See also: last week's Kernel page.

Kernel development

H. J. Lu continues to work on the NFS code. Bugs have been dying in massive numbers, and the NFS subsystem is beginning to look like it is ready for a stable kernel release. Good NFS is crucial, and H.J. deserves many thanks for his efforts here. He does have a new version out for testing, give it a spin if you can.

Stephen Williams posted an interesting comparison of the Linux and NT device driver interfaces. It is a thorough and fair-minded work. Give it a read and see how two worlds compare.

A last minute thrashup of the network driver interface is in the works. This change comes from discontentment on Linus's part with the use of a couple of the driver flags (dev->tbusy and dev->interrupt). While both of these flags had well-defined uses at one point, actual usage has drifted in multiple directions and consistency is hard to come by. The result is unpleasant driver bugs. (Linus summarized his complaints in this posting).

To fix this problem, both of those flags will disappear. The dev->tbusy flag, however, is still needed (it tells the network queueing system not to try to attempt to transmit further packets through that interface). A new flag will be created to fill this function, but with better-defined semantics. The result will be changes that ripple through a number of drivers, even though the code freeze is in effect. Much more extensive changes are waiting for 2.3.

Some people have noted that context switching times have increased in recent kernels. There are, as it turns out, a couple of reasons for that. First is that Linux has changed from using hardware context switching to a software switch scheme. This change was evidently needed to get around some other problems; see Linus's explanation for some gory details. The second reason is that the aforementioned software context switching scheme had a bug that made things worse. The fix is in 2.1.123; at this point context switching times should be pretty close to what they were back with the hardware scheme.

Streams in the Linux kernel? The question has come up yet again, for the same old reason: Netware uses streams. Most of the rest of the world gave up on streams some time ago, once their inevitably poor performance characteristics became clear. Thus, the lack of streams in Linux has inconvenienced very few people. Streams in future kernels remain unlikely, though some tweaks have been made recently making it easier to add them as a loadable module for those who must have them.

Since we're a weekly publication, chances are we'll be behind a rev or two on the kernel release by the time you read this page. Up-to-the-second information can always be found at LinuxHQ.

Next: Distributions