在 2021/1/31 下午8:51, Pavel Begunkov 写道:
On 29/01/2021 17:53, Hao Xu wrote:
So check if sqthread gets park flag right before schedule().
since ctx_list is always empty when this problem happens, here I put
kthread_should_park() before setting the wakeup flag(ctx_list is empty
so this for loop is fast), where is close enough to schedule(). The
problem doesn't show again in my repro testing after this fix.
Looks good, and I believe I saw syzbot reporting similar thing before.
Two nits below
Reported-by: Abaci <abaci@xxxxxxxxxxxxxxxxx>
Signed-off-by: Hao Xu <haoxu@xxxxxxxxxxxxxxxxx>
---
fs/io_uring.c | 3 +++
1 file changed, 3 insertions(+)
diff --git a/fs/io_uring.c b/fs/io_uring.c
index c07913ec0cca..444dc993157e 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -7132,6 +7132,9 @@ static int io_sq_thread(void *data)
}
}
How about killing btw a kthread_should_park() check few lines
above before prepare_to_wait? Parking is fairly rare, so we
don't need fast path for it.
Gotcha, It makes sense to kill that one if parking is rare.
+ if (kthread_should_park())
+ needs_sched = false;
+
if (needs_sched) {
if (needs_sched && !kthread_should_park())
Looks cleaner to me
Agree, It's cleaner.
I'll send a v3 soon. Thank you, Pavel.
Hao
list_for_each_entry(ctx, &sqd->ctx_list, sqd_list)
io_ring_set_wakeup_flag(ctx);
