LKML Summary Podcast

Audio: COMING SOON

For Tuesday, December 15th, 2009, I’m Jon Masters with a summary of today’s LKML traffic.

In today’s issue: ELF, power capping, OOM, and TSC.

ELF. Hatayama Daisuke posted a four part RFC patch series implementing an unofficial ELF (Executable Linking Format) extension known as “Extended Numbering”. This is (apparently) originally a Solaris hack that extends the possible number of program headers contained within an ELF file to greater than the 64K limit currently imposed by the standard in the case that the kernel dumper is dumping out a core file that contains a great many mmaps. It’s worth bearing in mind that the ELF standard has an unhealthy relation with a certain litigious corporation and hasn’t been updated in some time [that was at least my experience when I read both the 32 and 64 bit versions of the ELF spec and inquired who was maintaining it now.]

Power capping. Salman Qazi posted to let everyone know about a project within Google refered to as “power capping”. Briefly, this is an effort to break the longstanding relation between power supply and demand in the data center by overcommitting resources and running those resources more efficiently, by using such measures as “forced idling” of CPUs within the Linux kernel. The email thread was entitled “RFC: A proposal for power capping through forced idle in the Linux kernel”, and detaile a proposal for in-kernel assistance, an “idle cycle injector”, and so forth. Similar work has been done before, but it generated a fair amount of discussion. It would seem one of the ideas here is to allow datacenters to ride out temporary reductions in available power, rather than powering down systems. Arjan van de Ven would like to see all CPUs idling in unison, while Salman is interested in this being a userspace controlled policy decision.

OOM. Hiroyuki Kamezawa posted another version of his mm rss counting patches, intended to faciltate easier statistical gathering of RSS (Resident Set Size) data per task, and without scalability impact. This can be used to improve the OOM killer, and for other purposes. This version is less “invasive”.

TSC. Zachary Amsden posted a series of preparatory patches intended to unify support for the virtualized TSC emulation between VMX and SVM implementations within Linux. There is a lot of churn currently in this space as folks attempt to present a consistent TSC experience.

In today’s announcements: rt-tests version 0.57. Clark Williams announced that version 0.57 of the rt-tests package intended to provide various test cases for the preempt-rt kernel patch series, is now available from kernel.org. On a related note, Keith Mannthey (IBM) announced that the “ibm_rtl” driver is available, a module that tells various IBM BIOSes to turn off particular gregarious SMIs that can cause all manner of system latencies. I wonder if we need a generic framework for doing this and am considering it.

The latest kernel release was 2.6.32.

Jens Axboe reported a kexec boot regression affecting kernels after 2.6.32. In order to avoid excessive boot delays, Jens test boots new kernels through an kexec call in general, and this failed recently due to a patch from Yinghai Lu, according to the git bisect (a patch that was confirmed working was posted). Matthew Garrett also reported an ACPI (non-)boot regression.

Stephen Rothwell posted a linux-next tree for December 15th. Since Monday, the origin tree gained a build failure for which Stephen applied a patch, the md-current tree lost its conflict, the mips tree gained a conflict against Linus’ tree, the 52xx-and-virtex tree lost its build failure but gained a conflict against Linus’ tree, the acpi tree still has its build failure so Stephen used the version from Friday, the voltage tree gained 2 conflicts against Linus’ tree – one requiring a fixup and the other requiring a rollback of the voltage tree to the version from November 20th, the devicetree tree lost its build failure, the limits tree lost three conflicts, the tip tree lost its conflict, and the percpu tree lost its conflicts. Stephen repeated his usual “call for calm” in not merging patches intended for 2.6.34 until after the release of 2.6.33-rc1. The total subtree count remains steady at 155 trees in the latest compose.

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