Re: [RFC PATCH 04/31] cpufreq: make turbo settings to be configurable

[Oleksandr Tyshchenko <olekstysh@gmail.com>]

On Mon, Dec 4, 2017 at 1:58 PM, Andre Przywara <andre.przywara@arm.com> wrote:
> Hi,
Hi Andre

>
> ....
>
>> And the most important question is how to recognize in Xen on ARM
>> (using SCPI protocol) which frequencies are turbo-frequencies
>> actually? I couldn't find any information regarding that in protocol
>> description.
>
> So traditionally on ARM there is no notion of a "turbo" frequency. The
> idea is to expose the highest possible frequency, and let thermal
> throttling (possibly in hardware or in firmware) limit the frequency if
> the thermal budget is busted.
> Also in the ARM world it is expected that an OS has much better
> knowledge on how to handle frequencies, for instance when to give more
> power to the GPU and when to the CPU.
>
>> For DT-based CPUFreq it is not an issue, since there is a specific
>> property "turbo-mode" to mark corresponding OPPs. [1].
>> But neither SCPI DT bindings [2] nor the SCPI protocol itself [3]
>> mentions about it. Perhaps, additional command should be added to pass
>> such info.
>
> The DT binding you mentioned in Linux is a generic one.
> In general DT only describes non-discoverable properties. But for SCPI
> the OPPs are handled in the SCP and advertised via SCPI calls (3.2.9 Get
> DVFS Info, command 0x9).
> So the OPP table is not in the DT, and thus you don't have any way of
> detecting turbo frequencies.
> But as mentioned before, this is so by design, as ARM does not endorse
> the concept of turbo frequencies in general.
>
> Now with the advent of more "server-y" chips and ACPI, this might change
> in the future. For instance SCMI is designed to be closer to ACPI, so we
> might inherit some turbo notion from there.
>
> So we should not completely rule out the idea of turbo, but for a start
> we can somewhat assume that an ARM based system does not have turbo per se.

Thank you for the detailed explanation. I will take a look at SCMI
documentation.

>
> Cheers,
> Andre.
>
>> [1] https://www.kernel.org/doc/Documentation/devicetree/bindings/opp/opp.txt
>> [2] http://elixir.free-electrons.com/linux/v4.15-rc1/source/Documentation/devicetree/bindings/arm/arm,scpi.txt
>> [3] http://infocenter.arm.com/help/topic/com.arm.doc.dui0922g/scp_message_interface_v1_2_DUI0922G_en.pdf
>>
>>>
>>>
>>>> diff --git a/xen/drivers/pm/stat.c b/xen/drivers/pm/stat.c
>>>> index 2dbde1c..133e64d 100644
>>>> --- a/xen/drivers/pm/stat.c
>>>> +++ b/xen/drivers/pm/stat.c
>>>> @@ -290,7 +290,11 @@ static int get_cpufreq_para(struct xen_sysctl_pm_op *op)
>>>>              &op->u.get_para.u.ondemand.sampling_rate,
>>>>              &op->u.get_para.u.ondemand.up_threshold);
>>>>      }
>>>> +#ifdef CONFIG_HAS_CPU_TURBO
>>>>      op->u.get_para.turbo_enabled = cpufreq_get_turbo_status(op->cpuid);
>>>> +#else
>>>> +    op->u.get_para.turbo_enabled = 0;
>>>> +#endif
>>>>
>>>>      return ret;
>>>>  }
>>>> @@ -473,6 +477,7 @@ int do_pm_op(struct xen_sysctl_pm_op *op)
>>>>          break;
>>>>      }
>>>>
>>>> +#ifdef CONFIG_HAS_CPU_TURBO
>>>>      case XEN_SYSCTL_pm_op_enable_turbo:
>>>>      {
>>>>          ret = cpufreq_update_turbo(op->cpuid, CPUFREQ_TURBO_ENABLED);
>>>> @@ -484,6 +489,7 @@ int do_pm_op(struct xen_sysctl_pm_op *op)
>>>>          ret = cpufreq_update_turbo(op->cpuid, CPUFREQ_TURBO_DISABLED);
>>>>          break;
>>>>      }
>>>> +#endif /* CONFIG_HAS_CPU_TURBO */
>>>>
>>>>      default:
>>>>          printk("not defined sub-hypercall @ do_pm_op\n");
>>>> diff --git a/xen/include/xen/cpufreq.h b/xen/include/xen/cpufreq.h
>>>> index 30c70c9..2e0c16a 100644
>>>> --- a/xen/include/xen/cpufreq.h
>>>> +++ b/xen/include/xen/cpufreq.h
>>>> @@ -39,7 +39,9 @@ extern struct acpi_cpufreq_data *cpufreq_drv_data[NR_CPUS];
>>>>
>>>>  struct cpufreq_cpuinfo {
>>>>      unsigned int        max_freq;
>>>> +#ifdef CONFIG_HAS_CPU_TURBO
>>>>      unsigned int        second_max_freq;    /* P1 if Turbo Mode is on */
>>>> +#endif
>>>>      unsigned int        min_freq;
>>>>      unsigned int        transition_latency; /* in 10^(-9) s = nanoseconds */
>>>>  };
>>>> @@ -72,9 +74,11 @@ struct cpufreq_policy {
>>>>
>>>>      bool_t              resume; /* flag for cpufreq 1st run
>>>>                                   * S3 wakeup, hotplug cpu, etc */
>>>> +#ifdef CONFIG_HAS_CPU_TURBO
>>>>      s8                  turbo;  /* tristate flag: 0 for unsupported
>>>>                                   * -1 for disable, 1 for enabled
>>>>                                   * See CPUFREQ_TURBO_* below for defines */
>>>> +#endif
>>>>      bool_t              aperf_mperf; /* CPU has APERF/MPERF MSRs */
>>>>  };
>>>>  DECLARE_PER_CPU(struct cpufreq_policy *, cpufreq_cpu_policy);
>>>> @@ -138,8 +142,10 @@ extern int cpufreq_driver_getavg(unsigned int cpu, unsigned int flag);
>>>>  #define CPUFREQ_TURBO_UNSUPPORTED   0
>>>>  #define CPUFREQ_TURBO_ENABLED       1
>>>>
>>>> +#ifdef CONFIG_HAS_CPU_TURBO
>>>>  extern int cpufreq_update_turbo(int cpuid, int new_state);
>>>>  extern int cpufreq_get_turbo_status(int cpuid);
>>>> +#endif
>>>>
>>>>  static __inline__ int
>>>>  __cpufreq_governor(struct cpufreq_policy *policy, unsigned int event)
>>
>>
>>



-- 
Regards,

Oleksandr Tyshchenko

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xenproject.org
https://lists.xenproject.org/mailman/listinfo/xen-devel