On 25/01/2021 02:14, Joseph Qi wrote:
> On 1/23/21 5:49 AM, Pavel Begunkov wrote:
>> On 22/01/2021 09:45, Joseph Qi wrote:
>>> Seems this series can also resolve a possible deadlock case I'm looking
>>> into.
>>
>> It removes dead code, I believe your issue is solved by
>> ("io_uring: get rid of intermediate IORING_OP_CLOSE stage")
>> https://git.kernel.dk/cgit/linux-block/commit/?h=for-5.12/io_uring&id=7be8ba3b656cb4e0158b2c859b949f34a96aa94f
>>
> I've tested the above commit and the mentioned possible deadlock still
> exists.
I pasted a wrong patch, apparently had something else on my mind...
The intention was to point to a Jens' patch that made mentioned in
your report __io_uring_files_cancel() unreachable, though it was
dropped for unrelated reasons. So... the bug is still here at the
moment.
Anyway, thanks a lot for testing! Your reports are very helpful
>> Did you try this series in particular, or tested for-5.12/io_uring
>> and see that the issue is gone there?
>>
> I don't have this tree locally and it takes too long to clone it down.
> Will check once it is ready.
>
> Thanks,
> Joseph
>
>>> CPU0:
>>> ...
>>> io_kill_linked_timeout // &ctx->completion_lock
>>> io_commit_cqring
>>> __io_queue_deferred
>>> __io_queue_async_work
>>> io_wq_enqueue
>>> io_wqe_enqueue // &wqe->lock
>>>
>>> CPU1:
>>> ...
>>> __io_uring_files_cancel
>>> io_wq_cancel_cb
>>> io_wqe_cancel_pending_work // &wqe->lock
>>> io_cancel_task_cb // &ctx->completion_lock
>>>
--
Pavel Begunkov
