x86/reboot: Fix a warning message triggered by stop_other_cpus()
Feng Tang [Wed, 30 May 2012 15:15:41 +0000 (23:15 +0800)]
When rebooting our 24 CPU Westmere servers with 3.4-rc6, we
always see this warning msg:

Restarting system.
machine restart
------------[ cut here ]------------
WARNING: at arch/x86/kernel/smp.c:125
native_smp_send_reschedule+0x74/0xa7() Hardware name: X8DTN
Modules linked in: igb [last unloaded: scsi_wait_scan]
Pid: 1, comm: systemd-shutdow Not tainted 3.4.0-rc6+ #22
Call Trace:
 <IRQ>  [<ffffffff8102a41f>] warn_slowpath_common+0x7e/0x96
 [<ffffffff8102a44c>] warn_slowpath_null+0x15/0x17
 [<ffffffff81018cf7>] native_smp_send_reschedule+0x74/0xa7
 [<ffffffff810561c1>] trigger_load_balance+0x279/0x2a6
 [<ffffffff81050112>] scheduler_tick+0xe0/0xe9
 [<ffffffff81036768>] update_process_times+0x60/0x70
 [<ffffffff81062f2f>] tick_sched_timer+0x68/0x92
 [<ffffffff81046e33>] __run_hrtimer+0xb3/0x13c
 [<ffffffff81062ec7>] ? tick_nohz_handler+0xd0/0xd0
 [<ffffffff810474f2>] hrtimer_interrupt+0xdb/0x198
 [<ffffffff81019a35>] smp_apic_timer_interrupt+0x81/0x94
 [<ffffffff81655187>] apic_timer_interrupt+0x67/0x70
 <EOI>  [<ffffffff8101a3c4>] ? default_send_IPI_mask_allbutself_phys+0xb4/0xc4
 [<ffffffff8101c680>] physflat_send_IPI_allbutself+0x12/0x14
 [<ffffffff81018db4>] native_nmi_stop_other_cpus+0x8a/0xd6
 [<ffffffff810188ba>] native_machine_shutdown+0x50/0x67
 [<ffffffff81018926>] machine_shutdown+0xa/0xc
 [<ffffffff8101897e>] native_machine_restart+0x20/0x32
 [<ffffffff810189b0>] machine_restart+0xa/0xc
 [<ffffffff8103b196>] kernel_restart+0x47/0x4c
 [<ffffffff8103b2e6>] sys_reboot+0x13e/0x17c
 [<ffffffff8164e436>] ? _raw_spin_unlock_bh+0x10/0x12
 [<ffffffff810fcac9>] ? bdi_queue_work+0xcf/0xd8
 [<ffffffff810fe82f>] ? __bdi_start_writeback+0xae/0xb7
 [<ffffffff810e0d64>] ? iterate_supers+0xa3/0xb7
 [<ffffffff816547a2>] system_call_fastpath+0x16/0x1b
---[ end trace 320af5cb1cb60c5b ]---

The root cause seems to be the
default_send_IPI_mask_allbutself_phys() takes quite some time (I
measured it could be several ms) to complete sending NMIs to all
the other 23 CPUs, and for HZ=250/1000 system, the time is long
enough for a timer interrupt to happen, which will in turn
trigger to kick load balance to a stopped CPU and cause this
warning in native_smp_send_reschedule().

So disabling the local irq before stop_other_cpu() can fix this
problem (tested 25 times reboot ok), and it is fine as there
should be nobody caring the timer interrupt in such reboot

The latest 3.4 kernel slightly changes this behavior by sending
REBOOT_VECTOR first and only send NMI_VECTOR if the REBOOT_VCTOR
fails, and this patch is still needed to prevent the problem.

Signed-off-by: Feng Tang <feng.tang@intel.com>
Acked-by: Don Zickus <dzickus@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20120530231541.4c13433a@feng-i7
Signed-off-by: Ingo Molnar <mingo@kernel.org>


index 79c45af..25b48ed 100644 (file)
@@ -639,9 +639,11 @@ void native_machine_shutdown(void)
        set_cpus_allowed_ptr(current, cpumask_of(reboot_cpu_id));
-        * O.K Now that I'm on the appropriate processor,
-        * stop all of the others.
+        * O.K Now that I'm on the appropriate processor, stop all of the
+        * others. Also disable the local irq to not receive the per-cpu
+        * timer interrupt which may trigger scheduler's load balance.
+       local_irq_disable();