media: staging/intel-ipu3: Add imgu top level pci device driver

This patch adds support for the Intel IPU v3 as found
on Skylake and Kaby Lake SoCs.

The driver glues v4l2, css(camera sub system) and other
pieces together to perform its functions, it also loads
the IPU3 firmware binary as part of its initialization.

Signed-off-by: Yong Zhi <yong.zhi@intel.com>
Signed-off-by: Tomasz Figa <tfiga@chromium.org>
Signed-off-by: Sakari Ailus <sakari.ailus@linux.intel.com>
Signed-off-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>

5 files changed, 1044 insertions, 0 deletions

diff --git a/drivers/staging/media/ipu3/Kconfig b/drivers/staging/media/ipu3/Kconfig
new file mode 100644
index 000000000000..75cd889f18f7
--- /dev/null
+++ b/drivers/staging/media/ipu3/Kconfig
@@ -0,0 +1,14 @@
+config VIDEO_IPU3_IMGU
+	tristate "Intel ipu3-imgu driver"
+	depends on PCI && VIDEO_V4L2
+	depends on MEDIA_CONTROLLER && VIDEO_V4L2_SUBDEV_API
+	depends on X86
+	select IOMMU_IOVA
+	select VIDEOBUF2_DMA_SG
+	---help---
+	  This is the Video4Linux2 driver for Intel IPU3 image processing unit,
+	  found in Intel Skylake and Kaby Lake SoCs and used for processing
+	  images and video.
+
+	  Say Y or M here if you have a Skylake/Kaby Lake SoC with a MIPI
+	  camera. The module will be called ipu3-imgu.
diff --git a/drivers/staging/media/ipu3/Makefile b/drivers/staging/media/ipu3/Makefile
new file mode 100644
index 000000000000..fb146d178bd4
--- /dev/null
+++ b/drivers/staging/media/ipu3/Makefile
@@ -0,0 +1,11 @@
+#
+# Makefile for the IPU3 ImgU drivers
+#
+
+ipu3-imgu-objs += \
+		ipu3-mmu.o ipu3-dmamap.o \
+		ipu3-tables.o ipu3-css-pool.o \
+		ipu3-css-fw.o ipu3-css-params.o \
+		ipu3-css.o ipu3-v4l2.o ipu3.o
+
+obj-$(CONFIG_VIDEO_IPU3_IMGU) += ipu3-imgu.o
diff --git a/drivers/staging/media/ipu3/TODO b/drivers/staging/media/ipu3/TODO
new file mode 100644
index 000000000000..922b885f10a7
--- /dev/null
+++ b/drivers/staging/media/ipu3/TODO
@@ -0,0 +1,23 @@
+This is a list of things that need to be done to get this driver out of the
+staging directory.
+
+- Request API conversion. Remove of the dual pipeline and associate buffers
+  as well as formats and the binary used to a request. Remove the
+  opportunistic buffer management. (Sakari)
+
+- Using ENABLED and IMMUTABLE link flags for the links where those are
+  relevant. (Sakari)
+
+- Prefix imgu for all public APIs, i.e. change ipu3_v4l2_register() to
+  imgu_v4l2_register(). (Sakari)
+
+- Use V4L2_CTRL_TYPE_MENU for dual-pipe mode control. (Sakari)
+
+- IPU3 driver documentation (Laurent)
+  Add diagram in driver rst to describe output capability.
+  Comments on configuring v4l2 subdevs for CIO2 and ImgU.
+
+- uAPI documentation:
+  Further clarification on some ambiguities such as data type conversion of
+  IEFD CU inputs. (Sakari)
+  Move acronyms to doc-rst file. (Mauro)
diff --git a/drivers/staging/media/ipu3/ipu3.c b/drivers/staging/media/ipu3/ipu3.c
new file mode 100644
index 000000000000..3d0a34b86ff4
--- /dev/null
+++ b/drivers/staging/media/ipu3/ipu3.c
@@ -0,0 +1,844 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Copyright (C) 2017 - 2018 Intel Corporation
+ * Copyright 2017 Google LLC
+ *
+ * Based on Intel IPU4 driver.
+ *
+ */
+
+#include <linux/delay.h>
+#include <linux/interrupt.h>
+#include <linux/module.h>
+#include <linux/pm_runtime.h>
+
+#include "ipu3.h"
+#include "ipu3-dmamap.h"
+#include "ipu3-mmu.h"
+
+#define IMGU_PCI_ID			0x1919
+#define IMGU_PCI_BAR			0
+#define IMGU_DMA_MASK			DMA_BIT_MASK(39)
+#define IMGU_MAX_QUEUE_DEPTH		(2 + 2)
+
+/*
+ * pre-allocated buffer size for IMGU dummy buffers. Those
+ * values should be tuned to big enough to avoid buffer
+ * re-allocation when streaming to lower streaming latency.
+ */
+#define CSS_QUEUE_IN_BUF_SIZE		0
+#define CSS_QUEUE_PARAMS_BUF_SIZE	0
+#define CSS_QUEUE_OUT_BUF_SIZE		(4160 * 3120 * 12 / 8)
+#define CSS_QUEUE_VF_BUF_SIZE		(1920 * 1080 * 12 / 8)
+#define CSS_QUEUE_STAT_3A_BUF_SIZE	sizeof(struct ipu3_uapi_stats_3a)
+
+static const size_t css_queue_buf_size_map[IPU3_CSS_QUEUES] = {
+	[IPU3_CSS_QUEUE_IN] = CSS_QUEUE_IN_BUF_SIZE,
+	[IPU3_CSS_QUEUE_PARAMS] = CSS_QUEUE_PARAMS_BUF_SIZE,
+	[IPU3_CSS_QUEUE_OUT] = CSS_QUEUE_OUT_BUF_SIZE,
+	[IPU3_CSS_QUEUE_VF] = CSS_QUEUE_VF_BUF_SIZE,
+	[IPU3_CSS_QUEUE_STAT_3A] = CSS_QUEUE_STAT_3A_BUF_SIZE,
+};
+
+static const struct imgu_node_mapping imgu_node_map[IMGU_NODE_NUM] = {
+	[IMGU_NODE_IN] = {IPU3_CSS_QUEUE_IN, "input"},
+	[IMGU_NODE_PARAMS] = {IPU3_CSS_QUEUE_PARAMS, "parameters"},
+	[IMGU_NODE_OUT] = {IPU3_CSS_QUEUE_OUT, "output"},
+	[IMGU_NODE_VF] = {IPU3_CSS_QUEUE_VF, "viewfinder"},
+	[IMGU_NODE_PV] = {IPU3_CSS_QUEUE_VF, "postview"},
+	[IMGU_NODE_STAT_3A] = {IPU3_CSS_QUEUE_STAT_3A, "3a stat"},
+};
+
+unsigned int imgu_node_to_queue(unsigned int node)
+{
+	return imgu_node_map[node].css_queue;
+}
+
+unsigned int imgu_map_node(struct imgu_device *imgu, unsigned int css_queue)
+{
+	unsigned int i;
+
+	if (css_queue == IPU3_CSS_QUEUE_VF)
+		return imgu->nodes[IMGU_NODE_VF].enabled ?
+			IMGU_NODE_VF : IMGU_NODE_PV;
+
+	for (i = 0; i < IMGU_NODE_NUM; i++)
+		if (imgu_node_map[i].css_queue == css_queue)
+			break;
+
+	return i;
+}
+
+/**************** Dummy buffers ****************/
+
+static void imgu_dummybufs_cleanup(struct imgu_device *imgu)
+{
+	unsigned int i;
+
+	for (i = 0; i < IPU3_CSS_QUEUES; i++)
+		ipu3_dmamap_free(imgu, &imgu->queues[i].dmap);
+}
+
+static int imgu_dummybufs_preallocate(struct imgu_device *imgu)
+{
+	unsigned int i;
+	size_t size;
+
+	for (i = 0; i < IPU3_CSS_QUEUES; i++) {
+		size = css_queue_buf_size_map[i];
+		/*
+		 * Do not enable dummy buffers for master queue,
+		 * always require that real buffers from user are
+		 * available.
+		 */
+		if (i == IMGU_QUEUE_MASTER || size == 0)
+			continue;
+
+		if (!ipu3_dmamap_alloc(imgu, &imgu->queues[i].dmap, size)) {
+			imgu_dummybufs_cleanup(imgu);
+			return -ENOMEM;
+		}
+	}
+
+	return 0;
+}
+
+static int imgu_dummybufs_init(struct imgu_device *imgu)
+{
+	const struct v4l2_pix_format_mplane *mpix;
+	const struct v4l2_meta_format	*meta;
+	unsigned int i, j, node;
+	size_t size;
+
+	/* Allocate a dummy buffer for each queue where buffer is optional */
+	for (i = 0; i < IPU3_CSS_QUEUES; i++) {
+		node = imgu_map_node(imgu, i);
+		if (!imgu->queue_enabled[node] || i == IMGU_QUEUE_MASTER)
+			continue;
+
+		if (!imgu->nodes[IMGU_NODE_VF].enabled &&
+		    !imgu->nodes[IMGU_NODE_PV].enabled &&
+		    i == IPU3_CSS_QUEUE_VF)
+			/*
+			 * Do not enable dummy buffers for VF/PV if it is not
+			 * requested by the user.
+			 */
+			continue;
+
+		meta = &imgu->nodes[node].vdev_fmt.fmt.meta;
+		mpix = &imgu->nodes[node].vdev_fmt.fmt.pix_mp;
+
+		if (node == IMGU_NODE_STAT_3A || node == IMGU_NODE_PARAMS)
+			size = meta->buffersize;
+		else
+			size = mpix->plane_fmt[0].sizeimage;
+
+		if (ipu3_css_dma_buffer_resize(imgu, &imgu->queues[i].dmap,
+					       size)) {
+			imgu_dummybufs_cleanup(imgu);
+			return -ENOMEM;
+		}
+
+		for (j = 0; j < IMGU_MAX_QUEUE_DEPTH; j++)
+			ipu3_css_buf_init(&imgu->queues[i].dummybufs[j], i,
+					  imgu->queues[i].dmap.daddr);
+	}
+
+	return 0;
+}
+
+/* May be called from atomic context */
+static struct ipu3_css_buffer *imgu_dummybufs_get(struct imgu_device *imgu,
+						  int queue)
+{
+	unsigned int i;
+
+	/* dummybufs are not allocated for master q */
+	if (queue == IPU3_CSS_QUEUE_IN)
+		return NULL;
+
+	if (WARN_ON(!imgu->queues[queue].dmap.vaddr))
+		/* Buffer should not be allocated here */
+		return NULL;
+
+	for (i = 0; i < IMGU_MAX_QUEUE_DEPTH; i++)
+		if (ipu3_css_buf_state(&imgu->queues[queue].dummybufs[i]) !=
+			IPU3_CSS_BUFFER_QUEUED)
+			break;
+
+	if (i == IMGU_MAX_QUEUE_DEPTH)
+		return NULL;
+
+	ipu3_css_buf_init(&imgu->queues[queue].dummybufs[i], queue,
+			  imgu->queues[queue].dmap.daddr);
+
+	return &imgu->queues[queue].dummybufs[i];
+}
+
+/* Check if given buffer is a dummy buffer */
+static bool imgu_dummybufs_check(struct imgu_device *imgu,
+				 struct ipu3_css_buffer *buf)
+{
+	unsigned int i;
+
+	for (i = 0; i < IMGU_MAX_QUEUE_DEPTH; i++)
+		if (buf == &imgu->queues[buf->queue].dummybufs[i])
+			break;
+
+	return i < IMGU_MAX_QUEUE_DEPTH;
+}
+
+static void imgu_buffer_done(struct imgu_device *imgu, struct vb2_buffer *vb,
+			     enum vb2_buffer_state state)
+{
+	mutex_lock(&imgu->lock);
+	imgu_v4l2_buffer_done(vb, state);
+	mutex_unlock(&imgu->lock);
+}
+
+static struct ipu3_css_buffer *imgu_queue_getbuf(struct imgu_device *imgu,
+						 unsigned int node)
+{
+	struct imgu_buffer *buf;
+
+	if (WARN_ON(node >= IMGU_NODE_NUM))
+		return NULL;
+
+	/* Find first free buffer from the node */
+	list_for_each_entry(buf, &imgu->nodes[node].buffers, vid_buf.list) {
+		if (ipu3_css_buf_state(&buf->css_buf) == IPU3_CSS_BUFFER_NEW)
+			return &buf->css_buf;
+	}
+
+	/* There were no free buffers, try to return a dummy buffer */
+	return imgu_dummybufs_get(imgu, imgu_node_map[node].css_queue);
+}
+
+/*
+ * Queue as many buffers to CSS as possible. If all buffers don't fit into
+ * CSS buffer queues, they remain unqueued and will be queued later.
+ */
+int imgu_queue_buffers(struct imgu_device *imgu, bool initial)
+{
+	unsigned int node;
+	int r = 0;
+	struct imgu_buffer *ibuf;
+
+	if (!ipu3_css_is_streaming(&imgu->css))
+		return 0;
+
+	mutex_lock(&imgu->lock);
+
+	/* Buffer set is queued to FW only when input buffer is ready */
+	for (node = IMGU_NODE_NUM - 1;
+	     imgu_queue_getbuf(imgu, IMGU_NODE_IN);
+	     node = node ? node - 1 : IMGU_NODE_NUM - 1) {
+
+		if (node == IMGU_NODE_VF &&
+		    (imgu->css.pipe_id == IPU3_CSS_PIPE_ID_CAPTURE ||
+		     !imgu->nodes[IMGU_NODE_VF].enabled)) {
+			continue;
+		} else if (node == IMGU_NODE_PV &&
+			   (imgu->css.pipe_id == IPU3_CSS_PIPE_ID_VIDEO ||
+			    !imgu->nodes[IMGU_NODE_PV].enabled)) {
+			continue;
+		} else if (imgu->queue_enabled[node]) {
+			struct ipu3_css_buffer *buf =
+					imgu_queue_getbuf(imgu, node);
+			int dummy;
+
+			if (!buf)
+				break;
+
+			r = ipu3_css_buf_queue(&imgu->css, buf);
+			if (r)
+				break;
+			dummy = imgu_dummybufs_check(imgu, buf);
+			if (!dummy)
+				ibuf = container_of(buf, struct imgu_buffer,
+						    css_buf);
+			dev_dbg(&imgu->pci_dev->dev,
+				"queue %s %s buffer %d to css da: 0x%08x\n",
+				dummy ? "dummy" : "user",
+				imgu_node_map[node].name,
+				dummy ? 0 : ibuf->vid_buf.vbb.vb2_buf.index,
+				(u32)buf->daddr);
+		}
+	}
+	mutex_unlock(&imgu->lock);
+
+	if (r && r != -EBUSY)
+		goto failed;
+
+	return 0;
+
+failed:
+	/*
+	 * On error, mark all buffers as failed which are not
+	 * yet queued to CSS
+	 */
+	dev_err(&imgu->pci_dev->dev,
+		"failed to queue buffer to CSS on queue %i (%d)\n",
+		node, r);
+
+	if (initial)
+		/* If we were called from streamon(), no need to finish bufs */
+		return r;
+
+	for (node = 0; node < IMGU_NODE_NUM; node++) {
+		struct imgu_buffer *buf, *buf0;
+
+		if (!imgu->queue_enabled[node])
+			continue;	/* Skip disabled queues */
+
+		mutex_lock(&imgu->lock);
+		list_for_each_entry_safe(buf, buf0, &imgu->nodes[node].buffers,
+					 vid_buf.list) {
+			if (ipu3_css_buf_state(&buf->css_buf) ==
+					IPU3_CSS_BUFFER_QUEUED)
+				continue;	/* Was already queued, skip */
+
+			imgu_v4l2_buffer_done(&buf->vid_buf.vbb.vb2_buf,
+					      VB2_BUF_STATE_ERROR);
+		}
+		mutex_unlock(&imgu->lock);
+	}
+
+	return r;
+}
+
+static int imgu_powerup(struct imgu_device *imgu)
+{
+	int r;
+
+	r = ipu3_css_set_powerup(&imgu->pci_dev->dev, imgu->base);
+	if (r)
+		return r;
+
+	ipu3_mmu_resume(imgu->mmu);
+	return 0;
+}
+
+static void imgu_powerdown(struct imgu_device *imgu)
+{
+	ipu3_mmu_suspend(imgu->mmu);
+	ipu3_css_set_powerdown(&imgu->pci_dev->dev, imgu->base);
+}
+
+int imgu_s_stream(struct imgu_device *imgu, int enable)
+{
+	struct device *dev = &imgu->pci_dev->dev;
+	struct v4l2_pix_format_mplane *fmts[IPU3_CSS_QUEUES] = { NULL };
+	struct v4l2_rect *rects[IPU3_CSS_RECTS] = { NULL };
+	unsigned int i, node;
+	int r;
+
+	if (!enable) {
+		/* Stop streaming */
+		dev_dbg(dev, "stream off\n");
+		/* Block new buffers to be queued to CSS. */
+		atomic_set(&imgu->qbuf_barrier, 1);
+		ipu3_css_stop_streaming(&imgu->css);
+		synchronize_irq(imgu->pci_dev->irq);
+		atomic_set(&imgu->qbuf_barrier, 0);
+		imgu_powerdown(imgu);
+		pm_runtime_put(&imgu->pci_dev->dev);
+
+		return 0;
+	}
+
+	/* Start streaming */
+
+	dev_dbg(dev, "stream on\n");
+	for (i = 0; i < IMGU_NODE_NUM; i++)
+		imgu->queue_enabled[i] = imgu->nodes[i].enabled;
+
+	/*
+	 * CSS library expects that the following queues are
+	 * always enabled; if buffers are not provided to some of the
+	 * queues, it stalls due to lack of buffers.
+	 * Force the queues to be enabled and if the user really hasn't
+	 * enabled them, use dummy buffers.
+	 */
+	imgu->queue_enabled[IMGU_NODE_OUT] = true;
+	imgu->queue_enabled[IMGU_NODE_VF] = true;
+	imgu->queue_enabled[IMGU_NODE_PV] = true;
+	imgu->queue_enabled[IMGU_NODE_STAT_3A] = true;
+
+	/* This is handled specially */
+	imgu->queue_enabled[IPU3_CSS_QUEUE_PARAMS] = false;
+
+	/* Initialize CSS formats */
+	for (i = 0; i < IPU3_CSS_QUEUES; i++) {
+		node = imgu_map_node(imgu, i);
+		/* No need to reconfig meta nodes */
+		if (node == IMGU_NODE_STAT_3A || node == IMGU_NODE_PARAMS)
+			continue;
+		fmts[i] = imgu->queue_enabled[node] ?
+			&imgu->nodes[node].vdev_fmt.fmt.pix_mp : NULL;
+	}
+
+	/* Enable VF output only when VF or PV queue requested by user */
+	imgu->css.vf_output_en = IPU3_NODE_VF_DISABLED;
+	if (imgu->nodes[IMGU_NODE_VF].enabled)
+		imgu->css.vf_output_en = IPU3_NODE_VF_ENABLED;
+	else if (imgu->nodes[IMGU_NODE_PV].enabled)
+		imgu->css.vf_output_en = IPU3_NODE_PV_ENABLED;
+
+	rects[IPU3_CSS_RECT_EFFECTIVE] = &imgu->rect.eff;
+	rects[IPU3_CSS_RECT_BDS] = &imgu->rect.bds;
+	rects[IPU3_CSS_RECT_GDC] = &imgu->rect.gdc;
+
+	r = ipu3_css_fmt_set(&imgu->css, fmts, rects);
+	if (r) {
+		dev_err(dev, "failed to set initial formats (%d)", r);
+		return r;
+	}
+
+	/* Set Power */
+	r = pm_runtime_get_sync(dev);
+	if (r < 0) {
+		dev_err(dev, "failed to set imgu power\n");
+		pm_runtime_put(dev);
+		return r;
+	}
+
+	r = imgu_powerup(imgu);
+	if (r) {
+		dev_err(dev, "failed to power up imgu\n");
+		pm_runtime_put(dev);
+		return r;
+	}
+
+	/* Start CSS streaming */
+	r = ipu3_css_start_streaming(&imgu->css);
+	if (r) {
+		dev_err(dev, "failed to start css streaming (%d)", r);
+		goto fail_start_streaming;
+	}
+
+	/* Initialize dummy buffers */
+	r = imgu_dummybufs_init(imgu);
+	if (r) {
+		dev_err(dev, "failed to initialize dummy buffers (%d)", r);
+		goto fail_dummybufs;
+	}
+
+	/* Queue as many buffers from queue as possible */
+	r = imgu_queue_buffers(imgu, true);
+	if (r) {
+		dev_err(dev, "failed to queue initial buffers (%d)", r);
+		goto fail_queueing;
+	}
+
+	return 0;
+
+fail_queueing:
+	imgu_dummybufs_cleanup(imgu);
+fail_dummybufs:
+	ipu3_css_stop_streaming(&imgu->css);
+fail_start_streaming:
+	pm_runtime_put(dev);
+
+	return r;
+}
+
+static int imgu_video_nodes_init(struct imgu_device *imgu)
+{
+	struct v4l2_pix_format_mplane *fmts[IPU3_CSS_QUEUES] = { NULL };
+	struct v4l2_rect *rects[IPU3_CSS_RECTS] = { NULL };
+	unsigned int i;
+	int r;
+
+	imgu->buf_struct_size = sizeof(struct imgu_buffer);
+
+	for (i = 0; i < IMGU_NODE_NUM; i++) {
+		imgu->nodes[i].name = imgu_node_map[i].name;
+		imgu->nodes[i].output = i < IMGU_QUEUE_FIRST_INPUT;
+		imgu->nodes[i].immutable = false;
+		imgu->nodes[i].enabled = false;
+
+		if (i != IMGU_NODE_PARAMS && i != IMGU_NODE_STAT_3A)
+			fmts[imgu_node_map[i].css_queue] =
+				&imgu->nodes[i].vdev_fmt.fmt.pix_mp;
+		atomic_set(&imgu->nodes[i].sequence, 0);
+	}
+
+	/* Master queue is always enabled */
+	imgu->nodes[IMGU_QUEUE_MASTER].immutable = true;
+	imgu->nodes[IMGU_QUEUE_MASTER].enabled = true;
+
+	r = imgu_v4l2_register(imgu);
+	if (r)
+		return r;
+
+	/* Set initial formats and initialize formats of video nodes */
+	rects[IPU3_CSS_RECT_EFFECTIVE] = &imgu->rect.eff;
+	rects[IPU3_CSS_RECT_BDS] = &imgu->rect.bds;
+	ipu3_css_fmt_set(&imgu->css, fmts, rects);
+
+	/* Pre-allocate dummy buffers */
+	r = imgu_dummybufs_preallocate(imgu);
+	if (r) {
+		dev_err(&imgu->pci_dev->dev,
+			"failed to pre-allocate dummy buffers (%d)", r);
+		imgu_dummybufs_cleanup(imgu);
+		imgu_v4l2_unregister(imgu);
+	}
+
+	return 0;
+}
+
+static void imgu_video_nodes_exit(struct imgu_device *imgu)
+{
+	imgu_dummybufs_cleanup(imgu);
+	imgu_v4l2_unregister(imgu);
+}
+
+/**************** PCI interface ****************/
+
+static irqreturn_t imgu_isr_threaded(int irq, void *imgu_ptr)
+{
+	struct imgu_device *imgu = imgu_ptr;
+
+	/* Dequeue / queue buffers */
+	do {
+		u64 ns = ktime_get_ns();
+		struct ipu3_css_buffer *b;
+		struct imgu_buffer *buf;
+		unsigned int node;
+		bool dummy;
+
+		do {
+			mutex_lock(&imgu->lock);
+			b = ipu3_css_buf_dequeue(&imgu->css);
+			mutex_unlock(&imgu->lock);
+		} while (PTR_ERR(b) == -EAGAIN);
+
+		if (IS_ERR_OR_NULL(b)) {
+			if (!b || PTR_ERR(b) == -EBUSY)	/* All done */
+				break;
+			dev_err(&imgu->pci_dev->dev,
+				"failed to dequeue buffers (%ld)\n",
+				PTR_ERR(b));
+			break;
+		}
+
+		node = imgu_map_node(imgu, b->queue);
+		dummy = imgu_dummybufs_check(imgu, b);
+		if (!dummy)
+			buf = container_of(b, struct imgu_buffer, css_buf);
+		dev_dbg(&imgu->pci_dev->dev,
+			"dequeue %s %s buffer %d from css\n",
+			dummy ? "dummy" : "user",
+			imgu_node_map[node].name,
+			dummy ? 0 : buf->vid_buf.vbb.vb2_buf.index);
+
+		if (dummy)
+			/* It was a dummy buffer, skip it */
+			continue;
+
+		/* Fill vb2 buffer entries and tell it's ready */
+		if (!imgu->nodes[node].output) {
+			buf->vid_buf.vbb.vb2_buf.timestamp = ns;
+			buf->vid_buf.vbb.field = V4L2_FIELD_NONE;
+			buf->vid_buf.vbb.sequence =
+				atomic_inc_return(&imgu->nodes[node].sequence);
+		}
+		imgu_buffer_done(imgu, &buf->vid_buf.vbb.vb2_buf,
+				 ipu3_css_buf_state(&buf->css_buf) ==
+						    IPU3_CSS_BUFFER_DONE ?
+						    VB2_BUF_STATE_DONE :
+						    VB2_BUF_STATE_ERROR);
+		mutex_lock(&imgu->lock);
+		if (ipu3_css_queue_empty(&imgu->css))
+			wake_up_all(&imgu->buf_drain_wq);
+		mutex_unlock(&imgu->lock);
+	} while (1);
+
+	/*
+	 * Try to queue more buffers for CSS.
+	 * qbuf_barrier is used to disable new buffers
+	 * to be queued to CSS.
+	 */
+	if (!atomic_read(&imgu->qbuf_barrier))
+		imgu_queue_buffers(imgu, false);
+
+	return IRQ_HANDLED;
+}
+
+static irqreturn_t imgu_isr(int irq, void *imgu_ptr)
+{
+	struct imgu_device *imgu = imgu_ptr;
+
+	/* acknowledge interruption */
+	if (ipu3_css_irq_ack(&imgu->css) < 0)
+		return IRQ_NONE;
+
+	return IRQ_WAKE_THREAD;
+}
+
+static int imgu_pci_config_setup(struct pci_dev *dev)
+{
+	u16 pci_command;
+	int r = pci_enable_msi(dev);
+
+	if (r) {
+		dev_err(&dev->dev, "failed to enable MSI (%d)\n", r);
+		return r;
+	}
+
+	pci_read_config_word(dev, PCI_COMMAND, &pci_command);
+	pci_command |= PCI_COMMAND_MEMORY | PCI_COMMAND_MASTER |
+			PCI_COMMAND_INTX_DISABLE;
+	pci_write_config_word(dev, PCI_COMMAND, pci_command);
+
+	return 0;
+}
+
+static int imgu_pci_probe(struct pci_dev *pci_dev,
+			  const struct pci_device_id *id)
+{
+	struct imgu_device *imgu;
+	phys_addr_t phys;
+	unsigned long phys_len;
+	void __iomem *const *iomap;
+	int r;
+
+	imgu = devm_kzalloc(&pci_dev->dev, sizeof(*imgu), GFP_KERNEL);
+	if (!imgu)
+		return -ENOMEM;
+
+	imgu->pci_dev = pci_dev;
+
+	r = pcim_enable_device(pci_dev);
+	if (r) {
+		dev_err(&pci_dev->dev, "failed to enable device (%d)\n", r);
+		return r;
+	}
+
+	dev_info(&pci_dev->dev, "device 0x%x (rev: 0x%x)\n",
+		 pci_dev->device, pci_dev->revision);
+
+	phys = pci_resource_start(pci_dev, IMGU_PCI_BAR);
+	phys_len = pci_resource_len(pci_dev, IMGU_PCI_BAR);
+
+	r = pcim_iomap_regions(pci_dev, 1 << IMGU_PCI_BAR, pci_name(pci_dev));
+	if (r) {
+		dev_err(&pci_dev->dev, "failed to remap I/O memory (%d)\n", r);
+		return r;
+	}
+	dev_info(&pci_dev->dev, "physical base address %pap, %lu bytes\n",
+		 &phys, phys_len);
+
+	iomap = pcim_iomap_table(pci_dev);
+	if (!iomap) {
+		dev_err(&pci_dev->dev, "failed to iomap table\n");
+		return -ENODEV;
+	}
+
+	imgu->base = iomap[IMGU_PCI_BAR];
+
+	pci_set_drvdata(pci_dev, imgu);
+
+	pci_set_master(pci_dev);
+
+	r = dma_coerce_mask_and_coherent(&pci_dev->dev, IMGU_DMA_MASK);
+	if (r) {
+		dev_err(&pci_dev->dev, "failed to set DMA mask (%d)\n", r);
+		return -ENODEV;
+	}
+
+	r = imgu_pci_config_setup(pci_dev);
+	if (r)
+		return r;
+
+	mutex_init(&imgu->lock);
+	atomic_set(&imgu->qbuf_barrier, 0);
+	init_waitqueue_head(&imgu->buf_drain_wq);
+
+	r = ipu3_css_set_powerup(&pci_dev->dev, imgu->base);
+	if (r) {
+		dev_err(&pci_dev->dev,
+			"failed to power up CSS (%d)\n", r);
+		goto out_mutex_destroy;
+	}
+
+	imgu->mmu = ipu3_mmu_init(&pci_dev->dev, imgu->base);
+	if (IS_ERR(imgu->mmu)) {
+		r = PTR_ERR(imgu->mmu);
+		dev_err(&pci_dev->dev, "failed to initialize MMU (%d)\n", r);
+		goto out_css_powerdown;
+	}
+
+	r = ipu3_dmamap_init(imgu);
+	if (r) {
+		dev_err(&pci_dev->dev,
+			"failed to initialize DMA mapping (%d)\n", r);
+		goto out_mmu_exit;
+	}
+
+	/* ISP programming */
+	r = ipu3_css_init(&pci_dev->dev, &imgu->css, imgu->base, phys_len);
+	if (r) {
+		dev_err(&pci_dev->dev, "failed to initialize CSS (%d)\n", r);
+		goto out_dmamap_exit;
+	}
+
+	/* v4l2 sub-device registration */
+	r = imgu_video_nodes_init(imgu);
+	if (r) {
+		dev_err(&pci_dev->dev, "failed to create V4L2 devices (%d)\n",
+			r);
+		goto out_css_cleanup;
+	}
+
+	r = devm_request_threaded_irq(&pci_dev->dev, pci_dev->irq,
+				      imgu_isr, imgu_isr_threaded,
+				      IRQF_SHARED, IMGU_NAME, imgu);
+	if (r) {
+		dev_err(&pci_dev->dev, "failed to request IRQ (%d)\n", r);
+		goto out_video_exit;
+	}
+
+	pm_runtime_put_noidle(&pci_dev->dev);
+	pm_runtime_allow(&pci_dev->dev);
+
+	return 0;
+
+out_video_exit:
+	imgu_video_nodes_exit(imgu);
+out_css_cleanup:
+	ipu3_css_cleanup(&imgu->css);
+out_dmamap_exit:
+	ipu3_dmamap_exit(imgu);
+out_mmu_exit:
+	ipu3_mmu_exit(imgu->mmu);
+out_css_powerdown:
+	ipu3_css_set_powerdown(&pci_dev->dev, imgu->base);
+out_mutex_destroy:
+	mutex_destroy(&imgu->lock);
+
+	return r;
+}
+
+static void imgu_pci_remove(struct pci_dev *pci_dev)
+{
+	struct imgu_device *imgu = pci_get_drvdata(pci_dev);
+
+	pm_runtime_forbid(&pci_dev->dev);
+	pm_runtime_get_noresume(&pci_dev->dev);
+
+	imgu_video_nodes_exit(imgu);
+	ipu3_css_cleanup(&imgu->css);
+	ipu3_css_set_powerdown(&pci_dev->dev, imgu->base);
+	ipu3_dmamap_exit(imgu);
+	ipu3_mmu_exit(imgu->mmu);
+	mutex_destroy(&imgu->lock);
+}
+
+static int __maybe_unused imgu_suspend(struct device *dev)
+{
+	struct pci_dev *pci_dev = to_pci_dev(dev);
+	struct imgu_device *imgu = pci_get_drvdata(pci_dev);
+
+	dev_dbg(dev, "enter %s\n", __func__);
+	imgu->suspend_in_stream = ipu3_css_is_streaming(&imgu->css);
+	if (!imgu->suspend_in_stream)
+		goto out;
+	/* Block new buffers to be queued to CSS. */
+	atomic_set(&imgu->qbuf_barrier, 1);
+	/*
+	 * Wait for currently running irq handler to be done so that
+	 * no new buffers will be queued to fw later.
+	 */
+	synchronize_irq(pci_dev->irq);
+	/* Wait until all buffers in CSS are done. */
+	if (!wait_event_timeout(imgu->buf_drain_wq,
+	    ipu3_css_queue_empty(&imgu->css), msecs_to_jiffies(1000)))
+		dev_err(dev, "wait buffer drain timeout.\n");
+
+	ipu3_css_stop_streaming(&imgu->css);
+	atomic_set(&imgu->qbuf_barrier, 0);
+	imgu_powerdown(imgu);
+	pm_runtime_force_suspend(dev);
+out:
+	dev_dbg(dev, "leave %s\n", __func__);
+	return 0;
+}
+
+static int __maybe_unused imgu_resume(struct device *dev)
+{
+	struct pci_dev *pci_dev = to_pci_dev(dev);
+	struct imgu_device *imgu = pci_get_drvdata(pci_dev);
+	int r = 0;
+
+	dev_dbg(dev, "enter %s\n", __func__);
+
+	if (!imgu->suspend_in_stream)
+		goto out;
+
+	pm_runtime_force_resume(dev);
+
+	r = imgu_powerup(imgu);
+	if (r) {
+		dev_err(dev, "failed to power up imgu\n");
+		goto out;
+	}
+
+	/* Start CSS streaming */
+	r = ipu3_css_start_streaming(&imgu->css);
+	if (r) {
+		dev_err(dev, "failed to resume css streaming (%d)", r);
+		goto out;
+	}
+
+	r = imgu_queue_buffers(imgu, true);
+	if (r)
+		dev_err(dev, "failed to queue buffers (%d)", r);
+out:
+	dev_dbg(dev, "leave %s\n", __func__);
+
+	return r;
+}
+
+/*
+ * PCI rpm framework checks the existence of driver rpm callbacks.
+ * Place a dummy callback here to avoid rpm going into error state.
+ */
+static int imgu_rpm_dummy_cb(struct device *dev)
+{
+	return 0;
+}
+
+static const struct dev_pm_ops imgu_pm_ops = {
+	SET_RUNTIME_PM_OPS(&imgu_rpm_dummy_cb, &imgu_rpm_dummy_cb, NULL)
+	SET_SYSTEM_SLEEP_PM_OPS(&imgu_suspend, &imgu_resume)
+};
+
+static const struct pci_device_id imgu_pci_tbl[] = {
+	{ PCI_DEVICE(PCI_VENDOR_ID_INTEL, IMGU_PCI_ID) },
+	{ 0, }
+};
+
+MODULE_DEVICE_TABLE(pci, imgu_pci_tbl);
+
+static struct pci_driver imgu_pci_driver = {
+	.name = IMGU_NAME,
+	.id_table = imgu_pci_tbl,
+	.probe = imgu_pci_probe,
+	.remove = imgu_pci_remove,
+	.driver = {
+		.pm = &imgu_pm_ops,
+	},
+};
+
+module_pci_driver(imgu_pci_driver);
+
+MODULE_AUTHOR("Tuukka Toivonen <tuukka.toivonen@intel.com>");
+MODULE_AUTHOR("Tianshu Qiu <tian.shu.qiu@intel.com>");
+MODULE_AUTHOR("Jian Xu Zheng <jian.xu.zheng@intel.com>");
+MODULE_AUTHOR("Yuning Pu <yuning.pu@intel.com>");
+MODULE_AUTHOR("Yong Zhi <yong.zhi@intel.com>");
+MODULE_LICENSE("GPL v2");
+MODULE_DESCRIPTION("Intel ipu3_imgu PCI driver");
diff --git a/drivers/staging/media/ipu3/ipu3.h b/drivers/staging/media/ipu3/ipu3.h
new file mode 100644
index 000000000000..2e0c756a9e65
--- /dev/null
+++ b/drivers/staging/media/ipu3/ipu3.h
@@ -0,0 +1,152 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/* Copyright (C) 2018 Intel Corporation */
+
+#ifndef __IPU3_H
+#define __IPU3_H
+
+#include <linux/iova.h>
+#include <linux/pci.h>
+
+#include <media/v4l2-device.h>
+#include <media/videobuf2-dma-sg.h>
+
+#include "ipu3-css.h"
+
+#define IMGU_NAME			"ipu3-imgu"
+
+/*
+ * The semantics of the driver is that whenever there is a buffer available in
+ * master queue, the driver queues a buffer also to all other active nodes.
+ * If user space hasn't provided a buffer to all other video nodes first,
+ * the driver gets an internal dummy buffer and queues it.
+ */
+#define IMGU_QUEUE_MASTER		IPU3_CSS_QUEUE_IN
+#define IMGU_QUEUE_FIRST_INPUT		IPU3_CSS_QUEUE_OUT
+#define IMGU_MAX_QUEUE_DEPTH		(2 + 2)
+
+#define IMGU_NODE_IN			0 /* Input RAW image */
+#define IMGU_NODE_PARAMS		1 /* Input parameters */
+#define IMGU_NODE_OUT			2 /* Main output for still or video */
+#define IMGU_NODE_VF			3 /* Preview */
+#define IMGU_NODE_PV			4 /* Postview for still capture */
+#define IMGU_NODE_STAT_3A		5 /* 3A statistics */
+#define IMGU_NODE_NUM			6
+
+#define file_to_intel_ipu3_node(__file) \
+	container_of(video_devdata(__file), struct imgu_video_device, vdev)
+
+#define IPU3_INPUT_MIN_WIDTH		0U
+#define IPU3_INPUT_MIN_HEIGHT		0U
+#define IPU3_INPUT_MAX_WIDTH		5120U
+#define IPU3_INPUT_MAX_HEIGHT		38404U
+#define IPU3_OUTPUT_MIN_WIDTH		2U
+#define IPU3_OUTPUT_MIN_HEIGHT		2U
+#define IPU3_OUTPUT_MAX_WIDTH		4480U
+#define IPU3_OUTPUT_MAX_HEIGHT		34004U
+
+struct ipu3_vb2_buffer {
+	/* Public fields */
+	struct vb2_v4l2_buffer vbb;	/* Must be the first field */
+
+	/* Private fields */
+	struct list_head list;
+};
+
+struct imgu_buffer {
+	struct ipu3_vb2_buffer vid_buf;	/* Must be the first field */
+	struct ipu3_css_buffer css_buf;
+	struct ipu3_css_map map;
+};
+
+struct imgu_node_mapping {
+	unsigned int css_queue;
+	const char *name;
+};
+
+/**
+ * struct imgu_video_device
+ * each node registers as video device and maintains its
+ * own vb2_queue.
+ */
+struct imgu_video_device {
+	const char *name;
+	bool output;
+	bool immutable;		/* Can not be enabled/disabled */
+	bool enabled;
+	struct v4l2_format vdev_fmt;	/* Currently set format */
+
+	/* Private fields */
+	struct video_device vdev;
+	struct media_pad vdev_pad;
+	struct v4l2_mbus_framefmt pad_fmt;
+	struct vb2_queue vbq;
+	struct list_head buffers;
+	/* Protect vb2_queue and vdev structs*/
+	struct mutex lock;
+	atomic_t sequence;
+};
+
+/*
+ * imgu_device -- ImgU (Imaging Unit) driver
+ */
+struct imgu_device {
+	struct pci_dev *pci_dev;
+	void __iomem *base;
+
+	/* Internally enabled queues */
+	struct {
+		struct ipu3_css_map dmap;
+		struct ipu3_css_buffer dummybufs[IMGU_MAX_QUEUE_DEPTH];
+	} queues[IPU3_CSS_QUEUES];
+	struct imgu_video_device nodes[IMGU_NODE_NUM];
+	bool queue_enabled[IMGU_NODE_NUM];
+
+	/* Public fields, fill before registering */
+	unsigned int buf_struct_size;
+	bool streaming;		/* Public read only */
+	struct v4l2_ctrl_handler *ctrl_handler;
+
+	/* Private fields */
+	struct v4l2_device v4l2_dev;
+	struct media_device media_dev;
+	struct media_pipeline pipeline;
+	struct v4l2_subdev subdev;
+	struct media_pad subdev_pads[IMGU_NODE_NUM];
+	struct v4l2_file_operations v4l2_file_ops;
+
+	/* MMU driver for css */
+	struct ipu3_mmu_info *mmu;
+	struct iova_domain iova_domain;
+
+	/* css - Camera Sub-System */
+	struct ipu3_css css;
+
+	/*
+	 * Coarse-grained lock to protect
+	 * vid_buf.list and css->queue
+	 */
+	struct mutex lock;
+	/* Forbit streaming and buffer queuing during system suspend. */
+	atomic_t qbuf_barrier;
+	struct {
+		struct v4l2_rect eff; /* effective resolution */
+		struct v4l2_rect bds; /* bayer-domain scaled resolution*/
+		struct v4l2_rect gdc; /* gdc output resolution */
+	} rect;
+	/* Indicate if system suspend take place while imgu is streaming. */
+	bool suspend_in_stream;
+	/* Used to wait for FW buffer queue drain. */
+	wait_queue_head_t buf_drain_wq;
+};
+
+unsigned int imgu_node_to_queue(unsigned int node);
+unsigned int imgu_map_node(struct imgu_device *imgu, unsigned int css_queue);
+int imgu_queue_buffers(struct imgu_device *imgu, bool initial);
+
+int imgu_v4l2_register(struct imgu_device *dev);
+int imgu_v4l2_unregister(struct imgu_device *dev);
+void imgu_v4l2_buffer_done(struct vb2_buffer *vb, enum vb2_buffer_state state);
+
+int imgu_s_stream(struct imgu_device *imgu, int enable);
+
+#endif