Re: blktests failures with v6.19 kernel

[Chaitanya Kulkarni <chaitanyak@nvidia.com>]

On 2/16/26 02:26, Daniel Wagner wrote:
> Hi Chaitanya,
>
> On Sat, Feb 14, 2026 at 09:19:47PM +0000, Chaitanya Kulkarni wrote:
>> On 2/13/26 01:56, Daniel Wagner wrote:
>>> nvmet_fc_target_assoc_free runs in the nvmet_wq context and calls
>>>
>>>     nvmet_fc_delete_target_queue
>>>       nvmet_cq_put
>>>         nvmet_cq_destroy
>>>           nvmet_ctrl_put
>>>            nvmet_ctrl_free
>>>              flush_work(&ctrl->async_event_work);
>>>              cancel_work_sync(&ctrl->fatal_err_work);
>>>    
>>> The async_event_work could be running on nvmet_wq. So this deadlock is
>>> real. No idea how to fix it yet.
>>>
>> Can following patch be the potential fix for above issue as well ?
>> totally untested ...
> Yes this should work. I was not so happy adding a workqueue for this but
> after looking at nvme, this seems acceptable approach. Though, I'd make
> nvmet follow the nvme and instead adding an AEN workqueue, rather have a
> nvmet-reset-wq or nvmet_delete-wq.
>
> Thanks,
> Daniel


Thanks for looking into this.

The above patch has a cleanup bug, I've sent better patch, please have a look
hopefully we can get this merged this week.

-ck