Re: [crash] Re: [PATCH 0/4] x86_64: Optimize percpu accesses

[Mike Travis <travis@sgi.com>]

Ingo Molnar wrote:
> ok, i have integrated tip/x86/percpu-zerobased into tip/master briefly, 
> but it blew up almost immediately in testing, on two boxes.
> 
> one bad config is:
> 
>   http://redhat.com/~mingo/misc/config-Mon_Jul_28_17_35_00_CEST_2008.bad
> 
> failure pattern: it booted up fine to userspace and seemed function, but 
> then produced a spontaneous reboot while building a kernel, without any 
> log entries.
> 
> other bad config is:
> 
>   http://redhat.com/~mingo/misc/config-Mon_Jul_28_17_30_39_CEST_2008.bad
> 
> failure pattern: early crash at:
> 
>   PANIC: early exception 0e rip 10:fffffff817dfc1a error 0 cr2 28
> 
> which corresponds to:
> 
> ffffffff817dfc0f <machine_specific_memory_setup>:
> ffffffff817dfc0f:       48 8b 05 aa cf 04 00    mov    315306(%rip),%rax
> # ffffffff8182cbc0 <x86_quirks>
> ffffffff817dfc16:       55                      push   %rbp
> ffffffff817dfc17:       48 89 e5                mov    %rsp,%rbp
> ffffffff817dfc1a:       48 8b 40 28             mov    0x28(%rax),%rax  [*]
> ffffffff817dfc1e:       48 85 c0                test   %rax,%rax
> 
> i.e. RAX was zero.
> 
> i've pushed out the tip/tmp.x86/percpu-zerobased.bad branch which shows 
> the exact kernel that failed. It was generated by:
> 
>  git-checkout tip/master
>  git-merge tip/x86/percpu-zerobased
> 
> 	Ingo

Ok, thanks, I'll take a look.  There were some questions that I had
(and I should have RFC'd the patch since there are still questions.)

Mike