Re: [PATCH v2] perf trace: Fix missing handling of --call-graph dwarf
From: Arnaldo Carvalho de Melo
Date: Mon Jan 15 2018 - 10:02:39 EST
Em Mon, Jan 15, 2018 at 03:50:15PM +0100, Thomas-Mich Richter escreveu:
> On 01/15/2018 03:16 PM, Arnaldo Carvalho de Melo wrote:
> > Em Mon, Jan 15, 2018 at 10:57:52AM -0300, Arnaldo Carvalho de Melo escreveu:
> >>> [root@f27 perf]# ./perf trace --no-syscalls --max-stack 4
> >>> -e probe_libc:inet_pton/call-graph=dwarf/ ping -6 -c 1 ::1
> >>> PING ::1(::1) 56 data bytes
> >>> 64 bytes from ::1: icmp_seq=1 ttl=64 time=0.020 ms
> >>>
> >>> --- ::1 ping statistics ---
> >>> 1 packets transmitted, 1 received, 0% packet loss, time 0ms
> >>> rtt min/avg/max/mdev = 0.020/0.020/0.020/0.000 ms
> >>> 0.000 probe_libc:inet_pton:(7ffbc5f768a0))
> >>> __inet_pton (inlined)
> >>> gaih_inet.constprop.7 (/usr/lib64/libc-2.26.so)
> >>> __GI_getaddrinfo (inlined)
> >>> main (/usr/bin/ping)
> >>> [root@f27 perf]#
> >>>
> >>>
> >>> --> Dwarf call graph and --max-stack 4 is also honoured.
> >>
> >> [root@jouet ~]# perf trace --no-syscalls -e probe_libc:inet_pton/call-graph=dwarf,max-stack=4/
> >> perf trace --no-syscalls --max-stack 4 -e probe_libc:inet_pton/call-graph=dwarf/ ping -6 -c 1 ::1
> >> Value too large for defined data type
> >> [root@jouet ~]#
> >>
> >> Grrr.
> >
> >
> > Got this one fixed with the following patch:
> >
> > commit b78278e11f6992ca348a4b96aad3b2c0a9ecf0f0
>
> [...]
>
> >
> > diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c
> > index efa2e629a669..8f971a2301d1 100644
> > --- a/tools/perf/util/evsel.c
> > +++ b/tools/perf/util/evsel.c
> > @@ -731,14 +731,14 @@ static void apply_config_terms(struct perf_evsel *evsel,
> > struct perf_evsel_config_term *term;
> > struct list_head *config_terms = &evsel->config_terms;
> > struct perf_event_attr *attr = &evsel->attr;
> > - struct callchain_param param;
> > + /* callgraph default */
> > + struct callchain_param param = {
> > + .record_mode = callchain_param.record_mode,
> > + };
> > u32 dump_size = 0;
> > int max_stack = 0;
> > const char *callgraph_buf = NULL;
> >
> > - /* callgraph default */
> > - param.record_mode = callchain_param.record_mode;
> > -
> > list_for_each_entry(term, config_terms, list) {
> > switch (term->type) {
> > case PERF_EVSEL__CONFIG_TERM_PERIOD:
>
> This patch works for me.
> Here is the output on my s390x:
>
> [root@s8360047 perf]# ./perf trace --no-syscalls --max-stack 4
> -e probe_libc:inet_pton/call-graph=dwarf/ ping -6 -c 1 ::1
> PING ::1(::1) 56 data bytes
> 64 bytes from ::1: icmp_seq=1 ttl=64 time=0.070 ms
>
> --- ::1 ping statistics ---
> 1 packets transmitted, 1 received, 0% packet loss, time 0ms
> rtt min/avg/max/mdev = 0.070/0.070/0.070/0.000 ms
> 0.000 probe_libc:inet_pton:(3ffa70c2060))
> __GI___inet_pton (/usr/lib64/libc-2.26.so)
> gaih_inet (inlined)
> __GI_getaddrinfo (inlined)
> main (/usr/bin/ping)
> [root@s8360047 perf]# ./perf trace --no-syscalls --call-graph dwarf
> -e probe_libc:inet_pton -- ping -6 -c 1 ::1
> PING ::1(::1) 56 data bytes
> 64 bytes from ::1: icmp_seq=1 ttl=64 time=0.086 ms
>
> --- ::1 ping statistics ---
> 1 packets transmitted, 1 received, 0% packet loss, time 0ms
> rtt min/avg/max/mdev = 0.086/0.086/0.086/0.000 ms
> 0.000 probe_libc:inet_pton:(3ff93fc2060))
> __GI___inet_pton (/usr/lib64/libc-2.26.so)
> gaih_inet (inlined)
> __GI_getaddrinfo (inlined)
> main (/usr/bin/ping)
> __libc_start_main (/usr/lib64/libc-2.26.so)
> _start (/usr/bin/ping)
> [root@s8360047 perf]#
>
> [root@s8360047 perf]# ./perf trace --no-syscalls
> -e probe_libc:inet_pton/call-graph=dwarf,max-stack=3/ ping -6 -c 1 ::1
> PING ::1(::1) 56 data bytes
> 64 bytes from ::1: icmp_seq=1 ttl=64 time=0.066 ms
>
> --- ::1 ping statistics ---
> 1 packets transmitted, 1 received, 0% packet loss, time 0ms
> rtt min/avg/max/mdev = 0.066/0.066/0.066/0.000 ms
> 0.000 probe_libc:inet_pton:(3ffb82c2060))
> __GI___inet_pton (/usr/lib64/libc-2.26.so)
> gaih_inet (inlined)
> __GI_getaddrinfo (inlined)
> main (/usr/bin/ping)
> __libc_start_main (/usr/lib64/libc-2.26.so)
> _start (/usr/bin/ping)
> [root@s8360047 perf]#
>
> Not sure if this can work at all. Since dwarf stack unwinding is done in user space
> the attr.sample_max_stack set to 3 is useless in this case.
right, I'm working on this case now, this has to be honoured in
trace__resolve_callchain, that has to use evsel->attr.sample_max_stack
if it is set, otherwise fallback to trace->max_stack.
- Arnaldo