Re: [PATCH v17 00/14] Improve write performance for zoned UFS devices

[Damien Le Moal <dlemoal@kernel.org>]

On 1/16/25 7:46 AM, Bart Van Assche wrote:
> Hi Damien and Christoph,
> 
> This patch series improves small write IOPS by a factor of four (+300%) for
> zoned UFS devices on my test setup. Please help with reviewing this patch
> series.

Bart,

Given that this is a set of 14 patches, giving a more detailed summary here of
what the entire patch set intends to do and how it does it would really help
with the review. I do not want to have to reverse engineer your line of thoughts
to see if the patch set is correctly organized.

Also, patch 14 puts back a lot of the code that was recently removed. Not nice,
and its commit message is also far too short given the size of the patch. Please
spend more time explaining what the patches do and how they do it to facilitate
review.

I also fail to see why you treat request requeue as errors if you actually
expect them to happen... I do not think you need error handling and tracking of
the completion wp position: when you get a requeue event, requeue all requests
in the plug and set the plug wp position to the lowest sector of all the
requeued requests. That is necessarily the current location of the write
pointer. No need for re-introducing all the error handling for that, no ?

I am going to wait for you to resend this with a better "big picture"
explanation of what you are trying to do so that I do not randomly comment on
things that I am not sure I completely understand. Thanks.

-- 
Damien Le Moal
Western Digital Research