RE: [ofa-general] Re: [PATCH] core/cm: improverequestmessage interpretation of subnet local fields
>One thing though, if the subnet local = 0 and the incoming hop limit ><= 1 a GRH will still not be used. Your patch should probably also >force the hop limit to 2 in this case and include a note for later >fixup.. I think this case would be better handled by rejecting the REQ. - Sean ___ general mailing list [email protected] http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
Re: [ofa-general] Re: [PATCH] core/cm: improverequestmessage interpretation of subnet local fields
On Mon, Sep 17, 2007 at 02:29:15PM -0500, Jim Hall wrote: > Is it valid to have a CM request message with subnet local = 1 and hop > limit > 1? Oop, I hadn't considered the subnet local field. That should probably unconditionally control the GRH, like your patch. Hmm, you know, there is compliance statement C9-43.1.2 that is not reflected in the flow diagram of Figure 81, so the GRH presence/absence is explicitly matched. My bad. > C9-43.1.2: For RC, RD and UC services, if a received packet is > consistent with the configuration of the QP (or EEC) with respect to > the presence or absence of a GRH, then the packet shall be > considered to have passed the GRH check, subject to the remaining > GRH checks described in the rest of Section 9.6.1.2 GRH Checks on > page 274. One thing though, if the subnet local = 0 and the incoming hop limit <= 1 a GRH will still not be used. Your patch should probably also force the hop limit to 2 in this case and include a note for later fixup.. Jason ___ general mailing list [email protected] http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
Re: [ofa-general] Re: [PATCH] core/cm: improverequestmessage interpretation of subnet local fields
Is it valid to have a CM request message with subnet local = 1 and hop limit > 1? If so, then it's not clear in the spec on how each side of the QP creation are to determine if they use GRHs or not. Can Mellanox be configured to accept both (LRH,LRH_GRH) on a RC QP at the same time? - Jim - Original Message - From: "Sean Hefty" <[EMAIL PROTECTED]> To: "'Jason Gunthorpe'" <[EMAIL PROTECTED]>; "Sean Hefty" <[EMAIL PROTECTED]> Cc: Sent: Monday, September 17, 2007 1:39 PM Subject: RE: [ofa-general] Re: [PATCH] core/cm: improverequestmessage interpretation of subnet local fields >I'm with Hal on this - why does this cause a problem? There is no IB packet verification check that tests if a GRH is present, only if it is presen it must be valid - so how can an extra correctly filled in GRH cause anything but degraded performance? ib_init_ah_from_path() uses the hop_limit in the path record to determine if a GRH should be used. It sets the address handle attributes (used to configure the QP) based on hop_limit > 1. If hop_limit is set incorrectly in the CM REQ, the path record formed by the CM based on data carried in the REQ could have invalid GRH values. It's possible that this is an active side CM issue, but that's not clear to me. - Sean ___ general mailing list [email protected] http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general ___ general mailing list [email protected] http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
