linux/arch/mips/kernel/perf_event_mipsxx.c

/*
 * Linux performance counter support for MIPS.
 *
 * Copyright (C) 2010 MIPS Technologies, Inc.
 * Copyright (C) 2011 Cavium Networks, Inc.
 * Author: Deng-Cheng Zhu
 *
 * This code is based on the implementation for ARM, which is in turn
 * based on the sparc64 perf event code and the x86 code. Performance
 * counter access is based on the MIPS Oprofile code. And the callchain
 * support references the code of MIPS stacktrace.c.
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License version 2 as
 * published by the Free Software Foundation.
 */

#include <linux/cpumask.h>
#include <linux/interrupt.h>
#include <linux/smp.h>
#include <linux/kernel.h>
#include <linux/perf_event.h>
#include <linux/uaccess.h>

#include <asm/irq.h>
#include <asm/irq_regs.h>
#include <asm/stacktrace.h>
#include <asm/time.h> /* For perf_irq */

#define MIPS_MAX_HWEVENTS 4
#define MIPS_TCS_PER_COUNTER 2
#define MIPS_CPUID_TO_COUNTER_MASK (MIPS_TCS_PER_COUNTER - 1)

struct cpu_hw_events {
	/* Array of events on this cpu. */
	struct perf_event	*events[MIPS_MAX_HWEVENTS];

	/*
	 * Set the bit (indexed by the counter number) when the counter
	 * is used for an event.
	 */
	unsigned long		used_mask[BITS_TO_LONGS(MIPS_MAX_HWEVENTS)];

	/*
	 * Software copy of the control register for each performance counter.
	 * MIPS CPUs vary in performance counters. They use this differently,
	 * and even may not use it.
	 */
	unsigned int		saved_ctrl[MIPS_MAX_HWEVENTS];
};
DEFINE_PER_CPU(struct cpu_hw_events, cpu_hw_events) = {
	.saved_ctrl = {0},
};

/* The description of MIPS performance events. */
struct mips_perf_event {
	unsigned int event_id;
	/*
	 * MIPS performance counters are indexed starting from 0.
	 * CNTR_EVEN indicates the indexes of the counters to be used are
	 * even numbers.
	 */
	unsigned int cntr_mask;
	#define CNTR_EVEN	0x55555555
	#define CNTR_ODD	0xaaaaaaaa
	#define CNTR_ALL	0xffffffff
	enum {
		T  = 0,
		V  = 1,
		P  = 2,
	} range;
};

static struct mips_perf_event raw_event;
static DEFINE_MUTEX(raw_event_mutex);

#define C(x) PERF_COUNT_HW_CACHE_##x

struct mips_pmu {
	u64		max_period;
	u64		valid_count;
	u64		overflow;
	const char	*name;
	int		irq;
	u64		(*read_counter)(unsigned int idx);
	void		(*write_counter)(unsigned int idx, u64 val);
	const struct mips_perf_event *(*map_raw_event)(u64 config);
	const struct mips_perf_event (*general_event_map)[PERF_COUNT_HW_MAX];
	const struct mips_perf_event (*cache_event_map)
				[PERF_COUNT_HW_CACHE_MAX]
				[PERF_COUNT_HW_CACHE_OP_MAX]
				[PERF_COUNT_HW_CACHE_RESULT_MAX];
	unsigned int	num_counters;
};

static struct mips_pmu mipspmu;

#define M_PERFCTL_EVENT(event)		(((event) << MIPS_PERFCTRL_EVENT_S) & \
					 MIPS_PERFCTRL_EVENT)
#define M_PERFCTL_VPEID(vpe)		((vpe)	  << MIPS_PERFCTRL_VPEID_S)

#ifdef CONFIG_CPU_BMIPS5000
#define M_PERFCTL_MT_EN(filter)		0
#else /* !CONFIG_CPU_BMIPS5000 */
#define M_PERFCTL_MT_EN(filter)		(filter)
#endif /* CONFIG_CPU_BMIPS5000 */

#define	   M_TC_EN_ALL			M_PERFCTL_MT_EN(MIPS_PERFCTRL_MT_EN_ALL)
#define	   M_TC_EN_VPE			M_PERFCTL_MT_EN(MIPS_PERFCTRL_MT_EN_VPE)
#define	   M_TC_EN_TC			M_PERFCTL_MT_EN(MIPS_PERFCTRL_MT_EN_TC)

#define M_PERFCTL_COUNT_EVENT_WHENEVER	(MIPS_PERFCTRL_EXL |		\
					 MIPS_PERFCTRL_K |		\
					 MIPS_PERFCTRL_U |		\
					 MIPS_PERFCTRL_S |		\
					 MIPS_PERFCTRL_IE)

#ifdef CONFIG_MIPS_MT_SMP
#define M_PERFCTL_CONFIG_MASK		0x3fff801f
#else
#define M_PERFCTL_CONFIG_MASK		0x1f
#endif


#ifdef CONFIG_MIPS_PERF_SHARED_TC_COUNTERS
static DEFINE_RWLOCK(pmuint_rwlock);

#if defined(CONFIG_CPU_BMIPS5000)
#define vpe_id()	(cpu_has_mipsmt_pertccounters ? \
			 0 : (smp_processor_id() & MIPS_CPUID_TO_COUNTER_MASK))
#else
#define vpe_id()	(cpu_has_mipsmt_pertccounters ? \
			 0 : cpu_vpe_id(&current_cpu_data))
#endif

/* Copied from op_model_mipsxx.c */
static unsigned int vpe_shift(void)
{
	if (num_possible_cpus() > 1)
		return 1;

	return 0;
}

static unsigned int counters_total_to_per_cpu(unsigned int counters)
{
	return counters >> vpe_shift();
}

#else /* !CONFIG_MIPS_PERF_SHARED_TC_COUNTERS */
#define vpe_id()	0

#endif /* CONFIG_MIPS_PERF_SHARED_TC_COUNTERS */

static void resume_local_counters(void);
static void pause_local_counters(void);
static irqreturn_t mipsxx_pmu_handle_irq(int, void *);
static int mipsxx_pmu_handle_shared_irq(void);

static unsigned int mipsxx_pmu_swizzle_perf_idx(unsigned int idx)
{
	if (vpe_id() == 1)
		idx = (idx + 2) & 3;
	return idx;
}

static u64 mipsxx_pmu_read_counter(unsigned int idx)
{
	idx = mipsxx_pmu_swizzle_perf_idx(idx);

	switch (idx) {
	case 0:
		/*
		 * The counters are unsigned, we must cast to truncate
		 * off the high bits.
		 */
		return (u32)read_c0_perfcntr0();
	case 1:
		return (u32)read_c0_perfcntr1();
	case 2:
		return (u32)read_c0_perfcntr2();
	case 3:
		return (u32)read_c0_perfcntr3();
	default:
		WARN_ONCE(1, "Invalid performance counter number (%d)\n", idx);
		return 0;
	}
}

static u64 mipsxx_pmu_read_counter_64(unsigned int idx)
{
	idx = mipsxx_pmu_swizzle_perf_idx(idx);

	switch (idx) {
	case 0:
		return read_c0_perfcntr0_64();
	case 1:
		return read_c0_perfcntr1_64();
	case 2:
		return read_c0_perfcntr2_64();
	case 3:
		return read_c0_perfcntr3_64();
	default:
		WARN_ONCE(1, "Invalid performance counter number (%d)\n", idx);
		return 0;
	}
}

static void mipsxx_pmu_write_counter(unsigned int idx, u64 val)
{
	idx = mipsxx_pmu_swizzle_perf_idx(idx);

	switch (idx) {
	case 0:
		write_c0_perfcntr0(val);
		return;
	case 1:
		write_c0_perfcntr1(val);
		return;
	case 2:
		write_c0_perfcntr2(val);
		return;
	case 3:
		write_c0_perfcntr3(val);
		return;
	}
}

static void mipsxx_pmu_write_counter_64(unsigned int idx, u64 val)
{
	idx = mipsxx_pmu_swizzle_perf_idx(idx);

	switch (idx) {
	case 0:
		write_c0_perfcntr0_64(val);
		return;
	case 1:
		write_c0_perfcntr1_64(val);
		return;
	case 2:
		write_c0_perfcntr2_64(val);
		return;
	case 3:
		write_c0_perfcntr3_64(val);
		return;
	}
}

static unsigned int mipsxx_pmu_read_control(unsigned int idx)
{
	idx = mipsxx_pmu_swizzle_perf_idx(idx);

	switch (idx) {
	case 0:
		return read_c0_perfctrl0();
	case 1:
		return read_c0_perfctrl1();
	case 2:
		return read_c0_perfctrl2();
	case 3:
		return read_c0_perfctrl3();
	default:
		WARN_ONCE(1, "Invalid performance counter number (%d)\n", idx);
		return 0;
	}
}

static void mipsxx_pmu_write_control(unsigned int idx, unsigned int val)
{
	idx = mipsxx_pmu_swizzle_perf_idx(idx);

	switch (idx) {
	case 0:
		write_c0_perfctrl0(val);
		return;
	case 1:
		write_c0_perfctrl1(val);
		return;
	case 2:
		write_c0_perfctrl2(val);
		return;
	case 3:
		write_c0_perfctrl3(val);
		return;
	}
}

static int mipsxx_pmu_alloc_counter(struct cpu_hw_events *cpuc,
				    struct hw_perf_event *hwc)
{
	int i;

	/*
	 * We only need to care the counter mask. The range has been
	 * checked definitely.
	 */
	unsigned long cntr_mask = (hwc->event_base >> 8) & 0xffff;

	for (i = mipspmu.num_counters - 1; i >= 0; i--) {
		/*
		 * Note that some MIPS perf events can be counted by both
		 * even and odd counters, wheresas many other are only by
		 * even _or_ odd counters. This introduces an issue that
		 * when the former kind of event takes the counter the
		 * latter kind of event wants to use, then the "counter
		 * allocation" for the latter event will fail. In fact if
		 * they can be dynamically swapped, they both feel happy.
		 * But here we leave this issue alone for now.
		 */
		if (test_bit(i, &cntr_mask) &&
			!test_and_set_bit(i, cpuc->used_mask))
			return i;
	}

	return -EAGAIN;
}

static void mipsxx_pmu_enable_event(struct hw_perf_event *evt, int idx)
{
	struct perf_event *event = container_of(evt, struct perf_event, hw);
	struct cpu_hw_events *cpuc = this_cpu_ptr(&cpu_hw_events);
	unsigned int range = evt->event_base >> 24;

	WARN_ON(idx < 0 || idx >= mipspmu.num_counters);

	cpuc->saved_ctrl[idx] = M_PERFCTL_EVENT(evt->event_base & 0xff) |
		(evt->config_base & M_PERFCTL_CONFIG_MASK) |
		/* Make sure interrupt enabled. */
		MIPS_PERFCTRL_IE;

	if (IS_ENABLED(CONFIG_CPU_BMIPS5000)) {
		/* enable the counter for the calling thread */
		cpuc->saved_ctrl[idx] |=
			(1 << (12 + vpe_id())) | BRCM_PERFCTRL_TC;
	} else if (IS_ENABLED(CONFIG_MIPS_MT_SMP) && range > V) {
		/* The counter is processor wide. Set it up to count all TCs. */
		pr_debug("Enabling perf counter for all TCs\n");
		cpuc->saved_ctrl[idx] |= M_TC_EN_ALL;
	} else {
		unsigned int cpu, ctrl;

		/*
		 * Set up the counter for a particular CPU when event->cpu is
		 * a valid CPU number. Otherwise set up the counter for the CPU
		 * scheduling this thread.
		 */
		cpu = (event->cpu >= 0) ? event->cpu : smp_processor_id();

		ctrl = M_PERFCTL_VPEID(cpu_vpe_id(&cpu_data[cpu]));
		ctrl |= M_TC_EN_VPE;
		cpuc->saved_ctrl[idx] |= ctrl;
		pr_debug("Enabling perf counter for CPU%d\n", cpu);
	}
	/*
	 * We do not actually let the counter run. Leave it until start().
	 */
}

static void mipsxx_pmu_disable_event(int idx)
{
	struct cpu_hw_events *cpuc = this_cpu_ptr(&cpu_hw_events);
	unsigned long flags;

	WARN_ON(idx < 0 || idx >= mipspmu.num_counters);

	local_irq_save(flags);
	cpuc->saved_ctrl[idx] = mipsxx_pmu_read_control(idx) &
		~M_PERFCTL_COUNT_EVENT_WHENEVER;
	mipsxx_pmu_write_control(idx, cpuc->saved_ctrl[idx]);
	local_irq_restore(flags);
}

static int mipspmu_event_set_period(struct perf_event *event,
				    struct hw_perf_event *hwc,
				    int idx)
{
	u64 left = local64_read(&hwc->period_left);
	u64 period = hwc->sample_period;
	int ret = 0;

	if (unlikely((left + period) & (1ULL << 63))) {
		/* left underflowed by more than period. */
		left = period;
		local64_set(&hwc->period_left, left);
		hwc->last_period = period;
		ret = 1;
	} else	if (unlikely((left + period) <= period)) {
		/* left underflowed by less than period. */
		left += period;
		local64_set(&hwc->period_left, left);
		hwc->last_period = period;
		ret = 1;
	}

	if (left > mipspmu.max_period) {
		left = mipspmu.max_period;
		local64_set(&hwc->period_left, left);
	}

	local64_set(&hwc->prev_count, mipspmu.overflow - left);

	mipspmu.write_counter(idx, mipspmu.overflow - left);

	perf_event_update_userpage(event);

	return ret;
}

static void mipspmu_event_update(struct perf_event *event,
				 struct hw_perf_event *hwc,
				 int idx)
{
	u64 prev_raw_count, new_raw_count;
	u64 delta;

again:
	prev_raw_count = local64_read(&hwc->prev_count);
	new_raw_count = mipspmu.read_counter(idx);

	if (local64_cmpxchg(&hwc->prev_count, prev_raw_count,
				new_raw_count) != prev_raw_count)
		goto again;

	delta = new_raw_count - prev_raw_count;

	local64_add(delta, &event->count);
	local64_sub(delta, &hwc->period_left);
}

static void mipspmu_start(struct perf_event *event, int flags)
{
	struct hw_perf_event *hwc = &event->hw;

	if (flags & PERF_EF_RELOAD)
		WARN_ON_ONCE(!(hwc->state & PERF_HES_UPTODATE));

	hwc->state = 0;

	/* Set the period for the event. */
	mipspmu_event_set_period(event, hwc, hwc->idx);

	/* Enable the event. */
	mipsxx_pmu_enable_event(hwc, hwc->idx);
}

static void mipspmu_stop(struct perf_event *event, int flags)
{
	struct hw_perf_event *hwc = &event->hw;

	if (!(hwc->state & PERF_HES_STOPPED)) {
		/* We are working on a local event. */
		mipsxx_pmu_disable_event(hwc->idx);
		barrier();
		mipspmu_event_update(event, hwc, hwc->idx);
		hwc->state |= PERF_HES_STOPPED | PERF_HES_UPTODATE;
	}
}

static int mipspmu_add(struct perf_event *event, int flags)
{
	struct cpu_hw_events *cpuc = this_cpu_ptr(&cpu_hw_events);
	struct hw_perf_event *hwc = &event->hw;
	int idx;
	int err = 0;

	perf_pmu_disable(event->pmu);

	/* To look for a free counter for this event. */
	idx = mipsxx_pmu_alloc_counter(cpuc, hwc);
	if (idx < 0) {
		err = idx;
		goto out;
	}

	/*
	 * If there is an event in the counter we are going to use then
	 * make sure it is disabled.
	 */
	event->hw.idx = idx;
	mipsxx_pmu_disable_event(idx);
	cpuc->events[idx] = event;

	hwc->state = PERF_HES_STOPPED | PERF_HES_UPTODATE;
	if (flags & PERF_EF_START)
		mipspmu_start(event, PERF_EF_RELOAD);

	/* Propagate our changes to the userspace mapping. */
	perf_event_update_userpage(event);

out:
	perf_pmu_enable(event->pmu);
	return err;
}

static void mipspmu_del(struct perf_event *event, int flags)
{
	struct cpu_hw_events *cpuc = this_cpu_ptr(&cpu_hw_events);
	struct hw_perf_event *hwc = &event->hw;
	int idx = hwc->idx;

	WARN_ON(idx < 0 || idx >= mipspmu.num_counters);

	mipspmu_stop(event, PERF_EF_UPDATE);
	cpuc->events[idx] = NULL;
	clear_bit(idx, cpuc->used_mask);

	perf_event_update_userpage(event);
}

static void mipspmu_read(struct perf_event *event)
{
	struct hw_perf_event *hwc = &event->hw;

	/* Don't read disabled counters! */
	if (hwc->idx < 0)
		return;

	mipspmu_event_update(event, hwc, hwc->idx);
}

static void mipspmu_enable(struct pmu *pmu)
{
#ifdef CONFIG_MIPS_PERF_SHARED_TC_COUNTERS
	write_unlock(&pmuint_rwlock);
#endif
	resume_local_counters();
}

/*
 * MIPS performance counters can be per-TC. The control registers can
 * not be directly accessed across CPUs. Hence if we want to do global
 * control, we need cross CPU calls. on_each_cpu() can help us, but we
 * can not make sure this function is called with interrupts enabled. So
 * here we pause local counters and then grab a rwlock and leave the
 * counters on other CPUs alone. If any counter interrupt raises while
 * we own the write lock, simply pause local counters on that CPU and
 * spin in the handler. Also we know we won't be switched to another
 * CPU after pausing local counters and before grabbing the lock.
 */
static void mipspmu_disable(struct pmu *pmu)
{
	pause_local_counters();
#ifdef CONFIG_MIPS_PERF_SHARED_TC_COUNTERS
	write_lock(&pmuint_rwlock);
#endif
}

static atomic_t active_events = ATOMIC_INIT(0);
static DEFINE_MUTEX(pmu_reserve_mutex);
static int (*save_perf_irq)(void);

static int mipspmu_get_irq(void)
{
	int err;

	if (mipspmu.irq >= 0) {
		/* Request my own irq handler. */
		err = request_irq(mipspmu.irq, mipsxx_pmu_handle_irq,
				  IRQF_PERCPU | IRQF_NOBALANCING |
				  IRQF_NO_THREAD | IRQF_NO_SUSPEND |
				  IRQF_SHARED,
				  "mips_perf_pmu", &mipspmu);
		if (err) {
			pr_warn("Unable to request IRQ%d for MIPS performance counters!\n",
				mipspmu.irq);
		}
	} else if (cp0_perfcount_irq < 0) {
		/*
		 * We are sharing the irq number with the timer interrupt.
		 */
		save_perf_irq = perf_irq;
		perf_irq = mipsxx_pmu_handle_shared_irq;
		err = 0;
	} else {
		pr_warn("The platform hasn't properly defined its interrupt controller\n");
		err = -ENOENT;
	}

	return err;
}

static void mipspmu_free_irq(void)
{
	if (mipspmu.irq >= 0)
		free_irq(mipspmu.irq, &mipspmu);
	else if (cp0_perfcount_irq < 0)
		perf_irq = save_perf_irq;
}

/*
 * mipsxx/rm9000/loongson2 have different performance counters, they have
 * specific low-level init routines.
 */
static void reset_counters(void *arg);
static int __hw_perf_event_init(struct perf_event *event);

static void hw_perf_event_destroy(struct perf_event *event)
{
	if (atomic_dec_and_mutex_lock(&active_events,
				&pmu_reserve_mutex)) {
		/*
		 * We must not call the destroy function with interrupts
		 * disabled.
		 */
		on_each_cpu(reset_counters,
			(void *)(long)mipspmu.num_counters, 1);
		mipspmu_free_irq();
		mutex_unlock(&pmu_reserve_mutex);
	}
}

static int mipspmu_event_init(struct perf_event *event)
{
	int err = 0;

	/* does not support taken branch sampling */
	if (has_branch_stack(event))
		return -EOPNOTSUPP;

	switch (event->attr.type) {
	case PERF_TYPE_RAW:
	case PERF_TYPE_HARDWARE:
	case PERF_TYPE_HW_CACHE:
		break;

	default:
		return -ENOENT;
	}

	if (event->cpu >= 0 && !cpu_online(event->cpu))
		return -ENODEV;

	if (!atomic_inc_not_zero(&active_events)) {
		mutex_lock(&pmu_reserve_mutex);
		if (atomic_read(&active_events) == 0)
			err = mipspmu_get_irq();

		if (!err)
			atomic_inc(&active_events);
		mutex_unlock(&pmu_reserve_mutex);
	}

	if (err)
		return err;

	return __hw_perf_event_init(event);
}

static struct pmu pmu = {
	.pmu_enable	= mipspmu_enable,
	.pmu_disable	= mipspmu_disable,
	.event_init	= mipspmu_event_init,
	.add		= mipspmu_add,
	.del		= mipspmu_del,
	.start		= mipspmu_start,
	.stop		= mipspmu_stop,
	.read		= mipspmu_read,
};

static unsigned int mipspmu_perf_event_encode(const struct mips_perf_event *pev)
{
/*
 * Top 8 bits for range, next 16 bits for cntr_mask, lowest 8 bits for
 * event_id.
 */
#ifdef CONFIG_MIPS_MT_SMP
	if (num_possible_cpus() > 1)
		return ((unsigned int)pev->range << 24) |
			(pev->cntr_mask & 0xffff00) |
			(pev->event_id & 0xff);
	else
#endif /* CONFIG_MIPS_MT_SMP */
		return ((pev->cntr_mask & 0xffff00) |
			(pev->event_id & 0xff));
}

static const struct mips_perf_event *mipspmu_map_general_event(int idx)
{

	if ((*mipspmu.general_event_map)[idx].cntr_mask == 0)
		return ERR_PTR(-EOPNOTSUPP);
	return &(*mipspmu.general_event_map)[idx];
}

static const struct mips_perf_event *mipspmu_map_cache_event(u64 config)
{
	unsigned int cache_type, cache_op, cache_result;
	const struct mips_perf_event *pev;

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

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

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

	pev = &((*mipspmu.cache_event_map)
					[cache_type]
					[cache_op]
					[cache_result]);

	if (pev->cntr_mask == 0)
		return ERR_PTR(-EOPNOTSUPP);

	return pev;

}

static int validate_group(struct perf_event *event)
{
	struct perf_event *sibling, *leader = event->group_leader;
	struct cpu_hw_events fake_cpuc;

	memset(&fake_cpuc, 0, sizeof(fake_cpuc));

	if (mipsxx_pmu_alloc_counter(&fake_cpuc, &leader->hw) < 0)
		return -EINVAL;

	for_each_sibling_event(sibling, leader) {
		if (mipsxx_pmu_alloc_counter(&fake_cpuc, &sibling->hw) < 0)
			return -EINVAL;
	}

	if (mipsxx_pmu_alloc_counter(&fake_cpuc, &event->hw) < 0)
		return -EINVAL;

	return 0;
}

/* This is needed by specific irq handlers in perf_event_*.c */
static void handle_associated_event(struct cpu_hw_events *cpuc,
				    int idx, struct perf_sample_data *data,
				    struct pt_regs *regs)
{
	struct perf_event *event = cpuc->events[idx];
	struct hw_perf_event *hwc = &event->hw;

	mipspmu_event_update(event, hwc, idx);
	data->period = event->hw.last_period;
	if (!mipspmu_event_set_period(event, hwc, idx))
		return;

	if (perf_event_overflow(event, data, regs))
		mipsxx_pmu_disable_event(idx);
}


static int __n_counters(void)
{
	if (!cpu_has_perf)
		return 0;
	if (!(read_c0_perfctrl0() & MIPS_PERFCTRL_M))
		return 1;
	if (!(read_c0_perfctrl1() & MIPS_PERFCTRL_M))
		return 2;
	if (!(read_c0_perfctrl2() & MIPS_PERFCTRL_M))
		return 3;

	return 4;
}

static int n_counters(void)
{
	int counters;

	switch (current_cpu_type()) {
	case CPU_R10000:
		counters = 2;
		break;

	case CPU_R12000:
	case CPU_R14000:
	case CPU_R16000:
		counters = 4;
		break;

	default:
		counters = __n_counters();
	}

	return counters;
}

static void reset_counters(void *arg)
{
	int counters = (int)(long)arg;
	switch (counters) {
	case 4:
		mipsxx_pmu_write_control(3, 0);
		mipspmu.write_counter(3, 0);
	case 3:
		mipsxx_pmu_write_control(2, 0);
		mipspmu.write_counter(2, 0);
	case 2:
		mipsxx_pmu_write_control(1, 0);
		mipspmu.write_counter(1, 0);
	case 1:
		mipsxx_pmu_write_control(0, 0);
		mipspmu.write_counter(0, 0);
	}
}

/* 24K/34K/1004K/interAptiv/loongson1 cores share the same event map. */
static const struct mips_perf_event mipsxxcore_event_map
				[PERF_COUNT_HW_MAX] = {
	[PERF_COUNT_HW_CPU_CYCLES] = { 0x00, CNTR_EVEN | CNTR_ODD, P },
	[PERF_COUNT_HW_INSTRUCTIONS] = { 0x01, CNTR_EVEN | CNTR_ODD, T },
	[PERF_COUNT_HW_BRANCH_INSTRUCTIONS] = { 0x02, CNTR_EVEN, T },
	[PERF_COUNT_HW_BRANCH_MISSES] = { 0x02, CNTR_ODD, T },
};

/* 74K/proAptiv core has different branch event code. */
static const struct mips_perf_event mipsxxcore_event_map2
				[PERF_COUNT_HW_MAX] = {
	[PERF_COUNT_HW_CPU_CYCLES] = { 0x00, CNTR_EVEN | CNTR_ODD, P },
	[PERF_COUNT_HW_INSTRUCTIONS] = { 0x01, CNTR_EVEN | CNTR_ODD, T },
	[PERF_COUNT_HW_BRANCH_INSTRUCTIONS] = { 0x27, CNTR_EVEN, T },
	[PERF_COUNT_HW_BRANCH_MISSES] = { 0x27, CNTR_ODD, T },
};

static const struct mips_perf_event i6x00_event_map[PERF_COUNT_HW_MAX] = {
	[PERF_COUNT_HW_CPU_CYCLES]          = { 0x00, CNTR_EVEN | CNTR_ODD },
	[PERF_COUNT_HW_INSTRUCTIONS]        = { 0x01, CNTR_EVEN | CNTR_ODD },
	/* These only count dcache, not icache */
	[PERF_COUNT_HW_CACHE_REFERENCES]    = { 0x45, CNTR_EVEN | CNTR_ODD },
	[PERF_COUNT_HW_CACHE_MISSES]        = { 0x48, CNTR_EVEN | CNTR_ODD },
	[PERF_COUNT_HW_BRANCH_INSTRUCTIONS] = { 0x15, CNTR_EVEN | CNTR_ODD },
	[PERF_COUNT_HW_BRANCH_MISSES]       = { 0x16, CNTR_EVEN | CNTR_ODD },
};

static const struct mips_perf_event loongson3_event_map[PERF_COUNT_HW_MAX] = {
	[PERF_COUNT_HW_CPU_CYCLES] = { 0x00, CNTR_EVEN },
	[PERF_COUNT_HW_INSTRUCTIONS] = { 0x00, CNTR_ODD },
	[PERF_COUNT_HW_BRANCH_INSTRUCTIONS] = { 0x01, CNTR_EVEN },
	[PERF_COUNT_HW_BRANCH_MISSES] = { 0x01, CNTR_ODD },
};

static const struct mips_perf_event octeon_event_map[PERF_COUNT_HW_MAX] = {
	[PERF_COUNT_HW_CPU_CYCLES] = { 0x01, CNTR_ALL },
	[PERF_COUNT_HW_INSTRUCTIONS] = { 0x03, CNTR_ALL },
	[PERF_COUNT_HW_CACHE_REFERENCES] = { 0x2b, CNTR_ALL },
	[PERF_COUNT_HW_CACHE_MISSES] = { 0x2e, CNTR_ALL	 },
	[PERF_COUNT_HW_BRANCH_INSTRUCTIONS] = { 0x08, CNTR_ALL },
	[PERF_COUNT_HW_BRANCH_MISSES] = { 0x09, CNTR_ALL },
	[PERF_COUNT_HW_BUS_CYCLES] = { 0x25, CNTR_ALL },
};

static const struct mips_perf_event bmips5000_event_map
				[PERF_COUNT_HW_MAX] = {
	[PERF_COUNT_HW_CPU_CYCLES] = { 0x00, CNTR_EVEN | CNTR_ODD, T },
	[PERF_COUNT_HW_INSTRUCTIONS] = { 0x01, CNTR_EVEN | CNTR_ODD, T },
	[PERF_COUNT_HW_BRANCH_MISSES] = { 0x02, CNTR_ODD, T },
};

static const struct mips_perf_event xlp_event_map[PERF_COUNT_HW_MAX] = {
	[PERF_COUNT_HW_CPU_CYCLES] = { 0x01, CNTR_ALL },
	[PERF_COUNT_HW_INSTRUCTIONS] = { 0x18, CNTR_ALL }, /* PAPI_TOT_INS */
	[PERF_COUNT_HW_CACHE_REFERENCES] = { 0x04, CNTR_ALL }, /* PAPI_L1_ICA */
	[PERF_COUNT_HW_CACHE_MISSES] = { 0x07, CNTR_ALL }, /* PAPI_L1_ICM */
	[PERF_COUNT_HW_BRANCH_INSTRUCTIONS] = { 0x1b, CNTR_ALL }, /* PAPI_BR_CN */
	[PERF_COUNT_HW_BRANCH_MISSES] = { 0x1c, CNTR_ALL }, /* PAPI_BR_MSP */
};

/* 24K/34K/1004K/interAptiv/loongson1 cores share the same cache event map. */
static const struct mips_perf_event mipsxxcore_cache_map
				[PERF_COUNT_HW_CACHE_MAX]
				[PERF_COUNT_HW_CACHE_OP_MAX]
				[PERF_COUNT_HW_CACHE_RESULT_MAX] = {
[C(L1D)] = {
	/*
	 * Like some other architectures (e.g. ARM), the performance
	 * counters don't differentiate between read and write
	 * accesses/misses, so this isn't strictly correct, but it's the
	 * best we can do. Writes and reads get combined.
	 */
	[C(OP_READ)] = {
		[C(RESULT_ACCESS)]	= { 0x0a, CNTR_EVEN, T },
		[C(RESULT_MISS)]	= { 0x0b, CNTR_EVEN | CNTR_ODD, T },
	},
	[C(OP_WRITE)] = {
		[C(RESULT_ACCESS)]	= { 0x0a, CNTR_EVEN, T },
		[C(RESULT_MISS)]	= { 0x0b, CNTR_EVEN | CNTR_ODD, T },
	},
},
[C(L1I)] = {
	[C(OP_READ)] = {
		[C(RESULT_ACCESS)]	= { 0x09, CNTR_EVEN, T },
		[C(RESULT_MISS)]	= { 0x09, CNTR_ODD, T },
	},
	[C(OP_WRITE)] = {
		[C(RESULT_ACCESS)]	= { 0x09, CNTR_EVEN, T },
		[C(RESULT_MISS)]	= { 0x09, CNTR_ODD, T },
	},
	[C(OP_PREFETCH)] = {
		[C(RESULT_ACCESS)]	= { 0x14, CNTR_EVEN, T },
		/*
		 * Note that MIPS has only "hit" events countable for
		 * the prefetch operation.
		 */
	},
},
[C(LL)] = {
	[C(OP_READ)] = {
		[C(RESULT_ACCESS)]	= { 0x15, CNTR_ODD, P },
		[C(RESULT_MISS)]	= { 0x16, CNTR_EVEN, P },
	},
	[C(OP_WRITE)] = {
		[C(RESULT_ACCESS)]	= { 0x15, CNTR_ODD, P },
		[C(RESULT_MISS)]	= { 0x16, CNTR_EVEN, P },
	},
},
[C(DTLB)] = {
	[C(OP_READ)] = {
		[C(RESULT_ACCESS)]	= { 0x06, CNTR_EVEN, T },
		[C(RESULT_MISS)]	= { 0x06, CNTR_ODD, T },
	},
	[C(OP_WRITE)] = {
		[C(RESULT_ACCESS)]	= { 0x06, CNTR_EVEN, T },
		[C(RESULT_MISS)]	= { 0x06, CNTR_ODD, T },
	},
},
[C(ITLB)] = {
	[C(OP_READ)] = {
		[C(RESULT_ACCESS)]	= { 0x05, CNTR_EVEN, T },
		[C(RESULT_MISS)]	= { 0x05, CNTR_ODD, T },
	},
	[C(OP_WRITE)] = {
		[C(RESULT_ACCESS)]	= { 0x05, CNTR_EVEN, T },
		[C(RESULT_MISS)]	= { 0x05, CNTR_ODD, T },
	},
},
[C(BPU)] = {
	/* Using the same code for *HW_BRANCH* */
	[C(OP_READ)] = {
		[C(RESULT_ACCESS)]	= { 0x02, CNTR_EVEN, T },
		[C(RESULT_MISS)]	= { 0x02, CNTR_ODD, T },
	},
	[C(OP_WRITE)] = {
		[C(RESULT_ACCESS)]	= { 0x02, CNTR_EVEN, T },
		[C(RESULT_MISS)]	= { 0x02, CNTR_ODD, T },
	},
},
};

/* 74K/proAptiv core has completely different cache event map. */
static const struct mips_perf_event mipsxxcore_cache_map2
				[PERF_COUNT_HW_CACHE_MAX]
				[PERF_COUNT_HW_CACHE_OP_MAX]
				[PERF_COUNT_HW_CACHE_RESULT_MAX] = {
[C(L1D)] = {
	/*
	 * Like some other architectures (e.g. ARM), the performance
	 * counters don't differentiate between read and write
	 * accesses/misses, so this isn't strictly correct, but it's the
	 * best we can do. Writes and reads get combined.
	 */
	[C(OP_READ)] = {
		[C(RESULT_ACCESS)]	= { 0x17, CNTR_ODD, T },
		[C(RESULT_MISS)]	= { 0x18, CNTR_ODD, T },
	},
	[C(OP_WRITE)] = {
		[C(RESULT_ACCESS)]	= { 0x17, CNTR_ODD, T },
		[C(RESULT_MISS)]	= { 0x18, CNTR_ODD, T },
	},
},
[C(L1I)] = {
	[C(OP_READ)] = {
		[C(RESULT_ACCESS)]	= { 0x06, CNTR_EVEN, T },
		[C(RESULT_MISS)]	= { 0x06, CNTR_ODD, T },
	},
	[C(OP_WRITE)] = {
		[C(RESULT_ACCESS)]	= { 0x06, CNTR_EVEN, T },
		[C(RESULT_MISS)]	= { 0x06, CNTR_ODD, T },
	},
	[C(OP_PREFETCH)] = {
		[C(RESULT_ACCESS)]	= { 0x34, CNTR_EVEN, T },
		/*
		 * Note that MIPS has only "hit" events countable for
		 * the prefetch operation.
		 */
	},
},
[C(LL)] = {
	[C(OP_READ)] = {
		[C(RESULT_ACCESS)]	= { 0x1c, CNTR_ODD, P },
		[C(RESULT_MISS)]	= { 0x1d, CNTR_EVEN, P },
	},
	[C(OP_WRITE)] = {
		[C(RESULT_ACCESS)]	= { 0x1c, CNTR_ODD, P },
		[C(RESULT_MISS)]	= { 0x1d, CNTR_EVEN, P },
	},
},
/*
 * 74K core does not have specific DTLB events. proAptiv core has
 * "speculative" DTLB events which are numbered 0x63 (even/odd) and
 * not included here. One can use raw events if really needed.
 */
[C(ITLB)] = {
	[C(OP_READ)] = {
		[C(RESULT_ACCESS)]	= { 0x04, CNTR_EVEN, T },
		[C(RESULT_MISS)]	= { 0x04, CNTR_ODD, T },
	},
	[C(OP_WRITE)] = {
		[C(RESULT_ACCESS)]	= { 0x04, CNTR_EVEN, T },
		[C(RESULT_MISS)]	= { 0x04, CNTR_ODD, T },
	},
},
[C(BPU)] = {
	/* Using the same code for *HW_BRANCH* */
	[C(OP_READ)] = {
		[C(RESULT_ACCESS)]	= { 0x27, CNTR_EVEN, T },
		[C(RESULT_MISS)]	= { 0x27, CNTR_ODD, T },
	},
	[C(OP_WRITE)] = {
		[C(RESULT_ACCESS)]	= { 0x27, CNTR_EVEN, T },
		[C(RESULT_MISS)]	= { 0x27, CNTR_ODD, T },
	},
},
};

static const struct mips_perf_event i6x00_cache_map
				[PERF_COUNT_HW_CACHE_MAX]
				[PERF_COUNT_HW_CACHE_OP_MAX]
				[PERF_COUNT_HW_CACHE_RESULT_MAX] = {
[C(L1D)] = {
	[C(OP_READ)] = {
		[C(RESULT_ACCESS)]	= { 0x46, CNTR_EVEN | CNTR_ODD },
		[C(RESULT_MISS)]	= { 0x49, CNTR_EVEN | CNTR_ODD },
	},
	[C(OP_WRITE)] = {
		[C(RESULT_ACCESS)]	= { 0x47, CNTR_EVEN | CNTR_ODD },
		[C(RESULT_MISS)]	= { 0x4a, CNTR_EVEN | CNTR_ODD },
	},
},
[C(L1I)] = {
	[C(OP_READ)] = {
		[C(RESULT_ACCESS)]	= { 0x84, CNTR_EVEN | CNTR_ODD },
		[C(RESULT_MISS)]	= { 0x85, CNTR_EVEN | CNTR_ODD },
	},
},
[C(DTLB)] = {
	/* Can't distinguish read & write */
	[C(OP_READ)] = {
		[C(RESULT_ACCESS)]	= { 0x40, CNTR_EVEN | CNTR_ODD },
		[C(RESULT_MISS)]	= { 0x41, CNTR_EVEN | CNTR_ODD },
	},
	[C(OP_WRITE)] = {
		[C(RESULT_ACCESS)]	= { 0x40, CNTR_EVEN | CNTR_ODD },
		[C(RESULT_MISS)]	= { 0x41, CNTR_EVEN | CNTR_ODD },
	},
},
[C(BPU)] = {
	/* Conditional branches / mispredicted */
	[C(OP_READ)] = {
		[C(RESULT_ACCESS)]	= { 0x15, CNTR_EVEN | CNTR_ODD },
		[C(RESULT_MISS)]	= { 0x16, CNTR_EVEN | CNTR_ODD },
	},
},
};

static const struct mips_perf_event loongson3_cache_map
				[PERF_COUNT_HW_CACHE_MAX]
				[PERF_COUNT_HW_CACHE_OP_MAX]
				[PERF_COUNT_HW_CACHE_RESULT_MAX] = {
[C(L1D)] = {
	/*
	 * Like some other architectures (e.g. ARM), the performance
	 * counters don't differentiate between read and write
	 * accesses/misses, so this isn't strictly correct, but it's the
	 * best we can do. Writes and reads get combined.
	 */
	[C(OP_READ)] = {
		[C(RESULT_MISS)]        = { 0x04, CNTR_ODD },
	},
	[C(OP_WRITE)] = {
		[C(RESULT_MISS)]        = { 0x04, CNTR_ODD },
	},
},
[C(L1I)] = {
	[C(OP_READ)] = {
		[C(RESULT_MISS)]        = { 0x04, CNTR_EVEN },
	},
	[C(OP_WRITE)] = {
		[C(RESULT_MISS)]        = { 0x04, CNTR_EVEN },
	},
},
[C(DTLB)] = {
	[C(OP_READ)] = {
		[C(RESULT_MISS)]        = { 0x09, CNTR_ODD },
	},
	[C(OP_WRITE)] = {
		[C(RESULT_MISS)]        = { 0x09, CNTR_ODD },
	},
},
[C(ITLB)] = {
	[C(OP_READ)] = {
		[C(RESULT_MISS)]        = { 0x0c, CNTR_ODD },
	},
	[C(OP_WRITE)] = {
		[C(RESULT_MISS)]        = { 0x0c, CNTR_ODD },
	},
},
[C(BPU)] = {
	/* Using the same code for *HW_BRANCH* */
	[C(OP_READ)] = {
		[C(RESULT_ACCESS)]      = { 0x02, CNTR_EVEN },
		[C(RESULT_MISS)]        = { 0x02, CNTR_ODD },
	},
	[C(OP_WRITE)] = {
		[C(RESULT_ACCESS)]      = { 0x02, CNTR_EVEN },
		[C(RESULT_MISS)]        = { 0x02, CNTR_ODD },
	},
},
};

/* BMIPS5000 */
static const struct mips_perf_event bmips5000_cache_map
				[PERF_COUNT_HW_CACHE_MAX]
				[PERF_COUNT_HW_CACHE_OP_MAX]
				[PERF_COUNT_HW_CACHE_RESULT_MAX] = {
[C(L1D)] = {
	/*
	 * Like some other architectures (e.g. ARM), the performance
	 * counters don't differentiate between read and write
	 * accesses/misses, so this isn't strictly correct, but it's the
	 * best we can do. Writes and reads get combined.
	 */
	[C(OP_READ)] = {
		[C(RESULT_ACCESS)]	= { 12, CNTR_EVEN, T },
		[C(RESULT_MISS)]	= { 12, CNTR_ODD, T },
	},
	[C(OP_WRITE)] = {
		[C(RESULT_ACCESS)]	= { 12, CNTR_EVEN, T },
		[C(RESULT_MISS)]	= { 12, CNTR_ODD, T },
	},
},
[C(L1I)] = {
	[C(OP_READ)] = {
		[C(RESULT_ACCESS)]	= { 10, CNTR_EVEN, T },
		[C(RESULT_MISS)]	= { 10, CNTR_ODD, T },
	},
	[C(OP_WRITE)] = {
		[C(RESULT_ACCESS)]	= { 10, CNTR_EVEN, T },
		[C(RESULT_MISS)]	= { 10, CNTR_ODD, T },
	},
	[C(OP_PREFETCH)] = {
		[C(RESULT_ACCESS)]	= { 23, CNTR_EVEN, T },
		/*
		 * Note that MIPS has only "hit" events countable for
		 * the prefetch operation.
		 */
	},
},
[C(LL)] = {
	[C(OP_READ)] = {
		[C(RESULT_ACCESS)]	= { 28, CNTR_EVEN, P },
		[C(RESULT_MISS)]	= { 28, CNTR_ODD, P },
	},
	[C(OP_WRITE)] = {
		[C(RESULT_ACCESS)]	= { 28, CNTR_EVEN, P },
		[C(RESULT_MISS)]	= { 28, CNTR_ODD, P },
	},
},
[C(BPU)] = {
	/* Using the same code for *HW_BRANCH* */
	[C(OP_READ)] = {
		[C(RESULT_MISS)]	= { 0x02, CNTR_ODD, T },
	},
	[C(OP_WRITE)] = {
		[C(RESULT_MISS)]	= { 0x02, CNTR_ODD, T },
	},
},
};


static const struct mips_perf_event octeon_cache_map
				[PERF_COUNT_HW_CACHE_MAX]
				[PERF_COUNT_HW_CACHE_OP_MAX]
				[PERF_COUNT_HW_CACHE_RESULT_MAX] = {
[C(L1D)] = {
	[C(OP_READ)] = {
		[C(RESULT_ACCESS)]	= { 0x2b, CNTR_ALL },
		[C(RESULT_MISS)]	= { 0x2e, CNTR_ALL },
	},
	[C(OP_WRITE)] = {
		[C(RESULT_ACCESS)]	= { 0x30, CNTR_ALL },
	},
},
[C(L1I)] = {
	[C(OP_READ)] = {
		[C(RESULT_ACCESS)]	= { 0x18, CNTR_ALL },
	},
	[C(OP_PREFETCH)] = {
		[C(RESULT_ACCESS)]	= { 0x19, CNTR_ALL },
	},
},
[C(DTLB)] = {
	/*
	 * Only general DTLB misses are counted use the same event for
	 * read and write.
	 */
	[C(OP_READ)] = {
		[C(RESULT_MISS)]	= { 0x35, CNTR_ALL },
	},
	[C(OP_WRITE)] = {
		[C(RESULT_MISS)]	= { 0x35, CNTR_ALL },
	},
},
[C(ITLB)] = {
	[C(OP_READ)] = {
		[C(RESULT_MISS)]	= { 0x37, CNTR_ALL },
	},
},
};

static const struct mips_perf_event xlp_cache_map
				[PERF_COUNT_HW_CACHE_MAX]
				[PERF_COUNT_HW_CACHE_OP_MAX]
				[PERF_COUNT_HW_CACHE_RESULT_MAX] = {
[C(L1D)] = {
	[C(OP_READ)] = {
		[C(RESULT_ACCESS)]	= { 0x31, CNTR_ALL }, /* PAPI_L1_DCR */
		[C(RESULT_MISS)]	= { 0x30, CNTR_ALL }, /* PAPI_L1_LDM */
	},
	[C(OP_WRITE)] = {
		[C(RESULT_ACCESS)]	= { 0x2f, CNTR_ALL }, /* PAPI_L1_DCW */
		[C(RESULT_MISS)]	= { 0x2e, CNTR_ALL }, /* PAPI_L1_STM */
	},
},
[C(L1I)] = {
	[C(OP_READ)] = {
		[C(RESULT_ACCESS)]	= { 0x04, CNTR_ALL }, /* PAPI_L1_ICA */
		[C(RESULT_MISS)]	= { 0x07, CNTR_ALL }, /* PAPI_L1_ICM */
	},
},
[C(LL)] = {
	[C(OP_READ)] = {
		[C(RESULT_ACCESS)]	= { 0x35, CNTR_ALL }, /* PAPI_L2_DCR */
		[C(RESULT_MISS)]	= { 0x37, CNTR_ALL }, /* PAPI_L2_LDM */
	},
	[C(OP_WRITE)] = {
		[C(RESULT_ACCESS)]	= { 0x34, CNTR_ALL }, /* PAPI_L2_DCA */
		[C(RESULT_MISS)]	= { 0x36, CNTR_ALL }, /* PAPI_L2_DCM */
	},
},
[C(DTLB)] = {
	/*
	 * Only general DTLB misses are counted use the same event for
	 * read and write.
	 */
	[C(OP_READ)] = {
		[C(RESULT_MISS)]	= { 0x2d, CNTR_ALL }, /* PAPI_TLB_DM */
	},
	[C(OP_WRITE)] = {
		[C(RESULT_MISS)]	= { 0x2d, CNTR_ALL }, /* PAPI_TLB_DM */
	},
},
[C(ITLB)] = {
	[C(OP_READ)] = {
		[C(RESULT_MISS)]	= { 0x08, CNTR_ALL }, /* PAPI_TLB_IM */
	},
	[C(OP_WRITE)] = {
		[C(RESULT_MISS)]	= { 0x08, CNTR_ALL }, /* PAPI_TLB_IM */
	},
},
[C(BPU)] = {
	[C(OP_READ)] = {
		[C(RESULT_MISS)]	= { 0x25, CNTR_ALL },
	},
},
};

static int __hw_perf_event_init(struct perf_event *event)
{
	struct perf_event_attr *attr = &event->attr;
	struct hw_perf_event *hwc = &event->hw;
	const struct mips_perf_event *pev;
	int err;

	/* Returning MIPS event descriptor for generic perf event. */
	if (PERF_TYPE_HARDWARE == event->attr.type) {
		if (event->attr.config >= PERF_COUNT_HW_MAX)
			return -EINVAL;
		pev = mipspmu_map_general_event(event->attr.config);
	} else if (PERF_TYPE_HW_CACHE == event->attr.type) {
		pev = mipspmu_map_cache_event(event->attr.config);
	} else if (PERF_TYPE_RAW == event->attr.type) {
		/* We are working on the global raw event. */
		mutex_lock(&raw_event_mutex);
		pev = mipspmu.map_raw_event(event->attr.config);
	} else {
		/* The event type is not (yet) supported. */
		return -EOPNOTSUPP;
	}

	if (IS_ERR(pev)) {
		if (PERF_TYPE_RAW == event->attr.type)
			mutex_unlock(&raw_event_mutex);
		return PTR_ERR(pev);
	}

	/*
	 * We allow max flexibility on how each individual counter shared
	 * by the single CPU operates (the mode exclusion and the range).
	 */
	hwc->config_base = MIPS_PERFCTRL_IE;

	hwc->event_base = mipspmu_perf_event_encode(pev);
	if (PERF_TYPE_RAW == event->attr.type)
		mutex_unlock(&raw_event_mutex);

	if (!attr->exclude_user)
		hwc->config_base |= MIPS_PERFCTRL_U;
	if (!attr->exclude_kernel) {
		hwc->config_base |= MIPS_PERFCTRL_K;
		/* MIPS kernel mode: KSU == 00b || EXL == 1 || ERL == 1 */
		hwc->config_base |= MIPS_PERFCTRL_EXL;
	}
	if (!attr->exclude_hv)
		hwc->config_base |= MIPS_PERFCTRL_S;

	hwc->config_base &= M_PERFCTL_CONFIG_MASK;
	/*
	 * The event can belong to another cpu. We do not assign a local
	 * counter for it for now.
	 */
	hwc->idx = -1;
	hwc->config = 0;

	if (!hwc->sample_period) {
		hwc->sample_period  = mipspmu.max_period;
		hwc->last_period    = hwc->sample_period;
		local64_set(&hwc->period_left, hwc->sample_period);
	}

	err = 0;
	if (event->group_leader != event)
		err = validate_group(event);

	event->destroy = hw_perf_event_destroy;

	if (err)
		event->destroy(event);

	return err;
}

static void pause_local_counters(void)
{
	struct cpu_hw_events *cpuc = this_cpu_ptr(&cpu_hw_events);
	int ctr = mipspmu.num_counters;
	unsigned long flags;

	local_irq_save(flags);
	do {
		ctr--;
		cpuc->saved_ctrl[ctr] = mipsxx_pmu_read_control(ctr);
		mipsxx_pmu_write_control(ctr, cpuc->saved_ctrl[ctr] &
					 ~M_PERFCTL_COUNT_EVENT_WHENEVER);
	} while (ctr > 0);
	local_irq_restore(flags);
}

static void resume_local_counters(void)
{
	struct cpu_hw_events *cpuc = this_cpu_ptr(&cpu_hw_events);
	int ctr = mipspmu.num_counters;

	do {
		ctr--;
		mipsxx_pmu_write_control(ctr, cpuc->saved_ctrl[ctr]);
	} while (ctr > 0);
}

static int mipsxx_pmu_handle_shared_irq(void)
{
	struct cpu_hw_events *cpuc = this_cpu_ptr(&cpu_hw_events);
	struct perf_sample_data data;
	unsigned int counters = mipspmu.num_counters;
	u64 counter;
	int handled = IRQ_NONE;
	struct pt_regs *regs;

	if (cpu_has_perf_cntr_intr_bit && !(read_c0_cause() & CAUSEF_PCI))
		return handled;
	/*
	 * First we pause the local counters, so that when we are locked
	 * here, the counters are all paused. When it gets locked due to
	 * perf_disable(), the timer interrupt handler will be delayed.
	 *
	 * See also mipsxx_pmu_start().
	 */
	pause_local_counters();
#ifdef CONFIG_MIPS_PERF_SHARED_TC_COUNTERS
	read_lock(&pmuint_rwlock);
#endif

	regs = get_irq_regs();

	perf_sample_data_init(&data, 0, 0);

	switch (counters) {
#define HANDLE_COUNTER(n)						\
	case n + 1:							\
		if (test_bit(n, cpuc->used_mask)) {			\
			counter = mipspmu.read_counter(n);		\
			if (counter & mipspmu.overflow) {		\
				handle_associated_event(cpuc, n, &data, regs); \
				handled = IRQ_HANDLED;			\
			}						\
		}
	HANDLE_COUNTER(3)
	HANDLE_COUNTER(2)
	HANDLE_COUNTER(1)
	HANDLE_COUNTER(0)
	}

#ifdef CONFIG_MIPS_PERF_SHARED_TC_COUNTERS
	read_unlock(&pmuint_rwlock);
#endif
	resume_local_counters();

	/*
	 * Do all the work for the pending perf events. We can do this
	 * in here because the performance counter interrupt is a regular
	 * interrupt, not NMI.
	 */
	if (handled == IRQ_HANDLED)
		irq_work_run();

	return handled;
}

static irqreturn_t mipsxx_pmu_handle_irq(int irq, void *dev)
{
	return mipsxx_pmu_handle_shared_irq();
}

/* 24K */
#define IS_BOTH_COUNTERS_24K_EVENT(b)					\
	((b) == 0 || (b) == 1 || (b) == 11)

/* 34K */
#define IS_BOTH_COUNTERS_34K_EVENT(b)					\
	((b) == 0 || (b) == 1 || (b) == 11)
#ifdef CONFIG_MIPS_MT_SMP
#define IS_RANGE_P_34K_EVENT(r, b)					\
	((b) == 0 || (r) == 18 || (b) == 21 || (b) == 22 ||		\
	 (b) == 25 || (b) == 39 || (r) == 44 || (r) == 174 ||		\
	 (r) == 176 || ((b) >= 50 && (b) <= 55) ||			\
	 ((b) >= 64 && (b) <= 67))
#define IS_RANGE_V_34K_EVENT(r) ((r) == 47)
#endif

/* 74K */
#define IS_BOTH_COUNTERS_74K_EVENT(b)					\
	((b) == 0 || (b) == 1)

/* proAptiv */
#define IS_BOTH_COUNTERS_PROAPTIV_EVENT(b)				\
	((b) == 0 || (b) == 1)
/* P5600 */
#define IS_BOTH_COUNTERS_P5600_EVENT(b)					\
	((b) == 0 || (b) == 1)

/* 1004K */
#define IS_BOTH_COUNTERS_1004K_EVENT(b)					\
	((b) == 0 || (b) == 1 || (b) == 11)
#ifdef CONFIG_MIPS_MT_SMP
#define IS_RANGE_P_1004K_EVENT(r, b)					\
	((b) == 0 || (r) == 18 || (b) == 21 || (b) == 22 ||		\
	 (b) == 25 || (b) == 36 || (b) == 39 || (r) == 44 ||		\
	 (r) == 174 || (r) == 176 || ((b) >= 50 && (b) <= 59) ||	\
	 (r) == 188 || (b) == 61 || (b) == 62 ||			\
	 ((b) >= 64 && (b) <= 67))
#define IS_RANGE_V_1004K_EVENT(r)	((r) == 47)
#endif

/* interAptiv */
#define IS_BOTH_COUNTERS_INTERAPTIV_EVENT(b)				\
	((b) == 0 || (b) == 1 || (b) == 11)
#ifdef CONFIG_MIPS_MT_SMP
/* The P/V/T info is not provided for "(b) == 38" in SUM, assume P. */
#define IS_RANGE_P_INTERAPTIV_EVENT(r, b)				\
	((b) == 0 || (r) == 18 || (b) == 21 || (b) == 22 ||		\
	 (b) == 25 || (b) == 36 || (b) == 38 || (b) == 39 ||		\
	 (r) == 44 || (r) == 174 || (r) == 176 || ((b) >= 50 &&		\
	 (b) <= 59) || (r) == 188 || (b) == 61 || (b) == 62 ||		\
	 ((b) >= 64 && (b) <= 67))
#define IS_RANGE_V_INTERAPTIV_EVENT(r)	((r) == 47 || (r) == 175)
#endif

/* BMIPS5000 */
#define IS_BOTH_COUNTERS_BMIPS5000_EVENT(b)				\
	((b) == 0 || (b) == 1)


/*
 * For most cores the user can use 0-255 raw events, where 0-127 for the events
 * of even counters, and 128-255 for odd counters. Note that bit 7 is used to
 * indicate the even/odd bank selector. So, for example, when user wants to take
 * the Event Num of 15 for odd counters (by referring to the user manual), then
 * 128 needs to be added to 15 as the input for the event config, i.e., 143 (0x8F)
 * to be used.
 *
 * Some newer cores have even more events, in which case the user can use raw
 * events 0-511, where 0-255 are for the events of even counters, and 256-511
 * are for odd counters, so bit 8 is used to indicate the even/odd bank selector.
 */
static const struct mips_perf_event *mipsxx_pmu_map_raw_event(u64 config)
{
	/* currently most cores have 7-bit event numbers */
	unsigned int raw_id = config & 0xff;
	unsigned int base_id = raw_id & 0x7f;

	switch (current_cpu_type()) {
	case CPU_24K:
		if (IS_BOTH_COUNTERS_24K_EVENT(base_id))
			raw_event.cntr_mask = CNTR_EVEN | CNTR_ODD;
		else
			raw_event.cntr_mask =
				raw_id > 127 ? CNTR_ODD : CNTR_EVEN;
#ifdef CONFIG_MIPS_MT_SMP
		/*
		 * This is actually doing nothing. Non-multithreading
		 * CPUs will not check and calculate the range.
		 */
		raw_event.range = P;
#endif
		break;
	case CPU_34K:
		if (IS_BOTH_COUNTERS_34K_EVENT(base_id))
			raw_event.cntr_mask = CNTR_EVEN | CNTR_ODD;
		else
			raw_event.cntr_mask =
				raw_id > 127 ? CNTR_ODD : CNTR_EVEN;
#ifdef CONFIG_MIPS_MT_SMP
		if (IS_RANGE_P_34K_EVENT(raw_id, base_id))
			raw_event.range = P;
		else if (unlikely(IS_RANGE_V_34K_EVENT(raw_id)))
			raw_event.range = V;
		else
			raw_event.range = T;
#endif
		break;
	case CPU_74K:
	case CPU_1074K:
		if (IS_BOTH_COUNTERS_74K_EVENT(base_id))
			raw_event.cntr_mask = CNTR_EVEN | CNTR_ODD;
		else
			raw_event.cntr_mask =
				raw_id > 127 ? CNTR_ODD : CNTR_EVEN;
#ifdef CONFIG_MIPS_MT_SMP
		raw_event.range = P;
#endif
		break;
	case CPU_PROAPTIV:
		if (IS_BOTH_COUNTERS_PROAPTIV_EVENT(base_id))
			raw_event.cntr_mask = CNTR_EVEN | CNTR_ODD;
		else
			raw_event.cntr_mask =
				raw_id > 127 ? CNTR_ODD : CNTR_EVEN;
#ifdef CONFIG_MIPS_MT_SMP
		raw_event.range = P;
#endif
		break;
	case CPU_P5600:
	case CPU_P6600:
		/* 8-bit event numbers */
		raw_id = config & 0x1ff;
		base_id = raw_id & 0xff;
		if (IS_BOTH_COUNTERS_P5600_EVENT(base_id))
			raw_event.cntr_mask = CNTR_EVEN | CNTR_ODD;
		else
			raw_event.cntr_mask =
				raw_id > 255 ? CNTR_ODD : CNTR_EVEN;
#ifdef CONFIG_MIPS_MT_SMP
		raw_event.range = P;
#endif
		break;
	case CPU_I6400:
	case CPU_I6500:
		/* 8-bit event numbers */
		base_id = config & 0xff;
		raw_event.cntr_mask = CNTR_EVEN | CNTR_ODD;
		break;
	case CPU_1004K:
		if (IS_BOTH_COUNTERS_1004K_EVENT(base_id))
			raw_event.cntr_mask = CNTR_EVEN | CNTR_ODD;
		else
			raw_event.cntr_mask =
				raw_id > 127 ? CNTR_ODD : CNTR_EVEN;
#ifdef CONFIG_MIPS_MT_SMP
		if (IS_RANGE_P_1004K_EVENT(raw_id, base_id))
			raw_event.range = P;
		else if (unlikely(IS_RANGE_V_1004K_EVENT(raw_id)))
			raw_event.range = V;
		else
			raw_event.range = T;
#endif
		break;
	case CPU_INTERAPTIV:
		if (IS_BOTH_COUNTERS_INTERAPTIV_EVENT(base_id))
			raw_event.cntr_mask = CNTR_EVEN | CNTR_ODD;
		else
			raw_event.cntr_mask =
				raw_id > 127 ? CNTR_ODD : CNTR_EVEN;
#ifdef CONFIG_MIPS_MT_SMP
		if (IS_RANGE_P_INTERAPTIV_EVENT(raw_id, base_id))
			raw_event.range = P;
		else if (unlikely(IS_RANGE_V_INTERAPTIV_EVENT(raw_id)))
			raw_event.range = V;
		else
			raw_event.range = T;
#endif
		break;
	case CPU_BMIPS5000:
		if (IS_BOTH_COUNTERS_BMIPS5000_EVENT(base_id))
			raw_event.cntr_mask = CNTR_EVEN | CNTR_ODD;
		else
			raw_event.cntr_mask =
				raw_id > 127 ? CNTR_ODD : CNTR_EVEN;
		break;
	case CPU_LOONGSON3:
		raw_event.cntr_mask = raw_id > 127 ? CNTR_ODD : CNTR_EVEN;
	break;
	}

	raw_event.event_id = base_id;

	return &raw_event;
}

static const struct mips_perf_event *octeon_pmu_map_raw_event(u64 config)
{
	unsigned int raw_id = config & 0xff;
	unsigned int base_id = raw_id & 0x7f;


	raw_event.cntr_mask = CNTR_ALL;
	raw_event.event_id = base_id;

	if (current_cpu_type() == CPU_CAVIUM_OCTEON2) {
		if (base_id > 0x42)
			return ERR_PTR(-EOPNOTSUPP);
	} else {
		if (base_id > 0x3a)
			return ERR_PTR(-EOPNOTSUPP);
	}

	switch (base_id) {
	case 0x00:
	case 0x0f:
	case 0x1e:
	case 0x1f:
	case 0x2f:
	case 0x34:
	case 0x3b ... 0x3f:
		return ERR_PTR(-EOPNOTSUPP);
	default:
		break;
	}

	return &raw_event;
}

static const struct mips_perf_event *xlp_pmu_map_raw_event(u64 config)
{
	unsigned int raw_id = config & 0xff;

	/* Only 1-63 are defined */
	if ((raw_id < 0x01) || (raw_id > 0x3f))
		return ERR_PTR(-EOPNOTSUPP);

	raw_event.cntr_mask = CNTR_ALL;
	raw_event.event_id = raw_id;

	return &raw_event;
}

static int __init
init_hw_perf_events(void)
{
	int counters, irq;
	int counter_bits;

	pr_info("Performance counters: ");

	counters = n_counters();
	if (counters == 0) {
		pr_cont("No available PMU.\n");
		return -ENODEV;
	}

#ifdef CONFIG_MIPS_PERF_SHARED_TC_COUNTERS
	if (!cpu_has_mipsmt_pertccounters)
		counters = counters_total_to_per_cpu(counters);
#endif

	if (get_c0_perfcount_int)
		irq = get_c0_perfcount_int();
	else if (cp0_perfcount_irq >= 0)
		irq = MIPS_CPU_IRQ_BASE + cp0_perfcount_irq;
	else
		irq = -1;

	mipspmu.map_raw_event = mipsxx_pmu_map_raw_event;

	switch (current_cpu_type()) {
	case CPU_24K:
		mipspmu.name = "mips/24K";
		mipspmu.general_event_map = &mipsxxcore_event_map;
		mipspmu.cache_event_map = &mipsxxcore_cache_map;
		break;
	case CPU_34K:
		mipspmu.name = "mips/34K";
		mipspmu.general_event_map = &mipsxxcore_event_map;
		mipspmu.cache_event_map = &mipsxxcore_cache_map;
		break;
	case CPU_74K:
		mipspmu.name = "mips/74K";
		mipspmu.general_event_map = &mipsxxcore_event_map2;
		mipspmu.cache_event_map = &mipsxxcore_cache_map2;
		break;
	case CPU_PROAPTIV:
		mipspmu.name = "mips/proAptiv";
		mipspmu.general_event_map = &mipsxxcore_event_map2;
		mipspmu.cache_event_map = &mipsxxcore_cache_map2;
		break;
	case CPU_P5600:
		mipspmu.name = "mips/P5600";
		mipspmu.general_event_map = &mipsxxcore_event_map2;
		mipspmu.cache_event_map = &mipsxxcore_cache_map2;
		break;
	case CPU_P6600:
		mipspmu.name = "mips/P6600";
		mipspmu.general_event_map = &mipsxxcore_event_map2;
		mipspmu.cache_event_map = &mipsxxcore_cache_map2;
		break;
	case CPU_I6400:
		mipspmu.name = "mips/I6400";
		mipspmu.general_event_map = &i6x00_event_map;
		mipspmu.cache_event_map = &i6x00_cache_map;
		break;
	case CPU_I6500:
		mipspmu.name = "mips/I6500";
		mipspmu.general_event_map = &i6x00_event_map;
		mipspmu.cache_event_map = &i6x00_cache_map;
		break;
	case CPU_1004K:
		mipspmu.name = "mips/1004K";
		mipspmu.general_event_map = &mipsxxcore_event_map;
		mipspmu.cache_event_map = &mipsxxcore_cache_map;
		break;
	case CPU_1074K:
		mipspmu.name = "mips/1074K";
		mipspmu.general_event_map = &mipsxxcore_event_map;
		mipspmu.cache_event_map = &mipsxxcore_cache_map;
		break;
	case CPU_INTERAPTIV:
		mipspmu.name = "mips/interAptiv";
		mipspmu.general_event_map = &mipsxxcore_event_map;
		mipspmu.cache_event_map = &mipsxxcore_cache_map;
		break;
	case CPU_LOONGSON1:
		mipspmu.name = "mips/loongson1";
		mipspmu.general_event_map = &mipsxxcore_event_map;
		mipspmu.cache_event_map = &mipsxxcore_cache_map;
		break;
	case CPU_LOONGSON3:
		mipspmu.name = "mips/loongson3";
		mipspmu.general_event_map = &loongson3_event_map;
		mipspmu.cache_event_map = &loongson3_cache_map;
		break;
	case CPU_CAVIUM_OCTEON:
	case CPU_CAVIUM_OCTEON_PLUS:
	case CPU_CAVIUM_OCTEON2:
		mipspmu.name = "octeon";
		mipspmu.general_event_map = &octeon_event_map;
		mipspmu.cache_event_map = &octeon_cache_map;
		mipspmu.map_raw_event = octeon_pmu_map_raw_event;
		break;
	case CPU_BMIPS5000:
		mipspmu.name = "BMIPS5000";
		mipspmu.general_event_map = &bmips5000_event_map;
		mipspmu.cache_event_map = &bmips5000_cache_map;
		break;
	case CPU_XLP:
		mipspmu.name = "xlp";
		mipspmu.general_event_map = &xlp_event_map;
		mipspmu.cache_event_map = &xlp_cache_map;
		mipspmu.map_raw_event = xlp_pmu_map_raw_event;
		break;
	default:
		pr_cont("Either hardware does not support performance "
			"counters, or not yet implemented.\n");
		return -ENODEV;
	}

	mipspmu.num_counters = counters;
	mipspmu.irq = irq;

	if (read_c0_perfctrl0() & MIPS_PERFCTRL_W) {
		mipspmu.max_period = (1ULL << 63) - 1;
		mipspmu.valid_count = (1ULL << 63) - 1;
		mipspmu.overflow = 1ULL << 63;
		mipspmu.read_counter = mipsxx_pmu_read_counter_64;
		mipspmu.write_counter = mipsxx_pmu_write_counter_64;
		counter_bits = 64;
	} else {
		mipspmu.max_period = (1ULL << 31) - 1;
		mipspmu.valid_count = (1ULL << 31) - 1;
		mipspmu.overflow = 1ULL << 31;
		mipspmu.read_counter = mipsxx_pmu_read_counter;
		mipspmu.write_counter = mipsxx_pmu_write_counter;
		counter_bits = 32;
	}

	on_each_cpu(reset_counters, (void *)(long)counters, 1);

	pr_cont("%s PMU enabled, %d %d-bit counters available to each "
		"CPU, irq %d%s\n", mipspmu.name, counters, counter_bits, irq,
		irq < 0 ? " (share with timer interrupt)" : "");

	perf_pmu_register(&pmu, "cpu", PERF_TYPE_RAW);

	return 0;
}
early_initcall(init_hw_perf_events);