Re: [RFC] Registering non-contiguous memory

[Leon Romanovsky <leon-DgEjT+Ai2ygdnm+yROfE0A@public.gmane.org>]

[-- Attachment #1: Type: text/plain, Size: 1342 bytes --]

On Wed, Nov 01, 2017 at 10:56:25AM -0600, Jason Gunthorpe wrote:
> On Wed, Nov 01, 2017 at 11:11:42AM +0000, Alex Margolin wrote:
>
> >  struct verbs_context {
> >  	/*  "grows up" - new fields go here */
> > +	struct ib_mw * (*alloc_mw_ex)(struct ibv_mw_alloc_attr
> >  	*mw_alloc_attr);
>
> This patch is full of weird little mistakes like the above, wouldn't
> compile and doesn't really seem capture the proposed API.

Those RFCs are intended to present concept and implementation direction.
It is a little bit over-expectation to have working code and clean UAPI
at this stage.

>
> We are now asking for complete rdma-core patches before talking about
> merging new kernel uapi features. Please retry this RFC with the new
> requirement.

There are steps in development process, and first step before rushing
into implementation details is to talk about concept. This is exactly
what Alex did.

>
> This means a patch similar to the above, except functional, and all
> the man pages and documentation.

It already exists in this RFC, just in different format - inlined and
not in patch format.

>
> The text in the cover letter should go into either into a man page or
> a Documentation/ file.

At the implementation phase, yes.

>
> For very complex uapis like this, an simple example usage is also
> going to help.

Agree.

>
> Jason

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]