Initial import

4 files changed, 1310 insertions, 0 deletions

diff --git a/arch/powerpc/oprofile/cell/pr_util.h b/arch/powerpc/oprofile/cell/pr_util.h
new file mode 100644
index 000000000..964b93974
--- /dev/null
+++ b/arch/powerpc/oprofile/cell/pr_util.h
@@ -0,0 +1,114 @@
+ /*
+ * Cell Broadband Engine OProfile Support
+ *
+ * (C) Copyright IBM Corporation 2006
+ *
+ * Author: Maynard Johnson <maynardj@us.ibm.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.
+ */
+
+#ifndef PR_UTIL_H
+#define PR_UTIL_H
+
+#include <linux/cpumask.h>
+#include <linux/oprofile.h>
+#include <asm/cell-pmu.h>
+#include <asm/cell-regs.h>
+#include <asm/spu.h>
+
+/* Defines used for sync_start */
+#define SKIP_GENERIC_SYNC 0
+#define SYNC_START_ERROR -1
+#define DO_GENERIC_SYNC 1
+#define SPUS_PER_NODE   8
+#define DEFAULT_TIMER_EXPIRE  (HZ / 10)
+
+extern struct delayed_work spu_work;
+extern int spu_prof_running;
+
+#define TRACE_ARRAY_SIZE 1024
+
+extern spinlock_t oprof_spu_smpl_arry_lck;
+
+struct spu_overlay_info {	/* map of sections within an SPU overlay */
+	unsigned int vma;	/* SPU virtual memory address from elf */
+	unsigned int size;	/* size of section from elf */
+	unsigned int offset;	/* offset of section into elf file */
+	unsigned int buf;
+};
+
+struct vma_to_fileoffset_map {	/* map of sections within an SPU program */
+	struct vma_to_fileoffset_map *next;	/* list pointer */
+	unsigned int vma;	/* SPU virtual memory address from elf */
+	unsigned int size;	/* size of section from elf */
+	unsigned int offset;	/* offset of section into elf file */
+	unsigned int guard_ptr;
+	unsigned int guard_val;
+        /*
+	 * The guard pointer is an entry in the _ovly_buf_table,
+	 * computed using ovly.buf as the index into the table.  Since
+	 * ovly.buf values begin at '1' to reference the first (or 0th)
+	 * entry in the _ovly_buf_table, the computation subtracts 1
+	 * from ovly.buf.
+	 * The guard value is stored in the _ovly_buf_table entry and
+	 * is an index (starting at 1) back to the _ovly_table entry
+	 * that is pointing at this _ovly_buf_table entry.  So, for
+	 * example, for an overlay scenario with one overlay segment
+	 * and two overlay sections:
+	 *      - Section 1 points to the first entry of the
+	 *        _ovly_buf_table, which contains a guard value
+	 *        of '1', referencing the first (index=0) entry of
+	 *        _ovly_table.
+	 *      - Section 2 points to the second entry of the
+	 *        _ovly_buf_table, which contains a guard value
+	 *        of '2', referencing the second (index=1) entry of
+	 *        _ovly_table.
+	 */
+
+};
+
+struct spu_buffer {
+	int last_guard_val;
+	int ctx_sw_seen;
+	unsigned long *buff;
+	unsigned int head, tail;
+};
+
+
+/* The three functions below are for maintaining and accessing
+ * the vma-to-fileoffset map.
+ */
+struct vma_to_fileoffset_map *create_vma_map(const struct spu *spu,
+					     unsigned long objectid);
+unsigned int vma_map_lookup(struct vma_to_fileoffset_map *map,
+			    unsigned int vma, const struct spu *aSpu,
+			    int *grd_val);
+void vma_map_free(struct vma_to_fileoffset_map *map);
+
+/*
+ * Entry point for SPU profiling.
+ * cycles_reset is the SPU_CYCLES count value specified by the user.
+ */
+int start_spu_profiling_cycles(unsigned int cycles_reset);
+void start_spu_profiling_events(void);
+
+void stop_spu_profiling_cycles(void);
+void stop_spu_profiling_events(void);
+
+/* add the necessary profiling hooks */
+int spu_sync_start(void);
+
+/* remove the hooks */
+int spu_sync_stop(void);
+
+/* Record SPU program counter samples to the oprofile event buffer. */
+void spu_sync_buffer(int spu_num, unsigned int *samples,
+		     int num_samples);
+
+void set_spu_profiling_frequency(unsigned int freq_khz, unsigned int cycles_reset);
+
+#endif	  /* PR_UTIL_H */
diff --git a/arch/powerpc/oprofile/cell/spu_profiler.c b/arch/powerpc/oprofile/cell/spu_profiler.c
new file mode 100644
index 000000000..b129d007e
--- /dev/null
+++ b/arch/powerpc/oprofile/cell/spu_profiler.c
@@ -0,0 +1,252 @@
+/*
+ * Cell Broadband Engine OProfile Support
+ *
+ * (C) Copyright IBM Corporation 2006
+ *
+ * Authors: Maynard Johnson <maynardj@us.ibm.com>
+ *	    Carl Love <carll@us.ibm.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/hrtimer.h>
+#include <linux/smp.h>
+#include <linux/slab.h>
+#include <asm/cell-pmu.h>
+#include <asm/time.h>
+#include "pr_util.h"
+
+#define SCALE_SHIFT 14
+
+static u32 *samples;
+
+/* spu_prof_running is a flag used to indicate if spu profiling is enabled
+ * or not.  It is set by the routines start_spu_profiling_cycles() and
+ * start_spu_profiling_events().  The flag is cleared by the routines
+ * stop_spu_profiling_cycles() and stop_spu_profiling_events().  These
+ * routines are called via global_start() and global_stop() which are called in
+ * op_powerpc_start() and op_powerpc_stop().  These routines are called once
+ * per system as a result of the user starting/stopping oprofile.  Hence, only
+ * one CPU per user at a time will be changing  the value of spu_prof_running.
+ * In general, OProfile does not protect against multiple users trying to run
+ * OProfile at a time.
+ */
+int spu_prof_running;
+static unsigned int profiling_interval;
+
+#define NUM_SPU_BITS_TRBUF 16
+#define SPUS_PER_TB_ENTRY   4
+
+#define SPU_PC_MASK	     0xFFFF
+
+DEFINE_SPINLOCK(oprof_spu_smpl_arry_lck);
+unsigned long oprof_spu_smpl_arry_lck_flags;
+
+void set_spu_profiling_frequency(unsigned int freq_khz, unsigned int cycles_reset)
+{
+	unsigned long ns_per_cyc;
+
+	if (!freq_khz)
+		freq_khz = ppc_proc_freq/1000;
+
+	/* To calculate a timeout in nanoseconds, the basic
+	 * formula is ns = cycles_reset * (NSEC_PER_SEC / cpu frequency).
+	 * To avoid floating point math, we use the scale math
+	 * technique as described in linux/jiffies.h.  We use
+	 * a scale factor of SCALE_SHIFT, which provides 4 decimal places
+	 * of precision.  This is close enough for the purpose at hand.
+	 *
+	 * The value of the timeout should be small enough that the hw
+	 * trace buffer will not get more than about 1/3 full for the
+	 * maximum user specified (the LFSR value) hw sampling frequency.
+	 * This is to ensure the trace buffer will never fill even if the
+	 * kernel thread scheduling varies under a heavy system load.
+	 */
+
+	ns_per_cyc = (USEC_PER_SEC << SCALE_SHIFT)/freq_khz;
+	profiling_interval = (ns_per_cyc * cycles_reset) >> SCALE_SHIFT;
+
+}
+
+/*
+ * Extract SPU PC from trace buffer entry
+ */
+static void spu_pc_extract(int cpu, int entry)
+{
+	/* the trace buffer is 128 bits */
+	u64 trace_buffer[2];
+	u64 spu_mask;
+	int spu;
+
+	spu_mask = SPU_PC_MASK;
+
+	/* Each SPU PC is 16 bits; hence, four spus in each of
+	 * the two 64-bit buffer entries that make up the
+	 * 128-bit trace_buffer entry.	Process two 64-bit values
+	 * simultaneously.
+	 * trace[0] SPU PC contents are: 0 1 2 3
+	 * trace[1] SPU PC contents are: 4 5 6 7
+	 */
+
+	cbe_read_trace_buffer(cpu, trace_buffer);
+
+	for (spu = SPUS_PER_TB_ENTRY-1; spu >= 0; spu--) {
+		/* spu PC trace entry is upper 16 bits of the
+		 * 18 bit SPU program counter
+		 */
+		samples[spu * TRACE_ARRAY_SIZE + entry]
+			= (spu_mask & trace_buffer[0]) << 2;
+		samples[(spu + SPUS_PER_TB_ENTRY) * TRACE_ARRAY_SIZE + entry]
+			= (spu_mask & trace_buffer[1]) << 2;
+
+		trace_buffer[0] = trace_buffer[0] >> NUM_SPU_BITS_TRBUF;
+		trace_buffer[1] = trace_buffer[1] >> NUM_SPU_BITS_TRBUF;
+	}
+}
+
+static int cell_spu_pc_collection(int cpu)
+{
+	u32 trace_addr;
+	int entry;
+
+	/* process the collected SPU PC for the node */
+
+	entry = 0;
+
+	trace_addr = cbe_read_pm(cpu, trace_address);
+	while (!(trace_addr & CBE_PM_TRACE_BUF_EMPTY)) {
+		/* there is data in the trace buffer to process */
+		spu_pc_extract(cpu, entry);
+
+		entry++;
+
+		if (entry >= TRACE_ARRAY_SIZE)
+			/* spu_samples is full */
+			break;
+
+		trace_addr = cbe_read_pm(cpu, trace_address);
+	}
+
+	return entry;
+}
+
+
+static enum hrtimer_restart profile_spus(struct hrtimer *timer)
+{
+	ktime_t kt;
+	int cpu, node, k, num_samples, spu_num;
+
+	if (!spu_prof_running)
+		goto stop;
+
+	for_each_online_cpu(cpu) {
+		if (cbe_get_hw_thread_id(cpu))
+			continue;
+
+		node = cbe_cpu_to_node(cpu);
+
+		/* There should only be one kernel thread at a time processing
+		 * the samples.	 In the very unlikely case that the processing
+		 * is taking a very long time and multiple kernel threads are
+		 * started to process the samples.  Make sure only one kernel
+		 * thread is working on the samples array at a time.  The
+		 * sample array must be loaded and then processed for a given
+		 * cpu.	 The sample array is not per cpu.
+		 */
+		spin_lock_irqsave(&oprof_spu_smpl_arry_lck,
+				  oprof_spu_smpl_arry_lck_flags);
+		num_samples = cell_spu_pc_collection(cpu);
+
+		if (num_samples == 0) {
+			spin_unlock_irqrestore(&oprof_spu_smpl_arry_lck,
+					       oprof_spu_smpl_arry_lck_flags);
+			continue;
+		}
+
+		for (k = 0; k < SPUS_PER_NODE; k++) {
+			spu_num = k + (node * SPUS_PER_NODE);
+			spu_sync_buffer(spu_num,
+					samples + (k * TRACE_ARRAY_SIZE),
+					num_samples);
+		}
+
+		spin_unlock_irqrestore(&oprof_spu_smpl_arry_lck,
+				       oprof_spu_smpl_arry_lck_flags);
+
+	}
+	smp_wmb();	/* insure spu event buffer updates are written */
+			/* don't want events intermingled... */
+
+	kt = ktime_set(0, profiling_interval);
+	if (!spu_prof_running)
+		goto stop;
+	hrtimer_forward(timer, timer->base->get_time(), kt);
+	return HRTIMER_RESTART;
+
+ stop:
+	printk(KERN_INFO "SPU_PROF: spu-prof timer ending\n");
+	return HRTIMER_NORESTART;
+}
+
+static struct hrtimer timer;
+/*
+ * Entry point for SPU cycle profiling.
+ * NOTE:  SPU profiling is done system-wide, not per-CPU.
+ *
+ * cycles_reset is the count value specified by the user when
+ * setting up OProfile to count SPU_CYCLES.
+ */
+int start_spu_profiling_cycles(unsigned int cycles_reset)
+{
+	ktime_t kt;
+
+	pr_debug("timer resolution: %lu\n", TICK_NSEC);
+	kt = ktime_set(0, profiling_interval);
+	hrtimer_init(&timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
+	hrtimer_set_expires(&timer, kt);
+	timer.function = profile_spus;
+
+	/* Allocate arrays for collecting SPU PC samples */
+	samples = kzalloc(SPUS_PER_NODE *
+			  TRACE_ARRAY_SIZE * sizeof(u32), GFP_KERNEL);
+
+	if (!samples)
+		return -ENOMEM;
+
+	spu_prof_running = 1;
+	hrtimer_start(&timer, kt, HRTIMER_MODE_REL);
+	schedule_delayed_work(&spu_work, DEFAULT_TIMER_EXPIRE);
+
+	return 0;
+}
+
+/*
+ * Entry point for SPU event profiling.
+ * NOTE:  SPU profiling is done system-wide, not per-CPU.
+ *
+ * cycles_reset is the count value specified by the user when
+ * setting up OProfile to count SPU_CYCLES.
+ */
+void start_spu_profiling_events(void)
+{
+	spu_prof_running = 1;
+	schedule_delayed_work(&spu_work, DEFAULT_TIMER_EXPIRE);
+
+	return;
+}
+
+void stop_spu_profiling_cycles(void)
+{
+	spu_prof_running = 0;
+	hrtimer_cancel(&timer);
+	kfree(samples);
+	pr_debug("SPU_PROF: stop_spu_profiling_cycles issued\n");
+}
+
+void stop_spu_profiling_events(void)
+{
+	spu_prof_running = 0;
+}
diff --git a/arch/powerpc/oprofile/cell/spu_task_sync.c b/arch/powerpc/oprofile/cell/spu_task_sync.c
new file mode 100644
index 000000000..ed7b09770
--- /dev/null
+++ b/arch/powerpc/oprofile/cell/spu_task_sync.c
@@ -0,0 +1,661 @@
+/*
+ * Cell Broadband Engine OProfile Support
+ *
+ * (C) Copyright IBM Corporation 2006
+ *
+ * Author: Maynard Johnson <maynardj@us.ibm.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.
+ */
+
+/* The purpose of this file is to handle SPU event task switching
+ * and to record SPU context information into the OProfile
+ * event buffer.
+ *
+ * Additionally, the spu_sync_buffer function is provided as a helper
+ * for recoding actual SPU program counter samples to the event buffer.
+ */
+#include <linux/dcookies.h>
+#include <linux/kref.h>
+#include <linux/mm.h>
+#include <linux/fs.h>
+#include <linux/file.h>
+#include <linux/module.h>
+#include <linux/notifier.h>
+#include <linux/numa.h>
+#include <linux/oprofile.h>
+#include <linux/slab.h>
+#include <linux/spinlock.h>
+#include "pr_util.h"
+
+#define RELEASE_ALL 9999
+
+static DEFINE_SPINLOCK(buffer_lock);
+static DEFINE_SPINLOCK(cache_lock);
+static int num_spu_nodes;
+int spu_prof_num_nodes;
+
+struct spu_buffer spu_buff[MAX_NUMNODES * SPUS_PER_NODE];
+struct delayed_work spu_work;
+static unsigned max_spu_buff;
+
+static void spu_buff_add(unsigned long int value, int spu)
+{
+	/* spu buff is a circular buffer.  Add entries to the
+	 * head.  Head is the index to store the next value.
+	 * The buffer is full when there is one available entry
+	 * in the queue, i.e. head and tail can't be equal.
+	 * That way we can tell the difference between the
+	 * buffer being full versus empty.
+	 *
+	 *  ASSUPTION: the buffer_lock is held when this function
+	 *             is called to lock the buffer, head and tail.
+	 */
+	int full = 1;
+
+	if (spu_buff[spu].head >= spu_buff[spu].tail) {
+		if ((spu_buff[spu].head - spu_buff[spu].tail)
+		    <  (max_spu_buff - 1))
+			full = 0;
+
+	} else if (spu_buff[spu].tail > spu_buff[spu].head) {
+		if ((spu_buff[spu].tail - spu_buff[spu].head)
+		    > 1)
+			full = 0;
+	}
+
+	if (!full) {
+		spu_buff[spu].buff[spu_buff[spu].head] = value;
+		spu_buff[spu].head++;
+
+		if (spu_buff[spu].head >= max_spu_buff)
+			spu_buff[spu].head = 0;
+	} else {
+		/* From the user's perspective make the SPU buffer
+		 * size management/overflow look like we are using
+		 * per cpu buffers.  The user uses the same
+		 * per cpu parameter to adjust the SPU buffer size.
+		 * Increment the sample_lost_overflow to inform
+		 * the user the buffer size needs to be increased.
+		 */
+		oprofile_cpu_buffer_inc_smpl_lost();
+	}
+}
+
+/* This function copies the per SPU buffers to the
+ * OProfile kernel buffer.
+ */
+void sync_spu_buff(void)
+{
+	int spu;
+	unsigned long flags;
+	int curr_head;
+
+	for (spu = 0; spu < num_spu_nodes; spu++) {
+		/* In case there was an issue and the buffer didn't
+		 * get created skip it.
+		 */
+		if (spu_buff[spu].buff == NULL)
+			continue;
+
+		/* Hold the lock to make sure the head/tail
+		 * doesn't change while spu_buff_add() is
+		 * deciding if the buffer is full or not.
+		 * Being a little paranoid.
+		 */
+		spin_lock_irqsave(&buffer_lock, flags);
+		curr_head = spu_buff[spu].head;
+		spin_unlock_irqrestore(&buffer_lock, flags);
+
+		/* Transfer the current contents to the kernel buffer.
+		 * data can still be added to the head of the buffer.
+		 */
+		oprofile_put_buff(spu_buff[spu].buff,
+				  spu_buff[spu].tail,
+				  curr_head, max_spu_buff);
+
+		spin_lock_irqsave(&buffer_lock, flags);
+		spu_buff[spu].tail = curr_head;
+		spin_unlock_irqrestore(&buffer_lock, flags);
+	}
+
+}
+
+static void wq_sync_spu_buff(struct work_struct *work)
+{
+	/* move data from spu buffers to kernel buffer */
+	sync_spu_buff();
+
+	/* only reschedule if profiling is not done */
+	if (spu_prof_running)
+		schedule_delayed_work(&spu_work, DEFAULT_TIMER_EXPIRE);
+}
+
+/* Container for caching information about an active SPU task. */
+struct cached_info {
+	struct vma_to_fileoffset_map *map;
+	struct spu *the_spu;	/* needed to access pointer to local_store */
+	struct kref cache_ref;
+};
+
+static struct cached_info *spu_info[MAX_NUMNODES * 8];
+
+static void destroy_cached_info(struct kref *kref)
+{
+	struct cached_info *info;
+
+	info = container_of(kref, struct cached_info, cache_ref);
+	vma_map_free(info->map);
+	kfree(info);
+	module_put(THIS_MODULE);
+}
+
+/* Return the cached_info for the passed SPU number.
+ * ATTENTION:  Callers are responsible for obtaining the
+ *	       cache_lock if needed prior to invoking this function.
+ */
+static struct cached_info *get_cached_info(struct spu *the_spu, int spu_num)
+{
+	struct kref *ref;
+	struct cached_info *ret_info;
+
+	if (spu_num >= num_spu_nodes) {
+		printk(KERN_ERR "SPU_PROF: "
+		       "%s, line %d: Invalid index %d into spu info cache\n",
+		       __func__, __LINE__, spu_num);
+		ret_info = NULL;
+		goto out;
+	}
+	if (!spu_info[spu_num] && the_spu) {
+		ref = spu_get_profile_private_kref(the_spu->ctx);
+		if (ref) {
+			spu_info[spu_num] = container_of(ref, struct cached_info, cache_ref);
+			kref_get(&spu_info[spu_num]->cache_ref);
+		}
+	}
+
+	ret_info = spu_info[spu_num];
+ out:
+	return ret_info;
+}
+
+
+/* Looks for cached info for the passed spu.  If not found, the
+ * cached info is created for the passed spu.
+ * Returns 0 for success; otherwise, -1 for error.
+ */
+static int
+prepare_cached_spu_info(struct spu *spu, unsigned long objectId)
+{
+	unsigned long flags;
+	struct vma_to_fileoffset_map *new_map;
+	int retval = 0;
+	struct cached_info *info;
+
+	/* We won't bother getting cache_lock here since
+	 * don't do anything with the cached_info that's returned.
+	 */
+	info = get_cached_info(spu, spu->number);
+
+	if (info) {
+		pr_debug("Found cached SPU info.\n");
+		goto out;
+	}
+
+	/* Create cached_info and set spu_info[spu->number] to point to it.
+	 * spu->number is a system-wide value, not a per-node value.
+	 */
+	info = kzalloc(sizeof(struct cached_info), GFP_KERNEL);
+	if (!info) {
+		printk(KERN_ERR "SPU_PROF: "
+		       "%s, line %d: create vma_map failed\n",
+		       __func__, __LINE__);
+		retval = -ENOMEM;
+		goto err_alloc;
+	}
+	new_map = create_vma_map(spu, objectId);
+	if (!new_map) {
+		printk(KERN_ERR "SPU_PROF: "
+		       "%s, line %d: create vma_map failed\n",
+		       __func__, __LINE__);
+		retval = -ENOMEM;
+		goto err_alloc;
+	}
+
+	pr_debug("Created vma_map\n");
+	info->map = new_map;
+	info->the_spu = spu;
+	kref_init(&info->cache_ref);
+	spin_lock_irqsave(&cache_lock, flags);
+	spu_info[spu->number] = info;
+	/* Increment count before passing off ref to SPUFS. */
+	kref_get(&info->cache_ref);
+
+	/* We increment the module refcount here since SPUFS is
+	 * responsible for the final destruction of the cached_info,
+	 * and it must be able to access the destroy_cached_info()
+	 * function defined in the OProfile module.  We decrement
+	 * the module refcount in destroy_cached_info.
+	 */
+	try_module_get(THIS_MODULE);
+	spu_set_profile_private_kref(spu->ctx, &info->cache_ref,
+				destroy_cached_info);
+	spin_unlock_irqrestore(&cache_lock, flags);
+	goto out;
+
+err_alloc:
+	kfree(info);
+out:
+	return retval;
+}
+
+/*
+ * NOTE:  The caller is responsible for locking the
+ *	  cache_lock prior to calling this function.
+ */
+static int release_cached_info(int spu_index)
+{
+	int index, end;
+
+	if (spu_index == RELEASE_ALL) {
+		end = num_spu_nodes;
+		index = 0;
+	} else {
+		if (spu_index >= num_spu_nodes) {
+			printk(KERN_ERR "SPU_PROF: "
+				"%s, line %d: "
+				"Invalid index %d into spu info cache\n",
+				__func__, __LINE__, spu_index);
+			goto out;
+		}
+		end = spu_index + 1;
+		index = spu_index;
+	}
+	for (; index < end; index++) {
+		if (spu_info[index]) {
+			kref_put(&spu_info[index]->cache_ref,
+				 destroy_cached_info);
+			spu_info[index] = NULL;
+		}
+	}
+
+out:
+	return 0;
+}
+
+/* The source code for fast_get_dcookie was "borrowed"
+ * from drivers/oprofile/buffer_sync.c.
+ */
+
+/* Optimisation. We can manage without taking the dcookie sem
+ * because we cannot reach this code without at least one
+ * dcookie user still being registered (namely, the reader
+ * of the event buffer).
+ */
+static inline unsigned long fast_get_dcookie(struct path *path)
+{
+	unsigned long cookie;
+
+	if (path->dentry->d_flags & DCACHE_COOKIE)
+		return (unsigned long)path->dentry;
+	get_dcookie(path, &cookie);
+	return cookie;
+}
+
+/* Look up the dcookie for the task's mm->exe_file,
+ * which corresponds loosely to "application name". Also, determine
+ * the offset for the SPU ELF object.  If computed offset is
+ * non-zero, it implies an embedded SPU object; otherwise, it's a
+ * separate SPU binary, in which case we retrieve it's dcookie.
+ * For the embedded case, we must determine if SPU ELF is embedded
+ * in the executable application or another file (i.e., shared lib).
+ * If embedded in a shared lib, we must get the dcookie and return
+ * that to the caller.
+ */
+static unsigned long
+get_exec_dcookie_and_offset(struct spu *spu, unsigned int *offsetp,
+			    unsigned long *spu_bin_dcookie,
+			    unsigned long spu_ref)
+{
+	unsigned long app_cookie = 0;
+	unsigned int my_offset = 0;
+	struct vm_area_struct *vma;
+	struct file *exe_file;
+	struct mm_struct *mm = spu->mm;
+
+	if (!mm)
+		goto out;
+
+	exe_file = get_mm_exe_file(mm);
+	if (exe_file) {
+		app_cookie = fast_get_dcookie(&exe_file->f_path);
+		pr_debug("got dcookie for %pD\n", exe_file);
+		fput(exe_file);
+	}
+
+	down_read(&mm->mmap_sem);
+	for (vma = mm->mmap; vma; vma = vma->vm_next) {
+		if (vma->vm_start > spu_ref || vma->vm_end <= spu_ref)
+			continue;
+		my_offset = spu_ref - vma->vm_start;
+		if (!vma->vm_file)
+			goto fail_no_image_cookie;
+
+		pr_debug("Found spu ELF at %X(object-id:%lx) for file %pD\n",
+			 my_offset, spu_ref, vma->vm_file);
+		*offsetp = my_offset;
+		break;
+	}
+
+	*spu_bin_dcookie = fast_get_dcookie(&vma->vm_file->f_path);
+	pr_debug("got dcookie for %pD\n", vma->vm_file);
+
+	up_read(&mm->mmap_sem);
+
+out:
+	return app_cookie;
+
+fail_no_image_cookie:
+	up_read(&mm->mmap_sem);
+
+	printk(KERN_ERR "SPU_PROF: "
+		"%s, line %d: Cannot find dcookie for SPU binary\n",
+		__func__, __LINE__);
+	goto out;
+}
+
+
+
+/* This function finds or creates cached context information for the
+ * passed SPU and records SPU context information into the OProfile
+ * event buffer.
+ */
+static int process_context_switch(struct spu *spu, unsigned long objectId)
+{
+	unsigned long flags;
+	int retval;
+	unsigned int offset = 0;
+	unsigned long spu_cookie = 0, app_dcookie;
+
+	retval = prepare_cached_spu_info(spu, objectId);
+	if (retval)
+		goto out;
+
+	/* Get dcookie first because a mutex_lock is taken in that
+	 * code path, so interrupts must not be disabled.
+	 */
+	app_dcookie = get_exec_dcookie_and_offset(spu, &offset, &spu_cookie, objectId);
+	if (!app_dcookie || !spu_cookie) {
+		retval  = -ENOENT;
+		goto out;
+	}
+
+	/* Record context info in event buffer */
+	spin_lock_irqsave(&buffer_lock, flags);
+	spu_buff_add(ESCAPE_CODE, spu->number);
+	spu_buff_add(SPU_CTX_SWITCH_CODE, spu->number);
+	spu_buff_add(spu->number, spu->number);
+	spu_buff_add(spu->pid, spu->number);
+	spu_buff_add(spu->tgid, spu->number);
+	spu_buff_add(app_dcookie, spu->number);
+	spu_buff_add(spu_cookie, spu->number);
+	spu_buff_add(offset, spu->number);
+
+	/* Set flag to indicate SPU PC data can now be written out.  If
+	 * the SPU program counter data is seen before an SPU context
+	 * record is seen, the postprocessing will fail.
+	 */
+	spu_buff[spu->number].ctx_sw_seen = 1;
+
+	spin_unlock_irqrestore(&buffer_lock, flags);
+	smp_wmb();	/* insure spu event buffer updates are written */
+			/* don't want entries intermingled... */
+out:
+	return retval;
+}
+
+/*
+ * This function is invoked on either a bind_context or unbind_context.
+ * If called for an unbind_context, the val arg is 0; otherwise,
+ * it is the object-id value for the spu context.
+ * The data arg is of type 'struct spu *'.
+ */
+static int spu_active_notify(struct notifier_block *self, unsigned long val,
+				void *data)
+{
+	int retval;
+	unsigned long flags;
+	struct spu *the_spu = data;
+
+	pr_debug("SPU event notification arrived\n");
+	if (!val) {
+		spin_lock_irqsave(&cache_lock, flags);
+		retval = release_cached_info(the_spu->number);
+		spin_unlock_irqrestore(&cache_lock, flags);
+	} else {
+		retval = process_context_switch(the_spu, val);
+	}
+	return retval;
+}
+
+static struct notifier_block spu_active = {
+	.notifier_call = spu_active_notify,
+};
+
+static int number_of_online_nodes(void)
+{
+        u32 cpu; u32 tmp;
+        int nodes = 0;
+        for_each_online_cpu(cpu) {
+                tmp = cbe_cpu_to_node(cpu) + 1;
+                if (tmp > nodes)
+                        nodes++;
+        }
+        return nodes;
+}
+
+static int oprofile_spu_buff_create(void)
+{
+	int spu;
+
+	max_spu_buff = oprofile_get_cpu_buffer_size();
+
+	for (spu = 0; spu < num_spu_nodes; spu++) {
+		/* create circular buffers to store the data in.
+		 * use locks to manage accessing the buffers
+		 */
+		spu_buff[spu].head = 0;
+		spu_buff[spu].tail = 0;
+
+		/*
+		 * Create a buffer for each SPU.  Can't reliably
+		 * create a single buffer for all spus due to not
+		 * enough contiguous kernel memory.
+		 */
+
+		spu_buff[spu].buff = kzalloc((max_spu_buff
+					      * sizeof(unsigned long)),
+					     GFP_KERNEL);
+
+		if (!spu_buff[spu].buff) {
+			printk(KERN_ERR "SPU_PROF: "
+			       "%s, line %d:  oprofile_spu_buff_create "
+		       "failed to allocate spu buffer %d.\n",
+			       __func__, __LINE__, spu);
+
+			/* release the spu buffers that have been allocated */
+			while (spu >= 0) {
+				kfree(spu_buff[spu].buff);
+				spu_buff[spu].buff = 0;
+				spu--;
+			}
+			return -ENOMEM;
+		}
+	}
+	return 0;
+}
+
+/* The main purpose of this function is to synchronize
+ * OProfile with SPUFS by registering to be notified of
+ * SPU task switches.
+ *
+ * NOTE: When profiling SPUs, we must ensure that only
+ * spu_sync_start is invoked and not the generic sync_start
+ * in drivers/oprofile/oprof.c.	 A return value of
+ * SKIP_GENERIC_SYNC or SYNC_START_ERROR will
+ * accomplish this.
+ */
+int spu_sync_start(void)
+{
+	int spu;
+	int ret = SKIP_GENERIC_SYNC;
+	int register_ret;
+	unsigned long flags = 0;
+
+	spu_prof_num_nodes = number_of_online_nodes();
+	num_spu_nodes = spu_prof_num_nodes * 8;
+	INIT_DELAYED_WORK(&spu_work, wq_sync_spu_buff);
+
+	/* create buffer for storing the SPU data to put in
+	 * the kernel buffer.
+	 */
+	ret = oprofile_spu_buff_create();
+	if (ret)
+		goto out;
+
+	spin_lock_irqsave(&buffer_lock, flags);
+	for (spu = 0; spu < num_spu_nodes; spu++) {
+		spu_buff_add(ESCAPE_CODE, spu);
+		spu_buff_add(SPU_PROFILING_CODE, spu);
+		spu_buff_add(num_spu_nodes, spu);
+	}
+	spin_unlock_irqrestore(&buffer_lock, flags);
+
+	for (spu = 0; spu < num_spu_nodes; spu++) {
+		spu_buff[spu].ctx_sw_seen = 0;
+		spu_buff[spu].last_guard_val = 0;
+	}
+
+	/* Register for SPU events  */
+	register_ret = spu_switch_event_register(&spu_active);
+	if (register_ret) {
+		ret = SYNC_START_ERROR;
+		goto out;
+	}
+
+	pr_debug("spu_sync_start -- running.\n");
+out:
+	return ret;
+}
+
+/* Record SPU program counter samples to the oprofile event buffer. */
+void spu_sync_buffer(int spu_num, unsigned int *samples,
+		     int num_samples)
+{
+	unsigned long long file_offset;
+	unsigned long flags;
+	int i;
+	struct vma_to_fileoffset_map *map;
+	struct spu *the_spu;
+	unsigned long long spu_num_ll = spu_num;
+	unsigned long long spu_num_shifted = spu_num_ll << 32;
+	struct cached_info *c_info;
+
+	/* We need to obtain the cache_lock here because it's
+	 * possible that after getting the cached_info, the SPU job
+	 * corresponding to this cached_info may end, thus resulting
+	 * in the destruction of the cached_info.
+	 */
+	spin_lock_irqsave(&cache_lock, flags);
+	c_info = get_cached_info(NULL, spu_num);
+	if (!c_info) {
+		/* This legitimately happens when the SPU task ends before all
+		 * samples are recorded.
+		 * No big deal -- so we just drop a few samples.
+		 */
+		pr_debug("SPU_PROF: No cached SPU contex "
+			  "for SPU #%d. Dropping samples.\n", spu_num);
+		goto out;
+	}
+
+	map = c_info->map;
+	the_spu = c_info->the_spu;
+	spin_lock(&buffer_lock);
+	for (i = 0; i < num_samples; i++) {
+		unsigned int sample = *(samples+i);
+		int grd_val = 0;
+		file_offset = 0;
+		if (sample == 0)
+			continue;
+		file_offset = vma_map_lookup( map, sample, the_spu, &grd_val);
+
+		/* If overlays are used by this SPU application, the guard
+		 * value is non-zero, indicating which overlay section is in
+		 * use.	 We need to discard samples taken during the time
+		 * period which an overlay occurs (i.e., guard value changes).
+		 */
+		if (grd_val && grd_val != spu_buff[spu_num].last_guard_val) {
+			spu_buff[spu_num].last_guard_val = grd_val;
+			/* Drop the rest of the samples. */
+			break;
+		}
+
+		/* We must ensure that the SPU context switch has been written
+		 * out before samples for the SPU.  Otherwise, the SPU context
+		 * information is not available and the postprocessing of the
+		 * SPU PC will fail with no available anonymous map information.
+		 */
+		if (spu_buff[spu_num].ctx_sw_seen)
+			spu_buff_add((file_offset | spu_num_shifted),
+					 spu_num);
+	}
+	spin_unlock(&buffer_lock);
+out:
+	spin_unlock_irqrestore(&cache_lock, flags);
+}
+
+
+int spu_sync_stop(void)
+{
+	unsigned long flags = 0;
+	int ret;
+	int k;
+
+	ret = spu_switch_event_unregister(&spu_active);
+
+	if (ret)
+		printk(KERN_ERR "SPU_PROF: "
+		       "%s, line %d: spu_switch_event_unregister "	\
+		       "returned %d\n",
+		       __func__, __LINE__, ret);
+
+	/* flush any remaining data in the per SPU buffers */
+	sync_spu_buff();
+
+	spin_lock_irqsave(&cache_lock, flags);
+	ret = release_cached_info(RELEASE_ALL);
+	spin_unlock_irqrestore(&cache_lock, flags);
+
+	/* remove scheduled work queue item rather then waiting
+	 * for every queued entry to execute.  Then flush pending
+	 * system wide buffer to event buffer.
+	 */
+	cancel_delayed_work(&spu_work);
+
+	for (k = 0; k < num_spu_nodes; k++) {
+		spu_buff[k].ctx_sw_seen = 0;
+
+		/*
+		 * spu_sys_buff will be null if there was a problem
+		 * allocating the buffer.  Only delete if it exists.
+		 */
+		kfree(spu_buff[k].buff);
+		spu_buff[k].buff = 0;
+	}
+	pr_debug("spu_sync_stop -- done.\n");
+	return ret;
+}
+
diff --git a/arch/powerpc/oprofile/cell/vma_map.c b/arch/powerpc/oprofile/cell/vma_map.c
new file mode 100644
index 000000000..c579b1684
--- /dev/null
+++ b/arch/powerpc/oprofile/cell/vma_map.c
@@ -0,0 +1,283 @@
+/*
+ * Cell Broadband Engine OProfile Support
+ *
+ * (C) Copyright IBM Corporation 2006
+ *
+ * Author: Maynard Johnson <maynardj@us.ibm.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.
+ */
+
+/* The code in this source file is responsible for generating
+ * vma-to-fileOffset maps for both overlay and non-overlay SPU
+ * applications.
+ */
+
+#include <linux/mm.h>
+#include <linux/string.h>
+#include <linux/uaccess.h>
+#include <linux/elf.h>
+#include <linux/slab.h>
+#include "pr_util.h"
+
+
+void vma_map_free(struct vma_to_fileoffset_map *map)
+{
+	while (map) {
+		struct vma_to_fileoffset_map *next = map->next;
+		kfree(map);
+		map = next;
+	}
+}
+
+unsigned int
+vma_map_lookup(struct vma_to_fileoffset_map *map, unsigned int vma,
+	       const struct spu *aSpu, int *grd_val)
+{
+	/*
+	 * Default the offset to the physical address + a flag value.
+	 * Addresses of dynamically generated code can't be found in the vma
+	 * map.  For those addresses the flagged value will be sent on to
+	 * the user space tools so they can be reported rather than just
+	 * thrown away.
+	 */
+	u32 offset = 0x10000000 + vma;
+	u32 ovly_grd;
+
+	for (; map; map = map->next) {
+		if (vma < map->vma || vma >= map->vma + map->size)
+			continue;
+
+		if (map->guard_ptr) {
+			ovly_grd = *(u32 *)(aSpu->local_store + map->guard_ptr);
+			if (ovly_grd != map->guard_val)
+				continue;
+			*grd_val = ovly_grd;
+		}
+		offset = vma - map->vma + map->offset;
+		break;
+	}
+
+	return offset;
+}
+
+static struct vma_to_fileoffset_map *
+vma_map_add(struct vma_to_fileoffset_map *map, unsigned int vma,
+	    unsigned int size, unsigned int offset, unsigned int guard_ptr,
+	    unsigned int guard_val)
+{
+	struct vma_to_fileoffset_map *new =
+		kzalloc(sizeof(struct vma_to_fileoffset_map), GFP_KERNEL);
+	if (!new) {
+		printk(KERN_ERR "SPU_PROF: %s, line %d: malloc failed\n",
+		       __func__, __LINE__);
+		vma_map_free(map);
+		return NULL;
+	}
+
+	new->next = map;
+	new->vma = vma;
+	new->size = size;
+	new->offset = offset;
+	new->guard_ptr = guard_ptr;
+	new->guard_val = guard_val;
+
+	return new;
+}
+
+
+/* Parse SPE ELF header and generate a list of vma_maps.
+ * A pointer to the first vma_map in the generated list
+ * of vma_maps is returned.  */
+struct vma_to_fileoffset_map *create_vma_map(const struct spu *aSpu,
+					     unsigned long __spu_elf_start)
+{
+	static const unsigned char expected[EI_PAD] = {
+		[EI_MAG0] = ELFMAG0,
+		[EI_MAG1] = ELFMAG1,
+		[EI_MAG2] = ELFMAG2,
+		[EI_MAG3] = ELFMAG3,
+		[EI_CLASS] = ELFCLASS32,
+		[EI_DATA] = ELFDATA2MSB,
+		[EI_VERSION] = EV_CURRENT,
+		[EI_OSABI] = ELFOSABI_NONE
+	};
+
+	int grd_val;
+	struct vma_to_fileoffset_map *map = NULL;
+	void __user *spu_elf_start = (void __user *)__spu_elf_start;
+	struct spu_overlay_info ovly;
+	unsigned int overlay_tbl_offset = -1;
+	Elf32_Phdr __user *phdr_start;
+	Elf32_Shdr __user *shdr_start;
+	Elf32_Ehdr ehdr;
+	Elf32_Phdr phdr;
+	Elf32_Shdr shdr, shdr_str;
+	Elf32_Sym sym;
+	int i, j;
+	char name[32];
+
+	unsigned int ovly_table_sym = 0;
+	unsigned int ovly_buf_table_sym = 0;
+	unsigned int ovly_table_end_sym = 0;
+	unsigned int ovly_buf_table_end_sym = 0;
+	struct spu_overlay_info __user *ovly_table;
+	unsigned int n_ovlys;
+
+	/* Get and validate ELF header.	 */
+
+	if (copy_from_user(&ehdr, spu_elf_start, sizeof (ehdr)))
+		goto fail;
+
+	if (memcmp(ehdr.e_ident, expected, EI_PAD) != 0) {
+		printk(KERN_ERR "SPU_PROF: "
+		       "%s, line %d: Unexpected e_ident parsing SPU ELF\n",
+		       __func__, __LINE__);
+		goto fail;
+	}
+	if (ehdr.e_machine != EM_SPU) {
+		printk(KERN_ERR "SPU_PROF: "
+		       "%s, line %d: Unexpected e_machine parsing SPU ELF\n",
+		       __func__,  __LINE__);
+		goto fail;
+	}
+	if (ehdr.e_type != ET_EXEC) {
+		printk(KERN_ERR "SPU_PROF: "
+		       "%s, line %d: Unexpected e_type parsing SPU ELF\n",
+		       __func__, __LINE__);
+		goto fail;
+	}
+	phdr_start = spu_elf_start + ehdr.e_phoff;
+	shdr_start = spu_elf_start + ehdr.e_shoff;
+
+	/* Traverse program headers.  */
+	for (i = 0; i < ehdr.e_phnum; i++) {
+		if (copy_from_user(&phdr, phdr_start + i, sizeof(phdr)))
+			goto fail;
+
+		if (phdr.p_type != PT_LOAD)
+			continue;
+		if (phdr.p_flags & (1 << 27))
+			continue;
+
+		map = vma_map_add(map, phdr.p_vaddr, phdr.p_memsz,
+				  phdr.p_offset, 0, 0);
+		if (!map)
+			goto fail;
+	}
+
+	pr_debug("SPU_PROF: Created non-overlay maps\n");
+	/* Traverse section table and search for overlay-related symbols.  */
+	for (i = 0; i < ehdr.e_shnum; i++) {
+		if (copy_from_user(&shdr, shdr_start + i, sizeof(shdr)))
+			goto fail;
+
+		if (shdr.sh_type != SHT_SYMTAB)
+			continue;
+		if (shdr.sh_entsize != sizeof (sym))
+			continue;
+
+		if (copy_from_user(&shdr_str,
+				   shdr_start + shdr.sh_link,
+				   sizeof(shdr)))
+			goto fail;
+
+		if (shdr_str.sh_type != SHT_STRTAB)
+			goto fail;
+
+		for (j = 0; j < shdr.sh_size / sizeof (sym); j++) {
+			if (copy_from_user(&sym, spu_elf_start +
+						 shdr.sh_offset +
+						 j * sizeof (sym),
+					   sizeof (sym)))
+				goto fail;
+
+			if (copy_from_user(name, 
+					   spu_elf_start + shdr_str.sh_offset +
+					   sym.st_name,
+					   20))
+				goto fail;
+
+			if (memcmp(name, "_ovly_table", 12) == 0)
+				ovly_table_sym = sym.st_value;
+			if (memcmp(name, "_ovly_buf_table", 16) == 0)
+				ovly_buf_table_sym = sym.st_value;
+			if (memcmp(name, "_ovly_table_end", 16) == 0)
+				ovly_table_end_sym = sym.st_value;
+			if (memcmp(name, "_ovly_buf_table_end", 20) == 0)
+				ovly_buf_table_end_sym = sym.st_value;
+		}
+	}
+
+	/* If we don't have overlays, we're done.  */
+	if (ovly_table_sym == 0 || ovly_buf_table_sym == 0
+	    || ovly_table_end_sym == 0 || ovly_buf_table_end_sym == 0) {
+		pr_debug("SPU_PROF: No overlay table found\n");
+		goto out;
+	} else {
+		pr_debug("SPU_PROF: Overlay table found\n");
+	}
+
+	/* The _ovly_table symbol represents a table with one entry
+	 * per overlay section.	 The _ovly_buf_table symbol represents
+	 * a table with one entry per overlay region.
+	 * The struct spu_overlay_info gives the structure of the _ovly_table
+	 * entries.  The structure of _ovly_table_buf is simply one
+	 * u32 word per entry.
+	 */
+	overlay_tbl_offset = vma_map_lookup(map, ovly_table_sym,
+					    aSpu, &grd_val);
+	if (overlay_tbl_offset > 0x10000000) {
+		printk(KERN_ERR "SPU_PROF: "
+		       "%s, line %d: Error finding SPU overlay table\n",
+		       __func__, __LINE__);
+		goto fail;
+	}
+	ovly_table = spu_elf_start + overlay_tbl_offset;
+
+	n_ovlys = (ovly_table_end_sym -
+		   ovly_table_sym) / sizeof (ovly);
+
+	/* Traverse overlay table.  */
+	for (i = 0; i < n_ovlys; i++) {
+		if (copy_from_user(&ovly, ovly_table + i, sizeof (ovly)))
+			goto fail;
+
+		/* The ovly.vma/size/offset arguments are analogous to the same
+		 * arguments used above for non-overlay maps.  The final two
+		 * args are referred to as the guard pointer and the guard
+		 * value.
+		 * The guard pointer is an entry in the _ovly_buf_table,
+		 * computed using ovly.buf as the index into the table.	 Since
+		 * ovly.buf values begin at '1' to reference the first (or 0th)
+		 * entry in the _ovly_buf_table, the computation subtracts 1
+		 * from ovly.buf.
+		 * The guard value is stored in the _ovly_buf_table entry and
+		 * is an index (starting at 1) back to the _ovly_table entry
+		 * that is pointing at this _ovly_buf_table entry.  So, for
+		 * example, for an overlay scenario with one overlay segment
+		 * and two overlay sections:
+		 *	- Section 1 points to the first entry of the
+		 *	  _ovly_buf_table, which contains a guard value
+		 *	  of '1', referencing the first (index=0) entry of
+		 *	  _ovly_table.
+		 *	- Section 2 points to the second entry of the
+		 *	  _ovly_buf_table, which contains a guard value
+		 *	  of '2', referencing the second (index=1) entry of
+		 *	  _ovly_table.
+		 */
+		map = vma_map_add(map, ovly.vma, ovly.size, ovly.offset,
+				  ovly_buf_table_sym + (ovly.buf-1) * 4, i+1);
+		if (!map)
+			goto fail;
+	}
+	goto out;
+
+ fail:
+	map = NULL;
+ out:
+	return map;
+}