* Re: [PATCH V3 0/3] selftests: add ublk selftests
2025-03-01 1:11 ` [PATCH V3 0/3] selftests: add ublk selftests Ming Lei
@ 2025-03-01 2:19 ` Jens Axboe
0 siblings, 0 replies; 2+ messages in thread
From: Jens Axboe @ 2025-03-01 2:19 UTC (permalink / raw)
To: Ming Lei; +Cc: linux-block, linux-kselftest, Keith Busch, io-uring
On 2/28/25 6:11 PM, Ming Lei wrote:
> On Fri, Feb 28, 2025 at 09:37:47AM -0700, Jens Axboe wrote:
>> On 2/28/25 9:19 AM, Ming Lei wrote:
>>> Hello Jens,
>>>
>>> This patchset adds ublk kernel selftests, which is very handy for
>>> developer for verifying kernel change, especially ublk heavily depends
>>> on io_uring subsystem. Also it provides template for target implementation.
>>>
>>> Please consider it for v6.15.
>>
>> Can we add the zc bits to the liburing test case as well?
>
> OK, will unify the two tests and cover liburing too.
>
> BTW, would you like to consider to move liburing tests or part of them
> into kernel selftests?
>
> This way looks more friendly for kernel developer:
>
> - single repo, and single patchset can include both io_uring kernel
> patches and selftests change
>
> - easy to run test against same kernel repo
I have considered it, but at least the way I run the liburing tests, it
uses a bunch of different types of devices to get full coverage. It's
not really a fire-off-and-forget kind of setup. Yes you can run it like
that and not have it use any other fs or device, but you won't get full
coverage. The liburing regression tests are also meant to be run on ANY
kernel, not just the current kernel. Eg I do that for stable kernels.
As far as I can tell, the only win here would be that it'd be easier for
someone to run when making a kernel change. And that is a nice win
indeed. But there are so many downsides for me and the tests in general,
that I don't see that win as being nearly big enough to warrant
switching it over.
For new feature tests, I think adding that as kernel selftests may make
more sense - test that it works, test failure/error cases, etc.
> Also liburing development may be decoupled from io_uring kernel
> a bit.
It's already entirely decoupled from the kernel. At least as much as it
can be. Yes the uapi header is shared and synced across them, but
there's really no other dependency there and no version dependencies
between liburing and the kernel.
--
Jens Axboe
^ permalink raw reply [flat|nested] 2+ messages in thread