The project 'Threadprocs' enables the creation of thread-like processes (threadprocs) sharing one address space in a Linux environment on x86_64 and aarch64 architectures. Each threadproc operates independently with its own runtime library instance (libc), global variables, and executable, but pointers are shared across all threadprocs within the same virtual address space. This allows for zero-copy access to pointer-based data structures and direct dereferencing of these pointers without additional copying overhead. However, this shared memory model introduces challenges such as managing allocations and deallocations independently in each threadproc due to the lack of cross-threadproc memory management. Additionally, standard debugging tools like ptrace() are not supported, complicating troubleshooting and maintenance efforts.
- Linux (x86_64 and aarch64)
- Ensure all threadprocs are compiled as position-independent code using -fPIC or equivalent compiler flag.
- Avoid calling malloc() in one threadproc and free() in another by implementing proper memory management within each threadproc context.
- Do not use brk() or sbrk() for dynamic memory allocation due to their global nature across the shared address space.
- Set environment variables MALLOC_MMAP_THRESHOLD_=0 on the server to control glibc behavior.
Homelab stacks using Linux with x86_64 and aarch64 architectures might be impacted if experimenting with advanced memory management techniques. The primary impact would be in configurations where shared address space is utilized, affecting how processes interact and manage their own memory allocations.