Re: [PATCH] sched/fair: Don't balance task to its current running CPU

[Vincent Guittot]

On Mon, 29 May 2023 at 13:02, Yicong Yang <yangyicong@xxxxxxxxxx> wrote:
>
> Hi Vincent,
>
> On 2023/5/26 18:34, Vincent Guittot wrote:
> > On Fri, 26 May 2023 at 10:18, Yicong Yang <yangyicong@xxxxxxxxxx> wrote:
> >>
> >> On 2023/5/25 23:13, Vincent Guittot wrote:
> >>> On Wed, 24 May 2023 at 09:21, Yicong Yang <yangyicong@xxxxxxxxxx> wrote:
> >>>>
> >>>> From: Yicong Yang <yangyicong@xxxxxxxxxxxxx>
> >>>>
> >>>> We've run into the case that the balancer tries to balance a migration
> >>>> disabled task and trigger the warning in set_task_cpu() like below:
> >>>>
> >>>>  ------------[ cut here ]------------
> >>>>  WARNING: CPU: 7 PID: 0 at kernel/sched/core.c:3115 set_task_cpu+0x188/0x240
> >>>>  Modules linked in: hclgevf xt_CHECKSUM ipt_REJECT nf_reject_ipv4 <...snip>
> >>>>  CPU: 7 PID: 0 Comm: swapper/7 Kdump: loaded Tainted: G           O       6.1.0-rc4+ #1
> >>>>  Hardware name: Huawei TaiShan 2280 V2/BC82AMDC, BIOS 2280-V2 CS V5.B221.01 12/09/2021
> >>>>  pstate: 604000c9 (nZCv daIF +PAN -UAO -TCO -DIT -SSBS BTYPE=--)
> >>>>  pc : set_task_cpu+0x188/0x240
> >>>>  lr : load_balance+0x5d0/0xc60
> >>>>  sp : ffff80000803bc70
> >>>>  x29: ffff80000803bc70 x28: ffff004089e190e8 x27: ffff004089e19040
> >>>>  x26: ffff007effcabc38 x25: 0000000000000000 x24: 0000000000000001
> >>>>  x23: ffff80000803be84 x22: 000000000000000c x21: ffffb093e79e2a78
> >>>>  x20: 000000000000000c x19: ffff004089e19040 x18: 0000000000000000
> >>>>  x17: 0000000000001fad x16: 0000000000000030 x15: 0000000000000000
> >>>>  x14: 0000000000000003 x13: 0000000000000000 x12: 0000000000000000
> >>>>  x11: 0000000000000001 x10: 0000000000000400 x9 : ffffb093e4cee530
> >>>>  x8 : 00000000fffffffe x7 : 0000000000ce168a x6 : 000000000000013e
> >>>>  x5 : 00000000ffffffe1 x4 : 0000000000000001 x3 : 0000000000000b2a
> >>>>  x2 : 0000000000000b2a x1 : ffffb093e6d6c510 x0 : 0000000000000001
> >>>>  Call trace:
> >>>>   set_task_cpu+0x188/0x240
> >>>>   load_balance+0x5d0/0xc60
> >>>>   rebalance_domains+0x26c/0x380
> >>>>   _nohz_idle_balance.isra.0+0x1e0/0x370
> >>>>   run_rebalance_domains+0x6c/0x80
> >>>>   __do_softirq+0x128/0x3d8
> >>>>   ____do_softirq+0x18/0x24
> >>>>   call_on_irq_stack+0x2c/0x38
> >>>>   do_softirq_own_stack+0x24/0x3c
> >>>>   __irq_exit_rcu+0xcc/0xf4
> >>>>   irq_exit_rcu+0x18/0x24
> >>>>   el1_interrupt+0x4c/0xe4
> >>>>   el1h_64_irq_handler+0x18/0x2c
> >>>>   el1h_64_irq+0x74/0x78
> >>>>   arch_cpu_idle+0x18/0x4c
> >>>>   default_idle_call+0x58/0x194
> >>>>   do_idle+0x244/0x2b0
> >>>>   cpu_startup_entry+0x30/0x3c
> >>>>   secondary_start_kernel+0x14c/0x190
> >>>>   __secondary_switched+0xb0/0xb4
> >>>>  ---[ end trace 0000000000000000 ]---
> >>>>
> >>>> Further investigation shows that the warning is superfluous, the migration
> >>>> disabled task is just going to be migrated to its current running CPU.
> >>>> This is because that on load balance if the dst_cpu is not allowed by the
> >>>> task, we'll re-select a new_dst_cpu as a candidate. If no task can be
> >>>> balanced to dst_cpu we'll try to balance the task to the new_dst_cpu
> >>>> instead. In this case when the migration disabled task is not on CPU it
> >>>> only allows to run on its current CPU, load balance will select its
> >>>> current CPU as new_dst_cpu and later triggers the the warning above.
> >>>>
> >>>> This patch tries to solve this by not select the task's current running
> >>>> CPU as new_dst_cpu in the load balance.
> >>>>
> >>>> Signed-off-by: Yicong Yang <yangyicong@xxxxxxxxxxxxx>
> >>>> ---
> >>>> Thanks Valentin for the knowledge of migration disable. Previous discussion can
> >>>> be found at
> >>>> https://lore.kernel.org/all/20230313065759.39698-1-yangyicong@xxxxxxxxxx/
> >>>>
> >>>>  kernel/sched/fair.c | 3 ++-
> >>>>  1 file changed, 2 insertions(+), 1 deletion(-)
> >>>>
> >>>> diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
> >>>> index 7a1b1f855b96..3c4f3a244c1d 100644
> >>>> --- a/kernel/sched/fair.c
> >>>> +++ b/kernel/sched/fair.c
> >>>> @@ -8456,7 +8456,8 @@ int can_migrate_task(struct task_struct *p, struct lb_env *env)
> >>>>
> >>>>                 /* Prevent to re-select dst_cpu via env's CPUs: */
> >>>>                 for_each_cpu_and(cpu, env->dst_grpmask, env->cpus) {
> >>>> -                       if (cpumask_test_cpu(cpu, p->cpus_ptr)) {
> >>>> +                       if (cpumask_test_cpu(cpu, p->cpus_ptr) &&
> >>>> +                           cpu != env->src_cpu) {
> >>>
> >>> So I'm a bit surprised that src_cpu can be part of the dst_grpmask and
> >>> selected as new_dst_cpu. The only reason would be some numa
> >>> overlapping domains. Is it the case for you ?
> >>>
> >>
> >> It's a 2P 4 NUMA machine, the groups in the top NUMA domains are overlapped, for example for CPU64:
> >>
> >> [    3.147038] CPU64 attaching sched-domain(s):
> >> [    3.147040]  domain-0: span=64-67 level=CLS
> >> [    3.147043]   groups: 64:{ span=64 cap=1023 }, 65:{ span=65 cap=1023 }, 66:{ span=66 cap=1023 }, 67:{ span=67 }
> >> [    3.147056]   domain-1: span=64-95 level=MC
> >> [    3.147059]    groups: 64:{ span=64-67 cap=4093 }, 68:{ span=68-71 cap=4096 }, 72:{ span=72-75 cap=4096 }, 76:{ span=76-79 cap=4096 }, 80:{ span=80-83 cap=4096 }, 84:{ span=84-87 cap=4096 }, 88:{ span=88-91 cap=4096 }, 92:{ span=92-95 cap=4096 }
> >> [    3.147085]    domain-2: span=64-127 level=NUMA
> >> [    3.147087]     groups: 64:{ span=64-95 cap=32765 }, 96:{ span=96-127 cap=32767 }
> >> [    3.147095]     domain-3: span=0-31,64-127 level=NUMA
> >> [    3.147098]      groups: 64:{ span=64-127 cap=65532 }, 0:{ span=0-31 cap=32767 }
> >> [    3.147106]      domain-4: span=0-127 level=NUMA
> >> [    3.147109]       groups: 64:{ span=0-31,64-127 mask=64-95 cap=98300 }, 32:{ span=0-63 mask=32-63 cap=65531 }
> >>
> >
> > Thanks for confirming this.
> >
> > So I wonder if a better solution would be to make env->dst_grpmask =
> > group_balance_cpu(sd->groups) instead of
> > sched_group_span(sd->groups),. The behavior remains the same for non
> > overlapping groups because group_balance_cpu(sd->groups) ==
> > sched_group_span(sd->groups) in this case and for overlapping group,
> > we will try to find a dst_cpu that is not contained in src/busiest
> > group and the load balance will effectively pull load from the
> > busiest_group
> >
>
> I think this make sense to me. We've already limited the dst_cpu within the
> group_balance_mask(sd->groups) in should_we_balance() for periodical balance
> (but not for idle balance). The sg->sgc->cpumask is commented as "balance
> mask", so only the cpus in sg->sgc->cpumask can pull the task in the load balance.
> The newidle CPU maybe an exception, but also need to limit the new_dst_cpu
> int the sg->sgc->cpumask.

I think that will be okay. only cpus in sg->sgc->cpumask will make a
change in the load_balance so we should use this mask

>
> >
> >>>>                                 env->flags |= LBF_DST_PINNED;
> >>>>                                 env->new_dst_cpu = cpu;
> >>>>                                 break;
> >>>> --
> >>>> 2.24.0
> >>>>
> >>> .
> >>>
> > .
> >