[PATCH v2] base: dma-mapping: Postpone cpu addr translation on mmap()

[PATCH v2] base: dma-mapping: Postpone cpu addr translation on mmap()

[Jacopo Mondi]

Postpone calling virt_to_page() translation on memory locations not
guaranteed to be backed by a struct page. Try first to map memory from
device's coherent memory pool, then perform translation if that fails.

On some architectures, specifically SH when configured with SPARSEMEM
memory model, assuming a struct page is always assigned to a memory
address lead to unexpected hangs during the virtual to page address
translation. This patch fixes that specific issue but applies in the
general case too.

Suggested-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Jacopo Mondi <jacopo+renesas@jmondi.org>

---

It has now been clarified this patch does not resolve the issue, but only
mitigate it on platforms where dma_mmap_from_dev_coherent() succeeds and
delay page_to_pfn() faulty conversion.

A suggested proper solution would be not relying on dma_common_mmap() but
require all platforms to implement an mmap methods known to work, as noted
by Christoph in v1 review.

v1 -> v2:
- Save the 'pfn' temp variable performing the page_to_pfn() conversion in the
  remap_pfn_range() function call as suggested by Christoph.

---
 drivers/base/dma-mapping.c | 6 ++----
 1 file changed, 2 insertions(+), 4 deletions(-)

diff --git a/drivers/base/dma-mapping.c b/drivers/base/dma-mapping.c
index 3b11835..d82566d 100644
--- a/drivers/base/dma-mapping.c
+++ b/drivers/base/dma-mapping.c
@@ -226,7 +226,6 @@ int dma_common_mmap(struct device *dev, struct vm_area_struct *vma,
 #ifndef CONFIG_ARCH_NO_COHERENT_DMA_MMAP
 	unsigned long user_count = vma_pages(vma);
 	unsigned long count = PAGE_ALIGN(size) >> PAGE_SHIFT;
-	unsigned long pfn = page_to_pfn(virt_to_page(cpu_addr));
 	unsigned long off = vma->vm_pgoff;

 	vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
@@ -234,12 +233,11 @@ int dma_common_mmap(struct device *dev, struct vm_area_struct *vma,
 	if (dma_mmap_from_dev_coherent(dev, vma, cpu_addr, size, &ret))
 		return ret;

-	if (off < count && user_count <= (count - off)) {
+	if (off < count && user_count <= (count - off))
 		ret = remap_pfn_range(vma, vma->vm_start,
-				      pfn + off,
+				      page_to_pfn(virt_to_page(cpu_addr)) + off,
 				      user_count << PAGE_SHIFT,
 				      vma->vm_page_prot);
-	}
 #endif	/* !CONFIG_ARCH_NO_COHERENT_DMA_MMAP */

 	return ret;
--
2.7.4

Re: [PATCH v2] base: dma-mapping: Postpone cpu addr translation on mmap()

[Robin Murphy]

On 13/04/18 18:25, Jacopo Mondi wrote:
> Postpone calling virt_to_page() translation on memory locations not
> guaranteed to be backed by a struct page. Try first to map memory from
> device's coherent memory pool, then perform translation if that fails.
> 
> On some architectures, specifically SH when configured with SPARSEMEM
> memory model, assuming a struct page is always assigned to a memory
> address lead to unexpected hangs during the virtual to page address
> translation. This patch fixes that specific issue but applies in the
> general case too.

Reviewed-by: Robin Murphy <robin.murphy@arm.com>

> Suggested-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
> Signed-off-by: Jacopo Mondi <jacopo+renesas@jmondi.org>
> 
> ---
> 
> It has now been clarified this patch does not resolve the issue, but only
> mitigate it on platforms where dma_mmap_from_dev_coherent() succeeds and
> delay page_to_pfn() faulty conversion.
> 
> A suggested proper solution would be not relying on dma_common_mmap() but
> require all platforms to implement an mmap methods known to work, as noted
> by Christoph in v1 review.

Note that that "proper solution" should still involve having 
dma_common_mmap() since we certainly don't want an explosion of code 
duplication. It just means that architectures that do use it should be 
defining their dma_map_ops with an explicit ".mmap = dma_common_mmap" 
instead of relying on dma_mmap_attrs() calling it by default. Thus the 
more architectures this implementation *is* definitely safe for, the 
better :)

Robin.

> v1 -> v2:
> - Save the 'pfn' temp variable performing the page_to_pfn() conversion in the
>    remap_pfn_range() function call as suggested by Christoph.
> 
> ---
>   drivers/base/dma-mapping.c | 6 ++----
>   1 file changed, 2 insertions(+), 4 deletions(-)
> 
> diff --git a/drivers/base/dma-mapping.c b/drivers/base/dma-mapping.c
> index 3b11835..d82566d 100644
> --- a/drivers/base/dma-mapping.c
> +++ b/drivers/base/dma-mapping.c
> @@ -226,7 +226,6 @@ int dma_common_mmap(struct device *dev, struct vm_area_struct *vma,
>   #ifndef CONFIG_ARCH_NO_COHERENT_DMA_MMAP
>   	unsigned long user_count = vma_pages(vma);
>   	unsigned long count = PAGE_ALIGN(size) >> PAGE_SHIFT;
> -	unsigned long pfn = page_to_pfn(virt_to_page(cpu_addr));
>   	unsigned long off = vma->vm_pgoff;
> 
>   	vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
> @@ -234,12 +233,11 @@ int dma_common_mmap(struct device *dev, struct vm_area_struct *vma,
>   	if (dma_mmap_from_dev_coherent(dev, vma, cpu_addr, size, &ret))
>   		return ret;
> 
> -	if (off < count && user_count <= (count - off)) {
> +	if (off < count && user_count <= (count - off))
>   		ret = remap_pfn_range(vma, vma->vm_start,
> -				      pfn + off,
> +				      page_to_pfn(virt_to_page(cpu_addr)) + off,
>   				      user_count << PAGE_SHIFT,
>   				      vma->vm_page_prot);
> -	}
>   #endif	/* !CONFIG_ARCH_NO_COHERENT_DMA_MMAP */
> 
>   	return ret;
> --
> 2.7.4
> 

Re: [PATCH v2] base: dma-mapping: Postpone cpu addr translation on mmap()

[Christoph Hellwig]

Thanks,

applied to the dma-mapping tree for 4.17.