(In reply to comment #0)
> Description of problem:
> Kernel got deadlock in function rcu_check_quiescent_state(). Here's how the
> problem happen:
> 1. CPU_1 was processing rcu callbacks via __rcu_process_callbacks
> 2. CPU_2 call smp_call_function() to all other cpus in set_mtrr
> 3. All other cpus including CPU_1 were beginning to run ipi_handler() as a
> result of the smp all interruption. They all had already decreased the
> data.count and been waiting for CPU_2 setting gate to 1.
> 4. But interruption happened on CPU_2 before it set gate to 1. And then it
> entered __rcu_process_callbacks too, so it got deadlock because the spinlock
> was held by CPU_1.
> 
> Please additional info for detailed vmcore analysis.
> 
>  

Mark, can you try to reproduce using the debug kernel which has lockdep enabled?

Thanks,

P.

Created attachment 477896 [details]
RHEL5 fix for this issue

Mark, please apply and test.

Thanks,

P.

Prarit,
Thanks for your patch. I have built a test kernel and delivered it to customer. I will come back to you when I get feedback from customer.