Performance of SheevaPlug on 8-stable

[Rafal Jaworowski]

On 2010-03-07, at 20:56, Maks Verver wrote:

> On 03/07/2010 08:00 AM, Bernd Walter wrote:
>> That's probably just because of different CPUs.
>> I see a similar output on all of my systems with ARM920T CPU and
>> still there is something wrong.
> 
> That's strange indeed. I'm not sure if our problems are at all related
> (as in: caused by the same problem) as you seem to be using fairly
> different hardware.
> 
> In my case the kernel (at boot up) doesn't seem to even think caches are
> enabled, which gives me some hope that if they were, then they would
> work. In your case the kernel claims to enable them but then they don't
> work. Seems different to me.

The reason there is not output about cache ena/dis-able status for the Sheeva CPU is just a minor bug. Please try the patch below to get the status info at boot time:

Index: sys/arm/arm/identcpu.c                                                   
===================================================================             
--- sys/arm/arm/identcpu.c      (wersja 204729)                                 
+++ sys/arm/arm/identcpu.c      (kopia robocza)                                 
@@ -415,6 +415,7 @@ identify_arm_cpu(void)                                      
        case CPU_CLASS_ARM9EJS:                                                 
        case CPU_CLASS_ARM10E:                                                  
        case CPU_CLASS_ARM10EJ:                                                 
+       case CPU_CLASS_MARVELL:                                                 
        case CPU_CLASS_SA1:                                                     
        case CPU_CLASS_XSCALE:                                                  
        case CPU_CLASS_ARM11J:      

I would expect L1 caches are enabled on your system (please verify with the above patch) and it must be some other problem. If the caches were disabled the whole system would crawl (including networking). I also checked on another 6281-based dev board running with caches ON and observed the same ill effect.

Rafal