perf_counter.c 57.5 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12
/*
 * Performance counter core code
 *
 *  Copyright(C) 2008 Thomas Gleixner <tglx@linutronix.de>
 *  Copyright(C) 2008 Red Hat, Inc., Ingo Molnar
 *
 *  For licencing details see kernel-base/COPYING
 */

#include <linux/fs.h>
#include <linux/cpu.h>
#include <linux/smp.h>
13
#include <linux/file.h>
14 15 16 17 18 19 20
#include <linux/poll.h>
#include <linux/sysfs.h>
#include <linux/ptrace.h>
#include <linux/percpu.h>
#include <linux/uaccess.h>
#include <linux/syscalls.h>
#include <linux/anon_inodes.h>
21
#include <linux/kernel_stat.h>
22
#include <linux/perf_counter.h>
23 24
#include <linux/mm.h>
#include <linux/vmstat.h>
25
#include <linux/rculist.h>
26

27 28
#include <asm/irq_regs.h>

29 30 31 32 33
/*
 * Each CPU has a list of per CPU counters:
 */
DEFINE_PER_CPU(struct perf_cpu_context, perf_cpu_context);

34
int perf_max_counters __read_mostly = 1;
35 36 37 38 39 40 41 42 43 44 45
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:
 */
46
extern __weak const struct hw_perf_counter_ops *
Ingo Molnar's avatar
Ingo Molnar committed
47
hw_perf_counter_init(struct perf_counter *counter)
48
{
49
	return NULL;
50 51
}

52
u64 __weak hw_perf_save_disable(void)		{ return 0; }
53
void __weak hw_perf_restore(u64 ctrl)		{ barrier(); }
54
void __weak hw_perf_counter_setup(int cpu)	{ barrier(); }
55 56 57 58 59 60
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;
}
61

62 63
void __weak perf_counter_print_debug(void)	{ }

64 65 66 67 68 69 70 71 72 73 74 75 76 77
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);
	else
		list_add_tail(&counter->list_entry, &group_leader->sibling_list);
78 79

	list_add_rcu(&counter->event_entry, &ctx->event_list);
80 81 82 83 84 85 86 87
}

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);
88
	list_del_rcu(&counter->event_entry);
89 90 91 92 93 94 95 96 97

	/*
	 * 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) {

98
		list_move_tail(&sibling->list_entry, &ctx->counter_list);
99 100 101 102
		sibling->group_leader = sibling;
	}
}

103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121
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;
	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;
}

122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143
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;
}

144 145 146 147 148 149
/*
 * 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.
 */
150
static void __perf_counter_remove_from_context(void *info)
151 152 153 154
{
	struct perf_cpu_context *cpuctx = &__get_cpu_var(perf_cpu_context);
	struct perf_counter *counter = info;
	struct perf_counter_context *ctx = counter->ctx;
155
	unsigned long flags;
156
	u64 perf_flags;
157 158 159 160 161 162 163 164 165

	/*
	 * 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;

166 167
	curr_rq_lock_irq_save(&flags);
	spin_lock(&ctx->lock);
168

169 170 171
	counter_sched_out(counter, cpuctx, ctx);

	counter->task = NULL;
172 173 174 175 176 177
	ctx->nr_counters--;

	/*
	 * Protect the list operation against NMI by disabling the
	 * counters on a global level. NOP for non NMI based counters.
	 */
178
	perf_flags = hw_perf_save_disable();
179
	list_del_counter(counter, ctx);
180
	hw_perf_restore(perf_flags);
181 182 183 184 185 186 187 188 189 190 191

	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);
	}

192 193
	spin_unlock(&ctx->lock);
	curr_rq_unlock_irq_restore(&flags);
194 195 196 197 198 199
}


/*
 * Remove the counter from a task's (or a CPU's) list of counters.
 *
200
 * Must be called with counter->mutex and ctx->mutex held.
201 202 203 204
 *
 * CPU counters are removed with a smp call. For task counters we only
 * call when the task is on a CPU.
 */
205
static void perf_counter_remove_from_context(struct perf_counter *counter)
206 207 208 209 210 211 212 213 214 215
{
	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,
216
					 __perf_counter_remove_from_context,
217 218 219 220 221
					 counter, 1);
		return;
	}

retry:
222
	task_oncpu_function_call(task, __perf_counter_remove_from_context,
223 224 225 226 227 228
				 counter);

	spin_lock_irq(&ctx->lock);
	/*
	 * If the context is active we need to retry the smp call.
	 */
229
	if (ctx->nr_active && !list_empty(&counter->list_entry)) {
230 231 232 233 234 235
		spin_unlock_irq(&ctx->lock);
		goto retry;
	}

	/*
	 * The lock prevents that this context is scheduled in so we
236
	 * can remove the counter safely, if the call above did not
237 238
	 * succeed.
	 */
239
	if (!list_empty(&counter->list_entry)) {
240
		ctx->nr_counters--;
241
		list_del_counter(counter, ctx);
242 243 244 245 246
		counter->task = NULL;
	}
	spin_unlock_irq(&ctx->lock);
}

247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339
/*
 * 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) {
		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.
	 */
	if (counter->state == PERF_COUNTER_STATE_INACTIVE)
		counter->state = PERF_COUNTER_STATE_OFF;

	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);
}

340 341 342 343 344 345
static int
counter_sched_in(struct perf_counter *counter,
		 struct perf_cpu_context *cpuctx,
		 struct perf_counter_context *ctx,
		 int cpu)
{
346
	if (counter->state <= PERF_COUNTER_STATE_OFF)
347 348 349 350 351 352 353 354 355 356 357 358 359 360 361
		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;
	}

362 363
	if (!is_software_counter(counter))
		cpuctx->active_oncpu++;
364 365
	ctx->nr_active++;

366 367 368
	if (counter->hw_event.exclusive)
		cpuctx->exclusive = 1;

369 370 371
	return 0;
}

372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418
/*
 * 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;
	list_for_each_entry(counter, &leader->sibling_list, list_entry)
		if (!is_software_counter(counter))
			return 0;
	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;
}

419
/*
420
 * Cross CPU call to install and enable a performance counter
421 422 423 424 425 426
 */
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;
427
	struct perf_counter *leader = counter->group_leader;
428
	int cpu = smp_processor_id();
429
	unsigned long flags;
430
	u64 perf_flags;
431
	int err;
432 433 434 435 436 437 438 439 440

	/*
	 * 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;

441 442
	curr_rq_lock_irq_save(&flags);
	spin_lock(&ctx->lock);
443 444 445 446 447

	/*
	 * Protect the list operation against NMI by disabling the
	 * counters on a global level. NOP for non NMI based counters.
	 */
448
	perf_flags = hw_perf_save_disable();
449

450
	list_add_counter(counter, ctx);
451
	ctx->nr_counters++;
452
	counter->prev_state = PERF_COUNTER_STATE_OFF;
453

454 455 456 457 458 459 460 461
	/*
	 * 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;

462 463 464 465 466
	/*
	 * 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.
	 */
467
	if (!group_can_go_on(counter, cpuctx, 1))
468 469 470 471
		err = -EEXIST;
	else
		err = counter_sched_in(counter, cpuctx, ctx, cpu);

472 473 474 475 476 477 478 479 480 481 482
	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);
		if (leader->hw_event.pinned)
			leader->state = PERF_COUNTER_STATE_ERROR;
	}
483

484
	if (!err && !ctx->task && cpuctx->max_pertask)
485 486
		cpuctx->max_pertask--;

487
 unlock:
488 489
	hw_perf_restore(perf_flags);

490 491
	spin_unlock(&ctx->lock);
	curr_rq_unlock_irq_restore(&flags);
492 493 494 495 496 497 498 499 500 501 502
}

/*
 * 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.
503 504
 *
 * Must be called with ctx->mutex held.
505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531
 */
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.
	 */
532
	if (ctx->is_active && list_empty(&counter->list_entry)) {
533 534 535 536 537 538 539 540 541
		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.
	 */
542 543
	if (list_empty(&counter->list_entry)) {
		list_add_counter(counter, ctx);
544 545 546 547 548
		ctx->nr_counters++;
	}
	spin_unlock_irq(&ctx->lock);
}

549 550 551 552
/*
 * Cross CPU call to enable a performance counter
 */
static void __perf_counter_enable(void *info)
553
{
554 555 556 557 558 559
	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;
560

561 562 563 564 565
	/*
	 * 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)
566 567
		return;

568 569 570
	curr_rq_lock_irq_save(&flags);
	spin_lock(&ctx->lock);

571
	counter->prev_state = counter->state;
572 573 574
	if (counter->state >= PERF_COUNTER_STATE_INACTIVE)
		goto unlock;
	counter->state = PERF_COUNTER_STATE_INACTIVE;
575 576

	/*
577 578
	 * If the counter is in a group and isn't the group leader,
	 * then don't put it on unless the group is on.
579
	 */
580 581
	if (leader != counter && leader->state != PERF_COUNTER_STATE_ACTIVE)
		goto unlock;
582

583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674
	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);
		if (leader->hw_event.pinned)
			leader->state = PERF_COUNTER_STATE_ERROR;
	}

 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.
	 */
	if (counter->state == PERF_COUNTER_STATE_OFF)
		counter->state = PERF_COUNTER_STATE_INACTIVE;
 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);
675 676
}

677 678 679 680
void __perf_counter_sched_out(struct perf_counter_context *ctx,
			      struct perf_cpu_context *cpuctx)
{
	struct perf_counter *counter;
681
	u64 flags;
682

683 684
	spin_lock(&ctx->lock);
	ctx->is_active = 0;
685
	if (likely(!ctx->nr_counters))
686
		goto out;
687

688
	flags = hw_perf_save_disable();
689 690 691 692
	if (ctx->nr_active) {
		list_for_each_entry(counter, &ctx->counter_list, list_entry)
			group_sched_out(counter, cpuctx, ctx);
	}
693
	hw_perf_restore(flags);
694
 out:
695 696 697
	spin_unlock(&ctx->lock);
}

698 699 700 701 702 703
/*
 * 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.
 *
Ingo Molnar's avatar
Ingo Molnar committed
704
 * This does not protect us against NMI, but disable()
705 706 707 708 709 710 711 712 713 714 715 716
 * 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;

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

717 718
	__perf_counter_sched_out(ctx, cpuctx);

719 720 721
	cpuctx->task_ctx = NULL;
}

722
static void perf_counter_cpu_sched_out(struct perf_cpu_context *cpuctx)
723
{
724
	__perf_counter_sched_out(&cpuctx->ctx, cpuctx);
725 726
}

727
static int
728 729 730 731 732
group_sched_in(struct perf_counter *group_counter,
	       struct perf_cpu_context *cpuctx,
	       struct perf_counter_context *ctx,
	       int cpu)
{
733
	struct perf_counter *counter, *partial_group;
734 735 736 737 738 739 740 741
	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;
742

743
	group_counter->prev_state = group_counter->state;
744 745
	if (counter_sched_in(group_counter, cpuctx, ctx, cpu))
		return -EAGAIN;
746 747 748 749

	/*
	 * Schedule in siblings as one group (if any):
	 */
750
	list_for_each_entry(counter, &group_counter->sibling_list, list_entry) {
751
		counter->prev_state = counter->state;
752 753 754 755 756 757
		if (counter_sched_in(counter, cpuctx, ctx, cpu)) {
			partial_group = counter;
			goto group_error;
		}
	}

758
	return 0;
759 760 761 762 763 764 765 766 767 768

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);
769
	}
770
	counter_sched_out(group_counter, cpuctx, ctx);
771

772
	return -EAGAIN;
773 774
}

775 776 777
static void
__perf_counter_sched_in(struct perf_counter_context *ctx,
			struct perf_cpu_context *cpuctx, int cpu)
778 779
{
	struct perf_counter *counter;
780
	u64 flags;
781
	int can_add_hw = 1;
782

783 784
	spin_lock(&ctx->lock);
	ctx->is_active = 1;
785
	if (likely(!ctx->nr_counters))
786
		goto out;
787

788
	flags = hw_perf_save_disable();
789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811

	/*
	 * 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.
		 */
		if (counter->state == PERF_COUNTER_STATE_INACTIVE)
			counter->state = PERF_COUNTER_STATE_ERROR;
	}

812
	list_for_each_entry(counter, &ctx->counter_list, list_entry) {
813 814 815 816 817 818 819 820
		/*
		 * 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;

821 822 823 824
		/*
		 * Listen to the 'cpu' scheduling filter constraint
		 * of counters:
		 */
825 826 827
		if (counter->cpu != -1 && counter->cpu != cpu)
			continue;

828
		if (group_can_go_on(counter, cpuctx, can_add_hw)) {
829 830
			if (group_sched_in(counter, cpuctx, ctx, cpu))
				can_add_hw = 0;
831
		}
832
	}
833
	hw_perf_restore(flags);
834
 out:
835
	spin_unlock(&ctx->lock);
836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852
}

/*
 * 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;
853

854
	__perf_counter_sched_in(ctx, cpuctx, cpu);
855 856 857
	cpuctx->task_ctx = ctx;
}

858 859 860 861 862 863 864
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);
}

865 866 867 868 869
int perf_counter_task_disable(void)
{
	struct task_struct *curr = current;
	struct perf_counter_context *ctx = &curr->perf_counter_ctx;
	struct perf_counter *counter;
870
	unsigned long flags;
871 872 873 874 875 876
	u64 perf_flags;
	int cpu;

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

877
	curr_rq_lock_irq_save(&flags);
878 879
	cpu = smp_processor_id();

880 881 882
	/* force the update of the task clock: */
	__task_delta_exec(curr, 1);

883 884 885 886 887 888 889 890 891
	perf_counter_task_sched_out(curr, cpu);

	spin_lock(&ctx->lock);

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

892 893 894 895
	list_for_each_entry(counter, &ctx->counter_list, list_entry) {
		if (counter->state != PERF_COUNTER_STATE_ERROR)
			counter->state = PERF_COUNTER_STATE_OFF;
	}
896

897 898 899 900
	hw_perf_restore(perf_flags);

	spin_unlock(&ctx->lock);

901
	curr_rq_unlock_irq_restore(&flags);
902 903 904 905 906 907 908 909 910

	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;
911
	unsigned long flags;
912 913 914 915 916 917
	u64 perf_flags;
	int cpu;

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

918
	curr_rq_lock_irq_save(&flags);
919 920
	cpu = smp_processor_id();

921 922 923
	/* force the update of the task clock: */
	__task_delta_exec(curr, 1);

924 925
	perf_counter_task_sched_out(curr, cpu);

926 927 928 929 930 931 932 933
	spin_lock(&ctx->lock);

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

	list_for_each_entry(counter, &ctx->counter_list, list_entry) {
934
		if (counter->state > PERF_COUNTER_STATE_OFF)
935
			continue;
936
		counter->state = PERF_COUNTER_STATE_INACTIVE;
937
		counter->hw_event.disabled = 0;
938 939 940 941 942 943 944
	}
	hw_perf_restore(perf_flags);

	spin_unlock(&ctx->lock);

	perf_counter_task_sched_in(curr, cpu);

945
	curr_rq_unlock_irq_restore(&flags);
946 947 948 949

	return 0;
}

950 951 952 953
/*
 * Round-robin a context's counters:
 */
static void rotate_ctx(struct perf_counter_context *ctx)
954 955
{
	struct perf_counter *counter;
956
	u64 perf_flags;
957

958
	if (!ctx->nr_counters)
959 960 961 962
		return;

	spin_lock(&ctx->lock);
	/*
963
	 * Rotate the first entry last (works just fine for group counters too):
964
	 */
965
	perf_flags = hw_perf_save_disable();
966
	list_for_each_entry(counter, &ctx->counter_list, list_entry) {
967
		list_move_tail(&counter->list_entry, &ctx->counter_list);
968 969
		break;
	}
970
	hw_perf_restore(perf_flags);
971 972

	spin_unlock(&ctx->lock);
973 974 975 976 977 978 979 980 981 982 983
}

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);
984

985 986 987 988 989 990
	if (rotate_percpu)
		rotate_ctx(&cpuctx->ctx);
	rotate_ctx(ctx);

	if (rotate_percpu)
		perf_counter_cpu_sched_in(cpuctx, cpu);
991 992 993 994 995 996
	perf_counter_task_sched_in(curr, cpu);
}

/*
 * Cross CPU call to read the hardware counter
 */
Ingo Molnar's avatar
Ingo Molnar committed
997
static void __read(void *info)
998
{
Ingo Molnar's avatar
Ingo Molnar committed
999
	struct perf_counter *counter = info;
1000
	unsigned long flags;
Ingo Molnar's avatar
Ingo Molnar committed
1001

1002
	curr_rq_lock_irq_save(&flags);
Ingo Molnar's avatar
Ingo Molnar committed
1003
	counter->hw_ops->read(counter);
1004
	curr_rq_unlock_irq_restore(&flags);
1005 1006
}

1007
static u64 perf_counter_read(struct perf_counter *counter)
1008 1009 1010 1011 1012
{
	/*
	 * If counter is enabled and currently active on a CPU, update the
	 * value in the counter structure:
	 */
1013
	if (counter->state == PERF_COUNTER_STATE_ACTIVE) {
1014
		smp_call_function_single(counter->oncpu,
Ingo Molnar's avatar
Ingo Molnar committed
1015
					 __read, counter, 1);
1016 1017
	}

1018
	return atomic64_read(&counter->count);
1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065
}

/*
 * Cross CPU call to switch performance data pointers
 */
static void __perf_switch_irq_data(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;
	struct perf_data *oldirqdata = counter->irqdata;

	/*
	 * 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) {
		if (cpuctx->task_ctx != ctx)
			return;
		spin_lock(&ctx->lock);
	}

	/* Change the pointer NMI safe */
	atomic_long_set((atomic_long_t *)&counter->irqdata,
			(unsigned long) counter->usrdata);
	counter->usrdata = oldirqdata;

	if (ctx->task)
		spin_unlock(&ctx->lock);
}

static struct perf_data *perf_switch_irq_data(struct perf_counter *counter)
{
	struct perf_counter_context *ctx = counter->ctx;
	struct perf_data *oldirqdata = counter->irqdata;
	struct task_struct *task = ctx->task;

	if (!task) {
		smp_call_function_single(counter->cpu,
					 __perf_switch_irq_data,
					 counter, 1);
		return counter->usrdata;
	}

retry:
	spin_lock_irq(&ctx->lock);
1066
	if (counter->state != PERF_COUNTER_STATE_ACTIVE) {
1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141
		counter->irqdata = counter->usrdata;
		counter->usrdata = oldirqdata;
		spin_unlock_irq(&ctx->lock);
		return oldirqdata;
	}
	spin_unlock_irq(&ctx->lock);
	task_oncpu_function_call(task, __perf_switch_irq_data, counter);
	/* Might have failed, because task was scheduled out */
	if (counter->irqdata == oldirqdata)
		goto retry;

	return counter->usrdata;
}

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;
}

1142 1143 1144 1145 1146 1147 1148 1149
static void free_counter_rcu(struct rcu_head *head)
{
	struct perf_counter *counter;

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

1150 1151 1152 1153 1154 1155 1156 1157 1158 1159
/*
 * 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;

1160
	mutex_lock(&ctx->mutex);
1161 1162
	mutex_lock(&counter->mutex);

1163
	perf_counter_remove_from_context(counter);
1164 1165

	mutex_unlock(&counter->mutex);
1166
	mutex_unlock(&ctx->mutex);
1167

1168
	call_rcu(&counter->rcu_head, free_counter_rcu);
1169
	put_context(ctx);
1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184

	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)
{
	u64 cntval;

	if (count != sizeof(cntval))
		return -EINVAL;

1185 1186 1187 1188 1189 1190 1191 1192
	/*
	 * 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;

1193
	mutex_lock(&counter->mutex);
1194
	cntval = perf_counter_read(counter);
1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227
	mutex_unlock(&counter->mutex);

	return put_user(cntval, (u64 __user *) buf) ? -EFAULT : sizeof(cntval);
}

static ssize_t
perf_copy_usrdata(struct perf_data *usrdata, char __user *buf, size_t count)
{
	if (!usrdata->len)
		return 0;

	count = min(count, (size_t)usrdata->len);
	if (copy_to_user(buf, usrdata->data + usrdata->rd_idx, count))
		return -EFAULT;

	/* Adjust the counters */
	usrdata->len -= count;
	if (!usrdata->len)
		usrdata->rd_idx = 0;
	else
		usrdata->rd_idx += count;

	return count;
}

static ssize_t
perf_read_irq_data(struct perf_counter	*counter,
		   char __user		*buf,
		   size_t		count,
		   int			nonblocking)
{
	struct perf_data *irqdata, *usrdata;
	DECLARE_WAITQUEUE(wait, current);
1228
	ssize_t res, res2;
1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248

	irqdata = counter->irqdata;
	usrdata = counter->usrdata;

	if (usrdata->len + irqdata->len >= count)
		goto read_pending;

	if (nonblocking)
		return -EAGAIN;

	spin_lock_irq(&counter->waitq.lock);
	__add_wait_queue(&counter->waitq, &wait);
	for (;;) {
		set_current_state(TASK_INTERRUPTIBLE);
		if (usrdata->len + irqdata->len >= count)
			break;

		if (signal_pending(current))
			break;

1249 1250 1251
		if (counter->state == PERF_COUNTER_STATE_ERROR)
			break;

1252 1253 1254 1255 1256 1257 1258 1259
		spin_unlock_irq(&counter->waitq.lock);
		schedule();
		spin_lock_irq(&counter->waitq.lock);
	}
	__remove_wait_queue(&counter->waitq, &wait);
	__set_current_state(TASK_RUNNING);
	spin_unlock_irq(&counter->waitq.lock);

1260 1261
	if (usrdata->len + irqdata->len < count &&
	    counter->state != PERF_COUNTER_STATE_ERROR)
1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272
		return -ERESTARTSYS;
read_pending:
	mutex_lock(&counter->mutex);

	/* Drain pending data first: */
	res = perf_copy_usrdata(usrdata, buf, count);
	if (res < 0 || res == count)
		goto out;

	/* Switch irq buffer: */
	usrdata = perf_switch_irq_data(counter);
1273 1274
	res2 = perf_copy_usrdata(usrdata, buf + res, count - res);
	if (res2 < 0) {
1275 1276 1277
		if (!res)
			res = -EFAULT;
	} else {
1278
		res += res2;
1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290
	}
out:
	mutex_unlock(&counter->mutex);

	return res;
}

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

1291
	switch (counter->hw_event.record_type) {
1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318
	case PERF_RECORD_SIMPLE:
		return perf_read_hw(counter, buf, count);

	case PERF_RECORD_IRQ:
	case PERF_RECORD_GROUP:
		return perf_read_irq_data(counter, buf, count,
					  file->f_flags & O_NONBLOCK);
	}
	return -EINVAL;
}

static unsigned int perf_poll(struct file *file, poll_table *wait)
{
	struct perf_counter *counter = file->private_data;
	unsigned int events = 0;
	unsigned long flags;

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

	spin_lock_irqsave(&counter->waitq.lock, flags);
	if (counter->usrdata->len || counter->irqdata->len)
		events |= POLLIN;
	spin_unlock_irqrestore(&counter->waitq.lock, flags);

	return events;
}

1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336
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;
}

1337 1338 1339 1340
static const struct file_operations perf_fops = {
	.release		= perf_release,
	.read			= perf_read,
	.poll			= perf_poll,
1341 1342
	.unlocked_ioctl		= perf_ioctl,
	.compat_ioctl		= perf_ioctl,
1343 1344
};

1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 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
/*
 * 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);
}

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

static void perf_swcounter_store_irq(struct perf_counter *counter, u64 data)
{
	struct perf_data *irqdata = counter->irqdata;

	if (irqdata->len > PERF_DATA_BUFLEN - sizeof(u64)) {
		irqdata->overrun++;
	} else {
		u64 *p = (u64 *) &irqdata->data[irqdata->len];

		*p = data;
		irqdata->len += sizeof(u64);
	}
}

static void perf_swcounter_handle_group(struct perf_counter *sibling)
{
	struct perf_counter *counter, *group_leader = sibling->group_leader;

	list_for_each_entry(counter, &group_leader->sibling_list, list_entry) {
1412
		counter->hw_ops->read(counter);
1413 1414 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
		perf_swcounter_store_irq(sibling, counter->hw_event.type);
		perf_swcounter_store_irq(sibling, atomic64_read(&counter->count));
	}
}

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

	case PERF_RECORD_IRQ:
		perf_swcounter_store_irq(counter, instruction_pointer(regs));
		break;

	case PERF_RECORD_GROUP:
		perf_swcounter_handle_group(counter);
		break;
	}

	if (nmi) {
		counter->wakeup_pending = 1;
		set_tsk_thread_flag(current, TIF_PERF_COUNTERS);
	} else
		wake_up(&counter->waitq);
}

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 1467 1468 1469 1470 1471 1472
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)
		perf_swcounter_interrupt(counter, 0, regs);

	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)
{
	perf_swcounter_save_and_restart(counter);
	perf_swcounter_interrupt(counter, nmi, regs);
}

1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494
static int perf_swcounter_match(struct perf_counter *counter,
				enum hw_event_types event,
				struct pt_regs *regs)
{
	if (counter->state != PERF_COUNTER_STATE_ACTIVE)
		return 0;

	if (counter->hw_event.raw)
		return 0;

	if (counter->hw_event.type != event)
		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;
}

1495 1496 1497 1498 1499 1500 1501 1502
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);
}

1503 1504 1505 1506 1507 1508
static void perf_swcounter_ctx_event(struct perf_counter_context *ctx,
				     enum hw_event_types event, u64 nr,
				     int nmi, struct pt_regs *regs)
{
	struct perf_counter *counter;

1509
	if (list_empty(&ctx->event_list))
1510 1511
		return;

1512 1513
	rcu_read_lock();
	list_for_each_entry_rcu(counter, &ctx->event_list, event_entry) {
1514 1515
		if (perf_swcounter_match(counter, event, regs))
			perf_swcounter_add(counter, nr, nmi, regs);
1516
	}
1517
	rcu_read_unlock();
1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547
}

void perf_swcounter_event(enum hw_event_types event, u64 nr,
			  int nmi, struct pt_regs *regs)
{
	struct perf_cpu_context *cpuctx = &get_cpu_var(perf_cpu_context);

	perf_swcounter_ctx_event(&cpuctx->ctx, event, nr, nmi, regs);
	if (cpuctx->task_ctx)
		perf_swcounter_ctx_event(cpuctx->task_ctx, event, nr, nmi, regs);

	put_cpu_var(perf_cpu_context);
}

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);
}

1548 1549 1550 1551 1552 1553
static const struct hw_perf_counter_ops perf_ops_generic = {
	.enable		= perf_swcounter_enable,
	.disable	= perf_swcounter_disable,
	.read		= perf_swcounter_read,
};

1554 1555 1556 1557
/*
 * Software counter: cpu wall time clock
 */

1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569
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);
}

1570 1571 1572 1573 1574 1575
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));
1576 1577
	hrtimer_init(&hwc->hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
	hwc->hrtimer.function = perf_swcounter_hrtimer;
1578 1579 1580 1581 1582 1583 1584 1585 1586
	if (hwc->irq_period) {
		__hrtimer_start_range_ns(&hwc->hrtimer,
				ns_to_ktime(hwc->irq_period), 0,
				HRTIMER_MODE_REL, 0);
	}

	return 0;
}

1587 1588
static void cpu_clock_perf_counter_disable(struct perf_counter *counter)
{
1589
	hrtimer_cancel(&counter->hw.hrtimer);
1590
	cpu_clock_perf_counter_update(counter);
1591 1592 1593 1594
}

static void cpu_clock_perf_counter_read(struct perf_counter *counter)
{
1595
	cpu_clock_perf_counter_update(counter);
1596 1597 1598
}

static const struct hw_perf_counter_ops perf_ops_cpu_clock = {
Ingo Molnar's avatar
Ingo Molnar committed
1599 1600 1601
	.enable		= cpu_clock_perf_counter_enable,
	.disable	= cpu_clock_perf_counter_disable,
	.read		= cpu_clock_perf_counter_read,
1602 1603
};

1604 1605 1606 1607
/*
 * Software counter: task time clock
 */

1608 1609 1610 1611
/*
 * Called from within the scheduler:
 */
static u64 task_clock_perf_counter_val(struct perf_counter *counter, int update)
1612
{
1613 1614 1615 1616 1617 1618 1619 1620 1621 1622 1623
	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;
1624 1625 1626 1627 1628 1629 1630 1631 1632
	s64 delta;

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

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

	delta = now - prev;

	atomic64_add(delta, &counter->count);
1633 1634
}

1635
static int task_clock_perf_counter_enable(struct perf_counter *counter)
1636
{
1637 1638 1639
	struct hw_perf_counter *hwc = &counter->hw;

	atomic64_set(&hwc->prev_count, task_clock_perf_counter_val(counter, 0));
1640 1641
	hrtimer_init(&hwc->hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
	hwc->hrtimer.function = perf_swcounter_hrtimer;
1642 1643 1644 1645 1646
	if (hwc->irq_period) {
		__hrtimer_start_range_ns(&hwc->hrtimer,
				ns_to_ktime(hwc->irq_period), 0,
				HRTIMER_MODE_REL, 0);
	}
1647 1648

	return 0;
1649 1650 1651
}

static void task_clock_perf_counter_disable(struct perf_counter *counter)
1652
{
1653 1654 1655 1656
	hrtimer_cancel(&counter->hw.hrtimer);
	task_clock_perf_counter_update(counter,
			task_clock_perf_counter_val(counter, 0));
}
1657

1658 1659 1660 1661
static void task_clock_perf_counter_read(struct perf_counter *counter)
{
	task_clock_perf_counter_update(counter,
			task_clock_perf_counter_val(counter, 1));
1662 1663 1664
}

static const struct hw_perf_counter_ops perf_ops_task_clock = {
Ingo Molnar's avatar
Ingo Molnar committed
1665 1666 1667
	.enable		= task_clock_perf_counter_enable,
	.disable	= task_clock_perf_counter_disable,
	.read		= task_clock_perf_counter_read,
1668 1669
};

1670 1671 1672 1673
/*
 * Software counter: context switches
 */

1674
static u64 get_context_switches(struct perf_counter *counter)
1675
{
1676
	struct task_struct *curr = counter->ctx->task;
1677

1678 1679 1680
	if (curr)
		return curr->nvcsw + curr->nivcsw;
	return cpu_nr_switches(smp_processor_id());
1681 1682 1683 1684 1685 1686 1687 1688
}

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

	prev = atomic64_read(&counter->hw.prev_count);
1689
	now = get_context_switches(counter);
1690 1691 1692 1693 1694 1695 1696 1697 1698 1699 1700 1701 1702

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

	delta = now - prev;

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

static void context_switches_perf_counter_read(struct perf_counter *counter)
{
	context_switches_perf_counter_update(counter);
}

1703
static int context_switches_perf_counter_enable(struct perf_counter *counter)
1704
{
1705 1706 1707
	if (counter->prev_state <= PERF_COUNTER_STATE_OFF)
		atomic64_set(&counter->hw.prev_count,
			     get_context_switches(counter));
1708
	return 0;
1709 1710 1711 1712 1713 1714 1715 1716
}

static void context_switches_perf_counter_disable(struct perf_counter *counter)
{
	context_switches_perf_counter_update(counter);
}

static const struct hw_perf_counter_ops perf_ops_context_switches = {
Ingo Molnar's avatar
Ingo Molnar committed
1717 1718 1719
	.enable		= context_switches_perf_counter_enable,
	.disable	= context_switches_perf_counter_disable,
	.read		= context_switches_perf_counter_read,
1720 1721
};

1722 1723 1724 1725
/*
 * Software counter: cpu migrations
 */

1726
static inline u64 get_cpu_migrations(struct perf_counter *counter)
1727
{
1728 1729 1730 1731 1732
	struct task_struct *curr = counter->ctx->task;

	if (curr)
		return curr->se.nr_migrations;
	return cpu_nr_migrations(smp_processor_id());
1733 1734 1735 1736 1737 1738 1739 1740
}

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

	prev = atomic64_read(&counter->hw.prev_count);
1741
	now = get_cpu_migrations(counter);
1742 1743 1744 1745 1746 1747 1748 1749 1750 1751 1752 1753 1754

	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);
}

1755
static int cpu_migrations_perf_counter_enable(struct perf_counter *counter)
1756
{
1757 1758 1759
	if (counter->prev_state <= PERF_COUNTER_STATE_OFF)
		atomic64_set(&counter->hw.prev_count,
			     get_cpu_migrations(counter));
1760
	return 0;
1761 1762 1763 1764 1765 1766 1767 1768
}

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 = {
Ingo Molnar's avatar
Ingo Molnar committed
1769 1770 1771
	.enable		= cpu_migrations_perf_counter_enable,
	.disable	= cpu_migrations_perf_counter_disable,
	.read		= cpu_migrations_perf_counter_read,
1772 1773
};

1774 1775 1776
static const struct hw_perf_counter_ops *
sw_perf_counter_init(struct perf_counter *counter)
{
1777
	struct perf_counter_hw_event *hw_event = &counter->hw_event;
1778
	const struct hw_perf_counter_ops *hw_ops = NULL;
1779
	struct hw_perf_counter *hwc = &counter->hw;
1780

1781 1782 1783 1784 1785 1786 1787
	/*
	 * 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.
	 */
1788 1789
	switch (counter->hw_event.type) {
	case PERF_COUNT_CPU_CLOCK:
1790 1791 1792 1793
		hw_ops = &perf_ops_cpu_clock;

		if (hw_event->irq_period && hw_event->irq_period < 10000)
			hw_event->irq_period = 10000;
1794
		break;
1795
	case PERF_COUNT_TASK_CLOCK:
1796 1797 1798 1799 1800 1801 1802 1803
		/*
		 * 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;
1804 1805 1806

		if (hw_event->irq_period && hw_event->irq_period < 10000)
			hw_event->irq_period = 10000;
1807
		break;
1808
	case PERF_COUNT_PAGE_FAULTS:
1809 1810 1811
	case PERF_COUNT_PAGE_FAULTS_MIN:
	case PERF_COUNT_PAGE_FAULTS_MAJ:
		hw_ops = &perf_ops_generic;
1812
		break;
1813
	case PERF_COUNT_CONTEXT_SWITCHES:
1814 1815
		if (!counter->hw_event.exclude_kernel)
			hw_ops = &perf_ops_context_switches;
1816
		break;
1817
	case PERF_COUNT_CPU_MIGRATIONS:
1818 1819
		if (!counter->hw_event.exclude_kernel)
			hw_ops = &perf_ops_cpu_migrations;
1820
		break;
1821 1822 1823
	default:
		break;
	}
1824 1825 1826 1827

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

1828 1829 1830
	return hw_ops;
}

1831 1832 1833 1834
/*
 * Allocate and initialize a counter structure
 */
static struct perf_counter *
1835 1836
perf_counter_alloc(struct perf_counter_hw_event *hw_event,
		   int cpu,
1837
		   struct perf_counter_context *ctx,
1838 1839
		   struct perf_counter *group_leader,
		   gfp_t gfpflags)
1840
{
1841
	const struct hw_perf_counter_ops *hw_ops;
Ingo Molnar's avatar
Ingo Molnar committed
1842
	struct perf_counter *counter;
1843

1844
	counter = kzalloc(sizeof(*counter), gfpflags);
1845 1846 1847
	if (!counter)
		return NULL;

1848 1849 1850 1851 1852 1853 1854
	/*
	 * Single counters are their own group leaders, with an
	 * empty sibling list:
	 */
	if (!group_leader)
		group_leader = counter;

1855
	mutex_init(&counter->mutex);
1856
	INIT_LIST_HEAD(&counter->list_entry);
1857
	INIT_LIST_HEAD(&counter->event_entry);
1858
	INIT_LIST_HEAD(&counter->sibling_list);
1859 1860
	init_waitqueue_head(&counter->waitq);

1861 1862
	INIT_LIST_HEAD(&counter->child_list);

1863 1864 1865 1866 1867
	counter->irqdata		= &counter->data[0];
	counter->usrdata		= &counter->data[1];
	counter->cpu			= cpu;
	counter->hw_event		= *hw_event;
	counter->wakeup_pending		= 0;
1868
	counter->group_leader		= group_leader;
Ingo Molnar's avatar
Ingo Molnar committed
1869
	counter->hw_ops			= NULL;
1870
	counter->ctx			= ctx;
Ingo Molnar's avatar
Ingo Molnar committed
1871

1872
	counter->state = PERF_COUNTER_STATE_INACTIVE;
1873 1874 1875
	if (hw_event->disabled)
		counter->state = PERF_COUNTER_STATE_OFF;

1876 1877 1878
	hw_ops = NULL;
	if (!hw_event->raw && hw_event->type < 0)
		hw_ops = sw_perf_counter_init(counter);
1879
	else
1880 1881
		hw_ops = hw_perf_counter_init(counter);

Ingo Molnar's avatar
Ingo Molnar committed
1882 1883 1884 1885 1886
	if (!hw_ops) {
		kfree(counter);
		return NULL;
	}
	counter->hw_ops = hw_ops;
1887 1888 1889 1890 1891

	return counter;
}

/**
1892
 * sys_perf_counter_open - open a performance counter, associate it to a task/cpu
1893 1894
 *
 * @hw_event_uptr:	event type attributes for monitoring/sampling
1895
 * @pid:		target pid
1896 1897
 * @cpu:		target cpu
 * @group_fd:		group leader counter fd
1898
 */
1899
SYSCALL_DEFINE5(perf_counter_open,
1900
		const struct perf_counter_hw_event __user *, hw_event_uptr,
1901
		pid_t, pid, int, cpu, int, group_fd, unsigned long, flags)
1902
{
1903
	struct perf_counter *counter, *group_leader;
1904
	struct perf_counter_hw_event hw_event;
1905
	struct perf_counter_context *ctx;
1906
	struct file *counter_file = NULL;
1907 1908
	struct file *group_file = NULL;
	int fput_needed = 0;
1909
	int fput_needed2 = 0;
1910 1911
	int ret;

1912 1913 1914 1915
	/* for future expandability... */
	if (flags)
		return -EINVAL;

1916
	if (copy_from_user(&hw_event, hw_event_uptr, sizeof(hw_event)) != 0)
1917 1918
		return -EFAULT;

1919
	/*
1920 1921 1922 1923 1924 1925 1926 1927
	 * 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):
1928 1929 1930 1931 1932 1933
	 */
	group_leader = NULL;
	if (group_fd != -1) {
		ret = -EINVAL;
		group_file = fget_light(group_fd, &fput_needed);
		if (!group_file)
1934
			goto err_put_context;
1935
		if (group_file->f_op != &perf_fops)
1936
			goto err_put_context;
1937 1938 1939

		group_leader = group_file->private_data;
		/*
1940 1941 1942 1943 1944 1945 1946 1947
		 * 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:
1948
		 */
1949 1950
		if (group_leader->ctx != ctx)
			goto err_put_context;
1951 1952 1953 1954 1955
		/*
		 * Only a group leader can be exclusive or pinned
		 */
		if (hw_event.exclusive || hw_event.pinned)
			goto err_put_context;
1956 1957
	}

1958
	ret = -EINVAL;
1959 1960
	counter = perf_counter_alloc(&hw_event, cpu, ctx, group_leader,
				     GFP_KERNEL);
1961 1962 1963 1964 1965
	if (!counter)
		goto err_put_context;

	ret = anon_inode_getfd("[perf_counter]", &perf_fops, counter, 0);
	if (ret < 0)
1966 1967 1968 1969 1970 1971 1972
		goto err_free_put_context;

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

	counter->filp = counter_file;
1973
	mutex_lock(&ctx->mutex);
1974
	perf_install_in_context(ctx, counter, cpu);
1975
	mutex_unlock(&ctx->mutex);
1976 1977

	fput_light(counter_file, fput_needed2);
1978

1979 1980 1981
out_fput:
	fput_light(group_file, fput_needed);

1982 1983
	return ret;

1984
err_free_put_context:
1985 1986 1987 1988 1989
	kfree(counter);

err_put_context:
	put_context(ctx);

1990
	goto out_fput;
1991 1992
}

1993 1994 1995 1996 1997 1998 1999 2000 2001
/*
 * 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);
2002
	mutex_init(&ctx->mutex);
2003
	INIT_LIST_HEAD(&ctx->counter_list);
2004
	INIT_LIST_HEAD(&ctx->event_list);
2005 2006 2007 2008 2009 2010
	ctx->task = task;
}

/*
 * inherit a counter from parent task to child task:
 */
2011
static struct perf_counter *
2012 2013 2014 2015
inherit_counter(struct perf_counter *parent_counter,
	      struct task_struct *parent,
	      struct perf_counter_context *parent_ctx,
	      struct task_struct *child,
2016
	      struct perf_counter *group_leader,
2017 2018 2019 2020
	      struct perf_counter_context *child_ctx)
{
	struct perf_counter *child_counter;

2021 2022 2023 2024 2025 2026 2027 2028 2029
	/*
	 * 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;

2030
	child_counter = perf_counter_alloc(&parent_counter->hw_event,
2031 2032
					   parent_counter->cpu, child_ctx,
					   group_leader, GFP_KERNEL);
2033
	if (!child_counter)
2034
		return NULL;
2035 2036 2037 2038 2039 2040 2041 2042 2043 2044 2045 2046 2047 2048 2049 2050 2051 2052 2053 2054 2055 2056

	/*
	 * Link it up in the child's context:
	 */
	child_counter->task = child;
	list_add_counter(child_counter, child_ctx);
	child_ctx->nr_counters++;

	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);

2057 2058 2059 2060 2061 2062 2063 2064 2065 2066 2067 2068 2069 2070 2071 2072 2073 2074 2075 2076 2077 2078 2079 2080 2081 2082 2083 2084 2085 2086 2087 2088 2089 2090 2091 2092 2093 2094 2095
	/*
	 * 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;
	}
2096 2097 2098
	return 0;
}

2099 2100 2101 2102 2103 2104 2105 2106 2107 2108 2109 2110 2111 2112 2113 2114 2115 2116 2117 2118 2119 2120 2121 2122 2123 2124 2125
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);

	/*
	 * 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);
}

2126 2127 2128 2129 2130 2131
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;
2132
	struct perf_counter *sub, *tmp;
2133 2134

	/*
2135 2136 2137 2138 2139 2140
	 * 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)
2141
	 */
2142 2143 2144 2145
	if (child != current) {
		wait_task_inactive(child, 0);
		list_del_init(&child_counter->list_entry);
	} else {
2146
		struct perf_cpu_context *cpuctx;
2147 2148 2149 2150 2151 2152 2153 2154 2155 2156 2157
		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();
2158 2159 2160

		cpuctx = &__get_cpu_var(perf_cpu_context);

2161
		group_sched_out(child_counter, cpuctx, child_ctx);
2162

2163
		list_del_init(&child_counter->list_entry);
2164

2165
		child_ctx->nr_counters--;
2166

2167 2168 2169
		hw_perf_restore(perf_flags);
		curr_rq_unlock_irq_restore(&flags);
	}
2170 2171 2172 2173 2174 2175 2176

	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.
	 */
2177 2178 2179 2180
	if (parent_counter) {
		sync_child_counter(child_counter, parent_counter);
		list_for_each_entry_safe(sub, tmp, &child_counter->sibling_list,
					 list_entry) {
2181
			if (sub->parent) {
2182
				sync_child_counter(sub, sub->parent);
2183 2184
				kfree(sub);
			}
2185
		}
2186
		kfree(child_counter);
2187
	}
2188 2189 2190
}

/*
2191
 * When a child task exits, feed back counter values to parent counters.
2192
 *
2193
 * Note: we may be running in child context, but the PID is not hashed
2194 2195 2196 2197 2198 2199 2200 2201 2202 2203 2204 2205 2206 2207 2208 2209 2210 2211 2212 2213 2214 2215 2216
 * 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;
2217
	struct perf_counter *counter;
2218 2219 2220 2221 2222 2223 2224 2225 2226 2227 2228 2229 2230 2231 2232 2233 2234 2235 2236
	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.
	 */
2237
	mutex_lock(&parent_ctx->mutex);
2238 2239 2240 2241 2242 2243

	/*
	 * 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) {
2244
		if (!counter->hw_event.inherit)
2245 2246
			continue;

2247
		if (inherit_group(counter, parent,
2248 2249 2250 2251
				  parent_ctx, child, child_ctx))
			break;
	}

2252
	mutex_unlock(&parent_ctx->mutex);
2253 2254
}

2255
static void __cpuinit perf_counter_init_cpu(int cpu)
2256
{
2257
	struct perf_cpu_context *cpuctx;
2258

2259 2260
	cpuctx = &per_cpu(perf_cpu_context, cpu);
	__perf_counter_init_context(&cpuctx->ctx, NULL);
2261 2262

	mutex_lock(&perf_resource_mutex);
2263
	cpuctx->max_pertask = perf_max_counters - perf_reserved_percpu;
2264
	mutex_unlock(&perf_resource_mutex);
2265

2266
	hw_perf_counter_setup(cpu);
2267 2268 2269
}

#ifdef CONFIG_HOTPLUG_CPU
2270
static void __perf_counter_exit_cpu(void *info)
2271 2272 2273 2274 2275
{
	struct perf_cpu_context *cpuctx = &__get_cpu_var(perf_cpu_context);
	struct perf_counter_context *ctx = &cpuctx->ctx;
	struct perf_counter *counter, *tmp;

2276 2277
	list_for_each_entry_safe(counter, tmp, &ctx->counter_list, list_entry)
		__perf_counter_remove_from_context(counter);
2278
}
2279
static void perf_counter_exit_cpu(int cpu)
2280
{
2281 2282 2283 2284
	struct perf_cpu_context *cpuctx = &per_cpu(perf_cpu_context, cpu);
	struct perf_counter_context *ctx = &cpuctx->ctx;

	mutex_lock(&ctx->mutex);
2285
	smp_call_function_single(cpu, __perf_counter_exit_cpu, NULL, 1);
2286
	mutex_unlock(&ctx->mutex);
2287 2288
}
#else
2289
static inline void perf_counter_exit_cpu(int cpu) { }
2290 2291 2292 2293 2294 2295 2296 2297 2298 2299 2300
#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:
2301
		perf_counter_init_cpu(cpu);
2302 2303 2304 2305
		break;

	case CPU_DOWN_PREPARE:
	case CPU_DOWN_PREPARE_FROZEN:
2306
		perf_counter_exit_cpu(cpu);
2307 2308 2309 2310 2311 2312 2313 2314 2315 2316 2317 2318 2319 2320 2321 2322 2323 2324 2325 2326 2327 2328 2329 2330 2331 2332 2333 2334 2335 2336 2337 2338 2339 2340 2341 2342 2343 2344 2345 2346 2347 2348 2349 2350 2351 2352 2353 2354 2355 2356 2357 2358 2359 2360 2361 2362 2363 2364 2365 2366 2367 2368 2369 2370 2371 2372 2373 2374 2375 2376 2377 2378 2379 2380 2381 2382 2383 2384 2385 2386 2387 2388 2389 2390 2391 2392 2393 2394 2395 2396 2397 2398 2399 2400 2401 2402 2403 2404 2405 2406 2407 2408 2409 2410 2411 2412 2413 2414 2415 2416 2417 2418 2419
		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);