Re: [PATCH 1/2] [RESEND] sched: Fully integrate cpus_active_map and root-domain code

[Gregory Haskins <gregory.haskins@gmail.com>]

[-- Attachment #1: Type: text/plain, Size: 3874 bytes --]

Gregory Haskins wrote:
> Gregory Haskins wrote:
>> (Applies to 2.6.31-rc4)
>>
>> [
>> 	This patch was originaly sent about a year ago, but got dropped
>> 	presumably by accident.  Here is the original thread.
>>
>> 	http://lkml.org/lkml/2008/7/22/281
>>
>> 	At that time, Peter and Max acked it.  It has now been forward
>> 	ported to the new cpumask interface.  I will be so bold as to
>> 	carry their acks forward since the basic logic is the same.
>> 	However, a new acknowledgement, if they have the time to review,
>> 	would be ideal.
>>
>> 	I have tested this patch on a 4-way system using Max's recommended
>> 	"echo 0|1 > cpu1/online" technique and it appears to work properly
>> ]
>>
>> What: Reflect "active" cpus in the rq->rd->online field, instead of the
>> online_map.
>>
>> Motivation:  Things that use the root-domain code (such as cpupri) only
>> care about cpus classified as "active" anyway.  By synchronizing the
>> root-domain state with the active map, we allow several optimizations.
>>
>> For instance, we can remove an extra cpumask_and from the scheduler
>> hotpath by utilizing rq->rd->online (since it is now a cached version
>> of cpu_active_map & rq->rd->span).
>>
>> Signed-off-by: Gregory Haskins <ghaskins@novell.com>
>> Acked-by: Peter Zijlstra <peterz@infradead.org>
>> Acked-by: Max Krasnyansky <maxk@qualcomm.com>
>> ---
>>
>>  kernel/sched.c      |    2 +-
>>  kernel/sched_fair.c |   10 +++++++---
>>  kernel/sched_rt.c   |    7 -------
>>  3 files changed, 8 insertions(+), 11 deletions(-)
>>
>> diff --git a/kernel/sched.c b/kernel/sched.c
>> index 1a104e6..38a1526 100644
>> --- a/kernel/sched.c
>> +++ b/kernel/sched.c
>> @@ -7874,7 +7874,7 @@ static void rq_attach_root(struct rq *rq, struct root_domain *rd)
>>  	rq->rd = rd;
>>  
>>  	cpumask_set_cpu(rq->cpu, rd->span);
>> -	if (cpumask_test_cpu(rq->cpu, cpu_online_mask))
>> +	if (cpumask_test_cpu(rq->cpu, cpu_active_mask))
>>  		set_rq_online(rq);
>>  
>>  	spin_unlock_irqrestore(&rq->lock, flags);
>> diff --git a/kernel/sched_fair.c b/kernel/sched_fair.c
>> index 9ffb2b2..2b9cae6 100644
>> --- a/kernel/sched_fair.c
>> +++ b/kernel/sched_fair.c
>> @@ -1040,17 +1040,21 @@ static void yield_task_fair(struct rq *rq)
>>   * search starts with cpus closest then further out as needed,
>>   * so we always favor a closer, idle cpu.
>>   * Domains may include CPUs that are not usable for migration,
>> - * hence we need to mask them out (cpu_active_mask)
>> + * hence we need to mask them out (rq->rd->online)
>>   *
>>   * Returns the CPU we should wake onto.
>>   */
>>  #if defined(ARCH_HAS_SCHED_WAKE_IDLE)
>> +
>> +#define cpu_rd_active(cpu, rq) cpumask_test_cpu(cpu, rq->rd->online)
>> +
>>  static int wake_idle(int cpu, struct task_struct *p)
>>  {
>>  	struct sched_domain *sd;
>>  	int i;
>>  	unsigned int chosen_wakeup_cpu;
>>  	int this_cpu;
>> +	struct rq *task_rq = task_rq(p);
>>  
>>  	/*
>>  	 * At POWERSAVINGS_BALANCE_WAKEUP level, if both this_cpu and prev_cpu
>> @@ -1083,10 +1087,10 @@ static int wake_idle(int cpu, struct task_struct *p)
>>  	for_each_domain(cpu, sd) {
>>  		if ((sd->flags & SD_WAKE_IDLE)
>>  		    || ((sd->flags & SD_WAKE_IDLE_FAR)
>> -			&& !task_hot(p, task_rq(p)->clock, sd))) {
>> +			&& !task_hot(p, task_rq->clock, sd))) {
>>  			for_each_cpu_and(i, sched_domain_span(sd),
>>  					 &p->cpus_allowed) {
> 
> Hmm, something got suboptimal in translation from the original patch.
> 
> This would be better expressed as:
> 
> for_each_cpu_and(i, rq->rd->online, &p->cpus_allowed) {
> 	if (idle_cpu(i)...
> }


NM.  My first instinct was correct.

We need the result of sd->span, cpus_allowed, and active_map.  The
submitted logic (or the original) is correct, and my comment above is wrong.

Sorry for the noise,
-Greg


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 267 bytes --]