Re: [PATCH v3 07/11] x86/mm: Stop calling leave_mm() in idle code

[Andy Lutomirski]

On Wed, Jun 21, 2017 at 2:22 AM, Thomas Gleixner <tglx@xxxxxxxxxxxxx> wrote:
> On Tue, 20 Jun 2017, Andy Lutomirski wrote:
>> diff --git a/drivers/idle/intel_idle.c b/drivers/idle/intel_idle.c
>> index 216d7ec88c0c..2ae43f59091d 100644
>> --- a/drivers/idle/intel_idle.c
>> +++ b/drivers/idle/intel_idle.c
>> @@ -912,16 +912,15 @@ static __cpuidle int intel_idle(struct cpuidle_device *dev,
>>       struct cpuidle_state *state = &drv->states[index];
>>       unsigned long eax = flg2MWAIT(state->flags);
>>       unsigned int cstate;
>> -     int cpu = smp_processor_id();
>>
>>       cstate = (((eax) >> MWAIT_SUBSTATE_SIZE) & MWAIT_CSTATE_MASK) + 1;
>>
>>       /*
>> -      * leave_mm() to avoid costly and often unnecessary wakeups
>> -      * for flushing the user TLB's associated with the active mm.
>> +      * NB: if CPUIDLE_FLAG_TLB_FLUSHED is set, this idle transition
>> +      * will probably flush the TLB.  It's not guaranteed to flush
>> +      * the TLB, though, so it's not clear that we can do anything
>> +      * useful with this knowledge.
>
> So my understanding here is:
>
>       The C-state transition might flush the TLB, when cstate->flags has
>       CPUIDLE_FLAG_TLB_FLUSHED set. The idle transition already took the
>       CPU out of the set of CPUs which are remotely flushed, so the
>       knowledge about this potential flush is not useful for the kernels
>       view of the TLB state.

Indeed.  I assume the theory behind the old code was that leave_mm()
was expensive and that CPUIDLE_FLAG_TLB_FLUSHED would be a decent
heuristic for when to do it.