[Cbe-oss-dev] [PATCH 14/21] spufs: implement /proc/spu_loadavg
Jeremy Kerr
jk at ozlabs.org
Fri Jun 29 10:58:02 EST 2007
From: Christoph Hellwig <hch at lst.de>
Provide load averange information for spu context. The format
is identical to /proc/loadavg, which is also where a lot of code
and concepts is borrowed from.
Signed-off-by: Christoph Hellwig <hch at lst.de>
Signed-off-by: Arnd Bergmann <arnd.bergmann at de.ibm.com>
Signed-off-by: Jeremy Kerr <jk at ozlabs.org>
---
arch/powerpc/platforms/cell/spufs/context.c | 7 ++
arch/powerpc/platforms/cell/spufs/sched.c | 127 +++++++++++++++++++++++++--
arch/powerpc/platforms/cell/spufs/spufs.h | 1 +
3 files changed, 127 insertions(+), 8 deletions(-)
diff --git a/arch/powerpc/platforms/cell/spufs/context.c b/arch/powerpc/platforms/cell/spufs/context.c
index 6ff2a75..f623d96 100644
--- a/arch/powerpc/platforms/cell/spufs/context.c
+++ b/arch/powerpc/platforms/cell/spufs/context.c
@@ -23,10 +23,14 @@
#include <linux/fs.h>
#include <linux/mm.h>
#include <linux/slab.h>
+#include <asm/atomic.h>
#include <asm/spu.h>
#include <asm/spu_csa.h>
#include "spufs.h"
+
+atomic_t nr_spu_contexts = ATOMIC_INIT(0);
+
struct spu_context *alloc_spu_context(struct spu_gang *gang)
{
struct spu_context *ctx;
@@ -55,6 +59,8 @@ struct spu_context *alloc_spu_context(struct spu_gang *gang)
spu_gang_add_ctx(gang, ctx);
ctx->cpus_allowed = current->cpus_allowed;
spu_set_timeslice(ctx);
+
+ atomic_inc(&nr_spu_contexts);
goto out;
out_free:
kfree(ctx);
@@ -74,6 +80,7 @@ void destroy_spu_context(struct kref *kref)
if (ctx->gang)
spu_gang_remove_ctx(ctx->gang, ctx);
BUG_ON(!list_empty(&ctx->rq));
+ atomic_dec(&nr_spu_contexts);
kfree(ctx);
}
diff --git a/arch/powerpc/platforms/cell/spufs/sched.c b/arch/powerpc/platforms/cell/spufs/sched.c
index 5400675..9fc0930 100644
--- a/arch/powerpc/platforms/cell/spufs/sched.c
+++ b/arch/powerpc/platforms/cell/spufs/sched.c
@@ -36,6 +36,9 @@
#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>
@@ -50,8 +53,11 @@ struct spu_prio_array {
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 task_struct *spusched_task;
static struct timer_list spusched_timer;
@@ -169,14 +175,18 @@ static int node_allowed(struct spu_context *ctx, int node)
*/
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]--;
}
/**
@@ -275,6 +285,7 @@ static void __spu_add_to_rq(struct spu_context *ctx)
{
int prio = ctx->prio;
+ spu_prio->nr_waiting++;
list_add_tail(&ctx->rq, &spu_prio->runq[prio]);
set_bit(prio, spu_prio->bitmap);
}
@@ -283,8 +294,10 @@ static void __spu_del_from_rq(struct spu_context *ctx)
{
int prio = ctx->prio;
- if (!list_empty(&ctx->rq))
+ if (!list_empty(&ctx->rq)) {
list_del_init(&ctx->rq);
+ spu_prio->nr_waiting--;
+ }
if (list_empty(&spu_prio->runq[prio]))
clear_bit(prio, spu_prio->bitmap);
}
@@ -567,10 +580,56 @@ static void spusched_tick(struct spu_context *ctx)
}
}
+/**
+ * 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 nr_active = 0, node;
+
+ for (node = 0; node < MAX_NUMNODES; node++)
+ nr_active += spu_prio->nr_active[node];
+ nr_active += spu_prio->nr_waiting;
+
+ 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)
@@ -598,13 +657,52 @@ static int spusched_thread(void *unused)
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)
{
- int i;
+ struct proc_dir_entry *entry;
+ int err = -ENOMEM, i;
spu_prio = kzalloc(sizeof(struct spu_prio_array), GFP_KERNEL);
if (!spu_prio)
- return -ENOMEM;
+ goto out;
for (i = 0; i < MAX_PRIO; i++) {
INIT_LIST_HEAD(&spu_prio->runq[i]);
@@ -619,14 +717,25 @@ int __init spu_sched_init(void)
spusched_task = kthread_run(spusched_thread, NULL, "spusched");
if (IS_ERR(spusched_task)) {
- kfree(spu_prio);
- return PTR_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)
@@ -634,6 +743,8 @@ void __exit spu_sched_exit(void)
struct spu *spu, *tmp;
int node;
+ remove_proc_entry("spu_loadavg", NULL);
+
kthread_stop(spusched_task);
for (node = 0; node < MAX_NUMNODES; node++) {
diff --git a/arch/powerpc/platforms/cell/spufs/spufs.h b/arch/powerpc/platforms/cell/spufs/spufs.h
index 8ff16b4..7f5d0b2 100644
--- a/arch/powerpc/platforms/cell/spufs/spufs.h
+++ b/arch/powerpc/platforms/cell/spufs/spufs.h
@@ -176,6 +176,7 @@ void spu_gang_add_ctx(struct spu_gang *gang, struct spu_context *ctx);
int spufs_handle_class1(struct spu_context *ctx);
/* context management */
+extern atomic_t nr_spu_contexts;
static inline void spu_acquire(struct spu_context *ctx)
{
mutex_lock(&ctx->state_mutex);
--
1.5.0.rc4.g85b1
More information about the cbe-oss-dev
mailing list