1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * Coda multi-standard codec IP - BIT processor functions
4 *
5 * Copyright (C) 2012 Vista Silicon S.L.
6 * Javier Martin, <javier.martin@vista-silicon.com>
7 * Xavier Duret
8 * Copyright (C) 2012-2014 Philipp Zabel, Pengutronix
9 */
10
11 #include <linux/clk.h>
12 #include <linux/irqreturn.h>
13 #include <linux/kernel.h>
14 #include <linux/log2.h>
15 #include <linux/platform_device.h>
16 #include <linux/ratelimit.h>
17 #include <linux/reset.h>
18 #include <linux/slab.h>
19 #include <linux/videodev2.h>
20
21 #include <media/v4l2-common.h>
22 #include <media/v4l2-ctrls.h>
23 #include <media/v4l2-fh.h>
24 #include <media/v4l2-mem2mem.h>
25 #include <media/videobuf2-v4l2.h>
26 #include <media/videobuf2-dma-contig.h>
27 #include <media/videobuf2-vmalloc.h>
28
29 #include "coda.h"
30 #include "imx-vdoa.h"
31 #define CREATE_TRACE_POINTS
32 #include "trace.h"
33
34 #define CODA_PARA_BUF_SIZE (10 * 1024)
35 #define CODA7_PS_BUF_SIZE 0x28000
36 #define CODA9_PS_SAVE_SIZE (512 * 1024)
37
38 #define CODA_DEFAULT_GAMMA 4096
39 #define CODA9_DEFAULT_GAMMA 24576 /* 0.75 * 32768 */
40
41 static void coda_free_bitstream_buffer(struct coda_ctx *ctx);
42
coda_is_initialized(struct coda_dev * dev)43 static inline int coda_is_initialized(struct coda_dev *dev)
44 {
45 return coda_read(dev, CODA_REG_BIT_CUR_PC) != 0;
46 }
47
coda_isbusy(struct coda_dev * dev)48 static inline unsigned long coda_isbusy(struct coda_dev *dev)
49 {
50 return coda_read(dev, CODA_REG_BIT_BUSY);
51 }
52
coda_wait_timeout(struct coda_dev * dev)53 static int coda_wait_timeout(struct coda_dev *dev)
54 {
55 unsigned long timeout = jiffies + msecs_to_jiffies(1000);
56
57 while (coda_isbusy(dev)) {
58 if (time_after(jiffies, timeout))
59 return -ETIMEDOUT;
60 }
61 return 0;
62 }
63
coda_command_async(struct coda_ctx * ctx,int cmd)64 static void coda_command_async(struct coda_ctx *ctx, int cmd)
65 {
66 struct coda_dev *dev = ctx->dev;
67
68 if (dev->devtype->product == CODA_HX4 ||
69 dev->devtype->product == CODA_7541 ||
70 dev->devtype->product == CODA_960) {
71 /* Restore context related registers to CODA */
72 coda_write(dev, ctx->bit_stream_param,
73 CODA_REG_BIT_BIT_STREAM_PARAM);
74 coda_write(dev, ctx->frm_dis_flg,
75 CODA_REG_BIT_FRM_DIS_FLG(ctx->reg_idx));
76 coda_write(dev, ctx->frame_mem_ctrl,
77 CODA_REG_BIT_FRAME_MEM_CTRL);
78 coda_write(dev, ctx->workbuf.paddr, CODA_REG_BIT_WORK_BUF_ADDR);
79 }
80
81 if (dev->devtype->product == CODA_960) {
82 coda_write(dev, 1, CODA9_GDI_WPROT_ERR_CLR);
83 coda_write(dev, 0, CODA9_GDI_WPROT_RGN_EN);
84 }
85
86 coda_write(dev, CODA_REG_BIT_BUSY_FLAG, CODA_REG_BIT_BUSY);
87
88 coda_write(dev, ctx->idx, CODA_REG_BIT_RUN_INDEX);
89 coda_write(dev, ctx->params.codec_mode, CODA_REG_BIT_RUN_COD_STD);
90 coda_write(dev, ctx->params.codec_mode_aux, CODA7_REG_BIT_RUN_AUX_STD);
91
92 trace_coda_bit_run(ctx, cmd);
93
94 coda_write(dev, cmd, CODA_REG_BIT_RUN_COMMAND);
95 }
96
coda_command_sync(struct coda_ctx * ctx,int cmd)97 static int coda_command_sync(struct coda_ctx *ctx, int cmd)
98 {
99 struct coda_dev *dev = ctx->dev;
100 int ret;
101
102 lockdep_assert_held(&dev->coda_mutex);
103
104 coda_command_async(ctx, cmd);
105 ret = coda_wait_timeout(dev);
106 trace_coda_bit_done(ctx);
107
108 return ret;
109 }
110
coda_hw_reset(struct coda_ctx * ctx)111 int coda_hw_reset(struct coda_ctx *ctx)
112 {
113 struct coda_dev *dev = ctx->dev;
114 unsigned long timeout;
115 unsigned int idx;
116 int ret;
117
118 lockdep_assert_held(&dev->coda_mutex);
119
120 if (!dev->rstc)
121 return -ENOENT;
122
123 idx = coda_read(dev, CODA_REG_BIT_RUN_INDEX);
124
125 if (dev->devtype->product == CODA_960) {
126 timeout = jiffies + msecs_to_jiffies(100);
127 coda_write(dev, 0x11, CODA9_GDI_BUS_CTRL);
128 while (coda_read(dev, CODA9_GDI_BUS_STATUS) != 0x77) {
129 if (time_after(jiffies, timeout))
130 return -ETIME;
131 cpu_relax();
132 }
133 }
134
135 ret = reset_control_reset(dev->rstc);
136 if (ret < 0)
137 return ret;
138
139 if (dev->devtype->product == CODA_960)
140 coda_write(dev, 0x00, CODA9_GDI_BUS_CTRL);
141 coda_write(dev, CODA_REG_BIT_BUSY_FLAG, CODA_REG_BIT_BUSY);
142 coda_write(dev, CODA_REG_RUN_ENABLE, CODA_REG_BIT_CODE_RUN);
143 ret = coda_wait_timeout(dev);
144 coda_write(dev, idx, CODA_REG_BIT_RUN_INDEX);
145
146 return ret;
147 }
148
coda_kfifo_sync_from_device(struct coda_ctx * ctx)149 static void coda_kfifo_sync_from_device(struct coda_ctx *ctx)
150 {
151 struct __kfifo *kfifo = &ctx->bitstream_fifo.kfifo;
152 struct coda_dev *dev = ctx->dev;
153 u32 rd_ptr;
154
155 rd_ptr = coda_read(dev, CODA_REG_BIT_RD_PTR(ctx->reg_idx));
156 kfifo->out = (kfifo->in & ~kfifo->mask) |
157 (rd_ptr - ctx->bitstream.paddr);
158 if (kfifo->out > kfifo->in)
159 kfifo->out -= kfifo->mask + 1;
160 }
161
coda_kfifo_sync_to_device_full(struct coda_ctx * ctx)162 static void coda_kfifo_sync_to_device_full(struct coda_ctx *ctx)
163 {
164 struct __kfifo *kfifo = &ctx->bitstream_fifo.kfifo;
165 struct coda_dev *dev = ctx->dev;
166 u32 rd_ptr, wr_ptr;
167
168 rd_ptr = ctx->bitstream.paddr + (kfifo->out & kfifo->mask);
169 coda_write(dev, rd_ptr, CODA_REG_BIT_RD_PTR(ctx->reg_idx));
170 wr_ptr = ctx->bitstream.paddr + (kfifo->in & kfifo->mask);
171 coda_write(dev, wr_ptr, CODA_REG_BIT_WR_PTR(ctx->reg_idx));
172 }
173
coda_kfifo_sync_to_device_write(struct coda_ctx * ctx)174 static void coda_kfifo_sync_to_device_write(struct coda_ctx *ctx)
175 {
176 struct __kfifo *kfifo = &ctx->bitstream_fifo.kfifo;
177 struct coda_dev *dev = ctx->dev;
178 u32 wr_ptr;
179
180 wr_ptr = ctx->bitstream.paddr + (kfifo->in & kfifo->mask);
181 coda_write(dev, wr_ptr, CODA_REG_BIT_WR_PTR(ctx->reg_idx));
182 }
183
coda_h264_bitstream_pad(struct coda_ctx * ctx,u32 size)184 static int coda_h264_bitstream_pad(struct coda_ctx *ctx, u32 size)
185 {
186 unsigned char *buf;
187 u32 n;
188
189 if (size < 6)
190 size = 6;
191
192 buf = kmalloc(size, GFP_KERNEL);
193 if (!buf)
194 return -ENOMEM;
195
196 coda_h264_filler_nal(size, buf);
197 n = kfifo_in(&ctx->bitstream_fifo, buf, size);
198 kfree(buf);
199
200 return (n < size) ? -ENOSPC : 0;
201 }
202
coda_bitstream_flush(struct coda_ctx * ctx)203 int coda_bitstream_flush(struct coda_ctx *ctx)
204 {
205 int ret;
206
207 if (ctx->inst_type != CODA_INST_DECODER || !ctx->use_bit)
208 return 0;
209
210 ret = coda_command_sync(ctx, CODA_COMMAND_DEC_BUF_FLUSH);
211 if (ret < 0) {
212 v4l2_err(&ctx->dev->v4l2_dev, "failed to flush bitstream\n");
213 return ret;
214 }
215
216 kfifo_init(&ctx->bitstream_fifo, ctx->bitstream.vaddr,
217 ctx->bitstream.size);
218 coda_kfifo_sync_to_device_full(ctx);
219
220 return 0;
221 }
222
coda_bitstream_queue(struct coda_ctx * ctx,const u8 * buf,u32 size)223 static int coda_bitstream_queue(struct coda_ctx *ctx, const u8 *buf, u32 size)
224 {
225 u32 n = kfifo_in(&ctx->bitstream_fifo, buf, size);
226
227 return (n < size) ? -ENOSPC : 0;
228 }
229
coda_buffer_parse_headers(struct coda_ctx * ctx,struct vb2_v4l2_buffer * src_buf,u32 payload)230 static u32 coda_buffer_parse_headers(struct coda_ctx *ctx,
231 struct vb2_v4l2_buffer *src_buf,
232 u32 payload)
233 {
234 u8 *vaddr = vb2_plane_vaddr(&src_buf->vb2_buf, 0);
235 u32 size = 0;
236
237 switch (ctx->codec->src_fourcc) {
238 case V4L2_PIX_FMT_MPEG2:
239 size = coda_mpeg2_parse_headers(ctx, vaddr, payload);
240 break;
241 case V4L2_PIX_FMT_MPEG4:
242 size = coda_mpeg4_parse_headers(ctx, vaddr, payload);
243 break;
244 default:
245 break;
246 }
247
248 return size;
249 }
250
coda_bitstream_try_queue(struct coda_ctx * ctx,struct vb2_v4l2_buffer * src_buf)251 static bool coda_bitstream_try_queue(struct coda_ctx *ctx,
252 struct vb2_v4l2_buffer *src_buf)
253 {
254 unsigned long payload = vb2_get_plane_payload(&src_buf->vb2_buf, 0);
255 u8 *vaddr = vb2_plane_vaddr(&src_buf->vb2_buf, 0);
256 int ret;
257 int i;
258
259 if (coda_get_bitstream_payload(ctx) + payload + 512 >=
260 ctx->bitstream.size)
261 return false;
262
263 if (!vaddr) {
264 v4l2_err(&ctx->dev->v4l2_dev, "trying to queue empty buffer\n");
265 return true;
266 }
267
268 if (ctx->qsequence == 0 && payload < 512) {
269 /*
270 * Add padding after the first buffer, if it is too small to be
271 * fetched by the CODA, by repeating the headers. Without
272 * repeated headers, or the first frame already queued, decoder
273 * sequence initialization fails with error code 0x2000 on i.MX6
274 * or error code 0x1 on i.MX51.
275 */
276 u32 header_size = coda_buffer_parse_headers(ctx, src_buf,
277 payload);
278
279 if (header_size) {
280 coda_dbg(1, ctx, "pad with %u-byte header\n",
281 header_size);
282 for (i = payload; i < 512; i += header_size) {
283 ret = coda_bitstream_queue(ctx, vaddr,
284 header_size);
285 if (ret < 0) {
286 v4l2_err(&ctx->dev->v4l2_dev,
287 "bitstream buffer overflow\n");
288 return false;
289 }
290 if (ctx->dev->devtype->product == CODA_960)
291 break;
292 }
293 } else {
294 coda_dbg(1, ctx,
295 "could not parse header, sequence initialization might fail\n");
296 }
297
298 /* Add padding before the first buffer, if it is too small */
299 if (ctx->codec->src_fourcc == V4L2_PIX_FMT_H264)
300 coda_h264_bitstream_pad(ctx, 512 - payload);
301 }
302
303 ret = coda_bitstream_queue(ctx, vaddr, payload);
304 if (ret < 0) {
305 v4l2_err(&ctx->dev->v4l2_dev, "bitstream buffer overflow\n");
306 return false;
307 }
308
309 src_buf->sequence = ctx->qsequence++;
310
311 /* Sync read pointer to device */
312 if (ctx == v4l2_m2m_get_curr_priv(ctx->dev->m2m_dev))
313 coda_kfifo_sync_to_device_write(ctx);
314
315 /* Set the stream-end flag after the last buffer is queued */
316 if (src_buf->flags & V4L2_BUF_FLAG_LAST)
317 coda_bit_stream_end_flag(ctx);
318 ctx->hold = false;
319
320 return true;
321 }
322
coda_fill_bitstream(struct coda_ctx * ctx,struct list_head * buffer_list)323 void coda_fill_bitstream(struct coda_ctx *ctx, struct list_head *buffer_list)
324 {
325 struct vb2_v4l2_buffer *src_buf;
326 struct coda_buffer_meta *meta;
327 u32 start;
328
329 lockdep_assert_held(&ctx->bitstream_mutex);
330
331 if (ctx->bit_stream_param & CODA_BIT_STREAM_END_FLAG)
332 return;
333
334 while (v4l2_m2m_num_src_bufs_ready(ctx->fh.m2m_ctx) > 0) {
335 /*
336 * Only queue two JPEGs into the bitstream buffer to keep
337 * latency low. We need at least one complete buffer and the
338 * header of another buffer (for prescan) in the bitstream.
339 */
340 if (ctx->codec->src_fourcc == V4L2_PIX_FMT_JPEG &&
341 ctx->num_metas > 1)
342 break;
343
344 if (ctx->num_internal_frames &&
345 ctx->num_metas >= ctx->num_internal_frames) {
346 meta = list_first_entry(&ctx->buffer_meta_list,
347 struct coda_buffer_meta, list);
348
349 /*
350 * If we managed to fill in at least a full reorder
351 * window of buffers (num_internal_frames is a
352 * conservative estimate for this) and the bitstream
353 * prefetcher has at least 2 256 bytes periods beyond
354 * the first buffer to fetch, we can safely stop queuing
355 * in order to limit the decoder drain latency.
356 */
357 if (coda_bitstream_can_fetch_past(ctx, meta->end))
358 break;
359 }
360
361 src_buf = v4l2_m2m_next_src_buf(ctx->fh.m2m_ctx);
362
363 /* Drop frames that do not start/end with a SOI/EOI markers */
364 if (ctx->codec->src_fourcc == V4L2_PIX_FMT_JPEG &&
365 !coda_jpeg_check_buffer(ctx, &src_buf->vb2_buf)) {
366 v4l2_err(&ctx->dev->v4l2_dev,
367 "dropping invalid JPEG frame %d\n",
368 ctx->qsequence);
369 src_buf = v4l2_m2m_src_buf_remove(ctx->fh.m2m_ctx);
370 if (buffer_list) {
371 struct v4l2_m2m_buffer *m2m_buf;
372
373 m2m_buf = container_of(src_buf,
374 struct v4l2_m2m_buffer,
375 vb);
376 list_add_tail(&m2m_buf->list, buffer_list);
377 } else {
378 v4l2_m2m_buf_done(src_buf, VB2_BUF_STATE_ERROR);
379 }
380 continue;
381 }
382
383 /* Dump empty buffers */
384 if (!vb2_get_plane_payload(&src_buf->vb2_buf, 0)) {
385 src_buf = v4l2_m2m_src_buf_remove(ctx->fh.m2m_ctx);
386 v4l2_m2m_buf_done(src_buf, VB2_BUF_STATE_DONE);
387 continue;
388 }
389
390 /* Buffer start position */
391 start = ctx->bitstream_fifo.kfifo.in;
392
393 if (coda_bitstream_try_queue(ctx, src_buf)) {
394 /*
395 * Source buffer is queued in the bitstream ringbuffer;
396 * queue the timestamp and mark source buffer as done
397 */
398 src_buf = v4l2_m2m_src_buf_remove(ctx->fh.m2m_ctx);
399
400 meta = kmalloc(sizeof(*meta), GFP_KERNEL);
401 if (meta) {
402 meta->sequence = src_buf->sequence;
403 meta->timecode = src_buf->timecode;
404 meta->timestamp = src_buf->vb2_buf.timestamp;
405 meta->start = start;
406 meta->end = ctx->bitstream_fifo.kfifo.in;
407 meta->last = src_buf->flags & V4L2_BUF_FLAG_LAST;
408 if (meta->last)
409 coda_dbg(1, ctx, "marking last meta");
410 spin_lock(&ctx->buffer_meta_lock);
411 list_add_tail(&meta->list,
412 &ctx->buffer_meta_list);
413 ctx->num_metas++;
414 spin_unlock(&ctx->buffer_meta_lock);
415
416 trace_coda_bit_queue(ctx, src_buf, meta);
417 }
418
419 if (buffer_list) {
420 struct v4l2_m2m_buffer *m2m_buf;
421
422 m2m_buf = container_of(src_buf,
423 struct v4l2_m2m_buffer,
424 vb);
425 list_add_tail(&m2m_buf->list, buffer_list);
426 } else {
427 v4l2_m2m_buf_done(src_buf, VB2_BUF_STATE_DONE);
428 }
429 } else {
430 break;
431 }
432 }
433 }
434
coda_bit_stream_end_flag(struct coda_ctx * ctx)435 void coda_bit_stream_end_flag(struct coda_ctx *ctx)
436 {
437 struct coda_dev *dev = ctx->dev;
438
439 ctx->bit_stream_param |= CODA_BIT_STREAM_END_FLAG;
440
441 /* If this context is currently running, update the hardware flag */
442 if ((dev->devtype->product == CODA_960) &&
443 coda_isbusy(dev) &&
444 (ctx->idx == coda_read(dev, CODA_REG_BIT_RUN_INDEX))) {
445 coda_write(dev, ctx->bit_stream_param,
446 CODA_REG_BIT_BIT_STREAM_PARAM);
447 }
448 }
449
coda_parabuf_write(struct coda_ctx * ctx,int index,u32 value)450 static void coda_parabuf_write(struct coda_ctx *ctx, int index, u32 value)
451 {
452 struct coda_dev *dev = ctx->dev;
453 u32 *p = ctx->parabuf.vaddr;
454
455 if (dev->devtype->product == CODA_DX6)
456 p[index] = value;
457 else
458 p[index ^ 1] = value;
459 }
460
coda_alloc_context_buf(struct coda_ctx * ctx,struct coda_aux_buf * buf,size_t size,const char * name)461 static inline int coda_alloc_context_buf(struct coda_ctx *ctx,
462 struct coda_aux_buf *buf, size_t size,
463 const char *name)
464 {
465 return coda_alloc_aux_buf(ctx->dev, buf, size, name, ctx->debugfs_entry);
466 }
467
468
coda_free_framebuffers(struct coda_ctx * ctx)469 static void coda_free_framebuffers(struct coda_ctx *ctx)
470 {
471 int i;
472
473 for (i = 0; i < CODA_MAX_FRAMEBUFFERS; i++)
474 coda_free_aux_buf(ctx->dev, &ctx->internal_frames[i].buf);
475 }
476
coda_alloc_framebuffers(struct coda_ctx * ctx,struct coda_q_data * q_data,u32 fourcc)477 static int coda_alloc_framebuffers(struct coda_ctx *ctx,
478 struct coda_q_data *q_data, u32 fourcc)
479 {
480 struct coda_dev *dev = ctx->dev;
481 unsigned int ysize, ycbcr_size;
482 int ret;
483 int i;
484
485 if (ctx->codec->src_fourcc == V4L2_PIX_FMT_H264 ||
486 ctx->codec->dst_fourcc == V4L2_PIX_FMT_H264 ||
487 ctx->codec->src_fourcc == V4L2_PIX_FMT_MPEG4 ||
488 ctx->codec->dst_fourcc == V4L2_PIX_FMT_MPEG4)
489 ysize = round_up(q_data->rect.width, 16) *
490 round_up(q_data->rect.height, 16);
491 else
492 ysize = round_up(q_data->rect.width, 8) * q_data->rect.height;
493
494 if (ctx->tiled_map_type == GDI_TILED_FRAME_MB_RASTER_MAP)
495 ycbcr_size = round_up(ysize, 4096) + ysize / 2;
496 else
497 ycbcr_size = ysize + ysize / 2;
498
499 /* Allocate frame buffers */
500 for (i = 0; i < ctx->num_internal_frames; i++) {
501 size_t size = ycbcr_size;
502 char *name;
503
504 /* Add space for mvcol buffers */
505 if (dev->devtype->product != CODA_DX6 &&
506 (ctx->codec->src_fourcc == V4L2_PIX_FMT_H264 ||
507 (ctx->codec->src_fourcc == V4L2_PIX_FMT_MPEG4 && i == 0)))
508 size += ysize / 4;
509 name = kasprintf(GFP_KERNEL, "fb%d", i);
510 if (!name) {
511 coda_free_framebuffers(ctx);
512 return -ENOMEM;
513 }
514 ret = coda_alloc_context_buf(ctx, &ctx->internal_frames[i].buf,
515 size, name);
516 kfree(name);
517 if (ret < 0) {
518 coda_free_framebuffers(ctx);
519 return ret;
520 }
521 }
522
523 /* Register frame buffers in the parameter buffer */
524 for (i = 0; i < ctx->num_internal_frames; i++) {
525 u32 y, cb, cr, mvcol;
526
527 /* Start addresses of Y, Cb, Cr planes */
528 y = ctx->internal_frames[i].buf.paddr;
529 cb = y + ysize;
530 cr = y + ysize + ysize/4;
531 mvcol = y + ysize + ysize/4 + ysize/4;
532 if (ctx->tiled_map_type == GDI_TILED_FRAME_MB_RASTER_MAP) {
533 cb = round_up(cb, 4096);
534 mvcol = cb + ysize/2;
535 cr = 0;
536 /* Packed 20-bit MSB of base addresses */
537 /* YYYYYCCC, CCyyyyyc, cccc.... */
538 y = (y & 0xfffff000) | cb >> 20;
539 cb = (cb & 0x000ff000) << 12;
540 }
541 coda_parabuf_write(ctx, i * 3 + 0, y);
542 coda_parabuf_write(ctx, i * 3 + 1, cb);
543 coda_parabuf_write(ctx, i * 3 + 2, cr);
544
545 if (dev->devtype->product == CODA_DX6)
546 continue;
547
548 /* mvcol buffer for h.264 and mpeg4 */
549 if (ctx->codec->src_fourcc == V4L2_PIX_FMT_H264)
550 coda_parabuf_write(ctx, 96 + i, mvcol);
551 if (ctx->codec->src_fourcc == V4L2_PIX_FMT_MPEG4 && i == 0)
552 coda_parabuf_write(ctx, 97, mvcol);
553 }
554
555 return 0;
556 }
557
coda_free_context_buffers(struct coda_ctx * ctx)558 static void coda_free_context_buffers(struct coda_ctx *ctx)
559 {
560 struct coda_dev *dev = ctx->dev;
561
562 coda_free_aux_buf(dev, &ctx->slicebuf);
563 coda_free_aux_buf(dev, &ctx->psbuf);
564 if (dev->devtype->product != CODA_DX6)
565 coda_free_aux_buf(dev, &ctx->workbuf);
566 coda_free_aux_buf(dev, &ctx->parabuf);
567 }
568
coda_alloc_context_buffers(struct coda_ctx * ctx,struct coda_q_data * q_data)569 static int coda_alloc_context_buffers(struct coda_ctx *ctx,
570 struct coda_q_data *q_data)
571 {
572 struct coda_dev *dev = ctx->dev;
573 size_t size;
574 int ret;
575
576 if (!ctx->parabuf.vaddr) {
577 ret = coda_alloc_context_buf(ctx, &ctx->parabuf,
578 CODA_PARA_BUF_SIZE, "parabuf");
579 if (ret < 0)
580 return ret;
581 }
582
583 if (dev->devtype->product == CODA_DX6)
584 return 0;
585
586 if (!ctx->slicebuf.vaddr && q_data->fourcc == V4L2_PIX_FMT_H264) {
587 /* worst case slice size */
588 size = (DIV_ROUND_UP(q_data->rect.width, 16) *
589 DIV_ROUND_UP(q_data->rect.height, 16)) * 3200 / 8 + 512;
590 ret = coda_alloc_context_buf(ctx, &ctx->slicebuf, size,
591 "slicebuf");
592 if (ret < 0)
593 goto err;
594 }
595
596 if (!ctx->psbuf.vaddr && (dev->devtype->product == CODA_HX4 ||
597 dev->devtype->product == CODA_7541)) {
598 ret = coda_alloc_context_buf(ctx, &ctx->psbuf,
599 CODA7_PS_BUF_SIZE, "psbuf");
600 if (ret < 0)
601 goto err;
602 }
603
604 if (!ctx->workbuf.vaddr) {
605 size = dev->devtype->workbuf_size;
606 if (dev->devtype->product == CODA_960 &&
607 q_data->fourcc == V4L2_PIX_FMT_H264)
608 size += CODA9_PS_SAVE_SIZE;
609 ret = coda_alloc_context_buf(ctx, &ctx->workbuf, size,
610 "workbuf");
611 if (ret < 0)
612 goto err;
613 }
614
615 return 0;
616
617 err:
618 coda_free_context_buffers(ctx);
619 return ret;
620 }
621
coda_encode_header(struct coda_ctx * ctx,struct vb2_v4l2_buffer * buf,int header_code,u8 * header,int * size)622 static int coda_encode_header(struct coda_ctx *ctx, struct vb2_v4l2_buffer *buf,
623 int header_code, u8 *header, int *size)
624 {
625 struct vb2_buffer *vb = &buf->vb2_buf;
626 struct coda_dev *dev = ctx->dev;
627 struct coda_q_data *q_data_src;
628 struct v4l2_rect *r;
629 size_t bufsize;
630 int ret;
631 int i;
632
633 if (dev->devtype->product == CODA_960)
634 memset(vb2_plane_vaddr(vb, 0), 0, 64);
635
636 coda_write(dev, vb2_dma_contig_plane_dma_addr(vb, 0),
637 CODA_CMD_ENC_HEADER_BB_START);
638 bufsize = vb2_plane_size(vb, 0);
639 if (dev->devtype->product == CODA_960)
640 bufsize /= 1024;
641 coda_write(dev, bufsize, CODA_CMD_ENC_HEADER_BB_SIZE);
642 if (dev->devtype->product == CODA_960 &&
643 ctx->codec->dst_fourcc == V4L2_PIX_FMT_H264 &&
644 header_code == CODA_HEADER_H264_SPS) {
645 q_data_src = get_q_data(ctx, V4L2_BUF_TYPE_VIDEO_OUTPUT);
646 r = &q_data_src->rect;
647
648 if (r->width % 16 || r->height % 16) {
649 u32 crop_right = round_up(r->width, 16) - r->width;
650 u32 crop_bottom = round_up(r->height, 16) - r->height;
651
652 coda_write(dev, crop_right,
653 CODA9_CMD_ENC_HEADER_FRAME_CROP_H);
654 coda_write(dev, crop_bottom,
655 CODA9_CMD_ENC_HEADER_FRAME_CROP_V);
656 header_code |= CODA9_HEADER_FRAME_CROP;
657 }
658 }
659 coda_write(dev, header_code, CODA_CMD_ENC_HEADER_CODE);
660 ret = coda_command_sync(ctx, CODA_COMMAND_ENCODE_HEADER);
661 if (ret < 0) {
662 v4l2_err(&dev->v4l2_dev, "CODA_COMMAND_ENCODE_HEADER timeout\n");
663 return ret;
664 }
665
666 if (dev->devtype->product == CODA_960) {
667 for (i = 63; i > 0; i--)
668 if (((char *)vb2_plane_vaddr(vb, 0))[i] != 0)
669 break;
670 *size = i + 1;
671 } else {
672 *size = coda_read(dev, CODA_REG_BIT_WR_PTR(ctx->reg_idx)) -
673 coda_read(dev, CODA_CMD_ENC_HEADER_BB_START);
674 }
675 memcpy(header, vb2_plane_vaddr(vb, 0), *size);
676
677 return 0;
678 }
679
coda_slice_mode(struct coda_ctx * ctx)680 static u32 coda_slice_mode(struct coda_ctx *ctx)
681 {
682 int size, unit;
683
684 switch (ctx->params.slice_mode) {
685 case V4L2_MPEG_VIDEO_MULTI_SLICE_MODE_SINGLE:
686 default:
687 return 0;
688 case V4L2_MPEG_VIDEO_MULTI_SLICE_MODE_MAX_MB:
689 size = ctx->params.slice_max_mb;
690 unit = 1;
691 break;
692 case V4L2_MPEG_VIDEO_MULTI_SLICE_MODE_MAX_BYTES:
693 size = ctx->params.slice_max_bits;
694 unit = 0;
695 break;
696 }
697
698 return ((size & CODA_SLICING_SIZE_MASK) << CODA_SLICING_SIZE_OFFSET) |
699 ((unit & CODA_SLICING_UNIT_MASK) << CODA_SLICING_UNIT_OFFSET) |
700 ((1 & CODA_SLICING_MODE_MASK) << CODA_SLICING_MODE_OFFSET);
701 }
702
coda_enc_param_change(struct coda_ctx * ctx)703 static int coda_enc_param_change(struct coda_ctx *ctx)
704 {
705 struct coda_dev *dev = ctx->dev;
706 u32 change_enable = 0;
707 u32 success;
708 int ret;
709
710 if (ctx->params.gop_size_changed) {
711 change_enable |= CODA_PARAM_CHANGE_RC_GOP;
712 coda_write(dev, ctx->params.gop_size,
713 CODA_CMD_ENC_PARAM_RC_GOP);
714 ctx->gopcounter = ctx->params.gop_size - 1;
715 ctx->params.gop_size_changed = false;
716 }
717 if (ctx->params.h264_intra_qp_changed) {
718 coda_dbg(1, ctx, "parameter change: intra Qp %u\n",
719 ctx->params.h264_intra_qp);
720
721 if (ctx->params.bitrate) {
722 change_enable |= CODA_PARAM_CHANGE_RC_INTRA_QP;
723 coda_write(dev, ctx->params.h264_intra_qp,
724 CODA_CMD_ENC_PARAM_RC_INTRA_QP);
725 }
726 ctx->params.h264_intra_qp_changed = false;
727 }
728 if (ctx->params.bitrate_changed) {
729 coda_dbg(1, ctx, "parameter change: bitrate %u kbit/s\n",
730 ctx->params.bitrate);
731 change_enable |= CODA_PARAM_CHANGE_RC_BITRATE;
732 coda_write(dev, ctx->params.bitrate,
733 CODA_CMD_ENC_PARAM_RC_BITRATE);
734 ctx->params.bitrate_changed = false;
735 }
736 if (ctx->params.framerate_changed) {
737 coda_dbg(1, ctx, "parameter change: frame rate %u/%u Hz\n",
738 ctx->params.framerate & 0xffff,
739 (ctx->params.framerate >> 16) + 1);
740 change_enable |= CODA_PARAM_CHANGE_RC_FRAME_RATE;
741 coda_write(dev, ctx->params.framerate,
742 CODA_CMD_ENC_PARAM_RC_FRAME_RATE);
743 ctx->params.framerate_changed = false;
744 }
745 if (ctx->params.intra_refresh_changed) {
746 coda_dbg(1, ctx, "parameter change: intra refresh MBs %u\n",
747 ctx->params.intra_refresh);
748 change_enable |= CODA_PARAM_CHANGE_INTRA_MB_NUM;
749 coda_write(dev, ctx->params.intra_refresh,
750 CODA_CMD_ENC_PARAM_INTRA_MB_NUM);
751 ctx->params.intra_refresh_changed = false;
752 }
753 if (ctx->params.slice_mode_changed) {
754 change_enable |= CODA_PARAM_CHANGE_SLICE_MODE;
755 coda_write(dev, coda_slice_mode(ctx),
756 CODA_CMD_ENC_PARAM_SLICE_MODE);
757 ctx->params.slice_mode_changed = false;
758 }
759
760 if (!change_enable)
761 return 0;
762
763 coda_write(dev, change_enable, CODA_CMD_ENC_PARAM_CHANGE_ENABLE);
764
765 ret = coda_command_sync(ctx, CODA_COMMAND_RC_CHANGE_PARAMETER);
766 if (ret < 0)
767 return ret;
768
769 success = coda_read(dev, CODA_RET_ENC_PARAM_CHANGE_SUCCESS);
770 if (success != 1)
771 coda_dbg(1, ctx, "parameter change failed: %u\n", success);
772
773 return 0;
774 }
775
coda_iram_alloc(struct coda_iram_info * iram,size_t size)776 static phys_addr_t coda_iram_alloc(struct coda_iram_info *iram, size_t size)
777 {
778 phys_addr_t ret;
779
780 size = round_up(size, 1024);
781 if (size > iram->remaining)
782 return 0;
783 iram->remaining -= size;
784
785 ret = iram->next_paddr;
786 iram->next_paddr += size;
787
788 return ret;
789 }
790
coda_setup_iram(struct coda_ctx * ctx)791 static void coda_setup_iram(struct coda_ctx *ctx)
792 {
793 struct coda_iram_info *iram_info = &ctx->iram_info;
794 struct coda_dev *dev = ctx->dev;
795 int w64, w128;
796 int mb_width;
797 int dbk_bits;
798 int bit_bits;
799 int ip_bits;
800 int me_bits;
801
802 memset(iram_info, 0, sizeof(*iram_info));
803 iram_info->next_paddr = dev->iram.paddr;
804 iram_info->remaining = dev->iram.size;
805
806 if (!dev->iram.vaddr)
807 return;
808
809 switch (dev->devtype->product) {
810 case CODA_HX4:
811 dbk_bits = CODA7_USE_HOST_DBK_ENABLE;
812 bit_bits = CODA7_USE_HOST_BIT_ENABLE;
813 ip_bits = CODA7_USE_HOST_IP_ENABLE;
814 me_bits = CODA7_USE_HOST_ME_ENABLE;
815 break;
816 case CODA_7541:
817 dbk_bits = CODA7_USE_HOST_DBK_ENABLE | CODA7_USE_DBK_ENABLE;
818 bit_bits = CODA7_USE_HOST_BIT_ENABLE | CODA7_USE_BIT_ENABLE;
819 ip_bits = CODA7_USE_HOST_IP_ENABLE | CODA7_USE_IP_ENABLE;
820 me_bits = CODA7_USE_HOST_ME_ENABLE | CODA7_USE_ME_ENABLE;
821 break;
822 case CODA_960:
823 dbk_bits = CODA9_USE_HOST_DBK_ENABLE | CODA9_USE_DBK_ENABLE;
824 bit_bits = CODA9_USE_HOST_BIT_ENABLE | CODA7_USE_BIT_ENABLE;
825 ip_bits = CODA9_USE_HOST_IP_ENABLE | CODA7_USE_IP_ENABLE;
826 me_bits = 0;
827 break;
828 default: /* CODA_DX6 */
829 return;
830 }
831
832 if (ctx->inst_type == CODA_INST_ENCODER) {
833 struct coda_q_data *q_data_src;
834
835 q_data_src = get_q_data(ctx, V4L2_BUF_TYPE_VIDEO_OUTPUT);
836 mb_width = DIV_ROUND_UP(q_data_src->rect.width, 16);
837 w128 = mb_width * 128;
838 w64 = mb_width * 64;
839
840 /* Prioritize in case IRAM is too small for everything */
841 if (dev->devtype->product == CODA_HX4 ||
842 dev->devtype->product == CODA_7541) {
843 iram_info->search_ram_size = round_up(mb_width * 16 *
844 36 + 2048, 1024);
845 iram_info->search_ram_paddr = coda_iram_alloc(iram_info,
846 iram_info->search_ram_size);
847 if (!iram_info->search_ram_paddr) {
848 pr_err("IRAM is smaller than the search ram size\n");
849 goto out;
850 }
851 iram_info->axi_sram_use |= me_bits;
852 }
853
854 /* Only H.264BP and H.263P3 are considered */
855 iram_info->buf_dbk_y_use = coda_iram_alloc(iram_info, w64);
856 iram_info->buf_dbk_c_use = coda_iram_alloc(iram_info, w64);
857 if (!iram_info->buf_dbk_c_use)
858 goto out;
859 iram_info->axi_sram_use |= dbk_bits;
860
861 iram_info->buf_bit_use = coda_iram_alloc(iram_info, w128);
862 if (!iram_info->buf_bit_use)
863 goto out;
864 iram_info->axi_sram_use |= bit_bits;
865
866 iram_info->buf_ip_ac_dc_use = coda_iram_alloc(iram_info, w128);
867 if (!iram_info->buf_ip_ac_dc_use)
868 goto out;
869 iram_info->axi_sram_use |= ip_bits;
870
871 /* OVL and BTP disabled for encoder */
872 } else if (ctx->inst_type == CODA_INST_DECODER) {
873 struct coda_q_data *q_data_dst;
874
875 q_data_dst = get_q_data(ctx, V4L2_BUF_TYPE_VIDEO_CAPTURE);
876 mb_width = DIV_ROUND_UP(q_data_dst->width, 16);
877 w128 = mb_width * 128;
878
879 iram_info->buf_dbk_y_use = coda_iram_alloc(iram_info, w128);
880 iram_info->buf_dbk_c_use = coda_iram_alloc(iram_info, w128);
881 if (!iram_info->buf_dbk_c_use)
882 goto out;
883 iram_info->axi_sram_use |= dbk_bits;
884
885 iram_info->buf_bit_use = coda_iram_alloc(iram_info, w128);
886 if (!iram_info->buf_bit_use)
887 goto out;
888 iram_info->axi_sram_use |= bit_bits;
889
890 iram_info->buf_ip_ac_dc_use = coda_iram_alloc(iram_info, w128);
891 if (!iram_info->buf_ip_ac_dc_use)
892 goto out;
893 iram_info->axi_sram_use |= ip_bits;
894
895 /* OVL and BTP unused as there is no VC1 support yet */
896 }
897
898 out:
899 if (!(iram_info->axi_sram_use & CODA7_USE_HOST_IP_ENABLE))
900 coda_dbg(1, ctx, "IRAM smaller than needed\n");
901
902 if (dev->devtype->product == CODA_HX4 ||
903 dev->devtype->product == CODA_7541) {
904 /* TODO - Enabling these causes picture errors on CODA7541 */
905 if (ctx->inst_type == CODA_INST_DECODER) {
906 /* fw 1.4.50 */
907 iram_info->axi_sram_use &= ~(CODA7_USE_HOST_IP_ENABLE |
908 CODA7_USE_IP_ENABLE);
909 } else {
910 /* fw 13.4.29 */
911 iram_info->axi_sram_use &= ~(CODA7_USE_HOST_IP_ENABLE |
912 CODA7_USE_HOST_DBK_ENABLE |
913 CODA7_USE_IP_ENABLE |
914 CODA7_USE_DBK_ENABLE);
915 }
916 }
917 }
918
919 static u32 coda_supported_firmwares[] = {
920 CODA_FIRMWARE_VERNUM(CODA_DX6, 2, 2, 5),
921 CODA_FIRMWARE_VERNUM(CODA_HX4, 1, 4, 50),
922 CODA_FIRMWARE_VERNUM(CODA_7541, 1, 4, 50),
923 CODA_FIRMWARE_VERNUM(CODA_960, 2, 1, 5),
924 CODA_FIRMWARE_VERNUM(CODA_960, 2, 1, 9),
925 CODA_FIRMWARE_VERNUM(CODA_960, 2, 3, 10),
926 CODA_FIRMWARE_VERNUM(CODA_960, 3, 1, 1),
927 };
928
coda_firmware_supported(u32 vernum)929 static bool coda_firmware_supported(u32 vernum)
930 {
931 int i;
932
933 for (i = 0; i < ARRAY_SIZE(coda_supported_firmwares); i++)
934 if (vernum == coda_supported_firmwares[i])
935 return true;
936 return false;
937 }
938
coda_check_firmware(struct coda_dev * dev)939 int coda_check_firmware(struct coda_dev *dev)
940 {
941 u16 product, major, minor, release;
942 u32 data;
943 int ret;
944
945 ret = clk_prepare_enable(dev->clk_per);
946 if (ret)
947 goto err_clk_per;
948
949 ret = clk_prepare_enable(dev->clk_ahb);
950 if (ret)
951 goto err_clk_ahb;
952
953 coda_write(dev, 0, CODA_CMD_FIRMWARE_VERNUM);
954 coda_write(dev, CODA_REG_BIT_BUSY_FLAG, CODA_REG_BIT_BUSY);
955 coda_write(dev, 0, CODA_REG_BIT_RUN_INDEX);
956 coda_write(dev, 0, CODA_REG_BIT_RUN_COD_STD);
957 coda_write(dev, CODA_COMMAND_FIRMWARE_GET, CODA_REG_BIT_RUN_COMMAND);
958 if (coda_wait_timeout(dev)) {
959 v4l2_err(&dev->v4l2_dev, "firmware get command error\n");
960 ret = -EIO;
961 goto err_run_cmd;
962 }
963
964 if (dev->devtype->product == CODA_960) {
965 data = coda_read(dev, CODA9_CMD_FIRMWARE_CODE_REV);
966 v4l2_info(&dev->v4l2_dev, "Firmware code revision: %d\n",
967 data);
968 }
969
970 /* Check we are compatible with the loaded firmware */
971 data = coda_read(dev, CODA_CMD_FIRMWARE_VERNUM);
972 product = CODA_FIRMWARE_PRODUCT(data);
973 major = CODA_FIRMWARE_MAJOR(data);
974 minor = CODA_FIRMWARE_MINOR(data);
975 release = CODA_FIRMWARE_RELEASE(data);
976
977 clk_disable_unprepare(dev->clk_per);
978 clk_disable_unprepare(dev->clk_ahb);
979
980 if (product != dev->devtype->product) {
981 v4l2_err(&dev->v4l2_dev,
982 "Wrong firmware. Hw: %s, Fw: %s, Version: %u.%u.%u\n",
983 coda_product_name(dev->devtype->product),
984 coda_product_name(product), major, minor, release);
985 return -EINVAL;
986 }
987
988 v4l2_info(&dev->v4l2_dev, "Initialized %s.\n",
989 coda_product_name(product));
990
991 if (coda_firmware_supported(data)) {
992 v4l2_info(&dev->v4l2_dev, "Firmware version: %u.%u.%u\n",
993 major, minor, release);
994 } else {
995 v4l2_warn(&dev->v4l2_dev,
996 "Unsupported firmware version: %u.%u.%u\n",
997 major, minor, release);
998 }
999
1000 return 0;
1001
1002 err_run_cmd:
1003 clk_disable_unprepare(dev->clk_ahb);
1004 err_clk_ahb:
1005 clk_disable_unprepare(dev->clk_per);
1006 err_clk_per:
1007 return ret;
1008 }
1009
coda9_set_frame_cache(struct coda_ctx * ctx,u32 fourcc)1010 static void coda9_set_frame_cache(struct coda_ctx *ctx, u32 fourcc)
1011 {
1012 u32 cache_size, cache_config;
1013
1014 if (ctx->tiled_map_type == GDI_LINEAR_FRAME_MAP) {
1015 /* Luma 2x0 page, 2x6 cache, chroma 2x0 page, 2x4 cache size */
1016 cache_size = 0x20262024;
1017 cache_config = 2 << CODA9_CACHE_PAGEMERGE_OFFSET;
1018 } else {
1019 /* Luma 0x2 page, 4x4 cache, chroma 0x2 page, 4x3 cache size */
1020 cache_size = 0x02440243;
1021 cache_config = 1 << CODA9_CACHE_PAGEMERGE_OFFSET;
1022 }
1023 coda_write(ctx->dev, cache_size, CODA9_CMD_SET_FRAME_CACHE_SIZE);
1024 if (fourcc == V4L2_PIX_FMT_NV12 || fourcc == V4L2_PIX_FMT_YUYV) {
1025 cache_config |= 32 << CODA9_CACHE_LUMA_BUFFER_SIZE_OFFSET |
1026 16 << CODA9_CACHE_CR_BUFFER_SIZE_OFFSET |
1027 0 << CODA9_CACHE_CB_BUFFER_SIZE_OFFSET;
1028 } else {
1029 cache_config |= 32 << CODA9_CACHE_LUMA_BUFFER_SIZE_OFFSET |
1030 8 << CODA9_CACHE_CR_BUFFER_SIZE_OFFSET |
1031 8 << CODA9_CACHE_CB_BUFFER_SIZE_OFFSET;
1032 }
1033 coda_write(ctx->dev, cache_config, CODA9_CMD_SET_FRAME_CACHE_CONFIG);
1034 }
1035
1036 /*
1037 * Encoder context operations
1038 */
1039
coda_encoder_reqbufs(struct coda_ctx * ctx,struct v4l2_requestbuffers * rb)1040 static int coda_encoder_reqbufs(struct coda_ctx *ctx,
1041 struct v4l2_requestbuffers *rb)
1042 {
1043 struct coda_q_data *q_data_src;
1044 int ret;
1045
1046 if (rb->type != V4L2_BUF_TYPE_VIDEO_OUTPUT)
1047 return 0;
1048
1049 if (rb->count) {
1050 q_data_src = get_q_data(ctx, V4L2_BUF_TYPE_VIDEO_OUTPUT);
1051 ret = coda_alloc_context_buffers(ctx, q_data_src);
1052 if (ret < 0)
1053 return ret;
1054 } else {
1055 coda_free_context_buffers(ctx);
1056 }
1057
1058 return 0;
1059 }
1060
coda_start_encoding(struct coda_ctx * ctx)1061 static int coda_start_encoding(struct coda_ctx *ctx)
1062 {
1063 struct coda_dev *dev = ctx->dev;
1064 struct v4l2_device *v4l2_dev = &dev->v4l2_dev;
1065 struct coda_q_data *q_data_src, *q_data_dst;
1066 u32 bitstream_buf, bitstream_size;
1067 struct vb2_v4l2_buffer *buf;
1068 int gamma, ret, value;
1069 u32 dst_fourcc;
1070 int num_fb;
1071 u32 stride;
1072
1073 q_data_src = get_q_data(ctx, V4L2_BUF_TYPE_VIDEO_OUTPUT);
1074 q_data_dst = get_q_data(ctx, V4L2_BUF_TYPE_VIDEO_CAPTURE);
1075 dst_fourcc = q_data_dst->fourcc;
1076
1077 buf = v4l2_m2m_next_dst_buf(ctx->fh.m2m_ctx);
1078 bitstream_buf = vb2_dma_contig_plane_dma_addr(&buf->vb2_buf, 0);
1079 bitstream_size = q_data_dst->sizeimage;
1080
1081 if (!coda_is_initialized(dev)) {
1082 v4l2_err(v4l2_dev, "coda is not initialized.\n");
1083 return -EFAULT;
1084 }
1085
1086 if (dst_fourcc == V4L2_PIX_FMT_JPEG) {
1087 if (!ctx->params.jpeg_qmat_tab[0])
1088 ctx->params.jpeg_qmat_tab[0] = kmalloc(64, GFP_KERNEL);
1089 if (!ctx->params.jpeg_qmat_tab[1])
1090 ctx->params.jpeg_qmat_tab[1] = kmalloc(64, GFP_KERNEL);
1091 coda_set_jpeg_compression_quality(ctx, ctx->params.jpeg_quality);
1092 }
1093
1094 mutex_lock(&dev->coda_mutex);
1095
1096 coda_write(dev, ctx->parabuf.paddr, CODA_REG_BIT_PARA_BUF_ADDR);
1097 coda_write(dev, bitstream_buf, CODA_REG_BIT_RD_PTR(ctx->reg_idx));
1098 coda_write(dev, bitstream_buf, CODA_REG_BIT_WR_PTR(ctx->reg_idx));
1099 switch (dev->devtype->product) {
1100 case CODA_DX6:
1101 coda_write(dev, CODADX6_STREAM_BUF_DYNALLOC_EN |
1102 CODADX6_STREAM_BUF_PIC_RESET, CODA_REG_BIT_STREAM_CTRL);
1103 break;
1104 case CODA_960:
1105 coda_write(dev, 0, CODA9_GDI_WPROT_RGN_EN);
1106 fallthrough;
1107 case CODA_HX4:
1108 case CODA_7541:
1109 coda_write(dev, CODA7_STREAM_BUF_DYNALLOC_EN |
1110 CODA7_STREAM_BUF_PIC_RESET, CODA_REG_BIT_STREAM_CTRL);
1111 break;
1112 }
1113
1114 ctx->frame_mem_ctrl &= ~(CODA_FRAME_CHROMA_INTERLEAVE | (0x3 << 9) |
1115 CODA9_FRAME_TILED2LINEAR);
1116 if (q_data_src->fourcc == V4L2_PIX_FMT_NV12)
1117 ctx->frame_mem_ctrl |= CODA_FRAME_CHROMA_INTERLEAVE;
1118 if (ctx->tiled_map_type == GDI_TILED_FRAME_MB_RASTER_MAP)
1119 ctx->frame_mem_ctrl |= (0x3 << 9) | CODA9_FRAME_TILED2LINEAR;
1120 coda_write(dev, ctx->frame_mem_ctrl, CODA_REG_BIT_FRAME_MEM_CTRL);
1121
1122 if (dev->devtype->product == CODA_DX6) {
1123 /* Configure the coda */
1124 coda_write(dev, dev->iram.paddr,
1125 CODADX6_REG_BIT_SEARCH_RAM_BASE_ADDR);
1126 }
1127
1128 /* Could set rotation here if needed */
1129 value = 0;
1130 switch (dev->devtype->product) {
1131 case CODA_DX6:
1132 value = (q_data_src->rect.width & CODADX6_PICWIDTH_MASK)
1133 << CODADX6_PICWIDTH_OFFSET;
1134 value |= (q_data_src->rect.height & CODADX6_PICHEIGHT_MASK)
1135 << CODA_PICHEIGHT_OFFSET;
1136 break;
1137 case CODA_HX4:
1138 case CODA_7541:
1139 if (dst_fourcc == V4L2_PIX_FMT_H264) {
1140 value = (round_up(q_data_src->rect.width, 16) &
1141 CODA7_PICWIDTH_MASK) << CODA7_PICWIDTH_OFFSET;
1142 value |= (round_up(q_data_src->rect.height, 16) &
1143 CODA7_PICHEIGHT_MASK) << CODA_PICHEIGHT_OFFSET;
1144 break;
1145 }
1146 fallthrough;
1147 case CODA_960:
1148 value = (q_data_src->rect.width & CODA7_PICWIDTH_MASK)
1149 << CODA7_PICWIDTH_OFFSET;
1150 value |= (q_data_src->rect.height & CODA7_PICHEIGHT_MASK)
1151 << CODA_PICHEIGHT_OFFSET;
1152 }
1153 coda_write(dev, value, CODA_CMD_ENC_SEQ_SRC_SIZE);
1154 if (dst_fourcc == V4L2_PIX_FMT_JPEG)
1155 ctx->params.framerate = 0;
1156 coda_write(dev, ctx->params.framerate,
1157 CODA_CMD_ENC_SEQ_SRC_F_RATE);
1158
1159 ctx->params.codec_mode = ctx->codec->mode;
1160 switch (dst_fourcc) {
1161 case V4L2_PIX_FMT_MPEG4:
1162 if (dev->devtype->product == CODA_960)
1163 coda_write(dev, CODA9_STD_MPEG4,
1164 CODA_CMD_ENC_SEQ_COD_STD);
1165 else
1166 coda_write(dev, CODA_STD_MPEG4,
1167 CODA_CMD_ENC_SEQ_COD_STD);
1168 coda_write(dev, 0, CODA_CMD_ENC_SEQ_MP4_PARA);
1169 break;
1170 case V4L2_PIX_FMT_H264:
1171 if (dev->devtype->product == CODA_960)
1172 coda_write(dev, CODA9_STD_H264,
1173 CODA_CMD_ENC_SEQ_COD_STD);
1174 else
1175 coda_write(dev, CODA_STD_H264,
1176 CODA_CMD_ENC_SEQ_COD_STD);
1177 value = ((ctx->params.h264_disable_deblocking_filter_idc &
1178 CODA_264PARAM_DISABLEDEBLK_MASK) <<
1179 CODA_264PARAM_DISABLEDEBLK_OFFSET) |
1180 ((ctx->params.h264_slice_alpha_c0_offset_div2 &
1181 CODA_264PARAM_DEBLKFILTEROFFSETALPHA_MASK) <<
1182 CODA_264PARAM_DEBLKFILTEROFFSETALPHA_OFFSET) |
1183 ((ctx->params.h264_slice_beta_offset_div2 &
1184 CODA_264PARAM_DEBLKFILTEROFFSETBETA_MASK) <<
1185 CODA_264PARAM_DEBLKFILTEROFFSETBETA_OFFSET) |
1186 (ctx->params.h264_constrained_intra_pred_flag <<
1187 CODA_264PARAM_CONSTRAINEDINTRAPREDFLAG_OFFSET) |
1188 (ctx->params.h264_chroma_qp_index_offset &
1189 CODA_264PARAM_CHROMAQPOFFSET_MASK);
1190 coda_write(dev, value, CODA_CMD_ENC_SEQ_264_PARA);
1191 break;
1192 case V4L2_PIX_FMT_JPEG:
1193 coda_write(dev, 0, CODA_CMD_ENC_SEQ_JPG_PARA);
1194 coda_write(dev, ctx->params.jpeg_restart_interval,
1195 CODA_CMD_ENC_SEQ_JPG_RST_INTERVAL);
1196 coda_write(dev, 0, CODA_CMD_ENC_SEQ_JPG_THUMB_EN);
1197 coda_write(dev, 0, CODA_CMD_ENC_SEQ_JPG_THUMB_SIZE);
1198 coda_write(dev, 0, CODA_CMD_ENC_SEQ_JPG_THUMB_OFFSET);
1199
1200 coda_jpeg_write_tables(ctx);
1201 break;
1202 default:
1203 v4l2_err(v4l2_dev,
1204 "dst format (0x%08x) invalid.\n", dst_fourcc);
1205 ret = -EINVAL;
1206 goto out;
1207 }
1208
1209 /*
1210 * slice mode and GOP size registers are used for thumb size/offset
1211 * in JPEG mode
1212 */
1213 if (dst_fourcc != V4L2_PIX_FMT_JPEG) {
1214 value = coda_slice_mode(ctx);
1215 coda_write(dev, value, CODA_CMD_ENC_SEQ_SLICE_MODE);
1216 value = ctx->params.gop_size;
1217 coda_write(dev, value, CODA_CMD_ENC_SEQ_GOP_SIZE);
1218 }
1219
1220 if (ctx->params.bitrate && (ctx->params.frame_rc_enable ||
1221 ctx->params.mb_rc_enable)) {
1222 ctx->params.bitrate_changed = false;
1223 ctx->params.h264_intra_qp_changed = false;
1224
1225 /* Rate control enabled */
1226 value = (ctx->params.bitrate & CODA_RATECONTROL_BITRATE_MASK)
1227 << CODA_RATECONTROL_BITRATE_OFFSET;
1228 value |= 1 & CODA_RATECONTROL_ENABLE_MASK;
1229 value |= (ctx->params.vbv_delay &
1230 CODA_RATECONTROL_INITIALDELAY_MASK)
1231 << CODA_RATECONTROL_INITIALDELAY_OFFSET;
1232 if (dev->devtype->product == CODA_960)
1233 value |= BIT(31); /* disable autoskip */
1234 } else {
1235 value = 0;
1236 }
1237 coda_write(dev, value, CODA_CMD_ENC_SEQ_RC_PARA);
1238
1239 coda_write(dev, ctx->params.vbv_size, CODA_CMD_ENC_SEQ_RC_BUF_SIZE);
1240 coda_write(dev, ctx->params.intra_refresh,
1241 CODA_CMD_ENC_SEQ_INTRA_REFRESH);
1242
1243 coda_write(dev, bitstream_buf, CODA_CMD_ENC_SEQ_BB_START);
1244 coda_write(dev, bitstream_size / 1024, CODA_CMD_ENC_SEQ_BB_SIZE);
1245
1246
1247 value = 0;
1248 if (dev->devtype->product == CODA_960)
1249 gamma = CODA9_DEFAULT_GAMMA;
1250 else
1251 gamma = CODA_DEFAULT_GAMMA;
1252 if (gamma > 0) {
1253 coda_write(dev, (gamma & CODA_GAMMA_MASK) << CODA_GAMMA_OFFSET,
1254 CODA_CMD_ENC_SEQ_RC_GAMMA);
1255 }
1256
1257 if (ctx->params.h264_min_qp || ctx->params.h264_max_qp) {
1258 coda_write(dev,
1259 ctx->params.h264_min_qp << CODA_QPMIN_OFFSET |
1260 ctx->params.h264_max_qp << CODA_QPMAX_OFFSET,
1261 CODA_CMD_ENC_SEQ_RC_QP_MIN_MAX);
1262 }
1263 if (dev->devtype->product == CODA_960) {
1264 if (ctx->params.h264_max_qp)
1265 value |= 1 << CODA9_OPTION_RCQPMAX_OFFSET;
1266 if (CODA_DEFAULT_GAMMA > 0)
1267 value |= 1 << CODA9_OPTION_GAMMA_OFFSET;
1268 } else {
1269 if (CODA_DEFAULT_GAMMA > 0) {
1270 if (dev->devtype->product == CODA_DX6)
1271 value |= 1 << CODADX6_OPTION_GAMMA_OFFSET;
1272 else
1273 value |= 1 << CODA7_OPTION_GAMMA_OFFSET;
1274 }
1275 if (ctx->params.h264_min_qp)
1276 value |= 1 << CODA7_OPTION_RCQPMIN_OFFSET;
1277 if (ctx->params.h264_max_qp)
1278 value |= 1 << CODA7_OPTION_RCQPMAX_OFFSET;
1279 }
1280 coda_write(dev, value, CODA_CMD_ENC_SEQ_OPTION);
1281
1282 if (ctx->params.frame_rc_enable && !ctx->params.mb_rc_enable)
1283 value = 1;
1284 else
1285 value = 0;
1286 coda_write(dev, value, CODA_CMD_ENC_SEQ_RC_INTERVAL_MODE);
1287
1288 coda_setup_iram(ctx);
1289
1290 if (dst_fourcc == V4L2_PIX_FMT_H264) {
1291 switch (dev->devtype->product) {
1292 case CODA_DX6:
1293 value = FMO_SLICE_SAVE_BUF_SIZE << 7;
1294 coda_write(dev, value, CODADX6_CMD_ENC_SEQ_FMO);
1295 break;
1296 case CODA_HX4:
1297 case CODA_7541:
1298 coda_write(dev, ctx->iram_info.search_ram_paddr,
1299 CODA7_CMD_ENC_SEQ_SEARCH_BASE);
1300 coda_write(dev, ctx->iram_info.search_ram_size,
1301 CODA7_CMD_ENC_SEQ_SEARCH_SIZE);
1302 break;
1303 case CODA_960:
1304 coda_write(dev, 0, CODA9_CMD_ENC_SEQ_ME_OPTION);
1305 coda_write(dev, 0, CODA9_CMD_ENC_SEQ_INTRA_WEIGHT);
1306 }
1307 }
1308
1309 ret = coda_command_sync(ctx, CODA_COMMAND_SEQ_INIT);
1310 if (ret < 0) {
1311 v4l2_err(v4l2_dev, "CODA_COMMAND_SEQ_INIT timeout\n");
1312 goto out;
1313 }
1314
1315 if (coda_read(dev, CODA_RET_ENC_SEQ_SUCCESS) == 0) {
1316 v4l2_err(v4l2_dev, "CODA_COMMAND_SEQ_INIT failed\n");
1317 ret = -EFAULT;
1318 goto out;
1319 }
1320 ctx->initialized = 1;
1321
1322 if (dst_fourcc != V4L2_PIX_FMT_JPEG) {
1323 if (dev->devtype->product == CODA_960)
1324 ctx->num_internal_frames = 4;
1325 else
1326 ctx->num_internal_frames = 2;
1327 ret = coda_alloc_framebuffers(ctx, q_data_src, dst_fourcc);
1328 if (ret < 0) {
1329 v4l2_err(v4l2_dev, "failed to allocate framebuffers\n");
1330 goto out;
1331 }
1332 num_fb = 2;
1333 stride = q_data_src->bytesperline;
1334 } else {
1335 ctx->num_internal_frames = 0;
1336 num_fb = 0;
1337 stride = 0;
1338 }
1339 coda_write(dev, num_fb, CODA_CMD_SET_FRAME_BUF_NUM);
1340 coda_write(dev, stride, CODA_CMD_SET_FRAME_BUF_STRIDE);
1341
1342 if (dev->devtype->product == CODA_HX4 ||
1343 dev->devtype->product == CODA_7541) {
1344 coda_write(dev, q_data_src->bytesperline,
1345 CODA7_CMD_SET_FRAME_SOURCE_BUF_STRIDE);
1346 }
1347 if (dev->devtype->product != CODA_DX6) {
1348 coda_write(dev, ctx->iram_info.buf_bit_use,
1349 CODA7_CMD_SET_FRAME_AXI_BIT_ADDR);
1350 coda_write(dev, ctx->iram_info.buf_ip_ac_dc_use,
1351 CODA7_CMD_SET_FRAME_AXI_IPACDC_ADDR);
1352 coda_write(dev, ctx->iram_info.buf_dbk_y_use,
1353 CODA7_CMD_SET_FRAME_AXI_DBKY_ADDR);
1354 coda_write(dev, ctx->iram_info.buf_dbk_c_use,
1355 CODA7_CMD_SET_FRAME_AXI_DBKC_ADDR);
1356 coda_write(dev, ctx->iram_info.buf_ovl_use,
1357 CODA7_CMD_SET_FRAME_AXI_OVL_ADDR);
1358 if (dev->devtype->product == CODA_960) {
1359 coda_write(dev, ctx->iram_info.buf_btp_use,
1360 CODA9_CMD_SET_FRAME_AXI_BTP_ADDR);
1361
1362 coda9_set_frame_cache(ctx, q_data_src->fourcc);
1363
1364 /* FIXME */
1365 coda_write(dev, ctx->internal_frames[2].buf.paddr,
1366 CODA9_CMD_SET_FRAME_SUBSAMP_A);
1367 coda_write(dev, ctx->internal_frames[3].buf.paddr,
1368 CODA9_CMD_SET_FRAME_SUBSAMP_B);
1369 }
1370 }
1371
1372 ret = coda_command_sync(ctx, CODA_COMMAND_SET_FRAME_BUF);
1373 if (ret < 0) {
1374 v4l2_err(v4l2_dev, "CODA_COMMAND_SET_FRAME_BUF timeout\n");
1375 goto out;
1376 }
1377
1378 coda_dbg(1, ctx, "start encoding %dx%d %4.4s->%4.4s @ %d/%d Hz\n",
1379 q_data_src->rect.width, q_data_src->rect.height,
1380 (char *)&ctx->codec->src_fourcc, (char *)&dst_fourcc,
1381 ctx->params.framerate & 0xffff,
1382 (ctx->params.framerate >> 16) + 1);
1383
1384 /* Save stream headers */
1385 buf = v4l2_m2m_next_dst_buf(ctx->fh.m2m_ctx);
1386 switch (dst_fourcc) {
1387 case V4L2_PIX_FMT_H264:
1388 /*
1389 * Get SPS in the first frame and copy it to an
1390 * intermediate buffer.
1391 */
1392 ret = coda_encode_header(ctx, buf, CODA_HEADER_H264_SPS,
1393 &ctx->vpu_header[0][0],
1394 &ctx->vpu_header_size[0]);
1395 if (ret < 0)
1396 goto out;
1397
1398 /*
1399 * If visible width or height are not aligned to macroblock
1400 * size, the crop_right and crop_bottom SPS fields must be set
1401 * to the difference between visible and coded size. This is
1402 * only supported by CODA960 firmware. All others do not allow
1403 * writing frame cropping parameters, so we have to manually
1404 * fix up the SPS RBSP (Sequence Parameter Set Raw Byte
1405 * Sequence Payload) ourselves.
1406 */
1407 if (ctx->dev->devtype->product != CODA_960 &&
1408 ((q_data_src->rect.width % 16) ||
1409 (q_data_src->rect.height % 16))) {
1410 ret = coda_h264_sps_fixup(ctx, q_data_src->rect.width,
1411 q_data_src->rect.height,
1412 &ctx->vpu_header[0][0],
1413 &ctx->vpu_header_size[0],
1414 sizeof(ctx->vpu_header[0]));
1415 if (ret < 0)
1416 goto out;
1417 }
1418
1419 /*
1420 * Get PPS in the first frame and copy it to an
1421 * intermediate buffer.
1422 */
1423 ret = coda_encode_header(ctx, buf, CODA_HEADER_H264_PPS,
1424 &ctx->vpu_header[1][0],
1425 &ctx->vpu_header_size[1]);
1426 if (ret < 0)
1427 goto out;
1428
1429 /*
1430 * Length of H.264 headers is variable and thus it might not be
1431 * aligned for the coda to append the encoded frame. In that is
1432 * the case a filler NAL must be added to header 2.
1433 */
1434 ctx->vpu_header_size[2] = coda_h264_padding(
1435 (ctx->vpu_header_size[0] +
1436 ctx->vpu_header_size[1]),
1437 ctx->vpu_header[2]);
1438 break;
1439 case V4L2_PIX_FMT_MPEG4:
1440 /*
1441 * Get VOS in the first frame and copy it to an
1442 * intermediate buffer
1443 */
1444 ret = coda_encode_header(ctx, buf, CODA_HEADER_MP4V_VOS,
1445 &ctx->vpu_header[0][0],
1446 &ctx->vpu_header_size[0]);
1447 if (ret < 0)
1448 goto out;
1449
1450 ret = coda_encode_header(ctx, buf, CODA_HEADER_MP4V_VIS,
1451 &ctx->vpu_header[1][0],
1452 &ctx->vpu_header_size[1]);
1453 if (ret < 0)
1454 goto out;
1455
1456 ret = coda_encode_header(ctx, buf, CODA_HEADER_MP4V_VOL,
1457 &ctx->vpu_header[2][0],
1458 &ctx->vpu_header_size[2]);
1459 if (ret < 0)
1460 goto out;
1461 break;
1462 default:
1463 /* No more formats need to save headers at the moment */
1464 break;
1465 }
1466
1467 out:
1468 mutex_unlock(&dev->coda_mutex);
1469 return ret;
1470 }
1471
coda_prepare_encode(struct coda_ctx * ctx)1472 static int coda_prepare_encode(struct coda_ctx *ctx)
1473 {
1474 struct coda_q_data *q_data_src, *q_data_dst;
1475 struct vb2_v4l2_buffer *src_buf, *dst_buf;
1476 struct coda_dev *dev = ctx->dev;
1477 int force_ipicture;
1478 int quant_param = 0;
1479 u32 pic_stream_buffer_addr, pic_stream_buffer_size;
1480 u32 rot_mode = 0;
1481 u32 dst_fourcc;
1482 u32 reg;
1483 int ret;
1484
1485 ret = coda_enc_param_change(ctx);
1486 if (ret < 0) {
1487 v4l2_warn(&ctx->dev->v4l2_dev, "parameter change failed: %d\n",
1488 ret);
1489 }
1490
1491 src_buf = v4l2_m2m_next_src_buf(ctx->fh.m2m_ctx);
1492 dst_buf = v4l2_m2m_next_dst_buf(ctx->fh.m2m_ctx);
1493 q_data_src = get_q_data(ctx, V4L2_BUF_TYPE_VIDEO_OUTPUT);
1494 q_data_dst = get_q_data(ctx, V4L2_BUF_TYPE_VIDEO_CAPTURE);
1495 dst_fourcc = q_data_dst->fourcc;
1496
1497 src_buf->sequence = ctx->osequence;
1498 dst_buf->sequence = ctx->osequence;
1499 ctx->osequence++;
1500
1501 force_ipicture = ctx->params.force_ipicture;
1502 if (force_ipicture)
1503 ctx->params.force_ipicture = false;
1504 else if (ctx->params.gop_size != 0 &&
1505 (src_buf->sequence % ctx->params.gop_size) == 0)
1506 force_ipicture = 1;
1507
1508 /*
1509 * Workaround coda firmware BUG that only marks the first
1510 * frame as IDR. This is a problem for some decoders that can't
1511 * recover when a frame is lost.
1512 */
1513 if (!force_ipicture) {
1514 src_buf->flags |= V4L2_BUF_FLAG_PFRAME;
1515 src_buf->flags &= ~V4L2_BUF_FLAG_KEYFRAME;
1516 } else {
1517 src_buf->flags |= V4L2_BUF_FLAG_KEYFRAME;
1518 src_buf->flags &= ~V4L2_BUF_FLAG_PFRAME;
1519 }
1520
1521 if (dev->devtype->product == CODA_960)
1522 coda_set_gdi_regs(ctx);
1523
1524 /*
1525 * Copy headers in front of the first frame and forced I frames for
1526 * H.264 only. In MPEG4 they are already copied by the CODA.
1527 */
1528 if (src_buf->sequence == 0 || force_ipicture) {
1529 pic_stream_buffer_addr =
1530 vb2_dma_contig_plane_dma_addr(&dst_buf->vb2_buf, 0) +
1531 ctx->vpu_header_size[0] +
1532 ctx->vpu_header_size[1] +
1533 ctx->vpu_header_size[2];
1534 pic_stream_buffer_size = q_data_dst->sizeimage -
1535 ctx->vpu_header_size[0] -
1536 ctx->vpu_header_size[1] -
1537 ctx->vpu_header_size[2];
1538 memcpy(vb2_plane_vaddr(&dst_buf->vb2_buf, 0),
1539 &ctx->vpu_header[0][0], ctx->vpu_header_size[0]);
1540 memcpy(vb2_plane_vaddr(&dst_buf->vb2_buf, 0)
1541 + ctx->vpu_header_size[0], &ctx->vpu_header[1][0],
1542 ctx->vpu_header_size[1]);
1543 memcpy(vb2_plane_vaddr(&dst_buf->vb2_buf, 0)
1544 + ctx->vpu_header_size[0] + ctx->vpu_header_size[1],
1545 &ctx->vpu_header[2][0], ctx->vpu_header_size[2]);
1546 } else {
1547 pic_stream_buffer_addr =
1548 vb2_dma_contig_plane_dma_addr(&dst_buf->vb2_buf, 0);
1549 pic_stream_buffer_size = q_data_dst->sizeimage;
1550 }
1551
1552 if (force_ipicture) {
1553 switch (dst_fourcc) {
1554 case V4L2_PIX_FMT_H264:
1555 quant_param = ctx->params.h264_intra_qp;
1556 break;
1557 case V4L2_PIX_FMT_MPEG4:
1558 quant_param = ctx->params.mpeg4_intra_qp;
1559 break;
1560 case V4L2_PIX_FMT_JPEG:
1561 quant_param = 30;
1562 break;
1563 default:
1564 v4l2_warn(&ctx->dev->v4l2_dev,
1565 "cannot set intra qp, fmt not supported\n");
1566 break;
1567 }
1568 } else {
1569 switch (dst_fourcc) {
1570 case V4L2_PIX_FMT_H264:
1571 quant_param = ctx->params.h264_inter_qp;
1572 break;
1573 case V4L2_PIX_FMT_MPEG4:
1574 quant_param = ctx->params.mpeg4_inter_qp;
1575 break;
1576 default:
1577 v4l2_warn(&ctx->dev->v4l2_dev,
1578 "cannot set inter qp, fmt not supported\n");
1579 break;
1580 }
1581 }
1582
1583 /* submit */
1584 if (ctx->params.rot_mode)
1585 rot_mode = CODA_ROT_MIR_ENABLE | ctx->params.rot_mode;
1586 coda_write(dev, rot_mode, CODA_CMD_ENC_PIC_ROT_MODE);
1587 coda_write(dev, quant_param, CODA_CMD_ENC_PIC_QS);
1588
1589 if (dev->devtype->product == CODA_960) {
1590 coda_write(dev, 4/*FIXME: 0*/, CODA9_CMD_ENC_PIC_SRC_INDEX);
1591 coda_write(dev, q_data_src->bytesperline,
1592 CODA9_CMD_ENC_PIC_SRC_STRIDE);
1593 coda_write(dev, 0, CODA9_CMD_ENC_PIC_SUB_FRAME_SYNC);
1594
1595 reg = CODA9_CMD_ENC_PIC_SRC_ADDR_Y;
1596 } else {
1597 reg = CODA_CMD_ENC_PIC_SRC_ADDR_Y;
1598 }
1599 coda_write_base(ctx, q_data_src, src_buf, reg);
1600
1601 coda_write(dev, force_ipicture << 1 & 0x2,
1602 CODA_CMD_ENC_PIC_OPTION);
1603
1604 coda_write(dev, pic_stream_buffer_addr, CODA_CMD_ENC_PIC_BB_START);
1605 coda_write(dev, pic_stream_buffer_size / 1024,
1606 CODA_CMD_ENC_PIC_BB_SIZE);
1607
1608 if (!ctx->streamon_out) {
1609 /* After streamoff on the output side, set stream end flag */
1610 ctx->bit_stream_param |= CODA_BIT_STREAM_END_FLAG;
1611 coda_write(dev, ctx->bit_stream_param,
1612 CODA_REG_BIT_BIT_STREAM_PARAM);
1613 }
1614
1615 if (dev->devtype->product != CODA_DX6)
1616 coda_write(dev, ctx->iram_info.axi_sram_use,
1617 CODA7_REG_BIT_AXI_SRAM_USE);
1618
1619 trace_coda_enc_pic_run(ctx, src_buf);
1620
1621 coda_command_async(ctx, CODA_COMMAND_PIC_RUN);
1622
1623 return 0;
1624 }
1625
coda_frame_type_char(u32 flags)1626 static char coda_frame_type_char(u32 flags)
1627 {
1628 return (flags & V4L2_BUF_FLAG_KEYFRAME) ? 'I' :
1629 (flags & V4L2_BUF_FLAG_PFRAME) ? 'P' :
1630 (flags & V4L2_BUF_FLAG_BFRAME) ? 'B' : '?';
1631 }
1632
coda_finish_encode(struct coda_ctx * ctx)1633 static void coda_finish_encode(struct coda_ctx *ctx)
1634 {
1635 struct vb2_v4l2_buffer *src_buf, *dst_buf;
1636 struct coda_dev *dev = ctx->dev;
1637 u32 wr_ptr, start_ptr;
1638
1639 if (ctx->aborting)
1640 return;
1641
1642 /*
1643 * Lock to make sure that an encoder stop command running in parallel
1644 * will either already have marked src_buf as last, or it will wake up
1645 * the capture queue after the buffers are returned.
1646 */
1647 mutex_lock(&ctx->wakeup_mutex);
1648 src_buf = v4l2_m2m_src_buf_remove(ctx->fh.m2m_ctx);
1649 dst_buf = v4l2_m2m_next_dst_buf(ctx->fh.m2m_ctx);
1650
1651 trace_coda_enc_pic_done(ctx, dst_buf);
1652
1653 /* Get results from the coda */
1654 start_ptr = coda_read(dev, CODA_CMD_ENC_PIC_BB_START);
1655 wr_ptr = coda_read(dev, CODA_REG_BIT_WR_PTR(ctx->reg_idx));
1656
1657 /* Calculate bytesused field */
1658 if (dst_buf->sequence == 0 ||
1659 src_buf->flags & V4L2_BUF_FLAG_KEYFRAME) {
1660 vb2_set_plane_payload(&dst_buf->vb2_buf, 0, wr_ptr - start_ptr +
1661 ctx->vpu_header_size[0] +
1662 ctx->vpu_header_size[1] +
1663 ctx->vpu_header_size[2]);
1664 } else {
1665 vb2_set_plane_payload(&dst_buf->vb2_buf, 0, wr_ptr - start_ptr);
1666 }
1667
1668 coda_dbg(1, ctx, "frame size = %u\n", wr_ptr - start_ptr);
1669
1670 coda_read(dev, CODA_RET_ENC_PIC_SLICE_NUM);
1671 coda_read(dev, CODA_RET_ENC_PIC_FLAG);
1672
1673 dst_buf->flags &= ~(V4L2_BUF_FLAG_KEYFRAME |
1674 V4L2_BUF_FLAG_PFRAME |
1675 V4L2_BUF_FLAG_LAST);
1676 if (coda_read(dev, CODA_RET_ENC_PIC_TYPE) == 0)
1677 dst_buf->flags |= V4L2_BUF_FLAG_KEYFRAME;
1678 else
1679 dst_buf->flags |= V4L2_BUF_FLAG_PFRAME;
1680 dst_buf->flags |= src_buf->flags & V4L2_BUF_FLAG_LAST;
1681
1682 v4l2_m2m_buf_copy_metadata(src_buf, dst_buf, false);
1683
1684 v4l2_m2m_buf_done(src_buf, VB2_BUF_STATE_DONE);
1685
1686 dst_buf = v4l2_m2m_dst_buf_remove(ctx->fh.m2m_ctx);
1687 coda_m2m_buf_done(ctx, dst_buf, VB2_BUF_STATE_DONE);
1688 mutex_unlock(&ctx->wakeup_mutex);
1689
1690 ctx->gopcounter--;
1691 if (ctx->gopcounter < 0)
1692 ctx->gopcounter = ctx->params.gop_size - 1;
1693
1694 coda_dbg(1, ctx, "job finished: encoded %c frame (%d)%s\n",
1695 coda_frame_type_char(dst_buf->flags), dst_buf->sequence,
1696 (dst_buf->flags & V4L2_BUF_FLAG_LAST) ? " (last)" : "");
1697 }
1698
coda_seq_end_work(struct work_struct * work)1699 static void coda_seq_end_work(struct work_struct *work)
1700 {
1701 struct coda_ctx *ctx = container_of(work, struct coda_ctx, seq_end_work);
1702 struct coda_dev *dev = ctx->dev;
1703
1704 mutex_lock(&ctx->buffer_mutex);
1705 mutex_lock(&dev->coda_mutex);
1706
1707 if (ctx->initialized == 0)
1708 goto out;
1709
1710 coda_dbg(1, ctx, "%s: sent command 'SEQ_END' to coda\n", __func__);
1711 if (coda_command_sync(ctx, CODA_COMMAND_SEQ_END)) {
1712 v4l2_err(&dev->v4l2_dev,
1713 "CODA_COMMAND_SEQ_END failed\n");
1714 }
1715
1716 /*
1717 * FIXME: Sometimes h.264 encoding fails with 8-byte sequences missing
1718 * from the output stream after the h.264 decoder has run. Resetting the
1719 * hardware after the decoder has finished seems to help.
1720 */
1721 if (dev->devtype->product == CODA_960)
1722 coda_hw_reset(ctx);
1723
1724 kfifo_init(&ctx->bitstream_fifo,
1725 ctx->bitstream.vaddr, ctx->bitstream.size);
1726
1727 coda_free_framebuffers(ctx);
1728
1729 ctx->initialized = 0;
1730
1731 out:
1732 mutex_unlock(&dev->coda_mutex);
1733 mutex_unlock(&ctx->buffer_mutex);
1734 }
1735
coda_bit_release(struct coda_ctx * ctx)1736 static void coda_bit_release(struct coda_ctx *ctx)
1737 {
1738 mutex_lock(&ctx->buffer_mutex);
1739 coda_free_framebuffers(ctx);
1740 coda_free_context_buffers(ctx);
1741 coda_free_bitstream_buffer(ctx);
1742 mutex_unlock(&ctx->buffer_mutex);
1743 }
1744
1745 const struct coda_context_ops coda_bit_encode_ops = {
1746 .queue_init = coda_encoder_queue_init,
1747 .reqbufs = coda_encoder_reqbufs,
1748 .start_streaming = coda_start_encoding,
1749 .prepare_run = coda_prepare_encode,
1750 .finish_run = coda_finish_encode,
1751 .seq_end_work = coda_seq_end_work,
1752 .release = coda_bit_release,
1753 };
1754
1755 /*
1756 * Decoder context operations
1757 */
1758
coda_alloc_bitstream_buffer(struct coda_ctx * ctx,struct coda_q_data * q_data)1759 static int coda_alloc_bitstream_buffer(struct coda_ctx *ctx,
1760 struct coda_q_data *q_data)
1761 {
1762 if (ctx->bitstream.vaddr)
1763 return 0;
1764
1765 ctx->bitstream.size = roundup_pow_of_two(q_data->sizeimage * 2);
1766 ctx->bitstream.vaddr = dma_alloc_wc(ctx->dev->dev, ctx->bitstream.size,
1767 &ctx->bitstream.paddr, GFP_KERNEL);
1768 if (!ctx->bitstream.vaddr) {
1769 v4l2_err(&ctx->dev->v4l2_dev,
1770 "failed to allocate bitstream ringbuffer");
1771 return -ENOMEM;
1772 }
1773 kfifo_init(&ctx->bitstream_fifo,
1774 ctx->bitstream.vaddr, ctx->bitstream.size);
1775
1776 return 0;
1777 }
1778
coda_free_bitstream_buffer(struct coda_ctx * ctx)1779 static void coda_free_bitstream_buffer(struct coda_ctx *ctx)
1780 {
1781 if (ctx->bitstream.vaddr == NULL)
1782 return;
1783
1784 dma_free_wc(ctx->dev->dev, ctx->bitstream.size, ctx->bitstream.vaddr,
1785 ctx->bitstream.paddr);
1786 ctx->bitstream.vaddr = NULL;
1787 kfifo_init(&ctx->bitstream_fifo, NULL, 0);
1788 }
1789
coda_decoder_reqbufs(struct coda_ctx * ctx,struct v4l2_requestbuffers * rb)1790 static int coda_decoder_reqbufs(struct coda_ctx *ctx,
1791 struct v4l2_requestbuffers *rb)
1792 {
1793 struct coda_q_data *q_data_src;
1794 int ret;
1795
1796 if (rb->type != V4L2_BUF_TYPE_VIDEO_OUTPUT)
1797 return 0;
1798
1799 if (rb->count) {
1800 q_data_src = get_q_data(ctx, V4L2_BUF_TYPE_VIDEO_OUTPUT);
1801 ret = coda_alloc_context_buffers(ctx, q_data_src);
1802 if (ret < 0)
1803 return ret;
1804 ret = coda_alloc_bitstream_buffer(ctx, q_data_src);
1805 if (ret < 0) {
1806 coda_free_context_buffers(ctx);
1807 return ret;
1808 }
1809 } else {
1810 coda_free_bitstream_buffer(ctx);
1811 coda_free_context_buffers(ctx);
1812 }
1813
1814 return 0;
1815 }
1816
coda_reorder_enable(struct coda_ctx * ctx)1817 static bool coda_reorder_enable(struct coda_ctx *ctx)
1818 {
1819 struct coda_dev *dev = ctx->dev;
1820 int profile;
1821
1822 if (dev->devtype->product != CODA_HX4 &&
1823 dev->devtype->product != CODA_7541 &&
1824 dev->devtype->product != CODA_960)
1825 return false;
1826
1827 if (ctx->codec->src_fourcc == V4L2_PIX_FMT_JPEG)
1828 return false;
1829
1830 if (ctx->codec->src_fourcc != V4L2_PIX_FMT_H264)
1831 return true;
1832
1833 profile = coda_h264_profile(ctx->params.h264_profile_idc);
1834 if (profile < 0)
1835 v4l2_warn(&dev->v4l2_dev, "Unknown H264 Profile: %u\n",
1836 ctx->params.h264_profile_idc);
1837
1838 /* Baseline profile does not support reordering */
1839 return profile > V4L2_MPEG_VIDEO_H264_PROFILE_BASELINE;
1840 }
1841
coda_decoder_drop_used_metas(struct coda_ctx * ctx)1842 static void coda_decoder_drop_used_metas(struct coda_ctx *ctx)
1843 {
1844 struct coda_buffer_meta *meta, *tmp;
1845
1846 /*
1847 * All metas that end at or before the RD pointer (fifo out),
1848 * are now consumed by the VPU and should be released.
1849 */
1850 spin_lock(&ctx->buffer_meta_lock);
1851 list_for_each_entry_safe(meta, tmp, &ctx->buffer_meta_list, list) {
1852 if (ctx->bitstream_fifo.kfifo.out >= meta->end) {
1853 coda_dbg(2, ctx, "releasing meta: seq=%d start=%d end=%d\n",
1854 meta->sequence, meta->start, meta->end);
1855
1856 list_del(&meta->list);
1857 ctx->num_metas--;
1858 ctx->first_frame_sequence++;
1859 kfree(meta);
1860 }
1861 }
1862 spin_unlock(&ctx->buffer_meta_lock);
1863 }
1864
__coda_decoder_seq_init(struct coda_ctx * ctx)1865 static int __coda_decoder_seq_init(struct coda_ctx *ctx)
1866 {
1867 struct coda_q_data *q_data_src, *q_data_dst;
1868 u32 bitstream_buf, bitstream_size;
1869 struct coda_dev *dev = ctx->dev;
1870 int width, height;
1871 u32 src_fourcc, dst_fourcc;
1872 u32 val;
1873 int ret;
1874
1875 lockdep_assert_held(&dev->coda_mutex);
1876
1877 coda_dbg(1, ctx, "Video Data Order Adapter: %s\n",
1878 ctx->use_vdoa ? "Enabled" : "Disabled");
1879
1880 /* Start decoding */
1881 q_data_src = get_q_data(ctx, V4L2_BUF_TYPE_VIDEO_OUTPUT);
1882 q_data_dst = get_q_data(ctx, V4L2_BUF_TYPE_VIDEO_CAPTURE);
1883 bitstream_buf = ctx->bitstream.paddr;
1884 bitstream_size = ctx->bitstream.size;
1885 src_fourcc = q_data_src->fourcc;
1886 dst_fourcc = q_data_dst->fourcc;
1887
1888 /* Update coda bitstream read and write pointers from kfifo */
1889 coda_kfifo_sync_to_device_full(ctx);
1890
1891 ctx->frame_mem_ctrl &= ~(CODA_FRAME_CHROMA_INTERLEAVE | (0x3 << 9) |
1892 CODA9_FRAME_TILED2LINEAR);
1893 if (dst_fourcc == V4L2_PIX_FMT_NV12 || dst_fourcc == V4L2_PIX_FMT_YUYV)
1894 ctx->frame_mem_ctrl |= CODA_FRAME_CHROMA_INTERLEAVE;
1895 if (ctx->tiled_map_type == GDI_TILED_FRAME_MB_RASTER_MAP)
1896 ctx->frame_mem_ctrl |= (0x3 << 9) |
1897 ((ctx->use_vdoa) ? 0 : CODA9_FRAME_TILED2LINEAR);
1898 coda_write(dev, ctx->frame_mem_ctrl, CODA_REG_BIT_FRAME_MEM_CTRL);
1899
1900 ctx->display_idx = -1;
1901 ctx->frm_dis_flg = 0;
1902 coda_write(dev, 0, CODA_REG_BIT_FRM_DIS_FLG(ctx->reg_idx));
1903
1904 coda_write(dev, bitstream_buf, CODA_CMD_DEC_SEQ_BB_START);
1905 coda_write(dev, bitstream_size / 1024, CODA_CMD_DEC_SEQ_BB_SIZE);
1906 val = 0;
1907 if (coda_reorder_enable(ctx))
1908 val |= CODA_REORDER_ENABLE;
1909 if (ctx->codec->src_fourcc == V4L2_PIX_FMT_JPEG)
1910 val |= CODA_NO_INT_ENABLE;
1911 coda_write(dev, val, CODA_CMD_DEC_SEQ_OPTION);
1912
1913 ctx->params.codec_mode = ctx->codec->mode;
1914 if (dev->devtype->product == CODA_960 &&
1915 src_fourcc == V4L2_PIX_FMT_MPEG4)
1916 ctx->params.codec_mode_aux = CODA_MP4_AUX_MPEG4;
1917 else
1918 ctx->params.codec_mode_aux = 0;
1919 if (src_fourcc == V4L2_PIX_FMT_MPEG4) {
1920 coda_write(dev, CODA_MP4_CLASS_MPEG4,
1921 CODA_CMD_DEC_SEQ_MP4_ASP_CLASS);
1922 }
1923 if (src_fourcc == V4L2_PIX_FMT_H264) {
1924 if (dev->devtype->product == CODA_HX4 ||
1925 dev->devtype->product == CODA_7541) {
1926 coda_write(dev, ctx->psbuf.paddr,
1927 CODA_CMD_DEC_SEQ_PS_BB_START);
1928 coda_write(dev, (CODA7_PS_BUF_SIZE / 1024),
1929 CODA_CMD_DEC_SEQ_PS_BB_SIZE);
1930 }
1931 if (dev->devtype->product == CODA_960) {
1932 coda_write(dev, 0, CODA_CMD_DEC_SEQ_X264_MV_EN);
1933 coda_write(dev, 512, CODA_CMD_DEC_SEQ_SPP_CHUNK_SIZE);
1934 }
1935 }
1936 if (src_fourcc == V4L2_PIX_FMT_JPEG)
1937 coda_write(dev, 0, CODA_CMD_DEC_SEQ_JPG_THUMB_EN);
1938 if (dev->devtype->product != CODA_960)
1939 coda_write(dev, 0, CODA_CMD_DEC_SEQ_SRC_SIZE);
1940
1941 ctx->bit_stream_param = CODA_BIT_DEC_SEQ_INIT_ESCAPE;
1942 ret = coda_command_sync(ctx, CODA_COMMAND_SEQ_INIT);
1943 ctx->bit_stream_param = 0;
1944 if (ret) {
1945 v4l2_err(&dev->v4l2_dev, "CODA_COMMAND_SEQ_INIT timeout\n");
1946 return ret;
1947 }
1948 ctx->sequence_offset = ~0U;
1949 ctx->initialized = 1;
1950 ctx->first_frame_sequence = 0;
1951
1952 /* Update kfifo out pointer from coda bitstream read pointer */
1953 coda_kfifo_sync_from_device(ctx);
1954
1955 /*
1956 * After updating the read pointer, we need to check if
1957 * any metas are consumed and should be released.
1958 */
1959 coda_decoder_drop_used_metas(ctx);
1960
1961 if (coda_read(dev, CODA_RET_DEC_SEQ_SUCCESS) == 0) {
1962 v4l2_err(&dev->v4l2_dev,
1963 "CODA_COMMAND_SEQ_INIT failed, error code = 0x%x\n",
1964 coda_read(dev, CODA_RET_DEC_SEQ_ERR_REASON));
1965 return -EAGAIN;
1966 }
1967
1968 val = coda_read(dev, CODA_RET_DEC_SEQ_SRC_SIZE);
1969 if (dev->devtype->product == CODA_DX6) {
1970 width = (val >> CODADX6_PICWIDTH_OFFSET) & CODADX6_PICWIDTH_MASK;
1971 height = val & CODADX6_PICHEIGHT_MASK;
1972 } else {
1973 width = (val >> CODA7_PICWIDTH_OFFSET) & CODA7_PICWIDTH_MASK;
1974 height = val & CODA7_PICHEIGHT_MASK;
1975 }
1976
1977 if (width > q_data_dst->bytesperline || height > q_data_dst->height) {
1978 v4l2_err(&dev->v4l2_dev, "stream is %dx%d, not %dx%d\n",
1979 width, height, q_data_dst->bytesperline,
1980 q_data_dst->height);
1981 return -EINVAL;
1982 }
1983
1984 width = round_up(width, 16);
1985 height = round_up(height, 16);
1986
1987 coda_dbg(1, ctx, "start decoding: %dx%d\n", width, height);
1988
1989 ctx->num_internal_frames = coda_read(dev, CODA_RET_DEC_SEQ_FRAME_NEED);
1990 /*
1991 * If the VDOA is used, the decoder needs one additional frame,
1992 * because the frames are freed when the next frame is decoded.
1993 * Otherwise there are visible errors in the decoded frames (green
1994 * regions in displayed frames) and a broken order of frames (earlier
1995 * frames are sporadically displayed after later frames).
1996 */
1997 if (ctx->use_vdoa)
1998 ctx->num_internal_frames += 1;
1999 if (ctx->num_internal_frames > CODA_MAX_FRAMEBUFFERS) {
2000 v4l2_err(&dev->v4l2_dev,
2001 "not enough framebuffers to decode (%d < %d)\n",
2002 CODA_MAX_FRAMEBUFFERS, ctx->num_internal_frames);
2003 return -EINVAL;
2004 }
2005
2006 if (src_fourcc == V4L2_PIX_FMT_H264) {
2007 u32 left_right;
2008 u32 top_bottom;
2009
2010 left_right = coda_read(dev, CODA_RET_DEC_SEQ_CROP_LEFT_RIGHT);
2011 top_bottom = coda_read(dev, CODA_RET_DEC_SEQ_CROP_TOP_BOTTOM);
2012
2013 q_data_dst->rect.left = (left_right >> 10) & 0x3ff;
2014 q_data_dst->rect.top = (top_bottom >> 10) & 0x3ff;
2015 q_data_dst->rect.width = width - q_data_dst->rect.left -
2016 (left_right & 0x3ff);
2017 q_data_dst->rect.height = height - q_data_dst->rect.top -
2018 (top_bottom & 0x3ff);
2019 }
2020
2021 if (dev->devtype->product != CODA_DX6) {
2022 u8 profile, level;
2023
2024 val = coda_read(dev, CODA7_RET_DEC_SEQ_HEADER_REPORT);
2025 profile = val & 0xff;
2026 level = (val >> 8) & 0x7f;
2027
2028 if (profile || level)
2029 coda_update_profile_level_ctrls(ctx, profile, level);
2030 }
2031
2032 return 0;
2033 }
2034
coda_dec_seq_init_work(struct work_struct * work)2035 static void coda_dec_seq_init_work(struct work_struct *work)
2036 {
2037 struct coda_ctx *ctx = container_of(work,
2038 struct coda_ctx, seq_init_work);
2039 struct coda_dev *dev = ctx->dev;
2040
2041 mutex_lock(&ctx->buffer_mutex);
2042 mutex_lock(&dev->coda_mutex);
2043
2044 if (!ctx->initialized)
2045 __coda_decoder_seq_init(ctx);
2046
2047 mutex_unlock(&dev->coda_mutex);
2048 mutex_unlock(&ctx->buffer_mutex);
2049 }
2050
__coda_start_decoding(struct coda_ctx * ctx)2051 static int __coda_start_decoding(struct coda_ctx *ctx)
2052 {
2053 struct coda_q_data *q_data_src, *q_data_dst;
2054 struct coda_dev *dev = ctx->dev;
2055 u32 src_fourcc, dst_fourcc;
2056 int ret;
2057
2058 q_data_src = get_q_data(ctx, V4L2_BUF_TYPE_VIDEO_OUTPUT);
2059 q_data_dst = get_q_data(ctx, V4L2_BUF_TYPE_VIDEO_CAPTURE);
2060 src_fourcc = q_data_src->fourcc;
2061 dst_fourcc = q_data_dst->fourcc;
2062
2063 if (!ctx->initialized) {
2064 ret = __coda_decoder_seq_init(ctx);
2065 if (ret < 0)
2066 return ret;
2067 } else {
2068 ctx->frame_mem_ctrl &= ~(CODA_FRAME_CHROMA_INTERLEAVE | (0x3 << 9) |
2069 CODA9_FRAME_TILED2LINEAR);
2070 if (dst_fourcc == V4L2_PIX_FMT_NV12 || dst_fourcc == V4L2_PIX_FMT_YUYV)
2071 ctx->frame_mem_ctrl |= CODA_FRAME_CHROMA_INTERLEAVE;
2072 if (ctx->tiled_map_type == GDI_TILED_FRAME_MB_RASTER_MAP)
2073 ctx->frame_mem_ctrl |= (0x3 << 9) |
2074 ((ctx->use_vdoa) ? 0 : CODA9_FRAME_TILED2LINEAR);
2075 }
2076
2077 coda_write(dev, ctx->parabuf.paddr, CODA_REG_BIT_PARA_BUF_ADDR);
2078
2079 ret = coda_alloc_framebuffers(ctx, q_data_dst, src_fourcc);
2080 if (ret < 0) {
2081 v4l2_err(&dev->v4l2_dev, "failed to allocate framebuffers\n");
2082 return ret;
2083 }
2084
2085 /* Tell the decoder how many frame buffers we allocated. */
2086 coda_write(dev, ctx->num_internal_frames, CODA_CMD_SET_FRAME_BUF_NUM);
2087 coda_write(dev, round_up(q_data_dst->rect.width, 16),
2088 CODA_CMD_SET_FRAME_BUF_STRIDE);
2089
2090 if (dev->devtype->product != CODA_DX6) {
2091 /* Set secondary AXI IRAM */
2092 coda_setup_iram(ctx);
2093
2094 coda_write(dev, ctx->iram_info.buf_bit_use,
2095 CODA7_CMD_SET_FRAME_AXI_BIT_ADDR);
2096 coda_write(dev, ctx->iram_info.buf_ip_ac_dc_use,
2097 CODA7_CMD_SET_FRAME_AXI_IPACDC_ADDR);
2098 coda_write(dev, ctx->iram_info.buf_dbk_y_use,
2099 CODA7_CMD_SET_FRAME_AXI_DBKY_ADDR);
2100 coda_write(dev, ctx->iram_info.buf_dbk_c_use,
2101 CODA7_CMD_SET_FRAME_AXI_DBKC_ADDR);
2102 coda_write(dev, ctx->iram_info.buf_ovl_use,
2103 CODA7_CMD_SET_FRAME_AXI_OVL_ADDR);
2104 if (dev->devtype->product == CODA_960) {
2105 coda_write(dev, ctx->iram_info.buf_btp_use,
2106 CODA9_CMD_SET_FRAME_AXI_BTP_ADDR);
2107
2108 coda_write(dev, -1, CODA9_CMD_SET_FRAME_DELAY);
2109 coda9_set_frame_cache(ctx, dst_fourcc);
2110 }
2111 }
2112
2113 if (src_fourcc == V4L2_PIX_FMT_H264) {
2114 coda_write(dev, ctx->slicebuf.paddr,
2115 CODA_CMD_SET_FRAME_SLICE_BB_START);
2116 coda_write(dev, ctx->slicebuf.size / 1024,
2117 CODA_CMD_SET_FRAME_SLICE_BB_SIZE);
2118 }
2119
2120 if (dev->devtype->product == CODA_HX4 ||
2121 dev->devtype->product == CODA_7541) {
2122 int max_mb_x = 1920 / 16;
2123 int max_mb_y = 1088 / 16;
2124 int max_mb_num = max_mb_x * max_mb_y;
2125
2126 coda_write(dev, max_mb_num << 16 | max_mb_x << 8 | max_mb_y,
2127 CODA7_CMD_SET_FRAME_MAX_DEC_SIZE);
2128 } else if (dev->devtype->product == CODA_960) {
2129 int max_mb_x = 1920 / 16;
2130 int max_mb_y = 1088 / 16;
2131 int max_mb_num = max_mb_x * max_mb_y;
2132
2133 coda_write(dev, max_mb_num << 16 | max_mb_x << 8 | max_mb_y,
2134 CODA9_CMD_SET_FRAME_MAX_DEC_SIZE);
2135 }
2136
2137 if (coda_command_sync(ctx, CODA_COMMAND_SET_FRAME_BUF)) {
2138 v4l2_err(&ctx->dev->v4l2_dev,
2139 "CODA_COMMAND_SET_FRAME_BUF timeout\n");
2140 return -ETIMEDOUT;
2141 }
2142
2143 return 0;
2144 }
2145
coda_start_decoding(struct coda_ctx * ctx)2146 static int coda_start_decoding(struct coda_ctx *ctx)
2147 {
2148 struct coda_dev *dev = ctx->dev;
2149 int ret;
2150
2151 mutex_lock(&dev->coda_mutex);
2152 ret = __coda_start_decoding(ctx);
2153 mutex_unlock(&dev->coda_mutex);
2154
2155 return ret;
2156 }
2157
coda_prepare_decode(struct coda_ctx * ctx)2158 static int coda_prepare_decode(struct coda_ctx *ctx)
2159 {
2160 struct vb2_v4l2_buffer *dst_buf;
2161 struct coda_dev *dev = ctx->dev;
2162 struct coda_q_data *q_data_dst;
2163 struct coda_buffer_meta *meta;
2164 u32 rot_mode = 0;
2165 u32 reg_addr, reg_stride;
2166
2167 dst_buf = v4l2_m2m_next_dst_buf(ctx->fh.m2m_ctx);
2168 q_data_dst = get_q_data(ctx, V4L2_BUF_TYPE_VIDEO_CAPTURE);
2169
2170 /* Try to copy source buffer contents into the bitstream ringbuffer */
2171 mutex_lock(&ctx->bitstream_mutex);
2172 coda_fill_bitstream(ctx, NULL);
2173 mutex_unlock(&ctx->bitstream_mutex);
2174
2175 if (coda_get_bitstream_payload(ctx) < 512 &&
2176 (!(ctx->bit_stream_param & CODA_BIT_STREAM_END_FLAG))) {
2177 coda_dbg(1, ctx, "bitstream payload: %d, skipping\n",
2178 coda_get_bitstream_payload(ctx));
2179 return -EAGAIN;
2180 }
2181
2182 /* Run coda_start_decoding (again) if not yet initialized */
2183 if (!ctx->initialized) {
2184 int ret = __coda_start_decoding(ctx);
2185
2186 if (ret < 0) {
2187 v4l2_err(&dev->v4l2_dev, "failed to start decoding\n");
2188 return -EAGAIN;
2189 } else {
2190 ctx->initialized = 1;
2191 }
2192 }
2193
2194 if (dev->devtype->product == CODA_960)
2195 coda_set_gdi_regs(ctx);
2196
2197 if (ctx->use_vdoa &&
2198 ctx->display_idx >= 0 &&
2199 ctx->display_idx < ctx->num_internal_frames) {
2200 vdoa_device_run(ctx->vdoa,
2201 vb2_dma_contig_plane_dma_addr(&dst_buf->vb2_buf, 0),
2202 ctx->internal_frames[ctx->display_idx].buf.paddr);
2203 } else {
2204 if (dev->devtype->product == CODA_960) {
2205 /*
2206 * It was previously assumed that the CODA960 has an
2207 * internal list of 64 buffer entries that contains
2208 * both the registered internal frame buffers as well
2209 * as the rotator buffer output, and that the ROT_INDEX
2210 * register must be set to a value between the last
2211 * internal frame buffers' index and 64.
2212 * At least on firmware version 3.1.1 it turns out that
2213 * setting ROT_INDEX to any value >= 32 causes CODA
2214 * hangups that it can not recover from with the SRC VPU
2215 * reset.
2216 * It does appear to work however, to just set it to a
2217 * fixed value in the [ctx->num_internal_frames, 31]
2218 * range, for example CODA_MAX_FRAMEBUFFERS.
2219 */
2220 coda_write(dev, CODA_MAX_FRAMEBUFFERS,
2221 CODA9_CMD_DEC_PIC_ROT_INDEX);
2222
2223 reg_addr = CODA9_CMD_DEC_PIC_ROT_ADDR_Y;
2224 reg_stride = CODA9_CMD_DEC_PIC_ROT_STRIDE;
2225 } else {
2226 reg_addr = CODA_CMD_DEC_PIC_ROT_ADDR_Y;
2227 reg_stride = CODA_CMD_DEC_PIC_ROT_STRIDE;
2228 }
2229 coda_write_base(ctx, q_data_dst, dst_buf, reg_addr);
2230 coda_write(dev, q_data_dst->bytesperline, reg_stride);
2231
2232 rot_mode = CODA_ROT_MIR_ENABLE | ctx->params.rot_mode;
2233 }
2234
2235 coda_write(dev, rot_mode, CODA_CMD_DEC_PIC_ROT_MODE);
2236
2237 switch (dev->devtype->product) {
2238 case CODA_DX6:
2239 /* TBD */
2240 case CODA_HX4:
2241 case CODA_7541:
2242 coda_write(dev, CODA_PRE_SCAN_EN, CODA_CMD_DEC_PIC_OPTION);
2243 break;
2244 case CODA_960:
2245 /* 'hardcode to use interrupt disable mode'? */
2246 coda_write(dev, (1 << 10), CODA_CMD_DEC_PIC_OPTION);
2247 break;
2248 }
2249
2250 coda_write(dev, 0, CODA_CMD_DEC_PIC_SKIP_NUM);
2251
2252 coda_write(dev, 0, CODA_CMD_DEC_PIC_BB_START);
2253 coda_write(dev, 0, CODA_CMD_DEC_PIC_START_BYTE);
2254
2255 if (dev->devtype->product != CODA_DX6)
2256 coda_write(dev, ctx->iram_info.axi_sram_use,
2257 CODA7_REG_BIT_AXI_SRAM_USE);
2258
2259 spin_lock(&ctx->buffer_meta_lock);
2260 meta = list_first_entry_or_null(&ctx->buffer_meta_list,
2261 struct coda_buffer_meta, list);
2262
2263 if (meta && ctx->codec->src_fourcc == V4L2_PIX_FMT_JPEG) {
2264
2265 /* If this is the last buffer in the bitstream, add padding */
2266 if (meta->end == ctx->bitstream_fifo.kfifo.in) {
2267 static unsigned char buf[512];
2268 unsigned int pad;
2269
2270 /* Pad to multiple of 256 and then add 256 more */
2271 pad = ((0 - meta->end) & 0xff) + 256;
2272
2273 memset(buf, 0xff, sizeof(buf));
2274
2275 kfifo_in(&ctx->bitstream_fifo, buf, pad);
2276 }
2277 }
2278 spin_unlock(&ctx->buffer_meta_lock);
2279
2280 coda_kfifo_sync_to_device_full(ctx);
2281
2282 /* Clear decode success flag */
2283 coda_write(dev, 0, CODA_RET_DEC_PIC_SUCCESS);
2284
2285 /* Clear error return value */
2286 coda_write(dev, 0, CODA_RET_DEC_PIC_ERR_MB);
2287
2288 trace_coda_dec_pic_run(ctx, meta);
2289
2290 coda_command_async(ctx, CODA_COMMAND_PIC_RUN);
2291
2292 return 0;
2293 }
2294
coda_finish_decode(struct coda_ctx * ctx)2295 static void coda_finish_decode(struct coda_ctx *ctx)
2296 {
2297 struct coda_dev *dev = ctx->dev;
2298 struct coda_q_data *q_data_src;
2299 struct coda_q_data *q_data_dst;
2300 struct vb2_v4l2_buffer *dst_buf;
2301 struct coda_buffer_meta *meta;
2302 int width, height;
2303 int decoded_idx;
2304 int display_idx;
2305 struct coda_internal_frame *decoded_frame = NULL;
2306 u32 src_fourcc;
2307 int success;
2308 u32 err_mb;
2309 int err_vdoa = 0;
2310 u32 val;
2311
2312 if (ctx->aborting)
2313 return;
2314
2315 /* Update kfifo out pointer from coda bitstream read pointer */
2316 coda_kfifo_sync_from_device(ctx);
2317
2318 /*
2319 * in stream-end mode, the read pointer can overshoot the write pointer
2320 * by up to 512 bytes
2321 */
2322 if (ctx->bit_stream_param & CODA_BIT_STREAM_END_FLAG) {
2323 if (coda_get_bitstream_payload(ctx) >= ctx->bitstream.size - 512)
2324 kfifo_init(&ctx->bitstream_fifo,
2325 ctx->bitstream.vaddr, ctx->bitstream.size);
2326 }
2327
2328 q_data_src = get_q_data(ctx, V4L2_BUF_TYPE_VIDEO_OUTPUT);
2329 src_fourcc = q_data_src->fourcc;
2330
2331 val = coda_read(dev, CODA_RET_DEC_PIC_SUCCESS);
2332 if (val != 1)
2333 pr_err("DEC_PIC_SUCCESS = %d\n", val);
2334
2335 success = val & 0x1;
2336 if (!success)
2337 v4l2_err(&dev->v4l2_dev, "decode failed\n");
2338
2339 if (src_fourcc == V4L2_PIX_FMT_H264) {
2340 if (val & (1 << 3))
2341 v4l2_err(&dev->v4l2_dev,
2342 "insufficient PS buffer space (%d bytes)\n",
2343 ctx->psbuf.size);
2344 if (val & (1 << 2))
2345 v4l2_err(&dev->v4l2_dev,
2346 "insufficient slice buffer space (%d bytes)\n",
2347 ctx->slicebuf.size);
2348 }
2349
2350 val = coda_read(dev, CODA_RET_DEC_PIC_SIZE);
2351 width = (val >> 16) & 0xffff;
2352 height = val & 0xffff;
2353
2354 q_data_dst = get_q_data(ctx, V4L2_BUF_TYPE_VIDEO_CAPTURE);
2355
2356 /* frame crop information */
2357 if (src_fourcc == V4L2_PIX_FMT_H264) {
2358 u32 left_right;
2359 u32 top_bottom;
2360
2361 left_right = coda_read(dev, CODA_RET_DEC_PIC_CROP_LEFT_RIGHT);
2362 top_bottom = coda_read(dev, CODA_RET_DEC_PIC_CROP_TOP_BOTTOM);
2363
2364 if (left_right == 0xffffffff && top_bottom == 0xffffffff) {
2365 /* Keep current crop information */
2366 } else {
2367 struct v4l2_rect *rect = &q_data_dst->rect;
2368
2369 rect->left = left_right >> 16 & 0xffff;
2370 rect->top = top_bottom >> 16 & 0xffff;
2371 rect->width = width - rect->left -
2372 (left_right & 0xffff);
2373 rect->height = height - rect->top -
2374 (top_bottom & 0xffff);
2375 }
2376 } else {
2377 /* no cropping */
2378 }
2379
2380 err_mb = coda_read(dev, CODA_RET_DEC_PIC_ERR_MB);
2381 if (err_mb > 0) {
2382 if (__ratelimit(&dev->mb_err_rs))
2383 coda_dbg(1, ctx, "errors in %d macroblocks\n", err_mb);
2384 v4l2_ctrl_s_ctrl(ctx->mb_err_cnt_ctrl,
2385 v4l2_ctrl_g_ctrl(ctx->mb_err_cnt_ctrl) + err_mb);
2386 }
2387
2388 if (dev->devtype->product == CODA_HX4 ||
2389 dev->devtype->product == CODA_7541) {
2390 val = coda_read(dev, CODA_RET_DEC_PIC_OPTION);
2391 if (val == 0) {
2392 /* not enough bitstream data */
2393 coda_dbg(1, ctx, "prescan failed: %d\n", val);
2394 ctx->hold = true;
2395 return;
2396 }
2397 }
2398
2399 /* Wait until the VDOA finished writing the previous display frame */
2400 if (ctx->use_vdoa &&
2401 ctx->display_idx >= 0 &&
2402 ctx->display_idx < ctx->num_internal_frames) {
2403 err_vdoa = vdoa_wait_for_completion(ctx->vdoa);
2404 }
2405
2406 ctx->frm_dis_flg = coda_read(dev,
2407 CODA_REG_BIT_FRM_DIS_FLG(ctx->reg_idx));
2408
2409 /* The previous display frame was copied out and can be overwritten */
2410 if (ctx->display_idx >= 0 &&
2411 ctx->display_idx < ctx->num_internal_frames) {
2412 ctx->frm_dis_flg &= ~(1 << ctx->display_idx);
2413 coda_write(dev, ctx->frm_dis_flg,
2414 CODA_REG_BIT_FRM_DIS_FLG(ctx->reg_idx));
2415 }
2416
2417 /*
2418 * The index of the last decoded frame, not necessarily in
2419 * display order, and the index of the next display frame.
2420 * The latter could have been decoded in a previous run.
2421 */
2422 decoded_idx = coda_read(dev, CODA_RET_DEC_PIC_CUR_IDX);
2423 display_idx = coda_read(dev, CODA_RET_DEC_PIC_FRAME_IDX);
2424
2425 if (decoded_idx == -1) {
2426 /* no frame was decoded, but we might have a display frame */
2427 if (display_idx >= 0 && display_idx < ctx->num_internal_frames)
2428 ctx->sequence_offset++;
2429 else if (ctx->display_idx < 0)
2430 ctx->hold = true;
2431 } else if (decoded_idx == -2) {
2432 if (ctx->display_idx >= 0 &&
2433 ctx->display_idx < ctx->num_internal_frames)
2434 ctx->sequence_offset++;
2435 /* no frame was decoded, we still return remaining buffers */
2436 } else if (decoded_idx < 0 || decoded_idx >= ctx->num_internal_frames) {
2437 v4l2_err(&dev->v4l2_dev,
2438 "decoded frame index out of range: %d\n", decoded_idx);
2439 } else {
2440 int sequence;
2441
2442 decoded_frame = &ctx->internal_frames[decoded_idx];
2443
2444 val = coda_read(dev, CODA_RET_DEC_PIC_FRAME_NUM);
2445 if (ctx->sequence_offset == -1)
2446 ctx->sequence_offset = val;
2447
2448 sequence = val + ctx->first_frame_sequence
2449 - ctx->sequence_offset;
2450 spin_lock(&ctx->buffer_meta_lock);
2451 if (!list_empty(&ctx->buffer_meta_list)) {
2452 meta = list_first_entry(&ctx->buffer_meta_list,
2453 struct coda_buffer_meta, list);
2454 list_del(&meta->list);
2455 ctx->num_metas--;
2456 spin_unlock(&ctx->buffer_meta_lock);
2457 /*
2458 * Clamp counters to 16 bits for comparison, as the HW
2459 * counter rolls over at this point for h.264. This
2460 * may be different for other formats, but using 16 bits
2461 * should be enough to detect most errors and saves us
2462 * from doing different things based on the format.
2463 */
2464 if ((sequence & 0xffff) != (meta->sequence & 0xffff)) {
2465 v4l2_err(&dev->v4l2_dev,
2466 "sequence number mismatch (%d(%d) != %d)\n",
2467 sequence, ctx->sequence_offset,
2468 meta->sequence);
2469 }
2470 decoded_frame->meta = *meta;
2471 kfree(meta);
2472 } else {
2473 spin_unlock(&ctx->buffer_meta_lock);
2474 v4l2_err(&dev->v4l2_dev, "empty timestamp list!\n");
2475 memset(&decoded_frame->meta, 0,
2476 sizeof(struct coda_buffer_meta));
2477 decoded_frame->meta.sequence = sequence;
2478 decoded_frame->meta.last = false;
2479 ctx->sequence_offset++;
2480 }
2481
2482 trace_coda_dec_pic_done(ctx, &decoded_frame->meta);
2483
2484 val = coda_read(dev, CODA_RET_DEC_PIC_TYPE) & 0x7;
2485 decoded_frame->type = (val == 0) ? V4L2_BUF_FLAG_KEYFRAME :
2486 (val == 1) ? V4L2_BUF_FLAG_PFRAME :
2487 V4L2_BUF_FLAG_BFRAME;
2488
2489 decoded_frame->error = err_mb;
2490 }
2491
2492 if (display_idx == -1) {
2493 /*
2494 * no more frames to be decoded, but there could still
2495 * be rotator output to dequeue
2496 */
2497 ctx->hold = true;
2498 } else if (display_idx == -3) {
2499 /* possibly prescan failure */
2500 } else if (display_idx < 0 || display_idx >= ctx->num_internal_frames) {
2501 v4l2_err(&dev->v4l2_dev,
2502 "presentation frame index out of range: %d\n",
2503 display_idx);
2504 }
2505
2506 /* If a frame was copied out, return it */
2507 if (ctx->display_idx >= 0 &&
2508 ctx->display_idx < ctx->num_internal_frames) {
2509 struct coda_internal_frame *ready_frame;
2510
2511 ready_frame = &ctx->internal_frames[ctx->display_idx];
2512
2513 dst_buf = v4l2_m2m_dst_buf_remove(ctx->fh.m2m_ctx);
2514 dst_buf->sequence = ctx->osequence++;
2515
2516 dst_buf->field = V4L2_FIELD_NONE;
2517 dst_buf->flags &= ~(V4L2_BUF_FLAG_KEYFRAME |
2518 V4L2_BUF_FLAG_PFRAME |
2519 V4L2_BUF_FLAG_BFRAME);
2520 dst_buf->flags |= ready_frame->type;
2521 meta = &ready_frame->meta;
2522 if (meta->last && !coda_reorder_enable(ctx)) {
2523 /*
2524 * If this was the last decoded frame, and reordering
2525 * is disabled, this will be the last display frame.
2526 */
2527 coda_dbg(1, ctx, "last meta, marking as last frame\n");
2528 dst_buf->flags |= V4L2_BUF_FLAG_LAST;
2529 } else if (ctx->bit_stream_param & CODA_BIT_STREAM_END_FLAG &&
2530 display_idx == -1) {
2531 /*
2532 * If there is no designated presentation frame anymore,
2533 * this frame has to be the last one.
2534 */
2535 coda_dbg(1, ctx,
2536 "no more frames to return, marking as last frame\n");
2537 dst_buf->flags |= V4L2_BUF_FLAG_LAST;
2538 }
2539 dst_buf->timecode = meta->timecode;
2540 dst_buf->vb2_buf.timestamp = meta->timestamp;
2541
2542 trace_coda_dec_rot_done(ctx, dst_buf, meta);
2543
2544 vb2_set_plane_payload(&dst_buf->vb2_buf, 0,
2545 q_data_dst->sizeimage);
2546
2547 if (ready_frame->error || err_vdoa)
2548 coda_m2m_buf_done(ctx, dst_buf, VB2_BUF_STATE_ERROR);
2549 else
2550 coda_m2m_buf_done(ctx, dst_buf, VB2_BUF_STATE_DONE);
2551
2552 if (decoded_frame) {
2553 coda_dbg(1, ctx, "job finished: decoded %c frame %u, returned %c frame %u (%u/%u)%s\n",
2554 coda_frame_type_char(decoded_frame->type),
2555 decoded_frame->meta.sequence,
2556 coda_frame_type_char(dst_buf->flags),
2557 ready_frame->meta.sequence,
2558 dst_buf->sequence, ctx->qsequence,
2559 (dst_buf->flags & V4L2_BUF_FLAG_LAST) ?
2560 " (last)" : "");
2561 } else {
2562 coda_dbg(1, ctx, "job finished: no frame decoded (%d), returned %c frame %u (%u/%u)%s\n",
2563 decoded_idx,
2564 coda_frame_type_char(dst_buf->flags),
2565 ready_frame->meta.sequence,
2566 dst_buf->sequence, ctx->qsequence,
2567 (dst_buf->flags & V4L2_BUF_FLAG_LAST) ?
2568 " (last)" : "");
2569 }
2570 } else {
2571 if (decoded_frame) {
2572 coda_dbg(1, ctx, "job finished: decoded %c frame %u, no frame returned (%d)\n",
2573 coda_frame_type_char(decoded_frame->type),
2574 decoded_frame->meta.sequence,
2575 ctx->display_idx);
2576 } else {
2577 coda_dbg(1, ctx, "job finished: no frame decoded (%d) or returned (%d)\n",
2578 decoded_idx, ctx->display_idx);
2579 }
2580 }
2581
2582 /* The rotator will copy the current display frame next time */
2583 ctx->display_idx = display_idx;
2584
2585 /*
2586 * The current decode run might have brought the bitstream fill level
2587 * below the size where we can start the next decode run. As userspace
2588 * might have filled the output queue completely and might thus be
2589 * blocked, we can't rely on the next qbuf to trigger the bitstream
2590 * refill. Check if we have data to refill the bitstream now.
2591 */
2592 mutex_lock(&ctx->bitstream_mutex);
2593 coda_fill_bitstream(ctx, NULL);
2594 mutex_unlock(&ctx->bitstream_mutex);
2595 }
2596
coda_decode_timeout(struct coda_ctx * ctx)2597 static void coda_decode_timeout(struct coda_ctx *ctx)
2598 {
2599 struct vb2_v4l2_buffer *dst_buf;
2600
2601 /*
2602 * For now this only handles the case where we would deadlock with
2603 * userspace, i.e. userspace issued DEC_CMD_STOP and waits for EOS,
2604 * but after a failed decode run we would hold the context and wait for
2605 * userspace to queue more buffers.
2606 */
2607 if (!(ctx->bit_stream_param & CODA_BIT_STREAM_END_FLAG))
2608 return;
2609
2610 dst_buf = v4l2_m2m_dst_buf_remove(ctx->fh.m2m_ctx);
2611 dst_buf->sequence = ctx->qsequence - 1;
2612
2613 coda_m2m_buf_done(ctx, dst_buf, VB2_BUF_STATE_ERROR);
2614 }
2615
2616 const struct coda_context_ops coda_bit_decode_ops = {
2617 .queue_init = coda_decoder_queue_init,
2618 .reqbufs = coda_decoder_reqbufs,
2619 .start_streaming = coda_start_decoding,
2620 .prepare_run = coda_prepare_decode,
2621 .finish_run = coda_finish_decode,
2622 .run_timeout = coda_decode_timeout,
2623 .seq_init_work = coda_dec_seq_init_work,
2624 .seq_end_work = coda_seq_end_work,
2625 .release = coda_bit_release,
2626 };
2627
coda_irq_handler(int irq,void * data)2628 irqreturn_t coda_irq_handler(int irq, void *data)
2629 {
2630 struct coda_dev *dev = data;
2631 struct coda_ctx *ctx;
2632
2633 /* read status register to attend the IRQ */
2634 coda_read(dev, CODA_REG_BIT_INT_STATUS);
2635 coda_write(dev, 0, CODA_REG_BIT_INT_REASON);
2636 coda_write(dev, CODA_REG_BIT_INT_CLEAR_SET,
2637 CODA_REG_BIT_INT_CLEAR);
2638
2639 ctx = v4l2_m2m_get_curr_priv(dev->m2m_dev);
2640 if (ctx == NULL) {
2641 v4l2_err(&dev->v4l2_dev,
2642 "Instance released before the end of transaction\n");
2643 return IRQ_HANDLED;
2644 }
2645
2646 trace_coda_bit_done(ctx);
2647
2648 if (ctx->aborting) {
2649 coda_dbg(1, ctx, "task has been aborted\n");
2650 }
2651
2652 if (coda_isbusy(ctx->dev)) {
2653 coda_dbg(1, ctx, "coda is still busy!!!!\n");
2654 return IRQ_NONE;
2655 }
2656
2657 complete(&ctx->completion);
2658
2659 return IRQ_HANDLED;
2660 }
2661