xref: /openbmc/linux/drivers/iommu/amd/iommu_v2.c (revision 92c005a1)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) 2010-2012 Advanced Micro Devices, Inc.
4  * Author: Joerg Roedel <jroedel@suse.de>
5  */
6 
7 #define pr_fmt(fmt)     "AMD-Vi: " fmt
8 
9 #include <linux/refcount.h>
10 #include <linux/mmu_notifier.h>
11 #include <linux/amd-iommu.h>
12 #include <linux/mm_types.h>
13 #include <linux/profile.h>
14 #include <linux/module.h>
15 #include <linux/sched.h>
16 #include <linux/sched/mm.h>
17 #include <linux/wait.h>
18 #include <linux/pci.h>
19 #include <linux/gfp.h>
20 #include <linux/cc_platform.h>
21 
22 #include "amd_iommu.h"
23 
24 MODULE_LICENSE("GPL v2");
25 MODULE_AUTHOR("Joerg Roedel <jroedel@suse.de>");
26 
27 #define PRI_QUEUE_SIZE		512
28 
29 struct pri_queue {
30 	atomic_t inflight;
31 	bool finish;
32 	int status;
33 };
34 
35 struct pasid_state {
36 	struct list_head list;			/* For global state-list */
37 	refcount_t count;				/* Reference count */
38 	unsigned mmu_notifier_count;		/* Counting nested mmu_notifier
39 						   calls */
40 	struct mm_struct *mm;			/* mm_struct for the faults */
41 	struct mmu_notifier mn;                 /* mmu_notifier handle */
42 	struct pri_queue pri[PRI_QUEUE_SIZE];	/* PRI tag states */
43 	struct device_state *device_state;	/* Link to our device_state */
44 	u32 pasid;				/* PASID index */
45 	bool invalid;				/* Used during setup and
46 						   teardown of the pasid */
47 	spinlock_t lock;			/* Protect pri_queues and
48 						   mmu_notifer_count */
49 	wait_queue_head_t wq;			/* To wait for count == 0 */
50 };
51 
52 struct device_state {
53 	struct list_head list;
54 	u16 devid;
55 	atomic_t count;
56 	struct pci_dev *pdev;
57 	struct pasid_state **states;
58 	struct iommu_domain *domain;
59 	int pasid_levels;
60 	int max_pasids;
61 	amd_iommu_invalid_ppr_cb inv_ppr_cb;
62 	amd_iommu_invalidate_ctx inv_ctx_cb;
63 	spinlock_t lock;
64 	wait_queue_head_t wq;
65 };
66 
67 struct fault {
68 	struct work_struct work;
69 	struct device_state *dev_state;
70 	struct pasid_state *state;
71 	struct mm_struct *mm;
72 	u64 address;
73 	u32 pasid;
74 	u16 tag;
75 	u16 finish;
76 	u16 flags;
77 };
78 
79 static LIST_HEAD(state_list);
80 static DEFINE_SPINLOCK(state_lock);
81 
82 static struct workqueue_struct *iommu_wq;
83 
84 static void free_pasid_states(struct device_state *dev_state);
85 
86 static u16 device_id(struct pci_dev *pdev)
87 {
88 	u16 devid;
89 
90 	devid = pdev->bus->number;
91 	devid = (devid << 8) | pdev->devfn;
92 
93 	return devid;
94 }
95 
96 static struct device_state *__get_device_state(u16 devid)
97 {
98 	struct device_state *dev_state;
99 
100 	list_for_each_entry(dev_state, &state_list, list) {
101 		if (dev_state->devid == devid)
102 			return dev_state;
103 	}
104 
105 	return NULL;
106 }
107 
108 static struct device_state *get_device_state(u16 devid)
109 {
110 	struct device_state *dev_state;
111 	unsigned long flags;
112 
113 	spin_lock_irqsave(&state_lock, flags);
114 	dev_state = __get_device_state(devid);
115 	if (dev_state != NULL)
116 		atomic_inc(&dev_state->count);
117 	spin_unlock_irqrestore(&state_lock, flags);
118 
119 	return dev_state;
120 }
121 
122 static void free_device_state(struct device_state *dev_state)
123 {
124 	struct iommu_group *group;
125 
126 	/* Get rid of any remaining pasid states */
127 	free_pasid_states(dev_state);
128 
129 	/*
130 	 * Wait until the last reference is dropped before freeing
131 	 * the device state.
132 	 */
133 	wait_event(dev_state->wq, !atomic_read(&dev_state->count));
134 
135 	/*
136 	 * First detach device from domain - No more PRI requests will arrive
137 	 * from that device after it is unbound from the IOMMUv2 domain.
138 	 */
139 	group = iommu_group_get(&dev_state->pdev->dev);
140 	if (WARN_ON(!group))
141 		return;
142 
143 	iommu_detach_group(dev_state->domain, group);
144 
145 	iommu_group_put(group);
146 
147 	/* Everything is down now, free the IOMMUv2 domain */
148 	iommu_domain_free(dev_state->domain);
149 
150 	/* Finally get rid of the device-state */
151 	kfree(dev_state);
152 }
153 
154 static void put_device_state(struct device_state *dev_state)
155 {
156 	if (atomic_dec_and_test(&dev_state->count))
157 		wake_up(&dev_state->wq);
158 }
159 
160 /* Must be called under dev_state->lock */
161 static struct pasid_state **__get_pasid_state_ptr(struct device_state *dev_state,
162 						  u32 pasid, bool alloc)
163 {
164 	struct pasid_state **root, **ptr;
165 	int level, index;
166 
167 	level = dev_state->pasid_levels;
168 	root  = dev_state->states;
169 
170 	while (true) {
171 
172 		index = (pasid >> (9 * level)) & 0x1ff;
173 		ptr   = &root[index];
174 
175 		if (level == 0)
176 			break;
177 
178 		if (*ptr == NULL) {
179 			if (!alloc)
180 				return NULL;
181 
182 			*ptr = (void *)get_zeroed_page(GFP_ATOMIC);
183 			if (*ptr == NULL)
184 				return NULL;
185 		}
186 
187 		root   = (struct pasid_state **)*ptr;
188 		level -= 1;
189 	}
190 
191 	return ptr;
192 }
193 
194 static int set_pasid_state(struct device_state *dev_state,
195 			   struct pasid_state *pasid_state,
196 			   u32 pasid)
197 {
198 	struct pasid_state **ptr;
199 	unsigned long flags;
200 	int ret;
201 
202 	spin_lock_irqsave(&dev_state->lock, flags);
203 	ptr = __get_pasid_state_ptr(dev_state, pasid, true);
204 
205 	ret = -ENOMEM;
206 	if (ptr == NULL)
207 		goto out_unlock;
208 
209 	ret = -ENOMEM;
210 	if (*ptr != NULL)
211 		goto out_unlock;
212 
213 	*ptr = pasid_state;
214 
215 	ret = 0;
216 
217 out_unlock:
218 	spin_unlock_irqrestore(&dev_state->lock, flags);
219 
220 	return ret;
221 }
222 
223 static void clear_pasid_state(struct device_state *dev_state, u32 pasid)
224 {
225 	struct pasid_state **ptr;
226 	unsigned long flags;
227 
228 	spin_lock_irqsave(&dev_state->lock, flags);
229 	ptr = __get_pasid_state_ptr(dev_state, pasid, true);
230 
231 	if (ptr == NULL)
232 		goto out_unlock;
233 
234 	*ptr = NULL;
235 
236 out_unlock:
237 	spin_unlock_irqrestore(&dev_state->lock, flags);
238 }
239 
240 static struct pasid_state *get_pasid_state(struct device_state *dev_state,
241 					   u32 pasid)
242 {
243 	struct pasid_state **ptr, *ret = NULL;
244 	unsigned long flags;
245 
246 	spin_lock_irqsave(&dev_state->lock, flags);
247 	ptr = __get_pasid_state_ptr(dev_state, pasid, false);
248 
249 	if (ptr == NULL)
250 		goto out_unlock;
251 
252 	ret = *ptr;
253 	if (ret)
254 		refcount_inc(&ret->count);
255 
256 out_unlock:
257 	spin_unlock_irqrestore(&dev_state->lock, flags);
258 
259 	return ret;
260 }
261 
262 static void free_pasid_state(struct pasid_state *pasid_state)
263 {
264 	kfree(pasid_state);
265 }
266 
267 static void put_pasid_state(struct pasid_state *pasid_state)
268 {
269 	if (refcount_dec_and_test(&pasid_state->count))
270 		wake_up(&pasid_state->wq);
271 }
272 
273 static void put_pasid_state_wait(struct pasid_state *pasid_state)
274 {
275 	refcount_dec(&pasid_state->count);
276 	wait_event(pasid_state->wq, !refcount_read(&pasid_state->count));
277 	free_pasid_state(pasid_state);
278 }
279 
280 static void unbind_pasid(struct pasid_state *pasid_state)
281 {
282 	struct iommu_domain *domain;
283 
284 	domain = pasid_state->device_state->domain;
285 
286 	/*
287 	 * Mark pasid_state as invalid, no more faults will we added to the
288 	 * work queue after this is visible everywhere.
289 	 */
290 	pasid_state->invalid = true;
291 
292 	/* Make sure this is visible */
293 	smp_wmb();
294 
295 	/* After this the device/pasid can't access the mm anymore */
296 	amd_iommu_domain_clear_gcr3(domain, pasid_state->pasid);
297 
298 	/* Make sure no more pending faults are in the queue */
299 	flush_workqueue(iommu_wq);
300 }
301 
302 static void free_pasid_states_level1(struct pasid_state **tbl)
303 {
304 	int i;
305 
306 	for (i = 0; i < 512; ++i) {
307 		if (tbl[i] == NULL)
308 			continue;
309 
310 		free_page((unsigned long)tbl[i]);
311 	}
312 }
313 
314 static void free_pasid_states_level2(struct pasid_state **tbl)
315 {
316 	struct pasid_state **ptr;
317 	int i;
318 
319 	for (i = 0; i < 512; ++i) {
320 		if (tbl[i] == NULL)
321 			continue;
322 
323 		ptr = (struct pasid_state **)tbl[i];
324 		free_pasid_states_level1(ptr);
325 	}
326 }
327 
328 static void free_pasid_states(struct device_state *dev_state)
329 {
330 	struct pasid_state *pasid_state;
331 	int i;
332 
333 	for (i = 0; i < dev_state->max_pasids; ++i) {
334 		pasid_state = get_pasid_state(dev_state, i);
335 		if (pasid_state == NULL)
336 			continue;
337 
338 		put_pasid_state(pasid_state);
339 
340 		/*
341 		 * This will call the mn_release function and
342 		 * unbind the PASID
343 		 */
344 		mmu_notifier_unregister(&pasid_state->mn, pasid_state->mm);
345 
346 		put_pasid_state_wait(pasid_state); /* Reference taken in
347 						      amd_iommu_bind_pasid */
348 
349 		/* Drop reference taken in amd_iommu_bind_pasid */
350 		put_device_state(dev_state);
351 	}
352 
353 	if (dev_state->pasid_levels == 2)
354 		free_pasid_states_level2(dev_state->states);
355 	else if (dev_state->pasid_levels == 1)
356 		free_pasid_states_level1(dev_state->states);
357 	else
358 		BUG_ON(dev_state->pasid_levels != 0);
359 
360 	free_page((unsigned long)dev_state->states);
361 }
362 
363 static struct pasid_state *mn_to_state(struct mmu_notifier *mn)
364 {
365 	return container_of(mn, struct pasid_state, mn);
366 }
367 
368 static void mn_invalidate_range(struct mmu_notifier *mn,
369 				struct mm_struct *mm,
370 				unsigned long start, unsigned long end)
371 {
372 	struct pasid_state *pasid_state;
373 	struct device_state *dev_state;
374 
375 	pasid_state = mn_to_state(mn);
376 	dev_state   = pasid_state->device_state;
377 
378 	if ((start ^ (end - 1)) < PAGE_SIZE)
379 		amd_iommu_flush_page(dev_state->domain, pasid_state->pasid,
380 				     start);
381 	else
382 		amd_iommu_flush_tlb(dev_state->domain, pasid_state->pasid);
383 }
384 
385 static void mn_release(struct mmu_notifier *mn, struct mm_struct *mm)
386 {
387 	struct pasid_state *pasid_state;
388 	struct device_state *dev_state;
389 	bool run_inv_ctx_cb;
390 
391 	might_sleep();
392 
393 	pasid_state    = mn_to_state(mn);
394 	dev_state      = pasid_state->device_state;
395 	run_inv_ctx_cb = !pasid_state->invalid;
396 
397 	if (run_inv_ctx_cb && dev_state->inv_ctx_cb)
398 		dev_state->inv_ctx_cb(dev_state->pdev, pasid_state->pasid);
399 
400 	unbind_pasid(pasid_state);
401 }
402 
403 static const struct mmu_notifier_ops iommu_mn = {
404 	.release		= mn_release,
405 	.invalidate_range       = mn_invalidate_range,
406 };
407 
408 static void set_pri_tag_status(struct pasid_state *pasid_state,
409 			       u16 tag, int status)
410 {
411 	unsigned long flags;
412 
413 	spin_lock_irqsave(&pasid_state->lock, flags);
414 	pasid_state->pri[tag].status = status;
415 	spin_unlock_irqrestore(&pasid_state->lock, flags);
416 }
417 
418 static void finish_pri_tag(struct device_state *dev_state,
419 			   struct pasid_state *pasid_state,
420 			   u16 tag)
421 {
422 	unsigned long flags;
423 
424 	spin_lock_irqsave(&pasid_state->lock, flags);
425 	if (atomic_dec_and_test(&pasid_state->pri[tag].inflight) &&
426 	    pasid_state->pri[tag].finish) {
427 		amd_iommu_complete_ppr(dev_state->pdev, pasid_state->pasid,
428 				       pasid_state->pri[tag].status, tag);
429 		pasid_state->pri[tag].finish = false;
430 		pasid_state->pri[tag].status = PPR_SUCCESS;
431 	}
432 	spin_unlock_irqrestore(&pasid_state->lock, flags);
433 }
434 
435 static void handle_fault_error(struct fault *fault)
436 {
437 	int status;
438 
439 	if (!fault->dev_state->inv_ppr_cb) {
440 		set_pri_tag_status(fault->state, fault->tag, PPR_INVALID);
441 		return;
442 	}
443 
444 	status = fault->dev_state->inv_ppr_cb(fault->dev_state->pdev,
445 					      fault->pasid,
446 					      fault->address,
447 					      fault->flags);
448 	switch (status) {
449 	case AMD_IOMMU_INV_PRI_RSP_SUCCESS:
450 		set_pri_tag_status(fault->state, fault->tag, PPR_SUCCESS);
451 		break;
452 	case AMD_IOMMU_INV_PRI_RSP_INVALID:
453 		set_pri_tag_status(fault->state, fault->tag, PPR_INVALID);
454 		break;
455 	case AMD_IOMMU_INV_PRI_RSP_FAIL:
456 		set_pri_tag_status(fault->state, fault->tag, PPR_FAILURE);
457 		break;
458 	default:
459 		BUG();
460 	}
461 }
462 
463 static bool access_error(struct vm_area_struct *vma, struct fault *fault)
464 {
465 	unsigned long requested = 0;
466 
467 	if (fault->flags & PPR_FAULT_EXEC)
468 		requested |= VM_EXEC;
469 
470 	if (fault->flags & PPR_FAULT_READ)
471 		requested |= VM_READ;
472 
473 	if (fault->flags & PPR_FAULT_WRITE)
474 		requested |= VM_WRITE;
475 
476 	return (requested & ~vma->vm_flags) != 0;
477 }
478 
479 static void do_fault(struct work_struct *work)
480 {
481 	struct fault *fault = container_of(work, struct fault, work);
482 	struct vm_area_struct *vma;
483 	vm_fault_t ret = VM_FAULT_ERROR;
484 	unsigned int flags = 0;
485 	struct mm_struct *mm;
486 	u64 address;
487 
488 	mm = fault->state->mm;
489 	address = fault->address;
490 
491 	if (fault->flags & PPR_FAULT_USER)
492 		flags |= FAULT_FLAG_USER;
493 	if (fault->flags & PPR_FAULT_WRITE)
494 		flags |= FAULT_FLAG_WRITE;
495 	flags |= FAULT_FLAG_REMOTE;
496 
497 	mmap_read_lock(mm);
498 	vma = find_extend_vma(mm, address);
499 	if (!vma || address < vma->vm_start)
500 		/* failed to get a vma in the right range */
501 		goto out;
502 
503 	/* Check if we have the right permissions on the vma */
504 	if (access_error(vma, fault))
505 		goto out;
506 
507 	ret = handle_mm_fault(vma, address, flags, NULL);
508 out:
509 	mmap_read_unlock(mm);
510 
511 	if (ret & VM_FAULT_ERROR)
512 		/* failed to service fault */
513 		handle_fault_error(fault);
514 
515 	finish_pri_tag(fault->dev_state, fault->state, fault->tag);
516 
517 	put_pasid_state(fault->state);
518 
519 	kfree(fault);
520 }
521 
522 static int ppr_notifier(struct notifier_block *nb, unsigned long e, void *data)
523 {
524 	struct amd_iommu_fault *iommu_fault;
525 	struct pasid_state *pasid_state;
526 	struct device_state *dev_state;
527 	struct pci_dev *pdev = NULL;
528 	unsigned long flags;
529 	struct fault *fault;
530 	bool finish;
531 	u16 tag, devid;
532 	int ret;
533 
534 	iommu_fault = data;
535 	tag         = iommu_fault->tag & 0x1ff;
536 	finish      = (iommu_fault->tag >> 9) & 1;
537 
538 	devid = iommu_fault->device_id;
539 	pdev = pci_get_domain_bus_and_slot(0, PCI_BUS_NUM(devid),
540 					   devid & 0xff);
541 	if (!pdev)
542 		return -ENODEV;
543 
544 	ret = NOTIFY_DONE;
545 
546 	/* In kdump kernel pci dev is not initialized yet -> send INVALID */
547 	if (amd_iommu_is_attach_deferred(&pdev->dev)) {
548 		amd_iommu_complete_ppr(pdev, iommu_fault->pasid,
549 				       PPR_INVALID, tag);
550 		goto out;
551 	}
552 
553 	dev_state = get_device_state(iommu_fault->device_id);
554 	if (dev_state == NULL)
555 		goto out;
556 
557 	pasid_state = get_pasid_state(dev_state, iommu_fault->pasid);
558 	if (pasid_state == NULL || pasid_state->invalid) {
559 		/* We know the device but not the PASID -> send INVALID */
560 		amd_iommu_complete_ppr(dev_state->pdev, iommu_fault->pasid,
561 				       PPR_INVALID, tag);
562 		goto out_drop_state;
563 	}
564 
565 	spin_lock_irqsave(&pasid_state->lock, flags);
566 	atomic_inc(&pasid_state->pri[tag].inflight);
567 	if (finish)
568 		pasid_state->pri[tag].finish = true;
569 	spin_unlock_irqrestore(&pasid_state->lock, flags);
570 
571 	fault = kzalloc(sizeof(*fault), GFP_ATOMIC);
572 	if (fault == NULL) {
573 		/* We are OOM - send success and let the device re-fault */
574 		finish_pri_tag(dev_state, pasid_state, tag);
575 		goto out_drop_state;
576 	}
577 
578 	fault->dev_state = dev_state;
579 	fault->address   = iommu_fault->address;
580 	fault->state     = pasid_state;
581 	fault->tag       = tag;
582 	fault->finish    = finish;
583 	fault->pasid     = iommu_fault->pasid;
584 	fault->flags     = iommu_fault->flags;
585 	INIT_WORK(&fault->work, do_fault);
586 
587 	queue_work(iommu_wq, &fault->work);
588 
589 	ret = NOTIFY_OK;
590 
591 out_drop_state:
592 
593 	if (ret != NOTIFY_OK && pasid_state)
594 		put_pasid_state(pasid_state);
595 
596 	put_device_state(dev_state);
597 
598 out:
599 	return ret;
600 }
601 
602 static struct notifier_block ppr_nb = {
603 	.notifier_call = ppr_notifier,
604 };
605 
606 int amd_iommu_bind_pasid(struct pci_dev *pdev, u32 pasid,
607 			 struct task_struct *task)
608 {
609 	struct pasid_state *pasid_state;
610 	struct device_state *dev_state;
611 	struct mm_struct *mm;
612 	u16 devid;
613 	int ret;
614 
615 	might_sleep();
616 
617 	if (!amd_iommu_v2_supported())
618 		return -ENODEV;
619 
620 	devid     = device_id(pdev);
621 	dev_state = get_device_state(devid);
622 
623 	if (dev_state == NULL)
624 		return -EINVAL;
625 
626 	ret = -EINVAL;
627 	if (pasid >= dev_state->max_pasids)
628 		goto out;
629 
630 	ret = -ENOMEM;
631 	pasid_state = kzalloc(sizeof(*pasid_state), GFP_KERNEL);
632 	if (pasid_state == NULL)
633 		goto out;
634 
635 
636 	refcount_set(&pasid_state->count, 1);
637 	init_waitqueue_head(&pasid_state->wq);
638 	spin_lock_init(&pasid_state->lock);
639 
640 	mm                        = get_task_mm(task);
641 	pasid_state->mm           = mm;
642 	pasid_state->device_state = dev_state;
643 	pasid_state->pasid        = pasid;
644 	pasid_state->invalid      = true; /* Mark as valid only if we are
645 					     done with setting up the pasid */
646 	pasid_state->mn.ops       = &iommu_mn;
647 
648 	if (pasid_state->mm == NULL)
649 		goto out_free;
650 
651 	mmu_notifier_register(&pasid_state->mn, mm);
652 
653 	ret = set_pasid_state(dev_state, pasid_state, pasid);
654 	if (ret)
655 		goto out_unregister;
656 
657 	ret = amd_iommu_domain_set_gcr3(dev_state->domain, pasid,
658 					__pa(pasid_state->mm->pgd));
659 	if (ret)
660 		goto out_clear_state;
661 
662 	/* Now we are ready to handle faults */
663 	pasid_state->invalid = false;
664 
665 	/*
666 	 * Drop the reference to the mm_struct here. We rely on the
667 	 * mmu_notifier release call-back to inform us when the mm
668 	 * is going away.
669 	 */
670 	mmput(mm);
671 
672 	return 0;
673 
674 out_clear_state:
675 	clear_pasid_state(dev_state, pasid);
676 
677 out_unregister:
678 	mmu_notifier_unregister(&pasid_state->mn, mm);
679 	mmput(mm);
680 
681 out_free:
682 	free_pasid_state(pasid_state);
683 
684 out:
685 	put_device_state(dev_state);
686 
687 	return ret;
688 }
689 EXPORT_SYMBOL(amd_iommu_bind_pasid);
690 
691 void amd_iommu_unbind_pasid(struct pci_dev *pdev, u32 pasid)
692 {
693 	struct pasid_state *pasid_state;
694 	struct device_state *dev_state;
695 	u16 devid;
696 
697 	might_sleep();
698 
699 	if (!amd_iommu_v2_supported())
700 		return;
701 
702 	devid = device_id(pdev);
703 	dev_state = get_device_state(devid);
704 	if (dev_state == NULL)
705 		return;
706 
707 	if (pasid >= dev_state->max_pasids)
708 		goto out;
709 
710 	pasid_state = get_pasid_state(dev_state, pasid);
711 	if (pasid_state == NULL)
712 		goto out;
713 	/*
714 	 * Drop reference taken here. We are safe because we still hold
715 	 * the reference taken in the amd_iommu_bind_pasid function.
716 	 */
717 	put_pasid_state(pasid_state);
718 
719 	/* Clear the pasid state so that the pasid can be re-used */
720 	clear_pasid_state(dev_state, pasid_state->pasid);
721 
722 	/*
723 	 * Call mmu_notifier_unregister to drop our reference
724 	 * to pasid_state->mm
725 	 */
726 	mmu_notifier_unregister(&pasid_state->mn, pasid_state->mm);
727 
728 	put_pasid_state_wait(pasid_state); /* Reference taken in
729 					      amd_iommu_bind_pasid */
730 out:
731 	/* Drop reference taken in this function */
732 	put_device_state(dev_state);
733 
734 	/* Drop reference taken in amd_iommu_bind_pasid */
735 	put_device_state(dev_state);
736 }
737 EXPORT_SYMBOL(amd_iommu_unbind_pasid);
738 
739 int amd_iommu_init_device(struct pci_dev *pdev, int pasids)
740 {
741 	struct device_state *dev_state;
742 	struct iommu_group *group;
743 	unsigned long flags;
744 	int ret, tmp;
745 	u16 devid;
746 
747 	might_sleep();
748 
749 	/*
750 	 * When memory encryption is active the device is likely not in a
751 	 * direct-mapped domain. Forbid using IOMMUv2 functionality for now.
752 	 */
753 	if (cc_platform_has(CC_ATTR_MEM_ENCRYPT))
754 		return -ENODEV;
755 
756 	if (!amd_iommu_v2_supported())
757 		return -ENODEV;
758 
759 	if (pasids <= 0 || pasids > (PASID_MASK + 1))
760 		return -EINVAL;
761 
762 	devid = device_id(pdev);
763 
764 	dev_state = kzalloc(sizeof(*dev_state), GFP_KERNEL);
765 	if (dev_state == NULL)
766 		return -ENOMEM;
767 
768 	spin_lock_init(&dev_state->lock);
769 	init_waitqueue_head(&dev_state->wq);
770 	dev_state->pdev  = pdev;
771 	dev_state->devid = devid;
772 
773 	tmp = pasids;
774 	for (dev_state->pasid_levels = 0; (tmp - 1) & ~0x1ff; tmp >>= 9)
775 		dev_state->pasid_levels += 1;
776 
777 	atomic_set(&dev_state->count, 1);
778 	dev_state->max_pasids = pasids;
779 
780 	ret = -ENOMEM;
781 	dev_state->states = (void *)get_zeroed_page(GFP_KERNEL);
782 	if (dev_state->states == NULL)
783 		goto out_free_dev_state;
784 
785 	dev_state->domain = iommu_domain_alloc(&pci_bus_type);
786 	if (dev_state->domain == NULL)
787 		goto out_free_states;
788 
789 	amd_iommu_domain_direct_map(dev_state->domain);
790 
791 	ret = amd_iommu_domain_enable_v2(dev_state->domain, pasids);
792 	if (ret)
793 		goto out_free_domain;
794 
795 	group = iommu_group_get(&pdev->dev);
796 	if (!group) {
797 		ret = -EINVAL;
798 		goto out_free_domain;
799 	}
800 
801 	ret = iommu_attach_group(dev_state->domain, group);
802 	if (ret != 0)
803 		goto out_drop_group;
804 
805 	iommu_group_put(group);
806 
807 	spin_lock_irqsave(&state_lock, flags);
808 
809 	if (__get_device_state(devid) != NULL) {
810 		spin_unlock_irqrestore(&state_lock, flags);
811 		ret = -EBUSY;
812 		goto out_free_domain;
813 	}
814 
815 	list_add_tail(&dev_state->list, &state_list);
816 
817 	spin_unlock_irqrestore(&state_lock, flags);
818 
819 	return 0;
820 
821 out_drop_group:
822 	iommu_group_put(group);
823 
824 out_free_domain:
825 	iommu_domain_free(dev_state->domain);
826 
827 out_free_states:
828 	free_page((unsigned long)dev_state->states);
829 
830 out_free_dev_state:
831 	kfree(dev_state);
832 
833 	return ret;
834 }
835 EXPORT_SYMBOL(amd_iommu_init_device);
836 
837 void amd_iommu_free_device(struct pci_dev *pdev)
838 {
839 	struct device_state *dev_state;
840 	unsigned long flags;
841 	u16 devid;
842 
843 	if (!amd_iommu_v2_supported())
844 		return;
845 
846 	devid = device_id(pdev);
847 
848 	spin_lock_irqsave(&state_lock, flags);
849 
850 	dev_state = __get_device_state(devid);
851 	if (dev_state == NULL) {
852 		spin_unlock_irqrestore(&state_lock, flags);
853 		return;
854 	}
855 
856 	list_del(&dev_state->list);
857 
858 	spin_unlock_irqrestore(&state_lock, flags);
859 
860 	put_device_state(dev_state);
861 	free_device_state(dev_state);
862 }
863 EXPORT_SYMBOL(amd_iommu_free_device);
864 
865 int amd_iommu_set_invalid_ppr_cb(struct pci_dev *pdev,
866 				 amd_iommu_invalid_ppr_cb cb)
867 {
868 	struct device_state *dev_state;
869 	unsigned long flags;
870 	u16 devid;
871 	int ret;
872 
873 	if (!amd_iommu_v2_supported())
874 		return -ENODEV;
875 
876 	devid = device_id(pdev);
877 
878 	spin_lock_irqsave(&state_lock, flags);
879 
880 	ret = -EINVAL;
881 	dev_state = __get_device_state(devid);
882 	if (dev_state == NULL)
883 		goto out_unlock;
884 
885 	dev_state->inv_ppr_cb = cb;
886 
887 	ret = 0;
888 
889 out_unlock:
890 	spin_unlock_irqrestore(&state_lock, flags);
891 
892 	return ret;
893 }
894 EXPORT_SYMBOL(amd_iommu_set_invalid_ppr_cb);
895 
896 int amd_iommu_set_invalidate_ctx_cb(struct pci_dev *pdev,
897 				    amd_iommu_invalidate_ctx cb)
898 {
899 	struct device_state *dev_state;
900 	unsigned long flags;
901 	u16 devid;
902 	int ret;
903 
904 	if (!amd_iommu_v2_supported())
905 		return -ENODEV;
906 
907 	devid = device_id(pdev);
908 
909 	spin_lock_irqsave(&state_lock, flags);
910 
911 	ret = -EINVAL;
912 	dev_state = __get_device_state(devid);
913 	if (dev_state == NULL)
914 		goto out_unlock;
915 
916 	dev_state->inv_ctx_cb = cb;
917 
918 	ret = 0;
919 
920 out_unlock:
921 	spin_unlock_irqrestore(&state_lock, flags);
922 
923 	return ret;
924 }
925 EXPORT_SYMBOL(amd_iommu_set_invalidate_ctx_cb);
926 
927 static int __init amd_iommu_v2_init(void)
928 {
929 	int ret;
930 
931 	if (!amd_iommu_v2_supported()) {
932 		pr_info("AMD IOMMUv2 functionality not available on this system - This is not a bug.\n");
933 		/*
934 		 * Load anyway to provide the symbols to other modules
935 		 * which may use AMD IOMMUv2 optionally.
936 		 */
937 		return 0;
938 	}
939 
940 	ret = -ENOMEM;
941 	iommu_wq = alloc_workqueue("amd_iommu_v2", WQ_MEM_RECLAIM, 0);
942 	if (iommu_wq == NULL)
943 		goto out;
944 
945 	amd_iommu_register_ppr_notifier(&ppr_nb);
946 
947 	pr_info("AMD IOMMUv2 loaded and initialized\n");
948 
949 	return 0;
950 
951 out:
952 	return ret;
953 }
954 
955 static void __exit amd_iommu_v2_exit(void)
956 {
957 	struct device_state *dev_state, *next;
958 	unsigned long flags;
959 	LIST_HEAD(freelist);
960 
961 	if (!amd_iommu_v2_supported())
962 		return;
963 
964 	amd_iommu_unregister_ppr_notifier(&ppr_nb);
965 
966 	flush_workqueue(iommu_wq);
967 
968 	/*
969 	 * The loop below might call flush_workqueue(), so call
970 	 * destroy_workqueue() after it
971 	 */
972 	spin_lock_irqsave(&state_lock, flags);
973 
974 	list_for_each_entry_safe(dev_state, next, &state_list, list) {
975 		WARN_ON_ONCE(1);
976 
977 		put_device_state(dev_state);
978 		list_del(&dev_state->list);
979 		list_add_tail(&dev_state->list, &freelist);
980 	}
981 
982 	spin_unlock_irqrestore(&state_lock, flags);
983 
984 	/*
985 	 * Since free_device_state waits on the count to be zero,
986 	 * we need to free dev_state outside the spinlock.
987 	 */
988 	list_for_each_entry_safe(dev_state, next, &freelist, list) {
989 		list_del(&dev_state->list);
990 		free_device_state(dev_state);
991 	}
992 
993 	destroy_workqueue(iommu_wq);
994 }
995 
996 module_init(amd_iommu_v2_init);
997 module_exit(amd_iommu_v2_exit);
998