i915_gem.c 122.1 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
/*
 * Copyright © 2008 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.
 *
 * Authors:
 *    Eric Anholt <eric@anholt.net>
 *
 */

28
#include <drm/drmP.h>
29
#include <drm/drm_vma_manager.h>
30
#include <drm/i915_drm.h>
31
#include "i915_drv.h"
C
Chris Wilson 已提交
32
#include "i915_trace.h"
33
#include "intel_drv.h"
34
#include <linux/shmem_fs.h>
35
#include <linux/slab.h>
36
#include <linux/swap.h>
J
Jesse Barnes 已提交
37
#include <linux/pci.h>
38
#include <linux/dma-buf.h>
39

40
static void i915_gem_object_flush_gtt_write_domain(struct drm_i915_gem_object *obj);
41 42
static void i915_gem_object_flush_cpu_write_domain(struct drm_i915_gem_object *obj,
						   bool force);
43 44 45 46 47 48
static __must_check int
i915_gem_object_bind_to_vm(struct drm_i915_gem_object *obj,
			   struct i915_address_space *vm,
			   unsigned alignment,
			   bool map_and_fenceable,
			   bool nonblocking);
49 50
static int i915_gem_phys_pwrite(struct drm_device *dev,
				struct drm_i915_gem_object *obj,
51
				struct drm_i915_gem_pwrite *args,
52
				struct drm_file *file);
53

54 55 56 57 58 59
static void i915_gem_write_fence(struct drm_device *dev, int reg,
				 struct drm_i915_gem_object *obj);
static void i915_gem_object_update_fence(struct drm_i915_gem_object *obj,
					 struct drm_i915_fence_reg *fence,
					 bool enable);

60
static int i915_gem_inactive_shrink(struct shrinker *shrinker,
61
				    struct shrink_control *sc);
C
Chris Wilson 已提交
62 63
static long i915_gem_purge(struct drm_i915_private *dev_priv, long target);
static void i915_gem_shrink_all(struct drm_i915_private *dev_priv);
64
static void i915_gem_object_truncate(struct drm_i915_gem_object *obj);
65

66 67 68 69 70 71
static bool cpu_cache_is_coherent(struct drm_device *dev,
				  enum i915_cache_level level)
{
	return HAS_LLC(dev) || level != I915_CACHE_NONE;
}

72 73 74 75 76 77 78 79
static bool cpu_write_needs_clflush(struct drm_i915_gem_object *obj)
{
	if (!cpu_cache_is_coherent(obj->base.dev, obj->cache_level))
		return true;

	return obj->pin_display;
}

80 81 82 83 84 85 86 87
static inline void i915_gem_object_fence_lost(struct drm_i915_gem_object *obj)
{
	if (obj->tiling_mode)
		i915_gem_release_mmap(obj);

	/* As we do not have an associated fence register, we will force
	 * a tiling change if we ever need to acquire one.
	 */
88
	obj->fence_dirty = false;
89 90 91
	obj->fence_reg = I915_FENCE_REG_NONE;
}

92 93 94 95
/* some bookkeeping */
static void i915_gem_info_add_obj(struct drm_i915_private *dev_priv,
				  size_t size)
{
96
	spin_lock(&dev_priv->mm.object_stat_lock);
97 98
	dev_priv->mm.object_count++;
	dev_priv->mm.object_memory += size;
99
	spin_unlock(&dev_priv->mm.object_stat_lock);
100 101 102 103 104
}

static void i915_gem_info_remove_obj(struct drm_i915_private *dev_priv,
				     size_t size)
{
105
	spin_lock(&dev_priv->mm.object_stat_lock);
106 107
	dev_priv->mm.object_count--;
	dev_priv->mm.object_memory -= size;
108
	spin_unlock(&dev_priv->mm.object_stat_lock);
109 110
}

111
static int
112
i915_gem_wait_for_error(struct i915_gpu_error *error)
113 114 115
{
	int ret;

116 117
#define EXIT_COND (!i915_reset_in_progress(error) || \
		   i915_terminally_wedged(error))
118
	if (EXIT_COND)
119 120
		return 0;

121 122 123 124 125
	/*
	 * Only wait 10 seconds for the gpu reset to complete to avoid hanging
	 * userspace. If it takes that long something really bad is going on and
	 * we should simply try to bail out and fail as gracefully as possible.
	 */
126 127 128
	ret = wait_event_interruptible_timeout(error->reset_queue,
					       EXIT_COND,
					       10*HZ);
129 130 131 132
	if (ret == 0) {
		DRM_ERROR("Timed out waiting for the gpu reset to complete\n");
		return -EIO;
	} else if (ret < 0) {
133
		return ret;
134
	}
135
#undef EXIT_COND
136

137
	return 0;
138 139
}

140
int i915_mutex_lock_interruptible(struct drm_device *dev)
141
{
142
	struct drm_i915_private *dev_priv = dev->dev_private;
143 144
	int ret;

145
	ret = i915_gem_wait_for_error(&dev_priv->gpu_error);
146 147 148 149 150 151 152
	if (ret)
		return ret;

	ret = mutex_lock_interruptible(&dev->struct_mutex);
	if (ret)
		return ret;

153
	WARN_ON(i915_verify_lists(dev));
154 155
	return 0;
}
156

157
static inline bool
158
i915_gem_object_is_inactive(struct drm_i915_gem_object *obj)
159
{
160
	return i915_gem_obj_bound_any(obj) && !obj->active;
161 162
}

J
Jesse Barnes 已提交
163 164
int
i915_gem_init_ioctl(struct drm_device *dev, void *data,
165
		    struct drm_file *file)
J
Jesse Barnes 已提交
166
{
167
	struct drm_i915_private *dev_priv = dev->dev_private;
J
Jesse Barnes 已提交
168
	struct drm_i915_gem_init *args = data;
169

170 171 172
	if (drm_core_check_feature(dev, DRIVER_MODESET))
		return -ENODEV;

173 174 175
	if (args->gtt_start >= args->gtt_end ||
	    (args->gtt_end | args->gtt_start) & (PAGE_SIZE - 1))
		return -EINVAL;
J
Jesse Barnes 已提交
176

177 178 179 180
	/* GEM with user mode setting was never supported on ilk and later. */
	if (INTEL_INFO(dev)->gen >= 5)
		return -ENODEV;

J
Jesse Barnes 已提交
181
	mutex_lock(&dev->struct_mutex);
182 183
	i915_gem_setup_global_gtt(dev, args->gtt_start, args->gtt_end,
				  args->gtt_end);
184
	dev_priv->gtt.mappable_end = args->gtt_end;
185 186
	mutex_unlock(&dev->struct_mutex);

187
	return 0;
188 189
}

190 191
int
i915_gem_get_aperture_ioctl(struct drm_device *dev, void *data,
192
			    struct drm_file *file)
193
{
194
	struct drm_i915_private *dev_priv = dev->dev_private;
195
	struct drm_i915_gem_get_aperture *args = data;
196 197
	struct drm_i915_gem_object *obj;
	size_t pinned;
198

199
	pinned = 0;
200
	mutex_lock(&dev->struct_mutex);
201
	list_for_each_entry(obj, &dev_priv->mm.bound_list, global_list)
202
		if (obj->pin_count)
203
			pinned += i915_gem_obj_ggtt_size(obj);
204
	mutex_unlock(&dev->struct_mutex);
205

206
	args->aper_size = dev_priv->gtt.base.total;
207
	args->aper_available_size = args->aper_size - pinned;
208

209 210 211
	return 0;
}

212 213 214
void *i915_gem_object_alloc(struct drm_device *dev)
{
	struct drm_i915_private *dev_priv = dev->dev_private;
215
	return kmem_cache_zalloc(dev_priv->slab, GFP_KERNEL);
216 217 218 219 220 221 222 223
}

void i915_gem_object_free(struct drm_i915_gem_object *obj)
{
	struct drm_i915_private *dev_priv = obj->base.dev->dev_private;
	kmem_cache_free(dev_priv->slab, obj);
}

224 225 226 227 228
static int
i915_gem_create(struct drm_file *file,
		struct drm_device *dev,
		uint64_t size,
		uint32_t *handle_p)
229
{
230
	struct drm_i915_gem_object *obj;
231 232
	int ret;
	u32 handle;
233

234
	size = roundup(size, PAGE_SIZE);
235 236
	if (size == 0)
		return -EINVAL;
237 238

	/* Allocate the new object */
239
	obj = i915_gem_alloc_object(dev, size);
240 241 242
	if (obj == NULL)
		return -ENOMEM;

243
	ret = drm_gem_handle_create(file, &obj->base, &handle);
244
	/* drop reference from allocate - handle holds it now */
245 246 247
	drm_gem_object_unreference_unlocked(&obj->base);
	if (ret)
		return ret;
248

249
	*handle_p = handle;
250 251 252
	return 0;
}

253 254 255 256 257 258
int
i915_gem_dumb_create(struct drm_file *file,
		     struct drm_device *dev,
		     struct drm_mode_create_dumb *args)
{
	/* have to work out size/pitch and return them */
259
	args->pitch = ALIGN(args->width * ((args->bpp + 7) / 8), 64);
260 261 262 263 264 265 266 267 268 269 270 271 272
	args->size = args->pitch * args->height;
	return i915_gem_create(file, dev,
			       args->size, &args->handle);
}

/**
 * Creates a new mm object and returns a handle to it.
 */
int
i915_gem_create_ioctl(struct drm_device *dev, void *data,
		      struct drm_file *file)
{
	struct drm_i915_gem_create *args = data;
273

274 275 276 277
	return i915_gem_create(file, dev,
			       args->size, &args->handle);
}

278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303
static inline int
__copy_to_user_swizzled(char __user *cpu_vaddr,
			const char *gpu_vaddr, int gpu_offset,
			int length)
{
	int ret, cpu_offset = 0;

	while (length > 0) {
		int cacheline_end = ALIGN(gpu_offset + 1, 64);
		int this_length = min(cacheline_end - gpu_offset, length);
		int swizzled_gpu_offset = gpu_offset ^ 64;

		ret = __copy_to_user(cpu_vaddr + cpu_offset,
				     gpu_vaddr + swizzled_gpu_offset,
				     this_length);
		if (ret)
			return ret + length;

		cpu_offset += this_length;
		gpu_offset += this_length;
		length -= this_length;
	}

	return 0;
}

304
static inline int
305 306
__copy_from_user_swizzled(char *gpu_vaddr, int gpu_offset,
			  const char __user *cpu_vaddr,
307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329
			  int length)
{
	int ret, cpu_offset = 0;

	while (length > 0) {
		int cacheline_end = ALIGN(gpu_offset + 1, 64);
		int this_length = min(cacheline_end - gpu_offset, length);
		int swizzled_gpu_offset = gpu_offset ^ 64;

		ret = __copy_from_user(gpu_vaddr + swizzled_gpu_offset,
				       cpu_vaddr + cpu_offset,
				       this_length);
		if (ret)
			return ret + length;

		cpu_offset += this_length;
		gpu_offset += this_length;
		length -= this_length;
	}

	return 0;
}

330 331 332
/* Per-page copy function for the shmem pread fastpath.
 * Flushes invalid cachelines before reading the target if
 * needs_clflush is set. */
333
static int
334 335 336 337 338 339 340
shmem_pread_fast(struct page *page, int shmem_page_offset, int page_length,
		 char __user *user_data,
		 bool page_do_bit17_swizzling, bool needs_clflush)
{
	char *vaddr;
	int ret;

341
	if (unlikely(page_do_bit17_swizzling))
342 343 344 345 346 347 348 349 350 351 352
		return -EINVAL;

	vaddr = kmap_atomic(page);
	if (needs_clflush)
		drm_clflush_virt_range(vaddr + shmem_page_offset,
				       page_length);
	ret = __copy_to_user_inatomic(user_data,
				      vaddr + shmem_page_offset,
				      page_length);
	kunmap_atomic(vaddr);

353
	return ret ? -EFAULT : 0;
354 355
}

356 357 358 359
static void
shmem_clflush_swizzled_range(char *addr, unsigned long length,
			     bool swizzled)
{
360
	if (unlikely(swizzled)) {
361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377
		unsigned long start = (unsigned long) addr;
		unsigned long end = (unsigned long) addr + length;

		/* For swizzling simply ensure that we always flush both
		 * channels. Lame, but simple and it works. Swizzled
		 * pwrite/pread is far from a hotpath - current userspace
		 * doesn't use it at all. */
		start = round_down(start, 128);
		end = round_up(end, 128);

		drm_clflush_virt_range((void *)start, end - start);
	} else {
		drm_clflush_virt_range(addr, length);
	}

}

378 379 380 381 382 383 384 385 386 387 388 389
/* Only difference to the fast-path function is that this can handle bit17
 * and uses non-atomic copy and kmap functions. */
static int
shmem_pread_slow(struct page *page, int shmem_page_offset, int page_length,
		 char __user *user_data,
		 bool page_do_bit17_swizzling, bool needs_clflush)
{
	char *vaddr;
	int ret;

	vaddr = kmap(page);
	if (needs_clflush)
390 391 392
		shmem_clflush_swizzled_range(vaddr + shmem_page_offset,
					     page_length,
					     page_do_bit17_swizzling);
393 394 395 396 397 398 399 400 401 402 403

	if (page_do_bit17_swizzling)
		ret = __copy_to_user_swizzled(user_data,
					      vaddr, shmem_page_offset,
					      page_length);
	else
		ret = __copy_to_user(user_data,
				     vaddr + shmem_page_offset,
				     page_length);
	kunmap(page);

404
	return ret ? - EFAULT : 0;
405 406
}

407
static int
408 409 410 411
i915_gem_shmem_pread(struct drm_device *dev,
		     struct drm_i915_gem_object *obj,
		     struct drm_i915_gem_pread *args,
		     struct drm_file *file)
412
{
413
	char __user *user_data;
414
	ssize_t remain;
415
	loff_t offset;
416
	int shmem_page_offset, page_length, ret = 0;
417
	int obj_do_bit17_swizzling, page_do_bit17_swizzling;
418
	int prefaulted = 0;
419
	int needs_clflush = 0;
420
	struct sg_page_iter sg_iter;
421

V
Ville Syrjälä 已提交
422
	user_data = to_user_ptr(args->data_ptr);
423 424
	remain = args->size;

425
	obj_do_bit17_swizzling = i915_gem_object_needs_bit17_swizzle(obj);
426

427 428 429 430 431
	if (!(obj->base.read_domains & I915_GEM_DOMAIN_CPU)) {
		/* If we're not in the cpu read domain, set ourself into the gtt
		 * read domain and manually flush cachelines (if required). This
		 * optimizes for the case when the gpu will dirty the data
		 * anyway again before the next pread happens. */
432
		needs_clflush = !cpu_cache_is_coherent(dev, obj->cache_level);
433
		if (i915_gem_obj_bound_any(obj)) {
C
Chris Wilson 已提交
434 435 436 437
			ret = i915_gem_object_set_to_gtt_domain(obj, false);
			if (ret)
				return ret;
		}
438
	}
439

440 441 442 443 444 445
	ret = i915_gem_object_get_pages(obj);
	if (ret)
		return ret;

	i915_gem_object_pin_pages(obj);

446
	offset = args->offset;
447

448 449
	for_each_sg_page(obj->pages->sgl, &sg_iter, obj->pages->nents,
			 offset >> PAGE_SHIFT) {
450
		struct page *page = sg_page_iter_page(&sg_iter);
451 452 453 454

		if (remain <= 0)
			break;

455 456 457 458 459
		/* Operation in this page
		 *
		 * shmem_page_offset = offset within page in shmem file
		 * page_length = bytes to copy for this page
		 */
460
		shmem_page_offset = offset_in_page(offset);
461 462 463 464
		page_length = remain;
		if ((shmem_page_offset + page_length) > PAGE_SIZE)
			page_length = PAGE_SIZE - shmem_page_offset;

465 466 467
		page_do_bit17_swizzling = obj_do_bit17_swizzling &&
			(page_to_phys(page) & (1 << 17)) != 0;

468 469 470 471 472
		ret = shmem_pread_fast(page, shmem_page_offset, page_length,
				       user_data, page_do_bit17_swizzling,
				       needs_clflush);
		if (ret == 0)
			goto next_page;
473 474 475

		mutex_unlock(&dev->struct_mutex);

476
		if (likely(!i915_prefault_disable) && !prefaulted) {
477
			ret = fault_in_multipages_writeable(user_data, remain);
478 479 480 481 482 483 484
			/* Userspace is tricking us, but we've already clobbered
			 * its pages with the prefault and promised to write the
			 * data up to the first fault. Hence ignore any errors
			 * and just continue. */
			(void)ret;
			prefaulted = 1;
		}
485

486 487 488
		ret = shmem_pread_slow(page, shmem_page_offset, page_length,
				       user_data, page_do_bit17_swizzling,
				       needs_clflush);
489

490
		mutex_lock(&dev->struct_mutex);
491

492
next_page:
493 494
		mark_page_accessed(page);

495
		if (ret)
496 497
			goto out;

498
		remain -= page_length;
499
		user_data += page_length;
500 501 502
		offset += page_length;
	}

503
out:
504 505
	i915_gem_object_unpin_pages(obj);

506 507 508
	return ret;
}

509 510 511 512 513 514 515
/**
 * Reads data from the object referenced by handle.
 *
 * On error, the contents of *data are undefined.
 */
int
i915_gem_pread_ioctl(struct drm_device *dev, void *data,
516
		     struct drm_file *file)
517 518
{
	struct drm_i915_gem_pread *args = data;
519
	struct drm_i915_gem_object *obj;
520
	int ret = 0;
521

522 523 524 525
	if (args->size == 0)
		return 0;

	if (!access_ok(VERIFY_WRITE,
V
Ville Syrjälä 已提交
526
		       to_user_ptr(args->data_ptr),
527 528 529
		       args->size))
		return -EFAULT;

530
	ret = i915_mutex_lock_interruptible(dev);
531
	if (ret)
532
		return ret;
533

534
	obj = to_intel_bo(drm_gem_object_lookup(dev, file, args->handle));
535
	if (&obj->base == NULL) {
536 537
		ret = -ENOENT;
		goto unlock;
538
	}
539

540
	/* Bounds check source.  */
541 542
	if (args->offset > obj->base.size ||
	    args->size > obj->base.size - args->offset) {
C
Chris Wilson 已提交
543
		ret = -EINVAL;
544
		goto out;
C
Chris Wilson 已提交
545 546
	}

547 548 549 550 551 552 553 554
	/* prime objects have no backing filp to GEM pread/pwrite
	 * pages from.
	 */
	if (!obj->base.filp) {
		ret = -EINVAL;
		goto out;
	}

C
Chris Wilson 已提交
555 556
	trace_i915_gem_object_pread(obj, args->offset, args->size);

557
	ret = i915_gem_shmem_pread(dev, obj, args, file);
558

559
out:
560
	drm_gem_object_unreference(&obj->base);
561
unlock:
562
	mutex_unlock(&dev->struct_mutex);
563
	return ret;
564 565
}

566 567
/* This is the fast write path which cannot handle
 * page faults in the source data
568
 */
569 570 571 572 573 574

static inline int
fast_user_write(struct io_mapping *mapping,
		loff_t page_base, int page_offset,
		char __user *user_data,
		int length)
575
{
576 577
	void __iomem *vaddr_atomic;
	void *vaddr;
578
	unsigned long unwritten;
579

P
Peter Zijlstra 已提交
580
	vaddr_atomic = io_mapping_map_atomic_wc(mapping, page_base);
581 582 583
	/* We can use the cpu mem copy function because this is X86. */
	vaddr = (void __force*)vaddr_atomic + page_offset;
	unwritten = __copy_from_user_inatomic_nocache(vaddr,
584
						      user_data, length);
P
Peter Zijlstra 已提交
585
	io_mapping_unmap_atomic(vaddr_atomic);
586
	return unwritten;
587 588
}

589 590 591 592
/**
 * This is the fast pwrite path, where we copy the data directly from the
 * user into the GTT, uncached.
 */
593
static int
594 595
i915_gem_gtt_pwrite_fast(struct drm_device *dev,
			 struct drm_i915_gem_object *obj,
596
			 struct drm_i915_gem_pwrite *args,
597
			 struct drm_file *file)
598
{
599
	drm_i915_private_t *dev_priv = dev->dev_private;
600
	ssize_t remain;
601
	loff_t offset, page_base;
602
	char __user *user_data;
D
Daniel Vetter 已提交
603 604
	int page_offset, page_length, ret;

B
Ben Widawsky 已提交
605
	ret = i915_gem_obj_ggtt_pin(obj, 0, true, true);
D
Daniel Vetter 已提交
606 607 608 609 610 611 612 613 614 615
	if (ret)
		goto out;

	ret = i915_gem_object_set_to_gtt_domain(obj, true);
	if (ret)
		goto out_unpin;

	ret = i915_gem_object_put_fence(obj);
	if (ret)
		goto out_unpin;
616

V
Ville Syrjälä 已提交
617
	user_data = to_user_ptr(args->data_ptr);
618 619
	remain = args->size;

620
	offset = i915_gem_obj_ggtt_offset(obj) + args->offset;
621 622 623 624

	while (remain > 0) {
		/* Operation in this page
		 *
625 626 627
		 * page_base = page offset within aperture
		 * page_offset = offset within page
		 * page_length = bytes to copy for this page
628
		 */
629 630
		page_base = offset & PAGE_MASK;
		page_offset = offset_in_page(offset);
631 632 633 634 635
		page_length = remain;
		if ((page_offset + remain) > PAGE_SIZE)
			page_length = PAGE_SIZE - page_offset;

		/* If we get a fault while copying data, then (presumably) our
636 637
		 * source page isn't available.  Return the error and we'll
		 * retry in the slow path.
638
		 */
B
Ben Widawsky 已提交
639
		if (fast_user_write(dev_priv->gtt.mappable, page_base,
D
Daniel Vetter 已提交
640 641 642 643
				    page_offset, user_data, page_length)) {
			ret = -EFAULT;
			goto out_unpin;
		}
644

645 646 647
		remain -= page_length;
		user_data += page_length;
		offset += page_length;
648 649
	}

D
Daniel Vetter 已提交
650 651 652
out_unpin:
	i915_gem_object_unpin(obj);
out:
653
	return ret;
654 655
}

656 657 658 659
/* Per-page copy function for the shmem pwrite fastpath.
 * Flushes invalid cachelines before writing to the target if
 * needs_clflush_before is set and flushes out any written cachelines after
 * writing if needs_clflush is set. */
660
static int
661 662 663 664 665
shmem_pwrite_fast(struct page *page, int shmem_page_offset, int page_length,
		  char __user *user_data,
		  bool page_do_bit17_swizzling,
		  bool needs_clflush_before,
		  bool needs_clflush_after)
666
{
667
	char *vaddr;
668
	int ret;
669

670
	if (unlikely(page_do_bit17_swizzling))
671
		return -EINVAL;
672

673 674 675 676 677 678 679 680 681 682 683
	vaddr = kmap_atomic(page);
	if (needs_clflush_before)
		drm_clflush_virt_range(vaddr + shmem_page_offset,
				       page_length);
	ret = __copy_from_user_inatomic_nocache(vaddr + shmem_page_offset,
						user_data,
						page_length);
	if (needs_clflush_after)
		drm_clflush_virt_range(vaddr + shmem_page_offset,
				       page_length);
	kunmap_atomic(vaddr);
684

685
	return ret ? -EFAULT : 0;
686 687
}

688 689
/* Only difference to the fast-path function is that this can handle bit17
 * and uses non-atomic copy and kmap functions. */
690
static int
691 692 693 694 695
shmem_pwrite_slow(struct page *page, int shmem_page_offset, int page_length,
		  char __user *user_data,
		  bool page_do_bit17_swizzling,
		  bool needs_clflush_before,
		  bool needs_clflush_after)
696
{
697 698
	char *vaddr;
	int ret;
699

700
	vaddr = kmap(page);
701
	if (unlikely(needs_clflush_before || page_do_bit17_swizzling))
702 703 704
		shmem_clflush_swizzled_range(vaddr + shmem_page_offset,
					     page_length,
					     page_do_bit17_swizzling);
705 706
	if (page_do_bit17_swizzling)
		ret = __copy_from_user_swizzled(vaddr, shmem_page_offset,
707 708
						user_data,
						page_length);
709 710 711 712 713
	else
		ret = __copy_from_user(vaddr + shmem_page_offset,
				       user_data,
				       page_length);
	if (needs_clflush_after)
714 715 716
		shmem_clflush_swizzled_range(vaddr + shmem_page_offset,
					     page_length,
					     page_do_bit17_swizzling);
717
	kunmap(page);
718

719
	return ret ? -EFAULT : 0;
720 721 722
}

static int
723 724 725 726
i915_gem_shmem_pwrite(struct drm_device *dev,
		      struct drm_i915_gem_object *obj,
		      struct drm_i915_gem_pwrite *args,
		      struct drm_file *file)
727 728
{
	ssize_t remain;
729 730
	loff_t offset;
	char __user *user_data;
731
	int shmem_page_offset, page_length, ret = 0;
732
	int obj_do_bit17_swizzling, page_do_bit17_swizzling;
733
	int hit_slowpath = 0;
734 735
	int needs_clflush_after = 0;
	int needs_clflush_before = 0;
736
	struct sg_page_iter sg_iter;
737

V
Ville Syrjälä 已提交
738
	user_data = to_user_ptr(args->data_ptr);
739 740
	remain = args->size;

741
	obj_do_bit17_swizzling = i915_gem_object_needs_bit17_swizzle(obj);
742

743 744 745 746 747
	if (obj->base.write_domain != I915_GEM_DOMAIN_CPU) {
		/* If we're not in the cpu write domain, set ourself into the gtt
		 * write domain and manually flush cachelines (if required). This
		 * optimizes for the case when the gpu will use the data
		 * right away and we therefore have to clflush anyway. */
748
		needs_clflush_after = cpu_write_needs_clflush(obj);
749
		if (i915_gem_obj_bound_any(obj)) {
C
Chris Wilson 已提交
750 751 752 753
			ret = i915_gem_object_set_to_gtt_domain(obj, true);
			if (ret)
				return ret;
		}
754
	}
755 756 757 758 759
	/* Same trick applies to invalidate partially written cachelines read
	 * before writing. */
	if ((obj->base.read_domains & I915_GEM_DOMAIN_CPU) == 0)
		needs_clflush_before =
			!cpu_cache_is_coherent(dev, obj->cache_level);
760

761 762 763 764 765 766
	ret = i915_gem_object_get_pages(obj);
	if (ret)
		return ret;

	i915_gem_object_pin_pages(obj);

767
	offset = args->offset;
768
	obj->dirty = 1;
769

770 771
	for_each_sg_page(obj->pages->sgl, &sg_iter, obj->pages->nents,
			 offset >> PAGE_SHIFT) {
772
		struct page *page = sg_page_iter_page(&sg_iter);
773
		int partial_cacheline_write;
774

775 776 777
		if (remain <= 0)
			break;

778 779 780 781 782
		/* Operation in this page
		 *
		 * shmem_page_offset = offset within page in shmem file
		 * page_length = bytes to copy for this page
		 */
783
		shmem_page_offset = offset_in_page(offset);
784 785 786 787 788

		page_length = remain;
		if ((shmem_page_offset + page_length) > PAGE_SIZE)
			page_length = PAGE_SIZE - shmem_page_offset;

789 790 791 792 793 794 795
		/* If we don't overwrite a cacheline completely we need to be
		 * careful to have up-to-date data by first clflushing. Don't
		 * overcomplicate things and flush the entire patch. */
		partial_cacheline_write = needs_clflush_before &&
			((shmem_page_offset | page_length)
				& (boot_cpu_data.x86_clflush_size - 1));

796 797 798
		page_do_bit17_swizzling = obj_do_bit17_swizzling &&
			(page_to_phys(page) & (1 << 17)) != 0;

799 800 801 802 803 804
		ret = shmem_pwrite_fast(page, shmem_page_offset, page_length,
					user_data, page_do_bit17_swizzling,
					partial_cacheline_write,
					needs_clflush_after);
		if (ret == 0)
			goto next_page;
805 806 807

		hit_slowpath = 1;
		mutex_unlock(&dev->struct_mutex);
808 809 810 811
		ret = shmem_pwrite_slow(page, shmem_page_offset, page_length,
					user_data, page_do_bit17_swizzling,
					partial_cacheline_write,
					needs_clflush_after);
812

813
		mutex_lock(&dev->struct_mutex);
814

815
next_page:
816 817 818
		set_page_dirty(page);
		mark_page_accessed(page);

819
		if (ret)
820 821
			goto out;

822
		remain -= page_length;
823
		user_data += page_length;
824
		offset += page_length;
825 826
	}

827
out:
828 829
	i915_gem_object_unpin_pages(obj);

830
	if (hit_slowpath) {
831 832 833 834 835 836 837
		/*
		 * Fixup: Flush cpu caches in case we didn't flush the dirty
		 * cachelines in-line while writing and the object moved
		 * out of the cpu write domain while we've dropped the lock.
		 */
		if (!needs_clflush_after &&
		    obj->base.write_domain != I915_GEM_DOMAIN_CPU) {
838 839
			if (i915_gem_clflush_object(obj, obj->pin_display))
				i915_gem_chipset_flush(dev);
840
		}
841
	}
842

843
	if (needs_clflush_after)
844
		i915_gem_chipset_flush(dev);
845

846
	return ret;
847 848 849 850 851 852 853 854 855
}

/**
 * Writes data to the object referenced by handle.
 *
 * On error, the contents of the buffer that were to be modified are undefined.
 */
int
i915_gem_pwrite_ioctl(struct drm_device *dev, void *data,
856
		      struct drm_file *file)
857 858
{
	struct drm_i915_gem_pwrite *args = data;
859
	struct drm_i915_gem_object *obj;
860 861 862 863 864 865
	int ret;

	if (args->size == 0)
		return 0;

	if (!access_ok(VERIFY_READ,
V
Ville Syrjälä 已提交
866
		       to_user_ptr(args->data_ptr),
867 868 869
		       args->size))
		return -EFAULT;

870 871 872 873 874 875
	if (likely(!i915_prefault_disable)) {
		ret = fault_in_multipages_readable(to_user_ptr(args->data_ptr),
						   args->size);
		if (ret)
			return -EFAULT;
	}
876

877
	ret = i915_mutex_lock_interruptible(dev);
878
	if (ret)
879
		return ret;
880

881
	obj = to_intel_bo(drm_gem_object_lookup(dev, file, args->handle));
882
	if (&obj->base == NULL) {
883 884
		ret = -ENOENT;
		goto unlock;
885
	}
886

887
	/* Bounds check destination. */
888 889
	if (args->offset > obj->base.size ||
	    args->size > obj->base.size - args->offset) {
C
Chris Wilson 已提交
890
		ret = -EINVAL;
891
		goto out;
C
Chris Wilson 已提交
892 893
	}

894 895 896 897 898 899 900 901
	/* prime objects have no backing filp to GEM pread/pwrite
	 * pages from.
	 */
	if (!obj->base.filp) {
		ret = -EINVAL;
		goto out;
	}

C
Chris Wilson 已提交
902 903
	trace_i915_gem_object_pwrite(obj, args->offset, args->size);

D
Daniel Vetter 已提交
904
	ret = -EFAULT;
905 906 907 908 909 910
	/* We can only do the GTT pwrite on untiled buffers, as otherwise
	 * it would end up going through the fenced access, and we'll get
	 * different detiling behavior between reading and writing.
	 * pread/pwrite currently are reading and writing from the CPU
	 * perspective, requiring manual detiling by the client.
	 */
911
	if (obj->phys_obj) {
912
		ret = i915_gem_phys_pwrite(dev, obj, args, file);
913 914 915
		goto out;
	}

916 917 918
	if (obj->tiling_mode == I915_TILING_NONE &&
	    obj->base.write_domain != I915_GEM_DOMAIN_CPU &&
	    cpu_write_needs_clflush(obj)) {
919
		ret = i915_gem_gtt_pwrite_fast(dev, obj, args, file);
D
Daniel Vetter 已提交
920 921 922
		/* Note that the gtt paths might fail with non-page-backed user
		 * pointers (e.g. gtt mappings when moving data between
		 * textures). Fallback to the shmem path in that case. */
923
	}
924

925
	if (ret == -EFAULT || ret == -ENOSPC)
D
Daniel Vetter 已提交
926
		ret = i915_gem_shmem_pwrite(dev, obj, args, file);
927

928
out:
929
	drm_gem_object_unreference(&obj->base);
930
unlock:
931
	mutex_unlock(&dev->struct_mutex);
932 933 934
	return ret;
}

935
int
936
i915_gem_check_wedge(struct i915_gpu_error *error,
937 938
		     bool interruptible)
{
939
	if (i915_reset_in_progress(error)) {
940 941 942 943 944
		/* Non-interruptible callers can't handle -EAGAIN, hence return
		 * -EIO unconditionally for these. */
		if (!interruptible)
			return -EIO;

945 946
		/* Recovery complete, but the reset failed ... */
		if (i915_terminally_wedged(error))
947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966
			return -EIO;

		return -EAGAIN;
	}

	return 0;
}

/*
 * Compare seqno against outstanding lazy request. Emit a request if they are
 * equal.
 */
static int
i915_gem_check_olr(struct intel_ring_buffer *ring, u32 seqno)
{
	int ret;

	BUG_ON(!mutex_is_locked(&ring->dev->struct_mutex));

	ret = 0;
967
	if (seqno == ring->outstanding_lazy_seqno)
968
		ret = i915_add_request(ring, NULL);
969 970 971 972 973 974 975 976

	return ret;
}

/**
 * __wait_seqno - wait until execution of seqno has finished
 * @ring: the ring expected to report seqno
 * @seqno: duh!
977
 * @reset_counter: reset sequence associated with the given seqno
978 979 980
 * @interruptible: do an interruptible wait (normally yes)
 * @timeout: in - how long to wait (NULL forever); out - how much time remaining
 *
981 982 983 984 985 986 987
 * Note: It is of utmost importance that the passed in seqno and reset_counter
 * values have been read by the caller in an smp safe manner. Where read-side
 * locks are involved, it is sufficient to read the reset_counter before
 * unlocking the lock that protects the seqno. For lockless tricks, the
 * reset_counter _must_ be read before, and an appropriate smp_rmb must be
 * inserted.
 *
988 989 990 991
 * Returns 0 if the seqno was found within the alloted time. Else returns the
 * errno with remaining time filled in timeout argument.
 */
static int __wait_seqno(struct intel_ring_buffer *ring, u32 seqno,
992
			unsigned reset_counter,
993 994 995 996 997 998 999 1000 1001
			bool interruptible, struct timespec *timeout)
{
	drm_i915_private_t *dev_priv = ring->dev->dev_private;
	struct timespec before, now, wait_time={1,0};
	unsigned long timeout_jiffies;
	long end;
	bool wait_forever = true;
	int ret;

1002 1003
	WARN(dev_priv->pc8.irqs_disabled, "IRQs disabled\n");

1004 1005 1006 1007 1008 1009 1010 1011 1012 1013
	if (i915_seqno_passed(ring->get_seqno(ring, true), seqno))
		return 0;

	trace_i915_gem_request_wait_begin(ring, seqno);

	if (timeout != NULL) {
		wait_time = *timeout;
		wait_forever = false;
	}

1014
	timeout_jiffies = timespec_to_jiffies_timeout(&wait_time);
1015 1016 1017 1018 1019 1020 1021 1022 1023

	if (WARN_ON(!ring->irq_get(ring)))
		return -ENODEV;

	/* Record current time in case interrupted by signal, or wedged * */
	getrawmonotonic(&before);

#define EXIT_COND \
	(i915_seqno_passed(ring->get_seqno(ring, false), seqno) || \
1024 1025
	 i915_reset_in_progress(&dev_priv->gpu_error) || \
	 reset_counter != atomic_read(&dev_priv->gpu_error.reset_counter))
1026 1027 1028 1029 1030 1031 1032 1033 1034
	do {
		if (interruptible)
			end = wait_event_interruptible_timeout(ring->irq_queue,
							       EXIT_COND,
							       timeout_jiffies);
		else
			end = wait_event_timeout(ring->irq_queue, EXIT_COND,
						 timeout_jiffies);

1035 1036 1037 1038 1039 1040 1041
		/* We need to check whether any gpu reset happened in between
		 * the caller grabbing the seqno and now ... */
		if (reset_counter != atomic_read(&dev_priv->gpu_error.reset_counter))
			end = -EAGAIN;

		/* ... but upgrade the -EGAIN to an -EIO if the gpu is truely
		 * gone. */
1042
		ret = i915_gem_check_wedge(&dev_priv->gpu_error, interruptible);
1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055
		if (ret)
			end = ret;
	} while (end == 0 && wait_forever);

	getrawmonotonic(&now);

	ring->irq_put(ring);
	trace_i915_gem_request_wait_end(ring, seqno);
#undef EXIT_COND

	if (timeout) {
		struct timespec sleep_time = timespec_sub(now, before);
		*timeout = timespec_sub(*timeout, sleep_time);
1056 1057
		if (!timespec_valid(timeout)) /* i.e. negative time remains */
			set_normalized_timespec(timeout, 0, 0);
1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087
	}

	switch (end) {
	case -EIO:
	case -EAGAIN: /* Wedged */
	case -ERESTARTSYS: /* Signal */
		return (int)end;
	case 0: /* Timeout */
		return -ETIME;
	default: /* Completed */
		WARN_ON(end < 0); /* We're not aware of other errors */
		return 0;
	}
}

/**
 * Waits for a sequence number to be signaled, and cleans up the
 * request and object lists appropriately for that event.
 */
int
i915_wait_seqno(struct intel_ring_buffer *ring, uint32_t seqno)
{
	struct drm_device *dev = ring->dev;
	struct drm_i915_private *dev_priv = dev->dev_private;
	bool interruptible = dev_priv->mm.interruptible;
	int ret;

	BUG_ON(!mutex_is_locked(&dev->struct_mutex));
	BUG_ON(seqno == 0);

1088
	ret = i915_gem_check_wedge(&dev_priv->gpu_error, interruptible);
1089 1090 1091 1092 1093 1094 1095
	if (ret)
		return ret;

	ret = i915_gem_check_olr(ring, seqno);
	if (ret)
		return ret;

1096 1097 1098
	return __wait_seqno(ring, seqno,
			    atomic_read(&dev_priv->gpu_error.reset_counter),
			    interruptible, NULL);
1099 1100
}

1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119
static int
i915_gem_object_wait_rendering__tail(struct drm_i915_gem_object *obj,
				     struct intel_ring_buffer *ring)
{
	i915_gem_retire_requests_ring(ring);

	/* Manually manage the write flush as we may have not yet
	 * retired the buffer.
	 *
	 * Note that the last_write_seqno is always the earlier of
	 * the two (read/write) seqno, so if we haved successfully waited,
	 * we know we have passed the last write.
	 */
	obj->last_write_seqno = 0;
	obj->base.write_domain &= ~I915_GEM_GPU_DOMAINS;

	return 0;
}

1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139
/**
 * Ensures that all rendering to the object has completed and the object is
 * safe to unbind from the GTT or access from the CPU.
 */
static __must_check int
i915_gem_object_wait_rendering(struct drm_i915_gem_object *obj,
			       bool readonly)
{
	struct intel_ring_buffer *ring = obj->ring;
	u32 seqno;
	int ret;

	seqno = readonly ? obj->last_write_seqno : obj->last_read_seqno;
	if (seqno == 0)
		return 0;

	ret = i915_wait_seqno(ring, seqno);
	if (ret)
		return ret;

1140
	return i915_gem_object_wait_rendering__tail(obj, ring);
1141 1142
}

1143 1144 1145 1146 1147 1148 1149 1150 1151 1152
/* A nonblocking variant of the above wait. This is a highly dangerous routine
 * as the object state may change during this call.
 */
static __must_check int
i915_gem_object_wait_rendering__nonblocking(struct drm_i915_gem_object *obj,
					    bool readonly)
{
	struct drm_device *dev = obj->base.dev;
	struct drm_i915_private *dev_priv = dev->dev_private;
	struct intel_ring_buffer *ring = obj->ring;
1153
	unsigned reset_counter;
1154 1155 1156 1157 1158 1159 1160 1161 1162 1163
	u32 seqno;
	int ret;

	BUG_ON(!mutex_is_locked(&dev->struct_mutex));
	BUG_ON(!dev_priv->mm.interruptible);

	seqno = readonly ? obj->last_write_seqno : obj->last_read_seqno;
	if (seqno == 0)
		return 0;

1164
	ret = i915_gem_check_wedge(&dev_priv->gpu_error, true);
1165 1166 1167 1168 1169 1170 1171
	if (ret)
		return ret;

	ret = i915_gem_check_olr(ring, seqno);
	if (ret)
		return ret;

1172
	reset_counter = atomic_read(&dev_priv->gpu_error.reset_counter);
1173
	mutex_unlock(&dev->struct_mutex);
1174
	ret = __wait_seqno(ring, seqno, reset_counter, true, NULL);
1175
	mutex_lock(&dev->struct_mutex);
1176 1177
	if (ret)
		return ret;
1178

1179
	return i915_gem_object_wait_rendering__tail(obj, ring);
1180 1181
}

1182
/**
1183 1184
 * Called when user space prepares to use an object with the CPU, either
 * through the mmap ioctl's mapping or a GTT mapping.
1185 1186 1187
 */
int
i915_gem_set_domain_ioctl(struct drm_device *dev, void *data,
1188
			  struct drm_file *file)
1189 1190
{
	struct drm_i915_gem_set_domain *args = data;
1191
	struct drm_i915_gem_object *obj;
1192 1193
	uint32_t read_domains = args->read_domains;
	uint32_t write_domain = args->write_domain;
1194 1195
	int ret;

1196
	/* Only handle setting domains to types used by the CPU. */
1197
	if (write_domain & I915_GEM_GPU_DOMAINS)
1198 1199
		return -EINVAL;

1200
	if (read_domains & I915_GEM_GPU_DOMAINS)
1201 1202 1203 1204 1205 1206 1207 1208
		return -EINVAL;

	/* Having something in the write domain implies it's in the read
	 * domain, and only that read domain.  Enforce that in the request.
	 */
	if (write_domain != 0 && read_domains != write_domain)
		return -EINVAL;

1209
	ret = i915_mutex_lock_interruptible(dev);
1210
	if (ret)
1211
		return ret;
1212

1213
	obj = to_intel_bo(drm_gem_object_lookup(dev, file, args->handle));
1214
	if (&obj->base == NULL) {
1215 1216
		ret = -ENOENT;
		goto unlock;
1217
	}
1218

1219 1220 1221 1222 1223 1224 1225 1226
	/* Try to flush the object off the GPU without holding the lock.
	 * We will repeat the flush holding the lock in the normal manner
	 * to catch cases where we are gazumped.
	 */
	ret = i915_gem_object_wait_rendering__nonblocking(obj, !write_domain);
	if (ret)
		goto unref;

1227 1228
	if (read_domains & I915_GEM_DOMAIN_GTT) {
		ret = i915_gem_object_set_to_gtt_domain(obj, write_domain != 0);
1229 1230 1231 1232 1233 1234 1235

		/* Silently promote "you're not bound, there was nothing to do"
		 * to success, since the client was just asking us to
		 * make sure everything was done.
		 */
		if (ret == -EINVAL)
			ret = 0;
1236
	} else {
1237
		ret = i915_gem_object_set_to_cpu_domain(obj, write_domain != 0);
1238 1239
	}

1240
unref:
1241
	drm_gem_object_unreference(&obj->base);
1242
unlock:
1243 1244 1245 1246 1247 1248 1249 1250 1251
	mutex_unlock(&dev->struct_mutex);
	return ret;
}

/**
 * Called when user space has done writes to this buffer
 */
int
i915_gem_sw_finish_ioctl(struct drm_device *dev, void *data,
1252
			 struct drm_file *file)
1253 1254
{
	struct drm_i915_gem_sw_finish *args = data;
1255
	struct drm_i915_gem_object *obj;
1256 1257
	int ret = 0;

1258
	ret = i915_mutex_lock_interruptible(dev);
1259
	if (ret)
1260
		return ret;
1261

1262
	obj = to_intel_bo(drm_gem_object_lookup(dev, file, args->handle));
1263
	if (&obj->base == NULL) {
1264 1265
		ret = -ENOENT;
		goto unlock;
1266 1267 1268
	}

	/* Pinned buffers may be scanout, so flush the cache */
1269 1270
	if (obj->pin_display)
		i915_gem_object_flush_cpu_write_domain(obj, true);
1271

1272
	drm_gem_object_unreference(&obj->base);
1273
unlock:
1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286
	mutex_unlock(&dev->struct_mutex);
	return ret;
}

/**
 * Maps the contents of an object, returning the address it is mapped
 * into.
 *
 * While the mapping holds a reference on the contents of the object, it doesn't
 * imply a ref on the object itself.
 */
int
i915_gem_mmap_ioctl(struct drm_device *dev, void *data,
1287
		    struct drm_file *file)
1288 1289 1290 1291 1292
{
	struct drm_i915_gem_mmap *args = data;
	struct drm_gem_object *obj;
	unsigned long addr;

1293
	obj = drm_gem_object_lookup(dev, file, args->handle);
1294
	if (obj == NULL)
1295
		return -ENOENT;
1296

1297 1298 1299 1300 1301 1302 1303 1304
	/* prime objects have no backing filp to GEM mmap
	 * pages from.
	 */
	if (!obj->filp) {
		drm_gem_object_unreference_unlocked(obj);
		return -EINVAL;
	}

1305
	addr = vm_mmap(obj->filp, 0, args->size,
1306 1307
		       PROT_READ | PROT_WRITE, MAP_SHARED,
		       args->offset);
1308
	drm_gem_object_unreference_unlocked(obj);
1309 1310 1311 1312 1313 1314 1315 1316
	if (IS_ERR((void *)addr))
		return addr;

	args->addr_ptr = (uint64_t) addr;

	return 0;
}

1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334
/**
 * i915_gem_fault - fault a page into the GTT
 * vma: VMA in question
 * vmf: fault info
 *
 * The fault handler is set up by drm_gem_mmap() when a object is GTT mapped
 * from userspace.  The fault handler takes care of binding the object to
 * the GTT (if needed), allocating and programming a fence register (again,
 * only if needed based on whether the old reg is still valid or the object
 * is tiled) and inserting a new PTE into the faulting process.
 *
 * Note that the faulting process may involve evicting existing objects
 * from the GTT and/or fence registers to make room.  So performance may
 * suffer if the GTT working set is large or there are few fence registers
 * left.
 */
int i915_gem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
{
1335 1336
	struct drm_i915_gem_object *obj = to_intel_bo(vma->vm_private_data);
	struct drm_device *dev = obj->base.dev;
1337
	drm_i915_private_t *dev_priv = dev->dev_private;
1338 1339 1340
	pgoff_t page_offset;
	unsigned long pfn;
	int ret = 0;
1341
	bool write = !!(vmf->flags & FAULT_FLAG_WRITE);
1342 1343 1344 1345 1346

	/* We don't use vmf->pgoff since that has the fake offset */
	page_offset = ((unsigned long)vmf->virtual_address - vma->vm_start) >>
		PAGE_SHIFT;

1347 1348 1349
	ret = i915_mutex_lock_interruptible(dev);
	if (ret)
		goto out;
1350

C
Chris Wilson 已提交
1351 1352
	trace_i915_gem_object_fault(obj, page_offset, true, write);

1353 1354 1355 1356 1357 1358
	/* Access to snoopable pages through the GTT is incoherent. */
	if (obj->cache_level != I915_CACHE_NONE && !HAS_LLC(dev)) {
		ret = -EINVAL;
		goto unlock;
	}

1359
	/* Now bind it into the GTT if needed */
B
Ben Widawsky 已提交
1360
	ret = i915_gem_obj_ggtt_pin(obj,  0, true, false);
1361 1362
	if (ret)
		goto unlock;
1363

1364 1365 1366
	ret = i915_gem_object_set_to_gtt_domain(obj, write);
	if (ret)
		goto unpin;
1367

1368
	ret = i915_gem_object_get_fence(obj);
1369
	if (ret)
1370
		goto unpin;
1371

1372 1373
	obj->fault_mappable = true;

1374 1375 1376
	pfn = dev_priv->gtt.mappable_base + i915_gem_obj_ggtt_offset(obj);
	pfn >>= PAGE_SHIFT;
	pfn += page_offset;
1377 1378 1379

	/* Finally, remap it using the new GTT offset */
	ret = vm_insert_pfn(vma, (unsigned long)vmf->virtual_address, pfn);
1380 1381
unpin:
	i915_gem_object_unpin(obj);
1382
unlock:
1383
	mutex_unlock(&dev->struct_mutex);
1384
out:
1385
	switch (ret) {
1386
	case -EIO:
1387 1388 1389
		/* If this -EIO is due to a gpu hang, give the reset code a
		 * chance to clean up the mess. Otherwise return the proper
		 * SIGBUS. */
1390
		if (i915_terminally_wedged(&dev_priv->gpu_error))
1391
			return VM_FAULT_SIGBUS;
1392
	case -EAGAIN:
1393 1394 1395 1396 1397 1398 1399
		/* Give the error handler a chance to run and move the
		 * objects off the GPU active list. Next time we service the
		 * fault, we should be able to transition the page into the
		 * GTT without touching the GPU (and so avoid further
		 * EIO/EGAIN). If the GPU is wedged, then there is no issue
		 * with coherency, just lost writes.
		 */
1400
		set_need_resched();
1401 1402
	case 0:
	case -ERESTARTSYS:
1403
	case -EINTR:
1404 1405 1406 1407 1408
	case -EBUSY:
		/*
		 * EBUSY is ok: this just means that another thread
		 * already did the job.
		 */
1409
		return VM_FAULT_NOPAGE;
1410 1411
	case -ENOMEM:
		return VM_FAULT_OOM;
1412 1413
	case -ENOSPC:
		return VM_FAULT_SIGBUS;
1414
	default:
1415
		WARN_ONCE(ret, "unhandled error in i915_gem_fault: %i\n", ret);
1416
		return VM_FAULT_SIGBUS;
1417 1418 1419
	}
}

1420 1421 1422 1423
/**
 * i915_gem_release_mmap - remove physical page mappings
 * @obj: obj in question
 *
1424
 * Preserve the reservation of the mmapping with the DRM core code, but
1425 1426 1427 1428 1429 1430 1431 1432 1433
 * relinquish ownership of the pages back to the system.
 *
 * It is vital that we remove the page mapping if we have mapped a tiled
 * object through the GTT and then lose the fence register due to
 * resource pressure. Similarly if the object has been moved out of the
 * aperture, than pages mapped into userspace must be revoked. Removing the
 * mapping will then trigger a page fault on the next user access, allowing
 * fixup by i915_gem_fault().
 */
1434
void
1435
i915_gem_release_mmap(struct drm_i915_gem_object *obj)
1436
{
1437 1438
	if (!obj->fault_mappable)
		return;
1439

1440
	drm_vma_node_unmap(&obj->base.vma_node, obj->base.dev->dev_mapping);
1441
	obj->fault_mappable = false;
1442 1443
}

1444
uint32_t
1445
i915_gem_get_gtt_size(struct drm_device *dev, uint32_t size, int tiling_mode)
1446
{
1447
	uint32_t gtt_size;
1448 1449

	if (INTEL_INFO(dev)->gen >= 4 ||
1450 1451
	    tiling_mode == I915_TILING_NONE)
		return size;
1452 1453 1454

	/* Previous chips need a power-of-two fence region when tiling */
	if (INTEL_INFO(dev)->gen == 3)
1455
		gtt_size = 1024*1024;
1456
	else
1457
		gtt_size = 512*1024;
1458

1459 1460
	while (gtt_size < size)
		gtt_size <<= 1;
1461

1462
	return gtt_size;
1463 1464
}

1465 1466 1467 1468 1469
/**
 * i915_gem_get_gtt_alignment - return required GTT alignment for an object
 * @obj: object to check
 *
 * Return the required GTT alignment for an object, taking into account
1470
 * potential fence register mapping.
1471
 */
1472 1473 1474
uint32_t
i915_gem_get_gtt_alignment(struct drm_device *dev, uint32_t size,
			   int tiling_mode, bool fenced)
1475 1476 1477 1478 1479
{
	/*
	 * Minimum alignment is 4k (GTT page size), but might be greater
	 * if a fence register is needed for the object.
	 */
1480
	if (INTEL_INFO(dev)->gen >= 4 || (!fenced && IS_G33(dev)) ||
1481
	    tiling_mode == I915_TILING_NONE)
1482 1483
		return 4096;

1484 1485 1486 1487
	/*
	 * Previous chips need to be aligned to the size of the smallest
	 * fence register that can contain the object.
	 */
1488
	return i915_gem_get_gtt_size(dev, size, tiling_mode);
1489 1490
}

1491 1492 1493 1494 1495
static int i915_gem_object_create_mmap_offset(struct drm_i915_gem_object *obj)
{
	struct drm_i915_private *dev_priv = obj->base.dev->dev_private;
	int ret;

1496
	if (drm_vma_node_has_offset(&obj->base.vma_node))
1497 1498
		return 0;

1499 1500
	dev_priv->mm.shrinker_no_lock_stealing = true;

1501 1502
	ret = drm_gem_create_mmap_offset(&obj->base);
	if (ret != -ENOSPC)
1503
		goto out;
1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514

	/* Badly fragmented mmap space? The only way we can recover
	 * space is by destroying unwanted objects. We can't randomly release
	 * mmap_offsets as userspace expects them to be persistent for the
	 * lifetime of the objects. The closest we can is to release the
	 * offsets on purgeable objects by truncating it and marking it purged,
	 * which prevents userspace from ever using that object again.
	 */
	i915_gem_purge(dev_priv, obj->base.size >> PAGE_SHIFT);
	ret = drm_gem_create_mmap_offset(&obj->base);
	if (ret != -ENOSPC)
1515
		goto out;
1516 1517

	i915_gem_shrink_all(dev_priv);
1518 1519 1520 1521 1522
	ret = drm_gem_create_mmap_offset(&obj->base);
out:
	dev_priv->mm.shrinker_no_lock_stealing = false;

	return ret;
1523 1524 1525 1526 1527 1528 1529
}

static void i915_gem_object_free_mmap_offset(struct drm_i915_gem_object *obj)
{
	drm_gem_free_mmap_offset(&obj->base);
}

1530
int
1531 1532 1533 1534
i915_gem_mmap_gtt(struct drm_file *file,
		  struct drm_device *dev,
		  uint32_t handle,
		  uint64_t *offset)
1535
{
1536
	struct drm_i915_private *dev_priv = dev->dev_private;
1537
	struct drm_i915_gem_object *obj;
1538 1539
	int ret;

1540
	ret = i915_mutex_lock_interruptible(dev);
1541
	if (ret)
1542
		return ret;
1543

1544
	obj = to_intel_bo(drm_gem_object_lookup(dev, file, handle));
1545
	if (&obj->base == NULL) {
1546 1547 1548
		ret = -ENOENT;
		goto unlock;
	}
1549

B
Ben Widawsky 已提交
1550
	if (obj->base.size > dev_priv->gtt.mappable_end) {
1551
		ret = -E2BIG;
1552
		goto out;
1553 1554
	}

1555
	if (obj->madv != I915_MADV_WILLNEED) {
1556
		DRM_ERROR("Attempting to mmap a purgeable buffer\n");
1557 1558
		ret = -EINVAL;
		goto out;
1559 1560
	}

1561 1562 1563
	ret = i915_gem_object_create_mmap_offset(obj);
	if (ret)
		goto out;
1564

1565
	*offset = drm_vma_node_offset_addr(&obj->base.vma_node);
1566

1567
out:
1568
	drm_gem_object_unreference(&obj->base);
1569
unlock:
1570
	mutex_unlock(&dev->struct_mutex);
1571
	return ret;
1572 1573
}

1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597
/**
 * i915_gem_mmap_gtt_ioctl - prepare an object for GTT mmap'ing
 * @dev: DRM device
 * @data: GTT mapping ioctl data
 * @file: GEM object info
 *
 * Simply returns the fake offset to userspace so it can mmap it.
 * The mmap call will end up in drm_gem_mmap(), which will set things
 * up so we can get faults in the handler above.
 *
 * The fault handler will take care of binding the object into the GTT
 * (since it may have been evicted to make room for something), allocating
 * a fence register, and mapping the appropriate aperture address into
 * userspace.
 */
int
i915_gem_mmap_gtt_ioctl(struct drm_device *dev, void *data,
			struct drm_file *file)
{
	struct drm_i915_gem_mmap_gtt *args = data;

	return i915_gem_mmap_gtt(file, dev, args->handle, &args->offset);
}

D
Daniel Vetter 已提交
1598 1599 1600
/* Immediately discard the backing storage */
static void
i915_gem_object_truncate(struct drm_i915_gem_object *obj)
1601 1602 1603
{
	struct inode *inode;

1604
	i915_gem_object_free_mmap_offset(obj);
1605

1606 1607
	if (obj->base.filp == NULL)
		return;
1608

D
Daniel Vetter 已提交
1609 1610 1611 1612 1613
	/* Our goal here is to return as much of the memory as
	 * is possible back to the system as we are called from OOM.
	 * To do this we must instruct the shmfs to drop all of its
	 * backing pages, *now*.
	 */
A
Al Viro 已提交
1614
	inode = file_inode(obj->base.filp);
D
Daniel Vetter 已提交
1615
	shmem_truncate_range(inode, 0, (loff_t)-1);
1616

D
Daniel Vetter 已提交
1617 1618
	obj->madv = __I915_MADV_PURGED;
}
1619

D
Daniel Vetter 已提交
1620 1621 1622 1623
static inline int
i915_gem_object_is_purgeable(struct drm_i915_gem_object *obj)
{
	return obj->madv == I915_MADV_DONTNEED;
1624 1625
}

1626
static void
1627
i915_gem_object_put_pages_gtt(struct drm_i915_gem_object *obj)
1628
{
1629 1630
	struct sg_page_iter sg_iter;
	int ret;
1631

1632
	BUG_ON(obj->madv == __I915_MADV_PURGED);
1633

C
Chris Wilson 已提交
1634 1635 1636 1637 1638 1639
	ret = i915_gem_object_set_to_cpu_domain(obj, true);
	if (ret) {
		/* In the event of a disaster, abandon all caches and
		 * hope for the best.
		 */
		WARN_ON(ret != -EIO);
1640
		i915_gem_clflush_object(obj, true);
C
Chris Wilson 已提交
1641 1642 1643
		obj->base.read_domains = obj->base.write_domain = I915_GEM_DOMAIN_CPU;
	}

1644
	if (i915_gem_object_needs_bit17_swizzle(obj))
1645 1646
		i915_gem_object_save_bit_17_swizzle(obj);

1647 1648
	if (obj->madv == I915_MADV_DONTNEED)
		obj->dirty = 0;
1649

1650
	for_each_sg_page(obj->pages->sgl, &sg_iter, obj->pages->nents, 0) {
1651
		struct page *page = sg_page_iter_page(&sg_iter);
1652

1653
		if (obj->dirty)
1654
			set_page_dirty(page);
1655

1656
		if (obj->madv == I915_MADV_WILLNEED)
1657
			mark_page_accessed(page);
1658

1659
		page_cache_release(page);
1660
	}
1661
	obj->dirty = 0;
1662

1663 1664
	sg_free_table(obj->pages);
	kfree(obj->pages);
1665
}
C
Chris Wilson 已提交
1666

1667
int
1668 1669 1670 1671
i915_gem_object_put_pages(struct drm_i915_gem_object *obj)
{
	const struct drm_i915_gem_object_ops *ops = obj->ops;

1672
	if (obj->pages == NULL)
1673 1674
		return 0;

1675 1676 1677
	if (obj->pages_pin_count)
		return -EBUSY;

1678
	BUG_ON(i915_gem_obj_bound_any(obj));
B
Ben Widawsky 已提交
1679

1680 1681 1682
	/* ->put_pages might need to allocate memory for the bit17 swizzle
	 * array, hence protect them from being reaped by removing them from gtt
	 * lists early. */
1683
	list_del(&obj->global_list);
1684

1685
	ops->put_pages(obj);
1686
	obj->pages = NULL;
1687

C
Chris Wilson 已提交
1688 1689 1690 1691 1692 1693 1694
	if (i915_gem_object_is_purgeable(obj))
		i915_gem_object_truncate(obj);

	return 0;
}

static long
1695 1696
__i915_gem_shrink(struct drm_i915_private *dev_priv, long target,
		  bool purgeable_only)
C
Chris Wilson 已提交
1697 1698 1699 1700 1701 1702
{
	struct drm_i915_gem_object *obj, *next;
	long count = 0;

	list_for_each_entry_safe(obj, next,
				 &dev_priv->mm.unbound_list,
1703
				 global_list) {
1704
		if ((i915_gem_object_is_purgeable(obj) || !purgeable_only) &&
1705
		    i915_gem_object_put_pages(obj) == 0) {
C
Chris Wilson 已提交
1706 1707 1708 1709 1710 1711
			count += obj->base.size >> PAGE_SHIFT;
			if (count >= target)
				return count;
		}
	}

1712 1713 1714
	list_for_each_entry_safe(obj, next, &dev_priv->mm.bound_list,
				 global_list) {
		struct i915_vma *vma, *v;
1715 1716 1717 1718

		if (!i915_gem_object_is_purgeable(obj) && purgeable_only)
			continue;

1719 1720 1721
		list_for_each_entry_safe(vma, v, &obj->vma_list, vma_link)
			if (i915_vma_unbind(vma))
				break;
1722 1723

		if (!i915_gem_object_put_pages(obj)) {
C
Chris Wilson 已提交
1724 1725 1726 1727 1728 1729 1730 1731 1732
			count += obj->base.size >> PAGE_SHIFT;
			if (count >= target)
				return count;
		}
	}

	return count;
}

1733 1734 1735 1736 1737 1738
static long
i915_gem_purge(struct drm_i915_private *dev_priv, long target)
{
	return __i915_gem_shrink(dev_priv, target, true);
}

C
Chris Wilson 已提交
1739 1740 1741 1742 1743 1744 1745
static void
i915_gem_shrink_all(struct drm_i915_private *dev_priv)
{
	struct drm_i915_gem_object *obj, *next;

	i915_gem_evict_everything(dev_priv->dev);

1746 1747
	list_for_each_entry_safe(obj, next, &dev_priv->mm.unbound_list,
				 global_list)
1748
		i915_gem_object_put_pages(obj);
D
Daniel Vetter 已提交
1749 1750
}

1751
static int
C
Chris Wilson 已提交
1752
i915_gem_object_get_pages_gtt(struct drm_i915_gem_object *obj)
1753
{
C
Chris Wilson 已提交
1754
	struct drm_i915_private *dev_priv = obj->base.dev->dev_private;
1755 1756
	int page_count, i;
	struct address_space *mapping;
1757 1758
	struct sg_table *st;
	struct scatterlist *sg;
1759
	struct sg_page_iter sg_iter;
1760
	struct page *page;
1761
	unsigned long last_pfn = 0;	/* suppress gcc warning */
C
Chris Wilson 已提交
1762
	gfp_t gfp;
1763

C
Chris Wilson 已提交
1764 1765 1766 1767 1768 1769 1770
	/* Assert that the object is not currently in any GPU domain. As it
	 * wasn't in the GTT, there shouldn't be any way it could have been in
	 * a GPU cache
	 */
	BUG_ON(obj->base.read_domains & I915_GEM_GPU_DOMAINS);
	BUG_ON(obj->base.write_domain & I915_GEM_GPU_DOMAINS);

1771 1772 1773 1774
	st = kmalloc(sizeof(*st), GFP_KERNEL);
	if (st == NULL)
		return -ENOMEM;

1775
	page_count = obj->base.size / PAGE_SIZE;
1776 1777
	if (sg_alloc_table(st, page_count, GFP_KERNEL)) {
		kfree(st);
1778
		return -ENOMEM;
1779
	}
1780

1781 1782 1783 1784 1785
	/* Get the list of pages out of our struct file.  They'll be pinned
	 * at this point until we release them.
	 *
	 * Fail silently without starting the shrinker
	 */
A
Al Viro 已提交
1786
	mapping = file_inode(obj->base.filp)->i_mapping;
C
Chris Wilson 已提交
1787
	gfp = mapping_gfp_mask(mapping);
1788
	gfp |= __GFP_NORETRY | __GFP_NOWARN | __GFP_NO_KSWAPD;
C
Chris Wilson 已提交
1789
	gfp &= ~(__GFP_IO | __GFP_WAIT);
1790 1791 1792
	sg = st->sgl;
	st->nents = 0;
	for (i = 0; i < page_count; i++) {
C
Chris Wilson 已提交
1793 1794 1795 1796 1797 1798 1799 1800 1801 1802
		page = shmem_read_mapping_page_gfp(mapping, i, gfp);
		if (IS_ERR(page)) {
			i915_gem_purge(dev_priv, page_count);
			page = shmem_read_mapping_page_gfp(mapping, i, gfp);
		}
		if (IS_ERR(page)) {
			/* We've tried hard to allocate the memory by reaping
			 * our own buffer, now let the real VM do its job and
			 * go down in flames if truly OOM.
			 */
1803
			gfp &= ~(__GFP_NORETRY | __GFP_NOWARN | __GFP_NO_KSWAPD);
C
Chris Wilson 已提交
1804 1805 1806 1807 1808 1809 1810
			gfp |= __GFP_IO | __GFP_WAIT;

			i915_gem_shrink_all(dev_priv);
			page = shmem_read_mapping_page_gfp(mapping, i, gfp);
			if (IS_ERR(page))
				goto err_pages;

1811
			gfp |= __GFP_NORETRY | __GFP_NOWARN | __GFP_NO_KSWAPD;
C
Chris Wilson 已提交
1812 1813
			gfp &= ~(__GFP_IO | __GFP_WAIT);
		}
1814 1815 1816 1817 1818 1819 1820 1821
#ifdef CONFIG_SWIOTLB
		if (swiotlb_nr_tbl()) {
			st->nents++;
			sg_set_page(sg, page, PAGE_SIZE, 0);
			sg = sg_next(sg);
			continue;
		}
#endif
1822 1823 1824 1825 1826 1827 1828 1829 1830
		if (!i || page_to_pfn(page) != last_pfn + 1) {
			if (i)
				sg = sg_next(sg);
			st->nents++;
			sg_set_page(sg, page, PAGE_SIZE, 0);
		} else {
			sg->length += PAGE_SIZE;
		}
		last_pfn = page_to_pfn(page);
1831
	}
1832 1833 1834 1835
#ifdef CONFIG_SWIOTLB
	if (!swiotlb_nr_tbl())
#endif
		sg_mark_end(sg);
1836 1837
	obj->pages = st;

1838
	if (i915_gem_object_needs_bit17_swizzle(obj))
1839 1840 1841 1842 1843
		i915_gem_object_do_bit_17_swizzle(obj);

	return 0;

err_pages:
1844 1845
	sg_mark_end(sg);
	for_each_sg_page(st->sgl, &sg_iter, st->nents, 0)
1846
		page_cache_release(sg_page_iter_page(&sg_iter));
1847 1848
	sg_free_table(st);
	kfree(st);
1849
	return PTR_ERR(page);
1850 1851
}

1852 1853 1854 1855 1856 1857 1858 1859 1860 1861 1862 1863 1864 1865
/* Ensure that the associated pages are gathered from the backing storage
 * and pinned into our object. i915_gem_object_get_pages() may be called
 * multiple times before they are released by a single call to
 * i915_gem_object_put_pages() - once the pages are no longer referenced
 * either as a result of memory pressure (reaping pages under the shrinker)
 * or as the object is itself released.
 */
int
i915_gem_object_get_pages(struct drm_i915_gem_object *obj)
{
	struct drm_i915_private *dev_priv = obj->base.dev->dev_private;
	const struct drm_i915_gem_object_ops *ops = obj->ops;
	int ret;

1866
	if (obj->pages)
1867 1868
		return 0;

1869 1870 1871 1872 1873
	if (obj->madv != I915_MADV_WILLNEED) {
		DRM_ERROR("Attempting to obtain a purgeable object\n");
		return -EINVAL;
	}

1874 1875
	BUG_ON(obj->pages_pin_count);

1876 1877 1878 1879
	ret = ops->get_pages(obj);
	if (ret)
		return ret;

1880
	list_add_tail(&obj->global_list, &dev_priv->mm.unbound_list);
1881
	return 0;
1882 1883
}

1884
void
1885
i915_gem_object_move_to_active(struct drm_i915_gem_object *obj,
1886
			       struct intel_ring_buffer *ring)
1887
{
1888
	struct drm_device *dev = obj->base.dev;
1889
	struct drm_i915_private *dev_priv = dev->dev_private;
1890
	u32 seqno = intel_ring_get_seqno(ring);
1891

1892
	BUG_ON(ring == NULL);
1893 1894 1895 1896
	if (obj->ring != ring && obj->last_write_seqno) {
		/* Keep the seqno relative to the current ring */
		obj->last_write_seqno = seqno;
	}
1897
	obj->ring = ring;
1898 1899

	/* Add a reference if we're newly entering the active list. */
1900 1901 1902
	if (!obj->active) {
		drm_gem_object_reference(&obj->base);
		obj->active = 1;
1903
	}
1904

1905
	list_move_tail(&obj->ring_list, &ring->active_list);
1906

1907
	obj->last_read_seqno = seqno;
1908

1909
	if (obj->fenced_gpu_access) {
1910 1911
		obj->last_fenced_seqno = seqno;

1912 1913 1914 1915 1916 1917 1918 1919
		/* Bump MRU to take account of the delayed flush */
		if (obj->fence_reg != I915_FENCE_REG_NONE) {
			struct drm_i915_fence_reg *reg;

			reg = &dev_priv->fence_regs[obj->fence_reg];
			list_move_tail(&reg->lru_list,
				       &dev_priv->mm.fence_list);
		}
1920 1921 1922 1923 1924
	}
}

static void
i915_gem_object_move_to_inactive(struct drm_i915_gem_object *obj)
1925
{
B
Ben Widawsky 已提交
1926 1927 1928
	struct drm_i915_private *dev_priv = obj->base.dev->dev_private;
	struct i915_address_space *ggtt_vm = &dev_priv->gtt.base;
	struct i915_vma *vma = i915_gem_obj_to_vma(obj, ggtt_vm);
1929

1930
	BUG_ON(obj->base.write_domain & ~I915_GEM_GPU_DOMAINS);
1931
	BUG_ON(!obj->active);
1932

B
Ben Widawsky 已提交
1933
	list_move_tail(&vma->mm_list, &ggtt_vm->inactive_list);
1934

1935
	list_del_init(&obj->ring_list);
1936 1937
	obj->ring = NULL;

1938 1939 1940 1941 1942
	obj->last_read_seqno = 0;
	obj->last_write_seqno = 0;
	obj->base.write_domain = 0;

	obj->last_fenced_seqno = 0;
1943 1944 1945 1946 1947 1948
	obj->fenced_gpu_access = false;

	obj->active = 0;
	drm_gem_object_unreference(&obj->base);

	WARN_ON(i915_verify_lists(dev));
1949
}
1950

1951
static int
1952
i915_gem_init_seqno(struct drm_device *dev, u32 seqno)
1953
{
1954 1955 1956
	struct drm_i915_private *dev_priv = dev->dev_private;
	struct intel_ring_buffer *ring;
	int ret, i, j;
1957

1958
	/* Carefully retire all requests without writing to the rings */
1959
	for_each_ring(ring, dev_priv, i) {
1960 1961 1962
		ret = intel_ring_idle(ring);
		if (ret)
			return ret;
1963 1964
	}
	i915_gem_retire_requests(dev);
1965 1966

	/* Finally reset hw state */
1967
	for_each_ring(ring, dev_priv, i) {
1968
		intel_ring_init_seqno(ring, seqno);
1969

1970 1971 1972
		for (j = 0; j < ARRAY_SIZE(ring->sync_seqno); j++)
			ring->sync_seqno[j] = 0;
	}
1973

1974
	return 0;
1975 1976
}

1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002
int i915_gem_set_seqno(struct drm_device *dev, u32 seqno)
{
	struct drm_i915_private *dev_priv = dev->dev_private;
	int ret;

	if (seqno == 0)
		return -EINVAL;

	/* HWS page needs to be set less than what we
	 * will inject to ring
	 */
	ret = i915_gem_init_seqno(dev, seqno - 1);
	if (ret)
		return ret;

	/* Carefully set the last_seqno value so that wrap
	 * detection still works
	 */
	dev_priv->next_seqno = seqno;
	dev_priv->last_seqno = seqno - 1;
	if (dev_priv->last_seqno == 0)
		dev_priv->last_seqno--;

	return 0;
}

2003 2004
int
i915_gem_get_seqno(struct drm_device *dev, u32 *seqno)
2005
{
2006 2007 2008 2009
	struct drm_i915_private *dev_priv = dev->dev_private;

	/* reserve 0 for non-seqno */
	if (dev_priv->next_seqno == 0) {
2010
		int ret = i915_gem_init_seqno(dev, 0);
2011 2012
		if (ret)
			return ret;
2013

2014 2015
		dev_priv->next_seqno = 1;
	}
2016

2017
	*seqno = dev_priv->last_seqno = dev_priv->next_seqno++;
2018
	return 0;
2019 2020
}

2021 2022
int __i915_add_request(struct intel_ring_buffer *ring,
		       struct drm_file *file,
2023
		       struct drm_i915_gem_object *obj,
2024
		       u32 *out_seqno)
2025
{
C
Chris Wilson 已提交
2026
	drm_i915_private_t *dev_priv = ring->dev->dev_private;
2027
	struct drm_i915_gem_request *request;
2028
	u32 request_ring_position, request_start;
2029
	int was_empty;
2030 2031
	int ret;

2032
	request_start = intel_ring_get_tail(ring);
2033 2034 2035 2036 2037 2038 2039
	/*
	 * Emit any outstanding flushes - execbuf can fail to emit the flush
	 * after having emitted the batchbuffer command. Hence we need to fix
	 * things up similar to emitting the lazy request. The difference here
	 * is that the flush _must_ happen before the next request, no matter
	 * what.
	 */
2040 2041 2042
	ret = intel_ring_flush_all_caches(ring);
	if (ret)
		return ret;
2043

2044 2045
	request = ring->preallocated_lazy_request;
	if (WARN_ON(request == NULL))
2046
		return -ENOMEM;
2047

2048 2049 2050 2051 2052 2053 2054
	/* Record the position of the start of the request so that
	 * should we detect the updated seqno part-way through the
	 * GPU processing the request, we never over-estimate the
	 * position of the head.
	 */
	request_ring_position = intel_ring_get_tail(ring);

2055
	ret = ring->add_request(ring);
2056
	if (ret)
2057
		return ret;
2058

2059
	request->seqno = intel_ring_get_seqno(ring);
2060
	request->ring = ring;
2061
	request->head = request_start;
2062
	request->tail = request_ring_position;
2063 2064 2065 2066 2067 2068 2069

	/* Whilst this request exists, batch_obj will be on the
	 * active_list, and so will hold the active reference. Only when this
	 * request is retired will the the batch_obj be moved onto the
	 * inactive_list and lose its active reference. Hence we do not need
	 * to explicitly hold another reference here.
	 */
2070
	request->batch_obj = obj;
2071

2072 2073 2074 2075
	/* Hold a reference to the current context so that we can inspect
	 * it later in case a hangcheck error event fires.
	 */
	request->ctx = ring->last_context;
2076 2077 2078
	if (request->ctx)
		i915_gem_context_reference(request->ctx);

2079
	request->emitted_jiffies = jiffies;
2080 2081
	was_empty = list_empty(&ring->request_list);
	list_add_tail(&request->list, &ring->request_list);
2082
	request->file_priv = NULL;
2083

C
Chris Wilson 已提交
2084 2085 2086
	if (file) {
		struct drm_i915_file_private *file_priv = file->driver_priv;

2087
		spin_lock(&file_priv->mm.lock);
2088
		request->file_priv = file_priv;
2089
		list_add_tail(&request->client_list,
2090
			      &file_priv->mm.request_list);
2091
		spin_unlock(&file_priv->mm.lock);
2092
	}
2093

2094
	trace_i915_gem_request_add(ring, request->seqno);
2095
	ring->outstanding_lazy_seqno = 0;
2096
	ring->preallocated_lazy_request = NULL;
C
Chris Wilson 已提交
2097

2098
	if (!dev_priv->ums.mm_suspended) {
2099 2100
		i915_queue_hangcheck(ring->dev);

2101
		if (was_empty) {
2102
			queue_delayed_work(dev_priv->wq,
2103 2104
					   &dev_priv->mm.retire_work,
					   round_jiffies_up_relative(HZ));
2105 2106
			intel_mark_busy(dev_priv->dev);
		}
B
Ben Gamari 已提交
2107
	}
2108

2109
	if (out_seqno)
2110
		*out_seqno = request->seqno;
2111
	return 0;
2112 2113
}

2114 2115
static inline void
i915_gem_request_remove_from_client(struct drm_i915_gem_request *request)
2116
{
2117
	struct drm_i915_file_private *file_priv = request->file_priv;
2118

2119 2120
	if (!file_priv)
		return;
C
Chris Wilson 已提交
2121

2122
	spin_lock(&file_priv->mm.lock);
2123 2124 2125 2126
	if (request->file_priv) {
		list_del(&request->client_list);
		request->file_priv = NULL;
	}
2127
	spin_unlock(&file_priv->mm.lock);
2128 2129
}

2130 2131
static bool i915_head_inside_object(u32 acthd, struct drm_i915_gem_object *obj,
				    struct i915_address_space *vm)
2132
{
2133 2134
	if (acthd >= i915_gem_obj_offset(obj, vm) &&
	    acthd < i915_gem_obj_offset(obj, vm) + obj->base.size)
2135 2136 2137 2138 2139 2140 2141 2142 2143 2144 2145 2146 2147 2148 2149 2150 2151 2152 2153 2154 2155 2156
		return true;

	return false;
}

static bool i915_head_inside_request(const u32 acthd_unmasked,
				     const u32 request_start,
				     const u32 request_end)
{
	const u32 acthd = acthd_unmasked & HEAD_ADDR;

	if (request_start < request_end) {
		if (acthd >= request_start && acthd < request_end)
			return true;
	} else if (request_start > request_end) {
		if (acthd >= request_start || acthd < request_end)
			return true;
	}

	return false;
}

2157 2158 2159 2160 2161 2162 2163 2164 2165 2166 2167
static struct i915_address_space *
request_to_vm(struct drm_i915_gem_request *request)
{
	struct drm_i915_private *dev_priv = request->ring->dev->dev_private;
	struct i915_address_space *vm;

	vm = &dev_priv->gtt.base;

	return vm;
}

2168 2169 2170 2171 2172 2173 2174 2175
static bool i915_request_guilty(struct drm_i915_gem_request *request,
				const u32 acthd, bool *inside)
{
	/* There is a possibility that unmasked head address
	 * pointing inside the ring, matches the batch_obj address range.
	 * However this is extremely unlikely.
	 */
	if (request->batch_obj) {
2176 2177
		if (i915_head_inside_object(acthd, request->batch_obj,
					    request_to_vm(request))) {
2178 2179 2180 2181 2182 2183 2184 2185 2186 2187 2188 2189 2190 2191 2192 2193 2194 2195 2196
			*inside = true;
			return true;
		}
	}

	if (i915_head_inside_request(acthd, request->head, request->tail)) {
		*inside = false;
		return true;
	}

	return false;
}

static void i915_set_reset_status(struct intel_ring_buffer *ring,
				  struct drm_i915_gem_request *request,
				  u32 acthd)
{
	struct i915_ctx_hang_stats *hs = NULL;
	bool inside, guilty;
2197
	unsigned long offset = 0;
2198 2199 2200 2201

	/* Innocent until proven guilty */
	guilty = false;

2202 2203 2204 2205
	if (request->batch_obj)
		offset = i915_gem_obj_offset(request->batch_obj,
					     request_to_vm(request));

2206
	if (ring->hangcheck.action != HANGCHECK_WAIT &&
2207
	    i915_request_guilty(request, acthd, &inside)) {
2208
		DRM_ERROR("%s hung %s bo (0x%lx ctx %d) at 0x%x\n",
2209 2210
			  ring->name,
			  inside ? "inside" : "flushing",
2211
			  offset,
2212 2213 2214 2215 2216 2217 2218 2219 2220 2221 2222 2223 2224 2225 2226 2227 2228 2229 2230 2231 2232 2233
			  request->ctx ? request->ctx->id : 0,
			  acthd);

		guilty = true;
	}

	/* If contexts are disabled or this is the default context, use
	 * file_priv->reset_state
	 */
	if (request->ctx && request->ctx->id != DEFAULT_CONTEXT_ID)
		hs = &request->ctx->hang_stats;
	else if (request->file_priv)
		hs = &request->file_priv->hang_stats;

	if (hs) {
		if (guilty)
			hs->batch_active++;
		else
			hs->batch_pending++;
	}
}

2234 2235 2236 2237 2238 2239 2240 2241 2242 2243 2244
static void i915_gem_free_request(struct drm_i915_gem_request *request)
{
	list_del(&request->list);
	i915_gem_request_remove_from_client(request);

	if (request->ctx)
		i915_gem_context_unreference(request->ctx);

	kfree(request);
}

2245 2246
static void i915_gem_reset_ring_lists(struct drm_i915_private *dev_priv,
				      struct intel_ring_buffer *ring)
2247
{
2248 2249 2250 2251 2252 2253
	u32 completed_seqno;
	u32 acthd;

	acthd = intel_ring_get_active_head(ring);
	completed_seqno = ring->get_seqno(ring, false);

2254 2255
	while (!list_empty(&ring->request_list)) {
		struct drm_i915_gem_request *request;
2256

2257 2258 2259
		request = list_first_entry(&ring->request_list,
					   struct drm_i915_gem_request,
					   list);
2260

2261 2262 2263
		if (request->seqno > completed_seqno)
			i915_set_reset_status(ring, request, acthd);

2264
		i915_gem_free_request(request);
2265
	}
2266

2267
	while (!list_empty(&ring->active_list)) {
2268
		struct drm_i915_gem_object *obj;
2269

2270 2271 2272
		obj = list_first_entry(&ring->active_list,
				       struct drm_i915_gem_object,
				       ring_list);
2273

2274
		i915_gem_object_move_to_inactive(obj);
2275 2276 2277
	}
}

2278
void i915_gem_restore_fences(struct drm_device *dev)
2279 2280 2281 2282
{
	struct drm_i915_private *dev_priv = dev->dev_private;
	int i;

2283
	for (i = 0; i < dev_priv->num_fence_regs; i++) {
2284
		struct drm_i915_fence_reg *reg = &dev_priv->fence_regs[i];
2285

2286 2287 2288 2289 2290 2291 2292 2293 2294 2295
		/*
		 * Commit delayed tiling changes if we have an object still
		 * attached to the fence, otherwise just clear the fence.
		 */
		if (reg->obj) {
			i915_gem_object_update_fence(reg->obj, reg,
						     reg->obj->tiling_mode);
		} else {
			i915_gem_write_fence(dev, i, NULL);
		}
2296 2297 2298
	}
}

2299
void i915_gem_reset(struct drm_device *dev)
2300
{
2301
	struct drm_i915_private *dev_priv = dev->dev_private;
2302
	struct intel_ring_buffer *ring;
2303
	int i;
2304

2305 2306
	for_each_ring(ring, dev_priv, i)
		i915_gem_reset_ring_lists(dev_priv, ring);
2307

2308
	i915_gem_restore_fences(dev);
2309 2310 2311 2312 2313
}

/**
 * This function clears the request list as sequence numbers are passed.
 */
2314
void
C
Chris Wilson 已提交
2315
i915_gem_retire_requests_ring(struct intel_ring_buffer *ring)
2316 2317 2318
{
	uint32_t seqno;

C
Chris Wilson 已提交
2319
	if (list_empty(&ring->request_list))
2320 2321
		return;

C
Chris Wilson 已提交
2322
	WARN_ON(i915_verify_lists(ring->dev));
2323

2324
	seqno = ring->get_seqno(ring, true);
2325

2326
	while (!list_empty(&ring->request_list)) {
2327 2328
		struct drm_i915_gem_request *request;

2329
		request = list_first_entry(&ring->request_list,
2330 2331 2332
					   struct drm_i915_gem_request,
					   list);

2333
		if (!i915_seqno_passed(seqno, request->seqno))
2334 2335
			break;

C
Chris Wilson 已提交
2336
		trace_i915_gem_request_retire(ring, request->seqno);
2337 2338 2339 2340 2341 2342
		/* We know the GPU must have read the request to have
		 * sent us the seqno + interrupt, so use the position
		 * of tail of the request to update the last known position
		 * of the GPU head.
		 */
		ring->last_retired_head = request->tail;
2343

2344
		i915_gem_free_request(request);
2345
	}
2346

2347 2348 2349 2350
	/* Move any buffers on the active list that are no longer referenced
	 * by the ringbuffer to the flushing/inactive lists as appropriate.
	 */
	while (!list_empty(&ring->active_list)) {
2351
		struct drm_i915_gem_object *obj;
2352

2353
		obj = list_first_entry(&ring->active_list,
2354 2355
				      struct drm_i915_gem_object,
				      ring_list);
2356

2357
		if (!i915_seqno_passed(seqno, obj->last_read_seqno))
2358
			break;
2359

2360
		i915_gem_object_move_to_inactive(obj);
2361
	}
2362

C
Chris Wilson 已提交
2363 2364
	if (unlikely(ring->trace_irq_seqno &&
		     i915_seqno_passed(seqno, ring->trace_irq_seqno))) {
2365
		ring->irq_put(ring);
C
Chris Wilson 已提交
2366
		ring->trace_irq_seqno = 0;
2367
	}
2368

C
Chris Wilson 已提交
2369
	WARN_ON(i915_verify_lists(ring->dev));
2370 2371
}

2372 2373 2374 2375
void
i915_gem_retire_requests(struct drm_device *dev)
{
	drm_i915_private_t *dev_priv = dev->dev_private;
2376
	struct intel_ring_buffer *ring;
2377
	int i;
2378

2379 2380
	for_each_ring(ring, dev_priv, i)
		i915_gem_retire_requests_ring(ring);
2381 2382
}

2383
static void
2384 2385 2386 2387
i915_gem_retire_work_handler(struct work_struct *work)
{
	drm_i915_private_t *dev_priv;
	struct drm_device *dev;
2388
	struct intel_ring_buffer *ring;
2389 2390
	bool idle;
	int i;
2391 2392 2393 2394 2395

	dev_priv = container_of(work, drm_i915_private_t,
				mm.retire_work.work);
	dev = dev_priv->dev;

2396 2397
	/* Come back later if the device is busy... */
	if (!mutex_trylock(&dev->struct_mutex)) {
2398 2399
		queue_delayed_work(dev_priv->wq, &dev_priv->mm.retire_work,
				   round_jiffies_up_relative(HZ));
2400 2401
		return;
	}
2402

2403
	i915_gem_retire_requests(dev);
2404

2405 2406
	/* Send a periodic flush down the ring so we don't hold onto GEM
	 * objects indefinitely.
2407
	 */
2408
	idle = true;
2409
	for_each_ring(ring, dev_priv, i) {
2410
		if (ring->gpu_caches_dirty)
2411
			i915_add_request(ring, NULL);
2412 2413

		idle &= list_empty(&ring->request_list);
2414 2415
	}

2416
	if (!dev_priv->ums.mm_suspended && !idle)
2417 2418
		queue_delayed_work(dev_priv->wq, &dev_priv->mm.retire_work,
				   round_jiffies_up_relative(HZ));
2419 2420
	if (idle)
		intel_mark_idle(dev);
2421

2422 2423 2424
	mutex_unlock(&dev->struct_mutex);
}

2425 2426 2427 2428 2429 2430 2431 2432 2433 2434 2435
/**
 * Ensures that an object will eventually get non-busy by flushing any required
 * write domains, emitting any outstanding lazy request and retiring and
 * completed requests.
 */
static int
i915_gem_object_flush_active(struct drm_i915_gem_object *obj)
{
	int ret;

	if (obj->active) {
2436
		ret = i915_gem_check_olr(obj->ring, obj->last_read_seqno);
2437 2438 2439 2440 2441 2442 2443 2444 2445
		if (ret)
			return ret;

		i915_gem_retire_requests_ring(obj->ring);
	}

	return 0;
}

2446 2447 2448 2449 2450 2451 2452 2453 2454 2455 2456 2457 2458 2459 2460 2461 2462 2463 2464 2465 2466 2467 2468 2469 2470
/**
 * i915_gem_wait_ioctl - implements DRM_IOCTL_I915_GEM_WAIT
 * @DRM_IOCTL_ARGS: standard ioctl arguments
 *
 * Returns 0 if successful, else an error is returned with the remaining time in
 * the timeout parameter.
 *  -ETIME: object is still busy after timeout
 *  -ERESTARTSYS: signal interrupted the wait
 *  -ENONENT: object doesn't exist
 * Also possible, but rare:
 *  -EAGAIN: GPU wedged
 *  -ENOMEM: damn
 *  -ENODEV: Internal IRQ fail
 *  -E?: The add request failed
 *
 * The wait ioctl with a timeout of 0 reimplements the busy ioctl. With any
 * non-zero timeout parameter the wait ioctl will wait for the given number of
 * nanoseconds on an object becoming unbusy. Since the wait itself does so
 * without holding struct_mutex the object may become re-busied before this
 * function completes. A similar but shorter * race condition exists in the busy
 * ioctl
 */
int
i915_gem_wait_ioctl(struct drm_device *dev, void *data, struct drm_file *file)
{
2471
	drm_i915_private_t *dev_priv = dev->dev_private;
2472 2473 2474
	struct drm_i915_gem_wait *args = data;
	struct drm_i915_gem_object *obj;
	struct intel_ring_buffer *ring = NULL;
2475
	struct timespec timeout_stack, *timeout = NULL;
2476
	unsigned reset_counter;
2477 2478 2479
	u32 seqno = 0;
	int ret = 0;

2480 2481 2482 2483
	if (args->timeout_ns >= 0) {
		timeout_stack = ns_to_timespec(args->timeout_ns);
		timeout = &timeout_stack;
	}
2484 2485 2486 2487 2488 2489 2490 2491 2492 2493 2494

	ret = i915_mutex_lock_interruptible(dev);
	if (ret)
		return ret;

	obj = to_intel_bo(drm_gem_object_lookup(dev, file, args->bo_handle));
	if (&obj->base == NULL) {
		mutex_unlock(&dev->struct_mutex);
		return -ENOENT;
	}

2495 2496
	/* Need to make sure the object gets inactive eventually. */
	ret = i915_gem_object_flush_active(obj);
2497 2498 2499 2500
	if (ret)
		goto out;

	if (obj->active) {
2501
		seqno = obj->last_read_seqno;
2502 2503 2504 2505 2506 2507 2508 2509 2510 2511 2512 2513 2514 2515 2516
		ring = obj->ring;
	}

	if (seqno == 0)
		 goto out;

	/* Do this after OLR check to make sure we make forward progress polling
	 * on this IOCTL with a 0 timeout (like busy ioctl)
	 */
	if (!args->timeout_ns) {
		ret = -ETIME;
		goto out;
	}

	drm_gem_object_unreference(&obj->base);
2517
	reset_counter = atomic_read(&dev_priv->gpu_error.reset_counter);
2518 2519
	mutex_unlock(&dev->struct_mutex);

2520
	ret = __wait_seqno(ring, seqno, reset_counter, true, timeout);
2521
	if (timeout)
2522
		args->timeout_ns = timespec_to_ns(timeout);
2523 2524 2525 2526 2527 2528 2529 2530
	return ret;

out:
	drm_gem_object_unreference(&obj->base);
	mutex_unlock(&dev->struct_mutex);
	return ret;
}

2531 2532 2533 2534 2535 2536 2537 2538 2539 2540 2541 2542
/**
 * i915_gem_object_sync - sync an object to a ring.
 *
 * @obj: object which may be in use on another ring.
 * @to: ring we wish to use the object on. May be NULL.
 *
 * This code is meant to abstract object synchronization with the GPU.
 * Calling with NULL implies synchronizing the object with the CPU
 * rather than a particular GPU ring.
 *
 * Returns 0 if successful, else propagates up the lower layer error.
 */
2543 2544 2545 2546 2547 2548 2549 2550 2551 2552 2553
int
i915_gem_object_sync(struct drm_i915_gem_object *obj,
		     struct intel_ring_buffer *to)
{
	struct intel_ring_buffer *from = obj->ring;
	u32 seqno;
	int ret, idx;

	if (from == NULL || to == from)
		return 0;

2554
	if (to == NULL || !i915_semaphore_is_enabled(obj->base.dev))
2555
		return i915_gem_object_wait_rendering(obj, false);
2556 2557 2558

	idx = intel_ring_sync_index(from, to);

2559
	seqno = obj->last_read_seqno;
2560 2561 2562
	if (seqno <= from->sync_seqno[idx])
		return 0;

2563 2564 2565
	ret = i915_gem_check_olr(obj->ring, seqno);
	if (ret)
		return ret;
2566

2567
	ret = to->sync_to(to, from, seqno);
2568
	if (!ret)
2569 2570 2571 2572 2573
		/* We use last_read_seqno because sync_to()
		 * might have just caused seqno wrap under
		 * the radar.
		 */
		from->sync_seqno[idx] = obj->last_read_seqno;
2574

2575
	return ret;
2576 2577
}

2578 2579 2580 2581 2582 2583 2584
static void i915_gem_object_finish_gtt(struct drm_i915_gem_object *obj)
{
	u32 old_write_domain, old_read_domains;

	/* Force a pagefault for domain tracking on next user access */
	i915_gem_release_mmap(obj);

2585 2586 2587
	if ((obj->base.read_domains & I915_GEM_DOMAIN_GTT) == 0)
		return;

2588 2589 2590
	/* Wait for any direct GTT access to complete */
	mb();

2591 2592 2593 2594 2595 2596 2597 2598 2599 2600 2601
	old_read_domains = obj->base.read_domains;
	old_write_domain = obj->base.write_domain;

	obj->base.read_domains &= ~I915_GEM_DOMAIN_GTT;
	obj->base.write_domain &= ~I915_GEM_DOMAIN_GTT;

	trace_i915_gem_object_change_domain(obj,
					    old_read_domains,
					    old_write_domain);
}

2602
int i915_vma_unbind(struct i915_vma *vma)
2603
{
2604
	struct drm_i915_gem_object *obj = vma->obj;
2605
	drm_i915_private_t *dev_priv = obj->base.dev->dev_private;
2606
	int ret;
2607

2608 2609 2610
	/* For now we only ever use 1 vma per object */
	WARN_ON(!list_is_singular(&obj->vma_list));

2611
	if (list_empty(&vma->vma_link))
2612 2613
		return 0;

2614 2615 2616 2617 2618
	if (!drm_mm_node_allocated(&vma->node)) {
		i915_gem_vma_destroy(vma);

		return 0;
	}
2619

2620 2621
	if (obj->pin_count)
		return -EBUSY;
2622

2623 2624
	BUG_ON(obj->pages == NULL);

2625
	ret = i915_gem_object_finish_gpu(obj);
2626
	if (ret)
2627 2628 2629 2630 2631 2632
		return ret;
	/* Continue on if we fail due to EIO, the GPU is hung so we
	 * should be safe and we need to cleanup or else we might
	 * cause memory corruption through use-after-free.
	 */

2633
	i915_gem_object_finish_gtt(obj);
2634

2635
	/* release the fence reg _after_ flushing */
2636
	ret = i915_gem_object_put_fence(obj);
2637
	if (ret)
2638
		return ret;
2639

2640
	trace_i915_vma_unbind(vma);
C
Chris Wilson 已提交
2641

2642 2643
	if (obj->has_global_gtt_mapping)
		i915_gem_gtt_unbind_object(obj);
2644 2645 2646 2647
	if (obj->has_aliasing_ppgtt_mapping) {
		i915_ppgtt_unbind_object(dev_priv->mm.aliasing_ppgtt, obj);
		obj->has_aliasing_ppgtt_mapping = 0;
	}
2648
	i915_gem_gtt_finish_object(obj);
B
Ben Widawsky 已提交
2649
	i915_gem_object_unpin_pages(obj);
2650

B
Ben Widawsky 已提交
2651
	list_del(&vma->mm_list);
2652
	/* Avoid an unnecessary call to unbind on rebind. */
2653 2654
	if (i915_is_ggtt(vma->vm))
		obj->map_and_fenceable = true;
2655

B
Ben Widawsky 已提交
2656
	drm_mm_remove_node(&vma->node);
2657

B
Ben Widawsky 已提交
2658 2659 2660
	i915_gem_vma_destroy(vma);

	/* Since the unbound list is global, only move to that list if
2661
	 * no more VMAs exist. */
B
Ben Widawsky 已提交
2662 2663
	if (list_empty(&obj->vma_list))
		list_move_tail(&obj->global_list, &dev_priv->mm.unbound_list);
2664

2665
	return 0;
2666 2667
}

2668 2669 2670 2671 2672 2673 2674 2675 2676
/**
 * Unbinds an object from the global GTT aperture.
 */
int
i915_gem_object_ggtt_unbind(struct drm_i915_gem_object *obj)
{
	struct drm_i915_private *dev_priv = obj->base.dev->dev_private;
	struct i915_address_space *ggtt = &dev_priv->gtt.base;

2677
	if (!i915_gem_obj_ggtt_bound(obj))
2678 2679 2680 2681 2682 2683 2684 2685 2686 2687
		return 0;

	if (obj->pin_count)
		return -EBUSY;

	BUG_ON(obj->pages == NULL);

	return i915_vma_unbind(i915_gem_obj_to_vma(obj, ggtt));
}

2688
int i915_gpu_idle(struct drm_device *dev)
2689 2690
{
	drm_i915_private_t *dev_priv = dev->dev_private;
2691
	struct intel_ring_buffer *ring;
2692
	int ret, i;
2693 2694

	/* Flush everything onto the inactive list. */
2695
	for_each_ring(ring, dev_priv, i) {
2696 2697 2698 2699
		ret = i915_switch_context(ring, NULL, DEFAULT_CONTEXT_ID);
		if (ret)
			return ret;

2700
		ret = intel_ring_idle(ring);
2701 2702 2703
		if (ret)
			return ret;
	}
2704

2705
	return 0;
2706 2707
}

2708 2709
static void i965_write_fence_reg(struct drm_device *dev, int reg,
				 struct drm_i915_gem_object *obj)
2710 2711
{
	drm_i915_private_t *dev_priv = dev->dev_private;
2712 2713
	int fence_reg;
	int fence_pitch_shift;
2714

2715 2716 2717 2718 2719 2720 2721 2722
	if (INTEL_INFO(dev)->gen >= 6) {
		fence_reg = FENCE_REG_SANDYBRIDGE_0;
		fence_pitch_shift = SANDYBRIDGE_FENCE_PITCH_SHIFT;
	} else {
		fence_reg = FENCE_REG_965_0;
		fence_pitch_shift = I965_FENCE_PITCH_SHIFT;
	}

2723 2724 2725 2726 2727 2728 2729 2730 2731 2732 2733 2734 2735 2736
	fence_reg += reg * 8;

	/* To w/a incoherency with non-atomic 64-bit register updates,
	 * we split the 64-bit update into two 32-bit writes. In order
	 * for a partial fence not to be evaluated between writes, we
	 * precede the update with write to turn off the fence register,
	 * and only enable the fence as the last step.
	 *
	 * For extra levels of paranoia, we make sure each step lands
	 * before applying the next step.
	 */
	I915_WRITE(fence_reg, 0);
	POSTING_READ(fence_reg);

2737
	if (obj) {
2738
		u32 size = i915_gem_obj_ggtt_size(obj);
2739
		uint64_t val;
2740

2741
		val = (uint64_t)((i915_gem_obj_ggtt_offset(obj) + size - 4096) &
2742
				 0xfffff000) << 32;
2743
		val |= i915_gem_obj_ggtt_offset(obj) & 0xfffff000;
2744
		val |= (uint64_t)((obj->stride / 128) - 1) << fence_pitch_shift;
2745 2746 2747
		if (obj->tiling_mode == I915_TILING_Y)
			val |= 1 << I965_FENCE_TILING_Y_SHIFT;
		val |= I965_FENCE_REG_VALID;
2748

2749 2750 2751 2752 2753 2754 2755 2756 2757
		I915_WRITE(fence_reg + 4, val >> 32);
		POSTING_READ(fence_reg + 4);

		I915_WRITE(fence_reg + 0, val);
		POSTING_READ(fence_reg);
	} else {
		I915_WRITE(fence_reg + 4, 0);
		POSTING_READ(fence_reg + 4);
	}
2758 2759
}

2760 2761
static void i915_write_fence_reg(struct drm_device *dev, int reg,
				 struct drm_i915_gem_object *obj)
2762 2763
{
	drm_i915_private_t *dev_priv = dev->dev_private;
2764
	u32 val;
2765

2766
	if (obj) {
2767
		u32 size = i915_gem_obj_ggtt_size(obj);
2768 2769
		int pitch_val;
		int tile_width;
2770

2771
		WARN((i915_gem_obj_ggtt_offset(obj) & ~I915_FENCE_START_MASK) ||
2772
		     (size & -size) != size ||
2773 2774 2775
		     (i915_gem_obj_ggtt_offset(obj) & (size - 1)),
		     "object 0x%08lx [fenceable? %d] not 1M or pot-size (0x%08x) aligned\n",
		     i915_gem_obj_ggtt_offset(obj), obj->map_and_fenceable, size);
2776

2777 2778 2779 2780 2781 2782 2783 2784 2785
		if (obj->tiling_mode == I915_TILING_Y && HAS_128_BYTE_Y_TILING(dev))
			tile_width = 128;
		else
			tile_width = 512;

		/* Note: pitch better be a power of two tile widths */
		pitch_val = obj->stride / tile_width;
		pitch_val = ffs(pitch_val) - 1;

2786
		val = i915_gem_obj_ggtt_offset(obj);
2787 2788 2789 2790 2791 2792 2793 2794 2795 2796 2797 2798 2799 2800 2801
		if (obj->tiling_mode == I915_TILING_Y)
			val |= 1 << I830_FENCE_TILING_Y_SHIFT;
		val |= I915_FENCE_SIZE_BITS(size);
		val |= pitch_val << I830_FENCE_PITCH_SHIFT;
		val |= I830_FENCE_REG_VALID;
	} else
		val = 0;

	if (reg < 8)
		reg = FENCE_REG_830_0 + reg * 4;
	else
		reg = FENCE_REG_945_8 + (reg - 8) * 4;

	I915_WRITE(reg, val);
	POSTING_READ(reg);
2802 2803
}

2804 2805
static void i830_write_fence_reg(struct drm_device *dev, int reg,
				struct drm_i915_gem_object *obj)
2806 2807 2808 2809
{
	drm_i915_private_t *dev_priv = dev->dev_private;
	uint32_t val;

2810
	if (obj) {
2811
		u32 size = i915_gem_obj_ggtt_size(obj);
2812
		uint32_t pitch_val;
2813

2814
		WARN((i915_gem_obj_ggtt_offset(obj) & ~I830_FENCE_START_MASK) ||
2815
		     (size & -size) != size ||
2816 2817 2818
		     (i915_gem_obj_ggtt_offset(obj) & (size - 1)),
		     "object 0x%08lx not 512K or pot-size 0x%08x aligned\n",
		     i915_gem_obj_ggtt_offset(obj), size);
2819

2820 2821
		pitch_val = obj->stride / 128;
		pitch_val = ffs(pitch_val) - 1;
2822

2823
		val = i915_gem_obj_ggtt_offset(obj);
2824 2825 2826 2827 2828 2829 2830
		if (obj->tiling_mode == I915_TILING_Y)
			val |= 1 << I830_FENCE_TILING_Y_SHIFT;
		val |= I830_FENCE_SIZE_BITS(size);
		val |= pitch_val << I830_FENCE_PITCH_SHIFT;
		val |= I830_FENCE_REG_VALID;
	} else
		val = 0;
2831

2832 2833 2834 2835
	I915_WRITE(FENCE_REG_830_0 + reg * 4, val);
	POSTING_READ(FENCE_REG_830_0 + reg * 4);
}

2836 2837 2838 2839 2840
inline static bool i915_gem_object_needs_mb(struct drm_i915_gem_object *obj)
{
	return obj && obj->base.read_domains & I915_GEM_DOMAIN_GTT;
}

2841 2842 2843
static void i915_gem_write_fence(struct drm_device *dev, int reg,
				 struct drm_i915_gem_object *obj)
{
2844 2845 2846 2847 2848 2849 2850 2851
	struct drm_i915_private *dev_priv = dev->dev_private;

	/* Ensure that all CPU reads are completed before installing a fence
	 * and all writes before removing the fence.
	 */
	if (i915_gem_object_needs_mb(dev_priv->fence_regs[reg].obj))
		mb();

2852 2853 2854 2855
	WARN(obj && (!obj->stride || !obj->tiling_mode),
	     "bogus fence setup with stride: 0x%x, tiling mode: %i\n",
	     obj->stride, obj->tiling_mode);

2856 2857
	switch (INTEL_INFO(dev)->gen) {
	case 7:
2858
	case 6:
2859 2860 2861 2862
	case 5:
	case 4: i965_write_fence_reg(dev, reg, obj); break;
	case 3: i915_write_fence_reg(dev, reg, obj); break;
	case 2: i830_write_fence_reg(dev, reg, obj); break;
2863
	default: BUG();
2864
	}
2865 2866 2867 2868 2869 2870

	/* And similarly be paranoid that no direct access to this region
	 * is reordered to before the fence is installed.
	 */
	if (i915_gem_object_needs_mb(obj))
		mb();
2871 2872
}

2873 2874 2875 2876 2877 2878 2879 2880 2881 2882
static inline int fence_number(struct drm_i915_private *dev_priv,
			       struct drm_i915_fence_reg *fence)
{
	return fence - dev_priv->fence_regs;
}

static void i915_gem_object_update_fence(struct drm_i915_gem_object *obj,
					 struct drm_i915_fence_reg *fence,
					 bool enable)
{
2883
	struct drm_i915_private *dev_priv = obj->base.dev->dev_private;
2884 2885 2886
	int reg = fence_number(dev_priv, fence);

	i915_gem_write_fence(obj->base.dev, reg, enable ? obj : NULL);
2887 2888

	if (enable) {
2889
		obj->fence_reg = reg;
2890 2891 2892 2893 2894 2895 2896
		fence->obj = obj;
		list_move_tail(&fence->lru_list, &dev_priv->mm.fence_list);
	} else {
		obj->fence_reg = I915_FENCE_REG_NONE;
		fence->obj = NULL;
		list_del_init(&fence->lru_list);
	}
2897
	obj->fence_dirty = false;
2898 2899
}

2900
static int
2901
i915_gem_object_wait_fence(struct drm_i915_gem_object *obj)
2902
{
2903
	if (obj->last_fenced_seqno) {
2904
		int ret = i915_wait_seqno(obj->ring, obj->last_fenced_seqno);
2905 2906
		if (ret)
			return ret;
2907 2908 2909 2910

		obj->last_fenced_seqno = 0;
	}

2911
	obj->fenced_gpu_access = false;
2912 2913 2914 2915 2916 2917
	return 0;
}

int
i915_gem_object_put_fence(struct drm_i915_gem_object *obj)
{
2918
	struct drm_i915_private *dev_priv = obj->base.dev->dev_private;
2919
	struct drm_i915_fence_reg *fence;
2920 2921
	int ret;

2922
	ret = i915_gem_object_wait_fence(obj);
2923 2924 2925
	if (ret)
		return ret;

2926 2927
	if (obj->fence_reg == I915_FENCE_REG_NONE)
		return 0;
2928

2929 2930
	fence = &dev_priv->fence_regs[obj->fence_reg];

2931
	i915_gem_object_fence_lost(obj);
2932
	i915_gem_object_update_fence(obj, fence, false);
2933 2934 2935 2936 2937

	return 0;
}

static struct drm_i915_fence_reg *
C
Chris Wilson 已提交
2938
i915_find_fence_reg(struct drm_device *dev)
2939 2940
{
	struct drm_i915_private *dev_priv = dev->dev_private;
C
Chris Wilson 已提交
2941
	struct drm_i915_fence_reg *reg, *avail;
2942
	int i;
2943 2944

	/* First try to find a free reg */
2945
	avail = NULL;
2946 2947 2948
	for (i = dev_priv->fence_reg_start; i < dev_priv->num_fence_regs; i++) {
		reg = &dev_priv->fence_regs[i];
		if (!reg->obj)
2949
			return reg;
2950

2951
		if (!reg->pin_count)
2952
			avail = reg;
2953 2954
	}

2955 2956
	if (avail == NULL)
		return NULL;
2957 2958

	/* None available, try to steal one or wait for a user to finish */
2959
	list_for_each_entry(reg, &dev_priv->mm.fence_list, lru_list) {
2960
		if (reg->pin_count)
2961 2962
			continue;

C
Chris Wilson 已提交
2963
		return reg;
2964 2965
	}

C
Chris Wilson 已提交
2966
	return NULL;
2967 2968
}

2969
/**
2970
 * i915_gem_object_get_fence - set up fencing for an object
2971 2972 2973 2974 2975 2976 2977 2978 2979
 * @obj: object to map through a fence reg
 *
 * When mapping objects through the GTT, userspace wants to be able to write
 * to them without having to worry about swizzling if the object is tiled.
 * This function walks the fence regs looking for a free one for @obj,
 * stealing one if it can't find any.
 *
 * It then sets up the reg based on the object's properties: address, pitch
 * and tiling format.
2980 2981
 *
 * For an untiled surface, this removes any existing fence.
2982
 */
2983
int
2984
i915_gem_object_get_fence(struct drm_i915_gem_object *obj)
2985
{
2986
	struct drm_device *dev = obj->base.dev;
J
Jesse Barnes 已提交
2987
	struct drm_i915_private *dev_priv = dev->dev_private;
2988
	bool enable = obj->tiling_mode != I915_TILING_NONE;
2989
	struct drm_i915_fence_reg *reg;
2990
	int ret;
2991

2992 2993 2994
	/* Have we updated the tiling parameters upon the object and so
	 * will need to serialise the write to the associated fence register?
	 */
2995
	if (obj->fence_dirty) {
2996
		ret = i915_gem_object_wait_fence(obj);
2997 2998 2999
		if (ret)
			return ret;
	}
3000

3001
	/* Just update our place in the LRU if our fence is getting reused. */
3002 3003
	if (obj->fence_reg != I915_FENCE_REG_NONE) {
		reg = &dev_priv->fence_regs[obj->fence_reg];
3004
		if (!obj->fence_dirty) {
3005 3006 3007 3008 3009 3010 3011 3012
			list_move_tail(&reg->lru_list,
				       &dev_priv->mm.fence_list);
			return 0;
		}
	} else if (enable) {
		reg = i915_find_fence_reg(dev);
		if (reg == NULL)
			return -EDEADLK;
3013

3014 3015 3016
		if (reg->obj) {
			struct drm_i915_gem_object *old = reg->obj;

3017
			ret = i915_gem_object_wait_fence(old);
3018 3019 3020
			if (ret)
				return ret;

3021
			i915_gem_object_fence_lost(old);
3022
		}
3023
	} else
3024 3025
		return 0;

3026 3027
	i915_gem_object_update_fence(obj, reg, enable);

3028
	return 0;
3029 3030
}

3031 3032 3033 3034 3035 3036 3037 3038
static bool i915_gem_valid_gtt_space(struct drm_device *dev,
				     struct drm_mm_node *gtt_space,
				     unsigned long cache_level)
{
	struct drm_mm_node *other;

	/* On non-LLC machines we have to be careful when putting differing
	 * types of snoopable memory together to avoid the prefetcher
3039
	 * crossing memory domains and dying.
3040 3041 3042 3043
	 */
	if (HAS_LLC(dev))
		return true;

3044
	if (!drm_mm_node_allocated(gtt_space))
3045 3046 3047 3048 3049 3050 3051 3052 3053 3054 3055 3056 3057 3058 3059 3060 3061 3062 3063 3064 3065 3066 3067
		return true;

	if (list_empty(&gtt_space->node_list))
		return true;

	other = list_entry(gtt_space->node_list.prev, struct drm_mm_node, node_list);
	if (other->allocated && !other->hole_follows && other->color != cache_level)
		return false;

	other = list_entry(gtt_space->node_list.next, struct drm_mm_node, node_list);
	if (other->allocated && !gtt_space->hole_follows && other->color != cache_level)
		return false;

	return true;
}

static void i915_gem_verify_gtt(struct drm_device *dev)
{
#if WATCH_GTT
	struct drm_i915_private *dev_priv = dev->dev_private;
	struct drm_i915_gem_object *obj;
	int err = 0;

3068
	list_for_each_entry(obj, &dev_priv->mm.gtt_list, global_list) {
3069 3070 3071 3072 3073 3074 3075 3076
		if (obj->gtt_space == NULL) {
			printk(KERN_ERR "object found on GTT list with no space reserved\n");
			err++;
			continue;
		}

		if (obj->cache_level != obj->gtt_space->color) {
			printk(KERN_ERR "object reserved space [%08lx, %08lx] with wrong color, cache_level=%x, color=%lx\n",
3077 3078
			       i915_gem_obj_ggtt_offset(obj),
			       i915_gem_obj_ggtt_offset(obj) + i915_gem_obj_ggtt_size(obj),
3079 3080 3081 3082 3083 3084 3085 3086 3087 3088
			       obj->cache_level,
			       obj->gtt_space->color);
			err++;
			continue;
		}

		if (!i915_gem_valid_gtt_space(dev,
					      obj->gtt_space,
					      obj->cache_level)) {
			printk(KERN_ERR "invalid GTT space found at [%08lx, %08lx] - color=%x\n",
3089 3090
			       i915_gem_obj_ggtt_offset(obj),
			       i915_gem_obj_ggtt_offset(obj) + i915_gem_obj_ggtt_size(obj),
3091 3092 3093 3094 3095 3096 3097 3098 3099 3100
			       obj->cache_level);
			err++;
			continue;
		}
	}

	WARN_ON(err);
#endif
}

3101 3102 3103 3104
/**
 * Finds free space in the GTT aperture and binds the object there.
 */
static int
3105 3106 3107 3108 3109
i915_gem_object_bind_to_vm(struct drm_i915_gem_object *obj,
			   struct i915_address_space *vm,
			   unsigned alignment,
			   bool map_and_fenceable,
			   bool nonblocking)
3110
{
3111
	struct drm_device *dev = obj->base.dev;
3112
	drm_i915_private_t *dev_priv = dev->dev_private;
3113
	u32 size, fence_size, fence_alignment, unfenced_alignment;
3114 3115
	size_t gtt_max =
		map_and_fenceable ? dev_priv->gtt.mappable_end : vm->total;
B
Ben Widawsky 已提交
3116
	struct i915_vma *vma;
3117
	int ret;
3118

3119 3120 3121 3122 3123
	fence_size = i915_gem_get_gtt_size(dev,
					   obj->base.size,
					   obj->tiling_mode);
	fence_alignment = i915_gem_get_gtt_alignment(dev,
						     obj->base.size,
3124
						     obj->tiling_mode, true);
3125
	unfenced_alignment =
3126
		i915_gem_get_gtt_alignment(dev,
3127
						    obj->base.size,
3128
						    obj->tiling_mode, false);
3129

3130
	if (alignment == 0)
3131 3132
		alignment = map_and_fenceable ? fence_alignment :
						unfenced_alignment;
3133
	if (map_and_fenceable && alignment & (fence_alignment - 1)) {
3134 3135 3136 3137
		DRM_ERROR("Invalid object alignment requested %u\n", alignment);
		return -EINVAL;
	}

3138
	size = map_and_fenceable ? fence_size : obj->base.size;
3139

3140 3141 3142
	/* If the object is bigger than the entire aperture, reject it early
	 * before evicting everything in a vain attempt to find space.
	 */
3143
	if (obj->base.size > gtt_max) {
3144
		DRM_ERROR("Attempting to bind an object larger than the aperture: object=%zd > %s aperture=%zu\n",
3145 3146
			  obj->base.size,
			  map_and_fenceable ? "mappable" : "total",
3147
			  gtt_max);
3148 3149 3150
		return -E2BIG;
	}

3151
	ret = i915_gem_object_get_pages(obj);
C
Chris Wilson 已提交
3152 3153 3154
	if (ret)
		return ret;

3155 3156
	i915_gem_object_pin_pages(obj);

3157 3158
	BUG_ON(!i915_is_ggtt(vm));

3159
	vma = i915_gem_obj_lookup_or_create_vma(obj, vm);
3160
	if (IS_ERR(vma)) {
3161 3162
		ret = PTR_ERR(vma);
		goto err_unpin;
B
Ben Widawsky 已提交
3163 3164
	}

3165 3166 3167
	/* For now we only ever use 1 vma per object */
	WARN_ON(!list_is_singular(&obj->vma_list));

3168
search_free:
3169
	ret = drm_mm_insert_node_in_range_generic(&vm->mm, &vma->node,
3170
						  size, alignment,
3171 3172
						  obj->cache_level, 0, gtt_max,
						  DRM_MM_SEARCH_DEFAULT);
3173
	if (ret) {
3174
		ret = i915_gem_evict_something(dev, vm, size, alignment,
3175
					       obj->cache_level,
3176 3177
					       map_and_fenceable,
					       nonblocking);
3178 3179
		if (ret == 0)
			goto search_free;
3180

3181
		goto err_free_vma;
3182
	}
B
Ben Widawsky 已提交
3183
	if (WARN_ON(!i915_gem_valid_gtt_space(dev, &vma->node,
3184
					      obj->cache_level))) {
B
Ben Widawsky 已提交
3185
		ret = -EINVAL;
3186
		goto err_remove_node;
3187 3188
	}

3189
	ret = i915_gem_gtt_prepare_object(obj);
B
Ben Widawsky 已提交
3190
	if (ret)
3191
		goto err_remove_node;
3192

3193
	list_move_tail(&obj->global_list, &dev_priv->mm.bound_list);
B
Ben Widawsky 已提交
3194
	list_add_tail(&vma->mm_list, &vm->inactive_list);
3195

3196 3197
	if (i915_is_ggtt(vm)) {
		bool mappable, fenceable;
3198

3199 3200
		fenceable = (vma->node.size == fence_size &&
			     (vma->node.start & (fence_alignment - 1)) == 0);
3201

3202 3203
		mappable = (vma->node.start + obj->base.size <=
			    dev_priv->gtt.mappable_end);
3204

3205
		obj->map_and_fenceable = mappable && fenceable;
3206
	}
3207

3208
	WARN_ON(map_and_fenceable && !obj->map_and_fenceable);
3209

3210
	trace_i915_vma_bind(vma, map_and_fenceable);
3211
	i915_gem_verify_gtt(dev);
3212
	return 0;
B
Ben Widawsky 已提交
3213

3214
err_remove_node:
3215
	drm_mm_remove_node(&vma->node);
3216
err_free_vma:
B
Ben Widawsky 已提交
3217
	i915_gem_vma_destroy(vma);
3218
err_unpin:
B
Ben Widawsky 已提交
3219 3220
	i915_gem_object_unpin_pages(obj);
	return ret;
3221 3222
}

3223
bool
3224 3225
i915_gem_clflush_object(struct drm_i915_gem_object *obj,
			bool force)
3226 3227 3228 3229 3230
{
	/* If we don't have a page list set up, then we're not pinned
	 * to GPU, and we can ignore the cache flush because it'll happen
	 * again at bind time.
	 */
3231
	if (obj->pages == NULL)
3232
		return false;
3233

3234 3235 3236 3237 3238
	/*
	 * Stolen memory is always coherent with the GPU as it is explicitly
	 * marked as wc by the system, or the system is cache-coherent.
	 */
	if (obj->stolen)
3239
		return false;
3240

3241 3242 3243 3244 3245 3246 3247 3248
	/* If the GPU is snooping the contents of the CPU cache,
	 * we do not need to manually clear the CPU cache lines.  However,
	 * the caches are only snooped when the render cache is
	 * flushed/invalidated.  As we always have to emit invalidations
	 * and flushes when moving into and out of the RENDER domain, correct
	 * snooping behaviour occurs naturally as the result of our domain
	 * tracking.
	 */
3249
	if (!force && cpu_cache_is_coherent(obj->base.dev, obj->cache_level))
3250
		return false;
3251

C
Chris Wilson 已提交
3252
	trace_i915_gem_object_clflush(obj);
3253
	drm_clflush_sg(obj->pages);
3254 3255

	return true;
3256 3257 3258 3259
}

/** Flushes the GTT write domain for the object if it's dirty. */
static void
3260
i915_gem_object_flush_gtt_write_domain(struct drm_i915_gem_object *obj)
3261
{
C
Chris Wilson 已提交
3262 3263
	uint32_t old_write_domain;

3264
	if (obj->base.write_domain != I915_GEM_DOMAIN_GTT)
3265 3266
		return;

3267
	/* No actual flushing is required for the GTT write domain.  Writes
3268 3269
	 * to it immediately go to main memory as far as we know, so there's
	 * no chipset flush.  It also doesn't land in render cache.
3270 3271 3272 3273
	 *
	 * However, we do have to enforce the order so that all writes through
	 * the GTT land before any writes to the device, such as updates to
	 * the GATT itself.
3274
	 */
3275 3276
	wmb();

3277 3278
	old_write_domain = obj->base.write_domain;
	obj->base.write_domain = 0;
C
Chris Wilson 已提交
3279 3280

	trace_i915_gem_object_change_domain(obj,
3281
					    obj->base.read_domains,
C
Chris Wilson 已提交
3282
					    old_write_domain);
3283 3284 3285 3286
}

/** Flushes the CPU write domain for the object if it's dirty. */
static void
3287 3288
i915_gem_object_flush_cpu_write_domain(struct drm_i915_gem_object *obj,
				       bool force)
3289
{
C
Chris Wilson 已提交
3290
	uint32_t old_write_domain;
3291

3292
	if (obj->base.write_domain != I915_GEM_DOMAIN_CPU)
3293 3294
		return;

3295 3296 3297
	if (i915_gem_clflush_object(obj, force))
		i915_gem_chipset_flush(obj->base.dev);

3298 3299
	old_write_domain = obj->base.write_domain;
	obj->base.write_domain = 0;
C
Chris Wilson 已提交
3300 3301

	trace_i915_gem_object_change_domain(obj,
3302
					    obj->base.read_domains,
C
Chris Wilson 已提交
3303
					    old_write_domain);
3304 3305
}

3306 3307 3308 3309 3310 3311
/**
 * Moves a single object to the GTT read, and possibly write domain.
 *
 * This function returns when the move is complete, including waiting on
 * flushes to occur.
 */
J
Jesse Barnes 已提交
3312
int
3313
i915_gem_object_set_to_gtt_domain(struct drm_i915_gem_object *obj, bool write)
3314
{
3315
	drm_i915_private_t *dev_priv = obj->base.dev->dev_private;
C
Chris Wilson 已提交
3316
	uint32_t old_write_domain, old_read_domains;
3317
	int ret;
3318

3319
	/* Not valid to be called on unbound objects. */
3320
	if (!i915_gem_obj_bound_any(obj))
3321 3322
		return -EINVAL;

3323 3324 3325
	if (obj->base.write_domain == I915_GEM_DOMAIN_GTT)
		return 0;

3326
	ret = i915_gem_object_wait_rendering(obj, !write);
3327 3328 3329
	if (ret)
		return ret;

3330
	i915_gem_object_flush_cpu_write_domain(obj, false);
C
Chris Wilson 已提交
3331

3332 3333 3334 3335 3336 3337 3338
	/* Serialise direct access to this object with the barriers for
	 * coherent writes from the GPU, by effectively invalidating the
	 * GTT domain upon first access.
	 */
	if ((obj->base.read_domains & I915_GEM_DOMAIN_GTT) == 0)
		mb();

3339 3340
	old_write_domain = obj->base.write_domain;
	old_read_domains = obj->base.read_domains;
C
Chris Wilson 已提交
3341

3342 3343 3344
	/* It should now be out of any other write domains, and we can update
	 * the domain values for our changes.
	 */
3345 3346
	BUG_ON((obj->base.write_domain & ~I915_GEM_DOMAIN_GTT) != 0);
	obj->base.read_domains |= I915_GEM_DOMAIN_GTT;
3347
	if (write) {
3348 3349 3350
		obj->base.read_domains = I915_GEM_DOMAIN_GTT;
		obj->base.write_domain = I915_GEM_DOMAIN_GTT;
		obj->dirty = 1;
3351 3352
	}

C
Chris Wilson 已提交
3353 3354 3355 3356
	trace_i915_gem_object_change_domain(obj,
					    old_read_domains,
					    old_write_domain);

3357
	/* And bump the LRU for this access */
B
Ben Widawsky 已提交
3358 3359 3360 3361 3362 3363 3364 3365
	if (i915_gem_object_is_inactive(obj)) {
		struct i915_vma *vma = i915_gem_obj_to_vma(obj,
							   &dev_priv->gtt.base);
		if (vma)
			list_move_tail(&vma->mm_list,
				       &dev_priv->gtt.base.inactive_list);

	}
3366

3367 3368 3369
	return 0;
}

3370 3371 3372
int i915_gem_object_set_cache_level(struct drm_i915_gem_object *obj,
				    enum i915_cache_level cache_level)
{
3373 3374
	struct drm_device *dev = obj->base.dev;
	drm_i915_private_t *dev_priv = dev->dev_private;
3375
	struct i915_vma *vma;
3376 3377 3378 3379 3380 3381 3382 3383 3384 3385
	int ret;

	if (obj->cache_level == cache_level)
		return 0;

	if (obj->pin_count) {
		DRM_DEBUG("can not change the cache level of pinned objects\n");
		return -EBUSY;
	}

3386 3387
	list_for_each_entry(vma, &obj->vma_list, vma_link) {
		if (!i915_gem_valid_gtt_space(dev, &vma->node, cache_level)) {
3388
			ret = i915_vma_unbind(vma);
3389 3390 3391 3392 3393
			if (ret)
				return ret;

			break;
		}
3394 3395
	}

3396
	if (i915_gem_obj_bound_any(obj)) {
3397 3398 3399 3400 3401 3402 3403 3404 3405 3406
		ret = i915_gem_object_finish_gpu(obj);
		if (ret)
			return ret;

		i915_gem_object_finish_gtt(obj);

		/* Before SandyBridge, you could not use tiling or fence
		 * registers with snooped memory, so relinquish any fences
		 * currently pointing to our region in the aperture.
		 */
3407
		if (INTEL_INFO(dev)->gen < 6) {
3408 3409 3410 3411 3412
			ret = i915_gem_object_put_fence(obj);
			if (ret)
				return ret;
		}

3413 3414
		if (obj->has_global_gtt_mapping)
			i915_gem_gtt_bind_object(obj, cache_level);
3415 3416 3417
		if (obj->has_aliasing_ppgtt_mapping)
			i915_ppgtt_bind_object(dev_priv->mm.aliasing_ppgtt,
					       obj, cache_level);
3418 3419
	}

3420 3421 3422 3423 3424
	list_for_each_entry(vma, &obj->vma_list, vma_link)
		vma->node.color = cache_level;
	obj->cache_level = cache_level;

	if (cpu_write_needs_clflush(obj)) {
3425 3426 3427 3428 3429 3430 3431 3432 3433 3434 3435 3436 3437 3438 3439 3440 3441 3442 3443 3444 3445
		u32 old_read_domains, old_write_domain;

		/* If we're coming from LLC cached, then we haven't
		 * actually been tracking whether the data is in the
		 * CPU cache or not, since we only allow one bit set
		 * in obj->write_domain and have been skipping the clflushes.
		 * Just set it to the CPU cache for now.
		 */
		WARN_ON(obj->base.write_domain & ~I915_GEM_DOMAIN_CPU);

		old_read_domains = obj->base.read_domains;
		old_write_domain = obj->base.write_domain;

		obj->base.read_domains = I915_GEM_DOMAIN_CPU;
		obj->base.write_domain = I915_GEM_DOMAIN_CPU;

		trace_i915_gem_object_change_domain(obj,
						    old_read_domains,
						    old_write_domain);
	}

3446
	i915_gem_verify_gtt(dev);
3447 3448 3449
	return 0;
}

B
Ben Widawsky 已提交
3450 3451
int i915_gem_get_caching_ioctl(struct drm_device *dev, void *data,
			       struct drm_file *file)
3452
{
B
Ben Widawsky 已提交
3453
	struct drm_i915_gem_caching *args = data;
3454 3455 3456 3457 3458 3459 3460 3461 3462 3463 3464 3465 3466
	struct drm_i915_gem_object *obj;
	int ret;

	ret = i915_mutex_lock_interruptible(dev);
	if (ret)
		return ret;

	obj = to_intel_bo(drm_gem_object_lookup(dev, file, args->handle));
	if (&obj->base == NULL) {
		ret = -ENOENT;
		goto unlock;
	}

3467 3468 3469 3470 3471 3472
	switch (obj->cache_level) {
	case I915_CACHE_LLC:
	case I915_CACHE_L3_LLC:
		args->caching = I915_CACHING_CACHED;
		break;

3473 3474 3475 3476
	case I915_CACHE_WT:
		args->caching = I915_CACHING_DISPLAY;
		break;

3477 3478 3479 3480
	default:
		args->caching = I915_CACHING_NONE;
		break;
	}
3481 3482 3483 3484 3485 3486 3487

	drm_gem_object_unreference(&obj->base);
unlock:
	mutex_unlock(&dev->struct_mutex);
	return ret;
}

B
Ben Widawsky 已提交
3488 3489
int i915_gem_set_caching_ioctl(struct drm_device *dev, void *data,
			       struct drm_file *file)
3490
{
B
Ben Widawsky 已提交
3491
	struct drm_i915_gem_caching *args = data;
3492 3493 3494 3495
	struct drm_i915_gem_object *obj;
	enum i915_cache_level level;
	int ret;

B
Ben Widawsky 已提交
3496 3497
	switch (args->caching) {
	case I915_CACHING_NONE:
3498 3499
		level = I915_CACHE_NONE;
		break;
B
Ben Widawsky 已提交
3500
	case I915_CACHING_CACHED:
3501 3502
		level = I915_CACHE_LLC;
		break;
3503 3504 3505
	case I915_CACHING_DISPLAY:
		level = HAS_WT(dev) ? I915_CACHE_WT : I915_CACHE_NONE;
		break;
3506 3507 3508 3509
	default:
		return -EINVAL;
	}

B
Ben Widawsky 已提交
3510 3511 3512 3513
	ret = i915_mutex_lock_interruptible(dev);
	if (ret)
		return ret;

3514 3515 3516 3517 3518 3519 3520 3521 3522 3523 3524 3525 3526 3527
	obj = to_intel_bo(drm_gem_object_lookup(dev, file, args->handle));
	if (&obj->base == NULL) {
		ret = -ENOENT;
		goto unlock;
	}

	ret = i915_gem_object_set_cache_level(obj, level);

	drm_gem_object_unreference(&obj->base);
unlock:
	mutex_unlock(&dev->struct_mutex);
	return ret;
}

3528 3529 3530 3531 3532 3533 3534 3535 3536 3537 3538 3539 3540 3541 3542 3543
static bool is_pin_display(struct drm_i915_gem_object *obj)
{
	/* There are 3 sources that pin objects:
	 *   1. The display engine (scanouts, sprites, cursors);
	 *   2. Reservations for execbuffer;
	 *   3. The user.
	 *
	 * We can ignore reservations as we hold the struct_mutex and
	 * are only called outside of the reservation path.  The user
	 * can only increment pin_count once, and so if after
	 * subtracting the potential reference by the user, any pin_count
	 * remains, it must be due to another use by the display engine.
	 */
	return obj->pin_count - !!obj->user_pin_count;
}

3544
/*
3545 3546 3547
 * Prepare buffer for display plane (scanout, cursors, etc).
 * Can be called from an uninterruptible phase (modesetting) and allows
 * any flushes to be pipelined (for pageflips).
3548 3549
 */
int
3550 3551
i915_gem_object_pin_to_display_plane(struct drm_i915_gem_object *obj,
				     u32 alignment,
3552
				     struct intel_ring_buffer *pipelined)
3553
{
3554
	u32 old_read_domains, old_write_domain;
3555 3556
	int ret;

3557
	if (pipelined != obj->ring) {
3558 3559
		ret = i915_gem_object_sync(obj, pipelined);
		if (ret)
3560 3561 3562
			return ret;
	}

3563 3564 3565 3566 3567
	/* Mark the pin_display early so that we account for the
	 * display coherency whilst setting up the cache domains.
	 */
	obj->pin_display = true;

3568 3569 3570 3571 3572 3573 3574 3575 3576
	/* The display engine is not coherent with the LLC cache on gen6.  As
	 * a result, we make sure that the pinning that is about to occur is
	 * done with uncached PTEs. This is lowest common denominator for all
	 * chipsets.
	 *
	 * However for gen6+, we could do better by using the GFDT bit instead
	 * of uncaching, which would allow us to flush all the LLC-cached data
	 * with that bit in the PTE to main memory with just one PIPE_CONTROL.
	 */
3577 3578
	ret = i915_gem_object_set_cache_level(obj,
					      HAS_WT(obj->base.dev) ? I915_CACHE_WT : I915_CACHE_NONE);
3579
	if (ret)
3580
		goto err_unpin_display;
3581

3582 3583 3584 3585
	/* As the user may map the buffer once pinned in the display plane
	 * (e.g. libkms for the bootup splash), we have to ensure that we
	 * always use map_and_fenceable for all scanout buffers.
	 */
B
Ben Widawsky 已提交
3586
	ret = i915_gem_obj_ggtt_pin(obj, alignment, true, false);
3587
	if (ret)
3588
		goto err_unpin_display;
3589

3590
	i915_gem_object_flush_cpu_write_domain(obj, true);
3591

3592
	old_write_domain = obj->base.write_domain;
3593
	old_read_domains = obj->base.read_domains;
3594 3595 3596 3597

	/* It should now be out of any other write domains, and we can update
	 * the domain values for our changes.
	 */
3598
	obj->base.write_domain = 0;
3599
	obj->base.read_domains |= I915_GEM_DOMAIN_GTT;
3600 3601 3602

	trace_i915_gem_object_change_domain(obj,
					    old_read_domains,
3603
					    old_write_domain);
3604 3605

	return 0;
3606 3607 3608 3609 3610 3611 3612 3613 3614 3615 3616

err_unpin_display:
	obj->pin_display = is_pin_display(obj);
	return ret;
}

void
i915_gem_object_unpin_from_display_plane(struct drm_i915_gem_object *obj)
{
	i915_gem_object_unpin(obj);
	obj->pin_display = is_pin_display(obj);
3617 3618
}

3619
int
3620
i915_gem_object_finish_gpu(struct drm_i915_gem_object *obj)
3621
{
3622 3623
	int ret;

3624
	if ((obj->base.read_domains & I915_GEM_GPU_DOMAINS) == 0)
3625 3626
		return 0;

3627
	ret = i915_gem_object_wait_rendering(obj, false);
3628 3629 3630
	if (ret)
		return ret;

3631 3632
	/* Ensure that we invalidate the GPU's caches and TLBs. */
	obj->base.read_domains &= ~I915_GEM_GPU_DOMAINS;
3633
	return 0;
3634 3635
}

3636 3637 3638 3639 3640 3641
/**
 * Moves a single object to the CPU read, and possibly write domain.
 *
 * This function returns when the move is complete, including waiting on
 * flushes to occur.
 */
3642
int
3643
i915_gem_object_set_to_cpu_domain(struct drm_i915_gem_object *obj, bool write)
3644
{
C
Chris Wilson 已提交
3645
	uint32_t old_write_domain, old_read_domains;
3646 3647
	int ret;

3648 3649 3650
	if (obj->base.write_domain == I915_GEM_DOMAIN_CPU)
		return 0;

3651
	ret = i915_gem_object_wait_rendering(obj, !write);
3652 3653 3654
	if (ret)
		return ret;

3655
	i915_gem_object_flush_gtt_write_domain(obj);
3656

3657 3658
	old_write_domain = obj->base.write_domain;
	old_read_domains = obj->base.read_domains;
C
Chris Wilson 已提交
3659

3660
	/* Flush the CPU cache if it's still invalid. */
3661
	if ((obj->base.read_domains & I915_GEM_DOMAIN_CPU) == 0) {
3662
		i915_gem_clflush_object(obj, false);
3663

3664
		obj->base.read_domains |= I915_GEM_DOMAIN_CPU;
3665 3666 3667 3668 3669
	}

	/* It should now be out of any other write domains, and we can update
	 * the domain values for our changes.
	 */
3670
	BUG_ON((obj->base.write_domain & ~I915_GEM_DOMAIN_CPU) != 0);
3671 3672 3673 3674 3675

	/* If we're writing through the CPU, then the GPU read domains will
	 * need to be invalidated at next use.
	 */
	if (write) {
3676 3677
		obj->base.read_domains = I915_GEM_DOMAIN_CPU;
		obj->base.write_domain = I915_GEM_DOMAIN_CPU;
3678
	}
3679

C
Chris Wilson 已提交
3680 3681 3682 3683
	trace_i915_gem_object_change_domain(obj,
					    old_read_domains,
					    old_write_domain);

3684 3685 3686
	return 0;
}

3687 3688 3689
/* Throttle our rendering by waiting until the ring has completed our requests
 * emitted over 20 msec ago.
 *
3690 3691 3692 3693
 * Note that if we were to use the current jiffies each time around the loop,
 * we wouldn't escape the function with any frames outstanding if the time to
 * render a frame was over 20ms.
 *
3694 3695 3696
 * This should get us reasonable parallelism between CPU and GPU but also
 * relatively low latency when blocking on a particular request to finish.
 */
3697
static int
3698
i915_gem_ring_throttle(struct drm_device *dev, struct drm_file *file)
3699
{
3700 3701
	struct drm_i915_private *dev_priv = dev->dev_private;
	struct drm_i915_file_private *file_priv = file->driver_priv;
3702
	unsigned long recent_enough = jiffies - msecs_to_jiffies(20);
3703 3704
	struct drm_i915_gem_request *request;
	struct intel_ring_buffer *ring = NULL;
3705
	unsigned reset_counter;
3706 3707
	u32 seqno = 0;
	int ret;
3708

3709 3710 3711 3712 3713 3714 3715
	ret = i915_gem_wait_for_error(&dev_priv->gpu_error);
	if (ret)
		return ret;

	ret = i915_gem_check_wedge(&dev_priv->gpu_error, false);
	if (ret)
		return ret;
3716

3717
	spin_lock(&file_priv->mm.lock);
3718
	list_for_each_entry(request, &file_priv->mm.request_list, client_list) {
3719 3720
		if (time_after_eq(request->emitted_jiffies, recent_enough))
			break;
3721

3722 3723
		ring = request->ring;
		seqno = request->seqno;
3724
	}
3725
	reset_counter = atomic_read(&dev_priv->gpu_error.reset_counter);
3726
	spin_unlock(&file_priv->mm.lock);
3727

3728 3729
	if (seqno == 0)
		return 0;
3730

3731
	ret = __wait_seqno(ring, seqno, reset_counter, true, NULL);
3732 3733
	if (ret == 0)
		queue_delayed_work(dev_priv->wq, &dev_priv->mm.retire_work, 0);
3734 3735 3736 3737

	return ret;
}

3738
int
3739
i915_gem_object_pin(struct drm_i915_gem_object *obj,
B
Ben Widawsky 已提交
3740
		    struct i915_address_space *vm,
3741
		    uint32_t alignment,
3742 3743
		    bool map_and_fenceable,
		    bool nonblocking)
3744
{
3745
	struct i915_vma *vma;
3746 3747
	int ret;

3748 3749
	if (WARN_ON(obj->pin_count == DRM_I915_GEM_OBJECT_MAX_PIN_COUNT))
		return -EBUSY;
3750

3751 3752 3753 3754 3755 3756 3757
	WARN_ON(map_and_fenceable && !i915_is_ggtt(vm));

	vma = i915_gem_obj_to_vma(obj, vm);

	if (vma) {
		if ((alignment &&
		     vma->node.start & (alignment - 1)) ||
3758 3759
		    (map_and_fenceable && !obj->map_and_fenceable)) {
			WARN(obj->pin_count,
3760
			     "bo is already pinned with incorrect alignment:"
3761
			     " offset=%lx, req.alignment=%x, req.map_and_fenceable=%d,"
3762
			     " obj->map_and_fenceable=%d\n",
3763
			     i915_gem_obj_offset(obj, vm), alignment,
3764
			     map_and_fenceable,
3765
			     obj->map_and_fenceable);
3766
			ret = i915_vma_unbind(vma);
3767 3768 3769 3770 3771
			if (ret)
				return ret;
		}
	}

3772
	if (!i915_gem_obj_bound(obj, vm)) {
3773 3774
		struct drm_i915_private *dev_priv = obj->base.dev->dev_private;

3775 3776 3777
		ret = i915_gem_object_bind_to_vm(obj, vm, alignment,
						 map_and_fenceable,
						 nonblocking);
3778
		if (ret)
3779
			return ret;
3780 3781 3782

		if (!dev_priv->mm.aliasing_ppgtt)
			i915_gem_gtt_bind_object(obj, obj->cache_level);
3783
	}
J
Jesse Barnes 已提交
3784

3785 3786 3787
	if (!obj->has_global_gtt_mapping && map_and_fenceable)
		i915_gem_gtt_bind_object(obj, obj->cache_level);

3788
	obj->pin_count++;
3789
	obj->pin_mappable |= map_and_fenceable;
3790 3791 3792 3793 3794

	return 0;
}

void
3795
i915_gem_object_unpin(struct drm_i915_gem_object *obj)
3796
{
3797
	BUG_ON(obj->pin_count == 0);
3798
	BUG_ON(!i915_gem_obj_bound_any(obj));
3799

3800
	if (--obj->pin_count == 0)
3801
		obj->pin_mappable = false;
3802 3803 3804 3805
}

int
i915_gem_pin_ioctl(struct drm_device *dev, void *data,
3806
		   struct drm_file *file)
3807 3808
{
	struct drm_i915_gem_pin *args = data;
3809
	struct drm_i915_gem_object *obj;
3810 3811
	int ret;

3812 3813 3814
	ret = i915_mutex_lock_interruptible(dev);
	if (ret)
		return ret;
3815

3816
	obj = to_intel_bo(drm_gem_object_lookup(dev, file, args->handle));
3817
	if (&obj->base == NULL) {
3818 3819
		ret = -ENOENT;
		goto unlock;
3820 3821
	}

3822
	if (obj->madv != I915_MADV_WILLNEED) {
C
Chris Wilson 已提交
3823
		DRM_ERROR("Attempting to pin a purgeable buffer\n");
3824 3825
		ret = -EINVAL;
		goto out;
3826 3827
	}

3828
	if (obj->pin_filp != NULL && obj->pin_filp != file) {
J
Jesse Barnes 已提交
3829 3830
		DRM_ERROR("Already pinned in i915_gem_pin_ioctl(): %d\n",
			  args->handle);
3831 3832
		ret = -EINVAL;
		goto out;
J
Jesse Barnes 已提交
3833 3834
	}

3835
	if (obj->user_pin_count == 0) {
B
Ben Widawsky 已提交
3836
		ret = i915_gem_obj_ggtt_pin(obj, args->alignment, true, false);
3837 3838
		if (ret)
			goto out;
3839 3840
	}

3841 3842 3843
	obj->user_pin_count++;
	obj->pin_filp = file;

3844
	args->offset = i915_gem_obj_ggtt_offset(obj);
3845
out:
3846
	drm_gem_object_unreference(&obj->base);
3847
unlock:
3848
	mutex_unlock(&dev->struct_mutex);
3849
	return ret;
3850 3851 3852 3853
}

int
i915_gem_unpin_ioctl(struct drm_device *dev, void *data,
3854
		     struct drm_file *file)
3855 3856
{
	struct drm_i915_gem_pin *args = data;
3857
	struct drm_i915_gem_object *obj;
3858
	int ret;
3859

3860 3861 3862
	ret = i915_mutex_lock_interruptible(dev);
	if (ret)
		return ret;
3863

3864
	obj = to_intel_bo(drm_gem_object_lookup(dev, file, args->handle));
3865
	if (&obj->base == NULL) {
3866 3867
		ret = -ENOENT;
		goto unlock;
3868
	}
3869

3870
	if (obj->pin_filp != file) {
J
Jesse Barnes 已提交
3871 3872
		DRM_ERROR("Not pinned by caller in i915_gem_pin_ioctl(): %d\n",
			  args->handle);
3873 3874
		ret = -EINVAL;
		goto out;
J
Jesse Barnes 已提交
3875
	}
3876 3877 3878
	obj->user_pin_count--;
	if (obj->user_pin_count == 0) {
		obj->pin_filp = NULL;
J
Jesse Barnes 已提交
3879 3880
		i915_gem_object_unpin(obj);
	}
3881

3882
out:
3883
	drm_gem_object_unreference(&obj->base);
3884
unlock:
3885
	mutex_unlock(&dev->struct_mutex);
3886
	return ret;
3887 3888 3889 3890
}

int
i915_gem_busy_ioctl(struct drm_device *dev, void *data,
3891
		    struct drm_file *file)
3892 3893
{
	struct drm_i915_gem_busy *args = data;
3894
	struct drm_i915_gem_object *obj;
3895 3896
	int ret;

3897
	ret = i915_mutex_lock_interruptible(dev);
3898
	if (ret)
3899
		return ret;
3900

3901
	obj = to_intel_bo(drm_gem_object_lookup(dev, file, args->handle));
3902
	if (&obj->base == NULL) {
3903 3904
		ret = -ENOENT;
		goto unlock;
3905
	}
3906

3907 3908 3909 3910
	/* Count all active objects as busy, even if they are currently not used
	 * by the gpu. Users of this interface expect objects to eventually
	 * become non-busy without any further actions, therefore emit any
	 * necessary flushes here.
3911
	 */
3912
	ret = i915_gem_object_flush_active(obj);
3913

3914
	args->busy = obj->active;
3915 3916 3917 3918
	if (obj->ring) {
		BUILD_BUG_ON(I915_NUM_RINGS > 16);
		args->busy |= intel_ring_flag(obj->ring) << 16;
	}
3919

3920
	drm_gem_object_unreference(&obj->base);
3921
unlock:
3922
	mutex_unlock(&dev->struct_mutex);
3923
	return ret;
3924 3925 3926 3927 3928 3929
}

int
i915_gem_throttle_ioctl(struct drm_device *dev, void *data,
			struct drm_file *file_priv)
{
3930
	return i915_gem_ring_throttle(dev, file_priv);
3931 3932
}

3933 3934 3935 3936 3937
int
i915_gem_madvise_ioctl(struct drm_device *dev, void *data,
		       struct drm_file *file_priv)
{
	struct drm_i915_gem_madvise *args = data;
3938
	struct drm_i915_gem_object *obj;
3939
	int ret;
3940 3941 3942 3943 3944 3945 3946 3947 3948

	switch (args->madv) {
	case I915_MADV_DONTNEED:
	case I915_MADV_WILLNEED:
	    break;
	default:
	    return -EINVAL;
	}

3949 3950 3951 3952
	ret = i915_mutex_lock_interruptible(dev);
	if (ret)
		return ret;

3953
	obj = to_intel_bo(drm_gem_object_lookup(dev, file_priv, args->handle));
3954
	if (&obj->base == NULL) {
3955 3956
		ret = -ENOENT;
		goto unlock;
3957 3958
	}

3959
	if (obj->pin_count) {
3960 3961
		ret = -EINVAL;
		goto out;
3962 3963
	}

3964 3965
	if (obj->madv != __I915_MADV_PURGED)
		obj->madv = args->madv;
3966

C
Chris Wilson 已提交
3967 3968
	/* if the object is no longer attached, discard its backing storage */
	if (i915_gem_object_is_purgeable(obj) && obj->pages == NULL)
3969 3970
		i915_gem_object_truncate(obj);

3971
	args->retained = obj->madv != __I915_MADV_PURGED;
C
Chris Wilson 已提交
3972

3973
out:
3974
	drm_gem_object_unreference(&obj->base);
3975
unlock:
3976
	mutex_unlock(&dev->struct_mutex);
3977
	return ret;
3978 3979
}

3980 3981
void i915_gem_object_init(struct drm_i915_gem_object *obj,
			  const struct drm_i915_gem_object_ops *ops)
3982
{
3983
	INIT_LIST_HEAD(&obj->global_list);
3984
	INIT_LIST_HEAD(&obj->ring_list);
3985
	INIT_LIST_HEAD(&obj->obj_exec_link);
B
Ben Widawsky 已提交
3986
	INIT_LIST_HEAD(&obj->vma_list);
3987

3988 3989
	obj->ops = ops;

3990 3991 3992 3993 3994 3995 3996 3997
	obj->fence_reg = I915_FENCE_REG_NONE;
	obj->madv = I915_MADV_WILLNEED;
	/* Avoid an unnecessary call to unbind on the first bind. */
	obj->map_and_fenceable = true;

	i915_gem_info_add_obj(obj->base.dev->dev_private, obj->base.size);
}

3998 3999 4000 4001 4002
static const struct drm_i915_gem_object_ops i915_gem_object_ops = {
	.get_pages = i915_gem_object_get_pages_gtt,
	.put_pages = i915_gem_object_put_pages_gtt,
};

4003 4004
struct drm_i915_gem_object *i915_gem_alloc_object(struct drm_device *dev,
						  size_t size)
4005
{
4006
	struct drm_i915_gem_object *obj;
4007
	struct address_space *mapping;
D
Daniel Vetter 已提交
4008
	gfp_t mask;
4009

4010
	obj = i915_gem_object_alloc(dev);
4011 4012
	if (obj == NULL)
		return NULL;
4013

4014
	if (drm_gem_object_init(dev, &obj->base, size) != 0) {
4015
		i915_gem_object_free(obj);
4016 4017
		return NULL;
	}
4018

4019 4020 4021 4022 4023 4024 4025
	mask = GFP_HIGHUSER | __GFP_RECLAIMABLE;
	if (IS_CRESTLINE(dev) || IS_BROADWATER(dev)) {
		/* 965gm cannot relocate objects above 4GiB. */
		mask &= ~__GFP_HIGHMEM;
		mask |= __GFP_DMA32;
	}

A
Al Viro 已提交
4026
	mapping = file_inode(obj->base.filp)->i_mapping;
4027
	mapping_set_gfp_mask(mapping, mask);
4028

4029
	i915_gem_object_init(obj, &i915_gem_object_ops);
4030

4031 4032
	obj->base.write_domain = I915_GEM_DOMAIN_CPU;
	obj->base.read_domains = I915_GEM_DOMAIN_CPU;
4033

4034 4035
	if (HAS_LLC(dev)) {
		/* On some devices, we can have the GPU use the LLC (the CPU
4036 4037 4038 4039 4040 4041 4042 4043 4044 4045 4046 4047 4048 4049 4050
		 * cache) for about a 10% performance improvement
		 * compared to uncached.  Graphics requests other than
		 * display scanout are coherent with the CPU in
		 * accessing this cache.  This means in this mode we
		 * don't need to clflush on the CPU side, and on the
		 * GPU side we only need to flush internal caches to
		 * get data visible to the CPU.
		 *
		 * However, we maintain the display planes as UC, and so
		 * need to rebind when first used as such.
		 */
		obj->cache_level = I915_CACHE_LLC;
	} else
		obj->cache_level = I915_CACHE_NONE;

4051 4052
	trace_i915_gem_object_create(obj);

4053
	return obj;
4054 4055 4056 4057 4058
}

int i915_gem_init_object(struct drm_gem_object *obj)
{
	BUG();
4059

4060 4061 4062
	return 0;
}

4063
void i915_gem_free_object(struct drm_gem_object *gem_obj)
4064
{
4065
	struct drm_i915_gem_object *obj = to_intel_bo(gem_obj);
4066
	struct drm_device *dev = obj->base.dev;
4067
	drm_i915_private_t *dev_priv = dev->dev_private;
4068
	struct i915_vma *vma, *next;
4069

4070 4071
	trace_i915_gem_object_destroy(obj);

4072 4073 4074 4075
	if (obj->phys_obj)
		i915_gem_detach_phys_object(dev, obj);

	obj->pin_count = 0;
4076 4077 4078 4079 4080 4081 4082
	/* NB: 0 or 1 elements */
	WARN_ON(!list_empty(&obj->vma_list) &&
		!list_is_singular(&obj->vma_list));
	list_for_each_entry_safe(vma, next, &obj->vma_list, vma_link) {
		int ret = i915_vma_unbind(vma);
		if (WARN_ON(ret == -ERESTARTSYS)) {
			bool was_interruptible;
4083

4084 4085
			was_interruptible = dev_priv->mm.interruptible;
			dev_priv->mm.interruptible = false;
4086

4087
			WARN_ON(i915_vma_unbind(vma));
4088

4089 4090
			dev_priv->mm.interruptible = was_interruptible;
		}
4091 4092
	}

B
Ben Widawsky 已提交
4093 4094 4095 4096 4097
	/* Stolen objects don't hold a ref, but do hold pin count. Fix that up
	 * before progressing. */
	if (obj->stolen)
		i915_gem_object_unpin_pages(obj);

B
Ben Widawsky 已提交
4098 4099
	if (WARN_ON(obj->pages_pin_count))
		obj->pages_pin_count = 0;
4100
	i915_gem_object_put_pages(obj);
4101
	i915_gem_object_free_mmap_offset(obj);
4102
	i915_gem_object_release_stolen(obj);
4103

4104 4105
	BUG_ON(obj->pages);

4106 4107
	if (obj->base.import_attach)
		drm_prime_gem_destroy(&obj->base, NULL);
4108

4109 4110
	drm_gem_object_release(&obj->base);
	i915_gem_info_remove_obj(dev_priv, obj->base.size);
4111

4112
	kfree(obj->bit_17);
4113
	i915_gem_object_free(obj);
4114 4115
}

4116
struct i915_vma *i915_gem_obj_to_vma(struct drm_i915_gem_object *obj,
B
Ben Widawsky 已提交
4117
				     struct i915_address_space *vm)
4118 4119 4120 4121 4122 4123 4124 4125 4126 4127 4128
{
	struct i915_vma *vma;
	list_for_each_entry(vma, &obj->vma_list, vma_link)
		if (vma->vm == vm)
			return vma;

	return NULL;
}

static struct i915_vma *__i915_gem_vma_create(struct drm_i915_gem_object *obj,
					      struct i915_address_space *vm)
B
Ben Widawsky 已提交
4129 4130 4131 4132 4133 4134
{
	struct i915_vma *vma = kzalloc(sizeof(*vma), GFP_KERNEL);
	if (vma == NULL)
		return ERR_PTR(-ENOMEM);

	INIT_LIST_HEAD(&vma->vma_link);
B
Ben Widawsky 已提交
4135
	INIT_LIST_HEAD(&vma->mm_list);
4136
	INIT_LIST_HEAD(&vma->exec_list);
B
Ben Widawsky 已提交
4137 4138 4139
	vma->vm = vm;
	vma->obj = obj;

4140 4141 4142 4143 4144 4145
	/* Keep GGTT vmas first to make debug easier */
	if (i915_is_ggtt(vm))
		list_add(&vma->vma_link, &obj->vma_list);
	else
		list_add_tail(&vma->vma_link, &obj->vma_list);

B
Ben Widawsky 已提交
4146 4147 4148
	return vma;
}

4149 4150 4151 4152 4153 4154 4155 4156 4157 4158 4159 4160 4161
struct i915_vma *
i915_gem_obj_lookup_or_create_vma(struct drm_i915_gem_object *obj,
				  struct i915_address_space *vm)
{
	struct i915_vma *vma;

	vma = i915_gem_obj_to_vma(obj, vm);
	if (!vma)
		vma = __i915_gem_vma_create(obj, vm);

	return vma;
}

B
Ben Widawsky 已提交
4162 4163 4164
void i915_gem_vma_destroy(struct i915_vma *vma)
{
	WARN_ON(vma->node.allocated);
4165 4166 4167 4168 4169

	/* Keep the vma as a placeholder in the execbuffer reservation lists */
	if (!list_empty(&vma->exec_list))
		return;

4170 4171
	list_del(&vma->vma_link);

B
Ben Widawsky 已提交
4172 4173 4174
	kfree(vma);
}

4175 4176 4177 4178 4179
int
i915_gem_idle(struct drm_device *dev)
{
	drm_i915_private_t *dev_priv = dev->dev_private;
	int ret;
4180

4181
	if (dev_priv->ums.mm_suspended) {
4182 4183
		mutex_unlock(&dev->struct_mutex);
		return 0;
4184 4185
	}

4186
	ret = i915_gpu_idle(dev);
4187 4188
	if (ret) {
		mutex_unlock(&dev->struct_mutex);
4189
		return ret;
4190
	}
4191
	i915_gem_retire_requests(dev);
4192

4193
	/* Under UMS, be paranoid and evict. */
4194
	if (!drm_core_check_feature(dev, DRIVER_MODESET))
C
Chris Wilson 已提交
4195
		i915_gem_evict_everything(dev);
4196

4197
	del_timer_sync(&dev_priv->gpu_error.hangcheck_timer);
4198 4199

	i915_kernel_lost_context(dev);
4200
	i915_gem_cleanup_ringbuffer(dev);
4201 4202 4203 4204

	/* Cancel the retire work handler, which should be idle now. */
	cancel_delayed_work_sync(&dev_priv->mm.retire_work);

4205 4206 4207
	return 0;
}

B
Ben Widawsky 已提交
4208 4209 4210 4211 4212 4213
void i915_gem_l3_remap(struct drm_device *dev)
{
	drm_i915_private_t *dev_priv = dev->dev_private;
	u32 misccpctl;
	int i;

4214
	if (!HAS_L3_GPU_CACHE(dev))
B
Ben Widawsky 已提交
4215 4216
		return;

4217
	if (!dev_priv->l3_parity.remap_info)
B
Ben Widawsky 已提交
4218 4219 4220 4221 4222 4223 4224 4225
		return;

	misccpctl = I915_READ(GEN7_MISCCPCTL);
	I915_WRITE(GEN7_MISCCPCTL, misccpctl & ~GEN7_DOP_CLOCK_GATE_ENABLE);
	POSTING_READ(GEN7_MISCCPCTL);

	for (i = 0; i < GEN7_L3LOG_SIZE; i += 4) {
		u32 remap = I915_READ(GEN7_L3LOG_BASE + i);
4226
		if (remap && remap != dev_priv->l3_parity.remap_info[i/4])
B
Ben Widawsky 已提交
4227 4228
			DRM_DEBUG("0x%x was already programmed to %x\n",
				  GEN7_L3LOG_BASE + i, remap);
4229
		if (remap && !dev_priv->l3_parity.remap_info[i/4])
B
Ben Widawsky 已提交
4230
			DRM_DEBUG_DRIVER("Clearing remapped register\n");
4231
		I915_WRITE(GEN7_L3LOG_BASE + i, dev_priv->l3_parity.remap_info[i/4]);
B
Ben Widawsky 已提交
4232 4233 4234 4235 4236 4237 4238 4239
	}

	/* Make sure all the writes land before disabling dop clock gating */
	POSTING_READ(GEN7_L3LOG_BASE);

	I915_WRITE(GEN7_MISCCPCTL, misccpctl);
}

4240 4241 4242 4243
void i915_gem_init_swizzling(struct drm_device *dev)
{
	drm_i915_private_t *dev_priv = dev->dev_private;

4244
	if (INTEL_INFO(dev)->gen < 5 ||
4245 4246 4247 4248 4249 4250
	    dev_priv->mm.bit_6_swizzle_x == I915_BIT_6_SWIZZLE_NONE)
		return;

	I915_WRITE(DISP_ARB_CTL, I915_READ(DISP_ARB_CTL) |
				 DISP_TILE_SURFACE_SWIZZLING);

4251 4252 4253
	if (IS_GEN5(dev))
		return;

4254 4255
	I915_WRITE(TILECTL, I915_READ(TILECTL) | TILECTL_SWZCTL);
	if (IS_GEN6(dev))
4256
		I915_WRITE(ARB_MODE, _MASKED_BIT_ENABLE(ARB_MODE_SWIZZLE_SNB));
4257
	else if (IS_GEN7(dev))
4258
		I915_WRITE(ARB_MODE, _MASKED_BIT_ENABLE(ARB_MODE_SWIZZLE_IVB));
4259 4260
	else
		BUG();
4261
}
D
Daniel Vetter 已提交
4262

4263 4264 4265 4266 4267 4268 4269 4270 4271 4272 4273 4274 4275 4276 4277 4278
static bool
intel_enable_blt(struct drm_device *dev)
{
	if (!HAS_BLT(dev))
		return false;

	/* The blitter was dysfunctional on early prototypes */
	if (IS_GEN6(dev) && dev->pdev->revision < 8) {
		DRM_INFO("BLT not supported on this pre-production hardware;"
			 " graphics performance will be degraded.\n");
		return false;
	}

	return true;
}

4279
static int i915_gem_init_rings(struct drm_device *dev)
4280
{
4281
	struct drm_i915_private *dev_priv = dev->dev_private;
4282
	int ret;
4283

4284
	ret = intel_init_render_ring_buffer(dev);
4285
	if (ret)
4286
		return ret;
4287 4288

	if (HAS_BSD(dev)) {
4289
		ret = intel_init_bsd_ring_buffer(dev);
4290 4291
		if (ret)
			goto cleanup_render_ring;
4292
	}
4293

4294
	if (intel_enable_blt(dev)) {
4295 4296 4297 4298 4299
		ret = intel_init_blt_ring_buffer(dev);
		if (ret)
			goto cleanup_bsd_ring;
	}

B
Ben Widawsky 已提交
4300 4301 4302 4303 4304 4305 4306
	if (HAS_VEBOX(dev)) {
		ret = intel_init_vebox_ring_buffer(dev);
		if (ret)
			goto cleanup_blt_ring;
	}


4307
	ret = i915_gem_set_seqno(dev, ((u32)~0 - 0x1000));
4308
	if (ret)
B
Ben Widawsky 已提交
4309
		goto cleanup_vebox_ring;
4310 4311 4312

	return 0;

B
Ben Widawsky 已提交
4313 4314
cleanup_vebox_ring:
	intel_cleanup_ring_buffer(&dev_priv->ring[VECS]);
4315 4316 4317 4318 4319 4320 4321 4322 4323 4324 4325 4326 4327 4328 4329 4330 4331 4332 4333
cleanup_blt_ring:
	intel_cleanup_ring_buffer(&dev_priv->ring[BCS]);
cleanup_bsd_ring:
	intel_cleanup_ring_buffer(&dev_priv->ring[VCS]);
cleanup_render_ring:
	intel_cleanup_ring_buffer(&dev_priv->ring[RCS]);

	return ret;
}

int
i915_gem_init_hw(struct drm_device *dev)
{
	drm_i915_private_t *dev_priv = dev->dev_private;
	int ret;

	if (INTEL_INFO(dev)->gen < 6 && !intel_enable_gtt())
		return -EIO;

B
Ben Widawsky 已提交
4334
	if (dev_priv->ellc_size)
4335
		I915_WRITE(HSW_IDICR, I915_READ(HSW_IDICR) | IDIHASHMSK(0xf));
4336

4337 4338 4339 4340 4341
	if (IS_HSW_GT3(dev))
		I915_WRITE(MI_PREDICATE_RESULT_2, LOWER_SLICE_ENABLED);
	else
		I915_WRITE(MI_PREDICATE_RESULT_2, LOWER_SLICE_DISABLED);

4342 4343 4344 4345 4346 4347
	if (HAS_PCH_NOP(dev)) {
		u32 temp = I915_READ(GEN7_MSG_CTL);
		temp &= ~(WAIT_FOR_PCH_FLR_ACK | WAIT_FOR_PCH_RESET_ACK);
		I915_WRITE(GEN7_MSG_CTL, temp);
	}

4348 4349 4350 4351 4352
	i915_gem_l3_remap(dev);

	i915_gem_init_swizzling(dev);

	ret = i915_gem_init_rings(dev);
4353 4354 4355
	if (ret)
		return ret;

4356 4357 4358 4359 4360
	/*
	 * XXX: There was some w/a described somewhere suggesting loading
	 * contexts before PPGTT.
	 */
	i915_gem_context_init(dev);
4361 4362 4363 4364 4365 4366 4367
	if (dev_priv->mm.aliasing_ppgtt) {
		ret = dev_priv->mm.aliasing_ppgtt->enable(dev);
		if (ret) {
			i915_gem_cleanup_aliasing_ppgtt(dev);
			DRM_INFO("PPGTT enable failed. This is not fatal, but unexpected\n");
		}
	}
D
Daniel Vetter 已提交
4368

4369
	return 0;
4370 4371
}

4372 4373 4374 4375 4376 4377
int i915_gem_init(struct drm_device *dev)
{
	struct drm_i915_private *dev_priv = dev->dev_private;
	int ret;

	mutex_lock(&dev->struct_mutex);
4378 4379 4380 4381 4382 4383 4384 4385

	if (IS_VALLEYVIEW(dev)) {
		/* VLVA0 (potential hack), BIOS isn't actually waking us */
		I915_WRITE(VLV_GTLC_WAKE_CTRL, 1);
		if (wait_for((I915_READ(VLV_GTLC_PW_STATUS) & 1) == 1, 10))
			DRM_DEBUG_DRIVER("allow wake ack timed out\n");
	}

4386
	i915_gem_init_global_gtt(dev);
4387

4388 4389 4390 4391 4392 4393 4394
	ret = i915_gem_init_hw(dev);
	mutex_unlock(&dev->struct_mutex);
	if (ret) {
		i915_gem_cleanup_aliasing_ppgtt(dev);
		return ret;
	}

4395 4396 4397
	/* Allow hardware batchbuffers unless told otherwise, but not for KMS. */
	if (!drm_core_check_feature(dev, DRIVER_MODESET))
		dev_priv->dri1.allow_batchbuffer = 1;
4398 4399 4400
	return 0;
}

4401 4402 4403 4404
void
i915_gem_cleanup_ringbuffer(struct drm_device *dev)
{
	drm_i915_private_t *dev_priv = dev->dev_private;
4405
	struct intel_ring_buffer *ring;
4406
	int i;
4407

4408 4409
	for_each_ring(ring, dev_priv, i)
		intel_cleanup_ring_buffer(ring);
4410 4411
}

4412 4413 4414 4415
int
i915_gem_entervt_ioctl(struct drm_device *dev, void *data,
		       struct drm_file *file_priv)
{
4416
	struct drm_i915_private *dev_priv = dev->dev_private;
4417
	int ret;
4418

J
Jesse Barnes 已提交
4419 4420 4421
	if (drm_core_check_feature(dev, DRIVER_MODESET))
		return 0;

4422
	if (i915_reset_in_progress(&dev_priv->gpu_error)) {
4423
		DRM_ERROR("Reenabling wedged hardware, good luck\n");
4424
		atomic_set(&dev_priv->gpu_error.reset_counter, 0);
4425 4426 4427
	}

	mutex_lock(&dev->struct_mutex);
4428
	dev_priv->ums.mm_suspended = 0;
4429

4430
	ret = i915_gem_init_hw(dev);
4431 4432
	if (ret != 0) {
		mutex_unlock(&dev->struct_mutex);
4433
		return ret;
4434
	}
4435

4436
	BUG_ON(!list_empty(&dev_priv->gtt.base.active_list));
4437
	mutex_unlock(&dev->struct_mutex);
4438

4439 4440 4441
	ret = drm_irq_install(dev);
	if (ret)
		goto cleanup_ringbuffer;
4442

4443
	return 0;
4444 4445 4446 4447

cleanup_ringbuffer:
	mutex_lock(&dev->struct_mutex);
	i915_gem_cleanup_ringbuffer(dev);
4448
	dev_priv->ums.mm_suspended = 1;
4449 4450 4451
	mutex_unlock(&dev->struct_mutex);

	return ret;
4452 4453 4454 4455 4456 4457
}

int
i915_gem_leavevt_ioctl(struct drm_device *dev, void *data,
		       struct drm_file *file_priv)
{
4458 4459 4460
	struct drm_i915_private *dev_priv = dev->dev_private;
	int ret;

J
Jesse Barnes 已提交
4461 4462 4463
	if (drm_core_check_feature(dev, DRIVER_MODESET))
		return 0;

4464
	drm_irq_uninstall(dev);
4465 4466 4467 4468 4469 4470 4471 4472 4473 4474 4475 4476 4477

	mutex_lock(&dev->struct_mutex);
	ret =  i915_gem_idle(dev);

	/* Hack!  Don't let anybody do execbuf while we don't control the chip.
	 * We need to replace this with a semaphore, or something.
	 * And not confound ums.mm_suspended!
	 */
	if (ret != 0)
		dev_priv->ums.mm_suspended = 1;
	mutex_unlock(&dev->struct_mutex);

	return ret;
4478 4479 4480 4481 4482 4483 4484
}

void
i915_gem_lastclose(struct drm_device *dev)
{
	int ret;

4485 4486 4487
	if (drm_core_check_feature(dev, DRIVER_MODESET))
		return;

4488
	mutex_lock(&dev->struct_mutex);
4489 4490 4491
	ret = i915_gem_idle(dev);
	if (ret)
		DRM_ERROR("failed to idle hardware: %d\n", ret);
4492
	mutex_unlock(&dev->struct_mutex);
4493 4494
}

4495 4496 4497 4498 4499 4500 4501
static void
init_ring_lists(struct intel_ring_buffer *ring)
{
	INIT_LIST_HEAD(&ring->active_list);
	INIT_LIST_HEAD(&ring->request_list);
}

B
Ben Widawsky 已提交
4502 4503 4504 4505 4506 4507 4508 4509 4510 4511
static void i915_init_vm(struct drm_i915_private *dev_priv,
			 struct i915_address_space *vm)
{
	vm->dev = dev_priv->dev;
	INIT_LIST_HEAD(&vm->active_list);
	INIT_LIST_HEAD(&vm->inactive_list);
	INIT_LIST_HEAD(&vm->global_link);
	list_add(&vm->global_link, &dev_priv->vm_list);
}

4512 4513 4514 4515
void
i915_gem_load(struct drm_device *dev)
{
	drm_i915_private_t *dev_priv = dev->dev_private;
4516 4517 4518 4519 4520 4521 4522
	int i;

	dev_priv->slab =
		kmem_cache_create("i915_gem_object",
				  sizeof(struct drm_i915_gem_object), 0,
				  SLAB_HWCACHE_ALIGN,
				  NULL);
4523

B
Ben Widawsky 已提交
4524 4525 4526
	INIT_LIST_HEAD(&dev_priv->vm_list);
	i915_init_vm(dev_priv, &dev_priv->gtt.base);

C
Chris Wilson 已提交
4527 4528
	INIT_LIST_HEAD(&dev_priv->mm.unbound_list);
	INIT_LIST_HEAD(&dev_priv->mm.bound_list);
4529
	INIT_LIST_HEAD(&dev_priv->mm.fence_list);
4530 4531
	for (i = 0; i < I915_NUM_RINGS; i++)
		init_ring_lists(&dev_priv->ring[i]);
4532
	for (i = 0; i < I915_MAX_NUM_FENCES; i++)
4533
		INIT_LIST_HEAD(&dev_priv->fence_regs[i].lru_list);
4534 4535
	INIT_DELAYED_WORK(&dev_priv->mm.retire_work,
			  i915_gem_retire_work_handler);
4536
	init_waitqueue_head(&dev_priv->gpu_error.reset_queue);
4537

4538 4539
	/* On GEN3 we really need to make sure the ARB C3 LP bit is set */
	if (IS_GEN3(dev)) {
4540 4541
		I915_WRITE(MI_ARB_STATE,
			   _MASKED_BIT_ENABLE(MI_ARB_C3_LP_WRITE_ENABLE));
4542 4543
	}

4544 4545
	dev_priv->relative_constants_mode = I915_EXEC_CONSTANTS_REL_GENERAL;

4546
	/* Old X drivers will take 0-2 for front, back, depth buffers */
4547 4548
	if (!drm_core_check_feature(dev, DRIVER_MODESET))
		dev_priv->fence_reg_start = 3;
4549

4550 4551 4552
	if (INTEL_INFO(dev)->gen >= 7 && !IS_VALLEYVIEW(dev))
		dev_priv->num_fence_regs = 32;
	else if (INTEL_INFO(dev)->gen >= 4 || IS_I945G(dev) || IS_I945GM(dev) || IS_G33(dev))
4553 4554 4555 4556
		dev_priv->num_fence_regs = 16;
	else
		dev_priv->num_fence_regs = 8;

4557
	/* Initialize fence registers to zero */
4558 4559
	INIT_LIST_HEAD(&dev_priv->mm.fence_list);
	i915_gem_restore_fences(dev);
4560

4561
	i915_gem_detect_bit_6_swizzle(dev);
4562
	init_waitqueue_head(&dev_priv->pending_flip_queue);
4563

4564 4565
	dev_priv->mm.interruptible = true;

4566 4567 4568
	dev_priv->mm.inactive_shrinker.shrink = i915_gem_inactive_shrink;
	dev_priv->mm.inactive_shrinker.seeks = DEFAULT_SEEKS;
	register_shrinker(&dev_priv->mm.inactive_shrinker);
4569
}
4570 4571 4572 4573 4574

/*
 * Create a physically contiguous memory object for this object
 * e.g. for cursor + overlay regs
 */
4575 4576
static int i915_gem_init_phys_object(struct drm_device *dev,
				     int id, int size, int align)
4577 4578 4579 4580 4581 4582 4583 4584
{
	drm_i915_private_t *dev_priv = dev->dev_private;
	struct drm_i915_gem_phys_object *phys_obj;
	int ret;

	if (dev_priv->mm.phys_objs[id - 1] || !size)
		return 0;

4585
	phys_obj = kzalloc(sizeof(struct drm_i915_gem_phys_object), GFP_KERNEL);
4586 4587 4588 4589 4590
	if (!phys_obj)
		return -ENOMEM;

	phys_obj->id = id;

4591
	phys_obj->handle = drm_pci_alloc(dev, size, align);
4592 4593 4594 4595 4596 4597 4598 4599 4600 4601 4602 4603
	if (!phys_obj->handle) {
		ret = -ENOMEM;
		goto kfree_obj;
	}
#ifdef CONFIG_X86
	set_memory_wc((unsigned long)phys_obj->handle->vaddr, phys_obj->handle->size / PAGE_SIZE);
#endif

	dev_priv->mm.phys_objs[id - 1] = phys_obj;

	return 0;
kfree_obj:
4604
	kfree(phys_obj);
4605 4606 4607
	return ret;
}

4608
static void i915_gem_free_phys_object(struct drm_device *dev, int id)
4609 4610 4611 4612 4613 4614 4615 4616 4617 4618 4619 4620 4621 4622 4623 4624 4625 4626 4627 4628 4629 4630 4631 4632
{
	drm_i915_private_t *dev_priv = dev->dev_private;
	struct drm_i915_gem_phys_object *phys_obj;

	if (!dev_priv->mm.phys_objs[id - 1])
		return;

	phys_obj = dev_priv->mm.phys_objs[id - 1];
	if (phys_obj->cur_obj) {
		i915_gem_detach_phys_object(dev, phys_obj->cur_obj);
	}

#ifdef CONFIG_X86
	set_memory_wb((unsigned long)phys_obj->handle->vaddr, phys_obj->handle->size / PAGE_SIZE);
#endif
	drm_pci_free(dev, phys_obj->handle);
	kfree(phys_obj);
	dev_priv->mm.phys_objs[id - 1] = NULL;
}

void i915_gem_free_all_phys_object(struct drm_device *dev)
{
	int i;

4633
	for (i = I915_GEM_PHYS_CURSOR_0; i <= I915_MAX_PHYS_OBJECT; i++)
4634 4635 4636 4637
		i915_gem_free_phys_object(dev, i);
}

void i915_gem_detach_phys_object(struct drm_device *dev,
4638
				 struct drm_i915_gem_object *obj)
4639
{
A
Al Viro 已提交
4640
	struct address_space *mapping = file_inode(obj->base.filp)->i_mapping;
4641
	char *vaddr;
4642 4643 4644
	int i;
	int page_count;

4645
	if (!obj->phys_obj)
4646
		return;
4647
	vaddr = obj->phys_obj->handle->vaddr;
4648

4649
	page_count = obj->base.size / PAGE_SIZE;
4650
	for (i = 0; i < page_count; i++) {
4651
		struct page *page = shmem_read_mapping_page(mapping, i);
4652 4653 4654 4655 4656 4657 4658 4659 4660 4661 4662
		if (!IS_ERR(page)) {
			char *dst = kmap_atomic(page);
			memcpy(dst, vaddr + i*PAGE_SIZE, PAGE_SIZE);
			kunmap_atomic(dst);

			drm_clflush_pages(&page, 1);

			set_page_dirty(page);
			mark_page_accessed(page);
			page_cache_release(page);
		}
4663
	}
4664
	i915_gem_chipset_flush(dev);
4665

4666 4667
	obj->phys_obj->cur_obj = NULL;
	obj->phys_obj = NULL;
4668 4669 4670 4671
}

int
i915_gem_attach_phys_object(struct drm_device *dev,
4672
			    struct drm_i915_gem_object *obj,
4673 4674
			    int id,
			    int align)
4675
{
A
Al Viro 已提交
4676
	struct address_space *mapping = file_inode(obj->base.filp)->i_mapping;
4677 4678 4679 4680 4681 4682 4683 4684
	drm_i915_private_t *dev_priv = dev->dev_private;
	int ret = 0;
	int page_count;
	int i;

	if (id > I915_MAX_PHYS_OBJECT)
		return -EINVAL;

4685 4686
	if (obj->phys_obj) {
		if (obj->phys_obj->id == id)
4687 4688 4689 4690 4691 4692 4693
			return 0;
		i915_gem_detach_phys_object(dev, obj);
	}

	/* create a new object */
	if (!dev_priv->mm.phys_objs[id - 1]) {
		ret = i915_gem_init_phys_object(dev, id,
4694
						obj->base.size, align);
4695
		if (ret) {
4696 4697
			DRM_ERROR("failed to init phys object %d size: %zu\n",
				  id, obj->base.size);
4698
			return ret;
4699 4700 4701 4702
		}
	}

	/* bind to the object */
4703 4704
	obj->phys_obj = dev_priv->mm.phys_objs[id - 1];
	obj->phys_obj->cur_obj = obj;
4705

4706
	page_count = obj->base.size / PAGE_SIZE;
4707 4708

	for (i = 0; i < page_count; i++) {
4709 4710 4711
		struct page *page;
		char *dst, *src;

4712
		page = shmem_read_mapping_page(mapping, i);
4713 4714
		if (IS_ERR(page))
			return PTR_ERR(page);
4715

4716
		src = kmap_atomic(page);
4717
		dst = obj->phys_obj->handle->vaddr + (i * PAGE_SIZE);
4718
		memcpy(dst, src, PAGE_SIZE);
P
Peter Zijlstra 已提交
4719
		kunmap_atomic(src);
4720

4721 4722 4723
		mark_page_accessed(page);
		page_cache_release(page);
	}
4724

4725 4726 4727 4728
	return 0;
}

static int
4729 4730
i915_gem_phys_pwrite(struct drm_device *dev,
		     struct drm_i915_gem_object *obj,
4731 4732 4733
		     struct drm_i915_gem_pwrite *args,
		     struct drm_file *file_priv)
{
4734
	void *vaddr = obj->phys_obj->handle->vaddr + args->offset;
V
Ville Syrjälä 已提交
4735
	char __user *user_data = to_user_ptr(args->data_ptr);
4736

4737 4738 4739 4740 4741 4742 4743 4744 4745 4746 4747 4748 4749
	if (__copy_from_user_inatomic_nocache(vaddr, user_data, args->size)) {
		unsigned long unwritten;

		/* The physical object once assigned is fixed for the lifetime
		 * of the obj, so we can safely drop the lock and continue
		 * to access vaddr.
		 */
		mutex_unlock(&dev->struct_mutex);
		unwritten = copy_from_user(vaddr, user_data, args->size);
		mutex_lock(&dev->struct_mutex);
		if (unwritten)
			return -EFAULT;
	}
4750

4751
	i915_gem_chipset_flush(dev);
4752 4753
	return 0;
}
4754

4755
void i915_gem_release(struct drm_device *dev, struct drm_file *file)
4756
{
4757
	struct drm_i915_file_private *file_priv = file->driver_priv;
4758 4759 4760 4761 4762

	/* Clean up our request list when the client is going away, so that
	 * later retire_requests won't dereference our soon-to-be-gone
	 * file_priv.
	 */
4763
	spin_lock(&file_priv->mm.lock);
4764 4765 4766 4767 4768 4769 4770 4771 4772
	while (!list_empty(&file_priv->mm.request_list)) {
		struct drm_i915_gem_request *request;

		request = list_first_entry(&file_priv->mm.request_list,
					   struct drm_i915_gem_request,
					   client_list);
		list_del(&request->client_list);
		request->file_priv = NULL;
	}
4773
	spin_unlock(&file_priv->mm.lock);
4774
}
4775

4776 4777 4778 4779 4780 4781 4782 4783 4784 4785 4786 4787 4788
static bool mutex_is_locked_by(struct mutex *mutex, struct task_struct *task)
{
	if (!mutex_is_locked(mutex))
		return false;

#if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_MUTEXES)
	return mutex->owner == task;
#else
	/* Since UP may be pre-empted, we cannot assume that we own the lock */
	return false;
#endif
}

4789
static int
4790
i915_gem_inactive_shrink(struct shrinker *shrinker, struct shrink_control *sc)
4791
{
4792 4793 4794 4795 4796
	struct drm_i915_private *dev_priv =
		container_of(shrinker,
			     struct drm_i915_private,
			     mm.inactive_shrinker);
	struct drm_device *dev = dev_priv->dev;
C
Chris Wilson 已提交
4797
	struct drm_i915_gem_object *obj;
4798
	int nr_to_scan = sc->nr_to_scan;
4799
	bool unlock = true;
4800 4801
	int cnt;

4802 4803 4804 4805
	if (!mutex_trylock(&dev->struct_mutex)) {
		if (!mutex_is_locked_by(&dev->struct_mutex, current))
			return 0;

4806 4807 4808
		if (dev_priv->mm.shrinker_no_lock_stealing)
			return 0;

4809 4810
		unlock = false;
	}
4811

C
Chris Wilson 已提交
4812 4813
	if (nr_to_scan) {
		nr_to_scan -= i915_gem_purge(dev_priv, nr_to_scan);
4814 4815 4816
		if (nr_to_scan > 0)
			nr_to_scan -= __i915_gem_shrink(dev_priv, nr_to_scan,
							false);
C
Chris Wilson 已提交
4817 4818
		if (nr_to_scan > 0)
			i915_gem_shrink_all(dev_priv);
4819 4820
	}

4821
	cnt = 0;
4822
	list_for_each_entry(obj, &dev_priv->mm.unbound_list, global_list)
4823 4824
		if (obj->pages_pin_count == 0)
			cnt += obj->base.size >> PAGE_SHIFT;
4825 4826 4827 4828 4829

	list_for_each_entry(obj, &dev_priv->mm.bound_list, global_list) {
		if (obj->active)
			continue;

4830
		if (obj->pin_count == 0 && obj->pages_pin_count == 0)
C
Chris Wilson 已提交
4831
			cnt += obj->base.size >> PAGE_SHIFT;
4832
	}
4833

4834 4835
	if (unlock)
		mutex_unlock(&dev->struct_mutex);
C
Chris Wilson 已提交
4836
	return cnt;
4837
}
4838 4839 4840 4841 4842 4843 4844 4845 4846 4847 4848 4849 4850 4851 4852 4853 4854 4855 4856 4857 4858 4859 4860 4861 4862 4863

/* All the new VM stuff */
unsigned long i915_gem_obj_offset(struct drm_i915_gem_object *o,
				  struct i915_address_space *vm)
{
	struct drm_i915_private *dev_priv = o->base.dev->dev_private;
	struct i915_vma *vma;

	if (vm == &dev_priv->mm.aliasing_ppgtt->base)
		vm = &dev_priv->gtt.base;

	BUG_ON(list_empty(&o->vma_list));
	list_for_each_entry(vma, &o->vma_list, vma_link) {
		if (vma->vm == vm)
			return vma->node.start;

	}
	return -1;
}

bool i915_gem_obj_bound(struct drm_i915_gem_object *o,
			struct i915_address_space *vm)
{
	struct i915_vma *vma;

	list_for_each_entry(vma, &o->vma_list, vma_link)
4864
		if (vma->vm == vm && drm_mm_node_allocated(&vma->node))
4865 4866 4867 4868 4869 4870 4871 4872 4873 4874 4875 4876 4877 4878 4879 4880 4881 4882 4883 4884 4885 4886 4887 4888 4889 4890 4891 4892 4893 4894 4895 4896 4897 4898
			return true;

	return false;
}

bool i915_gem_obj_bound_any(struct drm_i915_gem_object *o)
{
	struct drm_i915_private *dev_priv = o->base.dev->dev_private;
	struct i915_address_space *vm;

	list_for_each_entry(vm, &dev_priv->vm_list, global_link)
		if (i915_gem_obj_bound(o, vm))
			return true;

	return false;
}

unsigned long i915_gem_obj_size(struct drm_i915_gem_object *o,
				struct i915_address_space *vm)
{
	struct drm_i915_private *dev_priv = o->base.dev->dev_private;
	struct i915_vma *vma;

	if (vm == &dev_priv->mm.aliasing_ppgtt->base)
		vm = &dev_priv->gtt.base;

	BUG_ON(list_empty(&o->vma_list));

	list_for_each_entry(vma, &o->vma_list, vma_link)
		if (vma->vm == vm)
			return vma->node.size;

	return 0;
}