As the documentation for dwfl_frame_pc says, frames that
are no activation frames need to have their program counter
decremented by one to properly find the function of the caller.
This fixes many cases where perf report currently attributes
the cost to the next line. I.e. I have code like this:
~~~~~~~~~~~~~~~
#include <thread>
#include <chrono>
using namespace std;
int main()
{
this_thread::sleep_for(chrono::milliseconds(1000));
this_thread::sleep_for(chrono::milliseconds(100));
this_thread::sleep_for(chrono::milliseconds(10));
return 0;
}
~~~~~~~~~~~~~~~
Now compile and record it:
~~~~~~~~~~~~~~~
g++ -std=c++11 -g -O2 test.cpp
echo 1 | sudo tee /proc/sys/kernel/sched_schedstats
perf record \
--event sched:sched_stat_sleep \
--event sched:sched_process_exit \
--event sched:sched_switch --call-graph=dwarf \
--output perf.data.raw \
./a.out
echo 0 | sudo tee /proc/sys/kernel/sched_schedstats
perf inject --sched-stat --input perf.data.raw --output perf.data
~~~~~~~~~~~~~~~
Before this patch, the report clearly shows the off-by-one issue.
Most notably, the last sleep invocation is incorrectly attributed
to the "return 0;" line:
~~~~~~~~~~~~~~~
Overhead Source:Line
........ ...........
100.00% core.c:0
|
---__schedule core.c:0
schedule
do_nanosleep hrtimer.c:0
hrtimer_nanosleep
sys_nanosleep
entry_SYSCALL_64_fastpath .tmp_entry_64.o:0
__nanosleep_nocancel .:0
std::this_thread::sleep_for<long, std::ratio<1l, 1000l> > thread:323
|
|--90.08%--main test.cpp:9
| __libc_start_main
| _start
|
|--9.01%--main test.cpp:10
| __libc_start_main
| _start
|
--0.91%--main test.cpp:13
__libc_start_main
_start
~~~~~~~~~~~~~~~
With this patch here applied, the issue is fixed. The report becomes
much more usable:
~~~~~~~~~~~~~~~
Overhead Source:Line
........ ...........
100.00% core.c:0
|
---__schedule core.c:0
schedule
do_nanosleep hrtimer.c:0
hrtimer_nanosleep
sys_nanosleep
entry_SYSCALL_64_fastpath .tmp_entry_64.o:0
__nanosleep_nocancel .:0
std::this_thread::sleep_for<long, std::ratio<1l, 1000l> > thread:323
|
|--90.08%--main test.cpp:8
| __libc_start_main
| _start
|
|--9.01%--main test.cpp:9
| __libc_start_main
| _start
|
--0.91%--main test.cpp:10
__libc_start_main
_start
~~~~~~~~~~~~~~~
Similarly it works for signal frames:
~~~~~~~~~~~~~~~
__noinline void bar(void)
{
volatile long cnt = 0;
for (cnt = 0; cnt < 100000000; cnt++);
}
__noinline void foo(void)
{
bar();
}
void sig_handler(int sig)
{
foo();
}
int main(void)
{
signal(SIGUSR1, sig_handler);
raise(SIGUSR1);
foo();
return 0;
}
~~~~~~~~~~~~~~~~
Before, the report wrongly points to `signal.c:29` after raise():
~~~~~~~~~~~~~~~~
$ perf report --stdio --no-children -g srcline -s srcline
...
100.00% signal.c:11
|
---bar signal.c:11
|
|--50.49%--main signal.c:29
| __libc_start_main
| _start
|
--49.51%--0x33a8f
raise .:0
main signal.c:29
__libc_start_main
_start
~~~~~~~~~~~~~~~~
With this patch in, the issue is fixed and we instead get:
~~~~~~~~~~~~~~~~
100.00% signal signal [.] bar
|
---bar signal.c:11
|
|--50.49%--main signal.c:29
| __libc_start_main
| _start
|
--49.51%--0x33a8f
raise .:0
main signal.c:27
__libc_start_main
_start
~~~~~~~~~~~~~~~~
Note how this patch fixes this issue for both unwinding methods, i.e.
both dwfl and libunwind. The former case is straight-forward thanks
to dwfl_frame_pc(). For libunwind, we replace the functionality via
unw_is_signal_frame() for any but the very first frame.
Signed-off-by: Milian Wolff <milian.wolff@kdab.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Cc: Arnaldo Carvalho de Melo <acme@kernel.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Yao Jin <yao.jin@linux.intel.com>
Cc: kernel-team@lge.com
Link: http://lkml.kernel.org/r/20170524062129.32529-4-namhyung@kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Using perf with call graph method dwarf fails to provide backtrace
support for stripped binary even though .gnu_debuglink points to *.dbg
flavor with properly populated debug symbols.
Problem is reproduced on ARM (v7, v8), kernels 3.14.y, 4.4.y and
4.10.rc3. Perf is configured with libunwind, and unwind dwarf support
[1]. Test code (stress_bt.c) can be found on [2].
Running (explicitly disable other unwinding methods):
$ gcc -g -o stress_bt -fomit-frame-pointer -fno-unwind-tables \
-fno-asynchronous-unwind-tables stress_bt.c
$ perf record -N --call-graph dwarf ./stress_bt
$ perf report
results in properly generated call graph. Stripping the binary and running
it results with missing call graph. Expected result is to have call graph:
$ gcc -g -o stress_bt -fomit-frame-pointer -fno-unwind-tables \
-fno-asynchronous-unwind-tables stress_bt.c
$ objcopy --only-keep-debug stress_bt stress_bt.dbg
$ objcopy --strip-debug stress_bt
$ objcopy --add-gnu-debuglink=stress_bt.dbg stress_bt
$ perf record -N --call-graph dwarf ./stress_bt
$ perf report
Problem is that perf doesn't try to read symbols pointed by gnu
debuglink. Patch adds checking, and reading of the symbols from
debuglink and symsrc. Order of the check is to first check within dso,
then check whether symsrc is defined and try to read from it. Finally,
debuglink is checked. Default locations of debug files are discussed in
[3] and [4]. Comments on RFC are on [5].
[1] https://wiki.linaro.org/LEG/Engineering/TOOLS/perf-callstack-unwinding
[2] [1]#Backtrace_stress_application
[3] https://sourceware.org/gdb/onlinedocs/gdb/Separate-Debug-Files.html
[4] https://sourceware.org/binutils/docs/binutils/objcopy.html
[5] https://lkml.org/lkml/2016/8/22/473
Signed-off-by: Matija Glavinic Pecotic <matija.glavinic-pecotic.ext@nokia.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Cc: Alexander Sverdlin <alexander.sverdlin@nokia.com>
Cc: Masami Hiramatsu <mhiramat@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Link: http://lkml.kernel.org/r/d309d40a-463f-482b-68e1-1465326efdc1@nokia.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
