[ofa-general] [PATCH] opensm/osm_ucast_lash: fix buffer overflow

[Sven-Arne Reinemo]

On Mon, 2008-10-13 at 13:45 -0500, Robert Pearson wrote:
> Thanks Hal,
> 
> I grabbed the head of tree and recompiled with some print statements thrown
> in. As you say the '<<' fixes the decoding correctly and ibsim is reporting
> 4 not 1 so life is good and I see no problems in ibsim or opensm.
> 
> I was looking at the binary init values for port info for switches and it
> looked like the op_vl was set to 1 but when I ran opensm over it I saw 4
> which is the correct answer. The last point is moot in this case.
> 
> I artificially relaxed the limit on the number of VLs in my case and found
> that I needed 12 VLs to route my problem with lash so it is not going to
> work since the switches only support 8 data vls.

With regards to the number of layers required, LASH is sensitive to the
port numbering used in regular topologies such as meshes and tori. If
the cabling is consistent (0=W, 1=E, 2=N, 3=S) LASH requires 1 VL for 2d
meshes of any size (equals dimension order routing) and 3 VLs for 2d
tori of any size. For 3d tori it requires 5 VLs. This is observed based
on simulations with IBMgtSim.

I do not know if this is relevant for your problem, but thought I should
mention it.

Regards,
Sven-Arne

> 
> Regards,
> 
> Bob
> -----Original Message-----
> From: Hal Rosenstock [mailto:hal.rosenstock at gmail.com] 
> Sent: Monday, October 13, 2008 11:58 AM
> To: Robert Pearson
> Cc: Sasha Khapyorsky; OpenIB
> Subject: Re: [ofa-general] [PATCH] opensm/osm_ucast_lash: fix buffer
> overflow
> 
> On Sun, Oct 12, 2008 at 7:55 PM, Robert Pearson
> <rpearson at systemfabricworks.com> wrote:
> > I spent a little time looking at osm_ucast_lash.c and ibsim.
> >
> > It looks like ibsim reports vl_cap = 4 and op_vl = 1 by default for a
> > switch.
> 
> Yes, ibsim only has one canned template for this. I think it is mainly
> that ibnetdiscover output format doesn't include this information
> currently and something needs to be assumed.
> 
> > Osm_ucast_lash.c computes the minimum over all switches of op_vl as
> > extracted from the port info mads starting from (5 which would correspond
> to
> >
> > It then uses the encoded value as though it was an integer instead of an
> > encoding of an integer which seems wrong.
> 
> osm_ucast_lash.c::discover_network_properties fixes this up properly:
>       vl_min = 1 << (vl_min - 1);
>       if (vl_min > 15)
>               vl_min = 15;
> 
> > I am not yet sure when the SM is supposed to set the op_vl field away from
> > 1. If later then you are using the wrong value and should be comparing to
> > the decoded value of vl_cap instead.
> 
> I don't understand what you mean here. vl_cap would never be right to be
> used.
> 
> -- Hal
> 
> > -----Original Message-----
> > From: general-bounces at lists.openfabrics.org
> > [mailto:general-bounces at lists.openfabrics.org] On Behalf Of Robert Pearson
> > Sent: Sunday, October 12, 2008 5:21 PM
> > To: 'Sasha Khapyorsky'; 'Hal Rosenstock'
> > Cc: 'OpenIB'
> > Subject: RE: [ofa-general] [PATCH] opensm/osm_ucast_lash: fix buffer
> > overflow
> >
> > How does lash know how many VLs are available? Especially, with ibsim.
> > Is there a way to have lash report the number of VLs required independent
> of
> > the type of switch used?
> >
> > -----Original Message-----
> > From: general-bounces at lists.openfabrics.org
> > [mailto:general-bounces at lists.openfabrics.org] On Behalf Of Sasha
> Khapyorsky
> > Sent: Thursday, October 09, 2008 2:06 PM
> > To: Hal Rosenstock
> > Cc: OpenIB
> > Subject: Re: [ofa-general] [PATCH] opensm/osm_ucast_lash: fix buffer
> > overflow
> >
> > On 07:04 Wed 08 Oct     , Hal Rosenstock wrote:
> >>
> >> Minor simplification as it seems like this could just be:
> >>
> >> if (++lanes_needed > p_lash->vl_min)
> >>   goto Error_Not_Enough_Lanes;
> >
> > Works for me. Thanks!
> >
> > Sasha
> > _______________________________________________
> > 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
> > _______________________________________________
> > 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
> >
> _______________________________________________
> 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