[PATCH] drm: exynos: fix for mapping non contig dma buffers

Inki Dae inki.dae at samsung.com
Thu Nov 1 21:44:14 PDT 2012


It's good patch. Right, there was my missing point. I thought each sgl would
always have 4k page. But dma mapping api, dma_alloc_alloc(), allocates
physical memory as continuously as possible so the sgl can have group of
pages.

Below is my comment.


> -----Original Message-----
> From: Rahul Sharma [mailto:rahul.sharma at samsung.com]
> Sent: Thursday, November 01, 2012 9:35 PM
> To: dri-devel at lists.freedesktop.org
> Cc: sw0312.kim at samsung.com; inki.dae at samsung.com; jy0922.shim at samsung.com;
> kyungmin.park at samsung.com; prashanth.g at samsung.com; joshi at samsung.com;
> s.shirish at samsung.com; r.sh.open at gmail.com; rahul.sharma at samsung.com
> Subject: [PATCH] drm: exynos: fix for mapping non contig dma buffers
> 
> This patch fixes the problem of mapping gem objects which are non-
> contigous
> dma buffers. These buffers are described using SG table and SG lists. Each
> valid SG List is pointing to a single page or group of pages which are
> physically contigous.
> 
> Current implementation just maps the first page of each SG List and leave
> the other pages unmapped, leading to a crash.
> 
> Given solution finds the page struct for all the pages in the SG list and
> map them one by one. This ensures all the pages of current SG list are
> mapped.
> Next SG list of the same buffer will be mapped when page fault occurs for
> the offset belongs to that list.
> 
> This patch is based on branch exynos-drm-next-iommu at
> git://git.kernel.org/pub/scm/linux/kernel/git/daeinki/drm-exynos.git
> 
> Signed-off-by: Rahul Sharma <rahul.sharma at samsung.com>
> ---
>  drivers/gpu/drm/exynos/exynos_drm_gem.c |   34
> +++++++++++++++++++++++++++---
>  1 files changed, 30 insertions(+), 4 deletions(-)
> 
> diff --git a/drivers/gpu/drm/exynos/exynos_drm_gem.c
> b/drivers/gpu/drm/exynos/exynos_drm_gem.c
> index 7057729..d2d6188 100644
> --- a/drivers/gpu/drm/exynos/exynos_drm_gem.c
> +++ b/drivers/gpu/drm/exynos/exynos_drm_gem.c
> @@ -95,17 +95,43 @@ static int exynos_drm_gem_map_buf(struct
> drm_gem_object *obj,
>  {
>  	struct exynos_drm_gem_obj *exynos_gem_obj = to_exynos_gem_obj(obj);
>  	struct exynos_drm_gem_buf *buf = exynos_gem_obj->buffer;
> +	struct scatterlist *sgl;
>  	unsigned long pfn;
> +	unsigned int phys_addr;
> +	int ret, i;
> 
>  	if (exynos_gem_obj->flags & EXYNOS_BO_NONCONTIG) {
>  		if (!buf->pages)
>  			return -EINTR;

Buf->pages should be checked in all cases. For this, see below comment.

> 
> -		pfn = page_to_pfn(buf->pages[page_offset++]);
> -	} else
> +		sgl = buf->sgt->sgl;
> +		for_each_sg(buf->sgt->sgl, sgl, buf->sgt->nents, i) {
> +			if (page_offset < (sgl->length >> PAGE_SHIFT))
> +				break;
> +			page_offset -=  (sgl->length >> PAGE_SHIFT);
> +		}
> +
> +		if (i >= buf->sgt->nents) {
> +			DRM_ERROR("invalid page offset\n");
> +			return -EINVAL;
> +		}
> +
> +		phys_addr = sg_phys(sgl);
> +
> +		for (i = 0; i < (sgl->length >> PAGE_SHIFT); i++) {
> +			pfn = __phys_to_pfn(phys_addr + (i << PAGE_SHIFT));
> +			ret = vm_insert_mixed(vma, f_vaddr + (i <<
PAGE_SHIFT),
> +						pfn);
> +			if (ret < 0) {
> +				DRM_ERROR("failed to map page.\n");
> +				return ret;
> +			}
> +		}
> +		return 0;
> +	} else {
>  		pfn = (buf->dma_addr >> PAGE_SHIFT) + page_offset;

It seems like that you missed EXYNOS_BO_CONTIG type testing. With iommu,
buf->dma_addr has device address but not physical address. So correct it
like below,
	pfn = page_to_pfn(buf->pages[0]) + page_offset;

And one more posting? :)

Thanks,
Inki Dae

> -
> -	return vm_insert_mixed(vma, f_vaddr, pfn);
> +		return vm_insert_mixed(vma, f_vaddr, pfn);
> +	}
>  }
> 
>  static int exynos_drm_gem_handle_create(struct drm_gem_object *obj,
> --
> 1.7.0.4



More information about the dri-devel mailing list