Re: [PATCH v5 04/13] iommufd/viommu: Add IOMMU_VIOMMU_ALLOC ioctl

[Nicolin Chen <nicolinc@nvidia.com>]

On Tue, Oct 29, 2024 at 11:54:36AM -0300, Jason Gunthorpe wrote:
> On Fri, Oct 25, 2024 at 04:49:44PM -0700, Nicolin Chen wrote:
> > +void iommufd_viommu_destroy(struct iommufd_object *obj)
> > +{
> > +	struct iommufd_viommu *viommu =
> > +		container_of(obj, struct iommufd_viommu, obj);
> > +
> > +	if (viommu->ops && viommu->ops->free)
> > +		viommu->ops->free(viommu);
> 
> Ops can't be null and free can't be null, that would mean there is a
> memory leak.

What if a driver doesn't have anything to free? You're suggesting
to force the driver to have an empty free function, right? We can
do that, if it is preferable:

void arm_vsmmu_free(struct iommufd_viommu *viommu)
{
}

> > +	refcount_dec(&viommu->hwpt->common.obj.users);
> 
> Don't touch viommu after freeing it

Drivers should only free their own stuff without touching the core:
"* @free: Free all driver-specific parts of an iommufd_viommu. The memory of the
 *        vIOMMU will be free-ed by iommufd core after calling this free op."

Then, viommu object is freed by the core after ->destroy(), right?

> Did you run selftests with kasn?

Yea, all passed with no WARN_ON.

Thanks
Nicolin