Re: [PATCH] x86: better CPU identification without the CPUID

[Krzysztof Helt <krzysztof.h1@wp.pl>]

On Sat, 13 Sep 2008 22:55:10 -0700
"Yinghai Lu" <yhlu.kernel@gmail.com> wrote:

> On Sat, Sep 13, 2008 at 9:43 PM, Yinghai Lu <yhlu.kernel@gmail.com> wrote:
> > On Sat, Sep 13, 2008 at 3:56 AM, Krzysztof Helt <krzysztof.h1@wp.pl> wrote:
> >> From: Krzysztof Helt <krzysztof.h1@wp.pl>
> >>
> >> cpus without the CPUID instruction are identified
> >> as general 386 or 486 while some cpus (mostly made
> >> by Cyrix) provide c_identify function which identify
> >> correctly older cpus using cpu specific registers).
> >>
> >> Cyrix cpus are even worse as 5x86 and 6x68 have
> >> the CPUID instruction disabled. The CPUID is
> >> enabled by the c_identify() but the c_identify
> >> is only called when the CPUID is available.
> >>
> >> Fix this by calling the c_identify() for all known
> >> cpu families if there is no the CPUID instruction
> 
> I updated it to tip/master and call that in early_identify_cpu. it
> should solve mtrr detection for Cyrix ...
> 


As I wrote I have no CPU to test this (Cyrix with ARRs). Finally, I am going
 to buy one or two of them during next two weeks so I will test.

I would like to postpone your patch until I (or someone else) test it.

There is some misunderstanding about the patch. There are two issues here.
The early enabling of mtrr registers and nicer and more accurate /proc/cpuinfo
content (and the CPU's name during boot). My patch was only about the latter.
It was not about the early identification or solving the mtrr problem (as without 
testing I am not sure I can correctly do this). Your patch resets the /proc/cpuinfo
content to the old status (before my patch).

> From: "Yinghai Lu" <yhlu.kernel@gmail.com>
>
> [PATCH] x86: identify_cpu_without_cpuid
> 
> need to call c_identify() for cpus without cpuid earlier...
> 
> Signed-off-by: Yinghai Lu <yhlu.kernel@gmail.com>


>  /*
>   * Do minimum CPU detection early.
>   * Fields really needed: vendor, cpuid_level, family, model, mask,
> @@ -503,18 +530,17 @@ static void __init early_identify_cpu(st
>  #endif
>  	c->x86_cache_alignment = c->x86_clflush_size;
>  
> -	if (!have_cpuid_p())
> -		return;
> -
>  	memset(&c->x86_capability, 0, sizeof c->x86_capability);
> -
>  	c->extended_cpuid_level = 0;
>  
> -	cpu_detect(c);
> -
> -	get_cpu_vendor(c);
> -
> -	get_cpu_cap(c);
> +	if (!have_cpuid_p()) {
> +		identify_cpu_without_cpuid(c);
> +		return;
> +	} else {
> +		cpu_detect(c);
> +		get_cpu_vendor(c);
> +		get_cpu_cap(c);
> +	}
>  
>  	if (this_cpu->c_early_init)
>  		this_cpu->c_early_init(c);

One wants to call c_early_init() here for Cyrix cpus- otherwise 
the mtrr caps are not set. The correct patch here is only:

 -
 -	if (!have_cpuid_p())
 -		return;

- 	    cpu_detect(c);

+	if (!have_cpuid_p())
+		enable_cpuid_on_some_cpus(); /* pseudo code for now */

+	if (have_cpuid_p())
+ 	    cpu_detect(c);


The enabling of the cpuid instruction should be added there but only the enabling.

> @@ -583,11 +609,13 @@ static void __cpuinit detect_nopl(struct
>  
>  static void __cpuinit generic_identify(struct cpuinfo_x86 *c)
>  {
> -	if (!have_cpuid_p())
> -		return;
> -

I have just checked:
http://git.kernel.org/?p=linux/kernel/git/x86/linux-2.6-tip.git

and it differs here. It is the same as in my patch (who is out of sync?).

> @@ -639,17 +667,6 @@ static void __cpuinit identify_cpu(struc
>  	c->x86_cache_alignment = c->x86_clflush_size;
>  	memset(&c->x86_capability, 0, sizeof c->x86_capability);
>  
> -	if (!have_cpuid_p()) {
> -		/*
> -		 * First of all, decide if this is a 486 or higher
> -		 * It's a 486 if we can modify the AC flag
> -		 */
> -		if (flag_is_changeable_p(X86_EFLAGS_AC))
> -			c->x86 = 4;
> -		else
> -			c->x86 = 3;
> -	}
> -
>  	generic_identify(c);
>  
>  	if (this_cpu->c_identify)

The identify_cpu can stay the way it was. No changes here needed.
My patch improves the /proc/cpuinfo content. The next patch should
only enable the cpuid instruction on cyrix.

Regards,
Krzysztof