1c2b38b27SPaolo Bonzini /* 2c2b38b27SPaolo Bonzini * Data plane event loop 3c2b38b27SPaolo Bonzini * 4c2b38b27SPaolo Bonzini * Copyright (c) 2003-2008 Fabrice Bellard 5c2b38b27SPaolo Bonzini * Copyright (c) 2009-2017 QEMU contributors 6c2b38b27SPaolo Bonzini * 7c2b38b27SPaolo Bonzini * Permission is hereby granted, free of charge, to any person obtaining a copy 8c2b38b27SPaolo Bonzini * of this software and associated documentation files (the "Software"), to deal 9c2b38b27SPaolo Bonzini * in the Software without restriction, including without limitation the rights 10c2b38b27SPaolo Bonzini * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell 11c2b38b27SPaolo Bonzini * copies of the Software, and to permit persons to whom the Software is 12c2b38b27SPaolo Bonzini * furnished to do so, subject to the following conditions: 13c2b38b27SPaolo Bonzini * 14c2b38b27SPaolo Bonzini * The above copyright notice and this permission notice shall be included in 15c2b38b27SPaolo Bonzini * all copies or substantial portions of the Software. 16c2b38b27SPaolo Bonzini * 17c2b38b27SPaolo Bonzini * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 18c2b38b27SPaolo Bonzini * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 19c2b38b27SPaolo Bonzini * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 20c2b38b27SPaolo Bonzini * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 21c2b38b27SPaolo Bonzini * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 22c2b38b27SPaolo Bonzini * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN 23c2b38b27SPaolo Bonzini * THE SOFTWARE. 24c2b38b27SPaolo Bonzini */ 25c2b38b27SPaolo Bonzini 26c2b38b27SPaolo Bonzini #include "qemu/osdep.h" 27c2b38b27SPaolo Bonzini #include "qapi/error.h" 28c2b38b27SPaolo Bonzini #include "block/aio.h" 29c2b38b27SPaolo Bonzini #include "block/thread-pool.h" 30587d82faSEmanuele Giuseppe Esposito #include "block/graph-lock.h" 31c2b38b27SPaolo Bonzini #include "qemu/main-loop.h" 32c2b38b27SPaolo Bonzini #include "qemu/atomic.h" 338c6b0356SStefan Hajnoczi #include "qemu/rcu_queue.h" 34c2b38b27SPaolo Bonzini #include "block/raw-aio.h" 350c330a73SPaolo Bonzini #include "qemu/coroutine_int.h" 3647b74464SStefan Hajnoczi #include "qemu/coroutine-tls.h" 3775bbe5e5SPavel Dovgalyuk #include "sysemu/cpu-timers.h" 380c330a73SPaolo Bonzini #include "trace.h" 39c2b38b27SPaolo Bonzini 40c2b38b27SPaolo Bonzini /***********************************************************/ 41c2b38b27SPaolo Bonzini /* bottom halves (can be seen as timers which expire ASAP) */ 42c2b38b27SPaolo Bonzini 438c6b0356SStefan Hajnoczi /* QEMUBH::flags values */ 448c6b0356SStefan Hajnoczi enum { 458c6b0356SStefan Hajnoczi /* Already enqueued and waiting for aio_bh_poll() */ 468c6b0356SStefan Hajnoczi BH_PENDING = (1 << 0), 478c6b0356SStefan Hajnoczi 488c6b0356SStefan Hajnoczi /* Invoke the callback */ 498c6b0356SStefan Hajnoczi BH_SCHEDULED = (1 << 1), 508c6b0356SStefan Hajnoczi 518c6b0356SStefan Hajnoczi /* Delete without invoking callback */ 528c6b0356SStefan Hajnoczi BH_DELETED = (1 << 2), 538c6b0356SStefan Hajnoczi 548c6b0356SStefan Hajnoczi /* Delete after invoking callback */ 558c6b0356SStefan Hajnoczi BH_ONESHOT = (1 << 3), 568c6b0356SStefan Hajnoczi 578c6b0356SStefan Hajnoczi /* Schedule periodically when the event loop is idle */ 588c6b0356SStefan Hajnoczi BH_IDLE = (1 << 4), 598c6b0356SStefan Hajnoczi }; 608c6b0356SStefan Hajnoczi 61c2b38b27SPaolo Bonzini struct QEMUBH { 62c2b38b27SPaolo Bonzini AioContext *ctx; 630f08586cSStefan Hajnoczi const char *name; 64c2b38b27SPaolo Bonzini QEMUBHFunc *cb; 65c2b38b27SPaolo Bonzini void *opaque; 668c6b0356SStefan Hajnoczi QSLIST_ENTRY(QEMUBH) next; 678c6b0356SStefan Hajnoczi unsigned flags; 689c86c97fSAlexander Bulekov MemReentrancyGuard *reentrancy_guard; 69c2b38b27SPaolo Bonzini }; 70c2b38b27SPaolo Bonzini 718c6b0356SStefan Hajnoczi /* Called concurrently from any thread */ 728c6b0356SStefan Hajnoczi static void aio_bh_enqueue(QEMUBH *bh, unsigned new_flags) 738c6b0356SStefan Hajnoczi { 748c6b0356SStefan Hajnoczi AioContext *ctx = bh->ctx; 758c6b0356SStefan Hajnoczi unsigned old_flags; 768c6b0356SStefan Hajnoczi 778c6b0356SStefan Hajnoczi /* 788dd48650SPaolo Bonzini * Synchronizes with atomic_fetch_and() in aio_bh_dequeue(), ensuring that 798dd48650SPaolo Bonzini * insertion starts after BH_PENDING is set. 808dd48650SPaolo Bonzini */ 818dd48650SPaolo Bonzini old_flags = qatomic_fetch_or(&bh->flags, BH_PENDING | new_flags); 828dd48650SPaolo Bonzini 838dd48650SPaolo Bonzini if (!(old_flags & BH_PENDING)) { 848dd48650SPaolo Bonzini /* 858dd48650SPaolo Bonzini * At this point the bottom half becomes visible to aio_bh_poll(). 868dd48650SPaolo Bonzini * This insertion thus synchronizes with QSLIST_MOVE_ATOMIC in 878dd48650SPaolo Bonzini * aio_bh_poll(), ensuring that: 888dd48650SPaolo Bonzini * 1. any writes needed by the callback are visible from the callback 898dd48650SPaolo Bonzini * after aio_bh_dequeue() returns bh. 908c6b0356SStefan Hajnoczi * 2. ctx is loaded before the callback has a chance to execute and bh 918c6b0356SStefan Hajnoczi * could be freed. 928c6b0356SStefan Hajnoczi */ 938c6b0356SStefan Hajnoczi QSLIST_INSERT_HEAD_ATOMIC(&ctx->bh_list, bh, next); 948c6b0356SStefan Hajnoczi } 958c6b0356SStefan Hajnoczi 968c6b0356SStefan Hajnoczi aio_notify(ctx); 9775bbe5e5SPavel Dovgalyuk /* 9875bbe5e5SPavel Dovgalyuk * Workaround for record/replay. 9975bbe5e5SPavel Dovgalyuk * vCPU execution should be suspended when new BH is set. 10075bbe5e5SPavel Dovgalyuk * This is needed to avoid guest timeouts caused 10175bbe5e5SPavel Dovgalyuk * by the long cycles of the execution. 10275bbe5e5SPavel Dovgalyuk */ 10375bbe5e5SPavel Dovgalyuk icount_notify_exit(); 1048c6b0356SStefan Hajnoczi } 1058c6b0356SStefan Hajnoczi 1068c6b0356SStefan Hajnoczi /* Only called from aio_bh_poll() and aio_ctx_finalize() */ 1078c6b0356SStefan Hajnoczi static QEMUBH *aio_bh_dequeue(BHList *head, unsigned *flags) 1088c6b0356SStefan Hajnoczi { 1098c6b0356SStefan Hajnoczi QEMUBH *bh = QSLIST_FIRST_RCU(head); 1108c6b0356SStefan Hajnoczi 1118c6b0356SStefan Hajnoczi if (!bh) { 1128c6b0356SStefan Hajnoczi return NULL; 1138c6b0356SStefan Hajnoczi } 1148c6b0356SStefan Hajnoczi 1158c6b0356SStefan Hajnoczi QSLIST_REMOVE_HEAD(head, next); 1168c6b0356SStefan Hajnoczi 1178c6b0356SStefan Hajnoczi /* 1188dd48650SPaolo Bonzini * Synchronizes with qatomic_fetch_or() in aio_bh_enqueue(), ensuring that 1198dd48650SPaolo Bonzini * the removal finishes before BH_PENDING is reset. 1208c6b0356SStefan Hajnoczi */ 121d73415a3SStefan Hajnoczi *flags = qatomic_fetch_and(&bh->flags, 1228c6b0356SStefan Hajnoczi ~(BH_PENDING | BH_SCHEDULED | BH_IDLE)); 1238c6b0356SStefan Hajnoczi return bh; 1248c6b0356SStefan Hajnoczi } 1258c6b0356SStefan Hajnoczi 1260f08586cSStefan Hajnoczi void aio_bh_schedule_oneshot_full(AioContext *ctx, QEMUBHFunc *cb, 1270f08586cSStefan Hajnoczi void *opaque, const char *name) 128c2b38b27SPaolo Bonzini { 129c2b38b27SPaolo Bonzini QEMUBH *bh; 130c2b38b27SPaolo Bonzini bh = g_new(QEMUBH, 1); 131c2b38b27SPaolo Bonzini *bh = (QEMUBH){ 132c2b38b27SPaolo Bonzini .ctx = ctx, 133c2b38b27SPaolo Bonzini .cb = cb, 134c2b38b27SPaolo Bonzini .opaque = opaque, 1350f08586cSStefan Hajnoczi .name = name, 136c2b38b27SPaolo Bonzini }; 1378c6b0356SStefan Hajnoczi aio_bh_enqueue(bh, BH_SCHEDULED | BH_ONESHOT); 138c2b38b27SPaolo Bonzini } 139c2b38b27SPaolo Bonzini 1400f08586cSStefan Hajnoczi QEMUBH *aio_bh_new_full(AioContext *ctx, QEMUBHFunc *cb, void *opaque, 1419c86c97fSAlexander Bulekov const char *name, MemReentrancyGuard *reentrancy_guard) 142c2b38b27SPaolo Bonzini { 143c2b38b27SPaolo Bonzini QEMUBH *bh; 144c2b38b27SPaolo Bonzini bh = g_new(QEMUBH, 1); 145c2b38b27SPaolo Bonzini *bh = (QEMUBH){ 146c2b38b27SPaolo Bonzini .ctx = ctx, 147c2b38b27SPaolo Bonzini .cb = cb, 148c2b38b27SPaolo Bonzini .opaque = opaque, 1490f08586cSStefan Hajnoczi .name = name, 1509c86c97fSAlexander Bulekov .reentrancy_guard = reentrancy_guard, 151c2b38b27SPaolo Bonzini }; 152c2b38b27SPaolo Bonzini return bh; 153c2b38b27SPaolo Bonzini } 154c2b38b27SPaolo Bonzini 155c2b38b27SPaolo Bonzini void aio_bh_call(QEMUBH *bh) 156c2b38b27SPaolo Bonzini { 1579c86c97fSAlexander Bulekov bool last_engaged_in_io = false; 1589c86c97fSAlexander Bulekov 159*7915bd06SAlexander Bulekov /* Make a copy of the guard-pointer as cb may free the bh */ 160*7915bd06SAlexander Bulekov MemReentrancyGuard *reentrancy_guard = bh->reentrancy_guard; 161*7915bd06SAlexander Bulekov if (reentrancy_guard) { 162*7915bd06SAlexander Bulekov last_engaged_in_io = reentrancy_guard->engaged_in_io; 163*7915bd06SAlexander Bulekov if (reentrancy_guard->engaged_in_io) { 1649c86c97fSAlexander Bulekov trace_reentrant_aio(bh->ctx, bh->name); 1659c86c97fSAlexander Bulekov } 166*7915bd06SAlexander Bulekov reentrancy_guard->engaged_in_io = true; 1679c86c97fSAlexander Bulekov } 1689c86c97fSAlexander Bulekov 169c2b38b27SPaolo Bonzini bh->cb(bh->opaque); 1709c86c97fSAlexander Bulekov 171*7915bd06SAlexander Bulekov if (reentrancy_guard) { 172*7915bd06SAlexander Bulekov reentrancy_guard->engaged_in_io = last_engaged_in_io; 1739c86c97fSAlexander Bulekov } 174c2b38b27SPaolo Bonzini } 175c2b38b27SPaolo Bonzini 1768c6b0356SStefan Hajnoczi /* Multiple occurrences of aio_bh_poll cannot be called concurrently. */ 177c2b38b27SPaolo Bonzini int aio_bh_poll(AioContext *ctx) 178c2b38b27SPaolo Bonzini { 1798c6b0356SStefan Hajnoczi BHListSlice slice; 1808c6b0356SStefan Hajnoczi BHListSlice *s; 1818c6b0356SStefan Hajnoczi int ret = 0; 182c2b38b27SPaolo Bonzini 1838dd48650SPaolo Bonzini /* Synchronizes with QSLIST_INSERT_HEAD_ATOMIC in aio_bh_enqueue(). */ 1848c6b0356SStefan Hajnoczi QSLIST_MOVE_ATOMIC(&slice.bh_list, &ctx->bh_list); 185d66ba6dcSCédric Le Goater 186d66ba6dcSCédric Le Goater /* 187d66ba6dcSCédric Le Goater * GCC13 [-Werror=dangling-pointer=] complains that the local variable 188d66ba6dcSCédric Le Goater * 'slice' is being stored in the global 'ctx->bh_slice_list' but the 189d66ba6dcSCédric Le Goater * list is emptied before this function returns. 190d66ba6dcSCédric Le Goater */ 191d66ba6dcSCédric Le Goater #if !defined(__clang__) 192d66ba6dcSCédric Le Goater #pragma GCC diagnostic push 193d66ba6dcSCédric Le Goater #pragma GCC diagnostic ignored "-Wpragmas" 194d66ba6dcSCédric Le Goater #pragma GCC diagnostic ignored "-Wdangling-pointer=" 195d66ba6dcSCédric Le Goater #endif 1968c6b0356SStefan Hajnoczi QSIMPLEQ_INSERT_TAIL(&ctx->bh_slice_list, &slice, next); 197d66ba6dcSCédric Le Goater #if !defined(__clang__) 198d66ba6dcSCédric Le Goater #pragma GCC diagnostic pop 199d66ba6dcSCédric Le Goater #endif 2008c6b0356SStefan Hajnoczi 2018c6b0356SStefan Hajnoczi while ((s = QSIMPLEQ_FIRST(&ctx->bh_slice_list))) { 2028c6b0356SStefan Hajnoczi QEMUBH *bh; 2038c6b0356SStefan Hajnoczi unsigned flags; 2048c6b0356SStefan Hajnoczi 2058c6b0356SStefan Hajnoczi bh = aio_bh_dequeue(&s->bh_list, &flags); 2068c6b0356SStefan Hajnoczi if (!bh) { 2078c6b0356SStefan Hajnoczi QSIMPLEQ_REMOVE_HEAD(&ctx->bh_slice_list, next); 2088c6b0356SStefan Hajnoczi continue; 2098c6b0356SStefan Hajnoczi } 2108c6b0356SStefan Hajnoczi 2118c6b0356SStefan Hajnoczi if ((flags & (BH_SCHEDULED | BH_DELETED)) == BH_SCHEDULED) { 212c2b38b27SPaolo Bonzini /* Idle BHs don't count as progress */ 2138c6b0356SStefan Hajnoczi if (!(flags & BH_IDLE)) { 214c2b38b27SPaolo Bonzini ret = 1; 215c2b38b27SPaolo Bonzini } 216c2b38b27SPaolo Bonzini aio_bh_call(bh); 217c2b38b27SPaolo Bonzini } 2188c6b0356SStefan Hajnoczi if (flags & (BH_DELETED | BH_ONESHOT)) { 219c2b38b27SPaolo Bonzini g_free(bh); 220c2b38b27SPaolo Bonzini } 221c2b38b27SPaolo Bonzini } 2228c6b0356SStefan Hajnoczi 223c2b38b27SPaolo Bonzini return ret; 224c2b38b27SPaolo Bonzini } 225c2b38b27SPaolo Bonzini 226c2b38b27SPaolo Bonzini void qemu_bh_schedule_idle(QEMUBH *bh) 227c2b38b27SPaolo Bonzini { 2288c6b0356SStefan Hajnoczi aio_bh_enqueue(bh, BH_SCHEDULED | BH_IDLE); 229c2b38b27SPaolo Bonzini } 230c2b38b27SPaolo Bonzini 231c2b38b27SPaolo Bonzini void qemu_bh_schedule(QEMUBH *bh) 232c2b38b27SPaolo Bonzini { 2338c6b0356SStefan Hajnoczi aio_bh_enqueue(bh, BH_SCHEDULED); 234c2b38b27SPaolo Bonzini } 235c2b38b27SPaolo Bonzini 236c2b38b27SPaolo Bonzini /* This func is async. 237c2b38b27SPaolo Bonzini */ 238c2b38b27SPaolo Bonzini void qemu_bh_cancel(QEMUBH *bh) 239c2b38b27SPaolo Bonzini { 240d73415a3SStefan Hajnoczi qatomic_and(&bh->flags, ~BH_SCHEDULED); 241c2b38b27SPaolo Bonzini } 242c2b38b27SPaolo Bonzini 243c2b38b27SPaolo Bonzini /* This func is async.The bottom half will do the delete action at the finial 244c2b38b27SPaolo Bonzini * end. 245c2b38b27SPaolo Bonzini */ 246c2b38b27SPaolo Bonzini void qemu_bh_delete(QEMUBH *bh) 247c2b38b27SPaolo Bonzini { 2488c6b0356SStefan Hajnoczi aio_bh_enqueue(bh, BH_DELETED); 249c2b38b27SPaolo Bonzini } 250c2b38b27SPaolo Bonzini 2518c6b0356SStefan Hajnoczi static int64_t aio_compute_bh_timeout(BHList *head, int timeout) 252c2b38b27SPaolo Bonzini { 253c2b38b27SPaolo Bonzini QEMUBH *bh; 254c2b38b27SPaolo Bonzini 2558c6b0356SStefan Hajnoczi QSLIST_FOREACH_RCU(bh, head, next) { 2568c6b0356SStefan Hajnoczi if ((bh->flags & (BH_SCHEDULED | BH_DELETED)) == BH_SCHEDULED) { 2578c6b0356SStefan Hajnoczi if (bh->flags & BH_IDLE) { 258c2b38b27SPaolo Bonzini /* idle bottom halves will be polled at least 259c2b38b27SPaolo Bonzini * every 10ms */ 260c2b38b27SPaolo Bonzini timeout = 10000000; 261c2b38b27SPaolo Bonzini } else { 262c2b38b27SPaolo Bonzini /* non-idle bottom halves will be executed 263c2b38b27SPaolo Bonzini * immediately */ 264c2b38b27SPaolo Bonzini return 0; 265c2b38b27SPaolo Bonzini } 266c2b38b27SPaolo Bonzini } 267c2b38b27SPaolo Bonzini } 268c2b38b27SPaolo Bonzini 2698c6b0356SStefan Hajnoczi return timeout; 2708c6b0356SStefan Hajnoczi } 2718c6b0356SStefan Hajnoczi 2728c6b0356SStefan Hajnoczi int64_t 2738c6b0356SStefan Hajnoczi aio_compute_timeout(AioContext *ctx) 2748c6b0356SStefan Hajnoczi { 2758c6b0356SStefan Hajnoczi BHListSlice *s; 2768c6b0356SStefan Hajnoczi int64_t deadline; 2778c6b0356SStefan Hajnoczi int timeout = -1; 2788c6b0356SStefan Hajnoczi 2798c6b0356SStefan Hajnoczi timeout = aio_compute_bh_timeout(&ctx->bh_list, timeout); 2808c6b0356SStefan Hajnoczi if (timeout == 0) { 2818c6b0356SStefan Hajnoczi return 0; 2828c6b0356SStefan Hajnoczi } 2838c6b0356SStefan Hajnoczi 2848c6b0356SStefan Hajnoczi QSIMPLEQ_FOREACH(s, &ctx->bh_slice_list, next) { 2858c6b0356SStefan Hajnoczi timeout = aio_compute_bh_timeout(&s->bh_list, timeout); 2868c6b0356SStefan Hajnoczi if (timeout == 0) { 2878c6b0356SStefan Hajnoczi return 0; 2888c6b0356SStefan Hajnoczi } 2898c6b0356SStefan Hajnoczi } 2908c6b0356SStefan Hajnoczi 291c2b38b27SPaolo Bonzini deadline = timerlistgroup_deadline_ns(&ctx->tlg); 292c2b38b27SPaolo Bonzini if (deadline == 0) { 293c2b38b27SPaolo Bonzini return 0; 294c2b38b27SPaolo Bonzini } else { 295c2b38b27SPaolo Bonzini return qemu_soonest_timeout(timeout, deadline); 296c2b38b27SPaolo Bonzini } 297c2b38b27SPaolo Bonzini } 298c2b38b27SPaolo Bonzini 299c2b38b27SPaolo Bonzini static gboolean 300c2b38b27SPaolo Bonzini aio_ctx_prepare(GSource *source, gint *timeout) 301c2b38b27SPaolo Bonzini { 302c2b38b27SPaolo Bonzini AioContext *ctx = (AioContext *) source; 303c2b38b27SPaolo Bonzini 304d73415a3SStefan Hajnoczi qatomic_set(&ctx->notify_me, qatomic_read(&ctx->notify_me) | 1); 3055710a3e0SPaolo Bonzini 3065710a3e0SPaolo Bonzini /* 3075710a3e0SPaolo Bonzini * Write ctx->notify_me before computing the timeout 3085710a3e0SPaolo Bonzini * (reading bottom half flags, etc.). Pairs with 3095710a3e0SPaolo Bonzini * smp_mb in aio_notify(). 3105710a3e0SPaolo Bonzini */ 3115710a3e0SPaolo Bonzini smp_mb(); 312c2b38b27SPaolo Bonzini 313c2b38b27SPaolo Bonzini /* We assume there is no timeout already supplied */ 314c2b38b27SPaolo Bonzini *timeout = qemu_timeout_ns_to_ms(aio_compute_timeout(ctx)); 315c2b38b27SPaolo Bonzini 316c2b38b27SPaolo Bonzini if (aio_prepare(ctx)) { 317c2b38b27SPaolo Bonzini *timeout = 0; 318c2b38b27SPaolo Bonzini } 319c2b38b27SPaolo Bonzini 320c2b38b27SPaolo Bonzini return *timeout == 0; 321c2b38b27SPaolo Bonzini } 322c2b38b27SPaolo Bonzini 323c2b38b27SPaolo Bonzini static gboolean 324c2b38b27SPaolo Bonzini aio_ctx_check(GSource *source) 325c2b38b27SPaolo Bonzini { 326c2b38b27SPaolo Bonzini AioContext *ctx = (AioContext *) source; 327c2b38b27SPaolo Bonzini QEMUBH *bh; 3288c6b0356SStefan Hajnoczi BHListSlice *s; 329c2b38b27SPaolo Bonzini 3305710a3e0SPaolo Bonzini /* Finish computing the timeout before clearing the flag. */ 331d73415a3SStefan Hajnoczi qatomic_store_release(&ctx->notify_me, qatomic_read(&ctx->notify_me) & ~1); 332c2b38b27SPaolo Bonzini aio_notify_accept(ctx); 333c2b38b27SPaolo Bonzini 3348c6b0356SStefan Hajnoczi QSLIST_FOREACH_RCU(bh, &ctx->bh_list, next) { 3358c6b0356SStefan Hajnoczi if ((bh->flags & (BH_SCHEDULED | BH_DELETED)) == BH_SCHEDULED) { 336c2b38b27SPaolo Bonzini return true; 337c2b38b27SPaolo Bonzini } 338c2b38b27SPaolo Bonzini } 3398c6b0356SStefan Hajnoczi 3408c6b0356SStefan Hajnoczi QSIMPLEQ_FOREACH(s, &ctx->bh_slice_list, next) { 3418c6b0356SStefan Hajnoczi QSLIST_FOREACH_RCU(bh, &s->bh_list, next) { 3428c6b0356SStefan Hajnoczi if ((bh->flags & (BH_SCHEDULED | BH_DELETED)) == BH_SCHEDULED) { 3438c6b0356SStefan Hajnoczi return true; 3448c6b0356SStefan Hajnoczi } 3458c6b0356SStefan Hajnoczi } 3468c6b0356SStefan Hajnoczi } 347c2b38b27SPaolo Bonzini return aio_pending(ctx) || (timerlistgroup_deadline_ns(&ctx->tlg) == 0); 348c2b38b27SPaolo Bonzini } 349c2b38b27SPaolo Bonzini 350c2b38b27SPaolo Bonzini static gboolean 351c2b38b27SPaolo Bonzini aio_ctx_dispatch(GSource *source, 352c2b38b27SPaolo Bonzini GSourceFunc callback, 353c2b38b27SPaolo Bonzini gpointer user_data) 354c2b38b27SPaolo Bonzini { 355c2b38b27SPaolo Bonzini AioContext *ctx = (AioContext *) source; 356c2b38b27SPaolo Bonzini 357c2b38b27SPaolo Bonzini assert(callback == NULL); 358a153bf52SPaolo Bonzini aio_dispatch(ctx); 359c2b38b27SPaolo Bonzini return true; 360c2b38b27SPaolo Bonzini } 361c2b38b27SPaolo Bonzini 362c2b38b27SPaolo Bonzini static void 363c2b38b27SPaolo Bonzini aio_ctx_finalize(GSource *source) 364c2b38b27SPaolo Bonzini { 365c2b38b27SPaolo Bonzini AioContext *ctx = (AioContext *) source; 3668c6b0356SStefan Hajnoczi QEMUBH *bh; 3678c6b0356SStefan Hajnoczi unsigned flags; 368c2b38b27SPaolo Bonzini 369c2b38b27SPaolo Bonzini thread_pool_free(ctx->thread_pool); 370c2b38b27SPaolo Bonzini 371c2b38b27SPaolo Bonzini #ifdef CONFIG_LINUX_AIO 372c2b38b27SPaolo Bonzini if (ctx->linux_aio) { 373c2b38b27SPaolo Bonzini laio_detach_aio_context(ctx->linux_aio, ctx); 374c2b38b27SPaolo Bonzini laio_cleanup(ctx->linux_aio); 375c2b38b27SPaolo Bonzini ctx->linux_aio = NULL; 376c2b38b27SPaolo Bonzini } 377c2b38b27SPaolo Bonzini #endif 378c2b38b27SPaolo Bonzini 379fcb7a4a4SAarushi Mehta #ifdef CONFIG_LINUX_IO_URING 380fcb7a4a4SAarushi Mehta if (ctx->linux_io_uring) { 381fcb7a4a4SAarushi Mehta luring_detach_aio_context(ctx->linux_io_uring, ctx); 382fcb7a4a4SAarushi Mehta luring_cleanup(ctx->linux_io_uring); 383fcb7a4a4SAarushi Mehta ctx->linux_io_uring = NULL; 384fcb7a4a4SAarushi Mehta } 385fcb7a4a4SAarushi Mehta #endif 386fcb7a4a4SAarushi Mehta 3870c330a73SPaolo Bonzini assert(QSLIST_EMPTY(&ctx->scheduled_coroutines)); 3880c330a73SPaolo Bonzini qemu_bh_delete(ctx->co_schedule_bh); 3890c330a73SPaolo Bonzini 3908c6b0356SStefan Hajnoczi /* There must be no aio_bh_poll() calls going on */ 3918c6b0356SStefan Hajnoczi assert(QSIMPLEQ_EMPTY(&ctx->bh_slice_list)); 392c2b38b27SPaolo Bonzini 3938c6b0356SStefan Hajnoczi while ((bh = aio_bh_dequeue(&ctx->bh_list, &flags))) { 394023ca420SStefan Hajnoczi /* 395023ca420SStefan Hajnoczi * qemu_bh_delete() must have been called on BHs in this AioContext. In 396023ca420SStefan Hajnoczi * many cases memory leaks, hangs, or inconsistent state occur when a 397023ca420SStefan Hajnoczi * BH is leaked because something still expects it to run. 398023ca420SStefan Hajnoczi * 399023ca420SStefan Hajnoczi * If you hit this, fix the lifecycle of the BH so that 400023ca420SStefan Hajnoczi * qemu_bh_delete() and any associated cleanup is called before the 401023ca420SStefan Hajnoczi * AioContext is finalized. 402023ca420SStefan Hajnoczi */ 403023ca420SStefan Hajnoczi if (unlikely(!(flags & BH_DELETED))) { 404023ca420SStefan Hajnoczi fprintf(stderr, "%s: BH '%s' leaked, aborting...\n", 405023ca420SStefan Hajnoczi __func__, bh->name); 406023ca420SStefan Hajnoczi abort(); 407023ca420SStefan Hajnoczi } 408c2b38b27SPaolo Bonzini 4098c6b0356SStefan Hajnoczi g_free(bh); 410c2b38b27SPaolo Bonzini } 411c2b38b27SPaolo Bonzini 412826cc324SStefan Hajnoczi aio_set_event_notifier(ctx, &ctx->notifier, false, NULL, NULL, NULL); 413c2b38b27SPaolo Bonzini event_notifier_cleanup(&ctx->notifier); 414c2b38b27SPaolo Bonzini qemu_rec_mutex_destroy(&ctx->lock); 415c2b38b27SPaolo Bonzini qemu_lockcnt_destroy(&ctx->list_lock); 416c2b38b27SPaolo Bonzini timerlistgroup_deinit(&ctx->tlg); 417587d82faSEmanuele Giuseppe Esposito unregister_aiocontext(ctx); 418cd0a6d2bSJie Wang aio_context_destroy(ctx); 419c2b38b27SPaolo Bonzini } 420c2b38b27SPaolo Bonzini 421c2b38b27SPaolo Bonzini static GSourceFuncs aio_source_funcs = { 422c2b38b27SPaolo Bonzini aio_ctx_prepare, 423c2b38b27SPaolo Bonzini aio_ctx_check, 424c2b38b27SPaolo Bonzini aio_ctx_dispatch, 425c2b38b27SPaolo Bonzini aio_ctx_finalize 426c2b38b27SPaolo Bonzini }; 427c2b38b27SPaolo Bonzini 428c2b38b27SPaolo Bonzini GSource *aio_get_g_source(AioContext *ctx) 429c2b38b27SPaolo Bonzini { 430ba607ca8SStefan Hajnoczi aio_context_use_g_source(ctx); 431c2b38b27SPaolo Bonzini g_source_ref(&ctx->source); 432c2b38b27SPaolo Bonzini return &ctx->source; 433c2b38b27SPaolo Bonzini } 434c2b38b27SPaolo Bonzini 435c2b38b27SPaolo Bonzini ThreadPool *aio_get_thread_pool(AioContext *ctx) 436c2b38b27SPaolo Bonzini { 437c2b38b27SPaolo Bonzini if (!ctx->thread_pool) { 438c2b38b27SPaolo Bonzini ctx->thread_pool = thread_pool_new(ctx); 439c2b38b27SPaolo Bonzini } 440c2b38b27SPaolo Bonzini return ctx->thread_pool; 441c2b38b27SPaolo Bonzini } 442c2b38b27SPaolo Bonzini 443c2b38b27SPaolo Bonzini #ifdef CONFIG_LINUX_AIO 444ed6e2161SNishanth Aravamudan LinuxAioState *aio_setup_linux_aio(AioContext *ctx, Error **errp) 445c2b38b27SPaolo Bonzini { 446c2b38b27SPaolo Bonzini if (!ctx->linux_aio) { 447ed6e2161SNishanth Aravamudan ctx->linux_aio = laio_init(errp); 448ed6e2161SNishanth Aravamudan if (ctx->linux_aio) { 449c2b38b27SPaolo Bonzini laio_attach_aio_context(ctx->linux_aio, ctx); 450c2b38b27SPaolo Bonzini } 451ed6e2161SNishanth Aravamudan } 452ed6e2161SNishanth Aravamudan return ctx->linux_aio; 453ed6e2161SNishanth Aravamudan } 454ed6e2161SNishanth Aravamudan 455ed6e2161SNishanth Aravamudan LinuxAioState *aio_get_linux_aio(AioContext *ctx) 456ed6e2161SNishanth Aravamudan { 457ed6e2161SNishanth Aravamudan assert(ctx->linux_aio); 458c2b38b27SPaolo Bonzini return ctx->linux_aio; 459c2b38b27SPaolo Bonzini } 460c2b38b27SPaolo Bonzini #endif 461c2b38b27SPaolo Bonzini 462fcb7a4a4SAarushi Mehta #ifdef CONFIG_LINUX_IO_URING 463fcb7a4a4SAarushi Mehta LuringState *aio_setup_linux_io_uring(AioContext *ctx, Error **errp) 464fcb7a4a4SAarushi Mehta { 465fcb7a4a4SAarushi Mehta if (ctx->linux_io_uring) { 466fcb7a4a4SAarushi Mehta return ctx->linux_io_uring; 467fcb7a4a4SAarushi Mehta } 468fcb7a4a4SAarushi Mehta 469fcb7a4a4SAarushi Mehta ctx->linux_io_uring = luring_init(errp); 470fcb7a4a4SAarushi Mehta if (!ctx->linux_io_uring) { 471fcb7a4a4SAarushi Mehta return NULL; 472fcb7a4a4SAarushi Mehta } 473fcb7a4a4SAarushi Mehta 474fcb7a4a4SAarushi Mehta luring_attach_aio_context(ctx->linux_io_uring, ctx); 475fcb7a4a4SAarushi Mehta return ctx->linux_io_uring; 476fcb7a4a4SAarushi Mehta } 477fcb7a4a4SAarushi Mehta 478fcb7a4a4SAarushi Mehta LuringState *aio_get_linux_io_uring(AioContext *ctx) 479fcb7a4a4SAarushi Mehta { 480fcb7a4a4SAarushi Mehta assert(ctx->linux_io_uring); 481fcb7a4a4SAarushi Mehta return ctx->linux_io_uring; 482fcb7a4a4SAarushi Mehta } 483fcb7a4a4SAarushi Mehta #endif 484fcb7a4a4SAarushi Mehta 485c2b38b27SPaolo Bonzini void aio_notify(AioContext *ctx) 486c2b38b27SPaolo Bonzini { 487601829f8SStefan Hajnoczi /* 4888dd48650SPaolo Bonzini * Write e.g. ctx->bh_list before writing ctx->notified. Pairs with 4898dd48650SPaolo Bonzini * smp_mb() in aio_notify_accept(). 490601829f8SStefan Hajnoczi */ 491601829f8SStefan Hajnoczi smp_wmb(); 492d73415a3SStefan Hajnoczi qatomic_set(&ctx->notified, true); 493601829f8SStefan Hajnoczi 494601829f8SStefan Hajnoczi /* 4958dd48650SPaolo Bonzini * Write ctx->notified (and also ctx->bh_list) before reading ctx->notify_me. 4968dd48650SPaolo Bonzini * Pairs with smp_mb() in aio_ctx_prepare or aio_poll. 497c2b38b27SPaolo Bonzini */ 498c2b38b27SPaolo Bonzini smp_mb(); 499d73415a3SStefan Hajnoczi if (qatomic_read(&ctx->notify_me)) { 500c2b38b27SPaolo Bonzini event_notifier_set(&ctx->notifier); 501c2b38b27SPaolo Bonzini } 502c2b38b27SPaolo Bonzini } 503c2b38b27SPaolo Bonzini 504c2b38b27SPaolo Bonzini void aio_notify_accept(AioContext *ctx) 505c2b38b27SPaolo Bonzini { 506d73415a3SStefan Hajnoczi qatomic_set(&ctx->notified, false); 507601829f8SStefan Hajnoczi 508601829f8SStefan Hajnoczi /* 5096229438cSPaolo Bonzini * Order reads of ctx->notified (in aio_context_notifier_poll()) and the 5106229438cSPaolo Bonzini * above clearing of ctx->notified before reads of e.g. bh->flags. Pairs 5116229438cSPaolo Bonzini * with smp_wmb() in aio_notify. 512601829f8SStefan Hajnoczi */ 513601829f8SStefan Hajnoczi smp_mb(); 514c2b38b27SPaolo Bonzini } 515c2b38b27SPaolo Bonzini 5163f53bc61SPaolo Bonzini static void aio_timerlist_notify(void *opaque, QEMUClockType type) 517c2b38b27SPaolo Bonzini { 518c2b38b27SPaolo Bonzini aio_notify(opaque); 519c2b38b27SPaolo Bonzini } 520c2b38b27SPaolo Bonzini 521601829f8SStefan Hajnoczi static void aio_context_notifier_cb(EventNotifier *e) 522c2b38b27SPaolo Bonzini { 523601829f8SStefan Hajnoczi AioContext *ctx = container_of(e, AioContext, notifier); 524601829f8SStefan Hajnoczi 525601829f8SStefan Hajnoczi event_notifier_test_and_clear(&ctx->notifier); 526c2b38b27SPaolo Bonzini } 527c2b38b27SPaolo Bonzini 528c2b38b27SPaolo Bonzini /* Returns true if aio_notify() was called (e.g. a BH was scheduled) */ 529c13be5a1SStefan Hajnoczi static bool aio_context_notifier_poll(void *opaque) 530c2b38b27SPaolo Bonzini { 531c2b38b27SPaolo Bonzini EventNotifier *e = opaque; 532c2b38b27SPaolo Bonzini AioContext *ctx = container_of(e, AioContext, notifier); 533c2b38b27SPaolo Bonzini 5346229438cSPaolo Bonzini /* 5356229438cSPaolo Bonzini * No need for load-acquire because we just want to kick the 5366229438cSPaolo Bonzini * event loop. aio_notify_accept() takes care of synchronizing 5376229438cSPaolo Bonzini * the event loop with the producers. 5386229438cSPaolo Bonzini */ 539d73415a3SStefan Hajnoczi return qatomic_read(&ctx->notified); 540c2b38b27SPaolo Bonzini } 541c2b38b27SPaolo Bonzini 542826cc324SStefan Hajnoczi static void aio_context_notifier_poll_ready(EventNotifier *e) 543826cc324SStefan Hajnoczi { 544826cc324SStefan Hajnoczi /* Do nothing, we just wanted to kick the event loop */ 545826cc324SStefan Hajnoczi } 546826cc324SStefan Hajnoczi 5470c330a73SPaolo Bonzini static void co_schedule_bh_cb(void *opaque) 5480c330a73SPaolo Bonzini { 5490c330a73SPaolo Bonzini AioContext *ctx = opaque; 5500c330a73SPaolo Bonzini QSLIST_HEAD(, Coroutine) straight, reversed; 5510c330a73SPaolo Bonzini 5520c330a73SPaolo Bonzini QSLIST_MOVE_ATOMIC(&reversed, &ctx->scheduled_coroutines); 5530c330a73SPaolo Bonzini QSLIST_INIT(&straight); 5540c330a73SPaolo Bonzini 5550c330a73SPaolo Bonzini while (!QSLIST_EMPTY(&reversed)) { 5560c330a73SPaolo Bonzini Coroutine *co = QSLIST_FIRST(&reversed); 5570c330a73SPaolo Bonzini QSLIST_REMOVE_HEAD(&reversed, co_scheduled_next); 5580c330a73SPaolo Bonzini QSLIST_INSERT_HEAD(&straight, co, co_scheduled_next); 5590c330a73SPaolo Bonzini } 5600c330a73SPaolo Bonzini 5610c330a73SPaolo Bonzini while (!QSLIST_EMPTY(&straight)) { 5620c330a73SPaolo Bonzini Coroutine *co = QSLIST_FIRST(&straight); 5630c330a73SPaolo Bonzini QSLIST_REMOVE_HEAD(&straight, co_scheduled_next); 5640c330a73SPaolo Bonzini trace_aio_co_schedule_bh_cb(ctx, co); 5651919631eSPaolo Bonzini aio_context_acquire(ctx); 5666133b39fSJeff Cody 5676133b39fSJeff Cody /* Protected by write barrier in qemu_aio_coroutine_enter */ 568d73415a3SStefan Hajnoczi qatomic_set(&co->scheduled, NULL); 5696808ae04SSergio Lopez qemu_aio_coroutine_enter(ctx, co); 5701919631eSPaolo Bonzini aio_context_release(ctx); 5710c330a73SPaolo Bonzini } 5720c330a73SPaolo Bonzini } 5730c330a73SPaolo Bonzini 574c2b38b27SPaolo Bonzini AioContext *aio_context_new(Error **errp) 575c2b38b27SPaolo Bonzini { 576c2b38b27SPaolo Bonzini int ret; 577c2b38b27SPaolo Bonzini AioContext *ctx; 578c2b38b27SPaolo Bonzini 579c2b38b27SPaolo Bonzini ctx = (AioContext *) g_source_new(&aio_source_funcs, sizeof(AioContext)); 5808c6b0356SStefan Hajnoczi QSLIST_INIT(&ctx->bh_list); 5818c6b0356SStefan Hajnoczi QSIMPLEQ_INIT(&ctx->bh_slice_list); 582c2b38b27SPaolo Bonzini aio_context_setup(ctx); 583c2b38b27SPaolo Bonzini 584c2b38b27SPaolo Bonzini ret = event_notifier_init(&ctx->notifier, false); 585c2b38b27SPaolo Bonzini if (ret < 0) { 586c2b38b27SPaolo Bonzini error_setg_errno(errp, -ret, "Failed to initialize event notifier"); 587c2b38b27SPaolo Bonzini goto fail; 588c2b38b27SPaolo Bonzini } 589c2b38b27SPaolo Bonzini g_source_set_can_recurse(&ctx->source, true); 590c2b38b27SPaolo Bonzini qemu_lockcnt_init(&ctx->list_lock); 5910c330a73SPaolo Bonzini 5920c330a73SPaolo Bonzini ctx->co_schedule_bh = aio_bh_new(ctx, co_schedule_bh_cb, ctx); 5930c330a73SPaolo Bonzini QSLIST_INIT(&ctx->scheduled_coroutines); 5940c330a73SPaolo Bonzini 595c2b38b27SPaolo Bonzini aio_set_event_notifier(ctx, &ctx->notifier, 596c2b38b27SPaolo Bonzini false, 597601829f8SStefan Hajnoczi aio_context_notifier_cb, 598826cc324SStefan Hajnoczi aio_context_notifier_poll, 599826cc324SStefan Hajnoczi aio_context_notifier_poll_ready); 600c2b38b27SPaolo Bonzini #ifdef CONFIG_LINUX_AIO 601c2b38b27SPaolo Bonzini ctx->linux_aio = NULL; 602c2b38b27SPaolo Bonzini #endif 603fcb7a4a4SAarushi Mehta 604fcb7a4a4SAarushi Mehta #ifdef CONFIG_LINUX_IO_URING 605fcb7a4a4SAarushi Mehta ctx->linux_io_uring = NULL; 606fcb7a4a4SAarushi Mehta #endif 607fcb7a4a4SAarushi Mehta 608c2b38b27SPaolo Bonzini ctx->thread_pool = NULL; 609c2b38b27SPaolo Bonzini qemu_rec_mutex_init(&ctx->lock); 610c2b38b27SPaolo Bonzini timerlistgroup_init(&ctx->tlg, aio_timerlist_notify, ctx); 611c2b38b27SPaolo Bonzini 612c2b38b27SPaolo Bonzini ctx->poll_ns = 0; 613c2b38b27SPaolo Bonzini ctx->poll_max_ns = 0; 614c2b38b27SPaolo Bonzini ctx->poll_grow = 0; 615c2b38b27SPaolo Bonzini ctx->poll_shrink = 0; 616c2b38b27SPaolo Bonzini 6171793ad02SStefano Garzarella ctx->aio_max_batch = 0; 6181793ad02SStefano Garzarella 61971ad4713SNicolas Saenz Julienne ctx->thread_pool_min = 0; 62071ad4713SNicolas Saenz Julienne ctx->thread_pool_max = THREAD_POOL_MAX_THREADS_DEFAULT; 62171ad4713SNicolas Saenz Julienne 622587d82faSEmanuele Giuseppe Esposito register_aiocontext(ctx); 623587d82faSEmanuele Giuseppe Esposito 624c2b38b27SPaolo Bonzini return ctx; 625c2b38b27SPaolo Bonzini fail: 626c2b38b27SPaolo Bonzini g_source_destroy(&ctx->source); 627c2b38b27SPaolo Bonzini return NULL; 628c2b38b27SPaolo Bonzini } 629c2b38b27SPaolo Bonzini 6300c330a73SPaolo Bonzini void aio_co_schedule(AioContext *ctx, Coroutine *co) 6310c330a73SPaolo Bonzini { 6320c330a73SPaolo Bonzini trace_aio_co_schedule(ctx, co); 633d73415a3SStefan Hajnoczi const char *scheduled = qatomic_cmpxchg(&co->scheduled, NULL, 6346133b39fSJeff Cody __func__); 6356133b39fSJeff Cody 6366133b39fSJeff Cody if (scheduled) { 6376133b39fSJeff Cody fprintf(stderr, 6386133b39fSJeff Cody "%s: Co-routine was already scheduled in '%s'\n", 6396133b39fSJeff Cody __func__, scheduled); 6406133b39fSJeff Cody abort(); 6416133b39fSJeff Cody } 6426133b39fSJeff Cody 643f0f81002SStefan Hajnoczi /* The coroutine might run and release the last ctx reference before we 644f0f81002SStefan Hajnoczi * invoke qemu_bh_schedule(). Take a reference to keep ctx alive until 645f0f81002SStefan Hajnoczi * we're done. 646f0f81002SStefan Hajnoczi */ 647f0f81002SStefan Hajnoczi aio_context_ref(ctx); 648f0f81002SStefan Hajnoczi 6490c330a73SPaolo Bonzini QSLIST_INSERT_HEAD_ATOMIC(&ctx->scheduled_coroutines, 6500c330a73SPaolo Bonzini co, co_scheduled_next); 6510c330a73SPaolo Bonzini qemu_bh_schedule(ctx->co_schedule_bh); 652f0f81002SStefan Hajnoczi 653f0f81002SStefan Hajnoczi aio_context_unref(ctx); 6540c330a73SPaolo Bonzini } 6550c330a73SPaolo Bonzini 65626b0b698SKevin Wolf typedef struct AioCoRescheduleSelf { 65726b0b698SKevin Wolf Coroutine *co; 65826b0b698SKevin Wolf AioContext *new_ctx; 65926b0b698SKevin Wolf } AioCoRescheduleSelf; 66026b0b698SKevin Wolf 66126b0b698SKevin Wolf static void aio_co_reschedule_self_bh(void *opaque) 66226b0b698SKevin Wolf { 66326b0b698SKevin Wolf AioCoRescheduleSelf *data = opaque; 66426b0b698SKevin Wolf aio_co_schedule(data->new_ctx, data->co); 66526b0b698SKevin Wolf } 66626b0b698SKevin Wolf 66726b0b698SKevin Wolf void coroutine_fn aio_co_reschedule_self(AioContext *new_ctx) 66826b0b698SKevin Wolf { 66926b0b698SKevin Wolf AioContext *old_ctx = qemu_get_current_aio_context(); 67026b0b698SKevin Wolf 67126b0b698SKevin Wolf if (old_ctx != new_ctx) { 67226b0b698SKevin Wolf AioCoRescheduleSelf data = { 67326b0b698SKevin Wolf .co = qemu_coroutine_self(), 67426b0b698SKevin Wolf .new_ctx = new_ctx, 67526b0b698SKevin Wolf }; 67626b0b698SKevin Wolf /* 67726b0b698SKevin Wolf * We can't directly schedule the coroutine in the target context 67826b0b698SKevin Wolf * because this would be racy: The other thread could try to enter the 67926b0b698SKevin Wolf * coroutine before it has yielded in this one. 68026b0b698SKevin Wolf */ 68126b0b698SKevin Wolf aio_bh_schedule_oneshot(old_ctx, aio_co_reschedule_self_bh, &data); 68226b0b698SKevin Wolf qemu_coroutine_yield(); 68326b0b698SKevin Wolf } 68426b0b698SKevin Wolf } 68526b0b698SKevin Wolf 68643695601SMarkus Armbruster void aio_co_wake(Coroutine *co) 6870c330a73SPaolo Bonzini { 6880c330a73SPaolo Bonzini AioContext *ctx; 6890c330a73SPaolo Bonzini 6900c330a73SPaolo Bonzini /* Read coroutine before co->ctx. Matches smp_wmb in 6910c330a73SPaolo Bonzini * qemu_coroutine_enter. 6920c330a73SPaolo Bonzini */ 6930c330a73SPaolo Bonzini smp_read_barrier_depends(); 694d73415a3SStefan Hajnoczi ctx = qatomic_read(&co->ctx); 6950c330a73SPaolo Bonzini 6968865852eSFam Zheng aio_co_enter(ctx, co); 6978865852eSFam Zheng } 6988865852eSFam Zheng 69943695601SMarkus Armbruster void aio_co_enter(AioContext *ctx, Coroutine *co) 7008865852eSFam Zheng { 7010c330a73SPaolo Bonzini if (ctx != qemu_get_current_aio_context()) { 7020c330a73SPaolo Bonzini aio_co_schedule(ctx, co); 7030c330a73SPaolo Bonzini return; 7040c330a73SPaolo Bonzini } 7050c330a73SPaolo Bonzini 7060c330a73SPaolo Bonzini if (qemu_in_coroutine()) { 7070c330a73SPaolo Bonzini Coroutine *self = qemu_coroutine_self(); 7080c330a73SPaolo Bonzini assert(self != co); 7090c330a73SPaolo Bonzini QSIMPLEQ_INSERT_TAIL(&self->co_queue_wakeup, co, co_queue_next); 7100c330a73SPaolo Bonzini } else { 7110c330a73SPaolo Bonzini aio_context_acquire(ctx); 7128865852eSFam Zheng qemu_aio_coroutine_enter(ctx, co); 7130c330a73SPaolo Bonzini aio_context_release(ctx); 7140c330a73SPaolo Bonzini } 7150c330a73SPaolo Bonzini } 7160c330a73SPaolo Bonzini 717c2b38b27SPaolo Bonzini void aio_context_ref(AioContext *ctx) 718c2b38b27SPaolo Bonzini { 719c2b38b27SPaolo Bonzini g_source_ref(&ctx->source); 720c2b38b27SPaolo Bonzini } 721c2b38b27SPaolo Bonzini 722c2b38b27SPaolo Bonzini void aio_context_unref(AioContext *ctx) 723c2b38b27SPaolo Bonzini { 724c2b38b27SPaolo Bonzini g_source_unref(&ctx->source); 725c2b38b27SPaolo Bonzini } 726c2b38b27SPaolo Bonzini 727c2b38b27SPaolo Bonzini void aio_context_acquire(AioContext *ctx) 728c2b38b27SPaolo Bonzini { 729c2b38b27SPaolo Bonzini qemu_rec_mutex_lock(&ctx->lock); 730c2b38b27SPaolo Bonzini } 731c2b38b27SPaolo Bonzini 732c2b38b27SPaolo Bonzini void aio_context_release(AioContext *ctx) 733c2b38b27SPaolo Bonzini { 734c2b38b27SPaolo Bonzini qemu_rec_mutex_unlock(&ctx->lock); 735c2b38b27SPaolo Bonzini } 7365f50be9bSPaolo Bonzini 73747b74464SStefan Hajnoczi QEMU_DEFINE_STATIC_CO_TLS(AioContext *, my_aiocontext) 7385f50be9bSPaolo Bonzini 7395f50be9bSPaolo Bonzini AioContext *qemu_get_current_aio_context(void) 7405f50be9bSPaolo Bonzini { 74147b74464SStefan Hajnoczi AioContext *ctx = get_my_aiocontext(); 74247b74464SStefan Hajnoczi if (ctx) { 74347b74464SStefan Hajnoczi return ctx; 7445f50be9bSPaolo Bonzini } 7455f50be9bSPaolo Bonzini if (qemu_mutex_iothread_locked()) { 7465f50be9bSPaolo Bonzini /* Possibly in a vCPU thread. */ 7475f50be9bSPaolo Bonzini return qemu_get_aio_context(); 7485f50be9bSPaolo Bonzini } 7495f50be9bSPaolo Bonzini return NULL; 7505f50be9bSPaolo Bonzini } 7515f50be9bSPaolo Bonzini 7525f50be9bSPaolo Bonzini void qemu_set_current_aio_context(AioContext *ctx) 7535f50be9bSPaolo Bonzini { 75447b74464SStefan Hajnoczi assert(!get_my_aiocontext()); 75547b74464SStefan Hajnoczi set_my_aiocontext(ctx); 7565f50be9bSPaolo Bonzini } 75771ad4713SNicolas Saenz Julienne 75871ad4713SNicolas Saenz Julienne void aio_context_set_thread_pool_params(AioContext *ctx, int64_t min, 75971ad4713SNicolas Saenz Julienne int64_t max, Error **errp) 76071ad4713SNicolas Saenz Julienne { 76171ad4713SNicolas Saenz Julienne 76271ad4713SNicolas Saenz Julienne if (min > max || !max || min > INT_MAX || max > INT_MAX) { 76371ad4713SNicolas Saenz Julienne error_setg(errp, "bad thread-pool-min/thread-pool-max values"); 76471ad4713SNicolas Saenz Julienne return; 76571ad4713SNicolas Saenz Julienne } 76671ad4713SNicolas Saenz Julienne 76771ad4713SNicolas Saenz Julienne ctx->thread_pool_min = min; 76871ad4713SNicolas Saenz Julienne ctx->thread_pool_max = max; 76971ad4713SNicolas Saenz Julienne 77071ad4713SNicolas Saenz Julienne if (ctx->thread_pool) { 77171ad4713SNicolas Saenz Julienne thread_pool_update_params(ctx->thread_pool, ctx); 77271ad4713SNicolas Saenz Julienne } 77371ad4713SNicolas Saenz Julienne } 774