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