core.c 210.4 KB
Newer Older
T
Thomas Gleixner 已提交
1
/*
I
Ingo Molnar 已提交
2
 * Performance events core code:
T
Thomas Gleixner 已提交
3
 *
4
 *  Copyright (C) 2008 Thomas Gleixner <tglx@linutronix.de>
5 6
 *  Copyright (C) 2008-2011 Red Hat, Inc., Ingo Molnar
 *  Copyright (C) 2008-2011 Red Hat, Inc., Peter Zijlstra <pzijlstr@redhat.com>
A
Al Viro 已提交
7
 *  Copyright  ©  2009 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
8
 *
I
Ingo Molnar 已提交
9
 * For licensing details see kernel-base/COPYING
T
Thomas Gleixner 已提交
10 11 12
 */

#include <linux/fs.h>
13
#include <linux/mm.h>
T
Thomas Gleixner 已提交
14 15
#include <linux/cpu.h>
#include <linux/smp.h>
P
Peter Zijlstra 已提交
16
#include <linux/idr.h>
17
#include <linux/file.h>
T
Thomas Gleixner 已提交
18
#include <linux/poll.h>
19
#include <linux/slab.h>
20
#include <linux/hash.h>
21
#include <linux/tick.h>
T
Thomas Gleixner 已提交
22
#include <linux/sysfs.h>
23
#include <linux/dcache.h>
T
Thomas Gleixner 已提交
24
#include <linux/percpu.h>
25
#include <linux/ptrace.h>
P
Peter Zijlstra 已提交
26
#include <linux/reboot.h>
27
#include <linux/vmstat.h>
P
Peter Zijlstra 已提交
28
#include <linux/device.h>
29
#include <linux/export.h>
30
#include <linux/vmalloc.h>
31 32
#include <linux/hardirq.h>
#include <linux/rculist.h>
T
Thomas Gleixner 已提交
33 34 35
#include <linux/uaccess.h>
#include <linux/syscalls.h>
#include <linux/anon_inodes.h>
I
Ingo Molnar 已提交
36
#include <linux/kernel_stat.h>
37
#include <linux/cgroup.h>
38
#include <linux/perf_event.h>
L
Li Zefan 已提交
39
#include <linux/ftrace_event.h>
40
#include <linux/hw_breakpoint.h>
41
#include <linux/mm_types.h>
42
#include <linux/module.h>
43
#include <linux/mman.h>
P
Pawel Moll 已提交
44
#include <linux/compat.h>
45 46
#include <linux/bpf.h>
#include <linux/filter.h>
T
Thomas Gleixner 已提交
47

48 49
#include "internal.h"

50 51
#include <asm/irq_regs.h>

52 53
static struct workqueue_struct *perf_wq;

54
struct remote_function_call {
55 56 57 58
	struct task_struct	*p;
	int			(*func)(void *info);
	void			*info;
	int			ret;
59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91
};

static void remote_function(void *data)
{
	struct remote_function_call *tfc = data;
	struct task_struct *p = tfc->p;

	if (p) {
		tfc->ret = -EAGAIN;
		if (task_cpu(p) != smp_processor_id() || !task_curr(p))
			return;
	}

	tfc->ret = tfc->func(tfc->info);
}

/**
 * task_function_call - call a function on the cpu on which a task runs
 * @p:		the task to evaluate
 * @func:	the function to be called
 * @info:	the function call argument
 *
 * Calls the function @func when the task is currently running. This might
 * be on the current CPU, which just calls the function directly
 *
 * returns: @func return value, or
 *	    -ESRCH  - when the process isn't running
 *	    -EAGAIN - when the process moved away
 */
static int
task_function_call(struct task_struct *p, int (*func) (void *info), void *info)
{
	struct remote_function_call data = {
92 93 94 95
		.p	= p,
		.func	= func,
		.info	= info,
		.ret	= -ESRCH, /* No such (running) process */
96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115
	};

	if (task_curr(p))
		smp_call_function_single(task_cpu(p), remote_function, &data, 1);

	return data.ret;
}

/**
 * cpu_function_call - call a function on the cpu
 * @func:	the function to be called
 * @info:	the function call argument
 *
 * Calls the function @func on the remote cpu.
 *
 * returns: @func return value or -ENXIO when the cpu is offline
 */
static int cpu_function_call(int cpu, int (*func) (void *info), void *info)
{
	struct remote_function_call data = {
116 117 118 119
		.p	= NULL,
		.func	= func,
		.info	= info,
		.ret	= -ENXIO, /* No such CPU */
120 121 122 123 124 125 126
	};

	smp_call_function_single(cpu, remote_function, &data, 1);

	return data.ret;
}

127 128 129 130 131 132 133
#define EVENT_OWNER_KERNEL ((void *) -1)

static bool is_kernel_event(struct perf_event *event)
{
	return event->owner == EVENT_OWNER_KERNEL;
}

S
Stephane Eranian 已提交
134 135
#define PERF_FLAG_ALL (PERF_FLAG_FD_NO_GROUP |\
		       PERF_FLAG_FD_OUTPUT  |\
136 137
		       PERF_FLAG_PID_CGROUP |\
		       PERF_FLAG_FD_CLOEXEC)
S
Stephane Eranian 已提交
138

139 140 141 142 143 144 145
/*
 * branch priv levels that need permission checks
 */
#define PERF_SAMPLE_BRANCH_PERM_PLM \
	(PERF_SAMPLE_BRANCH_KERNEL |\
	 PERF_SAMPLE_BRANCH_HV)

146 147 148 149 150 151
enum event_type_t {
	EVENT_FLEXIBLE = 0x1,
	EVENT_PINNED = 0x2,
	EVENT_ALL = EVENT_FLEXIBLE | EVENT_PINNED,
};

S
Stephane Eranian 已提交
152 153 154 155
/*
 * perf_sched_events : >0 events exist
 * perf_cgroup_events: >0 per-cpu cgroup events exist on this cpu
 */
156
struct static_key_deferred perf_sched_events __read_mostly;
S
Stephane Eranian 已提交
157
static DEFINE_PER_CPU(atomic_t, perf_cgroup_events);
158
static DEFINE_PER_CPU(int, perf_sched_cb_usages);
S
Stephane Eranian 已提交
159

160 161 162
static atomic_t nr_mmap_events __read_mostly;
static atomic_t nr_comm_events __read_mostly;
static atomic_t nr_task_events __read_mostly;
163
static atomic_t nr_freq_events __read_mostly;
164

P
Peter Zijlstra 已提交
165 166 167 168
static LIST_HEAD(pmus);
static DEFINE_MUTEX(pmus_lock);
static struct srcu_struct pmus_srcu;

169
/*
170
 * perf event paranoia level:
171 172
 *  -1 - not paranoid at all
 *   0 - disallow raw tracepoint access for unpriv
173
 *   1 - disallow cpu events for unpriv
174
 *   2 - disallow kernel profiling for unpriv
175
 */
176
int sysctl_perf_event_paranoid __read_mostly = 1;
177

178 179
/* Minimum for 512 kiB + 1 user control page */
int sysctl_perf_event_mlock __read_mostly = 512 + (PAGE_SIZE / 1024); /* 'free' kiB per user */
180 181

/*
182
 * max perf event sample rate
183
 */
184 185 186 187 188 189 190 191 192
#define DEFAULT_MAX_SAMPLE_RATE		100000
#define DEFAULT_SAMPLE_PERIOD_NS	(NSEC_PER_SEC / DEFAULT_MAX_SAMPLE_RATE)
#define DEFAULT_CPU_TIME_MAX_PERCENT	25

int sysctl_perf_event_sample_rate __read_mostly	= DEFAULT_MAX_SAMPLE_RATE;

static int max_samples_per_tick __read_mostly	= DIV_ROUND_UP(DEFAULT_MAX_SAMPLE_RATE, HZ);
static int perf_sample_period_ns __read_mostly	= DEFAULT_SAMPLE_PERIOD_NS;

P
Peter Zijlstra 已提交
193 194
static int perf_sample_allowed_ns __read_mostly =
	DEFAULT_SAMPLE_PERIOD_NS * DEFAULT_CPU_TIME_MAX_PERCENT / 100;
195 196 197 198 199 200

void update_perf_cpu_limits(void)
{
	u64 tmp = perf_sample_period_ns;

	tmp *= sysctl_perf_cpu_time_max_percent;
201
	do_div(tmp, 100);
P
Peter Zijlstra 已提交
202
	ACCESS_ONCE(perf_sample_allowed_ns) = tmp;
203
}
P
Peter Zijlstra 已提交
204

205 206
static int perf_rotate_context(struct perf_cpu_context *cpuctx);

P
Peter Zijlstra 已提交
207 208 209 210
int perf_proc_update_handler(struct ctl_table *table, int write,
		void __user *buffer, size_t *lenp,
		loff_t *ppos)
{
211
	int ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
P
Peter Zijlstra 已提交
212 213 214 215 216

	if (ret || !write)
		return ret;

	max_samples_per_tick = DIV_ROUND_UP(sysctl_perf_event_sample_rate, HZ);
217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234
	perf_sample_period_ns = NSEC_PER_SEC / sysctl_perf_event_sample_rate;
	update_perf_cpu_limits();

	return 0;
}

int sysctl_perf_cpu_time_max_percent __read_mostly = DEFAULT_CPU_TIME_MAX_PERCENT;

int perf_cpu_time_max_percent_handler(struct ctl_table *table, int write,
				void __user *buffer, size_t *lenp,
				loff_t *ppos)
{
	int ret = proc_dointvec(table, write, buffer, lenp, ppos);

	if (ret || !write)
		return ret;

	update_perf_cpu_limits();
P
Peter Zijlstra 已提交
235 236 237

	return 0;
}
238

239 240 241 242 243 244 245
/*
 * perf samples are done in some very critical code paths (NMIs).
 * If they take too much CPU time, the system can lock up and not
 * get any real work done.  This will drop the sample rate when
 * we detect that events are taking too long.
 */
#define NR_ACCUMULATED_SAMPLES 128
P
Peter Zijlstra 已提交
246
static DEFINE_PER_CPU(u64, running_sample_length);
247

248
static void perf_duration_warn(struct irq_work *w)
249
{
250
	u64 allowed_ns = ACCESS_ONCE(perf_sample_allowed_ns);
251
	u64 avg_local_sample_len;
252
	u64 local_samples_len;
253

254
	local_samples_len = __this_cpu_read(running_sample_length);
255 256 257 258 259
	avg_local_sample_len = local_samples_len/NR_ACCUMULATED_SAMPLES;

	printk_ratelimited(KERN_WARNING
			"perf interrupt took too long (%lld > %lld), lowering "
			"kernel.perf_event_max_sample_rate to %d\n",
260
			avg_local_sample_len, allowed_ns >> 1,
261 262 263 264 265 266 267
			sysctl_perf_event_sample_rate);
}

static DEFINE_IRQ_WORK(perf_duration_work, perf_duration_warn);

void perf_sample_event_took(u64 sample_len_ns)
{
P
Peter Zijlstra 已提交
268
	u64 allowed_ns = ACCESS_ONCE(perf_sample_allowed_ns);
269 270
	u64 avg_local_sample_len;
	u64 local_samples_len;
271

P
Peter Zijlstra 已提交
272
	if (allowed_ns == 0)
273 274 275
		return;

	/* decay the counter by 1 average sample */
276
	local_samples_len = __this_cpu_read(running_sample_length);
277 278
	local_samples_len -= local_samples_len/NR_ACCUMULATED_SAMPLES;
	local_samples_len += sample_len_ns;
279
	__this_cpu_write(running_sample_length, local_samples_len);
280 281 282 283 284 285 286 287

	/*
	 * note: this will be biased artifically low until we have
	 * seen NR_ACCUMULATED_SAMPLES.  Doing it this way keeps us
	 * from having to maintain a count.
	 */
	avg_local_sample_len = local_samples_len/NR_ACCUMULATED_SAMPLES;

P
Peter Zijlstra 已提交
288
	if (avg_local_sample_len <= allowed_ns)
289 290 291 292 293 294 295 296 297 298
		return;

	if (max_samples_per_tick <= 1)
		return;

	max_samples_per_tick = DIV_ROUND_UP(max_samples_per_tick, 2);
	sysctl_perf_event_sample_rate = max_samples_per_tick * HZ;
	perf_sample_period_ns = NSEC_PER_SEC / sysctl_perf_event_sample_rate;

	update_perf_cpu_limits();
299

300 301 302 303 304 305
	if (!irq_work_queue(&perf_duration_work)) {
		early_printk("perf interrupt took too long (%lld > %lld), lowering "
			     "kernel.perf_event_max_sample_rate to %d\n",
			     avg_local_sample_len, allowed_ns >> 1,
			     sysctl_perf_event_sample_rate);
	}
306 307
}

308
static atomic64_t perf_event_id;
309

310 311 312 313
static void cpu_ctx_sched_out(struct perf_cpu_context *cpuctx,
			      enum event_type_t event_type);

static void cpu_ctx_sched_in(struct perf_cpu_context *cpuctx,
S
Stephane Eranian 已提交
314 315 316 317 318
			     enum event_type_t event_type,
			     struct task_struct *task);

static void update_context_time(struct perf_event_context *ctx);
static u64 perf_event_time(struct perf_event *event);
319

320
void __weak perf_event_print_debug(void)	{ }
T
Thomas Gleixner 已提交
321

322
extern __weak const char *perf_pmu_name(void)
T
Thomas Gleixner 已提交
323
{
324
	return "pmu";
T
Thomas Gleixner 已提交
325 326
}

327 328 329 330 331
static inline u64 perf_clock(void)
{
	return local_clock();
}

332 333 334 335 336
static inline u64 perf_event_clock(struct perf_event *event)
{
	return event->clock();
}

S
Stephane Eranian 已提交
337 338 339 340 341 342
static inline struct perf_cpu_context *
__get_cpu_context(struct perf_event_context *ctx)
{
	return this_cpu_ptr(ctx->pmu->pmu_cpu_context);
}

343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358
static void perf_ctx_lock(struct perf_cpu_context *cpuctx,
			  struct perf_event_context *ctx)
{
	raw_spin_lock(&cpuctx->ctx.lock);
	if (ctx)
		raw_spin_lock(&ctx->lock);
}

static void perf_ctx_unlock(struct perf_cpu_context *cpuctx,
			    struct perf_event_context *ctx)
{
	if (ctx)
		raw_spin_unlock(&ctx->lock);
	raw_spin_unlock(&cpuctx->ctx.lock);
}

S
Stephane Eranian 已提交
359 360 361 362 363 364 365 366
#ifdef CONFIG_CGROUP_PERF

static inline bool
perf_cgroup_match(struct perf_event *event)
{
	struct perf_event_context *ctx = event->ctx;
	struct perf_cpu_context *cpuctx = __get_cpu_context(ctx);

367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382
	/* @event doesn't care about cgroup */
	if (!event->cgrp)
		return true;

	/* wants specific cgroup scope but @cpuctx isn't associated with any */
	if (!cpuctx->cgrp)
		return false;

	/*
	 * Cgroup scoping is recursive.  An event enabled for a cgroup is
	 * also enabled for all its descendant cgroups.  If @cpuctx's
	 * cgroup is a descendant of @event's (the test covers identity
	 * case), it's a match.
	 */
	return cgroup_is_descendant(cpuctx->cgrp->css.cgroup,
				    event->cgrp->css.cgroup);
S
Stephane Eranian 已提交
383 384 385 386
}

static inline void perf_detach_cgroup(struct perf_event *event)
{
Z
Zefan Li 已提交
387
	css_put(&event->cgrp->css);
S
Stephane Eranian 已提交
388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425
	event->cgrp = NULL;
}

static inline int is_cgroup_event(struct perf_event *event)
{
	return event->cgrp != NULL;
}

static inline u64 perf_cgroup_event_time(struct perf_event *event)
{
	struct perf_cgroup_info *t;

	t = per_cpu_ptr(event->cgrp->info, event->cpu);
	return t->time;
}

static inline void __update_cgrp_time(struct perf_cgroup *cgrp)
{
	struct perf_cgroup_info *info;
	u64 now;

	now = perf_clock();

	info = this_cpu_ptr(cgrp->info);

	info->time += now - info->timestamp;
	info->timestamp = now;
}

static inline void update_cgrp_time_from_cpuctx(struct perf_cpu_context *cpuctx)
{
	struct perf_cgroup *cgrp_out = cpuctx->cgrp;
	if (cgrp_out)
		__update_cgrp_time(cgrp_out);
}

static inline void update_cgrp_time_from_event(struct perf_event *event)
{
426 427
	struct perf_cgroup *cgrp;

S
Stephane Eranian 已提交
428
	/*
429 430
	 * ensure we access cgroup data only when needed and
	 * when we know the cgroup is pinned (css_get)
S
Stephane Eranian 已提交
431
	 */
432
	if (!is_cgroup_event(event))
S
Stephane Eranian 已提交
433 434
		return;

435 436 437 438 439 440
	cgrp = perf_cgroup_from_task(current);
	/*
	 * Do not update time when cgroup is not active
	 */
	if (cgrp == event->cgrp)
		__update_cgrp_time(event->cgrp);
S
Stephane Eranian 已提交
441 442 443
}

static inline void
444 445
perf_cgroup_set_timestamp(struct task_struct *task,
			  struct perf_event_context *ctx)
S
Stephane Eranian 已提交
446 447 448 449
{
	struct perf_cgroup *cgrp;
	struct perf_cgroup_info *info;

450 451 452 453 454 455
	/*
	 * ctx->lock held by caller
	 * ensure we do not access cgroup data
	 * unless we have the cgroup pinned (css_get)
	 */
	if (!task || !ctx->nr_cgroups)
S
Stephane Eranian 已提交
456 457 458 459
		return;

	cgrp = perf_cgroup_from_task(task);
	info = this_cpu_ptr(cgrp->info);
460
	info->timestamp = ctx->timestamp;
S
Stephane Eranian 已提交
461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492
}

#define PERF_CGROUP_SWOUT	0x1 /* cgroup switch out every event */
#define PERF_CGROUP_SWIN	0x2 /* cgroup switch in events based on task */

/*
 * reschedule events based on the cgroup constraint of task.
 *
 * mode SWOUT : schedule out everything
 * mode SWIN : schedule in based on cgroup for next
 */
void perf_cgroup_switch(struct task_struct *task, int mode)
{
	struct perf_cpu_context *cpuctx;
	struct pmu *pmu;
	unsigned long flags;

	/*
	 * disable interrupts to avoid geting nr_cgroup
	 * changes via __perf_event_disable(). Also
	 * avoids preemption.
	 */
	local_irq_save(flags);

	/*
	 * we reschedule only in the presence of cgroup
	 * constrained events.
	 */
	rcu_read_lock();

	list_for_each_entry_rcu(pmu, &pmus, entry) {
		cpuctx = this_cpu_ptr(pmu->pmu_cpu_context);
493 494
		if (cpuctx->unique_pmu != pmu)
			continue; /* ensure we process each cpuctx once */
S
Stephane Eranian 已提交
495 496 497 498 499 500 501 502 503

		/*
		 * perf_cgroup_events says at least one
		 * context on this CPU has cgroup events.
		 *
		 * ctx->nr_cgroups reports the number of cgroup
		 * events for a context.
		 */
		if (cpuctx->ctx.nr_cgroups > 0) {
504 505
			perf_ctx_lock(cpuctx, cpuctx->task_ctx);
			perf_pmu_disable(cpuctx->ctx.pmu);
S
Stephane Eranian 已提交
506 507 508 509 510 511 512 513 514 515 516

			if (mode & PERF_CGROUP_SWOUT) {
				cpu_ctx_sched_out(cpuctx, EVENT_ALL);
				/*
				 * must not be done before ctxswout due
				 * to event_filter_match() in event_sched_out()
				 */
				cpuctx->cgrp = NULL;
			}

			if (mode & PERF_CGROUP_SWIN) {
517
				WARN_ON_ONCE(cpuctx->cgrp);
518 519 520 521
				/*
				 * set cgrp before ctxsw in to allow
				 * event_filter_match() to not have to pass
				 * task around
S
Stephane Eranian 已提交
522 523 524 525
				 */
				cpuctx->cgrp = perf_cgroup_from_task(task);
				cpu_ctx_sched_in(cpuctx, EVENT_ALL, task);
			}
526 527
			perf_pmu_enable(cpuctx->ctx.pmu);
			perf_ctx_unlock(cpuctx, cpuctx->task_ctx);
S
Stephane Eranian 已提交
528 529 530 531 532 533 534 535
		}
	}

	rcu_read_unlock();

	local_irq_restore(flags);
}

536 537
static inline void perf_cgroup_sched_out(struct task_struct *task,
					 struct task_struct *next)
S
Stephane Eranian 已提交
538
{
539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560
	struct perf_cgroup *cgrp1;
	struct perf_cgroup *cgrp2 = NULL;

	/*
	 * we come here when we know perf_cgroup_events > 0
	 */
	cgrp1 = perf_cgroup_from_task(task);

	/*
	 * next is NULL when called from perf_event_enable_on_exec()
	 * that will systematically cause a cgroup_switch()
	 */
	if (next)
		cgrp2 = perf_cgroup_from_task(next);

	/*
	 * only schedule out current cgroup events if we know
	 * that we are switching to a different cgroup. Otherwise,
	 * do no touch the cgroup events.
	 */
	if (cgrp1 != cgrp2)
		perf_cgroup_switch(task, PERF_CGROUP_SWOUT);
S
Stephane Eranian 已提交
561 562
}

563 564
static inline void perf_cgroup_sched_in(struct task_struct *prev,
					struct task_struct *task)
S
Stephane Eranian 已提交
565
{
566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583
	struct perf_cgroup *cgrp1;
	struct perf_cgroup *cgrp2 = NULL;

	/*
	 * we come here when we know perf_cgroup_events > 0
	 */
	cgrp1 = perf_cgroup_from_task(task);

	/* prev can never be NULL */
	cgrp2 = perf_cgroup_from_task(prev);

	/*
	 * only need to schedule in cgroup events if we are changing
	 * cgroup during ctxsw. Cgroup events were not scheduled
	 * out of ctxsw out if that was not the case.
	 */
	if (cgrp1 != cgrp2)
		perf_cgroup_switch(task, PERF_CGROUP_SWIN);
S
Stephane Eranian 已提交
584 585 586 587 588 589 590 591
}

static inline int perf_cgroup_connect(int fd, struct perf_event *event,
				      struct perf_event_attr *attr,
				      struct perf_event *group_leader)
{
	struct perf_cgroup *cgrp;
	struct cgroup_subsys_state *css;
592 593
	struct fd f = fdget(fd);
	int ret = 0;
S
Stephane Eranian 已提交
594

595
	if (!f.file)
S
Stephane Eranian 已提交
596 597
		return -EBADF;

A
Al Viro 已提交
598
	css = css_tryget_online_from_dir(f.file->f_path.dentry,
599
					 &perf_event_cgrp_subsys);
600 601 602 603
	if (IS_ERR(css)) {
		ret = PTR_ERR(css);
		goto out;
	}
S
Stephane Eranian 已提交
604 605 606 607 608 609 610 611 612 613 614 615 616

	cgrp = container_of(css, struct perf_cgroup, css);
	event->cgrp = cgrp;

	/*
	 * all events in a group must monitor
	 * the same cgroup because a task belongs
	 * to only one perf cgroup at a time
	 */
	if (group_leader && group_leader->cgrp != cgrp) {
		perf_detach_cgroup(event);
		ret = -EINVAL;
	}
617
out:
618
	fdput(f);
S
Stephane Eranian 已提交
619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691
	return ret;
}

static inline void
perf_cgroup_set_shadow_time(struct perf_event *event, u64 now)
{
	struct perf_cgroup_info *t;
	t = per_cpu_ptr(event->cgrp->info, event->cpu);
	event->shadow_ctx_time = now - t->timestamp;
}

static inline void
perf_cgroup_defer_enabled(struct perf_event *event)
{
	/*
	 * when the current task's perf cgroup does not match
	 * the event's, we need to remember to call the
	 * perf_mark_enable() function the first time a task with
	 * a matching perf cgroup is scheduled in.
	 */
	if (is_cgroup_event(event) && !perf_cgroup_match(event))
		event->cgrp_defer_enabled = 1;
}

static inline void
perf_cgroup_mark_enabled(struct perf_event *event,
			 struct perf_event_context *ctx)
{
	struct perf_event *sub;
	u64 tstamp = perf_event_time(event);

	if (!event->cgrp_defer_enabled)
		return;

	event->cgrp_defer_enabled = 0;

	event->tstamp_enabled = tstamp - event->total_time_enabled;
	list_for_each_entry(sub, &event->sibling_list, group_entry) {
		if (sub->state >= PERF_EVENT_STATE_INACTIVE) {
			sub->tstamp_enabled = tstamp - sub->total_time_enabled;
			sub->cgrp_defer_enabled = 0;
		}
	}
}
#else /* !CONFIG_CGROUP_PERF */

static inline bool
perf_cgroup_match(struct perf_event *event)
{
	return true;
}

static inline void perf_detach_cgroup(struct perf_event *event)
{}

static inline int is_cgroup_event(struct perf_event *event)
{
	return 0;
}

static inline u64 perf_cgroup_event_cgrp_time(struct perf_event *event)
{
	return 0;
}

static inline void update_cgrp_time_from_event(struct perf_event *event)
{
}

static inline void update_cgrp_time_from_cpuctx(struct perf_cpu_context *cpuctx)
{
}

692 693
static inline void perf_cgroup_sched_out(struct task_struct *task,
					 struct task_struct *next)
S
Stephane Eranian 已提交
694 695 696
{
}

697 698
static inline void perf_cgroup_sched_in(struct task_struct *prev,
					struct task_struct *task)
S
Stephane Eranian 已提交
699 700 701 702 703 704 705 706 707 708 709
{
}

static inline int perf_cgroup_connect(pid_t pid, struct perf_event *event,
				      struct perf_event_attr *attr,
				      struct perf_event *group_leader)
{
	return -EINVAL;
}

static inline void
710 711
perf_cgroup_set_timestamp(struct task_struct *task,
			  struct perf_event_context *ctx)
S
Stephane Eranian 已提交
712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741
{
}

void
perf_cgroup_switch(struct task_struct *task, struct task_struct *next)
{
}

static inline void
perf_cgroup_set_shadow_time(struct perf_event *event, u64 now)
{
}

static inline u64 perf_cgroup_event_time(struct perf_event *event)
{
	return 0;
}

static inline void
perf_cgroup_defer_enabled(struct perf_event *event)
{
}

static inline void
perf_cgroup_mark_enabled(struct perf_event *event,
			 struct perf_event_context *ctx)
{
}
#endif

742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804
/*
 * set default to be dependent on timer tick just
 * like original code
 */
#define PERF_CPU_HRTIMER (1000 / HZ)
/*
 * function must be called with interrupts disbled
 */
static enum hrtimer_restart perf_cpu_hrtimer_handler(struct hrtimer *hr)
{
	struct perf_cpu_context *cpuctx;
	enum hrtimer_restart ret = HRTIMER_NORESTART;
	int rotations = 0;

	WARN_ON(!irqs_disabled());

	cpuctx = container_of(hr, struct perf_cpu_context, hrtimer);

	rotations = perf_rotate_context(cpuctx);

	/*
	 * arm timer if needed
	 */
	if (rotations) {
		hrtimer_forward_now(hr, cpuctx->hrtimer_interval);
		ret = HRTIMER_RESTART;
	}

	return ret;
}

/* CPU is going down */
void perf_cpu_hrtimer_cancel(int cpu)
{
	struct perf_cpu_context *cpuctx;
	struct pmu *pmu;
	unsigned long flags;

	if (WARN_ON(cpu != smp_processor_id()))
		return;

	local_irq_save(flags);

	rcu_read_lock();

	list_for_each_entry_rcu(pmu, &pmus, entry) {
		cpuctx = this_cpu_ptr(pmu->pmu_cpu_context);

		if (pmu->task_ctx_nr == perf_sw_context)
			continue;

		hrtimer_cancel(&cpuctx->hrtimer);
	}

	rcu_read_unlock();

	local_irq_restore(flags);
}

static void __perf_cpu_hrtimer_init(struct perf_cpu_context *cpuctx, int cpu)
{
	struct hrtimer *hr = &cpuctx->hrtimer;
	struct pmu *pmu = cpuctx->ctx.pmu;
805
	int timer;
806 807 808 809 810

	/* no multiplexing needed for SW PMU */
	if (pmu->task_ctx_nr == perf_sw_context)
		return;

811 812 813 814 815 816 817 818 819
	/*
	 * check default is sane, if not set then force to
	 * default interval (1/tick)
	 */
	timer = pmu->hrtimer_interval_ms;
	if (timer < 1)
		timer = pmu->hrtimer_interval_ms = PERF_CPU_HRTIMER;

	cpuctx->hrtimer_interval = ns_to_ktime(NSEC_PER_MSEC * timer);
820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841

	hrtimer_init(hr, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
	hr->function = perf_cpu_hrtimer_handler;
}

static void perf_cpu_hrtimer_restart(struct perf_cpu_context *cpuctx)
{
	struct hrtimer *hr = &cpuctx->hrtimer;
	struct pmu *pmu = cpuctx->ctx.pmu;

	/* not for SW PMU */
	if (pmu->task_ctx_nr == perf_sw_context)
		return;

	if (hrtimer_active(hr))
		return;

	if (!hrtimer_callback_running(hr))
		__hrtimer_start_range_ns(hr, cpuctx->hrtimer_interval,
					 0, HRTIMER_MODE_REL_PINNED, 0);
}

P
Peter Zijlstra 已提交
842
void perf_pmu_disable(struct pmu *pmu)
843
{
P
Peter Zijlstra 已提交
844 845 846
	int *count = this_cpu_ptr(pmu->pmu_disable_count);
	if (!(*count)++)
		pmu->pmu_disable(pmu);
847 848
}

P
Peter Zijlstra 已提交
849
void perf_pmu_enable(struct pmu *pmu)
850
{
P
Peter Zijlstra 已提交
851 852 853
	int *count = this_cpu_ptr(pmu->pmu_disable_count);
	if (!--(*count))
		pmu->pmu_enable(pmu);
854 855
}

856
static DEFINE_PER_CPU(struct list_head, active_ctx_list);
857 858

/*
859 860 861 862
 * perf_event_ctx_activate(), perf_event_ctx_deactivate(), and
 * perf_event_task_tick() are fully serialized because they're strictly cpu
 * affine and perf_event_ctx{activate,deactivate} are called with IRQs
 * disabled, while perf_event_task_tick is called from IRQ context.
863
 */
864
static void perf_event_ctx_activate(struct perf_event_context *ctx)
865
{
866
	struct list_head *head = this_cpu_ptr(&active_ctx_list);
867

868
	WARN_ON(!irqs_disabled());
869

870 871 872 873 874 875 876 877 878 879 880 881
	WARN_ON(!list_empty(&ctx->active_ctx_list));

	list_add(&ctx->active_ctx_list, head);
}

static void perf_event_ctx_deactivate(struct perf_event_context *ctx)
{
	WARN_ON(!irqs_disabled());

	WARN_ON(list_empty(&ctx->active_ctx_list));

	list_del_init(&ctx->active_ctx_list);
882 883
}

884
static void get_ctx(struct perf_event_context *ctx)
885
{
886
	WARN_ON(!atomic_inc_not_zero(&ctx->refcount));
887 888
}

889 890 891 892 893 894 895 896 897
static void free_ctx(struct rcu_head *head)
{
	struct perf_event_context *ctx;

	ctx = container_of(head, struct perf_event_context, rcu_head);
	kfree(ctx->task_ctx_data);
	kfree(ctx);
}

898
static void put_ctx(struct perf_event_context *ctx)
899
{
900 901 902
	if (atomic_dec_and_test(&ctx->refcount)) {
		if (ctx->parent_ctx)
			put_ctx(ctx->parent_ctx);
903 904
		if (ctx->task)
			put_task_struct(ctx->task);
905
		call_rcu(&ctx->rcu_head, free_ctx);
906
	}
907 908
}

P
Peter Zijlstra 已提交
909 910 911 912 913 914 915
/*
 * Because of perf_event::ctx migration in sys_perf_event_open::move_group and
 * perf_pmu_migrate_context() we need some magic.
 *
 * Those places that change perf_event::ctx will hold both
 * perf_event_ctx::mutex of the 'old' and 'new' ctx value.
 *
916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939
 * Lock ordering is by mutex address. There are two other sites where
 * perf_event_context::mutex nests and those are:
 *
 *  - perf_event_exit_task_context()	[ child , 0 ]
 *      __perf_event_exit_task()
 *        sync_child_event()
 *          put_event()			[ parent, 1 ]
 *
 *  - perf_event_init_context()		[ parent, 0 ]
 *      inherit_task_group()
 *        inherit_group()
 *          inherit_event()
 *            perf_event_alloc()
 *              perf_init_event()
 *                perf_try_init_event()	[ child , 1 ]
 *
 * While it appears there is an obvious deadlock here -- the parent and child
 * nesting levels are inverted between the two. This is in fact safe because
 * life-time rules separate them. That is an exiting task cannot fork, and a
 * spawning task cannot (yet) exit.
 *
 * But remember that that these are parent<->child context relations, and
 * migration does not affect children, therefore these two orderings should not
 * interact.
P
Peter Zijlstra 已提交
940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969
 *
 * The change in perf_event::ctx does not affect children (as claimed above)
 * because the sys_perf_event_open() case will install a new event and break
 * the ctx parent<->child relation, and perf_pmu_migrate_context() is only
 * concerned with cpuctx and that doesn't have children.
 *
 * The places that change perf_event::ctx will issue:
 *
 *   perf_remove_from_context();
 *   synchronize_rcu();
 *   perf_install_in_context();
 *
 * to affect the change. The remove_from_context() + synchronize_rcu() should
 * quiesce the event, after which we can install it in the new location. This
 * means that only external vectors (perf_fops, prctl) can perturb the event
 * while in transit. Therefore all such accessors should also acquire
 * perf_event_context::mutex to serialize against this.
 *
 * However; because event->ctx can change while we're waiting to acquire
 * ctx->mutex we must be careful and use the below perf_event_ctx_lock()
 * function.
 *
 * Lock order:
 *	task_struct::perf_event_mutex
 *	  perf_event_context::mutex
 *	    perf_event_context::lock
 *	    perf_event::child_mutex;
 *	    perf_event::mmap_mutex
 *	    mmap_sem
 */
P
Peter Zijlstra 已提交
970 971
static struct perf_event_context *
perf_event_ctx_lock_nested(struct perf_event *event, int nesting)
P
Peter Zijlstra 已提交
972 973 974 975 976 977 978 979 980 981 982 983
{
	struct perf_event_context *ctx;

again:
	rcu_read_lock();
	ctx = ACCESS_ONCE(event->ctx);
	if (!atomic_inc_not_zero(&ctx->refcount)) {
		rcu_read_unlock();
		goto again;
	}
	rcu_read_unlock();

P
Peter Zijlstra 已提交
984
	mutex_lock_nested(&ctx->mutex, nesting);
P
Peter Zijlstra 已提交
985 986 987 988 989 990 991 992 993
	if (event->ctx != ctx) {
		mutex_unlock(&ctx->mutex);
		put_ctx(ctx);
		goto again;
	}

	return ctx;
}

P
Peter Zijlstra 已提交
994 995 996 997 998 999
static inline struct perf_event_context *
perf_event_ctx_lock(struct perf_event *event)
{
	return perf_event_ctx_lock_nested(event, 0);
}

P
Peter Zijlstra 已提交
1000 1001 1002 1003 1004 1005 1006
static void perf_event_ctx_unlock(struct perf_event *event,
				  struct perf_event_context *ctx)
{
	mutex_unlock(&ctx->mutex);
	put_ctx(ctx);
}

1007 1008 1009 1010 1011 1012 1013
/*
 * This must be done under the ctx->lock, such as to serialize against
 * context_equiv(), therefore we cannot call put_ctx() since that might end up
 * calling scheduler related locks and ctx->lock nests inside those.
 */
static __must_check struct perf_event_context *
unclone_ctx(struct perf_event_context *ctx)
1014
{
1015 1016 1017 1018 1019
	struct perf_event_context *parent_ctx = ctx->parent_ctx;

	lockdep_assert_held(&ctx->lock);

	if (parent_ctx)
1020
		ctx->parent_ctx = NULL;
1021
	ctx->generation++;
1022 1023

	return parent_ctx;
1024 1025
}

1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047
static u32 perf_event_pid(struct perf_event *event, struct task_struct *p)
{
	/*
	 * only top level events have the pid namespace they were created in
	 */
	if (event->parent)
		event = event->parent;

	return task_tgid_nr_ns(p, event->ns);
}

static u32 perf_event_tid(struct perf_event *event, struct task_struct *p)
{
	/*
	 * only top level events have the pid namespace they were created in
	 */
	if (event->parent)
		event = event->parent;

	return task_pid_nr_ns(p, event->ns);
}

1048
/*
1049
 * If we inherit events we want to return the parent event id
1050 1051
 * to userspace.
 */
1052
static u64 primary_event_id(struct perf_event *event)
1053
{
1054
	u64 id = event->id;
1055

1056 1057
	if (event->parent)
		id = event->parent->id;
1058 1059 1060 1061

	return id;
}

1062
/*
1063
 * Get the perf_event_context for a task and lock it.
1064 1065 1066
 * This has to cope with with the fact that until it is locked,
 * the context could get moved to another task.
 */
1067
static struct perf_event_context *
P
Peter Zijlstra 已提交
1068
perf_lock_task_context(struct task_struct *task, int ctxn, unsigned long *flags)
1069
{
1070
	struct perf_event_context *ctx;
1071

P
Peter Zijlstra 已提交
1072
retry:
1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083
	/*
	 * One of the few rules of preemptible RCU is that one cannot do
	 * rcu_read_unlock() while holding a scheduler (or nested) lock when
	 * part of the read side critical section was preemptible -- see
	 * rcu_read_unlock_special().
	 *
	 * Since ctx->lock nests under rq->lock we must ensure the entire read
	 * side critical section is non-preemptible.
	 */
	preempt_disable();
	rcu_read_lock();
P
Peter Zijlstra 已提交
1084
	ctx = rcu_dereference(task->perf_event_ctxp[ctxn]);
1085 1086 1087 1088
	if (ctx) {
		/*
		 * If this context is a clone of another, it might
		 * get swapped for another underneath us by
1089
		 * perf_event_task_sched_out, though the
1090 1091 1092 1093 1094 1095
		 * rcu_read_lock() protects us from any context
		 * getting freed.  Lock the context and check if it
		 * got swapped before we could get the lock, and retry
		 * if so.  If we locked the right context, then it
		 * can't get swapped on us any more.
		 */
1096
		raw_spin_lock_irqsave(&ctx->lock, *flags);
P
Peter Zijlstra 已提交
1097
		if (ctx != rcu_dereference(task->perf_event_ctxp[ctxn])) {
1098
			raw_spin_unlock_irqrestore(&ctx->lock, *flags);
1099 1100
			rcu_read_unlock();
			preempt_enable();
1101 1102
			goto retry;
		}
1103 1104

		if (!atomic_inc_not_zero(&ctx->refcount)) {
1105
			raw_spin_unlock_irqrestore(&ctx->lock, *flags);
1106 1107
			ctx = NULL;
		}
1108 1109
	}
	rcu_read_unlock();
1110
	preempt_enable();
1111 1112 1113 1114 1115 1116 1117 1118
	return ctx;
}

/*
 * Get the context for a task and increment its pin_count so it
 * can't get swapped to another task.  This also increments its
 * reference count so that the context can't get freed.
 */
P
Peter Zijlstra 已提交
1119 1120
static struct perf_event_context *
perf_pin_task_context(struct task_struct *task, int ctxn)
1121
{
1122
	struct perf_event_context *ctx;
1123 1124
	unsigned long flags;

P
Peter Zijlstra 已提交
1125
	ctx = perf_lock_task_context(task, ctxn, &flags);
1126 1127
	if (ctx) {
		++ctx->pin_count;
1128
		raw_spin_unlock_irqrestore(&ctx->lock, flags);
1129 1130 1131 1132
	}
	return ctx;
}

1133
static void perf_unpin_context(struct perf_event_context *ctx)
1134 1135 1136
{
	unsigned long flags;

1137
	raw_spin_lock_irqsave(&ctx->lock, flags);
1138
	--ctx->pin_count;
1139
	raw_spin_unlock_irqrestore(&ctx->lock, flags);
1140 1141
}

1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152
/*
 * Update the record of the current time in a context.
 */
static void update_context_time(struct perf_event_context *ctx)
{
	u64 now = perf_clock();

	ctx->time += now - ctx->timestamp;
	ctx->timestamp = now;
}

1153 1154 1155
static u64 perf_event_time(struct perf_event *event)
{
	struct perf_event_context *ctx = event->ctx;
S
Stephane Eranian 已提交
1156 1157 1158 1159

	if (is_cgroup_event(event))
		return perf_cgroup_event_time(event);

1160 1161 1162
	return ctx ? ctx->time : 0;
}

1163 1164
/*
 * Update the total_time_enabled and total_time_running fields for a event.
1165
 * The caller of this function needs to hold the ctx->lock.
1166 1167 1168 1169 1170 1171 1172 1173 1174
 */
static void update_event_times(struct perf_event *event)
{
	struct perf_event_context *ctx = event->ctx;
	u64 run_end;

	if (event->state < PERF_EVENT_STATE_INACTIVE ||
	    event->group_leader->state < PERF_EVENT_STATE_INACTIVE)
		return;
S
Stephane Eranian 已提交
1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185
	/*
	 * in cgroup mode, time_enabled represents
	 * the time the event was enabled AND active
	 * tasks were in the monitored cgroup. This is
	 * independent of the activity of the context as
	 * there may be a mix of cgroup and non-cgroup events.
	 *
	 * That is why we treat cgroup events differently
	 * here.
	 */
	if (is_cgroup_event(event))
1186
		run_end = perf_cgroup_event_time(event);
S
Stephane Eranian 已提交
1187 1188
	else if (ctx->is_active)
		run_end = ctx->time;
1189 1190 1191 1192
	else
		run_end = event->tstamp_stopped;

	event->total_time_enabled = run_end - event->tstamp_enabled;
1193 1194 1195 1196

	if (event->state == PERF_EVENT_STATE_INACTIVE)
		run_end = event->tstamp_stopped;
	else
1197
		run_end = perf_event_time(event);
1198 1199

	event->total_time_running = run_end - event->tstamp_running;
S
Stephane Eranian 已提交
1200

1201 1202
}

1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214
/*
 * Update total_time_enabled and total_time_running for all events in a group.
 */
static void update_group_times(struct perf_event *leader)
{
	struct perf_event *event;

	update_event_times(leader);
	list_for_each_entry(event, &leader->sibling_list, group_entry)
		update_event_times(event);
}

1215 1216 1217 1218 1219 1220 1221 1222 1223
static struct list_head *
ctx_group_list(struct perf_event *event, struct perf_event_context *ctx)
{
	if (event->attr.pinned)
		return &ctx->pinned_groups;
	else
		return &ctx->flexible_groups;
}

1224
/*
1225
 * Add a event from the lists for its context.
1226 1227
 * Must be called with ctx->mutex and ctx->lock held.
 */
1228
static void
1229
list_add_event(struct perf_event *event, struct perf_event_context *ctx)
1230
{
1231 1232
	WARN_ON_ONCE(event->attach_state & PERF_ATTACH_CONTEXT);
	event->attach_state |= PERF_ATTACH_CONTEXT;
1233 1234

	/*
1235 1236 1237
	 * If we're a stand alone event or group leader, we go to the context
	 * list, group events are kept attached to the group so that
	 * perf_group_detach can, at all times, locate all siblings.
1238
	 */
1239
	if (event->group_leader == event) {
1240 1241
		struct list_head *list;

1242 1243 1244
		if (is_software_event(event))
			event->group_flags |= PERF_GROUP_SOFTWARE;

1245 1246
		list = ctx_group_list(event, ctx);
		list_add_tail(&event->group_entry, list);
P
Peter Zijlstra 已提交
1247
	}
P
Peter Zijlstra 已提交
1248

1249
	if (is_cgroup_event(event))
S
Stephane Eranian 已提交
1250 1251
		ctx->nr_cgroups++;

1252 1253 1254
	list_add_rcu(&event->event_entry, &ctx->event_list);
	ctx->nr_events++;
	if (event->attr.inherit_stat)
1255
		ctx->nr_stat++;
1256 1257

	ctx->generation++;
1258 1259
}

J
Jiri Olsa 已提交
1260 1261 1262 1263 1264 1265 1266 1267 1268
/*
 * Initialize event state based on the perf_event_attr::disabled.
 */
static inline void perf_event__state_init(struct perf_event *event)
{
	event->state = event->attr.disabled ? PERF_EVENT_STATE_OFF :
					      PERF_EVENT_STATE_INACTIVE;
}

1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307
/*
 * Called at perf_event creation and when events are attached/detached from a
 * group.
 */
static void perf_event__read_size(struct perf_event *event)
{
	int entry = sizeof(u64); /* value */
	int size = 0;
	int nr = 1;

	if (event->attr.read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
		size += sizeof(u64);

	if (event->attr.read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
		size += sizeof(u64);

	if (event->attr.read_format & PERF_FORMAT_ID)
		entry += sizeof(u64);

	if (event->attr.read_format & PERF_FORMAT_GROUP) {
		nr += event->group_leader->nr_siblings;
		size += sizeof(u64);
	}

	size += entry * nr;
	event->read_size = size;
}

static void perf_event__header_size(struct perf_event *event)
{
	struct perf_sample_data *data;
	u64 sample_type = event->attr.sample_type;
	u16 size = 0;

	perf_event__read_size(event);

	if (sample_type & PERF_SAMPLE_IP)
		size += sizeof(data->ip);

1308 1309 1310 1311 1312 1313
	if (sample_type & PERF_SAMPLE_ADDR)
		size += sizeof(data->addr);

	if (sample_type & PERF_SAMPLE_PERIOD)
		size += sizeof(data->period);

A
Andi Kleen 已提交
1314 1315 1316
	if (sample_type & PERF_SAMPLE_WEIGHT)
		size += sizeof(data->weight);

1317 1318 1319
	if (sample_type & PERF_SAMPLE_READ)
		size += event->read_size;

1320 1321 1322
	if (sample_type & PERF_SAMPLE_DATA_SRC)
		size += sizeof(data->data_src.val);

A
Andi Kleen 已提交
1323 1324 1325
	if (sample_type & PERF_SAMPLE_TRANSACTION)
		size += sizeof(data->txn);

1326 1327 1328 1329 1330 1331 1332 1333 1334
	event->header_size = size;
}

static void perf_event__id_header_size(struct perf_event *event)
{
	struct perf_sample_data *data;
	u64 sample_type = event->attr.sample_type;
	u16 size = 0;

1335 1336 1337 1338 1339 1340
	if (sample_type & PERF_SAMPLE_TID)
		size += sizeof(data->tid_entry);

	if (sample_type & PERF_SAMPLE_TIME)
		size += sizeof(data->time);

1341 1342 1343
	if (sample_type & PERF_SAMPLE_IDENTIFIER)
		size += sizeof(data->id);

1344 1345 1346 1347 1348 1349 1350 1351 1352
	if (sample_type & PERF_SAMPLE_ID)
		size += sizeof(data->id);

	if (sample_type & PERF_SAMPLE_STREAM_ID)
		size += sizeof(data->stream_id);

	if (sample_type & PERF_SAMPLE_CPU)
		size += sizeof(data->cpu_entry);

1353
	event->id_header_size = size;
1354 1355
}

1356 1357
static void perf_group_attach(struct perf_event *event)
{
1358
	struct perf_event *group_leader = event->group_leader, *pos;
1359

P
Peter Zijlstra 已提交
1360 1361 1362 1363 1364 1365
	/*
	 * We can have double attach due to group movement in perf_event_open.
	 */
	if (event->attach_state & PERF_ATTACH_GROUP)
		return;

1366 1367 1368 1369 1370
	event->attach_state |= PERF_ATTACH_GROUP;

	if (group_leader == event)
		return;

P
Peter Zijlstra 已提交
1371 1372
	WARN_ON_ONCE(group_leader->ctx != event->ctx);

1373 1374 1375 1376 1377 1378
	if (group_leader->group_flags & PERF_GROUP_SOFTWARE &&
			!is_software_event(event))
		group_leader->group_flags &= ~PERF_GROUP_SOFTWARE;

	list_add_tail(&event->group_entry, &group_leader->sibling_list);
	group_leader->nr_siblings++;
1379 1380 1381 1382 1383

	perf_event__header_size(group_leader);

	list_for_each_entry(pos, &group_leader->sibling_list, group_entry)
		perf_event__header_size(pos);
1384 1385
}

1386
/*
1387
 * Remove a event from the lists for its context.
1388
 * Must be called with ctx->mutex and ctx->lock held.
1389
 */
1390
static void
1391
list_del_event(struct perf_event *event, struct perf_event_context *ctx)
1392
{
1393
	struct perf_cpu_context *cpuctx;
P
Peter Zijlstra 已提交
1394 1395 1396 1397

	WARN_ON_ONCE(event->ctx != ctx);
	lockdep_assert_held(&ctx->lock);

1398 1399 1400 1401
	/*
	 * We can have double detach due to exit/hot-unplug + close.
	 */
	if (!(event->attach_state & PERF_ATTACH_CONTEXT))
1402
		return;
1403 1404 1405

	event->attach_state &= ~PERF_ATTACH_CONTEXT;

1406
	if (is_cgroup_event(event)) {
S
Stephane Eranian 已提交
1407
		ctx->nr_cgroups--;
1408 1409 1410 1411 1412 1413 1414 1415 1416
		cpuctx = __get_cpu_context(ctx);
		/*
		 * if there are no more cgroup events
		 * then cler cgrp to avoid stale pointer
		 * in update_cgrp_time_from_cpuctx()
		 */
		if (!ctx->nr_cgroups)
			cpuctx->cgrp = NULL;
	}
S
Stephane Eranian 已提交
1417

1418 1419
	ctx->nr_events--;
	if (event->attr.inherit_stat)
1420
		ctx->nr_stat--;
1421

1422
	list_del_rcu(&event->event_entry);
1423

1424 1425
	if (event->group_leader == event)
		list_del_init(&event->group_entry);
P
Peter Zijlstra 已提交
1426

1427
	update_group_times(event);
1428 1429 1430 1431 1432 1433 1434 1435 1436 1437

	/*
	 * If event was in error state, then keep it
	 * that way, otherwise bogus counts will be
	 * returned on read(). The only way to get out
	 * of error state is by explicit re-enabling
	 * of the event
	 */
	if (event->state > PERF_EVENT_STATE_OFF)
		event->state = PERF_EVENT_STATE_OFF;
1438 1439

	ctx->generation++;
1440 1441
}

1442
static void perf_group_detach(struct perf_event *event)
1443 1444
{
	struct perf_event *sibling, *tmp;
1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460
	struct list_head *list = NULL;

	/*
	 * We can have double detach due to exit/hot-unplug + close.
	 */
	if (!(event->attach_state & PERF_ATTACH_GROUP))
		return;

	event->attach_state &= ~PERF_ATTACH_GROUP;

	/*
	 * If this is a sibling, remove it from its group.
	 */
	if (event->group_leader != event) {
		list_del_init(&event->group_entry);
		event->group_leader->nr_siblings--;
1461
		goto out;
1462 1463 1464 1465
	}

	if (!list_empty(&event->group_entry))
		list = &event->group_entry;
1466

1467
	/*
1468 1469
	 * If this was a group event with sibling events then
	 * upgrade the siblings to singleton events by adding them
1470
	 * to whatever list we are on.
1471
	 */
1472
	list_for_each_entry_safe(sibling, tmp, &event->sibling_list, group_entry) {
1473 1474
		if (list)
			list_move_tail(&sibling->group_entry, list);
1475
		sibling->group_leader = sibling;
1476 1477 1478

		/* Inherit group flags from the previous leader */
		sibling->group_flags = event->group_flags;
P
Peter Zijlstra 已提交
1479 1480

		WARN_ON_ONCE(sibling->ctx != event->ctx);
1481
	}
1482 1483 1484 1485 1486 1487

out:
	perf_event__header_size(event->group_leader);

	list_for_each_entry(tmp, &event->group_leader->sibling_list, group_entry)
		perf_event__header_size(tmp);
1488 1489
}

1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528
/*
 * User event without the task.
 */
static bool is_orphaned_event(struct perf_event *event)
{
	return event && !is_kernel_event(event) && !event->owner;
}

/*
 * Event has a parent but parent's task finished and it's
 * alive only because of children holding refference.
 */
static bool is_orphaned_child(struct perf_event *event)
{
	return is_orphaned_event(event->parent);
}

static void orphans_remove_work(struct work_struct *work);

static void schedule_orphans_remove(struct perf_event_context *ctx)
{
	if (!ctx->task || ctx->orphans_remove_sched || !perf_wq)
		return;

	if (queue_delayed_work(perf_wq, &ctx->orphans_remove, 1)) {
		get_ctx(ctx);
		ctx->orphans_remove_sched = true;
	}
}

static int __init perf_workqueue_init(void)
{
	perf_wq = create_singlethread_workqueue("perf");
	WARN(!perf_wq, "failed to create perf workqueue\n");
	return perf_wq ? 0 : -1;
}

core_initcall(perf_workqueue_init);

1529 1530 1531
static inline int
event_filter_match(struct perf_event *event)
{
S
Stephane Eranian 已提交
1532 1533
	return (event->cpu == -1 || event->cpu == smp_processor_id())
	    && perf_cgroup_match(event);
1534 1535
}

1536 1537
static void
event_sched_out(struct perf_event *event,
1538
		  struct perf_cpu_context *cpuctx,
1539
		  struct perf_event_context *ctx)
1540
{
1541
	u64 tstamp = perf_event_time(event);
1542
	u64 delta;
P
Peter Zijlstra 已提交
1543 1544 1545 1546

	WARN_ON_ONCE(event->ctx != ctx);
	lockdep_assert_held(&ctx->lock);

1547 1548 1549 1550 1551 1552 1553 1554
	/*
	 * An event which could not be activated because of
	 * filter mismatch still needs to have its timings
	 * maintained, otherwise bogus information is return
	 * via read() for time_enabled, time_running:
	 */
	if (event->state == PERF_EVENT_STATE_INACTIVE
	    && !event_filter_match(event)) {
S
Stephane Eranian 已提交
1555
		delta = tstamp - event->tstamp_stopped;
1556
		event->tstamp_running += delta;
1557
		event->tstamp_stopped = tstamp;
1558 1559
	}

1560
	if (event->state != PERF_EVENT_STATE_ACTIVE)
1561
		return;
1562

1563 1564
	perf_pmu_disable(event->pmu);

1565 1566 1567 1568
	event->state = PERF_EVENT_STATE_INACTIVE;
	if (event->pending_disable) {
		event->pending_disable = 0;
		event->state = PERF_EVENT_STATE_OFF;
1569
	}
1570
	event->tstamp_stopped = tstamp;
P
Peter Zijlstra 已提交
1571
	event->pmu->del(event, 0);
1572
	event->oncpu = -1;
1573

1574
	if (!is_software_event(event))
1575
		cpuctx->active_oncpu--;
1576 1577
	if (!--ctx->nr_active)
		perf_event_ctx_deactivate(ctx);
1578 1579
	if (event->attr.freq && event->attr.sample_freq)
		ctx->nr_freq--;
1580
	if (event->attr.exclusive || !cpuctx->active_oncpu)
1581
		cpuctx->exclusive = 0;
1582

1583 1584 1585
	if (is_orphaned_child(event))
		schedule_orphans_remove(ctx);

1586
	perf_pmu_enable(event->pmu);
1587 1588
}

1589
static void
1590
group_sched_out(struct perf_event *group_event,
1591
		struct perf_cpu_context *cpuctx,
1592
		struct perf_event_context *ctx)
1593
{
1594
	struct perf_event *event;
1595
	int state = group_event->state;
1596

1597
	event_sched_out(group_event, cpuctx, ctx);
1598 1599 1600 1601

	/*
	 * Schedule out siblings (if any):
	 */
1602 1603
	list_for_each_entry(event, &group_event->sibling_list, group_entry)
		event_sched_out(event, cpuctx, ctx);
1604

1605
	if (state == PERF_EVENT_STATE_ACTIVE && group_event->attr.exclusive)
1606 1607 1608
		cpuctx->exclusive = 0;
}

1609 1610 1611 1612 1613
struct remove_event {
	struct perf_event *event;
	bool detach_group;
};

T
Thomas Gleixner 已提交
1614
/*
1615
 * Cross CPU call to remove a performance event
T
Thomas Gleixner 已提交
1616
 *
1617
 * We disable the event on the hardware level first. After that we
T
Thomas Gleixner 已提交
1618 1619
 * remove it from the context list.
 */
1620
static int __perf_remove_from_context(void *info)
T
Thomas Gleixner 已提交
1621
{
1622 1623
	struct remove_event *re = info;
	struct perf_event *event = re->event;
1624
	struct perf_event_context *ctx = event->ctx;
P
Peter Zijlstra 已提交
1625
	struct perf_cpu_context *cpuctx = __get_cpu_context(ctx);
T
Thomas Gleixner 已提交
1626

1627
	raw_spin_lock(&ctx->lock);
1628
	event_sched_out(event, cpuctx, ctx);
1629 1630
	if (re->detach_group)
		perf_group_detach(event);
1631
	list_del_event(event, ctx);
1632 1633 1634 1635
	if (!ctx->nr_events && cpuctx->task_ctx == ctx) {
		ctx->is_active = 0;
		cpuctx->task_ctx = NULL;
	}
1636
	raw_spin_unlock(&ctx->lock);
1637 1638

	return 0;
T
Thomas Gleixner 已提交
1639 1640 1641 1642
}


/*
1643
 * Remove the event from a task's (or a CPU's) list of events.
T
Thomas Gleixner 已提交
1644
 *
1645
 * CPU events are removed with a smp call. For task events we only
T
Thomas Gleixner 已提交
1646
 * call when the task is on a CPU.
1647
 *
1648 1649
 * If event->ctx is a cloned context, callers must make sure that
 * every task struct that event->ctx->task could possibly point to
1650 1651
 * remains valid.  This is OK when called from perf_release since
 * that only calls us on the top-level context, which can't be a clone.
1652
 * When called from perf_event_exit_task, it's OK because the
1653
 * context has been detached from its task.
T
Thomas Gleixner 已提交
1654
 */
1655
static void perf_remove_from_context(struct perf_event *event, bool detach_group)
T
Thomas Gleixner 已提交
1656
{
1657
	struct perf_event_context *ctx = event->ctx;
T
Thomas Gleixner 已提交
1658
	struct task_struct *task = ctx->task;
1659 1660 1661 1662
	struct remove_event re = {
		.event = event,
		.detach_group = detach_group,
	};
T
Thomas Gleixner 已提交
1663

1664 1665
	lockdep_assert_held(&ctx->mutex);

T
Thomas Gleixner 已提交
1666 1667
	if (!task) {
		/*
1668 1669 1670 1671
		 * Per cpu events are removed via an smp call. The removal can
		 * fail if the CPU is currently offline, but in that case we
		 * already called __perf_remove_from_context from
		 * perf_event_exit_cpu.
T
Thomas Gleixner 已提交
1672
		 */
1673
		cpu_function_call(event->cpu, __perf_remove_from_context, &re);
T
Thomas Gleixner 已提交
1674 1675 1676 1677
		return;
	}

retry:
1678
	if (!task_function_call(task, __perf_remove_from_context, &re))
1679
		return;
T
Thomas Gleixner 已提交
1680

1681
	raw_spin_lock_irq(&ctx->lock);
T
Thomas Gleixner 已提交
1682
	/*
1683 1684
	 * If we failed to find a running task, but find the context active now
	 * that we've acquired the ctx->lock, retry.
T
Thomas Gleixner 已提交
1685
	 */
1686
	if (ctx->is_active) {
1687
		raw_spin_unlock_irq(&ctx->lock);
1688 1689 1690 1691 1692
		/*
		 * Reload the task pointer, it might have been changed by
		 * a concurrent perf_event_context_sched_out().
		 */
		task = ctx->task;
T
Thomas Gleixner 已提交
1693 1694 1695 1696
		goto retry;
	}

	/*
1697 1698
	 * Since the task isn't running, its safe to remove the event, us
	 * holding the ctx->lock ensures the task won't get scheduled in.
T
Thomas Gleixner 已提交
1699
	 */
1700 1701
	if (detach_group)
		perf_group_detach(event);
1702
	list_del_event(event, ctx);
1703
	raw_spin_unlock_irq(&ctx->lock);
T
Thomas Gleixner 已提交
1704 1705
}

1706
/*
1707
 * Cross CPU call to disable a performance event
1708
 */
1709
int __perf_event_disable(void *info)
1710
{
1711 1712
	struct perf_event *event = info;
	struct perf_event_context *ctx = event->ctx;
P
Peter Zijlstra 已提交
1713
	struct perf_cpu_context *cpuctx = __get_cpu_context(ctx);
1714 1715

	/*
1716 1717
	 * If this is a per-task event, need to check whether this
	 * event's task is the current task on this cpu.
1718 1719 1720
	 *
	 * Can trigger due to concurrent perf_event_context_sched_out()
	 * flipping contexts around.
1721
	 */
1722
	if (ctx->task && cpuctx->task_ctx != ctx)
1723
		return -EINVAL;
1724

1725
	raw_spin_lock(&ctx->lock);
1726 1727

	/*
1728
	 * If the event is on, turn it off.
1729 1730
	 * If it is in error state, leave it in error state.
	 */
1731
	if (event->state >= PERF_EVENT_STATE_INACTIVE) {
1732
		update_context_time(ctx);
S
Stephane Eranian 已提交
1733
		update_cgrp_time_from_event(event);
1734 1735 1736
		update_group_times(event);
		if (event == event->group_leader)
			group_sched_out(event, cpuctx, ctx);
1737
		else
1738 1739
			event_sched_out(event, cpuctx, ctx);
		event->state = PERF_EVENT_STATE_OFF;
1740 1741
	}

1742
	raw_spin_unlock(&ctx->lock);
1743 1744

	return 0;
1745 1746 1747
}

/*
1748
 * Disable a event.
1749
 *
1750 1751
 * If event->ctx is a cloned context, callers must make sure that
 * every task struct that event->ctx->task could possibly point to
1752
 * remains valid.  This condition is satisifed when called through
1753 1754 1755 1756
 * perf_event_for_each_child or perf_event_for_each because they
 * hold the top-level event's child_mutex, so any descendant that
 * goes to exit will block in sync_child_event.
 * When called from perf_pending_event it's OK because event->ctx
1757
 * is the current context on this CPU and preemption is disabled,
1758
 * hence we can't get into perf_event_task_sched_out for this context.
1759
 */
P
Peter Zijlstra 已提交
1760
static void _perf_event_disable(struct perf_event *event)
1761
{
1762
	struct perf_event_context *ctx = event->ctx;
1763 1764 1765 1766
	struct task_struct *task = ctx->task;

	if (!task) {
		/*
1767
		 * Disable the event on the cpu that it's on
1768
		 */
1769
		cpu_function_call(event->cpu, __perf_event_disable, event);
1770 1771 1772
		return;
	}

P
Peter Zijlstra 已提交
1773
retry:
1774 1775
	if (!task_function_call(task, __perf_event_disable, event))
		return;
1776

1777
	raw_spin_lock_irq(&ctx->lock);
1778
	/*
1779
	 * If the event is still active, we need to retry the cross-call.
1780
	 */
1781
	if (event->state == PERF_EVENT_STATE_ACTIVE) {
1782
		raw_spin_unlock_irq(&ctx->lock);
1783 1784 1785 1786 1787
		/*
		 * Reload the task pointer, it might have been changed by
		 * a concurrent perf_event_context_sched_out().
		 */
		task = ctx->task;
1788 1789 1790 1791 1792 1793 1794
		goto retry;
	}

	/*
	 * Since we have the lock this context can't be scheduled
	 * in, so we can change the state safely.
	 */
1795 1796 1797
	if (event->state == PERF_EVENT_STATE_INACTIVE) {
		update_group_times(event);
		event->state = PERF_EVENT_STATE_OFF;
1798
	}
1799
	raw_spin_unlock_irq(&ctx->lock);
1800
}
P
Peter Zijlstra 已提交
1801 1802 1803 1804 1805 1806 1807 1808 1809 1810 1811 1812 1813

/*
 * Strictly speaking kernel users cannot create groups and therefore this
 * interface does not need the perf_event_ctx_lock() magic.
 */
void perf_event_disable(struct perf_event *event)
{
	struct perf_event_context *ctx;

	ctx = perf_event_ctx_lock(event);
	_perf_event_disable(event);
	perf_event_ctx_unlock(event, ctx);
}
1814
EXPORT_SYMBOL_GPL(perf_event_disable);
1815

S
Stephane Eranian 已提交
1816 1817 1818 1819 1820 1821 1822 1823 1824 1825 1826 1827 1828 1829 1830 1831 1832 1833 1834 1835 1836 1837 1838 1839 1840 1841 1842 1843 1844 1845 1846 1847 1848 1849 1850
static void perf_set_shadow_time(struct perf_event *event,
				 struct perf_event_context *ctx,
				 u64 tstamp)
{
	/*
	 * use the correct time source for the time snapshot
	 *
	 * We could get by without this by leveraging the
	 * fact that to get to this function, the caller
	 * has most likely already called update_context_time()
	 * and update_cgrp_time_xx() and thus both timestamp
	 * are identical (or very close). Given that tstamp is,
	 * already adjusted for cgroup, we could say that:
	 *    tstamp - ctx->timestamp
	 * is equivalent to
	 *    tstamp - cgrp->timestamp.
	 *
	 * Then, in perf_output_read(), the calculation would
	 * work with no changes because:
	 * - event is guaranteed scheduled in
	 * - no scheduled out in between
	 * - thus the timestamp would be the same
	 *
	 * But this is a bit hairy.
	 *
	 * So instead, we have an explicit cgroup call to remain
	 * within the time time source all along. We believe it
	 * is cleaner and simpler to understand.
	 */
	if (is_cgroup_event(event))
		perf_cgroup_set_shadow_time(event, tstamp);
	else
		event->shadow_ctx_time = tstamp - ctx->timestamp;
}

P
Peter Zijlstra 已提交
1851 1852 1853
#define MAX_INTERRUPTS (~0ULL)

static void perf_log_throttle(struct perf_event *event, int enable);
1854
static void perf_log_itrace_start(struct perf_event *event);
P
Peter Zijlstra 已提交
1855

1856
static int
1857
event_sched_in(struct perf_event *event,
1858
		 struct perf_cpu_context *cpuctx,
1859
		 struct perf_event_context *ctx)
1860
{
1861
	u64 tstamp = perf_event_time(event);
1862
	int ret = 0;
1863

1864 1865
	lockdep_assert_held(&ctx->lock);

1866
	if (event->state <= PERF_EVENT_STATE_OFF)
1867 1868
		return 0;

1869
	event->state = PERF_EVENT_STATE_ACTIVE;
1870
	event->oncpu = smp_processor_id();
P
Peter Zijlstra 已提交
1871 1872 1873 1874 1875 1876 1877 1878 1879 1880 1881

	/*
	 * Unthrottle events, since we scheduled we might have missed several
	 * ticks already, also for a heavily scheduling task there is little
	 * guarantee it'll get a tick in a timely manner.
	 */
	if (unlikely(event->hw.interrupts == MAX_INTERRUPTS)) {
		perf_log_throttle(event, 1);
		event->hw.interrupts = 0;
	}

1882 1883 1884 1885 1886
	/*
	 * The new state must be visible before we turn it on in the hardware:
	 */
	smp_wmb();

1887 1888
	perf_pmu_disable(event->pmu);

1889 1890 1891 1892
	event->tstamp_running += tstamp - event->tstamp_stopped;

	perf_set_shadow_time(event, ctx, tstamp);

1893 1894
	perf_log_itrace_start(event);

P
Peter Zijlstra 已提交
1895
	if (event->pmu->add(event, PERF_EF_START)) {
1896 1897
		event->state = PERF_EVENT_STATE_INACTIVE;
		event->oncpu = -1;
1898 1899
		ret = -EAGAIN;
		goto out;
1900 1901
	}

1902
	if (!is_software_event(event))
1903
		cpuctx->active_oncpu++;
1904 1905
	if (!ctx->nr_active++)
		perf_event_ctx_activate(ctx);
1906 1907
	if (event->attr.freq && event->attr.sample_freq)
		ctx->nr_freq++;
1908

1909
	if (event->attr.exclusive)
1910 1911
		cpuctx->exclusive = 1;

1912 1913 1914
	if (is_orphaned_child(event))
		schedule_orphans_remove(ctx);

1915 1916 1917 1918
out:
	perf_pmu_enable(event->pmu);

	return ret;
1919 1920
}

1921
static int
1922
group_sched_in(struct perf_event *group_event,
1923
	       struct perf_cpu_context *cpuctx,
1924
	       struct perf_event_context *ctx)
1925
{
1926
	struct perf_event *event, *partial_group = NULL;
P
Peter Zijlstra 已提交
1927
	struct pmu *pmu = ctx->pmu;
1928 1929
	u64 now = ctx->time;
	bool simulate = false;
1930

1931
	if (group_event->state == PERF_EVENT_STATE_OFF)
1932 1933
		return 0;

P
Peter Zijlstra 已提交
1934
	pmu->start_txn(pmu);
1935

1936
	if (event_sched_in(group_event, cpuctx, ctx)) {
P
Peter Zijlstra 已提交
1937
		pmu->cancel_txn(pmu);
1938
		perf_cpu_hrtimer_restart(cpuctx);
1939
		return -EAGAIN;
1940
	}
1941 1942 1943 1944

	/*
	 * Schedule in siblings as one group (if any):
	 */
1945
	list_for_each_entry(event, &group_event->sibling_list, group_entry) {
1946
		if (event_sched_in(event, cpuctx, ctx)) {
1947
			partial_group = event;
1948 1949 1950 1951
			goto group_error;
		}
	}

1952
	if (!pmu->commit_txn(pmu))
1953
		return 0;
1954

1955 1956 1957 1958
group_error:
	/*
	 * Groups can be scheduled in as one unit only, so undo any
	 * partial group before returning:
1959 1960 1961 1962 1963 1964 1965 1966 1967 1968
	 * The events up to the failed event are scheduled out normally,
	 * tstamp_stopped will be updated.
	 *
	 * The failed events and the remaining siblings need to have
	 * their timings updated as if they had gone thru event_sched_in()
	 * and event_sched_out(). This is required to get consistent timings
	 * across the group. This also takes care of the case where the group
	 * could never be scheduled by ensuring tstamp_stopped is set to mark
	 * the time the event was actually stopped, such that time delta
	 * calculation in update_event_times() is correct.
1969
	 */
1970 1971
	list_for_each_entry(event, &group_event->sibling_list, group_entry) {
		if (event == partial_group)
1972 1973 1974 1975 1976 1977 1978 1979
			simulate = true;

		if (simulate) {
			event->tstamp_running += now - event->tstamp_stopped;
			event->tstamp_stopped = now;
		} else {
			event_sched_out(event, cpuctx, ctx);
		}
1980
	}
1981
	event_sched_out(group_event, cpuctx, ctx);
1982

P
Peter Zijlstra 已提交
1983
	pmu->cancel_txn(pmu);
1984

1985 1986
	perf_cpu_hrtimer_restart(cpuctx);

1987 1988 1989
	return -EAGAIN;
}

1990
/*
1991
 * Work out whether we can put this event group on the CPU now.
1992
 */
1993
static int group_can_go_on(struct perf_event *event,
1994 1995 1996 1997
			   struct perf_cpu_context *cpuctx,
			   int can_add_hw)
{
	/*
1998
	 * Groups consisting entirely of software events can always go on.
1999
	 */
2000
	if (event->group_flags & PERF_GROUP_SOFTWARE)
2001 2002 2003
		return 1;
	/*
	 * If an exclusive group is already on, no other hardware
2004
	 * events can go on.
2005 2006 2007 2008 2009
	 */
	if (cpuctx->exclusive)
		return 0;
	/*
	 * If this group is exclusive and there are already
2010
	 * events on the CPU, it can't go on.
2011
	 */
2012
	if (event->attr.exclusive && cpuctx->active_oncpu)
2013 2014 2015 2016 2017 2018 2019 2020
		return 0;
	/*
	 * Otherwise, try to add it if all previous groups were able
	 * to go on.
	 */
	return can_add_hw;
}

2021 2022
static void add_event_to_ctx(struct perf_event *event,
			       struct perf_event_context *ctx)
2023
{
2024 2025
	u64 tstamp = perf_event_time(event);

2026
	list_add_event(event, ctx);
2027
	perf_group_attach(event);
2028 2029 2030
	event->tstamp_enabled = tstamp;
	event->tstamp_running = tstamp;
	event->tstamp_stopped = tstamp;
2031 2032
}

2033 2034 2035 2036 2037 2038
static void task_ctx_sched_out(struct perf_event_context *ctx);
static void
ctx_sched_in(struct perf_event_context *ctx,
	     struct perf_cpu_context *cpuctx,
	     enum event_type_t event_type,
	     struct task_struct *task);
2039

2040 2041 2042 2043 2044 2045 2046 2047 2048 2049 2050 2051
static void perf_event_sched_in(struct perf_cpu_context *cpuctx,
				struct perf_event_context *ctx,
				struct task_struct *task)
{
	cpu_ctx_sched_in(cpuctx, EVENT_PINNED, task);
	if (ctx)
		ctx_sched_in(ctx, cpuctx, EVENT_PINNED, task);
	cpu_ctx_sched_in(cpuctx, EVENT_FLEXIBLE, task);
	if (ctx)
		ctx_sched_in(ctx, cpuctx, EVENT_FLEXIBLE, task);
}

T
Thomas Gleixner 已提交
2052
/*
2053
 * Cross CPU call to install and enable a performance event
2054 2055
 *
 * Must be called with ctx->mutex held
T
Thomas Gleixner 已提交
2056
 */
2057
static int  __perf_install_in_context(void *info)
T
Thomas Gleixner 已提交
2058
{
2059 2060
	struct perf_event *event = info;
	struct perf_event_context *ctx = event->ctx;
P
Peter Zijlstra 已提交
2061
	struct perf_cpu_context *cpuctx = __get_cpu_context(ctx);
2062 2063 2064
	struct perf_event_context *task_ctx = cpuctx->task_ctx;
	struct task_struct *task = current;

2065
	perf_ctx_lock(cpuctx, task_ctx);
2066
	perf_pmu_disable(cpuctx->ctx.pmu);
T
Thomas Gleixner 已提交
2067 2068

	/*
2069
	 * If there was an active task_ctx schedule it out.
T
Thomas Gleixner 已提交
2070
	 */
2071
	if (task_ctx)
2072
		task_ctx_sched_out(task_ctx);
2073 2074 2075 2076 2077 2078 2079 2080 2081 2082 2083 2084 2085 2086

	/*
	 * If the context we're installing events in is not the
	 * active task_ctx, flip them.
	 */
	if (ctx->task && task_ctx != ctx) {
		if (task_ctx)
			raw_spin_unlock(&task_ctx->lock);
		raw_spin_lock(&ctx->lock);
		task_ctx = ctx;
	}

	if (task_ctx) {
		cpuctx->task_ctx = task_ctx;
2087 2088
		task = task_ctx->task;
	}
2089

2090
	cpu_ctx_sched_out(cpuctx, EVENT_ALL);
T
Thomas Gleixner 已提交
2091

2092
	update_context_time(ctx);
S
Stephane Eranian 已提交
2093 2094 2095 2096 2097 2098
	/*
	 * update cgrp time only if current cgrp
	 * matches event->cgrp. Must be done before
	 * calling add_event_to_ctx()
	 */
	update_cgrp_time_from_event(event);
T
Thomas Gleixner 已提交
2099

2100
	add_event_to_ctx(event, ctx);
T
Thomas Gleixner 已提交
2101

2102
	/*
2103
	 * Schedule everything back in
2104
	 */
2105
	perf_event_sched_in(cpuctx, task_ctx, task);
2106 2107 2108

	perf_pmu_enable(cpuctx->ctx.pmu);
	perf_ctx_unlock(cpuctx, task_ctx);
2109 2110

	return 0;
T
Thomas Gleixner 已提交
2111 2112 2113
}

/*
2114
 * Attach a performance event to a context
T
Thomas Gleixner 已提交
2115
 *
2116 2117
 * First we add the event to the list with the hardware enable bit
 * in event->hw_config cleared.
T
Thomas Gleixner 已提交
2118
 *
2119
 * If the event is attached to a task which is on a CPU we use a smp
T
Thomas Gleixner 已提交
2120 2121 2122 2123
 * call to enable it in the task context. The task might have been
 * scheduled away, but we check this in the smp call again.
 */
static void
2124 2125
perf_install_in_context(struct perf_event_context *ctx,
			struct perf_event *event,
T
Thomas Gleixner 已提交
2126 2127 2128 2129
			int cpu)
{
	struct task_struct *task = ctx->task;

2130 2131
	lockdep_assert_held(&ctx->mutex);

2132
	event->ctx = ctx;
2133 2134
	if (event->cpu != -1)
		event->cpu = cpu;
2135

T
Thomas Gleixner 已提交
2136 2137
	if (!task) {
		/*
2138
		 * Per cpu events are installed via an smp call and
2139
		 * the install is always successful.
T
Thomas Gleixner 已提交
2140
		 */
2141
		cpu_function_call(cpu, __perf_install_in_context, event);
T
Thomas Gleixner 已提交
2142 2143 2144 2145
		return;
	}

retry:
2146 2147
	if (!task_function_call(task, __perf_install_in_context, event))
		return;
T
Thomas Gleixner 已提交
2148

2149
	raw_spin_lock_irq(&ctx->lock);
T
Thomas Gleixner 已提交
2150
	/*
2151 2152
	 * If we failed to find a running task, but find the context active now
	 * that we've acquired the ctx->lock, retry.
T
Thomas Gleixner 已提交
2153
	 */
2154
	if (ctx->is_active) {
2155
		raw_spin_unlock_irq(&ctx->lock);
2156 2157 2158 2159 2160
		/*
		 * Reload the task pointer, it might have been changed by
		 * a concurrent perf_event_context_sched_out().
		 */
		task = ctx->task;
T
Thomas Gleixner 已提交
2161 2162 2163 2164
		goto retry;
	}

	/*
2165 2166
	 * Since the task isn't running, its safe to add the event, us holding
	 * the ctx->lock ensures the task won't get scheduled in.
T
Thomas Gleixner 已提交
2167
	 */
2168
	add_event_to_ctx(event, ctx);
2169
	raw_spin_unlock_irq(&ctx->lock);
T
Thomas Gleixner 已提交
2170 2171
}

2172
/*
2173
 * Put a event into inactive state and update time fields.
2174 2175 2176 2177 2178 2179
 * Enabling the leader of a group effectively enables all
 * the group members that aren't explicitly disabled, so we
 * have to update their ->tstamp_enabled also.
 * Note: this works for group members as well as group leaders
 * since the non-leader members' sibling_lists will be empty.
 */
2180
static void __perf_event_mark_enabled(struct perf_event *event)
2181
{
2182
	struct perf_event *sub;
2183
	u64 tstamp = perf_event_time(event);
2184

2185
	event->state = PERF_EVENT_STATE_INACTIVE;
2186
	event->tstamp_enabled = tstamp - event->total_time_enabled;
P
Peter Zijlstra 已提交
2187
	list_for_each_entry(sub, &event->sibling_list, group_entry) {
2188 2189
		if (sub->state >= PERF_EVENT_STATE_INACTIVE)
			sub->tstamp_enabled = tstamp - sub->total_time_enabled;
P
Peter Zijlstra 已提交
2190
	}
2191 2192
}

2193
/*
2194
 * Cross CPU call to enable a performance event
2195
 */
2196
static int __perf_event_enable(void *info)
2197
{
2198 2199 2200
	struct perf_event *event = info;
	struct perf_event_context *ctx = event->ctx;
	struct perf_event *leader = event->group_leader;
P
Peter Zijlstra 已提交
2201
	struct perf_cpu_context *cpuctx = __get_cpu_context(ctx);
2202
	int err;
2203

2204 2205 2206 2207 2208 2209 2210 2211 2212 2213
	/*
	 * There's a time window between 'ctx->is_active' check
	 * in perf_event_enable function and this place having:
	 *   - IRQs on
	 *   - ctx->lock unlocked
	 *
	 * where the task could be killed and 'ctx' deactivated
	 * by perf_event_exit_task.
	 */
	if (!ctx->is_active)
2214
		return -EINVAL;
2215

2216
	raw_spin_lock(&ctx->lock);
2217
	update_context_time(ctx);
2218

2219
	if (event->state >= PERF_EVENT_STATE_INACTIVE)
2220
		goto unlock;
S
Stephane Eranian 已提交
2221 2222 2223 2224

	/*
	 * set current task's cgroup time reference point
	 */
2225
	perf_cgroup_set_timestamp(current, ctx);
S
Stephane Eranian 已提交
2226

2227
	__perf_event_mark_enabled(event);
2228

S
Stephane Eranian 已提交
2229 2230 2231
	if (!event_filter_match(event)) {
		if (is_cgroup_event(event))
			perf_cgroup_defer_enabled(event);
2232
		goto unlock;
S
Stephane Eranian 已提交
2233
	}
2234

2235
	/*
2236
	 * If the event is in a group and isn't the group leader,
2237
	 * then don't put it on unless the group is on.
2238
	 */
2239
	if (leader != event && leader->state != PERF_EVENT_STATE_ACTIVE)
2240
		goto unlock;
2241

2242
	if (!group_can_go_on(event, cpuctx, 1)) {
2243
		err = -EEXIST;
2244
	} else {
2245
		if (event == leader)
2246
			err = group_sched_in(event, cpuctx, ctx);
2247
		else
2248
			err = event_sched_in(event, cpuctx, ctx);
2249
	}
2250 2251 2252

	if (err) {
		/*
2253
		 * If this event can't go on and it's part of a
2254 2255
		 * group, then the whole group has to come off.
		 */
2256
		if (leader != event) {
2257
			group_sched_out(leader, cpuctx, ctx);
2258 2259
			perf_cpu_hrtimer_restart(cpuctx);
		}
2260
		if (leader->attr.pinned) {
2261
			update_group_times(leader);
2262
			leader->state = PERF_EVENT_STATE_ERROR;
2263
		}
2264 2265
	}

P
Peter Zijlstra 已提交
2266
unlock:
2267
	raw_spin_unlock(&ctx->lock);
2268 2269

	return 0;
2270 2271 2272
}

/*
2273
 * Enable a event.
2274
 *
2275 2276
 * If event->ctx is a cloned context, callers must make sure that
 * every task struct that event->ctx->task could possibly point to
2277
 * remains valid.  This condition is satisfied when called through
2278 2279
 * perf_event_for_each_child or perf_event_for_each as described
 * for perf_event_disable.
2280
 */
P
Peter Zijlstra 已提交
2281
static void _perf_event_enable(struct perf_event *event)
2282
{
2283
	struct perf_event_context *ctx = event->ctx;
2284 2285 2286 2287
	struct task_struct *task = ctx->task;

	if (!task) {
		/*
2288
		 * Enable the event on the cpu that it's on
2289
		 */
2290
		cpu_function_call(event->cpu, __perf_event_enable, event);
2291 2292 2293
		return;
	}

2294
	raw_spin_lock_irq(&ctx->lock);
2295
	if (event->state >= PERF_EVENT_STATE_INACTIVE)
2296 2297 2298
		goto out;

	/*
2299 2300
	 * If the event is in error state, clear that first.
	 * That way, if we see the event in error state below, we
2301 2302 2303 2304
	 * know that it has gone back into error state, as distinct
	 * from the task having been scheduled away before the
	 * cross-call arrived.
	 */
2305 2306
	if (event->state == PERF_EVENT_STATE_ERROR)
		event->state = PERF_EVENT_STATE_OFF;
2307

P
Peter Zijlstra 已提交
2308
retry:
2309
	if (!ctx->is_active) {
2310
		__perf_event_mark_enabled(event);
2311 2312 2313
		goto out;
	}

2314
	raw_spin_unlock_irq(&ctx->lock);
2315 2316 2317

	if (!task_function_call(task, __perf_event_enable, event))
		return;
2318

2319
	raw_spin_lock_irq(&ctx->lock);
2320 2321

	/*
2322
	 * If the context is active and the event is still off,
2323 2324
	 * we need to retry the cross-call.
	 */
2325 2326 2327 2328 2329 2330
	if (ctx->is_active && event->state == PERF_EVENT_STATE_OFF) {
		/*
		 * task could have been flipped by a concurrent
		 * perf_event_context_sched_out()
		 */
		task = ctx->task;
2331
		goto retry;
2332
	}
2333

P
Peter Zijlstra 已提交
2334
out:
2335
	raw_spin_unlock_irq(&ctx->lock);
2336
}
P
Peter Zijlstra 已提交
2337 2338 2339 2340 2341 2342 2343 2344 2345 2346 2347 2348

/*
 * See perf_event_disable();
 */
void perf_event_enable(struct perf_event *event)
{
	struct perf_event_context *ctx;

	ctx = perf_event_ctx_lock(event);
	_perf_event_enable(event);
	perf_event_ctx_unlock(event, ctx);
}
2349
EXPORT_SYMBOL_GPL(perf_event_enable);
2350

P
Peter Zijlstra 已提交
2351
static int _perf_event_refresh(struct perf_event *event, int refresh)
2352
{
2353
	/*
2354
	 * not supported on inherited events
2355
	 */
2356
	if (event->attr.inherit || !is_sampling_event(event))
2357 2358
		return -EINVAL;

2359
	atomic_add(refresh, &event->event_limit);
P
Peter Zijlstra 已提交
2360
	_perf_event_enable(event);
2361 2362

	return 0;
2363
}
P
Peter Zijlstra 已提交
2364 2365 2366 2367 2368 2369 2370 2371 2372 2373 2374 2375 2376 2377 2378

/*
 * See perf_event_disable()
 */
int perf_event_refresh(struct perf_event *event, int refresh)
{
	struct perf_event_context *ctx;
	int ret;

	ctx = perf_event_ctx_lock(event);
	ret = _perf_event_refresh(event, refresh);
	perf_event_ctx_unlock(event, ctx);

	return ret;
}
2379
EXPORT_SYMBOL_GPL(perf_event_refresh);
2380

2381 2382 2383
static void ctx_sched_out(struct perf_event_context *ctx,
			  struct perf_cpu_context *cpuctx,
			  enum event_type_t event_type)
2384
{
2385
	struct perf_event *event;
2386
	int is_active = ctx->is_active;
2387

2388
	ctx->is_active &= ~event_type;
2389
	if (likely(!ctx->nr_events))
2390 2391
		return;

2392
	update_context_time(ctx);
S
Stephane Eranian 已提交
2393
	update_cgrp_time_from_cpuctx(cpuctx);
2394
	if (!ctx->nr_active)
2395
		return;
2396

P
Peter Zijlstra 已提交
2397
	perf_pmu_disable(ctx->pmu);
2398
	if ((is_active & EVENT_PINNED) && (event_type & EVENT_PINNED)) {
2399 2400
		list_for_each_entry(event, &ctx->pinned_groups, group_entry)
			group_sched_out(event, cpuctx, ctx);
P
Peter Zijlstra 已提交
2401
	}
2402

2403
	if ((is_active & EVENT_FLEXIBLE) && (event_type & EVENT_FLEXIBLE)) {
2404
		list_for_each_entry(event, &ctx->flexible_groups, group_entry)
2405
			group_sched_out(event, cpuctx, ctx);
P
Peter Zijlstra 已提交
2406
	}
P
Peter Zijlstra 已提交
2407
	perf_pmu_enable(ctx->pmu);
2408 2409
}

2410
/*
2411 2412 2413 2414 2415 2416
 * Test whether two contexts are equivalent, i.e. whether they have both been
 * cloned from the same version of the same context.
 *
 * Equivalence is measured using a generation number in the context that is
 * incremented on each modification to it; see unclone_ctx(), list_add_event()
 * and list_del_event().
2417
 */
2418 2419
static int context_equiv(struct perf_event_context *ctx1,
			 struct perf_event_context *ctx2)
2420
{
2421 2422 2423
	lockdep_assert_held(&ctx1->lock);
	lockdep_assert_held(&ctx2->lock);

2424 2425 2426 2427 2428 2429 2430 2431 2432 2433 2434 2435 2436 2437 2438 2439 2440 2441 2442 2443 2444 2445
	/* Pinning disables the swap optimization */
	if (ctx1->pin_count || ctx2->pin_count)
		return 0;

	/* If ctx1 is the parent of ctx2 */
	if (ctx1 == ctx2->parent_ctx && ctx1->generation == ctx2->parent_gen)
		return 1;

	/* If ctx2 is the parent of ctx1 */
	if (ctx1->parent_ctx == ctx2 && ctx1->parent_gen == ctx2->generation)
		return 1;

	/*
	 * If ctx1 and ctx2 have the same parent; we flatten the parent
	 * hierarchy, see perf_event_init_context().
	 */
	if (ctx1->parent_ctx && ctx1->parent_ctx == ctx2->parent_ctx &&
			ctx1->parent_gen == ctx2->parent_gen)
		return 1;

	/* Unmatched */
	return 0;
2446 2447
}

2448 2449
static void __perf_event_sync_stat(struct perf_event *event,
				     struct perf_event *next_event)
2450 2451 2452
{
	u64 value;

2453
	if (!event->attr.inherit_stat)
2454 2455 2456
		return;

	/*
2457
	 * Update the event value, we cannot use perf_event_read()
2458 2459
	 * because we're in the middle of a context switch and have IRQs
	 * disabled, which upsets smp_call_function_single(), however
2460
	 * we know the event must be on the current CPU, therefore we
2461 2462
	 * don't need to use it.
	 */
2463 2464
	switch (event->state) {
	case PERF_EVENT_STATE_ACTIVE:
2465 2466
		event->pmu->read(event);
		/* fall-through */
2467

2468 2469
	case PERF_EVENT_STATE_INACTIVE:
		update_event_times(event);
2470 2471 2472 2473 2474 2475 2476
		break;

	default:
		break;
	}

	/*
2477
	 * In order to keep per-task stats reliable we need to flip the event
2478 2479
	 * values when we flip the contexts.
	 */
2480 2481 2482
	value = local64_read(&next_event->count);
	value = local64_xchg(&event->count, value);
	local64_set(&next_event->count, value);
2483

2484 2485
	swap(event->total_time_enabled, next_event->total_time_enabled);
	swap(event->total_time_running, next_event->total_time_running);
2486

2487
	/*
2488
	 * Since we swizzled the values, update the user visible data too.
2489
	 */
2490 2491
	perf_event_update_userpage(event);
	perf_event_update_userpage(next_event);
2492 2493
}

2494 2495
static void perf_event_sync_stat(struct perf_event_context *ctx,
				   struct perf_event_context *next_ctx)
2496
{
2497
	struct perf_event *event, *next_event;
2498 2499 2500 2501

	if (!ctx->nr_stat)
		return;

2502 2503
	update_context_time(ctx);

2504 2505
	event = list_first_entry(&ctx->event_list,
				   struct perf_event, event_entry);
2506

2507 2508
	next_event = list_first_entry(&next_ctx->event_list,
					struct perf_event, event_entry);
2509

2510 2511
	while (&event->event_entry != &ctx->event_list &&
	       &next_event->event_entry != &next_ctx->event_list) {
2512

2513
		__perf_event_sync_stat(event, next_event);
2514

2515 2516
		event = list_next_entry(event, event_entry);
		next_event = list_next_entry(next_event, event_entry);
2517 2518 2519
	}
}

2520 2521
static void perf_event_context_sched_out(struct task_struct *task, int ctxn,
					 struct task_struct *next)
T
Thomas Gleixner 已提交
2522
{
P
Peter Zijlstra 已提交
2523
	struct perf_event_context *ctx = task->perf_event_ctxp[ctxn];
2524
	struct perf_event_context *next_ctx;
2525
	struct perf_event_context *parent, *next_parent;
P
Peter Zijlstra 已提交
2526
	struct perf_cpu_context *cpuctx;
2527
	int do_switch = 1;
T
Thomas Gleixner 已提交
2528

P
Peter Zijlstra 已提交
2529 2530
	if (likely(!ctx))
		return;
2531

P
Peter Zijlstra 已提交
2532 2533
	cpuctx = __get_cpu_context(ctx);
	if (!cpuctx->task_ctx)
T
Thomas Gleixner 已提交
2534 2535
		return;

2536
	rcu_read_lock();
P
Peter Zijlstra 已提交
2537
	next_ctx = next->perf_event_ctxp[ctxn];
2538 2539 2540 2541 2542 2543 2544
	if (!next_ctx)
		goto unlock;

	parent = rcu_dereference(ctx->parent_ctx);
	next_parent = rcu_dereference(next_ctx->parent_ctx);

	/* If neither context have a parent context; they cannot be clones. */
2545
	if (!parent && !next_parent)
2546 2547 2548
		goto unlock;

	if (next_parent == ctx || next_ctx == parent || next_parent == parent) {
2549 2550 2551 2552 2553 2554 2555 2556 2557
		/*
		 * Looks like the two contexts are clones, so we might be
		 * able to optimize the context switch.  We lock both
		 * contexts and check that they are clones under the
		 * lock (including re-checking that neither has been
		 * uncloned in the meantime).  It doesn't matter which
		 * order we take the locks because no other cpu could
		 * be trying to lock both of these tasks.
		 */
2558 2559
		raw_spin_lock(&ctx->lock);
		raw_spin_lock_nested(&next_ctx->lock, SINGLE_DEPTH_NESTING);
2560
		if (context_equiv(ctx, next_ctx)) {
2561 2562
			/*
			 * XXX do we need a memory barrier of sorts
2563
			 * wrt to rcu_dereference() of perf_event_ctxp
2564
			 */
P
Peter Zijlstra 已提交
2565 2566
			task->perf_event_ctxp[ctxn] = next_ctx;
			next->perf_event_ctxp[ctxn] = ctx;
2567 2568
			ctx->task = next;
			next_ctx->task = task;
2569 2570 2571

			swap(ctx->task_ctx_data, next_ctx->task_ctx_data);

2572
			do_switch = 0;
2573

2574
			perf_event_sync_stat(ctx, next_ctx);
2575
		}
2576 2577
		raw_spin_unlock(&next_ctx->lock);
		raw_spin_unlock(&ctx->lock);
2578
	}
2579
unlock:
2580
	rcu_read_unlock();
2581

2582
	if (do_switch) {
2583
		raw_spin_lock(&ctx->lock);
2584
		ctx_sched_out(ctx, cpuctx, EVENT_ALL);
2585
		cpuctx->task_ctx = NULL;
2586
		raw_spin_unlock(&ctx->lock);
2587
	}
T
Thomas Gleixner 已提交
2588 2589
}

2590 2591 2592 2593 2594 2595 2596 2597 2598 2599 2600 2601 2602 2603 2604 2605 2606 2607 2608 2609 2610 2611 2612 2613 2614 2615 2616 2617 2618 2619 2620 2621 2622 2623 2624 2625 2626 2627 2628 2629 2630 2631 2632 2633 2634 2635 2636 2637 2638 2639
void perf_sched_cb_dec(struct pmu *pmu)
{
	this_cpu_dec(perf_sched_cb_usages);
}

void perf_sched_cb_inc(struct pmu *pmu)
{
	this_cpu_inc(perf_sched_cb_usages);
}

/*
 * This function provides the context switch callback to the lower code
 * layer. It is invoked ONLY when the context switch callback is enabled.
 */
static void perf_pmu_sched_task(struct task_struct *prev,
				struct task_struct *next,
				bool sched_in)
{
	struct perf_cpu_context *cpuctx;
	struct pmu *pmu;
	unsigned long flags;

	if (prev == next)
		return;

	local_irq_save(flags);

	rcu_read_lock();

	list_for_each_entry_rcu(pmu, &pmus, entry) {
		if (pmu->sched_task) {
			cpuctx = this_cpu_ptr(pmu->pmu_cpu_context);

			perf_ctx_lock(cpuctx, cpuctx->task_ctx);

			perf_pmu_disable(pmu);

			pmu->sched_task(cpuctx->task_ctx, sched_in);

			perf_pmu_enable(pmu);

			perf_ctx_unlock(cpuctx, cpuctx->task_ctx);
		}
	}

	rcu_read_unlock();

	local_irq_restore(flags);
}

P
Peter Zijlstra 已提交
2640 2641 2642 2643 2644 2645 2646 2647 2648 2649 2650 2651 2652 2653
#define for_each_task_context_nr(ctxn)					\
	for ((ctxn) = 0; (ctxn) < perf_nr_task_contexts; (ctxn)++)

/*
 * Called from scheduler to remove the events of the current task,
 * with interrupts disabled.
 *
 * We stop each event and update the event value in event->count.
 *
 * This does not protect us against NMI, but disable()
 * sets the disabled bit in the control field of event _before_
 * accessing the event control register. If a NMI hits, then it will
 * not restart the event.
 */
2654 2655
void __perf_event_task_sched_out(struct task_struct *task,
				 struct task_struct *next)
P
Peter Zijlstra 已提交
2656 2657 2658
{
	int ctxn;

2659 2660 2661
	if (__this_cpu_read(perf_sched_cb_usages))
		perf_pmu_sched_task(task, next, false);

P
Peter Zijlstra 已提交
2662 2663
	for_each_task_context_nr(ctxn)
		perf_event_context_sched_out(task, ctxn, next);
S
Stephane Eranian 已提交
2664 2665 2666 2667 2668 2669

	/*
	 * if cgroup events exist on this CPU, then we need
	 * to check if we have to switch out PMU state.
	 * cgroup event are system-wide mode only
	 */
2670
	if (atomic_read(this_cpu_ptr(&perf_cgroup_events)))
2671
		perf_cgroup_sched_out(task, next);
P
Peter Zijlstra 已提交
2672 2673
}

2674
static void task_ctx_sched_out(struct perf_event_context *ctx)
2675
{
P
Peter Zijlstra 已提交
2676
	struct perf_cpu_context *cpuctx = __get_cpu_context(ctx);
2677

2678 2679
	if (!cpuctx->task_ctx)
		return;
2680 2681 2682 2683

	if (WARN_ON_ONCE(ctx != cpuctx->task_ctx))
		return;

2684
	ctx_sched_out(ctx, cpuctx, EVENT_ALL);
2685 2686 2687
	cpuctx->task_ctx = NULL;
}

2688 2689 2690 2691 2692 2693 2694
/*
 * Called with IRQs disabled
 */
static void cpu_ctx_sched_out(struct perf_cpu_context *cpuctx,
			      enum event_type_t event_type)
{
	ctx_sched_out(&cpuctx->ctx, cpuctx, event_type);
2695 2696
}

2697
static void
2698
ctx_pinned_sched_in(struct perf_event_context *ctx,
2699
		    struct perf_cpu_context *cpuctx)
T
Thomas Gleixner 已提交
2700
{
2701
	struct perf_event *event;
T
Thomas Gleixner 已提交
2702

2703 2704
	list_for_each_entry(event, &ctx->pinned_groups, group_entry) {
		if (event->state <= PERF_EVENT_STATE_OFF)
2705
			continue;
2706
		if (!event_filter_match(event))
2707 2708
			continue;

S
Stephane Eranian 已提交
2709 2710 2711 2712
		/* may need to reset tstamp_enabled */
		if (is_cgroup_event(event))
			perf_cgroup_mark_enabled(event, ctx);

2713
		if (group_can_go_on(event, cpuctx, 1))
2714
			group_sched_in(event, cpuctx, ctx);
2715 2716 2717 2718 2719

		/*
		 * If this pinned group hasn't been scheduled,
		 * put it in error state.
		 */
2720 2721 2722
		if (event->state == PERF_EVENT_STATE_INACTIVE) {
			update_group_times(event);
			event->state = PERF_EVENT_STATE_ERROR;
2723
		}
2724
	}
2725 2726 2727 2728
}

static void
ctx_flexible_sched_in(struct perf_event_context *ctx,
2729
		      struct perf_cpu_context *cpuctx)
2730 2731 2732
{
	struct perf_event *event;
	int can_add_hw = 1;
2733

2734 2735 2736
	list_for_each_entry(event, &ctx->flexible_groups, group_entry) {
		/* Ignore events in OFF or ERROR state */
		if (event->state <= PERF_EVENT_STATE_OFF)
2737
			continue;
2738 2739
		/*
		 * Listen to the 'cpu' scheduling filter constraint
2740
		 * of events:
2741
		 */
2742
		if (!event_filter_match(event))
T
Thomas Gleixner 已提交
2743 2744
			continue;

S
Stephane Eranian 已提交
2745 2746 2747 2748
		/* may need to reset tstamp_enabled */
		if (is_cgroup_event(event))
			perf_cgroup_mark_enabled(event, ctx);

P
Peter Zijlstra 已提交
2749
		if (group_can_go_on(event, cpuctx, can_add_hw)) {
2750
			if (group_sched_in(event, cpuctx, ctx))
2751
				can_add_hw = 0;
P
Peter Zijlstra 已提交
2752
		}
T
Thomas Gleixner 已提交
2753
	}
2754 2755 2756 2757 2758
}

static void
ctx_sched_in(struct perf_event_context *ctx,
	     struct perf_cpu_context *cpuctx,
S
Stephane Eranian 已提交
2759 2760
	     enum event_type_t event_type,
	     struct task_struct *task)
2761
{
S
Stephane Eranian 已提交
2762
	u64 now;
2763
	int is_active = ctx->is_active;
S
Stephane Eranian 已提交
2764

2765
	ctx->is_active |= event_type;
2766
	if (likely(!ctx->nr_events))
2767
		return;
2768

S
Stephane Eranian 已提交
2769 2770
	now = perf_clock();
	ctx->timestamp = now;
2771
	perf_cgroup_set_timestamp(task, ctx);
2772 2773 2774 2775
	/*
	 * First go through the list and put on any pinned groups
	 * in order to give them the best chance of going on.
	 */
2776
	if (!(is_active & EVENT_PINNED) && (event_type & EVENT_PINNED))
2777
		ctx_pinned_sched_in(ctx, cpuctx);
2778 2779

	/* Then walk through the lower prio flexible groups */
2780
	if (!(is_active & EVENT_FLEXIBLE) && (event_type & EVENT_FLEXIBLE))
2781
		ctx_flexible_sched_in(ctx, cpuctx);
2782 2783
}

2784
static void cpu_ctx_sched_in(struct perf_cpu_context *cpuctx,
S
Stephane Eranian 已提交
2785 2786
			     enum event_type_t event_type,
			     struct task_struct *task)
2787 2788 2789
{
	struct perf_event_context *ctx = &cpuctx->ctx;

S
Stephane Eranian 已提交
2790
	ctx_sched_in(ctx, cpuctx, event_type, task);
2791 2792
}

S
Stephane Eranian 已提交
2793 2794
static void perf_event_context_sched_in(struct perf_event_context *ctx,
					struct task_struct *task)
2795
{
P
Peter Zijlstra 已提交
2796
	struct perf_cpu_context *cpuctx;
2797

P
Peter Zijlstra 已提交
2798
	cpuctx = __get_cpu_context(ctx);
2799 2800 2801
	if (cpuctx->task_ctx == ctx)
		return;

2802
	perf_ctx_lock(cpuctx, ctx);
P
Peter Zijlstra 已提交
2803
	perf_pmu_disable(ctx->pmu);
2804 2805 2806 2807 2808 2809 2810
	/*
	 * We want to keep the following priority order:
	 * cpu pinned (that don't need to move), task pinned,
	 * cpu flexible, task flexible.
	 */
	cpu_ctx_sched_out(cpuctx, EVENT_FLEXIBLE);

2811 2812
	if (ctx->nr_events)
		cpuctx->task_ctx = ctx;
2813

2814 2815
	perf_event_sched_in(cpuctx, cpuctx->task_ctx, task);

2816 2817
	perf_pmu_enable(ctx->pmu);
	perf_ctx_unlock(cpuctx, ctx);
2818 2819
}

P
Peter Zijlstra 已提交
2820 2821 2822 2823 2824 2825 2826 2827 2828 2829 2830
/*
 * Called from scheduler to add the events of the current task
 * with interrupts disabled.
 *
 * We restore the event value and then enable it.
 *
 * This does not protect us against NMI, but enable()
 * sets the enabled bit in the control field of event _before_
 * accessing the event control register. If a NMI hits, then it will
 * keep the event running.
 */
2831 2832
void __perf_event_task_sched_in(struct task_struct *prev,
				struct task_struct *task)
P
Peter Zijlstra 已提交
2833 2834 2835 2836 2837 2838 2839 2840 2841
{
	struct perf_event_context *ctx;
	int ctxn;

	for_each_task_context_nr(ctxn) {
		ctx = task->perf_event_ctxp[ctxn];
		if (likely(!ctx))
			continue;

S
Stephane Eranian 已提交
2842
		perf_event_context_sched_in(ctx, task);
P
Peter Zijlstra 已提交
2843
	}
S
Stephane Eranian 已提交
2844 2845 2846 2847 2848
	/*
	 * if cgroup events exist on this CPU, then we need
	 * to check if we have to switch in PMU state.
	 * cgroup event are system-wide mode only
	 */
2849
	if (atomic_read(this_cpu_ptr(&perf_cgroup_events)))
2850
		perf_cgroup_sched_in(prev, task);
2851

2852 2853
	if (__this_cpu_read(perf_sched_cb_usages))
		perf_pmu_sched_task(prev, task, true);
2854 2855
}

2856 2857 2858 2859 2860 2861 2862 2863 2864 2865 2866 2867 2868 2869 2870 2871 2872 2873 2874 2875 2876 2877 2878 2879 2880 2881 2882
static u64 perf_calculate_period(struct perf_event *event, u64 nsec, u64 count)
{
	u64 frequency = event->attr.sample_freq;
	u64 sec = NSEC_PER_SEC;
	u64 divisor, dividend;

	int count_fls, nsec_fls, frequency_fls, sec_fls;

	count_fls = fls64(count);
	nsec_fls = fls64(nsec);
	frequency_fls = fls64(frequency);
	sec_fls = 30;

	/*
	 * We got @count in @nsec, with a target of sample_freq HZ
	 * the target period becomes:
	 *
	 *             @count * 10^9
	 * period = -------------------
	 *          @nsec * sample_freq
	 *
	 */

	/*
	 * Reduce accuracy by one bit such that @a and @b converge
	 * to a similar magnitude.
	 */
2883
#define REDUCE_FLS(a, b)		\
2884 2885 2886 2887 2888 2889 2890 2891 2892 2893 2894 2895 2896 2897 2898 2899 2900 2901 2902 2903 2904 2905 2906 2907 2908 2909 2910 2911 2912 2913 2914 2915 2916 2917 2918 2919 2920 2921 2922
do {					\
	if (a##_fls > b##_fls) {	\
		a >>= 1;		\
		a##_fls--;		\
	} else {			\
		b >>= 1;		\
		b##_fls--;		\
	}				\
} while (0)

	/*
	 * Reduce accuracy until either term fits in a u64, then proceed with
	 * the other, so that finally we can do a u64/u64 division.
	 */
	while (count_fls + sec_fls > 64 && nsec_fls + frequency_fls > 64) {
		REDUCE_FLS(nsec, frequency);
		REDUCE_FLS(sec, count);
	}

	if (count_fls + sec_fls > 64) {
		divisor = nsec * frequency;

		while (count_fls + sec_fls > 64) {
			REDUCE_FLS(count, sec);
			divisor >>= 1;
		}

		dividend = count * sec;
	} else {
		dividend = count * sec;

		while (nsec_fls + frequency_fls > 64) {
			REDUCE_FLS(nsec, frequency);
			dividend >>= 1;
		}

		divisor = nsec * frequency;
	}

2923 2924 2925
	if (!divisor)
		return dividend;

2926 2927 2928
	return div64_u64(dividend, divisor);
}

2929 2930 2931
static DEFINE_PER_CPU(int, perf_throttled_count);
static DEFINE_PER_CPU(u64, perf_throttled_seq);

2932
static void perf_adjust_period(struct perf_event *event, u64 nsec, u64 count, bool disable)
2933
{
2934
	struct hw_perf_event *hwc = &event->hw;
2935
	s64 period, sample_period;
2936 2937
	s64 delta;

2938
	period = perf_calculate_period(event, nsec, count);
2939 2940 2941 2942 2943 2944 2945 2946 2947 2948

	delta = (s64)(period - hwc->sample_period);
	delta = (delta + 7) / 8; /* low pass filter */

	sample_period = hwc->sample_period + delta;

	if (!sample_period)
		sample_period = 1;

	hwc->sample_period = sample_period;
2949

2950
	if (local64_read(&hwc->period_left) > 8*sample_period) {
2951 2952 2953
		if (disable)
			event->pmu->stop(event, PERF_EF_UPDATE);

2954
		local64_set(&hwc->period_left, 0);
2955 2956 2957

		if (disable)
			event->pmu->start(event, PERF_EF_RELOAD);
2958
	}
2959 2960
}

2961 2962 2963 2964 2965 2966 2967
/*
 * combine freq adjustment with unthrottling to avoid two passes over the
 * events. At the same time, make sure, having freq events does not change
 * the rate of unthrottling as that would introduce bias.
 */
static void perf_adjust_freq_unthr_context(struct perf_event_context *ctx,
					   int needs_unthr)
2968
{
2969 2970
	struct perf_event *event;
	struct hw_perf_event *hwc;
2971
	u64 now, period = TICK_NSEC;
2972
	s64 delta;
2973

2974 2975 2976 2977 2978 2979
	/*
	 * only need to iterate over all events iff:
	 * - context have events in frequency mode (needs freq adjust)
	 * - there are events to unthrottle on this cpu
	 */
	if (!(ctx->nr_freq || needs_unthr))
2980 2981
		return;

2982
	raw_spin_lock(&ctx->lock);
2983
	perf_pmu_disable(ctx->pmu);
2984

2985
	list_for_each_entry_rcu(event, &ctx->event_list, event_entry) {
2986
		if (event->state != PERF_EVENT_STATE_ACTIVE)
2987 2988
			continue;

2989
		if (!event_filter_match(event))
2990 2991
			continue;

2992 2993
		perf_pmu_disable(event->pmu);

2994
		hwc = &event->hw;
2995

2996
		if (hwc->interrupts == MAX_INTERRUPTS) {
2997
			hwc->interrupts = 0;
2998
			perf_log_throttle(event, 1);
P
Peter Zijlstra 已提交
2999
			event->pmu->start(event, 0);
3000 3001
		}

3002
		if (!event->attr.freq || !event->attr.sample_freq)
3003
			goto next;
3004

3005 3006 3007 3008 3009
		/*
		 * stop the event and update event->count
		 */
		event->pmu->stop(event, PERF_EF_UPDATE);

3010
		now = local64_read(&event->count);
3011 3012
		delta = now - hwc->freq_count_stamp;
		hwc->freq_count_stamp = now;
3013

3014 3015 3016
		/*
		 * restart the event
		 * reload only if value has changed
3017 3018 3019
		 * we have stopped the event so tell that
		 * to perf_adjust_period() to avoid stopping it
		 * twice.
3020
		 */
3021
		if (delta > 0)
3022
			perf_adjust_period(event, period, delta, false);
3023 3024

		event->pmu->start(event, delta > 0 ? PERF_EF_RELOAD : 0);
3025 3026
	next:
		perf_pmu_enable(event->pmu);
3027
	}
3028

3029
	perf_pmu_enable(ctx->pmu);
3030
	raw_spin_unlock(&ctx->lock);
3031 3032
}

3033
/*
3034
 * Round-robin a context's events:
3035
 */
3036
static void rotate_ctx(struct perf_event_context *ctx)
T
Thomas Gleixner 已提交
3037
{
3038 3039 3040 3041 3042 3043
	/*
	 * Rotate the first entry last of non-pinned groups. Rotation might be
	 * disabled by the inheritance code.
	 */
	if (!ctx->rotate_disable)
		list_rotate_left(&ctx->flexible_groups);
3044 3045
}

3046
static int perf_rotate_context(struct perf_cpu_context *cpuctx)
3047
{
P
Peter Zijlstra 已提交
3048
	struct perf_event_context *ctx = NULL;
3049
	int rotate = 0;
3050

3051 3052 3053 3054
	if (cpuctx->ctx.nr_events) {
		if (cpuctx->ctx.nr_events != cpuctx->ctx.nr_active)
			rotate = 1;
	}
3055

P
Peter Zijlstra 已提交
3056
	ctx = cpuctx->task_ctx;
3057 3058 3059 3060
	if (ctx && ctx->nr_events) {
		if (ctx->nr_events != ctx->nr_active)
			rotate = 1;
	}
3061

3062
	if (!rotate)
3063 3064
		goto done;

3065
	perf_ctx_lock(cpuctx, cpuctx->task_ctx);
P
Peter Zijlstra 已提交
3066
	perf_pmu_disable(cpuctx->ctx.pmu);
3067

3068 3069 3070
	cpu_ctx_sched_out(cpuctx, EVENT_FLEXIBLE);
	if (ctx)
		ctx_sched_out(ctx, cpuctx, EVENT_FLEXIBLE);
T
Thomas Gleixner 已提交
3071

3072 3073 3074
	rotate_ctx(&cpuctx->ctx);
	if (ctx)
		rotate_ctx(ctx);
3075

3076
	perf_event_sched_in(cpuctx, ctx, current);
3077

3078 3079
	perf_pmu_enable(cpuctx->ctx.pmu);
	perf_ctx_unlock(cpuctx, cpuctx->task_ctx);
3080
done:
3081 3082

	return rotate;
3083 3084
}

3085 3086 3087
#ifdef CONFIG_NO_HZ_FULL
bool perf_event_can_stop_tick(void)
{
3088
	if (atomic_read(&nr_freq_events) ||
3089
	    __this_cpu_read(perf_throttled_count))
3090
		return false;
3091 3092
	else
		return true;
3093 3094 3095
}
#endif

3096 3097
void perf_event_task_tick(void)
{
3098 3099
	struct list_head *head = this_cpu_ptr(&active_ctx_list);
	struct perf_event_context *ctx, *tmp;
3100
	int throttled;
3101

3102 3103
	WARN_ON(!irqs_disabled());

3104 3105 3106
	__this_cpu_inc(perf_throttled_seq);
	throttled = __this_cpu_xchg(perf_throttled_count, 0);

3107
	list_for_each_entry_safe(ctx, tmp, head, active_ctx_list)
3108
		perf_adjust_freq_unthr_context(ctx, throttled);
T
Thomas Gleixner 已提交
3109 3110
}

3111 3112 3113 3114 3115 3116 3117 3118 3119 3120
static int event_enable_on_exec(struct perf_event *event,
				struct perf_event_context *ctx)
{
	if (!event->attr.enable_on_exec)
		return 0;

	event->attr.enable_on_exec = 0;
	if (event->state >= PERF_EVENT_STATE_INACTIVE)
		return 0;

3121
	__perf_event_mark_enabled(event);
3122 3123 3124 3125

	return 1;
}

3126
/*
3127
 * Enable all of a task's events that have been marked enable-on-exec.
3128 3129
 * This expects task == current.
 */
P
Peter Zijlstra 已提交
3130
static void perf_event_enable_on_exec(struct perf_event_context *ctx)
3131
{
3132
	struct perf_event_context *clone_ctx = NULL;
3133
	struct perf_event *event;
3134 3135
	unsigned long flags;
	int enabled = 0;
3136
	int ret;
3137 3138

	local_irq_save(flags);
3139
	if (!ctx || !ctx->nr_events)
3140 3141
		goto out;

3142 3143 3144 3145 3146 3147 3148
	/*
	 * We must ctxsw out cgroup events to avoid conflict
	 * when invoking perf_task_event_sched_in() later on
	 * in this function. Otherwise we end up trying to
	 * ctxswin cgroup events which are already scheduled
	 * in.
	 */
3149
	perf_cgroup_sched_out(current, NULL);
3150

3151
	raw_spin_lock(&ctx->lock);
3152
	task_ctx_sched_out(ctx);
3153

3154
	list_for_each_entry(event, &ctx->event_list, event_entry) {
3155 3156 3157
		ret = event_enable_on_exec(event, ctx);
		if (ret)
			enabled = 1;
3158 3159 3160
	}

	/*
3161
	 * Unclone this context if we enabled any event.
3162
	 */
3163
	if (enabled)
3164
		clone_ctx = unclone_ctx(ctx);
3165

3166
	raw_spin_unlock(&ctx->lock);
3167

3168 3169 3170
	/*
	 * Also calls ctxswin for cgroup events, if any:
	 */
S
Stephane Eranian 已提交
3171
	perf_event_context_sched_in(ctx, ctx->task);
P
Peter Zijlstra 已提交
3172
out:
3173
	local_irq_restore(flags);
3174 3175 3176

	if (clone_ctx)
		put_ctx(clone_ctx);
3177 3178
}

3179 3180 3181 3182 3183 3184 3185 3186 3187 3188 3189 3190 3191 3192 3193 3194
void perf_event_exec(void)
{
	struct perf_event_context *ctx;
	int ctxn;

	rcu_read_lock();
	for_each_task_context_nr(ctxn) {
		ctx = current->perf_event_ctxp[ctxn];
		if (!ctx)
			continue;

		perf_event_enable_on_exec(ctx);
	}
	rcu_read_unlock();
}

T
Thomas Gleixner 已提交
3195
/*
3196
 * Cross CPU call to read the hardware event
T
Thomas Gleixner 已提交
3197
 */
3198
static void __perf_event_read(void *info)
T
Thomas Gleixner 已提交
3199
{
3200 3201
	struct perf_event *event = info;
	struct perf_event_context *ctx = event->ctx;
P
Peter Zijlstra 已提交
3202
	struct perf_cpu_context *cpuctx = __get_cpu_context(ctx);
I
Ingo Molnar 已提交
3203

3204 3205 3206 3207
	/*
	 * If this is a task context, we need to check whether it is
	 * the current task context of this cpu.  If not it has been
	 * scheduled out before the smp call arrived.  In that case
3208 3209
	 * event->count would have been updated to a recent sample
	 * when the event was scheduled out.
3210 3211 3212 3213
	 */
	if (ctx->task && cpuctx->task_ctx != ctx)
		return;

3214
	raw_spin_lock(&ctx->lock);
S
Stephane Eranian 已提交
3215
	if (ctx->is_active) {
3216
		update_context_time(ctx);
S
Stephane Eranian 已提交
3217 3218
		update_cgrp_time_from_event(event);
	}
3219
	update_event_times(event);
3220 3221
	if (event->state == PERF_EVENT_STATE_ACTIVE)
		event->pmu->read(event);
3222
	raw_spin_unlock(&ctx->lock);
T
Thomas Gleixner 已提交
3223 3224
}

P
Peter Zijlstra 已提交
3225 3226
static inline u64 perf_event_count(struct perf_event *event)
{
3227 3228 3229 3230
	if (event->pmu->count)
		return event->pmu->count(event);

	return __perf_event_count(event);
P
Peter Zijlstra 已提交
3231 3232
}

3233
static u64 perf_event_read(struct perf_event *event)
T
Thomas Gleixner 已提交
3234 3235
{
	/*
3236 3237
	 * If event is enabled and currently active on a CPU, update the
	 * value in the event structure:
T
Thomas Gleixner 已提交
3238
	 */
3239 3240 3241 3242
	if (event->state == PERF_EVENT_STATE_ACTIVE) {
		smp_call_function_single(event->oncpu,
					 __perf_event_read, event, 1);
	} else if (event->state == PERF_EVENT_STATE_INACTIVE) {
P
Peter Zijlstra 已提交
3243 3244 3245
		struct perf_event_context *ctx = event->ctx;
		unsigned long flags;

3246
		raw_spin_lock_irqsave(&ctx->lock, flags);
3247 3248 3249 3250 3251
		/*
		 * may read while context is not active
		 * (e.g., thread is blocked), in that case
		 * we cannot update context time
		 */
S
Stephane Eranian 已提交
3252
		if (ctx->is_active) {
3253
			update_context_time(ctx);
S
Stephane Eranian 已提交
3254 3255
			update_cgrp_time_from_event(event);
		}
3256
		update_event_times(event);
3257
		raw_spin_unlock_irqrestore(&ctx->lock, flags);
T
Thomas Gleixner 已提交
3258 3259
	}

P
Peter Zijlstra 已提交
3260
	return perf_event_count(event);
T
Thomas Gleixner 已提交
3261 3262
}

3263
/*
3264
 * Initialize the perf_event context in a task_struct:
3265
 */
3266
static void __perf_event_init_context(struct perf_event_context *ctx)
3267
{
3268
	raw_spin_lock_init(&ctx->lock);
3269
	mutex_init(&ctx->mutex);
3270
	INIT_LIST_HEAD(&ctx->active_ctx_list);
3271 3272
	INIT_LIST_HEAD(&ctx->pinned_groups);
	INIT_LIST_HEAD(&ctx->flexible_groups);
3273 3274
	INIT_LIST_HEAD(&ctx->event_list);
	atomic_set(&ctx->refcount, 1);
3275
	INIT_DELAYED_WORK(&ctx->orphans_remove, orphans_remove_work);
3276 3277 3278 3279 3280 3281 3282 3283 3284 3285 3286 3287 3288 3289 3290
}

static struct perf_event_context *
alloc_perf_context(struct pmu *pmu, struct task_struct *task)
{
	struct perf_event_context *ctx;

	ctx = kzalloc(sizeof(struct perf_event_context), GFP_KERNEL);
	if (!ctx)
		return NULL;

	__perf_event_init_context(ctx);
	if (task) {
		ctx->task = task;
		get_task_struct(task);
T
Thomas Gleixner 已提交
3291
	}
3292 3293 3294
	ctx->pmu = pmu;

	return ctx;
3295 3296
}

3297 3298 3299 3300 3301
static struct task_struct *
find_lively_task_by_vpid(pid_t vpid)
{
	struct task_struct *task;
	int err;
T
Thomas Gleixner 已提交
3302 3303

	rcu_read_lock();
3304
	if (!vpid)
T
Thomas Gleixner 已提交
3305 3306
		task = current;
	else
3307
		task = find_task_by_vpid(vpid);
T
Thomas Gleixner 已提交
3308 3309 3310 3311 3312 3313 3314 3315
	if (task)
		get_task_struct(task);
	rcu_read_unlock();

	if (!task)
		return ERR_PTR(-ESRCH);

	/* Reuse ptrace permission checks for now. */
3316 3317 3318 3319
	err = -EACCES;
	if (!ptrace_may_access(task, PTRACE_MODE_READ))
		goto errout;

3320 3321 3322 3323 3324 3325 3326
	return task;
errout:
	put_task_struct(task);
	return ERR_PTR(err);

}

3327 3328 3329
/*
 * Returns a matching context with refcount and pincount.
 */
P
Peter Zijlstra 已提交
3330
static struct perf_event_context *
3331 3332
find_get_context(struct pmu *pmu, struct task_struct *task,
		struct perf_event *event)
T
Thomas Gleixner 已提交
3333
{
3334
	struct perf_event_context *ctx, *clone_ctx = NULL;
3335
	struct perf_cpu_context *cpuctx;
3336
	void *task_ctx_data = NULL;
3337
	unsigned long flags;
P
Peter Zijlstra 已提交
3338
	int ctxn, err;
3339
	int cpu = event->cpu;
T
Thomas Gleixner 已提交
3340

3341
	if (!task) {
3342
		/* Must be root to operate on a CPU event: */
3343
		if (perf_paranoid_cpu() && !capable(CAP_SYS_ADMIN))
T
Thomas Gleixner 已提交
3344 3345 3346
			return ERR_PTR(-EACCES);

		/*
3347
		 * We could be clever and allow to attach a event to an
T
Thomas Gleixner 已提交
3348 3349 3350
		 * offline CPU and activate it when the CPU comes up, but
		 * that's for later.
		 */
3351
		if (!cpu_online(cpu))
T
Thomas Gleixner 已提交
3352 3353
			return ERR_PTR(-ENODEV);

P
Peter Zijlstra 已提交
3354
		cpuctx = per_cpu_ptr(pmu->pmu_cpu_context, cpu);
T
Thomas Gleixner 已提交
3355
		ctx = &cpuctx->ctx;
3356
		get_ctx(ctx);
3357
		++ctx->pin_count;
T
Thomas Gleixner 已提交
3358 3359 3360 3361

		return ctx;
	}

P
Peter Zijlstra 已提交
3362 3363 3364 3365 3366
	err = -EINVAL;
	ctxn = pmu->task_ctx_nr;
	if (ctxn < 0)
		goto errout;

3367 3368 3369 3370 3371 3372 3373 3374
	if (event->attach_state & PERF_ATTACH_TASK_DATA) {
		task_ctx_data = kzalloc(pmu->task_ctx_size, GFP_KERNEL);
		if (!task_ctx_data) {
			err = -ENOMEM;
			goto errout;
		}
	}

P
Peter Zijlstra 已提交
3375
retry:
P
Peter Zijlstra 已提交
3376
	ctx = perf_lock_task_context(task, ctxn, &flags);
3377
	if (ctx) {
3378
		clone_ctx = unclone_ctx(ctx);
3379
		++ctx->pin_count;
3380 3381 3382 3383 3384

		if (task_ctx_data && !ctx->task_ctx_data) {
			ctx->task_ctx_data = task_ctx_data;
			task_ctx_data = NULL;
		}
3385
		raw_spin_unlock_irqrestore(&ctx->lock, flags);
3386 3387 3388

		if (clone_ctx)
			put_ctx(clone_ctx);
3389
	} else {
3390
		ctx = alloc_perf_context(pmu, task);
3391 3392 3393
		err = -ENOMEM;
		if (!ctx)
			goto errout;
3394

3395 3396 3397 3398 3399
		if (task_ctx_data) {
			ctx->task_ctx_data = task_ctx_data;
			task_ctx_data = NULL;
		}

3400 3401 3402 3403 3404 3405 3406 3407 3408 3409
		err = 0;
		mutex_lock(&task->perf_event_mutex);
		/*
		 * If it has already passed perf_event_exit_task().
		 * we must see PF_EXITING, it takes this mutex too.
		 */
		if (task->flags & PF_EXITING)
			err = -ESRCH;
		else if (task->perf_event_ctxp[ctxn])
			err = -EAGAIN;
3410
		else {
3411
			get_ctx(ctx);
3412
			++ctx->pin_count;
3413
			rcu_assign_pointer(task->perf_event_ctxp[ctxn], ctx);
3414
		}
3415 3416 3417
		mutex_unlock(&task->perf_event_mutex);

		if (unlikely(err)) {
3418
			put_ctx(ctx);
3419 3420 3421 3422

			if (err == -EAGAIN)
				goto retry;
			goto errout;
3423 3424 3425
		}
	}

3426
	kfree(task_ctx_data);
T
Thomas Gleixner 已提交
3427
	return ctx;
3428

P
Peter Zijlstra 已提交
3429
errout:
3430
	kfree(task_ctx_data);
3431
	return ERR_PTR(err);
T
Thomas Gleixner 已提交
3432 3433
}

L
Li Zefan 已提交
3434
static void perf_event_free_filter(struct perf_event *event);
3435
static void perf_event_free_bpf_prog(struct perf_event *event);
L
Li Zefan 已提交
3436

3437
static void free_event_rcu(struct rcu_head *head)
P
Peter Zijlstra 已提交
3438
{
3439
	struct perf_event *event;
P
Peter Zijlstra 已提交
3440

3441 3442 3443
	event = container_of(head, struct perf_event, rcu_head);
	if (event->ns)
		put_pid_ns(event->ns);
L
Li Zefan 已提交
3444
	perf_event_free_filter(event);
3445
	perf_event_free_bpf_prog(event);
3446
	kfree(event);
P
Peter Zijlstra 已提交
3447 3448
}

3449 3450
static void ring_buffer_attach(struct perf_event *event,
			       struct ring_buffer *rb);
3451

3452
static void unaccount_event_cpu(struct perf_event *event, int cpu)
3453
{
3454 3455 3456 3457 3458 3459
	if (event->parent)
		return;

	if (is_cgroup_event(event))
		atomic_dec(&per_cpu(perf_cgroup_events, cpu));
}
3460

3461 3462 3463 3464 3465 3466 3467 3468 3469 3470 3471 3472 3473
static void unaccount_event(struct perf_event *event)
{
	if (event->parent)
		return;

	if (event->attach_state & PERF_ATTACH_TASK)
		static_key_slow_dec_deferred(&perf_sched_events);
	if (event->attr.mmap || event->attr.mmap_data)
		atomic_dec(&nr_mmap_events);
	if (event->attr.comm)
		atomic_dec(&nr_comm_events);
	if (event->attr.task)
		atomic_dec(&nr_task_events);
3474 3475
	if (event->attr.freq)
		atomic_dec(&nr_freq_events);
3476 3477 3478 3479 3480 3481 3482
	if (is_cgroup_event(event))
		static_key_slow_dec_deferred(&perf_sched_events);
	if (has_branch_stack(event))
		static_key_slow_dec_deferred(&perf_sched_events);

	unaccount_event_cpu(event, event->cpu);
}
3483

3484 3485 3486 3487 3488 3489 3490 3491 3492 3493 3494 3495 3496 3497 3498 3499 3500 3501 3502 3503 3504 3505 3506 3507 3508 3509 3510 3511 3512 3513 3514 3515 3516 3517 3518 3519 3520 3521 3522 3523 3524 3525 3526 3527 3528 3529 3530 3531 3532 3533 3534 3535 3536 3537 3538 3539 3540 3541 3542 3543 3544 3545 3546 3547 3548 3549 3550 3551 3552 3553 3554 3555 3556 3557 3558 3559 3560 3561 3562 3563 3564 3565 3566 3567 3568
/*
 * The following implement mutual exclusion of events on "exclusive" pmus
 * (PERF_PMU_CAP_EXCLUSIVE). Such pmus can only have one event scheduled
 * at a time, so we disallow creating events that might conflict, namely:
 *
 *  1) cpu-wide events in the presence of per-task events,
 *  2) per-task events in the presence of cpu-wide events,
 *  3) two matching events on the same context.
 *
 * The former two cases are handled in the allocation path (perf_event_alloc(),
 * __free_event()), the latter -- before the first perf_install_in_context().
 */
static int exclusive_event_init(struct perf_event *event)
{
	struct pmu *pmu = event->pmu;

	if (!(pmu->capabilities & PERF_PMU_CAP_EXCLUSIVE))
		return 0;

	/*
	 * Prevent co-existence of per-task and cpu-wide events on the
	 * same exclusive pmu.
	 *
	 * Negative pmu::exclusive_cnt means there are cpu-wide
	 * events on this "exclusive" pmu, positive means there are
	 * per-task events.
	 *
	 * Since this is called in perf_event_alloc() path, event::ctx
	 * doesn't exist yet; it is, however, safe to use PERF_ATTACH_TASK
	 * to mean "per-task event", because unlike other attach states it
	 * never gets cleared.
	 */
	if (event->attach_state & PERF_ATTACH_TASK) {
		if (!atomic_inc_unless_negative(&pmu->exclusive_cnt))
			return -EBUSY;
	} else {
		if (!atomic_dec_unless_positive(&pmu->exclusive_cnt))
			return -EBUSY;
	}

	return 0;
}

static void exclusive_event_destroy(struct perf_event *event)
{
	struct pmu *pmu = event->pmu;

	if (!(pmu->capabilities & PERF_PMU_CAP_EXCLUSIVE))
		return;

	/* see comment in exclusive_event_init() */
	if (event->attach_state & PERF_ATTACH_TASK)
		atomic_dec(&pmu->exclusive_cnt);
	else
		atomic_inc(&pmu->exclusive_cnt);
}

static bool exclusive_event_match(struct perf_event *e1, struct perf_event *e2)
{
	if ((e1->pmu->capabilities & PERF_PMU_CAP_EXCLUSIVE) &&
	    (e1->cpu == e2->cpu ||
	     e1->cpu == -1 ||
	     e2->cpu == -1))
		return true;
	return false;
}

/* Called under the same ctx::mutex as perf_install_in_context() */
static bool exclusive_event_installable(struct perf_event *event,
					struct perf_event_context *ctx)
{
	struct perf_event *iter_event;
	struct pmu *pmu = event->pmu;

	if (!(pmu->capabilities & PERF_PMU_CAP_EXCLUSIVE))
		return true;

	list_for_each_entry(iter_event, &ctx->event_list, event_entry) {
		if (exclusive_event_match(iter_event, event))
			return false;
	}

	return true;
}

3569 3570
static void __free_event(struct perf_event *event)
{
3571
	if (!event->parent) {
3572 3573
		if (event->attr.sample_type & PERF_SAMPLE_CALLCHAIN)
			put_callchain_buffers();
3574
	}
3575

3576 3577 3578 3579 3580 3581
	if (event->destroy)
		event->destroy(event);

	if (event->ctx)
		put_ctx(event->ctx);

3582 3583
	if (event->pmu) {
		exclusive_event_destroy(event);
3584
		module_put(event->pmu->module);
3585
	}
3586

3587 3588
	call_rcu(&event->rcu_head, free_event_rcu);
}
P
Peter Zijlstra 已提交
3589 3590

static void _free_event(struct perf_event *event)
3591
{
3592
	irq_work_sync(&event->pending);
3593

3594
	unaccount_event(event);
3595

3596
	if (event->rb) {
3597 3598 3599 3600 3601 3602 3603
		/*
		 * Can happen when we close an event with re-directed output.
		 *
		 * Since we have a 0 refcount, perf_mmap_close() will skip
		 * over us; possibly making our ring_buffer_put() the last.
		 */
		mutex_lock(&event->mmap_mutex);
3604
		ring_buffer_attach(event, NULL);
3605
		mutex_unlock(&event->mmap_mutex);
3606 3607
	}

S
Stephane Eranian 已提交
3608 3609 3610
	if (is_cgroup_event(event))
		perf_detach_cgroup(event);

3611
	__free_event(event);
3612 3613
}

P
Peter Zijlstra 已提交
3614 3615 3616 3617 3618
/*
 * Used to free events which have a known refcount of 1, such as in error paths
 * where the event isn't exposed yet and inherited events.
 */
static void free_event(struct perf_event *event)
T
Thomas Gleixner 已提交
3619
{
P
Peter Zijlstra 已提交
3620 3621 3622 3623 3624 3625
	if (WARN(atomic_long_cmpxchg(&event->refcount, 1, 0) != 1,
				"unexpected event refcount: %ld; ptr=%p\n",
				atomic_long_read(&event->refcount), event)) {
		/* leak to avoid use-after-free */
		return;
	}
T
Thomas Gleixner 已提交
3626

P
Peter Zijlstra 已提交
3627
	_free_event(event);
T
Thomas Gleixner 已提交
3628 3629
}

3630
/*
3631
 * Remove user event from the owner task.
3632
 */
3633
static void perf_remove_from_owner(struct perf_event *event)
3634
{
P
Peter Zijlstra 已提交
3635
	struct task_struct *owner;
3636

P
Peter Zijlstra 已提交
3637 3638 3639 3640 3641 3642 3643 3644 3645 3646 3647 3648 3649 3650 3651 3652 3653 3654 3655 3656
	rcu_read_lock();
	owner = ACCESS_ONCE(event->owner);
	/*
	 * Matches the smp_wmb() in perf_event_exit_task(). If we observe
	 * !owner it means the list deletion is complete and we can indeed
	 * free this event, otherwise we need to serialize on
	 * owner->perf_event_mutex.
	 */
	smp_read_barrier_depends();
	if (owner) {
		/*
		 * Since delayed_put_task_struct() also drops the last
		 * task reference we can safely take a new reference
		 * while holding the rcu_read_lock().
		 */
		get_task_struct(owner);
	}
	rcu_read_unlock();

	if (owner) {
P
Peter Zijlstra 已提交
3657 3658 3659 3660 3661 3662 3663 3664 3665 3666
		/*
		 * If we're here through perf_event_exit_task() we're already
		 * holding ctx->mutex which would be an inversion wrt. the
		 * normal lock order.
		 *
		 * However we can safely take this lock because its the child
		 * ctx->mutex.
		 */
		mutex_lock_nested(&owner->perf_event_mutex, SINGLE_DEPTH_NESTING);

P
Peter Zijlstra 已提交
3667 3668 3669 3670 3671 3672 3673 3674 3675 3676 3677
		/*
		 * We have to re-check the event->owner field, if it is cleared
		 * we raced with perf_event_exit_task(), acquiring the mutex
		 * ensured they're done, and we can proceed with freeing the
		 * event.
		 */
		if (event->owner)
			list_del_init(&event->owner_entry);
		mutex_unlock(&owner->perf_event_mutex);
		put_task_struct(owner);
	}
3678 3679 3680 3681
}

static void put_event(struct perf_event *event)
{
P
Peter Zijlstra 已提交
3682
	struct perf_event_context *ctx;
3683 3684 3685 3686 3687 3688

	if (!atomic_long_dec_and_test(&event->refcount))
		return;

	if (!is_kernel_event(event))
		perf_remove_from_owner(event);
P
Peter Zijlstra 已提交
3689

P
Peter Zijlstra 已提交
3690 3691 3692 3693 3694 3695 3696 3697 3698 3699 3700 3701
	/*
	 * There are two ways this annotation is useful:
	 *
	 *  1) there is a lock recursion from perf_event_exit_task
	 *     see the comment there.
	 *
	 *  2) there is a lock-inversion with mmap_sem through
	 *     perf_event_read_group(), which takes faults while
	 *     holding ctx->mutex, however this is called after
	 *     the last filedesc died, so there is no possibility
	 *     to trigger the AB-BA case.
	 */
P
Peter Zijlstra 已提交
3702 3703
	ctx = perf_event_ctx_lock_nested(event, SINGLE_DEPTH_NESTING);
	WARN_ON_ONCE(ctx->parent_ctx);
P
Peter Zijlstra 已提交
3704
	perf_remove_from_context(event, true);
L
Leon Yu 已提交
3705
	perf_event_ctx_unlock(event, ctx);
P
Peter Zijlstra 已提交
3706 3707

	_free_event(event);
3708 3709
}

P
Peter Zijlstra 已提交
3710 3711 3712 3713 3714 3715 3716
int perf_event_release_kernel(struct perf_event *event)
{
	put_event(event);
	return 0;
}
EXPORT_SYMBOL_GPL(perf_event_release_kernel);

3717 3718 3719
/*
 * Called when the last reference to the file is gone.
 */
3720 3721 3722 3723
static int perf_release(struct inode *inode, struct file *file)
{
	put_event(file->private_data);
	return 0;
3724 3725
}

3726 3727 3728 3729 3730 3731 3732 3733 3734 3735 3736 3737 3738 3739 3740 3741 3742 3743 3744 3745 3746 3747 3748 3749 3750 3751 3752 3753 3754 3755 3756 3757 3758 3759 3760 3761
/*
 * Remove all orphanes events from the context.
 */
static void orphans_remove_work(struct work_struct *work)
{
	struct perf_event_context *ctx;
	struct perf_event *event, *tmp;

	ctx = container_of(work, struct perf_event_context,
			   orphans_remove.work);

	mutex_lock(&ctx->mutex);
	list_for_each_entry_safe(event, tmp, &ctx->event_list, event_entry) {
		struct perf_event *parent_event = event->parent;

		if (!is_orphaned_child(event))
			continue;

		perf_remove_from_context(event, true);

		mutex_lock(&parent_event->child_mutex);
		list_del_init(&event->child_list);
		mutex_unlock(&parent_event->child_mutex);

		free_event(event);
		put_event(parent_event);
	}

	raw_spin_lock_irq(&ctx->lock);
	ctx->orphans_remove_sched = false;
	raw_spin_unlock_irq(&ctx->lock);
	mutex_unlock(&ctx->mutex);

	put_ctx(ctx);
}

3762
u64 perf_event_read_value(struct perf_event *event, u64 *enabled, u64 *running)
3763
{
3764
	struct perf_event *child;
3765 3766
	u64 total = 0;

3767 3768 3769
	*enabled = 0;
	*running = 0;

3770
	mutex_lock(&event->child_mutex);
3771
	total += perf_event_read(event);
3772 3773 3774 3775 3776 3777
	*enabled += event->total_time_enabled +
			atomic64_read(&event->child_total_time_enabled);
	*running += event->total_time_running +
			atomic64_read(&event->child_total_time_running);

	list_for_each_entry(child, &event->child_list, child_list) {
3778
		total += perf_event_read(child);
3779 3780 3781
		*enabled += child->total_time_enabled;
		*running += child->total_time_running;
	}
3782
	mutex_unlock(&event->child_mutex);
3783 3784 3785

	return total;
}
3786
EXPORT_SYMBOL_GPL(perf_event_read_value);
3787

3788
static int perf_event_read_group(struct perf_event *event,
3789 3790
				   u64 read_format, char __user *buf)
{
3791
	struct perf_event *leader = event->group_leader, *sub;
3792
	struct perf_event_context *ctx = leader->ctx;
P
Peter Zijlstra 已提交
3793
	int n = 0, size = 0, ret;
3794
	u64 count, enabled, running;
P
Peter Zijlstra 已提交
3795 3796 3797
	u64 values[5];

	lockdep_assert_held(&ctx->mutex);
3798

3799
	count = perf_event_read_value(leader, &enabled, &running);
3800 3801

	values[n++] = 1 + leader->nr_siblings;
3802 3803 3804 3805
	if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
		values[n++] = enabled;
	if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
		values[n++] = running;
3806 3807 3808
	values[n++] = count;
	if (read_format & PERF_FORMAT_ID)
		values[n++] = primary_event_id(leader);
3809 3810 3811 3812

	size = n * sizeof(u64);

	if (copy_to_user(buf, values, size))
P
Peter Zijlstra 已提交
3813
		return -EFAULT;
3814

3815
	ret = size;
3816

3817
	list_for_each_entry(sub, &leader->sibling_list, group_entry) {
3818
		n = 0;
3819

3820
		values[n++] = perf_event_read_value(sub, &enabled, &running);
3821 3822 3823 3824 3825
		if (read_format & PERF_FORMAT_ID)
			values[n++] = primary_event_id(sub);

		size = n * sizeof(u64);

3826
		if (copy_to_user(buf + ret, values, size)) {
P
Peter Zijlstra 已提交
3827
			return -EFAULT;
3828
		}
3829 3830

		ret += size;
3831 3832
	}

3833
	return ret;
3834 3835
}

3836
static int perf_event_read_one(struct perf_event *event,
3837 3838
				 u64 read_format, char __user *buf)
{
3839
	u64 enabled, running;
3840 3841 3842
	u64 values[4];
	int n = 0;

3843 3844 3845 3846 3847
	values[n++] = perf_event_read_value(event, &enabled, &running);
	if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
		values[n++] = enabled;
	if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
		values[n++] = running;
3848
	if (read_format & PERF_FORMAT_ID)
3849
		values[n++] = primary_event_id(event);
3850 3851 3852 3853 3854 3855 3856

	if (copy_to_user(buf, values, n * sizeof(u64)))
		return -EFAULT;

	return n * sizeof(u64);
}

3857 3858 3859 3860 3861 3862 3863 3864 3865 3866 3867 3868 3869
static bool is_event_hup(struct perf_event *event)
{
	bool no_children;

	if (event->state != PERF_EVENT_STATE_EXIT)
		return false;

	mutex_lock(&event->child_mutex);
	no_children = list_empty(&event->child_list);
	mutex_unlock(&event->child_mutex);
	return no_children;
}

T
Thomas Gleixner 已提交
3870
/*
3871
 * Read the performance event - simple non blocking version for now
T
Thomas Gleixner 已提交
3872 3873
 */
static ssize_t
3874
perf_read_hw(struct perf_event *event, char __user *buf, size_t count)
T
Thomas Gleixner 已提交
3875
{
3876
	u64 read_format = event->attr.read_format;
3877
	int ret;
T
Thomas Gleixner 已提交
3878

3879
	/*
3880
	 * Return end-of-file for a read on a event that is in
3881 3882 3883
	 * error state (i.e. because it was pinned but it couldn't be
	 * scheduled on to the CPU at some point).
	 */
3884
	if (event->state == PERF_EVENT_STATE_ERROR)
3885 3886
		return 0;

3887
	if (count < event->read_size)
3888 3889
		return -ENOSPC;

3890
	WARN_ON_ONCE(event->ctx->parent_ctx);
3891
	if (read_format & PERF_FORMAT_GROUP)
3892
		ret = perf_event_read_group(event, read_format, buf);
3893
	else
3894
		ret = perf_event_read_one(event, read_format, buf);
T
Thomas Gleixner 已提交
3895

3896
	return ret;
T
Thomas Gleixner 已提交
3897 3898 3899 3900 3901
}

static ssize_t
perf_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
{
3902
	struct perf_event *event = file->private_data;
P
Peter Zijlstra 已提交
3903 3904
	struct perf_event_context *ctx;
	int ret;
T
Thomas Gleixner 已提交
3905

P
Peter Zijlstra 已提交
3906 3907 3908 3909 3910
	ctx = perf_event_ctx_lock(event);
	ret = perf_read_hw(event, buf, count);
	perf_event_ctx_unlock(event, ctx);

	return ret;
T
Thomas Gleixner 已提交
3911 3912 3913 3914
}

static unsigned int perf_poll(struct file *file, poll_table *wait)
{
3915
	struct perf_event *event = file->private_data;
3916
	struct ring_buffer *rb;
3917
	unsigned int events = POLLHUP;
P
Peter Zijlstra 已提交
3918

3919
	poll_wait(file, &event->waitq, wait);
3920

3921
	if (is_event_hup(event))
3922
		return events;
P
Peter Zijlstra 已提交
3923

3924
	/*
3925 3926
	 * Pin the event->rb by taking event->mmap_mutex; otherwise
	 * perf_event_set_output() can swizzle our rb and make us miss wakeups.
3927 3928
	 */
	mutex_lock(&event->mmap_mutex);
3929 3930
	rb = event->rb;
	if (rb)
3931
		events = atomic_xchg(&rb->poll, 0);
3932
	mutex_unlock(&event->mmap_mutex);
T
Thomas Gleixner 已提交
3933 3934 3935
	return events;
}

P
Peter Zijlstra 已提交
3936
static void _perf_event_reset(struct perf_event *event)
3937
{
3938
	(void)perf_event_read(event);
3939
	local64_set(&event->count, 0);
3940
	perf_event_update_userpage(event);
P
Peter Zijlstra 已提交
3941 3942
}

3943
/*
3944 3945 3946 3947
 * Holding the top-level event's child_mutex means that any
 * descendant process that has inherited this event will block
 * in sync_child_event if it goes to exit, thus satisfying the
 * task existence requirements of perf_event_enable/disable.
3948
 */
3949 3950
static void perf_event_for_each_child(struct perf_event *event,
					void (*func)(struct perf_event *))
P
Peter Zijlstra 已提交
3951
{
3952
	struct perf_event *child;
P
Peter Zijlstra 已提交
3953

3954
	WARN_ON_ONCE(event->ctx->parent_ctx);
P
Peter Zijlstra 已提交
3955

3956 3957 3958
	mutex_lock(&event->child_mutex);
	func(event);
	list_for_each_entry(child, &event->child_list, child_list)
P
Peter Zijlstra 已提交
3959
		func(child);
3960
	mutex_unlock(&event->child_mutex);
P
Peter Zijlstra 已提交
3961 3962
}

3963 3964
static void perf_event_for_each(struct perf_event *event,
				  void (*func)(struct perf_event *))
P
Peter Zijlstra 已提交
3965
{
3966 3967
	struct perf_event_context *ctx = event->ctx;
	struct perf_event *sibling;
P
Peter Zijlstra 已提交
3968

P
Peter Zijlstra 已提交
3969 3970
	lockdep_assert_held(&ctx->mutex);

3971
	event = event->group_leader;
3972

3973 3974
	perf_event_for_each_child(event, func);
	list_for_each_entry(sibling, &event->sibling_list, group_entry)
3975
		perf_event_for_each_child(sibling, func);
3976 3977
}

3978
static int perf_event_period(struct perf_event *event, u64 __user *arg)
3979
{
3980
	struct perf_event_context *ctx = event->ctx;
3981
	int ret = 0, active;
3982 3983
	u64 value;

3984
	if (!is_sampling_event(event))
3985 3986
		return -EINVAL;

3987
	if (copy_from_user(&value, arg, sizeof(value)))
3988 3989 3990 3991 3992
		return -EFAULT;

	if (!value)
		return -EINVAL;

3993
	raw_spin_lock_irq(&ctx->lock);
3994 3995
	if (event->attr.freq) {
		if (value > sysctl_perf_event_sample_rate) {
3996 3997 3998 3999
			ret = -EINVAL;
			goto unlock;
		}

4000
		event->attr.sample_freq = value;
4001
	} else {
4002 4003
		event->attr.sample_period = value;
		event->hw.sample_period = value;
4004
	}
4005 4006 4007 4008 4009 4010 4011 4012 4013 4014 4015 4016 4017 4018

	active = (event->state == PERF_EVENT_STATE_ACTIVE);
	if (active) {
		perf_pmu_disable(ctx->pmu);
		event->pmu->stop(event, PERF_EF_UPDATE);
	}

	local64_set(&event->hw.period_left, 0);

	if (active) {
		event->pmu->start(event, PERF_EF_RELOAD);
		perf_pmu_enable(ctx->pmu);
	}

4019
unlock:
4020
	raw_spin_unlock_irq(&ctx->lock);
4021 4022 4023 4024

	return ret;
}

4025 4026
static const struct file_operations perf_fops;

4027
static inline int perf_fget_light(int fd, struct fd *p)
4028
{
4029 4030 4031
	struct fd f = fdget(fd);
	if (!f.file)
		return -EBADF;
4032

4033 4034 4035
	if (f.file->f_op != &perf_fops) {
		fdput(f);
		return -EBADF;
4036
	}
4037 4038
	*p = f;
	return 0;
4039 4040 4041 4042
}

static int perf_event_set_output(struct perf_event *event,
				 struct perf_event *output_event);
L
Li Zefan 已提交
4043
static int perf_event_set_filter(struct perf_event *event, void __user *arg);
4044
static int perf_event_set_bpf_prog(struct perf_event *event, u32 prog_fd);
4045

P
Peter Zijlstra 已提交
4046
static long _perf_ioctl(struct perf_event *event, unsigned int cmd, unsigned long arg)
4047
{
4048
	void (*func)(struct perf_event *);
P
Peter Zijlstra 已提交
4049
	u32 flags = arg;
4050 4051

	switch (cmd) {
4052
	case PERF_EVENT_IOC_ENABLE:
P
Peter Zijlstra 已提交
4053
		func = _perf_event_enable;
4054
		break;
4055
	case PERF_EVENT_IOC_DISABLE:
P
Peter Zijlstra 已提交
4056
		func = _perf_event_disable;
4057
		break;
4058
	case PERF_EVENT_IOC_RESET:
P
Peter Zijlstra 已提交
4059
		func = _perf_event_reset;
4060
		break;
P
Peter Zijlstra 已提交
4061

4062
	case PERF_EVENT_IOC_REFRESH:
P
Peter Zijlstra 已提交
4063
		return _perf_event_refresh(event, arg);
4064

4065 4066
	case PERF_EVENT_IOC_PERIOD:
		return perf_event_period(event, (u64 __user *)arg);
4067

4068 4069 4070 4071 4072 4073 4074 4075 4076
	case PERF_EVENT_IOC_ID:
	{
		u64 id = primary_event_id(event);

		if (copy_to_user((void __user *)arg, &id, sizeof(id)))
			return -EFAULT;
		return 0;
	}

4077
	case PERF_EVENT_IOC_SET_OUTPUT:
4078 4079 4080
	{
		int ret;
		if (arg != -1) {
4081 4082 4083 4084 4085 4086 4087 4088 4089 4090
			struct perf_event *output_event;
			struct fd output;
			ret = perf_fget_light(arg, &output);
			if (ret)
				return ret;
			output_event = output.file->private_data;
			ret = perf_event_set_output(event, output_event);
			fdput(output);
		} else {
			ret = perf_event_set_output(event, NULL);
4091 4092 4093
		}
		return ret;
	}
4094

L
Li Zefan 已提交
4095 4096 4097
	case PERF_EVENT_IOC_SET_FILTER:
		return perf_event_set_filter(event, (void __user *)arg);

4098 4099 4100
	case PERF_EVENT_IOC_SET_BPF:
		return perf_event_set_bpf_prog(event, arg);

4101
	default:
P
Peter Zijlstra 已提交
4102
		return -ENOTTY;
4103
	}
P
Peter Zijlstra 已提交
4104 4105

	if (flags & PERF_IOC_FLAG_GROUP)
4106
		perf_event_for_each(event, func);
P
Peter Zijlstra 已提交
4107
	else
4108
		perf_event_for_each_child(event, func);
P
Peter Zijlstra 已提交
4109 4110

	return 0;
4111 4112
}

P
Peter Zijlstra 已提交
4113 4114 4115 4116 4117 4118 4119 4120 4121 4122 4123 4124 4125
static long perf_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
{
	struct perf_event *event = file->private_data;
	struct perf_event_context *ctx;
	long ret;

	ctx = perf_event_ctx_lock(event);
	ret = _perf_ioctl(event, cmd, arg);
	perf_event_ctx_unlock(event, ctx);

	return ret;
}

P
Pawel Moll 已提交
4126 4127 4128 4129 4130 4131 4132 4133 4134 4135 4136 4137 4138 4139 4140 4141 4142 4143 4144 4145
#ifdef CONFIG_COMPAT
static long perf_compat_ioctl(struct file *file, unsigned int cmd,
				unsigned long arg)
{
	switch (_IOC_NR(cmd)) {
	case _IOC_NR(PERF_EVENT_IOC_SET_FILTER):
	case _IOC_NR(PERF_EVENT_IOC_ID):
		/* Fix up pointer size (usually 4 -> 8 in 32-on-64-bit case */
		if (_IOC_SIZE(cmd) == sizeof(compat_uptr_t)) {
			cmd &= ~IOCSIZE_MASK;
			cmd |= sizeof(void *) << IOCSIZE_SHIFT;
		}
		break;
	}
	return perf_ioctl(file, cmd, arg);
}
#else
# define perf_compat_ioctl NULL
#endif

4146
int perf_event_task_enable(void)
4147
{
P
Peter Zijlstra 已提交
4148
	struct perf_event_context *ctx;
4149
	struct perf_event *event;
4150

4151
	mutex_lock(&current->perf_event_mutex);
P
Peter Zijlstra 已提交
4152 4153 4154 4155 4156
	list_for_each_entry(event, &current->perf_event_list, owner_entry) {
		ctx = perf_event_ctx_lock(event);
		perf_event_for_each_child(event, _perf_event_enable);
		perf_event_ctx_unlock(event, ctx);
	}
4157
	mutex_unlock(&current->perf_event_mutex);
4158 4159 4160 4161

	return 0;
}

4162
int perf_event_task_disable(void)
4163
{
P
Peter Zijlstra 已提交
4164
	struct perf_event_context *ctx;
4165
	struct perf_event *event;
4166

4167
	mutex_lock(&current->perf_event_mutex);
P
Peter Zijlstra 已提交
4168 4169 4170 4171 4172
	list_for_each_entry(event, &current->perf_event_list, owner_entry) {
		ctx = perf_event_ctx_lock(event);
		perf_event_for_each_child(event, _perf_event_disable);
		perf_event_ctx_unlock(event, ctx);
	}
4173
	mutex_unlock(&current->perf_event_mutex);
4174 4175 4176 4177

	return 0;
}

4178
static int perf_event_index(struct perf_event *event)
4179
{
P
Peter Zijlstra 已提交
4180 4181 4182
	if (event->hw.state & PERF_HES_STOPPED)
		return 0;

4183
	if (event->state != PERF_EVENT_STATE_ACTIVE)
4184 4185
		return 0;

4186
	return event->pmu->event_idx(event);
4187 4188
}

4189
static void calc_timer_values(struct perf_event *event,
4190
				u64 *now,
4191 4192
				u64 *enabled,
				u64 *running)
4193
{
4194
	u64 ctx_time;
4195

4196 4197
	*now = perf_clock();
	ctx_time = event->shadow_ctx_time + *now;
4198 4199 4200 4201
	*enabled = ctx_time - event->tstamp_enabled;
	*running = ctx_time - event->tstamp_running;
}

4202 4203 4204 4205 4206 4207 4208 4209 4210 4211 4212 4213 4214 4215 4216
static void perf_event_init_userpage(struct perf_event *event)
{
	struct perf_event_mmap_page *userpg;
	struct ring_buffer *rb;

	rcu_read_lock();
	rb = rcu_dereference(event->rb);
	if (!rb)
		goto unlock;

	userpg = rb->user_page;

	/* Allow new userspace to detect that bit 0 is deprecated */
	userpg->cap_bit0_is_deprecated = 1;
	userpg->size = offsetof(struct perf_event_mmap_page, __reserved);
4217 4218
	userpg->data_offset = PAGE_SIZE;
	userpg->data_size = perf_data_size(rb);
4219 4220 4221 4222 4223

unlock:
	rcu_read_unlock();
}

4224 4225
void __weak arch_perf_update_userpage(
	struct perf_event *event, struct perf_event_mmap_page *userpg, u64 now)
4226 4227 4228
{
}

4229 4230 4231 4232 4233
/*
 * Callers need to ensure there can be no nesting of this function, otherwise
 * the seqlock logic goes bad. We can not serialize this because the arch
 * code calls this from NMI context.
 */
4234
void perf_event_update_userpage(struct perf_event *event)
4235
{
4236
	struct perf_event_mmap_page *userpg;
4237
	struct ring_buffer *rb;
4238
	u64 enabled, running, now;
4239 4240

	rcu_read_lock();
4241 4242 4243 4244
	rb = rcu_dereference(event->rb);
	if (!rb)
		goto unlock;

4245 4246 4247 4248 4249 4250 4251 4252 4253
	/*
	 * compute total_time_enabled, total_time_running
	 * based on snapshot values taken when the event
	 * was last scheduled in.
	 *
	 * we cannot simply called update_context_time()
	 * because of locking issue as we can be called in
	 * NMI context
	 */
4254
	calc_timer_values(event, &now, &enabled, &running);
4255

4256
	userpg = rb->user_page;
4257 4258 4259 4260 4261
	/*
	 * Disable preemption so as to not let the corresponding user-space
	 * spin too long if we get preempted.
	 */
	preempt_disable();
4262
	++userpg->lock;
4263
	barrier();
4264
	userpg->index = perf_event_index(event);
P
Peter Zijlstra 已提交
4265
	userpg->offset = perf_event_count(event);
4266
	if (userpg->index)
4267
		userpg->offset -= local64_read(&event->hw.prev_count);
4268

4269
	userpg->time_enabled = enabled +
4270
			atomic64_read(&event->child_total_time_enabled);
4271

4272
	userpg->time_running = running +
4273
			atomic64_read(&event->child_total_time_running);
4274

4275
	arch_perf_update_userpage(event, userpg, now);
4276

4277
	barrier();
4278
	++userpg->lock;
4279
	preempt_enable();
4280
unlock:
4281
	rcu_read_unlock();
4282 4283
}

4284 4285 4286
static int perf_mmap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
{
	struct perf_event *event = vma->vm_file->private_data;
4287
	struct ring_buffer *rb;
4288 4289 4290 4291 4292 4293 4294 4295 4296
	int ret = VM_FAULT_SIGBUS;

	if (vmf->flags & FAULT_FLAG_MKWRITE) {
		if (vmf->pgoff == 0)
			ret = 0;
		return ret;
	}

	rcu_read_lock();
4297 4298
	rb = rcu_dereference(event->rb);
	if (!rb)
4299 4300 4301 4302 4303
		goto unlock;

	if (vmf->pgoff && (vmf->flags & FAULT_FLAG_WRITE))
		goto unlock;

4304
	vmf->page = perf_mmap_to_page(rb, vmf->pgoff);
4305 4306 4307 4308 4309 4310 4311 4312 4313 4314 4315 4316 4317 4318
	if (!vmf->page)
		goto unlock;

	get_page(vmf->page);
	vmf->page->mapping = vma->vm_file->f_mapping;
	vmf->page->index   = vmf->pgoff;

	ret = 0;
unlock:
	rcu_read_unlock();

	return ret;
}

4319 4320 4321
static void ring_buffer_attach(struct perf_event *event,
			       struct ring_buffer *rb)
{
4322
	struct ring_buffer *old_rb = NULL;
4323 4324
	unsigned long flags;

4325 4326 4327 4328 4329 4330
	if (event->rb) {
		/*
		 * Should be impossible, we set this when removing
		 * event->rb_entry and wait/clear when adding event->rb_entry.
		 */
		WARN_ON_ONCE(event->rcu_pending);
4331

4332 4333 4334
		old_rb = event->rb;
		event->rcu_batches = get_state_synchronize_rcu();
		event->rcu_pending = 1;
4335

4336 4337 4338 4339
		spin_lock_irqsave(&old_rb->event_lock, flags);
		list_del_rcu(&event->rb_entry);
		spin_unlock_irqrestore(&old_rb->event_lock, flags);
	}
4340

4341 4342 4343 4344
	if (event->rcu_pending && rb) {
		cond_synchronize_rcu(event->rcu_batches);
		event->rcu_pending = 0;
	}
4345

4346 4347 4348 4349 4350 4351 4352 4353 4354 4355 4356 4357 4358 4359 4360 4361 4362
	if (rb) {
		spin_lock_irqsave(&rb->event_lock, flags);
		list_add_rcu(&event->rb_entry, &rb->event_list);
		spin_unlock_irqrestore(&rb->event_lock, flags);
	}

	rcu_assign_pointer(event->rb, rb);

	if (old_rb) {
		ring_buffer_put(old_rb);
		/*
		 * Since we detached before setting the new rb, so that we
		 * could attach the new rb, we could have missed a wakeup.
		 * Provide it now.
		 */
		wake_up_all(&event->waitq);
	}
4363 4364 4365 4366 4367 4368 4369 4370
}

static void ring_buffer_wakeup(struct perf_event *event)
{
	struct ring_buffer *rb;

	rcu_read_lock();
	rb = rcu_dereference(event->rb);
4371 4372 4373 4374
	if (rb) {
		list_for_each_entry_rcu(event, &rb->event_list, rb_entry)
			wake_up_all(&event->waitq);
	}
4375 4376 4377
	rcu_read_unlock();
}

4378
static void rb_free_rcu(struct rcu_head *rcu_head)
4379
{
4380
	struct ring_buffer *rb;
4381

4382 4383
	rb = container_of(rcu_head, struct ring_buffer, rcu_head);
	rb_free(rb);
4384 4385
}

4386
struct ring_buffer *ring_buffer_get(struct perf_event *event)
4387
{
4388
	struct ring_buffer *rb;
4389

4390
	rcu_read_lock();
4391 4392 4393 4394
	rb = rcu_dereference(event->rb);
	if (rb) {
		if (!atomic_inc_not_zero(&rb->refcount))
			rb = NULL;
4395 4396 4397
	}
	rcu_read_unlock();

4398
	return rb;
4399 4400
}

4401
void ring_buffer_put(struct ring_buffer *rb)
4402
{
4403
	if (!atomic_dec_and_test(&rb->refcount))
4404
		return;
4405

4406
	WARN_ON_ONCE(!list_empty(&rb->event_list));
4407

4408
	call_rcu(&rb->rcu_head, rb_free_rcu);
4409 4410 4411 4412
}

static void perf_mmap_open(struct vm_area_struct *vma)
{
4413
	struct perf_event *event = vma->vm_file->private_data;
4414

4415
	atomic_inc(&event->mmap_count);
4416
	atomic_inc(&event->rb->mmap_count);
4417

4418 4419 4420
	if (vma->vm_pgoff)
		atomic_inc(&event->rb->aux_mmap_count);

4421 4422
	if (event->pmu->event_mapped)
		event->pmu->event_mapped(event);
4423 4424
}

4425 4426 4427 4428 4429 4430 4431 4432
/*
 * A buffer can be mmap()ed multiple times; either directly through the same
 * event, or through other events by use of perf_event_set_output().
 *
 * In order to undo the VM accounting done by perf_mmap() we need to destroy
 * the buffer here, where we still have a VM context. This means we need
 * to detach all events redirecting to us.
 */
4433 4434
static void perf_mmap_close(struct vm_area_struct *vma)
{
4435
	struct perf_event *event = vma->vm_file->private_data;
4436

4437
	struct ring_buffer *rb = ring_buffer_get(event);
4438 4439 4440
	struct user_struct *mmap_user = rb->mmap_user;
	int mmap_locked = rb->mmap_locked;
	unsigned long size = perf_data_size(rb);
4441

4442 4443 4444
	if (event->pmu->event_unmapped)
		event->pmu->event_unmapped(event);

4445 4446 4447 4448 4449 4450 4451 4452 4453 4454 4455 4456 4457 4458
	/*
	 * rb->aux_mmap_count will always drop before rb->mmap_count and
	 * event->mmap_count, so it is ok to use event->mmap_mutex to
	 * serialize with perf_mmap here.
	 */
	if (rb_has_aux(rb) && vma->vm_pgoff == rb->aux_pgoff &&
	    atomic_dec_and_mutex_lock(&rb->aux_mmap_count, &event->mmap_mutex)) {
		atomic_long_sub(rb->aux_nr_pages, &mmap_user->locked_vm);
		vma->vm_mm->pinned_vm -= rb->aux_mmap_locked;

		rb_free_aux(rb);
		mutex_unlock(&event->mmap_mutex);
	}

4459 4460 4461
	atomic_dec(&rb->mmap_count);

	if (!atomic_dec_and_mutex_lock(&event->mmap_count, &event->mmap_mutex))
4462
		goto out_put;
4463

4464
	ring_buffer_attach(event, NULL);
4465 4466 4467
	mutex_unlock(&event->mmap_mutex);

	/* If there's still other mmap()s of this buffer, we're done. */
4468 4469
	if (atomic_read(&rb->mmap_count))
		goto out_put;
4470

4471 4472 4473 4474 4475 4476 4477 4478 4479 4480 4481 4482 4483 4484 4485 4486
	/*
	 * No other mmap()s, detach from all other events that might redirect
	 * into the now unreachable buffer. Somewhat complicated by the
	 * fact that rb::event_lock otherwise nests inside mmap_mutex.
	 */
again:
	rcu_read_lock();
	list_for_each_entry_rcu(event, &rb->event_list, rb_entry) {
		if (!atomic_long_inc_not_zero(&event->refcount)) {
			/*
			 * This event is en-route to free_event() which will
			 * detach it and remove it from the list.
			 */
			continue;
		}
		rcu_read_unlock();
4487

4488 4489 4490 4491 4492 4493 4494 4495 4496 4497 4498
		mutex_lock(&event->mmap_mutex);
		/*
		 * Check we didn't race with perf_event_set_output() which can
		 * swizzle the rb from under us while we were waiting to
		 * acquire mmap_mutex.
		 *
		 * If we find a different rb; ignore this event, a next
		 * iteration will no longer find it on the list. We have to
		 * still restart the iteration to make sure we're not now
		 * iterating the wrong list.
		 */
4499 4500 4501
		if (event->rb == rb)
			ring_buffer_attach(event, NULL);

4502
		mutex_unlock(&event->mmap_mutex);
4503
		put_event(event);
4504

4505 4506 4507 4508 4509
		/*
		 * Restart the iteration; either we're on the wrong list or
		 * destroyed its integrity by doing a deletion.
		 */
		goto again;
4510
	}
4511 4512 4513 4514 4515 4516 4517 4518 4519 4520 4521 4522 4523 4524 4525
	rcu_read_unlock();

	/*
	 * It could be there's still a few 0-ref events on the list; they'll
	 * get cleaned up by free_event() -- they'll also still have their
	 * ref on the rb and will free it whenever they are done with it.
	 *
	 * Aside from that, this buffer is 'fully' detached and unmapped,
	 * undo the VM accounting.
	 */

	atomic_long_sub((size >> PAGE_SHIFT) + 1, &mmap_user->locked_vm);
	vma->vm_mm->pinned_vm -= mmap_locked;
	free_uid(mmap_user);

4526
out_put:
4527
	ring_buffer_put(rb); /* could be last */
4528 4529
}

4530
static const struct vm_operations_struct perf_mmap_vmops = {
4531
	.open		= perf_mmap_open,
4532
	.close		= perf_mmap_close, /* non mergable */
4533 4534
	.fault		= perf_mmap_fault,
	.page_mkwrite	= perf_mmap_fault,
4535 4536 4537 4538
};

static int perf_mmap(struct file *file, struct vm_area_struct *vma)
{
4539
	struct perf_event *event = file->private_data;
4540
	unsigned long user_locked, user_lock_limit;
4541
	struct user_struct *user = current_user();
4542
	unsigned long locked, lock_limit;
4543
	struct ring_buffer *rb = NULL;
4544 4545
	unsigned long vma_size;
	unsigned long nr_pages;
4546
	long user_extra = 0, extra = 0;
4547
	int ret = 0, flags = 0;
4548

4549 4550 4551
	/*
	 * Don't allow mmap() of inherited per-task counters. This would
	 * create a performance issue due to all children writing to the
4552
	 * same rb.
4553 4554 4555 4556
	 */
	if (event->cpu == -1 && event->attr.inherit)
		return -EINVAL;

4557
	if (!(vma->vm_flags & VM_SHARED))
4558
		return -EINVAL;
4559 4560

	vma_size = vma->vm_end - vma->vm_start;
4561 4562 4563 4564 4565 4566 4567 4568 4569 4570 4571 4572 4573 4574 4575 4576 4577 4578 4579 4580 4581 4582 4583 4584 4585 4586 4587 4588 4589 4590 4591 4592 4593 4594 4595 4596 4597 4598 4599 4600 4601 4602 4603 4604 4605 4606 4607 4608 4609 4610 4611 4612 4613 4614 4615 4616 4617 4618 4619 4620

	if (vma->vm_pgoff == 0) {
		nr_pages = (vma_size / PAGE_SIZE) - 1;
	} else {
		/*
		 * AUX area mapping: if rb->aux_nr_pages != 0, it's already
		 * mapped, all subsequent mappings should have the same size
		 * and offset. Must be above the normal perf buffer.
		 */
		u64 aux_offset, aux_size;

		if (!event->rb)
			return -EINVAL;

		nr_pages = vma_size / PAGE_SIZE;

		mutex_lock(&event->mmap_mutex);
		ret = -EINVAL;

		rb = event->rb;
		if (!rb)
			goto aux_unlock;

		aux_offset = ACCESS_ONCE(rb->user_page->aux_offset);
		aux_size = ACCESS_ONCE(rb->user_page->aux_size);

		if (aux_offset < perf_data_size(rb) + PAGE_SIZE)
			goto aux_unlock;

		if (aux_offset != vma->vm_pgoff << PAGE_SHIFT)
			goto aux_unlock;

		/* already mapped with a different offset */
		if (rb_has_aux(rb) && rb->aux_pgoff != vma->vm_pgoff)
			goto aux_unlock;

		if (aux_size != vma_size || aux_size != nr_pages * PAGE_SIZE)
			goto aux_unlock;

		/* already mapped with a different size */
		if (rb_has_aux(rb) && rb->aux_nr_pages != nr_pages)
			goto aux_unlock;

		if (!is_power_of_2(nr_pages))
			goto aux_unlock;

		if (!atomic_inc_not_zero(&rb->mmap_count))
			goto aux_unlock;

		if (rb_has_aux(rb)) {
			atomic_inc(&rb->aux_mmap_count);
			ret = 0;
			goto unlock;
		}

		atomic_set(&rb->aux_mmap_count, 1);
		user_extra = nr_pages;

		goto accounting;
	}
4621

4622
	/*
4623
	 * If we have rb pages ensure they're a power-of-two number, so we
4624 4625
	 * can do bitmasks instead of modulo.
	 */
4626
	if (nr_pages != 0 && !is_power_of_2(nr_pages))
4627 4628
		return -EINVAL;

4629
	if (vma_size != PAGE_SIZE * (1 + nr_pages))
4630 4631
		return -EINVAL;

4632
	WARN_ON_ONCE(event->ctx->parent_ctx);
4633
again:
4634
	mutex_lock(&event->mmap_mutex);
4635
	if (event->rb) {
4636
		if (event->rb->nr_pages != nr_pages) {
4637
			ret = -EINVAL;
4638 4639 4640 4641 4642 4643 4644 4645 4646 4647 4648 4649 4650
			goto unlock;
		}

		if (!atomic_inc_not_zero(&event->rb->mmap_count)) {
			/*
			 * Raced against perf_mmap_close() through
			 * perf_event_set_output(). Try again, hope for better
			 * luck.
			 */
			mutex_unlock(&event->mmap_mutex);
			goto again;
		}

4651 4652 4653
		goto unlock;
	}

4654
	user_extra = nr_pages + 1;
4655 4656

accounting:
4657
	user_lock_limit = sysctl_perf_event_mlock >> (PAGE_SHIFT - 10);
I
Ingo Molnar 已提交
4658 4659 4660 4661 4662 4663

	/*
	 * Increase the limit linearly with more CPUs:
	 */
	user_lock_limit *= num_online_cpus();

4664
	user_locked = atomic_long_read(&user->locked_vm) + user_extra;
4665

4666 4667
	if (user_locked > user_lock_limit)
		extra = user_locked - user_lock_limit;
4668

4669
	lock_limit = rlimit(RLIMIT_MEMLOCK);
4670
	lock_limit >>= PAGE_SHIFT;
4671
	locked = vma->vm_mm->pinned_vm + extra;
4672

4673 4674
	if ((locked > lock_limit) && perf_paranoid_tracepoint_raw() &&
		!capable(CAP_IPC_LOCK)) {
4675 4676 4677
		ret = -EPERM;
		goto unlock;
	}
4678

4679
	WARN_ON(!rb && event->rb);
4680

4681
	if (vma->vm_flags & VM_WRITE)
4682
		flags |= RING_BUFFER_WRITABLE;
4683

4684
	if (!rb) {
4685 4686 4687
		rb = rb_alloc(nr_pages,
			      event->attr.watermark ? event->attr.wakeup_watermark : 0,
			      event->cpu, flags);
P
Peter Zijlstra 已提交
4688

4689 4690 4691 4692
		if (!rb) {
			ret = -ENOMEM;
			goto unlock;
		}
4693

4694 4695 4696
		atomic_set(&rb->mmap_count, 1);
		rb->mmap_user = get_current_user();
		rb->mmap_locked = extra;
P
Peter Zijlstra 已提交
4697

4698
		ring_buffer_attach(event, rb);
4699

4700 4701 4702
		perf_event_init_userpage(event);
		perf_event_update_userpage(event);
	} else {
4703 4704
		ret = rb_alloc_aux(rb, event, vma->vm_pgoff, nr_pages,
				   event->attr.aux_watermark, flags);
4705 4706 4707
		if (!ret)
			rb->aux_mmap_locked = extra;
	}
4708

4709
unlock:
4710 4711 4712 4713
	if (!ret) {
		atomic_long_add(user_extra, &user->locked_vm);
		vma->vm_mm->pinned_vm += extra;

4714
		atomic_inc(&event->mmap_count);
4715 4716 4717 4718
	} else if (rb) {
		atomic_dec(&rb->mmap_count);
	}
aux_unlock:
4719
	mutex_unlock(&event->mmap_mutex);
4720

4721 4722 4723 4724
	/*
	 * Since pinned accounting is per vm we cannot allow fork() to copy our
	 * vma.
	 */
P
Peter Zijlstra 已提交
4725
	vma->vm_flags |= VM_DONTCOPY | VM_DONTEXPAND | VM_DONTDUMP;
4726
	vma->vm_ops = &perf_mmap_vmops;
4727

4728 4729 4730
	if (event->pmu->event_mapped)
		event->pmu->event_mapped(event);

4731
	return ret;
4732 4733
}

P
Peter Zijlstra 已提交
4734 4735
static int perf_fasync(int fd, struct file *filp, int on)
{
A
Al Viro 已提交
4736
	struct inode *inode = file_inode(filp);
4737
	struct perf_event *event = filp->private_data;
P
Peter Zijlstra 已提交
4738 4739 4740
	int retval;

	mutex_lock(&inode->i_mutex);
4741
	retval = fasync_helper(fd, filp, on, &event->fasync);
P
Peter Zijlstra 已提交
4742 4743 4744 4745 4746 4747 4748 4749
	mutex_unlock(&inode->i_mutex);

	if (retval < 0)
		return retval;

	return 0;
}

T
Thomas Gleixner 已提交
4750
static const struct file_operations perf_fops = {
4751
	.llseek			= no_llseek,
T
Thomas Gleixner 已提交
4752 4753 4754
	.release		= perf_release,
	.read			= perf_read,
	.poll			= perf_poll,
4755
	.unlocked_ioctl		= perf_ioctl,
P
Pawel Moll 已提交
4756
	.compat_ioctl		= perf_compat_ioctl,
4757
	.mmap			= perf_mmap,
P
Peter Zijlstra 已提交
4758
	.fasync			= perf_fasync,
T
Thomas Gleixner 已提交
4759 4760
};

4761
/*
4762
 * Perf event wakeup
4763 4764 4765 4766 4767
 *
 * If there's data, ensure we set the poll() state and publish everything
 * to user-space before waking everybody up.
 */

4768
void perf_event_wakeup(struct perf_event *event)
4769
{
4770
	ring_buffer_wakeup(event);
4771

4772 4773 4774
	if (event->pending_kill) {
		kill_fasync(&event->fasync, SIGIO, event->pending_kill);
		event->pending_kill = 0;
4775
	}
4776 4777
}

4778
static void perf_pending_event(struct irq_work *entry)
4779
{
4780 4781
	struct perf_event *event = container_of(entry,
			struct perf_event, pending);
4782 4783 4784 4785 4786 4787 4788
	int rctx;

	rctx = perf_swevent_get_recursion_context();
	/*
	 * If we 'fail' here, that's OK, it means recursion is already disabled
	 * and we won't recurse 'further'.
	 */
4789

4790 4791 4792
	if (event->pending_disable) {
		event->pending_disable = 0;
		__perf_event_disable(event);
4793 4794
	}

4795 4796 4797
	if (event->pending_wakeup) {
		event->pending_wakeup = 0;
		perf_event_wakeup(event);
4798
	}
4799 4800 4801

	if (rctx >= 0)
		perf_swevent_put_recursion_context(rctx);
4802 4803
}

4804 4805 4806 4807 4808 4809 4810 4811 4812 4813 4814 4815 4816 4817 4818 4819 4820 4821 4822 4823 4824
/*
 * We assume there is only KVM supporting the callbacks.
 * Later on, we might change it to a list if there is
 * another virtualization implementation supporting the callbacks.
 */
struct perf_guest_info_callbacks *perf_guest_cbs;

int perf_register_guest_info_callbacks(struct perf_guest_info_callbacks *cbs)
{
	perf_guest_cbs = cbs;
	return 0;
}
EXPORT_SYMBOL_GPL(perf_register_guest_info_callbacks);

int perf_unregister_guest_info_callbacks(struct perf_guest_info_callbacks *cbs)
{
	perf_guest_cbs = NULL;
	return 0;
}
EXPORT_SYMBOL_GPL(perf_unregister_guest_info_callbacks);

4825 4826 4827 4828 4829 4830 4831 4832 4833 4834 4835 4836 4837 4838 4839
static void
perf_output_sample_regs(struct perf_output_handle *handle,
			struct pt_regs *regs, u64 mask)
{
	int bit;

	for_each_set_bit(bit, (const unsigned long *) &mask,
			 sizeof(mask) * BITS_PER_BYTE) {
		u64 val;

		val = perf_reg_value(regs, bit);
		perf_output_put(handle, val);
	}
}

4840
static void perf_sample_regs_user(struct perf_regs *regs_user,
4841 4842
				  struct pt_regs *regs,
				  struct pt_regs *regs_user_copy)
4843
{
4844 4845
	if (user_mode(regs)) {
		regs_user->abi = perf_reg_abi(current);
4846
		regs_user->regs = regs;
4847 4848
	} else if (current->mm) {
		perf_get_regs_user(regs_user, regs, regs_user_copy);
4849 4850 4851
	} else {
		regs_user->abi = PERF_SAMPLE_REGS_ABI_NONE;
		regs_user->regs = NULL;
4852 4853 4854
	}
}

4855 4856 4857 4858 4859 4860 4861 4862
static void perf_sample_regs_intr(struct perf_regs *regs_intr,
				  struct pt_regs *regs)
{
	regs_intr->regs = regs;
	regs_intr->abi  = perf_reg_abi(current);
}


4863 4864 4865 4866 4867 4868 4869 4870 4871 4872 4873 4874 4875 4876 4877 4878 4879 4880 4881 4882 4883 4884 4885 4886 4887 4888 4889 4890 4891 4892 4893 4894 4895 4896 4897 4898 4899 4900 4901 4902 4903 4904 4905 4906 4907 4908 4909 4910 4911 4912 4913 4914 4915 4916 4917 4918 4919 4920 4921 4922 4923 4924 4925 4926 4927 4928 4929 4930 4931 4932 4933 4934 4935 4936 4937 4938 4939 4940 4941 4942 4943 4944 4945 4946 4947 4948 4949 4950 4951 4952 4953 4954 4955 4956 4957
/*
 * Get remaining task size from user stack pointer.
 *
 * It'd be better to take stack vma map and limit this more
 * precisly, but there's no way to get it safely under interrupt,
 * so using TASK_SIZE as limit.
 */
static u64 perf_ustack_task_size(struct pt_regs *regs)
{
	unsigned long addr = perf_user_stack_pointer(regs);

	if (!addr || addr >= TASK_SIZE)
		return 0;

	return TASK_SIZE - addr;
}

static u16
perf_sample_ustack_size(u16 stack_size, u16 header_size,
			struct pt_regs *regs)
{
	u64 task_size;

	/* No regs, no stack pointer, no dump. */
	if (!regs)
		return 0;

	/*
	 * Check if we fit in with the requested stack size into the:
	 * - TASK_SIZE
	 *   If we don't, we limit the size to the TASK_SIZE.
	 *
	 * - remaining sample size
	 *   If we don't, we customize the stack size to
	 *   fit in to the remaining sample size.
	 */

	task_size  = min((u64) USHRT_MAX, perf_ustack_task_size(regs));
	stack_size = min(stack_size, (u16) task_size);

	/* Current header size plus static size and dynamic size. */
	header_size += 2 * sizeof(u64);

	/* Do we fit in with the current stack dump size? */
	if ((u16) (header_size + stack_size) < header_size) {
		/*
		 * If we overflow the maximum size for the sample,
		 * we customize the stack dump size to fit in.
		 */
		stack_size = USHRT_MAX - header_size - sizeof(u64);
		stack_size = round_up(stack_size, sizeof(u64));
	}

	return stack_size;
}

static void
perf_output_sample_ustack(struct perf_output_handle *handle, u64 dump_size,
			  struct pt_regs *regs)
{
	/* Case of a kernel thread, nothing to dump */
	if (!regs) {
		u64 size = 0;
		perf_output_put(handle, size);
	} else {
		unsigned long sp;
		unsigned int rem;
		u64 dyn_size;

		/*
		 * We dump:
		 * static size
		 *   - the size requested by user or the best one we can fit
		 *     in to the sample max size
		 * data
		 *   - user stack dump data
		 * dynamic size
		 *   - the actual dumped size
		 */

		/* Static size. */
		perf_output_put(handle, dump_size);

		/* Data. */
		sp = perf_user_stack_pointer(regs);
		rem = __output_copy_user(handle, (void *) sp, dump_size);
		dyn_size = dump_size - rem;

		perf_output_skip(handle, rem);

		/* Dynamic size. */
		perf_output_put(handle, dyn_size);
	}
}

4958 4959 4960
static void __perf_event_header__init_id(struct perf_event_header *header,
					 struct perf_sample_data *data,
					 struct perf_event *event)
4961 4962 4963 4964 4965 4966 4967 4968 4969 4970 4971 4972 4973
{
	u64 sample_type = event->attr.sample_type;

	data->type = sample_type;
	header->size += event->id_header_size;

	if (sample_type & PERF_SAMPLE_TID) {
		/* namespace issues */
		data->tid_entry.pid = perf_event_pid(event, current);
		data->tid_entry.tid = perf_event_tid(event, current);
	}

	if (sample_type & PERF_SAMPLE_TIME)
4974
		data->time = perf_event_clock(event);
4975

4976
	if (sample_type & (PERF_SAMPLE_ID | PERF_SAMPLE_IDENTIFIER))
4977 4978 4979 4980 4981 4982 4983 4984 4985 4986 4987
		data->id = primary_event_id(event);

	if (sample_type & PERF_SAMPLE_STREAM_ID)
		data->stream_id = event->id;

	if (sample_type & PERF_SAMPLE_CPU) {
		data->cpu_entry.cpu	 = raw_smp_processor_id();
		data->cpu_entry.reserved = 0;
	}
}

4988 4989 4990
void perf_event_header__init_id(struct perf_event_header *header,
				struct perf_sample_data *data,
				struct perf_event *event)
4991 4992 4993 4994 4995 4996 4997 4998 4999 5000 5001 5002 5003 5004 5005 5006 5007 5008 5009 5010 5011 5012 5013 5014
{
	if (event->attr.sample_id_all)
		__perf_event_header__init_id(header, data, event);
}

static void __perf_event__output_id_sample(struct perf_output_handle *handle,
					   struct perf_sample_data *data)
{
	u64 sample_type = data->type;

	if (sample_type & PERF_SAMPLE_TID)
		perf_output_put(handle, data->tid_entry);

	if (sample_type & PERF_SAMPLE_TIME)
		perf_output_put(handle, data->time);

	if (sample_type & PERF_SAMPLE_ID)
		perf_output_put(handle, data->id);

	if (sample_type & PERF_SAMPLE_STREAM_ID)
		perf_output_put(handle, data->stream_id);

	if (sample_type & PERF_SAMPLE_CPU)
		perf_output_put(handle, data->cpu_entry);
5015 5016 5017

	if (sample_type & PERF_SAMPLE_IDENTIFIER)
		perf_output_put(handle, data->id);
5018 5019
}

5020 5021 5022
void perf_event__output_id_sample(struct perf_event *event,
				  struct perf_output_handle *handle,
				  struct perf_sample_data *sample)
5023 5024 5025 5026 5027
{
	if (event->attr.sample_id_all)
		__perf_event__output_id_sample(handle, sample);
}

5028
static void perf_output_read_one(struct perf_output_handle *handle,
5029 5030
				 struct perf_event *event,
				 u64 enabled, u64 running)
5031
{
5032
	u64 read_format = event->attr.read_format;
5033 5034 5035
	u64 values[4];
	int n = 0;

P
Peter Zijlstra 已提交
5036
	values[n++] = perf_event_count(event);
5037
	if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) {
5038
		values[n++] = enabled +
5039
			atomic64_read(&event->child_total_time_enabled);
5040 5041
	}
	if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) {
5042
		values[n++] = running +
5043
			atomic64_read(&event->child_total_time_running);
5044 5045
	}
	if (read_format & PERF_FORMAT_ID)
5046
		values[n++] = primary_event_id(event);
5047

5048
	__output_copy(handle, values, n * sizeof(u64));
5049 5050 5051
}

/*
5052
 * XXX PERF_FORMAT_GROUP vs inherited events seems difficult.
5053 5054
 */
static void perf_output_read_group(struct perf_output_handle *handle,
5055 5056
			    struct perf_event *event,
			    u64 enabled, u64 running)
5057
{
5058 5059
	struct perf_event *leader = event->group_leader, *sub;
	u64 read_format = event->attr.read_format;
5060 5061 5062 5063 5064 5065
	u64 values[5];
	int n = 0;

	values[n++] = 1 + leader->nr_siblings;

	if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
5066
		values[n++] = enabled;
5067 5068

	if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
5069
		values[n++] = running;
5070

5071
	if (leader != event)
5072 5073
		leader->pmu->read(leader);

P
Peter Zijlstra 已提交
5074
	values[n++] = perf_event_count(leader);
5075
	if (read_format & PERF_FORMAT_ID)
5076
		values[n++] = primary_event_id(leader);
5077

5078
	__output_copy(handle, values, n * sizeof(u64));
5079

5080
	list_for_each_entry(sub, &leader->sibling_list, group_entry) {
5081 5082
		n = 0;

5083 5084
		if ((sub != event) &&
		    (sub->state == PERF_EVENT_STATE_ACTIVE))
5085 5086
			sub->pmu->read(sub);

P
Peter Zijlstra 已提交
5087
		values[n++] = perf_event_count(sub);
5088
		if (read_format & PERF_FORMAT_ID)
5089
			values[n++] = primary_event_id(sub);
5090

5091
		__output_copy(handle, values, n * sizeof(u64));
5092 5093 5094
	}
}

5095 5096 5097
#define PERF_FORMAT_TOTAL_TIMES (PERF_FORMAT_TOTAL_TIME_ENABLED|\
				 PERF_FORMAT_TOTAL_TIME_RUNNING)

5098
static void perf_output_read(struct perf_output_handle *handle,
5099
			     struct perf_event *event)
5100
{
5101
	u64 enabled = 0, running = 0, now;
5102 5103 5104 5105 5106 5107 5108 5109 5110 5111 5112
	u64 read_format = event->attr.read_format;

	/*
	 * compute total_time_enabled, total_time_running
	 * based on snapshot values taken when the event
	 * was last scheduled in.
	 *
	 * we cannot simply called update_context_time()
	 * because of locking issue as we are called in
	 * NMI context
	 */
5113
	if (read_format & PERF_FORMAT_TOTAL_TIMES)
5114
		calc_timer_values(event, &now, &enabled, &running);
5115

5116
	if (event->attr.read_format & PERF_FORMAT_GROUP)
5117
		perf_output_read_group(handle, event, enabled, running);
5118
	else
5119
		perf_output_read_one(handle, event, enabled, running);
5120 5121
}

5122 5123 5124
void perf_output_sample(struct perf_output_handle *handle,
			struct perf_event_header *header,
			struct perf_sample_data *data,
5125
			struct perf_event *event)
5126 5127 5128 5129 5130
{
	u64 sample_type = data->type;

	perf_output_put(handle, *header);

5131 5132 5133
	if (sample_type & PERF_SAMPLE_IDENTIFIER)
		perf_output_put(handle, data->id);

5134 5135 5136 5137 5138 5139 5140 5141 5142 5143 5144 5145 5146 5147 5148 5149 5150 5151 5152 5153 5154 5155 5156 5157 5158
	if (sample_type & PERF_SAMPLE_IP)
		perf_output_put(handle, data->ip);

	if (sample_type & PERF_SAMPLE_TID)
		perf_output_put(handle, data->tid_entry);

	if (sample_type & PERF_SAMPLE_TIME)
		perf_output_put(handle, data->time);

	if (sample_type & PERF_SAMPLE_ADDR)
		perf_output_put(handle, data->addr);

	if (sample_type & PERF_SAMPLE_ID)
		perf_output_put(handle, data->id);

	if (sample_type & PERF_SAMPLE_STREAM_ID)
		perf_output_put(handle, data->stream_id);

	if (sample_type & PERF_SAMPLE_CPU)
		perf_output_put(handle, data->cpu_entry);

	if (sample_type & PERF_SAMPLE_PERIOD)
		perf_output_put(handle, data->period);

	if (sample_type & PERF_SAMPLE_READ)
5159
		perf_output_read(handle, event);
5160 5161 5162 5163 5164 5165 5166 5167 5168 5169

	if (sample_type & PERF_SAMPLE_CALLCHAIN) {
		if (data->callchain) {
			int size = 1;

			if (data->callchain)
				size += data->callchain->nr;

			size *= sizeof(u64);

5170
			__output_copy(handle, data->callchain, size);
5171 5172 5173 5174 5175 5176 5177 5178 5179
		} else {
			u64 nr = 0;
			perf_output_put(handle, nr);
		}
	}

	if (sample_type & PERF_SAMPLE_RAW) {
		if (data->raw) {
			perf_output_put(handle, data->raw->size);
5180 5181
			__output_copy(handle, data->raw->data,
					   data->raw->size);
5182 5183 5184 5185 5186 5187 5188 5189 5190 5191 5192
		} else {
			struct {
				u32	size;
				u32	data;
			} raw = {
				.size = sizeof(u32),
				.data = 0,
			};
			perf_output_put(handle, raw);
		}
	}
5193

5194 5195 5196 5197 5198 5199 5200 5201 5202 5203 5204 5205 5206 5207 5208 5209 5210
	if (sample_type & PERF_SAMPLE_BRANCH_STACK) {
		if (data->br_stack) {
			size_t size;

			size = data->br_stack->nr
			     * sizeof(struct perf_branch_entry);

			perf_output_put(handle, data->br_stack->nr);
			perf_output_copy(handle, data->br_stack->entries, size);
		} else {
			/*
			 * we always store at least the value of nr
			 */
			u64 nr = 0;
			perf_output_put(handle, nr);
		}
	}
5211 5212 5213 5214 5215 5216 5217 5218 5219 5220 5221 5222 5223 5224 5225 5226 5227

	if (sample_type & PERF_SAMPLE_REGS_USER) {
		u64 abi = data->regs_user.abi;

		/*
		 * If there are no regs to dump, notice it through
		 * first u64 being zero (PERF_SAMPLE_REGS_ABI_NONE).
		 */
		perf_output_put(handle, abi);

		if (abi) {
			u64 mask = event->attr.sample_regs_user;
			perf_output_sample_regs(handle,
						data->regs_user.regs,
						mask);
		}
	}
5228

5229
	if (sample_type & PERF_SAMPLE_STACK_USER) {
5230 5231 5232
		perf_output_sample_ustack(handle,
					  data->stack_user_size,
					  data->regs_user.regs);
5233
	}
A
Andi Kleen 已提交
5234 5235 5236

	if (sample_type & PERF_SAMPLE_WEIGHT)
		perf_output_put(handle, data->weight);
5237 5238 5239

	if (sample_type & PERF_SAMPLE_DATA_SRC)
		perf_output_put(handle, data->data_src.val);
5240

A
Andi Kleen 已提交
5241 5242 5243
	if (sample_type & PERF_SAMPLE_TRANSACTION)
		perf_output_put(handle, data->txn);

5244 5245 5246 5247 5248 5249 5250 5251 5252 5253 5254 5255 5256 5257 5258 5259 5260
	if (sample_type & PERF_SAMPLE_REGS_INTR) {
		u64 abi = data->regs_intr.abi;
		/*
		 * If there are no regs to dump, notice it through
		 * first u64 being zero (PERF_SAMPLE_REGS_ABI_NONE).
		 */
		perf_output_put(handle, abi);

		if (abi) {
			u64 mask = event->attr.sample_regs_intr;

			perf_output_sample_regs(handle,
						data->regs_intr.regs,
						mask);
		}
	}

5261 5262 5263 5264 5265 5266 5267 5268 5269 5270 5271 5272 5273
	if (!event->attr.watermark) {
		int wakeup_events = event->attr.wakeup_events;

		if (wakeup_events) {
			struct ring_buffer *rb = handle->rb;
			int events = local_inc_return(&rb->events);

			if (events >= wakeup_events) {
				local_sub(wakeup_events, &rb->events);
				local_inc(&rb->wakeup);
			}
		}
	}
5274 5275 5276 5277
}

void perf_prepare_sample(struct perf_event_header *header,
			 struct perf_sample_data *data,
5278
			 struct perf_event *event,
5279
			 struct pt_regs *regs)
5280
{
5281
	u64 sample_type = event->attr.sample_type;
5282

5283
	header->type = PERF_RECORD_SAMPLE;
5284
	header->size = sizeof(*header) + event->header_size;
5285 5286 5287

	header->misc = 0;
	header->misc |= perf_misc_flags(regs);
5288

5289
	__perf_event_header__init_id(header, data, event);
5290

5291
	if (sample_type & PERF_SAMPLE_IP)
5292 5293
		data->ip = perf_instruction_pointer(regs);

5294
	if (sample_type & PERF_SAMPLE_CALLCHAIN) {
5295
		int size = 1;
5296

5297
		data->callchain = perf_callchain(event, regs);
5298 5299 5300 5301 5302

		if (data->callchain)
			size += data->callchain->nr;

		header->size += size * sizeof(u64);
5303 5304
	}

5305
	if (sample_type & PERF_SAMPLE_RAW) {
5306 5307 5308 5309 5310 5311 5312 5313
		int size = sizeof(u32);

		if (data->raw)
			size += data->raw->size;
		else
			size += sizeof(u32);

		WARN_ON_ONCE(size & (sizeof(u64)-1));
5314
		header->size += size;
5315
	}
5316 5317 5318 5319 5320 5321 5322 5323 5324

	if (sample_type & PERF_SAMPLE_BRANCH_STACK) {
		int size = sizeof(u64); /* nr */
		if (data->br_stack) {
			size += data->br_stack->nr
			      * sizeof(struct perf_branch_entry);
		}
		header->size += size;
	}
5325

5326
	if (sample_type & (PERF_SAMPLE_REGS_USER | PERF_SAMPLE_STACK_USER))
5327 5328
		perf_sample_regs_user(&data->regs_user, regs,
				      &data->regs_user_copy);
5329

5330 5331 5332 5333 5334 5335 5336 5337 5338 5339 5340
	if (sample_type & PERF_SAMPLE_REGS_USER) {
		/* regs dump ABI info */
		int size = sizeof(u64);

		if (data->regs_user.regs) {
			u64 mask = event->attr.sample_regs_user;
			size += hweight64(mask) * sizeof(u64);
		}

		header->size += size;
	}
5341 5342 5343 5344 5345 5346 5347 5348 5349 5350 5351 5352

	if (sample_type & PERF_SAMPLE_STACK_USER) {
		/*
		 * Either we need PERF_SAMPLE_STACK_USER bit to be allways
		 * processed as the last one or have additional check added
		 * in case new sample type is added, because we could eat
		 * up the rest of the sample size.
		 */
		u16 stack_size = event->attr.sample_stack_user;
		u16 size = sizeof(u64);

		stack_size = perf_sample_ustack_size(stack_size, header->size,
5353
						     data->regs_user.regs);
5354 5355 5356 5357 5358 5359 5360 5361 5362 5363 5364 5365

		/*
		 * If there is something to dump, add space for the dump
		 * itself and for the field that tells the dynamic size,
		 * which is how many have been actually dumped.
		 */
		if (stack_size)
			size += sizeof(u64) + stack_size;

		data->stack_user_size = stack_size;
		header->size += size;
	}
5366 5367 5368 5369 5370 5371 5372 5373 5374 5375 5376 5377 5378 5379 5380

	if (sample_type & PERF_SAMPLE_REGS_INTR) {
		/* regs dump ABI info */
		int size = sizeof(u64);

		perf_sample_regs_intr(&data->regs_intr, regs);

		if (data->regs_intr.regs) {
			u64 mask = event->attr.sample_regs_intr;

			size += hweight64(mask) * sizeof(u64);
		}

		header->size += size;
	}
5381
}
5382

5383
static void perf_event_output(struct perf_event *event,
5384 5385 5386 5387 5388
				struct perf_sample_data *data,
				struct pt_regs *regs)
{
	struct perf_output_handle handle;
	struct perf_event_header header;
5389

5390 5391 5392
	/* protect the callchain buffers */
	rcu_read_lock();

5393
	perf_prepare_sample(&header, data, event, regs);
P
Peter Zijlstra 已提交
5394

5395
	if (perf_output_begin(&handle, event, header.size))
5396
		goto exit;
5397

5398
	perf_output_sample(&handle, &header, data, event);
5399

5400
	perf_output_end(&handle);
5401 5402 5403

exit:
	rcu_read_unlock();
5404 5405
}

5406
/*
5407
 * read event_id
5408 5409 5410 5411 5412 5413 5414 5415 5416 5417
 */

struct perf_read_event {
	struct perf_event_header	header;

	u32				pid;
	u32				tid;
};

static void
5418
perf_event_read_event(struct perf_event *event,
5419 5420 5421
			struct task_struct *task)
{
	struct perf_output_handle handle;
5422
	struct perf_sample_data sample;
5423
	struct perf_read_event read_event = {
5424
		.header = {
5425
			.type = PERF_RECORD_READ,
5426
			.misc = 0,
5427
			.size = sizeof(read_event) + event->read_size,
5428
		},
5429 5430
		.pid = perf_event_pid(event, task),
		.tid = perf_event_tid(event, task),
5431
	};
5432
	int ret;
5433

5434
	perf_event_header__init_id(&read_event.header, &sample, event);
5435
	ret = perf_output_begin(&handle, event, read_event.header.size);
5436 5437 5438
	if (ret)
		return;

5439
	perf_output_put(&handle, read_event);
5440
	perf_output_read(&handle, event);
5441
	perf_event__output_id_sample(event, &handle, &sample);
5442

5443 5444 5445
	perf_output_end(&handle);
}

5446 5447 5448 5449 5450 5451 5452 5453 5454 5455 5456 5457 5458 5459
typedef void (perf_event_aux_output_cb)(struct perf_event *event, void *data);

static void
perf_event_aux_ctx(struct perf_event_context *ctx,
		   perf_event_aux_output_cb output,
		   void *data)
{
	struct perf_event *event;

	list_for_each_entry_rcu(event, &ctx->event_list, event_entry) {
		if (event->state < PERF_EVENT_STATE_INACTIVE)
			continue;
		if (!event_filter_match(event))
			continue;
5460
		output(event, data);
5461 5462 5463 5464
	}
}

static void
5465
perf_event_aux(perf_event_aux_output_cb output, void *data,
5466 5467 5468 5469 5470 5471 5472 5473 5474 5475 5476 5477
	       struct perf_event_context *task_ctx)
{
	struct perf_cpu_context *cpuctx;
	struct perf_event_context *ctx;
	struct pmu *pmu;
	int ctxn;

	rcu_read_lock();
	list_for_each_entry_rcu(pmu, &pmus, entry) {
		cpuctx = get_cpu_ptr(pmu->pmu_cpu_context);
		if (cpuctx->unique_pmu != pmu)
			goto next;
5478
		perf_event_aux_ctx(&cpuctx->ctx, output, data);
5479 5480 5481 5482 5483 5484 5485
		if (task_ctx)
			goto next;
		ctxn = pmu->task_ctx_nr;
		if (ctxn < 0)
			goto next;
		ctx = rcu_dereference(current->perf_event_ctxp[ctxn]);
		if (ctx)
5486
			perf_event_aux_ctx(ctx, output, data);
5487 5488 5489 5490 5491 5492
next:
		put_cpu_ptr(pmu->pmu_cpu_context);
	}

	if (task_ctx) {
		preempt_disable();
5493
		perf_event_aux_ctx(task_ctx, output, data);
5494 5495 5496 5497 5498
		preempt_enable();
	}
	rcu_read_unlock();
}

P
Peter Zijlstra 已提交
5499
/*
P
Peter Zijlstra 已提交
5500 5501
 * task tracking -- fork/exit
 *
5502
 * enabled by: attr.comm | attr.mmap | attr.mmap2 | attr.mmap_data | attr.task
P
Peter Zijlstra 已提交
5503 5504
 */

P
Peter Zijlstra 已提交
5505
struct perf_task_event {
5506
	struct task_struct		*task;
5507
	struct perf_event_context	*task_ctx;
P
Peter Zijlstra 已提交
5508 5509 5510 5511 5512 5513

	struct {
		struct perf_event_header	header;

		u32				pid;
		u32				ppid;
P
Peter Zijlstra 已提交
5514 5515
		u32				tid;
		u32				ptid;
5516
		u64				time;
5517
	} event_id;
P
Peter Zijlstra 已提交
5518 5519
};

5520 5521
static int perf_event_task_match(struct perf_event *event)
{
5522 5523 5524
	return event->attr.comm  || event->attr.mmap ||
	       event->attr.mmap2 || event->attr.mmap_data ||
	       event->attr.task;
5525 5526
}

5527
static void perf_event_task_output(struct perf_event *event,
5528
				   void *data)
P
Peter Zijlstra 已提交
5529
{
5530
	struct perf_task_event *task_event = data;
P
Peter Zijlstra 已提交
5531
	struct perf_output_handle handle;
5532
	struct perf_sample_data	sample;
P
Peter Zijlstra 已提交
5533
	struct task_struct *task = task_event->task;
5534
	int ret, size = task_event->event_id.header.size;
5535

5536 5537 5538
	if (!perf_event_task_match(event))
		return;

5539
	perf_event_header__init_id(&task_event->event_id.header, &sample, event);
P
Peter Zijlstra 已提交
5540

5541
	ret = perf_output_begin(&handle, event,
5542
				task_event->event_id.header.size);
5543
	if (ret)
5544
		goto out;
P
Peter Zijlstra 已提交
5545

5546 5547
	task_event->event_id.pid = perf_event_pid(event, task);
	task_event->event_id.ppid = perf_event_pid(event, current);
P
Peter Zijlstra 已提交
5548

5549 5550
	task_event->event_id.tid = perf_event_tid(event, task);
	task_event->event_id.ptid = perf_event_tid(event, current);
P
Peter Zijlstra 已提交
5551

5552 5553
	task_event->event_id.time = perf_event_clock(event);

5554
	perf_output_put(&handle, task_event->event_id);
5555

5556 5557
	perf_event__output_id_sample(event, &handle, &sample);

P
Peter Zijlstra 已提交
5558
	perf_output_end(&handle);
5559 5560
out:
	task_event->event_id.header.size = size;
P
Peter Zijlstra 已提交
5561 5562
}

5563 5564
static void perf_event_task(struct task_struct *task,
			      struct perf_event_context *task_ctx,
5565
			      int new)
P
Peter Zijlstra 已提交
5566
{
P
Peter Zijlstra 已提交
5567
	struct perf_task_event task_event;
P
Peter Zijlstra 已提交
5568

5569 5570 5571
	if (!atomic_read(&nr_comm_events) &&
	    !atomic_read(&nr_mmap_events) &&
	    !atomic_read(&nr_task_events))
P
Peter Zijlstra 已提交
5572 5573
		return;

P
Peter Zijlstra 已提交
5574
	task_event = (struct perf_task_event){
5575 5576
		.task	  = task,
		.task_ctx = task_ctx,
5577
		.event_id    = {
P
Peter Zijlstra 已提交
5578
			.header = {
5579
				.type = new ? PERF_RECORD_FORK : PERF_RECORD_EXIT,
5580
				.misc = 0,
5581
				.size = sizeof(task_event.event_id),
P
Peter Zijlstra 已提交
5582
			},
5583 5584
			/* .pid  */
			/* .ppid */
P
Peter Zijlstra 已提交
5585 5586
			/* .tid  */
			/* .ptid */
5587
			/* .time */
P
Peter Zijlstra 已提交
5588 5589 5590
		},
	};

5591
	perf_event_aux(perf_event_task_output,
5592 5593
		       &task_event,
		       task_ctx);
P
Peter Zijlstra 已提交
5594 5595
}

5596
void perf_event_fork(struct task_struct *task)
P
Peter Zijlstra 已提交
5597
{
5598
	perf_event_task(task, NULL, 1);
P
Peter Zijlstra 已提交
5599 5600
}

5601 5602 5603 5604 5605
/*
 * comm tracking
 */

struct perf_comm_event {
5606 5607
	struct task_struct	*task;
	char			*comm;
5608 5609 5610 5611 5612 5613 5614
	int			comm_size;

	struct {
		struct perf_event_header	header;

		u32				pid;
		u32				tid;
5615
	} event_id;
5616 5617
};

5618 5619 5620 5621 5622
static int perf_event_comm_match(struct perf_event *event)
{
	return event->attr.comm;
}

5623
static void perf_event_comm_output(struct perf_event *event,
5624
				   void *data)
5625
{
5626
	struct perf_comm_event *comm_event = data;
5627
	struct perf_output_handle handle;
5628
	struct perf_sample_data sample;
5629
	int size = comm_event->event_id.header.size;
5630 5631
	int ret;

5632 5633 5634
	if (!perf_event_comm_match(event))
		return;

5635 5636
	perf_event_header__init_id(&comm_event->event_id.header, &sample, event);
	ret = perf_output_begin(&handle, event,
5637
				comm_event->event_id.header.size);
5638 5639

	if (ret)
5640
		goto out;
5641

5642 5643
	comm_event->event_id.pid = perf_event_pid(event, comm_event->task);
	comm_event->event_id.tid = perf_event_tid(event, comm_event->task);
5644

5645
	perf_output_put(&handle, comm_event->event_id);
5646
	__output_copy(&handle, comm_event->comm,
5647
				   comm_event->comm_size);
5648 5649 5650

	perf_event__output_id_sample(event, &handle, &sample);

5651
	perf_output_end(&handle);
5652 5653
out:
	comm_event->event_id.header.size = size;
5654 5655
}

5656
static void perf_event_comm_event(struct perf_comm_event *comm_event)
5657
{
5658
	char comm[TASK_COMM_LEN];
5659 5660
	unsigned int size;

5661
	memset(comm, 0, sizeof(comm));
5662
	strlcpy(comm, comm_event->task->comm, sizeof(comm));
5663
	size = ALIGN(strlen(comm)+1, sizeof(u64));
5664 5665 5666 5667

	comm_event->comm = comm;
	comm_event->comm_size = size;

5668
	comm_event->event_id.header.size = sizeof(comm_event->event_id) + size;
P
Peter Zijlstra 已提交
5669

5670
	perf_event_aux(perf_event_comm_output,
5671 5672
		       comm_event,
		       NULL);
5673 5674
}

5675
void perf_event_comm(struct task_struct *task, bool exec)
5676
{
5677 5678
	struct perf_comm_event comm_event;

5679
	if (!atomic_read(&nr_comm_events))
5680
		return;
5681

5682
	comm_event = (struct perf_comm_event){
5683
		.task	= task,
5684 5685
		/* .comm      */
		/* .comm_size */
5686
		.event_id  = {
5687
			.header = {
5688
				.type = PERF_RECORD_COMM,
5689
				.misc = exec ? PERF_RECORD_MISC_COMM_EXEC : 0,
5690 5691 5692 5693
				/* .size */
			},
			/* .pid */
			/* .tid */
5694 5695 5696
		},
	};

5697
	perf_event_comm_event(&comm_event);
5698 5699
}

5700 5701 5702 5703 5704
/*
 * mmap tracking
 */

struct perf_mmap_event {
5705 5706 5707 5708
	struct vm_area_struct	*vma;

	const char		*file_name;
	int			file_size;
5709 5710 5711
	int			maj, min;
	u64			ino;
	u64			ino_generation;
5712
	u32			prot, flags;
5713 5714 5715 5716 5717 5718 5719 5720 5721

	struct {
		struct perf_event_header	header;

		u32				pid;
		u32				tid;
		u64				start;
		u64				len;
		u64				pgoff;
5722
	} event_id;
5723 5724
};

5725 5726 5727 5728 5729 5730 5731 5732
static int perf_event_mmap_match(struct perf_event *event,
				 void *data)
{
	struct perf_mmap_event *mmap_event = data;
	struct vm_area_struct *vma = mmap_event->vma;
	int executable = vma->vm_flags & VM_EXEC;

	return (!executable && event->attr.mmap_data) ||
5733
	       (executable && (event->attr.mmap || event->attr.mmap2));
5734 5735
}

5736
static void perf_event_mmap_output(struct perf_event *event,
5737
				   void *data)
5738
{
5739
	struct perf_mmap_event *mmap_event = data;
5740
	struct perf_output_handle handle;
5741
	struct perf_sample_data sample;
5742
	int size = mmap_event->event_id.header.size;
5743
	int ret;
5744

5745 5746 5747
	if (!perf_event_mmap_match(event, data))
		return;

5748 5749 5750 5751 5752
	if (event->attr.mmap2) {
		mmap_event->event_id.header.type = PERF_RECORD_MMAP2;
		mmap_event->event_id.header.size += sizeof(mmap_event->maj);
		mmap_event->event_id.header.size += sizeof(mmap_event->min);
		mmap_event->event_id.header.size += sizeof(mmap_event->ino);
5753
		mmap_event->event_id.header.size += sizeof(mmap_event->ino_generation);
5754 5755
		mmap_event->event_id.header.size += sizeof(mmap_event->prot);
		mmap_event->event_id.header.size += sizeof(mmap_event->flags);
5756 5757
	}

5758 5759
	perf_event_header__init_id(&mmap_event->event_id.header, &sample, event);
	ret = perf_output_begin(&handle, event,
5760
				mmap_event->event_id.header.size);
5761
	if (ret)
5762
		goto out;
5763

5764 5765
	mmap_event->event_id.pid = perf_event_pid(event, current);
	mmap_event->event_id.tid = perf_event_tid(event, current);
5766

5767
	perf_output_put(&handle, mmap_event->event_id);
5768 5769 5770 5771 5772 5773

	if (event->attr.mmap2) {
		perf_output_put(&handle, mmap_event->maj);
		perf_output_put(&handle, mmap_event->min);
		perf_output_put(&handle, mmap_event->ino);
		perf_output_put(&handle, mmap_event->ino_generation);
5774 5775
		perf_output_put(&handle, mmap_event->prot);
		perf_output_put(&handle, mmap_event->flags);
5776 5777
	}

5778
	__output_copy(&handle, mmap_event->file_name,
5779
				   mmap_event->file_size);
5780 5781 5782

	perf_event__output_id_sample(event, &handle, &sample);

5783
	perf_output_end(&handle);
5784 5785
out:
	mmap_event->event_id.header.size = size;
5786 5787
}

5788
static void perf_event_mmap_event(struct perf_mmap_event *mmap_event)
5789
{
5790 5791
	struct vm_area_struct *vma = mmap_event->vma;
	struct file *file = vma->vm_file;
5792 5793
	int maj = 0, min = 0;
	u64 ino = 0, gen = 0;
5794
	u32 prot = 0, flags = 0;
5795 5796 5797
	unsigned int size;
	char tmp[16];
	char *buf = NULL;
5798
	char *name;
5799

5800
	if (file) {
5801 5802
		struct inode *inode;
		dev_t dev;
5803

5804
		buf = kmalloc(PATH_MAX, GFP_KERNEL);
5805
		if (!buf) {
5806 5807
			name = "//enomem";
			goto cpy_name;
5808
		}
5809
		/*
5810
		 * d_path() works from the end of the rb backwards, so we
5811 5812 5813
		 * need to add enough zero bytes after the string to handle
		 * the 64bit alignment we do later.
		 */
5814
		name = d_path(&file->f_path, buf, PATH_MAX - sizeof(u64));
5815
		if (IS_ERR(name)) {
5816 5817
			name = "//toolong";
			goto cpy_name;
5818
		}
5819 5820 5821 5822 5823 5824
		inode = file_inode(vma->vm_file);
		dev = inode->i_sb->s_dev;
		ino = inode->i_ino;
		gen = inode->i_generation;
		maj = MAJOR(dev);
		min = MINOR(dev);
5825 5826 5827 5828 5829 5830 5831 5832 5833 5834 5835 5836 5837 5838 5839 5840 5841 5842 5843 5844 5845 5846

		if (vma->vm_flags & VM_READ)
			prot |= PROT_READ;
		if (vma->vm_flags & VM_WRITE)
			prot |= PROT_WRITE;
		if (vma->vm_flags & VM_EXEC)
			prot |= PROT_EXEC;

		if (vma->vm_flags & VM_MAYSHARE)
			flags = MAP_SHARED;
		else
			flags = MAP_PRIVATE;

		if (vma->vm_flags & VM_DENYWRITE)
			flags |= MAP_DENYWRITE;
		if (vma->vm_flags & VM_MAYEXEC)
			flags |= MAP_EXECUTABLE;
		if (vma->vm_flags & VM_LOCKED)
			flags |= MAP_LOCKED;
		if (vma->vm_flags & VM_HUGETLB)
			flags |= MAP_HUGETLB;

5847
		goto got_name;
5848
	} else {
5849 5850 5851 5852 5853 5854
		if (vma->vm_ops && vma->vm_ops->name) {
			name = (char *) vma->vm_ops->name(vma);
			if (name)
				goto cpy_name;
		}

5855
		name = (char *)arch_vma_name(vma);
5856 5857
		if (name)
			goto cpy_name;
5858

5859
		if (vma->vm_start <= vma->vm_mm->start_brk &&
5860
				vma->vm_end >= vma->vm_mm->brk) {
5861 5862
			name = "[heap]";
			goto cpy_name;
5863 5864
		}
		if (vma->vm_start <= vma->vm_mm->start_stack &&
5865
				vma->vm_end >= vma->vm_mm->start_stack) {
5866 5867
			name = "[stack]";
			goto cpy_name;
5868 5869
		}

5870 5871
		name = "//anon";
		goto cpy_name;
5872 5873
	}

5874 5875 5876
cpy_name:
	strlcpy(tmp, name, sizeof(tmp));
	name = tmp;
5877
got_name:
5878 5879 5880 5881 5882 5883 5884 5885
	/*
	 * Since our buffer works in 8 byte units we need to align our string
	 * size to a multiple of 8. However, we must guarantee the tail end is
	 * zero'd out to avoid leaking random bits to userspace.
	 */
	size = strlen(name)+1;
	while (!IS_ALIGNED(size, sizeof(u64)))
		name[size++] = '\0';
5886 5887 5888

	mmap_event->file_name = name;
	mmap_event->file_size = size;
5889 5890 5891 5892
	mmap_event->maj = maj;
	mmap_event->min = min;
	mmap_event->ino = ino;
	mmap_event->ino_generation = gen;
5893 5894
	mmap_event->prot = prot;
	mmap_event->flags = flags;
5895

5896 5897 5898
	if (!(vma->vm_flags & VM_EXEC))
		mmap_event->event_id.header.misc |= PERF_RECORD_MISC_MMAP_DATA;

5899
	mmap_event->event_id.header.size = sizeof(mmap_event->event_id) + size;
5900

5901
	perf_event_aux(perf_event_mmap_output,
5902 5903
		       mmap_event,
		       NULL);
5904

5905 5906 5907
	kfree(buf);
}

5908
void perf_event_mmap(struct vm_area_struct *vma)
5909
{
5910 5911
	struct perf_mmap_event mmap_event;

5912
	if (!atomic_read(&nr_mmap_events))
5913 5914 5915
		return;

	mmap_event = (struct perf_mmap_event){
5916
		.vma	= vma,
5917 5918
		/* .file_name */
		/* .file_size */
5919
		.event_id  = {
5920
			.header = {
5921
				.type = PERF_RECORD_MMAP,
5922
				.misc = PERF_RECORD_MISC_USER,
5923 5924 5925 5926
				/* .size */
			},
			/* .pid */
			/* .tid */
5927 5928
			.start  = vma->vm_start,
			.len    = vma->vm_end - vma->vm_start,
5929
			.pgoff  = (u64)vma->vm_pgoff << PAGE_SHIFT,
5930
		},
5931 5932 5933 5934
		/* .maj (attr_mmap2 only) */
		/* .min (attr_mmap2 only) */
		/* .ino (attr_mmap2 only) */
		/* .ino_generation (attr_mmap2 only) */
5935 5936
		/* .prot (attr_mmap2 only) */
		/* .flags (attr_mmap2 only) */
5937 5938
	};

5939
	perf_event_mmap_event(&mmap_event);
5940 5941
}

A
Alexander Shishkin 已提交
5942 5943 5944 5945 5946 5947 5948 5949 5950 5951 5952 5953 5954 5955 5956 5957 5958 5959 5960 5961 5962 5963 5964 5965 5966 5967 5968 5969 5970 5971 5972 5973 5974 5975
void perf_event_aux_event(struct perf_event *event, unsigned long head,
			  unsigned long size, u64 flags)
{
	struct perf_output_handle handle;
	struct perf_sample_data sample;
	struct perf_aux_event {
		struct perf_event_header	header;
		u64				offset;
		u64				size;
		u64				flags;
	} rec = {
		.header = {
			.type = PERF_RECORD_AUX,
			.misc = 0,
			.size = sizeof(rec),
		},
		.offset		= head,
		.size		= size,
		.flags		= flags,
	};
	int ret;

	perf_event_header__init_id(&rec.header, &sample, event);
	ret = perf_output_begin(&handle, event, rec.header.size);

	if (ret)
		return;

	perf_output_put(&handle, rec);
	perf_event__output_id_sample(event, &handle, &sample);

	perf_output_end(&handle);
}

5976 5977 5978 5979
/*
 * IRQ throttle logging
 */

5980
static void perf_log_throttle(struct perf_event *event, int enable)
5981 5982
{
	struct perf_output_handle handle;
5983
	struct perf_sample_data sample;
5984 5985 5986 5987 5988
	int ret;

	struct {
		struct perf_event_header	header;
		u64				time;
5989
		u64				id;
5990
		u64				stream_id;
5991 5992
	} throttle_event = {
		.header = {
5993
			.type = PERF_RECORD_THROTTLE,
5994 5995 5996
			.misc = 0,
			.size = sizeof(throttle_event),
		},
5997
		.time		= perf_event_clock(event),
5998 5999
		.id		= primary_event_id(event),
		.stream_id	= event->id,
6000 6001
	};

6002
	if (enable)
6003
		throttle_event.header.type = PERF_RECORD_UNTHROTTLE;
6004

6005 6006 6007
	perf_event_header__init_id(&throttle_event.header, &sample, event);

	ret = perf_output_begin(&handle, event,
6008
				throttle_event.header.size);
6009 6010 6011 6012
	if (ret)
		return;

	perf_output_put(&handle, throttle_event);
6013
	perf_event__output_id_sample(event, &handle, &sample);
6014 6015 6016
	perf_output_end(&handle);
}

6017 6018 6019 6020 6021 6022 6023 6024 6025 6026 6027 6028 6029 6030 6031 6032 6033 6034 6035 6036 6037 6038 6039 6040 6041 6042 6043 6044 6045 6046 6047 6048 6049 6050 6051 6052 6053 6054
static void perf_log_itrace_start(struct perf_event *event)
{
	struct perf_output_handle handle;
	struct perf_sample_data sample;
	struct perf_aux_event {
		struct perf_event_header        header;
		u32				pid;
		u32				tid;
	} rec;
	int ret;

	if (event->parent)
		event = event->parent;

	if (!(event->pmu->capabilities & PERF_PMU_CAP_ITRACE) ||
	    event->hw.itrace_started)
		return;

	event->hw.itrace_started = 1;

	rec.header.type	= PERF_RECORD_ITRACE_START;
	rec.header.misc	= 0;
	rec.header.size	= sizeof(rec);
	rec.pid	= perf_event_pid(event, current);
	rec.tid	= perf_event_tid(event, current);

	perf_event_header__init_id(&rec.header, &sample, event);
	ret = perf_output_begin(&handle, event, rec.header.size);

	if (ret)
		return;

	perf_output_put(&handle, rec);
	perf_event__output_id_sample(event, &handle, &sample);

	perf_output_end(&handle);
}

6055
/*
6056
 * Generic event overflow handling, sampling.
6057 6058
 */

6059
static int __perf_event_overflow(struct perf_event *event,
6060 6061
				   int throttle, struct perf_sample_data *data,
				   struct pt_regs *regs)
6062
{
6063 6064
	int events = atomic_read(&event->event_limit);
	struct hw_perf_event *hwc = &event->hw;
6065
	u64 seq;
6066 6067
	int ret = 0;

6068 6069 6070 6071 6072 6073 6074
	/*
	 * Non-sampling counters might still use the PMI to fold short
	 * hardware counters, ignore those.
	 */
	if (unlikely(!is_sampling_event(event)))
		return 0;

6075 6076 6077 6078 6079 6080 6081 6082 6083
	seq = __this_cpu_read(perf_throttled_seq);
	if (seq != hwc->interrupts_seq) {
		hwc->interrupts_seq = seq;
		hwc->interrupts = 1;
	} else {
		hwc->interrupts++;
		if (unlikely(throttle
			     && hwc->interrupts >= max_samples_per_tick)) {
			__this_cpu_inc(perf_throttled_count);
P
Peter Zijlstra 已提交
6084 6085
			hwc->interrupts = MAX_INTERRUPTS;
			perf_log_throttle(event, 0);
6086
			tick_nohz_full_kick();
6087 6088
			ret = 1;
		}
6089
	}
6090

6091
	if (event->attr.freq) {
P
Peter Zijlstra 已提交
6092
		u64 now = perf_clock();
6093
		s64 delta = now - hwc->freq_time_stamp;
6094

6095
		hwc->freq_time_stamp = now;
6096

6097
		if (delta > 0 && delta < 2*TICK_NSEC)
6098
			perf_adjust_period(event, delta, hwc->last_period, true);
6099 6100
	}

6101 6102
	/*
	 * XXX event_limit might not quite work as expected on inherited
6103
	 * events
6104 6105
	 */

6106 6107
	event->pending_kill = POLL_IN;
	if (events && atomic_dec_and_test(&event->event_limit)) {
6108
		ret = 1;
6109
		event->pending_kill = POLL_HUP;
6110 6111
		event->pending_disable = 1;
		irq_work_queue(&event->pending);
6112 6113
	}

6114
	if (event->overflow_handler)
6115
		event->overflow_handler(event, data, regs);
6116
	else
6117
		perf_event_output(event, data, regs);
6118

P
Peter Zijlstra 已提交
6119
	if (event->fasync && event->pending_kill) {
6120 6121
		event->pending_wakeup = 1;
		irq_work_queue(&event->pending);
P
Peter Zijlstra 已提交
6122 6123
	}

6124
	return ret;
6125 6126
}

6127
int perf_event_overflow(struct perf_event *event,
6128 6129
			  struct perf_sample_data *data,
			  struct pt_regs *regs)
6130
{
6131
	return __perf_event_overflow(event, 1, data, regs);
6132 6133
}

6134
/*
6135
 * Generic software event infrastructure
6136 6137
 */

6138 6139 6140 6141 6142 6143 6144
struct swevent_htable {
	struct swevent_hlist		*swevent_hlist;
	struct mutex			hlist_mutex;
	int				hlist_refcount;

	/* Recursion avoidance in each contexts */
	int				recursion[PERF_NR_CONTEXTS];
6145 6146 6147

	/* Keeps track of cpu being initialized/exited */
	bool				online;
6148 6149 6150 6151
};

static DEFINE_PER_CPU(struct swevent_htable, swevent_htable);

6152
/*
6153 6154
 * We directly increment event->count and keep a second value in
 * event->hw.period_left to count intervals. This period event
6155 6156 6157 6158
 * is kept in the range [-sample_period, 0] so that we can use the
 * sign as trigger.
 */

6159
u64 perf_swevent_set_period(struct perf_event *event)
6160
{
6161
	struct hw_perf_event *hwc = &event->hw;
6162 6163 6164 6165 6166
	u64 period = hwc->last_period;
	u64 nr, offset;
	s64 old, val;

	hwc->last_period = hwc->sample_period;
6167 6168

again:
6169
	old = val = local64_read(&hwc->period_left);
6170 6171
	if (val < 0)
		return 0;
6172

6173 6174 6175
	nr = div64_u64(period + val, period);
	offset = nr * period;
	val -= offset;
6176
	if (local64_cmpxchg(&hwc->period_left, old, val) != old)
6177
		goto again;
6178

6179
	return nr;
6180 6181
}

6182
static void perf_swevent_overflow(struct perf_event *event, u64 overflow,
6183
				    struct perf_sample_data *data,
6184
				    struct pt_regs *regs)
6185
{
6186
	struct hw_perf_event *hwc = &event->hw;
6187
	int throttle = 0;
6188

6189 6190
	if (!overflow)
		overflow = perf_swevent_set_period(event);
6191

6192 6193
	if (hwc->interrupts == MAX_INTERRUPTS)
		return;
6194

6195
	for (; overflow; overflow--) {
6196
		if (__perf_event_overflow(event, throttle,
6197
					    data, regs)) {
6198 6199 6200 6201 6202 6203
			/*
			 * We inhibit the overflow from happening when
			 * hwc->interrupts == MAX_INTERRUPTS.
			 */
			break;
		}
6204
		throttle = 1;
6205
	}
6206 6207
}

P
Peter Zijlstra 已提交
6208
static void perf_swevent_event(struct perf_event *event, u64 nr,
6209
			       struct perf_sample_data *data,
6210
			       struct pt_regs *regs)
6211
{
6212
	struct hw_perf_event *hwc = &event->hw;
6213

6214
	local64_add(nr, &event->count);
6215

6216 6217 6218
	if (!regs)
		return;

6219
	if (!is_sampling_event(event))
6220
		return;
6221

6222 6223 6224 6225 6226 6227
	if ((event->attr.sample_type & PERF_SAMPLE_PERIOD) && !event->attr.freq) {
		data->period = nr;
		return perf_swevent_overflow(event, 1, data, regs);
	} else
		data->period = event->hw.last_period;

6228
	if (nr == 1 && hwc->sample_period == 1 && !event->attr.freq)
6229
		return perf_swevent_overflow(event, 1, data, regs);
6230

6231
	if (local64_add_negative(nr, &hwc->period_left))
6232
		return;
6233

6234
	perf_swevent_overflow(event, 0, data, regs);
6235 6236
}

6237 6238 6239
static int perf_exclude_event(struct perf_event *event,
			      struct pt_regs *regs)
{
P
Peter Zijlstra 已提交
6240
	if (event->hw.state & PERF_HES_STOPPED)
6241
		return 1;
P
Peter Zijlstra 已提交
6242

6243 6244 6245 6246 6247 6248 6249 6250 6251 6252 6253
	if (regs) {
		if (event->attr.exclude_user && user_mode(regs))
			return 1;

		if (event->attr.exclude_kernel && !user_mode(regs))
			return 1;
	}

	return 0;
}

6254
static int perf_swevent_match(struct perf_event *event,
P
Peter Zijlstra 已提交
6255
				enum perf_type_id type,
L
Li Zefan 已提交
6256 6257 6258
				u32 event_id,
				struct perf_sample_data *data,
				struct pt_regs *regs)
6259
{
6260
	if (event->attr.type != type)
6261
		return 0;
6262

6263
	if (event->attr.config != event_id)
6264 6265
		return 0;

6266 6267
	if (perf_exclude_event(event, regs))
		return 0;
6268 6269 6270 6271

	return 1;
}

6272 6273 6274 6275 6276 6277 6278
static inline u64 swevent_hash(u64 type, u32 event_id)
{
	u64 val = event_id | (type << 32);

	return hash_64(val, SWEVENT_HLIST_BITS);
}

6279 6280
static inline struct hlist_head *
__find_swevent_head(struct swevent_hlist *hlist, u64 type, u32 event_id)
6281
{
6282 6283 6284 6285
	u64 hash = swevent_hash(type, event_id);

	return &hlist->heads[hash];
}
6286

6287 6288
/* For the read side: events when they trigger */
static inline struct hlist_head *
6289
find_swevent_head_rcu(struct swevent_htable *swhash, u64 type, u32 event_id)
6290 6291
{
	struct swevent_hlist *hlist;
6292

6293
	hlist = rcu_dereference(swhash->swevent_hlist);
6294 6295 6296
	if (!hlist)
		return NULL;

6297 6298 6299 6300 6301
	return __find_swevent_head(hlist, type, event_id);
}

/* For the event head insertion and removal in the hlist */
static inline struct hlist_head *
6302
find_swevent_head(struct swevent_htable *swhash, struct perf_event *event)
6303 6304 6305 6306 6307 6308 6309 6310 6311 6312
{
	struct swevent_hlist *hlist;
	u32 event_id = event->attr.config;
	u64 type = event->attr.type;

	/*
	 * Event scheduling is always serialized against hlist allocation
	 * and release. Which makes the protected version suitable here.
	 * The context lock guarantees that.
	 */
6313
	hlist = rcu_dereference_protected(swhash->swevent_hlist,
6314 6315 6316 6317 6318
					  lockdep_is_held(&event->ctx->lock));
	if (!hlist)
		return NULL;

	return __find_swevent_head(hlist, type, event_id);
6319 6320 6321
}

static void do_perf_sw_event(enum perf_type_id type, u32 event_id,
6322
				    u64 nr,
6323 6324
				    struct perf_sample_data *data,
				    struct pt_regs *regs)
6325
{
6326
	struct swevent_htable *swhash = this_cpu_ptr(&swevent_htable);
6327
	struct perf_event *event;
6328
	struct hlist_head *head;
6329

6330
	rcu_read_lock();
6331
	head = find_swevent_head_rcu(swhash, type, event_id);
6332 6333 6334
	if (!head)
		goto end;

6335
	hlist_for_each_entry_rcu(event, head, hlist_entry) {
L
Li Zefan 已提交
6336
		if (perf_swevent_match(event, type, event_id, data, regs))
6337
			perf_swevent_event(event, nr, data, regs);
6338
	}
6339 6340
end:
	rcu_read_unlock();
6341 6342
}

6343 6344
DEFINE_PER_CPU(struct pt_regs, __perf_regs[4]);

6345
int perf_swevent_get_recursion_context(void)
P
Peter Zijlstra 已提交
6346
{
6347
	struct swevent_htable *swhash = this_cpu_ptr(&swevent_htable);
P
Peter Zijlstra 已提交
6348

6349
	return get_recursion_context(swhash->recursion);
P
Peter Zijlstra 已提交
6350
}
I
Ingo Molnar 已提交
6351
EXPORT_SYMBOL_GPL(perf_swevent_get_recursion_context);
P
Peter Zijlstra 已提交
6352

6353
inline void perf_swevent_put_recursion_context(int rctx)
6354
{
6355
	struct swevent_htable *swhash = this_cpu_ptr(&swevent_htable);
6356

6357
	put_recursion_context(swhash->recursion, rctx);
6358
}
6359

6360
void ___perf_sw_event(u32 event_id, u64 nr, struct pt_regs *regs, u64 addr)
6361
{
6362
	struct perf_sample_data data;
6363

6364
	if (WARN_ON_ONCE(!regs))
6365
		return;
6366

6367
	perf_sample_data_init(&data, addr, 0);
6368
	do_perf_sw_event(PERF_TYPE_SOFTWARE, event_id, nr, &data, regs);
6369 6370 6371 6372 6373 6374 6375 6376 6377 6378 6379 6380
}

void __perf_sw_event(u32 event_id, u64 nr, struct pt_regs *regs, u64 addr)
{
	int rctx;

	preempt_disable_notrace();
	rctx = perf_swevent_get_recursion_context();
	if (unlikely(rctx < 0))
		goto fail;

	___perf_sw_event(event_id, nr, regs, addr);
6381 6382

	perf_swevent_put_recursion_context(rctx);
6383
fail:
6384
	preempt_enable_notrace();
6385 6386
}

6387
static void perf_swevent_read(struct perf_event *event)
6388 6389 6390
{
}

P
Peter Zijlstra 已提交
6391
static int perf_swevent_add(struct perf_event *event, int flags)
6392
{
6393
	struct swevent_htable *swhash = this_cpu_ptr(&swevent_htable);
6394
	struct hw_perf_event *hwc = &event->hw;
6395 6396
	struct hlist_head *head;

6397
	if (is_sampling_event(event)) {
6398
		hwc->last_period = hwc->sample_period;
6399
		perf_swevent_set_period(event);
6400
	}
6401

P
Peter Zijlstra 已提交
6402 6403
	hwc->state = !(flags & PERF_EF_START);

6404
	head = find_swevent_head(swhash, event);
6405 6406 6407 6408 6409 6410
	if (!head) {
		/*
		 * We can race with cpu hotplug code. Do not
		 * WARN if the cpu just got unplugged.
		 */
		WARN_ON_ONCE(swhash->online);
6411
		return -EINVAL;
6412
	}
6413 6414

	hlist_add_head_rcu(&event->hlist_entry, head);
6415
	perf_event_update_userpage(event);
6416

6417 6418 6419
	return 0;
}

P
Peter Zijlstra 已提交
6420
static void perf_swevent_del(struct perf_event *event, int flags)
6421
{
6422
	hlist_del_rcu(&event->hlist_entry);
6423 6424
}

P
Peter Zijlstra 已提交
6425
static void perf_swevent_start(struct perf_event *event, int flags)
6426
{
P
Peter Zijlstra 已提交
6427
	event->hw.state = 0;
6428
}
I
Ingo Molnar 已提交
6429

P
Peter Zijlstra 已提交
6430
static void perf_swevent_stop(struct perf_event *event, int flags)
6431
{
P
Peter Zijlstra 已提交
6432
	event->hw.state = PERF_HES_STOPPED;
6433 6434
}

6435 6436
/* Deref the hlist from the update side */
static inline struct swevent_hlist *
6437
swevent_hlist_deref(struct swevent_htable *swhash)
6438
{
6439 6440
	return rcu_dereference_protected(swhash->swevent_hlist,
					 lockdep_is_held(&swhash->hlist_mutex));
6441 6442
}

6443
static void swevent_hlist_release(struct swevent_htable *swhash)
6444
{
6445
	struct swevent_hlist *hlist = swevent_hlist_deref(swhash);
6446

6447
	if (!hlist)
6448 6449
		return;

6450
	RCU_INIT_POINTER(swhash->swevent_hlist, NULL);
6451
	kfree_rcu(hlist, rcu_head);
6452 6453 6454 6455
}

static void swevent_hlist_put_cpu(struct perf_event *event, int cpu)
{
6456
	struct swevent_htable *swhash = &per_cpu(swevent_htable, cpu);
6457

6458
	mutex_lock(&swhash->hlist_mutex);
6459

6460 6461
	if (!--swhash->hlist_refcount)
		swevent_hlist_release(swhash);
6462

6463
	mutex_unlock(&swhash->hlist_mutex);
6464 6465 6466 6467 6468 6469 6470 6471 6472 6473 6474 6475
}

static void swevent_hlist_put(struct perf_event *event)
{
	int cpu;

	for_each_possible_cpu(cpu)
		swevent_hlist_put_cpu(event, cpu);
}

static int swevent_hlist_get_cpu(struct perf_event *event, int cpu)
{
6476
	struct swevent_htable *swhash = &per_cpu(swevent_htable, cpu);
6477 6478
	int err = 0;

6479
	mutex_lock(&swhash->hlist_mutex);
6480

6481
	if (!swevent_hlist_deref(swhash) && cpu_online(cpu)) {
6482 6483 6484 6485 6486 6487 6488
		struct swevent_hlist *hlist;

		hlist = kzalloc(sizeof(*hlist), GFP_KERNEL);
		if (!hlist) {
			err = -ENOMEM;
			goto exit;
		}
6489
		rcu_assign_pointer(swhash->swevent_hlist, hlist);
6490
	}
6491
	swhash->hlist_refcount++;
P
Peter Zijlstra 已提交
6492
exit:
6493
	mutex_unlock(&swhash->hlist_mutex);
6494 6495 6496 6497 6498 6499 6500 6501 6502 6503 6504 6505 6506 6507 6508 6509 6510 6511 6512 6513

	return err;
}

static int swevent_hlist_get(struct perf_event *event)
{
	int err;
	int cpu, failed_cpu;

	get_online_cpus();
	for_each_possible_cpu(cpu) {
		err = swevent_hlist_get_cpu(event, cpu);
		if (err) {
			failed_cpu = cpu;
			goto fail;
		}
	}
	put_online_cpus();

	return 0;
P
Peter Zijlstra 已提交
6514
fail:
6515 6516 6517 6518 6519 6520 6521 6522 6523 6524
	for_each_possible_cpu(cpu) {
		if (cpu == failed_cpu)
			break;
		swevent_hlist_put_cpu(event, cpu);
	}

	put_online_cpus();
	return err;
}

6525
struct static_key perf_swevent_enabled[PERF_COUNT_SW_MAX];
6526

6527 6528 6529
static void sw_perf_event_destroy(struct perf_event *event)
{
	u64 event_id = event->attr.config;
6530

6531 6532
	WARN_ON(event->parent);

6533
	static_key_slow_dec(&perf_swevent_enabled[event_id]);
6534 6535 6536 6537 6538
	swevent_hlist_put(event);
}

static int perf_swevent_init(struct perf_event *event)
{
6539
	u64 event_id = event->attr.config;
6540 6541 6542 6543

	if (event->attr.type != PERF_TYPE_SOFTWARE)
		return -ENOENT;

6544 6545 6546 6547 6548 6549
	/*
	 * no branch sampling for software events
	 */
	if (has_branch_stack(event))
		return -EOPNOTSUPP;

6550 6551 6552 6553 6554 6555 6556 6557 6558
	switch (event_id) {
	case PERF_COUNT_SW_CPU_CLOCK:
	case PERF_COUNT_SW_TASK_CLOCK:
		return -ENOENT;

	default:
		break;
	}

6559
	if (event_id >= PERF_COUNT_SW_MAX)
6560 6561 6562 6563 6564 6565 6566 6567 6568
		return -ENOENT;

	if (!event->parent) {
		int err;

		err = swevent_hlist_get(event);
		if (err)
			return err;

6569
		static_key_slow_inc(&perf_swevent_enabled[event_id]);
6570 6571 6572 6573 6574 6575 6576
		event->destroy = sw_perf_event_destroy;
	}

	return 0;
}

static struct pmu perf_swevent = {
6577
	.task_ctx_nr	= perf_sw_context,
6578

6579 6580
	.capabilities	= PERF_PMU_CAP_NO_NMI,

6581
	.event_init	= perf_swevent_init,
P
Peter Zijlstra 已提交
6582 6583 6584 6585
	.add		= perf_swevent_add,
	.del		= perf_swevent_del,
	.start		= perf_swevent_start,
	.stop		= perf_swevent_stop,
6586 6587 6588
	.read		= perf_swevent_read,
};

6589 6590
#ifdef CONFIG_EVENT_TRACING

6591 6592 6593 6594 6595 6596 6597 6598 6599 6600 6601 6602 6603 6604
static int perf_tp_filter_match(struct perf_event *event,
				struct perf_sample_data *data)
{
	void *record = data->raw->data;

	if (likely(!event->filter) || filter_match_preds(event->filter, record))
		return 1;
	return 0;
}

static int perf_tp_event_match(struct perf_event *event,
				struct perf_sample_data *data,
				struct pt_regs *regs)
{
6605 6606
	if (event->hw.state & PERF_HES_STOPPED)
		return 0;
6607 6608 6609 6610
	/*
	 * All tracepoints are from kernel-space.
	 */
	if (event->attr.exclude_kernel)
6611 6612 6613 6614 6615 6616 6617 6618 6619
		return 0;

	if (!perf_tp_filter_match(event, data))
		return 0;

	return 1;
}

void perf_tp_event(u64 addr, u64 count, void *record, int entry_size,
6620 6621
		   struct pt_regs *regs, struct hlist_head *head, int rctx,
		   struct task_struct *task)
6622 6623
{
	struct perf_sample_data data;
6624 6625
	struct perf_event *event;

6626 6627 6628 6629 6630
	struct perf_raw_record raw = {
		.size = entry_size,
		.data = record,
	};

6631
	perf_sample_data_init(&data, addr, 0);
6632 6633
	data.raw = &raw;

6634
	hlist_for_each_entry_rcu(event, head, hlist_entry) {
6635
		if (perf_tp_event_match(event, &data, regs))
6636
			perf_swevent_event(event, count, &data, regs);
6637
	}
6638

6639 6640 6641 6642 6643 6644 6645 6646 6647 6648 6649 6650 6651 6652 6653 6654 6655 6656 6657 6658 6659 6660 6661 6662 6663
	/*
	 * If we got specified a target task, also iterate its context and
	 * deliver this event there too.
	 */
	if (task && task != current) {
		struct perf_event_context *ctx;
		struct trace_entry *entry = record;

		rcu_read_lock();
		ctx = rcu_dereference(task->perf_event_ctxp[perf_sw_context]);
		if (!ctx)
			goto unlock;

		list_for_each_entry_rcu(event, &ctx->event_list, event_entry) {
			if (event->attr.type != PERF_TYPE_TRACEPOINT)
				continue;
			if (event->attr.config != entry->type)
				continue;
			if (perf_tp_event_match(event, &data, regs))
				perf_swevent_event(event, count, &data, regs);
		}
unlock:
		rcu_read_unlock();
	}

6664
	perf_swevent_put_recursion_context(rctx);
6665 6666 6667
}
EXPORT_SYMBOL_GPL(perf_tp_event);

6668
static void tp_perf_event_destroy(struct perf_event *event)
6669
{
6670
	perf_trace_destroy(event);
6671 6672
}

6673
static int perf_tp_event_init(struct perf_event *event)
6674
{
6675 6676
	int err;

6677 6678 6679
	if (event->attr.type != PERF_TYPE_TRACEPOINT)
		return -ENOENT;

6680 6681 6682 6683 6684 6685
	/*
	 * no branch sampling for tracepoint events
	 */
	if (has_branch_stack(event))
		return -EOPNOTSUPP;

6686 6687
	err = perf_trace_init(event);
	if (err)
6688
		return err;
6689

6690
	event->destroy = tp_perf_event_destroy;
6691

6692 6693 6694 6695
	return 0;
}

static struct pmu perf_tracepoint = {
6696 6697
	.task_ctx_nr	= perf_sw_context,

6698
	.event_init	= perf_tp_event_init,
P
Peter Zijlstra 已提交
6699 6700 6701 6702
	.add		= perf_trace_add,
	.del		= perf_trace_del,
	.start		= perf_swevent_start,
	.stop		= perf_swevent_stop,
6703 6704 6705 6706 6707
	.read		= perf_swevent_read,
};

static inline void perf_tp_register(void)
{
P
Peter Zijlstra 已提交
6708
	perf_pmu_register(&perf_tracepoint, "tracepoint", PERF_TYPE_TRACEPOINT);
6709
}
L
Li Zefan 已提交
6710 6711 6712 6713 6714 6715 6716 6717 6718 6719 6720 6721 6722 6723 6724 6725 6726 6727 6728 6729 6730 6731 6732 6733

static int perf_event_set_filter(struct perf_event *event, void __user *arg)
{
	char *filter_str;
	int ret;

	if (event->attr.type != PERF_TYPE_TRACEPOINT)
		return -EINVAL;

	filter_str = strndup_user(arg, PAGE_SIZE);
	if (IS_ERR(filter_str))
		return PTR_ERR(filter_str);

	ret = ftrace_profile_set_filter(event, event->attr.config, filter_str);

	kfree(filter_str);
	return ret;
}

static void perf_event_free_filter(struct perf_event *event)
{
	ftrace_profile_free_filter(event);
}

6734 6735 6736 6737 6738 6739 6740 6741 6742 6743 6744 6745 6746 6747 6748 6749 6750 6751
static int perf_event_set_bpf_prog(struct perf_event *event, u32 prog_fd)
{
	struct bpf_prog *prog;

	if (event->attr.type != PERF_TYPE_TRACEPOINT)
		return -EINVAL;

	if (event->tp_event->prog)
		return -EEXIST;

	if (!(event->tp_event->flags & TRACE_EVENT_FL_KPROBE))
		/* bpf programs can only be attached to kprobes */
		return -EINVAL;

	prog = bpf_prog_get(prog_fd);
	if (IS_ERR(prog))
		return PTR_ERR(prog);

6752
	if (prog->type != BPF_PROG_TYPE_KPROBE) {
6753 6754 6755 6756 6757 6758 6759 6760 6761 6762 6763 6764 6765 6766 6767 6768 6769 6770 6771 6772 6773 6774 6775 6776
		/* valid fd, but invalid bpf program type */
		bpf_prog_put(prog);
		return -EINVAL;
	}

	event->tp_event->prog = prog;

	return 0;
}

static void perf_event_free_bpf_prog(struct perf_event *event)
{
	struct bpf_prog *prog;

	if (!event->tp_event)
		return;

	prog = event->tp_event->prog;
	if (prog) {
		event->tp_event->prog = NULL;
		bpf_prog_put(prog);
	}
}

6777
#else
L
Li Zefan 已提交
6778

6779
static inline void perf_tp_register(void)
6780 6781
{
}
L
Li Zefan 已提交
6782 6783 6784 6785 6786 6787 6788 6789 6790 6791

static int perf_event_set_filter(struct perf_event *event, void __user *arg)
{
	return -ENOENT;
}

static void perf_event_free_filter(struct perf_event *event)
{
}

6792 6793 6794 6795 6796 6797 6798 6799
static int perf_event_set_bpf_prog(struct perf_event *event, u32 prog_fd)
{
	return -ENOENT;
}

static void perf_event_free_bpf_prog(struct perf_event *event)
{
}
6800
#endif /* CONFIG_EVENT_TRACING */
6801

6802
#ifdef CONFIG_HAVE_HW_BREAKPOINT
6803
void perf_bp_event(struct perf_event *bp, void *data)
6804
{
6805 6806 6807
	struct perf_sample_data sample;
	struct pt_regs *regs = data;

6808
	perf_sample_data_init(&sample, bp->attr.bp_addr, 0);
6809

P
Peter Zijlstra 已提交
6810
	if (!bp->hw.state && !perf_exclude_event(bp, regs))
6811
		perf_swevent_event(bp, 1, &sample, regs);
6812 6813 6814
}
#endif

6815 6816 6817
/*
 * hrtimer based swevent callback
 */
6818

6819
static enum hrtimer_restart perf_swevent_hrtimer(struct hrtimer *hrtimer)
6820
{
6821 6822 6823 6824 6825
	enum hrtimer_restart ret = HRTIMER_RESTART;
	struct perf_sample_data data;
	struct pt_regs *regs;
	struct perf_event *event;
	u64 period;
6826

6827
	event = container_of(hrtimer, struct perf_event, hw.hrtimer);
P
Peter Zijlstra 已提交
6828 6829 6830 6831

	if (event->state != PERF_EVENT_STATE_ACTIVE)
		return HRTIMER_NORESTART;

6832
	event->pmu->read(event);
6833

6834
	perf_sample_data_init(&data, 0, event->hw.last_period);
6835 6836 6837
	regs = get_irq_regs();

	if (regs && !perf_exclude_event(event, regs)) {
6838
		if (!(event->attr.exclude_idle && is_idle_task(current)))
6839
			if (__perf_event_overflow(event, 1, &data, regs))
6840 6841
				ret = HRTIMER_NORESTART;
	}
6842

6843 6844
	period = max_t(u64, 10000, event->hw.sample_period);
	hrtimer_forward_now(hrtimer, ns_to_ktime(period));
6845

6846
	return ret;
6847 6848
}

6849
static void perf_swevent_start_hrtimer(struct perf_event *event)
6850
{
6851
	struct hw_perf_event *hwc = &event->hw;
6852 6853 6854 6855
	s64 period;

	if (!is_sampling_event(event))
		return;
6856

6857 6858 6859 6860
	period = local64_read(&hwc->period_left);
	if (period) {
		if (period < 0)
			period = 10000;
P
Peter Zijlstra 已提交
6861

6862 6863 6864 6865 6866
		local64_set(&hwc->period_left, 0);
	} else {
		period = max_t(u64, 10000, hwc->sample_period);
	}
	__hrtimer_start_range_ns(&hwc->hrtimer,
6867
				ns_to_ktime(period), 0,
6868
				HRTIMER_MODE_REL_PINNED, 0);
6869
}
6870 6871

static void perf_swevent_cancel_hrtimer(struct perf_event *event)
6872
{
6873 6874
	struct hw_perf_event *hwc = &event->hw;

6875
	if (is_sampling_event(event)) {
6876
		ktime_t remaining = hrtimer_get_remaining(&hwc->hrtimer);
P
Peter Zijlstra 已提交
6877
		local64_set(&hwc->period_left, ktime_to_ns(remaining));
6878 6879 6880

		hrtimer_cancel(&hwc->hrtimer);
	}
6881 6882
}

P
Peter Zijlstra 已提交
6883 6884 6885 6886 6887 6888 6889 6890 6891 6892 6893 6894 6895 6896 6897 6898 6899 6900 6901 6902
static void perf_swevent_init_hrtimer(struct perf_event *event)
{
	struct hw_perf_event *hwc = &event->hw;

	if (!is_sampling_event(event))
		return;

	hrtimer_init(&hwc->hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
	hwc->hrtimer.function = perf_swevent_hrtimer;

	/*
	 * Since hrtimers have a fixed rate, we can do a static freq->period
	 * mapping and avoid the whole period adjust feedback stuff.
	 */
	if (event->attr.freq) {
		long freq = event->attr.sample_freq;

		event->attr.sample_period = NSEC_PER_SEC / freq;
		hwc->sample_period = event->attr.sample_period;
		local64_set(&hwc->period_left, hwc->sample_period);
6903
		hwc->last_period = hwc->sample_period;
P
Peter Zijlstra 已提交
6904 6905 6906 6907
		event->attr.freq = 0;
	}
}

6908 6909 6910 6911 6912
/*
 * Software event: cpu wall time clock
 */

static void cpu_clock_event_update(struct perf_event *event)
6913
{
6914 6915 6916
	s64 prev;
	u64 now;

P
Peter Zijlstra 已提交
6917
	now = local_clock();
6918 6919
	prev = local64_xchg(&event->hw.prev_count, now);
	local64_add(now - prev, &event->count);
6920 6921
}

P
Peter Zijlstra 已提交
6922
static void cpu_clock_event_start(struct perf_event *event, int flags)
6923
{
P
Peter Zijlstra 已提交
6924
	local64_set(&event->hw.prev_count, local_clock());
6925 6926 6927
	perf_swevent_start_hrtimer(event);
}

P
Peter Zijlstra 已提交
6928
static void cpu_clock_event_stop(struct perf_event *event, int flags)
6929
{
6930 6931 6932
	perf_swevent_cancel_hrtimer(event);
	cpu_clock_event_update(event);
}
6933

P
Peter Zijlstra 已提交
6934 6935 6936 6937
static int cpu_clock_event_add(struct perf_event *event, int flags)
{
	if (flags & PERF_EF_START)
		cpu_clock_event_start(event, flags);
6938
	perf_event_update_userpage(event);
P
Peter Zijlstra 已提交
6939 6940 6941 6942 6943 6944 6945 6946 6947

	return 0;
}

static void cpu_clock_event_del(struct perf_event *event, int flags)
{
	cpu_clock_event_stop(event, flags);
}

6948 6949 6950 6951
static void cpu_clock_event_read(struct perf_event *event)
{
	cpu_clock_event_update(event);
}
6952

6953 6954 6955 6956 6957 6958 6959 6960
static int cpu_clock_event_init(struct perf_event *event)
{
	if (event->attr.type != PERF_TYPE_SOFTWARE)
		return -ENOENT;

	if (event->attr.config != PERF_COUNT_SW_CPU_CLOCK)
		return -ENOENT;

6961 6962 6963 6964 6965 6966
	/*
	 * no branch sampling for software events
	 */
	if (has_branch_stack(event))
		return -EOPNOTSUPP;

P
Peter Zijlstra 已提交
6967 6968
	perf_swevent_init_hrtimer(event);

6969
	return 0;
6970 6971
}

6972
static struct pmu perf_cpu_clock = {
6973 6974
	.task_ctx_nr	= perf_sw_context,

6975 6976
	.capabilities	= PERF_PMU_CAP_NO_NMI,

6977
	.event_init	= cpu_clock_event_init,
P
Peter Zijlstra 已提交
6978 6979 6980 6981
	.add		= cpu_clock_event_add,
	.del		= cpu_clock_event_del,
	.start		= cpu_clock_event_start,
	.stop		= cpu_clock_event_stop,
6982 6983 6984 6985 6986 6987 6988 6989
	.read		= cpu_clock_event_read,
};

/*
 * Software event: task time clock
 */

static void task_clock_event_update(struct perf_event *event, u64 now)
6990
{
6991 6992
	u64 prev;
	s64 delta;
6993

6994 6995 6996 6997
	prev = local64_xchg(&event->hw.prev_count, now);
	delta = now - prev;
	local64_add(delta, &event->count);
}
6998

P
Peter Zijlstra 已提交
6999
static void task_clock_event_start(struct perf_event *event, int flags)
7000
{
P
Peter Zijlstra 已提交
7001
	local64_set(&event->hw.prev_count, event->ctx->time);
7002 7003 7004
	perf_swevent_start_hrtimer(event);
}

P
Peter Zijlstra 已提交
7005
static void task_clock_event_stop(struct perf_event *event, int flags)
7006 7007 7008
{
	perf_swevent_cancel_hrtimer(event);
	task_clock_event_update(event, event->ctx->time);
P
Peter Zijlstra 已提交
7009 7010 7011 7012 7013 7014
}

static int task_clock_event_add(struct perf_event *event, int flags)
{
	if (flags & PERF_EF_START)
		task_clock_event_start(event, flags);
7015
	perf_event_update_userpage(event);
7016

P
Peter Zijlstra 已提交
7017 7018 7019 7020 7021 7022
	return 0;
}

static void task_clock_event_del(struct perf_event *event, int flags)
{
	task_clock_event_stop(event, PERF_EF_UPDATE);
7023 7024 7025 7026
}

static void task_clock_event_read(struct perf_event *event)
{
7027 7028 7029
	u64 now = perf_clock();
	u64 delta = now - event->ctx->timestamp;
	u64 time = event->ctx->time + delta;
7030 7031 7032 7033 7034

	task_clock_event_update(event, time);
}

static int task_clock_event_init(struct perf_event *event)
L
Li Zefan 已提交
7035
{
7036 7037 7038 7039 7040 7041
	if (event->attr.type != PERF_TYPE_SOFTWARE)
		return -ENOENT;

	if (event->attr.config != PERF_COUNT_SW_TASK_CLOCK)
		return -ENOENT;

7042 7043 7044 7045 7046 7047
	/*
	 * no branch sampling for software events
	 */
	if (has_branch_stack(event))
		return -EOPNOTSUPP;

P
Peter Zijlstra 已提交
7048 7049
	perf_swevent_init_hrtimer(event);

7050
	return 0;
L
Li Zefan 已提交
7051 7052
}

7053
static struct pmu perf_task_clock = {
7054 7055
	.task_ctx_nr	= perf_sw_context,

7056 7057
	.capabilities	= PERF_PMU_CAP_NO_NMI,

7058
	.event_init	= task_clock_event_init,
P
Peter Zijlstra 已提交
7059 7060 7061 7062
	.add		= task_clock_event_add,
	.del		= task_clock_event_del,
	.start		= task_clock_event_start,
	.stop		= task_clock_event_stop,
7063 7064
	.read		= task_clock_event_read,
};
L
Li Zefan 已提交
7065

P
Peter Zijlstra 已提交
7066
static void perf_pmu_nop_void(struct pmu *pmu)
7067 7068
{
}
L
Li Zefan 已提交
7069

P
Peter Zijlstra 已提交
7070
static int perf_pmu_nop_int(struct pmu *pmu)
L
Li Zefan 已提交
7071
{
P
Peter Zijlstra 已提交
7072
	return 0;
L
Li Zefan 已提交
7073 7074
}

P
Peter Zijlstra 已提交
7075
static void perf_pmu_start_txn(struct pmu *pmu)
L
Li Zefan 已提交
7076
{
P
Peter Zijlstra 已提交
7077
	perf_pmu_disable(pmu);
L
Li Zefan 已提交
7078 7079
}

P
Peter Zijlstra 已提交
7080 7081 7082 7083 7084
static int perf_pmu_commit_txn(struct pmu *pmu)
{
	perf_pmu_enable(pmu);
	return 0;
}
7085

P
Peter Zijlstra 已提交
7086
static void perf_pmu_cancel_txn(struct pmu *pmu)
7087
{
P
Peter Zijlstra 已提交
7088
	perf_pmu_enable(pmu);
7089 7090
}

7091 7092
static int perf_event_idx_default(struct perf_event *event)
{
7093
	return 0;
7094 7095
}

P
Peter Zijlstra 已提交
7096 7097 7098 7099
/*
 * Ensures all contexts with the same task_ctx_nr have the same
 * pmu_cpu_context too.
 */
7100
static struct perf_cpu_context __percpu *find_pmu_context(int ctxn)
7101
{
P
Peter Zijlstra 已提交
7102
	struct pmu *pmu;
7103

P
Peter Zijlstra 已提交
7104 7105
	if (ctxn < 0)
		return NULL;
7106

P
Peter Zijlstra 已提交
7107 7108 7109 7110
	list_for_each_entry(pmu, &pmus, entry) {
		if (pmu->task_ctx_nr == ctxn)
			return pmu->pmu_cpu_context;
	}
7111

P
Peter Zijlstra 已提交
7112
	return NULL;
7113 7114
}

7115
static void update_pmu_context(struct pmu *pmu, struct pmu *old_pmu)
7116
{
7117 7118 7119 7120 7121 7122 7123
	int cpu;

	for_each_possible_cpu(cpu) {
		struct perf_cpu_context *cpuctx;

		cpuctx = per_cpu_ptr(pmu->pmu_cpu_context, cpu);

7124 7125
		if (cpuctx->unique_pmu == old_pmu)
			cpuctx->unique_pmu = pmu;
7126 7127 7128 7129 7130 7131
	}
}

static void free_pmu_context(struct pmu *pmu)
{
	struct pmu *i;
7132

P
Peter Zijlstra 已提交
7133
	mutex_lock(&pmus_lock);
7134
	/*
P
Peter Zijlstra 已提交
7135
	 * Like a real lame refcount.
7136
	 */
7137 7138 7139
	list_for_each_entry(i, &pmus, entry) {
		if (i->pmu_cpu_context == pmu->pmu_cpu_context) {
			update_pmu_context(i, pmu);
P
Peter Zijlstra 已提交
7140
			goto out;
7141
		}
P
Peter Zijlstra 已提交
7142
	}
7143

7144
	free_percpu(pmu->pmu_cpu_context);
P
Peter Zijlstra 已提交
7145 7146
out:
	mutex_unlock(&pmus_lock);
7147
}
P
Peter Zijlstra 已提交
7148
static struct idr pmu_idr;
7149

P
Peter Zijlstra 已提交
7150 7151 7152 7153 7154 7155 7156
static ssize_t
type_show(struct device *dev, struct device_attribute *attr, char *page)
{
	struct pmu *pmu = dev_get_drvdata(dev);

	return snprintf(page, PAGE_SIZE-1, "%d\n", pmu->type);
}
7157
static DEVICE_ATTR_RO(type);
P
Peter Zijlstra 已提交
7158

7159 7160 7161 7162 7163 7164 7165 7166 7167 7168 7169 7170 7171 7172 7173 7174 7175 7176 7177 7178 7179 7180 7181 7182 7183 7184 7185 7186 7187 7188 7189 7190 7191 7192 7193 7194 7195 7196 7197 7198 7199 7200 7201
static ssize_t
perf_event_mux_interval_ms_show(struct device *dev,
				struct device_attribute *attr,
				char *page)
{
	struct pmu *pmu = dev_get_drvdata(dev);

	return snprintf(page, PAGE_SIZE-1, "%d\n", pmu->hrtimer_interval_ms);
}

static ssize_t
perf_event_mux_interval_ms_store(struct device *dev,
				 struct device_attribute *attr,
				 const char *buf, size_t count)
{
	struct pmu *pmu = dev_get_drvdata(dev);
	int timer, cpu, ret;

	ret = kstrtoint(buf, 0, &timer);
	if (ret)
		return ret;

	if (timer < 1)
		return -EINVAL;

	/* same value, noting to do */
	if (timer == pmu->hrtimer_interval_ms)
		return count;

	pmu->hrtimer_interval_ms = timer;

	/* update all cpuctx for this PMU */
	for_each_possible_cpu(cpu) {
		struct perf_cpu_context *cpuctx;
		cpuctx = per_cpu_ptr(pmu->pmu_cpu_context, cpu);
		cpuctx->hrtimer_interval = ns_to_ktime(NSEC_PER_MSEC * timer);

		if (hrtimer_active(&cpuctx->hrtimer))
			hrtimer_forward_now(&cpuctx->hrtimer, cpuctx->hrtimer_interval);
	}

	return count;
}
7202
static DEVICE_ATTR_RW(perf_event_mux_interval_ms);
7203

7204 7205 7206 7207
static struct attribute *pmu_dev_attrs[] = {
	&dev_attr_type.attr,
	&dev_attr_perf_event_mux_interval_ms.attr,
	NULL,
P
Peter Zijlstra 已提交
7208
};
7209
ATTRIBUTE_GROUPS(pmu_dev);
P
Peter Zijlstra 已提交
7210 7211 7212 7213

static int pmu_bus_running;
static struct bus_type pmu_bus = {
	.name		= "event_source",
7214
	.dev_groups	= pmu_dev_groups,
P
Peter Zijlstra 已提交
7215 7216 7217 7218 7219 7220 7221 7222 7223 7224 7225 7226 7227 7228 7229
};

static void pmu_dev_release(struct device *dev)
{
	kfree(dev);
}

static int pmu_dev_alloc(struct pmu *pmu)
{
	int ret = -ENOMEM;

	pmu->dev = kzalloc(sizeof(struct device), GFP_KERNEL);
	if (!pmu->dev)
		goto out;

7230
	pmu->dev->groups = pmu->attr_groups;
P
Peter Zijlstra 已提交
7231 7232 7233 7234 7235 7236 7237 7238 7239 7240 7241 7242 7243 7244 7245 7246 7247 7248 7249 7250
	device_initialize(pmu->dev);
	ret = dev_set_name(pmu->dev, "%s", pmu->name);
	if (ret)
		goto free_dev;

	dev_set_drvdata(pmu->dev, pmu);
	pmu->dev->bus = &pmu_bus;
	pmu->dev->release = pmu_dev_release;
	ret = device_add(pmu->dev);
	if (ret)
		goto free_dev;

out:
	return ret;

free_dev:
	put_device(pmu->dev);
	goto out;
}

7251
static struct lock_class_key cpuctx_mutex;
7252
static struct lock_class_key cpuctx_lock;
7253

7254
int perf_pmu_register(struct pmu *pmu, const char *name, int type)
7255
{
P
Peter Zijlstra 已提交
7256
	int cpu, ret;
7257

7258
	mutex_lock(&pmus_lock);
P
Peter Zijlstra 已提交
7259 7260 7261 7262
	ret = -ENOMEM;
	pmu->pmu_disable_count = alloc_percpu(int);
	if (!pmu->pmu_disable_count)
		goto unlock;
7263

P
Peter Zijlstra 已提交
7264 7265 7266 7267 7268 7269
	pmu->type = -1;
	if (!name)
		goto skip_type;
	pmu->name = name;

	if (type < 0) {
T
Tejun Heo 已提交
7270 7271 7272
		type = idr_alloc(&pmu_idr, pmu, PERF_TYPE_MAX, 0, GFP_KERNEL);
		if (type < 0) {
			ret = type;
P
Peter Zijlstra 已提交
7273 7274 7275 7276 7277
			goto free_pdc;
		}
	}
	pmu->type = type;

P
Peter Zijlstra 已提交
7278 7279 7280 7281 7282 7283
	if (pmu_bus_running) {
		ret = pmu_dev_alloc(pmu);
		if (ret)
			goto free_idr;
	}

P
Peter Zijlstra 已提交
7284
skip_type:
P
Peter Zijlstra 已提交
7285 7286 7287
	pmu->pmu_cpu_context = find_pmu_context(pmu->task_ctx_nr);
	if (pmu->pmu_cpu_context)
		goto got_cpu_context;
7288

W
Wei Yongjun 已提交
7289
	ret = -ENOMEM;
P
Peter Zijlstra 已提交
7290 7291
	pmu->pmu_cpu_context = alloc_percpu(struct perf_cpu_context);
	if (!pmu->pmu_cpu_context)
P
Peter Zijlstra 已提交
7292
		goto free_dev;
7293

P
Peter Zijlstra 已提交
7294 7295 7296 7297
	for_each_possible_cpu(cpu) {
		struct perf_cpu_context *cpuctx;

		cpuctx = per_cpu_ptr(pmu->pmu_cpu_context, cpu);
7298
		__perf_event_init_context(&cpuctx->ctx);
7299
		lockdep_set_class(&cpuctx->ctx.mutex, &cpuctx_mutex);
7300
		lockdep_set_class(&cpuctx->ctx.lock, &cpuctx_lock);
P
Peter Zijlstra 已提交
7301
		cpuctx->ctx.pmu = pmu;
7302 7303 7304

		__perf_cpu_hrtimer_init(cpuctx, cpu);

7305
		cpuctx->unique_pmu = pmu;
P
Peter Zijlstra 已提交
7306
	}
7307

P
Peter Zijlstra 已提交
7308
got_cpu_context:
P
Peter Zijlstra 已提交
7309 7310 7311 7312 7313 7314 7315 7316 7317 7318 7319 7320 7321 7322
	if (!pmu->start_txn) {
		if (pmu->pmu_enable) {
			/*
			 * If we have pmu_enable/pmu_disable calls, install
			 * transaction stubs that use that to try and batch
			 * hardware accesses.
			 */
			pmu->start_txn  = perf_pmu_start_txn;
			pmu->commit_txn = perf_pmu_commit_txn;
			pmu->cancel_txn = perf_pmu_cancel_txn;
		} else {
			pmu->start_txn  = perf_pmu_nop_void;
			pmu->commit_txn = perf_pmu_nop_int;
			pmu->cancel_txn = perf_pmu_nop_void;
7323
		}
7324
	}
7325

P
Peter Zijlstra 已提交
7326 7327 7328 7329 7330
	if (!pmu->pmu_enable) {
		pmu->pmu_enable  = perf_pmu_nop_void;
		pmu->pmu_disable = perf_pmu_nop_void;
	}

7331 7332 7333
	if (!pmu->event_idx)
		pmu->event_idx = perf_event_idx_default;

7334
	list_add_rcu(&pmu->entry, &pmus);
7335
	atomic_set(&pmu->exclusive_cnt, 0);
P
Peter Zijlstra 已提交
7336 7337
	ret = 0;
unlock:
7338 7339
	mutex_unlock(&pmus_lock);

P
Peter Zijlstra 已提交
7340
	return ret;
P
Peter Zijlstra 已提交
7341

P
Peter Zijlstra 已提交
7342 7343 7344 7345
free_dev:
	device_del(pmu->dev);
	put_device(pmu->dev);

P
Peter Zijlstra 已提交
7346 7347 7348 7349
free_idr:
	if (pmu->type >= PERF_TYPE_MAX)
		idr_remove(&pmu_idr, pmu->type);

P
Peter Zijlstra 已提交
7350 7351 7352
free_pdc:
	free_percpu(pmu->pmu_disable_count);
	goto unlock;
7353
}
7354
EXPORT_SYMBOL_GPL(perf_pmu_register);
7355

7356
void perf_pmu_unregister(struct pmu *pmu)
7357
{
7358 7359 7360
	mutex_lock(&pmus_lock);
	list_del_rcu(&pmu->entry);
	mutex_unlock(&pmus_lock);
7361

7362
	/*
P
Peter Zijlstra 已提交
7363 7364
	 * We dereference the pmu list under both SRCU and regular RCU, so
	 * synchronize against both of those.
7365
	 */
7366
	synchronize_srcu(&pmus_srcu);
P
Peter Zijlstra 已提交
7367
	synchronize_rcu();
7368

P
Peter Zijlstra 已提交
7369
	free_percpu(pmu->pmu_disable_count);
P
Peter Zijlstra 已提交
7370 7371
	if (pmu->type >= PERF_TYPE_MAX)
		idr_remove(&pmu_idr, pmu->type);
P
Peter Zijlstra 已提交
7372 7373
	device_del(pmu->dev);
	put_device(pmu->dev);
7374
	free_pmu_context(pmu);
7375
}
7376
EXPORT_SYMBOL_GPL(perf_pmu_unregister);
7377

7378 7379
static int perf_try_init_event(struct pmu *pmu, struct perf_event *event)
{
P
Peter Zijlstra 已提交
7380
	struct perf_event_context *ctx = NULL;
7381 7382 7383 7384
	int ret;

	if (!try_module_get(pmu->module))
		return -ENODEV;
P
Peter Zijlstra 已提交
7385 7386

	if (event->group_leader != event) {
7387 7388 7389 7390 7391 7392
		/*
		 * This ctx->mutex can nest when we're called through
		 * inheritance. See the perf_event_ctx_lock_nested() comment.
		 */
		ctx = perf_event_ctx_lock_nested(event->group_leader,
						 SINGLE_DEPTH_NESTING);
P
Peter Zijlstra 已提交
7393 7394 7395
		BUG_ON(!ctx);
	}

7396 7397
	event->pmu = pmu;
	ret = pmu->event_init(event);
P
Peter Zijlstra 已提交
7398 7399 7400 7401

	if (ctx)
		perf_event_ctx_unlock(event->group_leader, ctx);

7402 7403 7404 7405 7406 7407
	if (ret)
		module_put(pmu->module);

	return ret;
}

7408 7409 7410 7411
struct pmu *perf_init_event(struct perf_event *event)
{
	struct pmu *pmu = NULL;
	int idx;
7412
	int ret;
7413 7414

	idx = srcu_read_lock(&pmus_srcu);
P
Peter Zijlstra 已提交
7415 7416 7417 7418

	rcu_read_lock();
	pmu = idr_find(&pmu_idr, event->attr.type);
	rcu_read_unlock();
7419
	if (pmu) {
7420
		ret = perf_try_init_event(pmu, event);
7421 7422
		if (ret)
			pmu = ERR_PTR(ret);
P
Peter Zijlstra 已提交
7423
		goto unlock;
7424
	}
P
Peter Zijlstra 已提交
7425

7426
	list_for_each_entry_rcu(pmu, &pmus, entry) {
7427
		ret = perf_try_init_event(pmu, event);
7428
		if (!ret)
P
Peter Zijlstra 已提交
7429
			goto unlock;
7430

7431 7432
		if (ret != -ENOENT) {
			pmu = ERR_PTR(ret);
P
Peter Zijlstra 已提交
7433
			goto unlock;
7434
		}
7435
	}
P
Peter Zijlstra 已提交
7436 7437
	pmu = ERR_PTR(-ENOENT);
unlock:
7438
	srcu_read_unlock(&pmus_srcu, idx);
7439

7440
	return pmu;
7441 7442
}

7443 7444 7445 7446 7447 7448 7449 7450 7451
static void account_event_cpu(struct perf_event *event, int cpu)
{
	if (event->parent)
		return;

	if (is_cgroup_event(event))
		atomic_inc(&per_cpu(perf_cgroup_events, cpu));
}

7452 7453
static void account_event(struct perf_event *event)
{
7454 7455 7456
	if (event->parent)
		return;

7457 7458 7459 7460 7461 7462 7463 7464
	if (event->attach_state & PERF_ATTACH_TASK)
		static_key_slow_inc(&perf_sched_events.key);
	if (event->attr.mmap || event->attr.mmap_data)
		atomic_inc(&nr_mmap_events);
	if (event->attr.comm)
		atomic_inc(&nr_comm_events);
	if (event->attr.task)
		atomic_inc(&nr_task_events);
7465 7466 7467 7468
	if (event->attr.freq) {
		if (atomic_inc_return(&nr_freq_events) == 1)
			tick_nohz_full_kick_all();
	}
7469
	if (has_branch_stack(event))
7470
		static_key_slow_inc(&perf_sched_events.key);
7471
	if (is_cgroup_event(event))
7472
		static_key_slow_inc(&perf_sched_events.key);
7473 7474

	account_event_cpu(event, event->cpu);
7475 7476
}

T
Thomas Gleixner 已提交
7477
/*
7478
 * Allocate and initialize a event structure
T
Thomas Gleixner 已提交
7479
 */
7480
static struct perf_event *
7481
perf_event_alloc(struct perf_event_attr *attr, int cpu,
7482 7483 7484
		 struct task_struct *task,
		 struct perf_event *group_leader,
		 struct perf_event *parent_event,
7485
		 perf_overflow_handler_t overflow_handler,
7486
		 void *context, int cgroup_fd)
T
Thomas Gleixner 已提交
7487
{
P
Peter Zijlstra 已提交
7488
	struct pmu *pmu;
7489 7490
	struct perf_event *event;
	struct hw_perf_event *hwc;
7491
	long err = -EINVAL;
T
Thomas Gleixner 已提交
7492

7493 7494 7495 7496 7497
	if ((unsigned)cpu >= nr_cpu_ids) {
		if (!task || cpu != -1)
			return ERR_PTR(-EINVAL);
	}

7498
	event = kzalloc(sizeof(*event), GFP_KERNEL);
7499
	if (!event)
7500
		return ERR_PTR(-ENOMEM);
T
Thomas Gleixner 已提交
7501

7502
	/*
7503
	 * Single events are their own group leaders, with an
7504 7505 7506
	 * empty sibling list:
	 */
	if (!group_leader)
7507
		group_leader = event;
7508

7509 7510
	mutex_init(&event->child_mutex);
	INIT_LIST_HEAD(&event->child_list);
7511

7512 7513 7514
	INIT_LIST_HEAD(&event->group_entry);
	INIT_LIST_HEAD(&event->event_entry);
	INIT_LIST_HEAD(&event->sibling_list);
7515
	INIT_LIST_HEAD(&event->rb_entry);
7516
	INIT_LIST_HEAD(&event->active_entry);
7517 7518
	INIT_HLIST_NODE(&event->hlist_entry);

7519

7520
	init_waitqueue_head(&event->waitq);
7521
	init_irq_work(&event->pending, perf_pending_event);
T
Thomas Gleixner 已提交
7522

7523
	mutex_init(&event->mmap_mutex);
7524

7525
	atomic_long_set(&event->refcount, 1);
7526 7527 7528 7529 7530
	event->cpu		= cpu;
	event->attr		= *attr;
	event->group_leader	= group_leader;
	event->pmu		= NULL;
	event->oncpu		= -1;
7531

7532
	event->parent		= parent_event;
7533

7534
	event->ns		= get_pid_ns(task_active_pid_ns(current));
7535
	event->id		= atomic64_inc_return(&perf_event_id);
7536

7537
	event->state		= PERF_EVENT_STATE_INACTIVE;
7538

7539 7540 7541
	if (task) {
		event->attach_state = PERF_ATTACH_TASK;
		/*
7542 7543 7544
		 * XXX pmu::event_init needs to know what task to account to
		 * and we cannot use the ctx information because we need the
		 * pmu before we get a ctx.
7545
		 */
7546
		event->hw.target = task;
7547 7548
	}

7549 7550 7551 7552
	event->clock = &local_clock;
	if (parent_event)
		event->clock = parent_event->clock;

7553
	if (!overflow_handler && parent_event) {
7554
		overflow_handler = parent_event->overflow_handler;
7555 7556
		context = parent_event->overflow_handler_context;
	}
7557

7558
	event->overflow_handler	= overflow_handler;
7559
	event->overflow_handler_context = context;
7560

J
Jiri Olsa 已提交
7561
	perf_event__state_init(event);
7562

7563
	pmu = NULL;
7564

7565
	hwc = &event->hw;
7566
	hwc->sample_period = attr->sample_period;
7567
	if (attr->freq && attr->sample_freq)
7568
		hwc->sample_period = 1;
7569
	hwc->last_period = hwc->sample_period;
7570

7571
	local64_set(&hwc->period_left, hwc->sample_period);
7572

7573
	/*
7574
	 * we currently do not support PERF_FORMAT_GROUP on inherited events
7575
	 */
7576
	if (attr->inherit && (attr->read_format & PERF_FORMAT_GROUP))
7577
		goto err_ns;
7578 7579 7580

	if (!has_branch_stack(event))
		event->attr.branch_sample_type = 0;
7581

7582 7583 7584 7585 7586 7587
	if (cgroup_fd != -1) {
		err = perf_cgroup_connect(cgroup_fd, event, attr, group_leader);
		if (err)
			goto err_ns;
	}

7588
	pmu = perf_init_event(event);
7589
	if (!pmu)
7590 7591
		goto err_ns;
	else if (IS_ERR(pmu)) {
7592
		err = PTR_ERR(pmu);
7593
		goto err_ns;
I
Ingo Molnar 已提交
7594
	}
7595

7596 7597 7598 7599
	err = exclusive_event_init(event);
	if (err)
		goto err_pmu;

7600
	if (!event->parent) {
7601 7602
		if (event->attr.sample_type & PERF_SAMPLE_CALLCHAIN) {
			err = get_callchain_buffers();
7603
			if (err)
7604
				goto err_per_task;
7605
		}
7606
	}
7607

7608
	return event;
7609

7610 7611 7612
err_per_task:
	exclusive_event_destroy(event);

7613 7614 7615
err_pmu:
	if (event->destroy)
		event->destroy(event);
7616
	module_put(pmu->module);
7617
err_ns:
7618 7619
	if (is_cgroup_event(event))
		perf_detach_cgroup(event);
7620 7621 7622 7623 7624
	if (event->ns)
		put_pid_ns(event->ns);
	kfree(event);

	return ERR_PTR(err);
T
Thomas Gleixner 已提交
7625 7626
}

7627 7628
static int perf_copy_attr(struct perf_event_attr __user *uattr,
			  struct perf_event_attr *attr)
7629 7630
{
	u32 size;
7631
	int ret;
7632 7633 7634 7635 7636 7637 7638 7639 7640 7641 7642 7643 7644 7645 7646 7647 7648 7649 7650 7651 7652 7653 7654 7655

	if (!access_ok(VERIFY_WRITE, uattr, PERF_ATTR_SIZE_VER0))
		return -EFAULT;

	/*
	 * zero the full structure, so that a short copy will be nice.
	 */
	memset(attr, 0, sizeof(*attr));

	ret = get_user(size, &uattr->size);
	if (ret)
		return ret;

	if (size > PAGE_SIZE)	/* silly large */
		goto err_size;

	if (!size)		/* abi compat */
		size = PERF_ATTR_SIZE_VER0;

	if (size < PERF_ATTR_SIZE_VER0)
		goto err_size;

	/*
	 * If we're handed a bigger struct than we know of,
7656 7657 7658
	 * ensure all the unknown bits are 0 - i.e. new
	 * user-space does not rely on any kernel feature
	 * extensions we dont know about yet.
7659 7660
	 */
	if (size > sizeof(*attr)) {
7661 7662 7663
		unsigned char __user *addr;
		unsigned char __user *end;
		unsigned char val;
7664

7665 7666
		addr = (void __user *)uattr + sizeof(*attr);
		end  = (void __user *)uattr + size;
7667

7668
		for (; addr < end; addr++) {
7669 7670 7671 7672 7673 7674
			ret = get_user(val, addr);
			if (ret)
				return ret;
			if (val)
				goto err_size;
		}
7675
		size = sizeof(*attr);
7676 7677 7678 7679 7680 7681
	}

	ret = copy_from_user(attr, uattr, size);
	if (ret)
		return -EFAULT;

7682
	if (attr->__reserved_1)
7683 7684 7685 7686 7687 7688 7689 7690
		return -EINVAL;

	if (attr->sample_type & ~(PERF_SAMPLE_MAX-1))
		return -EINVAL;

	if (attr->read_format & ~(PERF_FORMAT_MAX-1))
		return -EINVAL;

7691 7692 7693 7694 7695 7696 7697 7698 7699 7700 7701 7702 7703 7704 7705 7706 7707 7708 7709 7710 7711 7712 7713 7714 7715 7716 7717 7718
	if (attr->sample_type & PERF_SAMPLE_BRANCH_STACK) {
		u64 mask = attr->branch_sample_type;

		/* only using defined bits */
		if (mask & ~(PERF_SAMPLE_BRANCH_MAX-1))
			return -EINVAL;

		/* at least one branch bit must be set */
		if (!(mask & ~PERF_SAMPLE_BRANCH_PLM_ALL))
			return -EINVAL;

		/* propagate priv level, when not set for branch */
		if (!(mask & PERF_SAMPLE_BRANCH_PLM_ALL)) {

			/* exclude_kernel checked on syscall entry */
			if (!attr->exclude_kernel)
				mask |= PERF_SAMPLE_BRANCH_KERNEL;

			if (!attr->exclude_user)
				mask |= PERF_SAMPLE_BRANCH_USER;

			if (!attr->exclude_hv)
				mask |= PERF_SAMPLE_BRANCH_HV;
			/*
			 * adjust user setting (for HW filter setup)
			 */
			attr->branch_sample_type = mask;
		}
7719 7720
		/* privileged levels capture (kernel, hv): check permissions */
		if ((mask & PERF_SAMPLE_BRANCH_PERM_PLM)
7721 7722
		    && perf_paranoid_kernel() && !capable(CAP_SYS_ADMIN))
			return -EACCES;
7723
	}
7724

7725
	if (attr->sample_type & PERF_SAMPLE_REGS_USER) {
7726
		ret = perf_reg_validate(attr->sample_regs_user);
7727 7728 7729 7730 7731 7732 7733 7734 7735 7736 7737 7738 7739 7740 7741 7742 7743 7744
		if (ret)
			return ret;
	}

	if (attr->sample_type & PERF_SAMPLE_STACK_USER) {
		if (!arch_perf_have_user_stack_dump())
			return -ENOSYS;

		/*
		 * We have __u32 type for the size, but so far
		 * we can only use __u16 as maximum due to the
		 * __u16 sample size limit.
		 */
		if (attr->sample_stack_user >= USHRT_MAX)
			ret = -EINVAL;
		else if (!IS_ALIGNED(attr->sample_stack_user, sizeof(u64)))
			ret = -EINVAL;
	}
7745

7746 7747
	if (attr->sample_type & PERF_SAMPLE_REGS_INTR)
		ret = perf_reg_validate(attr->sample_regs_intr);
7748 7749 7750 7751 7752 7753 7754 7755 7756
out:
	return ret;

err_size:
	put_user(sizeof(*attr), &uattr->size);
	ret = -E2BIG;
	goto out;
}

7757 7758
static int
perf_event_set_output(struct perf_event *event, struct perf_event *output_event)
7759
{
7760
	struct ring_buffer *rb = NULL;
7761 7762
	int ret = -EINVAL;

7763
	if (!output_event)
7764 7765
		goto set;

7766 7767
	/* don't allow circular references */
	if (event == output_event)
7768 7769
		goto out;

7770 7771 7772 7773 7774 7775 7776
	/*
	 * Don't allow cross-cpu buffers
	 */
	if (output_event->cpu != event->cpu)
		goto out;

	/*
7777
	 * If its not a per-cpu rb, it must be the same task.
7778 7779 7780 7781
	 */
	if (output_event->cpu == -1 && output_event->ctx != event->ctx)
		goto out;

7782 7783 7784 7785 7786 7787
	/*
	 * Mixing clocks in the same buffer is trouble you don't need.
	 */
	if (output_event->clock != event->clock)
		goto out;

7788 7789 7790 7791 7792 7793 7794
	/*
	 * If both events generate aux data, they must be on the same PMU
	 */
	if (has_aux(event) && has_aux(output_event) &&
	    event->pmu != output_event->pmu)
		goto out;

7795
set:
7796
	mutex_lock(&event->mmap_mutex);
7797 7798 7799
	/* Can't redirect output if we've got an active mmap() */
	if (atomic_read(&event->mmap_count))
		goto unlock;
7800

7801
	if (output_event) {
7802 7803 7804
		/* get the rb we want to redirect to */
		rb = ring_buffer_get(output_event);
		if (!rb)
7805
			goto unlock;
7806 7807
	}

7808
	ring_buffer_attach(event, rb);
7809

7810
	ret = 0;
7811 7812 7813
unlock:
	mutex_unlock(&event->mmap_mutex);

7814 7815 7816 7817
out:
	return ret;
}

P
Peter Zijlstra 已提交
7818 7819 7820 7821 7822 7823 7824 7825 7826
static void mutex_lock_double(struct mutex *a, struct mutex *b)
{
	if (b < a)
		swap(a, b);

	mutex_lock(a);
	mutex_lock_nested(b, SINGLE_DEPTH_NESTING);
}

7827 7828 7829 7830 7831 7832 7833 7834 7835 7836 7837 7838 7839 7840 7841 7842 7843 7844 7845 7846 7847 7848 7849 7850 7851 7852 7853 7854 7855 7856 7857 7858 7859 7860 7861 7862 7863
static int perf_event_set_clock(struct perf_event *event, clockid_t clk_id)
{
	bool nmi_safe = false;

	switch (clk_id) {
	case CLOCK_MONOTONIC:
		event->clock = &ktime_get_mono_fast_ns;
		nmi_safe = true;
		break;

	case CLOCK_MONOTONIC_RAW:
		event->clock = &ktime_get_raw_fast_ns;
		nmi_safe = true;
		break;

	case CLOCK_REALTIME:
		event->clock = &ktime_get_real_ns;
		break;

	case CLOCK_BOOTTIME:
		event->clock = &ktime_get_boot_ns;
		break;

	case CLOCK_TAI:
		event->clock = &ktime_get_tai_ns;
		break;

	default:
		return -EINVAL;
	}

	if (!nmi_safe && !(event->pmu->capabilities & PERF_PMU_CAP_NO_NMI))
		return -EINVAL;

	return 0;
}

T
Thomas Gleixner 已提交
7864
/**
7865
 * sys_perf_event_open - open a performance event, associate it to a task/cpu
I
Ingo Molnar 已提交
7866
 *
7867
 * @attr_uptr:	event_id type attributes for monitoring/sampling
T
Thomas Gleixner 已提交
7868
 * @pid:		target pid
I
Ingo Molnar 已提交
7869
 * @cpu:		target cpu
7870
 * @group_fd:		group leader event fd
T
Thomas Gleixner 已提交
7871
 */
7872 7873
SYSCALL_DEFINE5(perf_event_open,
		struct perf_event_attr __user *, attr_uptr,
7874
		pid_t, pid, int, cpu, int, group_fd, unsigned long, flags)
T
Thomas Gleixner 已提交
7875
{
7876 7877
	struct perf_event *group_leader = NULL, *output_event = NULL;
	struct perf_event *event, *sibling;
7878
	struct perf_event_attr attr;
P
Peter Zijlstra 已提交
7879
	struct perf_event_context *ctx, *uninitialized_var(gctx);
7880
	struct file *event_file = NULL;
7881
	struct fd group = {NULL, 0};
M
Matt Helsley 已提交
7882
	struct task_struct *task = NULL;
7883
	struct pmu *pmu;
7884
	int event_fd;
7885
	int move_group = 0;
7886
	int err;
7887
	int f_flags = O_RDWR;
7888
	int cgroup_fd = -1;
T
Thomas Gleixner 已提交
7889

7890
	/* for future expandability... */
S
Stephane Eranian 已提交
7891
	if (flags & ~PERF_FLAG_ALL)
7892 7893
		return -EINVAL;

7894 7895 7896
	err = perf_copy_attr(attr_uptr, &attr);
	if (err)
		return err;
7897

7898 7899 7900 7901 7902
	if (!attr.exclude_kernel) {
		if (perf_paranoid_kernel() && !capable(CAP_SYS_ADMIN))
			return -EACCES;
	}

7903
	if (attr.freq) {
7904
		if (attr.sample_freq > sysctl_perf_event_sample_rate)
7905
			return -EINVAL;
7906 7907 7908
	} else {
		if (attr.sample_period & (1ULL << 63))
			return -EINVAL;
7909 7910
	}

S
Stephane Eranian 已提交
7911 7912 7913 7914 7915 7916 7917 7918 7919
	/*
	 * In cgroup mode, the pid argument is used to pass the fd
	 * opened to the cgroup directory in cgroupfs. The cpu argument
	 * designates the cpu on which to monitor threads from that
	 * cgroup.
	 */
	if ((flags & PERF_FLAG_PID_CGROUP) && (pid == -1 || cpu == -1))
		return -EINVAL;

7920 7921 7922 7923
	if (flags & PERF_FLAG_FD_CLOEXEC)
		f_flags |= O_CLOEXEC;

	event_fd = get_unused_fd_flags(f_flags);
7924 7925 7926
	if (event_fd < 0)
		return event_fd;

7927
	if (group_fd != -1) {
7928 7929
		err = perf_fget_light(group_fd, &group);
		if (err)
7930
			goto err_fd;
7931
		group_leader = group.file->private_data;
7932 7933 7934 7935 7936 7937
		if (flags & PERF_FLAG_FD_OUTPUT)
			output_event = group_leader;
		if (flags & PERF_FLAG_FD_NO_GROUP)
			group_leader = NULL;
	}

S
Stephane Eranian 已提交
7938
	if (pid != -1 && !(flags & PERF_FLAG_PID_CGROUP)) {
7939 7940 7941 7942 7943 7944 7945
		task = find_lively_task_by_vpid(pid);
		if (IS_ERR(task)) {
			err = PTR_ERR(task);
			goto err_group_fd;
		}
	}

7946 7947 7948 7949 7950 7951
	if (task && group_leader &&
	    group_leader->attr.inherit != attr.inherit) {
		err = -EINVAL;
		goto err_task;
	}

7952 7953
	get_online_cpus();

7954 7955 7956
	if (flags & PERF_FLAG_PID_CGROUP)
		cgroup_fd = pid;

7957
	event = perf_event_alloc(&attr, cpu, task, group_leader, NULL,
7958
				 NULL, NULL, cgroup_fd);
7959 7960
	if (IS_ERR(event)) {
		err = PTR_ERR(event);
7961
		goto err_cpus;
7962 7963
	}

7964 7965 7966 7967 7968 7969 7970
	if (is_sampling_event(event)) {
		if (event->pmu->capabilities & PERF_PMU_CAP_NO_INTERRUPT) {
			err = -ENOTSUPP;
			goto err_alloc;
		}
	}

7971 7972
	account_event(event);

7973 7974 7975 7976 7977
	/*
	 * Special case software events and allow them to be part of
	 * any hardware group.
	 */
	pmu = event->pmu;
7978

7979 7980 7981 7982 7983 7984
	if (attr.use_clockid) {
		err = perf_event_set_clock(event, attr.clockid);
		if (err)
			goto err_alloc;
	}

7985 7986 7987 7988 7989 7990 7991 7992 7993 7994 7995 7996 7997 7998 7999 8000 8001 8002 8003 8004 8005 8006
	if (group_leader &&
	    (is_software_event(event) != is_software_event(group_leader))) {
		if (is_software_event(event)) {
			/*
			 * If event and group_leader are not both a software
			 * event, and event is, then group leader is not.
			 *
			 * Allow the addition of software events to !software
			 * groups, this is safe because software events never
			 * fail to schedule.
			 */
			pmu = group_leader->pmu;
		} else if (is_software_event(group_leader) &&
			   (group_leader->group_flags & PERF_GROUP_SOFTWARE)) {
			/*
			 * In case the group is a pure software group, and we
			 * try to add a hardware event, move the whole group to
			 * the hardware context.
			 */
			move_group = 1;
		}
	}
8007 8008 8009 8010

	/*
	 * Get the target context (task or percpu):
	 */
8011
	ctx = find_get_context(pmu, task, event);
8012 8013
	if (IS_ERR(ctx)) {
		err = PTR_ERR(ctx);
8014
		goto err_alloc;
8015 8016
	}

8017 8018 8019 8020 8021
	if ((pmu->capabilities & PERF_PMU_CAP_EXCLUSIVE) && group_leader) {
		err = -EBUSY;
		goto err_context;
	}

8022 8023 8024 8025 8026
	if (task) {
		put_task_struct(task);
		task = NULL;
	}

I
Ingo Molnar 已提交
8027
	/*
8028
	 * Look up the group leader (we will attach this event to it):
8029
	 */
8030
	if (group_leader) {
8031
		err = -EINVAL;
8032 8033

		/*
I
Ingo Molnar 已提交
8034 8035 8036 8037
		 * Do not allow a recursive hierarchy (this new sibling
		 * becoming part of another group-sibling):
		 */
		if (group_leader->group_leader != group_leader)
8038
			goto err_context;
8039 8040 8041 8042 8043

		/* All events in a group should have the same clock */
		if (group_leader->clock != event->clock)
			goto err_context;

I
Ingo Molnar 已提交
8044 8045 8046
		/*
		 * Do not allow to attach to a group in a different
		 * task or CPU context:
8047
		 */
8048
		if (move_group) {
8049 8050 8051 8052 8053 8054 8055 8056 8057 8058 8059 8060 8061
			/*
			 * Make sure we're both on the same task, or both
			 * per-cpu events.
			 */
			if (group_leader->ctx->task != ctx->task)
				goto err_context;

			/*
			 * Make sure we're both events for the same CPU;
			 * grouping events for different CPUs is broken; since
			 * you can never concurrently schedule them anyhow.
			 */
			if (group_leader->cpu != event->cpu)
8062 8063 8064 8065 8066 8067
				goto err_context;
		} else {
			if (group_leader->ctx != ctx)
				goto err_context;
		}

8068 8069 8070
		/*
		 * Only a group leader can be exclusive or pinned
		 */
8071
		if (attr.exclusive || attr.pinned)
8072
			goto err_context;
8073 8074 8075 8076 8077
	}

	if (output_event) {
		err = perf_event_set_output(event, output_event);
		if (err)
8078
			goto err_context;
8079
	}
T
Thomas Gleixner 已提交
8080

8081 8082
	event_file = anon_inode_getfile("[perf_event]", &perf_fops, event,
					f_flags);
8083 8084
	if (IS_ERR(event_file)) {
		err = PTR_ERR(event_file);
8085
		goto err_context;
8086
	}
8087

8088
	if (move_group) {
P
Peter Zijlstra 已提交
8089 8090 8091 8092 8093 8094 8095
		gctx = group_leader->ctx;

		/*
		 * See perf_event_ctx_lock() for comments on the details
		 * of swizzling perf_event::ctx.
		 */
		mutex_lock_double(&gctx->mutex, &ctx->mutex);
8096

8097
		perf_remove_from_context(group_leader, false);
J
Jiri Olsa 已提交
8098

8099 8100
		list_for_each_entry(sibling, &group_leader->sibling_list,
				    group_entry) {
8101
			perf_remove_from_context(sibling, false);
8102 8103
			put_ctx(gctx);
		}
P
Peter Zijlstra 已提交
8104 8105
	} else {
		mutex_lock(&ctx->mutex);
8106
	}
8107

8108
	WARN_ON_ONCE(ctx->parent_ctx);
8109 8110

	if (move_group) {
P
Peter Zijlstra 已提交
8111 8112 8113 8114
		/*
		 * Wait for everybody to stop referencing the events through
		 * the old lists, before installing it on new lists.
		 */
8115
		synchronize_rcu();
P
Peter Zijlstra 已提交
8116

8117 8118 8119 8120 8121 8122 8123 8124 8125 8126
		/*
		 * Install the group siblings before the group leader.
		 *
		 * Because a group leader will try and install the entire group
		 * (through the sibling list, which is still in-tact), we can
		 * end up with siblings installed in the wrong context.
		 *
		 * By installing siblings first we NO-OP because they're not
		 * reachable through the group lists.
		 */
8127 8128
		list_for_each_entry(sibling, &group_leader->sibling_list,
				    group_entry) {
8129
			perf_event__state_init(sibling);
8130
			perf_install_in_context(ctx, sibling, sibling->cpu);
8131 8132
			get_ctx(ctx);
		}
8133 8134 8135 8136 8137 8138 8139 8140 8141

		/*
		 * Removing from the context ends up with disabled
		 * event. What we want here is event in the initial
		 * startup state, ready to be add into new context.
		 */
		perf_event__state_init(group_leader);
		perf_install_in_context(ctx, group_leader, group_leader->cpu);
		get_ctx(ctx);
8142 8143
	}

8144 8145 8146 8147 8148 8149 8150
	if (!exclusive_event_installable(event, ctx)) {
		err = -EBUSY;
		mutex_unlock(&ctx->mutex);
		fput(event_file);
		goto err_context;
	}

8151
	perf_install_in_context(ctx, event, event->cpu);
8152
	perf_unpin_context(ctx);
P
Peter Zijlstra 已提交
8153 8154 8155 8156 8157

	if (move_group) {
		mutex_unlock(&gctx->mutex);
		put_ctx(gctx);
	}
8158
	mutex_unlock(&ctx->mutex);
8159

8160 8161
	put_online_cpus();

8162
	event->owner = current;
P
Peter Zijlstra 已提交
8163

8164 8165 8166
	mutex_lock(&current->perf_event_mutex);
	list_add_tail(&event->owner_entry, &current->perf_event_list);
	mutex_unlock(&current->perf_event_mutex);
8167

8168 8169 8170 8171
	/*
	 * Precalculate sample_data sizes
	 */
	perf_event__header_size(event);
8172
	perf_event__id_header_size(event);
8173

8174 8175 8176 8177 8178 8179
	/*
	 * Drop the reference on the group_event after placing the
	 * new event on the sibling_list. This ensures destruction
	 * of the group leader will find the pointer to itself in
	 * perf_group_detach().
	 */
8180
	fdput(group);
8181 8182
	fd_install(event_fd, event_file);
	return event_fd;
T
Thomas Gleixner 已提交
8183

8184
err_context:
8185
	perf_unpin_context(ctx);
8186
	put_ctx(ctx);
8187
err_alloc:
8188
	free_event(event);
8189
err_cpus:
8190
	put_online_cpus();
8191
err_task:
P
Peter Zijlstra 已提交
8192 8193
	if (task)
		put_task_struct(task);
8194
err_group_fd:
8195
	fdput(group);
8196 8197
err_fd:
	put_unused_fd(event_fd);
8198
	return err;
T
Thomas Gleixner 已提交
8199 8200
}

8201 8202 8203 8204 8205
/**
 * perf_event_create_kernel_counter
 *
 * @attr: attributes of the counter to create
 * @cpu: cpu in which the counter is bound
M
Matt Helsley 已提交
8206
 * @task: task to profile (NULL for percpu)
8207 8208 8209
 */
struct perf_event *
perf_event_create_kernel_counter(struct perf_event_attr *attr, int cpu,
M
Matt Helsley 已提交
8210
				 struct task_struct *task,
8211 8212
				 perf_overflow_handler_t overflow_handler,
				 void *context)
8213 8214
{
	struct perf_event_context *ctx;
8215
	struct perf_event *event;
8216
	int err;
8217

8218 8219 8220
	/*
	 * Get the target context (task or percpu):
	 */
8221

8222
	event = perf_event_alloc(attr, cpu, task, NULL, NULL,
8223
				 overflow_handler, context, -1);
8224 8225 8226 8227
	if (IS_ERR(event)) {
		err = PTR_ERR(event);
		goto err;
	}
8228

8229 8230 8231
	/* Mark owner so we could distinguish it from user events. */
	event->owner = EVENT_OWNER_KERNEL;

8232 8233
	account_event(event);

8234
	ctx = find_get_context(event->pmu, task, event);
8235 8236
	if (IS_ERR(ctx)) {
		err = PTR_ERR(ctx);
8237
		goto err_free;
8238
	}
8239 8240 8241

	WARN_ON_ONCE(ctx->parent_ctx);
	mutex_lock(&ctx->mutex);
8242 8243 8244 8245 8246 8247 8248 8249
	if (!exclusive_event_installable(event, ctx)) {
		mutex_unlock(&ctx->mutex);
		perf_unpin_context(ctx);
		put_ctx(ctx);
		err = -EBUSY;
		goto err_free;
	}

8250
	perf_install_in_context(ctx, event, cpu);
8251
	perf_unpin_context(ctx);
8252 8253 8254 8255
	mutex_unlock(&ctx->mutex);

	return event;

8256 8257 8258
err_free:
	free_event(event);
err:
8259
	return ERR_PTR(err);
8260
}
8261
EXPORT_SYMBOL_GPL(perf_event_create_kernel_counter);
8262

8263 8264 8265 8266 8267 8268 8269 8270 8271 8272
void perf_pmu_migrate_context(struct pmu *pmu, int src_cpu, int dst_cpu)
{
	struct perf_event_context *src_ctx;
	struct perf_event_context *dst_ctx;
	struct perf_event *event, *tmp;
	LIST_HEAD(events);

	src_ctx = &per_cpu_ptr(pmu->pmu_cpu_context, src_cpu)->ctx;
	dst_ctx = &per_cpu_ptr(pmu->pmu_cpu_context, dst_cpu)->ctx;

P
Peter Zijlstra 已提交
8273 8274 8275 8276 8277
	/*
	 * See perf_event_ctx_lock() for comments on the details
	 * of swizzling perf_event::ctx.
	 */
	mutex_lock_double(&src_ctx->mutex, &dst_ctx->mutex);
8278 8279
	list_for_each_entry_safe(event, tmp, &src_ctx->event_list,
				 event_entry) {
8280
		perf_remove_from_context(event, false);
8281
		unaccount_event_cpu(event, src_cpu);
8282
		put_ctx(src_ctx);
8283
		list_add(&event->migrate_entry, &events);
8284 8285
	}

8286 8287 8288
	/*
	 * Wait for the events to quiesce before re-instating them.
	 */
8289 8290
	synchronize_rcu();

8291 8292 8293 8294 8295 8296 8297 8298 8299 8300 8301 8302 8303 8304 8305 8306 8307 8308 8309 8310 8311 8312 8313 8314
	/*
	 * Re-instate events in 2 passes.
	 *
	 * Skip over group leaders and only install siblings on this first
	 * pass, siblings will not get enabled without a leader, however a
	 * leader will enable its siblings, even if those are still on the old
	 * context.
	 */
	list_for_each_entry_safe(event, tmp, &events, migrate_entry) {
		if (event->group_leader == event)
			continue;

		list_del(&event->migrate_entry);
		if (event->state >= PERF_EVENT_STATE_OFF)
			event->state = PERF_EVENT_STATE_INACTIVE;
		account_event_cpu(event, dst_cpu);
		perf_install_in_context(dst_ctx, event, dst_cpu);
		get_ctx(dst_ctx);
	}

	/*
	 * Once all the siblings are setup properly, install the group leaders
	 * to make it go.
	 */
8315 8316
	list_for_each_entry_safe(event, tmp, &events, migrate_entry) {
		list_del(&event->migrate_entry);
8317 8318
		if (event->state >= PERF_EVENT_STATE_OFF)
			event->state = PERF_EVENT_STATE_INACTIVE;
8319
		account_event_cpu(event, dst_cpu);
8320 8321 8322 8323
		perf_install_in_context(dst_ctx, event, dst_cpu);
		get_ctx(dst_ctx);
	}
	mutex_unlock(&dst_ctx->mutex);
P
Peter Zijlstra 已提交
8324
	mutex_unlock(&src_ctx->mutex);
8325 8326 8327
}
EXPORT_SYMBOL_GPL(perf_pmu_migrate_context);

8328
static void sync_child_event(struct perf_event *child_event,
8329
			       struct task_struct *child)
8330
{
8331
	struct perf_event *parent_event = child_event->parent;
8332
	u64 child_val;
8333

8334 8335
	if (child_event->attr.inherit_stat)
		perf_event_read_event(child_event, child);
8336

P
Peter Zijlstra 已提交
8337
	child_val = perf_event_count(child_event);
8338 8339 8340 8341

	/*
	 * Add back the child's count to the parent's count:
	 */
8342
	atomic64_add(child_val, &parent_event->child_count);
8343 8344 8345 8346
	atomic64_add(child_event->total_time_enabled,
		     &parent_event->child_total_time_enabled);
	atomic64_add(child_event->total_time_running,
		     &parent_event->child_total_time_running);
8347 8348

	/*
8349
	 * Remove this event from the parent's list
8350
	 */
8351 8352 8353 8354
	WARN_ON_ONCE(parent_event->ctx->parent_ctx);
	mutex_lock(&parent_event->child_mutex);
	list_del_init(&child_event->child_list);
	mutex_unlock(&parent_event->child_mutex);
8355

8356 8357 8358 8359 8360 8361
	/*
	 * Make sure user/parent get notified, that we just
	 * lost one event.
	 */
	perf_event_wakeup(parent_event);

8362
	/*
8363
	 * Release the parent event, if this was the last
8364 8365
	 * reference to it.
	 */
8366
	put_event(parent_event);
8367 8368
}

8369
static void
8370 8371
__perf_event_exit_task(struct perf_event *child_event,
			 struct perf_event_context *child_ctx,
8372
			 struct task_struct *child)
8373
{
8374 8375 8376 8377 8378 8379 8380 8381 8382 8383 8384 8385 8386
	/*
	 * Do not destroy the 'original' grouping; because of the context
	 * switch optimization the original events could've ended up in a
	 * random child task.
	 *
	 * If we were to destroy the original group, all group related
	 * operations would cease to function properly after this random
	 * child dies.
	 *
	 * Do destroy all inherited groups, we don't care about those
	 * and being thorough is better.
	 */
	perf_remove_from_context(child_event, !!child_event->parent);
8387

8388
	/*
8389
	 * It can happen that the parent exits first, and has events
8390
	 * that are still around due to the child reference. These
8391
	 * events need to be zapped.
8392
	 */
8393
	if (child_event->parent) {
8394 8395
		sync_child_event(child_event, child);
		free_event(child_event);
8396 8397 8398
	} else {
		child_event->state = PERF_EVENT_STATE_EXIT;
		perf_event_wakeup(child_event);
8399
	}
8400 8401
}

P
Peter Zijlstra 已提交
8402
static void perf_event_exit_task_context(struct task_struct *child, int ctxn)
8403
{
8404
	struct perf_event *child_event, *next;
8405
	struct perf_event_context *child_ctx, *clone_ctx = NULL;
8406
	unsigned long flags;
8407

P
Peter Zijlstra 已提交
8408
	if (likely(!child->perf_event_ctxp[ctxn])) {
8409
		perf_event_task(child, NULL, 0);
8410
		return;
P
Peter Zijlstra 已提交
8411
	}
8412

8413
	local_irq_save(flags);
8414 8415 8416 8417 8418 8419
	/*
	 * We can't reschedule here because interrupts are disabled,
	 * and either child is current or it is a task that can't be
	 * scheduled, so we are now safe from rescheduling changing
	 * our context.
	 */
8420
	child_ctx = rcu_dereference_raw(child->perf_event_ctxp[ctxn]);
8421 8422 8423

	/*
	 * Take the context lock here so that if find_get_context is
8424
	 * reading child->perf_event_ctxp, we wait until it has
8425 8426
	 * incremented the context's refcount before we do put_ctx below.
	 */
8427
	raw_spin_lock(&child_ctx->lock);
8428
	task_ctx_sched_out(child_ctx);
P
Peter Zijlstra 已提交
8429
	child->perf_event_ctxp[ctxn] = NULL;
8430

8431 8432 8433
	/*
	 * If this context is a clone; unclone it so it can't get
	 * swapped to another process while we're removing all
8434
	 * the events from it.
8435
	 */
8436
	clone_ctx = unclone_ctx(child_ctx);
8437
	update_context_time(child_ctx);
8438
	raw_spin_unlock_irqrestore(&child_ctx->lock, flags);
P
Peter Zijlstra 已提交
8439

8440 8441
	if (clone_ctx)
		put_ctx(clone_ctx);
8442

P
Peter Zijlstra 已提交
8443
	/*
8444 8445 8446
	 * Report the task dead after unscheduling the events so that we
	 * won't get any samples after PERF_RECORD_EXIT. We can however still
	 * get a few PERF_RECORD_READ events.
P
Peter Zijlstra 已提交
8447
	 */
8448
	perf_event_task(child, child_ctx, 0);
8449

8450 8451 8452
	/*
	 * We can recurse on the same lock type through:
	 *
8453 8454
	 *   __perf_event_exit_task()
	 *     sync_child_event()
8455 8456
	 *       put_event()
	 *         mutex_lock(&ctx->mutex)
8457 8458 8459
	 *
	 * But since its the parent context it won't be the same instance.
	 */
8460
	mutex_lock(&child_ctx->mutex);
8461

8462
	list_for_each_entry_safe(child_event, next, &child_ctx->event_list, event_entry)
8463
		__perf_event_exit_task(child_event, child_ctx, child);
8464

8465 8466 8467
	mutex_unlock(&child_ctx->mutex);

	put_ctx(child_ctx);
8468 8469
}

P
Peter Zijlstra 已提交
8470 8471 8472 8473 8474
/*
 * When a child task exits, feed back event values to parent events.
 */
void perf_event_exit_task(struct task_struct *child)
{
P
Peter Zijlstra 已提交
8475
	struct perf_event *event, *tmp;
P
Peter Zijlstra 已提交
8476 8477
	int ctxn;

P
Peter Zijlstra 已提交
8478 8479 8480 8481 8482 8483 8484 8485 8486 8487 8488 8489 8490 8491 8492
	mutex_lock(&child->perf_event_mutex);
	list_for_each_entry_safe(event, tmp, &child->perf_event_list,
				 owner_entry) {
		list_del_init(&event->owner_entry);

		/*
		 * Ensure the list deletion is visible before we clear
		 * the owner, closes a race against perf_release() where
		 * we need to serialize on the owner->perf_event_mutex.
		 */
		smp_wmb();
		event->owner = NULL;
	}
	mutex_unlock(&child->perf_event_mutex);

P
Peter Zijlstra 已提交
8493 8494 8495 8496
	for_each_task_context_nr(ctxn)
		perf_event_exit_task_context(child, ctxn);
}

8497 8498 8499 8500 8501 8502 8503 8504 8505 8506 8507 8508
static void perf_free_event(struct perf_event *event,
			    struct perf_event_context *ctx)
{
	struct perf_event *parent = event->parent;

	if (WARN_ON_ONCE(!parent))
		return;

	mutex_lock(&parent->child_mutex);
	list_del_init(&event->child_list);
	mutex_unlock(&parent->child_mutex);

8509
	put_event(parent);
8510

P
Peter Zijlstra 已提交
8511
	raw_spin_lock_irq(&ctx->lock);
8512
	perf_group_detach(event);
8513
	list_del_event(event, ctx);
P
Peter Zijlstra 已提交
8514
	raw_spin_unlock_irq(&ctx->lock);
8515 8516 8517
	free_event(event);
}

8518
/*
P
Peter Zijlstra 已提交
8519
 * Free an unexposed, unused context as created by inheritance by
P
Peter Zijlstra 已提交
8520
 * perf_event_init_task below, used by fork() in case of fail.
P
Peter Zijlstra 已提交
8521 8522 8523
 *
 * Not all locks are strictly required, but take them anyway to be nice and
 * help out with the lockdep assertions.
8524
 */
8525
void perf_event_free_task(struct task_struct *task)
8526
{
P
Peter Zijlstra 已提交
8527
	struct perf_event_context *ctx;
8528
	struct perf_event *event, *tmp;
P
Peter Zijlstra 已提交
8529
	int ctxn;
8530

P
Peter Zijlstra 已提交
8531 8532 8533 8534
	for_each_task_context_nr(ctxn) {
		ctx = task->perf_event_ctxp[ctxn];
		if (!ctx)
			continue;
8535

P
Peter Zijlstra 已提交
8536
		mutex_lock(&ctx->mutex);
8537
again:
P
Peter Zijlstra 已提交
8538 8539 8540
		list_for_each_entry_safe(event, tmp, &ctx->pinned_groups,
				group_entry)
			perf_free_event(event, ctx);
8541

P
Peter Zijlstra 已提交
8542 8543 8544
		list_for_each_entry_safe(event, tmp, &ctx->flexible_groups,
				group_entry)
			perf_free_event(event, ctx);
8545

P
Peter Zijlstra 已提交
8546 8547 8548
		if (!list_empty(&ctx->pinned_groups) ||
				!list_empty(&ctx->flexible_groups))
			goto again;
8549

P
Peter Zijlstra 已提交
8550
		mutex_unlock(&ctx->mutex);
8551

P
Peter Zijlstra 已提交
8552 8553
		put_ctx(ctx);
	}
8554 8555
}

8556 8557 8558 8559 8560 8561 8562 8563
void perf_event_delayed_put(struct task_struct *task)
{
	int ctxn;

	for_each_task_context_nr(ctxn)
		WARN_ON_ONCE(task->perf_event_ctxp[ctxn]);
}

P
Peter Zijlstra 已提交
8564 8565 8566 8567 8568 8569 8570 8571 8572 8573 8574
/*
 * inherit a event from parent task to child task:
 */
static struct perf_event *
inherit_event(struct perf_event *parent_event,
	      struct task_struct *parent,
	      struct perf_event_context *parent_ctx,
	      struct task_struct *child,
	      struct perf_event *group_leader,
	      struct perf_event_context *child_ctx)
{
8575
	enum perf_event_active_state parent_state = parent_event->state;
P
Peter Zijlstra 已提交
8576
	struct perf_event *child_event;
8577
	unsigned long flags;
P
Peter Zijlstra 已提交
8578 8579 8580 8581 8582 8583 8584 8585 8586 8587 8588 8589

	/*
	 * Instead of creating recursive hierarchies of events,
	 * we link inherited events back to the original parent,
	 * which has a filp for sure, which we use as the reference
	 * count:
	 */
	if (parent_event->parent)
		parent_event = parent_event->parent;

	child_event = perf_event_alloc(&parent_event->attr,
					   parent_event->cpu,
8590
					   child,
P
Peter Zijlstra 已提交
8591
					   group_leader, parent_event,
8592
					   NULL, NULL, -1);
P
Peter Zijlstra 已提交
8593 8594
	if (IS_ERR(child_event))
		return child_event;
8595

8596 8597
	if (is_orphaned_event(parent_event) ||
	    !atomic_long_inc_not_zero(&parent_event->refcount)) {
8598 8599 8600 8601
		free_event(child_event);
		return NULL;
	}

P
Peter Zijlstra 已提交
8602 8603 8604 8605 8606 8607 8608
	get_ctx(child_ctx);

	/*
	 * Make the child state follow the state of the parent event,
	 * not its attr.disabled bit.  We hold the parent's mutex,
	 * so we won't race with perf_event_{en, dis}able_family.
	 */
8609
	if (parent_state >= PERF_EVENT_STATE_INACTIVE)
P
Peter Zijlstra 已提交
8610 8611 8612 8613 8614 8615 8616 8617 8618 8619 8620 8621 8622 8623 8624 8625
		child_event->state = PERF_EVENT_STATE_INACTIVE;
	else
		child_event->state = PERF_EVENT_STATE_OFF;

	if (parent_event->attr.freq) {
		u64 sample_period = parent_event->hw.sample_period;
		struct hw_perf_event *hwc = &child_event->hw;

		hwc->sample_period = sample_period;
		hwc->last_period   = sample_period;

		local64_set(&hwc->period_left, sample_period);
	}

	child_event->ctx = child_ctx;
	child_event->overflow_handler = parent_event->overflow_handler;
8626 8627
	child_event->overflow_handler_context
		= parent_event->overflow_handler_context;
P
Peter Zijlstra 已提交
8628

8629 8630 8631 8632
	/*
	 * Precalculate sample_data sizes
	 */
	perf_event__header_size(child_event);
8633
	perf_event__id_header_size(child_event);
8634

P
Peter Zijlstra 已提交
8635 8636 8637
	/*
	 * Link it up in the child's context:
	 */
8638
	raw_spin_lock_irqsave(&child_ctx->lock, flags);
P
Peter Zijlstra 已提交
8639
	add_event_to_ctx(child_event, child_ctx);
8640
	raw_spin_unlock_irqrestore(&child_ctx->lock, flags);
P
Peter Zijlstra 已提交
8641 8642 8643 8644 8645 8646 8647 8648 8649 8650 8651 8652 8653 8654 8655 8656 8657 8658 8659 8660 8661 8662 8663 8664 8665 8666 8667 8668 8669 8670 8671 8672 8673

	/*
	 * Link this into the parent event's child list
	 */
	WARN_ON_ONCE(parent_event->ctx->parent_ctx);
	mutex_lock(&parent_event->child_mutex);
	list_add_tail(&child_event->child_list, &parent_event->child_list);
	mutex_unlock(&parent_event->child_mutex);

	return child_event;
}

static int inherit_group(struct perf_event *parent_event,
	      struct task_struct *parent,
	      struct perf_event_context *parent_ctx,
	      struct task_struct *child,
	      struct perf_event_context *child_ctx)
{
	struct perf_event *leader;
	struct perf_event *sub;
	struct perf_event *child_ctr;

	leader = inherit_event(parent_event, parent, parent_ctx,
				 child, NULL, child_ctx);
	if (IS_ERR(leader))
		return PTR_ERR(leader);
	list_for_each_entry(sub, &parent_event->sibling_list, group_entry) {
		child_ctr = inherit_event(sub, parent, parent_ctx,
					    child, leader, child_ctx);
		if (IS_ERR(child_ctr))
			return PTR_ERR(child_ctr);
	}
	return 0;
8674 8675 8676 8677 8678
}

static int
inherit_task_group(struct perf_event *event, struct task_struct *parent,
		   struct perf_event_context *parent_ctx,
P
Peter Zijlstra 已提交
8679
		   struct task_struct *child, int ctxn,
8680 8681 8682
		   int *inherited_all)
{
	int ret;
P
Peter Zijlstra 已提交
8683
	struct perf_event_context *child_ctx;
8684 8685 8686 8687

	if (!event->attr.inherit) {
		*inherited_all = 0;
		return 0;
8688 8689
	}

8690
	child_ctx = child->perf_event_ctxp[ctxn];
8691 8692 8693 8694 8695 8696 8697
	if (!child_ctx) {
		/*
		 * This is executed from the parent task context, so
		 * inherit events that have been marked for cloning.
		 * First allocate and initialize a context for the
		 * child.
		 */
8698

8699
		child_ctx = alloc_perf_context(parent_ctx->pmu, child);
8700 8701
		if (!child_ctx)
			return -ENOMEM;
8702

P
Peter Zijlstra 已提交
8703
		child->perf_event_ctxp[ctxn] = child_ctx;
8704 8705 8706 8707 8708 8709 8710 8711 8712
	}

	ret = inherit_group(event, parent, parent_ctx,
			    child, child_ctx);

	if (ret)
		*inherited_all = 0;

	return ret;
8713 8714
}

8715
/*
8716
 * Initialize the perf_event context in task_struct
8717
 */
8718
static int perf_event_init_context(struct task_struct *child, int ctxn)
8719
{
8720
	struct perf_event_context *child_ctx, *parent_ctx;
8721 8722
	struct perf_event_context *cloned_ctx;
	struct perf_event *event;
8723
	struct task_struct *parent = current;
8724
	int inherited_all = 1;
8725
	unsigned long flags;
8726
	int ret = 0;
8727

P
Peter Zijlstra 已提交
8728
	if (likely(!parent->perf_event_ctxp[ctxn]))
8729 8730
		return 0;

8731
	/*
8732 8733
	 * If the parent's context is a clone, pin it so it won't get
	 * swapped under us.
8734
	 */
P
Peter Zijlstra 已提交
8735
	parent_ctx = perf_pin_task_context(parent, ctxn);
8736 8737
	if (!parent_ctx)
		return 0;
8738

8739 8740 8741 8742 8743 8744 8745
	/*
	 * No need to check if parent_ctx != NULL here; since we saw
	 * it non-NULL earlier, the only reason for it to become NULL
	 * is if we exit, and since we're currently in the middle of
	 * a fork we can't be exiting at the same time.
	 */

8746 8747 8748 8749
	/*
	 * Lock the parent list. No need to lock the child - not PID
	 * hashed yet and not running, so nobody can access it.
	 */
8750
	mutex_lock(&parent_ctx->mutex);
8751 8752 8753 8754 8755

	/*
	 * We dont have to disable NMIs - we are only looking at
	 * the list, not manipulating it:
	 */
8756
	list_for_each_entry(event, &parent_ctx->pinned_groups, group_entry) {
P
Peter Zijlstra 已提交
8757 8758
		ret = inherit_task_group(event, parent, parent_ctx,
					 child, ctxn, &inherited_all);
8759 8760 8761
		if (ret)
			break;
	}
8762

8763 8764 8765 8766 8767 8768 8769 8770 8771
	/*
	 * We can't hold ctx->lock when iterating the ->flexible_group list due
	 * to allocations, but we need to prevent rotation because
	 * rotate_ctx() will change the list from interrupt context.
	 */
	raw_spin_lock_irqsave(&parent_ctx->lock, flags);
	parent_ctx->rotate_disable = 1;
	raw_spin_unlock_irqrestore(&parent_ctx->lock, flags);

8772
	list_for_each_entry(event, &parent_ctx->flexible_groups, group_entry) {
P
Peter Zijlstra 已提交
8773 8774
		ret = inherit_task_group(event, parent, parent_ctx,
					 child, ctxn, &inherited_all);
8775
		if (ret)
8776
			break;
8777 8778
	}

8779 8780 8781
	raw_spin_lock_irqsave(&parent_ctx->lock, flags);
	parent_ctx->rotate_disable = 0;

P
Peter Zijlstra 已提交
8782
	child_ctx = child->perf_event_ctxp[ctxn];
8783

8784
	if (child_ctx && inherited_all) {
8785 8786 8787
		/*
		 * Mark the child context as a clone of the parent
		 * context, or of whatever the parent is a clone of.
P
Peter Zijlstra 已提交
8788 8789 8790
		 *
		 * Note that if the parent is a clone, the holding of
		 * parent_ctx->lock avoids it from being uncloned.
8791
		 */
P
Peter Zijlstra 已提交
8792
		cloned_ctx = parent_ctx->parent_ctx;
8793 8794
		if (cloned_ctx) {
			child_ctx->parent_ctx = cloned_ctx;
8795
			child_ctx->parent_gen = parent_ctx->parent_gen;
8796 8797 8798 8799 8800
		} else {
			child_ctx->parent_ctx = parent_ctx;
			child_ctx->parent_gen = parent_ctx->generation;
		}
		get_ctx(child_ctx->parent_ctx);
8801 8802
	}

P
Peter Zijlstra 已提交
8803
	raw_spin_unlock_irqrestore(&parent_ctx->lock, flags);
8804
	mutex_unlock(&parent_ctx->mutex);
8805

8806
	perf_unpin_context(parent_ctx);
8807
	put_ctx(parent_ctx);
8808

8809
	return ret;
8810 8811
}

P
Peter Zijlstra 已提交
8812 8813 8814 8815 8816 8817 8818
/*
 * Initialize the perf_event context in task_struct
 */
int perf_event_init_task(struct task_struct *child)
{
	int ctxn, ret;

8819 8820 8821 8822
	memset(child->perf_event_ctxp, 0, sizeof(child->perf_event_ctxp));
	mutex_init(&child->perf_event_mutex);
	INIT_LIST_HEAD(&child->perf_event_list);

P
Peter Zijlstra 已提交
8823 8824
	for_each_task_context_nr(ctxn) {
		ret = perf_event_init_context(child, ctxn);
P
Peter Zijlstra 已提交
8825 8826
		if (ret) {
			perf_event_free_task(child);
P
Peter Zijlstra 已提交
8827
			return ret;
P
Peter Zijlstra 已提交
8828
		}
P
Peter Zijlstra 已提交
8829 8830 8831 8832 8833
	}

	return 0;
}

8834 8835
static void __init perf_event_init_all_cpus(void)
{
8836
	struct swevent_htable *swhash;
8837 8838 8839
	int cpu;

	for_each_possible_cpu(cpu) {
8840 8841
		swhash = &per_cpu(swevent_htable, cpu);
		mutex_init(&swhash->hlist_mutex);
8842
		INIT_LIST_HEAD(&per_cpu(active_ctx_list, cpu));
8843 8844 8845
	}
}

8846
static void perf_event_init_cpu(int cpu)
T
Thomas Gleixner 已提交
8847
{
P
Peter Zijlstra 已提交
8848
	struct swevent_htable *swhash = &per_cpu(swevent_htable, cpu);
T
Thomas Gleixner 已提交
8849

8850
	mutex_lock(&swhash->hlist_mutex);
8851
	swhash->online = true;
8852
	if (swhash->hlist_refcount > 0) {
8853 8854
		struct swevent_hlist *hlist;

8855 8856 8857
		hlist = kzalloc_node(sizeof(*hlist), GFP_KERNEL, cpu_to_node(cpu));
		WARN_ON(!hlist);
		rcu_assign_pointer(swhash->swevent_hlist, hlist);
8858
	}
8859
	mutex_unlock(&swhash->hlist_mutex);
T
Thomas Gleixner 已提交
8860 8861
}

P
Peter Zijlstra 已提交
8862
#if defined CONFIG_HOTPLUG_CPU || defined CONFIG_KEXEC
P
Peter Zijlstra 已提交
8863
static void __perf_event_exit_context(void *__info)
T
Thomas Gleixner 已提交
8864
{
8865
	struct remove_event re = { .detach_group = true };
P
Peter Zijlstra 已提交
8866
	struct perf_event_context *ctx = __info;
T
Thomas Gleixner 已提交
8867

P
Peter Zijlstra 已提交
8868
	rcu_read_lock();
8869 8870
	list_for_each_entry_rcu(re.event, &ctx->event_list, event_entry)
		__perf_remove_from_context(&re);
P
Peter Zijlstra 已提交
8871
	rcu_read_unlock();
T
Thomas Gleixner 已提交
8872
}
P
Peter Zijlstra 已提交
8873 8874 8875 8876 8877 8878 8879 8880 8881

static void perf_event_exit_cpu_context(int cpu)
{
	struct perf_event_context *ctx;
	struct pmu *pmu;
	int idx;

	idx = srcu_read_lock(&pmus_srcu);
	list_for_each_entry_rcu(pmu, &pmus, entry) {
8882
		ctx = &per_cpu_ptr(pmu->pmu_cpu_context, cpu)->ctx;
P
Peter Zijlstra 已提交
8883 8884 8885 8886 8887 8888 8889 8890

		mutex_lock(&ctx->mutex);
		smp_call_function_single(cpu, __perf_event_exit_context, ctx, 1);
		mutex_unlock(&ctx->mutex);
	}
	srcu_read_unlock(&pmus_srcu, idx);
}

8891
static void perf_event_exit_cpu(int cpu)
T
Thomas Gleixner 已提交
8892
{
8893
	struct swevent_htable *swhash = &per_cpu(swevent_htable, cpu);
8894

P
Peter Zijlstra 已提交
8895 8896
	perf_event_exit_cpu_context(cpu);

8897
	mutex_lock(&swhash->hlist_mutex);
8898
	swhash->online = false;
8899 8900
	swevent_hlist_release(swhash);
	mutex_unlock(&swhash->hlist_mutex);
T
Thomas Gleixner 已提交
8901 8902
}
#else
8903
static inline void perf_event_exit_cpu(int cpu) { }
T
Thomas Gleixner 已提交
8904 8905
#endif

P
Peter Zijlstra 已提交
8906 8907 8908 8909 8910 8911 8912 8913 8914 8915 8916 8917 8918 8919 8920 8921 8922 8923 8924 8925
static int
perf_reboot(struct notifier_block *notifier, unsigned long val, void *v)
{
	int cpu;

	for_each_online_cpu(cpu)
		perf_event_exit_cpu(cpu);

	return NOTIFY_OK;
}

/*
 * Run the perf reboot notifier at the very last possible moment so that
 * the generic watchdog code runs as long as possible.
 */
static struct notifier_block perf_reboot_notifier = {
	.notifier_call = perf_reboot,
	.priority = INT_MIN,
};

8926
static int
T
Thomas Gleixner 已提交
8927 8928 8929 8930
perf_cpu_notify(struct notifier_block *self, unsigned long action, void *hcpu)
{
	unsigned int cpu = (long)hcpu;

8931
	switch (action & ~CPU_TASKS_FROZEN) {
T
Thomas Gleixner 已提交
8932 8933

	case CPU_UP_PREPARE:
P
Peter Zijlstra 已提交
8934
	case CPU_DOWN_FAILED:
8935
		perf_event_init_cpu(cpu);
T
Thomas Gleixner 已提交
8936 8937
		break;

P
Peter Zijlstra 已提交
8938
	case CPU_UP_CANCELED:
T
Thomas Gleixner 已提交
8939
	case CPU_DOWN_PREPARE:
8940
		perf_event_exit_cpu(cpu);
T
Thomas Gleixner 已提交
8941 8942 8943 8944 8945 8946 8947 8948
		break;
	default:
		break;
	}

	return NOTIFY_OK;
}

8949
void __init perf_event_init(void)
T
Thomas Gleixner 已提交
8950
{
8951 8952
	int ret;

P
Peter Zijlstra 已提交
8953 8954
	idr_init(&pmu_idr);

8955
	perf_event_init_all_cpus();
8956
	init_srcu_struct(&pmus_srcu);
P
Peter Zijlstra 已提交
8957 8958 8959
	perf_pmu_register(&perf_swevent, "software", PERF_TYPE_SOFTWARE);
	perf_pmu_register(&perf_cpu_clock, NULL, -1);
	perf_pmu_register(&perf_task_clock, NULL, -1);
8960 8961
	perf_tp_register();
	perf_cpu_notifier(perf_cpu_notify);
P
Peter Zijlstra 已提交
8962
	register_reboot_notifier(&perf_reboot_notifier);
8963 8964 8965

	ret = init_hw_breakpoint();
	WARN(ret, "hw_breakpoint initialization failed with: %d", ret);
8966 8967 8968

	/* do not patch jump label more than once per second */
	jump_label_rate_limit(&perf_sched_events, HZ);
8969 8970 8971 8972 8973 8974 8975

	/*
	 * Build time assertion that we keep the data_head at the intended
	 * location.  IOW, validation we got the __reserved[] size right.
	 */
	BUILD_BUG_ON((offsetof(struct perf_event_mmap_page, data_head))
		     != 1024);
T
Thomas Gleixner 已提交
8976
}
P
Peter Zijlstra 已提交
8977

8978 8979 8980 8981 8982 8983 8984 8985 8986 8987 8988 8989
ssize_t perf_event_sysfs_show(struct device *dev, struct device_attribute *attr,
			      char *page)
{
	struct perf_pmu_events_attr *pmu_attr =
		container_of(attr, struct perf_pmu_events_attr, attr);

	if (pmu_attr->event_str)
		return sprintf(page, "%s\n", pmu_attr->event_str);

	return 0;
}

P
Peter Zijlstra 已提交
8990 8991 8992 8993 8994 8995 8996 8997 8998 8999 9000 9001 9002 9003 9004 9005 9006 9007 9008 9009 9010 9011 9012 9013 9014 9015 9016
static int __init perf_event_sysfs_init(void)
{
	struct pmu *pmu;
	int ret;

	mutex_lock(&pmus_lock);

	ret = bus_register(&pmu_bus);
	if (ret)
		goto unlock;

	list_for_each_entry(pmu, &pmus, entry) {
		if (!pmu->name || pmu->type < 0)
			continue;

		ret = pmu_dev_alloc(pmu);
		WARN(ret, "Failed to register pmu: %s, reason %d\n", pmu->name, ret);
	}
	pmu_bus_running = 1;
	ret = 0;

unlock:
	mutex_unlock(&pmus_lock);

	return ret;
}
device_initcall(perf_event_sysfs_init);
S
Stephane Eranian 已提交
9017 9018

#ifdef CONFIG_CGROUP_PERF
9019 9020
static struct cgroup_subsys_state *
perf_cgroup_css_alloc(struct cgroup_subsys_state *parent_css)
S
Stephane Eranian 已提交
9021 9022 9023
{
	struct perf_cgroup *jc;

9024
	jc = kzalloc(sizeof(*jc), GFP_KERNEL);
S
Stephane Eranian 已提交
9025 9026 9027 9028 9029 9030 9031 9032 9033 9034 9035 9036
	if (!jc)
		return ERR_PTR(-ENOMEM);

	jc->info = alloc_percpu(struct perf_cgroup_info);
	if (!jc->info) {
		kfree(jc);
		return ERR_PTR(-ENOMEM);
	}

	return &jc->css;
}

9037
static void perf_cgroup_css_free(struct cgroup_subsys_state *css)
S
Stephane Eranian 已提交
9038
{
9039 9040
	struct perf_cgroup *jc = container_of(css, struct perf_cgroup, css);

S
Stephane Eranian 已提交
9041 9042 9043 9044 9045 9046 9047 9048 9049 9050 9051
	free_percpu(jc->info);
	kfree(jc);
}

static int __perf_cgroup_move(void *info)
{
	struct task_struct *task = info;
	perf_cgroup_switch(task, PERF_CGROUP_SWOUT | PERF_CGROUP_SWIN);
	return 0;
}

9052 9053
static void perf_cgroup_attach(struct cgroup_subsys_state *css,
			       struct cgroup_taskset *tset)
S
Stephane Eranian 已提交
9054
{
9055 9056
	struct task_struct *task;

9057
	cgroup_taskset_for_each(task, tset)
9058
		task_function_call(task, __perf_cgroup_move, task);
S
Stephane Eranian 已提交
9059 9060
}

9061 9062
static void perf_cgroup_exit(struct cgroup_subsys_state *css,
			     struct cgroup_subsys_state *old_css,
9063
			     struct task_struct *task)
S
Stephane Eranian 已提交
9064 9065 9066 9067 9068 9069 9070 9071 9072
{
	/*
	 * cgroup_exit() is called in the copy_process() failure path.
	 * Ignore this case since the task hasn't ran yet, this avoids
	 * trying to poke a half freed task state from generic code.
	 */
	if (!(task->flags & PF_EXITING))
		return;

9073
	task_function_call(task, __perf_cgroup_move, task);
S
Stephane Eranian 已提交
9074 9075
}

9076
struct cgroup_subsys perf_event_cgrp_subsys = {
9077 9078
	.css_alloc	= perf_cgroup_css_alloc,
	.css_free	= perf_cgroup_css_free,
9079
	.exit		= perf_cgroup_exit,
9080
	.attach		= perf_cgroup_attach,
S
Stephane Eranian 已提交
9081 9082
};
#endif /* CONFIG_CGROUP_PERF */