| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-ID: <8fcba5d4-eea5-d7c4-2bf7-482321b333b7@huawei.com>
Date: Tue, 30 May 2023 22:39:24 +0800
From: Yicong Yang <yangyicong@...wei.com>
To: Peter Zijlstra <peterz@...radead.org>
CC: <yangyicong@...ilicon.com>, <mingo@...hat.com>,
<juri.lelli@...hat.com>, <vincent.guittot@...aro.org>,
<dietmar.eggemann@....com>, <tim.c.chen@...ux.intel.com>,
<yu.c.chen@...el.com>, <gautham.shenoy@....com>, <mgorman@...e.de>,
<vschneid@...hat.com>, <linux-kernel@...r.kernel.org>,
<linux-arm-kernel@...ts.infradead.org>, <rostedt@...dmis.org>,
<bsegall@...gle.com>, <bristot@...hat.com>,
<prime.zeng@...wei.com>, <jonathan.cameron@...wei.com>,
<ego@...ux.vnet.ibm.com>, <srikar@...ux.vnet.ibm.com>,
<linuxarm@...wei.com>, <21cnbao@...il.com>,
<kprateek.nayak@....com>, <wuyun.abel@...edance.com>,
Barry Song <song.bao.hua@...ilicon.com>
Subject: Re: [PATCH v8 2/2] sched/fair: Scan cluster before scanning LLC in
wake-up path
On 2023/5/30 19:55, Peter Zijlstra wrote:
> On Tue, May 30, 2023 at 03:02:53PM +0800, Yicong Yang wrote:
>
>> diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
>> index 373ff5f55884..b8c129ed8b47 100644
>> --- a/kernel/sched/fair.c
>> +++ b/kernel/sched/fair.c
>> @@ -6994,6 +6994,30 @@ static int select_idle_cpu(struct task_struct *p, struct sched_domain *sd, bool
>> }
>> }
>>
>> + if (static_branch_unlikely(&sched_cluster_active)) {
>> + struct sched_domain *sdc = rcu_dereference(per_cpu(sd_cluster, target));
>> +
>> + if (sdc) {
>> + for_each_cpu_wrap(cpu, sched_domain_span(sdc), target + 1) {
>> + if (!cpumask_test_cpu(cpu, cpus))
>> + continue;
>> +
>> + if (has_idle_core) {
>> + i = select_idle_core(p, cpu, cpus, &idle_cpu);
>> + if ((unsigned int)i < nr_cpumask_bits)
>> + return i;
>> + } else {
>> + if (--nr <= 0)
>> + return -1;
>> + idle_cpu = __select_idle_cpu(cpu, p);
>> + if ((unsigned int)idle_cpu < nr_cpumask_bits)
>> + return idle_cpu;
>> + }
>> + }
>> + cpumask_andnot(cpus, cpus, sched_domain_span(sdc));
>> + }
>> + }
>
> Would not this:
>
> --- a/kernel/sched/fair.c
> +++ b/kernel/sched/fair.c
> @@ -6994,6 +6994,29 @@ static int select_idle_cpu(struct task_s
> }
> }
>
> + if (static_branch_unlikely(&sched_cluster_active)) {
> + struct sched_group *sg = sd->groups;
> + if (sg->flags & SD_CLUSTER) {
> + for_each_cpu_wrap(cpu, sched_group_span(sg), target+1) {
> + if (!cpumask_test_cpu(cpu, cpus))
> + continue;
> +
> + if (has_idle_core) {
> + i = select_idle_core(p, cpu, cpus, &idle_cpu);
> + if ((unsigned)i < nr_cpumask_bits)
> + return 1;
> + } else {
> + if (--nr <= 0)
> + return -1;
> + idle_cpu = __select_idle_cpu(cpu, p);
> + if ((unsigned)idle_cpu < nr_cpumask_bits)
> + return idle_cpu;
> + }
> + }
> + cpumask_andnot(cpus, cpus, sched_group_span(sg));
> + }
> + }
> +
> for_each_cpu_wrap(cpu, cpus, target + 1) {
> if (has_idle_core) {
> i = select_idle_core(p, cpu, cpus, &idle_cpu);
>
> also work? Then we can avoid the extra sd_cluster per-cpu variable.
>
I thought it will be fine since sg->flags is derived from the child domain. But practically it doesn't.
Tested on a 2P Skylake server with no clusters, add some debug messages to see how sg->flags appears:
diff --git a/kernel/sched/topology.c b/kernel/sched/topology.c
index 69968ed9ffb9..5c443b74abf5 100644
--- a/kernel/sched/topology.c
+++ b/kernel/sched/topology.c
@@ -90,8 +90,8 @@ static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level,
cpumask_or(groupmask, groupmask, sched_group_span(group));
- printk(KERN_CONT " %d:{ span=%*pbl",
- group->sgc->id,
+ printk(KERN_CONT " %d:{ cluster: %s span=%*pbl",
+ group->sgc->id, group->flags & SD_CLUSTER ? "true" : "false",
cpumask_pr_args(sched_group_span(group)));
if ((sd->flags & SD_OVERLAP) &&
Unfortunately the result doesn't match what I expected, the MC domain's sg->flags still marked
as cluster:
[ 8.886099] CPU0 attaching sched-domain(s):
[ 8.889539] domain-0: span=0,40 level=SMT
[ 8.893538] groups: 0:{ cluster: false span=0 }, 40:{ cluster: false span=40 }
[ 8.897538] domain-1: span=0-19,40-59 level=MC
[ 8.901538] groups: 0:{ cluster: true span=0,40 cap=2048 }, 1:{ cluster: true span=1,41 cap=2048 }, 2:{ cluster: true span=2,42 cap=2048 }, 3:{ cluster: true span=3,43 cap=2048 }, 4:{ cluster: true span=4,44 cap=2048 }, 5:{ cluster: true span=5,45 cap=2048 }, 6:{ cluster: true span=6,46 cap=2048 }, 7:{ cluster: true span=7,47 cap=2048 }, 8:{ cluster: true span=8,48 cap=2048 }, 9:{ cluster: true span=9,49 cap=2048 }, 10:{ cluster: true span=10,50 cap=2048 }, 11:{ cluster: true span=11,51 cap=2048 }, 12:{ cluster: true span=12,52 cap=2048 }, 13:{ cluster: true span=13,53 cap=2048 }, 14:{ cluster: true span=14,54 cap=2048 }, 15:{ cluster: true span=15,55 cap=2048 }, 16:{ cluster: true span=16,56 cap=2048 }, 17:{ cluster: true span=17,57 cap=2048 }, 18:{ cluster: true span=18,58 cap=2048 }, 19:{ cluster: true span=19,59 cap=2048 }
[ 8.905538] domain-2: span=0-79 level=NUMA
[ 8.909538] groups: 0:{ cluster: false span=0-19,40-59 cap=40960 }, 20:{ cluster: false span=20-39,60-79 cap=40960 }
I assume we didn't handle the sg->flags correctly on the domain degeneration. Simply checked the code seems
we've already make sg->flags = 0 on degeneration, maybe I need to check where's wrong.
Thanks,
Yicong
Powered by blists - more mailing lists