LKML Summary Podcast

Audio: http://media.libsyn.com/media/jcm/linux_kernel_podcast_20091201.mp3

The LKML summary podcast is about to enter its 7th month, and recently passed its 100,000th download. Thanks for making the podcast popular and for listening to the semi-daily updates. I am trying to do it daily so as to avoid the insanity of catching up on several weeks worth of podcasts last weekend!

For Tuesday, December 1st, 2009, I’m Jon Masters with a summary of today’s LKML traffic.

In today’s issue: Allocators, coding style, and interrupts.

Allocators. Matt Mackall replied to an ongoing thread concerning a previous “false positive” lockdep warning in the SLOB allocator (an allocator optimized for handling very small amounts of system RAM – less than 16MB) that had warped into a discussion of merging SLOB into an existing allocator and shoving it somehow under the CONFIG_EMBEDDED Kconfig hierarchy. Matt noted that there was little anecdotal evidence as to the difficulties of maintaining multiple allocators and that SLOB “is the least mergeable of the set”. He was quite annoyed that a thread discussing lockdep had gotten so off topic. Others (including Christoph Lameter) were less convinced of widespread SLOB use.

CodingStyle. Following on from the previous criticisms of changes in the net tree that Joe Perches had been involved with and which had raised a concern as to proper coding style conventions in the kernel tree, Joe send another “stylistic cleanups” patch, this time targeting the floppy driver. He didn’t express much enthusiasm for its chances of being merged, however.

Interrupts. Alan Cox noted that the 8390 network controller (and perhaps therefore other devices) is a “single thread device” in which one must switch its internal addressing to page 0 in order to determine interrupt state. Thus a busy multi-procressor system cannot easily probe the device to find out whether a given interrupt came from it without disturbing its state. This was part of the larger ongoing discussion recently revived by Thomas Gleixner as to the future ways in which Linux will handle interrupts – an increasingly threaded interrupt world in which the small quiescent handler used by the threaded IRQ implementation would run with interrupts disabled.

Finally today, Jamie Iles posted some initial questions as he begins to poke at providing support for performance counters on ARMv6. In particular, since ARM revisions to date don’t do atomic 64-bit operations, he wonders whether using spinlocked versions would introduce a huge performance hit.

The latest kernel release is 2.6.32-rc8.

Stephen Rothwell posted a linux-next tree for December 1st. Since Monday the origin tree gained a build failure for which Stephen reverted the offending commit, and a build failure correction patch from Ingo Molnar was added. Of the trees, pxa, and trivial lost issues while powerpc, net, mtd, and percpu gained some conflicts. The pcmcia tree lost two conflicts but gained another against the arm tree. The total sub-tree count remains steady at 154.

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