[PATCH] iommu/amd: Propagate IOVA allocation failure

[PATCH] iommu/amd: Propagate IOVA allocation failure

[Robin Murphy]

Unlike the old allocator, alloc_iova_fast() will return 0 if it failed
to allocate a PFN. Since the callers of dma_ops_alloc_iova() would end
up treating that as a valid address, translate it to the DMA_ERROR_CODE
that they would expect.

Fixes: 256e4621c21a ("iommu/amd: Make use of the generic IOVA allocator")
Signed-off-by: Robin Murphy <robin.murphy-5wv7dgnIgG8@public.gmane.org>
---

Just something I spotted whilst comparing dma_map_page() callchains...

 drivers/iommu/amd_iommu.c | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/drivers/iommu/amd_iommu.c b/drivers/iommu/amd_iommu.c
index 63cacf5d6cf2..489dc302899e 100644
--- a/drivers/iommu/amd_iommu.c
+++ b/drivers/iommu/amd_iommu.c
@@ -1555,6 +1555,9 @@ static unsigned long dma_ops_alloc_iova(struct device *dev,
 	if (!pfn)
 		pfn = alloc_iova_fast(&dma_dom->iovad, pages, IOVA_PFN(dma_mask));
 
+	if (!pfn)
+		return DMA_ERROR_CODE;
+
 	return (pfn << PAGE_SHIFT);
 }
 
-- 
2.12.2.dirty

Re: [PATCH] iommu/amd: Propagate IOVA allocation failure

[Joerg Roedel]

On Fri, May 26, 2017 at 07:31:26PM +0100, Robin Murphy wrote:
> Unlike the old allocator, alloc_iova_fast() will return 0 if it failed
> to allocate a PFN. Since the callers of dma_ops_alloc_iova() would end
> up treating that as a valid address, translate it to the DMA_ERROR_CODE
> that they would expect.
> 
> Fixes: 256e4621c21a ("iommu/amd: Make use of the generic IOVA allocator")
> Signed-off-by: Robin Murphy <robin.murphy-5wv7dgnIgG8@public.gmane.org>
> ---
> 
> Just something I spotted whilst comparing dma_map_page() callchains...
> 
>  drivers/iommu/amd_iommu.c | 3 +++
>  1 file changed, 3 insertions(+)
> 
> diff --git a/drivers/iommu/amd_iommu.c b/drivers/iommu/amd_iommu.c
> index 63cacf5d6cf2..489dc302899e 100644
> --- a/drivers/iommu/amd_iommu.c
> +++ b/drivers/iommu/amd_iommu.c
> @@ -1555,6 +1555,9 @@ static unsigned long dma_ops_alloc_iova(struct device *dev,
>  	if (!pfn)
>  		pfn = alloc_iova_fast(&dma_dom->iovad, pages, IOVA_PFN(dma_mask));
>  
> +	if (!pfn)
> +		return DMA_ERROR_CODE;
> +

That shouldn't make a difference on x86 because the DMA_ERROR_CODE is 0
as well.


	Joerg

Re: [PATCH] iommu/amd: Propagate IOVA allocation failure

[Robin Murphy]

On 30/05/17 10:48, Joerg Roedel wrote:
> On Fri, May 26, 2017 at 07:31:26PM +0100, Robin Murphy wrote:
>> Unlike the old allocator, alloc_iova_fast() will return 0 if it failed
>> to allocate a PFN. Since the callers of dma_ops_alloc_iova() would end
>> up treating that as a valid address, translate it to the DMA_ERROR_CODE
>> that they would expect.
>>
>> Fixes: 256e4621c21a ("iommu/amd: Make use of the generic IOVA allocator")
>> Signed-off-by: Robin Murphy <robin.murphy-5wv7dgnIgG8@public.gmane.org>
>> ---
>>
>> Just something I spotted whilst comparing dma_map_page() callchains...
>>
>>  drivers/iommu/amd_iommu.c | 3 +++
>>  1 file changed, 3 insertions(+)
>>
>> diff --git a/drivers/iommu/amd_iommu.c b/drivers/iommu/amd_iommu.c
>> index 63cacf5d6cf2..489dc302899e 100644
>> --- a/drivers/iommu/amd_iommu.c
>> +++ b/drivers/iommu/amd_iommu.c
>> @@ -1555,6 +1555,9 @@ static unsigned long dma_ops_alloc_iova(struct device *dev,
>>  	if (!pfn)
>>  		pfn = alloc_iova_fast(&dma_dom->iovad, pages, IOVA_PFN(dma_mask));
>>  
>> +	if (!pfn)
>> +		return DMA_ERROR_CODE;
>> +
> 
> That shouldn't make a difference on x86 because the DMA_ERROR_CODE is 0
> as well.

Ha, now that I did manage to overlook. Oh well, never mind then :)

Thanks,
Robin.

> 
> 
> 	Joerg
>