iommufd: Add enforced_cache_coherency to iommufd_hw_pagetable_alloc()

Logically the HWPT should have the coherency set properly for the device
that it is being created for when it is created.

This was happening implicitly if the immediate_attach was set because
iommufd_hw_pagetable_attach() does it as the first thing.

Do it unconditionally so !immediate_attach works properly.

Link: https://lore.kernel.org/r/9-v8-6659224517ea+532-iommufd_alloc_jgg@nvidia.com
Reviewed-by: Kevin Tian <kevin.tian@intel.com>
Reviewed-by: Lu Baolu <baolu.lu@linux.intel.com>
Tested-by: Nicolin Chen <nicolinc@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

3 files changed, 32 insertions, 15 deletions

diff --git a/drivers/iommu/iommufd/device.c b/drivers/iommu/iommufd/device.c
index 89cef2d2a2b1..08bba99f2fb0 100644
--- a/drivers/iommu/iommufd/device.c
+++ b/drivers/iommu/iommufd/device.c
@@ -338,22 +338,11 @@ int iommufd_hw_pagetable_attach(struct iommufd_hw_pagetable *hwpt,
 		goto err_unlock;
 	}
 
-	/*
-	 * Try to upgrade the domain we have, it is an iommu driver bug to
-	 * report IOMMU_CAP_ENFORCE_CACHE_COHERENCY but fail
-	 * enforce_cache_coherency when there are no devices attached to the
-	 * domain.
-	 */
-	if (idev->enforce_cache_coherency && !hwpt->enforce_cache_coherency) {
-		if (hwpt->domain->ops->enforce_cache_coherency)
-			hwpt->enforce_cache_coherency =
-				hwpt->domain->ops->enforce_cache_coherency(
-					hwpt->domain);
-		if (!hwpt->enforce_cache_coherency) {
-			WARN_ON(list_empty(&idev->igroup->device_list));
-			rc = -EINVAL;
+	/* Try to upgrade the domain we have */
+	if (idev->enforce_cache_coherency) {
+		rc = iommufd_hw_pagetable_enforce_cc(hwpt);
+		if (rc)
 			goto err_unlock;
-		}
 	}
 
 	rc = iopt_table_enforce_dev_resv_regions(&hwpt->ioas->iopt, idev->dev,
diff --git a/drivers/iommu/iommufd/hw_pagetable.c b/drivers/iommu/iommufd/hw_pagetable.c
index bdb76cdb1dc3..e0699d7f4c64 100644
--- a/drivers/iommu/iommufd/hw_pagetable.c
+++ b/drivers/iommu/iommufd/hw_pagetable.c
@@ -25,6 +25,20 @@ void iommufd_hw_pagetable_destroy(struct iommufd_object *obj)
 	refcount_dec(&hwpt->ioas->obj.users);
 }
 
+int iommufd_hw_pagetable_enforce_cc(struct iommufd_hw_pagetable *hwpt)
+{
+	if (hwpt->enforce_cache_coherency)
+		return 0;
+
+	if (hwpt->domain->ops->enforce_cache_coherency)
+		hwpt->enforce_cache_coherency =
+			hwpt->domain->ops->enforce_cache_coherency(
+				hwpt->domain);
+	if (!hwpt->enforce_cache_coherency)
+		return -EINVAL;
+	return 0;
+}
+
 /**
  * iommufd_hw_pagetable_alloc() - Get an iommu_domain for a device
  * @ictx: iommufd context
@@ -61,6 +75,19 @@ iommufd_hw_pagetable_alloc(struct iommufd_ctx *ictx, struct iommufd_ioas *ioas,
 	}
 
 	/*
+	 * Set the coherency mode before we do iopt_table_add_domain() as some
+	 * iommus have a per-PTE bit that controls it and need to decide before
+	 * doing any maps. It is an iommu driver bug to report
+	 * IOMMU_CAP_ENFORCE_CACHE_COHERENCY but fail enforce_cache_coherency on
+	 * a new domain.
+	 */
+	if (idev->enforce_cache_coherency) {
+		rc = iommufd_hw_pagetable_enforce_cc(hwpt);
+		if (WARN_ON(rc))
+			goto out_abort;
+	}
+
+	/*
 	 * immediate_attach exists only to accommodate iommu drivers that cannot
 	 * directly allocate a domain. These drivers do not finish creating the
 	 * domain until attach is completed. Thus we must have this call
diff --git a/drivers/iommu/iommufd/iommufd_private.h b/drivers/iommu/iommufd/iommufd_private.h
index 8bdbef8bd878..a6210f51b956 100644
--- a/drivers/iommu/iommufd/iommufd_private.h
+++ b/drivers/iommu/iommufd/iommufd_private.h
@@ -254,6 +254,7 @@ struct iommufd_hw_pagetable {
 struct iommufd_hw_pagetable *
 iommufd_hw_pagetable_alloc(struct iommufd_ctx *ictx, struct iommufd_ioas *ioas,
 			   struct iommufd_device *idev, bool immediate_attach);
+int iommufd_hw_pagetable_enforce_cc(struct iommufd_hw_pagetable *hwpt);
 int iommufd_hw_pagetable_attach(struct iommufd_hw_pagetable *hwpt,
 				struct iommufd_device *idev);
 struct iommufd_hw_pagetable *