OpenIoT & ELC 2016

Memory barriers are hard, poorly understood, and easy to mess up, creating every so subtle bugs in system software. They are, nonetheless, necessary, enabling lockless, non-blocking algorithms and data structures on SMP systems, and therefore very much a reality in the Linux kernel. For a while now, embedded systems are becoming multi-core and therefore are exposed to the pros and cons in SMP concurrent scenarios influenced by barriers.

This presentation hopes to update the audience on the increasingly changing barrier interfaces in the kernel, illustrating a number of new calls that build upon and extend the basic read/write semantics to very ad-hoc situations. Understanding when and how to use them are important when integrating new functionality in the kernel that make use of barriers.