xref: /openbmc/qemu/block/throttle-groups.c (revision 121d0712)
1 /*
2  * QEMU block throttling group infrastructure
3  *
4  * Copyright (C) Nodalink, EURL. 2014
5  * Copyright (C) Igalia, S.L. 2015
6  *
7  * Authors:
8  *   Benoît Canet <benoit.canet@nodalink.com>
9  *   Alberto Garcia <berto@igalia.com>
10  *
11  * This program is free software; you can redistribute it and/or
12  * modify it under the terms of the GNU General Public License as
13  * published by the Free Software Foundation; either version 2 or
14  * (at your option) version 3 of the License.
15  *
16  * This program is distributed in the hope that it will be useful,
17  * but WITHOUT ANY WARRANTY; without even the implied warranty of
18  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
19  * GNU General Public License for more details.
20  *
21  * You should have received a copy of the GNU General Public License
22  * along with this program; if not, see <http://www.gnu.org/licenses/>.
23  */
24 
25 #include "qemu/osdep.h"
26 #include "sysemu/block-backend.h"
27 #include "block/throttle-groups.h"
28 #include "qemu/queue.h"
29 #include "qemu/thread.h"
30 #include "sysemu/qtest.h"
31 
32 /* The ThrottleGroup structure (with its ThrottleState) is shared
33  * among different BlockBackends and it's independent from
34  * AioContext, so in order to use it from different threads it needs
35  * its own locking.
36  *
37  * This locking is however handled internally in this file, so it's
38  * transparent to outside users.
39  *
40  * The whole ThrottleGroup structure is private and invisible to
41  * outside users, that only use it through its ThrottleState.
42  *
43  * In addition to the ThrottleGroup structure, BlockBackendPublic has
44  * fields that need to be accessed by other members of the group and
45  * therefore also need to be protected by this lock. Once a
46  * BlockBackend is registered in a group those fields can be accessed
47  * by other threads any time.
48  *
49  * Again, all this is handled internally and is mostly transparent to
50  * the outside. The 'throttle_timers' field however has an additional
51  * constraint because it may be temporarily invalid (see for example
52  * bdrv_set_aio_context()). Therefore in this file a thread will
53  * access some other BlockBackend's timers only after verifying that
54  * that BlockBackend has throttled requests in the queue.
55  */
56 typedef struct ThrottleGroup {
57     char *name; /* This is constant during the lifetime of the group */
58 
59     QemuMutex lock; /* This lock protects the following four fields */
60     ThrottleState ts;
61     QLIST_HEAD(, BlockBackendPublic) head;
62     BlockBackend *tokens[2];
63     bool any_timer_armed[2];
64 
65     /* These two are protected by the global throttle_groups_lock */
66     unsigned refcount;
67     QTAILQ_ENTRY(ThrottleGroup) list;
68 } ThrottleGroup;
69 
70 static QemuMutex throttle_groups_lock;
71 static QTAILQ_HEAD(, ThrottleGroup) throttle_groups =
72     QTAILQ_HEAD_INITIALIZER(throttle_groups);
73 
74 /* Increments the reference count of a ThrottleGroup given its name.
75  *
76  * If no ThrottleGroup is found with the given name a new one is
77  * created.
78  *
79  * @name: the name of the ThrottleGroup
80  * @ret:  the ThrottleState member of the ThrottleGroup
81  */
82 ThrottleState *throttle_group_incref(const char *name)
83 {
84     ThrottleGroup *tg = NULL;
85     ThrottleGroup *iter;
86 
87     qemu_mutex_lock(&throttle_groups_lock);
88 
89     /* Look for an existing group with that name */
90     QTAILQ_FOREACH(iter, &throttle_groups, list) {
91         if (!strcmp(name, iter->name)) {
92             tg = iter;
93             break;
94         }
95     }
96 
97     /* Create a new one if not found */
98     if (!tg) {
99         tg = g_new0(ThrottleGroup, 1);
100         tg->name = g_strdup(name);
101         qemu_mutex_init(&tg->lock);
102         throttle_init(&tg->ts);
103         QLIST_INIT(&tg->head);
104 
105         QTAILQ_INSERT_TAIL(&throttle_groups, tg, list);
106     }
107 
108     tg->refcount++;
109 
110     qemu_mutex_unlock(&throttle_groups_lock);
111 
112     return &tg->ts;
113 }
114 
115 /* Decrease the reference count of a ThrottleGroup.
116  *
117  * When the reference count reaches zero the ThrottleGroup is
118  * destroyed.
119  *
120  * @ts:  The ThrottleGroup to unref, given by its ThrottleState member
121  */
122 void throttle_group_unref(ThrottleState *ts)
123 {
124     ThrottleGroup *tg = container_of(ts, ThrottleGroup, ts);
125 
126     qemu_mutex_lock(&throttle_groups_lock);
127     if (--tg->refcount == 0) {
128         QTAILQ_REMOVE(&throttle_groups, tg, list);
129         qemu_mutex_destroy(&tg->lock);
130         g_free(tg->name);
131         g_free(tg);
132     }
133     qemu_mutex_unlock(&throttle_groups_lock);
134 }
135 
136 /* Get the name from a BlockBackend's ThrottleGroup. The name (and the pointer)
137  * is guaranteed to remain constant during the lifetime of the group.
138  *
139  * @blk:  a BlockBackend that is member of a throttling group
140  * @ret:  the name of the group.
141  */
142 const char *throttle_group_get_name(BlockBackend *blk)
143 {
144     BlockBackendPublic *blkp = blk_get_public(blk);
145     ThrottleGroup *tg = container_of(blkp->throttle_state, ThrottleGroup, ts);
146     return tg->name;
147 }
148 
149 /* Return the next BlockBackend in the round-robin sequence, simulating a
150  * circular list.
151  *
152  * This assumes that tg->lock is held.
153  *
154  * @blk: the current BlockBackend
155  * @ret: the next BlockBackend in the sequence
156  */
157 static BlockBackend *throttle_group_next_blk(BlockBackend *blk)
158 {
159     BlockBackendPublic *blkp = blk_get_public(blk);
160     ThrottleState *ts = blkp->throttle_state;
161     ThrottleGroup *tg = container_of(ts, ThrottleGroup, ts);
162     BlockBackendPublic *next = QLIST_NEXT(blkp, round_robin);
163 
164     if (!next) {
165         next = QLIST_FIRST(&tg->head);
166     }
167 
168     return blk_by_public(next);
169 }
170 
171 /* Return the next BlockBackend in the round-robin sequence with pending I/O
172  * requests.
173  *
174  * This assumes that tg->lock is held.
175  *
176  * @blk:       the current BlockBackend
177  * @is_write:  the type of operation (read/write)
178  * @ret:       the next BlockBackend with pending requests, or blk if there is
179  *             none.
180  */
181 static BlockBackend *next_throttle_token(BlockBackend *blk, bool is_write)
182 {
183     BlockBackendPublic *blkp = blk_get_public(blk);
184     ThrottleGroup *tg = container_of(blkp->throttle_state, ThrottleGroup, ts);
185     BlockBackend *token, *start;
186 
187     start = token = tg->tokens[is_write];
188 
189     /* get next bs round in round robin style */
190     token = throttle_group_next_blk(token);
191     while (token != start && !blkp->pending_reqs[is_write]) {
192         token = throttle_group_next_blk(token);
193     }
194 
195     /* If no IO are queued for scheduling on the next round robin token
196      * then decide the token is the current bs because chances are
197      * the current bs get the current request queued.
198      */
199     if (token == start && !blkp->pending_reqs[is_write]) {
200         token = blk;
201     }
202 
203     return token;
204 }
205 
206 /* Check if the next I/O request for a BlockBackend needs to be throttled or
207  * not. If there's no timer set in this group, set one and update the token
208  * accordingly.
209  *
210  * This assumes that tg->lock is held.
211  *
212  * @blk:        the current BlockBackend
213  * @is_write:   the type of operation (read/write)
214  * @ret:        whether the I/O request needs to be throttled or not
215  */
216 static bool throttle_group_schedule_timer(BlockBackend *blk, bool is_write)
217 {
218     BlockBackendPublic *blkp = blk_get_public(blk);
219     ThrottleState *ts = blkp->throttle_state;
220     ThrottleTimers *tt = &blkp->throttle_timers;
221     ThrottleGroup *tg = container_of(ts, ThrottleGroup, ts);
222     bool must_wait;
223 
224     if (blkp->io_limits_disabled) {
225         return false;
226     }
227 
228     /* Check if any of the timers in this group is already armed */
229     if (tg->any_timer_armed[is_write]) {
230         return true;
231     }
232 
233     must_wait = throttle_schedule_timer(ts, tt, is_write);
234 
235     /* If a timer just got armed, set blk as the current token */
236     if (must_wait) {
237         tg->tokens[is_write] = blk;
238         tg->any_timer_armed[is_write] = true;
239     }
240 
241     return must_wait;
242 }
243 
244 /* Look for the next pending I/O request and schedule it.
245  *
246  * This assumes that tg->lock is held.
247  *
248  * @blk:       the current BlockBackend
249  * @is_write:  the type of operation (read/write)
250  */
251 static void schedule_next_request(BlockBackend *blk, bool is_write)
252 {
253     BlockBackendPublic *blkp = blk_get_public(blk);
254     ThrottleGroup *tg = container_of(blkp->throttle_state, ThrottleGroup, ts);
255     bool must_wait;
256     BlockBackend *token;
257 
258     /* Check if there's any pending request to schedule next */
259     token = next_throttle_token(blk, is_write);
260     if (!blkp->pending_reqs[is_write]) {
261         return;
262     }
263 
264     /* Set a timer for the request if it needs to be throttled */
265     must_wait = throttle_group_schedule_timer(token, is_write);
266 
267     /* If it doesn't have to wait, queue it for immediate execution */
268     if (!must_wait) {
269         /* Give preference to requests from the current blk */
270         if (qemu_in_coroutine() &&
271             qemu_co_queue_next(&blkp->throttled_reqs[is_write])) {
272             token = blk;
273         } else {
274             ThrottleTimers *tt = &blkp->throttle_timers;
275             int64_t now = qemu_clock_get_ns(tt->clock_type);
276             timer_mod(tt->timers[is_write], now + 1);
277             tg->any_timer_armed[is_write] = true;
278         }
279         tg->tokens[is_write] = token;
280     }
281 }
282 
283 /* Check if an I/O request needs to be throttled, wait and set a timer
284  * if necessary, and schedule the next request using a round robin
285  * algorithm.
286  *
287  * @blk:       the current BlockBackend
288  * @bytes:     the number of bytes for this I/O
289  * @is_write:  the type of operation (read/write)
290  */
291 void coroutine_fn throttle_group_co_io_limits_intercept(BlockBackend *blk,
292                                                         unsigned int bytes,
293                                                         bool is_write)
294 {
295     bool must_wait;
296     BlockBackend *token;
297 
298     BlockBackendPublic *blkp = blk_get_public(blk);
299     ThrottleGroup *tg = container_of(blkp->throttle_state, ThrottleGroup, ts);
300     qemu_mutex_lock(&tg->lock);
301 
302     /* First we check if this I/O has to be throttled. */
303     token = next_throttle_token(blk, is_write);
304     must_wait = throttle_group_schedule_timer(token, is_write);
305 
306     /* Wait if there's a timer set or queued requests of this type */
307     if (must_wait || blkp->pending_reqs[is_write]) {
308         blkp->pending_reqs[is_write]++;
309         qemu_mutex_unlock(&tg->lock);
310         qemu_co_queue_wait(&blkp->throttled_reqs[is_write]);
311         qemu_mutex_lock(&tg->lock);
312         blkp->pending_reqs[is_write]--;
313     }
314 
315     /* The I/O will be executed, so do the accounting */
316     throttle_account(blkp->throttle_state, is_write, bytes);
317 
318     /* Schedule the next request */
319     schedule_next_request(blk, is_write);
320 
321     qemu_mutex_unlock(&tg->lock);
322 }
323 
324 void throttle_group_restart_blk(BlockBackend *blk)
325 {
326     BlockBackendPublic *blkp = blk_get_public(blk);
327     int i;
328 
329     for (i = 0; i < 2; i++) {
330         while (qemu_co_enter_next(&blkp->throttled_reqs[i])) {
331             ;
332         }
333     }
334 }
335 
336 /* Update the throttle configuration for a particular group. Similar
337  * to throttle_config(), but guarantees atomicity within the
338  * throttling group.
339  *
340  * @blk: a BlockBackend that is a member of the group
341  * @cfg: the configuration to set
342  */
343 void throttle_group_config(BlockBackend *blk, ThrottleConfig *cfg)
344 {
345     BlockBackendPublic *blkp = blk_get_public(blk);
346     ThrottleTimers *tt = &blkp->throttle_timers;
347     ThrottleState *ts = blkp->throttle_state;
348     ThrottleGroup *tg = container_of(ts, ThrottleGroup, ts);
349     qemu_mutex_lock(&tg->lock);
350     /* throttle_config() cancels the timers */
351     if (timer_pending(tt->timers[0])) {
352         tg->any_timer_armed[0] = false;
353     }
354     if (timer_pending(tt->timers[1])) {
355         tg->any_timer_armed[1] = false;
356     }
357     throttle_config(ts, tt, cfg);
358     qemu_mutex_unlock(&tg->lock);
359 
360     qemu_co_enter_next(&blkp->throttled_reqs[0]);
361     qemu_co_enter_next(&blkp->throttled_reqs[1]);
362 }
363 
364 /* Get the throttle configuration from a particular group. Similar to
365  * throttle_get_config(), but guarantees atomicity within the
366  * throttling group.
367  *
368  * @blk: a BlockBackend that is a member of the group
369  * @cfg: the configuration will be written here
370  */
371 void throttle_group_get_config(BlockBackend *blk, ThrottleConfig *cfg)
372 {
373     BlockBackendPublic *blkp = blk_get_public(blk);
374     ThrottleState *ts = blkp->throttle_state;
375     ThrottleGroup *tg = container_of(ts, ThrottleGroup, ts);
376     qemu_mutex_lock(&tg->lock);
377     throttle_get_config(ts, cfg);
378     qemu_mutex_unlock(&tg->lock);
379 }
380 
381 /* ThrottleTimers callback. This wakes up a request that was waiting
382  * because it had been throttled.
383  *
384  * @blk:       the BlockBackend whose request had been throttled
385  * @is_write:  the type of operation (read/write)
386  */
387 static void timer_cb(BlockBackend *blk, bool is_write)
388 {
389     BlockBackendPublic *blkp = blk_get_public(blk);
390     ThrottleState *ts = blkp->throttle_state;
391     ThrottleGroup *tg = container_of(ts, ThrottleGroup, ts);
392     bool empty_queue;
393 
394     /* The timer has just been fired, so we can update the flag */
395     qemu_mutex_lock(&tg->lock);
396     tg->any_timer_armed[is_write] = false;
397     qemu_mutex_unlock(&tg->lock);
398 
399     /* Run the request that was waiting for this timer */
400     empty_queue = !qemu_co_enter_next(&blkp->throttled_reqs[is_write]);
401 
402     /* If the request queue was empty then we have to take care of
403      * scheduling the next one */
404     if (empty_queue) {
405         qemu_mutex_lock(&tg->lock);
406         schedule_next_request(blk, is_write);
407         qemu_mutex_unlock(&tg->lock);
408     }
409 }
410 
411 static void read_timer_cb(void *opaque)
412 {
413     timer_cb(opaque, false);
414 }
415 
416 static void write_timer_cb(void *opaque)
417 {
418     timer_cb(opaque, true);
419 }
420 
421 /* Register a BlockBackend in the throttling group, also initializing its
422  * timers and updating its throttle_state pointer to point to it. If a
423  * throttling group with that name does not exist yet, it will be created.
424  *
425  * @blk:       the BlockBackend to insert
426  * @groupname: the name of the group
427  */
428 void throttle_group_register_blk(BlockBackend *blk, const char *groupname)
429 {
430     int i;
431     BlockBackendPublic *blkp = blk_get_public(blk);
432     ThrottleState *ts = throttle_group_incref(groupname);
433     ThrottleGroup *tg = container_of(ts, ThrottleGroup, ts);
434     int clock_type = QEMU_CLOCK_REALTIME;
435 
436     if (qtest_enabled()) {
437         /* For testing block IO throttling only */
438         clock_type = QEMU_CLOCK_VIRTUAL;
439     }
440 
441     blkp->throttle_state = ts;
442 
443     qemu_mutex_lock(&tg->lock);
444     /* If the ThrottleGroup is new set this BlockBackend as the token */
445     for (i = 0; i < 2; i++) {
446         if (!tg->tokens[i]) {
447             tg->tokens[i] = blk;
448         }
449     }
450 
451     QLIST_INSERT_HEAD(&tg->head, blkp, round_robin);
452 
453     throttle_timers_init(&blkp->throttle_timers,
454                          blk_get_aio_context(blk),
455                          clock_type,
456                          read_timer_cb,
457                          write_timer_cb,
458                          blk);
459 
460     qemu_mutex_unlock(&tg->lock);
461 }
462 
463 /* Unregister a BlockBackend from its group, removing it from the list,
464  * destroying the timers and setting the throttle_state pointer to NULL.
465  *
466  * The BlockBackend must not have pending throttled requests, so the caller has
467  * to drain them first.
468  *
469  * The group will be destroyed if it's empty after this operation.
470  *
471  * @blk: the BlockBackend to remove
472  */
473 void throttle_group_unregister_blk(BlockBackend *blk)
474 {
475     BlockBackendPublic *blkp = blk_get_public(blk);
476     ThrottleGroup *tg = container_of(blkp->throttle_state, ThrottleGroup, ts);
477     int i;
478 
479     assert(blkp->pending_reqs[0] == 0 && blkp->pending_reqs[1] == 0);
480     assert(qemu_co_queue_empty(&blkp->throttled_reqs[0]));
481     assert(qemu_co_queue_empty(&blkp->throttled_reqs[1]));
482 
483     qemu_mutex_lock(&tg->lock);
484     for (i = 0; i < 2; i++) {
485         if (tg->tokens[i] == blk) {
486             BlockBackend *token = throttle_group_next_blk(blk);
487             /* Take care of the case where this is the last blk in the group */
488             if (token == blk) {
489                 token = NULL;
490             }
491             tg->tokens[i] = token;
492         }
493     }
494 
495     /* remove the current blk from the list */
496     QLIST_REMOVE(blkp, round_robin);
497     throttle_timers_destroy(&blkp->throttle_timers);
498     qemu_mutex_unlock(&tg->lock);
499 
500     throttle_group_unref(&tg->ts);
501     blkp->throttle_state = NULL;
502 }
503 
504 static void throttle_groups_init(void)
505 {
506     qemu_mutex_init(&throttle_groups_lock);
507 }
508 
509 block_init(throttle_groups_init);
510