drivers/media/platform/mtk-vcodec/venc/venc_vp8_if.c

   1 /*
   2  * Copyright (c) 2016 MediaTek Inc.
   3  * Author: Daniel Hsiao <daniel.hsiao@mediatek.com>
   4  *         PoChun Lin <pochun.lin@mediatek.com>
   5  *
   6  * This program is free software; you can redistribute it and/or
   7  * modify
   8  * it under the terms of the GNU General Public License version 2 as
   9  * published by the Free Software Foundation.
  10  *
  11  * This program is distributed in the hope that it will be useful,
  12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  14  * GNU General Public License for more details.
  15  */
  16
  17 #include <linux/interrupt.h>
  18 #include <linux/kernel.h>
  19 #include <linux/slab.h>
  20
  21 #include "../mtk_vcodec_drv.h"
  22 #include "../mtk_vcodec_util.h"
  23 #include "../mtk_vcodec_intr.h"
  24 #include "../mtk_vcodec_enc.h"
  25 #include "../mtk_vcodec_enc_pm.h"
  26 #include "../venc_drv_base.h"
  27 #include "../venc_ipi_msg.h"
  28 #include "../venc_vpu_if.h"
  29 #include "mtk_vpu.h"
  30
  31 #define VENC_BITSTREAM_FRAME_SIZE 0x0098
  32 #define VENC_BITSTREAM_HEADER_LEN 0x00e8
  33
  34 /* This ac_tag is vp8 frame tag. */
  35 #define MAX_AC_TAG_SIZE 10
  36
  37 /**
  38  * enum venc_vp8_vpu_work_buf - vp8 encoder buffer index
  39  */
  40 enum venc_vp8_vpu_work_buf {
  41         VENC_VP8_VPU_WORK_BUF_LUMA,
  42         VENC_VP8_VPU_WORK_BUF_LUMA2,
  43         VENC_VP8_VPU_WORK_BUF_LUMA3,
  44         VENC_VP8_VPU_WORK_BUF_CHROMA,
  45         VENC_VP8_VPU_WORK_BUF_CHROMA2,
  46         VENC_VP8_VPU_WORK_BUF_CHROMA3,
  47         VENC_VP8_VPU_WORK_BUF_MV_INFO,
  48         VENC_VP8_VPU_WORK_BUF_BS_HEADER,
  49         VENC_VP8_VPU_WORK_BUF_PROB_BUF,
  50         VENC_VP8_VPU_WORK_BUF_RC_INFO,
  51         VENC_VP8_VPU_WORK_BUF_RC_CODE,
  52         VENC_VP8_VPU_WORK_BUF_RC_CODE2,
  53         VENC_VP8_VPU_WORK_BUF_RC_CODE3,
  54         VENC_VP8_VPU_WORK_BUF_MAX,
  55 };
  56
  57 /*
  58  * struct venc_vp8_vpu_config - Structure for vp8 encoder configuration
  59  * @input_fourcc: input fourcc
  60  * @bitrate: target bitrate (in bps)
  61  * @pic_w: picture width. Picture size is visible stream resolution, in pixels,
  62  *         to be used for display purposes; must be smaller or equal to buffer
  63  *         size.
  64  * @pic_h: picture height
  65  * @buf_w: buffer width (with 16 alignment). Buffer size is stream resolution
  66  *         in pixels aligned to hardware requirements.
  67  * @buf_h: buffer height (with 16 alignment)
  68  * @gop_size: group of picture size (key frame)
  69  * @framerate: frame rate in fps
  70  * @ts_mode: temporal scalability mode (0: disable, 1: enable)
  71  *           support three temporal layers - 0: 7.5fps 1: 7.5fps 2: 15fps.
  72  */
  73 struct venc_vp8_vpu_config {
  74         u32 input_fourcc;
  75         u32 bitrate;
  76         u32 pic_w;
  77         u32 pic_h;
  78         u32 buf_w;
  79         u32 buf_h;
  80         u32 gop_size;
  81         u32 framerate;
  82         u32 ts_mode;
  83 };
  84
  85 /*
  86  * struct venc_vp8_vpu_buf -Structure for buffer information
  87  * @align: buffer alignment (in bytes)
  88  * @iova: IO virtual address
  89  * @vpua: VPU side memory addr which is used by RC_CODE
  90  * @size: buffer size (in bytes)
  91  */
  92 struct venc_vp8_vpu_buf {
  93         u32 align;
  94         u32 iova;
  95         u32 vpua;
  96         u32 size;
  97 };
  98
  99 /*
 100  * struct venc_vp8_vsi - Structure for VPU driver control and info share
 101  * This structure is allocated in VPU side and shared to AP side.
 102  * @config: vp8 encoder configuration
 103  * @work_bufs: working buffer information in VPU side
 104  * The work_bufs here is for storing the 'size' info shared to AP side.
 105  * The similar item in struct venc_vp8_inst is for memory allocation
 106  * in AP side. The AP driver will copy the 'size' from here to the one in
 107  * struct mtk_vcodec_mem, then invoke mtk_vcodec_mem_alloc to allocate
 108  * the buffer. After that, bypass the 'dma_addr' to the 'iova' field here for
 109  * register setting in VPU side.
 110  */
 111 struct venc_vp8_vsi {
 112         struct venc_vp8_vpu_config config;
 113         struct venc_vp8_vpu_buf work_bufs[VENC_VP8_VPU_WORK_BUF_MAX];
 114 };
 115
 116 /*
 117  * struct venc_vp8_inst - vp8 encoder AP driver instance
 118  * @hw_base: vp8 encoder hardware register base
 119  * @work_bufs: working buffer
 120  * @work_buf_allocated: working buffer allocated flag
 121  * @frm_cnt: encoded frame count, it's used for I-frame judgement and
 122  *           reset when force intra cmd received.
 123  * @ts_mode: temporal scalability mode (0: disable, 1: enable)
 124  *           support three temporal layers - 0: 7.5fps 1: 7.5fps 2: 15fps.
 125  * @vpu_inst: VPU instance to exchange information between AP and VPU
 126  * @vsi: driver structure allocated by VPU side and shared to AP side for
 127  *       control and info share
 128  * @ctx: context for v4l2 layer integration
 129  */
 130 struct venc_vp8_inst {
 131         void __iomem *hw_base;
 132         struct mtk_vcodec_mem work_bufs[VENC_VP8_VPU_WORK_BUF_MAX];
 133         bool work_buf_allocated;
 134         unsigned int frm_cnt;
 135         unsigned int ts_mode;
 136         struct venc_vpu_inst vpu_inst;
 137         struct venc_vp8_vsi *vsi;
 138         struct mtk_vcodec_ctx *ctx;
 139 };
 140
 141 static inline void vp8_enc_write_reg(struct venc_vp8_inst *inst, u32 addr,
 142                                      u32 val)
 143 {
 144         writel(val, inst->hw_base + addr);
 145 }
 146
 147 static inline u32 vp8_enc_read_reg(struct venc_vp8_inst *inst, u32 addr)
 148 {
 149         return readl(inst->hw_base + addr);
 150 }
 151
 152 static void vp8_enc_free_work_buf(struct venc_vp8_inst *inst)
 153 {
 154         int i;
 155
 156         mtk_vcodec_debug_enter(inst);
 157
 158         /* Buffers need to be freed by AP. */
 159         for (i = 0; i < VENC_VP8_VPU_WORK_BUF_MAX; i++) {
 160                 if ((inst->work_bufs[i].size == 0))
 161                         continue;
 162                 mtk_vcodec_mem_free(inst->ctx, &inst->work_bufs[i]);
 163         }
 164
 165         mtk_vcodec_debug_leave(inst);
 166 }
 167
 168 static int vp8_enc_alloc_work_buf(struct venc_vp8_inst *inst)
 169 {
 170         int i;
 171         int ret = 0;
 172         struct venc_vp8_vpu_buf *wb = inst->vsi->work_bufs;
 173
 174         mtk_vcodec_debug_enter(inst);
 175
 176         for (i = 0; i < VENC_VP8_VPU_WORK_BUF_MAX; i++) {
 177                 if ((wb[i].size == 0))
 178                         continue;
 179                 /*
 180                  * This 'wb' structure is set by VPU side and shared to AP for
 181                  * buffer allocation and IO virtual addr mapping. For most of
 182                  * the buffers, AP will allocate the buffer according to 'size'
 183                  * field and store the IO virtual addr in 'iova' field. For the
 184                  * RC_CODEx buffers, they are pre-allocated in the VPU side
 185                  * because they are inside VPU SRAM, and save the VPU addr in
 186                  * the 'vpua' field. The AP will translate the VPU addr to the
 187                  * corresponding IO virtual addr and store in 'iova' field.
 188                  */
 189                 inst->work_bufs[i].size = wb[i].size;
 190                 ret = mtk_vcodec_mem_alloc(inst->ctx, &inst->work_bufs[i]);
 191                 if (ret) {
 192                         mtk_vcodec_err(inst,
 193                                        "cannot alloc work_bufs[%d]", i);
 194                         goto err_alloc;
 195                 }
 196                 /*
 197                  * This RC_CODEx is pre-allocated by VPU and saved in VPU addr.
 198                  * So we need use memcpy to copy RC_CODEx from VPU addr into IO
 199                  * virtual addr in 'iova' field for reg setting in VPU side.
 200                  */
 201                 if (i == VENC_VP8_VPU_WORK_BUF_RC_CODE ||
 202                     i == VENC_VP8_VPU_WORK_BUF_RC_CODE2 ||
 203                     i == VENC_VP8_VPU_WORK_BUF_RC_CODE3) {
 204                         void *tmp_va;
 205
 206                         tmp_va = vpu_mapping_dm_addr(inst->vpu_inst.dev,
 207                                                      wb[i].vpua);
 208                         memcpy(inst->work_bufs[i].va, tmp_va, wb[i].size);
 209                 }
 210                 wb[i].iova = inst->work_bufs[i].dma_addr;
 211
 212                 mtk_vcodec_debug(inst,
 213                                  "work_bufs[%d] va=0x%p,iova=%pad,size=%zu",
 214                                  i, inst->work_bufs[i].va,
 215                                  &inst->work_bufs[i].dma_addr,
 216                                  inst->work_bufs[i].size);
 217         }
 218
 219         mtk_vcodec_debug_leave(inst);
 220
 221         return ret;
 222
 223 err_alloc:
 224         vp8_enc_free_work_buf(inst);
 225
 226         return ret;
 227 }
 228
 229 static unsigned int vp8_enc_wait_venc_done(struct venc_vp8_inst *inst)
 230 {
 231         unsigned int irq_status = 0;
 232         struct mtk_vcodec_ctx *ctx = (struct mtk_vcodec_ctx *)inst->ctx;
 233
 234         if (!mtk_vcodec_wait_for_done_ctx(ctx, MTK_INST_IRQ_RECEIVED,
 235                                           WAIT_INTR_TIMEOUT_MS)) {
 236                 irq_status = ctx->irq_status;
 237                 mtk_vcodec_debug(inst, "isr return %x", irq_status);
 238         }
 239         return irq_status;
 240 }
 241
 242 /*
 243  * Compose ac_tag, bitstream header and bitstream payload into
 244  * one bitstream buffer.
 245  */
 246 static int vp8_enc_compose_one_frame(struct venc_vp8_inst *inst,
 247                                      struct mtk_vcodec_mem *bs_buf,
 248                                      unsigned int *bs_size)
 249 {
 250         unsigned int not_key;
 251         u32 bs_frm_size;
 252         u32 bs_hdr_len;
 253         unsigned int ac_tag_size;
 254         u8 ac_tag[MAX_AC_TAG_SIZE];
 255         u32 tag;
 256
 257         bs_frm_size = vp8_enc_read_reg(inst, VENC_BITSTREAM_FRAME_SIZE);
 258         bs_hdr_len = vp8_enc_read_reg(inst, VENC_BITSTREAM_HEADER_LEN);
 259
 260         /* if a frame is key frame, not_key is 0 */
 261         not_key = !inst->vpu_inst.is_key_frm;
 262         tag = (bs_hdr_len << 5) | 0x10 | not_key;
 263         ac_tag[0] = tag & 0xff;
 264         ac_tag[1] = (tag >> 8) & 0xff;
 265         ac_tag[2] = (tag >> 16) & 0xff;
 266
 267         /* key frame */
 268         if (not_key == 0) {
 269                 ac_tag_size = MAX_AC_TAG_SIZE;
 270                 ac_tag[3] = 0x9d;
 271                 ac_tag[4] = 0x01;
 272                 ac_tag[5] = 0x2a;
 273                 ac_tag[6] = inst->vsi->config.pic_w;
 274                 ac_tag[7] = inst->vsi->config.pic_w >> 8;
 275                 ac_tag[8] = inst->vsi->config.pic_h;
 276                 ac_tag[9] = inst->vsi->config.pic_h >> 8;
 277         } else {
 278                 ac_tag_size = 3;
 279         }
 280
 281         if (bs_buf->size < bs_hdr_len + bs_frm_size + ac_tag_size) {
 282                 mtk_vcodec_err(inst, "bitstream buf size is too small(%zu)",
 283                                bs_buf->size);
 284                 return -EINVAL;
 285         }
 286
 287         /*
 288         * (1) The vp8 bitstream header and body are generated by the HW vp8
 289         * encoder separately at the same time. We cannot know the bitstream
 290         * header length in advance.
 291         * (2) From the vp8 spec, there is no stuffing byte allowed between the
 292         * ac tag, bitstream header and bitstream body.
 293         */
 294         memmove(bs_buf->va + bs_hdr_len + ac_tag_size,
 295                 bs_buf->va, bs_frm_size);
 296         memcpy(bs_buf->va + ac_tag_size,
 297                inst->work_bufs[VENC_VP8_VPU_WORK_BUF_BS_HEADER].va,
 298                bs_hdr_len);
 299         memcpy(bs_buf->va, ac_tag, ac_tag_size);
 300         *bs_size = bs_frm_size + bs_hdr_len + ac_tag_size;
 301
 302         return 0;
 303 }
 304
 305 static int vp8_enc_encode_frame(struct venc_vp8_inst *inst,
 306                                 struct venc_frm_buf *frm_buf,
 307                                 struct mtk_vcodec_mem *bs_buf,
 308                                 unsigned int *bs_size)
 309 {
 310         int ret = 0;
 311         unsigned int irq_status;
 312
 313         mtk_vcodec_debug(inst, "->frm_cnt=%d", inst->frm_cnt);
 314
 315         ret = vpu_enc_encode(&inst->vpu_inst, 0, frm_buf, bs_buf, bs_size);
 316         if (ret)
 317                 return ret;
 318
 319         irq_status = vp8_enc_wait_venc_done(inst);
 320         if (irq_status != MTK_VENC_IRQ_STATUS_FRM) {
 321                 mtk_vcodec_err(inst, "irq_status=%d failed", irq_status);
 322                 return -EIO;
 323         }
 324
 325         if (vp8_enc_compose_one_frame(inst, bs_buf, bs_size)) {
 326                 mtk_vcodec_err(inst, "vp8_enc_compose_one_frame failed");
 327                 return -EINVAL;
 328         }
 329
 330         inst->frm_cnt++;
 331         mtk_vcodec_debug(inst, "<-size=%d key_frm=%d", *bs_size,
 332                          inst->vpu_inst.is_key_frm);
 333
 334         return ret;
 335 }
 336
 337 static int vp8_enc_init(struct mtk_vcodec_ctx *ctx, unsigned long *handle)
 338 {
 339         int ret = 0;
 340         struct venc_vp8_inst *inst;
 341
 342         inst = kzalloc(sizeof(*inst), GFP_KERNEL);
 343         if (!inst)
 344                 return -ENOMEM;
 345
 346         inst->ctx = ctx;
 347         inst->vpu_inst.ctx = ctx;
 348         inst->vpu_inst.dev = ctx->dev->vpu_plat_dev;
 349         inst->vpu_inst.id = IPI_VENC_VP8;
 350         inst->hw_base = mtk_vcodec_get_reg_addr(inst->ctx, VENC_LT_SYS);
 351
 352         mtk_vcodec_debug_enter(inst);
 353
 354         ret = vpu_enc_init(&inst->vpu_inst);
 355
 356         inst->vsi = (struct venc_vp8_vsi *)inst->vpu_inst.vsi;
 357
 358         mtk_vcodec_debug_leave(inst);
 359
 360         if (ret)
 361                 kfree(inst);
 362         else
 363                 (*handle) = (unsigned long)inst;
 364
 365         return ret;
 366 }
 367
 368 static int vp8_enc_encode(unsigned long handle,
 369                           enum venc_start_opt opt,
 370                           struct venc_frm_buf *frm_buf,
 371                           struct mtk_vcodec_mem *bs_buf,
 372                           struct venc_done_result *result)
 373 {
 374         int ret = 0;
 375         struct venc_vp8_inst *inst = (struct venc_vp8_inst *)handle;
 376         struct mtk_vcodec_ctx *ctx = inst->ctx;
 377
 378         mtk_vcodec_debug_enter(inst);
 379
 380         enable_irq(ctx->dev->enc_lt_irq);
 381
 382         switch (opt) {
 383         case VENC_START_OPT_ENCODE_FRAME:
 384                 ret = vp8_enc_encode_frame(inst, frm_buf, bs_buf,
 385                                            &result->bs_size);
 386                 if (ret)
 387                         goto encode_err;
 388                 result->is_key_frm = inst->vpu_inst.is_key_frm;
 389                 break;
 390
 391         default:
 392                 mtk_vcodec_err(inst, "opt not support:%d", opt);
 393                 ret = -EINVAL;
 394                 break;
 395         }
 396
 397 encode_err:
 398
 399         disable_irq(ctx->dev->enc_lt_irq);
 400         mtk_vcodec_debug_leave(inst);
 401
 402         return ret;
 403 }
 404
 405 static int vp8_enc_set_param(unsigned long handle,
 406                              enum venc_set_param_type type,
 407                              struct venc_enc_param *enc_prm)
 408 {
 409         int ret = 0;
 410         struct venc_vp8_inst *inst = (struct venc_vp8_inst *)handle;
 411
 412         mtk_vcodec_debug(inst, "->type=%d", type);
 413
 414         switch (type) {
 415         case VENC_SET_PARAM_ENC:
 416                 inst->vsi->config.input_fourcc = enc_prm->input_yuv_fmt;
 417                 inst->vsi->config.bitrate = enc_prm->bitrate;
 418                 inst->vsi->config.pic_w = enc_prm->width;
 419                 inst->vsi->config.pic_h = enc_prm->height;
 420                 inst->vsi->config.buf_w = enc_prm->buf_width;
 421                 inst->vsi->config.buf_h = enc_prm->buf_height;
 422                 inst->vsi->config.gop_size = enc_prm->gop_size;
 423                 inst->vsi->config.framerate = enc_prm->frm_rate;
 424                 inst->vsi->config.ts_mode = inst->ts_mode;
 425                 ret = vpu_enc_set_param(&inst->vpu_inst, type, enc_prm);
 426                 if (ret)
 427                         break;
 428                 if (inst->work_buf_allocated) {
 429                         vp8_enc_free_work_buf(inst);
 430                         inst->work_buf_allocated = false;
 431                 }
 432                 ret = vp8_enc_alloc_work_buf(inst);
 433                 if (ret)
 434                         break;
 435                 inst->work_buf_allocated = true;
 436                 break;
 437
 438         /*
 439          * VENC_SET_PARAM_TS_MODE must be called before VENC_SET_PARAM_ENC
 440          */
 441         case VENC_SET_PARAM_TS_MODE:
 442                 inst->ts_mode = 1;
 443                 mtk_vcodec_debug(inst, "set ts_mode");
 444                 break;
 445
 446         default:
 447                 ret = vpu_enc_set_param(&inst->vpu_inst, type, enc_prm);
 448                 break;
 449         }
 450
 451         mtk_vcodec_debug_leave(inst);
 452
 453         return ret;
 454 }
 455
 456 static int vp8_enc_deinit(unsigned long handle)
 457 {
 458         int ret = 0;
 459         struct venc_vp8_inst *inst = (struct venc_vp8_inst *)handle;
 460
 461         mtk_vcodec_debug_enter(inst);
 462
 463         ret = vpu_enc_deinit(&inst->vpu_inst);
 464
 465         if (inst->work_buf_allocated)
 466                 vp8_enc_free_work_buf(inst);
 467
 468         mtk_vcodec_debug_leave(inst);
 469         kfree(inst);
 470
 471         return ret;
 472 }
 473
 474 static struct venc_common_if venc_vp8_if = {
 475         vp8_enc_init,
 476         vp8_enc_encode,
 477         vp8_enc_set_param,
 478         vp8_enc_deinit,
 479 };
 480
 481 struct venc_common_if *get_vp8_enc_comm_if(void);
 482
 483 struct venc_common_if *get_vp8_enc_comm_if(void)
 484 {
 485         return &venc_vp8_if;
 486 }