perf_event.c 42.6 KB
Newer Older
I
Ingo Molnar 已提交
1
/*
2
 * Performance events x86 architecture code
I
Ingo Molnar 已提交
3
 *
4 5 6 7 8
 *  Copyright (C) 2008 Thomas Gleixner <tglx@linutronix.de>
 *  Copyright (C) 2008-2009 Red Hat, Inc., Ingo Molnar
 *  Copyright (C) 2009 Jaswinder Singh Rajput
 *  Copyright (C) 2009 Advanced Micro Devices, Inc., Robert Richter
 *  Copyright (C) 2008-2009 Red Hat, Inc., Peter Zijlstra <pzijlstr@redhat.com>
9
 *  Copyright (C) 2009 Intel Corporation, <markus.t.metzger@intel.com>
10
 *  Copyright (C) 2009 Google, Inc., Stephane Eranian
I
Ingo Molnar 已提交
11 12 13 14
 *
 *  For licencing details see kernel-base/COPYING
 */

15
#include <linux/perf_event.h>
I
Ingo Molnar 已提交
16 17 18 19
#include <linux/capability.h>
#include <linux/notifier.h>
#include <linux/hardirq.h>
#include <linux/kprobes.h>
20
#include <linux/module.h>
I
Ingo Molnar 已提交
21 22
#include <linux/kdebug.h>
#include <linux/sched.h>
23
#include <linux/uaccess.h>
24
#include <linux/slab.h>
25
#include <linux/highmem.h>
26
#include <linux/cpu.h>
27
#include <linux/bitops.h>
I
Ingo Molnar 已提交
28 29

#include <asm/apic.h>
30
#include <asm/stacktrace.h>
P
Peter Zijlstra 已提交
31
#include <asm/nmi.h>
32
#include <asm/compat.h>
33
#include <asm/smp.h>
I
Ingo Molnar 已提交
34

35 36 37 38 39 40 41 42 43 44 45
#if 0
#undef wrmsrl
#define wrmsrl(msr, val) 					\
do {								\
	trace_printk("wrmsrl(%lx, %lx)\n", (unsigned long)(msr),\
			(unsigned long)(val));			\
	native_write_msr((msr), (u32)((u64)(val)), 		\
			(u32)((u64)(val) >> 32));		\
} while (0)
#endif

46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
/*
 * best effort, GUP based copy_from_user() that assumes IRQ or NMI context
 */
static unsigned long
copy_from_user_nmi(void *to, const void __user *from, unsigned long n)
{
	unsigned long offset, addr = (unsigned long)from;
	unsigned long size, len = 0;
	struct page *page;
	void *map;
	int ret;

	do {
		ret = __get_user_pages_fast(addr, 1, 0, &page);
		if (!ret)
			break;

		offset = addr & (PAGE_SIZE - 1);
		size = min(PAGE_SIZE - offset, n - len);

66
		map = kmap_atomic(page);
67
		memcpy(to, map+offset, size);
68
		kunmap_atomic(map);
69 70 71 72 73 74 75 76 77 78 79
		put_page(page);

		len  += size;
		to   += size;
		addr += size;

	} while (len < n);

	return len;
}

80
struct event_constraint {
81 82
	union {
		unsigned long	idxmsk[BITS_TO_LONGS(X86_PMC_IDX_MAX)];
83
		u64		idxmsk64;
84
	};
85 86
	u64	code;
	u64	cmask;
87
	int	weight;
88 89
};

90 91 92 93 94 95 96
struct amd_nb {
	int nb_id;  /* NorthBridge id */
	int refcnt; /* reference count */
	struct perf_event *owners[X86_PMC_IDX_MAX];
	struct event_constraint event_constraints[X86_PMC_IDX_MAX];
};

97 98
struct intel_percore;

99 100
#define MAX_LBR_ENTRIES		16

101
struct cpu_hw_events {
102 103 104
	/*
	 * Generic x86 PMC bits
	 */
105
	struct perf_event	*events[X86_PMC_IDX_MAX]; /* in counter order */
106
	unsigned long		active_mask[BITS_TO_LONGS(X86_PMC_IDX_MAX)];
107
	unsigned long		running[BITS_TO_LONGS(X86_PMC_IDX_MAX)];
108
	int			enabled;
I
Ingo Molnar 已提交
109

110 111
	int			n_events;
	int			n_added;
112
	int			n_txn;
113
	int			assign[X86_PMC_IDX_MAX]; /* event to counter assignment */
114
	u64			tags[X86_PMC_IDX_MAX];
115
	struct perf_event	*event_list[X86_PMC_IDX_MAX]; /* in enabled order */
116

117 118
	unsigned int		group_flag;

119 120 121 122 123 124
	/*
	 * Intel DebugStore bits
	 */
	struct debug_store	*ds;
	u64			pebs_enabled;

125 126 127 128 129 130 131 132
	/*
	 * Intel LBR bits
	 */
	int				lbr_users;
	void				*lbr_context;
	struct perf_branch_stack	lbr_stack;
	struct perf_branch_entry	lbr_entries[MAX_LBR_ENTRIES];

133 134 135 136 137 138 139
	/*
	 * Intel percore register state.
	 * Coordinate shared resources between HT threads.
	 */
	int				percore_used; /* Used by this CPU? */
	struct intel_percore		*per_core;

140 141 142
	/*
	 * AMD specific bits
	 */
143
	struct amd_nb		*amd_nb;
144 145
};

146
#define __EVENT_CONSTRAINT(c, n, m, w) {\
147
	{ .idxmsk64 = (n) },		\
148 149
	.code = (c),			\
	.cmask = (m),			\
150
	.weight = (w),			\
151
}
152

153 154 155
#define EVENT_CONSTRAINT(c, n, m)	\
	__EVENT_CONSTRAINT(c, n, m, HWEIGHT(n))

156 157 158
/*
 * Constraint on the Event code.
 */
159
#define INTEL_EVENT_CONSTRAINT(c, n)	\
160
	EVENT_CONSTRAINT(c, n, ARCH_PERFMON_EVENTSEL_EVENT)
161

162 163
/*
 * Constraint on the Event code + UMask + fixed-mask
164 165 166 167 168 169 170 171
 *
 * filter mask to validate fixed counter events.
 * the following filters disqualify for fixed counters:
 *  - inv
 *  - edge
 *  - cnt-mask
 *  The other filters are supported by fixed counters.
 *  The any-thread option is supported starting with v3.
172
 */
173
#define FIXED_EVENT_CONSTRAINT(c, n)	\
174
	EVENT_CONSTRAINT(c, (1ULL << (32+n)), X86_RAW_EVENT_MASK)
175

176 177 178
/*
 * Constraint on the Event code + UMask
 */
179
#define INTEL_UEVENT_CONSTRAINT(c, n)	\
180 181
	EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK)

182 183 184 185
#define EVENT_CONSTRAINT_END		\
	EVENT_CONSTRAINT(0, 0, 0)

#define for_each_event_constraint(e, c)	\
186
	for ((e) = (c); (e)->weight; (e)++)
187

188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209
/*
 * Extra registers for specific events.
 * Some events need large masks and require external MSRs.
 * Define a mapping to these extra registers.
 */
struct extra_reg {
	unsigned int		event;
	unsigned int		msr;
	u64			config_mask;
	u64			valid_mask;
};

#define EVENT_EXTRA_REG(e, ms, m, vm) {	\
	.event = (e),		\
	.msr = (ms),		\
	.config_mask = (m),	\
	.valid_mask = (vm),	\
	}
#define INTEL_EVENT_EXTRA_REG(event, msr, vm)	\
	EVENT_EXTRA_REG(event, msr, ARCH_PERFMON_EVENTSEL_EVENT, vm)
#define EVENT_EXTRA_END EVENT_EXTRA_REG(0, 0, 0, 0)

210 211 212 213 214 215 216 217 218 219 220
union perf_capabilities {
	struct {
		u64	lbr_format    : 6;
		u64	pebs_trap     : 1;
		u64	pebs_arch_reg : 1;
		u64	pebs_format   : 4;
		u64	smm_freeze    : 1;
	};
	u64	capabilities;
};

I
Ingo Molnar 已提交
221
/*
222
 * struct x86_pmu - generic x86 pmu
I
Ingo Molnar 已提交
223
 */
224
struct x86_pmu {
225 226 227
	/*
	 * Generic x86 PMC bits
	 */
228 229
	const char	*name;
	int		version;
230
	int		(*handle_irq)(struct pt_regs *);
231
	void		(*disable_all)(void);
232
	void		(*enable_all)(int added);
233 234
	void		(*enable)(struct perf_event *);
	void		(*disable)(struct perf_event *);
235
	int		(*hw_config)(struct perf_event *event);
236
	int		(*schedule_events)(struct cpu_hw_events *cpuc, int n, int *assign);
237 238
	unsigned	eventsel;
	unsigned	perfctr;
239
	u64		(*event_map)(int);
240
	int		max_events;
241 242 243 244
	int		num_counters;
	int		num_counters_fixed;
	int		cntval_bits;
	u64		cntval_mask;
245
	int		apic;
246
	u64		max_period;
247 248 249 250
	struct event_constraint *
			(*get_event_constraints)(struct cpu_hw_events *cpuc,
						 struct perf_event *event);

251 252
	void		(*put_event_constraints)(struct cpu_hw_events *cpuc,
						 struct perf_event *event);
253
	struct event_constraint *event_constraints;
254
	struct event_constraint *percore_constraints;
255
	void		(*quirks)(void);
256
	int		perfctr_second_write;
257

258
	int		(*cpu_prepare)(int cpu);
259 260 261
	void		(*cpu_starting)(int cpu);
	void		(*cpu_dying)(int cpu);
	void		(*cpu_dead)(int cpu);
262 263 264 265

	/*
	 * Intel Arch Perfmon v2+
	 */
266 267
	u64			intel_ctrl;
	union perf_capabilities intel_cap;
268 269 270 271 272

	/*
	 * Intel DebugStore bits
	 */
	int		bts, pebs;
273
	int		bts_active, pebs_active;
274 275 276
	int		pebs_record_size;
	void		(*drain_pebs)(struct pt_regs *regs);
	struct event_constraint *pebs_constraints;
277 278 279 280 281 282

	/*
	 * Intel LBR
	 */
	unsigned long	lbr_tos, lbr_from, lbr_to; /* MSR base regs       */
	int		lbr_nr;			   /* hardware stack size */
283 284 285 286 287

	/*
	 * Extra registers for events
	 */
	struct extra_reg *extra_regs;
288 289
};

290
static struct x86_pmu x86_pmu __read_mostly;
291

292
static DEFINE_PER_CPU(struct cpu_hw_events, cpu_hw_events) = {
293 294
	.enabled = 1,
};
I
Ingo Molnar 已提交
295

296
static int x86_perf_event_set_period(struct perf_event *event);
297

298
/*
299
 * Generalized hw caching related hw_event table, filled
300
 * in on a per model basis. A value of 0 means
301 302
 * 'not supported', -1 means 'hw_event makes no sense on
 * this CPU', any other value means the raw hw_event
303 304 305 306 307 308 309 310 311
 * ID.
 */

#define C(x) PERF_COUNT_HW_CACHE_##x

static u64 __read_mostly hw_cache_event_ids
				[PERF_COUNT_HW_CACHE_MAX]
				[PERF_COUNT_HW_CACHE_OP_MAX]
				[PERF_COUNT_HW_CACHE_RESULT_MAX];
312 313 314 315
static u64 __read_mostly hw_cache_extra_regs
				[PERF_COUNT_HW_CACHE_MAX]
				[PERF_COUNT_HW_CACHE_OP_MAX]
				[PERF_COUNT_HW_CACHE_RESULT_MAX];
316

317
/*
318 319
 * Propagate event elapsed time into the generic event.
 * Can only be executed on the CPU where the event is active.
320 321
 * Returns the delta events processed.
 */
322
static u64
323
x86_perf_event_update(struct perf_event *event)
324
{
325
	struct hw_perf_event *hwc = &event->hw;
326
	int shift = 64 - x86_pmu.cntval_bits;
327
	u64 prev_raw_count, new_raw_count;
328
	int idx = hwc->idx;
329
	s64 delta;
330

331 332 333
	if (idx == X86_PMC_IDX_FIXED_BTS)
		return 0;

334
	/*
335
	 * Careful: an NMI might modify the previous event value.
336 337 338
	 *
	 * Our tactic to handle this is to first atomically read and
	 * exchange a new raw count - then add that new-prev delta
339
	 * count to the generic event atomically:
340 341
	 */
again:
342
	prev_raw_count = local64_read(&hwc->prev_count);
343
	rdmsrl(hwc->event_base, new_raw_count);
344

345
	if (local64_cmpxchg(&hwc->prev_count, prev_raw_count,
346 347 348 349 350 351
					new_raw_count) != prev_raw_count)
		goto again;

	/*
	 * Now we have the new raw value and have updated the prev
	 * timestamp already. We can now calculate the elapsed delta
352
	 * (event-)time and add that to the generic event.
353 354
	 *
	 * Careful, not all hw sign-extends above the physical width
355
	 * of the count.
356
	 */
357 358
	delta = (new_raw_count << shift) - (prev_raw_count << shift);
	delta >>= shift;
359

360 361
	local64_add(delta, &event->count);
	local64_sub(delta, &hwc->period_left);
362 363

	return new_raw_count;
364 365
}

366 367 368 369 370 371 372 373
/* using X86_FEATURE_PERFCTR_CORE to later implement ALTERNATIVE() here */
static inline int x86_pmu_addr_offset(int index)
{
	if (boot_cpu_has(X86_FEATURE_PERFCTR_CORE))
		return index << 1;
	return index;
}

374 375
static inline unsigned int x86_pmu_config_addr(int index)
{
376
	return x86_pmu.eventsel + x86_pmu_addr_offset(index);
377 378 379 380
}

static inline unsigned int x86_pmu_event_addr(int index)
{
381
	return x86_pmu.perfctr + x86_pmu_addr_offset(index);
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
/*
 * Find and validate any extra registers to set up.
 */
static int x86_pmu_extra_regs(u64 config, struct perf_event *event)
{
	struct extra_reg *er;

	event->hw.extra_reg = 0;
	event->hw.extra_config = 0;

	if (!x86_pmu.extra_regs)
		return 0;

	for (er = x86_pmu.extra_regs; er->msr; er++) {
		if (er->event != (config & er->config_mask))
			continue;
		if (event->attr.config1 & ~er->valid_mask)
			return -EINVAL;
		event->hw.extra_reg = er->msr;
		event->hw.extra_config = event->attr.config1;
		break;
	}
	return 0;
}

409
static atomic_t active_events;
P
Peter Zijlstra 已提交
410 411
static DEFINE_MUTEX(pmc_reserve_mutex);

412 413
#ifdef CONFIG_X86_LOCAL_APIC

P
Peter Zijlstra 已提交
414 415 416 417
static bool reserve_pmc_hardware(void)
{
	int i;

418
	for (i = 0; i < x86_pmu.num_counters; i++) {
419
		if (!reserve_perfctr_nmi(x86_pmu_event_addr(i)))
P
Peter Zijlstra 已提交
420 421 422
			goto perfctr_fail;
	}

423
	for (i = 0; i < x86_pmu.num_counters; i++) {
424
		if (!reserve_evntsel_nmi(x86_pmu_config_addr(i)))
P
Peter Zijlstra 已提交
425 426 427 428 429 430 431
			goto eventsel_fail;
	}

	return true;

eventsel_fail:
	for (i--; i >= 0; i--)
432
		release_evntsel_nmi(x86_pmu_config_addr(i));
P
Peter Zijlstra 已提交
433

434
	i = x86_pmu.num_counters;
P
Peter Zijlstra 已提交
435 436 437

perfctr_fail:
	for (i--; i >= 0; i--)
438
		release_perfctr_nmi(x86_pmu_event_addr(i));
P
Peter Zijlstra 已提交
439 440 441 442 443 444 445 446

	return false;
}

static void release_pmc_hardware(void)
{
	int i;

447
	for (i = 0; i < x86_pmu.num_counters; i++) {
448 449
		release_perfctr_nmi(x86_pmu_event_addr(i));
		release_evntsel_nmi(x86_pmu_config_addr(i));
P
Peter Zijlstra 已提交
450 451 452
	}
}

453 454 455 456 457 458 459
#else

static bool reserve_pmc_hardware(void) { return true; }
static void release_pmc_hardware(void) {}

#endif

460 461 462
static bool check_hw_exists(void)
{
	u64 val, val_new = 0;
463
	int i, reg, ret = 0;
464

465 466 467 468 469
	/*
	 * Check to see if the BIOS enabled any of the counters, if so
	 * complain and bail.
	 */
	for (i = 0; i < x86_pmu.num_counters; i++) {
470
		reg = x86_pmu_config_addr(i);
471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493
		ret = rdmsrl_safe(reg, &val);
		if (ret)
			goto msr_fail;
		if (val & ARCH_PERFMON_EVENTSEL_ENABLE)
			goto bios_fail;
	}

	if (x86_pmu.num_counters_fixed) {
		reg = MSR_ARCH_PERFMON_FIXED_CTR_CTRL;
		ret = rdmsrl_safe(reg, &val);
		if (ret)
			goto msr_fail;
		for (i = 0; i < x86_pmu.num_counters_fixed; i++) {
			if (val & (0x03 << i*4))
				goto bios_fail;
		}
	}

	/*
	 * Now write a value and read it back to see if it matches,
	 * this is needed to detect certain hardware emulators (qemu/kvm)
	 * that don't trap on the MSR access and always return 0s.
	 */
494
	val = 0xabcdUL;
495 496
	ret = checking_wrmsrl(x86_pmu_event_addr(0), val);
	ret |= rdmsrl_safe(x86_pmu_event_addr(0), &val_new);
497
	if (ret || val != val_new)
498
		goto msr_fail;
499 500

	return true;
501 502

bios_fail:
503 504 505 506
	/*
	 * We still allow the PMU driver to operate:
	 */
	printk(KERN_CONT "Broken BIOS detected, complain to your hardware vendor.\n");
507
	printk(KERN_ERR FW_BUG "the BIOS has corrupted hw-PMU resources (MSR %x is %Lx)\n", reg, val);
508 509

	return true;
510 511 512

msr_fail:
	printk(KERN_CONT "Broken PMU hardware detected, using software events only.\n");
513

514
	return false;
515 516
}

517
static void reserve_ds_buffers(void);
518
static void release_ds_buffers(void);
519

520
static void hw_perf_event_destroy(struct perf_event *event)
P
Peter Zijlstra 已提交
521
{
522
	if (atomic_dec_and_mutex_lock(&active_events, &pmc_reserve_mutex)) {
P
Peter Zijlstra 已提交
523
		release_pmc_hardware();
524
		release_ds_buffers();
P
Peter Zijlstra 已提交
525 526 527 528
		mutex_unlock(&pmc_reserve_mutex);
	}
}

529 530 531 532 533
static inline int x86_pmu_initialized(void)
{
	return x86_pmu.handle_irq != NULL;
}

534
static inline int
535
set_ext_hw_attr(struct hw_perf_event *hwc, struct perf_event *event)
536
{
537
	struct perf_event_attr *attr = &event->attr;
538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563
	unsigned int cache_type, cache_op, cache_result;
	u64 config, val;

	config = attr->config;

	cache_type = (config >>  0) & 0xff;
	if (cache_type >= PERF_COUNT_HW_CACHE_MAX)
		return -EINVAL;

	cache_op = (config >>  8) & 0xff;
	if (cache_op >= PERF_COUNT_HW_CACHE_OP_MAX)
		return -EINVAL;

	cache_result = (config >> 16) & 0xff;
	if (cache_result >= PERF_COUNT_HW_CACHE_RESULT_MAX)
		return -EINVAL;

	val = hw_cache_event_ids[cache_type][cache_op][cache_result];

	if (val == 0)
		return -ENOENT;

	if (val == -1)
		return -EINVAL;

	hwc->config |= val;
564 565
	attr->config1 = hw_cache_extra_regs[cache_type][cache_op][cache_result];
	return x86_pmu_extra_regs(val, event);
566 567
}

568 569 570 571 572 573
static int x86_setup_perfctr(struct perf_event *event)
{
	struct perf_event_attr *attr = &event->attr;
	struct hw_perf_event *hwc = &event->hw;
	u64 config;

574
	if (!is_sampling_event(event)) {
575 576
		hwc->sample_period = x86_pmu.max_period;
		hwc->last_period = hwc->sample_period;
577
		local64_set(&hwc->period_left, hwc->sample_period);
578 579 580 581 582 583 584 585 586 587 588
	} else {
		/*
		 * If we have a PMU initialized but no APIC
		 * interrupts, we cannot sample hardware
		 * events (user-space has to fall back and
		 * sample via a hrtimer based software event):
		 */
		if (!x86_pmu.apic)
			return -EOPNOTSUPP;
	}

589 590 591 592
	/*
	 * Do not allow config1 (extended registers) to propagate,
	 * there's no sane user-space generalization yet:
	 */
593
	if (attr->type == PERF_TYPE_RAW)
594
		return 0;
595 596

	if (attr->type == PERF_TYPE_HW_CACHE)
597
		return set_ext_hw_attr(hwc, event);
598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615

	if (attr->config >= x86_pmu.max_events)
		return -EINVAL;

	/*
	 * The generic map:
	 */
	config = x86_pmu.event_map(attr->config);

	if (config == 0)
		return -ENOENT;

	if (config == -1LL)
		return -EINVAL;

	/*
	 * Branch tracing:
	 */
P
Peter Zijlstra 已提交
616 617
	if (attr->config == PERF_COUNT_HW_BRANCH_INSTRUCTIONS &&
	    !attr->freq && hwc->sample_period == 1) {
618
		/* BTS is not supported by this architecture. */
619
		if (!x86_pmu.bts_active)
620 621 622 623 624 625 626 627 628 629 630
			return -EOPNOTSUPP;

		/* BTS is currently only allowed for user-mode. */
		if (!attr->exclude_kernel)
			return -EOPNOTSUPP;
	}

	hwc->config |= config;

	return 0;
}
631

632
static int x86_pmu_hw_config(struct perf_event *event)
633
{
P
Peter Zijlstra 已提交
634 635 636 637
	if (event->attr.precise_ip) {
		int precise = 0;

		/* Support for constant skid */
638
		if (x86_pmu.pebs_active) {
P
Peter Zijlstra 已提交
639 640
			precise++;

641 642 643 644
			/* Support for IP fixup */
			if (x86_pmu.lbr_nr)
				precise++;
		}
P
Peter Zijlstra 已提交
645 646 647 648 649

		if (event->attr.precise_ip > precise)
			return -EOPNOTSUPP;
	}

650 651 652 653
	/*
	 * Generate PMC IRQs:
	 * (keep 'enabled' bit clear for now)
	 */
654
	event->hw.config = ARCH_PERFMON_EVENTSEL_INT;
655 656 657 658

	/*
	 * Count user and OS events unless requested not to
	 */
659 660 661 662
	if (!event->attr.exclude_user)
		event->hw.config |= ARCH_PERFMON_EVENTSEL_USR;
	if (!event->attr.exclude_kernel)
		event->hw.config |= ARCH_PERFMON_EVENTSEL_OS;
663

664 665
	if (event->attr.type == PERF_TYPE_RAW)
		event->hw.config |= event->attr.config & X86_RAW_EVENT_MASK;
666

667
	return x86_setup_perfctr(event);
668 669
}

I
Ingo Molnar 已提交
670
/*
671
 * Setup the hardware configuration for a given attr_type
I
Ingo Molnar 已提交
672
 */
673
static int __x86_pmu_event_init(struct perf_event *event)
I
Ingo Molnar 已提交
674
{
P
Peter Zijlstra 已提交
675
	int err;
I
Ingo Molnar 已提交
676

677 678
	if (!x86_pmu_initialized())
		return -ENODEV;
I
Ingo Molnar 已提交
679

P
Peter Zijlstra 已提交
680
	err = 0;
681
	if (!atomic_inc_not_zero(&active_events)) {
P
Peter Zijlstra 已提交
682
		mutex_lock(&pmc_reserve_mutex);
683
		if (atomic_read(&active_events) == 0) {
684 685
			if (!reserve_pmc_hardware())
				err = -EBUSY;
686 687
			else
				reserve_ds_buffers();
688 689
		}
		if (!err)
690
			atomic_inc(&active_events);
P
Peter Zijlstra 已提交
691 692 693 694 695
		mutex_unlock(&pmc_reserve_mutex);
	}
	if (err)
		return err;

696
	event->destroy = hw_perf_event_destroy;
697

698 699 700
	event->hw.idx = -1;
	event->hw.last_cpu = -1;
	event->hw.last_tag = ~0ULL;
701

702
	return x86_pmu.hw_config(event);
703 704
}

705
static void x86_pmu_disable_all(void)
706
{
707
	struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
708 709
	int idx;

710
	for (idx = 0; idx < x86_pmu.num_counters; idx++) {
711 712
		u64 val;

713
		if (!test_bit(idx, cpuc->active_mask))
714
			continue;
715
		rdmsrl(x86_pmu_config_addr(idx), val);
716
		if (!(val & ARCH_PERFMON_EVENTSEL_ENABLE))
717
			continue;
718
		val &= ~ARCH_PERFMON_EVENTSEL_ENABLE;
719
		wrmsrl(x86_pmu_config_addr(idx), val);
720 721 722
	}
}

P
Peter Zijlstra 已提交
723
static void x86_pmu_disable(struct pmu *pmu)
724
{
725 726
	struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);

727
	if (!x86_pmu_initialized())
728
		return;
729

730 731 732 733 734 735
	if (!cpuc->enabled)
		return;

	cpuc->n_added = 0;
	cpuc->enabled = 0;
	barrier();
736 737

	x86_pmu.disable_all();
738
}
I
Ingo Molnar 已提交
739

740 741 742
static inline void __x86_pmu_enable_event(struct hw_perf_event *hwc,
					  u64 enable_mask)
{
743 744
	if (hwc->extra_reg)
		wrmsrl(hwc->extra_reg, hwc->extra_config);
745
	wrmsrl(hwc->config_base, hwc->config | enable_mask);
746 747
}

748
static void x86_pmu_enable_all(int added)
749
{
750
	struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
751 752
	int idx;

753
	for (idx = 0; idx < x86_pmu.num_counters; idx++) {
754
		struct hw_perf_event *hwc = &cpuc->events[idx]->hw;
755

756
		if (!test_bit(idx, cpuc->active_mask))
757
			continue;
758

759
		__x86_pmu_enable_event(hwc, ARCH_PERFMON_EVENTSEL_ENABLE);
760 761 762
	}
}

P
Peter Zijlstra 已提交
763
static struct pmu pmu;
764 765 766 767 768 769 770 771

static inline int is_x86_event(struct perf_event *event)
{
	return event->pmu == &pmu;
}

static int x86_schedule_events(struct cpu_hw_events *cpuc, int n, int *assign)
{
772
	struct event_constraint *c, *constraints[X86_PMC_IDX_MAX];
773
	unsigned long used_mask[BITS_TO_LONGS(X86_PMC_IDX_MAX)];
774
	int i, j, w, wmax, num = 0;
775 776 777 778 779
	struct hw_perf_event *hwc;

	bitmap_zero(used_mask, X86_PMC_IDX_MAX);

	for (i = 0; i < n; i++) {
780 781
		c = x86_pmu.get_event_constraints(cpuc, cpuc->event_list[i]);
		constraints[i] = c;
782 783
	}

784 785 786
	/*
	 * fastpath, try to reuse previous register
	 */
787
	for (i = 0; i < n; i++) {
788
		hwc = &cpuc->event_list[i]->hw;
789
		c = constraints[i];
790 791 792 793 794 795

		/* never assigned */
		if (hwc->idx == -1)
			break;

		/* constraint still honored */
796
		if (!test_bit(hwc->idx, c->idxmsk))
797 798 799 800 801 802
			break;

		/* not already used */
		if (test_bit(hwc->idx, used_mask))
			break;

P
Peter Zijlstra 已提交
803
		__set_bit(hwc->idx, used_mask);
804 805 806
		if (assign)
			assign[i] = hwc->idx;
	}
807
	if (i == n)
808 809 810 811 812 813 814 815
		goto done;

	/*
	 * begin slow path
	 */

	bitmap_zero(used_mask, X86_PMC_IDX_MAX);

816 817 818 819 820 821 822 823 824
	/*
	 * weight = number of possible counters
	 *
	 * 1    = most constrained, only works on one counter
	 * wmax = least constrained, works on any counter
	 *
	 * assign events to counters starting with most
	 * constrained events.
	 */
825
	wmax = x86_pmu.num_counters;
826 827 828 829 830 831

	/*
	 * when fixed event counters are present,
	 * wmax is incremented by 1 to account
	 * for one more choice
	 */
832
	if (x86_pmu.num_counters_fixed)
833 834
		wmax++;

835
	for (w = 1, num = n; num && w <= wmax; w++) {
836
		/* for each event */
837
		for (i = 0; num && i < n; i++) {
838
			c = constraints[i];
839 840
			hwc = &cpuc->event_list[i]->hw;

841
			if (c->weight != w)
842 843
				continue;

844
			for_each_set_bit(j, c->idxmsk, X86_PMC_IDX_MAX) {
845 846 847 848 849 850 851
				if (!test_bit(j, used_mask))
					break;
			}

			if (j == X86_PMC_IDX_MAX)
				break;

P
Peter Zijlstra 已提交
852
			__set_bit(j, used_mask);
853

854 855 856 857 858
			if (assign)
				assign[i] = j;
			num--;
		}
	}
859
done:
860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881
	/*
	 * scheduling failed or is just a simulation,
	 * free resources if necessary
	 */
	if (!assign || num) {
		for (i = 0; i < n; i++) {
			if (x86_pmu.put_event_constraints)
				x86_pmu.put_event_constraints(cpuc, cpuc->event_list[i]);
		}
	}
	return num ? -ENOSPC : 0;
}

/*
 * dogrp: true if must collect siblings events (group)
 * returns total number of events and error code
 */
static int collect_events(struct cpu_hw_events *cpuc, struct perf_event *leader, bool dogrp)
{
	struct perf_event *event;
	int n, max_count;

882
	max_count = x86_pmu.num_counters + x86_pmu.num_counters_fixed;
883 884 885 886 887 888 889 890 891 892 893 894 895 896 897

	/* current number of events already accepted */
	n = cpuc->n_events;

	if (is_x86_event(leader)) {
		if (n >= max_count)
			return -ENOSPC;
		cpuc->event_list[n] = leader;
		n++;
	}
	if (!dogrp)
		return n;

	list_for_each_entry(event, &leader->sibling_list, group_entry) {
		if (!is_x86_event(event) ||
898
		    event->state <= PERF_EVENT_STATE_OFF)
899 900 901 902 903 904 905 906 907 908 909 910
			continue;

		if (n >= max_count)
			return -ENOSPC;

		cpuc->event_list[n] = event;
		n++;
	}
	return n;
}

static inline void x86_assign_hw_event(struct perf_event *event,
911
				struct cpu_hw_events *cpuc, int i)
912
{
913 914 915 916 917
	struct hw_perf_event *hwc = &event->hw;

	hwc->idx = cpuc->assign[i];
	hwc->last_cpu = smp_processor_id();
	hwc->last_tag = ++cpuc->tags[i];
918 919 920 921 922 923

	if (hwc->idx == X86_PMC_IDX_FIXED_BTS) {
		hwc->config_base = 0;
		hwc->event_base	= 0;
	} else if (hwc->idx >= X86_PMC_IDX_FIXED) {
		hwc->config_base = MSR_ARCH_PERFMON_FIXED_CTR_CTRL;
924
		hwc->event_base = MSR_ARCH_PERFMON_FIXED_CTR0 + (hwc->idx - X86_PMC_IDX_FIXED);
925
	} else {
926 927
		hwc->config_base = x86_pmu_config_addr(hwc->idx);
		hwc->event_base  = x86_pmu_event_addr(hwc->idx);
928 929 930
	}
}

931 932 933 934 935 936 937 938 939
static inline int match_prev_assignment(struct hw_perf_event *hwc,
					struct cpu_hw_events *cpuc,
					int i)
{
	return hwc->idx == cpuc->assign[i] &&
		hwc->last_cpu == smp_processor_id() &&
		hwc->last_tag == cpuc->tags[i];
}

P
Peter Zijlstra 已提交
940 941
static void x86_pmu_start(struct perf_event *event, int flags);
static void x86_pmu_stop(struct perf_event *event, int flags);
942

P
Peter Zijlstra 已提交
943
static void x86_pmu_enable(struct pmu *pmu)
944
{
945 946 947
	struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
	struct perf_event *event;
	struct hw_perf_event *hwc;
948
	int i, added = cpuc->n_added;
949

950
	if (!x86_pmu_initialized())
951
		return;
952 953 954 955

	if (cpuc->enabled)
		return;

956
	if (cpuc->n_added) {
957
		int n_running = cpuc->n_events - cpuc->n_added;
958 959 960 961 962 963 964
		/*
		 * apply assignment obtained either from
		 * hw_perf_group_sched_in() or x86_pmu_enable()
		 *
		 * step1: save events moving to new counters
		 * step2: reprogram moved events into new counters
		 */
965
		for (i = 0; i < n_running; i++) {
966 967 968
			event = cpuc->event_list[i];
			hwc = &event->hw;

969 970 971 972 973 974 975 976
			/*
			 * we can avoid reprogramming counter if:
			 * - assigned same counter as last time
			 * - running on same CPU as last time
			 * - no other event has used the counter since
			 */
			if (hwc->idx == -1 ||
			    match_prev_assignment(hwc, cpuc, i))
977 978
				continue;

P
Peter Zijlstra 已提交
979 980 981 982 983 984 985 986
			/*
			 * Ensure we don't accidentally enable a stopped
			 * counter simply because we rescheduled.
			 */
			if (hwc->state & PERF_HES_STOPPED)
				hwc->state |= PERF_HES_ARCH;

			x86_pmu_stop(event, PERF_EF_UPDATE);
987 988 989 990 991 992
		}

		for (i = 0; i < cpuc->n_events; i++) {
			event = cpuc->event_list[i];
			hwc = &event->hw;

993
			if (!match_prev_assignment(hwc, cpuc, i))
994
				x86_assign_hw_event(event, cpuc, i);
995 996
			else if (i < n_running)
				continue;
997

P
Peter Zijlstra 已提交
998 999 1000 1001
			if (hwc->state & PERF_HES_ARCH)
				continue;

			x86_pmu_start(event, PERF_EF_RELOAD);
1002 1003 1004 1005
		}
		cpuc->n_added = 0;
		perf_events_lapic_init();
	}
1006 1007 1008 1009

	cpuc->enabled = 1;
	barrier();

1010
	x86_pmu.enable_all(added);
1011 1012
}

1013
static inline void x86_pmu_disable_event(struct perf_event *event)
1014
{
1015
	struct hw_perf_event *hwc = &event->hw;
1016

1017
	wrmsrl(hwc->config_base, hwc->config);
1018 1019
}

1020
static DEFINE_PER_CPU(u64 [X86_PMC_IDX_MAX], pmc_prev_left);
I
Ingo Molnar 已提交
1021

1022 1023
/*
 * Set the next IRQ period, based on the hwc->period_left value.
1024
 * To be called with the event disabled in hw:
1025
 */
1026
static int
1027
x86_perf_event_set_period(struct perf_event *event)
I
Ingo Molnar 已提交
1028
{
1029
	struct hw_perf_event *hwc = &event->hw;
1030
	s64 left = local64_read(&hwc->period_left);
1031
	s64 period = hwc->sample_period;
1032
	int ret = 0, idx = hwc->idx;
1033

1034 1035 1036
	if (idx == X86_PMC_IDX_FIXED_BTS)
		return 0;

1037
	/*
1038
	 * If we are way outside a reasonable range then just skip forward:
1039 1040 1041
	 */
	if (unlikely(left <= -period)) {
		left = period;
1042
		local64_set(&hwc->period_left, left);
1043
		hwc->last_period = period;
1044
		ret = 1;
1045 1046 1047 1048
	}

	if (unlikely(left <= 0)) {
		left += period;
1049
		local64_set(&hwc->period_left, left);
1050
		hwc->last_period = period;
1051
		ret = 1;
1052
	}
1053
	/*
1054
	 * Quirk: certain CPUs dont like it if just 1 hw_event is left:
1055 1056 1057
	 */
	if (unlikely(left < 2))
		left = 2;
I
Ingo Molnar 已提交
1058

1059 1060 1061
	if (left > x86_pmu.max_period)
		left = x86_pmu.max_period;

1062
	per_cpu(pmc_prev_left[idx], smp_processor_id()) = left;
1063 1064

	/*
1065
	 * The hw event starts counting from this event offset,
1066 1067
	 * mark it to be able to extra future deltas:
	 */
1068
	local64_set(&hwc->prev_count, (u64)-left);
1069

1070
	wrmsrl(hwc->event_base, (u64)(-left) & x86_pmu.cntval_mask);
1071 1072 1073 1074 1075 1076 1077

	/*
	 * Due to erratum on certan cpu we need
	 * a second write to be sure the register
	 * is updated properly
	 */
	if (x86_pmu.perfctr_second_write) {
1078
		wrmsrl(hwc->event_base,
1079
			(u64)(-left) & x86_pmu.cntval_mask);
1080
	}
1081

1082
	perf_event_update_userpage(event);
1083

1084
	return ret;
1085 1086
}

1087
static void x86_pmu_enable_event(struct perf_event *event)
1088
{
T
Tejun Heo 已提交
1089
	if (__this_cpu_read(cpu_hw_events.enabled))
1090 1091
		__x86_pmu_enable_event(&event->hw,
				       ARCH_PERFMON_EVENTSEL_ENABLE);
I
Ingo Molnar 已提交
1092 1093
}

1094
/*
P
Peter Zijlstra 已提交
1095
 * Add a single event to the PMU.
1096 1097 1098
 *
 * The event is added to the group of enabled events
 * but only if it can be scehduled with existing events.
1099
 */
P
Peter Zijlstra 已提交
1100
static int x86_pmu_add(struct perf_event *event, int flags)
1101 1102
{
	struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
1103 1104 1105
	struct hw_perf_event *hwc;
	int assign[X86_PMC_IDX_MAX];
	int n, n0, ret;
1106

1107
	hwc = &event->hw;
1108

P
Peter Zijlstra 已提交
1109
	perf_pmu_disable(event->pmu);
1110
	n0 = cpuc->n_events;
1111 1112 1113
	ret = n = collect_events(cpuc, event, false);
	if (ret < 0)
		goto out;
1114

P
Peter Zijlstra 已提交
1115 1116 1117 1118
	hwc->state = PERF_HES_UPTODATE | PERF_HES_STOPPED;
	if (!(flags & PERF_EF_START))
		hwc->state |= PERF_HES_ARCH;

1119 1120
	/*
	 * If group events scheduling transaction was started,
L
Lucas De Marchi 已提交
1121
	 * skip the schedulability test here, it will be performed
P
Peter Zijlstra 已提交
1122
	 * at commit time (->commit_txn) as a whole
1123
	 */
1124
	if (cpuc->group_flag & PERF_EVENT_TXN)
1125
		goto done_collect;
1126

1127
	ret = x86_pmu.schedule_events(cpuc, n, assign);
1128
	if (ret)
1129
		goto out;
1130 1131 1132 1133 1134
	/*
	 * copy new assignment, now we know it is possible
	 * will be used by hw_perf_enable()
	 */
	memcpy(cpuc->assign, assign, n*sizeof(int));
1135

1136
done_collect:
1137
	cpuc->n_events = n;
1138
	cpuc->n_added += n - n0;
1139
	cpuc->n_txn += n - n0;
1140

1141 1142
	ret = 0;
out:
P
Peter Zijlstra 已提交
1143
	perf_pmu_enable(event->pmu);
1144
	return ret;
I
Ingo Molnar 已提交
1145 1146
}

P
Peter Zijlstra 已提交
1147
static void x86_pmu_start(struct perf_event *event, int flags)
1148
{
P
Peter Zijlstra 已提交
1149 1150 1151
	struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
	int idx = event->hw.idx;

P
Peter Zijlstra 已提交
1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163
	if (WARN_ON_ONCE(!(event->hw.state & PERF_HES_STOPPED)))
		return;

	if (WARN_ON_ONCE(idx == -1))
		return;

	if (flags & PERF_EF_RELOAD) {
		WARN_ON_ONCE(!(event->hw.state & PERF_HES_UPTODATE));
		x86_perf_event_set_period(event);
	}

	event->hw.state = 0;
1164

P
Peter Zijlstra 已提交
1165 1166
	cpuc->events[idx] = event;
	__set_bit(idx, cpuc->active_mask);
1167
	__set_bit(idx, cpuc->running);
1168
	x86_pmu.enable(event);
P
Peter Zijlstra 已提交
1169
	perf_event_update_userpage(event);
1170 1171
}

1172
void perf_event_print_debug(void)
I
Ingo Molnar 已提交
1173
{
1174
	u64 ctrl, status, overflow, pmc_ctrl, pmc_count, prev_left, fixed;
1175
	u64 pebs;
1176
	struct cpu_hw_events *cpuc;
1177
	unsigned long flags;
1178 1179
	int cpu, idx;

1180
	if (!x86_pmu.num_counters)
1181
		return;
I
Ingo Molnar 已提交
1182

1183
	local_irq_save(flags);
I
Ingo Molnar 已提交
1184 1185

	cpu = smp_processor_id();
1186
	cpuc = &per_cpu(cpu_hw_events, cpu);
I
Ingo Molnar 已提交
1187

1188
	if (x86_pmu.version >= 2) {
1189 1190 1191 1192
		rdmsrl(MSR_CORE_PERF_GLOBAL_CTRL, ctrl);
		rdmsrl(MSR_CORE_PERF_GLOBAL_STATUS, status);
		rdmsrl(MSR_CORE_PERF_GLOBAL_OVF_CTRL, overflow);
		rdmsrl(MSR_ARCH_PERFMON_FIXED_CTR_CTRL, fixed);
1193
		rdmsrl(MSR_IA32_PEBS_ENABLE, pebs);
1194 1195 1196 1197 1198 1199

		pr_info("\n");
		pr_info("CPU#%d: ctrl:       %016llx\n", cpu, ctrl);
		pr_info("CPU#%d: status:     %016llx\n", cpu, status);
		pr_info("CPU#%d: overflow:   %016llx\n", cpu, overflow);
		pr_info("CPU#%d: fixed:      %016llx\n", cpu, fixed);
1200
		pr_info("CPU#%d: pebs:       %016llx\n", cpu, pebs);
1201
	}
1202
	pr_info("CPU#%d: active:     %016llx\n", cpu, *(u64 *)cpuc->active_mask);
I
Ingo Molnar 已提交
1203

1204
	for (idx = 0; idx < x86_pmu.num_counters; idx++) {
1205 1206
		rdmsrl(x86_pmu_config_addr(idx), pmc_ctrl);
		rdmsrl(x86_pmu_event_addr(idx), pmc_count);
I
Ingo Molnar 已提交
1207

1208
		prev_left = per_cpu(pmc_prev_left[idx], cpu);
I
Ingo Molnar 已提交
1209

1210
		pr_info("CPU#%d:   gen-PMC%d ctrl:  %016llx\n",
I
Ingo Molnar 已提交
1211
			cpu, idx, pmc_ctrl);
1212
		pr_info("CPU#%d:   gen-PMC%d count: %016llx\n",
I
Ingo Molnar 已提交
1213
			cpu, idx, pmc_count);
1214
		pr_info("CPU#%d:   gen-PMC%d left:  %016llx\n",
1215
			cpu, idx, prev_left);
I
Ingo Molnar 已提交
1216
	}
1217
	for (idx = 0; idx < x86_pmu.num_counters_fixed; idx++) {
1218 1219
		rdmsrl(MSR_ARCH_PERFMON_FIXED_CTR0 + idx, pmc_count);

1220
		pr_info("CPU#%d: fixed-PMC%d count: %016llx\n",
1221 1222
			cpu, idx, pmc_count);
	}
1223
	local_irq_restore(flags);
I
Ingo Molnar 已提交
1224 1225
}

P
Peter Zijlstra 已提交
1226
static void x86_pmu_stop(struct perf_event *event, int flags)
I
Ingo Molnar 已提交
1227
{
1228
	struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
1229
	struct hw_perf_event *hwc = &event->hw;
I
Ingo Molnar 已提交
1230

P
Peter Zijlstra 已提交
1231 1232 1233 1234 1235 1236
	if (__test_and_clear_bit(hwc->idx, cpuc->active_mask)) {
		x86_pmu.disable(event);
		cpuc->events[hwc->idx] = NULL;
		WARN_ON_ONCE(hwc->state & PERF_HES_STOPPED);
		hwc->state |= PERF_HES_STOPPED;
	}
1237

P
Peter Zijlstra 已提交
1238 1239 1240 1241 1242 1243 1244 1245
	if ((flags & PERF_EF_UPDATE) && !(hwc->state & PERF_HES_UPTODATE)) {
		/*
		 * Drain the remaining delta count out of a event
		 * that we are disabling:
		 */
		x86_perf_event_update(event);
		hwc->state |= PERF_HES_UPTODATE;
	}
1246 1247
}

P
Peter Zijlstra 已提交
1248
static void x86_pmu_del(struct perf_event *event, int flags)
1249 1250 1251 1252
{
	struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
	int i;

1253 1254 1255 1256 1257
	/*
	 * If we're called during a txn, we don't need to do anything.
	 * The events never got scheduled and ->cancel_txn will truncate
	 * the event_list.
	 */
1258
	if (cpuc->group_flag & PERF_EVENT_TXN)
1259 1260
		return;

P
Peter Zijlstra 已提交
1261
	x86_pmu_stop(event, PERF_EF_UPDATE);
1262

1263 1264 1265 1266 1267 1268 1269 1270 1271 1272
	for (i = 0; i < cpuc->n_events; i++) {
		if (event == cpuc->event_list[i]) {

			if (x86_pmu.put_event_constraints)
				x86_pmu.put_event_constraints(cpuc, event);

			while (++i < cpuc->n_events)
				cpuc->event_list[i-1] = cpuc->event_list[i];

			--cpuc->n_events;
1273
			break;
1274 1275
		}
	}
1276
	perf_event_update_userpage(event);
I
Ingo Molnar 已提交
1277 1278
}

1279
static int x86_pmu_handle_irq(struct pt_regs *regs)
1280
{
1281
	struct perf_sample_data data;
1282 1283
	struct cpu_hw_events *cpuc;
	struct perf_event *event;
V
Vince Weaver 已提交
1284
	int idx, handled = 0;
1285 1286
	u64 val;

1287
	perf_sample_data_init(&data, 0);
1288

1289
	cpuc = &__get_cpu_var(cpu_hw_events);
1290

1291
	for (idx = 0; idx < x86_pmu.num_counters; idx++) {
1292 1293 1294 1295 1296 1297 1298 1299
		if (!test_bit(idx, cpuc->active_mask)) {
			/*
			 * Though we deactivated the counter some cpus
			 * might still deliver spurious interrupts still
			 * in flight. Catch them:
			 */
			if (__test_and_clear_bit(idx, cpuc->running))
				handled++;
1300
			continue;
1301
		}
1302

1303
		event = cpuc->events[idx];
1304

1305
		val = x86_perf_event_update(event);
1306
		if (val & (1ULL << (x86_pmu.cntval_bits - 1)))
1307
			continue;
1308

1309
		/*
1310
		 * event overflow
1311
		 */
1312
		handled++;
1313
		data.period	= event->hw.last_period;
1314

1315
		if (!x86_perf_event_set_period(event))
1316 1317
			continue;

1318
		if (perf_event_overflow(event, 1, &data, regs))
P
Peter Zijlstra 已提交
1319
			x86_pmu_stop(event, 0);
1320
	}
1321

1322 1323 1324
	if (handled)
		inc_irq_stat(apic_perf_irqs);

1325 1326
	return handled;
}
1327

1328
void perf_events_lapic_init(void)
I
Ingo Molnar 已提交
1329
{
1330
	if (!x86_pmu.apic || !x86_pmu_initialized())
I
Ingo Molnar 已提交
1331
		return;
1332

I
Ingo Molnar 已提交
1333
	/*
1334
	 * Always use NMI for PMU
I
Ingo Molnar 已提交
1335
	 */
1336
	apic_write(APIC_LVTPC, APIC_DM_NMI);
I
Ingo Molnar 已提交
1337 1338
}

1339 1340 1341 1342 1343 1344 1345
struct pmu_nmi_state {
	unsigned int	marked;
	int		handled;
};

static DEFINE_PER_CPU(struct pmu_nmi_state, pmu_nmi);

I
Ingo Molnar 已提交
1346
static int __kprobes
1347
perf_event_nmi_handler(struct notifier_block *self,
I
Ingo Molnar 已提交
1348 1349 1350
			 unsigned long cmd, void *__args)
{
	struct die_args *args = __args;
1351 1352
	unsigned int this_nmi;
	int handled;
1353

1354
	if (!atomic_read(&active_events))
1355 1356
		return NOTIFY_DONE;

1357 1358 1359
	switch (cmd) {
	case DIE_NMI:
		break;
1360 1361
	case DIE_NMIUNKNOWN:
		this_nmi = percpu_read(irq_stat.__nmi_count);
T
Tejun Heo 已提交
1362
		if (this_nmi != __this_cpu_read(pmu_nmi.marked))
1363 1364 1365 1366 1367 1368 1369 1370 1371 1372
			/* let the kernel handle the unknown nmi */
			return NOTIFY_DONE;
		/*
		 * This one is a PMU back-to-back nmi. Two events
		 * trigger 'simultaneously' raising two back-to-back
		 * NMIs. If the first NMI handles both, the latter
		 * will be empty and daze the CPU. So, we drop it to
		 * avoid false-positive 'unknown nmi' messages.
		 */
		return NOTIFY_STOP;
1373
	default:
I
Ingo Molnar 已提交
1374
		return NOTIFY_DONE;
1375
	}
I
Ingo Molnar 已提交
1376 1377

	apic_write(APIC_LVTPC, APIC_DM_NMI);
1378 1379 1380 1381 1382 1383 1384 1385

	handled = x86_pmu.handle_irq(args->regs);
	if (!handled)
		return NOTIFY_DONE;

	this_nmi = percpu_read(irq_stat.__nmi_count);
	if ((handled > 1) ||
		/* the next nmi could be a back-to-back nmi */
T
Tejun Heo 已提交
1386 1387
	    ((__this_cpu_read(pmu_nmi.marked) == this_nmi) &&
	     (__this_cpu_read(pmu_nmi.handled) > 1))) {
1388 1389 1390 1391 1392 1393 1394 1395 1396 1397
		/*
		 * We could have two subsequent back-to-back nmis: The
		 * first handles more than one counter, the 2nd
		 * handles only one counter and the 3rd handles no
		 * counter.
		 *
		 * This is the 2nd nmi because the previous was
		 * handling more than one counter. We will mark the
		 * next (3rd) and then drop it if unhandled.
		 */
T
Tejun Heo 已提交
1398 1399
		__this_cpu_write(pmu_nmi.marked, this_nmi + 1);
		__this_cpu_write(pmu_nmi.handled, handled);
1400
	}
I
Ingo Molnar 已提交
1401

1402
	return NOTIFY_STOP;
I
Ingo Molnar 已提交
1403 1404
}

1405 1406 1407
static __read_mostly struct notifier_block perf_event_nmi_notifier = {
	.notifier_call		= perf_event_nmi_handler,
	.next			= NULL,
1408
	.priority		= NMI_LOCAL_LOW_PRIOR,
1409 1410
};

1411
static struct event_constraint unconstrained;
1412
static struct event_constraint emptyconstraint;
1413 1414

static struct event_constraint *
1415
x86_get_event_constraints(struct cpu_hw_events *cpuc, struct perf_event *event)
1416
{
1417
	struct event_constraint *c;
1418 1419 1420

	if (x86_pmu.event_constraints) {
		for_each_event_constraint(c, x86_pmu.event_constraints) {
1421 1422
			if ((event->hw.config & c->cmask) == c->code)
				return c;
1423 1424
		}
	}
1425 1426

	return &unconstrained;
1427 1428
}

1429 1430
#include "perf_event_amd.c"
#include "perf_event_p6.c"
1431
#include "perf_event_p4.c"
1432
#include "perf_event_intel_lbr.c"
1433
#include "perf_event_intel_ds.c"
1434
#include "perf_event_intel.c"
1435

1436 1437 1438 1439
static int __cpuinit
x86_pmu_notifier(struct notifier_block *self, unsigned long action, void *hcpu)
{
	unsigned int cpu = (long)hcpu;
1440
	int ret = NOTIFY_OK;
1441 1442 1443 1444

	switch (action & ~CPU_TASKS_FROZEN) {
	case CPU_UP_PREPARE:
		if (x86_pmu.cpu_prepare)
1445
			ret = x86_pmu.cpu_prepare(cpu);
1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457
		break;

	case CPU_STARTING:
		if (x86_pmu.cpu_starting)
			x86_pmu.cpu_starting(cpu);
		break;

	case CPU_DYING:
		if (x86_pmu.cpu_dying)
			x86_pmu.cpu_dying(cpu);
		break;

1458
	case CPU_UP_CANCELED:
1459 1460 1461 1462 1463 1464 1465 1466 1467
	case CPU_DEAD:
		if (x86_pmu.cpu_dead)
			x86_pmu.cpu_dead(cpu);
		break;

	default:
		break;
	}

1468
	return ret;
1469 1470
}

1471 1472 1473 1474 1475 1476 1477 1478 1479 1480
static void __init pmu_check_apic(void)
{
	if (cpu_has_apic)
		return;

	x86_pmu.apic = 0;
	pr_info("no APIC, boot with the \"lapic\" boot parameter to force-enable it.\n");
	pr_info("no hardware sampling interrupt available.\n");
}

1481
static int __init init_hw_perf_events(void)
1482
{
1483
	struct event_constraint *c;
1484 1485
	int err;

1486
	pr_info("Performance Events: ");
1487

1488 1489
	switch (boot_cpu_data.x86_vendor) {
	case X86_VENDOR_INTEL:
1490
		err = intel_pmu_init();
1491
		break;
1492
	case X86_VENDOR_AMD:
1493
		err = amd_pmu_init();
1494
		break;
1495
	default:
1496
		return 0;
1497
	}
1498
	if (err != 0) {
1499
		pr_cont("no PMU driver, software events only.\n");
1500
		return 0;
1501
	}
1502

1503 1504
	pmu_check_apic();

1505
	/* sanity check that the hardware exists or is emulated */
1506
	if (!check_hw_exists())
1507
		return 0;
1508

1509
	pr_cont("%s PMU driver.\n", x86_pmu.name);
1510

1511 1512 1513
	if (x86_pmu.quirks)
		x86_pmu.quirks();

1514
	if (x86_pmu.num_counters > X86_PMC_MAX_GENERIC) {
1515
		WARN(1, KERN_ERR "hw perf events %d > max(%d), clipping!",
1516 1517
		     x86_pmu.num_counters, X86_PMC_MAX_GENERIC);
		x86_pmu.num_counters = X86_PMC_MAX_GENERIC;
I
Ingo Molnar 已提交
1518
	}
1519
	x86_pmu.intel_ctrl = (1 << x86_pmu.num_counters) - 1;
I
Ingo Molnar 已提交
1520

1521
	if (x86_pmu.num_counters_fixed > X86_PMC_MAX_FIXED) {
1522
		WARN(1, KERN_ERR "hw perf events fixed %d > max(%d), clipping!",
1523 1524
		     x86_pmu.num_counters_fixed, X86_PMC_MAX_FIXED);
		x86_pmu.num_counters_fixed = X86_PMC_MAX_FIXED;
1525
	}
1526

1527
	x86_pmu.intel_ctrl |=
1528
		((1LL << x86_pmu.num_counters_fixed)-1) << X86_PMC_IDX_FIXED;
I
Ingo Molnar 已提交
1529

1530 1531
	perf_events_lapic_init();
	register_die_notifier(&perf_event_nmi_notifier);
1532

1533
	unconstrained = (struct event_constraint)
1534 1535
		__EVENT_CONSTRAINT(0, (1ULL << x86_pmu.num_counters) - 1,
				   0, x86_pmu.num_counters);
1536

1537 1538
	if (x86_pmu.event_constraints) {
		for_each_event_constraint(c, x86_pmu.event_constraints) {
1539
			if (c->cmask != X86_RAW_EVENT_MASK)
1540 1541
				continue;

1542 1543
			c->idxmsk64 |= (1ULL << x86_pmu.num_counters) - 1;
			c->weight += x86_pmu.num_counters;
1544 1545 1546
		}
	}

I
Ingo Molnar 已提交
1547
	pr_info("... version:                %d\n",     x86_pmu.version);
1548 1549 1550
	pr_info("... bit width:              %d\n",     x86_pmu.cntval_bits);
	pr_info("... generic registers:      %d\n",     x86_pmu.num_counters);
	pr_info("... value mask:             %016Lx\n", x86_pmu.cntval_mask);
I
Ingo Molnar 已提交
1551
	pr_info("... max period:             %016Lx\n", x86_pmu.max_period);
1552
	pr_info("... fixed-purpose events:   %d\n",     x86_pmu.num_counters_fixed);
1553
	pr_info("... event mask:             %016Lx\n", x86_pmu.intel_ctrl);
1554

P
Peter Zijlstra 已提交
1555
	perf_pmu_register(&pmu, "cpu", PERF_TYPE_RAW);
1556
	perf_cpu_notifier(x86_pmu_notifier);
1557 1558

	return 0;
I
Ingo Molnar 已提交
1559
}
1560
early_initcall(init_hw_perf_events);
I
Ingo Molnar 已提交
1561

1562
static inline void x86_pmu_read(struct perf_event *event)
1563
{
1564
	x86_perf_event_update(event);
1565 1566
}

1567 1568 1569 1570 1571
/*
 * Start group events scheduling transaction
 * Set the flag to make pmu::enable() not perform the
 * schedulability test, it will be performed at commit time
 */
P
Peter Zijlstra 已提交
1572
static void x86_pmu_start_txn(struct pmu *pmu)
1573
{
P
Peter Zijlstra 已提交
1574
	perf_pmu_disable(pmu);
T
Tejun Heo 已提交
1575 1576
	__this_cpu_or(cpu_hw_events.group_flag, PERF_EVENT_TXN);
	__this_cpu_write(cpu_hw_events.n_txn, 0);
1577 1578 1579 1580 1581 1582 1583
}

/*
 * Stop group events scheduling transaction
 * Clear the flag and pmu::enable() will perform the
 * schedulability test.
 */
P
Peter Zijlstra 已提交
1584
static void x86_pmu_cancel_txn(struct pmu *pmu)
1585
{
T
Tejun Heo 已提交
1586
	__this_cpu_and(cpu_hw_events.group_flag, ~PERF_EVENT_TXN);
1587 1588 1589
	/*
	 * Truncate the collected events.
	 */
T
Tejun Heo 已提交
1590 1591
	__this_cpu_sub(cpu_hw_events.n_added, __this_cpu_read(cpu_hw_events.n_txn));
	__this_cpu_sub(cpu_hw_events.n_events, __this_cpu_read(cpu_hw_events.n_txn));
P
Peter Zijlstra 已提交
1592
	perf_pmu_enable(pmu);
1593 1594 1595 1596 1597 1598 1599
}

/*
 * Commit group events scheduling transaction
 * Perform the group schedulability test as a whole
 * Return 0 if success
 */
P
Peter Zijlstra 已提交
1600
static int x86_pmu_commit_txn(struct pmu *pmu)
1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613 1614 1615 1616 1617 1618 1619 1620
{
	struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
	int assign[X86_PMC_IDX_MAX];
	int n, ret;

	n = cpuc->n_events;

	if (!x86_pmu_initialized())
		return -EAGAIN;

	ret = x86_pmu.schedule_events(cpuc, n, assign);
	if (ret)
		return ret;

	/*
	 * copy new assignment, now we know it is possible
	 * will be used by hw_perf_enable()
	 */
	memcpy(cpuc->assign, assign, n*sizeof(int));

1621
	cpuc->group_flag &= ~PERF_EVENT_TXN;
P
Peter Zijlstra 已提交
1622
	perf_pmu_enable(pmu);
1623 1624 1625
	return 0;
}

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
/*
 * validate that we can schedule this event
 */
static int validate_event(struct perf_event *event)
{
	struct cpu_hw_events *fake_cpuc;
	struct event_constraint *c;
	int ret = 0;

	fake_cpuc = kmalloc(sizeof(*fake_cpuc), GFP_KERNEL | __GFP_ZERO);
	if (!fake_cpuc)
		return -ENOMEM;

	c = x86_pmu.get_event_constraints(fake_cpuc, event);

	if (!c || !c->weight)
		ret = -ENOSPC;

	if (x86_pmu.put_event_constraints)
		x86_pmu.put_event_constraints(fake_cpuc, event);

	kfree(fake_cpuc);

	return ret;
}

1652 1653 1654 1655
/*
 * validate a single event group
 *
 * validation include:
1656 1657 1658
 *	- check events are compatible which each other
 *	- events do not compete for the same counter
 *	- number of events <= number of counters
1659 1660 1661 1662
 *
 * validation ensures the group can be loaded onto the
 * PMU if it was the only group available.
 */
1663 1664
static int validate_group(struct perf_event *event)
{
1665
	struct perf_event *leader = event->group_leader;
1666 1667
	struct cpu_hw_events *fake_cpuc;
	int ret, n;
1668

1669 1670 1671 1672
	ret = -ENOMEM;
	fake_cpuc = kmalloc(sizeof(*fake_cpuc), GFP_KERNEL | __GFP_ZERO);
	if (!fake_cpuc)
		goto out;
1673

1674 1675 1676 1677 1678 1679
	/*
	 * the event is not yet connected with its
	 * siblings therefore we must first collect
	 * existing siblings, then add the new event
	 * before we can simulate the scheduling
	 */
1680 1681
	ret = -ENOSPC;
	n = collect_events(fake_cpuc, leader, true);
1682
	if (n < 0)
1683
		goto out_free;
1684

1685 1686
	fake_cpuc->n_events = n;
	n = collect_events(fake_cpuc, event, false);
1687
	if (n < 0)
1688
		goto out_free;
1689

1690
	fake_cpuc->n_events = n;
1691

1692
	ret = x86_pmu.schedule_events(fake_cpuc, n, NULL);
1693 1694 1695 1696 1697

out_free:
	kfree(fake_cpuc);
out:
	return ret;
1698 1699
}

1700
static int x86_pmu_event_init(struct perf_event *event)
I
Ingo Molnar 已提交
1701
{
P
Peter Zijlstra 已提交
1702
	struct pmu *tmp;
I
Ingo Molnar 已提交
1703 1704
	int err;

1705 1706 1707 1708 1709 1710 1711 1712 1713 1714 1715
	switch (event->attr.type) {
	case PERF_TYPE_RAW:
	case PERF_TYPE_HARDWARE:
	case PERF_TYPE_HW_CACHE:
		break;

	default:
		return -ENOENT;
	}

	err = __x86_pmu_event_init(event);
1716
	if (!err) {
1717 1718 1719 1720 1721 1722 1723 1724
		/*
		 * we temporarily connect event to its pmu
		 * such that validate_group() can classify
		 * it as an x86 event using is_x86_event()
		 */
		tmp = event->pmu;
		event->pmu = &pmu;

1725 1726
		if (event->group_leader != event)
			err = validate_group(event);
1727 1728
		else
			err = validate_event(event);
1729 1730

		event->pmu = tmp;
1731
	}
1732
	if (err) {
1733 1734
		if (event->destroy)
			event->destroy(event);
1735
	}
I
Ingo Molnar 已提交
1736

1737
	return err;
I
Ingo Molnar 已提交
1738
}
1739

1740
static struct pmu pmu = {
P
Peter Zijlstra 已提交
1741 1742 1743
	.pmu_enable	= x86_pmu_enable,
	.pmu_disable	= x86_pmu_disable,

1744
	.event_init	= x86_pmu_event_init,
P
Peter Zijlstra 已提交
1745 1746 1747

	.add		= x86_pmu_add,
	.del		= x86_pmu_del,
1748 1749 1750
	.start		= x86_pmu_start,
	.stop		= x86_pmu_stop,
	.read		= x86_pmu_read,
P
Peter Zijlstra 已提交
1751

1752 1753 1754 1755 1756
	.start_txn	= x86_pmu_start_txn,
	.cancel_txn	= x86_pmu_cancel_txn,
	.commit_txn	= x86_pmu_commit_txn,
};

1757 1758 1759 1760 1761 1762 1763 1764 1765 1766 1767 1768 1769 1770 1771 1772 1773
/*
 * callchain support
 */

static void
backtrace_warning_symbol(void *data, char *msg, unsigned long symbol)
{
	/* Ignore warnings */
}

static void backtrace_warning(void *data, char *msg)
{
	/* Ignore warnings */
}

static int backtrace_stack(void *data, char *name)
{
1774
	return 0;
1775 1776 1777 1778 1779 1780
}

static void backtrace_address(void *data, unsigned long addr, int reliable)
{
	struct perf_callchain_entry *entry = data;

1781
	perf_callchain_store(entry, addr);
1782 1783 1784 1785 1786 1787 1788
}

static const struct stacktrace_ops backtrace_ops = {
	.warning		= backtrace_warning,
	.warning_symbol		= backtrace_warning_symbol,
	.stack			= backtrace_stack,
	.address		= backtrace_address,
1789
	.walk_stack		= print_context_stack_bp,
1790 1791
};

1792 1793
void
perf_callchain_kernel(struct perf_callchain_entry *entry, struct pt_regs *regs)
1794
{
1795 1796
	if (perf_guest_cbs && perf_guest_cbs->is_in_guest()) {
		/* TODO: We don't support guest os callchain now */
1797
		return;
1798 1799
	}

1800
	perf_callchain_store(entry, regs->ip);
1801

1802
	dump_trace(NULL, regs, NULL, 0, &backtrace_ops, entry);
1803 1804
}

1805 1806 1807
#ifdef CONFIG_COMPAT
static inline int
perf_callchain_user32(struct pt_regs *regs, struct perf_callchain_entry *entry)
1808
{
1809 1810 1811
	/* 32-bit process in 64-bit kernel. */
	struct stack_frame_ia32 frame;
	const void __user *fp;
1812

1813 1814 1815 1816 1817 1818 1819 1820 1821 1822 1823 1824
	if (!test_thread_flag(TIF_IA32))
		return 0;

	fp = compat_ptr(regs->bp);
	while (entry->nr < PERF_MAX_STACK_DEPTH) {
		unsigned long bytes;
		frame.next_frame     = 0;
		frame.return_address = 0;

		bytes = copy_from_user_nmi(&frame, fp, sizeof(frame));
		if (bytes != sizeof(frame))
			break;
1825

1826 1827
		if (fp < compat_ptr(regs->sp))
			break;
1828

1829
		perf_callchain_store(entry, frame.return_address);
1830 1831 1832
		fp = compat_ptr(frame.next_frame);
	}
	return 1;
1833
}
1834 1835 1836 1837 1838 1839 1840
#else
static inline int
perf_callchain_user32(struct pt_regs *regs, struct perf_callchain_entry *entry)
{
    return 0;
}
#endif
1841

1842 1843
void
perf_callchain_user(struct perf_callchain_entry *entry, struct pt_regs *regs)
1844 1845 1846 1847
{
	struct stack_frame frame;
	const void __user *fp;

1848 1849
	if (perf_guest_cbs && perf_guest_cbs->is_in_guest()) {
		/* TODO: We don't support guest os callchain now */
1850
		return;
1851
	}
1852

1853
	fp = (void __user *)regs->bp;
1854

1855
	perf_callchain_store(entry, regs->ip);
1856

1857 1858 1859
	if (perf_callchain_user32(regs, entry))
		return;

1860
	while (entry->nr < PERF_MAX_STACK_DEPTH) {
1861
		unsigned long bytes;
1862
		frame.next_frame	     = NULL;
1863 1864
		frame.return_address = 0;

1865 1866
		bytes = copy_from_user_nmi(&frame, fp, sizeof(frame));
		if (bytes != sizeof(frame))
1867 1868
			break;

1869
		if ((unsigned long)fp < regs->sp)
1870 1871
			break;

1872
		perf_callchain_store(entry, frame.return_address);
1873
		fp = frame.next_frame;
1874 1875 1876
	}
}

1877 1878 1879
unsigned long perf_instruction_pointer(struct pt_regs *regs)
{
	unsigned long ip;
1880

1881 1882 1883 1884
	if (perf_guest_cbs && perf_guest_cbs->is_in_guest())
		ip = perf_guest_cbs->get_guest_ip();
	else
		ip = instruction_pointer(regs);
1885

1886 1887 1888 1889 1890 1891
	return ip;
}

unsigned long perf_misc_flags(struct pt_regs *regs)
{
	int misc = 0;
1892

1893
	if (perf_guest_cbs && perf_guest_cbs->is_in_guest()) {
1894 1895 1896 1897 1898 1899 1900 1901 1902 1903 1904
		if (perf_guest_cbs->is_user_mode())
			misc |= PERF_RECORD_MISC_GUEST_USER;
		else
			misc |= PERF_RECORD_MISC_GUEST_KERNEL;
	} else {
		if (user_mode(regs))
			misc |= PERF_RECORD_MISC_USER;
		else
			misc |= PERF_RECORD_MISC_KERNEL;
	}

1905
	if (regs->flags & PERF_EFLAGS_EXACT)
P
Peter Zijlstra 已提交
1906
		misc |= PERF_RECORD_MISC_EXACT_IP;
1907 1908 1909

	return misc;
}