perf/x86: Implement 64-bit counter support for IBS
[deliverable/linux.git] / arch / x86 / include / asm / perf_event.h
1 #ifndef _ASM_X86_PERF_EVENT_H
2 #define _ASM_X86_PERF_EVENT_H
3
4 /*
5 * Performance event hw details:
6 */
7
8 #define X86_PMC_MAX_GENERIC 32
9 #define X86_PMC_MAX_FIXED 3
10
11 #define X86_PMC_IDX_GENERIC 0
12 #define X86_PMC_IDX_FIXED 32
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_INT (1ULL << 20)
27 #define ARCH_PERFMON_EVENTSEL_ANY (1ULL << 21)
28 #define ARCH_PERFMON_EVENTSEL_ENABLE (1ULL << 22)
29 #define ARCH_PERFMON_EVENTSEL_INV (1ULL << 23)
30 #define ARCH_PERFMON_EVENTSEL_CMASK 0xFF000000ULL
31
32 #define AMD_PERFMON_EVENTSEL_GUESTONLY (1ULL << 40)
33 #define AMD_PERFMON_EVENTSEL_HOSTONLY (1ULL << 41)
34
35 #define AMD64_EVENTSEL_EVENT \
36 (ARCH_PERFMON_EVENTSEL_EVENT | (0x0FULL << 32))
37 #define INTEL_ARCH_EVENT_MASK \
38 (ARCH_PERFMON_EVENTSEL_UMASK | ARCH_PERFMON_EVENTSEL_EVENT)
39
40 #define X86_RAW_EVENT_MASK \
41 (ARCH_PERFMON_EVENTSEL_EVENT | \
42 ARCH_PERFMON_EVENTSEL_UMASK | \
43 ARCH_PERFMON_EVENTSEL_EDGE | \
44 ARCH_PERFMON_EVENTSEL_INV | \
45 ARCH_PERFMON_EVENTSEL_CMASK)
46 #define AMD64_RAW_EVENT_MASK \
47 (X86_RAW_EVENT_MASK | \
48 AMD64_EVENTSEL_EVENT)
49 #define AMD64_NUM_COUNTERS 4
50 #define AMD64_NUM_COUNTERS_F15H 6
51 #define AMD64_NUM_COUNTERS_MAX AMD64_NUM_COUNTERS_F15H
52
53 #define ARCH_PERFMON_UNHALTED_CORE_CYCLES_SEL 0x3c
54 #define ARCH_PERFMON_UNHALTED_CORE_CYCLES_UMASK (0x00 << 8)
55 #define ARCH_PERFMON_UNHALTED_CORE_CYCLES_INDEX 0
56 #define ARCH_PERFMON_UNHALTED_CORE_CYCLES_PRESENT \
57 (1 << (ARCH_PERFMON_UNHALTED_CORE_CYCLES_INDEX))
58
59 #define ARCH_PERFMON_BRANCH_MISSES_RETIRED 6
60 #define ARCH_PERFMON_EVENTS_COUNT 7
61
62 /*
63 * Intel "Architectural Performance Monitoring" CPUID
64 * detection/enumeration details:
65 */
66 union cpuid10_eax {
67 struct {
68 unsigned int version_id:8;
69 unsigned int num_counters:8;
70 unsigned int bit_width:8;
71 unsigned int mask_length:8;
72 } split;
73 unsigned int full;
74 };
75
76 union cpuid10_ebx {
77 struct {
78 unsigned int no_unhalted_core_cycles:1;
79 unsigned int no_instructions_retired:1;
80 unsigned int no_unhalted_reference_cycles:1;
81 unsigned int no_llc_reference:1;
82 unsigned int no_llc_misses:1;
83 unsigned int no_branch_instruction_retired:1;
84 unsigned int no_branch_misses_retired:1;
85 } split;
86 unsigned int full;
87 };
88
89 union cpuid10_edx {
90 struct {
91 unsigned int num_counters_fixed:5;
92 unsigned int bit_width_fixed:8;
93 unsigned int reserved:19;
94 } split;
95 unsigned int full;
96 };
97
98 struct x86_pmu_capability {
99 int version;
100 int num_counters_gp;
101 int num_counters_fixed;
102 int bit_width_gp;
103 int bit_width_fixed;
104 unsigned int events_mask;
105 int events_mask_len;
106 };
107
108 /*
109 * Fixed-purpose performance events:
110 */
111
112 /*
113 * All 3 fixed-mode PMCs are configured via this single MSR:
114 */
115 #define MSR_ARCH_PERFMON_FIXED_CTR_CTRL 0x38d
116
117 /*
118 * The counts are available in three separate MSRs:
119 */
120
121 /* Instr_Retired.Any: */
122 #define MSR_ARCH_PERFMON_FIXED_CTR0 0x309
123 #define X86_PMC_IDX_FIXED_INSTRUCTIONS (X86_PMC_IDX_FIXED + 0)
124
125 /* CPU_CLK_Unhalted.Core: */
126 #define MSR_ARCH_PERFMON_FIXED_CTR1 0x30a
127 #define X86_PMC_IDX_FIXED_CPU_CYCLES (X86_PMC_IDX_FIXED + 1)
128
129 /* CPU_CLK_Unhalted.Ref: */
130 #define MSR_ARCH_PERFMON_FIXED_CTR2 0x30b
131 #define X86_PMC_IDX_FIXED_REF_CYCLES (X86_PMC_IDX_FIXED + 2)
132 #define X86_PMC_MSK_FIXED_REF_CYCLES (1ULL << X86_PMC_IDX_FIXED_REF_CYCLES)
133
134 /*
135 * We model BTS tracing as another fixed-mode PMC.
136 *
137 * We choose a value in the middle of the fixed event range, since lower
138 * values are used by actual fixed events and higher values are used
139 * to indicate other overflow conditions in the PERF_GLOBAL_STATUS msr.
140 */
141 #define X86_PMC_IDX_FIXED_BTS (X86_PMC_IDX_FIXED + 16)
142
143 /*
144 * IBS cpuid feature detection
145 */
146
147 #define IBS_CPUID_FEATURES 0x8000001b
148
149 /*
150 * Same bit mask as for IBS cpuid feature flags (Fn8000_001B_EAX), but
151 * bit 0 is used to indicate the existence of IBS.
152 */
153 #define IBS_CAPS_AVAIL (1U<<0)
154 #define IBS_CAPS_FETCHSAM (1U<<1)
155 #define IBS_CAPS_OPSAM (1U<<2)
156 #define IBS_CAPS_RDWROPCNT (1U<<3)
157 #define IBS_CAPS_OPCNT (1U<<4)
158 #define IBS_CAPS_BRNTRGT (1U<<5)
159 #define IBS_CAPS_OPCNTEXT (1U<<6)
160
161 #define IBS_CAPS_DEFAULT (IBS_CAPS_AVAIL \
162 | IBS_CAPS_FETCHSAM \
163 | IBS_CAPS_OPSAM)
164
165 /*
166 * IBS APIC setup
167 */
168 #define IBSCTL 0x1cc
169 #define IBSCTL_LVT_OFFSET_VALID (1ULL<<8)
170 #define IBSCTL_LVT_OFFSET_MASK 0x0F
171
172 /* IbsFetchCtl bits/masks */
173 #define IBS_FETCH_RAND_EN (1ULL<<57)
174 #define IBS_FETCH_VAL (1ULL<<49)
175 #define IBS_FETCH_ENABLE (1ULL<<48)
176 #define IBS_FETCH_CNT 0xFFFF0000ULL
177 #define IBS_FETCH_MAX_CNT 0x0000FFFFULL
178
179 /* IbsOpCtl bits */
180 /* lower 4 bits of the current count are ignored: */
181 #define IBS_OP_CUR_CNT (0xFFFF0ULL<<32)
182 #define IBS_OP_CNT_CTL (1ULL<<19)
183 #define IBS_OP_VAL (1ULL<<18)
184 #define IBS_OP_ENABLE (1ULL<<17)
185 #define IBS_OP_MAX_CNT 0x0000FFFFULL
186 #define IBS_OP_MAX_CNT_EXT 0x007FFFFFULL /* not a register bit mask */
187
188 extern u32 get_ibs_caps(void);
189
190 #ifdef CONFIG_PERF_EVENTS
191 extern void perf_events_lapic_init(void);
192
193 /*
194 * Abuse bit 3 of the cpu eflags register to indicate proper PEBS IP fixups.
195 * This flag is otherwise unused and ABI specified to be 0, so nobody should
196 * care what we do with it.
197 */
198 #define PERF_EFLAGS_EXACT (1UL << 3)
199
200 struct pt_regs;
201 extern unsigned long perf_instruction_pointer(struct pt_regs *regs);
202 extern unsigned long perf_misc_flags(struct pt_regs *regs);
203 #define perf_misc_flags(regs) perf_misc_flags(regs)
204
205 #include <asm/stacktrace.h>
206
207 /*
208 * We abuse bit 3 from flags to pass exact information, see perf_misc_flags
209 * and the comment with PERF_EFLAGS_EXACT.
210 */
211 #define perf_arch_fetch_caller_regs(regs, __ip) { \
212 (regs)->ip = (__ip); \
213 (regs)->bp = caller_frame_pointer(); \
214 (regs)->cs = __KERNEL_CS; \
215 regs->flags = 0; \
216 asm volatile( \
217 _ASM_MOV "%%"_ASM_SP ", %0\n" \
218 : "=m" ((regs)->sp) \
219 :: "memory" \
220 ); \
221 }
222
223 struct perf_guest_switch_msr {
224 unsigned msr;
225 u64 host, guest;
226 };
227
228 extern struct perf_guest_switch_msr *perf_guest_get_msrs(int *nr);
229 extern void perf_get_x86_pmu_capability(struct x86_pmu_capability *cap);
230 #else
231 static inline perf_guest_switch_msr *perf_guest_get_msrs(int *nr)
232 {
233 *nr = 0;
234 return NULL;
235 }
236
237 static inline void perf_get_x86_pmu_capability(struct x86_pmu_capability *cap)
238 {
239 memset(cap, 0, sizeof(*cap));
240 }
241
242 static inline void perf_events_lapic_init(void) { }
243 #endif
244
245 #if defined(CONFIG_PERF_EVENTS) && defined(CONFIG_CPU_SUP_AMD)
246 extern void amd_pmu_enable_virt(void);
247 extern void amd_pmu_disable_virt(void);
248 #else
249 static inline void amd_pmu_enable_virt(void) { }
250 static inline void amd_pmu_disable_virt(void) { }
251 #endif
252
253 #endif /* _ASM_X86_PERF_EVENT_H */
This page took 0.072924 seconds and 5 git commands to generate.