[PATCH] powerpc: Fix WARN_ON in decrementer_check_overflow

[Hugh Dickins]

On Thu, 9 Feb 2012, Benjamin Herrenschmidt wrote:
> We use __get_cpu_var() which triggers a false positive warning
> in smp_processor_id() thinking interrupts are enabled (at this
> point, they are soft-enabled but hard-disabled).
> 
> Signed-off-by: Benjamin Herrenschmidt <benh at kernel.crashing.org>
> ---
> 
> I was initially planning to fix that with a more in-depth rework
> of how we do lazy irq disabling on powerpc, but that patch is
> becoming too complex for this release so I'll apply this as a
> stop-gap and leave the full rework for -next

Okay, thanks for the update, that's equivalent to the get_cpu_var
plus put_cpu_var patch I had generally been running with successfully.

(I was not at all confident that it was a sufficient fix, and have
to say "generally" above because one time I got something that looked
like recursive calls overflowing the stack, and IIRC something "irq"
did appear in each frame of the trace.  But probably no connection,
never seen again, and I don't recall even which -rc or -next it was.)

Hugh

>  
> diff --git a/arch/powerpc/kernel/irq.c b/arch/powerpc/kernel/irq.c
> index 701d4ac..01e2877 100644
> --- a/arch/powerpc/kernel/irq.c
> +++ b/arch/powerpc/kernel/irq.c
> @@ -118,10 +118,14 @@ static inline notrace void set_soft_enabled(unsigned long enable)
>  static inline notrace void decrementer_check_overflow(void)
>  {
>  	u64 now = get_tb_or_rtc();
> -	u64 *next_tb = &__get_cpu_var(decrementers_next_tb);
> +	u64 *next_tb;
> +
> +	preempt_disable();
> +	next_tb = &__get_cpu_var(decrementers_next_tb);
>  
>  	if (now >= *next_tb)
>  		set_dec(1);
> +	preempt_enable();
>  }
>  
>  notrace void arch_local_irq_restore(unsigned long en)