[PATCH v1] sched: adjust the layout of the cfs_bandwith structure to save memory

[PATCH v1] sched: adjust the layout of the cfs_bandwith structure to save memory

[zenghongling]

Using pahole, we can see that there are some padding holes
in the current cfs_bandwith structure. Adjusting the
layout of cfs_bandwith can reduce these holes,
resulting in the size of the structure decreasing
from 240 bytes to 232 bytes.
It is possible to make period_timer access one more cache
line.it is already touching so many cachelines and isn't
that kind of hot path.

Before:
struct cfs_bandwidth {
	raw_spinlock_t             lock;                 /*     0     4 */

	/* XXX 4 bytes hole, try to pack */

	ktime_t                    period;               /*     8     8 */
	u64                        quota;                /*    16     8 */
	u64                        runtime;              /*    24     8 */
	u64                        burst;                /*    32     8 */
	u64                        runtime_snap;         /*    40     8 */
	s64                        hierarchical_quota;   /*    48     8 */
	u8                         idle;                 /*    56     1 */
	u8                         period_active;        /*    57     1 */
	u8                         slack_started;        /*    58     1 */

	/* XXX 5 bytes hole, try to pack */

	/* --- cacheline 1 boundary (64 bytes) --- */
	struct hrtimer             period_timer;         /*    64     64 */

	/* --- cacheline 2 boundary (128 bytes) --- */
	struct hrtimer             slack_timer;          /*    128    64 */

	/* --- cacheline 3 boundary (192 bytes) --- */
	struct list_head           throttled_cfs_rq;     /*   192    16 */
	int                        nr_periods;           /*   208     4 */
	int                        nr_throttled;         /*   212     4 */
	int                        nr_burst;             /*   216     4 */

	/* XXX 4 bytes hole, try to pack */

	u64                        throttled_time;       /*   224     8 */
	u64                        burst_time;           /*   232     8 */

	/* size: 240, cachelines: 4, members: 18 */
	/* sum members: 227, holes: 3, sum holes: 13 */
	/* paddings: 2, sum paddings: 8 */
	/* forced alignments: 2, forced holes: 1, sum forced holes: 5 */
	/* last cacheline: 48 bytes */
}

After:
struct cfs_bandwidth {
	raw_spinlock_t             lock;                 /*     0     4 */
	u8                         idle;                 /*     4     1 */
	u8                         period_active;        /*     5     1 */
	u8                         slack_started;        /*     6     1 */

	/* XXX 1 byte hole, try to pack */

	ktime_t                    period;               /*     8     8 */
	u64                        quota;                /*    16     8 */
	u64                        runtime;              /*    24     8 */
	u64                        burst;                /*    32     8 */
	u64                        runtime_snap;         /*    40     8 */
	s64                        hierarchical_quota;   /*    48     8 */
	struct hrtimer             period_timer;         /*    56    64 */

	/* --- cacheline 1 boundary (64 bytes) was 56 bytes ago --- */
	struct hrtimer             slack_timer;          /*   120    64 */

	/* --- cacheline 2 boundary (128 bytes) was 56 bytes ago --- */
	struct list_head           throttled_cfs_rq;     /*   184    16 */
	/* --- cacheline 3 boundary (192 bytes) was 8 bytes ago --- */
	int                        nr_periods;           /*   200     4 */
	int                        nr_throttled;         /*   204     4 */
	int                        nr_burst;             /*   208     4 */

	/* XXX 4 bytes hole, try to pack */

	u64                        throttled_time;       /*   216     8 */
	u64                        burst_time;           /*   224     8 */

	/* size: 232, cachelines: 4, members: 18 */
	/* sum members: 227, holes: 2, sum holes: 5 */
	/* paddings: 2, sum paddings: 8 */
	/* forced alignments: 2 */
	/* last cacheline: 40 bytes */
}

Signed-off-by: zenghongling <zenghongling@kylinos.cn>

---
  v1:
  - change the commit.
---
---
 kernel/sched/sched.h | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h
index adfb6e3409d7..55ebd3d868e6 100644
--- a/kernel/sched/sched.h
+++ b/kernel/sched/sched.h
@@ -444,6 +444,9 @@ static inline u64 default_bw_period_us(void)
 struct cfs_bandwidth {
 #ifdef CONFIG_CFS_BANDWIDTH
 	raw_spinlock_t		lock;
+	u8                      idle;
+	u8                      period_active;
+	u8                      slack_started;
 	ktime_t			period;
 	u64			quota;
 	u64			runtime;
@@ -451,9 +454,6 @@ struct cfs_bandwidth {
 	u64			runtime_snap;
 	s64			hierarchical_quota;
 
-	u8			idle;
-	u8			period_active;
-	u8			slack_started;
 	struct hrtimer		period_timer;
 	struct hrtimer		slack_timer;
 	struct list_head	throttled_cfs_rq;
-- 
2.25.1

Re: [PATCH v1] sched: adjust the layout of the cfs_bandwith structure to save memory

[Benjamin Segall]

zenghongling <zenghongling@kylinos.cn> writes:

> Using pahole, we can see that there are some padding holes
> in the current cfs_bandwith structure. Adjusting the
> layout of cfs_bandwith can reduce these holes,
> resulting in the size of the structure decreasing
> from 240 bytes to 232 bytes.
> It is possible to make period_timer access one more cache
> line.it is already touching so many cachelines and isn't
> that kind of hot path.

Something more like: "This make make period_timer access one more cache
line, but it is already [...]"

And commits generally have a blank line between paragraphs.

With that,

Reviewed-by: Ben Segall <bsegall@google.com>