video: tegra: host: Throttle lower priority jobs

Implement per channel counter for jobs in each priority level. If
there are jobs active with higher priority than the one being
submitted, throttle.

Bug 926690

Change-Id: I5fed341e3f248325873b31d1c53bf57bf0a78074
Signed-off-by: Terje Bergstrom <tbergstrom@nvidia.com>
Reviewed-on: http://git-master/r/104939
Reviewed-by: Automatic_Commit_Validation_User
GVS: Gerrit_Virtual_Submit
Reviewed-by: Juha Tukkinen <jtukkinen@nvidia.com>
Reviewed-by: Mayuresh Kulkarni <mkulkarni@nvidia.com>

5 files changed, 63 insertions, 13 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)