Kernel 2.6.8.1-mm4: ptrace: "bad: scheduling while atomic!" followed by interrupt-disabling kernel panic

Kernel 2.6.8.1-mm4: ptrace: "bad: scheduling while atomic!" followed by interrupt-disabling kernel panic

[Joshua Wise]

Hi folks,

Recently I took the plunge and upgraded my desktop system (an ia32 
Athlon) from 2.4.x to 2.6.8.1-mm4. However, when I attempt to do a 
ptrace on a binary, my kernel spews a few loads of bad: scheduling while 
atomic!, and dies.

Initially, I watched at a console like so:

bluefire ~ # strace ls
execve("/bin/ls", ["ls"], [/* 42 vars */]) = 0
--- SIGSEGV (Segmentation fault) @ 0 (0) ---
Kernel panic - not syncing: Aiee, killing interrupt handler!

Not seeing the Oops or BUG that triggered it, I decided to try again, 
first doing a sysrq-[loglevel]8 to increase verbosity. Only having a pad 
of paper to note debugging output, I have clipped seemingly irrelevant 
bits, as notated by [...]. If you need more detail, please email me. 
Also, this kernel was not compiled with addr2line symbols. If you need 
line numbers, I can recompile it like that for you.

bluefire ~ # strace ls
bad: scheduling while atomic
  [<c01070ee>] dump_stack+0x1e/0x20
  [<c0385500>] schedule+0x4f0/0x50
  [<c0129531>] ptrace_notify_info+0x91/0xf0
  [<c01291b8>] get_signal_to_deliver[...]
  [...]
execve([...]) = 0
bad: scheduling while atomic!
  [...] dump_stack[...]
  [...] schedule[...]
  [...] ptrace_notify_info[...]
  [...]
--- SIGSEGV [...] ---
bad: scheduling while atomic!
  [...] dump_stack[...]
  [...] schedule[...]
  [...] sys_sched_yield[...]
  [...] coredump_wait[...]
  [...] do_coredump[...]
  [...] get_signal_to_deliver[...]
  [...]
Kernel panic - not syncing: Aiee, killing interrupt handler!

Does anyone have any insight?

Thanks,
joshua