Re: Linux v2.6.12-rc6

[Bill Davidsen <davidsen@tmr.com>]

Johannes Stezenbach wrote:
> Brice Goglin wrote:
> 
>>Johannes Stezenbach a écrit :
>>
>>>Johannes Stezenbach wrote:
>>>
>>>
>>>>Hype-threading stopped working for me (probably due to
>>>>me not enabling ACPI). dmesg output and .config attached.
>>>>-rc5 worked fine. The board is an Asus P4P800-Deluxe.
>>>>
>>>>dmesg: WARNING: 1 siblings found for CPU0, should be 2
>>>
>>>
>>>Indeed SMT works fine if I enable ACPI.
>>>Is SMT without ACPI not supported?
>>
>>You can pass acpi=ht into the kernel command line to disable
>>ACPI except the minimum required to get HT support.
> 
> 
> That's nice, but I was thinking along the lines of:
> 
> diff -ur linux-2.6.12-rc6.orig/arch/i386/Kconfig linux-2.6.12-rc6/arch/i386/Kconfig
> --- linux-2.6.12-rc6.orig/arch/i386/Kconfig	2005-06-06 23:16:27.000000000 +0200
> +++ linux-2.6.12-rc6/arch/i386/Kconfig	2005-06-08 13:35:08.000000000 +0200
> @@ -503,7 +503,7 @@
>  
>  config SCHED_SMT
>  	bool "SMT (Hyperthreading) scheduler support"
> -	depends on SMP
> +	depends on SMP && ACPI
>  	default off
>  	help
>  	  SMT scheduler support improves the CPU scheduler's decision making
> 
> Comments? Is this intended?

I would think that you can't do HT without ACPI, so there's no point in 
building in HT scheduling unless you can have HT.

Is that what you were asking? I was hoping someone else would comment.

Scheduling is getting harder and harder to get right... I have this 
thought of a Beowolf cluster of NUMA machines, with each node being HT 
multicore SMP. By "right" I meant "optimal," I'm sure that setup would 
do something reasonable with current scheduling.

-- 
    -bill davidsen (davidsen@tmr.com)
"The secret to procrastination is to put things off until the
  last possible moment - but no longer"  -me