[openib-general] NFS/RDMA client release for Linux 2.6.15

[Caitlin Bestler]

openib-general-bounces at openib.org wrote:
> Are we all talking about the same thing here? I think
> Christoph is just asking that the code use
> dma_map_[single|page|sg] instead of using page_to_phys or
> virt_to_phys. 
> 
> The rest of this talk is about memory registration strategies
> which is a different issue than having the code assume that
> bus address == phys addresses.
> 
Yes, that was the original point. And it is an important one.
But Roland's comment seemed to question the usage scenario,
and making sure everyone understands how applications use
RDMA (or perhaps more importantly want to use) is important.
Our choices on the verb layer APIs have very real impacts on
how applications work. I was disagreeing that windows and/or
FMRs are not of vital importance to many applications.

Part of building a viable long-term stack is shifting 
to thinking down from the application needs rather than
up from the hardware. A successful stack will eventually
support so many hardware options that the idea of working
up from the hardware will be self-evidently absurd. We
just need to start making that shift.