[PATCH-v2] kdump: Enable kdump if 2nd-kernel is loaded.

[Hidetoshi Seto]

Eric W. Biederman wrote:
> The Documentation for sysrq-c certainly needs to be updated.

No doubt about it.

> If I am doing development on a system I like oops's.  All of the
> information and nothing goes down.  I can get at /proc/kcore etc.
> 
> In a setting where I can't be Johnny on the spot and look at
> things a core dump is probably the best I can get.  In that scenario
> panic_on_oops sounds good.
> 
> As I read the current check it reads:
> If we are going to panic and not oops || panic_on_oops
> 	kexec_should_crash = true;
> 
> Which seems a very reasonable implementation of policy.

If I understand it,

a) panic
b) oops
c) SysRq (from keyboard interrupt)
d) echo c > /proc/sysrq-trigger

   a      b      c      d
common, w/o CONFIG_KEXEC:
 panic  oops   panic   oops  :
 panic  panic  panic   panic : w/ panic_on_oops

A) 2.6.30 + CONFIG_KEXEC:
 panic  oops   -       -     : w/o kdump-image
 panic  panic  -       -     : w/o kdump-image w/ panic_on_oops
 kdump  oops   kdump   kdump : w/ kdump-image
 kdump  kdump  kdump   kdump : w/ kdump-image w/ panic_on_oops

B) 2.6.31-rc1 + CONFIG_KEXEC:
 panic  oops   panic   oops  : w/o kdump-image
 panic  panic  panic   panic : w/o kdump-image w/ panic_on_oops
 kdump  oops   kdump   oops  : w/ kdump-image
 kdump  kdump  kdump   kdump : w/ kdump-image w/ panic_on_oops

C) After this patch + CONFIG_KEXEC:
 panic  oops   panic   oops  : w/o kdump-image
 panic  panic  panic   panic : w/o kdump-image w/ panic_on_oops
 kdump  kdump  kdump   kdump : w/ kdump-image
 kdump  kdump  kdump   kdump : w/ kdump-image w/ panic_on_oops

Ohmichi-san pointed "oops on d) with kdump-image" in B) and suggested
it should be "kdump."
And Eric pointed "kdump on b) without panic_on_oops" in C) and requested
it should be "oops."
(Plus, IMHO, "oops on d) without panic_on_oops" should be "panic.") 

... Right?

Then as I mentioned, use panic in sysrq would be one of solutions.
Are there any better fix?


Thanks,
H.Seto