The Daves Collective

Linux 2.6.0 (boast, boast)
2003-12-18 21:53

Released today, installed today! Yes, we are at home to Mister Smug-face. It's quite usable, perhaps I'll stick with it for now since there's a good-enough workround for each of the issues from when I tried -test11. (Fond memories of when 2.4.0 was released, within 24 hours I had that up and running on a 486DX33 with an ESDI disk, aaah)

d$ cat /proc/version
Linux version 2.6.0 (d@dsds) (gcc version 3.2.3) #1 Thu Dec 18 18:19:29 GMT 2003
d$ 

A first go at Linux 2.6
2003-12-08 12:00

Trying out 2.6.0-test11 using the following references - The post-halloween document and HowTo Upgrade To The 2.6 Kernel. Well it mostly worked: here's the dmesg from booting 2.6 and here's my .config. Visible changes were/are as follows -

1. Mouse acceleration within X much much faster, need to set it much slower
2. Stinking nvidia binary-only module not yet rewritten for 2.6 - an unofficial patch is at http://www.minion.de/ but easier to revert to 'nv' driver
3. Peculiar errors launching X (having reverted to nv driver - wtf??)
4. Migration modutils to module-init-tools - cos I use devfs, need /etc/modprobe.devfs, but not in Slackware 9.1 module-init-tools package - had to get source tarball, though since modprobe.devfs is just a static file in there I didn't need to build from source
5. add new sysfs to fstab
   /sys          /sys             sysfs     defaults                0 0
6. devfs no longer provides devpts filesystem at /dev/pts, now needs the old explicit fstab entry as if devfs not present
   /dev/pts      /dev/pts         devpts    gid=4,mode=620          0 0
7. usbdevfs is now renamed usbfs (although presumably there should be some replacement in /sys)
   /dev/usb      /proc/bus/usb    usbfs     defaults                0 0
8. Migration bdflush to pdflush - "warning: process `update' used the obsolete bdflush system call" - just needed to remove 'update' from rc_sysinit
9. syslogd: "process `syslogd' is using obsolete setsockopt SO_BSDCOMPAT" - no sign of a new or patched version of sysklogd, 1.4.1 now two and a half years old, could try migrating to syslog-ng but it's got previous and the "current stable version" is tagged -rc4. Hmmmmm.
10. 2.6 bzImage is much much bigger than 2.4 for similar config - why?
11. integrated version of i2c replaces /proc/bus/i2c with the new sysfs interface for which new userspace tools lm_sensors2 (in cvs) are needed. (Note the lm_sensors mailing list is here.)
12. the alsa emu10k1 driver in 2.6.0-test11 is quite some way behind recent alsa-driver releases (I upgraded 0.9.6 to 1.0-rc1 last week but the kernel seems to be older than 0.9.6), so it will mean manually reverting all the usual ghastly undocumented unversioned manual changes to the asound.state mixer settings, and then at some time in the future it'll all have to be painfully re-done yet again back to where I am now.

Note that the various fstab changes, the emu10k1 mixer settings and the X mouse acceleration are the only things I've noticed that make dual boot between 2.4 and 2.6 a bit grotty. But the main reason for not permanently going to 2.6 is still the criminally irresponsible great kernel header cockup.

Update - Prime Number Shitting Bear

It has come to my attention that the above list looks like Tux is crapping integers. This of course reminds us of the legendary Prime Number Shitting Bear (or Alkulukuja Paskova Karhu as they say in .fi). I just looked and it's still there! Yay!

For further drivel see the Index