Re: [PATCH rdma-next V2 5/5] IB/core: Integrate IB address resolution module into core

[Knut Omang <knut.omang-QHcLZuEGTsvQT0dZR+AlfA@public.gmane.org>]

On Mon, 2016-05-30 at 09:33 +0300, Leon Romanovsky wrote:
> On Mon, May 30, 2016 at 07:33:14AM +0200, Knut Omang wrote:
> > On Mon, 2016-05-23 at 14:09 +0000, Mark Bloch wrote:
> > > 
> > > > -----Original Message-----
> > > > From: Doug Ledford [mailto:dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org]
> > > > Sent: Monday, May 23, 2016 5:06 PM
> > > > To: leon-DgEjT+Ai2ygdnm+yROfE0A@public.gmane.org; Weiny, Ira <ira.weiny-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
> > > > Cc: Mark Bloch <markb-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>; Jason Gunthorpe
> > > > <jgunthorpe-ePGOBjL8dl3ta4EC/59zMFaTQe2KTcn/@public.gmane.org>; linux-rdma-u79uwXL29TY76Z2rM5mHXA@public.gmane.org;
> > > > hal.rosenstock-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org
> > > > Subject: Re: [PATCH rdma-next V2 5/5] IB/core: Integrate IB address
> > > > resolution module into core
> > > > 
> > > > On 5/23/2016 4:58 AM, Leon Romanovsky wrote:
> > > > > On Wed, May 18, 2016 at 07:36:03PM +0000, Weiny, Ira wrote:
> > > > > > > On 05/18/2016 02:28 PM, Mark Bloch wrote:
> > > > > > > > 
> > > > > > > > 
> > > > > > > > > -----Original Message-----
> > > > > > > > > From: Doug Ledford [mailto:dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org]
> > > > > > > > > Sent: Wednesday, May 18, 2016 8:59 PM
> > > > > > > > > To: Jason Gunthorpe <jgunthorpe-ePGOBjL8dl3ta4EC/59zMFaTQe2KTcn/@public.gmane.org>
> > > > > > > > > Cc: Mark Bloch <markb-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>; leon-DgEjT+Ai2ygdnm+yROfE0A@public.gmane.org;
> > > > > > > > > linux-
> > > > > > > > > rdma-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
> > > > > > > > > Subject: Re: [PATCH rdma-next V2 5/5] IB/core: Integrate
> > > > > > > > > IB address
> > > > > > > > > resolution module into core
> > > > > > > > > 
> > > > > > > > > On 05/18/2016 01:15 PM, Jason Gunthorpe wrote:
> > > > > > > > > > On Wed, May 18, 2016 at 10:51:02AM -0400, Doug Ledford
> > > > > > > > > > wrote:
> > > > > > > > > > > moves it into the ib_core and also keep the new
> > > > > > > > > > > ib_netlink stuff in
> > > > > > > > > > > ib_core.  So, fewer modules, not more.
> > > > > > > > > > 
> > > > > > > > > > What about getting rid of ib_sa as well so we can avoid
> > > > > > > > > > that dynamic
> > > > > > > > > > netlink registration patch too?
> > > > > > > > > 
> > > > > > > > > That's fine too.
> > > > > > > > Integrating ib_sa into ib_core also means to do the same
> > > > > > > > for ib_mad.
> > > > > > > > If you agree with it (ib_sa & ib_mad becoming part of
> > > > > > > > ib_core), I'll
> > > > create
> > > > > > > the patches.
> > > > > > > 
> > > > > > > Ira, Hal?
> > > > > > > 
> > 
> > Having just been through a development of a new HCA from scratch, I
> > found it very useful to have the separation between the core IB
> > functionality and what I perceive as clients of that core
> > infrastructure (ib_mad and ib_sa in particular). 
> > 
> > The separation feels natural, and it allows a more piecemeal approach -
> > a device can be loaded and tested with one WR at a time. Once ib_mad
> > gets loaded, a lot of "traffic" is created and just about the whole HCA
> > under development needs to work to support it.
> 
> Adding new device to IB is not an easy task and we clearly need to improve in
> that area, however these merges are targeted to the end-user of IB stack which
> constantly needs to load all these modules.

In a well configured system, this should all happen automatically -
this is why we have udev? End users should never have to struggle at
this level. 

> For the easy development, you can simply revert them in your tree.

:-)
Development and debugging needs aside: Keeping separation between
layers is still a good idea IMHO. 

> Hope it helps.
> 
> > 
> > IMHO, if these "clients" gets joined with ib_core, the path for
> > newcomers will be (even) steeper. I would appreciate if that particular
> > separation can be kept. I am fine with the other merges, and it might
> > as well be a good idea to join ib_mad and ib_sa,
> > 
> > Thanks,
> > Knut Omang
> > (lead developer of Oracle's new IB HCA, soon to appear here..:-) )
> 
> Terrific,
> We are eager to see it.

Good things come to those who wait :-) 
Still playing catch-up with the activity here as is evident from my
late comment in this thread..

Knut

> > 
> > > > > > I don't see a problem with that.
> > > > > 
> > > > > Doug,
> > > > > 
> > > > > Will it be acceptable to you if Mark base his patches on this
> > > > > assumption?
> > > > 
> > > > Yes.
> > > > 
> > > 
> > > Actually the new series is already posted, 
> > > http://marc.info/?l=linux-rdma&m=146366717406108&w=2
> > > if you can have a look at it'll be great.
> > > 
> > > Mark
> > > --
> > > 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