• Immutable Page
  • Info
  • Attachments

KernelPodcast20090514

For Thursday, May 14th 2009, I'm Jon Masters with a summary of today's LKML traffic.

In today's issue: Cross-platform device drivers, DRBD, KVM, Btrfs, and 2.6.30 deadlocks.

Cross-platform device drivers. Atul Mukker of LSI Corporation asked the kernel developers in LKML assembled what they thought about a new approach to driver architecting with cross platform compatibility in mind. There is a desire to have common driver code split between different Operating Systems, but in an Open Source friendly fashion - unlike certain other drivers. Jeff Garzik pointed out that Intel already do this, and so it can be done more or less "right", but it can also go very badly too - he itemized some of the ways.

DRBD. Neil Brown wants to give DRBD a proper review, and is therefore keen to ensure that he understands it. In order to do this, he posted a description of another implementation that could be made using md/raid1 + nbd (network block device) in order to provide similar functionality to DRBD. His description focused on data replication at the block level, as well as some specific alternations to the RAID1 level that would also be necessary.

KVM. Gregory Haskins posted version 8 of his patch series aimed at adding support for irqfd interrupt emulation for guests using the eventfd notification interface already supplied by the kernel. This patch update mostly supplies a few fixes to the versions from earlier in the week.

Btrfs. Chris Mason (lead Brtfs developer) posted some important updates for 2.6.30. These were small, but include a fix for improving the handling of certain IO failures when metadata reads fail because all copies of a given block are bad. He says, "There is still a bunch of work to do in that area, but this is a small start that avoids hopeless looping."

2.6.30 deadlocks. Jonathan Corbet (of LWN fame) posted a concerning bug report that his 2.6.30-rc builds were causing one of his systems to lock hard from time to time. It apparently isn't trivially reproducible but has started happening sometime since 2.6.29. The diagnostic output he provided from the 'lockdep' locking validator suggested some kind of internal lockdep bug caused because the number of lockdep chains was set too low. Peter Zijlstra said that Ingo Molnar had recently increased these limits in a patch set to go upstream.

In today's announcements: GlusterFS 2.0. Shehjar Tikoo posted an announcement about a heavily re-architectured version 2 release of GlusterFS. This GPLv3, filesystem agnostic cluster filesystem uses a regular Linux filesystem as its backend, but adds a layer of clustering on top. It is self-healing, features automatic data replication, is modular, etc. Further information is available at www.gluster.org. The stable kernel. 51 stable kernel 2.6.29 and 28 stable kernel 2.6.27 stable review patches were posted by Greg Kroah Hartman (comments needed by 8pm UTC on Saturday).

Linux Weekly News published the weekly "kernel page", which includes an excellent writeup of recent IO Containers discussion, as well as a new effort entitled "In Brief" that will cover ongoing issues and discussion threads.

The latest kernel release is still 2.6.30-rc5. A small number of fixes are waiting in Linus' tree for a likely impending 2.6.30-rc6 release.

Stephen Rothwell posted a linux-next tree for May 14th. Since Wednesday, the tree has gained fixes (via the sub-tree named 'fixes') for three net issues.

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 2009-05-15 06:14:52 by JonMasters