1 // SPDX-License-Identifier: GPL-2.0 2 #include <linux/perf_event.h> 3 #include <linux/nospec.h> 4 #include <asm/intel-family.h> 5 6 enum perf_msr_id { 7 PERF_MSR_TSC = 0, 8 PERF_MSR_APERF = 1, 9 PERF_MSR_MPERF = 2, 10 PERF_MSR_PPERF = 3, 11 PERF_MSR_SMI = 4, 12 PERF_MSR_PTSC = 5, 13 PERF_MSR_IRPERF = 6, 14 PERF_MSR_THERM = 7, 15 PERF_MSR_THERM_SNAP = 8, 16 PERF_MSR_THERM_UNIT = 9, 17 PERF_MSR_EVENT_MAX, 18 }; 19 20 static bool test_aperfmperf(int idx) 21 { 22 return boot_cpu_has(X86_FEATURE_APERFMPERF); 23 } 24 25 static bool test_ptsc(int idx) 26 { 27 return boot_cpu_has(X86_FEATURE_PTSC); 28 } 29 30 static bool test_irperf(int idx) 31 { 32 return boot_cpu_has(X86_FEATURE_IRPERF); 33 } 34 35 static bool test_therm_status(int idx) 36 { 37 return boot_cpu_has(X86_FEATURE_DTHERM); 38 } 39 40 static bool test_intel(int idx) 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_CORE: 63 case INTEL_FAM6_HASWELL_X: 64 case INTEL_FAM6_HASWELL_ULT: 65 case INTEL_FAM6_HASWELL_GT3E: 66 67 case INTEL_FAM6_BROADWELL_CORE: 68 case INTEL_FAM6_BROADWELL_XEON_D: 69 case INTEL_FAM6_BROADWELL_GT3E: 70 case INTEL_FAM6_BROADWELL_X: 71 72 case INTEL_FAM6_ATOM_SILVERMONT: 73 case INTEL_FAM6_ATOM_SILVERMONT_X: 74 case INTEL_FAM6_ATOM_AIRMONT: 75 76 case INTEL_FAM6_ATOM_GOLDMONT: 77 case INTEL_FAM6_ATOM_GOLDMONT_X: 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_MOBILE: 88 case INTEL_FAM6_SKYLAKE_DESKTOP: 89 case INTEL_FAM6_SKYLAKE_X: 90 case INTEL_FAM6_KABYLAKE_MOBILE: 91 case INTEL_FAM6_KABYLAKE_DESKTOP: 92 if (idx == PERF_MSR_SMI || idx == PERF_MSR_PPERF) 93 return true; 94 break; 95 } 96 97 return false; 98 } 99 100 struct perf_msr { 101 u64 msr; 102 struct perf_pmu_events_attr *attr; 103 bool (*test)(int idx); 104 }; 105 106 PMU_EVENT_ATTR_STRING(tsc, evattr_tsc, "event=0x00" ); 107 PMU_EVENT_ATTR_STRING(aperf, evattr_aperf, "event=0x01" ); 108 PMU_EVENT_ATTR_STRING(mperf, evattr_mperf, "event=0x02" ); 109 PMU_EVENT_ATTR_STRING(pperf, evattr_pperf, "event=0x03" ); 110 PMU_EVENT_ATTR_STRING(smi, evattr_smi, "event=0x04" ); 111 PMU_EVENT_ATTR_STRING(ptsc, evattr_ptsc, "event=0x05" ); 112 PMU_EVENT_ATTR_STRING(irperf, evattr_irperf, "event=0x06" ); 113 PMU_EVENT_ATTR_STRING(cpu_thermal_margin, evattr_therm, "event=0x07" ); 114 PMU_EVENT_ATTR_STRING(cpu_thermal_margin.snapshot, evattr_therm_snap, "1" ); 115 PMU_EVENT_ATTR_STRING(cpu_thermal_margin.unit, evattr_therm_unit, "C" ); 116 117 static struct perf_msr msr[] = { 118 [PERF_MSR_TSC] = { 0, &evattr_tsc, NULL, }, 119 [PERF_MSR_APERF] = { MSR_IA32_APERF, &evattr_aperf, test_aperfmperf, }, 120 [PERF_MSR_MPERF] = { MSR_IA32_MPERF, &evattr_mperf, test_aperfmperf, }, 121 [PERF_MSR_PPERF] = { MSR_PPERF, &evattr_pperf, test_intel, }, 122 [PERF_MSR_SMI] = { MSR_SMI_COUNT, &evattr_smi, test_intel, }, 123 [PERF_MSR_PTSC] = { MSR_F15H_PTSC, &evattr_ptsc, test_ptsc, }, 124 [PERF_MSR_IRPERF] = { MSR_F17H_IRPERF, &evattr_irperf, test_irperf, }, 125 [PERF_MSR_THERM] = { MSR_IA32_THERM_STATUS, &evattr_therm, test_therm_status, }, 126 [PERF_MSR_THERM_SNAP] = { MSR_IA32_THERM_STATUS, &evattr_therm_snap, test_therm_status, }, 127 [PERF_MSR_THERM_UNIT] = { MSR_IA32_THERM_STATUS, &evattr_therm_unit, test_therm_status, }, 128 }; 129 130 static struct attribute *events_attrs[PERF_MSR_EVENT_MAX + 1] = { 131 NULL, 132 }; 133 134 static struct attribute_group events_attr_group = { 135 .name = "events", 136 .attrs = events_attrs, 137 }; 138 139 PMU_FORMAT_ATTR(event, "config:0-63"); 140 static struct attribute *format_attrs[] = { 141 &format_attr_event.attr, 142 NULL, 143 }; 144 static struct attribute_group format_attr_group = { 145 .name = "format", 146 .attrs = format_attrs, 147 }; 148 149 static const struct attribute_group *attr_groups[] = { 150 &events_attr_group, 151 &format_attr_group, 152 NULL, 153 }; 154 155 static int msr_event_init(struct perf_event *event) 156 { 157 u64 cfg = event->attr.config; 158 159 if (event->attr.type != event->pmu->type) 160 return -ENOENT; 161 162 /* unsupported modes and filters */ 163 if (event->attr.exclude_user || 164 event->attr.exclude_kernel || 165 event->attr.exclude_hv || 166 event->attr.exclude_idle || 167 event->attr.exclude_host || 168 event->attr.exclude_guest || 169 event->attr.sample_period) /* no sampling */ 170 return -EINVAL; 171 172 if (cfg >= PERF_MSR_EVENT_MAX) 173 return -EINVAL; 174 175 cfg = array_index_nospec((unsigned long)cfg, PERF_MSR_EVENT_MAX); 176 177 if (!msr[cfg].attr) 178 return -EINVAL; 179 180 event->hw.idx = -1; 181 event->hw.event_base = msr[cfg].msr; 182 event->hw.config = cfg; 183 184 return 0; 185 } 186 187 static inline u64 msr_read_counter(struct perf_event *event) 188 { 189 u64 now; 190 191 if (event->hw.event_base) 192 rdmsrl(event->hw.event_base, now); 193 else 194 now = rdtsc_ordered(); 195 196 return now; 197 } 198 199 static void msr_event_update(struct perf_event *event) 200 { 201 u64 prev, now; 202 s64 delta; 203 204 /* Careful, an NMI might modify the previous event value: */ 205 again: 206 prev = local64_read(&event->hw.prev_count); 207 now = msr_read_counter(event); 208 209 if (local64_cmpxchg(&event->hw.prev_count, prev, now) != prev) 210 goto again; 211 212 delta = now - prev; 213 if (unlikely(event->hw.event_base == MSR_SMI_COUNT)) { 214 delta = sign_extend64(delta, 31); 215 local64_add(delta, &event->count); 216 } else if (unlikely(event->hw.event_base == MSR_IA32_THERM_STATUS)) { 217 /* If valid, extract digital readout, otherwise set to -1: */ 218 now = now & (1ULL << 31) ? (now >> 16) & 0x3f : -1; 219 local64_set(&event->count, now); 220 } else { 221 local64_add(delta, &event->count); 222 } 223 } 224 225 static void msr_event_start(struct perf_event *event, int flags) 226 { 227 u64 now = msr_read_counter(event); 228 229 local64_set(&event->hw.prev_count, now); 230 } 231 232 static void msr_event_stop(struct perf_event *event, int flags) 233 { 234 msr_event_update(event); 235 } 236 237 static void msr_event_del(struct perf_event *event, int flags) 238 { 239 msr_event_stop(event, PERF_EF_UPDATE); 240 } 241 242 static int msr_event_add(struct perf_event *event, int flags) 243 { 244 if (flags & PERF_EF_START) 245 msr_event_start(event, flags); 246 247 return 0; 248 } 249 250 static struct pmu pmu_msr = { 251 .task_ctx_nr = perf_sw_context, 252 .attr_groups = attr_groups, 253 .event_init = msr_event_init, 254 .add = msr_event_add, 255 .del = msr_event_del, 256 .start = msr_event_start, 257 .stop = msr_event_stop, 258 .read = msr_event_update, 259 .capabilities = PERF_PMU_CAP_NO_INTERRUPT, 260 }; 261 262 static int __init msr_init(void) 263 { 264 int i, j = 0; 265 266 if (!boot_cpu_has(X86_FEATURE_TSC)) { 267 pr_cont("no MSR PMU driver.\n"); 268 return 0; 269 } 270 271 /* Probe the MSRs. */ 272 for (i = PERF_MSR_TSC + 1; i < PERF_MSR_EVENT_MAX; i++) { 273 u64 val; 274 275 /* Virt sucks; you cannot tell if a R/O MSR is present :/ */ 276 if (!msr[i].test(i) || rdmsrl_safe(msr[i].msr, &val)) 277 msr[i].attr = NULL; 278 } 279 280 /* List remaining MSRs in the sysfs attrs. */ 281 for (i = 0; i < PERF_MSR_EVENT_MAX; i++) { 282 if (msr[i].attr) 283 events_attrs[j++] = &msr[i].attr->attr.attr; 284 } 285 events_attrs[j] = NULL; 286 287 perf_pmu_register(&pmu_msr, "msr", -1); 288 289 return 0; 290 } 291 device_initcall(msr_init); 292