[ofa-general] poll CQ failed -2 with connectX

Roland Dreier rdreier at cisco.com
Mon Nov 3 15:47:07 PST 2008


 > However, the userspace drivers used were indeed from OFED 1.3.1
 > and/or OFED 1.4, resulting in a mismatch between kernel-space and
 > userspace.
 > 
 > Specifically, ConnectX cards support XRC (Extended RC) in OFED 1.3.1
 > and OFED 1.4 (XRC was not present in OFED 1.2.5).  The 1.3.1 / 1.4
 > userspace libraries identified some of the QPs created by the OFED
 > 1.2.5 kernel modules as XRC QPs and returned an error as a result
 > (correctly indicating that these "XRC" qp's did not exist as XRC
 > qp's).

I think we need newer userspace to continue to work with old kernels;
it's a huge pain if someone needs to roll back userspace just to test
an older kernel (eg if bisecting a regression or something like that).

The simplest thing would be for libmlx4 to check if the kernel driver
reports the XRC capability, say when creating the first QP for a given
process, and treat the QPN bits appropriately depending on whether the
kernel supports XRC or not.

 - R.



More information about the general mailing list