Ubuntu
linux-source-2.6.15 package

add hotplug support for marvell discovery gigabit ethernet driver.

Bug #21792 reported by Sven Luther on 2005-09-16

Affects		Status	Importance	Assigned to	Milestone
	linux-source-2.6.15 (Ubuntu)	Fix Released	Medium	Ben Collins

Bug Description

As ubuntu moved away from using discover in favour of pure hotplug (or
derivative), the pegasos gigabit ethernet driver, mv643xx_eth, was no more
loaded automatically, which was problematic with u-i, and broke some other
stuff, like trying to use nfsroot over the gigabit ethernet port and other such
application that need the network interface.

This patch :

http://svn.debian.org/wsvn/kernel/dists/sid/linux-2.6/debian/patches-debian/powerpc-mv643xx-hotplug-support.patch?op=file&rev=0&sc=0

Scheduled for addition in the next round of debian kernels, adds the mv643xx_eth
module to the pci subsystem, in the same way that we used to map the discovery
II northbridge pci id in the discover method, and thus adds an hotplug event for
this device.

This is not an ideal solution, since we should instead create a builtin or
plateform kind of bus, where those devices reside, and add propper hotplug
events to all such devices, mostly used in the embedded world, but this is
probably something for after the ppc64/ppc reorganisation.

In any case, this code only affects hardware possessing a marvell discovery II
northbridge, and there is only the pegasos in the kind of market ubuntu supports
that uses it (others are probably routers and other such appliances).

Friendly,

Sven Luther

Revision history for this message

Ben Collins (ben-collins) wrote on 2005-12-11:

This bug has been fixed in an upcoming release kernel. Please keep an eye out
for linux-image-2.6.15-8.10, or higher in your next upgrade. If this upgrade
does not fix your problem, please reopen this bug report, and refresh any
information that was requested before.

Revision history for this message

Ben Collins (ben-collins) wrote on 2005-12-13:

Fixed in 2.6.15-8.10

Revision history for this message

Sven Luther (luther) wrote on 2005-12-19:

Yeah, much good does it do us for breezy support of the pegasos, unless you may
consider adding it in a bug fix release, which i would greatly appreciate, and
should be no risk for you given the nature of the patch.

Again this shows it is easier to push bug fixes upstream than to submit bug
reports, guess ubuntu is no different than debian here, at least for those
debian packages whose maintainers are sub-par or just too busy to take real care
of their packages.

Oh well,

Friendly,

Sven Luther

Revision history for this message

Colin Watson (cjwatson) wrote on 2005-12-19:

Sorry, but due to the high volume of the work required for kernel security
updates to our three supported stable releases alone, adding features is
unlikely to be high on the list. In any case, implementing the requested feature
in an update to breezy would require rebuilding the breezy installer, which
isn't something we have had the need to do for any stable release to date and so
we have no infrastructure for it. Furthermore, a working rebuild of the breezy
installer would require code changes to at least net-retriever to have it look
for udebs in breezy-updates, which is getting well out of scope for changes in
-updates.

It should be relatively easy for an interested third party to rebuild the breezy
kernel with this particular patch applied (although without the ABI changes
introduced by security fixes, to avoid having to change installer code to look
for udebs in breezy-updates) and publish a netboot installer image built against
that. I did something similar for Kickstart fixes to the hoary installer which
didn't make it into hoary, without ever opening the can of worms that uploading
a rebuilt hoary installer would have been.

Please refrain from making the type of comments in your second paragraph in
Ubuntu bug reports. This is not a means to discuss the mechanics of pushing
changes upstream, nor a forum for philosophical debate about the differences
between Debian and Ubuntu.

Revision history for this message

Sven Luther (luther) wrote on 2005-12-19:

Well, as i understand, a third party installer build should be ok, but the
resulting installed system will kill this patch at each kernel security upgrade,
so that this is not an option unless the patch is applied to the security tree
or whatever, you know best what i mean here, as we discussed this in the past.
So, please leave this bug open in breezy until breezy+1 is released or this
issue is fixed.

As for the fix itself, it is a few line patch, which has absolutely nil effect
on the ABI, nor on any ubuntu supported hardware for the pegasos, i did test
breezy beta and saw this as being a problem, and sent you a patch, so sorry for
feeling bitter that this never got applied or even considered, and when i
complained about this on #ubuntu-kernel, i got abuse and bullshit in return.

So, i guess there is no acceptable solution to this for breezy, so i guess this
bug should be kept open for breezy or whatever bugzilla will allow.

Workaround for users is to add mv643xx_eth to /etc/modules right now.

Friendly,

Sven Luther

Revision history for this message

Colin Watson (cjwatson) wrote on 2005-12-19:

Bugzilla has no version tracking support; there is no way to leave the bug open
for breezy without getting in the way of the kernel team's efforts to triage
bugs and get their bug list down to a manageable level.

Revision history for this message

Ben Collins (ben-collins) wrote on 2005-12-19:

(In reply to comment #3)
> Again this shows it is easier to push bug fixes upstream than to submit bug
> reports, guess ubuntu is no different than debian here, at least for those
> debian packages whose maintainers are sub-par or just too busy to take real care
> of their packages.

Yeah, this comment just got your request no where. In fact, it made me less
likely to help you out.

Really, it's not my decision. I only fix the bugs. Getting them in to an older
kernel is up to a select few.

As for this situation with backporting bug fixes, It's a very important issue
for me. I'd love to see things like this get backported. But because of
infrastructure, it's just too damn difficult. That is something I will address
for dapper+1 (too late to start now).