diff options
Diffstat (limited to 'arch/x86/oprofile/op_model_p4.c')
-rw-r--r-- | arch/x86/oprofile/op_model_p4.c | 175 |
1 files changed, 87 insertions, 88 deletions
diff --git a/arch/x86/oprofile/op_model_p4.c b/arch/x86/oprofile/op_model_p4.c index 56b4757a1f4..43ac5af338d 100644 --- a/arch/x86/oprofile/op_model_p4.c +++ b/arch/x86/oprofile/op_model_p4.c | |||
@@ -10,11 +10,12 @@ | |||
10 | 10 | ||
11 | #include <linux/oprofile.h> | 11 | #include <linux/oprofile.h> |
12 | #include <linux/smp.h> | 12 | #include <linux/smp.h> |
13 | #include <linux/ptrace.h> | ||
14 | #include <linux/nmi.h> | ||
13 | #include <asm/msr.h> | 15 | #include <asm/msr.h> |
14 | #include <asm/ptrace.h> | ||
15 | #include <asm/fixmap.h> | 16 | #include <asm/fixmap.h> |
16 | #include <asm/apic.h> | 17 | #include <asm/apic.h> |
17 | #include <asm/nmi.h> | 18 | |
18 | 19 | ||
19 | #include "op_x86_model.h" | 20 | #include "op_x86_model.h" |
20 | #include "op_counter.h" | 21 | #include "op_counter.h" |
@@ -40,7 +41,7 @@ static unsigned int num_controls = NUM_CONTROLS_NON_HT; | |||
40 | static inline void setup_num_counters(void) | 41 | static inline void setup_num_counters(void) |
41 | { | 42 | { |
42 | #ifdef CONFIG_SMP | 43 | #ifdef CONFIG_SMP |
43 | if (smp_num_siblings == 2){ | 44 | if (smp_num_siblings == 2) { |
44 | num_counters = NUM_COUNTERS_HT2; | 45 | num_counters = NUM_COUNTERS_HT2; |
45 | num_controls = NUM_CONTROLS_HT2; | 46 | num_controls = NUM_CONTROLS_HT2; |
46 | } | 47 | } |
@@ -86,7 +87,7 @@ struct p4_event_binding { | |||
86 | #define CTR_FLAME_2 (1 << 6) | 87 | #define CTR_FLAME_2 (1 << 6) |
87 | #define CTR_IQ_5 (1 << 7) | 88 | #define CTR_IQ_5 (1 << 7) |
88 | 89 | ||
89 | static struct p4_counter_binding p4_counters [NUM_COUNTERS_NON_HT] = { | 90 | static struct p4_counter_binding p4_counters[NUM_COUNTERS_NON_HT] = { |
90 | { CTR_BPU_0, MSR_P4_BPU_PERFCTR0, MSR_P4_BPU_CCCR0 }, | 91 | { CTR_BPU_0, MSR_P4_BPU_PERFCTR0, MSR_P4_BPU_CCCR0 }, |
91 | { CTR_MS_0, MSR_P4_MS_PERFCTR0, MSR_P4_MS_CCCR0 }, | 92 | { CTR_MS_0, MSR_P4_MS_PERFCTR0, MSR_P4_MS_CCCR0 }, |
92 | { CTR_FLAME_0, MSR_P4_FLAME_PERFCTR0, MSR_P4_FLAME_CCCR0 }, | 93 | { CTR_FLAME_0, MSR_P4_FLAME_PERFCTR0, MSR_P4_FLAME_CCCR0 }, |
@@ -97,32 +98,32 @@ static struct p4_counter_binding p4_counters [NUM_COUNTERS_NON_HT] = { | |||
97 | { CTR_IQ_5, MSR_P4_IQ_PERFCTR5, MSR_P4_IQ_CCCR5 } | 98 | { CTR_IQ_5, MSR_P4_IQ_PERFCTR5, MSR_P4_IQ_CCCR5 } |
98 | }; | 99 | }; |
99 | 100 | ||
100 | #define NUM_UNUSED_CCCRS NUM_CCCRS_NON_HT - NUM_COUNTERS_NON_HT | 101 | #define NUM_UNUSED_CCCRS (NUM_CCCRS_NON_HT - NUM_COUNTERS_NON_HT) |
101 | 102 | ||
102 | /* p4 event codes in libop/op_event.h are indices into this table. */ | 103 | /* p4 event codes in libop/op_event.h are indices into this table. */ |
103 | 104 | ||
104 | static struct p4_event_binding p4_events[NUM_EVENTS] = { | 105 | static struct p4_event_binding p4_events[NUM_EVENTS] = { |
105 | 106 | ||
106 | { /* BRANCH_RETIRED */ | 107 | { /* BRANCH_RETIRED */ |
107 | 0x05, 0x06, | 108 | 0x05, 0x06, |
108 | { {CTR_IQ_4, MSR_P4_CRU_ESCR2}, | 109 | { {CTR_IQ_4, MSR_P4_CRU_ESCR2}, |
109 | {CTR_IQ_5, MSR_P4_CRU_ESCR3} } | 110 | {CTR_IQ_5, MSR_P4_CRU_ESCR3} } |
110 | }, | 111 | }, |
111 | 112 | ||
112 | { /* MISPRED_BRANCH_RETIRED */ | 113 | { /* MISPRED_BRANCH_RETIRED */ |
113 | 0x04, 0x03, | 114 | 0x04, 0x03, |
114 | { { CTR_IQ_4, MSR_P4_CRU_ESCR0}, | 115 | { { CTR_IQ_4, MSR_P4_CRU_ESCR0}, |
115 | { CTR_IQ_5, MSR_P4_CRU_ESCR1} } | 116 | { CTR_IQ_5, MSR_P4_CRU_ESCR1} } |
116 | }, | 117 | }, |
117 | 118 | ||
118 | { /* TC_DELIVER_MODE */ | 119 | { /* TC_DELIVER_MODE */ |
119 | 0x01, 0x01, | 120 | 0x01, 0x01, |
120 | { { CTR_MS_0, MSR_P4_TC_ESCR0}, | 121 | { { CTR_MS_0, MSR_P4_TC_ESCR0}, |
121 | { CTR_MS_2, MSR_P4_TC_ESCR1} } | 122 | { CTR_MS_2, MSR_P4_TC_ESCR1} } |
122 | }, | 123 | }, |
123 | 124 | ||
124 | { /* BPU_FETCH_REQUEST */ | 125 | { /* BPU_FETCH_REQUEST */ |
125 | 0x00, 0x03, | 126 | 0x00, 0x03, |
126 | { { CTR_BPU_0, MSR_P4_BPU_ESCR0}, | 127 | { { CTR_BPU_0, MSR_P4_BPU_ESCR0}, |
127 | { CTR_BPU_2, MSR_P4_BPU_ESCR1} } | 128 | { CTR_BPU_2, MSR_P4_BPU_ESCR1} } |
128 | }, | 129 | }, |
@@ -146,7 +147,7 @@ static struct p4_event_binding p4_events[NUM_EVENTS] = { | |||
146 | }, | 147 | }, |
147 | 148 | ||
148 | { /* LOAD_PORT_REPLAY */ | 149 | { /* LOAD_PORT_REPLAY */ |
149 | 0x02, 0x04, | 150 | 0x02, 0x04, |
150 | { { CTR_FLAME_0, MSR_P4_SAAT_ESCR0}, | 151 | { { CTR_FLAME_0, MSR_P4_SAAT_ESCR0}, |
151 | { CTR_FLAME_2, MSR_P4_SAAT_ESCR1} } | 152 | { CTR_FLAME_2, MSR_P4_SAAT_ESCR1} } |
152 | }, | 153 | }, |
@@ -170,43 +171,43 @@ static struct p4_event_binding p4_events[NUM_EVENTS] = { | |||
170 | }, | 171 | }, |
171 | 172 | ||
172 | { /* BSQ_CACHE_REFERENCE */ | 173 | { /* BSQ_CACHE_REFERENCE */ |
173 | 0x07, 0x0c, | 174 | 0x07, 0x0c, |
174 | { { CTR_BPU_0, MSR_P4_BSU_ESCR0}, | 175 | { { CTR_BPU_0, MSR_P4_BSU_ESCR0}, |
175 | { CTR_BPU_2, MSR_P4_BSU_ESCR1} } | 176 | { CTR_BPU_2, MSR_P4_BSU_ESCR1} } |
176 | }, | 177 | }, |
177 | 178 | ||
178 | { /* IOQ_ALLOCATION */ | 179 | { /* IOQ_ALLOCATION */ |
179 | 0x06, 0x03, | 180 | 0x06, 0x03, |
180 | { { CTR_BPU_0, MSR_P4_FSB_ESCR0}, | 181 | { { CTR_BPU_0, MSR_P4_FSB_ESCR0}, |
181 | { 0, 0 } } | 182 | { 0, 0 } } |
182 | }, | 183 | }, |
183 | 184 | ||
184 | { /* IOQ_ACTIVE_ENTRIES */ | 185 | { /* IOQ_ACTIVE_ENTRIES */ |
185 | 0x06, 0x1a, | 186 | 0x06, 0x1a, |
186 | { { CTR_BPU_2, MSR_P4_FSB_ESCR1}, | 187 | { { CTR_BPU_2, MSR_P4_FSB_ESCR1}, |
187 | { 0, 0 } } | 188 | { 0, 0 } } |
188 | }, | 189 | }, |
189 | 190 | ||
190 | { /* FSB_DATA_ACTIVITY */ | 191 | { /* FSB_DATA_ACTIVITY */ |
191 | 0x06, 0x17, | 192 | 0x06, 0x17, |
192 | { { CTR_BPU_0, MSR_P4_FSB_ESCR0}, | 193 | { { CTR_BPU_0, MSR_P4_FSB_ESCR0}, |
193 | { CTR_BPU_2, MSR_P4_FSB_ESCR1} } | 194 | { CTR_BPU_2, MSR_P4_FSB_ESCR1} } |
194 | }, | 195 | }, |
195 | 196 | ||
196 | { /* BSQ_ALLOCATION */ | 197 | { /* BSQ_ALLOCATION */ |
197 | 0x07, 0x05, | 198 | 0x07, 0x05, |
198 | { { CTR_BPU_0, MSR_P4_BSU_ESCR0}, | 199 | { { CTR_BPU_0, MSR_P4_BSU_ESCR0}, |
199 | { 0, 0 } } | 200 | { 0, 0 } } |
200 | }, | 201 | }, |
201 | 202 | ||
202 | { /* BSQ_ACTIVE_ENTRIES */ | 203 | { /* BSQ_ACTIVE_ENTRIES */ |
203 | 0x07, 0x06, | 204 | 0x07, 0x06, |
204 | { { CTR_BPU_2, MSR_P4_BSU_ESCR1 /* guess */}, | 205 | { { CTR_BPU_2, MSR_P4_BSU_ESCR1 /* guess */}, |
205 | { 0, 0 } } | 206 | { 0, 0 } } |
206 | }, | 207 | }, |
207 | 208 | ||
208 | { /* X87_ASSIST */ | 209 | { /* X87_ASSIST */ |
209 | 0x05, 0x03, | 210 | 0x05, 0x03, |
210 | { { CTR_IQ_4, MSR_P4_CRU_ESCR2}, | 211 | { { CTR_IQ_4, MSR_P4_CRU_ESCR2}, |
211 | { CTR_IQ_5, MSR_P4_CRU_ESCR3} } | 212 | { CTR_IQ_5, MSR_P4_CRU_ESCR3} } |
212 | }, | 213 | }, |
@@ -216,21 +217,21 @@ static struct p4_event_binding p4_events[NUM_EVENTS] = { | |||
216 | { { CTR_FLAME_0, MSR_P4_FIRM_ESCR0}, | 217 | { { CTR_FLAME_0, MSR_P4_FIRM_ESCR0}, |
217 | { CTR_FLAME_2, MSR_P4_FIRM_ESCR1} } | 218 | { CTR_FLAME_2, MSR_P4_FIRM_ESCR1} } |
218 | }, | 219 | }, |
219 | 220 | ||
220 | { /* PACKED_SP_UOP */ | 221 | { /* PACKED_SP_UOP */ |
221 | 0x01, 0x08, | 222 | 0x01, 0x08, |
222 | { { CTR_FLAME_0, MSR_P4_FIRM_ESCR0}, | 223 | { { CTR_FLAME_0, MSR_P4_FIRM_ESCR0}, |
223 | { CTR_FLAME_2, MSR_P4_FIRM_ESCR1} } | 224 | { CTR_FLAME_2, MSR_P4_FIRM_ESCR1} } |
224 | }, | 225 | }, |
225 | 226 | ||
226 | { /* PACKED_DP_UOP */ | 227 | { /* PACKED_DP_UOP */ |
227 | 0x01, 0x0c, | 228 | 0x01, 0x0c, |
228 | { { CTR_FLAME_0, MSR_P4_FIRM_ESCR0}, | 229 | { { CTR_FLAME_0, MSR_P4_FIRM_ESCR0}, |
229 | { CTR_FLAME_2, MSR_P4_FIRM_ESCR1} } | 230 | { CTR_FLAME_2, MSR_P4_FIRM_ESCR1} } |
230 | }, | 231 | }, |
231 | 232 | ||
232 | { /* SCALAR_SP_UOP */ | 233 | { /* SCALAR_SP_UOP */ |
233 | 0x01, 0x0a, | 234 | 0x01, 0x0a, |
234 | { { CTR_FLAME_0, MSR_P4_FIRM_ESCR0}, | 235 | { { CTR_FLAME_0, MSR_P4_FIRM_ESCR0}, |
235 | { CTR_FLAME_2, MSR_P4_FIRM_ESCR1} } | 236 | { CTR_FLAME_2, MSR_P4_FIRM_ESCR1} } |
236 | }, | 237 | }, |
@@ -242,31 +243,31 @@ static struct p4_event_binding p4_events[NUM_EVENTS] = { | |||
242 | }, | 243 | }, |
243 | 244 | ||
244 | { /* 64BIT_MMX_UOP */ | 245 | { /* 64BIT_MMX_UOP */ |
245 | 0x01, 0x02, | 246 | 0x01, 0x02, |
246 | { { CTR_FLAME_0, MSR_P4_FIRM_ESCR0}, | 247 | { { CTR_FLAME_0, MSR_P4_FIRM_ESCR0}, |
247 | { CTR_FLAME_2, MSR_P4_FIRM_ESCR1} } | 248 | { CTR_FLAME_2, MSR_P4_FIRM_ESCR1} } |
248 | }, | 249 | }, |
249 | 250 | ||
250 | { /* 128BIT_MMX_UOP */ | 251 | { /* 128BIT_MMX_UOP */ |
251 | 0x01, 0x1a, | 252 | 0x01, 0x1a, |
252 | { { CTR_FLAME_0, MSR_P4_FIRM_ESCR0}, | 253 | { { CTR_FLAME_0, MSR_P4_FIRM_ESCR0}, |
253 | { CTR_FLAME_2, MSR_P4_FIRM_ESCR1} } | 254 | { CTR_FLAME_2, MSR_P4_FIRM_ESCR1} } |
254 | }, | 255 | }, |
255 | 256 | ||
256 | { /* X87_FP_UOP */ | 257 | { /* X87_FP_UOP */ |
257 | 0x01, 0x04, | 258 | 0x01, 0x04, |
258 | { { CTR_FLAME_0, MSR_P4_FIRM_ESCR0}, | 259 | { { CTR_FLAME_0, MSR_P4_FIRM_ESCR0}, |
259 | { CTR_FLAME_2, MSR_P4_FIRM_ESCR1} } | 260 | { CTR_FLAME_2, MSR_P4_FIRM_ESCR1} } |
260 | }, | 261 | }, |
261 | 262 | ||
262 | { /* X87_SIMD_MOVES_UOP */ | 263 | { /* X87_SIMD_MOVES_UOP */ |
263 | 0x01, 0x2e, | 264 | 0x01, 0x2e, |
264 | { { CTR_FLAME_0, MSR_P4_FIRM_ESCR0}, | 265 | { { CTR_FLAME_0, MSR_P4_FIRM_ESCR0}, |
265 | { CTR_FLAME_2, MSR_P4_FIRM_ESCR1} } | 266 | { CTR_FLAME_2, MSR_P4_FIRM_ESCR1} } |
266 | }, | 267 | }, |
267 | 268 | ||
268 | { /* MACHINE_CLEAR */ | 269 | { /* MACHINE_CLEAR */ |
269 | 0x05, 0x02, | 270 | 0x05, 0x02, |
270 | { { CTR_IQ_4, MSR_P4_CRU_ESCR2}, | 271 | { { CTR_IQ_4, MSR_P4_CRU_ESCR2}, |
271 | { CTR_IQ_5, MSR_P4_CRU_ESCR3} } | 272 | { CTR_IQ_5, MSR_P4_CRU_ESCR3} } |
272 | }, | 273 | }, |
@@ -276,9 +277,9 @@ static struct p4_event_binding p4_events[NUM_EVENTS] = { | |||
276 | { { CTR_BPU_0, MSR_P4_FSB_ESCR0}, | 277 | { { CTR_BPU_0, MSR_P4_FSB_ESCR0}, |
277 | { CTR_BPU_2, MSR_P4_FSB_ESCR1} } | 278 | { CTR_BPU_2, MSR_P4_FSB_ESCR1} } |
278 | }, | 279 | }, |
279 | 280 | ||
280 | { /* TC_MS_XFER */ | 281 | { /* TC_MS_XFER */ |
281 | 0x00, 0x05, | 282 | 0x00, 0x05, |
282 | { { CTR_MS_0, MSR_P4_MS_ESCR0}, | 283 | { { CTR_MS_0, MSR_P4_MS_ESCR0}, |
283 | { CTR_MS_2, MSR_P4_MS_ESCR1} } | 284 | { CTR_MS_2, MSR_P4_MS_ESCR1} } |
284 | }, | 285 | }, |
@@ -308,7 +309,7 @@ static struct p4_event_binding p4_events[NUM_EVENTS] = { | |||
308 | }, | 309 | }, |
309 | 310 | ||
310 | { /* INSTR_RETIRED */ | 311 | { /* INSTR_RETIRED */ |
311 | 0x04, 0x02, | 312 | 0x04, 0x02, |
312 | { { CTR_IQ_4, MSR_P4_CRU_ESCR0}, | 313 | { { CTR_IQ_4, MSR_P4_CRU_ESCR0}, |
313 | { CTR_IQ_5, MSR_P4_CRU_ESCR1} } | 314 | { CTR_IQ_5, MSR_P4_CRU_ESCR1} } |
314 | }, | 315 | }, |
@@ -319,14 +320,14 @@ static struct p4_event_binding p4_events[NUM_EVENTS] = { | |||
319 | { CTR_IQ_5, MSR_P4_CRU_ESCR1} } | 320 | { CTR_IQ_5, MSR_P4_CRU_ESCR1} } |
320 | }, | 321 | }, |
321 | 322 | ||
322 | { /* UOP_TYPE */ | 323 | { /* UOP_TYPE */ |
323 | 0x02, 0x02, | 324 | 0x02, 0x02, |
324 | { { CTR_IQ_4, MSR_P4_RAT_ESCR0}, | 325 | { { CTR_IQ_4, MSR_P4_RAT_ESCR0}, |
325 | { CTR_IQ_5, MSR_P4_RAT_ESCR1} } | 326 | { CTR_IQ_5, MSR_P4_RAT_ESCR1} } |
326 | }, | 327 | }, |
327 | 328 | ||
328 | { /* RETIRED_MISPRED_BRANCH_TYPE */ | 329 | { /* RETIRED_MISPRED_BRANCH_TYPE */ |
329 | 0x02, 0x05, | 330 | 0x02, 0x05, |
330 | { { CTR_MS_0, MSR_P4_TBPU_ESCR0}, | 331 | { { CTR_MS_0, MSR_P4_TBPU_ESCR0}, |
331 | { CTR_MS_2, MSR_P4_TBPU_ESCR1} } | 332 | { CTR_MS_2, MSR_P4_TBPU_ESCR1} } |
332 | }, | 333 | }, |
@@ -349,8 +350,8 @@ static struct p4_event_binding p4_events[NUM_EVENTS] = { | |||
349 | #define ESCR_SET_OS_1(escr, os) ((escr) |= (((os) & 1) << 1)) | 350 | #define ESCR_SET_OS_1(escr, os) ((escr) |= (((os) & 1) << 1)) |
350 | #define ESCR_SET_EVENT_SELECT(escr, sel) ((escr) |= (((sel) & 0x3f) << 25)) | 351 | #define ESCR_SET_EVENT_SELECT(escr, sel) ((escr) |= (((sel) & 0x3f) << 25)) |
351 | #define ESCR_SET_EVENT_MASK(escr, mask) ((escr) |= (((mask) & 0xffff) << 9)) | 352 | #define ESCR_SET_EVENT_MASK(escr, mask) ((escr) |= (((mask) & 0xffff) << 9)) |
352 | #define ESCR_READ(escr,high,ev,i) do {rdmsr(ev->bindings[(i)].escr_address, (escr), (high));} while (0) | 353 | #define ESCR_READ(escr, high, ev, i) do {rdmsr(ev->bindings[(i)].escr_address, (escr), (high)); } while (0) |
353 | #define ESCR_WRITE(escr,high,ev,i) do {wrmsr(ev->bindings[(i)].escr_address, (escr), (high));} while (0) | 354 | #define ESCR_WRITE(escr, high, ev, i) do {wrmsr(ev->bindings[(i)].escr_address, (escr), (high)); } while (0) |
354 | 355 | ||
355 | #define CCCR_RESERVED_BITS 0x38030FFF | 356 | #define CCCR_RESERVED_BITS 0x38030FFF |
356 | #define CCCR_CLEAR(cccr) ((cccr) &= CCCR_RESERVED_BITS) | 357 | #define CCCR_CLEAR(cccr) ((cccr) &= CCCR_RESERVED_BITS) |
@@ -360,15 +361,15 @@ static struct p4_event_binding p4_events[NUM_EVENTS] = { | |||
360 | #define CCCR_SET_PMI_OVF_1(cccr) ((cccr) |= (1<<27)) | 361 | #define CCCR_SET_PMI_OVF_1(cccr) ((cccr) |= (1<<27)) |
361 | #define CCCR_SET_ENABLE(cccr) ((cccr) |= (1<<12)) | 362 | #define CCCR_SET_ENABLE(cccr) ((cccr) |= (1<<12)) |
362 | #define CCCR_SET_DISABLE(cccr) ((cccr) &= ~(1<<12)) | 363 | #define CCCR_SET_DISABLE(cccr) ((cccr) &= ~(1<<12)) |
363 | #define CCCR_READ(low, high, i) do {rdmsr(p4_counters[(i)].cccr_address, (low), (high));} while (0) | 364 | #define CCCR_READ(low, high, i) do {rdmsr(p4_counters[(i)].cccr_address, (low), (high)); } while (0) |
364 | #define CCCR_WRITE(low, high, i) do {wrmsr(p4_counters[(i)].cccr_address, (low), (high));} while (0) | 365 | #define CCCR_WRITE(low, high, i) do {wrmsr(p4_counters[(i)].cccr_address, (low), (high)); } while (0) |
365 | #define CCCR_OVF_P(cccr) ((cccr) & (1U<<31)) | 366 | #define CCCR_OVF_P(cccr) ((cccr) & (1U<<31)) |
366 | #define CCCR_CLEAR_OVF(cccr) ((cccr) &= (~(1U<<31))) | 367 | #define CCCR_CLEAR_OVF(cccr) ((cccr) &= (~(1U<<31))) |
367 | 368 | ||
368 | #define CTRL_IS_RESERVED(msrs,c) (msrs->controls[(c)].addr ? 1 : 0) | 369 | #define CTRL_IS_RESERVED(msrs, c) (msrs->controls[(c)].addr ? 1 : 0) |
369 | #define CTR_IS_RESERVED(msrs,c) (msrs->counters[(c)].addr ? 1 : 0) | 370 | #define CTR_IS_RESERVED(msrs, c) (msrs->counters[(c)].addr ? 1 : 0) |
370 | #define CTR_READ(l,h,i) do {rdmsr(p4_counters[(i)].counter_address, (l), (h));} while (0) | 371 | #define CTR_READ(l, h, i) do {rdmsr(p4_counters[(i)].counter_address, (l), (h)); } while (0) |
371 | #define CTR_WRITE(l,i) do {wrmsr(p4_counters[(i)].counter_address, -(u32)(l), -1);} while (0) | 372 | #define CTR_WRITE(l, i) do {wrmsr(p4_counters[(i)].counter_address, -(u32)(l), -1); } while (0) |
372 | #define CTR_OVERFLOW_P(ctr) (!((ctr) & 0x80000000)) | 373 | #define CTR_OVERFLOW_P(ctr) (!((ctr) & 0x80000000)) |
373 | 374 | ||
374 | 375 | ||
@@ -380,7 +381,7 @@ static unsigned int get_stagger(void) | |||
380 | #ifdef CONFIG_SMP | 381 | #ifdef CONFIG_SMP |
381 | int cpu = smp_processor_id(); | 382 | int cpu = smp_processor_id(); |
382 | return (cpu != first_cpu(per_cpu(cpu_sibling_map, cpu))); | 383 | return (cpu != first_cpu(per_cpu(cpu_sibling_map, cpu))); |
383 | #endif | 384 | #endif |
384 | return 0; | 385 | return 0; |
385 | } | 386 | } |
386 | 387 | ||
@@ -395,25 +396,23 @@ static unsigned long reset_value[NUM_COUNTERS_NON_HT]; | |||
395 | 396 | ||
396 | static void p4_fill_in_addresses(struct op_msrs * const msrs) | 397 | static void p4_fill_in_addresses(struct op_msrs * const msrs) |
397 | { | 398 | { |
398 | unsigned int i; | 399 | unsigned int i; |
399 | unsigned int addr, cccraddr, stag; | 400 | unsigned int addr, cccraddr, stag; |
400 | 401 | ||
401 | setup_num_counters(); | 402 | setup_num_counters(); |
402 | stag = get_stagger(); | 403 | stag = get_stagger(); |
403 | 404 | ||
404 | /* initialize some registers */ | 405 | /* initialize some registers */ |
405 | for (i = 0; i < num_counters; ++i) { | 406 | for (i = 0; i < num_counters; ++i) |
406 | msrs->counters[i].addr = 0; | 407 | msrs->counters[i].addr = 0; |
407 | } | 408 | for (i = 0; i < num_controls; ++i) |
408 | for (i = 0; i < num_controls; ++i) { | ||
409 | msrs->controls[i].addr = 0; | 409 | msrs->controls[i].addr = 0; |
410 | } | 410 | |
411 | |||
412 | /* the counter & cccr registers we pay attention to */ | 411 | /* the counter & cccr registers we pay attention to */ |
413 | for (i = 0; i < num_counters; ++i) { | 412 | for (i = 0; i < num_counters; ++i) { |
414 | addr = p4_counters[VIRT_CTR(stag, i)].counter_address; | 413 | addr = p4_counters[VIRT_CTR(stag, i)].counter_address; |
415 | cccraddr = p4_counters[VIRT_CTR(stag, i)].cccr_address; | 414 | cccraddr = p4_counters[VIRT_CTR(stag, i)].cccr_address; |
416 | if (reserve_perfctr_nmi(addr)){ | 415 | if (reserve_perfctr_nmi(addr)) { |
417 | msrs->counters[i].addr = addr; | 416 | msrs->counters[i].addr = addr; |
418 | msrs->controls[i].addr = cccraddr; | 417 | msrs->controls[i].addr = cccraddr; |
419 | } | 418 | } |
@@ -447,22 +446,22 @@ static void p4_fill_in_addresses(struct op_msrs * const msrs) | |||
447 | if (reserve_evntsel_nmi(addr)) | 446 | if (reserve_evntsel_nmi(addr)) |
448 | msrs->controls[i].addr = addr; | 447 | msrs->controls[i].addr = addr; |
449 | } | 448 | } |
450 | 449 | ||
451 | for (addr = MSR_P4_MS_ESCR0 + stag; | 450 | for (addr = MSR_P4_MS_ESCR0 + stag; |
452 | addr <= MSR_P4_TC_ESCR1; ++i, addr += addr_increment()) { | 451 | addr <= MSR_P4_TC_ESCR1; ++i, addr += addr_increment()) { |
453 | if (reserve_evntsel_nmi(addr)) | 452 | if (reserve_evntsel_nmi(addr)) |
454 | msrs->controls[i].addr = addr; | 453 | msrs->controls[i].addr = addr; |
455 | } | 454 | } |
456 | 455 | ||
457 | for (addr = MSR_P4_IX_ESCR0 + stag; | 456 | for (addr = MSR_P4_IX_ESCR0 + stag; |
458 | addr <= MSR_P4_CRU_ESCR3; ++i, addr += addr_increment()) { | 457 | addr <= MSR_P4_CRU_ESCR3; ++i, addr += addr_increment()) { |
459 | if (reserve_evntsel_nmi(addr)) | 458 | if (reserve_evntsel_nmi(addr)) |
460 | msrs->controls[i].addr = addr; | 459 | msrs->controls[i].addr = addr; |
461 | } | 460 | } |
462 | 461 | ||
463 | /* there are 2 remaining non-contiguously located ESCRs */ | 462 | /* there are 2 remaining non-contiguously located ESCRs */ |
464 | 463 | ||
465 | if (num_counters == NUM_COUNTERS_NON_HT) { | 464 | if (num_counters == NUM_COUNTERS_NON_HT) { |
466 | /* standard non-HT CPUs handle both remaining ESCRs*/ | 465 | /* standard non-HT CPUs handle both remaining ESCRs*/ |
467 | if (reserve_evntsel_nmi(MSR_P4_CRU_ESCR5)) | 466 | if (reserve_evntsel_nmi(MSR_P4_CRU_ESCR5)) |
468 | msrs->controls[i++].addr = MSR_P4_CRU_ESCR5; | 467 | msrs->controls[i++].addr = MSR_P4_CRU_ESCR5; |
@@ -498,20 +497,20 @@ static void pmc_setup_one_p4_counter(unsigned int ctr) | |||
498 | unsigned int stag; | 497 | unsigned int stag; |
499 | 498 | ||
500 | stag = get_stagger(); | 499 | stag = get_stagger(); |
501 | 500 | ||
502 | /* convert from counter *number* to counter *bit* */ | 501 | /* convert from counter *number* to counter *bit* */ |
503 | counter_bit = 1 << VIRT_CTR(stag, ctr); | 502 | counter_bit = 1 << VIRT_CTR(stag, ctr); |
504 | 503 | ||
505 | /* find our event binding structure. */ | 504 | /* find our event binding structure. */ |
506 | if (counter_config[ctr].event <= 0 || counter_config[ctr].event > NUM_EVENTS) { | 505 | if (counter_config[ctr].event <= 0 || counter_config[ctr].event > NUM_EVENTS) { |
507 | printk(KERN_ERR | 506 | printk(KERN_ERR |
508 | "oprofile: P4 event code 0x%lx out of range\n", | 507 | "oprofile: P4 event code 0x%lx out of range\n", |
509 | counter_config[ctr].event); | 508 | counter_config[ctr].event); |
510 | return; | 509 | return; |
511 | } | 510 | } |
512 | 511 | ||
513 | ev = &(p4_events[counter_config[ctr].event - 1]); | 512 | ev = &(p4_events[counter_config[ctr].event - 1]); |
514 | 513 | ||
515 | for (i = 0; i < maxbind; i++) { | 514 | for (i = 0; i < maxbind; i++) { |
516 | if (ev->bindings[i].virt_counter & counter_bit) { | 515 | if (ev->bindings[i].virt_counter & counter_bit) { |
517 | 516 | ||
@@ -526,25 +525,24 @@ static void pmc_setup_one_p4_counter(unsigned int ctr) | |||
526 | ESCR_SET_OS_1(escr, counter_config[ctr].kernel); | 525 | ESCR_SET_OS_1(escr, counter_config[ctr].kernel); |
527 | } | 526 | } |
528 | ESCR_SET_EVENT_SELECT(escr, ev->event_select); | 527 | ESCR_SET_EVENT_SELECT(escr, ev->event_select); |
529 | ESCR_SET_EVENT_MASK(escr, counter_config[ctr].unit_mask); | 528 | ESCR_SET_EVENT_MASK(escr, counter_config[ctr].unit_mask); |
530 | ESCR_WRITE(escr, high, ev, i); | 529 | ESCR_WRITE(escr, high, ev, i); |
531 | 530 | ||
532 | /* modify CCCR */ | 531 | /* modify CCCR */ |
533 | CCCR_READ(cccr, high, VIRT_CTR(stag, ctr)); | 532 | CCCR_READ(cccr, high, VIRT_CTR(stag, ctr)); |
534 | CCCR_CLEAR(cccr); | 533 | CCCR_CLEAR(cccr); |
535 | CCCR_SET_REQUIRED_BITS(cccr); | 534 | CCCR_SET_REQUIRED_BITS(cccr); |
536 | CCCR_SET_ESCR_SELECT(cccr, ev->escr_select); | 535 | CCCR_SET_ESCR_SELECT(cccr, ev->escr_select); |
537 | if (stag == 0) { | 536 | if (stag == 0) |
538 | CCCR_SET_PMI_OVF_0(cccr); | 537 | CCCR_SET_PMI_OVF_0(cccr); |
539 | } else { | 538 | else |
540 | CCCR_SET_PMI_OVF_1(cccr); | 539 | CCCR_SET_PMI_OVF_1(cccr); |
541 | } | ||
542 | CCCR_WRITE(cccr, high, VIRT_CTR(stag, ctr)); | 540 | CCCR_WRITE(cccr, high, VIRT_CTR(stag, ctr)); |
543 | return; | 541 | return; |
544 | } | 542 | } |
545 | } | 543 | } |
546 | 544 | ||
547 | printk(KERN_ERR | 545 | printk(KERN_ERR |
548 | "oprofile: P4 event code 0x%lx no binding, stag %d ctr %d\n", | 546 | "oprofile: P4 event code 0x%lx no binding, stag %d ctr %d\n", |
549 | counter_config[ctr].event, stag, ctr); | 547 | counter_config[ctr].event, stag, ctr); |
550 | } | 548 | } |
@@ -559,14 +557,14 @@ static void p4_setup_ctrs(struct op_msrs const * const msrs) | |||
559 | stag = get_stagger(); | 557 | stag = get_stagger(); |
560 | 558 | ||
561 | rdmsr(MSR_IA32_MISC_ENABLE, low, high); | 559 | rdmsr(MSR_IA32_MISC_ENABLE, low, high); |
562 | if (! MISC_PMC_ENABLED_P(low)) { | 560 | if (!MISC_PMC_ENABLED_P(low)) { |
563 | printk(KERN_ERR "oprofile: P4 PMC not available\n"); | 561 | printk(KERN_ERR "oprofile: P4 PMC not available\n"); |
564 | return; | 562 | return; |
565 | } | 563 | } |
566 | 564 | ||
567 | /* clear the cccrs we will use */ | 565 | /* clear the cccrs we will use */ |
568 | for (i = 0 ; i < num_counters ; i++) { | 566 | for (i = 0 ; i < num_counters ; i++) { |
569 | if (unlikely(!CTRL_IS_RESERVED(msrs,i))) | 567 | if (unlikely(!CTRL_IS_RESERVED(msrs, i))) |
570 | continue; | 568 | continue; |
571 | rdmsr(p4_counters[VIRT_CTR(stag, i)].cccr_address, low, high); | 569 | rdmsr(p4_counters[VIRT_CTR(stag, i)].cccr_address, low, high); |
572 | CCCR_CLEAR(low); | 570 | CCCR_CLEAR(low); |
@@ -576,14 +574,14 @@ static void p4_setup_ctrs(struct op_msrs const * const msrs) | |||
576 | 574 | ||
577 | /* clear all escrs (including those outside our concern) */ | 575 | /* clear all escrs (including those outside our concern) */ |
578 | for (i = num_counters; i < num_controls; i++) { | 576 | for (i = num_counters; i < num_controls; i++) { |
579 | if (unlikely(!CTRL_IS_RESERVED(msrs,i))) | 577 | if (unlikely(!CTRL_IS_RESERVED(msrs, i))) |
580 | continue; | 578 | continue; |
581 | wrmsr(msrs->controls[i].addr, 0, 0); | 579 | wrmsr(msrs->controls[i].addr, 0, 0); |
582 | } | 580 | } |
583 | 581 | ||
584 | /* setup all counters */ | 582 | /* setup all counters */ |
585 | for (i = 0 ; i < num_counters ; ++i) { | 583 | for (i = 0 ; i < num_counters ; ++i) { |
586 | if ((counter_config[i].enabled) && (CTRL_IS_RESERVED(msrs,i))) { | 584 | if ((counter_config[i].enabled) && (CTRL_IS_RESERVED(msrs, i))) { |
587 | reset_value[i] = counter_config[i].count; | 585 | reset_value[i] = counter_config[i].count; |
588 | pmc_setup_one_p4_counter(i); | 586 | pmc_setup_one_p4_counter(i); |
589 | CTR_WRITE(counter_config[i].count, VIRT_CTR(stag, i)); | 587 | CTR_WRITE(counter_config[i].count, VIRT_CTR(stag, i)); |
@@ -603,11 +601,11 @@ static int p4_check_ctrs(struct pt_regs * const regs, | |||
603 | stag = get_stagger(); | 601 | stag = get_stagger(); |
604 | 602 | ||
605 | for (i = 0; i < num_counters; ++i) { | 603 | for (i = 0; i < num_counters; ++i) { |
606 | 604 | ||
607 | if (!reset_value[i]) | 605 | if (!reset_value[i]) |
608 | continue; | 606 | continue; |
609 | 607 | ||
610 | /* | 608 | /* |
611 | * there is some eccentricity in the hardware which | 609 | * there is some eccentricity in the hardware which |
612 | * requires that we perform 2 extra corrections: | 610 | * requires that we perform 2 extra corrections: |
613 | * | 611 | * |
@@ -616,24 +614,24 @@ static int p4_check_ctrs(struct pt_regs * const regs, | |||
616 | * | 614 | * |
617 | * - write the counter back twice to ensure it gets | 615 | * - write the counter back twice to ensure it gets |
618 | * updated properly. | 616 | * updated properly. |
619 | * | 617 | * |
620 | * the former seems to be related to extra NMIs happening | 618 | * the former seems to be related to extra NMIs happening |
621 | * during the current NMI; the latter is reported as errata | 619 | * during the current NMI; the latter is reported as errata |
622 | * N15 in intel doc 249199-029, pentium 4 specification | 620 | * N15 in intel doc 249199-029, pentium 4 specification |
623 | * update, though their suggested work-around does not | 621 | * update, though their suggested work-around does not |
624 | * appear to solve the problem. | 622 | * appear to solve the problem. |
625 | */ | 623 | */ |
626 | 624 | ||
627 | real = VIRT_CTR(stag, i); | 625 | real = VIRT_CTR(stag, i); |
628 | 626 | ||
629 | CCCR_READ(low, high, real); | 627 | CCCR_READ(low, high, real); |
630 | CTR_READ(ctr, high, real); | 628 | CTR_READ(ctr, high, real); |
631 | if (CCCR_OVF_P(low) || CTR_OVERFLOW_P(ctr)) { | 629 | if (CCCR_OVF_P(low) || CTR_OVERFLOW_P(ctr)) { |
632 | oprofile_add_sample(regs, i); | 630 | oprofile_add_sample(regs, i); |
633 | CTR_WRITE(reset_value[i], real); | 631 | CTR_WRITE(reset_value[i], real); |
634 | CCCR_CLEAR_OVF(low); | 632 | CCCR_CLEAR_OVF(low); |
635 | CCCR_WRITE(low, high, real); | 633 | CCCR_WRITE(low, high, real); |
636 | CTR_WRITE(reset_value[i], real); | 634 | CTR_WRITE(reset_value[i], real); |
637 | } | 635 | } |
638 | } | 636 | } |
639 | 637 | ||
@@ -683,15 +681,16 @@ static void p4_shutdown(struct op_msrs const * const msrs) | |||
683 | int i; | 681 | int i; |
684 | 682 | ||
685 | for (i = 0 ; i < num_counters ; ++i) { | 683 | for (i = 0 ; i < num_counters ; ++i) { |
686 | if (CTR_IS_RESERVED(msrs,i)) | 684 | if (CTR_IS_RESERVED(msrs, i)) |
687 | release_perfctr_nmi(msrs->counters[i].addr); | 685 | release_perfctr_nmi(msrs->counters[i].addr); |
688 | } | 686 | } |
689 | /* some of the control registers are specially reserved in | 687 | /* |
688 | * some of the control registers are specially reserved in | ||
690 | * conjunction with the counter registers (hence the starting offset). | 689 | * conjunction with the counter registers (hence the starting offset). |
691 | * This saves a few bits. | 690 | * This saves a few bits. |
692 | */ | 691 | */ |
693 | for (i = num_counters ; i < num_controls ; ++i) { | 692 | for (i = num_counters ; i < num_controls ; ++i) { |
694 | if (CTRL_IS_RESERVED(msrs,i)) | 693 | if (CTRL_IS_RESERVED(msrs, i)) |
695 | release_evntsel_nmi(msrs->controls[i].addr); | 694 | release_evntsel_nmi(msrs->controls[i].addr); |
696 | } | 695 | } |
697 | } | 696 | } |