summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorJens Axboe <axboe@kernel.dk>2023-12-13 08:58:15 -0700
committerJens Axboe <axboe@kernel.dk>2023-12-13 08:58:15 -0700
commit595e52284d24adc376890d3fc93bdca4707d9aca (patch)
tree7648aeef9e7ca87c0e2b9be537cf68dee084478d
parent705318a99a138c29a512a72c3e0043b3cd7f55f4 (diff)
io_uring/poll: don't enable lazy wake for POLLEXCLUSIVE
There are a few quirks around using lazy wake for poll unconditionally, and one of them is related the EPOLLEXCLUSIVE. Those may trigger exclusive wakeups, which wake a limited number of entries in the wait queue. If that wake number is less than the number of entries someone is waiting for (and that someone is also using DEFER_TASKRUN), then we can get stuck waiting for more entries while we should be processing the ones we already got. If we're doing exclusive poll waits, flag the request as not being compatible with lazy wakeups. Reported-by: Pavel Begunkov <asml.silence@gmail.com> Fixes: 6ce4a93dbb5b ("io_uring/poll: use IOU_F_TWQ_LAZY_WAKE for wakeups") Signed-off-by: Jens Axboe <axboe@kernel.dk>
-rw-r--r--include/linux/io_uring_types.h3
-rw-r--r--io_uring/poll.c20
2 files changed, 20 insertions, 3 deletions
diff --git a/include/linux/io_uring_types.h b/include/linux/io_uring_types.h
index 805bb635cdf5..239a4f68801b 100644
--- a/include/linux/io_uring_types.h
+++ b/include/linux/io_uring_types.h
@@ -434,6 +434,7 @@ enum {
/* keep async read/write and isreg together and in order */
REQ_F_SUPPORT_NOWAIT_BIT,
REQ_F_ISREG_BIT,
+ REQ_F_POLL_NO_LAZY_BIT,
/* not a real bit, just to check we're not overflowing the space */
__REQ_F_LAST_BIT,
@@ -501,6 +502,8 @@ enum {
REQ_F_CLEAR_POLLIN = BIT(REQ_F_CLEAR_POLLIN_BIT),
/* hashed into ->cancel_hash_locked, protected by ->uring_lock */
REQ_F_HASH_LOCKED = BIT(REQ_F_HASH_LOCKED_BIT),
+ /* don't use lazy poll wake for this request */
+ REQ_F_POLL_NO_LAZY = BIT(REQ_F_POLL_NO_LAZY_BIT),
};
typedef void (*io_req_tw_func_t)(struct io_kiocb *req, struct io_tw_state *ts);
diff --git a/io_uring/poll.c b/io_uring/poll.c
index d38d05edb4fa..d59b74a99d4e 100644
--- a/io_uring/poll.c
+++ b/io_uring/poll.c
@@ -366,11 +366,16 @@ void io_poll_task_func(struct io_kiocb *req, struct io_tw_state *ts)
static void __io_poll_execute(struct io_kiocb *req, int mask)
{
+ unsigned flags = 0;
+
io_req_set_res(req, mask, 0);
req->io_task_work.func = io_poll_task_func;
trace_io_uring_task_add(req, mask);
- __io_req_task_work_add(req, IOU_F_TWQ_LAZY_WAKE);
+
+ if (!(req->flags & REQ_F_POLL_NO_LAZY))
+ flags = IOU_F_TWQ_LAZY_WAKE;
+ __io_req_task_work_add(req, flags);
}
static inline void io_poll_execute(struct io_kiocb *req, int res)
@@ -526,10 +531,19 @@ static void __io_queue_proc(struct io_poll *poll, struct io_poll_table *pt,
poll->head = head;
poll->wait.private = (void *) wqe_private;
- if (poll->events & EPOLLEXCLUSIVE)
+ if (poll->events & EPOLLEXCLUSIVE) {
+ /*
+ * Exclusive waits may only wake a limited amount of entries
+ * rather than all of them, this may interfere with lazy
+ * wake if someone does wait(events > 1). Ensure we don't do
+ * lazy wake for those, as we need to process each one as they
+ * come in.
+ */
+ req->flags |= REQ_F_POLL_NO_LAZY;
add_wait_queue_exclusive(head, &poll->wait);
- else
+ } else {
add_wait_queue(head, &poll->wait);
+ }
}
static void io_poll_queue_proc(struct file *file, struct wait_queue_head *head,