[PATCH v2] ARM: mm: Do not invoke OOM for higher order IOMMU DMA allocations

[PATCH v2] ARM: mm: Do not invoke OOM for higher order IOMMU DMA allocations

[Tomasz Figa]

IOMMU should be able to use single pages as well as bigger blocks, so if
higher order allocations fail, we should not affect state of the system,
with events such as OOM killer, but rather fall back to order 0
allocations.

This patch changes the behavior of ARM IOMMU DMA allocator to use
__GFP_NORETRY, which bypasses OOM invocation, for orders higher than
zero and, only if that fails, fall back to normal order 0 allocation
which might invoke OOM killer.

Signed-off-by: Tomasz Figa <tfiga@chromium.org>
---
 arch/arm/mm/dma-mapping.c | 27 +++++++++++++++++++++------
 1 file changed, 21 insertions(+), 6 deletions(-)

Changes since v1:
(https://patchwork.kernel.org/patch/6015921/)
 - do not clear __GFP_NORETRY, as it might come from the caller,
 - s/positive order/order higher than 0/.

diff --git a/arch/arm/mm/dma-mapping.c b/arch/arm/mm/dma-mapping.c
index 83cd5ac..3f1ac51 100644
--- a/arch/arm/mm/dma-mapping.c
+++ b/arch/arm/mm/dma-mapping.c
@@ -1150,13 +1150,28 @@ static struct page **__iommu_alloc_buffer(struct device *dev, size_t size,
 	gfp |= __GFP_NOWARN | __GFP_HIGHMEM;
 
 	while (count) {
-		int j, order = __fls(count);
+		int j, order;
+
+		for (order = __fls(count); order > 0; --order) {
+			/*
+			 * We do not want OOM killer to be invoked as long
+			 * as we can fall back to single pages, so we force
+			 * __GFP_NORETRY for orders higher than zero.
+			 */
+			pages[i] = alloc_pages(gfp | __GFP_NORETRY, order);
+			if (pages[i])
+				break;
+		}
 
-		pages[i] = alloc_pages(gfp, order);
-		while (!pages[i] && order)
-			pages[i] = alloc_pages(gfp, --order);
-		if (!pages[i])
-			goto error;
+		if (!pages[i]) {
+			/*
+			 * Fall back to single page allocation.
+			 * Might invoke OOM killer as last resort.
+			 */
+			pages[i] = alloc_pages(gfp, 0);
+			if (!pages[i])
+				goto error;
+		}
 
 		if (order) {
 			split_page(pages[i], order);
-- 
2.2.0.rc0.207.ga3a616c

[PATCH v2] ARM: mm: Do not invoke OOM for higher order IOMMU DMA allocations

[Doug Anderson]

Tomasz,

On Mon, Mar 23, 2015 at 12:27 AM, Tomasz Figa <tfiga@chromium.org> wrote:
> IOMMU should be able to use single pages as well as bigger blocks, so if
> higher order allocations fail, we should not affect state of the system,
> with events such as OOM killer, but rather fall back to order 0
> allocations.
>
> This patch changes the behavior of ARM IOMMU DMA allocator to use
> __GFP_NORETRY, which bypasses OOM invocation, for orders higher than
> zero and, only if that fails, fall back to normal order 0 allocation
> which might invoke OOM killer.
>
> Signed-off-by: Tomasz Figa <tfiga@chromium.org>
> ---
>  arch/arm/mm/dma-mapping.c | 27 +++++++++++++++++++++------
>  1 file changed, 21 insertions(+), 6 deletions(-)

FWIW:

Reviewed-by: Doug Anderson <dianders@chromium.org>

[PATCH v2] ARM: mm: Do not invoke OOM for higher order IOMMU DMA allocations

[David Rientjes]

On Mon, 23 Mar 2015, Tomasz Figa wrote:

> IOMMU should be able to use single pages as well as bigger blocks, so if
> higher order allocations fail, we should not affect state of the system,
> with events such as OOM killer, but rather fall back to order 0
> allocations.
> 
> This patch changes the behavior of ARM IOMMU DMA allocator to use
> __GFP_NORETRY, which bypasses OOM invocation, for orders higher than
> zero and, only if that fails, fall back to normal order 0 allocation
> which might invoke OOM killer.
> 
> Signed-off-by: Tomasz Figa <tfiga@chromium.org>

Acked-by: David Rientjes <rientjes@google.com>

[PATCH v2] ARM: mm: Do not invoke OOM for higher order IOMMU DMA allocations

[Marek Szyprowski]

Hello,

On 2015-03-23 08:27, Tomasz Figa wrote:
> IOMMU should be able to use single pages as well as bigger blocks, so if
> higher order allocations fail, we should not affect state of the system,
> with events such as OOM killer, but rather fall back to order 0
> allocations.
>
> This patch changes the behavior of ARM IOMMU DMA allocator to use
> __GFP_NORETRY, which bypasses OOM invocation, for orders higher than
> zero and, only if that fails, fall back to normal order 0 allocation
> which might invoke OOM killer.
>
> Signed-off-by: Tomasz Figa <tfiga@chromium.org>

Acked-by: Marek Szyprowski <m.szyprowski@samsung.com>

> ---
>   arch/arm/mm/dma-mapping.c | 27 +++++++++++++++++++++------
>   1 file changed, 21 insertions(+), 6 deletions(-)
>
> Changes since v1:
> (https://patchwork.kernel.org/patch/6015921/)
>   - do not clear __GFP_NORETRY, as it might come from the caller,
>   - s/positive order/order higher than 0/.
>
> diff --git a/arch/arm/mm/dma-mapping.c b/arch/arm/mm/dma-mapping.c
> index 83cd5ac..3f1ac51 100644
> --- a/arch/arm/mm/dma-mapping.c
> +++ b/arch/arm/mm/dma-mapping.c
> @@ -1150,13 +1150,28 @@ static struct page **__iommu_alloc_buffer(struct device *dev, size_t size,
>   	gfp |= __GFP_NOWARN | __GFP_HIGHMEM;
>   
>   	while (count) {
> -		int j, order = __fls(count);
> +		int j, order;
> +
> +		for (order = __fls(count); order > 0; --order) {
> +			/*
> +			 * We do not want OOM killer to be invoked as long
> +			 * as we can fall back to single pages, so we force
> +			 * __GFP_NORETRY for orders higher than zero.
> +			 */
> +			pages[i] = alloc_pages(gfp | __GFP_NORETRY, order);
> +			if (pages[i])
> +				break;
> +		}
>   
> -		pages[i] = alloc_pages(gfp, order);
> -		while (!pages[i] && order)
> -			pages[i] = alloc_pages(gfp, --order);
> -		if (!pages[i])
> -			goto error;
> +		if (!pages[i]) {
> +			/*
> +			 * Fall back to single page allocation.
> +			 * Might invoke OOM killer as last resort.
> +			 */
> +			pages[i] = alloc_pages(gfp, 0);
> +			if (!pages[i])
> +				goto error;
> +		}
>   
>   		if (order) {
>   			split_page(pages[i], order);

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

[PATCH v2] ARM: mm: Do not invoke OOM for higher order IOMMU DMA allocations

[Ritesh Harjani]

Hi

On Mon, Mar 23, 2015 at 12:57 PM, Tomasz Figa <tfiga@chromium.org> wrote:
> IOMMU should be able to use single pages as well as bigger blocks, so if
> higher order allocations fail, we should not affect state of the system,
> with events such as OOM killer, but rather fall back to order 0
> allocations.
>
> This patch changes the behavior of ARM IOMMU DMA allocator to use
> __GFP_NORETRY, which bypasses OOM invocation, for orders higher than
> zero and, only if that fails, fall back to normal order 0 allocation
> which might invoke OOM killer.

Logical thing to do in IOMMU case :)
>
> Signed-off-by: Tomasz Figa <tfiga@chromium.org>
> ---
>  arch/arm/mm/dma-mapping.c | 27 +++++++++++++++++++++------
>  1 file changed, 21 insertions(+), 6 deletions(-)
>
> Changes since v1:
> (https://patchwork.kernel.org/patch/6015921/)
>  - do not clear __GFP_NORETRY, as it might come from the caller,
>  - s/positive order/order higher than 0/.
>
> diff --git a/arch/arm/mm/dma-mapping.c b/arch/arm/mm/dma-mapping.c
> index 83cd5ac..3f1ac51 100644
> --- a/arch/arm/mm/dma-mapping.c
> +++ b/arch/arm/mm/dma-mapping.c
> @@ -1150,13 +1150,28 @@ static struct page **__iommu_alloc_buffer(struct device *dev, size_t size,
>         gfp |= __GFP_NOWARN | __GFP_HIGHMEM;
>
>         while (count) {
> -               int j, order = __fls(count);
> +               int j, order;
> +
> +               for (order = __fls(count); order > 0; --order) {
> +                       /*
> +                        * We do not want OOM killer to be invoked as long
> +                        * as we can fall back to single pages, so we force
> +                        * __GFP_NORETRY for orders higher than zero.
> +                        */
> +                       pages[i] = alloc_pages(gfp | __GFP_NORETRY, order);
> +                       if (pages[i])
> +                               break;
> +               }
>
> -               pages[i] = alloc_pages(gfp, order);
> -               while (!pages[i] && order)
> -                       pages[i] = alloc_pages(gfp, --order);
> -               if (!pages[i])
> -                       goto error;
> +               if (!pages[i]) {
> +                       /*
> +                        * Fall back to single page allocation.
> +                        * Might invoke OOM killer as last resort.
> +                        */
> +                       pages[i] = alloc_pages(gfp, 0);
I think down the code in this while loop, i & count is being
calculated based on the "order" of allocation in the current
iteration.
Since value of order will be automatically 0 here if (!pages[i]) is
true then, why hard code order to value of 0 here.
Comment clearly says what this code is doing right?

I know it is just a minor thing. Don't know if it is relevant.

> +                       if (!pages[i])
> +                               goto error;
> +               }
>
>                 if (order) {
>                         split_page(pages[i], order);
> --
> 2.2.0.rc0.207.ga3a616c
>
>
> _______________________________________________
> linux-arm-kernel mailing list
> linux-arm-kernel at lists.infradead.org
> http://lists.infradead.org/mailman/listinfo/linux-arm-kernel


Thanks
Ritesh

[PATCH v2] ARM: mm: Do not invoke OOM for higher order IOMMU DMA allocations

[David Rientjes]

On Thu, 26 Mar 2015, Ritesh Harjani wrote:

> > diff --git a/arch/arm/mm/dma-mapping.c b/arch/arm/mm/dma-mapping.c
> > index 83cd5ac..3f1ac51 100644
> > --- a/arch/arm/mm/dma-mapping.c
> > +++ b/arch/arm/mm/dma-mapping.c
> > @@ -1150,13 +1150,28 @@ static struct page **__iommu_alloc_buffer(struct device *dev, size_t size,
> >         gfp |= __GFP_NOWARN | __GFP_HIGHMEM;
> >
> >         while (count) {
> > -               int j, order = __fls(count);
> > +               int j, order;
> > +
> > +               for (order = __fls(count); order > 0; --order) {
> > +                       /*
> > +                        * We do not want OOM killer to be invoked as long
> > +                        * as we can fall back to single pages, so we force
> > +                        * __GFP_NORETRY for orders higher than zero.
> > +                        */
> > +                       pages[i] = alloc_pages(gfp | __GFP_NORETRY, order);
> > +                       if (pages[i])
> > +                               break;
> > +               }
> >
> > -               pages[i] = alloc_pages(gfp, order);
> > -               while (!pages[i] && order)
> > -                       pages[i] = alloc_pages(gfp, --order);
> > -               if (!pages[i])
> > -                       goto error;
> > +               if (!pages[i]) {
> > +                       /*
> > +                        * Fall back to single page allocation.
> > +                        * Might invoke OOM killer as last resort.
> > +                        */
> > +                       pages[i] = alloc_pages(gfp, 0);
> I think down the code in this while loop, i & count is being
> calculated based on the "order" of allocation in the current
> iteration.
> Since value of order will be automatically 0 here if (!pages[i]) is
> true then, why hard code order to value of 0 here.
> Comment clearly says what this code is doing right?
> 

Gcc is smart enough to know that order == 0 here, the code generation on 
arm will be the same, so this is only a matter of how the source looks.  
To me, it doesn't make a lot of sense to write it as alloc_pages(gfp, 
order) when order is always equal to 0.  I think it's clearer the way that 
Tomasz wrote it.

> I know it is just a minor thing. Don't know if it is relevant.
> 
> > +                       if (!pages[i])
> > +                               goto error;
> > +               }
> >
> >                 if (order) {
> >                         split_page(pages[i], order);

[PATCH v2] ARM: mm: Do not invoke OOM for higher order IOMMU DMA allocations

[Ritesh Harjani]

On Thu, Mar 26, 2015 at 6:04 AM, David Rientjes <rientjes@google.com> wrote:
> On Thu, 26 Mar 2015, Ritesh Harjani wrote:
>
>> > diff --git a/arch/arm/mm/dma-mapping.c b/arch/arm/mm/dma-mapping.c
>> > index 83cd5ac..3f1ac51 100644
>> > --- a/arch/arm/mm/dma-mapping.c
>> > +++ b/arch/arm/mm/dma-mapping.c
>> > @@ -1150,13 +1150,28 @@ static struct page **__iommu_alloc_buffer(struct device *dev, size_t size,
>> >         gfp |= __GFP_NOWARN | __GFP_HIGHMEM;
>> >
>> >         while (count) {
>> > -               int j, order = __fls(count);
>> > +               int j, order;
>> > +
>> > +               for (order = __fls(count); order > 0; --order) {
>> > +                       /*
>> > +                        * We do not want OOM killer to be invoked as long
>> > +                        * as we can fall back to single pages, so we force
>> > +                        * __GFP_NORETRY for orders higher than zero.
>> > +                        */
>> > +                       pages[i] = alloc_pages(gfp | __GFP_NORETRY, order);
>> > +                       if (pages[i])
>> > +                               break;
>> > +               }
>> >
>> > -               pages[i] = alloc_pages(gfp, order);
>> > -               while (!pages[i] && order)
>> > -                       pages[i] = alloc_pages(gfp, --order);
>> > -               if (!pages[i])
>> > -                       goto error;
>> > +               if (!pages[i]) {
>> > +                       /*
>> > +                        * Fall back to single page allocation.
>> > +                        * Might invoke OOM killer as last resort.
>> > +                        */
>> > +                       pages[i] = alloc_pages(gfp, 0);
>> I think down the code in this while loop, i & count is being
>> calculated based on the "order" of allocation in the current
>> iteration.
>> Since value of order will be automatically 0 here if (!pages[i]) is
>> true then, why hard code order to value of 0 here.
>> Comment clearly says what this code is doing right?
>>
>
> Gcc is smart enough to know that order == 0 here, the code generation on
> arm will be the same, so this is only a matter of how the source looks.

Agreed.
> To me, it doesn't make a lot of sense to write it as alloc_pages(gfp,
> order) when order is always equal to 0.  I think it's clearer the way that
> Tomasz wrote it.

Ok
>
>> I know it is just a minor thing. Don't know if it is relevant.
>>
>> > +                       if (!pages[i])
>> > +                               goto error;
>> > +               }
>> >
>> >                 if (order) {
>> >                         split_page(pages[i], order);

Thanks
Ritesh