[POWERPC] spufs: Add spu stats in sysfs

Export spu statistics in sysfs.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Arnd Bergmann <arnd.bergmann@de.ibm.com>
Signed-off-by: Jeremy Kerr <jk@ozlabs.org>
Signed-off-by: Paul Mackerras <paulus@samba.org>

arch/powerpc/platforms/cell/spu_base.c

@@ -585,6 +585,9 @@ static int __init create_spu(void *data)
 	spin_unlock_irqrestore(&spu_list_lock, flags);
 	mutex_unlock(&spu_mutex);
 
+	spu->stats.utilization_state = SPU_UTIL_IDLE;
+	spu->stats.tstamp = jiffies;
+
 	goto out;
 
 out_free_irqs:
@@ -597,6 +600,45 @@ out:
 	return ret;
 }
 
+static const char *spu_state_names[] = {
+	"user", "system", "iowait", "idle"
+};
+
+static unsigned long long spu_acct_time(struct spu *spu,
+		enum spu_utilization_state state)
+{
+	unsigned long long time = spu->stats.times[state];
+
+	if (spu->stats.utilization_state == state)
+		time += jiffies - spu->stats.tstamp;
+
+	return jiffies_to_msecs(time);
+}
+
+
+static ssize_t spu_stat_show(struct sys_device *sysdev, char *buf)
+{
+	struct spu *spu = container_of(sysdev, struct spu, sysdev);
+
+	return sprintf(buf, "%s %llu %llu %llu %llu "
+		      "%llu %llu %llu %llu %llu %llu %llu %llu\n",
+		spu_state_names[spu->stats.utilization_state],
+		spu_acct_time(spu, SPU_UTIL_USER),
+		spu_acct_time(spu, SPU_UTIL_SYSTEM),
+		spu_acct_time(spu, SPU_UTIL_IOWAIT),
+		spu_acct_time(spu, SPU_UTIL_IDLE),
+		spu->stats.vol_ctx_switch,
+		spu->stats.invol_ctx_switch,
+		spu->stats.slb_flt,
+		spu->stats.hash_flt,
+		spu->stats.min_flt,
+		spu->stats.maj_flt,
+		spu->stats.class2_intr,
+		spu->stats.libassist);
+}
+
+static SYSDEV_ATTR(stat, 0644, spu_stat_show, NULL);
+
 static int __init init_spu_base(void)
 {
 	int i, ret = 0;
@@ -622,6 +664,8 @@ static int __init init_spu_base(void)
 
 	xmon_register_spus(&spu_full_list);
 
+	spu_add_sysdev_attr(&attr_stat);
+
 	return 0;
 
  out_unregister_sysdev_class:

arch/powerpc/platforms/cell/spufs/fault.c

@@ -187,6 +187,10 @@ int spufs_handle_class1(struct spu_context *ctx)
 		dsisr, ctx->state);
 
 	ctx->stats.hash_flt++;
+	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);
@@ -212,6 +216,12 @@ int spufs_handle_class1(struct spu_context *ctx)
 			ctx->stats.min_flt++;
 		else
 			ctx->stats.maj_flt++;
+		if (ctx->state == SPU_STATE_RUNNABLE) {
+			if (flt == VM_FAULT_MINOR)
+				ctx->spu->stats.min_flt++;
+			else
+				ctx->spu->stats.maj_flt++;
+		}
 
 		if (ctx->spu)
 			ctx->ops->restart_dma(ctx);

arch/powerpc/platforms/cell/spufs/run.c

@@ -352,7 +352,8 @@ long spufs_run_spu(struct file *file, struct spu_context *ctx,
 				       SPU_STATUS_SINGLE_STEP)));
 
 	if ((status & SPU_STATUS_STOPPED_BY_STOP) &&
-	    (((status >> SPU_STOP_STATUS_SHIFT) & 0x3f00) == 0x2100))
+	    (((status >> SPU_STOP_STATUS_SHIFT) & 0x3f00) == 0x2100) &&
+	    (ctx->state == SPU_STATE_RUNNABLE))
 		ctx->stats.libassist++;
 
 	ctx->ops->master_stop(ctx);

arch/powerpc/platforms/cell/spufs/sched.c

@@ -251,6 +251,7 @@ 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);
 }
 
 /**
@@ -263,6 +264,8 @@ 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);
 
+	spu_switch_state(spu, SPU_UTIL_IDLE);
+
 	spu_switch_notify(spu, NULL);
 	spu_unmap_mappings(ctx);
 	spu_save(&ctx->csa, spu);
@@ -426,6 +429,7 @@ static struct spu *find_victim(struct spu_context *ctx)
 			spu_remove_from_active_list(spu);
 			spu_unbind_context(spu, victim);
 			victim->stats.invol_ctx_switch++;
+			spu->stats.invol_ctx_switch++;
 			mutex_unlock(&victim->state_mutex);
 			/*
 			 * We need to break out of the wait loop in spu_run
@@ -526,6 +530,7 @@ static int __spu_deactivate(struct spu_context *ctx, int force, int max_prio)
 			spu_remove_from_active_list(spu);
 			spu_unbind_context(spu, ctx);
 			ctx->stats.vol_ctx_switch++;
+			spu->stats.vol_ctx_switch++;
 			spu_free(spu);
 			if (new)
 				wake_up(&new->stop_wq);
@@ -572,8 +577,10 @@ void spu_yield(struct spu_context *ctx)
 		mutex_lock(&ctx->state_mutex);
 		if (__spu_deactivate(ctx, 0, MAX_PRIO))
 			spuctx_switch_state(ctx, SPUCTX_UTIL_USER);
-		else
+		else {
 			spuctx_switch_state(ctx, SPUCTX_UTIL_LOADED);
+			spu_switch_state(ctx->spu, SPU_UTIL_USER);
+		}
 		mutex_unlock(&ctx->state_mutex);
 	}
 }
@@ -603,6 +610,7 @@ static void spusched_tick(struct spu_context *ctx)
 			__spu_remove_from_active_list(spu);
 			spu_unbind_context(spu, ctx);
 			ctx->stats.invol_ctx_switch++;
+			spu->stats.invol_ctx_switch++;
 			spu_free(spu);
 			wake_up(&new->stop_wq);
 			/*

arch/powerpc/platforms/cell/spufs/spufs.h

@@ -307,4 +307,17 @@ static inline void spuctx_switch_state(struct spu_context *ctx,
 	}
 }
 
+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;
+
+		spu->stats.times[spu->stats.utilization_state] +=
+				 curtime - spu->stats.tstamp;
+		spu->stats.tstamp = curtime;
+		spu->stats.utilization_state = new_state;
+	}
+}
+
 #endif

include/asm-powerpc/spu.h

@@ -106,6 +106,14 @@ struct spu_context;
 struct spu_runqueue;
 struct device_node;
 
+enum spu_utilization_state {
+	SPU_UTIL_SYSTEM,
+	SPU_UTIL_USER,
+	SPU_UTIL_IOWAIT,
+	SPU_UTIL_IDLE,
+	SPU_UTIL_MAX
+};
+
 struct spu {
 	const char *name;
 	unsigned long local_store_phys;
@@ -159,8 +167,17 @@ struct spu {
 
 	struct {
 		/* protected by interrupt reentrancy */
+		enum spu_utilization_state utilization_state;
+		unsigned long tstamp;		/* time of last ctx switch */
+		unsigned long times[SPU_UTIL_MAX];
+		unsigned long long vol_ctx_switch;
+		unsigned long long invol_ctx_switch;
+		unsigned long long min_flt;
+		unsigned long long maj_flt;
+		unsigned long long hash_flt;
 		unsigned long long slb_flt;
 		unsigned long long class2_intr;
+		unsigned long long libassist;
 	} stats;
 };