> In article <linux.kernel.Pine.LNX.3.91.970718152603.26810F-100000@sigil.csc.com>,
> Teunis Peters <teunis@usa.net> wrote:
> >On Fri, 18 Jul 1997, Mike Jagdis wrote:
> >
> >> On Thu, 17 Jul 1997, Aaron Tiensivu wrote:
> >>
> >> > Coming up in the kernel, I think the whole chip definition section should
> >> > be overhauled into some type of sassy multi-selection deal.
> >>
> >> This is excessive. You don't need this level of detail and putting
> >> processor options into config rather than looking at the processor
> >> type at run time makes portable kernel images tricky. Anyway my
> >> non-Intel patch does something along these lines.
> >
> >I think this kind of detail is _GOOD_!
> >(priorities in terms of importance to _ME_ :)
>
> Well, it's bad in that it tosses more knowledge about the processor
> into the build stage; if you're building general purpose kernels
> (like a distribution, or for a network appliance like WebShield)
> it means that the code taking advantage of the features of the
> processor will be left off the kernel.
You have a point - there are some features that must be chosen at
compile-time when dealing with CPU optimizations.
Most of these 'features' are alpha-level and unknown stability. Most have
warnings about potential danger.....
On the other hand [when dealing with Intel clones], having kernel support
for multiple CPU support so bugfixes for different CPU's can be chosen at
runtime.
Doesn't the kernel already do this for buggy CPU's like pentiums,
pentium-pros, pentium-II's and the like? Why not extend it for the
clones? [are any known bugs for, say, cyrix 6x86's?]
On dealings with other platforms - Being able to select 'Alpha' as a
processor (eg) would be nice and not just restrict to Intel-types....
Why not [for main screen]:
Intel -> [386-series] -> math-co support?
bugfixes? (they had some :)
[486-series] -> math-co support?
486-clone? (eg. 386 with delusions)
bugfixes - list of bug patches to include
[586-series] -> pentium
-> pentium optimizations?
5x86 (486 with delusions)
-> 586 optimizations?
AMD ????
-> ??? optimizations?
pentium-MMX (who cares?)
-> MMX optimizations?
(network speedup :)
bugfixes - list of bug patches to include
[686-series] -> pentium-pro
AMD K6
-> K6 optimizations
6x86 (Cyrix, IBM)
-> 6x86 optimizations
pentium-II
bugfixes - list of bug patches to include
optimizations - applicable to all
(MMX with most?)
Alpha -> (?)
Sparc -> sparcstation
sparc-64
(?)
Mips -> I don't know...
PowerPC -> native (aren't there multiple motherboard-interfaces?)
mklinux (or something like that)
68K -> Amiga
Atari
Macintosh -> series of supported platforms
... and the like...
then make dep/make config could select 'asm' and 'arch' options based on
selected CPU.
It's just an idea, yes?
If anyone wants to see this, I could check into building it...
[The only platforms I can access though are 486/SX, 6x86 Cyrix, and
486/DX - the powerbook won't run Linux (I prefer MacOS on it)]
Please let me know!
Wind to your wings!
- Teunis