2023-02-10 00:37:36

by Zqiang

[permalink] [raw]
Subject: [PATCH v3] 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 <[email protected]>
---
kernel/sched/isolation.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/kernel/sched/isolation.c b/kernel/sched/isolation.c
index 373d42c707bc..edfba557a2e1 100644
--- a/kernel/sched/isolation.c
+++ b/kernel/sched/isolation.c
@@ -46,7 +46,8 @@ 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);
+ return (cpu >= nr_cpu_ids) ? smp_processor_id() : cpu;
}
}
return smp_processor_id();
--
2.25.1



2023-02-11 08:28:58

by Qiuxu Zhuo

[permalink] [raw]
Subject: RE: [PATCH v3] sched/isolation: Fix illegal CPU value by housekeeping_any_cpu() return

> From: Zqiang <[email protected]>
> Sent: Friday, February 10, 2023 8:40 AM
> To: [email protected]; [email protected]; [email protected];
> [email protected]; [email protected]; [email protected];
> [email protected]; [email protected]
> Subject: [PATCH v3] 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

Hi Qiang,

Did some quick testing using the same kernel parameters and the reproducing steps as yours:

1) If not apply this v3, the kernel was panic like you found.

2) If apply this v3, the kernel did NOT panic and worked well.
But a WARNING call trace [1] was thrown.
Not sure whether [1] was another issue.

[1]
[ 2445.396928] smpboot: CPU 2 is now offline
[ 2445.399084] CPU2 attaching NULL sched-domain.
[ 2445.399091] CPU3 attaching NULL sched-domain.
[ 2445.399202] CPU3 attaching NULL sched-domain.
[ 2445.399208] root domain span: 3 (max cpu_capacity = 1024)
[ 2449.731424] process 672 (tuned) no longer affine to cpu3
[ 2449.733332] process 509 (systemd-journal) no longer affine to cpu3
[ 2449.742278] process 541 (systemd-udevd) no longer affine to cpu3
[ 2449.745409] process 760 (bash) no longer affine to cpu3
[ 2449.748550] smpboot: CPU 3 is now offline
[ 2449.755129] CPU3 attaching NULL sched-domain.
[ 2449.755194] ------------[ cut here ]------------
[ 2449.756296] WARNING: CPU: 0 PID: 483 at kernel/sched/topology.c:2257 build_sched_domains+0x104c/0x1430
[ 2449.758227] Modules linked in: rfkill sunrpc psmouse i2c_piix4 atkbd libps2 vivaldi_fmap serio_raw virtio_net net_failover failover sr_mod cdrom i8042 qemu_fw_cfg pata_acpi ipmi_devintf ipmi_msghandler
[ 2449.760804] CPU: 0 PID: 483 Comm: kworker/3:6 Not tainted 6.2.0-rc7-rcu+ #21
[ 2449.761820] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 0.0.0 02/06/2015
[ 2449.762931] Workqueue: events cpuset_hotplug_workfn
[ 2449.763676] RIP: 0010:build_sched_domains+0x104c/0x1430
[ 2449.764465] Code: 45 98 f4 ff ff ff 0f 84 1a f8 ff ff 48 8b 7d 90 31 f6 e8 17 48 ff ff e9 0a f8 ff ff 0f 0b e9 01 fe ff ff 0f 0b e9 b6 fb ff ff <0f> 0b c7 45 98 f4 ff ff ff e9 3f f7 ff ff 48 c7 45 90 00 00 00 00
[ 2449.766934] RSP: 0000:ffffab51c08f7c00 EFLAGS: 00010246
[ 2449.767568] process 737 (tuned) no longer affine to cpu3
[ 2449.768378] RAX: 0000000000000004 RBX: 0000000000000004 RCX: 0000000000000000
[ 2449.769079] RDX: 0000000000000040 RSI: 0000000000000004 RDI: ffff9486442d7f08
[ 2449.769785] RBP: ffffab51c08f7ca0 R08: 0000000000000000 R09: 0000000000000000
[ 2449.770501] R10: 0000000000000190 R11: ffffab51c08f7ab8 R12: 0000000000000001
[ 2449.771227] R13: 0000000000000000 R14: ffff9486424379c0 R15: 0000000000000001
[ 2449.771920] FS: 0000000000000000(0000) GS:ffff948777c00000(0000) knlGS:0000000000000000
[ 2449.772714] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[ 2449.773303] CR2: 000055ed0e0ed158 CR3: 000000010091a002 CR4: 0000000000370ef0
[ 2449.774011] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[ 2449.774725] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
[ 2449.775437] Call Trace:
[ 2449.775752] <TASK>
[ 2449.776053] ? cpu_attach_domain+0x3d7/0x810
[ 2449.776532] ? wait_for_completion+0xff/0x110
[ 2449.777015] partition_sched_domains_locked+0x1e7/0x3a0
[ 2449.777554] rebuild_sched_domains_locked+0x545/0x800
[ 2449.778032] ? rcu_sync_enter+0x6b/0xc0
[ 2449.778377] rebuild_sched_domains+0x1a/0x40
[ 2449.778728] cpuset_hotplug_workfn+0x18a/0xe10
[ 2449.779105] ? balance_push+0x51/0x110
[ 2449.779444] ? finish_task_switch+0x85/0x2c0
[ 2449.779810] ? __schedule+0x2f7/0x9f0
[ 2449.780134] process_one_work+0x1cd/0x3e0
[ 2449.780495] worker_thread+0x32/0x380
[ 2449.781436] ? process_one_work+0x3e0/0x3e0
[ 2449.782006] kthread+0xe8/0x110
[ 2449.782478] ? kthread_complete_and_exit+0x20/0x20
[ 2449.783067] ret_from_fork+0x1f/0x30
[ 2449.783566] </TASK>
[ 2449.783953] ---[ end trace 0000000000000000 ]---
[ 2449.789269] process 741 (tuned) no longer affine to cpu3
[ 2449.794191] process 759 (sshd) no longer affine to cpu3
[ 2450.188215] process 732 (in:imjournal) no longer affine to cpu3
[ 2450.188457] process 733 (rs:main Q:Reg) no longer affine to cpu3
[ 2453.011183] process 659 (gmain) no longer affine to cpu3
[ 2465.517178] select_fallback_rq: 1 callbacks suppressed
[ 2465.517185] process 605 (rpcbind) no longer affine to cpu3
[ 2479.794154] process 652 (chronyd) no longer affine to cpu2
...

2023-02-11 11:21:39

by Zqiang

[permalink] [raw]
Subject: RE: [PATCH v3] sched/isolation: Fix illegal CPU value by housekeeping_any_cpu() return


> From: Zqiang <[email protected]>
> Sent: Friday, February 10, 2023 8:40 AM
> To: [email protected]; [email protected]; [email protected];
> [email protected]; [email protected]; [email protected];
> [email protected]; [email protected]
> Subject: [PATCH v3] 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
>
>Hi Qiang,
>
>Did some quick testing using the same kernel parameters and the reproducing steps as yours:
>
>1) If not apply this v3, the kernel was panic like you found.
>
>2) If apply this v3, the kernel did NOT panic and worked well.
> But a WARNING call trace [1] was thrown.
> Not sure whether [1] was another issue.

Thanks for testing, yes this is another issue, if enable CONFIG_CPUSETS options
will trigger follow calltrace.

Thanks
Zqiang


>
>[1]
>[ 2445.396928] smpboot: CPU 2 is now offline
>[ 2445.399084] CPU2 attaching NULL sched-domain.
[ 2445.399091] CPU3 attaching NULL sched-domain.
[ 2445.399202] CPU3 attaching NULL sched-domain.
[ 2445.399208] root domain span: 3 (max cpu_capacity = 1024)
[ 2449.731424] process 672 (tuned) no longer affine to cpu3
[ 2449.733332] process 509 (systemd-journal) no longer affine to cpu3
[ 2449.742278] process 541 (systemd-udevd) no longer affine to cpu3
[ 2449.745409] process 760 (bash) no longer affine to cpu3
[ 2449.748550] smpboot: CPU 3 is now offline
[ 2449.755129] CPU3 attaching NULL sched-domain.
[ 2449.755194] ------------[ cut here ]------------
[ 2449.756296] WARNING: CPU: 0 PID: 483 at kernel/sched/topology.c:2257 build_sched_domains+0x104c/0x1430
[ 2449.758227] Modules linked in: rfkill sunrpc psmouse i2c_piix4 atkbd libps2 vivaldi_fmap serio_raw virtio_net net_failover failover sr_mod cdrom i8042 qemu_fw_cfg pata_acpi ipmi_devintf ipmi_msghandler
[ 2449.760804] CPU: 0 PID: 483 Comm: kworker/3:6 Not tainted 6.2.0-rc7-rcu+ #21
[ 2449.761820] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 0.0.0 02/06/2015
[ 2449.762931] Workqueue: events cpuset_hotplug_workfn
[ 2449.763676] RIP: 0010:build_sched_domains+0x104c/0x1430
[ 2449.764465] Code: 45 98 f4 ff ff ff 0f 84 1a f8 ff ff 48 8b 7d 90 31 f6 e8 17 48 ff ff e9 0a f8 ff ff 0f 0b e9 01 fe ff ff 0f 0b e9 b6 fb ff ff <0f> 0b c7 45 98 f4 ff ff ff e9 3f f7 ff ff 48 c7 45 90 00 00 00 00
[ 2449.766934] RSP: 0000:ffffab51c08f7c00 EFLAGS: 00010246
[ 2449.767568] process 737 (tuned) no longer affine to cpu3
[ 2449.768378] RAX: 0000000000000004 RBX: 0000000000000004 RCX: 0000000000000000
[ 2449.769079] RDX: 0000000000000040 RSI: 0000000000000004 RDI: ffff9486442d7f08
[ 2449.769785] RBP: ffffab51c08f7ca0 R08: 0000000000000000 R09: 0000000000000000
[ 2449.770501] R10: 0000000000000190 R11: ffffab51c08f7ab8 R12: 0000000000000001
[ 2449.771227] R13: 0000000000000000 R14: ffff9486424379c0 R15: 0000000000000001
[ 2449.771920] FS: 0000000000000000(0000) GS:ffff948777c00000(0000) knlGS:0000000000000000
[ 2449.772714] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[ 2449.773303] CR2: 000055ed0e0ed158 CR3: 000000010091a002 CR4: 0000000000370ef0
[ 2449.774011] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[ 2449.774725] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
[ 2449.775437] Call Trace:
[ 2449.775752] <TASK>
[ 2449.776053] ? cpu_attach_domain+0x3d7/0x810
[ 2449.776532] ? wait_for_completion+0xff/0x110
[ 2449.777015] partition_sched_domains_locked+0x1e7/0x3a0
[ 2449.777554] rebuild_sched_domains_locked+0x545/0x800
[ 2449.778032] ? rcu_sync_enter+0x6b/0xc0
[ 2449.778377] rebuild_sched_domains+0x1a/0x40
[ 2449.778728] cpuset_hotplug_workfn+0x18a/0xe10
[ 2449.779105] ? balance_push+0x51/0x110
[ 2449.779444] ? finish_task_switch+0x85/0x2c0
[ 2449.779810] ? __schedule+0x2f7/0x9f0
[ 2449.780134] process_one_work+0x1cd/0x3e0
[ 2449.780495] worker_thread+0x32/0x380
[ 2449.781436] ? process_one_work+0x3e0/0x3e0
[ 2449.782006] kthread+0xe8/0x110
[ 2449.782478] ? kthread_complete_and_exit+0x20/0x20
[ 2449.783067] ret_from_fork+0x1f/0x30
[ 2449.783566] </TASK>
[ 2449.783953] ---[ end trace 0000000000000000 ]---
[ 2449.789269] process 741 (tuned) no longer affine to cpu3
[ 2449.794191] process 759 (sshd) no longer affine to cpu3
[ 2450.188215] process 732 (in:imjournal) no longer affine to cpu3
[ 2450.188457] process 733 (rs:main Q:Reg) no longer affine to cpu3
[ 2453.011183] process 659 (gmain) no longer affine to cpu3
[ 2465.517178] select_fallback_rq: 1 callbacks suppressed
[ 2465.517185] process 605 (rpcbind) no longer affine to cpu3
[ 2479.794154] process 652 (chronyd) no longer affine to cpu2
...