1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Copyright (c) 2016 MediaTek Inc.
4 * Author: PC Chen <pc.chen@mediatek.com>
5 */
6
7 #include <linux/module.h>
8 #include <linux/slab.h>
9
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"
16
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)
21
22 #define BUF_PREDICTION_SZ (32 * 1024)
23
24 #define MB_UNIT_LEN 16
25
26 /* motion vector size (bytes) for every macro block */
27 #define HW_MB_STORE_SZ 64
28
29 #define H264_MAX_FB_NUM 17
30 #define HDR_PARSING_BUF_SZ 1024
31
32 #define DEC_ERR_RET(ret) ((ret) >> 16)
33 #define H264_ERR_NOT_VALID 3
34
35 /**
36 * struct h264_fb - h264 decode frame buffer information
37 * @vdec_fb_va : virtual address of struct vdec_fb
38 * @y_fb_dma : dma address of Y frame buffer (luma)
39 * @c_fb_dma : dma address of C frame buffer (chroma)
40 * @poc : picture order count of frame buffer
41 * @reserved : for 8 bytes alignment
42 */
43 struct h264_fb {
44 uint64_t vdec_fb_va;
45 uint64_t y_fb_dma;
46 uint64_t c_fb_dma;
47 int32_t poc;
48 uint32_t reserved;
49 };
50
51 /**
52 * struct h264_ring_fb_list - ring frame buffer list
53 * @fb_list : frame buffer array
54 * @read_idx : read index
55 * @write_idx : write index
56 * @count : buffer count in list
57 * @reserved : for 8 bytes alignment
58 */
59 struct h264_ring_fb_list {
60 struct h264_fb fb_list[H264_MAX_FB_NUM];
61 unsigned int read_idx;
62 unsigned int write_idx;
63 unsigned int count;
64 unsigned int reserved;
65 };
66
67 /**
68 * struct vdec_h264_dec_info - decode information
69 * @dpb_sz : decoding picture buffer size
70 * @resolution_changed : resolution change happen
71 * @realloc_mv_buf : flag to notify driver to re-allocate mv buffer
72 * @reserved : for 8 bytes alignment
73 * @bs_dma : Input bit-stream buffer dma address
74 * @y_fb_dma : Y frame buffer dma address
75 * @c_fb_dma : C frame buffer dma address
76 * @vdec_fb_va : VDEC frame buffer struct virtual address
77 */
78 struct vdec_h264_dec_info {
79 uint32_t dpb_sz;
80 uint32_t resolution_changed;
81 uint32_t realloc_mv_buf;
82 uint32_t reserved;
83 uint64_t bs_dma;
84 uint64_t y_fb_dma;
85 uint64_t c_fb_dma;
86 uint64_t vdec_fb_va;
87 };
88
89 /**
90 * struct vdec_h264_vsi - shared memory for decode information exchange
91 * between VPU and Host.
92 * The memory is allocated by VPU then mapping to Host
93 * in vpu_dec_init() and freed in vpu_dec_deinit()
94 * by VPU.
95 * AP-W/R : AP is writer/reader on this item
96 * VPU-W/R: VPU is write/reader on this item
97 * @hdr_buf : Header parsing buffer (AP-W, VPU-R)
98 * @pred_buf_dma : HW working predication buffer dma address (AP-W, VPU-R)
99 * @mv_buf_dma : HW working motion vector buffer dma address (AP-W, VPU-R)
100 * @list_free : free frame buffer ring list (AP-W/R, VPU-W)
101 * @list_disp : display frame buffer ring list (AP-R, VPU-W)
102 * @dec : decode information (AP-R, VPU-W)
103 * @pic : picture information (AP-R, VPU-W)
104 * @crop : crop information (AP-R, VPU-W)
105 */
106 struct vdec_h264_vsi {
107 unsigned char hdr_buf[HDR_PARSING_BUF_SZ];
108 uint64_t pred_buf_dma;
109 uint64_t mv_buf_dma[H264_MAX_FB_NUM];
110 struct h264_ring_fb_list list_free;
111 struct h264_ring_fb_list list_disp;
112 struct vdec_h264_dec_info dec;
113 struct vdec_pic_info pic;
114 struct v4l2_rect crop;
115 };
116
117 /**
118 * struct vdec_h264_inst - h264 decoder instance
119 * @num_nalu : how many nalus be decoded
120 * @ctx : point to mtk_vcodec_ctx
121 * @pred_buf : HW working predication buffer
122 * @mv_buf : HW working motion vector buffer
123 * @vpu : VPU instance
124 * @vsi : VPU shared information
125 */
126 struct vdec_h264_inst {
127 unsigned int num_nalu;
128 struct mtk_vcodec_ctx *ctx;
129 struct mtk_vcodec_mem pred_buf;
130 struct mtk_vcodec_mem mv_buf[H264_MAX_FB_NUM];
131 struct vdec_vpu_inst vpu;
132 struct vdec_h264_vsi *vsi;
133 };
134
get_mv_buf_size(unsigned int width,unsigned int height)135 static unsigned int get_mv_buf_size(unsigned int width, unsigned int height)
136 {
137 return HW_MB_STORE_SZ * (width/MB_UNIT_LEN) * (height/MB_UNIT_LEN);
138 }
139
allocate_predication_buf(struct vdec_h264_inst * inst)140 static int allocate_predication_buf(struct vdec_h264_inst *inst)
141 {
142 int err = 0;
143
144 inst->pred_buf.size = BUF_PREDICTION_SZ;
145 err = mtk_vcodec_mem_alloc(inst->ctx, &inst->pred_buf);
146 if (err) {
147 mtk_vcodec_err(inst, "failed to allocate ppl buf");
148 return err;
149 }
150
151 inst->vsi->pred_buf_dma = inst->pred_buf.dma_addr;
152 return 0;
153 }
154
free_predication_buf(struct vdec_h264_inst * inst)155 static void free_predication_buf(struct vdec_h264_inst *inst)
156 {
157 struct mtk_vcodec_mem *mem = NULL;
158
159 mtk_vcodec_debug_enter(inst);
160
161 inst->vsi->pred_buf_dma = 0;
162 mem = &inst->pred_buf;
163 if (mem->va)
164 mtk_vcodec_mem_free(inst->ctx, mem);
165 }
166
alloc_mv_buf(struct vdec_h264_inst * inst,struct vdec_pic_info * pic)167 static int alloc_mv_buf(struct vdec_h264_inst *inst, struct vdec_pic_info *pic)
168 {
169 int i;
170 int err;
171 struct mtk_vcodec_mem *mem = NULL;
172 unsigned int buf_sz = get_mv_buf_size(pic->buf_w, pic->buf_h);
173
174 for (i = 0; i < H264_MAX_FB_NUM; i++) {
175 mem = &inst->mv_buf[i];
176 if (mem->va)
177 mtk_vcodec_mem_free(inst->ctx, mem);
178 mem->size = buf_sz;
179 err = mtk_vcodec_mem_alloc(inst->ctx, mem);
180 if (err) {
181 mtk_vcodec_err(inst, "failed to allocate mv buf");
182 return err;
183 }
184 inst->vsi->mv_buf_dma[i] = mem->dma_addr;
185 }
186
187 return 0;
188 }
189
free_mv_buf(struct vdec_h264_inst * inst)190 static void free_mv_buf(struct vdec_h264_inst *inst)
191 {
192 int i;
193 struct mtk_vcodec_mem *mem = NULL;
194
195 for (i = 0; i < H264_MAX_FB_NUM; i++) {
196 inst->vsi->mv_buf_dma[i] = 0;
197 mem = &inst->mv_buf[i];
198 if (mem->va)
199 mtk_vcodec_mem_free(inst->ctx, mem);
200 }
201 }
202
check_list_validity(struct vdec_h264_inst * inst,bool disp_list)203 static int check_list_validity(struct vdec_h264_inst *inst, bool disp_list)
204 {
205 struct h264_ring_fb_list *list;
206
207 list = disp_list ? &inst->vsi->list_disp : &inst->vsi->list_free;
208
209 if (list->count > H264_MAX_FB_NUM ||
210 list->read_idx >= H264_MAX_FB_NUM ||
211 list->write_idx >= H264_MAX_FB_NUM) {
212 mtk_vcodec_err(inst, "%s list err: cnt=%d r_idx=%d w_idx=%d",
213 disp_list ? "disp" : "free", list->count,
214 list->read_idx, list->write_idx);
215 return -EINVAL;
216 }
217
218 return 0;
219 }
220
put_fb_to_free(struct vdec_h264_inst * inst,struct vdec_fb * fb)221 static void put_fb_to_free(struct vdec_h264_inst *inst, struct vdec_fb *fb)
222 {
223 struct h264_ring_fb_list *list;
224
225 if (fb) {
226 if (check_list_validity(inst, false))
227 return;
228
229 list = &inst->vsi->list_free;
230 if (list->count == H264_MAX_FB_NUM) {
231 mtk_vcodec_err(inst, "[FB] put fb free_list full");
232 return;
233 }
234
235 mtk_vcodec_debug(inst, "[FB] put fb into free_list @(%p, %llx)",
236 fb->base_y.va, (u64)fb->base_y.dma_addr);
237
238 list->fb_list[list->write_idx].vdec_fb_va = (u64)(uintptr_t)fb;
239 list->write_idx = (list->write_idx == H264_MAX_FB_NUM - 1) ?
240 0 : list->write_idx + 1;
241 list->count++;
242 }
243 }
244
get_pic_info(struct vdec_h264_inst * inst,struct vdec_pic_info * pic)245 static void get_pic_info(struct vdec_h264_inst *inst,
246 struct vdec_pic_info *pic)
247 {
248 *pic = inst->vsi->pic;
249 mtk_vcodec_debug(inst, "pic(%d, %d), buf(%d, %d)",
250 pic->pic_w, pic->pic_h, pic->buf_w, pic->buf_h);
251 mtk_vcodec_debug(inst, "fb size: Y(%d), C(%d)",
252 pic->fb_sz[0], pic->fb_sz[1]);
253 }
254
get_crop_info(struct vdec_h264_inst * inst,struct v4l2_rect * cr)255 static void get_crop_info(struct vdec_h264_inst *inst, struct v4l2_rect *cr)
256 {
257 cr->left = inst->vsi->crop.left;
258 cr->top = inst->vsi->crop.top;
259 cr->width = inst->vsi->crop.width;
260 cr->height = inst->vsi->crop.height;
261
262 mtk_vcodec_debug(inst, "l=%d, t=%d, w=%d, h=%d",
263 cr->left, cr->top, cr->width, cr->height);
264 }
265
get_dpb_size(struct vdec_h264_inst * inst,unsigned int * dpb_sz)266 static void get_dpb_size(struct vdec_h264_inst *inst, unsigned int *dpb_sz)
267 {
268 *dpb_sz = inst->vsi->dec.dpb_sz;
269 mtk_vcodec_debug(inst, "sz=%d", *dpb_sz);
270 }
271
vdec_h264_init(struct mtk_vcodec_ctx * ctx)272 static int vdec_h264_init(struct mtk_vcodec_ctx *ctx)
273 {
274 struct vdec_h264_inst *inst = NULL;
275 int err;
276
277 inst = kzalloc(sizeof(*inst), GFP_KERNEL);
278 if (!inst)
279 return -ENOMEM;
280
281 inst->ctx = ctx;
282
283 inst->vpu.id = IPI_VDEC_H264;
284 inst->vpu.ctx = ctx;
285
286 err = vpu_dec_init(&inst->vpu);
287 if (err) {
288 mtk_vcodec_err(inst, "vdec_h264 init err=%d", err);
289 goto error_free_inst;
290 }
291
292 inst->vsi = (struct vdec_h264_vsi *)inst->vpu.vsi;
293 err = allocate_predication_buf(inst);
294 if (err)
295 goto error_deinit;
296
297 mtk_vcodec_debug(inst, "H264 Instance >> %p", inst);
298
299 ctx->drv_handle = inst;
300 return 0;
301
302 error_deinit:
303 vpu_dec_deinit(&inst->vpu);
304
305 error_free_inst:
306 kfree(inst);
307 return err;
308 }
309
vdec_h264_deinit(void * h_vdec)310 static void vdec_h264_deinit(void *h_vdec)
311 {
312 struct vdec_h264_inst *inst = (struct vdec_h264_inst *)h_vdec;
313
314 mtk_vcodec_debug_enter(inst);
315
316 vpu_dec_deinit(&inst->vpu);
317 free_predication_buf(inst);
318 free_mv_buf(inst);
319
320 kfree(inst);
321 }
322
find_start_code(unsigned char * data,unsigned int data_sz)323 static int find_start_code(unsigned char *data, unsigned int data_sz)
324 {
325 if (data_sz > 3 && data[0] == 0 && data[1] == 0 && data[2] == 1)
326 return 3;
327
328 if (data_sz > 4 && data[0] == 0 && data[1] == 0 && data[2] == 0 &&
329 data[3] == 1)
330 return 4;
331
332 return -1;
333 }
334
vdec_h264_decode(void * h_vdec,struct mtk_vcodec_mem * bs,struct vdec_fb * fb,bool * res_chg)335 static int vdec_h264_decode(void *h_vdec, struct mtk_vcodec_mem *bs,
336 struct vdec_fb *fb, bool *res_chg)
337 {
338 struct vdec_h264_inst *inst = (struct vdec_h264_inst *)h_vdec;
339 struct vdec_vpu_inst *vpu = &inst->vpu;
340 int nal_start_idx = 0;
341 int err = 0;
342 unsigned int nal_start;
343 unsigned int nal_type;
344 unsigned char *buf;
345 unsigned int buf_sz;
346 unsigned int data[2];
347 uint64_t vdec_fb_va = (u64)(uintptr_t)fb;
348 uint64_t y_fb_dma = fb ? (u64)fb->base_y.dma_addr : 0;
349 uint64_t c_fb_dma = fb ? (u64)fb->base_c.dma_addr : 0;
350
351 mtk_vcodec_debug(inst, "+ [%d] FB y_dma=%llx c_dma=%llx va=%p",
352 ++inst->num_nalu, y_fb_dma, c_fb_dma, fb);
353
354 /* bs NULL means flush decoder */
355 if (bs == NULL)
356 return vpu_dec_reset(vpu);
357
358 buf = (unsigned char *)bs->va;
359 buf_sz = bs->size;
360 nal_start_idx = find_start_code(buf, buf_sz);
361 if (nal_start_idx < 0) {
362 mtk_vcodec_err(inst, "invalid nal start code");
363 err = -EIO;
364 goto err_free_fb_out;
365 }
366
367 nal_start = buf[nal_start_idx];
368 nal_type = NAL_TYPE(buf[nal_start_idx]);
369 mtk_vcodec_debug(inst, "\n + NALU[%d] type %d +\n", inst->num_nalu,
370 nal_type);
371
372 if (nal_type == NAL_H264_PPS) {
373 buf_sz -= nal_start_idx;
374 if (buf_sz > HDR_PARSING_BUF_SZ) {
375 err = -EILSEQ;
376 goto err_free_fb_out;
377 }
378 memcpy(inst->vsi->hdr_buf, buf + nal_start_idx, buf_sz);
379 }
380
381 inst->vsi->dec.bs_dma = (uint64_t)bs->dma_addr;
382 inst->vsi->dec.y_fb_dma = y_fb_dma;
383 inst->vsi->dec.c_fb_dma = c_fb_dma;
384 inst->vsi->dec.vdec_fb_va = vdec_fb_va;
385
386 data[0] = buf_sz;
387 data[1] = nal_start;
388 err = vpu_dec_start(vpu, data, 2);
389 if (err) {
390 if (err > 0 && (DEC_ERR_RET(err) == H264_ERR_NOT_VALID)) {
391 mtk_vcodec_err(inst, "- error bitstream - err = %d -",
392 err);
393 err = -EIO;
394 }
395 goto err_free_fb_out;
396 }
397
398 *res_chg = inst->vsi->dec.resolution_changed;
399 if (*res_chg) {
400 struct vdec_pic_info pic;
401
402 mtk_vcodec_debug(inst, "- resolution changed -");
403 get_pic_info(inst, &pic);
404
405 if (inst->vsi->dec.realloc_mv_buf) {
406 err = alloc_mv_buf(inst, &pic);
407 if (err)
408 goto err_free_fb_out;
409 }
410 }
411
412 if (nal_type == NAL_NON_IDR_SLICE || nal_type == NAL_IDR_SLICE) {
413 /* wait decoder done interrupt */
414 err = mtk_vcodec_wait_for_done_ctx(inst->ctx,
415 MTK_INST_IRQ_RECEIVED,
416 WAIT_INTR_TIMEOUT_MS);
417 if (err)
418 goto err_free_fb_out;
419
420 vpu_dec_end(vpu);
421 }
422
423 mtk_vcodec_debug(inst, "\n - NALU[%d] type=%d -\n", inst->num_nalu,
424 nal_type);
425 return 0;
426
427 err_free_fb_out:
428 put_fb_to_free(inst, fb);
429 mtk_vcodec_err(inst, "\n - NALU[%d] err=%d -\n", inst->num_nalu, err);
430 return err;
431 }
432
vdec_h264_get_fb(struct vdec_h264_inst * inst,struct h264_ring_fb_list * list,bool disp_list,struct vdec_fb ** out_fb)433 static void vdec_h264_get_fb(struct vdec_h264_inst *inst,
434 struct h264_ring_fb_list *list,
435 bool disp_list, struct vdec_fb **out_fb)
436 {
437 struct vdec_fb *fb;
438
439 if (check_list_validity(inst, disp_list))
440 return;
441
442 if (list->count == 0) {
443 mtk_vcodec_debug(inst, "[FB] there is no %s fb",
444 disp_list ? "disp" : "free");
445 *out_fb = NULL;
446 return;
447 }
448
449 fb = (struct vdec_fb *)
450 (uintptr_t)list->fb_list[list->read_idx].vdec_fb_va;
451 fb->status |= (disp_list ? FB_ST_DISPLAY : FB_ST_FREE);
452
453 *out_fb = fb;
454 mtk_vcodec_debug(inst, "[FB] get %s fb st=%d poc=%d %llx",
455 disp_list ? "disp" : "free",
456 fb->status, list->fb_list[list->read_idx].poc,
457 list->fb_list[list->read_idx].vdec_fb_va);
458
459 list->read_idx = (list->read_idx == H264_MAX_FB_NUM - 1) ?
460 0 : list->read_idx + 1;
461 list->count--;
462 }
463
vdec_h264_get_param(void * h_vdec,enum vdec_get_param_type type,void * out)464 static int vdec_h264_get_param(void *h_vdec, enum vdec_get_param_type type,
465 void *out)
466 {
467 struct vdec_h264_inst *inst = (struct vdec_h264_inst *)h_vdec;
468
469 switch (type) {
470 case GET_PARAM_DISP_FRAME_BUFFER:
471 vdec_h264_get_fb(inst, &inst->vsi->list_disp, true, out);
472 break;
473
474 case GET_PARAM_FREE_FRAME_BUFFER:
475 vdec_h264_get_fb(inst, &inst->vsi->list_free, false, out);
476 break;
477
478 case GET_PARAM_PIC_INFO:
479 get_pic_info(inst, out);
480 break;
481
482 case GET_PARAM_DPB_SIZE:
483 get_dpb_size(inst, out);
484 break;
485
486 case GET_PARAM_CROP_INFO:
487 get_crop_info(inst, out);
488 break;
489
490 default:
491 mtk_vcodec_err(inst, "invalid get parameter type=%d", type);
492 return -EINVAL;
493 }
494
495 return 0;
496 }
497
498 const struct vdec_common_if vdec_h264_if = {
499 .init = vdec_h264_init,
500 .decode = vdec_h264_decode,
501 .get_param = vdec_h264_get_param,
502 .deinit = vdec_h264_deinit,
503 };
504