Re: Intel iAPX 432

In article <[EMAIL PROTECTED]>,
Jan C. Vorbrüggen  <[EMAIL PROTECTED]> wrote:
>> (Later we realised the flip-side of our decision -- priority inversions,
>> which did bite on a few rare occasions.  Oh well.)
>
>How does buffering help with priority inversion? I believe it can even
>make things worse (see Mars Pathfinder, e.g., in comp.risks archives).

I didn't mean to imply that.  Buffering can complicate the situation
considerably.  Indeed, it may even be a good thing that rendezvous
leads to in-your-face priority inversion, so when something gets stuck,
it is easy to figure out why, and deal with it.

Btw, we do have a second IPC primitive, Signal, which causes an interrupt
and can pass a small fixed amount of information to another process.  This
is typically used to request a rendezvous, at which point an unbounded 
amount of information can be exchanged.  It can also be used to break out
of a rendezvous, which is easy, since our rendezvous is stateless.  (It is
one of those instructions that the kernel can break out of as if the instr
had never been attempted.)

Michel.