Re: [PATCH v2] io-wq: handle hashed writes in chains

From: Pavel Begunkov <asml.silence@gmail.com>
To: Jens Axboe <axboe@kernel.dk>, io-uring <io-uring@vger.kernel.org>
Subject: Re: [PATCH v2] io-wq: handle hashed writes in chains
Date: Sun, 22 Mar 2020 20:37:51 +0300	[thread overview]
Message-ID: <097a0386-bd60-3d37-16ad-4053edb3c12b@gmail.com> (raw)
In-Reply-To: <1c0d0978-0824-b896-d100-e0b7664ba81a@kernel.dk>

On 22/03/2020 20:08, Jens Axboe wrote:
>>> +				 */
>>> +				if (!work) {
>>> +					work = wq_first_entry(&list,
>>> +							      struct io_wq_work,
>>> +							      list);
>>
>> Wouldn't it just drop a linked request? Probably works because of the
>> comment above.
> 
> Only drops if if we always override 'work', if it's already set we use
> 'work' and don't grab from the list. So that should work for links.
> 

Oh, right, the NULL check is just above.

>>> +					if (work) {
>>> +						wq_node_del(&list, &work->list);
>>
>> There is a bug, apparently from one of my commits, where it do
>> io_assign_current_work() but then re-enqueue and reassign new work,
>> though there is a gap for cancel to happen, which would screw
>> everything up.
>>
>> I'll send a patch, so it'd be more clear. However, this is a good
>> point to look after for this as well.
> 
> Saw it, I'll apply when you have the Fixes line. I'll respin this one
> after.
> 

-- 
Pavel Begunkov