1 // SPDX-License-Identifier: GPL-2.0 2 #include <linux/perf_event.h> 3 #include <linux/sysfs.h> 4 #include <linux/nospec.h> 5 #include <asm/intel-family.h> 6 #include "probe.h" 7 8 enum perf_msr_id { 9 PERF_MSR_TSC = 0, 10 PERF_MSR_APERF = 1, 11 PERF_MSR_MPERF = 2, 12 PERF_MSR_PPERF = 3, 13 PERF_MSR_SMI = 4, 14 PERF_MSR_PTSC = 5, 15 PERF_MSR_IRPERF = 6, 16 PERF_MSR_THERM = 7, 17 PERF_MSR_EVENT_MAX, 18 }; 19 20 static bool test_aperfmperf(int idx, void *data) 21 { 22 return boot_cpu_has(X86_FEATURE_APERFMPERF); 23 } 24 25 static bool test_ptsc(int idx, void *data) 26 { 27 return boot_cpu_has(X86_FEATURE_PTSC); 28 } 29 30 static bool test_irperf(int idx, void *data) 31 { 32 return boot_cpu_has(X86_FEATURE_IRPERF); 33 } 34 35 static bool test_therm_status(int idx, void *data) 36 { 37 return boot_cpu_has(X86_FEATURE_DTHERM); 38 } 39 40 static bool test_intel(int idx, void *data) 41 { 42 if (boot_cpu_data.x86_vendor != X86_VENDOR_INTEL || 43 boot_cpu_data.x86 != 6) 44 return false; 45 46 switch (boot_cpu_data.x86_model) { 47 case INTEL_FAM6_NEHALEM: 48 case INTEL_FAM6_NEHALEM_G: 49 case INTEL_FAM6_NEHALEM_EP: 50 case INTEL_FAM6_NEHALEM_EX: 51 52 case INTEL_FAM6_WESTMERE: 53 case INTEL_FAM6_WESTMERE_EP: 54 case INTEL_FAM6_WESTMERE_EX: 55 56 case INTEL_FAM6_SANDYBRIDGE: 57 case INTEL_FAM6_SANDYBRIDGE_X: 58 59 case INTEL_FAM6_IVYBRIDGE: 60 case INTEL_FAM6_IVYBRIDGE_X: 61 62 case INTEL_FAM6_HASWELL: 63 case INTEL_FAM6_HASWELL_X: 64 case INTEL_FAM6_HASWELL_L: 65 case INTEL_FAM6_HASWELL_G: 66 67 case INTEL_FAM6_BROADWELL: 68 case INTEL_FAM6_BROADWELL_D: 69 case INTEL_FAM6_BROADWELL_G: 70 case INTEL_FAM6_BROADWELL_X: 71 72 case INTEL_FAM6_ATOM_SILVERMONT: 73 case INTEL_FAM6_ATOM_SILVERMONT_D: 74 case INTEL_FAM6_ATOM_AIRMONT: 75 76 case INTEL_FAM6_ATOM_GOLDMONT: 77 case INTEL_FAM6_ATOM_GOLDMONT_D: 78 79 case INTEL_FAM6_ATOM_GOLDMONT_PLUS: 80 81 case INTEL_FAM6_XEON_PHI_KNL: 82 case INTEL_FAM6_XEON_PHI_KNM: 83 if (idx == PERF_MSR_SMI) 84 return true; 85 break; 86 87 case INTEL_FAM6_SKYLAKE_L: 88 case INTEL_FAM6_SKYLAKE: 89 case INTEL_FAM6_SKYLAKE_X: 90 case INTEL_FAM6_KABYLAKE_L: 91 case INTEL_FAM6_KABYLAKE: 92 case INTEL_FAM6_ICELAKE_L: 93 if (idx == PERF_MSR_SMI || idx == PERF_MSR_PPERF) 94 return true; 95 break; 96 } 97 98 return false; 99 } 100 101 PMU_EVENT_ATTR_STRING(tsc, attr_tsc, "event=0x00" ); 102 PMU_EVENT_ATTR_STRING(aperf, attr_aperf, "event=0x01" ); 103 PMU_EVENT_ATTR_STRING(mperf, attr_mperf, "event=0x02" ); 104 PMU_EVENT_ATTR_STRING(pperf, attr_pperf, "event=0x03" ); 105 PMU_EVENT_ATTR_STRING(smi, attr_smi, "event=0x04" ); 106 PMU_EVENT_ATTR_STRING(ptsc, attr_ptsc, "event=0x05" ); 107 PMU_EVENT_ATTR_STRING(irperf, attr_irperf, "event=0x06" ); 108 PMU_EVENT_ATTR_STRING(cpu_thermal_margin, attr_therm, "event=0x07" ); 109 PMU_EVENT_ATTR_STRING(cpu_thermal_margin.snapshot, attr_therm_snap, "1" ); 110 PMU_EVENT_ATTR_STRING(cpu_thermal_margin.unit, attr_therm_unit, "C" ); 111 112 static unsigned long msr_mask; 113 114 PMU_EVENT_GROUP(events, aperf); 115 PMU_EVENT_GROUP(events, mperf); 116 PMU_EVENT_GROUP(events, pperf); 117 PMU_EVENT_GROUP(events, smi); 118 PMU_EVENT_GROUP(events, ptsc); 119 PMU_EVENT_GROUP(events, irperf); 120 121 static struct attribute *attrs_therm[] = { 122 &attr_therm.attr.attr, 123 &attr_therm_snap.attr.attr, 124 &attr_therm_unit.attr.attr, 125 NULL, 126 }; 127 128 static struct attribute_group group_therm = { 129 .name = "events", 130 .attrs = attrs_therm, 131 }; 132 133 static struct perf_msr msr[] = { 134 [PERF_MSR_TSC] = { .no_check = true, }, 135 [PERF_MSR_APERF] = { MSR_IA32_APERF, &group_aperf, test_aperfmperf, }, 136 [PERF_MSR_MPERF] = { MSR_IA32_MPERF, &group_mperf, test_aperfmperf, }, 137 [PERF_MSR_PPERF] = { MSR_PPERF, &group_pperf, test_intel, }, 138 [PERF_MSR_SMI] = { MSR_SMI_COUNT, &group_smi, test_intel, }, 139 [PERF_MSR_PTSC] = { MSR_F15H_PTSC, &group_ptsc, test_ptsc, }, 140 [PERF_MSR_IRPERF] = { MSR_F17H_IRPERF, &group_irperf, test_irperf, }, 141 [PERF_MSR_THERM] = { MSR_IA32_THERM_STATUS, &group_therm, test_therm_status, }, 142 }; 143 144 static struct attribute *events_attrs[] = { 145 &attr_tsc.attr.attr, 146 NULL, 147 }; 148 149 static struct attribute_group events_attr_group = { 150 .name = "events", 151 .attrs = events_attrs, 152 }; 153 154 PMU_FORMAT_ATTR(event, "config:0-63"); 155 static struct attribute *format_attrs[] = { 156 &format_attr_event.attr, 157 NULL, 158 }; 159 static struct attribute_group format_attr_group = { 160 .name = "format", 161 .attrs = format_attrs, 162 }; 163 164 static const struct attribute_group *attr_groups[] = { 165 &events_attr_group, 166 &format_attr_group, 167 NULL, 168 }; 169 170 static const struct attribute_group *attr_update[] = { 171 &group_aperf, 172 &group_mperf, 173 &group_pperf, 174 &group_smi, 175 &group_ptsc, 176 &group_irperf, 177 &group_therm, 178 NULL, 179 }; 180 181 static int msr_event_init(struct perf_event *event) 182 { 183 u64 cfg = event->attr.config; 184 185 if (event->attr.type != event->pmu->type) 186 return -ENOENT; 187 188 /* unsupported modes and filters */ 189 if (event->attr.sample_period) /* no sampling */ 190 return -EINVAL; 191 192 if (cfg >= PERF_MSR_EVENT_MAX) 193 return -EINVAL; 194 195 cfg = array_index_nospec((unsigned long)cfg, PERF_MSR_EVENT_MAX); 196 197 if (!(msr_mask & (1 << cfg))) 198 return -EINVAL; 199 200 event->hw.idx = -1; 201 event->hw.event_base = msr[cfg].msr; 202 event->hw.config = cfg; 203 204 return 0; 205 } 206 207 static inline u64 msr_read_counter(struct perf_event *event) 208 { 209 u64 now; 210 211 if (event->hw.event_base) 212 rdmsrl(event->hw.event_base, now); 213 else 214 now = rdtsc_ordered(); 215 216 return now; 217 } 218 219 static void msr_event_update(struct perf_event *event) 220 { 221 u64 prev, now; 222 s64 delta; 223 224 /* Careful, an NMI might modify the previous event value: */ 225 again: 226 prev = local64_read(&event->hw.prev_count); 227 now = msr_read_counter(event); 228 229 if (local64_cmpxchg(&event->hw.prev_count, prev, now) != prev) 230 goto again; 231 232 delta = now - prev; 233 if (unlikely(event->hw.event_base == MSR_SMI_COUNT)) { 234 delta = sign_extend64(delta, 31); 235 local64_add(delta, &event->count); 236 } else if (unlikely(event->hw.event_base == MSR_IA32_THERM_STATUS)) { 237 /* If valid, extract digital readout, otherwise set to -1: */ 238 now = now & (1ULL << 31) ? (now >> 16) & 0x3f : -1; 239 local64_set(&event->count, now); 240 } else { 241 local64_add(delta, &event->count); 242 } 243 } 244 245 static void msr_event_start(struct perf_event *event, int flags) 246 { 247 u64 now = msr_read_counter(event); 248 249 local64_set(&event->hw.prev_count, now); 250 } 251 252 static void msr_event_stop(struct perf_event *event, int flags) 253 { 254 msr_event_update(event); 255 } 256 257 static void msr_event_del(struct perf_event *event, int flags) 258 { 259 msr_event_stop(event, PERF_EF_UPDATE); 260 } 261 262 static int msr_event_add(struct perf_event *event, int flags) 263 { 264 if (flags & PERF_EF_START) 265 msr_event_start(event, flags); 266 267 return 0; 268 } 269 270 static struct pmu pmu_msr = { 271 .task_ctx_nr = perf_sw_context, 272 .attr_groups = attr_groups, 273 .event_init = msr_event_init, 274 .add = msr_event_add, 275 .del = msr_event_del, 276 .start = msr_event_start, 277 .stop = msr_event_stop, 278 .read = msr_event_update, 279 .capabilities = PERF_PMU_CAP_NO_INTERRUPT | PERF_PMU_CAP_NO_EXCLUDE, 280 .attr_update = attr_update, 281 }; 282 283 static int __init msr_init(void) 284 { 285 if (!boot_cpu_has(X86_FEATURE_TSC)) { 286 pr_cont("no MSR PMU driver.\n"); 287 return 0; 288 } 289 290 msr_mask = perf_msr_probe(msr, PERF_MSR_EVENT_MAX, true, NULL); 291 292 perf_pmu_register(&pmu_msr, "msr", -1); 293 294 return 0; 295 } 296 device_initcall(msr_init); 297