Re: [PATCH] dma-mapping: don't clear GFP_ZERO in dma_alloc_attrs

[Evgeniy Didin]

Adding linux-snps and linux-arch mailing lists.

> Revert the clearing of __GFP_ZERO in dma_alloc_attrs and move it to
> dma_direct_alloc for now.ÂÂWhile most common architectures always zero dma
> cohereny allocations (and x86 did so since day one) this is not documented
> and at least arc and s390 do not zero without the explicit __GFP_ZERO
> argument.
This patch fixed Ethernet issues on ARC HSDK.
https://www.spinics.net/lists/kernel/msg2762054.html

Tested-by: Evgeniy Didin <Evgeniy.Didin@xxxxxxxxxxxx>
> Fixes: 57bf5a8963f8 ("dma-mapping: clear harmful GFP_* flags in common code")
> Reported-by: Evgeniy Didin <Evgeniy.Didin@xxxxxxxxxxxx>
> Reported-by: Sebastian Ott <sebott@xxxxxxxxxxxxxxxxxx>
> Signed-off-by: Christoph Hellwig <hch@xxxxxx>
> ---
> Âinclude/linux/dma-mapping.h | 8 ++------
> Âlib/dma-direct.cÂÂÂÂÂÂÂÂÂÂÂÂ| 3 +++
> Â2 files changed, 5 insertions(+), 6 deletions(-)
> 
> diff --git a/include/linux/dma-mapping.h b/include/linux/dma-mapping.h
> index eb9eab4ecd6d..12fedcba9a9a 100644
> --- a/include/linux/dma-mapping.h
> +++ b/include/linux/dma-mapping.h
> @@ -518,12 +518,8 @@ static inline void *dma_alloc_attrs(struct device *dev, size_t size,
> Â	if (dma_alloc_from_dev_coherent(dev, size, dma_handle, &cpu_addr))
> Â		return cpu_addr;
> Â
> -	/*
> -	Â* Let the implementation decide on the zone to allocate from, and
> -	Â* decide on the way of zeroing the memory given that the memory
> -	Â* returned should always be zeroed.
> -	Â*/
> -	flag &= ~(__GFP_DMA | __GFP_DMA32 | __GFP_HIGHMEM | __GFP_ZERO);
> +	/* let the implementation decide on the zone to allocate from: */
> +	flag &= ~(__GFP_DMA | __GFP_DMA32 | __GFP_HIGHMEM);
> 
> Â	if (!arch_dma_alloc_attrs(&dev, &flag))
> Â		return NULL;
> diff --git a/lib/dma-direct.c b/lib/dma-direct.c
> index 1277d293d4da..c0bba30fef0a 100644
> --- a/lib/dma-direct.c
> +++ b/lib/dma-direct.c
> @@ -59,6 +59,9 @@ void *dma_direct_alloc(struct device *dev, size_t size, dma_addr_t *dma_handle,
> Â	struct page *page = NULL;
> Â	void *ret;
> Â
> +	/* we always manually zero the memory once we are done: */
> +	gfp &= ~__GFP_ZERO;
> +
> Â	/* GFP_DMA32 and GFP_DMA are no ops without the corresponding zones: */
> Â	if (dev->coherent_dma_mask <= DMA_BIT_MASK(ARCH_ZONE_DMA_BITS))
> Â		gfp |= GFP_DMA;