perf_event.c 42.5 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);
578 579 580 581 582 583 584 585 586 587 588 589
	} 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;
	}

	if (attr->type == PERF_TYPE_RAW)
590
		return x86_pmu_extra_regs(event->attr.config, event);
591 592

	if (attr->type == PERF_TYPE_HW_CACHE)
593
		return set_ext_hw_attr(hwc, event);
594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614

	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:
	 */
	if ((attr->config == PERF_COUNT_HW_BRANCH_INSTRUCTIONS) &&
	    (hwc->sample_period == 1)) {
		/* BTS is not supported by this architecture. */
615
		if (!x86_pmu.bts_active)
616 617 618 619 620 621 622 623 624 625 626
			return -EOPNOTSUPP;

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

	hwc->config |= config;

	return 0;
}
627

628
static int x86_pmu_hw_config(struct perf_event *event)
629
{
P
Peter Zijlstra 已提交
630 631 632 633
	if (event->attr.precise_ip) {
		int precise = 0;

		/* Support for constant skid */
634
		if (x86_pmu.pebs_active) {
P
Peter Zijlstra 已提交
635 636
			precise++;

637 638 639 640
			/* Support for IP fixup */
			if (x86_pmu.lbr_nr)
				precise++;
		}
P
Peter Zijlstra 已提交
641 642 643 644 645

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

646 647 648 649
	/*
	 * Generate PMC IRQs:
	 * (keep 'enabled' bit clear for now)
	 */
650
	event->hw.config = ARCH_PERFMON_EVENTSEL_INT;
651 652 653 654

	/*
	 * Count user and OS events unless requested not to
	 */
655 656 657 658
	if (!event->attr.exclude_user)
		event->hw.config |= ARCH_PERFMON_EVENTSEL_USR;
	if (!event->attr.exclude_kernel)
		event->hw.config |= ARCH_PERFMON_EVENTSEL_OS;
659

660 661
	if (event->attr.type == PERF_TYPE_RAW)
		event->hw.config |= event->attr.config & X86_RAW_EVENT_MASK;
662

663
	return x86_setup_perfctr(event);
664 665
}

I
Ingo Molnar 已提交
666
/*
667
 * Setup the hardware configuration for a given attr_type
I
Ingo Molnar 已提交
668
 */
669
static int __x86_pmu_event_init(struct perf_event *event)
I
Ingo Molnar 已提交
670
{
P
Peter Zijlstra 已提交
671
	int err;
I
Ingo Molnar 已提交
672

673 674
	if (!x86_pmu_initialized())
		return -ENODEV;
I
Ingo Molnar 已提交
675

P
Peter Zijlstra 已提交
676
	err = 0;
677
	if (!atomic_inc_not_zero(&active_events)) {
P
Peter Zijlstra 已提交
678
		mutex_lock(&pmc_reserve_mutex);
679
		if (atomic_read(&active_events) == 0) {
680 681
			if (!reserve_pmc_hardware())
				err = -EBUSY;
682 683
			else
				reserve_ds_buffers();
684 685
		}
		if (!err)
686
			atomic_inc(&active_events);
P
Peter Zijlstra 已提交
687 688 689 690 691
		mutex_unlock(&pmc_reserve_mutex);
	}
	if (err)
		return err;

692
	event->destroy = hw_perf_event_destroy;
693

694 695 696
	event->hw.idx = -1;
	event->hw.last_cpu = -1;
	event->hw.last_tag = ~0ULL;
697

698
	return x86_pmu.hw_config(event);
699 700
}

701
static void x86_pmu_disable_all(void)
702
{
703
	struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
704 705
	int idx;

706
	for (idx = 0; idx < x86_pmu.num_counters; idx++) {
707 708
		u64 val;

709
		if (!test_bit(idx, cpuc->active_mask))
710
			continue;
711
		rdmsrl(x86_pmu_config_addr(idx), val);
712
		if (!(val & ARCH_PERFMON_EVENTSEL_ENABLE))
713
			continue;
714
		val &= ~ARCH_PERFMON_EVENTSEL_ENABLE;
715
		wrmsrl(x86_pmu_config_addr(idx), val);
716 717 718
	}
}

P
Peter Zijlstra 已提交
719
static void x86_pmu_disable(struct pmu *pmu)
720
{
721 722
	struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);

723
	if (!x86_pmu_initialized())
724
		return;
725

726 727 728 729 730 731
	if (!cpuc->enabled)
		return;

	cpuc->n_added = 0;
	cpuc->enabled = 0;
	barrier();
732 733

	x86_pmu.disable_all();
734
}
I
Ingo Molnar 已提交
735

736 737 738
static inline void __x86_pmu_enable_event(struct hw_perf_event *hwc,
					  u64 enable_mask)
{
739 740
	if (hwc->extra_reg)
		wrmsrl(hwc->extra_reg, hwc->extra_config);
741
	wrmsrl(hwc->config_base, hwc->config | enable_mask);
742 743
}

744
static void x86_pmu_enable_all(int added)
745
{
746
	struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
747 748
	int idx;

749
	for (idx = 0; idx < x86_pmu.num_counters; idx++) {
750
		struct hw_perf_event *hwc = &cpuc->events[idx]->hw;
751

752
		if (!test_bit(idx, cpuc->active_mask))
753
			continue;
754

755
		__x86_pmu_enable_event(hwc, ARCH_PERFMON_EVENTSEL_ENABLE);
756 757 758
	}
}

P
Peter Zijlstra 已提交
759
static struct pmu pmu;
760 761 762 763 764 765 766 767

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)
{
768
	struct event_constraint *c, *constraints[X86_PMC_IDX_MAX];
769
	unsigned long used_mask[BITS_TO_LONGS(X86_PMC_IDX_MAX)];
770
	int i, j, w, wmax, num = 0;
771 772 773 774 775
	struct hw_perf_event *hwc;

	bitmap_zero(used_mask, X86_PMC_IDX_MAX);

	for (i = 0; i < n; i++) {
776 777
		c = x86_pmu.get_event_constraints(cpuc, cpuc->event_list[i]);
		constraints[i] = c;
778 779
	}

780 781 782
	/*
	 * fastpath, try to reuse previous register
	 */
783
	for (i = 0; i < n; i++) {
784
		hwc = &cpuc->event_list[i]->hw;
785
		c = constraints[i];
786 787 788 789 790 791

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

		/* constraint still honored */
792
		if (!test_bit(hwc->idx, c->idxmsk))
793 794 795 796 797 798
			break;

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

P
Peter Zijlstra 已提交
799
		__set_bit(hwc->idx, used_mask);
800 801 802
		if (assign)
			assign[i] = hwc->idx;
	}
803
	if (i == n)
804 805 806 807 808 809 810 811
		goto done;

	/*
	 * begin slow path
	 */

	bitmap_zero(used_mask, X86_PMC_IDX_MAX);

812 813 814 815 816 817 818 819 820
	/*
	 * 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.
	 */
821
	wmax = x86_pmu.num_counters;
822 823 824 825 826 827

	/*
	 * when fixed event counters are present,
	 * wmax is incremented by 1 to account
	 * for one more choice
	 */
828
	if (x86_pmu.num_counters_fixed)
829 830
		wmax++;

831
	for (w = 1, num = n; num && w <= wmax; w++) {
832
		/* for each event */
833
		for (i = 0; num && i < n; i++) {
834
			c = constraints[i];
835 836
			hwc = &cpuc->event_list[i]->hw;

837
			if (c->weight != w)
838 839
				continue;

840
			for_each_set_bit(j, c->idxmsk, X86_PMC_IDX_MAX) {
841 842 843 844 845 846 847
				if (!test_bit(j, used_mask))
					break;
			}

			if (j == X86_PMC_IDX_MAX)
				break;

P
Peter Zijlstra 已提交
848
			__set_bit(j, used_mask);
849

850 851 852 853 854
			if (assign)
				assign[i] = j;
			num--;
		}
	}
855
done:
856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877
	/*
	 * 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;

878
	max_count = x86_pmu.num_counters + x86_pmu.num_counters_fixed;
879 880 881 882 883 884 885 886 887 888 889 890 891 892 893

	/* 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) ||
894
		    event->state <= PERF_EVENT_STATE_OFF)
895 896 897 898 899 900 901 902 903 904 905 906
			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,
907
				struct cpu_hw_events *cpuc, int i)
908
{
909 910 911 912 913
	struct hw_perf_event *hwc = &event->hw;

	hwc->idx = cpuc->assign[i];
	hwc->last_cpu = smp_processor_id();
	hwc->last_tag = ++cpuc->tags[i];
914 915 916 917 918 919

	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;
920
		hwc->event_base = MSR_ARCH_PERFMON_FIXED_CTR0 + (hwc->idx - X86_PMC_IDX_FIXED);
921
	} else {
922 923
		hwc->config_base = x86_pmu_config_addr(hwc->idx);
		hwc->event_base  = x86_pmu_event_addr(hwc->idx);
924 925 926
	}
}

927 928 929 930 931 932 933 934 935
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
936 937
static void x86_pmu_start(struct perf_event *event, int flags);
static void x86_pmu_stop(struct perf_event *event, int flags);
938

P
Peter Zijlstra 已提交
939
static void x86_pmu_enable(struct pmu *pmu)
940
{
941 942 943
	struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
	struct perf_event *event;
	struct hw_perf_event *hwc;
944
	int i, added = cpuc->n_added;
945

946
	if (!x86_pmu_initialized())
947
		return;
948 949 950 951

	if (cpuc->enabled)
		return;

952
	if (cpuc->n_added) {
953
		int n_running = cpuc->n_events - cpuc->n_added;
954 955 956 957 958 959 960
		/*
		 * 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
		 */
961
		for (i = 0; i < n_running; i++) {
962 963 964
			event = cpuc->event_list[i];
			hwc = &event->hw;

965 966 967 968 969 970 971 972
			/*
			 * 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))
973 974
				continue;

P
Peter Zijlstra 已提交
975 976 977 978 979 980 981 982
			/*
			 * 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);
983 984 985 986 987 988
		}

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

989
			if (!match_prev_assignment(hwc, cpuc, i))
990
				x86_assign_hw_event(event, cpuc, i);
991 992
			else if (i < n_running)
				continue;
993

P
Peter Zijlstra 已提交
994 995 996 997
			if (hwc->state & PERF_HES_ARCH)
				continue;

			x86_pmu_start(event, PERF_EF_RELOAD);
998 999 1000 1001
		}
		cpuc->n_added = 0;
		perf_events_lapic_init();
	}
1002 1003 1004 1005

	cpuc->enabled = 1;
	barrier();

1006
	x86_pmu.enable_all(added);
1007 1008
}

1009
static inline void x86_pmu_disable_event(struct perf_event *event)
1010
{
1011
	struct hw_perf_event *hwc = &event->hw;
1012

1013
	wrmsrl(hwc->config_base, hwc->config);
1014 1015
}

1016
static DEFINE_PER_CPU(u64 [X86_PMC_IDX_MAX], pmc_prev_left);
I
Ingo Molnar 已提交
1017

1018 1019
/*
 * Set the next IRQ period, based on the hwc->period_left value.
1020
 * To be called with the event disabled in hw:
1021
 */
1022
static int
1023
x86_perf_event_set_period(struct perf_event *event)
I
Ingo Molnar 已提交
1024
{
1025
	struct hw_perf_event *hwc = &event->hw;
1026
	s64 left = local64_read(&hwc->period_left);
1027
	s64 period = hwc->sample_period;
1028
	int ret = 0, idx = hwc->idx;
1029

1030 1031 1032
	if (idx == X86_PMC_IDX_FIXED_BTS)
		return 0;

1033
	/*
1034
	 * If we are way outside a reasonable range then just skip forward:
1035 1036 1037
	 */
	if (unlikely(left <= -period)) {
		left = period;
1038
		local64_set(&hwc->period_left, left);
1039
		hwc->last_period = period;
1040
		ret = 1;
1041 1042 1043 1044
	}

	if (unlikely(left <= 0)) {
		left += period;
1045
		local64_set(&hwc->period_left, left);
1046
		hwc->last_period = period;
1047
		ret = 1;
1048
	}
1049
	/*
1050
	 * Quirk: certain CPUs dont like it if just 1 hw_event is left:
1051 1052 1053
	 */
	if (unlikely(left < 2))
		left = 2;
I
Ingo Molnar 已提交
1054

1055 1056 1057
	if (left > x86_pmu.max_period)
		left = x86_pmu.max_period;

1058
	per_cpu(pmc_prev_left[idx], smp_processor_id()) = left;
1059 1060

	/*
1061
	 * The hw event starts counting from this event offset,
1062 1063
	 * mark it to be able to extra future deltas:
	 */
1064
	local64_set(&hwc->prev_count, (u64)-left);
1065

1066
	wrmsrl(hwc->event_base, (u64)(-left) & x86_pmu.cntval_mask);
1067 1068 1069 1070 1071 1072 1073

	/*
	 * 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) {
1074
		wrmsrl(hwc->event_base,
1075
			(u64)(-left) & x86_pmu.cntval_mask);
1076
	}
1077

1078
	perf_event_update_userpage(event);
1079

1080
	return ret;
1081 1082
}

1083
static void x86_pmu_enable_event(struct perf_event *event)
1084
{
T
Tejun Heo 已提交
1085
	if (__this_cpu_read(cpu_hw_events.enabled))
1086 1087
		__x86_pmu_enable_event(&event->hw,
				       ARCH_PERFMON_EVENTSEL_ENABLE);
I
Ingo Molnar 已提交
1088 1089
}

1090
/*
P
Peter Zijlstra 已提交
1091
 * Add a single event to the PMU.
1092 1093 1094
 *
 * The event is added to the group of enabled events
 * but only if it can be scehduled with existing events.
1095
 */
P
Peter Zijlstra 已提交
1096
static int x86_pmu_add(struct perf_event *event, int flags)
1097 1098
{
	struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
1099 1100 1101
	struct hw_perf_event *hwc;
	int assign[X86_PMC_IDX_MAX];
	int n, n0, ret;
1102

1103
	hwc = &event->hw;
1104

P
Peter Zijlstra 已提交
1105
	perf_pmu_disable(event->pmu);
1106
	n0 = cpuc->n_events;
1107 1108 1109
	ret = n = collect_events(cpuc, event, false);
	if (ret < 0)
		goto out;
1110

P
Peter Zijlstra 已提交
1111 1112 1113 1114
	hwc->state = PERF_HES_UPTODATE | PERF_HES_STOPPED;
	if (!(flags & PERF_EF_START))
		hwc->state |= PERF_HES_ARCH;

1115 1116
	/*
	 * If group events scheduling transaction was started,
L
Lucas De Marchi 已提交
1117
	 * skip the schedulability test here, it will be performed
P
Peter Zijlstra 已提交
1118
	 * at commit time (->commit_txn) as a whole
1119
	 */
1120
	if (cpuc->group_flag & PERF_EVENT_TXN)
1121
		goto done_collect;
1122

1123
	ret = x86_pmu.schedule_events(cpuc, n, assign);
1124
	if (ret)
1125
		goto out;
1126 1127 1128 1129 1130
	/*
	 * copy new assignment, now we know it is possible
	 * will be used by hw_perf_enable()
	 */
	memcpy(cpuc->assign, assign, n*sizeof(int));
1131

1132
done_collect:
1133
	cpuc->n_events = n;
1134
	cpuc->n_added += n - n0;
1135
	cpuc->n_txn += n - n0;
1136

1137 1138
	ret = 0;
out:
P
Peter Zijlstra 已提交
1139
	perf_pmu_enable(event->pmu);
1140
	return ret;
I
Ingo Molnar 已提交
1141 1142
}

P
Peter Zijlstra 已提交
1143
static void x86_pmu_start(struct perf_event *event, int flags)
1144
{
P
Peter Zijlstra 已提交
1145 1146 1147
	struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
	int idx = event->hw.idx;

P
Peter Zijlstra 已提交
1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159
	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;
1160

P
Peter Zijlstra 已提交
1161 1162
	cpuc->events[idx] = event;
	__set_bit(idx, cpuc->active_mask);
1163
	__set_bit(idx, cpuc->running);
1164
	x86_pmu.enable(event);
P
Peter Zijlstra 已提交
1165
	perf_event_update_userpage(event);
1166 1167
}

1168
void perf_event_print_debug(void)
I
Ingo Molnar 已提交
1169
{
1170
	u64 ctrl, status, overflow, pmc_ctrl, pmc_count, prev_left, fixed;
1171
	u64 pebs;
1172
	struct cpu_hw_events *cpuc;
1173
	unsigned long flags;
1174 1175
	int cpu, idx;

1176
	if (!x86_pmu.num_counters)
1177
		return;
I
Ingo Molnar 已提交
1178

1179
	local_irq_save(flags);
I
Ingo Molnar 已提交
1180 1181

	cpu = smp_processor_id();
1182
	cpuc = &per_cpu(cpu_hw_events, cpu);
I
Ingo Molnar 已提交
1183

1184
	if (x86_pmu.version >= 2) {
1185 1186 1187 1188
		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);
1189
		rdmsrl(MSR_IA32_PEBS_ENABLE, pebs);
1190 1191 1192 1193 1194 1195

		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);
1196
		pr_info("CPU#%d: pebs:       %016llx\n", cpu, pebs);
1197
	}
1198
	pr_info("CPU#%d: active:     %016llx\n", cpu, *(u64 *)cpuc->active_mask);
I
Ingo Molnar 已提交
1199

1200
	for (idx = 0; idx < x86_pmu.num_counters; idx++) {
1201 1202
		rdmsrl(x86_pmu_config_addr(idx), pmc_ctrl);
		rdmsrl(x86_pmu_event_addr(idx), pmc_count);
I
Ingo Molnar 已提交
1203

1204
		prev_left = per_cpu(pmc_prev_left[idx], cpu);
I
Ingo Molnar 已提交
1205

1206
		pr_info("CPU#%d:   gen-PMC%d ctrl:  %016llx\n",
I
Ingo Molnar 已提交
1207
			cpu, idx, pmc_ctrl);
1208
		pr_info("CPU#%d:   gen-PMC%d count: %016llx\n",
I
Ingo Molnar 已提交
1209
			cpu, idx, pmc_count);
1210
		pr_info("CPU#%d:   gen-PMC%d left:  %016llx\n",
1211
			cpu, idx, prev_left);
I
Ingo Molnar 已提交
1212
	}
1213
	for (idx = 0; idx < x86_pmu.num_counters_fixed; idx++) {
1214 1215
		rdmsrl(MSR_ARCH_PERFMON_FIXED_CTR0 + idx, pmc_count);

1216
		pr_info("CPU#%d: fixed-PMC%d count: %016llx\n",
1217 1218
			cpu, idx, pmc_count);
	}
1219
	local_irq_restore(flags);
I
Ingo Molnar 已提交
1220 1221
}

P
Peter Zijlstra 已提交
1222
static void x86_pmu_stop(struct perf_event *event, int flags)
I
Ingo Molnar 已提交
1223
{
1224
	struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
1225
	struct hw_perf_event *hwc = &event->hw;
I
Ingo Molnar 已提交
1226

P
Peter Zijlstra 已提交
1227 1228 1229 1230 1231 1232
	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;
	}
1233

P
Peter Zijlstra 已提交
1234 1235 1236 1237 1238 1239 1240 1241
	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;
	}
1242 1243
}

P
Peter Zijlstra 已提交
1244
static void x86_pmu_del(struct perf_event *event, int flags)
1245 1246 1247 1248
{
	struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
	int i;

1249 1250 1251 1252 1253
	/*
	 * 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.
	 */
1254
	if (cpuc->group_flag & PERF_EVENT_TXN)
1255 1256
		return;

P
Peter Zijlstra 已提交
1257
	x86_pmu_stop(event, PERF_EF_UPDATE);
1258

1259 1260 1261 1262 1263 1264 1265 1266 1267 1268
	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;
1269
			break;
1270 1271
		}
	}
1272
	perf_event_update_userpage(event);
I
Ingo Molnar 已提交
1273 1274
}

1275
static int x86_pmu_handle_irq(struct pt_regs *regs)
1276
{
1277
	struct perf_sample_data data;
1278 1279
	struct cpu_hw_events *cpuc;
	struct perf_event *event;
V
Vince Weaver 已提交
1280
	int idx, handled = 0;
1281 1282
	u64 val;

1283
	perf_sample_data_init(&data, 0);
1284

1285
	cpuc = &__get_cpu_var(cpu_hw_events);
1286

1287
	for (idx = 0; idx < x86_pmu.num_counters; idx++) {
1288 1289 1290 1291 1292 1293 1294 1295
		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++;
1296
			continue;
1297
		}
1298

1299
		event = cpuc->events[idx];
1300

1301
		val = x86_perf_event_update(event);
1302
		if (val & (1ULL << (x86_pmu.cntval_bits - 1)))
1303
			continue;
1304

1305
		/*
1306
		 * event overflow
1307
		 */
1308
		handled++;
1309
		data.period	= event->hw.last_period;
1310

1311
		if (!x86_perf_event_set_period(event))
1312 1313
			continue;

1314
		if (perf_event_overflow(event, 1, &data, regs))
P
Peter Zijlstra 已提交
1315
			x86_pmu_stop(event, 0);
1316
	}
1317

1318 1319 1320
	if (handled)
		inc_irq_stat(apic_perf_irqs);

1321 1322
	return handled;
}
1323

1324
void perf_events_lapic_init(void)
I
Ingo Molnar 已提交
1325
{
1326
	if (!x86_pmu.apic || !x86_pmu_initialized())
I
Ingo Molnar 已提交
1327
		return;
1328

I
Ingo Molnar 已提交
1329
	/*
1330
	 * Always use NMI for PMU
I
Ingo Molnar 已提交
1331
	 */
1332
	apic_write(APIC_LVTPC, APIC_DM_NMI);
I
Ingo Molnar 已提交
1333 1334
}

1335 1336 1337 1338 1339 1340 1341
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
1342
static int __kprobes
1343
perf_event_nmi_handler(struct notifier_block *self,
I
Ingo Molnar 已提交
1344 1345 1346
			 unsigned long cmd, void *__args)
{
	struct die_args *args = __args;
1347 1348
	unsigned int this_nmi;
	int handled;
1349

1350
	if (!atomic_read(&active_events))
1351 1352
		return NOTIFY_DONE;

1353 1354 1355
	switch (cmd) {
	case DIE_NMI:
		break;
1356 1357
	case DIE_NMIUNKNOWN:
		this_nmi = percpu_read(irq_stat.__nmi_count);
T
Tejun Heo 已提交
1358
		if (this_nmi != __this_cpu_read(pmu_nmi.marked))
1359 1360 1361 1362 1363 1364 1365 1366 1367 1368
			/* 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;
1369
	default:
I
Ingo Molnar 已提交
1370
		return NOTIFY_DONE;
1371
	}
I
Ingo Molnar 已提交
1372 1373

	apic_write(APIC_LVTPC, APIC_DM_NMI);
1374 1375 1376 1377 1378 1379 1380 1381

	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
1382 1383
	    ((__this_cpu_read(pmu_nmi.marked) == this_nmi) &&
	     (__this_cpu_read(pmu_nmi.handled) > 1))) {
1384 1385 1386 1387 1388 1389 1390 1391 1392 1393
		/*
		 * 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
1394 1395
		__this_cpu_write(pmu_nmi.marked, this_nmi + 1);
		__this_cpu_write(pmu_nmi.handled, handled);
1396
	}
I
Ingo Molnar 已提交
1397

1398
	return NOTIFY_STOP;
I
Ingo Molnar 已提交
1399 1400
}

1401 1402 1403
static __read_mostly struct notifier_block perf_event_nmi_notifier = {
	.notifier_call		= perf_event_nmi_handler,
	.next			= NULL,
1404
	.priority		= NMI_LOCAL_LOW_PRIOR,
1405 1406
};

1407
static struct event_constraint unconstrained;
1408
static struct event_constraint emptyconstraint;
1409 1410

static struct event_constraint *
1411
x86_get_event_constraints(struct cpu_hw_events *cpuc, struct perf_event *event)
1412
{
1413
	struct event_constraint *c;
1414 1415 1416

	if (x86_pmu.event_constraints) {
		for_each_event_constraint(c, x86_pmu.event_constraints) {
1417 1418
			if ((event->hw.config & c->cmask) == c->code)
				return c;
1419 1420
		}
	}
1421 1422

	return &unconstrained;
1423 1424
}

1425 1426
#include "perf_event_amd.c"
#include "perf_event_p6.c"
1427
#include "perf_event_p4.c"
1428
#include "perf_event_intel_lbr.c"
1429
#include "perf_event_intel_ds.c"
1430
#include "perf_event_intel.c"
1431

1432 1433 1434 1435
static int __cpuinit
x86_pmu_notifier(struct notifier_block *self, unsigned long action, void *hcpu)
{
	unsigned int cpu = (long)hcpu;
1436
	int ret = NOTIFY_OK;
1437 1438 1439 1440

	switch (action & ~CPU_TASKS_FROZEN) {
	case CPU_UP_PREPARE:
		if (x86_pmu.cpu_prepare)
1441
			ret = x86_pmu.cpu_prepare(cpu);
1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453
		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;

1454
	case CPU_UP_CANCELED:
1455 1456 1457 1458 1459 1460 1461 1462 1463
	case CPU_DEAD:
		if (x86_pmu.cpu_dead)
			x86_pmu.cpu_dead(cpu);
		break;

	default:
		break;
	}

1464
	return ret;
1465 1466
}

1467 1468 1469 1470 1471 1472 1473 1474 1475 1476
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");
}

1477
static int __init init_hw_perf_events(void)
1478
{
1479
	struct event_constraint *c;
1480 1481
	int err;

1482
	pr_info("Performance Events: ");
1483

1484 1485
	switch (boot_cpu_data.x86_vendor) {
	case X86_VENDOR_INTEL:
1486
		err = intel_pmu_init();
1487
		break;
1488
	case X86_VENDOR_AMD:
1489
		err = amd_pmu_init();
1490
		break;
1491
	default:
1492
		return 0;
1493
	}
1494
	if (err != 0) {
1495
		pr_cont("no PMU driver, software events only.\n");
1496
		return 0;
1497
	}
1498

1499 1500
	pmu_check_apic();

1501
	/* sanity check that the hardware exists or is emulated */
1502
	if (!check_hw_exists())
1503
		return 0;
1504

1505
	pr_cont("%s PMU driver.\n", x86_pmu.name);
1506

1507 1508 1509
	if (x86_pmu.quirks)
		x86_pmu.quirks();

1510
	if (x86_pmu.num_counters > X86_PMC_MAX_GENERIC) {
1511
		WARN(1, KERN_ERR "hw perf events %d > max(%d), clipping!",
1512 1513
		     x86_pmu.num_counters, X86_PMC_MAX_GENERIC);
		x86_pmu.num_counters = X86_PMC_MAX_GENERIC;
I
Ingo Molnar 已提交
1514
	}
1515
	x86_pmu.intel_ctrl = (1 << x86_pmu.num_counters) - 1;
I
Ingo Molnar 已提交
1516

1517
	if (x86_pmu.num_counters_fixed > X86_PMC_MAX_FIXED) {
1518
		WARN(1, KERN_ERR "hw perf events fixed %d > max(%d), clipping!",
1519 1520
		     x86_pmu.num_counters_fixed, X86_PMC_MAX_FIXED);
		x86_pmu.num_counters_fixed = X86_PMC_MAX_FIXED;
1521
	}
1522

1523
	x86_pmu.intel_ctrl |=
1524
		((1LL << x86_pmu.num_counters_fixed)-1) << X86_PMC_IDX_FIXED;
I
Ingo Molnar 已提交
1525

1526 1527
	perf_events_lapic_init();
	register_die_notifier(&perf_event_nmi_notifier);
1528

1529
	unconstrained = (struct event_constraint)
1530 1531
		__EVENT_CONSTRAINT(0, (1ULL << x86_pmu.num_counters) - 1,
				   0, x86_pmu.num_counters);
1532

1533 1534
	if (x86_pmu.event_constraints) {
		for_each_event_constraint(c, x86_pmu.event_constraints) {
1535
			if (c->cmask != X86_RAW_EVENT_MASK)
1536 1537
				continue;

1538 1539
			c->idxmsk64 |= (1ULL << x86_pmu.num_counters) - 1;
			c->weight += x86_pmu.num_counters;
1540 1541 1542
		}
	}

I
Ingo Molnar 已提交
1543
	pr_info("... version:                %d\n",     x86_pmu.version);
1544 1545 1546
	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
1547
	pr_info("... max period:             %016Lx\n", x86_pmu.max_period);
1548
	pr_info("... fixed-purpose events:   %d\n",     x86_pmu.num_counters_fixed);
1549
	pr_info("... event mask:             %016Lx\n", x86_pmu.intel_ctrl);
1550

P
Peter Zijlstra 已提交
1551
	perf_pmu_register(&pmu, "cpu", PERF_TYPE_RAW);
1552
	perf_cpu_notifier(x86_pmu_notifier);
1553 1554

	return 0;
I
Ingo Molnar 已提交
1555
}
1556
early_initcall(init_hw_perf_events);
I
Ingo Molnar 已提交
1557

1558
static inline void x86_pmu_read(struct perf_event *event)
1559
{
1560
	x86_perf_event_update(event);
1561 1562
}

1563 1564 1565 1566 1567
/*
 * 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
1568
static void x86_pmu_start_txn(struct pmu *pmu)
1569
{
P
Peter Zijlstra 已提交
1570
	perf_pmu_disable(pmu);
T
Tejun Heo 已提交
1571 1572
	__this_cpu_or(cpu_hw_events.group_flag, PERF_EVENT_TXN);
	__this_cpu_write(cpu_hw_events.n_txn, 0);
1573 1574 1575 1576 1577 1578 1579
}

/*
 * Stop group events scheduling transaction
 * Clear the flag and pmu::enable() will perform the
 * schedulability test.
 */
P
Peter Zijlstra 已提交
1580
static void x86_pmu_cancel_txn(struct pmu *pmu)
1581
{
T
Tejun Heo 已提交
1582
	__this_cpu_and(cpu_hw_events.group_flag, ~PERF_EVENT_TXN);
1583 1584 1585
	/*
	 * Truncate the collected events.
	 */
T
Tejun Heo 已提交
1586 1587
	__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
1588
	perf_pmu_enable(pmu);
1589 1590 1591 1592 1593 1594 1595
}

/*
 * Commit group events scheduling transaction
 * Perform the group schedulability test as a whole
 * Return 0 if success
 */
P
Peter Zijlstra 已提交
1596
static int x86_pmu_commit_txn(struct pmu *pmu)
1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613 1614 1615 1616
{
	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));

1617
	cpuc->group_flag &= ~PERF_EVENT_TXN;
P
Peter Zijlstra 已提交
1618
	perf_pmu_enable(pmu);
1619 1620 1621
	return 0;
}

1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647
/*
 * 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;
}

1648 1649 1650 1651
/*
 * validate a single event group
 *
 * validation include:
1652 1653 1654
 *	- check events are compatible which each other
 *	- events do not compete for the same counter
 *	- number of events <= number of counters
1655 1656 1657 1658
 *
 * validation ensures the group can be loaded onto the
 * PMU if it was the only group available.
 */
1659 1660
static int validate_group(struct perf_event *event)
{
1661
	struct perf_event *leader = event->group_leader;
1662 1663
	struct cpu_hw_events *fake_cpuc;
	int ret, n;
1664

1665 1666 1667 1668
	ret = -ENOMEM;
	fake_cpuc = kmalloc(sizeof(*fake_cpuc), GFP_KERNEL | __GFP_ZERO);
	if (!fake_cpuc)
		goto out;
1669

1670 1671 1672 1673 1674 1675
	/*
	 * 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
	 */
1676 1677
	ret = -ENOSPC;
	n = collect_events(fake_cpuc, leader, true);
1678
	if (n < 0)
1679
		goto out_free;
1680

1681 1682
	fake_cpuc->n_events = n;
	n = collect_events(fake_cpuc, event, false);
1683
	if (n < 0)
1684
		goto out_free;
1685

1686
	fake_cpuc->n_events = n;
1687

1688
	ret = x86_pmu.schedule_events(fake_cpuc, n, NULL);
1689 1690 1691 1692 1693

out_free:
	kfree(fake_cpuc);
out:
	return ret;
1694 1695
}

1696
static int x86_pmu_event_init(struct perf_event *event)
I
Ingo Molnar 已提交
1697
{
P
Peter Zijlstra 已提交
1698
	struct pmu *tmp;
I
Ingo Molnar 已提交
1699 1700
	int err;

1701 1702 1703 1704 1705 1706 1707 1708 1709 1710 1711
	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);
1712
	if (!err) {
1713 1714 1715 1716 1717 1718 1719 1720
		/*
		 * 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;

1721 1722
		if (event->group_leader != event)
			err = validate_group(event);
1723 1724
		else
			err = validate_event(event);
1725 1726

		event->pmu = tmp;
1727
	}
1728
	if (err) {
1729 1730
		if (event->destroy)
			event->destroy(event);
1731
	}
I
Ingo Molnar 已提交
1732

1733
	return err;
I
Ingo Molnar 已提交
1734
}
1735

1736
static struct pmu pmu = {
P
Peter Zijlstra 已提交
1737 1738 1739
	.pmu_enable	= x86_pmu_enable,
	.pmu_disable	= x86_pmu_disable,

1740
	.event_init	= x86_pmu_event_init,
P
Peter Zijlstra 已提交
1741 1742 1743

	.add		= x86_pmu_add,
	.del		= x86_pmu_del,
1744 1745 1746
	.start		= x86_pmu_start,
	.stop		= x86_pmu_stop,
	.read		= x86_pmu_read,
P
Peter Zijlstra 已提交
1747

1748 1749 1750 1751 1752
	.start_txn	= x86_pmu_start_txn,
	.cancel_txn	= x86_pmu_cancel_txn,
	.commit_txn	= x86_pmu_commit_txn,
};

1753 1754 1755 1756 1757 1758 1759 1760 1761 1762 1763 1764 1765 1766 1767 1768 1769
/*
 * 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)
{
1770
	return 0;
1771 1772 1773 1774 1775 1776
}

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

1777
	perf_callchain_store(entry, addr);
1778 1779 1780 1781 1782 1783 1784
}

static const struct stacktrace_ops backtrace_ops = {
	.warning		= backtrace_warning,
	.warning_symbol		= backtrace_warning_symbol,
	.stack			= backtrace_stack,
	.address		= backtrace_address,
1785
	.walk_stack		= print_context_stack_bp,
1786 1787
};

1788 1789
void
perf_callchain_kernel(struct perf_callchain_entry *entry, struct pt_regs *regs)
1790
{
1791 1792
	if (perf_guest_cbs && perf_guest_cbs->is_in_guest()) {
		/* TODO: We don't support guest os callchain now */
1793
		return;
1794 1795
	}

1796
	perf_callchain_store(entry, regs->ip);
1797

1798
	dump_trace(NULL, regs, NULL, 0, &backtrace_ops, entry);
1799 1800
}

1801 1802 1803
#ifdef CONFIG_COMPAT
static inline int
perf_callchain_user32(struct pt_regs *regs, struct perf_callchain_entry *entry)
1804
{
1805 1806 1807
	/* 32-bit process in 64-bit kernel. */
	struct stack_frame_ia32 frame;
	const void __user *fp;
1808

1809 1810 1811 1812 1813 1814 1815 1816 1817 1818 1819 1820
	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;
1821

1822 1823
		if (fp < compat_ptr(regs->sp))
			break;
1824

1825
		perf_callchain_store(entry, frame.return_address);
1826 1827 1828
		fp = compat_ptr(frame.next_frame);
	}
	return 1;
1829
}
1830 1831 1832 1833 1834 1835 1836
#else
static inline int
perf_callchain_user32(struct pt_regs *regs, struct perf_callchain_entry *entry)
{
    return 0;
}
#endif
1837

1838 1839
void
perf_callchain_user(struct perf_callchain_entry *entry, struct pt_regs *regs)
1840 1841 1842 1843
{
	struct stack_frame frame;
	const void __user *fp;

1844 1845
	if (perf_guest_cbs && perf_guest_cbs->is_in_guest()) {
		/* TODO: We don't support guest os callchain now */
1846
		return;
1847
	}
1848

1849
	fp = (void __user *)regs->bp;
1850

1851
	perf_callchain_store(entry, regs->ip);
1852

1853 1854 1855
	if (perf_callchain_user32(regs, entry))
		return;

1856
	while (entry->nr < PERF_MAX_STACK_DEPTH) {
1857
		unsigned long bytes;
1858
		frame.next_frame	     = NULL;
1859 1860
		frame.return_address = 0;

1861 1862
		bytes = copy_from_user_nmi(&frame, fp, sizeof(frame));
		if (bytes != sizeof(frame))
1863 1864
			break;

1865
		if ((unsigned long)fp < regs->sp)
1866 1867
			break;

1868
		perf_callchain_store(entry, frame.return_address);
1869
		fp = frame.next_frame;
1870 1871 1872
	}
}

1873 1874 1875
unsigned long perf_instruction_pointer(struct pt_regs *regs)
{
	unsigned long ip;
1876

1877 1878 1879 1880
	if (perf_guest_cbs && perf_guest_cbs->is_in_guest())
		ip = perf_guest_cbs->get_guest_ip();
	else
		ip = instruction_pointer(regs);
1881

1882 1883 1884 1885 1886 1887
	return ip;
}

unsigned long perf_misc_flags(struct pt_regs *regs)
{
	int misc = 0;
1888

1889
	if (perf_guest_cbs && perf_guest_cbs->is_in_guest()) {
1890 1891 1892 1893 1894 1895 1896 1897 1898 1899 1900
		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;
	}

1901
	if (regs->flags & PERF_EFLAGS_EXACT)
P
Peter Zijlstra 已提交
1902
		misc |= PERF_RECORD_MISC_EXACT_IP;
1903 1904 1905

	return misc;
}