]> git.baikalelectronics.ru Git - kernel.git/commit
nohz: Restore NMI safe local irq work for local nohz kick
authorFrederic Weisbecker <fweisbec@gmail.com>
Wed, 13 Aug 2014 16:50:16 +0000 (18:50 +0200)
committerFrederic Weisbecker <fweisbec@gmail.com>
Thu, 4 Sep 2014 20:35:59 +0000 (22:35 +0200)
commit8c705c0016129f772a9b9a02d3ba24a09365dc84
treea0a62cada1b7b74ab8771b50fec2b8902cef07e6
parent2e5c8ca83cd8c7326618ad75fd27258ed226d7b9
nohz: Restore NMI safe local irq work for local nohz kick

The local nohz kick is currently used by perf which needs it to be
NMI-safe. Recent commit though (f45f9587f1dd1e89dd1c89561c9572f777c5bcb7)
changed its implementation to fire the local kick using the remote kick
API. It was convenient to make the code more generic but the remote kick
isn't NMI-safe.

As a result:

WARNING: CPU: 3 PID: 18062 at kernel/irq_work.c:72 irq_work_queue_on+0x11e/0x140()
CPU: 3 PID: 18062 Comm: trinity-subchil Not tainted 3.16.0+ #34
0000000000000009 00000000903774d1 ffff880244e06c00 ffffffff9a7f1e37
0000000000000000 ffff880244e06c38 ffffffff9a0791dd ffff880244fce180
0000000000000003 ffff880244e06d58 ffff880244e06ef8 0000000000000000
Call Trace:
<NMI>  [<ffffffff9a7f1e37>] dump_stack+0x4e/0x7a
[<ffffffff9a0791dd>] warn_slowpath_common+0x7d/0xa0
[<ffffffff9a07930a>] warn_slowpath_null+0x1a/0x20
[<ffffffff9a17ca1e>] irq_work_queue_on+0x11e/0x140
[<ffffffff9a10a2c7>] tick_nohz_full_kick_cpu+0x57/0x90
[<ffffffff9a186cd5>] __perf_event_overflow+0x275/0x350
[<ffffffff9a184f80>] ? perf_event_task_disable+0xa0/0xa0
[<ffffffff9a01a4cf>] ? x86_perf_event_set_period+0xbf/0x150
[<ffffffff9a187934>] perf_event_overflow+0x14/0x20
[<ffffffff9a020386>] intel_pmu_handle_irq+0x206/0x410
[<ffffffff9a0b54d3>] ? arch_vtime_task_switch+0x63/0x130
[<ffffffff9a01937b>] perf_event_nmi_handler+0x2b/0x50
[<ffffffff9a007b72>] nmi_handle+0xd2/0x390
[<ffffffff9a007aa5>] ? nmi_handle+0x5/0x390
[<ffffffff9a0d131b>] ? lock_release+0xab/0x330
[<ffffffff9a008062>] default_do_nmi+0x72/0x1c0
[<ffffffff9a0c925f>] ? cpuacct_account_field+0xcf/0x200
[<ffffffff9a008268>] do_nmi+0xb8/0x100

Lets fix this by restoring the use of local irq work for the nohz local
kick.

Reported-by: Catalin Iacob <iacobcatalin@gmail.com>
Reported-and-tested-by: Dave Jones <davej@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
include/linux/tick.h
kernel/time/tick-sched.c