[ofa-general] Bugs in opensm/libvendor
Hal Rosenstock
hal.rosenstock at gmail.com
Mon Dec 15 08:31:26 PST 2008
On Mon, Dec 15, 2008 at 11:25 AM, Mike Heinz <michael.heinz at qlogic.com> wrote:
> As long as any API change permits interoperation with spec-compliant
> SMs, that's fine.
That's exactly the idea; Allow for both (spec compliant and enhanced)
modes of operation.
-- Hal
>At the moment, the infinibanddiags tool, saquery, will
> fail when looking for path records on SMs other than OpenSM; that was
> the problem that actually sent me down this path.
>
> --
> Michael Heinz
> Principal Engineer, Qlogic Corporation
> King of Prussia, Pennsylvania
>
> -----Original Message-----
> From: Hal Rosenstock [mailto:hal.rosenstock at gmail.com]
> Sent: Monday, December 15, 2008 10:54 AM
> To: Mike Heinz
> Cc: general at lists.openfabrics.org
> Subject: Re: [ofa-general] Bugs in opensm/libvendor
>
> Mike,
>
> On Mon, Dec 15, 2008 at 10:41 AM, Mike Heinz <michael.heinz at qlogic.com>
> wrote:
>> Hal,
>>
>> I could be wrong, but as I understand it, this function does not
>> permit the user to request more than one path. The API takes a query
>> request which contains pair of lids or guids, but does not have a
>> field for specifying the number of paths.
>
> You're right; the path requests do not currently support num_paths
> unlike multipath requests. But there is reliance on this "feature". So
> the current API needs enhancement to support this IMO to preserve
> backward compatibility.
>
>> --
>> Michael Heinz
>> Principal Engineer, Qlogic Corporation King of Prussia, Pennsylvania
>>
>> -----Original Message-----
>> From: Hal Rosenstock [mailto:hal.rosenstock at gmail.com]
>> Sent: Monday, December 15, 2008 10:31 AM
>> To: Mike Heinz
>> Cc: general at lists.openfabrics.org; John Russo
>> Subject: Re: [ofa-general] Bugs in opensm/libvendor
>>
>> On Wed, Dec 10, 2008 at 1:31 PM, Mike Heinz <michael.heinz at qlogic.com>
>> wrote:
>>> While experimenting with the APIs in opensm/libvendor,
>
> In tree or out of tree application(s) ?
>
>> I was unable to
>>
>>> get the path record queries to work. Reviewing the error logs from
>>> the
>>
>>> SM,
>>
>> Which SM(s) ?
>
> Which SM(s) ?
>
> -- Hal
>
>>> I discovered that the APIs were not setting the required num_path
>> field.
>>> Here's the fix:
>>
>> The approach used breaks backward compatibility which IMO should be
>> preserved.
>> I think a better approach is as follows:
>> 1. Set num_paths in the application(s) as desired 2. In the library,
>> check for num_paths 0 or not. If not, then set the numbpath compmask
>> bit.
>>
>> -- Hal
>>
>>> --- osm_vendor_ibumad_sa.bak 2008-12-10 13:21:22.000000000 -0500
>>> +++ osm_vendor_ibumad_sa.c 2008-12-10 13:24:42.000000000 -0500
>>> @@ -615,7 +615,7 @@
>>> sa_mad_data.attr_offset =
>>> ib_get_attr_offset(sizeof(ib_path_rec_t));
>>> sa_mad_data.comp_mask =
>>> - (IB_PR_COMPMASK_DGID | IB_PR_COMPMASK_SGID);
>>> + (IB_PR_COMPMASK_DGID | IB_PR_COMPMASK_SGID |
>>> IB_PR_COMPMASK_NUMBPATH);
>>> sa_mad_data.p_attr = &path_rec;
>>> ib_gid_set_default(&path_rec.dgid,
>>> ((osmv_guid_pair_t *)
>>> (p_query_req-> @@ -625,6 +625,7 @@
>>> ((osmv_guid_pair_t *)
>> (p_query_req->
>>>
>> p_query_input))->
>>> src_guid);
>>> + path_rec.num_path = 1;
>>> break;
>>>
>>> case OSMV_QUERY_PATH_REC_BY_GIDS:
>>> @@ -634,7 +635,7 @@
>>> sa_mad_data.attr_offset =
>>> ib_get_attr_offset(sizeof(ib_path_rec_t));
>>> sa_mad_data.comp_mask =
>>> - (IB_PR_COMPMASK_DGID | IB_PR_COMPMASK_SGID);
>>> + (IB_PR_COMPMASK_DGID | IB_PR_COMPMASK_SGID |
>>> IB_PR_COMPMASK_NUMBPATH);
>>> sa_mad_data.p_attr = &path_rec;
>>> memcpy(&path_rec.dgid,
>>> &((osmv_gid_pair_t *)
>>> (p_query_req->p_query_input))-> @@ -642,6 +643,7 @@
>>> memcpy(&path_rec.sgid,
>>> &((osmv_gid_pair_t *)
>> (p_query_req->p_query_input))->
>>> src_gid, sizeof(ib_gid_t));
>>> + path_rec.num_path = 1;
>>> break;
>>>
>>> case OSMV_QUERY_PATH_REC_BY_LIDS:
>>> @@ -652,13 +654,14 @@
>>> sa_mad_data.attr_offset =
>>> ib_get_attr_offset(sizeof(ib_path_rec_t));
>>> sa_mad_data.comp_mask =
>>> - (IB_PR_COMPMASK_DLID | IB_PR_COMPMASK_SLID);
>>> + (IB_PR_COMPMASK_DLID | IB_PR_COMPMASK_SLID |
>>> IB_PR_COMPMASK_NUMBPATH);
>>> sa_mad_data.p_attr = &path_rec;
>>> path_rec.dlid =
>>> ((osmv_lid_pair_t *)
>> (p_query_req->p_query_input))->
>>> dest_lid;
>>> path_rec.slid =
>>> ((osmv_lid_pair_t *)
>>> (p_query_req->p_query_input))->src_lid;
>>> + path_rec.num_path = 1;
>>> break;
>>>
>>> case OSMV_QUERY_UD_MULTICAST_SET:
>>> --- osm_vendor_mlx_sa.bak 2008-12-10 13:21:10.000000000 -0500
>>> +++ osm_vendor_mlx_sa.c 2008-12-10 13:24:07.000000000 -0500
>>> @@ -743,7 +743,7 @@
>>> sa_mad_data.attr_offset =
>>> ib_get_attr_offset(sizeof(ib_path_rec_t));
>>> sa_mad_data.comp_mask =
>>> - (IB_PR_COMPMASK_DGID | IB_PR_COMPMASK_SGID);
>>> + (IB_PR_COMPMASK_DGID | IB_PR_COMPMASK_SGID |
>>> IB_PR_COMPMASK_NUMBPATH);
>>> sa_mad_data.p_attr = &path_rec;
>>> ib_gid_set_default(&path_rec.dgid,
>>> ((osmv_guid_pair_t *)
>>> (p_query_req-> @@ -753,6 +753,7 @@
>>> ((osmv_guid_pair_t *)
>> (p_query_req->
>>>
>> p_query_input))->
>>> src_guid);
>>> + path_rec.num_path = 1;
>>> break;
>>>
>>> case OSMV_QUERY_PATH_REC_BY_GIDS:
>>> @@ -763,7 +764,7 @@
>>> sa_mad_data.attr_offset =
>>> ib_get_attr_offset(sizeof(ib_path_rec_t));
>>> sa_mad_data.comp_mask =
>>> - (IB_PR_COMPMASK_DGID | IB_PR_COMPMASK_SGID);
>>> + (IB_PR_COMPMASK_DGID | IB_PR_COMPMASK_SGID |
>>> IB_PR_COMPMASK_NUMBPATH);
>>> sa_mad_data.p_attr = &path_rec;
>>> memcpy(&path_rec.dgid,
>>> &((osmv_gid_pair_t *)
>>> (p_query_req->p_query_input))-> @@ -771,6 +772,7 @@
>>> memcpy(&path_rec.sgid,
>>> &((osmv_gid_pair_t *)
>> (p_query_req->p_query_input))->
>>> src_gid, sizeof(ib_gid_t));
>>> + path_rec.num_path = 1;
>>> break;
>>>
>>> case OSMV_QUERY_PATH_REC_BY_LIDS:
>>> @@ -789,6 +791,7 @@
>>> dest_lid;
>>> path_rec.slid =
>>> ((osmv_lid_pair_t *)
>>> (p_query_req->p_query_input))->src_lid;
>>> + path_rec.num_path = 1;
>>> break;
>>>
>>> case OSMV_QUERY_UD_MULTICAST_SET:
>>>
>>> --
>>> Michael Heinz
>>> Principal Engineer, Qlogic Corporation King of Prussia, Pennsylvania
>>>
>>> _______________________________________________
>>> general mailing list
>>> general at lists.openfabrics.org
>>> http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general
>>>
>>> To unsubscribe, please visit
>>> http://openib.org/mailman/listinfo/openib-general
>>>
>>
>
More information about the general
mailing list