Skip site navigation (1)Skip section navigation (2)
Date:      Mon, 08 Aug 2016 20:27:04 +0000
From:      bugzilla-noreply@freebsd.org
To:        freebsd-bugs@FreeBSD.org
Subject:   [Bug 211528] [request] Mellanox ConnectX-3 drivers compiled into the base system
Message-ID:  <bug-211528-8-Yol60lPFIq@https.bugs.freebsd.org/bugzilla/>
In-Reply-To: <bug-211528-8@https.bugs.freebsd.org/bugzilla/>
References:  <bug-211528-8@https.bugs.freebsd.org/bugzilla/>

next in thread | previous in thread | raw e-mail | index | archive | help
https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=3D211528

Matthew Macy <mmacy@nextbsd.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |mmacy@nextbsd.org

--- Comment #4 from Matthew Macy <mmacy@nextbsd.org> ---
(In reply to Ben RUBSON from comment #3)
There are two problems with this:

The one that gives me say:
1) The linuxkpi that drm-next uses is very different from the linuxkpi in H=
EAD.
Adding COMPAT_LINUXKPI to HEAD would essentially preclude ever supporting
anything newer than Haswell on FreeBSD with a GENERIC kernel config. It's s=
afe
to say that there are a lot more users of X11 on Intel hardware built within
the last 3 years than there are IB users.

The one that really bother me:
2) Once a driver is in GENERIC there is no established mechanism for removi=
ng
it.  The kernel is reaching upwards of 30 MB. Unlike Linux and everything e=
lse,
FreeBSD has no namespacing for modules statically linked in to the kernel. =
I've
hit namespace conflicts between the linuxkpi and 20 year old raid drivers. =
The
driver has no reason to be there now and should have its own namespace even=
 if
it does.

FreeBSD is already several times slower booting than linux. A large part of
this is due to the fact that it runs the probe routine for every PCI driver=
 at
every PCI node in the device tree.

Thus with any driver being added to GENERIC, the question should be: "is th=
is
device needed to work out of the box?" For drivers like em or igb the answe=
r is
yes. For most things the answer is no.=20


I assume that you're not really asking to further slow down boot times, blo=
at
the kernel, and create a perpetual burden on developers. What you probably
really want is for device support to automatically load the modules as need=
ed.
If that's the case, take a look at Warner's initial work and see if you're =
able
to apply that to mlx:=20
http://bsdimp.blogspot.com/2016/01/details-on-coming-automatic-module.html

--=20
You are receiving this mail because:
You are the assignee for the bug.=



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?bug-211528-8-Yol60lPFIq>