[Qemu-devel] Re: [PATCH] x86: Reboot CPU on triple fault - Version 4

[Jan Kiszka <jan.kiszka@siemens.com>]

Anthony Liguori wrote:
> Jan Kiszka wrote:
>> Anthony Liguori wrote:
>>  
>>> Why is it so important to log triple faults?  Printing something to
>>> stderr instead of logfile is just wrong.
>>>
>>> What is your use-case?  I think you've got the wrong solution to the
>>> problem and I can't find an adequate description of the problem your
>>> trying to solve in this thread.
>>>     
>>
>> You may not always sit next to your VM when it decides to reboot due to
>> a _real_ problem that caused a triple fault.
>>   
> 
> You can always check within the guest to see if it's rebooted (via
> uptime for instance).

But you won't find the CPU state on triple fault there.

> 
> It's extremely unlikely you'll ever see an OS triple fault in the wild
> unless you're doing kernel development.  Triple faulting requires a bad
> IDT or a really bad page table both of which are not something an OS is
> likely to do by accident.  If your OS is triple faulting, I highly doubt
> it's just going to reboot and everything's going to be okay.

There are various OSes out there in the wild. Not all of them conform to
common assumptions about how OSes typically look like. And once you
start moving things under a different roof (like QEMU), you are better
off logging such /potentially/ critical events (specifically if that
roof is a bit smaller due to missing segment limit and type checks).
That's at least our situation ATM.

> 
> Besides, the -no-reboot flag should already satisfy your needs if you're
> really concerned about it.

Nope, because we also have the case where a triple fault is
intentionally triggered to perform a reboot.

Jan

-- 
Siemens AG, Corporate Technology, CT SE 2
Corporate Competence Center Embedded Linux