[PATCH 0/2] iommufd: Misc cleanups per iommufd iopf merging

[PATCH 0/2] iommufd: Misc cleanups per iommufd iopf merging

[Yi Liu]

Hi,

Two misc patches per the iommufd iopf merging. Please feel free comment.

Regards,
	Yi Liu

Yi Liu (2):
  iommufd: Avoid duplicated __iommu_group_set_core_domain() call
  iommu: Set iommu_attach_handle->domain in core

 drivers/iommu/iommu.c                   | 1 +
 drivers/iommu/iommufd/fault.c           | 1 -
 drivers/iommu/iommufd/iommufd_private.h | 4 +++-
 3 files changed, 4 insertions(+), 2 deletions(-)

-- 
2.34.1

[PATCH 1/2] iommufd: Avoid duplicated __iommu_group_set_core_domain() call

[Yi Liu]

For the fault-capable hwpts, the iommufd_hwpt_detach_device() calls both
iommufd_fault_domain_detach_dev() and iommu_detach_group(). This would have
duplicated __iommu_group_set_core_domain() call since both functions call
it in the end. This looks no harm as the __iommu_group_set_core_domain()
returns if the new domain equals to the existing one. But it makes sense to
avoid such duplicated calls in caller side.

Signed-off-by: Yi Liu <yi.l.liu@intel.com>
---
 drivers/iommu/iommufd/iommufd_private.h | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/drivers/iommu/iommufd/iommufd_private.h b/drivers/iommu/iommufd/iommufd_private.h
index 92efe30a8f0d..1141c0633dc9 100644
--- a/drivers/iommu/iommufd/iommufd_private.h
+++ b/drivers/iommu/iommufd/iommufd_private.h
@@ -490,8 +490,10 @@ static inline int iommufd_hwpt_attach_device(struct iommufd_hw_pagetable *hwpt,
 static inline void iommufd_hwpt_detach_device(struct iommufd_hw_pagetable *hwpt,
 					      struct iommufd_device *idev)
 {
-	if (hwpt->fault)
+	if (hwpt->fault) {
 		iommufd_fault_domain_detach_dev(hwpt, idev);
+		return;
+	}
 
 	iommu_detach_group(hwpt->domain, idev->igroup->group);
 }
-- 
2.34.1

[PATCH 2/2] iommu: Set iommu_attach_handle->domain in core

[Yi Liu]

The IOMMU core sets the iommu_attach_handle->domain for the
iommu_attach_group_handle() path, while the iommu_replace_group_handle()
sets it on the caller side. Make the two paths aligned on it.

Signed-off-by: Yi Liu <yi.l.liu@intel.com>
---
 drivers/iommu/iommu.c         | 1 +
 drivers/iommu/iommufd/fault.c | 1 -
 2 files changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/iommu/iommu.c b/drivers/iommu/iommu.c
index ed6c5cb60c5a..83c8e617a2c5 100644
--- a/drivers/iommu/iommu.c
+++ b/drivers/iommu/iommu.c
@@ -3578,6 +3578,7 @@ int iommu_replace_group_handle(struct iommu_group *group,
 		ret = xa_reserve(&group->pasid_array, IOMMU_NO_PASID, GFP_KERNEL);
 		if (ret)
 			goto err_unlock;
+		handle->domain = new_domain;
 	}
 
 	ret = __iommu_group_set_domain(group, new_domain);
diff --git a/drivers/iommu/iommufd/fault.c b/drivers/iommu/iommufd/fault.c
index a643d5c7c535..c4715261f2c7 100644
--- a/drivers/iommu/iommufd/fault.c
+++ b/drivers/iommu/iommufd/fault.c
@@ -161,7 +161,6 @@ static int __fault_domain_replace_dev(struct iommufd_device *idev,
 		if (!handle)
 			return -ENOMEM;
 
-		handle->handle.domain = hwpt->domain;
 		handle->idev = idev;
 		ret = iommu_replace_group_handle(idev->igroup->group,
 						 hwpt->domain, &handle->handle);
-- 
2.34.1

Re: [PATCH 1/2] iommufd: Avoid duplicated __iommu_group_set_core_domain() call

[Baolu Lu]

On 2024/9/8 19:42, Yi Liu wrote:
> For the fault-capable hwpts, the iommufd_hwpt_detach_device() calls both
> iommufd_fault_domain_detach_dev() and iommu_detach_group(). This would have
> duplicated __iommu_group_set_core_domain() call since both functions call
> it in the end. This looks no harm as the __iommu_group_set_core_domain()
> returns if the new domain equals to the existing one. But it makes sense to
> avoid such duplicated calls in caller side.
> 
> Signed-off-by: Yi Liu<yi.l.liu@intel.com>
> ---
>   drivers/iommu/iommufd/iommufd_private.h | 4 +++-
>   1 file changed, 3 insertions(+), 1 deletion(-)

Reviewed-by: Lu Baolu <baolu.lu@linux.intel.com>

Thanks,
baolu

Re: [PATCH 2/2] iommu: Set iommu_attach_handle->domain in core

[Baolu Lu]

On 2024/9/8 19:42, Yi Liu wrote:
> The IOMMU core sets the iommu_attach_handle->domain for the
> iommu_attach_group_handle() path, while the iommu_replace_group_handle()
> sets it on the caller side. Make the two paths aligned on it.
> 
> Signed-off-by: Yi Liu<yi.l.liu@intel.com>
> ---
>   drivers/iommu/iommu.c         | 1 +
>   drivers/iommu/iommufd/fault.c | 1 -
>   2 files changed, 1 insertion(+), 1 deletion(-)

Reviewed-by: Lu Baolu <baolu.lu@linux.intel.com>

Thanks,
baolu

RE: [PATCH 1/2] iommufd: Avoid duplicated __iommu_group_set_core_domain() call

[Tian, Kevin]

> From: Liu, Yi L <yi.l.liu@intel.com>
> Sent: Sunday, September 8, 2024 7:43 PM
> 
> For the fault-capable hwpts, the iommufd_hwpt_detach_device() calls both
> iommufd_fault_domain_detach_dev() and iommu_detach_group(). This
> would have
> duplicated __iommu_group_set_core_domain() call since both functions call
> it in the end. This looks no harm as the __iommu_group_set_core_domain()
> returns if the new domain equals to the existing one. But it makes sense to
> avoid such duplicated calls in caller side.
> 
> Signed-off-by: Yi Liu <yi.l.liu@intel.com>

Reviewed-by: Kevin Tian <kevin.tian@intel.com>

RE: [PATCH 2/2] iommu: Set iommu_attach_handle->domain in core

[Tian, Kevin]

> From: Liu, Yi L <yi.l.liu@intel.com>
> Sent: Sunday, September 8, 2024 7:43 PM
> 
> The IOMMU core sets the iommu_attach_handle->domain for the
> iommu_attach_group_handle() path, while the
> iommu_replace_group_handle()
> sets it on the caller side. Make the two paths aligned on it.
> 
> Signed-off-by: Yi Liu <yi.l.liu@intel.com>

Reviewed-by: Kevin Tian <kevin.tian@intel.com>

Re: [PATCH 0/2] iommufd: Misc cleanups per iommufd iopf merging

[Jason Gunthorpe]

On Sun, Sep 08, 2024 at 04:42:54AM -0700, Yi Liu wrote:
> Hi,
> 
> Two misc patches per the iommufd iopf merging. Please feel free comment.
> 
> Regards,
> 	Yi Liu
> 
> Yi Liu (2):
>   iommufd: Avoid duplicated __iommu_group_set_core_domain() call
>   iommu: Set iommu_attach_handle->domain in core

Applied to for-next, thanks

Jason