On 7/23/21 11:36 AM, Jens Axboe wrote:

On 7/23/21 11:32 AM, Jens Axboe wrote:

quoted

Outside of that, we're not submitting off release, only killing anything
pending. The only odd case there is iopoll, but that doesn't resubmit
here.

OK perhaps I'm wrong on this one - if we have a pending iopoll request,
and we run into the rare case of needing resubmit, we are doing that off
the release path and that should not happen. Hence it could potentially
happen for iosched and/or low queue depth devices, if you are using a
ring for pure polling. I'll patch that up.

Will send out two patches for this. Note that I don't see this being a
real issue, as we explicitly gave the ring fd to another task, and being
that this is purely for read/write, it would result in -EFAULT anyway.

-- 
Jens Axboe