• Immutable Page
  • Info
  • Attachments

KernelPodcast20090512

For Tuesday, May 12th 2009, I'm Jon Masters with a summary of today's LKML traffic.

In today's issue: vmscan, irqfd, OOM, and Xen.

vmscan. Ongoing work is taking place to improve the page evict code in situations such as a typical desktop environment, and in particular to protect referenced PROT_EXEC pages from being displaced by large single-use file IO or other similar activities. Wu Fengguang posted a patch that aims to make mapped executable pages a first class citizen by keeping them on the active list as long as they are referenced during each full scan of the active list. The patch interacts with another recent one from Rik van Riel that aimed to explicitly evict use-once pages (for example, those due to an updatedb operation) by preference.

irqfd. Gregory Haskins posted version 6 (and later in the day, versions 7, 7.1, and version 7.2) of a patch aimed at adding irqfd support to the Kernel Virtual Machine (KVM) hypervisor. This patch adds a mechanism for injecting a specific emulated interrupt into a running guest. Any valid signal on the irqfd will translate into an interrupt seen by the running virtual machine when the next availabe interrupt window is reached.

OOM. Christoph Lameter took issue at ongoing use of the term "Out of memory" to describe the process whereby the kernel fails to perform a memory reclaim and must kill something to free up memory for future and ongoing allocations. When "out of memory" occurs, the system may not actually be out of memory at all, but the kernel was certainly unable to make use of whatever was available at the time of the OOM. Christoph says that he's seen scores of people add memory to systems or lower ulimit values for users after receiving warnings from the OOM killer, expensive actions and not necessarily helpful.

Xen. Jeremy Fitzhardinge reposted some Dom0 core changes and requested inclusion against a specific topic branch to be created within Ingo's x86 tree on approval. The changes are varied and include some basic functionality required of the hypervisor (such as basic IO permissions), and some cleanups. Jeremy also posted updated Xen Dom0 PCI access patches, swiotlb updates, Xen APIC hooks, a Xen /proc/mtrr implemenation, and likely more. Jeremy has been aggressively pursuing the upstreaming of Xen Dom0 support recently, to complement the existing support for DomU guests built into the kernel.

In today's announcements: Thomas Gleixner posted version rt12 of the RT patches against stable kernel 2.6.29.3. This is a simple rebase featuring loadavg calculation optimization and some futex fixes from Thomas and Peter Zjilstra. Andrew Morton release an "mm of the moment": 2009-05-12-16-51. It contains a large number of patches against 2.6.30-rc5. Kay Sievers announced udev version 142 had been released. The release furthers the ongoing effort at consolidation and cleanup, and also includes an updated libudev for which no major changes of interface are planned. Version 2.6 of the alternative Gujin bootloader was released.

The latest kernel release remains 2.6.30-rc5, which was released by Linus on Friday afternoon, just over a week after the release before that. A number of additional patches are now sitting in Linus' tree, leading to a likely release of a 2.6.30-rc6 kernel at the end of this week.

Stephen Rothwell posted a linux-next tree for May 12th. Since Monday's tree, the driver-core tree gained a build failure for which Stephen reverted the offending commit, and the staging tree lost a conflict. There remain 136 trees comprising the linux-next tree at this time.

That's a summary of today's LKML traffic. For further information visit kernel.org. I'm Jon Masters.

Tell others about this page:

last edited 2009-05-13 06:19:12 by JonMasters