[openib-general] [PATCH for-2.6.18] IB/cma: option to limitMTU to 1K

Hal Rosenstock halr at voltaire.com
Wed Sep 13 14:57:27 PDT 2006


On Wed, 2006-09-13 at 17:49, Michael S. Tsirkin wrote:
> Quoting r. Hal Rosenstock <halr at voltaire.com>:
> > Subject: Re: [PATCH for-2.6.18] IB/cma: option to limitMTU to 1K
> > 
> > On Wed, 2006-09-13 at 17:09, Michael S. Tsirkin wrote:
> > > Quoting r. Hal Rosenstock <halr at voltaire.com>:
> > > > Subject: Re: [PATCH for-2.6.18] IB/cma: option to limitMTU to 1K
> > > > 
> > > > On Wed, 2006-09-13 at 15:05, Michael S. Tsirkin wrote:
> > > > > Quoting r. Hal Rosenstock <halr at voltaire.com>:
> > > > > > > If I understood Michael's comment properly, this will have the side
> > > > > > > effect that IPoIB won't work since IPoIB requires 2K MTUs.  As far as
> > > > > > > I know, there is no way to specify whether a path is needed for UD vs.
> > > > > > > RC in the path query.
> > > > > > 
> > > > > > I don't know how either. I don't think it can be done (at least
> > > > > > currently per the standard).
> > > > > 
> > > > > We don't really need to know whether path is for RC or UD QP.
> > > > > IPoIB needs MTU >= 2K so it should set mtu selector to >= 2K.
> > > > 
> > > > That's the default and not the minimum MTU (for IPoIB).
> > > 
> > > How isn't it?
> > 
> > Look at RFC 4391 as to the requirement.
> 
> I'm talking about our implementation not the spec.

Don't we risk interop issues by relying on things not required in the
spec ?

> > > By default, IPoIB reports 2K MTU to linux.
> > > So it will get 2K packets, and since IB swiches
> > > can not fragment packets, they will simply get dropped.
> > 
> > With ifconfig, the MTU can be changed. Fragmentation is at the IP layer
> > in the end station stack, not the IB switches.
> 
> AFAIK linux won't fragment packets that do not exceed MTU and MSS.
> 
> > > I conclude that IPoIB by default requires minimum mtu of 2K.
> > > Right?
> > 
> > Not minimum.
> > 
> > > And it's not a problem since all HCAs support 2K.
> > 
> > or more but it could be less per the RFC.
> 
> Again, if IPoIB implementation does not need 2K mtu there's
> no problem to give it 1K in path. If it wants 2K MTU it must
> set selector accordingly.





More information about the general mailing list