Re: [PATCH v2 0/5] fast poll multishot mode

From: Jens Axboe <axboe@kernel.dk>
To: Pavel Begunkov <asml.silence@gmail.com>,
	Hao Xu <haoxu.linux@gmail.com>,
	io-uring@vger.kernel.org
Cc: linux-kernel@vger.kernel.org
Subject: Re: [PATCH v2 0/5] fast poll multishot mode
Date: Fri, 6 May 2022 10:03:58 -0600	[thread overview]
Message-ID: <05e99d9e-4c66-6d7c-604c-1e52d8d0b5b2@kernel.dk> (raw)
In-Reply-To: <3f940dad-73ce-4ea6-dc76-f877c64dbb9a@gmail.com>

On 5/6/22 10:01 AM, Pavel Begunkov wrote:
> On 5/6/22 15:18, Jens Axboe wrote:
>> On 5/6/22 1:36 AM, Hao Xu wrote:
>>> Hi All,
>>> I actually had a question about the current poll code, from the code it
>>> seems when we cancel a poll-like request, it will ignore the existing
>>> events and just raise a -ECANCELED cqe though I haven't tested it. Is
>>> this by design or am I missing something?
>>
>> That's by design, but honestly I don't think anyone considered the case
>> where it's being canceled but has events already. For that case, I think
>> we should follow the usual logic of only returning an error (canceled)
>> if we don't have events, if we have events just return them. For
>> multi-shot, obviously also terminate, but same logic there.
> 
> Why would we care? It's inherently racy in any case and any
> user not handling this is already screwed.

We don't really need to care, but the normal logic is "return error if
we have no result, return result otherwise". No reason this should be
any different imho, it's not really a correctness thing.

-- 
Jens Axboe