diff options
Diffstat (limited to 'kernel/perf_event.c')
-rw-r--r-- | kernel/perf_event.c | 105 |
1 files changed, 55 insertions, 50 deletions
diff --git a/kernel/perf_event.c b/kernel/perf_event.c index 4393b9e73740..3d1552d3c12b 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c | |||
@@ -15,6 +15,7 @@ | |||
15 | #include <linux/smp.h> | 15 | #include <linux/smp.h> |
16 | #include <linux/file.h> | 16 | #include <linux/file.h> |
17 | #include <linux/poll.h> | 17 | #include <linux/poll.h> |
18 | #include <linux/slab.h> | ||
18 | #include <linux/sysfs.h> | 19 | #include <linux/sysfs.h> |
19 | #include <linux/dcache.h> | 20 | #include <linux/dcache.h> |
20 | #include <linux/percpu.h> | 21 | #include <linux/percpu.h> |
@@ -81,10 +82,6 @@ extern __weak const struct pmu *hw_perf_event_init(struct perf_event *event) | |||
81 | void __weak hw_perf_disable(void) { barrier(); } | 82 | void __weak hw_perf_disable(void) { barrier(); } |
82 | void __weak hw_perf_enable(void) { barrier(); } | 83 | void __weak hw_perf_enable(void) { barrier(); } |
83 | 84 | ||
84 | void __weak hw_perf_event_setup(int cpu) { barrier(); } | ||
85 | void __weak hw_perf_event_setup_online(int cpu) { barrier(); } | ||
86 | void __weak hw_perf_event_setup_offline(int cpu) { barrier(); } | ||
87 | |||
88 | int __weak | 85 | int __weak |
89 | hw_perf_group_sched_in(struct perf_event *group_leader, | 86 | hw_perf_group_sched_in(struct perf_event *group_leader, |
90 | struct perf_cpu_context *cpuctx, | 87 | struct perf_cpu_context *cpuctx, |
@@ -97,25 +94,15 @@ void __weak perf_event_print_debug(void) { } | |||
97 | 94 | ||
98 | static DEFINE_PER_CPU(int, perf_disable_count); | 95 | static DEFINE_PER_CPU(int, perf_disable_count); |
99 | 96 | ||
100 | void __perf_disable(void) | ||
101 | { | ||
102 | __get_cpu_var(perf_disable_count)++; | ||
103 | } | ||
104 | |||
105 | bool __perf_enable(void) | ||
106 | { | ||
107 | return !--__get_cpu_var(perf_disable_count); | ||
108 | } | ||
109 | |||
110 | void perf_disable(void) | 97 | void perf_disable(void) |
111 | { | 98 | { |
112 | __perf_disable(); | 99 | if (!__get_cpu_var(perf_disable_count)++) |
113 | hw_perf_disable(); | 100 | hw_perf_disable(); |
114 | } | 101 | } |
115 | 102 | ||
116 | void perf_enable(void) | 103 | void perf_enable(void) |
117 | { | 104 | { |
118 | if (__perf_enable()) | 105 | if (!--__get_cpu_var(perf_disable_count)) |
119 | hw_perf_enable(); | 106 | hw_perf_enable(); |
120 | } | 107 | } |
121 | 108 | ||
@@ -1178,11 +1165,9 @@ void perf_event_task_sched_out(struct task_struct *task, | |||
1178 | struct perf_event_context *ctx = task->perf_event_ctxp; | 1165 | struct perf_event_context *ctx = task->perf_event_ctxp; |
1179 | struct perf_event_context *next_ctx; | 1166 | struct perf_event_context *next_ctx; |
1180 | struct perf_event_context *parent; | 1167 | struct perf_event_context *parent; |
1181 | struct pt_regs *regs; | ||
1182 | int do_switch = 1; | 1168 | int do_switch = 1; |
1183 | 1169 | ||
1184 | regs = task_pt_regs(task); | 1170 | perf_sw_event(PERF_COUNT_SW_CONTEXT_SWITCHES, 1, 1, NULL, 0); |
1185 | perf_sw_event(PERF_COUNT_SW_CONTEXT_SWITCHES, 1, 1, regs, 0); | ||
1186 | 1171 | ||
1187 | if (likely(!ctx || !cpuctx->task_ctx)) | 1172 | if (likely(!ctx || !cpuctx->task_ctx)) |
1188 | return; | 1173 | return; |
@@ -1538,12 +1523,15 @@ static void perf_ctx_adjust_freq(struct perf_event_context *ctx) | |||
1538 | */ | 1523 | */ |
1539 | if (interrupts == MAX_INTERRUPTS) { | 1524 | if (interrupts == MAX_INTERRUPTS) { |
1540 | perf_log_throttle(event, 1); | 1525 | perf_log_throttle(event, 1); |
1526 | perf_disable(); | ||
1541 | event->pmu->unthrottle(event); | 1527 | event->pmu->unthrottle(event); |
1528 | perf_enable(); | ||
1542 | } | 1529 | } |
1543 | 1530 | ||
1544 | if (!event->attr.freq || !event->attr.sample_freq) | 1531 | if (!event->attr.freq || !event->attr.sample_freq) |
1545 | continue; | 1532 | continue; |
1546 | 1533 | ||
1534 | perf_disable(); | ||
1547 | event->pmu->read(event); | 1535 | event->pmu->read(event); |
1548 | now = atomic64_read(&event->count); | 1536 | now = atomic64_read(&event->count); |
1549 | delta = now - hwc->freq_count_stamp; | 1537 | delta = now - hwc->freq_count_stamp; |
@@ -1551,6 +1539,7 @@ static void perf_ctx_adjust_freq(struct perf_event_context *ctx) | |||
1551 | 1539 | ||
1552 | if (delta > 0) | 1540 | if (delta > 0) |
1553 | perf_adjust_period(event, TICK_NSEC, delta); | 1541 | perf_adjust_period(event, TICK_NSEC, delta); |
1542 | perf_enable(); | ||
1554 | } | 1543 | } |
1555 | raw_spin_unlock(&ctx->lock); | 1544 | raw_spin_unlock(&ctx->lock); |
1556 | } | 1545 | } |
@@ -1560,9 +1549,6 @@ static void perf_ctx_adjust_freq(struct perf_event_context *ctx) | |||
1560 | */ | 1549 | */ |
1561 | static void rotate_ctx(struct perf_event_context *ctx) | 1550 | static void rotate_ctx(struct perf_event_context *ctx) |
1562 | { | 1551 | { |
1563 | if (!ctx->nr_events) | ||
1564 | return; | ||
1565 | |||
1566 | raw_spin_lock(&ctx->lock); | 1552 | raw_spin_lock(&ctx->lock); |
1567 | 1553 | ||
1568 | /* Rotate the first entry last of non-pinned groups */ | 1554 | /* Rotate the first entry last of non-pinned groups */ |
@@ -1575,19 +1561,28 @@ void perf_event_task_tick(struct task_struct *curr) | |||
1575 | { | 1561 | { |
1576 | struct perf_cpu_context *cpuctx; | 1562 | struct perf_cpu_context *cpuctx; |
1577 | struct perf_event_context *ctx; | 1563 | struct perf_event_context *ctx; |
1564 | int rotate = 0; | ||
1578 | 1565 | ||
1579 | if (!atomic_read(&nr_events)) | 1566 | if (!atomic_read(&nr_events)) |
1580 | return; | 1567 | return; |
1581 | 1568 | ||
1582 | cpuctx = &__get_cpu_var(perf_cpu_context); | 1569 | cpuctx = &__get_cpu_var(perf_cpu_context); |
1583 | ctx = curr->perf_event_ctxp; | 1570 | if (cpuctx->ctx.nr_events && |
1571 | cpuctx->ctx.nr_events != cpuctx->ctx.nr_active) | ||
1572 | rotate = 1; | ||
1584 | 1573 | ||
1585 | perf_disable(); | 1574 | ctx = curr->perf_event_ctxp; |
1575 | if (ctx && ctx->nr_events && ctx->nr_events != ctx->nr_active) | ||
1576 | rotate = 1; | ||
1586 | 1577 | ||
1587 | perf_ctx_adjust_freq(&cpuctx->ctx); | 1578 | perf_ctx_adjust_freq(&cpuctx->ctx); |
1588 | if (ctx) | 1579 | if (ctx) |
1589 | perf_ctx_adjust_freq(ctx); | 1580 | perf_ctx_adjust_freq(ctx); |
1590 | 1581 | ||
1582 | if (!rotate) | ||
1583 | return; | ||
1584 | |||
1585 | perf_disable(); | ||
1591 | cpu_ctx_sched_out(cpuctx, EVENT_FLEXIBLE); | 1586 | cpu_ctx_sched_out(cpuctx, EVENT_FLEXIBLE); |
1592 | if (ctx) | 1587 | if (ctx) |
1593 | task_ctx_sched_out(ctx, EVENT_FLEXIBLE); | 1588 | task_ctx_sched_out(ctx, EVENT_FLEXIBLE); |
@@ -1599,7 +1594,6 @@ void perf_event_task_tick(struct task_struct *curr) | |||
1599 | cpu_ctx_sched_in(cpuctx, EVENT_FLEXIBLE); | 1594 | cpu_ctx_sched_in(cpuctx, EVENT_FLEXIBLE); |
1600 | if (ctx) | 1595 | if (ctx) |
1601 | task_ctx_sched_in(curr, EVENT_FLEXIBLE); | 1596 | task_ctx_sched_in(curr, EVENT_FLEXIBLE); |
1602 | |||
1603 | perf_enable(); | 1597 | perf_enable(); |
1604 | } | 1598 | } |
1605 | 1599 | ||
@@ -2791,6 +2785,12 @@ __weak struct perf_callchain_entry *perf_callchain(struct pt_regs *regs) | |||
2791 | return NULL; | 2785 | return NULL; |
2792 | } | 2786 | } |
2793 | 2787 | ||
2788 | __weak | ||
2789 | void perf_arch_fetch_caller_regs(struct pt_regs *regs, unsigned long ip, int skip) | ||
2790 | { | ||
2791 | } | ||
2792 | |||
2793 | |||
2794 | /* | 2794 | /* |
2795 | * Output | 2795 | * Output |
2796 | */ | 2796 | */ |
@@ -3376,15 +3376,23 @@ static void perf_event_task_output(struct perf_event *event, | |||
3376 | struct perf_task_event *task_event) | 3376 | struct perf_task_event *task_event) |
3377 | { | 3377 | { |
3378 | struct perf_output_handle handle; | 3378 | struct perf_output_handle handle; |
3379 | int size; | ||
3380 | struct task_struct *task = task_event->task; | 3379 | struct task_struct *task = task_event->task; |
3381 | int ret; | 3380 | unsigned long flags; |
3381 | int size, ret; | ||
3382 | |||
3383 | /* | ||
3384 | * If this CPU attempts to acquire an rq lock held by a CPU spinning | ||
3385 | * in perf_output_lock() from interrupt context, it's game over. | ||
3386 | */ | ||
3387 | local_irq_save(flags); | ||
3382 | 3388 | ||
3383 | size = task_event->event_id.header.size; | 3389 | size = task_event->event_id.header.size; |
3384 | ret = perf_output_begin(&handle, event, size, 0, 0); | 3390 | ret = perf_output_begin(&handle, event, size, 0, 0); |
3385 | 3391 | ||
3386 | if (ret) | 3392 | if (ret) { |
3393 | local_irq_restore(flags); | ||
3387 | return; | 3394 | return; |
3395 | } | ||
3388 | 3396 | ||
3389 | task_event->event_id.pid = perf_event_pid(event, task); | 3397 | task_event->event_id.pid = perf_event_pid(event, task); |
3390 | task_event->event_id.ppid = perf_event_pid(event, current); | 3398 | task_event->event_id.ppid = perf_event_pid(event, current); |
@@ -3395,6 +3403,7 @@ static void perf_event_task_output(struct perf_event *event, | |||
3395 | perf_output_put(&handle, task_event->event_id); | 3403 | perf_output_put(&handle, task_event->event_id); |
3396 | 3404 | ||
3397 | perf_output_end(&handle); | 3405 | perf_output_end(&handle); |
3406 | local_irq_restore(flags); | ||
3398 | } | 3407 | } |
3399 | 3408 | ||
3400 | static int perf_event_task_match(struct perf_event *event) | 3409 | static int perf_event_task_match(struct perf_event *event) |
@@ -4318,9 +4327,8 @@ static const struct pmu perf_ops_task_clock = { | |||
4318 | #ifdef CONFIG_EVENT_TRACING | 4327 | #ifdef CONFIG_EVENT_TRACING |
4319 | 4328 | ||
4320 | void perf_tp_event(int event_id, u64 addr, u64 count, void *record, | 4329 | void perf_tp_event(int event_id, u64 addr, u64 count, void *record, |
4321 | int entry_size) | 4330 | int entry_size, struct pt_regs *regs) |
4322 | { | 4331 | { |
4323 | struct pt_regs *regs = get_irq_regs(); | ||
4324 | struct perf_sample_data data; | 4332 | struct perf_sample_data data; |
4325 | struct perf_raw_record raw = { | 4333 | struct perf_raw_record raw = { |
4326 | .size = entry_size, | 4334 | .size = entry_size, |
@@ -4330,12 +4338,9 @@ void perf_tp_event(int event_id, u64 addr, u64 count, void *record, | |||
4330 | perf_sample_data_init(&data, addr); | 4338 | perf_sample_data_init(&data, addr); |
4331 | data.raw = &raw; | 4339 | data.raw = &raw; |
4332 | 4340 | ||
4333 | if (!regs) | ||
4334 | regs = task_pt_regs(current); | ||
4335 | |||
4336 | /* Trace events already protected against recursion */ | 4341 | /* Trace events already protected against recursion */ |
4337 | do_perf_sw_event(PERF_TYPE_TRACEPOINT, event_id, count, 1, | 4342 | do_perf_sw_event(PERF_TYPE_TRACEPOINT, event_id, count, 1, |
4338 | &data, regs); | 4343 | &data, regs); |
4339 | } | 4344 | } |
4340 | EXPORT_SYMBOL_GPL(perf_tp_event); | 4345 | EXPORT_SYMBOL_GPL(perf_tp_event); |
4341 | 4346 | ||
@@ -4351,7 +4356,7 @@ static int perf_tp_event_match(struct perf_event *event, | |||
4351 | 4356 | ||
4352 | static void tp_perf_event_destroy(struct perf_event *event) | 4357 | static void tp_perf_event_destroy(struct perf_event *event) |
4353 | { | 4358 | { |
4354 | ftrace_profile_disable(event->attr.config); | 4359 | perf_trace_disable(event->attr.config); |
4355 | } | 4360 | } |
4356 | 4361 | ||
4357 | static const struct pmu *tp_perf_event_init(struct perf_event *event) | 4362 | static const struct pmu *tp_perf_event_init(struct perf_event *event) |
@@ -4365,7 +4370,7 @@ static const struct pmu *tp_perf_event_init(struct perf_event *event) | |||
4365 | !capable(CAP_SYS_ADMIN)) | 4370 | !capable(CAP_SYS_ADMIN)) |
4366 | return ERR_PTR(-EPERM); | 4371 | return ERR_PTR(-EPERM); |
4367 | 4372 | ||
4368 | if (ftrace_profile_enable(event->attr.config)) | 4373 | if (perf_trace_enable(event->attr.config)) |
4369 | return NULL; | 4374 | return NULL; |
4370 | 4375 | ||
4371 | event->destroy = tp_perf_event_destroy; | 4376 | event->destroy = tp_perf_event_destroy; |
@@ -4892,7 +4897,7 @@ err_fput_free_put_context: | |||
4892 | 4897 | ||
4893 | err_free_put_context: | 4898 | err_free_put_context: |
4894 | if (err < 0) | 4899 | if (err < 0) |
4895 | kfree(event); | 4900 | free_event(event); |
4896 | 4901 | ||
4897 | err_put_context: | 4902 | err_put_context: |
4898 | if (err < 0) | 4903 | if (err < 0) |
@@ -5372,18 +5377,26 @@ int perf_event_init_task(struct task_struct *child) | |||
5372 | return ret; | 5377 | return ret; |
5373 | } | 5378 | } |
5374 | 5379 | ||
5380 | static void __init perf_event_init_all_cpus(void) | ||
5381 | { | ||
5382 | int cpu; | ||
5383 | struct perf_cpu_context *cpuctx; | ||
5384 | |||
5385 | for_each_possible_cpu(cpu) { | ||
5386 | cpuctx = &per_cpu(perf_cpu_context, cpu); | ||
5387 | __perf_event_init_context(&cpuctx->ctx, NULL); | ||
5388 | } | ||
5389 | } | ||
5390 | |||
5375 | static void __cpuinit perf_event_init_cpu(int cpu) | 5391 | static void __cpuinit perf_event_init_cpu(int cpu) |
5376 | { | 5392 | { |
5377 | struct perf_cpu_context *cpuctx; | 5393 | struct perf_cpu_context *cpuctx; |
5378 | 5394 | ||
5379 | cpuctx = &per_cpu(perf_cpu_context, cpu); | 5395 | cpuctx = &per_cpu(perf_cpu_context, cpu); |
5380 | __perf_event_init_context(&cpuctx->ctx, NULL); | ||
5381 | 5396 | ||
5382 | spin_lock(&perf_resource_lock); | 5397 | spin_lock(&perf_resource_lock); |
5383 | cpuctx->max_pertask = perf_max_events - perf_reserved_percpu; | 5398 | cpuctx->max_pertask = perf_max_events - perf_reserved_percpu; |
5384 | spin_unlock(&perf_resource_lock); | 5399 | spin_unlock(&perf_resource_lock); |
5385 | |||
5386 | hw_perf_event_setup(cpu); | ||
5387 | } | 5400 | } |
5388 | 5401 | ||
5389 | #ifdef CONFIG_HOTPLUG_CPU | 5402 | #ifdef CONFIG_HOTPLUG_CPU |
@@ -5423,20 +5436,11 @@ perf_cpu_notify(struct notifier_block *self, unsigned long action, void *hcpu) | |||
5423 | perf_event_init_cpu(cpu); | 5436 | perf_event_init_cpu(cpu); |
5424 | break; | 5437 | break; |
5425 | 5438 | ||
5426 | case CPU_ONLINE: | ||
5427 | case CPU_ONLINE_FROZEN: | ||
5428 | hw_perf_event_setup_online(cpu); | ||
5429 | break; | ||
5430 | |||
5431 | case CPU_DOWN_PREPARE: | 5439 | case CPU_DOWN_PREPARE: |
5432 | case CPU_DOWN_PREPARE_FROZEN: | 5440 | case CPU_DOWN_PREPARE_FROZEN: |
5433 | perf_event_exit_cpu(cpu); | 5441 | perf_event_exit_cpu(cpu); |
5434 | break; | 5442 | break; |
5435 | 5443 | ||
5436 | case CPU_DEAD: | ||
5437 | hw_perf_event_setup_offline(cpu); | ||
5438 | break; | ||
5439 | |||
5440 | default: | 5444 | default: |
5441 | break; | 5445 | break; |
5442 | } | 5446 | } |
@@ -5454,6 +5458,7 @@ static struct notifier_block __cpuinitdata perf_cpu_nb = { | |||
5454 | 5458 | ||
5455 | void __init perf_event_init(void) | 5459 | void __init perf_event_init(void) |
5456 | { | 5460 | { |
5461 | perf_event_init_all_cpus(); | ||
5457 | perf_cpu_notify(&perf_cpu_nb, (unsigned long)CPU_UP_PREPARE, | 5462 | perf_cpu_notify(&perf_cpu_nb, (unsigned long)CPU_UP_PREPARE, |
5458 | (void *)(long)smp_processor_id()); | 5463 | (void *)(long)smp_processor_id()); |
5459 | perf_cpu_notify(&perf_cpu_nb, (unsigned long)CPU_ONLINE, | 5464 | perf_cpu_notify(&perf_cpu_nb, (unsigned long)CPU_ONLINE, |