Re: [PATCH v5] sched/clock: Avoid false sharing for sched_clock_irqtime

["Guo, Wangyang" <wangyang.guo@intel.com>]

On 1/27/2026 11:38 AM, K Prateek Nayak wrote:
> Hello Wangyang,
> 
> On 1/27/2026 8:46 AM, Wangyang Guo wrote:
>> @@ -238,6 +239,8 @@ static int __init sched_clock_init_late(void)
>>   
>>   	if (__sched_clock_stable_early)
>>   		__set_sched_clock_stable();
>> +	else
>> +		disable_sched_clock_irqtime();  /* disable if clock unstable. */
> 
> nit. I think we should check for irqtime_enabled() before since
> static_key_disable() would grab the cpus_read_lock() unnecessarily even
> if irqtime wasn't enabled - possible with PA-RISC where a slow processor
> registers the generic sched clock not eabling irqtime and then marks the
> sched_clock unstable on SMP which would hit this without having irqtime
> enabled.
> 
> Same is case with "tsc=noirqtime" and then tsc turns unstable at early
> boot.

Added in v6 as below:
https://lore.kernel.org/all/20260127044159.2254247-1-wangyang.guo@intel.com/

@@ -238,6 +239,8 @@ static int __init sched_clock_init_late(void)

  	if (__sched_clock_stable_early)
  		__set_sched_clock_stable();
+	else if (irqtime_enabled())
+		disable_sched_clock_irqtime();  /* disable if clock unstable. */

  	return 0;
  }