Issues do exist with reader-writer spinlocks. One typical issue with it is that, unfortunately, writers can starve when blocking on several readers. Think about it: let's say that three reader threads currently have the reader-writer lock. Now, a writer comes along wanting the lock. It has to wait until all three readers perform the unlock. But what if, in the interim, more readers come along (which is entirely possible)? This becomes a disaster for the writer, who has to now wait even longer – in effect, starve. (Carefully instrumenting or profiling the code paths involved might be necessary to figure out whether this is indeed the case.)
Not only that, cache effects – known as cache ping-pong – can and do occur quite often when several reader threads on different CPU cores are reading the same shared state in parallel (while holding the reader-writer lock); we in fact discuss this in the Cache effects and false...
