Re: BUG-lockdep and freeze (was: Arrr! Linux 2.6.18)

[Andi Kleen <ak@suse.de>]

On Saturday 30 September 2006 23:56, Linus Torvalds wrote:
> 
> On Sat, 30 Sep 2006, Andi Kleen wrote:
> > 
> > Anyways, I guess we need even more validation in the fallback code,
> > but just terminating the kernel thread stacks should fix that particular case.
> 
> Why not just add the simple validation?
> 
> A kernel stack is one page in size. If you move to another page, you 
> terminate. It's that simple.

No, it's not. On x86-64 it can be three or more stacks nested in
complicated ways (process stack, interrupt stack, exception stack)
The exception stack can happen multiple times.
 
> What if the kernel stack is corrupt? Buffer overruns do that.
> 
> This patch seems to just paper over the _real_ problem, namely the fact 
> that the stack tracer code doesn't actually validate any of its arguments.

It has pretty good sanity checking by first using __get_user for the stack
data, and the regularly double checking the EIPs by looking them up
in CFI. If it can't find them it will abort.

> The old unwinder (well, at least for x86, and I assume x86-64 used that as 
> the beginning point) didn't have this problem at all, exactly because it 
> couldn't get on the wrong stack-page in the first place.

In this particular case what happened is that the dwarf2 unwinder
ended and then the fallback was in the wrong page and couldn't handle 
it.
 
> The old code literally had:
> 
> 	static inline int valid_stack_ptr(struct thread_info *tinfo, void *p)
> 	{
> 	        return  p > (void *)tinfo &&
> 	                p < (void *)tinfo + THREAD_SIZE - 3;
> 	}
> 
> and would refuse to touch anything that wasn't in the stack page. It was 
> simple, AND WE NEVER _EVER_ HAD A BUG RELATED TO IT, AFAIK.

That was before interrupt stacks were introduced. With that it is significantly
more complicated. On x86-64 even more because there are exception stacks.

-Andi