Linux in the news
All in one big page
See also: last week's Kernel page.
The current development kernel release is 2.3.48. This large (4.5MB) patch is mostly dedicated to a major merge of MIPS and MIPS64 support; it now can run on a wide series of SGI systems, MIPS-based DECstations, and even a Russian embedded system called the "Baget." It also has an experimental 3ware hardware ATA RAID driver, PS/2 ESDI disk support, Matrox I2C support, a driver for FORE Systems 200E ATM adapters, and a rework of the Tulip ethernet driver.
There is a 2.3.49 pre-patch available. It includes more work on interrupt handling; a new bonded network driver; USB enhancements for IBM cameras, Pegasus ethernet, and Wacom tablets; a rework of the Reva frame buffer driver; and a new "big reader" spinlock mechanism.
Those interested in the interrupt work may want to look at this rather technical message from Ingo Molnar on what has been done. Essentially, the work is aimed at increased performance and scalability, especially on multiprocessor systems.
The current stable kernel release is still 2.2.14. The 2.2.15 pre-patch is up to version 2.2.15pre12. This version of the patch is intended to lead to the final cleanup; Alan intends a pre13 toward the end of the week, and the real 2.2.15 kernel sometime next week, "if things go OK."
Announcing: the Linux Console Project. James Simmons sent out this announcement for a new project which seeks "to design a new multihead console system for linux." There had been some discussion in linux-kernel about the status of the current console driver (which appears to be unmaintained) and the desirability of some new features - like different video modes for each virtual terminal. The project already has a shiny new SourceForge page; Mr. Simmons would like to have something ready to merge into the 2.5 development series at an early date.
Devfs, /proc, and magic files. Back in 2.3.46, a new driver (by Tigran Aivazian) showed up which provides the (somewhat scary) ability to rewrite the microcode in Intel P6 processors. If you want to mess with the internal wiring of your processor, just load the appropriate module (wiring this driver into a monolithic kernel would be a risky thing to do) and shovel your new microcode into /proc/driver/microcode. All there is to it.
This driver drew a number of immediate complaints. The ability to hose your processor doesn't seem to bother people; the real sticking point is the creation of another magic file in /proc. What is coming out is that quite a few people really do not like the proc filesystem at all, and do not wish to see more capabilities added to it.
What are the problems with /proc? Common gripes include: it has gone past its initial scope of providing process information; it is bloated and slow; the files and their formats are not standardized and subject to frequent change; the internal API is convoluted; the use of ASCII strings requires applications to do too much parsing; changes to permissions on /proc entries are not persistent; and so on. /proc's detractors are especially annoyed that it is increasingly hard to run a system without /proc. Once upon a time, /proc was supposed to be an optional feature.
One suggestion that has been made is that the microcode update driver should move to devfs instead, now that devfs is in the kernel. At that point, it would be necessary to use devfs to make use of this driver. That has upset a number of people; you see, devfs is supposed to be an entirely optional feature. Devfs author Richard Gooch has always stepped very carefully around that issue until now; his more recent words on the subject have drawn some criticism.
This issue will be difficult to resolve in a way that satisfies everybody. Anything that makes devfs mandatory for system operation will draw fierce opposition. While /proc is a bit messier than it should be, changing it will prove difficult - many applications depend on its current form. Nonetheless, some sort of more coherent approach to the "magic files" interface to the kernel will become necessary at some point. That sort of interface is just to convenient to get rid of. Expect to see this fight come back again.
Other patches and updates released this week include:
Section Editor: Jonathan Corbet
March 2, 2000