Eric Lee / smarc-fsl-linux-kernel

Download as
Browse Code ยป

Commit 61f0c1dcaaac71faabac6ef7c839b29f20204bea

Authored by Jens Axboe
1 parent 963b72fc66
Exists in master and in 39 other branches 8mp-imx_5.4.70_2.3.0, 8qm-imx_5.4.70_2.3.0, emb_imx_lf-5.15.y, emb_lf-6.1.y, 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, imx_4.1.15_1.0.0_ga, pitx_8mp_lf-5.10.y, rt-smarc-imx_4.1.15_1.0.0_ga, rt_linux_5.15.71, smarc-8m-android-11.0.0_2.0.0, smarc-imx6_4.14.98_2.0.0_ga, smarc-imx6_4.9.88_2.0.0_ga, smarc-imx7_4.14.98_2.0.0_ga, smarc-imx7_4.9.11_1.0.0_ga, smarc-imx7_4.9.88_2.0.0_ga, smarc-imx_3.10.53_1.1.0_ga, smarc-imx_3.14.28_1.0.0_ga, smarc-imx_4.1.15_1.0.0_ga, smarc-imx_4.9.11_1.0.0_ga, smarc-imx_4.9.51_imx8m_ga, smarc-imx_4.9.88_2.0.0_ga, smarc-m6.0.1_2.1.0-ga, smarc-n7.1.2_2.0.0-ga, smarc-rel_imx_4.1.15_1.2.0_ga, smarc_8m_00d0_imx_4.14.98_2.0.0_ga, smarc_8m_imx_4.14.78_1.0.0_ga, smarc_8m_imx_4.14.98_2.0.0_ga, smarc_8m_imx_4.19.35_1.1.0, smarc_8mm_imx_4.14.78_1.0.0_ga, smarc_8mm_imx_4.14.98_2.0.0_ga, smarc_8mm_imx_4.19.35_1.1.0, smarc_8mm_imx_5.4.24_2.1.0, smarc_8mp_lf-5.10.y, smarc_8mq_imx_5.4.24_2.1.0, smarc_8mq_lf-5.10.y, smarc_imx_lf-5.15.y

cfq-iosched: use assigned slice sync value, not default 

We should use the sysfs modified slice sync value, in case it differs
from the default.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

Showing 1 changed file with 4 additions and 3 deletions Inline Diff

1 /* 1 /*
2 * CFQ, or complete fairness queueing, disk scheduler. 2 * CFQ, or complete fairness queueing, disk scheduler.
3 * 3 *
4 * Based on ideas from a previously unfinished io 4 * Based on ideas from a previously unfinished io
5 * scheduler (round robin per-process disk scheduling) and Andrea Arcangeli. 5 * scheduler (round robin per-process disk scheduling) and Andrea Arcangeli.
6 * 6 *
7 * Copyright (C) 2003 Jens Axboe <axboe@kernel.dk> 7 * Copyright (C) 2003 Jens Axboe <axboe@kernel.dk>
8 */ 8 */
9 #include <linux/module.h> 9 #include <linux/module.h>
10 #include <linux/blkdev.h> 10 #include <linux/blkdev.h>
11 #include <linux/elevator.h> 11 #include <linux/elevator.h>
12 #include <linux/rbtree.h> 12 #include <linux/rbtree.h>
13 #include <linux/ioprio.h> 13 #include <linux/ioprio.h>
14 #include <linux/blktrace_api.h> 14 #include <linux/blktrace_api.h>
15 15
16 /* 16 /*
17 * tunables 17 * tunables
18 */ 18 */
19 /* max queue in one round of service */ 19 /* max queue in one round of service */
20 static const int cfq_quantum = 4; 20 static const int cfq_quantum = 4;
21 static const int cfq_fifo_expire[2] = { HZ / 4, HZ / 8 }; 21 static const int cfq_fifo_expire[2] = { HZ / 4, HZ / 8 };
22 /* maximum backwards seek, in KiB */ 22 /* maximum backwards seek, in KiB */
23 static const int cfq_back_max = 16 * 1024; 23 static const int cfq_back_max = 16 * 1024;
24 /* penalty of a backwards seek */ 24 /* penalty of a backwards seek */
25 static const int cfq_back_penalty = 2; 25 static const int cfq_back_penalty = 2;
26 static const int cfq_slice_sync = HZ / 10; 26 static const int cfq_slice_sync = HZ / 10;
27 static int cfq_slice_async = HZ / 25; 27 static int cfq_slice_async = HZ / 25;
28 static const int cfq_slice_async_rq = 2; 28 static const int cfq_slice_async_rq = 2;
29 static int cfq_slice_idle = HZ / 125; 29 static int cfq_slice_idle = HZ / 125;
30 30
31 /* 31 /*
32 * offset from end of service tree 32 * offset from end of service tree
33 */ 33 */
34 #define CFQ_IDLE_DELAY (HZ / 5) 34 #define CFQ_IDLE_DELAY (HZ / 5)
35 35
36 /* 36 /*
37 * below this threshold, we consider thinktime immediate 37 * below this threshold, we consider thinktime immediate
38 */ 38 */
39 #define CFQ_MIN_TT (2) 39 #define CFQ_MIN_TT (2)
40 40
41 #define CFQ_SLICE_SCALE (5) 41 #define CFQ_SLICE_SCALE (5)
42 #define CFQ_HW_QUEUE_MIN (5) 42 #define CFQ_HW_QUEUE_MIN (5)
43 43
44 #define RQ_CIC(rq) \ 44 #define RQ_CIC(rq) \
45 ((struct cfq_io_context *) (rq)->elevator_private) 45 ((struct cfq_io_context *) (rq)->elevator_private)
46 #define RQ_CFQQ(rq) (struct cfq_queue *) ((rq)->elevator_private2) 46 #define RQ_CFQQ(rq) (struct cfq_queue *) ((rq)->elevator_private2)
47 47
48 static struct kmem_cache *cfq_pool; 48 static struct kmem_cache *cfq_pool;
49 static struct kmem_cache *cfq_ioc_pool; 49 static struct kmem_cache *cfq_ioc_pool;
50 50
51 static DEFINE_PER_CPU(unsigned long, cfq_ioc_count); 51 static DEFINE_PER_CPU(unsigned long, cfq_ioc_count);
52 static struct completion *ioc_gone; 52 static struct completion *ioc_gone;
53 static DEFINE_SPINLOCK(ioc_gone_lock); 53 static DEFINE_SPINLOCK(ioc_gone_lock);
54 54
55 #define CFQ_PRIO_LISTS IOPRIO_BE_NR 55 #define CFQ_PRIO_LISTS IOPRIO_BE_NR
56 #define cfq_class_idle(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_IDLE) 56 #define cfq_class_idle(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_IDLE)
57 #define cfq_class_rt(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_RT) 57 #define cfq_class_rt(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_RT)
58 58
59 #define sample_valid(samples) ((samples) > 80) 59 #define sample_valid(samples) ((samples) > 80)
60 60
61 /* 61 /*
62 * Most of our rbtree usage is for sorting with min extraction, so 62 * Most of our rbtree usage is for sorting with min extraction, so
63 * if we cache the leftmost node we don't have to walk down the tree 63 * if we cache the leftmost node we don't have to walk down the tree
64 * to find it. Idea borrowed from Ingo Molnars CFS scheduler. We should 64 * to find it. Idea borrowed from Ingo Molnars CFS scheduler. We should
65 * move this into the elevator for the rq sorting as well. 65 * move this into the elevator for the rq sorting as well.
66 */ 66 */
67 struct cfq_rb_root { 67 struct cfq_rb_root {
68 struct rb_root rb; 68 struct rb_root rb;
69 struct rb_node *left; 69 struct rb_node *left;
70 }; 70 };
71 #define CFQ_RB_ROOT (struct cfq_rb_root) { RB_ROOT, NULL, } 71 #define CFQ_RB_ROOT (struct cfq_rb_root) { RB_ROOT, NULL, }
72 72
73 /* 73 /*
74 * Per process-grouping structure 74 * Per process-grouping structure
75 */ 75 */
76 struct cfq_queue { 76 struct cfq_queue {
77 /* reference count */ 77 /* reference count */
78 atomic_t ref; 78 atomic_t ref;
79 /* various state flags, see below */ 79 /* various state flags, see below */
80 unsigned int flags; 80 unsigned int flags;
81 /* parent cfq_data */ 81 /* parent cfq_data */
82 struct cfq_data *cfqd; 82 struct cfq_data *cfqd;
83 /* service_tree member */ 83 /* service_tree member */
84 struct rb_node rb_node; 84 struct rb_node rb_node;
85 /* service_tree key */ 85 /* service_tree key */
86 unsigned long rb_key; 86 unsigned long rb_key;
87 /* prio tree member */ 87 /* prio tree member */
88 struct rb_node p_node; 88 struct rb_node p_node;
89 /* prio tree root we belong to, if any */ 89 /* prio tree root we belong to, if any */
90 struct rb_root *p_root; 90 struct rb_root *p_root;
91 /* sorted list of pending requests */ 91 /* sorted list of pending requests */
92 struct rb_root sort_list; 92 struct rb_root sort_list;
93 /* if fifo isn't expired, next request to serve */ 93 /* if fifo isn't expired, next request to serve */
94 struct request *next_rq; 94 struct request *next_rq;
95 /* requests queued in sort_list */ 95 /* requests queued in sort_list */
96 int queued[2]; 96 int queued[2];
97 /* currently allocated requests */ 97 /* currently allocated requests */
98 int allocated[2]; 98 int allocated[2];
99 /* fifo list of requests in sort_list */ 99 /* fifo list of requests in sort_list */
100 struct list_head fifo; 100 struct list_head fifo;
101 101
102 unsigned long slice_end; 102 unsigned long slice_end;
103 long slice_resid; 103 long slice_resid;
104 unsigned int slice_dispatch; 104 unsigned int slice_dispatch;
105 105
106 /* pending metadata requests */ 106 /* pending metadata requests */
107 int meta_pending; 107 int meta_pending;
108 /* number of requests that are on the dispatch list or inside driver */ 108 /* number of requests that are on the dispatch list or inside driver */
109 int dispatched; 109 int dispatched;
110 110
111 /* io prio of this group */ 111 /* io prio of this group */
112 unsigned short ioprio, org_ioprio; 112 unsigned short ioprio, org_ioprio;
113 unsigned short ioprio_class, org_ioprio_class; 113 unsigned short ioprio_class, org_ioprio_class;
114 114
115 pid_t pid; 115 pid_t pid;
116 }; 116 };
117 117
118 /* 118 /*
119 * Per block device queue structure 119 * Per block device queue structure
120 */ 120 */
121 struct cfq_data { 121 struct cfq_data {
122 struct request_queue *queue; 122 struct request_queue *queue;
123 123
124 /* 124 /*
125 * rr list of queues with requests and the count of them 125 * rr list of queues with requests and the count of them
126 */ 126 */
127 struct cfq_rb_root service_tree; 127 struct cfq_rb_root service_tree;
128 128
129 /* 129 /*
130 * Each priority tree is sorted by next_request position. These 130 * Each priority tree is sorted by next_request position. These
131 * trees are used when determining if two or more queues are 131 * trees are used when determining if two or more queues are
132 * interleaving requests (see cfq_close_cooperator). 132 * interleaving requests (see cfq_close_cooperator).
133 */ 133 */
134 struct rb_root prio_trees[CFQ_PRIO_LISTS]; 134 struct rb_root prio_trees[CFQ_PRIO_LISTS];
135 135
136 unsigned int busy_queues; 136 unsigned int busy_queues;
137 137
138 int rq_in_driver[2]; 138 int rq_in_driver[2];
139 int sync_flight; 139 int sync_flight;
140 140
141 /* 141 /*
142 * queue-depth detection 142 * queue-depth detection
143 */ 143 */
144 int rq_queued; 144 int rq_queued;
145 int hw_tag; 145 int hw_tag;
146 int hw_tag_samples; 146 int hw_tag_samples;
147 int rq_in_driver_peak; 147 int rq_in_driver_peak;
148 148
149 /* 149 /*
150 * idle window management 150 * idle window management
151 */ 151 */
152 struct timer_list idle_slice_timer; 152 struct timer_list idle_slice_timer;
153 struct delayed_work unplug_work; 153 struct delayed_work unplug_work;
154 154
155 struct cfq_queue *active_queue; 155 struct cfq_queue *active_queue;
156 struct cfq_io_context *active_cic; 156 struct cfq_io_context *active_cic;
157 157
158 /* 158 /*
159 * async queue for each priority case 159 * async queue for each priority case
160 */ 160 */
161 struct cfq_queue *async_cfqq[2][IOPRIO_BE_NR]; 161 struct cfq_queue *async_cfqq[2][IOPRIO_BE_NR];
162 struct cfq_queue *async_idle_cfqq; 162 struct cfq_queue *async_idle_cfqq;
163 163
164 sector_t last_position; 164 sector_t last_position;
165 165
166 /* 166 /*
167 * tunables, see top of file 167 * tunables, see top of file
168 */ 168 */
169 unsigned int cfq_quantum; 169 unsigned int cfq_quantum;
170 unsigned int cfq_fifo_expire[2]; 170 unsigned int cfq_fifo_expire[2];
171 unsigned int cfq_back_penalty; 171 unsigned int cfq_back_penalty;
172 unsigned int cfq_back_max; 172 unsigned int cfq_back_max;
173 unsigned int cfq_slice[2]; 173 unsigned int cfq_slice[2];
174 unsigned int cfq_slice_async_rq; 174 unsigned int cfq_slice_async_rq;
175 unsigned int cfq_slice_idle; 175 unsigned int cfq_slice_idle;
176 unsigned int cfq_latency; 176 unsigned int cfq_latency;
177 177
178 struct list_head cic_list; 178 struct list_head cic_list;
179 179
180 /* 180 /*
181 * Fallback dummy cfqq for extreme OOM conditions 181 * Fallback dummy cfqq for extreme OOM conditions
182 */ 182 */
183 struct cfq_queue oom_cfqq; 183 struct cfq_queue oom_cfqq;
184 184
185 unsigned long last_end_sync_rq; 185 unsigned long last_end_sync_rq;
186 }; 186 };
187 187
188 enum cfqq_state_flags { 188 enum cfqq_state_flags {
189 CFQ_CFQQ_FLAG_on_rr = 0, /* on round-robin busy list */ 189 CFQ_CFQQ_FLAG_on_rr = 0, /* on round-robin busy list */
190 CFQ_CFQQ_FLAG_wait_request, /* waiting for a request */ 190 CFQ_CFQQ_FLAG_wait_request, /* waiting for a request */
191 CFQ_CFQQ_FLAG_must_dispatch, /* must be allowed a dispatch */ 191 CFQ_CFQQ_FLAG_must_dispatch, /* must be allowed a dispatch */
192 CFQ_CFQQ_FLAG_must_alloc_slice, /* per-slice must_alloc flag */ 192 CFQ_CFQQ_FLAG_must_alloc_slice, /* per-slice must_alloc flag */
193 CFQ_CFQQ_FLAG_fifo_expire, /* FIFO checked in this slice */ 193 CFQ_CFQQ_FLAG_fifo_expire, /* FIFO checked in this slice */
194 CFQ_CFQQ_FLAG_idle_window, /* slice idling enabled */ 194 CFQ_CFQQ_FLAG_idle_window, /* slice idling enabled */
195 CFQ_CFQQ_FLAG_prio_changed, /* task priority has changed */ 195 CFQ_CFQQ_FLAG_prio_changed, /* task priority has changed */
196 CFQ_CFQQ_FLAG_slice_new, /* no requests dispatched in slice */ 196 CFQ_CFQQ_FLAG_slice_new, /* no requests dispatched in slice */
197 CFQ_CFQQ_FLAG_sync, /* synchronous queue */ 197 CFQ_CFQQ_FLAG_sync, /* synchronous queue */
198 CFQ_CFQQ_FLAG_coop, /* has done a coop jump of the queue */ 198 CFQ_CFQQ_FLAG_coop, /* has done a coop jump of the queue */
199 }; 199 };
200 200
201 #define CFQ_CFQQ_FNS(name) \ 201 #define CFQ_CFQQ_FNS(name) \
202 static inline void cfq_mark_cfqq_##name(struct cfq_queue *cfqq) \ 202 static inline void cfq_mark_cfqq_##name(struct cfq_queue *cfqq) \
203 { \ 203 { \
204 (cfqq)->flags |= (1 << CFQ_CFQQ_FLAG_##name); \ 204 (cfqq)->flags |= (1 << CFQ_CFQQ_FLAG_##name); \
205 } \ 205 } \
206 static inline void cfq_clear_cfqq_##name(struct cfq_queue *cfqq) \ 206 static inline void cfq_clear_cfqq_##name(struct cfq_queue *cfqq) \
207 { \ 207 { \
208 (cfqq)->flags &= ~(1 << CFQ_CFQQ_FLAG_##name); \ 208 (cfqq)->flags &= ~(1 << CFQ_CFQQ_FLAG_##name); \
209 } \ 209 } \
210 static inline int cfq_cfqq_##name(const struct cfq_queue *cfqq) \ 210 static inline int cfq_cfqq_##name(const struct cfq_queue *cfqq) \
211 { \ 211 { \
212 return ((cfqq)->flags & (1 << CFQ_CFQQ_FLAG_##name)) != 0; \ 212 return ((cfqq)->flags & (1 << CFQ_CFQQ_FLAG_##name)) != 0; \
213 } 213 }
214 214
215 CFQ_CFQQ_FNS(on_rr); 215 CFQ_CFQQ_FNS(on_rr);
216 CFQ_CFQQ_FNS(wait_request); 216 CFQ_CFQQ_FNS(wait_request);
217 CFQ_CFQQ_FNS(must_dispatch); 217 CFQ_CFQQ_FNS(must_dispatch);
218 CFQ_CFQQ_FNS(must_alloc_slice); 218 CFQ_CFQQ_FNS(must_alloc_slice);
219 CFQ_CFQQ_FNS(fifo_expire); 219 CFQ_CFQQ_FNS(fifo_expire);
220 CFQ_CFQQ_FNS(idle_window); 220 CFQ_CFQQ_FNS(idle_window);
221 CFQ_CFQQ_FNS(prio_changed); 221 CFQ_CFQQ_FNS(prio_changed);
222 CFQ_CFQQ_FNS(slice_new); 222 CFQ_CFQQ_FNS(slice_new);
223 CFQ_CFQQ_FNS(sync); 223 CFQ_CFQQ_FNS(sync);
224 CFQ_CFQQ_FNS(coop); 224 CFQ_CFQQ_FNS(coop);
225 #undef CFQ_CFQQ_FNS 225 #undef CFQ_CFQQ_FNS
226 226
227 #define cfq_log_cfqq(cfqd, cfqq, fmt, args...) \ 227 #define cfq_log_cfqq(cfqd, cfqq, fmt, args...) \
228 blk_add_trace_msg((cfqd)->queue, "cfq%d " fmt, (cfqq)->pid, ##args) 228 blk_add_trace_msg((cfqd)->queue, "cfq%d " fmt, (cfqq)->pid, ##args)
229 #define cfq_log(cfqd, fmt, args...) \ 229 #define cfq_log(cfqd, fmt, args...) \
230 blk_add_trace_msg((cfqd)->queue, "cfq " fmt, ##args) 230 blk_add_trace_msg((cfqd)->queue, "cfq " fmt, ##args)
231 231
232 static void cfq_dispatch_insert(struct request_queue *, struct request *); 232 static void cfq_dispatch_insert(struct request_queue *, struct request *);
233 static struct cfq_queue *cfq_get_queue(struct cfq_data *, int, 233 static struct cfq_queue *cfq_get_queue(struct cfq_data *, int,
234 struct io_context *, gfp_t); 234 struct io_context *, gfp_t);
235 static struct cfq_io_context *cfq_cic_lookup(struct cfq_data *, 235 static struct cfq_io_context *cfq_cic_lookup(struct cfq_data *,
236 struct io_context *); 236 struct io_context *);
237 237
238 static inline int rq_in_driver(struct cfq_data *cfqd) 238 static inline int rq_in_driver(struct cfq_data *cfqd)
239 { 239 {
240 return cfqd->rq_in_driver[0] + cfqd->rq_in_driver[1]; 240 return cfqd->rq_in_driver[0] + cfqd->rq_in_driver[1];
241 } 241 }
242 242
243 static inline struct cfq_queue *cic_to_cfqq(struct cfq_io_context *cic, 243 static inline struct cfq_queue *cic_to_cfqq(struct cfq_io_context *cic,
244 int is_sync) 244 int is_sync)
245 { 245 {
246 return cic->cfqq[!!is_sync]; 246 return cic->cfqq[!!is_sync];
247 } 247 }
248 248
249 static inline void cic_set_cfqq(struct cfq_io_context *cic, 249 static inline void cic_set_cfqq(struct cfq_io_context *cic,
250 struct cfq_queue *cfqq, int is_sync) 250 struct cfq_queue *cfqq, int is_sync)
251 { 251 {
252 cic->cfqq[!!is_sync] = cfqq; 252 cic->cfqq[!!is_sync] = cfqq;
253 } 253 }
254 254
255 /* 255 /*
256 * We regard a request as SYNC, if it's either a read or has the SYNC bit 256 * We regard a request as SYNC, if it's either a read or has the SYNC bit
257 * set (in which case it could also be direct WRITE). 257 * set (in which case it could also be direct WRITE).
258 */ 258 */
259 static inline int cfq_bio_sync(struct bio *bio) 259 static inline int cfq_bio_sync(struct bio *bio)
260 { 260 {
261 if (bio_data_dir(bio) == READ || bio_rw_flagged(bio, BIO_RW_SYNCIO)) 261 if (bio_data_dir(bio) == READ || bio_rw_flagged(bio, BIO_RW_SYNCIO))
262 return 1; 262 return 1;
263 263
264 return 0; 264 return 0;
265 } 265 }
266 266
267 /* 267 /*
268 * scheduler run of queue, if there are requests pending and no one in the 268 * scheduler run of queue, if there are requests pending and no one in the
269 * driver that will restart queueing 269 * driver that will restart queueing
270 */ 270 */
271 static inline void cfq_schedule_dispatch(struct cfq_data *cfqd, 271 static inline void cfq_schedule_dispatch(struct cfq_data *cfqd,
272 unsigned long delay) 272 unsigned long delay)
273 { 273 {
274 if (cfqd->busy_queues) { 274 if (cfqd->busy_queues) {
275 cfq_log(cfqd, "schedule dispatch"); 275 cfq_log(cfqd, "schedule dispatch");
276 kblockd_schedule_delayed_work(cfqd->queue, &cfqd->unplug_work, 276 kblockd_schedule_delayed_work(cfqd->queue, &cfqd->unplug_work,
277 delay); 277 delay);
278 } 278 }
279 } 279 }
280 280
281 static int cfq_queue_empty(struct request_queue *q) 281 static int cfq_queue_empty(struct request_queue *q)
282 { 282 {
283 struct cfq_data *cfqd = q->elevator->elevator_data; 283 struct cfq_data *cfqd = q->elevator->elevator_data;
284 284
285 return !cfqd->busy_queues; 285 return !cfqd->busy_queues;
286 } 286 }
287 287
288 /* 288 /*
289 * Scale schedule slice based on io priority. Use the sync time slice only 289 * Scale schedule slice based on io priority. Use the sync time slice only
290 * if a queue is marked sync and has sync io queued. A sync queue with async 290 * if a queue is marked sync and has sync io queued. A sync queue with async
291 * io only, should not get full sync slice length. 291 * io only, should not get full sync slice length.
292 */ 292 */
293 static inline int cfq_prio_slice(struct cfq_data *cfqd, int sync, 293 static inline int cfq_prio_slice(struct cfq_data *cfqd, int sync,
294 unsigned short prio) 294 unsigned short prio)
295 { 295 {
296 const int base_slice = cfqd->cfq_slice[sync]; 296 const int base_slice = cfqd->cfq_slice[sync];
297 297
298 WARN_ON(prio >= IOPRIO_BE_NR); 298 WARN_ON(prio >= IOPRIO_BE_NR);
299 299
300 return base_slice + (base_slice/CFQ_SLICE_SCALE * (4 - prio)); 300 return base_slice + (base_slice/CFQ_SLICE_SCALE * (4 - prio));
301 } 301 }
302 302
303 static inline int 303 static inline int
304 cfq_prio_to_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) 304 cfq_prio_to_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq)
305 { 305 {
306 return cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio); 306 return cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio);
307 } 307 }
308 308
309 static inline void 309 static inline void
310 cfq_set_prio_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) 310 cfq_set_prio_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq)
311 { 311 {
312 cfqq->slice_end = cfq_prio_to_slice(cfqd, cfqq) + jiffies; 312 cfqq->slice_end = cfq_prio_to_slice(cfqd, cfqq) + jiffies;
313 cfq_log_cfqq(cfqd, cfqq, "set_slice=%lu", cfqq->slice_end - jiffies); 313 cfq_log_cfqq(cfqd, cfqq, "set_slice=%lu", cfqq->slice_end - jiffies);
314 } 314 }
315 315
316 /* 316 /*
317 * We need to wrap this check in cfq_cfqq_slice_new(), since ->slice_end 317 * We need to wrap this check in cfq_cfqq_slice_new(), since ->slice_end
318 * isn't valid until the first request from the dispatch is activated 318 * isn't valid until the first request from the dispatch is activated
319 * and the slice time set. 319 * and the slice time set.
320 */ 320 */
321 static inline int cfq_slice_used(struct cfq_queue *cfqq) 321 static inline int cfq_slice_used(struct cfq_queue *cfqq)
322 { 322 {
323 if (cfq_cfqq_slice_new(cfqq)) 323 if (cfq_cfqq_slice_new(cfqq))
324 return 0; 324 return 0;
325 if (time_before(jiffies, cfqq->slice_end)) 325 if (time_before(jiffies, cfqq->slice_end))
326 return 0; 326 return 0;
327 327
328 return 1; 328 return 1;
329 } 329 }
330 330
331 /* 331 /*
332 * Lifted from AS - choose which of rq1 and rq2 that is best served now. 332 * Lifted from AS - choose which of rq1 and rq2 that is best served now.
333 * We choose the request that is closest to the head right now. Distance 333 * We choose the request that is closest to the head right now. Distance
334 * behind the head is penalized and only allowed to a certain extent. 334 * behind the head is penalized and only allowed to a certain extent.
335 */ 335 */
336 static struct request * 336 static struct request *
337 cfq_choose_req(struct cfq_data *cfqd, struct request *rq1, struct request *rq2) 337 cfq_choose_req(struct cfq_data *cfqd, struct request *rq1, struct request *rq2)
338 { 338 {
339 sector_t last, s1, s2, d1 = 0, d2 = 0; 339 sector_t last, s1, s2, d1 = 0, d2 = 0;
340 unsigned long back_max; 340 unsigned long back_max;
341 #define CFQ_RQ1_WRAP 0x01 /* request 1 wraps */ 341 #define CFQ_RQ1_WRAP 0x01 /* request 1 wraps */
342 #define CFQ_RQ2_WRAP 0x02 /* request 2 wraps */ 342 #define CFQ_RQ2_WRAP 0x02 /* request 2 wraps */
343 unsigned wrap = 0; /* bit mask: requests behind the disk head? */ 343 unsigned wrap = 0; /* bit mask: requests behind the disk head? */
344 344
345 if (rq1 == NULL || rq1 == rq2) 345 if (rq1 == NULL || rq1 == rq2)
346 return rq2; 346 return rq2;
347 if (rq2 == NULL) 347 if (rq2 == NULL)
348 return rq1; 348 return rq1;
349 349
350 if (rq_is_sync(rq1) && !rq_is_sync(rq2)) 350 if (rq_is_sync(rq1) && !rq_is_sync(rq2))
351 return rq1; 351 return rq1;
352 else if (rq_is_sync(rq2) && !rq_is_sync(rq1)) 352 else if (rq_is_sync(rq2) && !rq_is_sync(rq1))
353 return rq2; 353 return rq2;
354 if (rq_is_meta(rq1) && !rq_is_meta(rq2)) 354 if (rq_is_meta(rq1) && !rq_is_meta(rq2))
355 return rq1; 355 return rq1;
356 else if (rq_is_meta(rq2) && !rq_is_meta(rq1)) 356 else if (rq_is_meta(rq2) && !rq_is_meta(rq1))
357 return rq2; 357 return rq2;
358 358
359 s1 = blk_rq_pos(rq1); 359 s1 = blk_rq_pos(rq1);
360 s2 = blk_rq_pos(rq2); 360 s2 = blk_rq_pos(rq2);
361 361
362 last = cfqd->last_position; 362 last = cfqd->last_position;
363 363
364 /* 364 /*
365 * by definition, 1KiB is 2 sectors 365 * by definition, 1KiB is 2 sectors
366 */ 366 */
367 back_max = cfqd->cfq_back_max * 2; 367 back_max = cfqd->cfq_back_max * 2;
368 368
369 /* 369 /*
370 * Strict one way elevator _except_ in the case where we allow 370 * Strict one way elevator _except_ in the case where we allow
371 * short backward seeks which are biased as twice the cost of a 371 * short backward seeks which are biased as twice the cost of a
372 * similar forward seek. 372 * similar forward seek.
373 */ 373 */
374 if (s1 >= last) 374 if (s1 >= last)
375 d1 = s1 - last; 375 d1 = s1 - last;
376 else if (s1 + back_max >= last) 376 else if (s1 + back_max >= last)
377 d1 = (last - s1) * cfqd->cfq_back_penalty; 377 d1 = (last - s1) * cfqd->cfq_back_penalty;
378 else 378 else
379 wrap |= CFQ_RQ1_WRAP; 379 wrap |= CFQ_RQ1_WRAP;
380 380
381 if (s2 >= last) 381 if (s2 >= last)
382 d2 = s2 - last; 382 d2 = s2 - last;
383 else if (s2 + back_max >= last) 383 else if (s2 + back_max >= last)
384 d2 = (last - s2) * cfqd->cfq_back_penalty; 384 d2 = (last - s2) * cfqd->cfq_back_penalty;
385 else 385 else
386 wrap |= CFQ_RQ2_WRAP; 386 wrap |= CFQ_RQ2_WRAP;
387 387
388 /* Found required data */ 388 /* Found required data */
389 389
390 /* 390 /*
391 * By doing switch() on the bit mask "wrap" we avoid having to 391 * By doing switch() on the bit mask "wrap" we avoid having to
392 * check two variables for all permutations: --> faster! 392 * check two variables for all permutations: --> faster!
393 */ 393 */
394 switch (wrap) { 394 switch (wrap) {
395 case 0: /* common case for CFQ: rq1 and rq2 not wrapped */ 395 case 0: /* common case for CFQ: rq1 and rq2 not wrapped */
396 if (d1 < d2) 396 if (d1 < d2)
397 return rq1; 397 return rq1;
398 else if (d2 < d1) 398 else if (d2 < d1)
399 return rq2; 399 return rq2;
400 else { 400 else {
401 if (s1 >= s2) 401 if (s1 >= s2)
402 return rq1; 402 return rq1;
403 else 403 else
404 return rq2; 404 return rq2;
405 } 405 }
406 406
407 case CFQ_RQ2_WRAP: 407 case CFQ_RQ2_WRAP:
408 return rq1; 408 return rq1;
409 case CFQ_RQ1_WRAP: 409 case CFQ_RQ1_WRAP:
410 return rq2; 410 return rq2;
411 case (CFQ_RQ1_WRAP|CFQ_RQ2_WRAP): /* both rqs wrapped */ 411 case (CFQ_RQ1_WRAP|CFQ_RQ2_WRAP): /* both rqs wrapped */
412 default: 412 default:
413 /* 413 /*
414 * Since both rqs are wrapped, 414 * Since both rqs are wrapped,
415 * start with the one that's further behind head 415 * start with the one that's further behind head
416 * (--> only *one* back seek required), 416 * (--> only *one* back seek required),
417 * since back seek takes more time than forward. 417 * since back seek takes more time than forward.
418 */ 418 */
419 if (s1 <= s2) 419 if (s1 <= s2)
420 return rq1; 420 return rq1;
421 else 421 else
422 return rq2; 422 return rq2;
423 } 423 }
424 } 424 }
425 425
426 /* 426 /*
427 * The below is leftmost cache rbtree addon 427 * The below is leftmost cache rbtree addon
428 */ 428 */
429 static struct cfq_queue *cfq_rb_first(struct cfq_rb_root *root) 429 static struct cfq_queue *cfq_rb_first(struct cfq_rb_root *root)
430 { 430 {
431 if (!root->left) 431 if (!root->left)
432 root->left = rb_first(&root->rb); 432 root->left = rb_first(&root->rb);
433 433
434 if (root->left) 434 if (root->left)
435 return rb_entry(root->left, struct cfq_queue, rb_node); 435 return rb_entry(root->left, struct cfq_queue, rb_node);
436 436
437 return NULL; 437 return NULL;
438 } 438 }
439 439
440 static void rb_erase_init(struct rb_node *n, struct rb_root *root) 440 static void rb_erase_init(struct rb_node *n, struct rb_root *root)
441 { 441 {
442 rb_erase(n, root); 442 rb_erase(n, root);
443 RB_CLEAR_NODE(n); 443 RB_CLEAR_NODE(n);
444 } 444 }
445 445
446 static void cfq_rb_erase(struct rb_node *n, struct cfq_rb_root *root) 446 static void cfq_rb_erase(struct rb_node *n, struct cfq_rb_root *root)
447 { 447 {
448 if (root->left == n) 448 if (root->left == n)
449 root->left = NULL; 449 root->left = NULL;
450 rb_erase_init(n, &root->rb); 450 rb_erase_init(n, &root->rb);
451 } 451 }
452 452
453 /* 453 /*
454 * would be nice to take fifo expire time into account as well 454 * would be nice to take fifo expire time into account as well
455 */ 455 */
456 static struct request * 456 static struct request *
457 cfq_find_next_rq(struct cfq_data *cfqd, struct cfq_queue *cfqq, 457 cfq_find_next_rq(struct cfq_data *cfqd, struct cfq_queue *cfqq,
458 struct request *last) 458 struct request *last)
459 { 459 {
460 struct rb_node *rbnext = rb_next(&last->rb_node); 460 struct rb_node *rbnext = rb_next(&last->rb_node);
461 struct rb_node *rbprev = rb_prev(&last->rb_node); 461 struct rb_node *rbprev = rb_prev(&last->rb_node);
462 struct request *next = NULL, *prev = NULL; 462 struct request *next = NULL, *prev = NULL;
463 463
464 BUG_ON(RB_EMPTY_NODE(&last->rb_node)); 464 BUG_ON(RB_EMPTY_NODE(&last->rb_node));
465 465
466 if (rbprev) 466 if (rbprev)
467 prev = rb_entry_rq(rbprev); 467 prev = rb_entry_rq(rbprev);
468 468
469 if (rbnext) 469 if (rbnext)
470 next = rb_entry_rq(rbnext); 470 next = rb_entry_rq(rbnext);
471 else { 471 else {
472 rbnext = rb_first(&cfqq->sort_list); 472 rbnext = rb_first(&cfqq->sort_list);
473 if (rbnext && rbnext != &last->rb_node) 473 if (rbnext && rbnext != &last->rb_node)
474 next = rb_entry_rq(rbnext); 474 next = rb_entry_rq(rbnext);
475 } 475 }
476 476
477 return cfq_choose_req(cfqd, next, prev); 477 return cfq_choose_req(cfqd, next, prev);
478 } 478 }
479 479
480 static unsigned long cfq_slice_offset(struct cfq_data *cfqd, 480 static unsigned long cfq_slice_offset(struct cfq_data *cfqd,
481 struct cfq_queue *cfqq) 481 struct cfq_queue *cfqq)
482 { 482 {
483 /* 483 /*
484 * just an approximation, should be ok. 484 * just an approximation, should be ok.
485 */ 485 */
486 return (cfqd->busy_queues - 1) * (cfq_prio_slice(cfqd, 1, 0) - 486 return (cfqd->busy_queues - 1) * (cfq_prio_slice(cfqd, 1, 0) -
487 cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio)); 487 cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio));
488 } 488 }
489 489
490 /* 490 /*
491 * The cfqd->service_tree holds all pending cfq_queue's that have 491 * The cfqd->service_tree holds all pending cfq_queue's that have
492 * requests waiting to be processed. It is sorted in the order that 492 * requests waiting to be processed. It is sorted in the order that
493 * we will service the queues. 493 * we will service the queues.
494 */ 494 */
495 static void cfq_service_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq, 495 static void cfq_service_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq,
496 int add_front) 496 int add_front)
497 { 497 {
498 struct rb_node **p, *parent; 498 struct rb_node **p, *parent;
499 struct cfq_queue *__cfqq; 499 struct cfq_queue *__cfqq;
500 unsigned long rb_key; 500 unsigned long rb_key;
501 int left; 501 int left;
502 502
503 if (cfq_class_idle(cfqq)) { 503 if (cfq_class_idle(cfqq)) {
504 rb_key = CFQ_IDLE_DELAY; 504 rb_key = CFQ_IDLE_DELAY;
505 parent = rb_last(&cfqd->service_tree.rb); 505 parent = rb_last(&cfqd->service_tree.rb);
506 if (parent && parent != &cfqq->rb_node) { 506 if (parent && parent != &cfqq->rb_node) {
507 __cfqq = rb_entry(parent, struct cfq_queue, rb_node); 507 __cfqq = rb_entry(parent, struct cfq_queue, rb_node);
508 rb_key += __cfqq->rb_key; 508 rb_key += __cfqq->rb_key;
509 } else 509 } else
510 rb_key += jiffies; 510 rb_key += jiffies;
511 } else if (!add_front) { 511 } else if (!add_front) {
512 rb_key = cfq_slice_offset(cfqd, cfqq) + jiffies; 512 rb_key = cfq_slice_offset(cfqd, cfqq) + jiffies;
513 rb_key += cfqq->slice_resid; 513 rb_key += cfqq->slice_resid;
514 cfqq->slice_resid = 0; 514 cfqq->slice_resid = 0;
515 } else 515 } else
516 rb_key = 0; 516 rb_key = 0;
517 517
518 if (!RB_EMPTY_NODE(&cfqq->rb_node)) { 518 if (!RB_EMPTY_NODE(&cfqq->rb_node)) {
519 /* 519 /*
520 * same position, nothing more to do 520 * same position, nothing more to do
521 */ 521 */
522 if (rb_key == cfqq->rb_key) 522 if (rb_key == cfqq->rb_key)
523 return; 523 return;
524 524
525 cfq_rb_erase(&cfqq->rb_node, &cfqd->service_tree); 525 cfq_rb_erase(&cfqq->rb_node, &cfqd->service_tree);
526 } 526 }
527 527
528 left = 1; 528 left = 1;
529 parent = NULL; 529 parent = NULL;
530 p = &cfqd->service_tree.rb.rb_node; 530 p = &cfqd->service_tree.rb.rb_node;
531 while (*p) { 531 while (*p) {
532 struct rb_node **n; 532 struct rb_node **n;
533 533
534 parent = *p; 534 parent = *p;
535 __cfqq = rb_entry(parent, struct cfq_queue, rb_node); 535 __cfqq = rb_entry(parent, struct cfq_queue, rb_node);
536 536
537 /* 537 /*
538 * sort RT queues first, we always want to give 538 * sort RT queues first, we always want to give
539 * preference to them. IDLE queues goes to the back. 539 * preference to them. IDLE queues goes to the back.
540 * after that, sort on the next service time. 540 * after that, sort on the next service time.
541 */ 541 */
542 if (cfq_class_rt(cfqq) > cfq_class_rt(__cfqq)) 542 if (cfq_class_rt(cfqq) > cfq_class_rt(__cfqq))
543 n = &(*p)->rb_left; 543 n = &(*p)->rb_left;
544 else if (cfq_class_rt(cfqq) < cfq_class_rt(__cfqq)) 544 else if (cfq_class_rt(cfqq) < cfq_class_rt(__cfqq))
545 n = &(*p)->rb_right; 545 n = &(*p)->rb_right;
546 else if (cfq_class_idle(cfqq) < cfq_class_idle(__cfqq)) 546 else if (cfq_class_idle(cfqq) < cfq_class_idle(__cfqq))
547 n = &(*p)->rb_left; 547 n = &(*p)->rb_left;
548 else if (cfq_class_idle(cfqq) > cfq_class_idle(__cfqq)) 548 else if (cfq_class_idle(cfqq) > cfq_class_idle(__cfqq))
549 n = &(*p)->rb_right; 549 n = &(*p)->rb_right;
550 else if (rb_key < __cfqq->rb_key) 550 else if (rb_key < __cfqq->rb_key)
551 n = &(*p)->rb_left; 551 n = &(*p)->rb_left;
552 else 552 else
553 n = &(*p)->rb_right; 553 n = &(*p)->rb_right;
554 554
555 if (n == &(*p)->rb_right) 555 if (n == &(*p)->rb_right)
556 left = 0; 556 left = 0;
557 557
558 p = n; 558 p = n;
559 } 559 }
560 560
561 if (left) 561 if (left)
562 cfqd->service_tree.left = &cfqq->rb_node; 562 cfqd->service_tree.left = &cfqq->rb_node;
563 563
564 cfqq->rb_key = rb_key; 564 cfqq->rb_key = rb_key;
565 rb_link_node(&cfqq->rb_node, parent, p); 565 rb_link_node(&cfqq->rb_node, parent, p);
566 rb_insert_color(&cfqq->rb_node, &cfqd->service_tree.rb); 566 rb_insert_color(&cfqq->rb_node, &cfqd->service_tree.rb);
567 } 567 }
568 568
569 static struct cfq_queue * 569 static struct cfq_queue *
570 cfq_prio_tree_lookup(struct cfq_data *cfqd, struct rb_root *root, 570 cfq_prio_tree_lookup(struct cfq_data *cfqd, struct rb_root *root,
571 sector_t sector, struct rb_node **ret_parent, 571 sector_t sector, struct rb_node **ret_parent,
572 struct rb_node ***rb_link) 572 struct rb_node ***rb_link)
573 { 573 {
574 struct rb_node **p, *parent; 574 struct rb_node **p, *parent;
575 struct cfq_queue *cfqq = NULL; 575 struct cfq_queue *cfqq = NULL;
576 576
577 parent = NULL; 577 parent = NULL;
578 p = &root->rb_node; 578 p = &root->rb_node;
579 while (*p) { 579 while (*p) {
580 struct rb_node **n; 580 struct rb_node **n;
581 581
582 parent = *p; 582 parent = *p;
583 cfqq = rb_entry(parent, struct cfq_queue, p_node); 583 cfqq = rb_entry(parent, struct cfq_queue, p_node);
584 584
585 /* 585 /*
586 * Sort strictly based on sector. Smallest to the left, 586 * Sort strictly based on sector. Smallest to the left,
587 * largest to the right. 587 * largest to the right.
588 */ 588 */
589 if (sector > blk_rq_pos(cfqq->next_rq)) 589 if (sector > blk_rq_pos(cfqq->next_rq))
590 n = &(*p)->rb_right; 590 n = &(*p)->rb_right;
591 else if (sector < blk_rq_pos(cfqq->next_rq)) 591 else if (sector < blk_rq_pos(cfqq->next_rq))
592 n = &(*p)->rb_left; 592 n = &(*p)->rb_left;
593 else 593 else
594 break; 594 break;
595 p = n; 595 p = n;
596 cfqq = NULL; 596 cfqq = NULL;
597 } 597 }
598 598
599 *ret_parent = parent; 599 *ret_parent = parent;
600 if (rb_link) 600 if (rb_link)
601 *rb_link = p; 601 *rb_link = p;
602 return cfqq; 602 return cfqq;
603 } 603 }
604 604
605 static void cfq_prio_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq) 605 static void cfq_prio_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq)
606 { 606 {
607 struct rb_node **p, *parent; 607 struct rb_node **p, *parent;
608 struct cfq_queue *__cfqq; 608 struct cfq_queue *__cfqq;
609 609
610 if (cfqq->p_root) { 610 if (cfqq->p_root) {
611 rb_erase(&cfqq->p_node, cfqq->p_root); 611 rb_erase(&cfqq->p_node, cfqq->p_root);
612 cfqq->p_root = NULL; 612 cfqq->p_root = NULL;
613 } 613 }
614 614
615 if (cfq_class_idle(cfqq)) 615 if (cfq_class_idle(cfqq))
616 return; 616 return;
617 if (!cfqq->next_rq) 617 if (!cfqq->next_rq)
618 return; 618 return;
619 619
620 cfqq->p_root = &cfqd->prio_trees[cfqq->org_ioprio]; 620 cfqq->p_root = &cfqd->prio_trees[cfqq->org_ioprio];
621 __cfqq = cfq_prio_tree_lookup(cfqd, cfqq->p_root, 621 __cfqq = cfq_prio_tree_lookup(cfqd, cfqq->p_root,
622 blk_rq_pos(cfqq->next_rq), &parent, &p); 622 blk_rq_pos(cfqq->next_rq), &parent, &p);
623 if (!__cfqq) { 623 if (!__cfqq) {
624 rb_link_node(&cfqq->p_node, parent, p); 624 rb_link_node(&cfqq->p_node, parent, p);
625 rb_insert_color(&cfqq->p_node, cfqq->p_root); 625 rb_insert_color(&cfqq->p_node, cfqq->p_root);
626 } else 626 } else
627 cfqq->p_root = NULL; 627 cfqq->p_root = NULL;
628 } 628 }
629 629
630 /* 630 /*
631 * Update cfqq's position in the service tree. 631 * Update cfqq's position in the service tree.
632 */ 632 */
633 static void cfq_resort_rr_list(struct cfq_data *cfqd, struct cfq_queue *cfqq) 633 static void cfq_resort_rr_list(struct cfq_data *cfqd, struct cfq_queue *cfqq)
634 { 634 {
635 /* 635 /*
636 * Resorting requires the cfqq to be on the RR list already. 636 * Resorting requires the cfqq to be on the RR list already.
637 */ 637 */
638 if (cfq_cfqq_on_rr(cfqq)) { 638 if (cfq_cfqq_on_rr(cfqq)) {
639 cfq_service_tree_add(cfqd, cfqq, 0); 639 cfq_service_tree_add(cfqd, cfqq, 0);
640 cfq_prio_tree_add(cfqd, cfqq); 640 cfq_prio_tree_add(cfqd, cfqq);
641 } 641 }
642 } 642 }
643 643
644 /* 644 /*
645 * add to busy list of queues for service, trying to be fair in ordering 645 * add to busy list of queues for service, trying to be fair in ordering
646 * the pending list according to last request service 646 * the pending list according to last request service
647 */ 647 */
648 static void cfq_add_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) 648 static void cfq_add_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq)
649 { 649 {
650 cfq_log_cfqq(cfqd, cfqq, "add_to_rr"); 650 cfq_log_cfqq(cfqd, cfqq, "add_to_rr");
651 BUG_ON(cfq_cfqq_on_rr(cfqq)); 651 BUG_ON(cfq_cfqq_on_rr(cfqq));
652 cfq_mark_cfqq_on_rr(cfqq); 652 cfq_mark_cfqq_on_rr(cfqq);
653 cfqd->busy_queues++; 653 cfqd->busy_queues++;
654 654
655 cfq_resort_rr_list(cfqd, cfqq); 655 cfq_resort_rr_list(cfqd, cfqq);
656 } 656 }
657 657
658 /* 658 /*
659 * Called when the cfqq no longer has requests pending, remove it from 659 * Called when the cfqq no longer has requests pending, remove it from
660 * the service tree. 660 * the service tree.
661 */ 661 */
662 static void cfq_del_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) 662 static void cfq_del_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq)
663 { 663 {
664 cfq_log_cfqq(cfqd, cfqq, "del_from_rr"); 664 cfq_log_cfqq(cfqd, cfqq, "del_from_rr");
665 BUG_ON(!cfq_cfqq_on_rr(cfqq)); 665 BUG_ON(!cfq_cfqq_on_rr(cfqq));
666 cfq_clear_cfqq_on_rr(cfqq); 666 cfq_clear_cfqq_on_rr(cfqq);
667 667
668 if (!RB_EMPTY_NODE(&cfqq->rb_node)) 668 if (!RB_EMPTY_NODE(&cfqq->rb_node))
669 cfq_rb_erase(&cfqq->rb_node, &cfqd->service_tree); 669 cfq_rb_erase(&cfqq->rb_node, &cfqd->service_tree);
670 if (cfqq->p_root) { 670 if (cfqq->p_root) {
671 rb_erase(&cfqq->p_node, cfqq->p_root); 671 rb_erase(&cfqq->p_node, cfqq->p_root);
672 cfqq->p_root = NULL; 672 cfqq->p_root = NULL;
673 } 673 }
674 674
675 BUG_ON(!cfqd->busy_queues); 675 BUG_ON(!cfqd->busy_queues);
676 cfqd->busy_queues--; 676 cfqd->busy_queues--;
677 } 677 }
678 678
679 /* 679 /*
680 * rb tree support functions 680 * rb tree support functions
681 */ 681 */
682 static void cfq_del_rq_rb(struct request *rq) 682 static void cfq_del_rq_rb(struct request *rq)
683 { 683 {
684 struct cfq_queue *cfqq = RQ_CFQQ(rq); 684 struct cfq_queue *cfqq = RQ_CFQQ(rq);
685 struct cfq_data *cfqd = cfqq->cfqd; 685 struct cfq_data *cfqd = cfqq->cfqd;
686 const int sync = rq_is_sync(rq); 686 const int sync = rq_is_sync(rq);
687 687
688 BUG_ON(!cfqq->queued[sync]); 688 BUG_ON(!cfqq->queued[sync]);
689 cfqq->queued[sync]--; 689 cfqq->queued[sync]--;
690 690
691 elv_rb_del(&cfqq->sort_list, rq); 691 elv_rb_del(&cfqq->sort_list, rq);
692 692
693 if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list)) 693 if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list))
694 cfq_del_cfqq_rr(cfqd, cfqq); 694 cfq_del_cfqq_rr(cfqd, cfqq);
695 } 695 }
696 696
697 static void cfq_add_rq_rb(struct request *rq) 697 static void cfq_add_rq_rb(struct request *rq)
698 { 698 {
699 struct cfq_queue *cfqq = RQ_CFQQ(rq); 699 struct cfq_queue *cfqq = RQ_CFQQ(rq);
700 struct cfq_data *cfqd = cfqq->cfqd; 700 struct cfq_data *cfqd = cfqq->cfqd;
701 struct request *__alias, *prev; 701 struct request *__alias, *prev;
702 702
703 cfqq->queued[rq_is_sync(rq)]++; 703 cfqq->queued[rq_is_sync(rq)]++;
704 704
705 /* 705 /*
706 * looks a little odd, but the first insert might return an alias. 706 * looks a little odd, but the first insert might return an alias.
707 * if that happens, put the alias on the dispatch list 707 * if that happens, put the alias on the dispatch list
708 */ 708 */
709 while ((__alias = elv_rb_add(&cfqq->sort_list, rq)) != NULL) 709 while ((__alias = elv_rb_add(&cfqq->sort_list, rq)) != NULL)
710 cfq_dispatch_insert(cfqd->queue, __alias); 710 cfq_dispatch_insert(cfqd->queue, __alias);
711 711
712 if (!cfq_cfqq_on_rr(cfqq)) 712 if (!cfq_cfqq_on_rr(cfqq))
713 cfq_add_cfqq_rr(cfqd, cfqq); 713 cfq_add_cfqq_rr(cfqd, cfqq);
714 714
715 /* 715 /*
716 * check if this request is a better next-serve candidate 716 * check if this request is a better next-serve candidate
717 */ 717 */
718 prev = cfqq->next_rq; 718 prev = cfqq->next_rq;
719 cfqq->next_rq = cfq_choose_req(cfqd, cfqq->next_rq, rq); 719 cfqq->next_rq = cfq_choose_req(cfqd, cfqq->next_rq, rq);
720 720
721 /* 721 /*
722 * adjust priority tree position, if ->next_rq changes 722 * adjust priority tree position, if ->next_rq changes
723 */ 723 */
724 if (prev != cfqq->next_rq) 724 if (prev != cfqq->next_rq)
725 cfq_prio_tree_add(cfqd, cfqq); 725 cfq_prio_tree_add(cfqd, cfqq);
726 726
727 BUG_ON(!cfqq->next_rq); 727 BUG_ON(!cfqq->next_rq);
728 } 728 }
729 729
730 static void cfq_reposition_rq_rb(struct cfq_queue *cfqq, struct request *rq) 730 static void cfq_reposition_rq_rb(struct cfq_queue *cfqq, struct request *rq)
731 { 731 {
732 elv_rb_del(&cfqq->sort_list, rq); 732 elv_rb_del(&cfqq->sort_list, rq);
733 cfqq->queued[rq_is_sync(rq)]--; 733 cfqq->queued[rq_is_sync(rq)]--;
734 cfq_add_rq_rb(rq); 734 cfq_add_rq_rb(rq);
735 } 735 }
736 736
737 static struct request * 737 static struct request *
738 cfq_find_rq_fmerge(struct cfq_data *cfqd, struct bio *bio) 738 cfq_find_rq_fmerge(struct cfq_data *cfqd, struct bio *bio)
739 { 739 {
740 struct task_struct *tsk = current; 740 struct task_struct *tsk = current;
741 struct cfq_io_context *cic; 741 struct cfq_io_context *cic;
742 struct cfq_queue *cfqq; 742 struct cfq_queue *cfqq;
743 743
744 cic = cfq_cic_lookup(cfqd, tsk->io_context); 744 cic = cfq_cic_lookup(cfqd, tsk->io_context);
745 if (!cic) 745 if (!cic)
746 return NULL; 746 return NULL;
747 747
748 cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio)); 748 cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio));
749 if (cfqq) { 749 if (cfqq) {
750 sector_t sector = bio->bi_sector + bio_sectors(bio); 750 sector_t sector = bio->bi_sector + bio_sectors(bio);
751 751
752 return elv_rb_find(&cfqq->sort_list, sector); 752 return elv_rb_find(&cfqq->sort_list, sector);
753 } 753 }
754 754
755 return NULL; 755 return NULL;
756 } 756 }
757 757
758 static void cfq_activate_request(struct request_queue *q, struct request *rq) 758 static void cfq_activate_request(struct request_queue *q, struct request *rq)
759 { 759 {
760 struct cfq_data *cfqd = q->elevator->elevator_data; 760 struct cfq_data *cfqd = q->elevator->elevator_data;
761 761
762 cfqd->rq_in_driver[rq_is_sync(rq)]++; 762 cfqd->rq_in_driver[rq_is_sync(rq)]++;
763 cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "activate rq, drv=%d", 763 cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "activate rq, drv=%d",
764 rq_in_driver(cfqd)); 764 rq_in_driver(cfqd));
765 765
766 cfqd->last_position = blk_rq_pos(rq) + blk_rq_sectors(rq); 766 cfqd->last_position = blk_rq_pos(rq) + blk_rq_sectors(rq);
767 } 767 }
768 768
769 static void cfq_deactivate_request(struct request_queue *q, struct request *rq) 769 static void cfq_deactivate_request(struct request_queue *q, struct request *rq)
770 { 770 {
771 struct cfq_data *cfqd = q->elevator->elevator_data; 771 struct cfq_data *cfqd = q->elevator->elevator_data;
772 const int sync = rq_is_sync(rq); 772 const int sync = rq_is_sync(rq);
773 773
774 WARN_ON(!cfqd->rq_in_driver[sync]); 774 WARN_ON(!cfqd->rq_in_driver[sync]);
775 cfqd->rq_in_driver[sync]--; 775 cfqd->rq_in_driver[sync]--;
776 cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "deactivate rq, drv=%d", 776 cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "deactivate rq, drv=%d",
777 rq_in_driver(cfqd)); 777 rq_in_driver(cfqd));
778 } 778 }
779 779
780 static void cfq_remove_request(struct request *rq) 780 static void cfq_remove_request(struct request *rq)
781 { 781 {
782 struct cfq_queue *cfqq = RQ_CFQQ(rq); 782 struct cfq_queue *cfqq = RQ_CFQQ(rq);
783 783
784 if (cfqq->next_rq == rq) 784 if (cfqq->next_rq == rq)
785 cfqq->next_rq = cfq_find_next_rq(cfqq->cfqd, cfqq, rq); 785 cfqq->next_rq = cfq_find_next_rq(cfqq->cfqd, cfqq, rq);
786 786
787 list_del_init(&rq->queuelist); 787 list_del_init(&rq->queuelist);
788 cfq_del_rq_rb(rq); 788 cfq_del_rq_rb(rq);
789 789
790 cfqq->cfqd->rq_queued--; 790 cfqq->cfqd->rq_queued--;
791 if (rq_is_meta(rq)) { 791 if (rq_is_meta(rq)) {
792 WARN_ON(!cfqq->meta_pending); 792 WARN_ON(!cfqq->meta_pending);
793 cfqq->meta_pending--; 793 cfqq->meta_pending--;
794 } 794 }
795 } 795 }
796 796
797 static int cfq_merge(struct request_queue *q, struct request **req, 797 static int cfq_merge(struct request_queue *q, struct request **req,
798 struct bio *bio) 798 struct bio *bio)
799 { 799 {
800 struct cfq_data *cfqd = q->elevator->elevator_data; 800 struct cfq_data *cfqd = q->elevator->elevator_data;
801 struct request *__rq; 801 struct request *__rq;
802 802
803 __rq = cfq_find_rq_fmerge(cfqd, bio); 803 __rq = cfq_find_rq_fmerge(cfqd, bio);
804 if (__rq && elv_rq_merge_ok(__rq, bio)) { 804 if (__rq && elv_rq_merge_ok(__rq, bio)) {
805 *req = __rq; 805 *req = __rq;
806 return ELEVATOR_FRONT_MERGE; 806 return ELEVATOR_FRONT_MERGE;
807 } 807 }
808 808
809 return ELEVATOR_NO_MERGE; 809 return ELEVATOR_NO_MERGE;
810 } 810 }
811 811
812 static void cfq_merged_request(struct request_queue *q, struct request *req, 812 static void cfq_merged_request(struct request_queue *q, struct request *req,
813 int type) 813 int type)
814 { 814 {
815 if (type == ELEVATOR_FRONT_MERGE) { 815 if (type == ELEVATOR_FRONT_MERGE) {
816 struct cfq_queue *cfqq = RQ_CFQQ(req); 816 struct cfq_queue *cfqq = RQ_CFQQ(req);
817 817
818 cfq_reposition_rq_rb(cfqq, req); 818 cfq_reposition_rq_rb(cfqq, req);
819 } 819 }
820 } 820 }
821 821
822 static void 822 static void
823 cfq_merged_requests(struct request_queue *q, struct request *rq, 823 cfq_merged_requests(struct request_queue *q, struct request *rq,
824 struct request *next) 824 struct request *next)
825 { 825 {
826 /* 826 /*
827 * reposition in fifo if next is older than rq 827 * reposition in fifo if next is older than rq
828 */ 828 */
829 if (!list_empty(&rq->queuelist) && !list_empty(&next->queuelist) && 829 if (!list_empty(&rq->queuelist) && !list_empty(&next->queuelist) &&
830 time_before(next->start_time, rq->start_time)) 830 time_before(next->start_time, rq->start_time))
831 list_move(&rq->queuelist, &next->queuelist); 831 list_move(&rq->queuelist, &next->queuelist);
832 832
833 cfq_remove_request(next); 833 cfq_remove_request(next);
834 } 834 }
835 835
836 static int cfq_allow_merge(struct request_queue *q, struct request *rq, 836 static int cfq_allow_merge(struct request_queue *q, struct request *rq,
837 struct bio *bio) 837 struct bio *bio)
838 { 838 {
839 struct cfq_data *cfqd = q->elevator->elevator_data; 839 struct cfq_data *cfqd = q->elevator->elevator_data;
840 struct cfq_io_context *cic; 840 struct cfq_io_context *cic;
841 struct cfq_queue *cfqq; 841 struct cfq_queue *cfqq;
842 842
843 /* 843 /*
844 * Disallow merge of a sync bio into an async request. 844 * Disallow merge of a sync bio into an async request.
845 */ 845 */
846 if (cfq_bio_sync(bio) && !rq_is_sync(rq)) 846 if (cfq_bio_sync(bio) && !rq_is_sync(rq))
847 return 0; 847 return 0;
848 848
849 /* 849 /*
850 * Lookup the cfqq that this bio will be queued with. Allow 850 * Lookup the cfqq that this bio will be queued with. Allow
851 * merge only if rq is queued there. 851 * merge only if rq is queued there.
852 */ 852 */
853 cic = cfq_cic_lookup(cfqd, current->io_context); 853 cic = cfq_cic_lookup(cfqd, current->io_context);
854 if (!cic) 854 if (!cic)
855 return 0; 855 return 0;
856 856
857 cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio)); 857 cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio));
858 if (cfqq == RQ_CFQQ(rq)) 858 if (cfqq == RQ_CFQQ(rq))
859 return 1; 859 return 1;
860 860
861 return 0; 861 return 0;
862 } 862 }
863 863
864 static void __cfq_set_active_queue(struct cfq_data *cfqd, 864 static void __cfq_set_active_queue(struct cfq_data *cfqd,
865 struct cfq_queue *cfqq) 865 struct cfq_queue *cfqq)
866 { 866 {
867 if (cfqq) { 867 if (cfqq) {
868 cfq_log_cfqq(cfqd, cfqq, "set_active"); 868 cfq_log_cfqq(cfqd, cfqq, "set_active");
869 cfqq->slice_end = 0; 869 cfqq->slice_end = 0;
870 cfqq->slice_dispatch = 0; 870 cfqq->slice_dispatch = 0;
871 871
872 cfq_clear_cfqq_wait_request(cfqq); 872 cfq_clear_cfqq_wait_request(cfqq);
873 cfq_clear_cfqq_must_dispatch(cfqq); 873 cfq_clear_cfqq_must_dispatch(cfqq);
874 cfq_clear_cfqq_must_alloc_slice(cfqq); 874 cfq_clear_cfqq_must_alloc_slice(cfqq);
875 cfq_clear_cfqq_fifo_expire(cfqq); 875 cfq_clear_cfqq_fifo_expire(cfqq);
876 cfq_mark_cfqq_slice_new(cfqq); 876 cfq_mark_cfqq_slice_new(cfqq);
877 877
878 del_timer(&cfqd->idle_slice_timer); 878 del_timer(&cfqd->idle_slice_timer);
879 } 879 }
880 880
881 cfqd->active_queue = cfqq; 881 cfqd->active_queue = cfqq;
882 } 882 }
883 883
884 /* 884 /*
885 * current cfqq expired its slice (or was too idle), select new one 885 * current cfqq expired its slice (or was too idle), select new one
886 */ 886 */
887 static void 887 static void
888 __cfq_slice_expired(struct cfq_data *cfqd, struct cfq_queue *cfqq, 888 __cfq_slice_expired(struct cfq_data *cfqd, struct cfq_queue *cfqq,
889 int timed_out) 889 int timed_out)
890 { 890 {
891 cfq_log_cfqq(cfqd, cfqq, "slice expired t=%d", timed_out); 891 cfq_log_cfqq(cfqd, cfqq, "slice expired t=%d", timed_out);
892 892
893 if (cfq_cfqq_wait_request(cfqq)) 893 if (cfq_cfqq_wait_request(cfqq))
894 del_timer(&cfqd->idle_slice_timer); 894 del_timer(&cfqd->idle_slice_timer);
895 895
896 cfq_clear_cfqq_wait_request(cfqq); 896 cfq_clear_cfqq_wait_request(cfqq);
897 897
898 /* 898 /*
899 * store what was left of this slice, if the queue idled/timed out 899 * store what was left of this slice, if the queue idled/timed out
900 */ 900 */
901 if (timed_out && !cfq_cfqq_slice_new(cfqq)) { 901 if (timed_out && !cfq_cfqq_slice_new(cfqq)) {
902 cfqq->slice_resid = cfqq->slice_end - jiffies; 902 cfqq->slice_resid = cfqq->slice_end - jiffies;
903 cfq_log_cfqq(cfqd, cfqq, "resid=%ld", cfqq->slice_resid); 903 cfq_log_cfqq(cfqd, cfqq, "resid=%ld", cfqq->slice_resid);
904 } 904 }
905 905
906 cfq_resort_rr_list(cfqd, cfqq); 906 cfq_resort_rr_list(cfqd, cfqq);
907 907
908 if (cfqq == cfqd->active_queue) 908 if (cfqq == cfqd->active_queue)
909 cfqd->active_queue = NULL; 909 cfqd->active_queue = NULL;
910 910
911 if (cfqd->active_cic) { 911 if (cfqd->active_cic) {
912 put_io_context(cfqd->active_cic->ioc); 912 put_io_context(cfqd->active_cic->ioc);
913 cfqd->active_cic = NULL; 913 cfqd->active_cic = NULL;
914 } 914 }
915 } 915 }
916 916
917 static inline void cfq_slice_expired(struct cfq_data *cfqd, int timed_out) 917 static inline void cfq_slice_expired(struct cfq_data *cfqd, int timed_out)
918 { 918 {
919 struct cfq_queue *cfqq = cfqd->active_queue; 919 struct cfq_queue *cfqq = cfqd->active_queue;
920 920
921 if (cfqq) 921 if (cfqq)
922 __cfq_slice_expired(cfqd, cfqq, timed_out); 922 __cfq_slice_expired(cfqd, cfqq, timed_out);
923 } 923 }
924 924
925 /* 925 /*
926 * Get next queue for service. Unless we have a queue preemption, 926 * Get next queue for service. Unless we have a queue preemption,
927 * we'll simply select the first cfqq in the service tree. 927 * we'll simply select the first cfqq in the service tree.
928 */ 928 */
929 static struct cfq_queue *cfq_get_next_queue(struct cfq_data *cfqd) 929 static struct cfq_queue *cfq_get_next_queue(struct cfq_data *cfqd)
930 { 930 {
931 if (RB_EMPTY_ROOT(&cfqd->service_tree.rb)) 931 if (RB_EMPTY_ROOT(&cfqd->service_tree.rb))
932 return NULL; 932 return NULL;
933 933
934 return cfq_rb_first(&cfqd->service_tree); 934 return cfq_rb_first(&cfqd->service_tree);
935 } 935 }
936 936
937 /* 937 /*
938 * Get and set a new active queue for service. 938 * Get and set a new active queue for service.
939 */ 939 */
940 static struct cfq_queue *cfq_set_active_queue(struct cfq_data *cfqd, 940 static struct cfq_queue *cfq_set_active_queue(struct cfq_data *cfqd,
941 struct cfq_queue *cfqq) 941 struct cfq_queue *cfqq)
942 { 942 {
943 if (!cfqq) { 943 if (!cfqq) {
944 cfqq = cfq_get_next_queue(cfqd); 944 cfqq = cfq_get_next_queue(cfqd);
945 if (cfqq) 945 if (cfqq)
946 cfq_clear_cfqq_coop(cfqq); 946 cfq_clear_cfqq_coop(cfqq);
947 } 947 }
948 948
949 __cfq_set_active_queue(cfqd, cfqq); 949 __cfq_set_active_queue(cfqd, cfqq);
950 return cfqq; 950 return cfqq;
951 } 951 }
952 952
953 static inline sector_t cfq_dist_from_last(struct cfq_data *cfqd, 953 static inline sector_t cfq_dist_from_last(struct cfq_data *cfqd,
954 struct request *rq) 954 struct request *rq)
955 { 955 {
956 if (blk_rq_pos(rq) >= cfqd->last_position) 956 if (blk_rq_pos(rq) >= cfqd->last_position)
957 return blk_rq_pos(rq) - cfqd->last_position; 957 return blk_rq_pos(rq) - cfqd->last_position;
958 else 958 else
959 return cfqd->last_position - blk_rq_pos(rq); 959 return cfqd->last_position - blk_rq_pos(rq);
960 } 960 }
961 961
962 #define CIC_SEEK_THR 8 * 1024 962 #define CIC_SEEK_THR 8 * 1024
963 #define CIC_SEEKY(cic) ((cic)->seek_mean > CIC_SEEK_THR) 963 #define CIC_SEEKY(cic) ((cic)->seek_mean > CIC_SEEK_THR)
964 964
965 static inline int cfq_rq_close(struct cfq_data *cfqd, struct request *rq) 965 static inline int cfq_rq_close(struct cfq_data *cfqd, struct request *rq)
966 { 966 {
967 struct cfq_io_context *cic = cfqd->active_cic; 967 struct cfq_io_context *cic = cfqd->active_cic;
968 sector_t sdist = cic->seek_mean; 968 sector_t sdist = cic->seek_mean;
969 969
970 if (!sample_valid(cic->seek_samples)) 970 if (!sample_valid(cic->seek_samples))
971 sdist = CIC_SEEK_THR; 971 sdist = CIC_SEEK_THR;
972 972
973 return cfq_dist_from_last(cfqd, rq) <= sdist; 973 return cfq_dist_from_last(cfqd, rq) <= sdist;
974 } 974 }
975 975
976 static struct cfq_queue *cfqq_close(struct cfq_data *cfqd, 976 static struct cfq_queue *cfqq_close(struct cfq_data *cfqd,
977 struct cfq_queue *cur_cfqq) 977 struct cfq_queue *cur_cfqq)
978 { 978 {
979 struct rb_root *root = &cfqd->prio_trees[cur_cfqq->org_ioprio]; 979 struct rb_root *root = &cfqd->prio_trees[cur_cfqq->org_ioprio];
980 struct rb_node *parent, *node; 980 struct rb_node *parent, *node;
981 struct cfq_queue *__cfqq; 981 struct cfq_queue *__cfqq;
982 sector_t sector = cfqd->last_position; 982 sector_t sector = cfqd->last_position;
983 983
984 if (RB_EMPTY_ROOT(root)) 984 if (RB_EMPTY_ROOT(root))
985 return NULL; 985 return NULL;
986 986
987 /* 987 /*
988 * First, if we find a request starting at the end of the last 988 * First, if we find a request starting at the end of the last
989 * request, choose it. 989 * request, choose it.
990 */ 990 */
991 __cfqq = cfq_prio_tree_lookup(cfqd, root, sector, &parent, NULL); 991 __cfqq = cfq_prio_tree_lookup(cfqd, root, sector, &parent, NULL);
992 if (__cfqq) 992 if (__cfqq)
993 return __cfqq; 993 return __cfqq;
994 994
995 /* 995 /*
996 * If the exact sector wasn't found, the parent of the NULL leaf 996 * If the exact sector wasn't found, the parent of the NULL leaf
997 * will contain the closest sector. 997 * will contain the closest sector.
998 */ 998 */
999 __cfqq = rb_entry(parent, struct cfq_queue, p_node); 999 __cfqq = rb_entry(parent, struct cfq_queue, p_node);
1000 if (cfq_rq_close(cfqd, __cfqq->next_rq)) 1000 if (cfq_rq_close(cfqd, __cfqq->next_rq))
1001 return __cfqq; 1001 return __cfqq;
1002 1002
1003 if (blk_rq_pos(__cfqq->next_rq) < sector) 1003 if (blk_rq_pos(__cfqq->next_rq) < sector)
1004 node = rb_next(&__cfqq->p_node); 1004 node = rb_next(&__cfqq->p_node);
1005 else 1005 else
1006 node = rb_prev(&__cfqq->p_node); 1006 node = rb_prev(&__cfqq->p_node);
1007 if (!node) 1007 if (!node)
1008 return NULL; 1008 return NULL;
1009 1009
1010 __cfqq = rb_entry(node, struct cfq_queue, p_node); 1010 __cfqq = rb_entry(node, struct cfq_queue, p_node);
1011 if (cfq_rq_close(cfqd, __cfqq->next_rq)) 1011 if (cfq_rq_close(cfqd, __cfqq->next_rq))
1012 return __cfqq; 1012 return __cfqq;
1013 1013
1014 return NULL; 1014 return NULL;
1015 } 1015 }
1016 1016
1017 /* 1017 /*
1018 * cfqd - obvious 1018 * cfqd - obvious
1019 * cur_cfqq - passed in so that we don't decide that the current queue is 1019 * cur_cfqq - passed in so that we don't decide that the current queue is
1020 * closely cooperating with itself. 1020 * closely cooperating with itself.
1021 * 1021 *
1022 * So, basically we're assuming that that cur_cfqq has dispatched at least 1022 * So, basically we're assuming that that cur_cfqq has dispatched at least
1023 * one request, and that cfqd->last_position reflects a position on the disk 1023 * one request, and that cfqd->last_position reflects a position on the disk
1024 * associated with the I/O issued by cur_cfqq. I'm not sure this is a valid 1024 * associated with the I/O issued by cur_cfqq. I'm not sure this is a valid
1025 * assumption. 1025 * assumption.
1026 */ 1026 */
1027 static struct cfq_queue *cfq_close_cooperator(struct cfq_data *cfqd, 1027 static struct cfq_queue *cfq_close_cooperator(struct cfq_data *cfqd,
1028 struct cfq_queue *cur_cfqq, 1028 struct cfq_queue *cur_cfqq,
1029 int probe) 1029 int probe)
1030 { 1030 {
1031 struct cfq_queue *cfqq; 1031 struct cfq_queue *cfqq;
1032 1032
1033 /* 1033 /*
1034 * A valid cfq_io_context is necessary to compare requests against 1034 * A valid cfq_io_context is necessary to compare requests against
1035 * the seek_mean of the current cfqq. 1035 * the seek_mean of the current cfqq.
1036 */ 1036 */
1037 if (!cfqd->active_cic) 1037 if (!cfqd->active_cic)
1038 return NULL; 1038 return NULL;
1039 1039
1040 /* 1040 /*
1041 * We should notice if some of the queues are cooperating, eg 1041 * We should notice if some of the queues are cooperating, eg
1042 * working closely on the same area of the disk. In that case, 1042 * working closely on the same area of the disk. In that case,
1043 * we can group them together and don't waste time idling. 1043 * we can group them together and don't waste time idling.
1044 */ 1044 */
1045 cfqq = cfqq_close(cfqd, cur_cfqq); 1045 cfqq = cfqq_close(cfqd, cur_cfqq);
1046 if (!cfqq) 1046 if (!cfqq)
1047 return NULL; 1047 return NULL;
1048 1048
1049 if (cfq_cfqq_coop(cfqq)) 1049 if (cfq_cfqq_coop(cfqq))
1050 return NULL; 1050 return NULL;
1051 1051
1052 if (!probe) 1052 if (!probe)
1053 cfq_mark_cfqq_coop(cfqq); 1053 cfq_mark_cfqq_coop(cfqq);
1054 return cfqq; 1054 return cfqq;
1055 } 1055 }
1056 1056
1057 static void cfq_arm_slice_timer(struct cfq_data *cfqd) 1057 static void cfq_arm_slice_timer(struct cfq_data *cfqd)
1058 { 1058 {
1059 struct cfq_queue *cfqq = cfqd->active_queue; 1059 struct cfq_queue *cfqq = cfqd->active_queue;
1060 struct cfq_io_context *cic; 1060 struct cfq_io_context *cic;
1061 unsigned long sl; 1061 unsigned long sl;
1062 1062
1063 /* 1063 /*
1064 * SSD device without seek penalty, disable idling. But only do so 1064 * SSD device without seek penalty, disable idling. But only do so
1065 * for devices that support queuing, otherwise we still have a problem 1065 * for devices that support queuing, otherwise we still have a problem
1066 * with sync vs async workloads. 1066 * with sync vs async workloads.
1067 */ 1067 */
1068 if (blk_queue_nonrot(cfqd->queue) && cfqd->hw_tag) 1068 if (blk_queue_nonrot(cfqd->queue) && cfqd->hw_tag)
1069 return; 1069 return;
1070 1070
1071 WARN_ON(!RB_EMPTY_ROOT(&cfqq->sort_list)); 1071 WARN_ON(!RB_EMPTY_ROOT(&cfqq->sort_list));
1072 WARN_ON(cfq_cfqq_slice_new(cfqq)); 1072 WARN_ON(cfq_cfqq_slice_new(cfqq));
1073 1073
1074 /* 1074 /*
1075 * idle is disabled, either manually or by past process history 1075 * idle is disabled, either manually or by past process history
1076 */ 1076 */
1077 if (!cfqd->cfq_slice_idle || !cfq_cfqq_idle_window(cfqq)) 1077 if (!cfqd->cfq_slice_idle || !cfq_cfqq_idle_window(cfqq))
1078 return; 1078 return;
1079 1079
1080 /* 1080 /*
1081 * still requests with the driver, don't idle 1081 * still requests with the driver, don't idle
1082 */ 1082 */
1083 if (rq_in_driver(cfqd)) 1083 if (rq_in_driver(cfqd))
1084 return; 1084 return;
1085 1085
1086 /* 1086 /*
1087 * task has exited, don't wait 1087 * task has exited, don't wait
1088 */ 1088 */
1089 cic = cfqd->active_cic; 1089 cic = cfqd->active_cic;
1090 if (!cic || !atomic_read(&cic->ioc->nr_tasks)) 1090 if (!cic || !atomic_read(&cic->ioc->nr_tasks))
1091 return; 1091 return;
1092 1092
1093 cfq_mark_cfqq_wait_request(cfqq); 1093 cfq_mark_cfqq_wait_request(cfqq);
1094 1094
1095 /* 1095 /*
1096 * we don't want to idle for seeks, but we do want to allow 1096 * we don't want to idle for seeks, but we do want to allow
1097 * fair distribution of slice time for a process doing back-to-back 1097 * fair distribution of slice time for a process doing back-to-back
1098 * seeks. so allow a little bit of time for him to submit a new rq 1098 * seeks. so allow a little bit of time for him to submit a new rq
1099 */ 1099 */
1100 sl = cfqd->cfq_slice_idle; 1100 sl = cfqd->cfq_slice_idle;
1101 if (sample_valid(cic->seek_samples) && CIC_SEEKY(cic)) 1101 if (sample_valid(cic->seek_samples) && CIC_SEEKY(cic))
1102 sl = min(sl, msecs_to_jiffies(CFQ_MIN_TT)); 1102 sl = min(sl, msecs_to_jiffies(CFQ_MIN_TT));
1103 1103
1104 mod_timer(&cfqd->idle_slice_timer, jiffies + sl); 1104 mod_timer(&cfqd->idle_slice_timer, jiffies + sl);
1105 cfq_log_cfqq(cfqd, cfqq, "arm_idle: %lu", sl); 1105 cfq_log_cfqq(cfqd, cfqq, "arm_idle: %lu", sl);
1106 } 1106 }
1107 1107
1108 /* 1108 /*
1109 * Move request from internal lists to the request queue dispatch list. 1109 * Move request from internal lists to the request queue dispatch list.
1110 */ 1110 */
1111 static void cfq_dispatch_insert(struct request_queue *q, struct request *rq) 1111 static void cfq_dispatch_insert(struct request_queue *q, struct request *rq)
1112 { 1112 {
1113 struct cfq_data *cfqd = q->elevator->elevator_data; 1113 struct cfq_data *cfqd = q->elevator->elevator_data;
1114 struct cfq_queue *cfqq = RQ_CFQQ(rq); 1114 struct cfq_queue *cfqq = RQ_CFQQ(rq);
1115 1115
1116 cfq_log_cfqq(cfqd, cfqq, "dispatch_insert"); 1116 cfq_log_cfqq(cfqd, cfqq, "dispatch_insert");
1117 1117
1118 cfqq->next_rq = cfq_find_next_rq(cfqd, cfqq, rq); 1118 cfqq->next_rq = cfq_find_next_rq(cfqd, cfqq, rq);
1119 cfq_remove_request(rq); 1119 cfq_remove_request(rq);
1120 cfqq->dispatched++; 1120 cfqq->dispatched++;
1121 elv_dispatch_sort(q, rq); 1121 elv_dispatch_sort(q, rq);
1122 1122
1123 if (cfq_cfqq_sync(cfqq)) 1123 if (cfq_cfqq_sync(cfqq))
1124 cfqd->sync_flight++; 1124 cfqd->sync_flight++;
1125 } 1125 }
1126 1126
1127 /* 1127 /*
1128 * return expired entry, or NULL to just start from scratch in rbtree 1128 * return expired entry, or NULL to just start from scratch in rbtree
1129 */ 1129 */
1130 static struct request *cfq_check_fifo(struct cfq_queue *cfqq) 1130 static struct request *cfq_check_fifo(struct cfq_queue *cfqq)
1131 { 1131 {
1132 struct cfq_data *cfqd = cfqq->cfqd; 1132 struct cfq_data *cfqd = cfqq->cfqd;
1133 struct request *rq; 1133 struct request *rq;
1134 int fifo; 1134 int fifo;
1135 1135
1136 if (cfq_cfqq_fifo_expire(cfqq)) 1136 if (cfq_cfqq_fifo_expire(cfqq))
1137 return NULL; 1137 return NULL;
1138 1138
1139 cfq_mark_cfqq_fifo_expire(cfqq); 1139 cfq_mark_cfqq_fifo_expire(cfqq);
1140 1140
1141 if (list_empty(&cfqq->fifo)) 1141 if (list_empty(&cfqq->fifo))
1142 return NULL; 1142 return NULL;
1143 1143
1144 fifo = cfq_cfqq_sync(cfqq); 1144 fifo = cfq_cfqq_sync(cfqq);
1145 rq = rq_entry_fifo(cfqq->fifo.next); 1145 rq = rq_entry_fifo(cfqq->fifo.next);
1146 1146
1147 if (time_before(jiffies, rq->start_time + cfqd->cfq_fifo_expire[fifo])) 1147 if (time_before(jiffies, rq->start_time + cfqd->cfq_fifo_expire[fifo]))
1148 rq = NULL; 1148 rq = NULL;
1149 1149
1150 cfq_log_cfqq(cfqd, cfqq, "fifo=%p", rq); 1150 cfq_log_cfqq(cfqd, cfqq, "fifo=%p", rq);
1151 return rq; 1151 return rq;
1152 } 1152 }
1153 1153
1154 static inline int 1154 static inline int
1155 cfq_prio_to_maxrq(struct cfq_data *cfqd, struct cfq_queue *cfqq) 1155 cfq_prio_to_maxrq(struct cfq_data *cfqd, struct cfq_queue *cfqq)
1156 { 1156 {
1157 const int base_rq = cfqd->cfq_slice_async_rq; 1157 const int base_rq = cfqd->cfq_slice_async_rq;
1158 1158
1159 WARN_ON(cfqq->ioprio >= IOPRIO_BE_NR); 1159 WARN_ON(cfqq->ioprio >= IOPRIO_BE_NR);
1160 1160
1161 return 2 * (base_rq + base_rq * (CFQ_PRIO_LISTS - 1 - cfqq->ioprio)); 1161 return 2 * (base_rq + base_rq * (CFQ_PRIO_LISTS - 1 - cfqq->ioprio));
1162 } 1162 }
1163 1163
1164 /* 1164 /*
1165 * Select a queue for service. If we have a current active queue, 1165 * Select a queue for service. If we have a current active queue,
1166 * check whether to continue servicing it, or retrieve and set a new one. 1166 * check whether to continue servicing it, or retrieve and set a new one.
1167 */ 1167 */
1168 static struct cfq_queue *cfq_select_queue(struct cfq_data *cfqd) 1168 static struct cfq_queue *cfq_select_queue(struct cfq_data *cfqd)
1169 { 1169 {
1170 struct cfq_queue *cfqq, *new_cfqq = NULL; 1170 struct cfq_queue *cfqq, *new_cfqq = NULL;
1171 1171
1172 cfqq = cfqd->active_queue; 1172 cfqq = cfqd->active_queue;
1173 if (!cfqq) 1173 if (!cfqq)
1174 goto new_queue; 1174 goto new_queue;
1175 1175
1176 /* 1176 /*
1177 * The active queue has run out of time, expire it and select new. 1177 * The active queue has run out of time, expire it and select new.
1178 */ 1178 */
1179 if (cfq_slice_used(cfqq) && !cfq_cfqq_must_dispatch(cfqq)) 1179 if (cfq_slice_used(cfqq) && !cfq_cfqq_must_dispatch(cfqq))
1180 goto expire; 1180 goto expire;
1181 1181
1182 /* 1182 /*
1183 * The active queue has requests and isn't expired, allow it to 1183 * The active queue has requests and isn't expired, allow it to
1184 * dispatch. 1184 * dispatch.
1185 */ 1185 */
1186 if (!RB_EMPTY_ROOT(&cfqq->sort_list)) 1186 if (!RB_EMPTY_ROOT(&cfqq->sort_list))
1187 goto keep_queue; 1187 goto keep_queue;
1188 1188
1189 /* 1189 /*
1190 * If another queue has a request waiting within our mean seek 1190 * If another queue has a request waiting within our mean seek
1191 * distance, let it run. The expire code will check for close 1191 * distance, let it run. The expire code will check for close
1192 * cooperators and put the close queue at the front of the service 1192 * cooperators and put the close queue at the front of the service
1193 * tree. 1193 * tree.
1194 */ 1194 */
1195 new_cfqq = cfq_close_cooperator(cfqd, cfqq, 0); 1195 new_cfqq = cfq_close_cooperator(cfqd, cfqq, 0);
1196 if (new_cfqq) 1196 if (new_cfqq)
1197 goto expire; 1197 goto expire;
1198 1198
1199 /* 1199 /*
1200 * No requests pending. If the active queue still has requests in 1200 * No requests pending. If the active queue still has requests in
1201 * flight or is idling for a new request, allow either of these 1201 * flight or is idling for a new request, allow either of these
1202 * conditions to happen (or time out) before selecting a new queue. 1202 * conditions to happen (or time out) before selecting a new queue.
1203 */ 1203 */
1204 if (timer_pending(&cfqd->idle_slice_timer) || 1204 if (timer_pending(&cfqd->idle_slice_timer) ||
1205 (cfqq->dispatched && cfq_cfqq_idle_window(cfqq))) { 1205 (cfqq->dispatched && cfq_cfqq_idle_window(cfqq))) {
1206 cfqq = NULL; 1206 cfqq = NULL;
1207 goto keep_queue; 1207 goto keep_queue;
1208 } 1208 }
1209 1209
1210 expire: 1210 expire:
1211 cfq_slice_expired(cfqd, 0); 1211 cfq_slice_expired(cfqd, 0);
1212 new_queue: 1212 new_queue:
1213 cfqq = cfq_set_active_queue(cfqd, new_cfqq); 1213 cfqq = cfq_set_active_queue(cfqd, new_cfqq);
1214 keep_queue: 1214 keep_queue:
1215 return cfqq; 1215 return cfqq;
1216 } 1216 }
1217 1217
1218 static int __cfq_forced_dispatch_cfqq(struct cfq_queue *cfqq) 1218 static int __cfq_forced_dispatch_cfqq(struct cfq_queue *cfqq)
1219 { 1219 {
1220 int dispatched = 0; 1220 int dispatched = 0;
1221 1221
1222 while (cfqq->next_rq) { 1222 while (cfqq->next_rq) {
1223 cfq_dispatch_insert(cfqq->cfqd->queue, cfqq->next_rq); 1223 cfq_dispatch_insert(cfqq->cfqd->queue, cfqq->next_rq);
1224 dispatched++; 1224 dispatched++;
1225 } 1225 }
1226 1226
1227 BUG_ON(!list_empty(&cfqq->fifo)); 1227 BUG_ON(!list_empty(&cfqq->fifo));
1228 return dispatched; 1228 return dispatched;
1229 } 1229 }
1230 1230
1231 /* 1231 /*
1232 * Drain our current requests. Used for barriers and when switching 1232 * Drain our current requests. Used for barriers and when switching
1233 * io schedulers on-the-fly. 1233 * io schedulers on-the-fly.
1234 */ 1234 */
1235 static int cfq_forced_dispatch(struct cfq_data *cfqd) 1235 static int cfq_forced_dispatch(struct cfq_data *cfqd)
1236 { 1236 {
1237 struct cfq_queue *cfqq; 1237 struct cfq_queue *cfqq;
1238 int dispatched = 0; 1238 int dispatched = 0;
1239 1239
1240 while ((cfqq = cfq_rb_first(&cfqd->service_tree)) != NULL) 1240 while ((cfqq = cfq_rb_first(&cfqd->service_tree)) != NULL)
1241 dispatched += __cfq_forced_dispatch_cfqq(cfqq); 1241 dispatched += __cfq_forced_dispatch_cfqq(cfqq);
1242 1242
1243 cfq_slice_expired(cfqd, 0); 1243 cfq_slice_expired(cfqd, 0);
1244 1244
1245 BUG_ON(cfqd->busy_queues); 1245 BUG_ON(cfqd->busy_queues);
1246 1246
1247 cfq_log(cfqd, "forced_dispatch=%d", dispatched); 1247 cfq_log(cfqd, "forced_dispatch=%d", dispatched);
1248 return dispatched; 1248 return dispatched;
1249 } 1249 }
1250 1250
1251 /* 1251 /*
1252 * Dispatch a request from cfqq, moving them to the request queue 1252 * Dispatch a request from cfqq, moving them to the request queue
1253 * dispatch list. 1253 * dispatch list.
1254 */ 1254 */
1255 static void cfq_dispatch_request(struct cfq_data *cfqd, struct cfq_queue *cfqq) 1255 static void cfq_dispatch_request(struct cfq_data *cfqd, struct cfq_queue *cfqq)
1256 { 1256 {
1257 struct request *rq; 1257 struct request *rq;
1258 1258
1259 BUG_ON(RB_EMPTY_ROOT(&cfqq->sort_list)); 1259 BUG_ON(RB_EMPTY_ROOT(&cfqq->sort_list));
1260 1260
1261 /* 1261 /*
1262 * follow expired path, else get first next available 1262 * follow expired path, else get first next available
1263 */ 1263 */
1264 rq = cfq_check_fifo(cfqq); 1264 rq = cfq_check_fifo(cfqq);
1265 if (!rq) 1265 if (!rq)
1266 rq = cfqq->next_rq; 1266 rq = cfqq->next_rq;
1267 1267
1268 /* 1268 /*
1269 * insert request into driver dispatch list 1269 * insert request into driver dispatch list
1270 */ 1270 */
1271 cfq_dispatch_insert(cfqd->queue, rq); 1271 cfq_dispatch_insert(cfqd->queue, rq);
1272 1272
1273 if (!cfqd->active_cic) { 1273 if (!cfqd->active_cic) {
1274 struct cfq_io_context *cic = RQ_CIC(rq); 1274 struct cfq_io_context *cic = RQ_CIC(rq);
1275 1275
1276 atomic_long_inc(&cic->ioc->refcount); 1276 atomic_long_inc(&cic->ioc->refcount);
1277 cfqd->active_cic = cic; 1277 cfqd->active_cic = cic;
1278 } 1278 }
1279 } 1279 }
1280 1280
1281 /* 1281 /*
1282 * Find the cfqq that we need to service and move a request from that to the 1282 * Find the cfqq that we need to service and move a request from that to the
1283 * dispatch list 1283 * dispatch list
1284 */ 1284 */
1285 static int cfq_dispatch_requests(struct request_queue *q, int force) 1285 static int cfq_dispatch_requests(struct request_queue *q, int force)
1286 { 1286 {
1287 struct cfq_data *cfqd = q->elevator->elevator_data; 1287 struct cfq_data *cfqd = q->elevator->elevator_data;
1288 struct cfq_queue *cfqq; 1288 struct cfq_queue *cfqq;
1289 unsigned int max_dispatch; 1289 unsigned int max_dispatch;
1290 1290
1291 if (!cfqd->busy_queues) 1291 if (!cfqd->busy_queues)
1292 return 0; 1292 return 0;
1293 1293
1294 if (unlikely(force)) 1294 if (unlikely(force))
1295 return cfq_forced_dispatch(cfqd); 1295 return cfq_forced_dispatch(cfqd);
1296 1296
1297 cfqq = cfq_select_queue(cfqd); 1297 cfqq = cfq_select_queue(cfqd);
1298 if (!cfqq) 1298 if (!cfqq)
1299 return 0; 1299 return 0;
1300 1300
1301 /* 1301 /*
1302 * Drain async requests before we start sync IO 1302 * Drain async requests before we start sync IO
1303 */ 1303 */
1304 if (cfq_cfqq_idle_window(cfqq) && cfqd->rq_in_driver[BLK_RW_ASYNC]) 1304 if (cfq_cfqq_idle_window(cfqq) && cfqd->rq_in_driver[BLK_RW_ASYNC])
1305 return 0; 1305 return 0;
1306 1306
1307 /* 1307 /*
1308 * If this is an async queue and we have sync IO in flight, let it wait 1308 * If this is an async queue and we have sync IO in flight, let it wait
1309 */ 1309 */
1310 if (cfqd->sync_flight && !cfq_cfqq_sync(cfqq)) 1310 if (cfqd->sync_flight && !cfq_cfqq_sync(cfqq))
1311 return 0; 1311 return 0;
1312 1312
1313 max_dispatch = cfqd->cfq_quantum; 1313 max_dispatch = cfqd->cfq_quantum;
1314 if (cfq_class_idle(cfqq)) 1314 if (cfq_class_idle(cfqq))
1315 max_dispatch = 1; 1315 max_dispatch = 1;
1316 1316
1317 /* 1317 /*
1318 * Does this cfqq already have too much IO in flight? 1318 * Does this cfqq already have too much IO in flight?
1319 */ 1319 */
1320 if (cfqq->dispatched >= max_dispatch) { 1320 if (cfqq->dispatched >= max_dispatch) {
1321 /* 1321 /*
1322 * idle queue must always only have a single IO in flight 1322 * idle queue must always only have a single IO in flight
1323 */ 1323 */
1324 if (cfq_class_idle(cfqq)) 1324 if (cfq_class_idle(cfqq))
1325 return 0; 1325 return 0;
1326 1326
1327 /* 1327 /*
1328 * We have other queues, don't allow more IO from this one 1328 * We have other queues, don't allow more IO from this one
1329 */ 1329 */
1330 if (cfqd->busy_queues > 1) 1330 if (cfqd->busy_queues > 1)
1331 return 0; 1331 return 0;
1332 1332
1333 /* 1333 /*
1334 * Sole queue user, allow bigger slice 1334 * Sole queue user, allow bigger slice
1335 */ 1335 */
1336 max_dispatch *= 4; 1336 max_dispatch *= 4;
1337 } 1337 }
1338 1338
1339 /* 1339 /*
1340 * Async queues must wait a bit before being allowed dispatch. 1340 * Async queues must wait a bit before being allowed dispatch.
1341 * We also ramp up the dispatch depth gradually for async IO, 1341 * We also ramp up the dispatch depth gradually for async IO,
1342 * based on the last sync IO we serviced 1342 * based on the last sync IO we serviced
1343 */ 1343 */
1344 if (!cfq_cfqq_sync(cfqq) && cfqd->cfq_latency) { 1344 if (!cfq_cfqq_sync(cfqq) && cfqd->cfq_latency) {
1345 unsigned long last_sync = jiffies - cfqd->last_end_sync_rq; 1345 unsigned long last_sync = jiffies - cfqd->last_end_sync_rq;
1346 unsigned int depth; 1346 unsigned int depth;
1347 1347
1348 /* 1348 /*
1349 * must wait a bit longer 1349 * must wait a bit longer
1350 */ 1350 */
1351 if (last_sync < cfq_slice_sync) { 1351 if (last_sync < cfqd->cfq_slice[1]) {
1352 cfq_schedule_dispatch(cfqd, cfq_slice_sync - last_sync); 1352 cfq_schedule_dispatch(cfqd,
1353 cfqd->cfq_slice[1] - last_sync);
1353 return 0; 1354 return 0;
1354 } 1355 }
1355 1356
1356 depth = last_sync / cfq_slice_sync; 1357 depth = last_sync / cfqd->cfq_slice[1];
1357 if (depth < max_dispatch) 1358 if (depth < max_dispatch)
1358 max_dispatch = depth; 1359 max_dispatch = depth;
1359 } 1360 }
1360 1361
1361 if (cfqq->dispatched >= max_dispatch) 1362 if (cfqq->dispatched >= max_dispatch)
1362 return 0; 1363 return 0;
1363 1364
1364 /* 1365 /*
1365 * Dispatch a request from this cfqq 1366 * Dispatch a request from this cfqq
1366 */ 1367 */
1367 cfq_dispatch_request(cfqd, cfqq); 1368 cfq_dispatch_request(cfqd, cfqq);
1368 cfqq->slice_dispatch++; 1369 cfqq->slice_dispatch++;
1369 cfq_clear_cfqq_must_dispatch(cfqq); 1370 cfq_clear_cfqq_must_dispatch(cfqq);
1370 1371
1371 /* 1372 /*
1372 * expire an async queue immediately if it has used up its slice. idle 1373 * expire an async queue immediately if it has used up its slice. idle
1373 * queue always expire after 1 dispatch round. 1374 * queue always expire after 1 dispatch round.
1374 */ 1375 */
1375 if (cfqd->busy_queues > 1 && ((!cfq_cfqq_sync(cfqq) && 1376 if (cfqd->busy_queues > 1 && ((!cfq_cfqq_sync(cfqq) &&
1376 cfqq->slice_dispatch >= cfq_prio_to_maxrq(cfqd, cfqq)) || 1377 cfqq->slice_dispatch >= cfq_prio_to_maxrq(cfqd, cfqq)) ||
1377 cfq_class_idle(cfqq))) { 1378 cfq_class_idle(cfqq))) {
1378 cfqq->slice_end = jiffies + 1; 1379 cfqq->slice_end = jiffies + 1;
1379 cfq_slice_expired(cfqd, 0); 1380 cfq_slice_expired(cfqd, 0);
1380 } 1381 }
1381 1382
1382 cfq_log_cfqq(cfqd, cfqq, "dispatched a request"); 1383 cfq_log_cfqq(cfqd, cfqq, "dispatched a request");
1383 return 1; 1384 return 1;
1384 } 1385 }
1385 1386
1386 /* 1387 /*
1387 * task holds one reference to the queue, dropped when task exits. each rq 1388 * task holds one reference to the queue, dropped when task exits. each rq
1388 * in-flight on this queue also holds a reference, dropped when rq is freed. 1389 * in-flight on this queue also holds a reference, dropped when rq is freed.
1389 * 1390 *
1390 * queue lock must be held here. 1391 * queue lock must be held here.
1391 */ 1392 */
1392 static void cfq_put_queue(struct cfq_queue *cfqq) 1393 static void cfq_put_queue(struct cfq_queue *cfqq)
1393 { 1394 {
1394 struct cfq_data *cfqd = cfqq->cfqd; 1395 struct cfq_data *cfqd = cfqq->cfqd;
1395 1396
1396 BUG_ON(atomic_read(&cfqq->ref) <= 0); 1397 BUG_ON(atomic_read(&cfqq->ref) <= 0);
1397 1398
1398 if (!atomic_dec_and_test(&cfqq->ref)) 1399 if (!atomic_dec_and_test(&cfqq->ref))
1399 return; 1400 return;
1400 1401
1401 cfq_log_cfqq(cfqd, cfqq, "put_queue"); 1402 cfq_log_cfqq(cfqd, cfqq, "put_queue");
1402 BUG_ON(rb_first(&cfqq->sort_list)); 1403 BUG_ON(rb_first(&cfqq->sort_list));
1403 BUG_ON(cfqq->allocated[READ] + cfqq->allocated[WRITE]); 1404 BUG_ON(cfqq->allocated[READ] + cfqq->allocated[WRITE]);
1404 BUG_ON(cfq_cfqq_on_rr(cfqq)); 1405 BUG_ON(cfq_cfqq_on_rr(cfqq));
1405 1406
1406 if (unlikely(cfqd->active_queue == cfqq)) { 1407 if (unlikely(cfqd->active_queue == cfqq)) {
1407 __cfq_slice_expired(cfqd, cfqq, 0); 1408 __cfq_slice_expired(cfqd, cfqq, 0);
1408 cfq_schedule_dispatch(cfqd, 0); 1409 cfq_schedule_dispatch(cfqd, 0);
1409 } 1410 }
1410 1411
1411 kmem_cache_free(cfq_pool, cfqq); 1412 kmem_cache_free(cfq_pool, cfqq);
1412 } 1413 }
1413 1414
1414 /* 1415 /*
1415 * Must always be called with the rcu_read_lock() held 1416 * Must always be called with the rcu_read_lock() held
1416 */ 1417 */
1417 static void 1418 static void
1418 __call_for_each_cic(struct io_context *ioc, 1419 __call_for_each_cic(struct io_context *ioc,
1419 void (*func)(struct io_context *, struct cfq_io_context *)) 1420 void (*func)(struct io_context *, struct cfq_io_context *))
1420 { 1421 {
1421 struct cfq_io_context *cic; 1422 struct cfq_io_context *cic;
1422 struct hlist_node *n; 1423 struct hlist_node *n;
1423 1424
1424 hlist_for_each_entry_rcu(cic, n, &ioc->cic_list, cic_list) 1425 hlist_for_each_entry_rcu(cic, n, &ioc->cic_list, cic_list)
1425 func(ioc, cic); 1426 func(ioc, cic);
1426 } 1427 }
1427 1428
1428 /* 1429 /*
1429 * Call func for each cic attached to this ioc. 1430 * Call func for each cic attached to this ioc.
1430 */ 1431 */
1431 static void 1432 static void
1432 call_for_each_cic(struct io_context *ioc, 1433 call_for_each_cic(struct io_context *ioc,
1433 void (*func)(struct io_context *, struct cfq_io_context *)) 1434 void (*func)(struct io_context *, struct cfq_io_context *))
1434 { 1435 {
1435 rcu_read_lock(); 1436 rcu_read_lock();
1436 __call_for_each_cic(ioc, func); 1437 __call_for_each_cic(ioc, func);
1437 rcu_read_unlock(); 1438 rcu_read_unlock();
1438 } 1439 }
1439 1440
1440 static void cfq_cic_free_rcu(struct rcu_head *head) 1441 static void cfq_cic_free_rcu(struct rcu_head *head)
1441 { 1442 {
1442 struct cfq_io_context *cic; 1443 struct cfq_io_context *cic;
1443 1444
1444 cic = container_of(head, struct cfq_io_context, rcu_head); 1445 cic = container_of(head, struct cfq_io_context, rcu_head);
1445 1446
1446 kmem_cache_free(cfq_ioc_pool, cic); 1447 kmem_cache_free(cfq_ioc_pool, cic);
1447 elv_ioc_count_dec(cfq_ioc_count); 1448 elv_ioc_count_dec(cfq_ioc_count);
1448 1449
1449 if (ioc_gone) { 1450 if (ioc_gone) {
1450 /* 1451 /*
1451 * CFQ scheduler is exiting, grab exit lock and check 1452 * CFQ scheduler is exiting, grab exit lock and check
1452 * the pending io context count. If it hits zero, 1453 * the pending io context count. If it hits zero,
1453 * complete ioc_gone and set it back to NULL 1454 * complete ioc_gone and set it back to NULL
1454 */ 1455 */
1455 spin_lock(&ioc_gone_lock); 1456 spin_lock(&ioc_gone_lock);
1456 if (ioc_gone && !elv_ioc_count_read(cfq_ioc_count)) { 1457 if (ioc_gone && !elv_ioc_count_read(cfq_ioc_count)) {
1457 complete(ioc_gone); 1458 complete(ioc_gone);
1458 ioc_gone = NULL; 1459 ioc_gone = NULL;
1459 } 1460 }
1460 spin_unlock(&ioc_gone_lock); 1461 spin_unlock(&ioc_gone_lock);
1461 } 1462 }
1462 } 1463 }
1463 1464
1464 static void cfq_cic_free(struct cfq_io_context *cic) 1465 static void cfq_cic_free(struct cfq_io_context *cic)
1465 { 1466 {
1466 call_rcu(&cic->rcu_head, cfq_cic_free_rcu); 1467 call_rcu(&cic->rcu_head, cfq_cic_free_rcu);
1467 } 1468 }
1468 1469
1469 static void cic_free_func(struct io_context *ioc, struct cfq_io_context *cic) 1470 static void cic_free_func(struct io_context *ioc, struct cfq_io_context *cic)
1470 { 1471 {
1471 unsigned long flags; 1472 unsigned long flags;
1472 1473
1473 BUG_ON(!cic->dead_key); 1474 BUG_ON(!cic->dead_key);
1474 1475
1475 spin_lock_irqsave(&ioc->lock, flags); 1476 spin_lock_irqsave(&ioc->lock, flags);
1476 radix_tree_delete(&ioc->radix_root, cic->dead_key); 1477 radix_tree_delete(&ioc->radix_root, cic->dead_key);
1477 hlist_del_rcu(&cic->cic_list); 1478 hlist_del_rcu(&cic->cic_list);
1478 spin_unlock_irqrestore(&ioc->lock, flags); 1479 spin_unlock_irqrestore(&ioc->lock, flags);
1479 1480
1480 cfq_cic_free(cic); 1481 cfq_cic_free(cic);
1481 } 1482 }
1482 1483
1483 /* 1484 /*
1484 * Must be called with rcu_read_lock() held or preemption otherwise disabled. 1485 * Must be called with rcu_read_lock() held or preemption otherwise disabled.
1485 * Only two callers of this - ->dtor() which is called with the rcu_read_lock(), 1486 * Only two callers of this - ->dtor() which is called with the rcu_read_lock(),
1486 * and ->trim() which is called with the task lock held 1487 * and ->trim() which is called with the task lock held
1487 */ 1488 */
1488 static void cfq_free_io_context(struct io_context *ioc) 1489 static void cfq_free_io_context(struct io_context *ioc)
1489 { 1490 {
1490 /* 1491 /*
1491 * ioc->refcount is zero here, or we are called from elv_unregister(), 1492 * ioc->refcount is zero here, or we are called from elv_unregister(),
1492 * so no more cic's are allowed to be linked into this ioc. So it 1493 * so no more cic's are allowed to be linked into this ioc. So it
1493 * should be ok to iterate over the known list, we will see all cic's 1494 * should be ok to iterate over the known list, we will see all cic's
1494 * since no new ones are added. 1495 * since no new ones are added.
1495 */ 1496 */
1496 __call_for_each_cic(ioc, cic_free_func); 1497 __call_for_each_cic(ioc, cic_free_func);
1497 } 1498 }
1498 1499
1499 static void cfq_exit_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq) 1500 static void cfq_exit_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq)
1500 { 1501 {
1501 if (unlikely(cfqq == cfqd->active_queue)) { 1502 if (unlikely(cfqq == cfqd->active_queue)) {
1502 __cfq_slice_expired(cfqd, cfqq, 0); 1503 __cfq_slice_expired(cfqd, cfqq, 0);
1503 cfq_schedule_dispatch(cfqd, 0); 1504 cfq_schedule_dispatch(cfqd, 0);
1504 } 1505 }
1505 1506
1506 cfq_put_queue(cfqq); 1507 cfq_put_queue(cfqq);
1507 } 1508 }
1508 1509
1509 static void __cfq_exit_single_io_context(struct cfq_data *cfqd, 1510 static void __cfq_exit_single_io_context(struct cfq_data *cfqd,
1510 struct cfq_io_context *cic) 1511 struct cfq_io_context *cic)
1511 { 1512 {
1512 struct io_context *ioc = cic->ioc; 1513 struct io_context *ioc = cic->ioc;
1513 1514
1514 list_del_init(&cic->queue_list); 1515 list_del_init(&cic->queue_list);
1515 1516
1516 /* 1517 /*
1517 * Make sure key == NULL is seen for dead queues 1518 * Make sure key == NULL is seen for dead queues
1518 */ 1519 */
1519 smp_wmb(); 1520 smp_wmb();
1520 cic->dead_key = (unsigned long) cic->key; 1521 cic->dead_key = (unsigned long) cic->key;
1521 cic->key = NULL; 1522 cic->key = NULL;
1522 1523
1523 if (ioc->ioc_data == cic) 1524 if (ioc->ioc_data == cic)
1524 rcu_assign_pointer(ioc->ioc_data, NULL); 1525 rcu_assign_pointer(ioc->ioc_data, NULL);
1525 1526
1526 if (cic->cfqq[BLK_RW_ASYNC]) { 1527 if (cic->cfqq[BLK_RW_ASYNC]) {
1527 cfq_exit_cfqq(cfqd, cic->cfqq[BLK_RW_ASYNC]); 1528 cfq_exit_cfqq(cfqd, cic->cfqq[BLK_RW_ASYNC]);
1528 cic->cfqq[BLK_RW_ASYNC] = NULL; 1529 cic->cfqq[BLK_RW_ASYNC] = NULL;
1529 } 1530 }
1530 1531
1531 if (cic->cfqq[BLK_RW_SYNC]) { 1532 if (cic->cfqq[BLK_RW_SYNC]) {
1532 cfq_exit_cfqq(cfqd, cic->cfqq[BLK_RW_SYNC]); 1533 cfq_exit_cfqq(cfqd, cic->cfqq[BLK_RW_SYNC]);
1533 cic->cfqq[BLK_RW_SYNC] = NULL; 1534 cic->cfqq[BLK_RW_SYNC] = NULL;
1534 } 1535 }
1535 } 1536 }
1536 1537
1537 static void cfq_exit_single_io_context(struct io_context *ioc, 1538 static void cfq_exit_single_io_context(struct io_context *ioc,
1538 struct cfq_io_context *cic) 1539 struct cfq_io_context *cic)
1539 { 1540 {
1540 struct cfq_data *cfqd = cic->key; 1541 struct cfq_data *cfqd = cic->key;
1541 1542
1542 if (cfqd) { 1543 if (cfqd) {
1543 struct request_queue *q = cfqd->queue; 1544 struct request_queue *q = cfqd->queue;
1544 unsigned long flags; 1545 unsigned long flags;
1545 1546
1546 spin_lock_irqsave(q->queue_lock, flags); 1547 spin_lock_irqsave(q->queue_lock, flags);
1547 1548
1548 /* 1549 /*
1549 * Ensure we get a fresh copy of the ->key to prevent 1550 * Ensure we get a fresh copy of the ->key to prevent
1550 * race between exiting task and queue 1551 * race between exiting task and queue
1551 */ 1552 */
1552 smp_read_barrier_depends(); 1553 smp_read_barrier_depends();
1553 if (cic->key) 1554 if (cic->key)
1554 __cfq_exit_single_io_context(cfqd, cic); 1555 __cfq_exit_single_io_context(cfqd, cic);
1555 1556
1556 spin_unlock_irqrestore(q->queue_lock, flags); 1557 spin_unlock_irqrestore(q->queue_lock, flags);
1557 } 1558 }
1558 } 1559 }
1559 1560
1560 /* 1561 /*
1561 * The process that ioc belongs to has exited, we need to clean up 1562 * The process that ioc belongs to has exited, we need to clean up
1562 * and put the internal structures we have that belongs to that process. 1563 * and put the internal structures we have that belongs to that process.
1563 */ 1564 */
1564 static void cfq_exit_io_context(struct io_context *ioc) 1565 static void cfq_exit_io_context(struct io_context *ioc)
1565 { 1566 {
1566 call_for_each_cic(ioc, cfq_exit_single_io_context); 1567 call_for_each_cic(ioc, cfq_exit_single_io_context);
1567 } 1568 }
1568 1569
1569 static struct cfq_io_context * 1570 static struct cfq_io_context *
1570 cfq_alloc_io_context(struct cfq_data *cfqd, gfp_t gfp_mask) 1571 cfq_alloc_io_context(struct cfq_data *cfqd, gfp_t gfp_mask)
1571 { 1572 {
1572 struct cfq_io_context *cic; 1573 struct cfq_io_context *cic;
1573 1574
1574 cic = kmem_cache_alloc_node(cfq_ioc_pool, gfp_mask | __GFP_ZERO, 1575 cic = kmem_cache_alloc_node(cfq_ioc_pool, gfp_mask | __GFP_ZERO,
1575 cfqd->queue->node); 1576 cfqd->queue->node);
1576 if (cic) { 1577 if (cic) {
1577 cic->last_end_request = jiffies; 1578 cic->last_end_request = jiffies;
1578 INIT_LIST_HEAD(&cic->queue_list); 1579 INIT_LIST_HEAD(&cic->queue_list);
1579 INIT_HLIST_NODE(&cic->cic_list); 1580 INIT_HLIST_NODE(&cic->cic_list);
1580 cic->dtor = cfq_free_io_context; 1581 cic->dtor = cfq_free_io_context;
1581 cic->exit = cfq_exit_io_context; 1582 cic->exit = cfq_exit_io_context;
1582 elv_ioc_count_inc(cfq_ioc_count); 1583 elv_ioc_count_inc(cfq_ioc_count);
1583 } 1584 }
1584 1585
1585 return cic; 1586 return cic;
1586 } 1587 }
1587 1588
1588 static void cfq_init_prio_data(struct cfq_queue *cfqq, struct io_context *ioc) 1589 static void cfq_init_prio_data(struct cfq_queue *cfqq, struct io_context *ioc)
1589 { 1590 {
1590 struct task_struct *tsk = current; 1591 struct task_struct *tsk = current;
1591 int ioprio_class; 1592 int ioprio_class;
1592 1593
1593 if (!cfq_cfqq_prio_changed(cfqq)) 1594 if (!cfq_cfqq_prio_changed(cfqq))
1594 return; 1595 return;
1595 1596
1596 ioprio_class = IOPRIO_PRIO_CLASS(ioc->ioprio); 1597 ioprio_class = IOPRIO_PRIO_CLASS(ioc->ioprio);
1597 switch (ioprio_class) { 1598 switch (ioprio_class) {
1598 default: 1599 default:
1599 printk(KERN_ERR "cfq: bad prio %x\n", ioprio_class); 1600 printk(KERN_ERR "cfq: bad prio %x\n", ioprio_class);
1600 case IOPRIO_CLASS_NONE: 1601 case IOPRIO_CLASS_NONE:
1601 /* 1602 /*
1602 * no prio set, inherit CPU scheduling settings 1603 * no prio set, inherit CPU scheduling settings
1603 */ 1604 */
1604 cfqq->ioprio = task_nice_ioprio(tsk); 1605 cfqq->ioprio = task_nice_ioprio(tsk);
1605 cfqq->ioprio_class = task_nice_ioclass(tsk); 1606 cfqq->ioprio_class = task_nice_ioclass(tsk);
1606 break; 1607 break;
1607 case IOPRIO_CLASS_RT: 1608 case IOPRIO_CLASS_RT:
1608 cfqq->ioprio = task_ioprio(ioc); 1609 cfqq->ioprio = task_ioprio(ioc);
1609 cfqq->ioprio_class = IOPRIO_CLASS_RT; 1610 cfqq->ioprio_class = IOPRIO_CLASS_RT;
1610 break; 1611 break;
1611 case IOPRIO_CLASS_BE: 1612 case IOPRIO_CLASS_BE:
1612 cfqq->ioprio = task_ioprio(ioc); 1613 cfqq->ioprio = task_ioprio(ioc);
1613 cfqq->ioprio_class = IOPRIO_CLASS_BE; 1614 cfqq->ioprio_class = IOPRIO_CLASS_BE;
1614 break; 1615 break;
1615 case IOPRIO_CLASS_IDLE: 1616 case IOPRIO_CLASS_IDLE:
1616 cfqq->ioprio_class = IOPRIO_CLASS_IDLE; 1617 cfqq->ioprio_class = IOPRIO_CLASS_IDLE;
1617 cfqq->ioprio = 7; 1618 cfqq->ioprio = 7;
1618 cfq_clear_cfqq_idle_window(cfqq); 1619 cfq_clear_cfqq_idle_window(cfqq);
1619 break; 1620 break;
1620 } 1621 }
1621 1622
1622 /* 1623 /*
1623 * keep track of original prio settings in case we have to temporarily 1624 * keep track of original prio settings in case we have to temporarily
1624 * elevate the priority of this queue 1625 * elevate the priority of this queue
1625 */ 1626 */
1626 cfqq->org_ioprio = cfqq->ioprio; 1627 cfqq->org_ioprio = cfqq->ioprio;
1627 cfqq->org_ioprio_class = cfqq->ioprio_class; 1628 cfqq->org_ioprio_class = cfqq->ioprio_class;
1628 cfq_clear_cfqq_prio_changed(cfqq); 1629 cfq_clear_cfqq_prio_changed(cfqq);
1629 } 1630 }
1630 1631
1631 static void changed_ioprio(struct io_context *ioc, struct cfq_io_context *cic) 1632 static void changed_ioprio(struct io_context *ioc, struct cfq_io_context *cic)
1632 { 1633 {
1633 struct cfq_data *cfqd = cic->key; 1634 struct cfq_data *cfqd = cic->key;
1634 struct cfq_queue *cfqq; 1635 struct cfq_queue *cfqq;
1635 unsigned long flags; 1636 unsigned long flags;
1636 1637
1637 if (unlikely(!cfqd)) 1638 if (unlikely(!cfqd))
1638 return; 1639 return;
1639 1640
1640 spin_lock_irqsave(cfqd->queue->queue_lock, flags); 1641 spin_lock_irqsave(cfqd->queue->queue_lock, flags);
1641 1642
1642 cfqq = cic->cfqq[BLK_RW_ASYNC]; 1643 cfqq = cic->cfqq[BLK_RW_ASYNC];
1643 if (cfqq) { 1644 if (cfqq) {
1644 struct cfq_queue *new_cfqq; 1645 struct cfq_queue *new_cfqq;
1645 new_cfqq = cfq_get_queue(cfqd, BLK_RW_ASYNC, cic->ioc, 1646 new_cfqq = cfq_get_queue(cfqd, BLK_RW_ASYNC, cic->ioc,
1646 GFP_ATOMIC); 1647 GFP_ATOMIC);
1647 if (new_cfqq) { 1648 if (new_cfqq) {
1648 cic->cfqq[BLK_RW_ASYNC] = new_cfqq; 1649 cic->cfqq[BLK_RW_ASYNC] = new_cfqq;
1649 cfq_put_queue(cfqq); 1650 cfq_put_queue(cfqq);
1650 } 1651 }
1651 } 1652 }
1652 1653
1653 cfqq = cic->cfqq[BLK_RW_SYNC]; 1654 cfqq = cic->cfqq[BLK_RW_SYNC];
1654 if (cfqq) 1655 if (cfqq)
1655 cfq_mark_cfqq_prio_changed(cfqq); 1656 cfq_mark_cfqq_prio_changed(cfqq);
1656 1657
1657 spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); 1658 spin_unlock_irqrestore(cfqd->queue->queue_lock, flags);
1658 } 1659 }
1659 1660
1660 static void cfq_ioc_set_ioprio(struct io_context *ioc) 1661 static void cfq_ioc_set_ioprio(struct io_context *ioc)
1661 { 1662 {
1662 call_for_each_cic(ioc, changed_ioprio); 1663 call_for_each_cic(ioc, changed_ioprio);
1663 ioc->ioprio_changed = 0; 1664 ioc->ioprio_changed = 0;
1664 } 1665 }
1665 1666
1666 static void cfq_init_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq, 1667 static void cfq_init_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq,
1667 pid_t pid, int is_sync) 1668 pid_t pid, int is_sync)
1668 { 1669 {
1669 RB_CLEAR_NODE(&cfqq->rb_node); 1670 RB_CLEAR_NODE(&cfqq->rb_node);
1670 RB_CLEAR_NODE(&cfqq->p_node); 1671 RB_CLEAR_NODE(&cfqq->p_node);
1671 INIT_LIST_HEAD(&cfqq->fifo); 1672 INIT_LIST_HEAD(&cfqq->fifo);
1672 1673
1673 atomic_set(&cfqq->ref, 0); 1674 atomic_set(&cfqq->ref, 0);
1674 cfqq->cfqd = cfqd; 1675 cfqq->cfqd = cfqd;
1675 1676
1676 cfq_mark_cfqq_prio_changed(cfqq); 1677 cfq_mark_cfqq_prio_changed(cfqq);
1677 1678
1678 if (is_sync) { 1679 if (is_sync) {
1679 if (!cfq_class_idle(cfqq)) 1680 if (!cfq_class_idle(cfqq))
1680 cfq_mark_cfqq_idle_window(cfqq); 1681 cfq_mark_cfqq_idle_window(cfqq);
1681 cfq_mark_cfqq_sync(cfqq); 1682 cfq_mark_cfqq_sync(cfqq);
1682 } 1683 }
1683 cfqq->pid = pid; 1684 cfqq->pid = pid;
1684 } 1685 }
1685 1686
1686 static struct cfq_queue * 1687 static struct cfq_queue *
1687 cfq_find_alloc_queue(struct cfq_data *cfqd, int is_sync, 1688 cfq_find_alloc_queue(struct cfq_data *cfqd, int is_sync,
1688 struct io_context *ioc, gfp_t gfp_mask) 1689 struct io_context *ioc, gfp_t gfp_mask)
1689 { 1690 {
1690 struct cfq_queue *cfqq, *new_cfqq = NULL; 1691 struct cfq_queue *cfqq, *new_cfqq = NULL;
1691 struct cfq_io_context *cic; 1692 struct cfq_io_context *cic;
1692 1693
1693 retry: 1694 retry:
1694 cic = cfq_cic_lookup(cfqd, ioc); 1695 cic = cfq_cic_lookup(cfqd, ioc);
1695 /* cic always exists here */ 1696 /* cic always exists here */
1696 cfqq = cic_to_cfqq(cic, is_sync); 1697 cfqq = cic_to_cfqq(cic, is_sync);
1697 1698
1698 /* 1699 /*
1699 * Always try a new alloc if we fell back to the OOM cfqq 1700 * Always try a new alloc if we fell back to the OOM cfqq
1700 * originally, since it should just be a temporary situation. 1701 * originally, since it should just be a temporary situation.
1701 */ 1702 */
1702 if (!cfqq || cfqq == &cfqd->oom_cfqq) { 1703 if (!cfqq || cfqq == &cfqd->oom_cfqq) {
1703 cfqq = NULL; 1704 cfqq = NULL;
1704 if (new_cfqq) { 1705 if (new_cfqq) {
1705 cfqq = new_cfqq; 1706 cfqq = new_cfqq;
1706 new_cfqq = NULL; 1707 new_cfqq = NULL;
1707 } else if (gfp_mask & __GFP_WAIT) { 1708 } else if (gfp_mask & __GFP_WAIT) {
1708 spin_unlock_irq(cfqd->queue->queue_lock); 1709 spin_unlock_irq(cfqd->queue->queue_lock);
1709 new_cfqq = kmem_cache_alloc_node(cfq_pool, 1710 new_cfqq = kmem_cache_alloc_node(cfq_pool,
1710 gfp_mask | __GFP_ZERO, 1711 gfp_mask | __GFP_ZERO,
1711 cfqd->queue->node); 1712 cfqd->queue->node);
1712 spin_lock_irq(cfqd->queue->queue_lock); 1713 spin_lock_irq(cfqd->queue->queue_lock);
1713 if (new_cfqq) 1714 if (new_cfqq)
1714 goto retry; 1715 goto retry;
1715 } else { 1716 } else {
1716 cfqq = kmem_cache_alloc_node(cfq_pool, 1717 cfqq = kmem_cache_alloc_node(cfq_pool,
1717 gfp_mask | __GFP_ZERO, 1718 gfp_mask | __GFP_ZERO,
1718 cfqd->queue->node); 1719 cfqd->queue->node);
1719 } 1720 }
1720 1721
1721 if (cfqq) { 1722 if (cfqq) {
1722 cfq_init_cfqq(cfqd, cfqq, current->pid, is_sync); 1723 cfq_init_cfqq(cfqd, cfqq, current->pid, is_sync);
1723 cfq_init_prio_data(cfqq, ioc); 1724 cfq_init_prio_data(cfqq, ioc);
1724 cfq_log_cfqq(cfqd, cfqq, "alloced"); 1725 cfq_log_cfqq(cfqd, cfqq, "alloced");
1725 } else 1726 } else
1726 cfqq = &cfqd->oom_cfqq; 1727 cfqq = &cfqd->oom_cfqq;
1727 } 1728 }
1728 1729
1729 if (new_cfqq) 1730 if (new_cfqq)
1730 kmem_cache_free(cfq_pool, new_cfqq); 1731 kmem_cache_free(cfq_pool, new_cfqq);
1731 1732
1732 return cfqq; 1733 return cfqq;
1733 } 1734 }
1734 1735
1735 static struct cfq_queue ** 1736 static struct cfq_queue **
1736 cfq_async_queue_prio(struct cfq_data *cfqd, int ioprio_class, int ioprio) 1737 cfq_async_queue_prio(struct cfq_data *cfqd, int ioprio_class, int ioprio)
1737 { 1738 {
1738 switch (ioprio_class) { 1739 switch (ioprio_class) {
1739 case IOPRIO_CLASS_RT: 1740 case IOPRIO_CLASS_RT:
1740 return &cfqd->async_cfqq[0][ioprio]; 1741 return &cfqd->async_cfqq[0][ioprio];
1741 case IOPRIO_CLASS_BE: 1742 case IOPRIO_CLASS_BE:
1742 return &cfqd->async_cfqq[1][ioprio]; 1743 return &cfqd->async_cfqq[1][ioprio];
1743 case IOPRIO_CLASS_IDLE: 1744 case IOPRIO_CLASS_IDLE:
1744 return &cfqd->async_idle_cfqq; 1745 return &cfqd->async_idle_cfqq;
1745 default: 1746 default:
1746 BUG(); 1747 BUG();
1747 } 1748 }
1748 } 1749 }
1749 1750
1750 static struct cfq_queue * 1751 static struct cfq_queue *
1751 cfq_get_queue(struct cfq_data *cfqd, int is_sync, struct io_context *ioc, 1752 cfq_get_queue(struct cfq_data *cfqd, int is_sync, struct io_context *ioc,
1752 gfp_t gfp_mask) 1753 gfp_t gfp_mask)
1753 { 1754 {
1754 const int ioprio = task_ioprio(ioc); 1755 const int ioprio = task_ioprio(ioc);
1755 const int ioprio_class = task_ioprio_class(ioc); 1756 const int ioprio_class = task_ioprio_class(ioc);
1756 struct cfq_queue **async_cfqq = NULL; 1757 struct cfq_queue **async_cfqq = NULL;
1757 struct cfq_queue *cfqq = NULL; 1758 struct cfq_queue *cfqq = NULL;
1758 1759
1759 if (!is_sync) { 1760 if (!is_sync) {
1760 async_cfqq = cfq_async_queue_prio(cfqd, ioprio_class, ioprio); 1761 async_cfqq = cfq_async_queue_prio(cfqd, ioprio_class, ioprio);
1761 cfqq = *async_cfqq; 1762 cfqq = *async_cfqq;
1762 } 1763 }
1763 1764
1764 if (!cfqq) 1765 if (!cfqq)
1765 cfqq = cfq_find_alloc_queue(cfqd, is_sync, ioc, gfp_mask); 1766 cfqq = cfq_find_alloc_queue(cfqd, is_sync, ioc, gfp_mask);
1766 1767
1767 /* 1768 /*
1768 * pin the queue now that it's allocated, scheduler exit will prune it 1769 * pin the queue now that it's allocated, scheduler exit will prune it
1769 */ 1770 */
1770 if (!is_sync && !(*async_cfqq)) { 1771 if (!is_sync && !(*async_cfqq)) {
1771 atomic_inc(&cfqq->ref); 1772 atomic_inc(&cfqq->ref);
1772 *async_cfqq = cfqq; 1773 *async_cfqq = cfqq;
1773 } 1774 }
1774 1775
1775 atomic_inc(&cfqq->ref); 1776 atomic_inc(&cfqq->ref);
1776 return cfqq; 1777 return cfqq;
1777 } 1778 }
1778 1779
1779 /* 1780 /*
1780 * We drop cfq io contexts lazily, so we may find a dead one. 1781 * We drop cfq io contexts lazily, so we may find a dead one.
1781 */ 1782 */
1782 static void 1783 static void
1783 cfq_drop_dead_cic(struct cfq_data *cfqd, struct io_context *ioc, 1784 cfq_drop_dead_cic(struct cfq_data *cfqd, struct io_context *ioc,
1784 struct cfq_io_context *cic) 1785 struct cfq_io_context *cic)
1785 { 1786 {
1786 unsigned long flags; 1787 unsigned long flags;
1787 1788
1788 WARN_ON(!list_empty(&cic->queue_list)); 1789 WARN_ON(!list_empty(&cic->queue_list));
1789 1790
1790 spin_lock_irqsave(&ioc->lock, flags); 1791 spin_lock_irqsave(&ioc->lock, flags);
1791 1792
1792 BUG_ON(ioc->ioc_data == cic); 1793 BUG_ON(ioc->ioc_data == cic);
1793 1794
1794 radix_tree_delete(&ioc->radix_root, (unsigned long) cfqd); 1795 radix_tree_delete(&ioc->radix_root, (unsigned long) cfqd);
1795 hlist_del_rcu(&cic->cic_list); 1796 hlist_del_rcu(&cic->cic_list);
1796 spin_unlock_irqrestore(&ioc->lock, flags); 1797 spin_unlock_irqrestore(&ioc->lock, flags);
1797 1798
1798 cfq_cic_free(cic); 1799 cfq_cic_free(cic);
1799 } 1800 }
1800 1801
1801 static struct cfq_io_context * 1802 static struct cfq_io_context *
1802 cfq_cic_lookup(struct cfq_data *cfqd, struct io_context *ioc) 1803 cfq_cic_lookup(struct cfq_data *cfqd, struct io_context *ioc)
1803 { 1804 {
1804 struct cfq_io_context *cic; 1805 struct cfq_io_context *cic;
1805 unsigned long flags; 1806 unsigned long flags;
1806 void *k; 1807 void *k;
1807 1808
1808 if (unlikely(!ioc)) 1809 if (unlikely(!ioc))
1809 return NULL; 1810 return NULL;
1810 1811
1811 rcu_read_lock(); 1812 rcu_read_lock();
1812 1813
1813 /* 1814 /*
1814 * we maintain a last-hit cache, to avoid browsing over the tree 1815 * we maintain a last-hit cache, to avoid browsing over the tree
1815 */ 1816 */
1816 cic = rcu_dereference(ioc->ioc_data); 1817 cic = rcu_dereference(ioc->ioc_data);
1817 if (cic && cic->key == cfqd) { 1818 if (cic && cic->key == cfqd) {
1818 rcu_read_unlock(); 1819 rcu_read_unlock();
1819 return cic; 1820 return cic;
1820 } 1821 }
1821 1822
1822 do { 1823 do {
1823 cic = radix_tree_lookup(&ioc->radix_root, (unsigned long) cfqd); 1824 cic = radix_tree_lookup(&ioc->radix_root, (unsigned long) cfqd);
1824 rcu_read_unlock(); 1825 rcu_read_unlock();
1825 if (!cic) 1826 if (!cic)
1826 break; 1827 break;
1827 /* ->key must be copied to avoid race with cfq_exit_queue() */ 1828 /* ->key must be copied to avoid race with cfq_exit_queue() */
1828 k = cic->key; 1829 k = cic->key;
1829 if (unlikely(!k)) { 1830 if (unlikely(!k)) {
1830 cfq_drop_dead_cic(cfqd, ioc, cic); 1831 cfq_drop_dead_cic(cfqd, ioc, cic);
1831 rcu_read_lock(); 1832 rcu_read_lock();
1832 continue; 1833 continue;
1833 } 1834 }
1834 1835
1835 spin_lock_irqsave(&ioc->lock, flags); 1836 spin_lock_irqsave(&ioc->lock, flags);
1836 rcu_assign_pointer(ioc->ioc_data, cic); 1837 rcu_assign_pointer(ioc->ioc_data, cic);
1837 spin_unlock_irqrestore(&ioc->lock, flags); 1838 spin_unlock_irqrestore(&ioc->lock, flags);
1838 break; 1839 break;
1839 } while (1); 1840 } while (1);
1840 1841
1841 return cic; 1842 return cic;
1842 } 1843 }
1843 1844
1844 /* 1845 /*
1845 * Add cic into ioc, using cfqd as the search key. This enables us to lookup 1846 * Add cic into ioc, using cfqd as the search key. This enables us to lookup
1846 * the process specific cfq io context when entered from the block layer. 1847 * the process specific cfq io context when entered from the block layer.
1847 * Also adds the cic to a per-cfqd list, used when this queue is removed. 1848 * Also adds the cic to a per-cfqd list, used when this queue is removed.
1848 */ 1849 */
1849 static int cfq_cic_link(struct cfq_data *cfqd, struct io_context *ioc, 1850 static int cfq_cic_link(struct cfq_data *cfqd, struct io_context *ioc,
1850 struct cfq_io_context *cic, gfp_t gfp_mask) 1851 struct cfq_io_context *cic, gfp_t gfp_mask)
1851 { 1852 {
1852 unsigned long flags; 1853 unsigned long flags;
1853 int ret; 1854 int ret;
1854 1855
1855 ret = radix_tree_preload(gfp_mask); 1856 ret = radix_tree_preload(gfp_mask);
1856 if (!ret) { 1857 if (!ret) {
1857 cic->ioc = ioc; 1858 cic->ioc = ioc;
1858 cic->key = cfqd; 1859 cic->key = cfqd;
1859 1860
1860 spin_lock_irqsave(&ioc->lock, flags); 1861 spin_lock_irqsave(&ioc->lock, flags);
1861 ret = radix_tree_insert(&ioc->radix_root, 1862 ret = radix_tree_insert(&ioc->radix_root,
1862 (unsigned long) cfqd, cic); 1863 (unsigned long) cfqd, cic);
1863 if (!ret) 1864 if (!ret)
1864 hlist_add_head_rcu(&cic->cic_list, &ioc->cic_list); 1865 hlist_add_head_rcu(&cic->cic_list, &ioc->cic_list);
1865 spin_unlock_irqrestore(&ioc->lock, flags); 1866 spin_unlock_irqrestore(&ioc->lock, flags);
1866 1867
1867 radix_tree_preload_end(); 1868 radix_tree_preload_end();
1868 1869
1869 if (!ret) { 1870 if (!ret) {
1870 spin_lock_irqsave(cfqd->queue->queue_lock, flags); 1871 spin_lock_irqsave(cfqd->queue->queue_lock, flags);
1871 list_add(&cic->queue_list, &cfqd->cic_list); 1872 list_add(&cic->queue_list, &cfqd->cic_list);
1872 spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); 1873 spin_unlock_irqrestore(cfqd->queue->queue_lock, flags);
1873 } 1874 }
1874 } 1875 }
1875 1876
1876 if (ret) 1877 if (ret)
1877 printk(KERN_ERR "cfq: cic link failed!\n"); 1878 printk(KERN_ERR "cfq: cic link failed!\n");
1878 1879
1879 return ret; 1880 return ret;
1880 } 1881 }
1881 1882
1882 /* 1883 /*
1883 * Setup general io context and cfq io context. There can be several cfq 1884 * Setup general io context and cfq io context. There can be several cfq
1884 * io contexts per general io context, if this process is doing io to more 1885 * io contexts per general io context, if this process is doing io to more
1885 * than one device managed by cfq. 1886 * than one device managed by cfq.
1886 */ 1887 */
1887 static struct cfq_io_context * 1888 static struct cfq_io_context *
1888 cfq_get_io_context(struct cfq_data *cfqd, gfp_t gfp_mask) 1889 cfq_get_io_context(struct cfq_data *cfqd, gfp_t gfp_mask)
1889 { 1890 {
1890 struct io_context *ioc = NULL; 1891 struct io_context *ioc = NULL;
1891 struct cfq_io_context *cic; 1892 struct cfq_io_context *cic;
1892 1893
1893 might_sleep_if(gfp_mask & __GFP_WAIT); 1894 might_sleep_if(gfp_mask & __GFP_WAIT);
1894 1895
1895 ioc = get_io_context(gfp_mask, cfqd->queue->node); 1896 ioc = get_io_context(gfp_mask, cfqd->queue->node);
1896 if (!ioc) 1897 if (!ioc)
1897 return NULL; 1898 return NULL;
1898 1899
1899 cic = cfq_cic_lookup(cfqd, ioc); 1900 cic = cfq_cic_lookup(cfqd, ioc);
1900 if (cic) 1901 if (cic)
1901 goto out; 1902 goto out;
1902 1903
1903 cic = cfq_alloc_io_context(cfqd, gfp_mask); 1904 cic = cfq_alloc_io_context(cfqd, gfp_mask);
1904 if (cic == NULL) 1905 if (cic == NULL)
1905 goto err; 1906 goto err;
1906 1907
1907 if (cfq_cic_link(cfqd, ioc, cic, gfp_mask)) 1908 if (cfq_cic_link(cfqd, ioc, cic, gfp_mask))
1908 goto err_free; 1909 goto err_free;
1909 1910
1910 out: 1911 out:
1911 smp_read_barrier_depends(); 1912 smp_read_barrier_depends();
1912 if (unlikely(ioc->ioprio_changed)) 1913 if (unlikely(ioc->ioprio_changed))
1913 cfq_ioc_set_ioprio(ioc); 1914 cfq_ioc_set_ioprio(ioc);
1914 1915
1915 return cic; 1916 return cic;
1916 err_free: 1917 err_free:
1917 cfq_cic_free(cic); 1918 cfq_cic_free(cic);
1918 err: 1919 err:
1919 put_io_context(ioc); 1920 put_io_context(ioc);
1920 return NULL; 1921 return NULL;
1921 } 1922 }
1922 1923
1923 static void 1924 static void
1924 cfq_update_io_thinktime(struct cfq_data *cfqd, struct cfq_io_context *cic) 1925 cfq_update_io_thinktime(struct cfq_data *cfqd, struct cfq_io_context *cic)
1925 { 1926 {
1926 unsigned long elapsed = jiffies - cic->last_end_request; 1927 unsigned long elapsed = jiffies - cic->last_end_request;
1927 unsigned long ttime = min(elapsed, 2UL * cfqd->cfq_slice_idle); 1928 unsigned long ttime = min(elapsed, 2UL * cfqd->cfq_slice_idle);
1928 1929
1929 cic->ttime_samples = (7*cic->ttime_samples + 256) / 8; 1930 cic->ttime_samples = (7*cic->ttime_samples + 256) / 8;
1930 cic->ttime_total = (7*cic->ttime_total + 256*ttime) / 8; 1931 cic->ttime_total = (7*cic->ttime_total + 256*ttime) / 8;
1931 cic->ttime_mean = (cic->ttime_total + 128) / cic->ttime_samples; 1932 cic->ttime_mean = (cic->ttime_total + 128) / cic->ttime_samples;
1932 } 1933 }
1933 1934
1934 static void 1935 static void
1935 cfq_update_io_seektime(struct cfq_data *cfqd, struct cfq_io_context *cic, 1936 cfq_update_io_seektime(struct cfq_data *cfqd, struct cfq_io_context *cic,
1936 struct request *rq) 1937 struct request *rq)
1937 { 1938 {
1938 sector_t sdist; 1939 sector_t sdist;
1939 u64 total; 1940 u64 total;
1940 1941
1941 if (!cic->last_request_pos) 1942 if (!cic->last_request_pos)
1942 sdist = 0; 1943 sdist = 0;
1943 else if (cic->last_request_pos < blk_rq_pos(rq)) 1944 else if (cic->last_request_pos < blk_rq_pos(rq))
1944 sdist = blk_rq_pos(rq) - cic->last_request_pos; 1945 sdist = blk_rq_pos(rq) - cic->last_request_pos;
1945 else 1946 else
1946 sdist = cic->last_request_pos - blk_rq_pos(rq); 1947 sdist = cic->last_request_pos - blk_rq_pos(rq);
1947 1948
1948 /* 1949 /*
1949 * Don't allow the seek distance to get too large from the 1950 * Don't allow the seek distance to get too large from the
1950 * odd fragment, pagein, etc 1951 * odd fragment, pagein, etc
1951 */ 1952 */
1952 if (cic->seek_samples <= 60) /* second&third seek */ 1953 if (cic->seek_samples <= 60) /* second&third seek */
1953 sdist = min(sdist, (cic->seek_mean * 4) + 2*1024*1024); 1954 sdist = min(sdist, (cic->seek_mean * 4) + 2*1024*1024);
1954 else 1955 else
1955 sdist = min(sdist, (cic->seek_mean * 4) + 2*1024*64); 1956 sdist = min(sdist, (cic->seek_mean * 4) + 2*1024*64);
1956 1957
1957 cic->seek_samples = (7*cic->seek_samples + 256) / 8; 1958 cic->seek_samples = (7*cic->seek_samples + 256) / 8;
1958 cic->seek_total = (7*cic->seek_total + (u64)256*sdist) / 8; 1959 cic->seek_total = (7*cic->seek_total + (u64)256*sdist) / 8;
1959 total = cic->seek_total + (cic->seek_samples/2); 1960 total = cic->seek_total + (cic->seek_samples/2);
1960 do_div(total, cic->seek_samples); 1961 do_div(total, cic->seek_samples);
1961 cic->seek_mean = (sector_t)total; 1962 cic->seek_mean = (sector_t)total;
1962 } 1963 }
1963 1964
1964 /* 1965 /*
1965 * Disable idle window if the process thinks too long or seeks so much that 1966 * Disable idle window if the process thinks too long or seeks so much that
1966 * it doesn't matter 1967 * it doesn't matter
1967 */ 1968 */
1968 static void 1969 static void
1969 cfq_update_idle_window(struct cfq_data *cfqd, struct cfq_queue *cfqq, 1970 cfq_update_idle_window(struct cfq_data *cfqd, struct cfq_queue *cfqq,
1970 struct cfq_io_context *cic) 1971 struct cfq_io_context *cic)
1971 { 1972 {
1972 int old_idle, enable_idle; 1973 int old_idle, enable_idle;
1973 1974
1974 /* 1975 /*
1975 * Don't idle for async or idle io prio class 1976 * Don't idle for async or idle io prio class
1976 */ 1977 */
1977 if (!cfq_cfqq_sync(cfqq) || cfq_class_idle(cfqq)) 1978 if (!cfq_cfqq_sync(cfqq) || cfq_class_idle(cfqq))
1978 return; 1979 return;
1979 1980
1980 enable_idle = old_idle = cfq_cfqq_idle_window(cfqq); 1981 enable_idle = old_idle = cfq_cfqq_idle_window(cfqq);
1981 1982
1982 if (!atomic_read(&cic->ioc->nr_tasks) || !cfqd->cfq_slice_idle || 1983 if (!atomic_read(&cic->ioc->nr_tasks) || !cfqd->cfq_slice_idle ||
1983 (!cfqd->cfq_latency && cfqd->hw_tag && CIC_SEEKY(cic))) 1984 (!cfqd->cfq_latency && cfqd->hw_tag && CIC_SEEKY(cic)))
1984 enable_idle = 0; 1985 enable_idle = 0;
1985 else if (sample_valid(cic->ttime_samples)) { 1986 else if (sample_valid(cic->ttime_samples)) {
1986 if (cic->ttime_mean > cfqd->cfq_slice_idle) 1987 if (cic->ttime_mean > cfqd->cfq_slice_idle)
1987 enable_idle = 0; 1988 enable_idle = 0;
1988 else 1989 else
1989 enable_idle = 1; 1990 enable_idle = 1;
1990 } 1991 }
1991 1992
1992 if (old_idle != enable_idle) { 1993 if (old_idle != enable_idle) {
1993 cfq_log_cfqq(cfqd, cfqq, "idle=%d", enable_idle); 1994 cfq_log_cfqq(cfqd, cfqq, "idle=%d", enable_idle);
1994 if (enable_idle) 1995 if (enable_idle)
1995 cfq_mark_cfqq_idle_window(cfqq); 1996 cfq_mark_cfqq_idle_window(cfqq);
1996 else 1997 else
1997 cfq_clear_cfqq_idle_window(cfqq); 1998 cfq_clear_cfqq_idle_window(cfqq);
1998 } 1999 }
1999 } 2000 }
2000 2001
2001 /* 2002 /*
2002 * Check if new_cfqq should preempt the currently active queue. Return 0 for 2003 * Check if new_cfqq should preempt the currently active queue. Return 0 for
2003 * no or if we aren't sure, a 1 will cause a preempt. 2004 * no or if we aren't sure, a 1 will cause a preempt.
2004 */ 2005 */
2005 static int 2006 static int
2006 cfq_should_preempt(struct cfq_data *cfqd, struct cfq_queue *new_cfqq, 2007 cfq_should_preempt(struct cfq_data *cfqd, struct cfq_queue *new_cfqq,
2007 struct request *rq) 2008 struct request *rq)
2008 { 2009 {
2009 struct cfq_queue *cfqq; 2010 struct cfq_queue *cfqq;
2010 2011
2011 cfqq = cfqd->active_queue; 2012 cfqq = cfqd->active_queue;
2012 if (!cfqq) 2013 if (!cfqq)
2013 return 0; 2014 return 0;
2014 2015
2015 if (cfq_slice_used(cfqq)) 2016 if (cfq_slice_used(cfqq))
2016 return 1; 2017 return 1;
2017 2018
2018 if (cfq_class_idle(new_cfqq)) 2019 if (cfq_class_idle(new_cfqq))
2019 return 0; 2020 return 0;
2020 2021
2021 if (cfq_class_idle(cfqq)) 2022 if (cfq_class_idle(cfqq))
2022 return 1; 2023 return 1;
2023 2024
2024 /* 2025 /*
2025 * if the new request is sync, but the currently running queue is 2026 * if the new request is sync, but the currently running queue is
2026 * not, let the sync request have priority. 2027 * not, let the sync request have priority.
2027 */ 2028 */
2028 if (rq_is_sync(rq) && !cfq_cfqq_sync(cfqq)) 2029 if (rq_is_sync(rq) && !cfq_cfqq_sync(cfqq))
2029 return 1; 2030 return 1;
2030 2031
2031 /* 2032 /*
2032 * So both queues are sync. Let the new request get disk time if 2033 * So both queues are sync. Let the new request get disk time if
2033 * it's a metadata request and the current queue is doing regular IO. 2034 * it's a metadata request and the current queue is doing regular IO.
2034 */ 2035 */
2035 if (rq_is_meta(rq) && !cfqq->meta_pending) 2036 if (rq_is_meta(rq) && !cfqq->meta_pending)
2036 return 1; 2037 return 1;
2037 2038
2038 /* 2039 /*
2039 * Allow an RT request to pre-empt an ongoing non-RT cfqq timeslice. 2040 * Allow an RT request to pre-empt an ongoing non-RT cfqq timeslice.
2040 */ 2041 */
2041 if (cfq_class_rt(new_cfqq) && !cfq_class_rt(cfqq)) 2042 if (cfq_class_rt(new_cfqq) && !cfq_class_rt(cfqq))
2042 return 1; 2043 return 1;
2043 2044
2044 if (!cfqd->active_cic || !cfq_cfqq_wait_request(cfqq)) 2045 if (!cfqd->active_cic || !cfq_cfqq_wait_request(cfqq))
2045 return 0; 2046 return 0;
2046 2047
2047 /* 2048 /*
2048 * if this request is as-good as one we would expect from the 2049 * if this request is as-good as one we would expect from the
2049 * current cfqq, let it preempt 2050 * current cfqq, let it preempt
2050 */ 2051 */
2051 if (cfq_rq_close(cfqd, rq)) 2052 if (cfq_rq_close(cfqd, rq))
2052 return 1; 2053 return 1;
2053 2054
2054 return 0; 2055 return 0;
2055 } 2056 }
2056 2057
2057 /* 2058 /*
2058 * cfqq preempts the active queue. if we allowed preempt with no slice left, 2059 * cfqq preempts the active queue. if we allowed preempt with no slice left,
2059 * let it have half of its nominal slice. 2060 * let it have half of its nominal slice.
2060 */ 2061 */
2061 static void cfq_preempt_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq) 2062 static void cfq_preempt_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq)
2062 { 2063 {
2063 cfq_log_cfqq(cfqd, cfqq, "preempt"); 2064 cfq_log_cfqq(cfqd, cfqq, "preempt");
2064 cfq_slice_expired(cfqd, 1); 2065 cfq_slice_expired(cfqd, 1);
2065 2066
2066 /* 2067 /*
2067 * Put the new queue at the front of the of the current list, 2068 * Put the new queue at the front of the of the current list,
2068 * so we know that it will be selected next. 2069 * so we know that it will be selected next.
2069 */ 2070 */
2070 BUG_ON(!cfq_cfqq_on_rr(cfqq)); 2071 BUG_ON(!cfq_cfqq_on_rr(cfqq));
2071 2072
2072 cfq_service_tree_add(cfqd, cfqq, 1); 2073 cfq_service_tree_add(cfqd, cfqq, 1);
2073 2074
2074 cfqq->slice_end = 0; 2075 cfqq->slice_end = 0;
2075 cfq_mark_cfqq_slice_new(cfqq); 2076 cfq_mark_cfqq_slice_new(cfqq);
2076 } 2077 }
2077 2078
2078 /* 2079 /*
2079 * Called when a new fs request (rq) is added (to cfqq). Check if there's 2080 * Called when a new fs request (rq) is added (to cfqq). Check if there's
2080 * something we should do about it 2081 * something we should do about it
2081 */ 2082 */
2082 static void 2083 static void
2083 cfq_rq_enqueued(struct cfq_data *cfqd, struct cfq_queue *cfqq, 2084 cfq_rq_enqueued(struct cfq_data *cfqd, struct cfq_queue *cfqq,
2084 struct request *rq) 2085 struct request *rq)
2085 { 2086 {
2086 struct cfq_io_context *cic = RQ_CIC(rq); 2087 struct cfq_io_context *cic = RQ_CIC(rq);
2087 2088
2088 cfqd->rq_queued++; 2089 cfqd->rq_queued++;
2089 if (rq_is_meta(rq)) 2090 if (rq_is_meta(rq))
2090 cfqq->meta_pending++; 2091 cfqq->meta_pending++;
2091 2092
2092 cfq_update_io_thinktime(cfqd, cic); 2093 cfq_update_io_thinktime(cfqd, cic);
2093 cfq_update_io_seektime(cfqd, cic, rq); 2094 cfq_update_io_seektime(cfqd, cic, rq);
2094 cfq_update_idle_window(cfqd, cfqq, cic); 2095 cfq_update_idle_window(cfqd, cfqq, cic);
2095 2096
2096 cic->last_request_pos = blk_rq_pos(rq) + blk_rq_sectors(rq); 2097 cic->last_request_pos = blk_rq_pos(rq) + blk_rq_sectors(rq);
2097 2098
2098 if (cfqq == cfqd->active_queue) { 2099 if (cfqq == cfqd->active_queue) {
2099 /* 2100 /*
2100 * Remember that we saw a request from this process, but 2101 * Remember that we saw a request from this process, but
2101 * don't start queuing just yet. Otherwise we risk seeing lots 2102 * don't start queuing just yet. Otherwise we risk seeing lots
2102 * of tiny requests, because we disrupt the normal plugging 2103 * of tiny requests, because we disrupt the normal plugging
2103 * and merging. If the request is already larger than a single 2104 * and merging. If the request is already larger than a single
2104 * page, let it rip immediately. For that case we assume that 2105 * page, let it rip immediately. For that case we assume that
2105 * merging is already done. Ditto for a busy system that 2106 * merging is already done. Ditto for a busy system that
2106 * has other work pending, don't risk delaying until the 2107 * has other work pending, don't risk delaying until the
2107 * idle timer unplug to continue working. 2108 * idle timer unplug to continue working.
2108 */ 2109 */
2109 if (cfq_cfqq_wait_request(cfqq)) { 2110 if (cfq_cfqq_wait_request(cfqq)) {
2110 if (blk_rq_bytes(rq) > PAGE_CACHE_SIZE || 2111 if (blk_rq_bytes(rq) > PAGE_CACHE_SIZE ||
2111 cfqd->busy_queues > 1) { 2112 cfqd->busy_queues > 1) {
2112 del_timer(&cfqd->idle_slice_timer); 2113 del_timer(&cfqd->idle_slice_timer);
2113 __blk_run_queue(cfqd->queue); 2114 __blk_run_queue(cfqd->queue);
2114 } 2115 }
2115 cfq_mark_cfqq_must_dispatch(cfqq); 2116 cfq_mark_cfqq_must_dispatch(cfqq);
2116 } 2117 }
2117 } else if (cfq_should_preempt(cfqd, cfqq, rq)) { 2118 } else if (cfq_should_preempt(cfqd, cfqq, rq)) {
2118 /* 2119 /*
2119 * not the active queue - expire current slice if it is 2120 * not the active queue - expire current slice if it is
2120 * idle and has expired it's mean thinktime or this new queue 2121 * idle and has expired it's mean thinktime or this new queue
2121 * has some old slice time left and is of higher priority or 2122 * has some old slice time left and is of higher priority or
2122 * this new queue is RT and the current one is BE 2123 * this new queue is RT and the current one is BE
2123 */ 2124 */
2124 cfq_preempt_queue(cfqd, cfqq); 2125 cfq_preempt_queue(cfqd, cfqq);
2125 __blk_run_queue(cfqd->queue); 2126 __blk_run_queue(cfqd->queue);
2126 } 2127 }
2127 } 2128 }
2128 2129
2129 static void cfq_insert_request(struct request_queue *q, struct request *rq) 2130 static void cfq_insert_request(struct request_queue *q, struct request *rq)
2130 { 2131 {
2131 struct cfq_data *cfqd = q->elevator->elevator_data; 2132 struct cfq_data *cfqd = q->elevator->elevator_data;
2132 struct cfq_queue *cfqq = RQ_CFQQ(rq); 2133 struct cfq_queue *cfqq = RQ_CFQQ(rq);
2133 2134
2134 cfq_log_cfqq(cfqd, cfqq, "insert_request"); 2135 cfq_log_cfqq(cfqd, cfqq, "insert_request");
2135 cfq_init_prio_data(cfqq, RQ_CIC(rq)->ioc); 2136 cfq_init_prio_data(cfqq, RQ_CIC(rq)->ioc);
2136 2137
2137 cfq_add_rq_rb(rq); 2138 cfq_add_rq_rb(rq);
2138 2139
2139 list_add_tail(&rq->queuelist, &cfqq->fifo); 2140 list_add_tail(&rq->queuelist, &cfqq->fifo);
2140 2141
2141 cfq_rq_enqueued(cfqd, cfqq, rq); 2142 cfq_rq_enqueued(cfqd, cfqq, rq);
2142 } 2143 }
2143 2144
2144 /* 2145 /*
2145 * Update hw_tag based on peak queue depth over 50 samples under 2146 * Update hw_tag based on peak queue depth over 50 samples under
2146 * sufficient load. 2147 * sufficient load.
2147 */ 2148 */
2148 static void cfq_update_hw_tag(struct cfq_data *cfqd) 2149 static void cfq_update_hw_tag(struct cfq_data *cfqd)
2149 { 2150 {
2150 if (rq_in_driver(cfqd) > cfqd->rq_in_driver_peak) 2151 if (rq_in_driver(cfqd) > cfqd->rq_in_driver_peak)
2151 cfqd->rq_in_driver_peak = rq_in_driver(cfqd); 2152 cfqd->rq_in_driver_peak = rq_in_driver(cfqd);
2152 2153
2153 if (cfqd->rq_queued <= CFQ_HW_QUEUE_MIN && 2154 if (cfqd->rq_queued <= CFQ_HW_QUEUE_MIN &&
2154 rq_in_driver(cfqd) <= CFQ_HW_QUEUE_MIN) 2155 rq_in_driver(cfqd) <= CFQ_HW_QUEUE_MIN)
2155 return; 2156 return;
2156 2157
2157 if (cfqd->hw_tag_samples++ < 50) 2158 if (cfqd->hw_tag_samples++ < 50)
2158 return; 2159 return;
2159 2160
2160 if (cfqd->rq_in_driver_peak >= CFQ_HW_QUEUE_MIN) 2161 if (cfqd->rq_in_driver_peak >= CFQ_HW_QUEUE_MIN)
2161 cfqd->hw_tag = 1; 2162 cfqd->hw_tag = 1;
2162 else 2163 else
2163 cfqd->hw_tag = 0; 2164 cfqd->hw_tag = 0;
2164 2165
2165 cfqd->hw_tag_samples = 0; 2166 cfqd->hw_tag_samples = 0;
2166 cfqd->rq_in_driver_peak = 0; 2167 cfqd->rq_in_driver_peak = 0;
2167 } 2168 }
2168 2169
2169 static void cfq_completed_request(struct request_queue *q, struct request *rq) 2170 static void cfq_completed_request(struct request_queue *q, struct request *rq)
2170 { 2171 {
2171 struct cfq_queue *cfqq = RQ_CFQQ(rq); 2172 struct cfq_queue *cfqq = RQ_CFQQ(rq);
2172 struct cfq_data *cfqd = cfqq->cfqd; 2173 struct cfq_data *cfqd = cfqq->cfqd;
2173 const int sync = rq_is_sync(rq); 2174 const int sync = rq_is_sync(rq);
2174 unsigned long now; 2175 unsigned long now;
2175 2176
2176 now = jiffies; 2177 now = jiffies;
2177 cfq_log_cfqq(cfqd, cfqq, "complete"); 2178 cfq_log_cfqq(cfqd, cfqq, "complete");
2178 2179
2179 cfq_update_hw_tag(cfqd); 2180 cfq_update_hw_tag(cfqd);
2180 2181
2181 WARN_ON(!cfqd->rq_in_driver[sync]); 2182 WARN_ON(!cfqd->rq_in_driver[sync]);
2182 WARN_ON(!cfqq->dispatched); 2183 WARN_ON(!cfqq->dispatched);
2183 cfqd->rq_in_driver[sync]--; 2184 cfqd->rq_in_driver[sync]--;
2184 cfqq->dispatched--; 2185 cfqq->dispatched--;
2185 2186
2186 if (cfq_cfqq_sync(cfqq)) 2187 if (cfq_cfqq_sync(cfqq))
2187 cfqd->sync_flight--; 2188 cfqd->sync_flight--;
2188 2189
2189 if (sync) { 2190 if (sync) {
2190 RQ_CIC(rq)->last_end_request = now; 2191 RQ_CIC(rq)->last_end_request = now;
2191 cfqd->last_end_sync_rq = now; 2192 cfqd->last_end_sync_rq = now;
2192 } 2193 }
2193 2194
2194 /* 2195 /*
2195 * If this is the active queue, check if it needs to be expired, 2196 * If this is the active queue, check if it needs to be expired,
2196 * or if we want to idle in case it has no pending requests. 2197 * or if we want to idle in case it has no pending requests.
2197 */ 2198 */
2198 if (cfqd->active_queue == cfqq) { 2199 if (cfqd->active_queue == cfqq) {
2199 const bool cfqq_empty = RB_EMPTY_ROOT(&cfqq->sort_list); 2200 const bool cfqq_empty = RB_EMPTY_ROOT(&cfqq->sort_list);
2200 2201
2201 if (cfq_cfqq_slice_new(cfqq)) { 2202 if (cfq_cfqq_slice_new(cfqq)) {
2202 cfq_set_prio_slice(cfqd, cfqq); 2203 cfq_set_prio_slice(cfqd, cfqq);
2203 cfq_clear_cfqq_slice_new(cfqq); 2204 cfq_clear_cfqq_slice_new(cfqq);
2204 } 2205 }
2205 /* 2206 /*
2206 * If there are no requests waiting in this queue, and 2207 * If there are no requests waiting in this queue, and
2207 * there are other queues ready to issue requests, AND 2208 * there are other queues ready to issue requests, AND
2208 * those other queues are issuing requests within our 2209 * those other queues are issuing requests within our
2209 * mean seek distance, give them a chance to run instead 2210 * mean seek distance, give them a chance to run instead
2210 * of idling. 2211 * of idling.
2211 */ 2212 */
2212 if (cfq_slice_used(cfqq) || cfq_class_idle(cfqq)) 2213 if (cfq_slice_used(cfqq) || cfq_class_idle(cfqq))
2213 cfq_slice_expired(cfqd, 1); 2214 cfq_slice_expired(cfqd, 1);
2214 else if (cfqq_empty && !cfq_close_cooperator(cfqd, cfqq, 1) && 2215 else if (cfqq_empty && !cfq_close_cooperator(cfqd, cfqq, 1) &&
2215 sync && !rq_noidle(rq)) 2216 sync && !rq_noidle(rq))
2216 cfq_arm_slice_timer(cfqd); 2217 cfq_arm_slice_timer(cfqd);
2217 } 2218 }
2218 2219
2219 if (!rq_in_driver(cfqd)) 2220 if (!rq_in_driver(cfqd))
2220 cfq_schedule_dispatch(cfqd, 0); 2221 cfq_schedule_dispatch(cfqd, 0);
2221 } 2222 }
2222 2223
2223 /* 2224 /*
2224 * we temporarily boost lower priority queues if they are holding fs exclusive 2225 * we temporarily boost lower priority queues if they are holding fs exclusive
2225 * resources. they are boosted to normal prio (CLASS_BE/4) 2226 * resources. they are boosted to normal prio (CLASS_BE/4)
2226 */ 2227 */
2227 static void cfq_prio_boost(struct cfq_queue *cfqq) 2228 static void cfq_prio_boost(struct cfq_queue *cfqq)
2228 { 2229 {
2229 if (has_fs_excl()) { 2230 if (has_fs_excl()) {
2230 /* 2231 /*
2231 * boost idle prio on transactions that would lock out other 2232 * boost idle prio on transactions that would lock out other
2232 * users of the filesystem 2233 * users of the filesystem
2233 */ 2234 */
2234 if (cfq_class_idle(cfqq)) 2235 if (cfq_class_idle(cfqq))
2235 cfqq->ioprio_class = IOPRIO_CLASS_BE; 2236 cfqq->ioprio_class = IOPRIO_CLASS_BE;
2236 if (cfqq->ioprio > IOPRIO_NORM) 2237 if (cfqq->ioprio > IOPRIO_NORM)
2237 cfqq->ioprio = IOPRIO_NORM; 2238 cfqq->ioprio = IOPRIO_NORM;
2238 } else { 2239 } else {
2239 /* 2240 /*
2240 * check if we need to unboost the queue 2241 * check if we need to unboost the queue
2241 */ 2242 */
2242 if (cfqq->ioprio_class != cfqq->org_ioprio_class) 2243 if (cfqq->ioprio_class != cfqq->org_ioprio_class)
2243 cfqq->ioprio_class = cfqq->org_ioprio_class; 2244 cfqq->ioprio_class = cfqq->org_ioprio_class;
2244 if (cfqq->ioprio != cfqq->org_ioprio) 2245 if (cfqq->ioprio != cfqq->org_ioprio)
2245 cfqq->ioprio = cfqq->org_ioprio; 2246 cfqq->ioprio = cfqq->org_ioprio;
2246 } 2247 }
2247 } 2248 }
2248 2249
2249 static inline int __cfq_may_queue(struct cfq_queue *cfqq) 2250 static inline int __cfq_may_queue(struct cfq_queue *cfqq)
2250 { 2251 {
2251 if (cfq_cfqq_wait_request(cfqq) && !cfq_cfqq_must_alloc_slice(cfqq)) { 2252 if (cfq_cfqq_wait_request(cfqq) && !cfq_cfqq_must_alloc_slice(cfqq)) {
2252 cfq_mark_cfqq_must_alloc_slice(cfqq); 2253 cfq_mark_cfqq_must_alloc_slice(cfqq);
2253 return ELV_MQUEUE_MUST; 2254 return ELV_MQUEUE_MUST;
2254 } 2255 }
2255 2256
2256 return ELV_MQUEUE_MAY; 2257 return ELV_MQUEUE_MAY;
2257 } 2258 }
2258 2259
2259 static int cfq_may_queue(struct request_queue *q, int rw) 2260 static int cfq_may_queue(struct request_queue *q, int rw)
2260 { 2261 {
2261 struct cfq_data *cfqd = q->elevator->elevator_data; 2262 struct cfq_data *cfqd = q->elevator->elevator_data;
2262 struct task_struct *tsk = current; 2263 struct task_struct *tsk = current;
2263 struct cfq_io_context *cic; 2264 struct cfq_io_context *cic;
2264 struct cfq_queue *cfqq; 2265 struct cfq_queue *cfqq;
2265 2266
2266 /* 2267 /*
2267 * don't force setup of a queue from here, as a call to may_queue 2268 * don't force setup of a queue from here, as a call to may_queue
2268 * does not necessarily imply that a request actually will be queued. 2269 * does not necessarily imply that a request actually will be queued.
2269 * so just lookup a possibly existing queue, or return 'may queue' 2270 * so just lookup a possibly existing queue, or return 'may queue'
2270 * if that fails 2271 * if that fails
2271 */ 2272 */
2272 cic = cfq_cic_lookup(cfqd, tsk->io_context); 2273 cic = cfq_cic_lookup(cfqd, tsk->io_context);
2273 if (!cic) 2274 if (!cic)
2274 return ELV_MQUEUE_MAY; 2275 return ELV_MQUEUE_MAY;
2275 2276
2276 cfqq = cic_to_cfqq(cic, rw_is_sync(rw)); 2277 cfqq = cic_to_cfqq(cic, rw_is_sync(rw));
2277 if (cfqq) { 2278 if (cfqq) {
2278 cfq_init_prio_data(cfqq, cic->ioc); 2279 cfq_init_prio_data(cfqq, cic->ioc);
2279 cfq_prio_boost(cfqq); 2280 cfq_prio_boost(cfqq);
2280 2281
2281 return __cfq_may_queue(cfqq); 2282 return __cfq_may_queue(cfqq);
2282 } 2283 }
2283 2284
2284 return ELV_MQUEUE_MAY; 2285 return ELV_MQUEUE_MAY;
2285 } 2286 }
2286 2287
2287 /* 2288 /*
2288 * queue lock held here 2289 * queue lock held here
2289 */ 2290 */
2290 static void cfq_put_request(struct request *rq) 2291 static void cfq_put_request(struct request *rq)
2291 { 2292 {
2292 struct cfq_queue *cfqq = RQ_CFQQ(rq); 2293 struct cfq_queue *cfqq = RQ_CFQQ(rq);
2293 2294
2294 if (cfqq) { 2295 if (cfqq) {
2295 const int rw = rq_data_dir(rq); 2296 const int rw = rq_data_dir(rq);
2296 2297
2297 BUG_ON(!cfqq->allocated[rw]); 2298 BUG_ON(!cfqq->allocated[rw]);
2298 cfqq->allocated[rw]--; 2299 cfqq->allocated[rw]--;
2299 2300
2300 put_io_context(RQ_CIC(rq)->ioc); 2301 put_io_context(RQ_CIC(rq)->ioc);
2301 2302
2302 rq->elevator_private = NULL; 2303 rq->elevator_private = NULL;
2303 rq->elevator_private2 = NULL; 2304 rq->elevator_private2 = NULL;
2304 2305
2305 cfq_put_queue(cfqq); 2306 cfq_put_queue(cfqq);
2306 } 2307 }
2307 } 2308 }
2308 2309
2309 /* 2310 /*
2310 * Allocate cfq data structures associated with this request. 2311 * Allocate cfq data structures associated with this request.
2311 */ 2312 */
2312 static int 2313 static int
2313 cfq_set_request(struct request_queue *q, struct request *rq, gfp_t gfp_mask) 2314 cfq_set_request(struct request_queue *q, struct request *rq, gfp_t gfp_mask)
2314 { 2315 {
2315 struct cfq_data *cfqd = q->elevator->elevator_data; 2316 struct cfq_data *cfqd = q->elevator->elevator_data;
2316 struct cfq_io_context *cic; 2317 struct cfq_io_context *cic;
2317 const int rw = rq_data_dir(rq); 2318 const int rw = rq_data_dir(rq);
2318 const int is_sync = rq_is_sync(rq); 2319 const int is_sync = rq_is_sync(rq);
2319 struct cfq_queue *cfqq; 2320 struct cfq_queue *cfqq;
2320 unsigned long flags; 2321 unsigned long flags;
2321 2322
2322 might_sleep_if(gfp_mask & __GFP_WAIT); 2323 might_sleep_if(gfp_mask & __GFP_WAIT);
2323 2324
2324 cic = cfq_get_io_context(cfqd, gfp_mask); 2325 cic = cfq_get_io_context(cfqd, gfp_mask);
2325 2326
2326 spin_lock_irqsave(q->queue_lock, flags); 2327 spin_lock_irqsave(q->queue_lock, flags);
2327 2328
2328 if (!cic) 2329 if (!cic)
2329 goto queue_fail; 2330 goto queue_fail;
2330 2331
2331 cfqq = cic_to_cfqq(cic, is_sync); 2332 cfqq = cic_to_cfqq(cic, is_sync);
2332 if (!cfqq || cfqq == &cfqd->oom_cfqq) { 2333 if (!cfqq || cfqq == &cfqd->oom_cfqq) {
2333 cfqq = cfq_get_queue(cfqd, is_sync, cic->ioc, gfp_mask); 2334 cfqq = cfq_get_queue(cfqd, is_sync, cic->ioc, gfp_mask);
2334 cic_set_cfqq(cic, cfqq, is_sync); 2335 cic_set_cfqq(cic, cfqq, is_sync);
2335 } 2336 }
2336 2337
2337 cfqq->allocated[rw]++; 2338 cfqq->allocated[rw]++;
2338 atomic_inc(&cfqq->ref); 2339 atomic_inc(&cfqq->ref);
2339 2340
2340 spin_unlock_irqrestore(q->queue_lock, flags); 2341 spin_unlock_irqrestore(q->queue_lock, flags);
2341 2342
2342 rq->elevator_private = cic; 2343 rq->elevator_private = cic;
2343 rq->elevator_private2 = cfqq; 2344 rq->elevator_private2 = cfqq;
2344 return 0; 2345 return 0;
2345 2346
2346 queue_fail: 2347 queue_fail:
2347 if (cic) 2348 if (cic)
2348 put_io_context(cic->ioc); 2349 put_io_context(cic->ioc);
2349 2350
2350 cfq_schedule_dispatch(cfqd, 0); 2351 cfq_schedule_dispatch(cfqd, 0);
2351 spin_unlock_irqrestore(q->queue_lock, flags); 2352 spin_unlock_irqrestore(q->queue_lock, flags);
2352 cfq_log(cfqd, "set_request fail"); 2353 cfq_log(cfqd, "set_request fail");
2353 return 1; 2354 return 1;
2354 } 2355 }
2355 2356
2356 static void cfq_kick_queue(struct work_struct *work) 2357 static void cfq_kick_queue(struct work_struct *work)
2357 { 2358 {
2358 struct cfq_data *cfqd = 2359 struct cfq_data *cfqd =
2359 container_of(work, struct cfq_data, unplug_work.work); 2360 container_of(work, struct cfq_data, unplug_work.work);
2360 struct request_queue *q = cfqd->queue; 2361 struct request_queue *q = cfqd->queue;
2361 2362
2362 spin_lock_irq(q->queue_lock); 2363 spin_lock_irq(q->queue_lock);
2363 __blk_run_queue(cfqd->queue); 2364 __blk_run_queue(cfqd->queue);
2364 spin_unlock_irq(q->queue_lock); 2365 spin_unlock_irq(q->queue_lock);
2365 } 2366 }
2366 2367
2367 /* 2368 /*
2368 * Timer running if the active_queue is currently idling inside its time slice 2369 * Timer running if the active_queue is currently idling inside its time slice
2369 */ 2370 */
2370 static void cfq_idle_slice_timer(unsigned long data) 2371 static void cfq_idle_slice_timer(unsigned long data)
2371 { 2372 {
2372 struct cfq_data *cfqd = (struct cfq_data *) data; 2373 struct cfq_data *cfqd = (struct cfq_data *) data;
2373 struct cfq_queue *cfqq; 2374 struct cfq_queue *cfqq;
2374 unsigned long flags; 2375 unsigned long flags;
2375 int timed_out = 1; 2376 int timed_out = 1;
2376 2377
2377 cfq_log(cfqd, "idle timer fired"); 2378 cfq_log(cfqd, "idle timer fired");
2378 2379
2379 spin_lock_irqsave(cfqd->queue->queue_lock, flags); 2380 spin_lock_irqsave(cfqd->queue->queue_lock, flags);
2380 2381
2381 cfqq = cfqd->active_queue; 2382 cfqq = cfqd->active_queue;
2382 if (cfqq) { 2383 if (cfqq) {
2383 timed_out = 0; 2384 timed_out = 0;
2384 2385
2385 /* 2386 /*
2386 * We saw a request before the queue expired, let it through 2387 * We saw a request before the queue expired, let it through
2387 */ 2388 */
2388 if (cfq_cfqq_must_dispatch(cfqq)) 2389 if (cfq_cfqq_must_dispatch(cfqq))
2389 goto out_kick; 2390 goto out_kick;
2390 2391
2391 /* 2392 /*
2392 * expired 2393 * expired
2393 */ 2394 */
2394 if (cfq_slice_used(cfqq)) 2395 if (cfq_slice_used(cfqq))
2395 goto expire; 2396 goto expire;
2396 2397
2397 /* 2398 /*
2398 * only expire and reinvoke request handler, if there are 2399 * only expire and reinvoke request handler, if there are
2399 * other queues with pending requests 2400 * other queues with pending requests
2400 */ 2401 */
2401 if (!cfqd->busy_queues) 2402 if (!cfqd->busy_queues)
2402 goto out_cont; 2403 goto out_cont;
2403 2404
2404 /* 2405 /*
2405 * not expired and it has a request pending, let it dispatch 2406 * not expired and it has a request pending, let it dispatch
2406 */ 2407 */
2407 if (!RB_EMPTY_ROOT(&cfqq->sort_list)) 2408 if (!RB_EMPTY_ROOT(&cfqq->sort_list))
2408 goto out_kick; 2409 goto out_kick;
2409 } 2410 }
2410 expire: 2411 expire:
2411 cfq_slice_expired(cfqd, timed_out); 2412 cfq_slice_expired(cfqd, timed_out);
2412 out_kick: 2413 out_kick:
2413 cfq_schedule_dispatch(cfqd, 0); 2414 cfq_schedule_dispatch(cfqd, 0);
2414 out_cont: 2415 out_cont:
2415 spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); 2416 spin_unlock_irqrestore(cfqd->queue->queue_lock, flags);
2416 } 2417 }
2417 2418
2418 static void cfq_shutdown_timer_wq(struct cfq_data *cfqd) 2419 static void cfq_shutdown_timer_wq(struct cfq_data *cfqd)
2419 { 2420 {
2420 del_timer_sync(&cfqd->idle_slice_timer); 2421 del_timer_sync(&cfqd->idle_slice_timer);
2421 cancel_delayed_work_sync(&cfqd->unplug_work); 2422 cancel_delayed_work_sync(&cfqd->unplug_work);
2422 } 2423 }
2423 2424
2424 static void cfq_put_async_queues(struct cfq_data *cfqd) 2425 static void cfq_put_async_queues(struct cfq_data *cfqd)
2425 { 2426 {
2426 int i; 2427 int i;
2427 2428
2428 for (i = 0; i < IOPRIO_BE_NR; i++) { 2429 for (i = 0; i < IOPRIO_BE_NR; i++) {
2429 if (cfqd->async_cfqq[0][i]) 2430 if (cfqd->async_cfqq[0][i])
2430 cfq_put_queue(cfqd->async_cfqq[0][i]); 2431 cfq_put_queue(cfqd->async_cfqq[0][i]);
2431 if (cfqd->async_cfqq[1][i]) 2432 if (cfqd->async_cfqq[1][i])
2432 cfq_put_queue(cfqd->async_cfqq[1][i]); 2433 cfq_put_queue(cfqd->async_cfqq[1][i]);
2433 } 2434 }
2434 2435
2435 if (cfqd->async_idle_cfqq) 2436 if (cfqd->async_idle_cfqq)
2436 cfq_put_queue(cfqd->async_idle_cfqq); 2437 cfq_put_queue(cfqd->async_idle_cfqq);
2437 } 2438 }
2438 2439
2439 static void cfq_exit_queue(struct elevator_queue *e) 2440 static void cfq_exit_queue(struct elevator_queue *e)
2440 { 2441 {
2441 struct cfq_data *cfqd = e->elevator_data; 2442 struct cfq_data *cfqd = e->elevator_data;
2442 struct request_queue *q = cfqd->queue; 2443 struct request_queue *q = cfqd->queue;
2443 2444
2444 cfq_shutdown_timer_wq(cfqd); 2445 cfq_shutdown_timer_wq(cfqd);
2445 2446
2446 spin_lock_irq(q->queue_lock); 2447 spin_lock_irq(q->queue_lock);
2447 2448
2448 if (cfqd->active_queue) 2449 if (cfqd->active_queue)
2449 __cfq_slice_expired(cfqd, cfqd->active_queue, 0); 2450 __cfq_slice_expired(cfqd, cfqd->active_queue, 0);
2450 2451
2451 while (!list_empty(&cfqd->cic_list)) { 2452 while (!list_empty(&cfqd->cic_list)) {
2452 struct cfq_io_context *cic = list_entry(cfqd->cic_list.next, 2453 struct cfq_io_context *cic = list_entry(cfqd->cic_list.next,
2453 struct cfq_io_context, 2454 struct cfq_io_context,
2454 queue_list); 2455 queue_list);
2455 2456
2456 __cfq_exit_single_io_context(cfqd, cic); 2457 __cfq_exit_single_io_context(cfqd, cic);
2457 } 2458 }
2458 2459
2459 cfq_put_async_queues(cfqd); 2460 cfq_put_async_queues(cfqd);
2460 2461
2461 spin_unlock_irq(q->queue_lock); 2462 spin_unlock_irq(q->queue_lock);
2462 2463
2463 cfq_shutdown_timer_wq(cfqd); 2464 cfq_shutdown_timer_wq(cfqd);
2464 2465
2465 kfree(cfqd); 2466 kfree(cfqd);
2466 } 2467 }
2467 2468
2468 static void *cfq_init_queue(struct request_queue *q) 2469 static void *cfq_init_queue(struct request_queue *q)
2469 { 2470 {
2470 struct cfq_data *cfqd; 2471 struct cfq_data *cfqd;
2471 int i; 2472 int i;
2472 2473
2473 cfqd = kmalloc_node(sizeof(*cfqd), GFP_KERNEL | __GFP_ZERO, q->node); 2474 cfqd = kmalloc_node(sizeof(*cfqd), GFP_KERNEL | __GFP_ZERO, q->node);
2474 if (!cfqd) 2475 if (!cfqd)
2475 return NULL; 2476 return NULL;
2476 2477
2477 cfqd->service_tree = CFQ_RB_ROOT; 2478 cfqd->service_tree = CFQ_RB_ROOT;
2478 2479
2479 /* 2480 /*
2480 * Not strictly needed (since RB_ROOT just clears the node and we 2481 * Not strictly needed (since RB_ROOT just clears the node and we
2481 * zeroed cfqd on alloc), but better be safe in case someone decides 2482 * zeroed cfqd on alloc), but better be safe in case someone decides
2482 * to add magic to the rb code 2483 * to add magic to the rb code
2483 */ 2484 */
2484 for (i = 0; i < CFQ_PRIO_LISTS; i++) 2485 for (i = 0; i < CFQ_PRIO_LISTS; i++)
2485 cfqd->prio_trees[i] = RB_ROOT; 2486 cfqd->prio_trees[i] = RB_ROOT;
2486 2487
2487 /* 2488 /*
2488 * Our fallback cfqq if cfq_find_alloc_queue() runs into OOM issues. 2489 * Our fallback cfqq if cfq_find_alloc_queue() runs into OOM issues.
2489 * Grab a permanent reference to it, so that the normal code flow 2490 * Grab a permanent reference to it, so that the normal code flow
2490 * will not attempt to free it. 2491 * will not attempt to free it.
2491 */ 2492 */
2492 cfq_init_cfqq(cfqd, &cfqd->oom_cfqq, 1, 0); 2493 cfq_init_cfqq(cfqd, &cfqd->oom_cfqq, 1, 0);
2493 atomic_inc(&cfqd->oom_cfqq.ref); 2494 atomic_inc(&cfqd->oom_cfqq.ref);
2494 2495
2495 INIT_LIST_HEAD(&cfqd->cic_list); 2496 INIT_LIST_HEAD(&cfqd->cic_list);
2496 2497
2497 cfqd->queue = q; 2498 cfqd->queue = q;
2498 2499
2499 init_timer(&cfqd->idle_slice_timer); 2500 init_timer(&cfqd->idle_slice_timer);
2500 cfqd->idle_slice_timer.function = cfq_idle_slice_timer; 2501 cfqd->idle_slice_timer.function = cfq_idle_slice_timer;
2501 cfqd->idle_slice_timer.data = (unsigned long) cfqd; 2502 cfqd->idle_slice_timer.data = (unsigned long) cfqd;
2502 2503
2503 INIT_DELAYED_WORK(&cfqd->unplug_work, cfq_kick_queue); 2504 INIT_DELAYED_WORK(&cfqd->unplug_work, cfq_kick_queue);
2504 2505
2505 cfqd->cfq_quantum = cfq_quantum; 2506 cfqd->cfq_quantum = cfq_quantum;
2506 cfqd->cfq_fifo_expire[0] = cfq_fifo_expire[0]; 2507 cfqd->cfq_fifo_expire[0] = cfq_fifo_expire[0];
2507 cfqd->cfq_fifo_expire[1] = cfq_fifo_expire[1]; 2508 cfqd->cfq_fifo_expire[1] = cfq_fifo_expire[1];
2508 cfqd->cfq_back_max = cfq_back_max; 2509 cfqd->cfq_back_max = cfq_back_max;
2509 cfqd->cfq_back_penalty = cfq_back_penalty; 2510 cfqd->cfq_back_penalty = cfq_back_penalty;
2510 cfqd->cfq_slice[0] = cfq_slice_async; 2511 cfqd->cfq_slice[0] = cfq_slice_async;
2511 cfqd->cfq_slice[1] = cfq_slice_sync; 2512 cfqd->cfq_slice[1] = cfq_slice_sync;
2512 cfqd->cfq_slice_async_rq = cfq_slice_async_rq; 2513 cfqd->cfq_slice_async_rq = cfq_slice_async_rq;
2513 cfqd->cfq_slice_idle = cfq_slice_idle; 2514 cfqd->cfq_slice_idle = cfq_slice_idle;
2514 cfqd->cfq_latency = 1; 2515 cfqd->cfq_latency = 1;
2515 cfqd->hw_tag = 1; 2516 cfqd->hw_tag = 1;
2516 cfqd->last_end_sync_rq = jiffies; 2517 cfqd->last_end_sync_rq = jiffies;
2517 return cfqd; 2518 return cfqd;
2518 } 2519 }
2519 2520
2520 static void cfq_slab_kill(void) 2521 static void cfq_slab_kill(void)
2521 { 2522 {
2522 /* 2523 /*
2523 * Caller already ensured that pending RCU callbacks are completed, 2524 * Caller already ensured that pending RCU callbacks are completed,
2524 * so we should have no busy allocations at this point. 2525 * so we should have no busy allocations at this point.
2525 */ 2526 */
2526 if (cfq_pool) 2527 if (cfq_pool)
2527 kmem_cache_destroy(cfq_pool); 2528 kmem_cache_destroy(cfq_pool);
2528 if (cfq_ioc_pool) 2529 if (cfq_ioc_pool)
2529 kmem_cache_destroy(cfq_ioc_pool); 2530 kmem_cache_destroy(cfq_ioc_pool);
2530 } 2531 }
2531 2532
2532 static int __init cfq_slab_setup(void) 2533 static int __init cfq_slab_setup(void)
2533 { 2534 {
2534 cfq_pool = KMEM_CACHE(cfq_queue, 0); 2535 cfq_pool = KMEM_CACHE(cfq_queue, 0);
2535 if (!cfq_pool) 2536 if (!cfq_pool)
2536 goto fail; 2537 goto fail;
2537 2538
2538 cfq_ioc_pool = KMEM_CACHE(cfq_io_context, 0); 2539 cfq_ioc_pool = KMEM_CACHE(cfq_io_context, 0);
2539 if (!cfq_ioc_pool) 2540 if (!cfq_ioc_pool)
2540 goto fail; 2541 goto fail;
2541 2542
2542 return 0; 2543 return 0;
2543 fail: 2544 fail:
2544 cfq_slab_kill(); 2545 cfq_slab_kill();
2545 return -ENOMEM; 2546 return -ENOMEM;
2546 } 2547 }
2547 2548
2548 /* 2549 /*
2549 * sysfs parts below --> 2550 * sysfs parts below -->
2550 */ 2551 */
2551 static ssize_t 2552 static ssize_t
2552 cfq_var_show(unsigned int var, char *page) 2553 cfq_var_show(unsigned int var, char *page)
2553 { 2554 {
2554 return sprintf(page, "%d\n", var); 2555 return sprintf(page, "%d\n", var);
2555 } 2556 }
2556 2557
2557 static ssize_t 2558 static ssize_t
2558 cfq_var_store(unsigned int *var, const char *page, size_t count) 2559 cfq_var_store(unsigned int *var, const char *page, size_t count)
2559 { 2560 {
2560 char *p = (char *) page; 2561 char *p = (char *) page;
2561 2562
2562 *var = simple_strtoul(p, &p, 10); 2563 *var = simple_strtoul(p, &p, 10);
2563 return count; 2564 return count;
2564 } 2565 }
2565 2566
2566 #define SHOW_FUNCTION(__FUNC, __VAR, __CONV) \ 2567 #define SHOW_FUNCTION(__FUNC, __VAR, __CONV) \
2567 static ssize_t __FUNC(struct elevator_queue *e, char *page) \ 2568 static ssize_t __FUNC(struct elevator_queue *e, char *page) \
2568 { \ 2569 { \
2569 struct cfq_data *cfqd = e->elevator_data; \ 2570 struct cfq_data *cfqd = e->elevator_data; \
2570 unsigned int __data = __VAR; \ 2571 unsigned int __data = __VAR; \
2571 if (__CONV) \ 2572 if (__CONV) \
2572 __data = jiffies_to_msecs(__data); \ 2573 __data = jiffies_to_msecs(__data); \
2573 return cfq_var_show(__data, (page)); \ 2574 return cfq_var_show(__data, (page)); \
2574 } 2575 }
2575 SHOW_FUNCTION(cfq_quantum_show, cfqd->cfq_quantum, 0); 2576 SHOW_FUNCTION(cfq_quantum_show, cfqd->cfq_quantum, 0);
2576 SHOW_FUNCTION(cfq_fifo_expire_sync_show, cfqd->cfq_fifo_expire[1], 1); 2577 SHOW_FUNCTION(cfq_fifo_expire_sync_show, cfqd->cfq_fifo_expire[1], 1);
2577 SHOW_FUNCTION(cfq_fifo_expire_async_show, cfqd->cfq_fifo_expire[0], 1); 2578 SHOW_FUNCTION(cfq_fifo_expire_async_show, cfqd->cfq_fifo_expire[0], 1);
2578 SHOW_FUNCTION(cfq_back_seek_max_show, cfqd->cfq_back_max, 0); 2579 SHOW_FUNCTION(cfq_back_seek_max_show, cfqd->cfq_back_max, 0);
2579 SHOW_FUNCTION(cfq_back_seek_penalty_show, cfqd->cfq_back_penalty, 0); 2580 SHOW_FUNCTION(cfq_back_seek_penalty_show, cfqd->cfq_back_penalty, 0);
2580 SHOW_FUNCTION(cfq_slice_idle_show, cfqd->cfq_slice_idle, 1); 2581 SHOW_FUNCTION(cfq_slice_idle_show, cfqd->cfq_slice_idle, 1);
2581 SHOW_FUNCTION(cfq_slice_sync_show, cfqd->cfq_slice[1], 1); 2582 SHOW_FUNCTION(cfq_slice_sync_show, cfqd->cfq_slice[1], 1);
2582 SHOW_FUNCTION(cfq_slice_async_show, cfqd->cfq_slice[0], 1); 2583 SHOW_FUNCTION(cfq_slice_async_show, cfqd->cfq_slice[0], 1);
2583 SHOW_FUNCTION(cfq_slice_async_rq_show, cfqd->cfq_slice_async_rq, 0); 2584 SHOW_FUNCTION(cfq_slice_async_rq_show, cfqd->cfq_slice_async_rq, 0);
2584 SHOW_FUNCTION(cfq_low_latency_show, cfqd->cfq_latency, 0); 2585 SHOW_FUNCTION(cfq_low_latency_show, cfqd->cfq_latency, 0);
2585 #undef SHOW_FUNCTION 2586 #undef SHOW_FUNCTION
2586 2587
2587 #define STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, __CONV) \ 2588 #define STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, __CONV) \
2588 static ssize_t __FUNC(struct elevator_queue *e, const char *page, size_t count) \ 2589 static ssize_t __FUNC(struct elevator_queue *e, const char *page, size_t count) \
2589 { \ 2590 { \
2590 struct cfq_data *cfqd = e->elevator_data; \ 2591 struct cfq_data *cfqd = e->elevator_data; \
2591 unsigned int __data; \ 2592 unsigned int __data; \
2592 int ret = cfq_var_store(&__data, (page), count); \ 2593 int ret = cfq_var_store(&__data, (page), count); \
2593 if (__data < (MIN)) \ 2594 if (__data < (MIN)) \
2594 __data = (MIN); \ 2595 __data = (MIN); \
2595 else if (__data > (MAX)) \ 2596 else if (__data > (MAX)) \
2596 __data = (MAX); \ 2597 __data = (MAX); \
2597 if (__CONV) \ 2598 if (__CONV) \
2598 *(__PTR) = msecs_to_jiffies(__data); \ 2599 *(__PTR) = msecs_to_jiffies(__data); \
2599 else \ 2600 else \
2600 *(__PTR) = __data; \ 2601 *(__PTR) = __data; \
2601 return ret; \ 2602 return ret; \
2602 } 2603 }
2603 STORE_FUNCTION(cfq_quantum_store, &cfqd->cfq_quantum, 1, UINT_MAX, 0); 2604 STORE_FUNCTION(cfq_quantum_store, &cfqd->cfq_quantum, 1, UINT_MAX, 0);
2604 STORE_FUNCTION(cfq_fifo_expire_sync_store, &cfqd->cfq_fifo_expire[1], 1, 2605 STORE_FUNCTION(cfq_fifo_expire_sync_store, &cfqd->cfq_fifo_expire[1], 1,
2605 UINT_MAX, 1); 2606 UINT_MAX, 1);
2606 STORE_FUNCTION(cfq_fifo_expire_async_store, &cfqd->cfq_fifo_expire[0], 1, 2607 STORE_FUNCTION(cfq_fifo_expire_async_store, &cfqd->cfq_fifo_expire[0], 1,
2607 UINT_MAX, 1); 2608 UINT_MAX, 1);
2608 STORE_FUNCTION(cfq_back_seek_max_store, &cfqd->cfq_back_max, 0, UINT_MAX, 0); 2609 STORE_FUNCTION(cfq_back_seek_max_store, &cfqd->cfq_back_max, 0, UINT_MAX, 0);
2609 STORE_FUNCTION(cfq_back_seek_penalty_store, &cfqd->cfq_back_penalty, 1, 2610 STORE_FUNCTION(cfq_back_seek_penalty_store, &cfqd->cfq_back_penalty, 1,
2610 UINT_MAX, 0); 2611 UINT_MAX, 0);
2611 STORE_FUNCTION(cfq_slice_idle_store, &cfqd->cfq_slice_idle, 0, UINT_MAX, 1); 2612 STORE_FUNCTION(cfq_slice_idle_store, &cfqd->cfq_slice_idle, 0, UINT_MAX, 1);
2612 STORE_FUNCTION(cfq_slice_sync_store, &cfqd->cfq_slice[1], 1, UINT_MAX, 1); 2613 STORE_FUNCTION(cfq_slice_sync_store, &cfqd->cfq_slice[1], 1, UINT_MAX, 1);
2613 STORE_FUNCTION(cfq_slice_async_store, &cfqd->cfq_slice[0], 1, UINT_MAX, 1); 2614 STORE_FUNCTION(cfq_slice_async_store, &cfqd->cfq_slice[0], 1, UINT_MAX, 1);
2614 STORE_FUNCTION(cfq_slice_async_rq_store, &cfqd->cfq_slice_async_rq, 1, 2615 STORE_FUNCTION(cfq_slice_async_rq_store, &cfqd->cfq_slice_async_rq, 1,
2615 UINT_MAX, 0); 2616 UINT_MAX, 0);
2616 STORE_FUNCTION(cfq_low_latency_store, &cfqd->cfq_latency, 0, 1, 0); 2617 STORE_FUNCTION(cfq_low_latency_store, &cfqd->cfq_latency, 0, 1, 0);
2617 #undef STORE_FUNCTION 2618 #undef STORE_FUNCTION
2618 2619
2619 #define CFQ_ATTR(name) \ 2620 #define CFQ_ATTR(name) \
2620 __ATTR(name, S_IRUGO|S_IWUSR, cfq_##name##_show, cfq_##name##_store) 2621 __ATTR(name, S_IRUGO|S_IWUSR, cfq_##name##_show, cfq_##name##_store)
2621 2622
2622 static struct elv_fs_entry cfq_attrs[] = { 2623 static struct elv_fs_entry cfq_attrs[] = {
2623 CFQ_ATTR(quantum), 2624 CFQ_ATTR(quantum),
2624 CFQ_ATTR(fifo_expire_sync), 2625 CFQ_ATTR(fifo_expire_sync),
2625 CFQ_ATTR(fifo_expire_async), 2626 CFQ_ATTR(fifo_expire_async),
2626 CFQ_ATTR(back_seek_max), 2627 CFQ_ATTR(back_seek_max),
2627 CFQ_ATTR(back_seek_penalty), 2628 CFQ_ATTR(back_seek_penalty),
2628 CFQ_ATTR(slice_sync), 2629 CFQ_ATTR(slice_sync),
2629 CFQ_ATTR(slice_async), 2630 CFQ_ATTR(slice_async),
2630 CFQ_ATTR(slice_async_rq), 2631 CFQ_ATTR(slice_async_rq),
2631 CFQ_ATTR(slice_idle), 2632 CFQ_ATTR(slice_idle),
2632 CFQ_ATTR(low_latency), 2633 CFQ_ATTR(low_latency),
2633 __ATTR_NULL 2634 __ATTR_NULL
2634 }; 2635 };
2635 2636
2636 static struct elevator_type iosched_cfq = { 2637 static struct elevator_type iosched_cfq = {
2637 .ops = { 2638 .ops = {
2638 .elevator_merge_fn = cfq_merge, 2639 .elevator_merge_fn = cfq_merge,
2639 .elevator_merged_fn = cfq_merged_request, 2640 .elevator_merged_fn = cfq_merged_request,
2640 .elevator_merge_req_fn = cfq_merged_requests, 2641 .elevator_merge_req_fn = cfq_merged_requests,
2641 .elevator_allow_merge_fn = cfq_allow_merge, 2642 .elevator_allow_merge_fn = cfq_allow_merge,
2642 .elevator_dispatch_fn = cfq_dispatch_requests, 2643 .elevator_dispatch_fn = cfq_dispatch_requests,
2643 .elevator_add_req_fn = cfq_insert_request, 2644 .elevator_add_req_fn = cfq_insert_request,
2644 .elevator_activate_req_fn = cfq_activate_request, 2645 .elevator_activate_req_fn = cfq_activate_request,
2645 .elevator_deactivate_req_fn = cfq_deactivate_request, 2646 .elevator_deactivate_req_fn = cfq_deactivate_request,
2646 .elevator_queue_empty_fn = cfq_queue_empty, 2647 .elevator_queue_empty_fn = cfq_queue_empty,
2647 .elevator_completed_req_fn = cfq_completed_request, 2648 .elevator_completed_req_fn = cfq_completed_request,
2648 .elevator_former_req_fn = elv_rb_former_request, 2649 .elevator_former_req_fn = elv_rb_former_request,
2649 .elevator_latter_req_fn = elv_rb_latter_request, 2650 .elevator_latter_req_fn = elv_rb_latter_request,
2650 .elevator_set_req_fn = cfq_set_request, 2651 .elevator_set_req_fn = cfq_set_request,
2651 .elevator_put_req_fn = cfq_put_request, 2652 .elevator_put_req_fn = cfq_put_request,
2652 .elevator_may_queue_fn = cfq_may_queue, 2653 .elevator_may_queue_fn = cfq_may_queue,
2653 .elevator_init_fn = cfq_init_queue, 2654 .elevator_init_fn = cfq_init_queue,
2654 .elevator_exit_fn = cfq_exit_queue, 2655 .elevator_exit_fn = cfq_exit_queue,
2655 .trim = cfq_free_io_context, 2656 .trim = cfq_free_io_context,
2656 }, 2657 },
2657 .elevator_attrs = cfq_attrs, 2658 .elevator_attrs = cfq_attrs,
2658 .elevator_name = "cfq", 2659 .elevator_name = "cfq",
2659 .elevator_owner = THIS_MODULE, 2660 .elevator_owner = THIS_MODULE,
2660 }; 2661 };
2661 2662
2662 static int __init cfq_init(void) 2663 static int __init cfq_init(void)
2663 { 2664 {
2664 /* 2665 /*
2665 * could be 0 on HZ < 1000 setups 2666 * could be 0 on HZ < 1000 setups
2666 */ 2667 */
2667 if (!cfq_slice_async) 2668 if (!cfq_slice_async)
2668 cfq_slice_async = 1; 2669 cfq_slice_async = 1;
2669 if (!cfq_slice_idle) 2670 if (!cfq_slice_idle)
2670 cfq_slice_idle = 1; 2671 cfq_slice_idle = 1;
2671 2672
2672 if (cfq_slab_setup()) 2673 if (cfq_slab_setup())
2673 return -ENOMEM; 2674 return -ENOMEM;
2674 2675
2675 elv_register(&iosched_cfq); 2676 elv_register(&iosched_cfq);
2676 2677
2677 return 0; 2678 return 0;
2678 } 2679 }
2679 2680
2680 static void __exit cfq_exit(void) 2681 static void __exit cfq_exit(void)
2681 { 2682 {
2682 DECLARE_COMPLETION_ONSTACK(all_gone); 2683 DECLARE_COMPLETION_ONSTACK(all_gone);
2683 elv_unregister(&iosched_cfq); 2684 elv_unregister(&iosched_cfq);
2684 ioc_gone = &all_gone; 2685 ioc_gone = &all_gone;
2685 /* ioc_gone's update must be visible before reading ioc_count */ 2686 /* ioc_gone's update must be visible before reading ioc_count */
2686 smp_wmb(); 2687 smp_wmb();
2687 2688
2688 /* 2689 /*
2689 * this also protects us from entering cfq_slab_kill() with 2690 * this also protects us from entering cfq_slab_kill() with
2690 * pending RCU callbacks 2691 * pending RCU callbacks
2691 */ 2692 */
2692 if (elv_ioc_count_read(cfq_ioc_count)) 2693 if (elv_ioc_count_read(cfq_ioc_count))
2693 wait_for_completion(&all_gone); 2694 wait_for_completion(&all_gone);
2694 cfq_slab_kill(); 2695 cfq_slab_kill();
2695 } 2696 }
2696 2697
2697 module_init(cfq_init); 2698 module_init(cfq_init);
2698 module_exit(cfq_exit); 2699 module_exit(cfq_exit);
2699 2700
2700 MODULE_AUTHOR("Jens Axboe"); 2701 MODULE_AUTHOR("Jens Axboe");
2701 MODULE_LICENSE("GPL"); 2702 MODULE_LICENSE("GPL");
2702 MODULE_DESCRIPTION("Completely Fair Queueing IO scheduler"); 2703 MODULE_DESCRIPTION("Completely Fair Queueing IO scheduler");
2703 2704