diff mbox

[PATCHv7,9/9] ARM: dma-mapping: add support for IOMMU mapper

Message ID 20120329101927.8ab6b1993475b7e16ae2258f@nvidia.com
State New
Headers show

Commit Message

Hiroshi Doyu March 29, 2012, 7:19 a.m. UTC
Hi Marek,

On Wed, 29 Feb 2012 16:04:22 +0100
Marek Szyprowski <m.szyprowski@samsung.com> wrote:

> This patch add a complete implementation of DMA-mapping API for
> devices that have IOMMU support. All DMA-mapping calls are supported.
> 
> This patch contains some of the code kindly provided by Krishna Reddy
> <vdumpa@nvidia.com> and Andrzej Pietrasiewicz <andrzej.p@samsung.com>
> 
> Signed-off-by: Marek Szyprowski <m.szyprowski@samsung.com>
> Signed-off-by: Kyungmin Park <kyungmin.park@samsung.com>
> Reviewed-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
> ---
>  arch/arm/Kconfig                 |    8 +
>  arch/arm/include/asm/device.h    |    3 +
>  arch/arm/include/asm/dma-iommu.h |   34 ++
>  arch/arm/mm/dma-mapping.c        |  726 +++++++++++++++++++++++++++++++++++++-
>  arch/arm/mm/vmregion.h           |    2 +-
>  5 files changed, 758 insertions(+), 15 deletions(-)
>  create mode 100644 arch/arm/include/asm/dma-iommu.h
> 

<snip>

> +/*
> + * Map a part of the scatter-gather list into contiguous io address space
> + */
> +static int __map_sg_chunk(struct device *dev, struct scatterlist *sg,
> +                         size_t size, dma_addr_t *handle,
> +                         enum dma_data_direction dir)
> +{
> +       struct dma_iommu_mapping *mapping = dev->archdata.mapping;
> +       dma_addr_t iova, iova_base;
> +       int ret = 0;
> +       unsigned int count;
> +       struct scatterlist *s;
> +
> +       size = PAGE_ALIGN(size);
> +       *handle = ARM_DMA_ERROR;
> +
> +       iova_base = iova = __alloc_iova(mapping, size);
> +       if (iova == ARM_DMA_ERROR)
> +               return -ENOMEM;
> +
> +       for (count = 0, s = sg; count < (size >> PAGE_SHIFT); s = sg_next(s))
> +       {
> +               phys_addr_t phys = page_to_phys(sg_page(s));
> +               unsigned int len = PAGE_ALIGN(s->offset + s->length);
> +
> +               if (!arch_is_coherent())
> +                       __dma_page_cpu_to_dev(sg_page(s), s->offset, s->length, dir);
> +
> +               ret = iommu_map(mapping->domain, iova, phys, len, 0);
> +               if (ret < 0)
> +                       goto fail;
> +               count += len >> PAGE_SHIFT;
> +               iova += len;
> +       }
> +       *handle = iova_base;
> +
> +       return 0;
> +fail:
> +       iommu_unmap(mapping->domain, iova_base, count * PAGE_SIZE);
> +       __free_iova(mapping, iova_base, size);
> +       return ret;
> +}

Do we need to set dma_address as below?

From e8bcc3cdac5375b5d7f5ac5b3f716a11c1008f38 Mon Sep 17 00:00:00 2001
From: Hiroshi DOYU <hdoyu@nvidia.com>
Date: Thu, 29 Mar 2012 09:59:04 +0300
Subject: [PATCH 1/1] ARM: dma-mapping: Fix dma_address in sglist

s->dma_address wasn't set at mapping.

Signed-off-by: Hiroshi DOYU <hdoyu@nvidia.com>
---
 arch/arm/mm/dma-mapping.c |    2 ++
 1 files changed, 2 insertions(+), 0 deletions(-)

Comments

Marek Szyprowski March 29, 2012, 8 a.m. UTC | #1
Hello,

On Thursday, March 29, 2012 9:19 AM Hiroshi Doyu wrote:

> On Wed, 29 Feb 2012 16:04:22 +0100
> Marek Szyprowski <m.szyprowski@samsung.com> wrote:
> 
> > This patch add a complete implementation of DMA-mapping API for
> > devices that have IOMMU support. All DMA-mapping calls are supported.
> >
> > This patch contains some of the code kindly provided by Krishna Reddy
> > <vdumpa@nvidia.com> and Andrzej Pietrasiewicz <andrzej.p@samsung.com>
> >
> > Signed-off-by: Marek Szyprowski <m.szyprowski@samsung.com>
> > Signed-off-by: Kyungmin Park <kyungmin.park@samsung.com>
> > Reviewed-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
> > ---
> >  arch/arm/Kconfig                 |    8 +
> >  arch/arm/include/asm/device.h    |    3 +
> >  arch/arm/include/asm/dma-iommu.h |   34 ++
> >  arch/arm/mm/dma-mapping.c        |  726 +++++++++++++++++++++++++++++++++++++-
> >  arch/arm/mm/vmregion.h           |    2 +-
> >  5 files changed, 758 insertions(+), 15 deletions(-)
> >  create mode 100644 arch/arm/include/asm/dma-iommu.h
> >
> 
> <snip>
> 
> > +/*
> > + * Map a part of the scatter-gather list into contiguous io address space
> > + */
> > +static int __map_sg_chunk(struct device *dev, struct scatterlist *sg,
> > +                         size_t size, dma_addr_t *handle,
> > +                         enum dma_data_direction dir)
> > +{
> > +       struct dma_iommu_mapping *mapping = dev->archdata.mapping;
> > +       dma_addr_t iova, iova_base;
> > +       int ret = 0;
> > +       unsigned int count;
> > +       struct scatterlist *s;
> > +
> > +       size = PAGE_ALIGN(size);
> > +       *handle = ARM_DMA_ERROR;
> > +
> > +       iova_base = iova = __alloc_iova(mapping, size);
> > +       if (iova == ARM_DMA_ERROR)
> > +               return -ENOMEM;
> > +
> > +       for (count = 0, s = sg; count < (size >> PAGE_SHIFT); s = sg_next(s))
> > +       {
> > +               phys_addr_t phys = page_to_phys(sg_page(s));
> > +               unsigned int len = PAGE_ALIGN(s->offset + s->length);
> > +
> > +               if (!arch_is_coherent())
> > +                       __dma_page_cpu_to_dev(sg_page(s), s->offset, s->length, dir);
> > +
> > +               ret = iommu_map(mapping->domain, iova, phys, len, 0);
> > +               if (ret < 0)
> > +                       goto fail;
> > +               count += len >> PAGE_SHIFT;
> > +               iova += len;
> > +       }
> > +       *handle = iova_base;
> > +
> > +       return 0;
> > +fail:
> > +       iommu_unmap(mapping->domain, iova_base, count * PAGE_SIZE);
> > +       __free_iova(mapping, iova_base, size);
> > +       return ret;
> > +}
> 
> Do we need to set dma_address as below?

Nope, this one is not correct. Please check the arm_iommu_map_sg() function. It calls 
__map_sg_chunk() and gives it &dma->dma_address as one of the arguments, so the dma 
address is correctly stored in the scatter list. Please note that scatterlist is really
non-trivial structure and information about physical memory pages/chunks is disjoint 
from the information about dma address space chunks, although both are stored on the 
same list. In arm_iommu_map_sg() 's' pointer is used for physical memory chunks and 
'dma' pointer is used for dma address space chunks. 

The number of dma address space chunks (returned by arm_iommu_map_sq) can be lower 
than the number of physical memory chunks (given as nents argument).

In case of IOMMU you usually construct the scatter list in such a way, that you get
only one dma address chunk (so in fact you get a buffer mapped contiguously in 
virtual io address space).

> From e8bcc3cdac5375b5d7f5ac5b3f716a11c1008f38 Mon Sep 17 00:00:00 2001
> From: Hiroshi DOYU <hdoyu@nvidia.com>
> Date: Thu, 29 Mar 2012 09:59:04 +0300
> Subject: [PATCH 1/1] ARM: dma-mapping: Fix dma_address in sglist
> 
> s->dma_address wasn't set at mapping.
> 
> Signed-off-by: Hiroshi DOYU <hdoyu@nvidia.com>
> ---
>  arch/arm/mm/dma-mapping.c |    2 ++
>  1 files changed, 2 insertions(+), 0 deletions(-)
> 
> diff --git a/arch/arm/mm/dma-mapping.c b/arch/arm/mm/dma-mapping.c
> index 3347c37..11a9d65 100644
> --- a/arch/arm/mm/dma-mapping.c
> +++ b/arch/arm/mm/dma-mapping.c
> @@ -1111,6 +1111,8 @@ static int __map_sg_chunk(struct device *dev, struct scatterlist *sg,
>  		ret = iommu_map(mapping->domain, iova, phys, len, 0);
>  		if (ret < 0)
>  			goto fail;
> +		s->dma_address = iova;
> +
>  		count += len >> PAGE_SHIFT;
>  		iova += len;
>  	}

The above patch is not needed at all.

Best regards
Krishna Reddy March 30, 2012, 2:24 a.m. UTC | #2
Hi,
I have found a bug in arm_iommu_map_sg(). 

> +int arm_iommu_map_sg(struct device *dev, struct scatterlist *sg, int nents,
> +		     enum dma_data_direction dir, struct dma_attrs *attrs) {
> +	struct scatterlist *s = sg, *dma = sg, *start = sg;
> +	int i, count = 0;
> +	unsigned int offset = s->offset;
> +	unsigned int size = s->offset + s->length;
> +	unsigned int max = dma_get_max_seg_size(dev);
> +
> +	for (i = 1; i < nents; i++) {
> +		s->dma_address = ARM_DMA_ERROR;
> +		s->dma_length = 0;
> +
> +		s = sg_next(s);

With above code, the last sg element's dma_length is not getting set to zero.
This causing additional incorrect  unmapping during arm_iommu_unmap_sg call and 
leading to random crashes.
The order of above three lines should be as follows.
		s = sg_next(s);

		s->dma_address = ARM_DMA_ERROR;
		s->dma_length = 0;


-KR
--nvpublic
Marek Szyprowski March 30, 2012, 6:30 a.m. UTC | #3
Hello,

On Friday, March 30, 2012 4:24 AM Krishna Reddy wrote:

> Hi,
> I have found a bug in arm_iommu_map_sg().
> 
> > +int arm_iommu_map_sg(struct device *dev, struct scatterlist *sg, int nents,
> > +		     enum dma_data_direction dir, struct dma_attrs *attrs) {
> > +	struct scatterlist *s = sg, *dma = sg, *start = sg;
> > +	int i, count = 0;
> > +	unsigned int offset = s->offset;
> > +	unsigned int size = s->offset + s->length;
> > +	unsigned int max = dma_get_max_seg_size(dev);
> > +
> > +	for (i = 1; i < nents; i++) {
> > +		s->dma_address = ARM_DMA_ERROR;
> > +		s->dma_length = 0;
> > +
> > +		s = sg_next(s);
> 
> With above code, the last sg element's dma_length is not getting set to zero.
> This causing additional incorrect  unmapping during arm_iommu_unmap_sg call and
> leading to random crashes.
> The order of above three lines should be as follows.
> 		s = sg_next(s);
> 
> 		s->dma_address = ARM_DMA_ERROR;
> 		s->dma_length = 0;
> 

You are right, the order of those lines must be reversed. In all my test codes the 
scatter list was initially cleared, so I missed this typical off-by-one error. 
Thanks for spotting it!

Best regards
diff mbox

Patch

diff --git a/arch/arm/mm/dma-mapping.c b/arch/arm/mm/dma-mapping.c
index 3347c37..11a9d65 100644
--- a/arch/arm/mm/dma-mapping.c
+++ b/arch/arm/mm/dma-mapping.c
@@ -1111,6 +1111,8 @@  static int __map_sg_chunk(struct device *dev, struct scatterlist *sg,
 		ret = iommu_map(mapping->domain, iova, phys, len, 0);
 		if (ret < 0)
 			goto fail;
+		s->dma_address = iova;
+
 		count += len >> PAGE_SHIFT;
 		iova += len;
 	}