Re: [PATCH 04/19] cpufreq: amd: introduce a new amd pstate driver to support future processors

[Huang Rui]

On Mon, Sep 13, 2021 at 04:56:24PM +0800, Peter Zijlstra wrote:
> On Mon, Sep 13, 2021 at 04:11:34PM +0800, Huang Rui wrote:
> > On Thu, Sep 09, 2021 at 11:01:41PM +0800, Peter Zijlstra wrote:
> 
> > > What is the purpose of this seemingly pointless indirection? Showing off
> > > how good AMD hardware is at doing retpolines or something?
> > 
> > Hi Petter,
> > 
> > Thanks to look at our codes again. We adopt your suggestion which raised
> > about two year ago that using the kernel governors such as schedutil to
> > manage frequency control for new cpufreq driver.
> 
> Indeed, no objections there :-)
> 
> > We will have two approaches (it depends on different AMD processor
> > hardware) to implement the amd-pstate driver. (Please see details in Patch
> > 19)
> 
> Patch 19 is RST and as such I will not read it. But I think you're
> referring to patch 6, which adds another amd_pstate_perf_funcs instance,
> which I seem to have missed the last time.

Yes, right. No problem. ;-)

> 
> As such, perhaps you could do with something like the below.
> 
> > 1) Full MSR Support
> > If current hardware has the full MSR support, we register "pstate_funcs"
> > callback functions to implement the MSR operations to control the clocks.
> 
> What's the WRMSR cost for those? I've not really kept track of the MSR
> costs on AMD platforms, but on Intel it has (luckily) been coming down
> quite a bit.

Good to know this, I didn't have a chance to give a check. May I know how
did you test this latency? But MSR is new hardware design for this
solution, as designer mentioned, the WRMSR is low-latency register model is
faster than ACPI AML code interpreter.

> 
> > 2) Shared Memory Support
> > If current hardware doesn't have the full MSR support, that means it only
> > provides share memory support. We will leverage APIs in cppc_acpi libs with
> > "cppc_funcs" to implement the target function for the frequency control.
> 
> Right, the mailbox thing. How is the performance of this vs MSR accesses?

I will give a check. If you have a existing test method that can be used, I
can check it quickly.

> 
> > The mainly reasons that we proposed a new amd-pstate driver, not use the
> > existing acpi-freq or cppc-cpufreq driver are below:
> 
> I wasn't really questioning that, much seems similar to having
> intel-pstate, but since you brought it up, a few questions: -)

Thank you!

> 
> > 1. As mentioned above, amd-pstate driver can implement
> > fast_switch/adjust_perf function with full MSR operations that have better
> > performance for schedutil and other governors.
> 
> Why couldn't the existing cppc-cpufreq grow this?

Because fast_switch can adjust the frequency directly in the interrupt
context, if we use the acpi cppc handling with shared memory solution, it
will have a deadlock. So fast switch needs the control with registers
directly like acpi-cpufreq and intel-pstate.

> 
> > 2. We will implement the AMD specific features such as Energy Performance
> > Preference, Preferred Core, and etc. in the amd-pstate driver next step.
> 
> That's the ITMT stuff, right?

Similar with ITMT. :-)

> 
> 
> ---
> 
> --- a/drivers/cpufreq/amd-pstate.c
> +++ b/drivers/cpufreq/amd-pstate.c
> @@ -79,14 +79,6 @@ struct amd_cpudata {
>  	bool	boost_supported;
>  };
>  
> -struct amd_pstate_perf_funcs {
> -	int (*enable)(bool enable);
> -	int (*init_perf)(struct amd_cpudata *cpudata);
> -	void (*update_perf)(struct amd_cpudata *cpudata,
> -			    u32 min_perf, u32 des_perf,
> -			    u32 max_perf, bool fast_switch);
> -};
> -
>  static inline int pstate_enable(bool enable)
>  {
>  	return wrmsrl_safe(MSR_AMD_CPPC_ENABLE, enable ? 1 : 0);
> @@ -105,13 +97,12 @@ static int cppc_enable(bool enable)
>  	return ret;
>  }
>  
> -static int
> -amd_pstate_enable(struct amd_pstate_perf_funcs *funcs, bool enable)
> -{
> -	if (!funcs)
> -		return -EINVAL;
> +static DEFINE_STATIC_CALL(amd_pstate_enable, pstate_enable);
>  
> -	return funcs->enable(enable);
> +static inline int
> +amd_pstate_enable(bool enable)
> +{
> +	return static_call(amd_pstate_enable)(enable);
>  }
>  
>  static int pstate_init_perf(struct amd_cpudata *cpudata)
> @@ -154,14 +145,11 @@ static int cppc_init_perf(struct amd_cpu
>  	return 0;
>  }
>  
> -static int amd_pstate_init_perf(struct amd_cpudata *cpudata)
> -{
> -	struct amd_pstate_perf_funcs *funcs = cpufreq_get_driver_data();
> +static DEFINE_STATIC_CALL(amd_pstate_init_perf, pstate_init_perf);
>  
> -	if (!funcs)
> -		return -EINVAL;
> -
> -	return funcs->init_perf(cpudata);
> +static inline int amd_pstate_init_perf(struct amd_cpudata *cpudata)
> +{
> +	return static_call(amd_pstate_init_perf)(cpudata);
>  }
>  
>  static void pstate_update_perf(struct amd_cpudata *cpudata,
> @@ -188,19 +176,14 @@ static void cppc_update_perf(struct amd_
>  	cppc_set_perf(cpudata->cpu, &perf_ctrls);
>  }
>  
> -static int
> +static DEFINE_STATIC_CALL(amd_pstate_update_perf, pstate_update_perf);
> +
> +static inline int
>  amd_pstate_update_perf(struct amd_cpudata *cpudata, u32 min_perf,
>  		       u32 des_perf, u32 max_perf, bool fast_switch)
>  {
> -	struct amd_pstate_perf_funcs *funcs = cpufreq_get_driver_data();
> -
> -	if (!funcs)
> -		return -EINVAL;
> -
> -	funcs->update_perf(cpudata, min_perf, des_perf,
> -			   max_perf, fast_switch);
> -
> -	return 0;
> +	return static_call(amd_pstate_update_perf)(cpudata, min_perf, des_perf,
> +						   max_perf, fast_switch);
>  }
>  
>  static int
> @@ -465,18 +448,6 @@ static int amd_pstate_init_freqs_in_cpud
>  	return 0;
>  }
>  
> -static struct amd_pstate_perf_funcs pstate_funcs = {
> -	.enable = pstate_enable,
> -	.init_perf = pstate_init_perf,
> -	.update_perf = pstate_update_perf,
> -};
> -
> -static struct amd_pstate_perf_funcs cppc_funcs = {
> -	.enable = cppc_enable,
> -	.init_perf = cppc_init_perf,
> -	.update_perf = cppc_update_perf,
> -};
> -
>  static int amd_pstate_cpu_init(struct cpufreq_policy *policy)
>  {
>  	int min_freq, max_freq, nominal_freq, lowest_nonlinear_freq, ret;
> @@ -749,7 +720,6 @@ static struct cpufreq_driver amd_pstate_
>  static int __init amd_pstate_init(void)
>  {
>  	int ret;
> -	struct amd_pstate_perf_funcs *funcs;
>  
>  	if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD)
>  		return -ENODEV;
> @@ -768,22 +738,21 @@ static int __init amd_pstate_init(void)
>  	if (boot_cpu_has(X86_FEATURE_AMD_CPPC_EXT)) {
>  		pr_debug("%s, AMD CPPC extension functionality is supported\n",
>  			 __func__);
> -		funcs = &pstate_funcs;
>  		amd_pstate_driver.adjust_perf = amd_pstate_adjust_perf;
>  	} else {
> -		funcs = &cppc_funcs;
> +		static_call_update(amd_pstate_enable, cppc_enable);
> +		static_call_update(amd_pstate_init_perf, cppc_init_perf);
> +		static_call_update(amd_pstate_update_perf, cppc_update_perf);

Thanks again for detailed example, I will update to this approach at V2.

Best Regards,
Ray

>  	}
>  
>  	/* enable amd pstate feature */
> -	ret = amd_pstate_enable(funcs, true);
> +	ret = amd_pstate_enable(true);
>  	if (ret) {
>  		pr_err("%s, failed to enable amd-pstate with return %d\n",
>  		       __func__, ret);
>  		return ret;
>  	}
>  
> -	amd_pstate_driver.driver_data = funcs;
> -
>  	ret = cpufreq_register_driver(&amd_pstate_driver);
>  	if (ret) {
>  		pr_err("%s, return %d\n", __func__, ret);
> @@ -795,13 +764,8 @@ static int __init amd_pstate_init(void)
>  
>  static void __exit amd_pstate_exit(void)
>  {
> -	struct amd_pstate_perf_funcs *funcs;
> -
> -	funcs = cpufreq_get_driver_data();
> -
>  	cpufreq_unregister_driver(&amd_pstate_driver);
> -
> -	amd_pstate_enable(funcs, false);
> +	amd_pstate_enable(false);
>  }
>  
>  module_init(amd_pstate_init);