2013-03-06 19:07:45

by Seiji Aguchi

[permalink] [raw]
Subject: [PATCH]Skip unnecessary WARN_ON in panic case

[Problem]

When kernel panics, unnecessary WARN_ON() may be printed after panic messages in a following scenario.
- A panicked cpu stops other cpus via smp_send_stop().
- Other cpus turn to be offline in stop_this_cpu().
- The panicked cpu enables interrupt.
- native_smp_send_reschedule() is called via a timer interrupt on the panicked cpu.
- The panicked cpu tries to send a reschedule IPI to other cpus
- The panicked cpu hits WARN_ON() because other cpus have already been offlined.

If an user has just a VGA console, panic messages may be missed because they are floated outside a screen
due to messages of the WARN_ON().
In this case, it could be difficult to investigate the reason why a kernel panicked.

Here is an actual result of the scenario above.

SysRq : Trigger a crash
BUG: unable to handle kernel NULL pointer dereference at (null)
IP: [<ffffffff81346056>] sysrq_handle_crash+0x16/0x20
PGD 127f0c067 PUD 11c7fd067 PMD 0
Oops: 0002 [#1] SMP
Modules linked in: ebtable_nat ebtables xt_CHECKSUM iptable_mangle bridge autofs4 sunrpc 8021q garp stp llc cpufreq_ondemand ipt_REJECT nf_conntrack_ipv4 nf_defrag_ipv4 iptable_filter ip_tables ip6t_REJECT nf_conntrack_ipv6 nf_defrag_ipv6 xt_state nf_conntrack ip6table_filter ip6_tables ipv6 vfat fat vhost_net macvtap macvlan tun uinput sg iTCO_wdt iTCO_vendor_support dcdbas acpi_cpufreq freq_table mperf coretemp kvm_intel kvm crc32c_intel microcode pcspkr i7core_edac edac_core lpc_ich mfd_core bnx2 ext4(F) mbcache(F) jbd2(F) sr_mod(F) cdrom(F) sd_mod(F) crc_t10dif(F) pata_acpi(F) ata_generic(F) ata_piix(F) dm_mirror(F) dm_region_hash(F) dm_log(F) dm_mod(F)
CPU 2
Pid: 4225, comm: bash Tainted: GF 3.9.0-rc1+ #9 Dell Inc. PowerEdge T310/02P9X9
RIP: 0010:[<ffffffff81346056>] [<ffffffff81346056>] sysrq_handle_crash+0x16/0x20
RSP: 0018:ffff8801298b1e18 EFLAGS: 00010096
RAX: 000000000000000f RBX: 0000000000000063 RCX: ffff88013f24fb10
RDX: 0000000000000000 RSI: ffff88013f24df08 RDI: 0000000000000063
RBP: ffff8801298b1e18 R08: 0000000000000003 R09: 00000000000115e4
R10: 0000000000000371 R11: 0000000000000372 R12: ffffffff81aa5a40
R13: 0000000000000286 R14: 0000000000000007 R15: 0000000000000000
FS: 00007fcaed1b1700(0000) GS:ffff88013f240000(0000) knlGS:0000000000000000
CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 0000000000000000 CR3: 0000000128d91000 CR4: 00000000000007e0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 0000000000000400
Process bash (pid: 4225, threadinfo ffff8801298b0000, task ffff880128d15580)
Stack:
ffff8801298b1e58 ffffffff81346969 ffff8801298b1f38 0000000000000002
ffff880129605e80 00007fcaed1c2000 0000000000000002 fffffffffffffffb
ffff8801298b1e88 ffffffff81346a1a ffff8801298b1eb8 00007fcaed1c2000
Call Trace:
[<ffffffff81346969>] __handle_sysrq+0x129/0x190
[<ffffffff81346a1a>] write_sysrq_trigger+0x4a/0x50
[<ffffffff811e4cb9>] proc_reg_write+0x79/0xb0
[<ffffffff8117f104>] vfs_write+0xb4/0x130
[<ffffffff8117f8cf>] sys_write+0x5f/0xa0
[<ffffffff81570a19>] system_call_fastpath+0x16/0x1b
Code: 48 81 c7 08 08 00 00 e8 c9 1b 22 00 31 c0 e9 62 ff ff ff 90 90 55 48 89 e5 66 66 66 66 90 c7 05 cd 02 9f 00 01 00 00 00 0f ae f8 <c6> 04 25 00 00 00 00 01 c9 c3 55 48 89 e5 66 66 66 66 90 8d 47
RIP [<ffffffff81346056>] sysrq_handle_crash+0x16/0x20
RSP <ffff8801298b1e18>
CR2: 0000000000000000
---[ end trace b3d5243c59d80623 ]---
Kernel panic - not syncing: Fatal exception
------------[ cut here ]------------
WARNING: at arch/x86/kernel/smp.c:123 native_smp_send_reschedule+0x5c/0x60()
Hardware name: PowerEdge T310
Modules linked in: ebtable_nat ebtables xt_CHECKSUM iptable_mangle bridge autofs4 sunrpc 8021q garp stp llc cpufreq_ondemand ipt_REJECT nf_conntrack_ipv4 nf_defrag_ipv4 iptable_filter ip_tables ip6t_REJECT nf_conntrack_ipv6 nf_defrag_ipv6 xt_state nf_conntrack ip6table_filter ip6_tables ipv6 vfat fat vhost_net macvtap macvlan tun uinput sg iTCO_wdt iTCO_vendor_support dcdbas acpi_cpufreq freq_table mperf coretemp kvm_intel kvm crc32c_intel microcode pcspkr i7core_edac edac_core lpc_ich mfd_core bnx2 ext4(F) mbcache(F) jbd2(F) sr_mod(F) cdrom(F) sd_mod(F) crc_t10dif(F) pata_acpi(F) ata_generic(F) ata_piix(F) dm_mirror(F) dm_region_hash(F) dm_log(F) dm_mod(F)
Pid: 4225, comm: bash Tainted: GF D 3.9.0-rc1+ #9
Call Trace:
<IRQ> [<ffffffff81056daf>] warn_slowpath_common+0x7f/0xc0
[<ffffffff81056e0a>] warn_slowpath_null+0x1a/0x20
[<ffffffff8103918c>] native_smp_send_reschedule+0x5c/0x60
[<ffffffff81092976>] trigger_load_balance+0x1c6/0x240
[<ffffffff81089fff>] scheduler_tick+0x10f/0x140
[<ffffffff81066f29>] update_process_times+0x69/0x80
[<ffffffff810b1859>] tick_sched_handle+0x39/0x80
[<ffffffff810b1a84>] tick_sched_timer+0x54/0x90
[<ffffffff8107f1a3>] __run_hrtimer+0x83/0x1d0
[<ffffffff810b1a30>] ? tick_nohz_handler+0xc0/0xc0
[<ffffffff8107f576>] hrtimer_interrupt+0xf6/0x240
[<ffffffff810435d6>] hpet_interrupt_handler+0x16/0x40
[<ffffffff810e0f8d>] handle_irq_event_percpu+0x6d/0x210
[<ffffffff8105f745>] ? __do_softirq+0x165/0x260
[<ffffffff810e1172>] handle_irq_event+0x42/0x70
[<ffffffff810e48c9>] handle_edge_irq+0x69/0x120
[<ffffffff810164ac>] handle_irq+0x5c/0x150
[<ffffffff8105f57b>] ? irq_enter+0x1b/0x80
[<ffffffff8157265d>] do_IRQ+0x5d/0xe0
[<ffffffff8156822d>] common_interrupt+0x6d/0x6d
<EOI> [<ffffffff815649d0>] ? panic+0x19c/0x1e4
[<ffffffff8156492d>] ? panic+0xf9/0x1e4
[<ffffffff81569114>] oops_end+0xe4/0x100
[<ffffffff8104aa7e>] no_context+0x11e/0x1f0
[<ffffffff8104ac7d>] __bad_area_nosemaphore+0x12d/0x230
[<ffffffff8104adee>] bad_area+0x4e/0x60
[<ffffffff8156c04e>] __do_page_fault+0x43e/0x490
[<ffffffff81058b33>] ? call_console_drivers.clone.3+0xa3/0x110
[<ffffffff81080f4f>] ? up+0x2f/0x50
[<ffffffff81058404>] ? wake_up_klogd+0x34/0x40
[<ffffffff81058f9d>] ? console_unlock+0x25d/0x290
[<ffffffff8156c0ae>] do_page_fault+0xe/0x10
[<ffffffff81568518>] page_fault+0x28/0x30
[<ffffffff81346056>] ? sysrq_handle_crash+0x16/0x20
[<ffffffff81346969>] __handle_sysrq+0x129/0x190
[<ffffffff81346a1a>] write_sysrq_trigger+0x4a/0x50
[<ffffffff811e4cb9>] proc_reg_write+0x79/0xb0
[<ffffffff8117f104>] vfs_write+0xb4/0x130
[<ffffffff8117f8cf>] sys_write+0x5f/0xa0
[<ffffffff81570a19>] system_call_fastpath+0x16/0x1b
---[ end trace b3d5243c59d80624 ]---

[Solution]

Skip WARN_ON() when a panicked cpu which is set to stopping_cpu in
smp_send_stop() calls native_smp_send_reschedule().

Signed-off-by: Seiji Aguchi <[email protected]>
---
arch/x86/kernel/smp.c | 6 +++++-
1 files changed, 5 insertions(+), 1 deletions(-)

diff --git a/arch/x86/kernel/smp.c b/arch/x86/kernel/smp.c
index 48d2b7d..35168b1 100644
--- a/arch/x86/kernel/smp.c
+++ b/arch/x86/kernel/smp.c
@@ -120,7 +120,11 @@ static bool smp_no_nmi_ipi = false;
static void native_smp_send_reschedule(int cpu)
{
if (unlikely(cpu_is_offline(cpu))) {
- WARN_ON(1);
+ /*
+ * Skip WARN_ON() if cpu is stopping
+ * to avoid printing spurious messages.
+ */
+ WARN_ON(raw_smp_processor_id() != atomic_read(&stopping_cpu));
return;
}
apic->send_IPI_mask(cpumask_of(cpu), RESCHEDULE_VECTOR);
--
1.7.1





2013-03-11 08:28:57

by Ingo Molnar

[permalink] [raw]
Subject: Re: [PATCH]Skip unnecessary WARN_ON in panic case


* Seiji Aguchi <[email protected]> wrote:

> [Problem]
>
> When kernel panics, unnecessary WARN_ON() may be printed after panic messages in a following scenario.
> - A panicked cpu stops other cpus via smp_send_stop().
> - Other cpus turn to be offline in stop_this_cpu().
> - The panicked cpu enables interrupt.
> - native_smp_send_reschedule() is called via a timer interrupt on the panicked cpu.
> - The panicked cpu tries to send a reschedule IPI to other cpus
> - The panicked cpu hits WARN_ON() because other cpus have already been offlined.
>
> If an user has just a VGA console, panic messages may be missed because
> they are floated outside a screen due to messages of the WARN_ON(). In
> this case, it could be difficult to investigate the reason why a kernel
> panicked.

So it appears the problem occurs because the other CPUs are not offlined
correctly, and still have pending work which may generate wakeup IPIs to
them?

The fix would be to either offline them properly - or, if we don't want to
do that to keep panicking simple, why do we mark them offline? They aren't
really offline in a proper way.

Thanks,

Ingo

2013-03-15 15:43:17

by Seiji Aguchi

[permalink] [raw]
Subject: RE: [PATCH]Skip unnecessary WARN_ON in panic case

Thank you for giving me the feedback.

>
> So it appears the problem occurs because the other CPUs are not offlined correctly, and still have pending work which may generate
> wakeup IPIs to them?

Right.

>
> The fix would be to either offline them properly - or, if we don't want to do that to keep panicking simple, why do we mark them
> offline? They aren't really offline in a proper way.

To offline cpus properly, it should be done with threads like stop_machine().
But I don't think it is reasonable to kick the threads in panic case.
Therefore, I will make a patch keeping cpus online.
Also, to inform a panicked cpu that other cpus stopped surely,
I will introduce a new cpu_mask like cpu_stopped_mask.

Seiji