linux/drivers/media/platform/coda/coda-bit.c
<<
>>
Prefs
   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
  41static void coda_free_bitstream_buffer(struct coda_ctx *ctx);
  42
  43static inline int coda_is_initialized(struct coda_dev *dev)
  44{
  45        return coda_read(dev, CODA_REG_BIT_CUR_PC) != 0;
  46}
  47
  48static inline unsigned long coda_isbusy(struct coda_dev *dev)
  49{
  50        return coda_read(dev, CODA_REG_BIT_BUSY);
  51}
  52
  53static 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
  64static 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
  97static 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
 111int 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
 149static 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
 162static 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
 174static 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
 184static 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
 203int 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
 223static 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
 230static 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
 251static 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
 323void 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        if (ctx->bit_stream_param & CODA_BIT_STREAM_END_FLAG)
 330                return;
 331
 332        while (v4l2_m2m_num_src_bufs_ready(ctx->fh.m2m_ctx) > 0) {
 333                /*
 334                 * Only queue two JPEGs into the bitstream buffer to keep
 335                 * latency low. We need at least one complete buffer and the
 336                 * header of another buffer (for prescan) in the bitstream.
 337                 */
 338                if (ctx->codec->src_fourcc == V4L2_PIX_FMT_JPEG &&
 339                    ctx->num_metas > 1)
 340                        break;
 341
 342                if (ctx->num_internal_frames &&
 343                    ctx->num_metas >= ctx->num_internal_frames) {
 344                        meta = list_first_entry(&ctx->buffer_meta_list,
 345                                                struct coda_buffer_meta, list);
 346
 347                        /*
 348                         * If we managed to fill in at least a full reorder
 349                         * window of buffers (num_internal_frames is a
 350                         * conservative estimate for this) and the bitstream
 351                         * prefetcher has at least 2 256 bytes periods beyond
 352                         * the first buffer to fetch, we can safely stop queuing
 353                         * in order to limit the decoder drain latency.
 354                         */
 355                        if (coda_bitstream_can_fetch_past(ctx, meta->end))
 356                                break;
 357                }
 358
 359                src_buf = v4l2_m2m_next_src_buf(ctx->fh.m2m_ctx);
 360
 361                /* Drop frames that do not start/end with a SOI/EOI markers */
 362                if (ctx->codec->src_fourcc == V4L2_PIX_FMT_JPEG &&
 363                    !coda_jpeg_check_buffer(ctx, &src_buf->vb2_buf)) {
 364                        v4l2_err(&ctx->dev->v4l2_dev,
 365                                 "dropping invalid JPEG frame %d\n",
 366                                 ctx->qsequence);
 367                        src_buf = v4l2_m2m_src_buf_remove(ctx->fh.m2m_ctx);
 368                        if (buffer_list) {
 369                                struct v4l2_m2m_buffer *m2m_buf;
 370
 371                                m2m_buf = container_of(src_buf,
 372                                                       struct v4l2_m2m_buffer,
 373                                                       vb);
 374                                list_add_tail(&m2m_buf->list, buffer_list);
 375                        } else {
 376                                v4l2_m2m_buf_done(src_buf, VB2_BUF_STATE_ERROR);
 377                        }
 378                        continue;
 379                }
 380
 381                /* Dump empty buffers */
 382                if (!vb2_get_plane_payload(&src_buf->vb2_buf, 0)) {
 383                        src_buf = v4l2_m2m_src_buf_remove(ctx->fh.m2m_ctx);
 384                        v4l2_m2m_buf_done(src_buf, VB2_BUF_STATE_DONE);
 385                        continue;
 386                }
 387
 388                /* Buffer start position */
 389                start = ctx->bitstream_fifo.kfifo.in;
 390
 391                if (coda_bitstream_try_queue(ctx, src_buf)) {
 392                        /*
 393                         * Source buffer is queued in the bitstream ringbuffer;
 394                         * queue the timestamp and mark source buffer as done
 395                         */
 396                        src_buf = v4l2_m2m_src_buf_remove(ctx->fh.m2m_ctx);
 397
 398                        meta = kmalloc(sizeof(*meta), GFP_KERNEL);
 399                        if (meta) {
 400                                meta->sequence = src_buf->sequence;
 401                                meta->timecode = src_buf->timecode;
 402                                meta->timestamp = src_buf->vb2_buf.timestamp;
 403                                meta->start = start;
 404                                meta->end = ctx->bitstream_fifo.kfifo.in;
 405                                meta->last = src_buf->flags & V4L2_BUF_FLAG_LAST;
 406                                if (meta->last)
 407                                        coda_dbg(1, ctx, "marking last meta");
 408                                spin_lock(&ctx->buffer_meta_lock);
 409                                list_add_tail(&meta->list,
 410                                              &ctx->buffer_meta_list);
 411                                ctx->num_metas++;
 412                                spin_unlock(&ctx->buffer_meta_lock);
 413
 414                                trace_coda_bit_queue(ctx, src_buf, meta);
 415                        }
 416
 417                        if (buffer_list) {
 418                                struct v4l2_m2m_buffer *m2m_buf;
 419
 420                                m2m_buf = container_of(src_buf,
 421                                                       struct v4l2_m2m_buffer,
 422                                                       vb);
 423                                list_add_tail(&m2m_buf->list, buffer_list);
 424                        } else {
 425                                v4l2_m2m_buf_done(src_buf, VB2_BUF_STATE_DONE);
 426                        }
 427                } else {
 428                        break;
 429                }
 430        }
 431}
 432
 433void coda_bit_stream_end_flag(struct coda_ctx *ctx)
 434{
 435        struct coda_dev *dev = ctx->dev;
 436
 437        ctx->bit_stream_param |= CODA_BIT_STREAM_END_FLAG;
 438
 439        /* If this context is currently running, update the hardware flag */
 440        if ((dev->devtype->product == CODA_960) &&
 441            coda_isbusy(dev) &&
 442            (ctx->idx == coda_read(dev, CODA_REG_BIT_RUN_INDEX))) {
 443                coda_write(dev, ctx->bit_stream_param,
 444                           CODA_REG_BIT_BIT_STREAM_PARAM);
 445        }
 446}
 447
 448static void coda_parabuf_write(struct coda_ctx *ctx, int index, u32 value)
 449{
 450        struct coda_dev *dev = ctx->dev;
 451        u32 *p = ctx->parabuf.vaddr;
 452
 453        if (dev->devtype->product == CODA_DX6)
 454                p[index] = value;
 455        else
 456                p[index ^ 1] = value;
 457}
 458
 459static inline int coda_alloc_context_buf(struct coda_ctx *ctx,
 460                                         struct coda_aux_buf *buf, size_t size,
 461                                         const char *name)
 462{
 463        return coda_alloc_aux_buf(ctx->dev, buf, size, name, ctx->debugfs_entry);
 464}
 465
 466
 467static void coda_free_framebuffers(struct coda_ctx *ctx)
 468{
 469        int i;
 470
 471        for (i = 0; i < CODA_MAX_FRAMEBUFFERS; i++)
 472                coda_free_aux_buf(ctx->dev, &ctx->internal_frames[i].buf);
 473}
 474
 475static int coda_alloc_framebuffers(struct coda_ctx *ctx,
 476                                   struct coda_q_data *q_data, u32 fourcc)
 477{
 478        struct coda_dev *dev = ctx->dev;
 479        unsigned int ysize, ycbcr_size;
 480        int ret;
 481        int i;
 482
 483        if (ctx->codec->src_fourcc == V4L2_PIX_FMT_H264 ||
 484            ctx->codec->dst_fourcc == V4L2_PIX_FMT_H264 ||
 485            ctx->codec->src_fourcc == V4L2_PIX_FMT_MPEG4 ||
 486            ctx->codec->dst_fourcc == V4L2_PIX_FMT_MPEG4)
 487                ysize = round_up(q_data->rect.width, 16) *
 488                        round_up(q_data->rect.height, 16);
 489        else
 490                ysize = round_up(q_data->rect.width, 8) * q_data->rect.height;
 491
 492        if (ctx->tiled_map_type == GDI_TILED_FRAME_MB_RASTER_MAP)
 493                ycbcr_size = round_up(ysize, 4096) + ysize / 2;
 494        else
 495                ycbcr_size = ysize + ysize / 2;
 496
 497        /* Allocate frame buffers */
 498        for (i = 0; i < ctx->num_internal_frames; i++) {
 499                size_t size = ycbcr_size;
 500                char *name;
 501
 502                /* Add space for mvcol buffers */
 503                if (dev->devtype->product != CODA_DX6 &&
 504                    (ctx->codec->src_fourcc == V4L2_PIX_FMT_H264 ||
 505                     (ctx->codec->src_fourcc == V4L2_PIX_FMT_MPEG4 && i == 0)))
 506                        size += ysize / 4;
 507                name = kasprintf(GFP_KERNEL, "fb%d", i);
 508                if (!name) {
 509                        coda_free_framebuffers(ctx);
 510                        return -ENOMEM;
 511                }
 512                ret = coda_alloc_context_buf(ctx, &ctx->internal_frames[i].buf,
 513                                             size, name);
 514                kfree(name);
 515                if (ret < 0) {
 516                        coda_free_framebuffers(ctx);
 517                        return ret;
 518                }
 519        }
 520
 521        /* Register frame buffers in the parameter buffer */
 522        for (i = 0; i < ctx->num_internal_frames; i++) {
 523                u32 y, cb, cr, mvcol;
 524
 525                /* Start addresses of Y, Cb, Cr planes */
 526                y = ctx->internal_frames[i].buf.paddr;
 527                cb = y + ysize;
 528                cr = y + ysize + ysize/4;
 529                mvcol = y + ysize + ysize/4 + ysize/4;
 530                if (ctx->tiled_map_type == GDI_TILED_FRAME_MB_RASTER_MAP) {
 531                        cb = round_up(cb, 4096);
 532                        mvcol = cb + ysize/2;
 533                        cr = 0;
 534                        /* Packed 20-bit MSB of base addresses */
 535                        /* YYYYYCCC, CCyyyyyc, cccc.... */
 536                        y = (y & 0xfffff000) | cb >> 20;
 537                        cb = (cb & 0x000ff000) << 12;
 538                }
 539                coda_parabuf_write(ctx, i * 3 + 0, y);
 540                coda_parabuf_write(ctx, i * 3 + 1, cb);
 541                coda_parabuf_write(ctx, i * 3 + 2, cr);
 542
 543                if (dev->devtype->product == CODA_DX6)
 544                        continue;
 545
 546                /* mvcol buffer for h.264 and mpeg4 */
 547                if (ctx->codec->src_fourcc == V4L2_PIX_FMT_H264)
 548                        coda_parabuf_write(ctx, 96 + i, mvcol);
 549                if (ctx->codec->src_fourcc == V4L2_PIX_FMT_MPEG4 && i == 0)
 550                        coda_parabuf_write(ctx, 97, mvcol);
 551        }
 552
 553        return 0;
 554}
 555
 556static void coda_free_context_buffers(struct coda_ctx *ctx)
 557{
 558        struct coda_dev *dev = ctx->dev;
 559
 560        coda_free_aux_buf(dev, &ctx->slicebuf);
 561        coda_free_aux_buf(dev, &ctx->psbuf);
 562        if (dev->devtype->product != CODA_DX6)
 563                coda_free_aux_buf(dev, &ctx->workbuf);
 564        coda_free_aux_buf(dev, &ctx->parabuf);
 565}
 566
 567static int coda_alloc_context_buffers(struct coda_ctx *ctx,
 568                                      struct coda_q_data *q_data)
 569{
 570        struct coda_dev *dev = ctx->dev;
 571        size_t size;
 572        int ret;
 573
 574        if (!ctx->parabuf.vaddr) {
 575                ret = coda_alloc_context_buf(ctx, &ctx->parabuf,
 576                                             CODA_PARA_BUF_SIZE, "parabuf");
 577                if (ret < 0)
 578                        return ret;
 579        }
 580
 581        if (dev->devtype->product == CODA_DX6)
 582                return 0;
 583
 584        if (!ctx->slicebuf.vaddr && q_data->fourcc == V4L2_PIX_FMT_H264) {
 585                /* worst case slice size */
 586                size = (DIV_ROUND_UP(q_data->rect.width, 16) *
 587                        DIV_ROUND_UP(q_data->rect.height, 16)) * 3200 / 8 + 512;
 588                ret = coda_alloc_context_buf(ctx, &ctx->slicebuf, size,
 589                                             "slicebuf");
 590                if (ret < 0)
 591                        goto err;
 592        }
 593
 594        if (!ctx->psbuf.vaddr && (dev->devtype->product == CODA_HX4 ||
 595                                  dev->devtype->product == CODA_7541)) {
 596                ret = coda_alloc_context_buf(ctx, &ctx->psbuf,
 597                                             CODA7_PS_BUF_SIZE, "psbuf");
 598                if (ret < 0)
 599                        goto err;
 600        }
 601
 602        if (!ctx->workbuf.vaddr) {
 603                size = dev->devtype->workbuf_size;
 604                if (dev->devtype->product == CODA_960 &&
 605                    q_data->fourcc == V4L2_PIX_FMT_H264)
 606                        size += CODA9_PS_SAVE_SIZE;
 607                ret = coda_alloc_context_buf(ctx, &ctx->workbuf, size,
 608                                             "workbuf");
 609                if (ret < 0)
 610                        goto err;
 611        }
 612
 613        return 0;
 614
 615err:
 616        coda_free_context_buffers(ctx);
 617        return ret;
 618}
 619
 620static int coda_encode_header(struct coda_ctx *ctx, struct vb2_v4l2_buffer *buf,
 621                              int header_code, u8 *header, int *size)
 622{
 623        struct vb2_buffer *vb = &buf->vb2_buf;
 624        struct coda_dev *dev = ctx->dev;
 625        struct coda_q_data *q_data_src;
 626        struct v4l2_rect *r;
 627        size_t bufsize;
 628        int ret;
 629        int i;
 630
 631        if (dev->devtype->product == CODA_960)
 632                memset(vb2_plane_vaddr(vb, 0), 0, 64);
 633
 634        coda_write(dev, vb2_dma_contig_plane_dma_addr(vb, 0),
 635                   CODA_CMD_ENC_HEADER_BB_START);
 636        bufsize = vb2_plane_size(vb, 0);
 637        if (dev->devtype->product == CODA_960)
 638                bufsize /= 1024;
 639        coda_write(dev, bufsize, CODA_CMD_ENC_HEADER_BB_SIZE);
 640        if (dev->devtype->product == CODA_960 &&
 641            ctx->codec->dst_fourcc == V4L2_PIX_FMT_H264 &&
 642            header_code == CODA_HEADER_H264_SPS) {
 643                q_data_src = get_q_data(ctx, V4L2_BUF_TYPE_VIDEO_OUTPUT);
 644                r = &q_data_src->rect;
 645
 646                if (r->width % 16 || r->height % 16) {
 647                        u32 crop_right = round_up(r->width, 16) -  r->width;
 648                        u32 crop_bottom = round_up(r->height, 16) - r->height;
 649
 650                        coda_write(dev, crop_right,
 651                                   CODA9_CMD_ENC_HEADER_FRAME_CROP_H);
 652                        coda_write(dev, crop_bottom,
 653                                   CODA9_CMD_ENC_HEADER_FRAME_CROP_V);
 654                        header_code |= CODA9_HEADER_FRAME_CROP;
 655                }
 656        }
 657        coda_write(dev, header_code, CODA_CMD_ENC_HEADER_CODE);
 658        ret = coda_command_sync(ctx, CODA_COMMAND_ENCODE_HEADER);
 659        if (ret < 0) {
 660                v4l2_err(&dev->v4l2_dev, "CODA_COMMAND_ENCODE_HEADER timeout\n");
 661                return ret;
 662        }
 663
 664        if (dev->devtype->product == CODA_960) {
 665                for (i = 63; i > 0; i--)
 666                        if (((char *)vb2_plane_vaddr(vb, 0))[i] != 0)
 667                                break;
 668                *size = i + 1;
 669        } else {
 670                *size = coda_read(dev, CODA_REG_BIT_WR_PTR(ctx->reg_idx)) -
 671                        coda_read(dev, CODA_CMD_ENC_HEADER_BB_START);
 672        }
 673        memcpy(header, vb2_plane_vaddr(vb, 0), *size);
 674
 675        return 0;
 676}
 677
 678static u32 coda_slice_mode(struct coda_ctx *ctx)
 679{
 680        int size, unit;
 681
 682        switch (ctx->params.slice_mode) {
 683        case V4L2_MPEG_VIDEO_MULTI_SLICE_MODE_SINGLE:
 684        default:
 685                return 0;
 686        case V4L2_MPEG_VIDEO_MULTI_SLICE_MODE_MAX_MB:
 687                size = ctx->params.slice_max_mb;
 688                unit = 1;
 689                break;
 690        case V4L2_MPEG_VIDEO_MULTI_SLICE_MODE_MAX_BYTES:
 691                size = ctx->params.slice_max_bits;
 692                unit = 0;
 693                break;
 694        }
 695
 696        return ((size & CODA_SLICING_SIZE_MASK) << CODA_SLICING_SIZE_OFFSET) |
 697               ((unit & CODA_SLICING_UNIT_MASK) << CODA_SLICING_UNIT_OFFSET) |
 698               ((1 & CODA_SLICING_MODE_MASK) << CODA_SLICING_MODE_OFFSET);
 699}
 700
 701static int coda_enc_param_change(struct coda_ctx *ctx)
 702{
 703        struct coda_dev *dev = ctx->dev;
 704        u32 change_enable = 0;
 705        u32 success;
 706        int ret;
 707
 708        if (ctx->params.gop_size_changed) {
 709                change_enable |= CODA_PARAM_CHANGE_RC_GOP;
 710                coda_write(dev, ctx->params.gop_size,
 711                           CODA_CMD_ENC_PARAM_RC_GOP);
 712                ctx->gopcounter = ctx->params.gop_size - 1;
 713                ctx->params.gop_size_changed = false;
 714        }
 715        if (ctx->params.h264_intra_qp_changed) {
 716                coda_dbg(1, ctx, "parameter change: intra Qp %u\n",
 717                         ctx->params.h264_intra_qp);
 718
 719                if (ctx->params.bitrate) {
 720                        change_enable |= CODA_PARAM_CHANGE_RC_INTRA_QP;
 721                        coda_write(dev, ctx->params.h264_intra_qp,
 722                                   CODA_CMD_ENC_PARAM_RC_INTRA_QP);
 723                }
 724                ctx->params.h264_intra_qp_changed = false;
 725        }
 726        if (ctx->params.bitrate_changed) {
 727                coda_dbg(1, ctx, "parameter change: bitrate %u kbit/s\n",
 728                         ctx->params.bitrate);
 729                change_enable |= CODA_PARAM_CHANGE_RC_BITRATE;
 730                coda_write(dev, ctx->params.bitrate,
 731                           CODA_CMD_ENC_PARAM_RC_BITRATE);
 732                ctx->params.bitrate_changed = false;
 733        }
 734        if (ctx->params.framerate_changed) {
 735                coda_dbg(1, ctx, "parameter change: frame rate %u/%u Hz\n",
 736                         ctx->params.framerate & 0xffff,
 737                         (ctx->params.framerate >> 16) + 1);
 738                change_enable |= CODA_PARAM_CHANGE_RC_FRAME_RATE;
 739                coda_write(dev, ctx->params.framerate,
 740                           CODA_CMD_ENC_PARAM_RC_FRAME_RATE);
 741                ctx->params.framerate_changed = false;
 742        }
 743        if (ctx->params.intra_refresh_changed) {
 744                coda_dbg(1, ctx, "parameter change: intra refresh MBs %u\n",
 745                         ctx->params.intra_refresh);
 746                change_enable |= CODA_PARAM_CHANGE_INTRA_MB_NUM;
 747                coda_write(dev, ctx->params.intra_refresh,
 748                           CODA_CMD_ENC_PARAM_INTRA_MB_NUM);
 749                ctx->params.intra_refresh_changed = false;
 750        }
 751        if (ctx->params.slice_mode_changed) {
 752                change_enable |= CODA_PARAM_CHANGE_SLICE_MODE;
 753                coda_write(dev, coda_slice_mode(ctx),
 754                           CODA_CMD_ENC_PARAM_SLICE_MODE);
 755                ctx->params.slice_mode_changed = false;
 756        }
 757
 758        if (!change_enable)
 759                return 0;
 760
 761        coda_write(dev, change_enable, CODA_CMD_ENC_PARAM_CHANGE_ENABLE);
 762
 763        ret = coda_command_sync(ctx, CODA_COMMAND_RC_CHANGE_PARAMETER);
 764        if (ret < 0)
 765                return ret;
 766
 767        success = coda_read(dev, CODA_RET_ENC_PARAM_CHANGE_SUCCESS);
 768        if (success != 1)
 769                coda_dbg(1, ctx, "parameter change failed: %u\n", success);
 770
 771        return 0;
 772}
 773
 774static phys_addr_t coda_iram_alloc(struct coda_iram_info *iram, size_t size)
 775{
 776        phys_addr_t ret;
 777
 778        size = round_up(size, 1024);
 779        if (size > iram->remaining)
 780                return 0;
 781        iram->remaining -= size;
 782
 783        ret = iram->next_paddr;
 784        iram->next_paddr += size;
 785
 786        return ret;
 787}
 788
 789static void coda_setup_iram(struct coda_ctx *ctx)
 790{
 791        struct coda_iram_info *iram_info = &ctx->iram_info;
 792        struct coda_dev *dev = ctx->dev;
 793        int w64, w128;
 794        int mb_width;
 795        int dbk_bits;
 796        int bit_bits;
 797        int ip_bits;
 798        int me_bits;
 799
 800        memset(iram_info, 0, sizeof(*iram_info));
 801        iram_info->next_paddr = dev->iram.paddr;
 802        iram_info->remaining = dev->iram.size;
 803
 804        if (!dev->iram.vaddr)
 805                return;
 806
 807        switch (dev->devtype->product) {
 808        case CODA_HX4:
 809                dbk_bits = CODA7_USE_HOST_DBK_ENABLE;
 810                bit_bits = CODA7_USE_HOST_BIT_ENABLE;
 811                ip_bits = CODA7_USE_HOST_IP_ENABLE;
 812                me_bits = CODA7_USE_HOST_ME_ENABLE;
 813                break;
 814        case CODA_7541:
 815                dbk_bits = CODA7_USE_HOST_DBK_ENABLE | CODA7_USE_DBK_ENABLE;
 816                bit_bits = CODA7_USE_HOST_BIT_ENABLE | CODA7_USE_BIT_ENABLE;
 817                ip_bits = CODA7_USE_HOST_IP_ENABLE | CODA7_USE_IP_ENABLE;
 818                me_bits = CODA7_USE_HOST_ME_ENABLE | CODA7_USE_ME_ENABLE;
 819                break;
 820        case CODA_960:
 821                dbk_bits = CODA9_USE_HOST_DBK_ENABLE | CODA9_USE_DBK_ENABLE;
 822                bit_bits = CODA9_USE_HOST_BIT_ENABLE | CODA7_USE_BIT_ENABLE;
 823                ip_bits = CODA9_USE_HOST_IP_ENABLE | CODA7_USE_IP_ENABLE;
 824                me_bits = 0;
 825                break;
 826        default: /* CODA_DX6 */
 827                return;
 828        }
 829
 830        if (ctx->inst_type == CODA_INST_ENCODER) {
 831                struct coda_q_data *q_data_src;
 832
 833                q_data_src = get_q_data(ctx, V4L2_BUF_TYPE_VIDEO_OUTPUT);
 834                mb_width = DIV_ROUND_UP(q_data_src->rect.width, 16);
 835                w128 = mb_width * 128;
 836                w64 = mb_width * 64;
 837
 838                /* Prioritize in case IRAM is too small for everything */
 839                if (dev->devtype->product == CODA_HX4 ||
 840                    dev->devtype->product == CODA_7541) {
 841                        iram_info->search_ram_size = round_up(mb_width * 16 *
 842                                                              36 + 2048, 1024);
 843                        iram_info->search_ram_paddr = coda_iram_alloc(iram_info,
 844                                                iram_info->search_ram_size);
 845                        if (!iram_info->search_ram_paddr) {
 846                                pr_err("IRAM is smaller than the search ram size\n");
 847                                goto out;
 848                        }
 849                        iram_info->axi_sram_use |= me_bits;
 850                }
 851
 852                /* Only H.264BP and H.263P3 are considered */
 853                iram_info->buf_dbk_y_use = coda_iram_alloc(iram_info, w64);
 854                iram_info->buf_dbk_c_use = coda_iram_alloc(iram_info, w64);
 855                if (!iram_info->buf_dbk_c_use)
 856                        goto out;
 857                iram_info->axi_sram_use |= dbk_bits;
 858
 859                iram_info->buf_bit_use = coda_iram_alloc(iram_info, w128);
 860                if (!iram_info->buf_bit_use)
 861                        goto out;
 862                iram_info->axi_sram_use |= bit_bits;
 863
 864                iram_info->buf_ip_ac_dc_use = coda_iram_alloc(iram_info, w128);
 865                if (!iram_info->buf_ip_ac_dc_use)
 866                        goto out;
 867                iram_info->axi_sram_use |= ip_bits;
 868
 869                /* OVL and BTP disabled for encoder */
 870        } else if (ctx->inst_type == CODA_INST_DECODER) {
 871                struct coda_q_data *q_data_dst;
 872
 873                q_data_dst = get_q_data(ctx, V4L2_BUF_TYPE_VIDEO_CAPTURE);
 874                mb_width = DIV_ROUND_UP(q_data_dst->width, 16);
 875                w128 = mb_width * 128;
 876
 877                iram_info->buf_dbk_y_use = coda_iram_alloc(iram_info, w128);
 878                iram_info->buf_dbk_c_use = coda_iram_alloc(iram_info, w128);
 879                if (!iram_info->buf_dbk_c_use)
 880                        goto out;
 881                iram_info->axi_sram_use |= dbk_bits;
 882
 883                iram_info->buf_bit_use = coda_iram_alloc(iram_info, w128);
 884                if (!iram_info->buf_bit_use)
 885                        goto out;
 886                iram_info->axi_sram_use |= bit_bits;
 887
 888                iram_info->buf_ip_ac_dc_use = coda_iram_alloc(iram_info, w128);
 889                if (!iram_info->buf_ip_ac_dc_use)
 890                        goto out;
 891                iram_info->axi_sram_use |= ip_bits;
 892
 893                /* OVL and BTP unused as there is no VC1 support yet */
 894        }
 895
 896out:
 897        if (!(iram_info->axi_sram_use & CODA7_USE_HOST_IP_ENABLE))
 898                coda_dbg(1, ctx, "IRAM smaller than needed\n");
 899
 900        if (dev->devtype->product == CODA_HX4 ||
 901            dev->devtype->product == CODA_7541) {
 902                /* TODO - Enabling these causes picture errors on CODA7541 */
 903                if (ctx->inst_type == CODA_INST_DECODER) {
 904                        /* fw 1.4.50 */
 905                        iram_info->axi_sram_use &= ~(CODA7_USE_HOST_IP_ENABLE |
 906                                                     CODA7_USE_IP_ENABLE);
 907                } else {
 908                        /* fw 13.4.29 */
 909                        iram_info->axi_sram_use &= ~(CODA7_USE_HOST_IP_ENABLE |
 910                                                     CODA7_USE_HOST_DBK_ENABLE |
 911                                                     CODA7_USE_IP_ENABLE |
 912                                                     CODA7_USE_DBK_ENABLE);
 913                }
 914        }
 915}
 916
 917static u32 coda_supported_firmwares[] = {
 918        CODA_FIRMWARE_VERNUM(CODA_DX6, 2, 2, 5),
 919        CODA_FIRMWARE_VERNUM(CODA_HX4, 1, 4, 50),
 920        CODA_FIRMWARE_VERNUM(CODA_7541, 1, 4, 50),
 921        CODA_FIRMWARE_VERNUM(CODA_960, 2, 1, 5),
 922        CODA_FIRMWARE_VERNUM(CODA_960, 2, 1, 9),
 923        CODA_FIRMWARE_VERNUM(CODA_960, 2, 3, 10),
 924        CODA_FIRMWARE_VERNUM(CODA_960, 3, 1, 1),
 925};
 926
 927static bool coda_firmware_supported(u32 vernum)
 928{
 929        int i;
 930
 931        for (i = 0; i < ARRAY_SIZE(coda_supported_firmwares); i++)
 932                if (vernum == coda_supported_firmwares[i])
 933                        return true;
 934        return false;
 935}
 936
 937int coda_check_firmware(struct coda_dev *dev)
 938{
 939        u16 product, major, minor, release;
 940        u32 data;
 941        int ret;
 942
 943        ret = clk_prepare_enable(dev->clk_per);
 944        if (ret)
 945                goto err_clk_per;
 946
 947        ret = clk_prepare_enable(dev->clk_ahb);
 948        if (ret)
 949                goto err_clk_ahb;
 950
 951        coda_write(dev, 0, CODA_CMD_FIRMWARE_VERNUM);
 952        coda_write(dev, CODA_REG_BIT_BUSY_FLAG, CODA_REG_BIT_BUSY);
 953        coda_write(dev, 0, CODA_REG_BIT_RUN_INDEX);
 954        coda_write(dev, 0, CODA_REG_BIT_RUN_COD_STD);
 955        coda_write(dev, CODA_COMMAND_FIRMWARE_GET, CODA_REG_BIT_RUN_COMMAND);
 956        if (coda_wait_timeout(dev)) {
 957                v4l2_err(&dev->v4l2_dev, "firmware get command error\n");
 958                ret = -EIO;
 959                goto err_run_cmd;
 960        }
 961
 962        if (dev->devtype->product == CODA_960) {
 963                data = coda_read(dev, CODA9_CMD_FIRMWARE_CODE_REV);
 964                v4l2_info(&dev->v4l2_dev, "Firmware code revision: %d\n",
 965                          data);
 966        }
 967
 968        /* Check we are compatible with the loaded firmware */
 969        data = coda_read(dev, CODA_CMD_FIRMWARE_VERNUM);
 970        product = CODA_FIRMWARE_PRODUCT(data);
 971        major = CODA_FIRMWARE_MAJOR(data);
 972        minor = CODA_FIRMWARE_MINOR(data);
 973        release = CODA_FIRMWARE_RELEASE(data);
 974
 975        clk_disable_unprepare(dev->clk_per);
 976        clk_disable_unprepare(dev->clk_ahb);
 977
 978        if (product != dev->devtype->product) {
 979                v4l2_err(&dev->v4l2_dev,
 980                         "Wrong firmware. Hw: %s, Fw: %s, Version: %u.%u.%u\n",
 981                         coda_product_name(dev->devtype->product),
 982                         coda_product_name(product), major, minor, release);
 983                return -EINVAL;
 984        }
 985
 986        v4l2_info(&dev->v4l2_dev, "Initialized %s.\n",
 987                  coda_product_name(product));
 988
 989        if (coda_firmware_supported(data)) {
 990                v4l2_info(&dev->v4l2_dev, "Firmware version: %u.%u.%u\n",
 991                          major, minor, release);
 992        } else {
 993                v4l2_warn(&dev->v4l2_dev,
 994                          "Unsupported firmware version: %u.%u.%u\n",
 995                          major, minor, release);
 996        }
 997
 998        return 0;
 999
1000err_run_cmd:
1001        clk_disable_unprepare(dev->clk_ahb);
1002err_clk_ahb:
1003        clk_disable_unprepare(dev->clk_per);
1004err_clk_per:
1005        return ret;
1006}
1007
1008static void coda9_set_frame_cache(struct coda_ctx *ctx, u32 fourcc)
1009{
1010        u32 cache_size, cache_config;
1011
1012        if (ctx->tiled_map_type == GDI_LINEAR_FRAME_MAP) {
1013                /* Luma 2x0 page, 2x6 cache, chroma 2x0 page, 2x4 cache size */
1014                cache_size = 0x20262024;
1015                cache_config = 2 << CODA9_CACHE_PAGEMERGE_OFFSET;
1016        } else {
1017                /* Luma 0x2 page, 4x4 cache, chroma 0x2 page, 4x3 cache size */
1018                cache_size = 0x02440243;
1019                cache_config = 1 << CODA9_CACHE_PAGEMERGE_OFFSET;
1020        }
1021        coda_write(ctx->dev, cache_size, CODA9_CMD_SET_FRAME_CACHE_SIZE);
1022        if (fourcc == V4L2_PIX_FMT_NV12 || fourcc == V4L2_PIX_FMT_YUYV) {
1023                cache_config |= 32 << CODA9_CACHE_LUMA_BUFFER_SIZE_OFFSET |
1024                                16 << CODA9_CACHE_CR_BUFFER_SIZE_OFFSET |
1025                                0 << CODA9_CACHE_CB_BUFFER_SIZE_OFFSET;
1026        } else {
1027                cache_config |= 32 << CODA9_CACHE_LUMA_BUFFER_SIZE_OFFSET |
1028                                8 << CODA9_CACHE_CR_BUFFER_SIZE_OFFSET |
1029                                8 << CODA9_CACHE_CB_BUFFER_SIZE_OFFSET;
1030        }
1031        coda_write(ctx->dev, cache_config, CODA9_CMD_SET_FRAME_CACHE_CONFIG);
1032}
1033
1034/*
1035 * Encoder context operations
1036 */
1037
1038static int coda_encoder_reqbufs(struct coda_ctx *ctx,
1039                                struct v4l2_requestbuffers *rb)
1040{
1041        struct coda_q_data *q_data_src;
1042        int ret;
1043
1044        if (rb->type != V4L2_BUF_TYPE_VIDEO_OUTPUT)
1045                return 0;
1046
1047        if (rb->count) {
1048                q_data_src = get_q_data(ctx, V4L2_BUF_TYPE_VIDEO_OUTPUT);
1049                ret = coda_alloc_context_buffers(ctx, q_data_src);
1050                if (ret < 0)
1051                        return ret;
1052        } else {
1053                coda_free_context_buffers(ctx);
1054        }
1055
1056        return 0;
1057}
1058
1059static int coda_start_encoding(struct coda_ctx *ctx)
1060{
1061        struct coda_dev *dev = ctx->dev;
1062        struct v4l2_device *v4l2_dev = &dev->v4l2_dev;
1063        struct coda_q_data *q_data_src, *q_data_dst;
1064        u32 bitstream_buf, bitstream_size;
1065        struct vb2_v4l2_buffer *buf;
1066        int gamma, ret, value;
1067        u32 dst_fourcc;
1068        int num_fb;
1069        u32 stride;
1070
1071        q_data_src = get_q_data(ctx, V4L2_BUF_TYPE_VIDEO_OUTPUT);
1072        q_data_dst = get_q_data(ctx, V4L2_BUF_TYPE_VIDEO_CAPTURE);
1073        dst_fourcc = q_data_dst->fourcc;
1074
1075        buf = v4l2_m2m_next_dst_buf(ctx->fh.m2m_ctx);
1076        bitstream_buf = vb2_dma_contig_plane_dma_addr(&buf->vb2_buf, 0);
1077        bitstream_size = q_data_dst->sizeimage;
1078
1079        if (!coda_is_initialized(dev)) {
1080                v4l2_err(v4l2_dev, "coda is not initialized.\n");
1081                return -EFAULT;
1082        }
1083
1084        if (dst_fourcc == V4L2_PIX_FMT_JPEG) {
1085                if (!ctx->params.jpeg_qmat_tab[0])
1086                        ctx->params.jpeg_qmat_tab[0] = kmalloc(64, GFP_KERNEL);
1087                if (!ctx->params.jpeg_qmat_tab[1])
1088                        ctx->params.jpeg_qmat_tab[1] = kmalloc(64, GFP_KERNEL);
1089                coda_set_jpeg_compression_quality(ctx, ctx->params.jpeg_quality);
1090        }
1091
1092        mutex_lock(&dev->coda_mutex);
1093
1094        coda_write(dev, ctx->parabuf.paddr, CODA_REG_BIT_PARA_BUF_ADDR);
1095        coda_write(dev, bitstream_buf, CODA_REG_BIT_RD_PTR(ctx->reg_idx));
1096        coda_write(dev, bitstream_buf, CODA_REG_BIT_WR_PTR(ctx->reg_idx));
1097        switch (dev->devtype->product) {
1098        case CODA_DX6:
1099                coda_write(dev, CODADX6_STREAM_BUF_DYNALLOC_EN |
1100                        CODADX6_STREAM_BUF_PIC_RESET, CODA_REG_BIT_STREAM_CTRL);
1101                break;
1102        case CODA_960:
1103                coda_write(dev, 0, CODA9_GDI_WPROT_RGN_EN);
1104                fallthrough;
1105        case CODA_HX4:
1106        case CODA_7541:
1107                coda_write(dev, CODA7_STREAM_BUF_DYNALLOC_EN |
1108                        CODA7_STREAM_BUF_PIC_RESET, CODA_REG_BIT_STREAM_CTRL);
1109                break;
1110        }
1111
1112        ctx->frame_mem_ctrl &= ~(CODA_FRAME_CHROMA_INTERLEAVE | (0x3 << 9) |
1113                                 CODA9_FRAME_TILED2LINEAR);
1114        if (q_data_src->fourcc == V4L2_PIX_FMT_NV12)
1115                ctx->frame_mem_ctrl |= CODA_FRAME_CHROMA_INTERLEAVE;
1116        if (ctx->tiled_map_type == GDI_TILED_FRAME_MB_RASTER_MAP)
1117                ctx->frame_mem_ctrl |= (0x3 << 9) | CODA9_FRAME_TILED2LINEAR;
1118        coda_write(dev, ctx->frame_mem_ctrl, CODA_REG_BIT_FRAME_MEM_CTRL);
1119
1120        if (dev->devtype->product == CODA_DX6) {
1121                /* Configure the coda */
1122                coda_write(dev, dev->iram.paddr,
1123                           CODADX6_REG_BIT_SEARCH_RAM_BASE_ADDR);
1124        }
1125
1126        /* Could set rotation here if needed */
1127        value = 0;
1128        switch (dev->devtype->product) {
1129        case CODA_DX6:
1130                value = (q_data_src->rect.width & CODADX6_PICWIDTH_MASK)
1131                        << CODADX6_PICWIDTH_OFFSET;
1132                value |= (q_data_src->rect.height & CODADX6_PICHEIGHT_MASK)
1133                         << CODA_PICHEIGHT_OFFSET;
1134                break;
1135        case CODA_HX4:
1136        case CODA_7541:
1137                if (dst_fourcc == V4L2_PIX_FMT_H264) {
1138                        value = (round_up(q_data_src->rect.width, 16) &
1139                                 CODA7_PICWIDTH_MASK) << CODA7_PICWIDTH_OFFSET;
1140                        value |= (round_up(q_data_src->rect.height, 16) &
1141                                 CODA7_PICHEIGHT_MASK) << CODA_PICHEIGHT_OFFSET;
1142                        break;
1143                }
1144                fallthrough;
1145        case CODA_960:
1146                value = (q_data_src->rect.width & CODA7_PICWIDTH_MASK)
1147                        << CODA7_PICWIDTH_OFFSET;
1148                value |= (q_data_src->rect.height & CODA7_PICHEIGHT_MASK)
1149                         << CODA_PICHEIGHT_OFFSET;
1150        }
1151        coda_write(dev, value, CODA_CMD_ENC_SEQ_SRC_SIZE);
1152        if (dst_fourcc == V4L2_PIX_FMT_JPEG)
1153                ctx->params.framerate = 0;
1154        coda_write(dev, ctx->params.framerate,
1155                   CODA_CMD_ENC_SEQ_SRC_F_RATE);
1156
1157        ctx->params.codec_mode = ctx->codec->mode;
1158        switch (dst_fourcc) {
1159        case V4L2_PIX_FMT_MPEG4:
1160                if (dev->devtype->product == CODA_960)
1161                        coda_write(dev, CODA9_STD_MPEG4,
1162                                   CODA_CMD_ENC_SEQ_COD_STD);
1163                else
1164                        coda_write(dev, CODA_STD_MPEG4,
1165                                   CODA_CMD_ENC_SEQ_COD_STD);
1166                coda_write(dev, 0, CODA_CMD_ENC_SEQ_MP4_PARA);
1167                break;
1168        case V4L2_PIX_FMT_H264:
1169                if (dev->devtype->product == CODA_960)
1170                        coda_write(dev, CODA9_STD_H264,
1171                                   CODA_CMD_ENC_SEQ_COD_STD);
1172                else
1173                        coda_write(dev, CODA_STD_H264,
1174                                   CODA_CMD_ENC_SEQ_COD_STD);
1175                value = ((ctx->params.h264_disable_deblocking_filter_idc &
1176                          CODA_264PARAM_DISABLEDEBLK_MASK) <<
1177                         CODA_264PARAM_DISABLEDEBLK_OFFSET) |
1178                        ((ctx->params.h264_slice_alpha_c0_offset_div2 &
1179                          CODA_264PARAM_DEBLKFILTEROFFSETALPHA_MASK) <<
1180                         CODA_264PARAM_DEBLKFILTEROFFSETALPHA_OFFSET) |
1181                        ((ctx->params.h264_slice_beta_offset_div2 &
1182                          CODA_264PARAM_DEBLKFILTEROFFSETBETA_MASK) <<
1183                         CODA_264PARAM_DEBLKFILTEROFFSETBETA_OFFSET) |
1184                        (ctx->params.h264_constrained_intra_pred_flag <<
1185                         CODA_264PARAM_CONSTRAINEDINTRAPREDFLAG_OFFSET) |
1186                        (ctx->params.h264_chroma_qp_index_offset &
1187                         CODA_264PARAM_CHROMAQPOFFSET_MASK);
1188                coda_write(dev, value, CODA_CMD_ENC_SEQ_264_PARA);
1189                break;
1190        case V4L2_PIX_FMT_JPEG:
1191                coda_write(dev, 0, CODA_CMD_ENC_SEQ_JPG_PARA);
1192                coda_write(dev, ctx->params.jpeg_restart_interval,
1193                                CODA_CMD_ENC_SEQ_JPG_RST_INTERVAL);
1194                coda_write(dev, 0, CODA_CMD_ENC_SEQ_JPG_THUMB_EN);
1195                coda_write(dev, 0, CODA_CMD_ENC_SEQ_JPG_THUMB_SIZE);
1196                coda_write(dev, 0, CODA_CMD_ENC_SEQ_JPG_THUMB_OFFSET);
1197
1198                coda_jpeg_write_tables(ctx);
1199                break;
1200        default:
1201                v4l2_err(v4l2_dev,
1202                         "dst format (0x%08x) invalid.\n", dst_fourcc);
1203                ret = -EINVAL;
1204                goto out;
1205        }
1206
1207        /*
1208         * slice mode and GOP size registers are used for thumb size/offset
1209         * in JPEG mode
1210         */
1211        if (dst_fourcc != V4L2_PIX_FMT_JPEG) {
1212                value = coda_slice_mode(ctx);
1213                coda_write(dev, value, CODA_CMD_ENC_SEQ_SLICE_MODE);
1214                value = ctx->params.gop_size;
1215                coda_write(dev, value, CODA_CMD_ENC_SEQ_GOP_SIZE);
1216        }
1217
1218        if (ctx->params.bitrate && (ctx->params.frame_rc_enable ||
1219                                    ctx->params.mb_rc_enable)) {
1220                ctx->params.bitrate_changed = false;
1221                ctx->params.h264_intra_qp_changed = false;
1222
1223                /* Rate control enabled */
1224                value = (ctx->params.bitrate & CODA_RATECONTROL_BITRATE_MASK)
1225                        << CODA_RATECONTROL_BITRATE_OFFSET;
1226                value |=  1 & CODA_RATECONTROL_ENABLE_MASK;
1227                value |= (ctx->params.vbv_delay &
1228                          CODA_RATECONTROL_INITIALDELAY_MASK)
1229                         << CODA_RATECONTROL_INITIALDELAY_OFFSET;
1230                if (dev->devtype->product == CODA_960)
1231                        value |= BIT(31); /* disable autoskip */
1232        } else {
1233                value = 0;
1234        }
1235        coda_write(dev, value, CODA_CMD_ENC_SEQ_RC_PARA);
1236
1237        coda_write(dev, ctx->params.vbv_size, CODA_CMD_ENC_SEQ_RC_BUF_SIZE);
1238        coda_write(dev, ctx->params.intra_refresh,
1239                   CODA_CMD_ENC_SEQ_INTRA_REFRESH);
1240
1241        coda_write(dev, bitstream_buf, CODA_CMD_ENC_SEQ_BB_START);
1242        coda_write(dev, bitstream_size / 1024, CODA_CMD_ENC_SEQ_BB_SIZE);
1243
1244
1245        value = 0;
1246        if (dev->devtype->product == CODA_960)
1247                gamma = CODA9_DEFAULT_GAMMA;
1248        else
1249                gamma = CODA_DEFAULT_GAMMA;
1250        if (gamma > 0) {
1251                coda_write(dev, (gamma & CODA_GAMMA_MASK) << CODA_GAMMA_OFFSET,
1252                           CODA_CMD_ENC_SEQ_RC_GAMMA);
1253        }
1254
1255        if (ctx->params.h264_min_qp || ctx->params.h264_max_qp) {
1256                coda_write(dev,
1257                           ctx->params.h264_min_qp << CODA_QPMIN_OFFSET |
1258                           ctx->params.h264_max_qp << CODA_QPMAX_OFFSET,
1259                           CODA_CMD_ENC_SEQ_RC_QP_MIN_MAX);
1260        }
1261        if (dev->devtype->product == CODA_960) {
1262                if (ctx->params.h264_max_qp)
1263                        value |= 1 << CODA9_OPTION_RCQPMAX_OFFSET;
1264                if (CODA_DEFAULT_GAMMA > 0)
1265                        value |= 1 << CODA9_OPTION_GAMMA_OFFSET;
1266        } else {
1267                if (CODA_DEFAULT_GAMMA > 0) {
1268                        if (dev->devtype->product == CODA_DX6)
1269                                value |= 1 << CODADX6_OPTION_GAMMA_OFFSET;
1270                        else
1271                                value |= 1 << CODA7_OPTION_GAMMA_OFFSET;
1272                }
1273                if (ctx->params.h264_min_qp)
1274                        value |= 1 << CODA7_OPTION_RCQPMIN_OFFSET;
1275                if (ctx->params.h264_max_qp)
1276                        value |= 1 << CODA7_OPTION_RCQPMAX_OFFSET;
1277        }
1278        coda_write(dev, value, CODA_CMD_ENC_SEQ_OPTION);
1279
1280        if (ctx->params.frame_rc_enable && !ctx->params.mb_rc_enable)
1281                value = 1;
1282        else
1283                value = 0;
1284        coda_write(dev, value, CODA_CMD_ENC_SEQ_RC_INTERVAL_MODE);
1285
1286        coda_setup_iram(ctx);
1287
1288        if (dst_fourcc == V4L2_PIX_FMT_H264) {
1289                switch (dev->devtype->product) {
1290                case CODA_DX6:
1291                        value = FMO_SLICE_SAVE_BUF_SIZE << 7;
1292                        coda_write(dev, value, CODADX6_CMD_ENC_SEQ_FMO);
1293                        break;
1294                case CODA_HX4:
1295                case CODA_7541:
1296                        coda_write(dev, ctx->iram_info.search_ram_paddr,
1297                                        CODA7_CMD_ENC_SEQ_SEARCH_BASE);
1298                        coda_write(dev, ctx->iram_info.search_ram_size,
1299                                        CODA7_CMD_ENC_SEQ_SEARCH_SIZE);
1300                        break;
1301                case CODA_960:
1302                        coda_write(dev, 0, CODA9_CMD_ENC_SEQ_ME_OPTION);
1303                        coda_write(dev, 0, CODA9_CMD_ENC_SEQ_INTRA_WEIGHT);
1304                }
1305        }
1306
1307        ret = coda_command_sync(ctx, CODA_COMMAND_SEQ_INIT);
1308        if (ret < 0) {
1309                v4l2_err(v4l2_dev, "CODA_COMMAND_SEQ_INIT timeout\n");
1310                goto out;
1311        }
1312
1313        if (coda_read(dev, CODA_RET_ENC_SEQ_SUCCESS) == 0) {
1314                v4l2_err(v4l2_dev, "CODA_COMMAND_SEQ_INIT failed\n");
1315                ret = -EFAULT;
1316                goto out;
1317        }
1318        ctx->initialized = 1;
1319
1320        if (dst_fourcc != V4L2_PIX_FMT_JPEG) {
1321                if (dev->devtype->product == CODA_960)
1322                        ctx->num_internal_frames = 4;
1323                else
1324                        ctx->num_internal_frames = 2;
1325                ret = coda_alloc_framebuffers(ctx, q_data_src, dst_fourcc);
1326                if (ret < 0) {
1327                        v4l2_err(v4l2_dev, "failed to allocate framebuffers\n");
1328                        goto out;
1329                }
1330                num_fb = 2;
1331                stride = q_data_src->bytesperline;
1332        } else {
1333                ctx->num_internal_frames = 0;
1334                num_fb = 0;
1335                stride = 0;
1336        }
1337        coda_write(dev, num_fb, CODA_CMD_SET_FRAME_BUF_NUM);
1338        coda_write(dev, stride, CODA_CMD_SET_FRAME_BUF_STRIDE);
1339
1340        if (dev->devtype->product == CODA_HX4 ||
1341            dev->devtype->product == CODA_7541) {
1342                coda_write(dev, q_data_src->bytesperline,
1343                                CODA7_CMD_SET_FRAME_SOURCE_BUF_STRIDE);
1344        }
1345        if (dev->devtype->product != CODA_DX6) {
1346                coda_write(dev, ctx->iram_info.buf_bit_use,
1347                                CODA7_CMD_SET_FRAME_AXI_BIT_ADDR);
1348                coda_write(dev, ctx->iram_info.buf_ip_ac_dc_use,
1349                                CODA7_CMD_SET_FRAME_AXI_IPACDC_ADDR);
1350                coda_write(dev, ctx->iram_info.buf_dbk_y_use,
1351                                CODA7_CMD_SET_FRAME_AXI_DBKY_ADDR);
1352                coda_write(dev, ctx->iram_info.buf_dbk_c_use,
1353                                CODA7_CMD_SET_FRAME_AXI_DBKC_ADDR);
1354                coda_write(dev, ctx->iram_info.buf_ovl_use,
1355                                CODA7_CMD_SET_FRAME_AXI_OVL_ADDR);
1356                if (dev->devtype->product == CODA_960) {
1357                        coda_write(dev, ctx->iram_info.buf_btp_use,
1358                                        CODA9_CMD_SET_FRAME_AXI_BTP_ADDR);
1359
1360                        coda9_set_frame_cache(ctx, q_data_src->fourcc);
1361
1362                        /* FIXME */
1363                        coda_write(dev, ctx->internal_frames[2].buf.paddr,
1364                                   CODA9_CMD_SET_FRAME_SUBSAMP_A);
1365                        coda_write(dev, ctx->internal_frames[3].buf.paddr,
1366                                   CODA9_CMD_SET_FRAME_SUBSAMP_B);
1367                }
1368        }
1369
1370        ret = coda_command_sync(ctx, CODA_COMMAND_SET_FRAME_BUF);
1371        if (ret < 0) {
1372                v4l2_err(v4l2_dev, "CODA_COMMAND_SET_FRAME_BUF timeout\n");
1373                goto out;
1374        }
1375
1376        coda_dbg(1, ctx, "start encoding %dx%d %4.4s->%4.4s @ %d/%d Hz\n",
1377                 q_data_src->rect.width, q_data_src->rect.height,
1378                 (char *)&ctx->codec->src_fourcc, (char *)&dst_fourcc,
1379                 ctx->params.framerate & 0xffff,
1380                 (ctx->params.framerate >> 16) + 1);
1381
1382        /* Save stream headers */
1383        buf = v4l2_m2m_next_dst_buf(ctx->fh.m2m_ctx);
1384        switch (dst_fourcc) {
1385        case V4L2_PIX_FMT_H264:
1386                /*
1387                 * Get SPS in the first frame and copy it to an
1388                 * intermediate buffer.
1389                 */
1390                ret = coda_encode_header(ctx, buf, CODA_HEADER_H264_SPS,
1391                                         &ctx->vpu_header[0][0],
1392                                         &ctx->vpu_header_size[0]);
1393                if (ret < 0)
1394                        goto out;
1395
1396                /*
1397                 * If visible width or height are not aligned to macroblock
1398                 * size, the crop_right and crop_bottom SPS fields must be set
1399                 * to the difference between visible and coded size.  This is
1400                 * only supported by CODA960 firmware. All others do not allow
1401                 * writing frame cropping parameters, so we have to manually
1402                 * fix up the SPS RBSP (Sequence Parameter Set Raw Byte
1403                 * Sequence Payload) ourselves.
1404                 */
1405                if (ctx->dev->devtype->product != CODA_960 &&
1406                    ((q_data_src->rect.width % 16) ||
1407                     (q_data_src->rect.height % 16))) {
1408                        ret = coda_h264_sps_fixup(ctx, q_data_src->rect.width,
1409                                                  q_data_src->rect.height,
1410                                                  &ctx->vpu_header[0][0],
1411                                                  &ctx->vpu_header_size[0],
1412                                                  sizeof(ctx->vpu_header[0]));
1413                        if (ret < 0)
1414                                goto out;
1415                }
1416
1417                /*
1418                 * Get PPS in the first frame and copy it to an
1419                 * intermediate buffer.
1420                 */
1421                ret = coda_encode_header(ctx, buf, CODA_HEADER_H264_PPS,
1422                                         &ctx->vpu_header[1][0],
1423                                         &ctx->vpu_header_size[1]);
1424                if (ret < 0)
1425                        goto out;
1426
1427                /*
1428                 * Length of H.264 headers is variable and thus it might not be
1429                 * aligned for the coda to append the encoded frame. In that is
1430                 * the case a filler NAL must be added to header 2.
1431                 */
1432                ctx->vpu_header_size[2] = coda_h264_padding(
1433                                        (ctx->vpu_header_size[0] +
1434                                         ctx->vpu_header_size[1]),
1435                                         ctx->vpu_header[2]);
1436                break;
1437        case V4L2_PIX_FMT_MPEG4:
1438                /*
1439                 * Get VOS in the first frame and copy it to an
1440                 * intermediate buffer
1441                 */
1442                ret = coda_encode_header(ctx, buf, CODA_HEADER_MP4V_VOS,
1443                                         &ctx->vpu_header[0][0],
1444                                         &ctx->vpu_header_size[0]);
1445                if (ret < 0)
1446                        goto out;
1447
1448                ret = coda_encode_header(ctx, buf, CODA_HEADER_MP4V_VIS,
1449                                         &ctx->vpu_header[1][0],
1450                                         &ctx->vpu_header_size[1]);
1451                if (ret < 0)
1452                        goto out;
1453
1454                ret = coda_encode_header(ctx, buf, CODA_HEADER_MP4V_VOL,
1455                                         &ctx->vpu_header[2][0],
1456                                         &ctx->vpu_header_size[2]);
1457                if (ret < 0)
1458                        goto out;
1459                break;
1460        default:
1461                /* No more formats need to save headers at the moment */
1462                break;
1463        }
1464
1465out:
1466        mutex_unlock(&dev->coda_mutex);
1467        return ret;
1468}
1469
1470static int coda_prepare_encode(struct coda_ctx *ctx)
1471{
1472        struct coda_q_data *q_data_src, *q_data_dst;
1473        struct vb2_v4l2_buffer *src_buf, *dst_buf;
1474        struct coda_dev *dev = ctx->dev;
1475        int force_ipicture;
1476        int quant_param = 0;
1477        u32 pic_stream_buffer_addr, pic_stream_buffer_size;
1478        u32 rot_mode = 0;
1479        u32 dst_fourcc;
1480        u32 reg;
1481        int ret;
1482
1483        ret = coda_enc_param_change(ctx);
1484        if (ret < 0) {
1485                v4l2_warn(&ctx->dev->v4l2_dev, "parameter change failed: %d\n",
1486                          ret);
1487        }
1488
1489        src_buf = v4l2_m2m_next_src_buf(ctx->fh.m2m_ctx);
1490        dst_buf = v4l2_m2m_next_dst_buf(ctx->fh.m2m_ctx);
1491        q_data_src = get_q_data(ctx, V4L2_BUF_TYPE_VIDEO_OUTPUT);
1492        q_data_dst = get_q_data(ctx, V4L2_BUF_TYPE_VIDEO_CAPTURE);
1493        dst_fourcc = q_data_dst->fourcc;
1494
1495        src_buf->sequence = ctx->osequence;
1496        dst_buf->sequence = ctx->osequence;
1497        ctx->osequence++;
1498
1499        force_ipicture = ctx->params.force_ipicture;
1500        if (force_ipicture)
1501                ctx->params.force_ipicture = false;
1502        else if (ctx->params.gop_size != 0 &&
1503                 (src_buf->sequence % ctx->params.gop_size) == 0)
1504                force_ipicture = 1;
1505
1506        /*
1507         * Workaround coda firmware BUG that only marks the first
1508         * frame as IDR. This is a problem for some decoders that can't
1509         * recover when a frame is lost.
1510         */
1511        if (!force_ipicture) {
1512                src_buf->flags |= V4L2_BUF_FLAG_PFRAME;
1513                src_buf->flags &= ~V4L2_BUF_FLAG_KEYFRAME;
1514        } else {
1515                src_buf->flags |= V4L2_BUF_FLAG_KEYFRAME;
1516                src_buf->flags &= ~V4L2_BUF_FLAG_PFRAME;
1517        }
1518
1519        if (dev->devtype->product == CODA_960)
1520                coda_set_gdi_regs(ctx);
1521
1522        /*
1523         * Copy headers in front of the first frame and forced I frames for
1524         * H.264 only. In MPEG4 they are already copied by the CODA.
1525         */
1526        if (src_buf->sequence == 0 || force_ipicture) {
1527                pic_stream_buffer_addr =
1528                        vb2_dma_contig_plane_dma_addr(&dst_buf->vb2_buf, 0) +
1529                        ctx->vpu_header_size[0] +
1530                        ctx->vpu_header_size[1] +
1531                        ctx->vpu_header_size[2];
1532                pic_stream_buffer_size = q_data_dst->sizeimage -
1533                        ctx->vpu_header_size[0] -
1534                        ctx->vpu_header_size[1] -
1535                        ctx->vpu_header_size[2];
1536                memcpy(vb2_plane_vaddr(&dst_buf->vb2_buf, 0),
1537                       &ctx->vpu_header[0][0], ctx->vpu_header_size[0]);
1538                memcpy(vb2_plane_vaddr(&dst_buf->vb2_buf, 0)
1539                        + ctx->vpu_header_size[0], &ctx->vpu_header[1][0],
1540                        ctx->vpu_header_size[1]);
1541                memcpy(vb2_plane_vaddr(&dst_buf->vb2_buf, 0)
1542                        + ctx->vpu_header_size[0] + ctx->vpu_header_size[1],
1543                        &ctx->vpu_header[2][0], ctx->vpu_header_size[2]);
1544        } else {
1545                pic_stream_buffer_addr =
1546                        vb2_dma_contig_plane_dma_addr(&dst_buf->vb2_buf, 0);
1547                pic_stream_buffer_size = q_data_dst->sizeimage;
1548        }
1549
1550        if (force_ipicture) {
1551                switch (dst_fourcc) {
1552                case V4L2_PIX_FMT_H264:
1553                        quant_param = ctx->params.h264_intra_qp;
1554                        break;
1555                case V4L2_PIX_FMT_MPEG4:
1556                        quant_param = ctx->params.mpeg4_intra_qp;
1557                        break;
1558                case V4L2_PIX_FMT_JPEG:
1559                        quant_param = 30;
1560                        break;
1561                default:
1562                        v4l2_warn(&ctx->dev->v4l2_dev,
1563                                "cannot set intra qp, fmt not supported\n");
1564                        break;
1565                }
1566        } else {
1567                switch (dst_fourcc) {
1568                case V4L2_PIX_FMT_H264:
1569                        quant_param = ctx->params.h264_inter_qp;
1570                        break;
1571                case V4L2_PIX_FMT_MPEG4:
1572                        quant_param = ctx->params.mpeg4_inter_qp;
1573                        break;
1574                default:
1575                        v4l2_warn(&ctx->dev->v4l2_dev,
1576                                "cannot set inter qp, fmt not supported\n");
1577                        break;
1578                }
1579        }
1580
1581        /* submit */
1582        if (ctx->params.rot_mode)
1583                rot_mode = CODA_ROT_MIR_ENABLE | ctx->params.rot_mode;
1584        coda_write(dev, rot_mode, CODA_CMD_ENC_PIC_ROT_MODE);
1585        coda_write(dev, quant_param, CODA_CMD_ENC_PIC_QS);
1586
1587        if (dev->devtype->product == CODA_960) {
1588                coda_write(dev, 4/*FIXME: 0*/, CODA9_CMD_ENC_PIC_SRC_INDEX);
1589                coda_write(dev, q_data_src->bytesperline,
1590                           CODA9_CMD_ENC_PIC_SRC_STRIDE);
1591                coda_write(dev, 0, CODA9_CMD_ENC_PIC_SUB_FRAME_SYNC);
1592
1593                reg = CODA9_CMD_ENC_PIC_SRC_ADDR_Y;
1594        } else {
1595                reg = CODA_CMD_ENC_PIC_SRC_ADDR_Y;
1596        }
1597        coda_write_base(ctx, q_data_src, src_buf, reg);
1598
1599        coda_write(dev, force_ipicture << 1 & 0x2,
1600                   CODA_CMD_ENC_PIC_OPTION);
1601
1602        coda_write(dev, pic_stream_buffer_addr, CODA_CMD_ENC_PIC_BB_START);
1603        coda_write(dev, pic_stream_buffer_size / 1024,
1604                   CODA_CMD_ENC_PIC_BB_SIZE);
1605
1606        if (!ctx->streamon_out) {
1607                /* After streamoff on the output side, set stream end flag */
1608                ctx->bit_stream_param |= CODA_BIT_STREAM_END_FLAG;
1609                coda_write(dev, ctx->bit_stream_param,
1610                           CODA_REG_BIT_BIT_STREAM_PARAM);
1611        }
1612
1613        if (dev->devtype->product != CODA_DX6)
1614                coda_write(dev, ctx->iram_info.axi_sram_use,
1615                                CODA7_REG_BIT_AXI_SRAM_USE);
1616
1617        trace_coda_enc_pic_run(ctx, src_buf);
1618
1619        coda_command_async(ctx, CODA_COMMAND_PIC_RUN);
1620
1621        return 0;
1622}
1623
1624static char coda_frame_type_char(u32 flags)
1625{
1626        return (flags & V4L2_BUF_FLAG_KEYFRAME) ? 'I' :
1627               (flags & V4L2_BUF_FLAG_PFRAME) ? 'P' :
1628               (flags & V4L2_BUF_FLAG_BFRAME) ? 'B' : '?';
1629}
1630
1631static void coda_finish_encode(struct coda_ctx *ctx)
1632{
1633        struct vb2_v4l2_buffer *src_buf, *dst_buf;
1634        struct coda_dev *dev = ctx->dev;
1635        u32 wr_ptr, start_ptr;
1636
1637        if (ctx->aborting)
1638                return;
1639
1640        /*
1641         * Lock to make sure that an encoder stop command running in parallel
1642         * will either already have marked src_buf as last, or it will wake up
1643         * the capture queue after the buffers are returned.
1644         */
1645        mutex_lock(&ctx->wakeup_mutex);
1646        src_buf = v4l2_m2m_src_buf_remove(ctx->fh.m2m_ctx);
1647        dst_buf = v4l2_m2m_next_dst_buf(ctx->fh.m2m_ctx);
1648
1649        trace_coda_enc_pic_done(ctx, dst_buf);
1650
1651        /* Get results from the coda */
1652        start_ptr = coda_read(dev, CODA_CMD_ENC_PIC_BB_START);
1653        wr_ptr = coda_read(dev, CODA_REG_BIT_WR_PTR(ctx->reg_idx));
1654
1655        /* Calculate bytesused field */
1656        if (dst_buf->sequence == 0 ||
1657            src_buf->flags & V4L2_BUF_FLAG_KEYFRAME) {
1658                vb2_set_plane_payload(&dst_buf->vb2_buf, 0, wr_ptr - start_ptr +
1659                                        ctx->vpu_header_size[0] +
1660                                        ctx->vpu_header_size[1] +
1661                                        ctx->vpu_header_size[2]);
1662        } else {
1663                vb2_set_plane_payload(&dst_buf->vb2_buf, 0, wr_ptr - start_ptr);
1664        }
1665
1666        coda_dbg(1, ctx, "frame size = %u\n", wr_ptr - start_ptr);
1667
1668        coda_read(dev, CODA_RET_ENC_PIC_SLICE_NUM);
1669        coda_read(dev, CODA_RET_ENC_PIC_FLAG);
1670
1671        dst_buf->flags &= ~(V4L2_BUF_FLAG_KEYFRAME |
1672                            V4L2_BUF_FLAG_PFRAME |
1673                            V4L2_BUF_FLAG_LAST);
1674        if (coda_read(dev, CODA_RET_ENC_PIC_TYPE) == 0)
1675                dst_buf->flags |= V4L2_BUF_FLAG_KEYFRAME;
1676        else
1677                dst_buf->flags |= V4L2_BUF_FLAG_PFRAME;
1678        dst_buf->flags |= src_buf->flags & V4L2_BUF_FLAG_LAST;
1679
1680        v4l2_m2m_buf_copy_metadata(src_buf, dst_buf, false);
1681
1682        v4l2_m2m_buf_done(src_buf, VB2_BUF_STATE_DONE);
1683
1684        dst_buf = v4l2_m2m_dst_buf_remove(ctx->fh.m2m_ctx);
1685        coda_m2m_buf_done(ctx, dst_buf, VB2_BUF_STATE_DONE);
1686        mutex_unlock(&ctx->wakeup_mutex);
1687
1688        ctx->gopcounter--;
1689        if (ctx->gopcounter < 0)
1690                ctx->gopcounter = ctx->params.gop_size - 1;
1691
1692        coda_dbg(1, ctx, "job finished: encoded %c frame (%d)%s\n",
1693                 coda_frame_type_char(dst_buf->flags), dst_buf->sequence,
1694                 (dst_buf->flags & V4L2_BUF_FLAG_LAST) ? " (last)" : "");
1695}
1696
1697static void coda_seq_end_work(struct work_struct *work)
1698{
1699        struct coda_ctx *ctx = container_of(work, struct coda_ctx, seq_end_work);
1700        struct coda_dev *dev = ctx->dev;
1701
1702        mutex_lock(&ctx->buffer_mutex);
1703        mutex_lock(&dev->coda_mutex);
1704
1705        if (ctx->initialized == 0)
1706                goto out;
1707
1708        coda_dbg(1, ctx, "%s: sent command 'SEQ_END' to coda\n", __func__);
1709        if (coda_command_sync(ctx, CODA_COMMAND_SEQ_END)) {
1710                v4l2_err(&dev->v4l2_dev,
1711                         "CODA_COMMAND_SEQ_END failed\n");
1712        }
1713
1714        /*
1715         * FIXME: Sometimes h.264 encoding fails with 8-byte sequences missing
1716         * from the output stream after the h.264 decoder has run. Resetting the
1717         * hardware after the decoder has finished seems to help.
1718         */
1719        if (dev->devtype->product == CODA_960)
1720                coda_hw_reset(ctx);
1721
1722        kfifo_init(&ctx->bitstream_fifo,
1723                ctx->bitstream.vaddr, ctx->bitstream.size);
1724
1725        coda_free_framebuffers(ctx);
1726
1727        ctx->initialized = 0;
1728
1729out:
1730        mutex_unlock(&dev->coda_mutex);
1731        mutex_unlock(&ctx->buffer_mutex);
1732}
1733
1734static void coda_bit_release(struct coda_ctx *ctx)
1735{
1736        mutex_lock(&ctx->buffer_mutex);
1737        coda_free_framebuffers(ctx);
1738        coda_free_context_buffers(ctx);
1739        coda_free_bitstream_buffer(ctx);
1740        mutex_unlock(&ctx->buffer_mutex);
1741}
1742
1743const struct coda_context_ops coda_bit_encode_ops = {
1744        .queue_init = coda_encoder_queue_init,
1745        .reqbufs = coda_encoder_reqbufs,
1746        .start_streaming = coda_start_encoding,
1747        .prepare_run = coda_prepare_encode,
1748        .finish_run = coda_finish_encode,
1749        .seq_end_work = coda_seq_end_work,
1750        .release = coda_bit_release,
1751};
1752
1753/*
1754 * Decoder context operations
1755 */
1756
1757static int coda_alloc_bitstream_buffer(struct coda_ctx *ctx,
1758                                       struct coda_q_data *q_data)
1759{
1760        if (ctx->bitstream.vaddr)
1761                return 0;
1762
1763        ctx->bitstream.size = roundup_pow_of_two(q_data->sizeimage * 2);
1764        ctx->bitstream.vaddr = dma_alloc_wc(ctx->dev->dev, ctx->bitstream.size,
1765                                            &ctx->bitstream.paddr, GFP_KERNEL);
1766        if (!ctx->bitstream.vaddr) {
1767                v4l2_err(&ctx->dev->v4l2_dev,
1768                         "failed to allocate bitstream ringbuffer");
1769                return -ENOMEM;
1770        }
1771        kfifo_init(&ctx->bitstream_fifo,
1772                   ctx->bitstream.vaddr, ctx->bitstream.size);
1773
1774        return 0;
1775}
1776
1777static void coda_free_bitstream_buffer(struct coda_ctx *ctx)
1778{
1779        if (ctx->bitstream.vaddr == NULL)
1780                return;
1781
1782        dma_free_wc(ctx->dev->dev, ctx->bitstream.size, ctx->bitstream.vaddr,
1783                    ctx->bitstream.paddr);
1784        ctx->bitstream.vaddr = NULL;
1785        kfifo_init(&ctx->bitstream_fifo, NULL, 0);
1786}
1787
1788static int coda_decoder_reqbufs(struct coda_ctx *ctx,
1789                                struct v4l2_requestbuffers *rb)
1790{
1791        struct coda_q_data *q_data_src;
1792        int ret;
1793
1794        if (rb->type != V4L2_BUF_TYPE_VIDEO_OUTPUT)
1795                return 0;
1796
1797        if (rb->count) {
1798                q_data_src = get_q_data(ctx, V4L2_BUF_TYPE_VIDEO_OUTPUT);
1799                ret = coda_alloc_context_buffers(ctx, q_data_src);
1800                if (ret < 0)
1801                        return ret;
1802                ret = coda_alloc_bitstream_buffer(ctx, q_data_src);
1803                if (ret < 0) {
1804                        coda_free_context_buffers(ctx);
1805                        return ret;
1806                }
1807        } else {
1808                coda_free_bitstream_buffer(ctx);
1809                coda_free_context_buffers(ctx);
1810        }
1811
1812        return 0;
1813}
1814
1815static bool coda_reorder_enable(struct coda_ctx *ctx)
1816{
1817        struct coda_dev *dev = ctx->dev;
1818        int profile;
1819
1820        if (dev->devtype->product != CODA_HX4 &&
1821            dev->devtype->product != CODA_7541 &&
1822            dev->devtype->product != CODA_960)
1823                return false;
1824
1825        if (ctx->codec->src_fourcc == V4L2_PIX_FMT_JPEG)
1826                return false;
1827
1828        if (ctx->codec->src_fourcc != V4L2_PIX_FMT_H264)
1829                return true;
1830
1831        profile = coda_h264_profile(ctx->params.h264_profile_idc);
1832        if (profile < 0)
1833                v4l2_warn(&dev->v4l2_dev, "Unknown H264 Profile: %u\n",
1834                          ctx->params.h264_profile_idc);
1835
1836        /* Baseline profile does not support reordering */
1837        return profile > V4L2_MPEG_VIDEO_H264_PROFILE_BASELINE;
1838}
1839
1840static void coda_decoder_drop_used_metas(struct coda_ctx *ctx)
1841{
1842        struct coda_buffer_meta *meta, *tmp;
1843
1844        /*
1845         * All metas that end at or before the RD pointer (fifo out),
1846         * are now consumed by the VPU and should be released.
1847         */
1848        spin_lock(&ctx->buffer_meta_lock);
1849        list_for_each_entry_safe(meta, tmp, &ctx->buffer_meta_list, list) {
1850                if (ctx->bitstream_fifo.kfifo.out >= meta->end) {
1851                        coda_dbg(2, ctx, "releasing meta: seq=%d start=%d end=%d\n",
1852                                 meta->sequence, meta->start, meta->end);
1853
1854                        list_del(&meta->list);
1855                        ctx->num_metas--;
1856                        ctx->first_frame_sequence++;
1857                        kfree(meta);
1858                }
1859        }
1860        spin_unlock(&ctx->buffer_meta_lock);
1861}
1862
1863static int __coda_decoder_seq_init(struct coda_ctx *ctx)
1864{
1865        struct coda_q_data *q_data_src, *q_data_dst;
1866        u32 bitstream_buf, bitstream_size;
1867        struct coda_dev *dev = ctx->dev;
1868        int width, height;
1869        u32 src_fourcc, dst_fourcc;
1870        u32 val;
1871        int ret;
1872
1873        lockdep_assert_held(&dev->coda_mutex);
1874
1875        coda_dbg(1, ctx, "Video Data Order Adapter: %s\n",
1876                 ctx->use_vdoa ? "Enabled" : "Disabled");
1877
1878        /* Start decoding */
1879        q_data_src = get_q_data(ctx, V4L2_BUF_TYPE_VIDEO_OUTPUT);
1880        q_data_dst = get_q_data(ctx, V4L2_BUF_TYPE_VIDEO_CAPTURE);
1881        bitstream_buf = ctx->bitstream.paddr;
1882        bitstream_size = ctx->bitstream.size;
1883        src_fourcc = q_data_src->fourcc;
1884        dst_fourcc = q_data_dst->fourcc;
1885
1886        /* Update coda bitstream read and write pointers from kfifo */
1887        coda_kfifo_sync_to_device_full(ctx);
1888
1889        ctx->frame_mem_ctrl &= ~(CODA_FRAME_CHROMA_INTERLEAVE | (0x3 << 9) |
1890                                 CODA9_FRAME_TILED2LINEAR);
1891        if (dst_fourcc == V4L2_PIX_FMT_NV12 || dst_fourcc == V4L2_PIX_FMT_YUYV)
1892                ctx->frame_mem_ctrl |= CODA_FRAME_CHROMA_INTERLEAVE;
1893        if (ctx->tiled_map_type == GDI_TILED_FRAME_MB_RASTER_MAP)
1894                ctx->frame_mem_ctrl |= (0x3 << 9) |
1895                        ((ctx->use_vdoa) ? 0 : CODA9_FRAME_TILED2LINEAR);
1896        coda_write(dev, ctx->frame_mem_ctrl, CODA_REG_BIT_FRAME_MEM_CTRL);
1897
1898        ctx->display_idx = -1;
1899        ctx->frm_dis_flg = 0;
1900        coda_write(dev, 0, CODA_REG_BIT_FRM_DIS_FLG(ctx->reg_idx));
1901
1902        coda_write(dev, bitstream_buf, CODA_CMD_DEC_SEQ_BB_START);
1903        coda_write(dev, bitstream_size / 1024, CODA_CMD_DEC_SEQ_BB_SIZE);
1904        val = 0;
1905        if (coda_reorder_enable(ctx))
1906                val |= CODA_REORDER_ENABLE;
1907        if (ctx->codec->src_fourcc == V4L2_PIX_FMT_JPEG)
1908                val |= CODA_NO_INT_ENABLE;
1909        coda_write(dev, val, CODA_CMD_DEC_SEQ_OPTION);
1910
1911        ctx->params.codec_mode = ctx->codec->mode;
1912        if (dev->devtype->product == CODA_960 &&
1913            src_fourcc == V4L2_PIX_FMT_MPEG4)
1914                ctx->params.codec_mode_aux = CODA_MP4_AUX_MPEG4;
1915        else
1916                ctx->params.codec_mode_aux = 0;
1917        if (src_fourcc == V4L2_PIX_FMT_MPEG4) {
1918                coda_write(dev, CODA_MP4_CLASS_MPEG4,
1919                           CODA_CMD_DEC_SEQ_MP4_ASP_CLASS);
1920        }
1921        if (src_fourcc == V4L2_PIX_FMT_H264) {
1922                if (dev->devtype->product == CODA_HX4 ||
1923                    dev->devtype->product == CODA_7541) {
1924                        coda_write(dev, ctx->psbuf.paddr,
1925                                        CODA_CMD_DEC_SEQ_PS_BB_START);
1926                        coda_write(dev, (CODA7_PS_BUF_SIZE / 1024),
1927                                        CODA_CMD_DEC_SEQ_PS_BB_SIZE);
1928                }
1929                if (dev->devtype->product == CODA_960) {
1930                        coda_write(dev, 0, CODA_CMD_DEC_SEQ_X264_MV_EN);
1931                        coda_write(dev, 512, CODA_CMD_DEC_SEQ_SPP_CHUNK_SIZE);
1932                }
1933        }
1934        if (src_fourcc == V4L2_PIX_FMT_JPEG)
1935                coda_write(dev, 0, CODA_CMD_DEC_SEQ_JPG_THUMB_EN);
1936        if (dev->devtype->product != CODA_960)
1937                coda_write(dev, 0, CODA_CMD_DEC_SEQ_SRC_SIZE);
1938
1939        ctx->bit_stream_param = CODA_BIT_DEC_SEQ_INIT_ESCAPE;
1940        ret = coda_command_sync(ctx, CODA_COMMAND_SEQ_INIT);
1941        ctx->bit_stream_param = 0;
1942        if (ret) {
1943                v4l2_err(&dev->v4l2_dev, "CODA_COMMAND_SEQ_INIT timeout\n");
1944                return ret;
1945        }
1946        ctx->sequence_offset = ~0U;
1947        ctx->initialized = 1;
1948        ctx->first_frame_sequence = 0;
1949
1950        /* Update kfifo out pointer from coda bitstream read pointer */
1951        coda_kfifo_sync_from_device(ctx);
1952
1953        /*
1954         * After updating the read pointer, we need to check if
1955         * any metas are consumed and should be released.
1956         */
1957        coda_decoder_drop_used_metas(ctx);
1958
1959        if (coda_read(dev, CODA_RET_DEC_SEQ_SUCCESS) == 0) {
1960                v4l2_err(&dev->v4l2_dev,
1961                        "CODA_COMMAND_SEQ_INIT failed, error code = 0x%x\n",
1962                        coda_read(dev, CODA_RET_DEC_SEQ_ERR_REASON));
1963                return -EAGAIN;
1964        }
1965
1966        val = coda_read(dev, CODA_RET_DEC_SEQ_SRC_SIZE);
1967        if (dev->devtype->product == CODA_DX6) {
1968                width = (val >> CODADX6_PICWIDTH_OFFSET) & CODADX6_PICWIDTH_MASK;
1969                height = val & CODADX6_PICHEIGHT_MASK;
1970        } else {
1971                width = (val >> CODA7_PICWIDTH_OFFSET) & CODA7_PICWIDTH_MASK;
1972                height = val & CODA7_PICHEIGHT_MASK;
1973        }
1974
1975        if (width > q_data_dst->bytesperline || height > q_data_dst->height) {
1976                v4l2_err(&dev->v4l2_dev, "stream is %dx%d, not %dx%d\n",
1977                         width, height, q_data_dst->bytesperline,
1978                         q_data_dst->height);
1979                return -EINVAL;
1980        }
1981
1982        width = round_up(width, 16);
1983        height = round_up(height, 16);
1984
1985        coda_dbg(1, ctx, "start decoding: %dx%d\n", width, height);
1986
1987        ctx->num_internal_frames = coda_read(dev, CODA_RET_DEC_SEQ_FRAME_NEED);
1988        /*
1989         * If the VDOA is used, the decoder needs one additional frame,
1990         * because the frames are freed when the next frame is decoded.
1991         * Otherwise there are visible errors in the decoded frames (green
1992         * regions in displayed frames) and a broken order of frames (earlier
1993         * frames are sporadically displayed after later frames).
1994         */
1995        if (ctx->use_vdoa)
1996                ctx->num_internal_frames += 1;
1997        if (ctx->num_internal_frames > CODA_MAX_FRAMEBUFFERS) {
1998                v4l2_err(&dev->v4l2_dev,
1999                         "not enough framebuffers to decode (%d < %d)\n",
2000                         CODA_MAX_FRAMEBUFFERS, ctx->num_internal_frames);
2001                return -EINVAL;
2002        }
2003
2004        if (src_fourcc == V4L2_PIX_FMT_H264) {
2005                u32 left_right;
2006                u32 top_bottom;
2007
2008                left_right = coda_read(dev, CODA_RET_DEC_SEQ_CROP_LEFT_RIGHT);
2009                top_bottom = coda_read(dev, CODA_RET_DEC_SEQ_CROP_TOP_BOTTOM);
2010
2011                q_data_dst->rect.left = (left_right >> 10) & 0x3ff;
2012                q_data_dst->rect.top = (top_bottom >> 10) & 0x3ff;
2013                q_data_dst->rect.width = width - q_data_dst->rect.left -
2014                                         (left_right & 0x3ff);
2015                q_data_dst->rect.height = height - q_data_dst->rect.top -
2016                                          (top_bottom & 0x3ff);
2017        }
2018
2019        if (dev->devtype->product != CODA_DX6) {
2020                u8 profile, level;
2021
2022                val = coda_read(dev, CODA7_RET_DEC_SEQ_HEADER_REPORT);
2023                profile = val & 0xff;
2024                level = (val >> 8) & 0x7f;
2025
2026                if (profile || level)
2027                        coda_update_profile_level_ctrls(ctx, profile, level);
2028        }
2029
2030        return 0;
2031}
2032
2033static void coda_dec_seq_init_work(struct work_struct *work)
2034{
2035        struct coda_ctx *ctx = container_of(work,
2036                                            struct coda_ctx, seq_init_work);
2037        struct coda_dev *dev = ctx->dev;
2038
2039        mutex_lock(&ctx->buffer_mutex);
2040        mutex_lock(&dev->coda_mutex);
2041
2042        if (!ctx->initialized)
2043                __coda_decoder_seq_init(ctx);
2044
2045        mutex_unlock(&dev->coda_mutex);
2046        mutex_unlock(&ctx->buffer_mutex);
2047}
2048
2049static int __coda_start_decoding(struct coda_ctx *ctx)
2050{
2051        struct coda_q_data *q_data_src, *q_data_dst;
2052        struct coda_dev *dev = ctx->dev;
2053        u32 src_fourcc, dst_fourcc;
2054        int ret;
2055
2056        if (!ctx->initialized) {
2057                ret = __coda_decoder_seq_init(ctx);
2058                if (ret < 0)
2059                        return ret;
2060        }
2061
2062        q_data_src = get_q_data(ctx, V4L2_BUF_TYPE_VIDEO_OUTPUT);
2063        q_data_dst = get_q_data(ctx, V4L2_BUF_TYPE_VIDEO_CAPTURE);
2064        src_fourcc = q_data_src->fourcc;
2065        dst_fourcc = q_data_dst->fourcc;
2066
2067        coda_write(dev, ctx->parabuf.paddr, CODA_REG_BIT_PARA_BUF_ADDR);
2068
2069        ret = coda_alloc_framebuffers(ctx, q_data_dst, src_fourcc);
2070        if (ret < 0) {
2071                v4l2_err(&dev->v4l2_dev, "failed to allocate framebuffers\n");
2072                return ret;
2073        }
2074
2075        /* Tell the decoder how many frame buffers we allocated. */
2076        coda_write(dev, ctx->num_internal_frames, CODA_CMD_SET_FRAME_BUF_NUM);
2077        coda_write(dev, round_up(q_data_dst->rect.width, 16),
2078                   CODA_CMD_SET_FRAME_BUF_STRIDE);
2079
2080        if (dev->devtype->product != CODA_DX6) {
2081                /* Set secondary AXI IRAM */
2082                coda_setup_iram(ctx);
2083
2084                coda_write(dev, ctx->iram_info.buf_bit_use,
2085                                CODA7_CMD_SET_FRAME_AXI_BIT_ADDR);
2086                coda_write(dev, ctx->iram_info.buf_ip_ac_dc_use,
2087                                CODA7_CMD_SET_FRAME_AXI_IPACDC_ADDR);
2088                coda_write(dev, ctx->iram_info.buf_dbk_y_use,
2089                                CODA7_CMD_SET_FRAME_AXI_DBKY_ADDR);
2090                coda_write(dev, ctx->iram_info.buf_dbk_c_use,
2091                                CODA7_CMD_SET_FRAME_AXI_DBKC_ADDR);
2092                coda_write(dev, ctx->iram_info.buf_ovl_use,
2093                                CODA7_CMD_SET_FRAME_AXI_OVL_ADDR);
2094                if (dev->devtype->product == CODA_960) {
2095                        coda_write(dev, ctx->iram_info.buf_btp_use,
2096                                        CODA9_CMD_SET_FRAME_AXI_BTP_ADDR);
2097
2098                        coda_write(dev, -1, CODA9_CMD_SET_FRAME_DELAY);
2099                        coda9_set_frame_cache(ctx, dst_fourcc);
2100                }
2101        }
2102
2103        if (src_fourcc == V4L2_PIX_FMT_H264) {
2104                coda_write(dev, ctx->slicebuf.paddr,
2105                                CODA_CMD_SET_FRAME_SLICE_BB_START);
2106                coda_write(dev, ctx->slicebuf.size / 1024,
2107                                CODA_CMD_SET_FRAME_SLICE_BB_SIZE);
2108        }
2109
2110        if (dev->devtype->product == CODA_HX4 ||
2111            dev->devtype->product == CODA_7541) {
2112                int max_mb_x = 1920 / 16;
2113                int max_mb_y = 1088 / 16;
2114                int max_mb_num = max_mb_x * max_mb_y;
2115
2116                coda_write(dev, max_mb_num << 16 | max_mb_x << 8 | max_mb_y,
2117                                CODA7_CMD_SET_FRAME_MAX_DEC_SIZE);
2118        } else if (dev->devtype->product == CODA_960) {
2119                int max_mb_x = 1920 / 16;
2120                int max_mb_y = 1088 / 16;
2121                int max_mb_num = max_mb_x * max_mb_y;
2122
2123                coda_write(dev, max_mb_num << 16 | max_mb_x << 8 | max_mb_y,
2124                                CODA9_CMD_SET_FRAME_MAX_DEC_SIZE);
2125        }
2126
2127        if (coda_command_sync(ctx, CODA_COMMAND_SET_FRAME_BUF)) {
2128                v4l2_err(&ctx->dev->v4l2_dev,
2129                         "CODA_COMMAND_SET_FRAME_BUF timeout\n");
2130                return -ETIMEDOUT;
2131        }
2132
2133        return 0;
2134}
2135
2136static int coda_start_decoding(struct coda_ctx *ctx)
2137{
2138        struct coda_dev *dev = ctx->dev;
2139        int ret;
2140
2141        mutex_lock(&dev->coda_mutex);
2142        ret = __coda_start_decoding(ctx);
2143        mutex_unlock(&dev->coda_mutex);
2144
2145        return ret;
2146}
2147
2148static int coda_prepare_decode(struct coda_ctx *ctx)
2149{
2150        struct vb2_v4l2_buffer *dst_buf;
2151        struct coda_dev *dev = ctx->dev;
2152        struct coda_q_data *q_data_dst;
2153        struct coda_buffer_meta *meta;
2154        u32 rot_mode = 0;
2155        u32 reg_addr, reg_stride;
2156
2157        dst_buf = v4l2_m2m_next_dst_buf(ctx->fh.m2m_ctx);
2158        q_data_dst = get_q_data(ctx, V4L2_BUF_TYPE_VIDEO_CAPTURE);
2159
2160        /* Try to copy source buffer contents into the bitstream ringbuffer */
2161        mutex_lock(&ctx->bitstream_mutex);
2162        coda_fill_bitstream(ctx, NULL);
2163        mutex_unlock(&ctx->bitstream_mutex);
2164
2165        if (coda_get_bitstream_payload(ctx) < 512 &&
2166            (!(ctx->bit_stream_param & CODA_BIT_STREAM_END_FLAG))) {
2167                coda_dbg(1, ctx, "bitstream payload: %d, skipping\n",
2168                         coda_get_bitstream_payload(ctx));
2169                v4l2_m2m_job_finish(ctx->dev->m2m_dev, ctx->fh.m2m_ctx);
2170                return -EAGAIN;
2171        }
2172
2173        /* Run coda_start_decoding (again) if not yet initialized */
2174        if (!ctx->initialized) {
2175                int ret = __coda_start_decoding(ctx);
2176
2177                if (ret < 0) {
2178                        v4l2_err(&dev->v4l2_dev, "failed to start decoding\n");
2179                        v4l2_m2m_job_finish(ctx->dev->m2m_dev, ctx->fh.m2m_ctx);
2180                        return -EAGAIN;
2181                } else {
2182                        ctx->initialized = 1;
2183                }
2184        }
2185
2186        if (dev->devtype->product == CODA_960)
2187                coda_set_gdi_regs(ctx);
2188
2189        if (ctx->use_vdoa &&
2190            ctx->display_idx >= 0 &&
2191            ctx->display_idx < ctx->num_internal_frames) {
2192                vdoa_device_run(ctx->vdoa,
2193                                vb2_dma_contig_plane_dma_addr(&dst_buf->vb2_buf, 0),
2194                                ctx->internal_frames[ctx->display_idx].buf.paddr);
2195        } else {
2196                if (dev->devtype->product == CODA_960) {
2197                        /*
2198                         * It was previously assumed that the CODA960 has an
2199                         * internal list of 64 buffer entries that contains
2200                         * both the registered internal frame buffers as well
2201                         * as the rotator buffer output, and that the ROT_INDEX
2202                         * register must be set to a value between the last
2203                         * internal frame buffers' index and 64.
2204                         * At least on firmware version 3.1.1 it turns out that
2205                         * setting ROT_INDEX to any value >= 32 causes CODA
2206                         * hangups that it can not recover from with the SRC VPU
2207                         * reset.
2208                         * It does appear to work however, to just set it to a
2209                         * fixed value in the [ctx->num_internal_frames, 31]
2210                         * range, for example CODA_MAX_FRAMEBUFFERS.
2211                         */
2212                        coda_write(dev, CODA_MAX_FRAMEBUFFERS,
2213                                   CODA9_CMD_DEC_PIC_ROT_INDEX);
2214
2215                        reg_addr = CODA9_CMD_DEC_PIC_ROT_ADDR_Y;
2216                        reg_stride = CODA9_CMD_DEC_PIC_ROT_STRIDE;
2217                } else {
2218                        reg_addr = CODA_CMD_DEC_PIC_ROT_ADDR_Y;
2219                        reg_stride = CODA_CMD_DEC_PIC_ROT_STRIDE;
2220                }
2221                coda_write_base(ctx, q_data_dst, dst_buf, reg_addr);
2222                coda_write(dev, q_data_dst->bytesperline, reg_stride);
2223
2224                rot_mode = CODA_ROT_MIR_ENABLE | ctx->params.rot_mode;
2225        }
2226
2227        coda_write(dev, rot_mode, CODA_CMD_DEC_PIC_ROT_MODE);
2228
2229        switch (dev->devtype->product) {
2230        case CODA_DX6:
2231                /* TBD */
2232        case CODA_HX4:
2233        case CODA_7541:
2234                coda_write(dev, CODA_PRE_SCAN_EN, CODA_CMD_DEC_PIC_OPTION);
2235                break;
2236        case CODA_960:
2237                /* 'hardcode to use interrupt disable mode'? */
2238                coda_write(dev, (1 << 10), CODA_CMD_DEC_PIC_OPTION);
2239                break;
2240        }
2241
2242        coda_write(dev, 0, CODA_CMD_DEC_PIC_SKIP_NUM);
2243
2244        coda_write(dev, 0, CODA_CMD_DEC_PIC_BB_START);
2245        coda_write(dev, 0, CODA_CMD_DEC_PIC_START_BYTE);
2246
2247        if (dev->devtype->product != CODA_DX6)
2248                coda_write(dev, ctx->iram_info.axi_sram_use,
2249                                CODA7_REG_BIT_AXI_SRAM_USE);
2250
2251        spin_lock(&ctx->buffer_meta_lock);
2252        meta = list_first_entry_or_null(&ctx->buffer_meta_list,
2253                                        struct coda_buffer_meta, list);
2254
2255        if (meta && ctx->codec->src_fourcc == V4L2_PIX_FMT_JPEG) {
2256
2257                /* If this is the last buffer in the bitstream, add padding */
2258                if (meta->end == ctx->bitstream_fifo.kfifo.in) {
2259                        static unsigned char buf[512];
2260                        unsigned int pad;
2261
2262                        /* Pad to multiple of 256 and then add 256 more */
2263                        pad = ((0 - meta->end) & 0xff) + 256;
2264
2265                        memset(buf, 0xff, sizeof(buf));
2266
2267                        kfifo_in(&ctx->bitstream_fifo, buf, pad);
2268                }
2269        }
2270        spin_unlock(&ctx->buffer_meta_lock);
2271
2272        coda_kfifo_sync_to_device_full(ctx);
2273
2274        /* Clear decode success flag */
2275        coda_write(dev, 0, CODA_RET_DEC_PIC_SUCCESS);
2276
2277        /* Clear error return value */
2278        coda_write(dev, 0, CODA_RET_DEC_PIC_ERR_MB);
2279
2280        trace_coda_dec_pic_run(ctx, meta);
2281
2282        coda_command_async(ctx, CODA_COMMAND_PIC_RUN);
2283
2284        return 0;
2285}
2286
2287static void coda_finish_decode(struct coda_ctx *ctx)
2288{
2289        struct coda_dev *dev = ctx->dev;
2290        struct coda_q_data *q_data_src;
2291        struct coda_q_data *q_data_dst;
2292        struct vb2_v4l2_buffer *dst_buf;
2293        struct coda_buffer_meta *meta;
2294        int width, height;
2295        int decoded_idx;
2296        int display_idx;
2297        struct coda_internal_frame *decoded_frame = NULL;
2298        u32 src_fourcc;
2299        int success;
2300        u32 err_mb;
2301        int err_vdoa = 0;
2302        u32 val;
2303
2304        if (ctx->aborting)
2305                return;
2306
2307        /* Update kfifo out pointer from coda bitstream read pointer */
2308        coda_kfifo_sync_from_device(ctx);
2309
2310        /*
2311         * in stream-end mode, the read pointer can overshoot the write pointer
2312         * by up to 512 bytes
2313         */
2314        if (ctx->bit_stream_param & CODA_BIT_STREAM_END_FLAG) {
2315                if (coda_get_bitstream_payload(ctx) >= ctx->bitstream.size - 512)
2316                        kfifo_init(&ctx->bitstream_fifo,
2317                                ctx->bitstream.vaddr, ctx->bitstream.size);
2318        }
2319
2320        q_data_src = get_q_data(ctx, V4L2_BUF_TYPE_VIDEO_OUTPUT);
2321        src_fourcc = q_data_src->fourcc;
2322
2323        val = coda_read(dev, CODA_RET_DEC_PIC_SUCCESS);
2324        if (val != 1)
2325                pr_err("DEC_PIC_SUCCESS = %d\n", val);
2326
2327        success = val & 0x1;
2328        if (!success)
2329                v4l2_err(&dev->v4l2_dev, "decode failed\n");
2330
2331        if (src_fourcc == V4L2_PIX_FMT_H264) {
2332                if (val & (1 << 3))
2333                        v4l2_err(&dev->v4l2_dev,
2334                                 "insufficient PS buffer space (%d bytes)\n",
2335                                 ctx->psbuf.size);
2336                if (val & (1 << 2))
2337                        v4l2_err(&dev->v4l2_dev,
2338                                 "insufficient slice buffer space (%d bytes)\n",
2339                                 ctx->slicebuf.size);
2340        }
2341
2342        val = coda_read(dev, CODA_RET_DEC_PIC_SIZE);
2343        width = (val >> 16) & 0xffff;
2344        height = val & 0xffff;
2345
2346        q_data_dst = get_q_data(ctx, V4L2_BUF_TYPE_VIDEO_CAPTURE);
2347
2348        /* frame crop information */
2349        if (src_fourcc == V4L2_PIX_FMT_H264) {
2350                u32 left_right;
2351                u32 top_bottom;
2352
2353                left_right = coda_read(dev, CODA_RET_DEC_PIC_CROP_LEFT_RIGHT);
2354                top_bottom = coda_read(dev, CODA_RET_DEC_PIC_CROP_TOP_BOTTOM);
2355
2356                if (left_right == 0xffffffff && top_bottom == 0xffffffff) {
2357                        /* Keep current crop information */
2358                } else {
2359                        struct v4l2_rect *rect = &q_data_dst->rect;
2360
2361                        rect->left = left_right >> 16 & 0xffff;
2362                        rect->top = top_bottom >> 16 & 0xffff;
2363                        rect->width = width - rect->left -
2364                                      (left_right & 0xffff);
2365                        rect->height = height - rect->top -
2366                                       (top_bottom & 0xffff);
2367                }
2368        } else {
2369                /* no cropping */
2370        }
2371
2372        err_mb = coda_read(dev, CODA_RET_DEC_PIC_ERR_MB);
2373        if (err_mb > 0) {
2374                if (__ratelimit(&dev->mb_err_rs))
2375                        coda_dbg(1, ctx, "errors in %d macroblocks\n", err_mb);
2376                v4l2_ctrl_s_ctrl(ctx->mb_err_cnt_ctrl,
2377                                 v4l2_ctrl_g_ctrl(ctx->mb_err_cnt_ctrl) + err_mb);
2378        }
2379
2380        if (dev->devtype->product == CODA_HX4 ||
2381            dev->devtype->product == CODA_7541) {
2382                val = coda_read(dev, CODA_RET_DEC_PIC_OPTION);
2383                if (val == 0) {
2384                        /* not enough bitstream data */
2385                        coda_dbg(1, ctx, "prescan failed: %d\n", val);
2386                        ctx->hold = true;
2387                        return;
2388                }
2389        }
2390
2391        /* Wait until the VDOA finished writing the previous display frame */
2392        if (ctx->use_vdoa &&
2393            ctx->display_idx >= 0 &&
2394            ctx->display_idx < ctx->num_internal_frames) {
2395                err_vdoa = vdoa_wait_for_completion(ctx->vdoa);
2396        }
2397
2398        ctx->frm_dis_flg = coda_read(dev,
2399                                     CODA_REG_BIT_FRM_DIS_FLG(ctx->reg_idx));
2400
2401        /* The previous display frame was copied out and can be overwritten */
2402        if (ctx->display_idx >= 0 &&
2403            ctx->display_idx < ctx->num_internal_frames) {
2404                ctx->frm_dis_flg &= ~(1 << ctx->display_idx);
2405                coda_write(dev, ctx->frm_dis_flg,
2406                                CODA_REG_BIT_FRM_DIS_FLG(ctx->reg_idx));
2407        }
2408
2409        /*
2410         * The index of the last decoded frame, not necessarily in
2411         * display order, and the index of the next display frame.
2412         * The latter could have been decoded in a previous run.
2413         */
2414        decoded_idx = coda_read(dev, CODA_RET_DEC_PIC_CUR_IDX);
2415        display_idx = coda_read(dev, CODA_RET_DEC_PIC_FRAME_IDX);
2416
2417        if (decoded_idx == -1) {
2418                /* no frame was decoded, but we might have a display frame */
2419                if (display_idx >= 0 && display_idx < ctx->num_internal_frames)
2420                        ctx->sequence_offset++;
2421                else if (ctx->display_idx < 0)
2422                        ctx->hold = true;
2423        } else if (decoded_idx == -2) {
2424                if (ctx->display_idx >= 0 &&
2425                    ctx->display_idx < ctx->num_internal_frames)
2426                        ctx->sequence_offset++;
2427                /* no frame was decoded, we still return remaining buffers */
2428        } else if (decoded_idx < 0 || decoded_idx >= ctx->num_internal_frames) {
2429                v4l2_err(&dev->v4l2_dev,
2430                         "decoded frame index out of range: %d\n", decoded_idx);
2431        } else {
2432                int sequence;
2433
2434                decoded_frame = &ctx->internal_frames[decoded_idx];
2435
2436                val = coda_read(dev, CODA_RET_DEC_PIC_FRAME_NUM);
2437                if (ctx->sequence_offset == -1)
2438                        ctx->sequence_offset = val;
2439
2440                sequence = val + ctx->first_frame_sequence
2441                               - ctx->sequence_offset;
2442                spin_lock(&ctx->buffer_meta_lock);
2443                if (!list_empty(&ctx->buffer_meta_list)) {
2444                        meta = list_first_entry(&ctx->buffer_meta_list,
2445                                              struct coda_buffer_meta, list);
2446                        list_del(&meta->list);
2447                        ctx->num_metas--;
2448                        spin_unlock(&ctx->buffer_meta_lock);
2449                        /*
2450                         * Clamp counters to 16 bits for comparison, as the HW
2451                         * counter rolls over at this point for h.264. This
2452                         * may be different for other formats, but using 16 bits
2453                         * should be enough to detect most errors and saves us
2454                         * from doing different things based on the format.
2455                         */
2456                        if ((sequence & 0xffff) != (meta->sequence & 0xffff)) {
2457                                v4l2_err(&dev->v4l2_dev,
2458                                         "sequence number mismatch (%d(%d) != %d)\n",
2459                                         sequence, ctx->sequence_offset,
2460                                         meta->sequence);
2461                        }
2462                        decoded_frame->meta = *meta;
2463                        kfree(meta);
2464                } else {
2465                        spin_unlock(&ctx->buffer_meta_lock);
2466                        v4l2_err(&dev->v4l2_dev, "empty timestamp list!\n");
2467                        memset(&decoded_frame->meta, 0,
2468                               sizeof(struct coda_buffer_meta));
2469                        decoded_frame->meta.sequence = sequence;
2470                        decoded_frame->meta.last = false;
2471                        ctx->sequence_offset++;
2472                }
2473
2474                trace_coda_dec_pic_done(ctx, &decoded_frame->meta);
2475
2476                val = coda_read(dev, CODA_RET_DEC_PIC_TYPE) & 0x7;
2477                decoded_frame->type = (val == 0) ? V4L2_BUF_FLAG_KEYFRAME :
2478                                      (val == 1) ? V4L2_BUF_FLAG_PFRAME :
2479                                                   V4L2_BUF_FLAG_BFRAME;
2480
2481                decoded_frame->error = err_mb;
2482        }
2483
2484        if (display_idx == -1) {
2485                /*
2486                 * no more frames to be decoded, but there could still
2487                 * be rotator output to dequeue
2488                 */
2489                ctx->hold = true;
2490        } else if (display_idx == -3) {
2491                /* possibly prescan failure */
2492        } else if (display_idx < 0 || display_idx >= ctx->num_internal_frames) {
2493                v4l2_err(&dev->v4l2_dev,
2494                         "presentation frame index out of range: %d\n",
2495                         display_idx);
2496        }
2497
2498        /* If a frame was copied out, return it */
2499        if (ctx->display_idx >= 0 &&
2500            ctx->display_idx < ctx->num_internal_frames) {
2501                struct coda_internal_frame *ready_frame;
2502
2503                ready_frame = &ctx->internal_frames[ctx->display_idx];
2504
2505                dst_buf = v4l2_m2m_dst_buf_remove(ctx->fh.m2m_ctx);
2506                dst_buf->sequence = ctx->osequence++;
2507
2508                dst_buf->field = V4L2_FIELD_NONE;
2509                dst_buf->flags &= ~(V4L2_BUF_FLAG_KEYFRAME |
2510                                             V4L2_BUF_FLAG_PFRAME |
2511                                             V4L2_BUF_FLAG_BFRAME);
2512                dst_buf->flags |= ready_frame->type;
2513                meta = &ready_frame->meta;
2514                if (meta->last && !coda_reorder_enable(ctx)) {
2515                        /*
2516                         * If this was the last decoded frame, and reordering
2517                         * is disabled, this will be the last display frame.
2518                         */
2519                        coda_dbg(1, ctx, "last meta, marking as last frame\n");
2520                        dst_buf->flags |= V4L2_BUF_FLAG_LAST;
2521                } else if (ctx->bit_stream_param & CODA_BIT_STREAM_END_FLAG &&
2522                           display_idx == -1) {
2523                        /*
2524                         * If there is no designated presentation frame anymore,
2525                         * this frame has to be the last one.
2526                         */
2527                        coda_dbg(1, ctx,
2528                                 "no more frames to return, marking as last frame\n");
2529                        dst_buf->flags |= V4L2_BUF_FLAG_LAST;
2530                }
2531                dst_buf->timecode = meta->timecode;
2532                dst_buf->vb2_buf.timestamp = meta->timestamp;
2533
2534                trace_coda_dec_rot_done(ctx, dst_buf, meta);
2535
2536                vb2_set_plane_payload(&dst_buf->vb2_buf, 0,
2537                                      q_data_dst->sizeimage);
2538
2539                if (ready_frame->error || err_vdoa)
2540                        coda_m2m_buf_done(ctx, dst_buf, VB2_BUF_STATE_ERROR);
2541                else
2542                        coda_m2m_buf_done(ctx, dst_buf, VB2_BUF_STATE_DONE);
2543
2544                if (decoded_frame) {
2545                        coda_dbg(1, ctx, "job finished: decoded %c frame %u, returned %c frame %u (%u/%u)%s\n",
2546                                 coda_frame_type_char(decoded_frame->type),
2547                                 decoded_frame->meta.sequence,
2548                                 coda_frame_type_char(dst_buf->flags),
2549                                 ready_frame->meta.sequence,
2550                                 dst_buf->sequence, ctx->qsequence,
2551                                 (dst_buf->flags & V4L2_BUF_FLAG_LAST) ?
2552                                 " (last)" : "");
2553                } else {
2554                        coda_dbg(1, ctx, "job finished: no frame decoded (%d), returned %c frame %u (%u/%u)%s\n",
2555                                 decoded_idx,
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                }
2562        } else {
2563                if (decoded_frame) {
2564                        coda_dbg(1, ctx, "job finished: decoded %c frame %u, no frame returned (%d)\n",
2565                                 coda_frame_type_char(decoded_frame->type),
2566                                 decoded_frame->meta.sequence,
2567                                 ctx->display_idx);
2568                } else {
2569                        coda_dbg(1, ctx, "job finished: no frame decoded (%d) or returned (%d)\n",
2570                                 decoded_idx, ctx->display_idx);
2571                }
2572        }
2573
2574        /* The rotator will copy the current display frame next time */
2575        ctx->display_idx = display_idx;
2576
2577        /*
2578         * The current decode run might have brought the bitstream fill level
2579         * below the size where we can start the next decode run. As userspace
2580         * might have filled the output queue completely and might thus be
2581         * blocked, we can't rely on the next qbuf to trigger the bitstream
2582         * refill. Check if we have data to refill the bitstream now.
2583         */
2584        mutex_lock(&ctx->bitstream_mutex);
2585        coda_fill_bitstream(ctx, NULL);
2586        mutex_unlock(&ctx->bitstream_mutex);
2587}
2588
2589static void coda_decode_timeout(struct coda_ctx *ctx)
2590{
2591        struct vb2_v4l2_buffer *dst_buf;
2592
2593        /*
2594         * For now this only handles the case where we would deadlock with
2595         * userspace, i.e. userspace issued DEC_CMD_STOP and waits for EOS,
2596         * but after a failed decode run we would hold the context and wait for
2597         * userspace to queue more buffers.
2598         */
2599        if (!(ctx->bit_stream_param & CODA_BIT_STREAM_END_FLAG))
2600                return;
2601
2602        dst_buf = v4l2_m2m_dst_buf_remove(ctx->fh.m2m_ctx);
2603        dst_buf->sequence = ctx->qsequence - 1;
2604
2605        coda_m2m_buf_done(ctx, dst_buf, VB2_BUF_STATE_ERROR);
2606}
2607
2608const struct coda_context_ops coda_bit_decode_ops = {
2609        .queue_init = coda_decoder_queue_init,
2610        .reqbufs = coda_decoder_reqbufs,
2611        .start_streaming = coda_start_decoding,
2612        .prepare_run = coda_prepare_decode,
2613        .finish_run = coda_finish_decode,
2614        .run_timeout = coda_decode_timeout,
2615        .seq_init_work = coda_dec_seq_init_work,
2616        .seq_end_work = coda_seq_end_work,
2617        .release = coda_bit_release,
2618};
2619
2620irqreturn_t coda_irq_handler(int irq, void *data)
2621{
2622        struct coda_dev *dev = data;
2623        struct coda_ctx *ctx;
2624
2625        /* read status register to attend the IRQ */
2626        coda_read(dev, CODA_REG_BIT_INT_STATUS);
2627        coda_write(dev, 0, CODA_REG_BIT_INT_REASON);
2628        coda_write(dev, CODA_REG_BIT_INT_CLEAR_SET,
2629                      CODA_REG_BIT_INT_CLEAR);
2630
2631        ctx = v4l2_m2m_get_curr_priv(dev->m2m_dev);
2632        if (ctx == NULL) {
2633                v4l2_err(&dev->v4l2_dev,
2634                         "Instance released before the end of transaction\n");
2635                return IRQ_HANDLED;
2636        }
2637
2638        trace_coda_bit_done(ctx);
2639
2640        if (ctx->aborting) {
2641                coda_dbg(1, ctx, "task has been aborted\n");
2642        }
2643
2644        if (coda_isbusy(ctx->dev)) {
2645                coda_dbg(1, ctx, "coda is still busy!!!!\n");
2646                return IRQ_NONE;
2647        }
2648
2649        complete(&ctx->completion);
2650
2651        return IRQ_HANDLED;
2652}
2653