I am trying to use perf to profile my Android application.
With help of fplutil project I managed to install perf on Android device, run perf record and then perf report on host machine.
Usually when I run perf report -g I expect to see call graph similar to this one:
47.91% a.out a.out [.] f5()
|
--- f5()
|
|--53.82%-- f4()
| |
| |--56.16%-- f3()
| | |
| | |--49.14%-- f2()
| | | |
| | | |--52.92%-- f1()
| | | | a()
| | | | main
| | | | __libc_start_main
| | | |
| | | --47.08%-- a()
| | | main
| | | __libc_start_main
However when I execute perf report for data collected on Android device I do not see proper call graph, like if stack is not unwinding properly:
4.98% app_name my_lib.so [.] void quat_apply<double>(double*, double const*, double const*, int, double)
|
--- 0xbee75fe4
|
--- 0xbee760c4
|
--- 0xbee75c0c
What should I do to replace these hex addresses with function names?
I am already building my code with -g and -fno-omit-frame-pointer, am I missing something else?
Bingo!
-mapcs-frame
Generate a stack frame that is compliant with the ARM Procedure Call Standard for all functions, even if this is not strictly necessary for correct execution of the code. Specifying -fomit-frame-pointer with this option causes the stack frames not to be generated for leaf functions. The default is -mno-apcs-frame. This option is deprecated.
Although option is marked as deprecated, it fixed call graph for me.