On 10/31/2018 7:37 AM, Håkon Bugge wrote:
> Hi,
>
> Somewhat confused. We have in ibacm/include/acm_mad.h:
>
> struct ib_sa_mad {
> []
> __be16 status;
> []
> };
>
> In ibacm/prov/acmp/src/acmp.c, acmp_process_join_resp():
>
> if (mad->status) {
> acm_log(0, "ERROR - join response status 0x%x\n", mad->status);
> goto out;
> }
>
> So, I thought, simple enough, the above is an endian conversion bug, and the code should read:
>
> if (be16toh(mad->status)) {
> acm_log(0, "ERROR - join response status 0x%x\n", be16toh(mad->status));
> goto out;
> }
>
> On a real system, I saw:
>
> acmp_process_join_resp: ERROR - join response status 0x2
>
> So, since the conversion is missing, the status value is actually 0x200.
>
> Looking at IBTA, MAD Common Status Field Bit Values, status 0x200 is (bit #0) not busy, (bit #1) no redirection, (bits #2-4) no invalid fields, and a Class Specific status of 2, which from Table SA MAD Class-Specific Status Encodings is ERR_REQ_INVALID.
>
> Is this the correct interpretation?
Yes.
> If so, should the status print divide the Common and Class-Specific in two? Or are we only interested in the Class-Specific error here?
I think it should handle both common and SA class specific errors here.
-- Hal
>
> Thxs, Håkon
>
>
>
