i915_gem_userptr.c 21.6 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
/*
 * Copyright © 2012-2014 Intel Corporation
 *
 * 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, sublicense,
 * 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 NONINFRINGEMENT.  IN NO EVENT SHALL
 * THE AUTHORS OR COPYRIGHT HOLDERS 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.
 *
 */

25 26
#include <drm/drmP.h>
#include <drm/i915_drm.h>
27 28 29 30 31 32 33 34
#include "i915_drv.h"
#include "i915_trace.h"
#include "intel_drv.h"
#include <linux/mmu_context.h>
#include <linux/mmu_notifier.h>
#include <linux/mempolicy.h>
#include <linux/swap.h>

35 36 37 38 39 40 41 42 43
struct i915_mm_struct {
	struct mm_struct *mm;
	struct drm_device *dev;
	struct i915_mmu_notifier *mn;
	struct hlist_node node;
	struct kref kref;
	struct work_struct work;
};

44 45 46 47 48 49 50 51 52 53 54
#if defined(CONFIG_MMU_NOTIFIER)
#include <linux/interval_tree.h>

struct i915_mmu_notifier {
	spinlock_t lock;
	struct hlist_node node;
	struct mmu_notifier mn;
	struct rb_root objects;
};

struct i915_mmu_object {
55
	struct i915_mmu_notifier *mn;
56
	struct drm_i915_gem_object *obj;
57
	struct interval_tree_node it;
58
	struct list_head link;
59
	struct work_struct work;
60
	bool attached;
61 62
};

63
static void cancel_userptr(struct work_struct *work)
64
{
65 66
	struct i915_mmu_object *mo = container_of(work, typeof(*mo), work);
	struct drm_i915_gem_object *obj = mo->obj;
67 68 69 70 71 72 73 74 75 76 77 78 79 80
	struct drm_device *dev = obj->base.dev;

	mutex_lock(&dev->struct_mutex);
	/* Cancel any active worker and force us to re-evaluate gup */
	obj->userptr.work = NULL;

	if (obj->pages != NULL) {
		struct drm_i915_private *dev_priv = to_i915(dev);
		struct i915_vma *vma, *tmp;
		bool was_interruptible;

		was_interruptible = dev_priv->mm.interruptible;
		dev_priv->mm.interruptible = false;

81
		list_for_each_entry_safe(vma, tmp, &obj->vma_list, obj_link) {
82 83 84 85 86 87 88 89 90 91 92 93
			int ret = i915_vma_unbind(vma);
			WARN_ON(ret && ret != -EIO);
		}
		WARN_ON(i915_gem_object_put_pages(obj));

		dev_priv->mm.interruptible = was_interruptible;
	}

	drm_gem_object_unreference(&obj->base);
	mutex_unlock(&dev->struct_mutex);
}

94
static void add_object(struct i915_mmu_object *mo)
95
{
96 97
	if (mo->attached)
		return;
98

99 100 101 102 103 104 105 106 107 108 109
	interval_tree_insert(&mo->it, &mo->mn->objects);
	mo->attached = true;
}

static void del_object(struct i915_mmu_object *mo)
{
	if (!mo->attached)
		return;

	interval_tree_remove(&mo->it, &mo->mn->objects);
	mo->attached = false;
110 111
}

112 113 114 115 116
static void i915_gem_userptr_mn_invalidate_range_start(struct mmu_notifier *_mn,
						       struct mm_struct *mm,
						       unsigned long start,
						       unsigned long end)
{
117 118 119
	struct i915_mmu_notifier *mn =
		container_of(_mn, struct i915_mmu_notifier, mn);
	struct i915_mmu_object *mo;
120 121 122 123 124
	struct interval_tree_node *it;
	LIST_HEAD(cancelled);

	if (RB_EMPTY_ROOT(&mn->objects))
		return;
125 126 127 128 129

	/* interval ranges are inclusive, but invalidate range is exclusive */
	end--;

	spin_lock(&mn->lock);
130 131 132 133 134 135 136 137 138 139 140 141 142 143
	it = interval_tree_iter_first(&mn->objects, start, end);
	while (it) {
		/* The mmu_object is released late when destroying the
		 * GEM object so it is entirely possible to gain a
		 * reference on an object in the process of being freed
		 * since our serialisation is via the spinlock and not
		 * the struct_mutex - and consequently use it after it
		 * is freed and then double free it. To prevent that
		 * use-after-free we only acquire a reference on the
		 * object if it is not in the process of being destroyed.
		 */
		mo = container_of(it, struct i915_mmu_object, it);
		if (kref_get_unless_zero(&mo->obj->base.refcount))
			schedule_work(&mo->work);
144

145 146
		list_add(&mo->link, &cancelled);
		it = interval_tree_iter_next(it, start, end);
147
	}
148 149
	list_for_each_entry(mo, &cancelled, link)
		del_object(mo);
150
	spin_unlock(&mn->lock);
151 152 153 154 155 156 157
}

static const struct mmu_notifier_ops i915_gem_userptr_notifier = {
	.invalidate_range_start = i915_gem_userptr_mn_invalidate_range_start,
};

static struct i915_mmu_notifier *
158
i915_mmu_notifier_create(struct mm_struct *mm)
159
{
160
	struct i915_mmu_notifier *mn;
161 162
	int ret;

163 164
	mn = kmalloc(sizeof(*mn), GFP_KERNEL);
	if (mn == NULL)
165 166
		return ERR_PTR(-ENOMEM);

167 168 169 170 171 172
	spin_lock_init(&mn->lock);
	mn->mn.ops = &i915_gem_userptr_notifier;
	mn->objects = RB_ROOT;

	 /* Protected by mmap_sem (write-lock) */
	ret = __mmu_notifier_register(&mn->mn, mm);
173
	if (ret) {
174
		kfree(mn);
175 176 177
		return ERR_PTR(ret);
	}

178
	return mn;
179 180 181 182 183
}

static void
i915_gem_userptr_release__mmu_notifier(struct drm_i915_gem_object *obj)
{
184
	struct i915_mmu_object *mo;
185

186 187
	mo = obj->userptr.mmu_object;
	if (mo == NULL)
188 189
		return;

190 191 192
	spin_lock(&mo->mn->lock);
	del_object(mo);
	spin_unlock(&mo->mn->lock);
193 194 195 196 197 198 199 200
	kfree(mo);

	obj->userptr.mmu_object = NULL;
}

static struct i915_mmu_notifier *
i915_mmu_notifier_find(struct i915_mm_struct *mm)
{
201 202 203 204 205 206 207 208 209 210 211 212
	struct i915_mmu_notifier *mn = mm->mn;

	mn = mm->mn;
	if (mn)
		return mn;

	down_write(&mm->mm->mmap_sem);
	mutex_lock(&to_i915(mm->dev)->mm_lock);
	if ((mn = mm->mn) == NULL) {
		mn = i915_mmu_notifier_create(mm->mm);
		if (!IS_ERR(mn))
			mm->mn = mn;
213
	}
214 215 216 217
	mutex_unlock(&to_i915(mm->dev)->mm_lock);
	up_write(&mm->mm->mmap_sem);

	return mn;
218 219 220 221 222 223
}

static int
i915_gem_userptr_init__mmu_notifier(struct drm_i915_gem_object *obj,
				    unsigned flags)
{
224 225
	struct i915_mmu_notifier *mn;
	struct i915_mmu_object *mo;
226 227 228 229

	if (flags & I915_USERPTR_UNSYNCHRONIZED)
		return capable(CAP_SYS_ADMIN) ? 0 : -EPERM;

230 231
	if (WARN_ON(obj->userptr.mm == NULL))
		return -EINVAL;
232

233 234 235
	mn = i915_mmu_notifier_find(obj->userptr.mm);
	if (IS_ERR(mn))
		return PTR_ERR(mn);
236

237 238 239
	mo = kzalloc(sizeof(*mo), GFP_KERNEL);
	if (mo == NULL)
		return -ENOMEM;
240

241 242
	mo->mn = mn;
	mo->obj = obj;
243 244 245
	mo->it.start = obj->userptr.ptr;
	mo->it.last = obj->userptr.ptr + obj->base.size - 1;
	INIT_WORK(&mo->work, cancel_userptr);
246 247

	obj->userptr.mmu_object = mo;
248
	return 0;
249 250 251 252 253 254 255 256
}

static void
i915_mmu_notifier_free(struct i915_mmu_notifier *mn,
		       struct mm_struct *mm)
{
	if (mn == NULL)
		return;
257

258
	mmu_notifier_unregister(&mn->mn, mm);
259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280
	kfree(mn);
}

#else

static void
i915_gem_userptr_release__mmu_notifier(struct drm_i915_gem_object *obj)
{
}

static int
i915_gem_userptr_init__mmu_notifier(struct drm_i915_gem_object *obj,
				    unsigned flags)
{
	if ((flags & I915_USERPTR_UNSYNCHRONIZED) == 0)
		return -ENODEV;

	if (!capable(CAP_SYS_ADMIN))
		return -EPERM;

	return 0;
}
281 282 283 284 285 286 287

static void
i915_mmu_notifier_free(struct i915_mmu_notifier *mn,
		       struct mm_struct *mm)
{
}

288 289
#endif

290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382
static struct i915_mm_struct *
__i915_mm_struct_find(struct drm_i915_private *dev_priv, struct mm_struct *real)
{
	struct i915_mm_struct *mm;

	/* Protected by dev_priv->mm_lock */
	hash_for_each_possible(dev_priv->mm_structs, mm, node, (unsigned long)real)
		if (mm->mm == real)
			return mm;

	return NULL;
}

static int
i915_gem_userptr_init__mm_struct(struct drm_i915_gem_object *obj)
{
	struct drm_i915_private *dev_priv = to_i915(obj->base.dev);
	struct i915_mm_struct *mm;
	int ret = 0;

	/* During release of the GEM object we hold the struct_mutex. This
	 * precludes us from calling mmput() at that time as that may be
	 * the last reference and so call exit_mmap(). exit_mmap() will
	 * attempt to reap the vma, and if we were holding a GTT mmap
	 * would then call drm_gem_vm_close() and attempt to reacquire
	 * the struct mutex. So in order to avoid that recursion, we have
	 * to defer releasing the mm reference until after we drop the
	 * struct_mutex, i.e. we need to schedule a worker to do the clean
	 * up.
	 */
	mutex_lock(&dev_priv->mm_lock);
	mm = __i915_mm_struct_find(dev_priv, current->mm);
	if (mm == NULL) {
		mm = kmalloc(sizeof(*mm), GFP_KERNEL);
		if (mm == NULL) {
			ret = -ENOMEM;
			goto out;
		}

		kref_init(&mm->kref);
		mm->dev = obj->base.dev;

		mm->mm = current->mm;
		atomic_inc(&current->mm->mm_count);

		mm->mn = NULL;

		/* Protected by dev_priv->mm_lock */
		hash_add(dev_priv->mm_structs,
			 &mm->node, (unsigned long)mm->mm);
	} else
		kref_get(&mm->kref);

	obj->userptr.mm = mm;
out:
	mutex_unlock(&dev_priv->mm_lock);
	return ret;
}

static void
__i915_mm_struct_free__worker(struct work_struct *work)
{
	struct i915_mm_struct *mm = container_of(work, typeof(*mm), work);
	i915_mmu_notifier_free(mm->mn, mm->mm);
	mmdrop(mm->mm);
	kfree(mm);
}

static void
__i915_mm_struct_free(struct kref *kref)
{
	struct i915_mm_struct *mm = container_of(kref, typeof(*mm), kref);

	/* Protected by dev_priv->mm_lock */
	hash_del(&mm->node);
	mutex_unlock(&to_i915(mm->dev)->mm_lock);

	INIT_WORK(&mm->work, __i915_mm_struct_free__worker);
	schedule_work(&mm->work);
}

static void
i915_gem_userptr_release__mm_struct(struct drm_i915_gem_object *obj)
{
	if (obj->userptr.mm == NULL)
		return;

	kref_put_mutex(&obj->userptr.mm->kref,
		       __i915_mm_struct_free,
		       &to_i915(obj->base.dev)->mm_lock);
	obj->userptr.mm = NULL;
}

383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427
struct get_pages_work {
	struct work_struct work;
	struct drm_i915_gem_object *obj;
	struct task_struct *task;
};

#if IS_ENABLED(CONFIG_SWIOTLB)
#define swiotlb_active() swiotlb_nr_tbl()
#else
#define swiotlb_active() 0
#endif

static int
st_set_pages(struct sg_table **st, struct page **pvec, int num_pages)
{
	struct scatterlist *sg;
	int ret, n;

	*st = kmalloc(sizeof(**st), GFP_KERNEL);
	if (*st == NULL)
		return -ENOMEM;

	if (swiotlb_active()) {
		ret = sg_alloc_table(*st, num_pages, GFP_KERNEL);
		if (ret)
			goto err;

		for_each_sg((*st)->sgl, sg, num_pages, n)
			sg_set_page(sg, pvec[n], PAGE_SIZE, 0);
	} else {
		ret = sg_alloc_table_from_pages(*st, pvec, num_pages,
						0, num_pages << PAGE_SHIFT,
						GFP_KERNEL);
		if (ret)
			goto err;
	}

	return 0;

err:
	kfree(*st);
	*st = NULL;
	return ret;
}

I
Imre Deak 已提交
428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447
static int
__i915_gem_userptr_set_pages(struct drm_i915_gem_object *obj,
			     struct page **pvec, int num_pages)
{
	int ret;

	ret = st_set_pages(&obj->pages, pvec, num_pages);
	if (ret)
		return ret;

	ret = i915_gem_gtt_prepare_object(obj);
	if (ret) {
		sg_free_table(obj->pages);
		kfree(obj->pages);
		obj->pages = NULL;
	}

	return ret;
}

448
static int
449 450 451
__i915_gem_userptr_set_active(struct drm_i915_gem_object *obj,
			      bool value)
{
452 453
	int ret = 0;

454 455 456 457 458 459 460 461 462 463 464 465
	/* During mm_invalidate_range we need to cancel any userptr that
	 * overlaps the range being invalidated. Doing so requires the
	 * struct_mutex, and that risks recursion. In order to cause
	 * recursion, the user must alias the userptr address space with
	 * a GTT mmapping (possible with a MAP_FIXED) - then when we have
	 * to invalidate that mmaping, mm_invalidate_range is called with
	 * the userptr address *and* the struct_mutex held.  To prevent that
	 * we set a flag under the i915_mmu_notifier spinlock to indicate
	 * whether this object is valid.
	 */
#if defined(CONFIG_MMU_NOTIFIER)
	if (obj->userptr.mmu_object == NULL)
466
		return 0;
467 468

	spin_lock(&obj->userptr.mmu_object->mn->lock);
469 470 471
	/* In order to serialise get_pages with an outstanding
	 * cancel_userptr, we must drop the struct_mutex and try again.
	 */
472 473 474 475
	if (!value)
		del_object(obj->userptr.mmu_object);
	else if (!work_pending(&obj->userptr.mmu_object->work))
		add_object(obj->userptr.mmu_object);
476 477
	else
		ret = -EAGAIN;
478 479
	spin_unlock(&obj->userptr.mmu_object->mn->lock);
#endif
480 481

	return ret;
482 483
}

484 485 486 487 488 489
static void
__i915_gem_userptr_get_pages_worker(struct work_struct *_work)
{
	struct get_pages_work *work = container_of(_work, typeof(*work), work);
	struct drm_i915_gem_object *obj = work->obj;
	struct drm_device *dev = obj->base.dev;
490
	const int npages = obj->base.size >> PAGE_SHIFT;
491 492 493 494 495 496
	struct page **pvec;
	int pinned, ret;

	ret = -ENOMEM;
	pinned = 0;

497
	pvec = kmalloc(npages*sizeof(struct page *),
498 499
		       GFP_TEMPORARY | __GFP_NOWARN | __GFP_NORETRY);
	if (pvec == NULL)
500
		pvec = drm_malloc_ab(npages, sizeof(struct page *));
501
	if (pvec != NULL) {
502
		struct mm_struct *mm = obj->userptr.mm->mm;
503 504

		down_read(&mm->mmap_sem);
505
		while (pinned < npages) {
506 507 508 509 510
			ret = get_user_pages_remote(work->task, mm,
					obj->userptr.ptr + pinned * PAGE_SIZE,
					npages - pinned,
					!obj->userptr.read_only, 0,
					pvec + pinned, NULL);
511 512 513 514 515 516 517 518 519
			if (ret < 0)
				break;

			pinned += ret;
		}
		up_read(&mm->mmap_sem);
	}

	mutex_lock(&dev->struct_mutex);
520 521 522 523 524 525 526 527 528 529
	if (obj->userptr.work == &work->work) {
		if (pinned == npages) {
			ret = __i915_gem_userptr_set_pages(obj, pvec, npages);
			if (ret == 0) {
				list_add_tail(&obj->global_list,
					      &to_i915(dev)->mm.unbound_list);
				obj->get_page.sg = obj->pages->sgl;
				obj->get_page.last = 0;
				pinned = 0;
			}
530
		}
531
		obj->userptr.work = ERR_PTR(ret);
532 533
		if (ret)
			__i915_gem_userptr_set_active(obj, false);
534 535 536 537 538 539 540 541 542 543 544 545 546
	}

	obj->userptr.workers--;
	drm_gem_object_unreference(&obj->base);
	mutex_unlock(&dev->struct_mutex);

	release_pages(pvec, pinned, 0);
	drm_free_large(pvec);

	put_task_struct(work->task);
	kfree(work);
}

547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594
static int
__i915_gem_userptr_get_pages_schedule(struct drm_i915_gem_object *obj,
				      bool *active)
{
	struct get_pages_work *work;

	/* Spawn a worker so that we can acquire the
	 * user pages without holding our mutex. Access
	 * to the user pages requires mmap_sem, and we have
	 * a strict lock ordering of mmap_sem, struct_mutex -
	 * we already hold struct_mutex here and so cannot
	 * call gup without encountering a lock inversion.
	 *
	 * Userspace will keep on repeating the operation
	 * (thanks to EAGAIN) until either we hit the fast
	 * path or the worker completes. If the worker is
	 * cancelled or superseded, the task is still run
	 * but the results ignored. (This leads to
	 * complications that we may have a stray object
	 * refcount that we need to be wary of when
	 * checking for existing objects during creation.)
	 * If the worker encounters an error, it reports
	 * that error back to this function through
	 * obj->userptr.work = ERR_PTR.
	 */
	if (obj->userptr.workers >= I915_GEM_USERPTR_MAX_WORKERS)
		return -EAGAIN;

	work = kmalloc(sizeof(*work), GFP_KERNEL);
	if (work == NULL)
		return -ENOMEM;

	obj->userptr.work = &work->work;
	obj->userptr.workers++;

	work->obj = obj;
	drm_gem_object_reference(&obj->base);

	work->task = current;
	get_task_struct(work->task);

	INIT_WORK(&work->work, __i915_gem_userptr_get_pages_worker);
	schedule_work(&work->work);

	*active = true;
	return -EAGAIN;
}

595 596 597 598 599 600
static int
i915_gem_userptr_get_pages(struct drm_i915_gem_object *obj)
{
	const int num_pages = obj->base.size >> PAGE_SHIFT;
	struct page **pvec;
	int pinned, ret;
601
	bool active;
602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618

	/* If userspace should engineer that these pages are replaced in
	 * the vma between us binding this page into the GTT and completion
	 * of rendering... Their loss. If they change the mapping of their
	 * pages they need to create a new bo to point to the new vma.
	 *
	 * However, that still leaves open the possibility of the vma
	 * being copied upon fork. Which falls under the same userspace
	 * synchronisation issue as a regular bo, except that this time
	 * the process may not be expecting that a particular piece of
	 * memory is tied to the GPU.
	 *
	 * Fortunately, we can hook into the mmu_notifier in order to
	 * discard the page references prior to anything nasty happening
	 * to the vma (discard or cloning) which should prevent the more
	 * egregious cases from causing harm.
	 */
619 620 621 622 623 624 625 626 627 628 629
	if (IS_ERR(obj->userptr.work)) {
		/* active flag will have been dropped already by the worker */
		ret = PTR_ERR(obj->userptr.work);
		obj->userptr.work = NULL;
		return ret;
	}
	if (obj->userptr.work)
		/* active flag should still be held for the pending work */
		return -EAGAIN;

	/* Let the mmu-notifier know that we have begun and need cancellation */
630 631 632
	ret = __i915_gem_userptr_set_active(obj, true);
	if (ret)
		return ret;
633 634 635

	pvec = NULL;
	pinned = 0;
636
	if (obj->userptr.mm->mm == current->mm) {
637 638 639 640
		pvec = kmalloc(num_pages*sizeof(struct page *),
			       GFP_TEMPORARY | __GFP_NOWARN | __GFP_NORETRY);
		if (pvec == NULL) {
			pvec = drm_malloc_ab(num_pages, sizeof(struct page *));
641 642
			if (pvec == NULL) {
				__i915_gem_userptr_set_active(obj, false);
643
				return -ENOMEM;
644
			}
645 646 647 648 649
		}

		pinned = __get_user_pages_fast(obj->userptr.ptr, num_pages,
					       !obj->userptr.read_only, pvec);
	}
650 651 652 653 654 655 656

	active = false;
	if (pinned < 0)
		ret = pinned, pinned = 0;
	else if (pinned < num_pages)
		ret = __i915_gem_userptr_get_pages_schedule(obj, &active);
	else
I
Imre Deak 已提交
657
		ret = __i915_gem_userptr_set_pages(obj, pvec, num_pages);
658 659 660
	if (ret) {
		__i915_gem_userptr_set_active(obj, active);
		release_pages(pvec, pinned, 0);
661 662 663 664 665 666 667 668
	}
	drm_free_large(pvec);
	return ret;
}

static void
i915_gem_userptr_put_pages(struct drm_i915_gem_object *obj)
{
669
	struct sg_page_iter sg_iter;
670 671

	BUG_ON(obj->userptr.work != NULL);
672
	__i915_gem_userptr_set_active(obj, false);
673 674 675 676

	if (obj->madv != I915_MADV_WILLNEED)
		obj->dirty = 0;

I
Imre Deak 已提交
677 678
	i915_gem_gtt_finish_object(obj);

679 680
	for_each_sg_page(obj->pages->sgl, &sg_iter, obj->pages->nents, 0) {
		struct page *page = sg_page_iter_page(&sg_iter);
681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697

		if (obj->dirty)
			set_page_dirty(page);

		mark_page_accessed(page);
		page_cache_release(page);
	}
	obj->dirty = 0;

	sg_free_table(obj->pages);
	kfree(obj->pages);
}

static void
i915_gem_userptr_release(struct drm_i915_gem_object *obj)
{
	i915_gem_userptr_release__mmu_notifier(obj);
698
	i915_gem_userptr_release__mm_struct(obj);
699 700 701 702 703
}

static int
i915_gem_userptr_dmabuf_export(struct drm_i915_gem_object *obj)
{
704
	if (obj->userptr.mmu_object)
705 706 707 708 709 710
		return 0;

	return i915_gem_userptr_init__mmu_notifier(obj, 0);
}

static const struct drm_i915_gem_object_ops i915_gem_userptr_ops = {
711
	.flags = I915_GEM_OBJECT_HAS_STRUCT_PAGE,
712 713
	.get_pages = i915_gem_userptr_get_pages,
	.put_pages = i915_gem_userptr_put_pages,
714
	.dmabuf_export = i915_gem_userptr_dmabuf_export,
715 716 717 718 719 720 721 722 723 724
	.release = i915_gem_userptr_release,
};

/**
 * Creates a new mm object that wraps some normal memory from the process
 * context - user memory.
 *
 * We impose several restrictions upon the memory being mapped
 * into the GPU.
 * 1. It must be page aligned (both start/end addresses, i.e ptr and size).
725
 * 2. It must be normal system memory, not a pointer into another map of IO
726
 *    space (e.g. it must not be a GTT mmapping of another object).
727
 * 3. We only allow a bo as large as we could in theory map into the GTT,
728
 *    that is we limit the size to the total size of the GTT.
729
 * 4. The bo is marked as being snoopable. The backing pages are left
730 731 732 733 734 735 736 737 738 739 740 741 742 743
 *    accessible directly by the CPU, but reads and writes by the GPU may
 *    incur the cost of a snoop (unless you have an LLC architecture).
 *
 * Synchronisation between multiple users and the GPU is left to userspace
 * through the normal set-domain-ioctl. The kernel will enforce that the
 * GPU relinquishes the VMA before it is returned back to the system
 * i.e. upon free(), munmap() or process termination. However, the userspace
 * malloc() library may not immediately relinquish the VMA after free() and
 * instead reuse it whilst the GPU is still reading and writing to the VMA.
 * Caveat emptor.
 *
 * Also note, that the object created here is not currently a "first class"
 * object, in that several ioctls are banned. These are the CPU access
 * ioctls: mmap(), pwrite and pread. In practice, you are expected to use
744 745 746 747
 * direct access via your pointer rather than use those ioctls. Another
 * restriction is that we do not allow userptr surfaces to be pinned to the
 * hardware and so we reject any attempt to create a framebuffer out of a
 * userptr.
748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795
 *
 * If you think this is a good interface to use to pass GPU memory between
 * drivers, please use dma-buf instead. In fact, wherever possible use
 * dma-buf instead.
 */
int
i915_gem_userptr_ioctl(struct drm_device *dev, void *data, struct drm_file *file)
{
	struct drm_i915_gem_userptr *args = data;
	struct drm_i915_gem_object *obj;
	int ret;
	u32 handle;

	if (args->flags & ~(I915_USERPTR_READ_ONLY |
			    I915_USERPTR_UNSYNCHRONIZED))
		return -EINVAL;

	if (offset_in_page(args->user_ptr | args->user_size))
		return -EINVAL;

	if (!access_ok(args->flags & I915_USERPTR_READ_ONLY ? VERIFY_READ : VERIFY_WRITE,
		       (char __user *)(unsigned long)args->user_ptr, args->user_size))
		return -EFAULT;

	if (args->flags & I915_USERPTR_READ_ONLY) {
		/* On almost all of the current hw, we cannot tell the GPU that a
		 * page is readonly, so this is just a placeholder in the uAPI.
		 */
		return -ENODEV;
	}

	obj = i915_gem_object_alloc(dev);
	if (obj == NULL)
		return -ENOMEM;

	drm_gem_private_object_init(dev, &obj->base, args->user_size);
	i915_gem_object_init(obj, &i915_gem_userptr_ops);
	obj->cache_level = I915_CACHE_LLC;
	obj->base.write_domain = I915_GEM_DOMAIN_CPU;
	obj->base.read_domains = I915_GEM_DOMAIN_CPU;

	obj->userptr.ptr = args->user_ptr;
	obj->userptr.read_only = !!(args->flags & I915_USERPTR_READ_ONLY);

	/* And keep a pointer to the current->mm for resolving the user pages
	 * at binding. This means that we need to hook into the mmu_notifier
	 * in order to detect if the mmu is destroyed.
	 */
796 797
	ret = i915_gem_userptr_init__mm_struct(obj);
	if (ret == 0)
798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814
		ret = i915_gem_userptr_init__mmu_notifier(obj, args->flags);
	if (ret == 0)
		ret = drm_gem_handle_create(file, &obj->base, &handle);

	/* drop reference from allocate - handle holds it now */
	drm_gem_object_unreference_unlocked(&obj->base);
	if (ret)
		return ret;

	args->handle = handle;
	return 0;
}

int
i915_gem_init_userptr(struct drm_device *dev)
{
	struct drm_i915_private *dev_priv = to_i915(dev);
815 816
	mutex_init(&dev_priv->mm_lock);
	hash_init(dev_priv->mm_structs);
817 818
	return 0;
}