There are plenty of hard lock free algorithms, but an SPSC ring buffer is not exactly rocket science and it is a good introduction on the topic.
Why do you think that a GC is required for a queue? It is certainly very helpful to implement complex node-based intrusive concurrent data structures, but for queues, even node based and even MPMC, as long as each message is produced by one thread and consumed by another one (i.e. the typical use case), single ownership works just fine.
edit: I see you are talking about hidden locks in the allocator. That's fair, but even GC at some point might need to hit the kernel to increase the heap size.
Why do you think that a GC is required for a queue? It is certainly very helpful to implement complex node-based intrusive concurrent data structures, but for queues, even node based and even MPMC, as long as each message is produced by one thread and consumed by another one (i.e. the typical use case), single ownership works just fine.
edit: I see you are talking about hidden locks in the allocator. That's fair, but even GC at some point might need to hit the kernel to increase the heap size.