]>
Commit | Line | Data |
---|---|---|
86db1e29 JA |
1 | /* |
2 | * Functions related to io context handling | |
3 | */ | |
4 | #include <linux/kernel.h> | |
5 | #include <linux/module.h> | |
6 | #include <linux/init.h> | |
7 | #include <linux/bio.h> | |
8 | #include <linux/blkdev.h> | |
9 | #include <linux/bootmem.h> /* for max_pfn/max_low_pfn */ | |
5a0e3ad6 | 10 | #include <linux/slab.h> |
86db1e29 JA |
11 | |
12 | #include "blk.h" | |
13 | ||
14 | /* | |
15 | * For io context allocations | |
16 | */ | |
17 | static struct kmem_cache *iocontext_cachep; | |
18 | ||
6e736be7 TH |
19 | /** |
20 | * get_io_context - increment reference count to io_context | |
21 | * @ioc: io_context to get | |
22 | * | |
23 | * Increment reference count to @ioc. | |
24 | */ | |
25 | void get_io_context(struct io_context *ioc) | |
26 | { | |
27 | BUG_ON(atomic_long_read(&ioc->refcount) <= 0); | |
28 | atomic_long_inc(&ioc->refcount); | |
29 | } | |
30 | EXPORT_SYMBOL(get_io_context); | |
31 | ||
7e5a8794 TH |
32 | static void icq_free_icq_rcu(struct rcu_head *head) |
33 | { | |
34 | struct io_cq *icq = container_of(head, struct io_cq, __rcu_head); | |
35 | ||
36 | kmem_cache_free(icq->__rcu_icq_cache, icq); | |
37 | } | |
38 | ||
39 | /* | |
40 | * Exit and free an icq. Called with both ioc and q locked. | |
41 | */ | |
42 | static void ioc_exit_icq(struct io_cq *icq) | |
43 | { | |
44 | struct io_context *ioc = icq->ioc; | |
45 | struct request_queue *q = icq->q; | |
46 | struct elevator_type *et = q->elevator->type; | |
47 | ||
48 | lockdep_assert_held(&ioc->lock); | |
49 | lockdep_assert_held(q->queue_lock); | |
50 | ||
51 | radix_tree_delete(&ioc->icq_tree, icq->q->id); | |
52 | hlist_del_init(&icq->ioc_node); | |
53 | list_del_init(&icq->q_node); | |
54 | ||
55 | /* | |
56 | * Both setting lookup hint to and clearing it from @icq are done | |
57 | * under queue_lock. If it's not pointing to @icq now, it never | |
58 | * will. Hint assignment itself can race safely. | |
59 | */ | |
60 | if (rcu_dereference_raw(ioc->icq_hint) == icq) | |
61 | rcu_assign_pointer(ioc->icq_hint, NULL); | |
62 | ||
11a3122f | 63 | if (et->ops.elevator_exit_icq_fn) |
7e5a8794 | 64 | et->ops.elevator_exit_icq_fn(icq); |
7e5a8794 TH |
65 | |
66 | /* | |
67 | * @icq->q might have gone away by the time RCU callback runs | |
68 | * making it impossible to determine icq_cache. Record it in @icq. | |
69 | */ | |
70 | icq->__rcu_icq_cache = et->icq_cache; | |
71 | call_rcu(&icq->__rcu_head, icq_free_icq_rcu); | |
72 | } | |
73 | ||
b2efa052 TH |
74 | /* |
75 | * Slow path for ioc release in put_io_context(). Performs double-lock | |
c5869807 | 76 | * dancing to unlink all icq's and then frees ioc. |
b2efa052 TH |
77 | */ |
78 | static void ioc_release_fn(struct work_struct *work) | |
86db1e29 | 79 | { |
b2efa052 TH |
80 | struct io_context *ioc = container_of(work, struct io_context, |
81 | release_work); | |
d8c66c5d | 82 | unsigned long flags; |
b2efa052 | 83 | |
d8c66c5d TH |
84 | /* |
85 | * Exiting icq may call into put_io_context() through elevator | |
86 | * which will trigger lockdep warning. The ioc's are guaranteed to | |
87 | * be different, use a different locking subclass here. Use | |
88 | * irqsave variant as there's no spin_lock_irq_nested(). | |
89 | */ | |
90 | spin_lock_irqsave_nested(&ioc->lock, flags, 1); | |
b2efa052 | 91 | |
c5869807 TH |
92 | while (!hlist_empty(&ioc->icq_list)) { |
93 | struct io_cq *icq = hlist_entry(ioc->icq_list.first, | |
94 | struct io_cq, ioc_node); | |
2274b029 TH |
95 | struct request_queue *q = icq->q; |
96 | ||
97 | if (spin_trylock(q->queue_lock)) { | |
98 | ioc_exit_icq(icq); | |
99 | spin_unlock(q->queue_lock); | |
100 | } else { | |
101 | spin_unlock_irqrestore(&ioc->lock, flags); | |
102 | cpu_relax(); | |
103 | spin_lock_irqsave_nested(&ioc->lock, flags, 1); | |
b2efa052 | 104 | } |
b2efa052 | 105 | } |
ffc4e759 | 106 | |
2274b029 | 107 | spin_unlock_irqrestore(&ioc->lock, flags); |
b2efa052 TH |
108 | |
109 | kmem_cache_free(iocontext_cachep, ioc); | |
86db1e29 JA |
110 | } |
111 | ||
42ec57a8 TH |
112 | /** |
113 | * put_io_context - put a reference of io_context | |
114 | * @ioc: io_context to put | |
115 | * | |
116 | * Decrement reference count of @ioc and release it if the count reaches | |
11a3122f | 117 | * zero. |
86db1e29 | 118 | */ |
11a3122f | 119 | void put_io_context(struct io_context *ioc) |
86db1e29 | 120 | { |
b2efa052 TH |
121 | unsigned long flags; |
122 | ||
86db1e29 | 123 | if (ioc == NULL) |
42ec57a8 | 124 | return; |
86db1e29 | 125 | |
42ec57a8 | 126 | BUG_ON(atomic_long_read(&ioc->refcount) <= 0); |
86db1e29 | 127 | |
b2efa052 | 128 | /* |
11a3122f TH |
129 | * Releasing ioc requires reverse order double locking and we may |
130 | * already be holding a queue_lock. Do it asynchronously from wq. | |
b2efa052 | 131 | */ |
11a3122f TH |
132 | if (atomic_long_dec_and_test(&ioc->refcount)) { |
133 | spin_lock_irqsave(&ioc->lock, flags); | |
134 | if (!hlist_empty(&ioc->icq_list)) | |
135 | schedule_work(&ioc->release_work); | |
136 | spin_unlock_irqrestore(&ioc->lock, flags); | |
b2efa052 | 137 | } |
86db1e29 | 138 | } |
b2efa052 | 139 | EXPORT_SYMBOL(put_io_context); |
86db1e29 | 140 | |
27667c99 | 141 | /* Called by the exiting task */ |
b69f2292 | 142 | void exit_io_context(struct task_struct *task) |
86db1e29 JA |
143 | { |
144 | struct io_context *ioc; | |
145 | ||
b69f2292 LR |
146 | task_lock(task); |
147 | ioc = task->io_context; | |
148 | task->io_context = NULL; | |
149 | task_unlock(task); | |
86db1e29 | 150 | |
b2efa052 | 151 | atomic_dec(&ioc->nr_tasks); |
11a3122f | 152 | put_io_context(ioc); |
86db1e29 JA |
153 | } |
154 | ||
7e5a8794 TH |
155 | /** |
156 | * ioc_clear_queue - break any ioc association with the specified queue | |
157 | * @q: request_queue being cleared | |
158 | * | |
159 | * Walk @q->icq_list and exit all io_cq's. Must be called with @q locked. | |
160 | */ | |
161 | void ioc_clear_queue(struct request_queue *q) | |
162 | { | |
163 | lockdep_assert_held(q->queue_lock); | |
164 | ||
165 | while (!list_empty(&q->icq_list)) { | |
166 | struct io_cq *icq = list_entry(q->icq_list.next, | |
167 | struct io_cq, q_node); | |
168 | struct io_context *ioc = icq->ioc; | |
169 | ||
170 | spin_lock(&ioc->lock); | |
171 | ioc_exit_icq(icq); | |
172 | spin_unlock(&ioc->lock); | |
173 | } | |
174 | } | |
175 | ||
f2dbd76a TH |
176 | void create_io_context_slowpath(struct task_struct *task, gfp_t gfp_flags, |
177 | int node) | |
86db1e29 | 178 | { |
df415656 | 179 | struct io_context *ioc; |
86db1e29 | 180 | |
42ec57a8 TH |
181 | ioc = kmem_cache_alloc_node(iocontext_cachep, gfp_flags | __GFP_ZERO, |
182 | node); | |
183 | if (unlikely(!ioc)) | |
f2dbd76a | 184 | return; |
42ec57a8 TH |
185 | |
186 | /* initialize */ | |
187 | atomic_long_set(&ioc->refcount, 1); | |
188 | atomic_set(&ioc->nr_tasks, 1); | |
189 | spin_lock_init(&ioc->lock); | |
c5869807 TH |
190 | INIT_RADIX_TREE(&ioc->icq_tree, GFP_ATOMIC | __GFP_HIGH); |
191 | INIT_HLIST_HEAD(&ioc->icq_list); | |
b2efa052 | 192 | INIT_WORK(&ioc->release_work, ioc_release_fn); |
86db1e29 | 193 | |
fd638368 TH |
194 | /* |
195 | * Try to install. ioc shouldn't be installed if someone else | |
196 | * already did or @task, which isn't %current, is exiting. Note | |
197 | * that we need to allow ioc creation on exiting %current as exit | |
198 | * path may issue IOs from e.g. exit_files(). The exit path is | |
199 | * responsible for not issuing IO after exit_io_context(). | |
200 | */ | |
6e736be7 | 201 | task_lock(task); |
fd638368 TH |
202 | if (!task->io_context && |
203 | (task == current || !(task->flags & PF_EXITING))) | |
6e736be7 | 204 | task->io_context = ioc; |
f2dbd76a | 205 | else |
6e736be7 | 206 | kmem_cache_free(iocontext_cachep, ioc); |
6e736be7 | 207 | task_unlock(task); |
86db1e29 | 208 | } |
86db1e29 | 209 | |
6e736be7 TH |
210 | /** |
211 | * get_task_io_context - get io_context of a task | |
212 | * @task: task of interest | |
213 | * @gfp_flags: allocation flags, used if allocation is necessary | |
214 | * @node: allocation node, used if allocation is necessary | |
215 | * | |
216 | * Return io_context of @task. If it doesn't exist, it is created with | |
217 | * @gfp_flags and @node. The returned io_context has its reference count | |
218 | * incremented. | |
86db1e29 | 219 | * |
6e736be7 | 220 | * This function always goes through task_lock() and it's better to use |
f2dbd76a | 221 | * %current->io_context + get_io_context() for %current. |
86db1e29 | 222 | */ |
6e736be7 TH |
223 | struct io_context *get_task_io_context(struct task_struct *task, |
224 | gfp_t gfp_flags, int node) | |
86db1e29 | 225 | { |
6e736be7 | 226 | struct io_context *ioc; |
86db1e29 | 227 | |
6e736be7 TH |
228 | might_sleep_if(gfp_flags & __GFP_WAIT); |
229 | ||
f2dbd76a TH |
230 | do { |
231 | task_lock(task); | |
232 | ioc = task->io_context; | |
233 | if (likely(ioc)) { | |
234 | get_io_context(ioc); | |
235 | task_unlock(task); | |
236 | return ioc; | |
237 | } | |
6e736be7 | 238 | task_unlock(task); |
f2dbd76a | 239 | } while (create_io_context(task, gfp_flags, node)); |
6e736be7 | 240 | |
f2dbd76a | 241 | return NULL; |
86db1e29 | 242 | } |
6e736be7 | 243 | EXPORT_SYMBOL(get_task_io_context); |
86db1e29 | 244 | |
47fdd4ca TH |
245 | /** |
246 | * ioc_lookup_icq - lookup io_cq from ioc | |
247 | * @ioc: the associated io_context | |
248 | * @q: the associated request_queue | |
249 | * | |
250 | * Look up io_cq associated with @ioc - @q pair from @ioc. Must be called | |
251 | * with @q->queue_lock held. | |
252 | */ | |
253 | struct io_cq *ioc_lookup_icq(struct io_context *ioc, struct request_queue *q) | |
254 | { | |
255 | struct io_cq *icq; | |
256 | ||
257 | lockdep_assert_held(q->queue_lock); | |
258 | ||
259 | /* | |
260 | * icq's are indexed from @ioc using radix tree and hint pointer, | |
261 | * both of which are protected with RCU. All removals are done | |
262 | * holding both q and ioc locks, and we're holding q lock - if we | |
263 | * find a icq which points to us, it's guaranteed to be valid. | |
264 | */ | |
265 | rcu_read_lock(); | |
266 | icq = rcu_dereference(ioc->icq_hint); | |
267 | if (icq && icq->q == q) | |
268 | goto out; | |
269 | ||
270 | icq = radix_tree_lookup(&ioc->icq_tree, q->id); | |
271 | if (icq && icq->q == q) | |
272 | rcu_assign_pointer(ioc->icq_hint, icq); /* allowed to race */ | |
273 | else | |
274 | icq = NULL; | |
275 | out: | |
276 | rcu_read_unlock(); | |
277 | return icq; | |
278 | } | |
279 | EXPORT_SYMBOL(ioc_lookup_icq); | |
280 | ||
f1f8cc94 TH |
281 | /** |
282 | * ioc_create_icq - create and link io_cq | |
283 | * @q: request_queue of interest | |
284 | * @gfp_mask: allocation mask | |
285 | * | |
286 | * Make sure io_cq linking %current->io_context and @q exists. If either | |
287 | * io_context and/or icq don't exist, they will be created using @gfp_mask. | |
288 | * | |
289 | * The caller is responsible for ensuring @ioc won't go away and @q is | |
290 | * alive and will stay alive until this function returns. | |
291 | */ | |
292 | struct io_cq *ioc_create_icq(struct request_queue *q, gfp_t gfp_mask) | |
293 | { | |
294 | struct elevator_type *et = q->elevator->type; | |
295 | struct io_context *ioc; | |
296 | struct io_cq *icq; | |
297 | ||
298 | /* allocate stuff */ | |
299 | ioc = create_io_context(current, gfp_mask, q->node); | |
300 | if (!ioc) | |
301 | return NULL; | |
302 | ||
303 | icq = kmem_cache_alloc_node(et->icq_cache, gfp_mask | __GFP_ZERO, | |
304 | q->node); | |
305 | if (!icq) | |
306 | return NULL; | |
307 | ||
308 | if (radix_tree_preload(gfp_mask) < 0) { | |
309 | kmem_cache_free(et->icq_cache, icq); | |
310 | return NULL; | |
311 | } | |
312 | ||
313 | icq->ioc = ioc; | |
314 | icq->q = q; | |
315 | INIT_LIST_HEAD(&icq->q_node); | |
316 | INIT_HLIST_NODE(&icq->ioc_node); | |
317 | ||
318 | /* lock both q and ioc and try to link @icq */ | |
319 | spin_lock_irq(q->queue_lock); | |
320 | spin_lock(&ioc->lock); | |
321 | ||
322 | if (likely(!radix_tree_insert(&ioc->icq_tree, q->id, icq))) { | |
323 | hlist_add_head(&icq->ioc_node, &ioc->icq_list); | |
324 | list_add(&icq->q_node, &q->icq_list); | |
325 | if (et->ops.elevator_init_icq_fn) | |
326 | et->ops.elevator_init_icq_fn(icq); | |
327 | } else { | |
328 | kmem_cache_free(et->icq_cache, icq); | |
329 | icq = ioc_lookup_icq(ioc, q); | |
330 | if (!icq) | |
331 | printk(KERN_ERR "cfq: icq link failed!\n"); | |
332 | } | |
333 | ||
334 | spin_unlock(&ioc->lock); | |
335 | spin_unlock_irq(q->queue_lock); | |
336 | radix_tree_preload_end(); | |
337 | return icq; | |
338 | } | |
339 | ||
d705ae6b | 340 | void ioc_set_icq_flags(struct io_context *ioc, unsigned int flags) |
dc86900e | 341 | { |
c5869807 | 342 | struct io_cq *icq; |
dc86900e TH |
343 | struct hlist_node *n; |
344 | ||
c5869807 | 345 | hlist_for_each_entry(icq, n, &ioc->icq_list, ioc_node) |
d705ae6b | 346 | icq->flags |= flags; |
dc86900e TH |
347 | } |
348 | ||
349 | /** | |
350 | * ioc_ioprio_changed - notify ioprio change | |
351 | * @ioc: io_context of interest | |
352 | * @ioprio: new ioprio | |
353 | * | |
c5869807 TH |
354 | * @ioc's ioprio has changed to @ioprio. Set %ICQ_IOPRIO_CHANGED for all |
355 | * icq's. iosched is responsible for checking the bit and applying it on | |
dc86900e TH |
356 | * request issue path. |
357 | */ | |
358 | void ioc_ioprio_changed(struct io_context *ioc, int ioprio) | |
359 | { | |
360 | unsigned long flags; | |
361 | ||
362 | spin_lock_irqsave(&ioc->lock, flags); | |
363 | ioc->ioprio = ioprio; | |
d705ae6b | 364 | ioc_set_icq_flags(ioc, ICQ_IOPRIO_CHANGED); |
dc86900e TH |
365 | spin_unlock_irqrestore(&ioc->lock, flags); |
366 | } | |
367 | ||
368 | /** | |
369 | * ioc_cgroup_changed - notify cgroup change | |
370 | * @ioc: io_context of interest | |
371 | * | |
c5869807 | 372 | * @ioc's cgroup has changed. Set %ICQ_CGROUP_CHANGED for all icq's. |
dc86900e TH |
373 | * iosched is responsible for checking the bit and applying it on request |
374 | * issue path. | |
375 | */ | |
376 | void ioc_cgroup_changed(struct io_context *ioc) | |
377 | { | |
378 | unsigned long flags; | |
379 | ||
380 | spin_lock_irqsave(&ioc->lock, flags); | |
d705ae6b | 381 | ioc_set_icq_flags(ioc, ICQ_CGROUP_CHANGED); |
dc86900e TH |
382 | spin_unlock_irqrestore(&ioc->lock, flags); |
383 | } | |
64c42998 | 384 | EXPORT_SYMBOL(ioc_cgroup_changed); |
dc86900e | 385 | |
d705ae6b TH |
386 | /** |
387 | * icq_get_changed - fetch and clear icq changed mask | |
388 | * @icq: icq of interest | |
389 | * | |
390 | * Fetch and clear ICQ_*_CHANGED bits from @icq. Grabs and releases | |
391 | * @icq->ioc->lock. | |
392 | */ | |
393 | unsigned icq_get_changed(struct io_cq *icq) | |
394 | { | |
395 | unsigned int changed = 0; | |
396 | unsigned long flags; | |
397 | ||
398 | if (unlikely(icq->flags & ICQ_CHANGED_MASK)) { | |
399 | spin_lock_irqsave(&icq->ioc->lock, flags); | |
400 | changed = icq->flags & ICQ_CHANGED_MASK; | |
401 | icq->flags &= ~ICQ_CHANGED_MASK; | |
402 | spin_unlock_irqrestore(&icq->ioc->lock, flags); | |
403 | } | |
404 | return changed; | |
405 | } | |
406 | EXPORT_SYMBOL(icq_get_changed); | |
407 | ||
13341598 | 408 | static int __init blk_ioc_init(void) |
86db1e29 JA |
409 | { |
410 | iocontext_cachep = kmem_cache_create("blkdev_ioc", | |
411 | sizeof(struct io_context), 0, SLAB_PANIC, NULL); | |
412 | return 0; | |
413 | } | |
414 | subsys_initcall(blk_ioc_init); |