Skip to contentSkip to conversation

sched, arch_topology: Thermal pressure configuration cleanup

PATCH Applied Valentin Schneider 9 participants 18 messages Jul 12, 2020 – Jul 29, 2020

Subsystems (5)
lore
Share & tools
lore
#0 lore raw

v2 cover sched, arch_topology: Thermal pressure configuration cleanup

Hi folks,

This stems from this thread [1] on the list. TL;DR: the thermal pressure config has no helpful documentation, and figuring out if the right dependencies are in place is not easy for a regular user.

The current landscape also paints an odd picture: arch_set_thermal_pressure() is hardcoded in sched/core.c, and is *not* architecture-definable, while arch_get_thermal_pressure() is. Patch 1 is tackling this, the rest is Kconfig stuff.

Cheers,
Valentin

[1]: https://lkml.kernel.org/r/[email protected]

Revisions
=========

v1 -> v2
--------

o Remove cpufreq_cooling.c weak function; use #define stub in sched/topology.h (Vincent)
o Hinge arm SCHED_THERMAL_PRESSURE selection on ARM_CPU_TOPOLOGY

Valentin Schneider (3):
arch_topology, sched/core: Cleanup thermal pressure definition
sched: Cleanup SCHED_THERMAL_PRESSURE kconfig entry arm, arm64: Select CONFIG_SCHED_THERMAL_PRESSURE

 arch/arm/Kconfig                  |  1 +
 arch/arm/include/asm/topology.h   |  3 ++-
 arch/arm64/Kconfig                |  1 +
 arch/arm64/include/asm/topology.h |  3 ++-
 drivers/base/arch_topology.c      | 11 +++++++++++
 include/linux/arch_topology.h     |  4 ++--
 include/linux/sched/topology.h    |  7 +++++++
 init/Kconfig                      | 15 ++++++++++++++-
 kernel/sched/core.c               | 11 -----------
 9 files changed, 40 insertions(+), 16 deletions(-)

--
2.27.0

#1 lore raw

PATCH v2 1/3 arch_topology, sched/core: Cleanup thermal pressure definition

The following commit:

14533a16c46d ("thermal/cpu-cooling, sched/core: Move the arch_set_thermal_pressure() API to generic scheduler code")

moved the definition of arch_set_thermal_pressure() to sched/core.c, but kept its declaration in linux/arch_topology.h. When building e.g. an x86 kernel with CONFIG_SCHED_THERMAL_PRESSURE=y, cpufreq_cooling.c ends up getting the declaration of arch_set_thermal_pressure() from include/linux/arch_topology.h, which is somewhat awkward.

On top of this, sched/core.c unconditionally defines
o The thermal_pressure percpu variable
o arch_set_thermal_pressure()

while arch_scale_thermal_pressure() does nothing unless redefined by the architecture.

arch_*() functions are meant to be defined by architectures, so revert the aforementioned commit and re-implement it in a way that keeps arch_set_thermal_pressure() architecture-definable, and doesn't define the thermal pressure percpu variable for kernels that don't need it (CONFIG_SCHED_THERMAL_PRESSURE=n).

---
 arch/arm/include/asm/topology.h   |  3 ++-
 arch/arm64/include/asm/topology.h |  3 ++-
 drivers/base/arch_topology.c      | 11 +++++++++++
 include/linux/arch_topology.h     |  4 ++--
 include/linux/sched/topology.h    |  7 +++++++
 kernel/sched/core.c               | 11 -----------
 6 files changed, 24 insertions(+), 15 deletions(-)
arch/arm/include/asm/topology.h +2 -1
index 435aba289fc5..e0593cf095d0 100644
@@ -16,8 +16,9 @@
 /* Enable topology flag updates */
 #define arch_update_cpu_topology topology_update_cpu_topology
 
-/* Replace task scheduler's default thermal pressure retrieve API */
+/* Replace task scheduler's default thermal pressure API */
 #define arch_scale_thermal_pressure topology_get_thermal_pressure
+#define arch_set_thermal_pressure   topology_set_thermal_pressure
 
 #else
 
arch/arm64/include/asm/topology.h +2 -1
index 0cc835ddfcd1..e042f6527981 100644
@@ -34,8 +34,9 @@ void topology_scale_freq_tick(void);
 /* Enable topology flag updates */
 #define arch_update_cpu_topology topology_update_cpu_topology
 
-/* Replace task scheduler's default thermal pressure retrieve API */
+/* Replace task scheduler's default thermal pressure API */
 #define arch_scale_thermal_pressure topology_get_thermal_pressure
+#define arch_set_thermal_pressure   topology_set_thermal_pressure
 
 #include <asm-generic/topology.h>
 
drivers/base/arch_topology.c +11 -0
index 4d0a0038b476..d14cab7dfa3c 100644
@@ -54,6 +54,17 @@ void topology_set_cpu_scale(unsigned int cpu, unsigned long capacity)
 	per_cpu(cpu_scale, cpu) = capacity;
 }
 
+DEFINE_PER_CPU(unsigned long, thermal_pressure);
+
+void arch_set_thermal_pressure(const struct cpumask *cpus,
+			       unsigned long th_pressure)
+{
+	int cpu;
+
+	for_each_cpu(cpu, cpus)
+		WRITE_ONCE(per_cpu(thermal_pressure, cpu), th_pressure);
+}
+
 static ssize_t cpu_capacity_show(struct device *dev,
 				 struct device_attribute *attr,
 				 char *buf)
include/linux/arch_topology.h +2 -2
index 0566cb3314ef..69b1dabe39dc 100644
@@ -39,8 +39,8 @@ static inline unsigned long topology_get_thermal_pressure(int cpu)
 	return per_cpu(thermal_pressure, cpu);
 }
 
-void arch_set_thermal_pressure(struct cpumask *cpus,
-			       unsigned long th_pressure);
+void topology_set_thermal_pressure(const struct cpumask *cpus,
+				   unsigned long th_pressure);
 
 struct cpu_topology {
 	int thread_id;
include/linux/sched/topology.h +7 -0
index fb11091129b3..764222d637b7 100644
@@ -232,6 +232,13 @@ unsigned long arch_scale_thermal_pressure(int cpu)
 }
 #endif
 
+#ifndef arch_set_thermal_pressure
+static __always_inline
+void arch_set_thermal_pressure(const struct cpumask *cpus,
+			       unsigned long th_pressure)
+{ }
+#endif
+
 static inline int task_node(const struct task_struct *p)
 {
 	return cpu_to_node(task_cpu(p));
kernel/sched/core.c +0 -11
index ff0519551188..90b44f3840e4 100644
@@ -3731,17 +3731,6 @@ unsigned long long task_sched_runtime(struct task_struct *p)
 	return ns;
 }
 
-DEFINE_PER_CPU(unsigned long, thermal_pressure);
-
-void arch_set_thermal_pressure(struct cpumask *cpus,
-			       unsigned long th_pressure)
-{
-	int cpu;
-
-	for_each_cpu(cpu, cpus)
-		WRITE_ONCE(per_cpu(thermal_pressure, cpu), th_pressure);
-}
-
 /*
  * This function gets called by the timer code, with HZ frequency.
  * We call it with interrupts disabled.
Signature · 1 line
2.27.0
Reviewed-by on v2 1/3
#2 lore raw

Re: [PATCH v2 1/3] arch_topology, sched/core: Cleanup thermal pressure definition

Quoted 17 lines · Valentin Schneider

The following commit:

14533a16c46d ("thermal/cpu-cooling, sched/core: Move the arch_set_thermal_pressure() API to generic scheduler code")

moved the definition of arch_set_thermal_pressure() to sched/core.c, but kept its declaration in linux/arch_topology.h. When building e.g. an x86 kernel with CONFIG_SCHED_THERMAL_PRESSURE=y, cpufreq_cooling.c ends up getting the declaration of arch_set_thermal_pressure() from include/linux/arch_topology.h, which is somewhat awkward.

On top of this, sched/core.c unconditionally defines
o The thermal_pressure percpu variable
o arch_set_thermal_pressure()

while arch_scale_thermal_pressure() does nothing unless redefined by the architecture.

Quoted 8 lines

arch_*() functions are meant to be defined by architectures, so revert the aforementioned commit and re-implement it in a way that keeps arch_set_thermal_pressure() architecture-definable, and doesn't define the thermal pressure percpu variable for kernels that don't need it (CONFIG_SCHED_THERMAL_PRESSURE=n).

---

Quoted 116 lines
  arch/arm/include/asm/topology.h   |  3 ++-
  arch/arm64/include/asm/topology.h |  3 ++-
  drivers/base/arch_topology.c      | 11 +++++++++++
  include/linux/arch_topology.h     |  4 ++--
  include/linux/sched/topology.h    |  7 +++++++
  kernel/sched/core.c               | 11 -----------
  6 files changed, 24 insertions(+), 15 deletions(-)
arch/arm/include/asm/topology.h +2 -1
index 435aba289fc5..e0593cf095d0 100644
@@ -16,8 +16,9 @@
  /* Enable topology flag updates */
  #define arch_update_cpu_topology topology_update_cpu_topology
 
-/* Replace task scheduler's default thermal pressure retrieve API */
+/* Replace task scheduler's default thermal pressure API */
  #define arch_scale_thermal_pressure topology_get_thermal_pressure
+#define arch_set_thermal_pressure   topology_set_thermal_pressure
 
  #else
 
arch/arm64/include/asm/topology.h +2 -1
index 0cc835ddfcd1..e042f6527981 100644
@@ -34,8 +34,9 @@ void topology_scale_freq_tick(void);
  /* Enable topology flag updates */
  #define arch_update_cpu_topology topology_update_cpu_topology
 
-/* Replace task scheduler's default thermal pressure retrieve API */
+/* Replace task scheduler's default thermal pressure API */
  #define arch_scale_thermal_pressure topology_get_thermal_pressure
+#define arch_set_thermal_pressure   topology_set_thermal_pressure
 
  #include <asm-generic/topology.h>
 
drivers/base/arch_topology.c +11 -0
index 4d0a0038b476..d14cab7dfa3c 100644
@@ -54,6 +54,17 @@ void topology_set_cpu_scale(unsigned int cpu, unsigned long capacity)
  	per_cpu(cpu_scale, cpu) = capacity;
  }
 
+DEFINE_PER_CPU(unsigned long, thermal_pressure);
+
+void arch_set_thermal_pressure(const struct cpumask *cpus,
+			       unsigned long th_pressure)
+{
+	int cpu;
+
+	for_each_cpu(cpu, cpus)
+		WRITE_ONCE(per_cpu(thermal_pressure, cpu), th_pressure);
+}
+
  static ssize_t cpu_capacity_show(struct device *dev,
  				 struct device_attribute *attr,
  				 char *buf)
include/linux/arch_topology.h +2 -2
index 0566cb3314ef..69b1dabe39dc 100644
@@ -39,8 +39,8 @@ static inline unsigned long topology_get_thermal_pressure(int cpu)
  	return per_cpu(thermal_pressure, cpu);
  }
 
-void arch_set_thermal_pressure(struct cpumask *cpus,
-			       unsigned long th_pressure);
+void topology_set_thermal_pressure(const struct cpumask *cpus,
+				   unsigned long th_pressure);
 
  struct cpu_topology {
  	int thread_id;
include/linux/sched/topology.h +7 -0
index fb11091129b3..764222d637b7 100644
@@ -232,6 +232,13 @@ unsigned long arch_scale_thermal_pressure(int cpu)
  }
  #endif
 
+#ifndef arch_set_thermal_pressure
+static __always_inline
+void arch_set_thermal_pressure(const struct cpumask *cpus,
+			       unsigned long th_pressure)
+{ }
+#endif
+
  static inline int task_node(const struct task_struct *p)
  {
  	return cpu_to_node(task_cpu(p));
kernel/sched/core.c +0 -11
index ff0519551188..90b44f3840e4 100644
@@ -3731,17 +3731,6 @@ unsigned long long task_sched_runtime(struct task_struct *p)
  	return ns;
  }
 
-DEFINE_PER_CPU(unsigned long, thermal_pressure);
-
-void arch_set_thermal_pressure(struct cpumask *cpus,
-			       unsigned long th_pressure)
-{
-	int cpu;
-
-	for_each_cpu(cpu, cpus)
-		WRITE_ONCE(per_cpu(thermal_pressure, cpu), th_pressure);
-}
-
  /*
   * This function gets called by the timer code, with HZ frequency.
   * We call it with interrupts disabled.
Signature · 2 lines
Warm Regards
Thara
#3 lore raw

[tip: sched/core] arch_topology, sched/core: Cleanup thermal pressure definition

In reply to Valentin Schneider
The following commit has been merged into the sched/core branch of tip:

The following commit has been merged into the sched/core branch of tip:

Commit-ID:     25980c7a79af42f2daa73e2f475ebf4cbac8253e
Gitweb:        https://git.kernel.org/tip/25980c7a79af42f2daa73e2f475ebf4cbac8253e
Author:        Valentin Schneider <[email protected]>
AuthorDate:    Sun, 12 Jul 2020 17:59:15 +01:00
Committer:     Peter Zijlstra <[email protected]>
CommitterDate: Wed, 22 Jul 2020 10:22:05 +02:00

arch_topology, sched/core: Cleanup thermal pressure definition

The following commit:

14533a16c46d ("thermal/cpu-cooling, sched/core: Move the arch_set_thermal_pressure() API to generic scheduler code")

moved the definition of arch_set_thermal_pressure() to sched/core.c, but kept its declaration in linux/arch_topology.h. When building e.g. an x86 kernel with CONFIG_SCHED_THERMAL_PRESSURE=y, cpufreq_cooling.c ends up getting the declaration of arch_set_thermal_pressure() from include/linux/arch_topology.h, which is somewhat awkward.

On top of this, sched/core.c unconditionally defines
o The thermal_pressure percpu variable
o arch_set_thermal_pressure()

while arch_scale_thermal_pressure() does nothing unless redefined by the architecture.

arch_*() functions are meant to be defined by architectures, so revert the aforementioned commit and re-implement it in a way that keeps arch_set_thermal_pressure() architecture-definable, and doesn't define the thermal pressure percpu variable for kernels that don't need it (CONFIG_SCHED_THERMAL_PRESSURE=n).

---
 arch/arm/include/asm/topology.h   |  3 ++-
 arch/arm64/include/asm/topology.h |  3 ++-
 drivers/base/arch_topology.c      | 11 +++++++++++
 include/linux/arch_topology.h     |  4 ++--
 include/linux/sched/topology.h    |  7 +++++++
 kernel/sched/core.c               | 11 -----------
 6 files changed, 24 insertions(+), 15 deletions(-)
arch/arm/include/asm/topology.h +2 -1
index 435aba2..e0593cf 100644
@@ -16,8 +16,9 @@
 /* Enable topology flag updates */
 #define arch_update_cpu_topology topology_update_cpu_topology
 
-/* Replace task scheduler's default thermal pressure retrieve API */
+/* Replace task scheduler's default thermal pressure API */
 #define arch_scale_thermal_pressure topology_get_thermal_pressure
+#define arch_set_thermal_pressure   topology_set_thermal_pressure
 
 #else
 
arch/arm64/include/asm/topology.h +2 -1
index 0cc835d..e042f65 100644
@@ -34,8 +34,9 @@ void topology_scale_freq_tick(void);
 /* Enable topology flag updates */
 #define arch_update_cpu_topology topology_update_cpu_topology
 
-/* Replace task scheduler's default thermal pressure retrieve API */
+/* Replace task scheduler's default thermal pressure API */
 #define arch_scale_thermal_pressure topology_get_thermal_pressure
+#define arch_set_thermal_pressure   topology_set_thermal_pressure
 
 #include <asm-generic/topology.h>
 
drivers/base/arch_topology.c +11 -0
index 4d0a003..75f72d6 100644
@@ -54,6 +54,17 @@ void topology_set_cpu_scale(unsigned int cpu, unsigned long capacity)
 	per_cpu(cpu_scale, cpu) = capacity;
 }
 
+DEFINE_PER_CPU(unsigned long, thermal_pressure);
+
+void topology_set_thermal_pressure(const struct cpumask *cpus,
+			       unsigned long th_pressure)
+{
+	int cpu;
+
+	for_each_cpu(cpu, cpus)
+		WRITE_ONCE(per_cpu(thermal_pressure, cpu), th_pressure);
+}
+
 static ssize_t cpu_capacity_show(struct device *dev,
 				 struct device_attribute *attr,
 				 char *buf)
include/linux/arch_topology.h +2 -2
index 0566cb3..69b1dab 100644
@@ -39,8 +39,8 @@ static inline unsigned long topology_get_thermal_pressure(int cpu)
 	return per_cpu(thermal_pressure, cpu);
 }
 
-void arch_set_thermal_pressure(struct cpumask *cpus,
-			       unsigned long th_pressure);
+void topology_set_thermal_pressure(const struct cpumask *cpus,
+				   unsigned long th_pressure);
 
 struct cpu_topology {
 	int thread_id;
include/linux/sched/topology.h +7 -0
index fb11091..764222d 100644
@@ -232,6 +232,13 @@ unsigned long arch_scale_thermal_pressure(int cpu)
 }
 #endif
 
+#ifndef arch_set_thermal_pressure
+static __always_inline
+void arch_set_thermal_pressure(const struct cpumask *cpus,
+			       unsigned long th_pressure)
+{ }
+#endif
+
 static inline int task_node(const struct task_struct *p)
 {
 	return cpu_to_node(task_cpu(p));
kernel/sched/core.c +0 -11
index 12db8fb..bd8e521 100644
@@ -3869,17 +3869,6 @@ unsigned long long task_sched_runtime(struct task_struct *p)
 	return ns;
 }
 
-DEFINE_PER_CPU(unsigned long, thermal_pressure);
-
-void arch_set_thermal_pressure(struct cpumask *cpus,
-			       unsigned long th_pressure)
-{
-	int cpu;
-
-	for_each_cpu(cpu, cpus)
-		WRITE_ONCE(per_cpu(thermal_pressure, cpu), th_pressure);
-}
-
 /*
  * This function gets called by the timer code, with HZ frequency.
  * We call it with interrupts disabled.
#4 lore raw

PATCH v2 2/3 sched: Cleanup SCHED_THERMAL_PRESSURE kconfig entry

In reply to Valentin Schneider

As Russell pointed out [1], this option is severely lacking in the documentation department, and figuring out if one has the required dependencies to benefit from turning it on is not straightforward.

Make it non user-visible, and add a bit of help to it. While at it, make it depend on CPU_FREQ_THERMAL.

[1]: https://lkml.kernel.org/r/[email protected]

---
 init/Kconfig | 15 ++++++++++++++-
 1 file changed, 14 insertions(+), 1 deletion(-)
init/Kconfig +14 -1
index 0498af567f70..0a97d85568b2 100644
@@ -492,8 +492,21 @@ config HAVE_SCHED_AVG_IRQ
 	depends on SMP
 
 config SCHED_THERMAL_PRESSURE
-	bool "Enable periodic averaging of thermal pressure"
+	bool
 	depends on SMP
+	depends on CPU_FREQ_THERMAL
+	help
+	  Select this option to enable thermal pressure accounting in the
+	  scheduler. Thermal pressure is the value conveyed to the scheduler
+	  that reflects the reduction in CPU compute capacity resulted from
+	  thermal throttling. Thermal throttling occurs when the performance of
+	  a CPU is capped due to high operating temperatures.
+
+	  If selected, the scheduler will be able to balance tasks accordingly,
+	  i.e. put less load on throttled CPUs than on non/less throttled ones.
+
+	  This requires the architecture to implement
+	  arch_set_thermal_pressure() and arch_get_thermal_pressure().
 
 config BSD_PROCESS_ACCT
 	bool "BSD Process Accounting"
Signature · 1 line
2.27.0
#5 lore raw

[tip: sched/core] sched: Cleanup SCHED_THERMAL_PRESSURE kconfig entry

The following commit has been merged into the sched/core branch of tip:

The following commit has been merged into the sched/core branch of tip:

Commit-ID:     98eb401d09301d8a38c31cc8851ba95ac9385c8f
Gitweb:        https://git.kernel.org/tip/98eb401d09301d8a38c31cc8851ba95ac9385c8f
Author:        Valentin Schneider <[email protected]>
AuthorDate:    Sun, 12 Jul 2020 17:59:16 +01:00
Committer:     Peter Zijlstra <[email protected]>
CommitterDate: Wed, 22 Jul 2020 10:22:06 +02:00

sched: Cleanup SCHED_THERMAL_PRESSURE kconfig entry

As Russell pointed out [1], this option is severely lacking in the documentation department, and figuring out if one has the required dependencies to benefit from turning it on is not straightforward.

Make it non user-visible, and add a bit of help to it. While at it, make it depend on CPU_FREQ_THERMAL.

[1]: https://lkml.kernel.org/r/[email protected]

---
 init/Kconfig | 15 ++++++++++++++-
 1 file changed, 14 insertions(+), 1 deletion(-)
init/Kconfig +14 -1
index 0498af5..0a97d85 100644
@@ -492,8 +492,21 @@ config HAVE_SCHED_AVG_IRQ
 	depends on SMP
 
 config SCHED_THERMAL_PRESSURE
-	bool "Enable periodic averaging of thermal pressure"
+	bool
 	depends on SMP
+	depends on CPU_FREQ_THERMAL
+	help
+	  Select this option to enable thermal pressure accounting in the
+	  scheduler. Thermal pressure is the value conveyed to the scheduler
+	  that reflects the reduction in CPU compute capacity resulted from
+	  thermal throttling. Thermal throttling occurs when the performance of
+	  a CPU is capped due to high operating temperatures.
+
+	  If selected, the scheduler will be able to balance tasks accordingly,
+	  i.e. put less load on throttled CPUs than on non/less throttled ones.
+
+	  This requires the architecture to implement
+	  arch_set_thermal_pressure() and arch_get_thermal_pressure().
 
 config BSD_PROCESS_ACCT
 	bool "BSD Process Accounting"
#6 lore raw

Re: [PATCH v2 2/3] sched: Cleanup SCHED_THERMAL_PRESSURE kconfig entry

In reply to Valentin Schneider
Quoted 32 lines · Valentin Schneider

As Russell pointed out [1], this option is severely lacking in the documentation department, and figuring out if one has the required dependencies to benefit from turning it on is not straightforward.

Make it non user-visible, and add a bit of help to it. While at it, make it depend on CPU_FREQ_THERMAL.

[1]: https://lkml.kernel.org/r/[email protected]

---
 init/Kconfig | 15 ++++++++++++++-
 1 file changed, 14 insertions(+), 1 deletion(-)
init/Kconfig +7 -1
index 0498af567f70..0a97d85568b2 100644
@@ -492,8 +492,21 @@ config HAVE_SCHED_AVG_IRQ
 	depends on SMP
 
 config SCHED_THERMAL_PRESSURE
-	bool "Enable periodic averaging of thermal pressure"
+	bool
 	depends on SMP
+	depends on CPU_FREQ_THERMAL
+	help
+	  Select this option to enable thermal pressure accounting in the
+	  scheduler. Thermal pressure is the value conveyed to the scheduler
+	  that reflects the reduction in CPU compute capacity resulted from
+	  thermal throttling. Thermal throttling occurs when the performance of

init/Kconfig config HAVE_SCHED_AVG_IRQ · line 512

Quoted 10 lines
init/Kconfig +7 -0
+	  a CPU is capped due to high operating temperatures.
+
+	  If selected, the scheduler will be able to balance tasks accordingly,
+	  i.e. put less load on throttled CPUs than on non/less throttled ones.
+
+	  This requires the architecture to implement
+	  arch_set_thermal_pressure() and arch_get_thermal_pressure().
 
 config BSD_PROCESS_ACCT
 	bool "BSD Process Accounting"

On arm64 linux-next (20200727),

https://gitlab.com/cailca/linux-mm/-/blob/master/arm64.config

WARNING: unmet direct dependencies detected for SCHED_THERMAL_PRESSURE
Depends on [n]: SMP [=y] && CPU_FREQ_THERMAL [=n] Selected by [y]:
- ARM64 [=y]

_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

#7 lore raw

Re: [PATCH v2 2/3] sched: Cleanup SCHED_THERMAL_PRESSURE kconfig entry

Quoted 46 lines · Qian Cai
Quoted 32 lines · Valentin Schneider

As Russell pointed out [1], this option is severely lacking in the documentation department, and figuring out if one has the required dependencies to benefit from turning it on is not straightforward.

Make it non user-visible, and add a bit of help to it. While at it, make it depend on CPU_FREQ_THERMAL.

[1]: https://lkml.kernel.org/r/[email protected]

---
 init/Kconfig | 15 ++++++++++++++-
 1 file changed, 14 insertions(+), 1 deletion(-)
init/Kconfig +7 -1
index 0498af567f70..0a97d85568b2 100644
@@ -492,8 +492,21 @@ config HAVE_SCHED_AVG_IRQ
 	depends on SMP
 
 config SCHED_THERMAL_PRESSURE
-	bool "Enable periodic averaging of thermal pressure"
+	bool
 	depends on SMP
+	depends on CPU_FREQ_THERMAL
+	help
+	  Select this option to enable thermal pressure accounting in the
+	  scheduler. Thermal pressure is the value conveyed to the scheduler
+	  that reflects the reduction in CPU compute capacity resulted from
+	  thermal throttling. Thermal throttling occurs when the performance of

init/Kconfig config HAVE_SCHED_AVG_IRQ · line 512

Quoted 10 lines
init/Kconfig +7 -0
+	  a CPU is capped due to high operating temperatures.
+
+	  If selected, the scheduler will be able to balance tasks accordingly,
+	  i.e. put less load on throttled CPUs than on non/less throttled ones.
+
+	  This requires the architecture to implement
+	  arch_set_thermal_pressure() and arch_get_thermal_pressure().
 
 config BSD_PROCESS_ACCT
 	bool "BSD Process Accounting"

On arm64 linux-next (20200727),

Quoted 6 lines

https://gitlab.com/cailca/linux-mm/-/blob/master/arm64.config

WARNING: unmet direct dependencies detected for SCHED_THERMAL_PRESSURE
Depends on [n]: SMP [=y] && CPU_FREQ_THERMAL [=n] Selected by [y]:
- ARM64 [=y]

Not sure, but:

(1) do we wan to let people enable SCHED_THERMAL_PRESSURE for arm64 so arm64 can potentially run w/o a CPU freq cooling device?

arch/arm64/Kconfig +0 -1
index 2d4abbc9f8d0..baffe8b66da2 100644
@@ -192,7 +192,6 @@ config ARM64
        select PCI_SYSCALL if PCI
        select POWER_RESET
        select POWER_SUPPLY
-       select SCHED_THERMAL_PRESSURE
        select SPARSE_IRQ
        select SWIOTLB
        select SYSCTL_EXCEPTION_TRACE
init/Kconfig +1 -1
index 37b089f87804..8b36e07fb230 100644
@@ -492,7 +492,7 @@ config HAVE_SCHED_AVG_IRQ
        depends on SMP
 
 config SCHED_THERMAL_PRESSURE
-       bool
+       bool "Thermal pressure accounting"
        depends on SMP
        depends on CPU_FREQ_THERMAL
        help

Or

(2) should SCHED_THERMAL_PRESSURE for arm64 be enabled by default?

But then it makes no sense to allow the removal of CONFIG_CPU_FREQ_THERMAL.

linux-next/master$ make ARCH=arm64 defconfig

// Remove CONFIG_CPU_FREQ_THERMAL
linux-next/master$ grep CPU_FREQ_THERMAL .config
# CONFIG_CPU_FREQ_THERMAL is not set

linux-next/master$ make
scripts/kconfig/conf --syncconfig Kconfig

WARNING: unmet direct dependencies detected for SCHED_THERMAL_PRESSURE
Depends on [n]: SMP [=y] && CPU_FREQ_THERMAL [=n] Selected by [y]:
- ARM64 [=y]

WARNING: unmet direct dependencies detected for SCHED_THERMAL_PRESSURE
Depends on [n]: SMP [=y] && CPU_FREQ_THERMAL [=n] Selected by [y]:
- ARM64 [=y]

WARNING: unmet direct dependencies detected for SCHED_THERMAL_PRESSURE
Depends on [n]: SMP [=y] && CPU_FREQ_THERMAL [=n] Selected by [y]:
- ARM64 [=y]
HOSTCC scripts/dtc/dtc.o

---

There is a similar issue with arm.

I would prefer for (1).

#8 lore raw

Re: [PATCH v2 2/3] sched: Cleanup SCHED_THERMAL_PRESSURE kconfig entry

Hi,

Quoted 116 lines · Dietmar Eggemann
Quoted 46 lines · Qian Cai
Quoted 32 lines · Valentin Schneider

As Russell pointed out [1], this option is severely lacking in the documentation department, and figuring out if one has the required dependencies to benefit from turning it on is not straightforward.

Make it non user-visible, and add a bit of help to it. While at it, make it depend on CPU_FREQ_THERMAL.

[1]: https://lkml.kernel.org/r/[email protected]

---
 init/Kconfig | 15 ++++++++++++++-
 1 file changed, 14 insertions(+), 1 deletion(-)
init/Kconfig +8 -1
index 0498af567f70..0a97d85568b2 100644
@@ -492,8 +492,21 @@ config HAVE_SCHED_AVG_IRQ
    depends on SMP
 
 config SCHED_THERMAL_PRESSURE
-	bool "Enable periodic averaging of thermal pressure"
+	bool
    depends on SMP
+	depends on CPU_FREQ_THERMAL
+	help
+	  Select this option to enable thermal pressure accounting in the
+	  scheduler. Thermal pressure is the value conveyed to the scheduler
+	  that reflects the reduction in CPU compute capacity resulted from
+	  thermal throttling. Thermal throttling occurs when the performance of
+	  a CPU is capped due to high operating temperatures.

init/Kconfig config HAVE_SCHED_AVG_IRQ

Quoted 10 lines
init/Kconfig +6 -1
+
+	  If selected, the scheduler will be able to balance tasks accordingly,
+	  i.e. put less load on throttled CPUs than on non/less throttled ones.
+
+	  This requires the architecture to implement
+	  arch_set_thermal_pressure() and arch_get_thermal_pressure().
 
 config BSD_PROCESS_ACCT
    bool "BSD Process Accounting"
--

On arm64 linux-next (20200727),

Quoted 6 lines

https://gitlab.com/cailca/linux-mm/-/blob/master/arm64.config

WARNING: unmet direct dependencies detected for SCHED_THERMAL_PRESSURE
Depends on [n]: SMP [=y] && CPU_FREQ_THERMAL [=n] Selected by [y]:
- ARM64 [=y]

Not sure, but:

(1) do we wan to let people enable SCHED_THERMAL_PRESSURE for arm64 so arm64 can potentially run w/o a CPU freq cooling device?

arch/arm64/Kconfig +0 -1
index 2d4abbc9f8d0..baffe8b66da2 100644
@@ -192,7 +192,6 @@ config ARM64
        select PCI_SYSCALL if PCI
        select POWER_RESET
        select POWER_SUPPLY
-       select SCHED_THERMAL_PRESSURE
        select SPARSE_IRQ
        select SWIOTLB
        select SYSCTL_EXCEPTION_TRACE
init/Kconfig +1 -1
index 37b089f87804..8b36e07fb230 100644
@@ -492,7 +492,7 @@ config HAVE_SCHED_AVG_IRQ
        depends on SMP
 
 config SCHED_THERMAL_PRESSURE
-       bool
+       bool "Thermal pressure accounting"
        depends on SMP
        depends on CPU_FREQ_THERMAL
        help

Or

(2) should SCHED_THERMAL_PRESSURE for arm64 be enabled by default?

But then it makes no sense to allow the removal of CONFIG_CPU_FREQ_THERMAL.

linux-next/master$ make ARCH=arm64 defconfig

// Remove CONFIG_CPU_FREQ_THERMAL
linux-next/master$ grep CPU_FREQ_THERMAL .config
# CONFIG_CPU_FREQ_THERMAL is not set

linux-next/master$ make
scripts/kconfig/conf --syncconfig Kconfig

WARNING: unmet direct dependencies detected for SCHED_THERMAL_PRESSURE
Depends on [n]: SMP [=y] && CPU_FREQ_THERMAL [=n] Selected by [y]:
- ARM64 [=y]

WARNING: unmet direct dependencies detected for SCHED_THERMAL_PRESSURE
Depends on [n]: SMP [=y] && CPU_FREQ_THERMAL [=n] Selected by [y]:
- ARM64 [=y]

WARNING: unmet direct dependencies detected for SCHED_THERMAL_PRESSURE
Depends on [n]: SMP [=y] && CPU_FREQ_THERMAL [=n] Selected by [y]:
- ARM64 [=y]
HOSTCC scripts/dtc/dtc.o

Quoted 5 lines

---

There is a similar issue with arm.

I would prefer for (1).

I went for having SCHED_THERMAL_PRESSURE in arm64/Kconfig because of where the discussion went in the original thread ([1] in the changelog).

One point is that selecting this option requires having the right infrastructure in place (arch_{set, scale}_thermal_pressure() must be redefined by the architecture), which cannot be easily expressed in Kconfig terms. Russell's point was that this is difficult for a lambda user to make sense of, and Vincent argued that this option should simply be selected at architecture level, which, given the context, makes sense IMO.

We could change the arch Kconfig into

select SCHED_THERMAL_PRESSURE if CPU_FREQ_THERMAL

but that seems redundant; this dependency is already expressed in SCHED_THERMAL_PRESSURE's definition. Is there a proper pattern to select some Kconfig option only if all of its dependencies are met?

#9 lore raw

Re: [PATCH v2 2/3] sched: Cleanup SCHED_THERMAL_PRESSURE kconfig entry

Quoted 6 lines · Valentin Schneider

Hi,

Quoted 2 lines · Dietmar Eggemann
Quoted 1 line · Qian Cai

On Sun, Jul 12, 2020 at 05:59:16PM +0100, Valentin Schneider wrote:

[...]

Quoted 10 lines

I went for having SCHED_THERMAL_PRESSURE in arm64/Kconfig because of where the discussion went in the original thread ([1] in the changelog).

One point is that selecting this option requires having the right infrastructure in place (arch_{set, scale}_thermal_pressure() must be redefined by the architecture), which cannot be easily expressed in Kconfig terms. Russell's point was that this is difficult for a lambda user to make sense of, and Vincent argued that this option should simply be selected at architecture level, which, given the context, makes sense IMO.

Quoted 7 lines

We could change the arch Kconfig into

select SCHED_THERMAL_PRESSURE if CPU_FREQ_THERMAL

but that seems redundant; this dependency is already expressed in SCHED_THERMAL_PRESSURE's definition. Is there a proper pattern to select some Kconfig option only if all of its dependencies are met?

The warning when disabling CPU_FREQ_THERMAL after make defconfig disappears, so this should be OK.

arch/arm/Kconfig +1 -1
index 939c4d6bbc2e..a677e71b3d5f 100644
@@ -46,7 +46,7 @@ config ARM
        select EDAC_ATOMIC_SCRUB
        select GENERIC_ALLOCATOR
        select GENERIC_ARCH_TOPOLOGY if ARM_CPU_TOPOLOGY
-       select SCHED_THERMAL_PRESSURE if ARM_CPU_TOPOLOGY
+       select SCHED_THERMAL_PRESSURE if ARM_CPU_TOPOLOGY && CPU_FREQ_THERMAL
        select GENERIC_ATOMIC64 if CPU_V7M || CPU_V6 || !CPU_32v6K || !AEABI
        select GENERIC_CLOCKEVENTS_BROADCAST if SMP
        select GENERIC_CPU_AUTOPROBE
arch/arm64/Kconfig +1 -1
index c403e6f5db86..59ae16f8b941 100644
@@ -192,7 +192,7 @@ config ARM64
        select PCI_SYSCALL if PCI
        select POWER_RESET
        select POWER_SUPPLY
-       select SCHED_THERMAL_PRESSURE
+       select SCHED_THERMAL_PRESSURE if CPU_FREQ_THERMAL
        select SPARSE_IRQ
        select SWIOTLB
        select SYSCTL_EXCEPTION_TRAC
#10 lore raw

Re: [PATCH v2 2/3] sched: Cleanup SCHED_THERMAL_PRESSURE kconfig entry

In reply to Valentin Schneider
Quoted 7 lines · Valentin Schneider

We could change the arch Kconfig into

select SCHED_THERMAL_PRESSURE if CPU_FREQ_THERMAL

but that seems redundant; this dependency is already expressed in SCHED_THERMAL_PRESSURE's definition. Is there a proper pattern to select some Kconfig option only if all of its dependencies are met?

How about something like this (totally untested):

---8<---

arch/arm/Kconfig +0 -1
index 939c4d6bbc2e..2ac74904a3ce 100644
@@ -46,7 +46,6 @@ config ARM
 	select EDAC_ATOMIC_SCRUB
 	select GENERIC_ALLOCATOR
 	select GENERIC_ARCH_TOPOLOGY if ARM_CPU_TOPOLOGY
-	select SCHED_THERMAL_PRESSURE if ARM_CPU_TOPOLOGY
 	select GENERIC_ATOMIC64 if CPU_V7M || CPU_V6 || !CPU_32v6K || !AEABI
 	select GENERIC_CLOCKEVENTS_BROADCAST if SMP
 	select GENERIC_CPU_AUTOPROBE
arch/arm64/Kconfig +0 -1
index c403e6f5db86..66dc41fd49f2 100644
@@ -192,7 +192,6 @@ config ARM64
 	select PCI_SYSCALL if PCI
 	select POWER_RESET
 	select POWER_SUPPLY
-	select SCHED_THERMAL_PRESSURE
 	select SPARSE_IRQ
 	select SWIOTLB
 	select SYSCTL_EXCEPTION_TRACE
init/Kconfig +1 -0
index 0a97d85568b2..c2e1f3ac527e 100644
@@ -493,6 +493,7 @@ config HAVE_SCHED_AVG_IRQ
 
 config SCHED_THERMAL_PRESSURE
 	bool
+	default y if ARM64 || ARM_CPU_TOPOLOGY
 	depends on SMP
 	depends on CPU_FREQ_THERMAL
 	help

--->8---

Thanks,
Quentin

#11 lore raw

Re: [PATCH v2 2/3] sched: Cleanup SCHED_THERMAL_PRESSURE kconfig entry

Quoted 42 lines · Quentin Perret
Quoted 7 lines · Valentin Schneider

We could change the arch Kconfig into

select SCHED_THERMAL_PRESSURE if CPU_FREQ_THERMAL

but that seems redundant; this dependency is already expressed in SCHED_THERMAL_PRESSURE's definition. Is there a proper pattern to select some Kconfig option only if all of its dependencies are met?

How about something like this (totally untested):

---8<---

arch/arm/Kconfig +0 -1
index 939c4d6bbc2e..2ac74904a3ce 100644
@@ -46,7 +46,6 @@ config ARM
      select EDAC_ATOMIC_SCRUB
      select GENERIC_ALLOCATOR
      select GENERIC_ARCH_TOPOLOGY if ARM_CPU_TOPOLOGY
-	select SCHED_THERMAL_PRESSURE if ARM_CPU_TOPOLOGY
      select GENERIC_ATOMIC64 if CPU_V7M || CPU_V6 || !CPU_32v6K || !AEABI
      select GENERIC_CLOCKEVENTS_BROADCAST if SMP
      select GENERIC_CPU_AUTOPROBE
arch/arm64/Kconfig +0 -1
index c403e6f5db86..66dc41fd49f2 100644
@@ -192,7 +192,6 @@ config ARM64
      select PCI_SYSCALL if PCI
      select POWER_RESET
      select POWER_SUPPLY
-	select SCHED_THERMAL_PRESSURE
      select SPARSE_IRQ
      select SWIOTLB
      select SYSCTL_EXCEPTION_TRACE
init/Kconfig +0 -0
index 0a97d85568b2..c2e1f3ac527e 100644

init/Kconfig config HAVE_SCHED_AVG_IRQ

Quoted 8 lines
init/Kconfig +1 -1
@@ -493,6 +493,7 @@ config HAVE_SCHED_AVG_IRQ
 
 config SCHED_THERMAL_PRESSURE
      bool
+	default y if ARM64 || ARM_CPU_TOPOLOGY
      depends on SMP
      depends on CPU_FREQ_THERMAL
      help
--->8---

That does seem to do just what I was looking for, thanks!

Quoted 2 lines

Thanks,
Quentin

#12 lore raw

PATCH v2 3/3 arm, arm64: Select CONFIG_SCHED_THERMAL_PRESSURE

In reply to Valentin Schneider

This option now correctly depends on CPU_FREQ_THERMAL, so select it on the architectures that implement the required functions, arch_set_thermal_pressure() and arch_get_thermal_pressure().

---
 arch/arm/Kconfig   | 1 +
 arch/arm64/Kconfig | 1 +
 2 files changed, 2 insertions(+)
arch/arm/Kconfig +1 -0
index 2ac74904a3ce..939c4d6bbc2e 100644
@@ -46,6 +46,7 @@ config ARM
 	select EDAC_ATOMIC_SCRUB
 	select GENERIC_ALLOCATOR
 	select GENERIC_ARCH_TOPOLOGY if ARM_CPU_TOPOLOGY
+	select SCHED_THERMAL_PRESSURE if ARM_CPU_TOPOLOGY
 	select GENERIC_ATOMIC64 if CPU_V7M || CPU_V6 || !CPU_32v6K || !AEABI
 	select GENERIC_CLOCKEVENTS_BROADCAST if SMP
 	select GENERIC_CPU_AUTOPROBE
arch/arm64/Kconfig +1 -0
index 66dc41fd49f2..96d478fb7a2e 100644
@@ -100,6 +100,7 @@ config ARM64
 	select FRAME_POINTER
 	select GENERIC_ALLOCATOR
 	select GENERIC_ARCH_TOPOLOGY
+	select SCHED_THERMAL_PRESSURE
 	select GENERIC_CLOCKEVENTS
 	select GENERIC_CLOCKEVENTS_BROADCAST
 	select GENERIC_CPU_AUTOPROBE
Signature · 1 line
2.27.0
maintainer Acked-by on v2 3/3
#13 lore raw

Re: [PATCH v2 3/3] arm, arm64: Select CONFIG_SCHED_THERMAL_PRESSURE

arch/arm64/Kconfig config ARM64 · line 106

Quoted 12 lines · Valentin Schneider
arch/arm64/Kconfig +1 -0
index 66dc41fd49f2..96d478fb7a2e 100644
@@ -100,6 +100,7 @@ config ARM64
 	select FRAME_POINTER
 	select GENERIC_ALLOCATOR
 	select GENERIC_ARCH_TOPOLOGY
+	select SCHED_THERMAL_PRESSURE
 	select GENERIC_CLOCKEVENTS
 	select GENERIC_CLOCKEVENTS_BROADCAST
 	select GENERIC_CPU_AUTOPROBE

We tend to keep these in alphabetical order. Otherwise,

#14 lore raw

Re: [PATCH v2 3/3] arm, arm64: Select CONFIG_SCHED_THERMAL_PRESSURE

Quoted 16 lines · Catalin Marinas

arch/arm64/Kconfig config ARM64 · line 106

Quoted 12 lines · Valentin Schneider
arch/arm64/Kconfig +1 -0
index 66dc41fd49f2..96d478fb7a2e 100644
@@ -100,6 +100,7 @@ config ARM64
     select FRAME_POINTER
     select GENERIC_ALLOCATOR
     select GENERIC_ARCH_TOPOLOGY
+	select SCHED_THERMAL_PRESSURE
     select GENERIC_CLOCKEVENTS
     select GENERIC_CLOCKEVENTS_BROADCAST
     select GENERIC_CPU_AUTOPROBE

We tend to keep these in alphabetical order. Otherwise,

Quoted 1 line

Thanks! Will reorder that in v3.

#15 lore raw

[tip: sched/core] arm, arm64: Select CONFIG_SCHED_THERMAL_PRESSURE

In reply to Valentin Schneider
The following commit has been merged into the sched/core branch of tip:

The following commit has been merged into the sched/core branch of tip:

Commit-ID:     e17ae7fea8714caa743aa0d5e446a25a999ad726
Gitweb:        https://git.kernel.org/tip/e17ae7fea8714caa743aa0d5e446a25a999ad726
Author:        Valentin Schneider <[email protected]>
AuthorDate:    Sun, 12 Jul 2020 17:59:17 +01:00
Committer:     Peter Zijlstra <[email protected]>
CommitterDate: Wed, 22 Jul 2020 10:22:06 +02:00

arm, arm64: Select CONFIG_SCHED_THERMAL_PRESSURE

This option now correctly depends on CPU_FREQ_THERMAL, so select it on the architectures that implement the required functions, arch_set_thermal_pressure() and arch_get_thermal_pressure().

---
 arch/arm/Kconfig   | 1 +
 arch/arm64/Kconfig | 1 +
 2 files changed, 2 insertions(+)
arch/arm/Kconfig +1 -0
index 2ac7490..939c4d6 100644
@@ -46,6 +46,7 @@ config ARM
 	select EDAC_ATOMIC_SCRUB
 	select GENERIC_ALLOCATOR
 	select GENERIC_ARCH_TOPOLOGY if ARM_CPU_TOPOLOGY
+	select SCHED_THERMAL_PRESSURE if ARM_CPU_TOPOLOGY
 	select GENERIC_ATOMIC64 if CPU_V7M || CPU_V6 || !CPU_32v6K || !AEABI
 	select GENERIC_CLOCKEVENTS_BROADCAST if SMP
 	select GENERIC_CPU_AUTOPROBE
arch/arm64/Kconfig +1 -0
index 66dc41f..c403e6f 100644
@@ -192,6 +192,7 @@ config ARM64
 	select PCI_SYSCALL if PCI
 	select POWER_RESET
 	select POWER_SUPPLY
+	select SCHED_THERMAL_PRESSURE
 	select SPARSE_IRQ
 	select SWIOTLB
 	select SYSCTL_EXCEPTION_TRACE
maintainer
#16 lore raw
In reply to Valentin Schneider
Quoted 4 lines · Valentin Schneider

Valentin Schneider (3):
arch_topology, sched/core: Cleanup thermal pressure definition
sched: Cleanup SCHED_THERMAL_PRESSURE kconfig entry arm, arm64: Select CONFIG_SCHED_THERMAL_PRESSURE

Thanks!

maintainer Reviewed-by on the v2 series
#17 lore raw
In reply to Valentin Schneider
Quoted 17 lines · Valentin Schneider

Hi folks,

This stems from this thread [1] on the list. TL;DR: the thermal pressure config has no helpful documentation, and figuring out if the right dependencies are in place is not easy for a regular user.

The current landscape also paints an odd picture: arch_set_thermal_pressure() is hardcoded in sched/core.c, and is *not* architecture-definable, while arch_get_thermal_pressure() is. Patch 1 is tackling this, the rest is Kconfig stuff.

Cheers,
Valentin

[1]: https://lkml.kernel.org/r/[email protected]

Quoted 8 lines

Revisions
=========

v1 -> v2
--------

o Remove cpufreq_cooling.c weak function; use #define stub in sched/topology.h (Vincent)

Looks good to me.

Quoted 21 lines

o Hinge arm SCHED_THERMAL_PRESSURE selection on ARM_CPU_TOPOLOGY

Valentin Schneider (3):
arch_topology, sched/core: Cleanup thermal pressure definition
sched: Cleanup SCHED_THERMAL_PRESSURE kconfig entry arm, arm64: Select CONFIG_SCHED_THERMAL_PRESSURE

 arch/arm/Kconfig                  |  1 +
 arch/arm/include/asm/topology.h   |  3 ++-
 arch/arm64/Kconfig                |  1 +
 arch/arm64/include/asm/topology.h |  3 ++-
 drivers/base/arch_topology.c      | 11 +++++++++++
 include/linux/arch_topology.h     |  4 ++--
 include/linux/sched/topology.h    |  7 +++++++
 init/Kconfig                      | 15 ++++++++++++++-
 kernel/sched/core.c               | 11 -----------
 9 files changed, 40 insertions(+), 16 deletions(-)

--
2.27.0