1 /* 2 * Copyright 2016 Advanced Micro Devices, Inc. 3 * 4 * Permission is hereby granted, free of charge, to any person obtaining a 5 * copy of this software and associated documentation files (the "Software"), 6 * to deal in the Software without restriction, including without limitation 7 * the rights to use, copy, modify, merge, publish, distribute, sublicense, 8 * and/or sell copies of the Software, and to permit persons to whom the 9 * Software is furnished to do so, subject to the following conditions: 10 * 11 * The above copyright notice and this permission notice shall be included in 12 * all copies or substantial portions of the Software. 13 * 14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 15 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 16 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 17 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR 18 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, 19 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR 20 * OTHER DEALINGS IN THE SOFTWARE. 21 * 22 * Authors: Christian König 23 */ 24 25 #include <linux/dma-mapping.h> 26 #include <drm/ttm/ttm_range_manager.h> 27 #include <drm/drm_drv.h> 28 #include <drm/drm_buddy.h> 29 30 #include "amdgpu.h" 31 #include "amdgpu_vm.h" 32 #include "amdgpu_res_cursor.h" 33 #include "atom.h" 34 35 #define AMDGPU_MAX_SG_SEGMENT_SIZE (2UL << 30) 36 37 struct amdgpu_vram_reservation { 38 u64 start; 39 u64 size; 40 struct list_head allocated; 41 struct list_head blocks; 42 }; 43 44 static inline struct amdgpu_vram_mgr * 45 to_vram_mgr(struct ttm_resource_manager *man) 46 { 47 return container_of(man, struct amdgpu_vram_mgr, manager); 48 } 49 50 static inline struct amdgpu_device * 51 to_amdgpu_device(struct amdgpu_vram_mgr *mgr) 52 { 53 return container_of(mgr, struct amdgpu_device, mman.vram_mgr); 54 } 55 56 static inline struct gpu_buddy_block * 57 amdgpu_vram_mgr_first_block(struct list_head *list) 58 { 59 return list_first_entry_or_null(list, struct gpu_buddy_block, link); 60 } 61 62 static inline bool amdgpu_is_vram_mgr_blocks_contiguous(struct list_head *head) 63 { 64 struct gpu_buddy_block *block; 65 u64 start, size; 66 67 block = amdgpu_vram_mgr_first_block(head); 68 if (!block) 69 return false; 70 71 while (head != block->link.next) { 72 start = amdgpu_vram_mgr_block_start(block); 73 size = amdgpu_vram_mgr_block_size(block); 74 75 block = list_entry(block->link.next, struct gpu_buddy_block, link); 76 if (start + size != amdgpu_vram_mgr_block_start(block)) 77 return false; 78 } 79 80 return true; 81 } 82 83 static inline u64 amdgpu_vram_mgr_blocks_size(struct list_head *head) 84 { 85 struct gpu_buddy_block *block; 86 u64 size = 0; 87 88 list_for_each_entry(block, head, link) 89 size += amdgpu_vram_mgr_block_size(block); 90 91 return size; 92 } 93 94 /** 95 * DOC: mem_info_vram_total 96 * 97 * The amdgpu driver provides a sysfs API for reporting current total VRAM 98 * available on the device 99 * The file mem_info_vram_total is used for this and returns the total 100 * amount of VRAM in bytes 101 */ 102 static ssize_t amdgpu_mem_info_vram_total_show(struct device *dev, 103 struct device_attribute *attr, char *buf) 104 { 105 struct drm_device *ddev = dev_get_drvdata(dev); 106 struct amdgpu_device *adev = drm_to_adev(ddev); 107 108 return sysfs_emit(buf, "%llu\n", adev->gmc.real_vram_size); 109 } 110 111 /** 112 * DOC: mem_info_vis_vram_total 113 * 114 * The amdgpu driver provides a sysfs API for reporting current total 115 * visible VRAM available on the device 116 * The file mem_info_vis_vram_total is used for this and returns the total 117 * amount of visible VRAM in bytes 118 */ 119 static ssize_t amdgpu_mem_info_vis_vram_total_show(struct device *dev, 120 struct device_attribute *attr, char *buf) 121 { 122 struct drm_device *ddev = dev_get_drvdata(dev); 123 struct amdgpu_device *adev = drm_to_adev(ddev); 124 125 return sysfs_emit(buf, "%llu\n", adev->gmc.visible_vram_size); 126 } 127 128 /** 129 * DOC: mem_info_vram_used 130 * 131 * The amdgpu driver provides a sysfs API for reporting current total VRAM 132 * available on the device 133 * The file mem_info_vram_used is used for this and returns the total 134 * amount of currently used VRAM in bytes 135 */ 136 static ssize_t amdgpu_mem_info_vram_used_show(struct device *dev, 137 struct device_attribute *attr, 138 char *buf) 139 { 140 struct drm_device *ddev = dev_get_drvdata(dev); 141 struct amdgpu_device *adev = drm_to_adev(ddev); 142 struct ttm_resource_manager *man = &adev->mman.vram_mgr.manager; 143 144 return sysfs_emit(buf, "%llu\n", ttm_resource_manager_usage(man)); 145 } 146 147 /** 148 * DOC: mem_info_vis_vram_used 149 * 150 * The amdgpu driver provides a sysfs API for reporting current total of 151 * used visible VRAM 152 * The file mem_info_vis_vram_used is used for this and returns the total 153 * amount of currently used visible VRAM in bytes 154 */ 155 static ssize_t amdgpu_mem_info_vis_vram_used_show(struct device *dev, 156 struct device_attribute *attr, 157 char *buf) 158 { 159 struct drm_device *ddev = dev_get_drvdata(dev); 160 struct amdgpu_device *adev = drm_to_adev(ddev); 161 162 return sysfs_emit(buf, "%llu\n", 163 amdgpu_vram_mgr_vis_usage(&adev->mman.vram_mgr)); 164 } 165 166 /** 167 * DOC: mem_info_vram_vendor 168 * 169 * The amdgpu driver provides a sysfs API for reporting the vendor of the 170 * installed VRAM 171 * The file mem_info_vram_vendor is used for this and returns the name of the 172 * vendor. 173 */ 174 static ssize_t amdgpu_mem_info_vram_vendor(struct device *dev, 175 struct device_attribute *attr, 176 char *buf) 177 { 178 struct drm_device *ddev = dev_get_drvdata(dev); 179 struct amdgpu_device *adev = drm_to_adev(ddev); 180 181 switch (adev->gmc.vram_vendor) { 182 case SAMSUNG: 183 return sysfs_emit(buf, "samsung\n"); 184 case INFINEON: 185 return sysfs_emit(buf, "infineon\n"); 186 case ELPIDA: 187 return sysfs_emit(buf, "elpida\n"); 188 case ETRON: 189 return sysfs_emit(buf, "etron\n"); 190 case NANYA: 191 return sysfs_emit(buf, "nanya\n"); 192 case HYNIX: 193 return sysfs_emit(buf, "hynix\n"); 194 case MOSEL: 195 return sysfs_emit(buf, "mosel\n"); 196 case WINBOND: 197 return sysfs_emit(buf, "winbond\n"); 198 case ESMT: 199 return sysfs_emit(buf, "esmt\n"); 200 case MICRON: 201 return sysfs_emit(buf, "micron\n"); 202 default: 203 return sysfs_emit(buf, "unknown\n"); 204 } 205 } 206 207 static DEVICE_ATTR(mem_info_vram_total, S_IRUGO, 208 amdgpu_mem_info_vram_total_show, NULL); 209 static DEVICE_ATTR(mem_info_vis_vram_total, S_IRUGO, 210 amdgpu_mem_info_vis_vram_total_show,NULL); 211 static DEVICE_ATTR(mem_info_vram_used, S_IRUGO, 212 amdgpu_mem_info_vram_used_show, NULL); 213 static DEVICE_ATTR(mem_info_vis_vram_used, S_IRUGO, 214 amdgpu_mem_info_vis_vram_used_show, NULL); 215 static DEVICE_ATTR(mem_info_vram_vendor, S_IRUGO, 216 amdgpu_mem_info_vram_vendor, NULL); 217 218 static struct attribute *amdgpu_vram_mgr_attributes[] = { 219 &dev_attr_mem_info_vram_total.attr, 220 &dev_attr_mem_info_vis_vram_total.attr, 221 &dev_attr_mem_info_vram_used.attr, 222 &dev_attr_mem_info_vis_vram_used.attr, 223 &dev_attr_mem_info_vram_vendor.attr, 224 NULL 225 }; 226 227 static umode_t amdgpu_vram_attrs_is_visible(struct kobject *kobj, 228 struct attribute *attr, int i) 229 { 230 struct device *dev = kobj_to_dev(kobj); 231 struct drm_device *ddev = dev_get_drvdata(dev); 232 struct amdgpu_device *adev = drm_to_adev(ddev); 233 234 if (attr == &dev_attr_mem_info_vram_vendor.attr && 235 !adev->gmc.vram_vendor) 236 return 0; 237 238 if (!ttm_resource_manager_used(&adev->mman.vram_mgr.manager)) 239 return 0; 240 241 return attr->mode; 242 } 243 244 const struct attribute_group amdgpu_vram_mgr_attr_group = { 245 .attrs = amdgpu_vram_mgr_attributes, 246 .is_visible = amdgpu_vram_attrs_is_visible 247 }; 248 249 /** 250 * amdgpu_vram_mgr_vis_size - Calculate visible block size 251 * 252 * @adev: amdgpu_device pointer 253 * @block: DRM BUDDY block structure 254 * 255 * Calculate how many bytes of the DRM BUDDY block are inside visible VRAM 256 */ 257 static u64 amdgpu_vram_mgr_vis_size(struct amdgpu_device *adev, 258 struct gpu_buddy_block *block) 259 { 260 u64 start = amdgpu_vram_mgr_block_start(block); 261 u64 end = start + amdgpu_vram_mgr_block_size(block); 262 263 if (start >= adev->gmc.visible_vram_size) 264 return 0; 265 266 return (end > adev->gmc.visible_vram_size ? 267 adev->gmc.visible_vram_size : end) - start; 268 } 269 270 /** 271 * amdgpu_vram_mgr_bo_visible_size - CPU visible BO size 272 * 273 * @bo: &amdgpu_bo buffer object (must be in VRAM) 274 * 275 * Returns: 276 * How much of the given &amdgpu_bo buffer object lies in CPU visible VRAM. 277 */ 278 u64 amdgpu_vram_mgr_bo_visible_size(struct amdgpu_bo *bo) 279 { 280 struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev); 281 struct ttm_resource *res = bo->tbo.resource; 282 struct amdgpu_vram_mgr_resource *vres = to_amdgpu_vram_mgr_resource(res); 283 struct gpu_buddy_block *block; 284 u64 usage = 0; 285 286 if (amdgpu_gmc_vram_full_visible(&adev->gmc)) 287 return amdgpu_bo_size(bo); 288 289 if (res->start >= adev->gmc.visible_vram_size >> PAGE_SHIFT) 290 return 0; 291 292 list_for_each_entry(block, &vres->blocks, link) 293 usage += amdgpu_vram_mgr_vis_size(adev, block); 294 295 return usage; 296 } 297 298 /* Commit the reservation of VRAM pages */ 299 static void amdgpu_vram_mgr_do_reserve(struct ttm_resource_manager *man) 300 { 301 struct amdgpu_vram_mgr *mgr = to_vram_mgr(man); 302 struct amdgpu_device *adev = to_amdgpu_device(mgr); 303 struct gpu_buddy *mm = &mgr->mm; 304 struct amdgpu_vram_reservation *rsv, *temp; 305 struct gpu_buddy_block *block; 306 uint64_t vis_usage; 307 308 list_for_each_entry_safe(rsv, temp, &mgr->reservations_pending, blocks) { 309 if (gpu_buddy_alloc_blocks(mm, rsv->start, rsv->start + rsv->size, 310 rsv->size, mm->chunk_size, &rsv->allocated, 311 GPU_BUDDY_RANGE_ALLOCATION)) 312 continue; 313 314 block = amdgpu_vram_mgr_first_block(&rsv->allocated); 315 if (!block) 316 continue; 317 318 dev_dbg(adev->dev, "Reservation 0x%llx - %lld, Succeeded\n", 319 rsv->start, rsv->size); 320 321 vis_usage = amdgpu_vram_mgr_vis_size(adev, block); 322 atomic64_add(vis_usage, &mgr->vis_usage); 323 spin_lock(&man->bdev->lru_lock); 324 man->usage += rsv->size; 325 spin_unlock(&man->bdev->lru_lock); 326 list_move(&rsv->blocks, &mgr->reserved_pages); 327 } 328 } 329 330 /** 331 * amdgpu_vram_mgr_reserve_range - Reserve a range from VRAM 332 * 333 * @mgr: amdgpu_vram_mgr pointer 334 * @start: start address of the range in VRAM 335 * @size: size of the range 336 * 337 * Reserve memory from start address with the specified size in VRAM 338 */ 339 int amdgpu_vram_mgr_reserve_range(struct amdgpu_vram_mgr *mgr, 340 uint64_t start, uint64_t size) 341 { 342 struct amdgpu_vram_reservation *rsv; 343 344 rsv = kzalloc_obj(*rsv); 345 if (!rsv) 346 return -ENOMEM; 347 348 INIT_LIST_HEAD(&rsv->allocated); 349 INIT_LIST_HEAD(&rsv->blocks); 350 351 rsv->start = start; 352 rsv->size = size; 353 354 mutex_lock(&mgr->lock); 355 list_add_tail(&rsv->blocks, &mgr->reservations_pending); 356 amdgpu_vram_mgr_do_reserve(&mgr->manager); 357 mutex_unlock(&mgr->lock); 358 359 return 0; 360 } 361 362 /** 363 * amdgpu_vram_mgr_query_page_status - query the reservation status 364 * 365 * @mgr: amdgpu_vram_mgr pointer 366 * @start: start address of a page in VRAM 367 * 368 * Returns: 369 * -EBUSY: the page is still hold and in pending list 370 * 0: the page has been reserved 371 * -ENOENT: the input page is not a reservation 372 */ 373 int amdgpu_vram_mgr_query_page_status(struct amdgpu_vram_mgr *mgr, 374 uint64_t start) 375 { 376 struct amdgpu_vram_reservation *rsv; 377 int ret; 378 379 mutex_lock(&mgr->lock); 380 381 list_for_each_entry(rsv, &mgr->reservations_pending, blocks) { 382 if (rsv->start <= start && 383 (start < (rsv->start + rsv->size))) { 384 ret = -EBUSY; 385 goto out; 386 } 387 } 388 389 list_for_each_entry(rsv, &mgr->reserved_pages, blocks) { 390 if (rsv->start <= start && 391 (start < (rsv->start + rsv->size))) { 392 ret = 0; 393 goto out; 394 } 395 } 396 397 ret = -ENOENT; 398 out: 399 mutex_unlock(&mgr->lock); 400 return ret; 401 } 402 403 int amdgpu_vram_mgr_query_address_block_info(struct amdgpu_vram_mgr *mgr, 404 uint64_t address, struct amdgpu_vram_block_info *info) 405 { 406 struct amdgpu_vram_mgr_resource *vres; 407 struct gpu_buddy_block *block; 408 u64 start, size; 409 int ret = -ENOENT; 410 411 mutex_lock(&mgr->lock); 412 list_for_each_entry(vres, &mgr->allocated_vres_list, vres_node) { 413 list_for_each_entry(block, &vres->blocks, link) { 414 start = amdgpu_vram_mgr_block_start(block); 415 size = amdgpu_vram_mgr_block_size(block); 416 if ((start <= address) && (address < (start + size))) { 417 info->start = start; 418 info->size = size; 419 memcpy(&info->task, &vres->task, sizeof(vres->task)); 420 ret = 0; 421 goto out; 422 } 423 } 424 } 425 426 out: 427 mutex_unlock(&mgr->lock); 428 429 return ret; 430 } 431 432 /** 433 * amdgpu_vram_mgr_new - allocate new ranges 434 * 435 * @man: TTM memory type manager 436 * @tbo: TTM BO we need this range for 437 * @place: placement flags and restrictions 438 * @res: the resulting mem object 439 * 440 * Allocate VRAM for the given BO. 441 */ 442 static int amdgpu_vram_mgr_new(struct ttm_resource_manager *man, 443 struct ttm_buffer_object *tbo, 444 const struct ttm_place *place, 445 struct ttm_resource **res) 446 { 447 struct amdgpu_vram_mgr *mgr = to_vram_mgr(man); 448 struct amdgpu_device *adev = to_amdgpu_device(mgr); 449 struct amdgpu_bo *bo = ttm_to_amdgpu_bo(tbo); 450 u64 vis_usage = 0, max_bytes, min_block_size; 451 struct amdgpu_vram_mgr_resource *vres; 452 u64 size, remaining_size, lpfn, fpfn; 453 unsigned int adjust_dcc_size = 0; 454 struct gpu_buddy *mm = &mgr->mm; 455 struct gpu_buddy_block *block; 456 unsigned long pages_per_block; 457 int r; 458 459 lpfn = (u64)place->lpfn << PAGE_SHIFT; 460 if (!lpfn || lpfn > man->size) 461 lpfn = man->size; 462 463 fpfn = (u64)place->fpfn << PAGE_SHIFT; 464 465 max_bytes = adev->gmc.mc_vram_size; 466 if (tbo->type != ttm_bo_type_kernel) 467 max_bytes -= AMDGPU_VM_RESERVED_VRAM; 468 469 if (bo->flags & AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS) { 470 pages_per_block = ~0ul; 471 } else { 472 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 473 pages_per_block = HPAGE_PMD_NR; 474 #else 475 /* default to 2MB */ 476 pages_per_block = 2UL << (20UL - PAGE_SHIFT); 477 #endif 478 pages_per_block = max_t(u32, pages_per_block, 479 tbo->page_alignment); 480 } 481 482 vres = kzalloc_obj(*vres); 483 if (!vres) 484 return -ENOMEM; 485 486 ttm_resource_init(tbo, place, &vres->base); 487 488 /* bail out quickly if there's likely not enough VRAM for this BO */ 489 if (ttm_resource_manager_usage(man) > max_bytes) { 490 r = -ENOSPC; 491 goto error_fini; 492 } 493 494 INIT_LIST_HEAD(&vres->blocks); 495 496 if (place->flags & TTM_PL_FLAG_TOPDOWN) 497 vres->flags |= GPU_BUDDY_TOPDOWN_ALLOCATION; 498 499 if (bo->flags & AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS) 500 vres->flags |= GPU_BUDDY_CONTIGUOUS_ALLOCATION; 501 502 if (bo->flags & AMDGPU_GEM_CREATE_VRAM_CLEARED) 503 vres->flags |= GPU_BUDDY_CLEAR_ALLOCATION; 504 505 if (fpfn || lpfn != mgr->mm.size) 506 /* Allocate blocks in desired range */ 507 vres->flags |= GPU_BUDDY_RANGE_ALLOCATION; 508 509 if (bo->flags & AMDGPU_GEM_CREATE_GFX12_DCC && 510 adev->gmc.gmc_funcs->get_dcc_alignment) 511 adjust_dcc_size = amdgpu_gmc_get_dcc_alignment(adev); 512 513 remaining_size = (u64)vres->base.size; 514 if (bo->flags & AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS && adjust_dcc_size) { 515 unsigned int dcc_size; 516 517 dcc_size = roundup_pow_of_two(vres->base.size + adjust_dcc_size); 518 remaining_size = (u64)dcc_size; 519 520 vres->flags |= GPU_BUDDY_TRIM_DISABLE; 521 } 522 523 mutex_lock(&mgr->lock); 524 while (remaining_size) { 525 if (tbo->page_alignment) 526 min_block_size = (u64)tbo->page_alignment << PAGE_SHIFT; 527 else 528 min_block_size = mgr->default_page_size; 529 530 size = remaining_size; 531 532 if (bo->flags & AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS && adjust_dcc_size) 533 min_block_size = size; 534 else if ((size >= (u64)pages_per_block << PAGE_SHIFT) && 535 !(size & (((u64)pages_per_block << PAGE_SHIFT) - 1))) 536 min_block_size = (u64)pages_per_block << PAGE_SHIFT; 537 538 BUG_ON(min_block_size < mm->chunk_size); 539 540 r = gpu_buddy_alloc_blocks(mm, fpfn, 541 lpfn, 542 size, 543 min_block_size, 544 &vres->blocks, 545 vres->flags); 546 547 if (unlikely(r == -ENOSPC) && pages_per_block == ~0ul && 548 !(place->flags & TTM_PL_FLAG_CONTIGUOUS)) { 549 vres->flags &= ~GPU_BUDDY_CONTIGUOUS_ALLOCATION; 550 pages_per_block = max_t(u32, 2UL << (20UL - PAGE_SHIFT), 551 tbo->page_alignment); 552 553 continue; 554 } 555 556 if (unlikely(r)) 557 goto error_free_blocks; 558 559 if (size > remaining_size) 560 remaining_size = 0; 561 else 562 remaining_size -= size; 563 } 564 565 vres->task.pid = task_pid_nr(current); 566 get_task_comm(vres->task.comm, current); 567 list_add_tail(&vres->vres_node, &mgr->allocated_vres_list); 568 569 if (bo->flags & AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS && adjust_dcc_size) { 570 struct gpu_buddy_block *dcc_block; 571 unsigned long dcc_start; 572 u64 trim_start; 573 574 dcc_block = amdgpu_vram_mgr_first_block(&vres->blocks); 575 /* Adjust the start address for DCC buffers only */ 576 dcc_start = 577 roundup((unsigned long)amdgpu_vram_mgr_block_start(dcc_block), 578 adjust_dcc_size); 579 trim_start = (u64)dcc_start; 580 gpu_buddy_block_trim(mm, &trim_start, 581 (u64)vres->base.size, 582 &vres->blocks); 583 } 584 mutex_unlock(&mgr->lock); 585 586 vres->base.start = 0; 587 size = max_t(u64, amdgpu_vram_mgr_blocks_size(&vres->blocks), 588 vres->base.size); 589 list_for_each_entry(block, &vres->blocks, link) { 590 unsigned long start; 591 592 start = amdgpu_vram_mgr_block_start(block) + 593 amdgpu_vram_mgr_block_size(block); 594 start >>= PAGE_SHIFT; 595 596 if (start > PFN_UP(size)) 597 start -= PFN_UP(size); 598 else 599 start = 0; 600 vres->base.start = max(vres->base.start, start); 601 602 vis_usage += amdgpu_vram_mgr_vis_size(adev, block); 603 } 604 605 if (amdgpu_is_vram_mgr_blocks_contiguous(&vres->blocks)) 606 vres->base.placement |= TTM_PL_FLAG_CONTIGUOUS; 607 608 if (adev->gmc.xgmi.connected_to_cpu) 609 vres->base.bus.caching = ttm_cached; 610 else 611 vres->base.bus.caching = ttm_write_combined; 612 613 atomic64_add(vis_usage, &mgr->vis_usage); 614 *res = &vres->base; 615 return 0; 616 617 error_free_blocks: 618 gpu_buddy_free_list(mm, &vres->blocks, 0); 619 mutex_unlock(&mgr->lock); 620 error_fini: 621 ttm_resource_fini(man, &vres->base); 622 kfree(vres); 623 624 return r; 625 } 626 627 /** 628 * amdgpu_vram_mgr_del - free ranges 629 * 630 * @man: TTM memory type manager 631 * @res: TTM memory object 632 * 633 * Free the allocated VRAM again. 634 */ 635 static void amdgpu_vram_mgr_del(struct ttm_resource_manager *man, 636 struct ttm_resource *res) 637 { 638 struct amdgpu_vram_mgr_resource *vres = to_amdgpu_vram_mgr_resource(res); 639 struct amdgpu_vram_mgr *mgr = to_vram_mgr(man); 640 struct amdgpu_device *adev = to_amdgpu_device(mgr); 641 struct gpu_buddy *mm = &mgr->mm; 642 struct gpu_buddy_block *block; 643 uint64_t vis_usage = 0; 644 645 mutex_lock(&mgr->lock); 646 647 list_del(&vres->vres_node); 648 memset(&vres->task, 0, sizeof(vres->task)); 649 650 list_for_each_entry(block, &vres->blocks, link) 651 vis_usage += amdgpu_vram_mgr_vis_size(adev, block); 652 653 gpu_buddy_free_list(mm, &vres->blocks, vres->flags); 654 amdgpu_vram_mgr_do_reserve(man); 655 mutex_unlock(&mgr->lock); 656 657 atomic64_sub(vis_usage, &mgr->vis_usage); 658 659 ttm_resource_fini(man, res); 660 kfree(vres); 661 } 662 663 /** 664 * amdgpu_vram_mgr_alloc_sgt - allocate and fill a sg table 665 * 666 * @adev: amdgpu device pointer 667 * @res: TTM memory object 668 * @offset: byte offset from the base of VRAM BO 669 * @length: number of bytes to export in sg_table 670 * @dev: the other device 671 * @dir: dma direction 672 * @sgt: resulting sg table 673 * 674 * Allocate and fill a sg table from a VRAM allocation. 675 */ 676 int amdgpu_vram_mgr_alloc_sgt(struct amdgpu_device *adev, 677 struct ttm_resource *res, 678 u64 offset, u64 length, 679 struct device *dev, 680 enum dma_data_direction dir, 681 struct sg_table **sgt) 682 { 683 struct amdgpu_res_cursor cursor; 684 struct scatterlist *sg; 685 int num_entries = 0; 686 int i, r; 687 688 *sgt = kmalloc_obj(**sgt); 689 if (!*sgt) 690 return -ENOMEM; 691 692 /* Determine the number of GPU_BUDDY blocks to export */ 693 amdgpu_res_first(res, offset, length, &cursor); 694 while (cursor.remaining) { 695 num_entries++; 696 amdgpu_res_next(&cursor, min(cursor.size, AMDGPU_MAX_SG_SEGMENT_SIZE)); 697 } 698 699 r = sg_alloc_table(*sgt, num_entries, GFP_KERNEL); 700 if (r) 701 goto error_free; 702 703 /* Initialize scatterlist nodes of sg_table */ 704 for_each_sgtable_sg((*sgt), sg, i) 705 sg->length = 0; 706 707 /* 708 * Walk down GPU_BUDDY blocks to populate scatterlist nodes 709 * @note: Use iterator api to get first the GPU_BUDDY block 710 * and the number of bytes from it. Access the following 711 * GPU_BUDDY block(s) if more buffer needs to exported 712 */ 713 amdgpu_res_first(res, offset, length, &cursor); 714 for_each_sgtable_sg((*sgt), sg, i) { 715 phys_addr_t phys = cursor.start + adev->gmc.aper_base; 716 unsigned long size = min(cursor.size, AMDGPU_MAX_SG_SEGMENT_SIZE); 717 dma_addr_t addr; 718 719 addr = dma_map_resource(dev, phys, size, dir, 720 DMA_ATTR_SKIP_CPU_SYNC); 721 r = dma_mapping_error(dev, addr); 722 if (r) 723 goto error_unmap; 724 725 sg_set_page(sg, NULL, size, 0); 726 sg_dma_address(sg) = addr; 727 sg_dma_len(sg) = size; 728 729 amdgpu_res_next(&cursor, size); 730 } 731 732 return 0; 733 734 error_unmap: 735 for_each_sgtable_sg((*sgt), sg, i) { 736 if (!sg->length) 737 continue; 738 739 dma_unmap_resource(dev, sg->dma_address, 740 sg->length, dir, 741 DMA_ATTR_SKIP_CPU_SYNC); 742 } 743 sg_free_table(*sgt); 744 745 error_free: 746 kfree(*sgt); 747 return r; 748 } 749 750 /** 751 * amdgpu_vram_mgr_free_sgt - allocate and fill a sg table 752 * 753 * @dev: device pointer 754 * @dir: data direction of resource to unmap 755 * @sgt: sg table to free 756 * 757 * Free a previously allocate sg table. 758 */ 759 void amdgpu_vram_mgr_free_sgt(struct device *dev, 760 enum dma_data_direction dir, 761 struct sg_table *sgt) 762 { 763 struct scatterlist *sg; 764 int i; 765 766 for_each_sgtable_sg(sgt, sg, i) 767 dma_unmap_resource(dev, sg->dma_address, 768 sg->length, dir, 769 DMA_ATTR_SKIP_CPU_SYNC); 770 sg_free_table(sgt); 771 kfree(sgt); 772 } 773 774 /** 775 * amdgpu_vram_mgr_vis_usage - how many bytes are used in the visible part 776 * 777 * @mgr: amdgpu_vram_mgr pointer 778 * 779 * Returns how many bytes are used in the visible part of VRAM 780 */ 781 uint64_t amdgpu_vram_mgr_vis_usage(struct amdgpu_vram_mgr *mgr) 782 { 783 return atomic64_read(&mgr->vis_usage); 784 } 785 786 /** 787 * amdgpu_vram_mgr_clear_reset_blocks - reset clear blocks 788 * 789 * @adev: amdgpu device pointer 790 * 791 * Reset the cleared drm buddy blocks. 792 */ 793 void amdgpu_vram_mgr_clear_reset_blocks(struct amdgpu_device *adev) 794 { 795 struct amdgpu_vram_mgr *mgr = &adev->mman.vram_mgr; 796 struct gpu_buddy *mm = &mgr->mm; 797 798 mutex_lock(&mgr->lock); 799 gpu_buddy_reset_clear(mm, false); 800 mutex_unlock(&mgr->lock); 801 } 802 803 /** 804 * amdgpu_vram_mgr_intersects - test each drm buddy block for intersection 805 * 806 * @man: TTM memory type manager 807 * @res: The resource to test 808 * @place: The place to test against 809 * @size: Size of the new allocation 810 * 811 * Test each drm buddy block for intersection for eviction decision. 812 */ 813 static bool amdgpu_vram_mgr_intersects(struct ttm_resource_manager *man, 814 struct ttm_resource *res, 815 const struct ttm_place *place, 816 size_t size) 817 { 818 struct amdgpu_vram_mgr_resource *mgr = to_amdgpu_vram_mgr_resource(res); 819 struct gpu_buddy_block *block; 820 821 /* Check each drm buddy block individually */ 822 list_for_each_entry(block, &mgr->blocks, link) { 823 unsigned long fpfn = 824 amdgpu_vram_mgr_block_start(block) >> PAGE_SHIFT; 825 unsigned long lpfn = fpfn + 826 (amdgpu_vram_mgr_block_size(block) >> PAGE_SHIFT); 827 828 if (place->fpfn < lpfn && 829 (!place->lpfn || place->lpfn > fpfn)) 830 return true; 831 } 832 833 return false; 834 } 835 836 /** 837 * amdgpu_vram_mgr_compatible - test each drm buddy block for compatibility 838 * 839 * @man: TTM memory type manager 840 * @res: The resource to test 841 * @place: The place to test against 842 * @size: Size of the new allocation 843 * 844 * Test each drm buddy block for placement compatibility. 845 */ 846 static bool amdgpu_vram_mgr_compatible(struct ttm_resource_manager *man, 847 struct ttm_resource *res, 848 const struct ttm_place *place, 849 size_t size) 850 { 851 struct amdgpu_vram_mgr_resource *mgr = to_amdgpu_vram_mgr_resource(res); 852 struct gpu_buddy_block *block; 853 854 /* Check each drm buddy block individually */ 855 list_for_each_entry(block, &mgr->blocks, link) { 856 unsigned long fpfn = 857 amdgpu_vram_mgr_block_start(block) >> PAGE_SHIFT; 858 unsigned long lpfn = fpfn + 859 (amdgpu_vram_mgr_block_size(block) >> PAGE_SHIFT); 860 861 if (fpfn < place->fpfn || 862 (place->lpfn && lpfn > place->lpfn)) 863 return false; 864 } 865 866 return true; 867 } 868 869 /** 870 * amdgpu_vram_mgr_debug - dump VRAM table 871 * 872 * @man: TTM memory type manager 873 * @printer: DRM printer to use 874 * 875 * Dump the table content using printk. 876 */ 877 static void amdgpu_vram_mgr_debug(struct ttm_resource_manager *man, 878 struct drm_printer *printer) 879 { 880 struct amdgpu_vram_mgr *mgr = to_vram_mgr(man); 881 struct gpu_buddy *mm = &mgr->mm; 882 struct amdgpu_vram_reservation *rsv; 883 884 drm_printf(printer, " vis usage:%llu\n", 885 amdgpu_vram_mgr_vis_usage(mgr)); 886 887 mutex_lock(&mgr->lock); 888 drm_printf(printer, "default_page_size: %lluKiB\n", 889 mgr->default_page_size >> 10); 890 891 drm_buddy_print(mm, printer); 892 893 drm_printf(printer, "reserved:\n"); 894 list_for_each_entry(rsv, &mgr->reserved_pages, blocks) 895 drm_printf(printer, "%#018llx-%#018llx: %llu\n", 896 rsv->start, rsv->start + rsv->size, rsv->size); 897 mutex_unlock(&mgr->lock); 898 } 899 900 static const struct ttm_resource_manager_func amdgpu_vram_mgr_func = { 901 .alloc = amdgpu_vram_mgr_new, 902 .free = amdgpu_vram_mgr_del, 903 .intersects = amdgpu_vram_mgr_intersects, 904 .compatible = amdgpu_vram_mgr_compatible, 905 .debug = amdgpu_vram_mgr_debug 906 }; 907 908 /** 909 * amdgpu_vram_mgr_init - init VRAM manager and DRM MM 910 * 911 * @adev: amdgpu_device pointer 912 * 913 * Allocate and initialize the VRAM manager. 914 */ 915 int amdgpu_vram_mgr_init(struct amdgpu_device *adev) 916 { 917 struct amdgpu_vram_mgr *mgr = &adev->mman.vram_mgr; 918 struct ttm_resource_manager *man = &mgr->manager; 919 int err; 920 921 man->cg = drmm_cgroup_register_region(adev_to_drm(adev), "vram", adev->gmc.real_vram_size); 922 if (IS_ERR(man->cg)) 923 return PTR_ERR(man->cg); 924 ttm_resource_manager_init(man, &adev->mman.bdev, 925 adev->gmc.real_vram_size); 926 927 mutex_init(&mgr->lock); 928 INIT_LIST_HEAD(&mgr->reservations_pending); 929 INIT_LIST_HEAD(&mgr->reserved_pages); 930 INIT_LIST_HEAD(&mgr->allocated_vres_list); 931 mgr->default_page_size = PAGE_SIZE; 932 933 man->func = &amdgpu_vram_mgr_func; 934 err = gpu_buddy_init(&mgr->mm, man->size, PAGE_SIZE); 935 if (err) 936 return err; 937 938 ttm_set_driver_manager(&adev->mman.bdev, TTM_PL_VRAM, &mgr->manager); 939 ttm_resource_manager_set_used(man, true); 940 return 0; 941 } 942 943 /** 944 * amdgpu_vram_mgr_fini - free and destroy VRAM manager 945 * 946 * @adev: amdgpu_device pointer 947 * 948 * Destroy and free the VRAM manager, returns -EBUSY if ranges are still 949 * allocated inside it. 950 */ 951 void amdgpu_vram_mgr_fini(struct amdgpu_device *adev) 952 { 953 struct amdgpu_vram_mgr *mgr = &adev->mman.vram_mgr; 954 struct ttm_resource_manager *man = &mgr->manager; 955 int ret; 956 struct amdgpu_vram_reservation *rsv, *temp; 957 958 ttm_resource_manager_set_used(man, false); 959 960 ret = ttm_resource_manager_evict_all(&adev->mman.bdev, man); 961 if (ret) 962 return; 963 964 mutex_lock(&mgr->lock); 965 list_for_each_entry_safe(rsv, temp, &mgr->reservations_pending, blocks) 966 kfree(rsv); 967 968 list_for_each_entry_safe(rsv, temp, &mgr->reserved_pages, blocks) { 969 gpu_buddy_free_list(&mgr->mm, &rsv->allocated, 0); 970 kfree(rsv); 971 } 972 if (!adev->gmc.is_app_apu) 973 gpu_buddy_fini(&mgr->mm); 974 mutex_unlock(&mgr->lock); 975 976 ttm_resource_manager_cleanup(man); 977 ttm_set_driver_manager(&adev->mman.bdev, TTM_PL_VRAM, NULL); 978 } 979