1 /* SPDX-License-Identifier: GPL-2.0 */ 2 #ifndef _ASM_X86_PERF_EVENT_H 3 #define _ASM_X86_PERF_EVENT_H 4 5 /* 6 * Performance event hw details: 7 */ 8 9 #define INTEL_PMC_MAX_GENERIC 32 10 #define INTEL_PMC_MAX_FIXED 4 11 #define INTEL_PMC_IDX_FIXED 32 12 13 #define X86_PMC_IDX_MAX 64 14 15 #define MSR_ARCH_PERFMON_PERFCTR0 0xc1 16 #define MSR_ARCH_PERFMON_PERFCTR1 0xc2 17 18 #define MSR_ARCH_PERFMON_EVENTSEL0 0x186 19 #define MSR_ARCH_PERFMON_EVENTSEL1 0x187 20 21 #define ARCH_PERFMON_EVENTSEL_EVENT 0x000000FFULL 22 #define ARCH_PERFMON_EVENTSEL_UMASK 0x0000FF00ULL 23 #define ARCH_PERFMON_EVENTSEL_USR (1ULL << 16) 24 #define ARCH_PERFMON_EVENTSEL_OS (1ULL << 17) 25 #define ARCH_PERFMON_EVENTSEL_EDGE (1ULL << 18) 26 #define ARCH_PERFMON_EVENTSEL_PIN_CONTROL (1ULL << 19) 27 #define ARCH_PERFMON_EVENTSEL_INT (1ULL << 20) 28 #define ARCH_PERFMON_EVENTSEL_ANY (1ULL << 21) 29 #define ARCH_PERFMON_EVENTSEL_ENABLE (1ULL << 22) 30 #define ARCH_PERFMON_EVENTSEL_INV (1ULL << 23) 31 #define ARCH_PERFMON_EVENTSEL_CMASK 0xFF000000ULL 32 33 #define HSW_IN_TX (1ULL << 32) 34 #define HSW_IN_TX_CHECKPOINTED (1ULL << 33) 35 #define ICL_EVENTSEL_ADAPTIVE (1ULL << 34) 36 #define ICL_FIXED_0_ADAPTIVE (1ULL << 32) 37 38 #define AMD64_EVENTSEL_INT_CORE_ENABLE (1ULL << 36) 39 #define AMD64_EVENTSEL_GUESTONLY (1ULL << 40) 40 #define AMD64_EVENTSEL_HOSTONLY (1ULL << 41) 41 42 #define AMD64_EVENTSEL_INT_CORE_SEL_SHIFT 37 43 #define AMD64_EVENTSEL_INT_CORE_SEL_MASK \ 44 (0xFULL << AMD64_EVENTSEL_INT_CORE_SEL_SHIFT) 45 46 #define AMD64_EVENTSEL_EVENT \ 47 (ARCH_PERFMON_EVENTSEL_EVENT | (0x0FULL << 32)) 48 #define INTEL_ARCH_EVENT_MASK \ 49 (ARCH_PERFMON_EVENTSEL_UMASK | ARCH_PERFMON_EVENTSEL_EVENT) 50 51 #define AMD64_L3_SLICE_SHIFT 48 52 #define AMD64_L3_SLICE_MASK \ 53 (0xFULL << AMD64_L3_SLICE_SHIFT) 54 #define AMD64_L3_SLICEID_MASK \ 55 (0x7ULL << AMD64_L3_SLICE_SHIFT) 56 57 #define AMD64_L3_THREAD_SHIFT 56 58 #define AMD64_L3_THREAD_MASK \ 59 (0xFFULL << AMD64_L3_THREAD_SHIFT) 60 #define AMD64_L3_F19H_THREAD_MASK \ 61 (0x3ULL << AMD64_L3_THREAD_SHIFT) 62 63 #define AMD64_L3_EN_ALL_CORES BIT_ULL(47) 64 #define AMD64_L3_EN_ALL_SLICES BIT_ULL(46) 65 66 #define AMD64_L3_COREID_SHIFT 42 67 #define AMD64_L3_COREID_MASK \ 68 (0x7ULL << AMD64_L3_COREID_SHIFT) 69 70 #define X86_RAW_EVENT_MASK \ 71 (ARCH_PERFMON_EVENTSEL_EVENT | \ 72 ARCH_PERFMON_EVENTSEL_UMASK | \ 73 ARCH_PERFMON_EVENTSEL_EDGE | \ 74 ARCH_PERFMON_EVENTSEL_INV | \ 75 ARCH_PERFMON_EVENTSEL_CMASK) 76 #define X86_ALL_EVENT_FLAGS \ 77 (ARCH_PERFMON_EVENTSEL_EDGE | \ 78 ARCH_PERFMON_EVENTSEL_INV | \ 79 ARCH_PERFMON_EVENTSEL_CMASK | \ 80 ARCH_PERFMON_EVENTSEL_ANY | \ 81 ARCH_PERFMON_EVENTSEL_PIN_CONTROL | \ 82 HSW_IN_TX | \ 83 HSW_IN_TX_CHECKPOINTED) 84 #define AMD64_RAW_EVENT_MASK \ 85 (X86_RAW_EVENT_MASK | \ 86 AMD64_EVENTSEL_EVENT) 87 #define AMD64_RAW_EVENT_MASK_NB \ 88 (AMD64_EVENTSEL_EVENT | \ 89 ARCH_PERFMON_EVENTSEL_UMASK) 90 #define AMD64_NUM_COUNTERS 4 91 #define AMD64_NUM_COUNTERS_CORE 6 92 #define AMD64_NUM_COUNTERS_NB 4 93 94 #define ARCH_PERFMON_UNHALTED_CORE_CYCLES_SEL 0x3c 95 #define ARCH_PERFMON_UNHALTED_CORE_CYCLES_UMASK (0x00 << 8) 96 #define ARCH_PERFMON_UNHALTED_CORE_CYCLES_INDEX 0 97 #define ARCH_PERFMON_UNHALTED_CORE_CYCLES_PRESENT \ 98 (1 << (ARCH_PERFMON_UNHALTED_CORE_CYCLES_INDEX)) 99 100 #define ARCH_PERFMON_BRANCH_MISSES_RETIRED 6 101 #define ARCH_PERFMON_EVENTS_COUNT 7 102 103 #define PEBS_DATACFG_MEMINFO BIT_ULL(0) 104 #define PEBS_DATACFG_GP BIT_ULL(1) 105 #define PEBS_DATACFG_XMMS BIT_ULL(2) 106 #define PEBS_DATACFG_LBRS BIT_ULL(3) 107 #define PEBS_DATACFG_LBR_SHIFT 24 108 109 /* 110 * Intel "Architectural Performance Monitoring" CPUID 111 * detection/enumeration details: 112 */ 113 union cpuid10_eax { 114 struct { 115 unsigned int version_id:8; 116 unsigned int num_counters:8; 117 unsigned int bit_width:8; 118 unsigned int mask_length:8; 119 } split; 120 unsigned int full; 121 }; 122 123 union cpuid10_ebx { 124 struct { 125 unsigned int no_unhalted_core_cycles:1; 126 unsigned int no_instructions_retired:1; 127 unsigned int no_unhalted_reference_cycles:1; 128 unsigned int no_llc_reference:1; 129 unsigned int no_llc_misses:1; 130 unsigned int no_branch_instruction_retired:1; 131 unsigned int no_branch_misses_retired:1; 132 } split; 133 unsigned int full; 134 }; 135 136 union cpuid10_edx { 137 struct { 138 unsigned int num_counters_fixed:5; 139 unsigned int bit_width_fixed:8; 140 unsigned int reserved1:2; 141 unsigned int anythread_deprecated:1; 142 unsigned int reserved2:16; 143 } split; 144 unsigned int full; 145 }; 146 147 /* 148 * Intel Architectural LBR CPUID detection/enumeration details: 149 */ 150 union cpuid28_eax { 151 struct { 152 /* Supported LBR depth values */ 153 unsigned int lbr_depth_mask:8; 154 unsigned int reserved:22; 155 /* Deep C-state Reset */ 156 unsigned int lbr_deep_c_reset:1; 157 /* IP values contain LIP */ 158 unsigned int lbr_lip:1; 159 } split; 160 unsigned int full; 161 }; 162 163 union cpuid28_ebx { 164 struct { 165 /* CPL Filtering Supported */ 166 unsigned int lbr_cpl:1; 167 /* Branch Filtering Supported */ 168 unsigned int lbr_filter:1; 169 /* Call-stack Mode Supported */ 170 unsigned int lbr_call_stack:1; 171 } split; 172 unsigned int full; 173 }; 174 175 union cpuid28_ecx { 176 struct { 177 /* Mispredict Bit Supported */ 178 unsigned int lbr_mispred:1; 179 /* Timed LBRs Supported */ 180 unsigned int lbr_timed_lbr:1; 181 /* Branch Type Field Supported */ 182 unsigned int lbr_br_type:1; 183 } split; 184 unsigned int full; 185 }; 186 187 struct x86_pmu_capability { 188 int version; 189 int num_counters_gp; 190 int num_counters_fixed; 191 int bit_width_gp; 192 int bit_width_fixed; 193 unsigned int events_mask; 194 int events_mask_len; 195 }; 196 197 /* 198 * Fixed-purpose performance events: 199 */ 200 201 /* RDPMC offset for Fixed PMCs */ 202 #define INTEL_PMC_FIXED_RDPMC_BASE (1 << 30) 203 #define INTEL_PMC_FIXED_RDPMC_METRICS (1 << 29) 204 205 /* 206 * All the fixed-mode PMCs are configured via this single MSR: 207 */ 208 #define MSR_ARCH_PERFMON_FIXED_CTR_CTRL 0x38d 209 210 /* 211 * There is no event-code assigned to the fixed-mode PMCs. 212 * 213 * For a fixed-mode PMC, which has an equivalent event on a general-purpose 214 * PMC, the event-code of the equivalent event is used for the fixed-mode PMC, 215 * e.g., Instr_Retired.Any and CPU_CLK_Unhalted.Core. 216 * 217 * For a fixed-mode PMC, which doesn't have an equivalent event, a 218 * pseudo-encoding is used, e.g., CPU_CLK_Unhalted.Ref and TOPDOWN.SLOTS. 219 * The pseudo event-code for a fixed-mode PMC must be 0x00. 220 * The pseudo umask-code is 0xX. The X equals the index of the fixed 221 * counter + 1, e.g., the fixed counter 2 has the pseudo-encoding 0x0300. 222 * 223 * The counts are available in separate MSRs: 224 */ 225 226 /* Instr_Retired.Any: */ 227 #define MSR_ARCH_PERFMON_FIXED_CTR0 0x309 228 #define INTEL_PMC_IDX_FIXED_INSTRUCTIONS (INTEL_PMC_IDX_FIXED + 0) 229 230 /* CPU_CLK_Unhalted.Core: */ 231 #define MSR_ARCH_PERFMON_FIXED_CTR1 0x30a 232 #define INTEL_PMC_IDX_FIXED_CPU_CYCLES (INTEL_PMC_IDX_FIXED + 1) 233 234 /* CPU_CLK_Unhalted.Ref: event=0x00,umask=0x3 (pseudo-encoding) */ 235 #define MSR_ARCH_PERFMON_FIXED_CTR2 0x30b 236 #define INTEL_PMC_IDX_FIXED_REF_CYCLES (INTEL_PMC_IDX_FIXED + 2) 237 #define INTEL_PMC_MSK_FIXED_REF_CYCLES (1ULL << INTEL_PMC_IDX_FIXED_REF_CYCLES) 238 239 /* TOPDOWN.SLOTS: event=0x00,umask=0x4 (pseudo-encoding) */ 240 #define MSR_ARCH_PERFMON_FIXED_CTR3 0x30c 241 #define INTEL_PMC_IDX_FIXED_SLOTS (INTEL_PMC_IDX_FIXED + 3) 242 #define INTEL_PMC_MSK_FIXED_SLOTS (1ULL << INTEL_PMC_IDX_FIXED_SLOTS) 243 244 /* 245 * We model BTS tracing as another fixed-mode PMC. 246 * 247 * We choose the value 47 for the fixed index of BTS, since lower 248 * values are used by actual fixed events and higher values are used 249 * to indicate other overflow conditions in the PERF_GLOBAL_STATUS msr. 250 */ 251 #define INTEL_PMC_IDX_FIXED_BTS (INTEL_PMC_IDX_FIXED + 15) 252 253 /* 254 * The PERF_METRICS MSR is modeled as several magic fixed-mode PMCs, one for 255 * each TopDown metric event. 256 * 257 * Internally the TopDown metric events are mapped to the FxCtr 3 (SLOTS). 258 */ 259 #define INTEL_PMC_IDX_METRIC_BASE (INTEL_PMC_IDX_FIXED + 16) 260 #define INTEL_PMC_IDX_TD_RETIRING (INTEL_PMC_IDX_METRIC_BASE + 0) 261 #define INTEL_PMC_IDX_TD_BAD_SPEC (INTEL_PMC_IDX_METRIC_BASE + 1) 262 #define INTEL_PMC_IDX_TD_FE_BOUND (INTEL_PMC_IDX_METRIC_BASE + 2) 263 #define INTEL_PMC_IDX_TD_BE_BOUND (INTEL_PMC_IDX_METRIC_BASE + 3) 264 #define INTEL_PMC_IDX_METRIC_END INTEL_PMC_IDX_TD_BE_BOUND 265 #define INTEL_PMC_MSK_TOPDOWN ((0xfull << INTEL_PMC_IDX_METRIC_BASE) | \ 266 INTEL_PMC_MSK_FIXED_SLOTS) 267 268 /* 269 * There is no event-code assigned to the TopDown events. 270 * 271 * For the slots event, use the pseudo code of the fixed counter 3. 272 * 273 * For the metric events, the pseudo event-code is 0x00. 274 * The pseudo umask-code starts from the middle of the pseudo event 275 * space, 0x80. 276 */ 277 #define INTEL_TD_SLOTS 0x0400 /* TOPDOWN.SLOTS */ 278 /* Level 1 metrics */ 279 #define INTEL_TD_METRIC_RETIRING 0x8000 /* Retiring metric */ 280 #define INTEL_TD_METRIC_BAD_SPEC 0x8100 /* Bad speculation metric */ 281 #define INTEL_TD_METRIC_FE_BOUND 0x8200 /* FE bound metric */ 282 #define INTEL_TD_METRIC_BE_BOUND 0x8300 /* BE bound metric */ 283 #define INTEL_TD_METRIC_MAX INTEL_TD_METRIC_BE_BOUND 284 #define INTEL_TD_METRIC_NUM 4 285 286 static inline bool is_metric_idx(int idx) 287 { 288 return (unsigned)(idx - INTEL_PMC_IDX_METRIC_BASE) < INTEL_TD_METRIC_NUM; 289 } 290 291 static inline bool is_topdown_idx(int idx) 292 { 293 return is_metric_idx(idx) || idx == INTEL_PMC_IDX_FIXED_SLOTS; 294 } 295 296 #define INTEL_PMC_OTHER_TOPDOWN_BITS(bit) \ 297 (~(0x1ull << bit) & INTEL_PMC_MSK_TOPDOWN) 298 299 #define GLOBAL_STATUS_COND_CHG BIT_ULL(63) 300 #define GLOBAL_STATUS_BUFFER_OVF_BIT 62 301 #define GLOBAL_STATUS_BUFFER_OVF BIT_ULL(GLOBAL_STATUS_BUFFER_OVF_BIT) 302 #define GLOBAL_STATUS_UNC_OVF BIT_ULL(61) 303 #define GLOBAL_STATUS_ASIF BIT_ULL(60) 304 #define GLOBAL_STATUS_COUNTERS_FROZEN BIT_ULL(59) 305 #define GLOBAL_STATUS_LBRS_FROZEN_BIT 58 306 #define GLOBAL_STATUS_LBRS_FROZEN BIT_ULL(GLOBAL_STATUS_LBRS_FROZEN_BIT) 307 #define GLOBAL_STATUS_TRACE_TOPAPMI_BIT 55 308 #define GLOBAL_STATUS_TRACE_TOPAPMI BIT_ULL(GLOBAL_STATUS_TRACE_TOPAPMI_BIT) 309 #define GLOBAL_STATUS_PERF_METRICS_OVF_BIT 48 310 311 #define GLOBAL_CTRL_EN_PERF_METRICS 48 312 /* 313 * We model guest LBR event tracing as another fixed-mode PMC like BTS. 314 * 315 * We choose bit 58 because it's used to indicate LBR stack frozen state 316 * for architectural perfmon v4, also we unconditionally mask that bit in 317 * the handle_pmi_common(), so it'll never be set in the overflow handling. 318 * 319 * With this fake counter assigned, the guest LBR event user (such as KVM), 320 * can program the LBR registers on its own, and we don't actually do anything 321 * with then in the host context. 322 */ 323 #define INTEL_PMC_IDX_FIXED_VLBR (GLOBAL_STATUS_LBRS_FROZEN_BIT) 324 325 /* 326 * Pseudo-encoding the guest LBR event as event=0x00,umask=0x1b, 327 * since it would claim bit 58 which is effectively Fixed26. 328 */ 329 #define INTEL_FIXED_VLBR_EVENT 0x1b00 330 331 /* 332 * Adaptive PEBS v4 333 */ 334 335 struct pebs_basic { 336 u64 format_size; 337 u64 ip; 338 u64 applicable_counters; 339 u64 tsc; 340 }; 341 342 struct pebs_meminfo { 343 u64 address; 344 u64 aux; 345 u64 latency; 346 u64 tsx_tuning; 347 }; 348 349 struct pebs_gprs { 350 u64 flags, ip, ax, cx, dx, bx, sp, bp, si, di; 351 u64 r8, r9, r10, r11, r12, r13, r14, r15; 352 }; 353 354 struct pebs_xmm { 355 u64 xmm[16*2]; /* two entries for each register */ 356 }; 357 358 /* 359 * IBS cpuid feature detection 360 */ 361 362 #define IBS_CPUID_FEATURES 0x8000001b 363 364 /* 365 * Same bit mask as for IBS cpuid feature flags (Fn8000_001B_EAX), but 366 * bit 0 is used to indicate the existence of IBS. 367 */ 368 #define IBS_CAPS_AVAIL (1U<<0) 369 #define IBS_CAPS_FETCHSAM (1U<<1) 370 #define IBS_CAPS_OPSAM (1U<<2) 371 #define IBS_CAPS_RDWROPCNT (1U<<3) 372 #define IBS_CAPS_OPCNT (1U<<4) 373 #define IBS_CAPS_BRNTRGT (1U<<5) 374 #define IBS_CAPS_OPCNTEXT (1U<<6) 375 #define IBS_CAPS_RIPINVALIDCHK (1U<<7) 376 #define IBS_CAPS_OPBRNFUSE (1U<<8) 377 #define IBS_CAPS_FETCHCTLEXTD (1U<<9) 378 #define IBS_CAPS_OPDATA4 (1U<<10) 379 380 #define IBS_CAPS_DEFAULT (IBS_CAPS_AVAIL \ 381 | IBS_CAPS_FETCHSAM \ 382 | IBS_CAPS_OPSAM) 383 384 /* 385 * IBS APIC setup 386 */ 387 #define IBSCTL 0x1cc 388 #define IBSCTL_LVT_OFFSET_VALID (1ULL<<8) 389 #define IBSCTL_LVT_OFFSET_MASK 0x0F 390 391 /* IBS fetch bits/masks */ 392 #define IBS_FETCH_RAND_EN (1ULL<<57) 393 #define IBS_FETCH_VAL (1ULL<<49) 394 #define IBS_FETCH_ENABLE (1ULL<<48) 395 #define IBS_FETCH_CNT 0xFFFF0000ULL 396 #define IBS_FETCH_MAX_CNT 0x0000FFFFULL 397 398 /* 399 * IBS op bits/masks 400 * The lower 7 bits of the current count are random bits 401 * preloaded by hardware and ignored in software 402 */ 403 #define IBS_OP_CUR_CNT (0xFFF80ULL<<32) 404 #define IBS_OP_CUR_CNT_RAND (0x0007FULL<<32) 405 #define IBS_OP_CNT_CTL (1ULL<<19) 406 #define IBS_OP_VAL (1ULL<<18) 407 #define IBS_OP_ENABLE (1ULL<<17) 408 #define IBS_OP_MAX_CNT 0x0000FFFFULL 409 #define IBS_OP_MAX_CNT_EXT 0x007FFFFFULL /* not a register bit mask */ 410 #define IBS_OP_MAX_CNT_EXT_MASK (0x7FULL<<20) /* separate upper 7 bits */ 411 #define IBS_RIP_INVALID (1ULL<<38) 412 413 #ifdef CONFIG_X86_LOCAL_APIC 414 extern u32 get_ibs_caps(void); 415 #else 416 static inline u32 get_ibs_caps(void) { return 0; } 417 #endif 418 419 #ifdef CONFIG_PERF_EVENTS 420 extern void perf_events_lapic_init(void); 421 422 /* 423 * Abuse bits {3,5} of the cpu eflags register. These flags are otherwise 424 * unused and ABI specified to be 0, so nobody should care what we do with 425 * them. 426 * 427 * EXACT - the IP points to the exact instruction that triggered the 428 * event (HW bugs exempt). 429 * VM - original X86_VM_MASK; see set_linear_ip(). 430 */ 431 #define PERF_EFLAGS_EXACT (1UL << 3) 432 #define PERF_EFLAGS_VM (1UL << 5) 433 434 struct pt_regs; 435 struct x86_perf_regs { 436 struct pt_regs regs; 437 u64 *xmm_regs; 438 }; 439 440 extern unsigned long perf_instruction_pointer(struct pt_regs *regs); 441 extern unsigned long perf_misc_flags(struct pt_regs *regs); 442 #define perf_misc_flags(regs) perf_misc_flags(regs) 443 444 #include <asm/stacktrace.h> 445 446 /* 447 * We abuse bit 3 from flags to pass exact information, see perf_misc_flags 448 * and the comment with PERF_EFLAGS_EXACT. 449 */ 450 #define perf_arch_fetch_caller_regs(regs, __ip) { \ 451 (regs)->ip = (__ip); \ 452 (regs)->sp = (unsigned long)__builtin_frame_address(0); \ 453 (regs)->cs = __KERNEL_CS; \ 454 regs->flags = 0; \ 455 } 456 457 struct perf_guest_switch_msr { 458 unsigned msr; 459 u64 host, guest; 460 }; 461 462 struct x86_pmu_lbr { 463 unsigned int nr; 464 unsigned int from; 465 unsigned int to; 466 unsigned int info; 467 }; 468 469 extern void perf_get_x86_pmu_capability(struct x86_pmu_capability *cap); 470 extern void perf_check_microcode(void); 471 extern int x86_perf_rdpmc_index(struct perf_event *event); 472 #else 473 static inline void perf_get_x86_pmu_capability(struct x86_pmu_capability *cap) 474 { 475 memset(cap, 0, sizeof(*cap)); 476 } 477 478 static inline void perf_events_lapic_init(void) { } 479 static inline void perf_check_microcode(void) { } 480 #endif 481 482 #if defined(CONFIG_PERF_EVENTS) && defined(CONFIG_CPU_SUP_INTEL) 483 extern struct perf_guest_switch_msr *perf_guest_get_msrs(int *nr); 484 extern int x86_perf_get_lbr(struct x86_pmu_lbr *lbr); 485 #else 486 static inline struct perf_guest_switch_msr *perf_guest_get_msrs(int *nr) 487 { 488 *nr = 0; 489 return NULL; 490 } 491 static inline int x86_perf_get_lbr(struct x86_pmu_lbr *lbr) 492 { 493 return -1; 494 } 495 #endif 496 497 #ifdef CONFIG_CPU_SUP_INTEL 498 extern void intel_pt_handle_vmx(int on); 499 #else 500 static inline void intel_pt_handle_vmx(int on) 501 { 502 503 } 504 #endif 505 506 #if defined(CONFIG_PERF_EVENTS) && defined(CONFIG_CPU_SUP_AMD) 507 extern void amd_pmu_enable_virt(void); 508 extern void amd_pmu_disable_virt(void); 509 #else 510 static inline void amd_pmu_enable_virt(void) { } 511 static inline void amd_pmu_disable_virt(void) { } 512 #endif 513 514 #define arch_perf_out_copy_user copy_from_user_nmi 515 516 #endif /* _ASM_X86_PERF_EVENT_H */ 517