xref: /linux/drivers/iommu/s390-iommu.c (revision 7f71507851fc7764b36a3221839607d3a45c2025)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * IOMMU API for s390 PCI devices
4  *
5  * Copyright IBM Corp. 2015
6  * Author(s): Gerald Schaefer <gerald.schaefer@de.ibm.com>
7  */
8 
9 #include <linux/pci.h>
10 #include <linux/iommu.h>
11 #include <linux/iommu-helper.h>
12 #include <linux/sizes.h>
13 #include <linux/rculist.h>
14 #include <linux/rcupdate.h>
15 #include <asm/pci_dma.h>
16 
17 #include "dma-iommu.h"
18 
19 static const struct iommu_ops s390_iommu_ops;
20 
21 static struct kmem_cache *dma_region_table_cache;
22 static struct kmem_cache *dma_page_table_cache;
23 
24 static u64 s390_iommu_aperture;
25 static u32 s390_iommu_aperture_factor = 1;
26 
27 struct s390_domain {
28 	struct iommu_domain	domain;
29 	struct list_head	devices;
30 	struct zpci_iommu_ctrs	ctrs;
31 	unsigned long		*dma_table;
32 	spinlock_t		list_lock;
33 	struct rcu_head		rcu;
34 };
35 
36 static struct iommu_domain blocking_domain;
37 
38 static inline unsigned int calc_rtx(dma_addr_t ptr)
39 {
40 	return ((unsigned long)ptr >> ZPCI_RT_SHIFT) & ZPCI_INDEX_MASK;
41 }
42 
43 static inline unsigned int calc_sx(dma_addr_t ptr)
44 {
45 	return ((unsigned long)ptr >> ZPCI_ST_SHIFT) & ZPCI_INDEX_MASK;
46 }
47 
48 static inline unsigned int calc_px(dma_addr_t ptr)
49 {
50 	return ((unsigned long)ptr >> PAGE_SHIFT) & ZPCI_PT_MASK;
51 }
52 
53 static inline void set_pt_pfaa(unsigned long *entry, phys_addr_t pfaa)
54 {
55 	*entry &= ZPCI_PTE_FLAG_MASK;
56 	*entry |= (pfaa & ZPCI_PTE_ADDR_MASK);
57 }
58 
59 static inline void set_rt_sto(unsigned long *entry, phys_addr_t sto)
60 {
61 	*entry &= ZPCI_RTE_FLAG_MASK;
62 	*entry |= (sto & ZPCI_RTE_ADDR_MASK);
63 	*entry |= ZPCI_TABLE_TYPE_RTX;
64 }
65 
66 static inline void set_st_pto(unsigned long *entry, phys_addr_t pto)
67 {
68 	*entry &= ZPCI_STE_FLAG_MASK;
69 	*entry |= (pto & ZPCI_STE_ADDR_MASK);
70 	*entry |= ZPCI_TABLE_TYPE_SX;
71 }
72 
73 static inline void validate_rt_entry(unsigned long *entry)
74 {
75 	*entry &= ~ZPCI_TABLE_VALID_MASK;
76 	*entry &= ~ZPCI_TABLE_OFFSET_MASK;
77 	*entry |= ZPCI_TABLE_VALID;
78 	*entry |= ZPCI_TABLE_LEN_RTX;
79 }
80 
81 static inline void validate_st_entry(unsigned long *entry)
82 {
83 	*entry &= ~ZPCI_TABLE_VALID_MASK;
84 	*entry |= ZPCI_TABLE_VALID;
85 }
86 
87 static inline void invalidate_pt_entry(unsigned long *entry)
88 {
89 	WARN_ON_ONCE((*entry & ZPCI_PTE_VALID_MASK) == ZPCI_PTE_INVALID);
90 	*entry &= ~ZPCI_PTE_VALID_MASK;
91 	*entry |= ZPCI_PTE_INVALID;
92 }
93 
94 static inline void validate_pt_entry(unsigned long *entry)
95 {
96 	WARN_ON_ONCE((*entry & ZPCI_PTE_VALID_MASK) == ZPCI_PTE_VALID);
97 	*entry &= ~ZPCI_PTE_VALID_MASK;
98 	*entry |= ZPCI_PTE_VALID;
99 }
100 
101 static inline void entry_set_protected(unsigned long *entry)
102 {
103 	*entry &= ~ZPCI_TABLE_PROT_MASK;
104 	*entry |= ZPCI_TABLE_PROTECTED;
105 }
106 
107 static inline void entry_clr_protected(unsigned long *entry)
108 {
109 	*entry &= ~ZPCI_TABLE_PROT_MASK;
110 	*entry |= ZPCI_TABLE_UNPROTECTED;
111 }
112 
113 static inline int reg_entry_isvalid(unsigned long entry)
114 {
115 	return (entry & ZPCI_TABLE_VALID_MASK) == ZPCI_TABLE_VALID;
116 }
117 
118 static inline int pt_entry_isvalid(unsigned long entry)
119 {
120 	return (entry & ZPCI_PTE_VALID_MASK) == ZPCI_PTE_VALID;
121 }
122 
123 static inline unsigned long *get_rt_sto(unsigned long entry)
124 {
125 	if ((entry & ZPCI_TABLE_TYPE_MASK) == ZPCI_TABLE_TYPE_RTX)
126 		return phys_to_virt(entry & ZPCI_RTE_ADDR_MASK);
127 	else
128 		return NULL;
129 }
130 
131 static inline unsigned long *get_st_pto(unsigned long entry)
132 {
133 	if ((entry & ZPCI_TABLE_TYPE_MASK) == ZPCI_TABLE_TYPE_SX)
134 		return phys_to_virt(entry & ZPCI_STE_ADDR_MASK);
135 	else
136 		return NULL;
137 }
138 
139 static int __init dma_alloc_cpu_table_caches(void)
140 {
141 	dma_region_table_cache = kmem_cache_create("PCI_DMA_region_tables",
142 						   ZPCI_TABLE_SIZE,
143 						   ZPCI_TABLE_ALIGN,
144 						   0, NULL);
145 	if (!dma_region_table_cache)
146 		return -ENOMEM;
147 
148 	dma_page_table_cache = kmem_cache_create("PCI_DMA_page_tables",
149 						 ZPCI_PT_SIZE,
150 						 ZPCI_PT_ALIGN,
151 						 0, NULL);
152 	if (!dma_page_table_cache) {
153 		kmem_cache_destroy(dma_region_table_cache);
154 		return -ENOMEM;
155 	}
156 	return 0;
157 }
158 
159 static unsigned long *dma_alloc_cpu_table(gfp_t gfp)
160 {
161 	unsigned long *table, *entry;
162 
163 	table = kmem_cache_alloc(dma_region_table_cache, gfp);
164 	if (!table)
165 		return NULL;
166 
167 	for (entry = table; entry < table + ZPCI_TABLE_ENTRIES; entry++)
168 		*entry = ZPCI_TABLE_INVALID;
169 	return table;
170 }
171 
172 static void dma_free_cpu_table(void *table)
173 {
174 	kmem_cache_free(dma_region_table_cache, table);
175 }
176 
177 static void dma_free_page_table(void *table)
178 {
179 	kmem_cache_free(dma_page_table_cache, table);
180 }
181 
182 static void dma_free_seg_table(unsigned long entry)
183 {
184 	unsigned long *sto = get_rt_sto(entry);
185 	int sx;
186 
187 	for (sx = 0; sx < ZPCI_TABLE_ENTRIES; sx++)
188 		if (reg_entry_isvalid(sto[sx]))
189 			dma_free_page_table(get_st_pto(sto[sx]));
190 
191 	dma_free_cpu_table(sto);
192 }
193 
194 static void dma_cleanup_tables(unsigned long *table)
195 {
196 	int rtx;
197 
198 	if (!table)
199 		return;
200 
201 	for (rtx = 0; rtx < ZPCI_TABLE_ENTRIES; rtx++)
202 		if (reg_entry_isvalid(table[rtx]))
203 			dma_free_seg_table(table[rtx]);
204 
205 	dma_free_cpu_table(table);
206 }
207 
208 static unsigned long *dma_alloc_page_table(gfp_t gfp)
209 {
210 	unsigned long *table, *entry;
211 
212 	table = kmem_cache_alloc(dma_page_table_cache, gfp);
213 	if (!table)
214 		return NULL;
215 
216 	for (entry = table; entry < table + ZPCI_PT_ENTRIES; entry++)
217 		*entry = ZPCI_PTE_INVALID;
218 	return table;
219 }
220 
221 static unsigned long *dma_get_seg_table_origin(unsigned long *rtep, gfp_t gfp)
222 {
223 	unsigned long old_rte, rte;
224 	unsigned long *sto;
225 
226 	rte = READ_ONCE(*rtep);
227 	if (reg_entry_isvalid(rte)) {
228 		sto = get_rt_sto(rte);
229 	} else {
230 		sto = dma_alloc_cpu_table(gfp);
231 		if (!sto)
232 			return NULL;
233 
234 		set_rt_sto(&rte, virt_to_phys(sto));
235 		validate_rt_entry(&rte);
236 		entry_clr_protected(&rte);
237 
238 		old_rte = cmpxchg(rtep, ZPCI_TABLE_INVALID, rte);
239 		if (old_rte != ZPCI_TABLE_INVALID) {
240 			/* Somone else was faster, use theirs */
241 			dma_free_cpu_table(sto);
242 			sto = get_rt_sto(old_rte);
243 		}
244 	}
245 	return sto;
246 }
247 
248 static unsigned long *dma_get_page_table_origin(unsigned long *step, gfp_t gfp)
249 {
250 	unsigned long old_ste, ste;
251 	unsigned long *pto;
252 
253 	ste = READ_ONCE(*step);
254 	if (reg_entry_isvalid(ste)) {
255 		pto = get_st_pto(ste);
256 	} else {
257 		pto = dma_alloc_page_table(gfp);
258 		if (!pto)
259 			return NULL;
260 		set_st_pto(&ste, virt_to_phys(pto));
261 		validate_st_entry(&ste);
262 		entry_clr_protected(&ste);
263 
264 		old_ste = cmpxchg(step, ZPCI_TABLE_INVALID, ste);
265 		if (old_ste != ZPCI_TABLE_INVALID) {
266 			/* Somone else was faster, use theirs */
267 			dma_free_page_table(pto);
268 			pto = get_st_pto(old_ste);
269 		}
270 	}
271 	return pto;
272 }
273 
274 static unsigned long *dma_walk_cpu_trans(unsigned long *rto, dma_addr_t dma_addr, gfp_t gfp)
275 {
276 	unsigned long *sto, *pto;
277 	unsigned int rtx, sx, px;
278 
279 	rtx = calc_rtx(dma_addr);
280 	sto = dma_get_seg_table_origin(&rto[rtx], gfp);
281 	if (!sto)
282 		return NULL;
283 
284 	sx = calc_sx(dma_addr);
285 	pto = dma_get_page_table_origin(&sto[sx], gfp);
286 	if (!pto)
287 		return NULL;
288 
289 	px = calc_px(dma_addr);
290 	return &pto[px];
291 }
292 
293 static void dma_update_cpu_trans(unsigned long *ptep, phys_addr_t page_addr, int flags)
294 {
295 	unsigned long pte;
296 
297 	pte = READ_ONCE(*ptep);
298 	if (flags & ZPCI_PTE_INVALID) {
299 		invalidate_pt_entry(&pte);
300 	} else {
301 		set_pt_pfaa(&pte, page_addr);
302 		validate_pt_entry(&pte);
303 	}
304 
305 	if (flags & ZPCI_TABLE_PROTECTED)
306 		entry_set_protected(&pte);
307 	else
308 		entry_clr_protected(&pte);
309 
310 	xchg(ptep, pte);
311 }
312 
313 static struct s390_domain *to_s390_domain(struct iommu_domain *dom)
314 {
315 	return container_of(dom, struct s390_domain, domain);
316 }
317 
318 static bool s390_iommu_capable(struct device *dev, enum iommu_cap cap)
319 {
320 	struct zpci_dev *zdev = to_zpci_dev(dev);
321 
322 	switch (cap) {
323 	case IOMMU_CAP_CACHE_COHERENCY:
324 		return true;
325 	case IOMMU_CAP_DEFERRED_FLUSH:
326 		return zdev->pft != PCI_FUNC_TYPE_ISM;
327 	default:
328 		return false;
329 	}
330 }
331 
332 static struct iommu_domain *s390_domain_alloc_paging(struct device *dev)
333 {
334 	struct s390_domain *s390_domain;
335 
336 	s390_domain = kzalloc(sizeof(*s390_domain), GFP_KERNEL);
337 	if (!s390_domain)
338 		return NULL;
339 
340 	s390_domain->dma_table = dma_alloc_cpu_table(GFP_KERNEL);
341 	if (!s390_domain->dma_table) {
342 		kfree(s390_domain);
343 		return NULL;
344 	}
345 	s390_domain->domain.geometry.force_aperture = true;
346 	s390_domain->domain.geometry.aperture_start = 0;
347 	s390_domain->domain.geometry.aperture_end = ZPCI_TABLE_SIZE_RT - 1;
348 
349 	spin_lock_init(&s390_domain->list_lock);
350 	INIT_LIST_HEAD_RCU(&s390_domain->devices);
351 
352 	return &s390_domain->domain;
353 }
354 
355 static void s390_iommu_rcu_free_domain(struct rcu_head *head)
356 {
357 	struct s390_domain *s390_domain = container_of(head, struct s390_domain, rcu);
358 
359 	dma_cleanup_tables(s390_domain->dma_table);
360 	kfree(s390_domain);
361 }
362 
363 static void s390_domain_free(struct iommu_domain *domain)
364 {
365 	struct s390_domain *s390_domain = to_s390_domain(domain);
366 
367 	rcu_read_lock();
368 	WARN_ON(!list_empty(&s390_domain->devices));
369 	rcu_read_unlock();
370 
371 	call_rcu(&s390_domain->rcu, s390_iommu_rcu_free_domain);
372 }
373 
374 static void zdev_s390_domain_update(struct zpci_dev *zdev,
375 				    struct iommu_domain *domain)
376 {
377 	unsigned long flags;
378 
379 	spin_lock_irqsave(&zdev->dom_lock, flags);
380 	zdev->s390_domain = domain;
381 	spin_unlock_irqrestore(&zdev->dom_lock, flags);
382 }
383 
384 static int blocking_domain_attach_device(struct iommu_domain *domain,
385 					 struct device *dev)
386 {
387 	struct zpci_dev *zdev = to_zpci_dev(dev);
388 	struct s390_domain *s390_domain;
389 	unsigned long flags;
390 
391 	if (zdev->s390_domain->type == IOMMU_DOMAIN_BLOCKED)
392 		return 0;
393 
394 	s390_domain = to_s390_domain(zdev->s390_domain);
395 	spin_lock_irqsave(&s390_domain->list_lock, flags);
396 	list_del_rcu(&zdev->iommu_list);
397 	spin_unlock_irqrestore(&s390_domain->list_lock, flags);
398 
399 	zpci_unregister_ioat(zdev, 0);
400 	zdev->dma_table = NULL;
401 	zdev_s390_domain_update(zdev, domain);
402 
403 	return 0;
404 }
405 
406 static int s390_iommu_attach_device(struct iommu_domain *domain,
407 				    struct device *dev)
408 {
409 	struct s390_domain *s390_domain = to_s390_domain(domain);
410 	struct zpci_dev *zdev = to_zpci_dev(dev);
411 	unsigned long flags;
412 	u8 status;
413 	int cc;
414 
415 	if (!zdev)
416 		return -ENODEV;
417 
418 	if (WARN_ON(domain->geometry.aperture_start > zdev->end_dma ||
419 		domain->geometry.aperture_end < zdev->start_dma))
420 		return -EINVAL;
421 
422 	blocking_domain_attach_device(&blocking_domain, dev);
423 
424 	/* If we fail now DMA remains blocked via blocking domain */
425 	cc = zpci_register_ioat(zdev, 0, zdev->start_dma, zdev->end_dma,
426 				virt_to_phys(s390_domain->dma_table), &status);
427 	if (cc && status != ZPCI_PCI_ST_FUNC_NOT_AVAIL)
428 		return -EIO;
429 	zdev->dma_table = s390_domain->dma_table;
430 	zdev_s390_domain_update(zdev, domain);
431 
432 	spin_lock_irqsave(&s390_domain->list_lock, flags);
433 	list_add_rcu(&zdev->iommu_list, &s390_domain->devices);
434 	spin_unlock_irqrestore(&s390_domain->list_lock, flags);
435 
436 	return 0;
437 }
438 
439 static void s390_iommu_get_resv_regions(struct device *dev,
440 					struct list_head *list)
441 {
442 	struct zpci_dev *zdev = to_zpci_dev(dev);
443 	struct iommu_resv_region *region;
444 
445 	if (zdev->start_dma) {
446 		region = iommu_alloc_resv_region(0, zdev->start_dma, 0,
447 						 IOMMU_RESV_RESERVED, GFP_KERNEL);
448 		if (!region)
449 			return;
450 		list_add_tail(&region->list, list);
451 	}
452 
453 	if (zdev->end_dma < ZPCI_TABLE_SIZE_RT - 1) {
454 		region = iommu_alloc_resv_region(zdev->end_dma + 1,
455 						 ZPCI_TABLE_SIZE_RT - zdev->end_dma - 1,
456 						 0, IOMMU_RESV_RESERVED, GFP_KERNEL);
457 		if (!region)
458 			return;
459 		list_add_tail(&region->list, list);
460 	}
461 }
462 
463 static struct iommu_device *s390_iommu_probe_device(struct device *dev)
464 {
465 	struct zpci_dev *zdev;
466 
467 	if (!dev_is_pci(dev))
468 		return ERR_PTR(-ENODEV);
469 
470 	zdev = to_zpci_dev(dev);
471 
472 	if (zdev->start_dma > zdev->end_dma ||
473 	    zdev->start_dma > ZPCI_TABLE_SIZE_RT - 1)
474 		return ERR_PTR(-EINVAL);
475 
476 	if (zdev->end_dma > ZPCI_TABLE_SIZE_RT - 1)
477 		zdev->end_dma = ZPCI_TABLE_SIZE_RT - 1;
478 
479 	if (zdev->tlb_refresh)
480 		dev->iommu->shadow_on_flush = 1;
481 
482 	/* Start with DMA blocked */
483 	spin_lock_init(&zdev->dom_lock);
484 	zdev_s390_domain_update(zdev, &blocking_domain);
485 
486 	return &zdev->iommu_dev;
487 }
488 
489 static int zpci_refresh_all(struct zpci_dev *zdev)
490 {
491 	return zpci_refresh_trans((u64)zdev->fh << 32, zdev->start_dma,
492 				  zdev->end_dma - zdev->start_dma + 1);
493 }
494 
495 static void s390_iommu_flush_iotlb_all(struct iommu_domain *domain)
496 {
497 	struct s390_domain *s390_domain = to_s390_domain(domain);
498 	struct zpci_dev *zdev;
499 
500 	rcu_read_lock();
501 	list_for_each_entry_rcu(zdev, &s390_domain->devices, iommu_list) {
502 		atomic64_inc(&s390_domain->ctrs.global_rpcits);
503 		zpci_refresh_all(zdev);
504 	}
505 	rcu_read_unlock();
506 }
507 
508 static void s390_iommu_iotlb_sync(struct iommu_domain *domain,
509 				  struct iommu_iotlb_gather *gather)
510 {
511 	struct s390_domain *s390_domain = to_s390_domain(domain);
512 	size_t size = gather->end - gather->start + 1;
513 	struct zpci_dev *zdev;
514 
515 	/* If gather was never added to there is nothing to flush */
516 	if (!gather->end)
517 		return;
518 
519 	rcu_read_lock();
520 	list_for_each_entry_rcu(zdev, &s390_domain->devices, iommu_list) {
521 		atomic64_inc(&s390_domain->ctrs.sync_rpcits);
522 		zpci_refresh_trans((u64)zdev->fh << 32, gather->start,
523 				   size);
524 	}
525 	rcu_read_unlock();
526 }
527 
528 static int s390_iommu_iotlb_sync_map(struct iommu_domain *domain,
529 				     unsigned long iova, size_t size)
530 {
531 	struct s390_domain *s390_domain = to_s390_domain(domain);
532 	struct zpci_dev *zdev;
533 	int ret = 0;
534 
535 	rcu_read_lock();
536 	list_for_each_entry_rcu(zdev, &s390_domain->devices, iommu_list) {
537 		if (!zdev->tlb_refresh)
538 			continue;
539 		atomic64_inc(&s390_domain->ctrs.sync_map_rpcits);
540 		ret = zpci_refresh_trans((u64)zdev->fh << 32,
541 					 iova, size);
542 		/*
543 		 * let the hypervisor discover invalidated entries
544 		 * allowing it to free IOVAs and unpin pages
545 		 */
546 		if (ret == -ENOMEM) {
547 			ret = zpci_refresh_all(zdev);
548 			if (ret)
549 				break;
550 		}
551 	}
552 	rcu_read_unlock();
553 
554 	return ret;
555 }
556 
557 static int s390_iommu_validate_trans(struct s390_domain *s390_domain,
558 				     phys_addr_t pa, dma_addr_t dma_addr,
559 				     unsigned long nr_pages, int flags,
560 				     gfp_t gfp)
561 {
562 	phys_addr_t page_addr = pa & PAGE_MASK;
563 	unsigned long *entry;
564 	unsigned long i;
565 	int rc;
566 
567 	for (i = 0; i < nr_pages; i++) {
568 		entry = dma_walk_cpu_trans(s390_domain->dma_table, dma_addr,
569 					   gfp);
570 		if (unlikely(!entry)) {
571 			rc = -ENOMEM;
572 			goto undo_cpu_trans;
573 		}
574 		dma_update_cpu_trans(entry, page_addr, flags);
575 		page_addr += PAGE_SIZE;
576 		dma_addr += PAGE_SIZE;
577 	}
578 
579 	return 0;
580 
581 undo_cpu_trans:
582 	while (i-- > 0) {
583 		dma_addr -= PAGE_SIZE;
584 		entry = dma_walk_cpu_trans(s390_domain->dma_table,
585 					   dma_addr, gfp);
586 		if (!entry)
587 			break;
588 		dma_update_cpu_trans(entry, 0, ZPCI_PTE_INVALID);
589 	}
590 
591 	return rc;
592 }
593 
594 static int s390_iommu_invalidate_trans(struct s390_domain *s390_domain,
595 				       dma_addr_t dma_addr, unsigned long nr_pages)
596 {
597 	unsigned long *entry;
598 	unsigned long i;
599 	int rc = 0;
600 
601 	for (i = 0; i < nr_pages; i++) {
602 		entry = dma_walk_cpu_trans(s390_domain->dma_table, dma_addr,
603 					   GFP_ATOMIC);
604 		if (unlikely(!entry)) {
605 			rc = -EINVAL;
606 			break;
607 		}
608 		dma_update_cpu_trans(entry, 0, ZPCI_PTE_INVALID);
609 		dma_addr += PAGE_SIZE;
610 	}
611 
612 	return rc;
613 }
614 
615 static int s390_iommu_map_pages(struct iommu_domain *domain,
616 				unsigned long iova, phys_addr_t paddr,
617 				size_t pgsize, size_t pgcount,
618 				int prot, gfp_t gfp, size_t *mapped)
619 {
620 	struct s390_domain *s390_domain = to_s390_domain(domain);
621 	size_t size = pgcount << __ffs(pgsize);
622 	int flags = ZPCI_PTE_VALID, rc = 0;
623 
624 	if (pgsize != SZ_4K)
625 		return -EINVAL;
626 
627 	if (iova < s390_domain->domain.geometry.aperture_start ||
628 	    (iova + size - 1) > s390_domain->domain.geometry.aperture_end)
629 		return -EINVAL;
630 
631 	if (!IS_ALIGNED(iova | paddr, pgsize))
632 		return -EINVAL;
633 
634 	if (!(prot & IOMMU_WRITE))
635 		flags |= ZPCI_TABLE_PROTECTED;
636 
637 	rc = s390_iommu_validate_trans(s390_domain, paddr, iova,
638 				     pgcount, flags, gfp);
639 	if (!rc) {
640 		*mapped = size;
641 		atomic64_add(pgcount, &s390_domain->ctrs.mapped_pages);
642 	}
643 
644 	return rc;
645 }
646 
647 static phys_addr_t s390_iommu_iova_to_phys(struct iommu_domain *domain,
648 					   dma_addr_t iova)
649 {
650 	struct s390_domain *s390_domain = to_s390_domain(domain);
651 	unsigned long *rto, *sto, *pto;
652 	unsigned long ste, pte, rte;
653 	unsigned int rtx, sx, px;
654 	phys_addr_t phys = 0;
655 
656 	if (iova < domain->geometry.aperture_start ||
657 	    iova > domain->geometry.aperture_end)
658 		return 0;
659 
660 	rtx = calc_rtx(iova);
661 	sx = calc_sx(iova);
662 	px = calc_px(iova);
663 	rto = s390_domain->dma_table;
664 
665 	rte = READ_ONCE(rto[rtx]);
666 	if (reg_entry_isvalid(rte)) {
667 		sto = get_rt_sto(rte);
668 		ste = READ_ONCE(sto[sx]);
669 		if (reg_entry_isvalid(ste)) {
670 			pto = get_st_pto(ste);
671 			pte = READ_ONCE(pto[px]);
672 			if (pt_entry_isvalid(pte))
673 				phys = pte & ZPCI_PTE_ADDR_MASK;
674 		}
675 	}
676 
677 	return phys;
678 }
679 
680 static size_t s390_iommu_unmap_pages(struct iommu_domain *domain,
681 				     unsigned long iova,
682 				     size_t pgsize, size_t pgcount,
683 				     struct iommu_iotlb_gather *gather)
684 {
685 	struct s390_domain *s390_domain = to_s390_domain(domain);
686 	size_t size = pgcount << __ffs(pgsize);
687 	int rc;
688 
689 	if (WARN_ON(iova < s390_domain->domain.geometry.aperture_start ||
690 	    (iova + size - 1) > s390_domain->domain.geometry.aperture_end))
691 		return 0;
692 
693 	rc = s390_iommu_invalidate_trans(s390_domain, iova, pgcount);
694 	if (rc)
695 		return 0;
696 
697 	iommu_iotlb_gather_add_range(gather, iova, size);
698 	atomic64_add(pgcount, &s390_domain->ctrs.unmapped_pages);
699 
700 	return size;
701 }
702 
703 struct zpci_iommu_ctrs *zpci_get_iommu_ctrs(struct zpci_dev *zdev)
704 {
705 	struct s390_domain *s390_domain;
706 
707 	lockdep_assert_held(&zdev->dom_lock);
708 
709 	if (zdev->s390_domain->type == IOMMU_DOMAIN_BLOCKED)
710 		return NULL;
711 
712 	s390_domain = to_s390_domain(zdev->s390_domain);
713 	return &s390_domain->ctrs;
714 }
715 
716 int zpci_init_iommu(struct zpci_dev *zdev)
717 {
718 	u64 aperture_size;
719 	int rc = 0;
720 
721 	rc = iommu_device_sysfs_add(&zdev->iommu_dev, NULL, NULL,
722 				    "s390-iommu.%08x", zdev->fid);
723 	if (rc)
724 		goto out_err;
725 
726 	rc = iommu_device_register(&zdev->iommu_dev, &s390_iommu_ops, NULL);
727 	if (rc)
728 		goto out_sysfs;
729 
730 	zdev->start_dma = PAGE_ALIGN(zdev->start_dma);
731 	aperture_size = min3(s390_iommu_aperture,
732 			     ZPCI_TABLE_SIZE_RT - zdev->start_dma,
733 			     zdev->end_dma - zdev->start_dma + 1);
734 	zdev->end_dma = zdev->start_dma + aperture_size - 1;
735 
736 	return 0;
737 
738 out_sysfs:
739 	iommu_device_sysfs_remove(&zdev->iommu_dev);
740 
741 out_err:
742 	return rc;
743 }
744 
745 void zpci_destroy_iommu(struct zpci_dev *zdev)
746 {
747 	iommu_device_unregister(&zdev->iommu_dev);
748 	iommu_device_sysfs_remove(&zdev->iommu_dev);
749 }
750 
751 static int __init s390_iommu_setup(char *str)
752 {
753 	if (!strcmp(str, "strict")) {
754 		pr_warn("s390_iommu=strict deprecated; use iommu.strict=1 instead\n");
755 		iommu_set_dma_strict();
756 	}
757 	return 1;
758 }
759 
760 __setup("s390_iommu=", s390_iommu_setup);
761 
762 static int __init s390_iommu_aperture_setup(char *str)
763 {
764 	if (kstrtou32(str, 10, &s390_iommu_aperture_factor))
765 		s390_iommu_aperture_factor = 1;
766 	return 1;
767 }
768 
769 __setup("s390_iommu_aperture=", s390_iommu_aperture_setup);
770 
771 static int __init s390_iommu_init(void)
772 {
773 	int rc;
774 
775 	iommu_dma_forcedac = true;
776 	s390_iommu_aperture = (u64)virt_to_phys(high_memory);
777 	if (!s390_iommu_aperture_factor)
778 		s390_iommu_aperture = ULONG_MAX;
779 	else
780 		s390_iommu_aperture *= s390_iommu_aperture_factor;
781 
782 	rc = dma_alloc_cpu_table_caches();
783 	if (rc)
784 		return rc;
785 
786 	return rc;
787 }
788 subsys_initcall(s390_iommu_init);
789 
790 static struct iommu_domain blocking_domain = {
791 	.type = IOMMU_DOMAIN_BLOCKED,
792 	.ops = &(const struct iommu_domain_ops) {
793 		.attach_dev	= blocking_domain_attach_device,
794 	}
795 };
796 
797 static const struct iommu_ops s390_iommu_ops = {
798 	.blocked_domain		= &blocking_domain,
799 	.release_domain		= &blocking_domain,
800 	.capable = s390_iommu_capable,
801 	.domain_alloc_paging = s390_domain_alloc_paging,
802 	.probe_device = s390_iommu_probe_device,
803 	.device_group = generic_device_group,
804 	.pgsize_bitmap = SZ_4K,
805 	.get_resv_regions = s390_iommu_get_resv_regions,
806 	.default_domain_ops = &(const struct iommu_domain_ops) {
807 		.attach_dev	= s390_iommu_attach_device,
808 		.map_pages	= s390_iommu_map_pages,
809 		.unmap_pages	= s390_iommu_unmap_pages,
810 		.flush_iotlb_all = s390_iommu_flush_iotlb_all,
811 		.iotlb_sync      = s390_iommu_iotlb_sync,
812 		.iotlb_sync_map  = s390_iommu_iotlb_sync_map,
813 		.iova_to_phys	= s390_iommu_iova_to_phys,
814 		.free		= s390_domain_free,
815 	}
816 };
817