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
	wake_up_all(&bo->event_queue);
492 493

	/*
494 495 496 497 498
	 * Since the final reference to this bo may not be dropped by
	 * the current task we have to put a memory barrier here to make
	 * sure the changes done in this function are always visible.
	 *
	 * This function only needs protection against the final kref_put.
499
	 */
500
	smp_mb__before_atomic_dec();
501 502
}

503
static void ttm_bo_cleanup_refs_or_queue(struct ttm_buffer_object *bo)
504 505
{
	struct ttm_bo_device *bdev = bo->bdev;
506
	struct ttm_bo_global *glob = bo->glob;
507
	struct ttm_bo_driver *driver = bdev->driver;
508
	void *sync_obj = NULL;
509
	int put_count;
510 511
	int ret;

512 513 514
	spin_lock(&glob->lru_lock);
	ret = ttm_bo_reserve_locked(bo, false, true, false, 0);

515
	spin_lock(&bdev->fence_lock);
516
	(void) ttm_bo_wait(bo, false, false, true);
517
	if (!ret && !bo->sync_obj) {
518
		spin_unlock(&bdev->fence_lock);
519
		put_count = ttm_bo_del_from_lru(bo);
520

521
		spin_unlock(&glob->lru_lock);
522
		ttm_bo_cleanup_memtype_use(bo);
523

524
		ttm_bo_list_ref_sub(bo, put_count, true);
525

526
		return;
527
	}
528 529
	if (bo->sync_obj)
		sync_obj = driver->sync_obj_ref(bo->sync_obj);
530 531 532 533 534 535
	spin_unlock(&bdev->fence_lock);

	if (!ret) {
		atomic_set(&bo->reserved, 0);
		wake_up_all(&bo->event_queue);
	}
536 537 538 539 540

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

541
	if (sync_obj) {
542
		driver->sync_obj_flush(sync_obj);
543 544
		driver->sync_obj_unref(&sync_obj);
	}
545 546 547 548 549
	schedule_delayed_work(&bdev->wq,
			      ((HZ / 100) < 1) ? 1 : HZ / 100);
}

/**
550
 * function ttm_bo_cleanup_refs_and_unlock
551 552 553
 * If bo idle, remove from delayed- and lru lists, and unref.
 * If not idle, do nothing.
 *
554 555 556
 * Must be called with lru_lock and reservation held, this function
 * will drop both before returning.
 *
557 558 559 560
 * @interruptible         Any sleeps should occur interruptibly.
 * @no_wait_gpu           Never wait for gpu. Return -EBUSY instead.
 */

561 562 563
static int ttm_bo_cleanup_refs_and_unlock(struct ttm_buffer_object *bo,
					  bool interruptible,
					  bool no_wait_gpu)
564
{
565
	struct ttm_bo_device *bdev = bo->bdev;
566
	struct ttm_bo_driver *driver = bdev->driver;
567 568
	struct ttm_bo_global *glob = bo->glob;
	int put_count;
569
	int ret;
570

571
	spin_lock(&bdev->fence_lock);
572
	ret = ttm_bo_wait(bo, false, false, true);
573

574 575
	if (ret && !no_wait_gpu) {
		void *sync_obj;
576

577 578 579 580 581 582 583
		/*
		 * Take a reference to the fence and unreserve,
		 * at this point the buffer should be dead, so
		 * no new sync objects can be attached.
		 */
		sync_obj = driver->sync_obj_ref(&bo->sync_obj);
		spin_unlock(&bdev->fence_lock);
584

585 586
		atomic_set(&bo->reserved, 0);
		wake_up_all(&bo->event_queue);
587 588
		spin_unlock(&glob->lru_lock);

589 590 591
		ret = driver->sync_obj_wait(sync_obj, false, interruptible);
		driver->sync_obj_unref(&sync_obj);
		if (ret)
592 593
			return ret;

594 595 596 597 598 599 600 601 602 603
		/*
		 * remove sync_obj with ttm_bo_wait, the wait should be
		 * finished, and no new wait object should have been added.
		 */
		spin_lock(&bdev->fence_lock);
		ret = ttm_bo_wait(bo, false, false, true);
		WARN_ON(ret);
		spin_unlock(&bdev->fence_lock);
		if (ret)
			return ret;
604

605 606
		spin_lock(&glob->lru_lock);
		ret = ttm_bo_reserve_locked(bo, false, true, false, 0);
607

608 609 610 611 612 613 614 615 616 617 618 619 620 621
		/*
		 * We raced, and lost, someone else holds the reservation now,
		 * and is probably busy in ttm_bo_cleanup_memtype_use.
		 *
		 * Even if it's not the case, because we finished waiting any
		 * delayed destruction would succeed, so just return success
		 * here.
		 */
		if (ret) {
			spin_unlock(&glob->lru_lock);
			return 0;
		}
	} else
		spin_unlock(&bdev->fence_lock);
622

623
	if (ret || unlikely(list_empty(&bo->ddestroy))) {
624 625
		atomic_set(&bo->reserved, 0);
		wake_up_all(&bo->event_queue);
626
		spin_unlock(&glob->lru_lock);
627
		return ret;
628 629
	}

630 631 632 633 634 635 636
	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);

637
	ttm_bo_list_ref_sub(bo, put_count, true);
638 639

	return 0;
640 641 642 643 644 645 646 647 648
}

/**
 * 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)
{
649
	struct ttm_bo_global *glob = bdev->glob;
650 651
	struct ttm_buffer_object *entry = NULL;
	int ret = 0;
652

653
	spin_lock(&glob->lru_lock);
654 655 656 657 658 659 660 661 662 663 664 665 666
	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);
667 668 669
			kref_get(&nentry->list_kref);
		}

670 671 672 673 674 675 676
		ret = ttm_bo_reserve_locked(entry, false, !remove_all, false, 0);
		if (!ret)
			ret = ttm_bo_cleanup_refs_and_unlock(entry, false,
							     !remove_all);
		else
			spin_unlock(&glob->lru_lock);

677
		kref_put(&entry->list_kref, ttm_bo_release_list);
678 679 680 681
		entry = nentry;

		if (ret || !entry)
			goto out;
682

683
		spin_lock(&glob->lru_lock);
684
		if (list_empty(&entry->ddestroy))
685 686 687
			break;
	}

688 689 690 691 692
out_unlock:
	spin_unlock(&glob->lru_lock);
out:
	if (entry)
		kref_put(&entry->list_kref, ttm_bo_release_list);
693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711
	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;
712
	struct ttm_mem_type_manager *man = &bdev->man[bo->mem.mem_type];
713

714
	write_lock(&bdev->vm_lock);
715 716 717 718 719 720
	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);
721 722 723
	ttm_mem_io_lock(man, false);
	ttm_mem_io_free_vm(bo);
	ttm_mem_io_unlock(man);
724
	ttm_bo_cleanup_refs_or_queue(bo);
725 726 727 728 729 730 731 732 733 734 735 736
	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);

737 738 739 740 741 742 743 744 745 746 747 748 749 750
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);

751
static int ttm_bo_evict(struct ttm_buffer_object *bo, bool interruptible,
752
			bool no_wait_reserve, bool no_wait_gpu)
753 754 755
{
	struct ttm_bo_device *bdev = bo->bdev;
	struct ttm_mem_reg evict_mem;
756 757
	struct ttm_placement placement;
	int ret = 0;
758

759
	spin_lock(&bdev->fence_lock);
760
	ret = ttm_bo_wait(bo, false, interruptible, no_wait_gpu);
761
	spin_unlock(&bdev->fence_lock);
762

763
	if (unlikely(ret != 0)) {
764
		if (ret != -ERESTARTSYS) {
J
Joe Perches 已提交
765
			pr_err("Failed to expire sync object before buffer eviction\n");
766
		}
767 768 769
		goto out;
	}

770
	BUG_ON(!ttm_bo_is_reserved(bo));
771 772 773

	evict_mem = bo->mem;
	evict_mem.mm_node = NULL;
774 775
	evict_mem.bus.io_reserved_vm = false;
	evict_mem.bus.io_reserved_count = 0;
776

777 778 779 780
	placement.fpfn = 0;
	placement.lpfn = 0;
	placement.num_placement = 0;
	placement.num_busy_placement = 0;
781 782
	bdev->driver->evict_flags(bo, &placement);
	ret = ttm_bo_mem_space(bo, &placement, &evict_mem, interruptible,
783
				no_wait_reserve, no_wait_gpu);
784
	if (ret) {
785
		if (ret != -ERESTARTSYS) {
J
Joe Perches 已提交
786 787
			pr_err("Failed to find memory space for buffer 0x%p eviction\n",
			       bo);
788 789
			ttm_bo_mem_space_debug(bo, &placement);
		}
790 791 792 793
		goto out;
	}

	ret = ttm_bo_handle_move_mem(bo, &evict_mem, true, interruptible,
794
				     no_wait_reserve, no_wait_gpu);
795
	if (ret) {
796
		if (ret != -ERESTARTSYS)
J
Joe Perches 已提交
797
			pr_err("Buffer eviction failed\n");
798
		ttm_bo_mem_put(bo, &evict_mem);
799 800
		goto out;
	}
801 802 803 804 805 806 807
	bo->evicted = true;
out:
	return ret;
}

static int ttm_mem_evict_first(struct ttm_bo_device *bdev,
				uint32_t mem_type,
808 809
				bool interruptible, bool no_wait_reserve,
				bool no_wait_gpu)
810 811 812 813
{
	struct ttm_bo_global *glob = bdev->glob;
	struct ttm_mem_type_manager *man = &bdev->man[mem_type];
	struct ttm_buffer_object *bo;
814
	int ret = -EBUSY, put_count;
815

816
	spin_lock(&glob->lru_lock);
817 818 819 820 821 822 823
	list_for_each_entry(bo, &man->lru, lru) {
		ret = ttm_bo_reserve_locked(bo, false, true, false, 0);
		if (!ret)
			break;
	}

	if (ret) {
824
		spin_unlock(&glob->lru_lock);
825
		return ret;
826 827
	}

828
	kref_get(&bo->list_kref);
829

830
	if (!list_empty(&bo->ddestroy)) {
831 832
		ret = ttm_bo_cleanup_refs_and_unlock(bo, interruptible,
						     no_wait_gpu);
833
		kref_put(&bo->list_kref, ttm_bo_release_list);
834
		return ret;
835 836
	}

837
	put_count = ttm_bo_del_from_lru(bo);
838
	spin_unlock(&glob->lru_lock);
839 840 841

	BUG_ON(ret != 0);

842
	ttm_bo_list_ref_sub(bo, put_count, true);
843

844
	ret = ttm_bo_evict(bo, interruptible, no_wait_reserve, no_wait_gpu);
845
	ttm_bo_unreserve(bo);
846

847
	kref_put(&bo->list_kref, ttm_bo_release_list);
848 849 850
	return ret;
}

851 852
void ttm_bo_mem_put(struct ttm_buffer_object *bo, struct ttm_mem_reg *mem)
{
853
	struct ttm_mem_type_manager *man = &bo->bdev->man[mem->mem_type];
854

855 856
	if (mem->mm_node)
		(*man->func->put_node)(man, mem);
857 858 859
}
EXPORT_SYMBOL(ttm_bo_mem_put);

860 861 862 863
/**
 * 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.
 */
864 865 866 867
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,
868 869 870
					bool interruptible,
					bool no_wait_reserve,
					bool no_wait_gpu)
871
{
872
	struct ttm_bo_device *bdev = bo->bdev;
873 874 875 876
	struct ttm_mem_type_manager *man = &bdev->man[mem_type];
	int ret;

	do {
877
		ret = (*man->func->get_node)(man, bo, placement, mem);
878 879
		if (unlikely(ret != 0))
			return ret;
880
		if (mem->mm_node)
881
			break;
882
		ret = ttm_mem_evict_first(bdev, mem_type, interruptible,
883
						no_wait_reserve, no_wait_gpu);
884 885 886
		if (unlikely(ret != 0))
			return ret;
	} while (1);
887
	if (mem->mm_node == NULL)
888 889 890 891 892
		return -ENOMEM;
	mem->mem_type = mem_type;
	return 0;
}

893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917
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;
}

918 919
static bool ttm_bo_mt_compatible(struct ttm_mem_type_manager *man,
				 uint32_t mem_type,
920 921
				 uint32_t proposed_placement,
				 uint32_t *masked_placement)
922 923 924
{
	uint32_t cur_flags = ttm_bo_type_flags(mem_type);

925
	if ((cur_flags & proposed_placement & TTM_PL_MASK_MEM) == 0)
926 927
		return false;

928
	if ((proposed_placement & man->available_caching) == 0)
929 930
		return false;

931 932 933
	cur_flags |= (proposed_placement & man->available_caching);

	*masked_placement = cur_flags;
934 935 936 937 938 939 940 941 942 943 944 945
	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,
946 947
			struct ttm_placement *placement,
			struct ttm_mem_reg *mem,
948 949
			bool interruptible, bool no_wait_reserve,
			bool no_wait_gpu)
950 951 952 953 954 955 956
{
	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;
957
	bool has_erestartsys = false;
958
	int i, ret;
959 960

	mem->mm_node = NULL;
961
	for (i = 0; i < placement->num_placement; ++i) {
962 963 964 965
		ret = ttm_mem_type_from_flags(placement->placement[i],
						&mem_type);
		if (ret)
			return ret;
966 967 968
		man = &bdev->man[mem_type];

		type_ok = ttm_bo_mt_compatible(man,
969 970 971
						mem_type,
						placement->placement[i],
						&cur_flags);
972 973 974 975

		if (!type_ok)
			continue;

976 977
		cur_flags = ttm_bo_select_caching(man, bo->mem.placement,
						  cur_flags);
978 979 980 981 982 983
		/*
		 * 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);
984

985 986 987 988 989
		if (mem_type == TTM_PL_SYSTEM)
			break;

		if (man->has_type && man->use_type) {
			type_found = true;
990
			ret = (*man->func->get_node)(man, bo, placement, mem);
991 992
			if (unlikely(ret))
				return ret;
993
		}
994
		if (mem->mm_node)
995 996 997
			break;
	}

998
	if ((type_ok && (mem_type == TTM_PL_SYSTEM)) || mem->mm_node) {
999 1000 1001 1002 1003 1004 1005 1006
		mem->mem_type = mem_type;
		mem->placement = cur_flags;
		return 0;
	}

	if (!type_found)
		return -EINVAL;

1007 1008
	for (i = 0; i < placement->num_busy_placement; ++i) {
		ret = ttm_mem_type_from_flags(placement->busy_placement[i],
1009 1010 1011
						&mem_type);
		if (ret)
			return ret;
1012 1013 1014 1015
		man = &bdev->man[mem_type];
		if (!man->has_type)
			continue;
		if (!ttm_bo_mt_compatible(man,
1016
						mem_type,
1017
						placement->busy_placement[i],
1018
						&cur_flags))
1019 1020
			continue;

1021 1022
		cur_flags = ttm_bo_select_caching(man, bo->mem.placement,
						  cur_flags);
1023 1024 1025 1026
		/*
		 * Use the access and other non-mapping-related flag bits from
		 * the memory placement flags to the current flags
		 */
1027
		ttm_flag_masked(&cur_flags, placement->busy_placement[i],
1028
				~TTM_PL_MASK_MEMTYPE);
1029

1030 1031 1032 1033 1034 1035 1036 1037

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

1038
		ret = ttm_bo_mem_force_space(bo, mem_type, placement, mem,
1039
						interruptible, no_wait_reserve, no_wait_gpu);
1040 1041 1042 1043
		if (ret == 0 && mem->mm_node) {
			mem->placement = cur_flags;
			return 0;
		}
1044 1045
		if (ret == -ERESTARTSYS)
			has_erestartsys = true;
1046
	}
1047
	ret = (has_erestartsys) ? -ERESTARTSYS : -ENOMEM;
1048 1049 1050 1051 1052
	return ret;
}
EXPORT_SYMBOL(ttm_bo_mem_space);

int ttm_bo_move_buffer(struct ttm_buffer_object *bo,
1053
			struct ttm_placement *placement,
1054 1055
			bool interruptible, bool no_wait_reserve,
			bool no_wait_gpu)
1056 1057 1058
{
	int ret = 0;
	struct ttm_mem_reg mem;
1059
	struct ttm_bo_device *bdev = bo->bdev;
1060

1061
	BUG_ON(!ttm_bo_is_reserved(bo));
1062 1063 1064 1065 1066 1067

	/*
	 * FIXME: It's possible to pipeline buffer moves.
	 * Have the driver move function wait for idle when necessary,
	 * instead of doing it here.
	 */
1068
	spin_lock(&bdev->fence_lock);
1069
	ret = ttm_bo_wait(bo, false, interruptible, no_wait_gpu);
1070
	spin_unlock(&bdev->fence_lock);
1071 1072 1073 1074 1075
	if (ret)
		return ret;
	mem.num_pages = bo->num_pages;
	mem.size = mem.num_pages << PAGE_SHIFT;
	mem.page_alignment = bo->mem.page_alignment;
1076 1077
	mem.bus.io_reserved_vm = false;
	mem.bus.io_reserved_count = 0;
1078 1079 1080
	/*
	 * Determine where to move the buffer.
	 */
1081
	ret = ttm_bo_mem_space(bo, placement, &mem, interruptible, no_wait_reserve, no_wait_gpu);
1082 1083
	if (ret)
		goto out_unlock;
1084
	ret = ttm_bo_handle_move_mem(bo, &mem, false, interruptible, no_wait_reserve, no_wait_gpu);
1085
out_unlock:
1086 1087
	if (ret && mem.mm_node)
		ttm_bo_mem_put(bo, &mem);
1088 1089 1090
	return ret;
}

1091
static int ttm_bo_mem_compat(struct ttm_placement *placement,
1092 1093
			     struct ttm_mem_reg *mem)
{
1094
	int i;
1095

1096 1097 1098
	if (mem->mm_node && placement->lpfn != 0 &&
	    (mem->start < placement->fpfn ||
	     mem->start + mem->num_pages > placement->lpfn))
1099
		return -1;
1100 1101 1102 1103 1104 1105 1106 1107 1108

	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;
1109 1110
}

1111 1112
int ttm_bo_validate(struct ttm_buffer_object *bo,
			struct ttm_placement *placement,
1113 1114
			bool interruptible, bool no_wait_reserve,
			bool no_wait_gpu)
1115 1116 1117
{
	int ret;

1118
	BUG_ON(!ttm_bo_is_reserved(bo));
1119 1120 1121 1122 1123
	/* Check that range is valid */
	if (placement->lpfn || placement->fpfn)
		if (placement->fpfn > placement->lpfn ||
			(placement->lpfn - placement->fpfn) < bo->num_pages)
			return -EINVAL;
1124 1125 1126
	/*
	 * Check whether we need to move buffer.
	 */
1127 1128
	ret = ttm_bo_mem_compat(placement, &bo->mem);
	if (ret < 0) {
1129
		ret = ttm_bo_move_buffer(bo, placement, interruptible, no_wait_reserve, no_wait_gpu);
1130
		if (ret)
1131
			return ret;
1132 1133 1134 1135 1136 1137 1138
	} 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);
1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149
	}
	/*
	 * 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;
}
1150
EXPORT_SYMBOL(ttm_bo_validate);
1151

1152 1153
int ttm_bo_check_placement(struct ttm_buffer_object *bo,
				struct ttm_placement *placement)
1154
{
1155 1156
	BUG_ON((placement->fpfn || placement->lpfn) &&
	       (bo->mem.num_pages > (placement->lpfn - placement->fpfn)));
1157 1158 1159 1160

	return 0;
}

1161 1162 1163 1164 1165 1166 1167
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
1168
		struct file *persistent_swap_storage,
1169
		size_t acc_size,
1170
		struct sg_table *sg,
1171
		void (*destroy) (struct ttm_buffer_object *))
1172
{
1173
	int ret = 0;
1174
	unsigned long num_pages;
1175 1176 1177 1178
	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
1179
		pr_err("Out of kernel memory\n");
1180 1181 1182 1183 1184 1185
		if (destroy)
			(*destroy)(bo);
		else
			kfree(bo);
		return -ENOMEM;
	}
1186 1187 1188

	num_pages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
	if (num_pages == 0) {
J
Joe Perches 已提交
1189
		pr_err("Illegal buffer object size\n");
1190 1191 1192 1193
		if (destroy)
			(*destroy)(bo);
		else
			kfree(bo);
1194
		ttm_mem_global_free(mem_glob, acc_size);
1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206
		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);
1207
	INIT_LIST_HEAD(&bo->io_reserve_lru);
1208
	bo->bdev = bdev;
1209
	bo->glob = bdev->glob;
1210 1211
	bo->type = type;
	bo->num_pages = num_pages;
1212
	bo->mem.size = num_pages << PAGE_SHIFT;
1213 1214 1215 1216
	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;
1217 1218
	bo->mem.bus.io_reserved_vm = false;
	bo->mem.bus.io_reserved_count = 0;
1219 1220 1221
	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
1222
	bo->persistent_swap_storage = persistent_swap_storage;
1223
	bo->acc_size = acc_size;
1224
	bo->sg = sg;
1225
	atomic_inc(&bo->glob->bo_count);
1226

1227
	ret = ttm_bo_check_placement(bo, placement);
1228 1229 1230 1231 1232 1233 1234
	if (unlikely(ret != 0))
		goto out_err;

	/*
	 * For ttm_bo_type_device buffers, allocate
	 * address space from the device.
	 */
1235 1236
	if (bo->type == ttm_bo_type_device ||
	    bo->type == ttm_bo_type_sg) {
1237 1238 1239 1240 1241
		ret = ttm_bo_setup_vm(bo);
		if (ret)
			goto out_err;
	}

1242
	ret = ttm_bo_validate(bo, placement, interruptible, false, false);
1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254
	if (ret)
		goto out_err;

	ttm_bo_unreserve(bo);
	return 0;

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

	return ret;
}
1255
EXPORT_SYMBOL(ttm_bo_init);
1256

1257 1258 1259
size_t ttm_bo_acc_size(struct ttm_bo_device *bdev,
		       unsigned long bo_size,
		       unsigned struct_size)
1260
{
1261 1262
	unsigned npages = (PAGE_ALIGN(bo_size)) >> PAGE_SHIFT;
	size_t size = 0;
1263

1264 1265 1266 1267
	size += ttm_round_pot(struct_size);
	size += PAGE_ALIGN(npages * sizeof(void *));
	size += ttm_round_pot(sizeof(struct ttm_tt));
	return size;
1268
}
1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284
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);
1285

1286 1287 1288 1289 1290 1291
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
1292
			struct file *persistent_swap_storage,
1293
			struct ttm_buffer_object **p_bo)
1294 1295
{
	struct ttm_buffer_object *bo;
1296
	size_t acc_size;
1297
	int ret;
1298 1299

	bo = kzalloc(sizeof(*bo), GFP_KERNEL);
1300
	if (unlikely(bo == NULL))
1301 1302
		return -ENOMEM;

1303
	acc_size = ttm_bo_acc_size(bdev, size, sizeof(struct ttm_buffer_object));
1304
	ret = ttm_bo_init(bdev, bo, size, type, placement, page_alignment,
1305 1306
			  interruptible, persistent_swap_storage, acc_size,
			  NULL, NULL);
1307 1308 1309 1310 1311
	if (likely(ret == 0))
		*p_bo = bo;

	return ret;
}
T
Thomas Hellstrom 已提交
1312
EXPORT_SYMBOL(ttm_bo_create);
1313 1314

static int ttm_bo_force_list_clean(struct ttm_bo_device *bdev,
1315
					unsigned mem_type, bool allow_errors)
1316
{
1317
	struct ttm_mem_type_manager *man = &bdev->man[mem_type];
1318
	struct ttm_bo_global *glob = bdev->glob;
1319 1320 1321 1322 1323 1324
	int ret;

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

1325
	spin_lock(&glob->lru_lock);
1326
	while (!list_empty(&man->lru)) {
1327
		spin_unlock(&glob->lru_lock);
1328
		ret = ttm_mem_evict_first(bdev, mem_type, false, false, false);
1329 1330 1331 1332
		if (ret) {
			if (allow_errors) {
				return ret;
			} else {
J
Joe Perches 已提交
1333
				pr_err("Cleanup eviction failed\n");
1334 1335
			}
		}
1336
		spin_lock(&glob->lru_lock);
1337
	}
1338
	spin_unlock(&glob->lru_lock);
1339 1340 1341 1342 1343
	return 0;
}

int ttm_bo_clean_mm(struct ttm_bo_device *bdev, unsigned mem_type)
{
R
Roel Kluin 已提交
1344
	struct ttm_mem_type_manager *man;
1345 1346 1347
	int ret = -EINVAL;

	if (mem_type >= TTM_NUM_MEM_TYPES) {
J
Joe Perches 已提交
1348
		pr_err("Illegal memory type %d\n", mem_type);
1349 1350
		return ret;
	}
R
Roel Kluin 已提交
1351
	man = &bdev->man[mem_type];
1352 1353

	if (!man->has_type) {
J
Joe Perches 已提交
1354 1355
		pr_err("Trying to take down uninitialized memory manager type %u\n",
		       mem_type);
1356 1357 1358 1359 1360 1361 1362 1363
		return ret;
	}

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

	ret = 0;
	if (mem_type > 0) {
1364
		ttm_bo_force_list_clean(bdev, mem_type, false);
1365

1366
		ret = (*man->func->takedown)(man);
1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377
	}

	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
1378
		pr_err("Illegal memory manager memory type %u\n", mem_type);
1379 1380 1381 1382
		return -EINVAL;
	}

	if (!man->has_type) {
J
Joe Perches 已提交
1383
		pr_err("Memory type %u has not been initialized\n", mem_type);
1384 1385 1386
		return 0;
	}

1387
	return ttm_bo_force_list_clean(bdev, mem_type, true);
1388 1389 1390 1391
}
EXPORT_SYMBOL(ttm_bo_evict_mm);

int ttm_bo_init_mm(struct ttm_bo_device *bdev, unsigned type,
1392
			unsigned long p_size)
1393 1394 1395 1396
{
	int ret = -EINVAL;
	struct ttm_mem_type_manager *man;

1397
	BUG_ON(type >= TTM_NUM_MEM_TYPES);
1398
	man = &bdev->man[type];
1399
	BUG_ON(man->has_type);
1400 1401 1402 1403
	man->io_reserve_fastpath = true;
	man->use_io_reserve_lru = false;
	mutex_init(&man->io_reserve_mutex);
	INIT_LIST_HEAD(&man->io_reserve_lru);
1404 1405 1406 1407

	ret = bdev->driver->init_mem_type(bdev, type, man);
	if (ret)
		return ret;
1408
	man->bdev = bdev;
1409 1410 1411

	ret = 0;
	if (type != TTM_PL_SYSTEM) {
1412
		ret = (*man->func->init)(man, p_size);
1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425
		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);

1426 1427 1428 1429 1430 1431 1432 1433 1434 1435
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);
}

1436
void ttm_bo_global_release(struct drm_global_reference *ref)
1437 1438 1439 1440 1441 1442 1443 1444
{
	struct ttm_bo_global *glob = ref->object;

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

1445
int ttm_bo_global_init(struct drm_global_reference *ref)
1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467
{
	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
1468
		pr_err("Could not register buffer object swapout\n");
1469 1470 1471 1472 1473
		goto out_no_shrink;
	}

	atomic_set(&glob->bo_count, 0);

1474 1475
	ret = kobject_init_and_add(
		&glob->kobj, &ttm_bo_glob_kobj_type, ttm_get_kobj(), "buffer_objects");
1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487
	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);


1488 1489 1490 1491 1492
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;
1493
	struct ttm_bo_global *glob = bdev->glob;
1494 1495 1496 1497 1498 1499 1500

	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
1501 1502
				pr_err("DRM memory manager type %d is not clean\n",
				       i);
1503 1504 1505 1506 1507
			}
			man->has_type = false;
		}
	}

1508 1509 1510 1511
	mutex_lock(&glob->device_list_mutex);
	list_del(&bdev->device_list);
	mutex_unlock(&glob->device_list_mutex);

1512
	cancel_delayed_work_sync(&bdev->wq);
1513 1514 1515 1516

	while (ttm_bo_delayed_delete(bdev, true))
		;

1517
	spin_lock(&glob->lru_lock);
1518 1519 1520 1521 1522
	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");
1523
	spin_unlock(&glob->lru_lock);
1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534

	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,
1535 1536
		       struct ttm_bo_global *glob,
		       struct ttm_bo_driver *driver,
D
Dave Airlie 已提交
1537
		       uint64_t file_page_offset,
D
Dave Airlie 已提交
1538
		       bool need_dma32)
1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550
{
	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.
	 */
1551
	ret = ttm_bo_init_mm(bdev, TTM_PL_SYSTEM, 0);
1552
	if (unlikely(ret != 0))
1553
		goto out_no_sys;
1554 1555 1556 1557

	bdev->addr_space_rb = RB_ROOT;
	ret = drm_mm_init(&bdev->addr_space_mm, file_page_offset, 0x10000000);
	if (unlikely(ret != 0))
1558
		goto out_no_addr_mm;
1559 1560 1561 1562

	INIT_DELAYED_WORK(&bdev->wq, ttm_bo_delayed_workqueue);
	INIT_LIST_HEAD(&bdev->ddestroy);
	bdev->dev_mapping = NULL;
1563
	bdev->glob = glob;
D
Dave Airlie 已提交
1564
	bdev->need_dma32 = need_dma32;
1565
	bdev->val_seq = 0;
1566
	spin_lock_init(&bdev->fence_lock);
1567 1568 1569
	mutex_lock(&glob->device_list_mutex);
	list_add_tail(&bdev->device_list, &glob->device_list);
	mutex_unlock(&glob->device_list_mutex);
1570 1571

	return 0;
1572
out_no_addr_mm:
1573
	ttm_bo_clean_mm(bdev, 0);
1574
out_no_sys:
1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599
	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;
}

1600
void ttm_bo_unmap_virtual_locked(struct ttm_buffer_object *bo)
1601 1602 1603 1604 1605 1606 1607 1608
{
	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);
1609
	ttm_mem_io_free_vm(bo);
1610
}
1611 1612 1613 1614 1615 1616 1617 1618 1619

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);
1620
}
1621 1622


1623
EXPORT_SYMBOL(ttm_bo_unmap_virtual);
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 1691 1692 1693 1694 1695 1696 1697 1698

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,
1699
		bool lazy, bool interruptible, bool no_wait)
1700 1701
{
	struct ttm_bo_driver *driver = bo->bdev->driver;
1702
	struct ttm_bo_device *bdev = bo->bdev;
1703 1704 1705
	void *sync_obj;
	int ret = 0;

1706
	if (likely(bo->sync_obj == NULL))
1707 1708
		return 0;

1709
	while (bo->sync_obj) {
1710

1711
		if (driver->sync_obj_signaled(bo->sync_obj)) {
1712 1713 1714 1715 1716 1717
			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);
1718 1719 1720 1721 1722 1723
			continue;
		}

		if (no_wait)
			return -EBUSY;

1724
		sync_obj = driver->sync_obj_ref(bo->sync_obj);
1725
		spin_unlock(&bdev->fence_lock);
1726
		ret = driver->sync_obj_wait(sync_obj,
1727 1728 1729
					    lazy, interruptible);
		if (unlikely(ret != 0)) {
			driver->sync_obj_unref(&sync_obj);
1730
			spin_lock(&bdev->fence_lock);
1731 1732
			return ret;
		}
1733
		spin_lock(&bdev->fence_lock);
1734
		if (likely(bo->sync_obj == sync_obj)) {
1735 1736 1737 1738 1739 1740 1741 1742
			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);
1743
		} else {
1744
			spin_unlock(&bdev->fence_lock);
1745
			driver->sync_obj_unref(&sync_obj);
1746
			spin_lock(&bdev->fence_lock);
1747 1748 1749 1750 1751 1752 1753 1754
		}
	}
	return 0;
}
EXPORT_SYMBOL(ttm_bo_wait);

int ttm_bo_synccpu_write_grab(struct ttm_buffer_object *bo, bool no_wait)
{
1755
	struct ttm_bo_device *bdev = bo->bdev;
1756 1757 1758
	int ret = 0;

	/*
1759
	 * Using ttm_bo_reserve makes sure the lru lists are updated.
1760 1761 1762 1763 1764
	 */

	ret = ttm_bo_reserve(bo, true, no_wait, false, 0);
	if (unlikely(ret != 0))
		return ret;
1765
	spin_lock(&bdev->fence_lock);
1766
	ret = ttm_bo_wait(bo, false, true, no_wait);
1767
	spin_unlock(&bdev->fence_lock);
1768 1769 1770 1771 1772
	if (likely(ret == 0))
		atomic_inc(&bo->cpu_writers);
	ttm_bo_unreserve(bo);
	return ret;
}
1773
EXPORT_SYMBOL(ttm_bo_synccpu_write_grab);
1774 1775 1776

void ttm_bo_synccpu_write_release(struct ttm_buffer_object *bo)
{
1777
	atomic_dec(&bo->cpu_writers);
1778
}
1779
EXPORT_SYMBOL(ttm_bo_synccpu_write_release);
1780 1781 1782 1783 1784 1785 1786 1787

/**
 * 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)
{
1788 1789
	struct ttm_bo_global *glob =
	    container_of(shrink, struct ttm_bo_global, shrink);
1790 1791 1792 1793 1794
	struct ttm_buffer_object *bo;
	int ret = -EBUSY;
	int put_count;
	uint32_t swap_placement = (TTM_PL_FLAG_CACHED | TTM_PL_FLAG_SYSTEM);

1795
	spin_lock(&glob->lru_lock);
1796 1797 1798 1799 1800
	list_for_each_entry(bo, &glob->swap_lru, swap) {
		ret = ttm_bo_reserve_locked(bo, false, true, false, 0);
		if (!ret)
			break;
	}
1801

1802 1803 1804 1805
	if (ret) {
		spin_unlock(&glob->lru_lock);
		return ret;
	}
1806

1807
	kref_get(&bo->list_kref);
1808

1809 1810 1811 1812
	if (!list_empty(&bo->ddestroy)) {
		ret = ttm_bo_cleanup_refs_and_unlock(bo, false, false);
		kref_put(&bo->list_kref, ttm_bo_release_list);
		return ret;
1813 1814 1815
	}

	put_count = ttm_bo_del_from_lru(bo);
1816
	spin_unlock(&glob->lru_lock);
1817

1818
	ttm_bo_list_ref_sub(bo, put_count, true);
1819 1820 1821 1822 1823

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

1824
	spin_lock(&bo->bdev->fence_lock);
1825
	ret = ttm_bo_wait(bo, false, false, false);
1826
	spin_unlock(&bo->bdev->fence_lock);
1827 1828 1829 1830 1831 1832 1833 1834 1835 1836 1837 1838 1839

	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,
1840
					     false, false, false);
1841 1842 1843 1844 1845 1846 1847 1848 1849 1850 1851
		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.
	 */

1852 1853 1854
	if (bo->bdev->driver->swap_notify)
		bo->bdev->driver->swap_notify(bo);

J
Jan Engelhardt 已提交
1855
	ret = ttm_tt_swapout(bo->ttm, bo->persistent_swap_storage);
1856 1857 1858 1859 1860 1861 1862 1863 1864 1865 1866 1867 1868 1869 1870 1871
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)
{
1872
	while (ttm_bo_swapout(&bdev->glob->shrink) == 0)
1873 1874
		;
}
1875
EXPORT_SYMBOL(ttm_bo_swapout_all);