[ofa-general] PATCH 1/1 - fix kernel crash in mad.c when IB_MAD_RESULT_(SUCCESS|CONSUMED) returned

[Hal Rosenstock]

On Tue, 2008-05-20 at 15:41 -0700, Dave Olson wrote:
> On Tue, 20 May 2008, Hal Rosenstock wrote:
> 
> | On Tue, 2008-05-20 at 15:08 -0700, Dave Olson wrote:
> | > Ralph Campbell will submit this patch for ofed 1.3.1, also.
> | > 
> | > IB/MAD - fix crash when HCA returns IB_MAD_RESULT_SUCCESS|IB_MAD_RESULT_CONSUMED
> | > 
> | > This was observed with the hw/ipath driver, but could happen with any
> | > driver.  
> | 
> | Does this also occur with mthca/mlx4 ? Was the same thing that caused
> | this on ipath tried with either of these HCAs ?
> 
> No, it doesn't happen on mthca/mlx4, but presumably those drivers never
> return this value.  Yes, we tried it with them.  No problem is
> introduced on those by the change, either.
> 
> | > It's OFED bug 1027.
> | 
> | What's the port disable command which causes this crash ?
> 
> It was with the qlogic quicksilver command iba_portdisable
> and iba_portenable commands (they've been ported to work with OFED 1.3).
> 
> ibportstate refuses to work on non-switch nodes,

Right.

>  although in the past
> we've done local modifications to allow it to work on HCAs as well
> (we never submitted that change, because we assumed somebody explictly
> didn't want HCAs targetted).

Yes, it was a conscious choice discussed on the list. It's so people
don't shoot themselves in the foot as once a port is disabled, it might
not be so easy to enable it depending on the configuration (out of band
access might be needed to reenable). The only configurations where that
would be limiting would be CA<->CA and CA<->router without intervening
switches.

-- Hal

> Dave Olson
> dave.olson at qlogic.com
> _______________________________________________
> 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