[PATCH v3 0/3] Cleanup around iommu_set_fault_handler()

[PATCH v3 0/3] Cleanup around iommu_set_fault_handler()

[Jason Gunthorpe]

report_iommu_fault() is an older API that has been superseded by
iommu_report_device_fault() which is capable to support PRI.

Only four external drivers consume this via iommu_set_fault_handler()
  drivers/remoteproc/omap_remoteproc.c
  drivers/remoteproc/qcom_q6v5_adsp.c
  drivers/gpu/drm/msm
  drivers/infiniband/hw/usnic

Remove the use in usnic as the iommu driver logging is good enough these
days.

Remove generation support from the AMD iommu driver since msm and two
remoteproc drivers are not used on AMD x86 CPUs.

Fail iommu_set_fault_handler() if the underlying iommu driver does not
support it.

v3:
 - Go back to the v1 version of the AMD, removing the print was the right
   thing, explain in the commit message.
v2: https://patch.msgid.link/r/0-v2-25fc75484cab+ab-iommu_set_fault_jgg@nvidia.com
 - Revise commit messages
 - Move report_iommu_fault_supported into the struct iommu_ops
 - Put back the pr_err in the AMD driver
v1: https://patch.msgid.link/r/0-v1-391058a85f30+14b-iommu_set_fault_jgg@nvidia.com

Jason Gunthorpe (3):
  RDMA/usnic: Remove iommu_set_fault_handler()
  iommu/amd: Don't call report_iommu_fault()
  iommu: Allow drivers to say if they use report_iommu_fault()

 drivers/infiniband/hw/usnic/usnic_uiom.c | 13 -------------
 drivers/iommu/amd/iommu.c                | 24 ------------------------
 drivers/iommu/arm/arm-smmu/arm-smmu.c    |  1 +
 drivers/iommu/arm/arm-smmu/qcom_iommu.c  |  1 +
 drivers/iommu/iommu.c                    |  6 +++++-
 drivers/iommu/ipmmu-vmsa.c               |  1 +
 drivers/iommu/mtk_iommu.c                |  1 +
 drivers/iommu/mtk_iommu_v1.c             |  1 +
 drivers/iommu/omap-iommu.c               |  1 +
 drivers/iommu/rockchip-iommu.c           |  1 +
 drivers/iommu/sun50i-iommu.c             |  1 +
 include/linux/iommu.h                    |  3 +++
 12 files changed, 16 insertions(+), 38 deletions(-)


base-commit: 211ddde0823f1442e4ad052a2f30f050145ccada
-- 
2.43.0

[PATCH v3 1/3] RDMA/usnic: Remove iommu_set_fault_handler()

[Jason Gunthorpe]

The handler in usnic just prints a fault report message, the iommu drivers
all do a better job of that these days. Just remove the use of this old
API.

Reviewed-by: Lu Baolu <baolu.lu@linux.intel.com>
Reviewed-by: Kevin Tian <kevin.tian@intel.com>
Acked-by: Jason Gunthorpe <jgg@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
---
 drivers/infiniband/hw/usnic/usnic_uiom.c | 13 -------------
 1 file changed, 13 deletions(-)

diff --git a/drivers/infiniband/hw/usnic/usnic_uiom.c b/drivers/infiniband/hw/usnic/usnic_uiom.c
index 3fbf99757b1148..f7fb6246c83aba 100644
--- a/drivers/infiniband/hw/usnic/usnic_uiom.c
+++ b/drivers/infiniband/hw/usnic/usnic_uiom.c
@@ -51,17 +51,6 @@
 	((void *) &((struct usnic_uiom_chunk *) 0)->page_list[1] -	\
 	(void *) &((struct usnic_uiom_chunk *) 0)->page_list[0]))
 
-static int usnic_uiom_dma_fault(struct iommu_domain *domain,
-				struct device *dev,
-				unsigned long iova, int flags,
-				void *token)
-{
-	usnic_err("Device %s iommu fault domain 0x%p va 0x%lx flags 0x%x\n",
-		dev_name(dev),
-		domain, iova, flags);
-	return -ENOSYS;
-}
-
 static void usnic_uiom_put_pages(struct list_head *chunk_list, int dirty)
 {
 	struct usnic_uiom_chunk *chunk, *tmp;
@@ -450,8 +439,6 @@ struct usnic_uiom_pd *usnic_uiom_alloc_pd(struct device *dev)
 		return ERR_CAST(domain);
 	}
 
-	iommu_set_fault_handler(pd->domain, usnic_uiom_dma_fault, NULL);
-
 	spin_lock_init(&pd->lock);
 	INIT_LIST_HEAD(&pd->devs);
 
-- 
2.43.0

[PATCH v3 2/3] iommu/amd: Don't call report_iommu_fault()

[Jason Gunthorpe]

This old style API is only used by drivers/gpu/drm/msm,
drivers/remoteproc/omap_remoteproc.c, and
drivers/remoteproc/qcom_q6v5_adsp.c none are used on x86 HW.

Remove the dead code to discourage new users.

Also remove the domain == NULL print because it was intended to protect
against a NULL deref inside report_iommu_fault() which is no longer
possible.

Just always print the fault in the same format if it could get a dev_data.
There is no value to be gained by also printing if the domain is NULL. In
today's kernel when the dev_data is populated the domain will be made
!NULL very quickly during iommu device probing.

Reviewed-by: Vasant Hegde <vasant.hegde@amd.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
---
 drivers/iommu/amd/iommu.c | 24 ------------------------
 1 file changed, 24 deletions(-)

diff --git a/drivers/iommu/amd/iommu.c b/drivers/iommu/amd/iommu.c
index 2e1865daa1cee8..072c80bb2c2b3a 100644
--- a/drivers/iommu/amd/iommu.c
+++ b/drivers/iommu/amd/iommu.c
@@ -840,29 +840,6 @@ static void amd_iommu_report_page_fault(struct amd_iommu *iommu,
 		dev_data = dev_iommu_priv_get(&pdev->dev);
 
 	if (dev_data) {
-		/*
-		 * If this is a DMA fault (for which the I(nterrupt)
-		 * bit will be unset), allow report_iommu_fault() to
-		 * prevent logging it.
-		 */
-		if (IS_IOMMU_MEM_TRANSACTION(flags)) {
-			/* Device not attached to domain properly */
-			if (dev_data->domain == NULL) {
-				pr_err_ratelimited("Event logged [Device not attached to domain properly]\n");
-				pr_err_ratelimited("  device=%04x:%02x:%02x.%x domain=0x%04x\n",
-						   iommu->pci_seg->id, PCI_BUS_NUM(devid), PCI_SLOT(devid),
-						   PCI_FUNC(devid), domain_id);
-				goto out;
-			}
-
-			if (!report_iommu_fault(&dev_data->domain->domain,
-						&pdev->dev, address,
-						IS_WRITE_REQUEST(flags) ?
-							IOMMU_FAULT_WRITE :
-							IOMMU_FAULT_READ))
-				goto out;
-		}
-
 		if (__ratelimit(&dev_data->rs)) {
 			pci_err(pdev, "Event logged [IO_PAGE_FAULT domain=0x%04x address=0x%llx flags=0x%04x]\n",
 				domain_id, address, flags);
@@ -873,7 +850,6 @@ static void amd_iommu_report_page_fault(struct amd_iommu *iommu,
 			domain_id, address, flags);
 	}
 
-out:
 	if (pdev)
 		pci_dev_put(pdev);
 }
-- 
2.43.0

[PATCH v3 3/3] iommu: Allow drivers to say if they use report_iommu_fault()

[Jason Gunthorpe]

report_iommu_fault() is an older API that has been superseded by
iommu_report_device_fault() which is capable to support PRI.

Only two external drivers consume this, drivers/remoteproc and
drivers/gpu/drm/msm. Ideally they would move over to the new APIs, but for
now protect against accidentally mix and matching the wrong components.

The iommu drivers support either the old iommu_set_fault_handler() via the
driver calling report_iommu_fault(), or they are newer server focused
drivers that call iommu_report_device_fault().

Include a flag in the iommu_ops if the driver calls report_iommu_fault()
and block iommu_set_fault_handler() for domain's of iommu drivers that
can't support it.

Reviewed-by: Lu Baolu <baolu.lu@linux.intel.com>
Reviewed-by: Kevin Tian <kevin.tian@intel.com>
Reviewed-by: Vasant Hegde <vasant.hegde@amd.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
---
 drivers/iommu/arm/arm-smmu/arm-smmu.c   | 1 +
 drivers/iommu/arm/arm-smmu/qcom_iommu.c | 1 +
 drivers/iommu/iommu.c                   | 6 +++++-
 drivers/iommu/ipmmu-vmsa.c              | 1 +
 drivers/iommu/mtk_iommu.c               | 1 +
 drivers/iommu/mtk_iommu_v1.c            | 1 +
 drivers/iommu/omap-iommu.c              | 1 +
 drivers/iommu/rockchip-iommu.c          | 1 +
 drivers/iommu/sun50i-iommu.c            | 1 +
 include/linux/iommu.h                   | 3 +++
 10 files changed, 16 insertions(+), 1 deletion(-)

diff --git a/drivers/iommu/arm/arm-smmu/arm-smmu.c b/drivers/iommu/arm/arm-smmu/arm-smmu.c
index 4ced4b5bee4df3..f68a0576b9a3a6 100644
--- a/drivers/iommu/arm/arm-smmu/arm-smmu.c
+++ b/drivers/iommu/arm/arm-smmu/arm-smmu.c
@@ -1643,6 +1643,7 @@ static const struct iommu_ops arm_smmu_ops = {
 	.get_resv_regions	= arm_smmu_get_resv_regions,
 	.def_domain_type	= arm_smmu_def_domain_type,
 	.owner			= THIS_MODULE,
+	.report_iommu_fault_supported = true,
 	.default_domain_ops = &(const struct iommu_domain_ops) {
 		.attach_dev		= arm_smmu_attach_dev,
 		.map_pages		= arm_smmu_map_pages,
diff --git a/drivers/iommu/arm/arm-smmu/qcom_iommu.c b/drivers/iommu/arm/arm-smmu/qcom_iommu.c
index c5be95e560317e..173d5957796790 100644
--- a/drivers/iommu/arm/arm-smmu/qcom_iommu.c
+++ b/drivers/iommu/arm/arm-smmu/qcom_iommu.c
@@ -597,6 +597,7 @@ static const struct iommu_ops qcom_iommu_ops = {
 	.probe_device	= qcom_iommu_probe_device,
 	.device_group	= generic_device_group,
 	.of_xlate	= qcom_iommu_of_xlate,
+	.report_iommu_fault_supported = true,
 	.default_domain_ops = &(const struct iommu_domain_ops) {
 		.attach_dev	= qcom_iommu_attach_dev,
 		.map_pages	= qcom_iommu_map,
diff --git a/drivers/iommu/iommu.c b/drivers/iommu/iommu.c
index 59244c744eabd2..8d9086a37671e1 100644
--- a/drivers/iommu/iommu.c
+++ b/drivers/iommu/iommu.c
@@ -2005,6 +2005,9 @@ EXPORT_SYMBOL_GPL(iommu_group_has_isolated_msi);
  * This function should be used by IOMMU users which want to be notified
  * whenever an IOMMU fault happens.
  *
+ * This is a legacy API not supported by all drivers. New users should look
+ * to using domain->iopf_handler for the modern API.
+ *
  * The fault handler itself should return 0 on success, and an appropriate
  * error code otherwise.
  */
@@ -2012,7 +2015,8 @@ void iommu_set_fault_handler(struct iommu_domain *domain,
 					iommu_fault_handler_t handler,
 					void *token)
 {
-	if (WARN_ON(!domain || domain->cookie_type != IOMMU_COOKIE_NONE))
+	if (WARN_ON(!domain || domain->cookie_type != IOMMU_COOKIE_NONE ||
+		    !domain->owner->report_iommu_fault_supported))
 		return;
 
 	domain->cookie_type = IOMMU_COOKIE_FAULT_HANDLER;
diff --git a/drivers/iommu/ipmmu-vmsa.c b/drivers/iommu/ipmmu-vmsa.c
index ffa892f6571406..04a5836bf36000 100644
--- a/drivers/iommu/ipmmu-vmsa.c
+++ b/drivers/iommu/ipmmu-vmsa.c
@@ -884,6 +884,7 @@ static const struct iommu_ops ipmmu_ops = {
 	.device_group = IS_ENABLED(CONFIG_ARM) && !IS_ENABLED(CONFIG_IOMMU_DMA)
 			? generic_device_group : generic_single_device_group,
 	.of_xlate = ipmmu_of_xlate,
+	.report_iommu_fault_supported = true,
 	.default_domain_ops = &(const struct iommu_domain_ops) {
 		.attach_dev	= ipmmu_attach_device,
 		.map_pages	= ipmmu_map,
diff --git a/drivers/iommu/mtk_iommu.c b/drivers/iommu/mtk_iommu.c
index 0e0285348d2b8e..ccf6fdeb64a261 100644
--- a/drivers/iommu/mtk_iommu.c
+++ b/drivers/iommu/mtk_iommu.c
@@ -1018,6 +1018,7 @@ static const struct iommu_ops mtk_iommu_ops = {
 	.of_xlate	= mtk_iommu_of_xlate,
 	.get_resv_regions = mtk_iommu_get_resv_regions,
 	.owner		= THIS_MODULE,
+	.report_iommu_fault_supported = true,
 	.default_domain_ops = &(const struct iommu_domain_ops) {
 		.attach_dev	= mtk_iommu_attach_device,
 		.map_pages	= mtk_iommu_map,
diff --git a/drivers/iommu/mtk_iommu_v1.c b/drivers/iommu/mtk_iommu_v1.c
index 10cc0b1197e801..3dd6222d210921 100644
--- a/drivers/iommu/mtk_iommu_v1.c
+++ b/drivers/iommu/mtk_iommu_v1.c
@@ -581,6 +581,7 @@ static const struct iommu_ops mtk_iommu_v1_ops = {
 	.release_device	= mtk_iommu_v1_release_device,
 	.device_group	= generic_device_group,
 	.owner          = THIS_MODULE,
+	.report_iommu_fault_supported = true,
 	.default_domain_ops = &(const struct iommu_domain_ops) {
 		.attach_dev	= mtk_iommu_v1_attach_device,
 		.map_pages	= mtk_iommu_v1_map,
diff --git a/drivers/iommu/omap-iommu.c b/drivers/iommu/omap-iommu.c
index 5c6f5943f44b1f..7ba67600f32915 100644
--- a/drivers/iommu/omap-iommu.c
+++ b/drivers/iommu/omap-iommu.c
@@ -1723,6 +1723,7 @@ static const struct iommu_ops omap_iommu_ops = {
 	.release_device	= omap_iommu_release_device,
 	.device_group	= generic_single_device_group,
 	.of_xlate	= omap_iommu_of_xlate,
+	.report_iommu_fault_supported = true,
 	.default_domain_ops = &(const struct iommu_domain_ops) {
 		.attach_dev	= omap_iommu_attach_dev,
 		.map_pages	= omap_iommu_map,
diff --git a/drivers/iommu/rockchip-iommu.c b/drivers/iommu/rockchip-iommu.c
index 0861dd469bd866..932b5a62bac572 100644
--- a/drivers/iommu/rockchip-iommu.c
+++ b/drivers/iommu/rockchip-iommu.c
@@ -1173,6 +1173,7 @@ static const struct iommu_ops rk_iommu_ops = {
 	.release_device = rk_iommu_release_device,
 	.device_group = generic_single_device_group,
 	.of_xlate = rk_iommu_of_xlate,
+	.report_iommu_fault_supported = true,
 	.default_domain_ops = &(const struct iommu_domain_ops) {
 		.attach_dev	= rk_iommu_attach_device,
 		.map_pages	= rk_iommu_map,
diff --git a/drivers/iommu/sun50i-iommu.c b/drivers/iommu/sun50i-iommu.c
index de10b569d9a940..13c37b1be481d7 100644
--- a/drivers/iommu/sun50i-iommu.c
+++ b/drivers/iommu/sun50i-iommu.c
@@ -848,6 +848,7 @@ static const struct iommu_ops sun50i_iommu_ops = {
 	.domain_alloc_paging = sun50i_iommu_domain_alloc_paging,
 	.of_xlate	= sun50i_iommu_of_xlate,
 	.probe_device	= sun50i_iommu_probe_device,
+	.report_iommu_fault_supported = true,
 	.default_domain_ops = &(const struct iommu_domain_ops) {
 		.attach_dev	= sun50i_iommu_attach_device,
 		.flush_iotlb_all = sun50i_iommu_flush_iotlb_all,
diff --git a/include/linux/iommu.h b/include/linux/iommu.h
index c30d12e16473df..941e4971b05fd3 100644
--- a/include/linux/iommu.h
+++ b/include/linux/iommu.h
@@ -659,6 +659,8 @@ __iommu_copy_struct_to_user(const struct iommu_user_data *dst_data,
  *                    no user domain for each PASID and the I/O page faults are
  *                    forwarded through the user domain attached to the device
  *                    RID.
+ * @report_iommu_fault_supported: True if the domain supports
+ *                                iommu_set_fault_handler()
  */
 struct iommu_ops {
 	bool (*capable)(struct device *dev, enum iommu_cap);
@@ -710,6 +712,7 @@ struct iommu_ops {
 	struct iommu_domain *release_domain;
 	struct iommu_domain *default_domain;
 	u8 user_pasid_table:1;
+	u8 report_iommu_fault_supported:1;
 };
 
 /**
-- 
2.43.0

RE: [PATCH v3 2/3] iommu/amd: Don't call report_iommu_fault()

[Tian, Kevin]

> From: Jason Gunthorpe <jgg@nvidia.com>
> Sent: Friday, November 21, 2025 3:46 AM
> 
> This old style API is only used by drivers/gpu/drm/msm,
> drivers/remoteproc/omap_remoteproc.c, and
> drivers/remoteproc/qcom_q6v5_adsp.c none are used on x86 HW.
> 
> Remove the dead code to discourage new users.
> 
> Also remove the domain == NULL print because it was intended to protect
> against a NULL deref inside report_iommu_fault() which is no longer
> possible.
> 
> Just always print the fault in the same format if it could get a dev_data.
> There is no value to be gained by also printing if the domain is NULL. In
> today's kernel when the dev_data is populated the domain will be made
> !NULL very quickly during iommu device probing.
> 
> Reviewed-by: Vasant Hegde <vasant.hegde@amd.com>
> Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

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

Re: [PATCH v3 1/3] RDMA/usnic: Remove iommu_set_fault_handler()

[Leon Romanovsky]

On Thu, Nov 20, 2025 at 03:46:22PM -0400, Jason Gunthorpe wrote:
> The handler in usnic just prints a fault report message, the iommu drivers
> all do a better job of that these days. Just remove the use of this old
> API.
> 
> Reviewed-by: Lu Baolu <baolu.lu@linux.intel.com>
> Reviewed-by: Kevin Tian <kevin.tian@intel.com>
> Acked-by: Jason Gunthorpe <jgg@nvidia.com>
> Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
> ---
>  drivers/infiniband/hw/usnic/usnic_uiom.c | 13 -------------
>  1 file changed, 13 deletions(-)
>

I imagine that you want to merge this patch through iommu tree.

Thanks,
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>