On 16/11/2019 01:19, Pavel Begunkov wrote:
> On 16/11/2019 01:15, Jens Axboe wrote:
>> On 11/15/19 2:38 PM, Pavel Begunkov wrote:
>>> On 16/11/2019 00:16, Jens Axboe wrote:
>>>> On 11/15/19 12:34 PM, Jens Axboe wrote:
>>>>> How about something like this? Should work (and be valid) to have any
>>>>> sequence of timeout links, as long as there's something in front of it.
>>>>> Commit message has more details.
>>>>
>>>> Updated below (missed the sqe free), easiest to check out the repo
>>>> here:
>>>>
>>>> https://git.kernel.dk/cgit/linux-block/log/?h=for-5.5/io_uring-post
>>>>
>>>> as that will show the couple of prep patches, too. Let me know what
>>>> you think.
>>>>
>>>
>>> Sure,
>>>
>>> BTW, found "io_uring: make io_double_put_req() use normal completion
>>> path" in the tree. And it do exactly the same, what my patch was doing,
>>> the one which "blowed" the link test :)
>>
>> Hah yes, you are right, you never did resend it though. I'll get
>> rid of the one I have, and replace with your original (but with
>> the arguments fixed).
>>
> Just keep yours, it's better :)

Moreover, mine have one extra REQ_F_FAIL_LINK, which really
should not be there.

> 
>>> I'd add there "req->flags | REQ_F_FAIL_LINK" in-between failed
>>> io_req_defer() and calling io_double_put_req(). (in 2 places)
>>> Otherwise, even though a request failed, it will enqueue the rest
>>> of its link with io_queue_async_work().
>>
>> Good point, updating now.
>>
> 

-- 
Pavel Begunkov