Commit c4e7893ebc3a5c507b53f59b9de448db20849944
Committed by
Jens Axboe
1 parent
80bdf0c78f
Exists in
master
and in
4 other branches
cfq-iosched: blktrace print per slice sector stats
o Divyesh had gotten rid of this code in the past. I want to re-introduce it back as it helps me a lot during debugging. Reviewed-by: Jeff Moyer <jmoyer@redhat.com> Reviewed-by: Divyesh Shah <dpshah@google.com> Signed-off-by: Vivek Goyal <vgoyal@redhat.com> Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
Showing 1 changed file with 7 additions and 2 deletions Inline Diff
block/cfq-iosched.c
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/slab.h> | 10 | #include <linux/slab.h> |
11 | #include <linux/blkdev.h> | 11 | #include <linux/blkdev.h> |
12 | #include <linux/elevator.h> | 12 | #include <linux/elevator.h> |
13 | #include <linux/jiffies.h> | 13 | #include <linux/jiffies.h> |
14 | #include <linux/rbtree.h> | 14 | #include <linux/rbtree.h> |
15 | #include <linux/ioprio.h> | 15 | #include <linux/ioprio.h> |
16 | #include <linux/blktrace_api.h> | 16 | #include <linux/blktrace_api.h> |
17 | #include "cfq.h" | 17 | #include "cfq.h" |
18 | 18 | ||
19 | /* | 19 | /* |
20 | * tunables | 20 | * tunables |
21 | */ | 21 | */ |
22 | /* max queue in one round of service */ | 22 | /* max queue in one round of service */ |
23 | static const int cfq_quantum = 8; | 23 | static const int cfq_quantum = 8; |
24 | static const int cfq_fifo_expire[2] = { HZ / 4, HZ / 8 }; | 24 | static const int cfq_fifo_expire[2] = { HZ / 4, HZ / 8 }; |
25 | /* maximum backwards seek, in KiB */ | 25 | /* maximum backwards seek, in KiB */ |
26 | static const int cfq_back_max = 16 * 1024; | 26 | static const int cfq_back_max = 16 * 1024; |
27 | /* penalty of a backwards seek */ | 27 | /* penalty of a backwards seek */ |
28 | static const int cfq_back_penalty = 2; | 28 | static const int cfq_back_penalty = 2; |
29 | static const int cfq_slice_sync = HZ / 10; | 29 | static const int cfq_slice_sync = HZ / 10; |
30 | static int cfq_slice_async = HZ / 25; | 30 | static int cfq_slice_async = HZ / 25; |
31 | static const int cfq_slice_async_rq = 2; | 31 | static const int cfq_slice_async_rq = 2; |
32 | static int cfq_slice_idle = HZ / 125; | 32 | static int cfq_slice_idle = HZ / 125; |
33 | static int cfq_group_idle = HZ / 125; | 33 | static int cfq_group_idle = HZ / 125; |
34 | static const int cfq_target_latency = HZ * 3/10; /* 300 ms */ | 34 | static const int cfq_target_latency = HZ * 3/10; /* 300 ms */ |
35 | static const int cfq_hist_divisor = 4; | 35 | static const int cfq_hist_divisor = 4; |
36 | 36 | ||
37 | /* | 37 | /* |
38 | * offset from end of service tree | 38 | * offset from end of service tree |
39 | */ | 39 | */ |
40 | #define CFQ_IDLE_DELAY (HZ / 5) | 40 | #define CFQ_IDLE_DELAY (HZ / 5) |
41 | 41 | ||
42 | /* | 42 | /* |
43 | * below this threshold, we consider thinktime immediate | 43 | * below this threshold, we consider thinktime immediate |
44 | */ | 44 | */ |
45 | #define CFQ_MIN_TT (2) | 45 | #define CFQ_MIN_TT (2) |
46 | 46 | ||
47 | #define CFQ_SLICE_SCALE (5) | 47 | #define CFQ_SLICE_SCALE (5) |
48 | #define CFQ_HW_QUEUE_MIN (5) | 48 | #define CFQ_HW_QUEUE_MIN (5) |
49 | #define CFQ_SERVICE_SHIFT 12 | 49 | #define CFQ_SERVICE_SHIFT 12 |
50 | 50 | ||
51 | #define CFQQ_SEEK_THR (sector_t)(8 * 100) | 51 | #define CFQQ_SEEK_THR (sector_t)(8 * 100) |
52 | #define CFQQ_CLOSE_THR (sector_t)(8 * 1024) | 52 | #define CFQQ_CLOSE_THR (sector_t)(8 * 1024) |
53 | #define CFQQ_SECT_THR_NONROT (sector_t)(2 * 32) | 53 | #define CFQQ_SECT_THR_NONROT (sector_t)(2 * 32) |
54 | #define CFQQ_SEEKY(cfqq) (hweight32(cfqq->seek_history) > 32/8) | 54 | #define CFQQ_SEEKY(cfqq) (hweight32(cfqq->seek_history) > 32/8) |
55 | 55 | ||
56 | #define RQ_CIC(rq) \ | 56 | #define RQ_CIC(rq) \ |
57 | ((struct cfq_io_context *) (rq)->elevator_private) | 57 | ((struct cfq_io_context *) (rq)->elevator_private) |
58 | #define RQ_CFQQ(rq) (struct cfq_queue *) ((rq)->elevator_private2) | 58 | #define RQ_CFQQ(rq) (struct cfq_queue *) ((rq)->elevator_private2) |
59 | #define RQ_CFQG(rq) (struct cfq_group *) ((rq)->elevator_private3) | 59 | #define RQ_CFQG(rq) (struct cfq_group *) ((rq)->elevator_private3) |
60 | 60 | ||
61 | static struct kmem_cache *cfq_pool; | 61 | static struct kmem_cache *cfq_pool; |
62 | static struct kmem_cache *cfq_ioc_pool; | 62 | static struct kmem_cache *cfq_ioc_pool; |
63 | 63 | ||
64 | static DEFINE_PER_CPU(unsigned long, cfq_ioc_count); | 64 | static DEFINE_PER_CPU(unsigned long, cfq_ioc_count); |
65 | static struct completion *ioc_gone; | 65 | static struct completion *ioc_gone; |
66 | static DEFINE_SPINLOCK(ioc_gone_lock); | 66 | static DEFINE_SPINLOCK(ioc_gone_lock); |
67 | 67 | ||
68 | static DEFINE_SPINLOCK(cic_index_lock); | 68 | static DEFINE_SPINLOCK(cic_index_lock); |
69 | static DEFINE_IDA(cic_index_ida); | 69 | static DEFINE_IDA(cic_index_ida); |
70 | 70 | ||
71 | #define CFQ_PRIO_LISTS IOPRIO_BE_NR | 71 | #define CFQ_PRIO_LISTS IOPRIO_BE_NR |
72 | #define cfq_class_idle(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_IDLE) | 72 | #define cfq_class_idle(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_IDLE) |
73 | #define cfq_class_rt(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_RT) | 73 | #define cfq_class_rt(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_RT) |
74 | 74 | ||
75 | #define sample_valid(samples) ((samples) > 80) | 75 | #define sample_valid(samples) ((samples) > 80) |
76 | #define rb_entry_cfqg(node) rb_entry((node), struct cfq_group, rb_node) | 76 | #define rb_entry_cfqg(node) rb_entry((node), struct cfq_group, rb_node) |
77 | 77 | ||
78 | /* | 78 | /* |
79 | * Most of our rbtree usage is for sorting with min extraction, so | 79 | * Most of our rbtree usage is for sorting with min extraction, so |
80 | * if we cache the leftmost node we don't have to walk down the tree | 80 | * if we cache the leftmost node we don't have to walk down the tree |
81 | * to find it. Idea borrowed from Ingo Molnars CFS scheduler. We should | 81 | * to find it. Idea borrowed from Ingo Molnars CFS scheduler. We should |
82 | * move this into the elevator for the rq sorting as well. | 82 | * move this into the elevator for the rq sorting as well. |
83 | */ | 83 | */ |
84 | struct cfq_rb_root { | 84 | struct cfq_rb_root { |
85 | struct rb_root rb; | 85 | struct rb_root rb; |
86 | struct rb_node *left; | 86 | struct rb_node *left; |
87 | unsigned count; | 87 | unsigned count; |
88 | unsigned total_weight; | 88 | unsigned total_weight; |
89 | u64 min_vdisktime; | 89 | u64 min_vdisktime; |
90 | struct rb_node *active; | 90 | struct rb_node *active; |
91 | }; | 91 | }; |
92 | #define CFQ_RB_ROOT (struct cfq_rb_root) { .rb = RB_ROOT, .left = NULL, \ | 92 | #define CFQ_RB_ROOT (struct cfq_rb_root) { .rb = RB_ROOT, .left = NULL, \ |
93 | .count = 0, .min_vdisktime = 0, } | 93 | .count = 0, .min_vdisktime = 0, } |
94 | 94 | ||
95 | /* | 95 | /* |
96 | * Per process-grouping structure | 96 | * Per process-grouping structure |
97 | */ | 97 | */ |
98 | struct cfq_queue { | 98 | struct cfq_queue { |
99 | /* reference count */ | 99 | /* reference count */ |
100 | atomic_t ref; | 100 | atomic_t ref; |
101 | /* various state flags, see below */ | 101 | /* various state flags, see below */ |
102 | unsigned int flags; | 102 | unsigned int flags; |
103 | /* parent cfq_data */ | 103 | /* parent cfq_data */ |
104 | struct cfq_data *cfqd; | 104 | struct cfq_data *cfqd; |
105 | /* service_tree member */ | 105 | /* service_tree member */ |
106 | struct rb_node rb_node; | 106 | struct rb_node rb_node; |
107 | /* service_tree key */ | 107 | /* service_tree key */ |
108 | unsigned long rb_key; | 108 | unsigned long rb_key; |
109 | /* prio tree member */ | 109 | /* prio tree member */ |
110 | struct rb_node p_node; | 110 | struct rb_node p_node; |
111 | /* prio tree root we belong to, if any */ | 111 | /* prio tree root we belong to, if any */ |
112 | struct rb_root *p_root; | 112 | struct rb_root *p_root; |
113 | /* sorted list of pending requests */ | 113 | /* sorted list of pending requests */ |
114 | struct rb_root sort_list; | 114 | struct rb_root sort_list; |
115 | /* if fifo isn't expired, next request to serve */ | 115 | /* if fifo isn't expired, next request to serve */ |
116 | struct request *next_rq; | 116 | struct request *next_rq; |
117 | /* requests queued in sort_list */ | 117 | /* requests queued in sort_list */ |
118 | int queued[2]; | 118 | int queued[2]; |
119 | /* currently allocated requests */ | 119 | /* currently allocated requests */ |
120 | int allocated[2]; | 120 | int allocated[2]; |
121 | /* fifo list of requests in sort_list */ | 121 | /* fifo list of requests in sort_list */ |
122 | struct list_head fifo; | 122 | struct list_head fifo; |
123 | 123 | ||
124 | /* time when queue got scheduled in to dispatch first request. */ | 124 | /* time when queue got scheduled in to dispatch first request. */ |
125 | unsigned long dispatch_start; | 125 | unsigned long dispatch_start; |
126 | unsigned int allocated_slice; | 126 | unsigned int allocated_slice; |
127 | unsigned int slice_dispatch; | 127 | unsigned int slice_dispatch; |
128 | /* time when first request from queue completed and slice started. */ | 128 | /* time when first request from queue completed and slice started. */ |
129 | unsigned long slice_start; | 129 | unsigned long slice_start; |
130 | unsigned long slice_end; | 130 | unsigned long slice_end; |
131 | long slice_resid; | 131 | long slice_resid; |
132 | 132 | ||
133 | /* pending metadata requests */ | 133 | /* pending metadata requests */ |
134 | int meta_pending; | 134 | int meta_pending; |
135 | /* number of requests that are on the dispatch list or inside driver */ | 135 | /* number of requests that are on the dispatch list or inside driver */ |
136 | int dispatched; | 136 | int dispatched; |
137 | 137 | ||
138 | /* io prio of this group */ | 138 | /* io prio of this group */ |
139 | unsigned short ioprio, org_ioprio; | 139 | unsigned short ioprio, org_ioprio; |
140 | unsigned short ioprio_class, org_ioprio_class; | 140 | unsigned short ioprio_class, org_ioprio_class; |
141 | 141 | ||
142 | pid_t pid; | 142 | pid_t pid; |
143 | 143 | ||
144 | u32 seek_history; | 144 | u32 seek_history; |
145 | sector_t last_request_pos; | 145 | sector_t last_request_pos; |
146 | 146 | ||
147 | struct cfq_rb_root *service_tree; | 147 | struct cfq_rb_root *service_tree; |
148 | struct cfq_queue *new_cfqq; | 148 | struct cfq_queue *new_cfqq; |
149 | struct cfq_group *cfqg; | 149 | struct cfq_group *cfqg; |
150 | struct cfq_group *orig_cfqg; | 150 | struct cfq_group *orig_cfqg; |
151 | /* Number of sectors dispatched from queue in single dispatch round */ | ||
152 | unsigned long nr_sectors; | ||
151 | }; | 153 | }; |
152 | 154 | ||
153 | /* | 155 | /* |
154 | * First index in the service_trees. | 156 | * First index in the service_trees. |
155 | * IDLE is handled separately, so it has negative index | 157 | * IDLE is handled separately, so it has negative index |
156 | */ | 158 | */ |
157 | enum wl_prio_t { | 159 | enum wl_prio_t { |
158 | BE_WORKLOAD = 0, | 160 | BE_WORKLOAD = 0, |
159 | RT_WORKLOAD = 1, | 161 | RT_WORKLOAD = 1, |
160 | IDLE_WORKLOAD = 2, | 162 | IDLE_WORKLOAD = 2, |
161 | }; | 163 | }; |
162 | 164 | ||
163 | /* | 165 | /* |
164 | * Second index in the service_trees. | 166 | * Second index in the service_trees. |
165 | */ | 167 | */ |
166 | enum wl_type_t { | 168 | enum wl_type_t { |
167 | ASYNC_WORKLOAD = 0, | 169 | ASYNC_WORKLOAD = 0, |
168 | SYNC_NOIDLE_WORKLOAD = 1, | 170 | SYNC_NOIDLE_WORKLOAD = 1, |
169 | SYNC_WORKLOAD = 2 | 171 | SYNC_WORKLOAD = 2 |
170 | }; | 172 | }; |
171 | 173 | ||
172 | /* This is per cgroup per device grouping structure */ | 174 | /* This is per cgroup per device grouping structure */ |
173 | struct cfq_group { | 175 | struct cfq_group { |
174 | /* group service_tree member */ | 176 | /* group service_tree member */ |
175 | struct rb_node rb_node; | 177 | struct rb_node rb_node; |
176 | 178 | ||
177 | /* group service_tree key */ | 179 | /* group service_tree key */ |
178 | u64 vdisktime; | 180 | u64 vdisktime; |
179 | unsigned int weight; | 181 | unsigned int weight; |
180 | bool on_st; | 182 | bool on_st; |
181 | 183 | ||
182 | /* number of cfqq currently on this group */ | 184 | /* number of cfqq currently on this group */ |
183 | int nr_cfqq; | 185 | int nr_cfqq; |
184 | 186 | ||
185 | /* Per group busy queus average. Useful for workload slice calc. */ | 187 | /* Per group busy queus average. Useful for workload slice calc. */ |
186 | unsigned int busy_queues_avg[2]; | 188 | unsigned int busy_queues_avg[2]; |
187 | /* | 189 | /* |
188 | * rr lists of queues with requests, onle rr for each priority class. | 190 | * rr lists of queues with requests, onle rr for each priority class. |
189 | * Counts are embedded in the cfq_rb_root | 191 | * Counts are embedded in the cfq_rb_root |
190 | */ | 192 | */ |
191 | struct cfq_rb_root service_trees[2][3]; | 193 | struct cfq_rb_root service_trees[2][3]; |
192 | struct cfq_rb_root service_tree_idle; | 194 | struct cfq_rb_root service_tree_idle; |
193 | 195 | ||
194 | unsigned long saved_workload_slice; | 196 | unsigned long saved_workload_slice; |
195 | enum wl_type_t saved_workload; | 197 | enum wl_type_t saved_workload; |
196 | enum wl_prio_t saved_serving_prio; | 198 | enum wl_prio_t saved_serving_prio; |
197 | struct blkio_group blkg; | 199 | struct blkio_group blkg; |
198 | #ifdef CONFIG_CFQ_GROUP_IOSCHED | 200 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
199 | struct hlist_node cfqd_node; | 201 | struct hlist_node cfqd_node; |
200 | atomic_t ref; | 202 | atomic_t ref; |
201 | #endif | 203 | #endif |
202 | /* number of requests that are on the dispatch list or inside driver */ | 204 | /* number of requests that are on the dispatch list or inside driver */ |
203 | int dispatched; | 205 | int dispatched; |
204 | }; | 206 | }; |
205 | 207 | ||
206 | /* | 208 | /* |
207 | * Per block device queue structure | 209 | * Per block device queue structure |
208 | */ | 210 | */ |
209 | struct cfq_data { | 211 | struct cfq_data { |
210 | struct request_queue *queue; | 212 | struct request_queue *queue; |
211 | /* Root service tree for cfq_groups */ | 213 | /* Root service tree for cfq_groups */ |
212 | struct cfq_rb_root grp_service_tree; | 214 | struct cfq_rb_root grp_service_tree; |
213 | struct cfq_group root_group; | 215 | struct cfq_group root_group; |
214 | 216 | ||
215 | /* | 217 | /* |
216 | * The priority currently being served | 218 | * The priority currently being served |
217 | */ | 219 | */ |
218 | enum wl_prio_t serving_prio; | 220 | enum wl_prio_t serving_prio; |
219 | enum wl_type_t serving_type; | 221 | enum wl_type_t serving_type; |
220 | unsigned long workload_expires; | 222 | unsigned long workload_expires; |
221 | struct cfq_group *serving_group; | 223 | struct cfq_group *serving_group; |
222 | bool noidle_tree_requires_idle; | 224 | bool noidle_tree_requires_idle; |
223 | 225 | ||
224 | /* | 226 | /* |
225 | * Each priority tree is sorted by next_request position. These | 227 | * Each priority tree is sorted by next_request position. These |
226 | * trees are used when determining if two or more queues are | 228 | * trees are used when determining if two or more queues are |
227 | * interleaving requests (see cfq_close_cooperator). | 229 | * interleaving requests (see cfq_close_cooperator). |
228 | */ | 230 | */ |
229 | struct rb_root prio_trees[CFQ_PRIO_LISTS]; | 231 | struct rb_root prio_trees[CFQ_PRIO_LISTS]; |
230 | 232 | ||
231 | unsigned int busy_queues; | 233 | unsigned int busy_queues; |
232 | 234 | ||
233 | int rq_in_driver; | 235 | int rq_in_driver; |
234 | int rq_in_flight[2]; | 236 | int rq_in_flight[2]; |
235 | 237 | ||
236 | /* | 238 | /* |
237 | * queue-depth detection | 239 | * queue-depth detection |
238 | */ | 240 | */ |
239 | int rq_queued; | 241 | int rq_queued; |
240 | int hw_tag; | 242 | int hw_tag; |
241 | /* | 243 | /* |
242 | * hw_tag can be | 244 | * hw_tag can be |
243 | * -1 => indeterminate, (cfq will behave as if NCQ is present, to allow better detection) | 245 | * -1 => indeterminate, (cfq will behave as if NCQ is present, to allow better detection) |
244 | * 1 => NCQ is present (hw_tag_est_depth is the estimated max depth) | 246 | * 1 => NCQ is present (hw_tag_est_depth is the estimated max depth) |
245 | * 0 => no NCQ | 247 | * 0 => no NCQ |
246 | */ | 248 | */ |
247 | int hw_tag_est_depth; | 249 | int hw_tag_est_depth; |
248 | unsigned int hw_tag_samples; | 250 | unsigned int hw_tag_samples; |
249 | 251 | ||
250 | /* | 252 | /* |
251 | * idle window management | 253 | * idle window management |
252 | */ | 254 | */ |
253 | struct timer_list idle_slice_timer; | 255 | struct timer_list idle_slice_timer; |
254 | struct work_struct unplug_work; | 256 | struct work_struct unplug_work; |
255 | 257 | ||
256 | struct cfq_queue *active_queue; | 258 | struct cfq_queue *active_queue; |
257 | struct cfq_io_context *active_cic; | 259 | struct cfq_io_context *active_cic; |
258 | 260 | ||
259 | /* | 261 | /* |
260 | * async queue for each priority case | 262 | * async queue for each priority case |
261 | */ | 263 | */ |
262 | struct cfq_queue *async_cfqq[2][IOPRIO_BE_NR]; | 264 | struct cfq_queue *async_cfqq[2][IOPRIO_BE_NR]; |
263 | struct cfq_queue *async_idle_cfqq; | 265 | struct cfq_queue *async_idle_cfqq; |
264 | 266 | ||
265 | sector_t last_position; | 267 | sector_t last_position; |
266 | 268 | ||
267 | /* | 269 | /* |
268 | * tunables, see top of file | 270 | * tunables, see top of file |
269 | */ | 271 | */ |
270 | unsigned int cfq_quantum; | 272 | unsigned int cfq_quantum; |
271 | unsigned int cfq_fifo_expire[2]; | 273 | unsigned int cfq_fifo_expire[2]; |
272 | unsigned int cfq_back_penalty; | 274 | unsigned int cfq_back_penalty; |
273 | unsigned int cfq_back_max; | 275 | unsigned int cfq_back_max; |
274 | unsigned int cfq_slice[2]; | 276 | unsigned int cfq_slice[2]; |
275 | unsigned int cfq_slice_async_rq; | 277 | unsigned int cfq_slice_async_rq; |
276 | unsigned int cfq_slice_idle; | 278 | unsigned int cfq_slice_idle; |
277 | unsigned int cfq_group_idle; | 279 | unsigned int cfq_group_idle; |
278 | unsigned int cfq_latency; | 280 | unsigned int cfq_latency; |
279 | unsigned int cfq_group_isolation; | 281 | unsigned int cfq_group_isolation; |
280 | 282 | ||
281 | unsigned int cic_index; | 283 | unsigned int cic_index; |
282 | struct list_head cic_list; | 284 | struct list_head cic_list; |
283 | 285 | ||
284 | /* | 286 | /* |
285 | * Fallback dummy cfqq for extreme OOM conditions | 287 | * Fallback dummy cfqq for extreme OOM conditions |
286 | */ | 288 | */ |
287 | struct cfq_queue oom_cfqq; | 289 | struct cfq_queue oom_cfqq; |
288 | 290 | ||
289 | unsigned long last_delayed_sync; | 291 | unsigned long last_delayed_sync; |
290 | 292 | ||
291 | /* List of cfq groups being managed on this device*/ | 293 | /* List of cfq groups being managed on this device*/ |
292 | struct hlist_head cfqg_list; | 294 | struct hlist_head cfqg_list; |
293 | struct rcu_head rcu; | 295 | struct rcu_head rcu; |
294 | }; | 296 | }; |
295 | 297 | ||
296 | static struct cfq_group *cfq_get_next_cfqg(struct cfq_data *cfqd); | 298 | static struct cfq_group *cfq_get_next_cfqg(struct cfq_data *cfqd); |
297 | 299 | ||
298 | static struct cfq_rb_root *service_tree_for(struct cfq_group *cfqg, | 300 | static struct cfq_rb_root *service_tree_for(struct cfq_group *cfqg, |
299 | enum wl_prio_t prio, | 301 | enum wl_prio_t prio, |
300 | enum wl_type_t type) | 302 | enum wl_type_t type) |
301 | { | 303 | { |
302 | if (!cfqg) | 304 | if (!cfqg) |
303 | return NULL; | 305 | return NULL; |
304 | 306 | ||
305 | if (prio == IDLE_WORKLOAD) | 307 | if (prio == IDLE_WORKLOAD) |
306 | return &cfqg->service_tree_idle; | 308 | return &cfqg->service_tree_idle; |
307 | 309 | ||
308 | return &cfqg->service_trees[prio][type]; | 310 | return &cfqg->service_trees[prio][type]; |
309 | } | 311 | } |
310 | 312 | ||
311 | enum cfqq_state_flags { | 313 | enum cfqq_state_flags { |
312 | CFQ_CFQQ_FLAG_on_rr = 0, /* on round-robin busy list */ | 314 | CFQ_CFQQ_FLAG_on_rr = 0, /* on round-robin busy list */ |
313 | CFQ_CFQQ_FLAG_wait_request, /* waiting for a request */ | 315 | CFQ_CFQQ_FLAG_wait_request, /* waiting for a request */ |
314 | CFQ_CFQQ_FLAG_must_dispatch, /* must be allowed a dispatch */ | 316 | CFQ_CFQQ_FLAG_must_dispatch, /* must be allowed a dispatch */ |
315 | CFQ_CFQQ_FLAG_must_alloc_slice, /* per-slice must_alloc flag */ | 317 | CFQ_CFQQ_FLAG_must_alloc_slice, /* per-slice must_alloc flag */ |
316 | CFQ_CFQQ_FLAG_fifo_expire, /* FIFO checked in this slice */ | 318 | CFQ_CFQQ_FLAG_fifo_expire, /* FIFO checked in this slice */ |
317 | CFQ_CFQQ_FLAG_idle_window, /* slice idling enabled */ | 319 | CFQ_CFQQ_FLAG_idle_window, /* slice idling enabled */ |
318 | CFQ_CFQQ_FLAG_prio_changed, /* task priority has changed */ | 320 | CFQ_CFQQ_FLAG_prio_changed, /* task priority has changed */ |
319 | CFQ_CFQQ_FLAG_slice_new, /* no requests dispatched in slice */ | 321 | CFQ_CFQQ_FLAG_slice_new, /* no requests dispatched in slice */ |
320 | CFQ_CFQQ_FLAG_sync, /* synchronous queue */ | 322 | CFQ_CFQQ_FLAG_sync, /* synchronous queue */ |
321 | CFQ_CFQQ_FLAG_coop, /* cfqq is shared */ | 323 | CFQ_CFQQ_FLAG_coop, /* cfqq is shared */ |
322 | CFQ_CFQQ_FLAG_split_coop, /* shared cfqq will be splitted */ | 324 | CFQ_CFQQ_FLAG_split_coop, /* shared cfqq will be splitted */ |
323 | CFQ_CFQQ_FLAG_deep, /* sync cfqq experienced large depth */ | 325 | CFQ_CFQQ_FLAG_deep, /* sync cfqq experienced large depth */ |
324 | CFQ_CFQQ_FLAG_wait_busy, /* Waiting for next request */ | 326 | CFQ_CFQQ_FLAG_wait_busy, /* Waiting for next request */ |
325 | }; | 327 | }; |
326 | 328 | ||
327 | #define CFQ_CFQQ_FNS(name) \ | 329 | #define CFQ_CFQQ_FNS(name) \ |
328 | static inline void cfq_mark_cfqq_##name(struct cfq_queue *cfqq) \ | 330 | static inline void cfq_mark_cfqq_##name(struct cfq_queue *cfqq) \ |
329 | { \ | 331 | { \ |
330 | (cfqq)->flags |= (1 << CFQ_CFQQ_FLAG_##name); \ | 332 | (cfqq)->flags |= (1 << CFQ_CFQQ_FLAG_##name); \ |
331 | } \ | 333 | } \ |
332 | static inline void cfq_clear_cfqq_##name(struct cfq_queue *cfqq) \ | 334 | static inline void cfq_clear_cfqq_##name(struct cfq_queue *cfqq) \ |
333 | { \ | 335 | { \ |
334 | (cfqq)->flags &= ~(1 << CFQ_CFQQ_FLAG_##name); \ | 336 | (cfqq)->flags &= ~(1 << CFQ_CFQQ_FLAG_##name); \ |
335 | } \ | 337 | } \ |
336 | static inline int cfq_cfqq_##name(const struct cfq_queue *cfqq) \ | 338 | static inline int cfq_cfqq_##name(const struct cfq_queue *cfqq) \ |
337 | { \ | 339 | { \ |
338 | return ((cfqq)->flags & (1 << CFQ_CFQQ_FLAG_##name)) != 0; \ | 340 | return ((cfqq)->flags & (1 << CFQ_CFQQ_FLAG_##name)) != 0; \ |
339 | } | 341 | } |
340 | 342 | ||
341 | CFQ_CFQQ_FNS(on_rr); | 343 | CFQ_CFQQ_FNS(on_rr); |
342 | CFQ_CFQQ_FNS(wait_request); | 344 | CFQ_CFQQ_FNS(wait_request); |
343 | CFQ_CFQQ_FNS(must_dispatch); | 345 | CFQ_CFQQ_FNS(must_dispatch); |
344 | CFQ_CFQQ_FNS(must_alloc_slice); | 346 | CFQ_CFQQ_FNS(must_alloc_slice); |
345 | CFQ_CFQQ_FNS(fifo_expire); | 347 | CFQ_CFQQ_FNS(fifo_expire); |
346 | CFQ_CFQQ_FNS(idle_window); | 348 | CFQ_CFQQ_FNS(idle_window); |
347 | CFQ_CFQQ_FNS(prio_changed); | 349 | CFQ_CFQQ_FNS(prio_changed); |
348 | CFQ_CFQQ_FNS(slice_new); | 350 | CFQ_CFQQ_FNS(slice_new); |
349 | CFQ_CFQQ_FNS(sync); | 351 | CFQ_CFQQ_FNS(sync); |
350 | CFQ_CFQQ_FNS(coop); | 352 | CFQ_CFQQ_FNS(coop); |
351 | CFQ_CFQQ_FNS(split_coop); | 353 | CFQ_CFQQ_FNS(split_coop); |
352 | CFQ_CFQQ_FNS(deep); | 354 | CFQ_CFQQ_FNS(deep); |
353 | CFQ_CFQQ_FNS(wait_busy); | 355 | CFQ_CFQQ_FNS(wait_busy); |
354 | #undef CFQ_CFQQ_FNS | 356 | #undef CFQ_CFQQ_FNS |
355 | 357 | ||
356 | #ifdef CONFIG_CFQ_GROUP_IOSCHED | 358 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
357 | #define cfq_log_cfqq(cfqd, cfqq, fmt, args...) \ | 359 | #define cfq_log_cfqq(cfqd, cfqq, fmt, args...) \ |
358 | blk_add_trace_msg((cfqd)->queue, "cfq%d%c %s " fmt, (cfqq)->pid, \ | 360 | blk_add_trace_msg((cfqd)->queue, "cfq%d%c %s " fmt, (cfqq)->pid, \ |
359 | cfq_cfqq_sync((cfqq)) ? 'S' : 'A', \ | 361 | cfq_cfqq_sync((cfqq)) ? 'S' : 'A', \ |
360 | blkg_path(&(cfqq)->cfqg->blkg), ##args); | 362 | blkg_path(&(cfqq)->cfqg->blkg), ##args); |
361 | 363 | ||
362 | #define cfq_log_cfqg(cfqd, cfqg, fmt, args...) \ | 364 | #define cfq_log_cfqg(cfqd, cfqg, fmt, args...) \ |
363 | blk_add_trace_msg((cfqd)->queue, "%s " fmt, \ | 365 | blk_add_trace_msg((cfqd)->queue, "%s " fmt, \ |
364 | blkg_path(&(cfqg)->blkg), ##args); \ | 366 | blkg_path(&(cfqg)->blkg), ##args); \ |
365 | 367 | ||
366 | #else | 368 | #else |
367 | #define cfq_log_cfqq(cfqd, cfqq, fmt, args...) \ | 369 | #define cfq_log_cfqq(cfqd, cfqq, fmt, args...) \ |
368 | blk_add_trace_msg((cfqd)->queue, "cfq%d " fmt, (cfqq)->pid, ##args) | 370 | blk_add_trace_msg((cfqd)->queue, "cfq%d " fmt, (cfqq)->pid, ##args) |
369 | #define cfq_log_cfqg(cfqd, cfqg, fmt, args...) do {} while (0); | 371 | #define cfq_log_cfqg(cfqd, cfqg, fmt, args...) do {} while (0); |
370 | #endif | 372 | #endif |
371 | #define cfq_log(cfqd, fmt, args...) \ | 373 | #define cfq_log(cfqd, fmt, args...) \ |
372 | blk_add_trace_msg((cfqd)->queue, "cfq " fmt, ##args) | 374 | blk_add_trace_msg((cfqd)->queue, "cfq " fmt, ##args) |
373 | 375 | ||
374 | /* Traverses through cfq group service trees */ | 376 | /* Traverses through cfq group service trees */ |
375 | #define for_each_cfqg_st(cfqg, i, j, st) \ | 377 | #define for_each_cfqg_st(cfqg, i, j, st) \ |
376 | for (i = 0; i <= IDLE_WORKLOAD; i++) \ | 378 | for (i = 0; i <= IDLE_WORKLOAD; i++) \ |
377 | for (j = 0, st = i < IDLE_WORKLOAD ? &cfqg->service_trees[i][j]\ | 379 | for (j = 0, st = i < IDLE_WORKLOAD ? &cfqg->service_trees[i][j]\ |
378 | : &cfqg->service_tree_idle; \ | 380 | : &cfqg->service_tree_idle; \ |
379 | (i < IDLE_WORKLOAD && j <= SYNC_WORKLOAD) || \ | 381 | (i < IDLE_WORKLOAD && j <= SYNC_WORKLOAD) || \ |
380 | (i == IDLE_WORKLOAD && j == 0); \ | 382 | (i == IDLE_WORKLOAD && j == 0); \ |
381 | j++, st = i < IDLE_WORKLOAD ? \ | 383 | j++, st = i < IDLE_WORKLOAD ? \ |
382 | &cfqg->service_trees[i][j]: NULL) \ | 384 | &cfqg->service_trees[i][j]: NULL) \ |
383 | 385 | ||
384 | 386 | ||
385 | static inline bool iops_mode(struct cfq_data *cfqd) | 387 | static inline bool iops_mode(struct cfq_data *cfqd) |
386 | { | 388 | { |
387 | /* | 389 | /* |
388 | * If we are not idling on queues and it is a NCQ drive, parallel | 390 | * If we are not idling on queues and it is a NCQ drive, parallel |
389 | * execution of requests is on and measuring time is not possible | 391 | * execution of requests is on and measuring time is not possible |
390 | * in most of the cases until and unless we drive shallower queue | 392 | * in most of the cases until and unless we drive shallower queue |
391 | * depths and that becomes a performance bottleneck. In such cases | 393 | * depths and that becomes a performance bottleneck. In such cases |
392 | * switch to start providing fairness in terms of number of IOs. | 394 | * switch to start providing fairness in terms of number of IOs. |
393 | */ | 395 | */ |
394 | if (!cfqd->cfq_slice_idle && cfqd->hw_tag) | 396 | if (!cfqd->cfq_slice_idle && cfqd->hw_tag) |
395 | return true; | 397 | return true; |
396 | else | 398 | else |
397 | return false; | 399 | return false; |
398 | } | 400 | } |
399 | 401 | ||
400 | static inline enum wl_prio_t cfqq_prio(struct cfq_queue *cfqq) | 402 | static inline enum wl_prio_t cfqq_prio(struct cfq_queue *cfqq) |
401 | { | 403 | { |
402 | if (cfq_class_idle(cfqq)) | 404 | if (cfq_class_idle(cfqq)) |
403 | return IDLE_WORKLOAD; | 405 | return IDLE_WORKLOAD; |
404 | if (cfq_class_rt(cfqq)) | 406 | if (cfq_class_rt(cfqq)) |
405 | return RT_WORKLOAD; | 407 | return RT_WORKLOAD; |
406 | return BE_WORKLOAD; | 408 | return BE_WORKLOAD; |
407 | } | 409 | } |
408 | 410 | ||
409 | 411 | ||
410 | static enum wl_type_t cfqq_type(struct cfq_queue *cfqq) | 412 | static enum wl_type_t cfqq_type(struct cfq_queue *cfqq) |
411 | { | 413 | { |
412 | if (!cfq_cfqq_sync(cfqq)) | 414 | if (!cfq_cfqq_sync(cfqq)) |
413 | return ASYNC_WORKLOAD; | 415 | return ASYNC_WORKLOAD; |
414 | if (!cfq_cfqq_idle_window(cfqq)) | 416 | if (!cfq_cfqq_idle_window(cfqq)) |
415 | return SYNC_NOIDLE_WORKLOAD; | 417 | return SYNC_NOIDLE_WORKLOAD; |
416 | return SYNC_WORKLOAD; | 418 | return SYNC_WORKLOAD; |
417 | } | 419 | } |
418 | 420 | ||
419 | static inline int cfq_group_busy_queues_wl(enum wl_prio_t wl, | 421 | static inline int cfq_group_busy_queues_wl(enum wl_prio_t wl, |
420 | struct cfq_data *cfqd, | 422 | struct cfq_data *cfqd, |
421 | struct cfq_group *cfqg) | 423 | struct cfq_group *cfqg) |
422 | { | 424 | { |
423 | if (wl == IDLE_WORKLOAD) | 425 | if (wl == IDLE_WORKLOAD) |
424 | return cfqg->service_tree_idle.count; | 426 | return cfqg->service_tree_idle.count; |
425 | 427 | ||
426 | return cfqg->service_trees[wl][ASYNC_WORKLOAD].count | 428 | return cfqg->service_trees[wl][ASYNC_WORKLOAD].count |
427 | + cfqg->service_trees[wl][SYNC_NOIDLE_WORKLOAD].count | 429 | + cfqg->service_trees[wl][SYNC_NOIDLE_WORKLOAD].count |
428 | + cfqg->service_trees[wl][SYNC_WORKLOAD].count; | 430 | + cfqg->service_trees[wl][SYNC_WORKLOAD].count; |
429 | } | 431 | } |
430 | 432 | ||
431 | static inline int cfqg_busy_async_queues(struct cfq_data *cfqd, | 433 | static inline int cfqg_busy_async_queues(struct cfq_data *cfqd, |
432 | struct cfq_group *cfqg) | 434 | struct cfq_group *cfqg) |
433 | { | 435 | { |
434 | return cfqg->service_trees[RT_WORKLOAD][ASYNC_WORKLOAD].count | 436 | return cfqg->service_trees[RT_WORKLOAD][ASYNC_WORKLOAD].count |
435 | + cfqg->service_trees[BE_WORKLOAD][ASYNC_WORKLOAD].count; | 437 | + cfqg->service_trees[BE_WORKLOAD][ASYNC_WORKLOAD].count; |
436 | } | 438 | } |
437 | 439 | ||
438 | static void cfq_dispatch_insert(struct request_queue *, struct request *); | 440 | static void cfq_dispatch_insert(struct request_queue *, struct request *); |
439 | static struct cfq_queue *cfq_get_queue(struct cfq_data *, bool, | 441 | static struct cfq_queue *cfq_get_queue(struct cfq_data *, bool, |
440 | struct io_context *, gfp_t); | 442 | struct io_context *, gfp_t); |
441 | static struct cfq_io_context *cfq_cic_lookup(struct cfq_data *, | 443 | static struct cfq_io_context *cfq_cic_lookup(struct cfq_data *, |
442 | struct io_context *); | 444 | struct io_context *); |
443 | 445 | ||
444 | static inline struct cfq_queue *cic_to_cfqq(struct cfq_io_context *cic, | 446 | static inline struct cfq_queue *cic_to_cfqq(struct cfq_io_context *cic, |
445 | bool is_sync) | 447 | bool is_sync) |
446 | { | 448 | { |
447 | return cic->cfqq[is_sync]; | 449 | return cic->cfqq[is_sync]; |
448 | } | 450 | } |
449 | 451 | ||
450 | static inline void cic_set_cfqq(struct cfq_io_context *cic, | 452 | static inline void cic_set_cfqq(struct cfq_io_context *cic, |
451 | struct cfq_queue *cfqq, bool is_sync) | 453 | struct cfq_queue *cfqq, bool is_sync) |
452 | { | 454 | { |
453 | cic->cfqq[is_sync] = cfqq; | 455 | cic->cfqq[is_sync] = cfqq; |
454 | } | 456 | } |
455 | 457 | ||
456 | #define CIC_DEAD_KEY 1ul | 458 | #define CIC_DEAD_KEY 1ul |
457 | #define CIC_DEAD_INDEX_SHIFT 1 | 459 | #define CIC_DEAD_INDEX_SHIFT 1 |
458 | 460 | ||
459 | static inline void *cfqd_dead_key(struct cfq_data *cfqd) | 461 | static inline void *cfqd_dead_key(struct cfq_data *cfqd) |
460 | { | 462 | { |
461 | return (void *)(cfqd->cic_index << CIC_DEAD_INDEX_SHIFT | CIC_DEAD_KEY); | 463 | return (void *)(cfqd->cic_index << CIC_DEAD_INDEX_SHIFT | CIC_DEAD_KEY); |
462 | } | 464 | } |
463 | 465 | ||
464 | static inline struct cfq_data *cic_to_cfqd(struct cfq_io_context *cic) | 466 | static inline struct cfq_data *cic_to_cfqd(struct cfq_io_context *cic) |
465 | { | 467 | { |
466 | struct cfq_data *cfqd = cic->key; | 468 | struct cfq_data *cfqd = cic->key; |
467 | 469 | ||
468 | if (unlikely((unsigned long) cfqd & CIC_DEAD_KEY)) | 470 | if (unlikely((unsigned long) cfqd & CIC_DEAD_KEY)) |
469 | return NULL; | 471 | return NULL; |
470 | 472 | ||
471 | return cfqd; | 473 | return cfqd; |
472 | } | 474 | } |
473 | 475 | ||
474 | /* | 476 | /* |
475 | * We regard a request as SYNC, if it's either a read or has the SYNC bit | 477 | * We regard a request as SYNC, if it's either a read or has the SYNC bit |
476 | * set (in which case it could also be direct WRITE). | 478 | * set (in which case it could also be direct WRITE). |
477 | */ | 479 | */ |
478 | static inline bool cfq_bio_sync(struct bio *bio) | 480 | static inline bool cfq_bio_sync(struct bio *bio) |
479 | { | 481 | { |
480 | return bio_data_dir(bio) == READ || (bio->bi_rw & REQ_SYNC); | 482 | return bio_data_dir(bio) == READ || (bio->bi_rw & REQ_SYNC); |
481 | } | 483 | } |
482 | 484 | ||
483 | /* | 485 | /* |
484 | * scheduler run of queue, if there are requests pending and no one in the | 486 | * scheduler run of queue, if there are requests pending and no one in the |
485 | * driver that will restart queueing | 487 | * driver that will restart queueing |
486 | */ | 488 | */ |
487 | static inline void cfq_schedule_dispatch(struct cfq_data *cfqd) | 489 | static inline void cfq_schedule_dispatch(struct cfq_data *cfqd) |
488 | { | 490 | { |
489 | if (cfqd->busy_queues) { | 491 | if (cfqd->busy_queues) { |
490 | cfq_log(cfqd, "schedule dispatch"); | 492 | cfq_log(cfqd, "schedule dispatch"); |
491 | kblockd_schedule_work(cfqd->queue, &cfqd->unplug_work); | 493 | kblockd_schedule_work(cfqd->queue, &cfqd->unplug_work); |
492 | } | 494 | } |
493 | } | 495 | } |
494 | 496 | ||
495 | static int cfq_queue_empty(struct request_queue *q) | 497 | static int cfq_queue_empty(struct request_queue *q) |
496 | { | 498 | { |
497 | struct cfq_data *cfqd = q->elevator->elevator_data; | 499 | struct cfq_data *cfqd = q->elevator->elevator_data; |
498 | 500 | ||
499 | return !cfqd->rq_queued; | 501 | return !cfqd->rq_queued; |
500 | } | 502 | } |
501 | 503 | ||
502 | /* | 504 | /* |
503 | * Scale schedule slice based on io priority. Use the sync time slice only | 505 | * Scale schedule slice based on io priority. Use the sync time slice only |
504 | * if a queue is marked sync and has sync io queued. A sync queue with async | 506 | * if a queue is marked sync and has sync io queued. A sync queue with async |
505 | * io only, should not get full sync slice length. | 507 | * io only, should not get full sync slice length. |
506 | */ | 508 | */ |
507 | static inline int cfq_prio_slice(struct cfq_data *cfqd, bool sync, | 509 | static inline int cfq_prio_slice(struct cfq_data *cfqd, bool sync, |
508 | unsigned short prio) | 510 | unsigned short prio) |
509 | { | 511 | { |
510 | const int base_slice = cfqd->cfq_slice[sync]; | 512 | const int base_slice = cfqd->cfq_slice[sync]; |
511 | 513 | ||
512 | WARN_ON(prio >= IOPRIO_BE_NR); | 514 | WARN_ON(prio >= IOPRIO_BE_NR); |
513 | 515 | ||
514 | return base_slice + (base_slice/CFQ_SLICE_SCALE * (4 - prio)); | 516 | return base_slice + (base_slice/CFQ_SLICE_SCALE * (4 - prio)); |
515 | } | 517 | } |
516 | 518 | ||
517 | static inline int | 519 | static inline int |
518 | cfq_prio_to_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) | 520 | cfq_prio_to_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
519 | { | 521 | { |
520 | return cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio); | 522 | return cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio); |
521 | } | 523 | } |
522 | 524 | ||
523 | static inline u64 cfq_scale_slice(unsigned long delta, struct cfq_group *cfqg) | 525 | static inline u64 cfq_scale_slice(unsigned long delta, struct cfq_group *cfqg) |
524 | { | 526 | { |
525 | u64 d = delta << CFQ_SERVICE_SHIFT; | 527 | u64 d = delta << CFQ_SERVICE_SHIFT; |
526 | 528 | ||
527 | d = d * BLKIO_WEIGHT_DEFAULT; | 529 | d = d * BLKIO_WEIGHT_DEFAULT; |
528 | do_div(d, cfqg->weight); | 530 | do_div(d, cfqg->weight); |
529 | return d; | 531 | return d; |
530 | } | 532 | } |
531 | 533 | ||
532 | static inline u64 max_vdisktime(u64 min_vdisktime, u64 vdisktime) | 534 | static inline u64 max_vdisktime(u64 min_vdisktime, u64 vdisktime) |
533 | { | 535 | { |
534 | s64 delta = (s64)(vdisktime - min_vdisktime); | 536 | s64 delta = (s64)(vdisktime - min_vdisktime); |
535 | if (delta > 0) | 537 | if (delta > 0) |
536 | min_vdisktime = vdisktime; | 538 | min_vdisktime = vdisktime; |
537 | 539 | ||
538 | return min_vdisktime; | 540 | return min_vdisktime; |
539 | } | 541 | } |
540 | 542 | ||
541 | static inline u64 min_vdisktime(u64 min_vdisktime, u64 vdisktime) | 543 | static inline u64 min_vdisktime(u64 min_vdisktime, u64 vdisktime) |
542 | { | 544 | { |
543 | s64 delta = (s64)(vdisktime - min_vdisktime); | 545 | s64 delta = (s64)(vdisktime - min_vdisktime); |
544 | if (delta < 0) | 546 | if (delta < 0) |
545 | min_vdisktime = vdisktime; | 547 | min_vdisktime = vdisktime; |
546 | 548 | ||
547 | return min_vdisktime; | 549 | return min_vdisktime; |
548 | } | 550 | } |
549 | 551 | ||
550 | static void update_min_vdisktime(struct cfq_rb_root *st) | 552 | static void update_min_vdisktime(struct cfq_rb_root *st) |
551 | { | 553 | { |
552 | u64 vdisktime = st->min_vdisktime; | 554 | u64 vdisktime = st->min_vdisktime; |
553 | struct cfq_group *cfqg; | 555 | struct cfq_group *cfqg; |
554 | 556 | ||
555 | if (st->active) { | 557 | if (st->active) { |
556 | cfqg = rb_entry_cfqg(st->active); | 558 | cfqg = rb_entry_cfqg(st->active); |
557 | vdisktime = cfqg->vdisktime; | 559 | vdisktime = cfqg->vdisktime; |
558 | } | 560 | } |
559 | 561 | ||
560 | if (st->left) { | 562 | if (st->left) { |
561 | cfqg = rb_entry_cfqg(st->left); | 563 | cfqg = rb_entry_cfqg(st->left); |
562 | vdisktime = min_vdisktime(vdisktime, cfqg->vdisktime); | 564 | vdisktime = min_vdisktime(vdisktime, cfqg->vdisktime); |
563 | } | 565 | } |
564 | 566 | ||
565 | st->min_vdisktime = max_vdisktime(st->min_vdisktime, vdisktime); | 567 | st->min_vdisktime = max_vdisktime(st->min_vdisktime, vdisktime); |
566 | } | 568 | } |
567 | 569 | ||
568 | /* | 570 | /* |
569 | * get averaged number of queues of RT/BE priority. | 571 | * get averaged number of queues of RT/BE priority. |
570 | * average is updated, with a formula that gives more weight to higher numbers, | 572 | * average is updated, with a formula that gives more weight to higher numbers, |
571 | * to quickly follows sudden increases and decrease slowly | 573 | * to quickly follows sudden increases and decrease slowly |
572 | */ | 574 | */ |
573 | 575 | ||
574 | static inline unsigned cfq_group_get_avg_queues(struct cfq_data *cfqd, | 576 | static inline unsigned cfq_group_get_avg_queues(struct cfq_data *cfqd, |
575 | struct cfq_group *cfqg, bool rt) | 577 | struct cfq_group *cfqg, bool rt) |
576 | { | 578 | { |
577 | unsigned min_q, max_q; | 579 | unsigned min_q, max_q; |
578 | unsigned mult = cfq_hist_divisor - 1; | 580 | unsigned mult = cfq_hist_divisor - 1; |
579 | unsigned round = cfq_hist_divisor / 2; | 581 | unsigned round = cfq_hist_divisor / 2; |
580 | unsigned busy = cfq_group_busy_queues_wl(rt, cfqd, cfqg); | 582 | unsigned busy = cfq_group_busy_queues_wl(rt, cfqd, cfqg); |
581 | 583 | ||
582 | min_q = min(cfqg->busy_queues_avg[rt], busy); | 584 | min_q = min(cfqg->busy_queues_avg[rt], busy); |
583 | max_q = max(cfqg->busy_queues_avg[rt], busy); | 585 | max_q = max(cfqg->busy_queues_avg[rt], busy); |
584 | cfqg->busy_queues_avg[rt] = (mult * max_q + min_q + round) / | 586 | cfqg->busy_queues_avg[rt] = (mult * max_q + min_q + round) / |
585 | cfq_hist_divisor; | 587 | cfq_hist_divisor; |
586 | return cfqg->busy_queues_avg[rt]; | 588 | return cfqg->busy_queues_avg[rt]; |
587 | } | 589 | } |
588 | 590 | ||
589 | static inline unsigned | 591 | static inline unsigned |
590 | cfq_group_slice(struct cfq_data *cfqd, struct cfq_group *cfqg) | 592 | cfq_group_slice(struct cfq_data *cfqd, struct cfq_group *cfqg) |
591 | { | 593 | { |
592 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | 594 | struct cfq_rb_root *st = &cfqd->grp_service_tree; |
593 | 595 | ||
594 | return cfq_target_latency * cfqg->weight / st->total_weight; | 596 | return cfq_target_latency * cfqg->weight / st->total_weight; |
595 | } | 597 | } |
596 | 598 | ||
597 | static inline void | 599 | static inline void |
598 | cfq_set_prio_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) | 600 | cfq_set_prio_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
599 | { | 601 | { |
600 | unsigned slice = cfq_prio_to_slice(cfqd, cfqq); | 602 | unsigned slice = cfq_prio_to_slice(cfqd, cfqq); |
601 | if (cfqd->cfq_latency) { | 603 | if (cfqd->cfq_latency) { |
602 | /* | 604 | /* |
603 | * interested queues (we consider only the ones with the same | 605 | * interested queues (we consider only the ones with the same |
604 | * priority class in the cfq group) | 606 | * priority class in the cfq group) |
605 | */ | 607 | */ |
606 | unsigned iq = cfq_group_get_avg_queues(cfqd, cfqq->cfqg, | 608 | unsigned iq = cfq_group_get_avg_queues(cfqd, cfqq->cfqg, |
607 | cfq_class_rt(cfqq)); | 609 | cfq_class_rt(cfqq)); |
608 | unsigned sync_slice = cfqd->cfq_slice[1]; | 610 | unsigned sync_slice = cfqd->cfq_slice[1]; |
609 | unsigned expect_latency = sync_slice * iq; | 611 | unsigned expect_latency = sync_slice * iq; |
610 | unsigned group_slice = cfq_group_slice(cfqd, cfqq->cfqg); | 612 | unsigned group_slice = cfq_group_slice(cfqd, cfqq->cfqg); |
611 | 613 | ||
612 | if (expect_latency > group_slice) { | 614 | if (expect_latency > group_slice) { |
613 | unsigned base_low_slice = 2 * cfqd->cfq_slice_idle; | 615 | unsigned base_low_slice = 2 * cfqd->cfq_slice_idle; |
614 | /* scale low_slice according to IO priority | 616 | /* scale low_slice according to IO priority |
615 | * and sync vs async */ | 617 | * and sync vs async */ |
616 | unsigned low_slice = | 618 | unsigned low_slice = |
617 | min(slice, base_low_slice * slice / sync_slice); | 619 | min(slice, base_low_slice * slice / sync_slice); |
618 | /* the adapted slice value is scaled to fit all iqs | 620 | /* the adapted slice value is scaled to fit all iqs |
619 | * into the target latency */ | 621 | * into the target latency */ |
620 | slice = max(slice * group_slice / expect_latency, | 622 | slice = max(slice * group_slice / expect_latency, |
621 | low_slice); | 623 | low_slice); |
622 | } | 624 | } |
623 | } | 625 | } |
624 | cfqq->slice_start = jiffies; | 626 | cfqq->slice_start = jiffies; |
625 | cfqq->slice_end = jiffies + slice; | 627 | cfqq->slice_end = jiffies + slice; |
626 | cfqq->allocated_slice = slice; | 628 | cfqq->allocated_slice = slice; |
627 | cfq_log_cfqq(cfqd, cfqq, "set_slice=%lu", cfqq->slice_end - jiffies); | 629 | cfq_log_cfqq(cfqd, cfqq, "set_slice=%lu", cfqq->slice_end - jiffies); |
628 | } | 630 | } |
629 | 631 | ||
630 | /* | 632 | /* |
631 | * We need to wrap this check in cfq_cfqq_slice_new(), since ->slice_end | 633 | * We need to wrap this check in cfq_cfqq_slice_new(), since ->slice_end |
632 | * isn't valid until the first request from the dispatch is activated | 634 | * isn't valid until the first request from the dispatch is activated |
633 | * and the slice time set. | 635 | * and the slice time set. |
634 | */ | 636 | */ |
635 | static inline bool cfq_slice_used(struct cfq_queue *cfqq) | 637 | static inline bool cfq_slice_used(struct cfq_queue *cfqq) |
636 | { | 638 | { |
637 | if (cfq_cfqq_slice_new(cfqq)) | 639 | if (cfq_cfqq_slice_new(cfqq)) |
638 | return 0; | 640 | return 0; |
639 | if (time_before(jiffies, cfqq->slice_end)) | 641 | if (time_before(jiffies, cfqq->slice_end)) |
640 | return 0; | 642 | return 0; |
641 | 643 | ||
642 | return 1; | 644 | return 1; |
643 | } | 645 | } |
644 | 646 | ||
645 | /* | 647 | /* |
646 | * Lifted from AS - choose which of rq1 and rq2 that is best served now. | 648 | * Lifted from AS - choose which of rq1 and rq2 that is best served now. |
647 | * We choose the request that is closest to the head right now. Distance | 649 | * We choose the request that is closest to the head right now. Distance |
648 | * behind the head is penalized and only allowed to a certain extent. | 650 | * behind the head is penalized and only allowed to a certain extent. |
649 | */ | 651 | */ |
650 | static struct request * | 652 | static struct request * |
651 | cfq_choose_req(struct cfq_data *cfqd, struct request *rq1, struct request *rq2, sector_t last) | 653 | cfq_choose_req(struct cfq_data *cfqd, struct request *rq1, struct request *rq2, sector_t last) |
652 | { | 654 | { |
653 | sector_t s1, s2, d1 = 0, d2 = 0; | 655 | sector_t s1, s2, d1 = 0, d2 = 0; |
654 | unsigned long back_max; | 656 | unsigned long back_max; |
655 | #define CFQ_RQ1_WRAP 0x01 /* request 1 wraps */ | 657 | #define CFQ_RQ1_WRAP 0x01 /* request 1 wraps */ |
656 | #define CFQ_RQ2_WRAP 0x02 /* request 2 wraps */ | 658 | #define CFQ_RQ2_WRAP 0x02 /* request 2 wraps */ |
657 | unsigned wrap = 0; /* bit mask: requests behind the disk head? */ | 659 | unsigned wrap = 0; /* bit mask: requests behind the disk head? */ |
658 | 660 | ||
659 | if (rq1 == NULL || rq1 == rq2) | 661 | if (rq1 == NULL || rq1 == rq2) |
660 | return rq2; | 662 | return rq2; |
661 | if (rq2 == NULL) | 663 | if (rq2 == NULL) |
662 | return rq1; | 664 | return rq1; |
663 | 665 | ||
664 | if (rq_is_sync(rq1) && !rq_is_sync(rq2)) | 666 | if (rq_is_sync(rq1) && !rq_is_sync(rq2)) |
665 | return rq1; | 667 | return rq1; |
666 | else if (rq_is_sync(rq2) && !rq_is_sync(rq1)) | 668 | else if (rq_is_sync(rq2) && !rq_is_sync(rq1)) |
667 | return rq2; | 669 | return rq2; |
668 | if ((rq1->cmd_flags & REQ_META) && !(rq2->cmd_flags & REQ_META)) | 670 | if ((rq1->cmd_flags & REQ_META) && !(rq2->cmd_flags & REQ_META)) |
669 | return rq1; | 671 | return rq1; |
670 | else if ((rq2->cmd_flags & REQ_META) && | 672 | else if ((rq2->cmd_flags & REQ_META) && |
671 | !(rq1->cmd_flags & REQ_META)) | 673 | !(rq1->cmd_flags & REQ_META)) |
672 | return rq2; | 674 | return rq2; |
673 | 675 | ||
674 | s1 = blk_rq_pos(rq1); | 676 | s1 = blk_rq_pos(rq1); |
675 | s2 = blk_rq_pos(rq2); | 677 | s2 = blk_rq_pos(rq2); |
676 | 678 | ||
677 | /* | 679 | /* |
678 | * by definition, 1KiB is 2 sectors | 680 | * by definition, 1KiB is 2 sectors |
679 | */ | 681 | */ |
680 | back_max = cfqd->cfq_back_max * 2; | 682 | back_max = cfqd->cfq_back_max * 2; |
681 | 683 | ||
682 | /* | 684 | /* |
683 | * Strict one way elevator _except_ in the case where we allow | 685 | * Strict one way elevator _except_ in the case where we allow |
684 | * short backward seeks which are biased as twice the cost of a | 686 | * short backward seeks which are biased as twice the cost of a |
685 | * similar forward seek. | 687 | * similar forward seek. |
686 | */ | 688 | */ |
687 | if (s1 >= last) | 689 | if (s1 >= last) |
688 | d1 = s1 - last; | 690 | d1 = s1 - last; |
689 | else if (s1 + back_max >= last) | 691 | else if (s1 + back_max >= last) |
690 | d1 = (last - s1) * cfqd->cfq_back_penalty; | 692 | d1 = (last - s1) * cfqd->cfq_back_penalty; |
691 | else | 693 | else |
692 | wrap |= CFQ_RQ1_WRAP; | 694 | wrap |= CFQ_RQ1_WRAP; |
693 | 695 | ||
694 | if (s2 >= last) | 696 | if (s2 >= last) |
695 | d2 = s2 - last; | 697 | d2 = s2 - last; |
696 | else if (s2 + back_max >= last) | 698 | else if (s2 + back_max >= last) |
697 | d2 = (last - s2) * cfqd->cfq_back_penalty; | 699 | d2 = (last - s2) * cfqd->cfq_back_penalty; |
698 | else | 700 | else |
699 | wrap |= CFQ_RQ2_WRAP; | 701 | wrap |= CFQ_RQ2_WRAP; |
700 | 702 | ||
701 | /* Found required data */ | 703 | /* Found required data */ |
702 | 704 | ||
703 | /* | 705 | /* |
704 | * By doing switch() on the bit mask "wrap" we avoid having to | 706 | * By doing switch() on the bit mask "wrap" we avoid having to |
705 | * check two variables for all permutations: --> faster! | 707 | * check two variables for all permutations: --> faster! |
706 | */ | 708 | */ |
707 | switch (wrap) { | 709 | switch (wrap) { |
708 | case 0: /* common case for CFQ: rq1 and rq2 not wrapped */ | 710 | case 0: /* common case for CFQ: rq1 and rq2 not wrapped */ |
709 | if (d1 < d2) | 711 | if (d1 < d2) |
710 | return rq1; | 712 | return rq1; |
711 | else if (d2 < d1) | 713 | else if (d2 < d1) |
712 | return rq2; | 714 | return rq2; |
713 | else { | 715 | else { |
714 | if (s1 >= s2) | 716 | if (s1 >= s2) |
715 | return rq1; | 717 | return rq1; |
716 | else | 718 | else |
717 | return rq2; | 719 | return rq2; |
718 | } | 720 | } |
719 | 721 | ||
720 | case CFQ_RQ2_WRAP: | 722 | case CFQ_RQ2_WRAP: |
721 | return rq1; | 723 | return rq1; |
722 | case CFQ_RQ1_WRAP: | 724 | case CFQ_RQ1_WRAP: |
723 | return rq2; | 725 | return rq2; |
724 | case (CFQ_RQ1_WRAP|CFQ_RQ2_WRAP): /* both rqs wrapped */ | 726 | case (CFQ_RQ1_WRAP|CFQ_RQ2_WRAP): /* both rqs wrapped */ |
725 | default: | 727 | default: |
726 | /* | 728 | /* |
727 | * Since both rqs are wrapped, | 729 | * Since both rqs are wrapped, |
728 | * start with the one that's further behind head | 730 | * start with the one that's further behind head |
729 | * (--> only *one* back seek required), | 731 | * (--> only *one* back seek required), |
730 | * since back seek takes more time than forward. | 732 | * since back seek takes more time than forward. |
731 | */ | 733 | */ |
732 | if (s1 <= s2) | 734 | if (s1 <= s2) |
733 | return rq1; | 735 | return rq1; |
734 | else | 736 | else |
735 | return rq2; | 737 | return rq2; |
736 | } | 738 | } |
737 | } | 739 | } |
738 | 740 | ||
739 | /* | 741 | /* |
740 | * The below is leftmost cache rbtree addon | 742 | * The below is leftmost cache rbtree addon |
741 | */ | 743 | */ |
742 | static struct cfq_queue *cfq_rb_first(struct cfq_rb_root *root) | 744 | static struct cfq_queue *cfq_rb_first(struct cfq_rb_root *root) |
743 | { | 745 | { |
744 | /* Service tree is empty */ | 746 | /* Service tree is empty */ |
745 | if (!root->count) | 747 | if (!root->count) |
746 | return NULL; | 748 | return NULL; |
747 | 749 | ||
748 | if (!root->left) | 750 | if (!root->left) |
749 | root->left = rb_first(&root->rb); | 751 | root->left = rb_first(&root->rb); |
750 | 752 | ||
751 | if (root->left) | 753 | if (root->left) |
752 | return rb_entry(root->left, struct cfq_queue, rb_node); | 754 | return rb_entry(root->left, struct cfq_queue, rb_node); |
753 | 755 | ||
754 | return NULL; | 756 | return NULL; |
755 | } | 757 | } |
756 | 758 | ||
757 | static struct cfq_group *cfq_rb_first_group(struct cfq_rb_root *root) | 759 | static struct cfq_group *cfq_rb_first_group(struct cfq_rb_root *root) |
758 | { | 760 | { |
759 | if (!root->left) | 761 | if (!root->left) |
760 | root->left = rb_first(&root->rb); | 762 | root->left = rb_first(&root->rb); |
761 | 763 | ||
762 | if (root->left) | 764 | if (root->left) |
763 | return rb_entry_cfqg(root->left); | 765 | return rb_entry_cfqg(root->left); |
764 | 766 | ||
765 | return NULL; | 767 | return NULL; |
766 | } | 768 | } |
767 | 769 | ||
768 | static void rb_erase_init(struct rb_node *n, struct rb_root *root) | 770 | static void rb_erase_init(struct rb_node *n, struct rb_root *root) |
769 | { | 771 | { |
770 | rb_erase(n, root); | 772 | rb_erase(n, root); |
771 | RB_CLEAR_NODE(n); | 773 | RB_CLEAR_NODE(n); |
772 | } | 774 | } |
773 | 775 | ||
774 | static void cfq_rb_erase(struct rb_node *n, struct cfq_rb_root *root) | 776 | static void cfq_rb_erase(struct rb_node *n, struct cfq_rb_root *root) |
775 | { | 777 | { |
776 | if (root->left == n) | 778 | if (root->left == n) |
777 | root->left = NULL; | 779 | root->left = NULL; |
778 | rb_erase_init(n, &root->rb); | 780 | rb_erase_init(n, &root->rb); |
779 | --root->count; | 781 | --root->count; |
780 | } | 782 | } |
781 | 783 | ||
782 | /* | 784 | /* |
783 | * would be nice to take fifo expire time into account as well | 785 | * would be nice to take fifo expire time into account as well |
784 | */ | 786 | */ |
785 | static struct request * | 787 | static struct request * |
786 | cfq_find_next_rq(struct cfq_data *cfqd, struct cfq_queue *cfqq, | 788 | cfq_find_next_rq(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
787 | struct request *last) | 789 | struct request *last) |
788 | { | 790 | { |
789 | struct rb_node *rbnext = rb_next(&last->rb_node); | 791 | struct rb_node *rbnext = rb_next(&last->rb_node); |
790 | struct rb_node *rbprev = rb_prev(&last->rb_node); | 792 | struct rb_node *rbprev = rb_prev(&last->rb_node); |
791 | struct request *next = NULL, *prev = NULL; | 793 | struct request *next = NULL, *prev = NULL; |
792 | 794 | ||
793 | BUG_ON(RB_EMPTY_NODE(&last->rb_node)); | 795 | BUG_ON(RB_EMPTY_NODE(&last->rb_node)); |
794 | 796 | ||
795 | if (rbprev) | 797 | if (rbprev) |
796 | prev = rb_entry_rq(rbprev); | 798 | prev = rb_entry_rq(rbprev); |
797 | 799 | ||
798 | if (rbnext) | 800 | if (rbnext) |
799 | next = rb_entry_rq(rbnext); | 801 | next = rb_entry_rq(rbnext); |
800 | else { | 802 | else { |
801 | rbnext = rb_first(&cfqq->sort_list); | 803 | rbnext = rb_first(&cfqq->sort_list); |
802 | if (rbnext && rbnext != &last->rb_node) | 804 | if (rbnext && rbnext != &last->rb_node) |
803 | next = rb_entry_rq(rbnext); | 805 | next = rb_entry_rq(rbnext); |
804 | } | 806 | } |
805 | 807 | ||
806 | return cfq_choose_req(cfqd, next, prev, blk_rq_pos(last)); | 808 | return cfq_choose_req(cfqd, next, prev, blk_rq_pos(last)); |
807 | } | 809 | } |
808 | 810 | ||
809 | static unsigned long cfq_slice_offset(struct cfq_data *cfqd, | 811 | static unsigned long cfq_slice_offset(struct cfq_data *cfqd, |
810 | struct cfq_queue *cfqq) | 812 | struct cfq_queue *cfqq) |
811 | { | 813 | { |
812 | /* | 814 | /* |
813 | * just an approximation, should be ok. | 815 | * just an approximation, should be ok. |
814 | */ | 816 | */ |
815 | return (cfqq->cfqg->nr_cfqq - 1) * (cfq_prio_slice(cfqd, 1, 0) - | 817 | return (cfqq->cfqg->nr_cfqq - 1) * (cfq_prio_slice(cfqd, 1, 0) - |
816 | cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio)); | 818 | cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio)); |
817 | } | 819 | } |
818 | 820 | ||
819 | static inline s64 | 821 | static inline s64 |
820 | cfqg_key(struct cfq_rb_root *st, struct cfq_group *cfqg) | 822 | cfqg_key(struct cfq_rb_root *st, struct cfq_group *cfqg) |
821 | { | 823 | { |
822 | return cfqg->vdisktime - st->min_vdisktime; | 824 | return cfqg->vdisktime - st->min_vdisktime; |
823 | } | 825 | } |
824 | 826 | ||
825 | static void | 827 | static void |
826 | __cfq_group_service_tree_add(struct cfq_rb_root *st, struct cfq_group *cfqg) | 828 | __cfq_group_service_tree_add(struct cfq_rb_root *st, struct cfq_group *cfqg) |
827 | { | 829 | { |
828 | struct rb_node **node = &st->rb.rb_node; | 830 | struct rb_node **node = &st->rb.rb_node; |
829 | struct rb_node *parent = NULL; | 831 | struct rb_node *parent = NULL; |
830 | struct cfq_group *__cfqg; | 832 | struct cfq_group *__cfqg; |
831 | s64 key = cfqg_key(st, cfqg); | 833 | s64 key = cfqg_key(st, cfqg); |
832 | int left = 1; | 834 | int left = 1; |
833 | 835 | ||
834 | while (*node != NULL) { | 836 | while (*node != NULL) { |
835 | parent = *node; | 837 | parent = *node; |
836 | __cfqg = rb_entry_cfqg(parent); | 838 | __cfqg = rb_entry_cfqg(parent); |
837 | 839 | ||
838 | if (key < cfqg_key(st, __cfqg)) | 840 | if (key < cfqg_key(st, __cfqg)) |
839 | node = &parent->rb_left; | 841 | node = &parent->rb_left; |
840 | else { | 842 | else { |
841 | node = &parent->rb_right; | 843 | node = &parent->rb_right; |
842 | left = 0; | 844 | left = 0; |
843 | } | 845 | } |
844 | } | 846 | } |
845 | 847 | ||
846 | if (left) | 848 | if (left) |
847 | st->left = &cfqg->rb_node; | 849 | st->left = &cfqg->rb_node; |
848 | 850 | ||
849 | rb_link_node(&cfqg->rb_node, parent, node); | 851 | rb_link_node(&cfqg->rb_node, parent, node); |
850 | rb_insert_color(&cfqg->rb_node, &st->rb); | 852 | rb_insert_color(&cfqg->rb_node, &st->rb); |
851 | } | 853 | } |
852 | 854 | ||
853 | static void | 855 | static void |
854 | cfq_group_service_tree_add(struct cfq_data *cfqd, struct cfq_group *cfqg) | 856 | cfq_group_service_tree_add(struct cfq_data *cfqd, struct cfq_group *cfqg) |
855 | { | 857 | { |
856 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | 858 | struct cfq_rb_root *st = &cfqd->grp_service_tree; |
857 | struct cfq_group *__cfqg; | 859 | struct cfq_group *__cfqg; |
858 | struct rb_node *n; | 860 | struct rb_node *n; |
859 | 861 | ||
860 | cfqg->nr_cfqq++; | 862 | cfqg->nr_cfqq++; |
861 | if (cfqg->on_st) | 863 | if (cfqg->on_st) |
862 | return; | 864 | return; |
863 | 865 | ||
864 | /* | 866 | /* |
865 | * Currently put the group at the end. Later implement something | 867 | * Currently put the group at the end. Later implement something |
866 | * so that groups get lesser vtime based on their weights, so that | 868 | * so that groups get lesser vtime based on their weights, so that |
867 | * if group does not loose all if it was not continously backlogged. | 869 | * if group does not loose all if it was not continously backlogged. |
868 | */ | 870 | */ |
869 | n = rb_last(&st->rb); | 871 | n = rb_last(&st->rb); |
870 | if (n) { | 872 | if (n) { |
871 | __cfqg = rb_entry_cfqg(n); | 873 | __cfqg = rb_entry_cfqg(n); |
872 | cfqg->vdisktime = __cfqg->vdisktime + CFQ_IDLE_DELAY; | 874 | cfqg->vdisktime = __cfqg->vdisktime + CFQ_IDLE_DELAY; |
873 | } else | 875 | } else |
874 | cfqg->vdisktime = st->min_vdisktime; | 876 | cfqg->vdisktime = st->min_vdisktime; |
875 | 877 | ||
876 | __cfq_group_service_tree_add(st, cfqg); | 878 | __cfq_group_service_tree_add(st, cfqg); |
877 | cfqg->on_st = true; | 879 | cfqg->on_st = true; |
878 | st->total_weight += cfqg->weight; | 880 | st->total_weight += cfqg->weight; |
879 | } | 881 | } |
880 | 882 | ||
881 | static void | 883 | static void |
882 | cfq_group_service_tree_del(struct cfq_data *cfqd, struct cfq_group *cfqg) | 884 | cfq_group_service_tree_del(struct cfq_data *cfqd, struct cfq_group *cfqg) |
883 | { | 885 | { |
884 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | 886 | struct cfq_rb_root *st = &cfqd->grp_service_tree; |
885 | 887 | ||
886 | if (st->active == &cfqg->rb_node) | 888 | if (st->active == &cfqg->rb_node) |
887 | st->active = NULL; | 889 | st->active = NULL; |
888 | 890 | ||
889 | BUG_ON(cfqg->nr_cfqq < 1); | 891 | BUG_ON(cfqg->nr_cfqq < 1); |
890 | cfqg->nr_cfqq--; | 892 | cfqg->nr_cfqq--; |
891 | 893 | ||
892 | /* If there are other cfq queues under this group, don't delete it */ | 894 | /* If there are other cfq queues under this group, don't delete it */ |
893 | if (cfqg->nr_cfqq) | 895 | if (cfqg->nr_cfqq) |
894 | return; | 896 | return; |
895 | 897 | ||
896 | cfq_log_cfqg(cfqd, cfqg, "del_from_rr group"); | 898 | cfq_log_cfqg(cfqd, cfqg, "del_from_rr group"); |
897 | cfqg->on_st = false; | 899 | cfqg->on_st = false; |
898 | st->total_weight -= cfqg->weight; | 900 | st->total_weight -= cfqg->weight; |
899 | if (!RB_EMPTY_NODE(&cfqg->rb_node)) | 901 | if (!RB_EMPTY_NODE(&cfqg->rb_node)) |
900 | cfq_rb_erase(&cfqg->rb_node, st); | 902 | cfq_rb_erase(&cfqg->rb_node, st); |
901 | cfqg->saved_workload_slice = 0; | 903 | cfqg->saved_workload_slice = 0; |
902 | cfq_blkiocg_update_dequeue_stats(&cfqg->blkg, 1); | 904 | cfq_blkiocg_update_dequeue_stats(&cfqg->blkg, 1); |
903 | } | 905 | } |
904 | 906 | ||
905 | static inline unsigned int cfq_cfqq_slice_usage(struct cfq_queue *cfqq) | 907 | static inline unsigned int cfq_cfqq_slice_usage(struct cfq_queue *cfqq) |
906 | { | 908 | { |
907 | unsigned int slice_used; | 909 | unsigned int slice_used; |
908 | 910 | ||
909 | /* | 911 | /* |
910 | * Queue got expired before even a single request completed or | 912 | * Queue got expired before even a single request completed or |
911 | * got expired immediately after first request completion. | 913 | * got expired immediately after first request completion. |
912 | */ | 914 | */ |
913 | if (!cfqq->slice_start || cfqq->slice_start == jiffies) { | 915 | if (!cfqq->slice_start || cfqq->slice_start == jiffies) { |
914 | /* | 916 | /* |
915 | * Also charge the seek time incurred to the group, otherwise | 917 | * Also charge the seek time incurred to the group, otherwise |
916 | * if there are mutiple queues in the group, each can dispatch | 918 | * if there are mutiple queues in the group, each can dispatch |
917 | * a single request on seeky media and cause lots of seek time | 919 | * a single request on seeky media and cause lots of seek time |
918 | * and group will never know it. | 920 | * and group will never know it. |
919 | */ | 921 | */ |
920 | slice_used = max_t(unsigned, (jiffies - cfqq->dispatch_start), | 922 | slice_used = max_t(unsigned, (jiffies - cfqq->dispatch_start), |
921 | 1); | 923 | 1); |
922 | } else { | 924 | } else { |
923 | slice_used = jiffies - cfqq->slice_start; | 925 | slice_used = jiffies - cfqq->slice_start; |
924 | if (slice_used > cfqq->allocated_slice) | 926 | if (slice_used > cfqq->allocated_slice) |
925 | slice_used = cfqq->allocated_slice; | 927 | slice_used = cfqq->allocated_slice; |
926 | } | 928 | } |
927 | 929 | ||
928 | return slice_used; | 930 | return slice_used; |
929 | } | 931 | } |
930 | 932 | ||
931 | static void cfq_group_served(struct cfq_data *cfqd, struct cfq_group *cfqg, | 933 | static void cfq_group_served(struct cfq_data *cfqd, struct cfq_group *cfqg, |
932 | struct cfq_queue *cfqq) | 934 | struct cfq_queue *cfqq) |
933 | { | 935 | { |
934 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | 936 | struct cfq_rb_root *st = &cfqd->grp_service_tree; |
935 | unsigned int used_sl, charge; | 937 | unsigned int used_sl, charge; |
936 | int nr_sync = cfqg->nr_cfqq - cfqg_busy_async_queues(cfqd, cfqg) | 938 | int nr_sync = cfqg->nr_cfqq - cfqg_busy_async_queues(cfqd, cfqg) |
937 | - cfqg->service_tree_idle.count; | 939 | - cfqg->service_tree_idle.count; |
938 | 940 | ||
939 | BUG_ON(nr_sync < 0); | 941 | BUG_ON(nr_sync < 0); |
940 | used_sl = charge = cfq_cfqq_slice_usage(cfqq); | 942 | used_sl = charge = cfq_cfqq_slice_usage(cfqq); |
941 | 943 | ||
942 | if (iops_mode(cfqd)) | 944 | if (iops_mode(cfqd)) |
943 | charge = cfqq->slice_dispatch; | 945 | charge = cfqq->slice_dispatch; |
944 | else if (!cfq_cfqq_sync(cfqq) && !nr_sync) | 946 | else if (!cfq_cfqq_sync(cfqq) && !nr_sync) |
945 | charge = cfqq->allocated_slice; | 947 | charge = cfqq->allocated_slice; |
946 | 948 | ||
947 | /* Can't update vdisktime while group is on service tree */ | 949 | /* Can't update vdisktime while group is on service tree */ |
948 | cfq_rb_erase(&cfqg->rb_node, st); | 950 | cfq_rb_erase(&cfqg->rb_node, st); |
949 | cfqg->vdisktime += cfq_scale_slice(charge, cfqg); | 951 | cfqg->vdisktime += cfq_scale_slice(charge, cfqg); |
950 | __cfq_group_service_tree_add(st, cfqg); | 952 | __cfq_group_service_tree_add(st, cfqg); |
951 | 953 | ||
952 | /* This group is being expired. Save the context */ | 954 | /* This group is being expired. Save the context */ |
953 | if (time_after(cfqd->workload_expires, jiffies)) { | 955 | if (time_after(cfqd->workload_expires, jiffies)) { |
954 | cfqg->saved_workload_slice = cfqd->workload_expires | 956 | cfqg->saved_workload_slice = cfqd->workload_expires |
955 | - jiffies; | 957 | - jiffies; |
956 | cfqg->saved_workload = cfqd->serving_type; | 958 | cfqg->saved_workload = cfqd->serving_type; |
957 | cfqg->saved_serving_prio = cfqd->serving_prio; | 959 | cfqg->saved_serving_prio = cfqd->serving_prio; |
958 | } else | 960 | } else |
959 | cfqg->saved_workload_slice = 0; | 961 | cfqg->saved_workload_slice = 0; |
960 | 962 | ||
961 | cfq_log_cfqg(cfqd, cfqg, "served: vt=%llu min_vt=%llu", cfqg->vdisktime, | 963 | cfq_log_cfqg(cfqd, cfqg, "served: vt=%llu min_vt=%llu", cfqg->vdisktime, |
962 | st->min_vdisktime); | 964 | st->min_vdisktime); |
963 | cfq_log_cfqq(cfqq->cfqd, cfqq, "sl_used=%u disp=%u charge=%u iops=%u", | 965 | cfq_log_cfqq(cfqq->cfqd, cfqq, "sl_used=%u disp=%u charge=%u iops=%u" |
964 | used_sl, cfqq->slice_dispatch, charge, iops_mode(cfqd)); | 966 | " sect=%u", used_sl, cfqq->slice_dispatch, charge, |
967 | iops_mode(cfqd), cfqq->nr_sectors); | ||
965 | cfq_blkiocg_update_timeslice_used(&cfqg->blkg, used_sl); | 968 | cfq_blkiocg_update_timeslice_used(&cfqg->blkg, used_sl); |
966 | cfq_blkiocg_set_start_empty_time(&cfqg->blkg); | 969 | cfq_blkiocg_set_start_empty_time(&cfqg->blkg); |
967 | } | 970 | } |
968 | 971 | ||
969 | #ifdef CONFIG_CFQ_GROUP_IOSCHED | 972 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
970 | static inline struct cfq_group *cfqg_of_blkg(struct blkio_group *blkg) | 973 | static inline struct cfq_group *cfqg_of_blkg(struct blkio_group *blkg) |
971 | { | 974 | { |
972 | if (blkg) | 975 | if (blkg) |
973 | return container_of(blkg, struct cfq_group, blkg); | 976 | return container_of(blkg, struct cfq_group, blkg); |
974 | return NULL; | 977 | return NULL; |
975 | } | 978 | } |
976 | 979 | ||
977 | void | 980 | void |
978 | cfq_update_blkio_group_weight(struct blkio_group *blkg, unsigned int weight) | 981 | cfq_update_blkio_group_weight(struct blkio_group *blkg, unsigned int weight) |
979 | { | 982 | { |
980 | cfqg_of_blkg(blkg)->weight = weight; | 983 | cfqg_of_blkg(blkg)->weight = weight; |
981 | } | 984 | } |
982 | 985 | ||
983 | static struct cfq_group * | 986 | static struct cfq_group * |
984 | cfq_find_alloc_cfqg(struct cfq_data *cfqd, struct cgroup *cgroup, int create) | 987 | cfq_find_alloc_cfqg(struct cfq_data *cfqd, struct cgroup *cgroup, int create) |
985 | { | 988 | { |
986 | struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgroup); | 989 | struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgroup); |
987 | struct cfq_group *cfqg = NULL; | 990 | struct cfq_group *cfqg = NULL; |
988 | void *key = cfqd; | 991 | void *key = cfqd; |
989 | int i, j; | 992 | int i, j; |
990 | struct cfq_rb_root *st; | 993 | struct cfq_rb_root *st; |
991 | struct backing_dev_info *bdi = &cfqd->queue->backing_dev_info; | 994 | struct backing_dev_info *bdi = &cfqd->queue->backing_dev_info; |
992 | unsigned int major, minor; | 995 | unsigned int major, minor; |
993 | 996 | ||
994 | cfqg = cfqg_of_blkg(blkiocg_lookup_group(blkcg, key)); | 997 | cfqg = cfqg_of_blkg(blkiocg_lookup_group(blkcg, key)); |
995 | if (cfqg && !cfqg->blkg.dev && bdi->dev && dev_name(bdi->dev)) { | 998 | if (cfqg && !cfqg->blkg.dev && bdi->dev && dev_name(bdi->dev)) { |
996 | sscanf(dev_name(bdi->dev), "%u:%u", &major, &minor); | 999 | sscanf(dev_name(bdi->dev), "%u:%u", &major, &minor); |
997 | cfqg->blkg.dev = MKDEV(major, minor); | 1000 | cfqg->blkg.dev = MKDEV(major, minor); |
998 | goto done; | 1001 | goto done; |
999 | } | 1002 | } |
1000 | if (cfqg || !create) | 1003 | if (cfqg || !create) |
1001 | goto done; | 1004 | goto done; |
1002 | 1005 | ||
1003 | cfqg = kzalloc_node(sizeof(*cfqg), GFP_ATOMIC, cfqd->queue->node); | 1006 | cfqg = kzalloc_node(sizeof(*cfqg), GFP_ATOMIC, cfqd->queue->node); |
1004 | if (!cfqg) | 1007 | if (!cfqg) |
1005 | goto done; | 1008 | goto done; |
1006 | 1009 | ||
1007 | for_each_cfqg_st(cfqg, i, j, st) | 1010 | for_each_cfqg_st(cfqg, i, j, st) |
1008 | *st = CFQ_RB_ROOT; | 1011 | *st = CFQ_RB_ROOT; |
1009 | RB_CLEAR_NODE(&cfqg->rb_node); | 1012 | RB_CLEAR_NODE(&cfqg->rb_node); |
1010 | 1013 | ||
1011 | /* | 1014 | /* |
1012 | * Take the initial reference that will be released on destroy | 1015 | * Take the initial reference that will be released on destroy |
1013 | * This can be thought of a joint reference by cgroup and | 1016 | * This can be thought of a joint reference by cgroup and |
1014 | * elevator which will be dropped by either elevator exit | 1017 | * elevator which will be dropped by either elevator exit |
1015 | * or cgroup deletion path depending on who is exiting first. | 1018 | * or cgroup deletion path depending on who is exiting first. |
1016 | */ | 1019 | */ |
1017 | atomic_set(&cfqg->ref, 1); | 1020 | atomic_set(&cfqg->ref, 1); |
1018 | 1021 | ||
1019 | /* Add group onto cgroup list */ | 1022 | /* Add group onto cgroup list */ |
1020 | sscanf(dev_name(bdi->dev), "%u:%u", &major, &minor); | 1023 | sscanf(dev_name(bdi->dev), "%u:%u", &major, &minor); |
1021 | cfq_blkiocg_add_blkio_group(blkcg, &cfqg->blkg, (void *)cfqd, | 1024 | cfq_blkiocg_add_blkio_group(blkcg, &cfqg->blkg, (void *)cfqd, |
1022 | MKDEV(major, minor)); | 1025 | MKDEV(major, minor)); |
1023 | cfqg->weight = blkcg_get_weight(blkcg, cfqg->blkg.dev); | 1026 | cfqg->weight = blkcg_get_weight(blkcg, cfqg->blkg.dev); |
1024 | 1027 | ||
1025 | /* Add group on cfqd list */ | 1028 | /* Add group on cfqd list */ |
1026 | hlist_add_head(&cfqg->cfqd_node, &cfqd->cfqg_list); | 1029 | hlist_add_head(&cfqg->cfqd_node, &cfqd->cfqg_list); |
1027 | 1030 | ||
1028 | done: | 1031 | done: |
1029 | return cfqg; | 1032 | return cfqg; |
1030 | } | 1033 | } |
1031 | 1034 | ||
1032 | /* | 1035 | /* |
1033 | * Search for the cfq group current task belongs to. If create = 1, then also | 1036 | * Search for the cfq group current task belongs to. If create = 1, then also |
1034 | * create the cfq group if it does not exist. request_queue lock must be held. | 1037 | * create the cfq group if it does not exist. request_queue lock must be held. |
1035 | */ | 1038 | */ |
1036 | static struct cfq_group *cfq_get_cfqg(struct cfq_data *cfqd, int create) | 1039 | static struct cfq_group *cfq_get_cfqg(struct cfq_data *cfqd, int create) |
1037 | { | 1040 | { |
1038 | struct cgroup *cgroup; | 1041 | struct cgroup *cgroup; |
1039 | struct cfq_group *cfqg = NULL; | 1042 | struct cfq_group *cfqg = NULL; |
1040 | 1043 | ||
1041 | rcu_read_lock(); | 1044 | rcu_read_lock(); |
1042 | cgroup = task_cgroup(current, blkio_subsys_id); | 1045 | cgroup = task_cgroup(current, blkio_subsys_id); |
1043 | cfqg = cfq_find_alloc_cfqg(cfqd, cgroup, create); | 1046 | cfqg = cfq_find_alloc_cfqg(cfqd, cgroup, create); |
1044 | if (!cfqg && create) | 1047 | if (!cfqg && create) |
1045 | cfqg = &cfqd->root_group; | 1048 | cfqg = &cfqd->root_group; |
1046 | rcu_read_unlock(); | 1049 | rcu_read_unlock(); |
1047 | return cfqg; | 1050 | return cfqg; |
1048 | } | 1051 | } |
1049 | 1052 | ||
1050 | static inline struct cfq_group *cfq_ref_get_cfqg(struct cfq_group *cfqg) | 1053 | static inline struct cfq_group *cfq_ref_get_cfqg(struct cfq_group *cfqg) |
1051 | { | 1054 | { |
1052 | atomic_inc(&cfqg->ref); | 1055 | atomic_inc(&cfqg->ref); |
1053 | return cfqg; | 1056 | return cfqg; |
1054 | } | 1057 | } |
1055 | 1058 | ||
1056 | static void cfq_link_cfqq_cfqg(struct cfq_queue *cfqq, struct cfq_group *cfqg) | 1059 | static void cfq_link_cfqq_cfqg(struct cfq_queue *cfqq, struct cfq_group *cfqg) |
1057 | { | 1060 | { |
1058 | /* Currently, all async queues are mapped to root group */ | 1061 | /* Currently, all async queues are mapped to root group */ |
1059 | if (!cfq_cfqq_sync(cfqq)) | 1062 | if (!cfq_cfqq_sync(cfqq)) |
1060 | cfqg = &cfqq->cfqd->root_group; | 1063 | cfqg = &cfqq->cfqd->root_group; |
1061 | 1064 | ||
1062 | cfqq->cfqg = cfqg; | 1065 | cfqq->cfqg = cfqg; |
1063 | /* cfqq reference on cfqg */ | 1066 | /* cfqq reference on cfqg */ |
1064 | atomic_inc(&cfqq->cfqg->ref); | 1067 | atomic_inc(&cfqq->cfqg->ref); |
1065 | } | 1068 | } |
1066 | 1069 | ||
1067 | static void cfq_put_cfqg(struct cfq_group *cfqg) | 1070 | static void cfq_put_cfqg(struct cfq_group *cfqg) |
1068 | { | 1071 | { |
1069 | struct cfq_rb_root *st; | 1072 | struct cfq_rb_root *st; |
1070 | int i, j; | 1073 | int i, j; |
1071 | 1074 | ||
1072 | BUG_ON(atomic_read(&cfqg->ref) <= 0); | 1075 | BUG_ON(atomic_read(&cfqg->ref) <= 0); |
1073 | if (!atomic_dec_and_test(&cfqg->ref)) | 1076 | if (!atomic_dec_and_test(&cfqg->ref)) |
1074 | return; | 1077 | return; |
1075 | for_each_cfqg_st(cfqg, i, j, st) | 1078 | for_each_cfqg_st(cfqg, i, j, st) |
1076 | BUG_ON(!RB_EMPTY_ROOT(&st->rb) || st->active != NULL); | 1079 | BUG_ON(!RB_EMPTY_ROOT(&st->rb) || st->active != NULL); |
1077 | kfree(cfqg); | 1080 | kfree(cfqg); |
1078 | } | 1081 | } |
1079 | 1082 | ||
1080 | static void cfq_destroy_cfqg(struct cfq_data *cfqd, struct cfq_group *cfqg) | 1083 | static void cfq_destroy_cfqg(struct cfq_data *cfqd, struct cfq_group *cfqg) |
1081 | { | 1084 | { |
1082 | /* Something wrong if we are trying to remove same group twice */ | 1085 | /* Something wrong if we are trying to remove same group twice */ |
1083 | BUG_ON(hlist_unhashed(&cfqg->cfqd_node)); | 1086 | BUG_ON(hlist_unhashed(&cfqg->cfqd_node)); |
1084 | 1087 | ||
1085 | hlist_del_init(&cfqg->cfqd_node); | 1088 | hlist_del_init(&cfqg->cfqd_node); |
1086 | 1089 | ||
1087 | /* | 1090 | /* |
1088 | * Put the reference taken at the time of creation so that when all | 1091 | * Put the reference taken at the time of creation so that when all |
1089 | * queues are gone, group can be destroyed. | 1092 | * queues are gone, group can be destroyed. |
1090 | */ | 1093 | */ |
1091 | cfq_put_cfqg(cfqg); | 1094 | cfq_put_cfqg(cfqg); |
1092 | } | 1095 | } |
1093 | 1096 | ||
1094 | static void cfq_release_cfq_groups(struct cfq_data *cfqd) | 1097 | static void cfq_release_cfq_groups(struct cfq_data *cfqd) |
1095 | { | 1098 | { |
1096 | struct hlist_node *pos, *n; | 1099 | struct hlist_node *pos, *n; |
1097 | struct cfq_group *cfqg; | 1100 | struct cfq_group *cfqg; |
1098 | 1101 | ||
1099 | hlist_for_each_entry_safe(cfqg, pos, n, &cfqd->cfqg_list, cfqd_node) { | 1102 | hlist_for_each_entry_safe(cfqg, pos, n, &cfqd->cfqg_list, cfqd_node) { |
1100 | /* | 1103 | /* |
1101 | * If cgroup removal path got to blk_group first and removed | 1104 | * If cgroup removal path got to blk_group first and removed |
1102 | * it from cgroup list, then it will take care of destroying | 1105 | * it from cgroup list, then it will take care of destroying |
1103 | * cfqg also. | 1106 | * cfqg also. |
1104 | */ | 1107 | */ |
1105 | if (!cfq_blkiocg_del_blkio_group(&cfqg->blkg)) | 1108 | if (!cfq_blkiocg_del_blkio_group(&cfqg->blkg)) |
1106 | cfq_destroy_cfqg(cfqd, cfqg); | 1109 | cfq_destroy_cfqg(cfqd, cfqg); |
1107 | } | 1110 | } |
1108 | } | 1111 | } |
1109 | 1112 | ||
1110 | /* | 1113 | /* |
1111 | * Blk cgroup controller notification saying that blkio_group object is being | 1114 | * Blk cgroup controller notification saying that blkio_group object is being |
1112 | * delinked as associated cgroup object is going away. That also means that | 1115 | * delinked as associated cgroup object is going away. That also means that |
1113 | * no new IO will come in this group. So get rid of this group as soon as | 1116 | * no new IO will come in this group. So get rid of this group as soon as |
1114 | * any pending IO in the group is finished. | 1117 | * any pending IO in the group is finished. |
1115 | * | 1118 | * |
1116 | * This function is called under rcu_read_lock(). key is the rcu protected | 1119 | * This function is called under rcu_read_lock(). key is the rcu protected |
1117 | * pointer. That means "key" is a valid cfq_data pointer as long as we are rcu | 1120 | * pointer. That means "key" is a valid cfq_data pointer as long as we are rcu |
1118 | * read lock. | 1121 | * read lock. |
1119 | * | 1122 | * |
1120 | * "key" was fetched from blkio_group under blkio_cgroup->lock. That means | 1123 | * "key" was fetched from blkio_group under blkio_cgroup->lock. That means |
1121 | * it should not be NULL as even if elevator was exiting, cgroup deltion | 1124 | * it should not be NULL as even if elevator was exiting, cgroup deltion |
1122 | * path got to it first. | 1125 | * path got to it first. |
1123 | */ | 1126 | */ |
1124 | void cfq_unlink_blkio_group(void *key, struct blkio_group *blkg) | 1127 | void cfq_unlink_blkio_group(void *key, struct blkio_group *blkg) |
1125 | { | 1128 | { |
1126 | unsigned long flags; | 1129 | unsigned long flags; |
1127 | struct cfq_data *cfqd = key; | 1130 | struct cfq_data *cfqd = key; |
1128 | 1131 | ||
1129 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); | 1132 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); |
1130 | cfq_destroy_cfqg(cfqd, cfqg_of_blkg(blkg)); | 1133 | cfq_destroy_cfqg(cfqd, cfqg_of_blkg(blkg)); |
1131 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); | 1134 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); |
1132 | } | 1135 | } |
1133 | 1136 | ||
1134 | #else /* GROUP_IOSCHED */ | 1137 | #else /* GROUP_IOSCHED */ |
1135 | static struct cfq_group *cfq_get_cfqg(struct cfq_data *cfqd, int create) | 1138 | static struct cfq_group *cfq_get_cfqg(struct cfq_data *cfqd, int create) |
1136 | { | 1139 | { |
1137 | return &cfqd->root_group; | 1140 | return &cfqd->root_group; |
1138 | } | 1141 | } |
1139 | 1142 | ||
1140 | static inline struct cfq_group *cfq_ref_get_cfqg(struct cfq_group *cfqg) | 1143 | static inline struct cfq_group *cfq_ref_get_cfqg(struct cfq_group *cfqg) |
1141 | { | 1144 | { |
1142 | return cfqg; | 1145 | return cfqg; |
1143 | } | 1146 | } |
1144 | 1147 | ||
1145 | static inline void | 1148 | static inline void |
1146 | cfq_link_cfqq_cfqg(struct cfq_queue *cfqq, struct cfq_group *cfqg) { | 1149 | cfq_link_cfqq_cfqg(struct cfq_queue *cfqq, struct cfq_group *cfqg) { |
1147 | cfqq->cfqg = cfqg; | 1150 | cfqq->cfqg = cfqg; |
1148 | } | 1151 | } |
1149 | 1152 | ||
1150 | static void cfq_release_cfq_groups(struct cfq_data *cfqd) {} | 1153 | static void cfq_release_cfq_groups(struct cfq_data *cfqd) {} |
1151 | static inline void cfq_put_cfqg(struct cfq_group *cfqg) {} | 1154 | static inline void cfq_put_cfqg(struct cfq_group *cfqg) {} |
1152 | 1155 | ||
1153 | #endif /* GROUP_IOSCHED */ | 1156 | #endif /* GROUP_IOSCHED */ |
1154 | 1157 | ||
1155 | /* | 1158 | /* |
1156 | * The cfqd->service_trees holds all pending cfq_queue's that have | 1159 | * The cfqd->service_trees holds all pending cfq_queue's that have |
1157 | * requests waiting to be processed. It is sorted in the order that | 1160 | * requests waiting to be processed. It is sorted in the order that |
1158 | * we will service the queues. | 1161 | * we will service the queues. |
1159 | */ | 1162 | */ |
1160 | static void cfq_service_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq, | 1163 | static void cfq_service_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
1161 | bool add_front) | 1164 | bool add_front) |
1162 | { | 1165 | { |
1163 | struct rb_node **p, *parent; | 1166 | struct rb_node **p, *parent; |
1164 | struct cfq_queue *__cfqq; | 1167 | struct cfq_queue *__cfqq; |
1165 | unsigned long rb_key; | 1168 | unsigned long rb_key; |
1166 | struct cfq_rb_root *service_tree; | 1169 | struct cfq_rb_root *service_tree; |
1167 | int left; | 1170 | int left; |
1168 | int new_cfqq = 1; | 1171 | int new_cfqq = 1; |
1169 | int group_changed = 0; | 1172 | int group_changed = 0; |
1170 | 1173 | ||
1171 | #ifdef CONFIG_CFQ_GROUP_IOSCHED | 1174 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
1172 | if (!cfqd->cfq_group_isolation | 1175 | if (!cfqd->cfq_group_isolation |
1173 | && cfqq_type(cfqq) == SYNC_NOIDLE_WORKLOAD | 1176 | && cfqq_type(cfqq) == SYNC_NOIDLE_WORKLOAD |
1174 | && cfqq->cfqg && cfqq->cfqg != &cfqd->root_group) { | 1177 | && cfqq->cfqg && cfqq->cfqg != &cfqd->root_group) { |
1175 | /* Move this cfq to root group */ | 1178 | /* Move this cfq to root group */ |
1176 | cfq_log_cfqq(cfqd, cfqq, "moving to root group"); | 1179 | cfq_log_cfqq(cfqd, cfqq, "moving to root group"); |
1177 | if (!RB_EMPTY_NODE(&cfqq->rb_node)) | 1180 | if (!RB_EMPTY_NODE(&cfqq->rb_node)) |
1178 | cfq_group_service_tree_del(cfqd, cfqq->cfqg); | 1181 | cfq_group_service_tree_del(cfqd, cfqq->cfqg); |
1179 | cfqq->orig_cfqg = cfqq->cfqg; | 1182 | cfqq->orig_cfqg = cfqq->cfqg; |
1180 | cfqq->cfqg = &cfqd->root_group; | 1183 | cfqq->cfqg = &cfqd->root_group; |
1181 | atomic_inc(&cfqd->root_group.ref); | 1184 | atomic_inc(&cfqd->root_group.ref); |
1182 | group_changed = 1; | 1185 | group_changed = 1; |
1183 | } else if (!cfqd->cfq_group_isolation | 1186 | } else if (!cfqd->cfq_group_isolation |
1184 | && cfqq_type(cfqq) == SYNC_WORKLOAD && cfqq->orig_cfqg) { | 1187 | && cfqq_type(cfqq) == SYNC_WORKLOAD && cfqq->orig_cfqg) { |
1185 | /* cfqq is sequential now needs to go to its original group */ | 1188 | /* cfqq is sequential now needs to go to its original group */ |
1186 | BUG_ON(cfqq->cfqg != &cfqd->root_group); | 1189 | BUG_ON(cfqq->cfqg != &cfqd->root_group); |
1187 | if (!RB_EMPTY_NODE(&cfqq->rb_node)) | 1190 | if (!RB_EMPTY_NODE(&cfqq->rb_node)) |
1188 | cfq_group_service_tree_del(cfqd, cfqq->cfqg); | 1191 | cfq_group_service_tree_del(cfqd, cfqq->cfqg); |
1189 | cfq_put_cfqg(cfqq->cfqg); | 1192 | cfq_put_cfqg(cfqq->cfqg); |
1190 | cfqq->cfqg = cfqq->orig_cfqg; | 1193 | cfqq->cfqg = cfqq->orig_cfqg; |
1191 | cfqq->orig_cfqg = NULL; | 1194 | cfqq->orig_cfqg = NULL; |
1192 | group_changed = 1; | 1195 | group_changed = 1; |
1193 | cfq_log_cfqq(cfqd, cfqq, "moved to origin group"); | 1196 | cfq_log_cfqq(cfqd, cfqq, "moved to origin group"); |
1194 | } | 1197 | } |
1195 | #endif | 1198 | #endif |
1196 | 1199 | ||
1197 | service_tree = service_tree_for(cfqq->cfqg, cfqq_prio(cfqq), | 1200 | service_tree = service_tree_for(cfqq->cfqg, cfqq_prio(cfqq), |
1198 | cfqq_type(cfqq)); | 1201 | cfqq_type(cfqq)); |
1199 | if (cfq_class_idle(cfqq)) { | 1202 | if (cfq_class_idle(cfqq)) { |
1200 | rb_key = CFQ_IDLE_DELAY; | 1203 | rb_key = CFQ_IDLE_DELAY; |
1201 | parent = rb_last(&service_tree->rb); | 1204 | parent = rb_last(&service_tree->rb); |
1202 | if (parent && parent != &cfqq->rb_node) { | 1205 | if (parent && parent != &cfqq->rb_node) { |
1203 | __cfqq = rb_entry(parent, struct cfq_queue, rb_node); | 1206 | __cfqq = rb_entry(parent, struct cfq_queue, rb_node); |
1204 | rb_key += __cfqq->rb_key; | 1207 | rb_key += __cfqq->rb_key; |
1205 | } else | 1208 | } else |
1206 | rb_key += jiffies; | 1209 | rb_key += jiffies; |
1207 | } else if (!add_front) { | 1210 | } else if (!add_front) { |
1208 | /* | 1211 | /* |
1209 | * Get our rb key offset. Subtract any residual slice | 1212 | * Get our rb key offset. Subtract any residual slice |
1210 | * value carried from last service. A negative resid | 1213 | * value carried from last service. A negative resid |
1211 | * count indicates slice overrun, and this should position | 1214 | * count indicates slice overrun, and this should position |
1212 | * the next service time further away in the tree. | 1215 | * the next service time further away in the tree. |
1213 | */ | 1216 | */ |
1214 | rb_key = cfq_slice_offset(cfqd, cfqq) + jiffies; | 1217 | rb_key = cfq_slice_offset(cfqd, cfqq) + jiffies; |
1215 | rb_key -= cfqq->slice_resid; | 1218 | rb_key -= cfqq->slice_resid; |
1216 | cfqq->slice_resid = 0; | 1219 | cfqq->slice_resid = 0; |
1217 | } else { | 1220 | } else { |
1218 | rb_key = -HZ; | 1221 | rb_key = -HZ; |
1219 | __cfqq = cfq_rb_first(service_tree); | 1222 | __cfqq = cfq_rb_first(service_tree); |
1220 | rb_key += __cfqq ? __cfqq->rb_key : jiffies; | 1223 | rb_key += __cfqq ? __cfqq->rb_key : jiffies; |
1221 | } | 1224 | } |
1222 | 1225 | ||
1223 | if (!RB_EMPTY_NODE(&cfqq->rb_node)) { | 1226 | if (!RB_EMPTY_NODE(&cfqq->rb_node)) { |
1224 | new_cfqq = 0; | 1227 | new_cfqq = 0; |
1225 | /* | 1228 | /* |
1226 | * same position, nothing more to do | 1229 | * same position, nothing more to do |
1227 | */ | 1230 | */ |
1228 | if (rb_key == cfqq->rb_key && | 1231 | if (rb_key == cfqq->rb_key && |
1229 | cfqq->service_tree == service_tree) | 1232 | cfqq->service_tree == service_tree) |
1230 | return; | 1233 | return; |
1231 | 1234 | ||
1232 | cfq_rb_erase(&cfqq->rb_node, cfqq->service_tree); | 1235 | cfq_rb_erase(&cfqq->rb_node, cfqq->service_tree); |
1233 | cfqq->service_tree = NULL; | 1236 | cfqq->service_tree = NULL; |
1234 | } | 1237 | } |
1235 | 1238 | ||
1236 | left = 1; | 1239 | left = 1; |
1237 | parent = NULL; | 1240 | parent = NULL; |
1238 | cfqq->service_tree = service_tree; | 1241 | cfqq->service_tree = service_tree; |
1239 | p = &service_tree->rb.rb_node; | 1242 | p = &service_tree->rb.rb_node; |
1240 | while (*p) { | 1243 | while (*p) { |
1241 | struct rb_node **n; | 1244 | struct rb_node **n; |
1242 | 1245 | ||
1243 | parent = *p; | 1246 | parent = *p; |
1244 | __cfqq = rb_entry(parent, struct cfq_queue, rb_node); | 1247 | __cfqq = rb_entry(parent, struct cfq_queue, rb_node); |
1245 | 1248 | ||
1246 | /* | 1249 | /* |
1247 | * sort by key, that represents service time. | 1250 | * sort by key, that represents service time. |
1248 | */ | 1251 | */ |
1249 | if (time_before(rb_key, __cfqq->rb_key)) | 1252 | if (time_before(rb_key, __cfqq->rb_key)) |
1250 | n = &(*p)->rb_left; | 1253 | n = &(*p)->rb_left; |
1251 | else { | 1254 | else { |
1252 | n = &(*p)->rb_right; | 1255 | n = &(*p)->rb_right; |
1253 | left = 0; | 1256 | left = 0; |
1254 | } | 1257 | } |
1255 | 1258 | ||
1256 | p = n; | 1259 | p = n; |
1257 | } | 1260 | } |
1258 | 1261 | ||
1259 | if (left) | 1262 | if (left) |
1260 | service_tree->left = &cfqq->rb_node; | 1263 | service_tree->left = &cfqq->rb_node; |
1261 | 1264 | ||
1262 | cfqq->rb_key = rb_key; | 1265 | cfqq->rb_key = rb_key; |
1263 | rb_link_node(&cfqq->rb_node, parent, p); | 1266 | rb_link_node(&cfqq->rb_node, parent, p); |
1264 | rb_insert_color(&cfqq->rb_node, &service_tree->rb); | 1267 | rb_insert_color(&cfqq->rb_node, &service_tree->rb); |
1265 | service_tree->count++; | 1268 | service_tree->count++; |
1266 | if ((add_front || !new_cfqq) && !group_changed) | 1269 | if ((add_front || !new_cfqq) && !group_changed) |
1267 | return; | 1270 | return; |
1268 | cfq_group_service_tree_add(cfqd, cfqq->cfqg); | 1271 | cfq_group_service_tree_add(cfqd, cfqq->cfqg); |
1269 | } | 1272 | } |
1270 | 1273 | ||
1271 | static struct cfq_queue * | 1274 | static struct cfq_queue * |
1272 | cfq_prio_tree_lookup(struct cfq_data *cfqd, struct rb_root *root, | 1275 | cfq_prio_tree_lookup(struct cfq_data *cfqd, struct rb_root *root, |
1273 | sector_t sector, struct rb_node **ret_parent, | 1276 | sector_t sector, struct rb_node **ret_parent, |
1274 | struct rb_node ***rb_link) | 1277 | struct rb_node ***rb_link) |
1275 | { | 1278 | { |
1276 | struct rb_node **p, *parent; | 1279 | struct rb_node **p, *parent; |
1277 | struct cfq_queue *cfqq = NULL; | 1280 | struct cfq_queue *cfqq = NULL; |
1278 | 1281 | ||
1279 | parent = NULL; | 1282 | parent = NULL; |
1280 | p = &root->rb_node; | 1283 | p = &root->rb_node; |
1281 | while (*p) { | 1284 | while (*p) { |
1282 | struct rb_node **n; | 1285 | struct rb_node **n; |
1283 | 1286 | ||
1284 | parent = *p; | 1287 | parent = *p; |
1285 | cfqq = rb_entry(parent, struct cfq_queue, p_node); | 1288 | cfqq = rb_entry(parent, struct cfq_queue, p_node); |
1286 | 1289 | ||
1287 | /* | 1290 | /* |
1288 | * Sort strictly based on sector. Smallest to the left, | 1291 | * Sort strictly based on sector. Smallest to the left, |
1289 | * largest to the right. | 1292 | * largest to the right. |
1290 | */ | 1293 | */ |
1291 | if (sector > blk_rq_pos(cfqq->next_rq)) | 1294 | if (sector > blk_rq_pos(cfqq->next_rq)) |
1292 | n = &(*p)->rb_right; | 1295 | n = &(*p)->rb_right; |
1293 | else if (sector < blk_rq_pos(cfqq->next_rq)) | 1296 | else if (sector < blk_rq_pos(cfqq->next_rq)) |
1294 | n = &(*p)->rb_left; | 1297 | n = &(*p)->rb_left; |
1295 | else | 1298 | else |
1296 | break; | 1299 | break; |
1297 | p = n; | 1300 | p = n; |
1298 | cfqq = NULL; | 1301 | cfqq = NULL; |
1299 | } | 1302 | } |
1300 | 1303 | ||
1301 | *ret_parent = parent; | 1304 | *ret_parent = parent; |
1302 | if (rb_link) | 1305 | if (rb_link) |
1303 | *rb_link = p; | 1306 | *rb_link = p; |
1304 | return cfqq; | 1307 | return cfqq; |
1305 | } | 1308 | } |
1306 | 1309 | ||
1307 | static void cfq_prio_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq) | 1310 | static void cfq_prio_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1308 | { | 1311 | { |
1309 | struct rb_node **p, *parent; | 1312 | struct rb_node **p, *parent; |
1310 | struct cfq_queue *__cfqq; | 1313 | struct cfq_queue *__cfqq; |
1311 | 1314 | ||
1312 | if (cfqq->p_root) { | 1315 | if (cfqq->p_root) { |
1313 | rb_erase(&cfqq->p_node, cfqq->p_root); | 1316 | rb_erase(&cfqq->p_node, cfqq->p_root); |
1314 | cfqq->p_root = NULL; | 1317 | cfqq->p_root = NULL; |
1315 | } | 1318 | } |
1316 | 1319 | ||
1317 | if (cfq_class_idle(cfqq)) | 1320 | if (cfq_class_idle(cfqq)) |
1318 | return; | 1321 | return; |
1319 | if (!cfqq->next_rq) | 1322 | if (!cfqq->next_rq) |
1320 | return; | 1323 | return; |
1321 | 1324 | ||
1322 | cfqq->p_root = &cfqd->prio_trees[cfqq->org_ioprio]; | 1325 | cfqq->p_root = &cfqd->prio_trees[cfqq->org_ioprio]; |
1323 | __cfqq = cfq_prio_tree_lookup(cfqd, cfqq->p_root, | 1326 | __cfqq = cfq_prio_tree_lookup(cfqd, cfqq->p_root, |
1324 | blk_rq_pos(cfqq->next_rq), &parent, &p); | 1327 | blk_rq_pos(cfqq->next_rq), &parent, &p); |
1325 | if (!__cfqq) { | 1328 | if (!__cfqq) { |
1326 | rb_link_node(&cfqq->p_node, parent, p); | 1329 | rb_link_node(&cfqq->p_node, parent, p); |
1327 | rb_insert_color(&cfqq->p_node, cfqq->p_root); | 1330 | rb_insert_color(&cfqq->p_node, cfqq->p_root); |
1328 | } else | 1331 | } else |
1329 | cfqq->p_root = NULL; | 1332 | cfqq->p_root = NULL; |
1330 | } | 1333 | } |
1331 | 1334 | ||
1332 | /* | 1335 | /* |
1333 | * Update cfqq's position in the service tree. | 1336 | * Update cfqq's position in the service tree. |
1334 | */ | 1337 | */ |
1335 | static void cfq_resort_rr_list(struct cfq_data *cfqd, struct cfq_queue *cfqq) | 1338 | static void cfq_resort_rr_list(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1336 | { | 1339 | { |
1337 | /* | 1340 | /* |
1338 | * Resorting requires the cfqq to be on the RR list already. | 1341 | * Resorting requires the cfqq to be on the RR list already. |
1339 | */ | 1342 | */ |
1340 | if (cfq_cfqq_on_rr(cfqq)) { | 1343 | if (cfq_cfqq_on_rr(cfqq)) { |
1341 | cfq_service_tree_add(cfqd, cfqq, 0); | 1344 | cfq_service_tree_add(cfqd, cfqq, 0); |
1342 | cfq_prio_tree_add(cfqd, cfqq); | 1345 | cfq_prio_tree_add(cfqd, cfqq); |
1343 | } | 1346 | } |
1344 | } | 1347 | } |
1345 | 1348 | ||
1346 | /* | 1349 | /* |
1347 | * add to busy list of queues for service, trying to be fair in ordering | 1350 | * add to busy list of queues for service, trying to be fair in ordering |
1348 | * the pending list according to last request service | 1351 | * the pending list according to last request service |
1349 | */ | 1352 | */ |
1350 | static void cfq_add_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) | 1353 | static void cfq_add_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1351 | { | 1354 | { |
1352 | cfq_log_cfqq(cfqd, cfqq, "add_to_rr"); | 1355 | cfq_log_cfqq(cfqd, cfqq, "add_to_rr"); |
1353 | BUG_ON(cfq_cfqq_on_rr(cfqq)); | 1356 | BUG_ON(cfq_cfqq_on_rr(cfqq)); |
1354 | cfq_mark_cfqq_on_rr(cfqq); | 1357 | cfq_mark_cfqq_on_rr(cfqq); |
1355 | cfqd->busy_queues++; | 1358 | cfqd->busy_queues++; |
1356 | 1359 | ||
1357 | cfq_resort_rr_list(cfqd, cfqq); | 1360 | cfq_resort_rr_list(cfqd, cfqq); |
1358 | } | 1361 | } |
1359 | 1362 | ||
1360 | /* | 1363 | /* |
1361 | * Called when the cfqq no longer has requests pending, remove it from | 1364 | * Called when the cfqq no longer has requests pending, remove it from |
1362 | * the service tree. | 1365 | * the service tree. |
1363 | */ | 1366 | */ |
1364 | static void cfq_del_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) | 1367 | static void cfq_del_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1365 | { | 1368 | { |
1366 | cfq_log_cfqq(cfqd, cfqq, "del_from_rr"); | 1369 | cfq_log_cfqq(cfqd, cfqq, "del_from_rr"); |
1367 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); | 1370 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); |
1368 | cfq_clear_cfqq_on_rr(cfqq); | 1371 | cfq_clear_cfqq_on_rr(cfqq); |
1369 | 1372 | ||
1370 | if (!RB_EMPTY_NODE(&cfqq->rb_node)) { | 1373 | if (!RB_EMPTY_NODE(&cfqq->rb_node)) { |
1371 | cfq_rb_erase(&cfqq->rb_node, cfqq->service_tree); | 1374 | cfq_rb_erase(&cfqq->rb_node, cfqq->service_tree); |
1372 | cfqq->service_tree = NULL; | 1375 | cfqq->service_tree = NULL; |
1373 | } | 1376 | } |
1374 | if (cfqq->p_root) { | 1377 | if (cfqq->p_root) { |
1375 | rb_erase(&cfqq->p_node, cfqq->p_root); | 1378 | rb_erase(&cfqq->p_node, cfqq->p_root); |
1376 | cfqq->p_root = NULL; | 1379 | cfqq->p_root = NULL; |
1377 | } | 1380 | } |
1378 | 1381 | ||
1379 | cfq_group_service_tree_del(cfqd, cfqq->cfqg); | 1382 | cfq_group_service_tree_del(cfqd, cfqq->cfqg); |
1380 | BUG_ON(!cfqd->busy_queues); | 1383 | BUG_ON(!cfqd->busy_queues); |
1381 | cfqd->busy_queues--; | 1384 | cfqd->busy_queues--; |
1382 | } | 1385 | } |
1383 | 1386 | ||
1384 | /* | 1387 | /* |
1385 | * rb tree support functions | 1388 | * rb tree support functions |
1386 | */ | 1389 | */ |
1387 | static void cfq_del_rq_rb(struct request *rq) | 1390 | static void cfq_del_rq_rb(struct request *rq) |
1388 | { | 1391 | { |
1389 | struct cfq_queue *cfqq = RQ_CFQQ(rq); | 1392 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
1390 | const int sync = rq_is_sync(rq); | 1393 | const int sync = rq_is_sync(rq); |
1391 | 1394 | ||
1392 | BUG_ON(!cfqq->queued[sync]); | 1395 | BUG_ON(!cfqq->queued[sync]); |
1393 | cfqq->queued[sync]--; | 1396 | cfqq->queued[sync]--; |
1394 | 1397 | ||
1395 | elv_rb_del(&cfqq->sort_list, rq); | 1398 | elv_rb_del(&cfqq->sort_list, rq); |
1396 | 1399 | ||
1397 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list)) { | 1400 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list)) { |
1398 | /* | 1401 | /* |
1399 | * Queue will be deleted from service tree when we actually | 1402 | * Queue will be deleted from service tree when we actually |
1400 | * expire it later. Right now just remove it from prio tree | 1403 | * expire it later. Right now just remove it from prio tree |
1401 | * as it is empty. | 1404 | * as it is empty. |
1402 | */ | 1405 | */ |
1403 | if (cfqq->p_root) { | 1406 | if (cfqq->p_root) { |
1404 | rb_erase(&cfqq->p_node, cfqq->p_root); | 1407 | rb_erase(&cfqq->p_node, cfqq->p_root); |
1405 | cfqq->p_root = NULL; | 1408 | cfqq->p_root = NULL; |
1406 | } | 1409 | } |
1407 | } | 1410 | } |
1408 | } | 1411 | } |
1409 | 1412 | ||
1410 | static void cfq_add_rq_rb(struct request *rq) | 1413 | static void cfq_add_rq_rb(struct request *rq) |
1411 | { | 1414 | { |
1412 | struct cfq_queue *cfqq = RQ_CFQQ(rq); | 1415 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
1413 | struct cfq_data *cfqd = cfqq->cfqd; | 1416 | struct cfq_data *cfqd = cfqq->cfqd; |
1414 | struct request *__alias, *prev; | 1417 | struct request *__alias, *prev; |
1415 | 1418 | ||
1416 | cfqq->queued[rq_is_sync(rq)]++; | 1419 | cfqq->queued[rq_is_sync(rq)]++; |
1417 | 1420 | ||
1418 | /* | 1421 | /* |
1419 | * looks a little odd, but the first insert might return an alias. | 1422 | * looks a little odd, but the first insert might return an alias. |
1420 | * if that happens, put the alias on the dispatch list | 1423 | * if that happens, put the alias on the dispatch list |
1421 | */ | 1424 | */ |
1422 | while ((__alias = elv_rb_add(&cfqq->sort_list, rq)) != NULL) | 1425 | while ((__alias = elv_rb_add(&cfqq->sort_list, rq)) != NULL) |
1423 | cfq_dispatch_insert(cfqd->queue, __alias); | 1426 | cfq_dispatch_insert(cfqd->queue, __alias); |
1424 | 1427 | ||
1425 | if (!cfq_cfqq_on_rr(cfqq)) | 1428 | if (!cfq_cfqq_on_rr(cfqq)) |
1426 | cfq_add_cfqq_rr(cfqd, cfqq); | 1429 | cfq_add_cfqq_rr(cfqd, cfqq); |
1427 | 1430 | ||
1428 | /* | 1431 | /* |
1429 | * check if this request is a better next-serve candidate | 1432 | * check if this request is a better next-serve candidate |
1430 | */ | 1433 | */ |
1431 | prev = cfqq->next_rq; | 1434 | prev = cfqq->next_rq; |
1432 | cfqq->next_rq = cfq_choose_req(cfqd, cfqq->next_rq, rq, cfqd->last_position); | 1435 | cfqq->next_rq = cfq_choose_req(cfqd, cfqq->next_rq, rq, cfqd->last_position); |
1433 | 1436 | ||
1434 | /* | 1437 | /* |
1435 | * adjust priority tree position, if ->next_rq changes | 1438 | * adjust priority tree position, if ->next_rq changes |
1436 | */ | 1439 | */ |
1437 | if (prev != cfqq->next_rq) | 1440 | if (prev != cfqq->next_rq) |
1438 | cfq_prio_tree_add(cfqd, cfqq); | 1441 | cfq_prio_tree_add(cfqd, cfqq); |
1439 | 1442 | ||
1440 | BUG_ON(!cfqq->next_rq); | 1443 | BUG_ON(!cfqq->next_rq); |
1441 | } | 1444 | } |
1442 | 1445 | ||
1443 | static void cfq_reposition_rq_rb(struct cfq_queue *cfqq, struct request *rq) | 1446 | static void cfq_reposition_rq_rb(struct cfq_queue *cfqq, struct request *rq) |
1444 | { | 1447 | { |
1445 | elv_rb_del(&cfqq->sort_list, rq); | 1448 | elv_rb_del(&cfqq->sort_list, rq); |
1446 | cfqq->queued[rq_is_sync(rq)]--; | 1449 | cfqq->queued[rq_is_sync(rq)]--; |
1447 | cfq_blkiocg_update_io_remove_stats(&(RQ_CFQG(rq))->blkg, | 1450 | cfq_blkiocg_update_io_remove_stats(&(RQ_CFQG(rq))->blkg, |
1448 | rq_data_dir(rq), rq_is_sync(rq)); | 1451 | rq_data_dir(rq), rq_is_sync(rq)); |
1449 | cfq_add_rq_rb(rq); | 1452 | cfq_add_rq_rb(rq); |
1450 | cfq_blkiocg_update_io_add_stats(&(RQ_CFQG(rq))->blkg, | 1453 | cfq_blkiocg_update_io_add_stats(&(RQ_CFQG(rq))->blkg, |
1451 | &cfqq->cfqd->serving_group->blkg, rq_data_dir(rq), | 1454 | &cfqq->cfqd->serving_group->blkg, rq_data_dir(rq), |
1452 | rq_is_sync(rq)); | 1455 | rq_is_sync(rq)); |
1453 | } | 1456 | } |
1454 | 1457 | ||
1455 | static struct request * | 1458 | static struct request * |
1456 | cfq_find_rq_fmerge(struct cfq_data *cfqd, struct bio *bio) | 1459 | cfq_find_rq_fmerge(struct cfq_data *cfqd, struct bio *bio) |
1457 | { | 1460 | { |
1458 | struct task_struct *tsk = current; | 1461 | struct task_struct *tsk = current; |
1459 | struct cfq_io_context *cic; | 1462 | struct cfq_io_context *cic; |
1460 | struct cfq_queue *cfqq; | 1463 | struct cfq_queue *cfqq; |
1461 | 1464 | ||
1462 | cic = cfq_cic_lookup(cfqd, tsk->io_context); | 1465 | cic = cfq_cic_lookup(cfqd, tsk->io_context); |
1463 | if (!cic) | 1466 | if (!cic) |
1464 | return NULL; | 1467 | return NULL; |
1465 | 1468 | ||
1466 | cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio)); | 1469 | cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio)); |
1467 | if (cfqq) { | 1470 | if (cfqq) { |
1468 | sector_t sector = bio->bi_sector + bio_sectors(bio); | 1471 | sector_t sector = bio->bi_sector + bio_sectors(bio); |
1469 | 1472 | ||
1470 | return elv_rb_find(&cfqq->sort_list, sector); | 1473 | return elv_rb_find(&cfqq->sort_list, sector); |
1471 | } | 1474 | } |
1472 | 1475 | ||
1473 | return NULL; | 1476 | return NULL; |
1474 | } | 1477 | } |
1475 | 1478 | ||
1476 | static void cfq_activate_request(struct request_queue *q, struct request *rq) | 1479 | static void cfq_activate_request(struct request_queue *q, struct request *rq) |
1477 | { | 1480 | { |
1478 | struct cfq_data *cfqd = q->elevator->elevator_data; | 1481 | struct cfq_data *cfqd = q->elevator->elevator_data; |
1479 | 1482 | ||
1480 | cfqd->rq_in_driver++; | 1483 | cfqd->rq_in_driver++; |
1481 | cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "activate rq, drv=%d", | 1484 | cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "activate rq, drv=%d", |
1482 | cfqd->rq_in_driver); | 1485 | cfqd->rq_in_driver); |
1483 | 1486 | ||
1484 | cfqd->last_position = blk_rq_pos(rq) + blk_rq_sectors(rq); | 1487 | cfqd->last_position = blk_rq_pos(rq) + blk_rq_sectors(rq); |
1485 | } | 1488 | } |
1486 | 1489 | ||
1487 | static void cfq_deactivate_request(struct request_queue *q, struct request *rq) | 1490 | static void cfq_deactivate_request(struct request_queue *q, struct request *rq) |
1488 | { | 1491 | { |
1489 | struct cfq_data *cfqd = q->elevator->elevator_data; | 1492 | struct cfq_data *cfqd = q->elevator->elevator_data; |
1490 | 1493 | ||
1491 | WARN_ON(!cfqd->rq_in_driver); | 1494 | WARN_ON(!cfqd->rq_in_driver); |
1492 | cfqd->rq_in_driver--; | 1495 | cfqd->rq_in_driver--; |
1493 | cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "deactivate rq, drv=%d", | 1496 | cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "deactivate rq, drv=%d", |
1494 | cfqd->rq_in_driver); | 1497 | cfqd->rq_in_driver); |
1495 | } | 1498 | } |
1496 | 1499 | ||
1497 | static void cfq_remove_request(struct request *rq) | 1500 | static void cfq_remove_request(struct request *rq) |
1498 | { | 1501 | { |
1499 | struct cfq_queue *cfqq = RQ_CFQQ(rq); | 1502 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
1500 | 1503 | ||
1501 | if (cfqq->next_rq == rq) | 1504 | if (cfqq->next_rq == rq) |
1502 | cfqq->next_rq = cfq_find_next_rq(cfqq->cfqd, cfqq, rq); | 1505 | cfqq->next_rq = cfq_find_next_rq(cfqq->cfqd, cfqq, rq); |
1503 | 1506 | ||
1504 | list_del_init(&rq->queuelist); | 1507 | list_del_init(&rq->queuelist); |
1505 | cfq_del_rq_rb(rq); | 1508 | cfq_del_rq_rb(rq); |
1506 | 1509 | ||
1507 | cfqq->cfqd->rq_queued--; | 1510 | cfqq->cfqd->rq_queued--; |
1508 | cfq_blkiocg_update_io_remove_stats(&(RQ_CFQG(rq))->blkg, | 1511 | cfq_blkiocg_update_io_remove_stats(&(RQ_CFQG(rq))->blkg, |
1509 | rq_data_dir(rq), rq_is_sync(rq)); | 1512 | rq_data_dir(rq), rq_is_sync(rq)); |
1510 | if (rq->cmd_flags & REQ_META) { | 1513 | if (rq->cmd_flags & REQ_META) { |
1511 | WARN_ON(!cfqq->meta_pending); | 1514 | WARN_ON(!cfqq->meta_pending); |
1512 | cfqq->meta_pending--; | 1515 | cfqq->meta_pending--; |
1513 | } | 1516 | } |
1514 | } | 1517 | } |
1515 | 1518 | ||
1516 | static int cfq_merge(struct request_queue *q, struct request **req, | 1519 | static int cfq_merge(struct request_queue *q, struct request **req, |
1517 | struct bio *bio) | 1520 | struct bio *bio) |
1518 | { | 1521 | { |
1519 | struct cfq_data *cfqd = q->elevator->elevator_data; | 1522 | struct cfq_data *cfqd = q->elevator->elevator_data; |
1520 | struct request *__rq; | 1523 | struct request *__rq; |
1521 | 1524 | ||
1522 | __rq = cfq_find_rq_fmerge(cfqd, bio); | 1525 | __rq = cfq_find_rq_fmerge(cfqd, bio); |
1523 | if (__rq && elv_rq_merge_ok(__rq, bio)) { | 1526 | if (__rq && elv_rq_merge_ok(__rq, bio)) { |
1524 | *req = __rq; | 1527 | *req = __rq; |
1525 | return ELEVATOR_FRONT_MERGE; | 1528 | return ELEVATOR_FRONT_MERGE; |
1526 | } | 1529 | } |
1527 | 1530 | ||
1528 | return ELEVATOR_NO_MERGE; | 1531 | return ELEVATOR_NO_MERGE; |
1529 | } | 1532 | } |
1530 | 1533 | ||
1531 | static void cfq_merged_request(struct request_queue *q, struct request *req, | 1534 | static void cfq_merged_request(struct request_queue *q, struct request *req, |
1532 | int type) | 1535 | int type) |
1533 | { | 1536 | { |
1534 | if (type == ELEVATOR_FRONT_MERGE) { | 1537 | if (type == ELEVATOR_FRONT_MERGE) { |
1535 | struct cfq_queue *cfqq = RQ_CFQQ(req); | 1538 | struct cfq_queue *cfqq = RQ_CFQQ(req); |
1536 | 1539 | ||
1537 | cfq_reposition_rq_rb(cfqq, req); | 1540 | cfq_reposition_rq_rb(cfqq, req); |
1538 | } | 1541 | } |
1539 | } | 1542 | } |
1540 | 1543 | ||
1541 | static void cfq_bio_merged(struct request_queue *q, struct request *req, | 1544 | static void cfq_bio_merged(struct request_queue *q, struct request *req, |
1542 | struct bio *bio) | 1545 | struct bio *bio) |
1543 | { | 1546 | { |
1544 | cfq_blkiocg_update_io_merged_stats(&(RQ_CFQG(req))->blkg, | 1547 | cfq_blkiocg_update_io_merged_stats(&(RQ_CFQG(req))->blkg, |
1545 | bio_data_dir(bio), cfq_bio_sync(bio)); | 1548 | bio_data_dir(bio), cfq_bio_sync(bio)); |
1546 | } | 1549 | } |
1547 | 1550 | ||
1548 | static void | 1551 | static void |
1549 | cfq_merged_requests(struct request_queue *q, struct request *rq, | 1552 | cfq_merged_requests(struct request_queue *q, struct request *rq, |
1550 | struct request *next) | 1553 | struct request *next) |
1551 | { | 1554 | { |
1552 | struct cfq_queue *cfqq = RQ_CFQQ(rq); | 1555 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
1553 | /* | 1556 | /* |
1554 | * reposition in fifo if next is older than rq | 1557 | * reposition in fifo if next is older than rq |
1555 | */ | 1558 | */ |
1556 | if (!list_empty(&rq->queuelist) && !list_empty(&next->queuelist) && | 1559 | if (!list_empty(&rq->queuelist) && !list_empty(&next->queuelist) && |
1557 | time_before(rq_fifo_time(next), rq_fifo_time(rq))) { | 1560 | time_before(rq_fifo_time(next), rq_fifo_time(rq))) { |
1558 | list_move(&rq->queuelist, &next->queuelist); | 1561 | list_move(&rq->queuelist, &next->queuelist); |
1559 | rq_set_fifo_time(rq, rq_fifo_time(next)); | 1562 | rq_set_fifo_time(rq, rq_fifo_time(next)); |
1560 | } | 1563 | } |
1561 | 1564 | ||
1562 | if (cfqq->next_rq == next) | 1565 | if (cfqq->next_rq == next) |
1563 | cfqq->next_rq = rq; | 1566 | cfqq->next_rq = rq; |
1564 | cfq_remove_request(next); | 1567 | cfq_remove_request(next); |
1565 | cfq_blkiocg_update_io_merged_stats(&(RQ_CFQG(rq))->blkg, | 1568 | cfq_blkiocg_update_io_merged_stats(&(RQ_CFQG(rq))->blkg, |
1566 | rq_data_dir(next), rq_is_sync(next)); | 1569 | rq_data_dir(next), rq_is_sync(next)); |
1567 | } | 1570 | } |
1568 | 1571 | ||
1569 | static int cfq_allow_merge(struct request_queue *q, struct request *rq, | 1572 | static int cfq_allow_merge(struct request_queue *q, struct request *rq, |
1570 | struct bio *bio) | 1573 | struct bio *bio) |
1571 | { | 1574 | { |
1572 | struct cfq_data *cfqd = q->elevator->elevator_data; | 1575 | struct cfq_data *cfqd = q->elevator->elevator_data; |
1573 | struct cfq_io_context *cic; | 1576 | struct cfq_io_context *cic; |
1574 | struct cfq_queue *cfqq; | 1577 | struct cfq_queue *cfqq; |
1575 | 1578 | ||
1576 | /* | 1579 | /* |
1577 | * Disallow merge of a sync bio into an async request. | 1580 | * Disallow merge of a sync bio into an async request. |
1578 | */ | 1581 | */ |
1579 | if (cfq_bio_sync(bio) && !rq_is_sync(rq)) | 1582 | if (cfq_bio_sync(bio) && !rq_is_sync(rq)) |
1580 | return false; | 1583 | return false; |
1581 | 1584 | ||
1582 | /* | 1585 | /* |
1583 | * Lookup the cfqq that this bio will be queued with. Allow | 1586 | * Lookup the cfqq that this bio will be queued with. Allow |
1584 | * merge only if rq is queued there. | 1587 | * merge only if rq is queued there. |
1585 | */ | 1588 | */ |
1586 | cic = cfq_cic_lookup(cfqd, current->io_context); | 1589 | cic = cfq_cic_lookup(cfqd, current->io_context); |
1587 | if (!cic) | 1590 | if (!cic) |
1588 | return false; | 1591 | return false; |
1589 | 1592 | ||
1590 | cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio)); | 1593 | cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio)); |
1591 | return cfqq == RQ_CFQQ(rq); | 1594 | return cfqq == RQ_CFQQ(rq); |
1592 | } | 1595 | } |
1593 | 1596 | ||
1594 | static inline void cfq_del_timer(struct cfq_data *cfqd, struct cfq_queue *cfqq) | 1597 | static inline void cfq_del_timer(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1595 | { | 1598 | { |
1596 | del_timer(&cfqd->idle_slice_timer); | 1599 | del_timer(&cfqd->idle_slice_timer); |
1597 | cfq_blkiocg_update_idle_time_stats(&cfqq->cfqg->blkg); | 1600 | cfq_blkiocg_update_idle_time_stats(&cfqq->cfqg->blkg); |
1598 | } | 1601 | } |
1599 | 1602 | ||
1600 | static void __cfq_set_active_queue(struct cfq_data *cfqd, | 1603 | static void __cfq_set_active_queue(struct cfq_data *cfqd, |
1601 | struct cfq_queue *cfqq) | 1604 | struct cfq_queue *cfqq) |
1602 | { | 1605 | { |
1603 | if (cfqq) { | 1606 | if (cfqq) { |
1604 | cfq_log_cfqq(cfqd, cfqq, "set_active wl_prio:%d wl_type:%d", | 1607 | cfq_log_cfqq(cfqd, cfqq, "set_active wl_prio:%d wl_type:%d", |
1605 | cfqd->serving_prio, cfqd->serving_type); | 1608 | cfqd->serving_prio, cfqd->serving_type); |
1606 | cfq_blkiocg_update_avg_queue_size_stats(&cfqq->cfqg->blkg); | 1609 | cfq_blkiocg_update_avg_queue_size_stats(&cfqq->cfqg->blkg); |
1607 | cfqq->slice_start = 0; | 1610 | cfqq->slice_start = 0; |
1608 | cfqq->dispatch_start = jiffies; | 1611 | cfqq->dispatch_start = jiffies; |
1609 | cfqq->allocated_slice = 0; | 1612 | cfqq->allocated_slice = 0; |
1610 | cfqq->slice_end = 0; | 1613 | cfqq->slice_end = 0; |
1611 | cfqq->slice_dispatch = 0; | 1614 | cfqq->slice_dispatch = 0; |
1615 | cfqq->nr_sectors = 0; | ||
1612 | 1616 | ||
1613 | cfq_clear_cfqq_wait_request(cfqq); | 1617 | cfq_clear_cfqq_wait_request(cfqq); |
1614 | cfq_clear_cfqq_must_dispatch(cfqq); | 1618 | cfq_clear_cfqq_must_dispatch(cfqq); |
1615 | cfq_clear_cfqq_must_alloc_slice(cfqq); | 1619 | cfq_clear_cfqq_must_alloc_slice(cfqq); |
1616 | cfq_clear_cfqq_fifo_expire(cfqq); | 1620 | cfq_clear_cfqq_fifo_expire(cfqq); |
1617 | cfq_mark_cfqq_slice_new(cfqq); | 1621 | cfq_mark_cfqq_slice_new(cfqq); |
1618 | 1622 | ||
1619 | cfq_del_timer(cfqd, cfqq); | 1623 | cfq_del_timer(cfqd, cfqq); |
1620 | } | 1624 | } |
1621 | 1625 | ||
1622 | cfqd->active_queue = cfqq; | 1626 | cfqd->active_queue = cfqq; |
1623 | } | 1627 | } |
1624 | 1628 | ||
1625 | /* | 1629 | /* |
1626 | * current cfqq expired its slice (or was too idle), select new one | 1630 | * current cfqq expired its slice (or was too idle), select new one |
1627 | */ | 1631 | */ |
1628 | static void | 1632 | static void |
1629 | __cfq_slice_expired(struct cfq_data *cfqd, struct cfq_queue *cfqq, | 1633 | __cfq_slice_expired(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
1630 | bool timed_out) | 1634 | bool timed_out) |
1631 | { | 1635 | { |
1632 | cfq_log_cfqq(cfqd, cfqq, "slice expired t=%d", timed_out); | 1636 | cfq_log_cfqq(cfqd, cfqq, "slice expired t=%d", timed_out); |
1633 | 1637 | ||
1634 | if (cfq_cfqq_wait_request(cfqq)) | 1638 | if (cfq_cfqq_wait_request(cfqq)) |
1635 | cfq_del_timer(cfqd, cfqq); | 1639 | cfq_del_timer(cfqd, cfqq); |
1636 | 1640 | ||
1637 | cfq_clear_cfqq_wait_request(cfqq); | 1641 | cfq_clear_cfqq_wait_request(cfqq); |
1638 | cfq_clear_cfqq_wait_busy(cfqq); | 1642 | cfq_clear_cfqq_wait_busy(cfqq); |
1639 | 1643 | ||
1640 | /* | 1644 | /* |
1641 | * If this cfqq is shared between multiple processes, check to | 1645 | * If this cfqq is shared between multiple processes, check to |
1642 | * make sure that those processes are still issuing I/Os within | 1646 | * make sure that those processes are still issuing I/Os within |
1643 | * the mean seek distance. If not, it may be time to break the | 1647 | * the mean seek distance. If not, it may be time to break the |
1644 | * queues apart again. | 1648 | * queues apart again. |
1645 | */ | 1649 | */ |
1646 | if (cfq_cfqq_coop(cfqq) && CFQQ_SEEKY(cfqq)) | 1650 | if (cfq_cfqq_coop(cfqq) && CFQQ_SEEKY(cfqq)) |
1647 | cfq_mark_cfqq_split_coop(cfqq); | 1651 | cfq_mark_cfqq_split_coop(cfqq); |
1648 | 1652 | ||
1649 | /* | 1653 | /* |
1650 | * store what was left of this slice, if the queue idled/timed out | 1654 | * store what was left of this slice, if the queue idled/timed out |
1651 | */ | 1655 | */ |
1652 | if (timed_out && !cfq_cfqq_slice_new(cfqq)) { | 1656 | if (timed_out && !cfq_cfqq_slice_new(cfqq)) { |
1653 | cfqq->slice_resid = cfqq->slice_end - jiffies; | 1657 | cfqq->slice_resid = cfqq->slice_end - jiffies; |
1654 | cfq_log_cfqq(cfqd, cfqq, "resid=%ld", cfqq->slice_resid); | 1658 | cfq_log_cfqq(cfqd, cfqq, "resid=%ld", cfqq->slice_resid); |
1655 | } | 1659 | } |
1656 | 1660 | ||
1657 | cfq_group_served(cfqd, cfqq->cfqg, cfqq); | 1661 | cfq_group_served(cfqd, cfqq->cfqg, cfqq); |
1658 | 1662 | ||
1659 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list)) | 1663 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list)) |
1660 | cfq_del_cfqq_rr(cfqd, cfqq); | 1664 | cfq_del_cfqq_rr(cfqd, cfqq); |
1661 | 1665 | ||
1662 | cfq_resort_rr_list(cfqd, cfqq); | 1666 | cfq_resort_rr_list(cfqd, cfqq); |
1663 | 1667 | ||
1664 | if (cfqq == cfqd->active_queue) | 1668 | if (cfqq == cfqd->active_queue) |
1665 | cfqd->active_queue = NULL; | 1669 | cfqd->active_queue = NULL; |
1666 | 1670 | ||
1667 | if (&cfqq->cfqg->rb_node == cfqd->grp_service_tree.active) | 1671 | if (&cfqq->cfqg->rb_node == cfqd->grp_service_tree.active) |
1668 | cfqd->grp_service_tree.active = NULL; | 1672 | cfqd->grp_service_tree.active = NULL; |
1669 | 1673 | ||
1670 | if (cfqd->active_cic) { | 1674 | if (cfqd->active_cic) { |
1671 | put_io_context(cfqd->active_cic->ioc); | 1675 | put_io_context(cfqd->active_cic->ioc); |
1672 | cfqd->active_cic = NULL; | 1676 | cfqd->active_cic = NULL; |
1673 | } | 1677 | } |
1674 | } | 1678 | } |
1675 | 1679 | ||
1676 | static inline void cfq_slice_expired(struct cfq_data *cfqd, bool timed_out) | 1680 | static inline void cfq_slice_expired(struct cfq_data *cfqd, bool timed_out) |
1677 | { | 1681 | { |
1678 | struct cfq_queue *cfqq = cfqd->active_queue; | 1682 | struct cfq_queue *cfqq = cfqd->active_queue; |
1679 | 1683 | ||
1680 | if (cfqq) | 1684 | if (cfqq) |
1681 | __cfq_slice_expired(cfqd, cfqq, timed_out); | 1685 | __cfq_slice_expired(cfqd, cfqq, timed_out); |
1682 | } | 1686 | } |
1683 | 1687 | ||
1684 | /* | 1688 | /* |
1685 | * Get next queue for service. Unless we have a queue preemption, | 1689 | * Get next queue for service. Unless we have a queue preemption, |
1686 | * we'll simply select the first cfqq in the service tree. | 1690 | * we'll simply select the first cfqq in the service tree. |
1687 | */ | 1691 | */ |
1688 | static struct cfq_queue *cfq_get_next_queue(struct cfq_data *cfqd) | 1692 | static struct cfq_queue *cfq_get_next_queue(struct cfq_data *cfqd) |
1689 | { | 1693 | { |
1690 | struct cfq_rb_root *service_tree = | 1694 | struct cfq_rb_root *service_tree = |
1691 | service_tree_for(cfqd->serving_group, cfqd->serving_prio, | 1695 | service_tree_for(cfqd->serving_group, cfqd->serving_prio, |
1692 | cfqd->serving_type); | 1696 | cfqd->serving_type); |
1693 | 1697 | ||
1694 | if (!cfqd->rq_queued) | 1698 | if (!cfqd->rq_queued) |
1695 | return NULL; | 1699 | return NULL; |
1696 | 1700 | ||
1697 | /* There is nothing to dispatch */ | 1701 | /* There is nothing to dispatch */ |
1698 | if (!service_tree) | 1702 | if (!service_tree) |
1699 | return NULL; | 1703 | return NULL; |
1700 | if (RB_EMPTY_ROOT(&service_tree->rb)) | 1704 | if (RB_EMPTY_ROOT(&service_tree->rb)) |
1701 | return NULL; | 1705 | return NULL; |
1702 | return cfq_rb_first(service_tree); | 1706 | return cfq_rb_first(service_tree); |
1703 | } | 1707 | } |
1704 | 1708 | ||
1705 | static struct cfq_queue *cfq_get_next_queue_forced(struct cfq_data *cfqd) | 1709 | static struct cfq_queue *cfq_get_next_queue_forced(struct cfq_data *cfqd) |
1706 | { | 1710 | { |
1707 | struct cfq_group *cfqg; | 1711 | struct cfq_group *cfqg; |
1708 | struct cfq_queue *cfqq; | 1712 | struct cfq_queue *cfqq; |
1709 | int i, j; | 1713 | int i, j; |
1710 | struct cfq_rb_root *st; | 1714 | struct cfq_rb_root *st; |
1711 | 1715 | ||
1712 | if (!cfqd->rq_queued) | 1716 | if (!cfqd->rq_queued) |
1713 | return NULL; | 1717 | return NULL; |
1714 | 1718 | ||
1715 | cfqg = cfq_get_next_cfqg(cfqd); | 1719 | cfqg = cfq_get_next_cfqg(cfqd); |
1716 | if (!cfqg) | 1720 | if (!cfqg) |
1717 | return NULL; | 1721 | return NULL; |
1718 | 1722 | ||
1719 | for_each_cfqg_st(cfqg, i, j, st) | 1723 | for_each_cfqg_st(cfqg, i, j, st) |
1720 | if ((cfqq = cfq_rb_first(st)) != NULL) | 1724 | if ((cfqq = cfq_rb_first(st)) != NULL) |
1721 | return cfqq; | 1725 | return cfqq; |
1722 | return NULL; | 1726 | return NULL; |
1723 | } | 1727 | } |
1724 | 1728 | ||
1725 | /* | 1729 | /* |
1726 | * Get and set a new active queue for service. | 1730 | * Get and set a new active queue for service. |
1727 | */ | 1731 | */ |
1728 | static struct cfq_queue *cfq_set_active_queue(struct cfq_data *cfqd, | 1732 | static struct cfq_queue *cfq_set_active_queue(struct cfq_data *cfqd, |
1729 | struct cfq_queue *cfqq) | 1733 | struct cfq_queue *cfqq) |
1730 | { | 1734 | { |
1731 | if (!cfqq) | 1735 | if (!cfqq) |
1732 | cfqq = cfq_get_next_queue(cfqd); | 1736 | cfqq = cfq_get_next_queue(cfqd); |
1733 | 1737 | ||
1734 | __cfq_set_active_queue(cfqd, cfqq); | 1738 | __cfq_set_active_queue(cfqd, cfqq); |
1735 | return cfqq; | 1739 | return cfqq; |
1736 | } | 1740 | } |
1737 | 1741 | ||
1738 | static inline sector_t cfq_dist_from_last(struct cfq_data *cfqd, | 1742 | static inline sector_t cfq_dist_from_last(struct cfq_data *cfqd, |
1739 | struct request *rq) | 1743 | struct request *rq) |
1740 | { | 1744 | { |
1741 | if (blk_rq_pos(rq) >= cfqd->last_position) | 1745 | if (blk_rq_pos(rq) >= cfqd->last_position) |
1742 | return blk_rq_pos(rq) - cfqd->last_position; | 1746 | return blk_rq_pos(rq) - cfqd->last_position; |
1743 | else | 1747 | else |
1744 | return cfqd->last_position - blk_rq_pos(rq); | 1748 | return cfqd->last_position - blk_rq_pos(rq); |
1745 | } | 1749 | } |
1746 | 1750 | ||
1747 | static inline int cfq_rq_close(struct cfq_data *cfqd, struct cfq_queue *cfqq, | 1751 | static inline int cfq_rq_close(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
1748 | struct request *rq) | 1752 | struct request *rq) |
1749 | { | 1753 | { |
1750 | return cfq_dist_from_last(cfqd, rq) <= CFQQ_CLOSE_THR; | 1754 | return cfq_dist_from_last(cfqd, rq) <= CFQQ_CLOSE_THR; |
1751 | } | 1755 | } |
1752 | 1756 | ||
1753 | static struct cfq_queue *cfqq_close(struct cfq_data *cfqd, | 1757 | static struct cfq_queue *cfqq_close(struct cfq_data *cfqd, |
1754 | struct cfq_queue *cur_cfqq) | 1758 | struct cfq_queue *cur_cfqq) |
1755 | { | 1759 | { |
1756 | struct rb_root *root = &cfqd->prio_trees[cur_cfqq->org_ioprio]; | 1760 | struct rb_root *root = &cfqd->prio_trees[cur_cfqq->org_ioprio]; |
1757 | struct rb_node *parent, *node; | 1761 | struct rb_node *parent, *node; |
1758 | struct cfq_queue *__cfqq; | 1762 | struct cfq_queue *__cfqq; |
1759 | sector_t sector = cfqd->last_position; | 1763 | sector_t sector = cfqd->last_position; |
1760 | 1764 | ||
1761 | if (RB_EMPTY_ROOT(root)) | 1765 | if (RB_EMPTY_ROOT(root)) |
1762 | return NULL; | 1766 | return NULL; |
1763 | 1767 | ||
1764 | /* | 1768 | /* |
1765 | * First, if we find a request starting at the end of the last | 1769 | * First, if we find a request starting at the end of the last |
1766 | * request, choose it. | 1770 | * request, choose it. |
1767 | */ | 1771 | */ |
1768 | __cfqq = cfq_prio_tree_lookup(cfqd, root, sector, &parent, NULL); | 1772 | __cfqq = cfq_prio_tree_lookup(cfqd, root, sector, &parent, NULL); |
1769 | if (__cfqq) | 1773 | if (__cfqq) |
1770 | return __cfqq; | 1774 | return __cfqq; |
1771 | 1775 | ||
1772 | /* | 1776 | /* |
1773 | * If the exact sector wasn't found, the parent of the NULL leaf | 1777 | * If the exact sector wasn't found, the parent of the NULL leaf |
1774 | * will contain the closest sector. | 1778 | * will contain the closest sector. |
1775 | */ | 1779 | */ |
1776 | __cfqq = rb_entry(parent, struct cfq_queue, p_node); | 1780 | __cfqq = rb_entry(parent, struct cfq_queue, p_node); |
1777 | if (cfq_rq_close(cfqd, cur_cfqq, __cfqq->next_rq)) | 1781 | if (cfq_rq_close(cfqd, cur_cfqq, __cfqq->next_rq)) |
1778 | return __cfqq; | 1782 | return __cfqq; |
1779 | 1783 | ||
1780 | if (blk_rq_pos(__cfqq->next_rq) < sector) | 1784 | if (blk_rq_pos(__cfqq->next_rq) < sector) |
1781 | node = rb_next(&__cfqq->p_node); | 1785 | node = rb_next(&__cfqq->p_node); |
1782 | else | 1786 | else |
1783 | node = rb_prev(&__cfqq->p_node); | 1787 | node = rb_prev(&__cfqq->p_node); |
1784 | if (!node) | 1788 | if (!node) |
1785 | return NULL; | 1789 | return NULL; |
1786 | 1790 | ||
1787 | __cfqq = rb_entry(node, struct cfq_queue, p_node); | 1791 | __cfqq = rb_entry(node, struct cfq_queue, p_node); |
1788 | if (cfq_rq_close(cfqd, cur_cfqq, __cfqq->next_rq)) | 1792 | if (cfq_rq_close(cfqd, cur_cfqq, __cfqq->next_rq)) |
1789 | return __cfqq; | 1793 | return __cfqq; |
1790 | 1794 | ||
1791 | return NULL; | 1795 | return NULL; |
1792 | } | 1796 | } |
1793 | 1797 | ||
1794 | /* | 1798 | /* |
1795 | * cfqd - obvious | 1799 | * cfqd - obvious |
1796 | * cur_cfqq - passed in so that we don't decide that the current queue is | 1800 | * cur_cfqq - passed in so that we don't decide that the current queue is |
1797 | * closely cooperating with itself. | 1801 | * closely cooperating with itself. |
1798 | * | 1802 | * |
1799 | * So, basically we're assuming that that cur_cfqq has dispatched at least | 1803 | * So, basically we're assuming that that cur_cfqq has dispatched at least |
1800 | * one request, and that cfqd->last_position reflects a position on the disk | 1804 | * one request, and that cfqd->last_position reflects a position on the disk |
1801 | * associated with the I/O issued by cur_cfqq. I'm not sure this is a valid | 1805 | * associated with the I/O issued by cur_cfqq. I'm not sure this is a valid |
1802 | * assumption. | 1806 | * assumption. |
1803 | */ | 1807 | */ |
1804 | static struct cfq_queue *cfq_close_cooperator(struct cfq_data *cfqd, | 1808 | static struct cfq_queue *cfq_close_cooperator(struct cfq_data *cfqd, |
1805 | struct cfq_queue *cur_cfqq) | 1809 | struct cfq_queue *cur_cfqq) |
1806 | { | 1810 | { |
1807 | struct cfq_queue *cfqq; | 1811 | struct cfq_queue *cfqq; |
1808 | 1812 | ||
1809 | if (cfq_class_idle(cur_cfqq)) | 1813 | if (cfq_class_idle(cur_cfqq)) |
1810 | return NULL; | 1814 | return NULL; |
1811 | if (!cfq_cfqq_sync(cur_cfqq)) | 1815 | if (!cfq_cfqq_sync(cur_cfqq)) |
1812 | return NULL; | 1816 | return NULL; |
1813 | if (CFQQ_SEEKY(cur_cfqq)) | 1817 | if (CFQQ_SEEKY(cur_cfqq)) |
1814 | return NULL; | 1818 | return NULL; |
1815 | 1819 | ||
1816 | /* | 1820 | /* |
1817 | * Don't search priority tree if it's the only queue in the group. | 1821 | * Don't search priority tree if it's the only queue in the group. |
1818 | */ | 1822 | */ |
1819 | if (cur_cfqq->cfqg->nr_cfqq == 1) | 1823 | if (cur_cfqq->cfqg->nr_cfqq == 1) |
1820 | return NULL; | 1824 | return NULL; |
1821 | 1825 | ||
1822 | /* | 1826 | /* |
1823 | * We should notice if some of the queues are cooperating, eg | 1827 | * We should notice if some of the queues are cooperating, eg |
1824 | * working closely on the same area of the disk. In that case, | 1828 | * working closely on the same area of the disk. In that case, |
1825 | * we can group them together and don't waste time idling. | 1829 | * we can group them together and don't waste time idling. |
1826 | */ | 1830 | */ |
1827 | cfqq = cfqq_close(cfqd, cur_cfqq); | 1831 | cfqq = cfqq_close(cfqd, cur_cfqq); |
1828 | if (!cfqq) | 1832 | if (!cfqq) |
1829 | return NULL; | 1833 | return NULL; |
1830 | 1834 | ||
1831 | /* If new queue belongs to different cfq_group, don't choose it */ | 1835 | /* If new queue belongs to different cfq_group, don't choose it */ |
1832 | if (cur_cfqq->cfqg != cfqq->cfqg) | 1836 | if (cur_cfqq->cfqg != cfqq->cfqg) |
1833 | return NULL; | 1837 | return NULL; |
1834 | 1838 | ||
1835 | /* | 1839 | /* |
1836 | * It only makes sense to merge sync queues. | 1840 | * It only makes sense to merge sync queues. |
1837 | */ | 1841 | */ |
1838 | if (!cfq_cfqq_sync(cfqq)) | 1842 | if (!cfq_cfqq_sync(cfqq)) |
1839 | return NULL; | 1843 | return NULL; |
1840 | if (CFQQ_SEEKY(cfqq)) | 1844 | if (CFQQ_SEEKY(cfqq)) |
1841 | return NULL; | 1845 | return NULL; |
1842 | 1846 | ||
1843 | /* | 1847 | /* |
1844 | * Do not merge queues of different priority classes | 1848 | * Do not merge queues of different priority classes |
1845 | */ | 1849 | */ |
1846 | if (cfq_class_rt(cfqq) != cfq_class_rt(cur_cfqq)) | 1850 | if (cfq_class_rt(cfqq) != cfq_class_rt(cur_cfqq)) |
1847 | return NULL; | 1851 | return NULL; |
1848 | 1852 | ||
1849 | return cfqq; | 1853 | return cfqq; |
1850 | } | 1854 | } |
1851 | 1855 | ||
1852 | /* | 1856 | /* |
1853 | * Determine whether we should enforce idle window for this queue. | 1857 | * Determine whether we should enforce idle window for this queue. |
1854 | */ | 1858 | */ |
1855 | 1859 | ||
1856 | static bool cfq_should_idle(struct cfq_data *cfqd, struct cfq_queue *cfqq) | 1860 | static bool cfq_should_idle(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1857 | { | 1861 | { |
1858 | enum wl_prio_t prio = cfqq_prio(cfqq); | 1862 | enum wl_prio_t prio = cfqq_prio(cfqq); |
1859 | struct cfq_rb_root *service_tree = cfqq->service_tree; | 1863 | struct cfq_rb_root *service_tree = cfqq->service_tree; |
1860 | 1864 | ||
1861 | BUG_ON(!service_tree); | 1865 | BUG_ON(!service_tree); |
1862 | BUG_ON(!service_tree->count); | 1866 | BUG_ON(!service_tree->count); |
1863 | 1867 | ||
1864 | if (!cfqd->cfq_slice_idle) | 1868 | if (!cfqd->cfq_slice_idle) |
1865 | return false; | 1869 | return false; |
1866 | 1870 | ||
1867 | /* We never do for idle class queues. */ | 1871 | /* We never do for idle class queues. */ |
1868 | if (prio == IDLE_WORKLOAD) | 1872 | if (prio == IDLE_WORKLOAD) |
1869 | return false; | 1873 | return false; |
1870 | 1874 | ||
1871 | /* We do for queues that were marked with idle window flag. */ | 1875 | /* We do for queues that were marked with idle window flag. */ |
1872 | if (cfq_cfqq_idle_window(cfqq) && | 1876 | if (cfq_cfqq_idle_window(cfqq) && |
1873 | !(blk_queue_nonrot(cfqd->queue) && cfqd->hw_tag)) | 1877 | !(blk_queue_nonrot(cfqd->queue) && cfqd->hw_tag)) |
1874 | return true; | 1878 | return true; |
1875 | 1879 | ||
1876 | /* | 1880 | /* |
1877 | * Otherwise, we do only if they are the last ones | 1881 | * Otherwise, we do only if they are the last ones |
1878 | * in their service tree. | 1882 | * in their service tree. |
1879 | */ | 1883 | */ |
1880 | if (service_tree->count == 1 && cfq_cfqq_sync(cfqq)) | 1884 | if (service_tree->count == 1 && cfq_cfqq_sync(cfqq)) |
1881 | return 1; | 1885 | return 1; |
1882 | cfq_log_cfqq(cfqd, cfqq, "Not idling. st->count:%d", | 1886 | cfq_log_cfqq(cfqd, cfqq, "Not idling. st->count:%d", |
1883 | service_tree->count); | 1887 | service_tree->count); |
1884 | return 0; | 1888 | return 0; |
1885 | } | 1889 | } |
1886 | 1890 | ||
1887 | static void cfq_arm_slice_timer(struct cfq_data *cfqd) | 1891 | static void cfq_arm_slice_timer(struct cfq_data *cfqd) |
1888 | { | 1892 | { |
1889 | struct cfq_queue *cfqq = cfqd->active_queue; | 1893 | struct cfq_queue *cfqq = cfqd->active_queue; |
1890 | struct cfq_io_context *cic; | 1894 | struct cfq_io_context *cic; |
1891 | unsigned long sl, group_idle = 0; | 1895 | unsigned long sl, group_idle = 0; |
1892 | 1896 | ||
1893 | /* | 1897 | /* |
1894 | * SSD device without seek penalty, disable idling. But only do so | 1898 | * SSD device without seek penalty, disable idling. But only do so |
1895 | * for devices that support queuing, otherwise we still have a problem | 1899 | * for devices that support queuing, otherwise we still have a problem |
1896 | * with sync vs async workloads. | 1900 | * with sync vs async workloads. |
1897 | */ | 1901 | */ |
1898 | if (blk_queue_nonrot(cfqd->queue) && cfqd->hw_tag) | 1902 | if (blk_queue_nonrot(cfqd->queue) && cfqd->hw_tag) |
1899 | return; | 1903 | return; |
1900 | 1904 | ||
1901 | WARN_ON(!RB_EMPTY_ROOT(&cfqq->sort_list)); | 1905 | WARN_ON(!RB_EMPTY_ROOT(&cfqq->sort_list)); |
1902 | WARN_ON(cfq_cfqq_slice_new(cfqq)); | 1906 | WARN_ON(cfq_cfqq_slice_new(cfqq)); |
1903 | 1907 | ||
1904 | /* | 1908 | /* |
1905 | * idle is disabled, either manually or by past process history | 1909 | * idle is disabled, either manually or by past process history |
1906 | */ | 1910 | */ |
1907 | if (!cfq_should_idle(cfqd, cfqq)) { | 1911 | if (!cfq_should_idle(cfqd, cfqq)) { |
1908 | /* no queue idling. Check for group idling */ | 1912 | /* no queue idling. Check for group idling */ |
1909 | if (cfqd->cfq_group_idle) | 1913 | if (cfqd->cfq_group_idle) |
1910 | group_idle = cfqd->cfq_group_idle; | 1914 | group_idle = cfqd->cfq_group_idle; |
1911 | else | 1915 | else |
1912 | return; | 1916 | return; |
1913 | } | 1917 | } |
1914 | 1918 | ||
1915 | /* | 1919 | /* |
1916 | * still active requests from this queue, don't idle | 1920 | * still active requests from this queue, don't idle |
1917 | */ | 1921 | */ |
1918 | if (cfqq->dispatched) | 1922 | if (cfqq->dispatched) |
1919 | return; | 1923 | return; |
1920 | 1924 | ||
1921 | /* | 1925 | /* |
1922 | * task has exited, don't wait | 1926 | * task has exited, don't wait |
1923 | */ | 1927 | */ |
1924 | cic = cfqd->active_cic; | 1928 | cic = cfqd->active_cic; |
1925 | if (!cic || !atomic_read(&cic->ioc->nr_tasks)) | 1929 | if (!cic || !atomic_read(&cic->ioc->nr_tasks)) |
1926 | return; | 1930 | return; |
1927 | 1931 | ||
1928 | /* | 1932 | /* |
1929 | * If our average think time is larger than the remaining time | 1933 | * If our average think time is larger than the remaining time |
1930 | * slice, then don't idle. This avoids overrunning the allotted | 1934 | * slice, then don't idle. This avoids overrunning the allotted |
1931 | * time slice. | 1935 | * time slice. |
1932 | */ | 1936 | */ |
1933 | if (sample_valid(cic->ttime_samples) && | 1937 | if (sample_valid(cic->ttime_samples) && |
1934 | (cfqq->slice_end - jiffies < cic->ttime_mean)) { | 1938 | (cfqq->slice_end - jiffies < cic->ttime_mean)) { |
1935 | cfq_log_cfqq(cfqd, cfqq, "Not idling. think_time:%d", | 1939 | cfq_log_cfqq(cfqd, cfqq, "Not idling. think_time:%d", |
1936 | cic->ttime_mean); | 1940 | cic->ttime_mean); |
1937 | return; | 1941 | return; |
1938 | } | 1942 | } |
1939 | 1943 | ||
1940 | /* There are other queues in the group, don't do group idle */ | 1944 | /* There are other queues in the group, don't do group idle */ |
1941 | if (group_idle && cfqq->cfqg->nr_cfqq > 1) | 1945 | if (group_idle && cfqq->cfqg->nr_cfqq > 1) |
1942 | return; | 1946 | return; |
1943 | 1947 | ||
1944 | cfq_mark_cfqq_wait_request(cfqq); | 1948 | cfq_mark_cfqq_wait_request(cfqq); |
1945 | 1949 | ||
1946 | if (group_idle) | 1950 | if (group_idle) |
1947 | sl = cfqd->cfq_group_idle; | 1951 | sl = cfqd->cfq_group_idle; |
1948 | else | 1952 | else |
1949 | sl = cfqd->cfq_slice_idle; | 1953 | sl = cfqd->cfq_slice_idle; |
1950 | 1954 | ||
1951 | mod_timer(&cfqd->idle_slice_timer, jiffies + sl); | 1955 | mod_timer(&cfqd->idle_slice_timer, jiffies + sl); |
1952 | cfq_blkiocg_update_set_idle_time_stats(&cfqq->cfqg->blkg); | 1956 | cfq_blkiocg_update_set_idle_time_stats(&cfqq->cfqg->blkg); |
1953 | cfq_log_cfqq(cfqd, cfqq, "arm_idle: %lu group_idle: %d", sl, | 1957 | cfq_log_cfqq(cfqd, cfqq, "arm_idle: %lu group_idle: %d", sl, |
1954 | group_idle ? 1 : 0); | 1958 | group_idle ? 1 : 0); |
1955 | } | 1959 | } |
1956 | 1960 | ||
1957 | /* | 1961 | /* |
1958 | * Move request from internal lists to the request queue dispatch list. | 1962 | * Move request from internal lists to the request queue dispatch list. |
1959 | */ | 1963 | */ |
1960 | static void cfq_dispatch_insert(struct request_queue *q, struct request *rq) | 1964 | static void cfq_dispatch_insert(struct request_queue *q, struct request *rq) |
1961 | { | 1965 | { |
1962 | struct cfq_data *cfqd = q->elevator->elevator_data; | 1966 | struct cfq_data *cfqd = q->elevator->elevator_data; |
1963 | struct cfq_queue *cfqq = RQ_CFQQ(rq); | 1967 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
1964 | 1968 | ||
1965 | cfq_log_cfqq(cfqd, cfqq, "dispatch_insert"); | 1969 | cfq_log_cfqq(cfqd, cfqq, "dispatch_insert"); |
1966 | 1970 | ||
1967 | cfqq->next_rq = cfq_find_next_rq(cfqd, cfqq, rq); | 1971 | cfqq->next_rq = cfq_find_next_rq(cfqd, cfqq, rq); |
1968 | cfq_remove_request(rq); | 1972 | cfq_remove_request(rq); |
1969 | cfqq->dispatched++; | 1973 | cfqq->dispatched++; |
1970 | (RQ_CFQG(rq))->dispatched++; | 1974 | (RQ_CFQG(rq))->dispatched++; |
1971 | elv_dispatch_sort(q, rq); | 1975 | elv_dispatch_sort(q, rq); |
1972 | 1976 | ||
1973 | cfqd->rq_in_flight[cfq_cfqq_sync(cfqq)]++; | 1977 | cfqd->rq_in_flight[cfq_cfqq_sync(cfqq)]++; |
1978 | cfqq->nr_sectors += blk_rq_sectors(rq); | ||
1974 | cfq_blkiocg_update_dispatch_stats(&cfqq->cfqg->blkg, blk_rq_bytes(rq), | 1979 | cfq_blkiocg_update_dispatch_stats(&cfqq->cfqg->blkg, blk_rq_bytes(rq), |
1975 | rq_data_dir(rq), rq_is_sync(rq)); | 1980 | rq_data_dir(rq), rq_is_sync(rq)); |
1976 | } | 1981 | } |
1977 | 1982 | ||
1978 | /* | 1983 | /* |
1979 | * return expired entry, or NULL to just start from scratch in rbtree | 1984 | * return expired entry, or NULL to just start from scratch in rbtree |
1980 | */ | 1985 | */ |
1981 | static struct request *cfq_check_fifo(struct cfq_queue *cfqq) | 1986 | static struct request *cfq_check_fifo(struct cfq_queue *cfqq) |
1982 | { | 1987 | { |
1983 | struct request *rq = NULL; | 1988 | struct request *rq = NULL; |
1984 | 1989 | ||
1985 | if (cfq_cfqq_fifo_expire(cfqq)) | 1990 | if (cfq_cfqq_fifo_expire(cfqq)) |
1986 | return NULL; | 1991 | return NULL; |
1987 | 1992 | ||
1988 | cfq_mark_cfqq_fifo_expire(cfqq); | 1993 | cfq_mark_cfqq_fifo_expire(cfqq); |
1989 | 1994 | ||
1990 | if (list_empty(&cfqq->fifo)) | 1995 | if (list_empty(&cfqq->fifo)) |
1991 | return NULL; | 1996 | return NULL; |
1992 | 1997 | ||
1993 | rq = rq_entry_fifo(cfqq->fifo.next); | 1998 | rq = rq_entry_fifo(cfqq->fifo.next); |
1994 | if (time_before(jiffies, rq_fifo_time(rq))) | 1999 | if (time_before(jiffies, rq_fifo_time(rq))) |
1995 | rq = NULL; | 2000 | rq = NULL; |
1996 | 2001 | ||
1997 | cfq_log_cfqq(cfqq->cfqd, cfqq, "fifo=%p", rq); | 2002 | cfq_log_cfqq(cfqq->cfqd, cfqq, "fifo=%p", rq); |
1998 | return rq; | 2003 | return rq; |
1999 | } | 2004 | } |
2000 | 2005 | ||
2001 | static inline int | 2006 | static inline int |
2002 | cfq_prio_to_maxrq(struct cfq_data *cfqd, struct cfq_queue *cfqq) | 2007 | cfq_prio_to_maxrq(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
2003 | { | 2008 | { |
2004 | const int base_rq = cfqd->cfq_slice_async_rq; | 2009 | const int base_rq = cfqd->cfq_slice_async_rq; |
2005 | 2010 | ||
2006 | WARN_ON(cfqq->ioprio >= IOPRIO_BE_NR); | 2011 | WARN_ON(cfqq->ioprio >= IOPRIO_BE_NR); |
2007 | 2012 | ||
2008 | return 2 * (base_rq + base_rq * (CFQ_PRIO_LISTS - 1 - cfqq->ioprio)); | 2013 | return 2 * (base_rq + base_rq * (CFQ_PRIO_LISTS - 1 - cfqq->ioprio)); |
2009 | } | 2014 | } |
2010 | 2015 | ||
2011 | /* | 2016 | /* |
2012 | * Must be called with the queue_lock held. | 2017 | * Must be called with the queue_lock held. |
2013 | */ | 2018 | */ |
2014 | static int cfqq_process_refs(struct cfq_queue *cfqq) | 2019 | static int cfqq_process_refs(struct cfq_queue *cfqq) |
2015 | { | 2020 | { |
2016 | int process_refs, io_refs; | 2021 | int process_refs, io_refs; |
2017 | 2022 | ||
2018 | io_refs = cfqq->allocated[READ] + cfqq->allocated[WRITE]; | 2023 | io_refs = cfqq->allocated[READ] + cfqq->allocated[WRITE]; |
2019 | process_refs = atomic_read(&cfqq->ref) - io_refs; | 2024 | process_refs = atomic_read(&cfqq->ref) - io_refs; |
2020 | BUG_ON(process_refs < 0); | 2025 | BUG_ON(process_refs < 0); |
2021 | return process_refs; | 2026 | return process_refs; |
2022 | } | 2027 | } |
2023 | 2028 | ||
2024 | static void cfq_setup_merge(struct cfq_queue *cfqq, struct cfq_queue *new_cfqq) | 2029 | static void cfq_setup_merge(struct cfq_queue *cfqq, struct cfq_queue *new_cfqq) |
2025 | { | 2030 | { |
2026 | int process_refs, new_process_refs; | 2031 | int process_refs, new_process_refs; |
2027 | struct cfq_queue *__cfqq; | 2032 | struct cfq_queue *__cfqq; |
2028 | 2033 | ||
2029 | /* | 2034 | /* |
2030 | * If there are no process references on the new_cfqq, then it is | 2035 | * If there are no process references on the new_cfqq, then it is |
2031 | * unsafe to follow the ->new_cfqq chain as other cfqq's in the | 2036 | * unsafe to follow the ->new_cfqq chain as other cfqq's in the |
2032 | * chain may have dropped their last reference (not just their | 2037 | * chain may have dropped their last reference (not just their |
2033 | * last process reference). | 2038 | * last process reference). |
2034 | */ | 2039 | */ |
2035 | if (!cfqq_process_refs(new_cfqq)) | 2040 | if (!cfqq_process_refs(new_cfqq)) |
2036 | return; | 2041 | return; |
2037 | 2042 | ||
2038 | /* Avoid a circular list and skip interim queue merges */ | 2043 | /* Avoid a circular list and skip interim queue merges */ |
2039 | while ((__cfqq = new_cfqq->new_cfqq)) { | 2044 | while ((__cfqq = new_cfqq->new_cfqq)) { |
2040 | if (__cfqq == cfqq) | 2045 | if (__cfqq == cfqq) |
2041 | return; | 2046 | return; |
2042 | new_cfqq = __cfqq; | 2047 | new_cfqq = __cfqq; |
2043 | } | 2048 | } |
2044 | 2049 | ||
2045 | process_refs = cfqq_process_refs(cfqq); | 2050 | process_refs = cfqq_process_refs(cfqq); |
2046 | new_process_refs = cfqq_process_refs(new_cfqq); | 2051 | new_process_refs = cfqq_process_refs(new_cfqq); |
2047 | /* | 2052 | /* |
2048 | * If the process for the cfqq has gone away, there is no | 2053 | * If the process for the cfqq has gone away, there is no |
2049 | * sense in merging the queues. | 2054 | * sense in merging the queues. |
2050 | */ | 2055 | */ |
2051 | if (process_refs == 0 || new_process_refs == 0) | 2056 | if (process_refs == 0 || new_process_refs == 0) |
2052 | return; | 2057 | return; |
2053 | 2058 | ||
2054 | /* | 2059 | /* |
2055 | * Merge in the direction of the lesser amount of work. | 2060 | * Merge in the direction of the lesser amount of work. |
2056 | */ | 2061 | */ |
2057 | if (new_process_refs >= process_refs) { | 2062 | if (new_process_refs >= process_refs) { |
2058 | cfqq->new_cfqq = new_cfqq; | 2063 | cfqq->new_cfqq = new_cfqq; |
2059 | atomic_add(process_refs, &new_cfqq->ref); | 2064 | atomic_add(process_refs, &new_cfqq->ref); |
2060 | } else { | 2065 | } else { |
2061 | new_cfqq->new_cfqq = cfqq; | 2066 | new_cfqq->new_cfqq = cfqq; |
2062 | atomic_add(new_process_refs, &cfqq->ref); | 2067 | atomic_add(new_process_refs, &cfqq->ref); |
2063 | } | 2068 | } |
2064 | } | 2069 | } |
2065 | 2070 | ||
2066 | static enum wl_type_t cfq_choose_wl(struct cfq_data *cfqd, | 2071 | static enum wl_type_t cfq_choose_wl(struct cfq_data *cfqd, |
2067 | struct cfq_group *cfqg, enum wl_prio_t prio) | 2072 | struct cfq_group *cfqg, enum wl_prio_t prio) |
2068 | { | 2073 | { |
2069 | struct cfq_queue *queue; | 2074 | struct cfq_queue *queue; |
2070 | int i; | 2075 | int i; |
2071 | bool key_valid = false; | 2076 | bool key_valid = false; |
2072 | unsigned long lowest_key = 0; | 2077 | unsigned long lowest_key = 0; |
2073 | enum wl_type_t cur_best = SYNC_NOIDLE_WORKLOAD; | 2078 | enum wl_type_t cur_best = SYNC_NOIDLE_WORKLOAD; |
2074 | 2079 | ||
2075 | for (i = 0; i <= SYNC_WORKLOAD; ++i) { | 2080 | for (i = 0; i <= SYNC_WORKLOAD; ++i) { |
2076 | /* select the one with lowest rb_key */ | 2081 | /* select the one with lowest rb_key */ |
2077 | queue = cfq_rb_first(service_tree_for(cfqg, prio, i)); | 2082 | queue = cfq_rb_first(service_tree_for(cfqg, prio, i)); |
2078 | if (queue && | 2083 | if (queue && |
2079 | (!key_valid || time_before(queue->rb_key, lowest_key))) { | 2084 | (!key_valid || time_before(queue->rb_key, lowest_key))) { |
2080 | lowest_key = queue->rb_key; | 2085 | lowest_key = queue->rb_key; |
2081 | cur_best = i; | 2086 | cur_best = i; |
2082 | key_valid = true; | 2087 | key_valid = true; |
2083 | } | 2088 | } |
2084 | } | 2089 | } |
2085 | 2090 | ||
2086 | return cur_best; | 2091 | return cur_best; |
2087 | } | 2092 | } |
2088 | 2093 | ||
2089 | static void choose_service_tree(struct cfq_data *cfqd, struct cfq_group *cfqg) | 2094 | static void choose_service_tree(struct cfq_data *cfqd, struct cfq_group *cfqg) |
2090 | { | 2095 | { |
2091 | unsigned slice; | 2096 | unsigned slice; |
2092 | unsigned count; | 2097 | unsigned count; |
2093 | struct cfq_rb_root *st; | 2098 | struct cfq_rb_root *st; |
2094 | unsigned group_slice; | 2099 | unsigned group_slice; |
2095 | 2100 | ||
2096 | if (!cfqg) { | 2101 | if (!cfqg) { |
2097 | cfqd->serving_prio = IDLE_WORKLOAD; | 2102 | cfqd->serving_prio = IDLE_WORKLOAD; |
2098 | cfqd->workload_expires = jiffies + 1; | 2103 | cfqd->workload_expires = jiffies + 1; |
2099 | return; | 2104 | return; |
2100 | } | 2105 | } |
2101 | 2106 | ||
2102 | /* Choose next priority. RT > BE > IDLE */ | 2107 | /* Choose next priority. RT > BE > IDLE */ |
2103 | if (cfq_group_busy_queues_wl(RT_WORKLOAD, cfqd, cfqg)) | 2108 | if (cfq_group_busy_queues_wl(RT_WORKLOAD, cfqd, cfqg)) |
2104 | cfqd->serving_prio = RT_WORKLOAD; | 2109 | cfqd->serving_prio = RT_WORKLOAD; |
2105 | else if (cfq_group_busy_queues_wl(BE_WORKLOAD, cfqd, cfqg)) | 2110 | else if (cfq_group_busy_queues_wl(BE_WORKLOAD, cfqd, cfqg)) |
2106 | cfqd->serving_prio = BE_WORKLOAD; | 2111 | cfqd->serving_prio = BE_WORKLOAD; |
2107 | else { | 2112 | else { |
2108 | cfqd->serving_prio = IDLE_WORKLOAD; | 2113 | cfqd->serving_prio = IDLE_WORKLOAD; |
2109 | cfqd->workload_expires = jiffies + 1; | 2114 | cfqd->workload_expires = jiffies + 1; |
2110 | return; | 2115 | return; |
2111 | } | 2116 | } |
2112 | 2117 | ||
2113 | /* | 2118 | /* |
2114 | * For RT and BE, we have to choose also the type | 2119 | * For RT and BE, we have to choose also the type |
2115 | * (SYNC, SYNC_NOIDLE, ASYNC), and to compute a workload | 2120 | * (SYNC, SYNC_NOIDLE, ASYNC), and to compute a workload |
2116 | * expiration time | 2121 | * expiration time |
2117 | */ | 2122 | */ |
2118 | st = service_tree_for(cfqg, cfqd->serving_prio, cfqd->serving_type); | 2123 | st = service_tree_for(cfqg, cfqd->serving_prio, cfqd->serving_type); |
2119 | count = st->count; | 2124 | count = st->count; |
2120 | 2125 | ||
2121 | /* | 2126 | /* |
2122 | * check workload expiration, and that we still have other queues ready | 2127 | * check workload expiration, and that we still have other queues ready |
2123 | */ | 2128 | */ |
2124 | if (count && !time_after(jiffies, cfqd->workload_expires)) | 2129 | if (count && !time_after(jiffies, cfqd->workload_expires)) |
2125 | return; | 2130 | return; |
2126 | 2131 | ||
2127 | /* otherwise select new workload type */ | 2132 | /* otherwise select new workload type */ |
2128 | cfqd->serving_type = | 2133 | cfqd->serving_type = |
2129 | cfq_choose_wl(cfqd, cfqg, cfqd->serving_prio); | 2134 | cfq_choose_wl(cfqd, cfqg, cfqd->serving_prio); |
2130 | st = service_tree_for(cfqg, cfqd->serving_prio, cfqd->serving_type); | 2135 | st = service_tree_for(cfqg, cfqd->serving_prio, cfqd->serving_type); |
2131 | count = st->count; | 2136 | count = st->count; |
2132 | 2137 | ||
2133 | /* | 2138 | /* |
2134 | * the workload slice is computed as a fraction of target latency | 2139 | * the workload slice is computed as a fraction of target latency |
2135 | * proportional to the number of queues in that workload, over | 2140 | * proportional to the number of queues in that workload, over |
2136 | * all the queues in the same priority class | 2141 | * all the queues in the same priority class |
2137 | */ | 2142 | */ |
2138 | group_slice = cfq_group_slice(cfqd, cfqg); | 2143 | group_slice = cfq_group_slice(cfqd, cfqg); |
2139 | 2144 | ||
2140 | slice = group_slice * count / | 2145 | slice = group_slice * count / |
2141 | max_t(unsigned, cfqg->busy_queues_avg[cfqd->serving_prio], | 2146 | max_t(unsigned, cfqg->busy_queues_avg[cfqd->serving_prio], |
2142 | cfq_group_busy_queues_wl(cfqd->serving_prio, cfqd, cfqg)); | 2147 | cfq_group_busy_queues_wl(cfqd->serving_prio, cfqd, cfqg)); |
2143 | 2148 | ||
2144 | if (cfqd->serving_type == ASYNC_WORKLOAD) { | 2149 | if (cfqd->serving_type == ASYNC_WORKLOAD) { |
2145 | unsigned int tmp; | 2150 | unsigned int tmp; |
2146 | 2151 | ||
2147 | /* | 2152 | /* |
2148 | * Async queues are currently system wide. Just taking | 2153 | * Async queues are currently system wide. Just taking |
2149 | * proportion of queues with-in same group will lead to higher | 2154 | * proportion of queues with-in same group will lead to higher |
2150 | * async ratio system wide as generally root group is going | 2155 | * async ratio system wide as generally root group is going |
2151 | * to have higher weight. A more accurate thing would be to | 2156 | * to have higher weight. A more accurate thing would be to |
2152 | * calculate system wide asnc/sync ratio. | 2157 | * calculate system wide asnc/sync ratio. |
2153 | */ | 2158 | */ |
2154 | tmp = cfq_target_latency * cfqg_busy_async_queues(cfqd, cfqg); | 2159 | tmp = cfq_target_latency * cfqg_busy_async_queues(cfqd, cfqg); |
2155 | tmp = tmp/cfqd->busy_queues; | 2160 | tmp = tmp/cfqd->busy_queues; |
2156 | slice = min_t(unsigned, slice, tmp); | 2161 | slice = min_t(unsigned, slice, tmp); |
2157 | 2162 | ||
2158 | /* async workload slice is scaled down according to | 2163 | /* async workload slice is scaled down according to |
2159 | * the sync/async slice ratio. */ | 2164 | * the sync/async slice ratio. */ |
2160 | slice = slice * cfqd->cfq_slice[0] / cfqd->cfq_slice[1]; | 2165 | slice = slice * cfqd->cfq_slice[0] / cfqd->cfq_slice[1]; |
2161 | } else | 2166 | } else |
2162 | /* sync workload slice is at least 2 * cfq_slice_idle */ | 2167 | /* sync workload slice is at least 2 * cfq_slice_idle */ |
2163 | slice = max(slice, 2 * cfqd->cfq_slice_idle); | 2168 | slice = max(slice, 2 * cfqd->cfq_slice_idle); |
2164 | 2169 | ||
2165 | slice = max_t(unsigned, slice, CFQ_MIN_TT); | 2170 | slice = max_t(unsigned, slice, CFQ_MIN_TT); |
2166 | cfq_log(cfqd, "workload slice:%d", slice); | 2171 | cfq_log(cfqd, "workload slice:%d", slice); |
2167 | cfqd->workload_expires = jiffies + slice; | 2172 | cfqd->workload_expires = jiffies + slice; |
2168 | cfqd->noidle_tree_requires_idle = false; | 2173 | cfqd->noidle_tree_requires_idle = false; |
2169 | } | 2174 | } |
2170 | 2175 | ||
2171 | static struct cfq_group *cfq_get_next_cfqg(struct cfq_data *cfqd) | 2176 | static struct cfq_group *cfq_get_next_cfqg(struct cfq_data *cfqd) |
2172 | { | 2177 | { |
2173 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | 2178 | struct cfq_rb_root *st = &cfqd->grp_service_tree; |
2174 | struct cfq_group *cfqg; | 2179 | struct cfq_group *cfqg; |
2175 | 2180 | ||
2176 | if (RB_EMPTY_ROOT(&st->rb)) | 2181 | if (RB_EMPTY_ROOT(&st->rb)) |
2177 | return NULL; | 2182 | return NULL; |
2178 | cfqg = cfq_rb_first_group(st); | 2183 | cfqg = cfq_rb_first_group(st); |
2179 | st->active = &cfqg->rb_node; | 2184 | st->active = &cfqg->rb_node; |
2180 | update_min_vdisktime(st); | 2185 | update_min_vdisktime(st); |
2181 | return cfqg; | 2186 | return cfqg; |
2182 | } | 2187 | } |
2183 | 2188 | ||
2184 | static void cfq_choose_cfqg(struct cfq_data *cfqd) | 2189 | static void cfq_choose_cfqg(struct cfq_data *cfqd) |
2185 | { | 2190 | { |
2186 | struct cfq_group *cfqg = cfq_get_next_cfqg(cfqd); | 2191 | struct cfq_group *cfqg = cfq_get_next_cfqg(cfqd); |
2187 | 2192 | ||
2188 | cfqd->serving_group = cfqg; | 2193 | cfqd->serving_group = cfqg; |
2189 | 2194 | ||
2190 | /* Restore the workload type data */ | 2195 | /* Restore the workload type data */ |
2191 | if (cfqg->saved_workload_slice) { | 2196 | if (cfqg->saved_workload_slice) { |
2192 | cfqd->workload_expires = jiffies + cfqg->saved_workload_slice; | 2197 | cfqd->workload_expires = jiffies + cfqg->saved_workload_slice; |
2193 | cfqd->serving_type = cfqg->saved_workload; | 2198 | cfqd->serving_type = cfqg->saved_workload; |
2194 | cfqd->serving_prio = cfqg->saved_serving_prio; | 2199 | cfqd->serving_prio = cfqg->saved_serving_prio; |
2195 | } else | 2200 | } else |
2196 | cfqd->workload_expires = jiffies - 1; | 2201 | cfqd->workload_expires = jiffies - 1; |
2197 | 2202 | ||
2198 | choose_service_tree(cfqd, cfqg); | 2203 | choose_service_tree(cfqd, cfqg); |
2199 | } | 2204 | } |
2200 | 2205 | ||
2201 | /* | 2206 | /* |
2202 | * Select a queue for service. If we have a current active queue, | 2207 | * Select a queue for service. If we have a current active queue, |
2203 | * check whether to continue servicing it, or retrieve and set a new one. | 2208 | * check whether to continue servicing it, or retrieve and set a new one. |
2204 | */ | 2209 | */ |
2205 | static struct cfq_queue *cfq_select_queue(struct cfq_data *cfqd) | 2210 | static struct cfq_queue *cfq_select_queue(struct cfq_data *cfqd) |
2206 | { | 2211 | { |
2207 | struct cfq_queue *cfqq, *new_cfqq = NULL; | 2212 | struct cfq_queue *cfqq, *new_cfqq = NULL; |
2208 | 2213 | ||
2209 | cfqq = cfqd->active_queue; | 2214 | cfqq = cfqd->active_queue; |
2210 | if (!cfqq) | 2215 | if (!cfqq) |
2211 | goto new_queue; | 2216 | goto new_queue; |
2212 | 2217 | ||
2213 | if (!cfqd->rq_queued) | 2218 | if (!cfqd->rq_queued) |
2214 | return NULL; | 2219 | return NULL; |
2215 | 2220 | ||
2216 | /* | 2221 | /* |
2217 | * We were waiting for group to get backlogged. Expire the queue | 2222 | * We were waiting for group to get backlogged. Expire the queue |
2218 | */ | 2223 | */ |
2219 | if (cfq_cfqq_wait_busy(cfqq) && !RB_EMPTY_ROOT(&cfqq->sort_list)) | 2224 | if (cfq_cfqq_wait_busy(cfqq) && !RB_EMPTY_ROOT(&cfqq->sort_list)) |
2220 | goto expire; | 2225 | goto expire; |
2221 | 2226 | ||
2222 | /* | 2227 | /* |
2223 | * The active queue has run out of time, expire it and select new. | 2228 | * The active queue has run out of time, expire it and select new. |
2224 | */ | 2229 | */ |
2225 | if (cfq_slice_used(cfqq) && !cfq_cfqq_must_dispatch(cfqq)) { | 2230 | if (cfq_slice_used(cfqq) && !cfq_cfqq_must_dispatch(cfqq)) { |
2226 | /* | 2231 | /* |
2227 | * If slice had not expired at the completion of last request | 2232 | * If slice had not expired at the completion of last request |
2228 | * we might not have turned on wait_busy flag. Don't expire | 2233 | * we might not have turned on wait_busy flag. Don't expire |
2229 | * the queue yet. Allow the group to get backlogged. | 2234 | * the queue yet. Allow the group to get backlogged. |
2230 | * | 2235 | * |
2231 | * The very fact that we have used the slice, that means we | 2236 | * The very fact that we have used the slice, that means we |
2232 | * have been idling all along on this queue and it should be | 2237 | * have been idling all along on this queue and it should be |
2233 | * ok to wait for this request to complete. | 2238 | * ok to wait for this request to complete. |
2234 | */ | 2239 | */ |
2235 | if (cfqq->cfqg->nr_cfqq == 1 && RB_EMPTY_ROOT(&cfqq->sort_list) | 2240 | if (cfqq->cfqg->nr_cfqq == 1 && RB_EMPTY_ROOT(&cfqq->sort_list) |
2236 | && cfqq->dispatched && cfq_should_idle(cfqd, cfqq)) { | 2241 | && cfqq->dispatched && cfq_should_idle(cfqd, cfqq)) { |
2237 | cfqq = NULL; | 2242 | cfqq = NULL; |
2238 | goto keep_queue; | 2243 | goto keep_queue; |
2239 | } else | 2244 | } else |
2240 | goto check_group_idle; | 2245 | goto check_group_idle; |
2241 | } | 2246 | } |
2242 | 2247 | ||
2243 | /* | 2248 | /* |
2244 | * The active queue has requests and isn't expired, allow it to | 2249 | * The active queue has requests and isn't expired, allow it to |
2245 | * dispatch. | 2250 | * dispatch. |
2246 | */ | 2251 | */ |
2247 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) | 2252 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) |
2248 | goto keep_queue; | 2253 | goto keep_queue; |
2249 | 2254 | ||
2250 | /* | 2255 | /* |
2251 | * If another queue has a request waiting within our mean seek | 2256 | * If another queue has a request waiting within our mean seek |
2252 | * distance, let it run. The expire code will check for close | 2257 | * distance, let it run. The expire code will check for close |
2253 | * cooperators and put the close queue at the front of the service | 2258 | * cooperators and put the close queue at the front of the service |
2254 | * tree. If possible, merge the expiring queue with the new cfqq. | 2259 | * tree. If possible, merge the expiring queue with the new cfqq. |
2255 | */ | 2260 | */ |
2256 | new_cfqq = cfq_close_cooperator(cfqd, cfqq); | 2261 | new_cfqq = cfq_close_cooperator(cfqd, cfqq); |
2257 | if (new_cfqq) { | 2262 | if (new_cfqq) { |
2258 | if (!cfqq->new_cfqq) | 2263 | if (!cfqq->new_cfqq) |
2259 | cfq_setup_merge(cfqq, new_cfqq); | 2264 | cfq_setup_merge(cfqq, new_cfqq); |
2260 | goto expire; | 2265 | goto expire; |
2261 | } | 2266 | } |
2262 | 2267 | ||
2263 | /* | 2268 | /* |
2264 | * No requests pending. If the active queue still has requests in | 2269 | * No requests pending. If the active queue still has requests in |
2265 | * flight or is idling for a new request, allow either of these | 2270 | * flight or is idling for a new request, allow either of these |
2266 | * conditions to happen (or time out) before selecting a new queue. | 2271 | * conditions to happen (or time out) before selecting a new queue. |
2267 | */ | 2272 | */ |
2268 | if (timer_pending(&cfqd->idle_slice_timer)) { | 2273 | if (timer_pending(&cfqd->idle_slice_timer)) { |
2269 | cfqq = NULL; | 2274 | cfqq = NULL; |
2270 | goto keep_queue; | 2275 | goto keep_queue; |
2271 | } | 2276 | } |
2272 | 2277 | ||
2273 | if (cfqq->dispatched && cfq_should_idle(cfqd, cfqq)) { | 2278 | if (cfqq->dispatched && cfq_should_idle(cfqd, cfqq)) { |
2274 | cfqq = NULL; | 2279 | cfqq = NULL; |
2275 | goto keep_queue; | 2280 | goto keep_queue; |
2276 | } | 2281 | } |
2277 | 2282 | ||
2278 | /* | 2283 | /* |
2279 | * If group idle is enabled and there are requests dispatched from | 2284 | * If group idle is enabled and there are requests dispatched from |
2280 | * this group, wait for requests to complete. | 2285 | * this group, wait for requests to complete. |
2281 | */ | 2286 | */ |
2282 | check_group_idle: | 2287 | check_group_idle: |
2283 | if (cfqd->cfq_group_idle && cfqq->cfqg->nr_cfqq == 1 | 2288 | if (cfqd->cfq_group_idle && cfqq->cfqg->nr_cfqq == 1 |
2284 | && cfqq->cfqg->dispatched) { | 2289 | && cfqq->cfqg->dispatched) { |
2285 | cfqq = NULL; | 2290 | cfqq = NULL; |
2286 | goto keep_queue; | 2291 | goto keep_queue; |
2287 | } | 2292 | } |
2288 | 2293 | ||
2289 | expire: | 2294 | expire: |
2290 | cfq_slice_expired(cfqd, 0); | 2295 | cfq_slice_expired(cfqd, 0); |
2291 | new_queue: | 2296 | new_queue: |
2292 | /* | 2297 | /* |
2293 | * Current queue expired. Check if we have to switch to a new | 2298 | * Current queue expired. Check if we have to switch to a new |
2294 | * service tree | 2299 | * service tree |
2295 | */ | 2300 | */ |
2296 | if (!new_cfqq) | 2301 | if (!new_cfqq) |
2297 | cfq_choose_cfqg(cfqd); | 2302 | cfq_choose_cfqg(cfqd); |
2298 | 2303 | ||
2299 | cfqq = cfq_set_active_queue(cfqd, new_cfqq); | 2304 | cfqq = cfq_set_active_queue(cfqd, new_cfqq); |
2300 | keep_queue: | 2305 | keep_queue: |
2301 | return cfqq; | 2306 | return cfqq; |
2302 | } | 2307 | } |
2303 | 2308 | ||
2304 | static int __cfq_forced_dispatch_cfqq(struct cfq_queue *cfqq) | 2309 | static int __cfq_forced_dispatch_cfqq(struct cfq_queue *cfqq) |
2305 | { | 2310 | { |
2306 | int dispatched = 0; | 2311 | int dispatched = 0; |
2307 | 2312 | ||
2308 | while (cfqq->next_rq) { | 2313 | while (cfqq->next_rq) { |
2309 | cfq_dispatch_insert(cfqq->cfqd->queue, cfqq->next_rq); | 2314 | cfq_dispatch_insert(cfqq->cfqd->queue, cfqq->next_rq); |
2310 | dispatched++; | 2315 | dispatched++; |
2311 | } | 2316 | } |
2312 | 2317 | ||
2313 | BUG_ON(!list_empty(&cfqq->fifo)); | 2318 | BUG_ON(!list_empty(&cfqq->fifo)); |
2314 | 2319 | ||
2315 | /* By default cfqq is not expired if it is empty. Do it explicitly */ | 2320 | /* By default cfqq is not expired if it is empty. Do it explicitly */ |
2316 | __cfq_slice_expired(cfqq->cfqd, cfqq, 0); | 2321 | __cfq_slice_expired(cfqq->cfqd, cfqq, 0); |
2317 | return dispatched; | 2322 | return dispatched; |
2318 | } | 2323 | } |
2319 | 2324 | ||
2320 | /* | 2325 | /* |
2321 | * Drain our current requests. Used for barriers and when switching | 2326 | * Drain our current requests. Used for barriers and when switching |
2322 | * io schedulers on-the-fly. | 2327 | * io schedulers on-the-fly. |
2323 | */ | 2328 | */ |
2324 | static int cfq_forced_dispatch(struct cfq_data *cfqd) | 2329 | static int cfq_forced_dispatch(struct cfq_data *cfqd) |
2325 | { | 2330 | { |
2326 | struct cfq_queue *cfqq; | 2331 | struct cfq_queue *cfqq; |
2327 | int dispatched = 0; | 2332 | int dispatched = 0; |
2328 | 2333 | ||
2329 | /* Expire the timeslice of the current active queue first */ | 2334 | /* Expire the timeslice of the current active queue first */ |
2330 | cfq_slice_expired(cfqd, 0); | 2335 | cfq_slice_expired(cfqd, 0); |
2331 | while ((cfqq = cfq_get_next_queue_forced(cfqd)) != NULL) { | 2336 | while ((cfqq = cfq_get_next_queue_forced(cfqd)) != NULL) { |
2332 | __cfq_set_active_queue(cfqd, cfqq); | 2337 | __cfq_set_active_queue(cfqd, cfqq); |
2333 | dispatched += __cfq_forced_dispatch_cfqq(cfqq); | 2338 | dispatched += __cfq_forced_dispatch_cfqq(cfqq); |
2334 | } | 2339 | } |
2335 | 2340 | ||
2336 | BUG_ON(cfqd->busy_queues); | 2341 | BUG_ON(cfqd->busy_queues); |
2337 | 2342 | ||
2338 | cfq_log(cfqd, "forced_dispatch=%d", dispatched); | 2343 | cfq_log(cfqd, "forced_dispatch=%d", dispatched); |
2339 | return dispatched; | 2344 | return dispatched; |
2340 | } | 2345 | } |
2341 | 2346 | ||
2342 | static inline bool cfq_slice_used_soon(struct cfq_data *cfqd, | 2347 | static inline bool cfq_slice_used_soon(struct cfq_data *cfqd, |
2343 | struct cfq_queue *cfqq) | 2348 | struct cfq_queue *cfqq) |
2344 | { | 2349 | { |
2345 | /* the queue hasn't finished any request, can't estimate */ | 2350 | /* the queue hasn't finished any request, can't estimate */ |
2346 | if (cfq_cfqq_slice_new(cfqq)) | 2351 | if (cfq_cfqq_slice_new(cfqq)) |
2347 | return 1; | 2352 | return 1; |
2348 | if (time_after(jiffies + cfqd->cfq_slice_idle * cfqq->dispatched, | 2353 | if (time_after(jiffies + cfqd->cfq_slice_idle * cfqq->dispatched, |
2349 | cfqq->slice_end)) | 2354 | cfqq->slice_end)) |
2350 | return 1; | 2355 | return 1; |
2351 | 2356 | ||
2352 | return 0; | 2357 | return 0; |
2353 | } | 2358 | } |
2354 | 2359 | ||
2355 | static bool cfq_may_dispatch(struct cfq_data *cfqd, struct cfq_queue *cfqq) | 2360 | static bool cfq_may_dispatch(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
2356 | { | 2361 | { |
2357 | unsigned int max_dispatch; | 2362 | unsigned int max_dispatch; |
2358 | 2363 | ||
2359 | /* | 2364 | /* |
2360 | * Drain async requests before we start sync IO | 2365 | * Drain async requests before we start sync IO |
2361 | */ | 2366 | */ |
2362 | if (cfq_should_idle(cfqd, cfqq) && cfqd->rq_in_flight[BLK_RW_ASYNC]) | 2367 | if (cfq_should_idle(cfqd, cfqq) && cfqd->rq_in_flight[BLK_RW_ASYNC]) |
2363 | return false; | 2368 | return false; |
2364 | 2369 | ||
2365 | /* | 2370 | /* |
2366 | * If this is an async queue and we have sync IO in flight, let it wait | 2371 | * If this is an async queue and we have sync IO in flight, let it wait |
2367 | */ | 2372 | */ |
2368 | if (cfqd->rq_in_flight[BLK_RW_SYNC] && !cfq_cfqq_sync(cfqq)) | 2373 | if (cfqd->rq_in_flight[BLK_RW_SYNC] && !cfq_cfqq_sync(cfqq)) |
2369 | return false; | 2374 | return false; |
2370 | 2375 | ||
2371 | max_dispatch = max_t(unsigned int, cfqd->cfq_quantum / 2, 1); | 2376 | max_dispatch = max_t(unsigned int, cfqd->cfq_quantum / 2, 1); |
2372 | if (cfq_class_idle(cfqq)) | 2377 | if (cfq_class_idle(cfqq)) |
2373 | max_dispatch = 1; | 2378 | max_dispatch = 1; |
2374 | 2379 | ||
2375 | /* | 2380 | /* |
2376 | * Does this cfqq already have too much IO in flight? | 2381 | * Does this cfqq already have too much IO in flight? |
2377 | */ | 2382 | */ |
2378 | if (cfqq->dispatched >= max_dispatch) { | 2383 | if (cfqq->dispatched >= max_dispatch) { |
2379 | /* | 2384 | /* |
2380 | * idle queue must always only have a single IO in flight | 2385 | * idle queue must always only have a single IO in flight |
2381 | */ | 2386 | */ |
2382 | if (cfq_class_idle(cfqq)) | 2387 | if (cfq_class_idle(cfqq)) |
2383 | return false; | 2388 | return false; |
2384 | 2389 | ||
2385 | /* | 2390 | /* |
2386 | * We have other queues, don't allow more IO from this one | 2391 | * We have other queues, don't allow more IO from this one |
2387 | */ | 2392 | */ |
2388 | if (cfqd->busy_queues > 1 && cfq_slice_used_soon(cfqd, cfqq)) | 2393 | if (cfqd->busy_queues > 1 && cfq_slice_used_soon(cfqd, cfqq)) |
2389 | return false; | 2394 | return false; |
2390 | 2395 | ||
2391 | /* | 2396 | /* |
2392 | * Sole queue user, no limit | 2397 | * Sole queue user, no limit |
2393 | */ | 2398 | */ |
2394 | if (cfqd->busy_queues == 1) | 2399 | if (cfqd->busy_queues == 1) |
2395 | max_dispatch = -1; | 2400 | max_dispatch = -1; |
2396 | else | 2401 | else |
2397 | /* | 2402 | /* |
2398 | * Normally we start throttling cfqq when cfq_quantum/2 | 2403 | * Normally we start throttling cfqq when cfq_quantum/2 |
2399 | * requests have been dispatched. But we can drive | 2404 | * requests have been dispatched. But we can drive |
2400 | * deeper queue depths at the beginning of slice | 2405 | * deeper queue depths at the beginning of slice |
2401 | * subjected to upper limit of cfq_quantum. | 2406 | * subjected to upper limit of cfq_quantum. |
2402 | * */ | 2407 | * */ |
2403 | max_dispatch = cfqd->cfq_quantum; | 2408 | max_dispatch = cfqd->cfq_quantum; |
2404 | } | 2409 | } |
2405 | 2410 | ||
2406 | /* | 2411 | /* |
2407 | * Async queues must wait a bit before being allowed dispatch. | 2412 | * Async queues must wait a bit before being allowed dispatch. |
2408 | * We also ramp up the dispatch depth gradually for async IO, | 2413 | * We also ramp up the dispatch depth gradually for async IO, |
2409 | * based on the last sync IO we serviced | 2414 | * based on the last sync IO we serviced |
2410 | */ | 2415 | */ |
2411 | if (!cfq_cfqq_sync(cfqq) && cfqd->cfq_latency) { | 2416 | if (!cfq_cfqq_sync(cfqq) && cfqd->cfq_latency) { |
2412 | unsigned long last_sync = jiffies - cfqd->last_delayed_sync; | 2417 | unsigned long last_sync = jiffies - cfqd->last_delayed_sync; |
2413 | unsigned int depth; | 2418 | unsigned int depth; |
2414 | 2419 | ||
2415 | depth = last_sync / cfqd->cfq_slice[1]; | 2420 | depth = last_sync / cfqd->cfq_slice[1]; |
2416 | if (!depth && !cfqq->dispatched) | 2421 | if (!depth && !cfqq->dispatched) |
2417 | depth = 1; | 2422 | depth = 1; |
2418 | if (depth < max_dispatch) | 2423 | if (depth < max_dispatch) |
2419 | max_dispatch = depth; | 2424 | max_dispatch = depth; |
2420 | } | 2425 | } |
2421 | 2426 | ||
2422 | /* | 2427 | /* |
2423 | * If we're below the current max, allow a dispatch | 2428 | * If we're below the current max, allow a dispatch |
2424 | */ | 2429 | */ |
2425 | return cfqq->dispatched < max_dispatch; | 2430 | return cfqq->dispatched < max_dispatch; |
2426 | } | 2431 | } |
2427 | 2432 | ||
2428 | /* | 2433 | /* |
2429 | * Dispatch a request from cfqq, moving them to the request queue | 2434 | * Dispatch a request from cfqq, moving them to the request queue |
2430 | * dispatch list. | 2435 | * dispatch list. |
2431 | */ | 2436 | */ |
2432 | static bool cfq_dispatch_request(struct cfq_data *cfqd, struct cfq_queue *cfqq) | 2437 | static bool cfq_dispatch_request(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
2433 | { | 2438 | { |
2434 | struct request *rq; | 2439 | struct request *rq; |
2435 | 2440 | ||
2436 | BUG_ON(RB_EMPTY_ROOT(&cfqq->sort_list)); | 2441 | BUG_ON(RB_EMPTY_ROOT(&cfqq->sort_list)); |
2437 | 2442 | ||
2438 | if (!cfq_may_dispatch(cfqd, cfqq)) | 2443 | if (!cfq_may_dispatch(cfqd, cfqq)) |
2439 | return false; | 2444 | return false; |
2440 | 2445 | ||
2441 | /* | 2446 | /* |
2442 | * follow expired path, else get first next available | 2447 | * follow expired path, else get first next available |
2443 | */ | 2448 | */ |
2444 | rq = cfq_check_fifo(cfqq); | 2449 | rq = cfq_check_fifo(cfqq); |
2445 | if (!rq) | 2450 | if (!rq) |
2446 | rq = cfqq->next_rq; | 2451 | rq = cfqq->next_rq; |
2447 | 2452 | ||
2448 | /* | 2453 | /* |
2449 | * insert request into driver dispatch list | 2454 | * insert request into driver dispatch list |
2450 | */ | 2455 | */ |
2451 | cfq_dispatch_insert(cfqd->queue, rq); | 2456 | cfq_dispatch_insert(cfqd->queue, rq); |
2452 | 2457 | ||
2453 | if (!cfqd->active_cic) { | 2458 | if (!cfqd->active_cic) { |
2454 | struct cfq_io_context *cic = RQ_CIC(rq); | 2459 | struct cfq_io_context *cic = RQ_CIC(rq); |
2455 | 2460 | ||
2456 | atomic_long_inc(&cic->ioc->refcount); | 2461 | atomic_long_inc(&cic->ioc->refcount); |
2457 | cfqd->active_cic = cic; | 2462 | cfqd->active_cic = cic; |
2458 | } | 2463 | } |
2459 | 2464 | ||
2460 | return true; | 2465 | return true; |
2461 | } | 2466 | } |
2462 | 2467 | ||
2463 | /* | 2468 | /* |
2464 | * Find the cfqq that we need to service and move a request from that to the | 2469 | * Find the cfqq that we need to service and move a request from that to the |
2465 | * dispatch list | 2470 | * dispatch list |
2466 | */ | 2471 | */ |
2467 | static int cfq_dispatch_requests(struct request_queue *q, int force) | 2472 | static int cfq_dispatch_requests(struct request_queue *q, int force) |
2468 | { | 2473 | { |
2469 | struct cfq_data *cfqd = q->elevator->elevator_data; | 2474 | struct cfq_data *cfqd = q->elevator->elevator_data; |
2470 | struct cfq_queue *cfqq; | 2475 | struct cfq_queue *cfqq; |
2471 | 2476 | ||
2472 | if (!cfqd->busy_queues) | 2477 | if (!cfqd->busy_queues) |
2473 | return 0; | 2478 | return 0; |
2474 | 2479 | ||
2475 | if (unlikely(force)) | 2480 | if (unlikely(force)) |
2476 | return cfq_forced_dispatch(cfqd); | 2481 | return cfq_forced_dispatch(cfqd); |
2477 | 2482 | ||
2478 | cfqq = cfq_select_queue(cfqd); | 2483 | cfqq = cfq_select_queue(cfqd); |
2479 | if (!cfqq) | 2484 | if (!cfqq) |
2480 | return 0; | 2485 | return 0; |
2481 | 2486 | ||
2482 | /* | 2487 | /* |
2483 | * Dispatch a request from this cfqq, if it is allowed | 2488 | * Dispatch a request from this cfqq, if it is allowed |
2484 | */ | 2489 | */ |
2485 | if (!cfq_dispatch_request(cfqd, cfqq)) | 2490 | if (!cfq_dispatch_request(cfqd, cfqq)) |
2486 | return 0; | 2491 | return 0; |
2487 | 2492 | ||
2488 | cfqq->slice_dispatch++; | 2493 | cfqq->slice_dispatch++; |
2489 | cfq_clear_cfqq_must_dispatch(cfqq); | 2494 | cfq_clear_cfqq_must_dispatch(cfqq); |
2490 | 2495 | ||
2491 | /* | 2496 | /* |
2492 | * expire an async queue immediately if it has used up its slice. idle | 2497 | * expire an async queue immediately if it has used up its slice. idle |
2493 | * queue always expire after 1 dispatch round. | 2498 | * queue always expire after 1 dispatch round. |
2494 | */ | 2499 | */ |
2495 | if (cfqd->busy_queues > 1 && ((!cfq_cfqq_sync(cfqq) && | 2500 | if (cfqd->busy_queues > 1 && ((!cfq_cfqq_sync(cfqq) && |
2496 | cfqq->slice_dispatch >= cfq_prio_to_maxrq(cfqd, cfqq)) || | 2501 | cfqq->slice_dispatch >= cfq_prio_to_maxrq(cfqd, cfqq)) || |
2497 | cfq_class_idle(cfqq))) { | 2502 | cfq_class_idle(cfqq))) { |
2498 | cfqq->slice_end = jiffies + 1; | 2503 | cfqq->slice_end = jiffies + 1; |
2499 | cfq_slice_expired(cfqd, 0); | 2504 | cfq_slice_expired(cfqd, 0); |
2500 | } | 2505 | } |
2501 | 2506 | ||
2502 | cfq_log_cfqq(cfqd, cfqq, "dispatched a request"); | 2507 | cfq_log_cfqq(cfqd, cfqq, "dispatched a request"); |
2503 | return 1; | 2508 | return 1; |
2504 | } | 2509 | } |
2505 | 2510 | ||
2506 | /* | 2511 | /* |
2507 | * task holds one reference to the queue, dropped when task exits. each rq | 2512 | * task holds one reference to the queue, dropped when task exits. each rq |
2508 | * in-flight on this queue also holds a reference, dropped when rq is freed. | 2513 | * in-flight on this queue also holds a reference, dropped when rq is freed. |
2509 | * | 2514 | * |
2510 | * Each cfq queue took a reference on the parent group. Drop it now. | 2515 | * Each cfq queue took a reference on the parent group. Drop it now. |
2511 | * queue lock must be held here. | 2516 | * queue lock must be held here. |
2512 | */ | 2517 | */ |
2513 | static void cfq_put_queue(struct cfq_queue *cfqq) | 2518 | static void cfq_put_queue(struct cfq_queue *cfqq) |
2514 | { | 2519 | { |
2515 | struct cfq_data *cfqd = cfqq->cfqd; | 2520 | struct cfq_data *cfqd = cfqq->cfqd; |
2516 | struct cfq_group *cfqg, *orig_cfqg; | 2521 | struct cfq_group *cfqg, *orig_cfqg; |
2517 | 2522 | ||
2518 | BUG_ON(atomic_read(&cfqq->ref) <= 0); | 2523 | BUG_ON(atomic_read(&cfqq->ref) <= 0); |
2519 | 2524 | ||
2520 | if (!atomic_dec_and_test(&cfqq->ref)) | 2525 | if (!atomic_dec_and_test(&cfqq->ref)) |
2521 | return; | 2526 | return; |
2522 | 2527 | ||
2523 | cfq_log_cfqq(cfqd, cfqq, "put_queue"); | 2528 | cfq_log_cfqq(cfqd, cfqq, "put_queue"); |
2524 | BUG_ON(rb_first(&cfqq->sort_list)); | 2529 | BUG_ON(rb_first(&cfqq->sort_list)); |
2525 | BUG_ON(cfqq->allocated[READ] + cfqq->allocated[WRITE]); | 2530 | BUG_ON(cfqq->allocated[READ] + cfqq->allocated[WRITE]); |
2526 | cfqg = cfqq->cfqg; | 2531 | cfqg = cfqq->cfqg; |
2527 | orig_cfqg = cfqq->orig_cfqg; | 2532 | orig_cfqg = cfqq->orig_cfqg; |
2528 | 2533 | ||
2529 | if (unlikely(cfqd->active_queue == cfqq)) { | 2534 | if (unlikely(cfqd->active_queue == cfqq)) { |
2530 | __cfq_slice_expired(cfqd, cfqq, 0); | 2535 | __cfq_slice_expired(cfqd, cfqq, 0); |
2531 | cfq_schedule_dispatch(cfqd); | 2536 | cfq_schedule_dispatch(cfqd); |
2532 | } | 2537 | } |
2533 | 2538 | ||
2534 | BUG_ON(cfq_cfqq_on_rr(cfqq)); | 2539 | BUG_ON(cfq_cfqq_on_rr(cfqq)); |
2535 | kmem_cache_free(cfq_pool, cfqq); | 2540 | kmem_cache_free(cfq_pool, cfqq); |
2536 | cfq_put_cfqg(cfqg); | 2541 | cfq_put_cfqg(cfqg); |
2537 | if (orig_cfqg) | 2542 | if (orig_cfqg) |
2538 | cfq_put_cfqg(orig_cfqg); | 2543 | cfq_put_cfqg(orig_cfqg); |
2539 | } | 2544 | } |
2540 | 2545 | ||
2541 | /* | 2546 | /* |
2542 | * Must always be called with the rcu_read_lock() held | 2547 | * Must always be called with the rcu_read_lock() held |
2543 | */ | 2548 | */ |
2544 | static void | 2549 | static void |
2545 | __call_for_each_cic(struct io_context *ioc, | 2550 | __call_for_each_cic(struct io_context *ioc, |
2546 | void (*func)(struct io_context *, struct cfq_io_context *)) | 2551 | void (*func)(struct io_context *, struct cfq_io_context *)) |
2547 | { | 2552 | { |
2548 | struct cfq_io_context *cic; | 2553 | struct cfq_io_context *cic; |
2549 | struct hlist_node *n; | 2554 | struct hlist_node *n; |
2550 | 2555 | ||
2551 | hlist_for_each_entry_rcu(cic, n, &ioc->cic_list, cic_list) | 2556 | hlist_for_each_entry_rcu(cic, n, &ioc->cic_list, cic_list) |
2552 | func(ioc, cic); | 2557 | func(ioc, cic); |
2553 | } | 2558 | } |
2554 | 2559 | ||
2555 | /* | 2560 | /* |
2556 | * Call func for each cic attached to this ioc. | 2561 | * Call func for each cic attached to this ioc. |
2557 | */ | 2562 | */ |
2558 | static void | 2563 | static void |
2559 | call_for_each_cic(struct io_context *ioc, | 2564 | call_for_each_cic(struct io_context *ioc, |
2560 | void (*func)(struct io_context *, struct cfq_io_context *)) | 2565 | void (*func)(struct io_context *, struct cfq_io_context *)) |
2561 | { | 2566 | { |
2562 | rcu_read_lock(); | 2567 | rcu_read_lock(); |
2563 | __call_for_each_cic(ioc, func); | 2568 | __call_for_each_cic(ioc, func); |
2564 | rcu_read_unlock(); | 2569 | rcu_read_unlock(); |
2565 | } | 2570 | } |
2566 | 2571 | ||
2567 | static void cfq_cic_free_rcu(struct rcu_head *head) | 2572 | static void cfq_cic_free_rcu(struct rcu_head *head) |
2568 | { | 2573 | { |
2569 | struct cfq_io_context *cic; | 2574 | struct cfq_io_context *cic; |
2570 | 2575 | ||
2571 | cic = container_of(head, struct cfq_io_context, rcu_head); | 2576 | cic = container_of(head, struct cfq_io_context, rcu_head); |
2572 | 2577 | ||
2573 | kmem_cache_free(cfq_ioc_pool, cic); | 2578 | kmem_cache_free(cfq_ioc_pool, cic); |
2574 | elv_ioc_count_dec(cfq_ioc_count); | 2579 | elv_ioc_count_dec(cfq_ioc_count); |
2575 | 2580 | ||
2576 | if (ioc_gone) { | 2581 | if (ioc_gone) { |
2577 | /* | 2582 | /* |
2578 | * CFQ scheduler is exiting, grab exit lock and check | 2583 | * CFQ scheduler is exiting, grab exit lock and check |
2579 | * the pending io context count. If it hits zero, | 2584 | * the pending io context count. If it hits zero, |
2580 | * complete ioc_gone and set it back to NULL | 2585 | * complete ioc_gone and set it back to NULL |
2581 | */ | 2586 | */ |
2582 | spin_lock(&ioc_gone_lock); | 2587 | spin_lock(&ioc_gone_lock); |
2583 | if (ioc_gone && !elv_ioc_count_read(cfq_ioc_count)) { | 2588 | if (ioc_gone && !elv_ioc_count_read(cfq_ioc_count)) { |
2584 | complete(ioc_gone); | 2589 | complete(ioc_gone); |
2585 | ioc_gone = NULL; | 2590 | ioc_gone = NULL; |
2586 | } | 2591 | } |
2587 | spin_unlock(&ioc_gone_lock); | 2592 | spin_unlock(&ioc_gone_lock); |
2588 | } | 2593 | } |
2589 | } | 2594 | } |
2590 | 2595 | ||
2591 | static void cfq_cic_free(struct cfq_io_context *cic) | 2596 | static void cfq_cic_free(struct cfq_io_context *cic) |
2592 | { | 2597 | { |
2593 | call_rcu(&cic->rcu_head, cfq_cic_free_rcu); | 2598 | call_rcu(&cic->rcu_head, cfq_cic_free_rcu); |
2594 | } | 2599 | } |
2595 | 2600 | ||
2596 | static void cic_free_func(struct io_context *ioc, struct cfq_io_context *cic) | 2601 | static void cic_free_func(struct io_context *ioc, struct cfq_io_context *cic) |
2597 | { | 2602 | { |
2598 | unsigned long flags; | 2603 | unsigned long flags; |
2599 | unsigned long dead_key = (unsigned long) cic->key; | 2604 | unsigned long dead_key = (unsigned long) cic->key; |
2600 | 2605 | ||
2601 | BUG_ON(!(dead_key & CIC_DEAD_KEY)); | 2606 | BUG_ON(!(dead_key & CIC_DEAD_KEY)); |
2602 | 2607 | ||
2603 | spin_lock_irqsave(&ioc->lock, flags); | 2608 | spin_lock_irqsave(&ioc->lock, flags); |
2604 | radix_tree_delete(&ioc->radix_root, dead_key >> CIC_DEAD_INDEX_SHIFT); | 2609 | radix_tree_delete(&ioc->radix_root, dead_key >> CIC_DEAD_INDEX_SHIFT); |
2605 | hlist_del_rcu(&cic->cic_list); | 2610 | hlist_del_rcu(&cic->cic_list); |
2606 | spin_unlock_irqrestore(&ioc->lock, flags); | 2611 | spin_unlock_irqrestore(&ioc->lock, flags); |
2607 | 2612 | ||
2608 | cfq_cic_free(cic); | 2613 | cfq_cic_free(cic); |
2609 | } | 2614 | } |
2610 | 2615 | ||
2611 | /* | 2616 | /* |
2612 | * Must be called with rcu_read_lock() held or preemption otherwise disabled. | 2617 | * Must be called with rcu_read_lock() held or preemption otherwise disabled. |
2613 | * Only two callers of this - ->dtor() which is called with the rcu_read_lock(), | 2618 | * Only two callers of this - ->dtor() which is called with the rcu_read_lock(), |
2614 | * and ->trim() which is called with the task lock held | 2619 | * and ->trim() which is called with the task lock held |
2615 | */ | 2620 | */ |
2616 | static void cfq_free_io_context(struct io_context *ioc) | 2621 | static void cfq_free_io_context(struct io_context *ioc) |
2617 | { | 2622 | { |
2618 | /* | 2623 | /* |
2619 | * ioc->refcount is zero here, or we are called from elv_unregister(), | 2624 | * ioc->refcount is zero here, or we are called from elv_unregister(), |
2620 | * so no more cic's are allowed to be linked into this ioc. So it | 2625 | * so no more cic's are allowed to be linked into this ioc. So it |
2621 | * should be ok to iterate over the known list, we will see all cic's | 2626 | * should be ok to iterate over the known list, we will see all cic's |
2622 | * since no new ones are added. | 2627 | * since no new ones are added. |
2623 | */ | 2628 | */ |
2624 | __call_for_each_cic(ioc, cic_free_func); | 2629 | __call_for_each_cic(ioc, cic_free_func); |
2625 | } | 2630 | } |
2626 | 2631 | ||
2627 | static void cfq_put_cooperator(struct cfq_queue *cfqq) | 2632 | static void cfq_put_cooperator(struct cfq_queue *cfqq) |
2628 | { | 2633 | { |
2629 | struct cfq_queue *__cfqq, *next; | 2634 | struct cfq_queue *__cfqq, *next; |
2630 | 2635 | ||
2631 | /* | 2636 | /* |
2632 | * If this queue was scheduled to merge with another queue, be | 2637 | * If this queue was scheduled to merge with another queue, be |
2633 | * sure to drop the reference taken on that queue (and others in | 2638 | * sure to drop the reference taken on that queue (and others in |
2634 | * the merge chain). See cfq_setup_merge and cfq_merge_cfqqs. | 2639 | * the merge chain). See cfq_setup_merge and cfq_merge_cfqqs. |
2635 | */ | 2640 | */ |
2636 | __cfqq = cfqq->new_cfqq; | 2641 | __cfqq = cfqq->new_cfqq; |
2637 | while (__cfqq) { | 2642 | while (__cfqq) { |
2638 | if (__cfqq == cfqq) { | 2643 | if (__cfqq == cfqq) { |
2639 | WARN(1, "cfqq->new_cfqq loop detected\n"); | 2644 | WARN(1, "cfqq->new_cfqq loop detected\n"); |
2640 | break; | 2645 | break; |
2641 | } | 2646 | } |
2642 | next = __cfqq->new_cfqq; | 2647 | next = __cfqq->new_cfqq; |
2643 | cfq_put_queue(__cfqq); | 2648 | cfq_put_queue(__cfqq); |
2644 | __cfqq = next; | 2649 | __cfqq = next; |
2645 | } | 2650 | } |
2646 | } | 2651 | } |
2647 | 2652 | ||
2648 | static void cfq_exit_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq) | 2653 | static void cfq_exit_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
2649 | { | 2654 | { |
2650 | if (unlikely(cfqq == cfqd->active_queue)) { | 2655 | if (unlikely(cfqq == cfqd->active_queue)) { |
2651 | __cfq_slice_expired(cfqd, cfqq, 0); | 2656 | __cfq_slice_expired(cfqd, cfqq, 0); |
2652 | cfq_schedule_dispatch(cfqd); | 2657 | cfq_schedule_dispatch(cfqd); |
2653 | } | 2658 | } |
2654 | 2659 | ||
2655 | cfq_put_cooperator(cfqq); | 2660 | cfq_put_cooperator(cfqq); |
2656 | 2661 | ||
2657 | cfq_put_queue(cfqq); | 2662 | cfq_put_queue(cfqq); |
2658 | } | 2663 | } |
2659 | 2664 | ||
2660 | static void __cfq_exit_single_io_context(struct cfq_data *cfqd, | 2665 | static void __cfq_exit_single_io_context(struct cfq_data *cfqd, |
2661 | struct cfq_io_context *cic) | 2666 | struct cfq_io_context *cic) |
2662 | { | 2667 | { |
2663 | struct io_context *ioc = cic->ioc; | 2668 | struct io_context *ioc = cic->ioc; |
2664 | 2669 | ||
2665 | list_del_init(&cic->queue_list); | 2670 | list_del_init(&cic->queue_list); |
2666 | 2671 | ||
2667 | /* | 2672 | /* |
2668 | * Make sure dead mark is seen for dead queues | 2673 | * Make sure dead mark is seen for dead queues |
2669 | */ | 2674 | */ |
2670 | smp_wmb(); | 2675 | smp_wmb(); |
2671 | cic->key = cfqd_dead_key(cfqd); | 2676 | cic->key = cfqd_dead_key(cfqd); |
2672 | 2677 | ||
2673 | if (ioc->ioc_data == cic) | 2678 | if (ioc->ioc_data == cic) |
2674 | rcu_assign_pointer(ioc->ioc_data, NULL); | 2679 | rcu_assign_pointer(ioc->ioc_data, NULL); |
2675 | 2680 | ||
2676 | if (cic->cfqq[BLK_RW_ASYNC]) { | 2681 | if (cic->cfqq[BLK_RW_ASYNC]) { |
2677 | cfq_exit_cfqq(cfqd, cic->cfqq[BLK_RW_ASYNC]); | 2682 | cfq_exit_cfqq(cfqd, cic->cfqq[BLK_RW_ASYNC]); |
2678 | cic->cfqq[BLK_RW_ASYNC] = NULL; | 2683 | cic->cfqq[BLK_RW_ASYNC] = NULL; |
2679 | } | 2684 | } |
2680 | 2685 | ||
2681 | if (cic->cfqq[BLK_RW_SYNC]) { | 2686 | if (cic->cfqq[BLK_RW_SYNC]) { |
2682 | cfq_exit_cfqq(cfqd, cic->cfqq[BLK_RW_SYNC]); | 2687 | cfq_exit_cfqq(cfqd, cic->cfqq[BLK_RW_SYNC]); |
2683 | cic->cfqq[BLK_RW_SYNC] = NULL; | 2688 | cic->cfqq[BLK_RW_SYNC] = NULL; |
2684 | } | 2689 | } |
2685 | } | 2690 | } |
2686 | 2691 | ||
2687 | static void cfq_exit_single_io_context(struct io_context *ioc, | 2692 | static void cfq_exit_single_io_context(struct io_context *ioc, |
2688 | struct cfq_io_context *cic) | 2693 | struct cfq_io_context *cic) |
2689 | { | 2694 | { |
2690 | struct cfq_data *cfqd = cic_to_cfqd(cic); | 2695 | struct cfq_data *cfqd = cic_to_cfqd(cic); |
2691 | 2696 | ||
2692 | if (cfqd) { | 2697 | if (cfqd) { |
2693 | struct request_queue *q = cfqd->queue; | 2698 | struct request_queue *q = cfqd->queue; |
2694 | unsigned long flags; | 2699 | unsigned long flags; |
2695 | 2700 | ||
2696 | spin_lock_irqsave(q->queue_lock, flags); | 2701 | spin_lock_irqsave(q->queue_lock, flags); |
2697 | 2702 | ||
2698 | /* | 2703 | /* |
2699 | * Ensure we get a fresh copy of the ->key to prevent | 2704 | * Ensure we get a fresh copy of the ->key to prevent |
2700 | * race between exiting task and queue | 2705 | * race between exiting task and queue |
2701 | */ | 2706 | */ |
2702 | smp_read_barrier_depends(); | 2707 | smp_read_barrier_depends(); |
2703 | if (cic->key == cfqd) | 2708 | if (cic->key == cfqd) |
2704 | __cfq_exit_single_io_context(cfqd, cic); | 2709 | __cfq_exit_single_io_context(cfqd, cic); |
2705 | 2710 | ||
2706 | spin_unlock_irqrestore(q->queue_lock, flags); | 2711 | spin_unlock_irqrestore(q->queue_lock, flags); |
2707 | } | 2712 | } |
2708 | } | 2713 | } |
2709 | 2714 | ||
2710 | /* | 2715 | /* |
2711 | * The process that ioc belongs to has exited, we need to clean up | 2716 | * The process that ioc belongs to has exited, we need to clean up |
2712 | * and put the internal structures we have that belongs to that process. | 2717 | * and put the internal structures we have that belongs to that process. |
2713 | */ | 2718 | */ |
2714 | static void cfq_exit_io_context(struct io_context *ioc) | 2719 | static void cfq_exit_io_context(struct io_context *ioc) |
2715 | { | 2720 | { |
2716 | call_for_each_cic(ioc, cfq_exit_single_io_context); | 2721 | call_for_each_cic(ioc, cfq_exit_single_io_context); |
2717 | } | 2722 | } |
2718 | 2723 | ||
2719 | static struct cfq_io_context * | 2724 | static struct cfq_io_context * |
2720 | cfq_alloc_io_context(struct cfq_data *cfqd, gfp_t gfp_mask) | 2725 | cfq_alloc_io_context(struct cfq_data *cfqd, gfp_t gfp_mask) |
2721 | { | 2726 | { |
2722 | struct cfq_io_context *cic; | 2727 | struct cfq_io_context *cic; |
2723 | 2728 | ||
2724 | cic = kmem_cache_alloc_node(cfq_ioc_pool, gfp_mask | __GFP_ZERO, | 2729 | cic = kmem_cache_alloc_node(cfq_ioc_pool, gfp_mask | __GFP_ZERO, |
2725 | cfqd->queue->node); | 2730 | cfqd->queue->node); |
2726 | if (cic) { | 2731 | if (cic) { |
2727 | cic->last_end_request = jiffies; | 2732 | cic->last_end_request = jiffies; |
2728 | INIT_LIST_HEAD(&cic->queue_list); | 2733 | INIT_LIST_HEAD(&cic->queue_list); |
2729 | INIT_HLIST_NODE(&cic->cic_list); | 2734 | INIT_HLIST_NODE(&cic->cic_list); |
2730 | cic->dtor = cfq_free_io_context; | 2735 | cic->dtor = cfq_free_io_context; |
2731 | cic->exit = cfq_exit_io_context; | 2736 | cic->exit = cfq_exit_io_context; |
2732 | elv_ioc_count_inc(cfq_ioc_count); | 2737 | elv_ioc_count_inc(cfq_ioc_count); |
2733 | } | 2738 | } |
2734 | 2739 | ||
2735 | return cic; | 2740 | return cic; |
2736 | } | 2741 | } |
2737 | 2742 | ||
2738 | static void cfq_init_prio_data(struct cfq_queue *cfqq, struct io_context *ioc) | 2743 | static void cfq_init_prio_data(struct cfq_queue *cfqq, struct io_context *ioc) |
2739 | { | 2744 | { |
2740 | struct task_struct *tsk = current; | 2745 | struct task_struct *tsk = current; |
2741 | int ioprio_class; | 2746 | int ioprio_class; |
2742 | 2747 | ||
2743 | if (!cfq_cfqq_prio_changed(cfqq)) | 2748 | if (!cfq_cfqq_prio_changed(cfqq)) |
2744 | return; | 2749 | return; |
2745 | 2750 | ||
2746 | ioprio_class = IOPRIO_PRIO_CLASS(ioc->ioprio); | 2751 | ioprio_class = IOPRIO_PRIO_CLASS(ioc->ioprio); |
2747 | switch (ioprio_class) { | 2752 | switch (ioprio_class) { |
2748 | default: | 2753 | default: |
2749 | printk(KERN_ERR "cfq: bad prio %x\n", ioprio_class); | 2754 | printk(KERN_ERR "cfq: bad prio %x\n", ioprio_class); |
2750 | case IOPRIO_CLASS_NONE: | 2755 | case IOPRIO_CLASS_NONE: |
2751 | /* | 2756 | /* |
2752 | * no prio set, inherit CPU scheduling settings | 2757 | * no prio set, inherit CPU scheduling settings |
2753 | */ | 2758 | */ |
2754 | cfqq->ioprio = task_nice_ioprio(tsk); | 2759 | cfqq->ioprio = task_nice_ioprio(tsk); |
2755 | cfqq->ioprio_class = task_nice_ioclass(tsk); | 2760 | cfqq->ioprio_class = task_nice_ioclass(tsk); |
2756 | break; | 2761 | break; |
2757 | case IOPRIO_CLASS_RT: | 2762 | case IOPRIO_CLASS_RT: |
2758 | cfqq->ioprio = task_ioprio(ioc); | 2763 | cfqq->ioprio = task_ioprio(ioc); |
2759 | cfqq->ioprio_class = IOPRIO_CLASS_RT; | 2764 | cfqq->ioprio_class = IOPRIO_CLASS_RT; |
2760 | break; | 2765 | break; |
2761 | case IOPRIO_CLASS_BE: | 2766 | case IOPRIO_CLASS_BE: |
2762 | cfqq->ioprio = task_ioprio(ioc); | 2767 | cfqq->ioprio = task_ioprio(ioc); |
2763 | cfqq->ioprio_class = IOPRIO_CLASS_BE; | 2768 | cfqq->ioprio_class = IOPRIO_CLASS_BE; |
2764 | break; | 2769 | break; |
2765 | case IOPRIO_CLASS_IDLE: | 2770 | case IOPRIO_CLASS_IDLE: |
2766 | cfqq->ioprio_class = IOPRIO_CLASS_IDLE; | 2771 | cfqq->ioprio_class = IOPRIO_CLASS_IDLE; |
2767 | cfqq->ioprio = 7; | 2772 | cfqq->ioprio = 7; |
2768 | cfq_clear_cfqq_idle_window(cfqq); | 2773 | cfq_clear_cfqq_idle_window(cfqq); |
2769 | break; | 2774 | break; |
2770 | } | 2775 | } |
2771 | 2776 | ||
2772 | /* | 2777 | /* |
2773 | * keep track of original prio settings in case we have to temporarily | 2778 | * keep track of original prio settings in case we have to temporarily |
2774 | * elevate the priority of this queue | 2779 | * elevate the priority of this queue |
2775 | */ | 2780 | */ |
2776 | cfqq->org_ioprio = cfqq->ioprio; | 2781 | cfqq->org_ioprio = cfqq->ioprio; |
2777 | cfqq->org_ioprio_class = cfqq->ioprio_class; | 2782 | cfqq->org_ioprio_class = cfqq->ioprio_class; |
2778 | cfq_clear_cfqq_prio_changed(cfqq); | 2783 | cfq_clear_cfqq_prio_changed(cfqq); |
2779 | } | 2784 | } |
2780 | 2785 | ||
2781 | static void changed_ioprio(struct io_context *ioc, struct cfq_io_context *cic) | 2786 | static void changed_ioprio(struct io_context *ioc, struct cfq_io_context *cic) |
2782 | { | 2787 | { |
2783 | struct cfq_data *cfqd = cic_to_cfqd(cic); | 2788 | struct cfq_data *cfqd = cic_to_cfqd(cic); |
2784 | struct cfq_queue *cfqq; | 2789 | struct cfq_queue *cfqq; |
2785 | unsigned long flags; | 2790 | unsigned long flags; |
2786 | 2791 | ||
2787 | if (unlikely(!cfqd)) | 2792 | if (unlikely(!cfqd)) |
2788 | return; | 2793 | return; |
2789 | 2794 | ||
2790 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); | 2795 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); |
2791 | 2796 | ||
2792 | cfqq = cic->cfqq[BLK_RW_ASYNC]; | 2797 | cfqq = cic->cfqq[BLK_RW_ASYNC]; |
2793 | if (cfqq) { | 2798 | if (cfqq) { |
2794 | struct cfq_queue *new_cfqq; | 2799 | struct cfq_queue *new_cfqq; |
2795 | new_cfqq = cfq_get_queue(cfqd, BLK_RW_ASYNC, cic->ioc, | 2800 | new_cfqq = cfq_get_queue(cfqd, BLK_RW_ASYNC, cic->ioc, |
2796 | GFP_ATOMIC); | 2801 | GFP_ATOMIC); |
2797 | if (new_cfqq) { | 2802 | if (new_cfqq) { |
2798 | cic->cfqq[BLK_RW_ASYNC] = new_cfqq; | 2803 | cic->cfqq[BLK_RW_ASYNC] = new_cfqq; |
2799 | cfq_put_queue(cfqq); | 2804 | cfq_put_queue(cfqq); |
2800 | } | 2805 | } |
2801 | } | 2806 | } |
2802 | 2807 | ||
2803 | cfqq = cic->cfqq[BLK_RW_SYNC]; | 2808 | cfqq = cic->cfqq[BLK_RW_SYNC]; |
2804 | if (cfqq) | 2809 | if (cfqq) |
2805 | cfq_mark_cfqq_prio_changed(cfqq); | 2810 | cfq_mark_cfqq_prio_changed(cfqq); |
2806 | 2811 | ||
2807 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); | 2812 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); |
2808 | } | 2813 | } |
2809 | 2814 | ||
2810 | static void cfq_ioc_set_ioprio(struct io_context *ioc) | 2815 | static void cfq_ioc_set_ioprio(struct io_context *ioc) |
2811 | { | 2816 | { |
2812 | call_for_each_cic(ioc, changed_ioprio); | 2817 | call_for_each_cic(ioc, changed_ioprio); |
2813 | ioc->ioprio_changed = 0; | 2818 | ioc->ioprio_changed = 0; |
2814 | } | 2819 | } |
2815 | 2820 | ||
2816 | static void cfq_init_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq, | 2821 | static void cfq_init_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
2817 | pid_t pid, bool is_sync) | 2822 | pid_t pid, bool is_sync) |
2818 | { | 2823 | { |
2819 | RB_CLEAR_NODE(&cfqq->rb_node); | 2824 | RB_CLEAR_NODE(&cfqq->rb_node); |
2820 | RB_CLEAR_NODE(&cfqq->p_node); | 2825 | RB_CLEAR_NODE(&cfqq->p_node); |
2821 | INIT_LIST_HEAD(&cfqq->fifo); | 2826 | INIT_LIST_HEAD(&cfqq->fifo); |
2822 | 2827 | ||
2823 | atomic_set(&cfqq->ref, 0); | 2828 | atomic_set(&cfqq->ref, 0); |
2824 | cfqq->cfqd = cfqd; | 2829 | cfqq->cfqd = cfqd; |
2825 | 2830 | ||
2826 | cfq_mark_cfqq_prio_changed(cfqq); | 2831 | cfq_mark_cfqq_prio_changed(cfqq); |
2827 | 2832 | ||
2828 | if (is_sync) { | 2833 | if (is_sync) { |
2829 | if (!cfq_class_idle(cfqq)) | 2834 | if (!cfq_class_idle(cfqq)) |
2830 | cfq_mark_cfqq_idle_window(cfqq); | 2835 | cfq_mark_cfqq_idle_window(cfqq); |
2831 | cfq_mark_cfqq_sync(cfqq); | 2836 | cfq_mark_cfqq_sync(cfqq); |
2832 | } | 2837 | } |
2833 | cfqq->pid = pid; | 2838 | cfqq->pid = pid; |
2834 | } | 2839 | } |
2835 | 2840 | ||
2836 | #ifdef CONFIG_CFQ_GROUP_IOSCHED | 2841 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
2837 | static void changed_cgroup(struct io_context *ioc, struct cfq_io_context *cic) | 2842 | static void changed_cgroup(struct io_context *ioc, struct cfq_io_context *cic) |
2838 | { | 2843 | { |
2839 | struct cfq_queue *sync_cfqq = cic_to_cfqq(cic, 1); | 2844 | struct cfq_queue *sync_cfqq = cic_to_cfqq(cic, 1); |
2840 | struct cfq_data *cfqd = cic_to_cfqd(cic); | 2845 | struct cfq_data *cfqd = cic_to_cfqd(cic); |
2841 | unsigned long flags; | 2846 | unsigned long flags; |
2842 | struct request_queue *q; | 2847 | struct request_queue *q; |
2843 | 2848 | ||
2844 | if (unlikely(!cfqd)) | 2849 | if (unlikely(!cfqd)) |
2845 | return; | 2850 | return; |
2846 | 2851 | ||
2847 | q = cfqd->queue; | 2852 | q = cfqd->queue; |
2848 | 2853 | ||
2849 | spin_lock_irqsave(q->queue_lock, flags); | 2854 | spin_lock_irqsave(q->queue_lock, flags); |
2850 | 2855 | ||
2851 | if (sync_cfqq) { | 2856 | if (sync_cfqq) { |
2852 | /* | 2857 | /* |
2853 | * Drop reference to sync queue. A new sync queue will be | 2858 | * Drop reference to sync queue. A new sync queue will be |
2854 | * assigned in new group upon arrival of a fresh request. | 2859 | * assigned in new group upon arrival of a fresh request. |
2855 | */ | 2860 | */ |
2856 | cfq_log_cfqq(cfqd, sync_cfqq, "changed cgroup"); | 2861 | cfq_log_cfqq(cfqd, sync_cfqq, "changed cgroup"); |
2857 | cic_set_cfqq(cic, NULL, 1); | 2862 | cic_set_cfqq(cic, NULL, 1); |
2858 | cfq_put_queue(sync_cfqq); | 2863 | cfq_put_queue(sync_cfqq); |
2859 | } | 2864 | } |
2860 | 2865 | ||
2861 | spin_unlock_irqrestore(q->queue_lock, flags); | 2866 | spin_unlock_irqrestore(q->queue_lock, flags); |
2862 | } | 2867 | } |
2863 | 2868 | ||
2864 | static void cfq_ioc_set_cgroup(struct io_context *ioc) | 2869 | static void cfq_ioc_set_cgroup(struct io_context *ioc) |
2865 | { | 2870 | { |
2866 | call_for_each_cic(ioc, changed_cgroup); | 2871 | call_for_each_cic(ioc, changed_cgroup); |
2867 | ioc->cgroup_changed = 0; | 2872 | ioc->cgroup_changed = 0; |
2868 | } | 2873 | } |
2869 | #endif /* CONFIG_CFQ_GROUP_IOSCHED */ | 2874 | #endif /* CONFIG_CFQ_GROUP_IOSCHED */ |
2870 | 2875 | ||
2871 | static struct cfq_queue * | 2876 | static struct cfq_queue * |
2872 | cfq_find_alloc_queue(struct cfq_data *cfqd, bool is_sync, | 2877 | cfq_find_alloc_queue(struct cfq_data *cfqd, bool is_sync, |
2873 | struct io_context *ioc, gfp_t gfp_mask) | 2878 | struct io_context *ioc, gfp_t gfp_mask) |
2874 | { | 2879 | { |
2875 | struct cfq_queue *cfqq, *new_cfqq = NULL; | 2880 | struct cfq_queue *cfqq, *new_cfqq = NULL; |
2876 | struct cfq_io_context *cic; | 2881 | struct cfq_io_context *cic; |
2877 | struct cfq_group *cfqg; | 2882 | struct cfq_group *cfqg; |
2878 | 2883 | ||
2879 | retry: | 2884 | retry: |
2880 | cfqg = cfq_get_cfqg(cfqd, 1); | 2885 | cfqg = cfq_get_cfqg(cfqd, 1); |
2881 | cic = cfq_cic_lookup(cfqd, ioc); | 2886 | cic = cfq_cic_lookup(cfqd, ioc); |
2882 | /* cic always exists here */ | 2887 | /* cic always exists here */ |
2883 | cfqq = cic_to_cfqq(cic, is_sync); | 2888 | cfqq = cic_to_cfqq(cic, is_sync); |
2884 | 2889 | ||
2885 | /* | 2890 | /* |
2886 | * Always try a new alloc if we fell back to the OOM cfqq | 2891 | * Always try a new alloc if we fell back to the OOM cfqq |
2887 | * originally, since it should just be a temporary situation. | 2892 | * originally, since it should just be a temporary situation. |
2888 | */ | 2893 | */ |
2889 | if (!cfqq || cfqq == &cfqd->oom_cfqq) { | 2894 | if (!cfqq || cfqq == &cfqd->oom_cfqq) { |
2890 | cfqq = NULL; | 2895 | cfqq = NULL; |
2891 | if (new_cfqq) { | 2896 | if (new_cfqq) { |
2892 | cfqq = new_cfqq; | 2897 | cfqq = new_cfqq; |
2893 | new_cfqq = NULL; | 2898 | new_cfqq = NULL; |
2894 | } else if (gfp_mask & __GFP_WAIT) { | 2899 | } else if (gfp_mask & __GFP_WAIT) { |
2895 | spin_unlock_irq(cfqd->queue->queue_lock); | 2900 | spin_unlock_irq(cfqd->queue->queue_lock); |
2896 | new_cfqq = kmem_cache_alloc_node(cfq_pool, | 2901 | new_cfqq = kmem_cache_alloc_node(cfq_pool, |
2897 | gfp_mask | __GFP_ZERO, | 2902 | gfp_mask | __GFP_ZERO, |
2898 | cfqd->queue->node); | 2903 | cfqd->queue->node); |
2899 | spin_lock_irq(cfqd->queue->queue_lock); | 2904 | spin_lock_irq(cfqd->queue->queue_lock); |
2900 | if (new_cfqq) | 2905 | if (new_cfqq) |
2901 | goto retry; | 2906 | goto retry; |
2902 | } else { | 2907 | } else { |
2903 | cfqq = kmem_cache_alloc_node(cfq_pool, | 2908 | cfqq = kmem_cache_alloc_node(cfq_pool, |
2904 | gfp_mask | __GFP_ZERO, | 2909 | gfp_mask | __GFP_ZERO, |
2905 | cfqd->queue->node); | 2910 | cfqd->queue->node); |
2906 | } | 2911 | } |
2907 | 2912 | ||
2908 | if (cfqq) { | 2913 | if (cfqq) { |
2909 | cfq_init_cfqq(cfqd, cfqq, current->pid, is_sync); | 2914 | cfq_init_cfqq(cfqd, cfqq, current->pid, is_sync); |
2910 | cfq_init_prio_data(cfqq, ioc); | 2915 | cfq_init_prio_data(cfqq, ioc); |
2911 | cfq_link_cfqq_cfqg(cfqq, cfqg); | 2916 | cfq_link_cfqq_cfqg(cfqq, cfqg); |
2912 | cfq_log_cfqq(cfqd, cfqq, "alloced"); | 2917 | cfq_log_cfqq(cfqd, cfqq, "alloced"); |
2913 | } else | 2918 | } else |
2914 | cfqq = &cfqd->oom_cfqq; | 2919 | cfqq = &cfqd->oom_cfqq; |
2915 | } | 2920 | } |
2916 | 2921 | ||
2917 | if (new_cfqq) | 2922 | if (new_cfqq) |
2918 | kmem_cache_free(cfq_pool, new_cfqq); | 2923 | kmem_cache_free(cfq_pool, new_cfqq); |
2919 | 2924 | ||
2920 | return cfqq; | 2925 | return cfqq; |
2921 | } | 2926 | } |
2922 | 2927 | ||
2923 | static struct cfq_queue ** | 2928 | static struct cfq_queue ** |
2924 | cfq_async_queue_prio(struct cfq_data *cfqd, int ioprio_class, int ioprio) | 2929 | cfq_async_queue_prio(struct cfq_data *cfqd, int ioprio_class, int ioprio) |
2925 | { | 2930 | { |
2926 | switch (ioprio_class) { | 2931 | switch (ioprio_class) { |
2927 | case IOPRIO_CLASS_RT: | 2932 | case IOPRIO_CLASS_RT: |
2928 | return &cfqd->async_cfqq[0][ioprio]; | 2933 | return &cfqd->async_cfqq[0][ioprio]; |
2929 | case IOPRIO_CLASS_BE: | 2934 | case IOPRIO_CLASS_BE: |
2930 | return &cfqd->async_cfqq[1][ioprio]; | 2935 | return &cfqd->async_cfqq[1][ioprio]; |
2931 | case IOPRIO_CLASS_IDLE: | 2936 | case IOPRIO_CLASS_IDLE: |
2932 | return &cfqd->async_idle_cfqq; | 2937 | return &cfqd->async_idle_cfqq; |
2933 | default: | 2938 | default: |
2934 | BUG(); | 2939 | BUG(); |
2935 | } | 2940 | } |
2936 | } | 2941 | } |
2937 | 2942 | ||
2938 | static struct cfq_queue * | 2943 | static struct cfq_queue * |
2939 | cfq_get_queue(struct cfq_data *cfqd, bool is_sync, struct io_context *ioc, | 2944 | cfq_get_queue(struct cfq_data *cfqd, bool is_sync, struct io_context *ioc, |
2940 | gfp_t gfp_mask) | 2945 | gfp_t gfp_mask) |
2941 | { | 2946 | { |
2942 | const int ioprio = task_ioprio(ioc); | 2947 | const int ioprio = task_ioprio(ioc); |
2943 | const int ioprio_class = task_ioprio_class(ioc); | 2948 | const int ioprio_class = task_ioprio_class(ioc); |
2944 | struct cfq_queue **async_cfqq = NULL; | 2949 | struct cfq_queue **async_cfqq = NULL; |
2945 | struct cfq_queue *cfqq = NULL; | 2950 | struct cfq_queue *cfqq = NULL; |
2946 | 2951 | ||
2947 | if (!is_sync) { | 2952 | if (!is_sync) { |
2948 | async_cfqq = cfq_async_queue_prio(cfqd, ioprio_class, ioprio); | 2953 | async_cfqq = cfq_async_queue_prio(cfqd, ioprio_class, ioprio); |
2949 | cfqq = *async_cfqq; | 2954 | cfqq = *async_cfqq; |
2950 | } | 2955 | } |
2951 | 2956 | ||
2952 | if (!cfqq) | 2957 | if (!cfqq) |
2953 | cfqq = cfq_find_alloc_queue(cfqd, is_sync, ioc, gfp_mask); | 2958 | cfqq = cfq_find_alloc_queue(cfqd, is_sync, ioc, gfp_mask); |
2954 | 2959 | ||
2955 | /* | 2960 | /* |
2956 | * pin the queue now that it's allocated, scheduler exit will prune it | 2961 | * pin the queue now that it's allocated, scheduler exit will prune it |
2957 | */ | 2962 | */ |
2958 | if (!is_sync && !(*async_cfqq)) { | 2963 | if (!is_sync && !(*async_cfqq)) { |
2959 | atomic_inc(&cfqq->ref); | 2964 | atomic_inc(&cfqq->ref); |
2960 | *async_cfqq = cfqq; | 2965 | *async_cfqq = cfqq; |
2961 | } | 2966 | } |
2962 | 2967 | ||
2963 | atomic_inc(&cfqq->ref); | 2968 | atomic_inc(&cfqq->ref); |
2964 | return cfqq; | 2969 | return cfqq; |
2965 | } | 2970 | } |
2966 | 2971 | ||
2967 | /* | 2972 | /* |
2968 | * We drop cfq io contexts lazily, so we may find a dead one. | 2973 | * We drop cfq io contexts lazily, so we may find a dead one. |
2969 | */ | 2974 | */ |
2970 | static void | 2975 | static void |
2971 | cfq_drop_dead_cic(struct cfq_data *cfqd, struct io_context *ioc, | 2976 | cfq_drop_dead_cic(struct cfq_data *cfqd, struct io_context *ioc, |
2972 | struct cfq_io_context *cic) | 2977 | struct cfq_io_context *cic) |
2973 | { | 2978 | { |
2974 | unsigned long flags; | 2979 | unsigned long flags; |
2975 | 2980 | ||
2976 | WARN_ON(!list_empty(&cic->queue_list)); | 2981 | WARN_ON(!list_empty(&cic->queue_list)); |
2977 | BUG_ON(cic->key != cfqd_dead_key(cfqd)); | 2982 | BUG_ON(cic->key != cfqd_dead_key(cfqd)); |
2978 | 2983 | ||
2979 | spin_lock_irqsave(&ioc->lock, flags); | 2984 | spin_lock_irqsave(&ioc->lock, flags); |
2980 | 2985 | ||
2981 | BUG_ON(ioc->ioc_data == cic); | 2986 | BUG_ON(ioc->ioc_data == cic); |
2982 | 2987 | ||
2983 | radix_tree_delete(&ioc->radix_root, cfqd->cic_index); | 2988 | radix_tree_delete(&ioc->radix_root, cfqd->cic_index); |
2984 | hlist_del_rcu(&cic->cic_list); | 2989 | hlist_del_rcu(&cic->cic_list); |
2985 | spin_unlock_irqrestore(&ioc->lock, flags); | 2990 | spin_unlock_irqrestore(&ioc->lock, flags); |
2986 | 2991 | ||
2987 | cfq_cic_free(cic); | 2992 | cfq_cic_free(cic); |
2988 | } | 2993 | } |
2989 | 2994 | ||
2990 | static struct cfq_io_context * | 2995 | static struct cfq_io_context * |
2991 | cfq_cic_lookup(struct cfq_data *cfqd, struct io_context *ioc) | 2996 | cfq_cic_lookup(struct cfq_data *cfqd, struct io_context *ioc) |
2992 | { | 2997 | { |
2993 | struct cfq_io_context *cic; | 2998 | struct cfq_io_context *cic; |
2994 | unsigned long flags; | 2999 | unsigned long flags; |
2995 | 3000 | ||
2996 | if (unlikely(!ioc)) | 3001 | if (unlikely(!ioc)) |
2997 | return NULL; | 3002 | return NULL; |
2998 | 3003 | ||
2999 | rcu_read_lock(); | 3004 | rcu_read_lock(); |
3000 | 3005 | ||
3001 | /* | 3006 | /* |
3002 | * we maintain a last-hit cache, to avoid browsing over the tree | 3007 | * we maintain a last-hit cache, to avoid browsing over the tree |
3003 | */ | 3008 | */ |
3004 | cic = rcu_dereference(ioc->ioc_data); | 3009 | cic = rcu_dereference(ioc->ioc_data); |
3005 | if (cic && cic->key == cfqd) { | 3010 | if (cic && cic->key == cfqd) { |
3006 | rcu_read_unlock(); | 3011 | rcu_read_unlock(); |
3007 | return cic; | 3012 | return cic; |
3008 | } | 3013 | } |
3009 | 3014 | ||
3010 | do { | 3015 | do { |
3011 | cic = radix_tree_lookup(&ioc->radix_root, cfqd->cic_index); | 3016 | cic = radix_tree_lookup(&ioc->radix_root, cfqd->cic_index); |
3012 | rcu_read_unlock(); | 3017 | rcu_read_unlock(); |
3013 | if (!cic) | 3018 | if (!cic) |
3014 | break; | 3019 | break; |
3015 | if (unlikely(cic->key != cfqd)) { | 3020 | if (unlikely(cic->key != cfqd)) { |
3016 | cfq_drop_dead_cic(cfqd, ioc, cic); | 3021 | cfq_drop_dead_cic(cfqd, ioc, cic); |
3017 | rcu_read_lock(); | 3022 | rcu_read_lock(); |
3018 | continue; | 3023 | continue; |
3019 | } | 3024 | } |
3020 | 3025 | ||
3021 | spin_lock_irqsave(&ioc->lock, flags); | 3026 | spin_lock_irqsave(&ioc->lock, flags); |
3022 | rcu_assign_pointer(ioc->ioc_data, cic); | 3027 | rcu_assign_pointer(ioc->ioc_data, cic); |
3023 | spin_unlock_irqrestore(&ioc->lock, flags); | 3028 | spin_unlock_irqrestore(&ioc->lock, flags); |
3024 | break; | 3029 | break; |
3025 | } while (1); | 3030 | } while (1); |
3026 | 3031 | ||
3027 | return cic; | 3032 | return cic; |
3028 | } | 3033 | } |
3029 | 3034 | ||
3030 | /* | 3035 | /* |
3031 | * Add cic into ioc, using cfqd as the search key. This enables us to lookup | 3036 | * Add cic into ioc, using cfqd as the search key. This enables us to lookup |
3032 | * the process specific cfq io context when entered from the block layer. | 3037 | * the process specific cfq io context when entered from the block layer. |
3033 | * Also adds the cic to a per-cfqd list, used when this queue is removed. | 3038 | * Also adds the cic to a per-cfqd list, used when this queue is removed. |
3034 | */ | 3039 | */ |
3035 | static int cfq_cic_link(struct cfq_data *cfqd, struct io_context *ioc, | 3040 | static int cfq_cic_link(struct cfq_data *cfqd, struct io_context *ioc, |
3036 | struct cfq_io_context *cic, gfp_t gfp_mask) | 3041 | struct cfq_io_context *cic, gfp_t gfp_mask) |
3037 | { | 3042 | { |
3038 | unsigned long flags; | 3043 | unsigned long flags; |
3039 | int ret; | 3044 | int ret; |
3040 | 3045 | ||
3041 | ret = radix_tree_preload(gfp_mask); | 3046 | ret = radix_tree_preload(gfp_mask); |
3042 | if (!ret) { | 3047 | if (!ret) { |
3043 | cic->ioc = ioc; | 3048 | cic->ioc = ioc; |
3044 | cic->key = cfqd; | 3049 | cic->key = cfqd; |
3045 | 3050 | ||
3046 | spin_lock_irqsave(&ioc->lock, flags); | 3051 | spin_lock_irqsave(&ioc->lock, flags); |
3047 | ret = radix_tree_insert(&ioc->radix_root, | 3052 | ret = radix_tree_insert(&ioc->radix_root, |
3048 | cfqd->cic_index, cic); | 3053 | cfqd->cic_index, cic); |
3049 | if (!ret) | 3054 | if (!ret) |
3050 | hlist_add_head_rcu(&cic->cic_list, &ioc->cic_list); | 3055 | hlist_add_head_rcu(&cic->cic_list, &ioc->cic_list); |
3051 | spin_unlock_irqrestore(&ioc->lock, flags); | 3056 | spin_unlock_irqrestore(&ioc->lock, flags); |
3052 | 3057 | ||
3053 | radix_tree_preload_end(); | 3058 | radix_tree_preload_end(); |
3054 | 3059 | ||
3055 | if (!ret) { | 3060 | if (!ret) { |
3056 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); | 3061 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); |
3057 | list_add(&cic->queue_list, &cfqd->cic_list); | 3062 | list_add(&cic->queue_list, &cfqd->cic_list); |
3058 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); | 3063 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); |
3059 | } | 3064 | } |
3060 | } | 3065 | } |
3061 | 3066 | ||
3062 | if (ret) | 3067 | if (ret) |
3063 | printk(KERN_ERR "cfq: cic link failed!\n"); | 3068 | printk(KERN_ERR "cfq: cic link failed!\n"); |
3064 | 3069 | ||
3065 | return ret; | 3070 | return ret; |
3066 | } | 3071 | } |
3067 | 3072 | ||
3068 | /* | 3073 | /* |
3069 | * Setup general io context and cfq io context. There can be several cfq | 3074 | * Setup general io context and cfq io context. There can be several cfq |
3070 | * io contexts per general io context, if this process is doing io to more | 3075 | * io contexts per general io context, if this process is doing io to more |
3071 | * than one device managed by cfq. | 3076 | * than one device managed by cfq. |
3072 | */ | 3077 | */ |
3073 | static struct cfq_io_context * | 3078 | static struct cfq_io_context * |
3074 | cfq_get_io_context(struct cfq_data *cfqd, gfp_t gfp_mask) | 3079 | cfq_get_io_context(struct cfq_data *cfqd, gfp_t gfp_mask) |
3075 | { | 3080 | { |
3076 | struct io_context *ioc = NULL; | 3081 | struct io_context *ioc = NULL; |
3077 | struct cfq_io_context *cic; | 3082 | struct cfq_io_context *cic; |
3078 | 3083 | ||
3079 | might_sleep_if(gfp_mask & __GFP_WAIT); | 3084 | might_sleep_if(gfp_mask & __GFP_WAIT); |
3080 | 3085 | ||
3081 | ioc = get_io_context(gfp_mask, cfqd->queue->node); | 3086 | ioc = get_io_context(gfp_mask, cfqd->queue->node); |
3082 | if (!ioc) | 3087 | if (!ioc) |
3083 | return NULL; | 3088 | return NULL; |
3084 | 3089 | ||
3085 | cic = cfq_cic_lookup(cfqd, ioc); | 3090 | cic = cfq_cic_lookup(cfqd, ioc); |
3086 | if (cic) | 3091 | if (cic) |
3087 | goto out; | 3092 | goto out; |
3088 | 3093 | ||
3089 | cic = cfq_alloc_io_context(cfqd, gfp_mask); | 3094 | cic = cfq_alloc_io_context(cfqd, gfp_mask); |
3090 | if (cic == NULL) | 3095 | if (cic == NULL) |
3091 | goto err; | 3096 | goto err; |
3092 | 3097 | ||
3093 | if (cfq_cic_link(cfqd, ioc, cic, gfp_mask)) | 3098 | if (cfq_cic_link(cfqd, ioc, cic, gfp_mask)) |
3094 | goto err_free; | 3099 | goto err_free; |
3095 | 3100 | ||
3096 | out: | 3101 | out: |
3097 | smp_read_barrier_depends(); | 3102 | smp_read_barrier_depends(); |
3098 | if (unlikely(ioc->ioprio_changed)) | 3103 | if (unlikely(ioc->ioprio_changed)) |
3099 | cfq_ioc_set_ioprio(ioc); | 3104 | cfq_ioc_set_ioprio(ioc); |
3100 | 3105 | ||
3101 | #ifdef CONFIG_CFQ_GROUP_IOSCHED | 3106 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3102 | if (unlikely(ioc->cgroup_changed)) | 3107 | if (unlikely(ioc->cgroup_changed)) |
3103 | cfq_ioc_set_cgroup(ioc); | 3108 | cfq_ioc_set_cgroup(ioc); |
3104 | #endif | 3109 | #endif |
3105 | return cic; | 3110 | return cic; |
3106 | err_free: | 3111 | err_free: |
3107 | cfq_cic_free(cic); | 3112 | cfq_cic_free(cic); |
3108 | err: | 3113 | err: |
3109 | put_io_context(ioc); | 3114 | put_io_context(ioc); |
3110 | return NULL; | 3115 | return NULL; |
3111 | } | 3116 | } |
3112 | 3117 | ||
3113 | static void | 3118 | static void |
3114 | cfq_update_io_thinktime(struct cfq_data *cfqd, struct cfq_io_context *cic) | 3119 | cfq_update_io_thinktime(struct cfq_data *cfqd, struct cfq_io_context *cic) |
3115 | { | 3120 | { |
3116 | unsigned long elapsed = jiffies - cic->last_end_request; | 3121 | unsigned long elapsed = jiffies - cic->last_end_request; |
3117 | unsigned long ttime = min(elapsed, 2UL * cfqd->cfq_slice_idle); | 3122 | unsigned long ttime = min(elapsed, 2UL * cfqd->cfq_slice_idle); |
3118 | 3123 | ||
3119 | cic->ttime_samples = (7*cic->ttime_samples + 256) / 8; | 3124 | cic->ttime_samples = (7*cic->ttime_samples + 256) / 8; |
3120 | cic->ttime_total = (7*cic->ttime_total + 256*ttime) / 8; | 3125 | cic->ttime_total = (7*cic->ttime_total + 256*ttime) / 8; |
3121 | cic->ttime_mean = (cic->ttime_total + 128) / cic->ttime_samples; | 3126 | cic->ttime_mean = (cic->ttime_total + 128) / cic->ttime_samples; |
3122 | } | 3127 | } |
3123 | 3128 | ||
3124 | static void | 3129 | static void |
3125 | cfq_update_io_seektime(struct cfq_data *cfqd, struct cfq_queue *cfqq, | 3130 | cfq_update_io_seektime(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
3126 | struct request *rq) | 3131 | struct request *rq) |
3127 | { | 3132 | { |
3128 | sector_t sdist = 0; | 3133 | sector_t sdist = 0; |
3129 | sector_t n_sec = blk_rq_sectors(rq); | 3134 | sector_t n_sec = blk_rq_sectors(rq); |
3130 | if (cfqq->last_request_pos) { | 3135 | if (cfqq->last_request_pos) { |
3131 | if (cfqq->last_request_pos < blk_rq_pos(rq)) | 3136 | if (cfqq->last_request_pos < blk_rq_pos(rq)) |
3132 | sdist = blk_rq_pos(rq) - cfqq->last_request_pos; | 3137 | sdist = blk_rq_pos(rq) - cfqq->last_request_pos; |
3133 | else | 3138 | else |
3134 | sdist = cfqq->last_request_pos - blk_rq_pos(rq); | 3139 | sdist = cfqq->last_request_pos - blk_rq_pos(rq); |
3135 | } | 3140 | } |
3136 | 3141 | ||
3137 | cfqq->seek_history <<= 1; | 3142 | cfqq->seek_history <<= 1; |
3138 | if (blk_queue_nonrot(cfqd->queue)) | 3143 | if (blk_queue_nonrot(cfqd->queue)) |
3139 | cfqq->seek_history |= (n_sec < CFQQ_SECT_THR_NONROT); | 3144 | cfqq->seek_history |= (n_sec < CFQQ_SECT_THR_NONROT); |
3140 | else | 3145 | else |
3141 | cfqq->seek_history |= (sdist > CFQQ_SEEK_THR); | 3146 | cfqq->seek_history |= (sdist > CFQQ_SEEK_THR); |
3142 | } | 3147 | } |
3143 | 3148 | ||
3144 | /* | 3149 | /* |
3145 | * Disable idle window if the process thinks too long or seeks so much that | 3150 | * Disable idle window if the process thinks too long or seeks so much that |
3146 | * it doesn't matter | 3151 | * it doesn't matter |
3147 | */ | 3152 | */ |
3148 | static void | 3153 | static void |
3149 | cfq_update_idle_window(struct cfq_data *cfqd, struct cfq_queue *cfqq, | 3154 | cfq_update_idle_window(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
3150 | struct cfq_io_context *cic) | 3155 | struct cfq_io_context *cic) |
3151 | { | 3156 | { |
3152 | int old_idle, enable_idle; | 3157 | int old_idle, enable_idle; |
3153 | 3158 | ||
3154 | /* | 3159 | /* |
3155 | * Don't idle for async or idle io prio class | 3160 | * Don't idle for async or idle io prio class |
3156 | */ | 3161 | */ |
3157 | if (!cfq_cfqq_sync(cfqq) || cfq_class_idle(cfqq)) | 3162 | if (!cfq_cfqq_sync(cfqq) || cfq_class_idle(cfqq)) |
3158 | return; | 3163 | return; |
3159 | 3164 | ||
3160 | enable_idle = old_idle = cfq_cfqq_idle_window(cfqq); | 3165 | enable_idle = old_idle = cfq_cfqq_idle_window(cfqq); |
3161 | 3166 | ||
3162 | if (cfqq->queued[0] + cfqq->queued[1] >= 4) | 3167 | if (cfqq->queued[0] + cfqq->queued[1] >= 4) |
3163 | cfq_mark_cfqq_deep(cfqq); | 3168 | cfq_mark_cfqq_deep(cfqq); |
3164 | 3169 | ||
3165 | if (!atomic_read(&cic->ioc->nr_tasks) || !cfqd->cfq_slice_idle || | 3170 | if (!atomic_read(&cic->ioc->nr_tasks) || !cfqd->cfq_slice_idle || |
3166 | (!cfq_cfqq_deep(cfqq) && CFQQ_SEEKY(cfqq))) | 3171 | (!cfq_cfqq_deep(cfqq) && CFQQ_SEEKY(cfqq))) |
3167 | enable_idle = 0; | 3172 | enable_idle = 0; |
3168 | else if (sample_valid(cic->ttime_samples)) { | 3173 | else if (sample_valid(cic->ttime_samples)) { |
3169 | if (cic->ttime_mean > cfqd->cfq_slice_idle) | 3174 | if (cic->ttime_mean > cfqd->cfq_slice_idle) |
3170 | enable_idle = 0; | 3175 | enable_idle = 0; |
3171 | else | 3176 | else |
3172 | enable_idle = 1; | 3177 | enable_idle = 1; |
3173 | } | 3178 | } |
3174 | 3179 | ||
3175 | if (old_idle != enable_idle) { | 3180 | if (old_idle != enable_idle) { |
3176 | cfq_log_cfqq(cfqd, cfqq, "idle=%d", enable_idle); | 3181 | cfq_log_cfqq(cfqd, cfqq, "idle=%d", enable_idle); |
3177 | if (enable_idle) | 3182 | if (enable_idle) |
3178 | cfq_mark_cfqq_idle_window(cfqq); | 3183 | cfq_mark_cfqq_idle_window(cfqq); |
3179 | else | 3184 | else |
3180 | cfq_clear_cfqq_idle_window(cfqq); | 3185 | cfq_clear_cfqq_idle_window(cfqq); |
3181 | } | 3186 | } |
3182 | } | 3187 | } |
3183 | 3188 | ||
3184 | /* | 3189 | /* |
3185 | * Check if new_cfqq should preempt the currently active queue. Return 0 for | 3190 | * Check if new_cfqq should preempt the currently active queue. Return 0 for |
3186 | * no or if we aren't sure, a 1 will cause a preempt. | 3191 | * no or if we aren't sure, a 1 will cause a preempt. |
3187 | */ | 3192 | */ |
3188 | static bool | 3193 | static bool |
3189 | cfq_should_preempt(struct cfq_data *cfqd, struct cfq_queue *new_cfqq, | 3194 | cfq_should_preempt(struct cfq_data *cfqd, struct cfq_queue *new_cfqq, |
3190 | struct request *rq) | 3195 | struct request *rq) |
3191 | { | 3196 | { |
3192 | struct cfq_queue *cfqq; | 3197 | struct cfq_queue *cfqq; |
3193 | 3198 | ||
3194 | cfqq = cfqd->active_queue; | 3199 | cfqq = cfqd->active_queue; |
3195 | if (!cfqq) | 3200 | if (!cfqq) |
3196 | return false; | 3201 | return false; |
3197 | 3202 | ||
3198 | if (cfq_class_idle(new_cfqq)) | 3203 | if (cfq_class_idle(new_cfqq)) |
3199 | return false; | 3204 | return false; |
3200 | 3205 | ||
3201 | if (cfq_class_idle(cfqq)) | 3206 | if (cfq_class_idle(cfqq)) |
3202 | return true; | 3207 | return true; |
3203 | 3208 | ||
3204 | /* | 3209 | /* |
3205 | * Don't allow a non-RT request to preempt an ongoing RT cfqq timeslice. | 3210 | * Don't allow a non-RT request to preempt an ongoing RT cfqq timeslice. |
3206 | */ | 3211 | */ |
3207 | if (cfq_class_rt(cfqq) && !cfq_class_rt(new_cfqq)) | 3212 | if (cfq_class_rt(cfqq) && !cfq_class_rt(new_cfqq)) |
3208 | return false; | 3213 | return false; |
3209 | 3214 | ||
3210 | /* | 3215 | /* |
3211 | * if the new request is sync, but the currently running queue is | 3216 | * if the new request is sync, but the currently running queue is |
3212 | * not, let the sync request have priority. | 3217 | * not, let the sync request have priority. |
3213 | */ | 3218 | */ |
3214 | if (rq_is_sync(rq) && !cfq_cfqq_sync(cfqq)) | 3219 | if (rq_is_sync(rq) && !cfq_cfqq_sync(cfqq)) |
3215 | return true; | 3220 | return true; |
3216 | 3221 | ||
3217 | if (new_cfqq->cfqg != cfqq->cfqg) | 3222 | if (new_cfqq->cfqg != cfqq->cfqg) |
3218 | return false; | 3223 | return false; |
3219 | 3224 | ||
3220 | if (cfq_slice_used(cfqq)) | 3225 | if (cfq_slice_used(cfqq)) |
3221 | return true; | 3226 | return true; |
3222 | 3227 | ||
3223 | /* Allow preemption only if we are idling on sync-noidle tree */ | 3228 | /* Allow preemption only if we are idling on sync-noidle tree */ |
3224 | if (cfqd->serving_type == SYNC_NOIDLE_WORKLOAD && | 3229 | if (cfqd->serving_type == SYNC_NOIDLE_WORKLOAD && |
3225 | cfqq_type(new_cfqq) == SYNC_NOIDLE_WORKLOAD && | 3230 | cfqq_type(new_cfqq) == SYNC_NOIDLE_WORKLOAD && |
3226 | new_cfqq->service_tree->count == 2 && | 3231 | new_cfqq->service_tree->count == 2 && |
3227 | RB_EMPTY_ROOT(&cfqq->sort_list)) | 3232 | RB_EMPTY_ROOT(&cfqq->sort_list)) |
3228 | return true; | 3233 | return true; |
3229 | 3234 | ||
3230 | /* | 3235 | /* |
3231 | * So both queues are sync. Let the new request get disk time if | 3236 | * So both queues are sync. Let the new request get disk time if |
3232 | * it's a metadata request and the current queue is doing regular IO. | 3237 | * it's a metadata request and the current queue is doing regular IO. |
3233 | */ | 3238 | */ |
3234 | if ((rq->cmd_flags & REQ_META) && !cfqq->meta_pending) | 3239 | if ((rq->cmd_flags & REQ_META) && !cfqq->meta_pending) |
3235 | return true; | 3240 | return true; |
3236 | 3241 | ||
3237 | /* | 3242 | /* |
3238 | * Allow an RT request to pre-empt an ongoing non-RT cfqq timeslice. | 3243 | * Allow an RT request to pre-empt an ongoing non-RT cfqq timeslice. |
3239 | */ | 3244 | */ |
3240 | if (cfq_class_rt(new_cfqq) && !cfq_class_rt(cfqq)) | 3245 | if (cfq_class_rt(new_cfqq) && !cfq_class_rt(cfqq)) |
3241 | return true; | 3246 | return true; |
3242 | 3247 | ||
3243 | if (!cfqd->active_cic || !cfq_cfqq_wait_request(cfqq)) | 3248 | if (!cfqd->active_cic || !cfq_cfqq_wait_request(cfqq)) |
3244 | return false; | 3249 | return false; |
3245 | 3250 | ||
3246 | /* | 3251 | /* |
3247 | * if this request is as-good as one we would expect from the | 3252 | * if this request is as-good as one we would expect from the |
3248 | * current cfqq, let it preempt | 3253 | * current cfqq, let it preempt |
3249 | */ | 3254 | */ |
3250 | if (cfq_rq_close(cfqd, cfqq, rq)) | 3255 | if (cfq_rq_close(cfqd, cfqq, rq)) |
3251 | return true; | 3256 | return true; |
3252 | 3257 | ||
3253 | return false; | 3258 | return false; |
3254 | } | 3259 | } |
3255 | 3260 | ||
3256 | /* | 3261 | /* |
3257 | * cfqq preempts the active queue. if we allowed preempt with no slice left, | 3262 | * cfqq preempts the active queue. if we allowed preempt with no slice left, |
3258 | * let it have half of its nominal slice. | 3263 | * let it have half of its nominal slice. |
3259 | */ | 3264 | */ |
3260 | static void cfq_preempt_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq) | 3265 | static void cfq_preempt_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
3261 | { | 3266 | { |
3262 | cfq_log_cfqq(cfqd, cfqq, "preempt"); | 3267 | cfq_log_cfqq(cfqd, cfqq, "preempt"); |
3263 | cfq_slice_expired(cfqd, 1); | 3268 | cfq_slice_expired(cfqd, 1); |
3264 | 3269 | ||
3265 | /* | 3270 | /* |
3266 | * Put the new queue at the front of the of the current list, | 3271 | * Put the new queue at the front of the of the current list, |
3267 | * so we know that it will be selected next. | 3272 | * so we know that it will be selected next. |
3268 | */ | 3273 | */ |
3269 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); | 3274 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); |
3270 | 3275 | ||
3271 | cfq_service_tree_add(cfqd, cfqq, 1); | 3276 | cfq_service_tree_add(cfqd, cfqq, 1); |
3272 | 3277 | ||
3273 | cfqq->slice_end = 0; | 3278 | cfqq->slice_end = 0; |
3274 | cfq_mark_cfqq_slice_new(cfqq); | 3279 | cfq_mark_cfqq_slice_new(cfqq); |
3275 | } | 3280 | } |
3276 | 3281 | ||
3277 | /* | 3282 | /* |
3278 | * Called when a new fs request (rq) is added (to cfqq). Check if there's | 3283 | * Called when a new fs request (rq) is added (to cfqq). Check if there's |
3279 | * something we should do about it | 3284 | * something we should do about it |
3280 | */ | 3285 | */ |
3281 | static void | 3286 | static void |
3282 | cfq_rq_enqueued(struct cfq_data *cfqd, struct cfq_queue *cfqq, | 3287 | cfq_rq_enqueued(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
3283 | struct request *rq) | 3288 | struct request *rq) |
3284 | { | 3289 | { |
3285 | struct cfq_io_context *cic = RQ_CIC(rq); | 3290 | struct cfq_io_context *cic = RQ_CIC(rq); |
3286 | 3291 | ||
3287 | cfqd->rq_queued++; | 3292 | cfqd->rq_queued++; |
3288 | if (rq->cmd_flags & REQ_META) | 3293 | if (rq->cmd_flags & REQ_META) |
3289 | cfqq->meta_pending++; | 3294 | cfqq->meta_pending++; |
3290 | 3295 | ||
3291 | cfq_update_io_thinktime(cfqd, cic); | 3296 | cfq_update_io_thinktime(cfqd, cic); |
3292 | cfq_update_io_seektime(cfqd, cfqq, rq); | 3297 | cfq_update_io_seektime(cfqd, cfqq, rq); |
3293 | cfq_update_idle_window(cfqd, cfqq, cic); | 3298 | cfq_update_idle_window(cfqd, cfqq, cic); |
3294 | 3299 | ||
3295 | cfqq->last_request_pos = blk_rq_pos(rq) + blk_rq_sectors(rq); | 3300 | cfqq->last_request_pos = blk_rq_pos(rq) + blk_rq_sectors(rq); |
3296 | 3301 | ||
3297 | if (cfqq == cfqd->active_queue) { | 3302 | if (cfqq == cfqd->active_queue) { |
3298 | /* | 3303 | /* |
3299 | * Remember that we saw a request from this process, but | 3304 | * Remember that we saw a request from this process, but |
3300 | * don't start queuing just yet. Otherwise we risk seeing lots | 3305 | * don't start queuing just yet. Otherwise we risk seeing lots |
3301 | * of tiny requests, because we disrupt the normal plugging | 3306 | * of tiny requests, because we disrupt the normal plugging |
3302 | * and merging. If the request is already larger than a single | 3307 | * and merging. If the request is already larger than a single |
3303 | * page, let it rip immediately. For that case we assume that | 3308 | * page, let it rip immediately. For that case we assume that |
3304 | * merging is already done. Ditto for a busy system that | 3309 | * merging is already done. Ditto for a busy system that |
3305 | * has other work pending, don't risk delaying until the | 3310 | * has other work pending, don't risk delaying until the |
3306 | * idle timer unplug to continue working. | 3311 | * idle timer unplug to continue working. |
3307 | */ | 3312 | */ |
3308 | if (cfq_cfqq_wait_request(cfqq)) { | 3313 | if (cfq_cfqq_wait_request(cfqq)) { |
3309 | if (blk_rq_bytes(rq) > PAGE_CACHE_SIZE || | 3314 | if (blk_rq_bytes(rq) > PAGE_CACHE_SIZE || |
3310 | cfqd->busy_queues > 1) { | 3315 | cfqd->busy_queues > 1) { |
3311 | cfq_del_timer(cfqd, cfqq); | 3316 | cfq_del_timer(cfqd, cfqq); |
3312 | cfq_clear_cfqq_wait_request(cfqq); | 3317 | cfq_clear_cfqq_wait_request(cfqq); |
3313 | __blk_run_queue(cfqd->queue); | 3318 | __blk_run_queue(cfqd->queue); |
3314 | } else { | 3319 | } else { |
3315 | cfq_blkiocg_update_idle_time_stats( | 3320 | cfq_blkiocg_update_idle_time_stats( |
3316 | &cfqq->cfqg->blkg); | 3321 | &cfqq->cfqg->blkg); |
3317 | cfq_mark_cfqq_must_dispatch(cfqq); | 3322 | cfq_mark_cfqq_must_dispatch(cfqq); |
3318 | } | 3323 | } |
3319 | } | 3324 | } |
3320 | } else if (cfq_should_preempt(cfqd, cfqq, rq)) { | 3325 | } else if (cfq_should_preempt(cfqd, cfqq, rq)) { |
3321 | /* | 3326 | /* |
3322 | * not the active queue - expire current slice if it is | 3327 | * not the active queue - expire current slice if it is |
3323 | * idle and has expired it's mean thinktime or this new queue | 3328 | * idle and has expired it's mean thinktime or this new queue |
3324 | * has some old slice time left and is of higher priority or | 3329 | * has some old slice time left and is of higher priority or |
3325 | * this new queue is RT and the current one is BE | 3330 | * this new queue is RT and the current one is BE |
3326 | */ | 3331 | */ |
3327 | cfq_preempt_queue(cfqd, cfqq); | 3332 | cfq_preempt_queue(cfqd, cfqq); |
3328 | __blk_run_queue(cfqd->queue); | 3333 | __blk_run_queue(cfqd->queue); |
3329 | } | 3334 | } |
3330 | } | 3335 | } |
3331 | 3336 | ||
3332 | static void cfq_insert_request(struct request_queue *q, struct request *rq) | 3337 | static void cfq_insert_request(struct request_queue *q, struct request *rq) |
3333 | { | 3338 | { |
3334 | struct cfq_data *cfqd = q->elevator->elevator_data; | 3339 | struct cfq_data *cfqd = q->elevator->elevator_data; |
3335 | struct cfq_queue *cfqq = RQ_CFQQ(rq); | 3340 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
3336 | 3341 | ||
3337 | cfq_log_cfqq(cfqd, cfqq, "insert_request"); | 3342 | cfq_log_cfqq(cfqd, cfqq, "insert_request"); |
3338 | cfq_init_prio_data(cfqq, RQ_CIC(rq)->ioc); | 3343 | cfq_init_prio_data(cfqq, RQ_CIC(rq)->ioc); |
3339 | 3344 | ||
3340 | rq_set_fifo_time(rq, jiffies + cfqd->cfq_fifo_expire[rq_is_sync(rq)]); | 3345 | rq_set_fifo_time(rq, jiffies + cfqd->cfq_fifo_expire[rq_is_sync(rq)]); |
3341 | list_add_tail(&rq->queuelist, &cfqq->fifo); | 3346 | list_add_tail(&rq->queuelist, &cfqq->fifo); |
3342 | cfq_add_rq_rb(rq); | 3347 | cfq_add_rq_rb(rq); |
3343 | cfq_blkiocg_update_io_add_stats(&(RQ_CFQG(rq))->blkg, | 3348 | cfq_blkiocg_update_io_add_stats(&(RQ_CFQG(rq))->blkg, |
3344 | &cfqd->serving_group->blkg, rq_data_dir(rq), | 3349 | &cfqd->serving_group->blkg, rq_data_dir(rq), |
3345 | rq_is_sync(rq)); | 3350 | rq_is_sync(rq)); |
3346 | cfq_rq_enqueued(cfqd, cfqq, rq); | 3351 | cfq_rq_enqueued(cfqd, cfqq, rq); |
3347 | } | 3352 | } |
3348 | 3353 | ||
3349 | /* | 3354 | /* |
3350 | * Update hw_tag based on peak queue depth over 50 samples under | 3355 | * Update hw_tag based on peak queue depth over 50 samples under |
3351 | * sufficient load. | 3356 | * sufficient load. |
3352 | */ | 3357 | */ |
3353 | static void cfq_update_hw_tag(struct cfq_data *cfqd) | 3358 | static void cfq_update_hw_tag(struct cfq_data *cfqd) |
3354 | { | 3359 | { |
3355 | struct cfq_queue *cfqq = cfqd->active_queue; | 3360 | struct cfq_queue *cfqq = cfqd->active_queue; |
3356 | 3361 | ||
3357 | if (cfqd->rq_in_driver > cfqd->hw_tag_est_depth) | 3362 | if (cfqd->rq_in_driver > cfqd->hw_tag_est_depth) |
3358 | cfqd->hw_tag_est_depth = cfqd->rq_in_driver; | 3363 | cfqd->hw_tag_est_depth = cfqd->rq_in_driver; |
3359 | 3364 | ||
3360 | if (cfqd->hw_tag == 1) | 3365 | if (cfqd->hw_tag == 1) |
3361 | return; | 3366 | return; |
3362 | 3367 | ||
3363 | if (cfqd->rq_queued <= CFQ_HW_QUEUE_MIN && | 3368 | if (cfqd->rq_queued <= CFQ_HW_QUEUE_MIN && |
3364 | cfqd->rq_in_driver <= CFQ_HW_QUEUE_MIN) | 3369 | cfqd->rq_in_driver <= CFQ_HW_QUEUE_MIN) |
3365 | return; | 3370 | return; |
3366 | 3371 | ||
3367 | /* | 3372 | /* |
3368 | * If active queue hasn't enough requests and can idle, cfq might not | 3373 | * If active queue hasn't enough requests and can idle, cfq might not |
3369 | * dispatch sufficient requests to hardware. Don't zero hw_tag in this | 3374 | * dispatch sufficient requests to hardware. Don't zero hw_tag in this |
3370 | * case | 3375 | * case |
3371 | */ | 3376 | */ |
3372 | if (cfqq && cfq_cfqq_idle_window(cfqq) && | 3377 | if (cfqq && cfq_cfqq_idle_window(cfqq) && |
3373 | cfqq->dispatched + cfqq->queued[0] + cfqq->queued[1] < | 3378 | cfqq->dispatched + cfqq->queued[0] + cfqq->queued[1] < |
3374 | CFQ_HW_QUEUE_MIN && cfqd->rq_in_driver < CFQ_HW_QUEUE_MIN) | 3379 | CFQ_HW_QUEUE_MIN && cfqd->rq_in_driver < CFQ_HW_QUEUE_MIN) |
3375 | return; | 3380 | return; |
3376 | 3381 | ||
3377 | if (cfqd->hw_tag_samples++ < 50) | 3382 | if (cfqd->hw_tag_samples++ < 50) |
3378 | return; | 3383 | return; |
3379 | 3384 | ||
3380 | if (cfqd->hw_tag_est_depth >= CFQ_HW_QUEUE_MIN) | 3385 | if (cfqd->hw_tag_est_depth >= CFQ_HW_QUEUE_MIN) |
3381 | cfqd->hw_tag = 1; | 3386 | cfqd->hw_tag = 1; |
3382 | else | 3387 | else |
3383 | cfqd->hw_tag = 0; | 3388 | cfqd->hw_tag = 0; |
3384 | } | 3389 | } |
3385 | 3390 | ||
3386 | static bool cfq_should_wait_busy(struct cfq_data *cfqd, struct cfq_queue *cfqq) | 3391 | static bool cfq_should_wait_busy(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
3387 | { | 3392 | { |
3388 | struct cfq_io_context *cic = cfqd->active_cic; | 3393 | struct cfq_io_context *cic = cfqd->active_cic; |
3389 | 3394 | ||
3390 | /* If there are other queues in the group, don't wait */ | 3395 | /* If there are other queues in the group, don't wait */ |
3391 | if (cfqq->cfqg->nr_cfqq > 1) | 3396 | if (cfqq->cfqg->nr_cfqq > 1) |
3392 | return false; | 3397 | return false; |
3393 | 3398 | ||
3394 | if (cfq_slice_used(cfqq)) | 3399 | if (cfq_slice_used(cfqq)) |
3395 | return true; | 3400 | return true; |
3396 | 3401 | ||
3397 | /* if slice left is less than think time, wait busy */ | 3402 | /* if slice left is less than think time, wait busy */ |
3398 | if (cic && sample_valid(cic->ttime_samples) | 3403 | if (cic && sample_valid(cic->ttime_samples) |
3399 | && (cfqq->slice_end - jiffies < cic->ttime_mean)) | 3404 | && (cfqq->slice_end - jiffies < cic->ttime_mean)) |
3400 | return true; | 3405 | return true; |
3401 | 3406 | ||
3402 | /* | 3407 | /* |
3403 | * If think times is less than a jiffy than ttime_mean=0 and above | 3408 | * If think times is less than a jiffy than ttime_mean=0 and above |
3404 | * will not be true. It might happen that slice has not expired yet | 3409 | * will not be true. It might happen that slice has not expired yet |
3405 | * but will expire soon (4-5 ns) during select_queue(). To cover the | 3410 | * but will expire soon (4-5 ns) during select_queue(). To cover the |
3406 | * case where think time is less than a jiffy, mark the queue wait | 3411 | * case where think time is less than a jiffy, mark the queue wait |
3407 | * busy if only 1 jiffy is left in the slice. | 3412 | * busy if only 1 jiffy is left in the slice. |
3408 | */ | 3413 | */ |
3409 | if (cfqq->slice_end - jiffies == 1) | 3414 | if (cfqq->slice_end - jiffies == 1) |
3410 | return true; | 3415 | return true; |
3411 | 3416 | ||
3412 | return false; | 3417 | return false; |
3413 | } | 3418 | } |
3414 | 3419 | ||
3415 | static void cfq_completed_request(struct request_queue *q, struct request *rq) | 3420 | static void cfq_completed_request(struct request_queue *q, struct request *rq) |
3416 | { | 3421 | { |
3417 | struct cfq_queue *cfqq = RQ_CFQQ(rq); | 3422 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
3418 | struct cfq_data *cfqd = cfqq->cfqd; | 3423 | struct cfq_data *cfqd = cfqq->cfqd; |
3419 | const int sync = rq_is_sync(rq); | 3424 | const int sync = rq_is_sync(rq); |
3420 | unsigned long now; | 3425 | unsigned long now; |
3421 | 3426 | ||
3422 | now = jiffies; | 3427 | now = jiffies; |
3423 | cfq_log_cfqq(cfqd, cfqq, "complete rqnoidle %d", | 3428 | cfq_log_cfqq(cfqd, cfqq, "complete rqnoidle %d", |
3424 | !!(rq->cmd_flags & REQ_NOIDLE)); | 3429 | !!(rq->cmd_flags & REQ_NOIDLE)); |
3425 | 3430 | ||
3426 | cfq_update_hw_tag(cfqd); | 3431 | cfq_update_hw_tag(cfqd); |
3427 | 3432 | ||
3428 | WARN_ON(!cfqd->rq_in_driver); | 3433 | WARN_ON(!cfqd->rq_in_driver); |
3429 | WARN_ON(!cfqq->dispatched); | 3434 | WARN_ON(!cfqq->dispatched); |
3430 | cfqd->rq_in_driver--; | 3435 | cfqd->rq_in_driver--; |
3431 | cfqq->dispatched--; | 3436 | cfqq->dispatched--; |
3432 | (RQ_CFQG(rq))->dispatched--; | 3437 | (RQ_CFQG(rq))->dispatched--; |
3433 | cfq_blkiocg_update_completion_stats(&cfqq->cfqg->blkg, | 3438 | cfq_blkiocg_update_completion_stats(&cfqq->cfqg->blkg, |
3434 | rq_start_time_ns(rq), rq_io_start_time_ns(rq), | 3439 | rq_start_time_ns(rq), rq_io_start_time_ns(rq), |
3435 | rq_data_dir(rq), rq_is_sync(rq)); | 3440 | rq_data_dir(rq), rq_is_sync(rq)); |
3436 | 3441 | ||
3437 | cfqd->rq_in_flight[cfq_cfqq_sync(cfqq)]--; | 3442 | cfqd->rq_in_flight[cfq_cfqq_sync(cfqq)]--; |
3438 | 3443 | ||
3439 | if (sync) { | 3444 | if (sync) { |
3440 | RQ_CIC(rq)->last_end_request = now; | 3445 | RQ_CIC(rq)->last_end_request = now; |
3441 | if (!time_after(rq->start_time + cfqd->cfq_fifo_expire[1], now)) | 3446 | if (!time_after(rq->start_time + cfqd->cfq_fifo_expire[1], now)) |
3442 | cfqd->last_delayed_sync = now; | 3447 | cfqd->last_delayed_sync = now; |
3443 | } | 3448 | } |
3444 | 3449 | ||
3445 | /* | 3450 | /* |
3446 | * If this is the active queue, check if it needs to be expired, | 3451 | * If this is the active queue, check if it needs to be expired, |
3447 | * or if we want to idle in case it has no pending requests. | 3452 | * or if we want to idle in case it has no pending requests. |
3448 | */ | 3453 | */ |
3449 | if (cfqd->active_queue == cfqq) { | 3454 | if (cfqd->active_queue == cfqq) { |
3450 | const bool cfqq_empty = RB_EMPTY_ROOT(&cfqq->sort_list); | 3455 | const bool cfqq_empty = RB_EMPTY_ROOT(&cfqq->sort_list); |
3451 | 3456 | ||
3452 | if (cfq_cfqq_slice_new(cfqq)) { | 3457 | if (cfq_cfqq_slice_new(cfqq)) { |
3453 | cfq_set_prio_slice(cfqd, cfqq); | 3458 | cfq_set_prio_slice(cfqd, cfqq); |
3454 | cfq_clear_cfqq_slice_new(cfqq); | 3459 | cfq_clear_cfqq_slice_new(cfqq); |
3455 | } | 3460 | } |
3456 | 3461 | ||
3457 | /* | 3462 | /* |
3458 | * Should we wait for next request to come in before we expire | 3463 | * Should we wait for next request to come in before we expire |
3459 | * the queue. | 3464 | * the queue. |
3460 | */ | 3465 | */ |
3461 | if (cfq_should_wait_busy(cfqd, cfqq)) { | 3466 | if (cfq_should_wait_busy(cfqd, cfqq)) { |
3462 | unsigned long extend_sl = cfqd->cfq_slice_idle; | 3467 | unsigned long extend_sl = cfqd->cfq_slice_idle; |
3463 | if (!cfqd->cfq_slice_idle) | 3468 | if (!cfqd->cfq_slice_idle) |
3464 | extend_sl = cfqd->cfq_group_idle; | 3469 | extend_sl = cfqd->cfq_group_idle; |
3465 | cfqq->slice_end = jiffies + extend_sl; | 3470 | cfqq->slice_end = jiffies + extend_sl; |
3466 | cfq_mark_cfqq_wait_busy(cfqq); | 3471 | cfq_mark_cfqq_wait_busy(cfqq); |
3467 | cfq_log_cfqq(cfqd, cfqq, "will busy wait"); | 3472 | cfq_log_cfqq(cfqd, cfqq, "will busy wait"); |
3468 | } | 3473 | } |
3469 | 3474 | ||
3470 | /* | 3475 | /* |
3471 | * Idling is not enabled on: | 3476 | * Idling is not enabled on: |
3472 | * - expired queues | 3477 | * - expired queues |
3473 | * - idle-priority queues | 3478 | * - idle-priority queues |
3474 | * - async queues | 3479 | * - async queues |
3475 | * - queues with still some requests queued | 3480 | * - queues with still some requests queued |
3476 | * - when there is a close cooperator | 3481 | * - when there is a close cooperator |
3477 | */ | 3482 | */ |
3478 | if (cfq_slice_used(cfqq) || cfq_class_idle(cfqq)) | 3483 | if (cfq_slice_used(cfqq) || cfq_class_idle(cfqq)) |
3479 | cfq_slice_expired(cfqd, 1); | 3484 | cfq_slice_expired(cfqd, 1); |
3480 | else if (sync && cfqq_empty && | 3485 | else if (sync && cfqq_empty && |
3481 | !cfq_close_cooperator(cfqd, cfqq)) { | 3486 | !cfq_close_cooperator(cfqd, cfqq)) { |
3482 | cfqd->noidle_tree_requires_idle |= | 3487 | cfqd->noidle_tree_requires_idle |= |
3483 | !(rq->cmd_flags & REQ_NOIDLE); | 3488 | !(rq->cmd_flags & REQ_NOIDLE); |
3484 | /* | 3489 | /* |
3485 | * Idling is enabled for SYNC_WORKLOAD. | 3490 | * Idling is enabled for SYNC_WORKLOAD. |
3486 | * SYNC_NOIDLE_WORKLOAD idles at the end of the tree | 3491 | * SYNC_NOIDLE_WORKLOAD idles at the end of the tree |
3487 | * only if we processed at least one !REQ_NOIDLE request | 3492 | * only if we processed at least one !REQ_NOIDLE request |
3488 | */ | 3493 | */ |
3489 | if (cfqd->serving_type == SYNC_WORKLOAD | 3494 | if (cfqd->serving_type == SYNC_WORKLOAD |
3490 | || cfqd->noidle_tree_requires_idle | 3495 | || cfqd->noidle_tree_requires_idle |
3491 | || cfqq->cfqg->nr_cfqq == 1) | 3496 | || cfqq->cfqg->nr_cfqq == 1) |
3492 | cfq_arm_slice_timer(cfqd); | 3497 | cfq_arm_slice_timer(cfqd); |
3493 | } | 3498 | } |
3494 | } | 3499 | } |
3495 | 3500 | ||
3496 | if (!cfqd->rq_in_driver) | 3501 | if (!cfqd->rq_in_driver) |
3497 | cfq_schedule_dispatch(cfqd); | 3502 | cfq_schedule_dispatch(cfqd); |
3498 | } | 3503 | } |
3499 | 3504 | ||
3500 | /* | 3505 | /* |
3501 | * we temporarily boost lower priority queues if they are holding fs exclusive | 3506 | * we temporarily boost lower priority queues if they are holding fs exclusive |
3502 | * resources. they are boosted to normal prio (CLASS_BE/4) | 3507 | * resources. they are boosted to normal prio (CLASS_BE/4) |
3503 | */ | 3508 | */ |
3504 | static void cfq_prio_boost(struct cfq_queue *cfqq) | 3509 | static void cfq_prio_boost(struct cfq_queue *cfqq) |
3505 | { | 3510 | { |
3506 | if (has_fs_excl()) { | 3511 | if (has_fs_excl()) { |
3507 | /* | 3512 | /* |
3508 | * boost idle prio on transactions that would lock out other | 3513 | * boost idle prio on transactions that would lock out other |
3509 | * users of the filesystem | 3514 | * users of the filesystem |
3510 | */ | 3515 | */ |
3511 | if (cfq_class_idle(cfqq)) | 3516 | if (cfq_class_idle(cfqq)) |
3512 | cfqq->ioprio_class = IOPRIO_CLASS_BE; | 3517 | cfqq->ioprio_class = IOPRIO_CLASS_BE; |
3513 | if (cfqq->ioprio > IOPRIO_NORM) | 3518 | if (cfqq->ioprio > IOPRIO_NORM) |
3514 | cfqq->ioprio = IOPRIO_NORM; | 3519 | cfqq->ioprio = IOPRIO_NORM; |
3515 | } else { | 3520 | } else { |
3516 | /* | 3521 | /* |
3517 | * unboost the queue (if needed) | 3522 | * unboost the queue (if needed) |
3518 | */ | 3523 | */ |
3519 | cfqq->ioprio_class = cfqq->org_ioprio_class; | 3524 | cfqq->ioprio_class = cfqq->org_ioprio_class; |
3520 | cfqq->ioprio = cfqq->org_ioprio; | 3525 | cfqq->ioprio = cfqq->org_ioprio; |
3521 | } | 3526 | } |
3522 | } | 3527 | } |
3523 | 3528 | ||
3524 | static inline int __cfq_may_queue(struct cfq_queue *cfqq) | 3529 | static inline int __cfq_may_queue(struct cfq_queue *cfqq) |
3525 | { | 3530 | { |
3526 | if (cfq_cfqq_wait_request(cfqq) && !cfq_cfqq_must_alloc_slice(cfqq)) { | 3531 | if (cfq_cfqq_wait_request(cfqq) && !cfq_cfqq_must_alloc_slice(cfqq)) { |
3527 | cfq_mark_cfqq_must_alloc_slice(cfqq); | 3532 | cfq_mark_cfqq_must_alloc_slice(cfqq); |
3528 | return ELV_MQUEUE_MUST; | 3533 | return ELV_MQUEUE_MUST; |
3529 | } | 3534 | } |
3530 | 3535 | ||
3531 | return ELV_MQUEUE_MAY; | 3536 | return ELV_MQUEUE_MAY; |
3532 | } | 3537 | } |
3533 | 3538 | ||
3534 | static int cfq_may_queue(struct request_queue *q, int rw) | 3539 | static int cfq_may_queue(struct request_queue *q, int rw) |
3535 | { | 3540 | { |
3536 | struct cfq_data *cfqd = q->elevator->elevator_data; | 3541 | struct cfq_data *cfqd = q->elevator->elevator_data; |
3537 | struct task_struct *tsk = current; | 3542 | struct task_struct *tsk = current; |
3538 | struct cfq_io_context *cic; | 3543 | struct cfq_io_context *cic; |
3539 | struct cfq_queue *cfqq; | 3544 | struct cfq_queue *cfqq; |
3540 | 3545 | ||
3541 | /* | 3546 | /* |
3542 | * don't force setup of a queue from here, as a call to may_queue | 3547 | * don't force setup of a queue from here, as a call to may_queue |
3543 | * does not necessarily imply that a request actually will be queued. | 3548 | * does not necessarily imply that a request actually will be queued. |
3544 | * so just lookup a possibly existing queue, or return 'may queue' | 3549 | * so just lookup a possibly existing queue, or return 'may queue' |
3545 | * if that fails | 3550 | * if that fails |
3546 | */ | 3551 | */ |
3547 | cic = cfq_cic_lookup(cfqd, tsk->io_context); | 3552 | cic = cfq_cic_lookup(cfqd, tsk->io_context); |
3548 | if (!cic) | 3553 | if (!cic) |
3549 | return ELV_MQUEUE_MAY; | 3554 | return ELV_MQUEUE_MAY; |
3550 | 3555 | ||
3551 | cfqq = cic_to_cfqq(cic, rw_is_sync(rw)); | 3556 | cfqq = cic_to_cfqq(cic, rw_is_sync(rw)); |
3552 | if (cfqq) { | 3557 | if (cfqq) { |
3553 | cfq_init_prio_data(cfqq, cic->ioc); | 3558 | cfq_init_prio_data(cfqq, cic->ioc); |
3554 | cfq_prio_boost(cfqq); | 3559 | cfq_prio_boost(cfqq); |
3555 | 3560 | ||
3556 | return __cfq_may_queue(cfqq); | 3561 | return __cfq_may_queue(cfqq); |
3557 | } | 3562 | } |
3558 | 3563 | ||
3559 | return ELV_MQUEUE_MAY; | 3564 | return ELV_MQUEUE_MAY; |
3560 | } | 3565 | } |
3561 | 3566 | ||
3562 | /* | 3567 | /* |
3563 | * queue lock held here | 3568 | * queue lock held here |
3564 | */ | 3569 | */ |
3565 | static void cfq_put_request(struct request *rq) | 3570 | static void cfq_put_request(struct request *rq) |
3566 | { | 3571 | { |
3567 | struct cfq_queue *cfqq = RQ_CFQQ(rq); | 3572 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
3568 | 3573 | ||
3569 | if (cfqq) { | 3574 | if (cfqq) { |
3570 | const int rw = rq_data_dir(rq); | 3575 | const int rw = rq_data_dir(rq); |
3571 | 3576 | ||
3572 | BUG_ON(!cfqq->allocated[rw]); | 3577 | BUG_ON(!cfqq->allocated[rw]); |
3573 | cfqq->allocated[rw]--; | 3578 | cfqq->allocated[rw]--; |
3574 | 3579 | ||
3575 | put_io_context(RQ_CIC(rq)->ioc); | 3580 | put_io_context(RQ_CIC(rq)->ioc); |
3576 | 3581 | ||
3577 | rq->elevator_private = NULL; | 3582 | rq->elevator_private = NULL; |
3578 | rq->elevator_private2 = NULL; | 3583 | rq->elevator_private2 = NULL; |
3579 | 3584 | ||
3580 | /* Put down rq reference on cfqg */ | 3585 | /* Put down rq reference on cfqg */ |
3581 | cfq_put_cfqg(RQ_CFQG(rq)); | 3586 | cfq_put_cfqg(RQ_CFQG(rq)); |
3582 | rq->elevator_private3 = NULL; | 3587 | rq->elevator_private3 = NULL; |
3583 | 3588 | ||
3584 | cfq_put_queue(cfqq); | 3589 | cfq_put_queue(cfqq); |
3585 | } | 3590 | } |
3586 | } | 3591 | } |
3587 | 3592 | ||
3588 | static struct cfq_queue * | 3593 | static struct cfq_queue * |
3589 | cfq_merge_cfqqs(struct cfq_data *cfqd, struct cfq_io_context *cic, | 3594 | cfq_merge_cfqqs(struct cfq_data *cfqd, struct cfq_io_context *cic, |
3590 | struct cfq_queue *cfqq) | 3595 | struct cfq_queue *cfqq) |
3591 | { | 3596 | { |
3592 | cfq_log_cfqq(cfqd, cfqq, "merging with queue %p", cfqq->new_cfqq); | 3597 | cfq_log_cfqq(cfqd, cfqq, "merging with queue %p", cfqq->new_cfqq); |
3593 | cic_set_cfqq(cic, cfqq->new_cfqq, 1); | 3598 | cic_set_cfqq(cic, cfqq->new_cfqq, 1); |
3594 | cfq_mark_cfqq_coop(cfqq->new_cfqq); | 3599 | cfq_mark_cfqq_coop(cfqq->new_cfqq); |
3595 | cfq_put_queue(cfqq); | 3600 | cfq_put_queue(cfqq); |
3596 | return cic_to_cfqq(cic, 1); | 3601 | return cic_to_cfqq(cic, 1); |
3597 | } | 3602 | } |
3598 | 3603 | ||
3599 | /* | 3604 | /* |
3600 | * Returns NULL if a new cfqq should be allocated, or the old cfqq if this | 3605 | * Returns NULL if a new cfqq should be allocated, or the old cfqq if this |
3601 | * was the last process referring to said cfqq. | 3606 | * was the last process referring to said cfqq. |
3602 | */ | 3607 | */ |
3603 | static struct cfq_queue * | 3608 | static struct cfq_queue * |
3604 | split_cfqq(struct cfq_io_context *cic, struct cfq_queue *cfqq) | 3609 | split_cfqq(struct cfq_io_context *cic, struct cfq_queue *cfqq) |
3605 | { | 3610 | { |
3606 | if (cfqq_process_refs(cfqq) == 1) { | 3611 | if (cfqq_process_refs(cfqq) == 1) { |
3607 | cfqq->pid = current->pid; | 3612 | cfqq->pid = current->pid; |
3608 | cfq_clear_cfqq_coop(cfqq); | 3613 | cfq_clear_cfqq_coop(cfqq); |
3609 | cfq_clear_cfqq_split_coop(cfqq); | 3614 | cfq_clear_cfqq_split_coop(cfqq); |
3610 | return cfqq; | 3615 | return cfqq; |
3611 | } | 3616 | } |
3612 | 3617 | ||
3613 | cic_set_cfqq(cic, NULL, 1); | 3618 | cic_set_cfqq(cic, NULL, 1); |
3614 | 3619 | ||
3615 | cfq_put_cooperator(cfqq); | 3620 | cfq_put_cooperator(cfqq); |
3616 | 3621 | ||
3617 | cfq_put_queue(cfqq); | 3622 | cfq_put_queue(cfqq); |
3618 | return NULL; | 3623 | return NULL; |
3619 | } | 3624 | } |
3620 | /* | 3625 | /* |
3621 | * Allocate cfq data structures associated with this request. | 3626 | * Allocate cfq data structures associated with this request. |
3622 | */ | 3627 | */ |
3623 | static int | 3628 | static int |
3624 | cfq_set_request(struct request_queue *q, struct request *rq, gfp_t gfp_mask) | 3629 | cfq_set_request(struct request_queue *q, struct request *rq, gfp_t gfp_mask) |
3625 | { | 3630 | { |
3626 | struct cfq_data *cfqd = q->elevator->elevator_data; | 3631 | struct cfq_data *cfqd = q->elevator->elevator_data; |
3627 | struct cfq_io_context *cic; | 3632 | struct cfq_io_context *cic; |
3628 | const int rw = rq_data_dir(rq); | 3633 | const int rw = rq_data_dir(rq); |
3629 | const bool is_sync = rq_is_sync(rq); | 3634 | const bool is_sync = rq_is_sync(rq); |
3630 | struct cfq_queue *cfqq; | 3635 | struct cfq_queue *cfqq; |
3631 | unsigned long flags; | 3636 | unsigned long flags; |
3632 | 3637 | ||
3633 | might_sleep_if(gfp_mask & __GFP_WAIT); | 3638 | might_sleep_if(gfp_mask & __GFP_WAIT); |
3634 | 3639 | ||
3635 | cic = cfq_get_io_context(cfqd, gfp_mask); | 3640 | cic = cfq_get_io_context(cfqd, gfp_mask); |
3636 | 3641 | ||
3637 | spin_lock_irqsave(q->queue_lock, flags); | 3642 | spin_lock_irqsave(q->queue_lock, flags); |
3638 | 3643 | ||
3639 | if (!cic) | 3644 | if (!cic) |
3640 | goto queue_fail; | 3645 | goto queue_fail; |
3641 | 3646 | ||
3642 | new_queue: | 3647 | new_queue: |
3643 | cfqq = cic_to_cfqq(cic, is_sync); | 3648 | cfqq = cic_to_cfqq(cic, is_sync); |
3644 | if (!cfqq || cfqq == &cfqd->oom_cfqq) { | 3649 | if (!cfqq || cfqq == &cfqd->oom_cfqq) { |
3645 | cfqq = cfq_get_queue(cfqd, is_sync, cic->ioc, gfp_mask); | 3650 | cfqq = cfq_get_queue(cfqd, is_sync, cic->ioc, gfp_mask); |
3646 | cic_set_cfqq(cic, cfqq, is_sync); | 3651 | cic_set_cfqq(cic, cfqq, is_sync); |
3647 | } else { | 3652 | } else { |
3648 | /* | 3653 | /* |
3649 | * If the queue was seeky for too long, break it apart. | 3654 | * If the queue was seeky for too long, break it apart. |
3650 | */ | 3655 | */ |
3651 | if (cfq_cfqq_coop(cfqq) && cfq_cfqq_split_coop(cfqq)) { | 3656 | if (cfq_cfqq_coop(cfqq) && cfq_cfqq_split_coop(cfqq)) { |
3652 | cfq_log_cfqq(cfqd, cfqq, "breaking apart cfqq"); | 3657 | cfq_log_cfqq(cfqd, cfqq, "breaking apart cfqq"); |
3653 | cfqq = split_cfqq(cic, cfqq); | 3658 | cfqq = split_cfqq(cic, cfqq); |
3654 | if (!cfqq) | 3659 | if (!cfqq) |
3655 | goto new_queue; | 3660 | goto new_queue; |
3656 | } | 3661 | } |
3657 | 3662 | ||
3658 | /* | 3663 | /* |
3659 | * Check to see if this queue is scheduled to merge with | 3664 | * Check to see if this queue is scheduled to merge with |
3660 | * another, closely cooperating queue. The merging of | 3665 | * another, closely cooperating queue. The merging of |
3661 | * queues happens here as it must be done in process context. | 3666 | * queues happens here as it must be done in process context. |
3662 | * The reference on new_cfqq was taken in merge_cfqqs. | 3667 | * The reference on new_cfqq was taken in merge_cfqqs. |
3663 | */ | 3668 | */ |
3664 | if (cfqq->new_cfqq) | 3669 | if (cfqq->new_cfqq) |
3665 | cfqq = cfq_merge_cfqqs(cfqd, cic, cfqq); | 3670 | cfqq = cfq_merge_cfqqs(cfqd, cic, cfqq); |
3666 | } | 3671 | } |
3667 | 3672 | ||
3668 | cfqq->allocated[rw]++; | 3673 | cfqq->allocated[rw]++; |
3669 | atomic_inc(&cfqq->ref); | 3674 | atomic_inc(&cfqq->ref); |
3670 | 3675 | ||
3671 | spin_unlock_irqrestore(q->queue_lock, flags); | 3676 | spin_unlock_irqrestore(q->queue_lock, flags); |
3672 | 3677 | ||
3673 | rq->elevator_private = cic; | 3678 | rq->elevator_private = cic; |
3674 | rq->elevator_private2 = cfqq; | 3679 | rq->elevator_private2 = cfqq; |
3675 | rq->elevator_private3 = cfq_ref_get_cfqg(cfqq->cfqg); | 3680 | rq->elevator_private3 = cfq_ref_get_cfqg(cfqq->cfqg); |
3676 | return 0; | 3681 | return 0; |
3677 | 3682 | ||
3678 | queue_fail: | 3683 | queue_fail: |
3679 | if (cic) | 3684 | if (cic) |
3680 | put_io_context(cic->ioc); | 3685 | put_io_context(cic->ioc); |
3681 | 3686 | ||
3682 | cfq_schedule_dispatch(cfqd); | 3687 | cfq_schedule_dispatch(cfqd); |
3683 | spin_unlock_irqrestore(q->queue_lock, flags); | 3688 | spin_unlock_irqrestore(q->queue_lock, flags); |
3684 | cfq_log(cfqd, "set_request fail"); | 3689 | cfq_log(cfqd, "set_request fail"); |
3685 | return 1; | 3690 | return 1; |
3686 | } | 3691 | } |
3687 | 3692 | ||
3688 | static void cfq_kick_queue(struct work_struct *work) | 3693 | static void cfq_kick_queue(struct work_struct *work) |
3689 | { | 3694 | { |
3690 | struct cfq_data *cfqd = | 3695 | struct cfq_data *cfqd = |
3691 | container_of(work, struct cfq_data, unplug_work); | 3696 | container_of(work, struct cfq_data, unplug_work); |
3692 | struct request_queue *q = cfqd->queue; | 3697 | struct request_queue *q = cfqd->queue; |
3693 | 3698 | ||
3694 | spin_lock_irq(q->queue_lock); | 3699 | spin_lock_irq(q->queue_lock); |
3695 | __blk_run_queue(cfqd->queue); | 3700 | __blk_run_queue(cfqd->queue); |
3696 | spin_unlock_irq(q->queue_lock); | 3701 | spin_unlock_irq(q->queue_lock); |
3697 | } | 3702 | } |
3698 | 3703 | ||
3699 | /* | 3704 | /* |
3700 | * Timer running if the active_queue is currently idling inside its time slice | 3705 | * Timer running if the active_queue is currently idling inside its time slice |
3701 | */ | 3706 | */ |
3702 | static void cfq_idle_slice_timer(unsigned long data) | 3707 | static void cfq_idle_slice_timer(unsigned long data) |
3703 | { | 3708 | { |
3704 | struct cfq_data *cfqd = (struct cfq_data *) data; | 3709 | struct cfq_data *cfqd = (struct cfq_data *) data; |
3705 | struct cfq_queue *cfqq; | 3710 | struct cfq_queue *cfqq; |
3706 | unsigned long flags; | 3711 | unsigned long flags; |
3707 | int timed_out = 1; | 3712 | int timed_out = 1; |
3708 | 3713 | ||
3709 | cfq_log(cfqd, "idle timer fired"); | 3714 | cfq_log(cfqd, "idle timer fired"); |
3710 | 3715 | ||
3711 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); | 3716 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); |
3712 | 3717 | ||
3713 | cfqq = cfqd->active_queue; | 3718 | cfqq = cfqd->active_queue; |
3714 | if (cfqq) { | 3719 | if (cfqq) { |
3715 | timed_out = 0; | 3720 | timed_out = 0; |
3716 | 3721 | ||
3717 | /* | 3722 | /* |
3718 | * We saw a request before the queue expired, let it through | 3723 | * We saw a request before the queue expired, let it through |
3719 | */ | 3724 | */ |
3720 | if (cfq_cfqq_must_dispatch(cfqq)) | 3725 | if (cfq_cfqq_must_dispatch(cfqq)) |
3721 | goto out_kick; | 3726 | goto out_kick; |
3722 | 3727 | ||
3723 | /* | 3728 | /* |
3724 | * expired | 3729 | * expired |
3725 | */ | 3730 | */ |
3726 | if (cfq_slice_used(cfqq)) | 3731 | if (cfq_slice_used(cfqq)) |
3727 | goto expire; | 3732 | goto expire; |
3728 | 3733 | ||
3729 | /* | 3734 | /* |
3730 | * only expire and reinvoke request handler, if there are | 3735 | * only expire and reinvoke request handler, if there are |
3731 | * other queues with pending requests | 3736 | * other queues with pending requests |
3732 | */ | 3737 | */ |
3733 | if (!cfqd->busy_queues) | 3738 | if (!cfqd->busy_queues) |
3734 | goto out_cont; | 3739 | goto out_cont; |
3735 | 3740 | ||
3736 | /* | 3741 | /* |
3737 | * not expired and it has a request pending, let it dispatch | 3742 | * not expired and it has a request pending, let it dispatch |
3738 | */ | 3743 | */ |
3739 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) | 3744 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) |
3740 | goto out_kick; | 3745 | goto out_kick; |
3741 | 3746 | ||
3742 | /* | 3747 | /* |
3743 | * Queue depth flag is reset only when the idle didn't succeed | 3748 | * Queue depth flag is reset only when the idle didn't succeed |
3744 | */ | 3749 | */ |
3745 | cfq_clear_cfqq_deep(cfqq); | 3750 | cfq_clear_cfqq_deep(cfqq); |
3746 | } | 3751 | } |
3747 | expire: | 3752 | expire: |
3748 | cfq_slice_expired(cfqd, timed_out); | 3753 | cfq_slice_expired(cfqd, timed_out); |
3749 | out_kick: | 3754 | out_kick: |
3750 | cfq_schedule_dispatch(cfqd); | 3755 | cfq_schedule_dispatch(cfqd); |
3751 | out_cont: | 3756 | out_cont: |
3752 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); | 3757 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); |
3753 | } | 3758 | } |
3754 | 3759 | ||
3755 | static void cfq_shutdown_timer_wq(struct cfq_data *cfqd) | 3760 | static void cfq_shutdown_timer_wq(struct cfq_data *cfqd) |
3756 | { | 3761 | { |
3757 | del_timer_sync(&cfqd->idle_slice_timer); | 3762 | del_timer_sync(&cfqd->idle_slice_timer); |
3758 | cancel_work_sync(&cfqd->unplug_work); | 3763 | cancel_work_sync(&cfqd->unplug_work); |
3759 | } | 3764 | } |
3760 | 3765 | ||
3761 | static void cfq_put_async_queues(struct cfq_data *cfqd) | 3766 | static void cfq_put_async_queues(struct cfq_data *cfqd) |
3762 | { | 3767 | { |
3763 | int i; | 3768 | int i; |
3764 | 3769 | ||
3765 | for (i = 0; i < IOPRIO_BE_NR; i++) { | 3770 | for (i = 0; i < IOPRIO_BE_NR; i++) { |
3766 | if (cfqd->async_cfqq[0][i]) | 3771 | if (cfqd->async_cfqq[0][i]) |
3767 | cfq_put_queue(cfqd->async_cfqq[0][i]); | 3772 | cfq_put_queue(cfqd->async_cfqq[0][i]); |
3768 | if (cfqd->async_cfqq[1][i]) | 3773 | if (cfqd->async_cfqq[1][i]) |
3769 | cfq_put_queue(cfqd->async_cfqq[1][i]); | 3774 | cfq_put_queue(cfqd->async_cfqq[1][i]); |
3770 | } | 3775 | } |
3771 | 3776 | ||
3772 | if (cfqd->async_idle_cfqq) | 3777 | if (cfqd->async_idle_cfqq) |
3773 | cfq_put_queue(cfqd->async_idle_cfqq); | 3778 | cfq_put_queue(cfqd->async_idle_cfqq); |
3774 | } | 3779 | } |
3775 | 3780 | ||
3776 | static void cfq_cfqd_free(struct rcu_head *head) | 3781 | static void cfq_cfqd_free(struct rcu_head *head) |
3777 | { | 3782 | { |
3778 | kfree(container_of(head, struct cfq_data, rcu)); | 3783 | kfree(container_of(head, struct cfq_data, rcu)); |
3779 | } | 3784 | } |
3780 | 3785 | ||
3781 | static void cfq_exit_queue(struct elevator_queue *e) | 3786 | static void cfq_exit_queue(struct elevator_queue *e) |
3782 | { | 3787 | { |
3783 | struct cfq_data *cfqd = e->elevator_data; | 3788 | struct cfq_data *cfqd = e->elevator_data; |
3784 | struct request_queue *q = cfqd->queue; | 3789 | struct request_queue *q = cfqd->queue; |
3785 | 3790 | ||
3786 | cfq_shutdown_timer_wq(cfqd); | 3791 | cfq_shutdown_timer_wq(cfqd); |
3787 | 3792 | ||
3788 | spin_lock_irq(q->queue_lock); | 3793 | spin_lock_irq(q->queue_lock); |
3789 | 3794 | ||
3790 | if (cfqd->active_queue) | 3795 | if (cfqd->active_queue) |
3791 | __cfq_slice_expired(cfqd, cfqd->active_queue, 0); | 3796 | __cfq_slice_expired(cfqd, cfqd->active_queue, 0); |
3792 | 3797 | ||
3793 | while (!list_empty(&cfqd->cic_list)) { | 3798 | while (!list_empty(&cfqd->cic_list)) { |
3794 | struct cfq_io_context *cic = list_entry(cfqd->cic_list.next, | 3799 | struct cfq_io_context *cic = list_entry(cfqd->cic_list.next, |
3795 | struct cfq_io_context, | 3800 | struct cfq_io_context, |
3796 | queue_list); | 3801 | queue_list); |
3797 | 3802 | ||
3798 | __cfq_exit_single_io_context(cfqd, cic); | 3803 | __cfq_exit_single_io_context(cfqd, cic); |
3799 | } | 3804 | } |
3800 | 3805 | ||
3801 | cfq_put_async_queues(cfqd); | 3806 | cfq_put_async_queues(cfqd); |
3802 | cfq_release_cfq_groups(cfqd); | 3807 | cfq_release_cfq_groups(cfqd); |
3803 | cfq_blkiocg_del_blkio_group(&cfqd->root_group.blkg); | 3808 | cfq_blkiocg_del_blkio_group(&cfqd->root_group.blkg); |
3804 | 3809 | ||
3805 | spin_unlock_irq(q->queue_lock); | 3810 | spin_unlock_irq(q->queue_lock); |
3806 | 3811 | ||
3807 | cfq_shutdown_timer_wq(cfqd); | 3812 | cfq_shutdown_timer_wq(cfqd); |
3808 | 3813 | ||
3809 | spin_lock(&cic_index_lock); | 3814 | spin_lock(&cic_index_lock); |
3810 | ida_remove(&cic_index_ida, cfqd->cic_index); | 3815 | ida_remove(&cic_index_ida, cfqd->cic_index); |
3811 | spin_unlock(&cic_index_lock); | 3816 | spin_unlock(&cic_index_lock); |
3812 | 3817 | ||
3813 | /* Wait for cfqg->blkg->key accessors to exit their grace periods. */ | 3818 | /* Wait for cfqg->blkg->key accessors to exit their grace periods. */ |
3814 | call_rcu(&cfqd->rcu, cfq_cfqd_free); | 3819 | call_rcu(&cfqd->rcu, cfq_cfqd_free); |
3815 | } | 3820 | } |
3816 | 3821 | ||
3817 | static int cfq_alloc_cic_index(void) | 3822 | static int cfq_alloc_cic_index(void) |
3818 | { | 3823 | { |
3819 | int index, error; | 3824 | int index, error; |
3820 | 3825 | ||
3821 | do { | 3826 | do { |
3822 | if (!ida_pre_get(&cic_index_ida, GFP_KERNEL)) | 3827 | if (!ida_pre_get(&cic_index_ida, GFP_KERNEL)) |
3823 | return -ENOMEM; | 3828 | return -ENOMEM; |
3824 | 3829 | ||
3825 | spin_lock(&cic_index_lock); | 3830 | spin_lock(&cic_index_lock); |
3826 | error = ida_get_new(&cic_index_ida, &index); | 3831 | error = ida_get_new(&cic_index_ida, &index); |
3827 | spin_unlock(&cic_index_lock); | 3832 | spin_unlock(&cic_index_lock); |
3828 | if (error && error != -EAGAIN) | 3833 | if (error && error != -EAGAIN) |
3829 | return error; | 3834 | return error; |
3830 | } while (error); | 3835 | } while (error); |
3831 | 3836 | ||
3832 | return index; | 3837 | return index; |
3833 | } | 3838 | } |
3834 | 3839 | ||
3835 | static void *cfq_init_queue(struct request_queue *q) | 3840 | static void *cfq_init_queue(struct request_queue *q) |
3836 | { | 3841 | { |
3837 | struct cfq_data *cfqd; | 3842 | struct cfq_data *cfqd; |
3838 | int i, j; | 3843 | int i, j; |
3839 | struct cfq_group *cfqg; | 3844 | struct cfq_group *cfqg; |
3840 | struct cfq_rb_root *st; | 3845 | struct cfq_rb_root *st; |
3841 | 3846 | ||
3842 | i = cfq_alloc_cic_index(); | 3847 | i = cfq_alloc_cic_index(); |
3843 | if (i < 0) | 3848 | if (i < 0) |
3844 | return NULL; | 3849 | return NULL; |
3845 | 3850 | ||
3846 | cfqd = kmalloc_node(sizeof(*cfqd), GFP_KERNEL | __GFP_ZERO, q->node); | 3851 | cfqd = kmalloc_node(sizeof(*cfqd), GFP_KERNEL | __GFP_ZERO, q->node); |
3847 | if (!cfqd) | 3852 | if (!cfqd) |
3848 | return NULL; | 3853 | return NULL; |
3849 | 3854 | ||
3850 | cfqd->cic_index = i; | 3855 | cfqd->cic_index = i; |
3851 | 3856 | ||
3852 | /* Init root service tree */ | 3857 | /* Init root service tree */ |
3853 | cfqd->grp_service_tree = CFQ_RB_ROOT; | 3858 | cfqd->grp_service_tree = CFQ_RB_ROOT; |
3854 | 3859 | ||
3855 | /* Init root group */ | 3860 | /* Init root group */ |
3856 | cfqg = &cfqd->root_group; | 3861 | cfqg = &cfqd->root_group; |
3857 | for_each_cfqg_st(cfqg, i, j, st) | 3862 | for_each_cfqg_st(cfqg, i, j, st) |
3858 | *st = CFQ_RB_ROOT; | 3863 | *st = CFQ_RB_ROOT; |
3859 | RB_CLEAR_NODE(&cfqg->rb_node); | 3864 | RB_CLEAR_NODE(&cfqg->rb_node); |
3860 | 3865 | ||
3861 | /* Give preference to root group over other groups */ | 3866 | /* Give preference to root group over other groups */ |
3862 | cfqg->weight = 2*BLKIO_WEIGHT_DEFAULT; | 3867 | cfqg->weight = 2*BLKIO_WEIGHT_DEFAULT; |
3863 | 3868 | ||
3864 | #ifdef CONFIG_CFQ_GROUP_IOSCHED | 3869 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3865 | /* | 3870 | /* |
3866 | * Take a reference to root group which we never drop. This is just | 3871 | * Take a reference to root group which we never drop. This is just |
3867 | * to make sure that cfq_put_cfqg() does not try to kfree root group | 3872 | * to make sure that cfq_put_cfqg() does not try to kfree root group |
3868 | */ | 3873 | */ |
3869 | atomic_set(&cfqg->ref, 1); | 3874 | atomic_set(&cfqg->ref, 1); |
3870 | rcu_read_lock(); | 3875 | rcu_read_lock(); |
3871 | cfq_blkiocg_add_blkio_group(&blkio_root_cgroup, &cfqg->blkg, | 3876 | cfq_blkiocg_add_blkio_group(&blkio_root_cgroup, &cfqg->blkg, |
3872 | (void *)cfqd, 0); | 3877 | (void *)cfqd, 0); |
3873 | rcu_read_unlock(); | 3878 | rcu_read_unlock(); |
3874 | #endif | 3879 | #endif |
3875 | /* | 3880 | /* |
3876 | * Not strictly needed (since RB_ROOT just clears the node and we | 3881 | * Not strictly needed (since RB_ROOT just clears the node and we |
3877 | * zeroed cfqd on alloc), but better be safe in case someone decides | 3882 | * zeroed cfqd on alloc), but better be safe in case someone decides |
3878 | * to add magic to the rb code | 3883 | * to add magic to the rb code |
3879 | */ | 3884 | */ |
3880 | for (i = 0; i < CFQ_PRIO_LISTS; i++) | 3885 | for (i = 0; i < CFQ_PRIO_LISTS; i++) |
3881 | cfqd->prio_trees[i] = RB_ROOT; | 3886 | cfqd->prio_trees[i] = RB_ROOT; |
3882 | 3887 | ||
3883 | /* | 3888 | /* |
3884 | * Our fallback cfqq if cfq_find_alloc_queue() runs into OOM issues. | 3889 | * Our fallback cfqq if cfq_find_alloc_queue() runs into OOM issues. |
3885 | * Grab a permanent reference to it, so that the normal code flow | 3890 | * Grab a permanent reference to it, so that the normal code flow |
3886 | * will not attempt to free it. | 3891 | * will not attempt to free it. |
3887 | */ | 3892 | */ |
3888 | cfq_init_cfqq(cfqd, &cfqd->oom_cfqq, 1, 0); | 3893 | cfq_init_cfqq(cfqd, &cfqd->oom_cfqq, 1, 0); |
3889 | atomic_inc(&cfqd->oom_cfqq.ref); | 3894 | atomic_inc(&cfqd->oom_cfqq.ref); |
3890 | cfq_link_cfqq_cfqg(&cfqd->oom_cfqq, &cfqd->root_group); | 3895 | cfq_link_cfqq_cfqg(&cfqd->oom_cfqq, &cfqd->root_group); |
3891 | 3896 | ||
3892 | INIT_LIST_HEAD(&cfqd->cic_list); | 3897 | INIT_LIST_HEAD(&cfqd->cic_list); |
3893 | 3898 | ||
3894 | cfqd->queue = q; | 3899 | cfqd->queue = q; |
3895 | 3900 | ||
3896 | init_timer(&cfqd->idle_slice_timer); | 3901 | init_timer(&cfqd->idle_slice_timer); |
3897 | cfqd->idle_slice_timer.function = cfq_idle_slice_timer; | 3902 | cfqd->idle_slice_timer.function = cfq_idle_slice_timer; |
3898 | cfqd->idle_slice_timer.data = (unsigned long) cfqd; | 3903 | cfqd->idle_slice_timer.data = (unsigned long) cfqd; |
3899 | 3904 | ||
3900 | INIT_WORK(&cfqd->unplug_work, cfq_kick_queue); | 3905 | INIT_WORK(&cfqd->unplug_work, cfq_kick_queue); |
3901 | 3906 | ||
3902 | cfqd->cfq_quantum = cfq_quantum; | 3907 | cfqd->cfq_quantum = cfq_quantum; |
3903 | cfqd->cfq_fifo_expire[0] = cfq_fifo_expire[0]; | 3908 | cfqd->cfq_fifo_expire[0] = cfq_fifo_expire[0]; |
3904 | cfqd->cfq_fifo_expire[1] = cfq_fifo_expire[1]; | 3909 | cfqd->cfq_fifo_expire[1] = cfq_fifo_expire[1]; |
3905 | cfqd->cfq_back_max = cfq_back_max; | 3910 | cfqd->cfq_back_max = cfq_back_max; |
3906 | cfqd->cfq_back_penalty = cfq_back_penalty; | 3911 | cfqd->cfq_back_penalty = cfq_back_penalty; |
3907 | cfqd->cfq_slice[0] = cfq_slice_async; | 3912 | cfqd->cfq_slice[0] = cfq_slice_async; |
3908 | cfqd->cfq_slice[1] = cfq_slice_sync; | 3913 | cfqd->cfq_slice[1] = cfq_slice_sync; |
3909 | cfqd->cfq_slice_async_rq = cfq_slice_async_rq; | 3914 | cfqd->cfq_slice_async_rq = cfq_slice_async_rq; |
3910 | cfqd->cfq_slice_idle = cfq_slice_idle; | 3915 | cfqd->cfq_slice_idle = cfq_slice_idle; |
3911 | cfqd->cfq_group_idle = cfq_group_idle; | 3916 | cfqd->cfq_group_idle = cfq_group_idle; |
3912 | cfqd->cfq_latency = 1; | 3917 | cfqd->cfq_latency = 1; |
3913 | cfqd->cfq_group_isolation = 0; | 3918 | cfqd->cfq_group_isolation = 0; |
3914 | cfqd->hw_tag = -1; | 3919 | cfqd->hw_tag = -1; |
3915 | /* | 3920 | /* |
3916 | * we optimistically start assuming sync ops weren't delayed in last | 3921 | * we optimistically start assuming sync ops weren't delayed in last |
3917 | * second, in order to have larger depth for async operations. | 3922 | * second, in order to have larger depth for async operations. |
3918 | */ | 3923 | */ |
3919 | cfqd->last_delayed_sync = jiffies - HZ; | 3924 | cfqd->last_delayed_sync = jiffies - HZ; |
3920 | return cfqd; | 3925 | return cfqd; |
3921 | } | 3926 | } |
3922 | 3927 | ||
3923 | static void cfq_slab_kill(void) | 3928 | static void cfq_slab_kill(void) |
3924 | { | 3929 | { |
3925 | /* | 3930 | /* |
3926 | * Caller already ensured that pending RCU callbacks are completed, | 3931 | * Caller already ensured that pending RCU callbacks are completed, |
3927 | * so we should have no busy allocations at this point. | 3932 | * so we should have no busy allocations at this point. |
3928 | */ | 3933 | */ |
3929 | if (cfq_pool) | 3934 | if (cfq_pool) |
3930 | kmem_cache_destroy(cfq_pool); | 3935 | kmem_cache_destroy(cfq_pool); |
3931 | if (cfq_ioc_pool) | 3936 | if (cfq_ioc_pool) |
3932 | kmem_cache_destroy(cfq_ioc_pool); | 3937 | kmem_cache_destroy(cfq_ioc_pool); |
3933 | } | 3938 | } |
3934 | 3939 | ||
3935 | static int __init cfq_slab_setup(void) | 3940 | static int __init cfq_slab_setup(void) |
3936 | { | 3941 | { |
3937 | cfq_pool = KMEM_CACHE(cfq_queue, 0); | 3942 | cfq_pool = KMEM_CACHE(cfq_queue, 0); |
3938 | if (!cfq_pool) | 3943 | if (!cfq_pool) |
3939 | goto fail; | 3944 | goto fail; |
3940 | 3945 | ||
3941 | cfq_ioc_pool = KMEM_CACHE(cfq_io_context, 0); | 3946 | cfq_ioc_pool = KMEM_CACHE(cfq_io_context, 0); |
3942 | if (!cfq_ioc_pool) | 3947 | if (!cfq_ioc_pool) |
3943 | goto fail; | 3948 | goto fail; |
3944 | 3949 | ||
3945 | return 0; | 3950 | return 0; |
3946 | fail: | 3951 | fail: |
3947 | cfq_slab_kill(); | 3952 | cfq_slab_kill(); |
3948 | return -ENOMEM; | 3953 | return -ENOMEM; |
3949 | } | 3954 | } |
3950 | 3955 | ||
3951 | /* | 3956 | /* |
3952 | * sysfs parts below --> | 3957 | * sysfs parts below --> |
3953 | */ | 3958 | */ |
3954 | static ssize_t | 3959 | static ssize_t |
3955 | cfq_var_show(unsigned int var, char *page) | 3960 | cfq_var_show(unsigned int var, char *page) |
3956 | { | 3961 | { |
3957 | return sprintf(page, "%d\n", var); | 3962 | return sprintf(page, "%d\n", var); |
3958 | } | 3963 | } |
3959 | 3964 | ||
3960 | static ssize_t | 3965 | static ssize_t |
3961 | cfq_var_store(unsigned int *var, const char *page, size_t count) | 3966 | cfq_var_store(unsigned int *var, const char *page, size_t count) |
3962 | { | 3967 | { |
3963 | char *p = (char *) page; | 3968 | char *p = (char *) page; |
3964 | 3969 | ||
3965 | *var = simple_strtoul(p, &p, 10); | 3970 | *var = simple_strtoul(p, &p, 10); |
3966 | return count; | 3971 | return count; |
3967 | } | 3972 | } |
3968 | 3973 | ||
3969 | #define SHOW_FUNCTION(__FUNC, __VAR, __CONV) \ | 3974 | #define SHOW_FUNCTION(__FUNC, __VAR, __CONV) \ |
3970 | static ssize_t __FUNC(struct elevator_queue *e, char *page) \ | 3975 | static ssize_t __FUNC(struct elevator_queue *e, char *page) \ |
3971 | { \ | 3976 | { \ |
3972 | struct cfq_data *cfqd = e->elevator_data; \ | 3977 | struct cfq_data *cfqd = e->elevator_data; \ |
3973 | unsigned int __data = __VAR; \ | 3978 | unsigned int __data = __VAR; \ |
3974 | if (__CONV) \ | 3979 | if (__CONV) \ |
3975 | __data = jiffies_to_msecs(__data); \ | 3980 | __data = jiffies_to_msecs(__data); \ |
3976 | return cfq_var_show(__data, (page)); \ | 3981 | return cfq_var_show(__data, (page)); \ |
3977 | } | 3982 | } |
3978 | SHOW_FUNCTION(cfq_quantum_show, cfqd->cfq_quantum, 0); | 3983 | SHOW_FUNCTION(cfq_quantum_show, cfqd->cfq_quantum, 0); |
3979 | SHOW_FUNCTION(cfq_fifo_expire_sync_show, cfqd->cfq_fifo_expire[1], 1); | 3984 | SHOW_FUNCTION(cfq_fifo_expire_sync_show, cfqd->cfq_fifo_expire[1], 1); |
3980 | SHOW_FUNCTION(cfq_fifo_expire_async_show, cfqd->cfq_fifo_expire[0], 1); | 3985 | SHOW_FUNCTION(cfq_fifo_expire_async_show, cfqd->cfq_fifo_expire[0], 1); |
3981 | SHOW_FUNCTION(cfq_back_seek_max_show, cfqd->cfq_back_max, 0); | 3986 | SHOW_FUNCTION(cfq_back_seek_max_show, cfqd->cfq_back_max, 0); |
3982 | SHOW_FUNCTION(cfq_back_seek_penalty_show, cfqd->cfq_back_penalty, 0); | 3987 | SHOW_FUNCTION(cfq_back_seek_penalty_show, cfqd->cfq_back_penalty, 0); |
3983 | SHOW_FUNCTION(cfq_slice_idle_show, cfqd->cfq_slice_idle, 1); | 3988 | SHOW_FUNCTION(cfq_slice_idle_show, cfqd->cfq_slice_idle, 1); |
3984 | SHOW_FUNCTION(cfq_group_idle_show, cfqd->cfq_group_idle, 1); | 3989 | SHOW_FUNCTION(cfq_group_idle_show, cfqd->cfq_group_idle, 1); |
3985 | SHOW_FUNCTION(cfq_slice_sync_show, cfqd->cfq_slice[1], 1); | 3990 | SHOW_FUNCTION(cfq_slice_sync_show, cfqd->cfq_slice[1], 1); |
3986 | SHOW_FUNCTION(cfq_slice_async_show, cfqd->cfq_slice[0], 1); | 3991 | SHOW_FUNCTION(cfq_slice_async_show, cfqd->cfq_slice[0], 1); |
3987 | SHOW_FUNCTION(cfq_slice_async_rq_show, cfqd->cfq_slice_async_rq, 0); | 3992 | SHOW_FUNCTION(cfq_slice_async_rq_show, cfqd->cfq_slice_async_rq, 0); |
3988 | SHOW_FUNCTION(cfq_low_latency_show, cfqd->cfq_latency, 0); | 3993 | SHOW_FUNCTION(cfq_low_latency_show, cfqd->cfq_latency, 0); |
3989 | SHOW_FUNCTION(cfq_group_isolation_show, cfqd->cfq_group_isolation, 0); | 3994 | SHOW_FUNCTION(cfq_group_isolation_show, cfqd->cfq_group_isolation, 0); |
3990 | #undef SHOW_FUNCTION | 3995 | #undef SHOW_FUNCTION |
3991 | 3996 | ||
3992 | #define STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, __CONV) \ | 3997 | #define STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, __CONV) \ |
3993 | static ssize_t __FUNC(struct elevator_queue *e, const char *page, size_t count) \ | 3998 | static ssize_t __FUNC(struct elevator_queue *e, const char *page, size_t count) \ |
3994 | { \ | 3999 | { \ |
3995 | struct cfq_data *cfqd = e->elevator_data; \ | 4000 | struct cfq_data *cfqd = e->elevator_data; \ |
3996 | unsigned int __data; \ | 4001 | unsigned int __data; \ |
3997 | int ret = cfq_var_store(&__data, (page), count); \ | 4002 | int ret = cfq_var_store(&__data, (page), count); \ |
3998 | if (__data < (MIN)) \ | 4003 | if (__data < (MIN)) \ |
3999 | __data = (MIN); \ | 4004 | __data = (MIN); \ |
4000 | else if (__data > (MAX)) \ | 4005 | else if (__data > (MAX)) \ |
4001 | __data = (MAX); \ | 4006 | __data = (MAX); \ |
4002 | if (__CONV) \ | 4007 | if (__CONV) \ |
4003 | *(__PTR) = msecs_to_jiffies(__data); \ | 4008 | *(__PTR) = msecs_to_jiffies(__data); \ |
4004 | else \ | 4009 | else \ |
4005 | *(__PTR) = __data; \ | 4010 | *(__PTR) = __data; \ |
4006 | return ret; \ | 4011 | return ret; \ |
4007 | } | 4012 | } |
4008 | STORE_FUNCTION(cfq_quantum_store, &cfqd->cfq_quantum, 1, UINT_MAX, 0); | 4013 | STORE_FUNCTION(cfq_quantum_store, &cfqd->cfq_quantum, 1, UINT_MAX, 0); |
4009 | STORE_FUNCTION(cfq_fifo_expire_sync_store, &cfqd->cfq_fifo_expire[1], 1, | 4014 | STORE_FUNCTION(cfq_fifo_expire_sync_store, &cfqd->cfq_fifo_expire[1], 1, |
4010 | UINT_MAX, 1); | 4015 | UINT_MAX, 1); |
4011 | STORE_FUNCTION(cfq_fifo_expire_async_store, &cfqd->cfq_fifo_expire[0], 1, | 4016 | STORE_FUNCTION(cfq_fifo_expire_async_store, &cfqd->cfq_fifo_expire[0], 1, |
4012 | UINT_MAX, 1); | 4017 | UINT_MAX, 1); |
4013 | STORE_FUNCTION(cfq_back_seek_max_store, &cfqd->cfq_back_max, 0, UINT_MAX, 0); | 4018 | STORE_FUNCTION(cfq_back_seek_max_store, &cfqd->cfq_back_max, 0, UINT_MAX, 0); |
4014 | STORE_FUNCTION(cfq_back_seek_penalty_store, &cfqd->cfq_back_penalty, 1, | 4019 | STORE_FUNCTION(cfq_back_seek_penalty_store, &cfqd->cfq_back_penalty, 1, |
4015 | UINT_MAX, 0); | 4020 | UINT_MAX, 0); |
4016 | STORE_FUNCTION(cfq_slice_idle_store, &cfqd->cfq_slice_idle, 0, UINT_MAX, 1); | 4021 | STORE_FUNCTION(cfq_slice_idle_store, &cfqd->cfq_slice_idle, 0, UINT_MAX, 1); |
4017 | STORE_FUNCTION(cfq_group_idle_store, &cfqd->cfq_group_idle, 0, UINT_MAX, 1); | 4022 | STORE_FUNCTION(cfq_group_idle_store, &cfqd->cfq_group_idle, 0, UINT_MAX, 1); |
4018 | STORE_FUNCTION(cfq_slice_sync_store, &cfqd->cfq_slice[1], 1, UINT_MAX, 1); | 4023 | STORE_FUNCTION(cfq_slice_sync_store, &cfqd->cfq_slice[1], 1, UINT_MAX, 1); |
4019 | STORE_FUNCTION(cfq_slice_async_store, &cfqd->cfq_slice[0], 1, UINT_MAX, 1); | 4024 | STORE_FUNCTION(cfq_slice_async_store, &cfqd->cfq_slice[0], 1, UINT_MAX, 1); |
4020 | STORE_FUNCTION(cfq_slice_async_rq_store, &cfqd->cfq_slice_async_rq, 1, | 4025 | STORE_FUNCTION(cfq_slice_async_rq_store, &cfqd->cfq_slice_async_rq, 1, |
4021 | UINT_MAX, 0); | 4026 | UINT_MAX, 0); |
4022 | STORE_FUNCTION(cfq_low_latency_store, &cfqd->cfq_latency, 0, 1, 0); | 4027 | STORE_FUNCTION(cfq_low_latency_store, &cfqd->cfq_latency, 0, 1, 0); |
4023 | STORE_FUNCTION(cfq_group_isolation_store, &cfqd->cfq_group_isolation, 0, 1, 0); | 4028 | STORE_FUNCTION(cfq_group_isolation_store, &cfqd->cfq_group_isolation, 0, 1, 0); |
4024 | #undef STORE_FUNCTION | 4029 | #undef STORE_FUNCTION |
4025 | 4030 | ||
4026 | #define CFQ_ATTR(name) \ | 4031 | #define CFQ_ATTR(name) \ |
4027 | __ATTR(name, S_IRUGO|S_IWUSR, cfq_##name##_show, cfq_##name##_store) | 4032 | __ATTR(name, S_IRUGO|S_IWUSR, cfq_##name##_show, cfq_##name##_store) |
4028 | 4033 | ||
4029 | static struct elv_fs_entry cfq_attrs[] = { | 4034 | static struct elv_fs_entry cfq_attrs[] = { |
4030 | CFQ_ATTR(quantum), | 4035 | CFQ_ATTR(quantum), |
4031 | CFQ_ATTR(fifo_expire_sync), | 4036 | CFQ_ATTR(fifo_expire_sync), |
4032 | CFQ_ATTR(fifo_expire_async), | 4037 | CFQ_ATTR(fifo_expire_async), |
4033 | CFQ_ATTR(back_seek_max), | 4038 | CFQ_ATTR(back_seek_max), |
4034 | CFQ_ATTR(back_seek_penalty), | 4039 | CFQ_ATTR(back_seek_penalty), |
4035 | CFQ_ATTR(slice_sync), | 4040 | CFQ_ATTR(slice_sync), |
4036 | CFQ_ATTR(slice_async), | 4041 | CFQ_ATTR(slice_async), |
4037 | CFQ_ATTR(slice_async_rq), | 4042 | CFQ_ATTR(slice_async_rq), |
4038 | CFQ_ATTR(slice_idle), | 4043 | CFQ_ATTR(slice_idle), |
4039 | CFQ_ATTR(group_idle), | 4044 | CFQ_ATTR(group_idle), |
4040 | CFQ_ATTR(low_latency), | 4045 | CFQ_ATTR(low_latency), |
4041 | CFQ_ATTR(group_isolation), | 4046 | CFQ_ATTR(group_isolation), |
4042 | __ATTR_NULL | 4047 | __ATTR_NULL |
4043 | }; | 4048 | }; |
4044 | 4049 | ||
4045 | static struct elevator_type iosched_cfq = { | 4050 | static struct elevator_type iosched_cfq = { |
4046 | .ops = { | 4051 | .ops = { |
4047 | .elevator_merge_fn = cfq_merge, | 4052 | .elevator_merge_fn = cfq_merge, |
4048 | .elevator_merged_fn = cfq_merged_request, | 4053 | .elevator_merged_fn = cfq_merged_request, |
4049 | .elevator_merge_req_fn = cfq_merged_requests, | 4054 | .elevator_merge_req_fn = cfq_merged_requests, |
4050 | .elevator_allow_merge_fn = cfq_allow_merge, | 4055 | .elevator_allow_merge_fn = cfq_allow_merge, |
4051 | .elevator_bio_merged_fn = cfq_bio_merged, | 4056 | .elevator_bio_merged_fn = cfq_bio_merged, |
4052 | .elevator_dispatch_fn = cfq_dispatch_requests, | 4057 | .elevator_dispatch_fn = cfq_dispatch_requests, |
4053 | .elevator_add_req_fn = cfq_insert_request, | 4058 | .elevator_add_req_fn = cfq_insert_request, |
4054 | .elevator_activate_req_fn = cfq_activate_request, | 4059 | .elevator_activate_req_fn = cfq_activate_request, |
4055 | .elevator_deactivate_req_fn = cfq_deactivate_request, | 4060 | .elevator_deactivate_req_fn = cfq_deactivate_request, |
4056 | .elevator_queue_empty_fn = cfq_queue_empty, | 4061 | .elevator_queue_empty_fn = cfq_queue_empty, |
4057 | .elevator_completed_req_fn = cfq_completed_request, | 4062 | .elevator_completed_req_fn = cfq_completed_request, |
4058 | .elevator_former_req_fn = elv_rb_former_request, | 4063 | .elevator_former_req_fn = elv_rb_former_request, |
4059 | .elevator_latter_req_fn = elv_rb_latter_request, | 4064 | .elevator_latter_req_fn = elv_rb_latter_request, |
4060 | .elevator_set_req_fn = cfq_set_request, | 4065 | .elevator_set_req_fn = cfq_set_request, |
4061 | .elevator_put_req_fn = cfq_put_request, | 4066 | .elevator_put_req_fn = cfq_put_request, |
4062 | .elevator_may_queue_fn = cfq_may_queue, | 4067 | .elevator_may_queue_fn = cfq_may_queue, |
4063 | .elevator_init_fn = cfq_init_queue, | 4068 | .elevator_init_fn = cfq_init_queue, |
4064 | .elevator_exit_fn = cfq_exit_queue, | 4069 | .elevator_exit_fn = cfq_exit_queue, |
4065 | .trim = cfq_free_io_context, | 4070 | .trim = cfq_free_io_context, |
4066 | }, | 4071 | }, |
4067 | .elevator_attrs = cfq_attrs, | 4072 | .elevator_attrs = cfq_attrs, |
4068 | .elevator_name = "cfq", | 4073 | .elevator_name = "cfq", |
4069 | .elevator_owner = THIS_MODULE, | 4074 | .elevator_owner = THIS_MODULE, |
4070 | }; | 4075 | }; |
4071 | 4076 | ||
4072 | #ifdef CONFIG_CFQ_GROUP_IOSCHED | 4077 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
4073 | static struct blkio_policy_type blkio_policy_cfq = { | 4078 | static struct blkio_policy_type blkio_policy_cfq = { |
4074 | .ops = { | 4079 | .ops = { |
4075 | .blkio_unlink_group_fn = cfq_unlink_blkio_group, | 4080 | .blkio_unlink_group_fn = cfq_unlink_blkio_group, |
4076 | .blkio_update_group_weight_fn = cfq_update_blkio_group_weight, | 4081 | .blkio_update_group_weight_fn = cfq_update_blkio_group_weight, |
4077 | }, | 4082 | }, |
4078 | }; | 4083 | }; |
4079 | #else | 4084 | #else |
4080 | static struct blkio_policy_type blkio_policy_cfq; | 4085 | static struct blkio_policy_type blkio_policy_cfq; |
4081 | #endif | 4086 | #endif |
4082 | 4087 | ||
4083 | static int __init cfq_init(void) | 4088 | static int __init cfq_init(void) |
4084 | { | 4089 | { |
4085 | /* | 4090 | /* |
4086 | * could be 0 on HZ < 1000 setups | 4091 | * could be 0 on HZ < 1000 setups |
4087 | */ | 4092 | */ |
4088 | if (!cfq_slice_async) | 4093 | if (!cfq_slice_async) |
4089 | cfq_slice_async = 1; | 4094 | cfq_slice_async = 1; |
4090 | if (!cfq_slice_idle) | 4095 | if (!cfq_slice_idle) |
4091 | cfq_slice_idle = 1; | 4096 | cfq_slice_idle = 1; |
4092 | 4097 | ||
4093 | #ifdef CONFIG_CFQ_GROUP_IOSCHED | 4098 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
4094 | if (!cfq_group_idle) | 4099 | if (!cfq_group_idle) |
4095 | cfq_group_idle = 1; | 4100 | cfq_group_idle = 1; |
4096 | #else | 4101 | #else |
4097 | cfq_group_idle = 0; | 4102 | cfq_group_idle = 0; |
4098 | #endif | 4103 | #endif |
4099 | if (cfq_slab_setup()) | 4104 | if (cfq_slab_setup()) |
4100 | return -ENOMEM; | 4105 | return -ENOMEM; |
4101 | 4106 | ||
4102 | elv_register(&iosched_cfq); | 4107 | elv_register(&iosched_cfq); |
4103 | blkio_policy_register(&blkio_policy_cfq); | 4108 | blkio_policy_register(&blkio_policy_cfq); |
4104 | 4109 | ||
4105 | return 0; | 4110 | return 0; |
4106 | } | 4111 | } |
4107 | 4112 | ||
4108 | static void __exit cfq_exit(void) | 4113 | static void __exit cfq_exit(void) |
4109 | { | 4114 | { |
4110 | DECLARE_COMPLETION_ONSTACK(all_gone); | 4115 | DECLARE_COMPLETION_ONSTACK(all_gone); |
4111 | blkio_policy_unregister(&blkio_policy_cfq); | 4116 | blkio_policy_unregister(&blkio_policy_cfq); |
4112 | elv_unregister(&iosched_cfq); | 4117 | elv_unregister(&iosched_cfq); |
4113 | ioc_gone = &all_gone; | 4118 | ioc_gone = &all_gone; |
4114 | /* ioc_gone's update must be visible before reading ioc_count */ | 4119 | /* ioc_gone's update must be visible before reading ioc_count */ |
4115 | smp_wmb(); | 4120 | smp_wmb(); |
4116 | 4121 | ||
4117 | /* | 4122 | /* |
4118 | * this also protects us from entering cfq_slab_kill() with | 4123 | * this also protects us from entering cfq_slab_kill() with |
4119 | * pending RCU callbacks | 4124 | * pending RCU callbacks |
4120 | */ | 4125 | */ |
4121 | if (elv_ioc_count_read(cfq_ioc_count)) | 4126 | if (elv_ioc_count_read(cfq_ioc_count)) |
4122 | wait_for_completion(&all_gone); | 4127 | wait_for_completion(&all_gone); |
4123 | ida_destroy(&cic_index_ida); | 4128 | ida_destroy(&cic_index_ida); |
4124 | cfq_slab_kill(); | 4129 | cfq_slab_kill(); |
4125 | } | 4130 | } |
4126 | 4131 | ||
4127 | module_init(cfq_init); | 4132 | module_init(cfq_init); |
4128 | module_exit(cfq_exit); | 4133 | module_exit(cfq_exit); |
4129 | 4134 | ||
4130 | MODULE_AUTHOR("Jens Axboe"); | 4135 | MODULE_AUTHOR("Jens Axboe"); |
4131 | MODULE_LICENSE("GPL"); | 4136 | MODULE_LICENSE("GPL"); |
4132 | MODULE_DESCRIPTION("Completely Fair Queueing IO scheduler"); | 4137 | MODULE_DESCRIPTION("Completely Fair Queueing IO scheduler"); |
4133 | 4138 |