1c66ec88fSEmmanuel Vadot========================================== 2c66ec88fSEmmanuel VadotARM CPUs capacity bindings 3c66ec88fSEmmanuel Vadot========================================== 4c66ec88fSEmmanuel Vadot 5c66ec88fSEmmanuel Vadot========================================== 6c66ec88fSEmmanuel Vadot1 - Introduction 7c66ec88fSEmmanuel Vadot========================================== 8c66ec88fSEmmanuel Vadot 9c66ec88fSEmmanuel VadotARM systems may be configured to have cpus with different power/performance 10c66ec88fSEmmanuel Vadotcharacteristics within the same chip. In this case, additional information has 11c66ec88fSEmmanuel Vadotto be made available to the kernel for it to be aware of such differences and 12c66ec88fSEmmanuel Vadottake decisions accordingly. 13c66ec88fSEmmanuel Vadot 14c66ec88fSEmmanuel Vadot========================================== 15c66ec88fSEmmanuel Vadot2 - CPU capacity definition 16c66ec88fSEmmanuel Vadot========================================== 17c66ec88fSEmmanuel Vadot 18c66ec88fSEmmanuel VadotCPU capacity is a number that provides the scheduler information about CPUs 19c66ec88fSEmmanuel Vadotheterogeneity. Such heterogeneity can come from micro-architectural differences 20c66ec88fSEmmanuel Vadot(e.g., ARM big.LITTLE systems) or maximum frequency at which CPUs can run 21c66ec88fSEmmanuel Vadot(e.g., SMP systems with multiple frequency domains). Heterogeneity in this 22c66ec88fSEmmanuel Vadotcontext is about differing performance characteristics; this binding tries to 23c66ec88fSEmmanuel Vadotcapture a first-order approximation of the relative performance of CPUs. 24c66ec88fSEmmanuel Vadot 25c66ec88fSEmmanuel VadotCPU capacities are obtained by running a suitable benchmark. This binding makes 26c66ec88fSEmmanuel Vadotno guarantees on the validity or suitability of any particular benchmark, the 27c66ec88fSEmmanuel Vadotfinal capacity should, however, be: 28c66ec88fSEmmanuel Vadot 29c66ec88fSEmmanuel Vadot* A "single-threaded" or CPU affine benchmark 30c66ec88fSEmmanuel Vadot* Divided by the running frequency of the CPU executing the benchmark 31c66ec88fSEmmanuel Vadot* Not subject to dynamic frequency scaling of the CPU 32c66ec88fSEmmanuel Vadot 33c66ec88fSEmmanuel VadotFor the time being we however advise usage of the Dhrystone benchmark. What 34c66ec88fSEmmanuel Vadotabove thus becomes: 35c66ec88fSEmmanuel Vadot 36c66ec88fSEmmanuel VadotCPU capacities are obtained by running the Dhrystone benchmark on each CPU at 37c66ec88fSEmmanuel Vadotmax frequency (with caches enabled). The obtained DMIPS score is then divided 38c66ec88fSEmmanuel Vadotby the frequency (in MHz) at which the benchmark has been run, so that 39c66ec88fSEmmanuel VadotDMIPS/MHz are obtained. Such values are then normalized w.r.t. the highest 40c66ec88fSEmmanuel Vadotscore obtained in the system. 41c66ec88fSEmmanuel Vadot 42c66ec88fSEmmanuel Vadot========================================== 43c66ec88fSEmmanuel Vadot3 - capacity-dmips-mhz 44c66ec88fSEmmanuel Vadot========================================== 45c66ec88fSEmmanuel Vadot 46c66ec88fSEmmanuel Vadotcapacity-dmips-mhz is an optional cpu node [1] property: u32 value 47c66ec88fSEmmanuel Vadotrepresenting CPU capacity expressed in normalized DMIPS/MHz. At boot time, the 48c66ec88fSEmmanuel Vadotmaximum frequency available to the cpu is then used to calculate the capacity 49c66ec88fSEmmanuel Vadotvalue internally used by the kernel. 50c66ec88fSEmmanuel Vadot 51c66ec88fSEmmanuel Vadotcapacity-dmips-mhz property is all-or-nothing: if it is specified for a cpu 52c66ec88fSEmmanuel Vadotnode, it has to be specified for every other cpu nodes, or the system will 53c66ec88fSEmmanuel Vadotfall back to the default capacity value for every CPU. If cpufreq is not 54c66ec88fSEmmanuel Vadotavailable, final capacities are calculated by directly using capacity-dmips- 55c66ec88fSEmmanuel Vadotmhz values (normalized w.r.t. the highest value found while parsing the DT). 56c66ec88fSEmmanuel Vadot 57c66ec88fSEmmanuel Vadot=========================================== 58c66ec88fSEmmanuel Vadot4 - Examples 59c66ec88fSEmmanuel Vadot=========================================== 60c66ec88fSEmmanuel Vadot 61c66ec88fSEmmanuel VadotExample 1 (ARM 64-bit, 6-cpu system, two clusters): 62c66ec88fSEmmanuel VadotThe capacities-dmips-mhz or DMIPS/MHz values (scaled to 1024) 63c66ec88fSEmmanuel Vadotare 1024 and 578 for cluster0 and cluster1. Further normalization 64c66ec88fSEmmanuel Vadotis done by the operating system based on cluster0@max-freq=1100 and 65*c9ccf3a3SEmmanuel Vadotcluster1@max-freq=850, final capacities are 1024 for cluster0 and 66*c9ccf3a3SEmmanuel Vadot446 for cluster1 (578*850/1100). 67c66ec88fSEmmanuel Vadot 68c66ec88fSEmmanuel Vadotcpus { 69c66ec88fSEmmanuel Vadot #address-cells = <2>; 70c66ec88fSEmmanuel Vadot #size-cells = <0>; 71c66ec88fSEmmanuel Vadot 72c66ec88fSEmmanuel Vadot cpu-map { 73c66ec88fSEmmanuel Vadot cluster0 { 74c66ec88fSEmmanuel Vadot core0 { 75c66ec88fSEmmanuel Vadot cpu = <&A57_0>; 76c66ec88fSEmmanuel Vadot }; 77c66ec88fSEmmanuel Vadot core1 { 78c66ec88fSEmmanuel Vadot cpu = <&A57_1>; 79c66ec88fSEmmanuel Vadot }; 80c66ec88fSEmmanuel Vadot }; 81c66ec88fSEmmanuel Vadot 82c66ec88fSEmmanuel Vadot cluster1 { 83c66ec88fSEmmanuel Vadot core0 { 84c66ec88fSEmmanuel Vadot cpu = <&A53_0>; 85c66ec88fSEmmanuel Vadot }; 86c66ec88fSEmmanuel Vadot core1 { 87c66ec88fSEmmanuel Vadot cpu = <&A53_1>; 88c66ec88fSEmmanuel Vadot }; 89c66ec88fSEmmanuel Vadot core2 { 90c66ec88fSEmmanuel Vadot cpu = <&A53_2>; 91c66ec88fSEmmanuel Vadot }; 92c66ec88fSEmmanuel Vadot core3 { 93c66ec88fSEmmanuel Vadot cpu = <&A53_3>; 94c66ec88fSEmmanuel Vadot }; 95c66ec88fSEmmanuel Vadot }; 96c66ec88fSEmmanuel Vadot }; 97c66ec88fSEmmanuel Vadot 98c66ec88fSEmmanuel Vadot idle-states { 99c66ec88fSEmmanuel Vadot entry-method = "psci"; 100c66ec88fSEmmanuel Vadot 101c66ec88fSEmmanuel Vadot CPU_SLEEP_0: cpu-sleep-0 { 102c66ec88fSEmmanuel Vadot compatible = "arm,idle-state"; 103c66ec88fSEmmanuel Vadot arm,psci-suspend-param = <0x0010000>; 104c66ec88fSEmmanuel Vadot local-timer-stop; 105c66ec88fSEmmanuel Vadot entry-latency-us = <100>; 106c66ec88fSEmmanuel Vadot exit-latency-us = <250>; 107c66ec88fSEmmanuel Vadot min-residency-us = <150>; 108c66ec88fSEmmanuel Vadot }; 109c66ec88fSEmmanuel Vadot 110c66ec88fSEmmanuel Vadot CLUSTER_SLEEP_0: cluster-sleep-0 { 111c66ec88fSEmmanuel Vadot compatible = "arm,idle-state"; 112c66ec88fSEmmanuel Vadot arm,psci-suspend-param = <0x1010000>; 113c66ec88fSEmmanuel Vadot local-timer-stop; 114c66ec88fSEmmanuel Vadot entry-latency-us = <800>; 115c66ec88fSEmmanuel Vadot exit-latency-us = <700>; 116c66ec88fSEmmanuel Vadot min-residency-us = <2500>; 117c66ec88fSEmmanuel Vadot }; 118c66ec88fSEmmanuel Vadot }; 119c66ec88fSEmmanuel Vadot 120c66ec88fSEmmanuel Vadot A57_0: cpu@0 { 121c66ec88fSEmmanuel Vadot compatible = "arm,cortex-a57"; 122c66ec88fSEmmanuel Vadot reg = <0x0 0x0>; 123c66ec88fSEmmanuel Vadot device_type = "cpu"; 124c66ec88fSEmmanuel Vadot enable-method = "psci"; 125c66ec88fSEmmanuel Vadot next-level-cache = <&A57_L2>; 126c66ec88fSEmmanuel Vadot clocks = <&scpi_dvfs 0>; 127c66ec88fSEmmanuel Vadot cpu-idle-states = <&CPU_SLEEP_0 &CLUSTER_SLEEP_0>; 128c66ec88fSEmmanuel Vadot capacity-dmips-mhz = <1024>; 129c66ec88fSEmmanuel Vadot }; 130c66ec88fSEmmanuel Vadot 131c66ec88fSEmmanuel Vadot A57_1: cpu@1 { 132c66ec88fSEmmanuel Vadot compatible = "arm,cortex-a57"; 133c66ec88fSEmmanuel Vadot reg = <0x0 0x1>; 134c66ec88fSEmmanuel Vadot device_type = "cpu"; 135c66ec88fSEmmanuel Vadot enable-method = "psci"; 136c66ec88fSEmmanuel Vadot next-level-cache = <&A57_L2>; 137c66ec88fSEmmanuel Vadot clocks = <&scpi_dvfs 0>; 138c66ec88fSEmmanuel Vadot cpu-idle-states = <&CPU_SLEEP_0 &CLUSTER_SLEEP_0>; 139c66ec88fSEmmanuel Vadot capacity-dmips-mhz = <1024>; 140c66ec88fSEmmanuel Vadot }; 141c66ec88fSEmmanuel Vadot 142c66ec88fSEmmanuel Vadot A53_0: cpu@100 { 143c66ec88fSEmmanuel Vadot compatible = "arm,cortex-a53"; 144c66ec88fSEmmanuel Vadot reg = <0x0 0x100>; 145c66ec88fSEmmanuel Vadot device_type = "cpu"; 146c66ec88fSEmmanuel Vadot enable-method = "psci"; 147c66ec88fSEmmanuel Vadot next-level-cache = <&A53_L2>; 148c66ec88fSEmmanuel Vadot clocks = <&scpi_dvfs 1>; 149c66ec88fSEmmanuel Vadot cpu-idle-states = <&CPU_SLEEP_0 &CLUSTER_SLEEP_0>; 150c66ec88fSEmmanuel Vadot capacity-dmips-mhz = <578>; 151c66ec88fSEmmanuel Vadot }; 152c66ec88fSEmmanuel Vadot 153c66ec88fSEmmanuel Vadot A53_1: cpu@101 { 154c66ec88fSEmmanuel Vadot compatible = "arm,cortex-a53"; 155c66ec88fSEmmanuel Vadot reg = <0x0 0x101>; 156c66ec88fSEmmanuel Vadot device_type = "cpu"; 157c66ec88fSEmmanuel Vadot enable-method = "psci"; 158c66ec88fSEmmanuel Vadot next-level-cache = <&A53_L2>; 159c66ec88fSEmmanuel Vadot clocks = <&scpi_dvfs 1>; 160c66ec88fSEmmanuel Vadot cpu-idle-states = <&CPU_SLEEP_0 &CLUSTER_SLEEP_0>; 161c66ec88fSEmmanuel Vadot capacity-dmips-mhz = <578>; 162c66ec88fSEmmanuel Vadot }; 163c66ec88fSEmmanuel Vadot 164c66ec88fSEmmanuel Vadot A53_2: cpu@102 { 165c66ec88fSEmmanuel Vadot compatible = "arm,cortex-a53"; 166c66ec88fSEmmanuel Vadot reg = <0x0 0x102>; 167c66ec88fSEmmanuel Vadot device_type = "cpu"; 168c66ec88fSEmmanuel Vadot enable-method = "psci"; 169c66ec88fSEmmanuel Vadot next-level-cache = <&A53_L2>; 170c66ec88fSEmmanuel Vadot clocks = <&scpi_dvfs 1>; 171c66ec88fSEmmanuel Vadot cpu-idle-states = <&CPU_SLEEP_0 &CLUSTER_SLEEP_0>; 172c66ec88fSEmmanuel Vadot capacity-dmips-mhz = <578>; 173c66ec88fSEmmanuel Vadot }; 174c66ec88fSEmmanuel Vadot 175c66ec88fSEmmanuel Vadot A53_3: cpu@103 { 176c66ec88fSEmmanuel Vadot compatible = "arm,cortex-a53"; 177c66ec88fSEmmanuel Vadot reg = <0x0 0x103>; 178c66ec88fSEmmanuel Vadot device_type = "cpu"; 179c66ec88fSEmmanuel Vadot enable-method = "psci"; 180c66ec88fSEmmanuel Vadot next-level-cache = <&A53_L2>; 181c66ec88fSEmmanuel Vadot clocks = <&scpi_dvfs 1>; 182c66ec88fSEmmanuel Vadot cpu-idle-states = <&CPU_SLEEP_0 &CLUSTER_SLEEP_0>; 183c66ec88fSEmmanuel Vadot capacity-dmips-mhz = <578>; 184c66ec88fSEmmanuel Vadot }; 185c66ec88fSEmmanuel Vadot 186c66ec88fSEmmanuel Vadot A57_L2: l2-cache0 { 187c66ec88fSEmmanuel Vadot compatible = "cache"; 188c66ec88fSEmmanuel Vadot }; 189c66ec88fSEmmanuel Vadot 190c66ec88fSEmmanuel Vadot A53_L2: l2-cache1 { 191c66ec88fSEmmanuel Vadot compatible = "cache"; 192c66ec88fSEmmanuel Vadot }; 193c66ec88fSEmmanuel Vadot}; 194c66ec88fSEmmanuel Vadot 195c66ec88fSEmmanuel VadotExample 2 (ARM 32-bit, 4-cpu system, two clusters, 196c66ec88fSEmmanuel Vadot cpus 0,1@1GHz, cpus 2,3@500MHz): 197c66ec88fSEmmanuel Vadotcapacities-dmips-mhz are scaled w.r.t. 2 (cpu@0 and cpu@1), this means that first 198c66ec88fSEmmanuel Vadotcpu@0 and cpu@1 are twice fast than cpu@2 and cpu@3 (at the same frequency) 199c66ec88fSEmmanuel Vadot 200c66ec88fSEmmanuel Vadotcpus { 201c66ec88fSEmmanuel Vadot #address-cells = <1>; 202c66ec88fSEmmanuel Vadot #size-cells = <0>; 203c66ec88fSEmmanuel Vadot 204c66ec88fSEmmanuel Vadot cpu0: cpu@0 { 205c66ec88fSEmmanuel Vadot device_type = "cpu"; 206c66ec88fSEmmanuel Vadot compatible = "arm,cortex-a15"; 207c66ec88fSEmmanuel Vadot reg = <0>; 208c66ec88fSEmmanuel Vadot capacity-dmips-mhz = <2>; 209c66ec88fSEmmanuel Vadot }; 210c66ec88fSEmmanuel Vadot 211c66ec88fSEmmanuel Vadot cpu1: cpu@1 { 212c66ec88fSEmmanuel Vadot device_type = "cpu"; 213c66ec88fSEmmanuel Vadot compatible = "arm,cortex-a15"; 214c66ec88fSEmmanuel Vadot reg = <1>; 215c66ec88fSEmmanuel Vadot capacity-dmips-mhz = <2>; 216c66ec88fSEmmanuel Vadot }; 217c66ec88fSEmmanuel Vadot 218c66ec88fSEmmanuel Vadot cpu2: cpu@2 { 219c66ec88fSEmmanuel Vadot device_type = "cpu"; 220c66ec88fSEmmanuel Vadot compatible = "arm,cortex-a15"; 221c66ec88fSEmmanuel Vadot reg = <0x100>; 222c66ec88fSEmmanuel Vadot capacity-dmips-mhz = <1>; 223c66ec88fSEmmanuel Vadot }; 224c66ec88fSEmmanuel Vadot 225c66ec88fSEmmanuel Vadot cpu3: cpu@3 { 226c66ec88fSEmmanuel Vadot device_type = "cpu"; 227c66ec88fSEmmanuel Vadot compatible = "arm,cortex-a15"; 228c66ec88fSEmmanuel Vadot reg = <0x101>; 229c66ec88fSEmmanuel Vadot capacity-dmips-mhz = <1>; 230c66ec88fSEmmanuel Vadot }; 231c66ec88fSEmmanuel Vadot}; 232c66ec88fSEmmanuel Vadot 233c66ec88fSEmmanuel Vadot=========================================== 234c66ec88fSEmmanuel Vadot5 - References 235c66ec88fSEmmanuel Vadot=========================================== 236c66ec88fSEmmanuel Vadot 237c66ec88fSEmmanuel Vadot[1] ARM Linux Kernel documentation - CPUs bindings 238c66ec88fSEmmanuel Vadot Documentation/devicetree/bindings/arm/cpus.yaml 239