[RFC,V2,1/2] topology: Allow multiple entities to provide sched_freq_tick() callback

Message ID 28ade070dd80f6f22d6e8fce5db5f0142b428fa9.1608030508.git.viresh.kumar@linaro.org
State New
Headers show
Series
  • cpufreq: cppc: Add support for frequency invariance
Related show

Commit Message

Viresh Kumar Dec. 15, 2020, 11:16 a.m.
This patch attempts to make it generic enough so other parts of the
kernel can also provide their own implementation of scale_freq_tick()
callback, which is called by the scheduler periodically to update the
per-cpu freq_scale variable.

The implementations now need to provide struct scale_freq_tick_data for
the CPUs for which they have hardware counters available, and a callback
gets registered for each possible CPU in a per-cpu variable.

The AMU counters are updated to adapt to this and they take the highest
priority if they are available, i.e. they will be used instead of CPPC
based counters for example.

Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>

---
 arch/arm64/include/asm/topology.h |  8 +--
 arch/arm64/kernel/topology.c      | 89 ++++++++++++++-----------------
 drivers/base/arch_topology.c      | 56 +++++++++++++++++--
 include/linux/arch_topology.h     | 13 ++++-
 4 files changed, 104 insertions(+), 62 deletions(-)

-- 
2.25.0.rc1.19.g042ed3e048af

Comments

Ionela Voinescu Jan. 13, 2021, 4:18 p.m. | #1
Hi,

I focused for now on the design of the solution and mostly on the
generic code in arch_topology.c, as that will then impact the arm64
code. 

On Tuesday 15 Dec 2020 at 16:46:35 (+0530), Viresh Kumar wrote:
[..]
>  

> diff --git a/drivers/base/arch_topology.c b/drivers/base/arch_topology.c

> index de8587cc119e..07774fb47bc4 100644

> --- a/drivers/base/arch_topology.c

> +++ b/drivers/base/arch_topology.c

> @@ -21,17 +21,65 @@

>  #include <linux/sched.h>

>  #include <linux/smp.h>

>  

> +static DEFINE_PER_CPU(struct scale_freq_tick_data *, sft_data);

> +static struct cpumask scale_freq_tick_mask;

> +

> +static bool supports_scale_freq_tick(const struct cpumask *cpus)

> +{

> +	return cpumask_subset(cpus, &scale_freq_tick_mask);

> +}

> +

>  bool topology_scale_freq_invariant(void)

>  {

>  	return cpufreq_supports_freq_invariance() ||

> -	       arch_freq_counters_available(cpu_online_mask);

> +	       supports_scale_freq_tick(cpu_online_mask);

               ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
I think the names that involve "scale_freq_tick" are unfortunate and
unclear here. arch_scale_freq_tick() is a good wrapper as it refers to
some work being done for FIE on the tick, but for other functions and
structures here, any similar naming makes things unclear.

Basically all FIE related activity on the tick involves counters that
need regular sampling. So I'd suggest to keep to names that give a hint
about the source of information for frequency invariance, as it makes it
more clear what are the options you're choosing between for FIE.

s/supports_scale_freq_tick/supports_scale_freq_counters
s/scale_freq_tick_data/scale_freq_data
s/topology_set_scale_freq_tick/topology_set_scale_freq_source
s/topology_clear_scale_freq_tick/topology_clear_scale_freq_source
s/scale_freq_tick_mask/scale_freq_counters_mask

.. or similar.

> +}

> +

> +void topology_set_scale_freq_tick(struct scale_freq_tick_data *data,

> +				  const struct cpumask *cpus)

> +{

> +	struct scale_freq_tick_data *sftd;

> +	int cpu;

> +

> +	for_each_cpu(cpu, cpus) {

> +		sftd = per_cpu(sft_data, cpu);

> +

> +		/* Use AMU counters whenever possible */

> +		if (!sftd || sftd->source != SCALE_FREQ_SOURCE_AMU) {


I don't think the arch_topology driver should know anything about AMUs, as
they are an arm64 feature. SCALE_FREQ_SOURCE_ARCH might be better.
This way any architecture that uses the arch topology driver can register
some architected method which should take priority over all other
methods (at least until someone shows up wanting otherwise).

> +			per_cpu(sft_data, cpu) = data;

> +			cpumask_set_cpu(cpu, &scale_freq_tick_mask);

> +		}

> +	}

>  }

> +EXPORT_SYMBOL_GPL(topology_set_scale_freq_tick);

>  

> -__weak bool arch_freq_counters_available(const struct cpumask *cpus)

> +void topology_clear_scale_freq_tick(enum scale_freq_tick_source source,

> +				    const struct cpumask *cpus)

>  {

> -	return false;

> +	struct scale_freq_tick_data *sftd;

> +	int cpu;

> +

> +	for_each_cpu(cpu, cpus) {

> +		sftd = per_cpu(sft_data, cpu);

> +

> +		if (sftd && sftd->source == source) {

> +			per_cpu(sft_data, cpu) = NULL;

> +			cpumask_clear_cpu(cpu, &scale_freq_tick_mask);

> +		}

> +	}

>  }

> +EXPORT_SYMBOL_GPL(topology_clear_scale_freq_tick);

> +

> +void topology_scale_freq_tick(void)

> +{

> +	struct scale_freq_tick_data *sftd = *this_cpu_ptr(&sft_data);

> +

> +	if (sftd)

> +		sftd->scale_freq();

> +}


What do you think about having a single topology function that handles
all sources of invariance (cpufreq, arch counters, platform counters)?

Snippet of possible code:

"""
diff --git a/arch/arm64/include/asm/topology.h b/arch/arm64/include/asm/topology.h
index be6a53ba3e2d..d9b7221b8ea9 100644
--- a/arch/arm64/include/asm/topology.h
+++ b/arch/arm64/include/asm/topology.h
@@ -19,7 +19,7 @@ int pcibus_to_node(struct pci_bus *bus);
 void update_freq_counters_refs(void);
 
 /* Replace task scheduler's default frequency-invariance scale factor setting */
-#define arch_scale_freq_tick topology_scale_freq_tick
+#define arch_scale_freq_tick() topology_set_freq_scale(NULL, 0, 0)
 
 /* Replace task scheduler's default frequency-invariant accounting */
 #define arch_set_freq_scale topology_set_freq_scale
diff --git a/drivers/base/arch_topology.c b/drivers/base/arch_topology.c
index 07774fb47bc4..a669963e7b01 100644
--- a/drivers/base/arch_topology.c
+++ b/drivers/base/arch_topology.c
@@ -70,38 +70,27 @@ void topology_clear_scale_freq_tick(enum scale_freq_tick_source source,
 }
 EXPORT_SYMBOL_GPL(topology_clear_scale_freq_tick);
 
-void topology_scale_freq_tick(void)
-{
-	struct scale_freq_tick_data *sftd = *this_cpu_ptr(&sft_data);
-
-	if (sftd)
-		sftd->scale_freq();
-}
-
 DEFINE_PER_CPU(unsigned long, freq_scale) = SCHED_CAPACITY_SCALE;
 EXPORT_SYMBOL_GPL(freq_scale);
 
 void topology_set_freq_scale(const struct cpumask *cpus, unsigned long cur_freq,
 			     unsigned long max_freq)
 {
+	struct scale_freq_tick_data *sftd;
 	unsigned long scale;
 	int i;
 
-	if (WARN_ON_ONCE(!cur_freq || !max_freq))
-		return;
-
-	/*
-	 * If the use of counters for FIE is enabled, just return as we don't
-	 * want to update the scale factor with information from CPUFREQ.
-	 * Instead the scale factor will be updated from arch_scale_freq_tick.
-	 */
-	if (supports_scale_freq_tick(cpus))
-		return;
-
-	scale = (cur_freq << SCHED_CAPACITY_SHIFT) / max_freq;
-
-	for_each_cpu(i, cpus)
-		per_cpu(freq_scale, i) = scale;
+	/* Retrieve provided method for scale factor setting for current CPU */
+	if (!cpus) {
+		sftd = *this_cpu_ptr(&sft_data);
+		if (sftd)
+			sftd->scale_freq();
+	} else if (cur_freq && max_freq) {
+		/* Use arguments to compute and set scale factor for CPUs */
+		scale = (cur_freq << SCHED_CAPACITY_SHIFT) / max_freq;
+		for_each_cpu(i, cpus)
+			per_cpu(freq_scale, i) = scale;
+	}
 }
 
 DEFINE_PER_CPU(unsigned long, cpu_scale) = SCHED_CAPACITY_SCALE;
diff --git a/include/linux/arch_topology.h b/include/linux/arch_topology.h
index 09205b584ca5..9b41a90fc955 100644
--- a/include/linux/arch_topology.h
+++ b/include/linux/arch_topology.h
@@ -44,7 +44,6 @@ struct scale_freq_tick_data {
 	void (*scale_freq)(void);
 };
 
-void topology_scale_freq_tick(void);
 void topology_set_scale_freq_tick(struct scale_freq_tick_data *data, const struct cpumask *cpus);
 void topology_clear_scale_freq_tick(enum scale_freq_tick_source source, const struct cpumask *cpus);
 
"""

This basically says that if there is a method registered for a certain
CPU to set the scale factor, that method should be used. Otherwise,
rely on this default math of obtaining the scale factor based on the
current and maximum frequency arguments, if they are provided.

This could get us rid of supports_scale_freq_tick() and possibly
even the scale_freq_tick_mask. topology_set_scale_freq_tick() and
topology_clear_scale_freq_tick() could also be used to enable and
disable system level invariance (through a static key) so we don't have
to do cpumask checks in topology_scale_freq_invariant().

We might then add a static key to prevent those useless conditions being
evaluated on the tick. But we'll see possible optimisations later.

> +

>  DEFINE_PER_CPU(unsigned long, freq_scale) = SCHED_CAPACITY_SCALE;

> +EXPORT_SYMBOL_GPL(freq_scale);

>  

>  void topology_set_freq_scale(const struct cpumask *cpus, unsigned long cur_freq,

>  			     unsigned long max_freq)

> @@ -47,7 +95,7 @@ void topology_set_freq_scale(const struct cpumask *cpus, unsigned long cur_freq,

>  	 * want to update the scale factor with information from CPUFREQ.

>  	 * Instead the scale factor will be updated from arch_scale_freq_tick.

>  	 */

> -	if (arch_freq_counters_available(cpus))

> +	if (supports_scale_freq_tick(cpus))

>  		return;

>  

>  	scale = (cur_freq << SCHED_CAPACITY_SHIFT) / max_freq;

> diff --git a/include/linux/arch_topology.h b/include/linux/arch_topology.h

> index 0f6cd6b73a61..b2422ebef2dd 100644

> --- a/include/linux/arch_topology.h

> +++ b/include/linux/arch_topology.h

> @@ -34,7 +34,18 @@ void topology_set_freq_scale(const struct cpumask *cpus, unsigned long cur_freq,

>  			     unsigned long max_freq);

>  bool topology_scale_freq_invariant(void);

>  

> -bool arch_freq_counters_available(const struct cpumask *cpus);

> +enum scale_freq_tick_source {

> +	SCALE_FREQ_SOURCE_AMU,

> +};

> +

> +struct scale_freq_tick_data {

> +	enum scale_freq_tick_source source;

> +	void (*scale_freq)(void);

> +};


s/scale_freq/set_freq_scale?

Hope it helps,
Ionela.

> +

> +void topology_scale_freq_tick(void);

> +void topology_set_scale_freq_tick(struct scale_freq_tick_data *data, const struct cpumask *cpus);

> +void topology_clear_scale_freq_tick(enum scale_freq_tick_source source, const struct cpumask *cpus);

>  

>  DECLARE_PER_CPU(unsigned long, thermal_pressure);

>  

> -- 

> 2.25.0.rc1.19.g042ed3e048af

>
Viresh Kumar Jan. 15, 2021, 7:48 a.m. | #2
On 13-01-21, 16:18, Ionela Voinescu wrote:
> On Tuesday 15 Dec 2020 at 16:46:35 (+0530), Viresh Kumar wrote:

> > +void topology_scale_freq_tick(void)

> > +{

> > +	struct scale_freq_tick_data *sftd = *this_cpu_ptr(&sft_data);

> > +

> > +	if (sftd)

> > +		sftd->scale_freq();

> > +}

> 

> What do you think about having a single topology function that handles

> all sources of invariance (cpufreq, arch counters, platform counters)?


I think keeping them separate is better, both of these are called from
scheduler's context (hotpath) and adding any more unnecessary
conditionals there should be rather avoided. We could have given a
though to merging them if they were going to share some code, but that
is not the case here clearly. They are quite different.

This is how it looks now:

diff --git a/arch/arm64/include/asm/topology.h b/arch/arm64/include/asm/topology.h
index 3b8dca4eb08d..be6a53ba3e2d 100644
--- a/arch/arm64/include/asm/topology.h
+++ b/arch/arm64/include/asm/topology.h
@@ -17,15 +17,9 @@ int pcibus_to_node(struct pci_bus *bus);
 #include <linux/arch_topology.h>
 
 void update_freq_counters_refs(void);
-void topology_scale_freq_tick(void);
 
-#ifdef CONFIG_ARM64_AMU_EXTN
-/*
- * Replace task scheduler's default counter-based
- * frequency-invariance scale factor setting.
- */
+/* Replace task scheduler's default frequency-invariance scale factor setting */
 #define arch_scale_freq_tick topology_scale_freq_tick
-#endif /* CONFIG_ARM64_AMU_EXTN */
 
 /* Replace task scheduler's default frequency-invariant accounting */
 #define arch_set_freq_scale topology_set_freq_scale
diff --git a/arch/arm64/kernel/topology.c b/arch/arm64/kernel/topology.c
index e08a4126453a..1e47dfd465f8 100644
--- a/arch/arm64/kernel/topology.c
+++ b/arch/arm64/kernel/topology.c
@@ -199,8 +199,44 @@ static int freq_inv_set_max_ratio(int cpu, u64 max_rate, u64 ref_rate)
 	return 0;
 }
 
-static DEFINE_STATIC_KEY_FALSE(amu_fie_key);
-#define amu_freq_invariant() static_branch_unlikely(&amu_fie_key)
+static void amu_scale_freq_tick(void)
+{
+	u64 prev_core_cnt, prev_const_cnt;
+	u64 core_cnt, const_cnt, scale;
+
+	prev_const_cnt = this_cpu_read(arch_const_cycles_prev);
+	prev_core_cnt = this_cpu_read(arch_core_cycles_prev);
+
+	update_freq_counters_refs();
+
+	const_cnt = this_cpu_read(arch_const_cycles_prev);
+	core_cnt = this_cpu_read(arch_core_cycles_prev);
+
+	if (unlikely(core_cnt <= prev_core_cnt ||
+		     const_cnt <= prev_const_cnt))
+		return;
+
+	/*
+	 *	    /\core    arch_max_freq_scale
+	 * scale =  ------- * --------------------
+	 *	    /\const   SCHED_CAPACITY_SCALE
+	 *
+	 * See validate_cpu_freq_invariance_counters() for details on
+	 * arch_max_freq_scale and the use of SCHED_CAPACITY_SHIFT.
+	 */
+	scale = core_cnt - prev_core_cnt;
+	scale *= this_cpu_read(arch_max_freq_scale);
+	scale = div64_u64(scale >> SCHED_CAPACITY_SHIFT,
+			  const_cnt - prev_const_cnt);
+
+	scale = min_t(unsigned long, scale, SCHED_CAPACITY_SCALE);
+	this_cpu_write(freq_scale, (unsigned long)scale);
+}
+
+static struct scale_freq_data amu_sfd = {
+	.source = SCALE_FREQ_SOURCE_ARCH,
+	.set_freq_scale = amu_scale_freq_tick,
+};
 
 static void amu_fie_setup(const struct cpumask *cpus)
 {
@@ -227,7 +263,7 @@ static void amu_fie_setup(const struct cpumask *cpus)
 	if (!invariant && !cpumask_equal(amu_fie_cpus, cpu_present_mask))
 		return;
 
-	static_branch_enable(&amu_fie_key);
+	topology_set_scale_freq_source(&amu_sfd, amu_fie_cpus);
 
 	pr_debug("CPUs[%*pbl]: counters will be used for FIE.",
 		 cpumask_pr_args(cpus));
@@ -283,53 +319,6 @@ static int __init init_amu_fie(void)
 }
 core_initcall(init_amu_fie);
 
-bool arch_freq_counters_available(const struct cpumask *cpus)
-{
-	return amu_freq_invariant() &&
-	       cpumask_subset(cpus, amu_fie_cpus);
-}
-
-void topology_scale_freq_tick(void)
-{
-	u64 prev_core_cnt, prev_const_cnt;
-	u64 core_cnt, const_cnt, scale;
-	int cpu = smp_processor_id();
-
-	if (!amu_freq_invariant())
-		return;
-
-	if (!cpumask_test_cpu(cpu, amu_fie_cpus))
-		return;
-
-	prev_const_cnt = this_cpu_read(arch_const_cycles_prev);
-	prev_core_cnt = this_cpu_read(arch_core_cycles_prev);
-
-	update_freq_counters_refs();
-
-	const_cnt = this_cpu_read(arch_const_cycles_prev);
-	core_cnt = this_cpu_read(arch_core_cycles_prev);
-
-	if (unlikely(core_cnt <= prev_core_cnt ||
-		     const_cnt <= prev_const_cnt))
-		return;
-
-	/*
-	 *	    /\core    arch_max_freq_scale
-	 * scale =  ------- * --------------------
-	 *	    /\const   SCHED_CAPACITY_SCALE
-	 *
-	 * See validate_cpu_freq_invariance_counters() for details on
-	 * arch_max_freq_scale and the use of SCHED_CAPACITY_SHIFT.
-	 */
-	scale = core_cnt - prev_core_cnt;
-	scale *= this_cpu_read(arch_max_freq_scale);
-	scale = div64_u64(scale >> SCHED_CAPACITY_SHIFT,
-			  const_cnt - prev_const_cnt);
-
-	scale = min_t(unsigned long, scale, SCHED_CAPACITY_SCALE);
-	this_cpu_write(freq_scale, (unsigned long)scale);
-}
-
 #ifdef CONFIG_ACPI_CPPC_LIB
 #include <acpi/cppc_acpi.h>
 
diff --git a/drivers/base/arch_topology.c b/drivers/base/arch_topology.c
index de8587cc119e..e2115ea348dc 100644
--- a/drivers/base/arch_topology.c
+++ b/drivers/base/arch_topology.c
@@ -21,17 +21,65 @@
 #include <linux/sched.h>
 #include <linux/smp.h>
 
+static DEFINE_PER_CPU(struct scale_freq_data *, sft_data);
+static struct cpumask scale_freq_counters_mask;
+
+static bool supports_scale_freq_counters(const struct cpumask *cpus)
+{
+	return cpumask_subset(cpus, &scale_freq_counters_mask);
+}
+
 bool topology_scale_freq_invariant(void)
 {
 	return cpufreq_supports_freq_invariance() ||
-	       arch_freq_counters_available(cpu_online_mask);
+	       supports_scale_freq_counters(cpu_online_mask);
+}
+
+void topology_set_scale_freq_source(struct scale_freq_data *data,
+				  const struct cpumask *cpus)
+{
+	struct scale_freq_data *sfd;
+	int cpu;
+
+	for_each_cpu(cpu, cpus) {
+		sfd = per_cpu(sft_data, cpu);
+
+		/* Use ARCH provided counters whenever possible */
+		if (!sfd || sfd->source != SCALE_FREQ_SOURCE_ARCH) {
+			per_cpu(sft_data, cpu) = data;
+			cpumask_set_cpu(cpu, &scale_freq_counters_mask);
+		}
+	}
 }
+EXPORT_SYMBOL_GPL(topology_set_scale_freq_source);
 
-__weak bool arch_freq_counters_available(const struct cpumask *cpus)
+void topology_clear_scale_freq_source(enum scale_freq_source source,
+				    const struct cpumask *cpus)
 {
-	return false;
+	struct scale_freq_data *sfd;
+	int cpu;
+
+	for_each_cpu(cpu, cpus) {
+		sfd = per_cpu(sft_data, cpu);
+
+		if (sfd && sfd->source == source) {
+			per_cpu(sft_data, cpu) = NULL;
+			cpumask_clear_cpu(cpu, &scale_freq_counters_mask);
+		}
+	}
 }
+EXPORT_SYMBOL_GPL(topology_clear_scale_freq_source);
+
+void topology_scale_freq_tick(void)
+{
+	struct scale_freq_data *sfd = *this_cpu_ptr(&sft_data);
+
+	if (sfd)
+		sfd->set_freq_scale();
+}
+
 DEFINE_PER_CPU(unsigned long, freq_scale) = SCHED_CAPACITY_SCALE;
+EXPORT_SYMBOL_GPL(freq_scale);
 
 void topology_set_freq_scale(const struct cpumask *cpus, unsigned long cur_freq,
 			     unsigned long max_freq)
@@ -47,7 +95,7 @@ void topology_set_freq_scale(const struct cpumask *cpus, unsigned long cur_freq,
 	 * want to update the scale factor with information from CPUFREQ.
 	 * Instead the scale factor will be updated from arch_scale_freq_tick.
 	 */
-	if (arch_freq_counters_available(cpus))
+	if (supports_scale_freq_counters(cpus))
 		return;
 
 	scale = (cur_freq << SCHED_CAPACITY_SHIFT) / max_freq;
diff --git a/include/linux/arch_topology.h b/include/linux/arch_topology.h
index 0f6cd6b73a61..1bcbf8eff991 100644
--- a/include/linux/arch_topology.h
+++ b/include/linux/arch_topology.h
@@ -34,7 +34,18 @@ void topology_set_freq_scale(const struct cpumask *cpus, unsigned long cur_freq,
 			     unsigned long max_freq);
 bool topology_scale_freq_invariant(void);
 
-bool arch_freq_counters_available(const struct cpumask *cpus);
+enum scale_freq_source {
+	SCALE_FREQ_SOURCE_ARCH,
+};
+
+struct scale_freq_data {
+	enum scale_freq_source source;
+	void (*set_freq_scale)(void);
+};
+
+void topology_scale_freq_tick(void);
+void topology_set_scale_freq_source(struct scale_freq_data *data, const struct cpumask *cpus);
+void topology_clear_scale_freq_source(enum scale_freq_source source, const struct cpumask *cpus);
 
 DECLARE_PER_CPU(unsigned long, thermal_pressure);
 

-- 
viresh
Ionela Voinescu Jan. 19, 2021, 7:05 p.m. | #3
Hi,

On Friday 15 Jan 2021 at 13:18:47 (+0530), Viresh Kumar wrote:
> On 13-01-21, 16:18, Ionela Voinescu wrote:

> > On Tuesday 15 Dec 2020 at 16:46:35 (+0530), Viresh Kumar wrote:

> > > +void topology_scale_freq_tick(void)

> > > +{

> > > +	struct scale_freq_tick_data *sftd = *this_cpu_ptr(&sft_data);

> > > +

> > > +	if (sftd)

> > > +		sftd->scale_freq();

> > > +}

> > 

> > What do you think about having a single topology function that handles

> > all sources of invariance (cpufreq, arch counters, platform counters)?

> 

> I think keeping them separate is better, both of these are called from

> scheduler's context (hotpath) and adding any more unnecessary

> conditionals there should be rather avoided. We could have given a


It would be a single added condition on the hotpath for !cpus and I
think CPUs nowadays do a very good job optimising those.
Also, you lose a branch and a cpumask operation on the
arch_set_freq_scale path.

> though to merging them if they were going to share some code, but that

> is not the case here clearly. They are quite different.

> 


I think it would improve clarity a lot by having a single topology
function, with clear comments on which path is taken when.

In regards to them sharing code, there's not much code they could be
sharing, as one just does some simple math on provided values.
I attempted to have a single function for frequency invariance at some
point [1], as the logic for the computation is the same, but I could
never convince myself the clarity gained was worth the changes. But
I really like this version of your code for this purpose as well.

But I'll leave that to your judgement.

[1] https://gitlab.arm.com/linux-arm/linux-power/-/commit/b9277295475307051c34ca31aef28eac50115e1a

> This is how it looks now:

> 

> diff --git a/arch/arm64/include/asm/topology.h b/arch/arm64/include/asm/topology.h

> index 3b8dca4eb08d..be6a53ba3e2d 100644

> --- a/arch/arm64/include/asm/topology.h

> +++ b/arch/arm64/include/asm/topology.h

> @@ -17,15 +17,9 @@ int pcibus_to_node(struct pci_bus *bus);

>  #include <linux/arch_topology.h>

>  

>  void update_freq_counters_refs(void);

> -void topology_scale_freq_tick(void);

>  

> -#ifdef CONFIG_ARM64_AMU_EXTN

> -/*

> - * Replace task scheduler's default counter-based

> - * frequency-invariance scale factor setting.

> - */

> +/* Replace task scheduler's default frequency-invariance scale factor setting */

>  #define arch_scale_freq_tick topology_scale_freq_tick


You could probably move this define under the comment below, as both
arch functions do frequency-invariance accounting, but from different
sources.

> -#endif /* CONFIG_ARM64_AMU_EXTN */

>  

>  /* Replace task scheduler's default frequency-invariant accounting */

>  #define arch_set_freq_scale topology_set_freq_scale

> diff --git a/arch/arm64/kernel/topology.c b/arch/arm64/kernel/topology.c

> index e08a4126453a..1e47dfd465f8 100644

> --- a/arch/arm64/kernel/topology.c

> +++ b/arch/arm64/kernel/topology.c

> @@ -199,8 +199,44 @@ static int freq_inv_set_max_ratio(int cpu, u64 max_rate, u64 ref_rate)

>  	return 0;

>  }

>  

> -static DEFINE_STATIC_KEY_FALSE(amu_fie_key);

> -#define amu_freq_invariant() static_branch_unlikely(&amu_fie_key)

> +static void amu_scale_freq_tick(void)

> +{

> +	u64 prev_core_cnt, prev_const_cnt;

> +	u64 core_cnt, const_cnt, scale;

> +

> +	prev_const_cnt = this_cpu_read(arch_const_cycles_prev);

> +	prev_core_cnt = this_cpu_read(arch_core_cycles_prev);

> +

> +	update_freq_counters_refs();

> +

> +	const_cnt = this_cpu_read(arch_const_cycles_prev);

> +	core_cnt = this_cpu_read(arch_core_cycles_prev);

> +

> +	if (unlikely(core_cnt <= prev_core_cnt ||

> +		     const_cnt <= prev_const_cnt))

> +		return;

> +

> +	/*

> +	 *	    /\core    arch_max_freq_scale

> +	 * scale =  ------- * --------------------

> +	 *	    /\const   SCHED_CAPACITY_SCALE

> +	 *

> +	 * See validate_cpu_freq_invariance_counters() for details on

> +	 * arch_max_freq_scale and the use of SCHED_CAPACITY_SHIFT.

> +	 */

> +	scale = core_cnt - prev_core_cnt;

> +	scale *= this_cpu_read(arch_max_freq_scale);

> +	scale = div64_u64(scale >> SCHED_CAPACITY_SHIFT,

> +			  const_cnt - prev_const_cnt);

> +

> +	scale = min_t(unsigned long, scale, SCHED_CAPACITY_SCALE);

> +	this_cpu_write(freq_scale, (unsigned long)scale);

> +}

> +

> +static struct scale_freq_data amu_sfd = {

> +	.source = SCALE_FREQ_SOURCE_ARCH,

> +	.set_freq_scale = amu_scale_freq_tick,

> +};

>  

>  static void amu_fie_setup(const struct cpumask *cpus)

>  {

> @@ -227,7 +263,7 @@ static void amu_fie_setup(const struct cpumask *cpus)

>  	if (!invariant && !cpumask_equal(amu_fie_cpus, cpu_present_mask))

>  		return;

>  

> -	static_branch_enable(&amu_fie_key);

> +	topology_set_scale_freq_source(&amu_sfd, amu_fie_cpus);

>  

>  	pr_debug("CPUs[%*pbl]: counters will be used for FIE.",

>  		 cpumask_pr_args(cpus));

> @@ -283,53 +319,6 @@ static int __init init_amu_fie(void)

>  }

>  core_initcall(init_amu_fie);

>  

> -bool arch_freq_counters_available(const struct cpumask *cpus)

> -{

> -	return amu_freq_invariant() &&

> -	       cpumask_subset(cpus, amu_fie_cpus);

> -}

> -

> -void topology_scale_freq_tick(void)

> -{

> -	u64 prev_core_cnt, prev_const_cnt;

> -	u64 core_cnt, const_cnt, scale;

> -	int cpu = smp_processor_id();

> -

> -	if (!amu_freq_invariant())

> -		return;

> -

> -	if (!cpumask_test_cpu(cpu, amu_fie_cpus))

> -		return;

> -

> -	prev_const_cnt = this_cpu_read(arch_const_cycles_prev);

> -	prev_core_cnt = this_cpu_read(arch_core_cycles_prev);

> -

> -	update_freq_counters_refs();

> -

> -	const_cnt = this_cpu_read(arch_const_cycles_prev);

> -	core_cnt = this_cpu_read(arch_core_cycles_prev);

> -

> -	if (unlikely(core_cnt <= prev_core_cnt ||

> -		     const_cnt <= prev_const_cnt))

> -		return;

> -

> -	/*

> -	 *	    /\core    arch_max_freq_scale

> -	 * scale =  ------- * --------------------

> -	 *	    /\const   SCHED_CAPACITY_SCALE

> -	 *

> -	 * See validate_cpu_freq_invariance_counters() for details on

> -	 * arch_max_freq_scale and the use of SCHED_CAPACITY_SHIFT.

> -	 */

> -	scale = core_cnt - prev_core_cnt;

> -	scale *= this_cpu_read(arch_max_freq_scale);

> -	scale = div64_u64(scale >> SCHED_CAPACITY_SHIFT,

> -			  const_cnt - prev_const_cnt);

> -

> -	scale = min_t(unsigned long, scale, SCHED_CAPACITY_SCALE);

> -	this_cpu_write(freq_scale, (unsigned long)scale);

> -}

> -

>  #ifdef CONFIG_ACPI_CPPC_LIB

>  #include <acpi/cppc_acpi.h>

>  

> diff --git a/drivers/base/arch_topology.c b/drivers/base/arch_topology.c

> index de8587cc119e..e2115ea348dc 100644

> --- a/drivers/base/arch_topology.c

> +++ b/drivers/base/arch_topology.c

> @@ -21,17 +21,65 @@

>  #include <linux/sched.h>

>  #include <linux/smp.h>

>  

> +static DEFINE_PER_CPU(struct scale_freq_data *, sft_data);

> +static struct cpumask scale_freq_counters_mask;

> +

> +static bool supports_scale_freq_counters(const struct cpumask *cpus)

> +{

> +	return cpumask_subset(cpus, &scale_freq_counters_mask);

> +}

> +

>  bool topology_scale_freq_invariant(void)

>  {

>  	return cpufreq_supports_freq_invariance() ||

> -	       arch_freq_counters_available(cpu_online_mask);

> +	       supports_scale_freq_counters(cpu_online_mask);

> +}

> +

> +void topology_set_scale_freq_source(struct scale_freq_data *data,

> +				  const struct cpumask *cpus)

> +{

> +	struct scale_freq_data *sfd;

> +	int cpu;

> +

> +	for_each_cpu(cpu, cpus) {

> +		sfd = per_cpu(sft_data, cpu);

> +

> +		/* Use ARCH provided counters whenever possible */

> +		if (!sfd || sfd->source != SCALE_FREQ_SOURCE_ARCH) {

> +			per_cpu(sft_data, cpu) = data;

> +			cpumask_set_cpu(cpu, &scale_freq_counters_mask);

> +		}

> +	}

>  }

> +EXPORT_SYMBOL_GPL(topology_set_scale_freq_source);

>  

> -__weak bool arch_freq_counters_available(const struct cpumask *cpus)

> +void topology_clear_scale_freq_source(enum scale_freq_source source,

> +				    const struct cpumask *cpus)

>  {

> -	return false;

> +	struct scale_freq_data *sfd;

> +	int cpu;

> +

> +	for_each_cpu(cpu, cpus) {

> +		sfd = per_cpu(sft_data, cpu);

> +

> +		if (sfd && sfd->source == source) {

> +			per_cpu(sft_data, cpu) = NULL;

> +			cpumask_clear_cpu(cpu, &scale_freq_counters_mask);

> +		}

> +	}

>  }

> +EXPORT_SYMBOL_GPL(topology_clear_scale_freq_source);

> +

> +void topology_scale_freq_tick(void)

> +{

> +	struct scale_freq_data *sfd = *this_cpu_ptr(&sft_data);

> +

> +	if (sfd)

> +		sfd->set_freq_scale();

> +}

> +

>  DEFINE_PER_CPU(unsigned long, freq_scale) = SCHED_CAPACITY_SCALE;

> +EXPORT_SYMBOL_GPL(freq_scale);

>  

>  void topology_set_freq_scale(const struct cpumask *cpus, unsigned long cur_freq,

>  			     unsigned long max_freq)

> @@ -47,7 +95,7 @@ void topology_set_freq_scale(const struct cpumask *cpus, unsigned long cur_freq,

>  	 * want to update the scale factor with information from CPUFREQ.

>  	 * Instead the scale factor will be updated from arch_scale_freq_tick.

>  	 */

> -	if (arch_freq_counters_available(cpus))

> +	if (supports_scale_freq_counters(cpus))

>  		return;

>  

>  	scale = (cur_freq << SCHED_CAPACITY_SHIFT) / max_freq;

> diff --git a/include/linux/arch_topology.h b/include/linux/arch_topology.h

> index 0f6cd6b73a61..1bcbf8eff991 100644

> --- a/include/linux/arch_topology.h

> +++ b/include/linux/arch_topology.h

> @@ -34,7 +34,18 @@ void topology_set_freq_scale(const struct cpumask *cpus, unsigned long cur_freq,

>  			     unsigned long max_freq);

>  bool topology_scale_freq_invariant(void);

>  

> -bool arch_freq_counters_available(const struct cpumask *cpus);

> +enum scale_freq_source {


Maybe you can add SCALE_FREQ_SOURCE_CPUFREQ as the first value (=0).
That will give a hint of CPUFREQ being a source of information for FIE
and the default when counters are not supported, even if nothing gets
done with that value (for now at least).

> +	SCALE_FREQ_SOURCE_ARCH,

> +};

> +

> +struct scale_freq_data {

> +	enum scale_freq_source source;

> +	void (*set_freq_scale)(void);

> +};

> +

> +void topology_scale_freq_tick(void);

> +void topology_set_scale_freq_source(struct scale_freq_data *data, const struct cpumask *cpus);

> +void topology_clear_scale_freq_source(enum scale_freq_source source, const struct cpumask *cpus);

>  

>  DECLARE_PER_CPU(unsigned long, thermal_pressure);

>  


+1 on the rewrite!

Many thanks,
Ionela.

> 

> -- 

> viresh
Viresh Kumar Jan. 22, 2021, 7:44 a.m. | #4
On 19-01-21, 19:05, Ionela Voinescu wrote:
> I think it would improve clarity a lot by having a single topology

> function, with clear comments on which path is taken when.

> 

> In regards to them sharing code, there's not much code they could be

> sharing, as one just does some simple math on provided values.

> I attempted to have a single function for frequency invariance at some

> point [1], as the logic for the computation is the same, but I could

> never convince myself the clarity gained was worth the changes. But

> I really like this version of your code for this purpose as well.

> 

> But I'll leave that to your judgement.


Hmm, I don't really agree to merging those two paths into a single
routine unless we share something there. So I would like to skip that
for now :)

All other comments accepted.

Thanks.

-- 
viresh

Patch

diff --git a/arch/arm64/include/asm/topology.h b/arch/arm64/include/asm/topology.h
index 3b8dca4eb08d..be6a53ba3e2d 100644
--- a/arch/arm64/include/asm/topology.h
+++ b/arch/arm64/include/asm/topology.h
@@ -17,15 +17,9 @@  int pcibus_to_node(struct pci_bus *bus);
 #include <linux/arch_topology.h>
 
 void update_freq_counters_refs(void);
-void topology_scale_freq_tick(void);
 
-#ifdef CONFIG_ARM64_AMU_EXTN
-/*
- * Replace task scheduler's default counter-based
- * frequency-invariance scale factor setting.
- */
+/* Replace task scheduler's default frequency-invariance scale factor setting */
 #define arch_scale_freq_tick topology_scale_freq_tick
-#endif /* CONFIG_ARM64_AMU_EXTN */
 
 /* Replace task scheduler's default frequency-invariant accounting */
 #define arch_set_freq_scale topology_set_freq_scale
diff --git a/arch/arm64/kernel/topology.c b/arch/arm64/kernel/topology.c
index 0fc2b32ddb45..22f056821585 100644
--- a/arch/arm64/kernel/topology.c
+++ b/arch/arm64/kernel/topology.c
@@ -199,8 +199,44 @@  static int freq_inv_set_max_ratio(int cpu, u64 max_rate, u64 ref_rate)
 	return 0;
 }
 
-static DEFINE_STATIC_KEY_FALSE(amu_fie_key);
-#define amu_freq_invariant() static_branch_unlikely(&amu_fie_key)
+static void amu_scale_freq_tick(void)
+{
+	u64 prev_core_cnt, prev_const_cnt;
+	u64 core_cnt, const_cnt, scale;
+
+	prev_const_cnt = this_cpu_read(arch_const_cycles_prev);
+	prev_core_cnt = this_cpu_read(arch_core_cycles_prev);
+
+	update_freq_counters_refs();
+
+	const_cnt = this_cpu_read(arch_const_cycles_prev);
+	core_cnt = this_cpu_read(arch_core_cycles_prev);
+
+	if (unlikely(core_cnt <= prev_core_cnt ||
+		     const_cnt <= prev_const_cnt))
+		return;
+
+	/*
+	 *	    /\core    arch_max_freq_scale
+	 * scale =  ------- * --------------------
+	 *	    /\const   SCHED_CAPACITY_SCALE
+	 *
+	 * See validate_cpu_freq_invariance_counters() for details on
+	 * arch_max_freq_scale and the use of SCHED_CAPACITY_SHIFT.
+	 */
+	scale = core_cnt - prev_core_cnt;
+	scale *= this_cpu_read(arch_max_freq_scale);
+	scale = div64_u64(scale >> SCHED_CAPACITY_SHIFT,
+			  const_cnt - prev_const_cnt);
+
+	scale = min_t(unsigned long, scale, SCHED_CAPACITY_SCALE);
+	this_cpu_write(freq_scale, (unsigned long)scale);
+}
+
+static struct scale_freq_tick_data amu_sftd = {
+	.source = SCALE_FREQ_SOURCE_AMU,
+	.scale_freq = amu_scale_freq_tick,
+};
 
 static void amu_fie_setup(const struct cpumask *cpus)
 {
@@ -227,7 +263,7 @@  static void amu_fie_setup(const struct cpumask *cpus)
 	if (!invariant && !cpumask_equal(amu_fie_cpus, cpu_present_mask))
 		return;
 
-	static_branch_enable(&amu_fie_key);
+	topology_set_scale_freq_tick(&amu_sftd, amu_fie_cpus);
 
 	pr_info("CPUs[%*pbl]: counters will be used for FIE.",
 		cpumask_pr_args(amu_fie_cpus));
@@ -283,53 +319,6 @@  static int __init init_amu_fie(void)
 }
 core_initcall(init_amu_fie);
 
-bool arch_freq_counters_available(const struct cpumask *cpus)
-{
-	return amu_freq_invariant() &&
-	       cpumask_subset(cpus, amu_fie_cpus);
-}
-
-void topology_scale_freq_tick(void)
-{
-	u64 prev_core_cnt, prev_const_cnt;
-	u64 core_cnt, const_cnt, scale;
-	int cpu = smp_processor_id();
-
-	if (!amu_freq_invariant())
-		return;
-
-	if (!cpumask_test_cpu(cpu, amu_fie_cpus))
-		return;
-
-	prev_const_cnt = this_cpu_read(arch_const_cycles_prev);
-	prev_core_cnt = this_cpu_read(arch_core_cycles_prev);
-
-	update_freq_counters_refs();
-
-	const_cnt = this_cpu_read(arch_const_cycles_prev);
-	core_cnt = this_cpu_read(arch_core_cycles_prev);
-
-	if (unlikely(core_cnt <= prev_core_cnt ||
-		     const_cnt <= prev_const_cnt))
-		return;
-
-	/*
-	 *	    /\core    arch_max_freq_scale
-	 * scale =  ------- * --------------------
-	 *	    /\const   SCHED_CAPACITY_SCALE
-	 *
-	 * See validate_cpu_freq_invariance_counters() for details on
-	 * arch_max_freq_scale and the use of SCHED_CAPACITY_SHIFT.
-	 */
-	scale = core_cnt - prev_core_cnt;
-	scale *= this_cpu_read(arch_max_freq_scale);
-	scale = div64_u64(scale >> SCHED_CAPACITY_SHIFT,
-			  const_cnt - prev_const_cnt);
-
-	scale = min_t(unsigned long, scale, SCHED_CAPACITY_SCALE);
-	this_cpu_write(freq_scale, (unsigned long)scale);
-}
-
 #ifdef CONFIG_ACPI_CPPC_LIB
 #include <acpi/cppc_acpi.h>
 
diff --git a/drivers/base/arch_topology.c b/drivers/base/arch_topology.c
index de8587cc119e..07774fb47bc4 100644
--- a/drivers/base/arch_topology.c
+++ b/drivers/base/arch_topology.c
@@ -21,17 +21,65 @@ 
 #include <linux/sched.h>
 #include <linux/smp.h>
 
+static DEFINE_PER_CPU(struct scale_freq_tick_data *, sft_data);
+static struct cpumask scale_freq_tick_mask;
+
+static bool supports_scale_freq_tick(const struct cpumask *cpus)
+{
+	return cpumask_subset(cpus, &scale_freq_tick_mask);
+}
+
 bool topology_scale_freq_invariant(void)
 {
 	return cpufreq_supports_freq_invariance() ||
-	       arch_freq_counters_available(cpu_online_mask);
+	       supports_scale_freq_tick(cpu_online_mask);
+}
+
+void topology_set_scale_freq_tick(struct scale_freq_tick_data *data,
+				  const struct cpumask *cpus)
+{
+	struct scale_freq_tick_data *sftd;
+	int cpu;
+
+	for_each_cpu(cpu, cpus) {
+		sftd = per_cpu(sft_data, cpu);
+
+		/* Use AMU counters whenever possible */
+		if (!sftd || sftd->source != SCALE_FREQ_SOURCE_AMU) {
+			per_cpu(sft_data, cpu) = data;
+			cpumask_set_cpu(cpu, &scale_freq_tick_mask);
+		}
+	}
 }
+EXPORT_SYMBOL_GPL(topology_set_scale_freq_tick);
 
-__weak bool arch_freq_counters_available(const struct cpumask *cpus)
+void topology_clear_scale_freq_tick(enum scale_freq_tick_source source,
+				    const struct cpumask *cpus)
 {
-	return false;
+	struct scale_freq_tick_data *sftd;
+	int cpu;
+
+	for_each_cpu(cpu, cpus) {
+		sftd = per_cpu(sft_data, cpu);
+
+		if (sftd && sftd->source == source) {
+			per_cpu(sft_data, cpu) = NULL;
+			cpumask_clear_cpu(cpu, &scale_freq_tick_mask);
+		}
+	}
 }
+EXPORT_SYMBOL_GPL(topology_clear_scale_freq_tick);
+
+void topology_scale_freq_tick(void)
+{
+	struct scale_freq_tick_data *sftd = *this_cpu_ptr(&sft_data);
+
+	if (sftd)
+		sftd->scale_freq();
+}
+
 DEFINE_PER_CPU(unsigned long, freq_scale) = SCHED_CAPACITY_SCALE;
+EXPORT_SYMBOL_GPL(freq_scale);
 
 void topology_set_freq_scale(const struct cpumask *cpus, unsigned long cur_freq,
 			     unsigned long max_freq)
@@ -47,7 +95,7 @@  void topology_set_freq_scale(const struct cpumask *cpus, unsigned long cur_freq,
 	 * want to update the scale factor with information from CPUFREQ.
 	 * Instead the scale factor will be updated from arch_scale_freq_tick.
 	 */
-	if (arch_freq_counters_available(cpus))
+	if (supports_scale_freq_tick(cpus))
 		return;
 
 	scale = (cur_freq << SCHED_CAPACITY_SHIFT) / max_freq;
diff --git a/include/linux/arch_topology.h b/include/linux/arch_topology.h
index 0f6cd6b73a61..b2422ebef2dd 100644
--- a/include/linux/arch_topology.h
+++ b/include/linux/arch_topology.h
@@ -34,7 +34,18 @@  void topology_set_freq_scale(const struct cpumask *cpus, unsigned long cur_freq,
 			     unsigned long max_freq);
 bool topology_scale_freq_invariant(void);
 
-bool arch_freq_counters_available(const struct cpumask *cpus);
+enum scale_freq_tick_source {
+	SCALE_FREQ_SOURCE_AMU,
+};
+
+struct scale_freq_tick_data {
+	enum scale_freq_tick_source source;
+	void (*scale_freq)(void);
+};
+
+void topology_scale_freq_tick(void);
+void topology_set_scale_freq_tick(struct scale_freq_tick_data *data, const struct cpumask *cpus);
+void topology_clear_scale_freq_tick(enum scale_freq_tick_source source, const struct cpumask *cpus);
 
 DECLARE_PER_CPU(unsigned long, thermal_pressure);