Skip to contentSkip to conversation

iommu/vt-d: Fix bind svm with multiple devices

PATCH In review Jacob Pan 2 participants 3 messages May 8, 2019 – May 27, 2019 IOMMU

lore
Share & tools
lore

Jacob Pan

author
#0 lore raw

PATCH 1/1 iommu/vt-d: Fix bind svm with multiple devices

If multiple devices try to bind to the same mm/PASID, we need to set up first level PASID entries for all the devices. The current code does not consider this case which results in failed DMA for devices after the first bind.

---
 drivers/iommu/intel-svm.c | 15 +++++++++++++++
 1 file changed, 15 insertions(+)
drivers/iommu/intel-svm.c +15 -0
index 3a4b09a..f3d59d1 100644
@@ -357,6 +357,21 @@ int intel_svm_bind_mm(struct device *dev, int *pasid, int flags, struct svm_dev_
 		}
 
 		list_add_tail(&svm->list, &global_svm_list);
+	} else {
+		/*
+		 * Binding a new device with existing PASID, need to setup
+		 * the PASID entry.
+		 */
+		spin_lock(&iommu->lock);
+		ret = intel_pasid_setup_first_level(iommu, dev,
+						mm ? mm->pgd : init_mm.pgd,
+						svm->pasid, FLPT_DEFAULT_DID,
+						mm ? 0 : PASID_FLAG_SUPERVISOR_MODE);
+		spin_unlock(&iommu->lock);
+		if (ret) {
+			kfree(sdev);
+			goto out;
+		}
 	}
 	list_add_rcu(&sdev->list, &svm->devs);
 
Signature · 6 lines
2.7.4

_______________________________________________
iommu mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/iommu
author
#1 lore raw

Hi Joerg & David,

Any feedback on this one? Thanks.

Quoted 34 lines · On Wed, 8 May 2019 12:22:46 -0700 Jacob Pan <[email protected]>…

If multiple devices try to bind to the same mm/PASID, we need to set up first level PASID entries for all the devices. The current code does not consider this case which results in failed DMA for devices after the first bind.

---
 drivers/iommu/intel-svm.c | 15 +++++++++++++++
 1 file changed, 15 insertions(+)
drivers/iommu/intel-svm.c +3 -0
index 3a4b09a..f3d59d1 100644
@@ -357,6 +357,21 @@ int intel_svm_bind_mm(struct device *dev, int
 *pasid, int flags, struct svm_dev_ }
 
 		list_add_tail(&svm->list, &global_svm_list);
+	} else {
+		/*
+		 * Binding a new device with existing PASID, need to

setup

drivers/iommu/intel-svm.c +5 -0
+		 * the PASID entry.
+		 */
+		spin_lock(&iommu->lock);
+		ret = intel_pasid_setup_first_level(iommu, dev,
+						mm ? mm->pgd :

init_mm.pgd,

drivers/iommu/intel-svm.c +1 -0
+						svm->pasid,

FLPT_DEFAULT_DID,

drivers/iommu/intel-svm.c +1 -0
+						mm ? 0 :

PASID_FLAG_SUPERVISOR_MODE);

drivers/iommu/intel-svm.c int intel_svm_bind_mm(struct device *dev, int *pasid, int flags, struct svm_dev_ · line 376

Quoted 7 lines
drivers/iommu/intel-svm.c +5 -0
+		spin_unlock(&iommu->lock);
+		if (ret) {
+			kfree(sdev);
+			goto out;
+		}
 	}
 	list_add_rcu(&sdev->list, &svm->devs);

[Jacob Pan]

maintainer
#2 lore raw
In reply to Jacob Pan
Quoted 10 lines · Jacob Pan

If multiple devices try to bind to the same mm/PASID, we need to set up first level PASID entries for all the devices. The current code does not consider this case which results in failed DMA for devices after the first bind.

---
 drivers/iommu/intel-svm.c | 15 +++++++++++++++
 1 file changed, 15 insertions(+)

Applied for v5.3, thanks.