amdgpu_object.c 24.1 KB
Newer Older
A
Alex Deucher 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35
/*
 * Copyright 2009 Jerome Glisse.
 * All Rights Reserved.
 *
 * Permission is hereby granted, free of charge, to any person obtaining a
 * copy of this software and associated documentation files (the
 * "Software"), to deal in the Software without restriction, including
 * without limitation the rights to use, copy, modify, merge, publish,
 * distribute, sub license, and/or sell copies of the Software, and to
 * permit persons to whom the Software is furnished to do so, subject to
 * the following conditions:
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
 * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
 * USE OR OTHER DEALINGS IN THE SOFTWARE.
 *
 * The above copyright notice and this permission notice (including the
 * next paragraph) shall be included in all copies or substantial portions
 * of the Software.
 *
 */
/*
 * Authors:
 *    Jerome Glisse <glisse@freedesktop.org>
 *    Thomas Hellstrom <thomas-at-tungstengraphics-dot-com>
 *    Dave Airlie
 */
#include <linux/list.h>
#include <linux/slab.h>
#include <drm/drmP.h>
#include <drm/amdgpu_drm.h>
36
#include <drm/drm_cache.h>
A
Alex Deucher 已提交
37 38 39 40 41
#include "amdgpu.h"
#include "amdgpu_trace.h"

static void amdgpu_ttm_bo_destroy(struct ttm_buffer_object *tbo)
{
42
	struct amdgpu_device *adev = amdgpu_ttm_adev(tbo->bdev);
A
Alex Deucher 已提交
43 44 45 46
	struct amdgpu_bo *bo;

	bo = container_of(tbo, struct amdgpu_bo, tbo);

47
	amdgpu_bo_kunmap(bo);
A
Alex Deucher 已提交
48 49

	drm_gem_object_release(&bo->gem_base);
50
	amdgpu_bo_unref(&bo->parent);
51
	if (!list_empty(&bo->shadow_list)) {
52
		mutex_lock(&adev->shadow_list_lock);
53
		list_del_init(&bo->shadow_list);
54
		mutex_unlock(&adev->shadow_list_lock);
55
	}
A
Alex Deucher 已提交
56 57 58 59 60 61 62 63 64 65 66
	kfree(bo->metadata);
	kfree(bo);
}

bool amdgpu_ttm_bo_is_amdgpu_bo(struct ttm_buffer_object *bo)
{
	if (bo->destroy == &amdgpu_ttm_bo_destroy)
		return true;
	return false;
}

67
void amdgpu_ttm_placement_from_domain(struct amdgpu_bo *abo, u32 domain)
A
Alex Deucher 已提交
68
{
69 70 71 72
	struct amdgpu_device *adev = amdgpu_ttm_adev(abo->tbo.bdev);
	struct ttm_placement *placement = &abo->placement;
	struct ttm_place *places = abo->placements;
	u64 flags = abo->flags;
73
	u32 c = 0;
74

A
Alex Deucher 已提交
75
	if (domain & AMDGPU_GEM_DOMAIN_VRAM) {
76 77 78
		unsigned visible_pfn = adev->mc.visible_vram_size >> PAGE_SHIFT;

		places[c].fpfn = 0;
79
		places[c].lpfn = 0;
80
		places[c].flags = TTM_PL_FLAG_WC | TTM_PL_FLAG_UNCACHED |
81
			TTM_PL_FLAG_VRAM;
82

83 84 85 86
		if (flags & AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED)
			places[c].lpfn = visible_pfn;
		else
			places[c].flags |= TTM_PL_FLAG_TOPDOWN;
87 88 89

		if (flags & AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS)
			places[c].flags |= TTM_PL_FLAG_CONTIGUOUS;
90
		c++;
A
Alex Deucher 已提交
91 92 93
	}

	if (domain & AMDGPU_GEM_DOMAIN_GTT) {
94
		places[c].fpfn = 0;
95 96 97 98
		if (flags & AMDGPU_GEM_CREATE_SHADOW)
			places[c].lpfn = adev->mc.gart_size >> PAGE_SHIFT;
		else
			places[c].lpfn = 0;
99 100 101 102 103 104 105
		places[c].flags = TTM_PL_FLAG_TT;
		if (flags & AMDGPU_GEM_CREATE_CPU_GTT_USWC)
			places[c].flags |= TTM_PL_FLAG_WC |
				TTM_PL_FLAG_UNCACHED;
		else
			places[c].flags |= TTM_PL_FLAG_CACHED;
		c++;
A
Alex Deucher 已提交
106 107 108
	}

	if (domain & AMDGPU_GEM_DOMAIN_CPU) {
109 110 111 112 113 114 115 116 117
		places[c].fpfn = 0;
		places[c].lpfn = 0;
		places[c].flags = TTM_PL_FLAG_SYSTEM;
		if (flags & AMDGPU_GEM_CREATE_CPU_GTT_USWC)
			places[c].flags |= TTM_PL_FLAG_WC |
				TTM_PL_FLAG_UNCACHED;
		else
			places[c].flags |= TTM_PL_FLAG_CACHED;
		c++;
A
Alex Deucher 已提交
118 119 120
	}

	if (domain & AMDGPU_GEM_DOMAIN_GDS) {
121 122 123 124
		places[c].fpfn = 0;
		places[c].lpfn = 0;
		places[c].flags = TTM_PL_FLAG_UNCACHED | AMDGPU_PL_FLAG_GDS;
		c++;
A
Alex Deucher 已提交
125
	}
126

A
Alex Deucher 已提交
127
	if (domain & AMDGPU_GEM_DOMAIN_GWS) {
128 129 130 131
		places[c].fpfn = 0;
		places[c].lpfn = 0;
		places[c].flags = TTM_PL_FLAG_UNCACHED | AMDGPU_PL_FLAG_GWS;
		c++;
A
Alex Deucher 已提交
132
	}
133

A
Alex Deucher 已提交
134
	if (domain & AMDGPU_GEM_DOMAIN_OA) {
135 136 137 138
		places[c].fpfn = 0;
		places[c].lpfn = 0;
		places[c].flags = TTM_PL_FLAG_UNCACHED | AMDGPU_PL_FLAG_OA;
		c++;
A
Alex Deucher 已提交
139 140 141
	}

	if (!c) {
142 143 144 145
		places[c].fpfn = 0;
		places[c].lpfn = 0;
		places[c].flags = TTM_PL_MASK_CACHING | TTM_PL_FLAG_SYSTEM;
		c++;
A
Alex Deucher 已提交
146
	}
147

148
	placement->num_placement = c;
149
	placement->placement = places;
A
Alex Deucher 已提交
150

151 152
	placement->num_busy_placement = c;
	placement->busy_placement = places;
A
Alex Deucher 已提交
153 154
}

155
/**
156
 * amdgpu_bo_create_reserved - create reserved BO for kernel use
157 158 159 160 161 162 163 164 165
 *
 * @adev: amdgpu device object
 * @size: size for the new BO
 * @align: alignment for the new BO
 * @domain: where to place it
 * @bo_ptr: resulting BO
 * @gpu_addr: GPU addr of the pinned BO
 * @cpu_addr: optional CPU address mapping
 *
166 167
 * Allocates and pins a BO for kernel internal use, and returns it still
 * reserved.
168 169 170
 *
 * Returns 0 on success, negative error code otherwise.
 */
171 172 173 174
int amdgpu_bo_create_reserved(struct amdgpu_device *adev,
			      unsigned long size, int align,
			      u32 domain, struct amdgpu_bo **bo_ptr,
			      u64 *gpu_addr, void **cpu_addr)
175
{
176
	bool free = false;
177 178
	int r;

179 180 181 182
	if (!*bo_ptr) {
		r = amdgpu_bo_create(adev, size, align, true, domain,
				     AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED |
				     AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS,
183
				     NULL, NULL, 0, bo_ptr);
184 185 186 187 188 189
		if (r) {
			dev_err(adev->dev, "(%d) failed to allocate kernel bo\n",
				r);
			return r;
		}
		free = true;
190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217
	}

	r = amdgpu_bo_reserve(*bo_ptr, false);
	if (r) {
		dev_err(adev->dev, "(%d) failed to reserve kernel bo\n", r);
		goto error_free;
	}

	r = amdgpu_bo_pin(*bo_ptr, domain, gpu_addr);
	if (r) {
		dev_err(adev->dev, "(%d) kernel bo pin failed\n", r);
		goto error_unreserve;
	}

	if (cpu_addr) {
		r = amdgpu_bo_kmap(*bo_ptr, cpu_addr);
		if (r) {
			dev_err(adev->dev, "(%d) kernel bo map failed\n", r);
			goto error_unreserve;
		}
	}

	return 0;

error_unreserve:
	amdgpu_bo_unreserve(*bo_ptr);

error_free:
218 219
	if (free)
		amdgpu_bo_unref(bo_ptr);
220 221 222 223

	return r;
}

224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256
/**
 * amdgpu_bo_create_kernel - create BO for kernel use
 *
 * @adev: amdgpu device object
 * @size: size for the new BO
 * @align: alignment for the new BO
 * @domain: where to place it
 * @bo_ptr: resulting BO
 * @gpu_addr: GPU addr of the pinned BO
 * @cpu_addr: optional CPU address mapping
 *
 * Allocates and pins a BO for kernel internal use.
 *
 * Returns 0 on success, negative error code otherwise.
 */
int amdgpu_bo_create_kernel(struct amdgpu_device *adev,
			    unsigned long size, int align,
			    u32 domain, struct amdgpu_bo **bo_ptr,
			    u64 *gpu_addr, void **cpu_addr)
{
	int r;

	r = amdgpu_bo_create_reserved(adev, size, align, domain, bo_ptr,
				      gpu_addr, cpu_addr);

	if (r)
		return r;

	amdgpu_bo_unreserve(*bo_ptr);

	return 0;
}

257 258 259 260 261 262 263 264 265 266 267 268 269
/**
 * amdgpu_bo_free_kernel - free BO for kernel use
 *
 * @bo: amdgpu BO to free
 *
 * unmaps and unpin a BO for kernel internal use.
 */
void amdgpu_bo_free_kernel(struct amdgpu_bo **bo, u64 *gpu_addr,
			   void **cpu_addr)
{
	if (*bo == NULL)
		return;

270
	if (likely(amdgpu_bo_reserve(*bo, true) == 0)) {
271 272 273 274 275 276 277 278 279 280 281 282 283 284 285
		if (cpu_addr)
			amdgpu_bo_kunmap(*bo);

		amdgpu_bo_unpin(*bo);
		amdgpu_bo_unreserve(*bo);
	}
	amdgpu_bo_unref(bo);

	if (gpu_addr)
		*gpu_addr = 0;

	if (cpu_addr)
		*cpu_addr = NULL;
}

286 287 288 289 290 291 292
static int amdgpu_bo_do_create(struct amdgpu_device *adev,
			       unsigned long size, int byte_align,
			       bool kernel, u32 domain, u64 flags,
			       struct sg_table *sg,
			       struct reservation_object *resv,
			       uint64_t init_value,
			       struct amdgpu_bo **bo_ptr)
A
Alex Deucher 已提交
293 294 295 296
{
	struct amdgpu_bo *bo;
	enum ttm_bo_type type;
	unsigned long page_align;
297
	u64 initial_bytes_moved, bytes_moved;
A
Alex Deucher 已提交
298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323
	size_t acc_size;
	int r;

	page_align = roundup(byte_align, PAGE_SIZE) >> PAGE_SHIFT;
	size = ALIGN(size, PAGE_SIZE);

	if (kernel) {
		type = ttm_bo_type_kernel;
	} else if (sg) {
		type = ttm_bo_type_sg;
	} else {
		type = ttm_bo_type_device;
	}
	*bo_ptr = NULL;

	acc_size = ttm_bo_dma_acc_size(&adev->mman.bdev, size,
				       sizeof(struct amdgpu_bo));

	bo = kzalloc(sizeof(struct amdgpu_bo), GFP_KERNEL);
	if (bo == NULL)
		return -ENOMEM;
	r = drm_gem_object_init(adev->ddev, &bo->gem_base, size);
	if (unlikely(r)) {
		kfree(bo);
		return r;
	}
324
	INIT_LIST_HEAD(&bo->shadow_list);
A
Alex Deucher 已提交
325
	INIT_LIST_HEAD(&bo->va);
K
Kent Russell 已提交
326
	bo->preferred_domains = domain & (AMDGPU_GEM_DOMAIN_VRAM |
327 328 329 330 331
					 AMDGPU_GEM_DOMAIN_GTT |
					 AMDGPU_GEM_DOMAIN_CPU |
					 AMDGPU_GEM_DOMAIN_GDS |
					 AMDGPU_GEM_DOMAIN_GWS |
					 AMDGPU_GEM_DOMAIN_OA);
K
Kent Russell 已提交
332
	bo->allowed_domains = bo->preferred_domains;
333 334
	if (!kernel && bo->allowed_domains == AMDGPU_GEM_DOMAIN_VRAM)
		bo->allowed_domains |= AMDGPU_GEM_DOMAIN_GTT;
A
Alex Deucher 已提交
335 336

	bo->flags = flags;
337

338 339 340 341 342 343 344 345 346 347 348
#ifdef CONFIG_X86_32
	/* XXX: Write-combined CPU mappings of GTT seem broken on 32-bit
	 * See https://bugs.freedesktop.org/show_bug.cgi?id=84627
	 */
	bo->flags &= ~AMDGPU_GEM_CREATE_CPU_GTT_USWC;
#elif defined(CONFIG_X86) && !defined(CONFIG_X86_PAT)
	/* Don't try to enable write-combining when it can't work, or things
	 * may be slow
	 * See https://bugs.freedesktop.org/show_bug.cgi?id=88758
	 */

349
#ifndef CONFIG_COMPILE_TEST
350 351
#warning Please enable CONFIG_MTRR and CONFIG_X86_PAT for better performance \
	 thanks to write-combining
352
#endif
353 354 355 356 357 358

	if (bo->flags & AMDGPU_GEM_CREATE_CPU_GTT_USWC)
		DRM_INFO_ONCE("Please enable CONFIG_MTRR and CONFIG_X86_PAT for "
			      "better performance thanks to write-combining\n");
	bo->flags &= ~AMDGPU_GEM_CREATE_CPU_GTT_USWC;
#else
359 360 361 362 363
	/* For architectures that don't support WC memory,
	 * mask out the WC flag from the BO
	 */
	if (!drm_arch_can_wc_memory())
		bo->flags &= ~AMDGPU_GEM_CREATE_CPU_GTT_USWC;
364
#endif
365

366 367
	bo->tbo.bdev = &adev->mman.bdev;
	amdgpu_ttm_placement_from_domain(bo, domain);
368

369
	initial_bytes_moved = atomic64_read(&adev->num_bytes_moved);
370
	/* Kernel allocation are uninterruptible */
371 372 373
	r = ttm_bo_init_reserved(&adev->mman.bdev, &bo->tbo, size, type,
				 &bo->placement, page_align, !kernel, NULL,
				 acc_size, sg, resv, &amdgpu_ttm_bo_destroy);
374 375 376 377 378 379 380 381
	bytes_moved = atomic64_read(&adev->num_bytes_moved) -
		      initial_bytes_moved;
	if (adev->mc.visible_vram_size < adev->mc.real_vram_size &&
	    bo->tbo.mem.mem_type == TTM_PL_VRAM &&
	    bo->tbo.mem.start < adev->mc.visible_vram_size >> PAGE_SHIFT)
		amdgpu_cs_report_moved_bytes(adev, bytes_moved, bytes_moved);
	else
		amdgpu_cs_report_moved_bytes(adev, bytes_moved, 0);
382

383
	if (unlikely(r != 0))
A
Alex Deucher 已提交
384
		return r;
385

386
	if (kernel)
387
		bo->tbo.priority = 1;
388

389 390
	if (flags & AMDGPU_GEM_CREATE_VRAM_CLEARED &&
	    bo->tbo.mem.placement & TTM_PL_FLAG_VRAM) {
391
		struct dma_fence *fence;
392

393
		r = amdgpu_fill_buffer(bo, init_value, bo->tbo.resv, &fence);
394 395 396
		if (unlikely(r))
			goto fail_unreserve;

397
		amdgpu_bo_fence(bo, fence, false);
398 399 400
		dma_fence_put(bo->tbo.moving);
		bo->tbo.moving = dma_fence_get(fence);
		dma_fence_put(fence);
401
	}
402
	if (!resv)
403
		amdgpu_bo_unreserve(bo);
A
Alex Deucher 已提交
404 405 406 407
	*bo_ptr = bo;

	trace_amdgpu_bo_create(bo);

408 409 410 411
	/* Treat CPU_ACCESS_REQUIRED only as a hint if given by UMD */
	if (type == ttm_bo_type_device)
		bo->flags &= ~AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED;

A
Alex Deucher 已提交
412
	return 0;
413 414

fail_unreserve:
415 416
	if (!resv)
		ww_mutex_unlock(&bo->tbo.resv->lock);
417 418
	amdgpu_bo_unref(&bo);
	return r;
A
Alex Deucher 已提交
419 420
}

421 422 423 424 425 426 427 428 429
static int amdgpu_bo_create_shadow(struct amdgpu_device *adev,
				   unsigned long size, int byte_align,
				   struct amdgpu_bo *bo)
{
	int r;

	if (bo->shadow)
		return 0;

430 431 432 433 434 435
	r = amdgpu_bo_do_create(adev, size, byte_align, true,
				AMDGPU_GEM_DOMAIN_GTT,
				AMDGPU_GEM_CREATE_CPU_GTT_USWC |
				AMDGPU_GEM_CREATE_SHADOW,
				NULL, bo->tbo.resv, 0,
				&bo->shadow);
436
	if (!r) {
437
		bo->shadow->parent = amdgpu_bo_ref(bo);
438 439 440 441
		mutex_lock(&adev->shadow_list_lock);
		list_add_tail(&bo->shadow_list, &adev->shadow_list);
		mutex_unlock(&adev->shadow_list_lock);
	}
442 443 444 445

	return r;
}

446 447 448
/* init_value will only take effect when flags contains
 * AMDGPU_GEM_CREATE_VRAM_CLEARED.
 */
449 450 451
int amdgpu_bo_create(struct amdgpu_device *adev,
		     unsigned long size, int byte_align,
		     bool kernel, u32 domain, u64 flags,
452 453
		     struct sg_table *sg,
		     struct reservation_object *resv,
454
		     uint64_t init_value,
455
		     struct amdgpu_bo **bo_ptr)
456
{
457
	uint64_t parent_flags = flags & ~AMDGPU_GEM_CREATE_SHADOW;
458
	int r;
459

460 461
	r = amdgpu_bo_do_create(adev, size, byte_align, kernel, domain,
				parent_flags, sg, resv, init_value, bo_ptr);
462 463 464
	if (r)
		return r;

465 466 467 468
	if ((flags & AMDGPU_GEM_CREATE_SHADOW) && amdgpu_need_backup(adev)) {
		if (!resv)
			WARN_ON(reservation_object_lock((*bo_ptr)->tbo.resv,
							NULL));
469

470
		r = amdgpu_bo_create_shadow(adev, size, byte_align, (*bo_ptr));
471 472

		if (!resv)
473
			reservation_object_unlock((*bo_ptr)->tbo.resv);
474

475 476 477 478 479
		if (r)
			amdgpu_bo_unref(bo_ptr);
	}

	return r;
480 481
}

482 483 484 485
int amdgpu_bo_backup_to_shadow(struct amdgpu_device *adev,
			       struct amdgpu_ring *ring,
			       struct amdgpu_bo *bo,
			       struct reservation_object *resv,
486
			       struct dma_fence **fence,
487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505
			       bool direct)

{
	struct amdgpu_bo *shadow = bo->shadow;
	uint64_t bo_addr, shadow_addr;
	int r;

	if (!shadow)
		return -EINVAL;

	bo_addr = amdgpu_bo_gpu_offset(bo);
	shadow_addr = amdgpu_bo_gpu_offset(bo->shadow);

	r = reservation_object_reserve_shared(bo->tbo.resv);
	if (r)
		goto err;

	r = amdgpu_copy_buffer(ring, bo_addr, shadow_addr,
			       amdgpu_bo_size(bo), resv, fence,
506
			       direct, false);
507 508 509 510 511 512 513
	if (!r)
		amdgpu_bo_fence(bo, *fence, true);

err:
	return r;
}

514 515 516 517 518 519 520 521
int amdgpu_bo_validate(struct amdgpu_bo *bo)
{
	uint32_t domain;
	int r;

	if (bo->pin_count)
		return 0;

K
Kent Russell 已提交
522
	domain = bo->preferred_domains;
523 524 525 526 527 528 529 530 531 532 533 534

retry:
	amdgpu_ttm_placement_from_domain(bo, domain);
	r = ttm_bo_validate(&bo->tbo, &bo->placement, false, false);
	if (unlikely(r == -ENOMEM) && domain != bo->allowed_domains) {
		domain = bo->allowed_domains;
		goto retry;
	}

	return r;
}

535 536 537 538
int amdgpu_bo_restore_from_shadow(struct amdgpu_device *adev,
				  struct amdgpu_ring *ring,
				  struct amdgpu_bo *bo,
				  struct reservation_object *resv,
539
				  struct dma_fence **fence,
540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558
				  bool direct)

{
	struct amdgpu_bo *shadow = bo->shadow;
	uint64_t bo_addr, shadow_addr;
	int r;

	if (!shadow)
		return -EINVAL;

	bo_addr = amdgpu_bo_gpu_offset(bo);
	shadow_addr = amdgpu_bo_gpu_offset(bo->shadow);

	r = reservation_object_reserve_shared(bo->tbo.resv);
	if (r)
		goto err;

	r = amdgpu_copy_buffer(ring, shadow_addr, bo_addr,
			       amdgpu_bo_size(bo), resv, fence,
559
			       direct, false);
560 561 562 563 564 565 566
	if (!r)
		amdgpu_bo_fence(bo, *fence, true);

err:
	return r;
}

A
Alex Deucher 已提交
567 568
int amdgpu_bo_kmap(struct amdgpu_bo *bo, void **ptr)
{
569
	void *kptr;
570
	long r;
A
Alex Deucher 已提交
571

572 573 574
	if (bo->flags & AMDGPU_GEM_CREATE_NO_CPU_ACCESS)
		return -EPERM;

575 576 577 578
	kptr = amdgpu_bo_kptr(bo);
	if (kptr) {
		if (ptr)
			*ptr = kptr;
A
Alex Deucher 已提交
579 580
		return 0;
	}
581 582 583 584 585 586

	r = reservation_object_wait_timeout_rcu(bo->tbo.resv, false, false,
						MAX_SCHEDULE_TIMEOUT);
	if (r < 0)
		return r;

A
Alex Deucher 已提交
587
	r = ttm_bo_kmap(&bo->tbo, 0, bo->tbo.num_pages, &bo->kmap);
588
	if (r)
A
Alex Deucher 已提交
589
		return r;
590 591

	if (ptr)
592
		*ptr = amdgpu_bo_kptr(bo);
593

A
Alex Deucher 已提交
594 595 596
	return 0;
}

597 598 599 600 601 602 603
void *amdgpu_bo_kptr(struct amdgpu_bo *bo)
{
	bool is_iomem;

	return ttm_kmap_obj_virtual(&bo->kmap, &is_iomem);
}

A
Alex Deucher 已提交
604 605
void amdgpu_bo_kunmap(struct amdgpu_bo *bo)
{
606 607
	if (bo->kmap.bo)
		ttm_bo_kunmap(&bo->kmap);
A
Alex Deucher 已提交
608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631
}

struct amdgpu_bo *amdgpu_bo_ref(struct amdgpu_bo *bo)
{
	if (bo == NULL)
		return NULL;

	ttm_bo_reference(&bo->tbo);
	return bo;
}

void amdgpu_bo_unref(struct amdgpu_bo **bo)
{
	struct ttm_buffer_object *tbo;

	if ((*bo) == NULL)
		return;

	tbo = &((*bo)->tbo);
	ttm_bo_unref(&tbo);
	if (tbo == NULL)
		*bo = NULL;
}

632 633
int amdgpu_bo_pin_restricted(struct amdgpu_bo *bo, u32 domain,
			     u64 min_offset, u64 max_offset,
A
Alex Deucher 已提交
634 635
			     u64 *gpu_addr)
{
636
	struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev);
A
Alex Deucher 已提交
637
	int r, i;
638
	unsigned fpfn, lpfn;
A
Alex Deucher 已提交
639

640
	if (amdgpu_ttm_tt_get_usermm(bo->tbo.ttm))
A
Alex Deucher 已提交
641 642
		return -EPERM;

643 644 645
	if (WARN_ON_ONCE(min_offset > max_offset))
		return -EINVAL;

646 647 648 649
	/* A shared bo cannot be migrated to VRAM */
	if (bo->prime_shared_count && (domain == AMDGPU_GEM_DOMAIN_VRAM))
		return -EINVAL;

A
Alex Deucher 已提交
650
	if (bo->pin_count) {
651 652 653 654 655
		uint32_t mem_type = bo->tbo.mem.mem_type;

		if (domain != amdgpu_mem_type_to_domain(mem_type))
			return -EINVAL;

A
Alex Deucher 已提交
656 657 658 659 660
		bo->pin_count++;
		if (gpu_addr)
			*gpu_addr = amdgpu_bo_gpu_offset(bo);

		if (max_offset != 0) {
661
			u64 domain_start = bo->tbo.bdev->man[mem_type].gpu_offset;
A
Alex Deucher 已提交
662 663 664 665 666 667
			WARN_ON_ONCE(max_offset <
				     (amdgpu_bo_gpu_offset(bo) - domain_start));
		}

		return 0;
	}
668 669

	bo->flags |= AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS;
A
Alex Deucher 已提交
670 671 672 673
	amdgpu_ttm_placement_from_domain(bo, domain);
	for (i = 0; i < bo->placement.num_placement; i++) {
		/* force to pin into visible video ram */
		if ((bo->placements[i].flags & TTM_PL_FLAG_VRAM) &&
674
		    !(bo->flags & AMDGPU_GEM_CREATE_NO_CPU_ACCESS) &&
675
		    (!max_offset || max_offset >
676
		     adev->mc.visible_vram_size)) {
677
			if (WARN_ON_ONCE(min_offset >
678
					 adev->mc.visible_vram_size))
679 680
				return -EINVAL;
			fpfn = min_offset >> PAGE_SHIFT;
681
			lpfn = adev->mc.visible_vram_size >> PAGE_SHIFT;
682 683 684 685 686 687
		} else {
			fpfn = min_offset >> PAGE_SHIFT;
			lpfn = max_offset >> PAGE_SHIFT;
		}
		if (fpfn > bo->placements[i].fpfn)
			bo->placements[i].fpfn = fpfn;
688 689
		if (!bo->placements[i].lpfn ||
		    (lpfn && lpfn < bo->placements[i].lpfn))
690
			bo->placements[i].lpfn = lpfn;
A
Alex Deucher 已提交
691 692 693 694
		bo->placements[i].flags |= TTM_PL_FLAG_NO_EVICT;
	}

	r = ttm_bo_validate(&bo->tbo, &bo->placement, false, false);
695
	if (unlikely(r)) {
696
		dev_err(adev->dev, "%p pin failed\n", bo);
697 698 699 700
		goto error;
	}

	bo->pin_count = 1;
701 702 703 704 705 706
	if (gpu_addr != NULL) {
		r = amdgpu_ttm_bind(&bo->tbo, &bo->tbo.mem);
		if (unlikely(r)) {
			dev_err(adev->dev, "%p bind failed\n", bo);
			goto error;
		}
707
		*gpu_addr = amdgpu_bo_gpu_offset(bo);
708
	}
709
	if (domain == AMDGPU_GEM_DOMAIN_VRAM) {
710
		adev->vram_pin_size += amdgpu_bo_size(bo);
711
		if (bo->flags & AMDGPU_GEM_CREATE_NO_CPU_ACCESS)
712
			adev->invisible_pin_size += amdgpu_bo_size(bo);
713
	} else if (domain == AMDGPU_GEM_DOMAIN_GTT) {
714
		adev->gart_pin_size += amdgpu_bo_size(bo);
A
Alex Deucher 已提交
715
	}
716 717

error:
A
Alex Deucher 已提交
718 719 720 721 722
	return r;
}

int amdgpu_bo_pin(struct amdgpu_bo *bo, u32 domain, u64 *gpu_addr)
{
723
	return amdgpu_bo_pin_restricted(bo, domain, 0, 0, gpu_addr);
A
Alex Deucher 已提交
724 725 726 727
}

int amdgpu_bo_unpin(struct amdgpu_bo *bo)
{
728
	struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev);
A
Alex Deucher 已提交
729 730 731
	int r, i;

	if (!bo->pin_count) {
732
		dev_warn(adev->dev, "%p unpin not necessary\n", bo);
A
Alex Deucher 已提交
733 734 735 736 737 738 739 740 741 742
		return 0;
	}
	bo->pin_count--;
	if (bo->pin_count)
		return 0;
	for (i = 0; i < bo->placement.num_placement; i++) {
		bo->placements[i].lpfn = 0;
		bo->placements[i].flags &= ~TTM_PL_FLAG_NO_EVICT;
	}
	r = ttm_bo_validate(&bo->tbo, &bo->placement, false, false);
743
	if (unlikely(r)) {
744
		dev_err(adev->dev, "%p validate failed for unpin\n", bo);
745
		goto error;
A
Alex Deucher 已提交
746
	}
747 748

	if (bo->tbo.mem.mem_type == TTM_PL_VRAM) {
749
		adev->vram_pin_size -= amdgpu_bo_size(bo);
750
		if (bo->flags & AMDGPU_GEM_CREATE_NO_CPU_ACCESS)
751
			adev->invisible_pin_size -= amdgpu_bo_size(bo);
752
	} else if (bo->tbo.mem.mem_type == TTM_PL_TT) {
753
		adev->gart_pin_size -= amdgpu_bo_size(bo);
754 755 756
	}

error:
A
Alex Deucher 已提交
757 758 759 760 761 762
	return r;
}

int amdgpu_bo_evict_vram(struct amdgpu_device *adev)
{
	/* late 2.6.33 fix IGP hibernate - we need pm ops to do this correct */
763
	if (0 && (adev->flags & AMD_IS_APU)) {
A
Alex Deucher 已提交
764 765 766 767 768 769
		/* Useless to evict on IGP chips */
		return 0;
	}
	return ttm_bo_evict_mm(&adev->mman.bdev, TTM_PL_VRAM);
}

770 771 772 773 774 775 776 777 778 779 780
static const char *amdgpu_vram_names[] = {
	"UNKNOWN",
	"GDDR1",
	"DDR2",
	"GDDR3",
	"GDDR4",
	"GDDR5",
	"HBM",
	"DDR3"
};

A
Alex Deucher 已提交
781 782
int amdgpu_bo_init(struct amdgpu_device *adev)
{
783 784 785 786
	/* reserve PAT memory space to WC for VRAM */
	arch_io_reserve_memtype_wc(adev->mc.aper_base,
				   adev->mc.aper_size);

A
Alex Deucher 已提交
787 788 789 790 791 792
	/* Add an MTRR for the VRAM */
	adev->mc.vram_mtrr = arch_phys_wc_add(adev->mc.aper_base,
					      adev->mc.aper_size);
	DRM_INFO("Detected VRAM RAM=%lluM, BAR=%lluM\n",
		adev->mc.mc_vram_size >> 20,
		(unsigned long long)adev->mc.aper_size >> 20);
793 794
	DRM_INFO("RAM width %dbits %s\n",
		 adev->mc.vram_width, amdgpu_vram_names[adev->mc.vram_type]);
A
Alex Deucher 已提交
795 796 797 798 799 800 801
	return amdgpu_ttm_init(adev);
}

void amdgpu_bo_fini(struct amdgpu_device *adev)
{
	amdgpu_ttm_fini(adev);
	arch_phys_wc_del(adev->mc.vram_mtrr);
802
	arch_io_free_memtype_wc(adev->mc.aper_base, adev->mc.aper_size);
A
Alex Deucher 已提交
803 804 805 806 807 808 809 810 811 812
}

int amdgpu_bo_fbdev_mmap(struct amdgpu_bo *bo,
			     struct vm_area_struct *vma)
{
	return ttm_fbdev_mmap(vma, &bo->tbo);
}

int amdgpu_bo_set_tiling_flags(struct amdgpu_bo *bo, u64 tiling_flags)
{
813 814 815 816
	struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev);

	if (adev->family <= AMDGPU_FAMILY_CZ &&
	    AMDGPU_TILING_GET(tiling_flags, TILE_SPLIT) > 6)
A
Alex Deucher 已提交
817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838
		return -EINVAL;

	bo->tiling_flags = tiling_flags;
	return 0;
}

void amdgpu_bo_get_tiling_flags(struct amdgpu_bo *bo, u64 *tiling_flags)
{
	lockdep_assert_held(&bo->tbo.resv->lock.base);

	if (tiling_flags)
		*tiling_flags = bo->tiling_flags;
}

int amdgpu_bo_set_metadata (struct amdgpu_bo *bo, void *metadata,
			    uint32_t metadata_size, uint64_t flags)
{
	void *buffer;

	if (!metadata_size) {
		if (bo->metadata_size) {
			kfree(bo->metadata);
839
			bo->metadata = NULL;
A
Alex Deucher 已提交
840 841 842 843 844 845 846 847
			bo->metadata_size = 0;
		}
		return 0;
	}

	if (metadata == NULL)
		return -EINVAL;

848
	buffer = kmemdup(metadata, metadata_size, GFP_KERNEL);
A
Alex Deucher 已提交
849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883
	if (buffer == NULL)
		return -ENOMEM;

	kfree(bo->metadata);
	bo->metadata_flags = flags;
	bo->metadata = buffer;
	bo->metadata_size = metadata_size;

	return 0;
}

int amdgpu_bo_get_metadata(struct amdgpu_bo *bo, void *buffer,
			   size_t buffer_size, uint32_t *metadata_size,
			   uint64_t *flags)
{
	if (!buffer && !metadata_size)
		return -EINVAL;

	if (buffer) {
		if (buffer_size < bo->metadata_size)
			return -EINVAL;

		if (bo->metadata_size)
			memcpy(buffer, bo->metadata, bo->metadata_size);
	}

	if (metadata_size)
		*metadata_size = bo->metadata_size;
	if (flags)
		*flags = bo->metadata_flags;

	return 0;
}

void amdgpu_bo_move_notify(struct ttm_buffer_object *bo,
884
			   bool evict,
A
Alex Deucher 已提交
885 886
			   struct ttm_mem_reg *new_mem)
{
887
	struct amdgpu_device *adev = amdgpu_ttm_adev(bo->bdev);
888
	struct amdgpu_bo *abo;
889
	struct ttm_mem_reg *old_mem = &bo->mem;
A
Alex Deucher 已提交
890 891 892 893

	if (!amdgpu_ttm_bo_is_amdgpu_bo(bo))
		return;

894
	abo = container_of(bo, struct amdgpu_bo, tbo);
895
	amdgpu_vm_bo_invalidate(adev, abo, evict);
A
Alex Deucher 已提交
896

897 898
	amdgpu_bo_kunmap(abo);

899 900 901 902
	/* remember the eviction */
	if (evict)
		atomic64_inc(&adev->num_evictions);

A
Alex Deucher 已提交
903 904 905 906 907
	/* update statistics */
	if (!new_mem)
		return;

	/* move_notify is called before move happens */
908
	trace_amdgpu_ttm_bo_move(abo, new_mem->mem_type, old_mem->mem_type);
A
Alex Deucher 已提交
909 910 911 912
}

int amdgpu_bo_fault_reserve_notify(struct ttm_buffer_object *bo)
{
913
	struct amdgpu_device *adev = amdgpu_ttm_adev(bo->bdev);
914
	struct amdgpu_bo *abo;
915 916
	unsigned long offset, size;
	int r;
A
Alex Deucher 已提交
917 918 919

	if (!amdgpu_ttm_bo_is_amdgpu_bo(bo))
		return 0;
920 921

	abo = container_of(bo, struct amdgpu_bo, tbo);
922 923 924 925

	/* Remember that this BO was accessed by the CPU */
	abo->flags |= AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED;

926 927 928 929 930
	if (bo->mem.mem_type != TTM_PL_VRAM)
		return 0;

	size = bo->mem.num_pages << PAGE_SHIFT;
	offset = bo->mem.start << PAGE_SHIFT;
931
	if ((offset + size) <= adev->mc.visible_vram_size)
932 933
		return 0;

934 935 936 937
	/* Can't move a pinned BO to visible VRAM */
	if (abo->pin_count > 0)
		return -EINVAL;

938
	/* hurrah the memory is not visible ! */
939
	atomic64_inc(&adev->num_vram_cpu_page_faults);
940 941 942 943 944 945 946
	amdgpu_ttm_placement_from_domain(abo, AMDGPU_GEM_DOMAIN_VRAM |
					 AMDGPU_GEM_DOMAIN_GTT);

	/* Avoid costly evictions; only set GTT as a busy placement */
	abo->placement.num_busy_placement = 1;
	abo->placement.busy_placement = &abo->placements[1];

947
	r = ttm_bo_validate(bo, &abo->placement, false, false);
948
	if (unlikely(r != 0))
949 950 951 952
		return r;

	offset = bo->mem.start << PAGE_SHIFT;
	/* this should never happen */
953 954
	if (bo->mem.mem_type == TTM_PL_VRAM &&
	    (offset + size) > adev->mc.visible_vram_size)
955 956
		return -EINVAL;

A
Alex Deucher 已提交
957 958 959 960 961 962 963 964 965 966 967
	return 0;
}

/**
 * amdgpu_bo_fence - add fence to buffer object
 *
 * @bo: buffer object in question
 * @fence: fence to add
 * @shared: true if fence should be added shared
 *
 */
968
void amdgpu_bo_fence(struct amdgpu_bo *bo, struct dma_fence *fence,
A
Alex Deucher 已提交
969 970 971 972 973
		     bool shared)
{
	struct reservation_object *resv = bo->tbo.resv;

	if (shared)
974
		reservation_object_add_shared_fence(resv, fence);
A
Alex Deucher 已提交
975
	else
976
		reservation_object_add_excl_fence(resv, fence);
A
Alex Deucher 已提交
977
}
978 979 980 981 982 983 984 985 986 987 988 989 990

/**
 * amdgpu_bo_gpu_offset - return GPU offset of bo
 * @bo:	amdgpu object for which we query the offset
 *
 * Returns current GPU offset of the object.
 *
 * Note: object should either be pinned or reserved when calling this
 * function, it might be useful to add check for this for debugging.
 */
u64 amdgpu_bo_gpu_offset(struct amdgpu_bo *bo)
{
	WARN_ON_ONCE(bo->tbo.mem.mem_type == TTM_PL_SYSTEM);
991 992
	WARN_ON_ONCE(bo->tbo.mem.mem_type == TTM_PL_TT &&
		     !amdgpu_ttm_is_bound(bo->tbo.ttm));
993 994
	WARN_ON_ONCE(!ww_mutex_is_locked(&bo->tbo.resv->lock) &&
		     !bo->pin_count);
995
	WARN_ON_ONCE(bo->tbo.mem.start == AMDGPU_BO_INVALID_OFFSET);
996 997
	WARN_ON_ONCE(bo->tbo.mem.mem_type == TTM_PL_VRAM &&
		     !(bo->flags & AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS));
998 999 1000

	return bo->tbo.offset;
}