staging: media: Add Raspberry Pi V4L2 H265 decoder

raspberrypi inclusion
category: feature
bugzilla: 50432

--------------------------------

This driver is for the HEVC/H265 decoder block on the Raspberry
Pi 4, and conforms to the V4L2 stateless decoder API.
Signed-off-by: NJohn Cox <jc@kynesim.co.uk>
Signed-off-by: NFang Yafen <yafen@iscas.ac.cn>
Signed-off-by: NZheng Zengkai <zhengzengkai@huawei.com>

drivers/staging/media/Kconfig

@@ -34,6 +34,8 @@ source "drivers/staging/media/omap4iss/Kconfig"
 
 source "drivers/staging/media/rkvdec/Kconfig"
 
+source "drivers/staging/media/rpivid/Kconfig"
+
 source "drivers/staging/media/sunxi/Kconfig"
 
 source "drivers/staging/media/tegra-vde/Kconfig"

drivers/staging/media/Makefile

@@ -5,6 +5,7 @@ obj-$(CONFIG_VIDEO_IMX_MEDIA)	+= imx/
 obj-$(CONFIG_VIDEO_MESON_VDEC)	+= meson/vdec/
 obj-$(CONFIG_VIDEO_OMAP4)	+= omap4iss/
 obj-$(CONFIG_VIDEO_ROCKCHIP_VDEC)	+= rkvdec/
+obj-$(CONFIG_VIDEO_RPIVID)	+= rpivid/
 obj-$(CONFIG_VIDEO_SUNXI)	+= sunxi/
 obj-$(CONFIG_VIDEO_TEGRA)	+= tegra-video/
 obj-$(CONFIG_TEGRA_VDE)		+= tegra-vde/

drivers/staging/media/rpivid/Kconfig

+# SPDX-License-Identifier: GPL-2.0
+
+config VIDEO_RPIVID
+	tristate "Rpi H265 driver"
+	depends on VIDEO_DEV && VIDEO_V4L2
+	depends on MEDIA_CONTROLLER
+	depends on OF
+	depends on MEDIA_CONTROLLER_REQUEST_API
+	select VIDEOBUF2_DMA_CONTIG
+	select V4L2_MEM2MEM_DEV
+	help
+	  Support for the Rpi H265 h/w decoder.
+
+	  To compile this driver as a module, choose M here: the module
+	  will be called rpivid-hevc.
+

drivers/staging/media/rpivid/Makefile

+# SPDX-License-Identifier: GPL-2.0
+obj-$(CONFIG_VIDEO_RPIVID) += rpivid-hevc.o
+
+rpivid-hevc-y = rpivid.o rpivid_video.o rpivid_dec.o \
+		 rpivid_hw.o rpivid_h265.o

drivers/staging/media/rpivid/rpivid.c

+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Raspberry Pi HEVC driver
+ *
+ * Copyright (C) 2020 Raspberry Pi (Trading) Ltd
+ *
+ * Based on the Cedrus VPU driver, that is:
+ *
+ * Copyright (C) 2016 Florent Revest <florent.revest@free-electrons.com>
+ * Copyright (C) 2018 Paul Kocialkowski <paul.kocialkowski@bootlin.com>
+ * Copyright (C) 2018 Bootlin
+ */
+
+#include <linux/platform_device.h>
+#include <linux/module.h>
+#include <linux/of.h>
+
+#include <media/v4l2-device.h>
+#include <media/v4l2-ioctl.h>
+#include <media/v4l2-ctrls.h>
+#include <media/v4l2-mem2mem.h>
+
+#include "rpivid.h"
+#include "rpivid_video.h"
+#include "rpivid_hw.h"
+#include "rpivid_dec.h"
+
+/*
+ * Default /dev/videoN node number.
+ * Deliberately avoid the very low numbers as these are often taken by webcams
+ * etc, and simple apps tend to only go for /dev/video0.
+ */
+static int video_nr = 19;
+module_param(video_nr, int, 0644);
+MODULE_PARM_DESC(video_nr, "decoder video device number");
+
+static const struct rpivid_control rpivid_ctrls[] = {
+	{
+		.cfg = {
+			.id	= V4L2_CID_MPEG_VIDEO_HEVC_SPS,
+		},
+		.required	= true,
+	},
+	{
+		.cfg = {
+			.id	= V4L2_CID_MPEG_VIDEO_HEVC_PPS,
+		},
+		.required	= true,
+	},
+	{
+		.cfg = {
+			.id = V4L2_CID_MPEG_VIDEO_HEVC_SCALING_MATRIX,
+		},
+		.required	= false,
+	},
+	{
+		.cfg = {
+			.id	= V4L2_CID_MPEG_VIDEO_HEVC_SLICE_PARAMS,
+		},
+		.required	= true,
+	},
+	{
+		.cfg = {
+			.id	= V4L2_CID_MPEG_VIDEO_HEVC_DECODE_MODE,
+			.max	= V4L2_MPEG_VIDEO_HEVC_DECODE_MODE_SLICE_BASED,
+			.def	= V4L2_MPEG_VIDEO_HEVC_DECODE_MODE_SLICE_BASED,
+		},
+		.required	= false,
+	},
+	{
+		.cfg = {
+			.id	= V4L2_CID_MPEG_VIDEO_HEVC_START_CODE,
+			.max	= V4L2_MPEG_VIDEO_HEVC_START_CODE_NONE,
+			.def	= V4L2_MPEG_VIDEO_HEVC_START_CODE_NONE,
+		},
+		.required	= false,
+	},
+};
+
+#define rpivid_ctrls_COUNT	ARRAY_SIZE(rpivid_ctrls)
+
+void *rpivid_find_control_data(struct rpivid_ctx *ctx, u32 id)
+{
+	unsigned int i;
+
+	for (i = 0; ctx->ctrls[i]; i++)
+		if (ctx->ctrls[i]->id == id)
+			return ctx->ctrls[i]->p_cur.p;
+
+	return NULL;
+}
+
+static int rpivid_init_ctrls(struct rpivid_dev *dev, struct rpivid_ctx *ctx)
+{
+	struct v4l2_ctrl_handler *hdl = &ctx->hdl;
+	struct v4l2_ctrl *ctrl;
+	unsigned int ctrl_size;
+	unsigned int i;
+
+	v4l2_ctrl_handler_init(hdl, rpivid_ctrls_COUNT);
+	if (hdl->error) {
+		v4l2_err(&dev->v4l2_dev,
+			 "Failed to initialize control handler\n");
+		return hdl->error;
+	}
+
+	ctrl_size = sizeof(ctrl) * rpivid_ctrls_COUNT + 1;
+
+	ctx->ctrls = kzalloc(ctrl_size, GFP_KERNEL);
+	if (!ctx->ctrls)
+		return -ENOMEM;
+
+	for (i = 0; i < rpivid_ctrls_COUNT; i++) {
+		ctrl = v4l2_ctrl_new_custom(hdl, &rpivid_ctrls[i].cfg,
+					    NULL);
+		if (hdl->error) {
+			v4l2_err(&dev->v4l2_dev,
+				 "Failed to create new custom control id=%#x\n",
+				 rpivid_ctrls[i].cfg.id);
+
+			v4l2_ctrl_handler_free(hdl);
+			kfree(ctx->ctrls);
+			return hdl->error;
+		}
+
+		ctx->ctrls[i] = ctrl;
+	}
+
+	ctx->fh.ctrl_handler = hdl;
+	v4l2_ctrl_handler_setup(hdl);
+
+	return 0;
+}
+
+static int rpivid_request_validate(struct media_request *req)
+{
+	struct media_request_object *obj;
+	struct v4l2_ctrl_handler *parent_hdl, *hdl;
+	struct rpivid_ctx *ctx = NULL;
+	struct v4l2_ctrl *ctrl_test;
+	unsigned int count;
+	unsigned int i;
+
+	list_for_each_entry(obj, &req->objects, list) {
+		struct vb2_buffer *vb;
+
+		if (vb2_request_object_is_buffer(obj)) {
+			vb = container_of(obj, struct vb2_buffer, req_obj);
+			ctx = vb2_get_drv_priv(vb->vb2_queue);
+
+			break;
+		}
+	}
+
+	if (!ctx)
+		return -ENOENT;
+
+	count = vb2_request_buffer_cnt(req);
+	if (!count) {
+		v4l2_info(&ctx->dev->v4l2_dev,
+			  "No buffer was provided with the request\n");
+		return -ENOENT;
+	} else if (count > 1) {
+		v4l2_info(&ctx->dev->v4l2_dev,
+			  "More than one buffer was provided with the request\n");
+		return -EINVAL;
+	}
+
+	parent_hdl = &ctx->hdl;
+
+	hdl = v4l2_ctrl_request_hdl_find(req, parent_hdl);
+	if (!hdl) {
+		v4l2_info(&ctx->dev->v4l2_dev, "Missing codec control(s)\n");
+		return -ENOENT;
+	}
+
+	for (i = 0; i < rpivid_ctrls_COUNT; i++) {
+		if (!rpivid_ctrls[i].required)
+			continue;
+
+		ctrl_test =
+			v4l2_ctrl_request_hdl_ctrl_find(hdl,
+							rpivid_ctrls[i].cfg.id);
+		if (!ctrl_test) {
+			v4l2_info(&ctx->dev->v4l2_dev,
+				  "Missing required codec control\n");
+			return -ENOENT;
+		}
+	}
+
+	v4l2_ctrl_request_hdl_put(hdl);
+
+	return vb2_request_validate(req);
+}
+
+static int rpivid_open(struct file *file)
+{
+	struct rpivid_dev *dev = video_drvdata(file);
+	struct rpivid_ctx *ctx = NULL;
+	int ret;
+
+	if (mutex_lock_interruptible(&dev->dev_mutex))
+		return -ERESTARTSYS;
+
+	ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
+	if (!ctx) {
+		mutex_unlock(&dev->dev_mutex);
+		return -ENOMEM;
+	}
+
+	v4l2_fh_init(&ctx->fh, video_devdata(file));
+	file->private_data = &ctx->fh;
+	ctx->dev = dev;
+
+	ret = rpivid_init_ctrls(dev, ctx);
+	if (ret)
+		goto err_free;
+
+	ctx->fh.m2m_ctx = v4l2_m2m_ctx_init(dev->m2m_dev, ctx,
+					    &rpivid_queue_init);
+	if (IS_ERR(ctx->fh.m2m_ctx)) {
+		ret = PTR_ERR(ctx->fh.m2m_ctx);
+		goto err_ctrls;
+	}
+
+	/* The only bit of format info that we can guess now is H265 src
+	 * Everything else we need more info for
+	 */
+	ctx->src_fmt.pixelformat = RPIVID_SRC_PIXELFORMAT_DEFAULT;
+	rpivid_prepare_src_format(&ctx->src_fmt);
+
+	v4l2_fh_add(&ctx->fh);
+
+	mutex_unlock(&dev->dev_mutex);
+
+	return 0;
+
+err_ctrls:
+	v4l2_ctrl_handler_free(&ctx->hdl);
+err_free:
+	kfree(ctx);
+	mutex_unlock(&dev->dev_mutex);
+
+	return ret;
+}
+
+static int rpivid_release(struct file *file)
+{
+	struct rpivid_dev *dev = video_drvdata(file);
+	struct rpivid_ctx *ctx = container_of(file->private_data,
+					      struct rpivid_ctx, fh);
+
+	mutex_lock(&dev->dev_mutex);
+
+	v4l2_fh_del(&ctx->fh);
+	v4l2_m2m_ctx_release(ctx->fh.m2m_ctx);
+
+	v4l2_ctrl_handler_free(&ctx->hdl);
+	kfree(ctx->ctrls);
+
+	v4l2_fh_exit(&ctx->fh);
+
+	kfree(ctx);
+
+	mutex_unlock(&dev->dev_mutex);
+
+	return 0;
+}
+
+static const struct v4l2_file_operations rpivid_fops = {
+	.owner		= THIS_MODULE,
+	.open		= rpivid_open,
+	.release	= rpivid_release,
+	.poll		= v4l2_m2m_fop_poll,
+	.unlocked_ioctl	= video_ioctl2,
+	.mmap		= v4l2_m2m_fop_mmap,
+};
+
+static const struct video_device rpivid_video_device = {
+	.name		= RPIVID_NAME,
+	.vfl_dir	= VFL_DIR_M2M,
+	.fops		= &rpivid_fops,
+	.ioctl_ops	= &rpivid_ioctl_ops,
+	.minor		= -1,
+	.release	= video_device_release_empty,
+	.device_caps	= V4L2_CAP_VIDEO_M2M | V4L2_CAP_STREAMING,
+};
+
+static const struct v4l2_m2m_ops rpivid_m2m_ops = {
+	.device_run	= rpivid_device_run,
+};
+
+static const struct media_device_ops rpivid_m2m_media_ops = {
+	.req_validate	= rpivid_request_validate,
+	.req_queue	= v4l2_m2m_request_queue,
+};
+
+static int rpivid_probe(struct platform_device *pdev)
+{
+	struct rpivid_dev *dev;
+	struct video_device *vfd;
+	int ret;
+
+	dev = devm_kzalloc(&pdev->dev, sizeof(*dev), GFP_KERNEL);
+	if (!dev)
+		return -ENOMEM;
+
+	dev->vfd = rpivid_video_device;
+	dev->dev = &pdev->dev;
+	dev->pdev = pdev;
+
+	ret = 0;
+	ret = rpivid_hw_probe(dev);
+	if (ret) {
+		dev_err(&pdev->dev, "Failed to probe hardware\n");
+		return ret;
+	}
+
+	dev->dec_ops = &rpivid_dec_ops_h265;
+
+	mutex_init(&dev->dev_mutex);
+
+	ret = v4l2_device_register(&pdev->dev, &dev->v4l2_dev);
+	if (ret) {
+		dev_err(&pdev->dev, "Failed to register V4L2 device\n");
+		return ret;
+	}
+
+	vfd = &dev->vfd;
+	vfd->lock = &dev->dev_mutex;
+	vfd->v4l2_dev = &dev->v4l2_dev;
+
+	snprintf(vfd->name, sizeof(vfd->name), "%s", rpivid_video_device.name);
+	video_set_drvdata(vfd, dev);
+
+	dev->m2m_dev = v4l2_m2m_init(&rpivid_m2m_ops);
+	if (IS_ERR(dev->m2m_dev)) {
+		v4l2_err(&dev->v4l2_dev,
+			 "Failed to initialize V4L2 M2M device\n");
+		ret = PTR_ERR(dev->m2m_dev);
+
+		goto err_v4l2;
+	}
+
+	dev->mdev.dev = &pdev->dev;
+	strscpy(dev->mdev.model, RPIVID_NAME, sizeof(dev->mdev.model));
+	strscpy(dev->mdev.bus_info, "platform:" RPIVID_NAME,
+		sizeof(dev->mdev.bus_info));
+
+	media_device_init(&dev->mdev);
+	dev->mdev.ops = &rpivid_m2m_media_ops;
+	dev->v4l2_dev.mdev = &dev->mdev;
+
+	ret = video_register_device(vfd, VFL_TYPE_VIDEO, video_nr);
+	if (ret) {
+		v4l2_err(&dev->v4l2_dev, "Failed to register video device\n");
+		goto err_m2m;
+	}
+
+	v4l2_info(&dev->v4l2_dev,
+		  "Device registered as /dev/video%d\n", vfd->num);
+
+	ret = v4l2_m2m_register_media_controller(dev->m2m_dev, vfd,
+						 MEDIA_ENT_F_PROC_VIDEO_DECODER);
+	if (ret) {
+		v4l2_err(&dev->v4l2_dev,
+			 "Failed to initialize V4L2 M2M media controller\n");
+		goto err_video;
+	}
+
+	ret = media_device_register(&dev->mdev);
+	if (ret) {
+		v4l2_err(&dev->v4l2_dev, "Failed to register media device\n");
+		goto err_m2m_mc;
+	}
+
+	platform_set_drvdata(pdev, dev);
+
+	return 0;
+
+err_m2m_mc:
+	v4l2_m2m_unregister_media_controller(dev->m2m_dev);
+err_video:
+	video_unregister_device(&dev->vfd);
+err_m2m:
+	v4l2_m2m_release(dev->m2m_dev);
+err_v4l2:
+	v4l2_device_unregister(&dev->v4l2_dev);
+
+	return ret;
+}
+
+static int rpivid_remove(struct platform_device *pdev)
+{
+	struct rpivid_dev *dev = platform_get_drvdata(pdev);
+
+	if (media_devnode_is_registered(dev->mdev.devnode)) {
+		media_device_unregister(&dev->mdev);
+		v4l2_m2m_unregister_media_controller(dev->m2m_dev);
+		media_device_cleanup(&dev->mdev);
+	}
+
+	v4l2_m2m_release(dev->m2m_dev);
+	video_unregister_device(&dev->vfd);
+	v4l2_device_unregister(&dev->v4l2_dev);
+
+	rpivid_hw_remove(dev);
+
+	return 0;
+}
+
+static const struct of_device_id rpivid_dt_match[] = {
+	{
+		.compatible = "raspberrypi,rpivid-vid-decoder",
+	},
+	{ /* sentinel */ }
+};
+MODULE_DEVICE_TABLE(of, rpivid_dt_match);
+
+static struct platform_driver rpivid_driver = {
+	.probe		= rpivid_probe,
+	.remove		= rpivid_remove,
+	.driver		= {
+		.name = RPIVID_NAME,
+		.of_match_table	= of_match_ptr(rpivid_dt_match),
+	},
+};
+module_platform_driver(rpivid_driver);
+
+MODULE_LICENSE("GPL v2");
+MODULE_AUTHOR("John Cox <jc@kynesim.co.uk>");
+MODULE_DESCRIPTION("Raspberry Pi HEVC V4L2 driver");

drivers/staging/media/rpivid/rpivid.h

+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * Raspberry Pi HEVC driver
+ *
+ * Copyright (C) 2020 Raspberry Pi (Trading) Ltd
+ *
+ * Based on the Cedrus VPU driver, that is:
+ *
+ * Copyright (C) 2016 Florent Revest <florent.revest@free-electrons.com>
+ * Copyright (C) 2018 Paul Kocialkowski <paul.kocialkowski@bootlin.com>
+ * Copyright (C) 2018 Bootlin
+ */
+
+#ifndef _RPIVID_H_
+#define _RPIVID_H_
+
+#include <linux/clk.h>
+#include <linux/platform_device.h>
+#include <media/v4l2-ctrls.h>
+#include <media/v4l2-device.h>
+#include <media/v4l2-mem2mem.h>
+#include <media/videobuf2-v4l2.h>
+#include <media/videobuf2-dma-contig.h>
+
+#define OPT_DEBUG_POLL_IRQ  0
+
+#define RPIVID_NAME			"rpivid"
+
+#define RPIVID_CAPABILITY_UNTILED	BIT(0)
+#define RPIVID_CAPABILITY_H265_DEC	BIT(1)
+
+#define RPIVID_QUIRK_NO_DMA_OFFSET	BIT(0)
+
+#define RPIVID_SRC_PIXELFORMAT_DEFAULT	V4L2_PIX_FMT_HEVC_SLICE
+
+enum rpivid_irq_status {
+	RPIVID_IRQ_NONE,
+	RPIVID_IRQ_ERROR,
+	RPIVID_IRQ_OK,
+};
+
+struct rpivid_control {
+	struct v4l2_ctrl_config cfg;
+	unsigned char		required:1;
+};
+
+struct rpivid_h265_run {
+	const struct v4l2_ctrl_hevc_sps			*sps;
+	const struct v4l2_ctrl_hevc_pps			*pps;
+	const struct v4l2_ctrl_hevc_slice_params	*slice_params;
+	const struct v4l2_ctrl_hevc_scaling_matrix	*scaling_matrix;
+};
+
+struct rpivid_run {
+	struct vb2_v4l2_buffer	*src;
+	struct vb2_v4l2_buffer	*dst;
+
+	struct rpivid_h265_run	h265;
+};
+
+struct rpivid_buffer {
+	struct v4l2_m2m_buffer          m2m_buf;
+};
+
+struct rpivid_dec_state;
+struct rpivid_dec_env;
+#define RPIVID_DEC_ENV_COUNT 3
+
+struct rpivid_gptr {
+	size_t size;
+	__u8 *ptr;
+	dma_addr_t addr;
+	unsigned long attrs;
+};
+
+struct rpivid_dev;
+typedef void (*rpivid_irq_callback)(struct rpivid_dev *dev, void *ctx);
+
+struct rpivid_q_aux;
+#define RPIVID_AUX_ENT_COUNT VB2_MAX_FRAME
+
+#define RPIVID_P2BUF_COUNT 2
+
+struct rpivid_ctx {
+	struct v4l2_fh			fh;
+	struct rpivid_dev		*dev;
+
+	struct v4l2_pix_format		src_fmt;
+	struct v4l2_pix_format		dst_fmt;
+	int dst_fmt_set;
+
+	struct v4l2_ctrl_handler	hdl;
+	struct v4l2_ctrl		**ctrls;
+
+	/* Decode state - stateless decoder my *** */
+	/* state contains stuff that is only needed in phase0
+	 * it could be held in dec_env but that would be wasteful
+	 */
+	struct rpivid_dec_state *state;
+	struct rpivid_dec_env *dec0;
+
+	/* Spinlock protecting dec_free */
+	spinlock_t dec_lock;
+	struct rpivid_dec_env *dec_free;
+
+	struct rpivid_dec_env *dec_pool;
+
+	/* Some of these should be in dev */
+	struct rpivid_gptr bitbufs[1];  /* Will be 2 */
+	struct rpivid_gptr cmdbufs[1];  /* Will be 2 */
+	unsigned int p2idx;
+	atomic_t p2out;
+	struct rpivid_gptr pu_bufs[RPIVID_P2BUF_COUNT];
+	struct rpivid_gptr coeff_bufs[RPIVID_P2BUF_COUNT];
+
+	/* Spinlock protecting aux_free */
+	spinlock_t aux_lock;
+	struct rpivid_q_aux *aux_free;
+
+	struct rpivid_q_aux *aux_ents[RPIVID_AUX_ENT_COUNT];
+
+	unsigned int colmv_stride;
+	unsigned int colmv_picsize;
+};
+
+struct rpivid_dec_ops {
+	void (*setup)(struct rpivid_ctx *ctx, struct rpivid_run *run);
+	int (*start)(struct rpivid_ctx *ctx);
+	void (*stop)(struct rpivid_ctx *ctx);
+	void (*trigger)(struct rpivid_ctx *ctx);
+};
+
+struct rpivid_variant {
+	unsigned int	capabilities;
+	unsigned int	quirks;
+	unsigned int	mod_rate;
+};
+
+struct rpivid_hw_irq_ent;
+
+struct rpivid_hw_irq_ctrl {
+	/* Spinlock protecting claim and tail */
+	spinlock_t lock;
+	struct rpivid_hw_irq_ent *claim;
+	struct rpivid_hw_irq_ent *tail;
+
+	/* Ent for pending irq - also prevents sched */
+	struct rpivid_hw_irq_ent *irq;
+	/* Non-zero => do not start a new job - outer layer sched pending */
+	int no_sched;
+	/* Thread CB requested */
+	bool thread_reqed;
+};
+
+struct rpivid_dev {
+	struct v4l2_device	v4l2_dev;
+	struct video_device	vfd;
+	struct media_device	mdev;
+	struct media_pad	pad[2];
+	struct platform_device	*pdev;
+	struct device		*dev;
+	struct v4l2_m2m_dev	*m2m_dev;
+	struct rpivid_dec_ops	*dec_ops;
+
+	/* Device file mutex */
+	struct mutex		dev_mutex;
+
+	void __iomem		*base_irq;
+	void __iomem		*base_h265;
+
+	struct clk		*clock;
+
+	struct rpivid_hw_irq_ctrl ic_active1;
+	struct rpivid_hw_irq_ctrl ic_active2;
+};
+
+extern struct rpivid_dec_ops rpivid_dec_ops_h265;
+
+void *rpivid_find_control_data(struct rpivid_ctx *ctx, u32 id);
+
+#endif

drivers/staging/media/rpivid/rpivid_dec.c

+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Raspberry Pi HEVC driver
+ *
+ * Copyright (C) 2020 Raspberry Pi (Trading) Ltd
+ *
+ * Based on the Cedrus VPU driver, that is:
+ *
+ * Copyright (C) 2016 Florent Revest <florent.revest@free-electrons.com>
+ * Copyright (C) 2018 Paul Kocialkowski <paul.kocialkowski@bootlin.com>
+ * Copyright (C) 2018 Bootlin
+ */
+
+#include <media/v4l2-device.h>
+#include <media/v4l2-ioctl.h>
+#include <media/v4l2-event.h>
+#include <media/v4l2-mem2mem.h>
+
+#include "rpivid.h"
+#include "rpivid_dec.h"
+
+void rpivid_device_run(void *priv)
+{
+	struct rpivid_ctx *ctx = priv;
+	struct rpivid_dev *dev = ctx->dev;
+	struct rpivid_run run = {};
+	struct media_request *src_req;
+
+	run.src = v4l2_m2m_next_src_buf(ctx->fh.m2m_ctx);
+	run.dst = v4l2_m2m_next_dst_buf(ctx->fh.m2m_ctx);
+
+	if (!run.src || !run.dst) {
+		v4l2_err(&dev->v4l2_dev, "%s: Missing buffer: src=%p, dst=%p\n",
+			 __func__, run.src, run.dst);
+		/* We are stuffed - this probably won't dig us out of our
+		 * current situation but it is better than nothing
+		 */
+		v4l2_m2m_buf_done_and_job_finish(dev->m2m_dev, ctx->fh.m2m_ctx,
+						 VB2_BUF_STATE_ERROR);
+		return;
+	}
+
+	/* Apply request(s) controls if needed. */
+	src_req = run.src->vb2_buf.req_obj.req;
+
+	if (src_req)
+		v4l2_ctrl_request_setup(src_req, &ctx->hdl);
+
+	switch (ctx->src_fmt.pixelformat) {
+	case V4L2_PIX_FMT_HEVC_SLICE:
+		run.h265.sps =
+			rpivid_find_control_data(ctx,
+						 V4L2_CID_MPEG_VIDEO_HEVC_SPS);
+		run.h265.pps =
+			rpivid_find_control_data(ctx,
+						 V4L2_CID_MPEG_VIDEO_HEVC_PPS);
+		run.h265.slice_params =
+			rpivid_find_control_data(ctx,
+						 V4L2_CID_MPEG_VIDEO_HEVC_SLICE_PARAMS);
+		run.h265.scaling_matrix =
+			rpivid_find_control_data(ctx,
+						 V4L2_CID_MPEG_VIDEO_HEVC_SCALING_MATRIX);
+		break;
+
+	default:
+		break;
+	}
+
+	v4l2_m2m_buf_copy_metadata(run.src, run.dst, true);
+
+	dev->dec_ops->setup(ctx, &run);
+
+	/* Complete request(s) controls if needed. */
+
+	if (src_req)
+		v4l2_ctrl_request_complete(src_req, &ctx->hdl);
+
+	dev->dec_ops->trigger(ctx);
+}

drivers/staging/media/rpivid/rpivid_dec.h

+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * Raspberry Pi HEVC driver
+ *
+ * Copyright (C) 2020 Raspberry Pi (Trading) Ltd
+ *
+ * Based on the Cedrus VPU driver, that is:
+ *
+ * Copyright (C) 2016 Florent Revest <florent.revest@free-electrons.com>
+ * Copyright (C) 2018 Paul Kocialkowski <paul.kocialkowski@bootlin.com>
+ * Copyright (C) 2018 Bootlin
+ */
+
+#ifndef _RPIVID_DEC_H_
+#define _RPIVID_DEC_H_
+
+void rpivid_device_run(void *priv);
+
+#endif

drivers/staging/media/rpivid/rpivid_h265.c

+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * Raspberry Pi HEVC driver
+ *
+ * Copyright (C) 2020 Raspberry Pi (Trading) Ltd
+ *
+ * Based on the Cedrus VPU driver, that is:
+ *
+ * Copyright (C) 2016 Florent Revest <florent.revest@free-electrons.com>
+ * Copyright (C) 2018 Paul Kocialkowski <paul.kocialkowski@bootlin.com>
+ * Copyright (C) 2018 Bootlin
+ */
+
+#include <linux/delay.h>
+#include <linux/types.h>
+
+#include <media/videobuf2-dma-contig.h>
+
+#include "rpivid.h"
+#include "rpivid_hw.h"
+
+#define DEBUG_TRACE_P1_CMD 0
+#define DEBUG_TRACE_EXECUTION 0
+
+#if DEBUG_TRACE_EXECUTION
+#define xtrace_in(dev_, de_)\
+	v4l2_info(&(dev_)->v4l2_dev, "%s[%d]: in\n",   __func__,\
+		  (de_) == NULL ? -1 : (de_)->decode_order)
+#define xtrace_ok(dev_, de_)\
+	v4l2_info(&(dev_)->v4l2_dev, "%s[%d]: ok\n",   __func__,\
+		  (de_) == NULL ? -1 : (de_)->decode_order)
+#define xtrace_fin(dev_, de_)\
+	v4l2_info(&(dev_)->v4l2_dev, "%s[%d]: finish\n", __func__,\
+		  (de_) == NULL ? -1 : (de_)->decode_order)
+#define xtrace_fail(dev_, de_)\
+	v4l2_info(&(dev_)->v4l2_dev, "%s[%d]: FAIL\n", __func__,\
+		  (de_) == NULL ? -1 : (de_)->decode_order)
+#else
+#define xtrace_in(dev_, de_)
+#define xtrace_ok(dev_, de_)
+#define xtrace_fin(dev_, de_)
+#define xtrace_fail(dev_, de_)
+#endif
+
+enum hevc_slice_type {
+	HEVC_SLICE_B = 0,
+	HEVC_SLICE_P = 1,
+	HEVC_SLICE_I = 2,
+};
+
+enum hevc_layer { L0 = 0, L1 = 1 };
+
+static int gptr_alloc(struct rpivid_dev *const dev, struct rpivid_gptr *gptr,
+		      size_t size, unsigned long attrs)
+{
+	gptr->size = size;
+	gptr->attrs = attrs;
+	gptr->addr = 0;
+	gptr->ptr = dma_alloc_attrs(dev->dev, gptr->size, &gptr->addr,
+				    GFP_KERNEL, gptr->attrs);
+	return !gptr->ptr ? -ENOMEM : 0;
+}
+
+static void gptr_free(struct rpivid_dev *const dev,
+		      struct rpivid_gptr *const gptr)
+{
+	if (gptr->ptr)
+		dma_free_attrs(dev->dev, gptr->size, gptr->ptr, gptr->addr,
+			       gptr->attrs);
+	gptr->size = 0;
+	gptr->ptr = NULL;
+	gptr->addr = 0;
+	gptr->attrs = 0;
+}
+
+/* Realloc but do not copy */
+static int gptr_realloc_new(struct rpivid_dev * const dev,
+			    struct rpivid_gptr * const gptr, size_t size)
+{
+	if (size == gptr->size)
+		return 0;
+
+	if (gptr->ptr)
+		dma_free_attrs(dev->dev, gptr->size, gptr->ptr,
+			       gptr->addr, gptr->attrs);
+
+	gptr->addr = 0;
+	gptr->size = size;
+	gptr->ptr = dma_alloc_attrs(dev->dev, gptr->size,
+				    &gptr->addr, GFP_KERNEL, gptr->attrs);
+	return gptr->ptr ? 0 : -ENOMEM;
+}
+
+/* floor(log2(x)) */
+static unsigned int log2_size(size_t x)
+{
+	unsigned int n = 0;
+
+	if (x & ~0xffff) {
+		n += 16;
+		x >>= 16;
+	}
+	if (x & ~0xff) {
+		n += 8;
+		x >>= 8;
+	}
+	if (x & ~0xf) {
+		n += 4;
+		x >>= 4;
+	}
+	if (x & ~3) {
+		n += 2;
+		x >>= 2;
+	}
+	return (x & ~1) ? n + 1 : n;
+}
+
+static size_t round_up_size(const size_t x)
+{
+	/* Admit no size < 256 */
+	const unsigned int n = x < 256 ? 8 : log2_size(x) - 1;
+
+	return x >= (3 << n) ? 4 << n : (3 << n);
+}
+
+static size_t next_size(const size_t x)
+{
+	return round_up_size(x + 1);
+}
+
+#define NUM_SCALING_FACTORS 4064 /* Not a typo = 0xbe0 + 0x400 */
+
+#define AXI_BASE64 0
+
+#define PROB_BACKUP ((20 << 12) + (20 << 6) + (0 << 0))
+#define PROB_RELOAD ((20 << 12) + (20 << 0) + (0 << 6))
+
+#define HEVC_MAX_REFS V4L2_HEVC_DPB_ENTRIES_NUM_MAX
+
+//////////////////////////////////////////////////////////////////////////////
+
+struct rpi_cmd {
+	u32 addr;
+	u32 data;
+} __packed;
+
+struct rpivid_q_aux {
+	unsigned int refcount;
+	unsigned int q_index;
+	struct rpivid_q_aux *next;
+	struct rpivid_gptr col;
+};
+
+//////////////////////////////////////////////////////////////////////////////
+
+enum rpivid_decode_state {
+	RPIVID_DECODE_SLICE_START,
+	RPIVID_DECODE_SLICE_CONTINUE,
+	RPIVID_DECODE_ERROR_CONTINUE,
+	RPIVID_DECODE_ERROR_DONE,
+	RPIVID_DECODE_PHASE1,
+	RPIVID_DECODE_END,
+};
+
+struct rpivid_dec_env {
+	struct rpivid_ctx *ctx;
+	struct rpivid_dec_env *next;
+
+	enum rpivid_decode_state state;
+	unsigned int decode_order;
+	int p1_status;		/* P1 status - what to realloc */
+
+	struct rpivid_dec_env *phase_wait_q_next;
+
+	struct rpi_cmd *cmd_fifo;
+	unsigned int cmd_len, cmd_max;
+	unsigned int num_slice_msgs;
+	unsigned int pic_width_in_ctbs_y;
+	unsigned int pic_height_in_ctbs_y;
+	unsigned int dpbno_col;
+	u32 reg_slicestart;
+	int collocated_from_l0_flag;
+	unsigned int wpp_entry_x;
+	unsigned int wpp_entry_y;
+
+	u32 rpi_config2;
+	u32 rpi_framesize;
+	u32 rpi_currpoc;
+
+	struct vb2_v4l2_buffer *frame_buf; // Detached dest buffer
+	unsigned int frame_c_offset;
+	unsigned int frame_stride;
+	dma_addr_t frame_addr;
+	dma_addr_t ref_addrs[16];
+	struct rpivid_q_aux *frame_aux;
+	struct rpivid_q_aux *col_aux;
+
+	dma_addr_t pu_base_vc;
+	dma_addr_t coeff_base_vc;
+	u32 pu_stride;
+	u32 coeff_stride;
+
+	struct rpivid_gptr *bit_copy_gptr;
+	size_t bit_copy_len;
+	struct rpivid_gptr *cmd_copy_gptr;
+
+	u16 slice_msgs[2 * HEVC_MAX_REFS * 8 + 3];
+	u8 scaling_factors[NUM_SCALING_FACTORS];
+
+	struct rpivid_hw_irq_ent irq_ent;
+};
+
+#define member_size(type, member) sizeof(((type *)0)->member)
+
+struct rpivid_dec_state {
+	struct v4l2_ctrl_hevc_sps sps;
+	struct v4l2_ctrl_hevc_pps pps;
+
+	// Helper vars & tables derived from sps/pps
+	unsigned int log2_ctb_size; /* log2 width of a CTB */
+	unsigned int ctb_width; /* Width in CTBs */
+	unsigned int ctb_height; /* Height in CTBs */
+	unsigned int ctb_size; /* Pic area in CTBs */
+	unsigned int num_tile_columns;
+	unsigned int num_tile_rows;
+	u8 column_width[member_size(struct v4l2_ctrl_hevc_pps,
+				    column_width_minus1)];
+	u8 row_height[member_size(struct v4l2_ctrl_hevc_pps,
+				  row_height_minus1)];
+
+	int *col_bd;
+	int *row_bd;
+	int *ctb_addr_rs_to_ts;
+	int *ctb_addr_ts_to_rs;
+	int *tile_id;
+
+	// Aux starage for DPB
+	// Hold refs
+	struct rpivid_q_aux *ref_aux[HEVC_MAX_REFS];
+	struct rpivid_q_aux *frame_aux;
+
+	// Slice vars
+	unsigned int slice_idx;
+	bool frame_end;
+	bool slice_temporal_mvp;  /* Slice flag but constant for frame */
+
+	// Temp vars per run - don't actually need to persist
+	u8 *src_buf;
+	dma_addr_t src_addr;
+	const struct v4l2_ctrl_hevc_slice_params *sh;
+	unsigned int nb_refs[2];
+	unsigned int slice_qp;
+	unsigned int max_num_merge_cand; // 0 if I-slice
+	bool dependent_slice_segment_flag;
+};
+
+static inline int clip_int(const int x, const int lo, const int hi)
+{
+	return x < lo ? lo : x > hi ? hi : x;
+}
+
+//////////////////////////////////////////////////////////////////////////////
+// Phase 1 command and bit FIFOs
+
+#if DEBUG_TRACE_P1_CMD
+static int p1_z;
+#endif
+
+// ???? u16 addr - put in u32
+static int p1_apb_write(struct rpivid_dec_env *const de, const u16 addr,
+			const u32 data)
+{
+	if (de->cmd_len == de->cmd_max)
+		de->cmd_fifo =
+			krealloc(de->cmd_fifo,
+				 (de->cmd_max *= 2) * sizeof(struct rpi_cmd),
+				 GFP_KERNEL);
+	de->cmd_fifo[de->cmd_len].addr = addr;
+	de->cmd_fifo[de->cmd_len].data = data;
+
+#if DEBUG_TRACE_P1_CMD
+	if (++p1_z < 256) {
+		v4l2_info(&de->ctx->dev->v4l2_dev, "[%02x] %x %x\n",
+			  de->cmd_len, addr, data);
+	}
+#endif
+
+	return de->cmd_len++;
+}
+
+static int ctb_to_tile(unsigned int ctb, unsigned int *bd, int num)
+{
+	int i;
+
+	for (i = 1; ctb >= bd[i]; i++)
+		; // bd[] has num+1 elements; bd[0]=0;
+	return i - 1;
+}
+
+static int ctb_to_slice_w_h(unsigned int ctb, int ctb_size, int width,
+			    unsigned int *bd, int num)
+{
+	if (ctb < bd[num - 1])
+		return ctb_size;
+	else if (width % ctb_size)
+		return width % ctb_size;
+	else
+		return ctb_size;
+}
+
+static void aux_q_free(struct rpivid_ctx *const ctx,
+		       struct rpivid_q_aux *const aq)
+{
+	struct rpivid_dev *const dev = ctx->dev;
+
+	gptr_free(dev, &aq->col);
+	kfree(aq);
+}
+
+static struct rpivid_q_aux *aux_q_alloc(struct rpivid_ctx *const ctx)
+{
+	struct rpivid_dev *const dev = ctx->dev;
+	struct rpivid_q_aux *const aq = kzalloc(sizeof(*aq), GFP_KERNEL);
+
+	if (!aq)
+		return NULL;
+
+	aq->refcount = 1;
+	if (gptr_alloc(dev, &aq->col, ctx->colmv_picsize,
+		       DMA_ATTR_FORCE_CONTIGUOUS | DMA_ATTR_NO_KERNEL_MAPPING))
+		goto fail;
+
+	return aq;
+
+fail:
+	kfree(aq);
+	return NULL;
+}
+
+static struct rpivid_q_aux *aux_q_new(struct rpivid_ctx *const ctx,
+				      const unsigned int q_index)
+{
+	struct rpivid_q_aux *aq;
+	unsigned long lockflags;
+
+	spin_lock_irqsave(&ctx->aux_lock, lockflags);
+	aq = ctx->aux_free;
+	if (aq) {
+		ctx->aux_free = aq->next;
+		aq->next = NULL;
+		aq->refcount = 1;
+	}
+	spin_unlock_irqrestore(&ctx->aux_lock, lockflags);
+
+	if (!aq) {
+		aq = aux_q_alloc(ctx);
+		if (!aq)
+			return NULL;
+	}
+
+	aq->q_index = q_index;
+	ctx->aux_ents[q_index] = aq;
+	return aq;
+}
+
+static struct rpivid_q_aux *aux_q_ref(struct rpivid_ctx *const ctx,
+				      struct rpivid_q_aux *const aq)
+{
+	if (aq) {
+		unsigned long lockflags;
+
+		spin_lock_irqsave(&ctx->aux_lock, lockflags);
+
+		++aq->refcount;
+
+		spin_unlock_irqrestore(&ctx->aux_lock, lockflags);
+	}
+	return aq;
+}
+
+static void aux_q_release(struct rpivid_ctx *const ctx,
+			  struct rpivid_q_aux **const paq)
+{
+	struct rpivid_q_aux *const aq = *paq;
+	*paq = NULL;
+
+	if (aq) {
+		unsigned long lockflags;
+
+		spin_lock_irqsave(&ctx->aux_lock, lockflags);
+
+		if (--aq->refcount == 0) {
+			aq->next = ctx->aux_free;
+			ctx->aux_free = aq;
+			ctx->aux_ents[aq->q_index] = NULL;
+		}
+
+		spin_unlock_irqrestore(&ctx->aux_lock, lockflags);
+	}
+}
+
+static void aux_q_init(struct rpivid_ctx *const ctx)
+{
+	spin_lock_init(&ctx->aux_lock);
+	ctx->aux_free = NULL;
+}
+
+static void aux_q_uninit(struct rpivid_ctx *const ctx)
+{
+	struct rpivid_q_aux *aq;
+
+	ctx->colmv_picsize = 0;
+	ctx->colmv_stride = 0;
+	while ((aq = ctx->aux_free) != NULL) {
+		ctx->aux_free = aq->next;
+		aux_q_free(ctx, aq);
+	}
+}
+
+//////////////////////////////////////////////////////////////////////////////
+
+/*
+ * Initialisation process for context variables (CABAC init)
+ * see H.265 9.3.2.2
+ *
+ * N.B. If comparing with FFmpeg note that this h/w uses slightly different
+ * offsets to FFmpegs array
+ */
+
+/* Actual number of values */
+#define RPI_PROB_VALS 154U
+/* Rounded up as we copy words */
+#define RPI_PROB_ARRAY_SIZE ((154 + 3) & ~3)
+
+/* Initialiser values - see tables H.265 9-4 through 9-42 */
+static const u8 prob_init[3][156] = {
+	{
+		153, 200, 139, 141, 157, 154, 154, 154, 154, 154, 184, 154, 154,
+		154, 184, 63,  154, 154, 154, 154, 154, 154, 154, 154, 154, 154,
+		154, 154, 154, 153, 138, 138, 111, 141, 94,  138, 182, 154, 154,
+		154, 140, 92,  137, 138, 140, 152, 138, 139, 153, 74,  149, 92,
+		139, 107, 122, 152, 140, 179, 166, 182, 140, 227, 122, 197, 110,
+		110, 124, 125, 140, 153, 125, 127, 140, 109, 111, 143, 127, 111,
+		79,  108, 123, 63,  110, 110, 124, 125, 140, 153, 125, 127, 140,
+		109, 111, 143, 127, 111, 79,  108, 123, 63,  91,  171, 134, 141,
+		138, 153, 136, 167, 152, 152, 139, 139, 111, 111, 125, 110, 110,
+		94,  124, 108, 124, 107, 125, 141, 179, 153, 125, 107, 125, 141,
+		179, 153, 125, 107, 125, 141, 179, 153, 125, 140, 139, 182, 182,
+		152, 136, 152, 136, 153, 136, 139, 111, 136, 139, 111, 0,   0,
+	},
+	{
+		153, 185, 107, 139, 126, 197, 185, 201, 154, 149, 154, 139, 154,
+		154, 154, 152, 110, 122, 95,  79,  63,  31,  31,  153, 153, 168,
+		140, 198, 79,  124, 138, 94,  153, 111, 149, 107, 167, 154, 154,
+		154, 154, 196, 196, 167, 154, 152, 167, 182, 182, 134, 149, 136,
+		153, 121, 136, 137, 169, 194, 166, 167, 154, 167, 137, 182, 125,
+		110, 94,  110, 95,  79,  125, 111, 110, 78,  110, 111, 111, 95,
+		94,  108, 123, 108, 125, 110, 94,  110, 95,  79,  125, 111, 110,
+		78,  110, 111, 111, 95,  94,  108, 123, 108, 121, 140, 61,  154,
+		107, 167, 91,  122, 107, 167, 139, 139, 155, 154, 139, 153, 139,
+		123, 123, 63,  153, 166, 183, 140, 136, 153, 154, 166, 183, 140,
+		136, 153, 154, 166, 183, 140, 136, 153, 154, 170, 153, 123, 123,
+		107, 121, 107, 121, 167, 151, 183, 140, 151, 183, 140, 0,   0,
+	},
+	{
+		153, 160, 107, 139, 126, 197, 185, 201, 154, 134, 154, 139, 154,
+		154, 183, 152, 154, 137, 95,  79,  63,  31,  31,  153, 153, 168,
+		169, 198, 79,  224, 167, 122, 153, 111, 149, 92,  167, 154, 154,
+		154, 154, 196, 167, 167, 154, 152, 167, 182, 182, 134, 149, 136,
+		153, 121, 136, 122, 169, 208, 166, 167, 154, 152, 167, 182, 125,
+		110, 124, 110, 95,  94,  125, 111, 111, 79,  125, 126, 111, 111,
+		79,  108, 123, 93,  125, 110, 124, 110, 95,  94,  125, 111, 111,
+		79,  125, 126, 111, 111, 79,  108, 123, 93,  121, 140, 61,  154,
+		107, 167, 91,  107, 107, 167, 139, 139, 170, 154, 139, 153, 139,
+		123, 123, 63,  124, 166, 183, 140, 136, 153, 154, 166, 183, 140,
+		136, 153, 154, 166, 183, 140, 136, 153, 154, 170, 153, 138, 138,
+		122, 121, 122, 121, 167, 151, 183, 140, 151, 183, 140, 0,   0,
+	},
+};
+
+static void write_prob(struct rpivid_dec_env *const de,
+		       const struct rpivid_dec_state *const s)
+{
+	u8 dst[RPI_PROB_ARRAY_SIZE];
+
+	const unsigned int init_type =
+		((s->sh->flags & V4L2_HEVC_SLICE_PARAMS_FLAG_CABAC_INIT) != 0 &&
+		 s->sh->slice_type != HEVC_SLICE_I) ?
+			s->sh->slice_type + 1 :
+			2 - s->sh->slice_type;
+	const u8 *p = prob_init[init_type];
+	const int q = clip_int(s->slice_qp, 0, 51);
+	unsigned int i;
+
+	for (i = 0; i < RPI_PROB_VALS; i++) {
+		int init_value = p[i];
+		int m = (init_value >> 4) * 5 - 45;
+		int n = ((init_value & 15) << 3) - 16;
+		int pre = 2 * (((m * q) >> 4) + n) - 127;
+
+		pre ^= pre >> 31;
+		if (pre > 124)
+			pre = 124 + (pre & 1);
+		dst[i] = pre;
+	}
+	for (i = RPI_PROB_VALS; i != RPI_PROB_ARRAY_SIZE; ++i)
+		dst[i] = 0;
+
+	for (i = 0; i < RPI_PROB_ARRAY_SIZE; i += 4)
+		p1_apb_write(de, 0x1000 + i,
+			     dst[i] + (dst[i + 1] << 8) + (dst[i + 2] << 16) +
+				     (dst[i + 3] << 24));
+}
+
+static void write_scaling_factors(struct rpivid_dec_env *const de)
+{
+	int i;
+	const u8 *p = (u8 *)de->scaling_factors;
+
+	for (i = 0; i < NUM_SCALING_FACTORS; i += 4, p += 4)
+		p1_apb_write(de, 0x2000 + i,
+			     p[0] + (p[1] << 8) + (p[2] << 16) + (p[3] << 24));
+}
+
+static inline __u32 dma_to_axi_addr(dma_addr_t a)
+{
+	return (__u32)(a >> 6);
+}
+
+static void write_bitstream(struct rpivid_dec_env *const de,
+			    const struct rpivid_dec_state *const s)
+{
+	// Note that FFmpeg removes emulation prevention bytes, so this is
+	// matched in the configuration here.
+	// Whether that is the correct behaviour or not is not clear in the
+	// spec.
+	const int rpi_use_emu = 1;
+	unsigned int offset = s->sh->data_bit_offset / 8 + 1;
+	const unsigned int len = (s->sh->bit_size + 7) / 8 - offset;
+	dma_addr_t addr;
+
+	if (s->src_addr != 0) {
+		addr = s->src_addr + offset;
+	} else {
+		memcpy(de->bit_copy_gptr->ptr + de->bit_copy_len,
+		       s->src_buf + offset, len);
+		addr = de->bit_copy_gptr->addr + de->bit_copy_len;
+		de->bit_copy_len += (len + 63) & ~63;
+	}
+	offset = addr & 63;
+
+	p1_apb_write(de, RPI_BFBASE, dma_to_axi_addr(addr));
+	p1_apb_write(de, RPI_BFNUM, len);
+	p1_apb_write(de, RPI_BFCONTROL, offset + (1 << 7)); // Stop
+	p1_apb_write(de, RPI_BFCONTROL, offset + (rpi_use_emu << 6));
+}
+
+//////////////////////////////////////////////////////////////////////////////
+
+static void write_slice(struct rpivid_dec_env *const de,
+			const struct rpivid_dec_state *const s,
+			const unsigned int slice_w,
+			const unsigned int slice_h)
+{
+	u32 u32 = (s->sh->slice_type << 12) +
+		  (((s->sh->flags &
+		     V4L2_HEVC_SLICE_PARAMS_FLAG_SLICE_SAO_LUMA) != 0)
+		   << 14) +
+		  (((s->sh->flags &
+		     V4L2_HEVC_SLICE_PARAMS_FLAG_SLICE_SAO_CHROMA) != 0)
+		   << 15) +
+		  (slice_w << 17) + (slice_h << 24);
+
+	u32 |= (s->max_num_merge_cand << 0) + (s->nb_refs[L0] << 4) +
+	       (s->nb_refs[L1] << 8);
+
+	if (s->sh->slice_type == HEVC_SLICE_B)
+		u32 |= ((s->sh->flags &
+			 V4L2_HEVC_SLICE_PARAMS_FLAG_MVD_L1_ZERO) != 0)
+		       << 16;
+	p1_apb_write(de, RPI_SLICE, u32);
+}
+
+//////////////////////////////////////////////////////////////////////////////
+// Tiles mode
+
+static void new_entry_point(struct rpivid_dec_env *const de,
+			    const struct rpivid_dec_state *const s,
+			    const int do_bte,
+			    const int reset_qp_y, const int ctb_addr_ts)
+{
+	int ctb_col = s->ctb_addr_ts_to_rs[ctb_addr_ts] %
+							de->pic_width_in_ctbs_y;
+	int ctb_row = s->ctb_addr_ts_to_rs[ctb_addr_ts] /
+							de->pic_width_in_ctbs_y;
+
+	int tile_x = ctb_to_tile(ctb_col, s->col_bd, s->num_tile_columns);
+	int tile_y = ctb_to_tile(ctb_row, s->row_bd, s->num_tile_rows);
+
+	int endx = s->col_bd[tile_x + 1] - 1;
+	int endy = s->row_bd[tile_y + 1] - 1;
+
+	u8 slice_w = ctb_to_slice_w_h(ctb_col, 1 << s->log2_ctb_size,
+				      s->sps.pic_width_in_luma_samples,
+				      s->col_bd, s->num_tile_columns);
+	u8 slice_h = ctb_to_slice_w_h(ctb_row, 1 << s->log2_ctb_size,
+				      s->sps.pic_height_in_luma_samples,
+				      s->row_bd, s->num_tile_rows);
+
+	p1_apb_write(de, RPI_TILESTART,
+		     s->col_bd[tile_x] + (s->row_bd[tile_y] << 16));
+	p1_apb_write(de, RPI_TILEEND, endx + (endy << 16));
+
+	if (do_bte)
+		p1_apb_write(de, RPI_BEGINTILEEND, endx + (endy << 16));
+
+	write_slice(de, s, slice_w, slice_h);
+
+	if (reset_qp_y) {
+		unsigned int sps_qp_bd_offset =
+			6 * s->sps.bit_depth_luma_minus8;
+
+		p1_apb_write(de, RPI_QP, sps_qp_bd_offset + s->slice_qp);
+	}
+
+	p1_apb_write(de, RPI_MODE,
+		     (0xFFFF << 0) + (0x0 << 16) +
+			     ((tile_x == s->num_tile_columns - 1) << 17) +
+			     ((tile_y == s->num_tile_rows - 1) << 18));
+
+	p1_apb_write(de, RPI_CONTROL, (ctb_col << 0) + (ctb_row << 16));
+}
+
+//////////////////////////////////////////////////////////////////////////////
+
+static void new_slice_segment(struct rpivid_dec_env *const de,
+			      const struct rpivid_dec_state *const s)
+{
+	const struct v4l2_ctrl_hevc_sps *const sps = &s->sps;
+	const struct v4l2_ctrl_hevc_pps *const pps = &s->pps;
+
+	p1_apb_write(de,
+		     RPI_SPS0,
+		     ((sps->log2_min_luma_coding_block_size_minus3 + 3) << 0) |
+		     (s->log2_ctb_size << 4) |
+		     ((sps->log2_min_luma_transform_block_size_minus2 + 2)
+							<< 8) |
+		     ((sps->log2_min_luma_transform_block_size_minus2 + 2 +
+		       sps->log2_diff_max_min_luma_transform_block_size)
+						<< 12) |
+		     ((sps->bit_depth_luma_minus8 + 8) << 16) |
+		     ((sps->bit_depth_chroma_minus8 + 8) << 20) |
+		     (sps->max_transform_hierarchy_depth_intra << 24) |
+		     (sps->max_transform_hierarchy_depth_inter << 28));
+
+	p1_apb_write(de,
+		     RPI_SPS1,
+		     ((sps->pcm_sample_bit_depth_luma_minus1 + 1) << 0) |
+		     ((sps->pcm_sample_bit_depth_chroma_minus1 + 1) << 4) |
+		     ((sps->log2_min_pcm_luma_coding_block_size_minus3 + 3)
+						<< 8) |
+		     ((sps->log2_min_pcm_luma_coding_block_size_minus3 + 3 +
+		       sps->log2_diff_max_min_pcm_luma_coding_block_size)
+						<< 12) |
+		     (((sps->flags & V4L2_HEVC_SPS_FLAG_SEPARATE_COLOUR_PLANE) ?
+				0 : sps->chroma_format_idc) << 16) |
+		     ((!!(sps->flags & V4L2_HEVC_SPS_FLAG_AMP_ENABLED)) << 18) |
+		     ((!!(sps->flags & V4L2_HEVC_SPS_FLAG_PCM_ENABLED)) << 19) |
+		     ((!!(sps->flags & V4L2_HEVC_SPS_FLAG_SCALING_LIST_ENABLED))
+						<< 20) |
+		     ((!!(sps->flags &
+			   V4L2_HEVC_SPS_FLAG_STRONG_INTRA_SMOOTHING_ENABLED))
+						<< 21));
+
+	p1_apb_write(de,
+		     RPI_PPS,
+		     ((s->log2_ctb_size - pps->diff_cu_qp_delta_depth) << 0) |
+		     ((!!(pps->flags & V4L2_HEVC_PPS_FLAG_CU_QP_DELTA_ENABLED))
+						 << 4) |
+		     ((!!(pps->flags &
+				V4L2_HEVC_PPS_FLAG_TRANSQUANT_BYPASS_ENABLED))
+						 << 5) |
+		     ((!!(pps->flags & V4L2_HEVC_PPS_FLAG_TRANSFORM_SKIP_ENABLED))
+						 << 6) |
+		     ((!!(pps->flags &
+				V4L2_HEVC_PPS_FLAG_SIGN_DATA_HIDING_ENABLED))
+						<< 7) |
+		     (((pps->pps_cb_qp_offset + s->sh->slice_cb_qp_offset) & 255)
+						<< 8) |
+		     (((pps->pps_cr_qp_offset + s->sh->slice_cr_qp_offset) & 255)
+						<< 16) |
+		     ((!!(pps->flags &
+				V4L2_HEVC_PPS_FLAG_CONSTRAINED_INTRA_PRED))
+						<< 24));
+
+	if ((sps->flags & V4L2_HEVC_SPS_FLAG_SCALING_LIST_ENABLED) != 0)
+		write_scaling_factors(de);
+
+	if (!s->dependent_slice_segment_flag) {
+		int ctb_col = s->sh->slice_segment_addr %
+							de->pic_width_in_ctbs_y;
+		int ctb_row = s->sh->slice_segment_addr /
+							de->pic_width_in_ctbs_y;
+
+		de->reg_slicestart = (ctb_col << 0) + (ctb_row << 16);
+	}
+
+	p1_apb_write(de, RPI_SLICESTART, de->reg_slicestart);
+}
+
+//////////////////////////////////////////////////////////////////////////////
+// Slice messages
+
+static void msg_slice(struct rpivid_dec_env *const de, const u16 msg)
+{
+	de->slice_msgs[de->num_slice_msgs++] = msg;
+}
+
+static void program_slicecmds(struct rpivid_dec_env *const de,
+			      const int sliceid)
+{
+	int i;
+
+	p1_apb_write(de, RPI_SLICECMDS, de->num_slice_msgs + (sliceid << 8));
+
+	for (i = 0; i < de->num_slice_msgs; i++)
+		p1_apb_write(de, 0x4000 + 4 * i, de->slice_msgs[i] & 0xffff);
+}
+
+// NoBackwardPredictionFlag 8.3.5
+// Simply checks POCs
+static int has_backward(const struct v4l2_hevc_dpb_entry *const dpb,
+			const __u8 *const idx, const unsigned int n,
+			const unsigned int cur_poc)
+{
+	unsigned int i;
+
+	for (i = 0; i < n; ++i) {
+		// Compare mod 2^16
+		// We only get u16 pocs & 8.3.1 says
+		// "The bitstream shall not contain data that result in values
+		//  of DiffPicOrderCnt( picA, picB ) used in the decoding
+		//  process that are not in the range of −2^15 to 2^15 − 1,
+		//  inclusive."
+		if (((cur_poc - dpb[idx[i]].pic_order_cnt[0]) & 0x8000) != 0)
+			return 0;
+	}
+	return 1;
+}
+
+static void pre_slice_decode(struct rpivid_dec_env *const de,
+			     const struct rpivid_dec_state *const s)
+{
+	const struct v4l2_ctrl_hevc_slice_params *const sh = s->sh;
+	int weighted_pred_flag, idx;
+	u16 cmd_slice;
+	unsigned int collocated_from_l0_flag;
+
+	de->num_slice_msgs = 0;
+
+	cmd_slice = 0;
+	if (sh->slice_type == HEVC_SLICE_I)
+		cmd_slice = 1;
+	if (sh->slice_type == HEVC_SLICE_P)
+		cmd_slice = 2;
+	if (sh->slice_type == HEVC_SLICE_B)
+		cmd_slice = 3;
+
+	cmd_slice |= (s->nb_refs[L0] << 2) | (s->nb_refs[L1] << 6) |
+		     (s->max_num_merge_cand << 11);
+
+	collocated_from_l0_flag =
+		!s->slice_temporal_mvp ||
+		sh->slice_type != HEVC_SLICE_B ||
+		(sh->flags & V4L2_HEVC_SLICE_PARAMS_FLAG_COLLOCATED_FROM_L0);
+	cmd_slice |= collocated_from_l0_flag << 14;
+
+	if (sh->slice_type == HEVC_SLICE_P || sh->slice_type == HEVC_SLICE_B) {
+		// Flag to say all reference pictures are from the past
+		const int no_backward_pred_flag =
+			has_backward(sh->dpb, sh->ref_idx_l0, s->nb_refs[L0],
+				     sh->slice_pic_order_cnt) &&
+			has_backward(sh->dpb, sh->ref_idx_l1, s->nb_refs[L1],
+				     sh->slice_pic_order_cnt);
+		cmd_slice |= no_backward_pred_flag << 10;
+		msg_slice(de, cmd_slice);
+
+		if (s->slice_temporal_mvp) {
+			const __u8 *const rpl = collocated_from_l0_flag ?
+						sh->ref_idx_l0 : sh->ref_idx_l1;
+			de->dpbno_col = rpl[sh->collocated_ref_idx];
+			//v4l2_info(&de->ctx->dev->v4l2_dev,
+			//	    "L0=%d col_ref_idx=%d,
+			//          dpb_no=%d\n", collocated_from_l0_flag,
+			//          sh->collocated_ref_idx, de->dpbno_col);
+		}
+
+		// Write reference picture descriptions
+		weighted_pred_flag =
+			sh->slice_type == HEVC_SLICE_P ?
+				!!(s->pps.flags & V4L2_HEVC_PPS_FLAG_WEIGHTED_PRED) :
+				!!(s->pps.flags & V4L2_HEVC_PPS_FLAG_WEIGHTED_BIPRED);
+
+		for (idx = 0; idx < s->nb_refs[L0]; ++idx) {
+			unsigned int dpb_no = sh->ref_idx_l0[idx];
+			//v4l2_info(&de->ctx->dev->v4l2_dev,
+			//	  "L0[%d]=dpb[%d]\n", idx, dpb_no);
+
+			msg_slice(de,
+				  dpb_no |
+				  (sh->dpb[dpb_no].rps ==
+					V4L2_HEVC_DPB_ENTRY_RPS_LT_CURR ?
+						 (1 << 4) : 0) |
+				  (weighted_pred_flag ? (3 << 5) : 0));
+			msg_slice(de, sh->dpb[dpb_no].pic_order_cnt[0]);
+
+			if (weighted_pred_flag) {
+				const struct v4l2_hevc_pred_weight_table
+					*const w = &sh->pred_weight_table;
+				const int luma_weight_denom =
+					(1 << w->luma_log2_weight_denom);
+				const unsigned int chroma_log2_weight_denom =
+					(w->luma_log2_weight_denom +
+					 w->delta_chroma_log2_weight_denom);
+				const int chroma_weight_denom =
+					(1 << chroma_log2_weight_denom);
+
+				msg_slice(de,
+					  w->luma_log2_weight_denom |
+					  (((w->delta_luma_weight_l0[idx] +
+					     luma_weight_denom) & 0x1ff)
+						 << 3));
+				msg_slice(de, w->luma_offset_l0[idx] & 0xff);
+				msg_slice(de,
+					  chroma_log2_weight_denom |
+					  (((w->delta_chroma_weight_l0[idx][0] +
+					     chroma_weight_denom) & 0x1ff)
+						   << 3));
+				msg_slice(de,
+					  w->chroma_offset_l0[idx][0] & 0xff);
+				msg_slice(de,
+					  chroma_log2_weight_denom |
+					  (((w->delta_chroma_weight_l0[idx][1] +
+					     chroma_weight_denom) & 0x1ff)
+						   << 3));
+				msg_slice(de,
+					  w->chroma_offset_l0[idx][1] & 0xff);
+			}
+		}
+
+		for (idx = 0; idx < s->nb_refs[L1]; ++idx) {
+			unsigned int dpb_no = sh->ref_idx_l1[idx];
+			//v4l2_info(&de->ctx->dev->v4l2_dev,
+			//          "L1[%d]=dpb[%d]\n", idx, dpb_no);
+			msg_slice(de,
+				  dpb_no |
+				  (sh->dpb[dpb_no].rps ==
+					 V4L2_HEVC_DPB_ENTRY_RPS_LT_CURR ?
+						 (1 << 4) : 0) |
+					(weighted_pred_flag ? (3 << 5) : 0));
+			msg_slice(de, sh->dpb[dpb_no].pic_order_cnt[0]);
+			if (weighted_pred_flag) {
+				const struct v4l2_hevc_pred_weight_table
+					*const w = &sh->pred_weight_table;
+				const int luma_weight_denom =
+					(1 << w->luma_log2_weight_denom);
+				const unsigned int chroma_log2_weight_denom =
+					(w->luma_log2_weight_denom +
+					 w->delta_chroma_log2_weight_denom);
+				const int chroma_weight_denom =
+					(1 << chroma_log2_weight_denom);
+
+				msg_slice(de,
+					  w->luma_log2_weight_denom |
+					  (((w->delta_luma_weight_l1[idx] +
+					     luma_weight_denom) & 0x1ff) << 3));
+				msg_slice(de, w->luma_offset_l1[idx] & 0xff);
+				msg_slice(de,
+					  chroma_log2_weight_denom |
+					  (((w->delta_chroma_weight_l1[idx][0] +
+					     chroma_weight_denom) & 0x1ff)
+							<< 3));
+				msg_slice(de,
+					  w->chroma_offset_l1[idx][0] & 0xff);
+				msg_slice(de,
+					  chroma_log2_weight_denom |
+					  (((w->delta_chroma_weight_l1[idx][1] +
+					     chroma_weight_denom) & 0x1ff)
+						   << 3));
+				msg_slice(de,
+					  w->chroma_offset_l1[idx][1] & 0xff);
+			}
+		}
+	} else {
+		msg_slice(de, cmd_slice);
+	}
+
+	msg_slice(de,
+		  (sh->slice_beta_offset_div2 & 15) |
+		  ((sh->slice_tc_offset_div2 & 15) << 4) |
+		  ((sh->flags &
+		    V4L2_HEVC_SLICE_PARAMS_FLAG_SLICE_DEBLOCKING_FILTER_DISABLED) ?
+						1 << 8 : 0) |
+		  ((sh->flags &
+			  V4L2_HEVC_SLICE_PARAMS_FLAG_SLICE_LOOP_FILTER_ACROSS_SLICES_ENABLED) ?
+						1 << 9 : 0) |
+		  ((s->pps.flags &
+			  V4L2_HEVC_PPS_FLAG_LOOP_FILTER_ACROSS_TILES_ENABLED) ?
+						1 << 10 : 0));
+
+	msg_slice(de, ((sh->slice_cr_qp_offset & 31) << 5) +
+		       (sh->slice_cb_qp_offset & 31)); // CMD_QPOFF
+}
+
+//////////////////////////////////////////////////////////////////////////////
+// Write STATUS register with expected end CTU address of previous slice
+
+static void end_previous_slice(struct rpivid_dec_env *const de,
+			       const struct rpivid_dec_state *const s,
+			       const int ctb_addr_ts)
+{
+	int last_x =
+		s->ctb_addr_ts_to_rs[ctb_addr_ts - 1] % de->pic_width_in_ctbs_y;
+	int last_y =
+		s->ctb_addr_ts_to_rs[ctb_addr_ts - 1] / de->pic_width_in_ctbs_y;
+
+	p1_apb_write(de, RPI_STATUS, 1 + (last_x << 5) + (last_y << 18));
+}
+
+static void wpp_pause(struct rpivid_dec_env *const de, int ctb_row)
+{
+	p1_apb_write(de, RPI_STATUS, (ctb_row << 18) + 0x25);
+	p1_apb_write(de, RPI_TRANSFER, PROB_BACKUP);
+	p1_apb_write(de, RPI_MODE,
+		     ctb_row == de->pic_height_in_ctbs_y - 1 ?
+							0x70000 : 0x30000);
+	p1_apb_write(de, RPI_CONTROL, (ctb_row << 16) + 2);
+}
+
+static void wpp_end_previous_slice(struct rpivid_dec_env *const de,
+				   const struct rpivid_dec_state *const s,
+				   int ctb_addr_ts)
+{
+	int new_x = s->sh->slice_segment_addr % de->pic_width_in_ctbs_y;
+	int new_y = s->sh->slice_segment_addr / de->pic_width_in_ctbs_y;
+	int last_x =
+		s->ctb_addr_ts_to_rs[ctb_addr_ts - 1] % de->pic_width_in_ctbs_y;
+	int last_y =
+		s->ctb_addr_ts_to_rs[ctb_addr_ts - 1] / de->pic_width_in_ctbs_y;
+
+	if (de->wpp_entry_x < 2 && (de->wpp_entry_y < new_y || new_x > 2) &&
+	    de->pic_width_in_ctbs_y > 2)
+		wpp_pause(de, last_y);
+	p1_apb_write(de, RPI_STATUS, 1 + (last_x << 5) + (last_y << 18));
+	if (new_x == 2 || (de->pic_width_in_ctbs_y == 2 &&
+			   de->wpp_entry_y < new_y))
+		p1_apb_write(de, RPI_TRANSFER, PROB_BACKUP);
+}
+
+//////////////////////////////////////////////////////////////////////////////
+// Wavefront mode
+
+static void wpp_entry_point(struct rpivid_dec_env *const de,
+			    const struct rpivid_dec_state *const s,
+			    const int do_bte,
+			    const int reset_qp_y, const int ctb_addr_ts)
+{
+	int ctb_size = 1 << s->log2_ctb_size;
+	int ctb_addr_rs = s->ctb_addr_ts_to_rs[ctb_addr_ts];
+
+	int ctb_col = de->wpp_entry_x = ctb_addr_rs % de->pic_width_in_ctbs_y;
+	int ctb_row = de->wpp_entry_y = ctb_addr_rs / de->pic_width_in_ctbs_y;
+
+	int endx = de->pic_width_in_ctbs_y - 1;
+	int endy = ctb_row;
+
+	u8 slice_w = ctb_to_slice_w_h(ctb_col, ctb_size,
+				      s->sps.pic_width_in_luma_samples,
+				      s->col_bd, s->num_tile_columns);
+	u8 slice_h = ctb_to_slice_w_h(ctb_row, ctb_size,
+				      s->sps.pic_height_in_luma_samples,
+				      s->row_bd, s->num_tile_rows);
+
+	p1_apb_write(de, RPI_TILESTART, 0);
+	p1_apb_write(de, RPI_TILEEND, endx + (endy << 16));
+
+	if (do_bte)
+		p1_apb_write(de, RPI_BEGINTILEEND, endx + (endy << 16));
+
+	write_slice(de, s, slice_w,
+		    ctb_row == de->pic_height_in_ctbs_y - 1 ?
+							slice_h : ctb_size);
+
+	if (reset_qp_y) {
+		unsigned int sps_qp_bd_offset =
+			6 * s->sps.bit_depth_luma_minus8;
+
+		p1_apb_write(de, RPI_QP, sps_qp_bd_offset + s->slice_qp);
+	}
+
+	p1_apb_write(de, RPI_MODE,
+		     ctb_row == de->pic_height_in_ctbs_y - 1 ?
+							0x60001 : 0x20001);
+	p1_apb_write(de, RPI_CONTROL, (ctb_col << 0) + (ctb_row << 16));
+}
+
+//////////////////////////////////////////////////////////////////////////////
+// Wavefront mode
+
+static void wpp_decode_slice(struct rpivid_dec_env *const de,
+			     const struct rpivid_dec_state *const s,
+			     const struct v4l2_ctrl_hevc_slice_params *sh,
+			     int ctb_addr_ts)
+{
+	int i, reset_qp_y = 1;
+	int indep = !s->dependent_slice_segment_flag;
+	int ctb_col = s->sh->slice_segment_addr % de->pic_width_in_ctbs_y;
+
+	if (ctb_addr_ts)
+		wpp_end_previous_slice(de, s, ctb_addr_ts);
+	pre_slice_decode(de, s);
+	write_bitstream(de, s);
+	if (ctb_addr_ts == 0 || indep || de->pic_width_in_ctbs_y == 1)
+		write_prob(de, s);
+	else if (ctb_col == 0)
+		p1_apb_write(de, RPI_TRANSFER, PROB_RELOAD);
+	else
+		reset_qp_y = 0;
+	program_slicecmds(de, s->slice_idx);
+	new_slice_segment(de, s);
+	wpp_entry_point(de, s, indep, reset_qp_y, ctb_addr_ts);
+
+	for (i = 0; i < s->sh->num_entry_point_offsets; i++) {
+		int ctb_addr_rs = s->ctb_addr_ts_to_rs[ctb_addr_ts];
+		int ctb_row = ctb_addr_rs / de->pic_width_in_ctbs_y;
+		int last_x = de->pic_width_in_ctbs_y - 1;
+
+		if (de->pic_width_in_ctbs_y > 2)
+			wpp_pause(de, ctb_row);
+		p1_apb_write(de, RPI_STATUS,
+			     (ctb_row << 18) + (last_x << 5) + 2);
+		if (de->pic_width_in_ctbs_y == 2)
+			p1_apb_write(de, RPI_TRANSFER, PROB_BACKUP);
+		if (de->pic_width_in_ctbs_y == 1)
+			write_prob(de, s);
+		else
+			p1_apb_write(de, RPI_TRANSFER, PROB_RELOAD);
+		ctb_addr_ts += s->column_width[0];
+		wpp_entry_point(de, s, 0, 1, ctb_addr_ts);
+	}
+}
+
+//////////////////////////////////////////////////////////////////////////////
+// Tiles mode
+
+static void decode_slice(struct rpivid_dec_env *const de,
+			 const struct rpivid_dec_state *const s,
+			 const struct v4l2_ctrl_hevc_slice_params *const sh,
+			 int ctb_addr_ts)
+{
+	int i, reset_qp_y;
+
+	if (ctb_addr_ts)
+		end_previous_slice(de, s, ctb_addr_ts);
+
+	pre_slice_decode(de, s);
+	write_bitstream(de, s);
+
+#if DEBUG_TRACE_P1_CMD
+	if (p1_z < 256) {
+		v4l2_info(&de->ctx->dev->v4l2_dev,
+			  "TS=%d, tile=%d/%d, dss=%d, flags=%#llx\n",
+			  ctb_addr_ts, s->tile_id[ctb_addr_ts],
+			  s->tile_id[ctb_addr_ts - 1],
+			  s->dependent_slice_segment_flag, sh->flags);
+	}
+#endif
+
+	reset_qp_y = ctb_addr_ts == 0 ||
+		   s->tile_id[ctb_addr_ts] != s->tile_id[ctb_addr_ts - 1] ||
+		   !s->dependent_slice_segment_flag;
+	if (reset_qp_y)
+		write_prob(de, s);
+
+	program_slicecmds(de, s->slice_idx);
+	new_slice_segment(de, s);
+	new_entry_point(de, s, !s->dependent_slice_segment_flag, reset_qp_y,
+			ctb_addr_ts);
+
+	for (i = 0; i < s->sh->num_entry_point_offsets; i++) {
+		int ctb_addr_rs = s->ctb_addr_ts_to_rs[ctb_addr_ts];
+		int ctb_col = ctb_addr_rs % de->pic_width_in_ctbs_y;
+		int ctb_row = ctb_addr_rs / de->pic_width_in_ctbs_y;
+		int tile_x = ctb_to_tile(ctb_col, s->col_bd,
+					 s->num_tile_columns - 1);
+		int tile_y =
+			ctb_to_tile(ctb_row, s->row_bd, s->num_tile_rows - 1);
+		int last_x = s->col_bd[tile_x + 1] - 1;
+		int last_y = s->row_bd[tile_y + 1] - 1;
+
+		p1_apb_write(de, RPI_STATUS,
+			     2 + (last_x << 5) + (last_y << 18));
+		write_prob(de, s);
+		ctb_addr_ts += s->column_width[tile_x] * s->row_height[tile_y];
+		new_entry_point(de, s, 0, 1, ctb_addr_ts);
+	}
+}
+
+//////////////////////////////////////////////////////////////////////////////
+// Scaling factors
+
+static void expand_scaling_list(const unsigned int size_id,
+				const unsigned int matrix_id, u8 *const dst0,
+				const u8 *const src0, uint8_t dc)
+{
+	u8 *d;
+	unsigned int x, y;
+
+	// FIXME: matrix_id is unused ?
+	switch (size_id) {
+	case 0:
+		memcpy(dst0, src0, 16);
+		break;
+	case 1:
+		memcpy(dst0, src0, 64);
+		break;
+	case 2:
+		d = dst0;
+
+		for (y = 0; y != 16; y++) {
+			const u8 *s = src0 + (y >> 1) * 8;
+
+			for (x = 0; x != 8; ++x) {
+				*d++ = *s;
+				*d++ = *s++;
+			}
+		}
+		dst0[0] = dc;
+		break;
+	default:
+		d = dst0;
+
+		for (y = 0; y != 32; y++) {
+			const u8 *s = src0 + (y >> 2) * 8;
+
+			for (x = 0; x != 8; ++x) {
+				*d++ = *s;
+				*d++ = *s;
+				*d++ = *s;
+				*d++ = *s++;
+			}
+		}
+		dst0[0] = dc;
+		break;
+	}
+}
+
+static void populate_scaling_factors(const struct rpivid_run *const run,
+				     struct rpivid_dec_env *const de,
+				     const struct rpivid_dec_state *const s)
+{
+	const struct v4l2_ctrl_hevc_scaling_matrix *const sl =
+		run->h265.scaling_matrix;
+	// Array of constants for scaling factors
+	static const u32 scaling_factor_offsets[4][6] = {
+		// MID0    MID1    MID2    MID3    MID4    MID5
+		// SID0 (4x4)
+		{ 0x0000, 0x0010, 0x0020, 0x0030, 0x0040, 0x0050 },
+		// SID1 (8x8)
+		{ 0x0060, 0x00A0, 0x00E0, 0x0120, 0x0160, 0x01A0 },
+		// SID2 (16x16)
+		{ 0x01E0, 0x02E0, 0x03E0, 0x04E0, 0x05E0, 0x06E0 },
+		// SID3 (32x32)
+		{ 0x07E0, 0x0BE0, 0x0000, 0x0000, 0x0000, 0x0000 }
+	};
+
+	unsigned int mid;
+
+	for (mid = 0; mid < 6; mid++)
+		expand_scaling_list(0, mid,
+				    de->scaling_factors +
+					    scaling_factor_offsets[0][mid],
+				    sl->scaling_list_4x4[mid], 0);
+	for (mid = 0; mid < 6; mid++)
+		expand_scaling_list(1, mid,
+				    de->scaling_factors +
+					    scaling_factor_offsets[1][mid],
+				    sl->scaling_list_8x8[mid], 0);
+	for (mid = 0; mid < 6; mid++)
+		expand_scaling_list(2, mid,
+				    de->scaling_factors +
+					    scaling_factor_offsets[2][mid],
+				    sl->scaling_list_16x16[mid],
+				    sl->scaling_list_dc_coef_16x16[mid]);
+	for (mid = 0; mid < 2; mid += 1)
+		expand_scaling_list(3, mid,
+				    de->scaling_factors +
+					    scaling_factor_offsets[3][mid],
+				    sl->scaling_list_32x32[mid],
+				    sl->scaling_list_dc_coef_32x32[mid]);
+}
+
+static void free_ps_info(struct rpivid_dec_state *const s)
+{
+	kfree(s->ctb_addr_rs_to_ts);
+	s->ctb_addr_rs_to_ts = NULL;
+	kfree(s->ctb_addr_ts_to_rs);
+	s->ctb_addr_ts_to_rs = NULL;
+	kfree(s->tile_id);
+	s->tile_id = NULL;
+
+	kfree(s->col_bd);
+	s->col_bd = NULL;
+	kfree(s->row_bd);
+	s->row_bd = NULL;
+}
+
+static int updated_ps(struct rpivid_dec_state *const s)
+{
+	unsigned int ctb_addr_rs;
+	int j, x, y, tile_id;
+	unsigned int i;
+
+	free_ps_info(s);
+
+	// Inferred parameters
+	s->log2_ctb_size = s->sps.log2_min_luma_coding_block_size_minus3 + 3 +
+			   s->sps.log2_diff_max_min_luma_coding_block_size;
+
+	s->ctb_width = (s->sps.pic_width_in_luma_samples +
+			(1 << s->log2_ctb_size) - 1) >>
+		       s->log2_ctb_size;
+	s->ctb_height = (s->sps.pic_height_in_luma_samples +
+			 (1 << s->log2_ctb_size) - 1) >>
+			s->log2_ctb_size;
+	s->ctb_size = s->ctb_width * s->ctb_height;
+
+	// Inferred parameters
+
+	if (!(s->pps.flags & V4L2_HEVC_PPS_FLAG_TILES_ENABLED)) {
+		s->num_tile_columns = 1;
+		s->num_tile_rows = 1;
+		s->column_width[0] = s->ctb_width;
+		s->row_height[0] = s->ctb_height;
+	} else {
+		s->num_tile_columns = s->pps.num_tile_columns_minus1 + 1;
+		s->num_tile_rows = s->pps.num_tile_rows_minus1 + 1;
+		for (i = 0; i < s->num_tile_columns; ++i)
+			s->column_width[i] = s->pps.column_width_minus1[i] + 1;
+		for (i = 0; i < s->num_tile_rows; ++i)
+			s->row_height[i] = s->pps.row_height_minus1[i] + 1;
+	}
+
+	s->col_bd = kmalloc((s->num_tile_columns + 1) * sizeof(*s->col_bd),
+			    GFP_KERNEL);
+	s->row_bd = kmalloc((s->num_tile_rows + 1) * sizeof(*s->row_bd),
+			    GFP_KERNEL);
+
+	s->col_bd[0] = 0;
+	for (i = 0; i < s->num_tile_columns; i++)
+		s->col_bd[i + 1] = s->col_bd[i] + s->column_width[i];
+
+	s->row_bd[0] = 0;
+	for (i = 0; i < s->num_tile_rows; i++)
+		s->row_bd[i + 1] = s->row_bd[i] + s->row_height[i];
+
+	s->ctb_addr_rs_to_ts = kmalloc_array(s->ctb_size,
+					     sizeof(*s->ctb_addr_rs_to_ts),
+					     GFP_KERNEL);
+	s->ctb_addr_ts_to_rs = kmalloc_array(s->ctb_size,
+					     sizeof(*s->ctb_addr_ts_to_rs),
+					     GFP_KERNEL);
+	s->tile_id = kmalloc_array(s->ctb_size, sizeof(*s->tile_id),
+				   GFP_KERNEL);
+
+	for (ctb_addr_rs = 0; ctb_addr_rs < s->ctb_size; ctb_addr_rs++) {
+		int tb_x = ctb_addr_rs % s->ctb_width;
+		int tb_y = ctb_addr_rs / s->ctb_width;
+		int tile_x = 0;
+		int tile_y = 0;
+		int val = 0;
+
+		for (i = 0; i < s->num_tile_columns; i++) {
+			if (tb_x < s->col_bd[i + 1]) {
+				tile_x = i;
+				break;
+			}
+		}
+
+		for (i = 0; i < s->num_tile_rows; i++) {
+			if (tb_y < s->row_bd[i + 1]) {
+				tile_y = i;
+				break;
+			}
+		}
+
+		for (i = 0; i < tile_x; i++)
+			val += s->row_height[tile_y] * s->column_width[i];
+		for (i = 0; i < tile_y; i++)
+			val += s->ctb_width * s->row_height[i];
+
+		val += (tb_y - s->row_bd[tile_y]) * s->column_width[tile_x] +
+		       tb_x - s->col_bd[tile_x];
+
+		s->ctb_addr_rs_to_ts[ctb_addr_rs] = val;
+		s->ctb_addr_ts_to_rs[val] = ctb_addr_rs;
+	}
+
+	for (j = 0, tile_id = 0; j < s->num_tile_rows; j++)
+		for (i = 0; i < s->num_tile_columns; i++, tile_id++)
+			for (y = s->row_bd[j]; y < s->row_bd[j + 1]; y++)
+				for (x = s->col_bd[i];
+				     x < s->col_bd[i + 1];
+				     x++)
+					s->tile_id[s->ctb_addr_rs_to_ts
+							   [y * s->ctb_width +
+							    x]] = tile_id;
+
+	return 0;
+}
+
+static int frame_end(struct rpivid_dev *const dev,
+		     struct rpivid_dec_env *const de,
+		     const struct rpivid_dec_state *const s)
+{
+	const unsigned int last_x = s->col_bd[s->num_tile_columns] - 1;
+	const unsigned int last_y = s->row_bd[s->num_tile_rows] - 1;
+	size_t cmd_size;
+
+	if (s->pps.flags & V4L2_HEVC_PPS_FLAG_ENTROPY_CODING_SYNC_ENABLED) {
+		if (de->wpp_entry_x < 2 && de->pic_width_in_ctbs_y > 2)
+			wpp_pause(de, last_y);
+	}
+	p1_apb_write(de, RPI_STATUS, 1 + (last_x << 5) + (last_y << 18));
+
+	// Copy commands out to dma buf
+	cmd_size = de->cmd_len * sizeof(de->cmd_fifo[0]);
+
+	if (!de->cmd_copy_gptr->ptr || cmd_size > de->cmd_copy_gptr->size) {
+		size_t cmd_alloc = round_up_size(cmd_size);
+
+		if (gptr_realloc_new(dev, de->cmd_copy_gptr, cmd_alloc)) {
+			v4l2_err(&dev->v4l2_dev,
+				 "Alloc cmd buffer (%d): FAILED\n", cmd_alloc);
+			return -ENOMEM;
+		}
+		v4l2_info(&dev->v4l2_dev, "Alloc cmd buffer (%d): OK\n",
+			  cmd_alloc);
+	}
+
+	memcpy(de->cmd_copy_gptr->ptr, de->cmd_fifo, cmd_size);
+	return 0;
+}
+
+static void setup_colmv(struct rpivid_ctx *const ctx, struct rpivid_run *run,
+			struct rpivid_dec_state *const s)
+{
+	ctx->colmv_stride = ALIGN(s->sps.pic_width_in_luma_samples, 64);
+	ctx->colmv_picsize = ctx->colmv_stride *
+		(ALIGN(s->sps.pic_height_in_luma_samples, 64) >> 4);
+}
+
+// Can be called from irq context
+static struct rpivid_dec_env *dec_env_new(struct rpivid_ctx *const ctx)
+{
+	struct rpivid_dec_env *de;
+	unsigned long lock_flags;
+
+	spin_lock_irqsave(&ctx->dec_lock, lock_flags);
+
+	de = ctx->dec_free;
+	if (de) {
+		ctx->dec_free = de->next;
+		de->next = NULL;
+		de->state = RPIVID_DECODE_SLICE_START;
+	}
+
+	spin_unlock_irqrestore(&ctx->dec_lock, lock_flags);
+	return de;
+}
+
+// Can be called from irq context
+static void dec_env_delete(struct rpivid_dec_env *const de)
+{
+	struct rpivid_ctx * const ctx = de->ctx;
+	unsigned long lock_flags;
+
+	aux_q_release(ctx, &de->frame_aux);
+	aux_q_release(ctx, &de->col_aux);
+
+	spin_lock_irqsave(&ctx->dec_lock, lock_flags);
+
+	de->state = RPIVID_DECODE_END;
+	de->next = ctx->dec_free;
+	ctx->dec_free = de;
+
+	spin_unlock_irqrestore(&ctx->dec_lock, lock_flags);
+}
+
+static void dec_env_uninit(struct rpivid_ctx *const ctx)
+{
+	unsigned int i;
+
+	if (ctx->dec_pool) {
+		for (i = 0; i != RPIVID_DEC_ENV_COUNT; ++i) {
+			struct rpivid_dec_env *const de = ctx->dec_pool + i;
+
+			kfree(de->cmd_fifo);
+		}
+
+		kfree(ctx->dec_pool);
+	}
+
+	ctx->dec_pool = NULL;
+	ctx->dec_free = NULL;
+}
+
+static int dec_env_init(struct rpivid_ctx *const ctx)
+{
+	unsigned int i;
+
+	ctx->dec_pool = kzalloc(sizeof(*ctx->dec_pool) * RPIVID_DEC_ENV_COUNT,
+				GFP_KERNEL);
+	if (!ctx->dec_pool)
+		return -1;
+
+	spin_lock_init(&ctx->dec_lock);
+
+	// Build free chain
+	ctx->dec_free = ctx->dec_pool;
+	for (i = 0; i != RPIVID_DEC_ENV_COUNT - 1; ++i)
+		ctx->dec_pool[i].next = ctx->dec_pool + i + 1;
+
+	// Fill in other bits
+	for (i = 0; i != RPIVID_DEC_ENV_COUNT; ++i) {
+		struct rpivid_dec_env *const de = ctx->dec_pool + i;
+
+		de->ctx = ctx;
+		de->decode_order = i;
+		de->cmd_max = 1024;
+		de->cmd_fifo = kmalloc_array(de->cmd_max,
+					     sizeof(struct rpi_cmd),
+					     GFP_KERNEL);
+		if (!de->cmd_fifo)
+			goto fail;
+	}
+
+	return 0;
+
+fail:
+	dec_env_uninit(ctx);
+	return -1;
+}
+
+// Assume that we get exactly the same DPB for every slice
+// it makes no real sense otherwise
+#if V4L2_HEVC_DPB_ENTRIES_NUM_MAX > 16
+#error HEVC_DPB_ENTRIES > h/w slots
+#endif
+
+static u32 mk_config2(const struct rpivid_dec_state *const s)
+{
+	const struct v4l2_ctrl_hevc_sps *const sps = &s->sps;
+	const struct v4l2_ctrl_hevc_pps *const pps = &s->pps;
+	u32 c;
+	// BitDepthY
+	c = (sps->bit_depth_luma_minus8 + 8) << 0;
+	 // BitDepthC
+	c |= (sps->bit_depth_chroma_minus8 + 8) << 4;
+	 // BitDepthY
+	if (sps->bit_depth_luma_minus8)
+		c |= BIT(8);
+	// BitDepthC
+	if (sps->bit_depth_chroma_minus8)
+		c |= BIT(9);
+	c |= s->log2_ctb_size << 10;
+	if (pps->flags & V4L2_HEVC_PPS_FLAG_CONSTRAINED_INTRA_PRED)
+		c |= BIT(13);
+	if (sps->flags & V4L2_HEVC_SPS_FLAG_STRONG_INTRA_SMOOTHING_ENABLED)
+		c |= BIT(14);
+	if (sps->flags & V4L2_HEVC_SPS_FLAG_SPS_TEMPORAL_MVP_ENABLED)
+		c |= BIT(15); /* Write motion vectors to external memory */
+	c |= (pps->log2_parallel_merge_level_minus2 + 2) << 16;
+	if (s->slice_temporal_mvp)
+		c |= BIT(19);
+	if (sps->flags & V4L2_HEVC_SPS_FLAG_PCM_LOOP_FILTER_DISABLED)
+		c |= BIT(20);
+	c |= (pps->pps_cb_qp_offset & 31) << 21;
+	c |= (pps->pps_cr_qp_offset & 31) << 26;
+	return c;
+}
+
+static void rpivid_h265_setup(struct rpivid_ctx *ctx, struct rpivid_run *run)
+{
+	struct rpivid_dev *const dev = ctx->dev;
+	const struct v4l2_ctrl_hevc_slice_params *const sh =
+						run->h265.slice_params;
+	const struct v4l2_hevc_pred_weight_table *pred_weight_table;
+	struct rpivid_q_aux *dpb_q_aux[V4L2_HEVC_DPB_ENTRIES_NUM_MAX];
+	struct rpivid_dec_state *const s = ctx->state;
+	struct vb2_queue *vq;
+	struct rpivid_dec_env *de;
+	int ctb_addr_ts;
+	unsigned int i;
+	int use_aux;
+	bool slice_temporal_mvp;
+
+	pred_weight_table = &sh->pred_weight_table;
+
+	s->frame_end =
+		((run->src->flags & V4L2_BUF_FLAG_M2M_HOLD_CAPTURE_BUF) == 0);
+
+	de = ctx->dec0;
+	slice_temporal_mvp = (sh->flags &
+		   V4L2_HEVC_SLICE_PARAMS_FLAG_SLICE_TEMPORAL_MVP_ENABLED);
+
+	if (de && de->state != RPIVID_DECODE_END) {
+		++s->slice_idx;
+
+		switch (de->state) {
+		case RPIVID_DECODE_SLICE_CONTINUE:
+			// Expected state
+			break;
+		default:
+			v4l2_err(&dev->v4l2_dev, "%s: Unexpected state: %d\n",
+				 __func__, de->state);
+		/* FALLTHRU */
+		case RPIVID_DECODE_ERROR_CONTINUE:
+			// Uncleared error - fail now
+			goto fail;
+		}
+
+		if (s->slice_temporal_mvp != slice_temporal_mvp) {
+			v4l2_warn(&dev->v4l2_dev,
+				  "Slice Temporal MVP non-constant\n");
+			goto fail;
+		}
+	} else {
+		/* Frame start */
+		unsigned int ctb_size_y;
+		bool sps_changed = false;
+
+		if (memcmp(&s->sps, run->h265.sps, sizeof(s->sps)) != 0) {
+			/* SPS changed */
+			v4l2_info(&dev->v4l2_dev, "SPS changed\n");
+			memcpy(&s->sps, run->h265.sps, sizeof(s->sps));
+			sps_changed = true;
+		}
+		if (sps_changed ||
+		    memcmp(&s->pps, run->h265.pps, sizeof(s->pps)) != 0) {
+			/* SPS changed */
+			v4l2_info(&dev->v4l2_dev, "PPS changed\n");
+			memcpy(&s->pps, run->h265.pps, sizeof(s->pps));
+
+			/* Recalc stuff as required */
+			updated_ps(s);
+		}
+
+		de = dec_env_new(ctx);
+		if (!de) {
+			v4l2_err(&dev->v4l2_dev,
+				 "Failed to find free decode env\n");
+			goto fail;
+		}
+		ctx->dec0 = de;
+
+		ctb_size_y =
+			1U << (s->sps.log2_min_luma_coding_block_size_minus3 +
+			       3 +
+			       s->sps.log2_diff_max_min_luma_coding_block_size);
+
+		de->pic_width_in_ctbs_y =
+			(s->sps.pic_width_in_luma_samples + ctb_size_y - 1) /
+				ctb_size_y; // 7-15
+		de->pic_height_in_ctbs_y =
+			(s->sps.pic_height_in_luma_samples + ctb_size_y - 1) /
+				ctb_size_y; // 7-17
+		de->cmd_len = 0;
+		de->dpbno_col = ~0U;
+
+		de->bit_copy_gptr = ctx->bitbufs + 0;
+		de->bit_copy_len = 0;
+		de->cmd_copy_gptr = ctx->cmdbufs + 0;
+
+		de->frame_c_offset = ctx->dst_fmt.height * 128;
+		de->frame_stride = ctx->dst_fmt.bytesperline * 128;
+		de->frame_addr =
+			vb2_dma_contig_plane_dma_addr(&run->dst->vb2_buf, 0);
+		de->frame_aux = NULL;
+
+		if (s->sps.bit_depth_luma_minus8 !=
+		    s->sps.bit_depth_chroma_minus8) {
+			v4l2_warn(&dev->v4l2_dev,
+				  "Chroma depth (%d) != Luma depth (%d)\n",
+				  s->sps.bit_depth_chroma_minus8 + 8,
+				  s->sps.bit_depth_luma_minus8 + 8);
+			goto fail;
+		}
+		if (s->sps.bit_depth_luma_minus8 == 0) {
+			if (ctx->dst_fmt.pixelformat !=
+						V4L2_PIX_FMT_NV12_COL128) {
+				v4l2_err(&dev->v4l2_dev,
+					 "Pixel format %#x != NV12_COL128 for 8-bit output",
+					 ctx->dst_fmt.pixelformat);
+				goto fail;
+			}
+		} else if (s->sps.bit_depth_luma_minus8 == 2) {
+			if (ctx->dst_fmt.pixelformat !=
+						V4L2_PIX_FMT_NV12_10_COL128) {
+				v4l2_err(&dev->v4l2_dev,
+					 "Pixel format %#x != NV12_10_COL128 for 10-bit output",
+					 ctx->dst_fmt.pixelformat);
+				goto fail;
+			}
+		} else {
+			v4l2_warn(&dev->v4l2_dev,
+				  "Luma depth (%d) unsupported\n",
+				  s->sps.bit_depth_luma_minus8 + 8);
+			goto fail;
+		}
+		if (run->dst->vb2_buf.num_planes != 1) {
+			v4l2_warn(&dev->v4l2_dev, "Capture planes (%d) != 1\n",
+				  run->dst->vb2_buf.num_planes);
+			goto fail;
+		}
+		if (run->dst->planes[0].length <
+		    ctx->dst_fmt.sizeimage) {
+			v4l2_warn(&dev->v4l2_dev,
+				  "Capture plane[0] length (%d) < sizeimage (%d)\n",
+				  run->dst->planes[0].length,
+				  ctx->dst_fmt.sizeimage);
+			goto fail;
+		}
+
+		if (s->sps.pic_width_in_luma_samples > 4096 ||
+		    s->sps.pic_height_in_luma_samples > 4096) {
+			v4l2_warn(&dev->v4l2_dev,
+				  "Pic dimension (%dx%d) exeeds 4096\n",
+				  s->sps.pic_width_in_luma_samples,
+				  s->sps.pic_height_in_luma_samples);
+			goto fail;
+		}
+
+		// Fill in ref planes with our address s.t. if we mess
+		// up refs somehow then we still have a valid address
+		// entry
+		for (i = 0; i != 16; ++i)
+			de->ref_addrs[i] = de->frame_addr;
+
+		/*
+		 * Stash initial temporal_mvp flag
+		 * This must be the same for all pic slices (7.4.7.1)
+		 */
+		s->slice_temporal_mvp = slice_temporal_mvp;
+
+		// Phase 2 reg pre-calc
+		de->rpi_config2 = mk_config2(s);
+		de->rpi_framesize = (s->sps.pic_height_in_luma_samples << 16) |
+				    s->sps.pic_width_in_luma_samples;
+		de->rpi_currpoc = sh->slice_pic_order_cnt;
+
+		if (s->sps.flags &
+		    V4L2_HEVC_SPS_FLAG_SPS_TEMPORAL_MVP_ENABLED) {
+			setup_colmv(ctx, run, s);
+		}
+
+		s->slice_idx = 0;
+
+		if (sh->slice_segment_addr != 0) {
+			v4l2_warn(&dev->v4l2_dev,
+				  "New frame but segment_addr=%d\n",
+				  sh->slice_segment_addr);
+			goto fail;
+		}
+
+		/* Allocate a bitbuf if we need one - don't need one if single
+		 * slice as we can use the src buf directly
+		 */
+		if (!s->frame_end && !de->bit_copy_gptr->ptr) {
+			const size_t wxh = s->sps.pic_width_in_luma_samples *
+				s->sps.pic_height_in_luma_samples;
+			size_t bits_alloc;
+
+			/* Annex A gives a min compression of 2 @ lvl 3.1
+			 * (wxh <= 983040) and min 4 thereafter but avoid
+			 * the odity of 983041 having a lower limit than
+			 * 983040.
+			 * Multiply by 3/2 for 4:2:0
+			 */
+			bits_alloc = wxh < 983040 ? wxh * 3 / 4 :
+				wxh < 983040 * 2 ? 983040 * 3 / 4 :
+				wxh * 3 / 8;
+			bits_alloc = round_up_size(bits_alloc);
+
+			if (gptr_alloc(dev, de->bit_copy_gptr,
+				       bits_alloc,
+				       DMA_ATTR_FORCE_CONTIGUOUS) != 0) {
+				v4l2_err(&dev->v4l2_dev,
+					 "Unable to alloc buf (%d) for bit copy\n",
+					 bits_alloc);
+				goto fail;
+			}
+			v4l2_info(&dev->v4l2_dev,
+				  "Alloc buf (%d) for bit copy OK\n",
+				  bits_alloc);
+		}
+	}
+
+	// Pre calc a few things
+	s->src_addr =
+		!s->frame_end ?
+			0 :
+			vb2_dma_contig_plane_dma_addr(&run->src->vb2_buf, 0);
+	s->src_buf = s->src_addr != 0 ? NULL :
+					vb2_plane_vaddr(&run->src->vb2_buf, 0);
+	if (!s->src_addr && !s->src_buf) {
+		v4l2_err(&dev->v4l2_dev, "Failed to map src buffer\n");
+		goto fail;
+	}
+
+	s->sh = sh;
+	s->slice_qp = 26 + s->pps.init_qp_minus26 + s->sh->slice_qp_delta;
+	s->max_num_merge_cand = sh->slice_type == HEVC_SLICE_I ?
+					0 :
+					(5 - sh->five_minus_max_num_merge_cand);
+	// * SH DSS flag invented by me - but clearly needed
+	s->dependent_slice_segment_flag =
+		((sh->flags &
+		  V4L2_HEVC_SLICE_PARAMS_FLAG_DEPENDENT_SLICE_SEGMENT) != 0);
+
+	s->nb_refs[0] = (sh->slice_type == HEVC_SLICE_I) ?
+				0 :
+				sh->num_ref_idx_l0_active_minus1 + 1;
+	s->nb_refs[1] = (sh->slice_type != HEVC_SLICE_B) ?
+				0 :
+				sh->num_ref_idx_l1_active_minus1 + 1;
+
+	if (s->sps.flags & V4L2_HEVC_SPS_FLAG_SCALING_LIST_ENABLED)
+		populate_scaling_factors(run, de, s);
+
+	ctb_addr_ts = s->ctb_addr_rs_to_ts[sh->slice_segment_addr];
+
+	if ((s->pps.flags & V4L2_HEVC_PPS_FLAG_ENTROPY_CODING_SYNC_ENABLED))
+		wpp_decode_slice(de, s, sh, ctb_addr_ts);
+	else
+		decode_slice(de, s, sh, ctb_addr_ts);
+
+	if (!s->frame_end)
+		return;
+
+	// Frame end
+	memset(dpb_q_aux, 0,
+	       sizeof(*dpb_q_aux) * V4L2_HEVC_DPB_ENTRIES_NUM_MAX);
+	/*
+	 * Need Aux ents for all (ref) DPB ents if temporal MV could
+	 * be enabled for any pic
+	 * ** At the moment we have aux ents for all pics whether or not
+	 *    they are ref
+	 */
+	use_aux = ((s->sps.flags &
+		  V4L2_HEVC_SPS_FLAG_SPS_TEMPORAL_MVP_ENABLED) != 0);
+
+	// Locate ref frames
+	// At least in the current implementation this is constant across all
+	// slices. If this changes we will need idx mapping code.
+	// Uses sh so here rather than trigger
+
+	vq = v4l2_m2m_get_vq(ctx->fh.m2m_ctx, V4L2_BUF_TYPE_VIDEO_CAPTURE);
+
+	if (!vq) {
+		v4l2_err(&dev->v4l2_dev, "VQ gone!\n");
+		goto fail;
+	}
+
+	//        v4l2_info(&dev->v4l2_dev, "rpivid_h265_end of frame\n");
+	if (frame_end(dev, de, s))
+		goto fail;
+
+	for (i = 0; i < sh->num_active_dpb_entries; ++i) {
+		int buffer_index =
+			vb2_find_timestamp(vq, sh->dpb[i].timestamp, 0);
+		struct vb2_buffer *buf = buffer_index < 0 ?
+					NULL :
+					vb2_get_buffer(vq, buffer_index);
+
+		if (!buf) {
+			v4l2_warn(&dev->v4l2_dev,
+				  "Missing DPB ent %d, timestamp=%lld, index=%d\n",
+				  i, (long long)sh->dpb[i].timestamp,
+				  buffer_index);
+			continue;
+		}
+
+		if (use_aux) {
+			dpb_q_aux[i] = aux_q_ref(ctx,
+						 ctx->aux_ents[buffer_index]);
+			if (!dpb_q_aux[i])
+				v4l2_warn(&dev->v4l2_dev,
+					  "Missing DPB AUX ent %d index=%d\n",
+					  i, buffer_index);
+		}
+
+		de->ref_addrs[i] =
+			vb2_dma_contig_plane_dma_addr(buf, 0);
+	}
+
+	// Move DPB from temp
+	for (i = 0; i != V4L2_HEVC_DPB_ENTRIES_NUM_MAX; ++i) {
+		aux_q_release(ctx, &s->ref_aux[i]);
+		s->ref_aux[i] = dpb_q_aux[i];
+	}
+	// Unref the old frame aux too - it is either in the DPB or not
+	// now
+	aux_q_release(ctx, &s->frame_aux);
+
+	if (use_aux) {
+		// New frame so new aux ent
+		// ??? Do we need this if non-ref ??? can we tell
+		s->frame_aux = aux_q_new(ctx, run->dst->vb2_buf.index);
+
+		if (!s->frame_aux) {
+			v4l2_err(&dev->v4l2_dev,
+				 "Failed to obtain aux storage for frame\n");
+			goto fail;
+		}
+
+		de->frame_aux = aux_q_ref(ctx, s->frame_aux);
+	}
+
+	if (de->dpbno_col != ~0U) {
+		if (de->dpbno_col >= sh->num_active_dpb_entries) {
+			v4l2_err(&dev->v4l2_dev,
+				 "Col ref index %d >= %d\n",
+				 de->dpbno_col,
+				 sh->num_active_dpb_entries);
+		} else {
+			// Standard requires that the col pic is
+			// constant for the duration of the pic
+			// (text of collocated_ref_idx in H265-2 2018
+			// 7.4.7.1)
+
+			// Spot the collocated ref in passing
+			de->col_aux = aux_q_ref(ctx,
+						dpb_q_aux[de->dpbno_col]);
+
+			if (!de->col_aux) {
+				v4l2_warn(&dev->v4l2_dev,
+					  "Missing DPB ent for col\n");
+				// Probably need to abort if this fails
+				// as P2 may explode on bad data
+				goto fail;
+			}
+		}
+	}
+
+	de->state = RPIVID_DECODE_PHASE1;
+	return;
+
+fail:
+	if (de)
+		// Actual error reporting happens in Trigger
+		de->state = s->frame_end ? RPIVID_DECODE_ERROR_DONE :
+					   RPIVID_DECODE_ERROR_CONTINUE;
+}
+
+//////////////////////////////////////////////////////////////////////////////
+// Handle PU and COEFF stream overflow
+
+// Returns:
+// -1  Phase 1 decode error
+//  0  OK
+// >0  Out of space (bitmask)
+
+#define STATUS_COEFF_EXHAUSTED	8
+#define STATUS_PU_EXHAUSTED	16
+
+static int check_status(const struct rpivid_dev *const dev)
+{
+	const u32 cfstatus = apb_read(dev, RPI_CFSTATUS);
+	const u32 cfnum = apb_read(dev, RPI_CFNUM);
+	u32 status = apb_read(dev, RPI_STATUS);
+
+	// Handle PU and COEFF stream overflow
+
+	// this is the definition of successful completion of phase 1
+	// it assures that status register is zero and all blocks in each tile
+	// have completed
+	if (cfstatus == cfnum)
+		return 0;	//No error
+
+	status &= (STATUS_PU_EXHAUSTED | STATUS_COEFF_EXHAUSTED);
+	if (status)
+		return status;
+
+	return -1;
+}
+
+static void cb_phase2(struct rpivid_dev *const dev, void *v)
+{
+	struct rpivid_dec_env *const de = v;
+	struct rpivid_ctx *const ctx = de->ctx;
+
+	xtrace_in(dev, de);
+
+	v4l2_m2m_cap_buf_return(dev->m2m_dev, ctx->fh.m2m_ctx, de->frame_buf,
+				VB2_BUF_STATE_DONE);
+	de->frame_buf = NULL;
+
+	/* Delete de before finish as finish might immediately trigger a reuse
+	 * of de
+	 */
+	dec_env_delete(de);
+
+	if (atomic_add_return(-1, &ctx->p2out) >= RPIVID_P2BUF_COUNT - 1) {
+		xtrace_fin(dev, de);
+		v4l2_m2m_buf_done_and_job_finish(dev->m2m_dev, ctx->fh.m2m_ctx,
+						 VB2_BUF_STATE_DONE);
+	}
+
+	xtrace_ok(dev, de);
+}
+
+static void phase2_claimed(struct rpivid_dev *const dev, void *v)
+{
+	struct rpivid_dec_env *const de = v;
+	unsigned int i;
+
+	xtrace_in(dev, de);
+
+	apb_write_vc_addr(dev, RPI_PURBASE, de->pu_base_vc);
+	apb_write_vc_len(dev, RPI_PURSTRIDE, de->pu_stride);
+	apb_write_vc_addr(dev, RPI_COEFFRBASE, de->coeff_base_vc);
+	apb_write_vc_len(dev, RPI_COEFFRSTRIDE, de->coeff_stride);
+
+	apb_write_vc_addr(dev, RPI_OUTYBASE, de->frame_addr);
+	apb_write_vc_addr(dev, RPI_OUTCBASE,
+			  de->frame_addr + de->frame_c_offset);
+	apb_write_vc_len(dev, RPI_OUTYSTRIDE, de->frame_stride);
+	apb_write_vc_len(dev, RPI_OUTCSTRIDE, de->frame_stride);
+
+	//    v4l2_info(&dev->v4l2_dev, "Frame: Y=%llx, C=%llx, Stride=%x\n",
+	//              de->frame_addr, de->frame_addr + de->frame_c_offset,
+	//              de->frame_stride);
+
+	for (i = 0; i < 16; i++) {
+		// Strides are in fact unused but fill in anyway
+		apb_write_vc_addr(dev, 0x9000 + 16 * i, de->ref_addrs[i]);
+		apb_write_vc_len(dev, 0x9004 + 16 * i, de->frame_stride);
+		apb_write_vc_addr(dev, 0x9008 + 16 * i,
+				  de->ref_addrs[i] + de->frame_c_offset);
+		apb_write_vc_len(dev, 0x900C + 16 * i, de->frame_stride);
+	}
+
+	apb_write(dev, RPI_CONFIG2, de->rpi_config2);
+	apb_write(dev, RPI_FRAMESIZE, de->rpi_framesize);
+	apb_write(dev, RPI_CURRPOC, de->rpi_currpoc);
+	//    v4l2_info(&dev->v4l2_dev, "Config2=%#x, FrameSize=%#x, POC=%#x\n",
+	//    de->rpi_config2, de->rpi_framesize, de->rpi_currpoc);
+
+	// collocated reads/writes
+	apb_write_vc_len(dev, RPI_COLSTRIDE,
+			 de->ctx->colmv_stride); // Read vals
+	apb_write_vc_len(dev, RPI_MVSTRIDE,
+			 de->ctx->colmv_stride); // Write vals
+	apb_write_vc_addr(dev, RPI_MVBASE,
+			  !de->frame_aux ? 0 : de->frame_aux->col.addr);
+	apb_write_vc_addr(dev, RPI_COLBASE,
+			  !de->col_aux ? 0 : de->col_aux->col.addr);
+
+	//v4l2_info(&dev->v4l2_dev,
+	//	   "Mv=%llx, Col=%llx, Stride=%x, Buf=%llx->%llx\n",
+	//	   de->rpi_mvbase, de->rpi_colbase, de->ctx->colmv_stride,
+	//	   de->ctx->colmvbuf.addr, de->ctx->colmvbuf.addr +
+	//	   de->ctx->colmvbuf.size);
+
+	rpivid_hw_irq_active2_irq(dev, &de->irq_ent, cb_phase2, de);
+
+	apb_write_final(dev, RPI_NUMROWS, de->pic_height_in_ctbs_y);
+
+	xtrace_ok(dev, de);
+}
+
+static void phase1_claimed(struct rpivid_dev *const dev, void *v);
+
+static void phase1_thread(struct rpivid_dev *const dev, void *v)
+{
+	struct rpivid_dec_env *const de = v;
+	struct rpivid_ctx *const ctx = de->ctx;
+
+	struct rpivid_gptr *const pu_gptr = ctx->pu_bufs + ctx->p2idx;
+	struct rpivid_gptr *const coeff_gptr = ctx->coeff_bufs + ctx->p2idx;
+
+	xtrace_in(dev, de);
+
+	if (de->p1_status & STATUS_PU_EXHAUSTED) {
+		if (gptr_realloc_new(dev, pu_gptr, next_size(pu_gptr->size))) {
+			v4l2_err(&dev->v4l2_dev,
+				 "%s: PU realloc (%#x) failed\n",
+				 __func__, pu_gptr->size);
+			goto fail;
+		}
+		v4l2_info(&dev->v4l2_dev, "%s: PU realloc (%#x) OK\n",
+			  __func__, pu_gptr->size);
+	}
+
+	if (de->p1_status & STATUS_COEFF_EXHAUSTED) {
+		if (gptr_realloc_new(dev, coeff_gptr,
+				     next_size(coeff_gptr->size))) {
+			v4l2_err(&dev->v4l2_dev,
+				 "%s: Coeff realloc (%#x) failed\n",
+				 __func__, coeff_gptr->size);
+			goto fail;
+		}
+		v4l2_info(&dev->v4l2_dev, "%s: Coeff realloc (%#x) OK\n",
+			  __func__, coeff_gptr->size);
+	}
+
+	phase1_claimed(dev, de);
+	xtrace_ok(dev, de);
+	return;
+
+fail:
+	dec_env_delete(de);
+	xtrace_fin(dev, de);
+	v4l2_m2m_buf_done_and_job_finish(dev->m2m_dev, ctx->fh.m2m_ctx,
+					 VB2_BUF_STATE_ERROR);
+	xtrace_fail(dev, de);
+}
+
+/* Always called in irq context (this is good) */
+static void cb_phase1(struct rpivid_dev *const dev, void *v)
+{
+	struct rpivid_dec_env *const de = v;
+	struct rpivid_ctx *const ctx = de->ctx;
+
+	xtrace_in(dev, de);
+
+	de->p1_status = check_status(dev);
+	if (de->p1_status != 0) {
+		v4l2_info(&dev->v4l2_dev, "%s: Post wait: %#x\n",
+			  __func__, de->p1_status);
+
+		if (de->p1_status < 0)
+			goto fail;
+
+		/* Need to realloc - push onto a thread rather than IRQ */
+		rpivid_hw_irq_active1_thread(dev, &de->irq_ent,
+					     phase1_thread, de);
+		return;
+	}
+
+	/* After the frame-buf is detached it must be returned but from
+	 * this point onward (phase2_claimed, cb_phase2) there are no error
+	 * paths so the return at the end of cb_phase2 is all that is needed
+	 */
+	de->frame_buf = v4l2_m2m_cap_buf_detach(dev->m2m_dev, ctx->fh.m2m_ctx);
+	if (!de->frame_buf) {
+		v4l2_err(&dev->v4l2_dev, "%s: No detached buffer\n", __func__);
+		goto fail;
+	}
+
+	ctx->p2idx =
+		(ctx->p2idx + 1 >= RPIVID_P2BUF_COUNT) ? 0 : ctx->p2idx + 1;
+
+	// Enable the next setup if our Q isn't too big
+	if (atomic_add_return(1, &ctx->p2out) < RPIVID_P2BUF_COUNT) {
+		xtrace_fin(dev, de);
+		v4l2_m2m_buf_done_and_job_finish(dev->m2m_dev, ctx->fh.m2m_ctx,
+						 VB2_BUF_STATE_DONE);
+	}
+
+	rpivid_hw_irq_active2_claim(dev, &de->irq_ent, phase2_claimed, de);
+
+	xtrace_ok(dev, de);
+	return;
+
+fail:
+	dec_env_delete(de);
+	xtrace_fin(dev, de);
+	v4l2_m2m_buf_done_and_job_finish(dev->m2m_dev, ctx->fh.m2m_ctx,
+					 VB2_BUF_STATE_ERROR);
+	xtrace_fail(dev, de);
+}
+
+static void phase1_claimed(struct rpivid_dev *const dev, void *v)
+{
+	struct rpivid_dec_env *const de = v;
+	struct rpivid_ctx *const ctx = de->ctx;
+
+	const struct rpivid_gptr * const pu_gptr = ctx->pu_bufs + ctx->p2idx;
+	const struct rpivid_gptr * const coeff_gptr = ctx->coeff_bufs +
+						      ctx->p2idx;
+
+	xtrace_in(dev, de);
+
+	de->pu_base_vc = pu_gptr->addr;
+	de->pu_stride =
+		ALIGN_DOWN(pu_gptr->size / de->pic_height_in_ctbs_y, 64);
+
+	de->coeff_base_vc = coeff_gptr->addr;
+	de->coeff_stride =
+		ALIGN_DOWN(coeff_gptr->size / de->pic_height_in_ctbs_y, 64);
+
+	apb_write_vc_addr(dev, RPI_PUWBASE, de->pu_base_vc);
+	apb_write_vc_len(dev, RPI_PUWSTRIDE, de->pu_stride);
+	apb_write_vc_addr(dev, RPI_COEFFWBASE, de->coeff_base_vc);
+	apb_write_vc_len(dev, RPI_COEFFWSTRIDE, de->coeff_stride);
+
+	// Trigger command FIFO
+	apb_write(dev, RPI_CFNUM, de->cmd_len);
+
+	// Claim irq
+	rpivid_hw_irq_active1_irq(dev, &de->irq_ent, cb_phase1, de);
+
+	// And start the h/w
+	apb_write_vc_addr_final(dev, RPI_CFBASE, de->cmd_copy_gptr->addr);
+
+	xtrace_ok(dev, de);
+}
+
+static void dec_state_delete(struct rpivid_ctx *const ctx)
+{
+	unsigned int i;
+	struct rpivid_dec_state *const s = ctx->state;
+
+	if (!s)
+		return;
+	ctx->state = NULL;
+
+	free_ps_info(s);
+
+	for (i = 0; i != HEVC_MAX_REFS; ++i)
+		aux_q_release(ctx, &s->ref_aux[i]);
+	aux_q_release(ctx, &s->frame_aux);
+
+	kfree(s);
+}
+
+static void rpivid_h265_stop(struct rpivid_ctx *ctx)
+{
+	struct rpivid_dev *const dev = ctx->dev;
+	unsigned int i;
+
+	v4l2_info(&dev->v4l2_dev, "%s\n", __func__);
+
+	dec_env_uninit(ctx);
+	dec_state_delete(ctx);
+
+	// dec_env & state must be killed before this to release the buffer to
+	// the free pool
+	aux_q_uninit(ctx);
+
+	for (i = 0; i != ARRAY_SIZE(ctx->bitbufs); ++i)
+		gptr_free(dev, ctx->bitbufs + i);
+	for (i = 0; i != ARRAY_SIZE(ctx->cmdbufs); ++i)
+		gptr_free(dev, ctx->cmdbufs + i);
+	for (i = 0; i != ARRAY_SIZE(ctx->pu_bufs); ++i)
+		gptr_free(dev, ctx->pu_bufs + i);
+	for (i = 0; i != ARRAY_SIZE(ctx->coeff_bufs); ++i)
+		gptr_free(dev, ctx->coeff_bufs + i);
+}
+
+static int rpivid_h265_start(struct rpivid_ctx *ctx)
+{
+	struct rpivid_dev *const dev = ctx->dev;
+	unsigned int i;
+
+	unsigned int w = ctx->dst_fmt.width;
+	unsigned int h = ctx->dst_fmt.height;
+	unsigned int wxh;
+	size_t pu_alloc;
+	size_t coeff_alloc;
+
+	// Generate a sanitised WxH for memory alloc
+	// Assume HD if unset
+	if (w == 0)
+		w = 1920;
+	if (w > 4096)
+		w = 4096;
+	if (h == 0)
+		w = 1088;
+	if (h > 4096)
+		h = 4096;
+	wxh = w * h;
+
+	v4l2_info(&dev->v4l2_dev, "%s: (%dx%d)\n", __func__,
+		  ctx->dst_fmt.width, ctx->dst_fmt.height);
+
+	ctx->dec0 = NULL;
+	ctx->state = kzalloc(sizeof(*ctx->state), GFP_KERNEL);
+	if (!ctx->state) {
+		v4l2_err(&dev->v4l2_dev, "Failed to allocate decode state\n");
+		goto fail;
+	}
+
+	if (dec_env_init(ctx) != 0) {
+		v4l2_err(&dev->v4l2_dev, "Failed to allocate decode envs\n");
+		goto fail;
+	}
+
+	// 16k is plenty for most purposes but we will realloc if needed
+	for (i = 0; i != ARRAY_SIZE(ctx->cmdbufs); ++i) {
+		if (gptr_alloc(dev, ctx->cmdbufs + i, 0x4000,
+			       DMA_ATTR_FORCE_CONTIGUOUS))
+			goto fail;
+	}
+
+	// Finger in the air PU & Coeff alloc
+	// Will be realloced if too small
+	coeff_alloc = round_up_size(wxh);
+	pu_alloc = round_up_size(wxh / 4);
+	for (i = 0; i != ARRAY_SIZE(ctx->pu_bufs); ++i) {
+		// Don't actually need a kernel mapping here
+		if (gptr_alloc(dev, ctx->pu_bufs + i, pu_alloc,
+			       DMA_ATTR_FORCE_CONTIGUOUS |
+					DMA_ATTR_NO_KERNEL_MAPPING))
+			goto fail;
+		if (gptr_alloc(dev, ctx->coeff_bufs + i, coeff_alloc,
+			       DMA_ATTR_FORCE_CONTIGUOUS |
+					DMA_ATTR_NO_KERNEL_MAPPING))
+			goto fail;
+	}
+	aux_q_init(ctx);
+
+	return 0;
+
+fail:
+	rpivid_h265_stop(ctx);
+	return -ENOMEM;
+}
+
+static void rpivid_h265_trigger(struct rpivid_ctx *ctx)
+{
+	struct rpivid_dev *const dev = ctx->dev;
+	struct rpivid_dec_env *const de = ctx->dec0;
+
+	xtrace_in(dev, de);
+
+	switch (!de ? RPIVID_DECODE_ERROR_CONTINUE : de->state) {
+	case RPIVID_DECODE_SLICE_START:
+		de->state = RPIVID_DECODE_SLICE_CONTINUE;
+	/* FALLTHRU */
+	case RPIVID_DECODE_SLICE_CONTINUE:
+		v4l2_m2m_buf_done_and_job_finish(dev->m2m_dev, ctx->fh.m2m_ctx,
+						 VB2_BUF_STATE_DONE);
+		break;
+	default:
+		v4l2_err(&dev->v4l2_dev, "%s: Unexpected state: %d\n", __func__,
+			 de->state);
+	/* FALLTHRU */
+	case RPIVID_DECODE_ERROR_DONE:
+		ctx->dec0 = NULL;
+		dec_env_delete(de);
+	/* FALLTHRU */
+	case RPIVID_DECODE_ERROR_CONTINUE:
+		xtrace_fin(dev, de);
+		v4l2_m2m_buf_done_and_job_finish(dev->m2m_dev, ctx->fh.m2m_ctx,
+						 VB2_BUF_STATE_ERROR);
+		break;
+	case RPIVID_DECODE_PHASE1:
+		ctx->dec0 = NULL;
+		rpivid_hw_irq_active1_claim(dev, &de->irq_ent, phase1_claimed,
+					    de);
+		break;
+	}
+
+	xtrace_ok(dev, de);
+}
+
+struct rpivid_dec_ops rpivid_dec_ops_h265 = {
+	.setup = rpivid_h265_setup,
+	.start = rpivid_h265_start,
+	.stop = rpivid_h265_stop,
+	.trigger = rpivid_h265_trigger,
+};

drivers/staging/media/rpivid/rpivid_hw.c

+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Raspberry Pi HEVC driver
+ *
+ * Copyright (C) 2020 Raspberry Pi (Trading) Ltd
+ *
+ * Based on the Cedrus VPU driver, that is:
+ *
+ * Copyright (C) 2016 Florent Revest <florent.revest@free-electrons.com>
+ * Copyright (C) 2018 Paul Kocialkowski <paul.kocialkowski@bootlin.com>
+ * Copyright (C) 2018 Bootlin
+ */
+#include <linux/clk.h>
+#include <linux/component.h>
+#include <linux/dma-mapping.h>
+#include <linux/interrupt.h>
+#include <linux/io.h>
+#include <linux/of_reserved_mem.h>
+#include <linux/of_device.h>
+#include <linux/of_platform.h>
+#include <linux/platform_device.h>
+#include <linux/regmap.h>
+#include <linux/reset.h>
+
+#include <media/videobuf2-core.h>
+#include <media/v4l2-mem2mem.h>
+
+#include "rpivid.h"
+#include "rpivid_hw.h"
+
+static void pre_irq(struct rpivid_dev *dev, struct rpivid_hw_irq_ent *ient,
+		    rpivid_irq_callback cb, void *v,
+		    struct rpivid_hw_irq_ctrl *ictl)
+{
+	unsigned long flags;
+
+	if (ictl->irq) {
+		v4l2_err(&dev->v4l2_dev, "Attempt to claim IRQ when already claimed\n");
+		return;
+	}
+
+	ient->cb = cb;
+	ient->v = v;
+
+	// Not sure this lock is actually required
+	spin_lock_irqsave(&ictl->lock, flags);
+	ictl->irq = ient;
+	spin_unlock_irqrestore(&ictl->lock, flags);
+}
+
+static void sched_claim(struct rpivid_dev * const dev,
+			struct rpivid_hw_irq_ctrl * const ictl)
+{
+	for (;;) {
+		struct rpivid_hw_irq_ent *ient = NULL;
+		unsigned long flags;
+
+		spin_lock_irqsave(&ictl->lock, flags);
+
+		if (--ictl->no_sched <= 0) {
+			ient = ictl->claim;
+			if (!ictl->irq && ient) {
+				ictl->claim = ient->next;
+				ictl->no_sched = 1;
+			}
+		}
+
+		spin_unlock_irqrestore(&ictl->lock, flags);
+
+		if (!ient)
+			break;
+
+		ient->cb(dev, ient->v);
+	}
+}
+
+/* Should only ever be called from its own IRQ cb so no lock required */
+static void pre_thread(struct rpivid_dev *dev,
+		       struct rpivid_hw_irq_ent *ient,
+		       rpivid_irq_callback cb, void *v,
+		       struct rpivid_hw_irq_ctrl *ictl)
+{
+	ient->cb = cb;
+	ient->v = v;
+	ictl->irq = ient;
+	ictl->thread_reqed = true;
+	ictl->no_sched++;
+}
+
+// Called in irq context
+static void do_irq(struct rpivid_dev * const dev,
+		   struct rpivid_hw_irq_ctrl * const ictl)
+{
+	struct rpivid_hw_irq_ent *ient;
+	unsigned long flags;
+
+	spin_lock_irqsave(&ictl->lock, flags);
+	ient = ictl->irq;
+	if (ient) {
+		ictl->no_sched++;
+		ictl->irq = NULL;
+	}
+	spin_unlock_irqrestore(&ictl->lock, flags);
+
+	if (ient) {
+		ient->cb(dev, ient->v);
+
+		sched_claim(dev, ictl);
+	}
+}
+
+static void do_claim(struct rpivid_dev * const dev,
+		     struct rpivid_hw_irq_ent *ient,
+		     const rpivid_irq_callback cb, void * const v,
+		     struct rpivid_hw_irq_ctrl * const ictl)
+{
+	unsigned long flags;
+
+	ient->next = NULL;
+	ient->cb = cb;
+	ient->v = v;
+
+	spin_lock_irqsave(&ictl->lock, flags);
+
+	if (ictl->claim) {
+		// If we have a Q then add to end
+		ictl->tail->next = ient;
+		ictl->tail = ient;
+		ient = NULL;
+	} else if (ictl->no_sched || ictl->irq) {
+		// Empty Q but other activity in progress so Q
+		ictl->claim = ient;
+		ictl->tail = ient;
+		ient = NULL;
+	} else {
+		// Nothing else going on - schedule immediately and
+		// prevent anything else scheduling claims
+		ictl->no_sched = 1;
+	}
+
+	spin_unlock_irqrestore(&ictl->lock, flags);
+
+	if (ient) {
+		ient->cb(dev, ient->v);
+
+		sched_claim(dev, ictl);
+	}
+}
+
+static void ictl_init(struct rpivid_hw_irq_ctrl * const ictl)
+{
+	spin_lock_init(&ictl->lock);
+	ictl->claim = NULL;
+	ictl->tail = NULL;
+	ictl->irq = NULL;
+	ictl->no_sched = 0;
+}
+
+static void ictl_uninit(struct rpivid_hw_irq_ctrl * const ictl)
+{
+	// Nothing to do
+}
+
+#if !OPT_DEBUG_POLL_IRQ
+static irqreturn_t rpivid_irq_irq(int irq, void *data)
+{
+	struct rpivid_dev * const dev = data;
+	__u32 ictrl;
+
+	ictrl = irq_read(dev, ARG_IC_ICTRL);
+	if (!(ictrl & ARG_IC_ICTRL_ALL_IRQ_MASK)) {
+		v4l2_warn(&dev->v4l2_dev, "IRQ but no IRQ bits set\n");
+		return IRQ_NONE;
+	}
+
+	// Cancel any/all irqs
+	irq_write(dev, ARG_IC_ICTRL, ictrl & ~ARG_IC_ICTRL_SET_ZERO_MASK);
+
+	// Service Active2 before Active1 so Phase 1 can transition to Phase 2
+	// without delay
+	if (ictrl & ARG_IC_ICTRL_ACTIVE2_INT_SET)
+		do_irq(dev, &dev->ic_active2);
+	if (ictrl & ARG_IC_ICTRL_ACTIVE1_INT_SET)
+		do_irq(dev, &dev->ic_active1);
+
+	return dev->ic_active1.thread_reqed || dev->ic_active2.thread_reqed ?
+		IRQ_WAKE_THREAD : IRQ_HANDLED;
+}
+
+static void do_thread(struct rpivid_dev * const dev,
+		      struct rpivid_hw_irq_ctrl *const ictl)
+{
+	unsigned long flags;
+	struct rpivid_hw_irq_ent *ient = NULL;
+
+	spin_lock_irqsave(&ictl->lock, flags);
+
+	if (ictl->thread_reqed) {
+		ient = ictl->irq;
+		ictl->thread_reqed = false;
+		ictl->irq = NULL;
+	}
+
+	spin_unlock_irqrestore(&ictl->lock, flags);
+
+	if (ient) {
+		ient->cb(dev, ient->v);
+
+		sched_claim(dev, ictl);
+	}
+}
+
+static irqreturn_t rpivid_irq_thread(int irq, void *data)
+{
+	struct rpivid_dev * const dev = data;
+
+	do_thread(dev, &dev->ic_active1);
+	do_thread(dev, &dev->ic_active2);
+
+	return IRQ_HANDLED;
+}
+#endif
+
+/* May only be called from Active1 CB
+ * IRQs should not be expected until execution continues in the cb
+ */
+void rpivid_hw_irq_active1_thread(struct rpivid_dev *dev,
+				  struct rpivid_hw_irq_ent *ient,
+				  rpivid_irq_callback thread_cb, void *ctx)
+{
+	pre_thread(dev, ient, thread_cb, ctx, &dev->ic_active1);
+}
+
+void rpivid_hw_irq_active1_claim(struct rpivid_dev *dev,
+				 struct rpivid_hw_irq_ent *ient,
+				 rpivid_irq_callback ready_cb, void *ctx)
+{
+	do_claim(dev, ient, ready_cb, ctx, &dev->ic_active1);
+}
+
+void rpivid_hw_irq_active1_irq(struct rpivid_dev *dev,
+			       struct rpivid_hw_irq_ent *ient,
+			       rpivid_irq_callback irq_cb, void *ctx)
+{
+	pre_irq(dev, ient, irq_cb, ctx, &dev->ic_active1);
+}
+
+void rpivid_hw_irq_active2_claim(struct rpivid_dev *dev,
+				 struct rpivid_hw_irq_ent *ient,
+				 rpivid_irq_callback ready_cb, void *ctx)
+{
+	do_claim(dev, ient, ready_cb, ctx, &dev->ic_active2);
+}
+
+void rpivid_hw_irq_active2_irq(struct rpivid_dev *dev,
+			       struct rpivid_hw_irq_ent *ient,
+			       rpivid_irq_callback irq_cb, void *ctx)
+{
+	pre_irq(dev, ient, irq_cb, ctx, &dev->ic_active2);
+}
+
+int rpivid_hw_probe(struct rpivid_dev *dev)
+{
+	struct resource *res;
+	__u32 irq_stat;
+	int irq_dec;
+	int ret = 0;
+
+	ictl_init(&dev->ic_active1);
+	ictl_init(&dev->ic_active2);
+
+	res = platform_get_resource_byname(dev->pdev, IORESOURCE_MEM, "intc");
+	if (!res)
+		return -ENODEV;
+
+	dev->base_irq = devm_ioremap(dev->dev, res->start, resource_size(res));
+	if (IS_ERR(dev->base_irq))
+		return PTR_ERR(dev->base_irq);
+
+	res = platform_get_resource_byname(dev->pdev, IORESOURCE_MEM, "hevc");
+	if (!res)
+		return -ENODEV;
+
+	dev->base_h265 = devm_ioremap(dev->dev, res->start, resource_size(res));
+	if (IS_ERR(dev->base_h265))
+		return PTR_ERR(dev->base_h265);
+
+	dev->clock = devm_clk_get(&dev->pdev->dev, "hevc");
+	if (IS_ERR(dev->clock))
+		return PTR_ERR(dev->clock);
+
+	// Disable IRQs & reset anything pending
+	irq_write(dev, 0,
+		  ARG_IC_ICTRL_ACTIVE1_EN_SET | ARG_IC_ICTRL_ACTIVE2_EN_SET);
+	irq_stat = irq_read(dev, 0);
+	irq_write(dev, 0, irq_stat);
+
+#if !OPT_DEBUG_POLL_IRQ
+	irq_dec = platform_get_irq(dev->pdev, 0);
+	if (irq_dec <= 0)
+		return irq_dec;
+	ret = devm_request_threaded_irq(dev->dev, irq_dec,
+					rpivid_irq_irq,
+					rpivid_irq_thread,
+					0, dev_name(dev->dev), dev);
+	if (ret) {
+		dev_err(dev->dev, "Failed to request IRQ - %d\n", ret);
+
+		return ret;
+	}
+#endif
+	return ret;
+}
+
+void rpivid_hw_remove(struct rpivid_dev *dev)
+{
+	// IRQ auto freed on unload so no need to do it here
+	ictl_uninit(&dev->ic_active1);
+	ictl_uninit(&dev->ic_active2);
+}
+

drivers/staging/media/rpivid/rpivid_hw.h

+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * Raspberry Pi HEVC driver
+ *
+ * Copyright (C) 2020 Raspberry Pi (Trading) Ltd
+ *
+ * Based on the Cedrus VPU driver, that is:
+ *
+ * Copyright (C) 2016 Florent Revest <florent.revest@free-electrons.com>
+ * Copyright (C) 2018 Paul Kocialkowski <paul.kocialkowski@bootlin.com>
+ * Copyright (C) 2018 Bootlin
+ */
+
+#ifndef _RPIVID_HW_H_
+#define _RPIVID_HW_H_
+
+struct rpivid_hw_irq_ent {
+	struct rpivid_hw_irq_ent *next;
+	rpivid_irq_callback cb;
+	void *v;
+};
+
+/* Phase 1 Register offsets */
+
+#define RPI_SPS0 0
+#define RPI_SPS1 4
+#define RPI_PPS 8
+#define RPI_SLICE 12
+#define RPI_TILESTART 16
+#define RPI_TILEEND 20
+#define RPI_SLICESTART 24
+#define RPI_MODE 28
+#define RPI_LEFT0 32
+#define RPI_LEFT1 36
+#define RPI_LEFT2 40
+#define RPI_LEFT3 44
+#define RPI_QP 48
+#define RPI_CONTROL 52
+#define RPI_STATUS 56
+#define RPI_VERSION 60
+#define RPI_BFBASE 64
+#define RPI_BFNUM 68
+#define RPI_BFCONTROL 72
+#define RPI_BFSTATUS 76
+#define RPI_PUWBASE 80
+#define RPI_PUWSTRIDE 84
+#define RPI_COEFFWBASE 88
+#define RPI_COEFFWSTRIDE 92
+#define RPI_SLICECMDS 96
+#define RPI_BEGINTILEEND 100
+#define RPI_TRANSFER 104
+#define RPI_CFBASE 108
+#define RPI_CFNUM 112
+#define RPI_CFSTATUS 116
+
+/* Phase 2 Register offsets */
+
+#define RPI_PURBASE 0x8000
+#define RPI_PURSTRIDE 0x8004
+#define RPI_COEFFRBASE 0x8008
+#define RPI_COEFFRSTRIDE 0x800C
+#define RPI_NUMROWS 0x8010
+#define RPI_CONFIG2 0x8014
+#define RPI_OUTYBASE 0x8018
+#define RPI_OUTYSTRIDE 0x801C
+#define RPI_OUTCBASE 0x8020
+#define RPI_OUTCSTRIDE 0x8024
+#define RPI_STATUS2 0x8028
+#define RPI_FRAMESIZE 0x802C
+#define RPI_MVBASE 0x8030
+#define RPI_MVSTRIDE 0x8034
+#define RPI_COLBASE 0x8038
+#define RPI_COLSTRIDE 0x803C
+#define RPI_CURRPOC 0x8040
+
+/*
+ * Write a general register value
+ * Order is unimportant
+ */
+static inline void apb_write(const struct rpivid_dev * const dev,
+			     const unsigned int offset, const u32 val)
+{
+	writel_relaxed(val, dev->base_h265 + offset);
+}
+
+/* Write the final register value that actually starts the phase */
+static inline void apb_write_final(const struct rpivid_dev * const dev,
+				   const unsigned int offset, const u32 val)
+{
+	writel(val, dev->base_h265 + offset);
+}
+
+static inline u32 apb_read(const struct rpivid_dev * const dev,
+			   const unsigned int offset)
+{
+	return readl(dev->base_h265 + offset);
+}
+
+static inline void irq_write(const struct rpivid_dev * const dev,
+			     const unsigned int offset, const u32 val)
+{
+	writel(val, dev->base_irq + offset);
+}
+
+static inline u32 irq_read(const struct rpivid_dev * const dev,
+			   const unsigned int offset)
+{
+	return readl(dev->base_irq + offset);
+}
+
+static inline void apb_write_vc_addr(const struct rpivid_dev * const dev,
+				     const unsigned int offset,
+				     const dma_addr_t a)
+{
+	apb_write(dev, offset, (u32)(a >> 6));
+}
+
+static inline void apb_write_vc_addr_final(const struct rpivid_dev * const dev,
+					   const unsigned int offset,
+					   const dma_addr_t a)
+{
+	apb_write_final(dev, offset, (u32)(a >> 6));
+}
+
+static inline void apb_write_vc_len(const struct rpivid_dev * const dev,
+				    const unsigned int offset,
+				    const unsigned int x)
+{
+	apb_write(dev, offset, (x + 63) >> 6);
+}
+
+/* *ARG_IC_ICTRL - Interrupt control for ARGON Core*
+ * Offset (byte space) = 40'h2b10000
+ * Physical Address (byte space) = 40'h7eb10000
+ * Verilog Macro Address = `ARG_IC_REG_START + `ARGON_INTCTRL_ICTRL
+ * Reset Value = 32'b100x100x_100xxxxx_xxxxxxx0_x100x100
+ * Access = RW (32-bit only)
+ * Interrupt control logic for ARGON Core.
+ */
+#define ARG_IC_ICTRL 0
+
+/* acc=LWC ACTIVE1_INT FIELD ACCESS: LWC
+ *
+ * Interrupt 1
+ * This is set and held when an hevc_active1 interrupt edge is detected
+ * The polarity of the edge is set by the ACTIVE1_EDGE field
+ * Write a 1 to this bit to clear down the latched interrupt
+ * The latched interrupt is only enabled out onto the interrupt line if
+ * ACTIVE1_EN is set
+ * Reset value is *0* decimal.
+ */
+#define ARG_IC_ICTRL_ACTIVE1_INT_SET		BIT(0)
+
+/* ACTIVE1_EDGE Sets the polarity of the interrupt edge detection logic
+ * This logic detects edges of the hevc_active1 line from the argon core
+ * 0 = negedge, 1 = posedge
+ * Reset value is *0* decimal.
+ */
+#define ARG_IC_ICTRL_ACTIVE1_EDGE_SET		BIT(1)
+
+/* ACTIVE1_EN Enables ACTIVE1_INT out onto the argon interrupt line.
+ * If this isn't set, the interrupt logic will work but no interrupt will be
+ * set to the interrupt controller
+ * Reset value is *1* decimal.
+ *
+ * [JC] The above appears to be a lie - if unset then b0 is never set
+ */
+#define ARG_IC_ICTRL_ACTIVE1_EN_SET		BIT(2)
+
+/* acc=RO ACTIVE1_STATUS FIELD ACCESS: RO
+ *
+ * The current status of the hevc_active1 signal
+ */
+#define ARG_IC_ICTRL_ACTIVE1_STATUS_SET		BIT(3)
+
+/* acc=LWC ACTIVE2_INT FIELD ACCESS: LWC
+ *
+ * Interrupt 2
+ * This is set and held when an hevc_active2 interrupt edge is detected
+ * The polarity of the edge is set by the ACTIVE2_EDGE field
+ * Write a 1 to this bit to clear down the latched interrupt
+ * The latched interrupt is only enabled out onto the interrupt line if
+ * ACTIVE2_EN is set
+ * Reset value is *0* decimal.
+ */
+#define ARG_IC_ICTRL_ACTIVE2_INT_SET		BIT(4)
+
+/* ACTIVE2_EDGE Sets the polarity of the interrupt edge detection logic
+ * This logic detects edges of the hevc_active2 line from the argon core
+ * 0 = negedge, 1 = posedge
+ * Reset value is *0* decimal.
+ */
+#define ARG_IC_ICTRL_ACTIVE2_EDGE_SET		BIT(5)
+
+/* ACTIVE2_EN Enables ACTIVE2_INT out onto the argon interrupt line.
+ * If this isn't set, the interrupt logic will work but no interrupt will be
+ * set to the interrupt controller
+ * Reset value is *1* decimal.
+ */
+#define ARG_IC_ICTRL_ACTIVE2_EN_SET		BIT(6)
+
+/* acc=RO ACTIVE2_STATUS FIELD ACCESS: RO
+ *
+ * The current status of the hevc_active2 signal
+ */
+#define ARG_IC_ICTRL_ACTIVE2_STATUS_SET		BIT(7)
+
+/* TEST_INT Forces the argon int high for test purposes.
+ * Reset value is *0* decimal.
+ */
+#define ARG_IC_ICTRL_TEST_INT			BIT(8)
+#define ARG_IC_ICTRL_SPARE			BIT(9)
+
+/* acc=RO VP9_INTERRUPT_STATUS FIELD ACCESS: RO
+ *
+ * The current status of the vp9_interrupt signal
+ */
+#define ARG_IC_ICTRL_VP9_INTERRUPT_STATUS	BIT(10)
+
+/* AIO_INT_ENABLE 1 = Or the AIO int in with the Argon int so the VPU can see
+ * it
+ * 0 = the AIO int is masked. (It should still be connected to the GIC though).
+ */
+#define ARG_IC_ICTRL_AIO_INT_ENABLE		BIT(20)
+#define ARG_IC_ICTRL_H264_ACTIVE_INT		BIT(21)
+#define ARG_IC_ICTRL_H264_ACTIVE_EDGE		BIT(22)
+#define ARG_IC_ICTRL_H264_ACTIVE_EN		BIT(23)
+#define ARG_IC_ICTRL_H264_ACTIVE_STATUS		BIT(24)
+#define ARG_IC_ICTRL_H264_INTERRUPT_INT		BIT(25)
+#define ARG_IC_ICTRL_H264_INTERRUPT_EDGE	BIT(26)
+#define ARG_IC_ICTRL_H264_INTERRUPT_EN		BIT(27)
+
+/* acc=RO H264_INTERRUPT_STATUS FIELD ACCESS: RO
+ *
+ * The current status of the h264_interrupt signal
+ */
+#define ARG_IC_ICTRL_H264_INTERRUPT_STATUS	BIT(28)
+
+/* acc=LWC VP9_INTERRUPT_INT FIELD ACCESS: LWC
+ *
+ * Interrupt 1
+ * This is set and held when an vp9_interrupt interrupt edge is detected
+ * The polarity of the edge is set by the VP9_INTERRUPT_EDGE field
+ * Write a 1 to this bit to clear down the latched interrupt
+ * The latched interrupt is only enabled out onto the interrupt line if
+ * VP9_INTERRUPT_EN is set
+ * Reset value is *0* decimal.
+ */
+#define ARG_IC_ICTRL_VP9_INTERRUPT_INT		BIT(29)
+
+/* VP9_INTERRUPT_EDGE Sets the polarity of the interrupt edge detection logic
+ * This logic detects edges of the vp9_interrupt line from the argon h264 core
+ * 0 = negedge, 1 = posedge
+ * Reset value is *0* decimal.
+ */
+#define ARG_IC_ICTRL_VP9_INTERRUPT_EDGE		BIT(30)
+
+/* VP9_INTERRUPT_EN Enables VP9_INTERRUPT_INT out onto the argon interrupt line.
+ * If this isn't set, the interrupt logic will work but no interrupt will be
+ * set to the interrupt controller
+ * Reset value is *1* decimal.
+ */
+#define ARG_IC_ICTRL_VP9_INTERRUPT_EN		BIT(31)
+
+/* Bits 19:12, 11 reserved - read ?, write 0 */
+#define ARG_IC_ICTRL_SET_ZERO_MASK		((0xff << 12) | BIT(11))
+
+/* All IRQ bits */
+#define ARG_IC_ICTRL_ALL_IRQ_MASK   (\
+		ARG_IC_ICTRL_VP9_INTERRUPT_INT  |\
+		ARG_IC_ICTRL_H264_INTERRUPT_INT |\
+		ARG_IC_ICTRL_ACTIVE1_INT_SET    |\
+		ARG_IC_ICTRL_ACTIVE2_INT_SET)
+
+/* Auto release once all CBs called */
+void rpivid_hw_irq_active1_claim(struct rpivid_dev *dev,
+				 struct rpivid_hw_irq_ent *ient,
+				 rpivid_irq_callback ready_cb, void *ctx);
+/* May only be called in claim cb */
+void rpivid_hw_irq_active1_irq(struct rpivid_dev *dev,
+			       struct rpivid_hw_irq_ent *ient,
+			       rpivid_irq_callback irq_cb, void *ctx);
+/* May only be called in irq cb */
+void rpivid_hw_irq_active1_thread(struct rpivid_dev *dev,
+				  struct rpivid_hw_irq_ent *ient,
+				  rpivid_irq_callback thread_cb, void *ctx);
+
+/* Auto release once all CBs called */
+void rpivid_hw_irq_active2_claim(struct rpivid_dev *dev,
+				 struct rpivid_hw_irq_ent *ient,
+				 rpivid_irq_callback ready_cb, void *ctx);
+/* May only be called in claim cb */
+void rpivid_hw_irq_active2_irq(struct rpivid_dev *dev,
+			       struct rpivid_hw_irq_ent *ient,
+			       rpivid_irq_callback irq_cb, void *ctx);
+
+int rpivid_hw_probe(struct rpivid_dev *dev);
+void rpivid_hw_remove(struct rpivid_dev *dev);
+
+#endif

drivers/staging/media/rpivid/rpivid_video.c

+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Raspberry Pi HEVC driver
+ *
+ * Copyright (C) 2020 Raspberry Pi (Trading) Ltd
+ *
+ * Based on the Cedrus VPU driver, that is:
+ *
+ * Copyright (C) 2016 Florent Revest <florent.revest@free-electrons.com>
+ * Copyright (C) 2018 Paul Kocialkowski <paul.kocialkowski@bootlin.com>
+ * Copyright (C) 2018 Bootlin
+ */
+
+#include <media/videobuf2-dma-contig.h>
+#include <media/v4l2-device.h>
+#include <media/v4l2-ioctl.h>
+#include <media/v4l2-event.h>
+#include <media/v4l2-mem2mem.h>
+
+#include "rpivid.h"
+#include "rpivid_video.h"
+#include "rpivid_dec.h"
+
+#define RPIVID_DECODE_SRC	BIT(0)
+#define RPIVID_DECODE_DST	BIT(1)
+
+#define RPIVID_MIN_WIDTH	16U
+#define RPIVID_MIN_HEIGHT	16U
+#define RPIVID_MAX_WIDTH	4096U
+#define RPIVID_MAX_HEIGHT	4096U
+
+static inline struct rpivid_ctx *rpivid_file2ctx(struct file *file)
+{
+	return container_of(file->private_data, struct rpivid_ctx, fh);
+}
+
+/* constrain x to y,y*2 */
+static inline unsigned int constrain2x(unsigned int x, unsigned int y)
+{
+	return (x < y) ?
+			y :
+			(x > y * 2) ? y : x;
+}
+
+int rpivid_prepare_src_format(struct v4l2_pix_format *pix_fmt)
+{
+	if (pix_fmt->pixelformat != V4L2_PIX_FMT_HEVC_SLICE)
+		return -EINVAL;
+
+	/* Zero bytes per line for encoded source. */
+	pix_fmt->bytesperline = 0;
+	/* Choose some minimum size since this can't be 0 */
+	pix_fmt->sizeimage = max_t(u32, SZ_1K, pix_fmt->sizeimage);
+	pix_fmt->field = V4L2_FIELD_NONE;
+	return 0;
+}
+
+int rpivid_prepare_dst_format(struct v4l2_pix_format *pix_fmt)
+{
+	unsigned int width = pix_fmt->width;
+	unsigned int height = pix_fmt->height;
+	unsigned int sizeimage = pix_fmt->sizeimage;
+	unsigned int bytesperline = pix_fmt->bytesperline;
+
+	switch (pix_fmt->pixelformat) {
+	/* For column formats set bytesperline to column height (stride2) */
+	case V4L2_PIX_FMT_NV12_COL128:
+		/* Width rounds up to columns */
+		width = ALIGN(min(width, RPIVID_MAX_WIDTH), 128);
+
+		/* 16 aligned height - not sure we even need that */
+		height = ALIGN(height, 16);
+		/* column height
+		 * Accept suggested shape if at least min & < 2 * min
+		 */
+		bytesperline = constrain2x(bytesperline, height * 3 / 2);
+
+		/* image size
+		 * Again allow plausible variation in case added padding is
+		 * required
+		 */
+		sizeimage = constrain2x(sizeimage, bytesperline * width);
+		break;
+
+	case V4L2_PIX_FMT_NV12_10_COL128:
+		/* width in pixels (3 pels = 4 bytes) rounded to 128 byte
+		 * columns
+		 */
+		width = ALIGN(((min(width, RPIVID_MAX_WIDTH) + 2) / 3), 32) * 3;
+
+		/* 16-aligned height. */
+		height = ALIGN(height, 16);
+
+		/* column height
+		 * Accept suggested shape if at least min & < 2 * min
+		 */
+		bytesperline = constrain2x(bytesperline, height * 3 / 2);
+
+		/* image size
+		 * Again allow plausible variation in case added padding is
+		 * required
+		 */
+		sizeimage = constrain2x(sizeimage,
+					bytesperline * width * 4 / 3);
+		break;
+
+	default:
+		return -EINVAL;
+	}
+
+	pix_fmt->width = width;
+	pix_fmt->height = height;
+
+	pix_fmt->field = V4L2_FIELD_NONE;
+	pix_fmt->bytesperline = bytesperline;
+	pix_fmt->sizeimage = sizeimage;
+	return 0;
+}
+
+static int rpivid_querycap(struct file *file, void *priv,
+			   struct v4l2_capability *cap)
+{
+	strscpy(cap->driver, RPIVID_NAME, sizeof(cap->driver));
+	strscpy(cap->card, RPIVID_NAME, sizeof(cap->card));
+	snprintf(cap->bus_info, sizeof(cap->bus_info),
+		 "platform:%s", RPIVID_NAME);
+
+	return 0;
+}
+
+static int rpivid_enum_fmt_vid_out(struct file *file, void *priv,
+				   struct v4l2_fmtdesc *f)
+{
+	// Input formats
+
+	// H.265 Slice only currently
+	if (f->index == 0) {
+		f->pixelformat = V4L2_PIX_FMT_HEVC_SLICE;
+		return 0;
+	}
+
+	return -EINVAL;
+}
+
+static int rpivid_hevc_validate_sps(const struct v4l2_ctrl_hevc_sps * const sps)
+{
+	const unsigned int ctb_log2_size_y =
+			sps->log2_min_luma_coding_block_size_minus3 + 3 +
+			sps->log2_diff_max_min_luma_coding_block_size;
+	const unsigned int min_tb_log2_size_y =
+			sps->log2_min_luma_transform_block_size_minus2 + 2;
+	const unsigned int max_tb_log2_size_y = min_tb_log2_size_y +
+			sps->log2_diff_max_min_luma_transform_block_size;
+
+	/* Local limitations */
+	if (sps->pic_width_in_luma_samples < 32 ||
+	    sps->pic_width_in_luma_samples > 4096)
+		return 0;
+	if (sps->pic_height_in_luma_samples < 32 ||
+	    sps->pic_height_in_luma_samples > 4096)
+		return 0;
+	if (!(sps->bit_depth_luma_minus8 == 0 ||
+	      sps->bit_depth_luma_minus8 == 2))
+		return 0;
+	if (sps->bit_depth_luma_minus8 != sps->bit_depth_chroma_minus8)
+		return 0;
+	if (sps->chroma_format_idc != 1)
+		return 0;
+
+	/*  Limits from H.265 7.4.3.2.1 */
+	if (sps->log2_max_pic_order_cnt_lsb_minus4 > 12)
+		return 0;
+	if (sps->sps_max_dec_pic_buffering_minus1 > 15)
+		return 0;
+	if (sps->sps_max_num_reorder_pics >
+				sps->sps_max_dec_pic_buffering_minus1)
+		return 0;
+	if (ctb_log2_size_y > 6)
+		return 0;
+	if (max_tb_log2_size_y > 5)
+		return 0;
+	if (max_tb_log2_size_y > ctb_log2_size_y)
+		return 0;
+	if (sps->max_transform_hierarchy_depth_inter >
+				(ctb_log2_size_y - min_tb_log2_size_y))
+		return 0;
+	if (sps->max_transform_hierarchy_depth_intra >
+				(ctb_log2_size_y - min_tb_log2_size_y))
+		return 0;
+	/* Check pcm stuff */
+	if (sps->num_short_term_ref_pic_sets > 64)
+		return 0;
+	if (sps->num_long_term_ref_pics_sps > 32)
+		return 0;
+	return 1;
+}
+
+static inline int is_sps_set(const struct v4l2_ctrl_hevc_sps * const sps)
+{
+	return sps && sps->pic_width_in_luma_samples != 0;
+}
+
+static u32 pixelformat_from_sps(const struct v4l2_ctrl_hevc_sps * const sps,
+				const int index)
+{
+	u32 pf = 0;
+
+	// Use width 0 as a signifier of unsetness
+	if (!is_sps_set(sps)) {
+		/* Treat this as an error? For now return both */
+		if (index == 0)
+			pf = V4L2_PIX_FMT_NV12_COL128;
+		else if (index == 1)
+			pf = V4L2_PIX_FMT_NV12_10_COL128;
+	} else if (index == 0 && rpivid_hevc_validate_sps(sps)) {
+		if (sps->bit_depth_luma_minus8 == 0)
+			pf = V4L2_PIX_FMT_NV12_COL128;
+		else if (sps->bit_depth_luma_minus8 == 2)
+			pf = V4L2_PIX_FMT_NV12_10_COL128;
+	}
+
+	return pf;
+}
+
+static struct v4l2_pix_format
+rpivid_hevc_default_dst_fmt(struct rpivid_ctx * const ctx)
+{
+	const struct v4l2_ctrl_hevc_sps * const sps =
+		rpivid_find_control_data(ctx, V4L2_CID_MPEG_VIDEO_HEVC_SPS);
+	struct v4l2_pix_format pix_fmt = {
+		.width = sps->pic_width_in_luma_samples,
+		.height = sps->pic_height_in_luma_samples,
+		.pixelformat = pixelformat_from_sps(sps, 0)
+	};
+
+	rpivid_prepare_dst_format(&pix_fmt);
+	return pix_fmt;
+}
+
+static u32 rpivid_hevc_get_dst_pixelformat(struct rpivid_ctx * const ctx,
+					   const int index)
+{
+	const struct v4l2_ctrl_hevc_sps * const sps =
+		rpivid_find_control_data(ctx, V4L2_CID_MPEG_VIDEO_HEVC_SPS);
+
+	return pixelformat_from_sps(sps, index);
+}
+
+static int rpivid_enum_fmt_vid_cap(struct file *file, void *priv,
+				   struct v4l2_fmtdesc *f)
+{
+	struct rpivid_ctx * const ctx = rpivid_file2ctx(file);
+
+	const u32 pf = rpivid_hevc_get_dst_pixelformat(ctx, f->index);
+
+	if (pf == 0)
+		return -EINVAL;
+
+	f->pixelformat = pf;
+	return 0;
+}
+
+static int rpivid_g_fmt_vid_cap(struct file *file, void *priv,
+				struct v4l2_format *f)
+{
+	struct rpivid_ctx *ctx = rpivid_file2ctx(file);
+
+	if (!ctx->dst_fmt_set)
+		ctx->dst_fmt = rpivid_hevc_default_dst_fmt(ctx);
+	f->fmt.pix = ctx->dst_fmt;
+	return 0;
+}
+
+static int rpivid_g_fmt_vid_out(struct file *file, void *priv,
+				struct v4l2_format *f)
+{
+	struct rpivid_ctx *ctx = rpivid_file2ctx(file);
+
+	f->fmt.pix = ctx->src_fmt;
+	return 0;
+}
+
+static inline void copy_color(struct v4l2_pix_format *d,
+			      const struct v4l2_pix_format *s)
+{
+	d->colorspace   = s->colorspace;
+	d->xfer_func    = s->xfer_func;
+	d->ycbcr_enc    = s->ycbcr_enc;
+	d->quantization = s->quantization;
+}
+
+static int rpivid_try_fmt_vid_cap(struct file *file, void *priv,
+				  struct v4l2_format *f)
+{
+	struct rpivid_ctx *ctx = rpivid_file2ctx(file);
+	const struct v4l2_ctrl_hevc_sps * const sps =
+		rpivid_find_control_data(ctx, V4L2_CID_MPEG_VIDEO_HEVC_SPS);
+	u32 pixelformat;
+	int i;
+
+	/* Reject format types we don't support */
+	if (f->type != V4L2_BUF_TYPE_VIDEO_CAPTURE)
+		return -EINVAL;
+
+	for (i = 0; (pixelformat = pixelformat_from_sps(sps, i)) != 0; i++) {
+		if (f->fmt.pix.pixelformat == pixelformat)
+			break;
+	}
+
+	// If we can't use requested fmt then set to default
+	if (pixelformat == 0) {
+		pixelformat = pixelformat_from_sps(sps, 0);
+		// If we don't have a default then give up
+		if (pixelformat == 0)
+			return -EINVAL;
+	}
+
+	// We don't have any way of finding out colourspace so believe
+	// anything we are told - take anything set in src as a default
+	if (f->fmt.pix.colorspace == V4L2_COLORSPACE_DEFAULT)
+		copy_color(&f->fmt.pix, &ctx->src_fmt);
+
+	f->fmt.pix.pixelformat = pixelformat;
+	return rpivid_prepare_dst_format(&f->fmt.pix);
+}
+
+static int rpivid_try_fmt_vid_out(struct file *file, void *priv,
+				  struct v4l2_format *f)
+{
+	if (f->type != V4L2_BUF_TYPE_VIDEO_OUTPUT)
+		return -EINVAL;
+
+	if (rpivid_prepare_src_format(&f->fmt.pix)) {
+		// Set default src format
+		f->fmt.pix.pixelformat = RPIVID_SRC_PIXELFORMAT_DEFAULT;
+		rpivid_prepare_src_format(&f->fmt.pix);
+	}
+	return 0;
+}
+
+static int rpivid_s_fmt_vid_cap(struct file *file, void *priv,
+				struct v4l2_format *f)
+{
+	struct rpivid_ctx *ctx = rpivid_file2ctx(file);
+	struct vb2_queue *vq;
+	int ret;
+
+	vq = v4l2_m2m_get_vq(ctx->fh.m2m_ctx, f->type);
+	if (vb2_is_busy(vq))
+		return -EBUSY;
+
+	ret = rpivid_try_fmt_vid_cap(file, priv, f);
+	if (ret)
+		return ret;
+
+	ctx->dst_fmt = f->fmt.pix;
+	ctx->dst_fmt_set = 1;
+
+	return 0;
+}
+
+static int rpivid_s_fmt_vid_out(struct file *file, void *priv,
+				struct v4l2_format *f)
+{
+	struct rpivid_ctx *ctx = rpivid_file2ctx(file);
+	struct vb2_queue *vq;
+	int ret;
+
+	vq = v4l2_m2m_get_vq(ctx->fh.m2m_ctx, f->type);
+	if (vb2_is_busy(vq))
+		return -EBUSY;
+
+	ret = rpivid_try_fmt_vid_out(file, priv, f);
+	if (ret)
+		return ret;
+
+	ctx->src_fmt = f->fmt.pix;
+	ctx->dst_fmt_set = 0;  // Setting src invalidates dst
+
+	vq->subsystem_flags |=
+		VB2_V4L2_FL_SUPPORTS_M2M_HOLD_CAPTURE_BUF;
+
+	/* Propagate colorspace information to capture. */
+	copy_color(&ctx->dst_fmt, &f->fmt.pix);
+	return 0;
+}
+
+const struct v4l2_ioctl_ops rpivid_ioctl_ops = {
+	.vidioc_querycap		= rpivid_querycap,
+
+	.vidioc_enum_fmt_vid_cap	= rpivid_enum_fmt_vid_cap,
+	.vidioc_g_fmt_vid_cap		= rpivid_g_fmt_vid_cap,
+	.vidioc_try_fmt_vid_cap		= rpivid_try_fmt_vid_cap,
+	.vidioc_s_fmt_vid_cap		= rpivid_s_fmt_vid_cap,
+
+	.vidioc_enum_fmt_vid_out	= rpivid_enum_fmt_vid_out,
+	.vidioc_g_fmt_vid_out		= rpivid_g_fmt_vid_out,
+	.vidioc_try_fmt_vid_out		= rpivid_try_fmt_vid_out,
+	.vidioc_s_fmt_vid_out		= rpivid_s_fmt_vid_out,
+
+	.vidioc_reqbufs			= v4l2_m2m_ioctl_reqbufs,
+	.vidioc_querybuf		= v4l2_m2m_ioctl_querybuf,
+	.vidioc_qbuf			= v4l2_m2m_ioctl_qbuf,
+	.vidioc_dqbuf			= v4l2_m2m_ioctl_dqbuf,
+	.vidioc_prepare_buf		= v4l2_m2m_ioctl_prepare_buf,
+	.vidioc_create_bufs		= v4l2_m2m_ioctl_create_bufs,
+	.vidioc_expbuf			= v4l2_m2m_ioctl_expbuf,
+
+	.vidioc_streamon		= v4l2_m2m_ioctl_streamon,
+	.vidioc_streamoff		= v4l2_m2m_ioctl_streamoff,
+
+	.vidioc_try_decoder_cmd		= v4l2_m2m_ioctl_stateless_try_decoder_cmd,
+	.vidioc_decoder_cmd		= v4l2_m2m_ioctl_stateless_decoder_cmd,
+
+	.vidioc_subscribe_event		= v4l2_ctrl_subscribe_event,
+	.vidioc_unsubscribe_event	= v4l2_event_unsubscribe,
+};
+
+static int rpivid_queue_setup(struct vb2_queue *vq, unsigned int *nbufs,
+			      unsigned int *nplanes, unsigned int sizes[],
+			      struct device *alloc_devs[])
+{
+	struct rpivid_ctx *ctx = vb2_get_drv_priv(vq);
+	struct v4l2_pix_format *pix_fmt;
+
+	if (V4L2_TYPE_IS_OUTPUT(vq->type))
+		pix_fmt = &ctx->src_fmt;
+	else
+		pix_fmt = &ctx->dst_fmt;
+
+	if (*nplanes) {
+		if (sizes[0] < pix_fmt->sizeimage)
+			return -EINVAL;
+	} else {
+		sizes[0] = pix_fmt->sizeimage;
+		*nplanes = 1;
+	}
+
+	return 0;
+}
+
+static void rpivid_queue_cleanup(struct vb2_queue *vq, u32 state)
+{
+	struct rpivid_ctx *ctx = vb2_get_drv_priv(vq);
+	struct vb2_v4l2_buffer *vbuf;
+
+	for (;;) {
+		if (V4L2_TYPE_IS_OUTPUT(vq->type))
+			vbuf = v4l2_m2m_src_buf_remove(ctx->fh.m2m_ctx);
+		else
+			vbuf = v4l2_m2m_dst_buf_remove(ctx->fh.m2m_ctx);
+
+		if (!vbuf)
+			return;
+
+		v4l2_ctrl_request_complete(vbuf->vb2_buf.req_obj.req,
+					   &ctx->hdl);
+		v4l2_m2m_buf_done(vbuf, state);
+	}
+}
+
+static int rpivid_buf_out_validate(struct vb2_buffer *vb)
+{
+	struct vb2_v4l2_buffer *vbuf = to_vb2_v4l2_buffer(vb);
+
+	vbuf->field = V4L2_FIELD_NONE;
+	return 0;
+}
+
+static int rpivid_buf_prepare(struct vb2_buffer *vb)
+{
+	struct vb2_queue *vq = vb->vb2_queue;
+	struct rpivid_ctx *ctx = vb2_get_drv_priv(vq);
+	struct v4l2_pix_format *pix_fmt;
+
+	if (V4L2_TYPE_IS_OUTPUT(vq->type))
+		pix_fmt = &ctx->src_fmt;
+	else
+		pix_fmt = &ctx->dst_fmt;
+
+	if (vb2_plane_size(vb, 0) < pix_fmt->sizeimage)
+		return -EINVAL;
+
+	vb2_set_plane_payload(vb, 0, pix_fmt->sizeimage);
+
+	return 0;
+}
+
+static int rpivid_start_streaming(struct vb2_queue *vq, unsigned int count)
+{
+	struct rpivid_ctx *ctx = vb2_get_drv_priv(vq);
+	struct rpivid_dev *dev = ctx->dev;
+	int ret = 0;
+
+	if (ctx->src_fmt.pixelformat != V4L2_PIX_FMT_HEVC_SLICE)
+		return -EINVAL;
+
+	if (V4L2_TYPE_IS_OUTPUT(vq->type) && dev->dec_ops->start)
+		ret = dev->dec_ops->start(ctx);
+
+	ret = clk_set_rate(dev->clock, 500 * 1000 * 1000);
+	if (ret) {
+		dev_err(dev->dev, "Failed to set clock rate\n");
+		goto out;
+	}
+
+	ret = clk_prepare_enable(dev->clock);
+	if (ret)
+		dev_err(dev->dev, "Failed to enable clock\n");
+
+out:
+	if (ret)
+		rpivid_queue_cleanup(vq, VB2_BUF_STATE_QUEUED);
+
+	return ret;
+}
+
+static void rpivid_stop_streaming(struct vb2_queue *vq)
+{
+	struct rpivid_ctx *ctx = vb2_get_drv_priv(vq);
+	struct rpivid_dev *dev = ctx->dev;
+
+	if (V4L2_TYPE_IS_OUTPUT(vq->type) && dev->dec_ops->stop)
+		dev->dec_ops->stop(ctx);
+
+	rpivid_queue_cleanup(vq, VB2_BUF_STATE_ERROR);
+
+	clk_disable_unprepare(dev->clock);
+}
+
+static void rpivid_buf_queue(struct vb2_buffer *vb)
+{
+	struct vb2_v4l2_buffer *vbuf = to_vb2_v4l2_buffer(vb);
+	struct rpivid_ctx *ctx = vb2_get_drv_priv(vb->vb2_queue);
+
+	v4l2_m2m_buf_queue(ctx->fh.m2m_ctx, vbuf);
+}
+
+static void rpivid_buf_request_complete(struct vb2_buffer *vb)
+{
+	struct rpivid_ctx *ctx = vb2_get_drv_priv(vb->vb2_queue);
+
+	v4l2_ctrl_request_complete(vb->req_obj.req, &ctx->hdl);
+}
+
+static struct vb2_ops rpivid_qops = {
+	.queue_setup		= rpivid_queue_setup,
+	.buf_prepare		= rpivid_buf_prepare,
+	.buf_queue		= rpivid_buf_queue,
+	.buf_out_validate	= rpivid_buf_out_validate,
+	.buf_request_complete	= rpivid_buf_request_complete,
+	.start_streaming	= rpivid_start_streaming,
+	.stop_streaming		= rpivid_stop_streaming,
+	.wait_prepare		= vb2_ops_wait_prepare,
+	.wait_finish		= vb2_ops_wait_finish,
+};
+
+int rpivid_queue_init(void *priv, struct vb2_queue *src_vq,
+		      struct vb2_queue *dst_vq)
+{
+	struct rpivid_ctx *ctx = priv;
+	int ret;
+
+	src_vq->type = V4L2_BUF_TYPE_VIDEO_OUTPUT;
+	src_vq->io_modes = VB2_MMAP | VB2_DMABUF;
+	src_vq->drv_priv = ctx;
+	src_vq->buf_struct_size = sizeof(struct rpivid_buffer);
+	src_vq->min_buffers_needed = 1;
+	src_vq->ops = &rpivid_qops;
+	src_vq->mem_ops = &vb2_dma_contig_memops;
+	src_vq->timestamp_flags = V4L2_BUF_FLAG_TIMESTAMP_COPY;
+	src_vq->lock = &ctx->dev->dev_mutex;
+	src_vq->dev = ctx->dev->dev;
+	src_vq->supports_requests = true;
+	src_vq->requires_requests = true;
+
+	ret = vb2_queue_init(src_vq);
+	if (ret)
+		return ret;
+
+	dst_vq->type = V4L2_BUF_TYPE_VIDEO_CAPTURE;
+	dst_vq->io_modes = VB2_MMAP | VB2_DMABUF;
+	dst_vq->drv_priv = ctx;
+	dst_vq->buf_struct_size = sizeof(struct rpivid_buffer);
+	dst_vq->min_buffers_needed = 1;
+	dst_vq->ops = &rpivid_qops;
+	dst_vq->mem_ops = &vb2_dma_contig_memops;
+	dst_vq->timestamp_flags = V4L2_BUF_FLAG_TIMESTAMP_COPY;
+	dst_vq->lock = &ctx->dev->dev_mutex;
+	dst_vq->dev = ctx->dev->dev;
+
+	return vb2_queue_init(dst_vq);
+}

drivers/staging/media/rpivid/rpivid_video.h

+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * Raspberry Pi HEVC driver
+ *
+ * Copyright (C) 2020 Raspberry Pi (Trading) Ltd
+ *
+ * Based on the Cedrus VPU driver, that is:
+ *
+ * Copyright (C) 2016 Florent Revest <florent.revest@free-electrons.com>
+ * Copyright (C) 2018 Paul Kocialkowski <paul.kocialkowski@bootlin.com>
+ * Copyright (C) 2018 Bootlin
+ */
+
+#ifndef _RPIVID_VIDEO_H_
+#define _RPIVID_VIDEO_H_
+
+struct rpivid_format {
+	u32		pixelformat;
+	u32		directions;
+	unsigned int	capabilities;
+};
+
+extern const struct v4l2_ioctl_ops rpivid_ioctl_ops;
+
+int rpivid_queue_init(void *priv, struct vb2_queue *src_vq,
+		      struct vb2_queue *dst_vq);
+int rpivid_prepare_src_format(struct v4l2_pix_format *pix_fmt);
+int rpivid_prepare_dst_format(struct v4l2_pix_format *pix_fmt);
+
+#endif