• Immutable Page
  • Info
  • Attachments

KernelPodcast20090609

For Tuesday, June 9th 2009, I'm Jon Masters with a summary of today's LKML traff ic.

In today's issue: Linux 2.6.30, performance overhead, IO scheduler based IO controller, VIA Centaur CPUs, and procfs documentation.

Linux 2.6.30. Linus Torvalds finally announced the release of 2.6.30. Quoting Linus, "I'm sure we've missed something, and I know we have some regressions pending. At the same time, we do need the coverage of a [real] release, and on the whole it looks pretty good. We've fixed a few regressions in the last few days, and there's always 2.6.30.x". Linus refers to the kernelnewbies.org site writeup for further details.

Performance overhead. Continuing the previous discussion concerning the overhead of enabling certain kernel configuration options, Ingo Molnar pointed out that due to the way memory zones are allocated, we begin satisfying requests from the highmem zone first, which increases the need for kmap() mappings (on those x86-32 systems where this is applicable) if this memory happens to be used heavily in-kernel, for example for filesystem cacheing purposes, although - as Linus pointed out - it is intended this way in order to benefit the average case of userspace tasks, for which we obviously do not require a kmap/kunmap as the memory is directly mapped. The dicussion lead on to whether there is still a need for highmem kernels and kernel PAE support. The conclusion was that there is for the moment, since many 32-64 bit migrations have been offset by newer embedded users - though it was pointed out that many of these folks don't share bug reports on breakage.

IO Scheduler based IO controller. Vivek Goyal posted version 4 of his IO scheduler based IO controller patchset. This patchset aims to provide IO bandwidth control at the IO scheduler level, but does not directly address higher level (layered) logical devices - as before, Vivek includes some ideas about handling this, along with a link to a LWN article on the topic.

Converting block trace points to TRACE_EVENT. Li Zefan posted version 3 of a patchset intended to convert the block trace points over to TRACE_EVENT. As Li points out, doing so enables zero-copy and per-cpu splice() tracing, structured logging records exposed under debugfs, and arbitrary per-tracepoint filer expressions to be defined by the user. Li's benchmarks show that the performance hit of tracing is reduced marginally by using the new patches (especially because of the use of splice()), although it was already pretty small to begin with.

VIA Centaur CPUs. Harald Welte (who listeners may recall became the VIA Free and Open Source Software Liaison) posted a number of fixes for VIA/Centaur CPUs. He also responded to a discussion concerning setting the correct CPU frequency automatically in e_powersaver in which Michael Zick claimed that the cpuid instruction provides information about the Guaranteed Stable Frequency (GSF) expressing uncertainty that the CPUID provided anything more than the maximum clock frequency in ASCII.

Procfs documentation. Stefani Seibold posted a documentation update for procfs, covering the changes that have been made across recent releases. The update includes documentation of new /proc/self/status flags, as well as changes to the memory maps, including the newer smaps extension. The latter allows one to view the memory consumption for each of a process' mappings.

In today's announcements: Linux 2.6.30 was released on Tuesday. Christoph Hellwig posted an "XFS status update for May 2009", in which he noted a new release of xfsprogs and ongoing work on xfstests to allow those tests to be run on any filesystem, and not just XFS ones.

The latest kernel release is 2.6.30, which was released by Linus on Tuesday, at 8:36pm Pacific Daylight Time.

The next merge window won't really be open for a "day or two" because Linus is keen to see people actually test the 2.6.30 release first.

Andrew Morton posted an mm-of-the-moment snapshot for 2009-06-09-17-52.

The 2.6.29 -stable kernel. Greg K-H posted an 87-part patch review series in which he requested responses for 2.6.29.5-rc1 prior to June 11 at 09:00:00 UTC, which is likely to be after you hear this.

Stephen Rothwell posted a linux-next tree for May 9th. Since the previous compose last Thursday, two new trees were added ("irda" and "voyager"), one of which ("voyager") was immediately dropped due to a build problem. The tree still fails to build in a powerpc allyesconfig build configuration, and contains a large number of other conflicts. The total subtree count has increased to 144, given the addition of the two aforementioned items.

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 2009-06-11 07:12:52 by JonMasters