[ofa-general] RE: osm_physp_calc_link_ops question

[Eitan Zahavi]

Hi Hal,

Good idea !
 

> -----Original Message-----
> From: Sasha Khapyorsky [mailto:sashak at voltaire.com] 
> Sent: Thursday, July 26, 2007 12:11 AM
> To: Hal Rosenstock
> Cc: OpenFabrics General; Eitan Zahavi; Yevgeny Kliteynik
> Subject: Re: osm_physp_calc_link_ops question
> 
> Hi Hal,
> 
> On 13:53 Wed 25 Jul     , Hal Rosenstock wrote:
> > 
> >  Both osm_lid_mgr.c:__osm_lid_mgr_set_physp_pi and  
> > osm_link_mgr.c:__osm_link_mgr_set_physp_pi call  
> > osm_port.c:osm_physp_calc_link_op_vls. In the case where the remote 
> > end is  invalid, the local VLCap is used as the 
> OperationalVLs. When 
> > the VLCaps at  the two ends of the link do not match, this is not a 
> > good thing. It causes  trap storms on the flow control 
> watchdog timer 
> > expiring. Wouldn't it be  better to leave this field as is in this 
> > case or would that cause some other  problem ?
> > 
> >  Same thing might also be true for link MTU but not as critical.
> 
> Looks like good idea for me. Would you care about patch?
> 
> Sasha
>