Re: [PATCH V2] tmp patch to fix hotplug issue in CMCI storm

[Thomas Gleixner <tglx@linutronix.de>]

On Mon, 18 Jun 2012, Chen Gong wrote:
> index 92d8b5c..0493525 100644
> --- a/arch/x86/kernel/cpu/mcheck/mce_intel.c
> +++ b/arch/x86/kernel/cpu/mcheck/mce_intel.c
> @@ -40,6 +40,7 @@ static DEFINE_RAW_SPINLOCK(cmci_discover_lock);
>  static DEFINE_PER_CPU(unsigned long, cmci_time_stamp);
>  static DEFINE_PER_CPU(unsigned int, cmci_storm_cnt);
>  static DEFINE_PER_CPU(unsigned int, cmci_storm_state);
> +static DEFINE_PER_CPU(unsigned long, cmci_storm_hcpu_status);

Why do you insist on having another status variable, which does
actually nothing than obfuscate the code?

Look at the usage sites:

>  		__this_cpu_write(cmci_storm_state, CMCI_STORM_SUBSIDED);
> +		__this_cpu_write(cmci_storm_hcpu_status, CMCI_STORM_HCPU_NONE);

>  	__this_cpu_write(cmci_storm_state, CMCI_STORM_ACTIVE);
> +	__this_cpu_write(cmci_storm_hcpu_status, CMCI_STORM_HCPU_ACTIVE);

So it's a shadow variable of cmci_storm_state for no value.

And all you do with it is:

> +void mce_intel_hcpu_update(unsigned long cpu)
> +{
> +	unsigned long *status = &per_cpu(cmci_storm_hcpu_status, cpu);
> +
> +	if (*status == CMCI_STORM_HCPU_ACTIVE) {

This can be checked with the existing variable as well. And your check
leaves CMCI_STORM_SUBSIDED as a stale value around.

This simply wants to check

     	if (per_cpu(cmci_storm_state, cpu) == CMCI_STORM_ACTIVE)
		atomic_dec(&cmci_storm_on_cpus);

and unconditionally clear the state	   			      

	per_cpu(cmci_storm_state, cpu) = CMCI_STORM_NONE;

Right?