Re: crash in gart_unmap_page() with master snapshot (commit e1ef035d272e)

[Michal Kubecek]

On Fri, Jan 04, 2019 at 09:53:18AM +0100, Christoph Hellwig wrote:
> Hi Michal,
> 
> can you try the patch below?

The machine has been running with it for 12 hours now without any
apparent problem. Without the patch it crashed once after ~10 minutes
and once after ~3 hours (then I switched back to 4.20). Thanks a lot for
quick help.

Tested-by: Michal Kubecek <mkubecek@xxxxxxx>

> 
> ---
> From 6b22ae23a1971646dacc8a0ad313a6329a04cf98 Mon Sep 17 00:00:00 2001
> From: Christoph Hellwig <hch@xxxxxx>
> Date: Fri, 4 Jan 2019 09:50:33 +0100
> Subject: x86/amd_gart: fix unmapping of non-GART mappings
> 
> In many cases we don't have to create a GART mapping at all, which
> also means there is nothing to unmap.  Fix the range check that was
> incorrectly modified when removing the mapping_error method.
> 
> Fixes: 9e8aa6b546 ("x86/amd_gart: remove the mapping_error dma_map_ops method")
> Reported-by: Michal Kubecek <mkubecek@xxxxxxx>
> Signed-off-by: Christoph Hellwig <hch@xxxxxx>
> ---
>  arch/x86/kernel/amd_gart_64.c | 10 +++++++++-
>  1 file changed, 9 insertions(+), 1 deletion(-)
> 
> diff --git a/arch/x86/kernel/amd_gart_64.c b/arch/x86/kernel/amd_gart_64.c
> index e0ff3ac8c127..2c0aa34af69c 100644
> --- a/arch/x86/kernel/amd_gart_64.c
> +++ b/arch/x86/kernel/amd_gart_64.c
> @@ -256,7 +256,15 @@ static void gart_unmap_page(struct device *dev, dma_addr_t dma_addr,
>  	int npages;
>  	int i;
>  
> -	if (dma_addr == DMA_MAPPING_ERROR ||
> +	if (WARN_ON_ONCE(dma_addr == DMA_MAPPING_ERROR))
> +		return;
> +
> +	/*
> +	 * This driver will not always use a GART mapping, but might have
> +	 * created a direct mapping instead.  If that is the case there is
> +	 * nothing to unmap here.
> +	 */
> +	if (dma_addr < iommu_bus_base ||
>  	    dma_addr >= iommu_bus_base + iommu_size)
>  		return;
>  
> -- 
> 2.20.1
>