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

[Jason Gunthorpe <jgg@nvidia.com>]

On Tue, Jun 24, 2025 at 11:57:31PM +0000, Tian, Kevin wrote:
> > From: Jason Gunthorpe <jgg@nvidia.com>
> > Sent: Tuesday, June 24, 2025 10:54 PM
> > 
> > On Mon, Jun 23, 2025 at 05:49:45PM +0800, Xu Yilun wrote:
> > > +static void iommufd_device_remove_vdev(struct iommufd_device *idev)
> > > +{
> > > +	bool vdev_removing = false;
> > > +
> > > +	mutex_lock(&idev->igroup->lock);
> > > +	if (idev->vdev) {
> > > +		struct iommufd_vdevice *vdev;
> > > +
> > > +		vdev = iommufd_get_vdevice(idev->ictx, idev->vdev->obj.id);
> > > +		if (IS_ERR(vdev)) {
> > 
> > This incrs obj.users which will cause a concurrent
> > iommufd_object_remove() to fail with -EBUSY, which we are trying to
> > avoid.
> 
> concurrent remove means a user-initiated IOMMU_DESTROY, for which 
> failing with -EBUSY is expected as it doesn't wait for shortterm?

Yes a user IOMMU_DESTROY of the vdevice should not have a transient
EBUSY failure. Avoiding that is the purpose of the shorterm_users
mechanism.

> > Also you can hit a race where the tombstone has NULL'd the entry but
> > the racing destroy will then load the NULL with xas_load() and hit this:
> > 
> > 		if (WARN_ON(obj != to_destroy)) {
> 
> IOMMU_DESTROY doesn't provide to_destroy.

Right, but IOMMU_DESTROY thread could have already gone past the
xa_store(NULL) and then the kernel destroy thread could reach the
above WARN as it does use to_destroy.

Jason