Re: kernel memory registration

[Jason Gunthorpe <jgunthorpe-ePGOBjL8dl3ta4EC/59zMFaTQe2KTcn/@public.gmane.org>]

On Tue, Jul 14, 2015 at 11:24:26AM +0300, Sagi Grimberg wrote:
> 
> >
> >>Having a few schemes availabe in the core code that the driver can chose
> >>from seems like a much more sensible option.
> >
> >I think that makes sense, but several of the schemes we are working
> >with are effectively single-vendor schemes. Indirect MR and DIX are
> >good examples of things that only one vendor implements, and are not
> >standardized.
> >
> >In those cases it is hard to argue the core should provide support for
> >them.
> 
> I'm sorry but I disagree with the statement above. T10-DIF/DIX is a
> standard and *not* a vendor specific feature. Other storage protocols
> (FC, SAS, NVME) fully support it. So yes, the core *needs* to support it.

Sorry, I ment the exact details of how DIX is exposed through verbs is
only implemented by one vendor (eg IB_WR_REG_SIG_MR).

The core absolutely should provide an API for this, I'm not convinced
it should be at the MR level though....

> Moreover, the core stack poses memory alignment limitations for
> registration. The RDMA stack should allow an interface that can resolve
> these limitations if the device is capable of handling them. I'd say
> this is something all vendors should look at.

Sure would be nice..

Jason
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html