2024-03-22 14:17:59

by Shrikanth Hegde

[permalink] [raw]
Subject: [PATCH v2 2/2] sched/fair: Use helper functions to access rd->overload

rd->overload is accessed at multiple places. Instead it could use helper
get/set functions. This would make the code more readable and easy to
maintain.

No change in functionality intended.

Suggested-by: Qais Yousef <[email protected]>
Signed-off-by: Shrikanth Hegde <[email protected]>
Reviewed-by: Vincent Guittot <[email protected]>
---
kernel/sched/fair.c | 7 ++++---
kernel/sched/sched.h | 14 ++++++++++++--
2 files changed, 16 insertions(+), 5 deletions(-)

diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index eeebadd7d9ae..cee99c93e6a4 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -10621,8 +10621,9 @@ static inline void update_sd_lb_stats(struct lb_env *env, struct sd_lb_stats *sd

if (!env->sd->parent) {
/* update overload indicator if we are at root domain */
- if (READ_ONCE(env->dst_rq->rd->overload) != (sg_status & SG_OVERLOAD))
- WRITE_ONCE(env->dst_rq->rd->overload, sg_status & SG_OVERLOAD);
+ if (is_rd_overloaded(env->dst_rq->rd) != (sg_status & SG_OVERLOAD))
+ set_rd_overload_status(env->dst_rq->rd,
+ sg_status & SG_OVERLOAD);

/* Update over-utilization (tipping point, U >= 0) indicator */
set_rd_overutilized_status(env->dst_rq->rd,
@@ -12344,7 +12345,7 @@ static int sched_balance_newidle(struct rq *this_rq, struct rq_flags *rf)
rcu_read_lock();
sd = rcu_dereference_check_sched_domain(this_rq->sd);

- if (!READ_ONCE(this_rq->rd->overload) ||
+ if (!is_rd_overloaded(this_rq->rd) ||
(sd && this_rq->avg_idle < sd->max_newidle_lb_cost)) {

if (sd)
diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h
index 41024c1c49b4..c91eb8811bef 100644
--- a/kernel/sched/sched.h
+++ b/kernel/sched/sched.h
@@ -918,6 +918,16 @@ extern void rq_attach_root(struct rq *rq, struct root_domain *rd);
extern void sched_get_rd(struct root_domain *rd);
extern void sched_put_rd(struct root_domain *rd);

+static inline int is_rd_overloaded(struct root_domain *rd)
+{
+ return READ_ONCE(rd->overload);
+}
+
+static inline void set_rd_overload_status(struct root_domain *rd, int status)
+{
+ WRITE_ONCE(rd->overload, status);
+}
+
#ifdef HAVE_RT_PUSH_IPI
extern void rto_push_irq_work_func(struct irq_work *work);
#endif
@@ -2518,8 +2528,8 @@ static inline void add_nr_running(struct rq *rq, unsigned count)

#ifdef CONFIG_SMP
if (prev_nr < 2 && rq->nr_running >= 2) {
- if (!READ_ONCE(rq->rd->overload))
- WRITE_ONCE(rq->rd->overload, 1);
+ if (!is_rd_overloaded(rq->rd))
+ set_rd_overload_status(rq->rd, 1);
}
#endif

--
2.39.3



2024-03-24 01:16:37

by Qais Yousef

[permalink] [raw]
Subject: Re: [PATCH v2 2/2] sched/fair: Use helper functions to access rd->overload

On 03/22/24 19:46, Shrikanth Hegde wrote:
> rd->overload is accessed at multiple places. Instead it could use helper
> get/set functions. This would make the code more readable and easy to
> maintain.
>
> No change in functionality intended.
>
> Suggested-by: Qais Yousef <[email protected]>

Thanks for following up!

> Signed-off-by: Shrikanth Hegde <[email protected]>
> Reviewed-by: Vincent Guittot <[email protected]>
> ---
> kernel/sched/fair.c | 7 ++++---
> kernel/sched/sched.h | 14 ++++++++++++--
> 2 files changed, 16 insertions(+), 5 deletions(-)
>
> diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
> index eeebadd7d9ae..cee99c93e6a4 100644
> --- a/kernel/sched/fair.c
> +++ b/kernel/sched/fair.c
> @@ -10621,8 +10621,9 @@ static inline void update_sd_lb_stats(struct lb_env *env, struct sd_lb_stats *sd
>
> if (!env->sd->parent) {
> /* update overload indicator if we are at root domain */
> - if (READ_ONCE(env->dst_rq->rd->overload) != (sg_status & SG_OVERLOAD))
> - WRITE_ONCE(env->dst_rq->rd->overload, sg_status & SG_OVERLOAD);
> + if (is_rd_overloaded(env->dst_rq->rd) != (sg_status & SG_OVERLOAD))
> + set_rd_overload_status(env->dst_rq->rd,
> + sg_status & SG_OVERLOAD);

A bit picky, but..

Wouldn't it be better to encapsulate the check of whether we're writing a new
value inside set_rd_overload_status()? Only write if it the value changed and
all future users wouldn't care then.

I think no need to wrap the line too.

>
> /* Update over-utilization (tipping point, U >= 0) indicator */
> set_rd_overutilized_status(env->dst_rq->rd,
> @@ -12344,7 +12345,7 @@ static int sched_balance_newidle(struct rq *this_rq, struct rq_flags *rf)
> rcu_read_lock();
> sd = rcu_dereference_check_sched_domain(this_rq->sd);
>
> - if (!READ_ONCE(this_rq->rd->overload) ||
> + if (!is_rd_overloaded(this_rq->rd) ||
> (sd && this_rq->avg_idle < sd->max_newidle_lb_cost)) {
>
> if (sd)
> diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h
> index 41024c1c49b4..c91eb8811bef 100644
> --- a/kernel/sched/sched.h
> +++ b/kernel/sched/sched.h
> @@ -918,6 +918,16 @@ extern void rq_attach_root(struct rq *rq, struct root_domain *rd);
> extern void sched_get_rd(struct root_domain *rd);
> extern void sched_put_rd(struct root_domain *rd);
>
> +static inline int is_rd_overloaded(struct root_domain *rd)
> +{
> + return READ_ONCE(rd->overload);
> +}
> +
> +static inline void set_rd_overload_status(struct root_domain *rd, int status)
> +{
> + WRITE_ONCE(rd->overload, status);
> +}
> +
> #ifdef HAVE_RT_PUSH_IPI
> extern void rto_push_irq_work_func(struct irq_work *work);
> #endif
> @@ -2518,8 +2528,8 @@ static inline void add_nr_running(struct rq *rq, unsigned count)
>
> #ifdef CONFIG_SMP
> if (prev_nr < 2 && rq->nr_running >= 2) {
> - if (!READ_ONCE(rq->rd->overload))
> - WRITE_ONCE(rq->rd->overload, 1);
> + if (!is_rd_overloaded(rq->rd))
> + set_rd_overload_status(rq->rd, 1);

While at it, could you write SG_OVERLOAD instead of 1?

Both patches LGTM otherwise

Reviewed-by: Qais Yousef <[email protected]>


Thanks!

--
Qais Yousef

> }
> #endif
>
> --
> 2.39.3
>

2024-03-25 12:40:48

by Shrikanth Hegde

[permalink] [raw]
Subject: Re: [PATCH v2 2/2] sched/fair: Use helper functions to access rd->overload



On 3/24/24 6:46 AM, Qais Yousef wrote:
> On 03/22/24 19:46, Shrikanth Hegde wrote:

>>
>> if (!env->sd->parent) {
>> /* update overload indicator if we are at root domain */
>> - if (READ_ONCE(env->dst_rq->rd->overload) != (sg_status & SG_OVERLOAD))
>> - WRITE_ONCE(env->dst_rq->rd->overload, sg_status & SG_OVERLOAD);
>> + if (is_rd_overloaded(env->dst_rq->rd) != (sg_status & SG_OVERLOAD))
>> + set_rd_overload_status(env->dst_rq->rd,
>> + sg_status & SG_OVERLOAD);
>
> A bit picky, but..
>
> Wouldn't it be better to encapsulate the check of whether we're writing a new
> value inside set_rd_overload_status()? Only write if it the value changed and
> all future users wouldn't care then.
>
> I think no need to wrap the line too.
>

Yes. Makes sense all the places do the check and update, can be wrapped.
I will send out v3.

>>
>> /* Update over-utilization (tipping point, U >= 0) indicator */
>> set_rd_overutilized_status(env->dst_rq->rd,
>> @@ -12344,7 +12345,7 @@ static int sched_balance_newidle(struct rq *this_rq, struct rq_flags *rf)
>> rcu_read_lock();
>> sd = rcu_dereference_check_sched_domain(this_rq->sd);
>>
>> - if (!READ_ONCE(this_rq->rd->overload) ||
>> + if (!is_rd_overloaded(this_rq->rd) ||
>> (sd && this_rq->avg_idle < sd->max_newidle_lb_cost)) {
>>
>> if (sd)
>> @@ -2518,8 +2528,8 @@ static inline void add_nr_running(struct rq *rq, unsigned count)
>>
>> #ifdef CONFIG_SMP
>> if (prev_nr < 2 && rq->nr_running >= 2) {
>> - if (!READ_ONCE(rq->rd->overload))
>> - WRITE_ONCE(rq->rd->overload, 1);
>> + if (!is_rd_overloaded(rq->rd))
>> + set_rd_overload_status(rq->rd, 1);
>
> While at it, could you write SG_OVERLOAD instead of 1?
Done.
>
> Both patches LGTM otherwise
>
> Reviewed-by: Qais Yousef <[email protected]>
>
>