]> git.baikalelectronics.ru Git - kernel.git/commitdiff
io_uring: use private ctx wait queue entries for SQPOLL
authorJens Axboe <axboe@kernel.dk>
Wed, 2 Sep 2020 18:21:41 +0000 (12:21 -0600)
committerJens Axboe <axboe@kernel.dk>
Thu, 1 Oct 2020 02:32:33 +0000 (20:32 -0600)
This is in preparation to sharing the poller thread between rings. For
that we need per-ring wait_queue_entry storage, and we can't easily put
that on the stack if one thread is managing multiple rings.

We'll also be sharing the wait_queue_head across rings for the purposes
of wakeups, provide the usual private ring wait_queue_head for now but
make it a pointer so we can easily override it when sharing.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
fs/io_uring.c

index 26f819ae177d7f95d30f551f679822ad751c51c2..128ffa79d9d38d3c4b17150a4b79111d082fabb5 100644 (file)
@@ -287,7 +287,9 @@ struct io_ring_ctx {
        /* Only used for accounting purposes */
        struct mm_struct        *mm_account;
 
-       wait_queue_head_t       sqo_wait;
+       struct wait_queue_head  *sqo_wait;
+       struct wait_queue_head  __sqo_wait;
+       struct wait_queue_entry sqo_wait_entry;
 
        /*
         * If used, fixed file set. Writers must ensure that ->refs is dead,
@@ -1057,7 +1059,8 @@ static struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p)
                goto err;
 
        ctx->flags = p->flags;
-       init_waitqueue_head(&ctx->sqo_wait);
+       init_waitqueue_head(&ctx->__sqo_wait);
+       ctx->sqo_wait = &ctx->__sqo_wait;
        init_waitqueue_head(&ctx->cq_wait);
        INIT_LIST_HEAD(&ctx->cq_overflow_list);
        init_completion(&ctx->ref_comp);
@@ -1340,8 +1343,8 @@ static void io_cqring_ev_posted(struct io_ring_ctx *ctx)
 {
        if (waitqueue_active(&ctx->wait))
                wake_up(&ctx->wait);
-       if (waitqueue_active(&ctx->sqo_wait))
-               wake_up(&ctx->sqo_wait);
+       if (waitqueue_active(ctx->sqo_wait))
+               wake_up(ctx->sqo_wait);
        if (io_should_trigger_evfd(ctx))
                eventfd_signal(ctx->cq_ev_fd, 1);
 }
@@ -2448,9 +2451,8 @@ static void io_iopoll_req_issued(struct io_kiocb *req)
        else
                list_add_tail(&req->inflight_entry, &ctx->iopoll_list);
 
-       if ((ctx->flags & IORING_SETUP_SQPOLL) &&
-           wq_has_sleeper(&ctx->sqo_wait))
-               wake_up(&ctx->sqo_wait);
+       if ((ctx->flags & IORING_SETUP_SQPOLL) && wq_has_sleeper(ctx->sqo_wait))
+               wake_up(ctx->sqo_wait);
 }
 
 static void __io_state_file_put(struct io_submit_state *state)
@@ -6627,10 +6629,11 @@ static int io_sq_thread(void *data)
 {
        struct io_ring_ctx *ctx = data;
        const struct cred *old_cred;
-       DEFINE_WAIT(wait);
        unsigned long timeout;
        int ret = 0;
 
+       init_wait(&ctx->sqo_wait_entry);
+
        complete(&ctx->sq_thread_comp);
 
        old_cred = override_creds(ctx->creds);
@@ -6680,7 +6683,7 @@ static int io_sq_thread(void *data)
                                continue;
                        }
 
-                       prepare_to_wait(&ctx->sqo_wait, &wait,
+                       prepare_to_wait(ctx->sqo_wait, &ctx->sqo_wait_entry,
                                                TASK_INTERRUPTIBLE);
 
                        /*
@@ -6692,7 +6695,7 @@ static int io_sq_thread(void *data)
                         */
                        if ((ctx->flags & IORING_SETUP_IOPOLL) &&
                            !list_empty_careful(&ctx->iopoll_list)) {
-                               finish_wait(&ctx->sqo_wait, &wait);
+                               finish_wait(ctx->sqo_wait, &ctx->sqo_wait_entry);
                                continue;
                        }
 
@@ -6701,22 +6704,22 @@ static int io_sq_thread(void *data)
                        to_submit = io_sqring_entries(ctx);
                        if (!to_submit || ret == -EBUSY) {
                                if (kthread_should_park()) {
-                                       finish_wait(&ctx->sqo_wait, &wait);
+                                       finish_wait(ctx->sqo_wait, &ctx->sqo_wait_entry);
                                        break;
                                }
                                if (io_run_task_work()) {
-                                       finish_wait(&ctx->sqo_wait, &wait);
+                                       finish_wait(ctx->sqo_wait, &ctx->sqo_wait_entry);
                                        io_ring_clear_wakeup_flag(ctx);
                                        continue;
                                }
                                schedule();
-                               finish_wait(&ctx->sqo_wait, &wait);
+                               finish_wait(ctx->sqo_wait, &ctx->sqo_wait_entry);
 
                                io_ring_clear_wakeup_flag(ctx);
                                ret = 0;
                                continue;
                        }
-                       finish_wait(&ctx->sqo_wait, &wait);
+                       finish_wait(ctx->sqo_wait, &ctx->sqo_wait_entry);
 
                        io_ring_clear_wakeup_flag(ctx);
                }
@@ -8659,7 +8662,7 @@ SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit,
                if (!list_empty_careful(&ctx->cq_overflow_list))
                        io_cqring_overflow_flush(ctx, false, NULL, NULL);
                if (flags & IORING_ENTER_SQ_WAKEUP)
-                       wake_up(&ctx->sqo_wait);
+                       wake_up(ctx->sqo_wait);
                submitted = to_submit;
        } else if (to_submit) {
                ret = io_uring_add_task_file(f.file);