[Fwd: Re: Kernel 2.4 CPU Arch issues]

[Fwd: Re: Kernel 2.4 CPU Arch issues]

[William M. Quarles]

-------- Original Message --------
Subject: Re: Kernel 2.4 CPU Arch issues
Date: Sun, 20 Jul 2003 19:23:24 -0400
From: William M. Quarles <quarlewm@jmu.edu>
Organization: James Madison University
To: Mark Hahn <hahn@physics.mcmaster.ca>
References: 
<Pine.LNX.4.44.0307201920260.4900-100000@coffee.psychology.mcmaster.ca>

Mark Hahn wrote:
>>Something like this ?
> 
> 
> is there any real point?  what post-p6 features would you expect
> to make a difference?
> 

Well, you separated the Pentium and Pentium-MMX.  It's the exact same
difference between Pentium Pro and Pentium-II: MMX technology.  That's
the point.

-- 
William M. Quarles

quarlewm@jmu.edu
wquarles@bucknell.edu
walrus@bellsouth.net


-- 
William M. Quarles

quarlewm@jmu.edu
wquarles@bucknell.edu
walrus@bellsouth.net

Re: [Fwd: Re: Kernel 2.4 CPU Arch issues]

[Alan Cox]

On Llu, 2003-07-21 at 00:29, William M. Quarles wrote:
> Well, you separated the Pentium and Pentium-MMX.  It's the exact same
> difference between Pentium Pro and Pentium-II: MMX technology.  That's
> the point.

This makes no difference to the kernel. Splitting PPro would only make
sense for one reason. The Pentium Pro needs store barriers on
spin_unlock and friends, the PII and later do not. However if this was
done you'd also want to check for PPro boots with a PII kernel and panic
which isn't currently done

Re: [Fwd: Re: Kernel 2.4 CPU Arch issues]

[William M. Quarles]

Alan Cox wrote:
> On Llu, 2003-07-21 at 00:29, William M. Quarles wrote:
> 
>>Well, you separated the Pentium and Pentium-MMX.  It's the exact same
>>difference between Pentium Pro and Pentium-II: MMX technology.  That's
>>the point.
> 
> 
> This makes no difference to the kernel. Splitting PPro would only make
> sense for one reason. The Pentium Pro needs store barriers on
> spin_unlock and friends, the PII and later do not. However if this was
> done you'd also want to check for PPro boots with a PII kernel and panic
> which isn't currently done
> 

Well, wouldn't changing the gcc -march option and/or adding -mcpu 
options for the various processors in the Makefile make a difference, as 
the patchfile suggests?

-- 
William M. Quarles

quarlewm@jmu.edu
wquarles@bucknell.edu
walrus@bellsouth.net

Re: [Fwd: Re: Kernel 2.4 CPU Arch issues]

[Alan Cox]

On Llu, 2003-07-21 at 01:25, William M. Quarles wrote:
> Well, wouldn't changing the gcc -march option and/or adding -mcpu 
> options for the various processors in the Makefile make a difference, as 
> the patchfile suggests?

Currently - no. gcc knows a lot more processor names that require individual
unique optimisation

Re: [Fwd: Re: Kernel 2.4 CPU Arch issues]

[jw schultz]

On Mon, Jul 21, 2003 at 07:39:16AM +0100, Alan Cox wrote:
> On Llu, 2003-07-21 at 01:25, William M. Quarles wrote:
> > Well, wouldn't changing the gcc -march option and/or adding -mcpu 
> > options for the various processors in the Makefile make a difference, as 
> > the patchfile suggests?
> 
> Currently - no. gcc knows a lot more processor names that require individual
> unique optimisation

no && s/that/than/  


-- 
________________________________________________________________
	J.W. Schultz            Pegasystems Technologies
	email address:		jw@pegasys.ws

		Remember Cernan and Schmitt

Re: [Fwd: Re: Kernel 2.4 CPU Arch issues]

[J.A. Magallon]

On 07.21, Alan Cox wrote:
> On Llu, 2003-07-21 at 00:29, William M. Quarles wrote:
> > Well, you separated the Pentium and Pentium-MMX.  It's the exact same
> > difference between Pentium Pro and Pentium-II: MMX technology.  That's
> > the point.
> 
> This makes no difference to the kernel. Splitting PPro would only make
> sense for one reason. The Pentium Pro needs store barriers on
> spin_unlock and friends, the PII and later do not. However if this was
> done you'd also want to check for PPro boots with a PII kernel and panic
> which isn't currently done
> 

(sorry for the late answer, I have been out for a course...)

Is this done now ? Where is detected if you try to boot a P3-built kernel on a
PPro ?

AFAIK, P2 is more similar to a P3 than to a PPro in terms of architecture.
Some people say that P3=P2+SSE-50%cache.
As features like fxsr or cmov are detected independent of gcc flags
(/proc/cpuinfo...) I think this can affect on how gcc schedules instructions.
If I grep -r IUMIII * /usr/src/linux, I just get this:

/* Prefetch instructions for Pentium III and AMD Athlon */
#if defined(CONFIG_MPENTIUMIII) || defined (CONFIG_MPENTIUM4)

#define ARCH_HAS_PREFETCH
extern inline void prefetch(const void *x)
{
    __asm__ __volatile__ ("prefetchnta (%0)" : : "r"(x));
}                                                                               

And in terms of CONFIG_ flags, the differences from 686 to PIII are:

-  bool 'PGE extensions (not for Cyrix/Transmeta)' CONFIG_X86_PGE
+  define_bool CONFIG_X86_PGE y
-  define_bool CONFIG_X86_PPRO_FENCE y

In short:
- P3 is separate from 686 just for the prefetch and for the PPRO_FENCE
- P2 can kill also the PPRO_FENCE

so you I can suppose the main reason to separate P3 is that you trust gcc
to generate better code if it knows its a P3. So same applies to P2, as
gcc explicitely says that i686 is a pentiumpro, but says nothing about
being also a P2:

info gcc:

`-mcpu=CPU-TYPE'
     ...
     While picking a specific CPU-TYPE will schedule things
     appropriately for that particular chip, the compiler will not
     generate any code that does not run on the i386 without the
     `-march=CPU-TYPE' option being used.  `i586' is equivalent to
     `pentium' and `i686' is equivalent to `pentiumpro'.  `k6' and
     `athlon' are the AMD chips as opposed to the Intel ones.

;)

-- 
J.A. Magallon <jamagallon@able.es>      \                 Software is like sex:
werewolf.able.es                         \           It's better when it's free
Mandrake Linux release 9.2 (Cooker) for i586
Linux 2.4.22-pre7-jam1m (gcc 3.3.1 (Mandrake Linux 9.2 3.3.1-0.6mdk))