Re: [PATCH v3 for-next 02/33] IB/core: Add kref to IB devices

From: Matan Barak <matanb-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
To: Jason Gunthorpe
	<jgunthorpe-ePGOBjL8dl3ta4EC/59zMFaTQe2KTcn/@public.gmane.org>
Cc: "Hefty,
	Sean" <sean.hefty-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>,
	Matan Barak <matanb-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>,
	Or Gerlitz <gerlitz.or-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>,
	Somnath Kotur
	<somnath.kotur-laKkSmNT4hbQT0dZR+AlfA@public.gmane.org>,
	Roland Dreier <roland-DgEjT+Ai2ygdnm+yROfE0A@public.gmane.org>,
	"linux-rdma-u79uwXL29TY76Z2rM5mHXA@public.gmane.org"
	<linux-rdma-u79uwXL29TY76Z2rM5mHXA@public.gmane.org>
Subject: Re: [PATCH v3 for-next 02/33] IB/core: Add kref to IB devices
Date: Fri, 1 May 2015 09:41:55 +0300	[thread overview]
Message-ID: <CAAKD3BDj8MWhOu0wK7hRWHWf3Q6PmwYzr7Yp3oc6jpGBTs28JA@mail.gmail.com> (raw)
In-Reply-To: <20150430212842.GB7709-ePGOBjL8dl3ta4EC/59zMFaTQe2KTcn/@public.gmane.org>

On Fri, May 1, 2015 at 12:28 AM, Jason Gunthorpe
<jgunthorpe-ePGOBjL8dl3ta4EC/59zMFaTQe2KTcn/@public.gmane.org> wrote:
> On Thu, Apr 30, 2015 at 07:21:08PM +0000, Hefty, Sean wrote:
>> > But, how could RDMA CM work? Inbound CM messages will be filtered if
>> > the IP is not in the HW GID table?
>>
>> I'm not understanding the issue.
>>
>> If a device has some requirement to program its assigned IP
>> addresses into some HW table, I don't see why upper layers should be
>> made to care.
>
> Okay, I was only thinking about the first couple patches in this
> series.  Three drivers will have this HW gid table, so having a driver
> helper library in the common code makes sense to me.
>
> But then the series it just seems to go crazy - what is with all these
> net dev patches? Now we are doing something with bonding? And a lot of
> screwy looking rocev2 gid type stuff? And driver updates? And 33 patches worth?
>

Actually, we're in the middle of splitting this series into two series
- the first introduces
the roce_gid_cache management and the second will introduce RoCE V2.
The bonding changes are crucial - the mlx4 driver currently support bonding
and changing to a new mechanism without supporting bonding will cause
a regression.

> You are totally right, this GID index and GID type stuff is getting
> *everywhere*, and it is hard to follow the *why* of all the changes
> are really needed.
>
> Matan, if you want to progress this, then split it up.  Make a patch
> to library the existing roce GID table driver code and update the
> drivers. Just that bit alone has already spawned a huge discussion.
>

It's already in progress. The new series is expected to be a lot smaller.

> Work the other ancillary unrelated patches in small chuncks through
> their trees.
>

The DRV patch was already sent to net. Regarding the bonding and ipv6
patch - we can't change them without a good reason which relies in this
patchset.

> Then come back to rocev2 with an actual core API proposal patch set
> that can be discussed.
>
> Honestly, I'm not willing to even look at a patch set this big and
> rambly.

Ok, hopefully we'll have something soon enough.

>
> Jason

Thanks a lot for the review and your comments.

Matan

> --
> 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