On 10/25/19 3:36 AM, Pavel Begunkov wrote:

On 25/10/2019 12:31, Pavel Begunkov (Silence) wrote:

quoted

From: Pavel Begunkov <asml.silence@gmail.com>

io_sq_thread() processes sqes by 8 without considering links. As a
result, links will be randomely subdivided.

The easiest way to fix it is to call io_get_sqring() inside
io_submit_sqes() as do io_ring_submit().

Downsides:
1. This removes optimisation of not grabbing mm_struct for fixed files
2. It submitting all sqes in one go, without finer-grained sheduling
with cq processing.

Is this logic with not-grabbing mm and fixed files critical?
I want to put it back later after some cleanup.

Let's revisit that on top of the 5.5 branch of io_uring changes,
as that is being reworked anyway. I'd say it's not super critical,
as the hottest path should be polled IO with fixed buffers, and
it won't really change that.


-- 
Jens Axboe