Skip to contentSkip to conversation

v2.5-rc2

Powertop Sergey Senozhatsky 2 participants 3 messages Nov 12, 2013 1 message 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]

Sergey Senozhatsky

author
#1 lore raw
Quoted 7 lines · Kristen Carlson Accardi
Quoted 7 lines
Quoted 7 lines

Thanks for your report, I bisected this issue down to this commit:
commit 2b980e1b36cf6150308a9db07b958726dad36571

Author: Sergey Senozhatsky <sergey.senozhatsky(a)gmail.com>
Date:   Tue Oct 8 22:35:53 2013 +0300

traceevent: update libtraceevent code base (3.12)

Quoted 9 lines
Quoted 5 lines
Quoted 1 line

Sergey? See report below for how to duplicate.

hm... big problem, libtraceevent unable to support hrtimers event at the moment with arg construction in TP_printk()

OK, I am thinking that we'll consider this a showstopper and delay the release of 2.5 until it is fixed if we need to.

Hello,
sorry it took me so long to get back to this.

after short investigation, I don't think we need to postpone 2.5 release. yes, libtraceevent does not handle nicely
/** nor trace-cmd does

     kworker/1:0-3925  [001]  1626.266585: hrtimer_cancel:       hrtimer=0xffff880157c8df80
               X-2175  [000]  1626.266585: hrtimer_cancel:       hrtimer=0xffff880157c0df80
               X-2175  [000]  1626.266586: hrtimer_expire_entry: [FAILED TO PARSE] hrtimer=0xffff880157c0df80 now=1625348004324 function=0xffffffff810cadf0
     kworker/1:0-3925  [001]  1626.266586: hrtimer_expire_entry: [FAILED TO PARSE] hrtimer=0xffff880157c8df80 now=1625348004320 function=0xffffffff810cadf0
     kworker/1:0-3925  [001]  1626.266591: hrtimer_expire_exit:  hrtimer=0xffff880157c8df80
     kworker/1:0-3925  [001]  1626.266592: hrtimer_start:        [FAILED TO PARSE] hrtimer=0xffff880157c8df80 function=0xffffffff810cadf0 expires=1625352000000 softexpires=1625352000000
               X-2175  [000]  1626.266593: hrtimer_expire_exit:  hrtimer=0xffff880157c0df80
               X-2175  [000]  1626.266594: hrtimer_start:        [FAILED TO PARSE] hrtimer=0xffff880157c0df80 function=0xffffffff810cadf0 expires=1625352000000 softexpires=1625352000000
     kworker/1:0-3925  [001]  1626.266596: timer_cancel:         timer=0xffff880157c8ff28

**/
some of hrtimer events printk format [e.g. hrtimer_expire_entry]

	print fmt: "hrtimer=%p function=%pf now=%llu", REC->hrtimer, REC->function, (unsigned long long)(((ktime_t) { .tv64 = REC->now }).tv64)

having issues with arg construction { .tv64 = REC->now }.tv64 (and that unknown op '{' warning), however, we're able to parse hrtimer event and to handle it. /* as far as I understand, correct me if I'm wrong */

debug output from do_process perf_process_bundle::handle_trace_point()

[..]

Quoted 10 lines
Quoted 10 lines

hrtimer_expire_exit: pop_consumer: 0
timer done: 0x8683f90 5941
timer created: 0x8683f90 [timer]
hrtimer_expire_entry: push_consumer: 0
timer created: 0x8683f90 [timer]
hrtimer_expire_entry: push_consumer: 2
hrtimer_expire_exit: pop_consumer: 0
timer done: 0x8683f90 9857
hrtimer_expire_exit: pop_consumer: 2
timer done: 0x8683f90 12529

Quoted 8 lines
Quoted 8 lines

timer created: 0x8683f90 [timer]
hrtimer_expire_entry: push_consumer: 0
timer created: 0x8683f90 [timer]
hrtimer_expire_entry: push_consumer: 2
hrtimer_expire_exit: pop_consumer: 0
timer done: 0x8683f90 4481
hrtimer_expire_exit: pop_consumer: 2
timer done: 0x8683f90 6535

[..]

any objections?

	-ss
author
#2 lore raw
Quoted 24 lines · Sergey Senozhatsky
Quoted 7 lines · Kristen Carlson Accardi
Quoted 7 lines
Quoted 7 lines

Thanks for your report, I bisected this issue down to this commit:
commit 2b980e1b36cf6150308a9db07b958726dad36571

Author: Sergey Senozhatsky <sergey.senozhatsky(a)gmail.com>
Date:   Tue Oct 8 22:35:53 2013 +0300

traceevent: update libtraceevent code base (3.12)

Quoted 9 lines
Quoted 5 lines
Quoted 1 line

Sergey? See report below for how to duplicate.

hm... big problem, libtraceevent unable to support hrtimers event at the moment with arg construction in TP_printk()

OK, I am thinking that we'll consider this a showstopper and delay the release of 2.5 until it is fixed if we need to.

Hello,
sorry it took me so long to get back to this.

after short investigation, I don't think we need to postpone 2.5 release. yes, libtraceevent does not handle nicely
/** nor trace-cmd does

Quoted 9 lines
     kworker/1:0-3925  [001]  1626.266585: hrtimer_cancel:       hrtimer=0xffff880157c8df80
               X-2175  [000]  1626.266585: hrtimer_cancel:       hrtimer=0xffff880157c0df80
               X-2175  [000]  1626.266586: hrtimer_expire_entry: [FAILED TO PARSE] hrtimer=0xffff880157c0df80 now=1625348004324 function=0xffffffff810cadf0
     kworker/1:0-3925  [001]  1626.266586: hrtimer_expire_entry: [FAILED TO PARSE] hrtimer=0xffff880157c8df80 now=1625348004320 function=0xffffffff810cadf0
     kworker/1:0-3925  [001]  1626.266591: hrtimer_expire_exit:  hrtimer=0xffff880157c8df80
     kworker/1:0-3925  [001]  1626.266592: hrtimer_start:        [FAILED TO PARSE] hrtimer=0xffff880157c8df80 function=0xffffffff810cadf0 expires=1625352000000 softexpires=1625352000000
               X-2175  [000]  1626.266593: hrtimer_expire_exit:  hrtimer=0xffff880157c0df80
               X-2175  [000]  1626.266594: hrtimer_start:        [FAILED TO PARSE] hrtimer=0xffff880157c0df80 function=0xffffffff810cadf0 expires=1625352000000 softexpires=1625352000000
     kworker/1:0-3925  [001]  1626.266596: timer_cancel:         timer=0xffff880157c8ff28

taking my words about trace-cmd back.
trace-cmd can parse it, just need proper plugin placement [..]

         <idle>-0     [000]  7126.976512: hrtimer_cancel:       hrtimer=0xffff880157c0df80
              ls-21607 [001]  7126.976513: hrtimer_expire_entry: hrtimer=0xffff880157c8df80 now=7123856003259 function=tick_sched_timer/0x0
          <idle>-0     [000]  7126.976513: hrtimer_expire_entry: hrtimer=0xffff880157c0df80 now=7123856003132 function=tick_sched_timer/0x0
               X-2175  [003]  7126.976513: hrtimer_cancel:       hrtimer=0xffff880157d8df80
               X-2175  [003]  7126.976515: hrtimer_expire_entry: hrtimer=0xffff880157d8df80 now=7123856003882 function=tick_sched_timer/0x0
              ls-21607 [001]  7126.976518: hrtimer_expire_exit:  hrtimer=0xffff880157c8df80
          <idle>-0     [000]  7126.976519: hrtimer_expire_exit:  hrtimer=0xffff880157c0df80
              ls-21607 [001]  7126.976519: hrtimer_start:        hrtimer=0xffff880157c8df80 function=tick_sched_timer expires=7123860000000 softexpires=7123860000000
          <idle>-0     [000]  7126.976519: hrtimer_start:        hrtimer=0xffff880157c0df80 function=tick_sched_timer expires=7123860000000 softexpires=7123860000000
               X-2175  [003]  7126.976522: hrtimer_expire_exit:  hrtimer=0xffff880157d8df80

[..]

powertop does not use char function name representation, but instead uint64_t (passing to find_create_timer(uint64_t func)).

	-ss
Quoted 34 lines

**/
some of hrtimer events printk format [e.g. hrtimer_expire_entry]

	print fmt: "hrtimer=%p function=%pf now=%llu", REC->hrtimer, REC->function, (unsigned long long)(((ktime_t) { .tv64 = REC->now }).tv64)

having issues with arg construction { .tv64 = REC->now }.tv64 (and that unknown op '{' warning), however, we're able to parse hrtimer event and to handle it. /* as far as I understand, correct me if I'm wrong */

debug output from do_process perf_process_bundle::handle_trace_point()

[..]

Quoted 10 lines
Quoted 10 lines

hrtimer_expire_exit: pop_consumer: 0
timer done: 0x8683f90 5941
timer created: 0x8683f90 [timer]
hrtimer_expire_entry: push_consumer: 0
timer created: 0x8683f90 [timer]
hrtimer_expire_entry: push_consumer: 2
hrtimer_expire_exit: pop_consumer: 0
timer done: 0x8683f90 9857
hrtimer_expire_exit: pop_consumer: 2
timer done: 0x8683f90 12529

Quoted 8 lines
Quoted 8 lines

timer created: 0x8683f90 [timer]
hrtimer_expire_entry: push_consumer: 0
timer created: 0x8683f90 [timer]
hrtimer_expire_entry: push_consumer: 2
hrtimer_expire_exit: pop_consumer: 0
timer done: 0x8683f90 4481
hrtimer_expire_exit: pop_consumer: 2
timer done: 0x8683f90 6535

[..]

any objections?

	-ss
#3 lore raw
In reply to Sergey Senozhatsky
Quoted 62 lines · On Wed, 13 Nov 2013 00:08:28 +0300 Sergey Senozhatsky…
Quoted 7 lines · Kristen Carlson Accardi
Quoted 7 lines
Quoted 7 lines

Thanks for your report, I bisected this issue down to this commit:
commit 2b980e1b36cf6150308a9db07b958726dad36571

Author: Sergey Senozhatsky <sergey.senozhatsky(a)gmail.com>
Date:   Tue Oct 8 22:35:53 2013 +0300

traceevent: update libtraceevent code base (3.12)

Quoted 9 lines
Quoted 5 lines
Quoted 1 line

Sergey? See report below for how to duplicate.

hm... big problem, libtraceevent unable to support hrtimers event at the moment with arg construction in TP_printk()

OK, I am thinking that we'll consider this a showstopper and delay the release of 2.5 until it is fixed if we need to.

Hello,
sorry it took me so long to get back to this.

after short investigation, I don't think we need to postpone 2.5 release. yes, libtraceevent does not handle nicely
/** nor trace-cmd does

     kworker/1:0-3925  [001]  1626.266585: hrtimer_cancel:       hrtimer=0xffff880157c8df80
               X-2175  [000]  1626.266585: hrtimer_cancel:       hrtimer=0xffff880157c0df80
               X-2175  [000]  1626.266586: hrtimer_expire_entry: [FAILED TO PARSE] hrtimer=0xffff880157c0df80 now=1625348004324 function=0xffffffff810cadf0
     kworker/1:0-3925  [001]  1626.266586: hrtimer_expire_entry: [FAILED TO PARSE] hrtimer=0xffff880157c8df80 now=1625348004320 function=0xffffffff810cadf0
     kworker/1:0-3925  [001]  1626.266591: hrtimer_expire_exit:  hrtimer=0xffff880157c8df80
     kworker/1:0-3925  [001]  1626.266592: hrtimer_start:        [FAILED TO PARSE] hrtimer=0xffff880157c8df80 function=0xffffffff810cadf0 expires=1625352000000 softexpires=1625352000000
               X-2175  [000]  1626.266593: hrtimer_expire_exit:  hrtimer=0xffff880157c0df80
               X-2175  [000]  1626.266594: hrtimer_start:        [FAILED TO PARSE] hrtimer=0xffff880157c0df80 function=0xffffffff810cadf0 expires=1625352000000 softexpires=1625352000000
     kworker/1:0-3925  [001]  1626.266596: timer_cancel:         timer=0xffff880157c8ff28

**/
some of hrtimer events printk format [e.g. hrtimer_expire_entry]

	print fmt: "hrtimer=%p function=%pf now=%llu", REC->hrtimer, REC->function, (unsigned long long)(((ktime_t) { .tv64 = REC->now }).tv64)

having issues with arg construction { .tv64 = REC->now }.tv64 (and that unknown op '{' warning), however, we're able to parse hrtimer event and to handle it. /* as far as I understand, correct me if I'm wrong */

debug output from do_process perf_process_bundle::handle_trace_point()

[..]

Quoted 10 lines
Quoted 10 lines

hrtimer_expire_exit: pop_consumer: 0
timer done: 0x8683f90 5941
timer created: 0x8683f90 [timer]
hrtimer_expire_entry: push_consumer: 0
timer created: 0x8683f90 [timer]
hrtimer_expire_entry: push_consumer: 2
hrtimer_expire_exit: pop_consumer: 0
timer done: 0x8683f90 9857
hrtimer_expire_exit: pop_consumer: 2
timer done: 0x8683f90 12529

Quoted 8 lines
Quoted 8 lines

timer created: 0x8683f90 [timer]
hrtimer_expire_entry: push_consumer: 0
timer created: 0x8683f90 [timer]
hrtimer_expire_entry: push_consumer: 2
hrtimer_expire_exit: pop_consumer: 0
timer done: 0x8683f90 4481
hrtimer_expire_exit: pop_consumer: 2
timer done: 0x8683f90 6535

Quoted 5 lines

[..]

any objections?

	-ss

good, thanks! as long as we actually get the events it doesn't seem like a showstopper anymore to me :).