perf mem report segfaulting

perf mem report segfaulting

[Arnaldo Carvalho de Melo]

Hi,

	While trying to test my tmp.perf/ui_browser.horiz_scroll branch
with 'perf mem report' to allow that kind of browsing on the TUI, as of
now it works only on --stdio mode, using the right/left arrow keys, I
noticed this in acme/perf/core (but I saw it too in many previous
revisions, trying to bisect it):

[root@zoo ~]# perf mem record -a
^C[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 3.072 MB perf.data (647 samples) ]

[root@zoo ~]# perf mem report
perf: Segmentation fault
-------- backtrace --------
perf[0x52a08b]
/lib64/libc.so.6(+0x34960)[0x7f8726860960]
perf(addr_map_symbol__inc_samples+0xbb)[0x47abfb]
perf[0x42e93f]
perf(hist_entry_iter__add+0x102)[0x4d9632]
perf[0x42e7ba]
perf[0x4bc4c3]
perf[0x4bc9c1]
perf[0x4bf6b9]
perf(perf_session__process_events+0x390)[0x4be1b0]
perf(cmd_report+0x1038)[0x42faf8]
perf(cmd_mem+0x3ec)[0x44c69c]
perf[0x479113]
perf(main+0x60a)[0x420a5a]
/lib64/libc.so.6(__libc_start_main+0xf0)[0x7f872684bfe0]
perf[0x420b79]
[0x0]
[root@zoo ~]# perf evlist
cpu/mem-loads/pp
cpu/mem-stores/pp
[root@zoo ~]#

Have you noticed this? No time to look at this now :-\

- Arnaldo

Re: perf mem report segfaulting

[Jiri Olsa]

On Fri, Aug 28, 2015 at 02:04:14PM -0300, Arnaldo Carvalho de Melo wrote:
> 
> Hi,
> 
> 	While trying to test my tmp.perf/ui_browser.horiz_scroll branch
> with 'perf mem report' to allow that kind of browsing on the TUI, as of
> now it works only on --stdio mode, using the right/left arrow keys, I
> noticed this in acme/perf/core (but I saw it too in many previous
> revisions, trying to bisect it):
> 
> [root@zoo ~]# perf mem record -a
> ^C[ perf record: Woken up 1 times to write data ]
> [ perf record: Captured and wrote 3.072 MB perf.data (647 samples) ]
> 
> [root@zoo ~]# perf mem report
> perf: Segmentation fault
> -------- backtrace --------
> perf[0x52a08b]
> /lib64/libc.so.6(+0x34960)[0x7f8726860960]
> perf(addr_map_symbol__inc_samples+0xbb)[0x47abfb]
> perf[0x42e93f]
> perf(hist_entry_iter__add+0x102)[0x4d9632]
> perf[0x42e7ba]
> perf[0x4bc4c3]
> perf[0x4bc9c1]
> perf[0x4bf6b9]
> perf(perf_session__process_events+0x390)[0x4be1b0]
> perf(cmd_report+0x1038)[0x42faf8]
> perf(cmd_mem+0x3ec)[0x44c69c]
> perf[0x479113]
> perf(main+0x60a)[0x420a5a]
> /lib64/libc.so.6(__libc_start_main+0xf0)[0x7f872684bfe0]
> perf[0x420b79]
> [0x0]
> [root@zoo ~]# perf evlist
> cpu/mem-loads/pp
> cpu/mem-stores/pp
> [root@zoo ~]#
> 
> Have you noticed this? No time to look at this now :-\

cannot reproduce it on your perf/core.. did you navigate
to anything specific to trigger it?

jirka

Re: perf mem report segfaulting

[Arnaldo Carvalho de Melo]

Em Mon, Aug 31, 2015 at 10:19:57AM +0200, Jiri Olsa escreveu:
> On Fri, Aug 28, 2015 at 02:04:14PM -0300, Arnaldo Carvalho de Melo wrote:
> > 
> > Hi,
> > 
> > 	While trying to test my tmp.perf/ui_browser.horiz_scroll branch
> > with 'perf mem report' to allow that kind of browsing on the TUI, as of
> > now it works only on --stdio mode, using the right/left arrow keys, I
> > noticed this in acme/perf/core (but I saw it too in many previous
> > revisions, trying to bisect it):
> > 
> > [root@zoo ~]# perf mem record -a
> > ^C[ perf record: Woken up 1 times to write data ]
> > [ perf record: Captured and wrote 3.072 MB perf.data (647 samples) ]
> > 
> > [root@zoo ~]# perf mem report
> > perf: Segmentation fault
> > -------- backtrace --------
> > perf[0x52a08b]
> > /lib64/libc.so.6(+0x34960)[0x7f8726860960]
> > perf(addr_map_symbol__inc_samples+0xbb)[0x47abfb]
> > perf[0x42e93f]
> > perf(hist_entry_iter__add+0x102)[0x4d9632]
> > perf[0x42e7ba]
> > perf[0x4bc4c3]
> > perf[0x4bc9c1]
> > perf[0x4bf6b9]
> > perf(perf_session__process_events+0x390)[0x4be1b0]
> > perf(cmd_report+0x1038)[0x42faf8]
> > perf(cmd_mem+0x3ec)[0x44c69c]
> > perf[0x479113]
> > perf(main+0x60a)[0x420a5a]
> > /lib64/libc.so.6(__libc_start_main+0xf0)[0x7f872684bfe0]
> > perf[0x420b79]
> > [0x0]
> > [root@zoo ~]# perf evlist
> > cpu/mem-loads/pp
> > cpu/mem-stores/pp
> > [root@zoo ~]#
> > 
> > Have you noticed this? No time to look at this now :-\
> 
> cannot reproduce it on your perf/core.. did you navigate
> to anything specific to trigger it?

Nope, all I did is in the report, i.e. it segfaults straight away, right
after running 'perf mem report'.

Trying again, with what is in v4.2.0, i.e. upstream Linus, freshly built
(the tooling part, the kernel is a 4.2.0-rc one):

[root@zoo ~]# perf mem record -a
^C[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 3.260 MB perf.data (953 samples) ]

[root@zoo ~]# perf mem report
perf: Segmentation fault
-------- backtrace --------
perf[0x510c6b]
/lib64/libc.so.6(+0x34960)[0x7f27d6ed9960]
perf(addr_map_symbol__inc_samples+0xbb)[0x477ebb]
perf[0x42d017]
perf(hist_entry_iter__add+0x102)[0x4d4072]
perf[0x42ceda]
perf[0x4b7713]
perf[0x4b7bd1]
perf[0x4ba899]
perf(perf_session__process_events+0x390)[0x4b9390]
perf(cmd_report+0xfc9)[0x42e199]
perf(cmd_mem+0x3ec)[0x44aa6c]
perf[0x4767c3]
perf(main+0x60a)[0x41f19a]
/lib64/libc.so.6(__libc_start_main+0xf0)[0x7f27d6ec4fe0]
perf[0x41f2b9]
[0x0]
[root@zoo ~]# perf --version
perf version 4.2.g64291f
[root@zoo ~]# uname -a
Linux zoo 4.2.0-rc5+ #1 SMP Tue Aug 4 16:55:11 BRT 2015 x86_64 x86_64
x86_64 GNU/Linux
[root@zoo ~]# cat /etc/fedora-release 
Fedora release 21 (Twenty One)
[root@zoo ~]#

Re: perf mem report segfaulting

[Namhyung Kim]

Hi Arnaldo,

On Mon, Aug 31, 2015 at 11:23 PM, Arnaldo Carvalho de Melo
<acme@kernel.org> wrote:
> Em Mon, Aug 31, 2015 at 10:19:57AM +0200, Jiri Olsa escreveu:
>> On Fri, Aug 28, 2015 at 02:04:14PM -0300, Arnaldo Carvalho de Melo wrote:
>> >
>> > Hi,
>> >
>> >     While trying to test my tmp.perf/ui_browser.horiz_scroll branch
>> > with 'perf mem report' to allow that kind of browsing on the TUI, as of
>> > now it works only on --stdio mode, using the right/left arrow keys, I
>> > noticed this in acme/perf/core (but I saw it too in many previous
>> > revisions, trying to bisect it):
>> >
>> > [root@zoo ~]# perf mem record -a
>> > ^C[ perf record: Woken up 1 times to write data ]
>> > [ perf record: Captured and wrote 3.072 MB perf.data (647 samples) ]
>> >
>> > [root@zoo ~]# perf mem report
>> > perf: Segmentation fault
>> > -------- backtrace --------
>> > perf[0x52a08b]
>> > /lib64/libc.so.6(+0x34960)[0x7f8726860960]
>> > perf(addr_map_symbol__inc_samples+0xbb)[0x47abfb]
>> > perf[0x42e93f]
>> > perf(hist_entry_iter__add+0x102)[0x4d9632]
>> > perf[0x42e7ba]
>> > perf[0x4bc4c3]
>> > perf[0x4bc9c1]
>> > perf[0x4bf6b9]
>> > perf(perf_session__process_events+0x390)[0x4be1b0]
>> > perf(cmd_report+0x1038)[0x42faf8]
>> > perf(cmd_mem+0x3ec)[0x44c69c]
>> > perf[0x479113]
>> > perf(main+0x60a)[0x420a5a]
>> > /lib64/libc.so.6(__libc_start_main+0xf0)[0x7f872684bfe0]
>> > perf[0x420b79]
>> > [0x0]
>> > [root@zoo ~]# perf evlist
>> > cpu/mem-loads/pp
>> > cpu/mem-stores/pp
>> > [root@zoo ~]#
>> >
>> > Have you noticed this? No time to look at this now :-\
>>
>> cannot reproduce it on your perf/core.. did you navigate
>> to anything specific to trigger it?
>
> Nope, all I did is in the report, i.e. it segfaults straight away, right
> after running 'perf mem report'.
>
> Trying again, with what is in v4.2.0, i.e. upstream Linus, freshly built
> (the tooling part, the kernel is a 4.2.0-rc one):
>
> [root@zoo ~]# perf mem record -a
> ^C[ perf record: Woken up 1 times to write data ]
> [ perf record: Captured and wrote 3.260 MB perf.data (953 samples) ]
>
> [root@zoo ~]# perf mem report
> perf: Segmentation fault
> -------- backtrace --------
> perf[0x510c6b]
> /lib64/libc.so.6(+0x34960)[0x7f27d6ed9960]
> perf(addr_map_symbol__inc_samples+0xbb)[0x477ebb]
> perf[0x42d017]
> perf(hist_entry_iter__add+0x102)[0x4d4072]
> perf[0x42ceda]
> perf[0x4b7713]
> perf[0x4b7bd1]
> perf[0x4ba899]
> perf(perf_session__process_events+0x390)[0x4b9390]
> perf(cmd_report+0xfc9)[0x42e199]
> perf(cmd_mem+0x3ec)[0x44aa6c]
> perf[0x4767c3]
> perf(main+0x60a)[0x41f19a]
> /lib64/libc.so.6(__libc_start_main+0xf0)[0x7f27d6ec4fe0]
> perf[0x41f2b9]
> [0x0]
> [root@zoo ~]# perf --version
> perf version 4.2.g64291f
> [root@zoo ~]# uname -a
> Linux zoo 4.2.0-rc5+ #1 SMP Tue Aug 4 16:55:11 BRT 2015 x86_64 x86_64
> x86_64 GNU/Linux
> [root@zoo ~]# cat /etc/fedora-release
> Fedora release 21 (Twenty One)
> [root@zoo ~]#

I also cannot reproduce it on my system with current acme/perf/core.
The kernel version is 4.1.4 tho.

  $ uname -a
  Linux danjae 4.1.4-1-ARCH #1 SMP PREEMPT Mon Aug 3 21:30:37 UTC 2015
x86_64 GNU/Linux

  $ perf --version
  perf version 4.2.rc7.g2c07144

Thanks,
Namhyung

Re: perf mem report segfaulting

[Arnaldo Carvalho de Melo]

Em Tue, Sep 01, 2015 at 12:32:47AM +0900, Namhyung Kim escreveu:
> On Mon, Aug 31, 2015 at 11:23 PM, Arnaldo Carvalho de Melo
> > [root@zoo ~]# uname -a
> > Linux zoo 4.2.0-rc5+ #1 SMP Tue Aug 4 16:55:11 BRT 2015 x86_64 x86_64

> I also cannot reproduce it on my system with current acme/perf/core.
> The kernel version is 4.1.4 tho.
 
>   $ uname -a
>   Linux danjae 4.1.4-1-ARCH #1 SMP PREEMPT Mon Aug 3 21:30:37 UTC 2015
> x86_64 GNU/Linux
 
>   $ perf --version
>   perf version 4.2.rc7.g2c07144

Thanks for checking, that is strange, now with a 4.2.0 kernel while
using my a perf binary built from my perf/core branch I am not being
able to reproduce it... :-\

Will try rebooting back to that kernel to see if that is related...

- Arnaldo