1========================================== 2ARM CPUs capacity bindings 3========================================== 4 5========================================== 61 - Introduction 7========================================== 8 9ARM systems may be configured to have cpus with different power/performance 10characteristics within the same chip. In this case, additional information has 11to be made available to the kernel for it to be aware of such differences and 12take decisions accordingly. 13 14========================================== 152 - CPU capacity definition 16========================================== 17 18CPU capacity is a number that provides the scheduler information about CPUs 19heterogeneity. Such heterogeneity can come from micro-architectural differences 20(e.g., ARM big.LITTLE systems) or maximum frequency at which CPUs can run 21(e.g., SMP systems with multiple frequency domains). Heterogeneity in this 22context is about differing performance characteristics; this binding tries to 23capture a first-order approximation of the relative performance of CPUs. 24 25CPU capacities are obtained by running a suitable benchmark. This binding makes 26no guarantees on the validity or suitability of any particular benchmark, the 27final capacity should, however, be: 28 29* A "single-threaded" or CPU affine benchmark 30* Divided by the running frequency of the CPU executing the benchmark 31* Not subject to dynamic frequency scaling of the CPU 32 33For the time being we however advise usage of the Dhrystone benchmark. What 34above thus becomes: 35 36CPU capacities are obtained by running the Dhrystone benchmark on each CPU at 37max frequency (with caches enabled). The obtained DMIPS score is then divided 38by the frequency (in MHz) at which the benchmark has been run, so that 39DMIPS/MHz are obtained. Such values are then normalized w.r.t. the highest 40score obtained in the system. 41 42========================================== 433 - capacity-dmips-mhz 44========================================== 45 46capacity-dmips-mhz is an optional cpu node [1] property: u32 value 47representing CPU capacity expressed in normalized DMIPS/MHz. At boot time, the 48maximum frequency available to the cpu is then used to calculate the capacity 49value internally used by the kernel. 50 51capacity-dmips-mhz property is all-or-nothing: if it is specified for a cpu 52node, it has to be specified for every other cpu nodes, or the system will 53fall back to the default capacity value for every CPU. If cpufreq is not 54available, final capacities are calculated by directly using capacity-dmips- 55mhz values (normalized w.r.t. the highest value found while parsing the DT). 56 57=========================================== 584 - Examples 59=========================================== 60 61Example 1 (ARM 64-bit, 6-cpu system, two clusters): 62capacities-dmips-mhz are scaled w.r.t. 1024 (cpu@0 and cpu@1) 63supposing cluster0@max-freq=1100 and custer1@max-freq=850, 64final capacities are 1024 for cluster0 and 446 for cluster1 65 66cpus { 67 #address-cells = <2>; 68 #size-cells = <0>; 69 70 cpu-map { 71 cluster0 { 72 core0 { 73 cpu = <&A57_0>; 74 }; 75 core1 { 76 cpu = <&A57_1>; 77 }; 78 }; 79 80 cluster1 { 81 core0 { 82 cpu = <&A53_0>; 83 }; 84 core1 { 85 cpu = <&A53_1>; 86 }; 87 core2 { 88 cpu = <&A53_2>; 89 }; 90 core3 { 91 cpu = <&A53_3>; 92 }; 93 }; 94 }; 95 96 idle-states { 97 entry-method = "psci"; 98 99 CPU_SLEEP_0: cpu-sleep-0 { 100 compatible = "arm,idle-state"; 101 arm,psci-suspend-param = <0x0010000>; 102 local-timer-stop; 103 entry-latency-us = <100>; 104 exit-latency-us = <250>; 105 min-residency-us = <150>; 106 }; 107 108 CLUSTER_SLEEP_0: cluster-sleep-0 { 109 compatible = "arm,idle-state"; 110 arm,psci-suspend-param = <0x1010000>; 111 local-timer-stop; 112 entry-latency-us = <800>; 113 exit-latency-us = <700>; 114 min-residency-us = <2500>; 115 }; 116 }; 117 118 A57_0: cpu@0 { 119 compatible = "arm,cortex-a57","arm,armv8"; 120 reg = <0x0 0x0>; 121 device_type = "cpu"; 122 enable-method = "psci"; 123 next-level-cache = <&A57_L2>; 124 clocks = <&scpi_dvfs 0>; 125 cpu-idle-states = <&CPU_SLEEP_0 &CLUSTER_SLEEP_0>; 126 capacity-dmips-mhz = <1024>; 127 }; 128 129 A57_1: cpu@1 { 130 compatible = "arm,cortex-a57","arm,armv8"; 131 reg = <0x0 0x1>; 132 device_type = "cpu"; 133 enable-method = "psci"; 134 next-level-cache = <&A57_L2>; 135 clocks = <&scpi_dvfs 0>; 136 cpu-idle-states = <&CPU_SLEEP_0 &CLUSTER_SLEEP_0>; 137 capacity-dmips-mhz = <1024>; 138 }; 139 140 A53_0: cpu@100 { 141 compatible = "arm,cortex-a53","arm,armv8"; 142 reg = <0x0 0x100>; 143 device_type = "cpu"; 144 enable-method = "psci"; 145 next-level-cache = <&A53_L2>; 146 clocks = <&scpi_dvfs 1>; 147 cpu-idle-states = <&CPU_SLEEP_0 &CLUSTER_SLEEP_0>; 148 capacity-dmips-mhz = <578>; 149 }; 150 151 A53_1: cpu@101 { 152 compatible = "arm,cortex-a53","arm,armv8"; 153 reg = <0x0 0x101>; 154 device_type = "cpu"; 155 enable-method = "psci"; 156 next-level-cache = <&A53_L2>; 157 clocks = <&scpi_dvfs 1>; 158 cpu-idle-states = <&CPU_SLEEP_0 &CLUSTER_SLEEP_0>; 159 capacity-dmips-mhz = <578>; 160 }; 161 162 A53_2: cpu@102 { 163 compatible = "arm,cortex-a53","arm,armv8"; 164 reg = <0x0 0x102>; 165 device_type = "cpu"; 166 enable-method = "psci"; 167 next-level-cache = <&A53_L2>; 168 clocks = <&scpi_dvfs 1>; 169 cpu-idle-states = <&CPU_SLEEP_0 &CLUSTER_SLEEP_0>; 170 capacity-dmips-mhz = <578>; 171 }; 172 173 A53_3: cpu@103 { 174 compatible = "arm,cortex-a53","arm,armv8"; 175 reg = <0x0 0x103>; 176 device_type = "cpu"; 177 enable-method = "psci"; 178 next-level-cache = <&A53_L2>; 179 clocks = <&scpi_dvfs 1>; 180 cpu-idle-states = <&CPU_SLEEP_0 &CLUSTER_SLEEP_0>; 181 capacity-dmips-mhz = <578>; 182 }; 183 184 A57_L2: l2-cache0 { 185 compatible = "cache"; 186 }; 187 188 A53_L2: l2-cache1 { 189 compatible = "cache"; 190 }; 191}; 192 193Example 2 (ARM 32-bit, 4-cpu system, two clusters, 194 cpus 0,1@1GHz, cpus 2,3@500MHz): 195capacities-dmips-mhz are scaled w.r.t. 2 (cpu@0 and cpu@1), this means that first 196cpu@0 and cpu@1 are twice fast than cpu@2 and cpu@3 (at the same frequency) 197 198cpus { 199 #address-cells = <1>; 200 #size-cells = <0>; 201 202 cpu0: cpu@0 { 203 device_type = "cpu"; 204 compatible = "arm,cortex-a15"; 205 reg = <0>; 206 capacity-dmips-mhz = <2>; 207 }; 208 209 cpu1: cpu@1 { 210 device_type = "cpu"; 211 compatible = "arm,cortex-a15"; 212 reg = <1>; 213 capacity-dmips-mhz = <2>; 214 }; 215 216 cpu2: cpu@2 { 217 device_type = "cpu"; 218 compatible = "arm,cortex-a15"; 219 reg = <0x100>; 220 capacity-dmips-mhz = <1>; 221 }; 222 223 cpu3: cpu@3 { 224 device_type = "cpu"; 225 compatible = "arm,cortex-a15"; 226 reg = <0x101>; 227 capacity-dmips-mhz = <1>; 228 }; 229}; 230 231=========================================== 2325 - References 233=========================================== 234 235[1] ARM Linux Kernel documentation - CPUs bindings 236 Documentation/devicetree/bindings/arm/cpus.txt 237