[PATCH v3 2/2] sched: Remove the limitation of WF_ON_CPU on wakelist if wakee cpu is idle
From: Tianchen Ding
Date: Thu Jun 02 2022 - 00:07:16 EST
Wakelist can help avoid cache bouncing and offload the overhead of waker
cpu. So far, using wakelist within the same llc only happens on
WF_ON_CPU, and this limitation could be removed to further improve
wakeup performance.
The commit 518cd6234178 ("sched: Only queue remote wakeups when
crossing cache boundaries") disabled queuing tasks on wakelist when
the cpus share llc. This is because, at that time, the scheduler must
send IPIs to do ttwu_queue_wakelist. Nowadays, ttwu_queue_wakelist also
supports TIF_POLLING, so this is not a problem now when the wakee cpu is
in idle polling.
Benefits:
Queuing the task on idle cpu can help improving performance on waker cpu
and utilization on wakee cpu, and further improve locality because
the wakee cpu can handle its own rq. This patch helps improving rt on
our real java workloads where wakeup happens frequently.
Consider the normal condition (CPU0 and CPU1 share same llc)
Before this patch:
CPU0 CPU1
select_task_rq() idle
rq_lock(CPU1->rq)
enqueue_task(CPU1->rq)
notify CPU1 (by sending IPI or CPU1 polling)
resched()
After this patch:
CPU0 CPU1
select_task_rq() idle
add to wakelist of CPU1
notify CPU1 (by sending IPI or CPU1 polling)
rq_lock(CPU1->rq)
enqueue_task(CPU1->rq)
resched()
We see CPU0 can finish its work earlier. It only needs to put task to
wakelist and return.
While CPU1 is idle, so let itself handle its own runqueue data.
This patch brings no difference about IPI.
This patch only takes effect when the wakee cpu is:
1) idle polling
2) idle not polling
For 1), there will be no IPI with or without this patch.
For 2), there will always be an IPI before or after this patch.
Before this patch: waker cpu will enqueue task and check preempt. Since
"idle" will be sure to be preempted, waker cpu must send a resched IPI.
After this patch: waker cpu will put the task to the wakelist of wakee
cpu, and send an IPI.
Benchmark:
We've tested schbench, unixbench, and hachbench on both x86 and arm64.
On x86 (Intel Xeon Platinum 8269CY):
schbench -m 2 -t 8
Latency percentiles (usec) before after
50.0000th: 8 6
75.0000th: 10 7
90.0000th: 11 8
95.0000th: 12 8
*99.0000th: 15 10
99.5000th: 16 11
99.9000th: 20 14
Unixbench with full threads (104)
before after
Dhrystone 2 using register variables 3004614211 3004725417 0.00%
Double-Precision Whetstone 616764.3 617355.9 0.10%
Execl Throughput 26449.2 26468.6 0.07%
File Copy 1024 bufsize 2000 maxblocks 832763.3 824099.4 -1.04%
File Copy 256 bufsize 500 maxblocks 210718.7 211775.1 0.50%
File Copy 4096 bufsize 8000 maxblocks 2393528.2 2398755.4 0.22%
Pipe Throughput 144559102.7 144605068.8 0.03%
Pipe-based Context Switching 3192192.9 3571238.1 11.87%
Process Creation 95270.5 98865.4 3.77%
Shell Scripts (1 concurrent) 113780.6 113924.7 0.13%
Shell Scripts (8 concurrent) 15557.2 15508.9 -0.31%
System Call Overhead 5359984.1 5356711.4 -0.06%
hackbench -g 1 -l 100000
before after
Time 3.246 2.251
On arm64 (Ampere Altra):
schbench -m 2 -t 8
Latency percentiles (usec) before after
50.0000th: 14 10
75.0000th: 19 14
90.0000th: 22 16
95.0000th: 23 16
*99.0000th: 24 17
99.5000th: 24 17
99.9000th: 31 25
Unixbench with full threads (80)
before after
Dhrystone 2 using register variables 3536787968 3536476016 -0.01%
Double-Precision Whetstone 629370.6 629333.3 -0.01%
Execl Throughput 66615.9 66288.8 -0.49%
File Copy 1024 bufsize 2000 maxblocks 1038402.1 1050181.2 1.13%
File Copy 256 bufsize 500 maxblocks 311054.2 310317.2 -0.24%
File Copy 4096 bufsize 8000 maxblocks 2276795.6 2297703 0.92%
Pipe Throughput 130409359.9 130390848.7 -0.01%
Pipe-based Context Switching 3148440.7 3383705.1 7.47%
Process Creation 111574.3 119728.6 7.31%
Shell Scripts (1 concurrent) 122980.7 122657.4 -0.26%
Shell Scripts (8 concurrent) 17482.8 17476.8 -0.03%
System Call Overhead 4424103.4 4430062.6 0.13%
hackbench -g 1 -l 100000
before after
Time 4.217 2.916
Our patch has improvement on schbench, hackbench
and Pipe-based Context Switching of unixbench
when there exists idle cpus,
and no obvious regression on other tests of unixbench.
This can help improve rt in scenes where wakeup happens frequently.
Signed-off-by: Tianchen Ding <dtcccc@xxxxxxxxxxxxxxxxx>
---
kernel/sched/core.c | 27 ++++++++++++++++-----------
kernel/sched/sched.h | 1 -
2 files changed, 16 insertions(+), 12 deletions(-)
diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index a9efe134fbe5..ccb9e0fbf49b 100644
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -3808,7 +3808,7 @@ bool cpus_share_cache(int this_cpu, int that_cpu)
return per_cpu(sd_llc_id, this_cpu) == per_cpu(sd_llc_id, that_cpu);
}
-static inline bool ttwu_queue_cond(int cpu, int wake_flags)
+static inline bool ttwu_queue_cond(int cpu)
{
/*
* Do not complicate things with the async wake_list while the CPU is
@@ -3824,13 +3824,21 @@ static inline bool ttwu_queue_cond(int cpu, int wake_flags)
if (!cpus_share_cache(smp_processor_id(), cpu))
return true;
+ if (cpu == smp_processor_id())
+ return false;
+
/*
- * If the task is descheduling and the only running task on the
- * CPU then use the wakelist to offload the task activation to
- * the soon-to-be-idle CPU as the current CPU is likely busy.
- * nr_running is checked to avoid unnecessary task stacking.
+ * If the wakee cpu is idle, or the task is descheduling and the
+ * only running task on the CPU, then use the wakelist to offload
+ * the task activation to the idle (or soon-to-be-idle) CPU as
+ * the current CPU is likely busy. nr_running is checked to
+ * avoid unnecessary task stacking.
+ *
+ * Note that we can only get here with (wakee) p->on_rq=0,
+ * p->on_cpu can be whatever, we've done the dequeue, so
+ * the wakee has been accounted out of ->nr_running.
*/
- if ((wake_flags & WF_ON_CPU) && !cpu_rq(cpu)->nr_running)
+ if (!cpu_rq(cpu)->nr_running)
return true;
return false;
@@ -3838,10 +3846,7 @@ static inline bool ttwu_queue_cond(int cpu, int wake_flags)
static bool ttwu_queue_wakelist(struct task_struct *p, int cpu, int wake_flags)
{
- if (sched_feat(TTWU_QUEUE) && ttwu_queue_cond(cpu, wake_flags)) {
- if (WARN_ON_ONCE(cpu == smp_processor_id()))
- return false;
-
+ if (sched_feat(TTWU_QUEUE) && ttwu_queue_cond(cpu)) {
sched_clock_cpu(cpu); /* Sync clocks across CPUs */
__ttwu_queue_wakelist(p, cpu, wake_flags);
return true;
@@ -4163,7 +4168,7 @@ try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags)
* scheduling.
*/
if (smp_load_acquire(&p->on_cpu) &&
- ttwu_queue_wakelist(p, task_cpu(p), wake_flags | WF_ON_CPU))
+ ttwu_queue_wakelist(p, task_cpu(p), wake_flags))
goto unlock;
/*
diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h
index 01259611beb9..1e34bb4527fd 100644
--- a/kernel/sched/sched.h
+++ b/kernel/sched/sched.h
@@ -2039,7 +2039,6 @@ static inline int task_on_rq_migrating(struct task_struct *p)
#define WF_SYNC 0x10 /* Waker goes to sleep after wakeup */
#define WF_MIGRATED 0x20 /* Internal use, task got migrated */
-#define WF_ON_CPU 0x40 /* Wakee is on_cpu */
#ifdef CONFIG_SMP
static_assert(WF_EXEC == SD_BALANCE_EXEC);
--
2.27.0