Re: [PATCH v3 03/17] iommufd: Replace the hwpt->devices list with iommufd_group

[Jason Gunthorpe <jgg@nvidia.com>]

On Thu, Apr 20, 2023 at 06:15:16AM +0000, Tian, Kevin wrote:
> > > which internal list? group has a list for attached devices but regarding
> > > to hwpt it's stored in a single field igroup->hwpt.
> > 
> > It is added to
> > 
> > 	list_add_tail(&hwpt->hwpt_item, &hwpt->ioas->hwpt_list);
> 
> this is called under ioas->mutex.

Yes.. But.. that is troubled too, we are calling destroy under the
same mutex, there is a missing a fault point to catch it in the test,
and hwpt_alloc doesn't have the lock wide enough :\

So you want to argue that it is safe to do this:

   mutex_lock(&ioas->mutex);
   alloc
   attach
   detach
   abort
   mutex_unlock(&ioas->mutex);

Even if attach/detach lock/unlock the group mutex during their cycle?

It seems OK..

I don't see any places that Though I don't much like the locking
pattern where we succeed attach, drop all the locks and the fail and
then relock and do error unwind.. Sketchy..

Jason