6 files changed, 121 insertions, 17 deletions

diff --git a/drivers/video/tegra/host/bus_client.c b/drivers/video/tegra/host/bus_client.c
index 9b71542a48aa..87aa9c64d363 100644
--- a/drivers/video/tegra/host/bus_client.c
+++ b/drivers/video/tegra/host/bus_client.c
@@ -141,12 +141,6 @@ static int nvhost_channelopen(struct inode *inode, struct file *filp)
 	priv->clientid = atomic_add_return(1,
 			&nvhost_get_host(ch->dev)->clientid);
 	priv->timeout = MAX_STUCK_CHECK_COUNT * SYNCPT_CHECK_PERIOD;
-
-	priv->job = nvhost_job_alloc(ch, priv->hwctx, &priv->hdr,
-			NULL, priv->priority, priv->clientid);
-	if (!priv->job)
-		goto fail;
-
 	return 0;
 fail:
 	nvhost_channelrelease(inode, filp);
diff --git a/drivers/video/tegra/host/nvhost_cdma.c b/drivers/video/tegra/host/nvhost_cdma.c
index b1f138317cc1..c87415bf5ac2 100644
--- a/drivers/video/tegra/host/nvhost_cdma.c
+++ b/drivers/video/tegra/host/nvhost_cdma.c
@@ -53,6 +53,18 @@ static void add_to_sync_queue(struct nvhost_cdma *cdma,
 	job->num_slots = nr_slots;
 	nvhost_job_get(job);
 	list_add_tail(&job->list, &cdma->sync_queue);
+
+	switch (job->priority) {
+	case NVHOST_PRIORITY_HIGH:
+		cdma->high_prio_count++;
+		break;
+	case NVHOST_PRIORITY_MEDIUM:
+		cdma->med_prio_count++;
+		break;
+	case NVHOST_PRIORITY_LOW:
+		cdma->low_prio_count++;
+		break;
+	}
 }
 
 /**
@@ -200,6 +212,19 @@ static void update_cdma_locked(struct nvhost_cdma *cdma)
 		}
 
 		list_del(&job->list);
+
+		switch (job->priority) {
+		case NVHOST_PRIORITY_HIGH:
+			cdma->high_prio_count--;
+			break;
+		case NVHOST_PRIORITY_MEDIUM:
+			cdma->med_prio_count--;
+			break;
+		case NVHOST_PRIORITY_LOW:
+			cdma->low_prio_count--;
+			break;
+		}
+
 		nvhost_job_put(job);
 	}
 
@@ -466,6 +491,12 @@ void nvhost_cdma_end(struct nvhost_cdma *cdma,
 	if (job->timeout && was_idle)
 		cdma_start_timer_locked(cdma, job);
 
+	trace_nvhost_cdma_end(job->ch->dev->name,
+			job->priority,
+			job->ch->cdma.high_prio_count,
+			job->ch->cdma.med_prio_count,
+			job->ch->cdma.low_prio_count);
+
 	mutex_unlock(&cdma->lock);
 }
 
@@ -490,6 +521,8 @@ int nvhost_cdma_flush(struct nvhost_cdma *cdma, int timeout)
 	unsigned int space, err = 0;
 	unsigned long end_jiffies = jiffies + msecs_to_jiffies(timeout);
 
+	trace_nvhost_cdma_flush(cdma_to_channel(cdma)->dev->name, timeout);
+
 	/*
 	 * Wait for at most timeout ms. Recalculate timeout at each iteration
 	 * to better keep within given timeout.
diff --git a/drivers/video/tegra/host/nvhost_cdma.h b/drivers/video/tegra/host/nvhost_cdma.h
index 98393f0cc765..2056774a7bc7 100644
--- a/drivers/video/tegra/host/nvhost_cdma.h
+++ b/drivers/video/tegra/host/nvhost_cdma.h
@@ -99,6 +99,9 @@ struct nvhost_cdma {
 	struct buffer_timeout timeout;	/* channel's timeout state/wq */
 	bool running;
 	bool torndown;
+	int high_prio_count;
+	int med_prio_count;
+	int low_prio_count;
 };
 
 #define cdma_to_channel(cdma) container_of(cdma, struct nvhost_channel, cdma)
diff --git a/drivers/video/tegra/host/nvhost_channel.c b/drivers/video/tegra/host/nvhost_channel.c
index ef8886fe4652..ad303cf0a22d 100644
--- a/drivers/video/tegra/host/nvhost_channel.c
+++ b/drivers/video/tegra/host/nvhost_channel.c
@@ -51,10 +51,26 @@ int nvhost_channel_init(struct nvhost_channel *ch,
 
 int nvhost_channel_submit(struct nvhost_job *job)
 {
-	/* Low priority submits wait until sync queue is empty. Ignores result
-	 * from nvhost_cdma_flush, as we submit either when push buffer is
-	 * empty or when we reach the timeout. */
-	if (job->priority < NVHOST_PRIORITY_MEDIUM)
+	/*
+	 * Check if queue has higher priority jobs running. If so, wait until
+	 * queue is empty. Ignores result from nvhost_cdma_flush, as we submit
+	 * either when push buffer is empty or when we reach the timeout.
+	 */
+	int higher_count = 0;
+
+	switch (job->priority) {
+	case NVHOST_PRIORITY_HIGH:
+		higher_count = 0;
+		break;
+	case NVHOST_PRIORITY_MEDIUM:
+		higher_count = job->ch->cdma.high_prio_count;
+		break;
+	case NVHOST_PRIORITY_LOW:
+		higher_count = job->ch->cdma.high_prio_count
+			+ job->ch->cdma.med_prio_count;
+		break;
+	}
+	if (higher_count > 0)
 		(void)nvhost_cdma_flush(&job->ch->cdma,
 				NVHOST_CHANNEL_LOW_PRIO_MAX_WAIT);
 
diff --git a/drivers/video/tegra/host/nvhost_intr.c b/drivers/video/tegra/host/nvhost_intr.c
index ba821f694cb4..af2e3ad1bdb5 100644
--- a/drivers/video/tegra/host/nvhost_intr.c
+++ b/drivers/video/tegra/host/nvhost_intr.c
@@ -128,12 +128,16 @@ static void action_submit_complete(struct nvhost_waitlist *waiter)
 	struct nvhost_channel *channel = waiter->data;
 	int nr_completed = waiter->count;
 
+	nvhost_cdma_update(&channel->cdma);
+	nvhost_module_idle_mult(channel->dev, nr_completed);
+
 	/*  Add nr_completed to trace */
 	trace_nvhost_channel_submit_complete(channel->dev->name,
-			nr_completed, waiter->thresh);
+			nr_completed, waiter->thresh,
+			channel->cdma.high_prio_count,
+			channel->cdma.med_prio_count,
+			channel->cdma.low_prio_count);
 
-	nvhost_cdma_update(&channel->cdma);
-	nvhost_module_idle_mult(channel->dev, nr_completed);
 }
 
 static void action_ctxsave(struct nvhost_waitlist *waiter)
diff --git a/include/trace/events/nvhost.h b/include/trace/events/nvhost.h
index b4818f55844c..6506af44e576 100644
--- a/include/trace/events/nvhost.h
+++ b/include/trace/events/nvhost.h
@@ -138,6 +138,52 @@ TRACE_EVENT(nvhost_channel_write_cmdbuf,
 	  __entry->words, __entry->offset)
 );
 
+TRACE_EVENT(nvhost_cdma_end,
+	TP_PROTO(const char *name, int prio,
+		int hi_count, int med_count, int low_count),
+
+	TP_ARGS(name, prio, hi_count, med_count, low_count),
+
+	TP_STRUCT__entry(
+		__field(const char *, name)
+		__field(int, prio)
+		__field(int, hi_count)
+		__field(int, med_count)
+		__field(int, low_count)
+	),
+
+	TP_fast_assign(
+		__entry->name = name;
+		__entry->prio = prio;
+		__entry->hi_count = hi_count;
+		__entry->med_count = med_count;
+		__entry->low_count = low_count;
+	),
+
+	TP_printk("name=%s, prio=%d, hi=%d, med=%d, low=%d",
+		__entry->name, __entry->prio,
+		__entry->hi_count, __entry->med_count, __entry->low_count)
+);
+
+TRACE_EVENT(nvhost_cdma_flush,
+	TP_PROTO(const char *name, int timeout),
+
+	TP_ARGS(name, timeout),
+
+	TP_STRUCT__entry(
+		__field(const char *, name)
+		__field(int, timeout)
+	),
+
+	TP_fast_assign(
+		__entry->name = name;
+		__entry->timeout = timeout;
+	),
+
+	TP_printk("name=%s, timeout=%d",
+		__entry->name, __entry->timeout)
+);
+
 TRACE_EVENT(nvhost_cdma_push,
 	TP_PROTO(const char *name, u32 op1, u32 op2),
 
@@ -425,24 +471,32 @@ TRACE_EVENT(nvhost_channel_submitted,
 );
 
 TRACE_EVENT(nvhost_channel_submit_complete,
-	TP_PROTO(const char *name, int count, u32 thresh),
+	TP_PROTO(const char *name, int count, u32 thresh,
+		int hi_count, int med_count, int low_count),
 
-	TP_ARGS(name, count, thresh),
+	TP_ARGS(name, count, thresh, hi_count, med_count, low_count),
 
 	TP_STRUCT__entry(
 		__field(const char *, name)
 		__field(int, count)
 		__field(u32, thresh)
+		__field(int, hi_count)
+		__field(int, med_count)
+		__field(int, low_count)
 	),
 
 	TP_fast_assign(
 		__entry->name = name;
 		__entry->count = count;
 		__entry->thresh = thresh;
+		__entry->hi_count = hi_count;
+		__entry->med_count = med_count;
+		__entry->low_count = low_count;
 	),
 
-	TP_printk("name=%s, count=%d, thresh=%d",
-		__entry->name, __entry->count, __entry->thresh)
+	TP_printk("name=%s, count=%d, thresh=%d, hi=%d, med=%d, low=%d",
+		__entry->name, __entry->count, __entry->thresh,
+		__entry->hi_count, __entry->med_count, __entry->low_count)
 );
 
 TRACE_EVENT(nvhost_wait_cdma,