Skip to contentSkip to conversation

Linux 2.6.21: pmtmr losing time

DISCUSSION Joerg Sommrey 2 participants 6 messages Apr 30, 2007 – May 14, 2007

lore
Share & tools
lore
#0 lore raw

Hi all,

after switching to 2.6.21 the system clock sporadically loses time on my box (i386, Athlon MP).
It's always around 4.68 seconds and happened 7 times in the last 12 hours. A simple calculation (2 ^ ACPI_PM_MASK / PMTMR_TICKS_PER_SEC = 2 ^ 24 / 3579545 = 4.686968875) shows: There is almost exactly one pmtmr-cycle missing. Could this be caused by a pmtmr-wrap when the system is in a sleep state?

-jo

Signature · 1 line
-rw-r--r-- 1 jo users 62 2007-04-29 22:29 /home/jo/.signature
#1 lore raw
Quoted 9 lines · Joerg Sommrey

Hi all,

after switching to 2.6.21 the system clock sporadically loses time on my box (i386, Athlon MP).
It's always around 4.68 seconds and happened 7 times in the last 12 hours. A simple calculation (2 ^ ACPI_PM_MASK / PMTMR_TICKS_PER_SEC = 2 ^ 24 / 3579545 = 4.686968875) shows: There is almost exactly one pmtmr-cycle missing. Could this be caused by a pmtmr-wrap when the system is in a sleep state?

Hmm, looks like. That's strange we don't sleep 4.68 seconds. Can you provide me the output of /proc/timer_list please ?

tglx

#2 lore raw
Quoted 11 lines · Thomas Gleixner
Quoted 9 lines · Joerg Sommrey

Hi all,

after switching to 2.6.21 the system clock sporadically loses time on my box (i386, Athlon MP).
It's always around 4.68 seconds and happened 7 times in the last 12 hours. A simple calculation (2 ^ ACPI_PM_MASK / PMTMR_TICKS_PER_SEC = 2 ^ 24 / 3579545 = 4.686968875) shows: There is almost exactly one pmtmr-cycle missing. Could this be caused by a pmtmr-wrap when the system is in a sleep state?

Quoted 6 lines

Hmm, looks like. That's strange we don't sleep 4.68 seconds. Can you provide me the output of /proc/timer_list please ?

tglx

Here it is. Maybe this problem is related to the usage of the "experimental" amd76x_pm module?

-jo

Timer List Version: v0.3
HRTIMER_MAX_CLOCK_BASES: 2
now at 249154808025750 nsecs

cpu: 0
clock 0:
.index: 0
.resolution: 1 nsecs
.get_time: ktime_get_real
.offset: 1177701795305563440 nsecs
active timers:
clock 1:
.index: 1
.resolution: 1 nsecs
.get_time: ktime_get
.offset: 0 nsecs
active timers:
#0: <e4a79f00>, tick_sched_timer, S:01

 # expires at 249154808417850 nsecs [in 392100 nsecs]
 #1: <e4a79f00>, it_real_fn, S:01
 # expires at 249154832599897 nsecs [in 24574147 nsecs]
 #2: <e4a79f00>, hrtimer_wakeup, S:01
 # expires at 249154937316640 nsecs [in 129290890 nsecs]
 #3: <e4a79f00>, it_real_fn, S:01
 # expires at 249154940604253 nsecs [in 132578503 nsecs]
 #4: <e4a79f00>, it_real_fn, S:01
 # expires at 249156991584989 nsecs [in 2183559239 nsecs]
 #5: <e4a79f00>, hrtimer_wakeup, S:01
 # expires at 249163930201148 nsecs [in 9122175398 nsecs]
 #6: <e4a79f00>, hrtimer_wakeup, S:01
 # expires at 249163930619465 nsecs [in 9122593715 nsecs]
 #7: <e4a79f00>, it_real_fn, S:01
 # expires at 249164018722673 nsecs [in 9210696923 nsecs]
 #8: <e4a79f00>, it_real_fn, S:01
 # expires at 249164018756764 nsecs [in 9210731014 nsecs]
 #9: <e4a79f00>, hrtimer_wakeup, S:01
 # expires at 249166140491719 nsecs [in 11332465969 nsecs]
 #10: <e4a79f00>, it_real_fn, S:01
 # expires at 249168890475020 nsecs [in 14082449270 nsecs]
 #11: <e4a79f00>, it_real_fn, S:01
 # expires at 249216937518155 nsecs [in 62129492405 nsecs]
 #12: <e4a79f00>, it_real_fn, S:01
 # expires at 249694958542841 nsecs [in 540150517091 nsecs]
 #13: <e4a79f00>, hrtimer_wakeup, S:01
 # expires at 252071939585424 nsecs [in 2917131559674 nsecs]
 #14: <e4a79f00>, it_real_fn, S:01
 # expires at 277954353421786 nsecs [in 28799545396036 nsecs]
  .expires_next   : 249154808417850 nsecs
  .hres_active    : 1
  .nr_events      : 12571303
  .nohz_mode      : 2
  .idle_tick      : 249154808417850 nsecs
  .tick_stopped   : 0
  .idle_jiffies   : 74656449
  .idle_calls     : 31252991
  .idle_sleeps    : 18916982
  .idle_entrytime : 249154806452801 nsecs
  .idle_sleeptime : 202805663229475 nsecs
  .last_jiffies   : 74656449
  .next_jiffies   : 74656452
  .idle_expires   : 249154815084516 nsecs
jiffies: 74656449

cpu: 1
clock 0:
.index: 0
.resolution: 1 nsecs
.get_time: ktime_get_real
.offset: 1177701795305563440 nsecs
active timers:
clock 1:
.index: 1
.resolution: 1 nsecs
.get_time: ktime_get
.offset: 0 nsecs
active timers:
#0: <e4a79f00>, tick_sched_timer, S:01

 # expires at 249154808417850 nsecs [in 392100 nsecs]
 #1: <e4a79f00>, it_real_fn, S:01
 # expires at 249154824005495 nsecs [in 15979745 nsecs]
 #2: <e4a79f00>, hrtimer_wakeup, S:01
 # expires at 249154876721584 nsecs [in 68695834 nsecs]
 #3: <e4a79f00>, hrtimer_wakeup, S:01
 # expires at 249154876724658 nsecs [in 68698908 nsecs]
 #4: <e4a79f00>, it_real_fn, S:01
 # expires at 249156991445550 nsecs [in 2183419800 nsecs]
 #5: <e4a79f00>, hrtimer_wakeup, S:01
 # expires at 249158033258177 nsecs [in 3225232427 nsecs]
 #6: <e4a79f00>, it_real_fn, S:01
 # expires at 249158937910855 nsecs [in 4129885105 nsecs]
 #7: <e4a79f00>, hrtimer_wakeup, S:01
 # expires at 249160645185907 nsecs [in 5837160157 nsecs]
 #8: <e4a79f00>, hrtimer_wakeup, S:01
 # expires at 249288273241617 nsecs [in 133465215867 nsecs]
 #9: <e4a79f00>, it_real_fn, S:01
 # expires at 249688052597900 nsecs [in 533244572150 nsecs]
 #10: <e4a79f00>, hrtimer_wakeup, S:01
 # expires at 250127175002655 nsecs [in 972366976905 nsecs]
 #11: <e4a79f00>, hrtimer_wakeup, S:01
 # expires at 252087970262178 nsecs [in 2933162236428 nsecs]
  .expires_next   : 249154808417850 nsecs
  .hres_active    : 1
  .nr_events      : 13107829
  .nohz_mode      : 2
  .idle_tick      : 249154805084517 nsecs
  .tick_stopped   : 0
  .idle_jiffies   : 74656448
  .idle_calls     : 27583722
  .idle_sleeps    : 11208166
  .idle_entrytime : 249154805632934 nsecs
  .idle_sleeptime : 197905845372598 nsecs
  .last_jiffies   : 74656449
  .next_jiffies   : 74656450
  .idle_expires   : 249154935084504 nsecs
jiffies: 74656449

Tick Device: mode: 1
Clock Event Device: pit

 max_delta_ns:   27461866
 min_delta_ns:   12571
 mult:           5124677
 shift:          32
 mode:           3
 next_event:     9223372036854775807 nsecs

set_next_event: pit_next_event
set_mode: init_pit_timer
event_handler: tick_handle_oneshot_broadcast
tick_broadcast_mask: 00000000
tick_broadcast_oneshot_mask: 00000000

Tick Device: mode: 1
Clock Event Device: lapic

 max_delta_ns:   503300116
 min_delta_ns:   1000
 mult:           71585107
 shift:          32
 mode:           3
 next_event:     249154808417850 nsecs

set_next_event: lapic_next_event
set_mode: lapic_timer_setup
event_handler: hrtimer_interrupt

Tick Device: mode: 1
Clock Event Device: lapic

 max_delta_ns:   503300116
 min_delta_ns:   1000
 mult:           71585107
 shift:          32
 mode:           3
 next_event:     249154808417850 nsecs

set_next_event: lapic_next_event
set_mode: lapic_timer_setup
event_handler: hrtimer_interrupt

#3 lore raw
Quoted 2 lines · Joerg Sommrey

Here it is. Maybe this problem is related to the usage of the "experimental" amd76x_pm module?

Can you please verify what happens w/o that module ?

Thanks,

tglx

#4 lore raw
Quoted 6 lines · Thomas Gleixner
Quoted 2 lines · Joerg Sommrey

Here it is. Maybe this problem is related to the usage of the "experimental" amd76x_pm module?

Can you please verify what happens w/o that module ?

After rebooting the problem vanished for now. It first appeared after an uptime of about 3 days. I'll wait a few days. If it shows again, then I'll check without amd76x_pm.

-jo

Quoted 5 lines

Thanks,

tglx

Signature · 1 line
-rw-r--r-- 1 jo users 62 2007-05-01 01:11 /home/jo/.signature
#5 lore raw
Quoted 10 lines · Joerg Sommrey
Quoted 6 lines · Thomas Gleixner
Quoted 2 lines · Joerg Sommrey

Here it is. Maybe this problem is related to the usage of the "experimental" amd76x_pm module?

Can you please verify what happens w/o that module ?

After rebooting the problem vanished for now. It first appeared after an uptime of about 3 days. I'll wait a few days. If it shows again, then I'll check without amd76x_pm.

It really looks like amd76x_pm is causing the time loss. Sadly, I have no idea how to avoid this. What happens when the pm-timer wraps around while the processors are in C3 sleep state? How is this wrap detected at all? Is there anything I could put into amd76x_pm? It's no problem detecting a timer wrap there.

-jo

Signature · 1 line
-rw-r--r-- 1 jo users 62 2007-05-13 21:55 /home/jo/.signature