[openib-general] [PATCH RFC 1 of 5] IB/sa: QoS support

[Sean Hefty]

>  static const struct ib_field path_rec_table[] = {
> -	{ RESERVED,
> -	  .offset_words = 0,
> -	  .offset_bits  = 0,
> -	  .size_bits    = 32 },
> -	{ RESERVED,
> -	  .offset_words = 1,
> +	{ PATH_REC_FIELD(service_id),
> +	  .offset_words = 2,

I haven't read the proposed annex, but I would have expected this to be 
offset_words = 0.  Otherwise, it seems that the service_id would fall over the dgid.

>  	  .offset_bits  = 0,
> -	  .size_bits    = 32 },
> +	  .size_bits    = 64 },
>  	{ PATH_REC_FIELD(dgid),
>  	  .offset_words = 2,
>  	  .offset_bits  = 0,
> @@ -178,7 +174,7 @@ static const struct ib_field path_rec_ta
>  	  .offset_words = 12,
>  	  .offset_bits  = 16,
>  	  .size_bits    = 16 },
> -	{ RESERVED,
> +	{ PATH_REC_FIELD(priority),
>  	  .offset_words = 13,
>  	  .offset_bits  = 0,
>  	  .size_bits    = 12 },
> Index: linux-2.6/include/rdma/ib_sa.h
> ===================================================================
> --- linux-2.6.orig/include/rdma/ib_sa.h
> +++ linux-2.6/include/rdma/ib_sa.h
> @@ -109,8 +109,9 @@ enum ib_sa_selector {
>   * Reserved rows are indicated with comments to help maintainability.
>   */
>  
> -/* reserved:								 0 */
> -/* reserved:								 1 */
> +#define IB_SA_PATH_REC_SERVICE_ID		       (IB_SA_COMP_MASK( 0) | \
> +							IB_SA_COMP_MASK( 1))
> +

Does the annex redefine how component mask works, or does it make path records 
have a special case?  (I'll try to look at the annex;  it's just that this 
definition isn't obvious to me.)

- Sean