> In <Pine.LNX.3.96.981018115925.22009A-100000@z.ml.org> Gregory Maxwell (linker@z.ml.org) wrote:
>
> GM> My linux is SLLLOOOWWWW.. My linux is SLLLOOOWWWW..
>
> GM> The dists should be shipping two sets of kernels. Both with PPRO opt
> GM> turned on and MTRR support. This leaves out the i386. SO? What the hell
> GM> are you doing Running RedHat 5.5 w/ a 2.2 kernel on a i386?!? PPro opt
> GM> will work on i486+.
>
> I'm not so sure. AFAIK kernel compiled for Pentium (let alone PPro) could do
> some nasty things under 486 :-(( There are some difference in paging support...
No. They work fine. RTFM:
"This is the processor type of your CPU. This information is used for
optimizing purposes. In order to compile a kernel that can run on
all x86 CPU types (albeit not optimally fast), you can specify
"386" here. If you specify one of "486" or "Pentium" or "PPro",
then the kernel will run on all of these CPUs: 486 and Pentium
(=586) and Pentium Pro (=686). In rare cases, it can make sense to
specify "Pentium" even if running on a 486: the kernel will be smaller
but slower."
It also goes on to say that you should NOT run a SMP kernel on a UP
system.
> GM> MTRR support must be on w/ PPRO+ SMP because some MBs dont set MTRRs
> GM> correctly on both CPUs..
>
> MTRR should be compatible with 486-optimized kernel :-)
It is. However, more of most dists target audience are using chips that
like PPRO optimazations (i.e. K6, PII) today. Make it work for the lowend
while making it fast for the average case. If someone is running a i486
then they can take the slight performance hit of a PPRO optimized kernel.
> GM> Most people have UP. The install's failure modes should set the kernel to
> GM> UP in any case.
>
> Why ? How SMP support could harm in failure mode ? Only if it will prevent
> kernel from booting :-))
There are bugs that show up on SMP on UP. Idealy that should be fixed, but
I would consider it no more a priority then making ALPHA kernels work on
x86.
It also harms by making the system slower. We dont need mag reviews where
they show that NT5 is faster then Linux because linux installed a SMP
kernel.
> GM> :) I know how to do it. You enhance the boot loaded so it can boot a 'fat'
> GM> kernel. (i.e. all the ARCHS catted togeather) The same would work for SMP.
> GM> I'm not so sure I want 100megs of my HDD eaten up by a fat kernel and fat
> GM> modules.
>
> Old BIOS'es does not support more then 1.44" floppy image as "boot image" for
> bootable CD's :-((
The kernel need not be on the boot floppy image. All that needs to be
there is a first stage boot loader that knows how to read the kernel from
whereever it is (be it on a cd or on the network)..
> Hm. Why Linux should not be first such OS ?
Umm.. Because APM is designed in a SMP unfrendly way.. It is explicitly
not supposed to work.. It would be a bunch of terrible hacks to get it to
work.. And for what? So your computer can turn itself off? COME ON! :)
[I'm saying you could do poweroff under SMP]
> No. You could not call APM BIOS safely under SMP at all :-(( It's possible to
> stop second proc and call SMP (in theory) but it's unclear if this is worth
> such troubles. And this is not 2.2 issue anyway :-)
Umm.. Safty doesn't matter much when you are turning off the computer. I
imagine you can place real mode APM poweroff code at the right location in
memory and triple fault the cpu..
> GM> It is a big difference.
>
> When ? What kind of work you should do to see more then few per cent's
> difference in speed of SMP and UP kernel on UP comp ?
It increases memory usage.. Lots of critical code paths are slowed..
> It's not possible if we need to find "correct configurations" out of
> 1'000'000'000'000'000'000'000'000'000'000+ possible configurations (if we
> had only 100 options in kernel -- now we have more then 100 options there and
> thus more then 1'000'000'000'000'000'000'000'000'000'000 possible
> configurations) -- Earth simple does not enough have resources to do this :-((
> We should have few (no more then 5-10 and better only one) which will
> fit 99.9% users... With modules usage of course :-))
Is your head made of rock?
SMP and Arch is the ONLY thing that touches the ENTIRE KERNEL! IT CHANGES
HOW LOTS OF THINGS WORK! I'm not saying we should throw out modules..
I'm saying that we should treat SMP like a differnt ARCH cause that's
ALMOST what it is.
Enough! I'm right, your wrong! n to infinity! So THERE! HA! :P
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/