i915_gem_mman.c 20.8 KB
Newer Older
1 2 3 4 5 6 7
/*
 * SPDX-License-Identifier: MIT
 *
 * Copyright © 2014-2016 Intel Corporation
 */

#include <linux/mman.h>
8
#include <linux/pfn_t.h>
9 10
#include <linux/sizes.h>

11
#include "gt/intel_gt.h"
12
#include "gt/intel_gt_requests.h"
13

14 15 16 17
#include "i915_drv.h"
#include "i915_gem_gtt.h"
#include "i915_gem_ioctls.h"
#include "i915_gem_object.h"
18
#include "i915_gem_mman.h"
19
#include "i915_trace.h"
20
#include "i915_user_extensions.h"
21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149
#include "i915_vma.h"

static inline bool
__vma_matches(struct vm_area_struct *vma, struct file *filp,
	      unsigned long addr, unsigned long size)
{
	if (vma->vm_file != filp)
		return false;

	return vma->vm_start == addr &&
	       (vma->vm_end - vma->vm_start) == PAGE_ALIGN(size);
}

/**
 * i915_gem_mmap_ioctl - Maps the contents of an object, returning the address
 *			 it is mapped to.
 * @dev: drm device
 * @data: ioctl data blob
 * @file: drm file
 *
 * While the mapping holds a reference on the contents of the object, it doesn't
 * imply a ref on the object itself.
 *
 * IMPORTANT:
 *
 * DRM driver writers who look a this function as an example for how to do GEM
 * mmap support, please don't implement mmap support like here. The modern way
 * to implement DRM mmap support is with an mmap offset ioctl (like
 * i915_gem_mmap_gtt) and then using the mmap syscall on the DRM fd directly.
 * That way debug tooling like valgrind will understand what's going on, hiding
 * the mmap call in a driver private ioctl will break that. The i915 driver only
 * does cpu mmaps this way because we didn't know better.
 */
int
i915_gem_mmap_ioctl(struct drm_device *dev, void *data,
		    struct drm_file *file)
{
	struct drm_i915_gem_mmap *args = data;
	struct drm_i915_gem_object *obj;
	unsigned long addr;

	if (args->flags & ~(I915_MMAP_WC))
		return -EINVAL;

	if (args->flags & I915_MMAP_WC && !boot_cpu_has(X86_FEATURE_PAT))
		return -ENODEV;

	obj = i915_gem_object_lookup(file, args->handle);
	if (!obj)
		return -ENOENT;

	/* prime objects have no backing filp to GEM mmap
	 * pages from.
	 */
	if (!obj->base.filp) {
		addr = -ENXIO;
		goto err;
	}

	if (range_overflows(args->offset, args->size, (u64)obj->base.size)) {
		addr = -EINVAL;
		goto err;
	}

	addr = vm_mmap(obj->base.filp, 0, args->size,
		       PROT_READ | PROT_WRITE, MAP_SHARED,
		       args->offset);
	if (IS_ERR_VALUE(addr))
		goto err;

	if (args->flags & I915_MMAP_WC) {
		struct mm_struct *mm = current->mm;
		struct vm_area_struct *vma;

		if (down_write_killable(&mm->mmap_sem)) {
			addr = -EINTR;
			goto err;
		}
		vma = find_vma(mm, addr);
		if (vma && __vma_matches(vma, obj->base.filp, addr, args->size))
			vma->vm_page_prot =
				pgprot_writecombine(vm_get_page_prot(vma->vm_flags));
		else
			addr = -ENOMEM;
		up_write(&mm->mmap_sem);
		if (IS_ERR_VALUE(addr))
			goto err;
	}
	i915_gem_object_put(obj);

	args->addr_ptr = (u64)addr;
	return 0;

err:
	i915_gem_object_put(obj);
	return addr;
}

static unsigned int tile_row_pages(const struct drm_i915_gem_object *obj)
{
	return i915_gem_object_get_tile_row_size(obj) >> PAGE_SHIFT;
}

/**
 * i915_gem_mmap_gtt_version - report the current feature set for GTT mmaps
 *
 * A history of the GTT mmap interface:
 *
 * 0 - Everything had to fit into the GTT. Both parties of a memcpy had to
 *     aligned and suitable for fencing, and still fit into the available
 *     mappable space left by the pinned display objects. A classic problem
 *     we called the page-fault-of-doom where we would ping-pong between
 *     two objects that could not fit inside the GTT and so the memcpy
 *     would page one object in at the expense of the other between every
 *     single byte.
 *
 * 1 - Objects can be any size, and have any compatible fencing (X Y, or none
 *     as set via i915_gem_set_tiling() [DRM_I915_GEM_SET_TILING]). If the
 *     object is too large for the available space (or simply too large
 *     for the mappable aperture!), a view is created instead and faulted
 *     into userspace. (This view is aligned and sized appropriately for
 *     fenced access.)
 *
 * 2 - Recognise WC as a separate cache domain so that we can flush the
 *     delayed writes via GTT before performing direct access via WC.
 *
 * 3 - Remove implicit set-domain(GTT) and synchronisation on initial
 *     pagefault; swapin remains transparent.
 *
150 151 152
 * 4 - Support multiple fault handlers per object depending on object's
 *     backing storage (a.k.a. MMAP_OFFSET).
 *
153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179
 * Restrictions:
 *
 *  * snoopable objects cannot be accessed via the GTT. It can cause machine
 *    hangs on some architectures, corruption on others. An attempt to service
 *    a GTT page fault from a snoopable object will generate a SIGBUS.
 *
 *  * the object must be able to fit into RAM (physical memory, though no
 *    limited to the mappable aperture).
 *
 *
 * Caveats:
 *
 *  * a new GTT page fault will synchronize rendering from the GPU and flush
 *    all data to system memory. Subsequent access will not be synchronized.
 *
 *  * all mappings are revoked on runtime device suspend.
 *
 *  * there are only 8, 16 or 32 fence registers to share between all users
 *    (older machines require fence register for display and blitter access
 *    as well). Contention of the fence registers will cause the previous users
 *    to be unmapped and any new access will generate new page faults.
 *
 *  * running out of memory while servicing a fault may generate a SIGBUS,
 *    rather than the expected SIGSEGV.
 */
int i915_gem_mmap_gtt_version(void)
{
180
	return 4;
181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205
}

static inline struct i915_ggtt_view
compute_partial_view(const struct drm_i915_gem_object *obj,
		     pgoff_t page_offset,
		     unsigned int chunk)
{
	struct i915_ggtt_view view;

	if (i915_gem_object_is_tiled(obj))
		chunk = roundup(chunk, tile_row_pages(obj));

	view.type = I915_GGTT_VIEW_PARTIAL;
	view.partial.offset = rounddown(page_offset, chunk);
	view.partial.size =
		min_t(unsigned int, chunk,
		      (obj->base.size >> PAGE_SHIFT) - view.partial.offset);

	/* If the partial covers the entire object, just create a normal VMA. */
	if (chunk >= obj->base.size >> PAGE_SHIFT)
		view.type = I915_GGTT_VIEW_NORMAL;

	return view;
}

206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240
static vm_fault_t i915_error_to_vmf_fault(int err)
{
	switch (err) {
	default:
		WARN_ONCE(err, "unhandled error in %s: %i\n", __func__, err);
		/* fallthrough */
	case -EIO: /* shmemfs failure from swap device */
	case -EFAULT: /* purged object */
	case -ENODEV: /* bad object, how did you get here! */
		return VM_FAULT_SIGBUS;

	case -ENOSPC: /* shmemfs allocation failure */
	case -ENOMEM: /* our allocation failure */
		return VM_FAULT_OOM;

	case 0:
	case -EAGAIN:
	case -ERESTARTSYS:
	case -EINTR:
	case -EBUSY:
		/*
		 * EBUSY is ok: this just means that another thread
		 * already did the job.
		 */
		return VM_FAULT_NOPAGE;
	}
}

static vm_fault_t vm_fault_cpu(struct vm_fault *vmf)
{
	struct vm_area_struct *area = vmf->vma;
	struct i915_mmap_offset *mmo = area->vm_private_data;
	struct drm_i915_gem_object *obj = mmo->obj;
	int err;

241 242
	if (unlikely(!i915_gem_object_has_struct_page(obj)))
		return VM_FAULT_SIGBUS;
243 244

	/* Sanity check that we allow writing into this object */
245 246 247
	if (unlikely(i915_gem_object_is_readonly(obj) &&
		     area->vm_flags & VM_WRITE))
		return VM_FAULT_SIGBUS;
248 249 250

	err = i915_gem_object_pin_pages(obj);
	if (err)
251
		goto out;
252 253

	/* PTEs are revoked in obj->ops->put_pages() */
254 255 256
	err = remap_io_sg_page(area,
			       area->vm_start, area->vm_end - area->vm_start,
			       obj->mm.pages->sgl);
257

258
	if (area->vm_flags & VM_WRITE) {
259 260 261 262 263 264
		GEM_BUG_ON(!i915_gem_object_has_pinned_pages(obj));
		obj->mm.dirty = true;
	}

	i915_gem_object_unpin_pages(obj);

265 266
out:
	return i915_error_to_vmf_fault(err);
267 268 269
}

static vm_fault_t vm_fault_gtt(struct vm_fault *vmf)
270 271 272
{
#define MIN_CHUNK_PAGES (SZ_1M >> PAGE_SHIFT)
	struct vm_area_struct *area = vmf->vma;
273 274
	struct i915_mmap_offset *mmo = area->vm_private_data;
	struct drm_i915_gem_object *obj = mmo->obj;
275 276
	struct drm_device *dev = obj->base.dev;
	struct drm_i915_private *i915 = to_i915(dev);
277
	struct intel_runtime_pm *rpm = &i915->runtime_pm;
278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298
	struct i915_ggtt *ggtt = &i915->ggtt;
	bool write = area->vm_flags & VM_WRITE;
	intel_wakeref_t wakeref;
	struct i915_vma *vma;
	pgoff_t page_offset;
	int srcu;
	int ret;

	/* Sanity check that we allow writing into this object */
	if (i915_gem_object_is_readonly(obj) && write)
		return VM_FAULT_SIGBUS;

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

	trace_i915_gem_object_fault(obj, page_offset, true, write);

	ret = i915_gem_object_pin_pages(obj);
	if (ret)
		goto err;

299
	wakeref = intel_runtime_pm_get(rpm);
300

301 302
	ret = intel_gt_reset_trylock(ggtt->vm.gt, &srcu);
	if (ret)
303 304 305 306 307
		goto err_rpm;

	/* Now pin it into the GTT as needed */
	vma = i915_gem_object_ggtt_pin(obj, NULL, 0, 0,
				       PIN_MAPPABLE |
308
				       PIN_NONBLOCK /* NOWARN */ |
309
				       PIN_NOEVICT);
310 311 312 313 314 315
	if (IS_ERR(vma)) {
		/* Use a partial view if it is bigger than available space */
		struct i915_ggtt_view view =
			compute_partial_view(obj, page_offset, MIN_CHUNK_PAGES);
		unsigned int flags;

316
		flags = PIN_MAPPABLE | PIN_NOSEARCH;
317 318 319 320 321 322 323 324 325
		if (view.type == I915_GGTT_VIEW_NORMAL)
			flags |= PIN_NONBLOCK; /* avoid warnings for pinned */

		/*
		 * Userspace is now writing through an untracked VMA, abandon
		 * all hope that the hardware is able to track future writes.
		 */

		vma = i915_gem_object_ggtt_pin(obj, &view, 0, 0, flags);
326
		if (IS_ERR(vma)) {
327 328 329 330
			flags = PIN_MAPPABLE;
			view.type = I915_GGTT_VIEW_PARTIAL;
			vma = i915_gem_object_ggtt_pin(obj, &view, 0, 0, flags);
		}
331 332 333

		/* The entire mappable GGTT is pinned? Unexpected! */
		GEM_BUG_ON(vma == ERR_PTR(-ENOSPC));
334 335 336
	}
	if (IS_ERR(vma)) {
		ret = PTR_ERR(vma);
337 338 339 340 341 342 343
		goto err_reset;
	}

	/* Access to snoopable pages through the GTT is incoherent. */
	if (obj->cache_level != I915_CACHE_NONE && !HAS_LLC(i915)) {
		ret = -EFAULT;
		goto err_unpin;
344 345 346 347 348 349 350 351 352 353 354 355 356 357 358
	}

	ret = i915_vma_pin_fence(vma);
	if (ret)
		goto err_unpin;

	/* Finally, remap it using the new GTT offset */
	ret = remap_io_mapping(area,
			       area->vm_start + (vma->ggtt_view.partial.offset << PAGE_SHIFT),
			       (ggtt->gmadr.start + vma->node.start) >> PAGE_SHIFT,
			       min_t(u64, vma->size, area->vm_end - area->vm_start),
			       &ggtt->iomap);
	if (ret)
		goto err_fence;

359
	assert_rpm_wakelock_held(rpm);
360 361 362

	/* Mark as being mmapped into userspace for later revocation */
	mutex_lock(&i915->ggtt.vm.mutex);
363
	if (!i915_vma_set_userfault(vma) && !obj->userfault_count++)
364
		list_add(&obj->userfault_link, &i915->ggtt.userfault_list);
365 366
	mutex_unlock(&i915->ggtt.vm.mutex);

367 368 369
	/* Track the mmo associated with the fenced vma */
	vma->mmo = mmo;

370
	if (IS_ACTIVE(CONFIG_DRM_I915_USERFAULT_AUTOSUSPEND))
371
		intel_wakeref_auto(&i915->ggtt.userfault_wakeref,
372 373
				   msecs_to_jiffies_timeout(CONFIG_DRM_I915_USERFAULT_AUTOSUSPEND));

374 375 376 377 378
	if (write) {
		GEM_BUG_ON(!i915_gem_object_has_pinned_pages(obj));
		i915_vma_set_ggtt_write(vma);
		obj->mm.dirty = true;
	}
379 380 381 382 383 384

err_fence:
	i915_vma_unpin_fence(vma);
err_unpin:
	__i915_vma_unpin(vma);
err_reset:
385
	intel_gt_reset_unlock(ggtt->vm.gt, srcu);
386
err_rpm:
387
	intel_runtime_pm_put(rpm, wakeref);
388 389
	i915_gem_object_unpin_pages(obj);
err:
390
	return i915_error_to_vmf_fault(ret);
391 392
}

393
void __i915_gem_object_release_mmap_gtt(struct drm_i915_gem_object *obj)
394 395 396 397 398 399
{
	struct i915_vma *vma;

	GEM_BUG_ON(!obj->userfault_count);

	for_each_ggtt_vma(vma, obj)
400 401 402
		i915_vma_revoke_mmap(vma);

	GEM_BUG_ON(obj->userfault_count);
403 404
}

405
/*
406 407 408 409 410
 * 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
411
 * fixup by vm_fault_gtt().
412
 */
413
static void i915_gem_object_release_mmap_gtt(struct drm_i915_gem_object *obj)
414 415 416 417
{
	struct drm_i915_private *i915 = to_i915(obj->base.dev);
	intel_wakeref_t wakeref;

418 419
	/*
	 * Serialisation between user GTT access and our code depends upon
420 421 422 423 424 425 426
	 * revoking the CPU's PTE whilst the mutex is held. The next user
	 * pagefault then has to wait until we release the mutex.
	 *
	 * Note that RPM complicates somewhat by adding an additional
	 * requirement that operations to the GGTT be made holding the RPM
	 * wakeref.
	 */
427
	wakeref = intel_runtime_pm_get(&i915->runtime_pm);
428
	mutex_lock(&i915->ggtt.vm.mutex);
429 430 431 432

	if (!obj->userfault_count)
		goto out;

433
	__i915_gem_object_release_mmap_gtt(obj);
434

435 436
	/*
	 * Ensure that the CPU's PTE are revoked and there are not outstanding
437 438 439 440 441 442 443 444 445
	 * memory transactions from userspace before we return. The TLB
	 * flushing implied above by changing the PTE above *should* be
	 * sufficient, an extra barrier here just provides us with a bit
	 * of paranoid documentation about our requirement to serialise
	 * memory writes before touching registers / GSM.
	 */
	wmb();

out:
446
	mutex_unlock(&i915->ggtt.vm.mutex);
447
	intel_runtime_pm_put(&i915->runtime_pm, wakeref);
448 449
}

450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487
void i915_gem_object_release_mmap_offset(struct drm_i915_gem_object *obj)
{
	struct i915_mmap_offset *mmo;

	spin_lock(&obj->mmo.lock);
	list_for_each_entry(mmo, &obj->mmo.offsets, offset) {
		/*
		 * vma_node_unmap for GTT mmaps handled already in
		 * __i915_gem_object_release_mmap_gtt
		 */
		if (mmo->mmap_type == I915_MMAP_TYPE_GTT)
			continue;

		spin_unlock(&obj->mmo.lock);
		drm_vma_node_unmap(&mmo->vma_node,
				   obj->base.dev->anon_inode->i_mapping);
		spin_lock(&obj->mmo.lock);
	}
	spin_unlock(&obj->mmo.lock);
}

/**
 * i915_gem_object_release_mmap - remove physical page mappings
 * @obj: obj in question
 *
 * Preserve the reservation of the mmapping with the DRM core code, but
 * relinquish ownership of the pages back to the system.
 */
void i915_gem_object_release_mmap(struct drm_i915_gem_object *obj)
{
	i915_gem_object_release_mmap_gtt(obj);
	i915_gem_object_release_mmap_offset(obj);
}

static struct i915_mmap_offset *
mmap_offset_attach(struct drm_i915_gem_object *obj,
		   enum i915_mmap_type mmap_type,
		   struct drm_file *file)
488 489
{
	struct drm_i915_private *i915 = to_i915(obj->base.dev);
490
	struct i915_mmap_offset *mmo;
491 492
	int err;

493 494 495 496 497 498 499 500 501 502 503 504
	mmo = kmalloc(sizeof(*mmo), GFP_KERNEL);
	if (!mmo)
		return ERR_PTR(-ENOMEM);

	mmo->obj = obj;
	mmo->dev = obj->base.dev;
	mmo->file = file;
	mmo->mmap_type = mmap_type;
	drm_vma_node_reset(&mmo->vma_node);

	err = drm_vma_offset_add(mmo->dev->vma_offset_manager, &mmo->vma_node,
				 obj->base.size / PAGE_SIZE);
505
	if (likely(!err))
506
		goto out;
507 508

	/* Attempt to reap some mmap space from dead objects */
509
	err = intel_gt_retire_requests_timeout(&i915->gt, MAX_SCHEDULE_TIMEOUT);
510
	if (err)
511
		goto err;
512

513
	i915_gem_drain_freed_objects(i915);
514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531
	err = drm_vma_offset_add(mmo->dev->vma_offset_manager, &mmo->vma_node,
				 obj->base.size / PAGE_SIZE);
	if (err)
		goto err;

out:
	if (file)
		drm_vma_node_allow(&mmo->vma_node, file);

	spin_lock(&obj->mmo.lock);
	list_add(&mmo->offset, &obj->mmo.offsets);
	spin_unlock(&obj->mmo.lock);

	return mmo;

err:
	kfree(mmo);
	return ERR_PTR(err);
532 533
}

534 535 536 537 538
static int
__assign_mmap_offset(struct drm_file *file,
		     u32 handle,
		     enum i915_mmap_type mmap_type,
		     u64 *offset)
539 540
{
	struct drm_i915_gem_object *obj;
541 542
	struct i915_mmap_offset *mmo;
	int err;
543

544 545 546 547
	obj = i915_gem_object_lookup(file, handle);
	if (!obj)
		return -ENOENT;

548 549 550
	if (mmap_type == I915_MMAP_TYPE_GTT &&
	    i915_gem_object_never_bind_ggtt(obj)) {
		err = -ENODEV;
551 552 553
		goto out;
	}

554 555 556 557 558 559 560 561 562 563 564
	if (mmap_type != I915_MMAP_TYPE_GTT &&
	    !i915_gem_object_has_struct_page(obj)) {
		err = -ENODEV;
		goto out;
	}

	mmo = mmap_offset_attach(obj, mmap_type, file);
	if (IS_ERR(mmo)) {
		err = PTR_ERR(mmo);
		goto out;
	}
565

566 567
	*offset = drm_vma_node_offset_addr(&mmo->vma_node);
	err = 0;
568
out:
569
	i915_gem_object_put(obj);
570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588
	return err;
}

int
i915_gem_dumb_mmap_offset(struct drm_file *file,
			  struct drm_device *dev,
			  u32 handle,
			  u64 *offset)
{
	enum i915_mmap_type mmap_type;

	if (boot_cpu_has(X86_FEATURE_PAT))
		mmap_type = I915_MMAP_TYPE_WC;
	else if (!i915_ggtt_has_aperture(&to_i915(dev)->ggtt))
		return -ENODEV;
	else
		mmap_type = I915_MMAP_TYPE_GTT;

	return __assign_mmap_offset(file, handle, mmap_type, offset);
589 590 591
}

/**
592
 * i915_gem_mmap_offset_ioctl - prepare an object for GTT mmap'ing
593 594 595 596 597 598 599 600 601 602 603 604 605 606
 * @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
607 608
i915_gem_mmap_offset_ioctl(struct drm_device *dev, void *data,
			   struct drm_file *file)
609
{
610 611 612
	struct drm_i915_private *i915 = to_i915(dev);
	struct drm_i915_gem_mmap_offset *args = data;
	enum i915_mmap_type type;
613
	int err;
614

615 616 617 618 619 620 621 622 623
	/*
	 * Historically we failed to check args.pad and args.offset
	 * and so we cannot use those fields for user input and we cannot
	 * add -EINVAL for them as the ABI is fixed, i.e. old userspace
	 * may be feeding in garbage in those fields.
	 *
	 * if (args->pad) return -EINVAL; is verbotten!
	 */

624 625 626 627
	err = i915_user_extensions(u64_to_user_ptr(args->extensions),
				   NULL, 0, NULL);
	if (err)
		return err;
628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654

	switch (args->flags) {
	case I915_MMAP_OFFSET_GTT:
		if (!i915_ggtt_has_aperture(&i915->ggtt))
			return -ENODEV;
		type = I915_MMAP_TYPE_GTT;
		break;

	case I915_MMAP_OFFSET_WC:
		if (!boot_cpu_has(X86_FEATURE_PAT))
			return -ENODEV;
		type = I915_MMAP_TYPE_WC;
		break;

	case I915_MMAP_OFFSET_WB:
		type = I915_MMAP_TYPE_WB;
		break;

	case I915_MMAP_OFFSET_UC:
		if (!boot_cpu_has(X86_FEATURE_PAT))
			return -ENODEV;
		type = I915_MMAP_TYPE_UC;
		break;

	default:
		return -EINVAL;
	}
655

656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779
	return __assign_mmap_offset(file, args->handle, type, &args->offset);
}

static void vm_open(struct vm_area_struct *vma)
{
	struct i915_mmap_offset *mmo = vma->vm_private_data;
	struct drm_i915_gem_object *obj = mmo->obj;

	GEM_BUG_ON(!obj);
	i915_gem_object_get(obj);
}

static void vm_close(struct vm_area_struct *vma)
{
	struct i915_mmap_offset *mmo = vma->vm_private_data;
	struct drm_i915_gem_object *obj = mmo->obj;

	GEM_BUG_ON(!obj);
	i915_gem_object_put(obj);
}

static const struct vm_operations_struct vm_ops_gtt = {
	.fault = vm_fault_gtt,
	.open = vm_open,
	.close = vm_close,
};

static const struct vm_operations_struct vm_ops_cpu = {
	.fault = vm_fault_cpu,
	.open = vm_open,
	.close = vm_close,
};

/*
 * This overcomes the limitation in drm_gem_mmap's assignment of a
 * drm_gem_object as the vma->vm_private_data. Since we need to
 * be able to resolve multiple mmap offsets which could be tied
 * to a single gem object.
 */
int i915_gem_mmap(struct file *filp, struct vm_area_struct *vma)
{
	struct drm_vma_offset_node *node;
	struct drm_file *priv = filp->private_data;
	struct drm_device *dev = priv->minor->dev;
	struct i915_mmap_offset *mmo = NULL;
	struct drm_gem_object *obj = NULL;

	if (drm_dev_is_unplugged(dev))
		return -ENODEV;

	drm_vma_offset_lock_lookup(dev->vma_offset_manager);
	node = drm_vma_offset_exact_lookup_locked(dev->vma_offset_manager,
						  vma->vm_pgoff,
						  vma_pages(vma));
	if (likely(node)) {
		mmo = container_of(node, struct i915_mmap_offset,
				   vma_node);
		/*
		 * In our dependency chain, the drm_vma_offset_node
		 * depends on the validity of the mmo, which depends on
		 * the gem object. However the only reference we have
		 * at this point is the mmo (as the parent of the node).
		 * Try to check if the gem object was at least cleared.
		 */
		if (!mmo || !mmo->obj) {
			drm_vma_offset_unlock_lookup(dev->vma_offset_manager);
			return -EINVAL;
		}
		/*
		 * Skip 0-refcnted objects as it is in the process of being
		 * destroyed and will be invalid when the vma manager lock
		 * is released.
		 */
		obj = &mmo->obj->base;
		if (!kref_get_unless_zero(&obj->refcount))
			obj = NULL;
	}
	drm_vma_offset_unlock_lookup(dev->vma_offset_manager);
	if (!obj)
		return -EINVAL;

	if (!drm_vma_node_is_allowed(node, priv)) {
		drm_gem_object_put_unlocked(obj);
		return -EACCES;
	}

	if (i915_gem_object_is_readonly(to_intel_bo(obj))) {
		if (vma->vm_flags & VM_WRITE) {
			drm_gem_object_put_unlocked(obj);
			return -EINVAL;
		}
		vma->vm_flags &= ~VM_MAYWRITE;
	}

	vma->vm_flags |= VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP;
	vma->vm_private_data = mmo;

	switch (mmo->mmap_type) {
	case I915_MMAP_TYPE_WC:
		vma->vm_page_prot =
			pgprot_writecombine(vm_get_page_prot(vma->vm_flags));
		vma->vm_ops = &vm_ops_cpu;
		break;

	case I915_MMAP_TYPE_WB:
		vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
		vma->vm_ops = &vm_ops_cpu;
		break;

	case I915_MMAP_TYPE_UC:
		vma->vm_page_prot =
			pgprot_noncached(vm_get_page_prot(vma->vm_flags));
		vma->vm_ops = &vm_ops_cpu;
		break;

	case I915_MMAP_TYPE_GTT:
		vma->vm_page_prot =
			pgprot_writecombine(vm_get_page_prot(vma->vm_flags));
		vma->vm_ops = &vm_ops_gtt;
		break;
	}
	vma->vm_page_prot = pgprot_decrypted(vma->vm_page_prot);

	return 0;
780 781 782 783 784
}

#if IS_ENABLED(CONFIG_DRM_I915_SELFTEST)
#include "selftests/i915_gem_mman.c"
#endif