Memory providers multiplexing (Was: [PATCH net-next v4 4/5] page_pool: remove PP_FLAG_PAGE_FRAG flag)

[Jakub Kicinski]

On Tue, 11 Jul 2023 09:05:02 -0300 Jason Gunthorpe wrote:
> On Tue, Jul 11, 2023 at 07:04:45AM +0200, Christoph Hellwig wrote:
> > On Mon, Jul 10, 2023 at 09:59:06PM -0700, Jakub Kicinski wrote:  
> > > Noob question - how does RDMA integrate with the out of tree junk?
> > > AFAIU it's possible to run the "in-tree" RDMA stack and get "GPU
> > > direct".  
> > 
> > I don't care and it has absolutel no business being discussed here.

My question was genuine. If you think the code is dog shit I will 
make no argument for merging it. I just get hives from looking at
proprietary code so I was hoping someone could explain how the
proprietary stacks get it done.

> > FYI at leat iWarp is a totally open standard.  

And I'm sure someone cares about that. I care about open source.

I think people in the networking world have a deeper understanding 
of standardization processes, and their practical implication for
open source and hack-ability. If all usable implementations are
proprietary the standard could as well not exist.

That may be a little hard to understand for folks coming from storage
and fabric worlds where the interesting bits of the implementation was
pretty much always closed.

> So is Infiniband, Jakub has a unique definition of "proprietary".

For IB AFAIU there's only one practically usable vendor, such an
impressive ecosystem!!

> RDMA works with the AMD and Intel intree drivers using DMABUF without
> requiring struct pages using the DRM hacky scatterlist approach.

I see, thanks. We need pages primarily for refcounting. Avoiding all
the infamous problems with memory pins. Oh well.