aboutsummaryrefslogtreecommitdiffstats
path: root/arch
diff options
context:
space:
mode:
authorLinus Torvalds <torvalds@linux-foundation.org>2010-05-18 11:18:07 -0400
committerLinus Torvalds <torvalds@linux-foundation.org>2010-05-18 11:18:07 -0400
commit3aaf51ace5975050ab43c7d4d7e439e0ae7d13d7 (patch)
tree3ceb741d8b78c6dc78be3fd2e4f8aac443044787 /arch
parentf262af3d08d3fffc4e11277d3a177b2d67ea2aba (diff)
parentcc49b092d308f8ea8634134b0d95d831a88a674b (diff)
Merge branch 'oprofile-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip
* 'oprofile-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip: (24 commits) oprofile/x86: make AMD IBS hotplug capable oprofile/x86: notify cpus only when daemon is running oprofile/x86: reordering some functions oprofile/x86: stop disabled counters in nmi handler oprofile/x86: protect cpu hotplug sections oprofile/x86: remove CONFIG_SMP macros oprofile/x86: fix uninitialized counter usage during cpu hotplug oprofile/x86: remove duplicate IBS capability check oprofile/x86: move IBS code oprofile/x86: return -EBUSY if counters are already reserved oprofile/x86: moving shutdown functions oprofile/x86: reserve counter msrs pairwise oprofile/x86: rework error handler in nmi_setup() oprofile: update file list in MAINTAINERS file oprofile: protect from not being in an IRQ context oprofile: remove double ring buffering ring-buffer: Add lost event count to end of sub buffer tracing: Show the lost events in the trace_pipe output ring-buffer: Add place holder recording of dropped events tracing: Fix compile error in module tracepoints when MODULE_UNLOAD not set ...
Diffstat (limited to 'arch')
-rw-r--r--arch/x86/oprofile/nmi_int.c199
-rw-r--r--arch/x86/oprofile/op_model_amd.c280
-rw-r--r--arch/x86/oprofile/op_model_p4.c52
-rw-r--r--arch/x86/oprofile/op_model_ppro.c77
-rw-r--r--arch/x86/oprofile/op_x86_model.h4
5 files changed, 321 insertions, 291 deletions
diff --git a/arch/x86/oprofile/nmi_int.c b/arch/x86/oprofile/nmi_int.c
index 2c505ee71014..b28d2f1253bb 100644
--- a/arch/x86/oprofile/nmi_int.c
+++ b/arch/x86/oprofile/nmi_int.c
@@ -31,8 +31,9 @@ static struct op_x86_model_spec *model;
31static DEFINE_PER_CPU(struct op_msrs, cpu_msrs); 31static DEFINE_PER_CPU(struct op_msrs, cpu_msrs);
32static DEFINE_PER_CPU(unsigned long, saved_lvtpc); 32static DEFINE_PER_CPU(unsigned long, saved_lvtpc);
33 33
34/* 0 == registered but off, 1 == registered and on */ 34/* must be protected with get_online_cpus()/put_online_cpus(): */
35static int nmi_enabled = 0; 35static int nmi_enabled;
36static int ctr_running;
36 37
37struct op_counter_config counter_config[OP_MAX_COUNTER]; 38struct op_counter_config counter_config[OP_MAX_COUNTER];
38 39
@@ -61,12 +62,16 @@ static int profile_exceptions_notify(struct notifier_block *self,
61{ 62{
62 struct die_args *args = (struct die_args *)data; 63 struct die_args *args = (struct die_args *)data;
63 int ret = NOTIFY_DONE; 64 int ret = NOTIFY_DONE;
64 int cpu = smp_processor_id();
65 65
66 switch (val) { 66 switch (val) {
67 case DIE_NMI: 67 case DIE_NMI:
68 case DIE_NMI_IPI: 68 case DIE_NMI_IPI:
69 model->check_ctrs(args->regs, &per_cpu(cpu_msrs, cpu)); 69 if (ctr_running)
70 model->check_ctrs(args->regs, &__get_cpu_var(cpu_msrs));
71 else if (!nmi_enabled)
72 break;
73 else
74 model->stop(&__get_cpu_var(cpu_msrs));
70 ret = NOTIFY_STOP; 75 ret = NOTIFY_STOP;
71 break; 76 break;
72 default: 77 default:
@@ -95,24 +100,36 @@ static void nmi_cpu_save_registers(struct op_msrs *msrs)
95static void nmi_cpu_start(void *dummy) 100static void nmi_cpu_start(void *dummy)
96{ 101{
97 struct op_msrs const *msrs = &__get_cpu_var(cpu_msrs); 102 struct op_msrs const *msrs = &__get_cpu_var(cpu_msrs);
98 model->start(msrs); 103 if (!msrs->controls)
104 WARN_ON_ONCE(1);
105 else
106 model->start(msrs);
99} 107}
100 108
101static int nmi_start(void) 109static int nmi_start(void)
102{ 110{
111 get_online_cpus();
103 on_each_cpu(nmi_cpu_start, NULL, 1); 112 on_each_cpu(nmi_cpu_start, NULL, 1);
113 ctr_running = 1;
114 put_online_cpus();
104 return 0; 115 return 0;
105} 116}
106 117
107static void nmi_cpu_stop(void *dummy) 118static void nmi_cpu_stop(void *dummy)
108{ 119{
109 struct op_msrs const *msrs = &__get_cpu_var(cpu_msrs); 120 struct op_msrs const *msrs = &__get_cpu_var(cpu_msrs);
110 model->stop(msrs); 121 if (!msrs->controls)
122 WARN_ON_ONCE(1);
123 else
124 model->stop(msrs);
111} 125}
112 126
113static void nmi_stop(void) 127static void nmi_stop(void)
114{ 128{
129 get_online_cpus();
115 on_each_cpu(nmi_cpu_stop, NULL, 1); 130 on_each_cpu(nmi_cpu_stop, NULL, 1);
131 ctr_running = 0;
132 put_online_cpus();
116} 133}
117 134
118#ifdef CONFIG_OPROFILE_EVENT_MULTIPLEX 135#ifdef CONFIG_OPROFILE_EVENT_MULTIPLEX
@@ -252,7 +269,10 @@ static int nmi_switch_event(void)
252 if (nmi_multiplex_on() < 0) 269 if (nmi_multiplex_on() < 0)
253 return -EINVAL; /* not necessary */ 270 return -EINVAL; /* not necessary */
254 271
255 on_each_cpu(nmi_cpu_switch, NULL, 1); 272 get_online_cpus();
273 if (ctr_running)
274 on_each_cpu(nmi_cpu_switch, NULL, 1);
275 put_online_cpus();
256 276
257 return 0; 277 return 0;
258} 278}
@@ -295,6 +315,7 @@ static void free_msrs(void)
295 kfree(per_cpu(cpu_msrs, i).controls); 315 kfree(per_cpu(cpu_msrs, i).controls);
296 per_cpu(cpu_msrs, i).controls = NULL; 316 per_cpu(cpu_msrs, i).controls = NULL;
297 } 317 }
318 nmi_shutdown_mux();
298} 319}
299 320
300static int allocate_msrs(void) 321static int allocate_msrs(void)
@@ -307,14 +328,21 @@ static int allocate_msrs(void)
307 per_cpu(cpu_msrs, i).counters = kzalloc(counters_size, 328 per_cpu(cpu_msrs, i).counters = kzalloc(counters_size,
308 GFP_KERNEL); 329 GFP_KERNEL);
309 if (!per_cpu(cpu_msrs, i).counters) 330 if (!per_cpu(cpu_msrs, i).counters)
310 return 0; 331 goto fail;
311 per_cpu(cpu_msrs, i).controls = kzalloc(controls_size, 332 per_cpu(cpu_msrs, i).controls = kzalloc(controls_size,
312 GFP_KERNEL); 333 GFP_KERNEL);
313 if (!per_cpu(cpu_msrs, i).controls) 334 if (!per_cpu(cpu_msrs, i).controls)
314 return 0; 335 goto fail;
315 } 336 }
316 337
338 if (!nmi_setup_mux())
339 goto fail;
340
317 return 1; 341 return 1;
342
343fail:
344 free_msrs();
345 return 0;
318} 346}
319 347
320static void nmi_cpu_setup(void *dummy) 348static void nmi_cpu_setup(void *dummy)
@@ -336,49 +364,6 @@ static struct notifier_block profile_exceptions_nb = {
336 .priority = 2 364 .priority = 2
337}; 365};
338 366
339static int nmi_setup(void)
340{
341 int err = 0;
342 int cpu;
343
344 if (!allocate_msrs())
345 err = -ENOMEM;
346 else if (!nmi_setup_mux())
347 err = -ENOMEM;
348 else
349 err = register_die_notifier(&profile_exceptions_nb);
350
351 if (err) {
352 free_msrs();
353 nmi_shutdown_mux();
354 return err;
355 }
356
357 /* We need to serialize save and setup for HT because the subset
358 * of msrs are distinct for save and setup operations
359 */
360
361 /* Assume saved/restored counters are the same on all CPUs */
362 model->fill_in_addresses(&per_cpu(cpu_msrs, 0));
363 for_each_possible_cpu(cpu) {
364 if (!cpu)
365 continue;
366
367 memcpy(per_cpu(cpu_msrs, cpu).counters,
368 per_cpu(cpu_msrs, 0).counters,
369 sizeof(struct op_msr) * model->num_counters);
370
371 memcpy(per_cpu(cpu_msrs, cpu).controls,
372 per_cpu(cpu_msrs, 0).controls,
373 sizeof(struct op_msr) * model->num_controls);
374
375 mux_clone(cpu);
376 }
377 on_each_cpu(nmi_cpu_setup, NULL, 1);
378 nmi_enabled = 1;
379 return 0;
380}
381
382static void nmi_cpu_restore_registers(struct op_msrs *msrs) 367static void nmi_cpu_restore_registers(struct op_msrs *msrs)
383{ 368{
384 struct op_msr *counters = msrs->counters; 369 struct op_msr *counters = msrs->counters;
@@ -412,20 +397,24 @@ static void nmi_cpu_shutdown(void *dummy)
412 apic_write(APIC_LVTPC, per_cpu(saved_lvtpc, cpu)); 397 apic_write(APIC_LVTPC, per_cpu(saved_lvtpc, cpu));
413 apic_write(APIC_LVTERR, v); 398 apic_write(APIC_LVTERR, v);
414 nmi_cpu_restore_registers(msrs); 399 nmi_cpu_restore_registers(msrs);
400 if (model->cpu_down)
401 model->cpu_down();
415} 402}
416 403
417static void nmi_shutdown(void) 404static void nmi_cpu_up(void *dummy)
418{ 405{
419 struct op_msrs *msrs; 406 if (nmi_enabled)
407 nmi_cpu_setup(dummy);
408 if (ctr_running)
409 nmi_cpu_start(dummy);
410}
420 411
421 nmi_enabled = 0; 412static void nmi_cpu_down(void *dummy)
422 on_each_cpu(nmi_cpu_shutdown, NULL, 1); 413{
423 unregister_die_notifier(&profile_exceptions_nb); 414 if (ctr_running)
424 nmi_shutdown_mux(); 415 nmi_cpu_stop(dummy);
425 msrs = &get_cpu_var(cpu_msrs); 416 if (nmi_enabled)
426 model->shutdown(msrs); 417 nmi_cpu_shutdown(dummy);
427 free_msrs();
428 put_cpu_var(cpu_msrs);
429} 418}
430 419
431static int nmi_create_files(struct super_block *sb, struct dentry *root) 420static int nmi_create_files(struct super_block *sb, struct dentry *root)
@@ -457,7 +446,6 @@ static int nmi_create_files(struct super_block *sb, struct dentry *root)
457 return 0; 446 return 0;
458} 447}
459 448
460#ifdef CONFIG_SMP
461static int oprofile_cpu_notifier(struct notifier_block *b, unsigned long action, 449static int oprofile_cpu_notifier(struct notifier_block *b, unsigned long action,
462 void *data) 450 void *data)
463{ 451{
@@ -465,10 +453,10 @@ static int oprofile_cpu_notifier(struct notifier_block *b, unsigned long action,
465 switch (action) { 453 switch (action) {
466 case CPU_DOWN_FAILED: 454 case CPU_DOWN_FAILED:
467 case CPU_ONLINE: 455 case CPU_ONLINE:
468 smp_call_function_single(cpu, nmi_cpu_start, NULL, 0); 456 smp_call_function_single(cpu, nmi_cpu_up, NULL, 0);
469 break; 457 break;
470 case CPU_DOWN_PREPARE: 458 case CPU_DOWN_PREPARE:
471 smp_call_function_single(cpu, nmi_cpu_stop, NULL, 1); 459 smp_call_function_single(cpu, nmi_cpu_down, NULL, 1);
472 break; 460 break;
473 } 461 }
474 return NOTIFY_DONE; 462 return NOTIFY_DONE;
@@ -477,7 +465,75 @@ static int oprofile_cpu_notifier(struct notifier_block *b, unsigned long action,
477static struct notifier_block oprofile_cpu_nb = { 465static struct notifier_block oprofile_cpu_nb = {
478 .notifier_call = oprofile_cpu_notifier 466 .notifier_call = oprofile_cpu_notifier
479}; 467};
480#endif 468
469static int nmi_setup(void)
470{
471 int err = 0;
472 int cpu;
473
474 if (!allocate_msrs())
475 return -ENOMEM;
476
477 /* We need to serialize save and setup for HT because the subset
478 * of msrs are distinct for save and setup operations
479 */
480
481 /* Assume saved/restored counters are the same on all CPUs */
482 err = model->fill_in_addresses(&per_cpu(cpu_msrs, 0));
483 if (err)
484 goto fail;
485
486 for_each_possible_cpu(cpu) {
487 if (!cpu)
488 continue;
489
490 memcpy(per_cpu(cpu_msrs, cpu).counters,
491 per_cpu(cpu_msrs, 0).counters,
492 sizeof(struct op_msr) * model->num_counters);
493
494 memcpy(per_cpu(cpu_msrs, cpu).controls,
495 per_cpu(cpu_msrs, 0).controls,
496 sizeof(struct op_msr) * model->num_controls);
497
498 mux_clone(cpu);
499 }
500
501 nmi_enabled = 0;
502 ctr_running = 0;
503 barrier();
504 err = register_die_notifier(&profile_exceptions_nb);
505 if (err)
506 goto fail;
507
508 get_online_cpus();
509 register_cpu_notifier(&oprofile_cpu_nb);
510 on_each_cpu(nmi_cpu_setup, NULL, 1);
511 nmi_enabled = 1;
512 put_online_cpus();
513
514 return 0;
515fail:
516 free_msrs();
517 return err;
518}
519
520static void nmi_shutdown(void)
521{
522 struct op_msrs *msrs;
523
524 get_online_cpus();
525 unregister_cpu_notifier(&oprofile_cpu_nb);
526 on_each_cpu(nmi_cpu_shutdown, NULL, 1);
527 nmi_enabled = 0;
528 ctr_running = 0;
529 put_online_cpus();
530 barrier();
531 unregister_die_notifier(&profile_exceptions_nb);
532 msrs = &get_cpu_var(cpu_msrs);
533 model->shutdown(msrs);
534 free_msrs();
535 put_cpu_var(cpu_msrs);
536}
481 537
482#ifdef CONFIG_PM 538#ifdef CONFIG_PM
483 539
@@ -687,9 +743,6 @@ int __init op_nmi_init(struct oprofile_operations *ops)
687 return -ENODEV; 743 return -ENODEV;
688 } 744 }
689 745
690#ifdef CONFIG_SMP
691 register_cpu_notifier(&oprofile_cpu_nb);
692#endif
693 /* default values, can be overwritten by model */ 746 /* default values, can be overwritten by model */
694 ops->create_files = nmi_create_files; 747 ops->create_files = nmi_create_files;
695 ops->setup = nmi_setup; 748 ops->setup = nmi_setup;
@@ -716,12 +769,6 @@ int __init op_nmi_init(struct oprofile_operations *ops)
716 769
717void op_nmi_exit(void) 770void op_nmi_exit(void)
718{ 771{
719 if (using_nmi) { 772 if (using_nmi)
720 exit_sysfs(); 773 exit_sysfs();
721#ifdef CONFIG_SMP
722 unregister_cpu_notifier(&oprofile_cpu_nb);
723#endif
724 }
725 if (model->exit)
726 model->exit();
727} 774}
diff --git a/arch/x86/oprofile/op_model_amd.c b/arch/x86/oprofile/op_model_amd.c
index 090cbbec7dbd..b67a6b5aa8d4 100644
--- a/arch/x86/oprofile/op_model_amd.c
+++ b/arch/x86/oprofile/op_model_amd.c
@@ -30,13 +30,10 @@
30#include "op_counter.h" 30#include "op_counter.h"
31 31
32#define NUM_COUNTERS 4 32#define NUM_COUNTERS 4
33#define NUM_CONTROLS 4
34#ifdef CONFIG_OPROFILE_EVENT_MULTIPLEX 33#ifdef CONFIG_OPROFILE_EVENT_MULTIPLEX
35#define NUM_VIRT_COUNTERS 32 34#define NUM_VIRT_COUNTERS 32
36#define NUM_VIRT_CONTROLS 32
37#else 35#else
38#define NUM_VIRT_COUNTERS NUM_COUNTERS 36#define NUM_VIRT_COUNTERS NUM_COUNTERS
39#define NUM_VIRT_CONTROLS NUM_CONTROLS
40#endif 37#endif
41 38
42#define OP_EVENT_MASK 0x0FFF 39#define OP_EVENT_MASK 0x0FFF
@@ -105,102 +102,6 @@ static u32 get_ibs_caps(void)
105 return ibs_caps; 102 return ibs_caps;
106} 103}
107 104
108#ifdef CONFIG_OPROFILE_EVENT_MULTIPLEX
109
110static void op_mux_switch_ctrl(struct op_x86_model_spec const *model,
111 struct op_msrs const * const msrs)
112{
113 u64 val;
114 int i;
115
116 /* enable active counters */
117 for (i = 0; i < NUM_COUNTERS; ++i) {
118 int virt = op_x86_phys_to_virt(i);
119 if (!reset_value[virt])
120 continue;
121 rdmsrl(msrs->controls[i].addr, val);
122 val &= model->reserved;
123 val |= op_x86_get_ctrl(model, &counter_config[virt]);
124 wrmsrl(msrs->controls[i].addr, val);
125 }
126}
127
128#endif
129
130/* functions for op_amd_spec */
131
132static void op_amd_fill_in_addresses(struct op_msrs * const msrs)
133{
134 int i;
135
136 for (i = 0; i < NUM_COUNTERS; i++) {
137 if (reserve_perfctr_nmi(MSR_K7_PERFCTR0 + i))
138 msrs->counters[i].addr = MSR_K7_PERFCTR0 + i;
139 }
140
141 for (i = 0; i < NUM_CONTROLS; i++) {
142 if (reserve_evntsel_nmi(MSR_K7_EVNTSEL0 + i))
143 msrs->controls[i].addr = MSR_K7_EVNTSEL0 + i;
144 }
145}
146
147static void op_amd_setup_ctrs(struct op_x86_model_spec const *model,
148 struct op_msrs const * const msrs)
149{
150 u64 val;
151 int i;
152
153 /* setup reset_value */
154 for (i = 0; i < NUM_VIRT_COUNTERS; ++i) {
155 if (counter_config[i].enabled
156 && msrs->counters[op_x86_virt_to_phys(i)].addr)
157 reset_value[i] = counter_config[i].count;
158 else
159 reset_value[i] = 0;
160 }
161
162 /* clear all counters */
163 for (i = 0; i < NUM_CONTROLS; ++i) {
164 if (unlikely(!msrs->controls[i].addr)) {
165 if (counter_config[i].enabled && !smp_processor_id())
166 /*
167 * counter is reserved, this is on all
168 * cpus, so report only for cpu #0
169 */
170 op_x86_warn_reserved(i);
171 continue;
172 }
173 rdmsrl(msrs->controls[i].addr, val);
174 if (val & ARCH_PERFMON_EVENTSEL_ENABLE)
175 op_x86_warn_in_use(i);
176 val &= model->reserved;
177 wrmsrl(msrs->controls[i].addr, val);
178 }
179
180 /* avoid a false detection of ctr overflows in NMI handler */
181 for (i = 0; i < NUM_COUNTERS; ++i) {
182 if (unlikely(!msrs->counters[i].addr))
183 continue;
184 wrmsrl(msrs->counters[i].addr, -1LL);
185 }
186
187 /* enable active counters */
188 for (i = 0; i < NUM_COUNTERS; ++i) {
189 int virt = op_x86_phys_to_virt(i);
190 if (!reset_value[virt])
191 continue;
192
193 /* setup counter registers */
194 wrmsrl(msrs->counters[i].addr, -(u64)reset_value[virt]);
195
196 /* setup control registers */
197 rdmsrl(msrs->controls[i].addr, val);
198 val &= model->reserved;
199 val |= op_x86_get_ctrl(model, &counter_config[virt]);
200 wrmsrl(msrs->controls[i].addr, val);
201 }
202}
203
204/* 105/*
205 * 16-bit Linear Feedback Shift Register (LFSR) 106 * 16-bit Linear Feedback Shift Register (LFSR)
206 * 107 *
@@ -365,6 +266,125 @@ static void op_amd_stop_ibs(void)
365 wrmsrl(MSR_AMD64_IBSOPCTL, 0); 266 wrmsrl(MSR_AMD64_IBSOPCTL, 0);
366} 267}
367 268
269#ifdef CONFIG_OPROFILE_EVENT_MULTIPLEX
270
271static void op_mux_switch_ctrl(struct op_x86_model_spec const *model,
272 struct op_msrs const * const msrs)
273{
274 u64 val;
275 int i;
276
277 /* enable active counters */
278 for (i = 0; i < NUM_COUNTERS; ++i) {
279 int virt = op_x86_phys_to_virt(i);
280 if (!reset_value[virt])
281 continue;
282 rdmsrl(msrs->controls[i].addr, val);
283 val &= model->reserved;
284 val |= op_x86_get_ctrl(model, &counter_config[virt]);
285 wrmsrl(msrs->controls[i].addr, val);
286 }
287}
288
289#endif
290
291/* functions for op_amd_spec */
292
293static void op_amd_shutdown(struct op_msrs const * const msrs)
294{
295 int i;
296
297 for (i = 0; i < NUM_COUNTERS; ++i) {
298 if (!msrs->counters[i].addr)
299 continue;
300 release_perfctr_nmi(MSR_K7_PERFCTR0 + i);
301 release_evntsel_nmi(MSR_K7_EVNTSEL0 + i);
302 }
303}
304
305static int op_amd_fill_in_addresses(struct op_msrs * const msrs)
306{
307 int i;
308
309 for (i = 0; i < NUM_COUNTERS; i++) {
310 if (!reserve_perfctr_nmi(MSR_K7_PERFCTR0 + i))
311 goto fail;
312 if (!reserve_evntsel_nmi(MSR_K7_EVNTSEL0 + i)) {
313 release_perfctr_nmi(MSR_K7_PERFCTR0 + i);
314 goto fail;
315 }
316 /* both registers must be reserved */
317 msrs->counters[i].addr = MSR_K7_PERFCTR0 + i;
318 msrs->controls[i].addr = MSR_K7_EVNTSEL0 + i;
319 continue;
320 fail:
321 if (!counter_config[i].enabled)
322 continue;
323 op_x86_warn_reserved(i);
324 op_amd_shutdown(msrs);
325 return -EBUSY;
326 }
327
328 return 0;
329}
330
331static void op_amd_setup_ctrs(struct op_x86_model_spec const *model,
332 struct op_msrs const * const msrs)
333{
334 u64 val;
335 int i;
336
337 /* setup reset_value */
338 for (i = 0; i < NUM_VIRT_COUNTERS; ++i) {
339 if (counter_config[i].enabled
340 && msrs->counters[op_x86_virt_to_phys(i)].addr)
341 reset_value[i] = counter_config[i].count;
342 else
343 reset_value[i] = 0;
344 }
345
346 /* clear all counters */
347 for (i = 0; i < NUM_COUNTERS; ++i) {
348 if (!msrs->controls[i].addr)
349 continue;
350 rdmsrl(msrs->controls[i].addr, val);
351 if (val & ARCH_PERFMON_EVENTSEL_ENABLE)
352 op_x86_warn_in_use(i);
353 val &= model->reserved;
354 wrmsrl(msrs->controls[i].addr, val);
355 /*
356 * avoid a false detection of ctr overflows in NMI
357 * handler
358 */
359 wrmsrl(msrs->counters[i].addr, -1LL);
360 }
361
362 /* enable active counters */
363 for (i = 0; i < NUM_COUNTERS; ++i) {
364 int virt = op_x86_phys_to_virt(i);
365 if (!reset_value[virt])
366 continue;
367
368 /* setup counter registers */
369 wrmsrl(msrs->counters[i].addr, -(u64)reset_value[virt]);
370
371 /* setup control registers */
372 rdmsrl(msrs->controls[i].addr, val);
373 val &= model->reserved;
374 val |= op_x86_get_ctrl(model, &counter_config[virt]);
375 wrmsrl(msrs->controls[i].addr, val);
376 }
377
378 if (ibs_caps)
379 setup_APIC_eilvt_ibs(0, APIC_EILVT_MSG_NMI, 0);
380}
381
382static void op_amd_cpu_shutdown(void)
383{
384 if (ibs_caps)
385 setup_APIC_eilvt_ibs(0, APIC_EILVT_MSG_FIX, 1);
386}
387
368static int op_amd_check_ctrs(struct pt_regs * const regs, 388static int op_amd_check_ctrs(struct pt_regs * const regs,
369 struct op_msrs const * const msrs) 389 struct op_msrs const * const msrs)
370{ 390{
@@ -425,42 +445,16 @@ static void op_amd_stop(struct op_msrs const * const msrs)
425 op_amd_stop_ibs(); 445 op_amd_stop_ibs();
426} 446}
427 447
428static void op_amd_shutdown(struct op_msrs const * const msrs) 448static int __init_ibs_nmi(void)
429{
430 int i;
431
432 for (i = 0; i < NUM_COUNTERS; ++i) {
433 if (msrs->counters[i].addr)
434 release_perfctr_nmi(MSR_K7_PERFCTR0 + i);
435 }
436 for (i = 0; i < NUM_CONTROLS; ++i) {
437 if (msrs->controls[i].addr)
438 release_evntsel_nmi(MSR_K7_EVNTSEL0 + i);
439 }
440}
441
442static u8 ibs_eilvt_off;
443
444static inline void apic_init_ibs_nmi_per_cpu(void *arg)
445{
446 ibs_eilvt_off = setup_APIC_eilvt_ibs(0, APIC_EILVT_MSG_NMI, 0);
447}
448
449static inline void apic_clear_ibs_nmi_per_cpu(void *arg)
450{
451 setup_APIC_eilvt_ibs(0, APIC_EILVT_MSG_FIX, 1);
452}
453
454static int init_ibs_nmi(void)
455{ 449{
456#define IBSCTL_LVTOFFSETVAL (1 << 8) 450#define IBSCTL_LVTOFFSETVAL (1 << 8)
457#define IBSCTL 0x1cc 451#define IBSCTL 0x1cc
458 struct pci_dev *cpu_cfg; 452 struct pci_dev *cpu_cfg;
459 int nodes; 453 int nodes;
460 u32 value = 0; 454 u32 value = 0;
455 u8 ibs_eilvt_off;
461 456
462 /* per CPU setup */ 457 ibs_eilvt_off = setup_APIC_eilvt_ibs(0, APIC_EILVT_MSG_FIX, 1);
463 on_each_cpu(apic_init_ibs_nmi_per_cpu, NULL, 1);
464 458
465 nodes = 0; 459 nodes = 0;
466 cpu_cfg = NULL; 460 cpu_cfg = NULL;
@@ -490,22 +484,15 @@ static int init_ibs_nmi(void)
490 return 0; 484 return 0;
491} 485}
492 486
493/* uninitialize the APIC for the IBS interrupts if needed */
494static void clear_ibs_nmi(void)
495{
496 if (ibs_caps)
497 on_each_cpu(apic_clear_ibs_nmi_per_cpu, NULL, 1);
498}
499
500/* initialize the APIC for the IBS interrupts if available */ 487/* initialize the APIC for the IBS interrupts if available */
501static void ibs_init(void) 488static void init_ibs(void)
502{ 489{
503 ibs_caps = get_ibs_caps(); 490 ibs_caps = get_ibs_caps();
504 491
505 if (!ibs_caps) 492 if (!ibs_caps)
506 return; 493 return;
507 494
508 if (init_ibs_nmi()) { 495 if (__init_ibs_nmi()) {
509 ibs_caps = 0; 496 ibs_caps = 0;
510 return; 497 return;
511 } 498 }
@@ -514,14 +501,6 @@ static void ibs_init(void)
514 (unsigned)ibs_caps); 501 (unsigned)ibs_caps);
515} 502}
516 503
517static void ibs_exit(void)
518{
519 if (!ibs_caps)
520 return;
521
522 clear_ibs_nmi();
523}
524
525static int (*create_arch_files)(struct super_block *sb, struct dentry *root); 504static int (*create_arch_files)(struct super_block *sb, struct dentry *root);
526 505
527static int setup_ibs_files(struct super_block *sb, struct dentry *root) 506static int setup_ibs_files(struct super_block *sb, struct dentry *root)
@@ -570,27 +549,22 @@ static int setup_ibs_files(struct super_block *sb, struct dentry *root)
570 549
571static int op_amd_init(struct oprofile_operations *ops) 550static int op_amd_init(struct oprofile_operations *ops)
572{ 551{
573 ibs_init(); 552 init_ibs();
574 create_arch_files = ops->create_files; 553 create_arch_files = ops->create_files;
575 ops->create_files = setup_ibs_files; 554 ops->create_files = setup_ibs_files;
576 return 0; 555 return 0;
577} 556}
578 557
579static void op_amd_exit(void)
580{
581 ibs_exit();
582}
583
584struct op_x86_model_spec op_amd_spec = { 558struct op_x86_model_spec op_amd_spec = {
585 .num_counters = NUM_COUNTERS, 559 .num_counters = NUM_COUNTERS,
586 .num_controls = NUM_CONTROLS, 560 .num_controls = NUM_COUNTERS,
587 .num_virt_counters = NUM_VIRT_COUNTERS, 561 .num_virt_counters = NUM_VIRT_COUNTERS,
588 .reserved = MSR_AMD_EVENTSEL_RESERVED, 562 .reserved = MSR_AMD_EVENTSEL_RESERVED,
589 .event_mask = OP_EVENT_MASK, 563 .event_mask = OP_EVENT_MASK,
590 .init = op_amd_init, 564 .init = op_amd_init,
591 .exit = op_amd_exit,
592 .fill_in_addresses = &op_amd_fill_in_addresses, 565 .fill_in_addresses = &op_amd_fill_in_addresses,
593 .setup_ctrs = &op_amd_setup_ctrs, 566 .setup_ctrs = &op_amd_setup_ctrs,
567 .cpu_down = &op_amd_cpu_shutdown,
594 .check_ctrs = &op_amd_check_ctrs, 568 .check_ctrs = &op_amd_check_ctrs,
595 .start = &op_amd_start, 569 .start = &op_amd_start,
596 .stop = &op_amd_stop, 570 .stop = &op_amd_stop,
diff --git a/arch/x86/oprofile/op_model_p4.c b/arch/x86/oprofile/op_model_p4.c
index e6a160a4684a..182558dd5515 100644
--- a/arch/x86/oprofile/op_model_p4.c
+++ b/arch/x86/oprofile/op_model_p4.c
@@ -385,8 +385,26 @@ static unsigned int get_stagger(void)
385 385
386static unsigned long reset_value[NUM_COUNTERS_NON_HT]; 386static unsigned long reset_value[NUM_COUNTERS_NON_HT];
387 387
388static void p4_shutdown(struct op_msrs const * const msrs)
389{
390 int i;
388 391
389static void p4_fill_in_addresses(struct op_msrs * const msrs) 392 for (i = 0; i < num_counters; ++i) {
393 if (msrs->counters[i].addr)
394 release_perfctr_nmi(msrs->counters[i].addr);
395 }
396 /*
397 * some of the control registers are specially reserved in
398 * conjunction with the counter registers (hence the starting offset).
399 * This saves a few bits.
400 */
401 for (i = num_counters; i < num_controls; ++i) {
402 if (msrs->controls[i].addr)
403 release_evntsel_nmi(msrs->controls[i].addr);
404 }
405}
406
407static int p4_fill_in_addresses(struct op_msrs * const msrs)
390{ 408{
391 unsigned int i; 409 unsigned int i;
392 unsigned int addr, cccraddr, stag; 410 unsigned int addr, cccraddr, stag;
@@ -468,6 +486,18 @@ static void p4_fill_in_addresses(struct op_msrs * const msrs)
468 msrs->controls[i++].addr = MSR_P4_CRU_ESCR5; 486 msrs->controls[i++].addr = MSR_P4_CRU_ESCR5;
469 } 487 }
470 } 488 }
489
490 for (i = 0; i < num_counters; ++i) {
491 if (!counter_config[i].enabled)
492 continue;
493 if (msrs->controls[i].addr)
494 continue;
495 op_x86_warn_reserved(i);
496 p4_shutdown(msrs);
497 return -EBUSY;
498 }
499
500 return 0;
471} 501}
472 502
473 503
@@ -668,26 +698,6 @@ static void p4_stop(struct op_msrs const * const msrs)
668 } 698 }
669} 699}
670 700
671static void p4_shutdown(struct op_msrs const * const msrs)
672{
673 int i;
674
675 for (i = 0; i < num_counters; ++i) {
676 if (msrs->counters[i].addr)
677 release_perfctr_nmi(msrs->counters[i].addr);
678 }
679 /*
680 * some of the control registers are specially reserved in
681 * conjunction with the counter registers (hence the starting offset).
682 * This saves a few bits.
683 */
684 for (i = num_counters; i < num_controls; ++i) {
685 if (msrs->controls[i].addr)
686 release_evntsel_nmi(msrs->controls[i].addr);
687 }
688}
689
690
691#ifdef CONFIG_SMP 701#ifdef CONFIG_SMP
692struct op_x86_model_spec op_p4_ht2_spec = { 702struct op_x86_model_spec op_p4_ht2_spec = {
693 .num_counters = NUM_COUNTERS_HT2, 703 .num_counters = NUM_COUNTERS_HT2,
diff --git a/arch/x86/oprofile/op_model_ppro.c b/arch/x86/oprofile/op_model_ppro.c
index 2bf90fafa7b5..1fd17cfb956b 100644
--- a/arch/x86/oprofile/op_model_ppro.c
+++ b/arch/x86/oprofile/op_model_ppro.c
@@ -30,19 +30,46 @@ static int counter_width = 32;
30 30
31static u64 *reset_value; 31static u64 *reset_value;
32 32
33static void ppro_fill_in_addresses(struct op_msrs * const msrs) 33static void ppro_shutdown(struct op_msrs const * const msrs)
34{ 34{
35 int i; 35 int i;
36 36
37 for (i = 0; i < num_counters; i++) { 37 for (i = 0; i < num_counters; ++i) {
38 if (reserve_perfctr_nmi(MSR_P6_PERFCTR0 + i)) 38 if (!msrs->counters[i].addr)
39 msrs->counters[i].addr = MSR_P6_PERFCTR0 + i; 39 continue;
40 release_perfctr_nmi(MSR_P6_PERFCTR0 + i);
41 release_evntsel_nmi(MSR_P6_EVNTSEL0 + i);
42 }
43 if (reset_value) {
44 kfree(reset_value);
45 reset_value = NULL;
40 } 46 }
47}
48
49static int ppro_fill_in_addresses(struct op_msrs * const msrs)
50{
51 int i;
41 52
42 for (i = 0; i < num_counters; i++) { 53 for (i = 0; i < num_counters; i++) {
43 if (reserve_evntsel_nmi(MSR_P6_EVNTSEL0 + i)) 54 if (!reserve_perfctr_nmi(MSR_P6_PERFCTR0 + i))
44 msrs->controls[i].addr = MSR_P6_EVNTSEL0 + i; 55 goto fail;
56 if (!reserve_evntsel_nmi(MSR_P6_EVNTSEL0 + i)) {
57 release_perfctr_nmi(MSR_P6_PERFCTR0 + i);
58 goto fail;
59 }
60 /* both registers must be reserved */
61 msrs->counters[i].addr = MSR_P6_PERFCTR0 + i;
62 msrs->controls[i].addr = MSR_P6_EVNTSEL0 + i;
63 continue;
64 fail:
65 if (!counter_config[i].enabled)
66 continue;
67 op_x86_warn_reserved(i);
68 ppro_shutdown(msrs);
69 return -EBUSY;
45 } 70 }
71
72 return 0;
46} 73}
47 74
48 75
@@ -78,26 +105,17 @@ static void ppro_setup_ctrs(struct op_x86_model_spec const *model,
78 105
79 /* clear all counters */ 106 /* clear all counters */
80 for (i = 0; i < num_counters; ++i) { 107 for (i = 0; i < num_counters; ++i) {
81 if (unlikely(!msrs->controls[i].addr)) { 108 if (!msrs->controls[i].addr)
82 if (counter_config[i].enabled && !smp_processor_id())
83 /*
84 * counter is reserved, this is on all
85 * cpus, so report only for cpu #0
86 */
87 op_x86_warn_reserved(i);
88 continue; 109 continue;
89 }
90 rdmsrl(msrs->controls[i].addr, val); 110 rdmsrl(msrs->controls[i].addr, val);
91 if (val & ARCH_PERFMON_EVENTSEL_ENABLE) 111 if (val & ARCH_PERFMON_EVENTSEL_ENABLE)
92 op_x86_warn_in_use(i); 112 op_x86_warn_in_use(i);
93 val &= model->reserved; 113 val &= model->reserved;
94 wrmsrl(msrs->controls[i].addr, val); 114 wrmsrl(msrs->controls[i].addr, val);
95 } 115 /*
96 116 * avoid a false detection of ctr overflows in NMI *
97 /* avoid a false detection of ctr overflows in NMI handler */ 117 * handler
98 for (i = 0; i < num_counters; ++i) { 118 */
99 if (unlikely(!msrs->counters[i].addr))
100 continue;
101 wrmsrl(msrs->counters[i].addr, -1LL); 119 wrmsrl(msrs->counters[i].addr, -1LL);
102 } 120 }
103 121
@@ -189,25 +207,6 @@ static void ppro_stop(struct op_msrs const * const msrs)
189 } 207 }
190} 208}
191 209
192static void ppro_shutdown(struct op_msrs const * const msrs)
193{
194 int i;
195
196 for (i = 0; i < num_counters; ++i) {
197 if (msrs->counters[i].addr)
198 release_perfctr_nmi(MSR_P6_PERFCTR0 + i);
199 }
200 for (i = 0; i < num_counters; ++i) {
201 if (msrs->controls[i].addr)
202 release_evntsel_nmi(MSR_P6_EVNTSEL0 + i);
203 }
204 if (reset_value) {
205 kfree(reset_value);
206 reset_value = NULL;
207 }
208}
209
210
211struct op_x86_model_spec op_ppro_spec = { 210struct op_x86_model_spec op_ppro_spec = {
212 .num_counters = 2, 211 .num_counters = 2,
213 .num_controls = 2, 212 .num_controls = 2,
diff --git a/arch/x86/oprofile/op_x86_model.h b/arch/x86/oprofile/op_x86_model.h
index ff82a755edd4..89017fa1fd63 100644
--- a/arch/x86/oprofile/op_x86_model.h
+++ b/arch/x86/oprofile/op_x86_model.h
@@ -40,10 +40,10 @@ struct op_x86_model_spec {
40 u64 reserved; 40 u64 reserved;
41 u16 event_mask; 41 u16 event_mask;
42 int (*init)(struct oprofile_operations *ops); 42 int (*init)(struct oprofile_operations *ops);
43 void (*exit)(void); 43 int (*fill_in_addresses)(struct op_msrs * const msrs);
44 void (*fill_in_addresses)(struct op_msrs * const msrs);
45 void (*setup_ctrs)(struct op_x86_model_spec const *model, 44 void (*setup_ctrs)(struct op_x86_model_spec const *model,
46 struct op_msrs const * const msrs); 45 struct op_msrs const * const msrs);
46 void (*cpu_down)(void);
47 int (*check_ctrs)(struct pt_regs * const regs, 47 int (*check_ctrs)(struct pt_regs * const regs,
48 struct op_msrs const * const msrs); 48 struct op_msrs const * const msrs);
49 void (*start)(struct op_msrs const * const msrs); 49 void (*start)(struct op_msrs const * const msrs);