LKML Summary Podcast

Audio: http://media.libsyn.com/media/jcm/linux_kernel_podcast_20091115.mp3

For the weekend of November 15th, 2009, I’m Jon Masters with a summary of today’s LKML traffic.

In today’s issue: compat-ioctl, cpuinfo, ftrace, locking, perf, and s390x.

Compat-ioctl. Arnd Bergmann posted a series of RFC patches intended to put the compat-ioctl support on a resource diet. He implements this through inlining some functions, simplifying various structures, and passing pointers directly rather than through a layer of indirection.

Cpuinfo. Ingo Molnar had previously suggested clearing the “ht” flag within /proc/cpuinfo for a given CPU core if it only had one sibling since, as Ingo pointed out, “ht” is meaningless in such a context. Peter Anvin was inclined to agree, although was initially confused by another proposal to rename the “ht” flag entirely.

Ftrace. Zhangjin Wu posted a 16 part patch series implementing ftrace support for MIPS systems. This was version 8, and incorporated various feedback as well as removal of needless use of -mlong-calls GCC flags.

Locking. Frederic Weisbecker and Hitoshi Mitake had a dialogue about plans for a full “perf lock” tool that will be used with the locking tracepoints to capture runtime performance data for kernel locks. On a side note, the “lockdep” events were renamed to simply “lock” since they pertain solely to locking and unlocking (rather than dependencies).

Perf. Mitake Hitoshi posted a 4 part patch series implementing a new “mem” subsystem within the perf bench utility, intended to evaluate memory performance, as well as support for comparing various different memcpy algorithms on as applied to specific CPUs. This was timely, given a discussion of prefered memcpy algorithms for Intel Core2 vs. Intel Nehalem CPUs. Separately, Peter Zijlstra noted that performance counters are limited supply on physical hardware and are multiplexed in the case that more are asked for that are present, which results in a warning that results have been scaled. He also reminded everyone that the granularity of the round-robin performance counter switching is fixed at the timer tick, which can result in some zeroed results for some counters if they don’t have chance to get some CPU time.

s390x. Martin Schwidefsky sent a series of 52 patches intended for the next merge window (2.6.33). 27 of those patches rework the common-io-layer, 6 patches are for fault handler optimization, and 6 add support for cex3 crypto cards and “some other stuff”.

Finally today, Stephen Hemminger thought he had found the faulty commit that was causing synchronous ext2 mounted floppy disks to trigger an IO error on remove. The problem disappears when a seemingly unrelated tracing patch is reverted, suggesting that there may in fact be a timing bug. Andrew Morton felt that actually, the warning was fairly harmless noise from ext2 (which was unable to write back metadata changes). The discussion continued.

In today’s announcements: LTTng 0.168. Mathieu Desnoyers announced LTTng 0.168 had been released for kernel 2.6.31.6. It includes a bunch of optimizations and bug fixes, including a new periodic flush scheme.

The latest kernel release was 2.6.32-rc7.

Stephen Rothwell posted a linux-next tree for November 13th. Since Thursday, Linus’ tree lost its build failure, along with Rusty’s ‘rr’ tree, and the sysctl tree too. Meanwhile, the sound tree gained a conflict against the omap tree. Stephen remerged at the end to get Linus’ -rc7 tag. The total sub-tree count remained steady at 148 trees.

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