[openib-general] max_send_sge < max_sge

[Michael S. Tsirkin]

Quoting r. Talpey, Thomas <Thomas.Talpey at netapp.com>:
> Subject: Re: max_send_sge < max_sge
> 
> At 08:42 AM 6/28/2006, Michael S. Tsirkin wrote:
> >Quoting r. Talpey, Thomas <Thomas.Talpey at netapp.com>:
> >> Just some ideas. I feel your pain.
> >
> >Is there something that would make life easier for you?
> 
> A work-request-based IBTA1.2/iWARP-compliant FMR implementation.

Hmm. No an easy one :)
Just to clarify: what feature exactly do you want to use? The spec has 3
relevant compliance statements as far as I can see.  With respect to
fast registration:

	o10-37.2.6: If the HCA supports the Base Memory Management Extensions,
	the Fast Registration must take place before any subsequent Work
	Request on the same Send Queue is started.

It looks like Fast Registration can bypass previous work requests, so
existing FMR implementation already has this property I think.

So I am guessing that what you want is one of the
o10-37.2.19: Relaxed ordered, o10-37.2.20: Local Invalidate Fencing or
o10-37.2.21: Send with Invalidate.

Which one is it then?

-- 
MST