xref: /qemu/util/thread-pool.c (revision 6402cbbb)
1 /*
2  * QEMU block layer thread pool
3  *
4  * Copyright IBM, Corp. 2008
5  * Copyright Red Hat, Inc. 2012
6  *
7  * Authors:
8  *  Anthony Liguori   <aliguori@us.ibm.com>
9  *  Paolo Bonzini     <pbonzini@redhat.com>
10  *
11  * This work is licensed under the terms of the GNU GPL, version 2.  See
12  * the COPYING file in the top-level directory.
13  *
14  * Contributions after 2012-01-13 are licensed under the terms of the
15  * GNU GPL, version 2 or (at your option) any later version.
16  */
17 #include "qemu/osdep.h"
18 #include "qemu-common.h"
19 #include "qemu/queue.h"
20 #include "qemu/thread.h"
21 #include "qemu/coroutine.h"
22 #include "trace.h"
23 #include "block/thread-pool.h"
24 #include "qemu/main-loop.h"
25 
26 static void do_spawn_thread(ThreadPool *pool);
27 
28 typedef struct ThreadPoolElement ThreadPoolElement;
29 
30 enum ThreadState {
31     THREAD_QUEUED,
32     THREAD_ACTIVE,
33     THREAD_DONE,
34 };
35 
36 struct ThreadPoolElement {
37     BlockAIOCB common;
38     ThreadPool *pool;
39     ThreadPoolFunc *func;
40     void *arg;
41 
42     /* Moving state out of THREAD_QUEUED is protected by lock.  After
43      * that, only the worker thread can write to it.  Reads and writes
44      * of state and ret are ordered with memory barriers.
45      */
46     enum ThreadState state;
47     int ret;
48 
49     /* Access to this list is protected by lock.  */
50     QTAILQ_ENTRY(ThreadPoolElement) reqs;
51 
52     /* Access to this list is protected by the global mutex.  */
53     QLIST_ENTRY(ThreadPoolElement) all;
54 };
55 
56 struct ThreadPool {
57     AioContext *ctx;
58     QEMUBH *completion_bh;
59     QemuMutex lock;
60     QemuCond worker_stopped;
61     QemuSemaphore sem;
62     int max_threads;
63     QEMUBH *new_thread_bh;
64 
65     /* The following variables are only accessed from one AioContext. */
66     QLIST_HEAD(, ThreadPoolElement) head;
67 
68     /* The following variables are protected by lock.  */
69     QTAILQ_HEAD(, ThreadPoolElement) request_list;
70     int cur_threads;
71     int idle_threads;
72     int new_threads;     /* backlog of threads we need to create */
73     int pending_threads; /* threads created but not running yet */
74     bool stopping;
75 };
76 
77 static void *worker_thread(void *opaque)
78 {
79     ThreadPool *pool = opaque;
80 
81     qemu_mutex_lock(&pool->lock);
82     pool->pending_threads--;
83     do_spawn_thread(pool);
84 
85     while (!pool->stopping) {
86         ThreadPoolElement *req;
87         int ret;
88 
89         do {
90             pool->idle_threads++;
91             qemu_mutex_unlock(&pool->lock);
92             ret = qemu_sem_timedwait(&pool->sem, 10000);
93             qemu_mutex_lock(&pool->lock);
94             pool->idle_threads--;
95         } while (ret == -1 && !QTAILQ_EMPTY(&pool->request_list));
96         if (ret == -1 || pool->stopping) {
97             break;
98         }
99 
100         req = QTAILQ_FIRST(&pool->request_list);
101         QTAILQ_REMOVE(&pool->request_list, req, reqs);
102         req->state = THREAD_ACTIVE;
103         qemu_mutex_unlock(&pool->lock);
104 
105         ret = req->func(req->arg);
106 
107         req->ret = ret;
108         /* Write ret before state.  */
109         smp_wmb();
110         req->state = THREAD_DONE;
111 
112         qemu_mutex_lock(&pool->lock);
113 
114         qemu_bh_schedule(pool->completion_bh);
115     }
116 
117     pool->cur_threads--;
118     qemu_cond_signal(&pool->worker_stopped);
119     qemu_mutex_unlock(&pool->lock);
120     return NULL;
121 }
122 
123 static void do_spawn_thread(ThreadPool *pool)
124 {
125     QemuThread t;
126 
127     /* Runs with lock taken.  */
128     if (!pool->new_threads) {
129         return;
130     }
131 
132     pool->new_threads--;
133     pool->pending_threads++;
134 
135     qemu_thread_create(&t, "worker", worker_thread, pool, QEMU_THREAD_DETACHED);
136 }
137 
138 static void spawn_thread_bh_fn(void *opaque)
139 {
140     ThreadPool *pool = opaque;
141 
142     qemu_mutex_lock(&pool->lock);
143     do_spawn_thread(pool);
144     qemu_mutex_unlock(&pool->lock);
145 }
146 
147 static void spawn_thread(ThreadPool *pool)
148 {
149     pool->cur_threads++;
150     pool->new_threads++;
151     /* If there are threads being created, they will spawn new workers, so
152      * we don't spend time creating many threads in a loop holding a mutex or
153      * starving the current vcpu.
154      *
155      * If there are no idle threads, ask the main thread to create one, so we
156      * inherit the correct affinity instead of the vcpu affinity.
157      */
158     if (!pool->pending_threads) {
159         qemu_bh_schedule(pool->new_thread_bh);
160     }
161 }
162 
163 static void thread_pool_completion_bh(void *opaque)
164 {
165     ThreadPool *pool = opaque;
166     ThreadPoolElement *elem, *next;
167 
168     aio_context_acquire(pool->ctx);
169 restart:
170     QLIST_FOREACH_SAFE(elem, &pool->head, all, next) {
171         if (elem->state != THREAD_DONE) {
172             continue;
173         }
174 
175         trace_thread_pool_complete(pool, elem, elem->common.opaque,
176                                    elem->ret);
177         QLIST_REMOVE(elem, all);
178 
179         if (elem->common.cb) {
180             /* Read state before ret.  */
181             smp_rmb();
182 
183             /* Schedule ourselves in case elem->common.cb() calls aio_poll() to
184              * wait for another request that completed at the same time.
185              */
186             qemu_bh_schedule(pool->completion_bh);
187 
188             aio_context_release(pool->ctx);
189             elem->common.cb(elem->common.opaque, elem->ret);
190             aio_context_acquire(pool->ctx);
191 
192             /* We can safely cancel the completion_bh here regardless of someone
193              * else having scheduled it meanwhile because we reenter the
194              * completion function anyway (goto restart).
195              */
196             qemu_bh_cancel(pool->completion_bh);
197 
198             qemu_aio_unref(elem);
199             goto restart;
200         } else {
201             qemu_aio_unref(elem);
202         }
203     }
204     aio_context_release(pool->ctx);
205 }
206 
207 static void thread_pool_cancel(BlockAIOCB *acb)
208 {
209     ThreadPoolElement *elem = (ThreadPoolElement *)acb;
210     ThreadPool *pool = elem->pool;
211 
212     trace_thread_pool_cancel(elem, elem->common.opaque);
213 
214     qemu_mutex_lock(&pool->lock);
215     if (elem->state == THREAD_QUEUED &&
216         /* No thread has yet started working on elem. we can try to "steal"
217          * the item from the worker if we can get a signal from the
218          * semaphore.  Because this is non-blocking, we can do it with
219          * the lock taken and ensure that elem will remain THREAD_QUEUED.
220          */
221         qemu_sem_timedwait(&pool->sem, 0) == 0) {
222         QTAILQ_REMOVE(&pool->request_list, elem, reqs);
223         qemu_bh_schedule(pool->completion_bh);
224 
225         elem->state = THREAD_DONE;
226         elem->ret = -ECANCELED;
227     }
228 
229     qemu_mutex_unlock(&pool->lock);
230 }
231 
232 static AioContext *thread_pool_get_aio_context(BlockAIOCB *acb)
233 {
234     ThreadPoolElement *elem = (ThreadPoolElement *)acb;
235     ThreadPool *pool = elem->pool;
236     return pool->ctx;
237 }
238 
239 static const AIOCBInfo thread_pool_aiocb_info = {
240     .aiocb_size         = sizeof(ThreadPoolElement),
241     .cancel_async       = thread_pool_cancel,
242     .get_aio_context    = thread_pool_get_aio_context,
243 };
244 
245 BlockAIOCB *thread_pool_submit_aio(ThreadPool *pool,
246         ThreadPoolFunc *func, void *arg,
247         BlockCompletionFunc *cb, void *opaque)
248 {
249     ThreadPoolElement *req;
250 
251     req = qemu_aio_get(&thread_pool_aiocb_info, NULL, cb, opaque);
252     req->func = func;
253     req->arg = arg;
254     req->state = THREAD_QUEUED;
255     req->pool = pool;
256 
257     QLIST_INSERT_HEAD(&pool->head, req, all);
258 
259     trace_thread_pool_submit(pool, req, arg);
260 
261     qemu_mutex_lock(&pool->lock);
262     if (pool->idle_threads == 0 && pool->cur_threads < pool->max_threads) {
263         spawn_thread(pool);
264     }
265     QTAILQ_INSERT_TAIL(&pool->request_list, req, reqs);
266     qemu_mutex_unlock(&pool->lock);
267     qemu_sem_post(&pool->sem);
268     return &req->common;
269 }
270 
271 typedef struct ThreadPoolCo {
272     Coroutine *co;
273     int ret;
274 } ThreadPoolCo;
275 
276 static void thread_pool_co_cb(void *opaque, int ret)
277 {
278     ThreadPoolCo *co = opaque;
279 
280     co->ret = ret;
281     aio_co_wake(co->co);
282 }
283 
284 int coroutine_fn thread_pool_submit_co(ThreadPool *pool, ThreadPoolFunc *func,
285                                        void *arg)
286 {
287     ThreadPoolCo tpc = { .co = qemu_coroutine_self(), .ret = -EINPROGRESS };
288     assert(qemu_in_coroutine());
289     thread_pool_submit_aio(pool, func, arg, thread_pool_co_cb, &tpc);
290     qemu_coroutine_yield();
291     return tpc.ret;
292 }
293 
294 void thread_pool_submit(ThreadPool *pool, ThreadPoolFunc *func, void *arg)
295 {
296     thread_pool_submit_aio(pool, func, arg, NULL, NULL);
297 }
298 
299 static void thread_pool_init_one(ThreadPool *pool, AioContext *ctx)
300 {
301     if (!ctx) {
302         ctx = qemu_get_aio_context();
303     }
304 
305     memset(pool, 0, sizeof(*pool));
306     pool->ctx = ctx;
307     pool->completion_bh = aio_bh_new(ctx, thread_pool_completion_bh, pool);
308     qemu_mutex_init(&pool->lock);
309     qemu_cond_init(&pool->worker_stopped);
310     qemu_sem_init(&pool->sem, 0);
311     pool->max_threads = 64;
312     pool->new_thread_bh = aio_bh_new(ctx, spawn_thread_bh_fn, pool);
313 
314     QLIST_INIT(&pool->head);
315     QTAILQ_INIT(&pool->request_list);
316 }
317 
318 ThreadPool *thread_pool_new(AioContext *ctx)
319 {
320     ThreadPool *pool = g_new(ThreadPool, 1);
321     thread_pool_init_one(pool, ctx);
322     return pool;
323 }
324 
325 void thread_pool_free(ThreadPool *pool)
326 {
327     if (!pool) {
328         return;
329     }
330 
331     assert(QLIST_EMPTY(&pool->head));
332 
333     qemu_mutex_lock(&pool->lock);
334 
335     /* Stop new threads from spawning */
336     qemu_bh_delete(pool->new_thread_bh);
337     pool->cur_threads -= pool->new_threads;
338     pool->new_threads = 0;
339 
340     /* Wait for worker threads to terminate */
341     pool->stopping = true;
342     while (pool->cur_threads > 0) {
343         qemu_sem_post(&pool->sem);
344         qemu_cond_wait(&pool->worker_stopped, &pool->lock);
345     }
346 
347     qemu_mutex_unlock(&pool->lock);
348 
349     qemu_bh_delete(pool->completion_bh);
350     qemu_sem_destroy(&pool->sem);
351     qemu_cond_destroy(&pool->worker_stopped);
352     qemu_mutex_destroy(&pool->lock);
353     g_free(pool);
354 }
355