Re: Intel PT not work on 4.5.0-rc

[Arnaldo Carvalho de Melo]

Em Tue, Mar 01, 2016 at 03:20:35PM -0300, Arnaldo Carvalho de Melo escreveu:
> Em Tue, Mar 01, 2016 at 12:59:21PM -0300, Arnaldo Carvalho de Melo escreveu:
> > Anyway, 'strace' shows the 4 pairs of mmaps, but it still doesn't work, rebooting:
> 
> After I reboot, all works:
> 
> [root@jouet ~]# perf record -e intel_pt//u -a sleep 0.001
> [ perf record: Woken up 1 times to write data ]
> [ perf record: Captured and wrote 1.119 MB perf.data ]
> [root@jouet ~]# perf script | head -10
>   perf 3797 [002] 8136.590828: 1 branches:u:            0 [unknown] ([unknown]) =>     7fcb3f240707 __GI___ioctl (/usr/lib64/libc-2.22.so)
>   perf 3797 [002] 8136.590828: 1 branches:u: 7fcb3f24070f __GI___ioctl (/usr/lib64/libc-2.22.so) =>    497e56 perf_evsel__enable (/home/acme/bin/perf)
>   perf 3797 [002] 8136.590828: 1 branches:u:       497e58 perf_evsel__enable (/home/acme/bin/perf) =>  497e20 perf_evsel__enable (/home/acme/bin/perf)
>   perf 3797 [002] 8136.590828: 1 branches:u:       497e27 perf_evsel__enable (/home/acme/bin/perf) =>  497e80 perf_evsel__enable (/home/acme/bin/perf)
>   perf 3797 [002] 8136.590828: 1 branches:u:       497e87 perf_evsel__enable (/home/acme/bin/perf) =>  497e14 perf_evsel__enable (/home/acme/bin/perf)
>   perf 3797 [002] 8136.590828: 1 branches:u:       497e1b perf_evsel__enable (/home/acme/bin/perf) =>  497e29 perf_evsel__enable (/home/acme/bin/perf)
>   perf 3797 [002] 8136.590828: 1 branches:u:       497e51 perf_evsel__enable (/home/acme/bin/perf) =>  421d00 ioctl@plt (/home/acme/bin/perf)
>   perf 3797 [002] 8136.590828: 1 branches:u:       421d00 ioctl@plt (/home/acme/bin/perf) =>  7fcb3f240700 __GI___ioctl (/usr/lib64/libc-2.22.so)
>   perf 3797 [002] 8136.590828: 1 branches:u: 7fcb3f240705 __GI___ioctl (/usr/lib64/libc-2.22.so)   =>   0 [unknown] ([unknown])
>   perf 3797 [002] 8136.590834: 1 branches:u:            0 [unknown] ([unknown]) =>     7fcb3f240707 __GI___ioctl (/usr/lib64/libc-2.22.so)
> [root@jouet ~]# 
> [root@jouet ~]# perf report -D | tail -20
> .  0020:  00 00 00 00 00 00 00 00 07 14 b9 72 66 07 00 00  ...........rf...
> .  0030:  03 00 00 00 00 00 00 00 1e 00 00 00 00 00 00 00  ................
> 
> 3 8136592790535 0x11e688 [0x40]: PERF_RECORD_AUX offset: 0x28 size: 0x28 flags: 0 []
> 
> Aggregated stats: (excludes AUX area (e.g. instruction trace) decoded / synthesized events)
>            TOTAL events:       8859
>             MMAP events:        120
>             COMM events:        509
>             EXIT events:          1
>             FORK events:        507
>            MMAP2 events:       7673
>              AUX events:          8
>     ITRACE_START events:          4
>  SWITCH_CPU_WIDE events:         26
>   FINISHED_ROUND events:          2
>    AUXTRACE_INFO events:          1
>         AUXTRACE events:          8
> intel_pt//u stats:
> dummy:u stats:
> [root@jouet ~]# 
> 
> Trying a suspend/resume now...

Scrap that theory, it still works after two suspend/resume cycles :-\

- ARnaldo