perf_counter.c 71.2 KB
Newer Older
T
Thomas Gleixner 已提交
1 2 3 4 5 6
/*
 * Performance counter core code
 *
 *  Copyright(C) 2008 Thomas Gleixner <tglx@linutronix.de>
 *  Copyright(C) 2008 Red Hat, Inc., Ingo Molnar
 *
7 8
 *
 *  For licensing details see kernel-base/COPYING
T
Thomas Gleixner 已提交
9 10 11
 */

#include <linux/fs.h>
12
#include <linux/mm.h>
T
Thomas Gleixner 已提交
13 14
#include <linux/cpu.h>
#include <linux/smp.h>
15
#include <linux/file.h>
T
Thomas Gleixner 已提交
16 17 18 19
#include <linux/poll.h>
#include <linux/sysfs.h>
#include <linux/ptrace.h>
#include <linux/percpu.h>
20 21 22
#include <linux/vmstat.h>
#include <linux/hardirq.h>
#include <linux/rculist.h>
T
Thomas Gleixner 已提交
23 24 25
#include <linux/uaccess.h>
#include <linux/syscalls.h>
#include <linux/anon_inodes.h>
I
Ingo Molnar 已提交
26
#include <linux/kernel_stat.h>
T
Thomas Gleixner 已提交
27
#include <linux/perf_counter.h>
28
#include <linux/dcache.h>
T
Thomas Gleixner 已提交
29

30 31
#include <asm/irq_regs.h>

T
Thomas Gleixner 已提交
32 33 34 35 36
/*
 * Each CPU has a list of per CPU counters:
 */
DEFINE_PER_CPU(struct perf_cpu_context, perf_cpu_context);

37
int perf_max_counters __read_mostly = 1;
T
Thomas Gleixner 已提交
38 39 40 41 42 43 44 45 46 47 48
static int perf_reserved_percpu __read_mostly;
static int perf_overcommit __read_mostly = 1;

/*
 * Mutex for (sysadmin-configurable) counter reservations:
 */
static DEFINE_MUTEX(perf_resource_mutex);

/*
 * Architecture provided APIs - weak aliases:
 */
49
extern __weak const struct hw_perf_counter_ops *
I
Ingo Molnar 已提交
50
hw_perf_counter_init(struct perf_counter *counter)
T
Thomas Gleixner 已提交
51
{
52
	return NULL;
T
Thomas Gleixner 已提交
53 54
}

55
u64 __weak hw_perf_save_disable(void)		{ return 0; }
56
void __weak hw_perf_restore(u64 ctrl)		{ barrier(); }
57
void __weak hw_perf_counter_setup(int cpu)	{ barrier(); }
58 59 60 61 62 63
int __weak hw_perf_group_sched_in(struct perf_counter *group_leader,
	       struct perf_cpu_context *cpuctx,
	       struct perf_counter_context *ctx, int cpu)
{
	return 0;
}
T
Thomas Gleixner 已提交
64

65 66
void __weak perf_counter_print_debug(void)	{ }

67 68 69 70 71 72 73 74 75 76 77 78
static void
list_add_counter(struct perf_counter *counter, struct perf_counter_context *ctx)
{
	struct perf_counter *group_leader = counter->group_leader;

	/*
	 * Depending on whether it is a standalone or sibling counter,
	 * add it straight to the context's counter list, or to the group
	 * leader's sibling list:
	 */
	if (counter->group_leader == counter)
		list_add_tail(&counter->list_entry, &ctx->counter_list);
P
Peter Zijlstra 已提交
79
	else {
80
		list_add_tail(&counter->list_entry, &group_leader->sibling_list);
P
Peter Zijlstra 已提交
81 82
		group_leader->nr_siblings++;
	}
P
Peter Zijlstra 已提交
83 84

	list_add_rcu(&counter->event_entry, &ctx->event_list);
85 86 87 88 89 90 91 92
}

static void
list_del_counter(struct perf_counter *counter, struct perf_counter_context *ctx)
{
	struct perf_counter *sibling, *tmp;

	list_del_init(&counter->list_entry);
P
Peter Zijlstra 已提交
93
	list_del_rcu(&counter->event_entry);
94

P
Peter Zijlstra 已提交
95 96 97
	if (counter->group_leader != counter)
		counter->group_leader->nr_siblings--;

98 99 100 101 102 103 104 105
	/*
	 * If this was a group counter with sibling counters then
	 * upgrade the siblings to singleton counters by adding them
	 * to the context list directly:
	 */
	list_for_each_entry_safe(sibling, tmp,
				 &counter->sibling_list, list_entry) {

106
		list_move_tail(&sibling->list_entry, &ctx->counter_list);
107 108 109 110
		sibling->group_leader = sibling;
	}
}

111 112 113 114 115 116 117 118 119
static void
counter_sched_out(struct perf_counter *counter,
		  struct perf_cpu_context *cpuctx,
		  struct perf_counter_context *ctx)
{
	if (counter->state != PERF_COUNTER_STATE_ACTIVE)
		return;

	counter->state = PERF_COUNTER_STATE_INACTIVE;
120
	counter->tstamp_stopped = ctx->time_now;
121 122 123 124 125 126 127 128 129 130
	counter->hw_ops->disable(counter);
	counter->oncpu = -1;

	if (!is_software_counter(counter))
		cpuctx->active_oncpu--;
	ctx->nr_active--;
	if (counter->hw_event.exclusive || !cpuctx->active_oncpu)
		cpuctx->exclusive = 0;
}

131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152
static void
group_sched_out(struct perf_counter *group_counter,
		struct perf_cpu_context *cpuctx,
		struct perf_counter_context *ctx)
{
	struct perf_counter *counter;

	if (group_counter->state != PERF_COUNTER_STATE_ACTIVE)
		return;

	counter_sched_out(group_counter, cpuctx, ctx);

	/*
	 * Schedule out siblings (if any):
	 */
	list_for_each_entry(counter, &group_counter->sibling_list, list_entry)
		counter_sched_out(counter, cpuctx, ctx);

	if (group_counter->hw_event.exclusive)
		cpuctx->exclusive = 0;
}

T
Thomas Gleixner 已提交
153 154 155 156 157 158
/*
 * Cross CPU call to remove a performance counter
 *
 * We disable the counter on the hardware level first. After that we
 * remove it from the context list.
 */
159
static void __perf_counter_remove_from_context(void *info)
T
Thomas Gleixner 已提交
160 161 162 163
{
	struct perf_cpu_context *cpuctx = &__get_cpu_var(perf_cpu_context);
	struct perf_counter *counter = info;
	struct perf_counter_context *ctx = counter->ctx;
164
	unsigned long flags;
165
	u64 perf_flags;
T
Thomas Gleixner 已提交
166 167 168 169 170 171 172 173 174

	/*
	 * 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.
	 */
	if (ctx->task && cpuctx->task_ctx != ctx)
		return;

I
Ingo Molnar 已提交
175 176
	curr_rq_lock_irq_save(&flags);
	spin_lock(&ctx->lock);
T
Thomas Gleixner 已提交
177

178 179 180
	counter_sched_out(counter, cpuctx, ctx);

	counter->task = NULL;
T
Thomas Gleixner 已提交
181 182 183 184 185 186
	ctx->nr_counters--;

	/*
	 * Protect the list operation against NMI by disabling the
	 * counters on a global level. NOP for non NMI based counters.
	 */
187
	perf_flags = hw_perf_save_disable();
188
	list_del_counter(counter, ctx);
189
	hw_perf_restore(perf_flags);
T
Thomas Gleixner 已提交
190 191 192 193 194 195 196 197 198 199 200

	if (!ctx->task) {
		/*
		 * Allow more per task counters with respect to the
		 * reservation:
		 */
		cpuctx->max_pertask =
			min(perf_max_counters - ctx->nr_counters,
			    perf_max_counters - perf_reserved_percpu);
	}

I
Ingo Molnar 已提交
201 202
	spin_unlock(&ctx->lock);
	curr_rq_unlock_irq_restore(&flags);
T
Thomas Gleixner 已提交
203 204 205 206 207 208
}


/*
 * Remove the counter from a task's (or a CPU's) list of counters.
 *
209
 * Must be called with counter->mutex and ctx->mutex held.
T
Thomas Gleixner 已提交
210 211 212 213
 *
 * CPU counters are removed with a smp call. For task counters we only
 * call when the task is on a CPU.
 */
214
static void perf_counter_remove_from_context(struct perf_counter *counter)
T
Thomas Gleixner 已提交
215 216 217 218 219 220 221 222 223 224
{
	struct perf_counter_context *ctx = counter->ctx;
	struct task_struct *task = ctx->task;

	if (!task) {
		/*
		 * Per cpu counters are removed via an smp call and
		 * the removal is always sucessful.
		 */
		smp_call_function_single(counter->cpu,
225
					 __perf_counter_remove_from_context,
T
Thomas Gleixner 已提交
226 227 228 229 230
					 counter, 1);
		return;
	}

retry:
231
	task_oncpu_function_call(task, __perf_counter_remove_from_context,
T
Thomas Gleixner 已提交
232 233 234 235 236 237
				 counter);

	spin_lock_irq(&ctx->lock);
	/*
	 * If the context is active we need to retry the smp call.
	 */
238
	if (ctx->nr_active && !list_empty(&counter->list_entry)) {
T
Thomas Gleixner 已提交
239 240 241 242 243 244
		spin_unlock_irq(&ctx->lock);
		goto retry;
	}

	/*
	 * The lock prevents that this context is scheduled in so we
245
	 * can remove the counter safely, if the call above did not
T
Thomas Gleixner 已提交
246 247
	 * succeed.
	 */
248
	if (!list_empty(&counter->list_entry)) {
T
Thomas Gleixner 已提交
249
		ctx->nr_counters--;
250
		list_del_counter(counter, ctx);
T
Thomas Gleixner 已提交
251 252 253 254 255
		counter->task = NULL;
	}
	spin_unlock_irq(&ctx->lock);
}

256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309
/*
 * Get the current time for this context.
 * If this is a task context, we use the task's task clock,
 * or for a per-cpu context, we use the cpu clock.
 */
static u64 get_context_time(struct perf_counter_context *ctx, int update)
{
	struct task_struct *curr = ctx->task;

	if (!curr)
		return cpu_clock(smp_processor_id());

	return __task_delta_exec(curr, update) + curr->se.sum_exec_runtime;
}

/*
 * Update the record of the current time in a context.
 */
static void update_context_time(struct perf_counter_context *ctx, int update)
{
	ctx->time_now = get_context_time(ctx, update) - ctx->time_lost;
}

/*
 * Update the total_time_enabled and total_time_running fields for a counter.
 */
static void update_counter_times(struct perf_counter *counter)
{
	struct perf_counter_context *ctx = counter->ctx;
	u64 run_end;

	if (counter->state >= PERF_COUNTER_STATE_INACTIVE) {
		counter->total_time_enabled = ctx->time_now -
			counter->tstamp_enabled;
		if (counter->state == PERF_COUNTER_STATE_INACTIVE)
			run_end = counter->tstamp_stopped;
		else
			run_end = ctx->time_now;
		counter->total_time_running = run_end - counter->tstamp_running;
	}
}

/*
 * Update total_time_enabled and total_time_running for all counters in a group.
 */
static void update_group_times(struct perf_counter *leader)
{
	struct perf_counter *counter;

	update_counter_times(leader);
	list_for_each_entry(counter, &leader->sibling_list, list_entry)
		update_counter_times(counter);
}

310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334
/*
 * Cross CPU call to disable a performance counter
 */
static void __perf_counter_disable(void *info)
{
	struct perf_counter *counter = info;
	struct perf_cpu_context *cpuctx = &__get_cpu_var(perf_cpu_context);
	struct perf_counter_context *ctx = counter->ctx;
	unsigned long flags;

	/*
	 * If this is a per-task counter, need to check whether this
	 * counter's task is the current task on this cpu.
	 */
	if (ctx->task && cpuctx->task_ctx != ctx)
		return;

	curr_rq_lock_irq_save(&flags);
	spin_lock(&ctx->lock);

	/*
	 * If the counter is on, turn it off.
	 * If it is in error state, leave it in error state.
	 */
	if (counter->state >= PERF_COUNTER_STATE_INACTIVE) {
335 336
		update_context_time(ctx, 1);
		update_counter_times(counter);
337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380
		if (counter == counter->group_leader)
			group_sched_out(counter, cpuctx, ctx);
		else
			counter_sched_out(counter, cpuctx, ctx);
		counter->state = PERF_COUNTER_STATE_OFF;
	}

	spin_unlock(&ctx->lock);
	curr_rq_unlock_irq_restore(&flags);
}

/*
 * Disable a counter.
 */
static void perf_counter_disable(struct perf_counter *counter)
{
	struct perf_counter_context *ctx = counter->ctx;
	struct task_struct *task = ctx->task;

	if (!task) {
		/*
		 * Disable the counter on the cpu that it's on
		 */
		smp_call_function_single(counter->cpu, __perf_counter_disable,
					 counter, 1);
		return;
	}

 retry:
	task_oncpu_function_call(task, __perf_counter_disable, counter);

	spin_lock_irq(&ctx->lock);
	/*
	 * If the counter is still active, we need to retry the cross-call.
	 */
	if (counter->state == PERF_COUNTER_STATE_ACTIVE) {
		spin_unlock_irq(&ctx->lock);
		goto retry;
	}

	/*
	 * Since we have the lock this context can't be scheduled
	 * in, so we can change the state safely.
	 */
381 382
	if (counter->state == PERF_COUNTER_STATE_INACTIVE) {
		update_counter_times(counter);
383
		counter->state = PERF_COUNTER_STATE_OFF;
384
	}
385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406

	spin_unlock_irq(&ctx->lock);
}

/*
 * Disable a counter and all its children.
 */
static void perf_counter_disable_family(struct perf_counter *counter)
{
	struct perf_counter *child;

	perf_counter_disable(counter);

	/*
	 * Lock the mutex to protect the list of children
	 */
	mutex_lock(&counter->mutex);
	list_for_each_entry(child, &counter->child_list, child_list)
		perf_counter_disable(child);
	mutex_unlock(&counter->mutex);
}

407 408 409 410 411 412
static int
counter_sched_in(struct perf_counter *counter,
		 struct perf_cpu_context *cpuctx,
		 struct perf_counter_context *ctx,
		 int cpu)
{
413
	if (counter->state <= PERF_COUNTER_STATE_OFF)
414 415 416 417 418 419 420 421 422 423 424 425 426 427 428
		return 0;

	counter->state = PERF_COUNTER_STATE_ACTIVE;
	counter->oncpu = cpu;	/* TODO: put 'cpu' into cpuctx->cpu */
	/*
	 * The new state must be visible before we turn it on in the hardware:
	 */
	smp_wmb();

	if (counter->hw_ops->enable(counter)) {
		counter->state = PERF_COUNTER_STATE_INACTIVE;
		counter->oncpu = -1;
		return -EAGAIN;
	}

429 430
	counter->tstamp_running += ctx->time_now - counter->tstamp_stopped;

431 432
	if (!is_software_counter(counter))
		cpuctx->active_oncpu++;
433 434
	ctx->nr_active++;

435 436 437
	if (counter->hw_event.exclusive)
		cpuctx->exclusive = 1;

438 439 440
	return 0;
}

441 442 443 444 445 446 447 448 449 450
/*
 * Return 1 for a group consisting entirely of software counters,
 * 0 if the group contains any hardware counters.
 */
static int is_software_only_group(struct perf_counter *leader)
{
	struct perf_counter *counter;

	if (!is_software_counter(leader))
		return 0;
P
Peter Zijlstra 已提交
451

452 453 454
	list_for_each_entry(counter, &leader->sibling_list, list_entry)
		if (!is_software_counter(counter))
			return 0;
P
Peter Zijlstra 已提交
455

456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489
	return 1;
}

/*
 * Work out whether we can put this counter group on the CPU now.
 */
static int group_can_go_on(struct perf_counter *counter,
			   struct perf_cpu_context *cpuctx,
			   int can_add_hw)
{
	/*
	 * Groups consisting entirely of software counters can always go on.
	 */
	if (is_software_only_group(counter))
		return 1;
	/*
	 * If an exclusive group is already on, no other hardware
	 * counters can go on.
	 */
	if (cpuctx->exclusive)
		return 0;
	/*
	 * If this group is exclusive and there are already
	 * counters on the CPU, it can't go on.
	 */
	if (counter->hw_event.exclusive && cpuctx->active_oncpu)
		return 0;
	/*
	 * Otherwise, try to add it if all previous groups were able
	 * to go on.
	 */
	return can_add_hw;
}

490 491 492 493 494 495 496 497 498 499 500
static void add_counter_to_ctx(struct perf_counter *counter,
			       struct perf_counter_context *ctx)
{
	list_add_counter(counter, ctx);
	ctx->nr_counters++;
	counter->prev_state = PERF_COUNTER_STATE_OFF;
	counter->tstamp_enabled = ctx->time_now;
	counter->tstamp_running = ctx->time_now;
	counter->tstamp_stopped = ctx->time_now;
}

T
Thomas Gleixner 已提交
501
/*
502
 * Cross CPU call to install and enable a performance counter
T
Thomas Gleixner 已提交
503 504 505 506 507 508
 */
static void __perf_install_in_context(void *info)
{
	struct perf_cpu_context *cpuctx = &__get_cpu_var(perf_cpu_context);
	struct perf_counter *counter = info;
	struct perf_counter_context *ctx = counter->ctx;
509
	struct perf_counter *leader = counter->group_leader;
T
Thomas Gleixner 已提交
510
	int cpu = smp_processor_id();
511
	unsigned long flags;
512
	u64 perf_flags;
513
	int err;
T
Thomas Gleixner 已提交
514 515 516 517 518 519 520 521 522

	/*
	 * 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.
	 */
	if (ctx->task && cpuctx->task_ctx != ctx)
		return;

I
Ingo Molnar 已提交
523 524
	curr_rq_lock_irq_save(&flags);
	spin_lock(&ctx->lock);
525
	update_context_time(ctx, 1);
T
Thomas Gleixner 已提交
526 527 528 529 530

	/*
	 * Protect the list operation against NMI by disabling the
	 * counters on a global level. NOP for non NMI based counters.
	 */
531
	perf_flags = hw_perf_save_disable();
T
Thomas Gleixner 已提交
532

533
	add_counter_to_ctx(counter, ctx);
T
Thomas Gleixner 已提交
534

535 536 537 538 539 540 541 542
	/*
	 * Don't put the counter on if it is disabled or if
	 * it is in a group and the group isn't on.
	 */
	if (counter->state != PERF_COUNTER_STATE_INACTIVE ||
	    (leader != counter && leader->state != PERF_COUNTER_STATE_ACTIVE))
		goto unlock;

543 544 545 546 547
	/*
	 * An exclusive counter can't go on if there are already active
	 * hardware counters, and no hardware counter can go on if there
	 * is already an exclusive counter on.
	 */
548
	if (!group_can_go_on(counter, cpuctx, 1))
549 550 551 552
		err = -EEXIST;
	else
		err = counter_sched_in(counter, cpuctx, ctx, cpu);

553 554 555 556 557 558 559 560
	if (err) {
		/*
		 * This counter couldn't go on.  If it is in a group
		 * then we have to pull the whole group off.
		 * If the counter group is pinned then put it in error state.
		 */
		if (leader != counter)
			group_sched_out(leader, cpuctx, ctx);
561 562
		if (leader->hw_event.pinned) {
			update_group_times(leader);
563
			leader->state = PERF_COUNTER_STATE_ERROR;
564
		}
565
	}
T
Thomas Gleixner 已提交
566

567
	if (!err && !ctx->task && cpuctx->max_pertask)
T
Thomas Gleixner 已提交
568 569
		cpuctx->max_pertask--;

570
 unlock:
571 572
	hw_perf_restore(perf_flags);

I
Ingo Molnar 已提交
573 574
	spin_unlock(&ctx->lock);
	curr_rq_unlock_irq_restore(&flags);
T
Thomas Gleixner 已提交
575 576 577 578 579 580 581 582 583 584 585
}

/*
 * Attach a performance counter to a context
 *
 * First we add the counter to the list with the hardware enable bit
 * in counter->hw_config cleared.
 *
 * If the counter is attached to a task which is on a CPU we use a smp
 * call to enable it in the task context. The task might have been
 * scheduled away, but we check this in the smp call again.
586 587
 *
 * Must be called with ctx->mutex held.
T
Thomas Gleixner 已提交
588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614
 */
static void
perf_install_in_context(struct perf_counter_context *ctx,
			struct perf_counter *counter,
			int cpu)
{
	struct task_struct *task = ctx->task;

	if (!task) {
		/*
		 * Per cpu counters are installed via an smp call and
		 * the install is always sucessful.
		 */
		smp_call_function_single(cpu, __perf_install_in_context,
					 counter, 1);
		return;
	}

	counter->task = task;
retry:
	task_oncpu_function_call(task, __perf_install_in_context,
				 counter);

	spin_lock_irq(&ctx->lock);
	/*
	 * we need to retry the smp call.
	 */
615
	if (ctx->is_active && list_empty(&counter->list_entry)) {
T
Thomas Gleixner 已提交
616 617 618 619 620 621 622 623 624
		spin_unlock_irq(&ctx->lock);
		goto retry;
	}

	/*
	 * The lock prevents that this context is scheduled in so we
	 * can add the counter safely, if it the call above did not
	 * succeed.
	 */
625 626
	if (list_empty(&counter->list_entry))
		add_counter_to_ctx(counter, ctx);
T
Thomas Gleixner 已提交
627 628 629
	spin_unlock_irq(&ctx->lock);
}

630 631 632 633
/*
 * Cross CPU call to enable a performance counter
 */
static void __perf_counter_enable(void *info)
634
{
635 636 637 638 639 640
	struct perf_counter *counter = info;
	struct perf_cpu_context *cpuctx = &__get_cpu_var(perf_cpu_context);
	struct perf_counter_context *ctx = counter->ctx;
	struct perf_counter *leader = counter->group_leader;
	unsigned long flags;
	int err;
641

642 643 644 645 646
	/*
	 * If this is a per-task counter, need to check whether this
	 * counter's task is the current task on this cpu.
	 */
	if (ctx->task && cpuctx->task_ctx != ctx)
647 648
		return;

649 650
	curr_rq_lock_irq_save(&flags);
	spin_lock(&ctx->lock);
651
	update_context_time(ctx, 1);
652

653
	counter->prev_state = counter->state;
654 655 656
	if (counter->state >= PERF_COUNTER_STATE_INACTIVE)
		goto unlock;
	counter->state = PERF_COUNTER_STATE_INACTIVE;
657
	counter->tstamp_enabled = ctx->time_now - counter->total_time_enabled;
658 659

	/*
660 661
	 * If the counter is in a group and isn't the group leader,
	 * then don't put it on unless the group is on.
662
	 */
663 664
	if (leader != counter && leader->state != PERF_COUNTER_STATE_ACTIVE)
		goto unlock;
665

666 667 668 669 670 671 672 673 674 675 676 677 678
	if (!group_can_go_on(counter, cpuctx, 1))
		err = -EEXIST;
	else
		err = counter_sched_in(counter, cpuctx, ctx,
				       smp_processor_id());

	if (err) {
		/*
		 * If this counter can't go on and it's part of a
		 * group, then the whole group has to come off.
		 */
		if (leader != counter)
			group_sched_out(leader, cpuctx, ctx);
679 680
		if (leader->hw_event.pinned) {
			update_group_times(leader);
681
			leader->state = PERF_COUNTER_STATE_ERROR;
682
		}
683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737
	}

 unlock:
	spin_unlock(&ctx->lock);
	curr_rq_unlock_irq_restore(&flags);
}

/*
 * Enable a counter.
 */
static void perf_counter_enable(struct perf_counter *counter)
{
	struct perf_counter_context *ctx = counter->ctx;
	struct task_struct *task = ctx->task;

	if (!task) {
		/*
		 * Enable the counter on the cpu that it's on
		 */
		smp_call_function_single(counter->cpu, __perf_counter_enable,
					 counter, 1);
		return;
	}

	spin_lock_irq(&ctx->lock);
	if (counter->state >= PERF_COUNTER_STATE_INACTIVE)
		goto out;

	/*
	 * If the counter is in error state, clear that first.
	 * That way, if we see the counter in error state below, we
	 * know that it has gone back into error state, as distinct
	 * from the task having been scheduled away before the
	 * cross-call arrived.
	 */
	if (counter->state == PERF_COUNTER_STATE_ERROR)
		counter->state = PERF_COUNTER_STATE_OFF;

 retry:
	spin_unlock_irq(&ctx->lock);
	task_oncpu_function_call(task, __perf_counter_enable, counter);

	spin_lock_irq(&ctx->lock);

	/*
	 * If the context is active and the counter is still off,
	 * we need to retry the cross-call.
	 */
	if (ctx->is_active && counter->state == PERF_COUNTER_STATE_OFF)
		goto retry;

	/*
	 * Since we have the lock this context can't be scheduled
	 * in, so we can change the state safely.
	 */
738
	if (counter->state == PERF_COUNTER_STATE_OFF) {
739
		counter->state = PERF_COUNTER_STATE_INACTIVE;
740 741 742
		counter->tstamp_enabled = ctx->time_now -
			counter->total_time_enabled;
	}
743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762
 out:
	spin_unlock_irq(&ctx->lock);
}

/*
 * Enable a counter and all its children.
 */
static void perf_counter_enable_family(struct perf_counter *counter)
{
	struct perf_counter *child;

	perf_counter_enable(counter);

	/*
	 * Lock the mutex to protect the list of children
	 */
	mutex_lock(&counter->mutex);
	list_for_each_entry(child, &counter->child_list, child_list)
		perf_counter_enable(child);
	mutex_unlock(&counter->mutex);
763 764
}

765 766 767 768
void __perf_counter_sched_out(struct perf_counter_context *ctx,
			      struct perf_cpu_context *cpuctx)
{
	struct perf_counter *counter;
769
	u64 flags;
770

771 772
	spin_lock(&ctx->lock);
	ctx->is_active = 0;
773
	if (likely(!ctx->nr_counters))
774
		goto out;
775
	update_context_time(ctx, 0);
776

777
	flags = hw_perf_save_disable();
778 779 780 781
	if (ctx->nr_active) {
		list_for_each_entry(counter, &ctx->counter_list, list_entry)
			group_sched_out(counter, cpuctx, ctx);
	}
782
	hw_perf_restore(flags);
783
 out:
784 785 786
	spin_unlock(&ctx->lock);
}

T
Thomas Gleixner 已提交
787 788 789 790 791 792
/*
 * Called from scheduler to remove the counters of the current task,
 * with interrupts disabled.
 *
 * We stop each counter and update the counter value in counter->count.
 *
I
Ingo Molnar 已提交
793
 * This does not protect us against NMI, but disable()
T
Thomas Gleixner 已提交
794 795 796 797 798 799 800 801
 * sets the disabled bit in the control field of counter _before_
 * accessing the counter control register. If a NMI hits, then it will
 * not restart the counter.
 */
void perf_counter_task_sched_out(struct task_struct *task, int cpu)
{
	struct perf_cpu_context *cpuctx = &per_cpu(perf_cpu_context, cpu);
	struct perf_counter_context *ctx = &task->perf_counter_ctx;
802
	struct pt_regs *regs;
T
Thomas Gleixner 已提交
803 804 805 806

	if (likely(!cpuctx->task_ctx))
		return;

807 808
	regs = task_pt_regs(task);
	perf_swcounter_event(PERF_COUNT_CONTEXT_SWITCHES, 1, 1, regs);
809 810
	__perf_counter_sched_out(ctx, cpuctx);

T
Thomas Gleixner 已提交
811 812 813
	cpuctx->task_ctx = NULL;
}

814
static void perf_counter_cpu_sched_out(struct perf_cpu_context *cpuctx)
815
{
816
	__perf_counter_sched_out(&cpuctx->ctx, cpuctx);
817 818
}

I
Ingo Molnar 已提交
819
static int
820 821 822 823 824
group_sched_in(struct perf_counter *group_counter,
	       struct perf_cpu_context *cpuctx,
	       struct perf_counter_context *ctx,
	       int cpu)
{
825
	struct perf_counter *counter, *partial_group;
826 827 828 829 830 831 832 833
	int ret;

	if (group_counter->state == PERF_COUNTER_STATE_OFF)
		return 0;

	ret = hw_perf_group_sched_in(group_counter, cpuctx, ctx, cpu);
	if (ret)
		return ret < 0 ? ret : 0;
834

835
	group_counter->prev_state = group_counter->state;
836 837
	if (counter_sched_in(group_counter, cpuctx, ctx, cpu))
		return -EAGAIN;
838 839 840 841

	/*
	 * Schedule in siblings as one group (if any):
	 */
I
Ingo Molnar 已提交
842
	list_for_each_entry(counter, &group_counter->sibling_list, list_entry) {
843
		counter->prev_state = counter->state;
844 845 846 847 848 849
		if (counter_sched_in(counter, cpuctx, ctx, cpu)) {
			partial_group = counter;
			goto group_error;
		}
	}

850
	return 0;
851 852 853 854 855 856 857 858 859 860

group_error:
	/*
	 * Groups can be scheduled in as one unit only, so undo any
	 * partial group before returning:
	 */
	list_for_each_entry(counter, &group_counter->sibling_list, list_entry) {
		if (counter == partial_group)
			break;
		counter_sched_out(counter, cpuctx, ctx);
I
Ingo Molnar 已提交
861
	}
862
	counter_sched_out(group_counter, cpuctx, ctx);
I
Ingo Molnar 已提交
863

864
	return -EAGAIN;
865 866
}

867 868 869
static void
__perf_counter_sched_in(struct perf_counter_context *ctx,
			struct perf_cpu_context *cpuctx, int cpu)
T
Thomas Gleixner 已提交
870 871
{
	struct perf_counter *counter;
872
	u64 flags;
873
	int can_add_hw = 1;
T
Thomas Gleixner 已提交
874

875 876
	spin_lock(&ctx->lock);
	ctx->is_active = 1;
T
Thomas Gleixner 已提交
877
	if (likely(!ctx->nr_counters))
878
		goto out;
T
Thomas Gleixner 已提交
879

880 881 882 883 884 885 886
	/*
	 * Add any time since the last sched_out to the lost time
	 * so it doesn't get included in the total_time_enabled and
	 * total_time_running measures for counters in the context.
	 */
	ctx->time_lost = get_context_time(ctx, 0) - ctx->time_now;

887
	flags = hw_perf_save_disable();
888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906

	/*
	 * First go through the list and put on any pinned groups
	 * in order to give them the best chance of going on.
	 */
	list_for_each_entry(counter, &ctx->counter_list, list_entry) {
		if (counter->state <= PERF_COUNTER_STATE_OFF ||
		    !counter->hw_event.pinned)
			continue;
		if (counter->cpu != -1 && counter->cpu != cpu)
			continue;

		if (group_can_go_on(counter, cpuctx, 1))
			group_sched_in(counter, cpuctx, ctx, cpu);

		/*
		 * If this pinned group hasn't been scheduled,
		 * put it in error state.
		 */
907 908
		if (counter->state == PERF_COUNTER_STATE_INACTIVE) {
			update_group_times(counter);
909
			counter->state = PERF_COUNTER_STATE_ERROR;
910
		}
911 912
	}

913
	list_for_each_entry(counter, &ctx->counter_list, list_entry) {
914 915 916 917 918 919 920 921
		/*
		 * Ignore counters in OFF or ERROR state, and
		 * ignore pinned counters since we did them already.
		 */
		if (counter->state <= PERF_COUNTER_STATE_OFF ||
		    counter->hw_event.pinned)
			continue;

922 923 924 925
		/*
		 * Listen to the 'cpu' scheduling filter constraint
		 * of counters:
		 */
T
Thomas Gleixner 已提交
926 927 928
		if (counter->cpu != -1 && counter->cpu != cpu)
			continue;

929
		if (group_can_go_on(counter, cpuctx, can_add_hw)) {
930 931
			if (group_sched_in(counter, cpuctx, ctx, cpu))
				can_add_hw = 0;
932
		}
T
Thomas Gleixner 已提交
933
	}
934
	hw_perf_restore(flags);
935
 out:
T
Thomas Gleixner 已提交
936
	spin_unlock(&ctx->lock);
937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953
}

/*
 * Called from scheduler to add the counters of the current task
 * with interrupts disabled.
 *
 * We restore the counter value and then enable it.
 *
 * This does not protect us against NMI, but enable()
 * sets the enabled bit in the control field of counter _before_
 * accessing the counter control register. If a NMI hits, then it will
 * keep the counter running.
 */
void perf_counter_task_sched_in(struct task_struct *task, int cpu)
{
	struct perf_cpu_context *cpuctx = &per_cpu(perf_cpu_context, cpu);
	struct perf_counter_context *ctx = &task->perf_counter_ctx;
954

955
	__perf_counter_sched_in(ctx, cpuctx, cpu);
T
Thomas Gleixner 已提交
956 957 958
	cpuctx->task_ctx = ctx;
}

959 960 961 962 963 964 965
static void perf_counter_cpu_sched_in(struct perf_cpu_context *cpuctx, int cpu)
{
	struct perf_counter_context *ctx = &cpuctx->ctx;

	__perf_counter_sched_in(ctx, cpuctx, cpu);
}

966 967 968 969 970
int perf_counter_task_disable(void)
{
	struct task_struct *curr = current;
	struct perf_counter_context *ctx = &curr->perf_counter_ctx;
	struct perf_counter *counter;
I
Ingo Molnar 已提交
971
	unsigned long flags;
972 973 974 975 976 977
	u64 perf_flags;
	int cpu;

	if (likely(!ctx->nr_counters))
		return 0;

I
Ingo Molnar 已提交
978
	curr_rq_lock_irq_save(&flags);
979 980
	cpu = smp_processor_id();

I
Ingo Molnar 已提交
981 982 983
	/* force the update of the task clock: */
	__task_delta_exec(curr, 1);

984 985 986 987 988 989 990 991 992
	perf_counter_task_sched_out(curr, cpu);

	spin_lock(&ctx->lock);

	/*
	 * Disable all the counters:
	 */
	perf_flags = hw_perf_save_disable();

993
	list_for_each_entry(counter, &ctx->counter_list, list_entry) {
994 995
		if (counter->state != PERF_COUNTER_STATE_ERROR) {
			update_group_times(counter);
996
			counter->state = PERF_COUNTER_STATE_OFF;
997
		}
998
	}
999

1000 1001 1002 1003
	hw_perf_restore(perf_flags);

	spin_unlock(&ctx->lock);

I
Ingo Molnar 已提交
1004
	curr_rq_unlock_irq_restore(&flags);
1005 1006 1007 1008 1009 1010 1011 1012 1013

	return 0;
}

int perf_counter_task_enable(void)
{
	struct task_struct *curr = current;
	struct perf_counter_context *ctx = &curr->perf_counter_ctx;
	struct perf_counter *counter;
I
Ingo Molnar 已提交
1014
	unsigned long flags;
1015 1016 1017 1018 1019 1020
	u64 perf_flags;
	int cpu;

	if (likely(!ctx->nr_counters))
		return 0;

I
Ingo Molnar 已提交
1021
	curr_rq_lock_irq_save(&flags);
1022 1023
	cpu = smp_processor_id();

I
Ingo Molnar 已提交
1024 1025 1026
	/* force the update of the task clock: */
	__task_delta_exec(curr, 1);

1027 1028
	perf_counter_task_sched_out(curr, cpu);

1029 1030 1031 1032 1033 1034 1035 1036
	spin_lock(&ctx->lock);

	/*
	 * Disable all the counters:
	 */
	perf_flags = hw_perf_save_disable();

	list_for_each_entry(counter, &ctx->counter_list, list_entry) {
1037
		if (counter->state > PERF_COUNTER_STATE_OFF)
1038
			continue;
1039
		counter->state = PERF_COUNTER_STATE_INACTIVE;
1040 1041
		counter->tstamp_enabled = ctx->time_now -
			counter->total_time_enabled;
I
Ingo Molnar 已提交
1042
		counter->hw_event.disabled = 0;
1043 1044 1045 1046 1047 1048 1049
	}
	hw_perf_restore(perf_flags);

	spin_unlock(&ctx->lock);

	perf_counter_task_sched_in(curr, cpu);

I
Ingo Molnar 已提交
1050
	curr_rq_unlock_irq_restore(&flags);
1051 1052 1053 1054

	return 0;
}

1055 1056 1057 1058
/*
 * Round-robin a context's counters:
 */
static void rotate_ctx(struct perf_counter_context *ctx)
T
Thomas Gleixner 已提交
1059 1060
{
	struct perf_counter *counter;
1061
	u64 perf_flags;
T
Thomas Gleixner 已提交
1062

1063
	if (!ctx->nr_counters)
T
Thomas Gleixner 已提交
1064 1065 1066 1067
		return;

	spin_lock(&ctx->lock);
	/*
1068
	 * Rotate the first entry last (works just fine for group counters too):
T
Thomas Gleixner 已提交
1069
	 */
1070
	perf_flags = hw_perf_save_disable();
1071
	list_for_each_entry(counter, &ctx->counter_list, list_entry) {
1072
		list_move_tail(&counter->list_entry, &ctx->counter_list);
T
Thomas Gleixner 已提交
1073 1074
		break;
	}
1075
	hw_perf_restore(perf_flags);
T
Thomas Gleixner 已提交
1076 1077

	spin_unlock(&ctx->lock);
1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088
}

void perf_counter_task_tick(struct task_struct *curr, int cpu)
{
	struct perf_cpu_context *cpuctx = &per_cpu(perf_cpu_context, cpu);
	struct perf_counter_context *ctx = &curr->perf_counter_ctx;
	const int rotate_percpu = 0;

	if (rotate_percpu)
		perf_counter_cpu_sched_out(cpuctx);
	perf_counter_task_sched_out(curr, cpu);
T
Thomas Gleixner 已提交
1089

1090 1091 1092 1093 1094 1095
	if (rotate_percpu)
		rotate_ctx(&cpuctx->ctx);
	rotate_ctx(ctx);

	if (rotate_percpu)
		perf_counter_cpu_sched_in(cpuctx, cpu);
T
Thomas Gleixner 已提交
1096 1097 1098 1099 1100 1101
	perf_counter_task_sched_in(curr, cpu);
}

/*
 * Cross CPU call to read the hardware counter
 */
I
Ingo Molnar 已提交
1102
static void __read(void *info)
T
Thomas Gleixner 已提交
1103
{
I
Ingo Molnar 已提交
1104
	struct perf_counter *counter = info;
1105
	struct perf_counter_context *ctx = counter->ctx;
I
Ingo Molnar 已提交
1106
	unsigned long flags;
I
Ingo Molnar 已提交
1107

I
Ingo Molnar 已提交
1108
	curr_rq_lock_irq_save(&flags);
1109 1110
	if (ctx->is_active)
		update_context_time(ctx, 1);
I
Ingo Molnar 已提交
1111
	counter->hw_ops->read(counter);
1112
	update_counter_times(counter);
I
Ingo Molnar 已提交
1113
	curr_rq_unlock_irq_restore(&flags);
T
Thomas Gleixner 已提交
1114 1115
}

1116
static u64 perf_counter_read(struct perf_counter *counter)
T
Thomas Gleixner 已提交
1117 1118 1119 1120 1121
{
	/*
	 * If counter is enabled and currently active on a CPU, update the
	 * value in the counter structure:
	 */
1122
	if (counter->state == PERF_COUNTER_STATE_ACTIVE) {
T
Thomas Gleixner 已提交
1123
		smp_call_function_single(counter->oncpu,
I
Ingo Molnar 已提交
1124
					 __read, counter, 1);
1125 1126
	} else if (counter->state == PERF_COUNTER_STATE_INACTIVE) {
		update_counter_times(counter);
T
Thomas Gleixner 已提交
1127 1128
	}

1129
	return atomic64_read(&counter->count);
T
Thomas Gleixner 已提交
1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192
}

static void put_context(struct perf_counter_context *ctx)
{
	if (ctx->task)
		put_task_struct(ctx->task);
}

static struct perf_counter_context *find_get_context(pid_t pid, int cpu)
{
	struct perf_cpu_context *cpuctx;
	struct perf_counter_context *ctx;
	struct task_struct *task;

	/*
	 * If cpu is not a wildcard then this is a percpu counter:
	 */
	if (cpu != -1) {
		/* Must be root to operate on a CPU counter: */
		if (!capable(CAP_SYS_ADMIN))
			return ERR_PTR(-EACCES);

		if (cpu < 0 || cpu > num_possible_cpus())
			return ERR_PTR(-EINVAL);

		/*
		 * We could be clever and allow to attach a counter to an
		 * offline CPU and activate it when the CPU comes up, but
		 * that's for later.
		 */
		if (!cpu_isset(cpu, cpu_online_map))
			return ERR_PTR(-ENODEV);

		cpuctx = &per_cpu(perf_cpu_context, cpu);
		ctx = &cpuctx->ctx;

		return ctx;
	}

	rcu_read_lock();
	if (!pid)
		task = current;
	else
		task = find_task_by_vpid(pid);
	if (task)
		get_task_struct(task);
	rcu_read_unlock();

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

	ctx = &task->perf_counter_ctx;
	ctx->task = task;

	/* Reuse ptrace permission checks for now. */
	if (!ptrace_may_access(task, PTRACE_MODE_READ)) {
		put_context(ctx);
		return ERR_PTR(-EACCES);
	}

	return ctx;
}

P
Peter Zijlstra 已提交
1193 1194 1195 1196 1197 1198 1199 1200
static void free_counter_rcu(struct rcu_head *head)
{
	struct perf_counter *counter;

	counter = container_of(head, struct perf_counter, rcu_head);
	kfree(counter);
}

1201 1202
static void perf_pending_sync(struct perf_counter *counter);

1203 1204
static void free_counter(struct perf_counter *counter)
{
1205 1206
	perf_pending_sync(counter);

1207 1208 1209
	if (counter->destroy)
		counter->destroy(counter);

1210 1211 1212
	call_rcu(&counter->rcu_head, free_counter_rcu);
}

T
Thomas Gleixner 已提交
1213 1214 1215 1216 1217 1218 1219 1220 1221 1222
/*
 * Called when the last reference to the file is gone.
 */
static int perf_release(struct inode *inode, struct file *file)
{
	struct perf_counter *counter = file->private_data;
	struct perf_counter_context *ctx = counter->ctx;

	file->private_data = NULL;

1223
	mutex_lock(&ctx->mutex);
T
Thomas Gleixner 已提交
1224 1225
	mutex_lock(&counter->mutex);

1226
	perf_counter_remove_from_context(counter);
T
Thomas Gleixner 已提交
1227 1228

	mutex_unlock(&counter->mutex);
1229
	mutex_unlock(&ctx->mutex);
T
Thomas Gleixner 已提交
1230

1231
	free_counter(counter);
1232
	put_context(ctx);
T
Thomas Gleixner 已提交
1233 1234 1235 1236 1237 1238 1239 1240 1241 1242

	return 0;
}

/*
 * Read the performance counter - simple non blocking version for now
 */
static ssize_t
perf_read_hw(struct perf_counter *counter, char __user *buf, size_t count)
{
1243 1244
	u64 values[3];
	int n;
T
Thomas Gleixner 已提交
1245

1246 1247 1248 1249 1250 1251 1252 1253
	/*
	 * Return end-of-file for a read on a counter that is in
	 * error state (i.e. because it was pinned but it couldn't be
	 * scheduled on to the CPU at some point).
	 */
	if (counter->state == PERF_COUNTER_STATE_ERROR)
		return 0;

T
Thomas Gleixner 已提交
1254
	mutex_lock(&counter->mutex);
1255 1256 1257 1258 1259 1260 1261 1262
	values[0] = perf_counter_read(counter);
	n = 1;
	if (counter->hw_event.read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
		values[n++] = counter->total_time_enabled +
			atomic64_read(&counter->child_total_time_enabled);
	if (counter->hw_event.read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
		values[n++] = counter->total_time_running +
			atomic64_read(&counter->child_total_time_running);
T
Thomas Gleixner 已提交
1263 1264
	mutex_unlock(&counter->mutex);

1265 1266 1267 1268 1269 1270 1271 1272
	if (count < n * sizeof(u64))
		return -EINVAL;
	count = n * sizeof(u64);

	if (copy_to_user(buf, values, count))
		return -EFAULT;

	return count;
T
Thomas Gleixner 已提交
1273 1274 1275 1276 1277 1278 1279
}

static ssize_t
perf_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
{
	struct perf_counter *counter = file->private_data;

1280
	return perf_read_hw(counter, buf, count);
T
Thomas Gleixner 已提交
1281 1282 1283 1284 1285
}

static unsigned int perf_poll(struct file *file, poll_table *wait)
{
	struct perf_counter *counter = file->private_data;
P
Peter Zijlstra 已提交
1286 1287 1288 1289 1290 1291 1292 1293 1294 1295
	struct perf_mmap_data *data;
	unsigned int events;

	rcu_read_lock();
	data = rcu_dereference(counter->data);
	if (data)
		events = atomic_xchg(&data->wakeup, 0);
	else
		events = POLL_HUP;
	rcu_read_unlock();
T
Thomas Gleixner 已提交
1296 1297 1298 1299 1300 1301

	poll_wait(file, &counter->waitq, wait);

	return events;
}

1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319
static long perf_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
{
	struct perf_counter *counter = file->private_data;
	int err = 0;

	switch (cmd) {
	case PERF_COUNTER_IOC_ENABLE:
		perf_counter_enable_family(counter);
		break;
	case PERF_COUNTER_IOC_DISABLE:
		perf_counter_disable_family(counter);
		break;
	default:
		err = -ENOTTY;
	}
	return err;
}

1320 1321 1322 1323 1324 1325
/*
 * 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.
 */
void perf_counter_update_userpage(struct perf_counter *counter)
1326
{
1327 1328 1329 1330 1331 1332 1333 1334 1335
	struct perf_mmap_data *data;
	struct perf_counter_mmap_page *userpg;

	rcu_read_lock();
	data = rcu_dereference(counter->data);
	if (!data)
		goto unlock;

	userpg = data->user_page;
1336

1337 1338 1339 1340 1341
	/*
	 * Disable preemption so as to not let the corresponding user-space
	 * spin too long if we get preempted.
	 */
	preempt_disable();
1342 1343 1344 1345 1346 1347
	++userpg->lock;
	smp_wmb();
	userpg->index = counter->hw.idx;
	userpg->offset = atomic64_read(&counter->count);
	if (counter->state == PERF_COUNTER_STATE_ACTIVE)
		userpg->offset -= atomic64_read(&counter->hw.prev_count);
1348

1349 1350
	smp_wmb();
	++userpg->lock;
1351
	preempt_enable();
1352
unlock:
1353
	rcu_read_unlock();
1354 1355 1356 1357 1358
}

static int perf_mmap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
{
	struct perf_counter *counter = vma->vm_file->private_data;
1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370
	struct perf_mmap_data *data;
	int ret = VM_FAULT_SIGBUS;

	rcu_read_lock();
	data = rcu_dereference(counter->data);
	if (!data)
		goto unlock;

	if (vmf->pgoff == 0) {
		vmf->page = virt_to_page(data->user_page);
	} else {
		int nr = vmf->pgoff - 1;
1371

1372 1373
		if ((unsigned)nr > data->nr_pages)
			goto unlock;
1374

1375 1376
		vmf->page = virt_to_page(data->data_pages[nr]);
	}
1377
	get_page(vmf->page);
1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413
	ret = 0;
unlock:
	rcu_read_unlock();

	return ret;
}

static int perf_mmap_data_alloc(struct perf_counter *counter, int nr_pages)
{
	struct perf_mmap_data *data;
	unsigned long size;
	int i;

	WARN_ON(atomic_read(&counter->mmap_count));

	size = sizeof(struct perf_mmap_data);
	size += nr_pages * sizeof(void *);

	data = kzalloc(size, GFP_KERNEL);
	if (!data)
		goto fail;

	data->user_page = (void *)get_zeroed_page(GFP_KERNEL);
	if (!data->user_page)
		goto fail_user_page;

	for (i = 0; i < nr_pages; i++) {
		data->data_pages[i] = (void *)get_zeroed_page(GFP_KERNEL);
		if (!data->data_pages[i])
			goto fail_data_pages;
	}

	data->nr_pages = nr_pages;

	rcu_assign_pointer(counter->data, data);

1414
	return 0;
1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466

fail_data_pages:
	for (i--; i >= 0; i--)
		free_page((unsigned long)data->data_pages[i]);

	free_page((unsigned long)data->user_page);

fail_user_page:
	kfree(data);

fail:
	return -ENOMEM;
}

static void __perf_mmap_data_free(struct rcu_head *rcu_head)
{
	struct perf_mmap_data *data = container_of(rcu_head,
			struct perf_mmap_data, rcu_head);
	int i;

	free_page((unsigned long)data->user_page);
	for (i = 0; i < data->nr_pages; i++)
		free_page((unsigned long)data->data_pages[i]);
	kfree(data);
}

static void perf_mmap_data_free(struct perf_counter *counter)
{
	struct perf_mmap_data *data = counter->data;

	WARN_ON(atomic_read(&counter->mmap_count));

	rcu_assign_pointer(counter->data, NULL);
	call_rcu(&data->rcu_head, __perf_mmap_data_free);
}

static void perf_mmap_open(struct vm_area_struct *vma)
{
	struct perf_counter *counter = vma->vm_file->private_data;

	atomic_inc(&counter->mmap_count);
}

static void perf_mmap_close(struct vm_area_struct *vma)
{
	struct perf_counter *counter = vma->vm_file->private_data;

	if (atomic_dec_and_mutex_lock(&counter->mmap_count,
				      &counter->mmap_mutex)) {
		perf_mmap_data_free(counter);
		mutex_unlock(&counter->mmap_mutex);
	}
1467 1468 1469
}

static struct vm_operations_struct perf_mmap_vmops = {
1470 1471
	.open = perf_mmap_open,
	.close = perf_mmap_close,
1472 1473 1474 1475 1476 1477
	.fault = perf_mmap_fault,
};

static int perf_mmap(struct file *file, struct vm_area_struct *vma)
{
	struct perf_counter *counter = file->private_data;
1478 1479 1480 1481
	unsigned long vma_size;
	unsigned long nr_pages;
	unsigned long locked, lock_limit;
	int ret = 0;
1482 1483 1484

	if (!(vma->vm_flags & VM_SHARED) || (vma->vm_flags & VM_WRITE))
		return -EINVAL;
1485 1486 1487 1488

	vma_size = vma->vm_end - vma->vm_start;
	nr_pages = (vma_size / PAGE_SIZE) - 1;

1489 1490 1491 1492 1493
	/*
	 * If we have data pages ensure they're a power-of-two number, so we
	 * can do bitmasks instead of modulo.
	 */
	if (nr_pages != 0 && !is_power_of_2(nr_pages))
1494 1495
		return -EINVAL;

1496
	if (vma_size != PAGE_SIZE * (1 + nr_pages))
1497 1498
		return -EINVAL;

1499 1500
	if (vma->vm_pgoff != 0)
		return -EINVAL;
1501

1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520
	locked = vma_size >>  PAGE_SHIFT;
	locked += vma->vm_mm->locked_vm;

	lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur;
	lock_limit >>= PAGE_SHIFT;

	if ((locked > lock_limit) && !capable(CAP_IPC_LOCK))
		return -EPERM;

	mutex_lock(&counter->mmap_mutex);
	if (atomic_inc_not_zero(&counter->mmap_count))
		goto out;

	WARN_ON(counter->data);
	ret = perf_mmap_data_alloc(counter, nr_pages);
	if (!ret)
		atomic_set(&counter->mmap_count, 1);
out:
	mutex_unlock(&counter->mmap_mutex);
1521 1522 1523 1524

	vma->vm_flags &= ~VM_MAYWRITE;
	vma->vm_flags |= VM_RESERVED;
	vma->vm_ops = &perf_mmap_vmops;
1525 1526

	return ret;
1527 1528
}

T
Thomas Gleixner 已提交
1529 1530 1531 1532
static const struct file_operations perf_fops = {
	.release		= perf_release,
	.read			= perf_read,
	.poll			= perf_poll,
1533 1534
	.unlocked_ioctl		= perf_ioctl,
	.compat_ioctl		= perf_ioctl,
1535
	.mmap			= perf_mmap,
T
Thomas Gleixner 已提交
1536 1537
};

1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552
/*
 * Perf counter wakeup
 *
 * If there's data, ensure we set the poll() state and publish everything
 * to user-space before waking everybody up.
 */

void perf_counter_wakeup(struct perf_counter *counter)
{
	struct perf_mmap_data *data;

	rcu_read_lock();
	data = rcu_dereference(counter->data);
	if (data) {
		(void)atomic_xchg(&data->wakeup, POLL_IN);
1553 1554 1555 1556 1557 1558 1559
		/*
		 * Ensure all data writes are issued before updating the
		 * user-space data head information. The matching rmb()
		 * will be in userspace after reading this value.
		 */
		smp_wmb();
		data->user_page->data_head = atomic_read(&data->head);
1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613 1614 1615 1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655
	}
	rcu_read_unlock();

	wake_up_all(&counter->waitq);
}

/*
 * Pending wakeups
 *
 * Handle the case where we need to wakeup up from NMI (or rq->lock) context.
 *
 * The NMI bit means we cannot possibly take locks. Therefore, maintain a
 * single linked list and use cmpxchg() to add entries lockless.
 */

#define PENDING_TAIL ((struct perf_wakeup_entry *)-1UL)

static DEFINE_PER_CPU(struct perf_wakeup_entry *, perf_wakeup_head) = {
	PENDING_TAIL,
};

static void perf_pending_queue(struct perf_counter *counter)
{
	struct perf_wakeup_entry **head;
	struct perf_wakeup_entry *prev, *next;

	if (cmpxchg(&counter->wakeup.next, NULL, PENDING_TAIL) != NULL)
		return;

	head = &get_cpu_var(perf_wakeup_head);

	do {
		prev = counter->wakeup.next = *head;
		next = &counter->wakeup;
	} while (cmpxchg(head, prev, next) != prev);

	set_perf_counter_pending();

	put_cpu_var(perf_wakeup_head);
}

static int __perf_pending_run(void)
{
	struct perf_wakeup_entry *list;
	int nr = 0;

	list = xchg(&__get_cpu_var(perf_wakeup_head), PENDING_TAIL);
	while (list != PENDING_TAIL) {
		struct perf_counter *counter = container_of(list,
				struct perf_counter, wakeup);

		list = list->next;

		counter->wakeup.next = NULL;
		/*
		 * Ensure we observe the unqueue before we issue the wakeup,
		 * so that we won't be waiting forever.
		 * -- see perf_not_pending().
		 */
		smp_wmb();

		perf_counter_wakeup(counter);
		nr++;
	}

	return nr;
}

static inline int perf_not_pending(struct perf_counter *counter)
{
	/*
	 * If we flush on whatever cpu we run, there is a chance we don't
	 * need to wait.
	 */
	get_cpu();
	__perf_pending_run();
	put_cpu();

	/*
	 * Ensure we see the proper queue state before going to sleep
	 * so that we do not miss the wakeup. -- see perf_pending_handle()
	 */
	smp_rmb();
	return counter->wakeup.next == NULL;
}

static void perf_pending_sync(struct perf_counter *counter)
{
	wait_event(counter->waitq, perf_not_pending(counter));
}

void perf_counter_do_pending(void)
{
	__perf_pending_run();
}

1656 1657 1658 1659
/*
 * Output
 */

1660 1661 1662 1663
struct perf_output_handle {
	struct perf_counter	*counter;
	struct perf_mmap_data	*data;
	unsigned int		offset;
1664
	unsigned int		head;
1665 1666 1667 1668 1669
	int			wakeup;
};

static int perf_output_begin(struct perf_output_handle *handle,
			     struct perf_counter *counter, unsigned int size)
1670
{
1671
	struct perf_mmap_data *data;
1672
	unsigned int offset, head;
1673

1674 1675 1676 1677 1678 1679 1680 1681 1682 1683
	rcu_read_lock();
	data = rcu_dereference(counter->data);
	if (!data)
		goto out;

	if (!data->nr_pages)
		goto out;

	do {
		offset = head = atomic_read(&data->head);
P
Peter Zijlstra 已提交
1684
		head += size;
1685 1686
	} while (atomic_cmpxchg(&data->head, offset, head) != offset);

1687 1688 1689
	handle->counter	= counter;
	handle->data	= data;
	handle->offset	= offset;
1690
	handle->head	= head;
1691
	handle->wakeup	= (offset >> PAGE_SHIFT) != (head >> PAGE_SHIFT);
1692

1693
	return 0;
1694

1695 1696
out:
	rcu_read_unlock();
1697

1698 1699
	return -ENOSPC;
}
1700

1701 1702 1703 1704 1705 1706 1707 1708 1709 1710 1711 1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726 1727 1728
static void perf_output_copy(struct perf_output_handle *handle,
			     void *buf, unsigned int len)
{
	unsigned int pages_mask;
	unsigned int offset;
	unsigned int size;
	void **pages;

	offset		= handle->offset;
	pages_mask	= handle->data->nr_pages - 1;
	pages		= handle->data->data_pages;

	do {
		unsigned int page_offset;
		int nr;

		nr	    = (offset >> PAGE_SHIFT) & pages_mask;
		page_offset = offset & (PAGE_SIZE - 1);
		size	    = min_t(unsigned int, PAGE_SIZE - page_offset, len);

		memcpy(pages[nr] + page_offset, buf, size);

		len	    -= size;
		buf	    += size;
		offset	    += size;
	} while (len);

	handle->offset = offset;
1729 1730

	WARN_ON_ONCE(handle->offset > handle->head);
1731 1732
}

P
Peter Zijlstra 已提交
1733 1734 1735
#define perf_output_put(handle, x) \
	perf_output_copy((handle), &(x), sizeof(x))

1736 1737 1738
static void perf_output_end(struct perf_output_handle *handle, int nmi)
{
	if (handle->wakeup) {
1739 1740 1741 1742
		if (nmi)
			perf_pending_queue(handle->counter);
		else
			perf_counter_wakeup(handle->counter);
1743
	}
1744
	rcu_read_unlock();
1745 1746 1747 1748 1749 1750 1751
}

static int perf_output_write(struct perf_counter *counter, int nmi,
			     void *buf, ssize_t size)
{
	struct perf_output_handle handle;
	int ret;
1752

1753 1754 1755 1756 1757 1758 1759 1760
	ret = perf_output_begin(&handle, counter, size);
	if (ret)
		goto out;

	perf_output_copy(&handle, buf, size);
	perf_output_end(&handle, nmi);

out:
1761 1762 1763 1764 1765 1766
	return ret;
}

static void perf_output_simple(struct perf_counter *counter,
			       int nmi, struct pt_regs *regs)
{
1767
	unsigned int size;
P
Peter Zijlstra 已提交
1768 1769 1770
	struct {
		struct perf_event_header header;
		u64 ip;
1771
		u32 pid, tid;
P
Peter Zijlstra 已提交
1772
	} event;
1773

P
Peter Zijlstra 已提交
1774 1775
	event.header.type = PERF_EVENT_IP;
	event.ip = instruction_pointer(regs);
1776

1777 1778 1779 1780 1781 1782 1783 1784 1785 1786 1787 1788 1789 1790
	size = sizeof(event);

	if (counter->hw_event.include_tid) {
		/* namespace issues */
		event.pid = current->group_leader->pid;
		event.tid = current->pid;

		event.header.type |= __PERF_EVENT_TID;
	} else
		size -= sizeof(u64);

	event.header.size = size;

	perf_output_write(counter, nmi, &event, size);
1791 1792
}

1793
static void perf_output_group(struct perf_counter *counter, int nmi)
1794
{
P
Peter Zijlstra 已提交
1795 1796
	struct perf_output_handle handle;
	struct perf_event_header header;
1797
	struct perf_counter *leader, *sub;
P
Peter Zijlstra 已提交
1798 1799 1800 1801 1802 1803 1804 1805 1806 1807 1808 1809 1810 1811 1812 1813 1814
	unsigned int size;
	struct {
		u64 event;
		u64 counter;
	} entry;
	int ret;

	size = sizeof(header) + counter->nr_siblings * sizeof(entry);

	ret = perf_output_begin(&handle, counter, size);
	if (ret)
		return;

	header.type = PERF_EVENT_GROUP;
	header.size = size;

	perf_output_put(&handle, header);
1815 1816 1817 1818 1819

	leader = counter->group_leader;
	list_for_each_entry(sub, &leader->sibling_list, list_entry) {
		if (sub != counter)
			sub->hw_ops->read(sub);
1820 1821 1822 1823

		entry.event = sub->hw_event.config;
		entry.counter = atomic64_read(&sub->count);

P
Peter Zijlstra 已提交
1824
		perf_output_put(&handle, entry);
1825
	}
P
Peter Zijlstra 已提交
1826 1827

	perf_output_end(&handle, nmi);
1828 1829 1830 1831 1832 1833 1834 1835 1836 1837
}

void perf_counter_output(struct perf_counter *counter,
			 int nmi, struct pt_regs *regs)
{
	switch (counter->hw_event.record_type) {
	case PERF_RECORD_SIMPLE:
		return;

	case PERF_RECORD_IRQ:
1838
		perf_output_simple(counter, nmi, regs);
1839 1840 1841
		break;

	case PERF_RECORD_GROUP:
1842
		perf_output_group(counter, nmi);
1843 1844 1845 1846
		break;
	}
}

1847 1848 1849 1850 1851 1852 1853 1854 1855 1856 1857 1858 1859 1860 1861 1862 1863 1864 1865 1866 1867 1868 1869 1870 1871 1872 1873 1874 1875 1876 1877 1878 1879 1880 1881 1882 1883 1884 1885 1886 1887 1888 1889 1890 1891 1892 1893 1894 1895 1896 1897 1898 1899 1900 1901 1902 1903 1904 1905 1906 1907 1908 1909 1910 1911 1912 1913 1914 1915 1916 1917 1918 1919 1920 1921 1922 1923 1924 1925 1926 1927 1928 1929 1930 1931 1932 1933 1934 1935 1936 1937 1938 1939 1940 1941 1942 1943 1944 1945 1946 1947 1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990
/*
 * mmap tracking
 */

struct perf_mmap_event {
	struct file	*file;
	char		*file_name;
	int		file_size;

	struct {
		struct perf_event_header	header;

		u32				pid;
		u32				tid;
		u64				start;
		u64				len;
		u64				pgoff;
	} event;
};

static void perf_counter_mmap_output(struct perf_counter *counter,
				     struct perf_mmap_event *mmap_event)
{
	struct perf_output_handle handle;
	int size = mmap_event->event.header.size;
	int ret = perf_output_begin(&handle, counter, size);

	if (ret)
		return;

	perf_output_put(&handle, mmap_event->event);
	perf_output_copy(&handle, mmap_event->file_name,
				   mmap_event->file_size);
	perf_output_end(&handle, 0);
}

static int perf_counter_mmap_match(struct perf_counter *counter,
				   struct perf_mmap_event *mmap_event)
{
	if (counter->hw_event.mmap &&
	    mmap_event->event.header.type == PERF_EVENT_MMAP)
		return 1;

	if (counter->hw_event.munmap &&
	    mmap_event->event.header.type == PERF_EVENT_MUNMAP)
		return 1;

	return 0;
}

static void perf_counter_mmap_ctx(struct perf_counter_context *ctx,
				  struct perf_mmap_event *mmap_event)
{
	struct perf_counter *counter;

	if (system_state != SYSTEM_RUNNING || list_empty(&ctx->event_list))
		return;

	rcu_read_lock();
	list_for_each_entry_rcu(counter, &ctx->event_list, event_entry) {
		if (perf_counter_mmap_match(counter, mmap_event))
			perf_counter_mmap_output(counter, mmap_event);
	}
	rcu_read_unlock();
}

static void perf_counter_mmap_event(struct perf_mmap_event *mmap_event)
{
	struct perf_cpu_context *cpuctx;
	struct file *file = mmap_event->file;
	unsigned int size;
	char tmp[16];
	char *buf = NULL;
	char *name;

	if (file) {
		buf = kzalloc(PATH_MAX, GFP_KERNEL);
		if (!buf) {
			name = strncpy(tmp, "//enomem", sizeof(tmp));
			goto got_name;
		}
		name = dentry_path(file->f_dentry, buf, PATH_MAX);
		if (IS_ERR(name)) {
			name = strncpy(tmp, "//toolong", sizeof(tmp));
			goto got_name;
		}
	} else {
		name = strncpy(tmp, "//anon", sizeof(tmp));
		goto got_name;
	}

got_name:
	size = ALIGN(strlen(name), sizeof(u64));

	mmap_event->file_name = name;
	mmap_event->file_size = size;

	mmap_event->event.header.size = sizeof(mmap_event->event) + size;

	cpuctx = &get_cpu_var(perf_cpu_context);
	perf_counter_mmap_ctx(&cpuctx->ctx, mmap_event);
	put_cpu_var(perf_cpu_context);

	perf_counter_mmap_ctx(&current->perf_counter_ctx, mmap_event);

	kfree(buf);
}

void perf_counter_mmap(unsigned long addr, unsigned long len,
		       unsigned long pgoff, struct file *file)
{
	struct perf_mmap_event mmap_event = {
		.file   = file,
		.event  = {
			.header = { .type = PERF_EVENT_MMAP, },
			.pid	= current->group_leader->pid,
			.tid	= current->pid,
			.start  = addr,
			.len    = len,
			.pgoff  = pgoff,
		},
	};

	perf_counter_mmap_event(&mmap_event);
}

void perf_counter_munmap(unsigned long addr, unsigned long len,
			 unsigned long pgoff, struct file *file)
{
	struct perf_mmap_event mmap_event = {
		.file   = file,
		.event  = {
			.header = { .type = PERF_EVENT_MUNMAP, },
			.pid	= current->group_leader->pid,
			.tid	= current->pid,
			.start  = addr,
			.len    = len,
			.pgoff  = pgoff,
		},
	};

	perf_counter_mmap_event(&mmap_event);
}

1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024 2025 2026 2027 2028 2029 2030 2031 2032
/*
 * Generic software counter infrastructure
 */

static void perf_swcounter_update(struct perf_counter *counter)
{
	struct hw_perf_counter *hwc = &counter->hw;
	u64 prev, now;
	s64 delta;

again:
	prev = atomic64_read(&hwc->prev_count);
	now = atomic64_read(&hwc->count);
	if (atomic64_cmpxchg(&hwc->prev_count, prev, now) != prev)
		goto again;

	delta = now - prev;

	atomic64_add(delta, &counter->count);
	atomic64_sub(delta, &hwc->period_left);
}

static void perf_swcounter_set_period(struct perf_counter *counter)
{
	struct hw_perf_counter *hwc = &counter->hw;
	s64 left = atomic64_read(&hwc->period_left);
	s64 period = hwc->irq_period;

	if (unlikely(left <= -period)) {
		left = period;
		atomic64_set(&hwc->period_left, left);
	}

	if (unlikely(left <= 0)) {
		left += period;
		atomic64_add(period, &hwc->period_left);
	}

	atomic64_set(&hwc->prev_count, -left);
	atomic64_set(&hwc->count, -left);
}

2033 2034 2035 2036 2037 2038 2039 2040 2041 2042 2043 2044 2045 2046 2047 2048 2049 2050
static enum hrtimer_restart perf_swcounter_hrtimer(struct hrtimer *hrtimer)
{
	struct perf_counter *counter;
	struct pt_regs *regs;

	counter	= container_of(hrtimer, struct perf_counter, hw.hrtimer);
	counter->hw_ops->read(counter);

	regs = get_irq_regs();
	/*
	 * In case we exclude kernel IPs or are somehow not in interrupt
	 * context, provide the next best thing, the user IP.
	 */
	if ((counter->hw_event.exclude_kernel || !regs) &&
			!counter->hw_event.exclude_user)
		regs = task_pt_regs(current);

	if (regs)
2051
		perf_counter_output(counter, 0, regs);
2052 2053 2054 2055 2056 2057 2058 2059 2060

	hrtimer_forward_now(hrtimer, ns_to_ktime(counter->hw.irq_period));

	return HRTIMER_RESTART;
}

static void perf_swcounter_overflow(struct perf_counter *counter,
				    int nmi, struct pt_regs *regs)
{
2061 2062
	perf_swcounter_update(counter);
	perf_swcounter_set_period(counter);
2063
	perf_counter_output(counter, nmi, regs);
2064 2065
}

2066
static int perf_swcounter_match(struct perf_counter *counter,
2067 2068
				enum perf_event_types type,
				u32 event, struct pt_regs *regs)
2069 2070 2071 2072
{
	if (counter->state != PERF_COUNTER_STATE_ACTIVE)
		return 0;

2073
	if (perf_event_raw(&counter->hw_event))
2074 2075
		return 0;

2076
	if (perf_event_type(&counter->hw_event) != type)
2077 2078
		return 0;

2079
	if (perf_event_id(&counter->hw_event) != event)
2080 2081 2082 2083 2084 2085 2086 2087 2088 2089 2090
		return 0;

	if (counter->hw_event.exclude_user && user_mode(regs))
		return 0;

	if (counter->hw_event.exclude_kernel && !user_mode(regs))
		return 0;

	return 1;
}

2091 2092 2093 2094 2095 2096 2097 2098
static void perf_swcounter_add(struct perf_counter *counter, u64 nr,
			       int nmi, struct pt_regs *regs)
{
	int neg = atomic64_add_negative(nr, &counter->hw.count);
	if (counter->hw.irq_period && !neg)
		perf_swcounter_overflow(counter, nmi, regs);
}

2099
static void perf_swcounter_ctx_event(struct perf_counter_context *ctx,
2100 2101
				     enum perf_event_types type, u32 event,
				     u64 nr, int nmi, struct pt_regs *regs)
2102 2103 2104
{
	struct perf_counter *counter;

2105
	if (system_state != SYSTEM_RUNNING || list_empty(&ctx->event_list))
2106 2107
		return;

P
Peter Zijlstra 已提交
2108 2109
	rcu_read_lock();
	list_for_each_entry_rcu(counter, &ctx->event_list, event_entry) {
2110
		if (perf_swcounter_match(counter, type, event, regs))
2111
			perf_swcounter_add(counter, nr, nmi, regs);
2112
	}
P
Peter Zijlstra 已提交
2113
	rcu_read_unlock();
2114 2115
}

P
Peter Zijlstra 已提交
2116 2117 2118 2119 2120 2121 2122 2123 2124 2125 2126 2127 2128 2129
static int *perf_swcounter_recursion_context(struct perf_cpu_context *cpuctx)
{
	if (in_nmi())
		return &cpuctx->recursion[3];

	if (in_irq())
		return &cpuctx->recursion[2];

	if (in_softirq())
		return &cpuctx->recursion[1];

	return &cpuctx->recursion[0];
}

2130 2131
static void __perf_swcounter_event(enum perf_event_types type, u32 event,
				   u64 nr, int nmi, struct pt_regs *regs)
2132 2133
{
	struct perf_cpu_context *cpuctx = &get_cpu_var(perf_cpu_context);
P
Peter Zijlstra 已提交
2134 2135 2136 2137 2138 2139 2140
	int *recursion = perf_swcounter_recursion_context(cpuctx);

	if (*recursion)
		goto out;

	(*recursion)++;
	barrier();
2141

2142 2143 2144 2145 2146
	perf_swcounter_ctx_event(&cpuctx->ctx, type, event, nr, nmi, regs);
	if (cpuctx->task_ctx) {
		perf_swcounter_ctx_event(cpuctx->task_ctx, type, event,
				nr, nmi, regs);
	}
2147

P
Peter Zijlstra 已提交
2148 2149 2150 2151
	barrier();
	(*recursion)--;

out:
2152 2153 2154
	put_cpu_var(perf_cpu_context);
}

2155 2156 2157 2158 2159
void perf_swcounter_event(u32 event, u64 nr, int nmi, struct pt_regs *regs)
{
	__perf_swcounter_event(PERF_TYPE_SOFTWARE, event, nr, nmi, regs);
}

2160 2161 2162 2163 2164 2165 2166 2167 2168 2169 2170 2171 2172 2173 2174 2175
static void perf_swcounter_read(struct perf_counter *counter)
{
	perf_swcounter_update(counter);
}

static int perf_swcounter_enable(struct perf_counter *counter)
{
	perf_swcounter_set_period(counter);
	return 0;
}

static void perf_swcounter_disable(struct perf_counter *counter)
{
	perf_swcounter_update(counter);
}

2176 2177 2178 2179 2180 2181
static const struct hw_perf_counter_ops perf_ops_generic = {
	.enable		= perf_swcounter_enable,
	.disable	= perf_swcounter_disable,
	.read		= perf_swcounter_read,
};

2182 2183 2184 2185
/*
 * Software counter: cpu wall time clock
 */

2186 2187 2188 2189 2190 2191 2192 2193 2194 2195 2196 2197
static void cpu_clock_perf_counter_update(struct perf_counter *counter)
{
	int cpu = raw_smp_processor_id();
	s64 prev;
	u64 now;

	now = cpu_clock(cpu);
	prev = atomic64_read(&counter->hw.prev_count);
	atomic64_set(&counter->hw.prev_count, now);
	atomic64_add(now - prev, &counter->count);
}

2198 2199 2200 2201 2202 2203
static int cpu_clock_perf_counter_enable(struct perf_counter *counter)
{
	struct hw_perf_counter *hwc = &counter->hw;
	int cpu = raw_smp_processor_id();

	atomic64_set(&hwc->prev_count, cpu_clock(cpu));
2204 2205
	hrtimer_init(&hwc->hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
	hwc->hrtimer.function = perf_swcounter_hrtimer;
2206 2207 2208 2209 2210 2211 2212 2213 2214
	if (hwc->irq_period) {
		__hrtimer_start_range_ns(&hwc->hrtimer,
				ns_to_ktime(hwc->irq_period), 0,
				HRTIMER_MODE_REL, 0);
	}

	return 0;
}

2215 2216
static void cpu_clock_perf_counter_disable(struct perf_counter *counter)
{
2217
	hrtimer_cancel(&counter->hw.hrtimer);
2218
	cpu_clock_perf_counter_update(counter);
2219 2220 2221 2222
}

static void cpu_clock_perf_counter_read(struct perf_counter *counter)
{
2223
	cpu_clock_perf_counter_update(counter);
2224 2225 2226
}

static const struct hw_perf_counter_ops perf_ops_cpu_clock = {
I
Ingo Molnar 已提交
2227 2228 2229
	.enable		= cpu_clock_perf_counter_enable,
	.disable	= cpu_clock_perf_counter_disable,
	.read		= cpu_clock_perf_counter_read,
2230 2231
};

2232 2233 2234 2235
/*
 * Software counter: task time clock
 */

I
Ingo Molnar 已提交
2236 2237 2238 2239
/*
 * Called from within the scheduler:
 */
static u64 task_clock_perf_counter_val(struct perf_counter *counter, int update)
2240
{
I
Ingo Molnar 已提交
2241 2242 2243 2244 2245 2246 2247 2248 2249 2250 2251
	struct task_struct *curr = counter->task;
	u64 delta;

	delta = __task_delta_exec(curr, update);

	return curr->se.sum_exec_runtime + delta;
}

static void task_clock_perf_counter_update(struct perf_counter *counter, u64 now)
{
	u64 prev;
I
Ingo Molnar 已提交
2252 2253 2254 2255 2256 2257 2258 2259 2260
	s64 delta;

	prev = atomic64_read(&counter->hw.prev_count);

	atomic64_set(&counter->hw.prev_count, now);

	delta = now - prev;

	atomic64_add(delta, &counter->count);
2261 2262
}

2263
static int task_clock_perf_counter_enable(struct perf_counter *counter)
I
Ingo Molnar 已提交
2264
{
2265 2266 2267
	struct hw_perf_counter *hwc = &counter->hw;

	atomic64_set(&hwc->prev_count, task_clock_perf_counter_val(counter, 0));
2268 2269
	hrtimer_init(&hwc->hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
	hwc->hrtimer.function = perf_swcounter_hrtimer;
2270 2271 2272 2273 2274
	if (hwc->irq_period) {
		__hrtimer_start_range_ns(&hwc->hrtimer,
				ns_to_ktime(hwc->irq_period), 0,
				HRTIMER_MODE_REL, 0);
	}
2275 2276

	return 0;
I
Ingo Molnar 已提交
2277 2278 2279
}

static void task_clock_perf_counter_disable(struct perf_counter *counter)
2280
{
2281 2282 2283 2284
	hrtimer_cancel(&counter->hw.hrtimer);
	task_clock_perf_counter_update(counter,
			task_clock_perf_counter_val(counter, 0));
}
I
Ingo Molnar 已提交
2285

2286 2287 2288 2289
static void task_clock_perf_counter_read(struct perf_counter *counter)
{
	task_clock_perf_counter_update(counter,
			task_clock_perf_counter_val(counter, 1));
2290 2291 2292
}

static const struct hw_perf_counter_ops perf_ops_task_clock = {
I
Ingo Molnar 已提交
2293 2294 2295
	.enable		= task_clock_perf_counter_enable,
	.disable	= task_clock_perf_counter_disable,
	.read		= task_clock_perf_counter_read,
2296 2297
};

2298 2299 2300 2301
/*
 * Software counter: cpu migrations
 */

2302
static inline u64 get_cpu_migrations(struct perf_counter *counter)
2303
{
2304 2305 2306 2307 2308
	struct task_struct *curr = counter->ctx->task;

	if (curr)
		return curr->se.nr_migrations;
	return cpu_nr_migrations(smp_processor_id());
2309 2310 2311 2312 2313 2314 2315 2316
}

static void cpu_migrations_perf_counter_update(struct perf_counter *counter)
{
	u64 prev, now;
	s64 delta;

	prev = atomic64_read(&counter->hw.prev_count);
2317
	now = get_cpu_migrations(counter);
2318 2319 2320 2321 2322 2323 2324 2325 2326 2327 2328 2329 2330

	atomic64_set(&counter->hw.prev_count, now);

	delta = now - prev;

	atomic64_add(delta, &counter->count);
}

static void cpu_migrations_perf_counter_read(struct perf_counter *counter)
{
	cpu_migrations_perf_counter_update(counter);
}

2331
static int cpu_migrations_perf_counter_enable(struct perf_counter *counter)
2332
{
2333 2334 2335
	if (counter->prev_state <= PERF_COUNTER_STATE_OFF)
		atomic64_set(&counter->hw.prev_count,
			     get_cpu_migrations(counter));
2336
	return 0;
2337 2338 2339 2340 2341 2342 2343 2344
}

static void cpu_migrations_perf_counter_disable(struct perf_counter *counter)
{
	cpu_migrations_perf_counter_update(counter);
}

static const struct hw_perf_counter_ops perf_ops_cpu_migrations = {
I
Ingo Molnar 已提交
2345 2346 2347
	.enable		= cpu_migrations_perf_counter_enable,
	.disable	= cpu_migrations_perf_counter_disable,
	.read		= cpu_migrations_perf_counter_read,
2348 2349
};

2350 2351 2352
#ifdef CONFIG_EVENT_PROFILE
void perf_tpcounter_event(int event_id)
{
2353 2354 2355 2356 2357 2358
	struct pt_regs *regs = get_irq_regs();

	if (!regs)
		regs = task_pt_regs(current);

	__perf_swcounter_event(PERF_TYPE_TRACEPOINT, event_id, 1, 1, regs);
2359 2360 2361 2362 2363 2364 2365
}

extern int ftrace_profile_enable(int);
extern void ftrace_profile_disable(int);

static void tp_perf_counter_destroy(struct perf_counter *counter)
{
2366
	ftrace_profile_disable(perf_event_id(&counter->hw_event));
2367 2368 2369 2370 2371
}

static const struct hw_perf_counter_ops *
tp_perf_counter_init(struct perf_counter *counter)
{
2372
	int event_id = perf_event_id(&counter->hw_event);
2373 2374 2375 2376 2377 2378 2379
	int ret;

	ret = ftrace_profile_enable(event_id);
	if (ret)
		return NULL;

	counter->destroy = tp_perf_counter_destroy;
2380
	counter->hw.irq_period = counter->hw_event.irq_period;
2381 2382 2383 2384 2385 2386 2387 2388 2389 2390 2391

	return &perf_ops_generic;
}
#else
static const struct hw_perf_counter_ops *
tp_perf_counter_init(struct perf_counter *counter)
{
	return NULL;
}
#endif

2392 2393 2394
static const struct hw_perf_counter_ops *
sw_perf_counter_init(struct perf_counter *counter)
{
2395
	struct perf_counter_hw_event *hw_event = &counter->hw_event;
2396
	const struct hw_perf_counter_ops *hw_ops = NULL;
2397
	struct hw_perf_counter *hwc = &counter->hw;
2398

2399 2400 2401 2402 2403 2404 2405
	/*
	 * Software counters (currently) can't in general distinguish
	 * between user, kernel and hypervisor events.
	 * However, context switches and cpu migrations are considered
	 * to be kernel events, and page faults are never hypervisor
	 * events.
	 */
2406
	switch (perf_event_id(&counter->hw_event)) {
2407
	case PERF_COUNT_CPU_CLOCK:
2408 2409 2410 2411
		hw_ops = &perf_ops_cpu_clock;

		if (hw_event->irq_period && hw_event->irq_period < 10000)
			hw_event->irq_period = 10000;
2412
		break;
2413
	case PERF_COUNT_TASK_CLOCK:
2414 2415 2416 2417 2418 2419 2420 2421
		/*
		 * If the user instantiates this as a per-cpu counter,
		 * use the cpu_clock counter instead.
		 */
		if (counter->ctx->task)
			hw_ops = &perf_ops_task_clock;
		else
			hw_ops = &perf_ops_cpu_clock;
2422 2423 2424

		if (hw_event->irq_period && hw_event->irq_period < 10000)
			hw_event->irq_period = 10000;
2425
		break;
2426
	case PERF_COUNT_PAGE_FAULTS:
2427 2428
	case PERF_COUNT_PAGE_FAULTS_MIN:
	case PERF_COUNT_PAGE_FAULTS_MAJ:
2429
	case PERF_COUNT_CONTEXT_SWITCHES:
2430
		hw_ops = &perf_ops_generic;
2431
		break;
2432
	case PERF_COUNT_CPU_MIGRATIONS:
2433 2434
		if (!counter->hw_event.exclude_kernel)
			hw_ops = &perf_ops_cpu_migrations;
2435
		break;
2436
	}
2437 2438 2439 2440

	if (hw_ops)
		hwc->irq_period = hw_event->irq_period;

2441 2442 2443
	return hw_ops;
}

T
Thomas Gleixner 已提交
2444 2445 2446 2447
/*
 * Allocate and initialize a counter structure
 */
static struct perf_counter *
2448 2449
perf_counter_alloc(struct perf_counter_hw_event *hw_event,
		   int cpu,
2450
		   struct perf_counter_context *ctx,
2451 2452
		   struct perf_counter *group_leader,
		   gfp_t gfpflags)
T
Thomas Gleixner 已提交
2453
{
2454
	const struct hw_perf_counter_ops *hw_ops;
I
Ingo Molnar 已提交
2455
	struct perf_counter *counter;
T
Thomas Gleixner 已提交
2456

2457
	counter = kzalloc(sizeof(*counter), gfpflags);
T
Thomas Gleixner 已提交
2458 2459 2460
	if (!counter)
		return NULL;

2461 2462 2463 2464 2465 2466 2467
	/*
	 * Single counters are their own group leaders, with an
	 * empty sibling list:
	 */
	if (!group_leader)
		group_leader = counter;

T
Thomas Gleixner 已提交
2468
	mutex_init(&counter->mutex);
2469
	INIT_LIST_HEAD(&counter->list_entry);
P
Peter Zijlstra 已提交
2470
	INIT_LIST_HEAD(&counter->event_entry);
2471
	INIT_LIST_HEAD(&counter->sibling_list);
T
Thomas Gleixner 已提交
2472 2473
	init_waitqueue_head(&counter->waitq);

2474 2475
	mutex_init(&counter->mmap_mutex);

2476 2477
	INIT_LIST_HEAD(&counter->child_list);

I
Ingo Molnar 已提交
2478 2479
	counter->cpu			= cpu;
	counter->hw_event		= *hw_event;
2480
	counter->group_leader		= group_leader;
I
Ingo Molnar 已提交
2481
	counter->hw_ops			= NULL;
2482
	counter->ctx			= ctx;
I
Ingo Molnar 已提交
2483

2484
	counter->state = PERF_COUNTER_STATE_INACTIVE;
2485 2486 2487
	if (hw_event->disabled)
		counter->state = PERF_COUNTER_STATE_OFF;

2488
	hw_ops = NULL;
2489

2490
	if (perf_event_raw(hw_event)) {
2491
		hw_ops = hw_perf_counter_init(counter);
2492 2493 2494 2495
		goto done;
	}

	switch (perf_event_type(hw_event)) {
2496
	case PERF_TYPE_HARDWARE:
2497
		hw_ops = hw_perf_counter_init(counter);
2498 2499 2500 2501 2502 2503 2504 2505 2506 2507
		break;

	case PERF_TYPE_SOFTWARE:
		hw_ops = sw_perf_counter_init(counter);
		break;

	case PERF_TYPE_TRACEPOINT:
		hw_ops = tp_perf_counter_init(counter);
		break;
	}
2508

I
Ingo Molnar 已提交
2509 2510 2511 2512
	if (!hw_ops) {
		kfree(counter);
		return NULL;
	}
2513
done:
I
Ingo Molnar 已提交
2514
	counter->hw_ops = hw_ops;
T
Thomas Gleixner 已提交
2515 2516 2517 2518 2519

	return counter;
}

/**
2520
 * sys_perf_counter_open - open a performance counter, associate it to a task/cpu
I
Ingo Molnar 已提交
2521 2522
 *
 * @hw_event_uptr:	event type attributes for monitoring/sampling
T
Thomas Gleixner 已提交
2523
 * @pid:		target pid
I
Ingo Molnar 已提交
2524 2525
 * @cpu:		target cpu
 * @group_fd:		group leader counter fd
T
Thomas Gleixner 已提交
2526
 */
2527
SYSCALL_DEFINE5(perf_counter_open,
2528
		const struct perf_counter_hw_event __user *, hw_event_uptr,
2529
		pid_t, pid, int, cpu, int, group_fd, unsigned long, flags)
T
Thomas Gleixner 已提交
2530
{
2531
	struct perf_counter *counter, *group_leader;
I
Ingo Molnar 已提交
2532
	struct perf_counter_hw_event hw_event;
2533
	struct perf_counter_context *ctx;
2534
	struct file *counter_file = NULL;
2535 2536
	struct file *group_file = NULL;
	int fput_needed = 0;
2537
	int fput_needed2 = 0;
T
Thomas Gleixner 已提交
2538 2539
	int ret;

2540 2541 2542 2543
	/* for future expandability... */
	if (flags)
		return -EINVAL;

I
Ingo Molnar 已提交
2544
	if (copy_from_user(&hw_event, hw_event_uptr, sizeof(hw_event)) != 0)
2545 2546
		return -EFAULT;

2547
	/*
I
Ingo Molnar 已提交
2548 2549 2550 2551 2552 2553 2554 2555
	 * Get the target context (task or percpu):
	 */
	ctx = find_get_context(pid, cpu);
	if (IS_ERR(ctx))
		return PTR_ERR(ctx);

	/*
	 * Look up the group leader (we will attach this counter to it):
2556 2557 2558 2559 2560 2561
	 */
	group_leader = NULL;
	if (group_fd != -1) {
		ret = -EINVAL;
		group_file = fget_light(group_fd, &fput_needed);
		if (!group_file)
I
Ingo Molnar 已提交
2562
			goto err_put_context;
2563
		if (group_file->f_op != &perf_fops)
I
Ingo Molnar 已提交
2564
			goto err_put_context;
2565 2566 2567

		group_leader = group_file->private_data;
		/*
I
Ingo Molnar 已提交
2568 2569 2570 2571 2572 2573 2574 2575
		 * Do not allow a recursive hierarchy (this new sibling
		 * becoming part of another group-sibling):
		 */
		if (group_leader->group_leader != group_leader)
			goto err_put_context;
		/*
		 * Do not allow to attach to a group in a different
		 * task or CPU context:
2576
		 */
I
Ingo Molnar 已提交
2577 2578
		if (group_leader->ctx != ctx)
			goto err_put_context;
2579 2580 2581 2582 2583
		/*
		 * Only a group leader can be exclusive or pinned
		 */
		if (hw_event.exclusive || hw_event.pinned)
			goto err_put_context;
2584 2585
	}

2586
	ret = -EINVAL;
2587 2588
	counter = perf_counter_alloc(&hw_event, cpu, ctx, group_leader,
				     GFP_KERNEL);
T
Thomas Gleixner 已提交
2589 2590 2591 2592 2593
	if (!counter)
		goto err_put_context;

	ret = anon_inode_getfd("[perf_counter]", &perf_fops, counter, 0);
	if (ret < 0)
2594 2595 2596 2597 2598 2599 2600
		goto err_free_put_context;

	counter_file = fget_light(ret, &fput_needed2);
	if (!counter_file)
		goto err_free_put_context;

	counter->filp = counter_file;
2601
	mutex_lock(&ctx->mutex);
2602
	perf_install_in_context(ctx, counter, cpu);
2603
	mutex_unlock(&ctx->mutex);
2604 2605

	fput_light(counter_file, fput_needed2);
T
Thomas Gleixner 已提交
2606

2607 2608 2609
out_fput:
	fput_light(group_file, fput_needed);

T
Thomas Gleixner 已提交
2610 2611
	return ret;

2612
err_free_put_context:
T
Thomas Gleixner 已提交
2613 2614 2615 2616 2617
	kfree(counter);

err_put_context:
	put_context(ctx);

2618
	goto out_fput;
T
Thomas Gleixner 已提交
2619 2620
}

2621 2622 2623 2624 2625 2626 2627 2628 2629
/*
 * Initialize the perf_counter context in a task_struct:
 */
static void
__perf_counter_init_context(struct perf_counter_context *ctx,
			    struct task_struct *task)
{
	memset(ctx, 0, sizeof(*ctx));
	spin_lock_init(&ctx->lock);
2630
	mutex_init(&ctx->mutex);
2631
	INIT_LIST_HEAD(&ctx->counter_list);
P
Peter Zijlstra 已提交
2632
	INIT_LIST_HEAD(&ctx->event_list);
2633 2634 2635 2636 2637 2638
	ctx->task = task;
}

/*
 * inherit a counter from parent task to child task:
 */
2639
static struct perf_counter *
2640 2641 2642 2643
inherit_counter(struct perf_counter *parent_counter,
	      struct task_struct *parent,
	      struct perf_counter_context *parent_ctx,
	      struct task_struct *child,
2644
	      struct perf_counter *group_leader,
2645 2646 2647 2648
	      struct perf_counter_context *child_ctx)
{
	struct perf_counter *child_counter;

2649 2650 2651 2652 2653 2654 2655 2656 2657
	/*
	 * Instead of creating recursive hierarchies of counters,
	 * we link inherited counters back to the original parent,
	 * which has a filp for sure, which we use as the reference
	 * count:
	 */
	if (parent_counter->parent)
		parent_counter = parent_counter->parent;

2658
	child_counter = perf_counter_alloc(&parent_counter->hw_event,
2659 2660
					   parent_counter->cpu, child_ctx,
					   group_leader, GFP_KERNEL);
2661
	if (!child_counter)
2662
		return NULL;
2663 2664 2665 2666 2667

	/*
	 * Link it up in the child's context:
	 */
	child_counter->task = child;
2668
	add_counter_to_ctx(child_counter, child_ctx);
2669 2670 2671 2672 2673 2674 2675 2676 2677 2678 2679 2680 2681 2682 2683

	child_counter->parent = parent_counter;
	/*
	 * inherit into child's child as well:
	 */
	child_counter->hw_event.inherit = 1;

	/*
	 * Get a reference to the parent filp - we will fput it
	 * when the child counter exits. This is safe to do because
	 * we are in the parent and we know that the filp still
	 * exists and has a nonzero count:
	 */
	atomic_long_inc(&parent_counter->filp->f_count);

2684 2685 2686 2687 2688 2689 2690 2691 2692 2693 2694 2695 2696 2697 2698 2699 2700 2701 2702 2703 2704 2705 2706 2707 2708 2709 2710 2711 2712 2713 2714 2715 2716 2717 2718 2719 2720 2721 2722
	/*
	 * Link this into the parent counter's child list
	 */
	mutex_lock(&parent_counter->mutex);
	list_add_tail(&child_counter->child_list, &parent_counter->child_list);

	/*
	 * Make the child state follow the state of the parent counter,
	 * not its hw_event.disabled bit.  We hold the parent's mutex,
	 * so we won't race with perf_counter_{en,dis}able_family.
	 */
	if (parent_counter->state >= PERF_COUNTER_STATE_INACTIVE)
		child_counter->state = PERF_COUNTER_STATE_INACTIVE;
	else
		child_counter->state = PERF_COUNTER_STATE_OFF;

	mutex_unlock(&parent_counter->mutex);

	return child_counter;
}

static int inherit_group(struct perf_counter *parent_counter,
	      struct task_struct *parent,
	      struct perf_counter_context *parent_ctx,
	      struct task_struct *child,
	      struct perf_counter_context *child_ctx)
{
	struct perf_counter *leader;
	struct perf_counter *sub;

	leader = inherit_counter(parent_counter, parent, parent_ctx,
				 child, NULL, child_ctx);
	if (!leader)
		return -ENOMEM;
	list_for_each_entry(sub, &parent_counter->sibling_list, list_entry) {
		if (!inherit_counter(sub, parent, parent_ctx,
				     child, leader, child_ctx))
			return -ENOMEM;
	}
2723 2724 2725
	return 0;
}

2726 2727 2728 2729 2730 2731 2732 2733 2734 2735 2736 2737
static void sync_child_counter(struct perf_counter *child_counter,
			       struct perf_counter *parent_counter)
{
	u64 parent_val, child_val;

	parent_val = atomic64_read(&parent_counter->count);
	child_val = atomic64_read(&child_counter->count);

	/*
	 * Add back the child's count to the parent's count:
	 */
	atomic64_add(child_val, &parent_counter->count);
2738 2739 2740 2741
	atomic64_add(child_counter->total_time_enabled,
		     &parent_counter->child_total_time_enabled);
	atomic64_add(child_counter->total_time_running,
		     &parent_counter->child_total_time_running);
2742 2743 2744 2745 2746 2747 2748 2749 2750 2751 2752 2753 2754 2755 2756

	/*
	 * Remove this counter from the parent's list
	 */
	mutex_lock(&parent_counter->mutex);
	list_del_init(&child_counter->child_list);
	mutex_unlock(&parent_counter->mutex);

	/*
	 * Release the parent counter, if this was the last
	 * reference to it.
	 */
	fput(parent_counter->filp);
}

2757 2758 2759 2760 2761 2762
static void
__perf_counter_exit_task(struct task_struct *child,
			 struct perf_counter *child_counter,
			 struct perf_counter_context *child_ctx)
{
	struct perf_counter *parent_counter;
2763
	struct perf_counter *sub, *tmp;
2764 2765

	/*
2766 2767 2768 2769 2770 2771
	 * If we do not self-reap then we have to wait for the
	 * child task to unschedule (it will happen for sure),
	 * so that its counter is at its final count. (This
	 * condition triggers rarely - child tasks usually get
	 * off their CPU before the parent has a chance to
	 * get this far into the reaping action)
2772
	 */
2773 2774 2775
	if (child != current) {
		wait_task_inactive(child, 0);
		list_del_init(&child_counter->list_entry);
2776
		update_counter_times(child_counter);
2777
	} else {
2778
		struct perf_cpu_context *cpuctx;
2779 2780 2781 2782 2783 2784 2785 2786 2787 2788 2789
		unsigned long flags;
		u64 perf_flags;

		/*
		 * Disable and unlink this counter.
		 *
		 * Be careful about zapping the list - IRQ/NMI context
		 * could still be processing it:
		 */
		curr_rq_lock_irq_save(&flags);
		perf_flags = hw_perf_save_disable();
2790 2791 2792

		cpuctx = &__get_cpu_var(perf_cpu_context);

2793
		group_sched_out(child_counter, cpuctx, child_ctx);
2794
		update_counter_times(child_counter);
2795

2796
		list_del_init(&child_counter->list_entry);
2797

2798
		child_ctx->nr_counters--;
2799

2800 2801 2802
		hw_perf_restore(perf_flags);
		curr_rq_unlock_irq_restore(&flags);
	}
2803 2804 2805 2806 2807 2808 2809

	parent_counter = child_counter->parent;
	/*
	 * It can happen that parent exits first, and has counters
	 * that are still around due to the child reference. These
	 * counters need to be zapped - but otherwise linger.
	 */
2810 2811 2812 2813
	if (parent_counter) {
		sync_child_counter(child_counter, parent_counter);
		list_for_each_entry_safe(sub, tmp, &child_counter->sibling_list,
					 list_entry) {
2814
			if (sub->parent) {
2815
				sync_child_counter(sub, sub->parent);
2816
				free_counter(sub);
2817
			}
2818
		}
2819
		free_counter(child_counter);
2820
	}
2821 2822 2823
}

/*
2824
 * When a child task exits, feed back counter values to parent counters.
2825
 *
2826
 * Note: we may be running in child context, but the PID is not hashed
2827 2828 2829 2830 2831 2832 2833 2834 2835 2836 2837 2838 2839 2840 2841 2842 2843 2844 2845 2846 2847 2848 2849
 * anymore so new counters will not be added.
 */
void perf_counter_exit_task(struct task_struct *child)
{
	struct perf_counter *child_counter, *tmp;
	struct perf_counter_context *child_ctx;

	child_ctx = &child->perf_counter_ctx;

	if (likely(!child_ctx->nr_counters))
		return;

	list_for_each_entry_safe(child_counter, tmp, &child_ctx->counter_list,
				 list_entry)
		__perf_counter_exit_task(child, child_counter, child_ctx);
}

/*
 * Initialize the perf_counter context in task_struct
 */
void perf_counter_init_task(struct task_struct *child)
{
	struct perf_counter_context *child_ctx, *parent_ctx;
2850
	struct perf_counter *counter;
2851 2852 2853 2854 2855 2856 2857 2858 2859 2860 2861 2862 2863 2864 2865 2866 2867 2868 2869
	struct task_struct *parent = current;

	child_ctx  =  &child->perf_counter_ctx;
	parent_ctx = &parent->perf_counter_ctx;

	__perf_counter_init_context(child_ctx, child);

	/*
	 * This is executed from the parent task context, so inherit
	 * counters that have been marked for cloning:
	 */

	if (likely(!parent_ctx->nr_counters))
		return;

	/*
	 * Lock the parent list. No need to lock the child - not PID
	 * hashed yet and not running, so nobody can access it.
	 */
2870
	mutex_lock(&parent_ctx->mutex);
2871 2872 2873 2874 2875 2876

	/*
	 * We dont have to disable NMIs - we are only looking at
	 * the list, not manipulating it:
	 */
	list_for_each_entry(counter, &parent_ctx->counter_list, list_entry) {
2877
		if (!counter->hw_event.inherit)
2878 2879
			continue;

2880
		if (inherit_group(counter, parent,
2881 2882 2883 2884
				  parent_ctx, child, child_ctx))
			break;
	}

2885
	mutex_unlock(&parent_ctx->mutex);
2886 2887
}

2888
static void __cpuinit perf_counter_init_cpu(int cpu)
T
Thomas Gleixner 已提交
2889
{
2890
	struct perf_cpu_context *cpuctx;
T
Thomas Gleixner 已提交
2891

2892 2893
	cpuctx = &per_cpu(perf_cpu_context, cpu);
	__perf_counter_init_context(&cpuctx->ctx, NULL);
T
Thomas Gleixner 已提交
2894 2895

	mutex_lock(&perf_resource_mutex);
2896
	cpuctx->max_pertask = perf_max_counters - perf_reserved_percpu;
T
Thomas Gleixner 已提交
2897
	mutex_unlock(&perf_resource_mutex);
2898

2899
	hw_perf_counter_setup(cpu);
T
Thomas Gleixner 已提交
2900 2901 2902
}

#ifdef CONFIG_HOTPLUG_CPU
2903
static void __perf_counter_exit_cpu(void *info)
T
Thomas Gleixner 已提交
2904 2905 2906 2907 2908
{
	struct perf_cpu_context *cpuctx = &__get_cpu_var(perf_cpu_context);
	struct perf_counter_context *ctx = &cpuctx->ctx;
	struct perf_counter *counter, *tmp;

2909 2910
	list_for_each_entry_safe(counter, tmp, &ctx->counter_list, list_entry)
		__perf_counter_remove_from_context(counter);
T
Thomas Gleixner 已提交
2911
}
2912
static void perf_counter_exit_cpu(int cpu)
T
Thomas Gleixner 已提交
2913
{
2914 2915 2916 2917
	struct perf_cpu_context *cpuctx = &per_cpu(perf_cpu_context, cpu);
	struct perf_counter_context *ctx = &cpuctx->ctx;

	mutex_lock(&ctx->mutex);
2918
	smp_call_function_single(cpu, __perf_counter_exit_cpu, NULL, 1);
2919
	mutex_unlock(&ctx->mutex);
T
Thomas Gleixner 已提交
2920 2921
}
#else
2922
static inline void perf_counter_exit_cpu(int cpu) { }
T
Thomas Gleixner 已提交
2923 2924 2925 2926 2927 2928 2929 2930 2931 2932 2933
#endif

static int __cpuinit
perf_cpu_notify(struct notifier_block *self, unsigned long action, void *hcpu)
{
	unsigned int cpu = (long)hcpu;

	switch (action) {

	case CPU_UP_PREPARE:
	case CPU_UP_PREPARE_FROZEN:
2934
		perf_counter_init_cpu(cpu);
T
Thomas Gleixner 已提交
2935 2936 2937 2938
		break;

	case CPU_DOWN_PREPARE:
	case CPU_DOWN_PREPARE_FROZEN:
2939
		perf_counter_exit_cpu(cpu);
T
Thomas Gleixner 已提交
2940 2941 2942 2943 2944 2945 2946 2947 2948 2949 2950 2951 2952 2953 2954 2955 2956 2957 2958 2959 2960 2961 2962 2963 2964 2965 2966 2967 2968 2969 2970 2971 2972 2973 2974 2975 2976 2977 2978 2979 2980 2981 2982 2983 2984 2985 2986 2987 2988 2989 2990 2991 2992 2993 2994 2995 2996 2997 2998 2999 3000 3001 3002 3003 3004 3005 3006 3007 3008 3009 3010 3011 3012 3013 3014 3015 3016 3017 3018 3019 3020 3021 3022 3023 3024 3025 3026 3027 3028 3029 3030 3031 3032 3033 3034 3035 3036 3037 3038 3039 3040 3041 3042 3043 3044 3045 3046 3047 3048 3049 3050 3051 3052
		break;

	default:
		break;
	}

	return NOTIFY_OK;
}

static struct notifier_block __cpuinitdata perf_cpu_nb = {
	.notifier_call		= perf_cpu_notify,
};

static int __init perf_counter_init(void)
{
	perf_cpu_notify(&perf_cpu_nb, (unsigned long)CPU_UP_PREPARE,
			(void *)(long)smp_processor_id());
	register_cpu_notifier(&perf_cpu_nb);

	return 0;
}
early_initcall(perf_counter_init);

static ssize_t perf_show_reserve_percpu(struct sysdev_class *class, char *buf)
{
	return sprintf(buf, "%d\n", perf_reserved_percpu);
}

static ssize_t
perf_set_reserve_percpu(struct sysdev_class *class,
			const char *buf,
			size_t count)
{
	struct perf_cpu_context *cpuctx;
	unsigned long val;
	int err, cpu, mpt;

	err = strict_strtoul(buf, 10, &val);
	if (err)
		return err;
	if (val > perf_max_counters)
		return -EINVAL;

	mutex_lock(&perf_resource_mutex);
	perf_reserved_percpu = val;
	for_each_online_cpu(cpu) {
		cpuctx = &per_cpu(perf_cpu_context, cpu);
		spin_lock_irq(&cpuctx->ctx.lock);
		mpt = min(perf_max_counters - cpuctx->ctx.nr_counters,
			  perf_max_counters - perf_reserved_percpu);
		cpuctx->max_pertask = mpt;
		spin_unlock_irq(&cpuctx->ctx.lock);
	}
	mutex_unlock(&perf_resource_mutex);

	return count;
}

static ssize_t perf_show_overcommit(struct sysdev_class *class, char *buf)
{
	return sprintf(buf, "%d\n", perf_overcommit);
}

static ssize_t
perf_set_overcommit(struct sysdev_class *class, const char *buf, size_t count)
{
	unsigned long val;
	int err;

	err = strict_strtoul(buf, 10, &val);
	if (err)
		return err;
	if (val > 1)
		return -EINVAL;

	mutex_lock(&perf_resource_mutex);
	perf_overcommit = val;
	mutex_unlock(&perf_resource_mutex);

	return count;
}

static SYSDEV_CLASS_ATTR(
				reserve_percpu,
				0644,
				perf_show_reserve_percpu,
				perf_set_reserve_percpu
			);

static SYSDEV_CLASS_ATTR(
				overcommit,
				0644,
				perf_show_overcommit,
				perf_set_overcommit
			);

static struct attribute *perfclass_attrs[] = {
	&attr_reserve_percpu.attr,
	&attr_overcommit.attr,
	NULL
};

static struct attribute_group perfclass_attr_group = {
	.attrs			= perfclass_attrs,
	.name			= "perf_counters",
};

static int __init perf_counter_sysfs_init(void)
{
	return sysfs_create_group(&cpu_sysdev_class.kset.kobj,
				  &perfclass_attr_group);
}
device_initcall(perf_counter_sysfs_init);