This bug can be reproduced on Kunpeng arm64 and Phytium arm physical machines,
as well as in virtual machine environments, based on the linux-4.19.y stable
branch:
1. Check the number of CPUs on the system:
   nproc --all
   96

2. Add the parameter isolcpus=0-85 to the grub configuration,
update grub, and reboot.

3. Check the ksmd process:

   ps aux | grep -i ksmd
   root      502  0.0  0.0      0     0 ?        S    10:00   0:00 [ksmd]

   ps -o pid,psr,comm -p 502
   PID PSR COMMAND
   502   0 ksmd

4. Check the kthreadd process:

   ps aux | grep -i kthreadd
   root        2  0.0  0.0      0     0 ?        S    10:00   0:00 [kthreadd]

   ps -o pid,psr,comm -p 2
   PID PSR COMMAND
     2   0 kthreadd

From the output above, it can be seen that both ksmd and kthreadd are still
running on CPU0, which is unreasonable since CPU0 has been isolated.

Signed-off-by: wujing <realwujing@qq.com>
Signed-off-by: QiLiang Yuan <yuanql9@chinatelecom.cn>
---
 kernel/sched/fair.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index 0950cabfc1d0..454021ff70a1 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -6211,7 +6211,7 @@ static int select_idle_cpu(struct task_struct *p, struct sched_domain *sd, int t
 
 	this_sd = rcu_dereference(*this_cpu_ptr(&sd_llc));
 	if (!this_sd)
-		return -1;
+		return housekeeping_any_cpu(HK_FLAG_DOMAIN);
 
 	/*
 	 * Due to large variance we need a large fuzz factor; hackbench in
-- 
2.39.5

Your subject explicitly mentions arm64, however there is absolutely
nothing arm64 specific to this patch, as such the subject is just plain
wrong.

On Mon, Jan 06, 2025 at 05:04:03PM +0800, wujing wrote:
> This bug can be reproduced on Kunpeng arm64 and Phytium arm physical machines,
> as well as in virtual machine environments, based on the linux-4.19.y stable

Development does not happen on stable branches.

The reason this issue does not occur on x86 architecture is that x86 does not 
use the `select_idle_cpu` function for core selection. Instead, it uses the
`select_idle_smt` function. However, there is still an issue on x86 where 
isolated cores are not properly excluded. My other commit, 
`sched/fair: Correct CPU selection from isolated domain`, addresses the issue 
of isolated cores on x86.

> Your subject explicitly mentions arm64, however there is absolutely
> nothing arm64 specific to this patch, as such the subject is just plain
> wrong.

On x86 architecture, we have checked many machines in the cluster, and this 
issue does not exist.

> Your subject explicitly mentions arm64, however there is absolutely
> nothing arm64 specific to this patch, as such the subject is just plain
>wrong.