maks blog

maks blog, Hacks, fun around free software.

maximilian attems
email maks

Subscribe to a syndicated feed of my weblog, brought to you by the wonders of RSS.

There's more than one way to view this weblog; try these flavours on for size.

Other blogs in our institute or friends.

Wed, 02 Mar 2011

Cross distribution collaboration on longterm 2.6.32 linux-2.6

I'll collect here links to the various linux-2.6 distribution trees based on the longterm release 2.6.32. I won't go into details of bigger external patches (grsec, openvz, rt, xen, ..), which in consequence also aligned themselves on 2.6.32.

Important bug fixes usually are connected to some bug report, which may be publicly viewable. So from the patch and changelog entry one can usually assume if a certain patch satisfies the stable criteria and forward it for 2.6.32 longterm inclusion (+ other branches where it might apply). If the patch applies and compiles fine with the Debian tree one can assume that the patch will be fine for upstream 2.6.32 as Debian with small exceptions mostly follows the longterm release: 2.6.32 branch of Debian linux-2.6 (Comment: This is currently only a svn mirror but this bug is worked on for next release).

The canonical 2.6.32 linux-2.6 longterm repository is of course on maintained by gregkh. Opensuse publishes it's kernel-source on gitorious including all branches and the especially interesting SLE11-SP1 2.6.32 branch. Fedora was following till late Autumn 2.6.32 and the F12 branch has the relevant patches. Ubuntu released 10.04 with 2.6.32 as in a collaborative decision they also based their drm on 2.6.33 (same story as in Debian, thus particularly relevant for us). Oracle had a 2.6.32 that was maintained until Sept 2010 or such.

It is already bad that kernel source in Red Hat doesn't really follow upstream 2.6.32 longterm release itself. For 2.6.18 of course no such option existed, but for 2.6.32 this policy already shows a certain snobbery. Red Hat 6.0 Beta at least shipped kernel-2.6.32-37.el6.src.rpm with broken out patches - since then no patch series or git tree to be seen from RH. This strange move got since picked by lwn - RH "obfuscated" kernel source.

[/kernel] permanent link

Tue, 02 Nov 2010

"We'll Always Have Paris" linux-2.6

At the mini-DebConf Paris 2010 the Debian linux-2.6 team released latest 2.6.32-27 adding stable plus security fixes and drm/intel fixes from the Ubuntu shared drm 2.6.33 tree.

Beside the productive Debian kernel team meeting, whose minutes will be forthcoming, the mini Debconf had a great welcoming setup and friendly chats. We enjoyed an impulsive Saturday evening at Chatelet in middle of the funny Parisian Halloween mess.
Update: Photo of the last busy meeting session during lunch break that allowed us to finalise all Paris meeting topics.

[/kernel] permanent link

Tue, 06 Apr 2010

"Little Bang" 0.94 initramfs-tools release

Heavy Ion collisions try to recreate conditions very shortly after the Big Bang. Thus the created quark gluon plasma is often the Little Bang, due to recreating this very hot conditions.
The kernel Team already uploaded the linux-2.6 Big Bang release now follows initramfs-tools with the Little Bang. ;-)

Newer initramfs comes with lots of fixes and new features:

  • Faster boot thanks to Ubuntu merged patches using pre-cached boot scripts. It be cool if someone could time that and let us know?
  • Support for several compression schemes as set in initramfs.conf or overridden by appropriate mkinitramfs call.
  • Better MODULES=dep support /sys parsing (Still needs fixes for md).
  • Hooks for out of linux-2.6 make deb-pkg generated linux-images.
  • Call ipconfig with increasing timeouts to better fit with large scale clusters.
  • Use BOOTIF mac address passed from gpxe in setup_networking.
  • Out of the box support for devtmpfs. Hurra for devfs 2.0.
  • Loading of the netconsole module together with the specified args, when netconsole is set as bootarg.
  • Adds available, built network drivers in a dynamic way without need of hardcoded list.
  • Smaller and thus faster MODULES=most generated initramfs due to kicking the video drivers out of generic initramfs.
  • Last but not least don't suppress eventual modprobe errors.

Thanks for all the patches and useful input! Sorry for late release, will try to do it earlier more often and there is hope of an Ubuntu sync: view of initramfs-tools repo.
P.S.: cryptsetup needs fixup of #576488, udev can now start at a earlier stage of initramfs.

[/kernel] permanent link

Wed, 03 Mar 2010

2.6.32 sid updated and 2.6.33 in experimental

Experimental 2.6.33 will do the switch to UUID based root args, if you haven't switched already. Please test it out and report bugs on it, before we add the libata switch to squeeze 2.6.32.

2.6.32-9 includes and several other fixes. For the following 2.6.32-10 Ben Hutchings pulled in newer drm for lots of intel fixes: Status of kernel X drivers. It also features radeon and nouveau KMS modules. Current 2.6.32 is stabilizing well and we are seeing lots of external patches lining up.

Update: Sorry due to jet lag got aboves version number wrong. You might want also to checkout: The season I have mostly been building kernels.

[/kernel] permanent link

Tue, 20 Oct 2009

Debian Kernel Meeting

Vincent Sanders took notes during all our meetings at the Portland Linux Plumbers Conference: Debian Kernel Group Meeting. The condensed form has been posted today as Bits from the kernel team.

In the case of feedback I'd highly recommend to bring the Debian Kernel Mailinglist into the loop. The meeting decisions were done by the team as entity.

Responding to the deprecation of some external patches (Vserver, Xen Dom0): None of above patches have an upstream that supported the Lenny released version. Both have troublesome bugs in Lenny and thus are not in a condition one would expect from a stable release. If you want to help and have continued release of those beyond squeeze the answer is easy: Get them merged upstream. Openvz supports Lenny linux-2.6 version actively and promised to keep up with their work for Squeeze.

It has been a very productive meeting with lots of problems^Witems discussed. Interesting tracks for better cooperation between distributions, heavy technical tracks and loud BOFS. Quite some work has already been picked up since (Bug scripts, 2.6.31 experimental uploads, DEB_BUILD_OPTIONS=parallel=N support, package descriptions improvements, piuparts install fix, DFSG firmware clean, preempt, ..). So thanks a lot to Steve McIntyre (Debian Project Leader) for pushing the meeting, to Steve Langasek for setting it up on site and of course to everyone who contributed. Read aboves report for the full picture. :-)

[/kernel] permanent link

Tue, 24 Mar 2009

2.6.29 in sid

Latest Tasmanian Devil Linux 2.6.29 adds quite a lot of new features: Kernelnewbies 2.6.29 Overview.

Thanks a lot to the FTP Master Team for making possible the same day upload. Big thanks also for the regular testers of the rcX trunk snapshots.

[/kernel] permanent link

Thu, 05 Feb 2009

initramfs-tools new Lenny features

initramfs-tools design principle is to ship a generic initramfs. The design explicitly allows lots of customization through hook and boot scripts. A previous post focused on the history of initramfs-tools. Several contributors enhanced initramfs-tools since the Debian Etch release so the non exclusive new feature list:

  • The initramfs contains all drivers/{ata,block,ide,mmc,scsi} modules instead of a stupid hard coded list. This gives bonus points for getting more future proof.
  • Configurable small initramfs: The MODULES=dep mode walks /sys/ for the needed drivers of the specific box. This was the most wanted feature and is the new default for some Debian-Installer embedded architectures. The code seems in between well tested even if some corner cases might still emerge. The smaller initramfs is needed for specific bootloaders and helps on slow target boxes.
    For a even smaller initramfs you can kick busybox if you don't have an lvm2 root or don't need it's debugging tools.
  • Quicker initramfs build: The update-initramfs command uses dpkg trigger and thus not every postinst update-initramfs(8) call leads to a new initramfs build. Also the build time got improved.
  • Improved panic message (yes users need to be handhold at this point :).
  • Potential debug log is visible after call to init(8) inside /dev/.initramfs/.
  • update-initramfs got hardened against failure on update or ro partitons.
  • The klibc utilities got worked on: chroot, fstype, ipconfig, mount, nfsmount sync.
  • Fixed various bugs in NFS and networking setup due to incorrect option parsing.
  • Allow to blacklist modules on boot by cmdline bootarg blacklist=modulename.
  • Allow non root usage of mkinitramfs.

As bonus to some cleanup and docs update we have also seen a Ubuntu sync and are not too diverged.

[/kernel] permanent link

Tue, 22 Jul 2008

OpenVZ Debian Linux images

Thanks to the Virtuzzo developers for forward porting the OpenVZ patch to 2.6.26. It uses a lot of the already merged namespace features (For interested peoples linux-next carries interesting sysfs patches). The openvz upstream merge statistic is impressive.

The namespace solution allows low overhead, while still running multiple Linux distributions in individual containers on the same box. Control over the network vitalisation is gained as a resource management for CPU power, I/O bandwidth and disk space. Live migration including snapshotting is possible too.

The Debian 2.6.26 Linux images will feature not only better Xen vitalisation support, kvm port to new archs, but also an opvenvz flavour. Please don't forget to install proper vzctl userspace support as the correct dependencies have not yet been added. Supported archs will start with both x86 and the ia64 archs. Tomorrows linux-2.6 Debian trunk snapshots will already build openvz. Thanks for your feedback.

P.S.: 2.6.26 is the targeted kernel for Lenny, we hope to convince Debian installer people to pick it up soonest - 2.6.25 is a not so bad backup plan as release hasn't fully approved 2.6.26 before extensive unstable testing.

[/kernel] permanent link

Tue, 29 Jan 2008

2.6.24 in Sid

2.6.24 has been uploaded on Saturday and passed NEW on Monday. It will be soonest available for all archs. If we don't get enough test coverage or things go bad 2.6.22 will be the fallback for Etch+half. The images install in Lenny/testing just fine. So please give it a spin.

Known Issues are still missing VServer and Xen images, bnx2 + snd-cs46xx disabled due to firmware issue and ACPI userland still relying on old interfaces (for example #462305).

For more adventurous souls the kernel buildserver will continue to track latest Linus git daily snapshot (enhanced ext4 should land soon).

2.6.24 will also soon be pushed to Testing. I repeat this is not the real Etch+half kernel yet as there will be some config changes for it, but it is pretty close. Thanks for your 2.6.24-rc7 feedback.

[/kernel] permanent link

Fri, 11 Jan 2008

Testing Etch + 1/2 Kernel

2.6.24-rc7 is available on 5 architectures (amd64, i386, powerpc, s390, sparc). Please add the following line to your Lenny or Sid box sources.list and report back:

deb trunk main

This is not yet the real Etch + 1/2 kernel, but comes pretty close. According to our initial testing and user feedback 2.6.24 promises to be a great release. It will help Etch to have proper hardware support for newer hardware, has better power efficiency due to tickless kernel, enhanced wireless card support, the CFS scheduler and much more...

It is a Debian novelty to provide an supported optional Linux kernel upgrade for a stable release. This is not the final as due to incompatible user interface the Etch build will have the old firewire stack. Once aboves land in Sid and got broader coverage will announce the Etch snapshots. So please give it a shot on your Testing/Unstable boxes.

[/kernel] permanent link

Thu, 23 Aug 2007

linux-2.6 bugs cleanup part 2

Random notes on effective bugs cleanup on a bigger scale:

  • Ignore the subject. In 90% of the cases it is badly formulated and provides no clue.
  • Be on a good mood or don't even start.
  • Use different overviews (bts, mbox, web, reportbug-ng, ..) to not miss the easy stuff right on start.
  • Read the last message first. A please close message on the tail or an upstream nack keeps you from reading the previous ones. Obviously as curious person i often violate that rule.
  • Have prepared standard answers on FAQ: Debian kernel patch acceptance guidelines, report issue to Linux upstream, debug initramfs.
  • Be bold: A big wontfix list won't help anyone. Close right away.
  • In the ping mail directly write To the submitter. A Cc to the bug report makes it more personal.
  • Tag any bug that is against an old version with moreinfo in the ping mail. Gives the submitter a clear indication that this bug needs an update.
  • Sometimes a close acts better as ping, reopen quickly and use the info.
  • [/kernel] permanent link

    Thu, 02 Aug 2007

    linux-2.6 bugs cleanup part I

    xorg and started with a cleanup of their bug reports. It is really great that compared to 2-3 years ago major cores of the Debian OS in unstable are quite in sync with upstream. Thus it is easy to forward to upstream the interesting part of bugs.

    The linux-2.6 bugs count reached that weekend almost 850 open bugs. Seeing that it gets unmanageable I'll decided to get it down to a target of 100 bugs. I'll blog from time to time about the progress.

    The process consists of pinging quite some bug reports against old Linux images and closing a huge swap of duplicate or no longer relevant bug reports. I expect the signal to noise ratio getting better once the bug count gets denser. The current side effect was to forward 1 patch upstream and to get 2 easy low hanging fruits fixed in latest linux-2.6 trunk.

    Thanks to enabling CONFIG_BLK_DEV_IO_TRACE trunk gains support for the blktrace(8). blktrace does direct IO block layer tracing - see also blktrace guide.

    [/kernel] permanent link

    Tue, 07 Nov 2006

    linux-2.6 2.6.18 status
    2.6.18-4 was uploaded on Sunday and should be soonest available after dinstall run for most archs. The ia64 build failure is fixed in svn thanks to Thiemo Seufer (ths). alpha is waiting for an updated gcc (see patch in #397139). s390 is currently broken by vserver, patch is awaited soon. linux-latest will be updated tomorrow and expect soon a 2.6.18-5, once those issues are cleared.
    linux-image-2.6.18-2- is the Debian kernel team stabilized kernel. Please install it and report eventual bugs. If you know a patch from Linus git tree that fixes your problem even better name it. Thanks for your testing!
    We had been quite busy to feed stable with patches that fit the Documentation/stable_kernel_rules.txt. Goodies like Xen, drm-i965 or ahci backport are shared with Fedora Core 6 release. Beyond that for example the r8169 patch series or ccisss support for 2 TB volumes got backported. Out of reach are destabilizing patches like the new 2.6.19 ACPI patches. Further backports are planed for the SUSE reiserfs 2.6.19 patches and vorlon has a fix in the pipe for the vga console driver on alpha.

    [/kernel] permanent link

    Thu, 02 Nov 2006

    Early Userspace Fun
    For the sarge release official kernel-images from the newly formed Debian kernel team came together with initrd-tools. initrd-tools Maintainer was jbailey. It saw some care from tbm, vorlon and me before the actual release. Remembering the flow of bugs and installation-reports it accounted for a huge number of install failures plus lacked many features the Red Hat mkinitrd had. Ubuntu got hit more directly as hoary released with a newer kernel than 2.6.8 and due to the initrd-tools devfs requirement.
    initramfs-tools originated out of the need of a direct replacement. Inspired by an initramfs OLS talk jbailey had the goals of boot-time hardware detection using the nice and neat features the Linux 2.5 branch incorporated: The sysfs support of the Linux drivers allows an neat userspace daemon aka udev to coldplug them. udev is the replacement of a maze of slow shell scripts called hotplug package. The second big feature is the new initramfs format (the curious may want to read Documentation/filesystems/ramfs-rootfs-initramfs.txt of any current linux-2.6 source). For a boot loader initramfs or initrd is the same fish. The smaller and more efficient in kernel code makes the difference. The initramfs gets loaded much earlier, due to being only an cpio archive and not an fs allowing it to completely rule the rootfs search and mount. One initial assigned spec had the name "easy NFS root". The LTSP guys joined in and showed interest in incorporating their work.
    initramfs-tools got default in Debian together when the first common linux-2.6 (2.6.15) package reached etch, powered all the Ubuntu releases from Breezy on and is used by the grml live-cd. Since archive inclusion almost a year ago it got much better support of all the various linux boot parameters (see man initramfs-tools or the "official" Documentation/kernel-parameters.txt), keeps an backup initrd around and has better block-based bootloader support (lilo, ..). klibc has been ported to _all_ the Debian release archs where it is in use and compiles fine on m68k and sh.
    The initramfs-tools early userspace is very flexible based on it's hooks for initramfs build and boot. Thanks to alphix for the help on shaping the building blocks. There are initramfs hooks in cryptsetup, evms, dmraid, firmware-qlogic, lvm2 multipath, mdadm, usplash, .. Reviewers, uploaders and sponsors included fs, jbailey and Sesse. Most of the time it is sufficient and easy to test changes in qemu images (See slides workshop early userspace). The best wish of an early userspace is not to be remarked - aka happy booting. Nevertheless thanks a lot for the public support: Achieving Xen Disk encryption support in Etch, Popping my initramfs cherry, ..
    The latest initramfs-tools release seems almost ready for the upcoming etch release. The bug reports coming over from installation-reports are happily rare. Although the uuid based fstab generation solution is not yet integrated in the debian-installer. There are one or two known nitpicks left-over in the bts plus one bugger: udevsettle may exit to early, when not all the discs are ready for use. For grub there is a fail-over net, but which does not account for a to early md or lvm2 boot hook. Future initramfs-tools development will focus on an optional small klibc based Modules=dep hand walking /sys initramfs generation allowing better embedded support. grub2 will present run-time assembly challenges for the initramfs.

    [/kernel] permanent link

    Thu, 25 May 2006

    Kexec allows quickboot
    Kexec is handy for a quick and "real" test of the latest initramfs or the latest installed kernel. Install kexec-tools and be quite picky about how you pass the cmdline args. Be careful to launch the last execute from console. Sync and unmounting discs before is recommended. ipw2200 panics since ever on kexec, simple unload brings it back to life. When using in combination with initrd/iniramfs uname > 2.6.15 is recommended.

    kexec -l -t bzImage --command-line="ro root=/dev/sda1" \
    --initrd="/boot/initrd.img-2.6.16-1-686" /boot/vmlinuz-2.6.16-1-686
    kexec -e

    [/kernel] permanent link

    Sun, 26 Feb 2006

    gdth unmainted scsi driver
    _One_ disc had grown defects 2024. This made the controller hang and the fs are in poor state, what for is that ICP vortex controller?

    [/kernel] permanent link