ttm_bo.c 45.4 KB
Newer Older
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
/**************************************************************************
 *
 * Copyright (c) 2006-2009 VMware, Inc., Palo Alto, CA., USA
 * 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 above copyright notice and this permission notice (including the
 * next paragraph) shall be included in all copies or substantial portions
 * of the Software.
 *
 * 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.
 *
 **************************************************************************/
/*
 * Authors: Thomas Hellstrom <thellstrom-at-vmware-dot-com>
 */

J
Joe Perches 已提交
31 32
#define pr_fmt(fmt) "[TTM] " fmt

33 34 35
#include <drm/ttm/ttm_module.h>
#include <drm/ttm/ttm_bo_driver.h>
#include <drm/ttm/ttm_placement.h>
36 37 38 39 40 41
#include <linux/jiffies.h>
#include <linux/slab.h>
#include <linux/sched.h>
#include <linux/mm.h>
#include <linux/file.h>
#include <linux/module.h>
A
Arun Sharma 已提交
42
#include <linux/atomic.h>
43 44 45 46 47 48 49

#define TTM_ASSERT_LOCKED(param)
#define TTM_DEBUG(fmt, arg...)
#define TTM_BO_HASH_ORDER 13

static int ttm_bo_setup_vm(struct ttm_buffer_object *bo);
static int ttm_bo_swapout(struct ttm_mem_shrink *shrink);
50 51 52 53 54 55 56
static void ttm_bo_global_kobj_release(struct kobject *kobj);

static struct attribute ttm_bo_count = {
	.name = "bo_count",
	.mode = S_IRUGO
};

57 58 59 60 61 62 63 64 65 66 67 68
static inline int ttm_mem_type_from_flags(uint32_t flags, uint32_t *mem_type)
{
	int i;

	for (i = 0; i <= TTM_PL_PRIV5; i++)
		if (flags & (1 << i)) {
			*mem_type = i;
			return 0;
		}
	return -EINVAL;
}

69
static void ttm_mem_type_debug(struct ttm_bo_device *bdev, int mem_type)
70
{
71 72
	struct ttm_mem_type_manager *man = &bdev->man[mem_type];

J
Joe Perches 已提交
73 74 75 76 77 78 79
	pr_err("    has_type: %d\n", man->has_type);
	pr_err("    use_type: %d\n", man->use_type);
	pr_err("    flags: 0x%08X\n", man->flags);
	pr_err("    gpu_offset: 0x%08lX\n", man->gpu_offset);
	pr_err("    size: %llu\n", man->size);
	pr_err("    available_caching: 0x%08X\n", man->available_caching);
	pr_err("    default_caching: 0x%08X\n", man->default_caching);
80 81
	if (mem_type != TTM_PL_SYSTEM)
		(*man->func->debug)(man, TTM_PFX);
82 83 84 85 86 87 88
}

static void ttm_bo_mem_space_debug(struct ttm_buffer_object *bo,
					struct ttm_placement *placement)
{
	int i, ret, mem_type;

J
Joe Perches 已提交
89 90 91
	pr_err("No space for %p (%lu pages, %luK, %luM)\n",
	       bo, bo->mem.num_pages, bo->mem.size >> 10,
	       bo->mem.size >> 20);
92 93 94 95 96
	for (i = 0; i < placement->num_placement; i++) {
		ret = ttm_mem_type_from_flags(placement->placement[i],
						&mem_type);
		if (ret)
			return;
J
Joe Perches 已提交
97 98
		pr_err("  placement[%d]=0x%08X (%d)\n",
		       i, placement->placement[i], mem_type);
99
		ttm_mem_type_debug(bo->bdev, mem_type);
100 101 102
	}
}

103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118
static ssize_t ttm_bo_global_show(struct kobject *kobj,
				  struct attribute *attr,
				  char *buffer)
{
	struct ttm_bo_global *glob =
		container_of(kobj, struct ttm_bo_global, kobj);

	return snprintf(buffer, PAGE_SIZE, "%lu\n",
			(unsigned long) atomic_read(&glob->bo_count));
}

static struct attribute *ttm_bo_global_attrs[] = {
	&ttm_bo_count,
	NULL
};

119
static const struct sysfs_ops ttm_bo_global_ops = {
120 121 122 123 124 125 126 127 128
	.show = &ttm_bo_global_show
};

static struct kobj_type ttm_bo_glob_kobj_type  = {
	.release = &ttm_bo_global_kobj_release,
	.sysfs_ops = &ttm_bo_global_ops,
	.default_attrs = ttm_bo_global_attrs
};

129 130 131 132 133 134 135 136 137 138 139

static inline uint32_t ttm_bo_type_flags(unsigned type)
{
	return 1 << (type);
}

static void ttm_bo_release_list(struct kref *list_kref)
{
	struct ttm_buffer_object *bo =
	    container_of(list_kref, struct ttm_buffer_object, list_kref);
	struct ttm_bo_device *bdev = bo->bdev;
140
	size_t acc_size = bo->acc_size;
141 142 143 144 145 146 147 148 149 150 151

	BUG_ON(atomic_read(&bo->list_kref.refcount));
	BUG_ON(atomic_read(&bo->kref.refcount));
	BUG_ON(atomic_read(&bo->cpu_writers));
	BUG_ON(bo->sync_obj != NULL);
	BUG_ON(bo->mem.mm_node != NULL);
	BUG_ON(!list_empty(&bo->lru));
	BUG_ON(!list_empty(&bo->ddestroy));

	if (bo->ttm)
		ttm_tt_destroy(bo->ttm);
152
	atomic_dec(&bo->glob->bo_count);
153 154 155 156 157
	if (bo->destroy)
		bo->destroy(bo);
	else {
		kfree(bo);
	}
158
	ttm_mem_global_free(bdev->glob->mem_glob, acc_size);
159 160 161 162 163
}

int ttm_bo_wait_unreserved(struct ttm_buffer_object *bo, bool interruptible)
{
	if (interruptible) {
164
		return wait_event_interruptible(bo->event_queue,
165
					       !ttm_bo_is_reserved(bo));
166
	} else {
167
		wait_event(bo->event_queue, !ttm_bo_is_reserved(bo));
168
		return 0;
169 170
	}
}
171
EXPORT_SYMBOL(ttm_bo_wait_unreserved);
172

173
void ttm_bo_add_to_lru(struct ttm_buffer_object *bo)
174 175 176 177
{
	struct ttm_bo_device *bdev = bo->bdev;
	struct ttm_mem_type_manager *man;

178
	BUG_ON(!ttm_bo_is_reserved(bo));
179 180 181 182 183 184 185 186 187 188

	if (!(bo->mem.placement & TTM_PL_FLAG_NO_EVICT)) {

		BUG_ON(!list_empty(&bo->lru));

		man = &bdev->man[bo->mem.mem_type];
		list_add_tail(&bo->lru, &man->lru);
		kref_get(&bo->list_kref);

		if (bo->ttm != NULL) {
189
			list_add_tail(&bo->swap, &bo->glob->swap_lru);
190 191 192 193 194
			kref_get(&bo->list_kref);
		}
	}
}

195
int ttm_bo_del_from_lru(struct ttm_buffer_object *bo)
196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219
{
	int put_count = 0;

	if (!list_empty(&bo->swap)) {
		list_del_init(&bo->swap);
		++put_count;
	}
	if (!list_empty(&bo->lru)) {
		list_del_init(&bo->lru);
		++put_count;
	}

	/*
	 * TODO: Add a driver hook to delete from
	 * driver-specific LRU's here.
	 */

	return put_count;
}

int ttm_bo_reserve_locked(struct ttm_buffer_object *bo,
			  bool interruptible,
			  bool no_wait, bool use_sequence, uint32_t sequence)
{
220
	struct ttm_bo_global *glob = bo->glob;
221 222
	int ret;

223
	while (unlikely(atomic_read(&bo->reserved) != 0)) {
224 225 226
		/**
		 * Deadlock avoidance for multi-bo reserving.
		 */
227 228 229 230 231 232 233 234 235 236 237 238
		if (use_sequence && bo->seq_valid) {
			/**
			 * We've already reserved this one.
			 */
			if (unlikely(sequence == bo->val_seq))
				return -EDEADLK;
			/**
			 * Already reserved by a thread that will not back
			 * off for us. We need to back off.
			 */
			if (unlikely(sequence - bo->val_seq < (1 << 31)))
				return -EAGAIN;
239 240 241 242 243
		}

		if (no_wait)
			return -EBUSY;

244
		spin_unlock(&glob->lru_lock);
245
		ret = ttm_bo_wait_unreserved(bo, interruptible);
246
		spin_lock(&glob->lru_lock);
247 248 249 250 251

		if (unlikely(ret))
			return ret;
	}

252
	atomic_set(&bo->reserved, 1);
253
	if (use_sequence) {
254 255 256 257 258 259 260 261
		/**
		 * Wake up waiters that may need to recheck for deadlock,
		 * if we decreased the sequence number.
		 */
		if (unlikely((bo->val_seq - sequence < (1 << 31))
			     || !bo->seq_valid))
			wake_up_all(&bo->event_queue);

262 263 264 265 266 267 268 269 270 271 272 273 274 275 276
		bo->val_seq = sequence;
		bo->seq_valid = true;
	} else {
		bo->seq_valid = false;
	}

	return 0;
}
EXPORT_SYMBOL(ttm_bo_reserve);

static void ttm_bo_ref_bug(struct kref *list_kref)
{
	BUG();
}

277 278 279
void ttm_bo_list_ref_sub(struct ttm_buffer_object *bo, int count,
			 bool never_free)
{
280 281
	kref_sub(&bo->list_kref, count,
		 (never_free) ? ttm_bo_ref_bug : ttm_bo_release_list);
282 283
}

284 285 286 287
int ttm_bo_reserve(struct ttm_buffer_object *bo,
		   bool interruptible,
		   bool no_wait, bool use_sequence, uint32_t sequence)
{
288
	struct ttm_bo_global *glob = bo->glob;
289 290 291
	int put_count = 0;
	int ret;

292
	spin_lock(&glob->lru_lock);
293 294 295 296
	ret = ttm_bo_reserve_locked(bo, interruptible, no_wait, use_sequence,
				    sequence);
	if (likely(ret == 0))
		put_count = ttm_bo_del_from_lru(bo);
297
	spin_unlock(&glob->lru_lock);
298

299
	ttm_bo_list_ref_sub(bo, put_count, true);
300 301 302 303

	return ret;
}

304 305 306 307 308 309 310
void ttm_bo_unreserve_locked(struct ttm_buffer_object *bo)
{
	ttm_bo_add_to_lru(bo);
	atomic_set(&bo->reserved, 0);
	wake_up_all(&bo->event_queue);
}

311 312
void ttm_bo_unreserve(struct ttm_buffer_object *bo)
{
313
	struct ttm_bo_global *glob = bo->glob;
314

315
	spin_lock(&glob->lru_lock);
316
	ttm_bo_unreserve_locked(bo);
317
	spin_unlock(&glob->lru_lock);
318 319 320 321 322 323 324 325 326
}
EXPORT_SYMBOL(ttm_bo_unreserve);

/*
 * Call bo->mutex locked.
 */
static int ttm_bo_add_ttm(struct ttm_buffer_object *bo, bool zero_alloc)
{
	struct ttm_bo_device *bdev = bo->bdev;
327
	struct ttm_bo_global *glob = bo->glob;
328 329 330 331 332 333
	int ret = 0;
	uint32_t page_flags = 0;

	TTM_ASSERT_LOCKED(&bo->mutex);
	bo->ttm = NULL;

D
Dave Airlie 已提交
334 335 336
	if (bdev->need_dma32)
		page_flags |= TTM_PAGE_FLAG_DMA32;

337 338 339 340 341
	switch (bo->type) {
	case ttm_bo_type_device:
		if (zero_alloc)
			page_flags |= TTM_PAGE_FLAG_ZERO_ALLOC;
	case ttm_bo_type_kernel:
342 343
		bo->ttm = bdev->driver->ttm_tt_create(bdev, bo->num_pages << PAGE_SHIFT,
						      page_flags, glob->dummy_read_page);
344 345 346
		if (unlikely(bo->ttm == NULL))
			ret = -ENOMEM;
		break;
347 348 349 350 351 352 353 354 355 356
	case ttm_bo_type_sg:
		bo->ttm = bdev->driver->ttm_tt_create(bdev, bo->num_pages << PAGE_SHIFT,
						      page_flags | TTM_PAGE_FLAG_SG,
						      glob->dummy_read_page);
		if (unlikely(bo->ttm == NULL)) {
			ret = -ENOMEM;
			break;
		}
		bo->ttm->sg = bo->sg;
		break;
357
	default:
J
Joe Perches 已提交
358
		pr_err("Illegal buffer object type\n");
359 360 361 362 363 364 365 366 367
		ret = -EINVAL;
		break;
	}

	return ret;
}

static int ttm_bo_handle_move_mem(struct ttm_buffer_object *bo,
				  struct ttm_mem_reg *mem,
368 369
				  bool evict, bool interruptible,
				  bool no_wait_reserve, bool no_wait_gpu)
370 371 372 373 374 375 376 377 378
{
	struct ttm_bo_device *bdev = bo->bdev;
	bool old_is_pci = ttm_mem_reg_is_pci(bdev, &bo->mem);
	bool new_is_pci = ttm_mem_reg_is_pci(bdev, mem);
	struct ttm_mem_type_manager *old_man = &bdev->man[bo->mem.mem_type];
	struct ttm_mem_type_manager *new_man = &bdev->man[mem->mem_type];
	int ret = 0;

	if (old_is_pci || new_is_pci ||
379 380 381 382 383 384 385
	    ((mem->placement & bo->mem.placement & TTM_PL_MASK_CACHING) == 0)) {
		ret = ttm_mem_io_lock(old_man, true);
		if (unlikely(ret != 0))
			goto out_err;
		ttm_bo_unmap_virtual_locked(bo);
		ttm_mem_io_unlock(old_man);
	}
386 387 388 389 390

	/*
	 * Create and bind a ttm if required.
	 */

391 392
	if (!(new_man->flags & TTM_MEMTYPE_FLAG_FIXED)) {
		if (bo->ttm == NULL) {
393 394
			bool zero = !(old_man->flags & TTM_MEMTYPE_FLAG_FIXED);
			ret = ttm_bo_add_ttm(bo, zero);
395 396 397
			if (ret)
				goto out_err;
		}
398 399 400

		ret = ttm_tt_set_placement_caching(bo->ttm, mem->placement);
		if (ret)
401
			goto out_err;
402 403 404 405 406 407 408 409

		if (mem->mem_type != TTM_PL_SYSTEM) {
			ret = ttm_tt_bind(bo->ttm, mem);
			if (ret)
				goto out_err;
		}

		if (bo->mem.mem_type == TTM_PL_SYSTEM) {
410 411
			if (bdev->driver->move_notify)
				bdev->driver->move_notify(bo, mem);
412
			bo->mem = *mem;
413 414 415 416 417
			mem->mm_node = NULL;
			goto moved;
		}
	}

418 419 420
	if (bdev->driver->move_notify)
		bdev->driver->move_notify(bo, mem);

421 422
	if (!(old_man->flags & TTM_MEMTYPE_FLAG_FIXED) &&
	    !(new_man->flags & TTM_MEMTYPE_FLAG_FIXED))
423
		ret = ttm_bo_move_ttm(bo, evict, no_wait_reserve, no_wait_gpu, mem);
424 425
	else if (bdev->driver->move)
		ret = bdev->driver->move(bo, evict, interruptible,
426
					 no_wait_reserve, no_wait_gpu, mem);
427
	else
428
		ret = ttm_bo_move_memcpy(bo, evict, no_wait_reserve, no_wait_gpu, mem);
429

430 431 432 433 434 435 436 437
	if (ret) {
		if (bdev->driver->move_notify) {
			struct ttm_mem_reg tmp_mem = *mem;
			*mem = bo->mem;
			bo->mem = tmp_mem;
			bdev->driver->move_notify(bo, mem);
			bo->mem = *mem;
		}
438

439 440
		goto out_err;
	}
441

442 443 444 445
moved:
	if (bo->evicted) {
		ret = bdev->driver->invalidate_caches(bdev, bo->mem.placement);
		if (ret)
J
Joe Perches 已提交
446
			pr_err("Can not flush read caches\n");
447 448 449 450
		bo->evicted = false;
	}

	if (bo->mem.mm_node) {
451
		bo->offset = (bo->mem.start << PAGE_SHIFT) +
452 453
		    bdev->man[bo->mem.mem_type].gpu_offset;
		bo->cur_placement = bo->mem.placement;
454 455
	} else
		bo->offset = 0;
456 457 458 459 460 461 462 463 464 465 466 467 468 469

	return 0;

out_err:
	new_man = &bdev->man[bo->mem.mem_type];
	if ((new_man->flags & TTM_MEMTYPE_FLAG_FIXED) && bo->ttm) {
		ttm_tt_unbind(bo->ttm);
		ttm_tt_destroy(bo->ttm);
		bo->ttm = NULL;
	}

	return ret;
}

470
/**
471
 * Call bo::reserved.
472
 * Will release GPU memory type usage on destruction.
473 474 475
 * This is the place to put in driver specific hooks to release
 * driver private resources.
 * Will release the bo::reserved lock.
476 477 478 479
 */

static void ttm_bo_cleanup_memtype_use(struct ttm_buffer_object *bo)
{
480 481 482
	if (bo->bdev->driver->move_notify)
		bo->bdev->driver->move_notify(bo, NULL);

483 484 485 486 487
	if (bo->ttm) {
		ttm_tt_unbind(bo->ttm);
		ttm_tt_destroy(bo->ttm);
		bo->ttm = NULL;
	}
488
	ttm_bo_mem_put(bo, &bo->mem);
489 490

	atomic_set(&bo->reserved, 0);
491 492 493 494 495

	/*
	 * Make processes trying to reserve really pick it up.
	 */
	smp_mb__after_atomic_dec();
496 497 498
	wake_up_all(&bo->event_queue);
}

499
static void ttm_bo_cleanup_refs_or_queue(struct ttm_buffer_object *bo)
500 501
{
	struct ttm_bo_device *bdev = bo->bdev;
502
	struct ttm_bo_global *glob = bo->glob;
503
	struct ttm_bo_driver *driver = bdev->driver;
504
	void *sync_obj = NULL;
505
	int put_count;
506 507
	int ret;

508 509 510
	spin_lock(&glob->lru_lock);
	ret = ttm_bo_reserve_locked(bo, false, true, false, 0);

511
	spin_lock(&bdev->fence_lock);
512
	(void) ttm_bo_wait(bo, false, false, true);
513
	if (!ret && !bo->sync_obj) {
514
		spin_unlock(&bdev->fence_lock);
515
		put_count = ttm_bo_del_from_lru(bo);
516

517
		spin_unlock(&glob->lru_lock);
518
		ttm_bo_cleanup_memtype_use(bo);
519

520
		ttm_bo_list_ref_sub(bo, put_count, true);
521

522
		return;
523
	}
524 525
	if (bo->sync_obj)
		sync_obj = driver->sync_obj_ref(bo->sync_obj);
526 527 528 529 530 531
	spin_unlock(&bdev->fence_lock);

	if (!ret) {
		atomic_set(&bo->reserved, 0);
		wake_up_all(&bo->event_queue);
	}
532 533 534 535 536

	kref_get(&bo->list_kref);
	list_add_tail(&bo->ddestroy, &bdev->ddestroy);
	spin_unlock(&glob->lru_lock);

537
	if (sync_obj) {
538
		driver->sync_obj_flush(sync_obj);
539 540
		driver->sync_obj_unref(&sync_obj);
	}
541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559
	schedule_delayed_work(&bdev->wq,
			      ((HZ / 100) < 1) ? 1 : HZ / 100);
}

/**
 * function ttm_bo_cleanup_refs
 * If bo idle, remove from delayed- and lru lists, and unref.
 * If not idle, do nothing.
 *
 * @interruptible         Any sleeps should occur interruptibly.
 * @no_wait_reserve       Never wait for reserve. Return -EBUSY instead.
 * @no_wait_gpu           Never wait for gpu. Return -EBUSY instead.
 */

static int ttm_bo_cleanup_refs(struct ttm_buffer_object *bo,
			       bool interruptible,
			       bool no_wait_reserve,
			       bool no_wait_gpu)
{
560
	struct ttm_bo_device *bdev = bo->bdev;
561 562 563 564 565
	struct ttm_bo_global *glob = bo->glob;
	int put_count;
	int ret = 0;

retry:
566
	spin_lock(&bdev->fence_lock);
567
	ret = ttm_bo_wait(bo, false, interruptible, no_wait_gpu);
568
	spin_unlock(&bdev->fence_lock);
569 570 571 572

	if (unlikely(ret != 0))
		return ret;

573
retry_reserve:
574
	spin_lock(&glob->lru_lock);
575 576 577 578 579 580

	if (unlikely(list_empty(&bo->ddestroy))) {
		spin_unlock(&glob->lru_lock);
		return 0;
	}

581
	ret = ttm_bo_reserve_locked(bo, false, true, false, 0);
582

583
	if (unlikely(ret == -EBUSY)) {
584
		spin_unlock(&glob->lru_lock);
585 586 587 588 589 590
		if (likely(!no_wait_reserve))
			ret = ttm_bo_wait_unreserved(bo, interruptible);
		if (unlikely(ret != 0))
			return ret;

		goto retry_reserve;
591
	}
592

593 594
	BUG_ON(ret != 0);

595 596 597 598 599 600 601
	/**
	 * We can re-check for sync object without taking
	 * the bo::lock since setting the sync object requires
	 * also bo::reserved. A busy object at this point may
	 * be caused by another thread recently starting an accelerated
	 * eviction.
	 */
602

603 604 605
	if (unlikely(bo->sync_obj)) {
		atomic_set(&bo->reserved, 0);
		wake_up_all(&bo->event_queue);
606
		spin_unlock(&glob->lru_lock);
607
		goto retry;
608 609
	}

610 611 612 613 614 615 616
	put_count = ttm_bo_del_from_lru(bo);
	list_del_init(&bo->ddestroy);
	++put_count;

	spin_unlock(&glob->lru_lock);
	ttm_bo_cleanup_memtype_use(bo);

617
	ttm_bo_list_ref_sub(bo, put_count, true);
618 619

	return 0;
620 621 622 623 624 625 626 627 628
}

/**
 * Traverse the delayed list, and call ttm_bo_cleanup_refs on all
 * encountered buffers.
 */

static int ttm_bo_delayed_delete(struct ttm_bo_device *bdev, bool remove_all)
{
629
	struct ttm_bo_global *glob = bdev->glob;
630 631
	struct ttm_buffer_object *entry = NULL;
	int ret = 0;
632

633
	spin_lock(&glob->lru_lock);
634 635 636 637 638 639 640 641 642 643 644 645 646
	if (list_empty(&bdev->ddestroy))
		goto out_unlock;

	entry = list_first_entry(&bdev->ddestroy,
		struct ttm_buffer_object, ddestroy);
	kref_get(&entry->list_kref);

	for (;;) {
		struct ttm_buffer_object *nentry = NULL;

		if (entry->ddestroy.next != &bdev->ddestroy) {
			nentry = list_first_entry(&entry->ddestroy,
				struct ttm_buffer_object, ddestroy);
647 648 649
			kref_get(&nentry->list_kref);
		}

650
		spin_unlock(&glob->lru_lock);
651 652
		ret = ttm_bo_cleanup_refs(entry, false, !remove_all,
					  !remove_all);
653
		kref_put(&entry->list_kref, ttm_bo_release_list);
654 655 656 657
		entry = nentry;

		if (ret || !entry)
			goto out;
658

659
		spin_lock(&glob->lru_lock);
660
		if (list_empty(&entry->ddestroy))
661 662 663
			break;
	}

664 665 666 667 668
out_unlock:
	spin_unlock(&glob->lru_lock);
out:
	if (entry)
		kref_put(&entry->list_kref, ttm_bo_release_list);
669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687
	return ret;
}

static void ttm_bo_delayed_workqueue(struct work_struct *work)
{
	struct ttm_bo_device *bdev =
	    container_of(work, struct ttm_bo_device, wq.work);

	if (ttm_bo_delayed_delete(bdev, false)) {
		schedule_delayed_work(&bdev->wq,
				      ((HZ / 100) < 1) ? 1 : HZ / 100);
	}
}

static void ttm_bo_release(struct kref *kref)
{
	struct ttm_buffer_object *bo =
	    container_of(kref, struct ttm_buffer_object, kref);
	struct ttm_bo_device *bdev = bo->bdev;
688
	struct ttm_mem_type_manager *man = &bdev->man[bo->mem.mem_type];
689

690
	write_lock(&bdev->vm_lock);
691 692 693 694 695 696
	if (likely(bo->vm_node != NULL)) {
		rb_erase(&bo->vm_rb, &bdev->addr_space_rb);
		drm_mm_put_block(bo->vm_node);
		bo->vm_node = NULL;
	}
	write_unlock(&bdev->vm_lock);
697 698 699
	ttm_mem_io_lock(man, false);
	ttm_mem_io_free_vm(bo);
	ttm_mem_io_unlock(man);
700
	ttm_bo_cleanup_refs_or_queue(bo);
701 702 703 704 705 706 707 708 709 710 711 712
	kref_put(&bo->list_kref, ttm_bo_release_list);
}

void ttm_bo_unref(struct ttm_buffer_object **p_bo)
{
	struct ttm_buffer_object *bo = *p_bo;

	*p_bo = NULL;
	kref_put(&bo->kref, ttm_bo_release);
}
EXPORT_SYMBOL(ttm_bo_unref);

713 714 715 716 717 718 719 720 721 722 723 724 725 726
int ttm_bo_lock_delayed_workqueue(struct ttm_bo_device *bdev)
{
	return cancel_delayed_work_sync(&bdev->wq);
}
EXPORT_SYMBOL(ttm_bo_lock_delayed_workqueue);

void ttm_bo_unlock_delayed_workqueue(struct ttm_bo_device *bdev, int resched)
{
	if (resched)
		schedule_delayed_work(&bdev->wq,
				      ((HZ / 100) < 1) ? 1 : HZ / 100);
}
EXPORT_SYMBOL(ttm_bo_unlock_delayed_workqueue);

727
static int ttm_bo_evict(struct ttm_buffer_object *bo, bool interruptible,
728
			bool no_wait_reserve, bool no_wait_gpu)
729 730 731
{
	struct ttm_bo_device *bdev = bo->bdev;
	struct ttm_mem_reg evict_mem;
732 733
	struct ttm_placement placement;
	int ret = 0;
734

735
	spin_lock(&bdev->fence_lock);
736
	ret = ttm_bo_wait(bo, false, interruptible, no_wait_gpu);
737
	spin_unlock(&bdev->fence_lock);
738

739
	if (unlikely(ret != 0)) {
740
		if (ret != -ERESTARTSYS) {
J
Joe Perches 已提交
741
			pr_err("Failed to expire sync object before buffer eviction\n");
742
		}
743 744 745
		goto out;
	}

746
	BUG_ON(!ttm_bo_is_reserved(bo));
747 748 749

	evict_mem = bo->mem;
	evict_mem.mm_node = NULL;
750 751
	evict_mem.bus.io_reserved_vm = false;
	evict_mem.bus.io_reserved_count = 0;
752

753 754 755 756
	placement.fpfn = 0;
	placement.lpfn = 0;
	placement.num_placement = 0;
	placement.num_busy_placement = 0;
757 758
	bdev->driver->evict_flags(bo, &placement);
	ret = ttm_bo_mem_space(bo, &placement, &evict_mem, interruptible,
759
				no_wait_reserve, no_wait_gpu);
760
	if (ret) {
761
		if (ret != -ERESTARTSYS) {
J
Joe Perches 已提交
762 763
			pr_err("Failed to find memory space for buffer 0x%p eviction\n",
			       bo);
764 765
			ttm_bo_mem_space_debug(bo, &placement);
		}
766 767 768 769
		goto out;
	}

	ret = ttm_bo_handle_move_mem(bo, &evict_mem, true, interruptible,
770
				     no_wait_reserve, no_wait_gpu);
771
	if (ret) {
772
		if (ret != -ERESTARTSYS)
J
Joe Perches 已提交
773
			pr_err("Buffer eviction failed\n");
774
		ttm_bo_mem_put(bo, &evict_mem);
775 776
		goto out;
	}
777 778 779 780 781 782 783
	bo->evicted = true;
out:
	return ret;
}

static int ttm_mem_evict_first(struct ttm_bo_device *bdev,
				uint32_t mem_type,
784 785
				bool interruptible, bool no_wait_reserve,
				bool no_wait_gpu)
786 787 788 789 790
{
	struct ttm_bo_global *glob = bdev->glob;
	struct ttm_mem_type_manager *man = &bdev->man[mem_type];
	struct ttm_buffer_object *bo;
	int ret, put_count = 0;
791

792
retry:
793
	spin_lock(&glob->lru_lock);
794 795 796 797 798
	if (list_empty(&man->lru)) {
		spin_unlock(&glob->lru_lock);
		return -EBUSY;
	}

799 800
	bo = list_first_entry(&man->lru, struct ttm_buffer_object, lru);
	kref_get(&bo->list_kref);
801

802 803 804 805 806 807
	if (!list_empty(&bo->ddestroy)) {
		spin_unlock(&glob->lru_lock);
		ret = ttm_bo_cleanup_refs(bo, interruptible,
					  no_wait_reserve, no_wait_gpu);
		kref_put(&bo->list_kref, ttm_bo_release_list);

808
		return ret;
809 810
	}

T
Thomas Hellstrom 已提交
811
	ret = ttm_bo_reserve_locked(bo, false, true, false, 0);
812 813 814

	if (unlikely(ret == -EBUSY)) {
		spin_unlock(&glob->lru_lock);
T
Thomas Hellstrom 已提交
815
		if (likely(!no_wait_reserve))
816 817 818 819 820 821 822 823 824 825 826 827 828 829
			ret = ttm_bo_wait_unreserved(bo, interruptible);

		kref_put(&bo->list_kref, ttm_bo_release_list);

		/**
		 * We *need* to retry after releasing the lru lock.
		 */

		if (unlikely(ret != 0))
			return ret;
		goto retry;
	}

	put_count = ttm_bo_del_from_lru(bo);
830
	spin_unlock(&glob->lru_lock);
831 832 833

	BUG_ON(ret != 0);

834
	ttm_bo_list_ref_sub(bo, put_count, true);
835

836
	ret = ttm_bo_evict(bo, interruptible, no_wait_reserve, no_wait_gpu);
837
	ttm_bo_unreserve(bo);
838

839
	kref_put(&bo->list_kref, ttm_bo_release_list);
840 841 842
	return ret;
}

843 844
void ttm_bo_mem_put(struct ttm_buffer_object *bo, struct ttm_mem_reg *mem)
{
845
	struct ttm_mem_type_manager *man = &bo->bdev->man[mem->mem_type];
846

847 848
	if (mem->mm_node)
		(*man->func->put_node)(man, mem);
849 850 851
}
EXPORT_SYMBOL(ttm_bo_mem_put);

852 853 854 855
/**
 * Repeatedly evict memory from the LRU for @mem_type until we create enough
 * space, or we've evicted everything and there isn't enough space.
 */
856 857 858 859
static int ttm_bo_mem_force_space(struct ttm_buffer_object *bo,
					uint32_t mem_type,
					struct ttm_placement *placement,
					struct ttm_mem_reg *mem,
860 861 862
					bool interruptible,
					bool no_wait_reserve,
					bool no_wait_gpu)
863
{
864
	struct ttm_bo_device *bdev = bo->bdev;
865 866 867 868
	struct ttm_mem_type_manager *man = &bdev->man[mem_type];
	int ret;

	do {
869
		ret = (*man->func->get_node)(man, bo, placement, mem);
870 871
		if (unlikely(ret != 0))
			return ret;
872
		if (mem->mm_node)
873
			break;
874
		ret = ttm_mem_evict_first(bdev, mem_type, interruptible,
875
						no_wait_reserve, no_wait_gpu);
876 877 878
		if (unlikely(ret != 0))
			return ret;
	} while (1);
879
	if (mem->mm_node == NULL)
880 881 882 883 884
		return -ENOMEM;
	mem->mem_type = mem_type;
	return 0;
}

885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909
static uint32_t ttm_bo_select_caching(struct ttm_mem_type_manager *man,
				      uint32_t cur_placement,
				      uint32_t proposed_placement)
{
	uint32_t caching = proposed_placement & TTM_PL_MASK_CACHING;
	uint32_t result = proposed_placement & ~TTM_PL_MASK_CACHING;

	/**
	 * Keep current caching if possible.
	 */

	if ((cur_placement & caching) != 0)
		result |= (cur_placement & caching);
	else if ((man->default_caching & caching) != 0)
		result |= man->default_caching;
	else if ((TTM_PL_FLAG_CACHED & caching) != 0)
		result |= TTM_PL_FLAG_CACHED;
	else if ((TTM_PL_FLAG_WC & caching) != 0)
		result |= TTM_PL_FLAG_WC;
	else if ((TTM_PL_FLAG_UNCACHED & caching) != 0)
		result |= TTM_PL_FLAG_UNCACHED;

	return result;
}

910 911
static bool ttm_bo_mt_compatible(struct ttm_mem_type_manager *man,
				 uint32_t mem_type,
912 913
				 uint32_t proposed_placement,
				 uint32_t *masked_placement)
914 915 916
{
	uint32_t cur_flags = ttm_bo_type_flags(mem_type);

917
	if ((cur_flags & proposed_placement & TTM_PL_MASK_MEM) == 0)
918 919
		return false;

920
	if ((proposed_placement & man->available_caching) == 0)
921 922
		return false;

923 924 925
	cur_flags |= (proposed_placement & man->available_caching);

	*masked_placement = cur_flags;
926 927 928 929 930 931 932 933 934 935 936 937
	return true;
}

/**
 * Creates space for memory region @mem according to its type.
 *
 * This function first searches for free space in compatible memory types in
 * the priority order defined by the driver.  If free space isn't found, then
 * ttm_bo_mem_force_space is attempted in priority order to evict and find
 * space.
 */
int ttm_bo_mem_space(struct ttm_buffer_object *bo,
938 939
			struct ttm_placement *placement,
			struct ttm_mem_reg *mem,
940 941
			bool interruptible, bool no_wait_reserve,
			bool no_wait_gpu)
942 943 944 945 946 947 948
{
	struct ttm_bo_device *bdev = bo->bdev;
	struct ttm_mem_type_manager *man;
	uint32_t mem_type = TTM_PL_SYSTEM;
	uint32_t cur_flags = 0;
	bool type_found = false;
	bool type_ok = false;
949
	bool has_erestartsys = false;
950
	int i, ret;
951 952

	mem->mm_node = NULL;
953
	for (i = 0; i < placement->num_placement; ++i) {
954 955 956 957
		ret = ttm_mem_type_from_flags(placement->placement[i],
						&mem_type);
		if (ret)
			return ret;
958 959 960
		man = &bdev->man[mem_type];

		type_ok = ttm_bo_mt_compatible(man,
961 962 963
						mem_type,
						placement->placement[i],
						&cur_flags);
964 965 966 967

		if (!type_ok)
			continue;

968 969
		cur_flags = ttm_bo_select_caching(man, bo->mem.placement,
						  cur_flags);
970 971 972 973 974 975
		/*
		 * Use the access and other non-mapping-related flag bits from
		 * the memory placement flags to the current flags
		 */
		ttm_flag_masked(&cur_flags, placement->placement[i],
				~TTM_PL_MASK_MEMTYPE);
976

977 978 979 980 981
		if (mem_type == TTM_PL_SYSTEM)
			break;

		if (man->has_type && man->use_type) {
			type_found = true;
982
			ret = (*man->func->get_node)(man, bo, placement, mem);
983 984
			if (unlikely(ret))
				return ret;
985
		}
986
		if (mem->mm_node)
987 988 989
			break;
	}

990
	if ((type_ok && (mem_type == TTM_PL_SYSTEM)) || mem->mm_node) {
991 992 993 994 995 996 997 998
		mem->mem_type = mem_type;
		mem->placement = cur_flags;
		return 0;
	}

	if (!type_found)
		return -EINVAL;

999 1000
	for (i = 0; i < placement->num_busy_placement; ++i) {
		ret = ttm_mem_type_from_flags(placement->busy_placement[i],
1001 1002 1003
						&mem_type);
		if (ret)
			return ret;
1004 1005 1006 1007
		man = &bdev->man[mem_type];
		if (!man->has_type)
			continue;
		if (!ttm_bo_mt_compatible(man,
1008
						mem_type,
1009
						placement->busy_placement[i],
1010
						&cur_flags))
1011 1012
			continue;

1013 1014
		cur_flags = ttm_bo_select_caching(man, bo->mem.placement,
						  cur_flags);
1015 1016 1017 1018
		/*
		 * Use the access and other non-mapping-related flag bits from
		 * the memory placement flags to the current flags
		 */
1019
		ttm_flag_masked(&cur_flags, placement->busy_placement[i],
1020
				~TTM_PL_MASK_MEMTYPE);
1021

1022 1023 1024 1025 1026 1027 1028 1029

		if (mem_type == TTM_PL_SYSTEM) {
			mem->mem_type = mem_type;
			mem->placement = cur_flags;
			mem->mm_node = NULL;
			return 0;
		}

1030
		ret = ttm_bo_mem_force_space(bo, mem_type, placement, mem,
1031
						interruptible, no_wait_reserve, no_wait_gpu);
1032 1033 1034 1035
		if (ret == 0 && mem->mm_node) {
			mem->placement = cur_flags;
			return 0;
		}
1036 1037
		if (ret == -ERESTARTSYS)
			has_erestartsys = true;
1038
	}
1039
	ret = (has_erestartsys) ? -ERESTARTSYS : -ENOMEM;
1040 1041 1042 1043 1044
	return ret;
}
EXPORT_SYMBOL(ttm_bo_mem_space);

int ttm_bo_move_buffer(struct ttm_buffer_object *bo,
1045
			struct ttm_placement *placement,
1046 1047
			bool interruptible, bool no_wait_reserve,
			bool no_wait_gpu)
1048 1049 1050
{
	int ret = 0;
	struct ttm_mem_reg mem;
1051
	struct ttm_bo_device *bdev = bo->bdev;
1052

1053
	BUG_ON(!ttm_bo_is_reserved(bo));
1054 1055 1056 1057 1058 1059

	/*
	 * FIXME: It's possible to pipeline buffer moves.
	 * Have the driver move function wait for idle when necessary,
	 * instead of doing it here.
	 */
1060
	spin_lock(&bdev->fence_lock);
1061
	ret = ttm_bo_wait(bo, false, interruptible, no_wait_gpu);
1062
	spin_unlock(&bdev->fence_lock);
1063 1064 1065 1066 1067
	if (ret)
		return ret;
	mem.num_pages = bo->num_pages;
	mem.size = mem.num_pages << PAGE_SHIFT;
	mem.page_alignment = bo->mem.page_alignment;
1068 1069
	mem.bus.io_reserved_vm = false;
	mem.bus.io_reserved_count = 0;
1070 1071 1072
	/*
	 * Determine where to move the buffer.
	 */
1073
	ret = ttm_bo_mem_space(bo, placement, &mem, interruptible, no_wait_reserve, no_wait_gpu);
1074 1075
	if (ret)
		goto out_unlock;
1076
	ret = ttm_bo_handle_move_mem(bo, &mem, false, interruptible, no_wait_reserve, no_wait_gpu);
1077
out_unlock:
1078 1079
	if (ret && mem.mm_node)
		ttm_bo_mem_put(bo, &mem);
1080 1081 1082
	return ret;
}

1083
static int ttm_bo_mem_compat(struct ttm_placement *placement,
1084 1085
			     struct ttm_mem_reg *mem)
{
1086
	int i;
1087

1088 1089 1090
	if (mem->mm_node && placement->lpfn != 0 &&
	    (mem->start < placement->fpfn ||
	     mem->start + mem->num_pages > placement->lpfn))
1091
		return -1;
1092 1093 1094 1095 1096 1097 1098 1099 1100

	for (i = 0; i < placement->num_placement; i++) {
		if ((placement->placement[i] & mem->placement &
			TTM_PL_MASK_CACHING) &&
			(placement->placement[i] & mem->placement &
			TTM_PL_MASK_MEM))
			return i;
	}
	return -1;
1101 1102
}

1103 1104
int ttm_bo_validate(struct ttm_buffer_object *bo,
			struct ttm_placement *placement,
1105 1106
			bool interruptible, bool no_wait_reserve,
			bool no_wait_gpu)
1107 1108 1109
{
	int ret;

1110
	BUG_ON(!ttm_bo_is_reserved(bo));
1111 1112 1113 1114 1115
	/* Check that range is valid */
	if (placement->lpfn || placement->fpfn)
		if (placement->fpfn > placement->lpfn ||
			(placement->lpfn - placement->fpfn) < bo->num_pages)
			return -EINVAL;
1116 1117 1118
	/*
	 * Check whether we need to move buffer.
	 */
1119 1120
	ret = ttm_bo_mem_compat(placement, &bo->mem);
	if (ret < 0) {
1121
		ret = ttm_bo_move_buffer(bo, placement, interruptible, no_wait_reserve, no_wait_gpu);
1122
		if (ret)
1123
			return ret;
1124 1125 1126 1127 1128 1129 1130
	} else {
		/*
		 * Use the access and other non-mapping-related flag bits from
		 * the compatible memory placement flags to the active flags
		 */
		ttm_flag_masked(&bo->mem.placement, placement->placement[ret],
				~TTM_PL_MASK_MEMTYPE);
1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141
	}
	/*
	 * We might need to add a TTM.
	 */
	if (bo->mem.mem_type == TTM_PL_SYSTEM && bo->ttm == NULL) {
		ret = ttm_bo_add_ttm(bo, true);
		if (ret)
			return ret;
	}
	return 0;
}
1142
EXPORT_SYMBOL(ttm_bo_validate);
1143

1144 1145
int ttm_bo_check_placement(struct ttm_buffer_object *bo,
				struct ttm_placement *placement)
1146
{
1147 1148
	BUG_ON((placement->fpfn || placement->lpfn) &&
	       (bo->mem.num_pages > (placement->lpfn - placement->fpfn)));
1149 1150 1151 1152

	return 0;
}

1153 1154 1155 1156 1157 1158 1159
int ttm_bo_init(struct ttm_bo_device *bdev,
		struct ttm_buffer_object *bo,
		unsigned long size,
		enum ttm_bo_type type,
		struct ttm_placement *placement,
		uint32_t page_alignment,
		bool interruptible,
J
Jan Engelhardt 已提交
1160
		struct file *persistent_swap_storage,
1161
		size_t acc_size,
1162
		struct sg_table *sg,
1163
		void (*destroy) (struct ttm_buffer_object *))
1164
{
1165
	int ret = 0;
1166
	unsigned long num_pages;
1167 1168 1169 1170
	struct ttm_mem_global *mem_glob = bdev->glob->mem_glob;

	ret = ttm_mem_global_alloc(mem_glob, acc_size, false, false);
	if (ret) {
J
Joe Perches 已提交
1171
		pr_err("Out of kernel memory\n");
1172 1173 1174 1175 1176 1177
		if (destroy)
			(*destroy)(bo);
		else
			kfree(bo);
		return -ENOMEM;
	}
1178 1179 1180

	num_pages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
	if (num_pages == 0) {
J
Joe Perches 已提交
1181
		pr_err("Illegal buffer object size\n");
1182 1183 1184 1185
		if (destroy)
			(*destroy)(bo);
		else
			kfree(bo);
1186
		ttm_mem_global_free(mem_glob, acc_size);
1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198
		return -EINVAL;
	}
	bo->destroy = destroy;

	kref_init(&bo->kref);
	kref_init(&bo->list_kref);
	atomic_set(&bo->cpu_writers, 0);
	atomic_set(&bo->reserved, 1);
	init_waitqueue_head(&bo->event_queue);
	INIT_LIST_HEAD(&bo->lru);
	INIT_LIST_HEAD(&bo->ddestroy);
	INIT_LIST_HEAD(&bo->swap);
1199
	INIT_LIST_HEAD(&bo->io_reserve_lru);
1200
	bo->bdev = bdev;
1201
	bo->glob = bdev->glob;
1202 1203
	bo->type = type;
	bo->num_pages = num_pages;
1204
	bo->mem.size = num_pages << PAGE_SHIFT;
1205 1206 1207 1208
	bo->mem.mem_type = TTM_PL_SYSTEM;
	bo->mem.num_pages = bo->num_pages;
	bo->mem.mm_node = NULL;
	bo->mem.page_alignment = page_alignment;
1209 1210
	bo->mem.bus.io_reserved_vm = false;
	bo->mem.bus.io_reserved_count = 0;
1211 1212 1213
	bo->priv_flags = 0;
	bo->mem.placement = (TTM_PL_FLAG_SYSTEM | TTM_PL_FLAG_CACHED);
	bo->seq_valid = false;
J
Jan Engelhardt 已提交
1214
	bo->persistent_swap_storage = persistent_swap_storage;
1215
	bo->acc_size = acc_size;
1216
	bo->sg = sg;
1217
	atomic_inc(&bo->glob->bo_count);
1218

1219
	ret = ttm_bo_check_placement(bo, placement);
1220 1221 1222 1223 1224 1225 1226
	if (unlikely(ret != 0))
		goto out_err;

	/*
	 * For ttm_bo_type_device buffers, allocate
	 * address space from the device.
	 */
1227 1228
	if (bo->type == ttm_bo_type_device ||
	    bo->type == ttm_bo_type_sg) {
1229 1230 1231 1232 1233
		ret = ttm_bo_setup_vm(bo);
		if (ret)
			goto out_err;
	}

1234
	ret = ttm_bo_validate(bo, placement, interruptible, false, false);
1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246
	if (ret)
		goto out_err;

	ttm_bo_unreserve(bo);
	return 0;

out_err:
	ttm_bo_unreserve(bo);
	ttm_bo_unref(&bo);

	return ret;
}
1247
EXPORT_SYMBOL(ttm_bo_init);
1248

1249 1250 1251
size_t ttm_bo_acc_size(struct ttm_bo_device *bdev,
		       unsigned long bo_size,
		       unsigned struct_size)
1252
{
1253 1254
	unsigned npages = (PAGE_ALIGN(bo_size)) >> PAGE_SHIFT;
	size_t size = 0;
1255

1256 1257 1258 1259
	size += ttm_round_pot(struct_size);
	size += PAGE_ALIGN(npages * sizeof(void *));
	size += ttm_round_pot(sizeof(struct ttm_tt));
	return size;
1260
}
1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276
EXPORT_SYMBOL(ttm_bo_acc_size);

size_t ttm_bo_dma_acc_size(struct ttm_bo_device *bdev,
			   unsigned long bo_size,
			   unsigned struct_size)
{
	unsigned npages = (PAGE_ALIGN(bo_size)) >> PAGE_SHIFT;
	size_t size = 0;

	size += ttm_round_pot(struct_size);
	size += PAGE_ALIGN(npages * sizeof(void *));
	size += PAGE_ALIGN(npages * sizeof(dma_addr_t));
	size += ttm_round_pot(sizeof(struct ttm_dma_tt));
	return size;
}
EXPORT_SYMBOL(ttm_bo_dma_acc_size);
1277

1278 1279 1280 1281 1282 1283
int ttm_bo_create(struct ttm_bo_device *bdev,
			unsigned long size,
			enum ttm_bo_type type,
			struct ttm_placement *placement,
			uint32_t page_alignment,
			bool interruptible,
J
Jan Engelhardt 已提交
1284
			struct file *persistent_swap_storage,
1285
			struct ttm_buffer_object **p_bo)
1286 1287
{
	struct ttm_buffer_object *bo;
1288
	size_t acc_size;
1289
	int ret;
1290 1291

	bo = kzalloc(sizeof(*bo), GFP_KERNEL);
1292
	if (unlikely(bo == NULL))
1293 1294
		return -ENOMEM;

1295
	acc_size = ttm_bo_acc_size(bdev, size, sizeof(struct ttm_buffer_object));
1296
	ret = ttm_bo_init(bdev, bo, size, type, placement, page_alignment,
1297 1298
			  interruptible, persistent_swap_storage, acc_size,
			  NULL, NULL);
1299 1300 1301 1302 1303
	if (likely(ret == 0))
		*p_bo = bo;

	return ret;
}
T
Thomas Hellstrom 已提交
1304
EXPORT_SYMBOL(ttm_bo_create);
1305 1306

static int ttm_bo_force_list_clean(struct ttm_bo_device *bdev,
1307
					unsigned mem_type, bool allow_errors)
1308
{
1309
	struct ttm_mem_type_manager *man = &bdev->man[mem_type];
1310
	struct ttm_bo_global *glob = bdev->glob;
1311 1312 1313 1314 1315 1316
	int ret;

	/*
	 * Can't use standard list traversal since we're unlocking.
	 */

1317
	spin_lock(&glob->lru_lock);
1318
	while (!list_empty(&man->lru)) {
1319
		spin_unlock(&glob->lru_lock);
1320
		ret = ttm_mem_evict_first(bdev, mem_type, false, false, false);
1321 1322 1323 1324
		if (ret) {
			if (allow_errors) {
				return ret;
			} else {
J
Joe Perches 已提交
1325
				pr_err("Cleanup eviction failed\n");
1326 1327
			}
		}
1328
		spin_lock(&glob->lru_lock);
1329
	}
1330
	spin_unlock(&glob->lru_lock);
1331 1332 1333 1334 1335
	return 0;
}

int ttm_bo_clean_mm(struct ttm_bo_device *bdev, unsigned mem_type)
{
R
Roel Kluin 已提交
1336
	struct ttm_mem_type_manager *man;
1337 1338 1339
	int ret = -EINVAL;

	if (mem_type >= TTM_NUM_MEM_TYPES) {
J
Joe Perches 已提交
1340
		pr_err("Illegal memory type %d\n", mem_type);
1341 1342
		return ret;
	}
R
Roel Kluin 已提交
1343
	man = &bdev->man[mem_type];
1344 1345

	if (!man->has_type) {
J
Joe Perches 已提交
1346 1347
		pr_err("Trying to take down uninitialized memory manager type %u\n",
		       mem_type);
1348 1349 1350 1351 1352 1353 1354 1355
		return ret;
	}

	man->use_type = false;
	man->has_type = false;

	ret = 0;
	if (mem_type > 0) {
1356
		ttm_bo_force_list_clean(bdev, mem_type, false);
1357

1358
		ret = (*man->func->takedown)(man);
1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369
	}

	return ret;
}
EXPORT_SYMBOL(ttm_bo_clean_mm);

int ttm_bo_evict_mm(struct ttm_bo_device *bdev, unsigned mem_type)
{
	struct ttm_mem_type_manager *man = &bdev->man[mem_type];

	if (mem_type == 0 || mem_type >= TTM_NUM_MEM_TYPES) {
J
Joe Perches 已提交
1370
		pr_err("Illegal memory manager memory type %u\n", mem_type);
1371 1372 1373 1374
		return -EINVAL;
	}

	if (!man->has_type) {
J
Joe Perches 已提交
1375
		pr_err("Memory type %u has not been initialized\n", mem_type);
1376 1377 1378
		return 0;
	}

1379
	return ttm_bo_force_list_clean(bdev, mem_type, true);
1380 1381 1382 1383
}
EXPORT_SYMBOL(ttm_bo_evict_mm);

int ttm_bo_init_mm(struct ttm_bo_device *bdev, unsigned type,
1384
			unsigned long p_size)
1385 1386 1387 1388
{
	int ret = -EINVAL;
	struct ttm_mem_type_manager *man;

1389
	BUG_ON(type >= TTM_NUM_MEM_TYPES);
1390
	man = &bdev->man[type];
1391
	BUG_ON(man->has_type);
1392 1393 1394 1395
	man->io_reserve_fastpath = true;
	man->use_io_reserve_lru = false;
	mutex_init(&man->io_reserve_mutex);
	INIT_LIST_HEAD(&man->io_reserve_lru);
1396 1397 1398 1399

	ret = bdev->driver->init_mem_type(bdev, type, man);
	if (ret)
		return ret;
1400
	man->bdev = bdev;
1401 1402 1403

	ret = 0;
	if (type != TTM_PL_SYSTEM) {
1404
		ret = (*man->func->init)(man, p_size);
1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417
		if (ret)
			return ret;
	}
	man->has_type = true;
	man->use_type = true;
	man->size = p_size;

	INIT_LIST_HEAD(&man->lru);

	return 0;
}
EXPORT_SYMBOL(ttm_bo_init_mm);

1418 1419 1420 1421 1422 1423 1424 1425 1426 1427
static void ttm_bo_global_kobj_release(struct kobject *kobj)
{
	struct ttm_bo_global *glob =
		container_of(kobj, struct ttm_bo_global, kobj);

	ttm_mem_unregister_shrink(glob->mem_glob, &glob->shrink);
	__free_page(glob->dummy_read_page);
	kfree(glob);
}

1428
void ttm_bo_global_release(struct drm_global_reference *ref)
1429 1430 1431 1432 1433 1434 1435 1436
{
	struct ttm_bo_global *glob = ref->object;

	kobject_del(&glob->kobj);
	kobject_put(&glob->kobj);
}
EXPORT_SYMBOL(ttm_bo_global_release);

1437
int ttm_bo_global_init(struct drm_global_reference *ref)
1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459
{
	struct ttm_bo_global_ref *bo_ref =
		container_of(ref, struct ttm_bo_global_ref, ref);
	struct ttm_bo_global *glob = ref->object;
	int ret;

	mutex_init(&glob->device_list_mutex);
	spin_lock_init(&glob->lru_lock);
	glob->mem_glob = bo_ref->mem_glob;
	glob->dummy_read_page = alloc_page(__GFP_ZERO | GFP_DMA32);

	if (unlikely(glob->dummy_read_page == NULL)) {
		ret = -ENOMEM;
		goto out_no_drp;
	}

	INIT_LIST_HEAD(&glob->swap_lru);
	INIT_LIST_HEAD(&glob->device_list);

	ttm_mem_init_shrink(&glob->shrink, ttm_bo_swapout);
	ret = ttm_mem_register_shrink(glob->mem_glob, &glob->shrink);
	if (unlikely(ret != 0)) {
J
Joe Perches 已提交
1460
		pr_err("Could not register buffer object swapout\n");
1461 1462 1463 1464 1465
		goto out_no_shrink;
	}

	atomic_set(&glob->bo_count, 0);

1466 1467
	ret = kobject_init_and_add(
		&glob->kobj, &ttm_bo_glob_kobj_type, ttm_get_kobj(), "buffer_objects");
1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479
	if (unlikely(ret != 0))
		kobject_put(&glob->kobj);
	return ret;
out_no_shrink:
	__free_page(glob->dummy_read_page);
out_no_drp:
	kfree(glob);
	return ret;
}
EXPORT_SYMBOL(ttm_bo_global_init);


1480 1481 1482 1483 1484
int ttm_bo_device_release(struct ttm_bo_device *bdev)
{
	int ret = 0;
	unsigned i = TTM_NUM_MEM_TYPES;
	struct ttm_mem_type_manager *man;
1485
	struct ttm_bo_global *glob = bdev->glob;
1486 1487 1488 1489 1490 1491 1492

	while (i--) {
		man = &bdev->man[i];
		if (man->has_type) {
			man->use_type = false;
			if ((i != TTM_PL_SYSTEM) && ttm_bo_clean_mm(bdev, i)) {
				ret = -EBUSY;
J
Joe Perches 已提交
1493 1494
				pr_err("DRM memory manager type %d is not clean\n",
				       i);
1495 1496 1497 1498 1499
			}
			man->has_type = false;
		}
	}

1500 1501 1502 1503
	mutex_lock(&glob->device_list_mutex);
	list_del(&bdev->device_list);
	mutex_unlock(&glob->device_list_mutex);

1504
	cancel_delayed_work_sync(&bdev->wq);
1505 1506 1507 1508

	while (ttm_bo_delayed_delete(bdev, true))
		;

1509
	spin_lock(&glob->lru_lock);
1510 1511 1512 1513 1514
	if (list_empty(&bdev->ddestroy))
		TTM_DEBUG("Delayed destroy list was clean\n");

	if (list_empty(&bdev->man[0].lru))
		TTM_DEBUG("Swap list was clean\n");
1515
	spin_unlock(&glob->lru_lock);
1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526

	BUG_ON(!drm_mm_clean(&bdev->addr_space_mm));
	write_lock(&bdev->vm_lock);
	drm_mm_takedown(&bdev->addr_space_mm);
	write_unlock(&bdev->vm_lock);

	return ret;
}
EXPORT_SYMBOL(ttm_bo_device_release);

int ttm_bo_device_init(struct ttm_bo_device *bdev,
1527 1528
		       struct ttm_bo_global *glob,
		       struct ttm_bo_driver *driver,
D
Dave Airlie 已提交
1529
		       uint64_t file_page_offset,
D
Dave Airlie 已提交
1530
		       bool need_dma32)
1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542
{
	int ret = -EINVAL;

	rwlock_init(&bdev->vm_lock);
	bdev->driver = driver;

	memset(bdev->man, 0, sizeof(bdev->man));

	/*
	 * Initialize the system memory buffer type.
	 * Other types need to be driver / IOCTL initialized.
	 */
1543
	ret = ttm_bo_init_mm(bdev, TTM_PL_SYSTEM, 0);
1544
	if (unlikely(ret != 0))
1545
		goto out_no_sys;
1546 1547 1548 1549

	bdev->addr_space_rb = RB_ROOT;
	ret = drm_mm_init(&bdev->addr_space_mm, file_page_offset, 0x10000000);
	if (unlikely(ret != 0))
1550
		goto out_no_addr_mm;
1551 1552 1553 1554

	INIT_DELAYED_WORK(&bdev->wq, ttm_bo_delayed_workqueue);
	INIT_LIST_HEAD(&bdev->ddestroy);
	bdev->dev_mapping = NULL;
1555
	bdev->glob = glob;
D
Dave Airlie 已提交
1556
	bdev->need_dma32 = need_dma32;
1557
	bdev->val_seq = 0;
1558
	spin_lock_init(&bdev->fence_lock);
1559 1560 1561
	mutex_lock(&glob->device_list_mutex);
	list_add_tail(&bdev->device_list, &glob->device_list);
	mutex_unlock(&glob->device_list_mutex);
1562 1563

	return 0;
1564
out_no_addr_mm:
1565
	ttm_bo_clean_mm(bdev, 0);
1566
out_no_sys:
1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591
	return ret;
}
EXPORT_SYMBOL(ttm_bo_device_init);

/*
 * buffer object vm functions.
 */

bool ttm_mem_reg_is_pci(struct ttm_bo_device *bdev, struct ttm_mem_reg *mem)
{
	struct ttm_mem_type_manager *man = &bdev->man[mem->mem_type];

	if (!(man->flags & TTM_MEMTYPE_FLAG_FIXED)) {
		if (mem->mem_type == TTM_PL_SYSTEM)
			return false;

		if (man->flags & TTM_MEMTYPE_FLAG_CMA)
			return false;

		if (mem->placement & TTM_PL_FLAG_CACHED)
			return false;
	}
	return true;
}

1592
void ttm_bo_unmap_virtual_locked(struct ttm_buffer_object *bo)
1593 1594 1595 1596 1597 1598 1599 1600
{
	struct ttm_bo_device *bdev = bo->bdev;
	loff_t offset = (loff_t) bo->addr_space_offset;
	loff_t holelen = ((loff_t) bo->mem.num_pages) << PAGE_SHIFT;

	if (!bdev->dev_mapping)
		return;
	unmap_mapping_range(bdev->dev_mapping, offset, holelen, 1);
1601
	ttm_mem_io_free_vm(bo);
1602
}
1603 1604 1605 1606 1607 1608 1609 1610 1611

void ttm_bo_unmap_virtual(struct ttm_buffer_object *bo)
{
	struct ttm_bo_device *bdev = bo->bdev;
	struct ttm_mem_type_manager *man = &bdev->man[bo->mem.mem_type];

	ttm_mem_io_lock(man, false);
	ttm_bo_unmap_virtual_locked(bo);
	ttm_mem_io_unlock(man);
1612
}
1613 1614


1615
EXPORT_SYMBOL(ttm_bo_unmap_virtual);
1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661 1662 1663 1664 1665 1666 1667 1668 1669 1670 1671 1672 1673 1674 1675 1676 1677 1678 1679 1680 1681 1682 1683 1684 1685 1686 1687 1688 1689 1690

static void ttm_bo_vm_insert_rb(struct ttm_buffer_object *bo)
{
	struct ttm_bo_device *bdev = bo->bdev;
	struct rb_node **cur = &bdev->addr_space_rb.rb_node;
	struct rb_node *parent = NULL;
	struct ttm_buffer_object *cur_bo;
	unsigned long offset = bo->vm_node->start;
	unsigned long cur_offset;

	while (*cur) {
		parent = *cur;
		cur_bo = rb_entry(parent, struct ttm_buffer_object, vm_rb);
		cur_offset = cur_bo->vm_node->start;
		if (offset < cur_offset)
			cur = &parent->rb_left;
		else if (offset > cur_offset)
			cur = &parent->rb_right;
		else
			BUG();
	}

	rb_link_node(&bo->vm_rb, parent, cur);
	rb_insert_color(&bo->vm_rb, &bdev->addr_space_rb);
}

/**
 * ttm_bo_setup_vm:
 *
 * @bo: the buffer to allocate address space for
 *
 * Allocate address space in the drm device so that applications
 * can mmap the buffer and access the contents. This only
 * applies to ttm_bo_type_device objects as others are not
 * placed in the drm device address space.
 */

static int ttm_bo_setup_vm(struct ttm_buffer_object *bo)
{
	struct ttm_bo_device *bdev = bo->bdev;
	int ret;

retry_pre_get:
	ret = drm_mm_pre_get(&bdev->addr_space_mm);
	if (unlikely(ret != 0))
		return ret;

	write_lock(&bdev->vm_lock);
	bo->vm_node = drm_mm_search_free(&bdev->addr_space_mm,
					 bo->mem.num_pages, 0, 0);

	if (unlikely(bo->vm_node == NULL)) {
		ret = -ENOMEM;
		goto out_unlock;
	}

	bo->vm_node = drm_mm_get_block_atomic(bo->vm_node,
					      bo->mem.num_pages, 0);

	if (unlikely(bo->vm_node == NULL)) {
		write_unlock(&bdev->vm_lock);
		goto retry_pre_get;
	}

	ttm_bo_vm_insert_rb(bo);
	write_unlock(&bdev->vm_lock);
	bo->addr_space_offset = ((uint64_t) bo->vm_node->start) << PAGE_SHIFT;

	return 0;
out_unlock:
	write_unlock(&bdev->vm_lock);
	return ret;
}

int ttm_bo_wait(struct ttm_buffer_object *bo,
1691
		bool lazy, bool interruptible, bool no_wait)
1692 1693
{
	struct ttm_bo_driver *driver = bo->bdev->driver;
1694
	struct ttm_bo_device *bdev = bo->bdev;
1695 1696 1697
	void *sync_obj;
	int ret = 0;

1698
	if (likely(bo->sync_obj == NULL))
1699 1700
		return 0;

1701
	while (bo->sync_obj) {
1702

1703
		if (driver->sync_obj_signaled(bo->sync_obj)) {
1704 1705 1706 1707 1708 1709
			void *tmp_obj = bo->sync_obj;
			bo->sync_obj = NULL;
			clear_bit(TTM_BO_PRIV_FLAG_MOVING, &bo->priv_flags);
			spin_unlock(&bdev->fence_lock);
			driver->sync_obj_unref(&tmp_obj);
			spin_lock(&bdev->fence_lock);
1710 1711 1712 1713 1714 1715
			continue;
		}

		if (no_wait)
			return -EBUSY;

1716
		sync_obj = driver->sync_obj_ref(bo->sync_obj);
1717
		spin_unlock(&bdev->fence_lock);
1718
		ret = driver->sync_obj_wait(sync_obj,
1719 1720 1721
					    lazy, interruptible);
		if (unlikely(ret != 0)) {
			driver->sync_obj_unref(&sync_obj);
1722
			spin_lock(&bdev->fence_lock);
1723 1724
			return ret;
		}
1725
		spin_lock(&bdev->fence_lock);
1726
		if (likely(bo->sync_obj == sync_obj)) {
1727 1728 1729 1730 1731 1732 1733 1734
			void *tmp_obj = bo->sync_obj;
			bo->sync_obj = NULL;
			clear_bit(TTM_BO_PRIV_FLAG_MOVING,
				  &bo->priv_flags);
			spin_unlock(&bdev->fence_lock);
			driver->sync_obj_unref(&sync_obj);
			driver->sync_obj_unref(&tmp_obj);
			spin_lock(&bdev->fence_lock);
1735
		} else {
1736
			spin_unlock(&bdev->fence_lock);
1737
			driver->sync_obj_unref(&sync_obj);
1738
			spin_lock(&bdev->fence_lock);
1739 1740 1741 1742 1743 1744 1745 1746
		}
	}
	return 0;
}
EXPORT_SYMBOL(ttm_bo_wait);

int ttm_bo_synccpu_write_grab(struct ttm_buffer_object *bo, bool no_wait)
{
1747
	struct ttm_bo_device *bdev = bo->bdev;
1748 1749 1750
	int ret = 0;

	/*
1751
	 * Using ttm_bo_reserve makes sure the lru lists are updated.
1752 1753 1754 1755 1756
	 */

	ret = ttm_bo_reserve(bo, true, no_wait, false, 0);
	if (unlikely(ret != 0))
		return ret;
1757
	spin_lock(&bdev->fence_lock);
1758
	ret = ttm_bo_wait(bo, false, true, no_wait);
1759
	spin_unlock(&bdev->fence_lock);
1760 1761 1762 1763 1764
	if (likely(ret == 0))
		atomic_inc(&bo->cpu_writers);
	ttm_bo_unreserve(bo);
	return ret;
}
1765
EXPORT_SYMBOL(ttm_bo_synccpu_write_grab);
1766 1767 1768

void ttm_bo_synccpu_write_release(struct ttm_buffer_object *bo)
{
1769
	atomic_dec(&bo->cpu_writers);
1770
}
1771
EXPORT_SYMBOL(ttm_bo_synccpu_write_release);
1772 1773 1774 1775 1776 1777 1778 1779

/**
 * A buffer object shrink method that tries to swap out the first
 * buffer object on the bo_global::swap_lru list.
 */

static int ttm_bo_swapout(struct ttm_mem_shrink *shrink)
{
1780 1781
	struct ttm_bo_global *glob =
	    container_of(shrink, struct ttm_bo_global, shrink);
1782 1783 1784 1785 1786
	struct ttm_buffer_object *bo;
	int ret = -EBUSY;
	int put_count;
	uint32_t swap_placement = (TTM_PL_FLAG_CACHED | TTM_PL_FLAG_SYSTEM);

1787
	spin_lock(&glob->lru_lock);
1788
	while (ret == -EBUSY) {
1789 1790
		if (unlikely(list_empty(&glob->swap_lru))) {
			spin_unlock(&glob->lru_lock);
1791 1792 1793
			return -EBUSY;
		}

1794
		bo = list_first_entry(&glob->swap_lru,
1795 1796 1797
				      struct ttm_buffer_object, swap);
		kref_get(&bo->list_kref);

1798 1799 1800 1801
		if (!list_empty(&bo->ddestroy)) {
			spin_unlock(&glob->lru_lock);
			(void) ttm_bo_cleanup_refs(bo, false, false, false);
			kref_put(&bo->list_kref, ttm_bo_release_list);
1802
			spin_lock(&glob->lru_lock);
1803 1804 1805
			continue;
		}

1806 1807 1808 1809 1810 1811 1812 1813
		/**
		 * Reserve buffer. Since we unlock while sleeping, we need
		 * to re-check that nobody removed us from the swap-list while
		 * we slept.
		 */

		ret = ttm_bo_reserve_locked(bo, false, true, false, 0);
		if (unlikely(ret == -EBUSY)) {
1814
			spin_unlock(&glob->lru_lock);
1815 1816
			ttm_bo_wait_unreserved(bo, false);
			kref_put(&bo->list_kref, ttm_bo_release_list);
1817
			spin_lock(&glob->lru_lock);
1818 1819 1820 1821 1822
		}
	}

	BUG_ON(ret != 0);
	put_count = ttm_bo_del_from_lru(bo);
1823
	spin_unlock(&glob->lru_lock);
1824

1825
	ttm_bo_list_ref_sub(bo, put_count, true);
1826 1827 1828 1829 1830

	/**
	 * Wait for GPU, then move to system cached.
	 */

1831
	spin_lock(&bo->bdev->fence_lock);
1832
	ret = ttm_bo_wait(bo, false, false, false);
1833
	spin_unlock(&bo->bdev->fence_lock);
1834 1835 1836 1837 1838 1839 1840 1841 1842 1843 1844 1845 1846

	if (unlikely(ret != 0))
		goto out;

	if ((bo->mem.placement & swap_placement) != swap_placement) {
		struct ttm_mem_reg evict_mem;

		evict_mem = bo->mem;
		evict_mem.mm_node = NULL;
		evict_mem.placement = TTM_PL_FLAG_SYSTEM | TTM_PL_FLAG_CACHED;
		evict_mem.mem_type = TTM_PL_SYSTEM;

		ret = ttm_bo_handle_move_mem(bo, &evict_mem, true,
1847
					     false, false, false);
1848 1849 1850 1851 1852 1853 1854 1855 1856 1857 1858
		if (unlikely(ret != 0))
			goto out;
	}

	ttm_bo_unmap_virtual(bo);

	/**
	 * Swap out. Buffer will be swapped in again as soon as
	 * anyone tries to access a ttm page.
	 */

1859 1860 1861
	if (bo->bdev->driver->swap_notify)
		bo->bdev->driver->swap_notify(bo);

J
Jan Engelhardt 已提交
1862
	ret = ttm_tt_swapout(bo->ttm, bo->persistent_swap_storage);
1863 1864 1865 1866 1867 1868 1869 1870 1871 1872 1873 1874 1875 1876 1877 1878
out:

	/**
	 *
	 * Unreserve without putting on LRU to avoid swapping out an
	 * already swapped buffer.
	 */

	atomic_set(&bo->reserved, 0);
	wake_up_all(&bo->event_queue);
	kref_put(&bo->list_kref, ttm_bo_release_list);
	return ret;
}

void ttm_bo_swapout_all(struct ttm_bo_device *bdev)
{
1879
	while (ttm_bo_swapout(&bdev->glob->shrink) == 0)
1880 1881
		;
}
1882
EXPORT_SYMBOL(ttm_bo_swapout_all);