diff options
Diffstat (limited to 'arch/x86/oprofile/op_model_amd.c')
-rw-r--r-- | arch/x86/oprofile/op_model_amd.c | 76 |
1 files changed, 30 insertions, 46 deletions
diff --git a/arch/x86/oprofile/op_model_amd.c b/arch/x86/oprofile/op_model_amd.c index bbf2b68bcc5d..d9faf607b3a6 100644 --- a/arch/x86/oprofile/op_model_amd.c +++ b/arch/x86/oprofile/op_model_amd.c | |||
@@ -15,7 +15,6 @@ | |||
15 | #include <linux/oprofile.h> | 15 | #include <linux/oprofile.h> |
16 | #include <linux/device.h> | 16 | #include <linux/device.h> |
17 | #include <linux/pci.h> | 17 | #include <linux/pci.h> |
18 | #include <linux/percpu.h> | ||
19 | 18 | ||
20 | #include <asm/ptrace.h> | 19 | #include <asm/ptrace.h> |
21 | #include <asm/msr.h> | 20 | #include <asm/msr.h> |
@@ -24,10 +23,8 @@ | |||
24 | #include "op_x86_model.h" | 23 | #include "op_x86_model.h" |
25 | #include "op_counter.h" | 24 | #include "op_counter.h" |
26 | 25 | ||
27 | #define NUM_COUNTERS 32 | 26 | #define NUM_COUNTERS 4 |
28 | #define NUM_HARDWARE_COUNTERS 4 | 27 | #define NUM_CONTROLS 4 |
29 | #define NUM_CONTROLS 32 | ||
30 | #define NUM_HARDWARE_CONTROLS 4 | ||
31 | 28 | ||
32 | #define CTR_IS_RESERVED(msrs, c) (msrs->counters[(c)].addr ? 1 : 0) | 29 | #define CTR_IS_RESERVED(msrs, c) (msrs->counters[(c)].addr ? 1 : 0) |
33 | #define CTR_READ(l, h, msrs, c) do {rdmsr(msrs->counters[(c)].addr, (l), (h)); } while (0) | 30 | #define CTR_READ(l, h, msrs, c) do {rdmsr(msrs->counters[(c)].addr, (l), (h)); } while (0) |
@@ -51,7 +48,6 @@ | |||
51 | #define CTRL_SET_GUEST_ONLY(val, h) (val |= ((h & 1) << 8)) | 48 | #define CTRL_SET_GUEST_ONLY(val, h) (val |= ((h & 1) << 8)) |
52 | 49 | ||
53 | static unsigned long reset_value[NUM_COUNTERS]; | 50 | static unsigned long reset_value[NUM_COUNTERS]; |
54 | DECLARE_PER_CPU(int, switch_index); | ||
55 | 51 | ||
56 | #ifdef CONFIG_OPROFILE_IBS | 52 | #ifdef CONFIG_OPROFILE_IBS |
57 | 53 | ||
@@ -134,17 +130,15 @@ static void op_amd_fill_in_addresses(struct op_msrs * const msrs) | |||
134 | int i; | 130 | int i; |
135 | 131 | ||
136 | for (i = 0; i < NUM_COUNTERS; i++) { | 132 | for (i = 0; i < NUM_COUNTERS; i++) { |
137 | int hw_counter = i % NUM_HARDWARE_COUNTERS; | 133 | if (reserve_perfctr_nmi(MSR_K7_PERFCTR0 + i)) |
138 | if (reserve_perfctr_nmi(MSR_K7_PERFCTR0 + hw_counter)) | 134 | msrs->counters[i].addr = MSR_K7_PERFCTR0 + i; |
139 | msrs->counters[i].addr = MSR_K7_PERFCTR0 + hw_counter; | ||
140 | else | 135 | else |
141 | msrs->counters[i].addr = 0; | 136 | msrs->counters[i].addr = 0; |
142 | } | 137 | } |
143 | 138 | ||
144 | for (i = 0; i < NUM_CONTROLS; i++) { | 139 | for (i = 0; i < NUM_CONTROLS; i++) { |
145 | int hw_control = i % NUM_HARDWARE_CONTROLS; | 140 | if (reserve_evntsel_nmi(MSR_K7_EVNTSEL0 + i)) |
146 | if (reserve_evntsel_nmi(MSR_K7_EVNTSEL0 + hw_control)) | 141 | msrs->controls[i].addr = MSR_K7_EVNTSEL0 + i; |
147 | msrs->controls[i].addr = MSR_K7_EVNTSEL0 + hw_control; | ||
148 | else | 142 | else |
149 | msrs->controls[i].addr = 0; | 143 | msrs->controls[i].addr = 0; |
150 | } | 144 | } |
@@ -156,16 +150,8 @@ static void op_amd_setup_ctrs(struct op_msrs const * const msrs) | |||
156 | unsigned int low, high; | 150 | unsigned int low, high; |
157 | int i; | 151 | int i; |
158 | 152 | ||
159 | for (i = 0; i < NUM_HARDWARE_CONTROLS; ++i) { | ||
160 | int offset = i + __get_cpu_var(switch_index); | ||
161 | if (counter_config[offset].enabled) | ||
162 | reset_value[offset] = counter_config[offset].count; | ||
163 | else | ||
164 | reset_value[offset] = 0; | ||
165 | } | ||
166 | |||
167 | /* clear all counters */ | 153 | /* clear all counters */ |
168 | for (i = 0 ; i < NUM_HARDWARE_CONTROLS; ++i) { | 154 | for (i = 0 ; i < NUM_CONTROLS; ++i) { |
169 | if (unlikely(!CTRL_IS_RESERVED(msrs, i))) | 155 | if (unlikely(!CTRL_IS_RESERVED(msrs, i))) |
170 | continue; | 156 | continue; |
171 | CTRL_READ(low, high, msrs, i); | 157 | CTRL_READ(low, high, msrs, i); |
@@ -175,31 +161,34 @@ static void op_amd_setup_ctrs(struct op_msrs const * const msrs) | |||
175 | } | 161 | } |
176 | 162 | ||
177 | /* avoid a false detection of ctr overflows in NMI handler */ | 163 | /* avoid a false detection of ctr overflows in NMI handler */ |
178 | for (i = 0; i < NUM_HARDWARE_COUNTERS; ++i) { | 164 | for (i = 0; i < NUM_COUNTERS; ++i) { |
179 | if (unlikely(!CTR_IS_RESERVED(msrs, i))) | 165 | if (unlikely(!CTR_IS_RESERVED(msrs, i))) |
180 | continue; | 166 | continue; |
181 | CTR_WRITE(1, msrs, i); | 167 | CTR_WRITE(1, msrs, i); |
182 | } | 168 | } |
183 | 169 | ||
184 | /* enable active counters */ | 170 | /* enable active counters */ |
185 | for (i = 0; i < NUM_HARDWARE_COUNTERS; ++i) { | 171 | for (i = 0; i < NUM_COUNTERS; ++i) { |
186 | int offset = i + __get_cpu_var(switch_index); | 172 | if ((counter_config[i].enabled) && (CTR_IS_RESERVED(msrs, i))) { |
187 | if ((counter_config[offset].enabled) && (CTR_IS_RESERVED(msrs, i))) { | 173 | reset_value[i] = counter_config[i].count; |
188 | CTR_WRITE(counter_config[offset].count, msrs, i); | 174 | |
175 | CTR_WRITE(counter_config[i].count, msrs, i); | ||
189 | 176 | ||
190 | CTRL_READ(low, high, msrs, i); | 177 | CTRL_READ(low, high, msrs, i); |
191 | CTRL_CLEAR_LO(low); | 178 | CTRL_CLEAR_LO(low); |
192 | CTRL_CLEAR_HI(high); | 179 | CTRL_CLEAR_HI(high); |
193 | CTRL_SET_ENABLE(low); | 180 | CTRL_SET_ENABLE(low); |
194 | CTRL_SET_USR(low, counter_config[offset].user); | 181 | CTRL_SET_USR(low, counter_config[i].user); |
195 | CTRL_SET_KERN(low, counter_config[offset].kernel); | 182 | CTRL_SET_KERN(low, counter_config[i].kernel); |
196 | CTRL_SET_UM(low, counter_config[offset].unit_mask); | 183 | CTRL_SET_UM(low, counter_config[i].unit_mask); |
197 | CTRL_SET_EVENT_LOW(low, counter_config[offset].event); | 184 | CTRL_SET_EVENT_LOW(low, counter_config[i].event); |
198 | CTRL_SET_EVENT_HIGH(high, counter_config[offset].event); | 185 | CTRL_SET_EVENT_HIGH(high, counter_config[i].event); |
199 | CTRL_SET_HOST_ONLY(high, 0); | 186 | CTRL_SET_HOST_ONLY(high, 0); |
200 | CTRL_SET_GUEST_ONLY(high, 0); | 187 | CTRL_SET_GUEST_ONLY(high, 0); |
201 | 188 | ||
202 | CTRL_WRITE(low, high, msrs, i); | 189 | CTRL_WRITE(low, high, msrs, i); |
190 | } else { | ||
191 | reset_value[i] = 0; | ||
203 | } | 192 | } |
204 | } | 193 | } |
205 | } | 194 | } |
@@ -287,14 +276,13 @@ static int op_amd_check_ctrs(struct pt_regs * const regs, | |||
287 | unsigned int low, high; | 276 | unsigned int low, high; |
288 | int i; | 277 | int i; |
289 | 278 | ||
290 | for (i = 0 ; i < NUM_HARDWARE_COUNTERS ; ++i) { | 279 | for (i = 0 ; i < NUM_COUNTERS; ++i) { |
291 | int offset = i + __get_cpu_var(switch_index); | 280 | if (!reset_value[i]) |
292 | if (!reset_value[offset]) | ||
293 | continue; | 281 | continue; |
294 | CTR_READ(low, high, msrs, i); | 282 | CTR_READ(low, high, msrs, i); |
295 | if (CTR_OVERFLOWED(low)) { | 283 | if (CTR_OVERFLOWED(low)) { |
296 | oprofile_add_sample(regs, offset); | 284 | oprofile_add_sample(regs, i); |
297 | CTR_WRITE(reset_value[offset], msrs, i); | 285 | CTR_WRITE(reset_value[i], msrs, i); |
298 | } | 286 | } |
299 | } | 287 | } |
300 | 288 | ||
@@ -310,10 +298,8 @@ static void op_amd_start(struct op_msrs const * const msrs) | |||
310 | { | 298 | { |
311 | unsigned int low, high; | 299 | unsigned int low, high; |
312 | int i; | 300 | int i; |
313 | 301 | for (i = 0 ; i < NUM_COUNTERS ; ++i) { | |
314 | for (i = 0 ; i < NUM_HARDWARE_COUNTERS ; ++i) { | 302 | if (reset_value[i]) { |
315 | int offset = i + __get_cpu_var(switch_index); | ||
316 | if (reset_value[offset]) { | ||
317 | CTRL_READ(low, high, msrs, i); | 303 | CTRL_READ(low, high, msrs, i); |
318 | CTRL_SET_ACTIVE(low); | 304 | CTRL_SET_ACTIVE(low); |
319 | CTRL_WRITE(low, high, msrs, i); | 305 | CTRL_WRITE(low, high, msrs, i); |
@@ -343,8 +329,8 @@ static void op_amd_stop(struct op_msrs const * const msrs) | |||
343 | 329 | ||
344 | /* Subtle: stop on all counters to avoid race with | 330 | /* Subtle: stop on all counters to avoid race with |
345 | * setting our pm callback */ | 331 | * setting our pm callback */ |
346 | for (i = 0 ; i < NUM_HARDWARE_COUNTERS ; ++i) { | 332 | for (i = 0 ; i < NUM_COUNTERS ; ++i) { |
347 | if (!reset_value[i + per_cpu(switch_index, smp_processor_id())]) | 333 | if (!reset_value[i]) |
348 | continue; | 334 | continue; |
349 | CTRL_READ(low, high, msrs, i); | 335 | CTRL_READ(low, high, msrs, i); |
350 | CTRL_SET_INACTIVE(low); | 336 | CTRL_SET_INACTIVE(low); |
@@ -370,11 +356,11 @@ static void op_amd_shutdown(struct op_msrs const * const msrs) | |||
370 | { | 356 | { |
371 | int i; | 357 | int i; |
372 | 358 | ||
373 | for (i = 0 ; i < NUM_HARDWARE_COUNTERS ; ++i) { | 359 | for (i = 0 ; i < NUM_COUNTERS ; ++i) { |
374 | if (CTR_IS_RESERVED(msrs, i)) | 360 | if (CTR_IS_RESERVED(msrs, i)) |
375 | release_perfctr_nmi(MSR_K7_PERFCTR0 + i); | 361 | release_perfctr_nmi(MSR_K7_PERFCTR0 + i); |
376 | } | 362 | } |
377 | for (i = 0 ; i < NUM_HARDWARE_COUNTERS ; ++i) { | 363 | for (i = 0 ; i < NUM_CONTROLS ; ++i) { |
378 | if (CTRL_IS_RESERVED(msrs, i)) | 364 | if (CTRL_IS_RESERVED(msrs, i)) |
379 | release_evntsel_nmi(MSR_K7_EVNTSEL0 + i); | 365 | release_evntsel_nmi(MSR_K7_EVNTSEL0 + i); |
380 | } | 366 | } |
@@ -548,8 +534,6 @@ struct op_x86_model_spec const op_amd_spec = { | |||
548 | .exit = op_amd_exit, | 534 | .exit = op_amd_exit, |
549 | .num_counters = NUM_COUNTERS, | 535 | .num_counters = NUM_COUNTERS, |
550 | .num_controls = NUM_CONTROLS, | 536 | .num_controls = NUM_CONTROLS, |
551 | .num_hardware_counters = NUM_HARDWARE_COUNTERS, | ||
552 | .num_hardware_controls = NUM_HARDWARE_CONTROLS, | ||
553 | .fill_in_addresses = &op_amd_fill_in_addresses, | 537 | .fill_in_addresses = &op_amd_fill_in_addresses, |
554 | .setup_ctrs = &op_amd_setup_ctrs, | 538 | .setup_ctrs = &op_amd_setup_ctrs, |
555 | .check_ctrs = &op_amd_check_ctrs, | 539 | .check_ctrs = &op_amd_check_ctrs, |