See also: last week's Kernel page.

Kernel development

For those of you eager to stay on the bleeding edge, three new 2.4.0 prepatches were released this week, including 2.4.0test3-pre2, 2.4.0test3-pre3 and 2.4.0test-pre4. Both pre3 and pre4 were released today, pre4 within two hours of pre3.

We've produced basic listings of all the modified files in each patch:

• test3-pre2 (263 files)
• test3-pre3 (698 files)
• test3-pre4 (699 files)

The prepatches include architecture-specific fixes for just about every platform, including the S/390. They also include large numbers of driver updates, some fixes/improvements for various file systems and even some pokes at the memory management system.

Alan Cox has also released 2.4.0test2-ac2, with his patches against 2.4.0test2.

The current stable kernel release is still 2.2.16. Alan Cox's prepatches for 2.2.17 are up to 2.2.17pre10. Since security issues forced him to release 2.2.16 earlier than he might have liked, his priority for 2.2.17 is to gain a high level of stability and reliability. That means just important bug fixes are going in at this time.

Low-latency Linux vs RT-Linux. On Wednesday, June 28th, Paul Barton-Davis and a large group of other programmers sent a note to Linus Torvalds asking him to include Ingo's low latency patches, currently only available for the 2.2.X tree, in the mainstream Linux development tree. They explain in detail why the current latency of Linux is causing severe problems for programmers working in the audio+MIDI. Ingo's patches have helped them tremendously, but they do not believe they can get commercial support from large vendors for applications in this area if the support is not integrated into the main kernel.

This spawned an incredibly long, and sometimes acrimonious discussion on linux-kernel. One key point hammered on frequently was the question of why this same group could not use RTLinux for their needs. We won't try to rehash the arguments. In the long-run, the answer from Linus was "no", but a sympathetic one. His objection to Ingo's patches are not because he does not want to see latency in the mainstream kernel improved. He did not expect audio-MIDI developers to use RTLinux; in fact, he deprecated such work, since their needs are not truly for hard real-time. His concern is simply that Ingo's patch is currently too ugly (an opinion shared by Ingo as well). Ingo is now working on porting his patches to the 2.4.X series, where they can undergo more scrutiny and possibly some pieces of them may actually get incorporated, where deemed appropriate.

In the end, Paul Barton-Davis seems to believe that he, and the other programmers who joined in the original proposal, got what they needed. "The task ahead of those who want 1-5ms scheduling latencies is clearly to measure exactly where the problems are, and propose specific changes starting with the worst cases. Hopefully, as the summer (US) progresses, we'll get to that, and then by fall/winter (US), it will be possible to see just how far the acceptance rate of such changes has addressed the issue."

One interesting suggestion brought up was the possibility of creating a test for for drivers, similar to the "slab poisoning" test, to turn up poorly written or poorly designed device drivers that may be, driver by driver, adding to the latency problems in the kernel.

In the long-run, the two goals being examined are both important: keeping the kernel free of ugly kludges and keeping latency in the kernel as low as possible, without making hard real-time guarantees that are inappropriate for a general-purpose operating system. Unfortunately, more time will be required to meet both of those goals at once.

RTLinux availability. In case RT-Linux is something you've been considering, you may want to note that LinuxPPC and Yellow Dog Linux will be including RTLinux 3.0 in their distributions soon, and for x86, the RTLinux folks offer a Red Hat ISO image with the RTLinux kernel and modules "ready to go". Of course, people also have the option of RTLinux's competitor, the feature rich Real Time Application Interface (RTAI). Enjoy!

Other patches and updates released this week include:

• Ulrich Windl announced PPSkit-0.9.3. He received no comments from his pre-release last week, so this is a full release. PPSkit for 2.2.16 is a "general kernel patch to implement nanoseconds as time resolution".

• Version 1.3 of Mikael Pettersson's x86 performance-monitoring counters driver has been released.

• Eric Raymond has released both CML2 version 0.6.0 and version 0.6.1 this week. CML2 is a python-based, kernel configuration system.

• Richard Gooch has released devfs v99.17 (for the 2.2.x kernel series) and devfs v175against 2.4.0-test3-pre2.

• In response to the latency discussion, Roger Larsson released a version of his latency profiling patchagainst the latest kernel.

• Reiserfs 3.6.10 has been released. The new version has been ported to linux-2.4.0test2-ac2 and with some minor bug fixes.

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