Re: [PATCH v2 3/4] iommufd: Destroy vdevice on idevice destroy

[Xu Yilun <yilun.xu@linux.intel.com>]

On Wed, Jun 25, 2025 at 09:38:32AM -0300, Jason Gunthorpe wrote:
> On Wed, Jun 25, 2025 at 06:06:00PM +0800, Xu Yilun wrote:
> > > 	/*
> > > 	 * We don't know what thread is actually going to destroy the vdev, but
> > > 	 * once the vdev is destroyed the pointer is NULL'd. At this
> > > 	 * point idev->users is 0 so no other thread can set a new vdev.
> > > 	 */
> > > 	if (!wait_event_timeout(idev->ictx->destroy_wait,
> > > 				!READ_ONCE(idev->vdev),
> > > 				msecs_to_jiffies(60000)))
> > > 		pr_crit("Time out waiting for iommufd vdevice removed\n");
> > > }
> > > 
> > > Though there is a cleaner option here, you could do:
> > > 
> > > 	mutex_lock(&idev->igroup->lock);
> > > 	if (idev->vdev)
> > > 		iommufd_vdevice_abort(&idev->vdev->obj);
> > > 	mutex_unlock(&idev->igroup->lock);
> > > 
> > > And make it safe to call abort twice, eg by setting dev to NULL and
> > > checking for that. First thread to get to the igroup lock, either via
> > > iommufd_vdevice_destroy() or via the above will do the actual abort
> > > synchronously without any wait_event_timeout. That seems better??
> > 
> > I'm good to both options, but slightly tend not to make vdevice so
> > special from other objects, so still prefer the wait_event option.
> 
> The wait_event is a ugly hack though, even in its existing code. The
> above version is better because it doesn't have any failure mode and
> doesn't introduce any unlocked use of the idev->vdev which is easier
> to reason about, no READ_ONCE/WRITE_ONCE/etc
> 
> It sounds like you should largely leave the existing other parts the
> same as this v2, though can you try reorganize it to look a little
> more like the version I shared?

Sure. But may I confirm that your only want reentrant
iommufd_vdevice_abort() but not your iommufd_object_remove_tombstone()
changes?

To me, grab a shortterm_users but not a user is a new operation model. I
hesitate to add it when the existing refcount_inc(&obj->user) works for
this case.

Thanks,
Yilun

> 
> Jason
>