[Patch V3 0/2] iommu/arm-smmu: Fix races in iommu domain/group creation

[Patch V3 0/2] iommu/arm-smmu: Fix races in iommu domain/group creation

[Ashish Mhetre]

When two devices with same SID are getting probed concurrently through
iommu_probe_device(), the iommu_group and iommu_domain are allocated more
than once because they are not protected for concurrency. This is leading
to context faults when one device is accessing IOVA from other device.
Fix this by protecting iommu_domain and iommu_group creation with mutexes.

Changes in v3:
* Updated commit messages.
* Added Signed-off-by in patch 2.

Ashish Mhetre (1):
  iommu: Fix race condition during default domain allocation

Krishna Reddy (1):
  iommu/arm-smmu: Fix race condition during iommu_group creation

 drivers/iommu/arm/arm-smmu/arm-smmu.c | 6 +++++-
 drivers/iommu/iommu.c                 | 2 ++
 2 files changed, 7 insertions(+), 1 deletion(-)

-- 
2.7.4

[Patch V3 1/2] iommu: Fix race condition during default domain allocation

[Ashish Mhetre]

When two devices with same SID are getting probed concurrently through
iommu_probe_device(), the iommu_domain sometimes is getting allocated more
than once as call to iommu_alloc_default_domain() is not protected for
concurrency. Furthermore, it leads to each device holding a different
iommu_domain pointer, separate IOVA space and only one of the devices'
domain is used for translations from IOMMU. This causes accesses from other
device to fault or see incorrect translations.
Fix this by protecting iommu_alloc_default_domain() call with group->mutex
and let all devices with same SID share same iommu_domain.

Signed-off-by: Ashish Mhetre <amhetre@nvidia.com>
---
 drivers/iommu/iommu.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/drivers/iommu/iommu.c b/drivers/iommu/iommu.c
index 5419c4b..80c5a1c 100644
--- a/drivers/iommu/iommu.c
+++ b/drivers/iommu/iommu.c
@@ -273,7 +273,9 @@ int iommu_probe_device(struct device *dev)
 	 * support default domains, so the return value is not yet
 	 * checked.
 	 */
+	mutex_lock(&group->mutex);
 	iommu_alloc_default_domain(group, dev);
+	mutex_unlock(&group->mutex);
 
 	if (group->default_domain) {
 		ret = __iommu_attach_device(group->default_domain, dev);
-- 
2.7.4

[Patch V3 2/2] iommu/arm-smmu: Fix race condition during iommu_group creation

[Ashish Mhetre]

From: Krishna Reddy <vdumpa@nvidia.com>

When two devices with same SID are getting probed concurrently through
iommu_probe_device(), the iommu_group sometimes is getting allocated more
than once as call to arm_smmu_device_group() is not protected for
concurrency. Furthermore, it leads to each device holding a different
iommu_group and domain pointer, separate IOVA space and only one of the
devices' domain is used for translations from IOMMU. This causes accesses
from other device to fault or see incorrect translations.
Fix this by protecting iommu_group allocation from concurrency in
arm_smmu_device_group().

Signed-off-by: Krishna Reddy <vdumpa@nvidia.com>
Signed-off-by: Ashish Mhetre <amhetre@nvidia.com>
---
 drivers/iommu/arm/arm-smmu/arm-smmu.c | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)

diff --git a/drivers/iommu/arm/arm-smmu/arm-smmu.c b/drivers/iommu/arm/arm-smmu/arm-smmu.c
index dba15f3..f9c6648 100644
--- a/drivers/iommu/arm/arm-smmu/arm-smmu.c
+++ b/drivers/iommu/arm/arm-smmu/arm-smmu.c
@@ -1457,6 +1457,7 @@ static struct iommu_group *arm_smmu_device_group(struct device *dev)
 	struct iommu_group *group = NULL;
 	int i, idx;
 
+	mutex_lock(&smmu->stream_map_mutex);
 	for_each_cfg_sme(cfg, fwspec, i, idx) {
 		if (group && smmu->s2crs[idx].group &&
 		    group != smmu->s2crs[idx].group)
@@ -1465,8 +1466,10 @@ static struct iommu_group *arm_smmu_device_group(struct device *dev)
 		group = smmu->s2crs[idx].group;
 	}
 
-	if (group)
+	if (group) {
+		mutex_unlock(&smmu->stream_map_mutex);
 		return iommu_group_ref_get(group);
+	}
 
 	if (dev_is_pci(dev))
 		group = pci_device_group(dev);
@@ -1480,6 +1483,7 @@ static struct iommu_group *arm_smmu_device_group(struct device *dev)
 		for_each_cfg_sme(cfg, fwspec, i, idx)
 			smmu->s2crs[idx].group = group;
 
+	mutex_unlock(&smmu->stream_map_mutex);
 	return group;
 }
 
-- 
2.7.4

Re: [Patch V3 0/2] iommu/arm-smmu: Fix races in iommu domain/group creation

[Will Deacon]

On Tue, 10 Aug 2021 10:13:59 +0530, Ashish Mhetre wrote:
> When two devices with same SID are getting probed concurrently through
> iommu_probe_device(), the iommu_group and iommu_domain are allocated more
> than once because they are not protected for concurrency. This is leading
> to context faults when one device is accessing IOVA from other device.
> Fix this by protecting iommu_domain and iommu_group creation with mutexes.
> 
> Changes in v3:
> * Updated commit messages.
> * Added Signed-off-by in patch 2.
> 
> [...]

Applied to will (for-joerg/arm-smmu/updates), thanks!

[1/2] iommu: Fix race condition during default domain allocation
      https://git.kernel.org/will/c/211ff31b3d33
[2/2] iommu/arm-smmu: Fix race condition during iommu_group creation
      https://git.kernel.org/will/c/b1a1347912a7

Cheers,
-- 
Will

https://fixes.arm64.dev
https://next.arm64.dev
https://will.arm64.dev