xref: /openbmc/linux/drivers/iommu/intel/pasid.c (revision 0da908c2)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * intel-pasid.c - PASID idr, table and entry manipulation
4  *
5  * Copyright (C) 2018 Intel Corporation
6  *
7  * Author: Lu Baolu <baolu.lu@linux.intel.com>
8  */
9 
10 #define pr_fmt(fmt)	"DMAR: " fmt
11 
12 #include <linux/bitops.h>
13 #include <linux/cpufeature.h>
14 #include <linux/dmar.h>
15 #include <linux/iommu.h>
16 #include <linux/memory.h>
17 #include <linux/pci.h>
18 #include <linux/pci-ats.h>
19 #include <linux/spinlock.h>
20 
21 #include "iommu.h"
22 #include "pasid.h"
23 
24 /*
25  * Intel IOMMU system wide PASID name space:
26  */
27 u32 intel_pasid_max_id = PASID_MAX;
28 
29 int vcmd_alloc_pasid(struct intel_iommu *iommu, u32 *pasid)
30 {
31 	unsigned long flags;
32 	u8 status_code;
33 	int ret = 0;
34 	u64 res;
35 
36 	raw_spin_lock_irqsave(&iommu->register_lock, flags);
37 	dmar_writeq(iommu->reg + DMAR_VCMD_REG, VCMD_CMD_ALLOC);
38 	IOMMU_WAIT_OP(iommu, DMAR_VCRSP_REG, dmar_readq,
39 		      !(res & VCMD_VRSP_IP), res);
40 	raw_spin_unlock_irqrestore(&iommu->register_lock, flags);
41 
42 	status_code = VCMD_VRSP_SC(res);
43 	switch (status_code) {
44 	case VCMD_VRSP_SC_SUCCESS:
45 		*pasid = VCMD_VRSP_RESULT_PASID(res);
46 		break;
47 	case VCMD_VRSP_SC_NO_PASID_AVAIL:
48 		pr_info("IOMMU: %s: No PASID available\n", iommu->name);
49 		ret = -ENOSPC;
50 		break;
51 	default:
52 		ret = -ENODEV;
53 		pr_warn("IOMMU: %s: Unexpected error code %d\n",
54 			iommu->name, status_code);
55 	}
56 
57 	return ret;
58 }
59 
60 void vcmd_free_pasid(struct intel_iommu *iommu, u32 pasid)
61 {
62 	unsigned long flags;
63 	u8 status_code;
64 	u64 res;
65 
66 	raw_spin_lock_irqsave(&iommu->register_lock, flags);
67 	dmar_writeq(iommu->reg + DMAR_VCMD_REG,
68 		    VCMD_CMD_OPERAND(pasid) | VCMD_CMD_FREE);
69 	IOMMU_WAIT_OP(iommu, DMAR_VCRSP_REG, dmar_readq,
70 		      !(res & VCMD_VRSP_IP), res);
71 	raw_spin_unlock_irqrestore(&iommu->register_lock, flags);
72 
73 	status_code = VCMD_VRSP_SC(res);
74 	switch (status_code) {
75 	case VCMD_VRSP_SC_SUCCESS:
76 		break;
77 	case VCMD_VRSP_SC_INVALID_PASID:
78 		pr_info("IOMMU: %s: Invalid PASID\n", iommu->name);
79 		break;
80 	default:
81 		pr_warn("IOMMU: %s: Unexpected error code %d\n",
82 			iommu->name, status_code);
83 	}
84 }
85 
86 /*
87  * Per device pasid table management:
88  */
89 
90 /*
91  * Allocate a pasid table for @dev. It should be called in a
92  * single-thread context.
93  */
94 int intel_pasid_alloc_table(struct device *dev)
95 {
96 	struct device_domain_info *info;
97 	struct pasid_table *pasid_table;
98 	struct page *pages;
99 	u32 max_pasid = 0;
100 	int order, size;
101 
102 	might_sleep();
103 	info = dev_iommu_priv_get(dev);
104 	if (WARN_ON(!info || !dev_is_pci(dev)))
105 		return -ENODEV;
106 	if (WARN_ON(info->pasid_table))
107 		return -EEXIST;
108 
109 	pasid_table = kzalloc(sizeof(*pasid_table), GFP_KERNEL);
110 	if (!pasid_table)
111 		return -ENOMEM;
112 
113 	if (info->pasid_supported)
114 		max_pasid = min_t(u32, pci_max_pasids(to_pci_dev(dev)),
115 				  intel_pasid_max_id);
116 
117 	size = max_pasid >> (PASID_PDE_SHIFT - 3);
118 	order = size ? get_order(size) : 0;
119 	pages = alloc_pages_node(info->iommu->node,
120 				 GFP_KERNEL | __GFP_ZERO, order);
121 	if (!pages) {
122 		kfree(pasid_table);
123 		return -ENOMEM;
124 	}
125 
126 	pasid_table->table = page_address(pages);
127 	pasid_table->order = order;
128 	pasid_table->max_pasid = 1 << (order + PAGE_SHIFT + 3);
129 	info->pasid_table = pasid_table;
130 
131 	return 0;
132 }
133 
134 void intel_pasid_free_table(struct device *dev)
135 {
136 	struct device_domain_info *info;
137 	struct pasid_table *pasid_table;
138 	struct pasid_dir_entry *dir;
139 	struct pasid_entry *table;
140 	int i, max_pde;
141 
142 	info = dev_iommu_priv_get(dev);
143 	if (!info || !dev_is_pci(dev) || !info->pasid_table)
144 		return;
145 
146 	pasid_table = info->pasid_table;
147 	info->pasid_table = NULL;
148 
149 	/* Free scalable mode PASID directory tables: */
150 	dir = pasid_table->table;
151 	max_pde = pasid_table->max_pasid >> PASID_PDE_SHIFT;
152 	for (i = 0; i < max_pde; i++) {
153 		table = get_pasid_table_from_pde(&dir[i]);
154 		free_pgtable_page(table);
155 	}
156 
157 	free_pages((unsigned long)pasid_table->table, pasid_table->order);
158 	kfree(pasid_table);
159 }
160 
161 struct pasid_table *intel_pasid_get_table(struct device *dev)
162 {
163 	struct device_domain_info *info;
164 
165 	info = dev_iommu_priv_get(dev);
166 	if (!info)
167 		return NULL;
168 
169 	return info->pasid_table;
170 }
171 
172 static int intel_pasid_get_dev_max_id(struct device *dev)
173 {
174 	struct device_domain_info *info;
175 
176 	info = dev_iommu_priv_get(dev);
177 	if (!info || !info->pasid_table)
178 		return 0;
179 
180 	return info->pasid_table->max_pasid;
181 }
182 
183 static struct pasid_entry *intel_pasid_get_entry(struct device *dev, u32 pasid)
184 {
185 	struct device_domain_info *info;
186 	struct pasid_table *pasid_table;
187 	struct pasid_dir_entry *dir;
188 	struct pasid_entry *entries;
189 	int dir_index, index;
190 
191 	pasid_table = intel_pasid_get_table(dev);
192 	if (WARN_ON(!pasid_table || pasid >= intel_pasid_get_dev_max_id(dev)))
193 		return NULL;
194 
195 	dir = pasid_table->table;
196 	info = dev_iommu_priv_get(dev);
197 	dir_index = pasid >> PASID_PDE_SHIFT;
198 	index = pasid & PASID_PTE_MASK;
199 
200 retry:
201 	entries = get_pasid_table_from_pde(&dir[dir_index]);
202 	if (!entries) {
203 		entries = alloc_pgtable_page(info->iommu->node);
204 		if (!entries)
205 			return NULL;
206 
207 		/*
208 		 * The pasid directory table entry won't be freed after
209 		 * allocation. No worry about the race with free and
210 		 * clear. However, this entry might be populated by others
211 		 * while we are preparing it. Use theirs with a retry.
212 		 */
213 		if (cmpxchg64(&dir[dir_index].val, 0ULL,
214 			      (u64)virt_to_phys(entries) | PASID_PTE_PRESENT)) {
215 			free_pgtable_page(entries);
216 			goto retry;
217 		}
218 	}
219 
220 	return &entries[index];
221 }
222 
223 /*
224  * Interfaces for PASID table entry manipulation:
225  */
226 static inline void pasid_clear_entry(struct pasid_entry *pe)
227 {
228 	WRITE_ONCE(pe->val[0], 0);
229 	WRITE_ONCE(pe->val[1], 0);
230 	WRITE_ONCE(pe->val[2], 0);
231 	WRITE_ONCE(pe->val[3], 0);
232 	WRITE_ONCE(pe->val[4], 0);
233 	WRITE_ONCE(pe->val[5], 0);
234 	WRITE_ONCE(pe->val[6], 0);
235 	WRITE_ONCE(pe->val[7], 0);
236 }
237 
238 static inline void pasid_clear_entry_with_fpd(struct pasid_entry *pe)
239 {
240 	WRITE_ONCE(pe->val[0], PASID_PTE_FPD);
241 	WRITE_ONCE(pe->val[1], 0);
242 	WRITE_ONCE(pe->val[2], 0);
243 	WRITE_ONCE(pe->val[3], 0);
244 	WRITE_ONCE(pe->val[4], 0);
245 	WRITE_ONCE(pe->val[5], 0);
246 	WRITE_ONCE(pe->val[6], 0);
247 	WRITE_ONCE(pe->val[7], 0);
248 }
249 
250 static void
251 intel_pasid_clear_entry(struct device *dev, u32 pasid, bool fault_ignore)
252 {
253 	struct pasid_entry *pe;
254 
255 	pe = intel_pasid_get_entry(dev, pasid);
256 	if (WARN_ON(!pe))
257 		return;
258 
259 	if (fault_ignore && pasid_pte_is_present(pe))
260 		pasid_clear_entry_with_fpd(pe);
261 	else
262 		pasid_clear_entry(pe);
263 }
264 
265 static inline void pasid_set_bits(u64 *ptr, u64 mask, u64 bits)
266 {
267 	u64 old;
268 
269 	old = READ_ONCE(*ptr);
270 	WRITE_ONCE(*ptr, (old & ~mask) | bits);
271 }
272 
273 /*
274  * Setup the DID(Domain Identifier) field (Bit 64~79) of scalable mode
275  * PASID entry.
276  */
277 static inline void
278 pasid_set_domain_id(struct pasid_entry *pe, u64 value)
279 {
280 	pasid_set_bits(&pe->val[1], GENMASK_ULL(15, 0), value);
281 }
282 
283 /*
284  * Get domain ID value of a scalable mode PASID entry.
285  */
286 static inline u16
287 pasid_get_domain_id(struct pasid_entry *pe)
288 {
289 	return (u16)(READ_ONCE(pe->val[1]) & GENMASK_ULL(15, 0));
290 }
291 
292 /*
293  * Setup the SLPTPTR(Second Level Page Table Pointer) field (Bit 12~63)
294  * of a scalable mode PASID entry.
295  */
296 static inline void
297 pasid_set_slptr(struct pasid_entry *pe, u64 value)
298 {
299 	pasid_set_bits(&pe->val[0], VTD_PAGE_MASK, value);
300 }
301 
302 /*
303  * Setup the AW(Address Width) field (Bit 2~4) of a scalable mode PASID
304  * entry.
305  */
306 static inline void
307 pasid_set_address_width(struct pasid_entry *pe, u64 value)
308 {
309 	pasid_set_bits(&pe->val[0], GENMASK_ULL(4, 2), value << 2);
310 }
311 
312 /*
313  * Setup the PGTT(PASID Granular Translation Type) field (Bit 6~8)
314  * of a scalable mode PASID entry.
315  */
316 static inline void
317 pasid_set_translation_type(struct pasid_entry *pe, u64 value)
318 {
319 	pasid_set_bits(&pe->val[0], GENMASK_ULL(8, 6), value << 6);
320 }
321 
322 /*
323  * Enable fault processing by clearing the FPD(Fault Processing
324  * Disable) field (Bit 1) of a scalable mode PASID entry.
325  */
326 static inline void pasid_set_fault_enable(struct pasid_entry *pe)
327 {
328 	pasid_set_bits(&pe->val[0], 1 << 1, 0);
329 }
330 
331 /*
332  * Setup the SRE(Supervisor Request Enable) field (Bit 128) of a
333  * scalable mode PASID entry.
334  */
335 static inline void pasid_set_sre(struct pasid_entry *pe)
336 {
337 	pasid_set_bits(&pe->val[2], 1 << 0, 1);
338 }
339 
340 /*
341  * Setup the WPE(Write Protect Enable) field (Bit 132) of a
342  * scalable mode PASID entry.
343  */
344 static inline void pasid_set_wpe(struct pasid_entry *pe)
345 {
346 	pasid_set_bits(&pe->val[2], 1 << 4, 1 << 4);
347 }
348 
349 /*
350  * Setup the P(Present) field (Bit 0) of a scalable mode PASID
351  * entry.
352  */
353 static inline void pasid_set_present(struct pasid_entry *pe)
354 {
355 	pasid_set_bits(&pe->val[0], 1 << 0, 1);
356 }
357 
358 /*
359  * Setup Page Walk Snoop bit (Bit 87) of a scalable mode PASID
360  * entry.
361  */
362 static inline void pasid_set_page_snoop(struct pasid_entry *pe, bool value)
363 {
364 	pasid_set_bits(&pe->val[1], 1 << 23, value << 23);
365 }
366 
367 /*
368  * Setup the Page Snoop (PGSNP) field (Bit 88) of a scalable mode
369  * PASID entry.
370  */
371 static inline void
372 pasid_set_pgsnp(struct pasid_entry *pe)
373 {
374 	pasid_set_bits(&pe->val[1], 1ULL << 24, 1ULL << 24);
375 }
376 
377 /*
378  * Setup the First Level Page table Pointer field (Bit 140~191)
379  * of a scalable mode PASID entry.
380  */
381 static inline void
382 pasid_set_flptr(struct pasid_entry *pe, u64 value)
383 {
384 	pasid_set_bits(&pe->val[2], VTD_PAGE_MASK, value);
385 }
386 
387 /*
388  * Setup the First Level Paging Mode field (Bit 130~131) of a
389  * scalable mode PASID entry.
390  */
391 static inline void
392 pasid_set_flpm(struct pasid_entry *pe, u64 value)
393 {
394 	pasid_set_bits(&pe->val[2], GENMASK_ULL(3, 2), value << 2);
395 }
396 
397 static void
398 pasid_cache_invalidation_with_pasid(struct intel_iommu *iommu,
399 				    u16 did, u32 pasid)
400 {
401 	struct qi_desc desc;
402 
403 	desc.qw0 = QI_PC_DID(did) | QI_PC_GRAN(QI_PC_PASID_SEL) |
404 		QI_PC_PASID(pasid) | QI_PC_TYPE;
405 	desc.qw1 = 0;
406 	desc.qw2 = 0;
407 	desc.qw3 = 0;
408 
409 	qi_submit_sync(iommu, &desc, 1, 0);
410 }
411 
412 static void
413 devtlb_invalidation_with_pasid(struct intel_iommu *iommu,
414 			       struct device *dev, u32 pasid)
415 {
416 	struct device_domain_info *info;
417 	u16 sid, qdep, pfsid;
418 
419 	info = dev_iommu_priv_get(dev);
420 	if (!info || !info->ats_enabled)
421 		return;
422 
423 	sid = info->bus << 8 | info->devfn;
424 	qdep = info->ats_qdep;
425 	pfsid = info->pfsid;
426 
427 	/*
428 	 * When PASID 0 is used, it indicates RID2PASID(DMA request w/o PASID),
429 	 * devTLB flush w/o PASID should be used. For non-zero PASID under
430 	 * SVA usage, device could do DMA with multiple PASIDs. It is more
431 	 * efficient to flush devTLB specific to the PASID.
432 	 */
433 	if (pasid == PASID_RID2PASID)
434 		qi_flush_dev_iotlb(iommu, sid, pfsid, qdep, 0, 64 - VTD_PAGE_SHIFT);
435 	else
436 		qi_flush_dev_iotlb_pasid(iommu, sid, pfsid, pasid, qdep, 0, 64 - VTD_PAGE_SHIFT);
437 }
438 
439 void intel_pasid_tear_down_entry(struct intel_iommu *iommu, struct device *dev,
440 				 u32 pasid, bool fault_ignore)
441 {
442 	struct pasid_entry *pte;
443 	u16 did, pgtt;
444 
445 	spin_lock(&iommu->lock);
446 	pte = intel_pasid_get_entry(dev, pasid);
447 	if (WARN_ON(!pte) || !pasid_pte_is_present(pte)) {
448 		spin_unlock(&iommu->lock);
449 		return;
450 	}
451 
452 	did = pasid_get_domain_id(pte);
453 	pgtt = pasid_pte_get_pgtt(pte);
454 	intel_pasid_clear_entry(dev, pasid, fault_ignore);
455 	spin_unlock(&iommu->lock);
456 
457 	if (!ecap_coherent(iommu->ecap))
458 		clflush_cache_range(pte, sizeof(*pte));
459 
460 	pasid_cache_invalidation_with_pasid(iommu, did, pasid);
461 
462 	if (pgtt == PASID_ENTRY_PGTT_PT || pgtt == PASID_ENTRY_PGTT_FL_ONLY)
463 		qi_flush_piotlb(iommu, did, pasid, 0, -1, 0);
464 	else
465 		iommu->flush.flush_iotlb(iommu, did, 0, 0, DMA_TLB_DSI_FLUSH);
466 
467 	/* Device IOTLB doesn't need to be flushed in caching mode. */
468 	if (!cap_caching_mode(iommu->cap))
469 		devtlb_invalidation_with_pasid(iommu, dev, pasid);
470 }
471 
472 /*
473  * This function flushes cache for a newly setup pasid table entry.
474  * Caller of it should not modify the in-use pasid table entries.
475  */
476 static void pasid_flush_caches(struct intel_iommu *iommu,
477 				struct pasid_entry *pte,
478 			       u32 pasid, u16 did)
479 {
480 	if (!ecap_coherent(iommu->ecap))
481 		clflush_cache_range(pte, sizeof(*pte));
482 
483 	if (cap_caching_mode(iommu->cap)) {
484 		pasid_cache_invalidation_with_pasid(iommu, did, pasid);
485 		qi_flush_piotlb(iommu, did, pasid, 0, -1, 0);
486 	} else {
487 		iommu_flush_write_buffer(iommu);
488 	}
489 }
490 
491 /*
492  * Set up the scalable mode pasid table entry for first only
493  * translation type.
494  */
495 int intel_pasid_setup_first_level(struct intel_iommu *iommu,
496 				  struct device *dev, pgd_t *pgd,
497 				  u32 pasid, u16 did, int flags)
498 {
499 	struct pasid_entry *pte;
500 
501 	if (!ecap_flts(iommu->ecap)) {
502 		pr_err("No first level translation support on %s\n",
503 		       iommu->name);
504 		return -EINVAL;
505 	}
506 
507 	if (flags & PASID_FLAG_SUPERVISOR_MODE) {
508 #ifdef CONFIG_X86
509 		unsigned long cr0 = read_cr0();
510 
511 		/* CR0.WP is normally set but just to be sure */
512 		if (unlikely(!(cr0 & X86_CR0_WP))) {
513 			pr_err("No CPU write protect!\n");
514 			return -EINVAL;
515 		}
516 #endif
517 		if (!ecap_srs(iommu->ecap)) {
518 			pr_err("No supervisor request support on %s\n",
519 			       iommu->name);
520 			return -EINVAL;
521 		}
522 	}
523 
524 	if ((flags & PASID_FLAG_FL5LP) && !cap_fl5lp_support(iommu->cap)) {
525 		pr_err("No 5-level paging support for first-level on %s\n",
526 		       iommu->name);
527 		return -EINVAL;
528 	}
529 
530 	spin_lock(&iommu->lock);
531 	pte = intel_pasid_get_entry(dev, pasid);
532 	if (!pte) {
533 		spin_unlock(&iommu->lock);
534 		return -ENODEV;
535 	}
536 
537 	if (pasid_pte_is_present(pte)) {
538 		spin_unlock(&iommu->lock);
539 		return -EBUSY;
540 	}
541 
542 	pasid_clear_entry(pte);
543 
544 	/* Setup the first level page table pointer: */
545 	pasid_set_flptr(pte, (u64)__pa(pgd));
546 	if (flags & PASID_FLAG_SUPERVISOR_MODE) {
547 		pasid_set_sre(pte);
548 		pasid_set_wpe(pte);
549 	}
550 
551 	if (flags & PASID_FLAG_FL5LP)
552 		pasid_set_flpm(pte, 1);
553 
554 	if (flags & PASID_FLAG_PAGE_SNOOP)
555 		pasid_set_pgsnp(pte);
556 
557 	pasid_set_domain_id(pte, did);
558 	pasid_set_address_width(pte, iommu->agaw);
559 	pasid_set_page_snoop(pte, !!ecap_smpwc(iommu->ecap));
560 
561 	/* Setup Present and PASID Granular Transfer Type: */
562 	pasid_set_translation_type(pte, PASID_ENTRY_PGTT_FL_ONLY);
563 	pasid_set_present(pte);
564 	spin_unlock(&iommu->lock);
565 
566 	pasid_flush_caches(iommu, pte, pasid, did);
567 
568 	return 0;
569 }
570 
571 /*
572  * Skip top levels of page tables for iommu which has less agaw
573  * than default. Unnecessary for PT mode.
574  */
575 static inline int iommu_skip_agaw(struct dmar_domain *domain,
576 				  struct intel_iommu *iommu,
577 				  struct dma_pte **pgd)
578 {
579 	int agaw;
580 
581 	for (agaw = domain->agaw; agaw > iommu->agaw; agaw--) {
582 		*pgd = phys_to_virt(dma_pte_addr(*pgd));
583 		if (!dma_pte_present(*pgd))
584 			return -EINVAL;
585 	}
586 
587 	return agaw;
588 }
589 
590 /*
591  * Set up the scalable mode pasid entry for second only translation type.
592  */
593 int intel_pasid_setup_second_level(struct intel_iommu *iommu,
594 				   struct dmar_domain *domain,
595 				   struct device *dev, u32 pasid)
596 {
597 	struct pasid_entry *pte;
598 	struct dma_pte *pgd;
599 	u64 pgd_val;
600 	int agaw;
601 	u16 did;
602 
603 	/*
604 	 * If hardware advertises no support for second level
605 	 * translation, return directly.
606 	 */
607 	if (!ecap_slts(iommu->ecap)) {
608 		pr_err("No second level translation support on %s\n",
609 		       iommu->name);
610 		return -EINVAL;
611 	}
612 
613 	pgd = domain->pgd;
614 	agaw = iommu_skip_agaw(domain, iommu, &pgd);
615 	if (agaw < 0) {
616 		dev_err(dev, "Invalid domain page table\n");
617 		return -EINVAL;
618 	}
619 
620 	pgd_val = virt_to_phys(pgd);
621 	did = domain_id_iommu(domain, iommu);
622 
623 	spin_lock(&iommu->lock);
624 	pte = intel_pasid_get_entry(dev, pasid);
625 	if (!pte) {
626 		spin_unlock(&iommu->lock);
627 		return -ENODEV;
628 	}
629 
630 	if (pasid_pte_is_present(pte)) {
631 		spin_unlock(&iommu->lock);
632 		return -EBUSY;
633 	}
634 
635 	pasid_clear_entry(pte);
636 	pasid_set_domain_id(pte, did);
637 	pasid_set_slptr(pte, pgd_val);
638 	pasid_set_address_width(pte, agaw);
639 	pasid_set_translation_type(pte, PASID_ENTRY_PGTT_SL_ONLY);
640 	pasid_set_fault_enable(pte);
641 	pasid_set_page_snoop(pte, !!ecap_smpwc(iommu->ecap));
642 
643 	/*
644 	 * Since it is a second level only translation setup, we should
645 	 * set SRE bit as well (addresses are expected to be GPAs).
646 	 */
647 	if (pasid != PASID_RID2PASID && ecap_srs(iommu->ecap))
648 		pasid_set_sre(pte);
649 	pasid_set_present(pte);
650 	spin_unlock(&iommu->lock);
651 
652 	pasid_flush_caches(iommu, pte, pasid, did);
653 
654 	return 0;
655 }
656 
657 /*
658  * Set up the scalable mode pasid entry for passthrough translation type.
659  */
660 int intel_pasid_setup_pass_through(struct intel_iommu *iommu,
661 				   struct dmar_domain *domain,
662 				   struct device *dev, u32 pasid)
663 {
664 	u16 did = FLPT_DEFAULT_DID;
665 	struct pasid_entry *pte;
666 
667 	spin_lock(&iommu->lock);
668 	pte = intel_pasid_get_entry(dev, pasid);
669 	if (!pte) {
670 		spin_unlock(&iommu->lock);
671 		return -ENODEV;
672 	}
673 
674 	if (pasid_pte_is_present(pte)) {
675 		spin_unlock(&iommu->lock);
676 		return -EBUSY;
677 	}
678 
679 	pasid_clear_entry(pte);
680 	pasid_set_domain_id(pte, did);
681 	pasid_set_address_width(pte, iommu->agaw);
682 	pasid_set_translation_type(pte, PASID_ENTRY_PGTT_PT);
683 	pasid_set_fault_enable(pte);
684 	pasid_set_page_snoop(pte, !!ecap_smpwc(iommu->ecap));
685 
686 	/*
687 	 * We should set SRE bit as well since the addresses are expected
688 	 * to be GPAs.
689 	 */
690 	if (ecap_srs(iommu->ecap))
691 		pasid_set_sre(pte);
692 	pasid_set_present(pte);
693 	spin_unlock(&iommu->lock);
694 
695 	pasid_flush_caches(iommu, pte, pasid, did);
696 
697 	return 0;
698 }
699 
700 /*
701  * Set the page snoop control for a pasid entry which has been set up.
702  */
703 void intel_pasid_setup_page_snoop_control(struct intel_iommu *iommu,
704 					  struct device *dev, u32 pasid)
705 {
706 	struct pasid_entry *pte;
707 	u16 did;
708 
709 	spin_lock(&iommu->lock);
710 	pte = intel_pasid_get_entry(dev, pasid);
711 	if (WARN_ON(!pte || !pasid_pte_is_present(pte))) {
712 		spin_unlock(&iommu->lock);
713 		return;
714 	}
715 
716 	pasid_set_pgsnp(pte);
717 	did = pasid_get_domain_id(pte);
718 	spin_unlock(&iommu->lock);
719 
720 	if (!ecap_coherent(iommu->ecap))
721 		clflush_cache_range(pte, sizeof(*pte));
722 
723 	/*
724 	 * VT-d spec 3.4 table23 states guides for cache invalidation:
725 	 *
726 	 * - PASID-selective-within-Domain PASID-cache invalidation
727 	 * - PASID-selective PASID-based IOTLB invalidation
728 	 * - If (pasid is RID_PASID)
729 	 *    - Global Device-TLB invalidation to affected functions
730 	 *   Else
731 	 *    - PASID-based Device-TLB invalidation (with S=1 and
732 	 *      Addr[63:12]=0x7FFFFFFF_FFFFF) to affected functions
733 	 */
734 	pasid_cache_invalidation_with_pasid(iommu, did, pasid);
735 	qi_flush_piotlb(iommu, did, pasid, 0, -1, 0);
736 
737 	/* Device IOTLB doesn't need to be flushed in caching mode. */
738 	if (!cap_caching_mode(iommu->cap))
739 		devtlb_invalidation_with_pasid(iommu, dev, pasid);
740 }
741