heap_profile for my jni libs showing only first frame of stack trace about perfetto HOT 7 CLOSED

Hey! Would it be possible for you upload the .so so I can take a look? We unwind using DWARF data.

Feel free to join our Discord (https://discordapp.com/channels/629013440635207702/629013441096450058) for more real-time debugging help - I am fmayer there :)

from perfetto.

Also, could you attach the output of adb logcat | grep "perfetto\|heapprofd" so I can see if there is something suspicious there?

from perfetto.

04-28 09:49:18.836 19072 19072 I perfetto: perfetto_cmd.cc:604 Connected to the Perfetto traced service, starting tracing for 604800000 ms
04-28 09:49:18.839  7385  7385 I perfetto: tracing_service_impl.cc:529 Configured tracing, #sources:2, duration:604800000 ms, #buffers:1, total buffer size:32768 KB, total sessions:2
04-28 09:49:58.263 19072 19072 I perfetto: perfetto_cmd.cc:815 SIGINT/SIGTERM received: disabling tracing.
04-28 09:49:58.407  7386  7386 I perfetto: probes_producer.cc:314 Producer stop (id=7)
04-28 09:49:58.410 19072 19072 I perfetto: perfetto_cmd.cc:674 Trace written into the output file
04-28 09:49:58.425  7385  7385 I perfetto: tracing_service_impl.cc:1660 Tracing session 3 ended, total sessions:1
04-28 09:49:58.579  7102  7174 E perfetto: client.cc:375 Failed to send control socket byte. (errno: 32, Broken pipe)
04-28 09:49:58.579  7102 19125 E perfetto: client.cc:375 Failed to send control socket byte. (errno: 32, Broken pipe)
04-28 10:38:47.374  7600  7600 I perfetto: perfetto_cmd.cc:815 SIGINT/SIGTERM received: disabling tracing.
04-28 10:38:47.380  7386  7386 I perfetto: probes_producer.cc:314 Producer stop (id=4)
04-28 10:38:47.381  7600  7600 I perfetto: perfetto_cmd.cc:674 Trace written into the output file
04-28 10:38:47.391  7385  7385 I perfetto: tracing_service_impl.cc:1660 Tracing session 2 ended, total sessions:0
^C

I wasn't using PERFETTO_BINARY_PATH above, could that be part of the problem?

from perfetto.

Ad PERFETO_BINARY_PATH: That shouldn't be the cause, as that would still show the unsymbolized frames.

from perfetto.

The library had all the unwinding information stripped. We need one of

• .gnu_debugdata
• .eh_frame (+ preferably .eh_frame_hdr)
• .debug_frame.

We do not support frame-pointer unwinding.

@tmm1 could you add your changes to your build to this bug for posterity?

from perfetto.

So the issue was that the android build process strips all the jniLibs. I tried LOCAL_STRIP_MODE := none, but ultimately what worked was adding packagingOptions.doNotStrip "**/*.so" to my app/build.grade

See https://stackoverflow.com/questions/40355200/how-to-avoid-stripping-for-native-code-symbols-for-android-app

FYI, to anyone who ends up here:

The library had all the unwinding information stripped

You can check this yourself using readelf -S libname.so and see if one of the debug sections is listed. The list of known sections can be seen inside libunwindstack

• .gnu_debugdata

This is a compressed version of .debug_frame and can be generated with something like do_strip_keep_mini_debug_info

from perfetto.

ultimately what worked was adding packagingOptions.doNotStrip "**/*.so" to my app/build.grade

There is a way to do this with just debug builds now in AGP 7.0.0: https://issuetracker.google.com/issues/155215248#comment5

from perfetto.