On 2024/4/16 17:01, Tian, Kevin wrote:
From: Liu, Yi L <yi.l.liu@xxxxxxxxx>
Sent: Friday, April 12, 2024 4:21 PM
void vfio_iommufd_physical_unbind(struct vfio_device *vdev)
{
+ int pasid = 0;
+
lockdep_assert_held(&vdev->dev_set->lock);
+ while (!ida_is_empty(&vdev->pasids)) {
+ pasid = ida_get_lowest(&vdev->pasids, pasid, INT_MAX);
+ if (pasid < 0)
+ break;
WARN_ON as this shouldn't happen when ida is not empty.
ok.
+int vfio_iommufd_physical_pasid_attach_ioas(struct vfio_device *vdev,
+ u32 pasid, u32 *pt_id)
the name is too long. What about removing 'physical' as there is no
plan (unlikely) to support pasid on mdev?
I'm ok to do it.
+{
+ int rc;
+
+ lockdep_assert_held(&vdev->dev_set->lock);
+
+ if (WARN_ON(!vdev->iommufd_device))
+ return -EINVAL;
+
+ rc = ida_get_lowest(&vdev->pasids, pasid, pasid);
+ if (rc == pasid)
+ return iommufd_device_pasid_replace(vdev-
iommufd_device,
+ pasid, pt_id);
+
+ rc = iommufd_device_pasid_attach(vdev->iommufd_device, pasid,
pt_id);
+ if (rc)
+ return rc;
+
+ rc = ida_alloc_range(&vdev->pasids, pasid, pasid, GFP_KERNEL);
+ if (rc < 0) {
+ iommufd_device_pasid_detach(vdev->iommufd_device,
pasid);
+ return rc;
+ }
I'd do simple operation (ida_alloc_range()) first before doing attach.
But that means we rely on the ida_alloc_range() to return -ENOSPC to
indicate the pasid is allocated, hence this attach is actually a
replacement. This is easy to be broken if ida_alloc_range() returns
-ENOSPC for other reasons in future.
--
Regards,
Yi Liu
