Efficient way to debug art crash issue

250 views

Skip to first unread message

unread,

Apr 13, 2016, 2:35:14 PM4/13/16

to android-porting

When porting a new android device base on AOSP 5.0.1, we experience random crash across Java process.

From tombstone we see it mostly cause by GC check memory, such as

backtrace:

#00 pc 00127e90 /system/lib/libart.so (art::gc::collector::MarkSweep::ProcessMarkStack(bool)+183)

#01 pc 001284a1 /system/lib/libart.so (art::gc::collector::MarkSweep::MarkingPhase()+112)

#02 pc 0012856f /system/lib/libart.so (art::gc::collector::MarkSweep::RunPhases()+158)

#03 pc 0011fad1 /system/lib/libart.so (art::gc::collector::GarbageCollector::Run(art::gc::GcCause, bool)+232)

#04 pc 0013d1e7 /system/lib/libart.so (art::gc::Heap::CollectGarbageInternal(art::gc::collector::GcType, art::gc::GcCause, bool)+1398)

#05 pc 0013e5a9 /system/lib/libart.so (art::gc::Heap::ConcurrentGC(art::Thread*)+48)

#06 pc 000003df /data/dalvik-cache/arm/system@framework@boot.oat

Sometimes it fails on JNI check:

backtrace:

#00 pc 0003db54 /system/lib/libc.so (tgkill+12)

#01 pc 00016db9 /system/lib/libc.so (pthread_kill+52)

#02 pc 000179cf /system/lib/libc.so (raise+10)

#03 pc 00014171 /system/lib/libc.so (__libc_android_abort+36)

#04 pc 000124f8 /system/lib/libc.so (abort+4)

#05 pc 00216361 /system/lib/libart.so (art::Runtime::Abort()+160)

#06 pc 000a6d75 /system/lib/libart.so (art::LogMessage::~LogMessage()+1312)

#07 pc 000b09d5 /system/lib/libart.so (art::JniAbort(char const*, char const*)+1084)

#08 pc 000b0f1f /system/lib/libart.so (art::JniAbortF(char const*, char const*, ...)+58)

#09 pc 001c10bf /system/lib/libart.so (art::JNI::ReleaseIntArrayElements(_JNIEnv*, _jintArray*, int*, int)+470)

#10 pc 000140f1 /system/lib/libjavacore.so

The crash usually caused by access invalid memory or GC detect invalid access.

We know this may be caused by head corruption, but it is very hard to find who trashes the memory.