Re: x86 architecture questions

From: Scott Moore (samiam_at_moorecad.com)
Date: 03/26/04 

Date: Fri, 26 Mar 2004 06:56:50 +0000 (UTC)

"Robert Wessel" <robertwessel2@yahoo.com> wrote in message news:bea2590e.0403251548.28fbc898@posting.google.com...
> "David A. Caabeiro" <me@privacy.net> wrote in message news:<c3unk5$22br3c$1@ID-84876.news.uni-berlin.de>...
> > > I think my confusion comes from when you said:
> > >
> > > "Then you will find out that Intel and others are
> > > abandoning segmentation, ring privledges, and TSS
> > > faster than a cat abandoning a pepper sandwitch"
> > >
> > > I don't think Intel will ever abandon privilege levels. Perhaps you
> > > meant privilege levels 1 and 2? If applications were allowed to run
> > > so-called privileged instructions, how could an OS isolate processes?
> >
> > He's talking about privilege *rings*, which is a segmentation
> > unit check; he meant -I hope- removing all four rings.
> > As he stated, this is different from page-level protection.
>
>
> Not unless some other method is used to separate privileged from
> unprivileged code. Ring-0 is special in that you can actually execute
> the privileged instructions in ring-0 - that's not impacted by any of
> the page protection stuff.
>
>

See other replies in this thread. You don't need the priviledges in
paged management based code. The best way is if the kernel does not
even reveal any of its code to the target (user) processes. Each
run in their own virtual space. There is no "key" to protected space
required, because such space is not mapped in common to user space
unless the kernel wants that.

Yes, I am serious. Those features are clearly on their way out.
Intel is moving them to the emulation unit, which means the are
slower, perhaps a lot slower, and Intel is deprecating them. Keep
in mind, Intel is even now recommending that you set up your OWN
task state tables and switching code. Thats a big change for them.

Whats really odd is (to my understanding) Windows uses do-it-yourself
TSS and Linux uses the hardware supported TSS. You would think it
would be the opposite.

Bottom line: IBM pushed virtual memory in the 360 product line, and
MIT thought up segments and rings as a cheaper solution. Virtual
memory won. Its over.

Relevant Pages

  • Re: Crazy Apple! Crazier Mac kOOks.
    ... Intel iMac, sells for a few thousands more. ... A computer, slower, using ... Yes, the new pro towers are going to be faster, but the jump probably ... slower than on the new G5, but probably faster than on the Intel macs. ...
    (comp.sys.mac.advocacy)
  • Re: General slowness with using 64Gb HIGHMEM option on 32-bit kernel
    ... Intel Core 2 Duo E6600@2,4Ghz on Intel i965Q-based ... highmem option and rebooting into it, ... it's work after rc.M script sarted was 5-10 times slower then usual. ... compiled it with 64Gb highmem support. ...
    (Linux-Kernel)
  • Re: New version of Bresenhams Line Algorithm
    ... My processor is INTEL. ... a tiny bit slower than direct multiplication. ... Performance tests when you turn off optimisation are useless anyway, you might easily get situations where the slower code in debug gets much faster once you turn on optimisations. ...
    (rec.games.roguelike.development)
  • Re: Adobe apps languish under Rosetta.
    ... before making the new Intel Macs available? ... hardware that has superseded the previous hardware. ... software slower ... the Adobe apps I use slower, ...
    (comp.sys.mac.advocacy)