Re: [PATCH 1/2] io_uring: clear TIF_NOTIFY_SIGNAL when running task work

From: Nadav Amit
Date: Sun Aug 08 2021 - 13:31:38 EST




> On Aug 8, 2021, at 5:55 AM, Pavel Begunkov <asml.silence@xxxxxxxxx> wrote:
>
> On 8/8/21 1:13 AM, Nadav Amit wrote:
>> From: Nadav Amit <namit@xxxxxxxxxx>
>>
>> When using SQPOLL, the submission queue polling thread calls
>> task_work_run() to run queued work. However, when work is added with
>> TWA_SIGNAL - as done by io_uring itself - the TIF_NOTIFY_SIGNAL remains
>
> static int io_req_task_work_add(struct io_kiocb *req)
> {
> ...
> notify = (req->ctx->flags & IORING_SETUP_SQPOLL) ? TWA_NONE : TWA_SIGNAL;
> if (!task_work_add(tsk, &tctx->task_work, notify))
> ...
> }
>
> io_uring doesn't set TIF_NOTIFY_SIGNAL for SQPOLL. But if you see it, I'm
> rather curious who does.

I was saying io-uring, but I meant io-uring in the wider sense:
io_queue_worker_create().

Here is a call trace for when TWA_SIGNAL is used. io_queue_worker_create()
uses TWA_SIGNAL. It is called by io_wqe_dec_running(), and not shown due
to inlining:

[ 70.540761] Call Trace:
[ 70.541352] dump_stack+0x7d/0x9c
[ 70.541930] task_work_add.cold+0x9/0x12
[ 70.542591] io_wqe_dec_running+0xd6/0xf0
[ 70.543259] io_wq_worker_sleeping+0x3d/0x60
[ 70.544106] schedule+0xa0/0xc0
[ 70.544673] userfaultfd_read_iter+0x2c3/0x790
[ 70.545374] ? wake_up_q+0xa0/0xa0
[ 70.545887] io_iter_do_read+0x1e/0x40
[ 70.546531] io_read+0xdc/0x340
[ 70.547148] ? update_curr+0x72/0x1c0
[ 70.547887] ? update_load_avg+0x7c/0x600
[ 70.548538] ? __switch_to_xtra+0x10a/0x500
[ 70.549264] io_issue_sqe+0xd99/0x1840
[ 70.549887] ? lock_timer_base+0x72/0xa0
[ 70.550516] ? try_to_del_timer_sync+0x54/0x80
[ 70.551224] io_wq_submit_work+0x87/0xb0
[ 70.552001] io_worker_handle_work+0x2b5/0x4b0
[ 70.552705] io_wqe_worker+0xd6/0x2f0
[ 70.553364] ? recalc_sigpending+0x1c/0x50
[ 70.554074] ? io_worker_handle_work+0x4b0/0x4b0
[ 70.554813] ret_from_fork+0x22/0x30

Does it answer your question?