intel-svm.c 19.0 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
/*
 * Copyright © 2015 Intel Corporation.
 *
 * This program is free software; you can redistribute it and/or modify it
 * under the terms and conditions of the GNU General Public License,
 * version 2, as published by the Free Software Foundation.
 *
 * This program is distributed in the hope it will be useful, but WITHOUT
 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
 * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
 * more details.
 *
 * Authors: David Woodhouse <dwmw2@infradead.org>
 */

#include <linux/intel-iommu.h>
17 18
#include <linux/mmu_notifier.h>
#include <linux/sched.h>
19
#include <linux/sched/mm.h>
20 21 22 23 24
#include <linux/slab.h>
#include <linux/intel-svm.h>
#include <linux/rculist.h>
#include <linux/pci.h>
#include <linux/pci-ats.h>
25 26
#include <linux/dmar.h>
#include <linux/interrupt.h>
27
#include <asm/page.h>
28

29 30 31 32
#define PASID_ENTRY_P		BIT_ULL(0)
#define PASID_ENTRY_FLPM_5LP	BIT_ULL(9)
#define PASID_ENTRY_SRE		BIT_ULL(11)

33
static irqreturn_t prq_event_thread(int irq, void *d);
34 35 36 37

struct pasid_entry {
	u64 val;
};
38

39 40 41 42
struct pasid_state_entry {
	u64 val;
};

43 44 45 46 47
int intel_svm_alloc_pasid_tables(struct intel_iommu *iommu)
{
	struct page *pages;
	int order;

48 49 50 51
	if (cpu_feature_enabled(X86_FEATURE_GBPAGES) &&
			!cap_fl1gp_support(iommu->cap))
		return -EINVAL;

52 53 54 55
	if (cpu_feature_enabled(X86_FEATURE_LA57) &&
			!cap_5lp_support(iommu->cap))
		return -EINVAL;

56 57 58 59 60 61 62 63 64 65 66 67
	/* Start at 2 because it's defined as 2^(1+PSS) */
	iommu->pasid_max = 2 << ecap_pss(iommu->ecap);

	/* Eventually I'm promised we will get a multi-level PASID table
	 * and it won't have to be physically contiguous. Until then,
	 * limit the size because 8MiB contiguous allocations can be hard
	 * to come by. The limit of 0x20000, which is 1MiB for each of
	 * the PASID and PASID-state tables, is somewhat arbitrary. */
	if (iommu->pasid_max > 0x20000)
		iommu->pasid_max = 0x20000;

	order = get_order(sizeof(struct pasid_entry) * iommu->pasid_max);
68 69 70 71 72 73 74 75 76 77
	pages = alloc_pages(GFP_KERNEL | __GFP_ZERO, order);
	if (!pages) {
		pr_warn("IOMMU: %s: Failed to allocate PASID table\n",
			iommu->name);
		return -ENOMEM;
	}
	iommu->pasid_table = page_address(pages);
	pr_info("%s: Allocated order %d PASID table.\n", iommu->name, order);

	if (ecap_dis(iommu->ecap)) {
78 79
		/* Just making it explicit... */
		BUILD_BUG_ON(sizeof(struct pasid_entry) != sizeof(struct pasid_state_entry));
80 81 82 83 84 85 86 87
		pages = alloc_pages(GFP_KERNEL | __GFP_ZERO, order);
		if (pages)
			iommu->pasid_state_table = page_address(pages);
		else
			pr_warn("IOMMU: %s: Failed to allocate PASID state table\n",
				iommu->name);
	}

88 89
	idr_init(&iommu->pasid_idr);

90 91 92 93 94
	return 0;
}

int intel_svm_free_pasid_tables(struct intel_iommu *iommu)
{
95
	int order = get_order(sizeof(struct pasid_entry) * iommu->pasid_max);
96 97 98 99 100 101 102 103 104

	if (iommu->pasid_table) {
		free_pages((unsigned long)iommu->pasid_table, order);
		iommu->pasid_table = NULL;
	}
	if (iommu->pasid_state_table) {
		free_pages((unsigned long)iommu->pasid_state_table, order);
		iommu->pasid_state_table = NULL;
	}
105
	idr_destroy(&iommu->pasid_idr);
106 107
	return 0;
}
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
#define PRQ_ORDER 0

int intel_svm_enable_prq(struct intel_iommu *iommu)
{
	struct page *pages;
	int irq, ret;

	pages = alloc_pages(GFP_KERNEL | __GFP_ZERO, PRQ_ORDER);
	if (!pages) {
		pr_warn("IOMMU: %s: Failed to allocate page request queue\n",
			iommu->name);
		return -ENOMEM;
	}
	iommu->prq = page_address(pages);

	irq = dmar_alloc_hwirq(DMAR_UNITS_SUPPORTED + iommu->seq_id, iommu->node, iommu);
	if (irq <= 0) {
		pr_err("IOMMU: %s: Failed to create IRQ vector for page request queue\n",
		       iommu->name);
		ret = -EINVAL;
	err:
		free_pages((unsigned long)iommu->prq, PRQ_ORDER);
		iommu->prq = NULL;
		return ret;
	}
	iommu->pr_irq = irq;

	snprintf(iommu->prq_name, sizeof(iommu->prq_name), "dmar%d-prq", iommu->seq_id);

	ret = request_threaded_irq(irq, NULL, prq_event_thread, IRQF_ONESHOT,
				   iommu->prq_name, iommu);
	if (ret) {
		pr_err("IOMMU: %s: Failed to request IRQ for page request queue\n",
		       iommu->name);
		dmar_free_hwirq(irq);
144
		iommu->pr_irq = 0;
145 146 147 148 149 150 151 152 153 154 155 156 157 158 159
		goto err;
	}
	dmar_writeq(iommu->reg + DMAR_PQH_REG, 0ULL);
	dmar_writeq(iommu->reg + DMAR_PQT_REG, 0ULL);
	dmar_writeq(iommu->reg + DMAR_PQA_REG, virt_to_phys(iommu->prq) | PRQ_ORDER);

	return 0;
}

int intel_svm_finish_prq(struct intel_iommu *iommu)
{
	dmar_writeq(iommu->reg + DMAR_PQH_REG, 0ULL);
	dmar_writeq(iommu->reg + DMAR_PQT_REG, 0ULL);
	dmar_writeq(iommu->reg + DMAR_PQA_REG, 0ULL);

160 161 162 163 164
	if (iommu->pr_irq) {
		free_irq(iommu->pr_irq, iommu);
		dmar_free_hwirq(iommu->pr_irq);
		iommu->pr_irq = 0;
	}
165 166 167 168 169 170 171

	free_pages((unsigned long)iommu->prq, PRQ_ORDER);
	iommu->prq = NULL;

	return 0;
}

172
static void intel_flush_svm_range_dev (struct intel_svm *svm, struct intel_svm_dev *sdev,
173
				       unsigned long address, unsigned long pages, int ih, int gl)
174 175 176
{
	struct qi_desc desc;

177
	if (pages == -1) {
178 179 180 181 182 183 184 185 186
		/* For global kernel pages we have to flush them in *all* PASIDs
		 * because that's the only option the hardware gives us. Despite
		 * the fact that they are actually only accessible through one. */
		if (gl)
			desc.low = QI_EIOTLB_PASID(svm->pasid) | QI_EIOTLB_DID(sdev->did) |
				QI_EIOTLB_GRAN(QI_GRAN_ALL_ALL) | QI_EIOTLB_TYPE;
		else
			desc.low = QI_EIOTLB_PASID(svm->pasid) | QI_EIOTLB_DID(sdev->did) |
				QI_EIOTLB_GRAN(QI_GRAN_NONG_PASID) | QI_EIOTLB_TYPE;
187 188
		desc.high = 0;
	} else {
189 190
		int mask = ilog2(__roundup_pow_of_two(pages));

191 192
		desc.low = QI_EIOTLB_PASID(svm->pasid) | QI_EIOTLB_DID(sdev->did) |
			QI_EIOTLB_GRAN(QI_GRAN_PSI_PASID) | QI_EIOTLB_TYPE;
193
		desc.high = QI_EIOTLB_ADDR(address) | QI_EIOTLB_GL(gl) |
194 195 196 197 198 199 200
			QI_EIOTLB_IH(ih) | QI_EIOTLB_AM(mask);
	}
	qi_submit_sync(&desc, svm->iommu);

	if (sdev->dev_iotlb) {
		desc.low = QI_DEV_EIOTLB_PASID(svm->pasid) | QI_DEV_EIOTLB_SID(sdev->sid) |
			QI_DEV_EIOTLB_QDEP(sdev->qdep) | QI_DEIOTLB_TYPE;
201 202 203 204 205 206 207
		if (pages == -1) {
			desc.high = QI_DEV_EIOTLB_ADDR(-1ULL >> 1) | QI_DEV_EIOTLB_SIZE;
		} else if (pages > 1) {
			/* The least significant zero bit indicates the size. So,
			 * for example, an "address" value of 0x12345f000 will
			 * flush from 0x123440000 to 0x12347ffff (256KiB). */
			unsigned long last = address + ((unsigned long)(pages - 1) << VTD_PAGE_SHIFT);
208
			unsigned long mask = __rounddown_pow_of_two(address ^ last);
209 210

			desc.high = QI_DEV_EIOTLB_ADDR((address & ~mask) | (mask - 1)) | QI_DEV_EIOTLB_SIZE;
211 212 213 214 215 216 217 218
		} else {
			desc.high = QI_DEV_EIOTLB_ADDR(address);
		}
		qi_submit_sync(&desc, svm->iommu);
	}
}

static void intel_flush_svm_range(struct intel_svm *svm, unsigned long address,
219
				  unsigned long pages, int ih, int gl)
220 221 222
{
	struct intel_svm_dev *sdev;

223 224 225 226 227
	/* Try deferred invalidate if available */
	if (svm->iommu->pasid_state_table &&
	    !cmpxchg64(&svm->iommu->pasid_state_table[svm->pasid].val, 0, 1ULL << 63))
		return;

228 229
	rcu_read_lock();
	list_for_each_entry_rcu(sdev, &svm->devs, list)
230
		intel_flush_svm_range_dev(svm, sdev, address, pages, ih, gl);
231 232 233 234 235 236 237 238
	rcu_read_unlock();
}

static void intel_change_pte(struct mmu_notifier *mn, struct mm_struct *mm,
			     unsigned long address, pte_t pte)
{
	struct intel_svm *svm = container_of(mn, struct intel_svm, notifier);

239
	intel_flush_svm_range(svm, address, 1, 1, 0);
240 241 242 243 244 245 246 247 248 249
}

/* Pages have been freed at this point */
static void intel_invalidate_range(struct mmu_notifier *mn,
				   struct mm_struct *mm,
				   unsigned long start, unsigned long end)
{
	struct intel_svm *svm = container_of(mn, struct intel_svm, notifier);

	intel_flush_svm_range(svm, start,
250
			      (end - start + PAGE_SIZE - 1) >> VTD_PAGE_SHIFT, 0, 0);
251 252 253
}


254
static void intel_flush_pasid_dev(struct intel_svm *svm, struct intel_svm_dev *sdev, int pasid)
255 256 257 258
{
	struct qi_desc desc;

	desc.high = 0;
259
	desc.low = QI_PC_TYPE | QI_PC_DID(sdev->did) | QI_PC_PASID_SEL | QI_PC_PASID(pasid);
260 261 262 263 264 265 266

	qi_submit_sync(&desc, svm->iommu);
}

static void intel_mm_release(struct mmu_notifier *mn, struct mm_struct *mm)
{
	struct intel_svm *svm = container_of(mn, struct intel_svm, notifier);
267
	struct intel_svm_dev *sdev;
268

269 270 271 272 273 274 275 276 277 278 279 280
	/* This might end up being called from exit_mmap(), *before* the page
	 * tables are cleared. And __mmu_notifier_release() will delete us from
	 * the list of notifiers so that our invalidate_range() callback doesn't
	 * get called when the page tables are cleared. So we need to protect
	 * against hardware accessing those page tables.
	 *
	 * We do it by clearing the entry in the PASID table and then flushing
	 * the IOTLB and the PASID table caches. This might upset hardware;
	 * perhaps we'll want to point the PASID to a dummy PGD (like the zero
	 * page) so that we end up taking a fault that the hardware really
	 * *has* to handle gracefully without affecting other processes.
	 */
281
	svm->iommu->pasid_table[svm->pasid].val = 0;
282 283 284 285 286 287 288 289
	wmb();

	rcu_read_lock();
	list_for_each_entry_rcu(sdev, &svm->devs, list) {
		intel_flush_pasid_dev(svm, sdev, svm->pasid);
		intel_flush_svm_range_dev(svm, sdev, 0, -1, 0, !svm->mm);
	}
	rcu_read_unlock();
290 291 292 293

}

static const struct mmu_notifier_ops intel_mmuops = {
294
	.flags = MMU_INVALIDATE_DOES_NOT_BLOCK,
295 296 297 298 299 300
	.release = intel_mm_release,
	.change_pte = intel_change_pte,
	.invalidate_range = intel_invalidate_range,
};

static DEFINE_MUTEX(pasid_mutex);
301
static LIST_HEAD(global_svm_list);
302

303
int intel_svm_bind_mm(struct device *dev, int *pasid, int flags, struct svm_dev_ops *ops)
304 305 306 307
{
	struct intel_iommu *iommu = intel_svm_device_to_iommu(dev);
	struct intel_svm_dev *sdev;
	struct intel_svm *svm = NULL;
308
	struct mm_struct *mm = NULL;
309
	u64 pasid_entry_val;
310 311 312
	int pasid_max;
	int ret;

313
	if (!iommu || !iommu->pasid_table)
314 315 316 317 318 319 320 321 322
		return -EINVAL;

	if (dev_is_pci(dev)) {
		pasid_max = pci_max_pasids(to_pci_dev(dev));
		if (pasid_max < 0)
			return -EINVAL;
	} else
		pasid_max = 1 << 20;

323
	if (flags & SVM_FLAG_SUPERVISOR_MODE) {
324 325 326 327 328 329 330
		if (!ecap_srs(iommu->ecap))
			return -EINVAL;
	} else if (pasid) {
		mm = get_task_mm(current);
		BUG_ON(!mm);
	}

331
	mutex_lock(&pasid_mutex);
332
	if (pasid && !(flags & SVM_FLAG_PRIVATE_PASID)) {
333
		struct intel_svm *t;
334

335 336
		list_for_each_entry(t, &global_svm_list, list) {
			if (t->mm != mm || (t->flags & SVM_FLAG_PRIVATE_PASID))
337 338
				continue;

339
			svm = t;
340 341 342 343 344 345 346 347 348 349
			if (svm->pasid >= pasid_max) {
				dev_warn(dev,
					 "Limited PASID width. Cannot use existing PASID %d\n",
					 svm->pasid);
				ret = -ENOSPC;
				goto out;
			}

			list_for_each_entry(sdev, &svm->devs, list) {
				if (dev == sdev->dev) {
350 351 352 353
					if (sdev->ops != ops) {
						ret = -EBUSY;
						goto out;
					}
354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378
					sdev->users++;
					goto success;
				}
			}

			break;
		}
	}

	sdev = kzalloc(sizeof(*sdev), GFP_KERNEL);
	if (!sdev) {
		ret = -ENOMEM;
		goto out;
	}
	sdev->dev = dev;

	ret = intel_iommu_enable_pasid(iommu, sdev);
	if (ret || !pasid) {
		/* If they don't actually want to assign a PASID, this is
		 * just an enabling check/preparation. */
		kfree(sdev);
		goto out;
	}
	/* Finish the setup now we know we're keeping it */
	sdev->users = 1;
379
	sdev->ops = ops;
380 381 382 383 384 385 386 387 388 389 390
	init_rcu_head(&sdev->rcu);

	if (!svm) {
		svm = kzalloc(sizeof(*svm), GFP_KERNEL);
		if (!svm) {
			ret = -ENOMEM;
			kfree(sdev);
			goto out;
		}
		svm->iommu = iommu;

391 392
		if (pasid_max > iommu->pasid_max)
			pasid_max = iommu->pasid_max;
393

394 395 396 397
		/* Do not use PASID 0 in caching mode (virtualised IOMMU) */
		ret = idr_alloc(&iommu->pasid_idr, svm,
				!!cap_caching_mode(iommu->cap),
				pasid_max - 1, GFP_KERNEL);
398 399
		if (ret < 0) {
			kfree(svm);
400
			kfree(sdev);
401 402 403 404
			goto out;
		}
		svm->pasid = ret;
		svm->notifier.ops = &intel_mmuops;
405
		svm->mm = mm;
406
		svm->flags = flags;
407
		INIT_LIST_HEAD_RCU(&svm->devs);
408
		INIT_LIST_HEAD(&svm->list);
409
		ret = -ENOMEM;
410 411 412 413 414 415 416 417
		if (mm) {
			ret = mmu_notifier_register(&svm->notifier, mm);
			if (ret) {
				idr_remove(&svm->iommu->pasid_idr, svm->pasid);
				kfree(svm);
				kfree(sdev);
				goto out;
			}
418
			pasid_entry_val = (u64)__pa(mm->pgd) | PASID_ENTRY_P;
419
		} else
420 421 422 423 424 425 426
			pasid_entry_val = (u64)__pa(init_mm.pgd) |
					  PASID_ENTRY_P | PASID_ENTRY_SRE;
		if (cpu_feature_enabled(X86_FEATURE_LA57))
			pasid_entry_val |= PASID_ENTRY_FLPM_5LP;

		iommu->pasid_table[svm->pasid].val = pasid_entry_val;

427
		wmb();
428 429 430 431 432

		/*
		 * Flush PASID cache when a PASID table entry becomes
		 * present.
		 */
433
		if (cap_caching_mode(iommu->cap))
434
			intel_flush_pasid_dev(svm, sdev, svm->pasid);
435 436

		list_add_tail(&svm->list, &global_svm_list);
437 438 439 440 441 442 443 444
	}
	list_add_rcu(&sdev->list, &svm->devs);

 success:
	*pasid = svm->pasid;
	ret = 0;
 out:
	mutex_unlock(&pasid_mutex);
445 446
	if (mm)
		mmput(mm);
447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479
	return ret;
}
EXPORT_SYMBOL_GPL(intel_svm_bind_mm);

int intel_svm_unbind_mm(struct device *dev, int pasid)
{
	struct intel_svm_dev *sdev;
	struct intel_iommu *iommu;
	struct intel_svm *svm;
	int ret = -EINVAL;

	mutex_lock(&pasid_mutex);
	iommu = intel_svm_device_to_iommu(dev);
	if (!iommu || !iommu->pasid_table)
		goto out;

	svm = idr_find(&iommu->pasid_idr, pasid);
	if (!svm)
		goto out;

	list_for_each_entry(sdev, &svm->devs, list) {
		if (dev == sdev->dev) {
			ret = 0;
			sdev->users--;
			if (!sdev->users) {
				list_del_rcu(&sdev->list);
				/* Flush the PASID cache and IOTLB for this device.
				 * Note that we do depend on the hardware *not* using
				 * the PASID any more. Just as we depend on other
				 * devices never using PASIDs that they have no right
				 * to use. We have a *shared* PASID table, because it's
				 * large and has to be physically contiguous. So it's
				 * hard to be as defensive as we might like. */
480
				intel_flush_pasid_dev(svm, sdev, svm->pasid);
481
				intel_flush_svm_range_dev(svm, sdev, 0, -1, 0, !svm->mm);
482 483 484
				kfree_rcu(sdev, rcu);

				if (list_empty(&svm->devs)) {
485 486
					svm->iommu->pasid_table[svm->pasid].val = 0;
					wmb();
487 488

					idr_remove(&svm->iommu->pasid_idr, svm->pasid);
489
					if (svm->mm)
490 491
						mmu_notifier_unregister(&svm->notifier, svm->mm);

492 493
					list_del(&svm->list);

494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510
					/* We mandate that no page faults may be outstanding
					 * for the PASID when intel_svm_unbind_mm() is called.
					 * If that is not obeyed, subtle errors will happen.
					 * Let's make them less subtle... */
					memset(svm, 0x6b, sizeof(*svm));
					kfree(svm);
				}
			}
			break;
		}
	}
 out:
	mutex_unlock(&pasid_mutex);

	return ret;
}
EXPORT_SYMBOL_GPL(intel_svm_unbind_mm);
511

512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541
int intel_svm_is_pasid_valid(struct device *dev, int pasid)
{
	struct intel_iommu *iommu;
	struct intel_svm *svm;
	int ret = -EINVAL;

	mutex_lock(&pasid_mutex);
	iommu = intel_svm_device_to_iommu(dev);
	if (!iommu || !iommu->pasid_table)
		goto out;

	svm = idr_find(&iommu->pasid_idr, pasid);
	if (!svm)
		goto out;

	/* init_mm is used in this case */
	if (!svm->mm)
		ret = 1;
	else if (atomic_read(&svm->mm->mm_users) > 0)
		ret = 1;
	else
		ret = 0;

 out:
	mutex_unlock(&pasid_mutex);

	return ret;
}
EXPORT_SYMBOL_GPL(intel_svm_is_pasid_valid);

542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560
/* Page request queue descriptor */
struct page_req_dsc {
	u64 srr:1;
	u64 bof:1;
	u64 pasid_present:1;
	u64 lpig:1;
	u64 pasid:20;
	u64 bus:8;
	u64 private:23;
	u64 prg_index:9;
	u64 rd_req:1;
	u64 wr_req:1;
	u64 exe_req:1;
	u64 priv_req:1;
	u64 devfn:8;
	u64 addr:52;
};

#define PRQ_RING_MASK ((0x1000 << PRQ_ORDER) - 0x10)
561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577

static bool access_error(struct vm_area_struct *vma, struct page_req_dsc *req)
{
	unsigned long requested = 0;

	if (req->exe_req)
		requested |= VM_EXEC;

	if (req->rd_req)
		requested |= VM_READ;

	if (req->wr_req)
		requested |= VM_WRITE;

	return (requested & ~vma->vm_flags) != 0;
}

578 579 580 581 582 583 584 585
static bool is_canonical_address(u64 addr)
{
	int shift = 64 - (__VIRTUAL_MASK_SHIFT + 1);
	long saddr = (long) addr;

	return (((saddr << shift) >> shift) == saddr);
}

586 587 588 589 590 591
static irqreturn_t prq_event_thread(int irq, void *d)
{
	struct intel_iommu *iommu = d;
	struct intel_svm *svm = NULL;
	int head, tail, handled = 0;

592 593 594 595
	/* Clear PPR bit before reading head/tail registers, to
	 * ensure that we get a new interrupt if needed. */
	writel(DMA_PRS_PPR, iommu->reg + DMAR_PRS_REG);

596 597 598
	tail = dmar_readq(iommu->reg + DMAR_PQT_REG) & PRQ_RING_MASK;
	head = dmar_readq(iommu->reg + DMAR_PQH_REG) & PRQ_RING_MASK;
	while (head != tail) {
599
		struct intel_svm_dev *sdev;
600 601 602 603 604 605 606 607 608 609 610
		struct vm_area_struct *vma;
		struct page_req_dsc *req;
		struct qi_desc resp;
		int ret, result;
		u64 address;

		handled = 1;

		req = &iommu->prq[head / sizeof(*req)];

		result = QI_RESP_FAILURE;
611
		address = (u64)req->addr << VTD_PAGE_SHIFT;
612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630
		if (!req->pasid_present) {
			pr_err("%s: Page request without PASID: %08llx %08llx\n",
			       iommu->name, ((unsigned long long *)req)[0],
			       ((unsigned long long *)req)[1]);
			goto bad_req;
		}

		if (!svm || svm->pasid != req->pasid) {
			rcu_read_lock();
			svm = idr_find(&iommu->pasid_idr, req->pasid);
			/* It *can't* go away, because the driver is not permitted
			 * to unbind the mm while any page faults are outstanding.
			 * So we only need RCU to protect the internal idr code. */
			rcu_read_unlock();

			if (!svm) {
				pr_err("%s: Page request for invalid PASID %d: %08llx %08llx\n",
				       iommu->name, req->pasid, ((unsigned long long *)req)[0],
				       ((unsigned long long *)req)[1]);
631
				goto no_pasid;
632 633 634 635
			}
		}

		result = QI_RESP_INVALID;
636 637 638 639
		/* Since we're using init_mm.pgd directly, we should never take
		 * any faults on kernel addresses. */
		if (!svm->mm)
			goto bad_req;
640
		/* If the mm is already defunct, don't handle faults. */
V
Vegard Nossum 已提交
641
		if (!mmget_not_zero(svm->mm))
642
			goto bad_req;
643 644 645 646 647

		/* If address is not canonical, return invalid response */
		if (!is_canonical_address(address))
			goto bad_req;

648 649 650 651 652
		down_read(&svm->mm->mmap_sem);
		vma = find_extend_vma(svm->mm, address);
		if (!vma || address < vma->vm_start)
			goto invalid;

653 654 655
		if (access_error(vma, req))
			goto invalid;

656
		ret = handle_mm_fault(vma, address,
657 658 659 660 661 662 663
				      req->wr_req ? FAULT_FLAG_WRITE : 0);
		if (ret & VM_FAULT_ERROR)
			goto invalid;

		result = QI_RESP_SUCCESS;
	invalid:
		up_read(&svm->mm->mmap_sem);
664
		mmput(svm->mm);
665 666
	bad_req:
		/* Accounting for major/minor faults? */
667 668
		rcu_read_lock();
		list_for_each_entry_rcu(sdev, &svm->devs, list) {
669
			if (sdev->sid == PCI_DEVID(req->bus, req->devfn))
670 671 672 673 674 675 676 677 678 679 680 681
				break;
		}
		/* Other devices can go away, but the drivers are not permitted
		 * to unbind while any page faults might be in flight. So it's
		 * OK to drop the 'lock' here now we have it. */
		rcu_read_unlock();

		if (WARN_ON(&sdev->list == &svm->devs))
			sdev = NULL;

		if (sdev && sdev->ops && sdev->ops->fault_cb) {
			int rwxp = (req->rd_req << 3) | (req->wr_req << 2) |
682
				(req->exe_req << 1) | (req->priv_req);
683 684
			sdev->ops->fault_cb(sdev->dev, req->pasid, req->addr, req->private, rwxp, result);
		}
685 686 687 688 689
		/* We get here in the error case where the PASID lookup failed,
		   and these can be NULL. Do not use them below this point! */
		sdev = NULL;
		svm = NULL;
	no_pasid:
690 691 692 693 694 695 696 697 698
		if (req->lpig) {
			/* Page Group Response */
			resp.low = QI_PGRP_PASID(req->pasid) |
				QI_PGRP_DID((req->bus << 8) | req->devfn) |
				QI_PGRP_PASID_P(req->pasid_present) |
				QI_PGRP_RESP_TYPE;
			resp.high = QI_PGRP_IDX(req->prg_index) |
				QI_PGRP_PRIV(req->private) | QI_PGRP_RESP_CODE(result);

699
			qi_submit_sync(&resp, iommu);
700 701 702 703 704 705 706 707
		} else if (req->srr) {
			/* Page Stream Response */
			resp.low = QI_PSTRM_IDX(req->prg_index) |
				QI_PSTRM_PRIV(req->private) | QI_PSTRM_BUS(req->bus) |
				QI_PSTRM_PASID(req->pasid) | QI_PSTRM_RESP_TYPE;
			resp.high = QI_PSTRM_ADDR(address) | QI_PSTRM_DEVFN(req->devfn) |
				QI_PSTRM_RESP_CODE(result);

708
			qi_submit_sync(&resp, iommu);
709 710 711 712 713 714 715 716 717
		}

		head = (head + sizeof(*req)) & PRQ_RING_MASK;
	}

	dmar_writeq(iommu->reg + DMAR_PQH_REG, tail);

	return IRQ_RETVAL(handled);
}