[v2] sched/isolation: Fix illegal CPU value by housekeeping_any_cpu() return

  For kernels built with CONFIG_NO_HZ_FULL=y, running the following tests:

runqemu kvm slirp nographic qemuparams="-m 1024 -smp 4" bootparams=
"console=ttyS0 nohz_full=0,1 rcu_nocbs=0,1 sched_verbose" -d

root@qemux86-64:~# echo 0 > /sys/devices/system/cpu/cpu2/online
root@qemux86-64:~# echo 0 > /sys/devices/system/cpu/cpu3/online

[   22.838290] BUG: unable to handle page fault for address: ffffffff84cd48c0
[   22.839409] #PF: supervisor read access in kernel mode
[   22.840215] #PF: error_code(0x0000) - not-present page
[   22.841028] PGD 3e19067 P4D 3e19067 PUD 3e1a063 PMD 800ffffffb3ff062
[   22.841889] Oops: 0000 [#1] PREEMPT SMP DEBUG_PAGEALLOC KASAN PTI
[   22.842175] CPU: 0 PID: 16 Comm: rcu_preempt Not tainted 6.2.0-rc1-yocto-standard+ #658
[   22.842534] Hardware name: QEMU Standard PC (Q35 + ICH9, 2009),
               BIOS rel-1.16.0-0-gd239552ce722-prebuilt.qemu.o4
[   22.843036] RIP: 0010:do_raw_spin_trylock+0x70/0x120
[   22.843267] Code: 81 c7 00 f1 f1 f1 f1 c7 40 04 04 f3 f3 f3 65
               48 8b 04 25 28 00 00 00 48 89 45 e0 31 c0 e8 b8 0
[   22.844187] RSP: 0018:ffff8880072b7b30 EFLAGS: 00010046
[   22.844429] RAX: 0000000000000000 RBX: ffffffff84cd48c0 RCX: dffffc0000000000
[   22.844751] RDX: 0000000000000003 RSI: 0000000000000004 RDI: ffffffff84cd48c0
[   22.845074] RBP: ffff8880072b7ba8 R08: ffffffff811daa20 R09: fffffbfff099a919
[   22.845400] R10: ffffffff84cd48c3 R11: fffffbfff099a918 R12: 1ffff11000e56f66
[   22.845719] R13: ffffffff84cd48d8 R14: ffffffff84cd48c0 R15: ffff8880072b7cd8
[   22.846040] FS:  0000000000000000(0000) GS:ffff888035200000(0000) knlGS:0000000000000000
[   22.846403] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[   22.846667] CR2: ffffffff84cd48c0 CR3: 000000001036c000 CR4: 00000000001506f0
[   22.846988] Call Trace:
[   22.847105]  <TASK>
[   22.847208]  ? __pfx_do_raw_spin_trylock+0x10/0x10
[   22.847430]  ? rcu_read_unlock+0x26/0x80
[   22.847612]  ? trace_preempt_off+0x2a/0x130
[   22.847812]  _raw_spin_lock+0x41/0x80
[   22.847984]  ? schedule_timeout+0x242/0x580
[   22.848178]  schedule_timeout+0x242/0x580
[   22.848366]  ? __pfx_schedule_timeout+0x10/0x10
[   22.848575]  ? __pfx_do_raw_spin_trylock+0x10/0x10
[   22.848796]  ? __pfx_process_timeout+0x10/0x10
[   22.849005]  ? _raw_spin_unlock_irqrestore+0x46/0x80
[   22.849232]  ? prepare_to_swait_event+0xb8/0x210
[   22.849450]  rcu_gp_fqs_loop+0x66e/0xe70
[   22.849633]  ? rcu_gp_init+0x87c/0x1130
[   22.849813]  ? __pfx_rcu_gp_fqs_loop+0x10/0x10
[   22.850022]  ? _raw_spin_unlock_irqrestore+0x46/0x80
[   22.850251]  ? finish_swait+0xce/0x100
[   22.850429]  rcu_gp_kthread+0x2ea/0x6b0
[   22.850608]  ? __pfx_do_raw_spin_trylock+0x10/0x10
[   22.850829]  ? __pfx_rcu_gp_kthread+0x10/0x10
[   22.851039]  ? __kasan_check_read+0x11/0x20
[   22.851233]  ? __kthread_parkme+0xe8/0x110
[   22.851424]  ? __pfx_rcu_gp_kthread+0x10/0x10
[   22.851627]  kthread+0x172/0x1a0
[   22.851781]  ? __pfx_kthread+0x10/0x10
[   22.851956]  ret_from_fork+0x2c/0x50
[   22.852129]  </TASK>

schedule_timeout()
->__mod_timer()
 ->get_target_base(base, timer->flags)
   ->get_timer_cpu_base(tflags, get_nohz_timer_target());
     ->cpu = get_nohz_timer_target()
             ->housekeeping_any_cpu(HK_TYPE_TIMER)
                     /*housekeeping.cpumasks[type] is 2-3*/
		     /*cpu_online_mask is 0-1*/
               ->cpu = cpumask_any_and(housekeeping.cpumasks[type],
			cpu_online_mask);
             /*cpu value is 4*/
     ->new_base = per_cpu_ptr(&timer_bases[BASE_DEF], cpu);
   /*new_base is illegal address*/
 ->if (base != new_base)
   ->raw_spin_lock(&new_base->lock); ==> trigger Oops

This commit therefore add checks for cpumask_any_and() return values
in housekeeping_any_cpu(), if cpumask_any_and() returns an illegal CPU
value, the housekeeping_any_cpu() will return current CPU number.

Signed-off-by: Zqiang <qiang1.zhang@intel.com>
---
 kernel/sched/isolation.c | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)
  

Thanks for checking this!

On Fri, Feb 10, 2023 at 08:00:21AM +0800, Zqiang wrote:
> For kernels built with CONFIG_NO_HZ_FULL=y, running the following tests:
> 
> runqemu kvm slirp nographic qemuparams="-m 1024 -smp 4" bootparams=
> "console=ttyS0 nohz_full=0,1 rcu_nocbs=0,1 sched_verbose" -d

Has this ever worked? Again I'm tempted to just:

    git revert 08ae95f4fd3b38b257f5dc7e6507e071c27ba0d5

> 
> root@qemux86-64:~# echo 0 > /sys/devices/system/cpu/cpu2/online
> root@qemux86-64:~# echo 0 > /sys/devices/system/cpu/cpu3/online
> 
> [   22.838290] BUG: unable to handle page fault for address: ffffffff84cd48c0
  

On Thu, Feb 9, 2023 at 6:55 PM Zqiang <qiang1.zhang@intel.com> wrote:
>
> For kernels built with CONFIG_NO_HZ_FULL=y, running the following tests:
>
[...]
> This commit therefore add checks for cpumask_any_and() return values
> in housekeeping_any_cpu(), if cpumask_any_and() returns an illegal CPU
> value, the housekeeping_any_cpu() will return current CPU number.
>
> Signed-off-by: Zqiang <qiang1.zhang@intel.com>
> ---
>  kernel/sched/isolation.c | 6 +++++-
>  1 file changed, 5 insertions(+), 1 deletion(-)
>
> diff --git a/kernel/sched/isolation.c b/kernel/sched/isolation.c
> index 373d42c707bc..534397ab7a36 100644
> --- a/kernel/sched/isolation.c
> +++ b/kernel/sched/isolation.c
> @@ -46,7 +46,11 @@ int housekeeping_any_cpu(enum hk_type type)
>                         if (cpu < nr_cpu_ids)
>                                 return cpu;
>
> -                       return cpumask_any_and(housekeeping.cpumasks[type], cpu_online_mask);
> +                       cpu = cpumask_any_and(housekeeping.cpumasks[type], cpu_online_mask);
> +                       if (cpu >= nr_cpu_ids)
> +                               return smp_processor_id();
> +                       else
> +                               return cpu;

Nit: no need of "else", simply:

return (cpu >= nr_cpuids ? smp_processor_id() : cpu);

or

if (cpu >= nr_cpu_ids)
   return smp_processor_id();
return cpu;

Thanks.
  

>
> For kernels built with CONFIG_NO_HZ_FULL=y, running the following tests:
>
[...]
> This commit therefore add checks for cpumask_any_and() return values
> in housekeeping_any_cpu(), if cpumask_any_and() returns an illegal CPU
> value, the housekeeping_any_cpu() will return current CPU number.
>
> Signed-off-by: Zqiang <qiang1.zhang@intel.com>
> ---
>  kernel/sched/isolation.c | 6 +++++-
>  1 file changed, 5 insertions(+), 1 deletion(-)
>
> diff --git a/kernel/sched/isolation.c b/kernel/sched/isolation.c
> index 373d42c707bc..534397ab7a36 100644
> --- a/kernel/sched/isolation.c
> +++ b/kernel/sched/isolation.c
> @@ -46,7 +46,11 @@ int housekeeping_any_cpu(enum hk_type type)
>                         if (cpu < nr_cpu_ids)
>                                 return cpu;
>
> -                       return cpumask_any_and(housekeeping.cpumasks[type], cpu_online_mask);
> +                       cpu = cpumask_any_and(housekeeping.cpumasks[type], cpu_online_mask);
> +                       if (cpu >= nr_cpu_ids)
> +                               return smp_processor_id();
> +                       else
> +                               return cpu;
>
>Nit: no need of "else", simply:
>
>return (cpu >= nr_cpuids ? smp_processor_id() : cpu);

Thanks Joel, I have done the same modification in v3 version,  and this change is
also related to a 08ae95f4fd3b, after revert it, this calltrace will disappear.

Thanks
Zqiang


>
>or
>
>if (cpu >= nr_cpu_ids)
>   return smp_processor_id();
>return cpu;
>
>Thanks.