LKML Summary Podcast

Audio: http://media.libsyn.com/media/jcm/linux_kernel_podcast_20090714.mp3

Apologies folks, we’re playing post-Linux Symposium catchup at the moment.

For Tuesday, July 14th 2009, I’m Jon Masters with a summary of today’s LKML traffic.

In today’s issue: Hotplug, memory leaks, MPU, OOM and VGA.

Hotplug. Lai Jiangshan noticed that the CPU hotplug code temporarily changes the affinity of running tasks during hotplug operations. This breaks several user-visible behavioral expectations, especially if an error occurs during the hotplug operation, in which case the affinity may not be restored to the previous assigned configuration. The patch aims to rectify this behavior.

Memory leaks. Alexey Fisher, Pekka Enberg, and Catalin Marinas had a debate about the possible existence of as many as 60 memory leaks on Alexey’s system, especially affecting ext4. Catalin’s kmemleak checker was reporting that the system in question was leaking memory and ultimately decided that the leak warnings were legimate and not further unwanted noise from kmemleak itself. Catalin posted a new thread specifically on ext4 and requested that the ext4 developers provide clarification of their intended behavior in the code.

MPU (Memory Protection Units). Mike Frysinger posted a patch and accompanying explanation of the advanced features of modern MMU-less architectures. Even though these systems lack a full blown Memory Management Unit (MMU) and virtual memory (usually for reasons of cost and complexity), certain architectures – like the Blackfin – actually provide some protection against memory corruption even without virtual memory. The MPU allows protections against specific ranges of physical memory by programming processor registers, similar to e.g. BATs on other architectures.

OOM. Gene Haskett and Fengguang Wu continued debating the recent spate of OOM killer attacks in 2.6.31 rc kernels. Gene noted that his test box has been “boringly stable” with recent kernels and so he was going to reboot it as the “witching hour” was rapidly approaching with regard to finding OOM problems. Separately, Rik van Riel posted a patch that preserves a task’s oom_adj value on fork (clone), which seems like a very good idea to this author.

VGA. Tiago Vignatti posted a generic VGA arbiter implementation. This is necessary for legacy graphics cards providing mapping of legacy VGA IO addresses such that two cards both supporting legacy VGA do not fight for the addresseses in question. The patch includes documentation.

In today’s miscellaneous items: A fix to the readdir() implementation for procfs such that TGIDs not previously showing up in the directory listing would be shown (Nikanth Karthikesan), an implementation of gmtime and localtime for the kernel (based on glibc) from Zhao Lei (however this author wonders whether there isn’t already an implementation of this – it seems dubious there would not already be something available in the kernel time code), various core, x86, tracing, and timer fixes from Thomas Gleixner, infiniband fixes from Roland Drier, dlm fixes from David Teigland, some fairly serious networking fixes from David Miller (including a fix for missing methods in the netdev_ops network device driver conversion process by Stephen Hemminger et al), libata updates (Jeff Garzik), and some compilation warning fixes.

In today’s announcements: Pierre Ossman posted requesting a new maintainer for the MMC subsystem since he doesn’t have enough time these days to give it the love that he would like to give.

The latest kernel release was 2.6.31-rc3, which was release by Linus on the weekend.

Stephen Rothwell posted a linux-next tree for July 14th. Since Monday, Stephen dropped the USB (quilt import problem) and Staging (depends on USB) trees and the overall linux-next tree still fails to build in an allyesconfig build configuration on powerpc. Several other failures necessitated patches. The total sub-tree count in the day’s compose is steady at 132 trees.

That’s a summary of today’s LKML traffic. For further information visit kernel.org. I’m Jon Masters.

Audio: http://media.libsyn.com/media/jcm/linux_kernel_podcast_20090713.mp3

For Monday, July 13th 2009, I’m Jon Masters with a summary of today’s LKML traffic.

In today’s issue: Mailing lists, modules, and a new RC release.

Mailing Lists. Matti Aarnio posted to let everyone know that he is currently experimenting with a new list manager for vger.kernel.org. It is not based on various existing implementations – for example mailman, majordomo (the current, long in the tooth mailing list manager), etc. – and is instead a re-write using an SQL database to store all aspects of list information, and list management actions. If things go well, it will replace majordomo on vger and become the default mailing list manager for kernel.org mailing lists.

Modules. Andi Kleen followed up to the recently posted patches for RO/NX page protection enforced bits for loadable modules – which had devolved into a discussion as to the purpose of building loadable modules for kernels in cases where the distribution vendor enforces always loading the module – with a summary of the reasons why modules are still useful (especially disabling modules that have a problem, or replacing them). He revived the discussion of supplying Modules.builtin metadata concerning modules pre-compiled into the running kernel so that module loading tools can take appropriate action.

In today’s miscellaneous items: an interesting treatise on the various arguments for and against SFI being more ACPI-like (Peter Stuge – and for more information on SFI, take a look at Len Brown’s excellent, and highly explanatory Linux Symposium 2009 presentation), version 3 of a patch series implementing various helper OOM analysis logging functions, version 3 of another patch series implementing notifcation of memory usage limits for tasks, a fix to enable kdump if a second kernel is loaded and oops occurs with panic=oops set on the kernel command line (Ken’ichi Ohmichi), version 2 of a series of patches converting SPARC to asm-generic/dma-mapping-common.h and pci-dma-compat.h (Fujita Tomonori), and a suggestion from Mitchell Erblich that the SLAB allocator be changed to reap SLABs only after a configurable delay. Additionally, Andi Kleen posted a patch modifying a public (procfs) interface such that file size information on /proc will now include deleted (but still open) files in the calculation.

In today’s announcements: LTTng 0.146. Mathieu Desnoyers posted to let everyone know that LTTng 0.146 will add extra read-side sub-buffering for its flight recorder, which came complete with a technical summary, and version 1.0.1 of the SCST Target driver for QLogic 24xx adapters (Vladislav Bolkhovitin).

The latest kernel release is: 2.6.31-rc3, which was released by Linus on Monday evening PDT. Linus repeats his call that he wishes things would “calm down”, which – as his points out in the accounement – they have, to a degree.

Stephen Rothwell posted a linux-next tree for July 13th. Since Friday, the ia64 is now under control of Fenghua Yu, the tree still fails to build in an allyesconfig build configuration on powerpc, and several net gains in conflicts occured also. The total sub-tree count remains steady at 132 trees.

That’s a summary of today’s LKML traffic. For further information visit kernel.org. I’m Jon Masters.

Audio: http://media.libsyn.com/media/jcm/linux_kernel_podcast_20090712.mp3

Would you like to record a 30 second summary of your latest development efforts or favorite patch series? I’m at the Linux Symposium in Montreal, so come find me – I’m in the program(me). Look for the guy with too much coffee.

From Montreal, Canada, for the weekend of July 12th 2009, I’m Jon Masters with a summary of today’s LKML traffic.

In today’s issue: RT deadline scheduling, pipe fds, and SLAB replacements.

Deadline scheduling. Henrik Austad posted an RFC concerning a new scheduling policy/class intended to implement deadline driven scheduling, using a modified version of MLLF (Modified Least Laxity First) called Earliest Failure First (EFF), which orders tasks according to when they will miss their deadlines, not when the actual deadline is. The idea seems quite interesting and has triggered a large amount of discussion both on the LKML and also on the linux-rt Real Time mailing list, both on which the discussion was cross-posted. Amongst the concerns was the issue of priority inversion, but the notion of delaying until the last possible moment lead Peter Zjilstra to suggest that the algorithm would “utterly fall apart on overload and would thus be unsuited for soft-rt loads”.

Pipe fd leaks. Over the past few days, there has been some discussion concerning an fd leak for tasks calling pipe() with an invalid address. A fix was proposed by Changli Gao last week, and brought up again by Amerigo Wang, who suggested a trivial problem with the patch prevented Linus from merging it. In reality, though, it turns out that Linus was actually holding off taking this patch since he feels “If you give a bad area to pipe(), there’s no point in closing the file descriptors. It’s a user-space bug. You got your file descriptors, you just don’t know what the hell they are, because your program is [expletive]“.

SLAB. There is some discussion ongoing about last weeks’ “mudflap” GCC support patches – that aim to add support a feature of the GCC that can track down various memory complications. As part of this discussion – which wound up discussing the need for modifications to alternative SLAB-replacement allocators – came a reminder that one of these would be replacement for SLAB, SLQB, is facing some more delays due to “a bunch of problems” that Nick Piggin is currently working on for 2.6.32. For his part, Pekka Enberg stated that SLAB is going to die in the future, but that “we’re [not] going to blindly remove it if it performs better than the alternatives” in certain situations. In this case, those situations included networking benchmarks.

In today’s miscellaneous items: some perfcounter updates (Ingo Molnar – including the module symbol and annotation support), some ptrace updates for the new S+Core architecture (Liqin Chen), some trivial ALSA fixes (Takashi Iwai), a bunch (32) more MMC updates from Nokia (Adrian Hunter), a cond_resched() optimization removing one conditional in this hot path code (Peter Zijlstra – and some other optimization discussions were triggered as a result of this, even comment cleanup), the usual raft of fixes from Ingo Molnar (tracing, core), some nilsfs2 fixes (Ryusuke Konishi), a new version of the PAT patches (Venkatesh Pallipadi), some DRM fixes (Eric Anholt), MCE cleanups (Andi Kleen), version 9 of the soft limits for memory resource controller partches (Balbir Singh), some block/io bits (Jens Axboe), another version of the kernel command line parsing limitation reduction patch series (Daniel Mierswa), a new rb_for_each helper for generic red/black trees in the kernel library, a series of updates from Greg Kroah-Hartman (USB, driver core, staging), some hwmon fixes, some ftrace fixes (Rakib Mullick), and another general Xen Dom0 rant concerning its interaction with swiotlb.

Finally today, Ingo Molnar posted followed up to the posting of the (smells like ACPI) SFI implementation from Len Brown with a treatise on patch commit quality, and in particular, the use of patch title capitalization. It turns out Ingo has some pretty high standards, even for grammar.

The latest kernel release is 2.6.31-rc2, which was released by Linus on the 4th of July (current actual version post-this date is 2.6.31-rc3).

Stephen Rothwell posted a linux-next tree for July 10th. Since Thursday, the tree still fails to build in an allyesconfig build configuration on powerpc, Linus’ tree lost its build failure, and the ttydev tree lost a conflict. The total sub-tree count is steady at 132 trees in the current compose. There is some discussion suggesting an increase in OOM activity (on hackbench runs) for linux-next 0708 and more recent – the cause is being worked on.

That’s a summary of today’s LKML traffic. For further information visit kernel.org. I’m Jon Masters.

Audio: http://media.libsyn.com/media/jcm/linux_kernel_podcast_20090709.mp3

I’ve been in the air for 32 hours over the past few days, on a 3 day trip to Japan, apologies for lag, jetlag, and catchup I’m now playing.

For Thursday, July 9th 2009, I’m Jon Masters with a summary of today’s LKML traffic.

In today’s issue: Kmemleak, Mudflap, and OOM.

Kmemleak. Catalin Marinas posted a number of kmemleak fixes aimed at reducing the false positive leakage warnings down to near-zero (”making it usable”). The latest patches include support for large block allocations in SLUB and correct handling of bootmem allocations. Catalin also noted a potential leak in alloc_pid, as evidenced in a lost pid structure for an Xorg server started by gdm on Catalin’s system, and reported in a separate thread.

Mudflap. Janboe Ye posted an RFC patch implementing GCC mudflap support for the Linux kernel. GCC Mudflap is a feature of the compiler that will cause an application (linked against libmudflap) to check all “risky” pointer allocations, and aims to reduce all kinds of memory abuses. Although tools such as Kmemleak can already find many cases of memory leakage, kmemleak can’t do this after the SLAB in question has been freed. Janboe’s patch looks interesting, although I haven’t time to dig any further yet.

OOM. Rik van Riel posted a patch aimed at addressing some of the recent OOM situations, as touched upon in yesterday’s podcast. Rik noticed that vmscan can get horribly confused when too many tasks go into direct reclaim, and trigger an OOM situation that is caused because too few pages are on the LRU. Instead, Rik proposes limiting the number of tasks that may enter page reclaim to allow at most half of each inactive list to be isolated at any one time. In yet another semi-related VM thread, Kosaki Motohiro posted version 2 of his patches aimed at helping to track down OOM situations with more information presented to the user in the generated kernel log messages.

In today’s miscellaneous items: Version 3 of the Zero Page re-introduction patches (Kamezawa Hiroyuki) that were so well covered by Linux Weekly News recently, version 7 of the memory barriers for poll and receive network callbacks (Jiri Olsa – recall the original product bug report which lead to finding this generic problem back for version 1 of the patch), a fix for network device carrier detection for devices utilizing phylib (Anton Vorontsov), continued discussion of what makes a “good” tty driver, removal of the old (and quite deprecated) pre-x86 merge x86 MCE code (Andi Kleen), an added export of atomic64 symbols for non-64 arches (Roland Dreier), some networking updates (David Miller – including the memory barrier fixes from Jiri Olsa, and a number of other bug fixes), some ongoing discussion of the “tmem” patches in which arguments were put forward for alternative ways to provide a voltile cache to processes (especially virtual machines) without lots of extraneous copying, and version 11 of the kprobe based event tracer utilizing the ftrace tracing interface (Masami Hiramatsu). On another tracing related note, Robet Richter posted some OProfile cleanups.

In today’s announcements: linux-rt version 2.6.29.6-rt23. Thomas Gleixner announced version 2.6.29.6-rt23 of the linux-rt patchset. It contains some compiler fixes, a powerpc highmem fix, a smp_processor_id fix to hwlat, and some ktime_get speedups. It is also rebased upon 2.6.29.6.

The latest kernel release is 2.6.31-rc2, which was released by Linus on the 4th of July (except it isn’t – there’s a newer version since this time).

Stephen Rothwell posted a linux-next tree for July 9th. Since Wednesday, a new tree entitled “thumb-2″ has been added (presumably ARM related), the tree still fails to build in an allyesconfig build configuration on powerpc, a number of new conflicts appeared, and some build failures also. There are now a total of 132 sub-trees in today’s linux-next composition once more.

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

Audio: http://media.libsyn.com/media/jcm/linux_kernel_podcast_20090708.mp3

I’ve been in the air for 32 hours over the past few days, on a 3 day trip to Japan, apologies for lag, jetlag, and catchup I’m now playing.

For Wednesday, July 8th 2009, I’m Jon Masters with a summary of today’s LKML traffic.

In today’s issue: Modules, OOM, Lockless Ring Buffer, and VFAT.

Modules. In the second module-related item of the week, Siarhei Liakh posted a patch adding support for RODATA protection (via the RO/NX page protection bits) to kernel modules. This affords LKMs (Loadable/Linux Kernel Modules) the same protections as regular kernel code, using available hardware protection mechanisms for enforcement. Arjan van de Ven expressed his approval.

OOM. Another recent OOM cause was isolated in vmscan for systems without swap (or otherwise with more anon pages than file pages). Wu Fengguang says that he has tested the latest “vmscan don’t isolate too many pages” patch and that it solves his OOM, but now the process entering reclaim could sleep long enough to trigger a soft-lockup watchdog instead.

Lockless Ring Buffer. Steven Rostedt posted to let us know about his lockless ringbuffer, which has been discussed before and which is covered in an excellent story this week on Linux Weekly News. Do be sure to take a look at the page swap mechanism for which a patent application has been filed – and be sure to read Steven’s excellent paper/description on the process.

VFAT. Andrew Tridgell posted a summary of the VFAT discussions, and noted a new tree on kernel.org containing his previous proposed ‘workaround’. He isn’t confident that the current patches will make it into the kernel – for many reasons – but is willing to maintain the tree and hopes that he can address some of the technical concerns that have been raised before pushing for inclusion in the kernel once again.

In today’s miscellaneous items: a fix to elfcore.h to enable building UML (Parag Warudkar), some wireless tree updates (John Linville), a patch adding support for the security processor chip and a separate patch adding a driver for the “Restricted Access Region Register” (apparently an ability to restrict access to memory ranges from the CPU) on the Intel MID Platform (Mark Allyn), various cleanups to MAINTAINERS (Joe Perches), some minor powerpc cleanups (Ben Herrenschmidt), some input updates (Dmitry Torokhov), a conversion of ext2 to use the new truncate mechanism (Nick Piggin), a new uevent emmission on block device write protect attribution change, some Power Management updates (Rafael J. Wysocki), and some System 390 updates (Martin Schwidefsky).

Finally today, your author continues to be interested in the memory c-group notification patches, which should make it possible to implement proper memory reservations in the GNU C-library, if it is wired up that way around. I would love to know whether application developers are looking at this – imagine Firefox selectively relinquishing cached web pages from memory on demand.

In today’s announcements: SCST Target driver support for Emulex lpfc FC/FCoE adapters. Vladislav Bolkhovitin posted to announce support for lpfc adapters to be placed in ISCSI initiator and/or target mode using the lpfc SCST driver kit, available on scst.sourceforge.net.

The latest kernel release is 2.6.31-rc2, which was released by Linus over the 4th July weekend.

Stephen Rothwell posted a linux-next tree for July 8th. Since Tuesday, the “sfi” tree was undropped, the tree still fails to build in an allyesconfig build configuration on powerpc, and a number of other conflicts emerged. The total sub-tree count remains steady at 131 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.

Audio: http://media.libsyn.com/media/jcm/linux_kernel_podcast_20090707.mp3

I’m traveling in Tokyo, Japan this week and still playing catchup with LKML.

For Tuesday, July 7th 2009, I’m Jon Masters with a summary of today’s LKML traffic.

In today’s issue: (very MM heavy), Kmemleak, OOM, SLAB, and Transcendent memory.

Kmemleak. Catalin Marinas and others had a number of discussion concerning kmemleak. Amongst the discussion, it seems that the scanning threads can occasionally trigger RCU stall warnings (there are several patches floating around, including one that does some additional cond_resched() calls). Separately, Catalin posted a possible leak in request_firmware.

OOM. Kosaki Motohiro relied upon the latest “OOM analysis helper patches” to track down another potential cause of increased OOM situations in recent kernels. He discovered that the current reclaim logic doesn’t consider concurrent reclaim and might accidentally OOM on a number of systems.

SLAB. David Rientjes posted several iterations of a patch adding an option to disable high order debugging in the SLAB allocator. This effectively deals with situations such as an allocation of exactly PAGE_SIZE, which might result in the allocation being increased to cope with the debugging metadata when debugging was turned on. With David’s patch applied, the idea is to handle these situations and not add debugging information to the SLAB in this case. I’m pretty convinced that similar discussion in the past resulted in a net-dislike for selective debug, but maybe it will be different now.

Transcendent memory. It really does seem to be a VM heavy day today. Dan Magenheimer posted version 2 of his “transcendent memory” or ‘tmem’ patches. These are designed to allow Linux to make use of memory that is of an unknown and dynamically variable size – for example in virtual machine environments where the host provides some memory it may pull at any moment, but which may be helpful temporarily for use as a cache, or similar construct. The patch comes with a detailed summary, though I haven’t read all of it yet.

In today’s miscellaneous items: Version 10 of theioeventfd (formerly iosignalfd) patches (Gregory Haskins), a removal of an extraneous second access_ok() check on reads to /dev/zero (Nikanth Karthikesan), some ksym_tracer fixes (Li Zefan), a fix to resume from suspend when CONFIC_CC_STACKPROTECTOR is enabled (Peter Chubb), a reported regression (apparently since .30 and not recalled in 2.6.29) in leaking of key presses on switching [correction: "virtual terminals" - oops] virtual consoles (Andi Kleen, no doubt Alan will look at this during his pending cleanups – he did post some patches earlier, but they were mostly trivial cleanups), some more trivial cleanups from Jaswinder Singh Rajput, version 4 of a patch adding support for 1GB pages to KVM (Joerg Roedel), a P6 PMU patch that adds support for these older Pentium III systems to the performance counters work that Peter, Thomas, and Ingo are working on (Vince Weaver), a reworked version of memory usage limit notification for memcg (Vladislav Buzov) – which does look pretty cool, and version 2 of the smells-like-ACPI SFI (Simple Firmware Interface) support (Len Brown).

Finally today: It was going to happen sooner or later. Joao Correia posted a patch series increasing the lockdep limits in an effort to reduce the number of false positive warnings that are being generated.

The latest kernel release is: 2.6.31-rc2, which was released by Linus on the 4th of July.

Stephen Rothwell posted a linux-next tree for July 7th. Since Friday, the sfi tree has been dropped (but there was a dialog about this), the tree still fails to build in an allyesconfig build configuration (Stephen notes that this is due to a final link problem), the block tree gained a build failure for which Stephen applied a patch, and the suspend tree lost its conflict. The total sub-tree count in the compose remains steady today at 131 trees.

That’s a summary of today’s LKML traffic. For further information visit kernel.org. I’m Jon Masters.

Audio: http://media.libsyn.com/media/jcm/linux_kernel_podcast_20090706.mp3

I’m traveling in Tokyo, Japan, this week and surviving on a few hours sleep here and there so apologies that we’re still not quite up to date.

For Mon, July 6th 2009, I’m Jon Masters with a summary of today’s LKML traffic.

In today’s issue: DRBD, file truncations, modules, and pipes.

DRBD. Philipp Reisner posted a series of 16 patches implementing the latest iteration of his ‘drbd’ block device for High Availability clustering. He says “as the first bit of the DRBD patch already got upstream [...] it is time to get more of DRBD towards mainline”. Others, of course, may or may not agree.

File truncations. Amerigo Wang pointed out that, when the suid bit is set on a file, and a non-owner user having write permission attempts to truncate (and not append) the file, they get an EPERM error rather than the correct access to the file. Amerigo has a trivial patch that fixes this behavior to allow the truncation in this situation.

Modules. Sorry I missed this one before. Jan Beulich and Michal Marek discussed the recent patches to cleanup module CRC handling on 64-bit arches. Since they actually only need to generate 32-bit quantities, the use of 64-bit slots to store those quantities is corrected, and this now apparently breaks the dump-modversions feature of modprobe. Jan expressed some concern that userspace was reliant upon this not changing, but Rusty (and myself too) believe that this is a reasonable situation. I’ll post a fix when they figure out how they want to proceed with the CRC patches to the kernel.

Pipes. Ongoing discussion of extending pipe() to support NULL arguments is continuing with debate over position handling in open pipes. Changli Gao pointed out that pipe doesn’t refer to the “pos” when reading or writing, and directs those interested in discussing his patchset to look at the implementation of pipe_read() and pipe_write() first. Separately, Nick Piggin posted a new iteration of his patches for truncate operations, which adds a new ftruncate operation for inodes. Also in other news, Linda Walsh wondered aloud why 2GB read()s didn’t work on pipes, which Eric Dumazet followed up with a pointer to the fixed 2GB per-read limit. This exists even on x86_64 systems and might be a candidate for being dynamic.

In today’s miscellaneous items: A series of SPARC asm-generic conversion patches (Fujta Tomonori), bootmem kmemleak tracking (Catalin Marinas, which should help reduce the false positive warnings even further), a vote for making .governor() always be called with rwsem held in the cpufreq cleanups (Thomas Renninger), some Microblaze architectural cleanups from Michal Simek, some PCI fixes (Jesse Barnes, who says there’s “nothing too exiciting here”), and version 9 of the iosignalfd patch series (Gregory Haskins).

Finally today: We’re still seeing a few extraneous lockdep warnings (although one of those might go away thanks to a SLAB lockdep annotation fix posted from Pekka J Enberg) and crash reports on 2.6.31-rc2 (including one case of Xorg giving a blank screen in rc2, and instability in the tty layer, which Alan Cox is helping correct), so it’s worth putting in some testing, if you can.

In today’s announcements: Rafael J. Wysocki obliged everyone with another round of reported regression summaries for kernels 2.6.31-rc2. A total of 21 unresolved regressions are currently listed, and each has been accompanied with a separate email pinging the owner of the bug in kernel.org bugzilla.

The latest kernel release is: 2.6.31-rc2, which was released by Linus on the 4th of July.

Stephen Rothwell posted a linux-next tree for July 6th. Since Friday, the bkl-removal tree has been removed (since it has served its purpose), the sfi tree is still being dropped dur to build problems, the tree continues to fail to build in an allyesconfig build configuration on powerpc, and several other net gains in conflicts were recorded. With the removal of the bkl bits, the linux-next tree composition is down to 131 trees in today’s compose.

That’s a summary of today’s LKML traffic. For further information visit kernel.org. I’m Jon Masters.

Audio: http://media.libsyn.com/media/jcm/linux_kernel_podcast_20090705.mp3

I’m traveling in Tokyo, Japan between Monday and Friday this week (with a jolly 32 hours of flights in between) so am podcasting when I can. Today’s update gets us most of the way back toward being up to date.

From Tokyo, Japan, for the 4th July 2009 weekend, I’m a semi-jetlagged Jon Masters with with a summary of the weekend’s LKML traffic.

In today’s issue: Futexes, IO-MMU, Kmemleak, MM, OOM, VFAT.

Futexes. Mike Frysinger pondered aloud about futex support for architectures devoid of atomic instructions, such as is the case for the Blackfin port. He noted that existing atomic operations are implemented on Blackfin using an obvious interrupt disable/enable cycle, which could be used within futex_atomic_op_inuser() as is apparently the case for the SuperH port (via their asm/futex-irq.h header. Mike wondered why this couldn’t just be moved generically into the top level futex header, making futexes available to architectures such as Blackfin and SuperH by default.

IO-MMU and Trusted Boot. Joseph Cihula pointed out that systems using Intel TXT require that the IO-MMU be enabled in order to perform DMA operations. But the user is able to control this via kernel command line options, which can break TXT booting. Joseph had posted a patch to always force IO-MMU use in the case of TXT-enabled boot, but this was NACKed by a several others on the grounds that it intentionally goes against documented options and changes expected behavior. The current suggestion is simply to panic if the kernel is presented with incompatible TXT and IO-MMU options at boottime.

Kmemleak. Ingo Molnar followed up to an email from Catalin Marinas from last week (in which Catalin had sought advice and suggested that exiting a task with locks still held might be acceptable in the case of piped reads from the kmemleak interface, and that lockdep should not be generating a warning). Ingo understood the point Catalin was making, but re-asserted that “holding locks in user-space is almost always bad”, and that “I’ve yet to see a valid ‘need to hold this kernel lock in user-space’ case, and this does not seem to be such a case either.” Also on the subject of kmemleak, the background scanning process is now nice +10 by default, following the acceptance of another patch that Ingo Molnar signed off on at the same time. Separately Catalin posted a number of kmemleak fixes that are presumably intended for -rc3.

mm. Andrew Morton’s recent mm-of-the-moment for 2009-06-30-12-50 was dying early in boot for Valdis Kletnieks, who tracked it down to a miss-use of wake_up_interruptible in some recent reworks from Oleg Nesterov. Oleg followed up to say that he should have a fix available shortly, especially now that he is able to use the special __wake_up_parent() as an exported function.

OOM. Following up to recent discussion concerning noswap related patches triggering excessive OOM kill scenarios, and simply in reaction to the general mess that is trying to figure out exactly why an OOM occured, Kosaki Motohiro posted a “OOM analysis helper” patch series which adds a number of statistics to the output produced by the kernel on an OOM condition.

VFAT. Discussion surrounding the dual use of long and short filenames continues, with various members of the community chiming in. In part of the latest discussion, Andrew Tridgell and Jan debated a potential regression in the creation of short file names unreadable to certain devices with the latest patch series applied. Andrew suggested that the answer was actually to change the default behavior for 8.3 short names to use a Windows NT compatibility option, but that this would be a potentially incompatible behavior change requiring some further thought. Either way, the VFAT issue is still a hot one.

In today’s miscellaneous items: IO-MMU fixes for AMD (Joerg Roedel), a continued debate over extending pipe() to support NULL arguments (Changli Gao), a poll_wait call and receive callback race fix (Jiri Olsa), some late breaking parisc updates for 2.6.31 (Kyle McMartin – who notes that these were delayed as a few were being tracked down), questions about whether Core i7 frequency is working correctly (Felix von Leitner – Arjan van de Ven contends
that the person concerned should check with the output of PowerTOP to see what is actually happening), version 8 of the run-time PM framework patch for run-time power management of IO devices (Rafeal J. Wysocki), version 9 of the irqfd patch series (Gregory Haskins – fixing races, restoring DEASSIGN support, and a couple of other bugs), another version of x2APIC without interrupt remapping for KVM use (Gleb Natapov – the patch doesn’t note the previous one, but I assume it’s simply a new iteration), Firewire updates from Stefan Richter (and on that subject, Jun Koi also posted, asking where firedump and fireproxy debugging utilities had gone), even more Minix V3 related stuff from Doug Graham, and a followup from Jeff Arnold of Ksplice, Inc. clarifying some miss-conceptions about the true power of Ksplice. In this author’s opinion, too many people are jumping to conclusions concerning what Ksplice can and cannot do without reading Jeff’s excellent paper and actually trying it out for themselves.

Finally today, Daniel Mierswa posted concerning kernel command line escaping of quotation marks. It seems someone has finally gotten annoyed enough with current quote limitations to do something about it. I would finish with a discussion of the Voyager baiting over the past week, but LWN already did a pretty good job with that discussion summary.

In today’s announcements: Linux 2.6.31-rc2 (more on that in a moment). Robert P. J. Day posted announcing that he is writing a new column for Linux.com on kernel development, which is aimed mostly at newer kernel developers

The latest kernel release is 2.6.31-rc2, which was born on the 4th of July. While he was scared by the i915 updates, Linus seems otherwise fairly happy, although he does note that he will be lot stricter about patches for rc3, and he remains very concerned that folks actually report bugs, including regressions as often as they notice them. There are some concerns already with regard to various lockdep limits being exceeded in this latest kernel. Roland Dreier also noted an ext4 related lockdep problem with jbd2_handle, which Ted T’so thinks is a valid bug that needs some looking into.

Stephen Rothwell posted a linux-next tree for July 3rd. Since Thursday, the sfi tree is still dropped, the tree still fails to build in an allyesconfig powerpc build configuration, and the xfs tree gained a build failure so the version from the previous day was used instead. The total subtree count remains steady at 132 trees in today’s linux-next compose.

That’s a summary of today’s LKML traffic. For further information visit kernel.org. I’m Jon Masters.

Audio: http://media.libsyn.com/media/jcm/linux_kernel_podcast_20090702.mp3

For Thursday, July 2nd 2009, I’m Jon Masters with a summary of today’s LKML traffic.

In today’s issue: cgroups, kmemleak, OOM, VFAT, and the return of the Zero Page.

Cgroups. Paul Menage posted an RFC patch series intended to add Hierarchy Extensions to Cgroups. With the patch series applied, one gets named cgroup hierarchies, cgroup hieracrchies with no bound subsystems, and cgroup subsystems that be bound to multiple hierarchies. A number of example use cases are included in the patch series which contains a total of 9 patches.

Kmemleak. In an effort to reduce false positive warnings, Catalin Marinas posted a patch that would better handle objects allocated during a kmemleak scan. With the patch applied, kmemleak will check to see if an allocation happened after it began scanning a list. If so, it will re-scan the list again and repeat the allocation test before reporting any problems. If the system is simply too busy to never scan the list without it changing kmemleak gives up after a certain number of passes (25 in the posted patch). With these patches applied (and perhaps some others also – it was not specified which), Catalin is finding a lot of reports of iwlwifi leaking memory and is not sure whether these reports are still noise, or a legimate bug that needs some attention. Catalin adds “I’m not familiar with this code so any help is greatly appreciated”. Perhaps someone will help take a look at that driver.

OOM. Minchan Kim followed up to the ongoing debate about why exactly a specific patch (intended to affect only non-swap machines) had caused so many OOM-type of situation, with a theory that the patch actually improved performance of page reclaim to the point where the specific tests being used would subsequently expose the system to a fork bomb. Minchan contended that David Howells had merely been “lucky” in his previous use of an unnecessary routine and informed us that Rik van Riel is currently working on a throttling version of page reclaim which should help.

VFAT. Discussion continues about the VFAT implementation. James Bottomley and Alan Cox had a debate concerning the ways in which vendors do and do not carry patches out-of-tree. Alan’s point was largely that vendors always carry patches regardless of the wishes of the kernel community, so one more isn’t a big deal, whereas James counter-argued that this went against the general notion that the kernel community was against long-lived out-of-tree bits. James defended the involvement of the Linux Foundation against what he described as “conspiracy theories” of anything more sinister going on. Later, Jan Engelhardt, Ted T’so, and Andrew Tridgell had a dialog concerning a number of devices Jan had found that broken when given filesystems modified using the various VFAT patches currently going around.

Zero Page (again). Kamezawa Hiroyu posted concerning the removal (in 2.6.24, back in October of 2007) of zero page support (essentially allowing things like sparse unbacked array allocations in userspace that are otherwise contiguous), noting that many customers and users of his haven’t really noticed that it was removed because they’re still using kernels like the 2.6.18 kernel in Red Hat Enterprise Linux 5. Recently, he has also seen an uptick in intentional users of zero page (to which Avi Kivity later added KVM in terms of its migration process) and so suggests a re-implementation that fixes many of the reference-counting and other problems from the old one.

In today’s miscellaneous items: Alek Du posted version 2 of his GPIO driver for Intel Moorestown, Catalin Marinas experienced a problem fixing an issue with kmemleak in which piped reads from the kmemleak debugfs interface would result in a lock being held on return to userspace, fread later, but obviously generating a warning from the lockdep. Gregory Haskins posted version 9 of his irqfd patch series, including fixes and rebasing on a more recent KVM. Justin P. Mattock posted a bunch of SELinux updates (some labeled “non-trivial”, some simply typos and things of that nature), Vivek Goyal posted version 6 of his IO Scheduler based IO Controller patches (including mostly split out patches and some fixes), Doug Graham posted (with a Nortel address), a V3 Minix filesystem fix for big endian systems (or those with >64K inodes) – turns out someone does still use Minix – and several other generic fixes, Venkatesh Pallipadi posted some fixes to Dave Jones for cpufreq lockdep warnings, James Bottomley posted a small number of SCSI fixes against 2.6.31-rc1, Jonathan Cameron posted version 4 of his Industrial I/O Subsystem patches (which seem to include yet another ring buffer implementation? I didn’t check yet), Alan Cox followed up to Lennart Poettering’s previous VT_WAITACTIVE patch with a more generic event interface for Virtual Terminals, and Chris Mason posted a series of btrfs updates (mostly small bug fixes, but also a first step toward snapshot deletion from Yan Zheng), intended to still make it into 2.6.31-rc2.

Finally today. Kumar Gala posted asking Alan Cox if he had some good examples of users of the tty later to use as a reference in bringing an out-of-tree serial driver for the Avocent ESP-16 MI Serial Hubs (serial over Ethernet) up to scratch for mainline inclusion.

In today’s announcements: Karel Zak posted RC2 of util-linux-ng v2.16. It includes the moving of the libuuid library from e2fsprogs into util-linux-ng. Kay Sievers noted that it fails to build in a clean chroot due to an install hook hack that moves some files around during the build. And Jaswinder Singh Rajput helpfully mailed a number of people concerning various feature removal dates that were previously committed to and have lapsed or are pending. At least one of these reminders has resulted in the related feature being killed by a subsequent patch, and will hopefully lead others to contemplate likewise.

Ryo Tsuruta announced the IO Controller Mini-Summit in Japan in October, which will immediately precede the Linux Kernel Summit. This is an event that was hinted at previously, although details remain “sketchy” and it’s not entirely sure who will be there at this time.

The latest kernel release is 2.6.31-rc1, which was released by Linus last week.

Andrew Morton posted an mm-of-the-moment for 2009-07-02-19-57. Meanwhile, various users of the previous mm-of-the-moment have reported a few glitches.

Greg Kroah-Hartman announced releases 2.6.27.26, 2.6.29.6, and 2.6.30.1 of the kernel. He strongly encourages users to upgrade, and notes that the .29 update will be the last so users should migrate to 2.6.30 as soon as possible.

Stephen Rothwell posted a linux-next tree for July 2nd. Since Wednesday, the “sfi” tree is still dropped, the tree still hates powerpc allyesconfig, and several conflicts went away also. The total sub-tree count in the latest compose stands at a respectible 132 trees.

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

Audio: http://media.libsyn.com/media/jcm/linux_kernel_podcast_20090701.mp3

For Canada Day (Wednesday, July 1st 2009), I’m Jon Masters with a summary of today’s LKML traffic.

In today’s issue: Encrypting the page cache, VFAT, and XFS.

Encrypting the page cache. There is an RFC floating around from Jeremy Maitin-Shepard for TuxOnIce (the out-of-tree alternative suspend code) concerning encrypting the page cache in RAM on suspend-to-RAM so that the system cannot be cold booted into another kernel and its memory content analyzed. Such an attack is not too far fetched – research over the past few years has shown that it is not only feasible but is also actively being used. The only real concern surrounding this seems to be the overhead involved, though it is likely to be a configurable option, if Jeremy follows up to the RFC with some patches.

VFAT. Ongoing discussion of the VFAT patch proposed by Andrew Tridgell seems to have headed toward changes to VFAT support possibly necessitating that the additional configuration options be part of a new (perhaps aptly named) alternative VFAT filesystem. Posters have pointed out that some windows systems choke when working with these patches and we probably don’t want those systems to bluescreen when reading from disks written to using Linux systems. As an aside, this author is always amused by the configurable nature of the Windows bluescreen – I recall a much younger version of myself in college winding up Microsoft support one day by inquiring about the registry codes to set the bluescreen background colour. Childish for sure, but I recall it being quite fun at the time – there are sixteen glorious color options!

XFS. Christoph Hellwig posted an updated about XFS support, in which he noted that the 2.6.30 kernel had incorporated fixes for ENOSPC handling, and had additionally shrunk by 500 lines in the latest release. He notes that in the current merge window, quotaops went away (which simplifies quotas), and XFS dropped its own POSIX ACL implementation in favor of the generic in-kernel one. This author recently noticed Andreas updated his acl git tree with a note that the official location for those bits is now Savannaih, as Christoph mentions in his summary also. Seems like an exciting time for XFS again.

Miscellaneous fixes include: fixes to round_up/down from H Peter Anvin, some minor Super-H updates from Paul Mundt, some race fixes for irqfd/eventfd from Gregory Haskins, some fixes for FUSE (Miklos Szeredi) that include a couple of minor features which might not be allowed in rc2, cache events in perf from Jaswinder Singh Rajput (kudos to you, jaswinder, for your continued efforts to make cleanups and fixes to the kernel – it doesn’t all go unnoticed either), a per-GPIO sysfs symlinking naming patch from Jani Nikula, a scheduling while atomic bug introduced by kmemleak is fixed by a patch from Ingo Molnar, a disabling of CLONE_PARENT for the init task from Sukadev Bhattiprolu, a large number of networking bits from David Miller (including an important SCTP fix), some md updates from Neil Brown (implementing mostly the new ‘topology’ numbers), version 6 of the Intel Trusted Execution (Boot) Technology covered several times previously in this podcast, and some block bits for 2.6.31-rc2 from Jens Axboe (incuding a removal of __GFP_NOFAIL misuse in CFQ that was debated a little over last week). Steven Rostedt noticed that the gcov patches don’t protect against modification to vsyscall memory (as his ftrace patches do) so fail to boot on his test system, falling over inside the initrd init.

This podcast is now two months old. You’ll notice we’re a little behind again this week (it won’t get any easier next week as your author is in Japan only from Monday to Friday and will be fighting severe jetlag over the next two weekends in addition to being on planes for 32 hours and trying to write a chapter on the plane). It’s tricky to do a podcast every day, but I enjoy it, so I plan instead to continue to make a best effort – and occasionally live with being a few days out of sync. If you would like to see a guaranteed daily service, feel free to volunteer to help do that. Meanwhile, I’ll keep the podcasts coming, you just keep listening and sending me those nice mails. And if you happen to be in Tokyo on an evening next week, let’s have some coffee!

In today’s announcements: LTTng 0.142. Mathieu Desnoyers announced version 0.142 of LTTng. This includes a fix to text poke he posted about previously. Also, the Linux Test Project has been released for June 2009. Subrata Modak posted to let us know that 6 testcases have been added covering 5 new system calls, major syscall and powermanagement test fixes have been added, and a number of additional test fixes have been added. The healthy number of contributors to June 2009 LTP include the ever-involved Darren Hart and a large number of others also. And this is certainly a worthy project to help out on if you’ve got some spare cycles these days.

The latest kernel release is 2.6.31-rc1, which was released by Linus last week.

Greg Kroah-Hartman is preparing new .27, .29, and .30 stable releases for which he sent out 30, 35 and a whopping 108 proposed patches respectively for review on Tuesday evening ahead of a deadline on Friday morning.

Stephen Rothwell posted a linux-next tree for July 1st. He dropped the new “sfi” tree due to build problems, powerpc still fails allyesconfig, and the tree overall gained conflicts over the previous day. The total sub-tree count is now up to 132 trees in the current linux-next compose.

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