root should already exist */
8642 8648
8643 tg->parent = parent; 8649 tg->parent = parent;
8644 INIT_LIST_HEAD(&tg->children); 8650 INIT_LIST_HEAD(&tg->children);
8645 list_add_rcu(&tg->siblings, &parent->children); 8651 list_add_rcu(&tg->siblings, &parent->children);
8646 spin_unlock_irqrestore(&task_group_lock, flags); 8652 spin_unlock_irqrestore(&task_group_lock, flags);
8647 8653
8648 return tg; 8654 return tg;
8649 8655
8650 err: 8656 err:
8651 free_sched_group(tg); 8657 free_sched_group(tg);
8652 return ERR_PTR(-ENOMEM); 8658 return ERR_PTR(-ENOMEM);
8653 } 8659 }
8654 8660
8655 /* rcu callback to free various structures associated with a task group */ 8661 /* rcu callback to free various structures associated with a task group */
8656 static void free_sched_group_rcu(struct rcu_head *rhp) 8662 static void free_sched_group_rcu(struct rcu_head *rhp)
8657 { 8663 {
8658 /* now it should be safe to free those cfs_rqs */ 8664 /* now it should be safe to free those cfs_rqs */
8659 free_sched_group(container_of(rhp, struct task_group, rcu)); 8665 free_sched_group(container_of(rhp, struct task_group, rcu));
8660 } 8666 }
8661 8667
8662 /* Destroy runqueue etc associated with a task group */ 8668 /* Destroy runqueue etc associated with a task group */
8663 void sched_destroy_group(struct task_group *tg) 8669 void sched_destroy_group(struct task_group *tg)
8664 { 8670 {
8665 unsigned long flags; 8671 unsigned long flags;
8666 int i; 8672 int i;
8667 8673
8668 spin_lock_irqsave(&task_group_lock, flags); 8674 spin_lock_irqsave(&task_group_lock, flags);
8669 for_each_possible_cpu(i) { 8675 for_each_possible_cpu(i) {
8670 unregister_fair_sched_group(tg, i); 8676 unregister_fair_sched_group(tg, i);
8671 unregister_rt_sched_group(tg, i); 8677 unregister_rt_sched_group(tg, i);
8672 } 8678 }
8673 list_del_rcu(&tg->list); 8679 list_del_rcu(&tg->list);
8674 list_del_rcu(&tg->siblings); 8680 list_del_rcu(&tg->siblings);
8675 spin_unlock_irqrestore(&task_group_lock, flags); 8681 spin_unlock_irqrestore(&task_group_lock, flags);
8676 8682
8677 /* wait for possible concurrent references to cfs_rqs complete */ 8683 /* wait for possible concurrent references to cfs_rqs complete */
8678 call_rcu(&tg->rcu, free_sched_group_rcu); 8684 call_rcu(&tg->rcu, free_sched_group_rcu);
8679 } 8685 }
8680 8686
8681 /* change task's runqueue when it moves between groups. 8687 /* change task's runqueue when it moves between groups.
8682 * The caller of this function should have put the task in its new group 8688 * The caller of this function should have put the task in its new group
8683 * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to 8689 * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to
8684 * reflect its new group. 8690 * reflect its new group.
8685 */ 8691 */
8686 void sched_move_task(struct task_struct *tsk) 8692 void sched_move_task(struct task_struct *tsk)
8687 { 8693 {
8688 int on_rq, running; 8694 int on_rq, running;
8689 unsigned long flags; 8695 unsigned long flags;
8690 struct rq *rq; 8696 struct rq *rq;
8691 8697
8692 rq = task_rq_lock(tsk, &flags); 8698 rq = task_rq_lock(tsk, &flags);
8693 8699
8694 update_rq_clock(rq); 8700 update_rq_clock(rq);
8695 8701
8696 running = task_current(rq, tsk); 8702 running = task_current(rq, tsk);
8697 on_rq = tsk->se.on_rq; 8703 on_rq = tsk->se.on_rq;
8698 8704
8699 if (on_rq) 8705 if (on_rq)
8700 dequeue_task(rq, tsk, 0); 8706 dequeue_task(rq, tsk, 0);
8701 if (unlikely(running)) 8707 if (unlikely(running))
8702 tsk->sched_class->put_prev_task(rq, tsk); 8708 tsk->sched_class->put_prev_task(rq, tsk);
8703 8709
8704 set_task_rq(tsk, task_cpu(tsk)); 8710 set_task_rq(tsk, task_cpu(tsk));
8705 8711
8706 #ifdef CONFIG_FAIR_GROUP_SCHED 8712 #ifdef CONFIG_FAIR_GROUP_SCHED
8707 if (tsk->sched_class->moved_group) 8713 if (tsk->sched_class->moved_group)
8708 tsk->sched_class->moved_group(tsk); 8714 tsk->sched_class->moved_group(tsk);
8709 #endif 8715 #endif
8710 8716
8711 if (unlikely(running)) 8717 if (unlikely(running))
8712 tsk->sched_class->set_curr_task(rq); 8718 tsk->sched_class->set_curr_task(rq);
8713 if (on_rq) 8719 if (on_rq)
8714 enqueue_task(rq, tsk, 0); 8720 enqueue_task(rq, tsk, 0);
8715 8721
8716 task_rq_unlock(rq, &flags); 8722 task_rq_unlock(rq, &flags);
8717 } 8723 }
8718 #endif /* CONFIG_GROUP_SCHED */ 8724 #endif /* CONFIG_GROUP_SCHED */
8719 8725
8720 #ifdef CONFIG_FAIR_GROUP_SCHED 8726 #ifdef CONFIG_FAIR_GROUP_SCHED
8721 static void __set_se_shares(struct sched_entity *se, unsigned long shares) 8727 static void __set_se_shares(struct sched_entity *se, unsigned long shares)
8722 { 8728 {
8723 struct cfs_rq *cfs_rq = se->cfs_rq; 8729 struct cfs_rq *cfs_rq = se->cfs_rq;
8724 int on_rq; 8730 int on_rq;
8725 8731
8726 on_rq = se->on_rq; 8732 on_rq = se->on_rq;
8727 if (on_rq) 8733 if (on_rq)
8728 dequeue_entity(cfs_rq, se, 0); 8734 dequeue_entity(cfs_rq, se, 0);
8729 8735
8730 se->load.weight = shares; 8736 se->load.weight = shares;
8731 se->load.inv_weight = 0; 8737 se->load.inv_weight = 0;
8732 8738
8733 if (on_rq) 8739 if (on_rq)
8734 enqueue_entity(cfs_rq, se, 0); 8740 enqueue_entity(cfs_rq, se, 0);
8735 } 8741 }
8736 8742
8737 static void set_se_shares(struct sched_entity *se, unsigned long shares) 8743 static void set_se_shares(struct sched_entity *se, unsigned long shares)
8738 { 8744 {
8739 struct cfs_rq *cfs_rq = se->cfs_rq; 8745 struct cfs_rq *cfs_rq = se->cfs_rq;
8740 struct rq *rq = cfs_rq->rq; 8746 struct rq *rq = cfs_rq->rq;
8741 unsigned long flags; 8747 unsigned long flags;
8742 8748
8743 spin_lock_irqsave(&rq->lock, flags); 8749 spin_lock_irqsave(&rq->lock, flags);
8744 __set_se_shares(se, shares); 8750 __set_se_shares(se, shares);
8745 spin_unlock_irqrestore(&rq->lock, flags); 8751 spin_unlock_irqrestore(&rq->lock, flags);
8746 } 8752 }
8747 8753
8748 static DEFINE_MUTEX(shares_mutex); 8754 static DEFINE_MUTEX(shares_mutex);
8749 8755
8750 int sched_group_set_shares(struct task_group *tg, unsigned long shares) 8756 int sched_group_set_shares(struct task_group *tg, unsigned long shares)
8751 { 8757 {
8752 int i; 8758 int i;
8753 unsigned long flags; 8759 unsigned long flags;
8754 8760
8755 /* 8761 /*
8756 * We can't change the weight of the root cgroup. 8762 * We can't change the weight of the root cgroup.
8757 */ 8763 */
8758 if (!tg->se[0]) 8764 if (!tg->se[0])
8759 return -EINVAL; 8765 return -EINVAL;
8760 8766
8761 if (shares < MIN_SHARES) 8767 if (shares < MIN_SHARES)
8762 shares = MIN_SHARES; 8768 shares = MIN_SHARES;
8763 else if (shares > MAX_SHARES) 8769 else if (shares > MAX_SHARES)
8764 shares = MAX_SHARES; 8770 shares = MAX_SHARES;
8765 8771
8766 mutex_lock(&shares_mutex); 8772 mutex_lock(&shares_mutex);
8767 if (tg->shares == shares) 8773 if (tg->shares == shares)
8768 goto done; 8774 goto done;
8769 8775
8770 spin_lock_irqsave(&task_group_lock, flags); 8776 spin_lock_irqsave(&task_group_lock, flags);
8771 for_each_possible_cpu(i) 8777 for_each_possible_cpu(i)
8772 unregister_fair_sched_group(tg, i); 8778 unregister_fair_sched_group(tg, i);
8773 list_del_rcu(&tg->siblings); 8779 list_del_rcu(&tg->siblings);
8774 spin_unlock_irqrestore(&task_group_lock, flags); 8780 spin_unlock_irqrestore(&task_group_lock, flags);
8775 8781
8776 /* wait for any ongoing reference to this group to finish */ 8782 /* wait for any ongoing reference to this group to finish */
8777 synchronize_sched(); 8783 synchronize_sched();
8778 8784
8779 /* 8785 /*
8780 * Now we are free to modify the group's share on each cpu 8786 * Now we are free to modify the group's share on each cpu
8781 * w/o tripping rebalance_share or load_balance_fair. 8787 * w/o tripping rebalance_share or load_balance_fair.
8782 */ 8788 */
8783 tg->shares = shares; 8789 tg->shares = shares;
8784 for_each_possible_cpu(i) { 8790 for_each_possible_cpu(i) {
8785 /* 8791 /*
8786 * force a rebalance 8792 * force a rebalance
8787 */ 8793 */
8788 cfs_rq_set_shares(tg->cfs_rq[i], 0); 8794 cfs_rq_set_shares(tg->cfs_rq[i], 0);
8789 set_se_shares(tg->se[i], shares); 8795 set_se_shares(tg->se[i], shares);
8790 } 8796 }
8791 8797
8792 /* 8798 /*
8793 * Enable load balance activity on this group, by inserting it back on 8799 * Enable load balance activity on this group, by inserting it back on
8794 * each cpu's rq->leaf_cfs_rq_list. 8800 * each cpu's rq->leaf_cfs_rq_list.
8795 */ 8801 */
8796 spin_lock_irqsave(&task_group_lock, flags); 8802 spin_lock_irqsave(&task_group_lock, flags);
8797 for_each_possible_cpu(i) 8803 for_each_possible_cpu(i)
8798 register_fair_sched_group(tg, i); 8804 register_fair_sched_group(tg, i);
8799 list_add_rcu(&tg->siblings, &tg->parent->children); 8805 list_add_rcu(&tg->siblings, &tg->parent->children);
8800 spin_unlock_irqrestore(&task_group_lock, flags); 8806 spin_unlock_irqrestore(&task_group_lock, flags);
8801 done: 8807 done:
8802 mutex_unlock(&shares_mutex); 8808 mutex_unlock(&shares_mutex);
8803 return 0; 8809 return 0;
8804 } 8810 }
8805 8811
8806 unsigned long sched_group_shares(struct task_group *tg) 8812 unsigned long sched_group_shares(struct task_group *tg)
8807 { 8813 {
8808 return tg->shares; 8814 return tg->shares;
8809 } 8815 }
8810 #endif 8816 #endif
8811 8817
8812 #ifdef CONFIG_RT_GROUP_SCHED 8818 #ifdef CONFIG_RT_GROUP_SCHED
8813 /* 8819 /*
8814 * Ensure that the real time constraints are schedulable. 8820 * Ensure that the real time constraints are schedulable.
8815 */ 8821 */
8816 static DEFINE_MUTEX(rt_constraints_mutex); 8822 static DEFINE_MUTEX(rt_constraints_mutex);
8817 8823
8818 static unsigned long to_ratio(u64 period, u64 runtime) 8824 static unsigned long to_ratio(u64 period, u64 runtime)
8819 { 8825 {
8820 if (runtime == RUNTIME_INF) 8826 if (runtime == RUNTIME_INF)
8821 return 1ULL << 20; 8827 return 1ULL << 20;
8822 8828
8823 return div64_u64(runtime << 20, period); 8829 return div64_u64(runtime << 20, period);
8824 } 8830 }
8825 8831
8826 /* Must be called with tasklist_lock held */ 8832 /* Must be called with tasklist_lock held */
8827 static inline int tg_has_rt_tasks(struct task_group *tg) 8833 static inline int tg_has_rt_tasks(struct task_group *tg)
8828 { 8834 {
8829 struct task_struct *g, *p; 8835 struct task_struct *g, *p;
8830 8836
8831 do_each_thread(g, p) { 8837 do_each_thread(g, p) {
8832 if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg) 8838 if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg)
8833 return 1; 8839 return 1;
8834 } while_each_thread(g, p); 8840 } while_each_thread(g, p);
8835 8841
8836 return 0; 8842 return 0;
8837 } 8843 }
8838 8844
8839 struct rt_schedulable_data { 8845 struct rt_schedulable_data {
8840 struct task_group *tg; 8846 struct task_group *tg;
8841 u64 rt_period; 8847 u64 rt_period;
8842 u64 rt_runtime; 8848 u64 rt_runtime;
8843 }; 8849 };
8844 8850
8845 static int tg_schedulable(struct task_group *tg, void *data) 8851 static int tg_schedulable(struct task_group *tg, void *data)
8846 { 8852 {
8847 struct rt_schedulable_data *d = data; 8853 struct rt_schedulable_data *d = data;
8848 struct task_group *child; 8854 struct task_group *child;
8849 unsigned long total, sum = 0; 8855 unsigned long total, sum = 0;
8850 u64 period, runtime; 8856 u64 period, runtime;
8851 8857
8852 period = ktime_to_ns(tg->rt_bandwidth.rt_period); 8858 period = ktime_to_ns(tg->rt_bandwidth.rt_period);
8853 runtime = tg->rt_bandwidth.rt_runtime; 8859 runtime = tg->rt_bandwidth.rt_runtime;
8854 8860
8855 if (tg == d->tg) { 8861 if (tg == d->tg) {
8856 period = d->rt_period; 8862 period = d->rt_period;
8857 runtime = d->rt_runtime; 8863 runtime = d->rt_runtime;
8858 } 8864 }
8859 8865
8860 /* 8866 /*
8861 * Cannot have more runtime than the period. 8867 * Cannot have more runtime than the period.
8862 */ 8868 */
8863 if (runtime > period && runtime != RUNTIME_INF) 8869 if (runtime > period && runtime != RUNTIME_INF)
8864 return -EINVAL; 8870 return -EINVAL;
8865 8871
8866 /* 8872 /*
8867 * Ensure we don't starve existing RT tasks. 8873 * Ensure we don't starve existing RT tasks.
8868 */ 8874 */
8869 if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) 8875 if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg))
8870 return -EBUSY; 8876 return -EBUSY;
8871 8877
8872 total = to_ratio(period, runtime); 8878 total = to_ratio(period, runtime);
8873 8879
8874 /* 8880 /*
8875 * Nobody can have more than the global setting allows. 8881 * Nobody can have more than the global setting allows.
8876 */ 8882 */
8877 if (total > to_ratio(global_rt_period(), global_rt_runtime())) 8883 if (total > to_ratio(global_rt_period(), global_rt_runtime()))
8878 return -EINVAL; 8884 return -EINVAL;
8879 8885
8880 /* 8886 /*
8881 * The sum of our children's runtime should not exceed our own. 8887 * The sum of our children's runtime should not exceed our own.
8882 */ 8888 */
8883 list_for_each_entry_rcu(child, &tg->children, siblings) { 8889 list_for_each_entry_rcu(child, &tg->children, siblings) {
8884 period = ktime_to_ns(child->rt_bandwidth.rt_period); 8890 period = ktime_to_ns(child->rt_bandwidth.rt_period);
8885 runtime = child->rt_bandwidth.rt_runtime; 8891 runtime = child->rt_bandwidth.rt_runtime;
8886 8892
8887 if (child == d->tg) { 8893 if (child == d->tg) {
8888 period = d->rt_period; 8894 period = d->rt_period;
8889 runtime = d->rt_runtime; 8895 runtime = d->rt_runtime;
8890 } 8896 }
8891 8897
8892 sum += to_ratio(period, runtime); 8898 sum += to_ratio(period, runtime);
8893 } 8899 }
8894 8900
8895 if (sum > total) 8901 if (sum > total)
8896 return -EINVAL; 8902 return -EINVAL;
8897 8903
8898 return 0; 8904 return 0;
8899 } 8905 }
8900 8906
8901 static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) 8907 static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime)
8902 { 8908 {
8903 struct rt_schedulable_data data = { 8909 struct rt_schedulable_data data = {
8904 .tg = tg, 8910 .tg = tg,
8905 .rt_period = period, 8911 .rt_period = period,
8906 .rt_runtime = runtime, 8912 .rt_runtime = runtime,
8907 }; 8913 };
8908 8914
8909 return walk_tg_tree(tg_schedulable, tg_nop, &data); 8915 return walk_tg_tree(tg_schedulable, tg_nop, &data);
8910 } 8916 }
8911 8917
8912 static int tg_set_bandwidth(struct task_group *tg, 8918 static int tg_set_bandwidth(struct task_group *tg,
8913 u64 rt_period, u64 rt_runtime) 8919 u64 rt_period, u64 rt_runtime)
8914 { 8920 {
8915 int i, err = 0; 8921 int i, err = 0;
8916 8922
8917 mutex_lock(&rt_constraints_mutex); 8923 mutex_lock(&rt_constraints_mutex);
8918 read_lock(&tasklist_lock); 8924 read_lock(&tasklist_lock);
8919 err = __rt_schedulable(tg, rt_period, rt_runtime); 8925 err = __rt_schedulable(tg, rt_period, rt_runtime);
8920 if (err) 8926 if (err)
8921 goto unlock; 8927 goto unlock;
8922 8928
8923 spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); 8929 spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock);
8924 tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); 8930 tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period);
8925 tg->rt_bandwidth.rt_runtime = rt_runtime; 8931 tg->rt_bandwidth.rt_runtime = rt_runtime;
8926 8932
8927 for_each_possible_cpu(i) { 8933 for_each_possible_cpu(i) {
8928 struct rt_rq *rt_rq = tg->rt_rq[i]; 8934 struct rt_rq *rt_rq = tg->rt_rq[i];
8929 8935
8930 spin_lock(&rt_rq->rt_runtime_lock); 8936 spin_lock(&rt_rq->rt_runtime_lock);
8931 rt_rq->rt_runtime = rt_runtime; 8937 rt_rq->rt_runtime = rt_runtime;
8932 spin_unlock(&rt_rq->rt_runtime_lock); 8938 spin_unlock(&rt_rq->rt_runtime_lock);
8933 } 8939 }
8934 spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); 8940 spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock);
8935 unlock: 8941 unlock:
8936 read_unlock(&tasklist_lock); 8942 read_unlock(&tasklist_lock);
8937 mutex_unlock(&rt_constraints_mutex); 8943 mutex_unlock(&rt_constraints_mutex);
8938 8944
8939 return err; 8945 return err;
8940 } 8946 }
8941 8947
8942 int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) 8948 int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us)
8943 { 8949 {
8944 u64 rt_runtime, rt_period; 8950 u64 rt_runtime, rt_period;
8945 8951
8946 rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); 8952 rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period);
8947 rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; 8953 rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC;
8948 if (rt_runtime_us < 0) 8954 if (rt_runtime_us < 0)
8949 rt_runtime = RUNTIME_INF; 8955 rt_runtime = RUNTIME_INF;
8950 8956
8951 return tg_set_bandwidth(tg, rt_period, rt_runtime); 8957 return tg_set_bandwidth(tg, rt_period, rt_runtime);
8952 } 8958 }
8953 8959
8954 long sched_group_rt_runtime(struct task_group *tg) 8960 long sched_group_rt_runtime(struct task_group *tg)
8955 { 8961 {
8956 u64 rt_runtime_us; 8962 u64 rt_runtime_us;
8957 8963
8958 if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) 8964 if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF)
8959 return -1; 8965 return -1;
8960 8966
8961 rt_runtime_us = tg->rt_bandwidth.rt_runtime; 8967 rt_runtime_us = tg->rt_bandwidth.rt_runtime;
8962 do_div(rt_runtime_us, NSEC_PER_USEC); 8968 do_div(rt_runtime_us, NSEC_PER_USEC);
8963 return rt_runtime_us; 8969 return rt_runtime_us;
8964 } 8970 }
8965 8971
8966 int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) 8972 int sched_group_set_rt_period(struct task_group *tg, long rt_period_us)
8967 { 8973 {
8968 u64 rt_runtime, rt_period; 8974 u64 rt_runtime, rt_period;
8969 8975
8970 rt_period = (u64)rt_period_us * NSEC_PER_USEC; 8976 rt_period = (u64)rt_period_us * NSEC_PER_USEC;
8971 rt_runtime = tg->rt_bandwidth.rt_runtime; 8977 rt_runtime = tg->rt_bandwidth.rt_runtime;
8972 8978
8973 if (rt_period == 0) 8979 if (rt_period == 0)
8974 return -EINVAL; 8980 return -EINVAL;
8975 8981
8976 return tg_set_bandwidth(tg, rt_period, rt_runtime); 8982 return tg_set_bandwidth(tg, rt_period, rt_runtime);
8977 } 8983 }
8978 8984
8979 long sched_group_rt_period(struct task_group *tg) 8985 long sched_group_rt_period(struct task_group *tg)
8980 { 8986 {
8981 u64 rt_period_us; 8987 u64 rt_period_us;
8982 8988
8983 rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); 8989 rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period);
8984 do_div(rt_period_us, NSEC_PER_USEC); 8990 do_div(rt_period_us, NSEC_PER_USEC);
8985 return rt_period_us; 8991 return rt_period_us;
8986 } 8992 }
8987 8993
8988 static int sched_rt_global_constraints(void) 8994 static int sched_rt_global_constraints(void)
8989 { 8995 {
8990 u64 runtime, period; 8996 u64 runtime, period;
8991 int ret = 0; 8997 int ret = 0;
8992 8998
8993 if (sysctl_sched_rt_period <= 0) 8999 if (sysctl_sched_rt_period <= 0)
8994 return -EINVAL; 9000 return -EINVAL;
8995 9001
8996 runtime = global_rt_runtime(); 9002 runtime = global_rt_runtime();
8997 period = global_rt_period(); 9003 period = global_rt_period();
8998 9004
8999 /* 9005 /*
9000 * Sanity check on the sysctl variables. 9006 * Sanity check on the sysctl variables.
9001 */ 9007 */
9002 if (runtime > period && runtime != RUNTIME_INF) 9008 if (runtime > period && runtime != RUNTIME_INF)
9003 return -EINVAL; 9009 return -EINVAL;
9004 9010
9005 mutex_lock(&rt_constraints_mutex); 9011 mutex_lock(&rt_constraints_mutex);
9006 read_lock(&tasklist_lock); 9012 read_lock(&tasklist_lock);
9007 ret = __rt_schedulable(NULL, 0, 0); 9013 ret = __rt_schedulable(NULL, 0, 0);
9008 read_unlock(&tasklist_lock); 9014 read_unlock(&tasklist_lock);
9009 mutex_unlock(&rt_constraints_mutex); 9015 mutex_unlock(&rt_constraints_mutex);
9010 9016
9011 return ret; 9017 return ret;
9012 } 9018 }
9013 #else /* !CONFIG_RT_GROUP_SCHED */ 9019 #else /* !CONFIG_RT_GROUP_SCHED */
9014 static int sched_rt_global_constraints(void) 9020 static int sched_rt_global_constraints(void)
9015 { 9021 {
9016 unsigned long flags; 9022 unsigned long flags;
9017 int i; 9023 int i;
9018 9024
9019 if (sysctl_sched_rt_period <= 0) 9025 if (sysctl_sched_rt_period <= 0)
9020 return -EINVAL; 9026 return -EINVAL;
9021 9027
9022 spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); 9028 spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags);
9023 for_each_possible_cpu(i) { 9029 for_each_possible_cpu(i) {
9024 struct rt_rq *rt_rq = &cpu_rq(i)->rt; 9030 struct rt_rq *rt_rq = &cpu_rq(i)->rt;
9025 9031
9026 spin_lock(&rt_rq->rt_runtime_lock); 9032 spin_lock(&rt_rq->rt_runtime_lock);
9027 rt_rq->rt_runtime = global_rt_runtime(); 9033 rt_rq->rt_runtime = global_rt_runtime();
9028 spin_unlock(&rt_rq->rt_runtime_lock); 9034 spin_unlock(&rt_rq->rt_runtime_lock);
9029 } 9035 }
9030 spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); 9036 spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags);
9031 9037
9032 return 0; 9038 return 0;
9033 } 9039 }
9034 #endif /* CONFIG_RT_GROUP_SCHED */ 9040 #endif /* CONFIG_RT_GROUP_SCHED */
9035 9041
9036 int sched_rt_handler(struct ctl_table *table, int write, 9042 int sched_rt_handler(struct ctl_table *table, int write,
9037 struct file *filp, void __user *buffer, size_t *lenp, 9043 struct file *filp, void __user *buffer, size_t *lenp,
9038 loff_t *ppos) 9044 loff_t *ppos)
9039 { 9045 {
9040 int ret; 9046 int ret;
9041 int old_period, old_runtime; 9047 int old_period, old_runtime;
9042 static DEFINE_MUTEX(mutex); 9048 static DEFINE_MUTEX(mutex);
9043 9049
9044 mutex_lock(&mutex); 9050 mutex_lock(&mutex);
9045 old_period = sysctl_sched_rt_period; 9051 old_period = sysctl_sched_rt_period;
9046 old_runtime = sysctl_sched_rt_runtime; 9052 old_runtime = sysctl_sched_rt_runtime;
9047 9053
9048 ret = proc_dointvec(table, write, filp, buffer, lenp, ppos); 9054 ret = proc_dointvec(table, write, filp, buffer, lenp, ppos);
9049 9055
9050 if (!ret && write) { 9056 if (!ret && write) {
9051 ret = sched_rt_global_constraints(); 9057 ret = sched_rt_global_constraints();
9052 if (ret) { 9058 if (ret) {
9053 sysctl_sched_rt_period = old_period; 9059 sysctl_sched_rt_period = old_period;
9054 sysctl_sched_rt_runtime = old_runtime; 9060 sysctl_sched_rt_runtime = old_runtime;
9055 } else { 9061 } else {
9056 def_rt_bandwidth.rt_runtime = global_rt_runtime(); 9062 def_rt_bandwidth.rt_runtime = global_rt_runtime();
9057 def_rt_bandwidth.rt_period = 9063 def_rt_bandwidth.rt_period =
9058 ns_to_ktime(global_rt_period()); 9064 ns_to_ktime(global_rt_period());
9059 } 9065 }
9060 } 9066 }
9061 mutex_unlock(&mutex); 9067 mutex_unlock(&mutex);
9062 9068
9063 return ret; 9069 return ret;
9064 } 9070 }
9065 9071
9066 #ifdef CONFIG_CGROUP_SCHED 9072 #ifdef CONFIG_CGROUP_SCHED
9067 9073
9068 /* return corresponding task_group object of a cgroup */ 9074 /* return corresponding task_group object of a cgroup */
9069 static inline struct task_group *cgroup_tg(struct cgroup *cgrp) 9075 static inline struct task_group *cgroup_tg(struct cgroup *cgrp)
9070 { 9076 {
9071 return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), 9077 return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id),
9072 struct task_group, css); 9078 struct task_group, css);
9073 } 9079 }
9074 9080
9075 static struct cgroup_subsys_state * 9081 static struct cgroup_subsys_state *
9076 cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) 9082 cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp)
9077 { 9083 {
9078 struct task_group *tg, *parent; 9084 struct task_group *tg, *parent;
9079 9085
9080 if (!cgrp->parent) { 9086 if (!cgrp->parent) {
9081 /* This is early initialization for the top cgroup */ 9087 /* This is early initialization for the top cgroup */
9082 return &init_task_group.css; 9088 return &init_task_group.css;
9083 } 9089 }
9084 9090
9085 parent = cgroup_tg(cgrp->parent); 9091 parent = cgroup_tg(cgrp->parent);
9086 tg = sched_create_group(parent); 9092 tg = sched_create_group(parent);
9087 if (IS_ERR(tg)) 9093 if (IS_ERR(tg))
9088 return ERR_PTR(-ENOMEM); 9094 return ERR_PTR(-ENOMEM);
9089 9095
9090 return &tg->css; 9096 return &tg->css;
9091 } 9097 }
9092 9098
9093 static void 9099 static void
9094 cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) 9100 cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp)
9095 { 9101 {
9096 struct task_group *tg = cgroup_tg(cgrp); 9102 struct task_group *tg = cgroup_tg(cgrp);
9097 9103
9098 sched_destroy_group(tg); 9104 sched_destroy_group(tg);
9099 } 9105 }
9100 9106
9101 static int 9107 static int
9102 cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, 9108 cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp,
9103 struct task_struct *tsk) 9109 struct task_struct *tsk)
9104 { 9110 {
9105 #ifdef CONFIG_RT_GROUP_SCHED 9111 #ifdef CONFIG_RT_GROUP_SCHED
9106 /* Don't accept realtime tasks when there is no way for them to run */ 9112 /* Don't accept realtime tasks when there is no way for them to run */
9107 if (rt_task(tsk) && cgroup_tg(cgrp)->rt_bandwidth.rt_runtime == 0) 9113 if (rt_task(tsk) && cgroup_tg(cgrp)->rt_bandwidth.rt_runtime == 0)
9108 return -EINVAL; 9114 return -EINVAL;
9109 #else 9115 #else
9110 /* We don't support RT-tasks being in separate groups */ 9116 /* We don't support RT-tasks being in separate groups */
9111 if (tsk->sched_class != &fair_sched_class) 9117 if (tsk->sched_class != &fair_sched_class)
9112 return -EINVAL; 9118 return -EINVAL;
9113 #endif 9119 #endif
9114 9120
9115 return 0; 9121 return 0;
9116 } 9122 }
9117 9123
9118 static void 9124 static void
9119 cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, 9125 cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp,
9120 struct cgroup *old_cont, struct task_struct *tsk) 9126 struct cgroup *old_cont, struct task_struct *tsk)
9121 { 9127 {
9122 sched_move_task(tsk); 9128 sched_move_task(tsk);
9123 } 9129 }
9124 9130
9125 #ifdef CONFIG_FAIR_GROUP_SCHED 9131 #ifdef CONFIG_FAIR_GROUP_SCHED
9126 static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, 9132 static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype,
9127 u64 shareval) 9133 u64 shareval)
9128 { 9134 {
9129 return sched_group_set_shares(cgroup_tg(cgrp), shareval); 9135 return sched_group_set_shares(cgroup_tg(cgrp), shareval);
9130 } 9136 }
9131 9137
9132 static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) 9138 static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft)
9133 { 9139 {
9134 struct task_group *tg = cgroup_tg(cgrp); 9140 struct task_group *tg = cgroup_tg(cgrp);
9135 9141
9136 return (u64) tg->shares; 9142 return (u64) tg->shares;
9137 } 9143 }
9138 #endif /* CONFIG_FAIR_GROUP_SCHED */ 9144 #endif /* CONFIG_FAIR_GROUP_SCHED */
9139 9145
9140 #ifdef CONFIG_RT_GROUP_SCHED 9146 #ifdef CONFIG_RT_GROUP_SCHED
9141 static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, 9147 static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft,
9142 s64 val) 9148 s64 val)
9143 { 9149 {
9144 return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); 9150 return sched_group_set_rt_runtime(cgroup_tg(cgrp), val);
9145 } 9151 }
9146 9152
9147 static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) 9153 static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft)
9148 { 9154 {
9149 return sched_group_rt_runtime(cgroup_tg(cgrp)); 9155 return sched_group_rt_runtime(cgroup_tg(cgrp));
9150 } 9156 }
9151 9157
9152 static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, 9158 static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype,
9153 u64 rt_period_us) 9159 u64 rt_period_us)
9154 { 9160 {
9155 return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); 9161 return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us);
9156 } 9162 }
9157 9163
9158 static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) 9164 static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft)
9159 { 9165 {
9160 return sched_group_rt_period(cgroup_tg(cgrp)); 9166 return sched_group_rt_period(cgroup_tg(cgrp));
9161 } 9167 }
9162 #endif /* CONFIG_RT_GROUP_SCHED */ 9168 #endif /* CONFIG_RT_GROUP_SCHED */
9163 9169
9164 static struct cftype cpu_files[] = { 9170 static struct cftype cpu_files[] = {
9165 #ifdef CONFIG_FAIR_GROUP_SCHED 9171 #ifdef CONFIG_FAIR_GROUP_SCHED
9166 { 9172 {
9167 .name = "shares", 9173 .name = "shares",
9168 .read_u64 = cpu_shares_read_u64, 9174 .read_u64 = cpu_shares_read_u64,
9169 .write_u64 = cpu_shares_write_u64, 9175 .write_u64 = cpu_shares_write_u64,
9170 }, 9176 },
9171 #endif 9177 #endif
9172 #ifdef CONFIG_RT_GROUP_SCHED 9178 #ifdef CONFIG_RT_GROUP_SCHED
9173 { 9179 {
9174 .name = "rt_runtime_us", 9180 .name = "rt_runtime_us",
9175 .read_s64 = cpu_rt_runtime_read, 9181 .read_s64 = cpu_rt_runtime_read,
9176 .write_s64 = cpu_rt_runtime_write, 9182 .write_s64 = cpu_rt_runtime_write,
9177 }, 9183 },
9178 { 9184 {
9179 .name = "rt_period_us", 9185 .name = "rt_period_us",
9180 .read_u64 = cpu_rt_period_read_uint, 9186 .read_u64 = cpu_rt_period_read_uint,
9181 .write_u64 = cpu_rt_period_write_uint, 9187 .write_u64 = cpu_rt_period_write_uint,
9182 }, 9188 },
9183 #endif 9189 #endif
9184 }; 9190 };
9185 9191
9186 static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) 9192 static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont)
9187 { 9193 {
9188 return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); 9194 return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files));
9189 } 9195 }
9190 9196
9191 struct cgroup_subsys cpu_cgroup_subsys = { 9197 struct cgroup_subsys cpu_cgroup_subsys = {
9192 .name = "cpu", 9198 .name = "cpu",
9193 .create = cpu_cgroup_create, 9199 .create = cpu_cgroup_create,
9194 .destroy = cpu_cgroup_destroy, 9200 .destroy = cpu_cgroup_destroy,
9195 .can_attach = cpu_cgroup_can_attach, 9201 .can_attach = cpu_cgroup_can_attach,
9196 .attach = cpu_cgroup_attach, 9202 .attach = cpu_cgroup_attach,
9197 .populate = cpu_cgroup_populate, 9203 .populate = cpu_cgroup_populate,
9198 .subsys_id = cpu_cgroup_subsys_id, 9204 .subsys_id = cpu_cgroup_subsys_id,
9199 .early_init = 1, 9205 .early_init = 1,
9200 }; 9206 };
9201 9207
9202 #endif /* CONFIG_CGROUP_SCHED */ 9208 #endif /* CONFIG_CGROUP_SCHED */
9203 9209
9204 #ifdef CONFIG_CGROUP_CPUACCT 9210 #ifdef CONFIG_CGROUP_CPUACCT
9205 9211
9206 /* 9212 /*
9207 * CPU accounting code for task groups. 9213 * CPU accounting code for task groups.
9208 * 9214 *
9209 * Based on the work by Paul Menage (menage@google.com) and Balbir Singh 9215 * Based on the work by Paul Menage (menage@google.com) and Balbir Singh
9210 * (balbir@in.ibm.com). 9216 * (balbir@in.ibm.com).
9211 */ 9217 */
9212 9218
9213 /* track cpu usage of a group of tasks and its child groups */ 9219 /* track cpu usage of a group of tasks and its child groups */
9214 struct cpuacct { 9220 struct cpuacct {
9215 struct cgroup_subsys_state css; 9221 struct cgroup_subsys_state css;
9216 /* cpuusage holds pointer to a u64-type object on every cpu */ 9222 /* cpuusage holds pointer to a u64-type object on every cpu */
9217 u64 *cpuusage; 9223 u64 *cpuusage;
9218 struct cpuacct *parent; 9224 struct cpuacct *parent;
9219 }; 9225 };
9220 9226
9221 struct cgroup_subsys cpuacct_subsys; 9227 struct cgroup_subsys cpuacct_subsys;
9222 9228
9223 /* return cpu accounting group corresponding to this container */ 9229 /* return cpu accounting group corresponding to this container */
9224 static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp) 9230 static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp)
9225 { 9231 {
9226 return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id), 9232 return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id),
9227 struct cpuacct, css); 9233 struct cpuacct, css);
9228 } 9234 }
9229 9235
9230 /* return cpu accounting group to which this task belongs */ 9236 /* return cpu accounting group to which this task belongs */
9231 static inline struct cpuacct *task_ca(struct task_struct *tsk) 9237 static inline struct cpuacct *task_ca(struct task_struct *tsk)
9232 { 9238 {
9233 return container_of(task_subsys_state(tsk, cpuacct_subsys_id), 9239 return container_of(task_subsys_state(tsk, cpuacct_subsys_id),
9234 struct cpuacct, css); 9240 struct cpuacct, css);
9235 } 9241 }
9236 9242
9237 /* create a new cpu accounting group */ 9243 /* create a new cpu accounting group */
9238 static struct cgroup_subsys_state *cpuacct_create( 9244 static struct cgroup_subsys_state *cpuacct_create(
9239 struct cgroup_subsys *ss, struct cgroup *cgrp) 9245 struct cgroup_subsys *ss, struct cgroup *cgrp)
9240 { 9246 {
9241 struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); 9247 struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL);
9242 9248
9243 if (!ca) 9249 if (!ca)
9244 return ERR_PTR(-ENOMEM); 9250 return ERR_PTR(-ENOMEM);
9245 9251
9246 ca->cpuusage = alloc_percpu(u64); 9252 ca->cpuusage = alloc_percpu(u64);
9247 if (!ca->cpuusage) { 9253 if (!ca->cpuusage) {
9248 kfree(ca); 9254 kfree(ca);
9249 return ERR_PTR(-ENOMEM); 9255 return ERR_PTR(-ENOMEM);
9250 } 9256 }
9251 9257
9252 if (cgrp->parent) 9258 if (cgrp->parent)
9253 ca->parent = cgroup_ca(cgrp->parent); 9259 ca->parent = cgroup_ca(cgrp->parent);
9254 9260
9255 return &ca->css; 9261 return &ca->css;
9256 } 9262 }
9257 9263
9258 /* destroy an existing cpu accounting group */ 9264 /* destroy an existing cpu accounting group */
9259 static void 9265 static void
9260 cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) 9266 cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp)
9261 { 9267 {
9262 struct cpuacct *ca = cgroup_ca(cgrp); 9268 struct cpuacct *ca = cgroup_ca(cgrp);
9263 9269
9264 free_percpu(ca->cpuusage); 9270 free_percpu(ca->cpuusage);
9265 kfree(ca); 9271 kfree(ca);
9266 } 9272 }
9267 9273
9268 /* return total cpu usage (in nanoseconds) of a group */ 9274 /* return total cpu usage (in nanoseconds) of a group */
9269 static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) 9275 static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft)
9270 { 9276 {
9271 struct cpuacct *ca = cgroup_ca(cgrp); 9277 struct cpuacct *ca = cgroup_ca(cgrp);
9272 u64 totalcpuusage = 0; 9278 u64 totalcpuusage = 0;
9273 int i; 9279 int i;
9274 9280
9275 for_each_possible_cpu(i) { 9281 for_each_possible_cpu(i) {
9276 u64 *cpuusage = percpu_ptr(ca->cpuusage, i); 9282 u64 *cpuusage = percpu_ptr(ca->cpuusage, i);
9277 9283
9278 /* 9284 /*
9279 * Take rq->lock to make 64-bit addition safe on 32-bit 9285 * Take rq->lock to make 64-bit addition safe on 32-bit
9280 * platforms. 9286 * platforms.
9281 */ 9287 */
9282 spin_lock_irq(&cpu_rq(i)->lock); 9288 spin_lock_irq(&cpu_rq(i)->lock);
9283 totalcpuusage += *cpuusage; 9289 totalcpuusage += *cpuusage;
9284 spin_unlock_irq(&cpu_rq(i)->lock); 9290 spin_unlock_irq(&cpu_rq(i)->lock);
9285 } 9291 }
9286 9292
9287 return totalcpuusage; 9293 return totalcpuusage;
9288 } 9294 }
9289 9295
9290 static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, 9296 static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype,
9291 u64 reset) 9297 u64 reset)
9292 { 9298 {
9293 struct cpuacct *ca = cgroup_ca(cgrp); 9299 struct cpuacct *ca = cgroup_ca(cgrp);
9294 int err = 0; 9300 int err = 0;
9295 int i; 9301 int i;
9296 9302
9297 if (reset) { 9303 if (reset) {
9298 err = -EINVAL; 9304 err = -EINVAL;
9299 goto out; 9305 goto out;
9300 } 9306 }
9301 9307
9302 for_each_possible_cpu(i) { 9308 for_each_possible_cpu(i) {
9303 u64 *cpuusage = percpu_ptr(ca->cpuusage, i); 9309 u64 *cpuusage = percpu_ptr(ca->cpuusage, i);
9304 9310
9305 spin_lock_irq(&cpu_rq(i)->lock); 9311 spin_lock_irq(&cpu_rq(i)->lock);
9306 *cpuusage = 0; 9312 *cpuusage = 0;
9307 spin_unlock_irq(&cpu_rq(i)->lock); 9313 spin_unlock_irq(&cpu_rq(i)->lock);
9308 } 9314 }
9309 out: 9315 out:
9310 return err; 9316 return err;
9311 } 9317 }
9312 9318
9313 static struct cftype files[] = { 9319 static struct cftype files[] = {
9314 { 9320 {
9315 .name = "usage", 9321 .name = "usage",
9316 .read_u64 = cpuusage_read, 9322 .read_u64 = cpuusage_read,
9317 .write_u64 = cpuusage_write, 9323 .write_u64 = cpuusage_write,
9318 }, 9324 },
9319 }; 9325 };
9320 9326
9321 static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) 9327 static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp)
9322 { 9328 {
9323 return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files)); 9329 return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files));
9324 } 9330 }
9325 9331
9326 /* 9332 /*
9327 * charge this task's execution time to its accounting group. 9333 * charge this task's execution time to its accounting group.
9328 * 9334 *
9329 * called with rq->lock held. 9335 * called with rq->lock held.
9330 */ 9336 */
9331 static void cpuacct_charge(struct task_struct *tsk, u64 cputime) 9337 static void cpuacct_charge(struct task_struct *tsk, u64 cputime)
9332 { 9338 {
9333 struct cpuacct *ca; 9339 struct cpuacct *ca;
9334 int cpu; 9340 int cpu;
9335 9341
9336 if (!cpuacct_subsys.active) 9342 if (!cpuacct_subsys.active)
9337 return; 9343 return;
9338 9344
9339 cpu = task_cpu(tsk); 9345 cpu = task_cpu(tsk);
9340 ca = task_ca(tsk); 9346 ca = task_ca(tsk);
9341 9347
9342 for (; ca; ca = ca->parent) { 9348 for (; ca; ca = ca->parent) {
9343 u64 *cpuusage = percpu_ptr(ca->cpuusage, cpu); 9349 u64 *cpuusage = percpu_ptr(ca->cpuusage, cpu);
9344 *cpuusage += cputime; 9350 *cpuusage += cputime;
9345 } 9351 }
9346 } 9352 }
9347 9353
9348 struct cgroup_subsys cpuacct_subsys = { 9354 struct cgroup_subsys cpuacct_subsys = {
9349 .name = "cpuacct", 9355 .name = "cpuacct",
9350 .create = cpuacct_create, 9356 .create = cpuacct_create,
9351 .destroy = cpuacct_destroy, 9357 .destroy = cpuacct_destroy,
9352 .populate = cpuacct_populate, 9358 .populate = cpuacct_populate,
9353 .subsys_id = cpuacct_subsys_id, 9359 .subsys_id = cpuacct_subsys_id,
9354 }; 9360 };
9355 #endif /* CONFIG_CGROUP_CPUACCT */ 9361 #endif /* CONFIG_CGROUP_CPUACCT */
9356 9362