Re: [PATCH 0/3] ASoC: Move pcm writecombine dma buffer allocation to core

[Lars-Peter Clausen <lars@metafoo.de>]

On 06/29/2012 06:18 PM, Takashi Iwai wrote:
> At Fri, 29 Jun 2012 18:06:52 +0200,
> Lars-Peter Clausen wrote:
>>
>> On 06/29/2012 05:04 PM, Laxman Dewangan wrote:
>>> [...]
>>>
>>>
>>>> +/* allocate the coherent DMA pages */
>>>> +static void *snd_malloc_dev_pages(struct device *dev, size_t size,
>>>> dma_addr_t *dma)
>>>> +{
>>>> +    return __snd_malloc_dev_pages(dev, size, dma, dma_alloc_coherent);
>>>
>>> This does not get compiled in ARM because dma_alloc_coherant is macro
>>> defined as
>>>
>>> #define dma_alloc_coherent(d, s, h, f) dma_alloc_attrs(d, s, h, f, NULL)
>>>
>>> static inline void *dma_alloc_attrs(struct device *dev, size_t size,
>>>                                        dma_addr_t *dma_handle, gfp_t flag,
>>>                                        struct dma_attrs *attrs)
>>>
>>>
>>> I fixed this by
>>> static void *_dma_alloc_coherent(struct device *dev, size_t size,
>>>                 dma_addr_t *dma_handle, gfp_t flag)
>>> {
>>>         return dma_alloc_coherent(dev, size, dma_handle, flag);
>>> }
>>>
>>> /* allocate the coherent DMA pages */
>>> static void *snd_malloc_dev_pages(struct device *dev, size_t size,
>>> dma_addr_t *dma)
>>> {
>>>         return _snd_malloc_dev_pages(dev, size, dma, _dma_alloc_coherent);
>>> }
>>
>> I think all dma_alloc_* functions use dma_alloc_attrs internally these days.
>> I think it might be a better idea just to use dma_alloc_attrs in
>> __snd_malloc_dev_pages and just pass a dma_attrs struct instead of a function.
>>
>> Something like:
>>
>> static void *__snd_malloc_dev_pages(struct device *dev, size_t size,
>> 			    dma_addr_t *dma, struct dma_attrs *attrs)
>> {
>> ....
>> 	res = dma_alloc_attrs(dev, PAGE_SIZE << pg, dma, gfp_flags, attrs);
>> ....
>> }
>>
>>
>> static void *snd_malloc_dev_pages(struct device *dev, size_t size,
>> 	dma_addr_t *dma)
>> {
>> 	return __snd_malloc_dev_pages(dev, size, dma, NULL);
>> }
>>
>> static void *snd_malloc_dev_wc_pages(struct device *dev, size_t size,
>> 	dma_addr_t *dma)
>> {
>> 	DEFINE_DMA_ATTRS(attrs);
>> 	dma_set_attr(DMA_ATTR_WRITE_COMBINE, &attrs);
>>
>> 	return __snd_malloc_dev_pages(dev, size, dma, &attrs);
>> }
>>
>> Same applies for freeing the memory again.
> 
> Yes, that'd be cleaner.
> But I prefer the code still workable on 3.5-base kernel, which makes
> easier to test and merge for 3.6, at first. 

Actually that part is already in 3.5. But I've just had a closer look and it
looks as if even for 3.6 not all archs have been converted to
dma_alloc_attrs yet :/