Re: [PATCH] sched/fair: Fix wrong cpu selecting from isolated domain

[Srikar Dronamraju]

* xunlei <xlpang@xxxxxxxxxxxxxxxxx> [2020-08-25 10:11:24]:

> On 2020/8/24 PM9:38, Srikar Dronamraju wrote:
> > * Xunlei Pang <xlpang@xxxxxxxxxxxxxxxxx> [2020-08-24 20:30:19]:
> > 
> >> We've met problems that occasionally tasks with full cpumask
> >> (e.g. by putting it into a cpuset or setting to full affinity)
> >> were migrated to our isolated cpus in production environment.
> >>
> >> After some analysis, we found that it is due to the current
> >> select_idle_smt() not considering the sched_domain mask.
> >>
> >> Fix it by checking the valid domain mask in select_idle_smt().
> >>
> >> Fixes: 10e2f1acd010 ("sched/core: Rewrite and improve select_idle_siblings())
> >> Reported-by: Wetp Zhang <wetp.zy@xxxxxxxxxxxxxxxxx>
> >> Signed-off-by: Xunlei Pang <xlpang@xxxxxxxxxxxxxxxxx>
> >> ---
> >>  kernel/sched/fair.c | 9 +++++----
> >>  1 file changed, 5 insertions(+), 4 deletions(-)
> >>
> >> diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
> >> index 1a68a05..fa942c4 100644
> >> --- a/kernel/sched/fair.c
> >> +++ b/kernel/sched/fair.c
> >> @@ -6075,7 +6075,7 @@ static int select_idle_core(struct task_struct *p, struct sched_domain *sd, int
> >>  /*
> >>   * Scan the local SMT mask for idle CPUs.
> >>   */
> >> -static int select_idle_smt(struct task_struct *p, int target)
> >> +static int select_idle_smt(struct task_struct *p, struct sched_domain *sd, int target)
> >>  {
> >>  	int cpu;
> >>  
> >> @@ -6083,7 +6083,8 @@ static int select_idle_smt(struct task_struct *p, int target)
> >>  		return -1;
> >>  
> >>  	for_each_cpu(cpu, cpu_smt_mask(target)) {
> >> -		if (!cpumask_test_cpu(cpu, p->cpus_ptr))
> >> +		if (!cpumask_test_cpu(cpu, p->cpus_ptr) ||
> >> +		    !cpumask_test_cpu(cpu, sched_domain_span(sd)))
> >>  			continue;
> > 
> > Don't think this is right thing to do.  What if this task had set a cpumask
> > that doesn't cover all the cpus in this sched_domain_span(sd)

ah, right I  missed the 'or' part.
> 
> It doesn't matter, without this patch, it selects an idle cpu from:
> "cpu_smt_mask(target) and p->cpus_ptr"
> 
> with this patch, it selects an idle cpu from:
> "cpu_smt_mask(target) and p->cpus_ptr and sched_domain_span(sd)"
> 
> > 
> > cpu_smt_mask(target) would already limit to the sched_domain_span(sd) so I
> > am not sure how this can help?
> > 
> > 
> 
> Here is an example:
> CPU0 and CPU16 are hyper-thread pair, CPU16 is domain isolated. So its
> sd_llc doesn't contain CPU16, and cpu_smt_mask(0) is 0 and 16.
> 
> Then we have @target is 0, select_idle_smt() may return the isolated(and
> idle) CPU16 without this patch.

Okay.

-- 
Thanks and Regards
Srikar Dronamraju