1 /* 2 * Tegra host1x Job 3 * 4 * Copyright (c) 2010-2015, 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 EXPORT_SYMBOL(host1x_job_alloc); 79 80 struct host1x_job *host1x_job_get(struct host1x_job *job) 81 { 82 kref_get(&job->ref); 83 return job; 84 } 85 EXPORT_SYMBOL(host1x_job_get); 86 87 static void job_free(struct kref *ref) 88 { 89 struct host1x_job *job = container_of(ref, struct host1x_job, ref); 90 91 kfree(job); 92 } 93 94 void host1x_job_put(struct host1x_job *job) 95 { 96 kref_put(&job->ref, job_free); 97 } 98 EXPORT_SYMBOL(host1x_job_put); 99 100 void host1x_job_add_gather(struct host1x_job *job, struct host1x_bo *bo, 101 u32 words, u32 offset) 102 { 103 struct host1x_job_gather *cur_gather = &job->gathers[job->num_gathers]; 104 105 cur_gather->words = words; 106 cur_gather->bo = bo; 107 cur_gather->offset = offset; 108 job->num_gathers++; 109 } 110 EXPORT_SYMBOL(host1x_job_add_gather); 111 112 /* 113 * NULL an already satisfied WAIT_SYNCPT host method, by patching its 114 * args in the command stream. The method data is changed to reference 115 * a reserved (never given out or incr) HOST1X_SYNCPT_RESERVED syncpt 116 * with a matching threshold value of 0, so is guaranteed to be popped 117 * by the host HW. 118 */ 119 static void host1x_syncpt_patch_offset(struct host1x_syncpt *sp, 120 struct host1x_bo *h, u32 offset) 121 { 122 void *patch_addr = NULL; 123 124 /* patch the wait */ 125 patch_addr = host1x_bo_kmap(h, offset >> PAGE_SHIFT); 126 if (patch_addr) { 127 host1x_syncpt_patch_wait(sp, 128 patch_addr + (offset & ~PAGE_MASK)); 129 host1x_bo_kunmap(h, offset >> PAGE_SHIFT, patch_addr); 130 } else 131 pr_err("Could not map cmdbuf for wait check\n"); 132 } 133 134 /* 135 * Check driver supplied waitchk structs for syncpt thresholds 136 * that have already been satisfied and NULL the comparison (to 137 * avoid a wrap condition in the HW). 138 */ 139 static int do_waitchks(struct host1x_job *job, struct host1x *host, 140 struct host1x_job_gather *g) 141 { 142 struct host1x_bo *patch = g->bo; 143 int i; 144 145 /* compare syncpt vs wait threshold */ 146 for (i = 0; i < job->num_waitchk; i++) { 147 struct host1x_waitchk *wait = &job->waitchk[i]; 148 struct host1x_syncpt *sp = 149 host1x_syncpt_get(host, wait->syncpt_id); 150 151 /* validate syncpt id */ 152 if (wait->syncpt_id > host1x_syncpt_nb_pts(host)) 153 continue; 154 155 /* skip all other gathers */ 156 if (patch != wait->bo) 157 continue; 158 159 trace_host1x_syncpt_wait_check(wait->bo, wait->offset, 160 wait->syncpt_id, wait->thresh, 161 host1x_syncpt_read_min(sp)); 162 163 if (host1x_syncpt_is_expired(sp, wait->thresh)) { 164 dev_dbg(host->dev, 165 "drop WAIT id %u (%s) thresh 0x%x, min 0x%x\n", 166 wait->syncpt_id, sp->name, wait->thresh, 167 host1x_syncpt_read_min(sp)); 168 169 host1x_syncpt_patch_offset(sp, patch, 170 g->offset + wait->offset); 171 } 172 173 wait->bo = NULL; 174 } 175 176 return 0; 177 } 178 179 static unsigned int pin_job(struct host1x *host, struct host1x_job *job) 180 { 181 unsigned int i; 182 int err; 183 184 job->num_unpins = 0; 185 186 for (i = 0; i < job->num_relocs; i++) { 187 struct host1x_reloc *reloc = &job->relocarray[i]; 188 struct sg_table *sgt; 189 dma_addr_t phys_addr; 190 191 reloc->target.bo = host1x_bo_get(reloc->target.bo); 192 if (!reloc->target.bo) { 193 err = -EINVAL; 194 goto unpin; 195 } 196 197 phys_addr = host1x_bo_pin(reloc->target.bo, &sgt); 198 if (!phys_addr) { 199 err = -EINVAL; 200 goto unpin; 201 } 202 203 job->addr_phys[job->num_unpins] = phys_addr; 204 job->unpins[job->num_unpins].bo = reloc->target.bo; 205 job->unpins[job->num_unpins].sgt = sgt; 206 job->num_unpins++; 207 } 208 209 for (i = 0; i < job->num_gathers; i++) { 210 struct host1x_job_gather *g = &job->gathers[i]; 211 size_t gather_size = 0; 212 struct scatterlist *sg; 213 struct sg_table *sgt; 214 dma_addr_t phys_addr; 215 unsigned long shift; 216 struct iova *alloc; 217 unsigned int j; 218 219 g->bo = host1x_bo_get(g->bo); 220 if (!g->bo) { 221 err = -EINVAL; 222 goto unpin; 223 } 224 225 phys_addr = host1x_bo_pin(g->bo, &sgt); 226 if (!phys_addr) { 227 err = -EINVAL; 228 goto unpin; 229 } 230 231 if (!IS_ENABLED(CONFIG_TEGRA_HOST1X_FIREWALL) && host->domain) { 232 for_each_sg(sgt->sgl, sg, sgt->nents, j) 233 gather_size += sg->length; 234 gather_size = iova_align(&host->iova, gather_size); 235 236 shift = iova_shift(&host->iova); 237 alloc = alloc_iova(&host->iova, gather_size >> shift, 238 host->iova_end >> shift, true); 239 if (!alloc) { 240 err = -ENOMEM; 241 goto unpin; 242 } 243 244 err = iommu_map_sg(host->domain, 245 iova_dma_addr(&host->iova, alloc), 246 sgt->sgl, sgt->nents, IOMMU_READ); 247 if (err == 0) { 248 __free_iova(&host->iova, alloc); 249 err = -EINVAL; 250 goto unpin; 251 } 252 253 job->addr_phys[job->num_unpins] = 254 iova_dma_addr(&host->iova, alloc); 255 job->unpins[job->num_unpins].size = gather_size; 256 } else { 257 job->addr_phys[job->num_unpins] = phys_addr; 258 } 259 260 job->gather_addr_phys[i] = job->addr_phys[job->num_unpins]; 261 262 job->unpins[job->num_unpins].bo = g->bo; 263 job->unpins[job->num_unpins].sgt = sgt; 264 job->num_unpins++; 265 } 266 267 return 0; 268 269 unpin: 270 host1x_job_unpin(job); 271 return err; 272 } 273 274 static int do_relocs(struct host1x_job *job, struct host1x_job_gather *g) 275 { 276 int i = 0; 277 u32 last_page = ~0; 278 void *cmdbuf_page_addr = NULL; 279 struct host1x_bo *cmdbuf = g->bo; 280 281 /* pin & patch the relocs for one gather */ 282 for (i = 0; i < job->num_relocs; i++) { 283 struct host1x_reloc *reloc = &job->relocarray[i]; 284 u32 reloc_addr = (job->reloc_addr_phys[i] + 285 reloc->target.offset) >> reloc->shift; 286 u32 *target; 287 288 /* skip all other gathers */ 289 if (cmdbuf != reloc->cmdbuf.bo) 290 continue; 291 292 if (IS_ENABLED(CONFIG_TEGRA_HOST1X_FIREWALL)) { 293 target = (u32 *)job->gather_copy_mapped + 294 reloc->cmdbuf.offset / sizeof(u32) + 295 g->offset / sizeof(u32); 296 goto patch_reloc; 297 } 298 299 if (last_page != reloc->cmdbuf.offset >> PAGE_SHIFT) { 300 if (cmdbuf_page_addr) 301 host1x_bo_kunmap(cmdbuf, last_page, 302 cmdbuf_page_addr); 303 304 cmdbuf_page_addr = host1x_bo_kmap(cmdbuf, 305 reloc->cmdbuf.offset >> PAGE_SHIFT); 306 last_page = reloc->cmdbuf.offset >> PAGE_SHIFT; 307 308 if (unlikely(!cmdbuf_page_addr)) { 309 pr_err("Could not map cmdbuf for relocation\n"); 310 return -ENOMEM; 311 } 312 } 313 314 target = cmdbuf_page_addr + (reloc->cmdbuf.offset & ~PAGE_MASK); 315 patch_reloc: 316 *target = reloc_addr; 317 } 318 319 if (cmdbuf_page_addr) 320 host1x_bo_kunmap(cmdbuf, last_page, cmdbuf_page_addr); 321 322 return 0; 323 } 324 325 static bool check_reloc(struct host1x_reloc *reloc, struct host1x_bo *cmdbuf, 326 unsigned int offset) 327 { 328 offset *= sizeof(u32); 329 330 if (reloc->cmdbuf.bo != cmdbuf || reloc->cmdbuf.offset != offset) 331 return false; 332 333 return true; 334 } 335 336 struct host1x_firewall { 337 struct host1x_job *job; 338 struct device *dev; 339 340 unsigned int num_relocs; 341 struct host1x_reloc *reloc; 342 343 struct host1x_bo *cmdbuf; 344 unsigned int offset; 345 346 u32 words; 347 u32 class; 348 u32 reg; 349 u32 mask; 350 u32 count; 351 }; 352 353 static int check_register(struct host1x_firewall *fw, unsigned long offset) 354 { 355 if (fw->job->is_addr_reg(fw->dev, fw->class, offset)) { 356 if (!fw->num_relocs) 357 return -EINVAL; 358 359 if (!check_reloc(fw->reloc, fw->cmdbuf, fw->offset)) 360 return -EINVAL; 361 362 fw->num_relocs--; 363 fw->reloc++; 364 } 365 366 return 0; 367 } 368 369 static int check_mask(struct host1x_firewall *fw) 370 { 371 u32 mask = fw->mask; 372 u32 reg = fw->reg; 373 int ret; 374 375 while (mask) { 376 if (fw->words == 0) 377 return -EINVAL; 378 379 if (mask & 1) { 380 ret = check_register(fw, reg); 381 if (ret < 0) 382 return ret; 383 384 fw->words--; 385 fw->offset++; 386 } 387 mask >>= 1; 388 reg++; 389 } 390 391 return 0; 392 } 393 394 static int check_incr(struct host1x_firewall *fw) 395 { 396 u32 count = fw->count; 397 u32 reg = fw->reg; 398 int ret; 399 400 while (count) { 401 if (fw->words == 0) 402 return -EINVAL; 403 404 ret = check_register(fw, reg); 405 if (ret < 0) 406 return ret; 407 408 reg++; 409 fw->words--; 410 fw->offset++; 411 count--; 412 } 413 414 return 0; 415 } 416 417 static int check_nonincr(struct host1x_firewall *fw) 418 { 419 u32 count = fw->count; 420 int ret; 421 422 while (count) { 423 if (fw->words == 0) 424 return -EINVAL; 425 426 ret = check_register(fw, fw->reg); 427 if (ret < 0) 428 return ret; 429 430 fw->words--; 431 fw->offset++; 432 count--; 433 } 434 435 return 0; 436 } 437 438 static int validate(struct host1x_firewall *fw, struct host1x_job_gather *g) 439 { 440 u32 *cmdbuf_base = (u32 *)fw->job->gather_copy_mapped + 441 (g->offset / sizeof(u32)); 442 int err = 0; 443 444 if (!fw->job->is_addr_reg) 445 return 0; 446 447 fw->words = g->words; 448 fw->cmdbuf = g->bo; 449 fw->offset = 0; 450 451 while (fw->words && !err) { 452 u32 word = cmdbuf_base[fw->offset]; 453 u32 opcode = (word & 0xf0000000) >> 28; 454 455 fw->mask = 0; 456 fw->reg = 0; 457 fw->count = 0; 458 fw->words--; 459 fw->offset++; 460 461 switch (opcode) { 462 case 0: 463 fw->class = word >> 6 & 0x3ff; 464 fw->mask = word & 0x3f; 465 fw->reg = word >> 16 & 0xfff; 466 err = check_mask(fw); 467 if (err) 468 goto out; 469 break; 470 case 1: 471 fw->reg = word >> 16 & 0xfff; 472 fw->count = word & 0xffff; 473 err = check_incr(fw); 474 if (err) 475 goto out; 476 break; 477 478 case 2: 479 fw->reg = word >> 16 & 0xfff; 480 fw->count = word & 0xffff; 481 err = check_nonincr(fw); 482 if (err) 483 goto out; 484 break; 485 486 case 3: 487 fw->mask = word & 0xffff; 488 fw->reg = word >> 16 & 0xfff; 489 err = check_mask(fw); 490 if (err) 491 goto out; 492 break; 493 case 4: 494 case 5: 495 case 14: 496 break; 497 default: 498 err = -EINVAL; 499 break; 500 } 501 } 502 503 out: 504 return err; 505 } 506 507 static inline int copy_gathers(struct host1x_job *job, struct device *dev) 508 { 509 struct host1x_firewall fw; 510 size_t size = 0; 511 size_t offset = 0; 512 int i; 513 514 fw.job = job; 515 fw.dev = dev; 516 fw.reloc = job->relocarray; 517 fw.num_relocs = job->num_relocs; 518 fw.class = job->class; 519 520 for (i = 0; i < job->num_gathers; i++) { 521 struct host1x_job_gather *g = &job->gathers[i]; 522 523 size += g->words * sizeof(u32); 524 } 525 526 job->gather_copy_mapped = dma_alloc_wc(dev, size, &job->gather_copy, 527 GFP_KERNEL); 528 if (!job->gather_copy_mapped) { 529 job->gather_copy_mapped = NULL; 530 return -ENOMEM; 531 } 532 533 job->gather_copy_size = size; 534 535 for (i = 0; i < job->num_gathers; i++) { 536 struct host1x_job_gather *g = &job->gathers[i]; 537 void *gather; 538 539 /* Copy the gather */ 540 gather = host1x_bo_mmap(g->bo); 541 memcpy(job->gather_copy_mapped + offset, gather + g->offset, 542 g->words * sizeof(u32)); 543 host1x_bo_munmap(g->bo, gather); 544 545 /* Store the location in the buffer */ 546 g->base = job->gather_copy; 547 g->offset = offset; 548 549 /* Validate the job */ 550 if (validate(&fw, g)) 551 return -EINVAL; 552 553 offset += g->words * sizeof(u32); 554 } 555 556 /* No relocs should remain at this point */ 557 if (fw.num_relocs) 558 return -EINVAL; 559 560 return 0; 561 } 562 563 int host1x_job_pin(struct host1x_job *job, struct device *dev) 564 { 565 int err; 566 unsigned int i, j; 567 struct host1x *host = dev_get_drvdata(dev->parent); 568 DECLARE_BITMAP(waitchk_mask, host1x_syncpt_nb_pts(host)); 569 570 bitmap_zero(waitchk_mask, host1x_syncpt_nb_pts(host)); 571 for (i = 0; i < job->num_waitchk; i++) { 572 u32 syncpt_id = job->waitchk[i].syncpt_id; 573 574 if (syncpt_id < host1x_syncpt_nb_pts(host)) 575 set_bit(syncpt_id, waitchk_mask); 576 } 577 578 /* get current syncpt values for waitchk */ 579 for_each_set_bit(i, waitchk_mask, host1x_syncpt_nb_pts(host)) 580 host1x_syncpt_load(host->syncpt + i); 581 582 /* pin memory */ 583 err = pin_job(host, job); 584 if (err) 585 goto out; 586 587 if (IS_ENABLED(CONFIG_TEGRA_HOST1X_FIREWALL)) { 588 err = copy_gathers(job, dev); 589 if (err) 590 goto out; 591 } 592 593 /* patch gathers */ 594 for (i = 0; i < job->num_gathers; i++) { 595 struct host1x_job_gather *g = &job->gathers[i]; 596 597 /* process each gather mem only once */ 598 if (g->handled) 599 continue; 600 601 /* copy_gathers() sets gathers base if firewall is enabled */ 602 if (!IS_ENABLED(CONFIG_TEGRA_HOST1X_FIREWALL)) 603 g->base = job->gather_addr_phys[i]; 604 605 for (j = i + 1; j < job->num_gathers; j++) { 606 if (job->gathers[j].bo == g->bo) { 607 job->gathers[j].handled = true; 608 job->gathers[j].base = g->base; 609 } 610 } 611 612 err = do_relocs(job, g); 613 if (err) 614 break; 615 616 err = do_waitchks(job, host, g); 617 if (err) 618 break; 619 } 620 621 out: 622 if (err) 623 host1x_job_unpin(job); 624 wmb(); 625 626 return err; 627 } 628 EXPORT_SYMBOL(host1x_job_pin); 629 630 void host1x_job_unpin(struct host1x_job *job) 631 { 632 struct host1x *host = dev_get_drvdata(job->channel->dev->parent); 633 unsigned int i; 634 635 for (i = 0; i < job->num_unpins; i++) { 636 struct host1x_job_unpin_data *unpin = &job->unpins[i]; 637 638 if (!IS_ENABLED(CONFIG_TEGRA_HOST1X_FIREWALL) && host->domain) { 639 iommu_unmap(host->domain, job->addr_phys[i], 640 unpin->size); 641 free_iova(&host->iova, 642 iova_pfn(&host->iova, job->addr_phys[i])); 643 } 644 645 host1x_bo_unpin(unpin->bo, unpin->sgt); 646 host1x_bo_put(unpin->bo); 647 } 648 649 job->num_unpins = 0; 650 651 if (job->gather_copy_size) 652 dma_free_wc(job->channel->dev, job->gather_copy_size, 653 job->gather_copy_mapped, job->gather_copy); 654 } 655 EXPORT_SYMBOL(host1x_job_unpin); 656 657 /* 658 * Debug routine used to dump job entries 659 */ 660 void host1x_job_dump(struct device *dev, struct host1x_job *job) 661 { 662 dev_dbg(dev, " SYNCPT_ID %d\n", job->syncpt_id); 663 dev_dbg(dev, " SYNCPT_VAL %d\n", job->syncpt_end); 664 dev_dbg(dev, " FIRST_GET 0x%x\n", job->first_get); 665 dev_dbg(dev, " TIMEOUT %d\n", job->timeout); 666 dev_dbg(dev, " NUM_SLOTS %d\n", job->num_slots); 667 dev_dbg(dev, " NUM_HANDLES %d\n", job->num_unpins); 668 } 669