Minutes of the Debian linux-2.6 Group Meeting
Debian Kernel Group Meeting
===========================
Report by maximilian attems for the Debian Project Kernel List
The meetings were held on 3 sessions over the period of the
2 days during the Paris mini-Debconf, October 2010.
"These are the notes from the formal sessions, several informal
conversations happened which I have not made notes of, nor recorded
here." --Vince
=================================================================
Session 1
=========
1st meeting session from 10:00-12:30 on 30th October 2010
Present:
Bastian Blank (waldi)
Ben Hutchings (bwh)
maximilian attems (maks)
Feature patches for wheezy
--------------------------
Input by dann frazier was noted:
"He's still good w/ what we agreed to in Portland."
Vserver is not trying to upstream. Openvz can't anymore do checkpointing
since 2.6.18 and is as well staying out of vanilla. The virtualisiation
feature patches are thus in deprecated status, they will be provided as
is, but won't be reinstated for Wheezy. It was noted that past 2.6.32
Linux containers made several nice progresses. It was agreed on
that the lxc userspace is the future recommended way.
Concerning Xen: The upstream merge of dom0 and the impressive shrink of
the patch set led to the conclusion that for the wheezy kernel that
a separate Xen feature branch anymore might not be necessary anymore.
We think to be able to add the missing patches to linux-2.6 itself.
RT is a new interesting technology that people seem to ask for.
We appeal for interested maintainer to join in in order to have it
for Wheezy.
On the TODO is to reformulate the deprecation notice in the release notes
by dannf and maks.
Switch to git
-------------
The switch is agreed, welcomed and looked forward post Squeeze release.
Once the stabilisation and backport period leaves enough time for that.
3 submodules parts were suggested by waldi:
- config
- infrastructure (scripts, maintainer)
- linux-2.6 (maincode)
This would help not having to update infrastructure in separate branches
as we have to do now. The actual repository structure was left open for
further discussions.
Automated tests
---------------
This is a pressing task, we should know sooner if the specific
branch compiles and boots.
waldi shouts for reprepro, sbuild and wannbuild. Build server
needed between 6-12 hours and back then we didn't have debug yet.
People were using and testing our builds.
Action item by Ben to poke Vince for a Debian solution with
enough horse power and no bandwidth troubles.
Also the automated tests proposed at the kernel summit are to be watched.
Integrating kernel-wedge and d-i kernel packages
------------------------------------------------
We have more information then kernel-wedge plus Debian Installer
Team lacks manpower to keep it update to current linux-2.6.
The action item is to replace kernel-wedge and implement the
modules splitting properly by specific dirs (like drivers/ata or
drivers/block similar to initramfs-tools).
waldi volunteers to take a look on it and implement successor.
bwh will talk with Debian boot people including otavio.
make deb-pkg
------------
Most noted missing feature is the corresponding source to the
built deb. bwh started to look at it and plans to implement it.
maks queues a trivial patch to fix it for "paranoid" umasks.
It was further noted that thanks to the upstream linux-2.6
integration the script has seen fixes and thus improvements
by the general community. The handiness to have it around
in any git or tarball improves it's uptake.
Further documentation usage will be added to wiki or kernel
handbook by maks.
patch reduction for wheezy
--------------------------
Several patches got pushed or landed since 2.6.32, thus state
is in general good. Speakup landed finally in 2.6.37. We may
be able to use the new build configs. It was noted that noone
uses updateoldconfig thus reportallconfig beeing the interesting
target. bwh will resend his KBUILD_VERBOSE patch for both
versions of config.
We really hope to not have aufs, but instead the vfs union mounts.
lpar-console is an old patch from waldi that is actually a partial
revert on powerpc console. Action item from waldi to push it upstream.
Another go at sending doc-build-parallel.patch upstream, by bwh.
(bugfix/ia64/hardcode-arch-script-output.patch may become redundant ...,
sit it out). xorg people tell us to drop
i915-autoload-without-CONFIG_DRM_I915_KMS.patch
Ask aurel32 about sh4 patch, item by maks.
=================================================================
Session 2
=========
2nd meeting seession from 12:00-13:00 on 31st October 2010
3rd meeting seession from 14:00-14:30 on 31st October 2010
Present:
Bastian Blank (waldi)
Ben Hutchings (bwh)
Julien Cristau (jcristau)
maximilian attems (maks)
DRM updates in squeeze
----------------------
The shared tree with Ubuntu 2.6.33 stable tree was pointed out as
base, but might be missing stuff. maks will point gregkh to that
tree and see if he is interested to take some into 2.6.33 stable.
Ben did check the .34 stable tree patches and will check if we miss
some .35 drm ones that maybe are applicable.
Regression from -26 seems reproducible by bwh with graphical programs
using clutter. bwh will investigate those. It may have a link to the
new resume trouble people are reporting.
jcristau points out that upstream for 2.6.36 and newer has lots of
edp changes. He will investigate what can be backported as this
might be needed for several HP and Dell laptop's.
Debian intel xorg driver is missing patch to bail out earlier when
kms is off so the server will fall back to vesa, instead of erroring out.
There is an unknown status of nouveau on ppc and sparc. jcristau will
ask davem. Kicking out nv out of testing is consinderated.
Trouble is that sparc and powerpc have no vesa fb fallback but only
openfirmware fb. Radeon KMS is a known regression from UMS (can't
change backlight, and can't suspend/resume at least on powerpc).
xorg should currently remove modeset on these archs.
config wheezy checks
--------------------
maks will check for deprecated items so that we can turn them off early
in the release and not have the trouble at the end.
Ben proposed to split the generated linux images in general stuff and
a modules stuff that contains the rest. The security aspect would be
positive as many bugs relate to code that is not usually loaded
by the users on purpose (see rds or TIPC).
The big benefit would also be the space reduction of our generated
linux images. We have anyway to do the split of modules for d-i.
So it can be done here again.
waldi proposes to remove old untouched stuff like ax25 and atm.
waldi will to check for duplicate device id's, there might be still
enough lingering.
initramfs-tools dev report
--------------------------
The development since squeeze saw the addition of additional Maintainer
Michael Prokop. The MODULES=dep initramfs generation should be a robust
choice in between even for special roots settings. The netbooting
benefited from klibc ipconfig fixes of this spring on big clusters.
Boot speed is enhanced thanks to Ubuntu change in using build
precached script ordering. Several compressions are supported now.
klibc buffered stdio branch is scheduled to be merged for next upstream
version. It should allow to easier port userspace beyond kexec and
existing shells to klibc, which openembedded pushes. Together with new
devfs 2.0 and the force loading of the modules no other libc is required
in that minimal initramfs. Google uses klibc and pushes development.
dracut status in Debian is totally unmaintained, thus can't be really
added in linux-2.6 as alternative.
LSM: Enable AppArmor? as well as/instead of Tomoyo?
---------------------------------------------------
As the LSM need to be built we can't enable them. This needs a technical
solution were code can be disregarded as init sections or similar.
AppArmor seems more popular as Opensuse and Ubuntu uses it. Technicaly
Tomoyo is said to be cleaner.
Ben agreed to work on this.
NX bit emulation and 32-bit mmap randomization
----------------------------------------------
We don't want to carry intrusive patches.
The NX patch was rejected as such by upstream and thus we won't take
it either.
Currently we recommend PAE for bigger boxes but do not default to it.
Action item by bwh and waldi to default Debian Installer to it
and deprecate non PAE 686.
Upstream status of the other patch is unknown, maks will consult Kees.
#534964 memory cgroup support for linux containers?
---------------------------------------------------
bwh has patch to have cgroup mem functionaly but disabled by default.
The cost of it has not yet been evaluated. We need a server box
with good cooling to run benchmarks with it.
Action Item take by bwh as their seem to be a strange profileration
of recipes to recompile Debian linux-2.6 for Linux containers due to
this missing feature.
split linux-2.6/linux-kbuild-2.6 into easy/not-easy cross-buildable parts
-------------------------------------------------------------------------
The problem with cross builder is that it is easy to get a cross builder
that builds linux-2.6 kernel stuff right but not easy for userspace stuff.
Thus tools should move out along with kbuild.
Everyone agreed and waldi suggested to do it.
It was further noted by DPL that we will soon send out a press release
on the firmware status that we achieved for 2.6.32 Squeeeze.
thanks + happy hacking
--
maks
Reply to: