arm64: dma-mapping: implement dma_get_sgtable()

PATCH New Robin Murphy 2 participants 4 messages Jul 17, 2015 – Jul 20, 2015 arm64

Share & tools

Robin Murphy

author

2015-07-17 15:58 UTC #0 lore raw

PATCH 1/2 arm64: dma-mapping: implement dma_get_sgtable()

The default dma_common_get_sgtable() implementation relies on the CPU address of the buffer being a regular lowmem address. This is not always the case on arm64, since allocations from the various DMA pools may have remapped vmalloc addresses, rendering the use of virt_to_page() invalid.

Fix this by providing our own implementation based on the fact that we can safely derive a physical address from the DMA address in both cases.

CC Jon Medhurst <[email protected]>

Signed-off-by Robin Murphy <[email protected]>

---
 arch/arm64/mm/dma-mapping.c | 14 ++++++++++++++
 1 file changed, 14 insertions(+)

arch/arm64/mm/dma-mapping.c +14 -0

index d16a1ce..4b9b600 100644
@@ -337,10 +337,24 @@ static int __swiotlb_mmap(struct device *dev,
 	return __dma_common_mmap(dev, vma, cpu_addr, dma_addr, size);
 }
 
+int __swiotlb_get_sgtable(struct device *dev, struct sg_table *sgt,
+			  void *cpu_addr, dma_addr_t handle, size_t size,
+			  struct dma_attrs *attrs)
+{
+	int ret = sg_alloc_table(sgt, 1, GFP_KERNEL);
+
+	if (!ret)
+		sg_set_page(sgt->sgl, phys_to_page(dma_to_phys(dev, handle)),
+			    PAGE_ALIGN(size), 0);
+
+	return ret;
+}
+
 static struct dma_map_ops swiotlb_dma_ops = {
 	.alloc = __dma_alloc,
 	.free = __dma_free,
 	.mmap = __swiotlb_mmap,
+	.get_sgtable = __swiotlb_get_sgtable,
 	.map_page = __swiotlb_map_page,
 	.unmap_page = __swiotlb_unmap_page,
 	.map_sg = __swiotlb_map_sg_attrs,

Signature · 1 line

1.9.1

Robin Murphy

author

2015-07-17 15:58 UTC #1 lore raw

PATCH 2/2 arm64: dma-mapping: consolidate __swiotlb_mmap()

Since commit 9d3bfbb4df58 ("arm64: Combine coherent and non-coherent swiotlb dma_ops"), __dma_common_mmap() is no longer shared between two callers, so roll it into the remaining one.

Signed-off-by Robin Murphy <[email protected]>

---
 arch/arm64/mm/dma-mapping.c | 19 ++++++-------------
 1 file changed, 6 insertions(+), 13 deletions(-)

arch/arm64/mm/dma-mapping.c +6 ~~-13~~

index 4b9b600..07c6976 100644
@@ -303,9 +303,9 @@ static void __swiotlb_sync_sg_for_device(struct device *dev,
 				       sg->length, dir);
 }
 
-/* vma->vm_page_prot must be set appropriately before calling this function */
-static int __dma_common_mmap(struct device *dev, struct vm_area_struct *vma,
-			     void *cpu_addr, dma_addr_t dma_addr, size_t size)
+static int __swiotlb_mmap(struct device *dev, struct vm_area_struct *vma,
+			  void *cpu_addr, dma_addr_t dma_addr, size_t size,
+			  struct dma_attrs *attrs)
 {
 	int ret = -ENXIO;
 	unsigned long nr_vma_pages = (vma->vm_end - vma->vm_start) >>

@@ -314,6 +314,9 @@ static int __dma_common_mmap(struct device *dev, struct vm_area_struct *vma,
 	unsigned long pfn = dma_to_phys(dev, dma_addr) >> PAGE_SHIFT;
 	unsigned long off = vma->vm_pgoff;
 
+	vma->vm_page_prot = __get_dma_pgprot(attrs, vma->vm_page_prot,
+					     is_device_dma_coherent(dev));
+
 	if (dma_mmap_from_coherent(dev, vma, cpu_addr, size, &ret))
 		return ret;

@@ -327,16 +330,6 @@ static int __dma_common_mmap(struct device *dev, struct vm_area_struct *vma,
 	return ret;
 }
 
-static int __swiotlb_mmap(struct device *dev,
-			  struct vm_area_struct *vma,
-			  void *cpu_addr, dma_addr_t dma_addr, size_t size,
-			  struct dma_attrs *attrs)
-{
-	vma->vm_page_prot = __get_dma_pgprot(attrs, vma->vm_page_prot,
-					     is_device_dma_coherent(dev));
-	return __dma_common_mmap(dev, vma, cpu_addr, dma_addr, size);
-}
-
 int __swiotlb_get_sgtable(struct device *dev, struct sg_table *sgt,
 			  void *cpu_addr, dma_addr_t handle, size_t size,
 			  struct dma_attrs *attrs)

Signature · 1 line

1.9.1

Will Deacon

2015-07-20 16:36 UTC #2 lore raw

PATCH 1/2 arm64: dma-mapping: implement dma_get_sgtable()

In reply to Robin Murphy

Quoted 24 lines · Robin Murphy

The default dma_common_get_sgtable() implementation relies on the CPU address of the buffer being a regular lowmem address. This is not always the case on arm64, since allocations from the various DMA pools may have remapped vmalloc addresses, rendering the use of virt_to_page() invalid.
Fix this by providing our own implementation based on the fact that we can safely derive a physical address from the DMA address in both cases.
CC Jon Medhurst <[email protected]>
Signed-off-by Robin Murphy <[email protected]>
---
 arch/arm64/mm/dma-mapping.c | 14 ++++++++++++++
 1 file changed, 14 insertions(+)
arch/arm64/mm/dma-mapping.c +2 -0
index d16a1ce..4b9b600 100644
@@ -337,10 +337,24 @@ static int __swiotlb_mmap(struct device *dev,
 	return __dma_common_mmap(dev, vma, cpu_addr, dma_addr, size);
 }
 
+int __swiotlb_get_sgtable(struct device *dev, struct sg_table *sgt,
+			  void *cpu_addr, dma_addr_t handle, size_t size,

arch/arm64/mm/dma-mapping.c static int __swiotlb_mmap(struct device *dev, · line 351

Quoted 10 lines

arch/arm64/mm/dma-mapping.c +10 -0

+			  struct dma_attrs *attrs)
+{
+	int ret = sg_alloc_table(sgt, 1, GFP_KERNEL);
+
+	if (!ret)
+		sg_set_page(sgt->sgl, phys_to_page(dma_to_phys(dev, handle)),
+			    PAGE_ALIGN(size), 0);
+
+	return ret;
+}

Any reason not to do this in dma_common_get_sgtable?

Will

Robin Murphy

author

2015-07-20 17:00 UTC #3 lore raw

PATCH 1/2 arm64: dma-mapping: implement dma_get_sgtable()

Hi Will,

Quoted 23 lines · Will Deacon

Quoted 22 lines · Robin Murphy
The default dma_common_get_sgtable() implementation relies on the CPU address of the buffer being a regular lowmem address. This is not always the case on arm64, since allocations from the various DMA pools may have remapped vmalloc addresses, rendering the use of virt_to_page() invalid.
Fix this by providing our own implementation based on the fact that we can safely derive a physical address from the DMA address in both cases.
CC Jon Medhurst <[email protected]>
Signed-off-by Robin Murphy <[email protected]>
---
  arch/arm64/mm/dma-mapping.c | 14 ++++++++++++++
  1 file changed, 14 insertions(+)
arch/arm64/mm/dma-mapping.c +0 -0
index d16a1ce..4b9b600 100644
@@ -337,10 +337,24 @@ static int __swiotlb_mmap(struct device *dev,
  	return __dma_common_mmap(dev, vma, cpu_addr, dma_addr, size);
  }
 

Quoted 14 lines

arch/arm64/mm/dma-mapping.c static int __swiotlb_mmap(struct device *dev, · line 351

Quoted 12 lines

arch/arm64/mm/dma-mapping.c +12 -0

+int __swiotlb_get_sgtable(struct device *dev, struct sg_table *sgt,
+			  void *cpu_addr, dma_addr_t handle, size_t size,
+			  struct dma_attrs *attrs)
+{
+	int ret = sg_alloc_table(sgt, 1, GFP_KERNEL);
+
+	if (!ret)
+		sg_set_page(sgt->sgl, phys_to_page(dma_to_phys(dev, handle)),
+			    PAGE_ALIGN(size), 0);
+
+	return ret;
+}

Any reason not to do this in dma_common_get_sgtable?

Summarising the discussion over at [1], most architectures seem to depend on dma_common_get_sgtable, but only a handful implement dma_to_phys (plus this approach seems to match the original intent). There doesn't seem to be a nice solution for doing this in common code without a big cross-architecture patch, and it's somewhat questionable how widely this is actually needed.

Robin.

[1]:http://thread.gmane.org/gmane.linux.kernel/1998795

Quoted 3 lines

Will