Skip to contentSkip to conversation

cpufreq: intel_pstate: Adjust the energy model for hybrid systems

PATCH Superseded Rafael J. Wysocki 4 participants 12 messages Oct 15, 2025 – Oct 17, 2025

Subsystems (2)
lore
Share & tools
lore

Rafael J. Wysocki

author maintainer
#0 lore raw

v2 cover cpufreq: intel_pstate: Adjust the energy model for hybrid systems

Hi Everyone,

This supersedes

https://lore.kernel.org/linux-pm/[email protected]/

because the last patch in it resulted from some confusion. Sorry about that.

In this series patch [3/3] simply switches over the energy model callbacks to using CPU types retrieved from CPUID instead of scaling factors which is more robust and allows the energy model to be tuned more precisely. In particular, it will now allow P-cores to be populated when E-cores are utilized above 60% of their capacity (and not just in the "overutilized" situation or when there is a task whose utilization exceeds the E-core capacity).

The first two patches in the series are preliminary.

Thanks!

author maintainer
#1 lore raw

PATCH v2 1/3 cpufreq: intel_pstate: Add and use hybrid_get_cpu_type()

From: Rafael J. Wysocki <[email protected]>

Introduce a function for identifying the type of a given CPU in a hybrid system, called hybrid_get_cpu_type(), and use if for hybrid scaling factor determination in hwp_get_cpu_scaling().

---

v1 -> v2: No changes

---
 drivers/cpufreq/intel_pstate.c |   13 +++++++------
 1 file changed, 7 insertions(+), 6 deletions(-)
drivers/cpufreq/intel_pstate.c +7 -6
@@ -912,6 +912,11 @@ static struct freq_attr *hwp_cpufreq_att
 	[HWP_CPUFREQ_ATTR_COUNT] = NULL,
 };
 
+static u8 hybrid_get_cpu_type(unsigned int cpu)
+{
+	return cpu_data(cpu).topo.intel_type;
+}
+
 static bool no_cas __ro_after_init;
 
 static struct cpudata *hybrid_max_perf_cpu __read_mostly;
@@ -2298,18 +2303,14 @@ static int knl_get_turbo_pstate(int cpu)
 static int hwp_get_cpu_scaling(int cpu)
 {
 	if (hybrid_scaling_factor) {
-		struct cpuinfo_x86 *c = &cpu_data(cpu);
-		u8 cpu_type = c->topo.intel_type;
-
 		/*
 		 * Return the hybrid scaling factor for P-cores and use the
 		 * default core scaling for E-cores.
 		 */
-		if (cpu_type == INTEL_CPU_TYPE_CORE)
+		if (hybrid_get_cpu_type(cpu) == INTEL_CPU_TYPE_CORE)
 			return hybrid_scaling_factor;
 
-		if (cpu_type == INTEL_CPU_TYPE_ATOM)
-			return core_get_scaling();
+		return core_get_scaling();
 	}
 
 	/* Use core scaling on non-hybrid systems. */
author maintainer
#2 lore raw

PATCH v2 2/3 cpufreq: intel_pstate: Add and use hybrid_has_l3()

In reply to Rafael J. Wysocki

From: Rafael J. Wysocki <[email protected]>

Introduce a function for checking whether or not a given CPU has L3 cache, called hybrid_has_l3(), and use it in hybrid_get_cost() for computing cost coefficients associated with a given perf domain.

---

v1 -> v2: No changes

---
 drivers/cpufreq/intel_pstate.c |   30 ++++++++++++++++++------------
 1 file changed, 18 insertions(+), 12 deletions(-)
drivers/cpufreq/intel_pstate.c +18 -12
@@ -951,11 +951,26 @@ static int hybrid_active_power(struct de
 	return 0;
 }
 
+static bool hybrid_has_l3(unsigned int cpu)
+{
+	struct cpu_cacheinfo *cacheinfo = get_cpu_cacheinfo(cpu);
+	unsigned int i;
+
+	if (!cacheinfo)
+		return false;
+
+	for (i = 0; i < cacheinfo->num_leaves; i++) {
+		if (cacheinfo->info_list[i].level == 3)
+			return true;
+	}
+
+	return false;
+}
+
 static int hybrid_get_cost(struct device *dev, unsigned long freq,
 			   unsigned long *cost)
 {
 	struct pstate_data *pstate = &all_cpu_data[dev->id]->pstate;
-	struct cpu_cacheinfo *cacheinfo = get_cpu_cacheinfo(dev->id);
 
 	/*
 	 * The smaller the perf-to-frequency scaling factor, the larger the IPC
@@ -973,17 +988,8 @@ static int hybrid_get_cost(struct device
 	 * touching it in case some other CPUs of the same type can do the work
 	 * without it.
 	 */
-	if (cacheinfo) {
-		unsigned int i;
-
-		/* Check if L3 cache is there. */
-		for (i = 0; i < cacheinfo->num_leaves; i++) {
-			if (cacheinfo->info_list[i].level == 3) {
-				*cost += 2;
-				break;
-			}
-		}
-	}
+	if (hybrid_has_l3(dev->id))
+		*cost += 2;
 
 	return 0;
 }
author maintainer
#3 lore raw

PATCH v2 3/3 cpufreq: intel_pstate: hybrid: Adjust energy model rules

In reply to Rafael J. Wysocki

From: Rafael J. Wysocki <[email protected]>

Instead of using HWP-to-frequency scaling factors for computing cost coefficients in the energy model used on hybrid systems, which is fragile, rely on CPU type information that is easily accessible now and the information on whether or not L3 cache is present for this purpose.

This also allows the cost coefficients for P-cores to be adjusted so that they start to be populated somewhat earlier (that is, before E-cores are loaded up to their full capacity).

In addition to the above, replace an inaccurate comment regarding the reason why the freq value is added to the cost in hybrid_get_cost().

---
 drivers/cpufreq/intel_pstate.c |   37 +++++++++++++++----------------------
 1 file changed, 15 insertions(+), 22 deletions(-)
drivers/cpufreq/intel_pstate.c +15 -22
@@ -933,11 +933,8 @@ static int hybrid_active_power(struct de
 			       unsigned long *freq)
 {
 	/*
-	 * Create "utilization bins" of 0-40%, 40%-60%, 60%-80%, and 80%-100%
-	 * of the maximum capacity such that two CPUs of the same type will be
-	 * regarded as equally attractive if the utilization of each of them
-	 * falls into the same bin, which should prevent tasks from being
-	 * migrated between them too often.
+	 * Create four "states" corresponding to 40%, 60%, 80%, and 100% of the
+	 * full capacity.
 	 *
 	 * For this purpose, return the "frequency" of 2 for the first
 	 * performance level and otherwise leave the value set by the caller.
@@ -970,26 +967,22 @@ static bool hybrid_has_l3(unsigned int c
 static int hybrid_get_cost(struct device *dev, unsigned long freq,
 			   unsigned long *cost)
 {
-	struct pstate_data *pstate = &all_cpu_data[dev->id]->pstate;
-
+	/* Facilitate load balancing between CPUs of the same type. */
+	*cost = freq;
 	/*
-	 * The smaller the perf-to-frequency scaling factor, the larger the IPC
-	 * ratio between the given CPU and the least capable CPU in the system.
-	 * Regard that IPC ratio as the primary cost component and assume that
-	 * the scaling factors for different CPU types will differ by at least
-	 * 5% and they will not be above INTEL_PSTATE_CORE_SCALING.
+	 * Adjust the cost depending on CPU type.
 	 *
-	 * Add the freq value to the cost, so that the cost of running on CPUs
-	 * of the same type in different "utilization bins" is different.
-	 */
-	*cost = div_u64(100ULL * INTEL_PSTATE_CORE_SCALING, pstate->scaling) + freq;
-	/*
-	 * Increase the cost slightly for CPUs able to access L3 to avoid
-	 * touching it in case some other CPUs of the same type can do the work
-	 * without it.
+	 * The idea is to start loading up LPE-cores before E-cores and start
+	 * to populate E-cores when LPE-cores are utilized above 60% of the
+	 * capacity.  Similarly, P-cores start to be populated when E-cores are
+	 * utilized above 60% of the capacity.
 	 */
-	if (hybrid_has_l3(dev->id))
-		*cost += 2;
+	if (hybrid_get_cpu_type(dev->id) == INTEL_CPU_TYPE_ATOM) {
+		if (hybrid_has_l3(dev->id)) /* E-core */
+			*cost += 2;
+	} else { /* P-core */
+		*cost += 4;
+	}
 
 	return 0;
 }
#4 lore raw

Re: [PATCH v2 3/3] cpufreq: intel_pstate: hybrid: Adjust energy model rules

Quoted 62 lines · Rafael J. Wysocki

From: Rafael J. Wysocki <[email protected]>

Instead of using HWP-to-frequency scaling factors for computing cost coefficients in the energy model used on hybrid systems, which is fragile, rely on CPU type information that is easily accessible now and the information on whether or not L3 cache is present for this purpose.

This also allows the cost coefficients for P-cores to be adjusted so that they start to be populated somewhat earlier (that is, before E-cores are loaded up to their full capacity).

In addition to the above, replace an inaccurate comment regarding the reason why the freq value is added to the cost in hybrid_get_cost().

---
 drivers/cpufreq/intel_pstate.c |   37 +++++++++++++++----------------------
 1 file changed, 15 insertions(+), 22 deletions(-)
drivers/cpufreq/intel_pstate.c +8 -20
@@ -933,11 +933,8 @@ static int hybrid_active_power(struct de
 			       unsigned long *freq)
 {
 	/*
-	 * Create "utilization bins" of 0-40%, 40%-60%, 60%-80%, and 80%-100%
-	 * of the maximum capacity such that two CPUs of the same type will be
-	 * regarded as equally attractive if the utilization of each of them
-	 * falls into the same bin, which should prevent tasks from being
-	 * migrated between them too often.
+	 * Create four "states" corresponding to 40%, 60%, 80%, and 100% of the
+	 * full capacity.
 	 *
 	 * For this purpose, return the "frequency" of 2 for the first
 	 * performance level and otherwise leave the value set by the caller.
@@ -970,26 +967,22 @@ static bool hybrid_has_l3(unsigned int c
 static int hybrid_get_cost(struct device *dev, unsigned long freq,
 			   unsigned long *cost)
 {
-	struct pstate_data *pstate = &all_cpu_data[dev->id]->pstate;
-
+	/* Facilitate load balancing between CPUs of the same type. */
+	*cost = freq;
 	/*
-	 * The smaller the perf-to-frequency scaling factor, the larger the IPC
-	 * ratio between the given CPU and the least capable CPU in the system.
-	 * Regard that IPC ratio as the primary cost component and assume that
-	 * the scaling factors for different CPU types will differ by at least
-	 * 5% and they will not be above INTEL_PSTATE_CORE_SCALING.
+	 * Adjust the cost depending on CPU type.
 	 *
-	 * Add the freq value to the cost, so that the cost of running on CPUs
-	 * of the same type in different "utilization bins" is different.
-	 */
-	*cost = div_u64(100ULL * INTEL_PSTATE_CORE_SCALING, pstate->scaling) + freq;
-	/*
-	 * Increase the cost slightly for CPUs able to access L3 to avoid
-	 * touching it in case some other CPUs of the same type can do the work
-	 * without it.
+	 * The idea is to start loading up LPE-cores before E-cores and start
+	 * to populate E-cores when LPE-cores are utilized above 60% of the
+	 * capacity.  Similarly, P-cores start to be populated when E-cores are

drivers/cpufreq/intel_pstate.c static bool hybrid_has_l3(unsigned int c · line 985

Quoted 10 lines
drivers/cpufreq/intel_pstate.c +7 -2
+	 * utilized above 60% of the capacity.
 	 */
-	if (hybrid_has_l3(dev->id))
-		*cost += 2;
+	if (hybrid_get_cpu_type(dev->id) == INTEL_CPU_TYPE_ATOM) {
+		if (hybrid_has_l3(dev->id)) /* E-core */
+			*cost += 2;
+	} else { /* P-core */
+		*cost += 4;
+	}

Interesting, is there any reason in particular why you're looking to change this? Is it just performance because of the extra headroom? (I recall that your E-cores are always more efficient than your P-cores at comparable computing power).

How long does it take to trigger overutilized for you?
I still have the OU based on 'last idle time' observation patches lying around, although I haven't found the time to do more extensive testing if it doesn't regress some platform / workload combination. I will dust them off soon, although I'm not sure if they would help your case.
Happy to try though if you had a particular workload in mind!

author maintainer
#5 lore raw

Re: [PATCH v2 3/3] cpufreq: intel_pstate: hybrid: Adjust energy model rules

Quoted 70 lines · Christian Loehle
Quoted 68 lines · Rafael J. Wysocki

From: Rafael J. Wysocki <[email protected]>

Instead of using HWP-to-frequency scaling factors for computing cost coefficients in the energy model used on hybrid systems, which is fragile, rely on CPU type information that is easily accessible now and the information on whether or not L3 cache is present for this purpose.

This also allows the cost coefficients for P-cores to be adjusted so that they start to be populated somewhat earlier (that is, before E-cores are loaded up to their full capacity).

In addition to the above, replace an inaccurate comment regarding the reason why the freq value is added to the cost in hybrid_get_cost().

---
 drivers/cpufreq/intel_pstate.c |   37 +++++++++++++++----------------------
 1 file changed, 15 insertions(+), 22 deletions(-)
drivers/cpufreq/intel_pstate.c +11 -22
@@ -933,11 +933,8 @@ static int hybrid_active_power(struct de
                             unsigned long *freq)
 {
      /*
-      * Create "utilization bins" of 0-40%, 40%-60%, 60%-80%, and 80%-100%
-      * of the maximum capacity such that two CPUs of the same type will be
-      * regarded as equally attractive if the utilization of each of them
-      * falls into the same bin, which should prevent tasks from being
-      * migrated between them too often.
+      * Create four "states" corresponding to 40%, 60%, 80%, and 100% of the
+      * full capacity.
       *
       * For this purpose, return the "frequency" of 2 for the first
       * performance level and otherwise leave the value set by the caller.
@@ -970,26 +967,22 @@ static bool hybrid_has_l3(unsigned int c
 static int hybrid_get_cost(struct device *dev, unsigned long freq,
                         unsigned long *cost)
 {
-     struct pstate_data *pstate = &all_cpu_data[dev->id]->pstate;
-
+     /* Facilitate load balancing between CPUs of the same type. */
+     *cost = freq;
      /*
-      * The smaller the perf-to-frequency scaling factor, the larger the IPC
-      * ratio between the given CPU and the least capable CPU in the system.
-      * Regard that IPC ratio as the primary cost component and assume that
-      * the scaling factors for different CPU types will differ by at least
-      * 5% and they will not be above INTEL_PSTATE_CORE_SCALING.
+      * Adjust the cost depending on CPU type.
       *
-      * Add the freq value to the cost, so that the cost of running on CPUs
-      * of the same type in different "utilization bins" is different.
-      */
-     *cost = div_u64(100ULL * INTEL_PSTATE_CORE_SCALING, pstate->scaling) + freq;
-     /*
-      * Increase the cost slightly for CPUs able to access L3 to avoid
-      * touching it in case some other CPUs of the same type can do the work
-      * without it.
+      * The idea is to start loading up LPE-cores before E-cores and start
+      * to populate E-cores when LPE-cores are utilized above 60% of the
+      * capacity.  Similarly, P-cores start to be populated when E-cores are
+      * utilized above 60% of the capacity.
       */
-     if (hybrid_has_l3(dev->id))
-             *cost += 2;
+     if (hybrid_get_cpu_type(dev->id) == INTEL_CPU_TYPE_ATOM) {
+             if (hybrid_has_l3(dev->id)) /* E-core */
Quoted 8 lines

drivers/cpufreq/intel_pstate.c static bool hybrid_has_l3(unsigned int c · line 985

Quoted 4 lines
drivers/cpufreq/intel_pstate.c +4 -0
+                     *cost += 2;
+     } else { /* P-core */
+             *cost += 4;
+     }

Interesting, is there any reason in particular why you're looking to change this? Is it just performance because of the extra headroom? (I recall that your E-cores are always more efficient than your P-cores at comparable computing power).

Yes, it is performance mostly.

Quoted 1 line

How long does it take to trigger overutilized for you?

It depends, but sometimes it triggers really quickly due to LPE-core load spikes.

Quoted 5 lines

I still have the OU based on 'last idle time' observation patches lying around, although I haven't found the time to do more extensive testing if it doesn't regress some platform / workload combination. I will dust them off soon, although I'm not sure if they would help your case.
Happy to try though if you had a particular workload in mind!

OK, good to know, thanks!

#6 lore raw

Re: [PATCH v2 3/3] cpufreq: intel_pstate: hybrid: Adjust energy model rules

In reply to Rafael J. Wysocki

在 2025/10/15 21:48, Rafael J. Wysocki 写道:> From: Rafael J. Wysocki <[email protected]>

Quoted 64 lines

Instead of using HWP-to-frequency scaling factors for computing cost coefficients in the energy model used on hybrid systems, which is fragile, rely on CPU type information that is easily accessible now and the information on whether or not L3 cache is present for this purpose.

This also allows the cost coefficients for P-cores to be adjusted so that they start to be populated somewhat earlier (that is, before E-cores are loaded up to their full capacity).

In addition to the above, replace an inaccurate comment regarding the reason why the freq value is added to the cost in hybrid_get_cost().

---
  drivers/cpufreq/intel_pstate.c |   37 +++++++++++++++----------------------
  1 file changed, 15 insertions(+), 22 deletions(-)
drivers/cpufreq/intel_pstate.c +9 -21
@@ -933,11 +933,8 @@ static int hybrid_active_power(struct de
  			       unsigned long *freq)
  {
  	/*
-	 * Create "utilization bins" of 0-40%, 40%-60%, 60%-80%, and 80%-100%
-	 * of the maximum capacity such that two CPUs of the same type will be
-	 * regarded as equally attractive if the utilization of each of them
-	 * falls into the same bin, which should prevent tasks from being
-	 * migrated between them too often.
+	 * Create four "states" corresponding to 40%, 60%, 80%, and 100% of the
+	 * full capacity.
  	 *
  	 * For this purpose, return the "frequency" of 2 for the first
  	 * performance level and otherwise leave the value set by the caller.
@@ -970,26 +967,22 @@ static bool hybrid_has_l3(unsigned int c
  static int hybrid_get_cost(struct device *dev, unsigned long freq,
  			   unsigned long *cost)
  {
-	struct pstate_data *pstate = &all_cpu_data[dev->id]->pstate;
-
+	/* Facilitate load balancing between CPUs of the same type. */
+	*cost = freq;
  	/*
-	 * The smaller the perf-to-frequency scaling factor, the larger the IPC
-	 * ratio between the given CPU and the least capable CPU in the system.
-	 * Regard that IPC ratio as the primary cost component and assume that
-	 * the scaling factors for different CPU types will differ by at least
-	 * 5% and they will not be above INTEL_PSTATE_CORE_SCALING.
+	 * Adjust the cost depending on CPU type.
  	 *
-	 * Add the freq value to the cost, so that the cost of running on CPUs
-	 * of the same type in different "utilization bins" is different.
-	 */
-	*cost = div_u64(100ULL * INTEL_PSTATE_CORE_SCALING, pstate->scaling) + freq;
-	/*
-	 * Increase the cost slightly for CPUs able to access L3 to avoid
-	 * touching it in case some other CPUs of the same type can do the work
-	 * without it.
+	 * The idea is to start loading up LPE-cores before E-cores and start
+	 * to populate E-cores when LPE-cores are utilized above 60% of the
+	 * capacity.  Similarly, P-cores start to be populated when E-cores are
+	 * utilized above 60% of the capacity.
  	 */
-	if (hybrid_has_l3(dev->id))

drivers/cpufreq/intel_pstate.c static bool hybrid_has_l3(unsigned int c · line 988

Quoted 10 lines
drivers/cpufreq/intel_pstate.c +6 -1
-		*cost += 2;
+	if (hybrid_get_cpu_type(dev->id) == INTEL_CPU_TYPE_ATOM) {
+		if (hybrid_has_l3(dev->id)) /* E-core */
+			*cost += 2;
+	} else { /* P-core */
+		*cost += 4;
+	}
 
  	return 0;
  }

Hi Rafael J. Wysocki:

Is the increment of this cost for different types of CPUs by one instead of two?

cost by increment of 2:

          0~40%  40%~60%  60%~80% 80%~100
LPE-core    2       3        4      5
E-core      4       5        6      7
P-core      6       7        8      9

So, tasks only start being allocated to more powerful CPUs when utilization exceeds 80%, but by that point the system is already in an
overloaded state.

cost by increment of 1:

          0~40%  40%~60%  60%~80% 80%~100
LPE-core    2       3        4      5
E-core      3       4        5      6
P-core      4       5        6      7

This situation aligns with the description in your patch.

The idea of this patch looks good to me.

#7 lore raw

Re: [PATCH v2 3/3] cpufreq: intel_pstate: hybrid: Adjust energy model rules

Quoted 92 lines · Yaxiong Tian

在 2025/10/15 21:48, Rafael J. Wysocki 写道:> From: Rafael J. Wysocki <[email protected]>

Quoted 64 lines

Instead of using HWP-to-frequency scaling factors for computing cost coefficients in the energy model used on hybrid systems, which is fragile, rely on CPU type information that is easily accessible now and the information on whether or not L3 cache is present for this purpose.

This also allows the cost coefficients for P-cores to be adjusted so that they start to be populated somewhat earlier (that is, before E-cores are loaded up to their full capacity).

In addition to the above, replace an inaccurate comment regarding the reason why the freq value is added to the cost in hybrid_get_cost().

---
  drivers/cpufreq/intel_pstate.c |   37 +++++++++++++++----------------------
  1 file changed, 15 insertions(+), 22 deletions(-)
drivers/cpufreq/intel_pstate.c +9 -21
@@ -933,11 +933,8 @@ static int hybrid_active_power(struct de
  			       unsigned long *freq)
  {
  	/*
-	 * Create "utilization bins" of 0-40%, 40%-60%, 60%-80%, and 80%-100%
-	 * of the maximum capacity such that two CPUs of the same type will be
-	 * regarded as equally attractive if the utilization of each of them
-	 * falls into the same bin, which should prevent tasks from being
-	 * migrated between them too often.
+	 * Create four "states" corresponding to 40%, 60%, 80%, and 100% of the
+	 * full capacity.
  	 *
  	 * For this purpose, return the "frequency" of 2 for the first
  	 * performance level and otherwise leave the value set by the caller.
@@ -970,26 +967,22 @@ static bool hybrid_has_l3(unsigned int c
  static int hybrid_get_cost(struct device *dev, unsigned long freq,
  			   unsigned long *cost)
  {
-	struct pstate_data *pstate = &all_cpu_data[dev->id]->pstate;
-
+	/* Facilitate load balancing between CPUs of the same type. */
+	*cost = freq;
  	/*
-	 * The smaller the perf-to-frequency scaling factor, the larger the IPC
-	 * ratio between the given CPU and the least capable CPU in the system.
-	 * Regard that IPC ratio as the primary cost component and assume that
-	 * the scaling factors for different CPU types will differ by at least
-	 * 5% and they will not be above INTEL_PSTATE_CORE_SCALING.
+	 * Adjust the cost depending on CPU type.
  	 *
-	 * Add the freq value to the cost, so that the cost of running on CPUs
-	 * of the same type in different "utilization bins" is different.
-	 */
-	*cost = div_u64(100ULL * INTEL_PSTATE_CORE_SCALING, pstate->scaling) + freq;
-	/*
-	 * Increase the cost slightly for CPUs able to access L3 to avoid
-	 * touching it in case some other CPUs of the same type can do the work
-	 * without it.
+	 * The idea is to start loading up LPE-cores before E-cores and start
+	 * to populate E-cores when LPE-cores are utilized above 60% of the
+	 * capacity.  Similarly, P-cores start to be populated when E-cores are
+	 * utilized above 60% of the capacity.
  	 */
-	if (hybrid_has_l3(dev->id))

drivers/cpufreq/intel_pstate.c static bool hybrid_has_l3(unsigned int c · line 988

Quoted 10 lines
drivers/cpufreq/intel_pstate.c +6 -1
-		*cost += 2;
+	if (hybrid_get_cpu_type(dev->id) == INTEL_CPU_TYPE_ATOM) {
+		if (hybrid_has_l3(dev->id)) /* E-core */
+			*cost += 2;
+	} else { /* P-core */
+		*cost += 4;
+	}
 
  	return 0;
  }

Hi Rafael J. Wysocki:

Is the increment of this cost for different types of CPUs by one instead of two?

cost by increment of 2:

          0~40%  40%~60%  60%~80% 80%~100
LPE-core    2       3        4      5
E-core      4       5        6      7
P-core      6       7        8      9

So, tasks only start being allocated to more powerful CPUs when utilization exceeds 80%, but by that point the system is already in an
overloaded state.

cost by increment of 1:

Quoted 8 lines
          0~40%  40%~60%  60%~80% 80%~100
LPE-core    2       3        4      5
E-core      3       4        5      6
P-core      4       5        6      7

This situation aligns with the description in your patch.

The idea of this patch looks good to me.

Agreed if you want the threshold to be 60% for both it should be +1 for l3 and +2 for P-core.
Good catch!

author maintainer
#8 lore raw

Re: [PATCH v2 3/3] cpufreq: intel_pstate: hybrid: Adjust energy model rules

Quoted 99 lines · Christian Loehle
Quoted 97 lines · Yaxiong Tian

在 2025/10/15 21:48, Rafael J. Wysocki 写道:> From: Rafael J. Wysocki <[email protected]>

Quoted 64 lines

Instead of using HWP-to-frequency scaling factors for computing cost coefficients in the energy model used on hybrid systems, which is fragile, rely on CPU type information that is easily accessible now and the information on whether or not L3 cache is present for this purpose.

This also allows the cost coefficients for P-cores to be adjusted so that they start to be populated somewhat earlier (that is, before E-cores are loaded up to their full capacity).

In addition to the above, replace an inaccurate comment regarding the reason why the freq value is added to the cost in hybrid_get_cost().

---
  drivers/cpufreq/intel_pstate.c |   37 +++++++++++++++----------------------
  1 file changed, 15 insertions(+), 22 deletions(-)
drivers/cpufreq/intel_pstate.c +9 -21
@@ -933,11 +933,8 @@ static int hybrid_active_power(struct de
                            unsigned long *freq)
  {
     /*
-     * Create "utilization bins" of 0-40%, 40%-60%, 60%-80%, and 80%-100%
-     * of the maximum capacity such that two CPUs of the same type will be
-     * regarded as equally attractive if the utilization of each of them
-     * falls into the same bin, which should prevent tasks from being
-     * migrated between them too often.
+     * Create four "states" corresponding to 40%, 60%, 80%, and 100% of the
+     * full capacity.
      *
      * For this purpose, return the "frequency" of 2 for the first
      * performance level and otherwise leave the value set by the caller.
@@ -970,26 +967,22 @@ static bool hybrid_has_l3(unsigned int c
  static int hybrid_get_cost(struct device *dev, unsigned long freq,
                        unsigned long *cost)
  {
-    struct pstate_data *pstate = &all_cpu_data[dev->id]->pstate;
-
+    /* Facilitate load balancing between CPUs of the same type. */
+    *cost = freq;
     /*
-     * The smaller the perf-to-frequency scaling factor, the larger the IPC
-     * ratio between the given CPU and the least capable CPU in the system.
-     * Regard that IPC ratio as the primary cost component and assume that
-     * the scaling factors for different CPU types will differ by at least
-     * 5% and they will not be above INTEL_PSTATE_CORE_SCALING.
+     * Adjust the cost depending on CPU type.
      *
-     * Add the freq value to the cost, so that the cost of running on CPUs
-     * of the same type in different "utilization bins" is different.
-     */
-    *cost = div_u64(100ULL * INTEL_PSTATE_CORE_SCALING, pstate->scaling) + freq;
-    /*
-     * Increase the cost slightly for CPUs able to access L3 to avoid
-     * touching it in case some other CPUs of the same type can do the work
-     * without it.
+     * The idea is to start loading up LPE-cores before E-cores and start
+     * to populate E-cores when LPE-cores are utilized above 60% of the
+     * capacity.  Similarly, P-cores start to be populated when E-cores are
+     * utilized above 60% of the capacity.
      */
-    if (hybrid_has_l3(dev->id))

drivers/cpufreq/intel_pstate.c static bool hybrid_has_l3(unsigned int c · line 988

Quoted 10 lines
drivers/cpufreq/intel_pstate.c +6 -1
-            *cost += 2;
+    if (hybrid_get_cpu_type(dev->id) == INTEL_CPU_TYPE_ATOM) {
+            if (hybrid_has_l3(dev->id)) /* E-core */
+                    *cost += 2;
+    } else { /* P-core */
+            *cost += 4;
+    }
 
     return 0;
  }

Hi Rafael J. Wysocki:

Is the increment of this cost for different types of CPUs by one instead of two?

cost by increment of 2:

          0~40%  40%~60%  60%~80% 80%~100
LPE-core    2       3        4      5
E-core      4       5        6      7
P-core      6       7        8      9

So, tasks only start being allocated to more powerful CPUs when utilization exceeds 80%, but by that point the system is already in an
overloaded state.

cost by increment of 1:

          0~40%  40%~60%  60%~80% 80%~100
LPE-core    2       3        4      5
E-core      3       4        5      6
P-core      4       5        6      7
Quoted 7 lines
Quoted 3 lines

This situation aligns with the description in your patch.

The idea of this patch looks good to me.

Agreed if you want the threshold to be 60% for both it should be +1 for l3 and +2 for P-core.
Good catch!

OK, I'll send an update of this patch then, but I won't resend the first two patches in the series. I don't think that they are objectionable.

Reviewed-by
#9 lore raw

Re: [PATCH v2 3/3] cpufreq: intel_pstate: hybrid: Adjust energy model rules

Quoted 5 lines · Rafael J. Wysocki
Quoted 3 lines · Christian Loehle
Quoted 1 line · Yaxiong Tian

在 2025/10/15 21:48, Rafael J. Wysocki 写道:> From: Rafael J. Wysocki <[email protected]>

[...]

Quoted 25 lines
Quoted 25 lines
Quoted 24 lines

Hi Rafael J. Wysocki:

Is the increment of this cost for different types of CPUs by one instead of two?

cost by increment of 2:

          0~40%  40%~60%  60%~80% 80%~100
LPE-core    2       3        4      5
E-core      4       5        6      7
P-core      6       7        8      9

So, tasks only start being allocated to more powerful CPUs when utilization exceeds 80%, but by that point the system is already in an
overloaded state.

cost by increment of 1:

          0~40%  40%~60%  60%~80% 80%~100
LPE-core    2       3        4      5
E-core      3       4        5      6
P-core      4       5        6      7

This situation aligns with the description in your patch.

The idea of this patch looks good to me.

Quoted 7 lines
Quoted 3 lines

Agreed if you want the threshold to be 60% for both it should be +1 for l3 and +2 for P-core.
Good catch!

OK, I'll send an update of this patch then, but I won't resend the first two patches in the series. I don't think that they are objectionable.

FWIW: LGTM: tested on i7-13700K (nosmt):

$ cat /sys/devices/system/cpu/cpu{0,8,16}/cpu_capacity
1009
1024
623
root@gulliver:~# cat /sys/kernel/debug/energy_model/cpu0/ps:*/cost
6
7
8
9
root@gulliver:~# cat /sys/kernel/debug/energy_model/cpu8/ps:*/cost
6
7
8
9
root@gulliver:~# cat /sys/kernel/debug/energy_model/cpu16/ps:*/cost
4
5
6
7
author maintainer
#10 lore raw

PATCH v3 3/3 cpufreq: intel_pstate: hybrid: Adjust energy model rules

In reply to Rafael J. Wysocki

From: Rafael J. Wysocki <[email protected]>

Instead of using HWP-to-frequency scaling factors for computing cost coefficients in the energy model used on hybrid systems, which is fragile, rely on CPU type information that is easily accessible now and the information on whether or not L3 cache is present for this purpose.

This also allows the cost coefficients for P-cores to be adjusted so that they start to be populated somewhat earlier (that is, before E-cores are loaded up to their full capacity).

In addition to the above, replace an inaccurate comment regarding the reason why the freq value is added to the cost in hybrid_get_cost().

---

v2 -> v3:
* Reduce cost differences between CPU types by 1 (Tian, Christian)
* Add a tag from Dietmar

Note: The other patches in the series have not changed.

v1 -> v2: New patch

---
 drivers/cpufreq/intel_pstate.c |   35 ++++++++++++++---------------------
 1 file changed, 14 insertions(+), 21 deletions(-)
drivers/cpufreq/intel_pstate.c +14 -21
@@ -933,11 +933,8 @@ static int hybrid_active_power(struct de
 			       unsigned long *freq)
 {
 	/*
-	 * Create "utilization bins" of 0-40%, 40%-60%, 60%-80%, and 80%-100%
-	 * of the maximum capacity such that two CPUs of the same type will be
-	 * regarded as equally attractive if the utilization of each of them
-	 * falls into the same bin, which should prevent tasks from being
-	 * migrated between them too often.
+	 * Create four "states" corresponding to 40%, 60%, 80%, and 100% of the
+	 * full capacity.
 	 *
 	 * For this purpose, return the "frequency" of 2 for the first
 	 * performance level and otherwise leave the value set by the caller.
@@ -970,26 +967,22 @@ static bool hybrid_has_l3(unsigned int c
 static int hybrid_get_cost(struct device *dev, unsigned long freq,
 			   unsigned long *cost)
 {
-	struct pstate_data *pstate = &all_cpu_data[dev->id]->pstate;
-
+	/* Facilitate load balancing between CPUs of the same type. */
+	*cost = freq;
 	/*
-	 * The smaller the perf-to-frequency scaling factor, the larger the IPC
-	 * ratio between the given CPU and the least capable CPU in the system.
-	 * Regard that IPC ratio as the primary cost component and assume that
-	 * the scaling factors for different CPU types will differ by at least
-	 * 5% and they will not be above INTEL_PSTATE_CORE_SCALING.
+	 * Adjust the cost depending on CPU type.
 	 *
-	 * Add the freq value to the cost, so that the cost of running on CPUs
-	 * of the same type in different "utilization bins" is different.
-	 */
-	*cost = div_u64(100ULL * INTEL_PSTATE_CORE_SCALING, pstate->scaling) + freq;
-	/*
-	 * Increase the cost slightly for CPUs able to access L3 to avoid
-	 * touching it in case some other CPUs of the same type can do the work
-	 * without it.
+	 * The idea is to start loading up LPE-cores before E-cores and start
+	 * to populate E-cores when LPE-cores are utilized above 60% of the
+	 * capacity.  Similarly, P-cores start to be populated when E-cores are
+	 * utilized above 60% of the capacity.
 	 */
-	if (hybrid_has_l3(dev->id))
+	if (hybrid_get_cpu_type(dev->id) == INTEL_CPU_TYPE_ATOM) {
+		if (hybrid_has_l3(dev->id)) /* E-core */
+			*cost += 1;
+	} else { /* P-core */
 		*cost += 2;
+	}
 
 	return 0;
 }
Reviewed-by on v3 3/3
#11 lore raw

Re: [PATCH v3 3/3] cpufreq: intel_pstate: hybrid: Adjust energy model rules

Quoted 75 lines · Rafael J. Wysocki

From: Rafael J. Wysocki <[email protected]>

Instead of using HWP-to-frequency scaling factors for computing cost coefficients in the energy model used on hybrid systems, which is fragile, rely on CPU type information that is easily accessible now and the information on whether or not L3 cache is present for this purpose.

This also allows the cost coefficients for P-cores to be adjusted so that they start to be populated somewhat earlier (that is, before E-cores are loaded up to their full capacity).

In addition to the above, replace an inaccurate comment regarding the reason why the freq value is added to the cost in hybrid_get_cost().

---

v2 -> v3:
* Reduce cost differences between CPU types by 1 (Tian, Christian)
* Add a tag from Dietmar

Note: The other patches in the series have not changed.

v1 -> v2: New patch

---
  drivers/cpufreq/intel_pstate.c |   35 ++++++++++++++---------------------
  1 file changed, 14 insertions(+), 21 deletions(-)
drivers/cpufreq/intel_pstate.c +9 -20
@@ -933,11 +933,8 @@ static int hybrid_active_power(struct de
  			       unsigned long *freq)
  {
  	/*
-	 * Create "utilization bins" of 0-40%, 40%-60%, 60%-80%, and 80%-100%
-	 * of the maximum capacity such that two CPUs of the same type will be
-	 * regarded as equally attractive if the utilization of each of them
-	 * falls into the same bin, which should prevent tasks from being
-	 * migrated between them too often.
+	 * Create four "states" corresponding to 40%, 60%, 80%, and 100% of the
+	 * full capacity.
  	 *
  	 * For this purpose, return the "frequency" of 2 for the first
  	 * performance level and otherwise leave the value set by the caller.
@@ -970,26 +967,22 @@ static bool hybrid_has_l3(unsigned int c
  static int hybrid_get_cost(struct device *dev, unsigned long freq,
  			   unsigned long *cost)
  {
-	struct pstate_data *pstate = &all_cpu_data[dev->id]->pstate;
-
+	/* Facilitate load balancing between CPUs of the same type. */
+	*cost = freq;
  	/*
-	 * The smaller the perf-to-frequency scaling factor, the larger the IPC
-	 * ratio between the given CPU and the least capable CPU in the system.
-	 * Regard that IPC ratio as the primary cost component and assume that
-	 * the scaling factors for different CPU types will differ by at least
-	 * 5% and they will not be above INTEL_PSTATE_CORE_SCALING.
+	 * Adjust the cost depending on CPU type.
  	 *
-	 * Add the freq value to the cost, so that the cost of running on CPUs
-	 * of the same type in different "utilization bins" is different.
-	 */
-	*cost = div_u64(100ULL * INTEL_PSTATE_CORE_SCALING, pstate->scaling) + freq;
-	/*
-	 * Increase the cost slightly for CPUs able to access L3 to avoid
-	 * touching it in case some other CPUs of the same type can do the work
-	 * without it.
+	 * The idea is to start loading up LPE-cores before E-cores and start
+	 * to populate E-cores when LPE-cores are utilized above 60% of the
+	 * capacity.  Similarly, P-cores start to be populated when E-cores are
+	 * utilized above 60% of the capacity.
  	 */

drivers/cpufreq/intel_pstate.c static bool hybrid_has_l3(unsigned int c · line 988

Quoted 10 lines
drivers/cpufreq/intel_pstate.c +5 -1
-	if (hybrid_has_l3(dev->id))
+	if (hybrid_get_cpu_type(dev->id) == INTEL_CPU_TYPE_ATOM) {
+		if (hybrid_has_l3(dev->id)) /* E-core */
+			*cost += 1;
+	} else { /* P-core */
  		*cost += 2;
+	}
 
  	return 0;
  }

LGTM