[ofa-general] uDAPL libdat2.so version [PATCH] udapl v1 and v2 - dat_create_psp_any() seed value wrong

[Tang, Changqing]

Arlin:
        I am running today's OFED tarball uDAPL v1 version, pure RDMA works,
but if I switch to SRQ mode, I got segfault in dat_srq_create(), I checked
the parameters to dat_srq_create(), I don't see anything wrong:

Core was generated by `/mpiscratch/ctang/test/pp.x'.
Program terminated with signal 11, Segmentation fault.
#0  0x00002aaaabda5c3b in dat_srq_create () from /usr/lib64/libdat.so

(gdb) print hpmp_udapl->ia_handle
$7 = (DAT_IA_HANDLE) 0x1
(gdb) print hpmp_udapl->pz_handle
$8 = (DAT_PZ_HANDLE) 0xc4540e0
(gdb) print srq_attr
$9 = {max_recv_dtos = 16, max_recv_iov = 1, low_watermark = 0}
(gdb) print &srq_attr
$10 = (DAT_SRQ_ATTR *) 0x7fffe64fb760
(gdb) print &hpmp_udapl->srq_handle
$11 = (DAT_SRQ_HANDLE *) 0xc448bb8


Do you have any idea ?


Thanks.
--CQ



> -----Original Message-----
> From: Arlin Davis [mailto:ardavis at ichips.intel.com]
> Sent: Thursday, February 07, 2008 6:38 PM
> To: Tang, Changqing
> Cc: OpenFabrics General
> Subject: Re: [ofa-general] uDAPL libdat2.so version [PATCH]
> udapl v1 and v2 - dat_create_psp_any() seed value wrong
>
> Tang, Changqing wrote:
> > Yes, the error is from  dat_psp_create_any().
> >
> > After changing seed value to 1024, do I still get any
> warning message ?
>
> You may see some in-use warning messages. I noticed rdma_cm
> changed the return code from EBUSY to EADDRINUSE so the
> warning message is not suppressed like it should be.
>
> >
> > Can I get this fix from tomorow's tarball ?
> >
>
> OFED pulls from DAPL package releases so I would have to
> roll-up another package. In the meantime, you could install
> the dapl src rpm that comes with OFED, apply the patch, and
> rebuild the libraries to test the fix.
>
> -arlin
>