sched/cpufreq: Rename arch_update_thermal_pressure() => arch_update_hw_pressure()
authorVincent Guittot <vincent.guittot@linaro.org>
Tue, 26 Mar 2024 09:16:15 +0000 (10:16 +0100)
committerIngo Molnar <mingo@kernel.org>
Wed, 24 Apr 2024 10:08:01 +0000 (12:08 +0200)
Now that cpufreq provides a pressure value to the scheduler, rename
arch_update_thermal_pressure into HW pressure to reflect that it returns
a pressure applied by HW (i.e. with a high frequency change) and not
always related to thermal mitigation but also generated by max current
limitation as an example. Such high frequency signal needs filtering to be
smoothed and provide an value that reflects the average available capacity
into the scheduler time scale.

Signed-off-by: Vincent Guittot <vincent.guittot@linaro.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Tested-by: Lukasz Luba <lukasz.luba@arm.com>
Reviewed-by: Qais Yousef <qyousef@layalina.io>
Reviewed-by: Lukasz Luba <lukasz.luba@arm.com>
Link: https://lore.kernel.org/r/20240326091616.3696851-5-vincent.guittot@linaro.org
15 files changed:
arch/arm/include/asm/topology.h
arch/arm64/include/asm/topology.h
drivers/base/arch_topology.c
drivers/cpufreq/qcom-cpufreq-hw.c
include/linux/arch_topology.h
include/linux/sched/topology.h
include/trace/events/hw_pressure.h [new file with mode: 0644]
include/trace/events/sched.h
include/trace/events/thermal_pressure.h [deleted file]
init/Kconfig
kernel/sched/core.c
kernel/sched/fair.c
kernel/sched/pelt.c
kernel/sched/pelt.h
kernel/sched/sched.h

index 853c4f81ba4a5761da3135090ece7e9558f581a2..ad36b6570067467f94b056d8cc533d3e683101d0 100644 (file)
@@ -22,9 +22,9 @@
 /* Enable topology flag updates */
 #define arch_update_cpu_topology topology_update_cpu_topology
 
-/* Replace task scheduler's default thermal pressure API */
-#define arch_scale_thermal_pressure topology_get_thermal_pressure
-#define arch_update_thermal_pressure   topology_update_thermal_pressure
+/* Replace task scheduler's default HW pressure API */
+#define arch_scale_hw_pressure topology_get_hw_pressure
+#define arch_update_hw_pressure        topology_update_hw_pressure
 
 #else
 
index a323b109b9c44ba5fab1f2cfe2359dd6694c3a5a..0f6ef432fb8403c9264f191e6f6102d3044a20c4 100644 (file)
@@ -35,9 +35,9 @@ void update_freq_counters_refs(void);
 /* Enable topology flag updates */
 #define arch_update_cpu_topology topology_update_cpu_topology
 
-/* Replace task scheduler's default thermal pressure API */
-#define arch_scale_thermal_pressure topology_get_thermal_pressure
-#define arch_update_thermal_pressure   topology_update_thermal_pressure
+/* Replace task scheduler's default HW pressure API */
+#define arch_scale_hw_pressure topology_get_hw_pressure
+#define arch_update_hw_pressure        topology_update_hw_pressure
 
 #include <asm-generic/topology.h>
 
index 024b78a0cfc11bbba2f0bf3c32f21d55aa101d3d..0248912ff687513e8c45a5d594384d5d529a0045 100644 (file)
@@ -22,7 +22,7 @@
 #include <linux/units.h>
 
 #define CREATE_TRACE_POINTS
-#include <trace/events/thermal_pressure.h>
+#include <trace/events/hw_pressure.h>
 
 static DEFINE_PER_CPU(struct scale_freq_data __rcu *, sft_data);
 static struct cpumask scale_freq_counters_mask;
@@ -160,26 +160,26 @@ void topology_set_cpu_scale(unsigned int cpu, unsigned long capacity)
        per_cpu(cpu_scale, cpu) = capacity;
 }
 
-DEFINE_PER_CPU(unsigned long, thermal_pressure);
+DEFINE_PER_CPU(unsigned long, hw_pressure);
 
 /**
- * topology_update_thermal_pressure() - Update thermal pressure for CPUs
+ * topology_update_hw_pressure() - Update HW pressure for CPUs
  * @cpus        : The related CPUs for which capacity has been reduced
  * @capped_freq : The maximum allowed frequency that CPUs can run at
  *
- * Update the value of thermal pressure for all @cpus in the mask. The
+ * Update the value of HW pressure for all @cpus in the mask. The
  * cpumask should include all (online+offline) affected CPUs, to avoid
  * operating on stale data when hot-plug is used for some CPUs. The
  * @capped_freq reflects the currently allowed max CPUs frequency due to
- * thermal capping. It might be also a boost frequency value, which is bigger
+ * HW capping. It might be also a boost frequency value, which is bigger
  * than the internal 'capacity_freq_ref' max frequency. In such case the
  * pressure value should simply be removed, since this is an indication that
- * there is no thermal throttling. The @capped_freq must be provided in kHz.
+ * there is no HW throttling. The @capped_freq must be provided in kHz.
  */
-void topology_update_thermal_pressure(const struct cpumask *cpus,
+void topology_update_hw_pressure(const struct cpumask *cpus,
                                      unsigned long capped_freq)
 {
-       unsigned long max_capacity, capacity, th_pressure;
+       unsigned long max_capacity, capacity, hw_pressure;
        u32 max_freq;
        int cpu;
 
@@ -189,21 +189,21 @@ void topology_update_thermal_pressure(const struct cpumask *cpus,
 
        /*
         * Handle properly the boost frequencies, which should simply clean
-        * the thermal pressure value.
+        * the HW pressure value.
         */
        if (max_freq <= capped_freq)
                capacity = max_capacity;
        else
                capacity = mult_frac(max_capacity, capped_freq, max_freq);
 
-       th_pressure = max_capacity - capacity;
+       hw_pressure = max_capacity - capacity;
 
-       trace_thermal_pressure_update(cpu, th_pressure);
+       trace_hw_pressure_update(cpu, hw_pressure);
 
        for_each_cpu(cpu, cpus)
-               WRITE_ONCE(per_cpu(thermal_pressure, cpu), th_pressure);
+               WRITE_ONCE(per_cpu(hw_pressure, cpu), hw_pressure);
 }
-EXPORT_SYMBOL_GPL(topology_update_thermal_pressure);
+EXPORT_SYMBOL_GPL(topology_update_hw_pressure);
 
 static ssize_t cpu_capacity_show(struct device *dev,
                                 struct device_attribute *attr,
index 70b0f21968a010698794e90b86e397a6e669c73a..ec8df5496a0cf8a1d820357aea3d326472c0f361 100644 (file)
@@ -347,8 +347,8 @@ static void qcom_lmh_dcvs_notify(struct qcom_cpufreq_data *data)
 
        throttled_freq = freq_hz / HZ_PER_KHZ;
 
-       /* Update thermal pressure (the boost frequencies are accepted) */
-       arch_update_thermal_pressure(policy->related_cpus, throttled_freq);
+       /* Update HW pressure (the boost frequencies are accepted) */
+       arch_update_hw_pressure(policy->related_cpus, throttled_freq);
 
        /*
         * In the unlikely case policy is unregistered do not enable
index a63d61ca55afc82ec865de9de60d8fe667964fef..b721f360d75923bbc7d4da9900337127b0bd4212 100644 (file)
@@ -60,14 +60,14 @@ 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);
+DECLARE_PER_CPU(unsigned long, hw_pressure);
 
-static inline unsigned long topology_get_thermal_pressure(int cpu)
+static inline unsigned long topology_get_hw_pressure(int cpu)
 {
-       return per_cpu(thermal_pressure, cpu);
+       return per_cpu(hw_pressure, cpu);
 }
 
-void topology_update_thermal_pressure(const struct cpumask *cpus,
+void topology_update_hw_pressure(const struct cpumask *cpus,
                                      unsigned long capped_freq);
 
 struct cpu_topology {
index c8fe9bab981b215cb3d4d1af964e5db12b26673c..4237daa5ac7a259786a29a42fa60a3b6f7a8dba6 100644 (file)
@@ -270,17 +270,17 @@ unsigned long arch_scale_cpu_capacity(int cpu)
 }
 #endif
 
-#ifndef arch_scale_thermal_pressure
+#ifndef arch_scale_hw_pressure
 static __always_inline
-unsigned long arch_scale_thermal_pressure(int cpu)
+unsigned long arch_scale_hw_pressure(int cpu)
 {
        return 0;
 }
 #endif
 
-#ifndef arch_update_thermal_pressure
+#ifndef arch_update_hw_pressure
 static __always_inline
-void arch_update_thermal_pressure(const struct cpumask *cpus,
+void arch_update_hw_pressure(const struct cpumask *cpus,
                                  unsigned long capped_frequency)
 { }
 #endif
diff --git a/include/trace/events/hw_pressure.h b/include/trace/events/hw_pressure.h
new file mode 100644 (file)
index 0000000..b9cd688
--- /dev/null
@@ -0,0 +1,29 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#undef TRACE_SYSTEM
+#define TRACE_SYSTEM hw_pressure
+
+#if !defined(_TRACE_THERMAL_PRESSURE_H) || defined(TRACE_HEADER_MULTI_READ)
+#define _TRACE_THERMAL_PRESSURE_H
+
+#include <linux/tracepoint.h>
+
+TRACE_EVENT(hw_pressure_update,
+       TP_PROTO(int cpu, unsigned long hw_pressure),
+       TP_ARGS(cpu, hw_pressure),
+
+       TP_STRUCT__entry(
+               __field(unsigned long, hw_pressure)
+               __field(int, cpu)
+       ),
+
+       TP_fast_assign(
+               __entry->hw_pressure = hw_pressure;
+               __entry->cpu = cpu;
+       ),
+
+       TP_printk("cpu=%d hw_pressure=%lu", __entry->cpu, __entry->hw_pressure)
+);
+#endif /* _TRACE_THERMAL_PRESSURE_H */
+
+/* This part must be outside protection */
+#include <trace/define_trace.h>
index dbb01b4b745108993fcf28531eb93b6aa3b68ad3..d115d64c401134e799b4878019d3a55f3fb75017 100644 (file)
@@ -752,7 +752,7 @@ DECLARE_TRACE(pelt_dl_tp,
        TP_PROTO(struct rq *rq),
        TP_ARGS(rq));
 
-DECLARE_TRACE(pelt_thermal_tp,
+DECLARE_TRACE(pelt_hw_tp,
        TP_PROTO(struct rq *rq),
        TP_ARGS(rq));
 
diff --git a/include/trace/events/thermal_pressure.h b/include/trace/events/thermal_pressure.h
deleted file mode 100644 (file)
index b686802..0000000
+++ /dev/null
@@ -1,29 +0,0 @@
-/* SPDX-License-Identifier: GPL-2.0 */
-#undef TRACE_SYSTEM
-#define TRACE_SYSTEM thermal_pressure
-
-#if !defined(_TRACE_THERMAL_PRESSURE_H) || defined(TRACE_HEADER_MULTI_READ)
-#define _TRACE_THERMAL_PRESSURE_H
-
-#include <linux/tracepoint.h>
-
-TRACE_EVENT(thermal_pressure_update,
-       TP_PROTO(int cpu, unsigned long thermal_pressure),
-       TP_ARGS(cpu, thermal_pressure),
-
-       TP_STRUCT__entry(
-               __field(unsigned long, thermal_pressure)
-               __field(int, cpu)
-       ),
-
-       TP_fast_assign(
-               __entry->thermal_pressure = thermal_pressure;
-               __entry->cpu = cpu;
-       ),
-
-       TP_printk("cpu=%d thermal_pressure=%lu", __entry->cpu, __entry->thermal_pressure)
-);
-#endif /* _TRACE_THERMAL_PRESSURE_H */
-
-/* This part must be outside protection */
-#include <trace/define_trace.h>
index aa02aec6aa7d29432f1974003900fce14bf14452..f0c9117962eca9cf2b7164a564a8b5322916b52e 100644 (file)
@@ -547,24 +547,24 @@ config HAVE_SCHED_AVG_IRQ
        depends on IRQ_TIME_ACCOUNTING || PARAVIRT_TIME_ACCOUNTING
        depends on SMP
 
-config SCHED_THERMAL_PRESSURE
+config SCHED_HW_PRESSURE
        bool
        default y if ARM && ARM_CPU_TOPOLOGY
        default y if ARM64
        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
+         Select this option to enable HW pressure accounting in the
+         scheduler. HW 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.
+         HW throttling. HW throttling occurs when the performance of
+         a CPU is capped due to high operating temperatures as an example.
 
          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_update_thermal_pressure() and arch_scale_thermal_pressure().
+         arch_update_hw_pressure() and arch_scale_thermal_pressure().
 
 config BSD_PROCESS_ACCT
        bool "BSD Process Accounting"
index 0621e4ee31dec320436b58ee2a3a5c11d5b6572a..67a8302c3131e9fe946bb6b5191c4c8b4a94aa96 100644 (file)
@@ -108,7 +108,7 @@ EXPORT_TRACEPOINT_SYMBOL_GPL(pelt_rt_tp);
 EXPORT_TRACEPOINT_SYMBOL_GPL(pelt_dl_tp);
 EXPORT_TRACEPOINT_SYMBOL_GPL(pelt_irq_tp);
 EXPORT_TRACEPOINT_SYMBOL_GPL(pelt_se_tp);
-EXPORT_TRACEPOINT_SYMBOL_GPL(pelt_thermal_tp);
+EXPORT_TRACEPOINT_SYMBOL_GPL(pelt_hw_tp);
 EXPORT_TRACEPOINT_SYMBOL_GPL(sched_cpu_capacity_tp);
 EXPORT_TRACEPOINT_SYMBOL_GPL(sched_overutilized_tp);
 EXPORT_TRACEPOINT_SYMBOL_GPL(sched_util_est_cfs_tp);
@@ -5668,7 +5668,7 @@ void sched_tick(void)
        struct rq *rq = cpu_rq(cpu);
        struct task_struct *curr = rq->curr;
        struct rq_flags rf;
-       unsigned long thermal_pressure;
+       unsigned long hw_pressure;
        u64 resched_latency;
 
        if (housekeeping_cpu(cpu, HK_TYPE_TICK))
@@ -5679,8 +5679,8 @@ void sched_tick(void)
        rq_lock(rq, &rf);
 
        update_rq_clock(rq);
-       thermal_pressure = arch_scale_thermal_pressure(cpu_of(rq));
-       update_thermal_load_avg(rq_clock_thermal(rq), rq, thermal_pressure);
+       hw_pressure = arch_scale_hw_pressure(cpu_of(rq));
+       update_hw_load_avg(rq_clock_hw(rq), rq, hw_pressure);
        curr->sched_class->task_tick(rq, curr, 0);
        if (sched_feat(LATENCY_WARN))
                resched_latency = cpu_resched_latency(rq);
index 19199c119829caa3ae2cd051b86f66440dae108d..eef39ae3efcf0711647da24908cd7524c5a01b8c 100644 (file)
@@ -78,7 +78,7 @@ static unsigned int normalized_sysctl_sched_base_slice        = 750000ULL;
 
 const_debug unsigned int sysctl_sched_migration_cost   = 500000UL;
 
-int sched_thermal_decay_shift;
+int sched_hw_decay_shift;
 static int __init setup_sched_thermal_decay_shift(char *str)
 {
        int _shift = 0;
@@ -86,7 +86,7 @@ static int __init setup_sched_thermal_decay_shift(char *str)
        if (kstrtoint(str, 0, &_shift))
                pr_warn("Unable to set scheduler thermal pressure decay shift parameter\n");
 
-       sched_thermal_decay_shift = clamp(_shift, 0, 10);
+       sched_hw_decay_shift = clamp(_shift, 0, 10);
        return 1;
 }
 __setup("sched_thermal_decay_shift=", setup_sched_thermal_decay_shift);
@@ -4969,7 +4969,7 @@ static inline unsigned long get_actual_cpu_capacity(int cpu)
 {
        unsigned long capacity = arch_scale_cpu_capacity(cpu);
 
-       capacity -= max(thermal_load_avg(cpu_rq(cpu)), cpufreq_get_pressure(cpu));
+       capacity -= max(hw_load_avg(cpu_rq(cpu)), cpufreq_get_pressure(cpu));
 
        return capacity;
 }
@@ -5002,7 +5002,7 @@ static inline int util_fits_cpu(unsigned long util,
         * Similarly if a task is capped to arch_scale_cpu_capacity(little_cpu), it
         * should fit a little cpu even if there's some pressure.
         *
-        * Only exception is for thermal pressure since it has a direct impact
+        * Only exception is for HW or cpufreq pressure since it has a direct impact
         * on available OPP of the system.
         *
         * We honour it for uclamp_min only as a drop in performance level
@@ -9324,7 +9324,7 @@ static inline bool others_have_blocked(struct rq *rq)
        if (cpu_util_dl(rq))
                return true;
 
-       if (thermal_load_avg(rq))
+       if (hw_load_avg(rq))
                return true;
 
        if (cpu_util_irq(rq))
@@ -9354,7 +9354,7 @@ static bool __update_blocked_others(struct rq *rq, bool *done)
 {
        const struct sched_class *curr_class;
        u64 now = rq_clock_pelt(rq);
-       unsigned long thermal_pressure;
+       unsigned long hw_pressure;
        bool decayed;
 
        /*
@@ -9363,11 +9363,11 @@ static bool __update_blocked_others(struct rq *rq, bool *done)
         */
        curr_class = rq->curr->sched_class;
 
-       thermal_pressure = arch_scale_thermal_pressure(cpu_of(rq));
+       hw_pressure = arch_scale_hw_pressure(cpu_of(rq));
 
        decayed = update_rt_rq_load_avg(now, rq, curr_class == &rt_sched_class) |
                  update_dl_rq_load_avg(now, rq, curr_class == &dl_sched_class) |
-                 update_thermal_load_avg(rq_clock_thermal(rq), rq, thermal_pressure) |
+                 update_hw_load_avg(rq_clock_hw(rq), rq, hw_pressure) |
                  update_irq_load_avg(rq, 0);
 
        if (others_have_blocked(rq))
index 3a96da25b67c73a0cf60131ff21d77049e46e31a..ef00382de595f84eff39a63e4f4176426ae36d4c 100644 (file)
@@ -384,30 +384,30 @@ int update_dl_rq_load_avg(u64 now, struct rq *rq, int running)
        return 0;
 }
 
-#ifdef CONFIG_SCHED_THERMAL_PRESSURE
+#ifdef CONFIG_SCHED_HW_PRESSURE
 /*
- * thermal:
+ * hardware:
  *
  *   load_sum = \Sum se->avg.load_sum but se->avg.load_sum is not tracked
  *
  *   util_avg and runnable_load_avg are not supported and meaningless.
  *
  * Unlike rt/dl utilization tracking that track time spent by a cpu
- * running a rt/dl task through util_avg, the average thermal pressure is
- * tracked through load_avg. This is because thermal pressure signal is
+ * running a rt/dl task through util_avg, the average HW pressure is
+ * tracked through load_avg. This is because HW pressure signal is
  * time weighted "delta" capacity unlike util_avg which is binary.
  * "delta capacity" =  actual capacity  -
- *                     capped capacity a cpu due to a thermal event.
+ *                     capped capacity a cpu due to a HW event.
  */
 
-int update_thermal_load_avg(u64 now, struct rq *rq, u64 capacity)
+int update_hw_load_avg(u64 now, struct rq *rq, u64 capacity)
 {
-       if (___update_load_sum(now, &rq->avg_thermal,
+       if (___update_load_sum(now, &rq->avg_hw,
                               capacity,
                               capacity,
                               capacity)) {
-               ___update_load_avg(&rq->avg_thermal, 1);
-               trace_pelt_thermal_tp(rq);
+               ___update_load_avg(&rq->avg_hw, 1);
+               trace_pelt_hw_tp(rq);
                return 1;
        }
 
index 9e1083465fbc3d2f382e1439bc4cc7c3b2efc935..2150062949d43fc9c246d75e972f60fa62bec688 100644 (file)
@@ -7,21 +7,21 @@ int __update_load_avg_cfs_rq(u64 now, struct cfs_rq *cfs_rq);
 int update_rt_rq_load_avg(u64 now, struct rq *rq, int running);
 int update_dl_rq_load_avg(u64 now, struct rq *rq, int running);
 
-#ifdef CONFIG_SCHED_THERMAL_PRESSURE
-int update_thermal_load_avg(u64 now, struct rq *rq, u64 capacity);
+#ifdef CONFIG_SCHED_HW_PRESSURE
+int update_hw_load_avg(u64 now, struct rq *rq, u64 capacity);
 
-static inline u64 thermal_load_avg(struct rq *rq)
+static inline u64 hw_load_avg(struct rq *rq)
 {
-       return READ_ONCE(rq->avg_thermal.load_avg);
+       return READ_ONCE(rq->avg_hw.load_avg);
 }
 #else
 static inline int
-update_thermal_load_avg(u64 now, struct rq *rq, u64 capacity)
+update_hw_load_avg(u64 now, struct rq *rq, u64 capacity)
 {
        return 0;
 }
 
-static inline u64 thermal_load_avg(struct rq *rq)
+static inline u64 hw_load_avg(struct rq *rq)
 {
        return 0;
 }
@@ -202,12 +202,12 @@ update_dl_rq_load_avg(u64 now, struct rq *rq, int running)
 }
 
 static inline int
-update_thermal_load_avg(u64 now, struct rq *rq, u64 capacity)
+update_hw_load_avg(u64 now, struct rq *rq, u64 capacity)
 {
        return 0;
 }
 
-static inline u64 thermal_load_avg(struct rq *rq)
+static inline u64 hw_load_avg(struct rq *rq)
 {
        return 0;
 }
index 7c39dbf31f750cb7d92f49c547abfaff5a79f894..993edb02fb0dc04f73af5480e21abf8af159e7d5 100644 (file)
@@ -1108,8 +1108,8 @@ struct rq {
 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
        struct sched_avg        avg_irq;
 #endif
-#ifdef CONFIG_SCHED_THERMAL_PRESSURE
-       struct sched_avg        avg_thermal;
+#ifdef CONFIG_SCHED_HW_PRESSURE
+       struct sched_avg        avg_hw;
 #endif
        u64                     idle_stamp;
        u64                     avg_idle;
@@ -1561,11 +1561,11 @@ static inline u64 rq_clock_task(struct rq *rq)
  *     3                       256
  *     4                       512
  */
-extern int sched_thermal_decay_shift;
+extern int sched_hw_decay_shift;
 
-static inline u64 rq_clock_thermal(struct rq *rq)
+static inline u64 rq_clock_hw(struct rq *rq)
 {
-       return rq_clock_task(rq) >> sched_thermal_decay_shift;
+       return rq_clock_task(rq) >> sched_hw_decay_shift;
 }
 
 static inline void rq_clock_skip_update(struct rq *rq)