Kernel Log: First release candidate for Linux 2.6.39
Torvalds said that he considers 2.6.39 more of a solid version with a boring amount of progress – however, with ipset, the Xen network backend, as well as many new and improved drivers, the new kernel, which is expected to be released in late May or early June, does offer numerous improvements which matter to end users.
Fifteen days after releasing Linux 2.6.38, Linus Torvalds has closed the merge window of Linux 2.6.39, and completed the integration of this version's major changes, by issuing 2.6.39-rc1. At the end of his release email, Torvalds said "But on the whole I think this should be one of those 'solid, boring progress' releases."; however, his enthusiasm about the auto-grouping of processes and the VFS optimisations in 2.6.38 had been unusually strong.
There are actually changes galore as usual – this is also shown in the diffstat, which states that the kernel hackers already added, removed or changed more lines of code for the first release candidate of Linux 2.6.39 than they did for 2.6.38:
- 2.6.38: 9158 files changed, 748179 insertions(+), 455943 deletions(-)
- 2.6.39-rc1: 8995 files changed, 828740 insertions(+), 509303 deletions(-)
Among the reasons for this is the added support of the Unicore32 architecture that is used in various processors mainly sold in China. Also new are the Xen backend drivers for network communication, ipset support, the media controller subsystem, a platform driver for Samsung notebooks, a rudimentary KMS driver for Radeon HD 6900 graphics cards and batched discard functionality for the still experimental Btrfs. The Realtime developers submitted some patches that allow the kernel to process almost all threaded interrupts; in this context, Thomas Gleixner also did some major restructuring and cleaning up in the IRQ handling code of the various architectures supported by Linux. As expected, the last remaining components that use the Big Kernel Lock (BKL) have now been removed, allowing the kernel hackers to completely eradicate the lock via a commit entitled "BKL:That's all, folks".
The Kernel Log will comprehensively discuss these and numerous other new features in a series of "Coming in 2.6.39" articles over the next few weeks. Torvalds will probably release the next kernel version in late May or early June, if he and his fellow developers keep to their usual routine.
Several discussions on the LKML during the main development phase of 2.6.39 offer insights into various aspects of the kernel development process. For instance, Torvalds writes in his release email for RC1 that the kernel "at one point ate XFS file systems for breakfast, but that's fixed and it's hopefully all good now". This demonstrates one of the risks people take when testing a kernel version's release candidates – however, such serious issues only surface occasionally. They are typically fixed quickly to avoid scaring off the already small community of testers; in the Kernel Log author's experience, the danger of losing data is hardly any greater after RC2 than it is when manually installing a finished kernel version.
How Torvalds occasionally steps in to ensure sufficient attention is given to detail can be seen in his response to a Git-Pull request submitted by Rafael J. Wysocki. The request included patches that restructure the configuration options of the kernel's power management features. However, Torvalds didn't like various aspects: To ask users whether they want the hibernation interface is idiotic, he said, and added that the interface doesn't, and shouldn't, matter to users – they simply want hibernation support.