[crest]

It is a cool trick and even without virtual memory if the maximum read size has a low upper bound just copying a the first few bytes from the beginning to the end can be worth it to avoid the complexity and overhead of indexing into the ring buffer modulo its size.

This clever ring buffer has hidden costs that don't show up in microbenchmarks testing only the push() and pop() operations on an initialized ring buffer. These can be very high and impact other software as well.

The initialisation requires multiple system calls to setup the aliased memory mapping and it has to be a multiple of the page size. This can be very expensive on big systems e.g. shooting down the TLBs on all other CPUs with interprocessor interrupts to invalidate any potentially conflicting mappings. On other end small systems often using variations of virtual indexed data caches and may be forced to set the aliased ring buffer pages as uncacheable unless you can get the cache coloring correct. And even for very small ring buffers you also use at least two TLB entries. A "dumber" ring buffer implementation can share part of a large page with other data massively reducing the TLB pressure.

Abusing the MMU to implement an otherwise missing ring addressing mode is tempting and can be worth it for some use-cases on several common platforms, but be careful and measure the impact on the whole system.