Re: [bug] kernel panic when running ublk selftest on next-20250321

[Jens Axboe <axboe@kernel.dk>]

On 3/21/25 9:03 AM, Ming Lei wrote:
> On Fri, Mar 21, 2025 at 08:40:57AM -0600, Jens Axboe wrote:
>> On 3/21/25 8:40 AM, Pavel Begunkov wrote:
>>> On 3/21/25 14:33, Ming Lei wrote:
>>>> Hello,
>>>>
>>>> When running ublk selftest on today's next tree, the following kernel
>>>> panic is triggered immediately:
>>>
>>> Jens already stumbled on that one, it's likely because the
>>> cached structure is not zeroed on alloc. I believe the
>>> troubled patch is removed from the tree for now.
>>
>> Yep, ran into the same thing this morning, and yep the two last patches
>> in that series got dropped until this is resolved. Ming, if you use my
>> current for-next branch, it should all be good again.
> 
> Yeah, looks everything is good after reverting the following two patches:
> 
> Revert "io_uring/cmd: add iovec cache for commands"
> Revert "io_uring/cmd: introduce io_uring_cmd_import_fixed_vec"
> 
> BTW I feel io_uring_cmd_import_fixed_vec is useful for using ublk
> zc to write stacking target, plain readv/writev works just
> fine without zc.

I think Pavel is looking at the issue right now, so both may get
reintroduced for the 6.15 merge window. That said, we don't strictly
need the caching, we can always pile that on later. The vectored import
is going to be useful, agree, and can stand on its own.

-- 
Jens Axboe