Re: [dpdk-dev] [PATCH] librte_lpm: define tbl entry reversely for big endian

[Bruce Richardson <bruce.richardson@intel.com>]

On Wed, Mar 04, 2015 at 02:34:12PM +0800, xuelin.shi@freescale.com wrote:
> From: Xuelin Shi <xuelin.shi@freescale.com>
> 
> This module uses type conversion between struct and int.
> Also truncation and comparison is used with this int.
> It is not safe for different endian arch.
> 
> Add ifdef for big endian struct to fix this issue.
> 
> Signed-off-by: Xuelin Shi <xuelin.shi@freescale.com>
> ---
>  lib/librte_lpm/rte_lpm.h | 19 +++++++++++++++++++
>  1 file changed, 19 insertions(+)
> 
> diff --git a/lib/librte_lpm/rte_lpm.h b/lib/librte_lpm/rte_lpm.h
> index 1af150c..08a2859 100644
> --- a/lib/librte_lpm/rte_lpm.h
> +++ b/lib/librte_lpm/rte_lpm.h
> @@ -96,6 +96,7 @@ extern "C" {
>  /** Bitmask used to indicate successful lookup */
>  #define RTE_LPM_LOOKUP_SUCCESS          0x0100
>  
> +#if RTE_BYTE_ORDER == RTE_LITTLE_ENDIAN
>  /** @internal Tbl24 entry structure. */
>  struct rte_lpm_tbl24_entry {
>  	/* Stores Next hop or group index (i.e. gindex)into tbl8. */
> @@ -117,6 +118,24 @@ struct rte_lpm_tbl8_entry {
>  	uint8_t valid_group :1; /**< Group validation flag. */
>  	uint8_t depth       :6; /**< Rule depth. */
>  };
> +#else
> +struct rte_lpm_tbl24_entry {
> +	uint8_t depth 	    :6;
> +	uint8_t ext_entry   :1;
> +	uint8_t valid	    :1;
> +	union {
> +		uint8_t tbl8_gindex;
> +		uint8_t next_hop;
> +	};
> +};
> +
> +struct rte_lpm_tbl8_entry {
> +	uint8_t depth 	    :6;
> +	uint8_t valid_group :1;
> +	uint8_t valid	    :1;
> +	uint8_t next_hop;
> +};
> +#endif
>  
>  /** @internal Rule structure. */
>  struct rte_lpm_rule {
> -- 
> 1.9.1
> 

Get an error compiling this up (using clang on FreeBSD).

  CC rte_lpm.o
  In file included from /usr/home/bruce/dpdk.org/lib/librte_lpm/rte_lpm.c:57:
  /usr/home/bruce/dpdk.org/lib/librte_lpm/rte_lpm.h:99:5: fatal error: 'RTE_BYTE_ORDER' is not defined, evaluates to 0 [-Wundef]
#if RTE_BYTE_ORDER == RTE_LITTLE_ENDIAN
      ^
      1 error generated.

Adding "#include <rte_byteorder.h>" should fix the issue.

Existing unit tests on IA (little endian) pass fine there-after, but I think for
this patch it would be good to have an ack from someone who can validate on
a big endian system, since this is what this patch is meant to enable.

/Bruce