[mirror_ubuntu-hirsute-kernel.git] / arch / powerpc / platforms / cell / spufs / sched.c

/* sched.c - SPU scheduler.
 *
 * Copyright (C) IBM 2005
 * Author: Mark Nutter <mnutter@us.ibm.com>
 *
 * 2006-03-31	NUMA domains added.
 *
 * 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, or (at your option)
 * any later version.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with this program; if not, write to the Free Software
 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
 */

#undef DEBUG

#include <linux/module.h>
#include <linux/errno.h>
#include <linux/sched.h>
#include <linux/kernel.h>
#include <linux/mm.h>
#include <linux/completion.h>
#include <linux/vmalloc.h>
#include <linux/smp.h>
#include <linux/smp_lock.h>
#include <linux/stddef.h>
#include <linux/unistd.h>
#include <linux/numa.h>
#include <linux/mutex.h>
#include <linux/notifier.h>

#include <asm/io.h>
#include <asm/mmu_context.h>
#include <asm/spu.h>
#include <asm/spu_csa.h>
#include <asm/spu_priv1.h>
#include "spufs.h"

#define SPU_MIN_TIMESLICE 	(100 * HZ / 1000)

#define SPU_BITMAP_SIZE (((MAX_PRIO+BITS_PER_LONG)/BITS_PER_LONG)+1)
struct spu_prio_array {
	unsigned long bitmap[SPU_BITMAP_SIZE];
	wait_queue_head_t waitq[MAX_PRIO];
	struct list_head active_list[MAX_NUMNODES];
	struct mutex active_mutex[MAX_NUMNODES];
};

static struct spu_prio_array *spu_prio;

static inline int node_allowed(int node)
{
	cpumask_t mask;

	if (!nr_cpus_node(node))
		return 0;
	mask = node_to_cpumask(node);
	if (!cpus_intersects(mask, current->cpus_allowed))
		return 0;
	return 1;
}

static inline void mm_needs_global_tlbie(struct mm_struct *mm)
{
	int nr = (NR_CPUS > 1) ? NR_CPUS : NR_CPUS + 1;

	/* Global TLBIE broadcast required with SPEs. */
	__cpus_setall(&mm->cpu_vm_mask, nr);
}

static BLOCKING_NOTIFIER_HEAD(spu_switch_notifier);

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);
}

int spu_switch_event_register(struct notifier_block * n)
{
	return blocking_notifier_chain_register(&spu_switch_notifier, n);
}

int spu_switch_event_unregister(struct notifier_block * n)
{
	return blocking_notifier_chain_unregister(&spu_switch_notifier, n);
}


static inline void 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);
	spu->ctx = ctx;
	spu->flags = 0;
	ctx->spu = spu;
	ctx->ops = &spu_hw_ops;
	spu->pid = current->pid;
	spu->prio = current->prio;
	spu->mm = ctx->owner;
	mm_needs_global_tlbie(spu->mm);
	spu->ibox_callback = spufs_ibox_callback;
	spu->wbox_callback = spufs_wbox_callback;
	spu->stop_callback = spufs_stop_callback;
	spu->mfc_callback = spufs_mfc_callback;
	spu->dma_callback = spufs_dma_callback;
	mb();
	spu_unmap_mappings(ctx);
	spu_restore(&ctx->csa, spu);
	spu->timestamp = jiffies;
	spu_cpu_affinity_set(spu, raw_smp_processor_id());
	spu_switch_notify(spu, ctx);

	ctx->state = SPU_STATE_RUNNABLE;
}

static inline void 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);
	spu_switch_notify(spu, NULL);
	spu_unmap_mappings(ctx);
	spu_save(&ctx->csa, spu);
	spu->timestamp = jiffies;
	ctx->state = SPU_STATE_SAVED;
	spu->ibox_callback = NULL;
	spu->wbox_callback = NULL;
	spu->stop_callback = NULL;
	spu->mfc_callback = NULL;
	spu->dma_callback = NULL;
	spu->mm = NULL;
	spu->pid = 0;
	spu->prio = MAX_PRIO;
	ctx->ops = &spu_backing_ops;
	ctx->spu = NULL;
	spu->flags = 0;
	spu->ctx = NULL;
}

static inline void spu_add_wq(wait_queue_head_t * wq, wait_queue_t * wait,
			      int prio)
{
	prepare_to_wait_exclusive(wq, wait, TASK_INTERRUPTIBLE);
	set_bit(prio, spu_prio->bitmap);
}

static inline void spu_del_wq(wait_queue_head_t * wq, wait_queue_t * wait,
			      int prio)
{
	u64 flags;

	__set_current_state(TASK_RUNNING);

	spin_lock_irqsave(&wq->lock, flags);

	remove_wait_queue_locked(wq, wait);
	if (list_empty(&wq->task_list))
		clear_bit(prio, spu_prio->bitmap);

	spin_unlock_irqrestore(&wq->lock, flags);
}

static void spu_prio_wait(struct spu_context *ctx, u64 flags)
{
	int prio = current->prio;
	wait_queue_head_t *wq = &spu_prio->waitq[prio];
	DEFINE_WAIT(wait);

	if (ctx->spu)
		return;

	spu_add_wq(wq, &wait, prio);

	if (!signal_pending(current)) {
		up_write(&ctx->state_sema);
		pr_debug("%s: pid=%d prio=%d\n", __FUNCTION__,
			 current->pid, current->prio);
		schedule();
		down_write(&ctx->state_sema);
	}

	spu_del_wq(wq, &wait, prio);
}

static void spu_prio_wakeup(void)
{
	int best = sched_find_first_bit(spu_prio->bitmap);
	if (best < MAX_PRIO) {
		wait_queue_head_t *wq = &spu_prio->waitq[best];
		wake_up_interruptible_nr(wq, 1);
	}
}

static int get_active_spu(struct spu *spu)
{
	int node = spu->node;
	struct spu *tmp;
	int rc = 0;

	mutex_lock(&spu_prio->active_mutex[node]);
	list_for_each_entry(tmp, &spu_prio->active_list[node], list) {
		if (tmp == spu) {
			list_del_init(&spu->list);
			rc = 1;
			break;
		}
	}
	mutex_unlock(&spu_prio->active_mutex[node]);
	return rc;
}

static void put_active_spu(struct spu *spu)
{
	int node = spu->node;

	mutex_lock(&spu_prio->active_mutex[node]);
	list_add_tail(&spu->list, &spu_prio->active_list[node]);
	mutex_unlock(&spu_prio->active_mutex[node]);
}

static struct spu *spu_get_idle(struct spu_context *ctx, u64 flags)
{
	struct spu *spu = NULL;
	int node = cpu_to_node(raw_smp_processor_id());
	int n;

	for (n = 0; n < MAX_NUMNODES; n++, node++) {
		node = (node < MAX_NUMNODES) ? node : 0;
		if (!node_allowed(node))
			continue;
		spu = spu_alloc_node(node);
		if (spu)
			break;
	}
	return spu;
}

static inline struct spu *spu_get(struct spu_context *ctx, u64 flags)
{
	/* Future: spu_get_idle() if possible,
	 * otherwise try to preempt an active
	 * context.
	 */
	return spu_get_idle(ctx, flags);
}

/* The three externally callable interfaces
 * for the scheduler begin here.
 *
 *	spu_activate	- bind a context to SPU, waiting as needed.
 *	spu_deactivate	- unbind a context from its SPU.
 *	spu_yield	- yield an SPU if others are waiting.
 */

int spu_activate(struct spu_context *ctx, u64 flags)
{
	struct spu *spu;
	int ret = 0;

	for (;;) {
		if (ctx->spu)
			return 0;
		spu = spu_get(ctx, flags);
		if (spu != NULL) {
			if (ctx->spu != NULL) {
				spu_free(spu);
				spu_prio_wakeup();
				break;
			}
			bind_context(spu, ctx);
			put_active_spu(spu);
			break;
		}
		spu_prio_wait(ctx, flags);
		if (signal_pending(current)) {
			ret = -ERESTARTSYS;
			spu_prio_wakeup();
			break;
		}
	}
	return ret;
}

void spu_deactivate(struct spu_context *ctx)
{
	struct spu *spu;
	int needs_idle;

	spu = ctx->spu;
	if (!spu)
		return;
	needs_idle = get_active_spu(spu);
	unbind_context(spu, ctx);
	if (needs_idle) {
		spu_free(spu);
		spu_prio_wakeup();
	}
}

void spu_yield(struct spu_context *ctx)
{
	struct spu *spu;
	int need_yield = 0;

	if (down_write_trylock(&ctx->state_sema)) {
		if ((spu = ctx->spu) != NULL) {
			int best = sched_find_first_bit(spu_prio->bitmap);
			if (best < MAX_PRIO) {
				pr_debug("%s: yielding SPU %d NODE %d\n",
					 __FUNCTION__, spu->number, spu->node);
				spu_deactivate(ctx);
				need_yield = 1;
			} else {
				spu->prio = MAX_PRIO;
			}
		}
		up_write(&ctx->state_sema);
	}
	if (unlikely(need_yield))
		yield();
}

int __init spu_sched_init(void)
{
	int i;

	spu_prio = kzalloc(sizeof(struct spu_prio_array), GFP_KERNEL);
	if (!spu_prio) {
		printk(KERN_WARNING "%s: Unable to allocate priority queue.\n",
		       __FUNCTION__);
		return 1;
	}
	for (i = 0; i < MAX_PRIO; i++) {
		init_waitqueue_head(&spu_prio->waitq[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]);
	}
	return 0;
}

void __exit spu_sched_exit(void)
{
	struct spu *spu, *tmp;
	int node;

	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]);
	}
	kfree(spu_prio);
}
Commit	Line	Data
8b3d6663 AB	1	/* sched.c - SPU scheduler.
	2	*
	3	* Copyright (C) IBM 2005
	4	* Author: Mark Nutter <mnutter@us.ibm.com>
	5	*
a68cf983	6	* 2006-03-31 NUMA domains added.
8b3d6663 AB	7	*
	8	* This program is free software; you can redistribute it and/or modify
	9	* it under the terms of the GNU General Public License as published by
	10	* the Free Software Foundation; either version 2, or (at your option)
	11	* any later version.
	12	*
	13	* This program is distributed in the hope that it will be useful,
	14	* but WITHOUT ANY WARRANTY; without even the implied warranty of
	15	* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
	16	* GNU General Public License for more details.
	17	*
	18	* You should have received a copy of the GNU General Public License
	19	* along with this program; if not, write to the Free Software
	20	* Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
	21	*/
	22
3b3d22cb AB	23	#undef DEBUG
3b3d22cb AB	24
8b3d6663 AB	25	#include <linux/module.h>
	26	#include <linux/errno.h>
	27	#include <linux/sched.h>
	28	#include <linux/kernel.h>
	29	#include <linux/mm.h>
	30	#include <linux/completion.h>
	31	#include <linux/vmalloc.h>
	32	#include <linux/smp.h>
	33	#include <linux/smp_lock.h>
	34	#include <linux/stddef.h>
	35	#include <linux/unistd.h>
a68cf983 MN	36	#include <linux/numa.h>
a68cf983 MN	37	#include <linux/mutex.h>
86767277	38	#include <linux/notifier.h>
8b3d6663 AB	39
	40	#include <asm/io.h>
	41	#include <asm/mmu_context.h>
	42	#include <asm/spu.h>
	43	#include <asm/spu_csa.h>
a91942ae	44	#include <asm/spu_priv1.h>
8b3d6663 AB	45	#include "spufs.h"
8b3d6663 AB	46
7945a4a2	47	#define SPU_MIN_TIMESLICE (100 * HZ / 1000)
2a911f0b	48
8b3d6663 AB	49	#define SPU_BITMAP_SIZE (((MAX_PRIO+BITS_PER_LONG)/BITS_PER_LONG)+1)
8b3d6663 AB	50	struct spu_prio_array {
8b3d6663 AB	51	unsigned long bitmap[SPU_BITMAP_SIZE];
8b3d6663 AB	52	wait_queue_head_t waitq[MAX_PRIO];
a68cf983 MN	53	struct list_head active_list[MAX_NUMNODES];
a68cf983 MN	54	struct mutex active_mutex[MAX_NUMNODES];
8b3d6663 AB	55	};
8b3d6663 AB	56
a68cf983	57	static struct spu_prio_array *spu_prio;
8b3d6663	58
a68cf983	59	static inline int node_allowed(int node)
8b3d6663	60	{
a68cf983	61	cpumask_t mask;
8b3d6663	62
a68cf983 MN	63	if (!nr_cpus_node(node))
	64	return 0;
	65	mask = node_to_cpumask(node);
	66	if (!cpus_intersects(mask, current->cpus_allowed))
	67	return 0;
	68	return 1;
8b3d6663 AB	69	}
	70
	71	static inline void mm_needs_global_tlbie(struct mm_struct *mm)
	72	{
a68cf983 MN	73	int nr = (NR_CPUS > 1) ? NR_CPUS : NR_CPUS + 1;
a68cf983 MN	74
8b3d6663	75	/* Global TLBIE broadcast required with SPEs. */
a68cf983	76	__cpus_setall(&mm->cpu_vm_mask, nr);
8b3d6663 AB	77	}
8b3d6663 AB	78
86767277 AB	79	static BLOCKING_NOTIFIER_HEAD(spu_switch_notifier);
	80
	81	static void spu_switch_notify(struct spu spu, struct spu_context ctx)
	82	{
	83	blocking_notifier_call_chain(&spu_switch_notifier,
	84	ctx ? ctx->object_id : 0, spu);
	85	}
	86
	87	int spu_switch_event_register(struct notifier_block * n)
	88	{
	89	return blocking_notifier_chain_register(&spu_switch_notifier, n);
	90	}
	91
	92	int spu_switch_event_unregister(struct notifier_block * n)
	93	{
	94	return blocking_notifier_chain_unregister(&spu_switch_notifier, n);
	95	}
	96
	97
8b3d6663 AB	98	static inline void bind_context(struct spu spu, struct spu_context ctx)
8b3d6663 AB	99	{
a68cf983 MN	100	pr_debug("%s: pid=%d SPU=%d NODE=%d\n", __FUNCTION__, current->pid,
a68cf983 MN	101	spu->number, spu->node);
8b3d6663 AB	102	spu->ctx = ctx;
	103	spu->flags = 0;
	104	ctx->spu = spu;
	105	ctx->ops = &spu_hw_ops;
	106	spu->pid = current->pid;
	107	spu->prio = current->prio;
	108	spu->mm = ctx->owner;
	109	mm_needs_global_tlbie(spu->mm);
	110	spu->ibox_callback = spufs_ibox_callback;
	111	spu->wbox_callback = spufs_wbox_callback;
5110459f	112	spu->stop_callback = spufs_stop_callback;
a33a7d73	113	spu->mfc_callback = spufs_mfc_callback;
9add11da	114	spu->dma_callback = spufs_dma_callback;
8b3d6663	115	mb();
5110459f	116	spu_unmap_mappings(ctx);
8b3d6663	117	spu_restore(&ctx->csa, spu);
2a911f0b	118	spu->timestamp = jiffies;
a68cf983	119	spu_cpu_affinity_set(spu, raw_smp_processor_id());
86767277	120	spu_switch_notify(spu, ctx);
81998baf CH	121
81998baf CH	122	ctx->state = SPU_STATE_RUNNABLE;
8b3d6663 AB	123	}
	124
	125	static inline void unbind_context(struct spu spu, struct spu_context ctx)
	126	{
a68cf983 MN	127	pr_debug("%s: unbind pid=%d SPU=%d NODE=%d\n", __FUNCTION__,
a68cf983 MN	128	spu->pid, spu->number, spu->node);
86767277	129	spu_switch_notify(spu, NULL);
5110459f	130	spu_unmap_mappings(ctx);
8b3d6663	131	spu_save(&ctx->csa, spu);
2a911f0b	132	spu->timestamp = jiffies;
8b3d6663 AB	133	ctx->state = SPU_STATE_SAVED;
	134	spu->ibox_callback = NULL;
	135	spu->wbox_callback = NULL;
5110459f	136	spu->stop_callback = NULL;
a33a7d73	137	spu->mfc_callback = NULL;
9add11da	138	spu->dma_callback = NULL;
8b3d6663 AB	139	spu->mm = NULL;
	140	spu->pid = 0;
	141	spu->prio = MAX_PRIO;
	142	ctx->ops = &spu_backing_ops;
	143	ctx->spu = NULL;
2a911f0b	144	spu->flags = 0;
8b3d6663 AB	145	spu->ctx = NULL;
	146	}
	147
a68cf983 MN	148	static inline void spu_add_wq(wait_queue_head_t * wq, wait_queue_t * wait,
a68cf983 MN	149	int prio)
8b3d6663	150	{
a68cf983 MN	151	prepare_to_wait_exclusive(wq, wait, TASK_INTERRUPTIBLE);
a68cf983 MN	152	set_bit(prio, spu_prio->bitmap);
2a911f0b	153	}
5110459f	154
a68cf983 MN	155	static inline void spu_del_wq(wait_queue_head_t * wq, wait_queue_t * wait,
a68cf983 MN	156	int prio)
2a911f0b	157	{
a68cf983	158	u64 flags;
2a911f0b	159
a68cf983 MN	160	__set_current_state(TASK_RUNNING);
	161
	162	spin_lock_irqsave(&wq->lock, flags);
	163
	164	remove_wait_queue_locked(wq, wait);
	165	if (list_empty(&wq->task_list))
	166	clear_bit(prio, spu_prio->bitmap);
	167
	168	spin_unlock_irqrestore(&wq->lock, flags);
8b3d6663 AB	169	}
8b3d6663 AB	170
a68cf983	171	static void spu_prio_wait(struct spu_context *ctx, u64 flags)
8b3d6663	172	{
a68cf983 MN	173	int prio = current->prio;
	174	wait_queue_head_t *wq = &spu_prio->waitq[prio];
	175	DEFINE_WAIT(wait);
8b3d6663	176
a68cf983 MN	177	if (ctx->spu)
	178	return;
	179
	180	spu_add_wq(wq, &wait, prio);
	181
	182	if (!signal_pending(current)) {
	183	up_write(&ctx->state_sema);
	184	pr_debug("%s: pid=%d prio=%d\n", __FUNCTION__,
	185	current->pid, current->prio);
	186	schedule();
	187	down_write(&ctx->state_sema);
8b3d6663	188	}
a68cf983 MN	189
a68cf983 MN	190	spu_del_wq(wq, &wait, prio);
8b3d6663 AB	191	}
8b3d6663 AB	192
a68cf983	193	static void spu_prio_wakeup(void)
8b3d6663	194	{
a68cf983 MN	195	int best = sched_find_first_bit(spu_prio->bitmap);
	196	if (best < MAX_PRIO) {
	197	wait_queue_head_t *wq = &spu_prio->waitq[best];
	198	wake_up_interruptible_nr(wq, 1);
	199	}
8b3d6663 AB	200	}
	201
	202	static int get_active_spu(struct spu *spu)
	203	{
a68cf983	204	int node = spu->node;
8b3d6663 AB	205	struct spu *tmp;
	206	int rc = 0;
	207
a68cf983 MN	208	mutex_lock(&spu_prio->active_mutex[node]);
a68cf983 MN	209	list_for_each_entry(tmp, &spu_prio->active_list[node], list) {
8b3d6663	210	if (tmp == spu) {
a68cf983	211	list_del_init(&spu->list);
8b3d6663 AB	212	rc = 1;
	213	break;
	214	}
	215	}
a68cf983	216	mutex_unlock(&spu_prio->active_mutex[node]);
8b3d6663 AB	217	return rc;
	218	}
	219
	220	static void put_active_spu(struct spu *spu)
	221	{
a68cf983 MN	222	int node = spu->node;
	223
	224	mutex_lock(&spu_prio->active_mutex[node]);
	225	list_add_tail(&spu->list, &spu_prio->active_list[node]);
	226	mutex_unlock(&spu_prio->active_mutex[node]);
	227	}
	228
	229	static struct spu spu_get_idle(struct spu_context ctx, u64 flags)
	230	{
	231	struct spu *spu = NULL;
	232	int node = cpu_to_node(raw_smp_processor_id());
	233	int n;
	234
	235	for (n = 0; n < MAX_NUMNODES; n++, node++) {
	236	node = (node < MAX_NUMNODES) ? node : 0;
	237	if (!node_allowed(node))
	238	continue;
	239	spu = spu_alloc_node(node);
	240	if (spu)
	241	break;
	242	}
	243	return spu;
	244	}
8b3d6663	245
a68cf983 MN	246	static inline struct spu spu_get(struct spu_context ctx, u64 flags)
	247	{
	248	/* Future: spu_get_idle() if possible,
	249	* otherwise try to preempt an active
	250	* context.
	251	*/
	252	return spu_get_idle(ctx, flags);
8b3d6663 AB	253	}
8b3d6663 AB	254
a68cf983 MN	255	/* The three externally callable interfaces
a68cf983 MN	256	* for the scheduler begin here.
8b3d6663	257	*
a68cf983 MN	258	* spu_activate - bind a context to SPU, waiting as needed.
	259	* spu_deactivate - unbind a context from its SPU.
	260	* spu_yield - yield an SPU if others are waiting.
8b3d6663 AB	261	*/
	262
	263	int spu_activate(struct spu_context *ctx, u64 flags)
	264	{
	265	struct spu *spu;
a68cf983	266	int ret = 0;
8b3d6663	267
a68cf983 MN	268	for (;;) {
	269	if (ctx->spu)
	270	return 0;
	271	spu = spu_get(ctx, flags);
	272	if (spu != NULL) {
	273	if (ctx->spu != NULL) {
	274	spu_free(spu);
	275	spu_prio_wakeup();
	276	break;
	277	}
	278	bind_context(spu, ctx);
	279	put_active_spu(spu);
	280	break;
	281	}
	282	spu_prio_wait(ctx, flags);
	283	if (signal_pending(current)) {
	284	ret = -ERESTARTSYS;
	285	spu_prio_wakeup();
	286	break;
	287	}
	288	}
	289	return ret;
8b3d6663 AB	290	}
	291
	292	void spu_deactivate(struct spu_context *ctx)
	293	{
	294	struct spu *spu;
	295	int needs_idle;
	296
	297	spu = ctx->spu;
	298	if (!spu)
	299	return;
	300	needs_idle = get_active_spu(spu);
	301	unbind_context(spu, ctx);
a68cf983 MN	302	if (needs_idle) {
	303	spu_free(spu);
	304	spu_prio_wakeup();
	305	}
8b3d6663 AB	306	}
	307
	308	void spu_yield(struct spu_context *ctx)
	309	{
	310	struct spu *spu;
5110459f	311	int need_yield = 0;
8b3d6663	312
a68cf983 MN	313	if (down_write_trylock(&ctx->state_sema)) {
	314	if ((spu = ctx->spu) != NULL) {
	315	int best = sched_find_first_bit(spu_prio->bitmap);
	316	if (best < MAX_PRIO) {
	317	pr_debug("%s: yielding SPU %d NODE %d\n",
	318	__FUNCTION__, spu->number, spu->node);
	319	spu_deactivate(ctx);
a68cf983 MN	320	need_yield = 1;
	321	} else {
	322	spu->prio = MAX_PRIO;
	323	}
	324	}
	325	up_write(&ctx->state_sema);
8b3d6663	326	}
5110459f AB	327	if (unlikely(need_yield))
5110459f AB	328	yield();
8b3d6663 AB	329	}
	330
	331	int __init spu_sched_init(void)
	332	{
8b3d6663 AB	333	int i;
8b3d6663 AB	334
a68cf983 MN	335	spu_prio = kzalloc(sizeof(struct spu_prio_array), GFP_KERNEL);
	336	if (!spu_prio) {
	337	printk(KERN_WARNING "%s: Unable to allocate priority queue.\n",
8b3d6663 AB	338	__FUNCTION__);
	339	return 1;
	340	}
8b3d6663	341	for (i = 0; i < MAX_PRIO; i++) {
a68cf983 MN	342	init_waitqueue_head(&spu_prio->waitq[i]);
a68cf983 MN	343	__clear_bit(i, spu_prio->bitmap);
8b3d6663	344	}
a68cf983 MN	345	__set_bit(MAX_PRIO, spu_prio->bitmap);
	346	for (i = 0; i < MAX_NUMNODES; i++) {
	347	mutex_init(&spu_prio->active_mutex[i]);
	348	INIT_LIST_HEAD(&spu_prio->active_list[i]);
8b3d6663 AB	349	}
	350	return 0;
	351	}
	352
	353	void __exit spu_sched_exit(void)
	354	{
a68cf983 MN	355	struct spu spu, tmp;
	356	int node;
	357
	358	for (node = 0; node < MAX_NUMNODES; node++) {
	359	mutex_lock(&spu_prio->active_mutex[node]);
	360	list_for_each_entry_safe(spu, tmp, &spu_prio->active_list[node],
	361	list) {
	362	list_del_init(&spu->list);
	363	spu_free(spu);
	364	}
	365	mutex_unlock(&spu_prio->active_mutex[node]);
8b3d6663	366	}
a68cf983	367	kfree(spu_prio);
8b3d6663	368	}