1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _ASM_X86_PERF_EVENT_H
3 #define _ASM_X86_PERF_EVENT_H
6 * Performance event hw details:
9 #define INTEL_PMC_MAX_GENERIC 32
10 #define INTEL_PMC_MAX_FIXED 16
11 #define INTEL_PMC_IDX_FIXED 32
13 #define X86_PMC_IDX_MAX 64
15 #define MSR_ARCH_PERFMON_PERFCTR0 0xc1
16 #define MSR_ARCH_PERFMON_PERFCTR1 0xc2
18 #define MSR_ARCH_PERFMON_EVENTSEL0 0x186
19 #define MSR_ARCH_PERFMON_EVENTSEL1 0x187
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
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)
38 #define AMD64_EVENTSEL_INT_CORE_ENABLE (1ULL << 36)
39 #define AMD64_EVENTSEL_GUESTONLY (1ULL << 40)
40 #define AMD64_EVENTSEL_HOSTONLY (1ULL << 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)
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)
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)
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)
63 #define AMD64_L3_EN_ALL_CORES BIT_ULL(47)
64 #define AMD64_L3_EN_ALL_SLICES BIT_ULL(46)
66 #define AMD64_L3_COREID_SHIFT 42
67 #define AMD64_L3_COREID_MASK \
68 (0x7ULL << AMD64_L3_COREID_SHIFT)
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 | \
83 HSW_IN_TX_CHECKPOINTED)
84 #define AMD64_RAW_EVENT_MASK \
85 (X86_RAW_EVENT_MASK | \
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
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))
100 #define ARCH_PERFMON_BRANCH_MISSES_RETIRED 6
101 #define ARCH_PERFMON_EVENTS_COUNT 7
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
110 * Intel "Architectural Performance Monitoring" CPUID
111 * detection/enumeration details:
115 unsigned int version_id:8;
116 unsigned int num_counters:8;
117 unsigned int bit_width:8;
118 unsigned int mask_length:8;
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;
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;
148 * Intel Architectural LBR CPUID detection/enumeration details:
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;
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;
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;
187 struct x86_pmu_capability {
190 int num_counters_fixed;
193 unsigned int events_mask;
198 * Fixed-purpose performance events:
201 /* RDPMC offset for Fixed PMCs */
202 #define INTEL_PMC_FIXED_RDPMC_BASE (1 << 30)
203 #define INTEL_PMC_FIXED_RDPMC_METRICS (1 << 29)
206 * All the fixed-mode PMCs are configured via this single MSR:
208 #define MSR_ARCH_PERFMON_FIXED_CTR_CTRL 0x38d
211 * There is no event-code assigned to the fixed-mode PMCs.
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.
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.
223 * The counts are available in separate MSRs:
226 /* Instr_Retired.Any: */
227 #define MSR_ARCH_PERFMON_FIXED_CTR0 0x309
228 #define INTEL_PMC_IDX_FIXED_INSTRUCTIONS (INTEL_PMC_IDX_FIXED + 0)
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)
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)
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)
244 static inline bool use_fixed_pseudo_encoding(u64 code)
246 return !(code & 0xff);
250 * We model BTS tracing as another fixed-mode PMC.
252 * We choose the value 47 for the fixed index of BTS, since lower
253 * values are used by actual fixed events and higher values are used
254 * to indicate other overflow conditions in the PERF_GLOBAL_STATUS msr.
256 #define INTEL_PMC_IDX_FIXED_BTS (INTEL_PMC_IDX_FIXED + 15)
259 * The PERF_METRICS MSR is modeled as several magic fixed-mode PMCs, one for
260 * each TopDown metric event.
262 * Internally the TopDown metric events are mapped to the FxCtr 3 (SLOTS).
264 #define INTEL_PMC_IDX_METRIC_BASE (INTEL_PMC_IDX_FIXED + 16)
265 #define INTEL_PMC_IDX_TD_RETIRING (INTEL_PMC_IDX_METRIC_BASE + 0)
266 #define INTEL_PMC_IDX_TD_BAD_SPEC (INTEL_PMC_IDX_METRIC_BASE + 1)
267 #define INTEL_PMC_IDX_TD_FE_BOUND (INTEL_PMC_IDX_METRIC_BASE + 2)
268 #define INTEL_PMC_IDX_TD_BE_BOUND (INTEL_PMC_IDX_METRIC_BASE + 3)
269 #define INTEL_PMC_IDX_TD_HEAVY_OPS (INTEL_PMC_IDX_METRIC_BASE + 4)
270 #define INTEL_PMC_IDX_TD_BR_MISPREDICT (INTEL_PMC_IDX_METRIC_BASE + 5)
271 #define INTEL_PMC_IDX_TD_FETCH_LAT (INTEL_PMC_IDX_METRIC_BASE + 6)
272 #define INTEL_PMC_IDX_TD_MEM_BOUND (INTEL_PMC_IDX_METRIC_BASE + 7)
273 #define INTEL_PMC_IDX_METRIC_END INTEL_PMC_IDX_TD_MEM_BOUND
274 #define INTEL_PMC_MSK_TOPDOWN ((0xffull << INTEL_PMC_IDX_METRIC_BASE) | \
275 INTEL_PMC_MSK_FIXED_SLOTS)
278 * There is no event-code assigned to the TopDown events.
280 * For the slots event, use the pseudo code of the fixed counter 3.
282 * For the metric events, the pseudo event-code is 0x00.
283 * The pseudo umask-code starts from the middle of the pseudo event
286 #define INTEL_TD_SLOTS 0x0400 /* TOPDOWN.SLOTS */
287 /* Level 1 metrics */
288 #define INTEL_TD_METRIC_RETIRING 0x8000 /* Retiring metric */
289 #define INTEL_TD_METRIC_BAD_SPEC 0x8100 /* Bad speculation metric */
290 #define INTEL_TD_METRIC_FE_BOUND 0x8200 /* FE bound metric */
291 #define INTEL_TD_METRIC_BE_BOUND 0x8300 /* BE bound metric */
292 /* Level 2 metrics */
293 #define INTEL_TD_METRIC_HEAVY_OPS 0x8400 /* Heavy Operations metric */
294 #define INTEL_TD_METRIC_BR_MISPREDICT 0x8500 /* Branch Mispredict metric */
295 #define INTEL_TD_METRIC_FETCH_LAT 0x8600 /* Fetch Latency metric */
296 #define INTEL_TD_METRIC_MEM_BOUND 0x8700 /* Memory bound metric */
298 #define INTEL_TD_METRIC_MAX INTEL_TD_METRIC_MEM_BOUND
299 #define INTEL_TD_METRIC_NUM 8
301 static inline bool is_metric_idx(int idx)
303 return (unsigned)(idx - INTEL_PMC_IDX_METRIC_BASE) < INTEL_TD_METRIC_NUM;
306 static inline bool is_topdown_idx(int idx)
308 return is_metric_idx(idx) || idx == INTEL_PMC_IDX_FIXED_SLOTS;
311 #define INTEL_PMC_OTHER_TOPDOWN_BITS(bit) \
312 (~(0x1ull << bit) & INTEL_PMC_MSK_TOPDOWN)
314 #define GLOBAL_STATUS_COND_CHG BIT_ULL(63)
315 #define GLOBAL_STATUS_BUFFER_OVF_BIT 62
316 #define GLOBAL_STATUS_BUFFER_OVF BIT_ULL(GLOBAL_STATUS_BUFFER_OVF_BIT)
317 #define GLOBAL_STATUS_UNC_OVF BIT_ULL(61)
318 #define GLOBAL_STATUS_ASIF BIT_ULL(60)
319 #define GLOBAL_STATUS_COUNTERS_FROZEN BIT_ULL(59)
320 #define GLOBAL_STATUS_LBRS_FROZEN_BIT 58
321 #define GLOBAL_STATUS_LBRS_FROZEN BIT_ULL(GLOBAL_STATUS_LBRS_FROZEN_BIT)
322 #define GLOBAL_STATUS_TRACE_TOPAPMI_BIT 55
323 #define GLOBAL_STATUS_TRACE_TOPAPMI BIT_ULL(GLOBAL_STATUS_TRACE_TOPAPMI_BIT)
324 #define GLOBAL_STATUS_PERF_METRICS_OVF_BIT 48
326 #define GLOBAL_CTRL_EN_PERF_METRICS 48
328 * We model guest LBR event tracing as another fixed-mode PMC like BTS.
330 * We choose bit 58 because it's used to indicate LBR stack frozen state
331 * for architectural perfmon v4, also we unconditionally mask that bit in
332 * the handle_pmi_common(), so it'll never be set in the overflow handling.
334 * With this fake counter assigned, the guest LBR event user (such as KVM),
335 * can program the LBR registers on its own, and we don't actually do anything
336 * with then in the host context.
338 #define INTEL_PMC_IDX_FIXED_VLBR (GLOBAL_STATUS_LBRS_FROZEN_BIT)
341 * Pseudo-encoding the guest LBR event as event=0x00,umask=0x1b,
342 * since it would claim bit 58 which is effectively Fixed26.
344 #define INTEL_FIXED_VLBR_EVENT 0x1b00
353 u64 applicable_counters;
357 struct pebs_meminfo {
365 u64 flags, ip, ax, cx, dx, bx, sp, bp, si, di;
366 u64 r8, r9, r10, r11, r12, r13, r14, r15;
370 u64 xmm[16*2]; /* two entries for each register */
374 * IBS cpuid feature detection
377 #define IBS_CPUID_FEATURES 0x8000001b
380 * Same bit mask as for IBS cpuid feature flags (Fn8000_001B_EAX), but
381 * bit 0 is used to indicate the existence of IBS.
383 #define IBS_CAPS_AVAIL (1U<<0)
384 #define IBS_CAPS_FETCHSAM (1U<<1)
385 #define IBS_CAPS_OPSAM (1U<<2)
386 #define IBS_CAPS_RDWROPCNT (1U<<3)
387 #define IBS_CAPS_OPCNT (1U<<4)
388 #define IBS_CAPS_BRNTRGT (1U<<5)
389 #define IBS_CAPS_OPCNTEXT (1U<<6)
390 #define IBS_CAPS_RIPINVALIDCHK (1U<<7)
391 #define IBS_CAPS_OPBRNFUSE (1U<<8)
392 #define IBS_CAPS_FETCHCTLEXTD (1U<<9)
393 #define IBS_CAPS_OPDATA4 (1U<<10)
395 #define IBS_CAPS_DEFAULT (IBS_CAPS_AVAIL \
396 | IBS_CAPS_FETCHSAM \
403 #define IBSCTL_LVT_OFFSET_VALID (1ULL<<8)
404 #define IBSCTL_LVT_OFFSET_MASK 0x0F
406 /* IBS fetch bits/masks */
407 #define IBS_FETCH_RAND_EN (1ULL<<57)
408 #define IBS_FETCH_VAL (1ULL<<49)
409 #define IBS_FETCH_ENABLE (1ULL<<48)
410 #define IBS_FETCH_CNT 0xFFFF0000ULL
411 #define IBS_FETCH_MAX_CNT 0x0000FFFFULL
415 * The lower 7 bits of the current count are random bits
416 * preloaded by hardware and ignored in software
418 #define IBS_OP_CUR_CNT (0xFFF80ULL<<32)
419 #define IBS_OP_CUR_CNT_RAND (0x0007FULL<<32)
420 #define IBS_OP_CNT_CTL (1ULL<<19)
421 #define IBS_OP_VAL (1ULL<<18)
422 #define IBS_OP_ENABLE (1ULL<<17)
423 #define IBS_OP_MAX_CNT 0x0000FFFFULL
424 #define IBS_OP_MAX_CNT_EXT 0x007FFFFFULL /* not a register bit mask */
425 #define IBS_OP_MAX_CNT_EXT_MASK (0x7FULL<<20) /* separate upper 7 bits */
426 #define IBS_RIP_INVALID (1ULL<<38)
428 #ifdef CONFIG_X86_LOCAL_APIC
429 extern u32 get_ibs_caps(void);
431 static inline u32 get_ibs_caps(void) { return 0; }
434 #ifdef CONFIG_PERF_EVENTS
435 extern void perf_events_lapic_init(void);
438 * Abuse bits {3,5} of the cpu eflags register. These flags are otherwise
439 * unused and ABI specified to be 0, so nobody should care what we do with
442 * EXACT - the IP points to the exact instruction that triggered the
443 * event (HW bugs exempt).
444 * VM - original X86_VM_MASK; see set_linear_ip().
446 #define PERF_EFLAGS_EXACT (1UL << 3)
447 #define PERF_EFLAGS_VM (1UL << 5)
450 struct x86_perf_regs {
455 extern unsigned long perf_instruction_pointer(struct pt_regs *regs);
456 extern unsigned long perf_misc_flags(struct pt_regs *regs);
457 #define perf_misc_flags(regs) perf_misc_flags(regs)
459 #include <asm/stacktrace.h>
462 * We abuse bit 3 from flags to pass exact information, see perf_misc_flags
463 * and the comment with PERF_EFLAGS_EXACT.
465 #define perf_arch_fetch_caller_regs(regs, __ip) { \
466 (regs)->ip = (__ip); \
467 (regs)->sp = (unsigned long)__builtin_frame_address(0); \
468 (regs)->cs = __KERNEL_CS; \
472 struct perf_guest_switch_msr {
484 extern void perf_get_x86_pmu_capability(struct x86_pmu_capability *cap);
485 extern void perf_check_microcode(void);
486 extern void perf_clear_dirty_counters(void);
487 extern int x86_perf_rdpmc_index(struct perf_event *event);
489 static inline void perf_get_x86_pmu_capability(struct x86_pmu_capability *cap)
491 memset(cap, 0, sizeof(*cap));
494 static inline void perf_events_lapic_init(void) { }
495 static inline void perf_check_microcode(void) { }
498 #if defined(CONFIG_PERF_EVENTS) && defined(CONFIG_CPU_SUP_INTEL)
499 extern struct perf_guest_switch_msr *perf_guest_get_msrs(int *nr);
500 extern int x86_perf_get_lbr(struct x86_pmu_lbr *lbr);
502 struct perf_guest_switch_msr *perf_guest_get_msrs(int *nr);
503 static inline int x86_perf_get_lbr(struct x86_pmu_lbr *lbr)
509 #ifdef CONFIG_CPU_SUP_INTEL
510 extern void intel_pt_handle_vmx(int on);
512 static inline void intel_pt_handle_vmx(int on)
518 #if defined(CONFIG_PERF_EVENTS) && defined(CONFIG_CPU_SUP_AMD)
519 extern void amd_pmu_enable_virt(void);
520 extern void amd_pmu_disable_virt(void);
522 static inline void amd_pmu_enable_virt(void) { }
523 static inline void amd_pmu_disable_virt(void) { }
526 #define arch_perf_out_copy_user copy_from_user_nmi
528 #endif /* _ASM_X86_PERF_EVENT_H */