[PATCH] smpboot: cachesize comparison fix in smp_tune_scheduling()

[PATCH] smpboot: cachesize comparison fix in smp_tune_scheduling()

[Jarek Poplawski]

smpboot: cachesize comparison fix in smp_tune_scheduling()

boot_cpu_data.x86_cache_size is signed int and can be < 0 too.

PS: this function is removed from current -mm.


Signed-off-by: Jarek Poplawski <jarkao2@o2.pl>

---

diff -Nurp 2.6.22-rc2-git5-/arch/i386/kernel/smpboot.c 2.6.22-rc2-git5/arch/i386/kernel/smpboot.c
--- 2.6.22-rc2-git5-/arch/i386/kernel/smpboot.c	2007-05-24 09:37:11.000000000 +0200
+++ 2.6.22-rc2-git5/arch/i386/kernel/smpboot.c	2007-05-24 11:48:03.000000000 +0200
@@ -948,7 +948,7 @@ static void smp_tune_scheduling(void)
 	if (cpu_khz) {
 		cachesize = boot_cpu_data.x86_cache_size;
 
-		if (cachesize > 0)
+		if ((long)cachesize > 0)
 			max_cache_size = cachesize * 1024;
 	}
 }

Re: [PATCH] smpboot: cachesize comparison fix in smp_tune_scheduling()

[Andrew Morton]

On Thu, 24 May 2007 12:33:23 +0200
Jarek Poplawski <jarkao2@o2.pl> wrote:

> 
> smpboot: cachesize comparison fix in smp_tune_scheduling()
> 
> boot_cpu_data.x86_cache_size is signed int and can be < 0 too.
> 
> PS: this function is removed from current -mm.
> 
> 
> Signed-off-by: Jarek Poplawski <jarkao2@o2.pl>
> 
> ---
> 
> diff -Nurp 2.6.22-rc2-git5-/arch/i386/kernel/smpboot.c 2.6.22-rc2-git5/arch/i386/kernel/smpboot.c
> --- 2.6.22-rc2-git5-/arch/i386/kernel/smpboot.c	2007-05-24 09:37:11.000000000 +0200
> +++ 2.6.22-rc2-git5/arch/i386/kernel/smpboot.c	2007-05-24 11:48:03.000000000 +0200
> @@ -948,7 +948,7 @@ static void smp_tune_scheduling(void)
>  	if (cpu_khz) {
>  		cachesize = boot_cpu_data.x86_cache_size;
>  
> -		if (cachesize > 0)
> +		if ((long)cachesize > 0)
>  			max_cache_size = cachesize * 1024;
>  	}
>  }

Under what conditions can boot_cpu_data.x86_cache_size be negative?

Have negative values of boot_cpu_data.x86_cache_size been observed in
practice?

Thanks.

Re: [PATCH] smpboot: cachesize comparison fix in smp_tune_scheduling()

[Jarek Poplawski]

On Thu, May 24, 2007 at 04:02:07PM -0700, Andrew Morton wrote:
> On Thu, 24 May 2007 12:33:23 +0200
> Jarek Poplawski <jarkao2@o2.pl> wrote:
> 
> > 
> > smpboot: cachesize comparison fix in smp_tune_scheduling()
> > 
> > boot_cpu_data.x86_cache_size is signed int and can be < 0 too.
...
> Under what conditions can boot_cpu_data.x86_cache_size be negative?
> 
> Have negative values of boot_cpu_data.x86_cache_size been observed in
> practice?

Sorry, but IMHO every observations are only kind of illusions,
and should have no influence on any serious science, including
computer science.

But, from smpboot.c history:

"[PATCH] i386: Clean up smp_tune_scheduling()
Adrian Bunk [Thu, 7 Dec 2006 01:14:19 +0000 (02:14 +0100)]
- remove the write-only local variable "bandwidth"
- don't set "max_cache_size" in the (cachesize < 0) case:
  that's already handled in kernel/sched.c:measure_migration_cost()"

So, it seems such strange phenomenon could've been observed
long time ago...

Thanks,
Jarek P.