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

[Jesper Dangaard Brouer]

On 16/06/2023 21.21, Jakub Kicinski wrote:
> On Fri, 16 Jun 2023 20:59:12 +0200 Jesper Dangaard Brouer wrote:
> > +       if (mem_type == MEM_TYPE_PP_NETMEM)
> > +               pp_netmem_put_page(pp, page, allow_direct);
> > +       else
> > +               page_pool_put_full_page(pp, page, allow_direct);
>
> Interesting, what is the netmem type? I was thinking about extending
> page pool for other mem providers and what came to mind was either
> optionally replacing the free / alloc with a function pointer:
>
> https://github.com/torvalds/linux/commit/578ebda5607781c0abb26c1feae7ec8b83840768
>
> or wrapping the PP calls with static inlines which can direct to
> a different implementation completely (like zctap / io_uring zc).

I *LOVE* this idea!!!
It have been my master plan since day-1 to have other mem providers.
Notice how ZC xsk/AF_XDP have it's own memory allocator implementation.

The page_pool was never meant to be the final and best solution, I want
to see other, better and faster solutions competing with page_pool and
maybe some day replacing page_pool (I even see it as a success if PP get
depreciated and remove from the kernel due to a better solution).

See[1] how net/core/xdp.c simply have a switch statement
(is fast, because ASM wise it becomes a jump table):

 [1] 
https://github.com/torvalds/linux/blob/v6.4-rc6/net/core/xdp.c#L382-L402

> Former is better for huge pages, latter is better for IO mem
> (peer-to-peer DMA). I wonder if you have different use case which
> requires a different model :(

I want for the network stack SKBs (and XDP) to support different memory
types for the "head" frame and "data-frags". Eric have described this
idea before, that hardware will do header-split, and we/he can get TCP
data part is another page/frag, making it faster for TCP-streams, but
this can be used for much more.

My proposed use-cases involves more that TCP.  We can easily imagine
NVMe protocol header-split, and the data-frag could be a mem_type that
actually belongs to the harddisk (maybe CPU cannot even read this).  The
same scenario goes for GPU memory, which is for the AI use-case.  IIRC
then Jonathan have previously send patches for the GPU use-case.

I really hope we can work in this direction together,
--Jesper