1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * Coda multi-standard codec IP - JPEG support functions 4 * 5 * Copyright (C) 2014 Philipp Zabel, Pengutronix 6 */ 7 8 #include <asm/unaligned.h> 9 #include <linux/irqreturn.h> 10 #include <linux/kernel.h> 11 #include <linux/ktime.h> 12 #include <linux/slab.h> 13 #include <linux/swab.h> 14 #include <linux/videodev2.h> 15 16 #include <media/v4l2-common.h> 17 #include <media/v4l2-fh.h> 18 #include <media/v4l2-jpeg.h> 19 #include <media/v4l2-mem2mem.h> 20 #include <media/videobuf2-core.h> 21 #include <media/videobuf2-dma-contig.h> 22 23 #include "coda.h" 24 #include "trace.h" 25 26 #define SOI_MARKER 0xffd8 27 #define APP9_MARKER 0xffe9 28 #define DRI_MARKER 0xffdd 29 #define DQT_MARKER 0xffdb 30 #define DHT_MARKER 0xffc4 31 #define SOF_MARKER 0xffc0 32 #define SOS_MARKER 0xffda 33 #define EOI_MARKER 0xffd9 34 35 enum { 36 CODA9_JPEG_FORMAT_420, 37 CODA9_JPEG_FORMAT_422, 38 CODA9_JPEG_FORMAT_224, 39 CODA9_JPEG_FORMAT_444, 40 CODA9_JPEG_FORMAT_400, 41 }; 42 43 struct coda_huff_tab { 44 u8 luma_dc[16 + 12]; 45 u8 chroma_dc[16 + 12]; 46 u8 luma_ac[16 + 162]; 47 u8 chroma_ac[16 + 162]; 48 49 /* DC Luma, DC Chroma, AC Luma, AC Chroma */ 50 s16 min[4 * 16]; 51 s16 max[4 * 16]; 52 s8 ptr[4 * 16]; 53 }; 54 55 #define CODA9_JPEG_ENC_HUFF_DATA_SIZE (256 + 256 + 16 + 16) 56 57 /* 58 * Typical Huffman tables for 8-bit precision luminance and 59 * chrominance from JPEG ITU-T.81 (ISO/IEC 10918-1) Annex K.3 60 */ 61 62 static const unsigned char luma_dc[16 + 12] = { 63 /* bits */ 64 0x00, 0x01, 0x05, 0x01, 0x01, 0x01, 0x01, 0x01, 65 0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 66 /* values */ 67 0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, 68 0x08, 0x09, 0x0a, 0x0b, 69 }; 70 71 static const unsigned char chroma_dc[16 + 12] = { 72 /* bits */ 73 0x00, 0x03, 0x01, 0x01, 0x01, 0x01, 0x01, 0x01, 74 0x01, 0x01, 0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 75 /* values */ 76 0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, 77 0x08, 0x09, 0x0a, 0x0b, 78 }; 79 80 static const unsigned char luma_ac[16 + 162 + 2] = { 81 /* bits */ 82 0x00, 0x02, 0x01, 0x03, 0x03, 0x02, 0x04, 0x03, 83 0x05, 0x05, 0x04, 0x04, 0x00, 0x00, 0x01, 0x7d, 84 /* values */ 85 0x01, 0x02, 0x03, 0x00, 0x04, 0x11, 0x05, 0x12, 86 0x21, 0x31, 0x41, 0x06, 0x13, 0x51, 0x61, 0x07, 87 0x22, 0x71, 0x14, 0x32, 0x81, 0x91, 0xa1, 0x08, 88 0x23, 0x42, 0xb1, 0xc1, 0x15, 0x52, 0xd1, 0xf0, 89 0x24, 0x33, 0x62, 0x72, 0x82, 0x09, 0x0a, 0x16, 90 0x17, 0x18, 0x19, 0x1a, 0x25, 0x26, 0x27, 0x28, 91 0x29, 0x2a, 0x34, 0x35, 0x36, 0x37, 0x38, 0x39, 92 0x3a, 0x43, 0x44, 0x45, 0x46, 0x47, 0x48, 0x49, 93 0x4a, 0x53, 0x54, 0x55, 0x56, 0x57, 0x58, 0x59, 94 0x5a, 0x63, 0x64, 0x65, 0x66, 0x67, 0x68, 0x69, 95 0x6a, 0x73, 0x74, 0x75, 0x76, 0x77, 0x78, 0x79, 96 0x7a, 0x83, 0x84, 0x85, 0x86, 0x87, 0x88, 0x89, 97 0x8a, 0x92, 0x93, 0x94, 0x95, 0x96, 0x97, 0x98, 98 0x99, 0x9a, 0xa2, 0xa3, 0xa4, 0xa5, 0xa6, 0xa7, 99 0xa8, 0xa9, 0xaa, 0xb2, 0xb3, 0xb4, 0xb5, 0xb6, 100 0xb7, 0xb8, 0xb9, 0xba, 0xc2, 0xc3, 0xc4, 0xc5, 101 0xc6, 0xc7, 0xc8, 0xc9, 0xca, 0xd2, 0xd3, 0xd4, 102 0xd5, 0xd6, 0xd7, 0xd8, 0xd9, 0xda, 0xe1, 0xe2, 103 0xe3, 0xe4, 0xe5, 0xe6, 0xe7, 0xe8, 0xe9, 0xea, 104 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7, 0xf8, 105 0xf9, 0xfa, /* padded to 32-bit */ 106 }; 107 108 static const unsigned char chroma_ac[16 + 162 + 2] = { 109 /* bits */ 110 0x00, 0x02, 0x01, 0x02, 0x04, 0x04, 0x03, 0x04, 111 0x07, 0x05, 0x04, 0x04, 0x00, 0x01, 0x02, 0x77, 112 /* values */ 113 0x00, 0x01, 0x02, 0x03, 0x11, 0x04, 0x05, 0x21, 114 0x31, 0x06, 0x12, 0x41, 0x51, 0x07, 0x61, 0x71, 115 0x13, 0x22, 0x32, 0x81, 0x08, 0x14, 0x42, 0x91, 116 0xa1, 0xb1, 0xc1, 0x09, 0x23, 0x33, 0x52, 0xf0, 117 0x15, 0x62, 0x72, 0xd1, 0x0a, 0x16, 0x24, 0x34, 118 0xe1, 0x25, 0xf1, 0x17, 0x18, 0x19, 0x1a, 0x26, 119 0x27, 0x28, 0x29, 0x2a, 0x35, 0x36, 0x37, 0x38, 120 0x39, 0x3a, 0x43, 0x44, 0x45, 0x46, 0x47, 0x48, 121 0x49, 0x4a, 0x53, 0x54, 0x55, 0x56, 0x57, 0x58, 122 0x59, 0x5a, 0x63, 0x64, 0x65, 0x66, 0x67, 0x68, 123 0x69, 0x6a, 0x73, 0x74, 0x75, 0x76, 0x77, 0x78, 124 0x79, 0x7a, 0x82, 0x83, 0x84, 0x85, 0x86, 0x87, 125 0x88, 0x89, 0x8a, 0x92, 0x93, 0x94, 0x95, 0x96, 126 0x97, 0x98, 0x99, 0x9a, 0xa2, 0xa3, 0xa4, 0xa5, 127 0xa6, 0xa7, 0xa8, 0xa9, 0xaa, 0xb2, 0xb3, 0xb4, 128 0xb5, 0xb6, 0xb7, 0xb8, 0xb9, 0xba, 0xc2, 0xc3, 129 0xc4, 0xc5, 0xc6, 0xc7, 0xc8, 0xc9, 0xca, 0xd2, 130 0xd3, 0xd4, 0xd5, 0xd6, 0xd7, 0xd8, 0xd9, 0xda, 131 0xe2, 0xe3, 0xe4, 0xe5, 0xe6, 0xe7, 0xe8, 0xe9, 132 0xea, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7, 0xf8, 133 0xf9, 0xfa, /* padded to 32-bit */ 134 }; 135 136 /* 137 * Quantization tables for luminance and chrominance components in 138 * zig-zag scan order from the Freescale i.MX VPU libraries 139 */ 140 141 static unsigned char luma_q[64] = { 142 0x06, 0x04, 0x04, 0x04, 0x05, 0x04, 0x06, 0x05, 143 0x05, 0x06, 0x09, 0x06, 0x05, 0x06, 0x09, 0x0b, 144 0x08, 0x06, 0x06, 0x08, 0x0b, 0x0c, 0x0a, 0x0a, 145 0x0b, 0x0a, 0x0a, 0x0c, 0x10, 0x0c, 0x0c, 0x0c, 146 0x0c, 0x0c, 0x0c, 0x10, 0x0c, 0x0c, 0x0c, 0x0c, 147 0x0c, 0x0c, 0x0c, 0x0c, 0x0c, 0x0c, 0x0c, 0x0c, 148 0x0c, 0x0c, 0x0c, 0x0c, 0x0c, 0x0c, 0x0c, 0x0c, 149 0x0c, 0x0c, 0x0c, 0x0c, 0x0c, 0x0c, 0x0c, 0x0c, 150 }; 151 152 static unsigned char chroma_q[64] = { 153 0x07, 0x07, 0x07, 0x0d, 0x0c, 0x0d, 0x18, 0x10, 154 0x10, 0x18, 0x14, 0x0e, 0x0e, 0x0e, 0x14, 0x14, 155 0x0e, 0x0e, 0x0e, 0x0e, 0x14, 0x11, 0x0c, 0x0c, 156 0x0c, 0x0c, 0x0c, 0x11, 0x11, 0x0c, 0x0c, 0x0c, 157 0x0c, 0x0c, 0x0c, 0x11, 0x0c, 0x0c, 0x0c, 0x0c, 158 0x0c, 0x0c, 0x0c, 0x0c, 0x0c, 0x0c, 0x0c, 0x0c, 159 0x0c, 0x0c, 0x0c, 0x0c, 0x0c, 0x0c, 0x0c, 0x0c, 160 0x0c, 0x0c, 0x0c, 0x0c, 0x0c, 0x0c, 0x0c, 0x0c, 161 }; 162 163 static const unsigned char width_align[] = { 164 [CODA9_JPEG_FORMAT_420] = 16, 165 [CODA9_JPEG_FORMAT_422] = 16, 166 [CODA9_JPEG_FORMAT_224] = 8, 167 [CODA9_JPEG_FORMAT_444] = 8, 168 [CODA9_JPEG_FORMAT_400] = 8, 169 }; 170 171 static const unsigned char height_align[] = { 172 [CODA9_JPEG_FORMAT_420] = 16, 173 [CODA9_JPEG_FORMAT_422] = 8, 174 [CODA9_JPEG_FORMAT_224] = 16, 175 [CODA9_JPEG_FORMAT_444] = 8, 176 [CODA9_JPEG_FORMAT_400] = 8, 177 }; 178 179 static int coda9_jpeg_chroma_format(u32 pixfmt) 180 { 181 switch (pixfmt) { 182 case V4L2_PIX_FMT_YUV420: 183 case V4L2_PIX_FMT_NV12: 184 return CODA9_JPEG_FORMAT_420; 185 case V4L2_PIX_FMT_YUV422P: 186 return CODA9_JPEG_FORMAT_422; 187 case V4L2_PIX_FMT_YUV444: 188 return CODA9_JPEG_FORMAT_444; 189 case V4L2_PIX_FMT_GREY: 190 return CODA9_JPEG_FORMAT_400; 191 } 192 return -EINVAL; 193 } 194 195 struct coda_memcpy_desc { 196 int offset; 197 const void *src; 198 size_t len; 199 }; 200 201 static void coda_memcpy_parabuf(void *parabuf, 202 const struct coda_memcpy_desc *desc) 203 { 204 u32 *dst = parabuf + desc->offset; 205 const u32 *src = desc->src; 206 int len = desc->len / 4; 207 int i; 208 209 for (i = 0; i < len; i += 2) { 210 dst[i + 1] = swab32(src[i]); 211 dst[i] = swab32(src[i + 1]); 212 } 213 } 214 215 int coda_jpeg_write_tables(struct coda_ctx *ctx) 216 { 217 int i; 218 static const struct coda_memcpy_desc huff[8] = { 219 { 0, luma_dc, sizeof(luma_dc) }, 220 { 32, luma_ac, sizeof(luma_ac) }, 221 { 216, chroma_dc, sizeof(chroma_dc) }, 222 { 248, chroma_ac, sizeof(chroma_ac) }, 223 }; 224 struct coda_memcpy_desc qmat[3] = { 225 { 512, ctx->params.jpeg_qmat_tab[0], 64 }, 226 { 576, ctx->params.jpeg_qmat_tab[1], 64 }, 227 { 640, ctx->params.jpeg_qmat_tab[1], 64 }, 228 }; 229 230 /* Write huffman tables to parameter memory */ 231 for (i = 0; i < ARRAY_SIZE(huff); i++) 232 coda_memcpy_parabuf(ctx->parabuf.vaddr, huff + i); 233 234 /* Write Q-matrix to parameter memory */ 235 for (i = 0; i < ARRAY_SIZE(qmat); i++) 236 coda_memcpy_parabuf(ctx->parabuf.vaddr, qmat + i); 237 238 return 0; 239 } 240 241 bool coda_jpeg_check_buffer(struct coda_ctx *ctx, struct vb2_buffer *vb) 242 { 243 void *vaddr = vb2_plane_vaddr(vb, 0); 244 u16 soi, eoi; 245 int len, i; 246 247 soi = be16_to_cpup((__be16 *)vaddr); 248 if (soi != SOI_MARKER) 249 return false; 250 251 len = vb2_get_plane_payload(vb, 0); 252 vaddr += len - 2; 253 for (i = 0; i < 32; i++) { 254 eoi = be16_to_cpup((__be16 *)(vaddr - i)); 255 if (eoi == EOI_MARKER) { 256 if (i > 0) 257 vb2_set_plane_payload(vb, 0, len - i); 258 return true; 259 } 260 } 261 262 return false; 263 } 264 265 static int coda9_jpeg_gen_dec_huff_tab(struct coda_ctx *ctx, int tab_num); 266 267 int coda_jpeg_decode_header(struct coda_ctx *ctx, struct vb2_buffer *vb) 268 { 269 struct coda_dev *dev = ctx->dev; 270 u8 *buf = vb2_plane_vaddr(vb, 0); 271 size_t len = vb2_get_plane_payload(vb, 0); 272 struct v4l2_jpeg_scan_header scan_header; 273 struct v4l2_jpeg_reference quantization_tables[4] = { }; 274 struct v4l2_jpeg_reference huffman_tables[4] = { }; 275 struct v4l2_jpeg_header header = { 276 .scan = &scan_header, 277 .quantization_tables = quantization_tables, 278 .huffman_tables = huffman_tables, 279 }; 280 struct coda_q_data *q_data_src; 281 struct coda_huff_tab *huff_tab; 282 int i, j, ret; 283 284 ret = v4l2_jpeg_parse_header(buf, len, &header); 285 if (ret < 0) { 286 v4l2_err(&dev->v4l2_dev, "failed to parse JPEG header: %pe\n", 287 ERR_PTR(ret)); 288 return ret; 289 } 290 291 ctx->params.jpeg_restart_interval = header.restart_interval; 292 293 /* check frame header */ 294 if (header.frame.height > ctx->codec->max_h || 295 header.frame.width > ctx->codec->max_w) { 296 v4l2_err(&dev->v4l2_dev, "invalid dimensions: %dx%d\n", 297 header.frame.width, header.frame.height); 298 return -EINVAL; 299 } 300 301 q_data_src = get_q_data(ctx, V4L2_BUF_TYPE_VIDEO_OUTPUT); 302 if (header.frame.height != q_data_src->height || 303 header.frame.width != q_data_src->width) { 304 v4l2_err(&dev->v4l2_dev, 305 "dimensions don't match format: %dx%d\n", 306 header.frame.width, header.frame.height); 307 return -EINVAL; 308 } 309 310 if (header.frame.num_components != 3) { 311 v4l2_err(&dev->v4l2_dev, 312 "unsupported number of components: %d\n", 313 header.frame.num_components); 314 return -EINVAL; 315 } 316 317 /* install quantization tables */ 318 if (quantization_tables[3].start) { 319 v4l2_err(&dev->v4l2_dev, 320 "only 3 quantization tables supported\n"); 321 return -EINVAL; 322 } 323 for (i = 0; i < 3; i++) { 324 if (!quantization_tables[i].start) 325 continue; 326 if (quantization_tables[i].length != 64) { 327 v4l2_err(&dev->v4l2_dev, 328 "only 8-bit quantization tables supported\n"); 329 continue; 330 } 331 if (!ctx->params.jpeg_qmat_tab[i]) { 332 ctx->params.jpeg_qmat_tab[i] = kmalloc(64, GFP_KERNEL); 333 if (!ctx->params.jpeg_qmat_tab[i]) 334 return -ENOMEM; 335 } 336 memcpy(ctx->params.jpeg_qmat_tab[i], 337 quantization_tables[i].start, 64); 338 } 339 340 /* install Huffman tables */ 341 for (i = 0; i < 4; i++) { 342 if (!huffman_tables[i].start) { 343 v4l2_err(&dev->v4l2_dev, "missing Huffman table\n"); 344 return -EINVAL; 345 } 346 /* AC tables should be between 17 -> 178, DC between 17 -> 28 */ 347 if (huffman_tables[i].length < 17 || 348 huffman_tables[i].length > 178 || 349 ((i & 2) == 0 && huffman_tables[i].length > 28)) { 350 v4l2_err(&dev->v4l2_dev, 351 "invalid Huffman table %d length: %zu\n", 352 i, huffman_tables[i].length); 353 return -EINVAL; 354 } 355 } 356 huff_tab = ctx->params.jpeg_huff_tab; 357 if (!huff_tab) { 358 huff_tab = kzalloc(sizeof(struct coda_huff_tab), GFP_KERNEL); 359 if (!huff_tab) 360 return -ENOMEM; 361 ctx->params.jpeg_huff_tab = huff_tab; 362 } 363 364 memset(huff_tab, 0, sizeof(*huff_tab)); 365 memcpy(huff_tab->luma_dc, huffman_tables[0].start, huffman_tables[0].length); 366 memcpy(huff_tab->chroma_dc, huffman_tables[1].start, huffman_tables[1].length); 367 memcpy(huff_tab->luma_ac, huffman_tables[2].start, huffman_tables[2].length); 368 memcpy(huff_tab->chroma_ac, huffman_tables[3].start, huffman_tables[3].length); 369 370 /* check scan header */ 371 for (i = 0; i < scan_header.num_components; i++) { 372 struct v4l2_jpeg_scan_component_spec *scan_component; 373 374 scan_component = &scan_header.component[i]; 375 for (j = 0; j < header.frame.num_components; j++) { 376 if (header.frame.component[j].component_identifier == 377 scan_component->component_selector) 378 break; 379 } 380 if (j == header.frame.num_components) 381 continue; 382 383 ctx->params.jpeg_huff_dc_index[j] = 384 scan_component->dc_entropy_coding_table_selector; 385 ctx->params.jpeg_huff_ac_index[j] = 386 scan_component->ac_entropy_coding_table_selector; 387 } 388 389 /* Generate Huffman table information */ 390 for (i = 0; i < 4; i++) 391 coda9_jpeg_gen_dec_huff_tab(ctx, i); 392 393 /* start of entropy coded segment */ 394 ctx->jpeg_ecs_offset = header.ecs_offset; 395 396 switch (header.frame.subsampling) { 397 case V4L2_JPEG_CHROMA_SUBSAMPLING_420: 398 case V4L2_JPEG_CHROMA_SUBSAMPLING_422: 399 ctx->params.jpeg_chroma_subsampling = header.frame.subsampling; 400 break; 401 default: 402 v4l2_err(&dev->v4l2_dev, "chroma subsampling not supported: %d", 403 header.frame.subsampling); 404 return -EINVAL; 405 } 406 407 return 0; 408 } 409 410 static inline void coda9_jpeg_write_huff_values(struct coda_dev *dev, u8 *bits, 411 int num_values) 412 { 413 s8 *values = (s8 *)(bits + 16); 414 int huff_length, i; 415 416 for (huff_length = 0, i = 0; i < 16; i++) 417 huff_length += bits[i]; 418 for (i = huff_length; i < num_values; i++) 419 values[i] = -1; 420 for (i = 0; i < num_values; i++) 421 coda_write(dev, (s32)values[i], CODA9_REG_JPEG_HUFF_DATA); 422 } 423 424 static int coda9_jpeg_dec_huff_setup(struct coda_ctx *ctx) 425 { 426 struct coda_huff_tab *huff_tab = ctx->params.jpeg_huff_tab; 427 struct coda_dev *dev = ctx->dev; 428 s16 *huff_min = huff_tab->min; 429 s16 *huff_max = huff_tab->max; 430 s8 *huff_ptr = huff_tab->ptr; 431 int i; 432 433 /* MIN Tables */ 434 coda_write(dev, 0x003, CODA9_REG_JPEG_HUFF_CTRL); 435 coda_write(dev, 0x000, CODA9_REG_JPEG_HUFF_ADDR); 436 for (i = 0; i < 4 * 16; i++) 437 coda_write(dev, (s32)huff_min[i], CODA9_REG_JPEG_HUFF_DATA); 438 439 /* MAX Tables */ 440 coda_write(dev, 0x403, CODA9_REG_JPEG_HUFF_CTRL); 441 coda_write(dev, 0x440, CODA9_REG_JPEG_HUFF_ADDR); 442 for (i = 0; i < 4 * 16; i++) 443 coda_write(dev, (s32)huff_max[i], CODA9_REG_JPEG_HUFF_DATA); 444 445 /* PTR Tables */ 446 coda_write(dev, 0x803, CODA9_REG_JPEG_HUFF_CTRL); 447 coda_write(dev, 0x880, CODA9_REG_JPEG_HUFF_ADDR); 448 for (i = 0; i < 4 * 16; i++) 449 coda_write(dev, (s32)huff_ptr[i], CODA9_REG_JPEG_HUFF_DATA); 450 451 /* VAL Tables: DC Luma, DC Chroma, AC Luma, AC Chroma */ 452 coda_write(dev, 0xc03, CODA9_REG_JPEG_HUFF_CTRL); 453 coda9_jpeg_write_huff_values(dev, huff_tab->luma_dc, 12); 454 coda9_jpeg_write_huff_values(dev, huff_tab->chroma_dc, 12); 455 coda9_jpeg_write_huff_values(dev, huff_tab->luma_ac, 162); 456 coda9_jpeg_write_huff_values(dev, huff_tab->chroma_ac, 162); 457 coda_write(dev, 0x000, CODA9_REG_JPEG_HUFF_CTRL); 458 return 0; 459 } 460 461 static inline void coda9_jpeg_write_qmat_tab(struct coda_dev *dev, 462 u8 *qmat, int index) 463 { 464 int i; 465 466 coda_write(dev, index | 0x3, CODA9_REG_JPEG_QMAT_CTRL); 467 for (i = 0; i < 64; i++) 468 coda_write(dev, qmat[i], CODA9_REG_JPEG_QMAT_DATA); 469 coda_write(dev, 0, CODA9_REG_JPEG_QMAT_CTRL); 470 } 471 472 static void coda9_jpeg_qmat_setup(struct coda_ctx *ctx) 473 { 474 struct coda_dev *dev = ctx->dev; 475 int *qmat_index = ctx->params.jpeg_qmat_index; 476 u8 **qmat_tab = ctx->params.jpeg_qmat_tab; 477 478 coda9_jpeg_write_qmat_tab(dev, qmat_tab[qmat_index[0]], 0x00); 479 coda9_jpeg_write_qmat_tab(dev, qmat_tab[qmat_index[1]], 0x40); 480 coda9_jpeg_write_qmat_tab(dev, qmat_tab[qmat_index[2]], 0x80); 481 } 482 483 static void coda9_jpeg_dec_bbc_gbu_setup(struct coda_ctx *ctx, 484 struct vb2_buffer *buf, u32 ecs_offset) 485 { 486 struct coda_dev *dev = ctx->dev; 487 int page_ptr, word_ptr, bit_ptr; 488 u32 bbc_base_addr, end_addr; 489 int bbc_cur_pos; 490 int ret, val; 491 492 bbc_base_addr = vb2_dma_contig_plane_dma_addr(buf, 0); 493 end_addr = bbc_base_addr + vb2_get_plane_payload(buf, 0); 494 495 page_ptr = ecs_offset / 256; 496 word_ptr = (ecs_offset % 256) / 4; 497 if (page_ptr & 1) 498 word_ptr += 64; 499 bit_ptr = (ecs_offset % 4) * 8; 500 if (word_ptr & 1) 501 bit_ptr += 32; 502 word_ptr &= ~0x1; 503 504 coda_write(dev, end_addr, CODA9_REG_JPEG_BBC_WR_PTR); 505 coda_write(dev, bbc_base_addr, CODA9_REG_JPEG_BBC_BAS_ADDR); 506 507 /* Leave 3 256-byte page margin to avoid a BBC interrupt */ 508 coda_write(dev, end_addr + 256 * 3 + 256, CODA9_REG_JPEG_BBC_END_ADDR); 509 val = DIV_ROUND_UP(vb2_plane_size(buf, 0), 256) + 3; 510 coda_write(dev, BIT(31) | val, CODA9_REG_JPEG_BBC_STRM_CTRL); 511 512 bbc_cur_pos = page_ptr; 513 coda_write(dev, bbc_cur_pos, CODA9_REG_JPEG_BBC_CUR_POS); 514 coda_write(dev, bbc_base_addr + (bbc_cur_pos << 8), 515 CODA9_REG_JPEG_BBC_EXT_ADDR); 516 coda_write(dev, (bbc_cur_pos & 1) << 6, CODA9_REG_JPEG_BBC_INT_ADDR); 517 coda_write(dev, 64, CODA9_REG_JPEG_BBC_DATA_CNT); 518 coda_write(dev, 0, CODA9_REG_JPEG_BBC_COMMAND); 519 do { 520 ret = coda_read(dev, CODA9_REG_JPEG_BBC_BUSY); 521 } while (ret == 1); 522 523 bbc_cur_pos++; 524 coda_write(dev, bbc_cur_pos, CODA9_REG_JPEG_BBC_CUR_POS); 525 coda_write(dev, bbc_base_addr + (bbc_cur_pos << 8), 526 CODA9_REG_JPEG_BBC_EXT_ADDR); 527 coda_write(dev, (bbc_cur_pos & 1) << 6, CODA9_REG_JPEG_BBC_INT_ADDR); 528 coda_write(dev, 64, CODA9_REG_JPEG_BBC_DATA_CNT); 529 coda_write(dev, 0, CODA9_REG_JPEG_BBC_COMMAND); 530 do { 531 ret = coda_read(dev, CODA9_REG_JPEG_BBC_BUSY); 532 } while (ret == 1); 533 534 bbc_cur_pos++; 535 coda_write(dev, bbc_cur_pos, CODA9_REG_JPEG_BBC_CUR_POS); 536 coda_write(dev, 1, CODA9_REG_JPEG_BBC_CTRL); 537 538 coda_write(dev, 0, CODA9_REG_JPEG_GBU_TT_CNT); 539 coda_write(dev, word_ptr, CODA9_REG_JPEG_GBU_WD_PTR); 540 coda_write(dev, 0, CODA9_REG_JPEG_GBU_BBSR); 541 coda_write(dev, 127, CODA9_REG_JPEG_GBU_BBER); 542 if (page_ptr & 1) { 543 coda_write(dev, 0, CODA9_REG_JPEG_GBU_BBIR); 544 coda_write(dev, 0, CODA9_REG_JPEG_GBU_BBHR); 545 } else { 546 coda_write(dev, 64, CODA9_REG_JPEG_GBU_BBIR); 547 coda_write(dev, 64, CODA9_REG_JPEG_GBU_BBHR); 548 } 549 coda_write(dev, 4, CODA9_REG_JPEG_GBU_CTRL); 550 coda_write(dev, bit_ptr, CODA9_REG_JPEG_GBU_FF_RPTR); 551 coda_write(dev, 3, CODA9_REG_JPEG_GBU_CTRL); 552 } 553 554 static const int bus_req_num[] = { 555 [CODA9_JPEG_FORMAT_420] = 2, 556 [CODA9_JPEG_FORMAT_422] = 3, 557 [CODA9_JPEG_FORMAT_224] = 3, 558 [CODA9_JPEG_FORMAT_444] = 4, 559 [CODA9_JPEG_FORMAT_400] = 4, 560 }; 561 562 #define MCU_INFO(mcu_block_num, comp_num, comp0_info, comp1_info, comp2_info) \ 563 (((mcu_block_num) << CODA9_JPEG_MCU_BLOCK_NUM_OFFSET) | \ 564 ((comp_num) << CODA9_JPEG_COMP_NUM_OFFSET) | \ 565 ((comp0_info) << CODA9_JPEG_COMP0_INFO_OFFSET) | \ 566 ((comp1_info) << CODA9_JPEG_COMP1_INFO_OFFSET) | \ 567 ((comp2_info) << CODA9_JPEG_COMP2_INFO_OFFSET)) 568 569 static const u32 mcu_info[] = { 570 [CODA9_JPEG_FORMAT_420] = MCU_INFO(6, 3, 10, 5, 5), 571 [CODA9_JPEG_FORMAT_422] = MCU_INFO(4, 3, 9, 5, 5), 572 [CODA9_JPEG_FORMAT_224] = MCU_INFO(4, 3, 6, 5, 5), 573 [CODA9_JPEG_FORMAT_444] = MCU_INFO(3, 3, 5, 5, 5), 574 [CODA9_JPEG_FORMAT_400] = MCU_INFO(1, 1, 5, 0, 0), 575 }; 576 577 /* 578 * Convert Huffman table specifcations to tables of codes and code lengths. 579 * For reference, see JPEG ITU-T.81 (ISO/IEC 10918-1) [1] 580 * 581 * [1] https://www.w3.org/Graphics/JPEG/itu-t81.pdf 582 */ 583 static int coda9_jpeg_gen_enc_huff_tab(struct coda_ctx *ctx, int tab_num, 584 int *ehufsi, int *ehufco) 585 { 586 int i, j, k, lastk, si, code, maxsymbol; 587 const u8 *bits, *huffval; 588 struct { 589 int size[256]; 590 int code[256]; 591 } *huff; 592 static const unsigned char *huff_tabs[4] = { 593 luma_dc, luma_ac, chroma_dc, chroma_ac, 594 }; 595 int ret = -EINVAL; 596 597 huff = kzalloc(sizeof(*huff), GFP_KERNEL); 598 if (!huff) 599 return -ENOMEM; 600 601 bits = huff_tabs[tab_num]; 602 huffval = huff_tabs[tab_num] + 16; 603 604 maxsymbol = tab_num & 1 ? 256 : 16; 605 606 /* Figure C.1 - Generation of table of Huffman code sizes */ 607 k = 0; 608 for (i = 1; i <= 16; i++) { 609 j = bits[i - 1]; 610 if (k + j > maxsymbol) 611 goto out; 612 while (j--) 613 huff->size[k++] = i; 614 } 615 lastk = k; 616 617 /* Figure C.2 - Generation of table of Huffman codes */ 618 k = 0; 619 code = 0; 620 si = huff->size[0]; 621 while (k < lastk) { 622 while (huff->size[k] == si) { 623 huff->code[k++] = code; 624 code++; 625 } 626 if (code >= (1 << si)) 627 goto out; 628 code <<= 1; 629 si++; 630 } 631 632 /* Figure C.3 - Ordering procedure for encoding procedure code tables */ 633 for (k = 0; k < lastk; k++) { 634 i = huffval[k]; 635 if (i >= maxsymbol || ehufsi[i]) 636 goto out; 637 ehufco[i] = huff->code[k]; 638 ehufsi[i] = huff->size[k]; 639 } 640 641 ret = 0; 642 out: 643 kfree(huff); 644 return ret; 645 } 646 647 #define DC_TABLE_INDEX0 0 648 #define AC_TABLE_INDEX0 1 649 #define DC_TABLE_INDEX1 2 650 #define AC_TABLE_INDEX1 3 651 652 static u8 *coda9_jpeg_get_huff_bits(struct coda_ctx *ctx, int tab_num) 653 { 654 struct coda_huff_tab *huff_tab = ctx->params.jpeg_huff_tab; 655 656 if (!huff_tab) 657 return NULL; 658 659 switch (tab_num) { 660 case DC_TABLE_INDEX0: return huff_tab->luma_dc; 661 case AC_TABLE_INDEX0: return huff_tab->luma_ac; 662 case DC_TABLE_INDEX1: return huff_tab->chroma_dc; 663 case AC_TABLE_INDEX1: return huff_tab->chroma_ac; 664 } 665 666 return NULL; 667 } 668 669 static int coda9_jpeg_gen_dec_huff_tab(struct coda_ctx *ctx, int tab_num) 670 { 671 int ptr_cnt = 0, huff_code = 0, zero_flag = 0, data_flag = 0; 672 u8 *huff_bits; 673 s16 *huff_max; 674 s16 *huff_min; 675 s8 *huff_ptr; 676 int ofs; 677 int i; 678 679 huff_bits = coda9_jpeg_get_huff_bits(ctx, tab_num); 680 if (!huff_bits) 681 return -EINVAL; 682 683 /* DC/AC Luma, DC/AC Chroma -> DC Luma/Chroma, AC Luma/Chroma */ 684 ofs = ((tab_num & 1) << 1) | ((tab_num >> 1) & 1); 685 ofs *= 16; 686 687 huff_ptr = ctx->params.jpeg_huff_tab->ptr + ofs; 688 huff_max = ctx->params.jpeg_huff_tab->max + ofs; 689 huff_min = ctx->params.jpeg_huff_tab->min + ofs; 690 691 for (i = 0; i < 16; i++) { 692 if (huff_bits[i]) { 693 huff_ptr[i] = ptr_cnt; 694 ptr_cnt += huff_bits[i]; 695 huff_min[i] = huff_code; 696 huff_max[i] = huff_code + (huff_bits[i] - 1); 697 data_flag = 1; 698 zero_flag = 0; 699 } else { 700 huff_ptr[i] = -1; 701 huff_min[i] = -1; 702 huff_max[i] = -1; 703 zero_flag = 1; 704 } 705 706 if (data_flag == 1) { 707 if (zero_flag == 1) 708 huff_code <<= 1; 709 else 710 huff_code = (huff_max[i] + 1) << 1; 711 } 712 } 713 714 return 0; 715 } 716 717 static int coda9_jpeg_load_huff_tab(struct coda_ctx *ctx) 718 { 719 struct { 720 int size[4][256]; 721 int code[4][256]; 722 } *huff; 723 u32 *huff_data; 724 int i, j; 725 int ret; 726 727 huff = kzalloc(sizeof(*huff), GFP_KERNEL); 728 if (!huff) 729 return -ENOMEM; 730 731 /* Generate all four (luma/chroma DC/AC) code/size lookup tables */ 732 for (i = 0; i < 4; i++) { 733 ret = coda9_jpeg_gen_enc_huff_tab(ctx, i, huff->size[i], 734 huff->code[i]); 735 if (ret) 736 goto out; 737 } 738 739 if (!ctx->params.jpeg_huff_data) { 740 ctx->params.jpeg_huff_data = 741 kzalloc(sizeof(u32) * CODA9_JPEG_ENC_HUFF_DATA_SIZE, 742 GFP_KERNEL); 743 if (!ctx->params.jpeg_huff_data) { 744 ret = -ENOMEM; 745 goto out; 746 } 747 } 748 huff_data = ctx->params.jpeg_huff_data; 749 750 for (j = 0; j < 4; j++) { 751 /* Store Huffman lookup tables in AC0, AC1, DC0, DC1 order */ 752 int t = (j == 0) ? AC_TABLE_INDEX0 : 753 (j == 1) ? AC_TABLE_INDEX1 : 754 (j == 2) ? DC_TABLE_INDEX0 : 755 DC_TABLE_INDEX1; 756 /* DC tables only have 16 entries */ 757 int len = (j < 2) ? 256 : 16; 758 759 for (i = 0; i < len; i++) { 760 if (huff->size[t][i] == 0 && huff->code[t][i] == 0) 761 *(huff_data++) = 0; 762 else 763 *(huff_data++) = 764 ((huff->size[t][i] - 1) << 16) | 765 huff->code[t][i]; 766 } 767 } 768 769 ret = 0; 770 out: 771 kfree(huff); 772 return ret; 773 } 774 775 static void coda9_jpeg_write_huff_tab(struct coda_ctx *ctx) 776 { 777 struct coda_dev *dev = ctx->dev; 778 u32 *huff_data = ctx->params.jpeg_huff_data; 779 int i; 780 781 /* Write Huffman size/code lookup tables in AC0, AC1, DC0, DC1 order */ 782 coda_write(dev, 0x3, CODA9_REG_JPEG_HUFF_CTRL); 783 for (i = 0; i < CODA9_JPEG_ENC_HUFF_DATA_SIZE; i++) 784 coda_write(dev, *(huff_data++), CODA9_REG_JPEG_HUFF_DATA); 785 coda_write(dev, 0x0, CODA9_REG_JPEG_HUFF_CTRL); 786 } 787 788 static inline void coda9_jpeg_write_qmat_quotients(struct coda_dev *dev, 789 u8 *qmat, int index) 790 { 791 int i; 792 793 coda_write(dev, index | 0x3, CODA9_REG_JPEG_QMAT_CTRL); 794 for (i = 0; i < 64; i++) 795 coda_write(dev, 0x80000 / qmat[i], CODA9_REG_JPEG_QMAT_DATA); 796 coda_write(dev, index, CODA9_REG_JPEG_QMAT_CTRL); 797 } 798 799 static void coda9_jpeg_load_qmat_tab(struct coda_ctx *ctx) 800 { 801 struct coda_dev *dev = ctx->dev; 802 u8 *luma_tab; 803 u8 *chroma_tab; 804 805 luma_tab = ctx->params.jpeg_qmat_tab[0]; 806 if (!luma_tab) 807 luma_tab = luma_q; 808 809 chroma_tab = ctx->params.jpeg_qmat_tab[1]; 810 if (!chroma_tab) 811 chroma_tab = chroma_q; 812 813 coda9_jpeg_write_qmat_quotients(dev, luma_tab, 0x00); 814 coda9_jpeg_write_qmat_quotients(dev, chroma_tab, 0x40); 815 coda9_jpeg_write_qmat_quotients(dev, chroma_tab, 0x80); 816 } 817 818 struct coda_jpeg_stream { 819 u8 *curr; 820 u8 *end; 821 }; 822 823 static inline int coda_jpeg_put_byte(u8 byte, struct coda_jpeg_stream *stream) 824 { 825 if (stream->curr >= stream->end) 826 return -EINVAL; 827 828 *stream->curr++ = byte; 829 830 return 0; 831 } 832 833 static inline int coda_jpeg_put_word(u16 word, struct coda_jpeg_stream *stream) 834 { 835 if (stream->curr + sizeof(__be16) > stream->end) 836 return -EINVAL; 837 838 put_unaligned_be16(word, stream->curr); 839 stream->curr += sizeof(__be16); 840 841 return 0; 842 } 843 844 static int coda_jpeg_put_table(u16 marker, u8 index, const u8 *table, 845 size_t len, struct coda_jpeg_stream *stream) 846 { 847 int i, ret; 848 849 ret = coda_jpeg_put_word(marker, stream); 850 if (ret < 0) 851 return ret; 852 ret = coda_jpeg_put_word(3 + len, stream); 853 if (ret < 0) 854 return ret; 855 ret = coda_jpeg_put_byte(index, stream); 856 for (i = 0; i < len && ret == 0; i++) 857 ret = coda_jpeg_put_byte(table[i], stream); 858 859 return ret; 860 } 861 862 static int coda_jpeg_define_quantization_table(struct coda_ctx *ctx, u8 index, 863 struct coda_jpeg_stream *stream) 864 { 865 return coda_jpeg_put_table(DQT_MARKER, index, 866 ctx->params.jpeg_qmat_tab[index], 64, 867 stream); 868 } 869 870 static int coda_jpeg_define_huffman_table(u8 index, const u8 *table, size_t len, 871 struct coda_jpeg_stream *stream) 872 { 873 return coda_jpeg_put_table(DHT_MARKER, index, table, len, stream); 874 } 875 876 static int coda9_jpeg_encode_header(struct coda_ctx *ctx, int len, u8 *buf) 877 { 878 struct coda_jpeg_stream stream = { buf, buf + len }; 879 struct coda_q_data *q_data_src; 880 int chroma_format, comp_num; 881 int i, ret, pad; 882 883 q_data_src = get_q_data(ctx, V4L2_BUF_TYPE_VIDEO_OUTPUT); 884 chroma_format = coda9_jpeg_chroma_format(q_data_src->fourcc); 885 if (chroma_format < 0) 886 return 0; 887 888 /* Start Of Image */ 889 ret = coda_jpeg_put_word(SOI_MARKER, &stream); 890 if (ret < 0) 891 return ret; 892 893 /* Define Restart Interval */ 894 if (ctx->params.jpeg_restart_interval) { 895 ret = coda_jpeg_put_word(DRI_MARKER, &stream); 896 if (ret < 0) 897 return ret; 898 ret = coda_jpeg_put_word(4, &stream); 899 if (ret < 0) 900 return ret; 901 ret = coda_jpeg_put_word(ctx->params.jpeg_restart_interval, 902 &stream); 903 if (ret < 0) 904 return ret; 905 } 906 907 /* Define Quantization Tables */ 908 ret = coda_jpeg_define_quantization_table(ctx, 0x00, &stream); 909 if (ret < 0) 910 return ret; 911 if (chroma_format != CODA9_JPEG_FORMAT_400) { 912 ret = coda_jpeg_define_quantization_table(ctx, 0x01, &stream); 913 if (ret < 0) 914 return ret; 915 } 916 917 /* Define Huffman Tables */ 918 ret = coda_jpeg_define_huffman_table(0x00, luma_dc, 16 + 12, &stream); 919 if (ret < 0) 920 return ret; 921 ret = coda_jpeg_define_huffman_table(0x10, luma_ac, 16 + 162, &stream); 922 if (ret < 0) 923 return ret; 924 if (chroma_format != CODA9_JPEG_FORMAT_400) { 925 ret = coda_jpeg_define_huffman_table(0x01, chroma_dc, 16 + 12, 926 &stream); 927 if (ret < 0) 928 return ret; 929 ret = coda_jpeg_define_huffman_table(0x11, chroma_ac, 16 + 162, 930 &stream); 931 if (ret < 0) 932 return ret; 933 } 934 935 /* Start Of Frame */ 936 ret = coda_jpeg_put_word(SOF_MARKER, &stream); 937 if (ret < 0) 938 return ret; 939 comp_num = (chroma_format == CODA9_JPEG_FORMAT_400) ? 1 : 3; 940 ret = coda_jpeg_put_word(8 + comp_num * 3, &stream); 941 if (ret < 0) 942 return ret; 943 ret = coda_jpeg_put_byte(0x08, &stream); 944 if (ret < 0) 945 return ret; 946 ret = coda_jpeg_put_word(q_data_src->height, &stream); 947 if (ret < 0) 948 return ret; 949 ret = coda_jpeg_put_word(q_data_src->width, &stream); 950 if (ret < 0) 951 return ret; 952 ret = coda_jpeg_put_byte(comp_num, &stream); 953 if (ret < 0) 954 return ret; 955 for (i = 0; i < comp_num; i++) { 956 static unsigned char subsampling[5][3] = { 957 [CODA9_JPEG_FORMAT_420] = { 0x22, 0x11, 0x11 }, 958 [CODA9_JPEG_FORMAT_422] = { 0x21, 0x11, 0x11 }, 959 [CODA9_JPEG_FORMAT_224] = { 0x12, 0x11, 0x11 }, 960 [CODA9_JPEG_FORMAT_444] = { 0x11, 0x11, 0x11 }, 961 [CODA9_JPEG_FORMAT_400] = { 0x11 }, 962 }; 963 964 /* Component identifier, matches SOS */ 965 ret = coda_jpeg_put_byte(i + 1, &stream); 966 if (ret < 0) 967 return ret; 968 ret = coda_jpeg_put_byte(subsampling[chroma_format][i], 969 &stream); 970 if (ret < 0) 971 return ret; 972 /* Chroma table index */ 973 ret = coda_jpeg_put_byte((i == 0) ? 0 : 1, &stream); 974 if (ret < 0) 975 return ret; 976 } 977 978 /* Pad to multiple of 8 bytes */ 979 pad = (stream.curr - buf) % 8; 980 if (pad) { 981 pad = 8 - pad; 982 while (pad--) { 983 ret = coda_jpeg_put_byte(0x00, &stream); 984 if (ret < 0) 985 return ret; 986 } 987 } 988 989 return stream.curr - buf; 990 } 991 992 /* 993 * Scale quantization table using nonlinear scaling factor 994 * u8 qtab[64], scale [50,190] 995 */ 996 static void coda_scale_quant_table(u8 *q_tab, int scale) 997 { 998 unsigned int temp; 999 int i; 1000 1001 for (i = 0; i < 64; i++) { 1002 temp = DIV_ROUND_CLOSEST((unsigned int)q_tab[i] * scale, 100); 1003 if (temp <= 0) 1004 temp = 1; 1005 if (temp > 255) 1006 temp = 255; 1007 q_tab[i] = (unsigned char)temp; 1008 } 1009 } 1010 1011 void coda_set_jpeg_compression_quality(struct coda_ctx *ctx, int quality) 1012 { 1013 unsigned int scale; 1014 1015 ctx->params.jpeg_quality = quality; 1016 1017 /* Clip quality setting to [5,100] interval */ 1018 if (quality > 100) 1019 quality = 100; 1020 if (quality < 5) 1021 quality = 5; 1022 1023 /* 1024 * Non-linear scaling factor: 1025 * [5,50] -> [1000..100], [51,100] -> [98..0] 1026 */ 1027 if (quality < 50) 1028 scale = 5000 / quality; 1029 else 1030 scale = 200 - 2 * quality; 1031 1032 if (ctx->params.jpeg_qmat_tab[0]) { 1033 memcpy(ctx->params.jpeg_qmat_tab[0], luma_q, 64); 1034 coda_scale_quant_table(ctx->params.jpeg_qmat_tab[0], scale); 1035 } 1036 if (ctx->params.jpeg_qmat_tab[1]) { 1037 memcpy(ctx->params.jpeg_qmat_tab[1], chroma_q, 64); 1038 coda_scale_quant_table(ctx->params.jpeg_qmat_tab[1], scale); 1039 } 1040 } 1041 1042 /* 1043 * Encoder context operations 1044 */ 1045 1046 static int coda9_jpeg_start_encoding(struct coda_ctx *ctx) 1047 { 1048 struct coda_dev *dev = ctx->dev; 1049 int ret; 1050 1051 ret = coda9_jpeg_load_huff_tab(ctx); 1052 if (ret < 0) { 1053 v4l2_err(&dev->v4l2_dev, "error loading Huffman tables\n"); 1054 return ret; 1055 } 1056 if (!ctx->params.jpeg_qmat_tab[0]) 1057 ctx->params.jpeg_qmat_tab[0] = kmalloc(64, GFP_KERNEL); 1058 if (!ctx->params.jpeg_qmat_tab[1]) 1059 ctx->params.jpeg_qmat_tab[1] = kmalloc(64, GFP_KERNEL); 1060 coda_set_jpeg_compression_quality(ctx, ctx->params.jpeg_quality); 1061 1062 return 0; 1063 } 1064 1065 static int coda9_jpeg_prepare_encode(struct coda_ctx *ctx) 1066 { 1067 struct coda_q_data *q_data_src; 1068 struct vb2_v4l2_buffer *src_buf, *dst_buf; 1069 struct coda_dev *dev = ctx->dev; 1070 u32 start_addr, end_addr; 1071 u16 aligned_width, aligned_height; 1072 bool chroma_interleave; 1073 int chroma_format; 1074 int header_len; 1075 int ret; 1076 ktime_t timeout; 1077 1078 src_buf = v4l2_m2m_next_src_buf(ctx->fh.m2m_ctx); 1079 dst_buf = v4l2_m2m_next_dst_buf(ctx->fh.m2m_ctx); 1080 q_data_src = get_q_data(ctx, V4L2_BUF_TYPE_VIDEO_OUTPUT); 1081 1082 if (vb2_get_plane_payload(&src_buf->vb2_buf, 0) == 0) 1083 vb2_set_plane_payload(&src_buf->vb2_buf, 0, 1084 vb2_plane_size(&src_buf->vb2_buf, 0)); 1085 1086 src_buf->sequence = ctx->osequence; 1087 dst_buf->sequence = ctx->osequence; 1088 ctx->osequence++; 1089 1090 src_buf->flags |= V4L2_BUF_FLAG_KEYFRAME; 1091 src_buf->flags &= ~V4L2_BUF_FLAG_PFRAME; 1092 1093 coda_set_gdi_regs(ctx); 1094 1095 start_addr = vb2_dma_contig_plane_dma_addr(&dst_buf->vb2_buf, 0); 1096 end_addr = start_addr + vb2_plane_size(&dst_buf->vb2_buf, 0); 1097 1098 chroma_format = coda9_jpeg_chroma_format(q_data_src->fourcc); 1099 if (chroma_format < 0) 1100 return chroma_format; 1101 1102 /* Round image dimensions to multiple of MCU size */ 1103 aligned_width = round_up(q_data_src->width, width_align[chroma_format]); 1104 aligned_height = round_up(q_data_src->height, 1105 height_align[chroma_format]); 1106 if (aligned_width != q_data_src->bytesperline) { 1107 v4l2_err(&dev->v4l2_dev, "wrong stride: %d instead of %d\n", 1108 aligned_width, q_data_src->bytesperline); 1109 } 1110 1111 header_len = 1112 coda9_jpeg_encode_header(ctx, 1113 vb2_plane_size(&dst_buf->vb2_buf, 0), 1114 vb2_plane_vaddr(&dst_buf->vb2_buf, 0)); 1115 if (header_len < 0) 1116 return header_len; 1117 1118 coda_write(dev, start_addr + header_len, CODA9_REG_JPEG_BBC_BAS_ADDR); 1119 coda_write(dev, end_addr, CODA9_REG_JPEG_BBC_END_ADDR); 1120 coda_write(dev, start_addr + header_len, CODA9_REG_JPEG_BBC_WR_PTR); 1121 coda_write(dev, start_addr + header_len, CODA9_REG_JPEG_BBC_RD_PTR); 1122 coda_write(dev, 0, CODA9_REG_JPEG_BBC_CUR_POS); 1123 /* 64 words per 256-byte page */ 1124 coda_write(dev, 64, CODA9_REG_JPEG_BBC_DATA_CNT); 1125 coda_write(dev, start_addr, CODA9_REG_JPEG_BBC_EXT_ADDR); 1126 coda_write(dev, 0, CODA9_REG_JPEG_BBC_INT_ADDR); 1127 1128 coda_write(dev, 0, CODA9_REG_JPEG_GBU_BT_PTR); 1129 coda_write(dev, 0, CODA9_REG_JPEG_GBU_WD_PTR); 1130 coda_write(dev, 0, CODA9_REG_JPEG_GBU_BBSR); 1131 coda_write(dev, BIT(31) | ((end_addr - start_addr - header_len) / 256), 1132 CODA9_REG_JPEG_BBC_STRM_CTRL); 1133 coda_write(dev, 0, CODA9_REG_JPEG_GBU_CTRL); 1134 coda_write(dev, 0, CODA9_REG_JPEG_GBU_FF_RPTR); 1135 coda_write(dev, 127, CODA9_REG_JPEG_GBU_BBER); 1136 coda_write(dev, 64, CODA9_REG_JPEG_GBU_BBIR); 1137 coda_write(dev, 64, CODA9_REG_JPEG_GBU_BBHR); 1138 1139 chroma_interleave = (q_data_src->fourcc == V4L2_PIX_FMT_NV12); 1140 coda_write(dev, CODA9_JPEG_PIC_CTRL_TC_DIRECTION | 1141 CODA9_JPEG_PIC_CTRL_ENCODER_EN, CODA9_REG_JPEG_PIC_CTRL); 1142 coda_write(dev, 0, CODA9_REG_JPEG_SCL_INFO); 1143 coda_write(dev, chroma_interleave, CODA9_REG_JPEG_DPB_CONFIG); 1144 coda_write(dev, ctx->params.jpeg_restart_interval, 1145 CODA9_REG_JPEG_RST_INTVAL); 1146 coda_write(dev, 1, CODA9_REG_JPEG_BBC_CTRL); 1147 1148 coda_write(dev, bus_req_num[chroma_format], CODA9_REG_JPEG_OP_INFO); 1149 1150 coda9_jpeg_write_huff_tab(ctx); 1151 coda9_jpeg_load_qmat_tab(ctx); 1152 1153 if (ctx->params.rot_mode & CODA_ROT_90) { 1154 aligned_width = aligned_height; 1155 aligned_height = q_data_src->bytesperline; 1156 if (chroma_format == CODA9_JPEG_FORMAT_422) 1157 chroma_format = CODA9_JPEG_FORMAT_224; 1158 else if (chroma_format == CODA9_JPEG_FORMAT_224) 1159 chroma_format = CODA9_JPEG_FORMAT_422; 1160 } 1161 /* These need to be multiples of MCU size */ 1162 coda_write(dev, aligned_width << 16 | aligned_height, 1163 CODA9_REG_JPEG_PIC_SIZE); 1164 coda_write(dev, ctx->params.rot_mode ? 1165 (CODA_ROT_MIR_ENABLE | ctx->params.rot_mode) : 0, 1166 CODA9_REG_JPEG_ROT_INFO); 1167 1168 coda_write(dev, mcu_info[chroma_format], CODA9_REG_JPEG_MCU_INFO); 1169 1170 coda_write(dev, 1, CODA9_GDI_CONTROL); 1171 timeout = ktime_add_us(ktime_get(), 100000); 1172 do { 1173 ret = coda_read(dev, CODA9_GDI_STATUS); 1174 if (ktime_compare(ktime_get(), timeout) > 0) { 1175 v4l2_err(&dev->v4l2_dev, "timeout waiting for GDI\n"); 1176 return -ETIMEDOUT; 1177 } 1178 } while (!ret); 1179 1180 coda_write(dev, (chroma_format << 17) | (chroma_interleave << 16) | 1181 q_data_src->bytesperline, CODA9_GDI_INFO_CONTROL); 1182 /* The content of this register seems to be irrelevant: */ 1183 coda_write(dev, aligned_width << 16 | aligned_height, 1184 CODA9_GDI_INFO_PIC_SIZE); 1185 1186 coda_write_base(ctx, q_data_src, src_buf, CODA9_GDI_INFO_BASE_Y); 1187 1188 coda_write(dev, 0, CODA9_REG_JPEG_DPB_BASE00); 1189 coda_write(dev, 0, CODA9_GDI_CONTROL); 1190 coda_write(dev, 1, CODA9_GDI_PIC_INIT_HOST); 1191 1192 coda_write(dev, 1, CODA9_GDI_WPROT_ERR_CLR); 1193 coda_write(dev, 0, CODA9_GDI_WPROT_RGN_EN); 1194 1195 trace_coda_jpeg_run(ctx, src_buf); 1196 1197 coda_write(dev, 1, CODA9_REG_JPEG_PIC_START); 1198 1199 return 0; 1200 } 1201 1202 static void coda9_jpeg_finish_encode(struct coda_ctx *ctx) 1203 { 1204 struct vb2_v4l2_buffer *src_buf, *dst_buf; 1205 struct coda_dev *dev = ctx->dev; 1206 u32 wr_ptr, start_ptr; 1207 u32 err_mb; 1208 1209 if (ctx->aborting) { 1210 coda_write(ctx->dev, 0, CODA9_REG_JPEG_BBC_FLUSH_CMD); 1211 return; 1212 } 1213 1214 /* 1215 * Lock to make sure that an encoder stop command running in parallel 1216 * will either already have marked src_buf as last, or it will wake up 1217 * the capture queue after the buffers are returned. 1218 */ 1219 mutex_lock(&ctx->wakeup_mutex); 1220 src_buf = v4l2_m2m_src_buf_remove(ctx->fh.m2m_ctx); 1221 dst_buf = v4l2_m2m_dst_buf_remove(ctx->fh.m2m_ctx); 1222 1223 trace_coda_jpeg_done(ctx, dst_buf); 1224 1225 /* 1226 * Set plane payload to the number of bytes written out 1227 * by the JPEG processing unit 1228 */ 1229 start_ptr = vb2_dma_contig_plane_dma_addr(&dst_buf->vb2_buf, 0); 1230 wr_ptr = coda_read(dev, CODA9_REG_JPEG_BBC_WR_PTR); 1231 vb2_set_plane_payload(&dst_buf->vb2_buf, 0, wr_ptr - start_ptr); 1232 1233 err_mb = coda_read(dev, CODA9_REG_JPEG_PIC_ERRMB); 1234 if (err_mb) 1235 coda_dbg(1, ctx, "ERRMB: 0x%x\n", err_mb); 1236 1237 coda_write(dev, 0, CODA9_REG_JPEG_BBC_FLUSH_CMD); 1238 1239 dst_buf->flags &= ~(V4L2_BUF_FLAG_PFRAME | V4L2_BUF_FLAG_LAST); 1240 dst_buf->flags |= V4L2_BUF_FLAG_KEYFRAME; 1241 dst_buf->flags |= src_buf->flags & V4L2_BUF_FLAG_LAST; 1242 1243 v4l2_m2m_buf_copy_metadata(src_buf, dst_buf, false); 1244 1245 v4l2_m2m_buf_done(src_buf, VB2_BUF_STATE_DONE); 1246 coda_m2m_buf_done(ctx, dst_buf, err_mb ? VB2_BUF_STATE_ERROR : 1247 VB2_BUF_STATE_DONE); 1248 mutex_unlock(&ctx->wakeup_mutex); 1249 1250 coda_dbg(1, ctx, "job finished: encoded frame (%u)%s\n", 1251 dst_buf->sequence, 1252 (dst_buf->flags & V4L2_BUF_FLAG_LAST) ? " (last)" : ""); 1253 1254 /* 1255 * Reset JPEG processing unit after each encode run to work 1256 * around hangups when switching context between encoder and 1257 * decoder. 1258 */ 1259 coda_hw_reset(ctx); 1260 } 1261 1262 static void coda9_jpeg_encode_timeout(struct coda_ctx *ctx) 1263 { 1264 struct coda_dev *dev = ctx->dev; 1265 u32 end_addr, wr_ptr; 1266 1267 /* Handle missing BBC overflow interrupt via timeout */ 1268 end_addr = coda_read(dev, CODA9_REG_JPEG_BBC_END_ADDR); 1269 wr_ptr = coda_read(dev, CODA9_REG_JPEG_BBC_WR_PTR); 1270 if (wr_ptr >= end_addr - 256) { 1271 v4l2_err(&dev->v4l2_dev, "JPEG too large for capture buffer\n"); 1272 coda9_jpeg_finish_encode(ctx); 1273 return; 1274 } 1275 1276 coda_hw_reset(ctx); 1277 } 1278 1279 static void coda9_jpeg_release(struct coda_ctx *ctx) 1280 { 1281 int i; 1282 1283 if (ctx->params.jpeg_qmat_tab[0] == luma_q) 1284 ctx->params.jpeg_qmat_tab[0] = NULL; 1285 if (ctx->params.jpeg_qmat_tab[1] == chroma_q) 1286 ctx->params.jpeg_qmat_tab[1] = NULL; 1287 for (i = 0; i < 3; i++) 1288 kfree(ctx->params.jpeg_qmat_tab[i]); 1289 kfree(ctx->params.jpeg_huff_data); 1290 kfree(ctx->params.jpeg_huff_tab); 1291 } 1292 1293 const struct coda_context_ops coda9_jpeg_encode_ops = { 1294 .queue_init = coda_encoder_queue_init, 1295 .start_streaming = coda9_jpeg_start_encoding, 1296 .prepare_run = coda9_jpeg_prepare_encode, 1297 .finish_run = coda9_jpeg_finish_encode, 1298 .run_timeout = coda9_jpeg_encode_timeout, 1299 .release = coda9_jpeg_release, 1300 }; 1301 1302 /* 1303 * Decoder context operations 1304 */ 1305 1306 static int coda9_jpeg_start_decoding(struct coda_ctx *ctx) 1307 { 1308 ctx->params.jpeg_qmat_index[0] = 0; 1309 ctx->params.jpeg_qmat_index[1] = 1; 1310 ctx->params.jpeg_qmat_index[2] = 1; 1311 ctx->params.jpeg_qmat_tab[0] = luma_q; 1312 ctx->params.jpeg_qmat_tab[1] = chroma_q; 1313 /* nothing more to do here */ 1314 1315 /* TODO: we could already scan the first header to get the chroma 1316 * format. 1317 */ 1318 1319 return 0; 1320 } 1321 1322 static int coda9_jpeg_prepare_decode(struct coda_ctx *ctx) 1323 { 1324 struct coda_dev *dev = ctx->dev; 1325 int aligned_width, aligned_height; 1326 int chroma_format; 1327 int ret; 1328 u32 val, dst_fourcc; 1329 struct coda_q_data *q_data_src, *q_data_dst; 1330 struct vb2_v4l2_buffer *src_buf, *dst_buf; 1331 int chroma_interleave; 1332 int scl_hor_mode, scl_ver_mode; 1333 1334 src_buf = v4l2_m2m_next_src_buf(ctx->fh.m2m_ctx); 1335 dst_buf = v4l2_m2m_next_dst_buf(ctx->fh.m2m_ctx); 1336 q_data_src = get_q_data(ctx, V4L2_BUF_TYPE_VIDEO_OUTPUT); 1337 q_data_dst = get_q_data(ctx, V4L2_BUF_TYPE_VIDEO_CAPTURE); 1338 dst_fourcc = q_data_dst->fourcc; 1339 1340 scl_hor_mode = coda_jpeg_scale(q_data_src->width, q_data_dst->width); 1341 scl_ver_mode = coda_jpeg_scale(q_data_src->height, q_data_dst->height); 1342 1343 if (vb2_get_plane_payload(&src_buf->vb2_buf, 0) == 0) 1344 vb2_set_plane_payload(&src_buf->vb2_buf, 0, 1345 vb2_plane_size(&src_buf->vb2_buf, 0)); 1346 1347 chroma_format = coda9_jpeg_chroma_format(q_data_dst->fourcc); 1348 if (chroma_format < 0) 1349 return chroma_format; 1350 1351 ret = coda_jpeg_decode_header(ctx, &src_buf->vb2_buf); 1352 if (ret < 0) { 1353 src_buf = v4l2_m2m_src_buf_remove(ctx->fh.m2m_ctx); 1354 dst_buf = v4l2_m2m_dst_buf_remove(ctx->fh.m2m_ctx); 1355 v4l2_m2m_buf_done(src_buf, VB2_BUF_STATE_DONE); 1356 v4l2_m2m_buf_done(dst_buf, VB2_BUF_STATE_ERROR); 1357 1358 return ret; 1359 } 1360 1361 /* Round image dimensions to multiple of MCU size */ 1362 aligned_width = round_up(q_data_src->width, width_align[chroma_format]); 1363 aligned_height = round_up(q_data_src->height, height_align[chroma_format]); 1364 if (aligned_width != q_data_dst->bytesperline) { 1365 v4l2_err(&dev->v4l2_dev, "stride mismatch: %d != %d\n", 1366 aligned_width, q_data_dst->bytesperline); 1367 } 1368 1369 coda_set_gdi_regs(ctx); 1370 1371 val = ctx->params.jpeg_huff_ac_index[0] << 12 | 1372 ctx->params.jpeg_huff_ac_index[1] << 11 | 1373 ctx->params.jpeg_huff_ac_index[2] << 10 | 1374 ctx->params.jpeg_huff_dc_index[0] << 9 | 1375 ctx->params.jpeg_huff_dc_index[1] << 8 | 1376 ctx->params.jpeg_huff_dc_index[2] << 7; 1377 if (ctx->params.jpeg_huff_tab) 1378 val |= CODA9_JPEG_PIC_CTRL_USER_HUFFMAN_EN; 1379 coda_write(dev, val, CODA9_REG_JPEG_PIC_CTRL); 1380 1381 coda_write(dev, aligned_width << 16 | aligned_height, 1382 CODA9_REG_JPEG_PIC_SIZE); 1383 1384 chroma_interleave = (dst_fourcc == V4L2_PIX_FMT_NV12); 1385 coda_write(dev, 0, CODA9_REG_JPEG_ROT_INFO); 1386 coda_write(dev, bus_req_num[chroma_format], CODA9_REG_JPEG_OP_INFO); 1387 coda_write(dev, mcu_info[chroma_format], CODA9_REG_JPEG_MCU_INFO); 1388 if (scl_hor_mode || scl_ver_mode) 1389 val = CODA9_JPEG_SCL_ENABLE | (scl_hor_mode << 2) | scl_ver_mode; 1390 else 1391 val = 0; 1392 coda_write(dev, val, CODA9_REG_JPEG_SCL_INFO); 1393 coda_write(dev, chroma_interleave, CODA9_REG_JPEG_DPB_CONFIG); 1394 coda_write(dev, ctx->params.jpeg_restart_interval, 1395 CODA9_REG_JPEG_RST_INTVAL); 1396 1397 if (ctx->params.jpeg_huff_tab) { 1398 ret = coda9_jpeg_dec_huff_setup(ctx); 1399 if (ret < 0) { 1400 v4l2_err(&dev->v4l2_dev, 1401 "failed to set up Huffman tables: %d\n", ret); 1402 return ret; 1403 } 1404 } 1405 1406 coda9_jpeg_qmat_setup(ctx); 1407 1408 coda9_jpeg_dec_bbc_gbu_setup(ctx, &src_buf->vb2_buf, 1409 ctx->jpeg_ecs_offset); 1410 1411 coda_write(dev, 0, CODA9_REG_JPEG_RST_INDEX); 1412 coda_write(dev, 0, CODA9_REG_JPEG_RST_COUNT); 1413 1414 coda_write(dev, 0, CODA9_REG_JPEG_DPCM_DIFF_Y); 1415 coda_write(dev, 0, CODA9_REG_JPEG_DPCM_DIFF_CB); 1416 coda_write(dev, 0, CODA9_REG_JPEG_DPCM_DIFF_CR); 1417 1418 coda_write(dev, 0, CODA9_REG_JPEG_ROT_INFO); 1419 1420 coda_write(dev, 1, CODA9_GDI_CONTROL); 1421 do { 1422 ret = coda_read(dev, CODA9_GDI_STATUS); 1423 } while (!ret); 1424 1425 val = (chroma_format << 17) | (chroma_interleave << 16) | 1426 q_data_dst->bytesperline; 1427 if (ctx->tiled_map_type == GDI_TILED_FRAME_MB_RASTER_MAP) 1428 val |= 3 << 20; 1429 coda_write(dev, val, CODA9_GDI_INFO_CONTROL); 1430 1431 coda_write(dev, aligned_width << 16 | aligned_height, 1432 CODA9_GDI_INFO_PIC_SIZE); 1433 1434 coda_write_base(ctx, q_data_dst, dst_buf, CODA9_GDI_INFO_BASE_Y); 1435 1436 coda_write(dev, 0, CODA9_REG_JPEG_DPB_BASE00); 1437 coda_write(dev, 0, CODA9_GDI_CONTROL); 1438 coda_write(dev, 1, CODA9_GDI_PIC_INIT_HOST); 1439 1440 trace_coda_jpeg_run(ctx, src_buf); 1441 1442 coda_write(dev, 1, CODA9_REG_JPEG_PIC_START); 1443 1444 return 0; 1445 } 1446 1447 static void coda9_jpeg_finish_decode(struct coda_ctx *ctx) 1448 { 1449 struct coda_dev *dev = ctx->dev; 1450 struct vb2_v4l2_buffer *dst_buf, *src_buf; 1451 struct coda_q_data *q_data_dst; 1452 u32 err_mb; 1453 1454 err_mb = coda_read(dev, CODA9_REG_JPEG_PIC_ERRMB); 1455 if (err_mb) 1456 v4l2_err(&dev->v4l2_dev, "ERRMB: 0x%x\n", err_mb); 1457 1458 coda_write(dev, 0, CODA9_REG_JPEG_BBC_FLUSH_CMD); 1459 1460 /* 1461 * Lock to make sure that a decoder stop command running in parallel 1462 * will either already have marked src_buf as last, or it will wake up 1463 * the capture queue after the buffers are returned. 1464 */ 1465 mutex_lock(&ctx->wakeup_mutex); 1466 src_buf = v4l2_m2m_src_buf_remove(ctx->fh.m2m_ctx); 1467 dst_buf = v4l2_m2m_dst_buf_remove(ctx->fh.m2m_ctx); 1468 dst_buf->sequence = ctx->osequence++; 1469 1470 trace_coda_jpeg_done(ctx, dst_buf); 1471 1472 dst_buf->flags &= ~(V4L2_BUF_FLAG_PFRAME | V4L2_BUF_FLAG_LAST); 1473 dst_buf->flags |= V4L2_BUF_FLAG_KEYFRAME; 1474 dst_buf->flags |= src_buf->flags & V4L2_BUF_FLAG_LAST; 1475 1476 v4l2_m2m_buf_copy_metadata(src_buf, dst_buf, false); 1477 1478 q_data_dst = get_q_data(ctx, V4L2_BUF_TYPE_VIDEO_CAPTURE); 1479 vb2_set_plane_payload(&dst_buf->vb2_buf, 0, q_data_dst->sizeimage); 1480 1481 v4l2_m2m_buf_done(src_buf, VB2_BUF_STATE_DONE); 1482 coda_m2m_buf_done(ctx, dst_buf, err_mb ? VB2_BUF_STATE_ERROR : 1483 VB2_BUF_STATE_DONE); 1484 1485 mutex_unlock(&ctx->wakeup_mutex); 1486 1487 coda_dbg(1, ctx, "job finished: decoded frame (%u)%s\n", 1488 dst_buf->sequence, 1489 (dst_buf->flags & V4L2_BUF_FLAG_LAST) ? " (last)" : ""); 1490 1491 /* 1492 * Reset JPEG processing unit after each decode run to work 1493 * around hangups when switching context between encoder and 1494 * decoder. 1495 */ 1496 coda_hw_reset(ctx); 1497 } 1498 1499 const struct coda_context_ops coda9_jpeg_decode_ops = { 1500 .queue_init = coda_encoder_queue_init, /* non-bitstream operation */ 1501 .start_streaming = coda9_jpeg_start_decoding, 1502 .prepare_run = coda9_jpeg_prepare_decode, 1503 .finish_run = coda9_jpeg_finish_decode, 1504 .release = coda9_jpeg_release, 1505 }; 1506 1507 irqreturn_t coda9_jpeg_irq_handler(int irq, void *data) 1508 { 1509 struct coda_dev *dev = data; 1510 struct coda_ctx *ctx; 1511 int status; 1512 int err_mb; 1513 1514 status = coda_read(dev, CODA9_REG_JPEG_PIC_STATUS); 1515 if (status == 0) 1516 return IRQ_HANDLED; 1517 coda_write(dev, status, CODA9_REG_JPEG_PIC_STATUS); 1518 1519 if (status & CODA9_JPEG_STATUS_OVERFLOW) 1520 v4l2_err(&dev->v4l2_dev, "JPEG overflow\n"); 1521 1522 if (status & CODA9_JPEG_STATUS_BBC_INT) 1523 v4l2_err(&dev->v4l2_dev, "JPEG BBC interrupt\n"); 1524 1525 if (status & CODA9_JPEG_STATUS_ERROR) { 1526 v4l2_err(&dev->v4l2_dev, "JPEG error\n"); 1527 1528 err_mb = coda_read(dev, CODA9_REG_JPEG_PIC_ERRMB); 1529 if (err_mb) { 1530 v4l2_err(&dev->v4l2_dev, 1531 "ERRMB: 0x%x: rst idx %d, mcu pos (%d,%d)\n", 1532 err_mb, err_mb >> 24, (err_mb >> 12) & 0xfff, 1533 err_mb & 0xfff); 1534 } 1535 } 1536 1537 ctx = v4l2_m2m_get_curr_priv(dev->m2m_dev); 1538 if (!ctx) { 1539 v4l2_err(&dev->v4l2_dev, 1540 "Instance released before the end of transaction\n"); 1541 mutex_unlock(&dev->coda_mutex); 1542 return IRQ_HANDLED; 1543 } 1544 1545 complete(&ctx->completion); 1546 1547 return IRQ_HANDLED; 1548 } 1549