Debian Kernel Group Meeting =========================== Report by Vincent Sanders for Debian Project Kernel List The meeting was held as a series of 60 to 90 minute sessions over a four day period. The location was: Linux Plumbers Conference Mariott Waterfront, Portland, Oregon My primary role in the sessions was to ensure all points on the agenda were covered adequately, to attempt to keep focus within the discussion and to ensure the discussions remained civil. I attempted to ensure everyone had opportunity to contribute and ensure no-one was excluded because of a lack of communication skill in a group setting. These are the notes from the formal sessions, several informal conversations happened which I have not made notes of, nor recorded here. ================================================================= Session 1 ========= The first session was organized to convene in the conference area reception area at 17:00 local time. Most people who might possibly wish to contribute were in the area when the time and location were set. It has been suggested this information was not made sufficiently clear which cause some confusion and contribted to the late start. A initial group of nine members were in attendance at 17:00 and a suitable location in the front of one of the conference rooms was found. However the absence of Bastian Blank was noted and several attempts made to contact him. Luk Claes does not beleive Bastian was aware of the meeting. The meeting was started with the absence of Bastian who was going to arrive later. Start: Wednesday 23rd September 2009 17:20 (PDT) Present: Andres Salomon (dilinger) Luk Claes (luk) Steve Langasek (vorlon) Mark Brown (broonie) Dann Frazier (dannf) Troy Heber (troyh) Max Attems (maks) Ben Hutchings (bwh, Womble2) Vincent Sanders (kyllikki) Not present throughout: Bastian Blank (Waldi) Keith Packard Co-operation and version synchronization with other distributions ----------------------------------------------------------------- The Ubuntu developers made it clear their distribution is interested in syncing as a general aim. This was extended to mention it is actually a "corporate mandate" of Canonical. Steve Langasek as the Ubuntu release manager gave a generally positive indication as it would allow us all to share the burden of updates. A more general discussion of which kernels we might choose followed. Linux 2.6.31 has numerous issues and does not look good as a candidate for long term support or use as a release kernel. There was general agreement on this, the main issues being the number of regressions it shipped with and its generally poor state. If the kernel release cycle patterns from previous years remains constant, the general assent is that there would be a 2.6.32 release in late December or early January. Steve Langasek voiced concerns that the Debian release team still have no idea of freeze date and that Debian might slip much later into the new year. Some discussion occurred about the Debian release team might want to conclude the (currently) unrevealing consultation process and fix a date to allow greater collaboration. Luk seemed to smile but wouldn't make further comment. Ubuntu have made no final decision for 10.04 but highly unlikely to be 2.6.31 There was a discussion of what Greg K H will keep maintaining longer term Finally a consensus that 2.6.32 would be the "long term" supported version due to timings across several distributions. This decision was subject to change depending on freeze timeline/quality. Steve Langasek opined that we should try and say yes to that version for release. Vince was to go and talk with Greg K H about implications for long term support of 2.6.32 Separate firmware, what is left to do? -------------------------------------- This item seemed much less controversial and discussion was generally positive. The main topic was concerned with continued splitting of non free firmware and ensuring Debian kernels remain useful even with our DSFG challenges. Discussion about upstream firmware releases and general agreement was reached to talk to Dave Woodhouse about this (as he did most of the upstream firmware splits) . Action item for Vince and Max to talk to Dave Woodhouse. Brief discussion about bnx2/bnx2x firmware in initramfs and to fix kernel annotations. General discussion about firmware loading. Kernel Mode Setting transition ------------------------------ Keith Packard was Present for this part of teh meeting. There were concerns about initramfs and strategy to get X started in the right way, KMS for Intel drivers is a flag day with 2.9 release Lots of strategy was discussed but we ran out of time in this session. Session concludes 18:20 ==================================================================== Session 2 ========= Second session arranged for conference reception area at 17:30. Steve Langasek informs me he will be slightly late as he has something to attend to. Start: Thursday 24th September 2009 17:45 (PDT) Present: Andres Salomon (dilinger) Luk Claes (luk) Mark Brown (broonie) Dann Frazier (dannf) Max Attems (maks) Ben Hutchings (womble) Bastian Blank (Waldi) Vincent Sanders (kyllikki) Not present throughout: Steve Langasek (vorlon) Kees Cook KMS transition -------------- General technical disagreement of how it fits together. We resolved that KMS would be enabled at build time but disabled at run time by default, with the X packages using modprobe config files to enable it at run time. Action item that Steve Langasek will take the issue to X guys Feature patches --------------- Some of this discussion was easy but there were some somewhat impassioned arguments about specific patches. openvz ++++++ yes we can support, little comment Ben Hutchings said that he talked to some OpenVZ upstream developers about the likelihood of an up-to-date OpenVZ patch for a likely common Debian/Ubuntu LTS/RHEL kernel version, and they seemed keen to work on that. rt patchset +++++++++++ Ben Hutchings recalls that there was a comment that upstream (Steven Rostedt?) did not consider this to be in a production-ready state. no vserver +++++++ Dann Frazier willing to do the work with Micah Anderson Micah has agreed to handle the vserver<->kernel communication and testing, while Dann helps w/ the nuts & bolts of getting the patches applied. Bastian Blank was opposed to carrying the patch. It was agreed however users should be given sufficient warning of its deprication. Action item for Dann Frazier to investigate migration to Linux containers. Dann Frazier and Micah Anderson willing to integrate patches. Max will put Deprecation notice in squeeze release but looks like this patch will make squeeze if Dann Frazier gets the work merged. xen dom 0 +++++++++ Upstream development is splintered but Bastian Blank is willing to implement the patch for squeeze. Final decision on inclusion subject to building. We also agreed to have a deprecation notice for Xen dom 0 as well as vserver into the Release Notes. Action item for Luk Claes to add the deprecation notices to the Release Notes. IDE to libata decision ---------------------- Steve Langasek will look at what Ubuntu migration did, and send a mail, as there was general agreement that Debian can copy that. UDEV package did the upgrade Ben hutchings notes the "migration" and "upgrade" are about fixing device names in configuration files (/etc/fstab, /boot/grub/menu.lst etc). Consensus that non udev users get to fix it themselves. Steve Langasek to talk to Marco further. preemption ---------- A (very brief) discussion occurred but there appeared to be little or no disagreement and a decision to enable this feature was reached. OSS --- The LPC audio track has raised numerous issues with OSS and that it would be helpful if distributes can just get rid of its use. Turns out Fedora and Ubuntu already did. Also turns out Bastian Blank already disabled it for Debian too! Mark Brown has an action item to find and kill OSS users (thats what was in my notes. I *think* we just meant the software which uses the OSS interface ;-) bug triage and tagging ---------------------- current bug count is 600 to 700 Usertags were discussed but possibly unhelpful in the kernel packages model. We discussed the appalling quality of many bug reports and Ben Hutchings took an action item to improve reportbug as a technical thing we can do to improve matters. It was noted there were too many bugs without logs. A decision that the whole team needs to verify policy for bugs and patches, publish and *enforce* it. This is an action on the whole team. Ben Hutchings will post call for comments to mailing list with deadline. Session concludes 18:40 ==================================================================== Session 3 ========= Third session arranged for conference reception area at 17:30. Turns out this wasn't possible as the final keynote didn't get finished till 17:50, we started at the back of the keynote room but were forced to relocate due to noise halfway through. Start: Friday 25th September 2009 18:0 (PDT) Present: Andres Salomon (dilinger) Luk Claes (luk) Mark Brown (broonie) Dann Frazier (dannf) Max Attems (maks) Ben Hutchings (womble) Bastian Blank (Waldi) Vincent Sanders (kyllikki) Steve Langasek (vorlon) Andy Whitcroft (apw) Amit Kucheria (amitk) initramfs strategy ------------------ A general discussion on what (if any) changes to the initramfs strategy was started and concluded that we would strip down to block drivers but allow for framebuffer on non text console. Max and Steve Langasek were to have an out of band conversation about this and try and benefit from whats been done in ubuntu. Git --- The discussion started with most of the group agreeing git would be a better RCS for use with the kernel as it would integrate with upstream better. Ben Hutchings presented a fairly positive list of benefits and expressed his favour for a switch. Bastian Blank responded with a negative view because what we have now works fine and there is no reason to switch. He presented several issues and wants to wait for a working stratagy before using git. Steve Langasek made some points about the need to do a complete fresh checkout with the current SVN approach and Bastian would not comment further and took a somewhat closed posture. There was some discussion about our current model of development and tools and how they might be adapted for use with git. Several solutions were suggested and a reasonable technical discussion followed. Bastian when queried for his input responded with a statement that he doesn't know how git works and "i don't know" in relation to what impact it might have if we converted to git. The rest of the group knocked together a rough consensus where what we have today is directly transitioned to git and as a second stage we convert topic branches. The group decision is that we want to switch. An action item for Ben Hutchings, Max and Andres Salomon to investigate and implement; this group will keep the mailing list up to date with progress. Bastian Blank asked that his opposition to this process and decision to be noted. The canonical guys present who were representing the Ubuntu kernel team offered assistance and were very positive. coordination with release team and d-i -------------------------------------- A general decision that the udeb generation should probably be pushed back into the kernel source package (it is the only source package which doesn't generate its udebs directly and uses a second stage) Luk seemed happy with this proposal as it reduces the issues surrounding kernel-wedge Steve Langasek raised the point that, since one of the original arguments for the separate udeb packages was the need for the d-i team to asynchronously rev the udebs during d-i development without being blocked on the much slower kernel builds, we should establish empirically whether the d-i team has been making use of this feature - to forestall having a tempest-in-a-teapot flamewar. Testing migration is a burden/issue and the kernel team should be responsible Kernel ABI changes should be treated as a library transition for each version. Action for Luk Claes to try to reach consensus with the d-i Team about this. It was also noted that it is a challenge for ftpmasters to map binaries to source. Conclusion was to move onto next item which is really the main issue. out of tree modules ------------------- After a somewhat involved discussion taking into account the FTP masters extreme irritation about trying to match binaries to source by hand for the lenny release it was resolved to remove linux-modules-extra and -nonfree as they are an impossible to support approach. A few modules the project really want/must have will be placed directly into the linux-2.6 source The kernel team would endorse the use of dkms as a way for out-of-tree module maintainers to get their modules auto-built. Ben Hutchings to talk to Greg K H about extra modules being merged into staging tree. Leveraging upstream .deb building --------------------------------- This became a discussion about the general kernel packaging and how we might use the upstream provided facilities better. There was some discussion we have way too many ways to build a kernel. A general agreement was reached that "make-kpkg" should go away as it is no longer adequately maintained. Action item for Vince to tell Manoj it is deprecated and thank him for his contribution. Max agreed he would make the upstream in tree "make deb-pkg" packaging work and keep a general eye on it. Ben Hutchings and Max agreed they would make postinst better and rationalize some of this huge script. Ubuntu team noted that they are looking at doing the same, and general consensus was to combine efforts. New lists to co-ordinate ------------------------ There is a mailing list which we might resurrect for general distribution co-ordination email@example.com And a list the Debian kernel team might want to join to co-ordinate with Ubuntu firstname.lastname@example.org Session concludes at 19:00 ==================================================================== Session 4 ========= Fourth session arranged for Mariott hotel lobby at 09:00 We finally get sorted sat on the waterfront in the sunshine. Start: Saturday 26th September 2009 09:15 (PDT) Present: Max Attems Bastian Blank Dann Frazier Luk Claes Ben Hutchings Vincent Sanders Debug Packages -------------- Debugging information from current packages, not a separate configuration, useful for crash tool Aids some debugging scenarios though it was pointed out we cannot get users to submit data for the data we already generate. A *major* drawback is the data debs produced would total some 10 Gigabytes for all architectures and some arches simply could not build them. We talked about subsetting - data is more likely to be used on some archs/flavors (i386/amd64) and not all buildds are capable of this much data. I also noted that HP would like to see this to ease support of debugging customer issues on debian. Dann Frazier will talk to ftp masters and the buildd people to see if it is even feasible and report back to mailing list. Automated test -------------- Vince explained the kautobuild idea and infrastructure. General assent that it will be useful. Vince needs to work with Mark Hymers and Steve Gran to get a viable system actually up and running. Experimental ------------ Max agreed to do some upload 2.6.31 to experimental. Backports --------- Low priority idea and no action item considered Session concludes 09:50 ==================================================================== Action Items ============ 1) Vince was to go and talk with Greg K H about implications for long term support of 2.6.32 [[ I performed my action item, although Greg could not commit SUSE to any specific release it does seem that Fedora, Debian, Ubuntu and several other distributions have all said pretty much the same thing now and that 2.6.32 seems like "a nice one to settle on for longer term stable support" ]] 2) Vince and Max to talk to Dave Woodhouse about firmware. 3) Action item that Steve Langasek will take the KMS issue to X guys 4) Dann Frazier to follow trough on vserver patches. 5) Bastian Blank to follow through on xen dom 0 6) Luk Claes to add the deprecation notices to the Release Notes. 7) Steve Langasek to talk to Marco further. 8) Mark Brown has an action item to find and kill OSS users. 9) A decision that the whole team needs to verify policy for bugs and patches, publish and *enforce* it. This is an action on the whole team. Ben Hutchings will post call for comments to mailing list with deadline. 10) Max and Steve Langasek were to have an out of band conversation about initramfs stratagy and try and benefit from whats been done in ubuntu. 11) Ben Hutchings, Max and Andres to investigate and implement this group will keep the mailing list up to date with progress. 12) Luk Claes to try to reach consensus with the d-i Team about testing migration. 13) Ben Hutchings to talk to Greg K H about extra modules in staging. 14) Vince to tell Manoj its deprecated and thank him for his contribution. 15) Max agreed he would make the upstream in tree "make deb" packaging work and keep a general eye on it. 16) Ben Hutchings and Max agreed they would make postinst better and rationalize some of this huge script. 17) Dann Frazier will talk to ftp masters and the buildd people to see if debug packaging is feasible and report back to mailing list. 18) Vince needs to work with Mark Hymers and Steve Gran to get a viable autobuild system actually up and running. 19) Max agreed to do a 2.6.31 upload to experimental.
Description: Digital signature