update_idle_core() is only done for the case of sched_smt_present. but test_idle_cores() is done for all machines even those without smt. this could contribute to up 8%+ hackbench performance loss on a machine like kunpeng 920 which has no smt. this patch removes the redundant test_idle_cores() for non-smt machines.
we run the below hackbench with different -g parameter from 2 to 14, for each different g, we run the command 10 times and get the average time: $ numactl -N 0 hackbench -p -T -l 20000 -g $1
hackbench will report the time which is needed to complete a certain number of messages transmissions between a certain number of tasks, for example: $ numactl -N 0 hackbench -p -T -l 20000 -g 10 Running in threaded mode with 10 groups using 40 file descriptors each (== 400 tasks) Each sender will pass 20000 messages of 100 bytes
The below is the result of hackbench w/ and w/o this patch: g= 2 4 6 8 10 12 14 w/o: 1.8151 3.8499 5.5142 7.2491 9.0340 10.7345 12.0929 w/ : 1.8428 3.7436 5.4501 6.9522 8.2882 9.9535 11.3367 +4.1% +8.3% +7.3% +6.3%
Signed-off-by: Barry Song song.bao.hua@hisilicon.com --- kernel/sched/fair.c | 8 +++++--- 1 file changed, 5 insertions(+), 3 deletions(-)
diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c index 2e2ab1e..de42a32 100644 --- a/kernel/sched/fair.c +++ b/kernel/sched/fair.c @@ -6038,9 +6038,11 @@ static inline bool test_idle_cores(int cpu, bool def) { struct sched_domain_shared *sds;
- sds = rcu_dereference(per_cpu(sd_llc_shared, cpu)); - if (sds) - return READ_ONCE(sds->has_idle_cores); + if (static_branch_likely(&sched_smt_present)) { + sds = rcu_dereference(per_cpu(sd_llc_shared, cpu)); + if (sds) + return READ_ONCE(sds->has_idle_cores); + }
return def; }
Hi Barry,
On 2021/3/21 6:14, Barry Song wrote:
update_idle_core() is only done for the case of sched_smt_present. but test_idle_cores() is done for all machines even those without smt.
The patch looks good to me. May I know for what case we need to keep CONFIG_SCHED_SMT for non-smt machines?
Thanks, -Aubrey
this could contribute to up 8%+ hackbench performance loss on a machine like kunpeng 920 which has no smt. this patch removes the redundant test_idle_cores() for non-smt machines.
we run the below hackbench with different -g parameter from 2 to 14, for each different g, we run the command 10 times and get the average time: $ numactl -N 0 hackbench -p -T -l 20000 -g $1
hackbench will report the time which is needed to complete a certain number of messages transmissions between a certain number of tasks, for example: $ numactl -N 0 hackbench -p -T -l 20000 -g 10 Running in threaded mode with 10 groups using 40 file descriptors each (== 400 tasks) Each sender will pass 20000 messages of 100 bytes
The below is the result of hackbench w/ and w/o this patch: g= 2 4 6 8 10 12 14 w/o: 1.8151 3.8499 5.5142 7.2491 9.0340 10.7345 12.0929 w/ : 1.8428 3.7436 5.4501 6.9522 8.2882 9.9535 11.3367 +4.1% +8.3% +7.3% +6.3%
Signed-off-by: Barry Song song.bao.hua@hisilicon.com
kernel/sched/fair.c | 8 +++++--- 1 file changed, 5 insertions(+), 3 deletions(-)
diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c index 2e2ab1e..de42a32 100644 --- a/kernel/sched/fair.c +++ b/kernel/sched/fair.c @@ -6038,9 +6038,11 @@ static inline bool test_idle_cores(int cpu, bool def) { struct sched_domain_shared *sds;
- sds = rcu_dereference(per_cpu(sd_llc_shared, cpu));
- if (sds)
return READ_ONCE(sds->has_idle_cores);
if (static_branch_likely(&sched_smt_present)) {
sds = rcu_dereference(per_cpu(sd_llc_shared, cpu));if (sds)return READ_ONCE(sds->has_idle_cores);}
return def;
}
-----Original Message----- From: Li, Aubrey [mailto:aubrey.li@linux.intel.com] Sent: Monday, March 22, 2021 5:37 PM To: Song Bao Hua (Barry Song) song.bao.hua@hisilicon.com; vincent.guittot@linaro.org; mingo@redhat.com; peterz@infradead.org; juri.lelli@redhat.com; dietmar.eggemann@arm.com; rostedt@goodmis.org; bsegall@google.com; mgorman@suse.de Cc: valentin.schneider@arm.com; linux-arm-kernel@lists.infradead.org; linux-kernel@vger.kernel.org; xuwei (O) xuwei5@huawei.com; Zengtao (B) prime.zeng@hisilicon.com; guodong.xu@linaro.org; yangyicong yangyicong@huawei.com; Liguozhu (Kenneth) liguozhu@hisilicon.com; linuxarm@openeuler.org Subject: [Linuxarm] Re: [PATCH] sched/fair: remove redundant test_idle_cores for non-smt
Hi Barry,
On 2021/3/21 6:14, Barry Song wrote:
update_idle_core() is only done for the case of sched_smt_present. but test_idle_cores() is done for all machines even those without smt.
The patch looks good to me. May I know for what case we need to keep CONFIG_SCHED_SMT for non-smt machines?
Hi Aubrey,
I think the defconfig of arm64 has always enabled CONFIG_SCHED_SMT: https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/arch...
it is probably true for x86 as well.
I don't think Linux distribution will build a separate kernel for machines without smt. so basically the kernel depends on runtime topology parse to figure out if smt is present rather than depending on a rebuild.
Thanks, -Aubrey
this could contribute to up 8%+ hackbench performance loss on a machine like kunpeng 920 which has no smt. this patch removes the redundant test_idle_cores() for non-smt machines.
we run the below hackbench with different -g parameter from 2 to 14, for each different g, we run the command 10 times and get the average time: $ numactl -N 0 hackbench -p -T -l 20000 -g $1
hackbench will report the time which is needed to complete a certain number of messages transmissions between a certain number of tasks, for example: $ numactl -N 0 hackbench -p -T -l 20000 -g 10 Running in threaded mode with 10 groups using 40 file descriptors each (== 400 tasks) Each sender will pass 20000 messages of 100 bytes
The below is the result of hackbench w/ and w/o this patch: g= 2 4 6 8 10 12 14 w/o: 1.8151 3.8499 5.5142 7.2491 9.0340 10.7345 12.0929 w/ : 1.8428 3.7436 5.4501 6.9522 8.2882 9.9535 11.3367 +4.1% +8.3% +7.3% +6.3%
Signed-off-by: Barry Song song.bao.hua@hisilicon.com
kernel/sched/fair.c | 8 +++++--- 1 file changed, 5 insertions(+), 3 deletions(-)
diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c index 2e2ab1e..de42a32 100644 --- a/kernel/sched/fair.c +++ b/kernel/sched/fair.c @@ -6038,9 +6038,11 @@ static inline bool test_idle_cores(int cpu, bool def) { struct sched_domain_shared *sds;
- sds = rcu_dereference(per_cpu(sd_llc_shared, cpu));
- if (sds)
return READ_ONCE(sds->has_idle_cores);
if (static_branch_likely(&sched_smt_present)) {
sds = rcu_dereference(per_cpu(sd_llc_shared, cpu));if (sds)return READ_ONCE(sds->has_idle_cores);}
return def;
}
Thanks Barry
On Sun, Mar 21, 2021 at 11:14:32AM +1300, Barry Song wrote:
update_idle_core() is only done for the case of sched_smt_present. but test_idle_cores() is done for all machines even those without smt. this could contribute to up 8%+ hackbench performance loss on a machine like kunpeng 920 which has no smt. this patch removes the redundant test_idle_cores() for non-smt machines.
we run the below hackbench with different -g parameter from 2 to 14, for each different g, we run the command 10 times and get the average time: $ numactl -N 0 hackbench -p -T -l 20000 -g $1
hackbench will report the time which is needed to complete a certain number of messages transmissions between a certain number of tasks, for example: $ numactl -N 0 hackbench -p -T -l 20000 -g 10 Running in threaded mode with 10 groups using 40 file descriptors each (== 400 tasks) Each sender will pass 20000 messages of 100 bytes
The below is the result of hackbench w/ and w/o this patch: g= 2 4 6 8 10 12 14 w/o: 1.8151 3.8499 5.5142 7.2491 9.0340 10.7345 12.0929 w/ : 1.8428 3.7436 5.4501 6.9522 8.2882 9.9535 11.3367 +4.1% +8.3% +7.3% +6.3%
Signed-off-by: Barry Song song.bao.hua@hisilicon.com
Acked-by: Mel Gorman mgorman@suse.de
That said, the numa_idle_core() function then becomes slightly redundant. A possible follow up is to move the "idle_core >= 0" check in numa_idle_core() to its caller in update_numa_stats() and then remove the redundant check in !static_branch_likely(&sched_smt_present) check in numa_idle_core.
On Sun, Mar 21, 2021 at 11:14:32AM +1300, Barry Song wrote:
update_idle_core() is only done for the case of sched_smt_present. but test_idle_cores() is done for all machines even those without smt. this could contribute to up 8%+ hackbench performance loss on a machine like kunpeng 920 which has no smt. this patch removes the redundant test_idle_cores() for non-smt machines.
we run the below hackbench with different -g parameter from 2 to 14, for each different g, we run the command 10 times and get the average time: $ numactl -N 0 hackbench -p -T -l 20000 -g $1
hackbench will report the time which is needed to complete a certain number of messages transmissions between a certain number of tasks, for example: $ numactl -N 0 hackbench -p -T -l 20000 -g 10 Running in threaded mode with 10 groups using 40 file descriptors each (== 400 tasks) Each sender will pass 20000 messages of 100 bytes
The below is the result of hackbench w/ and w/o this patch: g= 2 4 6 8 10 12 14 w/o: 1.8151 3.8499 5.5142 7.2491 9.0340 10.7345 12.0929 w/ : 1.8428 3.7436 5.4501 6.9522 8.2882 9.9535 11.3367 +4.1% +8.3% +7.3% +6.3%
The patch looks obvious, but the Changelog and Subject needed a lot of help.
I've changed it like so:
--- Subject: sched/fair: Optimize test_idle_cores() for !SMT From: Barry Song song.bao.hua@hisilicon.com Date: Sun, 21 Mar 2021 11:14:32 +1300
From: Barry Song song.bao.hua@hisilicon.com
update_idle_core() is only done for the case of sched_smt_present. but test_idle_cores() is done for all machines even those without SMT.
This can contribute to up 8%+ hackbench performance loss on a machine like kunpeng 920 which has no SMT. This patch removes the redundant test_idle_cores() for !SMT machines.
Hackbench is ran with -g {2..14}, for each g it is ran 10 times to get an average.
$ numactl -N 0 hackbench -p -T -l 20000 -g $1
The below is the result of hackbench w/ and w/o this patch:
g= 2 4 6 8 10 12 14 w/o: 1.8151 3.8499 5.5142 7.2491 9.0340 10.7345 12.0929 w/ : 1.8428 3.7436 5.4501 6.9522 8.2882 9.9535 11.3367 +4.1% +8.3% +7.3% +6.3%
Signed-off-by: Barry Song song.bao.hua@hisilicon.com Signed-off-by: Peter Zijlstra (Intel) peterz@infradead.org Acked-by: Mel Gorman mgorman@suse.de Link: https://lkml.kernel.org/r/20210320221432.924-1-song.bao.hua@hisilicon.com --- kernel/sched/fair.c | 8 +++++--- 1 file changed, 5 insertions(+), 3 deletions(-)
--- a/kernel/sched/fair.c +++ b/kernel/sched/fair.c @@ -6038,9 +6038,11 @@ static inline bool test_idle_cores(int c { struct sched_domain_shared *sds;
- sds = rcu_dereference(per_cpu(sd_llc_shared, cpu)); - if (sds) - return READ_ONCE(sds->has_idle_cores); + if (static_branch_likely(&sched_smt_present)) { + sds = rcu_dereference(per_cpu(sd_llc_shared, cpu)); + if (sds) + return READ_ONCE(sds->has_idle_cores); + }
return def; }
On Sat, 20 Mar 2021 at 23:21, Barry Song song.bao.hua@hisilicon.com wrote:
update_idle_core() is only done for the case of sched_smt_present. but test_idle_cores() is done for all machines even those without smt. this could contribute to up 8%+ hackbench performance loss on a machine like kunpeng 920 which has no smt. this patch removes the redundant test_idle_cores() for non-smt machines.
we run the below hackbench with different -g parameter from 2 to 14, for each different g, we run the command 10 times and get the average time: $ numactl -N 0 hackbench -p -T -l 20000 -g $1
hackbench will report the time which is needed to complete a certain number of messages transmissions between a certain number of tasks, for example: $ numactl -N 0 hackbench -p -T -l 20000 -g 10 Running in threaded mode with 10 groups using 40 file descriptors each (== 400 tasks) Each sender will pass 20000 messages of 100 bytes
The below is the result of hackbench w/ and w/o this patch: g= 2 4 6 8 10 12 14 w/o: 1.8151 3.8499 5.5142 7.2491 9.0340 10.7345 12.0929 w/ : 1.8428 3.7436 5.4501 6.9522 8.2882 9.9535 11.3367 +4.1% +8.3% +7.3% +6.3%
Signed-off-by: Barry Song song.bao.hua@hisilicon.com
Reviewed-by: Vincent Guittot vincent.guittot@linaro.org
kernel/sched/fair.c | 8 +++++--- 1 file changed, 5 insertions(+), 3 deletions(-)
diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c index 2e2ab1e..de42a32 100644 --- a/kernel/sched/fair.c +++ b/kernel/sched/fair.c @@ -6038,9 +6038,11 @@ static inline bool test_idle_cores(int cpu, bool def) { struct sched_domain_shared *sds;
sds = rcu_dereference(per_cpu(sd_llc_shared, cpu));if (sds)return READ_ONCE(sds->has_idle_cores);
if (static_branch_likely(&sched_smt_present)) {sds = rcu_dereference(per_cpu(sd_llc_shared, cpu));if (sds)return READ_ONCE(sds->has_idle_cores);} return def;}
1.8.3.1
-
Barry Song -
Li, Aubrey -
Mel Gorman -
Peter Zijlstra -
Song Bao Hua (Barry Song)
-
Vincent Guittot