Android Fatal Signal 11 (Sigsegv) At 0X636F7D89 (Code=1). How Can It Be Tracked Down

Android Fatal signal 11 (SIGSEGV) at 0x636f7d89 (code=1). How can it be tracked down?

I found the problem. I don't think this will help a lot of others trying to track down their personal SIGSEGV, but mine (and it was very hard) was entirely related to this:

https://code.google.com/p/android/issues/detail?id=8709

The libcrypto.so in my dump kind of clued me in. I do a MD5 hash of packet data when trying to determine if I've already seen the packet, and skipping it if I had. I thought at one point this was an ugly threading issue related to tracking those hashes, but it turned out it was the java.security.MessageDigest class! It's not thread safe!

I swapped it out with a UID I was stuffing in every packet based on the device UUID and a timestamp. No problems since.

I guess the lesson I can impart to those that were in my situation is, even if you're a 100% Java application, pay attention to the native library and symbol noted in the crash dump for clues. Googling for SIGSEGV + the lib .so name will go a lot farther than the useless code=1, etc... Next think about where your Java app could touch native code, even if it's nothing you're doing. I made the mistake of assuming it was a Service + UI threading issue where the Canvas was drawing something that was null, (the most common case I Googled on SIGSEGV) and ignored the possibility it could have been completely related to code I wrote that was related to the lib .so in the crash dump. Naturally java.security would use a native component in libcrypto.so for speed, so once I clued in, I Googled for Android + SIGSEGV + libcrypto.so and found the documented issue.

Android Fatal signal 11 (SIGSEGV) at 0x636f7d89 (code=1). How can it be tracked down?

I found the problem. I don't think this will help a lot of others trying to track down their personal SIGSEGV, but mine (and it was very hard) was entirely related to this:

https://code.google.com/p/android/issues/detail?id=8709

The libcrypto.so in my dump kind of clued me in. I do a MD5 hash of packet data when trying to determine if I've already seen the packet, and skipping it if I had. I thought at one point this was an ugly threading issue related to tracking those hashes, but it turned out it was the java.security.MessageDigest class! It's not thread safe!

I swapped it out with a UID I was stuffing in every packet based on the device UUID and a timestamp. No problems since.

I guess the lesson I can impart to those that were in my situation is, even if you're a 100% Java application, pay attention to the native library and symbol noted in the crash dump for clues. Googling for SIGSEGV + the lib .so name will go a lot farther than the useless code=1, etc... Next think about where your Java app could touch native code, even if it's nothing you're doing. I made the mistake of assuming it was a Service + UI threading issue where the Canvas was drawing something that was null, (the most common case I Googled on SIGSEGV) and ignored the possibility it could have been completely related to code I wrote that was related to the lib .so in the crash dump. Naturally java.security would use a native component in libcrypto.so for speed, so once I clued in, I Googled for Android + SIGSEGV + libcrypto.so and found the documented issue.

How to solve Android Fatal signal 11 (SIGSEGV), code 1, fault addr 0x0 in tid xxxxx (Thread-X)?

After some discussion it became clear that the problem was with interaction with memory:

extern "C" 
jdouble 
JNICALL Java_com_foo(JNIEnv *env, jclass type, jlong addrRgba, jlong addrGray) { 
  Mat &mRgb = *(Mat *) addrRgba;
  Mat &mGray = *(Mat *) addrGray;

  return (jdouble) toGray(mRgb, mGray);
}

As a quick fix double toGray(Mat& rgb, Mat& gray); had to be changed to double toGray(Mat rgb, Mat gray)

Additional information can be found on topic CvMat deep copy

---

Related Topics