[ofa-general] [PATCH RFC] RDMA: New Memory Extensions.

Steve Wise swise at opengridcomputing.com
Thu May 15 07:16:41 PDT 2008



Talpey, Thomas wrote:
> At 07:49 PM 5/14/2008, Roland Dreier wrote:
>> also I wonder if it's clearer if we call this verb
>> ib_alloc_fast_reg_mr().
> 
> I have to disagree. Calling anything "fast" simply invites a "faster"
> thing to come along later. It's like calling something "new".
> 
> I say call it what it is - a work-request-based, alloc-phys-buffer-list,
> bind-pages-to-list, to-be-widely-supported memory registration.
> Obviously, the individual verbs need to be a bit more precise. :-)
> 
> Ralph - to answer your question who wants it, NFS/RDMA does, both
> client and server. I talked about requirements that it matches closely
> at Sonoma last month.
> 
> But Steve - aren't these capable of protecting memory at byte
> granularity? The word "page" in some of the names implies otherwise.
> 

The MR, once bound, defines the memory region at the byte granularity. 
The page list is just that: an array of DMA addresses to physical pages 
in memory.  The page list + the region length + the first byte offset 
define the region.

Steve.



More information about the general mailing list