Shrenuj Bansal | a419c79 | 2016-10-20 14:05:11 -0700 | [diff] [blame] | 1 | /* Copyright (c) 2013-2017, The Linux Foundation. All rights reserved. |
| 2 | * |
| 3 | * This program is free software; you can redistribute it and/or modify |
| 4 | * it under the terms of the GNU General Public License version 2 and |
| 5 | * only version 2 as published by the Free Software Foundation. |
| 6 | * |
| 7 | * This program is distributed in the hope that it will be useful, |
| 8 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 9 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 10 | * GNU General Public License for more details. |
| 11 | */ |
| 12 | |
| 13 | #include "kgsl.h" |
| 14 | #include "kgsl_sharedmem.h" |
| 15 | #include "kgsl_snapshot.h" |
| 16 | |
| 17 | #include "adreno.h" |
| 18 | #include "adreno_pm4types.h" |
| 19 | #include "a3xx_reg.h" |
| 20 | #include "adreno_cp_parser.h" |
| 21 | |
| 22 | #define MAX_IB_OBJS 1000 |
| 23 | #define NUM_SET_DRAW_GROUPS 32 |
| 24 | |
| 25 | struct set_draw_state { |
| 26 | uint64_t cmd_stream_addr; |
| 27 | uint64_t cmd_stream_dwords; |
| 28 | }; |
| 29 | |
| 30 | /* List of variables used when parsing an IB */ |
| 31 | struct ib_parser_variables { |
| 32 | /* List of registers containing addresses and their sizes */ |
| 33 | unsigned int cp_addr_regs[ADRENO_CP_ADDR_MAX]; |
| 34 | /* 32 groups of command streams in set draw state packets */ |
| 35 | struct set_draw_state set_draw_groups[NUM_SET_DRAW_GROUPS]; |
| 36 | }; |
| 37 | |
| 38 | /* |
| 39 | * Used for locating shader objects. This array holds the unit size of shader |
| 40 | * objects based on type and block of shader. The type can be 0 or 1 hence there |
| 41 | * are 2 columns and block can be 0-7 hence 7 rows. |
| 42 | */ |
| 43 | static int load_state_unit_sizes[7][2] = { |
| 44 | { 2, 4 }, |
| 45 | { 0, 1 }, |
| 46 | { 2, 4 }, |
| 47 | { 0, 1 }, |
| 48 | { 8, 2 }, |
| 49 | { 8, 2 }, |
| 50 | { 8, 2 }, |
| 51 | }; |
| 52 | |
| 53 | static int adreno_ib_find_objs(struct kgsl_device *device, |
| 54 | struct kgsl_process_private *process, |
| 55 | uint64_t gpuaddr, uint64_t dwords, |
Hareesh Gundu | 9c6b1fa | 2017-01-06 15:37:09 +0530 | [diff] [blame] | 56 | uint64_t ib2base, int obj_type, |
Shrenuj Bansal | a419c79 | 2016-10-20 14:05:11 -0700 | [diff] [blame] | 57 | struct adreno_ib_object_list *ib_obj_list, |
| 58 | int ib_level); |
| 59 | |
| 60 | static int ib_parse_set_draw_state(struct kgsl_device *device, |
| 61 | unsigned int *ptr, |
| 62 | struct kgsl_process_private *process, |
| 63 | struct adreno_ib_object_list *ib_obj_list, |
| 64 | struct ib_parser_variables *ib_parse_vars); |
| 65 | |
| 66 | static int ib_parse_type7_set_draw_state(struct kgsl_device *device, |
| 67 | unsigned int *ptr, |
| 68 | struct kgsl_process_private *process, |
| 69 | struct adreno_ib_object_list *ib_obj_list); |
| 70 | |
| 71 | /* |
| 72 | * adreno_ib_merge_range() - Increases the address range tracked by an ib |
| 73 | * object |
| 74 | * @ib_obj: The ib object |
| 75 | * @gpuaddr: The start address which is to be merged |
| 76 | * @size: Size of the merging address |
| 77 | */ |
| 78 | static void adreno_ib_merge_range(struct adreno_ib_object *ib_obj, |
| 79 | uint64_t gpuaddr, uint64_t size) |
| 80 | { |
| 81 | uint64_t addr_end1 = ib_obj->gpuaddr + ib_obj->size; |
| 82 | uint64_t addr_end2 = gpuaddr + size; |
| 83 | |
| 84 | if (gpuaddr < ib_obj->gpuaddr) |
| 85 | ib_obj->gpuaddr = gpuaddr; |
| 86 | if (addr_end2 > addr_end1) |
| 87 | ib_obj->size = addr_end2 - ib_obj->gpuaddr; |
| 88 | else |
| 89 | ib_obj->size = addr_end1 - ib_obj->gpuaddr; |
| 90 | } |
| 91 | |
| 92 | /* |
| 93 | * adreno_ib_check_overlap() - Checks if an address range overlap |
| 94 | * @gpuaddr: The start address range to check for overlap |
| 95 | * @size: Size of the address range |
| 96 | * @type: The type of address range |
| 97 | * @ib_obj_list: The list of address ranges to check for overlap |
| 98 | * |
| 99 | * Checks if an address range overlaps with a list of address ranges |
| 100 | * Returns the entry from list which overlaps else NULL |
| 101 | */ |
| 102 | static struct adreno_ib_object *adreno_ib_check_overlap(uint64_t gpuaddr, |
| 103 | uint64_t size, int type, |
| 104 | struct adreno_ib_object_list *ib_obj_list) |
| 105 | { |
| 106 | struct adreno_ib_object *ib_obj; |
| 107 | int i; |
| 108 | |
| 109 | for (i = 0; i < ib_obj_list->num_objs; i++) { |
| 110 | ib_obj = &(ib_obj_list->obj_list[i]); |
| 111 | if ((type == ib_obj->snapshot_obj_type) && |
| 112 | kgsl_addr_range_overlap(ib_obj->gpuaddr, ib_obj->size, |
| 113 | gpuaddr, size)) |
| 114 | /* regions overlap */ |
| 115 | return ib_obj; |
| 116 | } |
| 117 | return NULL; |
| 118 | } |
| 119 | |
| 120 | /* |
| 121 | * adreno_ib_add() - Add a gpuaddress range to list |
| 122 | * @process: Process in which the gpuaddress is mapped |
| 123 | * @type: The type of address range |
| 124 | * @ib_obj_list: List of the address ranges in which the given range is to be |
| 125 | * added |
| 126 | * |
| 127 | * Add a gpuaddress range as an ib object to a given list after checking if it |
| 128 | * overlaps with another entry on the list. If it conflicts then change the |
| 129 | * existing entry to incorporate this range |
| 130 | * |
| 131 | * Returns 0 on success else error code |
| 132 | */ |
| 133 | static int adreno_ib_add(struct kgsl_process_private *process, |
| 134 | uint64_t gpuaddr, int type, |
| 135 | struct adreno_ib_object_list *ib_obj_list) |
| 136 | { |
| 137 | uint64_t size; |
| 138 | struct adreno_ib_object *ib_obj; |
| 139 | struct kgsl_mem_entry *entry; |
| 140 | |
| 141 | if (ib_obj_list->num_objs >= MAX_IB_OBJS) |
| 142 | return -E2BIG; |
| 143 | |
| 144 | entry = kgsl_sharedmem_find(process, gpuaddr); |
| 145 | if (!entry) |
| 146 | /* |
| 147 | * Do not fail if gpuaddr not found, we can continue |
| 148 | * to search for other objects even if few objects are |
| 149 | * not found |
| 150 | */ |
| 151 | return 0; |
| 152 | |
| 153 | size = entry->memdesc.size; |
| 154 | gpuaddr = entry->memdesc.gpuaddr; |
| 155 | |
| 156 | ib_obj = adreno_ib_check_overlap(gpuaddr, size, type, ib_obj_list); |
| 157 | if (ib_obj) { |
| 158 | adreno_ib_merge_range(ib_obj, gpuaddr, size); |
| 159 | kgsl_mem_entry_put(entry); |
| 160 | } else { |
| 161 | adreno_ib_init_ib_obj(gpuaddr, size, type, entry, |
| 162 | &(ib_obj_list->obj_list[ib_obj_list->num_objs])); |
| 163 | ib_obj_list->num_objs++; |
| 164 | } |
| 165 | return 0; |
| 166 | } |
| 167 | |
| 168 | /* |
| 169 | * ib_save_mip_addresses() - Find mip addresses |
| 170 | * @pkt: Pointer to the packet in IB |
| 171 | * @process: The process in which IB is mapped |
| 172 | * @ib_obj_list: List in which any objects found are added |
| 173 | * |
| 174 | * Returns 0 on success else error code |
| 175 | */ |
| 176 | static int ib_save_mip_addresses(unsigned int *pkt, |
| 177 | struct kgsl_process_private *process, |
| 178 | struct adreno_ib_object_list *ib_obj_list) |
| 179 | { |
| 180 | int ret = 0; |
| 181 | int num_levels = (pkt[1] >> 22) & 0x03FF; |
| 182 | int i; |
| 183 | unsigned int *hostptr; |
| 184 | struct kgsl_mem_entry *ent; |
| 185 | unsigned int block, type; |
| 186 | int unitsize = 0; |
| 187 | |
| 188 | block = (pkt[1] >> 19) & 0x07; |
| 189 | type = pkt[2] & 0x03; |
| 190 | |
| 191 | if (type == 0) |
| 192 | unitsize = load_state_unit_sizes[block][0]; |
| 193 | else |
| 194 | unitsize = load_state_unit_sizes[block][1]; |
| 195 | |
| 196 | if (3 == block && 1 == type) { |
| 197 | uint64_t gpuaddr = pkt[2] & 0xFFFFFFFC; |
| 198 | uint64_t size = (num_levels * unitsize) << 2; |
| 199 | |
| 200 | ent = kgsl_sharedmem_find(process, gpuaddr); |
| 201 | if (ent == NULL) |
| 202 | return 0; |
| 203 | |
| 204 | if (!kgsl_gpuaddr_in_memdesc(&ent->memdesc, |
| 205 | gpuaddr, size)) { |
| 206 | kgsl_mem_entry_put(ent); |
| 207 | return 0; |
| 208 | } |
| 209 | |
| 210 | hostptr = kgsl_gpuaddr_to_vaddr(&ent->memdesc, gpuaddr); |
| 211 | if (hostptr != NULL) { |
| 212 | for (i = 0; i < num_levels; i++) { |
| 213 | ret = adreno_ib_add(process, hostptr[i], |
| 214 | SNAPSHOT_GPU_OBJECT_GENERIC, |
| 215 | ib_obj_list); |
| 216 | if (ret) |
| 217 | break; |
| 218 | } |
| 219 | } |
| 220 | |
| 221 | kgsl_memdesc_unmap(&ent->memdesc); |
| 222 | kgsl_mem_entry_put(ent); |
| 223 | } |
| 224 | return ret; |
| 225 | } |
| 226 | |
| 227 | /* |
| 228 | * ib_parse_load_state() - Parse load state packet |
| 229 | * @pkt: Pointer to the packet in IB |
| 230 | * @process: The pagetable in which the IB is mapped |
| 231 | * @ib_obj_list: List in which any objects found are added |
| 232 | * @ib_parse_vars: VAriable list that store temporary addressses |
| 233 | * |
| 234 | * Parse load state packet found in an IB and add any memory object found to |
| 235 | * a list |
| 236 | * Returns 0 on success else error code |
| 237 | */ |
| 238 | static int ib_parse_load_state(unsigned int *pkt, |
| 239 | struct kgsl_process_private *process, |
| 240 | struct adreno_ib_object_list *ib_obj_list, |
| 241 | struct ib_parser_variables *ib_parse_vars) |
| 242 | { |
| 243 | int ret = 0; |
| 244 | int i; |
| 245 | |
| 246 | /* |
| 247 | * The object here is to find indirect shaders i.e - shaders loaded from |
| 248 | * GPU memory instead of directly in the command. These should be added |
| 249 | * to the list of memory objects to dump. So look at the load state |
| 250 | * if the block is indirect (source = 4). If so then add the memory |
| 251 | * address to the list. The size of the object differs depending on the |
| 252 | * type per the load_state_unit_sizes array above. |
| 253 | */ |
| 254 | |
| 255 | if (type3_pkt_size(pkt[0]) < 2) |
| 256 | return 0; |
| 257 | |
| 258 | /* |
| 259 | * Anything from 3rd ordinal onwards of packet can be a memory object, |
| 260 | * no need to be fancy about parsing it, just save it if it looks |
| 261 | * like memory |
| 262 | */ |
| 263 | for (i = 0; i <= (type3_pkt_size(pkt[0]) - 2); i++) { |
| 264 | ret |= adreno_ib_add(process, pkt[2 + i] & 0xFFFFFFFC, |
| 265 | SNAPSHOT_GPU_OBJECT_GENERIC, |
| 266 | ib_obj_list); |
| 267 | if (ret) |
| 268 | break; |
| 269 | } |
| 270 | /* get the mip addresses */ |
| 271 | if (!ret) |
| 272 | ret = ib_save_mip_addresses(pkt, process, ib_obj_list); |
| 273 | return ret; |
| 274 | } |
| 275 | |
| 276 | /* |
| 277 | * This opcode sets the base addresses for the visibilty stream buffer and the |
| 278 | * visiblity stream size buffer. |
| 279 | */ |
| 280 | |
| 281 | static int ib_parse_set_bin_data(unsigned int *pkt, |
| 282 | struct kgsl_process_private *process, |
| 283 | struct adreno_ib_object_list *ib_obj_list, |
| 284 | struct ib_parser_variables *ib_parse_vars) |
| 285 | { |
| 286 | int ret = 0; |
| 287 | |
| 288 | if (type3_pkt_size(pkt[0]) < 2) |
| 289 | return 0; |
| 290 | |
| 291 | /* Visiblity stream buffer */ |
| 292 | ret = adreno_ib_add(process, pkt[1], |
| 293 | SNAPSHOT_GPU_OBJECT_GENERIC, ib_obj_list); |
| 294 | if (ret) |
| 295 | return ret; |
| 296 | |
| 297 | /* visiblity stream size buffer (fixed size 8 dwords) */ |
| 298 | ret = adreno_ib_add(process, pkt[2], |
| 299 | SNAPSHOT_GPU_OBJECT_GENERIC, ib_obj_list); |
| 300 | |
| 301 | return ret; |
| 302 | } |
| 303 | |
| 304 | /* |
| 305 | * This opcode writes to GPU memory - if the buffer is written to, there is a |
| 306 | * good chance that it would be valuable to capture in the snapshot, so mark all |
| 307 | * buffers that are written to as frozen |
| 308 | */ |
| 309 | |
| 310 | static int ib_parse_mem_write(unsigned int *pkt, |
| 311 | struct kgsl_process_private *process, |
| 312 | struct adreno_ib_object_list *ib_obj_list, |
| 313 | struct ib_parser_variables *ib_parse_vars) |
| 314 | { |
| 315 | if (type3_pkt_size(pkt[0]) < 1) |
| 316 | return 0; |
| 317 | |
| 318 | /* |
| 319 | * The address is where the data in the rest of this packet is written |
| 320 | * to, but since that might be an offset into the larger buffer we need |
| 321 | * to get the whole thing. Pass a size of 0 tocapture the entire buffer. |
| 322 | */ |
| 323 | |
| 324 | return adreno_ib_add(process, pkt[1] & 0xFFFFFFFC, |
| 325 | SNAPSHOT_GPU_OBJECT_GENERIC, ib_obj_list); |
| 326 | } |
| 327 | |
| 328 | /* |
| 329 | * ib_add_type0_entries() - Add memory objects to list |
| 330 | * @device: The device on which the IB will execute |
| 331 | * @process: The process in which IB is mapped |
| 332 | * @ib_obj_list: The list of gpu objects |
| 333 | * @ib_parse_vars: addresses ranges found in type0 packets |
| 334 | * |
| 335 | * Add memory objects to given list that are found in type0 packets |
| 336 | * Returns 0 on success else 0 |
| 337 | */ |
| 338 | static int ib_add_type0_entries(struct kgsl_device *device, |
| 339 | struct kgsl_process_private *process, |
| 340 | struct adreno_ib_object_list *ib_obj_list, |
| 341 | struct ib_parser_variables *ib_parse_vars) |
| 342 | { |
| 343 | struct adreno_device *adreno_dev = ADRENO_DEVICE(device); |
| 344 | int ret = 0; |
| 345 | int i; |
| 346 | int vfd_end; |
| 347 | unsigned int mask; |
| 348 | /* First up the visiblity stream buffer */ |
| 349 | if (adreno_is_a4xx(adreno_dev)) |
| 350 | mask = 0xFFFFFFFC; |
| 351 | else |
| 352 | mask = 0xFFFFFFFF; |
| 353 | for (i = ADRENO_CP_ADDR_VSC_PIPE_DATA_ADDRESS_0; |
| 354 | i < ADRENO_CP_ADDR_VSC_PIPE_DATA_LENGTH_7; i++) { |
| 355 | if (ib_parse_vars->cp_addr_regs[i]) { |
| 356 | ret = adreno_ib_add(process, |
| 357 | ib_parse_vars->cp_addr_regs[i] & mask, |
| 358 | SNAPSHOT_GPU_OBJECT_GENERIC, |
| 359 | ib_obj_list); |
| 360 | if (ret) |
| 361 | return ret; |
| 362 | ib_parse_vars->cp_addr_regs[i] = 0; |
| 363 | ib_parse_vars->cp_addr_regs[i + 1] = 0; |
| 364 | i++; |
| 365 | } |
| 366 | } |
| 367 | |
| 368 | vfd_end = adreno_is_a4xx(adreno_dev) ? |
| 369 | ADRENO_CP_ADDR_VFD_FETCH_INSTR_1_31 : |
| 370 | ADRENO_CP_ADDR_VFD_FETCH_INSTR_1_15; |
| 371 | for (i = ADRENO_CP_ADDR_VFD_FETCH_INSTR_1_0; |
| 372 | i <= vfd_end; i++) { |
| 373 | if (ib_parse_vars->cp_addr_regs[i]) { |
| 374 | ret = adreno_ib_add(process, |
| 375 | ib_parse_vars->cp_addr_regs[i], |
| 376 | SNAPSHOT_GPU_OBJECT_GENERIC, |
| 377 | ib_obj_list); |
| 378 | if (ret) |
| 379 | return ret; |
| 380 | ib_parse_vars->cp_addr_regs[i] = 0; |
| 381 | } |
| 382 | } |
| 383 | |
| 384 | if (ib_parse_vars->cp_addr_regs[ADRENO_CP_ADDR_VSC_SIZE_ADDRESS]) { |
| 385 | ret = adreno_ib_add(process, |
| 386 | ib_parse_vars->cp_addr_regs[ |
| 387 | ADRENO_CP_ADDR_VSC_SIZE_ADDRESS] & mask, |
| 388 | SNAPSHOT_GPU_OBJECT_GENERIC, ib_obj_list); |
| 389 | if (ret) |
| 390 | return ret; |
| 391 | ib_parse_vars->cp_addr_regs[ |
| 392 | ADRENO_CP_ADDR_VSC_SIZE_ADDRESS] = 0; |
| 393 | } |
| 394 | mask = 0xFFFFFFE0; |
| 395 | for (i = ADRENO_CP_ADDR_SP_VS_PVT_MEM_ADDR; |
| 396 | i <= ADRENO_CP_ADDR_SP_FS_OBJ_START_REG; i++) { |
| 397 | ret = adreno_ib_add(process, |
| 398 | ib_parse_vars->cp_addr_regs[i] & mask, |
| 399 | SNAPSHOT_GPU_OBJECT_GENERIC, ib_obj_list); |
| 400 | if (ret) |
| 401 | return ret; |
| 402 | ib_parse_vars->cp_addr_regs[i] = 0; |
| 403 | } |
| 404 | return ret; |
| 405 | } |
| 406 | /* |
| 407 | * The DRAW_INDX opcode sends a draw initator which starts a draw operation in |
| 408 | * the GPU, so this is the point where all the registers and buffers become |
| 409 | * "valid". The DRAW_INDX may also have an index buffer pointer that should be |
| 410 | * frozen with the others |
| 411 | */ |
| 412 | |
| 413 | static int ib_parse_draw_indx(struct kgsl_device *device, unsigned int *pkt, |
| 414 | struct kgsl_process_private *process, |
| 415 | struct adreno_ib_object_list *ib_obj_list, |
| 416 | struct ib_parser_variables *ib_parse_vars) |
| 417 | { |
| 418 | int ret = 0; |
| 419 | int i; |
| 420 | int opcode = cp_type3_opcode(pkt[0]); |
| 421 | |
| 422 | switch (opcode) { |
| 423 | case CP_DRAW_INDX: |
| 424 | if (type3_pkt_size(pkt[0]) > 3) { |
| 425 | ret = adreno_ib_add(process, |
| 426 | pkt[4], SNAPSHOT_GPU_OBJECT_GENERIC, |
| 427 | ib_obj_list); |
| 428 | } |
| 429 | break; |
| 430 | case CP_DRAW_INDX_OFFSET: |
| 431 | if (type3_pkt_size(pkt[0]) == 6) { |
| 432 | ret = adreno_ib_add(process, |
| 433 | pkt[5], SNAPSHOT_GPU_OBJECT_GENERIC, |
| 434 | ib_obj_list); |
| 435 | } |
| 436 | break; |
| 437 | case CP_DRAW_INDIRECT: |
| 438 | if (type3_pkt_size(pkt[0]) == 2) { |
| 439 | ret = adreno_ib_add(process, |
| 440 | pkt[2], SNAPSHOT_GPU_OBJECT_GENERIC, |
| 441 | ib_obj_list); |
| 442 | } |
| 443 | break; |
| 444 | case CP_DRAW_INDX_INDIRECT: |
| 445 | if (type3_pkt_size(pkt[0]) == 4) { |
| 446 | ret = adreno_ib_add(process, |
| 447 | pkt[2], SNAPSHOT_GPU_OBJECT_GENERIC, |
| 448 | ib_obj_list); |
| 449 | if (ret) |
| 450 | break; |
| 451 | ret = adreno_ib_add(process, |
| 452 | pkt[4], SNAPSHOT_GPU_OBJECT_GENERIC, |
| 453 | ib_obj_list); |
| 454 | } |
| 455 | break; |
| 456 | case CP_DRAW_AUTO: |
| 457 | if (type3_pkt_size(pkt[0]) == 6) { |
| 458 | ret = adreno_ib_add(process, |
| 459 | pkt[3], SNAPSHOT_GPU_OBJECT_GENERIC, |
| 460 | ib_obj_list); |
| 461 | if (ret) |
| 462 | break; |
| 463 | ret = adreno_ib_add(process, |
| 464 | pkt[4], SNAPSHOT_GPU_OBJECT_GENERIC, |
| 465 | ib_obj_list); |
| 466 | } |
| 467 | break; |
| 468 | } |
| 469 | |
| 470 | if (ret) |
| 471 | return ret; |
| 472 | /* |
| 473 | * All of the type0 writes are valid at a draw initiator, so freeze |
| 474 | * the various buffers that we are tracking |
| 475 | */ |
| 476 | ret = ib_add_type0_entries(device, process, ib_obj_list, |
| 477 | ib_parse_vars); |
| 478 | if (ret) |
| 479 | return ret; |
| 480 | /* Process set draw state command streams if any */ |
| 481 | for (i = 0; i < NUM_SET_DRAW_GROUPS; i++) { |
| 482 | if (!ib_parse_vars->set_draw_groups[i].cmd_stream_dwords) |
| 483 | continue; |
| 484 | ret = adreno_ib_find_objs(device, process, |
| 485 | ib_parse_vars->set_draw_groups[i].cmd_stream_addr, |
| 486 | ib_parse_vars->set_draw_groups[i].cmd_stream_dwords, |
Hareesh Gundu | 9c6b1fa | 2017-01-06 15:37:09 +0530 | [diff] [blame] | 487 | 0, SNAPSHOT_GPU_OBJECT_DRAW, |
Shrenuj Bansal | a419c79 | 2016-10-20 14:05:11 -0700 | [diff] [blame] | 488 | ib_obj_list, 2); |
| 489 | if (ret) |
| 490 | break; |
| 491 | } |
| 492 | return ret; |
| 493 | } |
| 494 | |
| 495 | /* |
| 496 | * Parse all the type7 opcode packets that may contain important information, |
| 497 | * such as additional GPU buffers to grab or a draw initator |
| 498 | */ |
| 499 | |
| 500 | static int ib_parse_type7(struct kgsl_device *device, unsigned int *ptr, |
| 501 | struct kgsl_process_private *process, |
| 502 | struct adreno_ib_object_list *ib_obj_list, |
| 503 | struct ib_parser_variables *ib_parse_vars) |
| 504 | { |
| 505 | int opcode = cp_type7_opcode(*ptr); |
| 506 | |
| 507 | switch (opcode) { |
| 508 | case CP_SET_DRAW_STATE: |
| 509 | return ib_parse_type7_set_draw_state(device, ptr, process, |
| 510 | ib_obj_list); |
| 511 | } |
| 512 | |
| 513 | return 0; |
| 514 | } |
| 515 | |
| 516 | /* |
| 517 | * Parse all the type3 opcode packets that may contain important information, |
| 518 | * such as additional GPU buffers to grab or a draw initator |
| 519 | */ |
| 520 | |
| 521 | static int ib_parse_type3(struct kgsl_device *device, unsigned int *ptr, |
| 522 | struct kgsl_process_private *process, |
| 523 | struct adreno_ib_object_list *ib_obj_list, |
| 524 | struct ib_parser_variables *ib_parse_vars) |
| 525 | { |
| 526 | int opcode = cp_type3_opcode(*ptr); |
| 527 | |
| 528 | switch (opcode) { |
| 529 | case CP_LOAD_STATE: |
| 530 | return ib_parse_load_state(ptr, process, ib_obj_list, |
| 531 | ib_parse_vars); |
| 532 | case CP_SET_BIN_DATA: |
| 533 | return ib_parse_set_bin_data(ptr, process, ib_obj_list, |
| 534 | ib_parse_vars); |
| 535 | case CP_MEM_WRITE: |
| 536 | return ib_parse_mem_write(ptr, process, ib_obj_list, |
| 537 | ib_parse_vars); |
| 538 | case CP_DRAW_INDX: |
| 539 | case CP_DRAW_INDX_OFFSET: |
| 540 | case CP_DRAW_INDIRECT: |
| 541 | case CP_DRAW_INDX_INDIRECT: |
| 542 | return ib_parse_draw_indx(device, ptr, process, ib_obj_list, |
| 543 | ib_parse_vars); |
| 544 | case CP_SET_DRAW_STATE: |
| 545 | return ib_parse_set_draw_state(device, ptr, process, |
| 546 | ib_obj_list, ib_parse_vars); |
| 547 | } |
| 548 | |
| 549 | return 0; |
| 550 | } |
| 551 | |
| 552 | /* |
| 553 | * Parse type0 packets found in the stream. Some of the registers that are |
| 554 | * written are clues for GPU buffers that we need to freeze. Register writes |
| 555 | * are considred valid when a draw initator is called, so just cache the values |
| 556 | * here and freeze them when a CP_DRAW_INDX is seen. This protects against |
| 557 | * needlessly caching buffers that won't be used during a draw call |
| 558 | */ |
| 559 | |
| 560 | static int ib_parse_type0(struct kgsl_device *device, unsigned int *ptr, |
| 561 | struct kgsl_process_private *process, |
| 562 | struct adreno_ib_object_list *ib_obj_list, |
| 563 | struct ib_parser_variables *ib_parse_vars) |
| 564 | { |
| 565 | struct adreno_device *adreno_dev = ADRENO_DEVICE(device); |
| 566 | int size = type0_pkt_size(*ptr); |
| 567 | int offset = type0_pkt_offset(*ptr); |
| 568 | int i; |
| 569 | int reg_index; |
| 570 | int ret = 0; |
| 571 | |
| 572 | for (i = 0; i < size; i++, offset++) { |
| 573 | /* Visiblity stream buffer */ |
| 574 | if (offset >= adreno_cp_parser_getreg(adreno_dev, |
| 575 | ADRENO_CP_ADDR_VSC_PIPE_DATA_ADDRESS_0) && |
| 576 | offset <= adreno_cp_parser_getreg(adreno_dev, |
| 577 | ADRENO_CP_ADDR_VSC_PIPE_DATA_LENGTH_7)) { |
| 578 | reg_index = adreno_cp_parser_regindex( |
| 579 | adreno_dev, offset, |
| 580 | ADRENO_CP_ADDR_VSC_PIPE_DATA_ADDRESS_0, |
| 581 | ADRENO_CP_ADDR_VSC_PIPE_DATA_LENGTH_7); |
| 582 | if (reg_index >= 0) |
| 583 | ib_parse_vars->cp_addr_regs[reg_index] = |
| 584 | ptr[i + 1]; |
| 585 | continue; |
| 586 | } else if ((offset >= adreno_cp_parser_getreg(adreno_dev, |
| 587 | ADRENO_CP_ADDR_VFD_FETCH_INSTR_1_0)) && |
| 588 | (offset <= adreno_cp_parser_getreg(adreno_dev, |
| 589 | ADRENO_CP_ADDR_VFD_FETCH_INSTR_1_15))) { |
| 590 | reg_index = adreno_cp_parser_regindex(adreno_dev, |
| 591 | offset, |
| 592 | ADRENO_CP_ADDR_VFD_FETCH_INSTR_1_0, |
| 593 | ADRENO_CP_ADDR_VFD_FETCH_INSTR_1_15); |
| 594 | if (reg_index >= 0) |
| 595 | ib_parse_vars->cp_addr_regs[reg_index] = |
| 596 | ptr[i + 1]; |
| 597 | continue; |
| 598 | } else if ((offset >= adreno_cp_parser_getreg(adreno_dev, |
| 599 | ADRENO_CP_ADDR_VFD_FETCH_INSTR_1_16)) && |
| 600 | (offset <= adreno_cp_parser_getreg(adreno_dev, |
| 601 | ADRENO_CP_ADDR_VFD_FETCH_INSTR_1_31))) { |
| 602 | reg_index = adreno_cp_parser_regindex(adreno_dev, |
| 603 | offset, |
| 604 | ADRENO_CP_ADDR_VFD_FETCH_INSTR_1_16, |
| 605 | ADRENO_CP_ADDR_VFD_FETCH_INSTR_1_31); |
| 606 | if (reg_index >= 0) |
| 607 | ib_parse_vars->cp_addr_regs[reg_index] = |
| 608 | ptr[i + 1]; |
| 609 | continue; |
| 610 | } else { |
| 611 | if (offset == |
| 612 | adreno_cp_parser_getreg(adreno_dev, |
| 613 | ADRENO_CP_ADDR_VSC_SIZE_ADDRESS)) |
| 614 | ib_parse_vars->cp_addr_regs[ |
| 615 | ADRENO_CP_ADDR_VSC_SIZE_ADDRESS] = |
| 616 | ptr[i + 1]; |
| 617 | else if (offset == adreno_cp_parser_getreg(adreno_dev, |
| 618 | ADRENO_CP_ADDR_SP_VS_PVT_MEM_ADDR)) |
| 619 | ib_parse_vars->cp_addr_regs[ |
| 620 | ADRENO_CP_ADDR_SP_VS_PVT_MEM_ADDR] = |
| 621 | ptr[i + 1]; |
| 622 | else if (offset == adreno_cp_parser_getreg(adreno_dev, |
| 623 | ADRENO_CP_ADDR_SP_FS_PVT_MEM_ADDR)) |
| 624 | ib_parse_vars->cp_addr_regs[ |
| 625 | ADRENO_CP_ADDR_SP_FS_PVT_MEM_ADDR] = |
| 626 | ptr[i + 1]; |
| 627 | else if (offset == adreno_cp_parser_getreg(adreno_dev, |
| 628 | ADRENO_CP_ADDR_SP_VS_OBJ_START_REG)) |
| 629 | ib_parse_vars->cp_addr_regs[ |
| 630 | ADRENO_CP_ADDR_SP_VS_OBJ_START_REG] = |
| 631 | ptr[i + 1]; |
| 632 | else if (offset == adreno_cp_parser_getreg(adreno_dev, |
| 633 | ADRENO_CP_ADDR_SP_FS_OBJ_START_REG)) |
| 634 | ib_parse_vars->cp_addr_regs[ |
| 635 | ADRENO_CP_ADDR_SP_FS_OBJ_START_REG] = |
| 636 | ptr[i + 1]; |
| 637 | else if ((offset == adreno_cp_parser_getreg(adreno_dev, |
| 638 | ADRENO_CP_UCHE_INVALIDATE0)) || |
| 639 | (offset == adreno_cp_parser_getreg(adreno_dev, |
| 640 | ADRENO_CP_UCHE_INVALIDATE1))) { |
| 641 | ret = adreno_ib_add(process, |
| 642 | ptr[i + 1] & 0xFFFFFFC0, |
| 643 | SNAPSHOT_GPU_OBJECT_GENERIC, |
| 644 | ib_obj_list); |
| 645 | if (ret) |
| 646 | break; |
| 647 | } |
| 648 | } |
| 649 | } |
| 650 | return ret; |
| 651 | } |
| 652 | |
| 653 | static int ib_parse_type7_set_draw_state(struct kgsl_device *device, |
| 654 | unsigned int *ptr, |
| 655 | struct kgsl_process_private *process, |
| 656 | struct adreno_ib_object_list *ib_obj_list) |
| 657 | { |
| 658 | int size = type7_pkt_size(*ptr); |
| 659 | int i; |
| 660 | int grp_id; |
| 661 | int ret = 0; |
| 662 | int flags; |
| 663 | uint64_t cmd_stream_dwords; |
| 664 | uint64_t cmd_stream_addr; |
| 665 | |
| 666 | /* |
| 667 | * size is the size of the packet that does not include the DWORD |
| 668 | * for the packet header, we only want to loop here through the |
| 669 | * packet parameters from ptr[1] till ptr[size] where ptr[0] is the |
| 670 | * packet header. In each loop we look at 3 DWORDS hence increment |
| 671 | * loop counter by 3 always |
| 672 | */ |
| 673 | for (i = 1; i <= size; i += 3) { |
| 674 | grp_id = (ptr[i] & 0x1F000000) >> 24; |
| 675 | /* take action based on flags */ |
| 676 | flags = (ptr[i] & 0x000F0000) >> 16; |
| 677 | |
| 678 | /* |
| 679 | * dirty flag or no flags both mean we need to load it for |
| 680 | * next draw. No flags is used when the group is activated |
| 681 | * or initialized for the first time in the IB |
| 682 | */ |
| 683 | if (flags & 0x1 || !flags) { |
| 684 | cmd_stream_dwords = ptr[i] & 0x0000FFFF; |
| 685 | cmd_stream_addr = ptr[i + 2]; |
| 686 | cmd_stream_addr = cmd_stream_addr << 32 | ptr[i + 1]; |
| 687 | if (cmd_stream_dwords) |
| 688 | ret = adreno_ib_find_objs(device, process, |
| 689 | cmd_stream_addr, cmd_stream_dwords, |
Hareesh Gundu | 9c6b1fa | 2017-01-06 15:37:09 +0530 | [diff] [blame] | 690 | 0, SNAPSHOT_GPU_OBJECT_DRAW, |
| 691 | ib_obj_list, 2); |
Shrenuj Bansal | a419c79 | 2016-10-20 14:05:11 -0700 | [diff] [blame] | 692 | if (ret) |
| 693 | break; |
| 694 | continue; |
| 695 | } |
| 696 | /* load immediate */ |
| 697 | if (flags & 0x8) { |
| 698 | uint64_t gpuaddr = ptr[i + 2]; |
| 699 | |
| 700 | gpuaddr = gpuaddr << 32 | ptr[i + 1]; |
| 701 | ret = adreno_ib_find_objs(device, process, |
| 702 | gpuaddr, (ptr[i] & 0x0000FFFF), |
Hareesh Gundu | 9c6b1fa | 2017-01-06 15:37:09 +0530 | [diff] [blame] | 703 | 0, SNAPSHOT_GPU_OBJECT_IB, |
Shrenuj Bansal | a419c79 | 2016-10-20 14:05:11 -0700 | [diff] [blame] | 704 | ib_obj_list, 2); |
| 705 | if (ret) |
| 706 | break; |
| 707 | } |
| 708 | } |
| 709 | return ret; |
| 710 | } |
| 711 | |
| 712 | static int ib_parse_set_draw_state(struct kgsl_device *device, |
| 713 | unsigned int *ptr, |
| 714 | struct kgsl_process_private *process, |
| 715 | struct adreno_ib_object_list *ib_obj_list, |
| 716 | struct ib_parser_variables *ib_parse_vars) |
| 717 | { |
| 718 | int size = type0_pkt_size(*ptr); |
| 719 | int i; |
| 720 | int grp_id; |
| 721 | int ret = 0; |
| 722 | int flags; |
| 723 | |
| 724 | /* |
| 725 | * size is the size of the packet that does not include the DWORD |
| 726 | * for the packet header, we only want to loop here through the |
| 727 | * packet parameters from ptr[1] till ptr[size] where ptr[0] is the |
| 728 | * packet header. In each loop we look at 2 DWORDS hence increment |
| 729 | * loop counter by 2 always |
| 730 | */ |
| 731 | for (i = 1; i <= size; i += 2) { |
| 732 | grp_id = (ptr[i] & 0x1F000000) >> 24; |
| 733 | /* take action based on flags */ |
| 734 | flags = (ptr[i] & 0x000F0000) >> 16; |
| 735 | /* Disable all groups */ |
| 736 | if (flags & 0x4) { |
| 737 | int j; |
| 738 | |
| 739 | for (j = 0; j < NUM_SET_DRAW_GROUPS; j++) |
| 740 | ib_parse_vars->set_draw_groups[j]. |
| 741 | cmd_stream_dwords = 0; |
| 742 | continue; |
| 743 | } |
| 744 | /* disable flag */ |
| 745 | if (flags & 0x2) { |
| 746 | ib_parse_vars->set_draw_groups[grp_id]. |
| 747 | cmd_stream_dwords = 0; |
| 748 | continue; |
| 749 | } |
| 750 | /* |
| 751 | * dirty flag or no flags both mean we need to load it for |
| 752 | * next draw. No flags is used when the group is activated |
| 753 | * or initialized for the first time in the IB |
| 754 | */ |
| 755 | if (flags & 0x1 || !flags) { |
| 756 | ib_parse_vars->set_draw_groups[grp_id]. |
| 757 | cmd_stream_dwords = ptr[i] & 0x0000FFFF; |
| 758 | ib_parse_vars->set_draw_groups[grp_id]. |
| 759 | cmd_stream_addr = ptr[i + 1]; |
| 760 | continue; |
| 761 | } |
| 762 | /* load immediate */ |
| 763 | if (flags & 0x8) { |
| 764 | ret = adreno_ib_find_objs(device, process, |
| 765 | ptr[i + 1], (ptr[i] & 0x0000FFFF), |
Hareesh Gundu | 9c6b1fa | 2017-01-06 15:37:09 +0530 | [diff] [blame] | 766 | 0, SNAPSHOT_GPU_OBJECT_IB, |
Shrenuj Bansal | a419c79 | 2016-10-20 14:05:11 -0700 | [diff] [blame] | 767 | ib_obj_list, 2); |
| 768 | if (ret) |
| 769 | break; |
| 770 | } |
| 771 | } |
| 772 | return ret; |
| 773 | } |
| 774 | |
| 775 | /* |
| 776 | * adreno_cp_parse_ib2() - Wrapper function around IB2 parsing |
| 777 | * @device: Device pointer |
| 778 | * @process: Process in which the IB is allocated |
| 779 | * @gpuaddr: IB2 gpuaddr |
| 780 | * @dwords: IB2 size in dwords |
Hareesh Gundu | 9c6b1fa | 2017-01-06 15:37:09 +0530 | [diff] [blame] | 781 | * @ib2base: Base address of active IB2 |
Shrenuj Bansal | a419c79 | 2016-10-20 14:05:11 -0700 | [diff] [blame] | 782 | * @ib_obj_list: List of objects found in IB |
| 783 | * @ib_level: The level from which function is called, either from IB1 or IB2 |
| 784 | * |
| 785 | * Function does some checks to ensure that IB2 parsing is called from IB1 |
| 786 | * and then calls the function to find objects in IB2. |
| 787 | */ |
| 788 | static int adreno_cp_parse_ib2(struct kgsl_device *device, |
| 789 | struct kgsl_process_private *process, |
Hareesh Gundu | 9c6b1fa | 2017-01-06 15:37:09 +0530 | [diff] [blame] | 790 | uint64_t gpuaddr, uint64_t dwords, uint64_t ib2base, |
Shrenuj Bansal | a419c79 | 2016-10-20 14:05:11 -0700 | [diff] [blame] | 791 | struct adreno_ib_object_list *ib_obj_list, |
| 792 | int ib_level) |
| 793 | { |
| 794 | int i; |
| 795 | |
| 796 | /* |
| 797 | * We can only expect an IB2 in IB1, if we are |
| 798 | * already processing an IB2 then return error |
| 799 | */ |
| 800 | if (ib_level == 2) |
| 801 | return -EINVAL; |
Hareesh Gundu | 9c6b1fa | 2017-01-06 15:37:09 +0530 | [diff] [blame] | 802 | |
| 803 | /* Save current IB2 statically */ |
| 804 | if (ib2base == gpuaddr) |
| 805 | kgsl_snapshot_push_object(process, gpuaddr, dwords); |
Shrenuj Bansal | a419c79 | 2016-10-20 14:05:11 -0700 | [diff] [blame] | 806 | /* |
| 807 | * only try to find sub objects iff this IB has |
| 808 | * not been processed already |
| 809 | */ |
| 810 | for (i = 0; i < ib_obj_list->num_objs; i++) { |
| 811 | struct adreno_ib_object *ib_obj = &(ib_obj_list->obj_list[i]); |
| 812 | |
| 813 | if ((ib_obj->snapshot_obj_type == SNAPSHOT_GPU_OBJECT_IB) && |
| 814 | (gpuaddr >= ib_obj->gpuaddr) && |
| 815 | (gpuaddr + dwords * sizeof(unsigned int) <= |
| 816 | ib_obj->gpuaddr + ib_obj->size)) |
| 817 | return 0; |
| 818 | } |
| 819 | |
Hareesh Gundu | 9c6b1fa | 2017-01-06 15:37:09 +0530 | [diff] [blame] | 820 | return adreno_ib_find_objs(device, process, gpuaddr, dwords, ib2base, |
Shrenuj Bansal | a419c79 | 2016-10-20 14:05:11 -0700 | [diff] [blame] | 821 | SNAPSHOT_GPU_OBJECT_IB, ib_obj_list, 2); |
| 822 | } |
| 823 | |
| 824 | /* |
| 825 | * adreno_ib_find_objs() - Find all IB objects in a given IB |
| 826 | * @device: The device pointer on which the IB executes |
| 827 | * @process: The process in which the IB and all contained objects are mapped. |
| 828 | * @gpuaddr: The gpu address of the IB |
Hareesh Gundu | 9c6b1fa | 2017-01-06 15:37:09 +0530 | [diff] [blame] | 829 | * @ib2base: IB2 base address |
Shrenuj Bansal | a419c79 | 2016-10-20 14:05:11 -0700 | [diff] [blame] | 830 | * @dwords: Size of ib in dwords |
| 831 | * @obj_type: The object type can be either an IB or a draw state sequence |
| 832 | * @ib_obj_list: The list in which the IB and the objects in it are added. |
| 833 | * @ib_level: Indicates if IB1 or IB2 is being processed |
| 834 | * |
| 835 | * Finds all IB objects in a given IB and puts then in a list. Can be called |
| 836 | * recursively for the IB2's in the IB1's |
| 837 | * Returns 0 on success else error code |
| 838 | */ |
| 839 | static int adreno_ib_find_objs(struct kgsl_device *device, |
| 840 | struct kgsl_process_private *process, |
| 841 | uint64_t gpuaddr, uint64_t dwords, |
Hareesh Gundu | 9c6b1fa | 2017-01-06 15:37:09 +0530 | [diff] [blame] | 842 | uint64_t ib2base, int obj_type, |
Shrenuj Bansal | a419c79 | 2016-10-20 14:05:11 -0700 | [diff] [blame] | 843 | struct adreno_ib_object_list *ib_obj_list, |
| 844 | int ib_level) |
| 845 | { |
| 846 | int ret = 0; |
| 847 | uint64_t rem = dwords; |
| 848 | int i; |
| 849 | struct ib_parser_variables ib_parse_vars; |
| 850 | unsigned int *src; |
| 851 | struct adreno_ib_object *ib_obj; |
| 852 | struct kgsl_mem_entry *entry; |
| 853 | struct adreno_device *adreno_dev = ADRENO_DEVICE(device); |
| 854 | |
| 855 | /* check that this IB is not already on list */ |
| 856 | for (i = 0; i < ib_obj_list->num_objs; i++) { |
| 857 | ib_obj = &(ib_obj_list->obj_list[i]); |
| 858 | if ((obj_type == ib_obj->snapshot_obj_type) && |
| 859 | (ib_obj->gpuaddr <= gpuaddr) && |
| 860 | ((ib_obj->gpuaddr + ib_obj->size) >= |
| 861 | (gpuaddr + (dwords << 2)))) |
| 862 | return 0; |
| 863 | } |
| 864 | |
| 865 | entry = kgsl_sharedmem_find(process, gpuaddr); |
| 866 | if (!entry) |
| 867 | return -EINVAL; |
| 868 | |
| 869 | if (!kgsl_gpuaddr_in_memdesc(&entry->memdesc, gpuaddr, (dwords << 2))) { |
| 870 | kgsl_mem_entry_put(entry); |
| 871 | return -EINVAL; |
| 872 | } |
| 873 | |
| 874 | src = kgsl_gpuaddr_to_vaddr(&entry->memdesc, gpuaddr); |
| 875 | if (!src) { |
| 876 | kgsl_mem_entry_put(entry); |
| 877 | return -EINVAL; |
| 878 | } |
| 879 | |
| 880 | memset(&ib_parse_vars, 0, sizeof(struct ib_parser_variables)); |
| 881 | |
| 882 | ret = adreno_ib_add(process, gpuaddr, obj_type, ib_obj_list); |
| 883 | if (ret) |
| 884 | goto done; |
| 885 | |
| 886 | for (i = 0; rem > 0; rem--, i++) { |
| 887 | int pktsize; |
| 888 | |
| 889 | if (pkt_is_type0(src[i])) |
| 890 | pktsize = type0_pkt_size(src[i]); |
| 891 | |
| 892 | else if (pkt_is_type3(src[i])) |
| 893 | pktsize = type3_pkt_size(src[i]); |
| 894 | |
| 895 | else if (pkt_is_type4(src[i])) |
| 896 | pktsize = type4_pkt_size(src[i]); |
| 897 | |
| 898 | else if (pkt_is_type7(src[i])) |
| 899 | pktsize = type7_pkt_size(src[i]); |
| 900 | |
| 901 | /* |
| 902 | * If the packet isn't a type 1, type 3, type 4 or type 7 then |
| 903 | * don't bother parsing it - it is likely corrupted |
| 904 | */ |
| 905 | else |
| 906 | break; |
| 907 | |
| 908 | if (((pkt_is_type0(src[i]) || pkt_is_type3(src[i])) && !pktsize) |
| 909 | || ((pktsize + 1) > rem)) |
| 910 | break; |
| 911 | |
| 912 | if (pkt_is_type3(src[i])) { |
| 913 | if (adreno_cmd_is_ib(adreno_dev, src[i])) { |
| 914 | uint64_t gpuaddrib2 = src[i + 1]; |
| 915 | uint64_t size = src[i + 2]; |
| 916 | |
| 917 | ret = adreno_cp_parse_ib2(device, process, |
Hareesh Gundu | 9c6b1fa | 2017-01-06 15:37:09 +0530 | [diff] [blame] | 918 | gpuaddrib2, size, ib2base, |
Shrenuj Bansal | a419c79 | 2016-10-20 14:05:11 -0700 | [diff] [blame] | 919 | ib_obj_list, ib_level); |
| 920 | if (ret) |
| 921 | goto done; |
| 922 | } else { |
| 923 | ret = ib_parse_type3(device, &src[i], process, |
| 924 | ib_obj_list, |
| 925 | &ib_parse_vars); |
| 926 | /* |
| 927 | * If the parse function failed (probably |
| 928 | * because of a bad decode) then bail out and |
| 929 | * just capture the binary IB data |
| 930 | */ |
| 931 | |
| 932 | if (ret) |
| 933 | goto done; |
| 934 | } |
| 935 | } |
| 936 | |
| 937 | else if (pkt_is_type7(src[i])) { |
| 938 | if (adreno_cmd_is_ib(adreno_dev, src[i])) { |
| 939 | uint64_t size = src[i + 3]; |
| 940 | uint64_t gpuaddrib2 = src[i + 2]; |
| 941 | |
| 942 | gpuaddrib2 = gpuaddrib2 << 32 | src[i + 1]; |
| 943 | |
| 944 | ret = adreno_cp_parse_ib2(device, process, |
Hareesh Gundu | 9c6b1fa | 2017-01-06 15:37:09 +0530 | [diff] [blame] | 945 | gpuaddrib2, size, ib2base, |
Shrenuj Bansal | a419c79 | 2016-10-20 14:05:11 -0700 | [diff] [blame] | 946 | ib_obj_list, ib_level); |
| 947 | if (ret) |
| 948 | goto done; |
| 949 | } else { |
| 950 | ret = ib_parse_type7(device, &src[i], process, |
| 951 | ib_obj_list, |
| 952 | &ib_parse_vars); |
| 953 | /* |
| 954 | * If the parse function failed (probably |
| 955 | * because of a bad decode) then bail out and |
| 956 | * just capture the binary IB data |
| 957 | */ |
| 958 | |
| 959 | if (ret) |
| 960 | goto done; |
| 961 | } |
| 962 | } |
| 963 | |
| 964 | else if (pkt_is_type0(src[i])) { |
| 965 | ret = ib_parse_type0(device, &src[i], process, |
| 966 | ib_obj_list, &ib_parse_vars); |
| 967 | if (ret) |
| 968 | goto done; |
| 969 | } |
| 970 | |
| 971 | i += pktsize; |
| 972 | rem -= pktsize; |
| 973 | } |
| 974 | |
| 975 | done: |
| 976 | /* |
| 977 | * For set draw objects there may not be a draw_indx packet at its end |
| 978 | * to signal that we need to save the found objects in it, so just save |
| 979 | * it here. |
| 980 | */ |
| 981 | if (!ret && SNAPSHOT_GPU_OBJECT_DRAW == obj_type) |
| 982 | ret = ib_add_type0_entries(device, process, ib_obj_list, |
| 983 | &ib_parse_vars); |
| 984 | |
| 985 | kgsl_memdesc_unmap(&entry->memdesc); |
| 986 | kgsl_mem_entry_put(entry); |
| 987 | return ret; |
| 988 | } |
| 989 | |
| 990 | |
| 991 | /* |
| 992 | * adreno_ib_create_object_list() - Find all the memory objects in IB |
| 993 | * @device: The device pointer on which the IB executes |
| 994 | * @process: The process in which the IB and all contained objects are mapped |
| 995 | * @gpuaddr: The gpu address of the IB |
| 996 | * @dwords: Size of ib in dwords |
Hareesh Gundu | 9c6b1fa | 2017-01-06 15:37:09 +0530 | [diff] [blame] | 997 | * @ib2base: Base address of active IB2 |
Shrenuj Bansal | a419c79 | 2016-10-20 14:05:11 -0700 | [diff] [blame] | 998 | * @ib_obj_list: The list in which the IB and the objects in it are added. |
| 999 | * |
| 1000 | * Find all the memory objects that an IB needs for execution and place |
| 1001 | * them in a list including the IB. |
| 1002 | * Returns the ib object list. On success 0 is returned, on failure error |
| 1003 | * code is returned along with number of objects that was saved before |
| 1004 | * error occurred. If no objects found then the list pointer is set to |
| 1005 | * NULL. |
| 1006 | */ |
| 1007 | int adreno_ib_create_object_list(struct kgsl_device *device, |
| 1008 | struct kgsl_process_private *process, |
Hareesh Gundu | 9c6b1fa | 2017-01-06 15:37:09 +0530 | [diff] [blame] | 1009 | uint64_t gpuaddr, uint64_t dwords, uint64_t ib2base, |
Shrenuj Bansal | a419c79 | 2016-10-20 14:05:11 -0700 | [diff] [blame] | 1010 | struct adreno_ib_object_list **out_ib_obj_list) |
| 1011 | { |
| 1012 | int ret = 0; |
| 1013 | struct adreno_ib_object_list *ib_obj_list; |
| 1014 | |
| 1015 | if (!out_ib_obj_list) |
| 1016 | return -EINVAL; |
| 1017 | |
| 1018 | *out_ib_obj_list = NULL; |
| 1019 | |
| 1020 | ib_obj_list = kzalloc(sizeof(*ib_obj_list), GFP_KERNEL); |
| 1021 | if (!ib_obj_list) |
| 1022 | return -ENOMEM; |
| 1023 | |
| 1024 | ib_obj_list->obj_list = vmalloc(MAX_IB_OBJS * |
| 1025 | sizeof(struct adreno_ib_object)); |
| 1026 | |
| 1027 | if (!ib_obj_list->obj_list) { |
| 1028 | kfree(ib_obj_list); |
| 1029 | return -ENOMEM; |
| 1030 | } |
| 1031 | |
Hareesh Gundu | 9c6b1fa | 2017-01-06 15:37:09 +0530 | [diff] [blame] | 1032 | ret = adreno_ib_find_objs(device, process, gpuaddr, dwords, ib2base, |
Shrenuj Bansal | a419c79 | 2016-10-20 14:05:11 -0700 | [diff] [blame] | 1033 | SNAPSHOT_GPU_OBJECT_IB, ib_obj_list, 1); |
| 1034 | |
| 1035 | /* Even if there was an error return the remaining objects found */ |
| 1036 | if (ib_obj_list->num_objs) |
| 1037 | *out_ib_obj_list = ib_obj_list; |
| 1038 | |
| 1039 | return ret; |
| 1040 | } |
| 1041 | |
| 1042 | /* |
| 1043 | * adreno_ib_destroy_obj_list() - Destroy an ib object list |
| 1044 | * @ib_obj_list: List to destroy |
| 1045 | * |
| 1046 | * Free up all resources used by an ib_obj_list |
| 1047 | */ |
| 1048 | void adreno_ib_destroy_obj_list(struct adreno_ib_object_list *ib_obj_list) |
| 1049 | { |
| 1050 | int i; |
| 1051 | |
| 1052 | if (!ib_obj_list) |
| 1053 | return; |
| 1054 | |
| 1055 | for (i = 0; i < ib_obj_list->num_objs; i++) { |
| 1056 | if (ib_obj_list->obj_list[i].entry) |
| 1057 | kgsl_mem_entry_put(ib_obj_list->obj_list[i].entry); |
| 1058 | } |
| 1059 | vfree(ib_obj_list->obj_list); |
| 1060 | kfree(ib_obj_list); |
| 1061 | } |