Re: [Xen-devel] [PATCH v4 10/11] x86/intel_pstate: support the use of intel_pstate in pmstat.c

["Jan Beulich" <JBeulich@xxxxxxxx>]

>>> On 09.09.15 at 12:35, <wei.w.wang@xxxxxxxxx> wrote:
> On 09/09/2015 18:10,  Jan Beulich wrote:
>>>> On 09.09.15 at 11:35, <wei.w.wang@xxxxxxxxx> wrote:
>>> Using the drinking vessel analogy, we are not putting milk and water 
>>> into the vessel at the same time. If the producer puts water into the 
>>> vessel, then the consumer simply consumes water; If the producer puts 
>>> milk into the vessel, then the consumer simply consumes milk. I think 
>>> we don't need to worry about what type of drinking is put inside the 
>>> vessel, because the vessel is just an intermediate place holding the 
>>>liquid between the producer and consumer - the consumer has the 
>>> privity of contract with the producer and it has the right logic to deal 
> with what's inside the vessel.
> 
>> This analogy breaks for a blind: How would he know whether there's water or 
> milk in there? He'd have to try it. 
>>Now what if we use >venom instead of milk in your analogy?
> 
>>Bottom line - the consumer needs to know what kind of data it is to expect to 
> consume.
> 
> There is a contract between the consumer and the producer. In our case, the 
> contract is "p_cpufreq->scaling_driver". Before the right consumer consumes 
> the 
> value of " uint32_t scaling_max_perf ", it goes through the check:
>  +    if (!strncmp(p_cpufreq->scaling_driver,
>  +                  "intel_pstate", CPUFREQ_NAME_LEN) )
> , where "intel_pstate" can be changed to other new driver names (contract) 
> in the future.

Which is precisely what I already said doesn't scale.

Jan


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel