LKML Summary Podcast

Audio: http://media.libsyn.com/media/jcm/linux_kernel_podcast_20091203.mp3

For Thursday, December 3rd, 2009, I’m Jon Masters with a summary of today’s LKML traffic.

In today’s issue: The 2.6.33 merge window, oops, and perf.

Merge window. The 2.6.32 release on Wednesday signaled the openeing of the 2.6.33 merge window. There have been a number of git tree postings so far, including security updates from James Morris, and GFS2 updates (previously announced though in pre-post warnings) from Steven Whitehouse. The trivial tree has what looks like a larger number of fixes than it does, but there are certainly a good number of cleanups present in Jiri Kosina’s tree. There were also the usual crazy number of git tree updates from Ingo Molnar (which at this point take up an entire screen in mutt) – including RCU-tiny and default RCU stall-detector enabled in Kconfig, and also a new feature in user-return-facility for fast returns to kernel mode only in KVM. Grant Likely also pushed a round of OpenFirmware device tree patches that everyone should get really excited about, because that’s cool stuff.

Oops. Donming Jin posted a question concerning kmsg_dump, a function called to ensure that an oops message is written synchronously before continuing with, for example, a panic inspired crash dump. He wonders if it should be called in oops_end rather than in oops_exit, to ensure ordering.

Perf. Part of the first round of pull requests for 2.6.33 (and amongst his many other git tree updates) were some perf patches from Ingo Molnar. These include support for a new “perf kmem” tool for SLAB allocator analysis, a new “perf probe” tool to utilize kprobes to dynamically probe kernel functions, “perf bench” to run micro-benchmarks in a “uniform way”, and new hardware breakpoints asbstraction and support in the perf tools. Perf really is becomming some kind of “swiss army knife” in the kernel at this point.

Finally today, someone finally answered Michael Gilbert concerning the state of fixes for a couple of CVEs first reported in 2004. It would seem that the cryptoloop code has some “major design/implementation weaknessess” vs. the two other forms of disk encryption already shipping in-kernel.

In today’s announcements: GIT version 1.6.5.4. Junio C Hamano announced the latest release of the GIT SCM as used by the kernel community. It includes a number of fixes, mostly fairly trivial (for example that “git help” used to require the current working directory to be under GIT control.

Lsscsi. Douglas Gilbert announced the latest version of lsscsi (0.23). The latest release of this utility will assume /sys as the mountpoint for sysfs rather than poking around in /proc/mounts by default. The utility is mostly used by those wishing to understand SCSI topology in their systems.

The Linux Test Project. Subrata Modak announced that the Linux Test Project for November 2009 has been released. The latest version features a number of highlights, including cleanup and stabilization of the new build system for distributions with older kernel releases and the addition of “EXECLTP”, a “more user friendly” front end script written in python.

The latest kernel release is 2.6.32.

Stephen Rothwell posted a linux-next tree for December 3rd. Since Wednesday, the new kbuild-current tree (with a new Kbuild maintainer) has been added, the powerpc tree lost its build failure, the ext3 tree gained a build failure for which Stephen reverted a commit, the net tree lost a conflict, the mtd tree gained a conflict against the mips tree, the tip tree lost a conflict, and the workqueues tree lost its conflicts. The total sub-tree count remains stead at 154.

Stephen notes his “usual call for calm” in saying that developers should not put stuff in their git trees intended to linux-next that targets 2.6.34 inclusion until after the first 2.6.33 -rc has gone out.

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

Audio: http://media.libsyn.com/media/jcm/linux_kernel_podcast_20091202.mp3

For Wednesday, December 2nd, 2009, I’m Jon Masters with a summary of today’s LKML traffic.

In today’s issue: Alignment, CFS, faults, utrace, and a new kernel release.

Alignment. Tejun Heo and Ingo Molnar had a debate about memory allocations (in the percpu allocator) and the issue of alignment. On UP (non-SMP) systems, the percpu allocator becomes a simple wrapper around kmalloc, but there’s no way to request a larger (than the machine’s minimum natural word) alignment with kmalloc, so users (such as workqueues) that want to rely upon a specific alignment in order to re-use low-order bits of struct pointers for flag purposes may need their own helper, as Tejun suggested in alloc_cwq.

CFS. Nick Piggin posted to let everyone know that he has been looking through some recent scheduler (CFS) patches and notices a number without any good justification for their inclusion, as well as a few that merely tweak tunables, to which he pondered “[a]re people really not reporting enough regressions against CFS that it is time to just tweak things?”.

Faults. Ingo Molnar and Sriphathi Kodi had a dialog about Sripathi’s recent work on futex fault injection. As this week’s Linux Weekly News points out, Ingo essentially rejected this patch because the patch is “butt-ugly” in the sense that it builds upon existing infrastructure Ingo isn’t fond of. There has been some debate over appropriate response when telling someone submitting patches that they have to also go rework and entire subsystem. It isn’t necessarily wrong to do that, but it can be very frustrating for those who don’t have a lot of time to hack on the kernel in the first place.

Utrace. Roland McGrath (who notes that he is traveling at the moment and is therefore going to be a little laggy on replies) provided some context for the merging of utrace and shared some of his frustrations in dealing with many arch maintainers giving conflicting advice, or just being uninterested and not really caring at all. In saying, “we want to do whatever the people want” (Roland and Oleg) are essentially saying they’re done with the open ended debate and perceived lack of action in moving things forward.

Finally today, Maxim Levitsky asked some questions about waiting for kernel events, to which Alan Cox replied with some useful advice, and Maciej W. Rozycki and Ingo Molnar had a debate about the correct possive form of the word “MIPS”, which this author reminds everyone is an acronym.

In today’s announcements: Linux 2.6.32. Linus Torvalds announced the release of the latest (2.6.32) version of the Linux kernel at 9:47pm Best Coast Time (PST). In noting that the release felt “long overdue” due to two empty weeks for Linus (at Kernel Summit and then a Thanksgiving week vacation) he added that the release was in fact less than three months following the 2.6.31 release and so wasn’t actually far from the normal schedule. There were relatively few changes since -rc8, with the notable exception of some last minute fixes to fscache. Linus wasn’t too pleased with the timing of those, and noted again that this is a release without a new filesystem “for once”. Of course, if LogFS gets merged next time, that will changed quite quickly.

GIT 1.6.6.rc1. Junio C Hamano posted the latest version (1.6.6.rc1) of the GIT SCM as used by the Linux kernel community. In the announcement, Junio notes a change to behavior in the “git fsck” command. It will now default to “git fsck –full” and check packfiles, which will mean that it takes longer to complete than it did previously. He also notes that the forthcoming 1.7.0 release will include a number of breaks to backward compatibility (i.e. use in scripts, rather than internal format changes). For example, “git send-email” will now do the right thing and not make deep threads by default, which has annoyed a number of people in the past.

The latest kernel release is 2.6.32.

Stephen Rothwell posted a linux-next tree for December 2nd. Since Wednesday the origin tree lost its build failures, the mips tree lost two conflicts, and the powerpc tree had a build failure for which a commit was revered. The total sub-tee count remains steady at 154 trees. Stephen mentions an interesting blog post (on neuling.org) with some linux-next stats.

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