Re: [RFC ABI V2 5/8] RDMA/core: Add new ioctl interface

From: Parav Pandit <pandit.parav-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
To: Christoph Lameter <cl-vYTEC60ixJUAvxtiuMwx3w@public.gmane.org>
Cc: Jason Gunthorpe
	<jgunthorpe-ePGOBjL8dl3ta4EC/59zMFaTQe2KTcn/@public.gmane.org>,
	Matan Barak <matanb-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>,
	linux-rdma-u79uwXL29TY76Z2rM5mHXA@public.gmane.org,
	Doug Ledford <dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>,
	Sean Hefty <sean.hefty-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>,
	Liran Liss <liranl-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>,
	Haggai Eran <haggaie-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>,
	Tal Alon <talal-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>,
	Majd Dibbiny <majd-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>,
	Leon Romanovsky <leonro-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
Subject: Re: [RFC ABI V2 5/8] RDMA/core: Add new ioctl interface
Date: Wed, 27 Jul 2016 22:58:31 +0530	[thread overview]
Message-ID: <CAG53R5XZ9RVg-2yGFbr-7To1SMbRoGG3xiZHf-vNnGM4K+R4pw@mail.gmail.com> (raw)
In-Reply-To: <alpine.DEB.2.20.1607270954150.25436-wcBtFHqTun5QOdAKl3ChDw@public.gmane.org>

On Wed, Jul 27, 2016 at 8:33 PM, Christoph Lameter <cl-vYTEC60ixJUAvxtiuMwx3w@public.gmane.org> wrote:
>
> On Mon, 25 Jul 2016, Jason Gunthorpe wrote:
>
> > I can't help that IBTA created a different scheme for device
> > aggregation, addressing, and basically everything else. 'standard
> > convention' in the kernel really just means ethernet, and this isn't
> > like ethernet, except superficially.
>
> Well you can create your own non ethernet protocol. Why is there not an IB
> ulp that allows basic messaging, listen() and all the other stuff using
> standard system calls? There should be a netdev. I already see devices
> in /sys/class/infiniband/*! But they are not listed in /proc/net/dev!!!
>
> Why can we not use these with regular syscalls? Could provide this basic
> support for native infiniband? And maybe offload some of the need for
> esoteric IB ioctl functions that could just be provided by the standard
> calls? F.e. statistics and device configuration would be simplified and possible using
> the standard tools. IB devices have MACs as well. We could reuse quite a
> bit of software there.
>

I recall this functionality is being called as SDP (Socket Direct
Protocol) in Annex-4 of IB spec. 1.3 which used be present in older
2.6.x kernel as well as an ULP.
Vaguely recall it as non zcopy implementation.

Few older papers on this:
1. Asynchronous Zero-copy Communication for Synchronous Sockets in the
Sockets Direct Protocol (SDP) over InfiniBand
2. Zero Copy Sockets Direct Protocol over InfiniBand - Preliminary
Implementation and Performance Analysis
3. Transparently Achieving Superior Socket Performance Using Zero Copy
Socket Direct Protocol over IB
4. One more paper from Microsoft, I don't have ready reference too
which did zcopy implementation

Parav

> I think Christoph H made the point already that the wire format needs to
> stay the same and also the user verbs API (to some degree) but below that
> we could do pretty much what we want.
> --
> 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
--
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