diff --git a/drivers/video/tegra/host/nvdla/dla_os_interface.h b/drivers/video/tegra/host/nvdla/dla_os_interface.h
index d2204aaa..d6efca15 100644
--- a/drivers/video/tegra/host/nvdla/dla_os_interface.h
+++ b/drivers/video/tegra/host/nvdla/dla_os_interface.h
@@ -87,6 +87,7 @@
 /* write actions */
 #define ACTION_WRITE_SEM	0x80
 #define ACTION_WRITE_TS_SEM	0x83
+#define ACTION_WRITE_TIMESTAMP  	0x87
 #define ACTION_WRITE_GOS	0xA0
 #define ACTION_WRITE_TASK_STATUS	0xC1
 
@@ -231,6 +232,17 @@ struct dla_action_task_status {
 	uint16_t status;
 } __attribute__ ((packed));
 
+/**
+ * Timestamp update action structure
+ *
+ * OPCODE = 0x87
+ *
+ * @address: Address to write timestamp
+ */
+struct dla_action_timestamp {
+	uint64_t address;
+} __attribute__ ((packed));
+
 /**
  * Status notifier structure
  *
diff --git a/drivers/video/tegra/host/nvdla/nvdla.h b/drivers/video/tegra/host/nvdla/nvdla.h
index 11e014bd..2aa3a6cc 100644
--- a/drivers/video/tegra/host/nvdla/nvdla.h
+++ b/drivers/video/tegra/host/nvdla/nvdla.h
@@ -103,6 +103,7 @@
 #define MAX_NUM_NVDLA_EMU_POSTFENCES	16
 #define MAX_NUM_NVDLA_IN_TASK_STATUS	MAX_NUM_NVDLA_PREFENCES
 #define MAX_NUM_NVDLA_OUT_TASK_STATUS	MAX_NUM_NVDLA_POSTFENCES
+#define MAX_NUM_NVDLA_OUT_TIMESTAMP  	32
 #define NUM_PROFILING_POSTACTION	1
 
 #define MAX_COMMANDS_PER_DEVICE		1
@@ -267,12 +268,16 @@ struct nvdla_task {
 	struct nvdla_status_notify in_task_status[MAX_NUM_NVDLA_IN_TASK_STATUS];
 	struct nvdla_status_notify sof_task_status[MAX_NUM_NVDLA_OUT_TASK_STATUS];
 	struct nvdla_status_notify eof_task_status[MAX_NUM_NVDLA_OUT_TASK_STATUS];
+	struct nvdla_mem_handle sof_timestamps[MAX_NUM_NVDLA_OUT_TIMESTAMP];
+	struct nvdla_mem_handle eof_timestamps[MAX_NUM_NVDLA_OUT_TIMESTAMP];
 	struct nvdla_mem_handle memory_handles[NVDLA_MAX_BUFFERS_PER_TASK];
 	u8 num_prefences;
 	u8 num_postfences;
 	u8 num_in_task_status;
 	u8 num_sof_task_status;
 	u8 num_eof_task_status;
+	u8 num_sof_timestamps;
+	u8 num_eof_timestamps;
 	u32 num_addresses;
 	u32 fence;
 	u32 fence_counter;
@@ -290,6 +295,8 @@ struct nvdla_task {
 	struct dma_buf *postfences_sem_dmabuf[MAX_NUM_NVDLA_POSTFENCES];
 	struct dma_buf *sof_task_status_dmabuf[MAX_NUM_NVDLA_OUT_TASK_STATUS];
 	struct dma_buf *eof_task_status_dmabuf[MAX_NUM_NVDLA_OUT_TASK_STATUS];
+	struct dma_buf *sof_timestamps_dmabuf[MAX_NUM_NVDLA_OUT_TIMESTAMP];
+	struct dma_buf *eof_timestamps_dmabuf[MAX_NUM_NVDLA_OUT_TIMESTAMP];
 };
 
 struct dla_mem_addr {
diff --git a/drivers/video/tegra/host/nvdla/nvdla_ioctl.c b/drivers/video/tegra/host/nvdla/nvdla_ioctl.c
index 04652af4..4987ee3b 100644
--- a/drivers/video/tegra/host/nvdla/nvdla_ioctl.c
+++ b/drivers/video/tegra/host/nvdla/nvdla_ioctl.c
@@ -396,6 +396,24 @@ static int nvdla_get_actions(struct nvdla_ioctl_submit_task *user_task,
 		goto fail;
 	}
 
+	/* get sof timestamps */
+	if (copy_from_user(task->sof_timestamps,
+		(void __user*)user_task->sof_timestamps,
+		(task->num_sof_timestamps * sizeof(struct nvdla_mem_handle)))) {
+		err = -EFAULT;
+		nvdla_dbg_err(pdev, "failed to copy sof timestamps");
+		goto fail;
+	}
+
+	/* get eof timestamps */
+	if (copy_from_user(task->eof_timestamps,
+		(void __user*)user_task->eof_timestamps,
+		(task->num_eof_timestamps * sizeof(struct nvdla_mem_handle)))) {
+		err = -EFAULT;
+		nvdla_dbg_err(pdev, "failed to copy eof timestamps");
+		goto fail;
+	}
+
 	nvdla_dbg_info(pdev, "copying actions done");
 
 fail:
@@ -640,6 +658,18 @@ static int nvdla_val_task_submit_input(struct nvdla_ioctl_submit_task *in_task)
 			MAX_NUM_NVDLA_OUT_TASK_STATUS);
 		return -EINVAL;
 	}
+	if (in_task->num_sof_timestamps > MAX_NUM_NVDLA_OUT_TIMESTAMP) {
+		pr_err("sof timestamps[%u] crossing expected[%d]\n",
+			in_task->num_sof_timestamps,
+			MAX_NUM_NVDLA_OUT_TIMESTAMP);
+		return -EINVAL;
+	}
+	if (in_task->num_eof_timestamps > MAX_NUM_NVDLA_OUT_TIMESTAMP) {
+		pr_err("eof timestamps[%u] crossing expected[%d]\n",
+			in_task->num_eof_timestamps,
+			MAX_NUM_NVDLA_OUT_TIMESTAMP);
+		return -EINVAL;
+	}
 	if (in_task->num_addresses < 1) {
 		pr_err("num addresses[%u] should be min one\n",
 				in_task->num_addresses);
@@ -683,6 +713,8 @@ static int nvdla_fill_task(struct nvdla_queue *queue,
 	task->num_in_task_status = local_task->num_input_task_status;
 	task->num_sof_task_status = local_task->num_sof_task_status;
 	task->num_eof_task_status = local_task->num_eof_task_status;
+	task->num_sof_timestamps = local_task->num_sof_timestamps;
+	task->num_eof_timestamps = local_task->num_eof_timestamps;
 	task->num_addresses = local_task->num_addresses;
 	task->timeout = local_task->timeout;
 
@@ -731,6 +763,9 @@ static void nvdla_dump_task(struct nvdla_task *task)
 			task->num_in_task_status,
 			task->num_sof_task_status,
 			task->num_eof_task_status);
+	nvdla_dbg_info(pdev, "num_sof_timestamps[%u] num_eof_timestamps[%u]",
+			task->num_sof_timestamps,
+			task->num_eof_timestamps);
 	nvdla_dbg_info(pdev, "num_addresses[%u]", task->num_addresses);
 
 	for (i = 0; i < task->num_prefences; i++) {
@@ -783,6 +818,20 @@ static void nvdla_dump_task(struct nvdla_task *task)
 				task->eof_task_status[i].status);
 	}
 
+	for (i = 0; i < task->num_sof_timestamps; i++) {
+		nvdla_dbg_info(pdev, "SOF timestamp[%d]:"
+				"handle[%u] offset[%u]",
+				i, task->sof_timestamps[i].handle,
+				task->sof_timestamps[i].offset);
+	}
+
+	for (i = 0; i < task->num_eof_timestamps; i++) {
+		nvdla_dbg_info(pdev, "EOF timestamp[%d]:"
+				"handle[%u] offset[%u]",
+				i, task->eof_timestamps[i].handle,
+				task->eof_timestamps[i].offset);
+	}
+
 	for (i = 0; i < task->num_addresses; i++) {
 		nvdla_dbg_info(pdev, "Memory Handles[%d]:"
 				"handle[%u] offset[%u]",
diff --git a/drivers/video/tegra/host/nvdla/nvdla_queue.c b/drivers/video/tegra/host/nvdla/nvdla_queue.c
index 0ce4e700..10c4b145 100644
--- a/drivers/video/tegra/host/nvdla/nvdla_queue.c
+++ b/drivers/video/tegra/host/nvdla/nvdla_queue.c
@@ -228,6 +228,25 @@ static int nvdla_unmap_task_memory(struct nvdla_task *task)
 	}
 	nvdla_dbg_fn(pdev, "all out task status unmaped");
 
+	/* unpin output timestamp memory */
+	for (ii = 0; ii < task->num_sof_timestamps; ii++) {
+		if (task->sof_timestamps[ii].handle) {
+			nvdla_buffer_submit_unpin(task->buffers,
+				&task->sof_timestamps_dmabuf[ii], 1);
+			dma_buf_put(task->sof_timestamps_dmabuf[ii]);
+		}
+	}
+
+	for (ii = 0; ii < task->num_eof_timestamps; ii++) {
+		if (task->eof_timestamps[ii].handle) {
+			nvdla_buffer_submit_unpin(task->buffers,
+				&task->eof_timestamps_dmabuf[ii], 1);
+			dma_buf_put(task->eof_timestamps_dmabuf[ii]);
+		}
+	}
+	nvdla_dbg_fn(pdev, "all out timestamps unmaped");
+
+
 	return 0;
 }
 
@@ -261,12 +280,15 @@ static void nvdla_task_syncpt_reset(struct nvhost_syncpt *syncpt,
 static inline int nvdla_get_max_preaction_size(void)
 {
 	return (((MAX_NUM_NVDLA_PREFENCES + MAX_NUM_NVDLA_IN_TASK_STATUS +
-			MAX_NUM_NVDLA_OUT_TASK_STATUS) *
+			MAX_NUM_NVDLA_OUT_TASK_STATUS +
+			MAX_NUM_NVDLA_OUT_TIMESTAMP) *
 		sizeof(struct dla_action_opcode)) +
 		(MAX_NUM_NVDLA_PREFENCES *
 			sizeof(struct dla_action_semaphore)) +
 		((MAX_NUM_NVDLA_IN_TASK_STATUS + MAX_NUM_NVDLA_OUT_TASK_STATUS) *
 			sizeof(struct dla_action_task_status)) +
+		(MAX_NUM_NVDLA_OUT_TIMESTAMP *
+			sizeof(struct dla_action_timestamp)) +
 		sizeof(struct dla_action_opcode));
 }
 
@@ -274,6 +296,7 @@ static inline int nvdla_get_max_postaction_size(void)
 {
 	return (((MAX_NUM_NVDLA_POSTFENCES +
 				MAX_NUM_NVDLA_OUT_TASK_STATUS +
+				MAX_NUM_NVDLA_OUT_TIMESTAMP +
 				NUM_PROFILING_POSTACTION) *
 		sizeof(struct dla_action_opcode)) +
 		(MAX_NUM_NVDLA_POSTFENCES *
@@ -281,6 +304,8 @@ static inline int nvdla_get_max_postaction_size(void)
 		((MAX_NUM_NVDLA_OUT_TASK_STATUS +
 			NUM_PROFILING_POSTACTION) *
 			sizeof(struct dla_action_task_status)) +
+		(MAX_NUM_NVDLA_OUT_TIMESTAMP *
+			sizeof(struct dla_action_timestamp)) +
 		sizeof(struct dla_action_opcode));
 }
 
@@ -436,6 +461,18 @@ static u8 *add_status_action(u8 *mem, uint8_t op, uint64_t addr,
 	return mem + sizeof(struct dla_action_task_status);
 }
 
+static u8 *add_timestamp_action(u8 *mem, uint8_t op, uint64_t addr)
+{
+	struct dla_action_timestamp *action;
+
+	mem = add_opcode(mem, op);
+
+	action = (struct dla_action_timestamp *)mem;
+	action->address = addr;
+
+	return mem + sizeof(struct dla_action_timestamp);
+}
+
 static u8 *add_gos_action(u8 *mem, uint8_t op, uint8_t index, uint16_t offset,
 				uint32_t value)
 {
@@ -918,6 +955,50 @@ fail:
 	return err;
 }
 
+static int nvdla_fill_timestamp_write_action(struct nvdla_task *task,
+	struct nvdla_mem_handle *timestamp,
+	struct dma_buf **dma_buf,
+	u8 **mem_next)
+{
+	int err = 0;
+
+	struct nvdla_buffers *buffers = task->buffers;
+	struct nvdla_queue *queue = task->queue;
+	struct platform_device *pdev = queue->pool->pdev;
+	dma_addr_t dma_addr;
+	size_t dma_size;
+
+	u8 *next = *mem_next;
+
+	nvdla_dbg_info(pdev, "h[%u] o[%u]",
+			timestamp->handle,
+			timestamp->offset);
+
+	*dma_buf = dma_buf_get(timestamp->handle);
+	if (IS_ERR_OR_NULL(*dma_buf)) {
+		*dma_buf = NULL;
+		nvdla_dbg_err(pdev, "fail to get buf");
+		err = -EINVAL;
+		goto fail;
+	}
+
+	if (nvdla_buffer_submit_pin(buffers,
+			dma_buf, 1, &dma_addr, &dma_size, NULL)) {
+		nvdla_dbg_err(pdev, "fail to pin timestamp");
+		err = -EINVAL;
+		goto fail;
+	}
+
+	next = add_timestamp_action(next, ACTION_WRITE_TIMESTAMP,
+			dma_addr + timestamp->offset);
+
+	*mem_next = next;
+
+fail:
+	return err;
+}
+
+
 static int nvdla_fill_postactions(struct nvdla_task *task)
 {
 	int err = 0;
@@ -941,6 +1022,20 @@ static int nvdla_fill_postactions(struct nvdla_task *task)
 	next = add_status_action(next, ACTION_WRITE_TASK_STATUS,
 		task->task_desc_pa + nvdla_profile_status_offset(task), 0);
 
+	/* fill eof timestamp actions */
+	for (i = 0; i < task->num_eof_timestamps; i++) {
+		err = nvdla_fill_timestamp_write_action(task,
+				&task->eof_timestamps[i],
+				&task->eof_timestamps_dmabuf[i],
+				&next);
+		if (err < 0) {
+			nvdla_dbg_err(pdev,
+				"failed to fill eof timestamp[%d]",
+				i);
+			goto fail;
+		}
+	}
+
 	/* fill output task status */
 	for (i = 0; i < task->num_eof_task_status; i++) {
 		err = nvdla_fill_taskstatus_write_action(task,
@@ -1043,6 +1138,20 @@ static int nvdla_fill_preactions(struct nvdla_task *task)
 		}
 	}
 
+	/* fill sof timestamp actions */
+	for (i = 0; i < task->num_sof_timestamps; i++) {
+		err = nvdla_fill_timestamp_write_action(task,
+				&task->sof_timestamps[i],
+				&task->sof_timestamps_dmabuf[i],
+				&next);
+		if (err < 0) {
+			nvdla_dbg_err(pdev,
+				"failed to fill sof timestamp[%d]",
+				i);
+			goto fail;
+		}
+	}
+
 	/* fill all preactions signals */
 	for (i = 0; i < task->num_prefences; i++) {
 		/* update action */
diff --git a/include/uapi/linux/nvhost_nvdla_ioctl.h b/include/uapi/linux/nvhost_nvdla_ioctl.h
index a3e750ac..3d85c544 100644
--- a/include/uapi/linux/nvhost_nvdla_ioctl.h
+++ b/include/uapi/linux/nvhost_nvdla_ioctl.h
@@ -79,7 +79,7 @@ struct nvdla_pin_unpin_args {
 struct nvdla_submit_args {
 	__u64 tasks;
 	__u16 num_tasks;
-#define MAX_TASKS_PER_SUBMIT		24
+#define MAX_TASKS_PER_SUBMIT		16
 #define NVDLA_SUBMIT_FLAGS_ATOMIC	(1 << 0)
 	__u16 flags;
 	__u32 version;
@@ -127,6 +127,8 @@ struct nvdla_mem_handle {
  * @num_input_task_status	number of input task status
  * @num_sof_task_status  	number of sof task status
  * @num_eof_task_status  	number of eof task status
+ * @num_sof_timestamps   	number of sof timestamp
+ * @num_eof_timestamps   	number of eof timestamp
  * @flags			flags for bitwise task info embeddeing
  * @reserved			reserved for future use
  * @prefences			pointer to pre-fence struct table
@@ -134,6 +136,8 @@ struct nvdla_mem_handle {
  * @input_task_status		pointer to input task status struct table
  * @sof_task_status  		pointer to sof task status struct table
  * @eof_task_status  		pointer to eof task status struct table
+ * @sof_timestamps   		pointer to sof timestamp handle list
+ * @eof_timestamps   		pointer to eof timestamp handle list
  * @num_addresses		total number of addressed passed in structure
  * @address_list		pointer to address list
  * @timeout			task timeout
@@ -145,7 +149,9 @@ struct nvdla_ioctl_submit_task {
 	__u8 num_input_task_status;
 	__u8 num_sof_task_status;
 	__u8 num_eof_task_status;
-	__u8 reserved0[3];
+	__u8 num_sof_timestamps;
+	__u8 num_eof_timestamps;
+	__u8 reserved0[1];
 #define NVDLA_MAX_BUFFERS_PER_TASK (6144)
 	__u32 num_addresses;
 	__u16 flags;
@@ -157,6 +163,8 @@ struct nvdla_ioctl_submit_task {
 	__u64 input_task_status;
 	__u64 sof_task_status;
 	__u64 eof_task_status;
+	__u64 sof_timestamps;
+	__u64 eof_timestamps;
 	__u64 address_list;
 	__u64 timeout;
 };