GC_allow_register_threads get stuck when it's called implicitly when DLL is loaded (on Windows)

[butterunderflow]

Recently, I try to make bdwgc get initialized when I load a library, so I wrote these code in my shared library test.cpp:

#include <iostream>
extern "C" {
extern void GC_init();
extern void GC_allow_register_threads();
}
void test() { std::cout << "Hello! I'm from a dll lib" << std::endl; }

class dllInitializer {
public:
  dllInitializer() {
    GC_init();
    GC_allow_register_threads();
  }
};
dllInitializer dllInitializerInstance;

and compile it to a dll library like this(the bdwgc/build/libgc.dll is compiled by cmake toolchain):

g++ test.cpp bdwgc/build/libgc.dll -o libtest.dll --shared

But unfortunately, when I load this library from my main executable main.cpp:

// compile this file with: 
//      g++ main.cpp libtest.dll -o main.exe

#include <cstdio>
void test();

int main()
{
    printf("Started!\n");
    test();
}

it get stuck:

$ ./main.exe
# it just hangs there

Then I enter gdb, it shows some information now, but the process still hangs, which looks like a dead lock:

GNU gdb (GDB) 11.2
Copyright (C) 2022 Free Software Foundation, Inc.
License GPLv3+: GNU GPL version 3 or later <http://gnu.org/licenses/gpl.html>
This is free software: you are free to change and redistribute it.
There is NO WARRANTY, to the extent permitted by law.
Type "show copying" and "show warranty" for details.
This GDB was configured as "x86_64-w64-mingw32".
Type "show configuration" for configuration details.
For bug reporting instructions, please see:
<https://www.gnu.org/software/gdb/bugs/>.
Find the GDB manual and other documentation resources online at:
    <http://www.gnu.org/software/gdb/documentation/>.

For help, type "help".
Type "apropos word" to search for commands related to "word"...
Reading symbols from main.exe...
(gdb) run
Starting program: <path-to-exe's-dir>main.exe
[New Thread 8116.0xae4c]
[New Thread 8116.0x7914]
[New Thread 8116.0x4e4c]
[New Thread 8116.0xa018]
[New Thread 8116.0x28dc]
[New Thread 8116.0x570]
[New Thread 8116.0x9cc0]
[New Thread 8116.0x7184]
[New Thread 8116.0x8324]
[New Thread 8116.0x6d60]
[New Thread 8116.0x7fd4]
[New Thread 8116.0xb0a0]
[New Thread 8116.0xa73c]
[New Thread 8116.0x7ad4]

Any help or suggestion is appreciated!

PS

The build configuration of bdwgc:

 cmake -B build -Denable_disclaim=ON -Denable_atomic_uncollectable=ON -Denable_java_finalization=ON -Denable_mmap=ON -Denable_munmap=ON -Denable_threads=ON -Denable_parallel_mark=ON -Denable_thread_local_alloc=ON  -DCMAKE_USE_WIN32_THREADS_INIT=true -G "MinGW Makefiles"

[butterunderflow]

I created a repository to reproduce the behavior in this issue. I hope that will illustrate my question more clear.

https://github.com/butterunderflow/bdwgc-issue-659-reproduce

[ivmai]

Is this reproduced on bdwgc master?

[butterunderflow]

Is this reproduced on bdwgc master?

I just tried on the latest master, the program still stuck at dll loading.

[butterunderflow]

Would anyone help me verify if this is a bug or if some configuration of mine is wrong?

[ivmai]

As mentioned in win32_threads.c, code invoked from a dll initializer should be lock-free, but code of GC_allow_register_threads() is not. I think, if you want to perform the GC initialization implicitly, you should compile bdwgc with -Denable_threads_discovery=ON and remove GC_init and GC_allow_register_threads() calls.