LKML Summary Podcast

Audio: http://media.libsyn.com/media/jcm/linux_kernel_podcast_20090804.mp3

Apologies for lagging again, what can I say? I spent most of the weekend working on work stuff and had a root canal last week to deal with. Here’s a hint to anyone considering avoiding the dentist for 6 years – it will catch up with you in the end. Especially if you’re British like this poor author.

For Tuesday, August 4th, 2009, I’m Jon Masters with a summary of today’s LKML traffic.

In today’s issue: Initdev, OOM, poisonous hardware, SIMPLE_PM_OPS, and TTY.

Initdev. If at first you don’t succeed. David VomLehn reposted a patch series intended to implement generic support for delaying the kernel boot waiting for devices to be enumerated and setup, as well as promptly continuing once the lack of an optional device has been determined. The idea behind these patches may be specific to Cisco, but they could help reduce the need for hacks such as scsi_wait_scan and “rootwait”. David says that the 7 part patch series could go in as a unit, but has split it out for easier review. And on that note, he thanks Alan Stern for his comments on the last posting.

OOM. Kosaki Motohiro brought up the change in oom_adj semantics again. This time, he describes the shared oom_adj for processes sharing a struct mm as a regression that requires fixing prior to 2.6.31 final. Other patches floating around add a new “oom_adj_child” that will affect only settings for subsequent processes, but it is not clear that Kosaki is happy with this. It certainly feels a little like someone needs to make a call on exactly how the oom_adj pseudofilesystem entries are going to be cleaned up, documented, and exposed through to userspace in a fashion that won’t cause livelocks.

Poisonous hardware. Andi Kleen posted the latest version of HWPOISON, which is a series of patches intended to support Intel MCA recovery (recovery from certain classes of hardware memory error). HWPOISON implements a high level machine check handler that can catch accesses to pages that have gone bad (mm/memory-failure.c) and can often do something about it. The latest version introduces a new VFS operation of “error_remove_page” that will trigger on a page used by a filesystem going bad. Andi requests further test.

SIMPLE_PM_OPS. Albin Tonnerre posted in regard to the ongoing migration to the new dev_pm_ops dynamic power management interface, suggesting that the current migration was too prone to regression (since all fields must be assigned in order for the new PM code to function), and proposing a SIMPLE_PM_OPS macro that can be used to initialize a dev_pm_ops with at least sane defaults.

TTY. Linus Torvalds continues to be fairly involved with the ongoing TTY layer saga (perhaps he should just own that stack and be done with it). He posted some ldisc locking rewrite patches, which (new TTY maintainer) Greg Kroah-Hartman picked up into the TTY tree, and then promptly sent back to Linus as a merge request. Greg noted, “As you wrote these, I think you know what they fix ”.

In today’s miscellaneous items: Some ALSA fixes (Takashi Iwai), some block bits from Jens Axboe (one pre-requisite topology fix, and removing the long overdue “experimental” label from bsg), a large number of networking fixes from David Miller (several regressions targeted), some performance counters fixes from Ingo Molnar (including a useful perf top fix from Arnaldo Carvalho de Melo to ignore mwait_idle_with_hints), some scheduler fixes (also Ingo Molnar), some timer fixes (also Ingo Molnar), a fair number of x86 fixes (also Ingo Molnar), version 5 of the GPIO regulator patches from Roger Quadros, ome NILFS2 fixes from Ryusuke Konishi, a “bundle of fixes” (most of which target s390 systems) for SCSI from James Bottomley, an analysis of lockdep behavior from Peter Zjilstra and David Howells, some patches to convert alpha to asm-generic from Christoph Helwig, the ability to export and unexport named GPIO devices in GPIOlib from Ben Dooks, a regression in AoE support from Bruno Premont, a new version 0.2 of Stefani Seibold’s “kfifo” API, some DRM fixes from Dave Airlie, a suggestion from Vivek Goyal that benchmarks of his IO scheduler based IO controller should be repeated several times and averaged (as some of the figures don’t seem to quite make consistent sense), an updated version 2 of the vbus_enet driver from Gregory Haskins as used in the Novell “AlacrityVM” fork of KVM (which uses the new “vbus”), some new trace events in the VM (Mel Gorman), and a rant from Pavel Machek that the linux-arm-kernel mailing list is not allowing him to crosspost with LKML when there are ARM regressions to discuss.

Finally today, Arnd Bergmann reminds us of the reason why one cannot perform a rename() over a directory symlink, by referenceing several Open Group specificiations.

The latest kernel release was 2.6.31-rc5 and still is current.

Andrew Morton posted an mm-of-the-moment for 2009-08-04-14-22. Several people encountered problems with the previous one, especially in SLQB.

Stephen Rothwell posted a linux-next tree for August 4th. Since Monday, a number of trees lost builds failures (md-current, cpufreq, tip, and oprofile), while the tree otherwise remained consistent at 138 sub-trees in the compose.

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