On Thu, 2010-04-15 at 16:12 +1000, Stephen Rothwell wrote:
>
> Caused by commit bd6d29c25bb1a24a4c160ec5de43e0004e01f72b ("lockstat:
> Make lockstat counting per cpu"). This added a WARN_ON_ONCE to
> debug_atomic_inc() which is called from trace_hardirqs_on_caller()
> with
> irqs enabled.
>
> Line 2301 is:
>
> if (unlikely(curr->hardirqs_enabled)) {
> debug_atomic_inc(redundant_hardirqs_on); <--- 2301
> return;
> }
>
> This is especially bad since on PowerPC, WARN_ON is a TRAP and the
> return
> path from the TRAP also calls trace_hardirqs_on_caller(), so the TRAP
> recurses ...
I think this is because our syscall entry pretty much force-enable irqs.
I remember deciding back then that getting lockdep balanced in that area
was tricky and I didn't do it to avoid adding more overhead to the
syscall path but I suppose I could revisit if necessary.
Cheers,
Ben.
--
To unsubscribe from this list: send the line "unsubscribe linux-next" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
