I implemented a IPC msg Q for a router company 15 years ago. Passing msg with different Threads/Process is not "rolling your own locks with atomic variables".

I implemented a IPC msg Q for a router company 15 years ago. Passing msg with different Threads/Process is not "rolling your own locks with atomic variables". Basically, each T/P need to receive msg will create MsgQ with msgQLen and return Q_id. msgQ_Send() Api will use atomic_add() to get the next slot on the msgQ to place the value and/or MsgPtr. There is no lock needed, only Atomic_add(). Performance wise, it works very well in SMP environments and I ported the library to vxWorks, Linux kernel, user space and QNX. I have tons of validation tests on the correctness of the message orders, contain and constantly run them on multiple 4 core Xeon machine to check the correctness and benchmark performance.