Are the PathRecord reserved components really reserved?

Are the PathRecord reserved components really reserved?

[Vangelis Tasoulas]

[-- Attachment #1: Type: text/plain, Size: 1435 bytes --]

Hello,

I am looking into the OFED code for the component mask bits of the
PathRecord requests and I need some further explanation.

The IB specification in section "15.2.5.16 PathRecord, table 207
PathRecord" defines the first two components as "Reserved". Each one of
them is 32 bits long and the description of the first one is "Reserved"
while the description of the second reserved component is "Offset for
alignment".

However, in the file "include/rdma/ib_sa.h" where the component mask
bits for the Path Records are defined, the 1st and 2nd component are
combined together under the name IB_SA_PATH_REC_SERVICE_ID.


#define IB_SA_PATH_REC_SERVICE_ID   (IB_SA_COMP_MASK( 0) |\
                                     IB_SA_COMP_MASK( 1))


The same definition exists in the OpenSM code in file
"include/iba/ib_types.h"


/* Path Record Component Masks */
#define  IB_PR_COMPMASK_SERVICEID_MSB     (CL_HTON64(((uint64_t)1)<<0))
#define  IB_PR_COMPMASK_SERVICEID_LSB     (CL_HTON64(((uint64_t)1)<<1))
.....
.....
#define  IB_PR_COMPMASK_SERVICEID (IB_PR_COMPMASK_SERVICEID_MSB | \
                                   IB_PR_COMPMASK_SERVICEID_LSB)


Why are these fields named as "SERVICE_ID"? I would expect something
like IB_SA_PATH_REC_RESERVED_1 and IB_SA_PATH_REC_RESERVED_2 according
to the specs. Are they really reserved or used, and if they are used,
when does this happen?

Vangelis


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 473 bytes --]

RE: Are the PathRecord reserved components really reserved?

[Weiny, Ira]

> 
> Hello,
> 
> I am looking into the OFED code for the component mask bits of the
> PathRecord requests and I need some further explanation.
> 
> The IB specification in section "15.2.5.16 PathRecord, table 207 PathRecord"
> defines the first two components as "Reserved". Each one of them is 32 bits
> long and the description of the first one is "Reserved"
> while the description of the second reserved component is "Offset for
> alignment".


The IBTA has published errata to the 1.2.1 specification which includes the use of many reserved fields.

I believe these errata are public on their web site.

Ira

> 
> However, in the file "include/rdma/ib_sa.h" where the component mask bits for
> the Path Records are defined, the 1st and 2nd component are combined
> together under the name IB_SA_PATH_REC_SERVICE_ID.
> 
> 
> #define IB_SA_PATH_REC_SERVICE_ID   (IB_SA_COMP_MASK( 0) |\
>                                      IB_SA_COMP_MASK( 1))
> 
> 
> The same definition exists in the OpenSM code in file "include/iba/ib_types.h"
> 
> 
> /* Path Record Component Masks */
> #define  IB_PR_COMPMASK_SERVICEID_MSB
> (CL_HTON64(((uint64_t)1)<<0))
> #define  IB_PR_COMPMASK_SERVICEID_LSB     (CL_HTON64(((uint64_t)1)<<1))
> .....
> .....
> #define  IB_PR_COMPMASK_SERVICEID (IB_PR_COMPMASK_SERVICEID_MSB |
> \
>                                    IB_PR_COMPMASK_SERVICEID_LSB)
> 
> 
> Why are these fields named as "SERVICE_ID"? I would expect something like
> IB_SA_PATH_REC_RESERVED_1 and IB_SA_PATH_REC_RESERVED_2 according
> to the specs. Are they really reserved or used, and if they are used, when does
> this happen?
> 
> Vangelis

Re: Are the PathRecord reserved components really reserved?

[Vangelis Tasoulas]

[-- Attachment #1: Type: text/plain, Size: 1850 bytes --]

Thanks Ira,

I found the corresponding document.

Vangelis

On 08/25/2014 04:49 PM, Weiny, Ira wrote:
>>
>> Hello,
>>
>> I am looking into the OFED code for the component mask bits of the
>> PathRecord requests and I need some further explanation.
>>
>> The IB specification in section "15.2.5.16 PathRecord, table 207 PathRecord"
>> defines the first two components as "Reserved". Each one of them is 32 bits
>> long and the description of the first one is "Reserved"
>> while the description of the second reserved component is "Offset for
>> alignment".
> 
> 
> The IBTA has published errata to the 1.2.1 specification which includes the use of many reserved fields.
> 
> I believe these errata are public on their web site.
> 
> Ira
> 
>>
>> However, in the file "include/rdma/ib_sa.h" where the component mask bits for
>> the Path Records are defined, the 1st and 2nd component are combined
>> together under the name IB_SA_PATH_REC_SERVICE_ID.
>>
>>
>> #define IB_SA_PATH_REC_SERVICE_ID   (IB_SA_COMP_MASK( 0) |\
>>                                      IB_SA_COMP_MASK( 1))
>>
>>
>> The same definition exists in the OpenSM code in file "include/iba/ib_types.h"
>>
>>
>> /* Path Record Component Masks */
>> #define  IB_PR_COMPMASK_SERVICEID_MSB
>> (CL_HTON64(((uint64_t)1)<<0))
>> #define  IB_PR_COMPMASK_SERVICEID_LSB     (CL_HTON64(((uint64_t)1)<<1))
>> .....
>> .....
>> #define  IB_PR_COMPMASK_SERVICEID (IB_PR_COMPMASK_SERVICEID_MSB |
>> \
>>                                    IB_PR_COMPMASK_SERVICEID_LSB)
>>
>>
>> Why are these fields named as "SERVICE_ID"? I would expect something like
>> IB_SA_PATH_REC_RESERVED_1 and IB_SA_PATH_REC_RESERVED_2 according
>> to the specs. Are they really reserved or used, and if they are used, when does
>> this happen?
>>
>> Vangelis
> 



[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 473 bytes --]