Friday

Most of the day has been used up in rebuilding kernels trying to find out where this deadlock bug occurs

(Strace fragment follows)

munmap(0x20000020000, 8192)             = 0
writev(3, ptrace: umoven: Input/output error 0x11ffff3f0,
1)               = 12
fcntl(3, F_GETFL)                       = 0x2 (flags
O_RDWR|O_LARGEFILE)
fcntl(3, F_SETFL, O_RDWR|O_NONBLOCK)    = 0
read(3, 0x11ffff4a0, 8)                 = -1 EDEADLK
(Resource deadlock avoided)

The major problem being that it affects sending xterms to other machines (evel itself if -display localhost:0) and prevents netscape/mozilla from browsing the web though lynx works.

This bug is just so bizzare. It seems to affect X11 communications, netscape and mozilla and nothing else yet I know this is a kernel bug (old kernel works, new kernel breaks the behavior) I'm hand applying all the changes the i386 kernel team made one at a time until I find the one that knackers the kernel in the above fashion. It's time consuming, and I'll have to come in tomorrow to finish it.