5 files changed, 151 insertions, 69 deletions

diff --git a/arch/powerpc/platforms/cell/spufs/file.c b/arch/powerpc/platforms/cell/spufs/file.c
index 010a51f59796..b73c369cc6f1 100644
--- a/arch/powerpc/platforms/cell/spufs/file.c
+++ b/arch/powerpc/platforms/cell/spufs/file.c
@@ -548,6 +548,11 @@ spufs_regs_read(struct file *file, char __user *buffer,
 	int ret;
 	struct spu_context *ctx = file->private_data;
 
+	/* pre-check for file position: if we'd return EOF, there's no point
+	 * causing a deschedule */
+	if (*pos >= sizeof(ctx->csa.lscsa->gprs))
+		return 0;
+
 	ret = spu_acquire_saved(ctx);
 	if (ret)
 		return ret;
@@ -2426,38 +2431,49 @@ static inline int spufs_switch_log_avail(struct spu_context *ctx)
 static int spufs_switch_log_open(struct inode *inode, struct file *file)
 {
 	struct spu_context *ctx = SPUFS_I(inode)->i_ctx;
+	int rc;
+
+	rc = spu_acquire(ctx);
+	if (rc)
+		return rc;
 
-	/*
-	 * We (ab-)use the mapping_lock here because it serves the similar
-	 * purpose for synchronizing open/close elsewhere.  Maybe it should
-	 * be renamed eventually.
-	 */
-	mutex_lock(&ctx->mapping_lock);
 	if (ctx->switch_log) {
-		spin_lock(&ctx->switch_log->lock);
-		ctx->switch_log->head = 0;
-		ctx->switch_log->tail = 0;
-		spin_unlock(&ctx->switch_log->lock);
-	} else {
-		/*
-		 * We allocate the switch log data structures on first open.
-		 * They will never be free because we assume a context will
-		 * be traced until it goes away.
-		 */
-		ctx->switch_log = kzalloc(sizeof(struct switch_log) +
-			SWITCH_LOG_BUFSIZE * sizeof(struct switch_log_entry),
-			GFP_KERNEL);
-		if (!ctx->switch_log)
-			goto out;
-		spin_lock_init(&ctx->switch_log->lock);
-		init_waitqueue_head(&ctx->switch_log->wait);
+		rc = -EBUSY;
+		goto out;
 	}
-	mutex_unlock(&ctx->mapping_lock);
+
+	ctx->switch_log = kmalloc(sizeof(struct switch_log) +
+		SWITCH_LOG_BUFSIZE * sizeof(struct switch_log_entry),
+		GFP_KERNEL);
+
+	if (!ctx->switch_log) {
+		rc = -ENOMEM;
+		goto out;
+	}
+
+	ctx->switch_log->head = ctx->switch_log->tail = 0;
+	init_waitqueue_head(&ctx->switch_log->wait);
+	rc = 0;
+
+out:
+	spu_release(ctx);
+	return rc;
+}
+
+static int spufs_switch_log_release(struct inode *inode, struct file *file)
+{
+	struct spu_context *ctx = SPUFS_I(inode)->i_ctx;
+	int rc;
+
+	rc = spu_acquire(ctx);
+	if (rc)
+		return rc;
+
+	kfree(ctx->switch_log);
+	ctx->switch_log = NULL;
+	spu_release(ctx);
 
 	return 0;
- out:
-	mutex_unlock(&ctx->mapping_lock);
-	return -ENOMEM;
 }
 
 static int switch_log_sprint(struct spu_context *ctx, char *tbuf, int n)
@@ -2485,42 +2501,54 @@ static ssize_t spufs_switch_log_read(struct file *file, char __user *buf,
 	if (!buf || len < 0)
 		return -EINVAL;
 
+	error = spu_acquire(ctx);
+	if (error)
+		return error;
+
 	while (cnt < len) {
 		char tbuf[128];
 		int width;
 
-		if (file->f_flags & O_NONBLOCK) {
-			if (spufs_switch_log_used(ctx) <= 0)
-				return cnt ? cnt : -EAGAIN;
-		} else {
-			/* Wait for data in buffer */
-			error = wait_event_interruptible(ctx->switch_log->wait,
-					spufs_switch_log_used(ctx) > 0);
-			if (error)
+		if (spufs_switch_log_used(ctx) == 0) {
+			if (cnt > 0) {
+				/* If there's data ready to go, we can
+				 * just return straight away */
+				break;
+
+			} else if (file->f_flags & O_NONBLOCK) {
+				error = -EAGAIN;
 				break;
-		}
 
-		spin_lock(&ctx->switch_log->lock);
-		if (ctx->switch_log->head == ctx->switch_log->tail) {
-			/* multiple readers race? */
-			spin_unlock(&ctx->switch_log->lock);
-			continue;
+			} else {
+				/* spufs_wait will drop the mutex and
+				 * re-acquire, but since we're in read(), the
+				 * file cannot be _released (and so
+				 * ctx->switch_log is stable).
+				 */
+				error = spufs_wait(ctx->switch_log->wait,
+						spufs_switch_log_used(ctx) > 0);
+
+				/* On error, spufs_wait returns without the
+				 * state mutex held */
+				if (error)
+					return error;
+
+				/* We may have had entries read from underneath
+				 * us while we dropped the mutex in spufs_wait,
+				 * so re-check */
+				if (spufs_switch_log_used(ctx) == 0)
+					continue;
+			}
 		}
 
 		width = switch_log_sprint(ctx, tbuf, sizeof(tbuf));
-		if (width < len) {
+		if (width < len)
 			ctx->switch_log->tail =
 				(ctx->switch_log->tail + 1) %
 				 SWITCH_LOG_BUFSIZE;
-		}
-
-		spin_unlock(&ctx->switch_log->lock);
-
-		/*
-		 * If the record is greater than space available return
-		 * partial buffer (so far)
-		 */
-		if (width >= len)
+		else
+			/* If the record is greater than space available return
+			 * partial buffer (so far) */
 			break;
 
 		error = copy_to_user(buf + cnt, tbuf, width);
@@ -2529,6 +2557,8 @@ static ssize_t spufs_switch_log_read(struct file *file, char __user *buf,
 		cnt += width;
 	}
 
+	spu_release(ctx);
+
 	return cnt == 0 ? error : cnt;
 }
 
@@ -2537,29 +2567,41 @@ static unsigned int spufs_switch_log_poll(struct file *file, poll_table *wait)
 	struct inode *inode = file->f_path.dentry->d_inode;
 	struct spu_context *ctx = SPUFS_I(inode)->i_ctx;
 	unsigned int mask = 0;
+	int rc;
 
 	poll_wait(file, &ctx->switch_log->wait, wait);
 
+	rc = spu_acquire(ctx);
+	if (rc)
+		return rc;
+
 	if (spufs_switch_log_used(ctx) > 0)
 		mask |= POLLIN;
 
+	spu_release(ctx);
+
 	return mask;
 }
 
 static const struct file_operations spufs_switch_log_fops = {
-	.owner	= THIS_MODULE,
-	.open	= spufs_switch_log_open,
-	.read	= spufs_switch_log_read,
-	.poll	= spufs_switch_log_poll,
+	.owner		= THIS_MODULE,
+	.open		= spufs_switch_log_open,
+	.read		= spufs_switch_log_read,
+	.poll		= spufs_switch_log_poll,
+	.release	= spufs_switch_log_release,
 };
 
+/**
+ * Log a context switch event to a switch log reader.
+ *
+ * Must be called with ctx->state_mutex held.
+ */
 void spu_switch_log_notify(struct spu *spu, struct spu_context *ctx,
 		u32 type, u32 val)
 {
 	if (!ctx->switch_log)
 		return;
 
-	spin_lock(&ctx->switch_log->lock);
 	if (spufs_switch_log_avail(ctx) > 1) {
 		struct switch_log_entry *p;
 
@@ -2573,7 +2615,6 @@ void spu_switch_log_notify(struct spu *spu, struct spu_context *ctx,
 		ctx->switch_log->head =
 			(ctx->switch_log->head + 1) % SWITCH_LOG_BUFSIZE;
 	}
-	spin_unlock(&ctx->switch_log->lock);
 
 	wake_up(&ctx->switch_log->wait);
 }
diff --git a/arch/powerpc/platforms/cell/spufs/run.c b/arch/powerpc/platforms/cell/spufs/run.c
index c9bb7cfd3dca..c58bd36b0c5b 100644
--- a/arch/powerpc/platforms/cell/spufs/run.c
+++ b/arch/powerpc/platforms/cell/spufs/run.c
@@ -249,6 +249,7 @@ static int spu_run_fini(struct spu_context *ctx, u32 *npc,
 
 	spuctx_switch_state(ctx, SPU_UTIL_IDLE_LOADED);
 	clear_bit(SPU_SCHED_SPU_RUN, &ctx->sched_flags);
+	spu_switch_log_notify(NULL, ctx, SWITCH_LOG_EXIT, *status);
 	spu_release(ctx);
 
 	if (signal_pending(current))
@@ -417,8 +418,6 @@ long spufs_run_spu(struct spu_context *ctx, u32 *npc, u32 *event)
 	ret = spu_run_fini(ctx, npc, &status);
 	spu_yield(ctx);
 
-	spu_switch_log_notify(NULL, ctx, SWITCH_LOG_EXIT, status);
-
 	if ((status & SPU_STATUS_STOPPED_BY_STOP) &&
 	    (((status >> SPU_STOP_STATUS_SHIFT) & 0x3f00) == 0x2100))
 		ctx->stats.libassist++;
diff --git a/arch/powerpc/platforms/cell/spufs/sched.c b/arch/powerpc/platforms/cell/spufs/sched.c
index 67595bc380dc..2ad914c47493 100644
--- a/arch/powerpc/platforms/cell/spufs/sched.c
+++ b/arch/powerpc/platforms/cell/spufs/sched.c
@@ -312,6 +312,15 @@ static struct spu *aff_ref_location(struct spu_context *ctx, int mem_aff,
 	 */
 	node = cpu_to_node(raw_smp_processor_id());
 	for (n = 0; n < MAX_NUMNODES; n++, node++) {
+		/*
+		 * "available_spus" counts how many spus are not potentially
+		 * going to be used by other affinity gangs whose reference
+		 * context is already in place. Although this code seeks to
+		 * avoid having affinity gangs with a summed amount of
+		 * contexts bigger than the amount of spus in the node,
+		 * this may happen sporadically. In this case, available_spus
+		 * becomes negative, which is harmless.
+		 */
 		int available_spus;
 
 		node = (node < MAX_NUMNODES) ? node : 0;
@@ -321,12 +330,10 @@ static struct spu *aff_ref_location(struct spu_context *ctx, int mem_aff,
 		available_spus = 0;
 		mutex_lock(&cbe_spu_info[node].list_mutex);
 		list_for_each_entry(spu, &cbe_spu_info[node].spus, cbe_list) {
-			if (spu->ctx && spu->ctx->gang
-					&& spu->ctx->aff_offset == 0)
-				available_spus -=
-					(spu->ctx->gang->contexts - 1);
-			else
-				available_spus++;
+			if (spu->ctx && spu->ctx->gang && !spu->ctx->aff_offset
+					&& spu->ctx->gang->aff_ref_spu)
+				available_spus -= spu->ctx->gang->contexts;
+			available_spus++;
 		}
 		if (available_spus < ctx->gang->contexts) {
 			mutex_unlock(&cbe_spu_info[node].list_mutex);
@@ -437,6 +444,11 @@ static void spu_unbind_context(struct spu *spu, struct spu_context *ctx)
 		atomic_dec(&cbe_spu_info[spu->node].reserved_spus);
 
 	if (ctx->gang)
+		/*
+		 * If ctx->gang->aff_sched_count is positive, SPU affinity is
+		 * being considered in this gang. Using atomic_dec_if_positive
+		 * allow us to skip an explicit check for affinity in this gang
+		 */
 		atomic_dec_if_positive(&ctx->gang->aff_sched_count);
 
 	spu_switch_notify(spu, NULL);
diff --git a/arch/powerpc/platforms/cell/spufs/spufs.h b/arch/powerpc/platforms/cell/spufs/spufs.h
index 8ae8ef9dfc22..15c62d3ca129 100644
--- a/arch/powerpc/platforms/cell/spufs/spufs.h
+++ b/arch/powerpc/platforms/cell/spufs/spufs.h
@@ -65,7 +65,6 @@ enum {
 };
 
 struct switch_log {
-	spinlock_t		lock;
 	wait_queue_head_t	wait;
 	unsigned long		head;
 	unsigned long		tail;
diff --git a/arch/powerpc/platforms/cell/spufs/sputrace.c b/arch/powerpc/platforms/cell/spufs/sputrace.c
index 2ece399f2862..d0b1f3f4d9c8 100644
--- a/arch/powerpc/platforms/cell/spufs/sputrace.c
+++ b/arch/powerpc/platforms/cell/spufs/sputrace.c
@@ -40,6 +40,7 @@ static DECLARE_WAIT_QUEUE_HEAD(sputrace_wait);
 static ktime_t sputrace_start;
 static unsigned long sputrace_head, sputrace_tail;
 static struct sputrace *sputrace_log;
+static int sputrace_logging;
 
 static int sputrace_used(void)
 {
@@ -79,6 +80,11 @@ static ssize_t sputrace_read(struct file *file, char __user *buf,
 		char tbuf[128];
 		int width;
 
+		/* If we have data ready to return, don't block waiting
+		 * for more */
+		if (cnt > 0 && sputrace_used() == 0)
+			break;
+
 		error = wait_event_interruptible(sputrace_wait,
 						 sputrace_used() > 0);
 		if (error)
@@ -109,24 +115,49 @@ static ssize_t sputrace_read(struct file *file, char __user *buf,
 
 static int sputrace_open(struct inode *inode, struct file *file)
 {
+	int rc;
+
 	spin_lock(&sputrace_lock);
+	if (sputrace_logging) {
+		rc = -EBUSY;
+		goto out;
+	}
+
+	sputrace_logging = 1;
 	sputrace_head = sputrace_tail = 0;
 	sputrace_start = ktime_get();
+	rc = 0;
+
+out:
 	spin_unlock(&sputrace_lock);
+	return rc;
+}
 
+static int sputrace_release(struct inode *inode, struct file *file)
+{
+	spin_lock(&sputrace_lock);
+	sputrace_logging = 0;
+	spin_unlock(&sputrace_lock);
 	return 0;
 }
 
 static const struct file_operations sputrace_fops = {
-	.owner	= THIS_MODULE,
-	.open	= sputrace_open,
-	.read	= sputrace_read,
+	.owner   = THIS_MODULE,
+	.open    = sputrace_open,
+	.read    = sputrace_read,
+	.release = sputrace_release,
 };
 
 static void sputrace_log_item(const char *name, struct spu_context *ctx,
 		struct spu *spu)
 {
 	spin_lock(&sputrace_lock);
+
+	if (!sputrace_logging) {
+		spin_unlock(&sputrace_lock);
+		return;
+	}
+
 	if (sputrace_avail() > 1) {
 		struct sputrace *t = sputrace_log + sputrace_head;