[mirror_ubuntu-kernels.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
 */
int hw_perf_counter_init(struct perf_counter *counter)
{
	struct hw_perf_counter *hwc = &counter->hw;
	u32 hw_event_type = counter->event.hw_event_type;

	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_type & PERF_COUNT_NMI)
			hwc->nmi = 1;
	}

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

	hwc->irq_period = counter->event.hw_event_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
	 */
	hw_event_type &= ~PERF_COUNT_NMI;
	if (hw_event_type == PERF_COUNT_RAW) {
		hwc->config |= counter->event.hw_raw_ctrl;
	} 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_ctrl(u64 ctrl)
{
	wrmsr(MSR_CORE_PERF_GLOBAL_CTRL, ctrl, 0);
}
EXPORT_SYMBOL_GPL(hw_perf_restore_ctrl);

u64 hw_perf_disable_all(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_disable_all);

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

void hw_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);

	__hw_perf_counter_disable(hwc, idx);

	cpuc->counters[idx] = counter;

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

#ifdef CONFIG_X86_64
static inline void atomic64_counter_set(struct perf_counter *counter, u64 val)
{
	atomic64_set(&counter->count, val);
}

static inline u64 atomic64_counter_read(struct perf_counter *counter)
{
	return atomic64_read(&counter->count);
}
#else
/*
 * Todo: add proper atomic64_t support to 32-bit x86:
 */
static inline void atomic64_counter_set(struct perf_counter *counter, u64 val64)
{
	u32 *val32 = (void *)&val64;

	atomic_set(counter->count32 + 0, *(val32 + 0));
	atomic_set(counter->count32 + 1, *(val32 + 1));
}

static inline u64 atomic64_counter_read(struct perf_counter *counter)
{
	return atomic_read(counter->count32 + 0) |
		(u64) atomic_read(counter->count32 + 1) << 32;
}
#endif

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();
}

void hw_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;

	__hw_perf_counter_disable(hwc, idx);

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

void hw_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)
		__hw_perf_counter_enable(hwc, idx);
}

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

	list_for_each_entry(counter, &ctx->counters, list) {
		if (counter->record_type != PERF_RECORD_SIMPLE ||
		    counter == leader)
			continue;

		if (counter->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(leader, counter->event.hw_event_type);
		perf_store_irq_data(leader, 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->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_store_irq_data(counter,
					    counter->event.hw_event_type);
			perf_store_irq_data(counter,
					    atomic64_counter_read(counter));
			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();
#ifdef CONFIG_X86_64
	add_pda(apic_perf_irqs, 1);
#else
	per_cpu(irq_stat, smp_processor_id()).apic_perf_irqs++;
#endif
	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;
}
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	*/
dfa7c899	59	int hw_perf_counter_init(struct perf_counter *counter)
241771ef IM	60	{
241771ef IM	61	struct hw_perf_counter *hwc = &counter->hw;
dfa7c899	62	u32 hw_event_type = counter->event.hw_event_type;
241771ef IM	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;
	80	if (hw_event_type & PERF_COUNT_NMI)
	81	hwc->nmi = 1;
	82	}
	83
	84	hwc->config_base = MSR_ARCH_PERFMON_EVENTSEL0;
	85	hwc->counter_base = MSR_ARCH_PERFMON_PERFCTR0;
	86
dfa7c899	87	hwc->irq_period = counter->event.hw_event_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
	96	hwc->next_count = -((s32) hwc->irq_period);
	97
	98	/*
dfa7c899	99	* Raw event type provide the config in the event structure
241771ef	100	*/
dfa7c899 TG	101	hw_event_type &= ~PERF_COUNT_NMI;
	102	if (hw_event_type == PERF_COUNT_RAW) {
	103	hwc->config \|= counter->event.hw_raw_ctrl;
241771ef	104	} else {
241771ef IM	105	if (hw_event_type >= max_intel_perfmon_events)
	106	return -EINVAL;
	107	/*
	108	* The generic map:
	109	*/
dfa7c899	110	hwc->config \|= intel_perfmon_event_map[hw_event_type];
241771ef	111	}
241771ef IM	112	counter->wakeup_pending = 0;
	113
	114	return 0;
	115	}
	116
241771ef IM	117	void hw_perf_enable_all(void)
241771ef IM	118	{
43874d23	119	wrmsr(MSR_CORE_PERF_GLOBAL_CTRL, perf_counter_mask, 0);
241771ef IM	120	}
241771ef IM	121
4ac13294	122	void hw_perf_restore_ctrl(u64 ctrl)
241771ef	123	{
4ac13294 TG	124	wrmsr(MSR_CORE_PERF_GLOBAL_CTRL, ctrl, 0);
	125	}
	126	EXPORT_SYMBOL_GPL(hw_perf_restore_ctrl);
	127
	128	u64 hw_perf_disable_all(void)
	129	{
	130	u64 ctrl;
	131
	132	rdmsrl(MSR_CORE_PERF_GLOBAL_CTRL, ctrl);
241771ef	133	wrmsr(MSR_CORE_PERF_GLOBAL_CTRL, 0, 0);
4ac13294	134	return ctrl;
241771ef	135	}
4ac13294	136	EXPORT_SYMBOL_GPL(hw_perf_disable_all);
241771ef	137
7e2ae347 IM	138	static inline void
	139	__hw_perf_counter_disable(struct hw_perf_counter *hwc, unsigned int idx)
	140	{
	141	wrmsr(hwc->config_base + idx, hwc->config, 0);
	142	}
	143
241771ef IM	144	static DEFINE_PER_CPU(u64, prev_next_count[MAX_HW_COUNTERS]);
241771ef IM	145
7e2ae347	146	static void __hw_perf_counter_set_period(struct hw_perf_counter *hwc, int idx)
241771ef IM	147	{
	148	per_cpu(prev_next_count[idx], smp_processor_id()) = hwc->next_count;
	149
	150	wrmsr(hwc->counter_base + idx, hwc->next_count, 0);
7e2ae347 IM	151	}
	152
	153	static void __hw_perf_counter_enable(struct hw_perf_counter *hwc, int idx)
	154	{
	155	wrmsr(hwc->config_base + idx,
	156	hwc->config \| ARCH_PERFMON_EVENTSEL0_ENABLE, 0);
241771ef IM	157	}
	158
	159	void hw_perf_counter_enable(struct perf_counter *counter)
	160	{
	161	struct cpu_hw_counters *cpuc = &__get_cpu_var(cpu_hw_counters);
	162	struct hw_perf_counter *hwc = &counter->hw;
	163	int idx = hwc->idx;
	164
	165	/* Try to get the previous counter again */
	166	if (test_and_set_bit(idx, cpuc->used)) {
	167	idx = find_first_zero_bit(cpuc->used, nr_hw_counters);
	168	set_bit(idx, cpuc->used);
	169	hwc->idx = idx;
	170	}
	171
	172	perf_counters_lapic_init(hwc->nmi);
	173
7e2ae347	174	__hw_perf_counter_disable(hwc, idx);
241771ef IM	175
241771ef IM	176	cpuc->counters[idx] = counter;
7e2ae347 IM	177
7e2ae347 IM	178	__hw_perf_counter_set_period(hwc, idx);
241771ef IM	179	__hw_perf_counter_enable(hwc, idx);
	180	}
	181
	182	#ifdef CONFIG_X86_64
	183	static inline void atomic64_counter_set(struct perf_counter *counter, u64 val)
	184	{
	185	atomic64_set(&counter->count, val);
	186	}
	187
	188	static inline u64 atomic64_counter_read(struct perf_counter *counter)
	189	{
	190	return atomic64_read(&counter->count);
	191	}
	192	#else
	193	/*
	194	* Todo: add proper atomic64_t support to 32-bit x86:
	195	*/
	196	static inline void atomic64_counter_set(struct perf_counter *counter, u64 val64)
	197	{
	198	u32 val32 = (void )&val64;
	199
	200	atomic_set(counter->count32 + 0, *(val32 + 0));
	201	atomic_set(counter->count32 + 1, *(val32 + 1));
	202	}
	203
	204	static inline u64 atomic64_counter_read(struct perf_counter *counter)
	205	{
	206	return atomic_read(counter->count32 + 0) \|
	207	(u64) atomic_read(counter->count32 + 1) << 32;
	208	}
	209	#endif
	210
	211	static void __hw_perf_save_counter(struct perf_counter *counter,
	212	struct hw_perf_counter *hwc, int idx)
	213	{
	214	s64 raw = -1;
	215	s64 delta;
241771ef IM	216
	217	/*
	218	* Get the raw hw counter value:
	219	*/
1e125676	220	rdmsrl(hwc->counter_base + idx, raw);
241771ef IM	221
	222	/*
	223	* Rebase it to zero (it started counting at -irq_period),
	224	* to see the delta since ->prev_count:
	225	*/
	226	delta = (s64)hwc->irq_period + (s64)(s32)raw;
	227
	228	atomic64_counter_set(counter, hwc->prev_count + delta);
	229
	230	/*
	231	* Adjust the ->prev_count offset - if we went beyond
	232	* irq_period of units, then we got an IRQ and the counter
	233	* was set back to -irq_period:
	234	*/
	235	while (delta >= (s64)hwc->irq_period) {
	236	hwc->prev_count += hwc->irq_period;
	237	delta -= (s64)hwc->irq_period;
	238	}
	239
	240	/*
	241	* Calculate the next raw counter value we'll write into
	242	* the counter at the next sched-in time:
	243	*/
	244	delta -= (s64)hwc->irq_period;
	245
	246	hwc->next_count = (s32)delta;
	247	}
	248
	249	void perf_counter_print_debug(void)
	250	{
	251	u64 ctrl, status, overflow, pmc_ctrl, pmc_count, next_count;
1e125676 IM	252	int cpu, idx;
	253
	254	if (!nr_hw_counters)
	255	return;
241771ef IM	256
	257	local_irq_disable();
	258
	259	cpu = smp_processor_id();
	260
1e125676 IM	261	rdmsrl(MSR_CORE_PERF_GLOBAL_CTRL, ctrl);
	262	rdmsrl(MSR_CORE_PERF_GLOBAL_STATUS, status);
	263	rdmsrl(MSR_CORE_PERF_GLOBAL_OVF_CTRL, overflow);
241771ef IM	264
	265	printk(KERN_INFO "\n");
	266	printk(KERN_INFO "CPU#%d: ctrl: %016llx\n", cpu, ctrl);
	267	printk(KERN_INFO "CPU#%d: status: %016llx\n", cpu, status);
	268	printk(KERN_INFO "CPU#%d: overflow: %016llx\n", cpu, overflow);
	269
	270	for (idx = 0; idx < nr_hw_counters; idx++) {
1e125676 IM	271	rdmsrl(MSR_ARCH_PERFMON_EVENTSEL0 + idx, pmc_ctrl);
1e125676 IM	272	rdmsrl(MSR_ARCH_PERFMON_PERFCTR0 + idx, pmc_count);
241771ef IM	273
	274	next_count = per_cpu(prev_next_count[idx], cpu);
	275
	276	printk(KERN_INFO "CPU#%d: PMC%d ctrl: %016llx\n",
	277	cpu, idx, pmc_ctrl);
	278	printk(KERN_INFO "CPU#%d: PMC%d count: %016llx\n",
	279	cpu, idx, pmc_count);
	280	printk(KERN_INFO "CPU#%d: PMC%d next: %016llx\n",
	281	cpu, idx, next_count);
	282	}
	283	local_irq_enable();
	284	}
	285
	286	void hw_perf_counter_disable(struct perf_counter *counter)
	287	{
	288	struct cpu_hw_counters *cpuc = &__get_cpu_var(cpu_hw_counters);
	289	struct hw_perf_counter *hwc = &counter->hw;
	290	unsigned int idx = hwc->idx;
	291
7e2ae347	292	__hw_perf_counter_disable(hwc, idx);
241771ef IM	293
	294	clear_bit(idx, cpuc->used);
	295	cpuc->counters[idx] = NULL;
	296	__hw_perf_save_counter(counter, hwc, idx);
	297	}
	298
	299	void hw_perf_counter_read(struct perf_counter *counter)
	300	{
	301	struct hw_perf_counter *hwc = &counter->hw;
	302	unsigned long addr = hwc->counter_base + hwc->idx;
	303	s64 offs, val = -1LL;
	304	s32 val32;
241771ef IM	305
	306	/* Careful: NMI might modify the counter offset */
	307	do {
	308	offs = hwc->prev_count;
1e125676	309	rdmsrl(addr, val);
241771ef IM	310	} while (offs != hwc->prev_count);
	311
	312	val32 = (s32) val;
	313	val = (s64)hwc->irq_period + (s64)val32;
	314	atomic64_counter_set(counter, hwc->prev_count + val);
	315	}
	316
	317	static void perf_store_irq_data(struct perf_counter *counter, u64 data)
	318	{
	319	struct perf_data *irqdata = counter->irqdata;
	320
	321	if (irqdata->len > PERF_DATA_BUFLEN - sizeof(u64)) {
	322	irqdata->overrun++;
	323	} else {
	324	u64 p = (u64 ) &irqdata->data[irqdata->len];
	325
	326	*p = data;
	327	irqdata->len += sizeof(u64);
	328	}
	329	}
	330
7e2ae347 IM	331	/*
	332	* NMI-safe enable method:
	333	*/
241771ef IM	334	static void perf_save_and_restart(struct perf_counter *counter)
	335	{
	336	struct hw_perf_counter *hwc = &counter->hw;
	337	int idx = hwc->idx;
7e2ae347	338	u64 pmc_ctrl;
241771ef	339
1e125676	340	rdmsrl(MSR_ARCH_PERFMON_EVENTSEL0 + idx, pmc_ctrl);
241771ef	341
7e2ae347 IM	342	__hw_perf_save_counter(counter, hwc, idx);
	343	__hw_perf_counter_set_period(hwc, idx);
	344
	345	if (pmc_ctrl & ARCH_PERFMON_EVENTSEL0_ENABLE)
241771ef	346	__hw_perf_counter_enable(hwc, idx);
241771ef IM	347	}
	348
	349	static void
	350	perf_handle_group(struct perf_counter leader, u64 status, u64 *overflown)
	351	{
	352	struct perf_counter_context *ctx = leader->ctx;
	353	struct perf_counter *counter;
	354	int bit;
	355
	356	list_for_each_entry(counter, &ctx->counters, list) {
	357	if (counter->record_type != PERF_RECORD_SIMPLE \|\|
	358	counter == leader)
	359	continue;
	360
	361	if (counter->active) {
	362	/*
	363	* When counter was not in the overflow mask, we have to
	364	* read it from hardware. We read it as well, when it
	365	* has not been read yet and clear the bit in the
	366	* status mask.
	367	*/
	368	bit = counter->hw.idx;
	369	if (!test_bit(bit, (unsigned long *) overflown) \|\|
	370	test_bit(bit, (unsigned long *) status)) {
	371	clear_bit(bit, (unsigned long *) status);
	372	perf_save_and_restart(counter);
	373	}
	374	}
dfa7c899	375	perf_store_irq_data(leader, counter->event.hw_event_type);
241771ef IM	376	perf_store_irq_data(leader, atomic64_counter_read(counter));
	377	}
	378	}
	379
	380	/*
	381	* This handler is triggered by the local APIC, so the APIC IRQ handling
	382	* rules apply:
	383	*/
	384	static void __smp_perf_counter_interrupt(struct pt_regs *regs, int nmi)
	385	{
	386	int bit, cpu = smp_processor_id();
43874d23	387	u64 ack, status, saved_global;
241771ef	388	struct cpu_hw_counters *cpuc;
43874d23 IM	389
43874d23 IM	390	rdmsrl(MSR_CORE_PERF_GLOBAL_CTRL, saved_global);
241771ef	391
241771ef IM	392	/* Disable counters globally */
	393	wrmsr(MSR_CORE_PERF_GLOBAL_CTRL, 0, 0);
	394	ack_APIC_irq();
	395
	396	cpuc = &per_cpu(cpu_hw_counters, cpu);
	397
87b9cf46 IM	398	rdmsrl(MSR_CORE_PERF_GLOBAL_STATUS, status);
	399	if (!status)
	400	goto out;
	401
241771ef IM	402	again:
	403	ack = status;
	404	for_each_bit(bit, (unsigned long *) &status, nr_hw_counters) {
	405	struct perf_counter *counter = cpuc->counters[bit];
	406
	407	clear_bit(bit, (unsigned long *) &status);
	408	if (!counter)
	409	continue;
	410
	411	perf_save_and_restart(counter);
	412
	413	switch (counter->record_type) {
	414	case PERF_RECORD_SIMPLE:
	415	continue;
	416	case PERF_RECORD_IRQ:
	417	perf_store_irq_data(counter, instruction_pointer(regs));
	418	break;
	419	case PERF_RECORD_GROUP:
dfa7c899 TG	420	perf_store_irq_data(counter,
dfa7c899 TG	421	counter->event.hw_event_type);
241771ef IM	422	perf_store_irq_data(counter,
	423	atomic64_counter_read(counter));
	424	perf_handle_group(counter, &status, &ack);
	425	break;
	426	}
	427	/*
	428	* From NMI context we cannot call into the scheduler to
	429	* do a task wakeup - but we mark these counters as
	430	* wakeup_pending and initate a wakeup callback:
	431	*/
	432	if (nmi) {
	433	counter->wakeup_pending = 1;
	434	set_tsk_thread_flag(current, TIF_PERF_COUNTERS);
	435	} else {
	436	wake_up(&counter->waitq);
	437	}
	438	}
	439
	440	wrmsr(MSR_CORE_PERF_GLOBAL_OVF_CTRL, ack, 0);
	441
	442	/*
	443	* Repeat if there is more work to be done:
	444	*/
	445	rdmsrl(MSR_CORE_PERF_GLOBAL_STATUS, status);
	446	if (status)
	447	goto again;
87b9cf46	448	out:
241771ef	449	/*
43874d23	450	* Restore - do not reenable when global enable is off:
241771ef	451	*/
43874d23	452	wrmsr(MSR_CORE_PERF_GLOBAL_CTRL, saved_global, 0);
241771ef IM	453	}
	454
	455	void smp_perf_counter_interrupt(struct pt_regs *regs)
	456	{
	457	irq_enter();
	458	#ifdef CONFIG_X86_64
	459	add_pda(apic_perf_irqs, 1);
	460	#else
	461	per_cpu(irq_stat, smp_processor_id()).apic_perf_irqs++;
	462	#endif
	463	apic_write(APIC_LVTPC, LOCAL_PERF_VECTOR);
	464	__smp_perf_counter_interrupt(regs, 0);
	465
	466	irq_exit();
	467	}
	468
	469	/*
	470	* This handler is triggered by NMI contexts:
	471	*/
	472	void perf_counter_notify(struct pt_regs *regs)
	473	{
	474	struct cpu_hw_counters *cpuc;
	475	unsigned long flags;
	476	int bit, cpu;
	477
	478	local_irq_save(flags);
	479	cpu = smp_processor_id();
	480	cpuc = &per_cpu(cpu_hw_counters, cpu);
	481
	482	for_each_bit(bit, cpuc->used, nr_hw_counters) {
	483	struct perf_counter *counter = cpuc->counters[bit];
	484
	485	if (!counter)
	486	continue;
	487
	488	if (counter->wakeup_pending) {
	489	counter->wakeup_pending = 0;
	490	wake_up(&counter->waitq);
	491	}
	492	}
	493
	494	local_irq_restore(flags);
	495	}
	496
	497	void __cpuinit perf_counters_lapic_init(int nmi)
	498	{
	499	u32 apic_val;
	500
	501	if (!perf_counters_initialized)
	502	return;
	503	/*
	504	* Enable the performance counter vector in the APIC LVT:
	505	*/
	506	apic_val = apic_read(APIC_LVTERR);
	507
	508	apic_write(APIC_LVTERR, apic_val \| APIC_LVT_MASKED);
	509	if (nmi)
	510	apic_write(APIC_LVTPC, APIC_DM_NMI);
	511	else
	512	apic_write(APIC_LVTPC, LOCAL_PERF_VECTOR);
	513	apic_write(APIC_LVTERR, apic_val);
	514	}
	515
	516	static int __kprobes
517	perf_counter_nmi_handler(struct notifier_block *self,
518	unsigned long cmd, void *__args)
519	{
520	struct die_args *args = __args;
521	struct pt_regs *regs;
522
523	if (likely(cmd != DIE_NMI_IPI))
524	return NOTIFY_DONE;
525
526	regs = args->regs;
527
528	apic_write(APIC_LVTPC, APIC_DM_NMI);
529	__smp_perf_counter_interrupt(regs, 1);
530
531	return NOTIFY_STOP;
532	}
533
534	static __read_mostly struct notifier_block perf_counter_nmi_notifier = {
535	.notifier_call = perf_counter_nmi_handler
536	};
537
538	void __init init_hw_perf_counters(void)
539	{
540	union cpuid10_eax eax;
541	unsigned int unused;
542	unsigned int ebx;
543
544	if (!cpu_has(&boot_cpu_data, X86_FEATURE_ARCH_PERFMON))
545	return;
546
547	/*
548	* Check whether the Architectural PerfMon supports
549	* Branch Misses Retired Event or not.
550	*/
551	cpuid(10, &(eax.full), &ebx, &unused, &unused);
552	if (eax.split.mask_length <= ARCH_PERFMON_BRANCH_MISSES_RETIRED)
553	return;
554
555	printk(KERN_INFO "Intel Performance Monitoring support detected.\n");
556
557	printk(KERN_INFO "... version: %d\n", eax.split.version_id);
558	printk(KERN_INFO "... num_counters: %d\n", eax.split.num_counters);
559	nr_hw_counters = eax.split.num_counters;
560	if (nr_hw_counters > MAX_HW_COUNTERS) {
561	nr_hw_counters = MAX_HW_COUNTERS;
562	WARN(1, KERN_ERR "hw perf counters %d > max(%d), clipping!",
563	nr_hw_counters, MAX_HW_COUNTERS);
564	}
565	perf_counter_mask = (1 << nr_hw_counters) - 1;
566	perf_max_counters = nr_hw_counters;
567
568	printk(KERN_INFO "... bit_width: %d\n", eax.split.bit_width);
569	printk(KERN_INFO "... mask_length: %d\n", eax.split.mask_length);
570
571	perf_counters_lapic_init(0);
572	register_die_notifier(&perf_counter_nmi_notifier);
573
574	perf_counters_initialized = true;
575	}