1

io_uring: remove unconditional looping in local task_work handling

If we have a ton of notifications coming in, we can be looping in here
for a long time. This can be problematic for various reasons, mostly
because we can starve userspace. If the application is waiting on N
events, then only re-run if we need more events.

Fixes: c0e0d6ba25 ("io_uring: add IORING_SETUP_DEFER_TASKRUN")
Signed-off-by: Jens Axboe <axboe@kernel.dk>
This commit is contained in:
Jens Axboe 2024-01-31 10:50:08 -07:00
parent 670d9d3df8
commit 9fe3eaea4a

View File

@ -1386,7 +1386,20 @@ static void __cold io_move_task_work_from_local(struct io_ring_ctx *ctx)
} }
} }
static int __io_run_local_work(struct io_ring_ctx *ctx, struct io_tw_state *ts) static bool io_run_local_work_continue(struct io_ring_ctx *ctx, int events,
int min_events)
{
if (llist_empty(&ctx->work_llist))
return false;
if (events < min_events)
return true;
if (ctx->flags & IORING_SETUP_TASKRUN_FLAG)
atomic_or(IORING_SQ_TASKRUN, &ctx->rings->sq_flags);
return false;
}
static int __io_run_local_work(struct io_ring_ctx *ctx, struct io_tw_state *ts,
int min_events)
{ {
struct llist_node *node; struct llist_node *node;
unsigned int loops = 0; unsigned int loops = 0;
@ -1414,18 +1427,20 @@ again:
} }
loops++; loops++;
if (!llist_empty(&ctx->work_llist)) if (io_run_local_work_continue(ctx, ret, min_events))
goto again; goto again;
if (ts->locked) { if (ts->locked) {
io_submit_flush_completions(ctx); io_submit_flush_completions(ctx);
if (!llist_empty(&ctx->work_llist)) if (io_run_local_work_continue(ctx, ret, min_events))
goto again; goto again;
} }
trace_io_uring_local_work_run(ctx, ret, loops); trace_io_uring_local_work_run(ctx, ret, loops);
return ret; return ret;
} }
static inline int io_run_local_work_locked(struct io_ring_ctx *ctx) static inline int io_run_local_work_locked(struct io_ring_ctx *ctx,
int min_events)
{ {
struct io_tw_state ts = { .locked = true, }; struct io_tw_state ts = { .locked = true, };
int ret; int ret;
@ -1433,20 +1448,20 @@ static inline int io_run_local_work_locked(struct io_ring_ctx *ctx)
if (llist_empty(&ctx->work_llist)) if (llist_empty(&ctx->work_llist))
return 0; return 0;
ret = __io_run_local_work(ctx, &ts); ret = __io_run_local_work(ctx, &ts, min_events);
/* shouldn't happen! */ /* shouldn't happen! */
if (WARN_ON_ONCE(!ts.locked)) if (WARN_ON_ONCE(!ts.locked))
mutex_lock(&ctx->uring_lock); mutex_lock(&ctx->uring_lock);
return ret; return ret;
} }
static int io_run_local_work(struct io_ring_ctx *ctx) static int io_run_local_work(struct io_ring_ctx *ctx, int min_events)
{ {
struct io_tw_state ts = {}; struct io_tw_state ts = {};
int ret; int ret;
ts.locked = mutex_trylock(&ctx->uring_lock); ts.locked = mutex_trylock(&ctx->uring_lock);
ret = __io_run_local_work(ctx, &ts); ret = __io_run_local_work(ctx, &ts, min_events);
if (ts.locked) if (ts.locked)
mutex_unlock(&ctx->uring_lock); mutex_unlock(&ctx->uring_lock);
@ -1642,7 +1657,7 @@ static int io_iopoll_check(struct io_ring_ctx *ctx, long min)
io_task_work_pending(ctx)) { io_task_work_pending(ctx)) {
u32 tail = ctx->cached_cq_tail; u32 tail = ctx->cached_cq_tail;
(void) io_run_local_work_locked(ctx); (void) io_run_local_work_locked(ctx, min);
if (task_work_pending(current) || if (task_work_pending(current) ||
wq_list_empty(&ctx->iopoll_list)) { wq_list_empty(&ctx->iopoll_list)) {
@ -2486,7 +2501,7 @@ int io_run_task_work_sig(struct io_ring_ctx *ctx)
{ {
if (!llist_empty(&ctx->work_llist)) { if (!llist_empty(&ctx->work_llist)) {
__set_current_state(TASK_RUNNING); __set_current_state(TASK_RUNNING);
if (io_run_local_work(ctx) > 0) if (io_run_local_work(ctx, INT_MAX) > 0)
return 0; return 0;
} }
if (io_run_task_work() > 0) if (io_run_task_work() > 0)
@ -2554,7 +2569,7 @@ static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events,
if (!io_allowed_run_tw(ctx)) if (!io_allowed_run_tw(ctx))
return -EEXIST; return -EEXIST;
if (!llist_empty(&ctx->work_llist)) if (!llist_empty(&ctx->work_llist))
io_run_local_work(ctx); io_run_local_work(ctx, min_events);
io_run_task_work(); io_run_task_work();
io_cqring_overflow_flush(ctx); io_cqring_overflow_flush(ctx);
/* if user messes with these they will just get an early return */ /* if user messes with these they will just get an early return */
@ -2592,11 +2607,10 @@ static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events,
trace_io_uring_cqring_wait(ctx, min_events); trace_io_uring_cqring_wait(ctx, min_events);
do { do {
int nr_wait = (int) iowq.cq_tail - READ_ONCE(ctx->rings->cq.tail);
unsigned long check_cq; unsigned long check_cq;
if (ctx->flags & IORING_SETUP_DEFER_TASKRUN) { if (ctx->flags & IORING_SETUP_DEFER_TASKRUN) {
int nr_wait = (int) iowq.cq_tail - READ_ONCE(ctx->rings->cq.tail);
atomic_set(&ctx->cq_wait_nr, nr_wait); atomic_set(&ctx->cq_wait_nr, nr_wait);
set_current_state(TASK_INTERRUPTIBLE); set_current_state(TASK_INTERRUPTIBLE);
} else { } else {
@ -2615,7 +2629,7 @@ static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events,
*/ */
io_run_task_work(); io_run_task_work();
if (!llist_empty(&ctx->work_llist)) if (!llist_empty(&ctx->work_llist))
io_run_local_work(ctx); io_run_local_work(ctx, nr_wait);
/* /*
* Non-local task_work will be run on exit to userspace, but * Non-local task_work will be run on exit to userspace, but
@ -3270,7 +3284,7 @@ static __cold bool io_uring_try_cancel_requests(struct io_ring_ctx *ctx,
if ((ctx->flags & IORING_SETUP_DEFER_TASKRUN) && if ((ctx->flags & IORING_SETUP_DEFER_TASKRUN) &&
io_allowed_defer_tw_run(ctx)) io_allowed_defer_tw_run(ctx))
ret |= io_run_local_work(ctx) > 0; ret |= io_run_local_work(ctx, INT_MAX) > 0;
ret |= io_cancel_defer_files(ctx, task, cancel_all); ret |= io_cancel_defer_files(ctx, task, cancel_all);
mutex_lock(&ctx->uring_lock); mutex_lock(&ctx->uring_lock);
ret |= io_poll_remove_all(ctx, task, cancel_all); ret |= io_poll_remove_all(ctx, task, cancel_all);
@ -3632,7 +3646,7 @@ SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit,
* it should handle ownership problems if any. * it should handle ownership problems if any.
*/ */
if (ctx->flags & IORING_SETUP_DEFER_TASKRUN) if (ctx->flags & IORING_SETUP_DEFER_TASKRUN)
(void)io_run_local_work_locked(ctx); (void)io_run_local_work_locked(ctx, min_complete);
} }
mutex_unlock(&ctx->uring_lock); mutex_unlock(&ctx->uring_lock);
} }