Search code examples
llvmllvm-irobjdumpbitcodeobjcopy

How to extract the __bitcode/__bundle section from a MachO binary

For a binary with embedded bitcode (i.e., compiled with -fembed-bitcode). How can i just extract that section such that I can use that bitcode file like any other bitcode file e.g, run opt or llvm-dis

As a testcase i have this hello-world program:

// hello.cpp
#include<iostream>

int main() {
  std::cout << "hello world";
  return 0;
}

compiled with: clang++ -O2 test.cpp -o test.o -fembed-bitcode -c

the objdump shows there is a __bitcode section in there:

objdump -h test.o 

test.o: file format Mach-O 64-bit x86-64

Sections:
Idx Name          Size      Address          Type
  0 __text        000002eb 0000000000000000 TEXT 
  1 __gcc_except_tab 00000068 00000000000002ec DATA 
  2 __cstring     0000000c 0000000000000354 DATA 
  3 __bitcode     00002bc0 0000000000000360 DATA 
  4 __cmdline     00000046 0000000000002f20 DATA 
  5 __compact_unwind 00000060 0000000000002f68 DATA 
  6 __eh_frame    000000d0 0000000000002fc8 DATA

Now as per https://github.com/llvm/llvm-project/blob/master/llvm/test/tools/llvm-objcopy/MachO/dump-section.test, I'm trying to extract the __bitcode section in the following way:

# Take1
./bin/llvm-objcopy --dump-section=DATA,__bitcode=a.bc test.o 
./bin/llvm-objcopy: error: 'test.o': section 'DATA,__bitcode' not found

# Take2
./bin/llvm-objcopy --dump-section=__bitcode=a.bc test.o 
./bin/llvm-objcopy: error: 'test.o': section '__bitcode' not found

# Take3
./bin/llvm-objcopy --dump-section __DATA,__bitcode=a.bc test.o 
./bin/llvm-objcopy: error: 'test.o': section '__DATA,__bitcode' not found

# Take4
./bin/llvm-objcopy --dump-section=__DATA,__bitcode=a.bc test.o 
./bin/llvm-objcopy: error: 'test.o': section '__DATA,__bitcode' not found

What am I missing here?

Solution

  • Actually the __bitcode is in LLVM section. so the following commands worked.

    llvm-objcopy --dump-section=__LLVM,__bitcode=a.bc test.o
segedit test.o -extract __LLVM __bitcode a.bc

    Seems like a bug in objdump, it doesn't recognize the LLVM section.