[llvm-bugs] [Bug 35371] New: [OrcJIT] Incorrect symbol resolution of libc function addresses on MacOS

[via llvm-bugs]

https://bugs.llvm.org/show_bug.cgi?id=35371

            Bug ID: 35371
           Summary: [OrcJIT] Incorrect symbol resolution of libc function
                    addresses on MacOS
           Product: libraries
           Version: 3.9
          Hardware: Macintosh
                OS: MacOS X
            Status: NEW
          Severity: normal
          Priority: P
         Component: OrcJIT
          Assignee: unassignedbugs at nondot.org
          Reporter: 1101.debian at gmail.com
                CC: llvm-bugs at lists.llvm.org

Created attachment 19450
  --> https://bugs.llvm.org/attachment.cgi?id=19450&action=edit
JIT runner source code

(Related to https://bugs.llvm.org/show_bug.cgi?id=35370).

I feed an object file to the ObjectLinkingLayer and the run the 'main' function
that is defined in the object file. It works on Linux but does not on MacOS.
On MacOS it simply crashes saying:

error: memory read failed for 0x8af86200

After debugging, I discovered what leads to this behavior. The crash happens
around this code:

    /// Disasm from LLDB
    0x1035a9287: cmpl   $0x1, -0x8(%rbp)
    0x1035a928b: movl   %eax, -0x24(%rbp)
    0x1035a928e: jne    0x1035a92b1
    0x1035a9294: leaq   -0x1190(%rip), %rdi
    0x1035a929b: movb   $0x0, %al
    0x1035a929d: callq  0x8adf0180

This assembly corresponds to the following code extracted from the object file
using otool (otool -tV main.o):

    /// Original code
    000000000032 callq  _printf
    000000000037 cmpl   $0x1, -0x8(%rbp)
    00000000003b movl   %eax, -0x24(%rbp)
    00000000003e jne    0x61
    000000000044 leaq   0x160(%rip), %rdi ## literal pool for: "no test
specified\n"
    00000000004b movb   $0x0, %al
    00000000004d callq  _printf


After few iterations, I discovered that the instruction in the original code
preceding the first instruction in the disassembled code from LLDB is always
has something to do with a libc function. In fact, if I remove any such calls
then everything works just fine.
Also, it works correctly on Linux.

Note: the code is working if I compile jit runner with address sanitizer
enabled.
Another note: I am not sure if it is the actual bug in the Orc itself or I am
doing something wrong.

I attach the JIT runner (main.cpp).
I compiled it using the following command:

clang++ `llvm-config --cxxflags` `llvm-config --libs` -lz -lncurses main.cpp -o
jitter

The I built simple 'hello world' program:

clang++ -c main.cpp

And fed it to the runner:

./jitter main.o

P.S. I tried to reproduce it using LLVM 5.0 but hit another problem:
https://bugs.llvm.org/show_bug.cgi?id=35370

-- 
You are receiving this mail because:
You are on the CC list for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-bugs/attachments/20171121/c4d07b7b/attachment-0001.html>