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