1 /* 2 * Tegra host1x Job 3 * 4 * Copyright (c) 2010-2013, NVIDIA Corporation. 5 * 6 * This program is free software; you can redistribute it and/or modify it 7 * under the terms and conditions of the GNU General Public License, 8 * version 2, as published by the Free Software Foundation. 9 * 10 * This program is distributed in the hope it will be useful, but WITHOUT 11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for 13 * more details. 14 * 15 * You should have received a copy of the GNU General Public License 16 * along with this program. If not, see <http://www.gnu.org/licenses/>. 17 */ 18 19 #include <linux/dma-mapping.h> 20 #include <linux/err.h> 21 #include <linux/host1x.h> 22 #include <linux/kref.h> 23 #include <linux/module.h> 24 #include <linux/scatterlist.h> 25 #include <linux/slab.h> 26 #include <linux/vmalloc.h> 27 #include <trace/events/host1x.h> 28 29 #include "channel.h" 30 #include "dev.h" 31 #include "job.h" 32 #include "syncpt.h" 33 34 struct host1x_job *host1x_job_alloc(struct host1x_channel *ch, 35 u32 num_cmdbufs, u32 num_relocs, 36 u32 num_waitchks) 37 { 38 struct host1x_job *job = NULL; 39 unsigned int num_unpins = num_cmdbufs + num_relocs; 40 u64 total; 41 void *mem; 42 43 /* Check that we're not going to overflow */ 44 total = sizeof(struct host1x_job) + 45 (u64)num_relocs * sizeof(struct host1x_reloc) + 46 (u64)num_unpins * sizeof(struct host1x_job_unpin_data) + 47 (u64)num_waitchks * sizeof(struct host1x_waitchk) + 48 (u64)num_cmdbufs * sizeof(struct host1x_job_gather) + 49 (u64)num_unpins * sizeof(dma_addr_t) + 50 (u64)num_unpins * sizeof(u32 *); 51 if (total > ULONG_MAX) 52 return NULL; 53 54 mem = job = kzalloc(total, GFP_KERNEL); 55 if (!job) 56 return NULL; 57 58 kref_init(&job->ref); 59 job->channel = ch; 60 61 /* Redistribute memory to the structs */ 62 mem += sizeof(struct host1x_job); 63 job->relocarray = num_relocs ? mem : NULL; 64 mem += num_relocs * sizeof(struct host1x_reloc); 65 job->unpins = num_unpins ? mem : NULL; 66 mem += num_unpins * sizeof(struct host1x_job_unpin_data); 67 job->waitchk = num_waitchks ? mem : NULL; 68 mem += num_waitchks * sizeof(struct host1x_waitchk); 69 job->gathers = num_cmdbufs ? mem : NULL; 70 mem += num_cmdbufs * sizeof(struct host1x_job_gather); 71 job->addr_phys = num_unpins ? mem : NULL; 72 73 job->reloc_addr_phys = job->addr_phys; 74 job->gather_addr_phys = &job->addr_phys[num_relocs]; 75 76 return job; 77 } 78 79 struct host1x_job *host1x_job_get(struct host1x_job *job) 80 { 81 kref_get(&job->ref); 82 return job; 83 } 84 85 static void job_free(struct kref *ref) 86 { 87 struct host1x_job *job = container_of(ref, struct host1x_job, ref); 88 89 kfree(job); 90 } 91 92 void host1x_job_put(struct host1x_job *job) 93 { 94 kref_put(&job->ref, job_free); 95 } 96 97 void host1x_job_add_gather(struct host1x_job *job, struct host1x_bo *bo, 98 u32 words, u32 offset) 99 { 100 struct host1x_job_gather *cur_gather = &job->gathers[job->num_gathers]; 101 102 cur_gather->words = words; 103 cur_gather->bo = bo; 104 cur_gather->offset = offset; 105 job->num_gathers++; 106 } 107 108 /* 109 * NULL an already satisfied WAIT_SYNCPT host method, by patching its 110 * args in the command stream. The method data is changed to reference 111 * a reserved (never given out or incr) HOST1X_SYNCPT_RESERVED syncpt 112 * with a matching threshold value of 0, so is guaranteed to be popped 113 * by the host HW. 114 */ 115 static void host1x_syncpt_patch_offset(struct host1x_syncpt *sp, 116 struct host1x_bo *h, u32 offset) 117 { 118 void *patch_addr = NULL; 119 120 /* patch the wait */ 121 patch_addr = host1x_bo_kmap(h, offset >> PAGE_SHIFT); 122 if (patch_addr) { 123 host1x_syncpt_patch_wait(sp, 124 patch_addr + (offset & ~PAGE_MASK)); 125 host1x_bo_kunmap(h, offset >> PAGE_SHIFT, patch_addr); 126 } else 127 pr_err("Could not map cmdbuf for wait check\n"); 128 } 129 130 /* 131 * Check driver supplied waitchk structs for syncpt thresholds 132 * that have already been satisfied and NULL the comparison (to 133 * avoid a wrap condition in the HW). 134 */ 135 static int do_waitchks(struct host1x_job *job, struct host1x *host, 136 struct host1x_bo *patch) 137 { 138 int i; 139 140 /* compare syncpt vs wait threshold */ 141 for (i = 0; i < job->num_waitchk; i++) { 142 struct host1x_waitchk *wait = &job->waitchk[i]; 143 struct host1x_syncpt *sp = 144 host1x_syncpt_get(host, wait->syncpt_id); 145 146 /* validate syncpt id */ 147 if (wait->syncpt_id > host1x_syncpt_nb_pts(host)) 148 continue; 149 150 /* skip all other gathers */ 151 if (patch != wait->bo) 152 continue; 153 154 trace_host1x_syncpt_wait_check(wait->bo, wait->offset, 155 wait->syncpt_id, wait->thresh, 156 host1x_syncpt_read_min(sp)); 157 158 if (host1x_syncpt_is_expired(sp, wait->thresh)) { 159 dev_dbg(host->dev, 160 "drop WAIT id %d (%s) thresh 0x%x, min 0x%x\n", 161 wait->syncpt_id, sp->name, wait->thresh, 162 host1x_syncpt_read_min(sp)); 163 164 host1x_syncpt_patch_offset(sp, patch, wait->offset); 165 } 166 167 wait->bo = NULL; 168 } 169 170 return 0; 171 } 172 173 static unsigned int pin_job(struct host1x_job *job) 174 { 175 unsigned int i; 176 177 job->num_unpins = 0; 178 179 for (i = 0; i < job->num_relocs; i++) { 180 struct host1x_reloc *reloc = &job->relocarray[i]; 181 struct sg_table *sgt; 182 dma_addr_t phys_addr; 183 184 reloc->target = host1x_bo_get(reloc->target); 185 if (!reloc->target) 186 goto unpin; 187 188 phys_addr = host1x_bo_pin(reloc->target, &sgt); 189 if (!phys_addr) 190 goto unpin; 191 192 job->addr_phys[job->num_unpins] = phys_addr; 193 job->unpins[job->num_unpins].bo = reloc->target; 194 job->unpins[job->num_unpins].sgt = sgt; 195 job->num_unpins++; 196 } 197 198 for (i = 0; i < job->num_gathers; i++) { 199 struct host1x_job_gather *g = &job->gathers[i]; 200 struct sg_table *sgt; 201 dma_addr_t phys_addr; 202 203 g->bo = host1x_bo_get(g->bo); 204 if (!g->bo) 205 goto unpin; 206 207 phys_addr = host1x_bo_pin(g->bo, &sgt); 208 if (!phys_addr) 209 goto unpin; 210 211 job->addr_phys[job->num_unpins] = phys_addr; 212 job->unpins[job->num_unpins].bo = g->bo; 213 job->unpins[job->num_unpins].sgt = sgt; 214 job->num_unpins++; 215 } 216 217 return job->num_unpins; 218 219 unpin: 220 host1x_job_unpin(job); 221 return 0; 222 } 223 224 static unsigned int do_relocs(struct host1x_job *job, struct host1x_bo *cmdbuf) 225 { 226 int i = 0; 227 u32 last_page = ~0; 228 void *cmdbuf_page_addr = NULL; 229 230 /* pin & patch the relocs for one gather */ 231 for (i = 0; i < job->num_relocs; i++) { 232 struct host1x_reloc *reloc = &job->relocarray[i]; 233 u32 reloc_addr = (job->reloc_addr_phys[i] + 234 reloc->target_offset) >> reloc->shift; 235 u32 *target; 236 237 /* skip all other gathers */ 238 if (cmdbuf != reloc->cmdbuf) 239 continue; 240 241 if (last_page != reloc->cmdbuf_offset >> PAGE_SHIFT) { 242 if (cmdbuf_page_addr) 243 host1x_bo_kunmap(cmdbuf, last_page, 244 cmdbuf_page_addr); 245 246 cmdbuf_page_addr = host1x_bo_kmap(cmdbuf, 247 reloc->cmdbuf_offset >> PAGE_SHIFT); 248 last_page = reloc->cmdbuf_offset >> PAGE_SHIFT; 249 250 if (unlikely(!cmdbuf_page_addr)) { 251 pr_err("Could not map cmdbuf for relocation\n"); 252 return -ENOMEM; 253 } 254 } 255 256 target = cmdbuf_page_addr + (reloc->cmdbuf_offset & ~PAGE_MASK); 257 *target = reloc_addr; 258 } 259 260 if (cmdbuf_page_addr) 261 host1x_bo_kunmap(cmdbuf, last_page, cmdbuf_page_addr); 262 263 return 0; 264 } 265 266 static bool check_reloc(struct host1x_reloc *reloc, struct host1x_bo *cmdbuf, 267 unsigned int offset) 268 { 269 offset *= sizeof(u32); 270 271 if (reloc->cmdbuf != cmdbuf || reloc->cmdbuf_offset != offset) 272 return false; 273 274 return true; 275 } 276 277 struct host1x_firewall { 278 struct host1x_job *job; 279 struct device *dev; 280 281 unsigned int num_relocs; 282 struct host1x_reloc *reloc; 283 284 struct host1x_bo *cmdbuf; 285 unsigned int offset; 286 287 u32 words; 288 u32 class; 289 u32 reg; 290 u32 mask; 291 u32 count; 292 }; 293 294 static int check_register(struct host1x_firewall *fw, unsigned long offset) 295 { 296 if (fw->job->is_addr_reg(fw->dev, fw->class, offset)) { 297 if (!fw->num_relocs) 298 return -EINVAL; 299 300 if (!check_reloc(fw->reloc, fw->cmdbuf, fw->offset)) 301 return -EINVAL; 302 303 fw->num_relocs--; 304 fw->reloc++; 305 } 306 307 return 0; 308 } 309 310 static int check_mask(struct host1x_firewall *fw) 311 { 312 u32 mask = fw->mask; 313 u32 reg = fw->reg; 314 int ret; 315 316 while (mask) { 317 if (fw->words == 0) 318 return -EINVAL; 319 320 if (mask & 1) { 321 ret = check_register(fw, reg); 322 if (ret < 0) 323 return ret; 324 325 fw->words--; 326 fw->offset++; 327 } 328 mask >>= 1; 329 reg++; 330 } 331 332 return 0; 333 } 334 335 static int check_incr(struct host1x_firewall *fw) 336 { 337 u32 count = fw->count; 338 u32 reg = fw->reg; 339 int ret; 340 341 while (count) { 342 if (fw->words == 0) 343 return -EINVAL; 344 345 ret = check_register(fw, reg); 346 if (ret < 0) 347 return ret; 348 349 reg++; 350 fw->words--; 351 fw->offset++; 352 count--; 353 } 354 355 return 0; 356 } 357 358 static int check_nonincr(struct host1x_firewall *fw) 359 { 360 u32 count = fw->count; 361 int ret; 362 363 while (count) { 364 if (fw->words == 0) 365 return -EINVAL; 366 367 ret = check_register(fw, fw->reg); 368 if (ret < 0) 369 return ret; 370 371 fw->words--; 372 fw->offset++; 373 count--; 374 } 375 376 return 0; 377 } 378 379 static int validate(struct host1x_firewall *fw, struct host1x_job_gather *g) 380 { 381 u32 *cmdbuf_base = (u32 *)fw->job->gather_copy_mapped + 382 (g->offset / sizeof(u32)); 383 int err = 0; 384 385 if (!fw->job->is_addr_reg) 386 return 0; 387 388 fw->words = g->words; 389 fw->cmdbuf = g->bo; 390 fw->offset = 0; 391 392 while (fw->words && !err) { 393 u32 word = cmdbuf_base[fw->offset]; 394 u32 opcode = (word & 0xf0000000) >> 28; 395 396 fw->mask = 0; 397 fw->reg = 0; 398 fw->count = 0; 399 fw->words--; 400 fw->offset++; 401 402 switch (opcode) { 403 case 0: 404 fw->class = word >> 6 & 0x3ff; 405 fw->mask = word & 0x3f; 406 fw->reg = word >> 16 & 0xfff; 407 err = check_mask(fw); 408 if (err) 409 goto out; 410 break; 411 case 1: 412 fw->reg = word >> 16 & 0xfff; 413 fw->count = word & 0xffff; 414 err = check_incr(fw); 415 if (err) 416 goto out; 417 break; 418 419 case 2: 420 fw->reg = word >> 16 & 0xfff; 421 fw->count = word & 0xffff; 422 err = check_nonincr(fw); 423 if (err) 424 goto out; 425 break; 426 427 case 3: 428 fw->mask = word & 0xffff; 429 fw->reg = word >> 16 & 0xfff; 430 err = check_mask(fw); 431 if (err) 432 goto out; 433 break; 434 case 4: 435 case 5: 436 case 14: 437 break; 438 default: 439 err = -EINVAL; 440 break; 441 } 442 } 443 444 out: 445 return err; 446 } 447 448 static inline int copy_gathers(struct host1x_job *job, struct device *dev) 449 { 450 struct host1x_firewall fw; 451 size_t size = 0; 452 size_t offset = 0; 453 int i; 454 455 fw.job = job; 456 fw.dev = dev; 457 fw.reloc = job->relocarray; 458 fw.num_relocs = job->num_relocs; 459 fw.class = 0; 460 461 for (i = 0; i < job->num_gathers; i++) { 462 struct host1x_job_gather *g = &job->gathers[i]; 463 size += g->words * sizeof(u32); 464 } 465 466 job->gather_copy_mapped = dma_alloc_writecombine(dev, size, 467 &job->gather_copy, 468 GFP_KERNEL); 469 if (!job->gather_copy_mapped) { 470 job->gather_copy_mapped = NULL; 471 return -ENOMEM; 472 } 473 474 job->gather_copy_size = size; 475 476 for (i = 0; i < job->num_gathers; i++) { 477 struct host1x_job_gather *g = &job->gathers[i]; 478 void *gather; 479 480 /* Copy the gather */ 481 gather = host1x_bo_mmap(g->bo); 482 memcpy(job->gather_copy_mapped + offset, gather + g->offset, 483 g->words * sizeof(u32)); 484 host1x_bo_munmap(g->bo, gather); 485 486 /* Store the location in the buffer */ 487 g->base = job->gather_copy; 488 g->offset = offset; 489 490 /* Validate the job */ 491 if (validate(&fw, g)) 492 return -EINVAL; 493 494 offset += g->words * sizeof(u32); 495 } 496 497 /* No relocs should remain at this point */ 498 if (fw.num_relocs) 499 return -EINVAL; 500 501 return 0; 502 } 503 504 int host1x_job_pin(struct host1x_job *job, struct device *dev) 505 { 506 int err; 507 unsigned int i, j; 508 struct host1x *host = dev_get_drvdata(dev->parent); 509 DECLARE_BITMAP(waitchk_mask, host1x_syncpt_nb_pts(host)); 510 511 bitmap_zero(waitchk_mask, host1x_syncpt_nb_pts(host)); 512 for (i = 0; i < job->num_waitchk; i++) { 513 u32 syncpt_id = job->waitchk[i].syncpt_id; 514 if (syncpt_id < host1x_syncpt_nb_pts(host)) 515 set_bit(syncpt_id, waitchk_mask); 516 } 517 518 /* get current syncpt values for waitchk */ 519 for_each_set_bit(i, waitchk_mask, host1x_syncpt_nb_pts(host)) 520 host1x_syncpt_load(host->syncpt + i); 521 522 /* pin memory */ 523 err = pin_job(job); 524 if (!err) 525 goto out; 526 527 /* patch gathers */ 528 for (i = 0; i < job->num_gathers; i++) { 529 struct host1x_job_gather *g = &job->gathers[i]; 530 531 /* process each gather mem only once */ 532 if (g->handled) 533 continue; 534 535 g->base = job->gather_addr_phys[i]; 536 537 for (j = 0; j < job->num_gathers; j++) 538 if (job->gathers[j].bo == g->bo) 539 job->gathers[j].handled = true; 540 541 err = do_relocs(job, g->bo); 542 if (err) 543 break; 544 545 err = do_waitchks(job, host, g->bo); 546 if (err) 547 break; 548 } 549 550 if (IS_ENABLED(CONFIG_TEGRA_HOST1X_FIREWALL) && !err) { 551 err = copy_gathers(job, dev); 552 if (err) { 553 host1x_job_unpin(job); 554 return err; 555 } 556 } 557 558 out: 559 wmb(); 560 561 return err; 562 } 563 564 void host1x_job_unpin(struct host1x_job *job) 565 { 566 unsigned int i; 567 568 for (i = 0; i < job->num_unpins; i++) { 569 struct host1x_job_unpin_data *unpin = &job->unpins[i]; 570 host1x_bo_unpin(unpin->bo, unpin->sgt); 571 host1x_bo_put(unpin->bo); 572 } 573 job->num_unpins = 0; 574 575 if (job->gather_copy_size) 576 dma_free_writecombine(job->channel->dev, job->gather_copy_size, 577 job->gather_copy_mapped, 578 job->gather_copy); 579 } 580 581 /* 582 * Debug routine used to dump job entries 583 */ 584 void host1x_job_dump(struct device *dev, struct host1x_job *job) 585 { 586 dev_dbg(dev, " SYNCPT_ID %d\n", job->syncpt_id); 587 dev_dbg(dev, " SYNCPT_VAL %d\n", job->syncpt_end); 588 dev_dbg(dev, " FIRST_GET 0x%x\n", job->first_get); 589 dev_dbg(dev, " TIMEOUT %d\n", job->timeout); 590 dev_dbg(dev, " NUM_SLOTS %d\n", job->num_slots); 591 dev_dbg(dev, " NUM_HANDLES %d\n", job->num_unpins); 592 } 593