Merge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/paulus/perfcounters into perfcounters/core

15 files changed, 1607 insertions, 10 deletions

diff --git a/arch/powerpc/include/asm/hw_irq.h b/arch/powerpc/include/asm/hw_irq.h
index f75a5fc64d2e..e10f151c3db6 100644
--- a/arch/powerpc/include/asm/hw_irq.h
+++ b/arch/powerpc/include/asm/hw_irq.h
@@ -131,5 +131,36 @@ static inline int irqs_disabled_flags(unsigned long flags)
  */
 struct hw_interrupt_type;
 
+#ifdef CONFIG_PERF_COUNTERS
+static inline unsigned long get_perf_counter_pending(void)
+{
+	unsigned long x;
+
+	asm volatile("lbz %0,%1(13)"
+		: "=r" (x)
+		: "i" (offsetof(struct paca_struct, perf_counter_pending)));
+	return x;
+}
+
+static inline void set_perf_counter_pending(int x)
+{
+	asm volatile("stb %0,%1(13)" : :
+		"r" (x),
+		"i" (offsetof(struct paca_struct, perf_counter_pending)));
+}
+
+extern void perf_counter_do_pending(void);
+
+#else
+
+static inline unsigned long get_perf_counter_pending(void)
+{
+	return 0;
+}
+
+static inline void set_perf_counter_pending(int x) {}
+static inline void perf_counter_do_pending(void) {}
+#endif /* CONFIG_PERF_COUNTERS */
+
 #endif	/* __KERNEL__ */
 #endif	/* _ASM_POWERPC_HW_IRQ_H */
diff --git a/arch/powerpc/include/asm/paca.h b/arch/powerpc/include/asm/paca.h
index 082b3aedf145..6ef055723019 100644
--- a/arch/powerpc/include/asm/paca.h
+++ b/arch/powerpc/include/asm/paca.h
@@ -99,6 +99,7 @@ struct paca_struct {
 	u8 soft_enabled;		/* irq soft-enable flag */
 	u8 hard_enabled;		/* set if irqs are enabled in MSR */
 	u8 io_sync;			/* writel() needs spin_unlock sync */
+	u8 perf_counter_pending;	/* PM interrupt while soft-disabled */
 
 	/* Stuff for accurate time accounting */
 	u64 user_time;			/* accumulated usermode TB ticks */
diff --git a/arch/powerpc/include/asm/perf_counter.h b/arch/powerpc/include/asm/perf_counter.h
new file mode 100644
index 000000000000..9d7ff6d7fb56
--- /dev/null
+++ b/arch/powerpc/include/asm/perf_counter.h
@@ -0,0 +1,72 @@
+/*
+ * Performance counter support - PowerPC-specific definitions.
+ *
+ * Copyright 2008-2009 Paul Mackerras, IBM Corporation.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version
+ * 2 of the License, or (at your option) any later version.
+ */
+#include <linux/types.h>
+
+#define MAX_HWCOUNTERS		8
+#define MAX_EVENT_ALTERNATIVES	8
+
+/*
+ * This struct provides the constants and functions needed to
+ * describe the PMU on a particular POWER-family CPU.
+ */
+struct power_pmu {
+	int	n_counter;
+	int	max_alternatives;
+	u64	add_fields;
+	u64	test_adder;
+	int	(*compute_mmcr)(unsigned int events[], int n_ev,
+				unsigned int hwc[], u64 mmcr[]);
+	int	(*get_constraint)(unsigned int event, u64 *mskp, u64 *valp);
+	int	(*get_alternatives)(unsigned int event, unsigned int alt[]);
+	void	(*disable_pmc)(unsigned int pmc, u64 mmcr[]);
+	int	n_generic;
+	int	*generic_events;
+};
+
+extern struct power_pmu *ppmu;
+
+/*
+ * The power_pmu.get_constraint function returns a 64-bit value and
+ * a 64-bit mask that express the constraints between this event and
+ * other events.
+ *
+ * The value and mask are divided up into (non-overlapping) bitfields
+ * of three different types:
+ *
+ * Select field: this expresses the constraint that some set of bits
+ * in MMCR* needs to be set to a specific value for this event.  For a
+ * select field, the mask contains 1s in every bit of the field, and
+ * the value contains a unique value for each possible setting of the
+ * MMCR* bits.  The constraint checking code will ensure that two events
+ * that set the same field in their masks have the same value in their
+ * value dwords.
+ *
+ * Add field: this expresses the constraint that there can be at most
+ * N events in a particular class.  A field of k bits can be used for
+ * N <= 2^(k-1) - 1.  The mask has the most significant bit of the field
+ * set (and the other bits 0), and the value has only the least significant
+ * bit of the field set.  In addition, the 'add_fields' and 'test_adder'
+ * in the struct power_pmu for this processor come into play.  The
+ * add_fields value contains 1 in the LSB of the field, and the
+ * test_adder contains 2^(k-1) - 1 - N in the field.
+ *
+ * NAND field: this expresses the constraint that you may not have events
+ * in all of a set of classes.  (For example, on PPC970, you can't select
+ * events from the FPU, ISU and IDU simultaneously, although any two are
+ * possible.)  For N classes, the field is N+1 bits wide, and each class
+ * is assigned one bit from the least-significant N bits.  The mask has
+ * only the most-significant bit set, and the value has only the bit
+ * for the event's class set.  The test_adder has the least significant
+ * bit set in the field.
+ *
+ * If an event is not subject to the constraint expressed by a particular
+ * field, then it will have 0 in both the mask and value for that field.
+ */
diff --git a/arch/powerpc/include/asm/systbl.h b/arch/powerpc/include/asm/systbl.h
index 803def236654..da300c4d2888 100644
--- a/arch/powerpc/include/asm/systbl.h
+++ b/arch/powerpc/include/asm/systbl.h
@@ -322,3 +322,4 @@ SYSCALL_SPU(epoll_create1)
 SYSCALL_SPU(dup3)
 SYSCALL_SPU(pipe2)
 SYSCALL(inotify_init1)
+SYSCALL(perf_counter_open)
diff --git a/arch/powerpc/include/asm/unistd.h b/arch/powerpc/include/asm/unistd.h
index e07d0c76ed77..7cef5afe89d8 100644
--- a/arch/powerpc/include/asm/unistd.h
+++ b/arch/powerpc/include/asm/unistd.h
@@ -341,10 +341,11 @@
 #define __NR_dup3		316
 #define __NR_pipe2		317
 #define __NR_inotify_init1	318
+#define __NR_perf_counter_open	319
 
 #ifdef __KERNEL__
 
-#define __NR_syscalls		319
+#define __NR_syscalls		320
 
 #define __NR__exit __NR_exit
 #define NR_syscalls	__NR_syscalls
diff --git a/arch/powerpc/kernel/Makefile b/arch/powerpc/kernel/Makefile
index 8d1a419df35d..7c941ec3b23e 100644
--- a/arch/powerpc/kernel/Makefile
+++ b/arch/powerpc/kernel/Makefile
@@ -94,6 +94,7 @@ obj-$(CONFIG_AUDIT)		+= audit.o
 obj64-$(CONFIG_AUDIT)		+= compat_audit.o
 
 obj-$(CONFIG_DYNAMIC_FTRACE)	+= ftrace.o
+obj-$(CONFIG_PERF_COUNTERS)	+= perf_counter.o ppc970-pmu.o power6-pmu.o
 
 obj-$(CONFIG_8XX_MINIMAL_FPEMU) += softemu8xx.o
 
diff --git a/arch/powerpc/kernel/asm-offsets.c b/arch/powerpc/kernel/asm-offsets.c
index 9937fe44555f..ce3f8f12f731 100644
--- a/arch/powerpc/kernel/asm-offsets.c
+++ b/arch/powerpc/kernel/asm-offsets.c
@@ -127,6 +127,7 @@ int main(void)
 	DEFINE(PACAKMSR, offsetof(struct paca_struct, kernel_msr));
 	DEFINE(PACASOFTIRQEN, offsetof(struct paca_struct, soft_enabled));
 	DEFINE(PACAHARDIRQEN, offsetof(struct paca_struct, hard_enabled));
+	DEFINE(PACAPERFPEND, offsetof(struct paca_struct, perf_counter_pending));
 	DEFINE(PACASLBCACHE, offsetof(struct paca_struct, slb_cache));
 	DEFINE(PACASLBCACHEPTR, offsetof(struct paca_struct, slb_cache_ptr));
 	DEFINE(PACACONTEXTID, offsetof(struct paca_struct, context.id));
diff --git a/arch/powerpc/kernel/entry_64.S b/arch/powerpc/kernel/entry_64.S
index 383ed6eb0085..f30b4e553c53 100644
--- a/arch/powerpc/kernel/entry_64.S
+++ b/arch/powerpc/kernel/entry_64.S
@@ -526,6 +526,15 @@ ALT_FW_FTR_SECTION_END_IFCLR(FW_FEATURE_ISERIES)
 2:
 	TRACE_AND_RESTORE_IRQ(r5);
 
+#ifdef CONFIG_PERF_COUNTERS
+	/* check paca->perf_counter_pending if we're enabling ints */
+	lbz	r3,PACAPERFPEND(r13)
+	and.	r3,r3,r5
+	beq	27f
+	bl	.perf_counter_do_pending
+27:
+#endif /* CONFIG_PERF_COUNTERS */
+
 	/* extract EE bit and use it to restore paca->hard_enabled */
 	ld	r3,_MSR(r1)
 	rldicl	r4,r3,49,63		/* r0 = (r3 >> 15) & 1 */
diff --git a/arch/powerpc/kernel/irq.c b/arch/powerpc/kernel/irq.c
index 23b8b5e36f98..f5ae4878ccef 100644
--- a/arch/powerpc/kernel/irq.c
+++ b/arch/powerpc/kernel/irq.c
@@ -104,6 +104,13 @@ static inline notrace void set_soft_enabled(unsigned long enable)
 	: : "r" (enable), "i" (offsetof(struct paca_struct, soft_enabled)));
 }
 
+#ifdef CONFIG_PERF_COUNTERS
+notrace void __weak perf_counter_do_pending(void)
+{
+	set_perf_counter_pending(0);
+}
+#endif
+
 notrace void raw_local_irq_restore(unsigned long en)
 {
 	/*
@@ -135,6 +142,9 @@ notrace void raw_local_irq_restore(unsigned long en)
 			iseries_handle_interrupts();
 	}
 
+	if (get_perf_counter_pending())
+		perf_counter_do_pending();
+
 	/*
 	 * if (get_paca()->hard_enabled) return;
 	 * But again we need to take care that gcc gets hard_enabled directly
diff --git a/arch/powerpc/kernel/perf_counter.c b/arch/powerpc/kernel/perf_counter.c
new file mode 100644
index 000000000000..df3fe057dee9
--- /dev/null
+++ b/arch/powerpc/kernel/perf_counter.c
@@ -0,0 +1,771 @@
+/*
+ * Performance counter support - powerpc architecture code
+ *
+ * Copyright 2008-2009 Paul Mackerras, IBM Corporation.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version
+ * 2 of the License, or (at your option) any later version.
+ */
+#include <linux/kernel.h>
+#include <linux/sched.h>
+#include <linux/perf_counter.h>
+#include <linux/percpu.h>
+#include <linux/hardirq.h>
+#include <asm/reg.h>
+#include <asm/pmc.h>
+
+struct cpu_hw_counters {
+	int n_counters;
+	int n_percpu;
+	int disabled;
+	int n_added;
+	struct perf_counter *counter[MAX_HWCOUNTERS];
+	unsigned int events[MAX_HWCOUNTERS];
+	u64 mmcr[3];
+};
+DEFINE_PER_CPU(struct cpu_hw_counters, cpu_hw_counters);
+
+struct power_pmu *ppmu;
+
+void perf_counter_print_debug(void)
+{
+}
+
+/*
+ * Return 1 for a software counter, 0 for a hardware counter
+ */
+static inline int is_software_counter(struct perf_counter *counter)
+{
+	return !counter->hw_event.raw && counter->hw_event.type < 0;
+}
+
+/*
+ * Read one performance monitor counter (PMC).
+ */
+static unsigned long read_pmc(int idx)
+{
+	unsigned long val;
+
+	switch (idx) {
+	case 1:
+		val = mfspr(SPRN_PMC1);
+		break;
+	case 2:
+		val = mfspr(SPRN_PMC2);
+		break;
+	case 3:
+		val = mfspr(SPRN_PMC3);
+		break;
+	case 4:
+		val = mfspr(SPRN_PMC4);
+		break;
+	case 5:
+		val = mfspr(SPRN_PMC5);
+		break;
+	case 6:
+		val = mfspr(SPRN_PMC6);
+		break;
+	case 7:
+		val = mfspr(SPRN_PMC7);
+		break;
+	case 8:
+		val = mfspr(SPRN_PMC8);
+		break;
+	default:
+		printk(KERN_ERR "oops trying to read PMC%d\n", idx);
+		val = 0;
+	}
+	return val;
+}
+
+/*
+ * Write one PMC.
+ */
+static void write_pmc(int idx, unsigned long val)
+{
+	switch (idx) {
+	case 1:
+		mtspr(SPRN_PMC1, val);
+		break;
+	case 2:
+		mtspr(SPRN_PMC2, val);
+		break;
+	case 3:
+		mtspr(SPRN_PMC3, val);
+		break;
+	case 4:
+		mtspr(SPRN_PMC4, val);
+		break;
+	case 5:
+		mtspr(SPRN_PMC5, val);
+		break;
+	case 6:
+		mtspr(SPRN_PMC6, val);
+		break;
+	case 7:
+		mtspr(SPRN_PMC7, val);
+		break;
+	case 8:
+		mtspr(SPRN_PMC8, val);
+		break;
+	default:
+		printk(KERN_ERR "oops trying to write PMC%d\n", idx);
+	}
+}
+
+/*
+ * Check if a set of events can all go on the PMU at once.
+ * If they can't, this will look at alternative codes for the events
+ * and see if any combination of alternative codes is feasible.
+ * The feasible set is returned in event[].
+ */
+static int power_check_constraints(unsigned int event[], int n_ev)
+{
+	u64 mask, value, nv;
+	unsigned int alternatives[MAX_HWCOUNTERS][MAX_EVENT_ALTERNATIVES];
+	u64 amasks[MAX_HWCOUNTERS][MAX_EVENT_ALTERNATIVES];
+	u64 avalues[MAX_HWCOUNTERS][MAX_EVENT_ALTERNATIVES];
+	u64 smasks[MAX_HWCOUNTERS], svalues[MAX_HWCOUNTERS];
+	int n_alt[MAX_HWCOUNTERS], choice[MAX_HWCOUNTERS];
+	int i, j;
+	u64 addf = ppmu->add_fields;
+	u64 tadd = ppmu->test_adder;
+
+	if (n_ev > ppmu->n_counter)
+		return -1;
+
+	/* First see if the events will go on as-is */
+	for (i = 0; i < n_ev; ++i) {
+		alternatives[i][0] = event[i];
+		if (ppmu->get_constraint(event[i], &amasks[i][0],
+					 &avalues[i][0]))
+			return -1;
+		choice[i] = 0;
+	}
+	value = mask = 0;
+	for (i = 0; i < n_ev; ++i) {
+		nv = (value | avalues[i][0]) + (value & avalues[i][0] & addf);
+		if ((((nv + tadd) ^ value) & mask) != 0 ||
+		    (((nv + tadd) ^ avalues[i][0]) & amasks[i][0]) != 0)
+			break;
+		value = nv;
+		mask |= amasks[i][0];
+	}
+	if (i == n_ev)
+		return 0;	/* all OK */
+
+	/* doesn't work, gather alternatives... */
+	if (!ppmu->get_alternatives)
+		return -1;
+	for (i = 0; i < n_ev; ++i) {
+		n_alt[i] = ppmu->get_alternatives(event[i], alternatives[i]);
+		for (j = 1; j < n_alt[i]; ++j)
+			ppmu->get_constraint(alternatives[i][j],
+					     &amasks[i][j], &avalues[i][j]);
+	}
+
+	/* enumerate all possibilities and see if any will work */
+	i = 0;
+	j = -1;
+	value = mask = nv = 0;
+	while (i < n_ev) {
+		if (j >= 0) {
+			/* we're backtracking, restore context */
+			value = svalues[i];
+			mask = smasks[i];
+			j = choice[i];
+		}
+		/*
+		 * See if any alternative k for event i,
+		 * where k > j, will satisfy the constraints.
+		 */
+		while (++j < n_alt[i]) {
+			nv = (value | avalues[i][j]) +
+				(value & avalues[i][j] & addf);
+			if ((((nv + tadd) ^ value) & mask) == 0 &&
+			    (((nv + tadd) ^ avalues[i][j])
+			     & amasks[i][j]) == 0)
+				break;
+		}
+		if (j >= n_alt[i]) {
+			/*
+			 * No feasible alternative, backtrack
+			 * to event i-1 and continue enumerating its
+			 * alternatives from where we got up to.
+			 */
+			if (--i < 0)
+				return -1;
+		} else {
+			/*
+			 * Found a feasible alternative for event i,
+			 * remember where we got up to with this event,
+			 * go on to the next event, and start with
+			 * the first alternative for it.
+			 */
+			choice[i] = j;
+			svalues[i] = value;
+			smasks[i] = mask;
+			value = nv;
+			mask |= amasks[i][j];
+			++i;
+			j = -1;
+		}
+	}
+
+	/* OK, we have a feasible combination, tell the caller the solution */
+	for (i = 0; i < n_ev; ++i)
+		event[i] = alternatives[i][choice[i]];
+	return 0;
+}
+
+static void power_perf_read(struct perf_counter *counter)
+{
+	long val, delta, prev;
+
+	if (!counter->hw.idx)
+		return;
+	/*
+	 * Performance monitor interrupts come even when interrupts
+	 * are soft-disabled, as long as interrupts are hard-enabled.
+	 * Therefore we treat them like NMIs.
+	 */
+	do {
+		prev = atomic64_read(&counter->hw.prev_count);
+		barrier();
+		val = read_pmc(counter->hw.idx);
+	} while (atomic64_cmpxchg(&counter->hw.prev_count, prev, val) != prev);
+
+	/* The counters are only 32 bits wide */
+	delta = (val - prev) & 0xfffffffful;
+	atomic64_add(delta, &counter->count);
+	atomic64_sub(delta, &counter->hw.period_left);
+}
+
+/*
+ * Disable all counters to prevent PMU interrupts and to allow
+ * counters to be added or removed.
+ */
+u64 hw_perf_save_disable(void)
+{
+	struct cpu_hw_counters *cpuhw;
+	unsigned long ret;
+	unsigned long flags;
+
+	local_irq_save(flags);
+	cpuhw = &__get_cpu_var(cpu_hw_counters);
+
+	ret = cpuhw->disabled;
+	if (!ret) {
+		cpuhw->disabled = 1;
+		cpuhw->n_added = 0;
+
+		/*
+		 * Set the 'freeze counters' bit.
+		 * The barrier is to make sure the mtspr has been
+		 * executed and the PMU has frozen the counters
+		 * before we return.
+		 */
+		mtspr(SPRN_MMCR0, mfspr(SPRN_MMCR0) | MMCR0_FC);
+		mb();
+	}
+	local_irq_restore(flags);
+	return ret;
+}
+
+/*
+ * Re-enable all counters if disable == 0.
+ * If we were previously disabled and counters were added, then
+ * put the new config on the PMU.
+ */
+void hw_perf_restore(u64 disable)
+{
+	struct perf_counter *counter;
+	struct cpu_hw_counters *cpuhw;
+	unsigned long flags;
+	long i;
+	unsigned long val;
+	s64 left;
+	unsigned int hwc_index[MAX_HWCOUNTERS];
+
+	if (disable)
+		return;
+	local_irq_save(flags);
+	cpuhw = &__get_cpu_var(cpu_hw_counters);
+	cpuhw->disabled = 0;
+
+	/*
+	 * If we didn't change anything, or only removed counters,
+	 * no need to recalculate MMCR* settings and reset the PMCs.
+	 * Just reenable the PMU with the current MMCR* settings
+	 * (possibly updated for removal of counters).
+	 */
+	if (!cpuhw->n_added) {
+		mtspr(SPRN_MMCRA, cpuhw->mmcr[2]);
+		mtspr(SPRN_MMCR1, cpuhw->mmcr[1]);
+		mtspr(SPRN_MMCR0, cpuhw->mmcr[0]);
+		goto out;
+	}
+
+	/*
+	 * Compute MMCR* values for the new set of counters
+	 */
+	if (ppmu->compute_mmcr(cpuhw->events, cpuhw->n_counters, hwc_index,
+			       cpuhw->mmcr)) {
+		/* shouldn't ever get here */
+		printk(KERN_ERR "oops compute_mmcr failed\n");
+		goto out;
+	}
+
+	/*
+	 * Write the new configuration to MMCR* with the freeze
+	 * bit set and set the hardware counters to their initial values.
+	 * Then unfreeze the counters.
+	 */
+	mtspr(SPRN_MMCRA, cpuhw->mmcr[2]);
+	mtspr(SPRN_MMCR1, cpuhw->mmcr[1]);
+	mtspr(SPRN_MMCR0, (cpuhw->mmcr[0] & ~(MMCR0_PMC1CE | MMCR0_PMCjCE))
+				| MMCR0_FC);
+
+	/*
+	 * Read off any pre-existing counters that need to move
+	 * to another PMC.
+	 */
+	for (i = 0; i < cpuhw->n_counters; ++i) {
+		counter = cpuhw->counter[i];
+		if (counter->hw.idx && counter->hw.idx != hwc_index[i] + 1) {
+			power_perf_read(counter);
+			write_pmc(counter->hw.idx, 0);
+			counter->hw.idx = 0;
+		}
+	}
+
+	/*
+	 * Initialize the PMCs for all the new and moved counters.
+	 */
+	for (i = 0; i < cpuhw->n_counters; ++i) {
+		counter = cpuhw->counter[i];
+		if (counter->hw.idx)
+			continue;
+		val = 0;
+		if (counter->hw_event.irq_period) {
+			left = atomic64_read(&counter->hw.period_left);
+			if (left < 0x80000000L)
+				val = 0x80000000L - left;
+		}
+		atomic64_set(&counter->hw.prev_count, val);
+		counter->hw.idx = hwc_index[i] + 1;
+		write_pmc(counter->hw.idx, val);
+	}
+	mb();
+	cpuhw->mmcr[0] |= MMCR0_PMXE | MMCR0_FCECE;
+	mtspr(SPRN_MMCR0, cpuhw->mmcr[0]);
+
+ out:
+	local_irq_restore(flags);
+}
+
+static int collect_events(struct perf_counter *group, int max_count,
+			  struct perf_counter *ctrs[], unsigned int *events)
+{
+	int n = 0;
+	struct perf_counter *counter;
+
+	if (!is_software_counter(group)) {
+		if (n >= max_count)
+			return -1;
+		ctrs[n] = group;
+		events[n++] = group->hw.config;
+	}
+	list_for_each_entry(counter, &group->sibling_list, list_entry) {
+		if (!is_software_counter(counter) &&
+		    counter->state != PERF_COUNTER_STATE_OFF) {
+			if (n >= max_count)
+				return -1;
+			ctrs[n] = counter;
+			events[n++] = counter->hw.config;
+		}
+	}
+	return n;
+}
+
+static void counter_sched_in(struct perf_counter *counter, int cpu)
+{
+	counter->state = PERF_COUNTER_STATE_ACTIVE;
+	counter->oncpu = cpu;
+	if (is_software_counter(counter))
+		counter->hw_ops->enable(counter);
+}
+
+/*
+ * Called to enable a whole group of counters.
+ * Returns 1 if the group was enabled, or -EAGAIN if it could not be.
+ * Assumes the caller has disabled interrupts and has
+ * frozen the PMU with hw_perf_save_disable.
+ */
+int hw_perf_group_sched_in(struct perf_counter *group_leader,
+	       struct perf_cpu_context *cpuctx,
+	       struct perf_counter_context *ctx, int cpu)
+{
+	struct cpu_hw_counters *cpuhw;
+	long i, n, n0;
+	struct perf_counter *sub;
+
+	cpuhw = &__get_cpu_var(cpu_hw_counters);
+	n0 = cpuhw->n_counters;
+	n = collect_events(group_leader, ppmu->n_counter - n0,
+			   &cpuhw->counter[n0], &cpuhw->events[n0]);
+	if (n < 0)
+		return -EAGAIN;
+	if (power_check_constraints(cpuhw->events, n + n0))
+		return -EAGAIN;
+	cpuhw->n_counters = n0 + n;
+	cpuhw->n_added += n;
+
+	/*
+	 * OK, this group can go on; update counter states etc.,
+	 * and enable any software counters
+	 */
+	for (i = n0; i < n0 + n; ++i)
+		cpuhw->counter[i]->hw.config = cpuhw->events[i];
+	n = 1;
+	counter_sched_in(group_leader, cpu);
+	list_for_each_entry(sub, &group_leader->sibling_list, list_entry) {
+		if (sub->state != PERF_COUNTER_STATE_OFF) {
+			counter_sched_in(sub, cpu);
+			++n;
+		}
+	}
+	cpuctx->active_oncpu += n;
+	ctx->nr_active += n;
+
+	return 1;
+}
+
+/*
+ * Add a counter to the PMU.
+ * If all counters are not already frozen, then we disable and
+ * re-enable the PMU in order to get hw_perf_restore to do the
+ * actual work of reconfiguring the PMU.
+ */
+static int power_perf_enable(struct perf_counter *counter)
+{
+	struct cpu_hw_counters *cpuhw;
+	unsigned long flags;
+	u64 pmudis;
+	int n0;
+	int ret = -EAGAIN;
+
+	local_irq_save(flags);
+	pmudis = hw_perf_save_disable();
+
+	/*
+	 * Add the counter to the list (if there is room)
+	 * and check whether the total set is still feasible.
+	 */
+	cpuhw = &__get_cpu_var(cpu_hw_counters);
+	n0 = cpuhw->n_counters;
+	if (n0 >= ppmu->n_counter)
+		goto out;
+	cpuhw->counter[n0] = counter;
+	cpuhw->events[n0] = counter->hw.config;
+	if (power_check_constraints(cpuhw->events, n0 + 1))
+		goto out;
+
+	counter->hw.config = cpuhw->events[n0];
+	++cpuhw->n_counters;
+	++cpuhw->n_added;
+
+	ret = 0;
+ out:
+	hw_perf_restore(pmudis);
+	local_irq_restore(flags);
+	return ret;
+}
+
+/*
+ * Remove a counter from the PMU.
+ */
+static void power_perf_disable(struct perf_counter *counter)
+{
+	struct cpu_hw_counters *cpuhw;
+	long i;
+	u64 pmudis;
+	unsigned long flags;
+
+	local_irq_save(flags);
+	pmudis = hw_perf_save_disable();
+
+	power_perf_read(counter);
+
+	cpuhw = &__get_cpu_var(cpu_hw_counters);
+	for (i = 0; i < cpuhw->n_counters; ++i) {
+		if (counter == cpuhw->counter[i]) {
+			while (++i < cpuhw->n_counters)
+				cpuhw->counter[i-1] = cpuhw->counter[i];
+			--cpuhw->n_counters;
+			ppmu->disable_pmc(counter->hw.idx - 1, cpuhw->mmcr);
+			write_pmc(counter->hw.idx, 0);
+			counter->hw.idx = 0;
+			break;
+		}
+	}
+	if (cpuhw->n_counters == 0) {
+		/* disable exceptions if no counters are running */
+		cpuhw->mmcr[0] &= ~(MMCR0_PMXE | MMCR0_FCECE);
+	}
+
+	hw_perf_restore(pmudis);
+	local_irq_restore(flags);
+}
+
+struct hw_perf_counter_ops power_perf_ops = {
+	.enable = power_perf_enable,
+	.disable = power_perf_disable,
+	.read = power_perf_read
+};
+
+const struct hw_perf_counter_ops *
+hw_perf_counter_init(struct perf_counter *counter)
+{
+	unsigned long ev;
+	struct perf_counter *ctrs[MAX_HWCOUNTERS];
+	unsigned int events[MAX_HWCOUNTERS];
+	int n;
+
+	if (!ppmu)
+		return NULL;
+	if ((s64)counter->hw_event.irq_period < 0)
+		return NULL;
+	ev = counter->hw_event.type;
+	if (!counter->hw_event.raw) {
+		if (ev >= ppmu->n_generic ||
+		    ppmu->generic_events[ev] == 0)
+			return NULL;
+		ev = ppmu->generic_events[ev];
+	}
+	counter->hw.config_base = ev;
+	counter->hw.idx = 0;
+
+	/*
+	 * If this is in a group, check if it can go on with all the
+	 * other hardware counters in the group.  We assume the counter
+	 * hasn't been linked into its leader's sibling list at this point.
+	 */
+	n = 0;
+	if (counter->group_leader != counter) {
+		n = collect_events(counter->group_leader, ppmu->n_counter - 1,
+				   ctrs, events);
+		if (n < 0)
+			return NULL;
+	}
+	events[n++] = ev;
+	if (power_check_constraints(events, n))
+		return NULL;
+
+	counter->hw.config = events[n - 1];
+	atomic64_set(&counter->hw.period_left, counter->hw_event.irq_period);
+	return &power_perf_ops;
+}
+
+/*
+ * Handle wakeups.
+ */
+void perf_counter_do_pending(void)
+{
+	int i;
+	struct cpu_hw_counters *cpuhw = &__get_cpu_var(cpu_hw_counters);
+	struct perf_counter *counter;
+
+	set_perf_counter_pending(0);
+	for (i = 0; i < cpuhw->n_counters; ++i) {
+		counter = cpuhw->counter[i];
+		if (counter && counter->wakeup_pending) {
+			counter->wakeup_pending = 0;
+			wake_up(&counter->waitq);
+		}
+	}
+}
+
+/*
+ * Record data for an irq counter.
+ * This function was lifted from the x86 code; maybe it should
+ * go in the core?
+ */
+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);
+	}
+}
+
+/*
+ * Record all the values of the counters in a group
+ */
+static void perf_handle_group(struct perf_counter *counter)
+{
+	struct perf_counter *leader, *sub;
+
+	leader = counter->group_leader;
+	list_for_each_entry(sub, &leader->sibling_list, list_entry) {
+		if (sub != counter)
+			sub->hw_ops->read(sub);
+		perf_store_irq_data(counter, sub->hw_event.type);
+		perf_store_irq_data(counter, atomic64_read(&sub->count));
+	}
+}
+
+/*
+ * A counter has overflowed; update its count and record
+ * things if requested.  Note that interrupts are hard-disabled
+ * here so there is no possibility of being interrupted.
+ */
+static void record_and_restart(struct perf_counter *counter, long val,
+			       struct pt_regs *regs)
+{
+	s64 prev, delta, left;
+	int record = 0;
+
+	/* we don't have to worry about interrupts here */
+	prev = atomic64_read(&counter->hw.prev_count);
+	delta = (val - prev) & 0xfffffffful;
+	atomic64_add(delta, &counter->count);
+
+	/*
+	 * See if the total period for this counter has expired,
+	 * and update for the next period.
+	 */
+	val = 0;
+	left = atomic64_read(&counter->hw.period_left) - delta;
+	if (counter->hw_event.irq_period) {
+		if (left <= 0) {
+			left += counter->hw_event.irq_period;
+			if (left <= 0)
+				left = counter->hw_event.irq_period;
+			record = 1;
+		}
+		if (left < 0x80000000L)
+			val = 0x80000000L - left;
+	}
+	write_pmc(counter->hw.idx, val);
+	atomic64_set(&counter->hw.prev_count, val);
+	atomic64_set(&counter->hw.period_left, left);
+
+	/*
+	 * Finally record data if requested.
+	 */
+	if (record) {
+		switch (counter->hw_event.record_type) {
+		case PERF_RECORD_SIMPLE:
+			break;
+		case PERF_RECORD_IRQ:
+			perf_store_irq_data(counter, instruction_pointer(regs));
+			counter->wakeup_pending = 1;
+			break;
+		case PERF_RECORD_GROUP:
+			perf_handle_group(counter);
+			counter->wakeup_pending = 1;
+			break;
+		}
+	}
+}
+
+/*
+ * Performance monitor interrupt stuff
+ */
+static void perf_counter_interrupt(struct pt_regs *regs)
+{
+	int i;
+	struct cpu_hw_counters *cpuhw = &__get_cpu_var(cpu_hw_counters);
+	struct perf_counter *counter;
+	long val;
+	int need_wakeup = 0, found = 0;
+
+	for (i = 0; i < cpuhw->n_counters; ++i) {
+		counter = cpuhw->counter[i];
+		val = read_pmc(counter->hw.idx);
+		if ((int)val < 0) {
+			/* counter has overflowed */
+			found = 1;
+			record_and_restart(counter, val, regs);
+			if (counter->wakeup_pending)
+				need_wakeup = 1;
+		}
+	}
+
+	/*
+	 * In case we didn't find and reset the counter that caused
+	 * the interrupt, scan all counters and reset any that are
+	 * negative, to avoid getting continual interrupts.
+	 * Any that we processed in the previous loop will not be negative.
+	 */
+	if (!found) {
+		for (i = 0; i < ppmu->n_counter; ++i) {
+			val = read_pmc(i + 1);
+			if ((int)val < 0)
+				write_pmc(i + 1, 0);
+		}
+	}
+
+	/*
+	 * Reset MMCR0 to its normal value.  This will set PMXE and
+	 * clear FC (freeze counters) and PMAO (perf mon alert occurred)
+	 * and thus allow interrupts to occur again.
+	 * XXX might want to use MSR.PM to keep the counters frozen until
+	 * we get back out of this interrupt.
+	 */
+	mtspr(SPRN_MMCR0, cpuhw->mmcr[0]);
+
+	/*
+	 * If we need a wakeup, check whether interrupts were soft-enabled
+	 * when we took the interrupt.  If they were, we can wake stuff up
+	 * immediately; otherwise we'll have to set a flag and do the
+	 * wakeup when interrupts get soft-enabled.
+	 */
+	if (need_wakeup) {
+		if (regs->softe) {
+			irq_enter();
+			perf_counter_do_pending();
+			irq_exit();
+		} else {
+			set_perf_counter_pending(1);
+		}
+	}
+}
+
+extern struct power_pmu ppc970_pmu;
+extern struct power_pmu power6_pmu;
+
+static int init_perf_counters(void)
+{
+	unsigned long pvr;
+
+	if (reserve_pmc_hardware(perf_counter_interrupt)) {
+		printk(KERN_ERR "Couldn't init performance monitor subsystem\n");
+		return -EBUSY;
+	}
+
+	/* XXX should get this from cputable */
+	pvr = mfspr(SPRN_PVR);
+	switch (PVR_VER(pvr)) {
+	case PV_970:
+	case PV_970FX:
+	case PV_970MP:
+		ppmu = &ppc970_pmu;
+		break;
+	case 0x3e:
+		ppmu = &power6_pmu;
+		break;
+	}
+	return 0;
+}
+
+arch_initcall(init_perf_counters);
diff --git a/arch/powerpc/kernel/power6-pmu.c b/arch/powerpc/kernel/power6-pmu.c
new file mode 100644
index 000000000000..b1f61f3c97bb
--- /dev/null
+++ b/arch/powerpc/kernel/power6-pmu.c
@@ -0,0 +1,283 @@
+/*
+ * Performance counter support for POWER6 processors.
+ *
+ * Copyright 2008-2009 Paul Mackerras, IBM Corporation.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version
+ * 2 of the License, or (at your option) any later version.
+ */
+#include <linux/kernel.h>
+#include <linux/perf_counter.h>
+#include <asm/reg.h>
+
+/*
+ * Bits in event code for POWER6
+ */
+#define PM_PMC_SH	20	/* PMC number (1-based) for direct events */
+#define PM_PMC_MSK	0x7
+#define PM_PMC_MSKS	(PM_PMC_MSK << PM_PMC_SH)
+#define PM_UNIT_SH	16	/* Unit event comes (TTMxSEL encoding) */
+#define PM_UNIT_MSK	0xf
+#define PM_UNIT_MSKS	(PM_UNIT_MSK << PM_UNIT_SH)
+#define PM_LLAV		0x8000	/* Load lookahead match value */
+#define PM_LLA		0x4000	/* Load lookahead match enable */
+#define PM_BYTE_SH	12	/* Byte of event bus to use */
+#define PM_BYTE_MSK	3
+#define PM_SUBUNIT_SH	8	/* Subunit event comes from (NEST_SEL enc.) */
+#define PM_SUBUNIT_MSK	7
+#define PM_SUBUNIT_MSKS	(PM_SUBUNIT_MSK << PM_SUBUNIT_SH)
+#define PM_PMCSEL_MSK	0xff	/* PMCxSEL value */
+#define PM_BUSEVENT_MSK	0xf3700
+
+/*
+ * Bits in MMCR1 for POWER6
+ */
+#define MMCR1_TTM0SEL_SH	60
+#define MMCR1_TTMSEL_SH(n)	(MMCR1_TTM0SEL_SH - (n) * 4)
+#define MMCR1_TTMSEL_MSK	0xf
+#define MMCR1_TTMSEL(m, n)	(((m) >> MMCR1_TTMSEL_SH(n)) & MMCR1_TTMSEL_MSK)
+#define MMCR1_NESTSEL_SH	45
+#define MMCR1_NESTSEL_MSK	0x7
+#define MMCR1_NESTSEL(m)	(((m) >> MMCR1_NESTSEL_SH) & MMCR1_NESTSEL_MSK)
+#define MMCR1_PMC1_LLA		((u64)1 << 44)
+#define MMCR1_PMC1_LLA_VALUE	((u64)1 << 39)
+#define MMCR1_PMC1_ADDR_SEL	((u64)1 << 35)
+#define MMCR1_PMC1SEL_SH	24
+#define MMCR1_PMCSEL_SH(n)	(MMCR1_PMC1SEL_SH - (n) * 8)
+#define MMCR1_PMCSEL_MSK	0xff
+
+/*
+ * Assign PMC numbers and compute MMCR1 value for a set of events
+ */
+static int p6_compute_mmcr(unsigned int event[], int n_ev,
+			   unsigned int hwc[], u64 mmcr[])
+{
+	u64 mmcr1 = 0;
+	int i;
+	unsigned int pmc, ev, b, u, s, psel;
+	unsigned int ttmset = 0;
+	unsigned int pmc_inuse = 0;
+
+	if (n_ev > 4)
+		return -1;
+	for (i = 0; i < n_ev; ++i) {
+		pmc = (event[i] >> PM_PMC_SH) & PM_PMC_MSK;
+		if (pmc) {
+			if (pmc_inuse & (1 << (pmc - 1)))
+				return -1;	/* collision! */
+			pmc_inuse |= 1 << (pmc - 1);
+		}
+	}
+	for (i = 0; i < n_ev; ++i) {
+		ev = event[i];
+		pmc = (ev >> PM_PMC_SH) & PM_PMC_MSK;
+		if (pmc) {
+			--pmc;
+		} else {
+			/* can go on any PMC; find a free one */
+			for (pmc = 0; pmc < 4; ++pmc)
+				if (!(pmc_inuse & (1 << pmc)))
+					break;
+			pmc_inuse |= 1 << pmc;
+		}
+		hwc[i] = pmc;
+		psel = ev & PM_PMCSEL_MSK;
+		if (ev & PM_BUSEVENT_MSK) {
+			/* this event uses the event bus */
+			b = (ev >> PM_BYTE_SH) & PM_BYTE_MSK;
+			u = (ev >> PM_UNIT_SH) & PM_UNIT_MSK;
+			/* check for conflict on this byte of event bus */
+			if ((ttmset & (1 << b)) && MMCR1_TTMSEL(mmcr1, b) != u)
+				return -1;
+			mmcr1 |= (u64)u << MMCR1_TTMSEL_SH(b);
+			ttmset |= 1 << b;
+			if (u == 5) {
+				/* Nest events have a further mux */
+				s = (ev >> PM_SUBUNIT_SH) & PM_SUBUNIT_MSK;
+				if ((ttmset & 0x10) &&
+				    MMCR1_NESTSEL(mmcr1) != s)
+					return -1;
+				ttmset |= 0x10;
+				mmcr1 |= (u64)s << MMCR1_NESTSEL_SH;
+			}
+			if (0x30 <= psel && psel <= 0x3d) {
+				/* these need the PMCx_ADDR_SEL bits */
+				if (b >= 2)
+					mmcr1 |= MMCR1_PMC1_ADDR_SEL >> pmc;
+			}
+			/* bus select values are different for PMC3/4 */
+			if (pmc >= 2 && (psel & 0x90) == 0x80)
+				psel ^= 0x20;
+		}
+		if (ev & PM_LLA) {
+			mmcr1 |= MMCR1_PMC1_LLA >> pmc;
+			if (ev & PM_LLAV)
+				mmcr1 |= MMCR1_PMC1_LLA_VALUE >> pmc;
+		}
+		mmcr1 |= (u64)psel << MMCR1_PMCSEL_SH(pmc);
+	}
+	mmcr[0] = 0;
+	if (pmc_inuse & 1)
+		mmcr[0] = MMCR0_PMC1CE;
+	if (pmc_inuse & 0xe)
+		mmcr[0] |= MMCR0_PMCjCE;
+	mmcr[1] = mmcr1;
+	mmcr[2] = 0;
+	return 0;
+}
+
+/*
+ * Layout of constraint bits:
+ *
+ *	0-1	add field: number of uses of PMC1 (max 1)
+ *	2-3, 4-5, 6-7: ditto for PMC2, 3, 4
+ *	8-10	select field: nest (subunit) event selector
+ *	16-19	select field: unit on byte 0 of event bus
+ *	20-23, 24-27, 28-31 ditto for bytes 1, 2, 3
+ */
+static int p6_get_constraint(unsigned int event, u64 *maskp, u64 *valp)
+{
+	int pmc, byte, sh;
+	unsigned int mask = 0, value = 0;
+
+	pmc = (event >> PM_PMC_SH) & PM_PMC_MSK;
+	if (pmc) {
+		if (pmc > 4)
+			return -1;
+		sh = (pmc - 1) * 2;
+		mask |= 2 << sh;
+		value |= 1 << sh;
+	}
+	if (event & PM_BUSEVENT_MSK) {
+		byte = (event >> PM_BYTE_SH) & PM_BYTE_MSK;
+		sh = byte * 4;
+		mask |= PM_UNIT_MSKS << sh;
+		value |= (event & PM_UNIT_MSKS) << sh;
+		if ((event & PM_UNIT_MSKS) == (5 << PM_UNIT_SH)) {
+			mask |= PM_SUBUNIT_MSKS;
+			value |= event & PM_SUBUNIT_MSKS;
+		}
+	}
+	*maskp = mask;
+	*valp = value;
+	return 0;
+}
+
+#define MAX_ALT	4	/* at most 4 alternatives for any event */
+
+static const unsigned int event_alternatives[][MAX_ALT] = {
+	{ 0x0130e8, 0x2000f6, 0x3000fc },	/* PM_PTEG_RELOAD_VALID */
+	{ 0x080080, 0x10000d, 0x30000c, 0x4000f0 }, /* PM_LD_MISS_L1 */
+	{ 0x080088, 0x200054, 0x3000f0 },	/* PM_ST_MISS_L1 */
+	{ 0x10000a, 0x2000f4 },			/* PM_RUN_CYC */
+	{ 0x10000b, 0x2000f5 },			/* PM_RUN_COUNT */
+	{ 0x10000e, 0x400010 },			/* PM_PURR */
+	{ 0x100010, 0x4000f8 },			/* PM_FLUSH */
+	{ 0x10001a, 0x200010 },			/* PM_MRK_INST_DISP */
+	{ 0x100026, 0x3000f8 },			/* PM_TB_BIT_TRANS */
+	{ 0x100054, 0x2000f0 },			/* PM_ST_FIN */
+	{ 0x100056, 0x2000fc },			/* PM_L1_ICACHE_MISS */
+	{ 0x1000f0, 0x40000a },			/* PM_INST_IMC_MATCH_CMPL */
+	{ 0x1000f8, 0x200008 },			/* PM_GCT_EMPTY_CYC */
+	{ 0x1000fc, 0x400006 },			/* PM_LSU_DERAT_MISS_CYC */
+	{ 0x20000e, 0x400007 },			/* PM_LSU_DERAT_MISS */
+	{ 0x200012, 0x300012 },			/* PM_INST_DISP */
+	{ 0x2000f2, 0x3000f2 },			/* PM_INST_DISP */
+	{ 0x2000f8, 0x300010 },			/* PM_EXT_INT */
+	{ 0x2000fe, 0x300056 },			/* PM_DATA_FROM_L2MISS */
+	{ 0x2d0030, 0x30001a },			/* PM_MRK_FPU_FIN */
+	{ 0x30000a, 0x400018 },			/* PM_MRK_INST_FIN */
+	{ 0x3000f6, 0x40000e },			/* PM_L1_DCACHE_RELOAD_VALID */
+	{ 0x3000fe, 0x400056 },			/* PM_DATA_FROM_L3MISS */
+};
+
+/*
+ * This could be made more efficient with a binary search on
+ * a presorted list, if necessary
+ */
+static int find_alternatives_list(unsigned int event)
+{
+	int i, j;
+	unsigned int alt;
+
+	for (i = 0; i < ARRAY_SIZE(event_alternatives); ++i) {
+		if (event < event_alternatives[i][0])
+			return -1;
+		for (j = 0; j < MAX_ALT; ++j) {
+			alt = event_alternatives[i][j];
+			if (!alt || event < alt)
+				break;
+			if (event == alt)
+				return i;
+		}
+	}
+	return -1;
+}
+
+static int p6_get_alternatives(unsigned int event, unsigned int alt[])
+{
+	int i, j;
+	unsigned int aevent, psel, pmc;
+	unsigned int nalt = 1;
+
+	alt[0] = event;
+
+	/* check the alternatives table */
+	i = find_alternatives_list(event);
+	if (i >= 0) {
+		/* copy out alternatives from list */
+		for (j = 0; j < MAX_ALT; ++j) {
+			aevent = event_alternatives[i][j];
+			if (!aevent)
+				break;
+			if (aevent != event)
+				alt[nalt++] = aevent;
+		}
+
+	} else {
+		/* Check for alternative ways of computing sum events */
+		/* PMCSEL 0x32 counter N == PMCSEL 0x34 counter 5-N */
+		psel = event & (PM_PMCSEL_MSK & ~1);	/* ignore edge bit */
+		pmc = (event >> PM_PMC_SH) & PM_PMC_MSK;
+		if (pmc && (psel == 0x32 || psel == 0x34))
+			alt[nalt++] = ((event ^ 0x6) & ~PM_PMC_MSKS) |
+				((5 - pmc) << PM_PMC_SH);
+
+		/* PMCSEL 0x38 counter N == PMCSEL 0x3a counter N+/-2 */
+		if (pmc && (psel == 0x38 || psel == 0x3a))
+			alt[nalt++] = ((event ^ 0x2) & ~PM_PMC_MSKS) |
+				((pmc > 2? pmc - 2: pmc + 2) << PM_PMC_SH);
+	}
+
+	return nalt;
+}
+
+static void p6_disable_pmc(unsigned int pmc, u64 mmcr[])
+{
+	/* Set PMCxSEL to 0 to disable PMCx */
+	mmcr[1] &= ~(0xffUL << MMCR1_PMCSEL_SH(pmc));
+}
+
+static int power6_generic_events[] = {
+	[PERF_COUNT_CPU_CYCLES] = 0x1e,
+	[PERF_COUNT_INSTRUCTIONS] = 2,
+	[PERF_COUNT_CACHE_REFERENCES] = 0x280030,	/* LD_REF_L1 */
+	[PERF_COUNT_CACHE_MISSES] = 0x30000c,		/* LD_MISS_L1 */
+	[PERF_COUNT_BRANCH_INSTRUCTIONS] = 0x410a0,	/* BR_PRED */ 
+	[PERF_COUNT_BRANCH_MISSES] = 0x400052,		/* BR_MPRED */
+};
+
+struct power_pmu power6_pmu = {
+	.n_counter = 4,
+	.max_alternatives = MAX_ALT,
+	.add_fields = 0x55,
+	.test_adder = 0,
+	.compute_mmcr = p6_compute_mmcr,
+	.get_constraint = p6_get_constraint,
+	.get_alternatives = p6_get_alternatives,
+	.disable_pmc = p6_disable_pmc,
+	.n_generic = ARRAY_SIZE(power6_generic_events),
+	.generic_events = power6_generic_events,
+};
diff --git a/arch/powerpc/kernel/ppc970-pmu.c b/arch/powerpc/kernel/ppc970-pmu.c
new file mode 100644
index 000000000000..c3256580be1a
--- /dev/null
+++ b/arch/powerpc/kernel/ppc970-pmu.c
@@ -0,0 +1,375 @@
+/*
+ * Performance counter support for PPC970-family processors.
+ *
+ * Copyright 2008-2009 Paul Mackerras, IBM Corporation.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version
+ * 2 of the License, or (at your option) any later version.
+ */
+#include <linux/string.h>
+#include <linux/perf_counter.h>
+#include <asm/reg.h>
+
+/*
+ * Bits in event code for PPC970
+ */
+#define PM_PMC_SH	12	/* PMC number (1-based) for direct events */
+#define PM_PMC_MSK	0xf
+#define PM_UNIT_SH	8	/* TTMMUX number and setting - unit select */
+#define PM_UNIT_MSK	0xf
+#define PM_BYTE_SH	4	/* Byte number of event bus to use */
+#define PM_BYTE_MSK	3
+#define PM_PMCSEL_MSK	0xf
+
+/* Values in PM_UNIT field */
+#define PM_NONE		0
+#define PM_FPU		1
+#define PM_VPU		2
+#define PM_ISU		3
+#define PM_IFU		4
+#define PM_IDU		5
+#define PM_STS		6
+#define PM_LSU0		7
+#define PM_LSU1U	8
+#define PM_LSU1L	9
+#define PM_LASTUNIT	9
+
+/*
+ * Bits in MMCR0 for PPC970
+ */
+#define MMCR0_PMC1SEL_SH	8
+#define MMCR0_PMC2SEL_SH	1
+#define MMCR_PMCSEL_MSK		0x1f
+
+/*
+ * Bits in MMCR1 for PPC970
+ */
+#define MMCR1_TTM0SEL_SH	62
+#define MMCR1_TTM1SEL_SH	59
+#define MMCR1_TTM3SEL_SH	53
+#define MMCR1_TTMSEL_MSK	3
+#define MMCR1_TD_CP_DBG0SEL_SH	50
+#define MMCR1_TD_CP_DBG1SEL_SH	48
+#define MMCR1_TD_CP_DBG2SEL_SH	46
+#define MMCR1_TD_CP_DBG3SEL_SH	44
+#define MMCR1_PMC1_ADDER_SEL_SH	39
+#define MMCR1_PMC2_ADDER_SEL_SH	38
+#define MMCR1_PMC6_ADDER_SEL_SH	37
+#define MMCR1_PMC5_ADDER_SEL_SH	36
+#define MMCR1_PMC8_ADDER_SEL_SH	35
+#define MMCR1_PMC7_ADDER_SEL_SH	34
+#define MMCR1_PMC3_ADDER_SEL_SH	33
+#define MMCR1_PMC4_ADDER_SEL_SH	32
+#define MMCR1_PMC3SEL_SH	27
+#define MMCR1_PMC4SEL_SH	22
+#define MMCR1_PMC5SEL_SH	17
+#define MMCR1_PMC6SEL_SH	12
+#define MMCR1_PMC7SEL_SH	7
+#define MMCR1_PMC8SEL_SH	2
+
+static short mmcr1_adder_bits[8] = {
+	MMCR1_PMC1_ADDER_SEL_SH,
+	MMCR1_PMC2_ADDER_SEL_SH,
+	MMCR1_PMC3_ADDER_SEL_SH,
+	MMCR1_PMC4_ADDER_SEL_SH,
+	MMCR1_PMC5_ADDER_SEL_SH,
+	MMCR1_PMC6_ADDER_SEL_SH,
+	MMCR1_PMC7_ADDER_SEL_SH,
+	MMCR1_PMC8_ADDER_SEL_SH
+};
+
+/*
+ * Bits in MMCRA
+ */
+
+/*
+ * Layout of constraint bits:
+ * 6666555555555544444444443333333333222222222211111111110000000000
+ * 3210987654321098765432109876543210987654321098765432109876543210
+ *                 <><>[  >[  >[  ><  ><  ><  ><  ><><><><><><><><>
+ *                 T0T1 UC  PS1 PS2 B0  B1  B2  B3 P1P2P3P4P5P6P7P8
+ *
+ * T0 - TTM0 constraint
+ *     46-47: TTM0SEL value (0=FPU, 2=IFU, 3=VPU) 0xC000_0000_0000
+ *
+ * T1 - TTM1 constraint
+ *     44-45: TTM1SEL value (0=IDU, 3=STS) 0x3000_0000_0000
+ *
+ * UC - unit constraint: can't have all three of FPU|IFU|VPU, ISU, IDU|STS
+ *     43: UC3 error 0x0800_0000_0000
+ *     42: FPU|IFU|VPU events needed 0x0400_0000_0000
+ *     41: ISU events needed 0x0200_0000_0000
+ *     40: IDU|STS events needed 0x0100_0000_0000
+ *
+ * PS1
+ *     39: PS1 error 0x0080_0000_0000
+ *     36-38: count of events needing PMC1/2/5/6 0x0070_0000_0000
+ *
+ * PS2
+ *     35: PS2 error 0x0008_0000_0000
+ *     32-34: count of events needing PMC3/4/7/8 0x0007_0000_0000
+ *
+ * B0
+ *     28-31: Byte 0 event source 0xf000_0000
+ *	      Encoding as for the event code
+ *
+ * B1, B2, B3
+ *     24-27, 20-23, 16-19: Byte 1, 2, 3 event sources
+ *
+ * P1
+ *     15: P1 error 0x8000
+ *     14-15: Count of events needing PMC1
+ *
+ * P2..P8
+ *     0-13: Count of events needing PMC2..PMC8
+ */
+
+/* Masks and values for using events from the various units */
+static u64 unit_cons[PM_LASTUNIT+1][2] = {
+	[PM_FPU] =   { 0xc80000000000ull, 0x040000000000ull },
+	[PM_VPU] =   { 0xc80000000000ull, 0xc40000000000ull },
+	[PM_ISU] =   { 0x080000000000ull, 0x020000000000ull },
+	[PM_IFU] =   { 0xc80000000000ull, 0x840000000000ull },
+	[PM_IDU] =   { 0x380000000000ull, 0x010000000000ull },
+	[PM_STS] =   { 0x380000000000ull, 0x310000000000ull },
+};
+
+static int p970_get_constraint(unsigned int event, u64 *maskp, u64 *valp)
+{
+	int pmc, byte, unit, sh;
+	u64 mask = 0, value = 0;
+	int grp = -1;
+
+	pmc = (event >> PM_PMC_SH) & PM_PMC_MSK;
+	if (pmc) {
+		if (pmc > 8)
+			return -1;
+		sh = (pmc - 1) * 2;
+		mask |= 2 << sh;
+		value |= 1 << sh;
+		grp = ((pmc - 1) >> 1) & 1;
+	}
+	unit = (event >> PM_UNIT_SH) & PM_UNIT_MSK;
+	if (unit) {
+		if (unit > PM_LASTUNIT)
+			return -1;
+		mask |= unit_cons[unit][0];
+		value |= unit_cons[unit][1];
+		byte = (event >> PM_BYTE_SH) & PM_BYTE_MSK;
+		/*
+		 * Bus events on bytes 0 and 2 can be counted
+		 * on PMC1/2/5/6; bytes 1 and 3 on PMC3/4/7/8.
+		 */
+		if (!pmc)
+			grp = byte & 1;
+		/* Set byte lane select field */
+		mask  |= 0xfULL << (28 - 4 * byte);
+		value |= (u64)unit << (28 - 4 * byte);
+	}
+	if (grp == 0) {
+		/* increment PMC1/2/5/6 field */
+		mask  |= 0x8000000000ull;
+		value |= 0x1000000000ull;
+	} else if (grp == 1) {
+		/* increment PMC3/4/7/8 field */
+		mask  |= 0x800000000ull;
+		value |= 0x100000000ull;
+	}
+	*maskp = mask;
+	*valp = value;
+	return 0;
+}
+
+static int p970_get_alternatives(unsigned int event, unsigned int alt[])
+{
+	alt[0] = event;
+
+	/* 2 alternatives for LSU empty */
+	if (event == 0x2002 || event == 0x3002) {
+		alt[1] = event ^ 0x1000;
+		return 2;
+	}
+		
+	return 1;
+}
+
+static int p970_compute_mmcr(unsigned int event[], int n_ev,
+			     unsigned int hwc[], u64 mmcr[])
+{
+	u64 mmcr0 = 0, mmcr1 = 0, mmcra = 0;
+	unsigned int pmc, unit, byte, psel;
+	unsigned int ttm, grp;
+	unsigned int pmc_inuse = 0;
+	unsigned int pmc_grp_use[2];
+	unsigned char busbyte[4];
+	unsigned char unituse[16];
+	unsigned char unitmap[] = { 0, 0<<3, 3<<3, 1<<3, 2<<3, 0|4, 3|4 };
+	unsigned char ttmuse[2];
+	unsigned char pmcsel[8];
+	int i;
+
+	if (n_ev > 8)
+		return -1;
+
+	/* First pass to count resource use */
+	pmc_grp_use[0] = pmc_grp_use[1] = 0;
+	memset(busbyte, 0, sizeof(busbyte));
+	memset(unituse, 0, sizeof(unituse));
+	for (i = 0; i < n_ev; ++i) {
+		pmc = (event[i] >> PM_PMC_SH) & PM_PMC_MSK;
+		if (pmc) {
+			if (pmc_inuse & (1 << (pmc - 1)))
+				return -1;
+			pmc_inuse |= 1 << (pmc - 1);
+			/* count 1/2/5/6 vs 3/4/7/8 use */
+			++pmc_grp_use[((pmc - 1) >> 1) & 1];
+		}
+		unit = (event[i] >> PM_UNIT_SH) & PM_UNIT_MSK;
+		byte = (event[i] >> PM_BYTE_SH) & PM_BYTE_MSK;
+		if (unit) {
+			if (unit > PM_LASTUNIT)
+				return -1;
+			if (!pmc)
+				++pmc_grp_use[byte & 1];
+			if (busbyte[byte] && busbyte[byte] != unit)
+				return -1;
+			busbyte[byte] = unit;
+			unituse[unit] = 1;
+		}
+	}
+	if (pmc_grp_use[0] > 4 || pmc_grp_use[1] > 4)
+		return -1;
+
+	/*
+	 * Assign resources and set multiplexer selects.
+	 *
+	 * PM_ISU can go either on TTM0 or TTM1, but that's the only
+	 * choice we have to deal with.
+	 */
+	if (unituse[PM_ISU] &
+	    (unituse[PM_FPU] | unituse[PM_IFU] | unituse[PM_VPU]))
+		unitmap[PM_ISU] = 2 | 4;	/* move ISU to TTM1 */
+	/* Set TTM[01]SEL fields. */
+	ttmuse[0] = ttmuse[1] = 0;
+	for (i = PM_FPU; i <= PM_STS; ++i) {
+		if (!unituse[i])
+			continue;
+		ttm = unitmap[i];
+		++ttmuse[(ttm >> 2) & 1];
+		mmcr1 |= (u64)(ttm & ~4) << MMCR1_TTM1SEL_SH;
+	}
+	/* Check only one unit per TTMx */
+	if (ttmuse[0] > 1 || ttmuse[1] > 1)
+		return -1;
+
+	/* Set byte lane select fields and TTM3SEL. */
+	for (byte = 0; byte < 4; ++byte) {
+		unit = busbyte[byte];
+		if (!unit)
+			continue;
+		if (unit <= PM_STS)
+			ttm = (unitmap[unit] >> 2) & 1;
+		else if (unit == PM_LSU0)
+			ttm = 2;
+		else {
+			ttm = 3;
+			if (unit == PM_LSU1L && byte >= 2)
+				mmcr1 |= 1ull << (MMCR1_TTM3SEL_SH + 3 - byte);
+		}
+		mmcr1 |= (u64)ttm << (MMCR1_TD_CP_DBG0SEL_SH - 2 * byte);
+	}
+
+	/* Second pass: assign PMCs, set PMCxSEL and PMCx_ADDER_SEL fields */
+	memset(pmcsel, 0x8, sizeof(pmcsel));	/* 8 means don't count */
+	for (i = 0; i < n_ev; ++i) {
+		pmc = (event[i] >> PM_PMC_SH) & PM_PMC_MSK;
+		unit = (event[i] >> PM_UNIT_SH) & PM_UNIT_MSK;
+		byte = (event[i] >> PM_BYTE_SH) & PM_BYTE_MSK;
+		psel = event[i] & PM_PMCSEL_MSK;
+		if (!pmc) {
+			/* Bus event or any-PMC direct event */
+			if (unit)
+				psel |= 0x10 | ((byte & 2) << 2);
+			else
+				psel |= 8;
+			for (pmc = 0; pmc < 8; ++pmc) {
+				if (pmc_inuse & (1 << pmc))
+					continue;
+				grp = (pmc >> 1) & 1;
+				if (unit) {
+					if (grp == (byte & 1))
+						break;
+				} else if (pmc_grp_use[grp] < 4) {
+					++pmc_grp_use[grp];
+					break;
+				}
+			}
+			pmc_inuse |= 1 << pmc;
+		} else {
+			/* Direct event */
+			--pmc;
+			if (psel == 0 && (byte & 2))
+				/* add events on higher-numbered bus */
+				mmcr1 |= 1ull << mmcr1_adder_bits[pmc];
+		}
+		pmcsel[pmc] = psel;
+		hwc[i] = pmc;
+	}
+	for (pmc = 0; pmc < 2; ++pmc)
+		mmcr0 |= pmcsel[pmc] << (MMCR0_PMC1SEL_SH - 7 * pmc);
+	for (; pmc < 8; ++pmc)
+		mmcr1 |= (u64)pmcsel[pmc] << (MMCR1_PMC3SEL_SH - 5 * (pmc - 2));
+	if (pmc_inuse & 1)
+		mmcr0 |= MMCR0_PMC1CE;
+	if (pmc_inuse & 0xfe)
+		mmcr0 |= MMCR0_PMCjCE;
+
+	mmcra |= 0x2000;	/* mark only one IOP per PPC instruction */
+
+	/* Return MMCRx values */
+	mmcr[0] = mmcr0;
+	mmcr[1] = mmcr1;
+	mmcr[2] = mmcra;
+	return 0;
+}
+
+static void p970_disable_pmc(unsigned int pmc, u64 mmcr[])
+{
+	int shift, i;
+
+	if (pmc <= 1) {
+		shift = MMCR0_PMC1SEL_SH - 7 * pmc;
+		i = 0;
+	} else {
+		shift = MMCR1_PMC3SEL_SH - 5 * (pmc - 2);
+		i = 1;
+	}
+	/*
+	 * Setting the PMCxSEL field to 0x08 disables PMC x.
+	 */
+	mmcr[i] = (mmcr[i] & ~(0x1fUL << shift)) | (0x08UL << shift);
+}
+
+static int ppc970_generic_events[] = {
+	[PERF_COUNT_CPU_CYCLES] = 7,
+	[PERF_COUNT_INSTRUCTIONS] = 1,
+	[PERF_COUNT_CACHE_REFERENCES] = 0x8810,		/* PM_LD_REF_L1 */
+	[PERF_COUNT_CACHE_MISSES] = 0x3810,		/* PM_LD_MISS_L1 */
+	[PERF_COUNT_BRANCH_INSTRUCTIONS] = 0x431,	/* PM_BR_ISSUED */
+	[PERF_COUNT_BRANCH_MISSES] = 0x327,		/* PM_GRP_BR_MPRED */
+};
+
+struct power_pmu ppc970_pmu = {
+	.n_counter = 8,
+	.max_alternatives = 2,
+	.add_fields = 0x001100005555ull,
+	.test_adder = 0x013300000000ull,
+	.compute_mmcr = p970_compute_mmcr,
+	.get_constraint = p970_get_constraint,
+	.get_alternatives = p970_get_alternatives,
+	.disable_pmc = p970_disable_pmc,
+	.n_generic = ARRAY_SIZE(ppc970_generic_events),
+	.generic_events = ppc970_generic_events,
+};
diff --git a/arch/powerpc/platforms/Kconfig.cputype b/arch/powerpc/platforms/Kconfig.cputype
index e868b5c50723..dc0f3c933518 100644
--- a/arch/powerpc/platforms/Kconfig.cputype
+++ b/arch/powerpc/platforms/Kconfig.cputype
@@ -1,6 +1,7 @@
 config PPC64
 	bool "64-bit kernel"
 	default n
+	select HAVE_PERF_COUNTERS
 	help
 	  This option selects whether a 32-bit or a 64-bit kernel
 	  will be built.
diff --git a/include/linux/perf_counter.h b/include/linux/perf_counter.h
index cc3a75a239a9..b21d1ea4c054 100644
--- a/include/linux/perf_counter.h
+++ b/include/linux/perf_counter.h
@@ -236,6 +236,9 @@ extern u64 hw_perf_save_disable(void);
 extern void hw_perf_restore(u64 ctrl);
 extern int perf_counter_task_disable(void);
 extern int perf_counter_task_enable(void);
+extern int hw_perf_group_sched_in(struct perf_counter *group_leader,
+	       struct perf_cpu_context *cpuctx,
+	       struct perf_counter_context *ctx, int cpu);
 
 #else
 static inline void
diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c
index 37f771691f93..4c0dccb756ad 100644
--- a/kernel/perf_counter.c
+++ b/kernel/perf_counter.c
@@ -41,12 +41,20 @@ static DEFINE_MUTEX(perf_resource_mutex);
 extern __weak const struct hw_perf_counter_ops *
 hw_perf_counter_init(struct perf_counter *counter)
 {
-	return ERR_PTR(-EINVAL);
+	return NULL;
 }
 
 u64 __weak hw_perf_save_disable(void)		{ return 0; }
 void __weak hw_perf_restore(u64 ctrl)		{ barrier(); }
 void __weak hw_perf_counter_setup(void)		{ barrier(); }
+int __weak hw_perf_group_sched_in(struct perf_counter *group_leader,
+	       struct perf_cpu_context *cpuctx,
+	       struct perf_counter_context *ctx, int cpu)
+{
+	return 0;
+}
+
+void __weak perf_counter_print_debug(void)	{ }
 
 static void
 list_add_counter(struct perf_counter *counter, struct perf_counter_context *ctx)
@@ -341,6 +349,9 @@ group_sched_out(struct perf_counter *group_counter,
 {
 	struct perf_counter *counter;
 
+	if (group_counter->state != PERF_COUNTER_STATE_ACTIVE)
+		return;
+
 	counter_sched_out(group_counter, cpuctx, ctx);
 
 	/*
@@ -354,15 +365,18 @@ void __perf_counter_sched_out(struct perf_counter_context *ctx,
 			      struct perf_cpu_context *cpuctx)
 {
 	struct perf_counter *counter;
+	u64 flags;
 
 	if (likely(!ctx->nr_counters))
 		return;
 
 	spin_lock(&ctx->lock);
+	flags = hw_perf_save_disable();
 	if (ctx->nr_active) {
 		list_for_each_entry(counter, &ctx->counter_list, list_entry)
 			group_sched_out(counter, cpuctx, ctx);
 	}
+	hw_perf_restore(flags);
 	spin_unlock(&ctx->lock);
 }
 
@@ -402,7 +416,14 @@ group_sched_in(struct perf_counter *group_counter,
 	       int cpu)
 {
 	struct perf_counter *counter, *partial_group;
-	int ret = 0;
+	int ret;
+
+	if (group_counter->state == PERF_COUNTER_STATE_OFF)
+		return 0;
+
+	ret = hw_perf_group_sched_in(group_counter, cpuctx, ctx, cpu);
+	if (ret)
+		return ret < 0 ? ret : 0;
 
 	if (counter_sched_in(group_counter, cpuctx, ctx, cpu))
 		return -EAGAIN;
@@ -415,10 +436,9 @@ group_sched_in(struct perf_counter *group_counter,
 			partial_group = counter;
 			goto group_error;
 		}
-		ret = -EAGAIN;
 	}
 
-	return ret;
+	return 0;
 
 group_error:
 	/*
@@ -440,11 +460,13 @@ __perf_counter_sched_in(struct perf_counter_context *ctx,
 			struct perf_cpu_context *cpuctx, int cpu)
 {
 	struct perf_counter *counter;
+	u64 flags;
 
 	if (likely(!ctx->nr_counters))
 		return;
 
 	spin_lock(&ctx->lock);
+	flags = hw_perf_save_disable();
 	list_for_each_entry(counter, &ctx->counter_list, list_entry) {
 		/*
 		 * Listen to the 'cpu' scheduling filter constraint
@@ -454,12 +476,13 @@ __perf_counter_sched_in(struct perf_counter_context *ctx,
 			continue;
 
 		/*
-		 * If we scheduled in a group atomically and
-		 * exclusively, break out:
+		 * If we scheduled in a group atomically and exclusively,
+		 * or if this group can't go on, break out:
 		 */
 		if (group_sched_in(counter, cpuctx, ctx, cpu))
 			break;
 	}
+	hw_perf_restore(flags);
 	spin_unlock(&ctx->lock);
 }
 
@@ -928,18 +951,32 @@ static const struct file_operations perf_fops = {
 
 static int cpu_clock_perf_counter_enable(struct perf_counter *counter)
 {
+	int cpu = raw_smp_processor_id();
+
+	atomic64_set(&counter->hw.prev_count, cpu_clock(cpu));
 	return 0;
 }
 
+static void cpu_clock_perf_counter_update(struct perf_counter *counter)
+{
+	int cpu = raw_smp_processor_id();
+	s64 prev;
+	u64 now;
+
+	now = cpu_clock(cpu);
+	prev = atomic64_read(&counter->hw.prev_count);
+	atomic64_set(&counter->hw.prev_count, now);
+	atomic64_add(now - prev, &counter->count);
+}
+
 static void cpu_clock_perf_counter_disable(struct perf_counter *counter)
 {
+	cpu_clock_perf_counter_update(counter);
 }
 
 static void cpu_clock_perf_counter_read(struct perf_counter *counter)
 {
-	int cpu = raw_smp_processor_id();
-
-	atomic64_set(&counter->count, cpu_clock(cpu));
+	cpu_clock_perf_counter_update(counter);
 }
 
 static const struct hw_perf_counter_ops perf_ops_cpu_clock = {