Re: [PATCH] MIPS: mm: Move some checks out of 'for' loop in DMA operations

[Markos Chandras <Markos.Chandras@imgtec.com>]

On 09/25/13 14:01, Jayachandran C wrote:
> The check cpu_needs_post_dma_flush() in mips_dma_sync_sg_for_cpu() and
> the check !plat_device_is_coherent() in mips_dma_sync_sg_for_device()
> can be moved outside the for loop.
>
> As a side effect, this also avoids a GCC bug that caused kernel compile
> to fail with the error:
>
> arch/mips/mm/dma-default.c: In function 'mips_dma_sync_sg_for_cpu':
> arch/mips/mm/dma-default.c:316:1: internal compiler error: in add_insn_before, at emit-rtl.c:3852
>
> This gcc failure is seen in Code Sourcery toolchains [e.g. gcc version
> 4.7.2 (Sourcery CodeBench Lite 2012.09-99)] after commit "MIPS: Optimize
> current_cpu_type() for better code."
> ---
>   arch/mips/mm/dma-default.c |   12 ++++--------
>   1 file changed, 4 insertions(+), 8 deletions(-)
>
> diff --git a/arch/mips/mm/dma-default.c b/arch/mips/mm/dma-default.c
> index e45e4b0..2e94185 100644
> --- a/arch/mips/mm/dma-default.c
> +++ b/arch/mips/mm/dma-default.c
> @@ -307,12 +307,10 @@ static void mips_dma_sync_sg_for_cpu(struct device *dev,
>   {
>   	int i;
>
> -	/* Make sure that gcc doesn't leave the empty loop body.  */
> -	for (i = 0; i < nelems; i++, sg++) {
> -		if (cpu_needs_post_dma_flush(dev))
> +	if (cpu_needs_post_dma_flush(dev))
> +		for (i = 0; i < nelems; i++, sg++)
>   			__dma_sync(sg_page(sg), sg->offset, sg->length,
>   				   direction);
> -	}
>   }
>
>   static void mips_dma_sync_sg_for_device(struct device *dev,
> @@ -320,12 +318,10 @@ static void mips_dma_sync_sg_for_device(struct device *dev,
>   {
>   	int i;
>
> -	/* Make sure that gcc doesn't leave the empty loop body.  */
> -	for (i = 0; i < nelems; i++, sg++) {
> -		if (!plat_device_is_coherent(dev))
> +	if (!plat_device_is_coherent(dev))
> +		for (i = 0; i < nelems; i++, sg++)
>   			__dma_sync(sg_page(sg), sg->offset, sg->length,
>   				   direction);
> -	}
>   }
>
>   int mips_dma_mapping_error(struct device *dev, dma_addr_t dma_addr)
>

Hi,

The patch looks good to me and it fixes the gcc problem indeed (the bug 
is also present in gcc-4.7.3  from the 2013.05 release)

Reviewed-by/Tested-by: Markos Chandras <markos.chandras@imgtec.com>