Messages in this thread | | | From | Bryan Wu <> | Date | Wed, 4 Jul 2012 15:29:14 +0800 | Subject | Re: [PATCH] leds: heartbeat: fix bug on panic |
| |
On Wed, Jul 4, 2012 at 3:11 PM, Alexander Holler <holler@ahsoftware.de> wrote: > Am 04.07.2012 09:05, schrieb Bryan Wu: > >> On Tue, Jul 3, 2012 at 2:35 PM, Alexander Holler <holler@ahsoftware.de> >> wrote: >>> >>> With commit 49dca5aebfdeadd4bf27b6cb4c60392147dc35a4 I introduced >>> a bug (visible if CONFIG_PROVE_RCU is enabled) which occures when a panic >>> has happened: >>> >>> [ 1526.520230] =============================== >>> [ 1526.520230] [ INFO: suspicious RCU usage. ] >>> [ 1526.520230] 3.5.0-rc1+ #12 Not tainted >>> [ 1526.520230] ------------------------------- >>> [ 1526.520230] /c/kernel-tests/mm/include/linux/rcupdate.h:436 Illegal >>> context switch in RCU read-side critical section! >>> [ 1526.520230] >>> [ 1526.520230] other info that might help us debug this: >>> [ 1526.520230] >>> [ 1526.520230] >>> [ 1526.520230] rcu_scheduler_active = 1, debug_locks = 0 >>> [ 1526.520230] 3 locks held by net.agent/3279: >>> [ 1526.520230] #0: (&mm->mmap_sem){++++++}, at: [<ffffffff82f85962>] >>> do_page_fault+0x193/0x390 >>> [ 1526.520230] #1: (panic_lock){+.+...}, at: [<ffffffff82ed2830>] >>> panic+0x37/0x1d3 >>> [ 1526.520230] #2: (rcu_read_lock){.+.+..}, at: [<ffffffff810b9b28>] >>> rcu_lock_acquire+0x0/0x29 >>> [ 1526.520230] >>> [ 1526.520230] stack backtrace: >>> [ 1526.520230] Pid: 3279, comm: net.agent Not tainted 3.5.0-rc1+ #12 >>> [ 1526.520230] Call Trace: >>> [ 1526.520230] [<ffffffff810e1570>] lockdep_rcu_suspicious+0x109/0x112 >>> [ 1526.520230] [<ffffffff810bfe3a>] rcu_preempt_sleep_check+0x45/0x47 >>> [ 1526.520230] [<ffffffff810bfe5a>] __might_sleep+0x1e/0x19a >>> [ 1526.520230] [<ffffffff82f8010e>] down_write+0x26/0x81 >>> [ 1526.520230] [<ffffffff8276a966>] led_trigger_unregister+0x1f/0x9c >>> [ 1526.520230] [<ffffffff8276def5>] heartbeat_reboot_notifier+0x15/0x19 >>> [ 1526.520230] [<ffffffff82f85bf5>] notifier_call_chain+0x96/0xcd >>> [ 1526.520230] [<ffffffff82f85cba>] >>> __atomic_notifier_call_chain+0x8e/0xff >>> [ 1526.520230] [<ffffffff81094b7c>] ? kmsg_dump+0x37/0x1eb >>> [ 1526.520230] [<ffffffff82f85d3f>] atomic_notifier_call_chain+0x14/0x16 >>> [ 1526.520230] [<ffffffff82ed28e1>] panic+0xe8/0x1d3 >>> [ 1526.520230] [<ffffffff811473e2>] out_of_memory+0x15d/0x1d3 >>> >>> So in case of a panic, now just turn of the LED. Other approaches like >>> scheduling a work to unregister the trigger aren't working because there >>> isn't much which still runs after a panic occured (except timers). >>> >>> Signed-off-by: Alexander Holler <holler@ahsoftware.de> >>> --- >>> drivers/leds/ledtrig-heartbeat.c | 16 +++++++++++++++- >>> 1 files changed, 15 insertions(+), 1 deletions(-) >>> >>> diff --git a/drivers/leds/ledtrig-heartbeat.c >>> b/drivers/leds/ledtrig-heartbeat.c >>> index 41dc76d..a019fbb 100644 >>> --- a/drivers/leds/ledtrig-heartbeat.c >>> +++ b/drivers/leds/ledtrig-heartbeat.c >>> @@ -21,6 +21,8 @@ >>> #include <linux/reboot.h> >>> #include "leds.h" >>> >>> +static int panic_heartbeats; >>> + >>> struct heartbeat_trig_data { >>> unsigned int phase; >>> unsigned int period; >>> @@ -34,6 +36,11 @@ static void led_heartbeat_function(unsigned long data) >>> unsigned long brightness = LED_OFF; >>> unsigned long delay = 0; >>> >>> + if (unlikely(panic_heartbeats)) { >>> + led_set_brightness(led_cdev, LED_OFF); >>> + return; >>> + } >>> + >>> /* acts like an actual heart beat -- ie thump-thump-pause... */ >>> switch (heartbeat_data->phase) { >>> case 0: >>> @@ -111,12 +118,19 @@ static int heartbeat_reboot_notifier(struct >>> notifier_block *nb, >>> return NOTIFY_DONE; >>> } >>> >>> +static int heartbeat_panic_notifier(struct notifier_block *nb, >>> + unsigned long code, void *unused) >>> +{ >>> + panic_heartbeats = 1; >> >> >> Can we just set LED as OFF and delete the timer here? because timer is >> also useless after a kernel panic. >> So we don't need this global static variable here. > > > No, the necessary information (heartbeat_trig_data) isn't available here. >
Yeah, looks like there is no way to pass heartbeat_trig_data information to the notifier call function. Anyway, I will apply this patch to my for-next branch.
Thanks, -- Bryan Wu <bryan.wu@canonical.com> Kernel Developer +86.186-168-78255 Mobile Canonical Ltd. www.canonical.com Ubuntu - Linux for human beings | www.ubuntu.com
| |