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 | * BenoƮt 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/throttle-options.h" |
29 | #include "qemu/main-loop.h" |
30 | #include "qemu/queue.h" |
31 | #include "qemu/thread.h" |
32 | #include "sysemu/qtest.h" |
33 | #include "qapi/error.h" |
34 | #include "qapi/qapi-visit-block-core.h" |
35 | #include "qom/object.h" |
36 | #include "qom/object_interfaces.h" |
37 | |
38 | static void throttle_group_obj_init(Object *obj); |
39 | static void throttle_group_obj_complete(UserCreatable *obj, Error **errp); |
40 | static void timer_cb(ThrottleGroupMember *tgm, bool is_write); |
41 | |
42 | /* The ThrottleGroup structure (with its ThrottleState) is shared |
43 | * among different ThrottleGroupMembers and it's independent from |
44 | * AioContext, so in order to use it from different threads it needs |
45 | * its own locking. |
46 | * |
47 | * This locking is however handled internally in this file, so it's |
48 | * transparent to outside users. |
49 | * |
50 | * The whole ThrottleGroup structure is private and invisible to |
51 | * outside users, that only use it through its ThrottleState. |
52 | * |
53 | * In addition to the ThrottleGroup structure, ThrottleGroupMember has |
54 | * fields that need to be accessed by other members of the group and |
55 | * therefore also need to be protected by this lock. Once a |
56 | * ThrottleGroupMember is registered in a group those fields can be accessed |
57 | * by other threads any time. |
58 | * |
59 | * Again, all this is handled internally and is mostly transparent to |
60 | * the outside. The 'throttle_timers' field however has an additional |
61 | * constraint because it may be temporarily invalid (see for example |
62 | * blk_set_aio_context()). Therefore in this file a thread will |
63 | * access some other ThrottleGroupMember's timers only after verifying that |
64 | * that ThrottleGroupMember has throttled requests in the queue. |
65 | */ |
66 | typedef struct ThrottleGroup { |
67 | Object parent_obj; |
68 | |
69 | /* refuse individual property change if initialization is complete */ |
70 | bool is_initialized; |
71 | char *name; /* This is constant during the lifetime of the group */ |
72 | |
73 | QemuMutex lock; /* This lock protects the following four fields */ |
74 | ThrottleState ts; |
75 | QLIST_HEAD(, ThrottleGroupMember) head; |
76 | ThrottleGroupMember *tokens[2]; |
77 | bool any_timer_armed[2]; |
78 | QEMUClockType clock_type; |
79 | |
80 | /* This field is protected by the global QEMU mutex */ |
81 | QTAILQ_ENTRY(ThrottleGroup) list; |
82 | } ThrottleGroup; |
83 | |
84 | /* This is protected by the global QEMU mutex */ |
85 | static QTAILQ_HEAD(, ThrottleGroup) throttle_groups = |
86 | QTAILQ_HEAD_INITIALIZER(throttle_groups); |
87 | |
88 | |
89 | /* This function reads throttle_groups and must be called under the global |
90 | * mutex. |
91 | */ |
92 | static ThrottleGroup *throttle_group_by_name(const char *name) |
93 | { |
94 | ThrottleGroup *iter; |
95 | |
96 | /* Look for an existing group with that name */ |
97 | QTAILQ_FOREACH(iter, &throttle_groups, list) { |
98 | if (!g_strcmp0(name, iter->name)) { |
99 | return iter; |
100 | } |
101 | } |
102 | |
103 | return NULL; |
104 | } |
105 | |
106 | /* This function reads throttle_groups and must be called under the global |
107 | * mutex. |
108 | */ |
109 | bool throttle_group_exists(const char *name) |
110 | { |
111 | return throttle_group_by_name(name) != NULL; |
112 | } |
113 | |
114 | /* Increments the reference count of a ThrottleGroup given its name. |
115 | * |
116 | * If no ThrottleGroup is found with the given name a new one is |
117 | * created. |
118 | * |
119 | * This function edits throttle_groups and must be called under the global |
120 | * mutex. |
121 | * |
122 | * @name: the name of the ThrottleGroup |
123 | * @ret: the ThrottleState member of the ThrottleGroup |
124 | */ |
125 | ThrottleState *throttle_group_incref(const char *name) |
126 | { |
127 | ThrottleGroup *tg = NULL; |
128 | |
129 | /* Look for an existing group with that name */ |
130 | tg = throttle_group_by_name(name); |
131 | |
132 | if (tg) { |
133 | object_ref(OBJECT(tg)); |
134 | } else { |
135 | /* Create a new one if not found */ |
136 | /* new ThrottleGroup obj will have a refcnt = 1 */ |
137 | tg = THROTTLE_GROUP(object_new(TYPE_THROTTLE_GROUP)); |
138 | tg->name = g_strdup(name); |
139 | throttle_group_obj_complete(USER_CREATABLE(tg), &error_abort); |
140 | } |
141 | |
142 | return &tg->ts; |
143 | } |
144 | |
145 | /* Decrease the reference count of a ThrottleGroup. |
146 | * |
147 | * When the reference count reaches zero the ThrottleGroup is |
148 | * destroyed. |
149 | * |
150 | * This function edits throttle_groups and must be called under the global |
151 | * mutex. |
152 | * |
153 | * @ts: The ThrottleGroup to unref, given by its ThrottleState member |
154 | */ |
155 | void throttle_group_unref(ThrottleState *ts) |
156 | { |
157 | ThrottleGroup *tg = container_of(ts, ThrottleGroup, ts); |
158 | object_unref(OBJECT(tg)); |
159 | } |
160 | |
161 | /* Get the name from a ThrottleGroupMember's group. The name (and the pointer) |
162 | * is guaranteed to remain constant during the lifetime of the group. |
163 | * |
164 | * @tgm: a ThrottleGroupMember |
165 | * @ret: the name of the group. |
166 | */ |
167 | const char *throttle_group_get_name(ThrottleGroupMember *tgm) |
168 | { |
169 | ThrottleGroup *tg = container_of(tgm->throttle_state, ThrottleGroup, ts); |
170 | return tg->name; |
171 | } |
172 | |
173 | /* Return the next ThrottleGroupMember in the round-robin sequence, simulating |
174 | * a circular list. |
175 | * |
176 | * This assumes that tg->lock is held. |
177 | * |
178 | * @tgm: the current ThrottleGroupMember |
179 | * @ret: the next ThrottleGroupMember in the sequence |
180 | */ |
181 | static ThrottleGroupMember *throttle_group_next_tgm(ThrottleGroupMember *tgm) |
182 | { |
183 | ThrottleState *ts = tgm->throttle_state; |
184 | ThrottleGroup *tg = container_of(ts, ThrottleGroup, ts); |
185 | ThrottleGroupMember *next = QLIST_NEXT(tgm, round_robin); |
186 | |
187 | if (!next) { |
188 | next = QLIST_FIRST(&tg->head); |
189 | } |
190 | |
191 | return next; |
192 | } |
193 | |
194 | /* |
195 | * Return whether a ThrottleGroupMember has pending requests. |
196 | * |
197 | * This assumes that tg->lock is held. |
198 | * |
199 | * @tgm: the ThrottleGroupMember |
200 | * @is_write: the type of operation (read/write) |
201 | * @ret: whether the ThrottleGroupMember has pending requests. |
202 | */ |
203 | static inline bool tgm_has_pending_reqs(ThrottleGroupMember *tgm, |
204 | bool is_write) |
205 | { |
206 | return tgm->pending_reqs[is_write]; |
207 | } |
208 | |
209 | /* Return the next ThrottleGroupMember in the round-robin sequence with pending |
210 | * I/O requests. |
211 | * |
212 | * This assumes that tg->lock is held. |
213 | * |
214 | * @tgm: the current ThrottleGroupMember |
215 | * @is_write: the type of operation (read/write) |
216 | * @ret: the next ThrottleGroupMember with pending requests, or tgm if |
217 | * there is none. |
218 | */ |
219 | static ThrottleGroupMember *next_throttle_token(ThrottleGroupMember *tgm, |
220 | bool is_write) |
221 | { |
222 | ThrottleState *ts = tgm->throttle_state; |
223 | ThrottleGroup *tg = container_of(ts, ThrottleGroup, ts); |
224 | ThrottleGroupMember *token, *start; |
225 | |
226 | /* If this member has its I/O limits disabled then it means that |
227 | * it's being drained. Skip the round-robin search and return tgm |
228 | * immediately if it has pending requests. Otherwise we could be |
229 | * forcing it to wait for other member's throttled requests. */ |
230 | if (tgm_has_pending_reqs(tgm, is_write) && |
231 | atomic_read(&tgm->io_limits_disabled)) { |
232 | return tgm; |
233 | } |
234 | |
235 | start = token = tg->tokens[is_write]; |
236 | |
237 | /* get next bs round in round robin style */ |
238 | token = throttle_group_next_tgm(token); |
239 | while (token != start && !tgm_has_pending_reqs(token, is_write)) { |
240 | token = throttle_group_next_tgm(token); |
241 | } |
242 | |
243 | /* If no IO are queued for scheduling on the next round robin token |
244 | * then decide the token is the current tgm because chances are |
245 | * the current tgm got the current request queued. |
246 | */ |
247 | if (token == start && !tgm_has_pending_reqs(token, is_write)) { |
248 | token = tgm; |
249 | } |
250 | |
251 | /* Either we return the original TGM, or one with pending requests */ |
252 | assert(token == tgm || tgm_has_pending_reqs(token, is_write)); |
253 | |
254 | return token; |
255 | } |
256 | |
257 | /* Check if the next I/O request for a ThrottleGroupMember needs to be |
258 | * throttled or not. If there's no timer set in this group, set one and update |
259 | * the token accordingly. |
260 | * |
261 | * This assumes that tg->lock is held. |
262 | * |
263 | * @tgm: the current ThrottleGroupMember |
264 | * @is_write: the type of operation (read/write) |
265 | * @ret: whether the I/O request needs to be throttled or not |
266 | */ |
267 | static bool throttle_group_schedule_timer(ThrottleGroupMember *tgm, |
268 | bool is_write) |
269 | { |
270 | ThrottleState *ts = tgm->throttle_state; |
271 | ThrottleGroup *tg = container_of(ts, ThrottleGroup, ts); |
272 | ThrottleTimers *tt = &tgm->throttle_timers; |
273 | bool must_wait; |
274 | |
275 | if (atomic_read(&tgm->io_limits_disabled)) { |
276 | return false; |
277 | } |
278 | |
279 | /* Check if any of the timers in this group is already armed */ |
280 | if (tg->any_timer_armed[is_write]) { |
281 | return true; |
282 | } |
283 | |
284 | must_wait = throttle_schedule_timer(ts, tt, is_write); |
285 | |
286 | /* If a timer just got armed, set tgm as the current token */ |
287 | if (must_wait) { |
288 | tg->tokens[is_write] = tgm; |
289 | tg->any_timer_armed[is_write] = true; |
290 | } |
291 | |
292 | return must_wait; |
293 | } |
294 | |
295 | /* Start the next pending I/O request for a ThrottleGroupMember. Return whether |
296 | * any request was actually pending. |
297 | * |
298 | * @tgm: the current ThrottleGroupMember |
299 | * @is_write: the type of operation (read/write) |
300 | */ |
301 | static bool coroutine_fn throttle_group_co_restart_queue(ThrottleGroupMember *tgm, |
302 | bool is_write) |
303 | { |
304 | bool ret; |
305 | |
306 | qemu_co_mutex_lock(&tgm->throttled_reqs_lock); |
307 | ret = qemu_co_queue_next(&tgm->throttled_reqs[is_write]); |
308 | qemu_co_mutex_unlock(&tgm->throttled_reqs_lock); |
309 | |
310 | return ret; |
311 | } |
312 | |
313 | /* Look for the next pending I/O request and schedule it. |
314 | * |
315 | * This assumes that tg->lock is held. |
316 | * |
317 | * @tgm: the current ThrottleGroupMember |
318 | * @is_write: the type of operation (read/write) |
319 | */ |
320 | static void schedule_next_request(ThrottleGroupMember *tgm, bool is_write) |
321 | { |
322 | ThrottleState *ts = tgm->throttle_state; |
323 | ThrottleGroup *tg = container_of(ts, ThrottleGroup, ts); |
324 | bool must_wait; |
325 | ThrottleGroupMember *token; |
326 | |
327 | /* Check if there's any pending request to schedule next */ |
328 | token = next_throttle_token(tgm, is_write); |
329 | if (!tgm_has_pending_reqs(token, is_write)) { |
330 | return; |
331 | } |
332 | |
333 | /* Set a timer for the request if it needs to be throttled */ |
334 | must_wait = throttle_group_schedule_timer(token, is_write); |
335 | |
336 | /* If it doesn't have to wait, queue it for immediate execution */ |
337 | if (!must_wait) { |
338 | /* Give preference to requests from the current tgm */ |
339 | if (qemu_in_coroutine() && |
340 | throttle_group_co_restart_queue(tgm, is_write)) { |
341 | token = tgm; |
342 | } else { |
343 | ThrottleTimers *tt = &token->throttle_timers; |
344 | int64_t now = qemu_clock_get_ns(tg->clock_type); |
345 | timer_mod(tt->timers[is_write], now); |
346 | tg->any_timer_armed[is_write] = true; |
347 | } |
348 | tg->tokens[is_write] = token; |
349 | } |
350 | } |
351 | |
352 | /* Check if an I/O request needs to be throttled, wait and set a timer |
353 | * if necessary, and schedule the next request using a round robin |
354 | * algorithm. |
355 | * |
356 | * @tgm: the current ThrottleGroupMember |
357 | * @bytes: the number of bytes for this I/O |
358 | * @is_write: the type of operation (read/write) |
359 | */ |
360 | void coroutine_fn throttle_group_co_io_limits_intercept(ThrottleGroupMember *tgm, |
361 | unsigned int bytes, |
362 | bool is_write) |
363 | { |
364 | bool must_wait; |
365 | ThrottleGroupMember *token; |
366 | ThrottleGroup *tg = container_of(tgm->throttle_state, ThrottleGroup, ts); |
367 | qemu_mutex_lock(&tg->lock); |
368 | |
369 | /* First we check if this I/O has to be throttled. */ |
370 | token = next_throttle_token(tgm, is_write); |
371 | must_wait = throttle_group_schedule_timer(token, is_write); |
372 | |
373 | /* Wait if there's a timer set or queued requests of this type */ |
374 | if (must_wait || tgm->pending_reqs[is_write]) { |
375 | tgm->pending_reqs[is_write]++; |
376 | qemu_mutex_unlock(&tg->lock); |
377 | qemu_co_mutex_lock(&tgm->throttled_reqs_lock); |
378 | qemu_co_queue_wait(&tgm->throttled_reqs[is_write], |
379 | &tgm->throttled_reqs_lock); |
380 | qemu_co_mutex_unlock(&tgm->throttled_reqs_lock); |
381 | qemu_mutex_lock(&tg->lock); |
382 | tgm->pending_reqs[is_write]--; |
383 | } |
384 | |
385 | /* The I/O will be executed, so do the accounting */ |
386 | throttle_account(tgm->throttle_state, is_write, bytes); |
387 | |
388 | /* Schedule the next request */ |
389 | schedule_next_request(tgm, is_write); |
390 | |
391 | qemu_mutex_unlock(&tg->lock); |
392 | } |
393 | |
394 | typedef struct { |
395 | ThrottleGroupMember *tgm; |
396 | bool is_write; |
397 | } RestartData; |
398 | |
399 | static void coroutine_fn throttle_group_restart_queue_entry(void *opaque) |
400 | { |
401 | RestartData *data = opaque; |
402 | ThrottleGroupMember *tgm = data->tgm; |
403 | ThrottleState *ts = tgm->throttle_state; |
404 | ThrottleGroup *tg = container_of(ts, ThrottleGroup, ts); |
405 | bool is_write = data->is_write; |
406 | bool empty_queue; |
407 | |
408 | empty_queue = !throttle_group_co_restart_queue(tgm, is_write); |
409 | |
410 | /* If the request queue was empty then we have to take care of |
411 | * scheduling the next one */ |
412 | if (empty_queue) { |
413 | qemu_mutex_lock(&tg->lock); |
414 | schedule_next_request(tgm, is_write); |
415 | qemu_mutex_unlock(&tg->lock); |
416 | } |
417 | |
418 | g_free(data); |
419 | |
420 | atomic_dec(&tgm->restart_pending); |
421 | aio_wait_kick(); |
422 | } |
423 | |
424 | static void throttle_group_restart_queue(ThrottleGroupMember *tgm, bool is_write) |
425 | { |
426 | Coroutine *co; |
427 | RestartData *rd = g_new0(RestartData, 1); |
428 | |
429 | rd->tgm = tgm; |
430 | rd->is_write = is_write; |
431 | |
432 | /* This function is called when a timer is fired or when |
433 | * throttle_group_restart_tgm() is called. Either way, there can |
434 | * be no timer pending on this tgm at this point */ |
435 | assert(!timer_pending(tgm->throttle_timers.timers[is_write])); |
436 | |
437 | atomic_inc(&tgm->restart_pending); |
438 | |
439 | co = qemu_coroutine_create(throttle_group_restart_queue_entry, rd); |
440 | aio_co_enter(tgm->aio_context, co); |
441 | } |
442 | |
443 | void throttle_group_restart_tgm(ThrottleGroupMember *tgm) |
444 | { |
445 | int i; |
446 | |
447 | if (tgm->throttle_state) { |
448 | for (i = 0; i < 2; i++) { |
449 | QEMUTimer *t = tgm->throttle_timers.timers[i]; |
450 | if (timer_pending(t)) { |
451 | /* If there's a pending timer on this tgm, fire it now */ |
452 | timer_del(t); |
453 | timer_cb(tgm, i); |
454 | } else { |
455 | /* Else run the next request from the queue manually */ |
456 | throttle_group_restart_queue(tgm, i); |
457 | } |
458 | } |
459 | } |
460 | } |
461 | |
462 | /* Update the throttle configuration for a particular group. Similar |
463 | * to throttle_config(), but guarantees atomicity within the |
464 | * throttling group. |
465 | * |
466 | * @tgm: a ThrottleGroupMember that is a member of the group |
467 | * @cfg: the configuration to set |
468 | */ |
469 | void throttle_group_config(ThrottleGroupMember *tgm, ThrottleConfig *cfg) |
470 | { |
471 | ThrottleState *ts = tgm->throttle_state; |
472 | ThrottleGroup *tg = container_of(ts, ThrottleGroup, ts); |
473 | qemu_mutex_lock(&tg->lock); |
474 | throttle_config(ts, tg->clock_type, cfg); |
475 | qemu_mutex_unlock(&tg->lock); |
476 | |
477 | throttle_group_restart_tgm(tgm); |
478 | } |
479 | |
480 | /* Get the throttle configuration from a particular group. Similar to |
481 | * throttle_get_config(), but guarantees atomicity within the |
482 | * throttling group. |
483 | * |
484 | * @tgm: a ThrottleGroupMember that is a member of the group |
485 | * @cfg: the configuration will be written here |
486 | */ |
487 | void throttle_group_get_config(ThrottleGroupMember *tgm, ThrottleConfig *cfg) |
488 | { |
489 | ThrottleState *ts = tgm->throttle_state; |
490 | ThrottleGroup *tg = container_of(ts, ThrottleGroup, ts); |
491 | qemu_mutex_lock(&tg->lock); |
492 | throttle_get_config(ts, cfg); |
493 | qemu_mutex_unlock(&tg->lock); |
494 | } |
495 | |
496 | /* ThrottleTimers callback. This wakes up a request that was waiting |
497 | * because it had been throttled. |
498 | * |
499 | * @tgm: the ThrottleGroupMember whose request had been throttled |
500 | * @is_write: the type of operation (read/write) |
501 | */ |
502 | static void timer_cb(ThrottleGroupMember *tgm, bool is_write) |
503 | { |
504 | ThrottleState *ts = tgm->throttle_state; |
505 | ThrottleGroup *tg = container_of(ts, ThrottleGroup, ts); |
506 | |
507 | /* The timer has just been fired, so we can update the flag */ |
508 | qemu_mutex_lock(&tg->lock); |
509 | tg->any_timer_armed[is_write] = false; |
510 | qemu_mutex_unlock(&tg->lock); |
511 | |
512 | /* Run the request that was waiting for this timer */ |
513 | throttle_group_restart_queue(tgm, is_write); |
514 | } |
515 | |
516 | static void read_timer_cb(void *opaque) |
517 | { |
518 | timer_cb(opaque, false); |
519 | } |
520 | |
521 | static void write_timer_cb(void *opaque) |
522 | { |
523 | timer_cb(opaque, true); |
524 | } |
525 | |
526 | /* Register a ThrottleGroupMember from the throttling group, also initializing |
527 | * its timers and updating its throttle_state pointer to point to it. If a |
528 | * throttling group with that name does not exist yet, it will be created. |
529 | * |
530 | * This function edits throttle_groups and must be called under the global |
531 | * mutex. |
532 | * |
533 | * @tgm: the ThrottleGroupMember to insert |
534 | * @groupname: the name of the group |
535 | * @ctx: the AioContext to use |
536 | */ |
537 | void throttle_group_register_tgm(ThrottleGroupMember *tgm, |
538 | const char *groupname, |
539 | AioContext *ctx) |
540 | { |
541 | int i; |
542 | ThrottleState *ts = throttle_group_incref(groupname); |
543 | ThrottleGroup *tg = container_of(ts, ThrottleGroup, ts); |
544 | |
545 | tgm->throttle_state = ts; |
546 | tgm->aio_context = ctx; |
547 | atomic_set(&tgm->restart_pending, 0); |
548 | |
549 | qemu_mutex_lock(&tg->lock); |
550 | /* If the ThrottleGroup is new set this ThrottleGroupMember as the token */ |
551 | for (i = 0; i < 2; i++) { |
552 | if (!tg->tokens[i]) { |
553 | tg->tokens[i] = tgm; |
554 | } |
555 | } |
556 | |
557 | QLIST_INSERT_HEAD(&tg->head, tgm, round_robin); |
558 | |
559 | throttle_timers_init(&tgm->throttle_timers, |
560 | tgm->aio_context, |
561 | tg->clock_type, |
562 | read_timer_cb, |
563 | write_timer_cb, |
564 | tgm); |
565 | qemu_co_mutex_init(&tgm->throttled_reqs_lock); |
566 | qemu_co_queue_init(&tgm->throttled_reqs[0]); |
567 | qemu_co_queue_init(&tgm->throttled_reqs[1]); |
568 | |
569 | qemu_mutex_unlock(&tg->lock); |
570 | } |
571 | |
572 | /* Unregister a ThrottleGroupMember from its group, removing it from the list, |
573 | * destroying the timers and setting the throttle_state pointer to NULL. |
574 | * |
575 | * The ThrottleGroupMember must not have pending throttled requests, so the |
576 | * caller has to drain them first. |
577 | * |
578 | * The group will be destroyed if it's empty after this operation. |
579 | * |
580 | * @tgm the ThrottleGroupMember to remove |
581 | */ |
582 | void throttle_group_unregister_tgm(ThrottleGroupMember *tgm) |
583 | { |
584 | ThrottleState *ts = tgm->throttle_state; |
585 | ThrottleGroup *tg = container_of(ts, ThrottleGroup, ts); |
586 | ThrottleGroupMember *token; |
587 | int i; |
588 | |
589 | if (!ts) { |
590 | /* Discard already unregistered tgm */ |
591 | return; |
592 | } |
593 | |
594 | /* Wait for throttle_group_restart_queue_entry() coroutines to finish */ |
595 | AIO_WAIT_WHILE(tgm->aio_context, atomic_read(&tgm->restart_pending) > 0); |
596 | |
597 | qemu_mutex_lock(&tg->lock); |
598 | for (i = 0; i < 2; i++) { |
599 | assert(tgm->pending_reqs[i] == 0); |
600 | assert(qemu_co_queue_empty(&tgm->throttled_reqs[i])); |
601 | assert(!timer_pending(tgm->throttle_timers.timers[i])); |
602 | if (tg->tokens[i] == tgm) { |
603 | token = throttle_group_next_tgm(tgm); |
604 | /* Take care of the case where this is the last tgm in the group */ |
605 | if (token == tgm) { |
606 | token = NULL; |
607 | } |
608 | tg->tokens[i] = token; |
609 | } |
610 | } |
611 | |
612 | /* remove the current tgm from the list */ |
613 | QLIST_REMOVE(tgm, round_robin); |
614 | throttle_timers_destroy(&tgm->throttle_timers); |
615 | qemu_mutex_unlock(&tg->lock); |
616 | |
617 | throttle_group_unref(&tg->ts); |
618 | tgm->throttle_state = NULL; |
619 | } |
620 | |
621 | void throttle_group_attach_aio_context(ThrottleGroupMember *tgm, |
622 | AioContext *new_context) |
623 | { |
624 | ThrottleTimers *tt = &tgm->throttle_timers; |
625 | throttle_timers_attach_aio_context(tt, new_context); |
626 | tgm->aio_context = new_context; |
627 | } |
628 | |
629 | void throttle_group_detach_aio_context(ThrottleGroupMember *tgm) |
630 | { |
631 | ThrottleGroup *tg = container_of(tgm->throttle_state, ThrottleGroup, ts); |
632 | ThrottleTimers *tt = &tgm->throttle_timers; |
633 | int i; |
634 | |
635 | /* Requests must have been drained */ |
636 | assert(tgm->pending_reqs[0] == 0 && tgm->pending_reqs[1] == 0); |
637 | assert(qemu_co_queue_empty(&tgm->throttled_reqs[0])); |
638 | assert(qemu_co_queue_empty(&tgm->throttled_reqs[1])); |
639 | |
640 | /* Kick off next ThrottleGroupMember, if necessary */ |
641 | qemu_mutex_lock(&tg->lock); |
642 | for (i = 0; i < 2; i++) { |
643 | if (timer_pending(tt->timers[i])) { |
644 | tg->any_timer_armed[i] = false; |
645 | schedule_next_request(tgm, i); |
646 | } |
647 | } |
648 | qemu_mutex_unlock(&tg->lock); |
649 | |
650 | throttle_timers_detach_aio_context(tt); |
651 | tgm->aio_context = NULL; |
652 | } |
653 | |
654 | #undef THROTTLE_OPT_PREFIX |
655 | #define THROTTLE_OPT_PREFIX "x-" |
656 | |
657 | /* Helper struct and array for QOM property setter/getter */ |
658 | typedef struct { |
659 | const char *name; |
660 | BucketType type; |
661 | enum { |
662 | AVG, |
663 | MAX, |
664 | BURST_LENGTH, |
665 | IOPS_SIZE, |
666 | } category; |
667 | } ThrottleParamInfo; |
668 | |
669 | static ThrottleParamInfo properties[] = { |
670 | { |
671 | THROTTLE_OPT_PREFIX QEMU_OPT_IOPS_TOTAL, |
672 | THROTTLE_OPS_TOTAL, AVG, |
673 | }, |
674 | { |
675 | THROTTLE_OPT_PREFIX QEMU_OPT_IOPS_TOTAL_MAX, |
676 | THROTTLE_OPS_TOTAL, MAX, |
677 | }, |
678 | { |
679 | THROTTLE_OPT_PREFIX QEMU_OPT_IOPS_TOTAL_MAX_LENGTH, |
680 | THROTTLE_OPS_TOTAL, BURST_LENGTH, |
681 | }, |
682 | { |
683 | THROTTLE_OPT_PREFIX QEMU_OPT_IOPS_READ, |
684 | THROTTLE_OPS_READ, AVG, |
685 | }, |
686 | { |
687 | THROTTLE_OPT_PREFIX QEMU_OPT_IOPS_READ_MAX, |
688 | THROTTLE_OPS_READ, MAX, |
689 | }, |
690 | { |
691 | THROTTLE_OPT_PREFIX QEMU_OPT_IOPS_READ_MAX_LENGTH, |
692 | THROTTLE_OPS_READ, BURST_LENGTH, |
693 | }, |
694 | { |
695 | THROTTLE_OPT_PREFIX QEMU_OPT_IOPS_WRITE, |
696 | THROTTLE_OPS_WRITE, AVG, |
697 | }, |
698 | { |
699 | THROTTLE_OPT_PREFIX QEMU_OPT_IOPS_WRITE_MAX, |
700 | THROTTLE_OPS_WRITE, MAX, |
701 | }, |
702 | { |
703 | THROTTLE_OPT_PREFIX QEMU_OPT_IOPS_WRITE_MAX_LENGTH, |
704 | THROTTLE_OPS_WRITE, BURST_LENGTH, |
705 | }, |
706 | { |
707 | THROTTLE_OPT_PREFIX QEMU_OPT_BPS_TOTAL, |
708 | THROTTLE_BPS_TOTAL, AVG, |
709 | }, |
710 | { |
711 | THROTTLE_OPT_PREFIX QEMU_OPT_BPS_TOTAL_MAX, |
712 | THROTTLE_BPS_TOTAL, MAX, |
713 | }, |
714 | { |
715 | THROTTLE_OPT_PREFIX QEMU_OPT_BPS_TOTAL_MAX_LENGTH, |
716 | THROTTLE_BPS_TOTAL, BURST_LENGTH, |
717 | }, |
718 | { |
719 | THROTTLE_OPT_PREFIX QEMU_OPT_BPS_READ, |
720 | THROTTLE_BPS_READ, AVG, |
721 | }, |
722 | { |
723 | THROTTLE_OPT_PREFIX QEMU_OPT_BPS_READ_MAX, |
724 | THROTTLE_BPS_READ, MAX, |
725 | }, |
726 | { |
727 | THROTTLE_OPT_PREFIX QEMU_OPT_BPS_READ_MAX_LENGTH, |
728 | THROTTLE_BPS_READ, BURST_LENGTH, |
729 | }, |
730 | { |
731 | THROTTLE_OPT_PREFIX QEMU_OPT_BPS_WRITE, |
732 | THROTTLE_BPS_WRITE, AVG, |
733 | }, |
734 | { |
735 | THROTTLE_OPT_PREFIX QEMU_OPT_BPS_WRITE_MAX, |
736 | THROTTLE_BPS_WRITE, MAX, |
737 | }, |
738 | { |
739 | THROTTLE_OPT_PREFIX QEMU_OPT_BPS_WRITE_MAX_LENGTH, |
740 | THROTTLE_BPS_WRITE, BURST_LENGTH, |
741 | }, |
742 | { |
743 | THROTTLE_OPT_PREFIX QEMU_OPT_IOPS_SIZE, |
744 | 0, IOPS_SIZE, |
745 | } |
746 | }; |
747 | |
748 | /* This function edits throttle_groups and must be called under the global |
749 | * mutex */ |
750 | static void throttle_group_obj_init(Object *obj) |
751 | { |
752 | ThrottleGroup *tg = THROTTLE_GROUP(obj); |
753 | |
754 | tg->clock_type = QEMU_CLOCK_REALTIME; |
755 | if (qtest_enabled()) { |
756 | /* For testing block IO throttling only */ |
757 | tg->clock_type = QEMU_CLOCK_VIRTUAL; |
758 | } |
759 | tg->is_initialized = false; |
760 | qemu_mutex_init(&tg->lock); |
761 | throttle_init(&tg->ts); |
762 | QLIST_INIT(&tg->head); |
763 | } |
764 | |
765 | /* This function edits throttle_groups and must be called under the global |
766 | * mutex */ |
767 | static void throttle_group_obj_complete(UserCreatable *obj, Error **errp) |
768 | { |
769 | ThrottleGroup *tg = THROTTLE_GROUP(obj); |
770 | ThrottleConfig cfg; |
771 | |
772 | /* set group name to object id if it exists */ |
773 | if (!tg->name && tg->parent_obj.parent) { |
774 | tg->name = object_get_canonical_path_component(OBJECT(obj)); |
775 | } |
776 | /* We must have a group name at this point */ |
777 | assert(tg->name); |
778 | |
779 | /* error if name is duplicate */ |
780 | if (throttle_group_exists(tg->name)) { |
781 | error_setg(errp, "A group with this name already exists" ); |
782 | return; |
783 | } |
784 | |
785 | /* check validity */ |
786 | throttle_get_config(&tg->ts, &cfg); |
787 | if (!throttle_is_valid(&cfg, errp)) { |
788 | return; |
789 | } |
790 | throttle_config(&tg->ts, tg->clock_type, &cfg); |
791 | QTAILQ_INSERT_TAIL(&throttle_groups, tg, list); |
792 | tg->is_initialized = true; |
793 | } |
794 | |
795 | /* This function edits throttle_groups and must be called under the global |
796 | * mutex */ |
797 | static void throttle_group_obj_finalize(Object *obj) |
798 | { |
799 | ThrottleGroup *tg = THROTTLE_GROUP(obj); |
800 | if (tg->is_initialized) { |
801 | QTAILQ_REMOVE(&throttle_groups, tg, list); |
802 | } |
803 | qemu_mutex_destroy(&tg->lock); |
804 | g_free(tg->name); |
805 | } |
806 | |
807 | static void throttle_group_set(Object *obj, Visitor *v, const char * name, |
808 | void *opaque, Error **errp) |
809 | |
810 | { |
811 | ThrottleGroup *tg = THROTTLE_GROUP(obj); |
812 | ThrottleConfig *cfg; |
813 | ThrottleParamInfo *info = opaque; |
814 | Error *local_err = NULL; |
815 | int64_t value; |
816 | |
817 | /* If we have finished initialization, don't accept individual property |
818 | * changes through QOM. Throttle configuration limits must be set in one |
819 | * transaction, as certain combinations are invalid. |
820 | */ |
821 | if (tg->is_initialized) { |
822 | error_setg(&local_err, "Property cannot be set after initialization" ); |
823 | goto ret; |
824 | } |
825 | |
826 | visit_type_int64(v, name, &value, &local_err); |
827 | if (local_err) { |
828 | goto ret; |
829 | } |
830 | if (value < 0) { |
831 | error_setg(&local_err, "Property values cannot be negative" ); |
832 | goto ret; |
833 | } |
834 | |
835 | cfg = &tg->ts.cfg; |
836 | switch (info->category) { |
837 | case AVG: |
838 | cfg->buckets[info->type].avg = value; |
839 | break; |
840 | case MAX: |
841 | cfg->buckets[info->type].max = value; |
842 | break; |
843 | case BURST_LENGTH: |
844 | if (value > UINT_MAX) { |
845 | error_setg(&local_err, "%s value must be in the" |
846 | "range [0, %u]" , info->name, UINT_MAX); |
847 | goto ret; |
848 | } |
849 | cfg->buckets[info->type].burst_length = value; |
850 | break; |
851 | case IOPS_SIZE: |
852 | cfg->op_size = value; |
853 | break; |
854 | } |
855 | |
856 | ret: |
857 | error_propagate(errp, local_err); |
858 | return; |
859 | |
860 | } |
861 | |
862 | static void throttle_group_get(Object *obj, Visitor *v, const char *name, |
863 | void *opaque, Error **errp) |
864 | { |
865 | ThrottleGroup *tg = THROTTLE_GROUP(obj); |
866 | ThrottleConfig cfg; |
867 | ThrottleParamInfo *info = opaque; |
868 | int64_t value; |
869 | |
870 | throttle_get_config(&tg->ts, &cfg); |
871 | switch (info->category) { |
872 | case AVG: |
873 | value = cfg.buckets[info->type].avg; |
874 | break; |
875 | case MAX: |
876 | value = cfg.buckets[info->type].max; |
877 | break; |
878 | case BURST_LENGTH: |
879 | value = cfg.buckets[info->type].burst_length; |
880 | break; |
881 | case IOPS_SIZE: |
882 | value = cfg.op_size; |
883 | break; |
884 | } |
885 | |
886 | visit_type_int64(v, name, &value, errp); |
887 | } |
888 | |
889 | static void throttle_group_set_limits(Object *obj, Visitor *v, |
890 | const char *name, void *opaque, |
891 | Error **errp) |
892 | |
893 | { |
894 | ThrottleGroup *tg = THROTTLE_GROUP(obj); |
895 | ThrottleConfig cfg; |
896 | ThrottleLimits arg = { 0 }; |
897 | ThrottleLimits *argp = &arg; |
898 | Error *local_err = NULL; |
899 | |
900 | visit_type_ThrottleLimits(v, name, &argp, &local_err); |
901 | if (local_err) { |
902 | goto ret; |
903 | } |
904 | qemu_mutex_lock(&tg->lock); |
905 | throttle_get_config(&tg->ts, &cfg); |
906 | throttle_limits_to_config(argp, &cfg, &local_err); |
907 | if (local_err) { |
908 | goto unlock; |
909 | } |
910 | throttle_config(&tg->ts, tg->clock_type, &cfg); |
911 | |
912 | unlock: |
913 | qemu_mutex_unlock(&tg->lock); |
914 | ret: |
915 | error_propagate(errp, local_err); |
916 | return; |
917 | } |
918 | |
919 | static void throttle_group_get_limits(Object *obj, Visitor *v, |
920 | const char *name, void *opaque, |
921 | Error **errp) |
922 | { |
923 | ThrottleGroup *tg = THROTTLE_GROUP(obj); |
924 | ThrottleConfig cfg; |
925 | ThrottleLimits arg = { 0 }; |
926 | ThrottleLimits *argp = &arg; |
927 | |
928 | qemu_mutex_lock(&tg->lock); |
929 | throttle_get_config(&tg->ts, &cfg); |
930 | qemu_mutex_unlock(&tg->lock); |
931 | |
932 | throttle_config_to_limits(&cfg, argp); |
933 | |
934 | visit_type_ThrottleLimits(v, name, &argp, errp); |
935 | } |
936 | |
937 | static bool throttle_group_can_be_deleted(UserCreatable *uc) |
938 | { |
939 | return OBJECT(uc)->ref == 1; |
940 | } |
941 | |
942 | static void throttle_group_obj_class_init(ObjectClass *klass, void *class_data) |
943 | { |
944 | size_t i = 0; |
945 | UserCreatableClass *ucc = USER_CREATABLE_CLASS(klass); |
946 | |
947 | ucc->complete = throttle_group_obj_complete; |
948 | ucc->can_be_deleted = throttle_group_can_be_deleted; |
949 | |
950 | /* individual properties */ |
951 | for (i = 0; i < sizeof(properties) / sizeof(ThrottleParamInfo); i++) { |
952 | object_class_property_add(klass, |
953 | properties[i].name, |
954 | "int" , |
955 | throttle_group_get, |
956 | throttle_group_set, |
957 | NULL, &properties[i], |
958 | &error_abort); |
959 | } |
960 | |
961 | /* ThrottleLimits */ |
962 | object_class_property_add(klass, |
963 | "limits" , "ThrottleLimits" , |
964 | throttle_group_get_limits, |
965 | throttle_group_set_limits, |
966 | NULL, NULL, |
967 | &error_abort); |
968 | } |
969 | |
970 | static const TypeInfo throttle_group_info = { |
971 | .name = TYPE_THROTTLE_GROUP, |
972 | .parent = TYPE_OBJECT, |
973 | .class_init = throttle_group_obj_class_init, |
974 | .instance_size = sizeof(ThrottleGroup), |
975 | .instance_init = throttle_group_obj_init, |
976 | .instance_finalize = throttle_group_obj_finalize, |
977 | .interfaces = (InterfaceInfo[]) { |
978 | { TYPE_USER_CREATABLE }, |
979 | { } |
980 | }, |
981 | }; |
982 | |
983 | static void throttle_groups_init(void) |
984 | { |
985 | type_register_static(&throttle_group_info); |
986 | } |
987 | |
988 | type_init(throttle_groups_init); |
989 | |