Re: [PATCH 3/3] swiotlb: use GFP_DMA32 instead of GFP_DMA in swiotlb_alloc_coherent

[Joerg Roedel <joro@8bytes.org>]

On Thu, Sep 04, 2008 at 03:04:25AM +0900, FUJITA Tomonori wrote:
> x86 sets up gfp flag propely so swiotlb_alloc_coherent doesn't need to
> use GFP_DMA unconditionally. It leads to allocation failures in some
> systems.
> 
> Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
> ---
>  lib/swiotlb.c |    7 +------
>  1 files changed, 1 insertions(+), 6 deletions(-)
> 
> diff --git a/lib/swiotlb.c b/lib/swiotlb.c
> index 977edbd..b88d7c2 100644
> --- a/lib/swiotlb.c
> +++ b/lib/swiotlb.c
> @@ -467,12 +467,7 @@ swiotlb_alloc_coherent(struct device *hwdev, size_t size,
>  	void *ret;
>  	int order = get_order(size);
>  
> -	/*
> -	 * XXX fix me: the DMA API should pass us an explicit DMA mask
> -	 * instead, or use ZONE_DMA32 (ia64 overloads ZONE_DMA to be a ~32
> -	 * bit range instead of a 16MB one).
> -	 */
> -	flags |= GFP_DMA;
> +	flags |= GFP_DMA32;

No. This is exactly the place where the swiotlb iommu implementation
should handle the dma_mask of the specific device. Unconditionally use
DMA32 here is not the best possible handling.

Joerg