On Jan 4, 2006, at 9:38 AM, jonathan@dsg.stanford.edu wrote:

> Oh, one more thing: to suggest using ras(9) is proof-positive that the
> suggester simply doesn't understand the problem domain.  Bershad's
> work on restartable atomic sequences, of which ras(9) is a
> re-implementation, gives atomic semantics to *USERSPACE PROCESSES*,
> implemented by adding "check for RAS collision during trap-to-OS"
> hooks and PC-unwind in one (or a very, very small integer number) of
> codepaths. Thus, RAS is not suitable for use *inside* a kernel.

Actually, it is possible to use RAS inside a kernel, and Solaris does  
(did) it on some platforms for low-level mutex operations on  
uniprocessor systems.

-- thorpej