1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2016 MediaTek Inc.
4 * Author: PC Chen <pc.chen@mediatek.com>
7 #include <linux/module.h>
8 #include <linux/slab.h>
10 #include "../vdec_drv_if.h"
11 #include "../mtk_vcodec_util.h"
12 #include "../mtk_vcodec_dec.h"
13 #include "../mtk_vcodec_intr.h"
14 #include "../vdec_vpu_if.h"
15 #include "../vdec_drv_base.h"
17 #define NAL_NON_IDR_SLICE 0x01
18 #define NAL_IDR_SLICE 0x05
19 #define NAL_H264_PPS 0x08
20 #define NAL_TYPE(value) ((value) & 0x1F)
22 #define BUF_PREDICTION_SZ (32 * 1024)
24 #define MB_UNIT_LEN 16
26 /* motion vector size (bytes) for every macro block */
27 #define HW_MB_STORE_SZ 64
29 #define H264_MAX_FB_NUM 17
30 #define HDR_PARSING_BUF_SZ 1024
33 * struct h264_fb - h264 decode frame buffer information
34 * @vdec_fb_va : virtual address of struct vdec_fb
35 * @y_fb_dma : dma address of Y frame buffer (luma)
36 * @c_fb_dma : dma address of C frame buffer (chroma)
37 * @poc : picture order count of frame buffer
38 * @reserved : for 8 bytes alignment
49 * struct h264_ring_fb_list - ring frame buffer list
50 * @fb_list : frame buffer array
51 * @read_idx : read index
52 * @write_idx : write index
53 * @count : buffer count in list
54 * @reserved : for 8 bytes alignment
56 struct h264_ring_fb_list {
57 struct h264_fb fb_list[H264_MAX_FB_NUM];
58 unsigned int read_idx;
59 unsigned int write_idx;
61 unsigned int reserved;
65 * struct vdec_h264_dec_info - decode information
66 * @dpb_sz : decoding picture buffer size
67 * @resolution_changed : resolution change happen
68 * @realloc_mv_buf : flag to notify driver to re-allocate mv buffer
69 * @reserved : for 8 bytes alignment
70 * @bs_dma : Input bit-stream buffer dma address
71 * @y_fb_dma : Y frame buffer dma address
72 * @c_fb_dma : C frame buffer dma address
73 * @vdec_fb_va : VDEC frame buffer struct virtual address
75 struct vdec_h264_dec_info {
77 uint32_t resolution_changed;
78 uint32_t realloc_mv_buf;
87 * struct vdec_h264_vsi - shared memory for decode information exchange
88 * between VPU and Host.
89 * The memory is allocated by VPU then mapping to Host
90 * in vpu_dec_init() and freed in vpu_dec_deinit()
92 * AP-W/R : AP is writer/reader on this item
93 * VPU-W/R: VPU is write/reader on this item
94 * @hdr_buf : Header parsing buffer (AP-W, VPU-R)
95 * @pred_buf_dma : HW working predication buffer dma address (AP-W, VPU-R)
96 * @mv_buf_dma : HW working motion vector buffer dma address (AP-W, VPU-R)
97 * @list_free : free frame buffer ring list (AP-W/R, VPU-W)
98 * @list_disp : display frame buffer ring list (AP-R, VPU-W)
99 * @dec : decode information (AP-R, VPU-W)
100 * @pic : picture information (AP-R, VPU-W)
101 * @crop : crop information (AP-R, VPU-W)
103 struct vdec_h264_vsi {
104 unsigned char hdr_buf[HDR_PARSING_BUF_SZ];
105 uint64_t pred_buf_dma;
106 uint64_t mv_buf_dma[H264_MAX_FB_NUM];
107 struct h264_ring_fb_list list_free;
108 struct h264_ring_fb_list list_disp;
109 struct vdec_h264_dec_info dec;
110 struct vdec_pic_info pic;
111 struct v4l2_rect crop;
115 * struct vdec_h264_inst - h264 decoder instance
116 * @num_nalu : how many nalus be decoded
117 * @ctx : point to mtk_vcodec_ctx
118 * @pred_buf : HW working predication buffer
119 * @mv_buf : HW working motion vector buffer
120 * @vpu : VPU instance
121 * @vsi : VPU shared information
123 struct vdec_h264_inst {
124 unsigned int num_nalu;
125 struct mtk_vcodec_ctx *ctx;
126 struct mtk_vcodec_mem pred_buf;
127 struct mtk_vcodec_mem mv_buf[H264_MAX_FB_NUM];
128 struct vdec_vpu_inst vpu;
129 struct vdec_h264_vsi *vsi;
132 static unsigned int get_mv_buf_size(unsigned int width, unsigned int height)
134 return HW_MB_STORE_SZ * (width/MB_UNIT_LEN) * (height/MB_UNIT_LEN);
137 static int allocate_predication_buf(struct vdec_h264_inst *inst)
141 inst->pred_buf.size = BUF_PREDICTION_SZ;
142 err = mtk_vcodec_mem_alloc(inst->ctx, &inst->pred_buf);
144 mtk_vcodec_err(inst, "failed to allocate ppl buf");
148 inst->vsi->pred_buf_dma = inst->pred_buf.dma_addr;
152 static void free_predication_buf(struct vdec_h264_inst *inst)
154 struct mtk_vcodec_mem *mem = NULL;
156 mtk_vcodec_debug_enter(inst);
158 inst->vsi->pred_buf_dma = 0;
159 mem = &inst->pred_buf;
161 mtk_vcodec_mem_free(inst->ctx, mem);
164 static int alloc_mv_buf(struct vdec_h264_inst *inst, struct vdec_pic_info *pic)
168 struct mtk_vcodec_mem *mem = NULL;
169 unsigned int buf_sz = get_mv_buf_size(pic->buf_w, pic->buf_h);
171 for (i = 0; i < H264_MAX_FB_NUM; i++) {
172 mem = &inst->mv_buf[i];
174 mtk_vcodec_mem_free(inst->ctx, mem);
176 err = mtk_vcodec_mem_alloc(inst->ctx, mem);
178 mtk_vcodec_err(inst, "failed to allocate mv buf");
181 inst->vsi->mv_buf_dma[i] = mem->dma_addr;
187 static void free_mv_buf(struct vdec_h264_inst *inst)
190 struct mtk_vcodec_mem *mem = NULL;
192 for (i = 0; i < H264_MAX_FB_NUM; i++) {
193 inst->vsi->mv_buf_dma[i] = 0;
194 mem = &inst->mv_buf[i];
196 mtk_vcodec_mem_free(inst->ctx, mem);
200 static int check_list_validity(struct vdec_h264_inst *inst, bool disp_list)
202 struct h264_ring_fb_list *list;
204 list = disp_list ? &inst->vsi->list_disp : &inst->vsi->list_free;
206 if (list->count > H264_MAX_FB_NUM ||
207 list->read_idx >= H264_MAX_FB_NUM ||
208 list->write_idx >= H264_MAX_FB_NUM) {
209 mtk_vcodec_err(inst, "%s list err: cnt=%d r_idx=%d w_idx=%d",
210 disp_list ? "disp" : "free", list->count,
211 list->read_idx, list->write_idx);
218 static void put_fb_to_free(struct vdec_h264_inst *inst, struct vdec_fb *fb)
220 struct h264_ring_fb_list *list;
223 if (check_list_validity(inst, false))
226 list = &inst->vsi->list_free;
227 if (list->count == H264_MAX_FB_NUM) {
228 mtk_vcodec_err(inst, "[FB] put fb free_list full");
232 mtk_vcodec_debug(inst, "[FB] put fb into free_list @(%p, %llx)",
233 fb->base_y.va, (u64)fb->base_y.dma_addr);
235 list->fb_list[list->write_idx].vdec_fb_va = (u64)(uintptr_t)fb;
236 list->write_idx = (list->write_idx == H264_MAX_FB_NUM - 1) ?
237 0 : list->write_idx + 1;
242 static void get_pic_info(struct vdec_h264_inst *inst,
243 struct vdec_pic_info *pic)
245 *pic = inst->vsi->pic;
246 mtk_vcodec_debug(inst, "pic(%d, %d), buf(%d, %d)",
247 pic->pic_w, pic->pic_h, pic->buf_w, pic->buf_h);
248 mtk_vcodec_debug(inst, "fb size: Y(%d), C(%d)",
249 pic->fb_sz[0], pic->fb_sz[1]);
252 static void get_crop_info(struct vdec_h264_inst *inst, struct v4l2_rect *cr)
254 cr->left = inst->vsi->crop.left;
255 cr->top = inst->vsi->crop.top;
256 cr->width = inst->vsi->crop.width;
257 cr->height = inst->vsi->crop.height;
259 mtk_vcodec_debug(inst, "l=%d, t=%d, w=%d, h=%d",
260 cr->left, cr->top, cr->width, cr->height);
263 static void get_dpb_size(struct vdec_h264_inst *inst, unsigned int *dpb_sz)
265 *dpb_sz = inst->vsi->dec.dpb_sz;
266 mtk_vcodec_debug(inst, "sz=%d", *dpb_sz);
269 static int vdec_h264_init(struct mtk_vcodec_ctx *ctx)
271 struct vdec_h264_inst *inst = NULL;
274 inst = kzalloc(sizeof(*inst), GFP_KERNEL);
280 inst->vpu.id = IPI_VDEC_H264;
281 inst->vpu.dev = ctx->dev->vpu_plat_dev;
283 inst->vpu.handler = vpu_dec_ipi_handler;
285 err = vpu_dec_init(&inst->vpu);
287 mtk_vcodec_err(inst, "vdec_h264 init err=%d", err);
288 goto error_free_inst;
291 inst->vsi = (struct vdec_h264_vsi *)inst->vpu.vsi;
292 err = allocate_predication_buf(inst);
296 mtk_vcodec_debug(inst, "H264 Instance >> %p", inst);
298 ctx->drv_handle = inst;
302 vpu_dec_deinit(&inst->vpu);
309 static void vdec_h264_deinit(void *h_vdec)
311 struct vdec_h264_inst *inst = (struct vdec_h264_inst *)h_vdec;
313 mtk_vcodec_debug_enter(inst);
315 vpu_dec_deinit(&inst->vpu);
316 free_predication_buf(inst);
322 static int find_start_code(unsigned char *data, unsigned int data_sz)
324 if (data_sz > 3 && data[0] == 0 && data[1] == 0 && data[2] == 1)
327 if (data_sz > 4 && data[0] == 0 && data[1] == 0 && data[2] == 0 &&
334 static int vdec_h264_decode(void *h_vdec, struct mtk_vcodec_mem *bs,
335 struct vdec_fb *fb, bool *res_chg)
337 struct vdec_h264_inst *inst = (struct vdec_h264_inst *)h_vdec;
338 struct vdec_vpu_inst *vpu = &inst->vpu;
339 int nal_start_idx = 0;
341 unsigned int nal_start;
342 unsigned int nal_type;
345 unsigned int data[2];
346 uint64_t vdec_fb_va = (u64)(uintptr_t)fb;
347 uint64_t y_fb_dma = fb ? (u64)fb->base_y.dma_addr : 0;
348 uint64_t c_fb_dma = fb ? (u64)fb->base_c.dma_addr : 0;
350 mtk_vcodec_debug(inst, "+ [%d] FB y_dma=%llx c_dma=%llx va=%p",
351 ++inst->num_nalu, y_fb_dma, c_fb_dma, fb);
353 /* bs NULL means flush decoder */
355 return vpu_dec_reset(vpu);
357 buf = (unsigned char *)bs->va;
359 nal_start_idx = find_start_code(buf, buf_sz);
360 if (nal_start_idx < 0)
361 goto err_free_fb_out;
363 nal_start = buf[nal_start_idx];
364 nal_type = NAL_TYPE(buf[nal_start_idx]);
365 mtk_vcodec_debug(inst, "\n + NALU[%d] type %d +\n", inst->num_nalu,
368 if (nal_type == NAL_H264_PPS) {
369 buf_sz -= nal_start_idx;
370 if (buf_sz > HDR_PARSING_BUF_SZ) {
372 goto err_free_fb_out;
374 memcpy(inst->vsi->hdr_buf, buf + nal_start_idx, buf_sz);
377 inst->vsi->dec.bs_dma = (uint64_t)bs->dma_addr;
378 inst->vsi->dec.y_fb_dma = y_fb_dma;
379 inst->vsi->dec.c_fb_dma = c_fb_dma;
380 inst->vsi->dec.vdec_fb_va = vdec_fb_va;
384 err = vpu_dec_start(vpu, data, 2);
386 goto err_free_fb_out;
388 *res_chg = inst->vsi->dec.resolution_changed;
390 struct vdec_pic_info pic;
392 mtk_vcodec_debug(inst, "- resolution changed -");
393 get_pic_info(inst, &pic);
395 if (inst->vsi->dec.realloc_mv_buf) {
396 err = alloc_mv_buf(inst, &pic);
398 goto err_free_fb_out;
402 if (nal_type == NAL_NON_IDR_SLICE || nal_type == NAL_IDR_SLICE) {
403 /* wait decoder done interrupt */
404 err = mtk_vcodec_wait_for_done_ctx(inst->ctx,
405 MTK_INST_IRQ_RECEIVED,
406 WAIT_INTR_TIMEOUT_MS);
408 goto err_free_fb_out;
413 mtk_vcodec_debug(inst, "\n - NALU[%d] type=%d -\n", inst->num_nalu,
418 put_fb_to_free(inst, fb);
419 mtk_vcodec_err(inst, "\n - NALU[%d] err=%d -\n", inst->num_nalu, err);
423 static void vdec_h264_get_fb(struct vdec_h264_inst *inst,
424 struct h264_ring_fb_list *list,
425 bool disp_list, struct vdec_fb **out_fb)
429 if (check_list_validity(inst, disp_list))
432 if (list->count == 0) {
433 mtk_vcodec_debug(inst, "[FB] there is no %s fb",
434 disp_list ? "disp" : "free");
439 fb = (struct vdec_fb *)
440 (uintptr_t)list->fb_list[list->read_idx].vdec_fb_va;
441 fb->status |= (disp_list ? FB_ST_DISPLAY : FB_ST_FREE);
444 mtk_vcodec_debug(inst, "[FB] get %s fb st=%d poc=%d %llx",
445 disp_list ? "disp" : "free",
446 fb->status, list->fb_list[list->read_idx].poc,
447 list->fb_list[list->read_idx].vdec_fb_va);
449 list->read_idx = (list->read_idx == H264_MAX_FB_NUM - 1) ?
450 0 : list->read_idx + 1;
454 static int vdec_h264_get_param(void *h_vdec, enum vdec_get_param_type type,
457 struct vdec_h264_inst *inst = (struct vdec_h264_inst *)h_vdec;
460 case GET_PARAM_DISP_FRAME_BUFFER:
461 vdec_h264_get_fb(inst, &inst->vsi->list_disp, true, out);
464 case GET_PARAM_FREE_FRAME_BUFFER:
465 vdec_h264_get_fb(inst, &inst->vsi->list_free, false, out);
468 case GET_PARAM_PIC_INFO:
469 get_pic_info(inst, out);
472 case GET_PARAM_DPB_SIZE:
473 get_dpb_size(inst, out);
476 case GET_PARAM_CROP_INFO:
477 get_crop_info(inst, out);
481 mtk_vcodec_err(inst, "invalid get parameter type=%d", type);
488 const struct vdec_common_if vdec_h264_if = {
489 .init = vdec_h264_init,
490 .decode = vdec_h264_decode,
491 .get_param = vdec_h264_get_param,
492 .deinit = vdec_h264_deinit,