🚀 go-pugleaf

In article <1992Dec25.033918.3246@beaver.cs.washington.edu> kolding@cs.washington.edu (Eric Koldinger) writes:
>There's a fundamental difference between hardware and software capabilities.
>Software capabilities can be forged, however improbable.

You may have seen such a system, but some software-based systems
really are fully secure. The OS keeps all of the unforgable bit
patterns in some system space. The user processes manipulate values
which the OS effectively treats as indexes. Forging an index gains a
process no abilities that it didn't have already.

>On an architecture like the AS/400, EVERY memory reference is
>checked against a capability, and individual bytes can be protected.
>Software capabilities are typically used to protect larger grained
>objects, and are only checked when that object is "bound" to your
>protection domain.

Agreed. However, conventional machines check every memory reference
with an MMU. On these, binding an object in === teaching the MMU
about the object. So, the real difference is that conventional MMUs
are page-oriented rather than byte-span oriented. [Well, the time
cost of a binding probably differs.]

There have been MMUs that allowed fractional pages - for example,
the front part, or the rear part, of a page could be missing.
The PDP11/45 supported this with a 64-byte granularity.

The AS/400 may support tiny objects, but does it gain much from doing
so?
-- 
Don		D.C.Lindsay	Carnegie Mellon Computer Science