|
Authored by: Wol on Thursday, September 20 2012 @ 06:47 PM EDT |
I expect you're thinking of Intel segmented memory.
If memory serves me right, each segment/offset pair on Primes guaranteed a
unique memory location - you couldn't have multiple pointers to the same
location.
The OS was below o4000, each user had their own user space at about o6000, and I
think o7000 and above was reserved for something else.
I never really understood all this stuff, but one of the things that stands out
from what I remember is that snapping a dynamic link was very fast, and coupled
with that a call into the operating system was EQUALLY as fast!
One of the big problems with linux and speed is that a user/kernel-space context
switch is relatively very expensive. With Primos it cost almost nothing.
And if (as I suspect it did) you could compile your kernel such that it insisted
on its writeable memory being in the o6000 address range, then security is
moderately easy to achieve - seeing as each user has their own separate 6000
range this code can't stomp over other users OR the kernel.
Okay, I'm sure other people here will tell me there were horrendous flaws in
Primos that let you break in (I gather security was actually pretty poor, thanks
to various bugs and/or design features), but the basic architecture design was
pretty good.
Cheers,
Wol[ Reply to This | Parent | # ]
|
- noooo! - Authored by: greed on Tuesday, September 25 2012 @ 06:25 PM EDT
|
|
|
|