Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > linux.kernel > #1409028 > unrolled thread

[RFC v2] Change dma_attrs from bitfield to unsigned long

Back to article view | Back to linux.kernel

Contents

  [RFC v2] Change dma_attrs from bitfield to unsigned long Krzysztof Kozlowski <k.kozlowski@samsung.com> - 2016-05-30 14:00 +0200
    Re: [RFC v2] dma-mapping: Use unsigned long for dma_attrs Christoph Hellwig <hch@infradead.org> - 2016-05-31 19:10 +0200
      Re: [RFC v2] dma-mapping: Use unsigned long for dma_attrs Krzysztof Kozlowski <k.kozlowski@samsung.com> - 2016-06-01 07:40 +0200
        Re: [RFC v2] dma-mapping: Use unsigned long for dma_attrs Christoph Hellwig <hch@infradead.org> - 2016-06-01 10:00 +0200
    Re: [RFC v2] dma-mapping: Use unsigned long for dma_attrs Krzysztof Kozlowski <k.kozlowski@samsung.com> - 2016-06-01 08:10 +0200

#1409028 — [RFC v2] Change dma_attrs from bitfield to unsigned long

Hi,


This is second attempt to bring some safeness to dma_attrs.
In v1 [0] I added const to data pointed by attrs. However Christoph
Hellwig suggested getting rid of struct dma_attrs in favor of
some simpler data type.


Benefits of unsigned long for dma_attrs:
1. This is just simpler.  Both in terms of reading the code and setting
   attributes.  Instead of initializing local attributes on the stack and
   passing pointer to it to dma_set_attr(), just set the bits.

2. It brings safeness and checking for const correctness because the
   attributes are passed by value.


The problem is that the patch touches the dma-mapping API so all
the users have to be converted (ARM and ARM64 as of now).

Any ideas to split it per smaller patches?

The work is not finished yet, I would rather get some feedback before
progressing further.


Best regards,
Krzysztof


Krzysztof Kozlowski (1):
  dma-mapping: Use unsigned long for dma_attrs

 Documentation/DMA-API.txt                 |   2 +-
 Documentation/DMA-attributes.txt          |   2 +-
 arch/arm/include/asm/dma-mapping.h        |  13 ++--
 arch/arm/include/asm/xen/page-coherent.h  |  16 ++---
 arch/arm/mm/dma-mapping.c                 |  82 +++++++++++------------
 arch/arm/xen/mm.c                         |   4 +-
 arch/arm64/mm/dma-mapping.c               |  57 ++++++++--------
 drivers/gpu/drm/exynos/exynos_drm_fbdev.c |   2 +-
 drivers/gpu/drm/exynos/exynos_drm_g2d.c   |   1 -
 drivers/gpu/drm/exynos/exynos_drm_gem.c   |  20 +++---
 drivers/gpu/drm/exynos/exynos_drm_gem.h   |   2 +-
 drivers/iommu/dma-iommu.c                 |   6 +-
 drivers/xen/swiotlb-xen.c                 |  14 ++--
 include/linux/dma-attrs.h                 |  71 --------------------
 include/linux/dma-iommu.h                 |   6 +-
 include/linux/dma-mapping.h               | 105 +++++++++++++++++-------------
 include/linux/swiotlb.h                   |  10 +--
 include/xen/swiotlb-xen.h                 |  12 ++--
 lib/dma-noop.c                            |   9 +--
 lib/swiotlb.c                             |  13 ++--
 20 files changed, 195 insertions(+), 252 deletions(-)
 delete mode 100644 include/linux/dma-attrs.h

-- 
1.9.1

[toc] | [next] | [standalone]

#1410449 — Re: [RFC v2] dma-mapping: Use unsigned long for dma_attrs

On Mon, May 30, 2016 at 01:54:06PM +0200, Krzysztof Kozlowski wrote:
> The dma-mapping core and the implementations do not change the
> DMA attributes passed by pointer.  Thus the pointer can point to const
> data.  However the attributes do not have to be a bitfield. Instead
> unsigned long will do fine:
> 
> 1. This is just simpler.  Both in terms of reading the code and setting
>    attributes.  Instead of initializing local attributes on the stack and
>    passing pointer to it to dma_set_attr(), just set the bits.
> 
> 2. It brings safeness and checking for const correctness because the
>    attributes are passed by value.
> 
> Please have in mind that this is RFC, not finished yet.  Only ARM and
> ARM64 are fixed (and not everywhere).
> However other API users also have to be converted which is quite
> intrusive.  I would rather avoid it until the overall approach is
> accepted.

This looks great!  Please continue doing the full conversion.

> +/**
> + * List of possible attributes associated with a DMA mapping. The semantics
> + * of each attribute should be defined in Documentation/DMA-attributes.txt.
> + */
> +#define DMA_ATTR_WRITE_BARRIER		BIT(1)
> +#define DMA_ATTR_WEAK_ORDERING		BIT(2)
> +#define DMA_ATTR_WRITE_COMBINE		BIT(3)
> +#define DMA_ATTR_NON_CONSISTENT		BIT(4)
> +#define DMA_ATTR_NO_KERNEL_MAPPING	BIT(5)
> +#define DMA_ATTR_SKIP_CPU_SYNC		BIT(6)
> +#define DMA_ATTR_FORCE_CONTIGUOUS	BIT(7)
> +#define DMA_ATTR_ALLOC_SINGLE_PAGES	BIT(8)

No really for this patch, but I would much prefer to document them next
to the code in the long run.  Also I really think these BIT() macros
are a distraction compared to the (1 << N) notation.

> +/**
> + * dma_get_attr - check for a specific attribute
> + * @attr: attribute to look for
> + * @attrs: attributes to check within
> + */
> +static inline bool dma_get_attr(unsigned long attr, unsigned long attrs)
> +{
> +	return !!(attr & attrs);
> +}

I'd just kill this helper, much easier to simply open code it in the
caller.

[toc] | [prev] | [next] | [standalone]

#1410826 — Re: [RFC v2] dma-mapping: Use unsigned long for dma_attrs

On 05/31/2016 07:04 PM, Christoph Hellwig wrote:
> On Mon, May 30, 2016 at 01:54:06PM +0200, Krzysztof Kozlowski wrote:
>> The dma-mapping core and the implementations do not change the
>> DMA attributes passed by pointer.  Thus the pointer can point to const
>> data.  However the attributes do not have to be a bitfield. Instead
>> unsigned long will do fine:
>>
>> 1. This is just simpler.  Both in terms of reading the code and setting
>>    attributes.  Instead of initializing local attributes on the stack and
>>    passing pointer to it to dma_set_attr(), just set the bits.
>>
>> 2. It brings safeness and checking for const correctness because the
>>    attributes are passed by value.
>>
>> Please have in mind that this is RFC, not finished yet.  Only ARM and
>> ARM64 are fixed (and not everywhere).
>> However other API users also have to be converted which is quite
>> intrusive.  I would rather avoid it until the overall approach is
>> accepted.
> 
> This looks great!  Please continue doing the full conversion.
> 
>> +/**
>> + * List of possible attributes associated with a DMA mapping. The semantics
>> + * of each attribute should be defined in Documentation/DMA-attributes.txt.
>> + */
>> +#define DMA_ATTR_WRITE_BARRIER		BIT(1)
>> +#define DMA_ATTR_WEAK_ORDERING		BIT(2)
>> +#define DMA_ATTR_WRITE_COMBINE		BIT(3)
>> +#define DMA_ATTR_NON_CONSISTENT		BIT(4)
>> +#define DMA_ATTR_NO_KERNEL_MAPPING	BIT(5)
>> +#define DMA_ATTR_SKIP_CPU_SYNC		BIT(6)
>> +#define DMA_ATTR_FORCE_CONTIGUOUS	BIT(7)
>> +#define DMA_ATTR_ALLOC_SINGLE_PAGES	BIT(8)
> 
> No really for this patch, but I would much prefer to document them next
> to the code in the long run.  Also I really think these BIT() macros
> are a distraction compared to the (1 << N) notation.

Not much difference to me but maybe plain number:
...	0x01u
...	0x02u
?

> 
>> +/**
>> + * dma_get_attr - check for a specific attribute
>> + * @attr: attribute to look for
>> + * @attrs: attributes to check within
>> + */
>> +static inline bool dma_get_attr(unsigned long attr, unsigned long attrs)
>> +{
>> +	return !!(attr & attrs);
>> +}
> 
> I'd just kill this helper, much easier to simply open code it in the
> caller.

Keeping it for now helps reducing the number of changes in the patch.
The patch will be quite big as it has to replace all the uses atomically.

I can get rid of the helper in consecutive patch.

Best regards,
Krzysztof

[toc] | [prev] | [next] | [standalone]

#1410910 — Re: [RFC v2] dma-mapping: Use unsigned long for dma_attrs

On Wed, Jun 01, 2016 at 07:36:42AM +0200, Krzysztof Kozlowski wrote:
> > No really for this patch, but I would much prefer to document them next
> > to the code in the long run.  Also I really think these BIT() macros
> > are a distraction compared to the (1 << N) notation.
> 
> Not much difference to me but maybe plain number:
> ...	0x01u
> ...	0x02u
> ?

I prefer the little bit shifts, but even the explicit values are much
better than the obsfucating macros :)  Anyway, your patch and in the end
all three methods will get the work done.

> > I'd just kill this helper, much easier to simply open code it in the
> > caller.
> 
> Keeping it for now helps reducing the number of changes in the patch.
> The patch will be quite big as it has to replace all the uses atomically.
> 
> I can get rid of the helper in consecutive patch.

Sounds fine.

[toc] | [prev] | [next] | [standalone]

#1410838 — Re: [RFC v2] dma-mapping: Use unsigned long for dma_attrs

On 05/31/2016 08:15 PM, Konrad Rzeszutek Wilk wrote:
> On Mon, May 30, 2016 at 01:54:06PM +0200, Krzysztof Kozlowski wrote:
>> The dma-mapping core and the implementations do not change the
>> DMA attributes passed by pointer.  Thus the pointer can point to const
>> data.  However the attributes do not have to be a bitfield. Instead
>> unsigned long will do fine:
>>
>> 1. This is just simpler.  Both in terms of reading the code and setting
>>    attributes.  Instead of initializing local attributes on the stack and
>>    passing pointer to it to dma_set_attr(), just set the bits.
>>
>> 2. It brings safeness and checking for const correctness because the
>>    attributes are passed by value.
> 
> 
> .. why not go the next step a do an enum? Perhaps that should be mentioned
> as part of the description?

These are additive flags so to me this would look a little bit weird:
enum dma_attr {
	DMA_ATTR_WRITE_BARRIER	= 0x1,
	DMA_ATTR_WEAK_ORDERING	= 0x2,
	DMA_ATTR_WRITE_COMBINE	= 0x4,
	DMA_ATTR_NON_CONSISTENT	= 0x8,
	...
}

It doesn't really look like enumeration.

Best regards,
Krzysztof

[toc] | [prev] | [standalone]

Back to top | Article view | linux.kernel

csiph-web