1 /* SPDX-License-Identifier: GPL-2.0 */ 2 #ifndef _ASM_X86_PERF_EVENT_H 3 #define _ASM_X86_PERF_EVENT_H 4 5 #include <linux/static_call.h> 6 7 /* 8 * Performance event hw details: 9 */ 10 11 #define INTEL_PMC_MAX_GENERIC 32 12 #define INTEL_PMC_MAX_FIXED 16 13 #define INTEL_PMC_IDX_FIXED 32 14 15 #define X86_PMC_IDX_MAX 64 16 17 #define MSR_ARCH_PERFMON_PERFCTR0 0xc1 18 #define MSR_ARCH_PERFMON_PERFCTR1 0xc2 19 20 #define MSR_ARCH_PERFMON_EVENTSEL0 0x186 21 #define MSR_ARCH_PERFMON_EVENTSEL1 0x187 22 23 #define ARCH_PERFMON_EVENTSEL_EVENT 0x000000FFULL 24 #define ARCH_PERFMON_EVENTSEL_UMASK 0x0000FF00ULL 25 #define ARCH_PERFMON_EVENTSEL_USR (1ULL << 16) 26 #define ARCH_PERFMON_EVENTSEL_OS (1ULL << 17) 27 #define ARCH_PERFMON_EVENTSEL_EDGE (1ULL << 18) 28 #define ARCH_PERFMON_EVENTSEL_PIN_CONTROL (1ULL << 19) 29 #define ARCH_PERFMON_EVENTSEL_INT (1ULL << 20) 30 #define ARCH_PERFMON_EVENTSEL_ANY (1ULL << 21) 31 #define ARCH_PERFMON_EVENTSEL_ENABLE (1ULL << 22) 32 #define ARCH_PERFMON_EVENTSEL_INV (1ULL << 23) 33 #define ARCH_PERFMON_EVENTSEL_CMASK 0xFF000000ULL 34 #define ARCH_PERFMON_EVENTSEL_BR_CNTR (1ULL << 35) 35 #define ARCH_PERFMON_EVENTSEL_EQ (1ULL << 36) 36 #define ARCH_PERFMON_EVENTSEL_UMASK2 (0xFFULL << 40) 37 38 #define INTEL_FIXED_BITS_STRIDE 4 39 #define INTEL_FIXED_0_KERNEL (1ULL << 0) 40 #define INTEL_FIXED_0_USER (1ULL << 1) 41 #define INTEL_FIXED_0_ANYTHREAD (1ULL << 2) 42 #define INTEL_FIXED_0_ENABLE_PMI (1ULL << 3) 43 #define INTEL_FIXED_3_METRICS_CLEAR (1ULL << 2) 44 45 #define HSW_IN_TX (1ULL << 32) 46 #define HSW_IN_TX_CHECKPOINTED (1ULL << 33) 47 #define ICL_EVENTSEL_ADAPTIVE (1ULL << 34) 48 #define ICL_FIXED_0_ADAPTIVE (1ULL << 32) 49 50 #define INTEL_FIXED_BITS_MASK \ 51 (INTEL_FIXED_0_KERNEL | INTEL_FIXED_0_USER | \ 52 INTEL_FIXED_0_ANYTHREAD | INTEL_FIXED_0_ENABLE_PMI | \ 53 ICL_FIXED_0_ADAPTIVE) 54 55 #define intel_fixed_bits_by_idx(_idx, _bits) \ 56 ((_bits) << ((_idx) * INTEL_FIXED_BITS_STRIDE)) 57 58 #define AMD64_EVENTSEL_INT_CORE_ENABLE (1ULL << 36) 59 #define AMD64_EVENTSEL_GUESTONLY (1ULL << 40) 60 #define AMD64_EVENTSEL_HOSTONLY (1ULL << 41) 61 62 #define AMD64_EVENTSEL_INT_CORE_SEL_SHIFT 37 63 #define AMD64_EVENTSEL_INT_CORE_SEL_MASK \ 64 (0xFULL << AMD64_EVENTSEL_INT_CORE_SEL_SHIFT) 65 66 #define AMD64_EVENTSEL_EVENT \ 67 (ARCH_PERFMON_EVENTSEL_EVENT | (0x0FULL << 32)) 68 #define INTEL_ARCH_EVENT_MASK \ 69 (ARCH_PERFMON_EVENTSEL_UMASK | ARCH_PERFMON_EVENTSEL_EVENT) 70 71 #define AMD64_L3_SLICE_SHIFT 48 72 #define AMD64_L3_SLICE_MASK \ 73 (0xFULL << AMD64_L3_SLICE_SHIFT) 74 #define AMD64_L3_SLICEID_MASK \ 75 (0x7ULL << AMD64_L3_SLICE_SHIFT) 76 77 #define AMD64_L3_THREAD_SHIFT 56 78 #define AMD64_L3_THREAD_MASK \ 79 (0xFFULL << AMD64_L3_THREAD_SHIFT) 80 #define AMD64_L3_F19H_THREAD_MASK \ 81 (0x3ULL << AMD64_L3_THREAD_SHIFT) 82 83 #define AMD64_L3_EN_ALL_CORES BIT_ULL(47) 84 #define AMD64_L3_EN_ALL_SLICES BIT_ULL(46) 85 86 #define AMD64_L3_COREID_SHIFT 42 87 #define AMD64_L3_COREID_MASK \ 88 (0x7ULL << AMD64_L3_COREID_SHIFT) 89 90 #define X86_RAW_EVENT_MASK \ 91 (ARCH_PERFMON_EVENTSEL_EVENT | \ 92 ARCH_PERFMON_EVENTSEL_UMASK | \ 93 ARCH_PERFMON_EVENTSEL_EDGE | \ 94 ARCH_PERFMON_EVENTSEL_INV | \ 95 ARCH_PERFMON_EVENTSEL_CMASK) 96 #define X86_ALL_EVENT_FLAGS \ 97 (ARCH_PERFMON_EVENTSEL_EDGE | \ 98 ARCH_PERFMON_EVENTSEL_INV | \ 99 ARCH_PERFMON_EVENTSEL_CMASK | \ 100 ARCH_PERFMON_EVENTSEL_ANY | \ 101 ARCH_PERFMON_EVENTSEL_PIN_CONTROL | \ 102 HSW_IN_TX | \ 103 HSW_IN_TX_CHECKPOINTED) 104 #define AMD64_RAW_EVENT_MASK \ 105 (X86_RAW_EVENT_MASK | \ 106 AMD64_EVENTSEL_EVENT) 107 #define AMD64_RAW_EVENT_MASK_NB \ 108 (AMD64_EVENTSEL_EVENT | \ 109 ARCH_PERFMON_EVENTSEL_UMASK) 110 111 #define AMD64_PERFMON_V2_EVENTSEL_EVENT_NB \ 112 (AMD64_EVENTSEL_EVENT | \ 113 GENMASK_ULL(37, 36)) 114 115 #define AMD64_PERFMON_V2_EVENTSEL_UMASK_NB \ 116 (ARCH_PERFMON_EVENTSEL_UMASK | \ 117 GENMASK_ULL(27, 24)) 118 119 #define AMD64_PERFMON_V2_RAW_EVENT_MASK_NB \ 120 (AMD64_PERFMON_V2_EVENTSEL_EVENT_NB | \ 121 AMD64_PERFMON_V2_EVENTSEL_UMASK_NB) 122 123 #define AMD64_PERFMON_V2_ENABLE_UMC BIT_ULL(31) 124 #define AMD64_PERFMON_V2_EVENTSEL_EVENT_UMC GENMASK_ULL(7, 0) 125 #define AMD64_PERFMON_V2_EVENTSEL_RDWRMASK_UMC GENMASK_ULL(9, 8) 126 #define AMD64_PERFMON_V2_RAW_EVENT_MASK_UMC \ 127 (AMD64_PERFMON_V2_EVENTSEL_EVENT_UMC | \ 128 AMD64_PERFMON_V2_EVENTSEL_RDWRMASK_UMC) 129 130 #define AMD64_NUM_COUNTERS 4 131 #define AMD64_NUM_COUNTERS_CORE 6 132 #define AMD64_NUM_COUNTERS_NB 4 133 134 #define ARCH_PERFMON_UNHALTED_CORE_CYCLES_SEL 0x3c 135 #define ARCH_PERFMON_UNHALTED_CORE_CYCLES_UMASK (0x00 << 8) 136 #define ARCH_PERFMON_UNHALTED_CORE_CYCLES_INDEX 0 137 #define ARCH_PERFMON_UNHALTED_CORE_CYCLES_PRESENT \ 138 (1 << (ARCH_PERFMON_UNHALTED_CORE_CYCLES_INDEX)) 139 140 #define ARCH_PERFMON_BRANCH_MISSES_RETIRED 6 141 #define ARCH_PERFMON_EVENTS_COUNT 7 142 143 #define PEBS_DATACFG_MEMINFO BIT_ULL(0) 144 #define PEBS_DATACFG_GP BIT_ULL(1) 145 #define PEBS_DATACFG_XMMS BIT_ULL(2) 146 #define PEBS_DATACFG_LBRS BIT_ULL(3) 147 #define PEBS_DATACFG_CNTR BIT_ULL(4) 148 #define PEBS_DATACFG_METRICS BIT_ULL(5) 149 #define PEBS_DATACFG_LBR_SHIFT 24 150 #define PEBS_DATACFG_CNTR_SHIFT 32 151 #define PEBS_DATACFG_CNTR_MASK GENMASK_ULL(15, 0) 152 #define PEBS_DATACFG_FIX_SHIFT 48 153 #define PEBS_DATACFG_FIX_MASK GENMASK_ULL(7, 0) 154 155 /* Steal the highest bit of pebs_data_cfg for SW usage */ 156 #define PEBS_UPDATE_DS_SW BIT_ULL(63) 157 158 /* 159 * Intel "Architectural Performance Monitoring" CPUID 160 * detection/enumeration details: 161 */ 162 union cpuid10_eax { 163 struct { 164 unsigned int version_id:8; 165 unsigned int num_counters:8; 166 unsigned int bit_width:8; 167 unsigned int mask_length:8; 168 } split; 169 unsigned int full; 170 }; 171 172 union cpuid10_ebx { 173 struct { 174 unsigned int no_unhalted_core_cycles:1; 175 unsigned int no_instructions_retired:1; 176 unsigned int no_unhalted_reference_cycles:1; 177 unsigned int no_llc_reference:1; 178 unsigned int no_llc_misses:1; 179 unsigned int no_branch_instruction_retired:1; 180 unsigned int no_branch_misses_retired:1; 181 } split; 182 unsigned int full; 183 }; 184 185 union cpuid10_edx { 186 struct { 187 unsigned int num_counters_fixed:5; 188 unsigned int bit_width_fixed:8; 189 unsigned int reserved1:2; 190 unsigned int anythread_deprecated:1; 191 unsigned int reserved2:16; 192 } split; 193 unsigned int full; 194 }; 195 196 /* 197 * Intel "Architectural Performance Monitoring extension" CPUID 198 * detection/enumeration details: 199 */ 200 #define ARCH_PERFMON_EXT_LEAF 0x00000023 201 #define ARCH_PERFMON_NUM_COUNTER_LEAF 0x1 202 #define ARCH_PERFMON_ACR_LEAF 0x2 203 #define ARCH_PERFMON_PEBS_CAP_LEAF 0x4 204 #define ARCH_PERFMON_PEBS_COUNTER_LEAF 0x5 205 206 union cpuid35_eax { 207 struct { 208 unsigned int leaf0:1; 209 /* Counters Sub-Leaf */ 210 unsigned int cntr_subleaf:1; 211 /* Auto Counter Reload Sub-Leaf */ 212 unsigned int acr_subleaf:1; 213 /* Events Sub-Leaf */ 214 unsigned int events_subleaf:1; 215 /* arch-PEBS Sub-Leaves */ 216 unsigned int pebs_caps_subleaf:1; 217 unsigned int pebs_cnts_subleaf:1; 218 unsigned int reserved:26; 219 } split; 220 unsigned int full; 221 }; 222 223 union cpuid35_ebx { 224 struct { 225 /* UnitMask2 Supported */ 226 unsigned int umask2:1; 227 /* EQ-bit Supported */ 228 unsigned int eq:1; 229 unsigned int reserved:30; 230 } split; 231 unsigned int full; 232 }; 233 234 /* 235 * Intel Architectural LBR CPUID detection/enumeration details: 236 */ 237 union cpuid28_eax { 238 struct { 239 /* Supported LBR depth values */ 240 unsigned int lbr_depth_mask:8; 241 unsigned int reserved:22; 242 /* Deep C-state Reset */ 243 unsigned int lbr_deep_c_reset:1; 244 /* IP values contain LIP */ 245 unsigned int lbr_lip:1; 246 } split; 247 unsigned int full; 248 }; 249 250 union cpuid28_ebx { 251 struct { 252 /* CPL Filtering Supported */ 253 unsigned int lbr_cpl:1; 254 /* Branch Filtering Supported */ 255 unsigned int lbr_filter:1; 256 /* Call-stack Mode Supported */ 257 unsigned int lbr_call_stack:1; 258 } split; 259 unsigned int full; 260 }; 261 262 union cpuid28_ecx { 263 struct { 264 /* Mispredict Bit Supported */ 265 unsigned int lbr_mispred:1; 266 /* Timed LBRs Supported */ 267 unsigned int lbr_timed_lbr:1; 268 /* Branch Type Field Supported */ 269 unsigned int lbr_br_type:1; 270 unsigned int reserved:13; 271 /* Branch counters (Event Logging) Supported */ 272 unsigned int lbr_counters:4; 273 } split; 274 unsigned int full; 275 }; 276 277 /* 278 * AMD "Extended Performance Monitoring and Debug" CPUID 279 * detection/enumeration details: 280 */ 281 union cpuid_0x80000022_ebx { 282 struct { 283 /* Number of Core Performance Counters */ 284 unsigned int num_core_pmc:4; 285 /* Number of available LBR Stack Entries */ 286 unsigned int lbr_v2_stack_sz:6; 287 /* Number of Data Fabric Counters */ 288 unsigned int num_df_pmc:6; 289 /* Number of Unified Memory Controller Counters */ 290 unsigned int num_umc_pmc:6; 291 } split; 292 unsigned int full; 293 }; 294 295 struct x86_pmu_capability { 296 int version; 297 int num_counters_gp; 298 int num_counters_fixed; 299 int bit_width_gp; 300 int bit_width_fixed; 301 unsigned int events_mask; 302 int events_mask_len; 303 unsigned int pebs_ept :1; 304 unsigned int mediated :1; 305 }; 306 307 /* 308 * Fixed-purpose performance events: 309 */ 310 311 /* RDPMC offset for Fixed PMCs */ 312 #define INTEL_PMC_FIXED_RDPMC_BASE (1 << 30) 313 #define INTEL_PMC_FIXED_RDPMC_METRICS (1 << 29) 314 315 /* 316 * All the fixed-mode PMCs are configured via this single MSR: 317 */ 318 #define MSR_ARCH_PERFMON_FIXED_CTR_CTRL 0x38d 319 320 /* 321 * There is no event-code assigned to the fixed-mode PMCs. 322 * 323 * For a fixed-mode PMC, which has an equivalent event on a general-purpose 324 * PMC, the event-code of the equivalent event is used for the fixed-mode PMC, 325 * e.g., Instr_Retired.Any and CPU_CLK_Unhalted.Core. 326 * 327 * For a fixed-mode PMC, which doesn't have an equivalent event, a 328 * pseudo-encoding is used, e.g., CPU_CLK_Unhalted.Ref and TOPDOWN.SLOTS. 329 * The pseudo event-code for a fixed-mode PMC must be 0x00. 330 * The pseudo umask-code is 0xX. The X equals the index of the fixed 331 * counter + 1, e.g., the fixed counter 2 has the pseudo-encoding 0x0300. 332 * 333 * The counts are available in separate MSRs: 334 */ 335 336 /* Instr_Retired.Any: */ 337 #define MSR_ARCH_PERFMON_FIXED_CTR0 0x309 338 #define INTEL_PMC_IDX_FIXED_INSTRUCTIONS (INTEL_PMC_IDX_FIXED + 0) 339 340 /* CPU_CLK_Unhalted.Core: */ 341 #define MSR_ARCH_PERFMON_FIXED_CTR1 0x30a 342 #define INTEL_PMC_IDX_FIXED_CPU_CYCLES (INTEL_PMC_IDX_FIXED + 1) 343 344 /* CPU_CLK_Unhalted.Ref: event=0x00,umask=0x3 (pseudo-encoding) */ 345 #define MSR_ARCH_PERFMON_FIXED_CTR2 0x30b 346 #define INTEL_PMC_IDX_FIXED_REF_CYCLES (INTEL_PMC_IDX_FIXED + 2) 347 #define INTEL_PMC_MSK_FIXED_REF_CYCLES (1ULL << INTEL_PMC_IDX_FIXED_REF_CYCLES) 348 349 /* TOPDOWN.SLOTS: event=0x00,umask=0x4 (pseudo-encoding) */ 350 #define MSR_ARCH_PERFMON_FIXED_CTR3 0x30c 351 #define INTEL_PMC_IDX_FIXED_SLOTS (INTEL_PMC_IDX_FIXED + 3) 352 #define INTEL_PMC_MSK_FIXED_SLOTS (1ULL << INTEL_PMC_IDX_FIXED_SLOTS) 353 354 /* TOPDOWN_BAD_SPECULATION.ALL: fixed counter 4 (Atom only) */ 355 /* TOPDOWN_FE_BOUND.ALL: fixed counter 5 (Atom only) */ 356 /* TOPDOWN_RETIRING.ALL: fixed counter 6 (Atom only) */ 357 358 static inline bool use_fixed_pseudo_encoding(u64 code) 359 { 360 return !(code & 0xff); 361 } 362 363 /* 364 * We model BTS tracing as another fixed-mode PMC. 365 * 366 * We choose the value 47 for the fixed index of BTS, since lower 367 * values are used by actual fixed events and higher values are used 368 * to indicate other overflow conditions in the PERF_GLOBAL_STATUS msr. 369 */ 370 #define INTEL_PMC_IDX_FIXED_BTS (INTEL_PMC_IDX_FIXED + 15) 371 372 /* 373 * The PERF_METRICS MSR is modeled as several magic fixed-mode PMCs, one for 374 * each TopDown metric event. 375 * 376 * Internally the TopDown metric events are mapped to the FxCtr 3 (SLOTS). 377 */ 378 #define INTEL_PMC_IDX_METRIC_BASE (INTEL_PMC_IDX_FIXED + 16) 379 #define INTEL_PMC_IDX_TD_RETIRING (INTEL_PMC_IDX_METRIC_BASE + 0) 380 #define INTEL_PMC_IDX_TD_BAD_SPEC (INTEL_PMC_IDX_METRIC_BASE + 1) 381 #define INTEL_PMC_IDX_TD_FE_BOUND (INTEL_PMC_IDX_METRIC_BASE + 2) 382 #define INTEL_PMC_IDX_TD_BE_BOUND (INTEL_PMC_IDX_METRIC_BASE + 3) 383 #define INTEL_PMC_IDX_TD_HEAVY_OPS (INTEL_PMC_IDX_METRIC_BASE + 4) 384 #define INTEL_PMC_IDX_TD_BR_MISPREDICT (INTEL_PMC_IDX_METRIC_BASE + 5) 385 #define INTEL_PMC_IDX_TD_FETCH_LAT (INTEL_PMC_IDX_METRIC_BASE + 6) 386 #define INTEL_PMC_IDX_TD_MEM_BOUND (INTEL_PMC_IDX_METRIC_BASE + 7) 387 #define INTEL_PMC_IDX_METRIC_END INTEL_PMC_IDX_TD_MEM_BOUND 388 #define INTEL_PMC_MSK_TOPDOWN ((0xffull << INTEL_PMC_IDX_METRIC_BASE) | \ 389 INTEL_PMC_MSK_FIXED_SLOTS) 390 391 /* 392 * There is no event-code assigned to the TopDown events. 393 * 394 * For the slots event, use the pseudo code of the fixed counter 3. 395 * 396 * For the metric events, the pseudo event-code is 0x00. 397 * The pseudo umask-code starts from the middle of the pseudo event 398 * space, 0x80. 399 */ 400 #define INTEL_TD_SLOTS 0x0400 /* TOPDOWN.SLOTS */ 401 /* Level 1 metrics */ 402 #define INTEL_TD_METRIC_RETIRING 0x8000 /* Retiring metric */ 403 #define INTEL_TD_METRIC_BAD_SPEC 0x8100 /* Bad speculation metric */ 404 #define INTEL_TD_METRIC_FE_BOUND 0x8200 /* FE bound metric */ 405 #define INTEL_TD_METRIC_BE_BOUND 0x8300 /* BE bound metric */ 406 /* Level 2 metrics */ 407 #define INTEL_TD_METRIC_HEAVY_OPS 0x8400 /* Heavy Operations metric */ 408 #define INTEL_TD_METRIC_BR_MISPREDICT 0x8500 /* Branch Mispredict metric */ 409 #define INTEL_TD_METRIC_FETCH_LAT 0x8600 /* Fetch Latency metric */ 410 #define INTEL_TD_METRIC_MEM_BOUND 0x8700 /* Memory bound metric */ 411 412 #define INTEL_TD_METRIC_MAX INTEL_TD_METRIC_MEM_BOUND 413 #define INTEL_TD_METRIC_NUM 8 414 415 #define INTEL_TD_CFG_METRIC_CLEAR_BIT 0 416 #define INTEL_TD_CFG_METRIC_CLEAR BIT_ULL(INTEL_TD_CFG_METRIC_CLEAR_BIT) 417 418 static inline bool is_metric_idx(int idx) 419 { 420 return (unsigned)(idx - INTEL_PMC_IDX_METRIC_BASE) < INTEL_TD_METRIC_NUM; 421 } 422 423 static inline bool is_topdown_idx(int idx) 424 { 425 return is_metric_idx(idx) || idx == INTEL_PMC_IDX_FIXED_SLOTS; 426 } 427 428 #define INTEL_PMC_OTHER_TOPDOWN_BITS(bit) \ 429 (~(0x1ull << bit) & INTEL_PMC_MSK_TOPDOWN) 430 431 #define GLOBAL_STATUS_COND_CHG BIT_ULL(63) 432 #define GLOBAL_STATUS_BUFFER_OVF_BIT 62 433 #define GLOBAL_STATUS_BUFFER_OVF BIT_ULL(GLOBAL_STATUS_BUFFER_OVF_BIT) 434 #define GLOBAL_STATUS_UNC_OVF BIT_ULL(61) 435 #define GLOBAL_STATUS_ASIF BIT_ULL(60) 436 #define GLOBAL_STATUS_COUNTERS_FROZEN BIT_ULL(59) 437 #define GLOBAL_STATUS_LBRS_FROZEN_BIT 58 438 #define GLOBAL_STATUS_LBRS_FROZEN BIT_ULL(GLOBAL_STATUS_LBRS_FROZEN_BIT) 439 #define GLOBAL_STATUS_TRACE_TOPAPMI_BIT 55 440 #define GLOBAL_STATUS_TRACE_TOPAPMI BIT_ULL(GLOBAL_STATUS_TRACE_TOPAPMI_BIT) 441 #define GLOBAL_STATUS_ARCH_PEBS_THRESHOLD_BIT 54 442 #define GLOBAL_STATUS_ARCH_PEBS_THRESHOLD BIT_ULL(GLOBAL_STATUS_ARCH_PEBS_THRESHOLD_BIT) 443 #define GLOBAL_STATUS_PERF_METRICS_OVF_BIT 48 444 445 #define GLOBAL_CTRL_EN_PERF_METRICS BIT_ULL(48) 446 /* 447 * We model guest LBR event tracing as another fixed-mode PMC like BTS. 448 * 449 * We choose bit 58 because it's used to indicate LBR stack frozen state 450 * for architectural perfmon v4, also we unconditionally mask that bit in 451 * the handle_pmi_common(), so it'll never be set in the overflow handling. 452 * 453 * With this fake counter assigned, the guest LBR event user (such as KVM), 454 * can program the LBR registers on its own, and we don't actually do anything 455 * with then in the host context. 456 */ 457 #define INTEL_PMC_IDX_FIXED_VLBR (GLOBAL_STATUS_LBRS_FROZEN_BIT) 458 459 /* 460 * Pseudo-encoding the guest LBR event as event=0x00,umask=0x1b, 461 * since it would claim bit 58 which is effectively Fixed26. 462 */ 463 #define INTEL_FIXED_VLBR_EVENT 0x1b00 464 465 /* 466 * Adaptive PEBS v4 467 */ 468 469 struct pebs_basic { 470 u64 format_group:32, 471 retire_latency:16, 472 format_size:16; 473 u64 ip; 474 u64 applicable_counters; 475 u64 tsc; 476 }; 477 478 struct pebs_meminfo { 479 u64 address; 480 u64 aux; 481 union { 482 /* pre Alder Lake */ 483 u64 mem_latency; 484 /* Alder Lake and later */ 485 struct { 486 u64 instr_latency:16; 487 u64 pad2:16; 488 u64 cache_latency:16; 489 u64 pad3:16; 490 }; 491 }; 492 u64 tsx_tuning; 493 }; 494 495 struct pebs_gprs { 496 u64 flags, ip, ax, cx, dx, bx, sp, bp, si, di; 497 u64 r8, r9, r10, r11, r12, r13, r14, r15; 498 }; 499 500 struct pebs_xmm { 501 u64 xmm[16*2]; /* two entries for each register */ 502 }; 503 504 struct pebs_cntr_header { 505 u32 cntr; 506 u32 fixed; 507 u32 metrics; 508 u32 reserved; 509 }; 510 511 #define INTEL_CNTR_METRICS 0x3 512 513 /* 514 * Arch PEBS 515 */ 516 union arch_pebs_index { 517 struct { 518 u64 rsvd:4, 519 wr:23, 520 rsvd2:4, 521 full:1, 522 en:1, 523 rsvd3:3, 524 thresh:23, 525 rsvd4:5; 526 }; 527 u64 whole; 528 }; 529 530 struct arch_pebs_header { 531 union { 532 u64 format; 533 struct { 534 u64 size:16, /* Record size */ 535 rsvd:14, 536 mode:1, /* 64BIT_MODE */ 537 cont:1, 538 rsvd2:3, 539 cntr:5, 540 lbr:2, 541 rsvd3:7, 542 xmm:1, 543 ymmh:1, 544 rsvd4:2, 545 opmask:1, 546 zmmh:1, 547 h16zmm:1, 548 rsvd5:5, 549 gpr:1, 550 aux:1, 551 basic:1; 552 }; 553 }; 554 u64 rsvd6; 555 }; 556 557 struct arch_pebs_basic { 558 u64 ip; 559 u64 applicable_counters; 560 u64 tsc; 561 u64 retire :16, /* Retire Latency */ 562 valid :1, 563 rsvd :47; 564 u64 rsvd2; 565 u64 rsvd3; 566 }; 567 568 struct arch_pebs_aux { 569 u64 address; 570 u64 rsvd; 571 u64 rsvd2; 572 u64 rsvd3; 573 u64 rsvd4; 574 u64 aux; 575 u64 instr_latency :16, 576 pad2 :16, 577 cache_latency :16, 578 pad3 :16; 579 u64 tsx_tuning; 580 }; 581 582 struct arch_pebs_gprs { 583 u64 flags, ip, ax, cx, dx, bx, sp, bp, si, di; 584 u64 r8, r9, r10, r11, r12, r13, r14, r15, ssp; 585 u64 rsvd; 586 }; 587 588 struct arch_pebs_xer_header { 589 u64 xstate; 590 u64 rsvd; 591 }; 592 593 #define ARCH_PEBS_LBR_NAN 0x0 594 #define ARCH_PEBS_LBR_NUM_8 0x1 595 #define ARCH_PEBS_LBR_NUM_16 0x2 596 #define ARCH_PEBS_LBR_NUM_VAR 0x3 597 #define ARCH_PEBS_BASE_LBR_ENTRIES 8 598 struct arch_pebs_lbr_header { 599 u64 rsvd; 600 u64 ctl; 601 u64 depth; 602 u64 ler_from; 603 u64 ler_to; 604 u64 ler_info; 605 }; 606 607 struct arch_pebs_cntr_header { 608 u32 cntr; 609 u32 fixed; 610 u32 metrics; 611 u32 reserved; 612 }; 613 614 /* 615 * AMD Extended Performance Monitoring and Debug cpuid feature detection 616 */ 617 #define EXT_PERFMON_DEBUG_FEATURES 0x80000022 618 619 /* 620 * IBS cpuid feature detection 621 */ 622 623 #define IBS_CPUID_FEATURES 0x8000001b 624 625 /* 626 * Same bit mask as for IBS cpuid feature flags (Fn8000_001B_EAX), but 627 * bit 0 is used to indicate the existence of IBS. 628 */ 629 #define IBS_CAPS_AVAIL (1U<<0) 630 #define IBS_CAPS_FETCHSAM (1U<<1) 631 #define IBS_CAPS_OPSAM (1U<<2) 632 #define IBS_CAPS_RDWROPCNT (1U<<3) 633 #define IBS_CAPS_OPCNT (1U<<4) 634 #define IBS_CAPS_BRNTRGT (1U<<5) 635 #define IBS_CAPS_OPCNTEXT (1U<<6) 636 #define IBS_CAPS_RIPINVALIDCHK (1U<<7) 637 #define IBS_CAPS_OPBRNFUSE (1U<<8) 638 #define IBS_CAPS_FETCHCTLEXTD (1U<<9) 639 #define IBS_CAPS_OPDATA4 (1U<<10) 640 #define IBS_CAPS_ZEN4 (1U<<11) 641 #define IBS_CAPS_OPLDLAT (1U<<12) 642 #define IBS_CAPS_OPDTLBPGSIZE (1U<<19) 643 644 #define IBS_CAPS_DEFAULT (IBS_CAPS_AVAIL \ 645 | IBS_CAPS_FETCHSAM \ 646 | IBS_CAPS_OPSAM) 647 648 /* 649 * IBS APIC setup 650 */ 651 #define IBSCTL 0x1cc 652 #define IBSCTL_LVT_OFFSET_VALID (1ULL<<8) 653 #define IBSCTL_LVT_OFFSET_MASK 0x0F 654 655 /* IBS fetch bits/masks */ 656 #define IBS_FETCH_L3MISSONLY (1ULL<<59) 657 #define IBS_FETCH_RAND_EN (1ULL<<57) 658 #define IBS_FETCH_VAL (1ULL<<49) 659 #define IBS_FETCH_ENABLE (1ULL<<48) 660 #define IBS_FETCH_CNT 0xFFFF0000ULL 661 #define IBS_FETCH_MAX_CNT 0x0000FFFFULL 662 663 /* 664 * IBS op bits/masks 665 * The lower 7 bits of the current count are random bits 666 * preloaded by hardware and ignored in software 667 */ 668 #define IBS_OP_LDLAT_EN (1ULL<<63) 669 #define IBS_OP_LDLAT_THRSH (0xFULL<<59) 670 #define IBS_OP_CUR_CNT (0xFFF80ULL<<32) 671 #define IBS_OP_CUR_CNT_RAND (0x0007FULL<<32) 672 #define IBS_OP_CUR_CNT_EXT_MASK (0x7FULL<<52) 673 #define IBS_OP_CNT_CTL (1ULL<<19) 674 #define IBS_OP_VAL (1ULL<<18) 675 #define IBS_OP_ENABLE (1ULL<<17) 676 #define IBS_OP_L3MISSONLY (1ULL<<16) 677 #define IBS_OP_MAX_CNT 0x0000FFFFULL 678 #define IBS_OP_MAX_CNT_EXT 0x007FFFFFULL /* not a register bit mask */ 679 #define IBS_OP_MAX_CNT_EXT_MASK (0x7FULL<<20) /* separate upper 7 bits */ 680 #define IBS_RIP_INVALID (1ULL<<38) 681 682 #ifdef CONFIG_X86_LOCAL_APIC 683 extern u32 get_ibs_caps(void); 684 extern int forward_event_to_ibs(struct perf_event *event); 685 #else 686 static inline u32 get_ibs_caps(void) { return 0; } 687 static inline int forward_event_to_ibs(struct perf_event *event) { return -ENOENT; } 688 #endif 689 690 #ifdef CONFIG_PERF_EVENTS 691 extern void perf_events_lapic_init(void); 692 693 /* 694 * Abuse bits {3,5} of the cpu eflags register. These flags are otherwise 695 * unused and ABI specified to be 0, so nobody should care what we do with 696 * them. 697 * 698 * EXACT - the IP points to the exact instruction that triggered the 699 * event (HW bugs exempt). 700 * VM - original X86_VM_MASK; see set_linear_ip(). 701 */ 702 #define PERF_EFLAGS_EXACT (1UL << 3) 703 #define PERF_EFLAGS_VM (1UL << 5) 704 705 struct pt_regs; 706 struct x86_perf_regs { 707 struct pt_regs regs; 708 u64 *xmm_regs; 709 }; 710 711 extern unsigned long perf_arch_instruction_pointer(struct pt_regs *regs); 712 extern unsigned long perf_arch_misc_flags(struct pt_regs *regs); 713 extern unsigned long perf_arch_guest_misc_flags(struct pt_regs *regs); 714 #define perf_arch_misc_flags(regs) perf_arch_misc_flags(regs) 715 #define perf_arch_guest_misc_flags(regs) perf_arch_guest_misc_flags(regs) 716 717 #include <asm/stacktrace.h> 718 719 /* 720 * We abuse bit 3 from flags to pass exact information, see 721 * perf_arch_misc_flags() and the comment with PERF_EFLAGS_EXACT. 722 */ 723 #define perf_arch_fetch_caller_regs(regs, __ip) { \ 724 (regs)->ip = (__ip); \ 725 (regs)->sp = (unsigned long)__builtin_frame_address(0); \ 726 (regs)->cs = __KERNEL_CS; \ 727 regs->flags = 0; \ 728 } 729 730 struct perf_guest_switch_msr { 731 unsigned msr; 732 u64 host, guest; 733 }; 734 735 struct x86_pmu_lbr { 736 unsigned int nr; 737 unsigned int from; 738 unsigned int to; 739 unsigned int info; 740 bool has_callstack; 741 }; 742 743 extern void perf_get_x86_pmu_capability(struct x86_pmu_capability *cap); 744 extern u64 perf_get_hw_event_config(int hw_event); 745 extern void perf_check_microcode(void); 746 extern void perf_clear_dirty_counters(void); 747 extern int x86_perf_rdpmc_index(struct perf_event *event); 748 #else 749 static inline void perf_get_x86_pmu_capability(struct x86_pmu_capability *cap) 750 { 751 memset(cap, 0, sizeof(*cap)); 752 } 753 754 static inline u64 perf_get_hw_event_config(int hw_event) 755 { 756 return 0; 757 } 758 759 static inline void perf_events_lapic_init(void) { } 760 static inline void perf_check_microcode(void) { } 761 #endif 762 763 #ifdef CONFIG_PERF_GUEST_MEDIATED_PMU 764 extern void perf_load_guest_lvtpc(u32 guest_lvtpc); 765 extern void perf_put_guest_lvtpc(void); 766 #endif 767 768 #if defined(CONFIG_PERF_EVENTS) && defined(CONFIG_CPU_SUP_INTEL) 769 extern struct perf_guest_switch_msr *perf_guest_get_msrs(int *nr, void *data); 770 extern void x86_perf_get_lbr(struct x86_pmu_lbr *lbr); 771 #else 772 struct perf_guest_switch_msr *perf_guest_get_msrs(int *nr, void *data); 773 static inline void x86_perf_get_lbr(struct x86_pmu_lbr *lbr) 774 { 775 memset(lbr, 0, sizeof(*lbr)); 776 } 777 #endif 778 779 #ifdef CONFIG_CPU_SUP_INTEL 780 extern void intel_pt_handle_vmx(int on); 781 #else 782 static inline void intel_pt_handle_vmx(int on) 783 { 784 785 } 786 #endif 787 788 #if defined(CONFIG_PERF_EVENTS) && defined(CONFIG_CPU_SUP_AMD) 789 extern void amd_pmu_enable_virt(void); 790 extern void amd_pmu_disable_virt(void); 791 792 #if defined(CONFIG_PERF_EVENTS_AMD_BRS) 793 794 #define PERF_NEEDS_LOPWR_CB 1 795 796 /* 797 * architectural low power callback impacts 798 * drivers/acpi/processor_idle.c 799 * drivers/acpi/acpi_pad.c 800 */ 801 extern void perf_amd_brs_lopwr_cb(bool lopwr_in); 802 803 DECLARE_STATIC_CALL(perf_lopwr_cb, perf_amd_brs_lopwr_cb); 804 805 static __always_inline void perf_lopwr_cb(bool lopwr_in) 806 { 807 static_call_mod(perf_lopwr_cb)(lopwr_in); 808 } 809 810 #endif /* PERF_NEEDS_LOPWR_CB */ 811 812 #else 813 static inline void amd_pmu_enable_virt(void) { } 814 static inline void amd_pmu_disable_virt(void) { } 815 #endif 816 817 #define arch_perf_out_copy_user copy_from_user_nmi 818 819 #endif /* _ASM_X86_PERF_EVENT_H */ 820