Re: [RFC PATCH] arm: dma-mapping: fallback allocation for cma failure

[Michal Nazarewicz]

On Mon, May 19 2014, Joonsoo Kim wrote:
> On Tue, May 20, 2014 at 02:57:47PM +0900, Gioh Kim wrote:
>> 
>> Thanks for your advise, Michal Nazarewicz.
>> 
>> Having discuss with Joonsoo, I'm adding fallback allocation after __alloc_from_contiguous().
>> The fallback allocation works if CMA kernel options is turned on but CMA size is zero.
>
> Hello, Gioh.
>
> I also mentioned the case where devices have their specific cma_area.
> It means that this device needs memory with some contraint.
> Although I'm not familiar with DMA infrastructure, I think that
> we should handle this case.
>
> How about below patch?
>
> ------------>8----------------
> diff --git a/arch/arm/mm/dma-mapping.c b/arch/arm/mm/dma-mapping.c
> index 6b00be1..4023434 100644
> --- a/arch/arm/mm/dma-mapping.c
> +++ b/arch/arm/mm/dma-mapping.c
> @@ -379,7 +379,7 @@ static int __init atomic_pool_init(void)
>  	unsigned long *bitmap;
>  	struct page *page;
>  	struct page **pages;
> -	void *ptr;
> +	void *ptr = NULL;
>  	int bitmap_size = BITS_TO_LONGS(nr_pages) * sizeof(long);
>  
>  	bitmap = kzalloc(bitmap_size, GFP_KERNEL);
> @@ -393,7 +393,8 @@ static int __init atomic_pool_init(void)
>  	if (IS_ENABLED(CONFIG_DMA_CMA))
>  		ptr = __alloc_from_contiguous(NULL, pool->size, prot, &page,
>  					      atomic_pool_init);
> -	else
> +
> +	if (!ptr)
>  		ptr = __alloc_remap_buffer(NULL, pool->size, gfp, prot, &page,
>  					   atomic_pool_init);
>  	if (ptr) {
> @@ -701,10 +702,22 @@ static void *__dma_alloc(struct device *dev, size_t size, dma_addr_t *handle,
>  		addr = __alloc_simple_buffer(dev, size, gfp, &page);
>  	else if (!(gfp & __GFP_WAIT))
>  		addr = __alloc_from_pool(size, &page);
> -	else if (!IS_ENABLED(CONFIG_DMA_CMA))
> -		addr = __alloc_remap_buffer(dev, size, gfp, prot, &page, caller);
> -	else
> -		addr = __alloc_from_contiguous(dev, size, prot, &page, caller);
> +	else {
> +		if (IS_ENABLED(CONFIG_DMA_CMA)) {
> +			addr = __alloc_from_contiguous(dev, size, prot,
> +							&page, caller);
> +			/*
> +			 * Device specific cma_area means that
> +			 * this device needs memory with some contraint.
> +			 * So, we can't fall through general remap allocation.
> +			 */
> +			if (!addr && dev && dev->cma_area)
> +				return NULL;
> +		}
> +
> +		addr = __alloc_remap_buffer(dev, size, gfp, prot,
> +							&page, caller);
> +	}

__arm_dma_free will have to be changed to handle the fallback as well.
But perhaps Marek is right and there should be no fallback for regular
allocations?  Than again, non-CMA allocation should be performed at
least in the case of cma=0.

>  
>  	if (addr)
>  		*handle = pfn_to_dma(dev, page_to_pfn(page));

-- 
Best regards,                                         _     _
.o. | Liege of Serenely Enlightened Majesty of      o' \,=./ `o
..o | Computer Science,  MichaÅ âmina86â Nazarewicz    (o o)
ooo +--<mpn@xxxxxxxxxx>--<xmpp:mina86@xxxxxxxxxx>--ooO--(_)--Ooo--

Attachment: signature.asc
Description: PGP signature