I started writing this entry the other day, but it was killed due to a#

Sat, 11 Jan 2003 04:43:03 +0000

I started writing this entry the other day, but it was killed due to a kernel problem.

My phone didn't come back, as such. Instead they gave me a new one, so that really ought to fix the problem. Thank you, Botley Road Oxford branch of Carphone Warehouse

All things considered, less done this week than I might have hoped to do. Having got the phone back, I obviously needed to get the numbers back into it - having cut and pasted the first two into cu, I decided to write a short perl program to do the rest automatically. Whee. Flashing caps lock light time. OK, let's do (almost) what I should have done instead and write it in lisp (I say "almost" because it's actually a shell script starting #!/usr/bin/clisp). Flash bang wallop #2. Hmm. Some poking around reveals that the device objects to having close() called, but as the oops is actually a full-blown kernel panic (in sched.c) and even stuff like magic sysrq only just about works, it's not going to be easy to get a log for.

(Some IRC people did recommend a serial console. Well, yes, but . If I had a serial port, I'd have plugged the phone in there in the first place and wouldn't need to be screwing around with USB.)

So I (forward ported to 2.4.21-pre3, wrote debian init scripts for, and) installed LKCD to get some kind of crash dump I could actually use. That seems to fixed things all by itself - at least, now I cannot replicate the problem.

The upside of all this is that 2.4.21-pre3 has some kind of sane memory usage patterns and once again I have a laptop that doesn't thrash itself for tens of seconds at a time on simple operations. My spies tell me that these days -ac kernels are using some variant of the rmap VM, and non-ac kernels are not. So, probably going to blame rmap. Maybe it's badly tuned. I still have LKCD installed for when I do manage to crash the machine using USB serial, but let's upload this diary entry first, shall we?