Arjan van de Ven wrote:
> Alan Jenkins wrote:
>>> I notice that the "states:" is empty???
>>>
>> That confirms what powertop said. It's what I would expect.
>
> yeah the bios doesn't export C-states to the OS.
>
>> However, even with the fix, my BIOS doesn't provide any *C* state
>> information. That's because the BIOS manages the different non-zero
>> C-states itself, and linux doesn't see them.
>>
>> This is known as "C1E",
>
> that is only true for AMD systems btw; "C1E" is defined by AMD to be
> "we'll break all the rules
> and pick the C-state ourselves", on Intel systems C1E is "we'll go to
> the lowest voltage in idle".
I have an Core Duo 2 myself. I was mainly thinking of the difference
between C1 on one core only, and C1 in both cores. The voltage has to be
the same in both cores, and AFAIK it's the BIOS that ensures that, not
linux. E.g. for core 1, idle/C1 maps to two different states, with the
lowest power state ("C1E") entered only if core 2 is also in C1.
I wasn't being very precise. Sorry if I've been spreading confusion.
> I'm not trying to be evil towards AMD by saying "we'll break all the
> rules", they really do break
> all the rules around C-states. (There are guarantees about what the OS
> can expect for the different
> C-states, C3 for example is very different from C2 and that again is
> very different from C1. This automagic
> going deeper gives you C3 behavior even though only C1 is asked for.
> This broke Linux quite severely
> and is the cause of many workarounds in the tickless idle code
> already, and I suspect more are to come
> still)
I'd heard of AMD C1E problems but not read the explanation. Isn't C1E
is invoked automatically on HLT? Some BIOS's don't even give you an
option to disable it. I'd have thought going into C3 on HLT would have
broken Windows (the versions that do use HLT).
Alan
Alan
_______________________________________________
Power mailing list
[email protected]
http://www.bughost.org/mailman/listinfo/power
