[openib-general] Re: SMI Patch

[Hal Rosenstock]

On Tue, 2004-11-16 at 12:42, Sean Hefty wrote:
> Hal Rosenstock wrote:
> 
> > Hi,
> > 
> > I have a patch which I think fixes the SMI issues for an end node. There
> > is more to be done for switch support but this hopefully is sufficient
> > for SM support. Can you please validate it before I check it in ? I'm a
> > little gun shy about breaking the tree after last week's debacle. If it
> > works in your configurations, I will check it in.
> 
> I applied the patch to my local repository, and it worked fine.

Thanks for checking this out.

> One item of note is that my test system is connected into a switched 
> fabric with opensm running on the source forge stack.  When I load the 
> openib stack, the port goes to INIT.  It doesn't go to ACTIVE until I 
> unplug and re-insert the cable.  This is true with or without this 
> patch.  (I'm connected to a 16-port Mellanox switch.)  The systems 
> running the source forge stack do not see this issue.

Can you see whether any packets received make it to
ib_mad_recv_done_handler when the port stays in INIT ?

It seems weird that a cable reinsertion would bring it back to life.
Sounds like some sort of initialization issue that gets fixed on a cable
reinsertion. Is this reproducible every time or intermittent ?

-- Hal

> 
> - Sean