Merge branch 'for-2.6.23' of master.kernel.org:/pub/scm/linux/kernel/git/arnd/cell-2.6

* 'for-2.6.23' of master.kernel.org:/pub/scm/linux/kernel/git/arnd/cell-2.6: (37 commits)
  [CELL] spufs: rework list management and associated locking
  [CELL] oprofile: add support to OProfile for profiling CELL BE SPUs
  [CELL] oprofile: enable SPU switch notification to detect currently active SPU tasks
  [CELL] spu_base: locking cleanup
  [CELL] cell: indexing of SPUs based on firmware vicinity properties
  [CELL] spufs: integration of SPE affinity with the scheduller
  [CELL] cell: add placement computation for scheduling of affinity contexts
  [CELL] spufs: extension of spu_create to support affinity definition
  [CELL] cell: add hardcoded spu vicinity information for QS20
  [CELL] cell: add vicinity information on spus
  [CELL] cell: add per BE structure with info about its SPUs
  [CELL] spufs: use find_first_bit() instead of sched_find_first_bit()
  [CELL] spufs: remove unused file argument from spufs_run_spu()
  [CELL] spufs: change decrementer restore timing
  [CELL] spufs: dont halt decrementer at restore step 47
  [CELL] spufs: limit saving MFC_CNTL bits
  [CELL] spufs: fix read and write for decr_status file
  [CELL] spufs: fix decr_status meanings
  [CELL] spufs: remove needless context save/restore code
  [CELL] spufs: fix array size of channel index
  ...

13 files changed, 932 insertions, 466 deletions

diff --git a/arch/powerpc/platforms/cell/spufs/context.c b/arch/powerpc/platforms/cell/spufs/context.c
index 6d7bd60f5380..6694f86d7000 100644
--- a/arch/powerpc/platforms/cell/spufs/context.c
+++ b/arch/powerpc/platforms/cell/spufs/context.c
@@ -22,6 +22,7 @@
 
 #include <linux/fs.h>
 #include <linux/mm.h>
+#include <linux/module.h>
 #include <linux/slab.h>
 #include <asm/atomic.h>
 #include <asm/spu.h>
@@ -55,12 +56,12 @@ struct spu_context *alloc_spu_context(struct spu_gang *gang)
 	ctx->ops = &spu_backing_ops;
 	ctx->owner = get_task_mm(current);
 	INIT_LIST_HEAD(&ctx->rq);
+	INIT_LIST_HEAD(&ctx->aff_list);
 	if (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;
+	ctx->stats.util_state = SPU_UTIL_IDLE_LOADED;
 
 	atomic_inc(&nr_spu_contexts);
 	goto out;
@@ -81,6 +82,8 @@ void destroy_spu_context(struct kref *kref)
 	spu_fini_csa(&ctx->csa);
 	if (ctx->gang)
 		spu_gang_remove_ctx(ctx->gang, ctx);
+	if (ctx->prof_priv_kref)
+		kref_put(ctx->prof_priv_kref, ctx->prof_priv_release);
 	BUG_ON(!list_empty(&ctx->rq));
 	atomic_dec(&nr_spu_contexts);
 	kfree(ctx);
@@ -166,6 +169,39 @@ int spu_acquire_runnable(struct spu_context *ctx, unsigned long flags)
 void spu_acquire_saved(struct spu_context *ctx)
 {
 	spu_acquire(ctx);
-	if (ctx->state != SPU_STATE_SAVED)
+	if (ctx->state != SPU_STATE_SAVED) {
+		set_bit(SPU_SCHED_WAS_ACTIVE, &ctx->sched_flags);
 		spu_deactivate(ctx);
+	}
+}
+
+/**
+ * spu_release_saved - unlock spu context and return it to the runqueue
+ * @ctx:	context to unlock
+ */
+void spu_release_saved(struct spu_context *ctx)
+{
+	BUG_ON(ctx->state != SPU_STATE_SAVED);
+
+	if (test_and_clear_bit(SPU_SCHED_WAS_ACTIVE, &ctx->sched_flags))
+		spu_activate(ctx, 0);
+
+	spu_release(ctx);
 }
+
+void spu_set_profile_private_kref(struct spu_context *ctx,
+				  struct kref *prof_info_kref,
+				  void ( * prof_info_release) (struct kref *kref))
+{
+	ctx->prof_priv_kref = prof_info_kref;
+	ctx->prof_priv_release = prof_info_release;
+}
+EXPORT_SYMBOL_GPL(spu_set_profile_private_kref);
+
+void *spu_get_profile_private_kref(struct spu_context *ctx)
+{
+	return ctx->prof_priv_kref;
+}
+EXPORT_SYMBOL_GPL(spu_get_profile_private_kref);
+
+
diff --git a/arch/powerpc/platforms/cell/spufs/coredump.c b/arch/powerpc/platforms/cell/spufs/coredump.c
index 5d9ad5a0307b..5e31799b1e3f 100644
--- a/arch/powerpc/platforms/cell/spufs/coredump.c
+++ b/arch/powerpc/platforms/cell/spufs/coredump.c
@@ -226,7 +226,7 @@ static void spufs_arch_write_notes(struct file *file)
 		spu_acquire_saved(ctx_info->ctx);
 		for (j = 0; j < spufs_coredump_num_notes; j++)
 			spufs_arch_write_note(ctx_info, j, file);
-		spu_release(ctx_info->ctx);
+		spu_release_saved(ctx_info->ctx);
 		list_del(&ctx_info->list);
 		kfree(ctx_info);
 	}
diff --git a/arch/powerpc/platforms/cell/spufs/fault.c b/arch/powerpc/platforms/cell/spufs/fault.c
index f53a07437472..917eab4be486 100644
--- a/arch/powerpc/platforms/cell/spufs/fault.c
+++ b/arch/powerpc/platforms/cell/spufs/fault.c
@@ -179,16 +179,14 @@ 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);
+	spuctx_switch_state(ctx, SPU_UTIL_IOWAIT);
 
 	pr_debug("ctx %p: ea %016lx, dsisr %016lx state %d\n", ctx, ea,
 		dsisr, ctx->state);
 
 	ctx->stats.hash_flt++;
-	if (ctx->state == SPU_STATE_RUNNABLE) {
+	if (ctx->state == SPU_STATE_RUNNABLE)
 		ctx->spu->stats.hash_flt++;
-		spu_switch_state(ctx->spu, SPU_UTIL_IOWAIT);
-	}
 
 	/* we must not hold the lock when entering spu_handle_mm_fault */
 	spu_release(ctx);
@@ -226,7 +224,7 @@ int spufs_handle_class1(struct spu_context *ctx)
 	} else
 		spufs_handle_dma_error(ctx, ea, SPE_EVENT_SPE_DATA_STORAGE);
 
-	spuctx_switch_state(ctx, SPUCTX_UTIL_SYSTEM);
+	spuctx_switch_state(ctx, SPU_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 c2814ea96af2..7de4e919687b 100644
--- a/arch/powerpc/platforms/cell/spufs/file.c
+++ b/arch/powerpc/platforms/cell/spufs/file.c
@@ -370,7 +370,7 @@ spufs_regs_read(struct file *file, char __user *buffer,
 
 	spu_acquire_saved(ctx);
 	ret = __spufs_regs_read(ctx, buffer, size, pos);
-	spu_release(ctx);
+	spu_release_saved(ctx);
 	return ret;
 }
 
@@ -392,7 +392,7 @@ spufs_regs_write(struct file *file, const char __user *buffer,
 	ret = copy_from_user(lscsa->gprs + *pos - size,
 			     buffer, size) ? -EFAULT : size;
 
-	spu_release(ctx);
+	spu_release_saved(ctx);
 	return ret;
 }
 
@@ -421,7 +421,7 @@ spufs_fpcr_read(struct file *file, char __user * buffer,
 
 	spu_acquire_saved(ctx);
 	ret = __spufs_fpcr_read(ctx, buffer, size, pos);
-	spu_release(ctx);
+	spu_release_saved(ctx);
 	return ret;
 }
 
@@ -443,7 +443,7 @@ spufs_fpcr_write(struct file *file, const char __user * buffer,
 	ret = copy_from_user((char *)&lscsa->fpcr + *pos - size,
 			     buffer, size) ? -EFAULT : size;
 
-	spu_release(ctx);
+	spu_release_saved(ctx);
 	return ret;
 }
 
@@ -868,7 +868,7 @@ static ssize_t spufs_signal1_read(struct file *file, char __user *buf,
 
 	spu_acquire_saved(ctx);
 	ret = __spufs_signal1_read(ctx, buf, len, pos);
-	spu_release(ctx);
+	spu_release_saved(ctx);
 
 	return ret;
 }
@@ -934,6 +934,13 @@ static const struct file_operations spufs_signal1_fops = {
 	.mmap = spufs_signal1_mmap,
 };
 
+static const struct file_operations spufs_signal1_nosched_fops = {
+	.open = spufs_signal1_open,
+	.release = spufs_signal1_release,
+	.write = spufs_signal1_write,
+	.mmap = spufs_signal1_mmap,
+};
+
 static int spufs_signal2_open(struct inode *inode, struct file *file)
 {
 	struct spufs_inode_info *i = SPUFS_I(inode);
@@ -992,7 +999,7 @@ static ssize_t spufs_signal2_read(struct file *file, char __user *buf,
 
 	spu_acquire_saved(ctx);
 	ret = __spufs_signal2_read(ctx, buf, len, pos);
-	spu_release(ctx);
+	spu_release_saved(ctx);
 
 	return ret;
 }
@@ -1062,6 +1069,13 @@ static const struct file_operations spufs_signal2_fops = {
 	.mmap = spufs_signal2_mmap,
 };
 
+static const struct file_operations spufs_signal2_nosched_fops = {
+	.open = spufs_signal2_open,
+	.release = spufs_signal2_release,
+	.write = spufs_signal2_write,
+	.mmap = spufs_signal2_mmap,
+};
+
 static void spufs_signal1_type_set(void *data, u64 val)
 {
 	struct spu_context *ctx = data;
@@ -1612,7 +1626,7 @@ static void spufs_decr_set(void *data, u64 val)
 	struct spu_lscsa *lscsa = ctx->csa.lscsa;
 	spu_acquire_saved(ctx);
 	lscsa->decr.slot[0] = (u32) val;
-	spu_release(ctx);
+	spu_release_saved(ctx);
 }
 
 static u64 __spufs_decr_get(void *data)
@@ -1628,7 +1642,7 @@ static u64 spufs_decr_get(void *data)
 	u64 ret;
 	spu_acquire_saved(ctx);
 	ret = __spufs_decr_get(data);
-	spu_release(ctx);
+	spu_release_saved(ctx);
 	return ret;
 }
 DEFINE_SIMPLE_ATTRIBUTE(spufs_decr_ops, spufs_decr_get, spufs_decr_set,
@@ -1637,17 +1651,21 @@ DEFINE_SIMPLE_ATTRIBUTE(spufs_decr_ops, spufs_decr_get, spufs_decr_set,
 static void spufs_decr_status_set(void *data, u64 val)
 {
 	struct spu_context *ctx = data;
-	struct spu_lscsa *lscsa = ctx->csa.lscsa;
 	spu_acquire_saved(ctx);
-	lscsa->decr_status.slot[0] = (u32) val;
-	spu_release(ctx);
+	if (val)
+		ctx->csa.priv2.mfc_control_RW |= MFC_CNTL_DECREMENTER_RUNNING;
+	else
+		ctx->csa.priv2.mfc_control_RW &= ~MFC_CNTL_DECREMENTER_RUNNING;
+	spu_release_saved(ctx);
 }
 
 static u64 __spufs_decr_status_get(void *data)
 {
 	struct spu_context *ctx = data;
-	struct spu_lscsa *lscsa = ctx->csa.lscsa;
-	return lscsa->decr_status.slot[0];
+	if (ctx->csa.priv2.mfc_control_RW & MFC_CNTL_DECREMENTER_RUNNING)
+		return SPU_DECR_STATUS_RUNNING;
+	else
+		return 0;
 }
 
 static u64 spufs_decr_status_get(void *data)
@@ -1656,7 +1674,7 @@ static u64 spufs_decr_status_get(void *data)
 	u64 ret;
 	spu_acquire_saved(ctx);
 	ret = __spufs_decr_status_get(data);
-	spu_release(ctx);
+	spu_release_saved(ctx);
 	return ret;
 }
 DEFINE_SIMPLE_ATTRIBUTE(spufs_decr_status_ops, spufs_decr_status_get,
@@ -1668,7 +1686,7 @@ static void spufs_event_mask_set(void *data, u64 val)
 	struct spu_lscsa *lscsa = ctx->csa.lscsa;
 	spu_acquire_saved(ctx);
 	lscsa->event_mask.slot[0] = (u32) val;
-	spu_release(ctx);
+	spu_release_saved(ctx);
 }
 
 static u64 __spufs_event_mask_get(void *data)
@@ -1684,7 +1702,7 @@ static u64 spufs_event_mask_get(void *data)
 	u64 ret;
 	spu_acquire_saved(ctx);
 	ret = __spufs_event_mask_get(data);
-	spu_release(ctx);
+	spu_release_saved(ctx);
 	return ret;
 }
 DEFINE_SIMPLE_ATTRIBUTE(spufs_event_mask_ops, spufs_event_mask_get,
@@ -1708,7 +1726,7 @@ static u64 spufs_event_status_get(void *data)
 
 	spu_acquire_saved(ctx);
 	ret = __spufs_event_status_get(data);
-	spu_release(ctx);
+	spu_release_saved(ctx);
 	return ret;
 }
 DEFINE_SIMPLE_ATTRIBUTE(spufs_event_status_ops, spufs_event_status_get,
@@ -1720,7 +1738,7 @@ static void spufs_srr0_set(void *data, u64 val)
 	struct spu_lscsa *lscsa = ctx->csa.lscsa;
 	spu_acquire_saved(ctx);
 	lscsa->srr0.slot[0] = (u32) val;
-	spu_release(ctx);
+	spu_release_saved(ctx);
 }
 
 static u64 spufs_srr0_get(void *data)
@@ -1730,7 +1748,7 @@ static u64 spufs_srr0_get(void *data)
 	u64 ret;
 	spu_acquire_saved(ctx);
 	ret = lscsa->srr0.slot[0];
-	spu_release(ctx);
+	spu_release_saved(ctx);
 	return ret;
 }
 DEFINE_SIMPLE_ATTRIBUTE(spufs_srr0_ops, spufs_srr0_get, spufs_srr0_set,
@@ -1786,7 +1804,7 @@ static u64 spufs_lslr_get(void *data)
 
 	spu_acquire_saved(ctx);
 	ret = __spufs_lslr_get(data);
-	spu_release(ctx);
+	spu_release_saved(ctx);
 
 	return ret;
 }
@@ -1850,7 +1868,7 @@ static ssize_t spufs_mbox_info_read(struct file *file, char __user *buf,
 	spin_lock(&ctx->csa.register_lock);
 	ret = __spufs_mbox_info_read(ctx, buf, len, pos);
 	spin_unlock(&ctx->csa.register_lock);
-	spu_release(ctx);
+	spu_release_saved(ctx);
 
 	return ret;
 }
@@ -1888,7 +1906,7 @@ static ssize_t spufs_ibox_info_read(struct file *file, char __user *buf,
 	spin_lock(&ctx->csa.register_lock);
 	ret = __spufs_ibox_info_read(ctx, buf, len, pos);
 	spin_unlock(&ctx->csa.register_lock);
-	spu_release(ctx);
+	spu_release_saved(ctx);
 
 	return ret;
 }
@@ -1929,7 +1947,7 @@ static ssize_t spufs_wbox_info_read(struct file *file, char __user *buf,
 	spin_lock(&ctx->csa.register_lock);
 	ret = __spufs_wbox_info_read(ctx, buf, len, pos);
 	spin_unlock(&ctx->csa.register_lock);
-	spu_release(ctx);
+	spu_release_saved(ctx);
 
 	return ret;
 }
@@ -1979,7 +1997,7 @@ static ssize_t spufs_dma_info_read(struct file *file, char __user *buf,
 	spin_lock(&ctx->csa.register_lock);
 	ret = __spufs_dma_info_read(ctx, buf, len, pos);
 	spin_unlock(&ctx->csa.register_lock);
-	spu_release(ctx);
+	spu_release_saved(ctx);
 
 	return ret;
 }
@@ -2030,7 +2048,7 @@ static ssize_t spufs_proxydma_info_read(struct file *file, char __user *buf,
 	spin_lock(&ctx->csa.register_lock);
 	ret = __spufs_proxydma_info_read(ctx, buf, len, pos);
 	spin_unlock(&ctx->csa.register_lock);
-	spu_release(ctx);
+	spu_release_saved(ctx);
 
 	return ret;
 }
@@ -2065,14 +2083,26 @@ static const char *ctx_state_names[] = {
 };
 
 static unsigned long long spufs_acct_time(struct spu_context *ctx,
-		enum spuctx_execution_state state)
+		enum spu_utilization_state state)
 {
-	unsigned long time = ctx->stats.times[state];
+	struct timespec ts;
+	unsigned long long time = ctx->stats.times[state];
 
-	if (ctx->stats.execution_state == state)
-		time += jiffies - ctx->stats.tstamp;
+	/*
+	 * In general, utilization statistics are updated by the controlling
+	 * thread as the spu context moves through various well defined
+	 * state transitions, but if the context is lazily loaded its
+	 * utilization statistics are not updated as the controlling thread
+	 * is not tightly coupled with the execution of the spu context.  We
+	 * calculate and apply the time delta from the last recorded state
+	 * of the spu context.
+	 */
+	if (ctx->spu && ctx->stats.util_state == state) {
+		ktime_get_ts(&ts);
+		time += timespec_to_ns(&ts) - ctx->stats.tstamp;
+	}
 
-	return jiffies_to_msecs(time);
+	return time / NSEC_PER_MSEC;
 }
 
 static unsigned long long spufs_slb_flts(struct spu_context *ctx)
@@ -2107,11 +2137,11 @@ static int spufs_show_stat(struct seq_file *s, void *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_state_names[ctx->stats.util_state],
+		spufs_acct_time(ctx, SPU_UTIL_USER),
+		spufs_acct_time(ctx, SPU_UTIL_SYSTEM),
+		spufs_acct_time(ctx, SPU_UTIL_IOWAIT),
+		spufs_acct_time(ctx, SPU_UTIL_IDLE_LOADED),
 		ctx->stats.vol_ctx_switch,
 		ctx->stats.invol_ctx_switch,
 		spufs_slb_flts(ctx),
@@ -2184,8 +2214,8 @@ struct tree_descr spufs_dir_nosched_contents[] = {
 	{ "mbox_stat", &spufs_mbox_stat_fops, 0444, },
 	{ "ibox_stat", &spufs_ibox_stat_fops, 0444, },
 	{ "wbox_stat", &spufs_wbox_stat_fops, 0444, },
-	{ "signal1", &spufs_signal1_fops, 0666, },
-	{ "signal2", &spufs_signal2_fops, 0666, },
+	{ "signal1", &spufs_signal1_nosched_fops, 0222, },
+	{ "signal2", &spufs_signal2_nosched_fops, 0222, },
 	{ "signal1_type", &spufs_signal1_type, 0666, },
 	{ "signal2_type", &spufs_signal2_type, 0666, },
 	{ "mss", &spufs_mss_fops, 0666, },
diff --git a/arch/powerpc/platforms/cell/spufs/gang.c b/arch/powerpc/platforms/cell/spufs/gang.c
index 212ea78f9051..71a443253021 100644
--- a/arch/powerpc/platforms/cell/spufs/gang.c
+++ b/arch/powerpc/platforms/cell/spufs/gang.c
@@ -35,7 +35,9 @@ struct spu_gang *alloc_spu_gang(void)
 
 	kref_init(&gang->kref);
 	mutex_init(&gang->mutex);
+	mutex_init(&gang->aff_mutex);
 	INIT_LIST_HEAD(&gang->list);
+	INIT_LIST_HEAD(&gang->aff_list_head);
 
 out:
 	return gang;
@@ -73,6 +75,10 @@ void spu_gang_remove_ctx(struct spu_gang *gang, struct spu_context *ctx)
 {
 	mutex_lock(&gang->mutex);
 	WARN_ON(ctx->gang != gang);
+	if (!list_empty(&ctx->aff_list)) {
+		list_del_init(&ctx->aff_list);
+		gang->aff_flags &= ~AFF_OFFSETS_SET;
+	}
 	list_del_init(&ctx->gang_list);
 	gang->contexts--;
 	mutex_unlock(&gang->mutex);
diff --git a/arch/powerpc/platforms/cell/spufs/inode.c b/arch/powerpc/platforms/cell/spufs/inode.c
index 7eb4d6cbcb74..b3d0dd118dd0 100644
--- a/arch/powerpc/platforms/cell/spufs/inode.c
+++ b/arch/powerpc/platforms/cell/spufs/inode.c
@@ -316,11 +316,107 @@ out:
 	return ret;
 }
 
-static int spufs_create_context(struct inode *inode,
-			struct dentry *dentry,
-			struct vfsmount *mnt, int flags, int mode)
+static struct spu_context *
+spufs_assert_affinity(unsigned int flags, struct spu_gang *gang,
+						struct file *filp)
+{
+	struct spu_context *tmp, *neighbor;
+	int count, node;
+	int aff_supp;
+
+	aff_supp = !list_empty(&(list_entry(cbe_spu_info[0].spus.next,
+					struct spu, cbe_list))->aff_list);
+
+	if (!aff_supp)
+		return ERR_PTR(-EINVAL);
+
+	if (flags & SPU_CREATE_GANG)
+		return ERR_PTR(-EINVAL);
+
+	if (flags & SPU_CREATE_AFFINITY_MEM &&
+	    gang->aff_ref_ctx &&
+	    gang->aff_ref_ctx->flags & SPU_CREATE_AFFINITY_MEM)
+		return ERR_PTR(-EEXIST);
+
+	if (gang->aff_flags & AFF_MERGED)
+		return ERR_PTR(-EBUSY);
+
+	neighbor = NULL;
+	if (flags & SPU_CREATE_AFFINITY_SPU) {
+		if (!filp || filp->f_op != &spufs_context_fops)
+			return ERR_PTR(-EINVAL);
+
+		neighbor = get_spu_context(
+				SPUFS_I(filp->f_dentry->d_inode)->i_ctx);
+
+		if (!list_empty(&neighbor->aff_list) && !(neighbor->aff_head) &&
+		    !list_is_last(&neighbor->aff_list, &gang->aff_list_head) &&
+		    !list_entry(neighbor->aff_list.next, struct spu_context,
+		    aff_list)->aff_head)
+			return ERR_PTR(-EEXIST);
+
+		if (gang != neighbor->gang)
+			return ERR_PTR(-EINVAL);
+
+		count = 1;
+		list_for_each_entry(tmp, &gang->aff_list_head, aff_list)
+			count++;
+		if (list_empty(&neighbor->aff_list))
+			count++;
+
+		for (node = 0; node < MAX_NUMNODES; node++) {
+			if ((cbe_spu_info[node].n_spus - atomic_read(
+				&cbe_spu_info[node].reserved_spus)) >= count)
+				break;
+		}
+
+		if (node == MAX_NUMNODES)
+			return ERR_PTR(-EEXIST);
+	}
+
+	return neighbor;
+}
+
+static void
+spufs_set_affinity(unsigned int flags, struct spu_context *ctx,
+					struct spu_context *neighbor)
+{
+	if (flags & SPU_CREATE_AFFINITY_MEM)
+		ctx->gang->aff_ref_ctx = ctx;
+
+	if (flags & SPU_CREATE_AFFINITY_SPU) {
+		if (list_empty(&neighbor->aff_list)) {
+			list_add_tail(&neighbor->aff_list,
+				&ctx->gang->aff_list_head);
+			neighbor->aff_head = 1;
+		}
+
+		if (list_is_last(&neighbor->aff_list, &ctx->gang->aff_list_head)
+		    || list_entry(neighbor->aff_list.next, struct spu_context,
+							aff_list)->aff_head) {
+			list_add(&ctx->aff_list, &neighbor->aff_list);
+		} else  {
+			list_add_tail(&ctx->aff_list, &neighbor->aff_list);
+			if (neighbor->aff_head) {
+				neighbor->aff_head = 0;
+				ctx->aff_head = 1;
+			}
+		}
+
+		if (!ctx->gang->aff_ref_ctx)
+			ctx->gang->aff_ref_ctx = ctx;
+	}
+}
+
+static int
+spufs_create_context(struct inode *inode, struct dentry *dentry,
+			struct vfsmount *mnt, int flags, int mode,
+			struct file *aff_filp)
 {
 	int ret;
+	int affinity;
+	struct spu_gang *gang;
+	struct spu_context *neighbor;
 
 	ret = -EPERM;
 	if ((flags & SPU_CREATE_NOSCHED) &&
@@ -336,9 +432,29 @@ static int spufs_create_context(struct inode *inode,
 	if ((flags & SPU_CREATE_ISOLATE) && !isolated_loader)
 		goto out_unlock;
 
+	gang = NULL;
+	neighbor = NULL;
+	affinity = flags & (SPU_CREATE_AFFINITY_MEM | SPU_CREATE_AFFINITY_SPU);
+	if (affinity) {
+		gang = SPUFS_I(inode)->i_gang;
+		ret = -EINVAL;
+		if (!gang)
+			goto out_unlock;
+		mutex_lock(&gang->aff_mutex);
+		neighbor = spufs_assert_affinity(flags, gang, aff_filp);
+		if (IS_ERR(neighbor)) {
+			ret = PTR_ERR(neighbor);
+			goto out_aff_unlock;
+		}
+	}
+
 	ret = spufs_mkdir(inode, dentry, flags, mode & S_IRWXUGO);
 	if (ret)
-		goto out_unlock;
+		goto out_aff_unlock;
+
+	if (affinity)
+		spufs_set_affinity(flags, SPUFS_I(dentry->d_inode)->i_ctx,
+								neighbor);
 
 	/*
 	 * get references for dget and mntget, will be released
@@ -352,6 +468,9 @@ static int spufs_create_context(struct inode *inode,
 		goto out;
 	}
 
+out_aff_unlock:
+	if (affinity)
+		mutex_unlock(&gang->aff_mutex);
 out_unlock:
 	mutex_unlock(&inode->i_mutex);
 out:
@@ -450,7 +569,8 @@ out:
 
 static struct file_system_type spufs_type;
 
-long spufs_create(struct nameidata *nd, unsigned int flags, mode_t mode)
+long spufs_create(struct nameidata *nd, unsigned int flags, mode_t mode,
+							struct file *filp)
 {
 	struct dentry *dentry;
 	int ret;
@@ -487,7 +607,7 @@ long spufs_create(struct nameidata *nd, unsigned int flags, mode_t mode)
 					dentry, nd->mnt, mode);
 	else
 		return spufs_create_context(nd->dentry->d_inode,
-					dentry, nd->mnt, flags, mode);
+					dentry, nd->mnt, flags, mode, filp);
 
 out_dput:
 	dput(dentry);
diff --git a/arch/powerpc/platforms/cell/spufs/run.c b/arch/powerpc/platforms/cell/spufs/run.c
index 58ae13b7de84..0b50fa5cb39d 100644
--- a/arch/powerpc/platforms/cell/spufs/run.c
+++ b/arch/powerpc/platforms/cell/spufs/run.c
@@ -18,15 +18,17 @@ void spufs_stop_callback(struct spu *spu)
 	wake_up_all(&ctx->stop_wq);
 }
 
-static inline int spu_stopped(struct spu_context *ctx, u32 * stat)
+static inline int spu_stopped(struct spu_context *ctx, u32 *stat)
 {
 	struct spu *spu;
 	u64 pte_fault;
 
 	*stat = ctx->ops->status_read(ctx);
-	if (ctx->state != SPU_STATE_RUNNABLE)
-		return 1;
+
 	spu = ctx->spu;
+	if (ctx->state != SPU_STATE_RUNNABLE ||
+	    test_bit(SPU_SCHED_NOTIFY_ACTIVE, &ctx->sched_flags))
+		return 1;
 	pte_fault = spu->dsisr &
 	    (MFC_DSISR_PTE_NOT_FOUND | MFC_DSISR_ACCESS_DENIED);
 	return (!(*stat & SPU_STATUS_RUNNING) || pte_fault || spu->class_0_pending) ?
@@ -124,8 +126,10 @@ out:
 	return ret;
 }
 
-static int spu_run_init(struct spu_context *ctx, u32 * npc)
+static int spu_run_init(struct spu_context *ctx, u32 *npc)
 {
+	spuctx_switch_state(ctx, SPU_UTIL_SYSTEM);
+
 	if (ctx->flags & SPU_CREATE_ISOLATE) {
 		unsigned long runcntl;
 
@@ -151,16 +155,20 @@ static int spu_run_init(struct spu_context *ctx, u32 * npc)
 		ctx->ops->runcntl_write(ctx, SPU_RUNCNTL_RUNNABLE);
 	}
 
+	spuctx_switch_state(ctx, SPU_UTIL_USER);
+
 	return 0;
 }
 
-static int spu_run_fini(struct spu_context *ctx, u32 * npc,
-			       u32 * status)
+static int spu_run_fini(struct spu_context *ctx, u32 *npc,
+			       u32 *status)
 {
 	int ret = 0;
 
 	*status = ctx->ops->status_read(ctx);
 	*npc = ctx->ops->npc_read(ctx);
+
+	spuctx_switch_state(ctx, SPU_UTIL_IDLE_LOADED);
 	spu_release(ctx);
 
 	if (signal_pending(current))
@@ -289,10 +297,10 @@ static inline int spu_process_events(struct spu_context *ctx)
 	return ret;
 }
 
-long spufs_run_spu(struct file *file, struct spu_context *ctx,
-		   u32 *npc, u32 *event)
+long spufs_run_spu(struct spu_context *ctx, u32 *npc, u32 *event)
 {
 	int ret;
+	struct spu *spu;
 	u32 status;
 
 	if (mutex_lock_interruptible(&ctx->run_mutex))
@@ -328,6 +336,17 @@ long spufs_run_spu(struct file *file, struct spu_context *ctx,
 		ret = spufs_wait(ctx->stop_wq, spu_stopped(ctx, &status));
 		if (unlikely(ret))
 			break;
+		spu = ctx->spu;
+		if (unlikely(test_and_clear_bit(SPU_SCHED_NOTIFY_ACTIVE,
+						&ctx->sched_flags))) {
+			if (!(status & SPU_STATUS_STOPPED_BY_STOP)) {
+				spu_switch_notify(spu, ctx);
+				continue;
+			}
+		}
+
+		spuctx_switch_state(ctx, SPU_UTIL_SYSTEM);
+
 		if ((status & SPU_STATUS_STOPPED_BY_STOP) &&
 		    (status >> SPU_STOP_STATUS_SHIFT == 0x2104)) {
 			ret = spu_process_callback(ctx);
@@ -356,6 +375,7 @@ long spufs_run_spu(struct file *file, struct spu_context *ctx,
 	    (ctx->state == SPU_STATE_RUNNABLE))
 		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 e5b4dd1db286..227968b4779d 100644
--- a/arch/powerpc/platforms/cell/spufs/sched.c
+++ b/arch/powerpc/platforms/cell/spufs/sched.c
@@ -51,9 +51,6 @@ struct spu_prio_array {
 	DECLARE_BITMAP(bitmap, MAX_PRIO);
 	struct list_head runq[MAX_PRIO];
 	spinlock_t runq_lock;
-	struct list_head active_list[MAX_NUMNODES];
-	struct mutex active_mutex[MAX_NUMNODES];
-	int nr_active[MAX_NUMNODES];
 	int nr_waiting;
 };
 
@@ -127,7 +124,7 @@ void __spu_update_sched_info(struct spu_context *ctx)
 	ctx->policy = current->policy;
 
 	/*
-	 * A lot of places that don't hold active_mutex poke into
+	 * A lot of places that don't hold list_mutex poke into
 	 * cpus_allowed, including grab_runnable_context which
 	 * already holds the runq_lock.  So abuse runq_lock
 	 * to protect this field aswell.
@@ -141,9 +138,9 @@ void spu_update_sched_info(struct spu_context *ctx)
 {
 	int node = ctx->spu->node;
 
-	mutex_lock(&spu_prio->active_mutex[node]);
+	mutex_lock(&cbe_spu_info[node].list_mutex);
 	__spu_update_sched_info(ctx);
-	mutex_unlock(&spu_prio->active_mutex[node]);
+	mutex_unlock(&cbe_spu_info[node].list_mutex);
 }
 
 static int __node_allowed(struct spu_context *ctx, int node)
@@ -169,56 +166,56 @@ static int node_allowed(struct spu_context *ctx, int node)
 	return rval;
 }
 
-/**
- * spu_add_to_active_list - add spu to active list
- * @spu:	spu to add to the active list
- */
-static void spu_add_to_active_list(struct spu *spu)
-{
-	int node = spu->node;
-
-	mutex_lock(&spu_prio->active_mutex[node]);
-	spu_prio->nr_active[node]++;
-	list_add_tail(&spu->list, &spu_prio->active_list[node]);
-	mutex_unlock(&spu_prio->active_mutex[node]);
-}
+static BLOCKING_NOTIFIER_HEAD(spu_switch_notifier);
 
-static void __spu_remove_from_active_list(struct spu *spu)
+void spu_switch_notify(struct spu *spu, struct spu_context *ctx)
 {
-	list_del_init(&spu->list);
-	spu_prio->nr_active[spu->node]--;
+	blocking_notifier_call_chain(&spu_switch_notifier,
+			    ctx ? ctx->object_id : 0, spu);
 }
 
-/**
- * spu_remove_from_active_list - remove spu from active list
- * @spu:       spu to remove from the active list
- */
-static void spu_remove_from_active_list(struct spu *spu)
+static void notify_spus_active(void)
 {
-	int node = spu->node;
-
-	mutex_lock(&spu_prio->active_mutex[node]);
-	__spu_remove_from_active_list(spu);
-	mutex_unlock(&spu_prio->active_mutex[node]);
-}
+	int node;
 
-static BLOCKING_NOTIFIER_HEAD(spu_switch_notifier);
+	/*
+	 * Wake up the active spu_contexts.
+	 *
+	 * When the awakened processes see their "notify_active" flag is set,
+	 * they will call spu_switch_notify();
+	 */
+	for_each_online_node(node) {
+		struct spu *spu;
 
-static void spu_switch_notify(struct spu *spu, struct spu_context *ctx)
-{
-	blocking_notifier_call_chain(&spu_switch_notifier,
-			    ctx ? ctx->object_id : 0, spu);
+		mutex_lock(&cbe_spu_info[node].list_mutex);
+		list_for_each_entry(spu, &cbe_spu_info[node].spus, cbe_list) {
+			if (spu->alloc_state != SPU_FREE) {
+				struct spu_context *ctx = spu->ctx;
+				set_bit(SPU_SCHED_NOTIFY_ACTIVE,
+					&ctx->sched_flags);
+				mb();
+				wake_up_all(&ctx->stop_wq);
+			}
+		}
+		mutex_unlock(&cbe_spu_info[node].list_mutex);
+	}
 }
 
 int spu_switch_event_register(struct notifier_block * n)
 {
-	return blocking_notifier_chain_register(&spu_switch_notifier, n);
+	int ret;
+	ret = blocking_notifier_chain_register(&spu_switch_notifier, n);
+	if (!ret)
+		notify_spus_active();
+	return ret;
 }
+EXPORT_SYMBOL_GPL(spu_switch_event_register);
 
 int spu_switch_event_unregister(struct notifier_block * n)
 {
 	return blocking_notifier_chain_unregister(&spu_switch_notifier, n);
 }
+EXPORT_SYMBOL_GPL(spu_switch_event_unregister);
 
 /**
  * spu_bind_context - bind spu context to physical spu
@@ -229,6 +226,12 @@ 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);
+	spuctx_switch_state(ctx, SPU_UTIL_SYSTEM);
+
+	if (ctx->flags & SPU_CREATE_NOSCHED)
+		atomic_inc(&cbe_spu_info[spu->node].reserved_spus);
+	if (!list_empty(&ctx->aff_list))
+		atomic_inc(&ctx->gang->aff_sched_count);
 
 	ctx->stats.slb_flt_base = spu->stats.slb_flt;
 	ctx->stats.class2_intr_base = spu->stats.class2_intr;
@@ -238,6 +241,7 @@ static void spu_bind_context(struct spu *spu, struct spu_context *ctx)
 	ctx->spu = spu;
 	ctx->ops = &spu_hw_ops;
 	spu->pid = current->pid;
+	spu->tgid = current->tgid;
 	spu_associate_mm(spu, ctx->owner);
 	spu->ibox_callback = spufs_ibox_callback;
 	spu->wbox_callback = spufs_wbox_callback;
@@ -251,7 +255,153 @@ static void spu_bind_context(struct spu *spu, struct spu_context *ctx)
 	spu_cpu_affinity_set(spu, raw_smp_processor_id());
 	spu_switch_notify(spu, ctx);
 	ctx->state = SPU_STATE_RUNNABLE;
-	spu_switch_state(spu, SPU_UTIL_SYSTEM);
+
+	spuctx_switch_state(ctx, SPU_UTIL_IDLE_LOADED);
+}
+
+/*
+ * Must be used with the list_mutex held.
+ */
+static inline int sched_spu(struct spu *spu)
+{
+	BUG_ON(!mutex_is_locked(&cbe_spu_info[spu->node].list_mutex));
+
+	return (!spu->ctx || !(spu->ctx->flags & SPU_CREATE_NOSCHED));
+}
+
+static void aff_merge_remaining_ctxs(struct spu_gang *gang)
+{
+	struct spu_context *ctx;
+
+	list_for_each_entry(ctx, &gang->aff_list_head, aff_list) {
+		if (list_empty(&ctx->aff_list))
+			list_add(&ctx->aff_list, &gang->aff_list_head);
+	}
+	gang->aff_flags |= AFF_MERGED;
+}
+
+static void aff_set_offsets(struct spu_gang *gang)
+{
+	struct spu_context *ctx;
+	int offset;
+
+	offset = -1;
+	list_for_each_entry_reverse(ctx, &gang->aff_ref_ctx->aff_list,
+								aff_list) {
+		if (&ctx->aff_list == &gang->aff_list_head)
+			break;
+		ctx->aff_offset = offset--;
+	}
+
+	offset = 0;
+	list_for_each_entry(ctx, gang->aff_ref_ctx->aff_list.prev, aff_list) {
+		if (&ctx->aff_list == &gang->aff_list_head)
+			break;
+		ctx->aff_offset = offset++;
+	}
+
+	gang->aff_flags |= AFF_OFFSETS_SET;
+}
+
+static struct spu *aff_ref_location(struct spu_context *ctx, int mem_aff,
+		 int group_size, int lowest_offset)
+{
+	struct spu *spu;
+	int node, n;
+
+	/*
+	 * TODO: A better algorithm could be used to find a good spu to be
+	 *       used as reference location for the ctxs chain.
+	 */
+	node = cpu_to_node(raw_smp_processor_id());
+	for (n = 0; n < MAX_NUMNODES; n++, node++) {
+		node = (node < MAX_NUMNODES) ? node : 0;
+		if (!node_allowed(ctx, node))
+			continue;
+		mutex_lock(&cbe_spu_info[node].list_mutex);
+		list_for_each_entry(spu, &cbe_spu_info[node].spus, cbe_list) {
+			if ((!mem_aff || spu->has_mem_affinity) &&
+							sched_spu(spu)) {
+				mutex_unlock(&cbe_spu_info[node].list_mutex);
+				return spu;
+			}
+		}
+		mutex_unlock(&cbe_spu_info[node].list_mutex);
+	}
+	return NULL;
+}
+
+static void aff_set_ref_point_location(struct spu_gang *gang)
+{
+	int mem_aff, gs, lowest_offset;
+	struct spu_context *ctx;
+	struct spu *tmp;
+
+	mem_aff = gang->aff_ref_ctx->flags & SPU_CREATE_AFFINITY_MEM;
+	lowest_offset = 0;
+	gs = 0;
+
+	list_for_each_entry(tmp, &gang->aff_list_head, aff_list)
+		gs++;
+
+	list_for_each_entry_reverse(ctx, &gang->aff_ref_ctx->aff_list,
+								aff_list) {
+		if (&ctx->aff_list == &gang->aff_list_head)
+			break;
+		lowest_offset = ctx->aff_offset;
+	}
+
+	gang->aff_ref_spu = aff_ref_location(ctx, mem_aff, gs, lowest_offset);
+}
+
+static struct spu *ctx_location(struct spu *ref, int offset, int node)
+{
+	struct spu *spu;
+
+	spu = NULL;
+	if (offset >= 0) {
+		list_for_each_entry(spu, ref->aff_list.prev, aff_list) {
+			BUG_ON(spu->node != node);
+			if (offset == 0)
+				break;
+			if (sched_spu(spu))
+				offset--;
+		}
+	} else {
+		list_for_each_entry_reverse(spu, ref->aff_list.next, aff_list) {
+			BUG_ON(spu->node != node);
+			if (offset == 0)
+				break;
+			if (sched_spu(spu))
+				offset++;
+		}
+	}
+
+	return spu;
+}
+
+/*
+ * affinity_check is called each time a context is going to be scheduled.
+ * It returns the spu ptr on which the context must run.
+ */
+static int has_affinity(struct spu_context *ctx)
+{
+	struct spu_gang *gang = ctx->gang;
+
+	if (list_empty(&ctx->aff_list))
+		return 0;
+
+	mutex_lock(&gang->aff_mutex);
+	if (!gang->aff_ref_spu) {
+		if (!(gang->aff_flags & AFF_MERGED))
+			aff_merge_remaining_ctxs(gang);
+		if (!(gang->aff_flags & AFF_OFFSETS_SET))
+			aff_set_offsets(gang);
+		aff_set_ref_point_location(gang);
+	}
+	mutex_unlock(&gang->aff_mutex);
+
+	return gang->aff_ref_spu != NULL;
 }
 
 /**
@@ -263,9 +413,13 @@ static void spu_unbind_context(struct spu *spu, struct spu_context *ctx)
 {
 	pr_debug("%s: unbind pid=%d SPU=%d NODE=%d\n", __FUNCTION__,
 		 spu->pid, spu->number, spu->node);
+	spuctx_switch_state(ctx, SPU_UTIL_SYSTEM);
 
-	spu_switch_state(spu, SPU_UTIL_IDLE);
-
+ 	if (spu->ctx->flags & SPU_CREATE_NOSCHED)
+		atomic_dec(&cbe_spu_info[spu->node].reserved_spus);
+ 	if (!list_empty(&ctx->aff_list))
+ 		if (atomic_dec_and_test(&ctx->gang->aff_sched_count))
+ 			ctx->gang->aff_ref_spu = NULL;
 	spu_switch_notify(spu, NULL);
 	spu_unmap_mappings(ctx);
 	spu_save(&ctx->csa, spu);
@@ -278,8 +432,8 @@ static void spu_unbind_context(struct spu *spu, struct spu_context *ctx)
 	spu->dma_callback = NULL;
 	spu_associate_mm(spu, NULL);
 	spu->pid = 0;
+	spu->tgid = 0;
 	ctx->ops = &spu_backing_ops;
-	ctx->spu = NULL;
 	spu->flags = 0;
 	spu->ctx = NULL;
 
@@ -287,6 +441,10 @@ static void spu_unbind_context(struct spu *spu, struct spu_context *ctx)
 		(spu->stats.slb_flt - ctx->stats.slb_flt_base);
 	ctx->stats.class2_intr +=
 		(spu->stats.class2_intr - ctx->stats.class2_intr_base);
+
+	/* This maps the underlying spu state to idle */
+	spuctx_switch_state(ctx, SPU_UTIL_IDLE_LOADED);
+	ctx->spu = NULL;
 }
 
 /**
@@ -352,18 +510,41 @@ static void spu_prio_wait(struct spu_context *ctx)
 
 static struct spu *spu_get_idle(struct spu_context *ctx)
 {
-	struct spu *spu = NULL;
-	int node = cpu_to_node(raw_smp_processor_id());
-	int n;
+	struct spu *spu;
+	int node, n;
+
+	if (has_affinity(ctx)) {
+		node = ctx->gang->aff_ref_spu->node;
 
+		mutex_lock(&cbe_spu_info[node].list_mutex);
+		spu = ctx_location(ctx->gang->aff_ref_spu, ctx->aff_offset, node);
+		if (spu && spu->alloc_state == SPU_FREE)
+			goto found;
+		mutex_unlock(&cbe_spu_info[node].list_mutex);
+		return NULL;
+	}
+
+	node = cpu_to_node(raw_smp_processor_id());
 	for (n = 0; n < MAX_NUMNODES; n++, node++) {
 		node = (node < MAX_NUMNODES) ? node : 0;
 		if (!node_allowed(ctx, node))
 			continue;
-		spu = spu_alloc_node(node);
-		if (spu)
-			break;
+
+		mutex_lock(&cbe_spu_info[node].list_mutex);
+		list_for_each_entry(spu, &cbe_spu_info[node].spus, cbe_list) {
+			if (spu->alloc_state == SPU_FREE)
+				goto found;
+		}
+		mutex_unlock(&cbe_spu_info[node].list_mutex);
 	}
+
+	return NULL;
+
+ found:
+	spu->alloc_state = SPU_USED;
+	mutex_unlock(&cbe_spu_info[node].list_mutex);
+	pr_debug("Got SPU %d %d\n", spu->number, spu->node);
+	spu_init_channels(spu);
 	return spu;
 }
 
@@ -393,15 +574,15 @@ static struct spu *find_victim(struct spu_context *ctx)
 		if (!node_allowed(ctx, node))
 			continue;
 
-		mutex_lock(&spu_prio->active_mutex[node]);
-		list_for_each_entry(spu, &spu_prio->active_list[node], list) {
+		mutex_lock(&cbe_spu_info[node].list_mutex);
+		list_for_each_entry(spu, &cbe_spu_info[node].spus, cbe_list) {
 			struct spu_context *tmp = spu->ctx;
 
 			if (tmp->prio > ctx->prio &&
 			    (!victim || tmp->prio > victim->prio))
 				victim = spu->ctx;
 		}
-		mutex_unlock(&spu_prio->active_mutex[node]);
+		mutex_unlock(&cbe_spu_info[node].list_mutex);
 
 		if (victim) {
 			/*
@@ -426,7 +607,11 @@ static struct spu *find_victim(struct spu_context *ctx)
 				victim = NULL;
 				goto restart;
 			}
-			spu_remove_from_active_list(spu);
+
+			mutex_lock(&cbe_spu_info[node].list_mutex);
+			cbe_spu_info[node].nr_active--;
+			mutex_unlock(&cbe_spu_info[node].list_mutex);
+
 			spu_unbind_context(spu, victim);
 			victim->stats.invol_ctx_switch++;
 			spu->stats.invol_ctx_switch++;
@@ -455,8 +640,6 @@ 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);
-
 	do {
 		struct spu *spu;
 
@@ -477,8 +660,12 @@ int spu_activate(struct spu_context *ctx, unsigned long flags)
 		if (!spu && rt_prio(ctx->prio))
 			spu = find_victim(ctx);
 		if (spu) {
+			int node = spu->node;
+
+			mutex_lock(&cbe_spu_info[node].list_mutex);
 			spu_bind_context(spu, ctx);
-			spu_add_to_active_list(spu);
+			cbe_spu_info[node].nr_active++;
+			mutex_unlock(&cbe_spu_info[node].list_mutex);
 			return 0;
 		}
 
@@ -500,7 +687,7 @@ static struct spu_context *grab_runnable_context(int prio, int node)
 	int best;
 
 	spin_lock(&spu_prio->runq_lock);
-	best = sched_find_first_bit(spu_prio->bitmap);
+	best = find_first_bit(spu_prio->bitmap, prio);
 	while (best < prio) {
 		struct list_head *rq = &spu_prio->runq[best];
 
@@ -527,11 +714,17 @@ static int __spu_deactivate(struct spu_context *ctx, int force, int max_prio)
 	if (spu) {
 		new = grab_runnable_context(max_prio, spu->node);
 		if (new || force) {
-			spu_remove_from_active_list(spu);
+			int node = spu->node;
+
+			mutex_lock(&cbe_spu_info[node].list_mutex);
 			spu_unbind_context(spu, ctx);
+			spu->alloc_state = SPU_FREE;
+			cbe_spu_info[node].nr_active--;
+			mutex_unlock(&cbe_spu_info[node].list_mutex);
+
 			ctx->stats.vol_ctx_switch++;
 			spu->stats.vol_ctx_switch++;
-			spu_free(spu);
+
 			if (new)
 				wake_up(&new->stop_wq);
 		}
@@ -550,21 +743,11 @@ static int __spu_deactivate(struct spu_context *ctx, int force, int max_prio)
  */
 void spu_deactivate(struct spu_context *ctx)
 {
-	/*
-	 * We must never reach this for a nosched context,
-	 * but handle the case gracefull instead of panicing.
-	 */
-	if (ctx->flags & SPU_CREATE_NOSCHED) {
-		WARN_ON(1);
-		return;
-	}
-
 	__spu_deactivate(ctx, 1, MAX_PRIO);
-	spuctx_switch_state(ctx, SPUCTX_UTIL_USER);
 }
 
 /**
- * spu_yield -  yield a physical spu if others are waiting
+ * spu_yield -	yield a physical spu if others are waiting
  * @ctx:	spu context to yield
  *
  * Check if there is a higher priority context waiting and if yes
@@ -575,17 +758,12 @@ void spu_yield(struct spu_context *ctx)
 {
 	if (!(ctx->flags & SPU_CREATE_NOSCHED)) {
 		mutex_lock(&ctx->state_mutex);
-		if (__spu_deactivate(ctx, 0, MAX_PRIO))
-			spuctx_switch_state(ctx, SPUCTX_UTIL_USER);
-		else {
-			spuctx_switch_state(ctx, SPUCTX_UTIL_LOADED);
-			spu_switch_state(ctx->spu, SPU_UTIL_USER);
-		}
+		__spu_deactivate(ctx, 0, MAX_PRIO);
 		mutex_unlock(&ctx->state_mutex);
 	}
 }
 
-static void spusched_tick(struct spu_context *ctx)
+static noinline void spusched_tick(struct spu_context *ctx)
 {
 	if (ctx->flags & SPU_CREATE_NOSCHED)
 		return;
@@ -596,7 +774,7 @@ static void spusched_tick(struct spu_context *ctx)
 		return;
 
 	/*
-	 * Unfortunately active_mutex ranks outside of state_mutex, so
+	 * Unfortunately list_mutex ranks outside of state_mutex, so
 	 * we have to trylock here.  If we fail give the context another
 	 * tick and try again.
 	 */
@@ -606,12 +784,11 @@ static void spusched_tick(struct spu_context *ctx)
 
 		new = grab_runnable_context(ctx->prio + 1, spu->node);
 		if (new) {
-
-			__spu_remove_from_active_list(spu);
 			spu_unbind_context(spu, ctx);
 			ctx->stats.invol_ctx_switch++;
 			spu->stats.invol_ctx_switch++;
-			spu_free(spu);
+			spu->alloc_state = SPU_FREE;
+			cbe_spu_info[spu->node].nr_active--;
 			wake_up(&new->stop_wq);
 			/*
 			 * We need to break out of the wait loop in
@@ -632,7 +809,7 @@ static void spusched_tick(struct spu_context *ctx)
  *
  * Return the number of tasks currently running or waiting to run.
  *
- * Note that we don't take runq_lock / active_mutex here.  Reading
+ * Note that we don't take runq_lock / list_mutex here.  Reading
  * a single 32bit value is atomic on powerpc, and we don't care
  * about memory ordering issues here.
  */
@@ -641,7 +818,7 @@ static unsigned long count_active_contexts(void)
 	int nr_active = 0, node;
 
 	for (node = 0; node < MAX_NUMNODES; node++)
-		nr_active += spu_prio->nr_active[node];
+		nr_active += cbe_spu_info[node].nr_active;
 	nr_active += spu_prio->nr_waiting;
 
 	return nr_active;
@@ -681,19 +858,18 @@ static void spusched_wake(unsigned long data)
 
 static int spusched_thread(void *unused)
 {
-	struct spu *spu, *next;
+	struct spu *spu;
 	int node;
 
 	while (!kthread_should_stop()) {
 		set_current_state(TASK_INTERRUPTIBLE);
 		schedule();
 		for (node = 0; node < MAX_NUMNODES; node++) {
-			mutex_lock(&spu_prio->active_mutex[node]);
-			list_for_each_entry_safe(spu, next,
-						 &spu_prio->active_list[node],
-						 list)
-				spusched_tick(spu->ctx);
-			mutex_unlock(&spu_prio->active_mutex[node]);
+			mutex_lock(&cbe_spu_info[node].list_mutex);
+			list_for_each_entry(spu, &cbe_spu_info[node].spus, cbe_list)
+				if (spu->ctx)
+					spusched_tick(spu->ctx);
+			mutex_unlock(&cbe_spu_info[node].list_mutex);
 		}
 	}
 
@@ -751,10 +927,9 @@ int __init spu_sched_init(void)
 		INIT_LIST_HEAD(&spu_prio->runq[i]);
 		__clear_bit(i, spu_prio->bitmap);
 	}
-	__set_bit(MAX_PRIO, spu_prio->bitmap);
 	for (i = 0; i < MAX_NUMNODES; i++) {
-		mutex_init(&spu_prio->active_mutex[i]);
-		INIT_LIST_HEAD(&spu_prio->active_list[i]);
+		mutex_init(&cbe_spu_info[i].list_mutex);
+		INIT_LIST_HEAD(&cbe_spu_info[i].spus);
 	}
 	spin_lock_init(&spu_prio->runq_lock);
 
@@ -783,9 +958,9 @@ int __init spu_sched_init(void)
 	return err;
 }
 
-void __exit spu_sched_exit(void)
+void spu_sched_exit(void)
 {
-	struct spu *spu, *tmp;
+	struct spu *spu;
 	int node;
 
 	remove_proc_entry("spu_loadavg", NULL);
@@ -794,13 +969,11 @@ void __exit spu_sched_exit(void)
 	kthread_stop(spusched_task);
 
 	for (node = 0; node < MAX_NUMNODES; node++) {
-		mutex_lock(&spu_prio->active_mutex[node]);
-		list_for_each_entry_safe(spu, tmp, &spu_prio->active_list[node],
-					 list) {
-			list_del_init(&spu->list);
-			spu_free(spu);
-		}
-		mutex_unlock(&spu_prio->active_mutex[node]);
+		mutex_lock(&cbe_spu_info[node].list_mutex);
+		list_for_each_entry(spu, &cbe_spu_info[node].spus, cbe_list)
+			if (spu->alloc_state != SPU_FREE)
+				spu->alloc_state = SPU_FREE;
+		mutex_unlock(&cbe_spu_info[node].list_mutex);
 	}
 	kfree(spu_prio);
 }
diff --git a/arch/powerpc/platforms/cell/spufs/spu_restore.c b/arch/powerpc/platforms/cell/spufs/spu_restore.c
index 4e19ed7a0756..21a9c952d88b 100644
--- a/arch/powerpc/platforms/cell/spufs/spu_restore.c
+++ b/arch/powerpc/platforms/cell/spufs/spu_restore.c
@@ -84,13 +84,13 @@ static inline void restore_decr(void)
 	unsigned int decr_running;
 	unsigned int decr;
 
-	/* Restore, Step 6:
+	/* Restore, Step 6(moved):
 	 *    If the LSCSA "decrementer running" flag is set
 	 *    then write the SPU_WrDec channel with the
 	 *    decrementer value from LSCSA.
 	 */
 	offset = LSCSA_QW_OFFSET(decr_status);
-	decr_running = regs_spill[offset].slot[0];
+	decr_running = regs_spill[offset].slot[0] & SPU_DECR_STATUS_RUNNING;
 	if (decr_running) {
 		offset = LSCSA_QW_OFFSET(decr);
 		decr = regs_spill[offset].slot[0];
@@ -318,10 +318,10 @@ int main()
 	build_dma_list(lscsa_ea);	/* Step 3.  */
 	restore_upper_240kb(lscsa_ea);	/* Step 4.  */
 					/* Step 5: done by 'exit'. */
-	restore_decr();			/* Step 6. */
 	enqueue_putllc(lscsa_ea);	/* Step 7. */
 	set_tag_update();		/* Step 8. */
 	read_tag_status();		/* Step 9. */
+	restore_decr();			/* moved Step 6. */
 	read_llar_status();		/* Step 10. */
 	write_ppu_mb();			/* Step 11. */
 	write_ppuint_mb();		/* Step 12. */
diff --git a/arch/powerpc/platforms/cell/spufs/spu_restore_dump.h_shipped b/arch/powerpc/platforms/cell/spufs/spu_restore_dump.h_shipped
index 15183d209b58..f383b027e8bf 100644
--- a/arch/powerpc/platforms/cell/spufs/spu_restore_dump.h_shipped
+++ b/arch/powerpc/platforms/cell/spufs/spu_restore_dump.h_shipped
@@ -10,7 +10,7 @@ static unsigned int spu_restore_code[]  __attribute__((__aligned__(128))) = {
 0x24fd8081,
 0x1cd80081,
 0x33001180,
-0x42030003,
+0x42034003,
 0x33800284,
 0x1c010204,
 0x40200000,
@@ -24,22 +24,22 @@ static unsigned int spu_restore_code[]  __attribute__((__aligned__(128))) = {
 0x23fffd84,
 0x1c100183,
 0x217ffa85,
-0x3080a000,
-0x3080a201,
-0x3080a402,
-0x3080a603,
-0x3080a804,
-0x3080aa05,
-0x3080ac06,
-0x3080ae07,
-0x3080b008,
-0x3080b209,
-0x3080b40a,
-0x3080b60b,
-0x3080b80c,
-0x3080ba0d,
-0x3080bc0e,
-0x3080be0f,
+0x3080b000,
+0x3080b201,
+0x3080b402,
+0x3080b603,
+0x3080b804,
+0x3080ba05,
+0x3080bc06,
+0x3080be07,
+0x3080c008,
+0x3080c209,
+0x3080c40a,
+0x3080c60b,
+0x3080c80c,
+0x3080ca0d,
+0x3080cc0e,
+0x3080ce0f,
 0x00003ffc,
 0x00000000,
 0x00000000,
@@ -48,19 +48,18 @@ static unsigned int spu_restore_code[]  __attribute__((__aligned__(128))) = {
 0x3ec00083,
 0xb0a14103,
 0x01a00204,
-0x3ec10082,
-0x4202800e,
-0x04000703,
-0xb0a14202,
-0x21a00803,
-0x3fbf028d,
-0x3f20068d,
-0x3fbe0682,
+0x3ec10083,
+0x4202c002,
+0xb0a14203,
+0x21a00802,
+0x3fbf028a,
+0x3f20050a,
+0x3fbe0502,
 0x3fe30102,
 0x21a00882,
-0x3f82028f,
-0x3fe3078f,
-0x3fbf0784,
+0x3f82028b,
+0x3fe3058b,
+0x3fbf0584,
 0x3f200204,
 0x3fbe0204,
 0x3fe30204,
@@ -75,252 +74,285 @@ static unsigned int spu_restore_code[]  __attribute__((__aligned__(128))) = {
 0x21a00083,
 0x40800082,
 0x21a00b02,
-0x10002818,
-0x42a00002,
-0x32800007,
-0x4207000c,
-0x18008208,
-0x40a0000b,
-0x4080020a,
-0x40800709,
-0x00200000,
-0x42070002,
-0x3ac30384,
+0x10002612,
+0x42a00003,
+0x42074006,
+0x1800c204,
+0x40a00008,
+0x40800789,
+0x1c010305,
+0x34000302,
 0x1cffc489,
-0x00200000,
-0x18008383,
-0x38830382,
-0x4cffc486,
-0x3ac28185,
-0xb0408584,
-0x28830382,
-0x1c020387,
-0x38828182,
-0xb0408405,
-0x1802c408,
-0x28828182,
-0x217ff886,
-0x04000583,
-0x21a00803,
-0x3fbe0682,
-0x3fe30102,
-0x04000106,
-0x21a00886,
-0x04000603,
-0x21a00903,
-0x40803c02,
-0x21a00982,
-0x40800003,
-0x04000184,
-0x21a00a04,
+0x3ec00303,
+0x3ec00287,
+0xb0408403,
+0x24000302,
+0x34000282,
+0x1c020306,
+0xb0408207,
+0x18020204,
+0x24000282,
+0x217ffa09,
+0x04000402,
+0x21a00802,
+0x3fbe0504,
+0x3fe30204,
+0x21a00884,
+0x42074002,
+0x21a00902,
+0x40803c03,
+0x21a00983,
+0x04000485,
+0x21a00a05,
 0x40802202,
 0x21a00a82,
-0x42028005,
-0x34208702,
-0x21002282,
-0x21a00804,
-0x21a00886,
-0x3fbf0782,
+0x21a00805,
+0x21a00884,
+0x3fbf0582,
 0x3f200102,
 0x3fbe0102,
 0x3fe30102,
 0x21a00902,
 0x40804003,
 0x21a00983,
-0x21a00a04,
+0x21a00a05,
 0x40805a02,
 0x21a00a82,
 0x40800083,
 0x21a00b83,
 0x01a00c02,
-0x01a00d83,
-0x3420c282,
+0x30809c03,
+0x34000182,
+0x14004102,
+0x21002082,
+0x01a00d82,
+0x3080a003,
+0x34000182,
 0x21a00e02,
-0x34210283,
-0x21a00f03,
-0x34200284,
-0x77400200,
-0x3421c282,
+0x3080a203,
+0x34000182,
+0x21a00f02,
+0x3080a403,
+0x34000182,
+0x77400100,
+0x3080a603,
+0x34000182,
 0x21a00702,
-0x34218283,
-0x21a00083,
-0x34214282,
+0x3080a803,
+0x34000182,
+0x21a00082,
+0x3080aa03,
+0x34000182,
 0x21a00b02,
-0x4200480c,
-0x00200000,
-0x1c010286,
-0x34220284,
-0x34220302,
-0x0f608203,
-0x5c024204,
-0x3b81810b,
-0x42013c02,
-0x00200000,
-0x18008185,
-0x38808183,
-0x3b814182,
-0x21004e84,
+0x4020007f,
+0x3080ae02,
+0x42004805,
+0x3080ac04,
+0x34000103,
+0x34000202,
+0x1cffc183,
+0x3b810106,
+0x0f608184,
+0x42013802,
+0x5c020183,
+0x38810102,
+0x3b810102,
+0x21000e83,
 0x4020007f,
 0x35000100,
-0x000004e0,
-0x000002a0,
-0x000002e8,
-0x00000428,
+0x00000470,
+0x000002f8,
+0x00000430,
 0x00000360,
-0x000002e8,
-0x000004a0,
-0x00000468,
+0x000002f8,
 0x000003c8,
+0x000004a8,
+0x00000298,
 0x00000360,
+0x00200000,
 0x409ffe02,
 0x30801203,
-0x40800204,
-0x3ec40085,
-0x10009c09,
-0x3ac10606,
-0xb060c105,
-0x4020007f,
-0x4020007f,
+0x40800208,
+0x3ec40084,
+0x40800407,
+0x3ac20289,
+0xb060c104,
+0x3ac1c284,
 0x20801203,
-0x38810602,
-0xb0408586,
-0x28810602,
-0x32004180,
-0x34204702,
+0x38820282,
+0x41004003,
+0xb0408189,
+0x28820282,
+0x3881c282,
+0xb0408304,
+0x2881c282,
+0x00400000,
+0x40800003,
+0x35000000,
+0x30809e03,
+0x34000182,
 0x21a00382,
 0x4020007f,
-0x327fdc80,
+0x327fde00,
 0x409ffe02,
 0x30801203,
-0x40800204,
-0x3ec40087,
-0x40800405,
-0x00200000,
-0x40800606,
-0x3ac10608,
-0x3ac14609,
-0x3ac1860a,
-0xb060c107,
+0x40800206,
+0x3ec40084,
+0x40800407,
+0x40800608,
+0x3ac1828a,
+0x3ac20289,
+0xb060c104,
+0x3ac1c284,
 0x20801203,
+0x38818282,
 0x41004003,
-0x38810602,
-0x4020007f,
-0xb0408188,
-0x4020007f,
-0x28810602,
-0x41201002,
-0x38814603,
-0x10009c09,
-0xb060c109,
-0x4020007f,
-0x28814603,
+0xb040818a,
+0x10005b0b,
+0x41201003,
+0x28818282,
+0x3881c282,
+0xb0408184,
 0x41193f83,
-0x38818602,
 0x60ffc003,
-0xb040818a,
-0x28818602,
-0x32003080,
+0x2881c282,
+0x38820282,
+0xb0408189,
+0x28820282,
+0x327fef80,
 0x409ffe02,
 0x30801203,
-0x40800204,
-0x3ec40087,
-0x41201008,
-0x10009c14,
-0x40800405,
-0x3ac10609,
-0x40800606,
-0x3ac1460a,
-0xb060c107,
-0x3ac1860b,
+0x40800207,
+0x3ec40086,
+0x4120100b,
+0x10005b14,
+0x40800404,
+0x3ac1c289,
+0x40800608,
+0xb060c106,
+0x3ac10286,
+0x3ac2028a,
 0x20801203,
-0x38810602,
-0xb0408409,
-0x28810602,
-0x38814603,
-0xb060c40a,
-0x4020007f,
-0x28814603,
+0x3881c282,
 0x41193f83,
-0x38818602,
 0x60ffc003,
-0xb040818b,
-0x28818602,
-0x32002380,
-0x409ffe02,
-0x30801204,
-0x40800205,
-0x3ec40083,
-0x40800406,
-0x3ac14607,
-0x3ac18608,
-0xb0810103,
-0x41004002,
-0x20801204,
-0x4020007f,
-0x38814603,
-0x10009c0b,
-0xb060c107,
-0x4020007f,
-0x4020007f,
-0x28814603,
-0x38818602,
-0x4020007f,
+0xb0408589,
+0x2881c282,
+0x38810282,
+0xb0408586,
+0x28810282,
+0x38820282,
+0xb040818a,
+0x28820282,
 0x4020007f,
-0xb0408588,
-0x28818602,
+0x327fe280,
+0x409ffe02,
+0x30801203,
+0x40800207,
+0x3ec40084,
+0x40800408,
+0x10005b14,
+0x40800609,
+0x3ac1c28a,
+0x3ac2028b,
+0xb060c104,
+0x3ac24284,
+0x20801203,
+0x41201003,
+0x3881c282,
+0xb040830a,
+0x2881c282,
+0x38820282,
+0xb040818b,
+0x41193f83,
+0x60ffc003,
+0x28820282,
+0x38824282,
+0xb0408184,
+0x28824282,
 0x4020007f,
-0x32001780,
+0x327fd580,
 0x409ffe02,
-0x1000640e,
-0x40800204,
+0x1000658e,
+0x40800206,
 0x30801203,
-0x40800405,
-0x3ec40087,
-0x40800606,
-0x3ac10608,
-0x3ac14609,
-0x3ac1860a,
-0xb060c107,
+0x40800407,
+0x3ec40084,
+0x40800608,
+0x3ac1828a,
+0x3ac20289,
+0xb060c104,
+0x3ac1c284,
 0x20801203,
 0x413d8003,
-0x38810602,
+0x38818282,
 0x4020007f,
-0x327fd780,
-0x409ffe02,
-0x10007f0c,
-0x40800205,
-0x30801204,
-0x40800406,
-0x3ec40083,
-0x3ac14607,
-0x3ac18608,
-0xb0810103,
-0x413d8002,
-0x20801204,
-0x38814603,
+0x327fd800,
+0x409ffe03,
+0x30801202,
+0x40800207,
+0x3ec40084,
+0x10005b09,
+0x3ac1c288,
+0xb0408184,
 0x4020007f,
-0x327feb80,
+0x4020007f,
+0x20801202,
+0x3881c282,
+0xb0408308,
+0x2881c282,
+0x327fc680,
 0x409ffe02,
+0x1000588b,
+0x40800208,
 0x30801203,
-0x40800204,
-0x3ec40087,
-0x40800405,
-0x1000650a,
-0x40800606,
-0x3ac10608,
-0x3ac14609,
-0x3ac1860a,
-0xb060c107,
+0x40800407,
+0x3ec40084,
+0x3ac20289,
+0xb060c104,
+0x3ac1c284,
 0x20801203,
-0x38810602,
-0xb0408588,
-0x4020007f,
-0x327fc980,
-0x00400000,
-0x40800003,
-0x4020007f,
-0x35000000,
+0x413d8003,
+0x38820282,
+0x327fbd80,
+0x00200000,
+0x00000da0,
+0x00000000,
+0x00000000,
+0x00000000,
+0x00000d90,
+0x00000000,
+0x00000000,
+0x00000000,
+0x00000db0,
+0x00000000,
+0x00000000,
+0x00000000,
+0x00000dc0,
+0x00000000,
+0x00000000,
+0x00000000,
+0x00000d80,
+0x00000000,
+0x00000000,
+0x00000000,
+0x00000df0,
+0x00000000,
+0x00000000,
+0x00000000,
+0x00000de0,
+0x00000000,
+0x00000000,
+0x00000000,
+0x00000dd0,
+0x00000000,
+0x00000000,
+0x00000000,
+0x00000e04,
+0x00000000,
+0x00000000,
 0x00000000,
+0x00000e00,
 0x00000000,
 0x00000000,
 0x00000000,
diff --git a/arch/powerpc/platforms/cell/spufs/spufs.h b/arch/powerpc/platforms/cell/spufs/spufs.h
index 08b3530288ac..8b20c0c1556f 100644
--- a/arch/powerpc/platforms/cell/spufs/spufs.h
+++ b/arch/powerpc/platforms/cell/spufs/spufs.h
@@ -40,17 +40,13 @@ 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
+enum {
+	SPU_SCHED_WAS_ACTIVE,	/* was active upon spu_acquire_saved()  */
+};
+
+/* ctx->sched_flags */
+enum {
+	SPU_SCHED_NOTIFY_ACTIVE,
 };
 
 struct spu_context {
@@ -89,6 +85,8 @@ struct spu_context {
 
 	struct list_head gang_list;
 	struct spu_gang *gang;
+	struct kref *prof_priv_kref;
+	void ( * prof_priv_release) (struct kref *kref);
 
 	/* owner thread */
 	pid_t tid;
@@ -104,9 +102,9 @@ struct spu_context {
 	/* 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];
+		enum spu_utilization_state util_state;
+		unsigned long long tstamp;	/* time of last state switch */
+		unsigned long long times[SPU_UTIL_MAX];
 		unsigned long long vol_ctx_switch;
 		unsigned long long invol_ctx_switch;
 		unsigned long long min_flt;
@@ -118,6 +116,10 @@ struct spu_context {
 		unsigned long long class2_intr_base; /* # at last ctx switch */
 		unsigned long long libassist;
 	} stats;
+
+	struct list_head aff_list;
+	int aff_head;
+	int aff_offset;
 };
 
 struct spu_gang {
@@ -125,8 +127,19 @@ struct spu_gang {
 	struct mutex mutex;
 	struct kref kref;
 	int contexts;
+
+	struct spu_context *aff_ref_ctx;
+	struct list_head aff_list_head;
+	struct mutex aff_mutex;
+	int aff_flags;
+	struct spu *aff_ref_spu;
+	atomic_t aff_sched_count;
 };
 
+/* Flag bits for spu_gang aff_flags */
+#define AFF_OFFSETS_SET		1
+#define AFF_MERGED		2
+
 struct mfc_dma_command {
 	int32_t pad;	/* reserved */
 	uint32_t lsa;	/* local storage address */
@@ -190,10 +203,9 @@ extern struct tree_descr spufs_dir_contents[];
 extern struct tree_descr spufs_dir_nosched_contents[];
 
 /* system call implementation */
-long spufs_run_spu(struct file *file,
-		   struct spu_context *ctx, u32 *npc, u32 *status);
-long spufs_create(struct nameidata *nd,
-			 unsigned int flags, mode_t mode);
+long spufs_run_spu(struct spu_context *ctx, u32 *npc, u32 *status);
+long spufs_create(struct nameidata *nd, unsigned int flags,
+			mode_t mode, struct file *filp);
 extern const struct file_operations spufs_context_fops;
 
 /* gang management */
@@ -206,6 +218,9 @@ void spu_gang_add_ctx(struct spu_gang *gang, struct spu_context *ctx);
 /* fault handling */
 int spufs_handle_class1(struct spu_context *ctx);
 
+/* affinity */
+struct spu *affinity_check(struct spu_context *ctx);
+
 /* context management */
 extern atomic_t nr_spu_contexts;
 static inline void spu_acquire(struct spu_context *ctx)
@@ -227,15 +242,17 @@ void spu_unmap_mappings(struct spu_context *ctx);
 void spu_forget(struct spu_context *ctx);
 int spu_acquire_runnable(struct spu_context *ctx, unsigned long flags);
 void spu_acquire_saved(struct spu_context *ctx);
+void spu_release_saved(struct spu_context *ctx);
 
 int spu_activate(struct spu_context *ctx, unsigned long flags);
 void spu_deactivate(struct spu_context *ctx);
 void spu_yield(struct spu_context *ctx);
+void spu_switch_notify(struct spu *spu, struct spu_context *ctx);
 void spu_set_timeslice(struct spu_context *ctx);
 void spu_update_sched_info(struct spu_context *ctx);
 void __spu_update_sched_info(struct spu_context *ctx);
 int __init spu_sched_init(void);
-void __exit spu_sched_exit(void);
+void spu_sched_exit(void);
 
 extern char *isolated_loader;
 
@@ -293,30 +310,34 @@ extern int spufs_coredump_num_notes;
  * line.
  */
 static inline void spuctx_switch_state(struct spu_context *ctx,
-		enum spuctx_execution_state new_state)
+		enum spu_utilization_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;
-	}
-}
+	unsigned long long curtime;
+	signed long long delta;
+	struct timespec ts;
+	struct spu *spu;
+	enum spu_utilization_state old_state;
 
-static inline void spu_switch_state(struct spu *spu,
-		enum spuctx_execution_state new_state)
-{
-	if (spu->stats.utilization_state != new_state) {
-		unsigned long curtime = jiffies;
+	ktime_get_ts(&ts);
+	curtime = timespec_to_ns(&ts);
+	delta = curtime - ctx->stats.tstamp;
 
-		spu->stats.times[spu->stats.utilization_state] +=
-				 curtime - spu->stats.tstamp;
+	WARN_ON(!mutex_is_locked(&ctx->state_mutex));
+	WARN_ON(delta < 0);
+
+	spu = ctx->spu;
+	old_state = ctx->stats.util_state;
+	ctx->stats.util_state = new_state;
+	ctx->stats.tstamp = curtime;
+
+	/*
+	 * Update the physical SPU utilization statistics.
+	 */
+	if (spu) {
+		ctx->stats.times[old_state] += delta;
+		spu->stats.times[old_state] += delta;
+		spu->stats.util_state = new_state;
 		spu->stats.tstamp = curtime;
-		spu->stats.utilization_state = new_state;
 	}
 }
 
diff --git a/arch/powerpc/platforms/cell/spufs/switch.c b/arch/powerpc/platforms/cell/spufs/switch.c
index 9c506ba08cdc..27ffdae98e5a 100644
--- a/arch/powerpc/platforms/cell/spufs/switch.c
+++ b/arch/powerpc/platforms/cell/spufs/switch.c
@@ -180,7 +180,7 @@ static inline void save_mfc_cntl(struct spu_state *csa, struct spu *spu)
 	case MFC_CNTL_SUSPEND_COMPLETE:
 		if (csa) {
 			csa->priv2.mfc_control_RW =
-				in_be64(&priv2->mfc_control_RW) |
+				MFC_CNTL_SUSPEND_MASK |
 				MFC_CNTL_SUSPEND_DMA_QUEUE;
 		}
 		break;
@@ -190,9 +190,7 @@ static inline void save_mfc_cntl(struct spu_state *csa, struct spu *spu)
 				  MFC_CNTL_SUSPEND_DMA_STATUS_MASK) ==
 				 MFC_CNTL_SUSPEND_COMPLETE);
 		if (csa) {
-			csa->priv2.mfc_control_RW =
-				in_be64(&priv2->mfc_control_RW) &
-				~MFC_CNTL_SUSPEND_DMA_QUEUE;
+			csa->priv2.mfc_control_RW = 0;
 		}
 		break;
 	}
@@ -251,16 +249,8 @@ static inline void save_mfc_decr(struct spu_state *csa, struct spu *spu)
 	 *     Read MFC_CNTL[Ds].  Update saved copy of
 	 *     CSA.MFC_CNTL[Ds].
 	 */
-	if (in_be64(&priv2->mfc_control_RW) & MFC_CNTL_DECREMENTER_RUNNING) {
-		csa->priv2.mfc_control_RW |= MFC_CNTL_DECREMENTER_RUNNING;
-		csa->suspend_time = get_cycles();
-		out_be64(&priv2->spu_chnlcntptr_RW, 7ULL);
-		eieio();
-		csa->spu_chnldata_RW[7] = in_be64(&priv2->spu_chnldata_RW);
-		eieio();
-	} else {
-		csa->priv2.mfc_control_RW &= ~MFC_CNTL_DECREMENTER_RUNNING;
-	}
+	csa->priv2.mfc_control_RW |=
+		in_be64(&priv2->mfc_control_RW) & MFC_CNTL_DECREMENTER_RUNNING;
 }
 
 static inline void halt_mfc_decr(struct spu_state *csa, struct spu *spu)
@@ -271,7 +261,8 @@ static inline void halt_mfc_decr(struct spu_state *csa, struct spu *spu)
 	 *     Write MFC_CNTL[Dh] set to a '1' to halt
 	 *     the decrementer.
 	 */
-	out_be64(&priv2->mfc_control_RW, MFC_CNTL_DECREMENTER_HALTED);
+	out_be64(&priv2->mfc_control_RW,
+		 MFC_CNTL_DECREMENTER_HALTED | MFC_CNTL_SUSPEND_MASK);
 	eieio();
 }
 
@@ -615,7 +606,7 @@ static inline void save_ppuint_mb(struct spu_state *csa, struct spu *spu)
 static inline void save_ch_part1(struct spu_state *csa, struct spu *spu)
 {
 	struct spu_priv2 __iomem *priv2 = spu->priv2;
-	u64 idx, ch_indices[7] = { 0UL, 3UL, 4UL, 24UL, 25UL, 27UL };
+	u64 idx, ch_indices[] = { 0UL, 3UL, 4UL, 24UL, 25UL, 27UL };
 	int i;
 
 	/* Save, Step 42:
@@ -626,7 +617,7 @@ static inline void save_ch_part1(struct spu_state *csa, struct spu *spu)
 	csa->spu_chnldata_RW[1] = in_be64(&priv2->spu_chnldata_RW);
 
 	/* Save the following CH: [0,3,4,24,25,27] */
-	for (i = 0; i < 7; i++) {
+	for (i = 0; i < ARRAY_SIZE(ch_indices); i++) {
 		idx = ch_indices[i];
 		out_be64(&priv2->spu_chnlcntptr_RW, idx);
 		eieio();
@@ -983,13 +974,13 @@ static inline void terminate_spu_app(struct spu_state *csa, struct spu *spu)
 	 */
 }
 
-static inline void suspend_mfc(struct spu_state *csa, struct spu *spu)
+static inline void suspend_mfc_and_halt_decr(struct spu_state *csa,
+		struct spu *spu)
 {
 	struct spu_priv2 __iomem *priv2 = spu->priv2;
 
 	/* Restore, Step 7:
-	 * Restore, Step 47.
-	 *     Write MFC_Cntl[Dh,Sc]='1','1' to suspend
+	 *     Write MFC_Cntl[Dh,Sc,Sm]='1','1','0' to suspend
 	 *     the queue and halt the decrementer.
 	 */
 	out_be64(&priv2->mfc_control_RW, MFC_CNTL_SUSPEND_DMA_QUEUE |
@@ -1090,7 +1081,7 @@ static inline void clear_spu_status(struct spu_state *csa, struct spu *spu)
 static inline void reset_ch_part1(struct spu_state *csa, struct spu *spu)
 {
 	struct spu_priv2 __iomem *priv2 = spu->priv2;
-	u64 ch_indices[7] = { 0UL, 3UL, 4UL, 24UL, 25UL, 27UL };
+	u64 ch_indices[] = { 0UL, 3UL, 4UL, 24UL, 25UL, 27UL };
 	u64 idx;
 	int i;
 
@@ -1102,7 +1093,7 @@ static inline void reset_ch_part1(struct spu_state *csa, struct spu *spu)
 	out_be64(&priv2->spu_chnldata_RW, 0UL);
 
 	/* Reset the following CH: [0,3,4,24,25,27] */
-	for (i = 0; i < 7; i++) {
+	for (i = 0; i < ARRAY_SIZE(ch_indices); i++) {
 		idx = ch_indices[i];
 		out_be64(&priv2->spu_chnlcntptr_RW, idx);
 		eieio();
@@ -1289,7 +1280,15 @@ static inline void setup_decr(struct spu_state *csa, struct spu *spu)
 		cycles_t resume_time = get_cycles();
 		cycles_t delta_time = resume_time - csa->suspend_time;
 
+		csa->lscsa->decr_status.slot[0] = SPU_DECR_STATUS_RUNNING;
+		if (csa->lscsa->decr.slot[0] < delta_time) {
+			csa->lscsa->decr_status.slot[0] |=
+				 SPU_DECR_STATUS_WRAPPED;
+		}
+
 		csa->lscsa->decr.slot[0] -= delta_time;
+	} else {
+		csa->lscsa->decr_status.slot[0] = 0;
 	}
 }
 
@@ -1398,6 +1397,18 @@ static inline void restore_ls_16kb(struct spu_state *csa, struct spu *spu)
 	send_mfc_dma(spu, addr, ls_offset, size, tag, rclass, cmd);
 }
 
+static inline void suspend_mfc(struct spu_state *csa, struct spu *spu)
+{
+	struct spu_priv2 __iomem *priv2 = spu->priv2;
+
+	/* Restore, Step 47.
+	 *     Write MFC_Cntl[Sc,Sm]='1','0' to suspend
+	 *     the queue.
+	 */
+	out_be64(&priv2->mfc_control_RW, MFC_CNTL_SUSPEND_DMA_QUEUE);
+	eieio();
+}
+
 static inline void clear_interrupts(struct spu_state *csa, struct spu *spu)
 {
 	/* Restore, Step 49:
@@ -1548,10 +1559,10 @@ static inline void restore_decr_wrapped(struct spu_state *csa, struct spu *spu)
 	 *     "wrapped" flag is set, OR in a '1' to
 	 *     CSA.SPU_Event_Status[Tm].
 	 */
-	if (csa->lscsa->decr_status.slot[0] == 1) {
+	if (csa->lscsa->decr_status.slot[0] & SPU_DECR_STATUS_WRAPPED) {
 		csa->spu_chnldata_RW[0] |= 0x20;
 	}
-	if ((csa->lscsa->decr_status.slot[0] == 1) &&
+	if ((csa->lscsa->decr_status.slot[0] & SPU_DECR_STATUS_WRAPPED) &&
 	    (csa->spu_chnlcnt_RW[0] == 0 &&
 	     ((csa->spu_chnldata_RW[2] & 0x20) == 0x0) &&
 	     ((csa->spu_chnldata_RW[0] & 0x20) != 0x1))) {
@@ -1562,18 +1573,13 @@ static inline void restore_decr_wrapped(struct spu_state *csa, struct spu *spu)
 static inline void restore_ch_part1(struct spu_state *csa, struct spu *spu)
 {
 	struct spu_priv2 __iomem *priv2 = spu->priv2;
-	u64 idx, ch_indices[7] = { 0UL, 3UL, 4UL, 24UL, 25UL, 27UL };
+	u64 idx, ch_indices[] = { 0UL, 3UL, 4UL, 24UL, 25UL, 27UL };
 	int i;
 
 	/* Restore, Step 59:
+	 *	Restore the following CH: [0,3,4,24,25,27]
 	 */
-
-	/* Restore CH 1 without count */
-	out_be64(&priv2->spu_chnlcntptr_RW, 1);
-	out_be64(&priv2->spu_chnldata_RW, csa->spu_chnldata_RW[1]);
-
-	/* Restore the following CH: [0,3,4,24,25,27] */
-	for (i = 0; i < 7; i++) {
+	for (i = 0; i < ARRAY_SIZE(ch_indices); i++) {
 		idx = ch_indices[i];
 		out_be64(&priv2->spu_chnlcntptr_RW, idx);
 		eieio();
@@ -1932,7 +1938,7 @@ static void harvest(struct spu_state *prev, struct spu *spu)
 	set_switch_pending(prev, spu);	        /* Step 5.  */
 	stop_spu_isolate(spu);			/* NEW.     */
 	remove_other_spu_access(prev, spu);	/* Step 6.  */
-	suspend_mfc(prev, spu);	                /* Step 7.  */
+	suspend_mfc_and_halt_decr(prev, spu);	/* Step 7.  */
 	wait_suspend_mfc_complete(prev, spu);	/* Step 8.  */
 	if (!suspend_spe(prev, spu))	        /* Step 9.  */
 		clear_spu_status(prev, spu);	/* Step 10. */
diff --git a/arch/powerpc/platforms/cell/spufs/syscalls.c b/arch/powerpc/platforms/cell/spufs/syscalls.c
index 8e37bdf4dfda..43f0fb88abbc 100644
--- a/arch/powerpc/platforms/cell/spufs/syscalls.c
+++ b/arch/powerpc/platforms/cell/spufs/syscalls.c
@@ -47,7 +47,7 @@ static long do_spu_run(struct file *filp,
 		goto out;
 
 	i = SPUFS_I(filp->f_path.dentry->d_inode);
-	ret = spufs_run_spu(filp, i->i_ctx, &npc, &status);
+	ret = spufs_run_spu(i->i_ctx, &npc, &status);
 
 	if (put_user(npc, unpc))
 		ret = -EFAULT;
@@ -76,8 +76,8 @@ asmlinkage long sys_spu_run(int fd, __u32 __user *unpc, __u32 __user *ustatus)
 }
 #endif
 
-asmlinkage long sys_spu_create(const char __user *pathname,
-					unsigned int flags, mode_t mode)
+asmlinkage long do_spu_create(const char __user *pathname, unsigned int flags,
+				mode_t mode, struct file *neighbor)
 {
 	char *tmp;
 	int ret;
@@ -90,7 +90,7 @@ asmlinkage long sys_spu_create(const char __user *pathname,
 		ret = path_lookup(tmp, LOOKUP_PARENT|
 				LOOKUP_OPEN|LOOKUP_CREATE, &nd);
 		if (!ret) {
-			ret = spufs_create(&nd, flags, mode);
+			ret = spufs_create(&nd, flags, mode, neighbor);
 			path_release(&nd);
 		}
 		putname(tmp);
@@ -99,8 +99,32 @@ asmlinkage long sys_spu_create(const char __user *pathname,
 	return ret;
 }
 
+#ifndef MODULE
+asmlinkage long sys_spu_create(const char __user *pathname, unsigned int flags,
+				mode_t mode, int neighbor_fd)
+{
+	int fput_needed;
+	struct file *neighbor;
+	long ret;
+
+	if (flags & SPU_CREATE_AFFINITY_SPU) {
+		ret = -EBADF;
+		neighbor = fget_light(neighbor_fd, &fput_needed);
+		if (neighbor) {
+			ret = do_spu_create(pathname, flags, mode, neighbor);
+			fput_light(neighbor, fput_needed);
+		}
+	}
+	else {
+		ret = do_spu_create(pathname, flags, mode, NULL);
+	}
+
+	return ret;
+}
+#endif
+
 struct spufs_calls spufs_calls = {
-	.create_thread = sys_spu_create,
+	.create_thread = do_spu_create,
 	.spu_run = do_spu_run,
 	.owner = THIS_MODULE,
 };