Skip to contentSkip to conversation

System panic under load with clockevents_program_event

Bug 15076 Thomas Gleixner 2 participants 4 messages Jan 25, 2010 – Jan 28, 2010 2 messages not in the archive

lore
Share & tools
lore

A message that is not in the archive (sent off-list or not yet received): [email protected]/

A message that is not in the archive (sent off-list or not yet received): [email protected]

#2 lore raw

Switched to email. Please reply to all instead of using the bugzilla interface.

Quoted 2 lines

--- Comment #4 from okias <[email protected]> 2010-01-22 10:17:25 --- and it's regression. Now I work on 2.6.32.3 and no problem.

That's a really weird one. The system is 50 min up and running and out of the blue it crashes in clockevents_program_event(). This function has been called a couple of thousand times before that point.

The only way to crash there is when *dev is pointing into nirwana. dev comes from

int tick_program_event(ktime_t expires, int force)

{
        struct clock_event_device *dev = __get_cpu_var(tick_cpu_device).evtdev;

according to the callchain. At this point nothing fiddles with tick_cpu_device.evtdev, so I suspect some really nasty memory corruption going on.

okias, can you please disable highmem support and verify whether the problem persists ?

Thanks,

tglx

#3 lore raw

Okey, I try without highmem soon.

2010/1/25, Thomas Gleixner <[email protected]>:

Quoted 29 lines

Switched to email. Please reply to all instead of using the bugzilla interface.

Quoted 2 lines

--- Comment #4 from okias <[email protected]> 2010-01-22 10:17:25 --- and it's regression. Now I work on 2.6.32.3 and no problem.

That's a really weird one. The system is 50 min up and running and out of the blue it crashes in clockevents_program_event(). This function has been called a couple of thousand times before that point.

The only way to crash there is when *dev is pointing into nirwana. dev comes from

int tick_program_event(ktime_t expires, int force) {
struct clock_event_device *dev =
__get_cpu_var(tick_cpu_device).evtdev;

according to the callchain. At this point nothing fiddles with tick_cpu_device.evtdev, so I suspect some really nasty memory corruption going on.

okias, can you please disable highmem support and verify whether the problem persists ?

Thanks,

tglx

Signature · 2 lines
Jabber/XMPP: [email protected]
SIP VoIP: sip:[email protected]
#4 lore raw

Lastest git without HIGHMEM look good. If no problems occur, then I try use kernel with HIGHMEM and see what change...

2010/1/25, okias <[email protected]>:

Quoted 38 lines

Okey, I try without highmem soon.

2010/1/25, Thomas Gleixner <[email protected]>:

Quoted 22 lines

Switched to email. Please reply to all instead of using the bugzilla interface.

Quoted 2 lines

--- Comment #4 from okias <[email protected]> 2010-01-22 10:17:25 --- and it's regression. Now I work on 2.6.32.3 and no problem.

That's a really weird one. The system is 50 min up and running and out of the blue it crashes in clockevents_program_event(). This function has been called a couple of thousand times before that point.

The only way to crash there is when *dev is pointing into nirwana. dev comes from

int tick_program_event(ktime_t expires, int force) {
struct clock_event_device *dev =
__get_cpu_var(tick_cpu_device).evtdev;

according to the callchain. At this point nothing fiddles with tick_cpu_device.evtdev, so I suspect some really nasty memory corruption going on.

Quoted 7 lines

okias, can you please disable highmem support and verify whether the problem persists ?

Thanks,

tglx

--
Jabber/XMPP: [email protected]
SIP VoIP: sip:[email protected]

Signature · 2 lines
Jabber/XMPP: [email protected]
SIP VoIP: sip:[email protected]
#5 lore raw

I'm afraid it freezed even without HIGHMEM. I left computer run (compilation KDE 4.3.95) and when I come back, lcd was in dpms mode and led was blinking. So it look like same problem. Maybe without HIGHMEM it takes longer until problem appear.

2010/1/26, okias <[email protected]>:

Quoted 48 lines

Lastest git without HIGHMEM look good. If no problems occur, then I try use kernel with HIGHMEM and see what change...

2010/1/25, okias <[email protected]>:

Quoted 30 lines

Okey, I try without highmem soon.

2010/1/25, Thomas Gleixner <[email protected]>:

Quoted 27 lines

Switched to email. Please reply to all instead of using the bugzilla interface.

Quoted 2 lines

--- Comment #4 from okias <[email protected]> 2010-01-22 10:17:25 --- and it's regression. Now I work on 2.6.32.3 and no problem.

That's a really weird one. The system is 50 min up and running and out of the blue it crashes in clockevents_program_event(). This function has been called a couple of thousand times before that point.

The only way to crash there is when *dev is pointing into nirwana. dev comes from

int tick_program_event(ktime_t expires, int force) {
struct clock_event_device *dev =
__get_cpu_var(tick_cpu_device).evtdev;

according to the callchain. At this point nothing fiddles with tick_cpu_device.evtdev, so I suspect some really nasty memory corruption going on.

okias, can you please disable highmem support and verify whether the problem persists ?

Thanks,

Quoted 8 lines
Quoted 2 lines

tglx

--
Jabber/XMPP: [email protected]
SIP VoIP: sip:[email protected]

--
Jabber/XMPP: [email protected]
SIP VoIP: sip:[email protected]

Signature · 2 lines
Jabber/XMPP: [email protected]
SIP VoIP: sip:[email protected]