[dpdk-dev] [PATCH] net/mlx5: fix build with strict alignment enabled

Slava Ovsiienko viacheslavo at mellanox.com
Tue Oct 15 17:06:06 CEST 2019


> -----Original Message-----
> From: Ali Alnubani <alialnu at mellanox.com>
> Sent: Tuesday, October 15, 2019 17:56
> To: dev at dpdk.org
> Cc: Slava Ovsiienko <viacheslavo at mellanox.com>; Jeremy Plsek
> <jplsek at iol.unh.edu>
> Subject: [PATCH] net/mlx5: fix build with strict alignment enabled
> 
> This patch converts some of the casts to unaligned integer types.
> 
> This fixes the following error, which is seen on x86, with gcc 7.4.0:
> 
> drivers/net/mlx5/mlx5_rxtx.c: In function ‘mlx5_tx_dseg_iptr’:
> drivers/net/mlx5/mlx5_rxtx.c:2740:9: error: passing argument 1 of ‘memcpy’
> makes pointer from integer without a cast [-Werror=int-conversion]
>   memcpy(dst, src, len);

This orphan line look strange:
>          ^~~
Could you, please, remove it?
The full compiler error message looks as non-appropriate either.
What do you think about replace one with description?

> The memcpy call is replaced with 2 copies of uint32 for better performance
> on ARM.
Replaced with completely unrolled copying for the data length up to 15B,
not only 2 copies of uint32.

I'm OK with this patch, after polishing commit message, you may append my
Acked-by to v2. Thanks.

With best regards, Slava

> 
> Fixes: 18a1c20044c0 ("net/mlx5: implement Tx burst template")
> 
> Reported-by: Jeremy Plsek <jplsek at iol.unh.edu>
> Signed-off-by: Ali Alnubani <alialnu at mellanox.com>
> ---
>  drivers/net/mlx5/mlx5_rxtx.c | 18 ++++++++++++------
>  1 file changed, 12 insertions(+), 6 deletions(-)
> 
> diff --git a/drivers/net/mlx5/mlx5_rxtx.c b/drivers/net/mlx5/mlx5_rxtx.c
> index 10d0ca116..d6a5106c5 100644
> --- a/drivers/net/mlx5/mlx5_rxtx.c
> +++ b/drivers/net/mlx5/mlx5_rxtx.c
> @@ -2747,27 +2747,33 @@ mlx5_tx_dseg_iptr(struct mlx5_txq_data
> *restrict txq,
>  	/* Unrolled implementation of generic rte_memcpy. */
>  	dst = (uintptr_t)&dseg->inline_data[0];
>  	src = (uintptr_t)buf;
> +	if (len & 0x08) {
>  #ifdef RTE_ARCH_STRICT_ALIGN
> -	memcpy(dst, src, len);
> +		assert(dst == RTE_PTR_ALIGN(dst, sizeof(uint32_t)));
> +		*(uint32_t *)dst = *(unaligned_uint32_t *)src;
> +		dst += sizeof(uint32_t);
> +		src += sizeof(uint32_t);
> +		*(uint32_t *)dst = *(unaligned_uint32_t *)src;
> +		dst += sizeof(uint32_t);
> +		src += sizeof(uint32_t);
>  #else
> -	if (len & 0x08) {
> -		*(uint64_t *)dst = *(uint64_t *)src;
> +		*(uint64_t *)dst = *(unaligned_uint64_t *)src;
>  		dst += sizeof(uint64_t);
>  		src += sizeof(uint64_t);
> +#endif
>  	}
>  	if (len & 0x04) {
> -		*(uint32_t *)dst = *(uint32_t *)src;
> +		*(uint32_t *)dst = *(unaligned_uint32_t *)src;
>  		dst += sizeof(uint32_t);
>  		src += sizeof(uint32_t);
>  	}
>  	if (len & 0x02) {
> -		*(uint16_t *)dst = *(uint16_t *)src;
> +		*(uint16_t *)dst = *(unaligned_uint16_t *)src;
>  		dst += sizeof(uint16_t);
>  		src += sizeof(uint16_t);
>  	}
>  	if (len & 0x01)
>  		*(uint8_t *)dst = *(uint8_t *)src;
> -#endif
>  }
> 
>  /**
> --
> 2.23.0



More information about the dev mailing list