diff --git a/arch/powerpc/platforms/cell/spu_base.c b/arch/powerpc/platforms/cell/spu_base.c
index cadcc64a8657..174bd9f911db 100644
--- a/arch/powerpc/platforms/cell/spu_base.c
+++ b/arch/powerpc/platforms/cell/spu_base.c
@@ -183,7 +183,7 @@ static int __spu_trap_data_seg(struct spu *spu, unsigned long ea)
 		spu->slb_replace = 0;
 
 	spu_restart_dma(spu);
-
+	spu->stats.slb_flt++;
 	return 0;
 }
 
@@ -332,6 +332,7 @@ spu_irq_class_2(int irq, void *data)
 	if (stat & 0x10) /* SPU mailbox threshold */
 		spu->wbox_callback(spu);
 
+	spu->stats.class2_intr++;
 	return stat ? IRQ_HANDLED : IRQ_NONE;
 }
 
diff --git a/arch/powerpc/platforms/cell/spufs/context.c b/arch/powerpc/platforms/cell/spufs/context.c
index f623d963fdc7..6d7bd60f5380 100644
--- a/arch/powerpc/platforms/cell/spufs/context.c
+++ b/arch/powerpc/platforms/cell/spufs/context.c
@@ -59,6 +59,8 @@ struct spu_context *alloc_spu_context(struct spu_gang *gang)
 		spu_gang_add_ctx(gang, ctx);
 	ctx->cpus_allowed = current->cpus_allowed;
 	spu_set_timeslice(ctx);
+	ctx->stats.execution_state = SPUCTX_UTIL_USER;
+	ctx->stats.tstamp = jiffies;
 
 	atomic_inc(&nr_spu_contexts);
 	goto out;
diff --git a/arch/powerpc/platforms/cell/spufs/fault.c b/arch/powerpc/platforms/cell/spufs/fault.c
index 0f75c07e29d8..3a9e49a24ec0 100644
--- a/arch/powerpc/platforms/cell/spufs/fault.c
+++ b/arch/powerpc/platforms/cell/spufs/fault.c
@@ -33,7 +33,8 @@
  * function. Currently, there are a few corner cases that we haven't had
  * to handle fortunately.
  */
-static int spu_handle_mm_fault(struct mm_struct *mm, unsigned long ea, unsigned long dsisr)
+static int spu_handle_mm_fault(struct mm_struct *mm, unsigned long ea,
+		unsigned long dsisr, unsigned *flt)
 {
 	struct vm_area_struct *vma;
 	unsigned long is_write;
@@ -73,7 +74,8 @@ static int spu_handle_mm_fault(struct mm_struct *mm, unsigned long ea, unsigned
 			goto bad_area;
 	}
 	ret = 0;
-	switch (handle_mm_fault(mm, vma, ea, is_write)) {
+	*flt = handle_mm_fault(mm, vma, ea, is_write);
+	switch (*flt) {
 	case VM_FAULT_MINOR:
 		current->min_flt++;
 		break;
@@ -153,6 +155,7 @@ int spufs_handle_class1(struct spu_context *ctx)
 {
 	u64 ea, dsisr, access;
 	unsigned long flags;
+	unsigned flt = 0;
 	int ret;
 
 	/*
@@ -178,9 +181,13 @@ int spufs_handle_class1(struct spu_context *ctx)
 	if (!(dsisr & (MFC_DSISR_PTE_NOT_FOUND | MFC_DSISR_ACCESS_DENIED)))
 		return 0;
 
+	spuctx_switch_state(ctx, SPUCTX_UTIL_IOWAIT);
+
 	pr_debug("ctx %p: ea %016lx, dsisr %016lx state %d\n", ctx, ea,
 		dsisr, ctx->state);
 
+	ctx->stats.hash_flt++;
+
 	/* we must not hold the lock when entering spu_handle_mm_fault */
 	spu_release(ctx);
 
@@ -192,7 +199,7 @@ int spufs_handle_class1(struct spu_context *ctx)
 
 	/* hashing failed, so try the actual fault handler */
 	if (ret)
-		ret = spu_handle_mm_fault(current->mm, ea, dsisr);
+		ret = spu_handle_mm_fault(current->mm, ea, dsisr, &flt);
 
 	spu_acquire(ctx);
 	/*
@@ -201,11 +208,17 @@ int spufs_handle_class1(struct spu_context *ctx)
 	 * In case of unhandled error report the problem to user space.
 	 */
 	if (!ret) {
+		if (flt == VM_FAULT_MINOR)
+			ctx->stats.min_flt++;
+		else
+			ctx->stats.maj_flt++;
+
 		if (ctx->spu)
 			ctx->ops->restart_dma(ctx);
 	} else
 		spufs_handle_dma_error(ctx, ea, SPE_EVENT_SPE_DATA_STORAGE);
 
+	spuctx_switch_state(ctx, SPUCTX_UTIL_SYSTEM);
 	return ret;
 }
 EXPORT_SYMBOL_GPL(spufs_handle_class1);
diff --git a/arch/powerpc/platforms/cell/spufs/file.c b/arch/powerpc/platforms/cell/spufs/file.c
index 2bb51ca51a6c..30f7b077f347 100644
--- a/arch/powerpc/platforms/cell/spufs/file.c
+++ b/arch/powerpc/platforms/cell/spufs/file.c
@@ -2059,6 +2059,83 @@ static const struct file_operations spufs_tid_fops = {
 	.release	= single_release,
 };
 
+static const char *ctx_state_names[] = {
+	"user", "system", "iowait", "loaded"
+};
+
+static unsigned long long spufs_acct_time(struct spu_context *ctx,
+		enum spuctx_execution_state state)
+{
+	unsigned long time = ctx->stats.times[state];
+
+	if (ctx->stats.execution_state == state)
+		time += jiffies - ctx->stats.tstamp;
+
+	return jiffies_to_msecs(time);
+}
+
+static unsigned long long spufs_slb_flts(struct spu_context *ctx)
+{
+	unsigned long long slb_flts = ctx->stats.slb_flt;
+
+	if (ctx->state == SPU_STATE_RUNNABLE) {
+		slb_flts += (ctx->spu->stats.slb_flt -
+			     ctx->stats.slb_flt_base);
+	}
+
+	return slb_flts;
+}
+
+static unsigned long long spufs_class2_intrs(struct spu_context *ctx)
+{
+	unsigned long long class2_intrs = ctx->stats.class2_intr;
+
+	if (ctx->state == SPU_STATE_RUNNABLE) {
+		class2_intrs += (ctx->spu->stats.class2_intr -
+				 ctx->stats.class2_intr_base);
+	}
+
+	return class2_intrs;
+}
+
+
+static int spufs_show_stat(struct seq_file *s, void *private)
+{
+	struct spu_context *ctx = s->private;
+
+	spu_acquire(ctx);
+	seq_printf(s, "%s %llu %llu %llu %llu "
+		      "%llu %llu %llu %llu %llu %llu %llu %llu\n",
+		ctx_state_names[ctx->stats.execution_state],
+		spufs_acct_time(ctx, SPUCTX_UTIL_USER),
+		spufs_acct_time(ctx, SPUCTX_UTIL_SYSTEM),
+		spufs_acct_time(ctx, SPUCTX_UTIL_IOWAIT),
+		spufs_acct_time(ctx, SPUCTX_UTIL_LOADED),
+		ctx->stats.vol_ctx_switch,
+		ctx->stats.invol_ctx_switch,
+		spufs_slb_flts(ctx),
+		ctx->stats.hash_flt,
+		ctx->stats.min_flt,
+		ctx->stats.maj_flt,
+		spufs_class2_intrs(ctx),
+		ctx->stats.libassist);
+	spu_release(ctx);
+	return 0;
+}
+
+static int spufs_stat_open(struct inode *inode, struct file *file)
+{
+	return single_open(file, spufs_show_stat, SPUFS_I(inode)->i_ctx);
+}
+
+static const struct file_operations spufs_stat_fops = {
+	.open		= spufs_stat_open,
+	.read		= seq_read,
+	.llseek		= seq_lseek,
+	.release	= single_release,
+};
+
+
 struct tree_descr spufs_dir_contents[] = {
 	{ "capabilities", &spufs_caps_fops, 0444, },
 	{ "mem",  &spufs_mem_fops,  0666, },
@@ -2093,6 +2170,7 @@ struct tree_descr spufs_dir_contents[] = {
 	{ "dma_info", &spufs_dma_info_fops, 0444, },
 	{ "proxydma_info", &spufs_proxydma_info_fops, 0444, },
 	{ "tid", &spufs_tid_fops, 0444, },
+	{ "stat", &spufs_stat_fops, 0444, },
 	{},
 };
 
@@ -2117,6 +2195,7 @@ struct tree_descr spufs_dir_nosched_contents[] = {
 	{ "phys-id", &spufs_id_ops, 0666, },
 	{ "object-id", &spufs_object_id_ops, 0666, },
 	{ "tid", &spufs_tid_fops, 0444, },
+	{ "stat", &spufs_stat_fops, 0444, },
 	{},
 };
 
diff --git a/arch/powerpc/platforms/cell/spufs/run.c b/arch/powerpc/platforms/cell/spufs/run.c
index 4e0db6ae0d5e..c69f63dd5f0a 100644
--- a/arch/powerpc/platforms/cell/spufs/run.c
+++ b/arch/powerpc/platforms/cell/spufs/run.c
@@ -351,6 +351,10 @@ long spufs_run_spu(struct file *file, struct spu_context *ctx,
 				      SPU_STATUS_STOPPED_BY_HALT |
 				       SPU_STATUS_SINGLE_STEP)));
 
+	if ((status & SPU_STATUS_STOPPED_BY_STOP) &&
+	    ((status >> SPU_STOP_STATUS_SHIFT) & 0x2100))
+		ctx->stats.libassist++;
+
 	ctx->ops->master_stop(ctx);
 	ret = spu_run_fini(ctx, npc, &status);
 	spu_yield(ctx);
diff --git a/arch/powerpc/platforms/cell/spufs/sched.c b/arch/powerpc/platforms/cell/spufs/sched.c
index 9fc09306c9ae..bb16c22360d5 100644
--- a/arch/powerpc/platforms/cell/spufs/sched.c
+++ b/arch/powerpc/platforms/cell/spufs/sched.c
@@ -229,6 +229,10 @@ static void spu_bind_context(struct spu *spu, struct spu_context *ctx)
 {
 	pr_debug("%s: pid=%d SPU=%d NODE=%d\n", __FUNCTION__, current->pid,
 		 spu->number, spu->node);
+
+	ctx->stats.slb_flt_base = spu->stats.slb_flt;
+	ctx->stats.class2_intr_base = spu->stats.class2_intr;
+
 	spu->ctx = ctx;
 	spu->flags = 0;
 	ctx->spu = spu;
@@ -275,6 +279,11 @@ static void spu_unbind_context(struct spu *spu, struct spu_context *ctx)
 	ctx->spu = NULL;
 	spu->flags = 0;
 	spu->ctx = NULL;
+
+	ctx->stats.slb_flt +=
+		(spu->stats.slb_flt - ctx->stats.slb_flt_base);
+	ctx->stats.class2_intr +=
+		(spu->stats.class2_intr - ctx->stats.class2_intr_base);
 }
 
 /**
@@ -400,6 +409,7 @@ static struct spu *find_victim(struct spu_context *ctx)
 			}
 			spu_remove_from_active_list(spu);
 			spu_unbind_context(spu, victim);
+			victim->stats.invol_ctx_switch++;
 			mutex_unlock(&victim->state_mutex);
 			/*
 			 * We need to break out of the wait loop in spu_run
@@ -425,6 +435,7 @@ static struct spu *find_victim(struct spu_context *ctx)
  */
 int spu_activate(struct spu_context *ctx, unsigned long flags)
 {
+	spuctx_switch_state(ctx, SPUCTX_UTIL_SYSTEM);
 
 	if (ctx->spu)
 		return 0;
@@ -492,6 +503,7 @@ static int __spu_deactivate(struct spu_context *ctx, int force, int max_prio)
 		if (new || force) {
 			spu_remove_from_active_list(spu);
 			spu_unbind_context(spu, ctx);
+			ctx->stats.vol_ctx_switch++;
 			spu_free(spu);
 			if (new)
 				wake_up(&new->stop_wq);
@@ -521,6 +533,7 @@ void spu_deactivate(struct spu_context *ctx)
 	}
 
 	__spu_deactivate(ctx, 1, MAX_PRIO);
+	spuctx_switch_state(ctx, SPUCTX_UTIL_USER);
 }
 
 /**
@@ -535,7 +548,10 @@ void spu_yield(struct spu_context *ctx)
 {
 	if (!(ctx->flags & SPU_CREATE_NOSCHED)) {
 		mutex_lock(&ctx->state_mutex);
-		__spu_deactivate(ctx, 0, MAX_PRIO);
+		if (__spu_deactivate(ctx, 0, MAX_PRIO))
+			spuctx_switch_state(ctx, SPUCTX_UTIL_USER);
+		else
+			spuctx_switch_state(ctx, SPUCTX_UTIL_LOADED);
 		mutex_unlock(&ctx->state_mutex);
 	}
 }
@@ -564,6 +580,7 @@ static void spusched_tick(struct spu_context *ctx)
 
 			__spu_remove_from_active_list(spu);
 			spu_unbind_context(spu, ctx);
+			ctx->stats.invol_ctx_switch++;
 			spu_free(spu);
 			wake_up(&new->stop_wq);
 			/*
diff --git a/arch/powerpc/platforms/cell/spufs/spufs.h b/arch/powerpc/platforms/cell/spufs/spufs.h
index 7f5d0b2fdea3..cd2b54f6e378 100644
--- a/arch/powerpc/platforms/cell/spufs/spufs.h
+++ b/arch/powerpc/platforms/cell/spufs/spufs.h
@@ -40,6 +40,19 @@ enum {
 struct spu_context_ops;
 struct spu_gang;
 
+/*
+ * This is the state for spu utilization reporting to userspace.
+ * Because this state is visible to userspace it must never change and needs
+ * to be kept strictly separate from any internal state kept by the kernel.
+ */
+enum spuctx_execution_state {
+	SPUCTX_UTIL_USER = 0,
+	SPUCTX_UTIL_SYSTEM,
+	SPUCTX_UTIL_IOWAIT,
+	SPUCTX_UTIL_LOADED,
+	SPUCTX_UTIL_MAX
+};
+
 struct spu_context {
 	struct spu *spu;		  /* pointer to a physical SPU */
 	struct spu_state csa;		  /* SPU context save area. */
@@ -87,6 +100,24 @@ struct spu_context {
 	cpumask_t cpus_allowed;
 	int policy;
 	int prio;
+
+	/* statistics */
+	struct {
+		/* updates protected by ctx->state_mutex */
+		enum spuctx_execution_state execution_state;
+		unsigned long tstamp;		/* time of last ctx switch */
+		unsigned long times[SPUCTX_UTIL_MAX];
+		unsigned long long vol_ctx_switch;
+		unsigned long long invol_ctx_switch;
+		unsigned long long min_flt;
+		unsigned long long maj_flt;
+		unsigned long long hash_flt;
+		unsigned long long slb_flt;
+		unsigned long long slb_flt_base; /* # at last ctx switch */
+		unsigned long long class2_intr;
+		unsigned long long class2_intr_base; /* # at last ctx switch */
+		unsigned long long libassist;
+	} stats;
 };
 
 struct spu_gang {
@@ -256,4 +287,24 @@ struct spufs_coredump_reader {
 extern struct spufs_coredump_reader spufs_coredump_read[];
 extern int spufs_coredump_num_notes;
 
+/*
+ * This function is a little bit too large for an inline, but
+ * as fault.c is built into the kernel we can't move it out of
+ * line.
+ */
+static inline void spuctx_switch_state(struct spu_context *ctx,
+		enum spuctx_execution_state new_state)
+{
+	WARN_ON(!mutex_is_locked(&ctx->state_mutex));
+
+	if (ctx->stats.execution_state != new_state) {
+		unsigned long curtime = jiffies;
+
+		ctx->stats.times[ctx->stats.execution_state] +=
+				 curtime - ctx->stats.tstamp;
+		ctx->stats.tstamp = curtime;
+		ctx->stats.execution_state = new_state;
+	}
+}
+
 #endif
diff --git a/include/asm-powerpc/spu.h b/include/asm-powerpc/spu.h
index 5f894b61e2d4..5957fcdda04c 100644
--- a/include/asm-powerpc/spu.h
+++ b/include/asm-powerpc/spu.h
@@ -156,6 +156,12 @@ struct spu {
 	u64 shadow_int_mask_RW[3];
 
 	struct sys_device sysdev;
+
+	struct {
+		/* protected by interrupt reentrancy */
+		unsigned long long slb_flt;
+		unsigned long long class2_intr;
+	} stats;
 };
 
 struct spu *spu_alloc(void);