6 files changed, 939 insertions, 0 deletions

diff --git a/arch/powerpc/oprofile/Kconfig b/arch/powerpc/oprofile/Kconfig
new file mode 100644
index 000000000000..19d37730b664
--- /dev/null
+++ b/arch/powerpc/oprofile/Kconfig
@@ -0,0 +1,23 @@
+
+menu "Profiling support"
+	depends on EXPERIMENTAL
+
+config PROFILING
+	bool "Profiling support (EXPERIMENTAL)"
+	help
+	  Say Y here to enable the extended profiling support mechanisms used
+	  by profilers such as OProfile.
+
+
+config OPROFILE
+	tristate "OProfile system profiling (EXPERIMENTAL)"
+	depends on PROFILING
+	help
+	  OProfile is a profiling system capable of profiling the
+	  whole system, include the kernel, kernel modules, libraries,
+	  and applications.
+
+	  If unsure, say N.
+
+endmenu
+
diff --git a/arch/powerpc/oprofile/Makefile b/arch/powerpc/oprofile/Makefile
new file mode 100644
index 000000000000..0782d0cca89c
--- /dev/null
+++ b/arch/powerpc/oprofile/Makefile
@@ -0,0 +1,11 @@
+obj-$(CONFIG_OPROFILE) += oprofile.o
+
+DRIVER_OBJS := $(addprefix ../../../drivers/oprofile/, \
+		oprof.o cpu_buffer.o buffer_sync.o \
+		event_buffer.o oprofile_files.o \
+		oprofilefs.o oprofile_stats.o \
+		timer_int.o )
+
+oprofile-y := $(DRIVER_OBJS) common.o
+oprofile-$(CONFIG_PPC64) += op_model_rs64.o op_model_power4.o
+oprofile-$(CONFIG_FSL_BOOKE) += op_model_fsl_booke.o
diff --git a/arch/powerpc/oprofile/common.c b/arch/powerpc/oprofile/common.c
new file mode 100644
index 000000000000..af2c05d20ba5
--- /dev/null
+++ b/arch/powerpc/oprofile/common.c
@@ -0,0 +1,195 @@
+/*
+ * PPC 64 oprofile support:
+ * Copyright (C) 2004 Anton Blanchard <anton@au.ibm.com>, IBM
+ * PPC 32 oprofile support: (based on PPC 64 support)
+ * Copyright (C) Freescale Semiconductor, Inc 2004
+ *	Author: Andy Fleming
+ *
+ * Based on alpha version.
+ *
+ * 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/oprofile.h>
+#ifndef __powerpc64__
+#include <linux/slab.h>
+#endif /* ! __powerpc64__ */
+#include <linux/init.h>
+#include <linux/smp.h>
+#include <linux/errno.h>
+#include <asm/ptrace.h>
+#include <asm/system.h>
+#include <asm/pmc.h>
+#include <asm/cputable.h>
+#include <asm/oprofile_impl.h>
+
+static struct op_powerpc_model *model;
+
+static struct op_counter_config ctr[OP_MAX_COUNTER];
+static struct op_system_config sys;
+
+#ifndef __powerpc64__
+static char *cpu_type;
+#endif /* ! __powerpc64__ */
+
+static void op_handle_interrupt(struct pt_regs *regs)
+{
+	model->handle_interrupt(regs, ctr);
+}
+
+static int op_powerpc_setup(void)
+{
+	int err;
+
+	/* Grab the hardware */
+	err = reserve_pmc_hardware(op_handle_interrupt);
+	if (err)
+		return err;
+
+	/* Pre-compute the values to stuff in the hardware registers.  */
+	model->reg_setup(ctr, &sys, model->num_counters);
+
+	/* Configure the registers on all cpus.  */
+#ifdef __powerpc64__
+	on_each_cpu(model->cpu_setup, NULL, 0, 1);
+#else /* __powerpc64__ */
+#if 0
+	/* FIXME: Make multi-cpu work */
+	on_each_cpu(model->reg_setup, NULL, 0, 1);
+#endif
+#endif /* __powerpc64__ */
+
+	return 0;
+}
+
+static void op_powerpc_shutdown(void)
+{
+	release_pmc_hardware();
+}
+
+static void op_powerpc_cpu_start(void *dummy)
+{
+	model->start(ctr);
+}
+
+static int op_powerpc_start(void)
+{
+	on_each_cpu(op_powerpc_cpu_start, NULL, 0, 1);
+	return 0;
+}
+
+static inline void op_powerpc_cpu_stop(void *dummy)
+{
+	model->stop();
+}
+
+static void op_powerpc_stop(void)
+{
+	on_each_cpu(op_powerpc_cpu_stop, NULL, 0, 1);
+}
+
+static int op_powerpc_create_files(struct super_block *sb, struct dentry *root)
+{
+	int i;
+
+#ifdef __powerpc64__
+	/*
+	 * There is one mmcr0, mmcr1 and mmcra for setting the events for
+	 * all of the counters.
+	 */
+	oprofilefs_create_ulong(sb, root, "mmcr0", &sys.mmcr0);
+	oprofilefs_create_ulong(sb, root, "mmcr1", &sys.mmcr1);
+	oprofilefs_create_ulong(sb, root, "mmcra", &sys.mmcra);
+#endif /* __powerpc64__ */
+
+	for (i = 0; i < model->num_counters; ++i) {
+		struct dentry *dir;
+		char buf[3];
+
+		snprintf(buf, sizeof buf, "%d", i);
+		dir = oprofilefs_mkdir(sb, root, buf);
+
+		oprofilefs_create_ulong(sb, dir, "enabled", &ctr[i].enabled);
+		oprofilefs_create_ulong(sb, dir, "event", &ctr[i].event);
+		oprofilefs_create_ulong(sb, dir, "count", &ctr[i].count);
+#ifdef __powerpc64__
+		/*
+		 * We dont support per counter user/kernel selection, but
+		 * we leave the entries because userspace expects them
+		 */
+#endif /* __powerpc64__ */
+		oprofilefs_create_ulong(sb, dir, "kernel", &ctr[i].kernel);
+		oprofilefs_create_ulong(sb, dir, "user", &ctr[i].user);
+
+#ifndef __powerpc64__
+		/* FIXME: Not sure if this is used */
+#endif /* ! __powerpc64__ */
+		oprofilefs_create_ulong(sb, dir, "unit_mask", &ctr[i].unit_mask);
+	}
+
+	oprofilefs_create_ulong(sb, root, "enable_kernel", &sys.enable_kernel);
+	oprofilefs_create_ulong(sb, root, "enable_user", &sys.enable_user);
+#ifdef __powerpc64__
+	oprofilefs_create_ulong(sb, root, "backtrace_spinlocks",
+				&sys.backtrace_spinlocks);
+#endif /* __powerpc64__ */
+
+	/* Default to tracing both kernel and user */
+	sys.enable_kernel = 1;
+	sys.enable_user = 1;
+#ifdef __powerpc64__
+	/* Turn on backtracing through spinlocks by default */
+	sys.backtrace_spinlocks = 1;
+#endif /* __powerpc64__ */
+
+	return 0;
+}
+
+int __init oprofile_arch_init(struct oprofile_operations *ops)
+{
+#ifndef __powerpc64__
+#ifdef CONFIG_FSL_BOOKE
+	model = &op_model_fsl_booke;
+#else
+	return -ENODEV;
+#endif
+
+	cpu_type = kmalloc(32, GFP_KERNEL);
+	if (NULL == cpu_type)
+		return -ENOMEM;
+
+	sprintf(cpu_type, "ppc/%s", cur_cpu_spec->cpu_name);
+
+	model->num_counters = cur_cpu_spec->num_pmcs;
+
+	ops->cpu_type = cpu_type;
+#else /* __powerpc64__ */
+	if (!cur_cpu_spec->oprofile_model || !cur_cpu_spec->oprofile_cpu_type)
+		return -ENODEV;
+	model = cur_cpu_spec->oprofile_model;
+	model->num_counters = cur_cpu_spec->num_pmcs;
+
+	ops->cpu_type = cur_cpu_spec->oprofile_cpu_type;
+#endif /* __powerpc64__ */
+	ops->create_files = op_powerpc_create_files;
+	ops->setup = op_powerpc_setup;
+	ops->shutdown = op_powerpc_shutdown;
+	ops->start = op_powerpc_start;
+	ops->stop = op_powerpc_stop;
+
+	printk(KERN_INFO "oprofile: using %s performance monitoring.\n",
+	       ops->cpu_type);
+
+	return 0;
+}
+
+void oprofile_arch_exit(void)
+{
+#ifndef __powerpc64__
+	kfree(cpu_type);
+	cpu_type = NULL;
+#endif /* ! __powerpc64__ */
+}
diff --git a/arch/powerpc/oprofile/op_model_fsl_booke.c b/arch/powerpc/oprofile/op_model_fsl_booke.c
new file mode 100644
index 000000000000..86124a94c9af
--- /dev/null
+++ b/arch/powerpc/oprofile/op_model_fsl_booke.c
@@ -0,0 +1,183 @@
+/*
+ * oprofile/op_model_e500.c
+ *
+ * Freescale Book-E oprofile support, based on ppc64 oprofile support
+ * Copyright (C) 2004 Anton Blanchard <anton@au.ibm.com>, IBM
+ *
+ * Copyright (c) 2004 Freescale Semiconductor, Inc
+ *
+ * Author: Andy Fleming
+ * Maintainer: Kumar Gala <Kumar.Gala@freescale.com>
+ *
+ * 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/oprofile.h>
+#include <linux/init.h>
+#include <linux/smp.h>
+#include <asm/ptrace.h>
+#include <asm/system.h>
+#include <asm/processor.h>
+#include <asm/cputable.h>
+#include <asm/reg_booke.h>
+#include <asm/page.h>
+#include <asm/pmc.h>
+#include <asm/oprofile_impl.h>
+
+static unsigned long reset_value[OP_MAX_COUNTER];
+
+static int num_counters;
+static int oprofile_running;
+
+static inline unsigned int ctr_read(unsigned int i)
+{
+	switch(i) {
+		case 0:
+			return mfpmr(PMRN_PMC0);
+		case 1:
+			return mfpmr(PMRN_PMC1);
+		case 2:
+			return mfpmr(PMRN_PMC2);
+		case 3:
+			return mfpmr(PMRN_PMC3);
+		default:
+			return 0;
+	}
+}
+
+static inline void ctr_write(unsigned int i, unsigned int val)
+{
+	switch(i) {
+		case 0:
+			mtpmr(PMRN_PMC0, val);
+			break;
+		case 1:
+			mtpmr(PMRN_PMC1, val);
+			break;
+		case 2:
+			mtpmr(PMRN_PMC2, val);
+			break;
+		case 3:
+			mtpmr(PMRN_PMC3, val);
+			break;
+		default:
+			break;
+	}
+}
+
+
+static void fsl_booke_reg_setup(struct op_counter_config *ctr,
+			     struct op_system_config *sys,
+			     int num_ctrs)
+{
+	int i;
+
+	num_counters = num_ctrs;
+
+	/* freeze all counters */
+	pmc_stop_ctrs();
+
+	/* Our counters count up, and "count" refers to
+	 * how much before the next interrupt, and we interrupt
+	 * on overflow.  So we calculate the starting value
+	 * which will give us "count" until overflow.
+	 * Then we set the events on the enabled counters */
+	for (i = 0; i < num_counters; ++i) {
+		reset_value[i] = 0x80000000UL - ctr[i].count;
+
+		init_pmc_stop(i);
+
+		set_pmc_event(i, ctr[i].event);
+
+		set_pmc_user_kernel(i, ctr[i].user, ctr[i].kernel);
+	}
+}
+
+static void fsl_booke_start(struct op_counter_config *ctr)
+{
+	int i;
+
+	mtmsr(mfmsr() | MSR_PMM);
+
+	for (i = 0; i < num_counters; ++i) {
+		if (ctr[i].enabled) {
+			ctr_write(i, reset_value[i]);
+			/* Set Each enabled counterd to only
+			 * count when the Mark bit is not set */
+			set_pmc_marked(i, 1, 0);
+			pmc_start_ctr(i, 1);
+		} else {
+			ctr_write(i, 0);
+
+			/* Set the ctr to be stopped */
+			pmc_start_ctr(i, 0);
+		}
+	}
+
+	/* Clear the freeze bit, and enable the interrupt.
+	 * The counters won't actually start until the rfi clears
+	 * the PMM bit */
+	pmc_start_ctrs(1);
+
+	oprofile_running = 1;
+
+	pr_debug("start on cpu %d, pmgc0 %x\n", smp_processor_id(),
+			mfpmr(PMRN_PMGC0));
+}
+
+static void fsl_booke_stop(void)
+{
+	/* freeze counters */
+	pmc_stop_ctrs();
+
+	oprofile_running = 0;
+
+	pr_debug("stop on cpu %d, pmgc0 %x\n", smp_processor_id(),
+			mfpmr(PMRN_PMGC0));
+
+	mb();
+}
+
+
+static void fsl_booke_handle_interrupt(struct pt_regs *regs,
+				    struct op_counter_config *ctr)
+{
+	unsigned long pc;
+	int is_kernel;
+	int val;
+	int i;
+
+	/* set the PMM bit (see comment below) */
+	mtmsr(mfmsr() | MSR_PMM);
+
+	pc = regs->nip;
+	is_kernel = (pc >= KERNELBASE);
+
+	for (i = 0; i < num_counters; ++i) {
+		val = ctr_read(i);
+		if (val < 0) {
+			if (oprofile_running && ctr[i].enabled) {
+				oprofile_add_pc(pc, is_kernel, i);
+				ctr_write(i, reset_value[i]);
+			} else {
+				ctr_write(i, 0);
+			}
+		}
+	}
+
+	/* The freeze bit was set by the interrupt. */
+	/* Clear the freeze bit, and reenable the interrupt.
+	 * The counters won't actually start until the rfi clears
+	 * the PMM bit */
+	pmc_start_ctrs(1);
+}
+
+struct op_powerpc_model op_model_fsl_booke = {
+	.reg_setup		= fsl_booke_reg_setup,
+	.start			= fsl_booke_start,
+	.stop			= fsl_booke_stop,
+	.handle_interrupt	= fsl_booke_handle_interrupt,
+};
diff --git a/arch/powerpc/oprofile/op_model_power4.c b/arch/powerpc/oprofile/op_model_power4.c
new file mode 100644
index 000000000000..886449315847
--- /dev/null
+++ b/arch/powerpc/oprofile/op_model_power4.c
@@ -0,0 +1,309 @@
+/*
+ * Copyright (C) 2004 Anton Blanchard <anton@au.ibm.com>, IBM
+ *
+ * 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/oprofile.h>
+#include <linux/init.h>
+#include <linux/smp.h>
+#include <asm/ptrace.h>
+#include <asm/system.h>
+#include <asm/processor.h>
+#include <asm/cputable.h>
+#include <asm/systemcfg.h>
+#include <asm/rtas.h>
+#include <asm/oprofile_impl.h>
+
+#define dbg(args...)
+
+static unsigned long reset_value[OP_MAX_COUNTER];
+
+static int oprofile_running;
+static int mmcra_has_sihv;
+
+/* mmcr values are set in power4_reg_setup, used in power4_cpu_setup */
+static u32 mmcr0_val;
+static u64 mmcr1_val;
+static u32 mmcra_val;
+
+/*
+ * Since we do not have an NMI, backtracing through spinlocks is
+ * only a best guess. In light of this, allow it to be disabled at
+ * runtime.
+ */
+static int backtrace_spinlocks;
+
+static void power4_reg_setup(struct op_counter_config *ctr,
+			     struct op_system_config *sys,
+			     int num_ctrs)
+{
+	int i;
+
+	/*
+	 * SIHV / SIPR bits are only implemented on POWER4+ (GQ) and above.
+	 * However we disable it on all POWER4 until we verify it works
+	 * (I was seeing some strange behaviour last time I tried).
+	 *
+	 * It has been verified to work on POWER5 so we enable it there.
+	 */
+	if (cpu_has_feature(CPU_FTR_MMCRA_SIHV))
+		mmcra_has_sihv = 1;
+
+	/*
+	 * The performance counter event settings are given in the mmcr0,
+	 * mmcr1 and mmcra values passed from the user in the
+	 * op_system_config structure (sys variable).
+	 */
+	mmcr0_val = sys->mmcr0;
+	mmcr1_val = sys->mmcr1;
+	mmcra_val = sys->mmcra;
+
+	backtrace_spinlocks = sys->backtrace_spinlocks;
+
+	for (i = 0; i < cur_cpu_spec->num_pmcs; ++i)
+		reset_value[i] = 0x80000000UL - ctr[i].count;
+
+	/* setup user and kernel profiling */
+	if (sys->enable_kernel)
+		mmcr0_val &= ~MMCR0_KERNEL_DISABLE;
+	else
+		mmcr0_val |= MMCR0_KERNEL_DISABLE;
+
+	if (sys->enable_user)
+		mmcr0_val &= ~MMCR0_PROBLEM_DISABLE;
+	else
+		mmcr0_val |= MMCR0_PROBLEM_DISABLE;
+}
+
+extern void ppc64_enable_pmcs(void);
+
+static void power4_cpu_setup(void *unused)
+{
+	unsigned int mmcr0 = mmcr0_val;
+	unsigned long mmcra = mmcra_val;
+
+	ppc64_enable_pmcs();
+
+	/* set the freeze bit */
+	mmcr0 |= MMCR0_FC;
+	mtspr(SPRN_MMCR0, mmcr0);
+
+	mmcr0 |= MMCR0_FCM1|MMCR0_PMXE|MMCR0_FCECE;
+	mmcr0 |= MMCR0_PMC1CE|MMCR0_PMCjCE;
+	mtspr(SPRN_MMCR0, mmcr0);
+
+	mtspr(SPRN_MMCR1, mmcr1_val);
+
+	mmcra |= MMCRA_SAMPLE_ENABLE;
+	mtspr(SPRN_MMCRA, mmcra);
+
+	dbg("setup on cpu %d, mmcr0 %lx\n", smp_processor_id(),
+	    mfspr(SPRN_MMCR0));
+	dbg("setup on cpu %d, mmcr1 %lx\n", smp_processor_id(),
+	    mfspr(SPRN_MMCR1));
+	dbg("setup on cpu %d, mmcra %lx\n", smp_processor_id(),
+	    mfspr(SPRN_MMCRA));
+}
+
+static void power4_start(struct op_counter_config *ctr)
+{
+	int i;
+	unsigned int mmcr0;
+
+	/* set the PMM bit (see comment below) */
+	mtmsrd(mfmsr() | MSR_PMM);
+
+	for (i = 0; i < cur_cpu_spec->num_pmcs; ++i) {
+		if (ctr[i].enabled) {
+			ctr_write(i, reset_value[i]);
+		} else {
+			ctr_write(i, 0);
+		}
+	}
+
+	mmcr0 = mfspr(SPRN_MMCR0);
+
+	/*
+	 * We must clear the PMAO bit on some (GQ) chips. Just do it
+	 * all the time
+	 */
+	mmcr0 &= ~MMCR0_PMAO;
+
+	/*
+	 * now clear the freeze bit, counting will not start until we
+	 * rfid from this excetion, because only at that point will
+	 * the PMM bit be cleared
+	 */
+	mmcr0 &= ~MMCR0_FC;
+	mtspr(SPRN_MMCR0, mmcr0);
+
+	oprofile_running = 1;
+
+	dbg("start on cpu %d, mmcr0 %x\n", smp_processor_id(), mmcr0);
+}
+
+static void power4_stop(void)
+{
+	unsigned int mmcr0;
+
+	/* freeze counters */
+	mmcr0 = mfspr(SPRN_MMCR0);
+	mmcr0 |= MMCR0_FC;
+	mtspr(SPRN_MMCR0, mmcr0);
+
+	oprofile_running = 0;
+
+	dbg("stop on cpu %d, mmcr0 %x\n", smp_processor_id(), mmcr0);
+
+	mb();
+}
+
+/* Fake functions used by canonicalize_pc */
+static void __attribute_used__ hypervisor_bucket(void)
+{
+}
+
+static void __attribute_used__ rtas_bucket(void)
+{
+}
+
+static void __attribute_used__ kernel_unknown_bucket(void)
+{
+}
+
+static unsigned long check_spinlock_pc(struct pt_regs *regs,
+				       unsigned long profile_pc)
+{
+	unsigned long pc = instruction_pointer(regs);
+
+	/*
+	 * If both the SIAR (sampled instruction) and the perfmon exception
+	 * occurred in a spinlock region then we account the sample to the
+	 * calling function. This isnt 100% correct, we really need soft
+	 * IRQ disable so we always get the perfmon exception at the
+	 * point at which the SIAR is set.
+	 */
+	if (backtrace_spinlocks && in_lock_functions(pc) &&
+			in_lock_functions(profile_pc))
+		return regs->link;
+	else
+		return profile_pc;
+}
+
+/*
+ * On GQ and newer the MMCRA stores the HV and PR bits at the time
+ * the SIAR was sampled. We use that to work out if the SIAR was sampled in
+ * the hypervisor, our exception vectors or RTAS.
+ */
+static unsigned long get_pc(struct pt_regs *regs)
+{
+	unsigned long pc = mfspr(SPRN_SIAR);
+	unsigned long mmcra;
+
+	/* Cant do much about it */
+	if (!mmcra_has_sihv)
+		return check_spinlock_pc(regs, pc);
+
+	mmcra = mfspr(SPRN_MMCRA);
+
+	/* Were we in the hypervisor? */
+	if ((systemcfg->platform == PLATFORM_PSERIES_LPAR) &&
+	    (mmcra & MMCRA_SIHV))
+		/* function descriptor madness */
+		return *((unsigned long *)hypervisor_bucket);
+
+	/* We were in userspace, nothing to do */
+	if (mmcra & MMCRA_SIPR)
+		return pc;
+
+#ifdef CONFIG_PPC_RTAS
+	/* Were we in RTAS? */
+	if (pc >= rtas.base && pc < (rtas.base + rtas.size))
+		/* function descriptor madness */
+		return *((unsigned long *)rtas_bucket);
+#endif
+
+	/* Were we in our exception vectors or SLB real mode miss handler? */
+	if (pc < 0x1000000UL)
+		return (unsigned long)__va(pc);
+
+	/* Not sure where we were */
+	if (pc < KERNELBASE)
+		/* function descriptor madness */
+		return *((unsigned long *)kernel_unknown_bucket);
+
+	return check_spinlock_pc(regs, pc);
+}
+
+static int get_kernel(unsigned long pc)
+{
+	int is_kernel;
+
+	if (!mmcra_has_sihv) {
+		is_kernel = (pc >= KERNELBASE);
+	} else {
+		unsigned long mmcra = mfspr(SPRN_MMCRA);
+		is_kernel = ((mmcra & MMCRA_SIPR) == 0);
+	}
+
+	return is_kernel;
+}
+
+static void power4_handle_interrupt(struct pt_regs *regs,
+				    struct op_counter_config *ctr)
+{
+	unsigned long pc;
+	int is_kernel;
+	int val;
+	int i;
+	unsigned int mmcr0;
+
+	pc = get_pc(regs);
+	is_kernel = get_kernel(pc);
+
+	/* set the PMM bit (see comment below) */
+	mtmsrd(mfmsr() | MSR_PMM);
+
+	for (i = 0; i < cur_cpu_spec->num_pmcs; ++i) {
+		val = ctr_read(i);
+		if (val < 0) {
+			if (oprofile_running && ctr[i].enabled) {
+				oprofile_add_pc(pc, is_kernel, i);
+				ctr_write(i, reset_value[i]);
+			} else {
+				ctr_write(i, 0);
+			}
+		}
+	}
+
+	mmcr0 = mfspr(SPRN_MMCR0);
+
+	/* reset the perfmon trigger */
+	mmcr0 |= MMCR0_PMXE;
+
+	/*
+	 * We must clear the PMAO bit on some (GQ) chips. Just do it
+	 * all the time
+	 */
+	mmcr0 &= ~MMCR0_PMAO;
+
+	/*
+	 * now clear the freeze bit, counting will not start until we
+	 * rfid from this exception, because only at that point will
+	 * the PMM bit be cleared
+	 */
+	mmcr0 &= ~MMCR0_FC;
+	mtspr(SPRN_MMCR0, mmcr0);
+}
+
+struct op_powerpc_model op_model_power4 = {
+	.reg_setup		= power4_reg_setup,
+	.cpu_setup		= power4_cpu_setup,
+	.start			= power4_start,
+	.stop			= power4_stop,
+	.handle_interrupt	= power4_handle_interrupt,
+};
diff --git a/arch/powerpc/oprofile/op_model_rs64.c b/arch/powerpc/oprofile/op_model_rs64.c
new file mode 100644
index 000000000000..e010b85996e8
--- /dev/null
+++ b/arch/powerpc/oprofile/op_model_rs64.c
@@ -0,0 +1,218 @@
+/*
+ * Copyright (C) 2004 Anton Blanchard <anton@au.ibm.com>, IBM
+ *
+ * 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/oprofile.h>
+#include <linux/init.h>
+#include <linux/smp.h>
+#include <asm/ptrace.h>
+#include <asm/system.h>
+#include <asm/processor.h>
+#include <asm/cputable.h>
+#include <asm/oprofile_impl.h>
+
+#define dbg(args...)
+
+static void ctrl_write(unsigned int i, unsigned int val)
+{
+	unsigned int tmp = 0;
+	unsigned long shift = 0, mask = 0;
+
+	dbg("ctrl_write %d %x\n", i, val);
+
+	switch(i) {
+	case 0:
+		tmp = mfspr(SPRN_MMCR0);
+		shift = 6;
+		mask = 0x7F;
+		break;
+	case 1:
+		tmp = mfspr(SPRN_MMCR0);
+		shift = 0;
+		mask = 0x3F;
+		break;
+	case 2:
+		tmp = mfspr(SPRN_MMCR1);
+		shift = 31 - 4;
+		mask = 0x1F;
+		break;
+	case 3:
+		tmp = mfspr(SPRN_MMCR1);
+		shift = 31 - 9;
+		mask = 0x1F;
+		break;
+	case 4:
+		tmp = mfspr(SPRN_MMCR1);
+		shift = 31 - 14;
+		mask = 0x1F;
+		break;
+	case 5:
+		tmp = mfspr(SPRN_MMCR1);
+		shift = 31 - 19;
+		mask = 0x1F;
+		break;
+	case 6:
+		tmp = mfspr(SPRN_MMCR1);
+		shift = 31 - 24;
+		mask = 0x1F;
+		break;
+	case 7:
+		tmp = mfspr(SPRN_MMCR1);
+		shift = 31 - 28;
+		mask = 0xF;
+		break;
+	}
+
+	tmp = tmp & ~(mask << shift);
+	tmp |= val << shift;
+
+	switch(i) {
+		case 0:
+		case 1:
+			mtspr(SPRN_MMCR0, tmp);
+			break;
+		default:
+			mtspr(SPRN_MMCR1, tmp);
+	}
+
+	dbg("ctrl_write mmcr0 %lx mmcr1 %lx\n", mfspr(SPRN_MMCR0),
+	       mfspr(SPRN_MMCR1));
+}
+
+static unsigned long reset_value[OP_MAX_COUNTER];
+
+static int num_counters;
+
+static void rs64_reg_setup(struct op_counter_config *ctr,
+			   struct op_system_config *sys,
+			   int num_ctrs)
+{
+	int i;
+
+	num_counters = num_ctrs;
+
+	for (i = 0; i < num_counters; ++i)
+		reset_value[i] = 0x80000000UL - ctr[i].count;
+
+	/* XXX setup user and kernel profiling */
+}
+
+static void rs64_cpu_setup(void *unused)
+{
+	unsigned int mmcr0;
+
+	/* reset MMCR0 and set the freeze bit */
+	mmcr0 = MMCR0_FC;
+	mtspr(SPRN_MMCR0, mmcr0);
+
+	/* reset MMCR1, MMCRA */
+	mtspr(SPRN_MMCR1, 0);
+
+	if (cpu_has_feature(CPU_FTR_MMCRA))
+		mtspr(SPRN_MMCRA, 0);
+
+	mmcr0 |= MMCR0_FCM1|MMCR0_PMXE|MMCR0_FCECE;
+	/* Only applies to POWER3, but should be safe on RS64 */
+	mmcr0 |= MMCR0_PMC1CE|MMCR0_PMCjCE;
+	mtspr(SPRN_MMCR0, mmcr0);
+
+	dbg("setup on cpu %d, mmcr0 %lx\n", smp_processor_id(),
+	    mfspr(SPRN_MMCR0));
+	dbg("setup on cpu %d, mmcr1 %lx\n", smp_processor_id(),
+	    mfspr(SPRN_MMCR1));
+}
+
+static void rs64_start(struct op_counter_config *ctr)
+{
+	int i;
+	unsigned int mmcr0;
+
+	/* set the PMM bit (see comment below) */
+	mtmsrd(mfmsr() | MSR_PMM);
+
+	for (i = 0; i < num_counters; ++i) {
+		if (ctr[i].enabled) {
+			ctr_write(i, reset_value[i]);
+			ctrl_write(i, ctr[i].event);
+		} else {
+			ctr_write(i, 0);
+		}
+	}
+
+	mmcr0 = mfspr(SPRN_MMCR0);
+
+	/*
+	 * now clear the freeze bit, counting will not start until we
+	 * rfid from this excetion, because only at that point will
+	 * the PMM bit be cleared
+	 */
+	mmcr0 &= ~MMCR0_FC;
+	mtspr(SPRN_MMCR0, mmcr0);
+
+	dbg("start on cpu %d, mmcr0 %x\n", smp_processor_id(), mmcr0);
+}
+
+static void rs64_stop(void)
+{
+	unsigned int mmcr0;
+
+	/* freeze counters */
+	mmcr0 = mfspr(SPRN_MMCR0);
+	mmcr0 |= MMCR0_FC;
+	mtspr(SPRN_MMCR0, mmcr0);
+
+	dbg("stop on cpu %d, mmcr0 %x\n", smp_processor_id(), mmcr0);
+
+	mb();
+}
+
+static void rs64_handle_interrupt(struct pt_regs *regs,
+				  struct op_counter_config *ctr)
+{
+	unsigned int mmcr0;
+	int val;
+	int i;
+	unsigned long pc = mfspr(SPRN_SIAR);
+	int is_kernel = (pc >= KERNELBASE);
+
+	/* set the PMM bit (see comment below) */
+	mtmsrd(mfmsr() | MSR_PMM);
+
+	for (i = 0; i < num_counters; ++i) {
+		val = ctr_read(i);
+		if (val < 0) {
+			if (ctr[i].enabled) {
+				oprofile_add_pc(pc, is_kernel, i);
+				ctr_write(i, reset_value[i]);
+			} else {
+				ctr_write(i, 0);
+			}
+		}
+	}
+
+	mmcr0 = mfspr(SPRN_MMCR0);
+
+	/* reset the perfmon trigger */
+	mmcr0 |= MMCR0_PMXE;
+
+	/*
+	 * now clear the freeze bit, counting will not start until we
+	 * rfid from this exception, because only at that point will
+	 * the PMM bit be cleared
+	 */
+	mmcr0 &= ~MMCR0_FC;
+	mtspr(SPRN_MMCR0, mmcr0);
+}
+
+struct op_powerpc_model op_model_rs64 = {
+	.reg_setup		= rs64_reg_setup,
+	.cpu_setup		= rs64_cpu_setup,
+	.start			= rs64_start,
+	.stop			= rs64_stop,
+	.handle_interrupt	= rs64_handle_interrupt,
+};