[deliverable/linux.git] / arch / x86 / kernel / cpu / perf_counter.c

/*
 * Performance counter x86 architecture code
 *
 *  Copyright(C) 2008 Thomas Gleixner <tglx@linutronix.de>
 *  Copyright(C) 2008 Red Hat, Inc., Ingo Molnar
 *
 *  For licencing details see kernel-base/COPYING
 */

#include <linux/perf_counter.h>
#include <linux/capability.h>
#include <linux/notifier.h>
#include <linux/hardirq.h>
#include <linux/kprobes.h>
#include <linux/module.h>
#include <linux/kdebug.h>
#include <linux/sched.h>

#include <asm/intel_arch_perfmon.h>
#include <asm/apic.h>

static bool perf_counters_initialized __read_mostly;

/*
 * Number of (generic) HW counters:
 */
static int nr_hw_counters __read_mostly;
static u32 perf_counter_mask __read_mostly;

/* No support for fixed function counters yet */

#define MAX_HW_COUNTERS		8

struct cpu_hw_counters {
	struct perf_counter	*counters[MAX_HW_COUNTERS];
	unsigned long		used[BITS_TO_LONGS(MAX_HW_COUNTERS)];
};

/*
 * Intel PerfMon v3. Used on Core2 and later.
 */
static DEFINE_PER_CPU(struct cpu_hw_counters, cpu_hw_counters);

const int intel_perfmon_event_map[] =
{
  [PERF_COUNT_CYCLES]			= 0x003c,
  [PERF_COUNT_INSTRUCTIONS]		= 0x00c0,
  [PERF_COUNT_CACHE_REFERENCES]		= 0x4f2e,
  [PERF_COUNT_CACHE_MISSES]		= 0x412e,
  [PERF_COUNT_BRANCH_INSTRUCTIONS]	= 0x00c4,
  [PERF_COUNT_BRANCH_MISSES]		= 0x00c5,
};

const int max_intel_perfmon_events = ARRAY_SIZE(intel_perfmon_event_map);

/*
 * Setup the hardware configuration for a given hw_event_type
 */
static int __hw_perf_counter_init(struct perf_counter *counter)
{
	struct perf_counter_hw_event *hw_event = &counter->hw_event;
	struct hw_perf_counter *hwc = &counter->hw;

	if (unlikely(!perf_counters_initialized))
		return -EINVAL;

	/*
	 * Count user events, and generate PMC IRQs:
	 * (keep 'enabled' bit clear for now)
	 */
	hwc->config = ARCH_PERFMON_EVENTSEL_USR | ARCH_PERFMON_EVENTSEL_INT;

	/*
	 * If privileged enough, count OS events too, and allow
	 * NMI events as well:
	 */
	hwc->nmi = 0;
	if (capable(CAP_SYS_ADMIN)) {
		hwc->config |= ARCH_PERFMON_EVENTSEL_OS;
		if (hw_event->nmi)
			hwc->nmi = 1;
	}

	hwc->config_base	= MSR_ARCH_PERFMON_EVENTSEL0;
	hwc->counter_base	= MSR_ARCH_PERFMON_PERFCTR0;

	hwc->irq_period		= hw_event->irq_period;
	/*
	 * Intel PMCs cannot be accessed sanely above 32 bit width,
	 * so we install an artificial 1<<31 period regardless of
	 * the generic counter period:
	 */
	if (!hwc->irq_period)
		hwc->irq_period = 0x7FFFFFFF;

	hwc->next_count	= -(s32)hwc->irq_period;

	/*
	 * Raw event type provide the config in the event structure
	 */
	if (hw_event->raw) {
		hwc->config |= hw_event->type;
	} else {
		if (hw_event->type >= max_intel_perfmon_events)
			return -EINVAL;
		/*
		 * The generic map:
		 */
		hwc->config |= intel_perfmon_event_map[hw_event->type];
	}
	counter->wakeup_pending = 0;

	return 0;
}

void hw_perf_enable_all(void)
{
	wrmsr(MSR_CORE_PERF_GLOBAL_CTRL, perf_counter_mask, 0);
}

void hw_perf_restore(u64 ctrl)
{
	wrmsr(MSR_CORE_PERF_GLOBAL_CTRL, ctrl, 0);
}
EXPORT_SYMBOL_GPL(hw_perf_restore);

u64 hw_perf_save_disable(void)
{
	u64 ctrl;

	rdmsrl(MSR_CORE_PERF_GLOBAL_CTRL, ctrl);
	wrmsr(MSR_CORE_PERF_GLOBAL_CTRL, 0, 0);
	return ctrl;
}
EXPORT_SYMBOL_GPL(hw_perf_save_disable);

static inline void
__x86_perf_counter_disable(struct hw_perf_counter *hwc, unsigned int idx)
{
	wrmsr(hwc->config_base + idx, hwc->config, 0);
}

static DEFINE_PER_CPU(u64, prev_next_count[MAX_HW_COUNTERS]);

static void __hw_perf_counter_set_period(struct hw_perf_counter *hwc, int idx)
{
	per_cpu(prev_next_count[idx], smp_processor_id()) = hwc->next_count;

	wrmsr(hwc->counter_base + idx, hwc->next_count, 0);
}

static void __x86_perf_counter_enable(struct hw_perf_counter *hwc, int idx)
{
	wrmsr(hwc->config_base + idx,
	      hwc->config | ARCH_PERFMON_EVENTSEL0_ENABLE, 0);
}

static void x86_perf_counter_enable(struct perf_counter *counter)
{
	struct cpu_hw_counters *cpuc = &__get_cpu_var(cpu_hw_counters);
	struct hw_perf_counter *hwc = &counter->hw;
	int idx = hwc->idx;

	/* Try to get the previous counter again */
	if (test_and_set_bit(idx, cpuc->used)) {
		idx = find_first_zero_bit(cpuc->used, nr_hw_counters);
		set_bit(idx, cpuc->used);
		hwc->idx = idx;
	}

	perf_counters_lapic_init(hwc->nmi);

	__x86_perf_counter_disable(hwc, idx);

	cpuc->counters[idx] = counter;

	__hw_perf_counter_set_period(hwc, idx);
	__x86_perf_counter_enable(hwc, idx);
}

static void __hw_perf_save_counter(struct perf_counter *counter,
				   struct hw_perf_counter *hwc, int idx)
{
	s64 raw = -1;
	s64 delta;

	/*
	 * Get the raw hw counter value:
	 */
	rdmsrl(hwc->counter_base + idx, raw);

	/*
	 * Rebase it to zero (it started counting at -irq_period),
	 * to see the delta since ->prev_count:
	 */
	delta = (s64)hwc->irq_period + (s64)(s32)raw;

	atomic64_counter_set(counter, hwc->prev_count + delta);

	/*
	 * Adjust the ->prev_count offset - if we went beyond
	 * irq_period of units, then we got an IRQ and the counter
	 * was set back to -irq_period:
	 */
	while (delta >= (s64)hwc->irq_period) {
		hwc->prev_count += hwc->irq_period;
		delta -= (s64)hwc->irq_period;
	}

	/*
	 * Calculate the next raw counter value we'll write into
	 * the counter at the next sched-in time:
	 */
	delta -= (s64)hwc->irq_period;

	hwc->next_count = (s32)delta;
}

void perf_counter_print_debug(void)
{
	u64 ctrl, status, overflow, pmc_ctrl, pmc_count, next_count;
	int cpu, idx;

	if (!nr_hw_counters)
		return;

	local_irq_disable();

	cpu = smp_processor_id();

	rdmsrl(MSR_CORE_PERF_GLOBAL_CTRL, ctrl);
	rdmsrl(MSR_CORE_PERF_GLOBAL_STATUS, status);
	rdmsrl(MSR_CORE_PERF_GLOBAL_OVF_CTRL, overflow);

	printk(KERN_INFO "\n");
	printk(KERN_INFO "CPU#%d: ctrl:       %016llx\n", cpu, ctrl);
	printk(KERN_INFO "CPU#%d: status:     %016llx\n", cpu, status);
	printk(KERN_INFO "CPU#%d: overflow:   %016llx\n", cpu, overflow);

	for (idx = 0; idx < nr_hw_counters; idx++) {
		rdmsrl(MSR_ARCH_PERFMON_EVENTSEL0 + idx, pmc_ctrl);
		rdmsrl(MSR_ARCH_PERFMON_PERFCTR0  + idx, pmc_count);

		next_count = per_cpu(prev_next_count[idx], cpu);

		printk(KERN_INFO "CPU#%d: PMC%d ctrl:  %016llx\n",
			cpu, idx, pmc_ctrl);
		printk(KERN_INFO "CPU#%d: PMC%d count: %016llx\n",
			cpu, idx, pmc_count);
		printk(KERN_INFO "CPU#%d: PMC%d next:  %016llx\n",
			cpu, idx, next_count);
	}
	local_irq_enable();
}

static void x86_perf_counter_disable(struct perf_counter *counter)
{
	struct cpu_hw_counters *cpuc = &__get_cpu_var(cpu_hw_counters);
	struct hw_perf_counter *hwc = &counter->hw;
	unsigned int idx = hwc->idx;

	__x86_perf_counter_disable(hwc, idx);

	clear_bit(idx, cpuc->used);
	cpuc->counters[idx] = NULL;
	__hw_perf_save_counter(counter, hwc, idx);
}

static void x86_perf_counter_read(struct perf_counter *counter)
{
	struct hw_perf_counter *hwc = &counter->hw;
	unsigned long addr = hwc->counter_base + hwc->idx;
	s64 offs, val = -1LL;
	s32 val32;

	/* Careful: NMI might modify the counter offset */
	do {
		offs = hwc->prev_count;
		rdmsrl(addr, val);
	} while (offs != hwc->prev_count);

	val32 = (s32) val;
	val = (s64)hwc->irq_period + (s64)val32;
	atomic64_counter_set(counter, hwc->prev_count + val);
}

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

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

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

/*
 * NMI-safe enable method:
 */
static void perf_save_and_restart(struct perf_counter *counter)
{
	struct hw_perf_counter *hwc = &counter->hw;
	int idx = hwc->idx;
	u64 pmc_ctrl;

	rdmsrl(MSR_ARCH_PERFMON_EVENTSEL0 + idx, pmc_ctrl);

	__hw_perf_save_counter(counter, hwc, idx);
	__hw_perf_counter_set_period(hwc, idx);

	if (pmc_ctrl & ARCH_PERFMON_EVENTSEL0_ENABLE)
		__x86_perf_counter_enable(hwc, idx);
}

static void
perf_handle_group(struct perf_counter *sibling, u64 *status, u64 *overflown)
{
	struct perf_counter *counter, *group_leader = sibling->group_leader;
	int bit;

	/*
	 * Store the counter's own timestamp first:
	 */
	perf_store_irq_data(sibling, sibling->hw_event.type);
	perf_store_irq_data(sibling, atomic64_counter_read(sibling));

	/*
	 * Then store sibling timestamps (if any):
	 */
	list_for_each_entry(counter, &group_leader->sibling_list, list_entry) {
		if (counter->state != PERF_COUNTER_STATE_ACTIVE) {
			/*
			 * When counter was not in the overflow mask, we have to
			 * read it from hardware. We read it as well, when it
			 * has not been read yet and clear the bit in the
			 * status mask.
			 */
			bit = counter->hw.idx;
			if (!test_bit(bit, (unsigned long *) overflown) ||
			    test_bit(bit, (unsigned long *) status)) {
				clear_bit(bit, (unsigned long *) status);
				perf_save_and_restart(counter);
			}
		}
		perf_store_irq_data(sibling, counter->hw_event.type);
		perf_store_irq_data(sibling, atomic64_counter_read(counter));
	}
}

/*
 * This handler is triggered by the local APIC, so the APIC IRQ handling
 * rules apply:
 */
static void __smp_perf_counter_interrupt(struct pt_regs *regs, int nmi)
{
	int bit, cpu = smp_processor_id();
	u64 ack, status, saved_global;
	struct cpu_hw_counters *cpuc;

	rdmsrl(MSR_CORE_PERF_GLOBAL_CTRL, saved_global);

	/* Disable counters globally */
	wrmsr(MSR_CORE_PERF_GLOBAL_CTRL, 0, 0);
	ack_APIC_irq();

	cpuc = &per_cpu(cpu_hw_counters, cpu);

	rdmsrl(MSR_CORE_PERF_GLOBAL_STATUS, status);
	if (!status)
		goto out;

again:
	ack = status;
	for_each_bit(bit, (unsigned long *) &status, nr_hw_counters) {
		struct perf_counter *counter = cpuc->counters[bit];

		clear_bit(bit, (unsigned long *) &status);
		if (!counter)
			continue;

		perf_save_and_restart(counter);

		switch (counter->hw_event.record_type) {
		case PERF_RECORD_SIMPLE:
			continue;
		case PERF_RECORD_IRQ:
			perf_store_irq_data(counter, instruction_pointer(regs));
			break;
		case PERF_RECORD_GROUP:
			perf_handle_group(counter, &status, &ack);
			break;
		}
		/*
		 * From NMI context we cannot call into the scheduler to
		 * do a task wakeup - but we mark these counters as
		 * wakeup_pending and initate a wakeup callback:
		 */
		if (nmi) {
			counter->wakeup_pending = 1;
			set_tsk_thread_flag(current, TIF_PERF_COUNTERS);
		} else {
			wake_up(&counter->waitq);
		}
	}

	wrmsr(MSR_CORE_PERF_GLOBAL_OVF_CTRL, ack, 0);

	/*
	 * Repeat if there is more work to be done:
	 */
	rdmsrl(MSR_CORE_PERF_GLOBAL_STATUS, status);
	if (status)
		goto again;
out:
	/*
	 * Restore - do not reenable when global enable is off:
	 */
	wrmsr(MSR_CORE_PERF_GLOBAL_CTRL, saved_global, 0);
}

void smp_perf_counter_interrupt(struct pt_regs *regs)
{
	irq_enter();
	inc_irq_stat(apic_perf_irqs);
	apic_write(APIC_LVTPC, LOCAL_PERF_VECTOR);
	__smp_perf_counter_interrupt(regs, 0);

	irq_exit();
}

/*
 * This handler is triggered by NMI contexts:
 */
void perf_counter_notify(struct pt_regs *regs)
{
	struct cpu_hw_counters *cpuc;
	unsigned long flags;
	int bit, cpu;

	local_irq_save(flags);
	cpu = smp_processor_id();
	cpuc = &per_cpu(cpu_hw_counters, cpu);

	for_each_bit(bit, cpuc->used, nr_hw_counters) {
		struct perf_counter *counter = cpuc->counters[bit];

		if (!counter)
			continue;

		if (counter->wakeup_pending) {
			counter->wakeup_pending = 0;
			wake_up(&counter->waitq);
		}
	}

	local_irq_restore(flags);
}

void __cpuinit perf_counters_lapic_init(int nmi)
{
	u32 apic_val;

	if (!perf_counters_initialized)
		return;
	/*
	 * Enable the performance counter vector in the APIC LVT:
	 */
	apic_val = apic_read(APIC_LVTERR);

	apic_write(APIC_LVTERR, apic_val | APIC_LVT_MASKED);
	if (nmi)
		apic_write(APIC_LVTPC, APIC_DM_NMI);
	else
		apic_write(APIC_LVTPC, LOCAL_PERF_VECTOR);
	apic_write(APIC_LVTERR, apic_val);
}

static int __kprobes
perf_counter_nmi_handler(struct notifier_block *self,
			 unsigned long cmd, void *__args)
{
	struct die_args *args = __args;
	struct pt_regs *regs;

	if (likely(cmd != DIE_NMI_IPI))
		return NOTIFY_DONE;

	regs = args->regs;

	apic_write(APIC_LVTPC, APIC_DM_NMI);
	__smp_perf_counter_interrupt(regs, 1);

	return NOTIFY_STOP;
}

static __read_mostly struct notifier_block perf_counter_nmi_notifier = {
	.notifier_call		= perf_counter_nmi_handler
};

void __init init_hw_perf_counters(void)
{
	union cpuid10_eax eax;
	unsigned int unused;
	unsigned int ebx;

	if (!cpu_has(&boot_cpu_data, X86_FEATURE_ARCH_PERFMON))
		return;

	/*
	 * Check whether the Architectural PerfMon supports
	 * Branch Misses Retired Event or not.
	 */
	cpuid(10, &(eax.full), &ebx, &unused, &unused);
	if (eax.split.mask_length <= ARCH_PERFMON_BRANCH_MISSES_RETIRED)
		return;

	printk(KERN_INFO "Intel Performance Monitoring support detected.\n");

	printk(KERN_INFO "... version:      %d\n", eax.split.version_id);
	printk(KERN_INFO "... num_counters: %d\n", eax.split.num_counters);
	nr_hw_counters = eax.split.num_counters;
	if (nr_hw_counters > MAX_HW_COUNTERS) {
		nr_hw_counters = MAX_HW_COUNTERS;
		WARN(1, KERN_ERR "hw perf counters %d > max(%d), clipping!",
			nr_hw_counters, MAX_HW_COUNTERS);
	}
	perf_counter_mask = (1 << nr_hw_counters) - 1;
	perf_max_counters = nr_hw_counters;

	printk(KERN_INFO "... bit_width:    %d\n", eax.split.bit_width);
	printk(KERN_INFO "... mask_length:  %d\n", eax.split.mask_length);

	perf_counters_lapic_init(0);
	register_die_notifier(&perf_counter_nmi_notifier);

	perf_counters_initialized = true;
}

static const struct hw_perf_counter_ops x86_perf_counter_ops = {
	.hw_perf_counter_enable		= x86_perf_counter_enable,
	.hw_perf_counter_disable	= x86_perf_counter_disable,
	.hw_perf_counter_read		= x86_perf_counter_read,
};

const struct hw_perf_counter_ops *
hw_perf_counter_init(struct perf_counter *counter)
{
	int err;

	err = __hw_perf_counter_init(counter);
	if (err)
		return NULL;

	return &x86_perf_counter_ops;
}
Commit	Line	Data
241771ef IM	1	/*
	2	* Performance counter x86 architecture code
	3	*
	4	* Copyright(C) 2008 Thomas Gleixner <tglx@linutronix.de>
	5	* Copyright(C) 2008 Red Hat, Inc., Ingo Molnar
	6	*
	7	* For licencing details see kernel-base/COPYING
	8	*/
	9
	10	#include <linux/perf_counter.h>
	11	#include <linux/capability.h>
	12	#include <linux/notifier.h>
	13	#include <linux/hardirq.h>
	14	#include <linux/kprobes.h>
4ac13294	15	#include <linux/module.h>
241771ef IM	16	#include <linux/kdebug.h>
	17	#include <linux/sched.h>
	18
	19	#include <asm/intel_arch_perfmon.h>
	20	#include <asm/apic.h>
	21
	22	static bool perf_counters_initialized __read_mostly;
	23
	24	/*
	25	* Number of (generic) HW counters:
	26	*/
	27	static int nr_hw_counters __read_mostly;
	28	static u32 perf_counter_mask __read_mostly;
	29
	30	/* No support for fixed function counters yet */
	31
	32	#define MAX_HW_COUNTERS 8
	33
	34	struct cpu_hw_counters {
	35	struct perf_counter *counters[MAX_HW_COUNTERS];
	36	unsigned long used[BITS_TO_LONGS(MAX_HW_COUNTERS)];
241771ef IM	37	};
	38
	39	/*
	40	* Intel PerfMon v3. Used on Core2 and later.
	41	*/
	42	static DEFINE_PER_CPU(struct cpu_hw_counters, cpu_hw_counters);
	43
	44	const int intel_perfmon_event_map[] =
	45	{
	46	[PERF_COUNT_CYCLES] = 0x003c,
	47	[PERF_COUNT_INSTRUCTIONS] = 0x00c0,
	48	[PERF_COUNT_CACHE_REFERENCES] = 0x4f2e,
	49	[PERF_COUNT_CACHE_MISSES] = 0x412e,
	50	[PERF_COUNT_BRANCH_INSTRUCTIONS] = 0x00c4,
	51	[PERF_COUNT_BRANCH_MISSES] = 0x00c5,
	52	};
	53
	54	const int max_intel_perfmon_events = ARRAY_SIZE(intel_perfmon_event_map);
	55
	56	/*
	57	* Setup the hardware configuration for a given hw_event_type
	58	*/
621a01ea	59	static int __hw_perf_counter_init(struct perf_counter *counter)
241771ef	60	{
9f66a381	61	struct perf_counter_hw_event *hw_event = &counter->hw_event;
241771ef IM	62	struct hw_perf_counter *hwc = &counter->hw;
	63
	64	if (unlikely(!perf_counters_initialized))
	65	return -EINVAL;
	66
	67	/*
	68	* Count user events, and generate PMC IRQs:
	69	* (keep 'enabled' bit clear for now)
	70	*/
	71	hwc->config = ARCH_PERFMON_EVENTSEL_USR \| ARCH_PERFMON_EVENTSEL_INT;
	72
	73	/*
	74	* If privileged enough, count OS events too, and allow
	75	* NMI events as well:
	76	*/
	77	hwc->nmi = 0;
	78	if (capable(CAP_SYS_ADMIN)) {
	79	hwc->config \|= ARCH_PERFMON_EVENTSEL_OS;
9f66a381	80	if (hw_event->nmi)
241771ef IM	81	hwc->nmi = 1;
	82	}
	83
9f66a381 IM	84	hwc->config_base = MSR_ARCH_PERFMON_EVENTSEL0;
9f66a381 IM	85	hwc->counter_base = MSR_ARCH_PERFMON_PERFCTR0;
241771ef	86
9f66a381	87	hwc->irq_period = hw_event->irq_period;
241771ef IM	88	/*
	89	* Intel PMCs cannot be accessed sanely above 32 bit width,
	90	* so we install an artificial 1<<31 period regardless of
	91	* the generic counter period:
	92	*/
	93	if (!hwc->irq_period)
	94	hwc->irq_period = 0x7FFFFFFF;
	95
9f66a381	96	hwc->next_count = -(s32)hwc->irq_period;
241771ef IM	97
241771ef IM	98	/*
dfa7c899	99	* Raw event type provide the config in the event structure
241771ef	100	*/
9f66a381 IM	101	if (hw_event->raw) {
9f66a381 IM	102	hwc->config \|= hw_event->type;
241771ef	103	} else {
9f66a381	104	if (hw_event->type >= max_intel_perfmon_events)
241771ef IM	105	return -EINVAL;
	106	/*
	107	* The generic map:
	108	*/
9f66a381	109	hwc->config \|= intel_perfmon_event_map[hw_event->type];
241771ef	110	}
241771ef IM	111	counter->wakeup_pending = 0;
	112
	113	return 0;
	114	}
	115
241771ef IM	116	void hw_perf_enable_all(void)
241771ef IM	117	{
43874d23	118	wrmsr(MSR_CORE_PERF_GLOBAL_CTRL, perf_counter_mask, 0);
241771ef IM	119	}
241771ef IM	120
01b2838c	121	void hw_perf_restore(u64 ctrl)
241771ef	122	{
4ac13294 TG	123	wrmsr(MSR_CORE_PERF_GLOBAL_CTRL, ctrl, 0);
4ac13294 TG	124	}
01b2838c	125	EXPORT_SYMBOL_GPL(hw_perf_restore);
4ac13294	126
01b2838c	127	u64 hw_perf_save_disable(void)
4ac13294 TG	128	{
	129	u64 ctrl;
	130
	131	rdmsrl(MSR_CORE_PERF_GLOBAL_CTRL, ctrl);
241771ef	132	wrmsr(MSR_CORE_PERF_GLOBAL_CTRL, 0, 0);
4ac13294	133	return ctrl;
241771ef	134	}
01b2838c	135	EXPORT_SYMBOL_GPL(hw_perf_save_disable);
241771ef	136
7e2ae347	137	static inline void
621a01ea	138	__x86_perf_counter_disable(struct hw_perf_counter *hwc, unsigned int idx)
7e2ae347 IM	139	{
	140	wrmsr(hwc->config_base + idx, hwc->config, 0);
	141	}
	142
241771ef IM	143	static DEFINE_PER_CPU(u64, prev_next_count[MAX_HW_COUNTERS]);
241771ef IM	144
7e2ae347	145	static void __hw_perf_counter_set_period(struct hw_perf_counter *hwc, int idx)
241771ef IM	146	{
	147	per_cpu(prev_next_count[idx], smp_processor_id()) = hwc->next_count;
	148
	149	wrmsr(hwc->counter_base + idx, hwc->next_count, 0);
7e2ae347 IM	150	}
7e2ae347 IM	151
621a01ea	152	static void __x86_perf_counter_enable(struct hw_perf_counter *hwc, int idx)
7e2ae347 IM	153	{
	154	wrmsr(hwc->config_base + idx,
	155	hwc->config \| ARCH_PERFMON_EVENTSEL0_ENABLE, 0);
241771ef IM	156	}
241771ef IM	157
621a01ea	158	static void x86_perf_counter_enable(struct perf_counter *counter)
241771ef IM	159	{
	160	struct cpu_hw_counters *cpuc = &__get_cpu_var(cpu_hw_counters);
	161	struct hw_perf_counter *hwc = &counter->hw;
	162	int idx = hwc->idx;
	163
	164	/* Try to get the previous counter again */
	165	if (test_and_set_bit(idx, cpuc->used)) {
	166	idx = find_first_zero_bit(cpuc->used, nr_hw_counters);
	167	set_bit(idx, cpuc->used);
	168	hwc->idx = idx;
	169	}
	170
	171	perf_counters_lapic_init(hwc->nmi);
	172
621a01ea	173	__x86_perf_counter_disable(hwc, idx);
241771ef IM	174
241771ef IM	175	cpuc->counters[idx] = counter;
7e2ae347 IM	176
7e2ae347 IM	177	__hw_perf_counter_set_period(hwc, idx);
621a01ea	178	__x86_perf_counter_enable(hwc, idx);
241771ef IM	179	}
241771ef IM	180
241771ef IM	181	static void __hw_perf_save_counter(struct perf_counter *counter,
	182	struct hw_perf_counter *hwc, int idx)
	183	{
	184	s64 raw = -1;
	185	s64 delta;
241771ef IM	186
	187	/*
	188	* Get the raw hw counter value:
	189	*/
1e125676	190	rdmsrl(hwc->counter_base + idx, raw);
241771ef IM	191
	192	/*
	193	* Rebase it to zero (it started counting at -irq_period),
	194	* to see the delta since ->prev_count:
	195	*/
	196	delta = (s64)hwc->irq_period + (s64)(s32)raw;
	197
	198	atomic64_counter_set(counter, hwc->prev_count + delta);
	199
	200	/*
	201	* Adjust the ->prev_count offset - if we went beyond
	202	* irq_period of units, then we got an IRQ and the counter
	203	* was set back to -irq_period:
	204	*/
	205	while (delta >= (s64)hwc->irq_period) {
	206	hwc->prev_count += hwc->irq_period;
	207	delta -= (s64)hwc->irq_period;
	208	}
	209
	210	/*
	211	* Calculate the next raw counter value we'll write into
	212	* the counter at the next sched-in time:
	213	*/
	214	delta -= (s64)hwc->irq_period;
	215
	216	hwc->next_count = (s32)delta;
	217	}
	218
	219	void perf_counter_print_debug(void)
	220	{
	221	u64 ctrl, status, overflow, pmc_ctrl, pmc_count, next_count;
1e125676 IM	222	int cpu, idx;
	223
	224	if (!nr_hw_counters)
	225	return;
241771ef IM	226
	227	local_irq_disable();
	228
	229	cpu = smp_processor_id();
	230
1e125676 IM	231	rdmsrl(MSR_CORE_PERF_GLOBAL_CTRL, ctrl);
	232	rdmsrl(MSR_CORE_PERF_GLOBAL_STATUS, status);
	233	rdmsrl(MSR_CORE_PERF_GLOBAL_OVF_CTRL, overflow);
241771ef IM	234
	235	printk(KERN_INFO "\n");
	236	printk(KERN_INFO "CPU#%d: ctrl: %016llx\n", cpu, ctrl);
	237	printk(KERN_INFO "CPU#%d: status: %016llx\n", cpu, status);
	238	printk(KERN_INFO "CPU#%d: overflow: %016llx\n", cpu, overflow);
	239
	240	for (idx = 0; idx < nr_hw_counters; idx++) {
1e125676 IM	241	rdmsrl(MSR_ARCH_PERFMON_EVENTSEL0 + idx, pmc_ctrl);
1e125676 IM	242	rdmsrl(MSR_ARCH_PERFMON_PERFCTR0 + idx, pmc_count);
241771ef IM	243
	244	next_count = per_cpu(prev_next_count[idx], cpu);
	245
	246	printk(KERN_INFO "CPU#%d: PMC%d ctrl: %016llx\n",
	247	cpu, idx, pmc_ctrl);
	248	printk(KERN_INFO "CPU#%d: PMC%d count: %016llx\n",
	249	cpu, idx, pmc_count);
	250	printk(KERN_INFO "CPU#%d: PMC%d next: %016llx\n",
	251	cpu, idx, next_count);
	252	}
	253	local_irq_enable();
	254	}
	255
621a01ea	256	static void x86_perf_counter_disable(struct perf_counter *counter)
241771ef IM	257	{
	258	struct cpu_hw_counters *cpuc = &__get_cpu_var(cpu_hw_counters);
	259	struct hw_perf_counter *hwc = &counter->hw;
	260	unsigned int idx = hwc->idx;
	261
621a01ea	262	__x86_perf_counter_disable(hwc, idx);
241771ef IM	263
	264	clear_bit(idx, cpuc->used);
	265	cpuc->counters[idx] = NULL;
	266	__hw_perf_save_counter(counter, hwc, idx);
	267	}
	268
621a01ea	269	static void x86_perf_counter_read(struct perf_counter *counter)
241771ef IM	270	{
	271	struct hw_perf_counter *hwc = &counter->hw;
	272	unsigned long addr = hwc->counter_base + hwc->idx;
	273	s64 offs, val = -1LL;
	274	s32 val32;
241771ef IM	275
	276	/* Careful: NMI might modify the counter offset */
	277	do {
	278	offs = hwc->prev_count;
1e125676	279	rdmsrl(addr, val);
241771ef IM	280	} while (offs != hwc->prev_count);
	281
	282	val32 = (s32) val;
5c92d124	283	val = (s64)hwc->irq_period + (s64)val32;
241771ef IM	284	atomic64_counter_set(counter, hwc->prev_count + val);
	285	}
	286
	287	static void perf_store_irq_data(struct perf_counter *counter, u64 data)
	288	{
	289	struct perf_data *irqdata = counter->irqdata;
	290
	291	if (irqdata->len > PERF_DATA_BUFLEN - sizeof(u64)) {
	292	irqdata->overrun++;
	293	} else {
	294	u64 p = (u64 ) &irqdata->data[irqdata->len];
	295
	296	*p = data;
	297	irqdata->len += sizeof(u64);
	298	}
	299	}
	300
7e2ae347 IM	301	/*
	302	* NMI-safe enable method:
	303	*/
241771ef IM	304	static void perf_save_and_restart(struct perf_counter *counter)
	305	{
	306	struct hw_perf_counter *hwc = &counter->hw;
	307	int idx = hwc->idx;
7e2ae347	308	u64 pmc_ctrl;
241771ef	309
1e125676	310	rdmsrl(MSR_ARCH_PERFMON_EVENTSEL0 + idx, pmc_ctrl);
241771ef	311
7e2ae347 IM	312	__hw_perf_save_counter(counter, hwc, idx);
	313	__hw_perf_counter_set_period(hwc, idx);
	314
	315	if (pmc_ctrl & ARCH_PERFMON_EVENTSEL0_ENABLE)
621a01ea	316	__x86_perf_counter_enable(hwc, idx);
241771ef IM	317	}
	318
	319	static void
04289bb9	320	perf_handle_group(struct perf_counter sibling, u64 status, u64 *overflown)
241771ef	321	{
04289bb9	322	struct perf_counter counter, group_leader = sibling->group_leader;
241771ef IM	323	int bit;
241771ef IM	324
04289bb9 IM	325	/*
	326	* Store the counter's own timestamp first:
	327	*/
	328	perf_store_irq_data(sibling, sibling->hw_event.type);
	329	perf_store_irq_data(sibling, atomic64_counter_read(sibling));
241771ef	330
04289bb9 IM	331	/*
	332	* Then store sibling timestamps (if any):
	333	*/
	334	list_for_each_entry(counter, &group_leader->sibling_list, list_entry) {
6a930700	335	if (counter->state != PERF_COUNTER_STATE_ACTIVE) {
241771ef IM	336	/*
	337	* When counter was not in the overflow mask, we have to
	338	* read it from hardware. We read it as well, when it
	339	* has not been read yet and clear the bit in the
	340	* status mask.
	341	*/
	342	bit = counter->hw.idx;
	343	if (!test_bit(bit, (unsigned long *) overflown) \|\|
	344	test_bit(bit, (unsigned long *) status)) {
	345	clear_bit(bit, (unsigned long *) status);
	346	perf_save_and_restart(counter);
	347	}
	348	}
04289bb9 IM	349	perf_store_irq_data(sibling, counter->hw_event.type);
04289bb9 IM	350	perf_store_irq_data(sibling, atomic64_counter_read(counter));
241771ef IM	351	}
	352	}
	353
	354	/*
	355	* This handler is triggered by the local APIC, so the APIC IRQ handling
	356	* rules apply:
	357	*/
	358	static void __smp_perf_counter_interrupt(struct pt_regs *regs, int nmi)
	359	{
	360	int bit, cpu = smp_processor_id();
43874d23	361	u64 ack, status, saved_global;
241771ef	362	struct cpu_hw_counters *cpuc;
43874d23 IM	363
43874d23 IM	364	rdmsrl(MSR_CORE_PERF_GLOBAL_CTRL, saved_global);
241771ef	365
241771ef IM	366	/* Disable counters globally */
	367	wrmsr(MSR_CORE_PERF_GLOBAL_CTRL, 0, 0);
	368	ack_APIC_irq();
	369
	370	cpuc = &per_cpu(cpu_hw_counters, cpu);
	371
87b9cf46 IM	372	rdmsrl(MSR_CORE_PERF_GLOBAL_STATUS, status);
	373	if (!status)
	374	goto out;
	375
241771ef IM	376	again:
	377	ack = status;
	378	for_each_bit(bit, (unsigned long *) &status, nr_hw_counters) {
	379	struct perf_counter *counter = cpuc->counters[bit];
	380
	381	clear_bit(bit, (unsigned long *) &status);
	382	if (!counter)
	383	continue;
	384
	385	perf_save_and_restart(counter);
	386
9f66a381	387	switch (counter->hw_event.record_type) {
241771ef IM	388	case PERF_RECORD_SIMPLE:
	389	continue;
	390	case PERF_RECORD_IRQ:
	391	perf_store_irq_data(counter, instruction_pointer(regs));
	392	break;
	393	case PERF_RECORD_GROUP:
241771ef IM	394	perf_handle_group(counter, &status, &ack);
	395	break;
	396	}
	397	/*
	398	* From NMI context we cannot call into the scheduler to
	399	* do a task wakeup - but we mark these counters as
	400	* wakeup_pending and initate a wakeup callback:
	401	*/
	402	if (nmi) {
	403	counter->wakeup_pending = 1;
	404	set_tsk_thread_flag(current, TIF_PERF_COUNTERS);
	405	} else {
	406	wake_up(&counter->waitq);
	407	}
	408	}
	409
	410	wrmsr(MSR_CORE_PERF_GLOBAL_OVF_CTRL, ack, 0);
	411
	412	/*
	413	* Repeat if there is more work to be done:
	414	*/
	415	rdmsrl(MSR_CORE_PERF_GLOBAL_STATUS, status);
	416	if (status)
	417	goto again;
87b9cf46	418	out:
241771ef	419	/*
43874d23	420	* Restore - do not reenable when global enable is off:
241771ef	421	*/
43874d23	422	wrmsr(MSR_CORE_PERF_GLOBAL_CTRL, saved_global, 0);
241771ef IM	423	}
	424
	425	void smp_perf_counter_interrupt(struct pt_regs *regs)
	426	{
	427	irq_enter();
92bf73e9	428	inc_irq_stat(apic_perf_irqs);
241771ef IM	429	apic_write(APIC_LVTPC, LOCAL_PERF_VECTOR);
	430	__smp_perf_counter_interrupt(regs, 0);
	431
	432	irq_exit();
	433	}
	434
	435	/*
	436	* This handler is triggered by NMI contexts:
	437	*/
	438	void perf_counter_notify(struct pt_regs *regs)
	439	{
	440	struct cpu_hw_counters *cpuc;
	441	unsigned long flags;
	442	int bit, cpu;
	443
	444	local_irq_save(flags);
	445	cpu = smp_processor_id();
	446	cpuc = &per_cpu(cpu_hw_counters, cpu);
	447
	448	for_each_bit(bit, cpuc->used, nr_hw_counters) {
	449	struct perf_counter *counter = cpuc->counters[bit];
	450
	451	if (!counter)
	452	continue;
	453
	454	if (counter->wakeup_pending) {
	455	counter->wakeup_pending = 0;
	456	wake_up(&counter->waitq);
	457	}
	458	}
	459
	460	local_irq_restore(flags);
	461	}
	462
	463	void __cpuinit perf_counters_lapic_init(int nmi)
	464	{
	465	u32 apic_val;
	466
	467	if (!perf_counters_initialized)
	468	return;
	469	/*
	470	* Enable the performance counter vector in the APIC LVT:
	471	*/
	472	apic_val = apic_read(APIC_LVTERR);
	473
	474	apic_write(APIC_LVTERR, apic_val \| APIC_LVT_MASKED);
	475	if (nmi)
	476	apic_write(APIC_LVTPC, APIC_DM_NMI);
	477	else
	478	apic_write(APIC_LVTPC, LOCAL_PERF_VECTOR);
	479	apic_write(APIC_LVTERR, apic_val);
	480	}
	481
	482	static int __kprobes
	483	perf_counter_nmi_handler(struct notifier_block *self,
	484	unsigned long cmd, void *__args)
	485	{
	486	struct die_args *args = __args;
	487	struct pt_regs *regs;
	488
	489	if (likely(cmd != DIE_NMI_IPI))
	490	return NOTIFY_DONE;
	491
	492	regs = args->regs;
493
494	apic_write(APIC_LVTPC, APIC_DM_NMI);
495	__smp_perf_counter_interrupt(regs, 1);
496
497	return NOTIFY_STOP;
498	}
499
500	static __read_mostly struct notifier_block perf_counter_nmi_notifier = {
501	.notifier_call = perf_counter_nmi_handler
502	};
503
504	void __init init_hw_perf_counters(void)
505	{
506	union cpuid10_eax eax;
507	unsigned int unused;
508	unsigned int ebx;
509
510	if (!cpu_has(&boot_cpu_data, X86_FEATURE_ARCH_PERFMON))
511	return;
512
513	/*
514	* Check whether the Architectural PerfMon supports
515	* Branch Misses Retired Event or not.
516	*/
517	cpuid(10, &(eax.full), &ebx, &unused, &unused);
518	if (eax.split.mask_length <= ARCH_PERFMON_BRANCH_MISSES_RETIRED)
519	return;
520
521	printk(KERN_INFO "Intel Performance Monitoring support detected.\n");
522
523	printk(KERN_INFO "... version: %d\n", eax.split.version_id);
524	printk(KERN_INFO "... num_counters: %d\n", eax.split.num_counters);
525	nr_hw_counters = eax.split.num_counters;
526	if (nr_hw_counters > MAX_HW_COUNTERS) {
527	nr_hw_counters = MAX_HW_COUNTERS;
528	WARN(1, KERN_ERR "hw perf counters %d > max(%d), clipping!",
529	nr_hw_counters, MAX_HW_COUNTERS);
530	}
531	perf_counter_mask = (1 << nr_hw_counters) - 1;
532	perf_max_counters = nr_hw_counters;
533
534	printk(KERN_INFO "... bit_width: %d\n", eax.split.bit_width);
535	printk(KERN_INFO "... mask_length: %d\n", eax.split.mask_length);
536
537	perf_counters_lapic_init(0);
538	register_die_notifier(&perf_counter_nmi_notifier);
539
540	perf_counters_initialized = true;
541	}
621a01ea	542
5c92d124	543	static const struct hw_perf_counter_ops x86_perf_counter_ops = {
621a01ea IM	544	.hw_perf_counter_enable = x86_perf_counter_enable,
	545	.hw_perf_counter_disable = x86_perf_counter_disable,
	546	.hw_perf_counter_read = x86_perf_counter_read,
	547	};
	548
5c92d124 IM	549	const struct hw_perf_counter_ops *
5c92d124 IM	550	hw_perf_counter_init(struct perf_counter *counter)
621a01ea IM	551	{
	552	int err;
	553
	554	err = __hw_perf_counter_init(counter);
	555	if (err)
	556	return NULL;
	557
	558	return &x86_perf_counter_ops;
	559	}