Re: [PATCH Resend v5] sched: fix init NOHZ_IDLE flag

[Frederic Weisbecker <fweisbec@gmail.com>]

On Wed, Apr 03, 2013 at 03:37:43PM +0200, Vincent Guittot wrote:
> On my smp platform which is made of 5 cores in 2 clusters, I have the
> nr_busy_cpu field of sched_group_power struct that is not null when the
> platform is fully idle. The root cause is:
> During the boot sequence, some CPUs reach the idle loop and set their
> NOHZ_IDLE flag while waiting for others CPUs to boot. But the nr_busy_cpus
> field is initialized later with the assumption that all CPUs are in the busy
> state whereas some CPUs have already set their NOHZ_IDLE flag.
> 
> More generally, the NOHZ_IDLE flag must be initialized when new sched_domains
> are created in order to ensure that NOHZ_IDLE and nr_busy_cpus are aligned.
> 
> This condition can be ensured by adding a synchronize_rcu between the
> destruction of old sched_domains and the creation of new ones so the NOHZ_IDLE
> flag will not be updated with old sched_domain once it has been initialized.
> But this solution introduces a additionnal latency in the rebuild sequence
> that is called during cpu hotplug.
> 
> As suggested by Frederic Weisbecker, another solution is to have the same
> rcu lifecycle for both NOHZ_IDLE and sched_domain struct. I have introduce
> a new sched_domain_rq struct that is the entry point for both sched_domains
> and objects that must follow the same lifecycle like NOHZ_IDLE flags. They
> will share the same RCU lifecycle and will be always synchronized.
> 
> The synchronization is done at the cost of :
>  - an additional indirection for accessing the first sched_domain level
>  - an additional indirection and a rcu_dereference before accessing to the
>    NOHZ_IDLE flag.
> 
> Change since v4:
>  - link both sched_domain and NOHZ_IDLE flag in one RCU object so
>    their states are always synchronized.
> 
> Change since V3;
>  - NOHZ flag is not cleared if a NULL domain is attached to the CPU
>  - Remove patch 2/2 which becomes useless with latest modifications
> 
> Change since V2:
>  - change the initialization to idle state instead of busy state so a CPU that
>    enters idle during the build of the sched_domain will not corrupt the
>    initialization state
> 
> Change since V1:
>  - remove the patch for SCHED softirq on an idle core use case as it was
>    a side effect of the other use cases.
> 
> Signed-off-by: Vincent Guittot <vincent.guittot@linaro.org>

Ok, the lockless scheme involving nr_buzy_cpus and rq flags seem to be correct now.
We can hope somebody will come up with a less complicated solution. But for now that's
the best fix I've seen.

I just have a few comments on details.

> ---
>  include/linux/sched.h |    6 +++
>  kernel/sched/core.c   |  105 ++++++++++++++++++++++++++++++++++++++++++++-----
>  kernel/sched/fair.c   |   35 +++++++++++------
>  kernel/sched/sched.h  |   24 +++++++++--
>  4 files changed, 145 insertions(+), 25 deletions(-)
> 
> diff --git a/include/linux/sched.h b/include/linux/sched.h
> index d35d2b6..2a52188 100644
> --- a/include/linux/sched.h
> +++ b/include/linux/sched.h
> @@ -959,6 +959,12 @@ struct sched_domain {
>  	unsigned long span[0];
>  };
> 
> +struct sched_domain_rq {
> +	struct sched_domain *sd;
> +	unsigned long flags;
> +	struct rcu_head rcu;	/* used during destruction */
> +};

So the reason for this level of indirection won't be intuitive for those
who read that code. Please add some comments that explain why we need
that. ie: because we need the object lifecycle of sched_power and flags to be the
same for the lockless scheme to work.

> +
>  static inline struct cpumask *sched_domain_span(struct sched_domain *sd)
>  {
>  	return to_cpumask(sd->span);
> diff --git a/kernel/sched/core.c b/kernel/sched/core.c
> index 7f12624..69e2313 100644
> --- a/kernel/sched/core.c
> +++ b/kernel/sched/core.c
> @@ -5602,6 +5602,15 @@ static void destroy_sched_domains(struct sched_domain *sd, int cpu)
>  		destroy_sched_domain(sd, cpu);
>  }
>  
> +static void destroy_sched_domain_rq(struct sched_domain_rq *sd_rq, int cpu)
> +{
> +	if (!sd_rq)
> +		return;
> +
> +	destroy_sched_domains(sd_rq->sd, cpu);
> +	kfree_rcu(sd_rq, rcu);
> +}
> +
>  /*
>   * Keep a special pointer to the highest sched_domain that has
>   * SD_SHARE_PKG_RESOURCE set (Last Level Cache Domain) for this
> @@ -5632,10 +5641,23 @@ static void update_top_cache_domain(int cpu)
>   * hold the hotplug lock.
>   */
>  static void
> -cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu)
> +cpu_attach_domain(struct sched_domain_rq *sd_rq, struct root_domain *rd,
> +		int cpu)
>  {
>  	struct rq *rq = cpu_rq(cpu);
> -	struct sched_domain *tmp;
> +	struct sched_domain_rq *tmp_rq;

old_sd_rq would be a clearer name.

> +	struct sched_domain *tmp, *sd = NULL;
> +
> +	/*
> +	 * If we don't have any sched_domain and associated object, we can
> +	 * directly jump to the attach sequence otherwise we try to degenerate
> +	 * the sched_domain
> +	 */
> +	if (!sd_rq)
> +		goto attach;
> +
> +	/* Get a pointer to the 1st sched_domain */
> +	sd = sd_rq->sd;
>  
>  	/* Remove the sched domains which do not contribute to scheduling. */
>  	for (tmp = sd; tmp; ) {
> @@ -5658,14 +5680,17 @@ cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu)
>  		destroy_sched_domain(tmp, cpu);
>  		if (sd)
>  			sd->child = NULL;
> +		/* update sched_domain_rq */
> +		sd_rq->sd = sd;
>  	}
>  
> +attach:
>  	sched_domain_debug(sd, cpu);
>  
>  	rq_attach_root(rq, rd);
> -	tmp = rq->sd;
> -	rcu_assign_pointer(rq->sd, sd);
> -	destroy_sched_domains(tmp, cpu);
> +	tmp_rq = rq->sd_rq;
> +	rcu_assign_pointer(rq->sd_rq, sd_rq);
> +	destroy_sched_domain_rq(tmp_rq, cpu);
>  
>  	update_top_cache_domain(cpu);

[...]
> +static void __sdrq_free(const struct cpumask *cpu_map, struct s_data *d)
> +{
> +	int j;
> +
> +	for_each_cpu(j, cpu_map)
> +		if (*per_cpu_ptr(d->sd_rq, j))
> +			kfree(*per_cpu_ptr(d->sd_rq, j));

kfree(NULL) works.

> +}
> +
> +static void build_sched_domain_rq(struct s_data *d, int cpu)
> +{
> +	struct sched_domain_rq *sd_rq;
> +	struct sched_domain *sd;
> +
> +	/* Attach sched_domain to sched_domain_rq */
> +	sd = *per_cpu_ptr(d->sd, cpu);
> +	sd_rq = *per_cpu_ptr(d->sd_rq, cpu);
> +	sd_rq->sd = sd;
> +	/* Init flags */
> +	set_bit(NOHZ_IDLE, sched_rq_flags(sd_rq));
> +}
> +
>  struct sched_domain *build_sched_domain(struct sched_domain_topology_level *tl,
>  		struct s_data *d, const struct cpumask *cpu_map,
>  		struct sched_domain_attr *attr, struct sched_domain *child,
> @@ -6495,6 +6572,7 @@ static int build_sched_domains(const struct cpumask *cpu_map,
>  			       struct sched_domain_attr *attr)
>  {
>  	enum s_alloc alloc_state = sa_none;
> +	struct sched_domain_rq *sd_rq;
>  	struct sched_domain *sd;
>  	struct s_data d;
>  	int i, ret = -ENOMEM;
> @@ -6547,11 +6625,18 @@ static int build_sched_domains(const struct cpumask *cpu_map,
>  		}
>  	}
>  
> +	/* Init objects that must follow the sched_domain lifecycle */
> +	for_each_cpu(i, cpu_map) {
> +		build_sched_domain_rq(&d, i);
> +	}

I suggest you put the above domain_rq initialization before the domain
initialization.

So that we have that more intuitively ordered initialization:

* set up domains_rq
* set up domains
* set up sched groups
* set up sched groups power

> +
>  	/* Attach the domains */
>  	rcu_read_lock();
>  	for_each_cpu(i, cpu_map) {
> -		sd = *per_cpu_ptr(d.sd, i);
> -		cpu_attach_domain(sd, d.rd, i);
> +		sd_rq = *per_cpu_ptr(d.sd_rq, i);
> +		cpu_attach_domain(sd_rq, d.rd, i);
> +		/* claim allocation of sched_domain_rq object */
> +		*per_cpu_ptr(d.sd_rq, i) = NULL;
>  	}
>  	rcu_read_unlock();
[...]
> @@ -5673,7 +5681,12 @@ static void run_rebalance_domains(struct softirq_action *h)
>  
>  static inline int on_null_domain(int cpu)
>  {
> -	return !rcu_dereference_sched(cpu_rq(cpu)->sd);
> +	struct sched_domain_rq *sd_rq =
> +		rcu_dereference_sched(cpu_rq(cpu)->sd_rq);
> +	struct sched_domain *sd = NULL;
> +	if (sd_rq)
> +		sd = sd_rq->sd;

Is it possible to have sd_rq->sd == NULL ?

> +	return !sd;
>  }
>  
>  /*
> diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h
> index cc03cfd..f589306 100644
> --- a/kernel/sched/sched.h
> +++ b/kernel/sched/sched.h
> @@ -417,7 +417,7 @@ struct rq {
>  
>  #ifdef CONFIG_SMP
>  	struct root_domain *rd;
> -	struct sched_domain *sd;
> +	struct sched_domain_rq *sd_rq;
>  
>  	unsigned long cpu_power;
>  
> @@ -505,21 +505,37 @@ DECLARE_PER_CPU(struct rq, runqueues);
>  
>  #ifdef CONFIG_SMP
>  
> -#define rcu_dereference_check_sched_domain(p) \
> +#define rcu_dereference_check_sched_domain_rq(p) \
>  	rcu_dereference_check((p), \
>  			      lockdep_is_held(&sched_domains_mutex))
>  
> +#define get_sched_domain_rq(cpu) \
> +	rcu_dereference_check_sched_domain_rq(cpu_rq(cpu)->sd_rq)

How about rcu_dereference_domain_rq()? It seems important to me that
we keep the rcu_dereference_*() naming so that we don't hide what really
happens there behind a more opaque naming.

I mean RCU is tricky to deal with and it's important not to obfuscate its use.

> +
> +#define rcu_dereference_check_sched_domain(cpu) ({ \
> +	struct sched_domain_rq *__sd_rq = get_sched_domain_rq(cpu); \
> +	struct sched_domain *__sd = NULL; \
> +	if (__sd_rq) \
> +		__sd = __sd_rq->sd; \

Same question for the NULL sd.

> +	__sd; \
> +})
> +
> +#define sched_rq_flags(sd_rq) (&sd_rq->flags)

Hm, why this "sched_" prefix? Maybe rq_domain_flags() ?

Thanks.