Re: BUG: sleeping function called from invalid context at arch/x86/mm/fault.c:1081

[Peter Zijlstra <peterz@infradead.org>]

On Tue, 2010-12-14 at 12:29 -0800, Linus Torvalds wrote:
> which looks like one of the "list_for_each_entry()" uses of that
> hash_head. In fact, building lockdep.lst and matching against that
> 0x2a6 constant (it's a warn_slowpath_null thing and is the line
> number), we can pinpoint exactly which one it seems to be. Afaik, it
> seems to be this code in look_up_lock_class:
> 
>         /*
>          * We can walk the hash lockfree, because the hash only
>          * grows, and we are careful when adding entries to the end:
>          */
>         list_for_each_entry(class, hash_head, hash_entry) {
>                 if (class->key == key) {
>                         WARN_ON_ONCE(class->name != lock->name);
>                         return class;
>                 }
>         }
> 
> Maybe the RCU list use in lockdep isn't entirely safe?
> 
> Added Peter and Ingo to the cc. It does look like some lockdep race.
> 
> Or then it's some general memory corruption, of course. That just
> happened to hit the lockdep data structures. That would explain your
> subsequent radeon GP fault too. 

Typically such crashes in lockdep are indeed indicative of a scribble,
typically a use after free with slab poisoning.

But I'll try and run through that code later today to see if I can spot
anything.