1/*
2 * Coda multi-standard codec IP
3 *
4 * Copyright (C) 2012 Vista Silicon S.L.
5 *    Javier Martin, <javier.martin@vista-silicon.com>
6 *    Xavier Duret
7 *
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License as published by
10 * the Free Software Foundation; either version 2 of the License, or
11 * (at your option) any later version.
12 */
13
14#include <linux/clk.h>
15#include <linux/debugfs.h>
16#include <linux/delay.h>
17#include <linux/firmware.h>
18#include <linux/gcd.h>
19#include <linux/genalloc.h>
20#include <linux/interrupt.h>
21#include <linux/io.h>
22#include <linux/irq.h>
23#include <linux/kfifo.h>
24#include <linux/module.h>
25#include <linux/of_device.h>
26#include <linux/platform_device.h>
27#include <linux/pm_runtime.h>
28#include <linux/slab.h>
29#include <linux/videodev2.h>
30#include <linux/of.h>
31#include <linux/platform_data/coda.h>
32#include <linux/reset.h>
33
34#include <media/v4l2-ctrls.h>
35#include <media/v4l2-device.h>
36#include <media/v4l2-event.h>
37#include <media/v4l2-ioctl.h>
38#include <media/v4l2-mem2mem.h>
39#include <media/videobuf2-v4l2.h>
40#include <media/videobuf2-dma-contig.h>
41#include <media/videobuf2-vmalloc.h>
42
43#include "coda.h"
44
45#define CODA_NAME		"coda"
46
47#define CODADX6_MAX_INSTANCES	4
48#define CODA_MAX_FORMATS	4
49
50#define CODA_ISRAM_SIZE	(2048 * 2)
51
52#define MIN_W 176
53#define MIN_H 144
54
55#define S_ALIGN		1 /* multiple of 2 */
56#define W_ALIGN		1 /* multiple of 2 */
57#define H_ALIGN		1 /* multiple of 2 */
58
59#define fh_to_ctx(__fh)	container_of(__fh, struct coda_ctx, fh)
60
61int coda_debug;
62module_param(coda_debug, int, 0644);
63MODULE_PARM_DESC(coda_debug, "Debug level (0-2)");
64
65static int disable_tiling;
66module_param(disable_tiling, int, 0644);
67MODULE_PARM_DESC(disable_tiling, "Disable tiled frame buffers");
68
69void coda_write(struct coda_dev *dev, u32 data, u32 reg)
70{
71	v4l2_dbg(2, coda_debug, &dev->v4l2_dev,
72		 "%s: data=0x%x, reg=0x%x\n", __func__, data, reg);
73	writel(data, dev->regs_base + reg);
74}
75
76unsigned int coda_read(struct coda_dev *dev, u32 reg)
77{
78	u32 data;
79
80	data = readl(dev->regs_base + reg);
81	v4l2_dbg(2, coda_debug, &dev->v4l2_dev,
82		 "%s: data=0x%x, reg=0x%x\n", __func__, data, reg);
83	return data;
84}
85
86void coda_write_base(struct coda_ctx *ctx, struct coda_q_data *q_data,
87		     struct vb2_v4l2_buffer *buf, unsigned int reg_y)
88{
89	u32 base_y = vb2_dma_contig_plane_dma_addr(&buf->vb2_buf, 0);
90	u32 base_cb, base_cr;
91
92	switch (q_data->fourcc) {
93	case V4L2_PIX_FMT_NV12:
94	case V4L2_PIX_FMT_YUV420:
95	default:
96		base_cb = base_y + q_data->bytesperline * q_data->height;
97		base_cr = base_cb + q_data->bytesperline * q_data->height / 4;
98		break;
99	case V4L2_PIX_FMT_YVU420:
100		/* Switch Cb and Cr for YVU420 format */
101		base_cr = base_y + q_data->bytesperline * q_data->height;
102		base_cb = base_cr + q_data->bytesperline * q_data->height / 4;
103		break;
104	case V4L2_PIX_FMT_YUV422P:
105		base_cb = base_y + q_data->bytesperline * q_data->height;
106		base_cr = base_cb + q_data->bytesperline * q_data->height / 2;
107	}
108
109	coda_write(ctx->dev, base_y, reg_y);
110	coda_write(ctx->dev, base_cb, reg_y + 4);
111	coda_write(ctx->dev, base_cr, reg_y + 8);
112}
113
114#define CODA_CODEC(mode, src_fourcc, dst_fourcc, max_w, max_h) \
115	{ mode, src_fourcc, dst_fourcc, max_w, max_h }
116
117/*
118 * Arrays of codecs supported by each given version of Coda:
119 *  i.MX27 -> codadx6
120 *  i.MX5x -> coda7
121 *  i.MX6  -> coda960
122 * Use V4L2_PIX_FMT_YUV420 as placeholder for all supported YUV 4:2:0 variants
123 */
124static const struct coda_codec codadx6_codecs[] = {
125	CODA_CODEC(CODADX6_MODE_ENCODE_H264, V4L2_PIX_FMT_YUV420, V4L2_PIX_FMT_H264,  720, 576),
126	CODA_CODEC(CODADX6_MODE_ENCODE_MP4,  V4L2_PIX_FMT_YUV420, V4L2_PIX_FMT_MPEG4, 720, 576),
127};
128
129static const struct coda_codec coda7_codecs[] = {
130	CODA_CODEC(CODA7_MODE_ENCODE_H264, V4L2_PIX_FMT_YUV420, V4L2_PIX_FMT_H264,   1280, 720),
131	CODA_CODEC(CODA7_MODE_ENCODE_MP4,  V4L2_PIX_FMT_YUV420, V4L2_PIX_FMT_MPEG4,  1280, 720),
132	CODA_CODEC(CODA7_MODE_ENCODE_MJPG, V4L2_PIX_FMT_YUV420, V4L2_PIX_FMT_JPEG,   8192, 8192),
133	CODA_CODEC(CODA7_MODE_DECODE_H264, V4L2_PIX_FMT_H264,   V4L2_PIX_FMT_YUV420, 1920, 1088),
134	CODA_CODEC(CODA7_MODE_DECODE_MP4,  V4L2_PIX_FMT_MPEG4,  V4L2_PIX_FMT_YUV420, 1920, 1088),
135	CODA_CODEC(CODA7_MODE_DECODE_MJPG, V4L2_PIX_FMT_JPEG,   V4L2_PIX_FMT_YUV420, 8192, 8192),
136};
137
138static const struct coda_codec coda9_codecs[] = {
139	CODA_CODEC(CODA9_MODE_ENCODE_H264, V4L2_PIX_FMT_YUV420, V4L2_PIX_FMT_H264,   1920, 1088),
140	CODA_CODEC(CODA9_MODE_ENCODE_MP4,  V4L2_PIX_FMT_YUV420, V4L2_PIX_FMT_MPEG4,  1920, 1088),
141	CODA_CODEC(CODA9_MODE_DECODE_H264, V4L2_PIX_FMT_H264,   V4L2_PIX_FMT_YUV420, 1920, 1088),
142	CODA_CODEC(CODA9_MODE_DECODE_MP4,  V4L2_PIX_FMT_MPEG4,  V4L2_PIX_FMT_YUV420, 1920, 1088),
143};
144
145struct coda_video_device {
146	const char *name;
147	enum coda_inst_type type;
148	const struct coda_context_ops *ops;
149	bool direct;
150	u32 src_formats[CODA_MAX_FORMATS];
151	u32 dst_formats[CODA_MAX_FORMATS];
152};
153
154static const struct coda_video_device coda_bit_encoder = {
155	.name = "coda-encoder",
156	.type = CODA_INST_ENCODER,
157	.ops = &coda_bit_encode_ops,
158	.src_formats = {
159		V4L2_PIX_FMT_NV12,
160		V4L2_PIX_FMT_YUV420,
161		V4L2_PIX_FMT_YVU420,
162	},
163	.dst_formats = {
164		V4L2_PIX_FMT_H264,
165		V4L2_PIX_FMT_MPEG4,
166	},
167};
168
169static const struct coda_video_device coda_bit_jpeg_encoder = {
170	.name = "coda-jpeg-encoder",
171	.type = CODA_INST_ENCODER,
172	.ops = &coda_bit_encode_ops,
173	.src_formats = {
174		V4L2_PIX_FMT_NV12,
175		V4L2_PIX_FMT_YUV420,
176		V4L2_PIX_FMT_YVU420,
177		V4L2_PIX_FMT_YUV422P,
178	},
179	.dst_formats = {
180		V4L2_PIX_FMT_JPEG,
181	},
182};
183
184static const struct coda_video_device coda_bit_decoder = {
185	.name = "coda-decoder",
186	.type = CODA_INST_DECODER,
187	.ops = &coda_bit_decode_ops,
188	.src_formats = {
189		V4L2_PIX_FMT_H264,
190		V4L2_PIX_FMT_MPEG4,
191	},
192	.dst_formats = {
193		V4L2_PIX_FMT_NV12,
194		V4L2_PIX_FMT_YUV420,
195		V4L2_PIX_FMT_YVU420,
196	},
197};
198
199static const struct coda_video_device coda_bit_jpeg_decoder = {
200	.name = "coda-jpeg-decoder",
201	.type = CODA_INST_DECODER,
202	.ops = &coda_bit_decode_ops,
203	.src_formats = {
204		V4L2_PIX_FMT_JPEG,
205	},
206	.dst_formats = {
207		V4L2_PIX_FMT_NV12,
208		V4L2_PIX_FMT_YUV420,
209		V4L2_PIX_FMT_YVU420,
210		V4L2_PIX_FMT_YUV422P,
211	},
212};
213
214static const struct coda_video_device *codadx6_video_devices[] = {
215	&coda_bit_encoder,
216};
217
218static const struct coda_video_device *coda7_video_devices[] = {
219	&coda_bit_jpeg_encoder,
220	&coda_bit_jpeg_decoder,
221	&coda_bit_encoder,
222	&coda_bit_decoder,
223};
224
225static const struct coda_video_device *coda9_video_devices[] = {
226	&coda_bit_encoder,
227	&coda_bit_decoder,
228};
229
230/*
231 * Normalize all supported YUV 4:2:0 formats to the value used in the codec
232 * tables.
233 */
234static u32 coda_format_normalize_yuv(u32 fourcc)
235{
236	switch (fourcc) {
237	case V4L2_PIX_FMT_NV12:
238	case V4L2_PIX_FMT_YUV420:
239	case V4L2_PIX_FMT_YVU420:
240	case V4L2_PIX_FMT_YUV422P:
241		return V4L2_PIX_FMT_YUV420;
242	default:
243		return fourcc;
244	}
245}
246
247static const struct coda_codec *coda_find_codec(struct coda_dev *dev,
248						int src_fourcc, int dst_fourcc)
249{
250	const struct coda_codec *codecs = dev->devtype->codecs;
251	int num_codecs = dev->devtype->num_codecs;
252	int k;
253
254	src_fourcc = coda_format_normalize_yuv(src_fourcc);
255	dst_fourcc = coda_format_normalize_yuv(dst_fourcc);
256	if (src_fourcc == dst_fourcc)
257		return NULL;
258
259	for (k = 0; k < num_codecs; k++) {
260		if (codecs[k].src_fourcc == src_fourcc &&
261		    codecs[k].dst_fourcc == dst_fourcc)
262			break;
263	}
264
265	if (k == num_codecs)
266		return NULL;
267
268	return &codecs[k];
269}
270
271static void coda_get_max_dimensions(struct coda_dev *dev,
272				    const struct coda_codec *codec,
273				    int *max_w, int *max_h)
274{
275	const struct coda_codec *codecs = dev->devtype->codecs;
276	int num_codecs = dev->devtype->num_codecs;
277	unsigned int w, h;
278	int k;
279
280	if (codec) {
281		w = codec->max_w;
282		h = codec->max_h;
283	} else {
284		for (k = 0, w = 0, h = 0; k < num_codecs; k++) {
285			w = max(w, codecs[k].max_w);
286			h = max(h, codecs[k].max_h);
287		}
288	}
289
290	if (max_w)
291		*max_w = w;
292	if (max_h)
293		*max_h = h;
294}
295
296const struct coda_video_device *to_coda_video_device(struct video_device *vdev)
297{
298	struct coda_dev *dev = video_get_drvdata(vdev);
299	unsigned int i = vdev - dev->vfd;
300
301	if (i >= dev->devtype->num_vdevs)
302		return NULL;
303
304	return dev->devtype->vdevs[i];
305}
306
307const char *coda_product_name(int product)
308{
309	static char buf[9];
310
311	switch (product) {
312	case CODA_DX6:
313		return "CodaDx6";
314	case CODA_7541:
315		return "CODA7541";
316	case CODA_960:
317		return "CODA960";
318	default:
319		snprintf(buf, sizeof(buf), "(0x%04x)", product);
320		return buf;
321	}
322}
323
324/*
325 * V4L2 ioctl() operations.
326 */
327static int coda_querycap(struct file *file, void *priv,
328			 struct v4l2_capability *cap)
329{
330	struct coda_ctx *ctx = fh_to_ctx(priv);
331
332	strlcpy(cap->driver, CODA_NAME, sizeof(cap->driver));
333	strlcpy(cap->card, coda_product_name(ctx->dev->devtype->product),
334		sizeof(cap->card));
335	strlcpy(cap->bus_info, "platform:" CODA_NAME, sizeof(cap->bus_info));
336	cap->device_caps = V4L2_CAP_VIDEO_M2M | V4L2_CAP_STREAMING;
337	cap->capabilities = cap->device_caps | V4L2_CAP_DEVICE_CAPS;
338
339	return 0;
340}
341
342static int coda_enum_fmt(struct file *file, void *priv,
343			 struct v4l2_fmtdesc *f)
344{
345	struct video_device *vdev = video_devdata(file);
346	const struct coda_video_device *cvd = to_coda_video_device(vdev);
347	const u32 *formats;
348
349	if (f->type == V4L2_BUF_TYPE_VIDEO_OUTPUT)
350		formats = cvd->src_formats;
351	else if (f->type == V4L2_BUF_TYPE_VIDEO_CAPTURE)
352		formats = cvd->dst_formats;
353	else
354		return -EINVAL;
355
356	if (f->index >= CODA_MAX_FORMATS || formats[f->index] == 0)
357		return -EINVAL;
358
359	f->pixelformat = formats[f->index];
360
361	return 0;
362}
363
364static int coda_g_fmt(struct file *file, void *priv,
365		      struct v4l2_format *f)
366{
367	struct coda_q_data *q_data;
368	struct coda_ctx *ctx = fh_to_ctx(priv);
369
370	q_data = get_q_data(ctx, f->type);
371	if (!q_data)
372		return -EINVAL;
373
374	f->fmt.pix.field	= V4L2_FIELD_NONE;
375	f->fmt.pix.pixelformat	= q_data->fourcc;
376	f->fmt.pix.width	= q_data->width;
377	f->fmt.pix.height	= q_data->height;
378	f->fmt.pix.bytesperline = q_data->bytesperline;
379
380	f->fmt.pix.sizeimage	= q_data->sizeimage;
381	if (f->fmt.pix.pixelformat == V4L2_PIX_FMT_JPEG)
382		f->fmt.pix.colorspace = V4L2_COLORSPACE_JPEG;
383	else
384		f->fmt.pix.colorspace = ctx->colorspace;
385
386	return 0;
387}
388
389static int coda_try_pixelformat(struct coda_ctx *ctx, struct v4l2_format *f)
390{
391	struct coda_q_data *q_data;
392	const u32 *formats;
393	int i;
394
395	if (f->type == V4L2_BUF_TYPE_VIDEO_OUTPUT)
396		formats = ctx->cvd->src_formats;
397	else if (f->type == V4L2_BUF_TYPE_VIDEO_CAPTURE)
398		formats = ctx->cvd->dst_formats;
399	else
400		return -EINVAL;
401
402	for (i = 0; i < CODA_MAX_FORMATS; i++) {
403		if (formats[i] == f->fmt.pix.pixelformat) {
404			f->fmt.pix.pixelformat = formats[i];
405			return 0;
406		}
407	}
408
409	/* Fall back to currently set pixelformat */
410	q_data = get_q_data(ctx, f->type);
411	f->fmt.pix.pixelformat = q_data->fourcc;
412
413	return 0;
414}
415
416static unsigned int coda_estimate_sizeimage(struct coda_ctx *ctx, u32 sizeimage,
417					    u32 width, u32 height)
418{
419	/*
420	 * This is a rough estimate for sensible compressed buffer
421	 * sizes (between 1 and 16 bits per pixel). This could be
422	 * improved by better format specific worst case estimates.
423	 */
424	return round_up(clamp(sizeimage, width * height / 8,
425					 width * height * 2), PAGE_SIZE);
426}
427
428static int coda_try_fmt(struct coda_ctx *ctx, const struct coda_codec *codec,
429			struct v4l2_format *f)
430{
431	struct coda_dev *dev = ctx->dev;
432	unsigned int max_w, max_h;
433	enum v4l2_field field;
434
435	field = f->fmt.pix.field;
436	if (field == V4L2_FIELD_ANY)
437		field = V4L2_FIELD_NONE;
438	else if (V4L2_FIELD_NONE != field)
439		return -EINVAL;
440
441	/* V4L2 specification suggests the driver corrects the format struct
442	 * if any of the dimensions is unsupported */
443	f->fmt.pix.field = field;
444
445	coda_get_max_dimensions(dev, codec, &max_w, &max_h);
446	v4l_bound_align_image(&f->fmt.pix.width, MIN_W, max_w, W_ALIGN,
447			      &f->fmt.pix.height, MIN_H, max_h, H_ALIGN,
448			      S_ALIGN);
449
450	switch (f->fmt.pix.pixelformat) {
451	case V4L2_PIX_FMT_NV12:
452	case V4L2_PIX_FMT_YUV420:
453	case V4L2_PIX_FMT_YVU420:
454		/*
455		 * Frame stride must be at least multiple of 8,
456		 * but multiple of 16 for h.264 or JPEG 4:2:x
457		 */
458		f->fmt.pix.bytesperline = round_up(f->fmt.pix.width, 16);
459		f->fmt.pix.sizeimage = f->fmt.pix.bytesperline *
460					f->fmt.pix.height * 3 / 2;
461		break;
462	case V4L2_PIX_FMT_YUV422P:
463		f->fmt.pix.bytesperline = round_up(f->fmt.pix.width, 16);
464		f->fmt.pix.sizeimage = f->fmt.pix.bytesperline *
465					f->fmt.pix.height * 2;
466		break;
467	case V4L2_PIX_FMT_JPEG:
468		f->fmt.pix.colorspace = V4L2_COLORSPACE_JPEG;
469		/* fallthrough */
470	case V4L2_PIX_FMT_H264:
471	case V4L2_PIX_FMT_MPEG4:
472		f->fmt.pix.bytesperline = 0;
473		f->fmt.pix.sizeimage = coda_estimate_sizeimage(ctx,
474							f->fmt.pix.sizeimage,
475							f->fmt.pix.width,
476							f->fmt.pix.height);
477		break;
478	default:
479		BUG();
480	}
481
482	return 0;
483}
484
485static int coda_try_fmt_vid_cap(struct file *file, void *priv,
486				struct v4l2_format *f)
487{
488	struct coda_ctx *ctx = fh_to_ctx(priv);
489	const struct coda_q_data *q_data_src;
490	const struct coda_codec *codec;
491	struct vb2_queue *src_vq;
492	int ret;
493
494	ret = coda_try_pixelformat(ctx, f);
495	if (ret < 0)
496		return ret;
497
498	q_data_src = get_q_data(ctx, V4L2_BUF_TYPE_VIDEO_OUTPUT);
499
500	/*
501	 * If the source format is already fixed, only allow the same output
502	 * resolution
503	 */
504	src_vq = v4l2_m2m_get_vq(ctx->fh.m2m_ctx, V4L2_BUF_TYPE_VIDEO_OUTPUT);
505	if (vb2_is_streaming(src_vq)) {
506		f->fmt.pix.width = q_data_src->width;
507		f->fmt.pix.height = q_data_src->height;
508	}
509
510	f->fmt.pix.colorspace = ctx->colorspace;
511
512	q_data_src = get_q_data(ctx, V4L2_BUF_TYPE_VIDEO_OUTPUT);
513	codec = coda_find_codec(ctx->dev, q_data_src->fourcc,
514				f->fmt.pix.pixelformat);
515	if (!codec)
516		return -EINVAL;
517
518	ret = coda_try_fmt(ctx, codec, f);
519	if (ret < 0)
520		return ret;
521
522	/* The h.264 decoder only returns complete 16x16 macroblocks */
523	if (codec && codec->src_fourcc == V4L2_PIX_FMT_H264) {
524		f->fmt.pix.width = f->fmt.pix.width;
525		f->fmt.pix.height = round_up(f->fmt.pix.height, 16);
526		f->fmt.pix.bytesperline = round_up(f->fmt.pix.width, 16);
527		f->fmt.pix.sizeimage = f->fmt.pix.bytesperline *
528				       f->fmt.pix.height * 3 / 2;
529	}
530
531	return 0;
532}
533
534static int coda_try_fmt_vid_out(struct file *file, void *priv,
535				struct v4l2_format *f)
536{
537	struct coda_ctx *ctx = fh_to_ctx(priv);
538	struct coda_dev *dev = ctx->dev;
539	const struct coda_q_data *q_data_dst;
540	const struct coda_codec *codec;
541	int ret;
542
543	ret = coda_try_pixelformat(ctx, f);
544	if (ret < 0)
545		return ret;
546
547	switch (f->fmt.pix.colorspace) {
548	case V4L2_COLORSPACE_REC709:
549	case V4L2_COLORSPACE_JPEG:
550		break;
551	default:
552		if (f->fmt.pix.pixelformat == V4L2_PIX_FMT_JPEG)
553			f->fmt.pix.colorspace = V4L2_COLORSPACE_JPEG;
554		else
555			f->fmt.pix.colorspace = V4L2_COLORSPACE_REC709;
556	}
557
558	q_data_dst = get_q_data(ctx, V4L2_BUF_TYPE_VIDEO_CAPTURE);
559	codec = coda_find_codec(dev, f->fmt.pix.pixelformat, q_data_dst->fourcc);
560
561	return coda_try_fmt(ctx, codec, f);
562}
563
564static int coda_s_fmt(struct coda_ctx *ctx, struct v4l2_format *f)
565{
566	struct coda_q_data *q_data;
567	struct vb2_queue *vq;
568
569	vq = v4l2_m2m_get_vq(ctx->fh.m2m_ctx, f->type);
570	if (!vq)
571		return -EINVAL;
572
573	q_data = get_q_data(ctx, f->type);
574	if (!q_data)
575		return -EINVAL;
576
577	if (vb2_is_busy(vq)) {
578		v4l2_err(&ctx->dev->v4l2_dev, "%s queue busy\n", __func__);
579		return -EBUSY;
580	}
581
582	q_data->fourcc = f->fmt.pix.pixelformat;
583	q_data->width = f->fmt.pix.width;
584	q_data->height = f->fmt.pix.height;
585	q_data->bytesperline = f->fmt.pix.bytesperline;
586	q_data->sizeimage = f->fmt.pix.sizeimage;
587	q_data->rect.left = 0;
588	q_data->rect.top = 0;
589	q_data->rect.width = f->fmt.pix.width;
590	q_data->rect.height = f->fmt.pix.height;
591
592	switch (f->fmt.pix.pixelformat) {
593	case V4L2_PIX_FMT_NV12:
594		if (f->type == V4L2_BUF_TYPE_VIDEO_OUTPUT) {
595			ctx->tiled_map_type = GDI_TILED_FRAME_MB_RASTER_MAP;
596			if (!disable_tiling)
597				break;
598		}
599		/* else fall through */
600	case V4L2_PIX_FMT_YUV420:
601	case V4L2_PIX_FMT_YVU420:
602		ctx->tiled_map_type = GDI_LINEAR_FRAME_MAP;
603		break;
604	default:
605		break;
606	}
607
608	v4l2_dbg(1, coda_debug, &ctx->dev->v4l2_dev,
609		"Setting format for type %d, wxh: %dx%d, fmt: %d\n",
610		f->type, q_data->width, q_data->height, q_data->fourcc);
611
612	return 0;
613}
614
615static int coda_s_fmt_vid_cap(struct file *file, void *priv,
616			      struct v4l2_format *f)
617{
618	struct coda_ctx *ctx = fh_to_ctx(priv);
619	int ret;
620
621	ret = coda_try_fmt_vid_cap(file, priv, f);
622	if (ret)
623		return ret;
624
625	return coda_s_fmt(ctx, f);
626}
627
628static int coda_s_fmt_vid_out(struct file *file, void *priv,
629			      struct v4l2_format *f)
630{
631	struct coda_ctx *ctx = fh_to_ctx(priv);
632	struct v4l2_format f_cap;
633	int ret;
634
635	ret = coda_try_fmt_vid_out(file, priv, f);
636	if (ret)
637		return ret;
638
639	ret = coda_s_fmt(ctx, f);
640	if (ret)
641		return ret;
642
643	ctx->colorspace = f->fmt.pix.colorspace;
644
645	memset(&f_cap, 0, sizeof(f_cap));
646	f_cap.type = V4L2_BUF_TYPE_VIDEO_CAPTURE;
647	coda_g_fmt(file, priv, &f_cap);
648	f_cap.fmt.pix.width = f->fmt.pix.width;
649	f_cap.fmt.pix.height = f->fmt.pix.height;
650
651	ret = coda_try_fmt_vid_cap(file, priv, &f_cap);
652	if (ret)
653		return ret;
654
655	return coda_s_fmt(ctx, &f_cap);
656}
657
658static int coda_reqbufs(struct file *file, void *priv,
659			struct v4l2_requestbuffers *rb)
660{
661	struct coda_ctx *ctx = fh_to_ctx(priv);
662	int ret;
663
664	ret = v4l2_m2m_reqbufs(file, ctx->fh.m2m_ctx, rb);
665	if (ret)
666		return ret;
667
668	/*
669	 * Allow to allocate instance specific per-context buffers, such as
670	 * bitstream ringbuffer, slice buffer, work buffer, etc. if needed.
671	 */
672	if (rb->type == V4L2_BUF_TYPE_VIDEO_OUTPUT && ctx->ops->reqbufs)
673		return ctx->ops->reqbufs(ctx, rb);
674
675	return 0;
676}
677
678static int coda_qbuf(struct file *file, void *priv,
679		     struct v4l2_buffer *buf)
680{
681	struct coda_ctx *ctx = fh_to_ctx(priv);
682
683	return v4l2_m2m_qbuf(file, ctx->fh.m2m_ctx, buf);
684}
685
686static bool coda_buf_is_end_of_stream(struct coda_ctx *ctx,
687				      struct vb2_v4l2_buffer *buf)
688{
689	struct vb2_queue *src_vq;
690
691	src_vq = v4l2_m2m_get_vq(ctx->fh.m2m_ctx, V4L2_BUF_TYPE_VIDEO_OUTPUT);
692
693	return ((ctx->bit_stream_param & CODA_BIT_STREAM_END_FLAG) &&
694		(buf->sequence == (ctx->qsequence - 1)));
695}
696
697void coda_m2m_buf_done(struct coda_ctx *ctx, struct vb2_v4l2_buffer *buf,
698		       enum vb2_buffer_state state)
699{
700	const struct v4l2_event eos_event = {
701		.type = V4L2_EVENT_EOS
702	};
703
704	if (coda_buf_is_end_of_stream(ctx, buf)) {
705		buf->flags |= V4L2_BUF_FLAG_LAST;
706
707		v4l2_event_queue_fh(&ctx->fh, &eos_event);
708	}
709
710	v4l2_m2m_buf_done(buf, state);
711}
712
713static int coda_g_selection(struct file *file, void *fh,
714			    struct v4l2_selection *s)
715{
716	struct coda_ctx *ctx = fh_to_ctx(fh);
717	struct coda_q_data *q_data;
718	struct v4l2_rect r, *rsel;
719
720	q_data = get_q_data(ctx, s->type);
721	if (!q_data)
722		return -EINVAL;
723
724	r.left = 0;
725	r.top = 0;
726	r.width = q_data->width;
727	r.height = q_data->height;
728	rsel = &q_data->rect;
729
730	switch (s->target) {
731	case V4L2_SEL_TGT_CROP_DEFAULT:
732	case V4L2_SEL_TGT_CROP_BOUNDS:
733		rsel = &r;
734		/* fallthrough */
735	case V4L2_SEL_TGT_CROP:
736		if (s->type != V4L2_BUF_TYPE_VIDEO_OUTPUT)
737			return -EINVAL;
738		break;
739	case V4L2_SEL_TGT_COMPOSE_BOUNDS:
740	case V4L2_SEL_TGT_COMPOSE_PADDED:
741		rsel = &r;
742		/* fallthrough */
743	case V4L2_SEL_TGT_COMPOSE:
744	case V4L2_SEL_TGT_COMPOSE_DEFAULT:
745		if (s->type != V4L2_BUF_TYPE_VIDEO_CAPTURE)
746			return -EINVAL;
747		break;
748	default:
749		return -EINVAL;
750	}
751
752	s->r = *rsel;
753
754	return 0;
755}
756
757static int coda_try_decoder_cmd(struct file *file, void *fh,
758				struct v4l2_decoder_cmd *dc)
759{
760	if (dc->cmd != V4L2_DEC_CMD_STOP)
761		return -EINVAL;
762
763	if (dc->flags & V4L2_DEC_CMD_STOP_TO_BLACK)
764		return -EINVAL;
765
766	if (!(dc->flags & V4L2_DEC_CMD_STOP_IMMEDIATELY) && (dc->stop.pts != 0))
767		return -EINVAL;
768
769	return 0;
770}
771
772static int coda_decoder_cmd(struct file *file, void *fh,
773			    struct v4l2_decoder_cmd *dc)
774{
775	struct coda_ctx *ctx = fh_to_ctx(fh);
776	int ret;
777
778	ret = coda_try_decoder_cmd(file, fh, dc);
779	if (ret < 0)
780		return ret;
781
782	/* Ignore decoder stop command silently in encoder context */
783	if (ctx->inst_type != CODA_INST_DECODER)
784		return 0;
785
786	/* Set the stream-end flag on this context */
787	coda_bit_stream_end_flag(ctx);
788	ctx->hold = false;
789	v4l2_m2m_try_schedule(ctx->fh.m2m_ctx);
790
791	return 0;
792}
793
794static int coda_g_parm(struct file *file, void *fh, struct v4l2_streamparm *a)
795{
796	struct coda_ctx *ctx = fh_to_ctx(fh);
797	struct v4l2_fract *tpf;
798
799	if (a->type != V4L2_BUF_TYPE_VIDEO_OUTPUT)
800		return -EINVAL;
801
802	a->parm.output.capability = V4L2_CAP_TIMEPERFRAME;
803	tpf = &a->parm.output.timeperframe;
804	tpf->denominator = ctx->params.framerate & CODA_FRATE_RES_MASK;
805	tpf->numerator = 1 + (ctx->params.framerate >>
806			      CODA_FRATE_DIV_OFFSET);
807
808	return 0;
809}
810
811/*
812 * Approximate timeperframe v4l2_fract with values that can be written
813 * into the 16-bit CODA_FRATE_DIV and CODA_FRATE_RES fields.
814 */
815static void coda_approximate_timeperframe(struct v4l2_fract *timeperframe)
816{
817	struct v4l2_fract s = *timeperframe;
818	struct v4l2_fract f0;
819	struct v4l2_fract f1 = { 1, 0 };
820	struct v4l2_fract f2 = { 0, 1 };
821	unsigned int i, div, s_denominator;
822
823	/* Lower bound is 1/65535 */
824	if (s.numerator == 0 || s.denominator / s.numerator > 65535) {
825		timeperframe->numerator = 1;
826		timeperframe->denominator = 65535;
827		return;
828	}
829
830	/* Upper bound is 65536/1, map everything above to infinity */
831	if (s.denominator == 0 || s.numerator / s.denominator > 65536) {
832		timeperframe->numerator = 1;
833		timeperframe->denominator = 0;
834		return;
835	}
836
837	/* Reduce fraction to lowest terms */
838	div = gcd(s.numerator, s.denominator);
839	if (div > 1) {
840		s.numerator /= div;
841		s.denominator /= div;
842	}
843
844	if (s.numerator <= 65536 && s.denominator < 65536) {
845		*timeperframe = s;
846		return;
847	}
848
849	/* Find successive convergents from continued fraction expansion */
850	while (f2.numerator <= 65536 && f2.denominator < 65536) {
851		f0 = f1;
852		f1 = f2;
853
854		/* Stop when f2 exactly equals timeperframe */
855		if (s.numerator == 0)
856			break;
857
858		i = s.denominator / s.numerator;
859
860		f2.numerator = f0.numerator + i * f1.numerator;
861		f2.denominator = f0.denominator + i * f2.denominator;
862
863		s_denominator = s.numerator;
864		s.numerator = s.denominator % s.numerator;
865		s.denominator = s_denominator;
866	}
867
868	*timeperframe = f1;
869}
870
871static uint32_t coda_timeperframe_to_frate(struct v4l2_fract *timeperframe)
872{
873	return ((timeperframe->numerator - 1) << CODA_FRATE_DIV_OFFSET) |
874		timeperframe->denominator;
875}
876
877static int coda_s_parm(struct file *file, void *fh, struct v4l2_streamparm *a)
878{
879	struct coda_ctx *ctx = fh_to_ctx(fh);
880	struct v4l2_fract *tpf;
881
882	if (a->type != V4L2_BUF_TYPE_VIDEO_OUTPUT)
883		return -EINVAL;
884
885	tpf = &a->parm.output.timeperframe;
886	coda_approximate_timeperframe(tpf);
887	ctx->params.framerate = coda_timeperframe_to_frate(tpf);
888
889	return 0;
890}
891
892static int coda_subscribe_event(struct v4l2_fh *fh,
893				const struct v4l2_event_subscription *sub)
894{
895	switch (sub->type) {
896	case V4L2_EVENT_EOS:
897		return v4l2_event_subscribe(fh, sub, 0, NULL);
898	default:
899		return v4l2_ctrl_subscribe_event(fh, sub);
900	}
901}
902
903static const struct v4l2_ioctl_ops coda_ioctl_ops = {
904	.vidioc_querycap	= coda_querycap,
905
906	.vidioc_enum_fmt_vid_cap = coda_enum_fmt,
907	.vidioc_g_fmt_vid_cap	= coda_g_fmt,
908	.vidioc_try_fmt_vid_cap	= coda_try_fmt_vid_cap,
909	.vidioc_s_fmt_vid_cap	= coda_s_fmt_vid_cap,
910
911	.vidioc_enum_fmt_vid_out = coda_enum_fmt,
912	.vidioc_g_fmt_vid_out	= coda_g_fmt,
913	.vidioc_try_fmt_vid_out	= coda_try_fmt_vid_out,
914	.vidioc_s_fmt_vid_out	= coda_s_fmt_vid_out,
915
916	.vidioc_reqbufs		= coda_reqbufs,
917	.vidioc_querybuf	= v4l2_m2m_ioctl_querybuf,
918
919	.vidioc_qbuf		= coda_qbuf,
920	.vidioc_expbuf		= v4l2_m2m_ioctl_expbuf,
921	.vidioc_dqbuf		= v4l2_m2m_ioctl_dqbuf,
922	.vidioc_create_bufs	= v4l2_m2m_ioctl_create_bufs,
923
924	.vidioc_streamon	= v4l2_m2m_ioctl_streamon,
925	.vidioc_streamoff	= v4l2_m2m_ioctl_streamoff,
926
927	.vidioc_g_selection	= coda_g_selection,
928
929	.vidioc_try_decoder_cmd	= coda_try_decoder_cmd,
930	.vidioc_decoder_cmd	= coda_decoder_cmd,
931
932	.vidioc_g_parm		= coda_g_parm,
933	.vidioc_s_parm		= coda_s_parm,
934
935	.vidioc_subscribe_event = coda_subscribe_event,
936	.vidioc_unsubscribe_event = v4l2_event_unsubscribe,
937};
938
939/*
940 * Mem-to-mem operations.
941 */
942
943static void coda_device_run(void *m2m_priv)
944{
945	struct coda_ctx *ctx = m2m_priv;
946	struct coda_dev *dev = ctx->dev;
947
948	queue_work(dev->workqueue, &ctx->pic_run_work);
949}
950
951static void coda_pic_run_work(struct work_struct *work)
952{
953	struct coda_ctx *ctx = container_of(work, struct coda_ctx, pic_run_work);
954	struct coda_dev *dev = ctx->dev;
955	int ret;
956
957	mutex_lock(&ctx->buffer_mutex);
958	mutex_lock(&dev->coda_mutex);
959
960	ret = ctx->ops->prepare_run(ctx);
961	if (ret < 0 && ctx->inst_type == CODA_INST_DECODER) {
962		mutex_unlock(&dev->coda_mutex);
963		mutex_unlock(&ctx->buffer_mutex);
964		/* job_finish scheduled by prepare_decode */
965		return;
966	}
967
968	if (!wait_for_completion_timeout(&ctx->completion,
969					 msecs_to_jiffies(1000))) {
970		dev_err(&dev->plat_dev->dev, "CODA PIC_RUN timeout\n");
971
972		ctx->hold = true;
973
974		coda_hw_reset(ctx);
975	} else if (!ctx->aborting) {
976		ctx->ops->finish_run(ctx);
977	}
978
979	if ((ctx->aborting || (!ctx->streamon_cap && !ctx->streamon_out)) &&
980	    ctx->ops->seq_end_work)
981		queue_work(dev->workqueue, &ctx->seq_end_work);
982
983	mutex_unlock(&dev->coda_mutex);
984	mutex_unlock(&ctx->buffer_mutex);
985
986	v4l2_m2m_job_finish(ctx->dev->m2m_dev, ctx->fh.m2m_ctx);
987}
988
989static int coda_job_ready(void *m2m_priv)
990{
991	struct coda_ctx *ctx = m2m_priv;
992	int src_bufs = v4l2_m2m_num_src_bufs_ready(ctx->fh.m2m_ctx);
993
994	/*
995	 * For both 'P' and 'key' frame cases 1 picture
996	 * and 1 frame are needed. In the decoder case,
997	 * the compressed frame can be in the bitstream.
998	 */
999	if (!src_bufs && ctx->inst_type != CODA_INST_DECODER) {
1000		v4l2_dbg(1, coda_debug, &ctx->dev->v4l2_dev,
1001			 "not ready: not enough video buffers.\n");
1002		return 0;
1003	}
1004
1005	if (!v4l2_m2m_num_dst_bufs_ready(ctx->fh.m2m_ctx)) {
1006		v4l2_dbg(1, coda_debug, &ctx->dev->v4l2_dev,
1007			 "not ready: not enough video capture buffers.\n");
1008		return 0;
1009	}
1010
1011	if (ctx->inst_type == CODA_INST_DECODER && ctx->use_bit) {
1012		bool stream_end = ctx->bit_stream_param &
1013				  CODA_BIT_STREAM_END_FLAG;
1014		int num_metas = ctx->num_metas;
1015
1016		if (ctx->hold && !src_bufs) {
1017			v4l2_dbg(1, coda_debug, &ctx->dev->v4l2_dev,
1018				 "%d: not ready: on hold for more buffers.\n",
1019				 ctx->idx);
1020			return 0;
1021		}
1022
1023		if (!stream_end && (num_metas + src_bufs) < 2) {
1024			v4l2_dbg(1, coda_debug, &ctx->dev->v4l2_dev,
1025				 "%d: not ready: need 2 buffers available (%d, %d)\n",
1026				 ctx->idx, num_metas, src_bufs);
1027			return 0;
1028		}
1029
1030
1031		if (!src_bufs && !stream_end &&
1032		    (coda_get_bitstream_payload(ctx) < 512)) {
1033			v4l2_dbg(1, coda_debug, &ctx->dev->v4l2_dev,
1034				 "%d: not ready: not enough bitstream data (%d).\n",
1035				 ctx->idx, coda_get_bitstream_payload(ctx));
1036			return 0;
1037		}
1038	}
1039
1040	if (ctx->aborting) {
1041		v4l2_dbg(1, coda_debug, &ctx->dev->v4l2_dev,
1042			 "not ready: aborting\n");
1043		return 0;
1044	}
1045
1046	v4l2_dbg(1, coda_debug, &ctx->dev->v4l2_dev,
1047			"job ready\n");
1048
1049	return 1;
1050}
1051
1052static void coda_job_abort(void *priv)
1053{
1054	struct coda_ctx *ctx = priv;
1055
1056	ctx->aborting = 1;
1057
1058	v4l2_dbg(1, coda_debug, &ctx->dev->v4l2_dev,
1059		 "Aborting task\n");
1060}
1061
1062static void coda_lock(void *m2m_priv)
1063{
1064	struct coda_ctx *ctx = m2m_priv;
1065	struct coda_dev *pcdev = ctx->dev;
1066
1067	mutex_lock(&pcdev->dev_mutex);
1068}
1069
1070static void coda_unlock(void *m2m_priv)
1071{
1072	struct coda_ctx *ctx = m2m_priv;
1073	struct coda_dev *pcdev = ctx->dev;
1074
1075	mutex_unlock(&pcdev->dev_mutex);
1076}
1077
1078static const struct v4l2_m2m_ops coda_m2m_ops = {
1079	.device_run	= coda_device_run,
1080	.job_ready	= coda_job_ready,
1081	.job_abort	= coda_job_abort,
1082	.lock		= coda_lock,
1083	.unlock		= coda_unlock,
1084};
1085
1086static void set_default_params(struct coda_ctx *ctx)
1087{
1088	unsigned int max_w, max_h, usize, csize;
1089
1090	ctx->codec = coda_find_codec(ctx->dev, ctx->cvd->src_formats[0],
1091				     ctx->cvd->dst_formats[0]);
1092	max_w = min(ctx->codec->max_w, 1920U);
1093	max_h = min(ctx->codec->max_h, 1088U);
1094	usize = max_w * max_h * 3 / 2;
1095	csize = coda_estimate_sizeimage(ctx, usize, max_w, max_h);
1096
1097	ctx->params.codec_mode = ctx->codec->mode;
1098	ctx->colorspace = V4L2_COLORSPACE_REC709;
1099	ctx->params.framerate = 30;
1100
1101	/* Default formats for output and input queues */
1102	ctx->q_data[V4L2_M2M_SRC].fourcc = ctx->cvd->src_formats[0];
1103	ctx->q_data[V4L2_M2M_DST].fourcc = ctx->cvd->dst_formats[0];
1104	ctx->q_data[V4L2_M2M_SRC].width = max_w;
1105	ctx->q_data[V4L2_M2M_SRC].height = max_h;
1106	ctx->q_data[V4L2_M2M_DST].width = max_w;
1107	ctx->q_data[V4L2_M2M_DST].height = max_h;
1108	if (ctx->codec->src_fourcc == V4L2_PIX_FMT_YUV420) {
1109		ctx->q_data[V4L2_M2M_SRC].bytesperline = max_w;
1110		ctx->q_data[V4L2_M2M_SRC].sizeimage = usize;
1111		ctx->q_data[V4L2_M2M_DST].bytesperline = 0;
1112		ctx->q_data[V4L2_M2M_DST].sizeimage = csize;
1113	} else {
1114		ctx->q_data[V4L2_M2M_SRC].bytesperline = 0;
1115		ctx->q_data[V4L2_M2M_SRC].sizeimage = csize;
1116		ctx->q_data[V4L2_M2M_DST].bytesperline = max_w;
1117		ctx->q_data[V4L2_M2M_DST].sizeimage = usize;
1118	}
1119	ctx->q_data[V4L2_M2M_SRC].rect.width = max_w;
1120	ctx->q_data[V4L2_M2M_SRC].rect.height = max_h;
1121	ctx->q_data[V4L2_M2M_DST].rect.width = max_w;
1122	ctx->q_data[V4L2_M2M_DST].rect.height = max_h;
1123
1124	/*
1125	 * Since the RBC2AXI logic only supports a single chroma plane,
1126	 * macroblock tiling only works for to NV12 pixel format.
1127	 */
1128	ctx->tiled_map_type = GDI_LINEAR_FRAME_MAP;
1129}
1130
1131/*
1132 * Queue operations
1133 */
1134static int coda_queue_setup(struct vb2_queue *vq, const void *parg,
1135				unsigned int *nbuffers, unsigned int *nplanes,
1136				unsigned int sizes[], void *alloc_ctxs[])
1137{
1138	struct coda_ctx *ctx = vb2_get_drv_priv(vq);
1139	struct coda_q_data *q_data;
1140	unsigned int size;
1141
1142	q_data = get_q_data(ctx, vq->type);
1143	size = q_data->sizeimage;
1144
1145	*nplanes = 1;
1146	sizes[0] = size;
1147
1148	/* Set to vb2-dma-contig allocator context, ignored by vb2-vmalloc */
1149	alloc_ctxs[0] = ctx->dev->alloc_ctx;
1150
1151	v4l2_dbg(1, coda_debug, &ctx->dev->v4l2_dev,
1152		 "get %d buffer(s) of size %d each.\n", *nbuffers, size);
1153
1154	return 0;
1155}
1156
1157static int coda_buf_prepare(struct vb2_buffer *vb)
1158{
1159	struct coda_ctx *ctx = vb2_get_drv_priv(vb->vb2_queue);
1160	struct coda_q_data *q_data;
1161
1162	q_data = get_q_data(ctx, vb->vb2_queue->type);
1163
1164	if (vb2_plane_size(vb, 0) < q_data->sizeimage) {
1165		v4l2_warn(&ctx->dev->v4l2_dev,
1166			  "%s data will not fit into plane (%lu < %lu)\n",
1167			  __func__, vb2_plane_size(vb, 0),
1168			  (long)q_data->sizeimage);
1169		return -EINVAL;
1170	}
1171
1172	return 0;
1173}
1174
1175static void coda_buf_queue(struct vb2_buffer *vb)
1176{
1177	struct vb2_v4l2_buffer *vbuf = to_vb2_v4l2_buffer(vb);
1178	struct coda_ctx *ctx = vb2_get_drv_priv(vb->vb2_queue);
1179	struct vb2_queue *vq = vb->vb2_queue;
1180	struct coda_q_data *q_data;
1181
1182	q_data = get_q_data(ctx, vb->vb2_queue->type);
1183
1184	/*
1185	 * In the decoder case, immediately try to copy the buffer into the
1186	 * bitstream ringbuffer and mark it as ready to be dequeued.
1187	 */
1188	if (ctx->bitstream.size && vq->type == V4L2_BUF_TYPE_VIDEO_OUTPUT) {
1189		/*
1190		 * For backwards compatibility, queuing an empty buffer marks
1191		 * the stream end
1192		 */
1193		if (vb2_get_plane_payload(vb, 0) == 0)
1194			coda_bit_stream_end_flag(ctx);
1195		mutex_lock(&ctx->bitstream_mutex);
1196		v4l2_m2m_buf_queue(ctx->fh.m2m_ctx, vbuf);
1197		if (vb2_is_streaming(vb->vb2_queue))
1198			coda_fill_bitstream(ctx, true);
1199		mutex_unlock(&ctx->bitstream_mutex);
1200	} else {
1201		v4l2_m2m_buf_queue(ctx->fh.m2m_ctx, vbuf);
1202	}
1203}
1204
1205int coda_alloc_aux_buf(struct coda_dev *dev, struct coda_aux_buf *buf,
1206		       size_t size, const char *name, struct dentry *parent)
1207{
1208	buf->vaddr = dma_alloc_coherent(&dev->plat_dev->dev, size, &buf->paddr,
1209					GFP_KERNEL);
1210	if (!buf->vaddr) {
1211		v4l2_err(&dev->v4l2_dev,
1212			 "Failed to allocate %s buffer of size %u\n",
1213			 name, size);
1214		return -ENOMEM;
1215	}
1216
1217	buf->size = size;
1218
1219	if (name && parent) {
1220		buf->blob.data = buf->vaddr;
1221		buf->blob.size = size;
1222		buf->dentry = debugfs_create_blob(name, 0644, parent,
1223						  &buf->blob);
1224		if (!buf->dentry)
1225			dev_warn(&dev->plat_dev->dev,
1226				 "failed to create debugfs entry %s\n", name);
1227	}
1228
1229	return 0;
1230}
1231
1232void coda_free_aux_buf(struct coda_dev *dev,
1233		       struct coda_aux_buf *buf)
1234{
1235	if (buf->vaddr) {
1236		dma_free_coherent(&dev->plat_dev->dev, buf->size,
1237				  buf->vaddr, buf->paddr);
1238		buf->vaddr = NULL;
1239		buf->size = 0;
1240		debugfs_remove(buf->dentry);
1241		buf->dentry = NULL;
1242	}
1243}
1244
1245static int coda_start_streaming(struct vb2_queue *q, unsigned int count)
1246{
1247	struct coda_ctx *ctx = vb2_get_drv_priv(q);
1248	struct v4l2_device *v4l2_dev = &ctx->dev->v4l2_dev;
1249	struct coda_q_data *q_data_src, *q_data_dst;
1250	struct vb2_v4l2_buffer *buf;
1251	int ret = 0;
1252
1253	q_data_src = get_q_data(ctx, V4L2_BUF_TYPE_VIDEO_OUTPUT);
1254	if (q->type == V4L2_BUF_TYPE_VIDEO_OUTPUT) {
1255		if (ctx->inst_type == CODA_INST_DECODER && ctx->use_bit) {
1256			/* copy the buffers that were queued before streamon */
1257			mutex_lock(&ctx->bitstream_mutex);
1258			coda_fill_bitstream(ctx, false);
1259			mutex_unlock(&ctx->bitstream_mutex);
1260
1261			if (coda_get_bitstream_payload(ctx) < 512) {
1262				ret = -EINVAL;
1263				goto err;
1264			}
1265		} else {
1266			if (count < 1) {
1267				ret = -EINVAL;
1268				goto err;
1269			}
1270		}
1271
1272		ctx->streamon_out = 1;
1273	} else {
1274		if (count < 1) {
1275			ret = -EINVAL;
1276			goto err;
1277		}
1278
1279		ctx->streamon_cap = 1;
1280	}
1281
1282	/* Don't start the coda unless both queues are on */
1283	if (!(ctx->streamon_out & ctx->streamon_cap))
1284		return 0;
1285
1286	q_data_dst = get_q_data(ctx, V4L2_BUF_TYPE_VIDEO_CAPTURE);
1287	if ((q_data_src->width != q_data_dst->width &&
1288	     round_up(q_data_src->width, 16) != q_data_dst->width) ||
1289	    (q_data_src->height != q_data_dst->height &&
1290	     round_up(q_data_src->height, 16) != q_data_dst->height)) {
1291		v4l2_err(v4l2_dev, "can't convert %dx%d to %dx%d\n",
1292			 q_data_src->width, q_data_src->height,
1293			 q_data_dst->width, q_data_dst->height);
1294		ret = -EINVAL;
1295		goto err;
1296	}
1297
1298	/* Allow BIT decoder device_run with no new buffers queued */
1299	if (ctx->inst_type == CODA_INST_DECODER && ctx->use_bit)
1300		v4l2_m2m_set_src_buffered(ctx->fh.m2m_ctx, true);
1301
1302	ctx->gopcounter = ctx->params.gop_size - 1;
1303
1304	ctx->codec = coda_find_codec(ctx->dev, q_data_src->fourcc,
1305				     q_data_dst->fourcc);
1306	if (!ctx->codec) {
1307		v4l2_err(v4l2_dev, "couldn't tell instance type.\n");
1308		ret = -EINVAL;
1309		goto err;
1310	}
1311
1312	if (q_data_dst->fourcc == V4L2_PIX_FMT_JPEG)
1313		ctx->params.gop_size = 1;
1314	ctx->gopcounter = ctx->params.gop_size - 1;
1315
1316	ret = ctx->ops->start_streaming(ctx);
1317	if (ctx->inst_type == CODA_INST_DECODER) {
1318		if (ret == -EAGAIN)
1319			return 0;
1320		else if (ret < 0)
1321			goto err;
1322	}
1323
1324	return ret;
1325
1326err:
1327	if (q->type == V4L2_BUF_TYPE_VIDEO_OUTPUT) {
1328		while ((buf = v4l2_m2m_src_buf_remove(ctx->fh.m2m_ctx)))
1329			v4l2_m2m_buf_done(buf, VB2_BUF_STATE_QUEUED);
1330	} else {
1331		while ((buf = v4l2_m2m_dst_buf_remove(ctx->fh.m2m_ctx)))
1332			v4l2_m2m_buf_done(buf, VB2_BUF_STATE_QUEUED);
1333	}
1334	return ret;
1335}
1336
1337static void coda_stop_streaming(struct vb2_queue *q)
1338{
1339	struct coda_ctx *ctx = vb2_get_drv_priv(q);
1340	struct coda_dev *dev = ctx->dev;
1341	struct vb2_v4l2_buffer *buf;
1342	unsigned long flags;
1343	bool stop;
1344
1345	stop = ctx->streamon_out && ctx->streamon_cap;
1346
1347	if (q->type == V4L2_BUF_TYPE_VIDEO_OUTPUT) {
1348		v4l2_dbg(1, coda_debug, &dev->v4l2_dev,
1349			 "%s: output\n", __func__);
1350		ctx->streamon_out = 0;
1351
1352		coda_bit_stream_end_flag(ctx);
1353
1354		ctx->qsequence = 0;
1355
1356		while ((buf = v4l2_m2m_src_buf_remove(ctx->fh.m2m_ctx)))
1357			v4l2_m2m_buf_done(buf, VB2_BUF_STATE_ERROR);
1358	} else {
1359		v4l2_dbg(1, coda_debug, &dev->v4l2_dev,
1360			 "%s: capture\n", __func__);
1361		ctx->streamon_cap = 0;
1362
1363		ctx->osequence = 0;
1364		ctx->sequence_offset = 0;
1365
1366		while ((buf = v4l2_m2m_dst_buf_remove(ctx->fh.m2m_ctx)))
1367			v4l2_m2m_buf_done(buf, VB2_BUF_STATE_ERROR);
1368	}
1369
1370	if (stop) {
1371		struct coda_buffer_meta *meta;
1372
1373		if (ctx->ops->seq_end_work) {
1374			queue_work(dev->workqueue, &ctx->seq_end_work);
1375			flush_work(&ctx->seq_end_work);
1376		}
1377		spin_lock_irqsave(&ctx->buffer_meta_lock, flags);
1378		while (!list_empty(&ctx->buffer_meta_list)) {
1379			meta = list_first_entry(&ctx->buffer_meta_list,
1380						struct coda_buffer_meta, list);
1381			list_del(&meta->list);
1382			kfree(meta);
1383		}
1384		ctx->num_metas = 0;
1385		spin_unlock_irqrestore(&ctx->buffer_meta_lock, flags);
1386		kfifo_init(&ctx->bitstream_fifo,
1387			ctx->bitstream.vaddr, ctx->bitstream.size);
1388		ctx->runcounter = 0;
1389		ctx->aborting = 0;
1390	}
1391
1392	if (!ctx->streamon_out && !ctx->streamon_cap)
1393		ctx->bit_stream_param &= ~CODA_BIT_STREAM_END_FLAG;
1394}
1395
1396static const struct vb2_ops coda_qops = {
1397	.queue_setup		= coda_queue_setup,
1398	.buf_prepare		= coda_buf_prepare,
1399	.buf_queue		= coda_buf_queue,
1400	.start_streaming	= coda_start_streaming,
1401	.stop_streaming		= coda_stop_streaming,
1402	.wait_prepare		= vb2_ops_wait_prepare,
1403	.wait_finish		= vb2_ops_wait_finish,
1404};
1405
1406static int coda_s_ctrl(struct v4l2_ctrl *ctrl)
1407{
1408	struct coda_ctx *ctx =
1409			container_of(ctrl->handler, struct coda_ctx, ctrls);
1410
1411	v4l2_dbg(1, coda_debug, &ctx->dev->v4l2_dev,
1412		 "s_ctrl: id = %d, val = %d\n", ctrl->id, ctrl->val);
1413
1414	switch (ctrl->id) {
1415	case V4L2_CID_HFLIP:
1416		if (ctrl->val)
1417			ctx->params.rot_mode |= CODA_MIR_HOR;
1418		else
1419			ctx->params.rot_mode &= ~CODA_MIR_HOR;
1420		break;
1421	case V4L2_CID_VFLIP:
1422		if (ctrl->val)
1423			ctx->params.rot_mode |= CODA_MIR_VER;
1424		else
1425			ctx->params.rot_mode &= ~CODA_MIR_VER;
1426		break;
1427	case V4L2_CID_MPEG_VIDEO_BITRATE:
1428		ctx->params.bitrate = ctrl->val / 1000;
1429		break;
1430	case V4L2_CID_MPEG_VIDEO_GOP_SIZE:
1431		ctx->params.gop_size = ctrl->val;
1432		break;
1433	case V4L2_CID_MPEG_VIDEO_H264_I_FRAME_QP:
1434		ctx->params.h264_intra_qp = ctrl->val;
1435		break;
1436	case V4L2_CID_MPEG_VIDEO_H264_P_FRAME_QP:
1437		ctx->params.h264_inter_qp = ctrl->val;
1438		break;
1439	case V4L2_CID_MPEG_VIDEO_H264_MIN_QP:
1440		ctx->params.h264_min_qp = ctrl->val;
1441		break;
1442	case V4L2_CID_MPEG_VIDEO_H264_MAX_QP:
1443		ctx->params.h264_max_qp = ctrl->val;
1444		break;
1445	case V4L2_CID_MPEG_VIDEO_H264_LOOP_FILTER_ALPHA:
1446		ctx->params.h264_deblk_alpha = ctrl->val;
1447		break;
1448	case V4L2_CID_MPEG_VIDEO_H264_LOOP_FILTER_BETA:
1449		ctx->params.h264_deblk_beta = ctrl->val;
1450		break;
1451	case V4L2_CID_MPEG_VIDEO_H264_LOOP_FILTER_MODE:
1452		ctx->params.h264_deblk_enabled = (ctrl->val ==
1453				V4L2_MPEG_VIDEO_H264_LOOP_FILTER_MODE_ENABLED);
1454		break;
1455	case V4L2_CID_MPEG_VIDEO_MPEG4_I_FRAME_QP:
1456		ctx->params.mpeg4_intra_qp = ctrl->val;
1457		break;
1458	case V4L2_CID_MPEG_VIDEO_MPEG4_P_FRAME_QP:
1459		ctx->params.mpeg4_inter_qp = ctrl->val;
1460		break;
1461	case V4L2_CID_MPEG_VIDEO_MULTI_SLICE_MODE:
1462		ctx->params.slice_mode = ctrl->val;
1463		break;
1464	case V4L2_CID_MPEG_VIDEO_MULTI_SLICE_MAX_MB:
1465		ctx->params.slice_max_mb = ctrl->val;
1466		break;
1467	case V4L2_CID_MPEG_VIDEO_MULTI_SLICE_MAX_BYTES:
1468		ctx->params.slice_max_bits = ctrl->val * 8;
1469		break;
1470	case V4L2_CID_MPEG_VIDEO_HEADER_MODE:
1471		break;
1472	case V4L2_CID_MPEG_VIDEO_CYCLIC_INTRA_REFRESH_MB:
1473		ctx->params.intra_refresh = ctrl->val;
1474		break;
1475	case V4L2_CID_JPEG_COMPRESSION_QUALITY:
1476		coda_set_jpeg_compression_quality(ctx, ctrl->val);
1477		break;
1478	case V4L2_CID_JPEG_RESTART_INTERVAL:
1479		ctx->params.jpeg_restart_interval = ctrl->val;
1480		break;
1481	case V4L2_CID_MPEG_VIDEO_VBV_DELAY:
1482		ctx->params.vbv_delay = ctrl->val;
1483		break;
1484	case V4L2_CID_MPEG_VIDEO_VBV_SIZE:
1485		ctx->params.vbv_size = min(ctrl->val * 8192, 0x7fffffff);
1486		break;
1487	default:
1488		v4l2_dbg(1, coda_debug, &ctx->dev->v4l2_dev,
1489			"Invalid control, id=%d, val=%d\n",
1490			ctrl->id, ctrl->val);
1491		return -EINVAL;
1492	}
1493
1494	return 0;
1495}
1496
1497static const struct v4l2_ctrl_ops coda_ctrl_ops = {
1498	.s_ctrl = coda_s_ctrl,
1499};
1500
1501static void coda_encode_ctrls(struct coda_ctx *ctx)
1502{
1503	v4l2_ctrl_new_std(&ctx->ctrls, &coda_ctrl_ops,
1504		V4L2_CID_MPEG_VIDEO_BITRATE, 0, 32767000, 1000, 0);
1505	v4l2_ctrl_new_std(&ctx->ctrls, &coda_ctrl_ops,
1506		V4L2_CID_MPEG_VIDEO_GOP_SIZE, 1, 60, 1, 16);
1507	v4l2_ctrl_new_std(&ctx->ctrls, &coda_ctrl_ops,
1508		V4L2_CID_MPEG_VIDEO_H264_I_FRAME_QP, 0, 51, 1, 25);
1509	v4l2_ctrl_new_std(&ctx->ctrls, &coda_ctrl_ops,
1510		V4L2_CID_MPEG_VIDEO_H264_P_FRAME_QP, 0, 51, 1, 25);
1511	if (ctx->dev->devtype->product != CODA_960) {
1512		v4l2_ctrl_new_std(&ctx->ctrls, &coda_ctrl_ops,
1513			V4L2_CID_MPEG_VIDEO_H264_MIN_QP, 0, 51, 1, 12);
1514	}
1515	v4l2_ctrl_new_std(&ctx->ctrls, &coda_ctrl_ops,
1516		V4L2_CID_MPEG_VIDEO_H264_MAX_QP, 0, 51, 1, 51);
1517	v4l2_ctrl_new_std(&ctx->ctrls, &coda_ctrl_ops,
1518		V4L2_CID_MPEG_VIDEO_H264_LOOP_FILTER_ALPHA, 0, 15, 1, 0);
1519	v4l2_ctrl_new_std(&ctx->ctrls, &coda_ctrl_ops,
1520		V4L2_CID_MPEG_VIDEO_H264_LOOP_FILTER_BETA, 0, 15, 1, 0);
1521	v4l2_ctrl_new_std_menu(&ctx->ctrls, &coda_ctrl_ops,
1522		V4L2_CID_MPEG_VIDEO_H264_LOOP_FILTER_MODE,
1523		V4L2_MPEG_VIDEO_H264_LOOP_FILTER_MODE_DISABLED, 0x0,
1524		V4L2_MPEG_VIDEO_H264_LOOP_FILTER_MODE_ENABLED);
1525	v4l2_ctrl_new_std(&ctx->ctrls, &coda_ctrl_ops,
1526		V4L2_CID_MPEG_VIDEO_MPEG4_I_FRAME_QP, 1, 31, 1, 2);
1527	v4l2_ctrl_new_std(&ctx->ctrls, &coda_ctrl_ops,
1528		V4L2_CID_MPEG_VIDEO_MPEG4_P_FRAME_QP, 1, 31, 1, 2);
1529	v4l2_ctrl_new_std_menu(&ctx->ctrls, &coda_ctrl_ops,
1530		V4L2_CID_MPEG_VIDEO_MULTI_SLICE_MODE,
1531		V4L2_MPEG_VIDEO_MULTI_SICE_MODE_MAX_BYTES, 0x0,
1532		V4L2_MPEG_VIDEO_MULTI_SLICE_MODE_SINGLE);
1533	v4l2_ctrl_new_std(&ctx->ctrls, &coda_ctrl_ops,
1534		V4L2_CID_MPEG_VIDEO_MULTI_SLICE_MAX_MB, 1, 0x3fffffff, 1, 1);
1535	v4l2_ctrl_new_std(&ctx->ctrls, &coda_ctrl_ops,
1536		V4L2_CID_MPEG_VIDEO_MULTI_SLICE_MAX_BYTES, 1, 0x3fffffff, 1,
1537		500);
1538	v4l2_ctrl_new_std_menu(&ctx->ctrls, &coda_ctrl_ops,
1539		V4L2_CID_MPEG_VIDEO_HEADER_MODE,
1540		V4L2_MPEG_VIDEO_HEADER_MODE_JOINED_WITH_1ST_FRAME,
1541		(1 << V4L2_MPEG_VIDEO_HEADER_MODE_SEPARATE),
1542		V4L2_MPEG_VIDEO_HEADER_MODE_JOINED_WITH_1ST_FRAME);
1543	v4l2_ctrl_new_std(&ctx->ctrls, &coda_ctrl_ops,
1544		V4L2_CID_MPEG_VIDEO_CYCLIC_INTRA_REFRESH_MB, 0,
1545		1920 * 1088 / 256, 1, 0);
1546	v4l2_ctrl_new_std(&ctx->ctrls, &coda_ctrl_ops,
1547		V4L2_CID_MPEG_VIDEO_VBV_DELAY, 0, 0x7fff, 1, 0);
1548	/*
1549	 * The maximum VBV size value is 0x7fffffff bits,
1550	 * one bit less than 262144 KiB
1551	 */
1552	v4l2_ctrl_new_std(&ctx->ctrls, &coda_ctrl_ops,
1553		V4L2_CID_MPEG_VIDEO_VBV_SIZE, 0, 262144, 1, 0);
1554}
1555
1556static void coda_jpeg_encode_ctrls(struct coda_ctx *ctx)
1557{
1558	v4l2_ctrl_new_std(&ctx->ctrls, &coda_ctrl_ops,
1559		V4L2_CID_JPEG_COMPRESSION_QUALITY, 5, 100, 1, 50);
1560	v4l2_ctrl_new_std(&ctx->ctrls, &coda_ctrl_ops,
1561		V4L2_CID_JPEG_RESTART_INTERVAL, 0, 100, 1, 0);
1562}
1563
1564static int coda_ctrls_setup(struct coda_ctx *ctx)
1565{
1566	v4l2_ctrl_handler_init(&ctx->ctrls, 2);
1567
1568	v4l2_ctrl_new_std(&ctx->ctrls, &coda_ctrl_ops,
1569		V4L2_CID_HFLIP, 0, 1, 1, 0);
1570	v4l2_ctrl_new_std(&ctx->ctrls, &coda_ctrl_ops,
1571		V4L2_CID_VFLIP, 0, 1, 1, 0);
1572	if (ctx->inst_type == CODA_INST_ENCODER) {
1573		if (ctx->cvd->dst_formats[0] == V4L2_PIX_FMT_JPEG)
1574			coda_jpeg_encode_ctrls(ctx);
1575		else
1576			coda_encode_ctrls(ctx);
1577	}
1578
1579	if (ctx->ctrls.error) {
1580		v4l2_err(&ctx->dev->v4l2_dev,
1581			"control initialization error (%d)",
1582			ctx->ctrls.error);
1583		return -EINVAL;
1584	}
1585
1586	return v4l2_ctrl_handler_setup(&ctx->ctrls);
1587}
1588
1589static int coda_queue_init(struct coda_ctx *ctx, struct vb2_queue *vq)
1590{
1591	vq->drv_priv = ctx;
1592	vq->ops = &coda_qops;
1593	vq->buf_struct_size = sizeof(struct v4l2_m2m_buffer);
1594	vq->timestamp_flags = V4L2_BUF_FLAG_TIMESTAMP_COPY;
1595	vq->lock = &ctx->dev->dev_mutex;
1596	/* One way to indicate end-of-stream for coda is to set the
1597	 * bytesused == 0. However by default videobuf2 handles bytesused
1598	 * equal to 0 as a special case and changes its value to the size
1599	 * of the buffer. Set the allow_zero_bytesused flag, so
1600	 * that videobuf2 will keep the value of bytesused intact.
1601	 */
1602	vq->allow_zero_bytesused = 1;
1603
1604	return vb2_queue_init(vq);
1605}
1606
1607int coda_encoder_queue_init(void *priv, struct vb2_queue *src_vq,
1608			    struct vb2_queue *dst_vq)
1609{
1610	int ret;
1611
1612	src_vq->type = V4L2_BUF_TYPE_VIDEO_OUTPUT;
1613	src_vq->io_modes = VB2_DMABUF | VB2_MMAP;
1614	src_vq->mem_ops = &vb2_dma_contig_memops;
1615
1616	ret = coda_queue_init(priv, src_vq);
1617	if (ret)
1618		return ret;
1619
1620	dst_vq->type = V4L2_BUF_TYPE_VIDEO_CAPTURE;
1621	dst_vq->io_modes = VB2_DMABUF | VB2_MMAP;
1622	dst_vq->mem_ops = &vb2_dma_contig_memops;
1623
1624	return coda_queue_init(priv, dst_vq);
1625}
1626
1627int coda_decoder_queue_init(void *priv, struct vb2_queue *src_vq,
1628			    struct vb2_queue *dst_vq)
1629{
1630	int ret;
1631
1632	src_vq->type = V4L2_BUF_TYPE_VIDEO_OUTPUT;
1633	src_vq->io_modes = VB2_DMABUF | VB2_MMAP | VB2_USERPTR;
1634	src_vq->mem_ops = &vb2_vmalloc_memops;
1635
1636	ret = coda_queue_init(priv, src_vq);
1637	if (ret)
1638		return ret;
1639
1640	dst_vq->type = V4L2_BUF_TYPE_VIDEO_CAPTURE;
1641	dst_vq->io_modes = VB2_DMABUF | VB2_MMAP;
1642	dst_vq->mem_ops = &vb2_dma_contig_memops;
1643
1644	return coda_queue_init(priv, dst_vq);
1645}
1646
1647static int coda_next_free_instance(struct coda_dev *dev)
1648{
1649	int idx = ffz(dev->instance_mask);
1650
1651	if ((idx < 0) ||
1652	    (dev->devtype->product == CODA_DX6 && idx > CODADX6_MAX_INSTANCES))
1653		return -EBUSY;
1654
1655	return idx;
1656}
1657
1658/*
1659 * File operations
1660 */
1661
1662static int coda_open(struct file *file)
1663{
1664	struct video_device *vdev = video_devdata(file);
1665	struct coda_dev *dev = video_get_drvdata(vdev);
1666	struct coda_ctx *ctx = NULL;
1667	char *name;
1668	int ret;
1669	int idx;
1670
1671	ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
1672	if (!ctx)
1673		return -ENOMEM;
1674
1675	idx = coda_next_free_instance(dev);
1676	if (idx < 0) {
1677		ret = idx;
1678		goto err_coda_max;
1679	}
1680	set_bit(idx, &dev->instance_mask);
1681
1682	name = kasprintf(GFP_KERNEL, "context%d", idx);
1683	if (!name) {
1684		ret = -ENOMEM;
1685		goto err_coda_name_init;
1686	}
1687
1688	ctx->debugfs_entry = debugfs_create_dir(name, dev->debugfs_root);
1689	kfree(name);
1690
1691	ctx->cvd = to_coda_video_device(vdev);
1692	ctx->inst_type = ctx->cvd->type;
1693	ctx->ops = ctx->cvd->ops;
1694	ctx->use_bit = !ctx->cvd->direct;
1695	init_completion(&ctx->completion);
1696	INIT_WORK(&ctx->pic_run_work, coda_pic_run_work);
1697	if (ctx->ops->seq_end_work)
1698		INIT_WORK(&ctx->seq_end_work, ctx->ops->seq_end_work);
1699	v4l2_fh_init(&ctx->fh, video_devdata(file));
1700	file->private_data = &ctx->fh;
1701	v4l2_fh_add(&ctx->fh);
1702	ctx->dev = dev;
1703	ctx->idx = idx;
1704	switch (dev->devtype->product) {
1705	case CODA_960:
1706		ctx->frame_mem_ctrl = 1 << 12;
1707		/* fallthrough */
1708	case CODA_7541:
1709		ctx->reg_idx = 0;
1710		break;
1711	default:
1712		ctx->reg_idx = idx;
1713	}
1714
1715	/* Power up and upload firmware if necessary */
1716	ret = pm_runtime_get_sync(&dev->plat_dev->dev);
1717	if (ret < 0) {
1718		v4l2_err(&dev->v4l2_dev, "failed to power up: %d\n", ret);
1719		goto err_pm_get;
1720	}
1721
1722	ret = clk_prepare_enable(dev->clk_per);
1723	if (ret)
1724		goto err_clk_per;
1725
1726	ret = clk_prepare_enable(dev->clk_ahb);
1727	if (ret)
1728		goto err_clk_ahb;
1729
1730	set_default_params(ctx);
1731	ctx->fh.m2m_ctx = v4l2_m2m_ctx_init(dev->m2m_dev, ctx,
1732					    ctx->ops->queue_init);
1733	if (IS_ERR(ctx->fh.m2m_ctx)) {
1734		ret = PTR_ERR(ctx->fh.m2m_ctx);
1735
1736		v4l2_err(&dev->v4l2_dev, "%s return error (%d)\n",
1737			 __func__, ret);
1738		goto err_ctx_init;
1739	}
1740
1741	ret = coda_ctrls_setup(ctx);
1742	if (ret) {
1743		v4l2_err(&dev->v4l2_dev, "failed to setup coda controls\n");
1744		goto err_ctrls_setup;
1745	}
1746
1747	ctx->fh.ctrl_handler = &ctx->ctrls;
1748
1749	mutex_init(&ctx->bitstream_mutex);
1750	mutex_init(&ctx->buffer_mutex);
1751	INIT_LIST_HEAD(&ctx->buffer_meta_list);
1752	spin_lock_init(&ctx->buffer_meta_lock);
1753
1754	coda_lock(ctx);
1755	list_add(&ctx->list, &dev->instances);
1756	coda_unlock(ctx);
1757
1758	v4l2_dbg(1, coda_debug, &dev->v4l2_dev, "Created instance %d (%p)\n",
1759		 ctx->idx, ctx);
1760
1761	return 0;
1762
1763err_ctrls_setup:
1764	v4l2_m2m_ctx_release(ctx->fh.m2m_ctx);
1765err_ctx_init:
1766	clk_disable_unprepare(dev->clk_ahb);
1767err_clk_ahb:
1768	clk_disable_unprepare(dev->clk_per);
1769err_clk_per:
1770	pm_runtime_put_sync(&dev->plat_dev->dev);
1771err_pm_get:
1772	v4l2_fh_del(&ctx->fh);
1773	v4l2_fh_exit(&ctx->fh);
1774	clear_bit(ctx->idx, &dev->instance_mask);
1775err_coda_name_init:
1776err_coda_max:
1777	kfree(ctx);
1778	return ret;
1779}
1780
1781static int coda_release(struct file *file)
1782{
1783	struct coda_dev *dev = video_drvdata(file);
1784	struct coda_ctx *ctx = fh_to_ctx(file->private_data);
1785
1786	v4l2_dbg(1, coda_debug, &dev->v4l2_dev, "Releasing instance %p\n",
1787		 ctx);
1788
1789	if (ctx->inst_type == CODA_INST_DECODER && ctx->use_bit)
1790		coda_bit_stream_end_flag(ctx);
1791
1792	/* If this instance is running, call .job_abort and wait for it to end */
1793	v4l2_m2m_ctx_release(ctx->fh.m2m_ctx);
1794
1795	/* In case the instance was not running, we still need to call SEQ_END */
1796	if (ctx->ops->seq_end_work) {
1797		queue_work(dev->workqueue, &ctx->seq_end_work);
1798		flush_work(&ctx->seq_end_work);
1799	}
1800
1801	coda_lock(ctx);
1802	list_del(&ctx->list);
1803	coda_unlock(ctx);
1804
1805	if (ctx->dev->devtype->product == CODA_DX6)
1806		coda_free_aux_buf(dev, &ctx->workbuf);
1807
1808	v4l2_ctrl_handler_free(&ctx->ctrls);
1809	clk_disable_unprepare(dev->clk_ahb);
1810	clk_disable_unprepare(dev->clk_per);
1811	pm_runtime_put_sync(&dev->plat_dev->dev);
1812	v4l2_fh_del(&ctx->fh);
1813	v4l2_fh_exit(&ctx->fh);
1814	clear_bit(ctx->idx, &dev->instance_mask);
1815	if (ctx->ops->release)
1816		ctx->ops->release(ctx);
1817	debugfs_remove_recursive(ctx->debugfs_entry);
1818	kfree(ctx);
1819
1820	return 0;
1821}
1822
1823static const struct v4l2_file_operations coda_fops = {
1824	.owner		= THIS_MODULE,
1825	.open		= coda_open,
1826	.release	= coda_release,
1827	.poll		= v4l2_m2m_fop_poll,
1828	.unlocked_ioctl	= video_ioctl2,
1829	.mmap		= v4l2_m2m_fop_mmap,
1830};
1831
1832static int coda_hw_init(struct coda_dev *dev)
1833{
1834	u32 data;
1835	u16 *p;
1836	int i, ret;
1837
1838	ret = clk_prepare_enable(dev->clk_per);
1839	if (ret)
1840		goto err_clk_per;
1841
1842	ret = clk_prepare_enable(dev->clk_ahb);
1843	if (ret)
1844		goto err_clk_ahb;
1845
1846	if (dev->rstc)
1847		reset_control_reset(dev->rstc);
1848
1849	/*
1850	 * Copy the first CODA_ISRAM_SIZE in the internal SRAM.
1851	 * The 16-bit chars in the code buffer are in memory access
1852	 * order, re-sort them to CODA order for register download.
1853	 * Data in this SRAM survives a reboot.
1854	 */
1855	p = (u16 *)dev->codebuf.vaddr;
1856	if (dev->devtype->product == CODA_DX6) {
1857		for (i = 0; i < (CODA_ISRAM_SIZE / 2); i++)  {
1858			data = CODA_DOWN_ADDRESS_SET(i) |
1859				CODA_DOWN_DATA_SET(p[i ^ 1]);
1860			coda_write(dev, data, CODA_REG_BIT_CODE_DOWN);
1861		}
1862	} else {
1863		for (i = 0; i < (CODA_ISRAM_SIZE / 2); i++) {
1864			data = CODA_DOWN_ADDRESS_SET(i) |
1865				CODA_DOWN_DATA_SET(p[round_down(i, 4) +
1866							3 - (i % 4)]);
1867			coda_write(dev, data, CODA_REG_BIT_CODE_DOWN);
1868		}
1869	}
1870
1871	/* Clear registers */
1872	for (i = 0; i < 64; i++)
1873		coda_write(dev, 0, CODA_REG_BIT_CODE_BUF_ADDR + i * 4);
1874
1875	/* Tell the BIT where to find everything it needs */
1876	if (dev->devtype->product == CODA_960 ||
1877	    dev->devtype->product == CODA_7541) {
1878		coda_write(dev, dev->tempbuf.paddr,
1879				CODA_REG_BIT_TEMP_BUF_ADDR);
1880		coda_write(dev, 0, CODA_REG_BIT_BIT_STREAM_PARAM);
1881	} else {
1882		coda_write(dev, dev->workbuf.paddr,
1883			      CODA_REG_BIT_WORK_BUF_ADDR);
1884	}
1885	coda_write(dev, dev->codebuf.paddr,
1886		      CODA_REG_BIT_CODE_BUF_ADDR);
1887	coda_write(dev, 0, CODA_REG_BIT_CODE_RUN);
1888
1889	/* Set default values */
1890	switch (dev->devtype->product) {
1891	case CODA_DX6:
1892		coda_write(dev, CODADX6_STREAM_BUF_PIC_FLUSH,
1893			   CODA_REG_BIT_STREAM_CTRL);
1894		break;
1895	default:
1896		coda_write(dev, CODA7_STREAM_BUF_PIC_FLUSH,
1897			   CODA_REG_BIT_STREAM_CTRL);
1898	}
1899	if (dev->devtype->product == CODA_960)
1900		coda_write(dev, 1 << 12, CODA_REG_BIT_FRAME_MEM_CTRL);
1901	else
1902		coda_write(dev, 0, CODA_REG_BIT_FRAME_MEM_CTRL);
1903
1904	if (dev->devtype->product != CODA_DX6)
1905		coda_write(dev, 0, CODA7_REG_BIT_AXI_SRAM_USE);
1906
1907	coda_write(dev, CODA_INT_INTERRUPT_ENABLE,
1908		      CODA_REG_BIT_INT_ENABLE);
1909
1910	/* Reset VPU and start processor */
1911	data = coda_read(dev, CODA_REG_BIT_CODE_RESET);
1912	data |= CODA_REG_RESET_ENABLE;
1913	coda_write(dev, data, CODA_REG_BIT_CODE_RESET);
1914	udelay(10);
1915	data &= ~CODA_REG_RESET_ENABLE;
1916	coda_write(dev, data, CODA_REG_BIT_CODE_RESET);
1917	coda_write(dev, CODA_REG_RUN_ENABLE, CODA_REG_BIT_CODE_RUN);
1918
1919	clk_disable_unprepare(dev->clk_ahb);
1920	clk_disable_unprepare(dev->clk_per);
1921
1922	return 0;
1923
1924err_clk_ahb:
1925	clk_disable_unprepare(dev->clk_per);
1926err_clk_per:
1927	return ret;
1928}
1929
1930static int coda_register_device(struct coda_dev *dev, int i)
1931{
1932	struct video_device *vfd = &dev->vfd[i];
1933
1934	if (i >= dev->devtype->num_vdevs)
1935		return -EINVAL;
1936
1937	strlcpy(vfd->name, dev->devtype->vdevs[i]->name, sizeof(vfd->name));
1938	vfd->fops	= &coda_fops;
1939	vfd->ioctl_ops	= &coda_ioctl_ops;
1940	vfd->release	= video_device_release_empty,
1941	vfd->lock	= &dev->dev_mutex;
1942	vfd->v4l2_dev	= &dev->v4l2_dev;
1943	vfd->vfl_dir	= VFL_DIR_M2M;
1944	video_set_drvdata(vfd, dev);
1945
1946	/* Not applicable, use the selection API instead */
1947	v4l2_disable_ioctl(vfd, VIDIOC_CROPCAP);
1948	v4l2_disable_ioctl(vfd, VIDIOC_G_CROP);
1949	v4l2_disable_ioctl(vfd, VIDIOC_S_CROP);
1950
1951	return video_register_device(vfd, VFL_TYPE_GRABBER, 0);
1952}
1953
1954static void coda_fw_callback(const struct firmware *fw, void *context)
1955{
1956	struct coda_dev *dev = context;
1957	struct platform_device *pdev = dev->plat_dev;
1958	int i, ret;
1959
1960	if (!fw) {
1961		v4l2_err(&dev->v4l2_dev, "firmware request failed\n");
1962		goto put_pm;
1963	}
1964
1965	/* allocate auxiliary per-device code buffer for the BIT processor */
1966	ret = coda_alloc_aux_buf(dev, &dev->codebuf, fw->size, "codebuf",
1967				 dev->debugfs_root);
1968	if (ret < 0)
1969		goto put_pm;
1970
1971	/* Copy the whole firmware image to the code buffer */
1972	memcpy(dev->codebuf.vaddr, fw->data, fw->size);
1973	release_firmware(fw);
1974
1975	ret = coda_hw_init(dev);
1976	if (ret < 0) {
1977		v4l2_err(&dev->v4l2_dev, "HW initialization failed\n");
1978		goto put_pm;
1979	}
1980
1981	ret = coda_check_firmware(dev);
1982	if (ret < 0)
1983		goto put_pm;
1984
1985	dev->alloc_ctx = vb2_dma_contig_init_ctx(&pdev->dev);
1986	if (IS_ERR(dev->alloc_ctx)) {
1987		v4l2_err(&dev->v4l2_dev, "Failed to alloc vb2 context\n");
1988		goto put_pm;
1989	}
1990
1991	dev->m2m_dev = v4l2_m2m_init(&coda_m2m_ops);
1992	if (IS_ERR(dev->m2m_dev)) {
1993		v4l2_err(&dev->v4l2_dev, "Failed to init mem2mem device\n");
1994		goto rel_ctx;
1995	}
1996
1997	for (i = 0; i < dev->devtype->num_vdevs; i++) {
1998		ret = coda_register_device(dev, i);
1999		if (ret) {
2000			v4l2_err(&dev->v4l2_dev,
2001				 "Failed to register %s video device: %d\n",
2002				 dev->devtype->vdevs[i]->name, ret);
2003			goto rel_vfd;
2004		}
2005	}
2006
2007	v4l2_info(&dev->v4l2_dev, "codec registered as /dev/video[%d-%d]\n",
2008		  dev->vfd[0].num, dev->vfd[i - 1].num);
2009
2010	pm_runtime_put_sync(&pdev->dev);
2011	return;
2012
2013rel_vfd:
2014	while (--i >= 0)
2015		video_unregister_device(&dev->vfd[i]);
2016	v4l2_m2m_release(dev->m2m_dev);
2017rel_ctx:
2018	vb2_dma_contig_cleanup_ctx(dev->alloc_ctx);
2019put_pm:
2020	pm_runtime_put_sync(&pdev->dev);
2021}
2022
2023static int coda_firmware_request(struct coda_dev *dev)
2024{
2025	char *fw = dev->devtype->firmware;
2026
2027	dev_dbg(&dev->plat_dev->dev, "requesting firmware '%s' for %s\n", fw,
2028		coda_product_name(dev->devtype->product));
2029
2030	return request_firmware_nowait(THIS_MODULE, true,
2031		fw, &dev->plat_dev->dev, GFP_KERNEL, dev, coda_fw_callback);
2032}
2033
2034enum coda_platform {
2035	CODA_IMX27,
2036	CODA_IMX53,
2037	CODA_IMX6Q,
2038	CODA_IMX6DL,
2039};
2040
2041static const struct coda_devtype coda_devdata[] = {
2042	[CODA_IMX27] = {
2043		.firmware     = "v4l-codadx6-imx27.bin",
2044		.product      = CODA_DX6,
2045		.codecs       = codadx6_codecs,
2046		.num_codecs   = ARRAY_SIZE(codadx6_codecs),
2047		.vdevs        = codadx6_video_devices,
2048		.num_vdevs    = ARRAY_SIZE(codadx6_video_devices),
2049		.workbuf_size = 288 * 1024 + FMO_SLICE_SAVE_BUF_SIZE * 8 * 1024,
2050		.iram_size    = 0xb000,
2051	},
2052	[CODA_IMX53] = {
2053		.firmware     = "v4l-coda7541-imx53.bin",
2054		.product      = CODA_7541,
2055		.codecs       = coda7_codecs,
2056		.num_codecs   = ARRAY_SIZE(coda7_codecs),
2057		.vdevs        = coda7_video_devices,
2058		.num_vdevs    = ARRAY_SIZE(coda7_video_devices),
2059		.workbuf_size = 128 * 1024,
2060		.tempbuf_size = 304 * 1024,
2061		.iram_size    = 0x14000,
2062	},
2063	[CODA_IMX6Q] = {
2064		.firmware     = "v4l-coda960-imx6q.bin",
2065		.product      = CODA_960,
2066		.codecs       = coda9_codecs,
2067		.num_codecs   = ARRAY_SIZE(coda9_codecs),
2068		.vdevs        = coda9_video_devices,
2069		.num_vdevs    = ARRAY_SIZE(coda9_video_devices),
2070		.workbuf_size = 80 * 1024,
2071		.tempbuf_size = 204 * 1024,
2072		.iram_size    = 0x21000,
2073	},
2074	[CODA_IMX6DL] = {
2075		.firmware     = "v4l-coda960-imx6dl.bin",
2076		.product      = CODA_960,
2077		.codecs       = coda9_codecs,
2078		.num_codecs   = ARRAY_SIZE(coda9_codecs),
2079		.vdevs        = coda9_video_devices,
2080		.num_vdevs    = ARRAY_SIZE(coda9_video_devices),
2081		.workbuf_size = 80 * 1024,
2082		.tempbuf_size = 204 * 1024,
2083		.iram_size    = 0x20000,
2084	},
2085};
2086
2087static struct platform_device_id coda_platform_ids[] = {
2088	{ .name = "coda-imx27", .driver_data = CODA_IMX27 },
2089	{ /* sentinel */ }
2090};
2091MODULE_DEVICE_TABLE(platform, coda_platform_ids);
2092
2093#ifdef CONFIG_OF
2094static const struct of_device_id coda_dt_ids[] = {
2095	{ .compatible = "fsl,imx27-vpu", .data = &coda_devdata[CODA_IMX27] },
2096	{ .compatible = "fsl,imx53-vpu", .data = &coda_devdata[CODA_IMX53] },
2097	{ .compatible = "fsl,imx6q-vpu", .data = &coda_devdata[CODA_IMX6Q] },
2098	{ .compatible = "fsl,imx6dl-vpu", .data = &coda_devdata[CODA_IMX6DL] },
2099	{ /* sentinel */ }
2100};
2101MODULE_DEVICE_TABLE(of, coda_dt_ids);
2102#endif
2103
2104static int coda_probe(struct platform_device *pdev)
2105{
2106	const struct of_device_id *of_id =
2107			of_match_device(of_match_ptr(coda_dt_ids), &pdev->dev);
2108	const struct platform_device_id *pdev_id;
2109	struct coda_platform_data *pdata = pdev->dev.platform_data;
2110	struct device_node *np = pdev->dev.of_node;
2111	struct gen_pool *pool;
2112	struct coda_dev *dev;
2113	struct resource *res;
2114	int ret, irq;
2115
2116	dev = devm_kzalloc(&pdev->dev, sizeof(*dev), GFP_KERNEL);
2117	if (!dev)
2118		return -ENOMEM;
2119
2120	pdev_id = of_id ? of_id->data : platform_get_device_id(pdev);
2121
2122	if (of_id)
2123		dev->devtype = of_id->data;
2124	else if (pdev_id)
2125		dev->devtype = &coda_devdata[pdev_id->driver_data];
2126	else
2127		return -EINVAL;
2128
2129	spin_lock_init(&dev->irqlock);
2130	INIT_LIST_HEAD(&dev->instances);
2131
2132	dev->plat_dev = pdev;
2133	dev->clk_per = devm_clk_get(&pdev->dev, "per");
2134	if (IS_ERR(dev->clk_per)) {
2135		dev_err(&pdev->dev, "Could not get per clock\n");
2136		return PTR_ERR(dev->clk_per);
2137	}
2138
2139	dev->clk_ahb = devm_clk_get(&pdev->dev, "ahb");
2140	if (IS_ERR(dev->clk_ahb)) {
2141		dev_err(&pdev->dev, "Could not get ahb clock\n");
2142		return PTR_ERR(dev->clk_ahb);
2143	}
2144
2145	/* Get  memory for physical registers */
2146	res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
2147	dev->regs_base = devm_ioremap_resource(&pdev->dev, res);
2148	if (IS_ERR(dev->regs_base))
2149		return PTR_ERR(dev->regs_base);
2150
2151	/* IRQ */
2152	irq = platform_get_irq_byname(pdev, "bit");
2153	if (irq < 0)
2154		irq = platform_get_irq(pdev, 0);
2155	if (irq < 0) {
2156		dev_err(&pdev->dev, "failed to get irq resource\n");
2157		return irq;
2158	}
2159
2160	ret = devm_request_threaded_irq(&pdev->dev, irq, NULL, coda_irq_handler,
2161			IRQF_ONESHOT, dev_name(&pdev->dev), dev);
2162	if (ret < 0) {
2163		dev_err(&pdev->dev, "failed to request irq: %d\n", ret);
2164		return ret;
2165	}
2166
2167	dev->rstc = devm_reset_control_get_optional(&pdev->dev, NULL);
2168	if (IS_ERR(dev->rstc)) {
2169		ret = PTR_ERR(dev->rstc);
2170		if (ret == -ENOENT || ret == -ENOSYS) {
2171			dev->rstc = NULL;
2172		} else {
2173			dev_err(&pdev->dev, "failed get reset control: %d\n",
2174				ret);
2175			return ret;
2176		}
2177	}
2178
2179	/* Get IRAM pool from device tree or platform data */
2180	pool = of_gen_pool_get(np, "iram", 0);
2181	if (!pool && pdata)
2182		pool = gen_pool_get(pdata->iram_dev, NULL);
2183	if (!pool) {
2184		dev_err(&pdev->dev, "iram pool not available\n");
2185		return -ENOMEM;
2186	}
2187	dev->iram_pool = pool;
2188
2189	ret = v4l2_device_register(&pdev->dev, &dev->v4l2_dev);
2190	if (ret)
2191		return ret;
2192
2193	mutex_init(&dev->dev_mutex);
2194	mutex_init(&dev->coda_mutex);
2195
2196	dev->debugfs_root = debugfs_create_dir("coda", NULL);
2197	if (!dev->debugfs_root)
2198		dev_warn(&pdev->dev, "failed to create debugfs root\n");
2199
2200	/* allocate auxiliary per-device buffers for the BIT processor */
2201	if (dev->devtype->product == CODA_DX6) {
2202		ret = coda_alloc_aux_buf(dev, &dev->workbuf,
2203					 dev->devtype->workbuf_size, "workbuf",
2204					 dev->debugfs_root);
2205		if (ret < 0)
2206			goto err_v4l2_register;
2207	}
2208
2209	if (dev->devtype->tempbuf_size) {
2210		ret = coda_alloc_aux_buf(dev, &dev->tempbuf,
2211					 dev->devtype->tempbuf_size, "tempbuf",
2212					 dev->debugfs_root);
2213		if (ret < 0)
2214			goto err_v4l2_register;
2215	}
2216
2217	dev->iram.size = dev->devtype->iram_size;
2218	dev->iram.vaddr = gen_pool_dma_alloc(dev->iram_pool, dev->iram.size,
2219					     &dev->iram.paddr);
2220	if (!dev->iram.vaddr) {
2221		dev_warn(&pdev->dev, "unable to alloc iram\n");
2222	} else {
2223		memset(dev->iram.vaddr, 0, dev->iram.size);
2224		dev->iram.blob.data = dev->iram.vaddr;
2225		dev->iram.blob.size = dev->iram.size;
2226		dev->iram.dentry = debugfs_create_blob("iram", 0644,
2227						       dev->debugfs_root,
2228						       &dev->iram.blob);
2229	}
2230
2231	dev->workqueue = alloc_workqueue("coda", WQ_UNBOUND | WQ_MEM_RECLAIM, 1);
2232	if (!dev->workqueue) {
2233		dev_err(&pdev->dev, "unable to alloc workqueue\n");
2234		ret = -ENOMEM;
2235		goto err_v4l2_register;
2236	}
2237
2238	platform_set_drvdata(pdev, dev);
2239
2240	/*
2241	 * Start activated so we can directly call coda_hw_init in
2242	 * coda_fw_callback regardless of whether CONFIG_PM is
2243	 * enabled or whether the device is associated with a PM domain.
2244	 */
2245	pm_runtime_get_noresume(&pdev->dev);
2246	pm_runtime_set_active(&pdev->dev);
2247	pm_runtime_enable(&pdev->dev);
2248
2249	return coda_firmware_request(dev);
2250
2251err_v4l2_register:
2252	v4l2_device_unregister(&dev->v4l2_dev);
2253	return ret;
2254}
2255
2256static int coda_remove(struct platform_device *pdev)
2257{
2258	struct coda_dev *dev = platform_get_drvdata(pdev);
2259	int i;
2260
2261	for (i = 0; i < ARRAY_SIZE(dev->vfd); i++) {
2262		if (video_get_drvdata(&dev->vfd[i]))
2263			video_unregister_device(&dev->vfd[i]);
2264	}
2265	if (dev->m2m_dev)
2266		v4l2_m2m_release(dev->m2m_dev);
2267	pm_runtime_disable(&pdev->dev);
2268	if (dev->alloc_ctx)
2269		vb2_dma_contig_cleanup_ctx(dev->alloc_ctx);
2270	v4l2_device_unregister(&dev->v4l2_dev);
2271	destroy_workqueue(dev->workqueue);
2272	if (dev->iram.vaddr)
2273		gen_pool_free(dev->iram_pool, (unsigned long)dev->iram.vaddr,
2274			      dev->iram.size);
2275	coda_free_aux_buf(dev, &dev->codebuf);
2276	coda_free_aux_buf(dev, &dev->tempbuf);
2277	coda_free_aux_buf(dev, &dev->workbuf);
2278	debugfs_remove_recursive(dev->debugfs_root);
2279	return 0;
2280}
2281
2282#ifdef CONFIG_PM
2283static int coda_runtime_resume(struct device *dev)
2284{
2285	struct coda_dev *cdev = dev_get_drvdata(dev);
2286	int ret = 0;
2287
2288	if (dev->pm_domain && cdev->codebuf.vaddr) {
2289		ret = coda_hw_init(cdev);
2290		if (ret)
2291			v4l2_err(&cdev->v4l2_dev, "HW initialization failed\n");
2292	}
2293
2294	return ret;
2295}
2296#endif
2297
2298static const struct dev_pm_ops coda_pm_ops = {
2299	SET_RUNTIME_PM_OPS(NULL, coda_runtime_resume, NULL)
2300};
2301
2302static struct platform_driver coda_driver = {
2303	.probe	= coda_probe,
2304	.remove	= coda_remove,
2305	.driver	= {
2306		.name	= CODA_NAME,
2307		.of_match_table = of_match_ptr(coda_dt_ids),
2308		.pm	= &coda_pm_ops,
2309	},
2310	.id_table = coda_platform_ids,
2311};
2312
2313module_platform_driver(coda_driver);
2314
2315MODULE_LICENSE("GPL");
2316MODULE_AUTHOR("Javier Martin <javier.martin@vista-silicon.com>");
2317MODULE_DESCRIPTION("Coda multi-standard codec V4L2 driver");
2318