[PATCH] ARM64: dma-mapping: fix IOMMU-based dma_mmap_atts() implementation

[PATCH] ARM64: dma-mapping: fix IOMMU-based dma_mmap_atts() implementation

[Marek Szyprowski]

IOMMU-based DMA-mapping implementation doesn't support
dma_declare_coherent() based regions, so there is no point to call
dma_mmap_from_coherent() from __iommu_mmap_atts(). This patch removes
useless code.

Signed-off-by: Marek Szyprowski <m.szyprowski@samsung.com>
---
 arch/arm64/mm/dma-mapping.c | 5 -----
 1 file changed, 5 deletions(-)

diff --git a/arch/arm64/mm/dma-mapping.c b/arch/arm64/mm/dma-mapping.c
index 331c4ca..573b1e6 100644
--- a/arch/arm64/mm/dma-mapping.c
+++ b/arch/arm64/mm/dma-mapping.c
@@ -636,14 +636,9 @@ static int __iommu_mmap_attrs(struct device *dev, struct vm_area_struct *vma,
 			      struct dma_attrs *attrs)
 {
 	struct vm_struct *area;
-	int ret;
 
 	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;
-
 	area = find_vm_area(cpu_addr);
 	if (WARN_ON(!area || !area->pages))
 		return -ENXIO;
-- 
1.9.2

[PATCH] ARM64: dma-mapping: fix IOMMU-based dma_mmap_atts() implementation

[Robin Murphy]

Hi Marek,

On 16/02/16 14:14, Marek Szyprowski wrote:
> IOMMU-based DMA-mapping implementation doesn't support
> dma_declare_coherent() based regions, so there is no point to call
> dma_mmap_from_coherent() from __iommu_mmap_atts(). This patch removes
> useless code.

I'm not sure that's the case - AFAICS a device could still have its own 
local memory despite being behind an IOMMU relative to the rest of the 
system. If it uses dma_declare_coherent() for that, the .alloc and .free 
callbacks here might be bypassed by dma_{alloc,free}_attrs() 
short-circuiting into dma_{alloc,release}_from_coherent(), but in the 
case of dma_mmap_attrs() it hits the .mmap callback first and only falls 
back to the from_coherent check in dma_common_mmap() if that isn't 
implemented.

Maybe the right thing to do would be to move this check (with its 
associated guards) out from dma_common_mmap() up to dma_mmap_attrs() to 
be consistent with alloc/free; so that it doesn't need to be 
reimplemented all over arch/{arm,arm64,mips} at all?

Robin.

> Signed-off-by: Marek Szyprowski <m.szyprowski@samsung.com>
> ---
>   arch/arm64/mm/dma-mapping.c | 5 -----
>   1 file changed, 5 deletions(-)
>
> diff --git a/arch/arm64/mm/dma-mapping.c b/arch/arm64/mm/dma-mapping.c
> index 331c4ca..573b1e6 100644
> --- a/arch/arm64/mm/dma-mapping.c
> +++ b/arch/arm64/mm/dma-mapping.c
> @@ -636,14 +636,9 @@ static int __iommu_mmap_attrs(struct device *dev, struct vm_area_struct *vma,
>   			      struct dma_attrs *attrs)
>   {
>   	struct vm_struct *area;
> -	int ret;
>
>   	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;
> -
>   	area = find_vm_area(cpu_addr);
>   	if (WARN_ON(!area || !area->pages))
>   		return -ENXIO;
>

[PATCH] ARM64: dma-mapping: fix IOMMU-based dma_mmap_atts() implementation

[Marek Szyprowski]

Hello,

On 2016-02-16 16:02, Robin Murphy wrote:
> Hi Marek,
>
> On 16/02/16 14:14, Marek Szyprowski wrote:
>> IOMMU-based DMA-mapping implementation doesn't support
>> dma_declare_coherent() based regions, so there is no point to call
>> dma_mmap_from_coherent() from __iommu_mmap_atts(). This patch removes
>> useless code.
>
> I'm not sure that's the case - AFAICS a device could still have its 
> own local memory despite being behind an IOMMU relative to the rest of 
> the system. If it uses dma_declare_coherent() for that, the .alloc and 
> .free callbacks here might be bypassed by dma_{alloc,free}_attrs() 
> short-circuiting into dma_{alloc,release}_from_coherent(), but in the 
> case of dma_mmap_attrs() it hits the .mmap callback first and only 
> falls back to the from_coherent check in dma_common_mmap() if that 
> isn't implemented.

Okay, I've missed that alloc/free_from_coherent is handled in generic
include/linux/dma-mapping.h. It is still a bit strange case with no real
user, but if this code is already there, then let it be.

>
> Maybe the right thing to do would be to move this check (with its 
> associated guards) out from dma_common_mmap() up to dma_mmap_attrs() 
> to be consistent with alloc/free; so that it doesn't need to be 
> reimplemented all over arch/{arm,arm64,mips} at all?
>

That might be a good idea, although I'm not sure how portable is
pgprot_dmacoherent(), which will be needed by such common code.

Please ignore my patch then.

> Robin.
>
>> Signed-off-by: Marek Szyprowski <m.szyprowski@samsung.com>
>> ---
>>   arch/arm64/mm/dma-mapping.c | 5 -----
>>   1 file changed, 5 deletions(-)
>>
>> diff --git a/arch/arm64/mm/dma-mapping.c b/arch/arm64/mm/dma-mapping.c
>> index 331c4ca..573b1e6 100644
>> --- a/arch/arm64/mm/dma-mapping.c
>> +++ b/arch/arm64/mm/dma-mapping.c
>> @@ -636,14 +636,9 @@ static int __iommu_mmap_attrs(struct device 
>> *dev, struct vm_area_struct *vma,
>>                     struct dma_attrs *attrs)
>>   {
>>       struct vm_struct *area;
>> -    int ret;
>>
>>       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;
>> -
>>       area = find_vm_area(cpu_addr);
>>       if (WARN_ON(!area || !area->pages))
>>           return -ENXIO;
>

Best regards
-- 
Marek Szyprowski, PhD
Samsung R&D Institute Poland