See also: last week's Kernel page.

Kernel development

The current kernel release is 2.4.8, which was announced by Linus on August 10. It includes much of the recent virtual memory work, many fixes, a new emu10k1 driver (see below), and a bunch of unspecified stuff from the "ac" releases.

On the prepatch front, Linus is, as of this writing, up to 2.4.9pre4; it contains a number of fixes, but, unless you've been bitten by a 2.4.8 bug or have a need to export VFAT filesystems over NFS, there's not too much that's exciting.

Alan Cox's latest is 2.4.8ac5. Included in this patch is a substantial set of IDE driver updates, a number of ReiserFS portability fixes, the integration of the intermezzo distributed filesystem code, the beginnings of the merge of the 64-bit PowerPC code, and a great many other fixes and updates. It's currently missing the virtual memory and emu10k1 code from 2.4.8. According to Alan: "This is a fairly experimental -ac so please treat it with care."

SB Live gets an update. The most controversial change in 2.4.8, perhaps, is the inclusion of a new emu10k1 (SB Live) driver. This patch was put together from the Creative Labs CVS archive by Robert Love. Mr. Love is not the official emu10k1 maintainer - but, then, it's not at all clear who is. The driver packaged with the kernel had not been updated for a year, and had a number of known problems.

Of course, the new driver also has some problems - in some cases breaking things for people who had a working system before. Stable kernels aren't really supposed to break things that way, so there has been a certain level of disgruntlement. Some of the new problems, at least, are easy to fix; those who build the driver as a module should go with the 2.4.9 prepatches or apply this patch. Some of the other problems may require some more serious debugging work.

It probably will not take very long before the inclusion of the new driver looks like a good thing to all involved. The old driver was clearly stagnant and in need of attention; the new one fixes a number of known problems, meaning that it actually works for some people for the first time. Development activity has already picked up, patches to enable more features and fix some problems, enable even more features, and add documentation have found their way onto the list. With luck, long-suffering Linux emu10k1 users will be much happier soon.

Where should patches go? As the "ac" series looks increasingly like the anteroom to the official Linux kernel, some kernel hackers have been known to wonder where their patches should go. Should they go to Alan, to Linus, or to both? Things are not helped by the fact that patches sent to Linus often seem to simply disappear into the void. An acknowledgement (or explicit rejection) from the Great Penguin is rare.

The answer from Linus is pretty straightforward: he wants to see the patches, even if he has silently dropped them in the past:

Re-sending is always the right thing to do. Sometimes it takes a few times, and you can add a small exasperated message at the top by the third time ("Don't you love me any more?").

This system evidently works for Linus, but it does leave some of the other developers wondering just what is going on. A more organized approach to patch management has been advocated many times, but it doesn't ever seem to happen. Linus likes things the way they are.

Logical Volume Manager updates. Heinz J. Mauelshagen has announced the 1.0 release of the Logical Volume Manager (LVM) for Linux. With this release, LVM is said to be stable and production ready. Note that, for those currently running an LVM release older than 0.9.1-beta8, there are some upgrade issues; be sure to read the instructions carefully.

Chris Mason has released a patch which makes LVM snapshots work properly with ReiserFS (and other filesystems as well). There is also an updated version that works with the recent superblock handling changes.

Other patches and updates released this week include:

• Jens Axboe has posted a new version of his patch which enables DMA I/O from high memory without the use of bounce buffers. The patch has since been revised and somewhat mixed up with some work by David Miller supporting 64-bit DMA operations. This work, when integrated (it's aimed at 2.4), should prove most helpful for performance on high-end systems.

• Richard Gooch has released devfs v186 and v187 which fix some bugs and make more drivers work with devfs. There is also a new devfsd release available.

• User-mode Linux 0.45-2.4.8 is out, with a number of bug fixes and some new features (such as the PowerPC port) as well.

• Ulrich Windl has released a version of his PPSkit (nanosecond timekeeping) patch for 2.4.7.

• A spinlock debugging patch was posted by Brent Baccala.

• A new security module patch has been posted by Greg Kroah-Hartman. Also available is a proposed system call interface for security modules.

• Rik van Riel has announced a wiki page for those who would like to discuss or complain about Linux memory management.

• We have an update from the Functionally Overloaded Linux Kernel project. It suffered a bit of a setback in its quest to build the most heavily endowed kernel when ext3 and PPC64 went into the "ac" series, but has fought back fiercely by adding pipe and socket ACLs and the CueCat driver. "Further work on finding obscure work is clearly necessary, though."

Section Editor: Jonathan Corbet

For other kernel news, see: Kernel traffic Kernel Newsflash Kernel Trap 2.5 Status

Other resources: L-K mailing list FAQ Linux-MM Linux Scalability Effort Kernel Newbies Linux Device Drivers

Next: Distributions