CODEWITHSUNDEEP
opencvsegmentation-fault
arvids
arvids

Reputation: 304

Architecture change caused C++ program segmentation fault

A code that uses OpenCV and Caffe has worked on all Linux devices that i have tested it on. However launching it on a successfully installed Jetson TX2 causes a segmentation fault with this stack trace:

nvidia@tegra-ubuntu:~/Desktop$ gdb ./main
GNU gdb (Ubuntu 7.11.1-0ubuntu1~16.5) 7.11.1
This GDB was configured as "aarch64-linux-gnu".    
Reading symbols from ./main...done.
(gdb) r
Starting program: /home/nvidia/Desktop/main
[Thread debugging using libthread_db enabled]
Using host libthread_db library "/lib/aarch64-linux-gnu/libthread_db.so.1".Program received signal SIGSEGV, Segmentation fault.
0x0000007fb5e5d14c in google::protobuf::Arena::AllocateAligned(std::type_info const*, unsigned long) () from /usr/local/lib/libopencv_dnn.so.3.3
(gdb) bt
#0  0x0000007fb5e5d14c in google::protobuf::Arena::AllocateAligned(std::type_info const*, unsigned long) () from /usr/local/lib/libopencv_dnn.so.3.3
#1  0x0000007fb5e5d248 in google::protobuf::Arena::AddListNode(void*, void ()(void)) ()
  from /usr/local/lib/libopencv_dnn.so.3.3
#2  0x0000007fb5eaaf34 in google::protobuf::FileDescriptorProto::New(google::protobuf::Arena*) const [clone .localalias.409] () from /usr/local/lib/libopencv_dnn.so.3.3
#3  0x0000007fad71bfc4 in google::protobuf::MessageLite::ParseFromArray(void const*, int)
   () from /usr/lib/aarch64-linux-gnu/libprotobuf.so.9
#4  0x0000007fad763e70 in google::protobuf::EncodedDescriptorDatabase::Add(void const*, int) () from /usr/lib/aarch64-linux-gnu/libprotobuf.so.9
#5  0x0000007fad726d30 in google::protobuf::DescriptorPool::InternalAddGeneratedFile(void const*, int) () from /usr/lib/aarch64-linux-gnu/libprotobuf.so.9
#6  0x0000007fad7560bc in google::protobuf::protobuf_AddDesc_google_2fprotobuf_2fdescriptor_2eproto() () from /usr/lib/aarch64-linux-gnu/libprotobuf.so.9
#7  0x0000007fb7fdfb18 in call_init (l=<optimized out>, argc=argc@entry=1,
   argv=argv@entry=0x7ffffff478, env=env@entry=0x7ffffff488) at dl-init.c:72
#8  0x0000007fb7fdfc60 in call_init (env=0x7ffffff488, argv=0x7ffffff478, argc=1,
   l=<optimized out>) at dl-init.c:30
#9  _dl_init (main_map=0x7fb8000190, argc=1, argv=0x7ffffff478, env=0x7ffffff488)
   at dl-init.c:120
#10 0x0000007fb7fd2d44 in _dl_start_user () from /lib/ld-linux-aarch64.so.1

Protobuf shows up a lot, so i installed protobuf 3.3 and recompiled everything, but that did not help either. Simple code examples using OpenCV do work and Caffe runtest was passed. How do i search for a solution to this segfault?

Arvids

Upvotes: 0

Views: 535

Answers (1)

Employed Russian
Employed Russian

Reputation: 213937

How do i search for a solution to this segfault?

You don't.

Instead of searching for a solution, you find it yourself, by debugging the problem.

The first step should likely be to install debug info package for libopencv_dnn.so.3.3, or building it from source, so you can understand where in the Arena allocator your code is crashing.

Just with any crash in malloc, the problem is most likely in the user code, not the Arena allocator itself. The problem could be a stray write (i.e. random corruption) or, more likely, the API mis-use (e.g. calling Arena::Deallocate on something that wasn't allocated from that Arena).

P.S. The bug likely exists on other architectures, but hasn't announced itself yet. Heap corruption bugs often do that.

Upvotes: 1

Related Questions