LKML Summary Podcast

Audio: http://media.libsyn.com/media/jcm/linux_kernel_podcast_20091119.mp3

For Thursday, November 19th, 2009, I’m Jon Masters with a summary of today’s LKML traffic.

In today’s issue: 4096+ CPUs, KVM, licensing, performance, procfs symlinks, zombie processes, and a new kernel -rc release.

4096+ cpus. Rusty Russell posted a 6 part patch series intended to complete the conversion of cpumask support over to supporting dynamic allocation for greater than 4096 CPUs. Now, using a configuration option and accessor functions, one can choose to support CONFIG_CPUMASK_OFFSTACK or not. Legacy on-stack uses (CONFIG_CPUMASK_OFFSTACK=n users don’t notice much except that the mm_cpumask and tsk_cpumask accessors are now required).

KVM. Avi Kivity posted 35 patches comprising batch 2 of a 2 part batch of patches intended to hit the forthcoming 2.6.33 merge window. Highlights were covered previously and include improved kernel context switching speed, better interoperation with other users of virtualization extensions, improved IRQ scaling, nested SVM improvements and tracing, improved cpufrequ integration, and spin loop detection on newer hardware. Your author continues to be scared by the productivity of the KVM team.

Licensing. Rusty Russell brought up the GPLv3 in asking Tim Abbott (of ksplice) about the license on the binary search library within lib/bsearch.c. He wanted to use it for SAMBA, which is GPLv3, and noted that this is one reason why all of his kernel code is explicitly licensed as “v2 or later”. He requested a small license change on that code so that it can be used in Samba without being re-implemented.

Performance. Ajay Patel posted a series of lmbench runs showing a comparison between 2.6.18 and 2.6.27 in which he showed what appeared to be a significant performance degradation between the two kernels. The kernels were actually not really an Apples to Apples comparison though, because one was a modified “Enterprise” kernel and the other was a Fedora kernel, which has a very different set of compilation options. In both cases, those kernels are now very old and so the results are hard to work with.

Procfs symlinks. Jeff Layton posted a 4th attempt at a patch to /proc symlink support, this time causing such symlinks to behave like regular symlinks. The existing code treated such symlinks in a special fashion and resulted in a failure of the path walking code to revalidate such dentries (so if the state of the associated file permissions changed during the running task, it could result in certain contrived security issues), which lead to various issues. The patch results in one user-visible change in the form of unlinked files no longer appearing in procfs for opening – so we can say goodbye to admin hacks that restore deleted critical system libraries on production systems.

Zombies. Nick Piggin posted to let everyone know that he experienced a zombie process during various ptracing on a recent kernel. He also seemed to lose job control around the same time. The task was last in proc_clear_tty.

In today’s announcements: Linux 2.6.32-rc8. Linus Torvalds announced release 2.6.32-rc8 of the Linux kernel at 2:56pm Best Coast Time (PST). As he puts it, “the way things are going, this will likely be the last -rc”. He points to the usual need for eyeballs on regressions, and notes that he’s away for the entire US Thanksgiving week and won’t be working on a release then.

The latest kernel release is 2.6.32-rc8.

Stephen Rothwell posted a linux-next tree for November 19th. Since Wednesday, the new, sound, and cpufreq trees lost conflicts, while the input and edac-amd trees gained conflicts. The total subtree count remains stready at 151.

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