LKML Summary Podcast

Audio: http://media.libsyn.com/media/jcm/linux_kernel_podcast_20100207.mp3

This podcast is brought to you by the awesome power of Jason Wessel’s kgdb patches, helping to support those who believe in kernel debuggers find hard to reach kernel bugs since 2009. Kernel debuggers: the way of the future.

For the weekend of February 7th, 2010, I’m Jon Masters with a summary of the week’s LKML traffic.

In today’s issue: Linux 2.6.33-rc7, regressions, Google Summer of Code, IMA, OOM, and sys_membarrier.

Linux 2.6.33-rc7. Linus Torvalds announced the 2.6.33-rc7 release of the Linux kernel on Saturday, February 6th, 2010 at 2:44pm (14:44) Best Coast Time (PST). In his announcement, Linus remarked, “I have to admit that I wish we had way fewer regressions listed by this time, so I hereby would like to point every developer to” a link to a recent post to the linux wireless mailing list archive on gmane.org showing a copy of a recent email from Rafael J. Wysocki detailing known kernel regressions between 2.6.32 and 2.6.33-rc6 as posted originally to the LKML. He added, “But we’ve certainly fixed a few things, and it’s been a week, so here’s -rc7″. Most of the changes are in PowerPC defconfigs (default configs), but there are even more i915 updates, radeon KMS updates, and lots of other smaller bits all over the tree. Linus also wondered (in another email) whether it was worth making the .gz files any more given that bzip2 has been around more than long enough by now. Some thought the gzip files were still useful on systems without bzip2 or for some really slow systems that apparently handle gzip files more easily.

Regressions. Rafael J. Wysocki followed up to Linus’ 2.6.33-rc7 announcement (as he had also done with 2.6.33-rc6) with a list of outstanding regressions beteen 2.6.32 and 2.6.33-rc7. There are currently 20 “unresolved” issues in the list of regressions given. Rafael also noted that Maciej Rutecki has, “generously volunteered to work on the tracking of kernel regressions”. The work done by Rafael (and now, hopefully Maciej also) is very valuable to the community and we really do owe them our gratitude for helping out. Arjan van de Ven also posted a list of oops and warning reports on kerneloops.org from the week, including a very common ext4/quota issue in Fedora.

Google Summer of Code. Luis Rodriguez stated that, “Google has confirmed it will have a Google Summer of Code for 2010″, then mentioned that last year’s effort (4 suggested projects, of which 3 were accepted) resulted in only one success. Witold Sowa followed up saying that he didn’t know he was the only student who completed his project, but that the work to add an AP mode to NetworkManager, “with use of wpa_supplicant’s newly developed AP mode” was relatively easy to accomplish and so he had worked on other things also. Apparently, the initial GSoC work is now available in NetworkManager. Nonetheless, it sounds as if Luis is keen to see a higher than 33% success rate if any entries are accepted this year under the Linux Foundation.

IMA. Mimi Zohar replied to an email from Shi Weihua concerning a NULL pointer deference bug in the IMA security code (ima_file_free), which Al Viro and others had previously discussed solutions for.

OOM. Lubos Lunak and David Rientjes resurrected the OOM killer discussion again after Lubos posted some analysis of various KDE processes running on his system, and wondered why the OOM killer uses VmSize rather than RSS to determine tasks that should be killed (in other words, why should it not favor tasks actually resident in memory at the time?). This discussion has been had recently, and David Rientjes explained that the kernel favors overall VmSize in its calculations so as to catch memory leakers as a preference (which are often not resident at the time). David did seem to like the suggestion of catching the the child with the highest badness calculation before killing its parent, and posted an untest patch. He also suggested that the KDE process tree example was “a textbook case for using /proc/pid/oom_adj to ensure a critical task, such as kdeinit is to you, is protected from getting selected for oom kill”. Lubos replied with some very good points about how simply setting oom_adj doesn’t scale, and Balbir Singh was amongst those still favoring a switch to RSS-like accounting but with support for shared pages (for example “PSS”) eventually. Rik van Riel noted that he had no strong opinion one way or the other. David posted various patches proposing an alternative fine grained oom_adj mechanism.

sys_membarrier. Mathieu Desnoyers posted a three part patch series implementing sys_membarrier, a new system call that can be used to “distribute the overhead of memory barriers asymmetrically”. In particular, he wants it for his urcu userspace RCU implementation (for use within the synchronize_rcu call). Sensibly, Mathieu proposes incremental additions to each architecture (even though he believes that it “should be portable to other architectures as-is”), reserving the system call numbers now, then implementing gradually.

In today’s miscellaneous items: Matti Aarnio posted to let everyone know that a recently discovered hole in the bayesian filtering system as used by the vger.kernel.org mailing list server to reduce SPAM has been plugged (it had been possible to reach the list using a specific “backend” majordomo domain), Catalin Marinas decided to simply patch the USB HCD driver that had resulted in cache coherency problems when using USB storage (and noted that a followup posting to linux-arch would call for a flush_dcache_range function), some miscallenous rewrites of obsolete syscall handlers to use generic versions from Christoph Hellwig, a request for an opinion on mergeing the kFIFO rewrite in 2.6.34 from Stefani Seibold, a potential issue with the kernel implementation of LZO compression reported by Nigel Cunningham (for which he will switch back to LZF in TuxOnIce again for the moment), Stephen Rothwell wondered aloud whether Linus would really be interested in taking the percpu changes currently sigging in percpu “next”, and Mathieu Desnoyers announced he is switching email from his academic address in Montreal (where he recently completed his PhD around LTTng) to a consulting firm he is involved with at http://efficios.com.

In today’s announcements: Greg Kroah-Hartman posted review patches for the 2.6.32.8 stable series kernel.

Scott James Remnant announced the release of upstart version 0.6.5. It includes a large number of fixes, amongst which is the completion of the splitting out of libnih into its own project. There is a new /sbin/reload command for reloading upstart daemons, a restored sync() before reboot, improved documentation, and more goodies.

Junio C Hamano announced version 1.7.0.rc2 of the Git SCM, which includes a number of forthcoming behavior changes as mentioned in this podcast when discussing the rc1 release from the previous week.

Subrata Modak announced that the Linux Test Project (LTP) for January 2010 has been released. It now contains over 3000 tests. Separately, Garrett Cooper noted a rather severe bug in the top level LTP Makefile that could result in an “rm -rf /” in the wrong circumstances, suggesting that all LTP users comment out three lines from that file.

Willy Tarreau (re-)announced the release of 2.4.37.9. The previos 2.4.37.8 hadn’t actually contained the required e1000 backport with a CVE fix that had triggered the previous release. Willy noted, “I don’t know how I managed to do that because it once was OK and I could successfully build it. Well, whatever I did, the result is wrong and the issue it was supposed to fix is still present in 2.4.37.8. So here comes 2.4.37.9 with the real fix this time”.

The latest kernel release is 2.6.33-rc7.

Andrew Morton posted an mm-of-the-moment (mmotm) for 2010-02-03-20-09.

That’s a summary of today’s Linux Kernel Mailing List traffic, for further information visit www.kernel.org. I’m Jon Masters.