#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 <linux/kthread.h>
+#include <linux/pid_namespace.h>
+#include <linux/proc_fs.h>
+#include <linux/seq_file.h>
#include <asm/io.h>
#include <asm/mmu_context.h>
#include <asm/spu_priv1.h>
#include "spufs.h"
-#define SPU_TIMESLICE (HZ)
-
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;
};
+static unsigned long spu_avenrun[3];
static struct spu_prio_array *spu_prio;
-static struct workqueue_struct *spu_sched_wq;
+static struct task_struct *spusched_task;
+static struct timer_list spusched_timer;
-static inline int node_allowed(int node)
-{
- cpumask_t mask;
+/*
+ * Priority of a normal, non-rt, non-niced'd process (aka nice level 0).
+ */
+#define NORMAL_PRIO 120
- if (!nr_cpus_node(node))
- return 0;
- mask = node_to_cpumask(node);
- if (!cpus_intersects(mask, current->cpus_allowed))
- return 0;
- return 1;
+/*
+ * Frequency of the spu scheduler tick. By default we do one SPU scheduler
+ * tick for every 10 CPU scheduler ticks.
+ */
+#define SPUSCHED_TICK (10)
+
+/*
+ * These are the 'tuning knobs' of the scheduler:
+ *
+ * Minimum timeslice is 5 msecs (or 1 spu scheduler tick, whichever is
+ * larger), default timeslice is 100 msecs, maximum timeslice is 800 msecs.
+ */
+#define MIN_SPU_TIMESLICE max(5 * HZ / (1000 * SPUSCHED_TICK), 1)
+#define DEF_SPU_TIMESLICE (100 * HZ / (1000 * SPUSCHED_TICK))
+
+#define MAX_USER_PRIO (MAX_PRIO - MAX_RT_PRIO)
+#define SCALE_PRIO(x, prio) \
+ max(x * (MAX_PRIO - prio) / (MAX_USER_PRIO / 2), MIN_SPU_TIMESLICE)
+
+/*
+ * scale user-nice values [ -20 ... 0 ... 19 ] to time slice values:
+ * [800ms ... 100ms ... 5ms]
+ *
+ * The higher a thread's priority, the bigger timeslices
+ * it gets during one round of execution. But even the lowest
+ * priority thread gets MIN_TIMESLICE worth of execution time.
+ */
+void spu_set_timeslice(struct spu_context *ctx)
+{
+ if (ctx->prio < NORMAL_PRIO)
+ ctx->time_slice = SCALE_PRIO(DEF_SPU_TIMESLICE * 4, ctx->prio);
+ else
+ ctx->time_slice = SCALE_PRIO(DEF_SPU_TIMESLICE, ctx->prio);
}
-void spu_start_tick(struct spu_context *ctx)
+/*
+ * Update scheduling information from the owning thread.
+ */
+void __spu_update_sched_info(struct spu_context *ctx)
{
- if (ctx->policy == SCHED_RR)
- queue_delayed_work(spu_sched_wq, &ctx->sched_work, SPU_TIMESLICE);
+ /*
+ * 32-Bit assignment are atomic on powerpc, and we don't care about
+ * memory ordering here because retriving the controlling thread is
+ * per defintion racy.
+ */
+ ctx->tid = current->pid;
+
+ /*
+ * We do our own priority calculations, so we normally want
+ * ->static_prio to start with. Unfortunately thies field
+ * contains junk for threads with a realtime scheduling
+ * policy so we have to look at ->prio in this case.
+ */
+ if (rt_prio(current->prio))
+ ctx->prio = current->prio;
+ else
+ ctx->prio = current->static_prio;
+ ctx->policy = current->policy;
+
+ /*
+ * A lot of places that don't hold active_mutex poke into
+ * cpus_allowed, including grab_runnable_context which
+ * already holds the runq_lock. So abuse runq_lock
+ * to protect this field aswell.
+ */
+ spin_lock(&spu_prio->runq_lock);
+ ctx->cpus_allowed = current->cpus_allowed;
+ spin_unlock(&spu_prio->runq_lock);
}
-void spu_stop_tick(struct spu_context *ctx)
+void spu_update_sched_info(struct spu_context *ctx)
{
- if (ctx->policy == SCHED_RR)
- cancel_delayed_work(&ctx->sched_work);
+ int node = ctx->spu->node;
+
+ mutex_lock(&spu_prio->active_mutex[node]);
+ __spu_update_sched_info(ctx);
+ mutex_unlock(&spu_prio->active_mutex[node]);
}
-void spu_sched_tick(struct work_struct *work)
+static int __node_allowed(struct spu_context *ctx, int node)
{
- struct spu_context *ctx =
- container_of(work, struct spu_context, sched_work.work);
- struct spu *spu;
- int rearm = 1;
+ if (nr_cpus_node(node)) {
+ cpumask_t mask = node_to_cpumask(node);
- mutex_lock(&ctx->state_mutex);
- spu = ctx->spu;
- if (spu) {
- int best = sched_find_first_bit(spu_prio->bitmap);
- if (best <= ctx->prio) {
- spu_deactivate(ctx);
- rearm = 0;
- }
+ if (cpus_intersects(mask, ctx->cpus_allowed))
+ return 1;
}
- mutex_unlock(&ctx->state_mutex);
- if (rearm)
- spu_start_tick(ctx);
+ return 0;
+}
+
+static int node_allowed(struct spu_context *ctx, int node)
+{
+ int rval;
+
+ spin_lock(&spu_prio->runq_lock);
+ rval = __node_allowed(ctx, node);
+ spin_unlock(&spu_prio->runq_lock);
+
+ return rval;
}
/**
*/
static void spu_add_to_active_list(struct spu *spu)
{
- mutex_lock(&spu_prio->active_mutex[spu->node]);
- list_add_tail(&spu->list, &spu_prio->active_list[spu->node]);
- mutex_unlock(&spu_prio->active_mutex[spu->node]);
+ 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 void __spu_remove_from_active_list(struct spu *spu)
+{
+ list_del_init(&spu->list);
+ spu_prio->nr_active[spu->node]--;
}
/**
int node = spu->node;
mutex_lock(&spu_prio->active_mutex[node]);
- list_del_init(&spu->list);
+ __spu_remove_from_active_list(spu);
mutex_unlock(&spu_prio->active_mutex[node]);
}
{
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;
spu->timestamp = jiffies;
spu_cpu_affinity_set(spu, raw_smp_processor_id());
spu_switch_notify(spu, ctx);
- spu_add_to_active_list(spu);
ctx->state = SPU_STATE_RUNNABLE;
}
pr_debug("%s: unbind pid=%d SPU=%d NODE=%d\n", __FUNCTION__,
spu->pid, spu->number, spu->node);
- spu_remove_from_active_list(spu);
spu_switch_notify(spu, NULL);
spu_unmap_mappings(ctx);
spu_save(&ctx->csa, spu);
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);
}
/**
* spu_add_to_rq - add a context to the runqueue
* @ctx: context to add
*/
-static void spu_add_to_rq(struct spu_context *ctx)
+static void __spu_add_to_rq(struct spu_context *ctx)
{
- spin_lock(&spu_prio->runq_lock);
- list_add_tail(&ctx->rq, &spu_prio->runq[ctx->prio]);
- set_bit(ctx->prio, spu_prio->bitmap);
- spin_unlock(&spu_prio->runq_lock);
-}
+ int prio = ctx->prio;
-/**
- * spu_del_from_rq - remove a context from the runqueue
- * @ctx: context to remove
- */
-static void spu_del_from_rq(struct spu_context *ctx)
-{
- spin_lock(&spu_prio->runq_lock);
- list_del_init(&ctx->rq);
- if (list_empty(&spu_prio->runq[ctx->prio]))
- clear_bit(ctx->prio, spu_prio->bitmap);
- spin_unlock(&spu_prio->runq_lock);
+ list_add_tail(&ctx->rq, &spu_prio->runq[prio]);
+ set_bit(prio, spu_prio->bitmap);
+ if (!spu_prio->nr_waiting++)
+ __mod_timer(&spusched_timer, jiffies + SPUSCHED_TICK);
}
-/**
- * spu_grab_context - remove one context from the runqueue
- * @prio: priority of the context to be removed
- *
- * This function removes one context from the runqueue for priority @prio.
- * If there is more than one context with the given priority the first
- * task on the runqueue will be taken.
- *
- * Returns the spu_context it just removed.
- *
- * Must be called with spu_prio->runq_lock held.
- */
-static struct spu_context *spu_grab_context(int prio)
+static void __spu_del_from_rq(struct spu_context *ctx)
{
- struct list_head *rq = &spu_prio->runq[prio];
+ int prio = ctx->prio;
+
+ if (!list_empty(&ctx->rq)) {
+ if (!--spu_prio->nr_waiting)
+ del_timer(&spusched_timer);
+ list_del_init(&ctx->rq);
- if (list_empty(rq))
- return NULL;
- return list_entry(rq->next, struct spu_context, rq);
+ if (list_empty(&spu_prio->runq[prio]))
+ clear_bit(prio, spu_prio->bitmap);
+ }
}
static void spu_prio_wait(struct spu_context *ctx)
{
DEFINE_WAIT(wait);
+ spin_lock(&spu_prio->runq_lock);
prepare_to_wait_exclusive(&ctx->stop_wq, &wait, TASK_INTERRUPTIBLE);
if (!signal_pending(current)) {
+ __spu_add_to_rq(ctx);
+ spin_unlock(&spu_prio->runq_lock);
mutex_unlock(&ctx->state_mutex);
schedule();
mutex_lock(&ctx->state_mutex);
+ spin_lock(&spu_prio->runq_lock);
+ __spu_del_from_rq(ctx);
}
+ spin_unlock(&spu_prio->runq_lock);
__set_current_state(TASK_RUNNING);
remove_wait_queue(&ctx->stop_wq, &wait);
}
-/**
- * spu_reschedule - try to find a runnable context for a spu
- * @spu: spu available
- *
- * This function is called whenever a spu becomes idle. It looks for the
- * most suitable runnable spu context and schedules it for execution.
- */
-static void spu_reschedule(struct spu *spu)
-{
- int best;
-
- spu_free(spu);
-
- spin_lock(&spu_prio->runq_lock);
- best = sched_find_first_bit(spu_prio->bitmap);
- if (best < MAX_PRIO) {
- struct spu_context *ctx = spu_grab_context(best);
- if (ctx)
- wake_up(&ctx->stop_wq);
- }
- spin_unlock(&spu_prio->runq_lock);
-}
-
static struct spu *spu_get_idle(struct spu_context *ctx)
{
struct spu *spu = NULL;
for (n = 0; n < MAX_NUMNODES; n++, node++) {
node = (node < MAX_NUMNODES) ? node : 0;
- if (!node_allowed(node))
+ if (!node_allowed(ctx, node))
continue;
spu = spu_alloc_node(node);
if (spu)
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(node))
+ if (!node_allowed(ctx, node))
continue;
mutex_lock(&spu_prio->active_mutex[node]);
list_for_each_entry(spu, &spu_prio->active_list[node], list) {
struct spu_context *tmp = spu->ctx;
- if (tmp->rt_priority < ctx->rt_priority &&
- (!victim || tmp->rt_priority < victim->rt_priority))
+ if (tmp->prio > ctx->prio &&
+ (!victim || tmp->prio > victim->prio))
victim = spu->ctx;
}
mutex_unlock(&spu_prio->active_mutex[node]);
victim = NULL;
goto restart;
}
+ 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
+ * manually to ensure this context gets put on the
+ * runqueue again ASAP.
+ */
+ wake_up(&victim->stop_wq);
return spu;
}
}
* @ctx: spu context to schedule
* @flags: flags (currently ignored)
*
- * Tries to find a free spu to run @ctx. If no free spu is availble
+ * Tries to find a free spu to run @ctx. If no free spu is available
* add the context to the runqueue so it gets woken up once an spu
* is available.
*/
int spu_activate(struct spu_context *ctx, unsigned long flags)
{
+ spuctx_switch_state(ctx, SPUCTX_UTIL_SYSTEM);
if (ctx->spu)
return 0;
* If this is a realtime thread we try to get it running by
* preempting a lower priority thread.
*/
- if (!spu && ctx->rt_priority)
+ if (!spu && rt_prio(ctx->prio))
spu = find_victim(ctx);
if (spu) {
spu_bind_context(spu, ctx);
+ spu_add_to_active_list(spu);
return 0;
}
- spu_add_to_rq(ctx);
spu_prio_wait(ctx);
- spu_del_from_rq(ctx);
} while (!signal_pending(current));
return -ERESTARTSYS;
}
+/**
+ * grab_runnable_context - try to find a runnable context
+ *
+ * Remove the highest priority context on the runqueue and return it
+ * to the caller. Returns %NULL if no runnable context was found.
+ */
+static struct spu_context *grab_runnable_context(int prio, int node)
+{
+ struct spu_context *ctx;
+ int best;
+
+ spin_lock(&spu_prio->runq_lock);
+ best = sched_find_first_bit(spu_prio->bitmap);
+ while (best < prio) {
+ struct list_head *rq = &spu_prio->runq[best];
+
+ list_for_each_entry(ctx, rq, rq) {
+ /* XXX(hch): check for affinity here aswell */
+ if (__node_allowed(ctx, node)) {
+ __spu_del_from_rq(ctx);
+ goto found;
+ }
+ }
+ best++;
+ }
+ ctx = NULL;
+ found:
+ spin_unlock(&spu_prio->runq_lock);
+ return ctx;
+}
+
+static int __spu_deactivate(struct spu_context *ctx, int force, int max_prio)
+{
+ struct spu *spu = ctx->spu;
+ struct spu_context *new = NULL;
+
+ if (spu) {
+ new = grab_runnable_context(max_prio, spu->node);
+ 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);
+ }
+
+ }
+
+ return new != NULL;
+}
+
/**
* spu_deactivate - unbind a context from it's physical spu
* @ctx: spu context to unbind
*/
void spu_deactivate(struct spu_context *ctx)
{
- struct spu *spu = ctx->spu;
-
- if (spu) {
- spu_unbind_context(spu, ctx);
- spu_reschedule(spu);
+ /*
+ * 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);
}
/**
*/
void spu_yield(struct spu_context *ctx)
{
- struct spu *spu;
- int need_yield = 0;
+ 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);
+ mutex_unlock(&ctx->state_mutex);
+ }
+}
+
+static void spusched_tick(struct spu_context *ctx)
+{
+ if (ctx->flags & SPU_CREATE_NOSCHED)
+ return;
+ if (ctx->policy == SCHED_FIFO)
+ return;
+
+ if (--ctx->time_slice)
+ return;
+ /*
+ * Unfortunately active_mutex ranks outside of state_mutex, so
+ * we have to trylock here. If we fail give the context another
+ * tick and try again.
+ */
if (mutex_trylock(&ctx->state_mutex)) {
- 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;
- }
+ struct spu *spu = ctx->spu;
+ struct spu_context *new;
+
+ 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_free(spu);
+ wake_up(&new->stop_wq);
+ /*
+ * We need to break out of the wait loop in
+ * spu_run manually to ensure this context
+ * gets put on the runqueue again ASAP.
+ */
+ wake_up(&ctx->stop_wq);
}
+ spu_set_timeslice(ctx);
mutex_unlock(&ctx->state_mutex);
+ } else {
+ ctx->time_slice++;
}
- if (unlikely(need_yield))
- yield();
}
-int __init spu_sched_init(void)
+/**
+ * count_active_contexts - count nr of active tasks
+ *
+ * Return the number of tasks currently running or waiting to run.
+ *
+ * Note that we don't take runq_lock / active_mutex here. Reading
+ * a single 32bit value is atomic on powerpc, and we don't care
+ * about memory ordering issues here.
+ */
+static unsigned long count_active_contexts(void)
{
- int i;
+ int nr_active = 0, node;
- spu_sched_wq = create_singlethread_workqueue("spusched");
- if (!spu_sched_wq)
- return 1;
+ for (node = 0; node < MAX_NUMNODES; node++)
+ nr_active += spu_prio->nr_active[node];
+ nr_active += spu_prio->nr_waiting;
- spu_prio = kzalloc(sizeof(struct spu_prio_array), GFP_KERNEL);
- if (!spu_prio) {
- printk(KERN_WARNING "%s: Unable to allocate priority queue.\n",
- __FUNCTION__);
- destroy_workqueue(spu_sched_wq);
- return 1;
+ return nr_active;
+}
+
+/**
+ * spu_calc_load - given tick count, update the avenrun load estimates.
+ * @tick: tick count
+ *
+ * No locking against reading these values from userspace, as for
+ * the CPU loadavg code.
+ */
+static void spu_calc_load(unsigned long ticks)
+{
+ unsigned long active_tasks; /* fixed-point */
+ static int count = LOAD_FREQ;
+
+ count -= ticks;
+
+ if (unlikely(count < 0)) {
+ active_tasks = count_active_contexts() * FIXED_1;
+ do {
+ CALC_LOAD(spu_avenrun[0], EXP_1, active_tasks);
+ CALC_LOAD(spu_avenrun[1], EXP_5, active_tasks);
+ CALC_LOAD(spu_avenrun[2], EXP_15, active_tasks);
+ count += LOAD_FREQ;
+ } while (count < 0);
}
+}
+
+static void spusched_wake(unsigned long data)
+{
+ mod_timer(&spusched_timer, jiffies + SPUSCHED_TICK);
+ wake_up_process(spusched_task);
+ spu_calc_load(SPUSCHED_TICK);
+}
+
+static int spusched_thread(void *unused)
+{
+ struct spu *spu, *next;
+ 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]);
+ }
+ }
+
+ return 0;
+}
+
+#define LOAD_INT(x) ((x) >> FSHIFT)
+#define LOAD_FRAC(x) LOAD_INT(((x) & (FIXED_1-1)) * 100)
+
+static int show_spu_loadavg(struct seq_file *s, void *private)
+{
+ int a, b, c;
+
+ a = spu_avenrun[0] + (FIXED_1/200);
+ b = spu_avenrun[1] + (FIXED_1/200);
+ c = spu_avenrun[2] + (FIXED_1/200);
+
+ /*
+ * Note that last_pid doesn't really make much sense for the
+ * SPU loadavg (it even seems very odd on the CPU side..),
+ * but we include it here to have a 100% compatible interface.
+ */
+ seq_printf(s, "%d.%02d %d.%02d %d.%02d %ld/%d %d\n",
+ LOAD_INT(a), LOAD_FRAC(a),
+ LOAD_INT(b), LOAD_FRAC(b),
+ LOAD_INT(c), LOAD_FRAC(c),
+ count_active_contexts(),
+ atomic_read(&nr_spu_contexts),
+ current->nsproxy->pid_ns->last_pid);
+ return 0;
+}
+
+static int spu_loadavg_open(struct inode *inode, struct file *file)
+{
+ return single_open(file, show_spu_loadavg, NULL);
+}
+
+static const struct file_operations spu_loadavg_fops = {
+ .open = spu_loadavg_open,
+ .read = seq_read,
+ .llseek = seq_lseek,
+ .release = single_release,
+};
+
+int __init spu_sched_init(void)
+{
+ struct proc_dir_entry *entry;
+ int err = -ENOMEM, i;
+
+ spu_prio = kzalloc(sizeof(struct spu_prio_array), GFP_KERNEL);
+ if (!spu_prio)
+ goto out;
+
for (i = 0; i < MAX_PRIO; i++) {
INIT_LIST_HEAD(&spu_prio->runq[i]);
__clear_bit(i, spu_prio->bitmap);
INIT_LIST_HEAD(&spu_prio->active_list[i]);
}
spin_lock_init(&spu_prio->runq_lock);
+
+ setup_timer(&spusched_timer, spusched_wake, 0);
+
+ spusched_task = kthread_run(spusched_thread, NULL, "spusched");
+ if (IS_ERR(spusched_task)) {
+ err = PTR_ERR(spusched_task);
+ goto out_free_spu_prio;
+ }
+
+ entry = create_proc_entry("spu_loadavg", 0, NULL);
+ if (!entry)
+ goto out_stop_kthread;
+ entry->proc_fops = &spu_loadavg_fops;
+
+ pr_debug("spusched: tick: %d, min ticks: %d, default ticks: %d\n",
+ SPUSCHED_TICK, MIN_SPU_TIMESLICE, DEF_SPU_TIMESLICE);
return 0;
+
+ out_stop_kthread:
+ kthread_stop(spusched_task);
+ out_free_spu_prio:
+ kfree(spu_prio);
+ out:
+ return err;
}
void __exit spu_sched_exit(void)
struct spu *spu, *tmp;
int node;
+ remove_proc_entry("spu_loadavg", NULL);
+
+ del_timer_sync(&spusched_timer);
+ 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],
mutex_unlock(&spu_prio->active_mutex[node]);
}
kfree(spu_prio);
- destroy_workqueue(spu_sched_wq);
}