Preparation to .map_page and .unmap_page removal

PATCH Superseded Leon Romanovsky 3 participants 17 messages Sep 16, 2025 – Sep 17, 2025

Subsystems (2)

Arm (32-bit) DMA mapping helpers

lore

Share & tools

lore

Leon Romanovsky

author

2025-09-16 07:32 UTC #0 lore raw

v3 cover Preparation to .map_page and .unmap_page removal

Changelog:
v3:
* Rewrote the series to allow combination of .map_resource and .map_page to one flow.
* Added two new patches to convert and remove .map_resource.
v2: https://lore.kernel.org/all/[email protected]/
* Added default "else" section without map_phys and map_page callbacks (impossible).
v1: https://lore.kernel.org/all/[email protected]
* Changed "else if" instead of "if".
v0: https://lore.kernel.org/all/[email protected]
---------------------------------------------------------------------

This is followup to "dma-mapping: migrate to physical address-based API" series https://lore.kernel.org/all/[email protected]

Thanks

Leon Romanovsky (4):
dma-mapping: prepare dma_map_ops to conversion to physical address
dma-mapping: convert dummy ops to physical address mapping
ARM: dma-mapping: Switch to physical address mapping callbacks
dma-mapping: remove unused mapping resource callbacks

 arch/arm/mm/dma-mapping.c   | 98 +++++++++----------------------------
 include/linux/dma-map-ops.h | 13 ++---
 kernel/dma/dummy.c          | 13 +++--
 kernel/dma/mapping.c        | 20 +++-----
 4 files changed, 43 insertions(+), 101 deletions(-)

Signature · 1 line

2.51.0

Leon Romanovsky

author

2025-09-16 07:32 UTC #3 lore raw

PATCH v3 2/4 dma-mapping: convert dummy ops to physical address mapping

From: Leon Romanovsky <[email protected]>

Change dma_dummy_map_page and dma_dummy_unmap_page routines to accept physical address and rename them.

Signed-off-by Leon Romanovsky <[email protected]>

---
 kernel/dma/dummy.c | 13 ++++++-------
 1 file changed, 6 insertions(+), 7 deletions(-)

kernel/dma/dummy.c +6 -7

index 92de80e5b057e..16a51736a2a39 100644
@@ -11,17 +11,16 @@ static int dma_dummy_mmap(struct device *dev, struct vm_area_struct *vma,
 	return -ENXIO;
 }
 
-static dma_addr_t dma_dummy_map_page(struct device *dev, struct page *page,
-		unsigned long offset, size_t size, enum dma_data_direction dir,
-		unsigned long attrs)
+static dma_addr_t dma_dummy_map_phys(struct device *dev, phys_addr_t phys,
+		size_t size, enum dma_data_direction dir, unsigned long attrs)
 {
 	return DMA_MAPPING_ERROR;
 }
-static void dma_dummy_unmap_page(struct device *dev, dma_addr_t dma_handle,
+static void dma_dummy_unmap_phys(struct device *dev, dma_addr_t dma_handle,
 		size_t size, enum dma_data_direction dir, unsigned long attrs)
 {
 	/*
-	 * Dummy ops doesn't support map_page, so unmap_page should never be
+	 * Dummy ops doesn't support map_phys, so unmap_page should never be
 	 * called.
 	 */
 	WARN_ON_ONCE(true);

@@ -51,8 +50,8 @@ static int dma_dummy_supported(struct device *hwdev, u64 mask)
 
 const struct dma_map_ops dma_dummy_ops = {
 	.mmap                   = dma_dummy_mmap,
-	.map_page               = dma_dummy_map_page,
-	.unmap_page             = dma_dummy_unmap_page,
+	.map_phys               = dma_dummy_map_phys,
+	.unmap_phys             = dma_dummy_unmap_phys,
 	.map_sg                 = dma_dummy_map_sg,
 	.unmap_sg               = dma_dummy_unmap_sg,
 	.dma_supported          = dma_dummy_supported,

Signature · 1 line

2.51.0

Leon Romanovsky

author

2025-09-16 07:32 UTC #1 lore raw

PATCH v3 1/4 dma-mapping: prepare dma_map_ops to conversion to physical address

In reply to Leon Romanovsky

From: Leon Romanovsky <[email protected]>

Add new .map_phys() and .unmap_phys() callbacks to dma_map_ops as a preparation to replace .map_page() and .unmap_page() respectively.

Signed-off-by Leon Romanovsky <[email protected]>

---
 include/linux/dma-map-ops.h | 7 +++++++
 kernel/dma/mapping.c        | 4 ++++
 2 files changed, 11 insertions(+)

include/linux/dma-map-ops.h +7 -0

index 71f5b30254159..25603cb273769 100644
@@ -37,6 +37,13 @@ struct dma_map_ops {
 	void (*unmap_page)(struct device *dev, dma_addr_t dma_handle,
 			size_t size, enum dma_data_direction dir,
 			unsigned long attrs);
+
+	dma_addr_t (*map_phys)(struct device *dev, phys_addr_t phys,
+			size_t size, enum dma_data_direction dir,
+			unsigned long attrs);
+	void (*unmap_phys)(struct device *dev, dma_addr_t dma_handle,
+			size_t size, enum dma_data_direction dir,
+			unsigned long attrs);
 	/*
 	 * map_sg should return a negative error code on error. See
 	 * dma_map_sgtable() for a list of appropriate error codes

kernel/dma/mapping.c +4 -0

index fe7472f13b106..4080aebe5debb 100644
@@ -169,6 +169,8 @@ dma_addr_t dma_map_phys(struct device *dev, phys_addr_t phys, size_t size,
 		addr = dma_direct_map_phys(dev, phys, size, dir, attrs);
 	else if (use_dma_iommu(dev))
 		addr = iommu_dma_map_phys(dev, phys, size, dir, attrs);
+	else if (ops->map_phys)
+		addr = ops->map_phys(dev, phys, size, dir, attrs);
 	else if (is_mmio) {
 		if (!ops->map_resource)
 			return DMA_MAPPING_ERROR;

@@ -223,6 +225,8 @@ void dma_unmap_phys(struct device *dev, dma_addr_t addr, size_t size,
 		dma_direct_unmap_phys(dev, addr, size, dir, attrs);
 	else if (use_dma_iommu(dev))
 		iommu_dma_unmap_phys(dev, addr, size, dir, attrs);
+	else if (ops->unmap_phys)
+		ops->unmap_phys(dev, addr, size, dir, attrs);
 	else if (is_mmio) {
 		if (ops->unmap_resource)
 			ops->unmap_resource(dev, addr, size, dir, attrs);

Signature · 1 line

2.51.0

Leon Romanovsky

author

2025-09-16 07:32 UTC #14 lore raw

PATCH v3 4/4 dma-mapping: remove unused mapping resource callbacks

In reply to Leon Romanovsky

From: Leon Romanovsky <[email protected]>

After ARM conversion to use physical addresses for the mapping, there is no in-kernel users for map_resource/unmap_resource callbacks, so remove it.

Signed-off-by Leon Romanovsky <[email protected]>

---
 include/linux/dma-map-ops.h |  6 ------
 kernel/dma/mapping.c        | 16 ++++------------
 2 files changed, 4 insertions(+), 18 deletions(-)

include/linux/dma-map-ops.h +0 -6

index 25603cb273769..a2ec1566aa270 100644
@@ -53,12 +53,6 @@ struct dma_map_ops {
 			enum dma_data_direction dir, unsigned long attrs);
 	void (*unmap_sg)(struct device *dev, struct scatterlist *sg, int nents,
 			enum dma_data_direction dir, unsigned long attrs);
-	dma_addr_t (*map_resource)(struct device *dev, phys_addr_t phys_addr,
-			size_t size, enum dma_data_direction dir,
-			unsigned long attrs);
-	void (*unmap_resource)(struct device *dev, dma_addr_t dma_handle,
-			size_t size, enum dma_data_direction dir,
-			unsigned long attrs);
 	void (*sync_single_for_cpu)(struct device *dev, dma_addr_t dma_handle,
 			size_t size, enum dma_data_direction dir);
 	void (*sync_single_for_device)(struct device *dev,

kernel/dma/mapping.c +4 ~~-12~~

index 4080aebe5debb..32a85bfdf873a 100644
@@ -157,7 +157,7 @@ dma_addr_t dma_map_phys(struct device *dev, phys_addr_t phys, size_t size,
 {
 	const struct dma_map_ops *ops = get_dma_ops(dev);
 	bool is_mmio = attrs & DMA_ATTR_MMIO;
-	dma_addr_t addr;
+	dma_addr_t addr = DMA_MAPPING_ERROR;
 
 	BUG_ON(!valid_dma_direction(dir));

@@ -171,18 +171,13 @@ dma_addr_t dma_map_phys(struct device *dev, phys_addr_t phys, size_t size,
 		addr = iommu_dma_map_phys(dev, phys, size, dir, attrs);
 	else if (ops->map_phys)
 		addr = ops->map_phys(dev, phys, size, dir, attrs);
-	else if (is_mmio) {
-		if (!ops->map_resource)
-			return DMA_MAPPING_ERROR;
-
-		addr = ops->map_resource(dev, phys, size, dir, attrs);
-	} else {
+	else if (!is_mmio && ops->map_page) {
 		struct page *page = phys_to_page(phys);
 		size_t offset = offset_in_page(phys);
 
 		/*
 		 * The dma_ops API contract for ops->map_page() requires
-		 * kmappable memory, while ops->map_resource() does not.
+		 * kmappable memory.
 		 */
 		addr = ops->map_page(dev, page, offset, size, dir, attrs);
 	}

@@ -227,10 +222,7 @@ void dma_unmap_phys(struct device *dev, dma_addr_t addr, size_t size,
 		iommu_dma_unmap_phys(dev, addr, size, dir, attrs);
 	else if (ops->unmap_phys)
 		ops->unmap_phys(dev, addr, size, dir, attrs);
-	else if (is_mmio) {
-		if (ops->unmap_resource)
-			ops->unmap_resource(dev, addr, size, dir, attrs);
-	} else
+	else
 		ops->unmap_page(dev, addr, size, dir, attrs);
 	trace_dma_unmap_phys(dev, addr, size, dir, attrs);
 	debug_dma_unmap_phys(dev, addr, size, dir);

Signature · 1 line

2.51.0

Leon Romanovsky

author

2025-09-16 07:32 UTC #5 lore raw

PATCH v3 3/4 ARM: dma-mapping: Switch to physical address mapping callbacks

In reply to Leon Romanovsky

From: Leon Romanovsky <[email protected]>

Combine resource and page mappings routines to one function, which handles both these flows at the same manner. This conversion allows us to remove .map_resource/.unmap_resource callbacks completely.

Signed-off-by Leon Romanovsky <[email protected]>

---
 arch/arm/mm/dma-mapping.c | 98 +++++++++------------------------------
 1 file changed, 22 insertions(+), 76 deletions(-)

arch/arm/mm/dma-mapping.c +22 ~~-76~~

index 88c2d68a69c9e..e0d65414b39e1 100644
@@ -737,6 +737,9 @@ static int __dma_info_to_prot(enum dma_data_direction dir, unsigned long attrs)
 	if (attrs & DMA_ATTR_PRIVILEGED)
 		prot |= IOMMU_PRIV;
 
+	if (attrs & DMA_ATTR_MMIO)
+		prot |= IOMMU_MMIO;
+
 	switch (dir) {
 	case DMA_BIDIRECTIONAL:
 		return prot | IOMMU_READ | IOMMU_WRITE;

@@ -1356,25 +1359,27 @@ static void arm_iommu_sync_sg_for_device(struct device *dev,
 }
 
 /**
- * arm_iommu_map_page
+ * arm_iommu_map_phys
  * @dev: valid struct device pointer
- * @page: page that buffer resides in
- * @offset: offset into page for start of buffer
+ * @phys: physical address that buffer resides in
  * @size: size of buffer to map
  * @dir: DMA transfer direction
+ * @attrs: DMA mapping attributes
  *
  * IOMMU aware version of arm_dma_map_page()
  */
-static dma_addr_t arm_iommu_map_page(struct device *dev, struct page *page,
-	     unsigned long offset, size_t size, enum dma_data_direction dir,
-	     unsigned long attrs)
+static dma_addr_t arm_iommu_map_phys(struct device *dev, phys_addr_t phys,
+	     size_t size, enum dma_data_direction dir, unsigned long attrs)
 {
 	struct dma_iommu_mapping *mapping = to_dma_iommu_mapping(dev);
+	int len = PAGE_ALIGN(size + offset_in_page(phys));
+	phys_addr_t addr = phys_addr & PAGE_MASK;
 	dma_addr_t dma_addr;
-	int ret, prot, len = PAGE_ALIGN(size + offset);
+	int ret, prot;
 
-	if (!dev->dma_coherent && !(attrs & DMA_ATTR_SKIP_CPU_SYNC))
-		__dma_page_cpu_to_dev(page, offset, size, dir);
+	if (!dev->dma_coherent &&
+	    !(attrs & (DMA_ATTR_SKIP_CPU_SYNC | DMA_ATTR_MMIO)))
+		__dma_page_cpu_to_dev(phys_to_page(phys), offset, size, dir);
 
 	dma_addr = __alloc_iova(mapping, len);
 	if (dma_addr == DMA_MAPPING_ERROR)

@@ -1382,8 +1387,7 @@ static dma_addr_t arm_iommu_map_page(struct device *dev, struct page *page,
 
 	prot = __dma_info_to_prot(dir, attrs);
 
-	ret = iommu_map(mapping->domain, dma_addr, page_to_phys(page), len,
-			prot, GFP_KERNEL);
+	ret = iommu_map(mapping->domain, dma_addr, addr, len, prot, GFP_KERNEL);
 	if (ret < 0)
 		goto fail;

@@ -1399,10 +1403,11 @@ static dma_addr_t arm_iommu_map_page(struct device *dev, struct page *page,
  * @handle: DMA address of buffer
  * @size: size of buffer (same as passed to dma_map_page)
  * @dir: DMA transfer direction (same as passed to dma_map_page)
+ * @attrs: DMA mapping attributes
  *
- * IOMMU aware version of arm_dma_unmap_page()
+ * IOMMU aware version of arm_dma_unmap_phys()
  */
-static void arm_iommu_unmap_page(struct device *dev, dma_addr_t handle,
+static void arm_iommu_unmap_phys(struct device *dev, dma_addr_t handle,
 		size_t size, enum dma_data_direction dir, unsigned long attrs)
 {
 	struct dma_iommu_mapping *mapping = to_dma_iommu_mapping(dev);

@@ -1414,7 +1419,8 @@ static void arm_iommu_unmap_page(struct device *dev, dma_addr_t handle,
 	if (!iova)
 		return;
 
-	if (!dev->dma_coherent && !(attrs & DMA_ATTR_SKIP_CPU_SYNC)) {
+	if (!dev->dma_coherent &&
+	    !(attrs & (DMA_ATTR_SKIP_CPU_SYNC | DMA_ATTR_MMIO))) {
 		page = phys_to_page(iommu_iova_to_phys(mapping->domain, iova));
 		__dma_page_dev_to_cpu(page, offset, size, dir);
 	}

@@ -1423,63 +1429,6 @@ static void arm_iommu_unmap_page(struct device *dev, dma_addr_t handle,
 	__free_iova(mapping, iova, len);
 }
 
-/**
- * arm_iommu_map_resource - map a device resource for DMA
- * @dev: valid struct device pointer
- * @phys_addr: physical address of resource
- * @size: size of resource to map
- * @dir: DMA transfer direction
- */
-static dma_addr_t arm_iommu_map_resource(struct device *dev,
-		phys_addr_t phys_addr, size_t size,
-		enum dma_data_direction dir, unsigned long attrs)
-{
-	struct dma_iommu_mapping *mapping = to_dma_iommu_mapping(dev);
-	dma_addr_t dma_addr;
-	int ret, prot;
-	phys_addr_t addr = phys_addr & PAGE_MASK;
-	unsigned int offset = phys_addr & ~PAGE_MASK;
-	size_t len = PAGE_ALIGN(size + offset);
-
-	dma_addr = __alloc_iova(mapping, len);
-	if (dma_addr == DMA_MAPPING_ERROR)
-		return dma_addr;
-
-	prot = __dma_info_to_prot(dir, attrs) | IOMMU_MMIO;
-
-	ret = iommu_map(mapping->domain, dma_addr, addr, len, prot, GFP_KERNEL);
-	if (ret < 0)
-		goto fail;
-
-	return dma_addr + offset;
-fail:
-	__free_iova(mapping, dma_addr, len);
-	return DMA_MAPPING_ERROR;
-}
-
-/**
- * arm_iommu_unmap_resource - unmap a device DMA resource
- * @dev: valid struct device pointer
- * @dma_handle: DMA address to resource
- * @size: size of resource to map
- * @dir: DMA transfer direction
- */
-static void arm_iommu_unmap_resource(struct device *dev, dma_addr_t dma_handle,
-		size_t size, enum dma_data_direction dir,
-		unsigned long attrs)
-{
-	struct dma_iommu_mapping *mapping = to_dma_iommu_mapping(dev);
-	dma_addr_t iova = dma_handle & PAGE_MASK;
-	unsigned int offset = dma_handle & ~PAGE_MASK;
-	size_t len = PAGE_ALIGN(size + offset);
-
-	if (!iova)
-		return;
-
-	iommu_unmap(mapping->domain, iova, len);
-	__free_iova(mapping, iova, len);
-}
-
 static void arm_iommu_sync_single_for_cpu(struct device *dev,
 		dma_addr_t handle, size_t size, enum dma_data_direction dir)
 {

@@ -1516,8 +1465,8 @@ static const struct dma_map_ops iommu_ops = {
 	.mmap		= arm_iommu_mmap_attrs,
 	.get_sgtable	= arm_iommu_get_sgtable,
 
-	.map_page		= arm_iommu_map_page,
-	.unmap_page		= arm_iommu_unmap_page,
+	.map_phys		= arm_iommu_map_phys,
+	.unmap_phys		= arm_iommu_unmap_phys,
 	.sync_single_for_cpu	= arm_iommu_sync_single_for_cpu,
 	.sync_single_for_device	= arm_iommu_sync_single_for_device,

@@ -1525,9 +1474,6 @@ static const struct dma_map_ops iommu_ops = {
 	.unmap_sg		= arm_iommu_unmap_sg,
 	.sync_sg_for_cpu	= arm_iommu_sync_sg_for_cpu,
 	.sync_sg_for_device	= arm_iommu_sync_sg_for_device,
-
-	.map_resource		= arm_iommu_map_resource,
-	.unmap_resource		= arm_iommu_unmap_resource,
 };
 
 /**

Signature · 1 line

2.51.0

Leon Romanovsky

2025-09-16 12:21 UTC #15 lore raw

Re: [PATCH v3 4/4] dma-mapping: remove unused mapping resource callbacks

In reply to Leon Romanovsky

Quoted 11 lines · Leon Romanovsky

From: Leon Romanovsky <[email protected]>
After ARM conversion to use physical addresses for the mapping, there is no in-kernel users for map_resource/unmap_resource callbacks, so remove it.
Signed-off-by Leon Romanovsky <[email protected]>
---
 include/linux/dma-map-ops.h |  6 ------
 kernel/dma/mapping.c        | 16 ++++------------
 2 files changed, 4 insertions(+), 18 deletions(-)

This patch is premature, I missed another map_resource user in drivers/xen/swiotlb-xen.c.

Thanks

Quoted 68 lines

include/linux/dma-map-ops.h +0 -6

index 25603cb273769..a2ec1566aa270 100644
@@ -53,12 +53,6 @@ struct dma_map_ops {
 			enum dma_data_direction dir, unsigned long attrs);
 	void (*unmap_sg)(struct device *dev, struct scatterlist *sg, int nents,
 			enum dma_data_direction dir, unsigned long attrs);
-	dma_addr_t (*map_resource)(struct device *dev, phys_addr_t phys_addr,
-			size_t size, enum dma_data_direction dir,
-			unsigned long attrs);
-	void (*unmap_resource)(struct device *dev, dma_addr_t dma_handle,
-			size_t size, enum dma_data_direction dir,
-			unsigned long attrs);
 	void (*sync_single_for_cpu)(struct device *dev, dma_addr_t dma_handle,
 			size_t size, enum dma_data_direction dir);
 	void (*sync_single_for_device)(struct device *dev,

kernel/dma/mapping.c +4 ~~-12~~

index 4080aebe5debb..32a85bfdf873a 100644
@@ -157,7 +157,7 @@ dma_addr_t dma_map_phys(struct device *dev, phys_addr_t phys, size_t size,
 {
 	const struct dma_map_ops *ops = get_dma_ops(dev);
 	bool is_mmio = attrs & DMA_ATTR_MMIO;
-	dma_addr_t addr;
+	dma_addr_t addr = DMA_MAPPING_ERROR;
 
 	BUG_ON(!valid_dma_direction(dir));

@@ -171,18 +171,13 @@ dma_addr_t dma_map_phys(struct device *dev, phys_addr_t phys, size_t size,
 		addr = iommu_dma_map_phys(dev, phys, size, dir, attrs);
 	else if (ops->map_phys)
 		addr = ops->map_phys(dev, phys, size, dir, attrs);
-	else if (is_mmio) {
-		if (!ops->map_resource)
-			return DMA_MAPPING_ERROR;
-
-		addr = ops->map_resource(dev, phys, size, dir, attrs);
-	} else {
+	else if (!is_mmio && ops->map_page) {
 		struct page *page = phys_to_page(phys);
 		size_t offset = offset_in_page(phys);
 
 		/*
 		 * The dma_ops API contract for ops->map_page() requires
-		 * kmappable memory, while ops->map_resource() does not.
+		 * kmappable memory.
 		 */
 		addr = ops->map_page(dev, page, offset, size, dir, attrs);
 	}

@@ -227,10 +222,7 @@ void dma_unmap_phys(struct device *dev, dma_addr_t addr, size_t size,
 		iommu_dma_unmap_phys(dev, addr, size, dir, attrs);
 	else if (ops->unmap_phys)
 		ops->unmap_phys(dev, addr, size, dir, attrs);
-	else if (is_mmio) {
-		if (ops->unmap_resource)
-			ops->unmap_resource(dev, addr, size, dir, attrs);
-	} else
+	else
 		ops->unmap_page(dev, addr, size, dir, attrs);
 	trace_dma_unmap_phys(dev, addr, size, dir, attrs);
 	debug_dma_unmap_phys(dev, addr, size, dir);

Signature · 1 line

2.51.0

Jason Gunthorpe

Reviewed-by on v3 1/4

2025-09-16 13:52 UTC #2 lore raw

Re: [PATCH v3 1/4] dma-mapping: prepare dma_map_ops to conversion to physical address

In reply to Leon Romanovsky

Quoted 10 lines · Leon Romanovsky

From: Leon Romanovsky <[email protected]>
Add new .map_phys() and .unmap_phys() callbacks to dma_map_ops as a preparation to replace .map_page() and .unmap_page() respectively.
Signed-off-by Leon Romanovsky <[email protected]>
---
 include/linux/dma-map-ops.h | 7 +++++++
 kernel/dma/mapping.c        | 4 ++++
 2 files changed, 11 insertions(+)

Reviewed-by Jason Gunthorpe <[email protected]>

Jason

Jason Gunthorpe

Reviewed-by on v3 2/4

2025-09-16 13:53 UTC #4 lore raw

Re: [PATCH v3 2/4] dma-mapping: convert dummy ops to physical address mapping

In reply to Leon Romanovsky

Quoted 9 lines · Leon Romanovsky

From: Leon Romanovsky <[email protected]>
Change dma_dummy_map_page and dma_dummy_unmap_page routines to accept physical address and rename them.
Signed-off-by Leon Romanovsky <[email protected]>
---
 kernel/dma/dummy.c | 13 ++++++-------
 1 file changed, 6 insertions(+), 7 deletions(-)

Reviewed-by Jason Gunthorpe <[email protected]>

Jason

Jason Gunthorpe

Reviewed-by on v3 3/4

2025-09-16 18:46 UTC #6 lore raw

Re: [PATCH v3 3/4] ARM: dma-mapping: Switch to physical address mapping callbacks

In reply to Leon Romanovsky

Review

Quoted 3 lines · Leon Romanovsky

(quoted diff) +3 -0

+	if (!dev->dma_coherent &&
+	    !(attrs & (DMA_ATTR_SKIP_CPU_SYNC | DMA_ATTR_MMIO)))
+		__dma_page_cpu_to_dev(phys_to_page(phys), offset, size, dir);

I'd keep going and get rid of the page here too, maybe as a second patch in this series:

arch/arm/mm/dma-mapping.c +10 ~~-15~~

index 88c2d68a69c9ee..a84d12cd0ba4a9 100644
@@ -624,16 +624,14 @@ static void __arm_dma_free(struct device *dev, size_t size, void *cpu_addr,
 	kfree(buf);
 }
 
-static void dma_cache_maint_page(struct page *page, unsigned long offset,
+static void dma_cache_maint_page(phys_addr_t paddr,
 	size_t size, enum dma_data_direction dir,
 	void (*op)(const void *, size_t, int))
 {
-	unsigned long pfn;
+	unsigned long pfn = paddr / PAGE_SIZE;
+	unsigned int offset = paddr % PAGE_SIZE;
 	size_t left = size;
 
-	pfn = page_to_pfn(page) + offset / PAGE_SIZE;
-	offset %= PAGE_SIZE;
-
 	/*
 	 * A single sg entry may refer to multiple physically contiguous
 	 * pages.  But we still need to process highmem pages individually.

@@ -644,17 +642,17 @@ static void dma_cache_maint_page(struct page *page, unsigned long offset,
 		size_t len = left;
 		void *vaddr;
 
-		page = pfn_to_page(pfn);
-
-		if (PageHighMem(page)) {
+		if (PhysHighMem(pfn << PAGE_SHIFT)) {
 			if (len + offset > PAGE_SIZE)
 				len = PAGE_SIZE - offset;
 
 			if (cache_is_vipt_nonaliasing()) {
-				vaddr = kmap_atomic(page);
+				vaddr = kmap_atomic_pfn(pfn);
 				op(vaddr + offset, len, dir);
 				kunmap_atomic(vaddr);
 			} else {
+				struct page *page = pfn_to_page(pfn);
+
 				vaddr = kmap_high_get(page);
 				if (vaddr) {
 					op(vaddr + offset, len, dir);

@@ -662,7 +660,7 @@ static void dma_cache_maint_page(struct page *page, unsigned long offset,
 				}
 			}
 		} else {
-			vaddr = page_address(page) + offset;
+			vaddr = phys_to_virt(pfn) + offset;
 			op(vaddr, len, dir);
 		}
 		offset = 0;

@@ -676,14 +674,11 @@ static void dma_cache_maint_page(struct page *page, unsigned long offset,
  * Note: Drivers should NOT use this function directly.
  * Use the driver DMA support - see dma-mapping.h (dma_sync_*)
  */
-static void __dma_page_cpu_to_dev(struct page *page, unsigned long off,
+static void __dma_page_cpu_to_dev(phys_addr_t paddr,
 	size_t size, enum dma_data_direction dir)
 {
-	phys_addr_t paddr;
+	dma_cache_maint_page(paddr, size, dir, dmac_map_area);
 
-	dma_cache_maint_page(page, off, size, dir, dmac_map_area);
-
-	paddr = page_to_phys(page) + off;
 	if (dir == DMA_FROM_DEVICE) {
 		outer_inv_range(paddr, paddr + size);
 	} else {

Review

Quoted 4 lines

(quoted diff) +2 -0

+	if (!dev->dma_coherent &&
+	    !(attrs & (DMA_ATTR_SKIP_CPU_SYNC | DMA_ATTR_MMIO))) {
 		page = phys_to_page(iommu_iova_to_phys(mapping->domain, iova));
 		__dma_page_dev_to_cpu(page, offset, size, dir);

Same treatment here..

Looks Ok though, I didn't notice any pitfalls

Reviewed-by Jason Gunthorpe <[email protected]>

Jason

Jason Gunthorpe

Reviewed-by on v3 4/4

2025-09-16 18:49 UTC #16 lore raw

Re: [PATCH v3 4/4] dma-mapping: remove unused mapping resource callbacks

In reply to Leon Romanovsky

Quoted 11 lines · Leon Romanovsky

From: Leon Romanovsky <[email protected]>
After ARM conversion to use physical addresses for the mapping, there is no in-kernel users for map_resource/unmap_resource callbacks, so remove it.
Signed-off-by Leon Romanovsky <[email protected]>
---
 include/linux/dma-map-ops.h |  6 ------
 kernel/dma/mapping.c        | 16 ++++------------
 2 files changed, 4 insertions(+), 18 deletions(-)

Reviewed-by Jason Gunthorpe <[email protected]>

Looks like xen_swiotlb_dma_ops will be easy enough to fix..

Jason

Leon Romanovsky

author

2025-09-17 10:36 UTC #7 lore raw

Re: [PATCH v3 3/4] ARM: dma-mapping: Switch to physical address mapping callbacks

In reply to Jason Gunthorpe

Quoted 7 lines · Jason Gunthorpe

Review
Quoted 3 lines · Leon Romanovsky
(quoted diff) +3 -0
+	if (!dev->dma_coherent &&
+	    !(attrs & (DMA_ATTR_SKIP_CPU_SYNC | DMA_ATTR_MMIO)))
+		__dma_page_cpu_to_dev(phys_to_page(phys), offset, size, dir);
I'd keep going and get rid of the page here too, maybe as a second patch in this series:

Thanks, it is always unclear how far to go with cleanups.

Quoted 87 lines

arch/arm/mm/dma-mapping.c +10 ~~-15~~

index 88c2d68a69c9ee..a84d12cd0ba4a9 100644
@@ -624,16 +624,14 @@ static void __arm_dma_free(struct device *dev, size_t size, void *cpu_addr,
 	kfree(buf);
 }
 
-static void dma_cache_maint_page(struct page *page, unsigned long offset,
+static void dma_cache_maint_page(phys_addr_t paddr,
 	size_t size, enum dma_data_direction dir,
 	void (*op)(const void *, size_t, int))
 {
-	unsigned long pfn;
+	unsigned long pfn = paddr / PAGE_SIZE;
+	unsigned int offset = paddr % PAGE_SIZE;
 	size_t left = size;
 
-	pfn = page_to_pfn(page) + offset / PAGE_SIZE;
-	offset %= PAGE_SIZE;
-
 	/*
 	 * A single sg entry may refer to multiple physically contiguous
 	 * pages.  But we still need to process highmem pages individually.

@@ -644,17 +642,17 @@ static void dma_cache_maint_page(struct page *page, unsigned long offset,
 		size_t len = left;
 		void *vaddr;
 
-		page = pfn_to_page(pfn);
-
-		if (PageHighMem(page)) {
+		if (PhysHighMem(pfn << PAGE_SHIFT)) {
 			if (len + offset > PAGE_SIZE)
 				len = PAGE_SIZE - offset;
 
 			if (cache_is_vipt_nonaliasing()) {
-				vaddr = kmap_atomic(page);
+				vaddr = kmap_atomic_pfn(pfn);
 				op(vaddr + offset, len, dir);
 				kunmap_atomic(vaddr);
 			} else {
+				struct page *page = pfn_to_page(pfn);
+
 				vaddr = kmap_high_get(page);
 				if (vaddr) {
 					op(vaddr + offset, len, dir);

@@ -662,7 +660,7 @@ static void dma_cache_maint_page(struct page *page, unsigned long offset,
 				}
 			}
 		} else {
-			vaddr = page_address(page) + offset;
+			vaddr = phys_to_virt(pfn) + offset;
 			op(vaddr, len, dir);
 		}
 		offset = 0;

@@ -676,14 +674,11 @@ static void dma_cache_maint_page(struct page *page, unsigned long offset,
  * Note: Drivers should NOT use this function directly.
  * Use the driver DMA support - see dma-mapping.h (dma_sync_*)
  */
-static void __dma_page_cpu_to_dev(struct page *page, unsigned long off,
+static void __dma_page_cpu_to_dev(phys_addr_t paddr,
 	size_t size, enum dma_data_direction dir)
 {
-	phys_addr_t paddr;
+	dma_cache_maint_page(paddr, size, dir, dmac_map_area);
 
-	dma_cache_maint_page(page, off, size, dir, dmac_map_area);
-
-	paddr = page_to_phys(page) + off;
 	if (dir == DMA_FROM_DEVICE) {
 		outer_inv_range(paddr, paddr + size);
 	} else {

Review

Quoted 5 lines

(quoted diff) +2 -0

+	if (!dev->dma_coherent &&
+	    !(attrs & (DMA_ATTR_SKIP_CPU_SYNC | DMA_ATTR_MMIO))) {
 		page = phys_to_page(iommu_iova_to_phys(mapping->domain, iova));
 		__dma_page_dev_to_cpu(page, offset, size, dir);

Same treatment here..

Looks Ok though, I didn't notice any pitfalls

Reviewed-by Jason Gunthorpe <[email protected]>

Jason

Jason Gunthorpe

2025-09-17 11:32 UTC #8 lore raw

Re: [PATCH v3 3/4] ARM: dma-mapping: Switch to physical address mapping callbacks

Quoted 10 lines · Leon Romanovsky

Quoted 7 lines · Jason Gunthorpe
Review
Quoted 3 lines · Leon Romanovsky
(quoted diff) +3 -0
+	if (!dev->dma_coherent &&
+	    !(attrs & (DMA_ATTR_SKIP_CPU_SYNC | DMA_ATTR_MMIO)))
+		__dma_page_cpu_to_dev(phys_to_page(phys), offset, size, dir);
I'd keep going and get rid of the page here too, maybe as a second patch in this series:
Thanks, it is always unclear how far to go with cleanups.

IMHO to maximally support what Matthew is working on I'd remove all the struct page things and prefer the pfn/phys variations from the MM side.

After this the only thing left is the kmap_high_get(), and I'm not sure what becomes of WANT_PAGE_VIRTUAL in a memdesc world..

Jason

Leon Romanovsky

author

2025-09-17 13:41 UTC #9 lore raw

Re: [PATCH v3 3/4] ARM: dma-mapping: Switch to physical address mapping callbacks

Quoted 10 lines · Jason Gunthorpe

Quoted 9 lines · Leon Romanovsky
Quoted 7 lines · Jason Gunthorpe
Review
Quoted 3 lines · Leon Romanovsky
(quoted diff) +3 -0
+	if (!dev->dma_coherent &&
+	    !(attrs & (DMA_ATTR_SKIP_CPU_SYNC | DMA_ATTR_MMIO)))
+		__dma_page_cpu_to_dev(phys_to_page(phys), offset, size, dir);
I'd keep going and get rid of the page here too, maybe as a second patch in this series:

Quoted 5 lines

Quoted 1 line
Thanks, it is always unclear how far to go with cleanups.
IMHO to maximally support what Matthew is working on I'd remove all the struct page things and prefer the pfn/phys variations from the MM side.

ok, my patches can be found here:
https://git.kernel.org/pub/scm/linux/kernel/git/leon/linux-rdma.git/log/?h=dmabuf-vfio

I converted "struct page" path from all archs with .map_page.

Thanks

Jason Gunthorpe

2025-09-17 13:58 UTC #10 lore raw

Re: [PATCH v3 3/4] ARM: dma-mapping: Switch to physical address mapping callbacks

Quoted 13 lines · Leon Romanovsky

Quoted 12 lines · Jason Gunthorpe
Quoted 10 lines · Leon Romanovsky
Quoted 7 lines · Jason Gunthorpe
Review
Quoted 3 lines · Leon Romanovsky
(quoted diff) +3 -0
+	if (!dev->dma_coherent &&
+	    !(attrs & (DMA_ATTR_SKIP_CPU_SYNC | DMA_ATTR_MMIO)))
+		__dma_page_cpu_to_dev(phys_to_page(phys), offset, size, dir);
I'd keep going and get rid of the page here too, maybe as a second patch in this series:
Thanks, it is always unclear how far to go with cleanups.

Quoted 8 lines

Quoted 3 lines
IMHO to maximally support what Matthew is working on I'd remove all the struct page things and prefer the pfn/phys variations from the MM side.
ok, my patches can be found here:
https://git.kernel.org/pub/scm/linux/kernel/git/leon/linux-rdma.git/log/?h=dmabuf-vfio
I converted "struct page" path from all archs with .map_page.

Yeah, that's right, I would split this into a few series

This group is all fairly trivial stuff that didn't use the struct page or virt address at all:

MIPS/jazzdma
x86
vdpa
xen: swiotlb: (though make a phys_to_xen_pfn() macro)

I'd also drop the ATTR_MMIO checks from x86 and jazzdma since the code is obviously fine with any phys.

vdpa can support ATTR_MMIO with the same iommu prot MMIO adjustment, that would be a good additional patch.

Then there are more conversions that don't use struct page or va but are not so trivial

alpha
parisc

Then the last ones sparc/power are quite complicated and do use __va/etc.

Jason

Leon Romanovsky

author

2025-09-17 18:46 UTC #11 lore raw

Re: [PATCH v3 3/4] ARM: dma-mapping: Switch to physical address mapping callbacks

Quoted 28 lines · Jason Gunthorpe

Quoted 13 lines · Leon Romanovsky
Quoted 12 lines · Jason Gunthorpe
Quoted 10 lines · Leon Romanovsky
Quoted 7 lines · Jason Gunthorpe
Review
Quoted 3 lines · Leon Romanovsky
(quoted diff) +3 -0
+	if (!dev->dma_coherent &&
+	    !(attrs & (DMA_ATTR_SKIP_CPU_SYNC | DMA_ATTR_MMIO)))
+		__dma_page_cpu_to_dev(phys_to_page(phys), offset, size, dir);
I'd keep going and get rid of the page here too, maybe as a second patch in this series:
Thanks, it is always unclear how far to go with cleanups.
Quoted 8 lines
Quoted 3 lines
IMHO to maximally support what Matthew is working on I'd remove all the struct page things and prefer the pfn/phys variations from the MM side.
ok, my patches can be found here:
https://git.kernel.org/pub/scm/linux/kernel/git/leon/linux-rdma.git/log/?h=dmabuf-vfio
I converted "struct page" path from all archs with .map_page.
Yeah, that's right, I would split this into a few series
This group is all fairly trivial stuff that didn't use the struct page or virt address at all:

Quoted 7 lines

MIPS/jazzdma
x86
vdpa
xen: swiotlb: (though make a phys_to_xen_pfn() macro)
I'd also drop the ATTR_MMIO checks from x86 and jazzdma since the code is obviously fine with any phys.

It is not trivial as it sounds.

Let's take as an example MIPS/jazzdma, should we call to vdma_alloc() in DMA_ATTR_MMIO case? or directly return "phys" as it is done in other architectures?

Quoted 3 lines

vdpa can support ATTR_MMIO with the same iommu prot MMIO adjustment, that would be a good additional patch.

There is "bounce page" in VDPA, it sounds like swiotlb to me, which doesn't work with MMIO.

Quoted 12 lines

Then there are more conversions that don't use struct page or va but are not so trivial
alpha
parisc
Then the last ones sparc/power are quite complicated and do use __va/etc.
Jason

Jason Gunthorpe

2025-09-17 19:08 UTC #12 lore raw

Re: [PATCH v3 3/4] ARM: dma-mapping: Switch to physical address mapping callbacks

Quoted 2 lines · Leon Romanovsky

Let's take as an example MIPS/jazzdma, should we call to vdma_alloc() in DMA_ATTR_MMIO case?

Yes, that is right, this is an iommu, so it has to be programmed.

Quoted 2 lines

or directly return "phys" as it is done in other
architectures?

Which?

Quoted 5 lines

Quoted 2 lines
vdpa can support ATTR_MMIO with the same iommu prot MMIO adjustment, that would be a good additional patch.
There is "bounce page" in VDPA, it sounds like swiotlb to me, which doesn't work with MMIO.

Oh OK, it eventually does call pfn_to_page().. This isn't the case I thought it was, there is another one that just routes to iommu_map

Jason

Leon Romanovsky

author

2025-09-17 19:46 UTC #13 lore raw

Re: [PATCH v3 3/4] ARM: dma-mapping: Switch to physical address mapping callbacks

Quoted 10 lines · Jason Gunthorpe

Quoted 2 lines · Leon Romanovsky
Let's take as an example MIPS/jazzdma, should we call to vdma_alloc() in DMA_ATTR_MMIO case?
Yes, that is right, this is an iommu, so it has to be programmed.
Quoted 2 lines
or directly return "phys" as it is done in other
architectures?
Which?

XEN does that.

https://git.kernel.org/pub/scm/linux/kernel/git/mszyprowski/linux.git/tree/drivers/xen/swiotlb-xen.c?h=dma-mapping-for-next#n395

static dma_addr_t xen_swiotlb_direct_map_resource(struct device *dev,
phys_addr_t paddr, size_t size, enum dma_data_direction dir, unsigned long attrs)

{
	dma_addr_t dma_addr = paddr;

	if (unlikely(!dma_capable(dev, dma_addr, size, false))) {
		dev_err_once(dev,
			     "DMA addr %pad+%zu overflow (mask %llx, bus limit %llx).\n",
			     &dma_addr, size, *dev->dma_mask, dev->bus_dma_limit);
		WARN_ON_ONCE(1);
		return DMA_MAPPING_ERROR;
	}

	return dma_addr;
}

Quoted 12 lines

Quoted 6 lines
Quoted 3 lines
vdpa can support ATTR_MMIO with the same iommu prot MMIO adjustment, that would be a good additional patch.
There is "bounce page" in VDPA, it sounds like swiotlb to me, which doesn't work with MMIO.
Oh OK, it eventually does call pfn_to_page().. This isn't the case I thought it was, there is another one that just routes to iommu_map
Jason