LKML Summary Podcast

Audio: http://media.libsyn.com/media/jcm/linux_kernel_podcast_20090823.mp3

You know the drill, so all together now: “Another week, another -rc kernel”.

For the weekend of August 23rd, 2009, I’m Jon Masters with a summary of today’s LKML traffic.

In today’s issue: Mailing lists, Offline scheduling, TuxOnIce, and x86.

Mailing Lists. Pavel Machek’s latest comments about the linux-arm-kernel mailing list (refering to a regular reminder mail sent to subscribers that he claimed was out of sync with recent discussions) obviously struck a raw nerve with Russell King, who decided to shut down those mailing lists immediately, telling everyone to refer to Pavel’s “extremely selfish attitude”. There have been recent efforts to move discussion onto an open (unmoderated) mailing list hosted on vger.kernel.org (there is already one – linux-arm – but it is not widely known and not in active use by the wider community of ARM developers) and it seemed as if this may finally happen this time, but instead David Woodhouse (with Russell’s blessing) setup alternative (open) lists on lists.infradread.org. Many other kernel hackers were unhappy with Russell’s actions – David Miller publicly stated that “a lot of us are tired of your crap, and like Alan Cox I can’t take you seriously at all”, while Ted T’So was somewhat more diplomatic but stated that he was “appalled” by the behavior.

Meanwhile, David Woodhouse, with a certain eloquance, debated the relative merits of automatically resubscribing all existing members of the old list to the new one, and especially whether some would be confused and flood the list with “please unsubscribe me” messages. David said “there’s no accounting for the stupidity of the human animal” and, ‘Even with the notification and the first major thread being “ARM Mailing lists have moved”, you’re right that there are probably a few people who are _so_ stupid that they can’t manage to work it out. I’ve sent a mail with words of one syllable or fewer to each list, which might hopefully help — but I’m sure there’ll still be some. When that happens, we’ll unsubscribe them — and send in the ninjas to ensure they don’t accidentally breed’. Catalin Marinas pointed out that the vger list will likely stick around for any time the new lists finally drive David mad.

Separately, there appeared to be some concensus that the Linux InfiBand/RDMA mailing list could be moved over to vger.kernel.org.

Offline scheduling (yes, that’s not a type). Raz Ben Yehuda mailed to let everyone know about some work ongoing at The Open University of Israel Department of Mathematics and Computer Science. They look at the problem of assigning dedicated processors in general purpose computing systems such as Linux, in which even a “dedicated” processor must peform a lot of work that relates to general system housekeeping, and other activities. Instead, they exploit the ability to offline a CPU and logically detach it from the running system, leaving it to run whatever dedicated code has been assigned to it. The example cited is that of offloading packet analysis in a firewall to one of these “offlined” cores, but there are many other potential uses. Some wondered why the same features couldn’t be achived with existing cpusets, virtualization, and other efforts. Ben’s reply included that this work could be used for “hard realtime” systems (this smells a lot like some of the other real time solutions available for Linux systems in that regard).

TuxOnIce. On Thursday, Nigel Cunningham had mailed to suggest various ways forward for the development of the “TuxOnIce” alternative suspend patchset. Nigel and others have not had enough time for development recently and don’t see that situation changing any time soon, and seemed to favor someone taking over development altogether. After a little silence, Jiri Slaby stepped up to the plate and suggested that he could help sheppard pieces of TuxOnIce into the mainline swsusp implementation to make major features more available. Others, including Xavier Gnata, offered their time in testing. Meanwhile, Nigel sent another oration in response, with a “todo list” included.

X86. Thomas Gleixner posted a 32 part patch series intended to refactor the setup code used in x86 to provide a base suited to embedded platforms, in light of the arrival of Intel Moorestown support, which Thomas says “indicate the arrival of the embedded nightmare to arch/x86″. As Thomas puts it, “Moorestown is a SoC with an x86 core and a bunch of random peripherals glued around it. It finally gets rid of legacy hardware like PIT, 8042 et. al. but on the other hand it introduces the full embedded horror by adding random peripheral IP cores as an replacement which are glued onto the x86 CPU with duct tape and other nasty tricks.” Thomas was particularly kind towards the design of the AHBT timer and then got down to the business of describing how the new patches refactored the setup code without replacing paravirt_ops.

In today’s miscellaneous items: a 46 part patch series representing part 3 of a 4 part patch series of KVM updates targeted for 2.6.32 from Avi Kivity, a bunch of staging patches from Bartlomiej Zolnierbiewicz, some “use printk_once” patch series’ from Marcin Slusarz, some s390x updates from Martin Schwidefsky, a PCI fix (correcting a PCI suspend/resume problem) from Jesse Barnes, a patch to reuse the boot-time mappings of fixed_addresses from Xiao Guangrong, a large number of RCU patches from Paul McKenney, an interesting munlock fix from Hiroaki Wakabayashi (who noticed that due to recent work to make get_user_pages interruptible, we can end up with a situation in which some pages passed to mlock are not actually pinned following a well-timed SIGKILL, and these will later result in a completely pointless page fault at munlock during exit), some tracing fixes for generic syscall events from Josh Stone, conversion to the new dev_pm_ops patches from Marek Vasut (which Nicolas Pitre questioned might be buggy), some SCSI fixes from James Bottomley (entirely mpt2sas fixes), the latest version of the flex_array patches from David Rientjes, an Intel Atom CPU configuration target from Tobias Doerffel, some m68k and m68knommu updates from Geert Utterhoeven (including support for Performance Counters), a backport of TREE_RCU to 2.6.27 from Paul McKenney (it had been pointed out previously that this was likely required for certain users of that kernel, which I believe is used by a particular Enterprise Linux kernel), another round of O_SYNC patches fro Jan Kara (17 patches implementing a single path for O_SYNC and standard syncing), a single wireless fix from John Linville (for rtl8187b parts), version two of Performance Counters support for IA64 from William Cohen (with the V2 email re-written by Ingo Molnar to include appropriate references to the previous version), some fixes to build Intel TXT (Trusted Boot Technology) on non-x86 platforms from Shane Wang, an informative mail from Gregory Haskins concerning “vbus design points” and how he is particularly proud of the design of the vbus shared-memory model, a fix to update_process_times from Peter Zijlstra that delays waking up softirqs from jiffy ticks in an attempt to fix broken runqueue balancing on recent RT kernels, an updated git repository from Dave Airlie containing the same DRM fixes but this time without some weird corruption, a btrfs fix from Jens Axboe that corrects a red/black (rb) tree corruption, version 4 of the automatic crashkernel patches from Amerigo Wang, ongoing discussion of the merits of sysfs and configfs over ioctls (mostly Avi Kivity, but some others also), and some updates to checkpatch from Andy Whitcroft.

In today’s announcements: Linux 2.6.31-rc7. Linus Torvards announced the release of version 2.6.31-rc7 of the Linux kernel. The latest rc includes many driver and architecture updates, with lots of other smaller fixes. Quoting Linus, there are “some inotify fixes here to, but I don’t think we’ve confirmed whether they help the (apparently very hard to trigger) oopes some people have seen. When Linus isn’t announcing new kernel releases, he’s giving advice on using ftrace to track down stack overflow issues – nice to see that ftrace is getting wide use and recommendation for all manner of situations.

2.6.31-rc6-rt6. Thomas Gleixner announced version 2.6.31-rc6-rt6 of the ongoing preempt-rt development. In the latest Real Time kernel, one can find a rebase to Linus’ recent git, and a load accounting/balancing fix from Peter Zijlstra. A problem with ARM highmem use remains unresolved.

Git version 1.6.4.1. Junio C Hamano announced the release of version 1.6.4.1 of the git SCM tool used in Linux kernel development (and by other projects). The latest version includes fixes to “git am”, documentation updates (e.g. for fast-forward), and a lot more besides.

Gujin GPL bootloader version 2.7. Etienne Lorrain announced the release of version 2.7 of the “Gujin” bootloader, which features support for several Linux distributions in the form of packages. This latest release can parse ISO format files containing filesystem images and extract kernels from within, allowing one to (in theory anyway) keep an image of a LiveCD around as a boot target. This doesn’t quite work (yet) for many standard live images, but one can see the direction this is heading.

The latest kernel release is 2.6.31-rc7, which was released by Linus Torvalds on Friday evening at 6:26pm PDT.

Several people have found issues with rc7 already, including a build error with the oom_adj reversion discovered by Geert Uytterhoeven, and lost hardware sensors support in the case of Gene Heskett. Jes Sorensen thought he had seen a problem in the scheduler, but it wasn’t easily reproduced. Ingo Molnar believed it might be a rare timing bug and asked for any logs, which didn’t exist. So whatever may or may not exist there probably will stick around.

Stephen Rothwell posted a linux-next tree for Friday August 21st. Since Thursday, the nfs tree gained a build failure, the drm tree lost all of its conflicts, and the agp tree gained a conflict against the powerpc tree. The sub-tree count remains steady today at 140 trees in the linux-next tree.

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