[blktest/nvme/058] Kernel OOPs while running nvme/058 tests

[Nilay Shroff]

On 8/26/25 2:38 PM, Ming Lei wrote:
> On Tue, Aug 26, 2025 at 02:00:56PM +0530, Venkat Rao Bagalkote wrote:
>> Greetings!!!
>>
>>
>> IBM CI has reported a kernel OOPs, while running blktest suite(nvme/058
>> test).
>>
>>
>> Kernel Repo:
>> https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
>>
>>
>> Traces:
>>
>>
>> [37496.800225] BUG: Kernel NULL pointer dereference at 0x00000000
>> [37496.800230] Faulting instruction address: 0xc0000000008a34b0
>> [37496.800235] Oops: Kernel access of bad area, sig: 11 [#1]
> 
> ...
> 
>> [37496.800365] GPR28: 0000000000000001 0000000000000001 c0000000b005c400
>> 0000000000000000
>> [37496.800424] NIP [c0000000008a34b0] __rq_qos_done_bio+0x3c/0x88
> 
> It looks regression from 370ac285f23a ("block: avoid cpu_hotplug_lock depedency on freeze_lock"),
> For nvme mpath, same bio crosses two drivers, so QUEUE_FLAG_QOS_ENABLED & q->rq_qos check can't
> be skipped.
> 
Thanks Ming for looking at it. And yes you were correct, we can't skip
QUEUE_FLAG_QOS_ENABLED & q->rq_qos for NVMe, However this issue only
manifests with NVMe multipath enabled, as that would create the stacked
NVMe devices. So shall I send the fix or are you going to send the patch
with fix?

Thanks,
--Nilay