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