xref: /openbmc/linux/drivers/iommu/intel/pasid.c (revision 31e67366)
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/intel-iommu.h>
16 #include <linux/iommu.h>
17 #include <linux/memory.h>
18 #include <linux/pci.h>
19 #include <linux/pci-ats.h>
20 #include <linux/spinlock.h>
21 
22 #include "pasid.h"
23 
24 /*
25  * Intel IOMMU system wide PASID name space:
26  */
27 static DEFINE_SPINLOCK(pasid_lock);
28 u32 intel_pasid_max_id = PASID_MAX;
29 
30 int vcmd_alloc_pasid(struct intel_iommu *iommu, u32 *pasid)
31 {
32 	unsigned long flags;
33 	u8 status_code;
34 	int ret = 0;
35 	u64 res;
36 
37 	raw_spin_lock_irqsave(&iommu->register_lock, flags);
38 	dmar_writeq(iommu->reg + DMAR_VCMD_REG, VCMD_CMD_ALLOC);
39 	IOMMU_WAIT_OP(iommu, DMAR_VCRSP_REG, dmar_readq,
40 		      !(res & VCMD_VRSP_IP), res);
41 	raw_spin_unlock_irqrestore(&iommu->register_lock, flags);
42 
43 	status_code = VCMD_VRSP_SC(res);
44 	switch (status_code) {
45 	case VCMD_VRSP_SC_SUCCESS:
46 		*pasid = VCMD_VRSP_RESULT_PASID(res);
47 		break;
48 	case VCMD_VRSP_SC_NO_PASID_AVAIL:
49 		pr_info("IOMMU: %s: No PASID available\n", iommu->name);
50 		ret = -ENOSPC;
51 		break;
52 	default:
53 		ret = -ENODEV;
54 		pr_warn("IOMMU: %s: Unexpected error code %d\n",
55 			iommu->name, status_code);
56 	}
57 
58 	return ret;
59 }
60 
61 void vcmd_free_pasid(struct intel_iommu *iommu, u32 pasid)
62 {
63 	unsigned long flags;
64 	u8 status_code;
65 	u64 res;
66 
67 	raw_spin_lock_irqsave(&iommu->register_lock, flags);
68 	dmar_writeq(iommu->reg + DMAR_VCMD_REG,
69 		    VCMD_CMD_OPERAND(pasid) | VCMD_CMD_FREE);
70 	IOMMU_WAIT_OP(iommu, DMAR_VCRSP_REG, dmar_readq,
71 		      !(res & VCMD_VRSP_IP), res);
72 	raw_spin_unlock_irqrestore(&iommu->register_lock, flags);
73 
74 	status_code = VCMD_VRSP_SC(res);
75 	switch (status_code) {
76 	case VCMD_VRSP_SC_SUCCESS:
77 		break;
78 	case VCMD_VRSP_SC_INVALID_PASID:
79 		pr_info("IOMMU: %s: Invalid PASID\n", iommu->name);
80 		break;
81 	default:
82 		pr_warn("IOMMU: %s: Unexpected error code %d\n",
83 			iommu->name, status_code);
84 	}
85 }
86 
87 /*
88  * Per device pasid table management:
89  */
90 static inline void
91 device_attach_pasid_table(struct device_domain_info *info,
92 			  struct pasid_table *pasid_table)
93 {
94 	info->pasid_table = pasid_table;
95 	list_add(&info->table, &pasid_table->dev);
96 }
97 
98 static inline void
99 device_detach_pasid_table(struct device_domain_info *info,
100 			  struct pasid_table *pasid_table)
101 {
102 	info->pasid_table = NULL;
103 	list_del(&info->table);
104 }
105 
106 struct pasid_table_opaque {
107 	struct pasid_table	**pasid_table;
108 	int			segment;
109 	int			bus;
110 	int			devfn;
111 };
112 
113 static int search_pasid_table(struct device_domain_info *info, void *opaque)
114 {
115 	struct pasid_table_opaque *data = opaque;
116 
117 	if (info->iommu->segment == data->segment &&
118 	    info->bus == data->bus &&
119 	    info->devfn == data->devfn &&
120 	    info->pasid_table) {
121 		*data->pasid_table = info->pasid_table;
122 		return 1;
123 	}
124 
125 	return 0;
126 }
127 
128 static int get_alias_pasid_table(struct pci_dev *pdev, u16 alias, void *opaque)
129 {
130 	struct pasid_table_opaque *data = opaque;
131 
132 	data->segment = pci_domain_nr(pdev->bus);
133 	data->bus = PCI_BUS_NUM(alias);
134 	data->devfn = alias & 0xff;
135 
136 	return for_each_device_domain(&search_pasid_table, data);
137 }
138 
139 /*
140  * Allocate a pasid table for @dev. It should be called in a
141  * single-thread context.
142  */
143 int intel_pasid_alloc_table(struct device *dev)
144 {
145 	struct device_domain_info *info;
146 	struct pasid_table *pasid_table;
147 	struct pasid_table_opaque data;
148 	struct page *pages;
149 	u32 max_pasid = 0;
150 	int ret, order;
151 	int size;
152 
153 	might_sleep();
154 	info = get_domain_info(dev);
155 	if (WARN_ON(!info || !dev_is_pci(dev) || info->pasid_table))
156 		return -EINVAL;
157 
158 	/* DMA alias device already has a pasid table, use it: */
159 	data.pasid_table = &pasid_table;
160 	ret = pci_for_each_dma_alias(to_pci_dev(dev),
161 				     &get_alias_pasid_table, &data);
162 	if (ret)
163 		goto attach_out;
164 
165 	pasid_table = kzalloc(sizeof(*pasid_table), GFP_KERNEL);
166 	if (!pasid_table)
167 		return -ENOMEM;
168 	INIT_LIST_HEAD(&pasid_table->dev);
169 
170 	if (info->pasid_supported)
171 		max_pasid = min_t(u32, pci_max_pasids(to_pci_dev(dev)),
172 				  intel_pasid_max_id);
173 
174 	size = max_pasid >> (PASID_PDE_SHIFT - 3);
175 	order = size ? get_order(size) : 0;
176 	pages = alloc_pages_node(info->iommu->node,
177 				 GFP_KERNEL | __GFP_ZERO, order);
178 	if (!pages) {
179 		kfree(pasid_table);
180 		return -ENOMEM;
181 	}
182 
183 	pasid_table->table = page_address(pages);
184 	pasid_table->order = order;
185 	pasid_table->max_pasid = 1 << (order + PAGE_SHIFT + 3);
186 
187 attach_out:
188 	device_attach_pasid_table(info, pasid_table);
189 
190 	return 0;
191 }
192 
193 void intel_pasid_free_table(struct device *dev)
194 {
195 	struct device_domain_info *info;
196 	struct pasid_table *pasid_table;
197 	struct pasid_dir_entry *dir;
198 	struct pasid_entry *table;
199 	int i, max_pde;
200 
201 	info = get_domain_info(dev);
202 	if (!info || !dev_is_pci(dev) || !info->pasid_table)
203 		return;
204 
205 	pasid_table = info->pasid_table;
206 	device_detach_pasid_table(info, pasid_table);
207 
208 	if (!list_empty(&pasid_table->dev))
209 		return;
210 
211 	/* Free scalable mode PASID directory tables: */
212 	dir = pasid_table->table;
213 	max_pde = pasid_table->max_pasid >> PASID_PDE_SHIFT;
214 	for (i = 0; i < max_pde; i++) {
215 		table = get_pasid_table_from_pde(&dir[i]);
216 		free_pgtable_page(table);
217 	}
218 
219 	free_pages((unsigned long)pasid_table->table, pasid_table->order);
220 	kfree(pasid_table);
221 }
222 
223 struct pasid_table *intel_pasid_get_table(struct device *dev)
224 {
225 	struct device_domain_info *info;
226 
227 	info = get_domain_info(dev);
228 	if (!info)
229 		return NULL;
230 
231 	return info->pasid_table;
232 }
233 
234 int intel_pasid_get_dev_max_id(struct device *dev)
235 {
236 	struct device_domain_info *info;
237 
238 	info = get_domain_info(dev);
239 	if (!info || !info->pasid_table)
240 		return 0;
241 
242 	return info->pasid_table->max_pasid;
243 }
244 
245 struct pasid_entry *intel_pasid_get_entry(struct device *dev, u32 pasid)
246 {
247 	struct device_domain_info *info;
248 	struct pasid_table *pasid_table;
249 	struct pasid_dir_entry *dir;
250 	struct pasid_entry *entries;
251 	int dir_index, index;
252 
253 	pasid_table = intel_pasid_get_table(dev);
254 	if (WARN_ON(!pasid_table || pasid >= intel_pasid_get_dev_max_id(dev)))
255 		return NULL;
256 
257 	dir = pasid_table->table;
258 	info = get_domain_info(dev);
259 	dir_index = pasid >> PASID_PDE_SHIFT;
260 	index = pasid & PASID_PTE_MASK;
261 
262 	spin_lock(&pasid_lock);
263 	entries = get_pasid_table_from_pde(&dir[dir_index]);
264 	if (!entries) {
265 		entries = alloc_pgtable_page(info->iommu->node);
266 		if (!entries) {
267 			spin_unlock(&pasid_lock);
268 			return NULL;
269 		}
270 
271 		WRITE_ONCE(dir[dir_index].val,
272 			   (u64)virt_to_phys(entries) | PASID_PTE_PRESENT);
273 	}
274 	spin_unlock(&pasid_lock);
275 
276 	return &entries[index];
277 }
278 
279 /*
280  * Interfaces for PASID table entry manipulation:
281  */
282 static inline void pasid_clear_entry(struct pasid_entry *pe)
283 {
284 	WRITE_ONCE(pe->val[0], 0);
285 	WRITE_ONCE(pe->val[1], 0);
286 	WRITE_ONCE(pe->val[2], 0);
287 	WRITE_ONCE(pe->val[3], 0);
288 	WRITE_ONCE(pe->val[4], 0);
289 	WRITE_ONCE(pe->val[5], 0);
290 	WRITE_ONCE(pe->val[6], 0);
291 	WRITE_ONCE(pe->val[7], 0);
292 }
293 
294 static inline void pasid_clear_entry_with_fpd(struct pasid_entry *pe)
295 {
296 	WRITE_ONCE(pe->val[0], PASID_PTE_FPD);
297 	WRITE_ONCE(pe->val[1], 0);
298 	WRITE_ONCE(pe->val[2], 0);
299 	WRITE_ONCE(pe->val[3], 0);
300 	WRITE_ONCE(pe->val[4], 0);
301 	WRITE_ONCE(pe->val[5], 0);
302 	WRITE_ONCE(pe->val[6], 0);
303 	WRITE_ONCE(pe->val[7], 0);
304 }
305 
306 static void
307 intel_pasid_clear_entry(struct device *dev, u32 pasid, bool fault_ignore)
308 {
309 	struct pasid_entry *pe;
310 
311 	pe = intel_pasid_get_entry(dev, pasid);
312 	if (WARN_ON(!pe))
313 		return;
314 
315 	if (fault_ignore && pasid_pte_is_present(pe))
316 		pasid_clear_entry_with_fpd(pe);
317 	else
318 		pasid_clear_entry(pe);
319 }
320 
321 static inline void pasid_set_bits(u64 *ptr, u64 mask, u64 bits)
322 {
323 	u64 old;
324 
325 	old = READ_ONCE(*ptr);
326 	WRITE_ONCE(*ptr, (old & ~mask) | bits);
327 }
328 
329 /*
330  * Setup the DID(Domain Identifier) field (Bit 64~79) of scalable mode
331  * PASID entry.
332  */
333 static inline void
334 pasid_set_domain_id(struct pasid_entry *pe, u64 value)
335 {
336 	pasid_set_bits(&pe->val[1], GENMASK_ULL(15, 0), value);
337 }
338 
339 /*
340  * Get domain ID value of a scalable mode PASID entry.
341  */
342 static inline u16
343 pasid_get_domain_id(struct pasid_entry *pe)
344 {
345 	return (u16)(READ_ONCE(pe->val[1]) & GENMASK_ULL(15, 0));
346 }
347 
348 /*
349  * Setup the SLPTPTR(Second Level Page Table Pointer) field (Bit 12~63)
350  * of a scalable mode PASID entry.
351  */
352 static inline void
353 pasid_set_slptr(struct pasid_entry *pe, u64 value)
354 {
355 	pasid_set_bits(&pe->val[0], VTD_PAGE_MASK, value);
356 }
357 
358 /*
359  * Setup the AW(Address Width) field (Bit 2~4) of a scalable mode PASID
360  * entry.
361  */
362 static inline void
363 pasid_set_address_width(struct pasid_entry *pe, u64 value)
364 {
365 	pasid_set_bits(&pe->val[0], GENMASK_ULL(4, 2), value << 2);
366 }
367 
368 /*
369  * Setup the PGTT(PASID Granular Translation Type) field (Bit 6~8)
370  * of a scalable mode PASID entry.
371  */
372 static inline void
373 pasid_set_translation_type(struct pasid_entry *pe, u64 value)
374 {
375 	pasid_set_bits(&pe->val[0], GENMASK_ULL(8, 6), value << 6);
376 }
377 
378 /*
379  * Enable fault processing by clearing the FPD(Fault Processing
380  * Disable) field (Bit 1) of a scalable mode PASID entry.
381  */
382 static inline void pasid_set_fault_enable(struct pasid_entry *pe)
383 {
384 	pasid_set_bits(&pe->val[0], 1 << 1, 0);
385 }
386 
387 /*
388  * Setup the SRE(Supervisor Request Enable) field (Bit 128) of a
389  * scalable mode PASID entry.
390  */
391 static inline void pasid_set_sre(struct pasid_entry *pe)
392 {
393 	pasid_set_bits(&pe->val[2], 1 << 0, 1);
394 }
395 
396 /*
397  * Setup the P(Present) field (Bit 0) of a scalable mode PASID
398  * entry.
399  */
400 static inline void pasid_set_present(struct pasid_entry *pe)
401 {
402 	pasid_set_bits(&pe->val[0], 1 << 0, 1);
403 }
404 
405 /*
406  * Setup Page Walk Snoop bit (Bit 87) of a scalable mode PASID
407  * entry.
408  */
409 static inline void pasid_set_page_snoop(struct pasid_entry *pe, bool value)
410 {
411 	pasid_set_bits(&pe->val[1], 1 << 23, value << 23);
412 }
413 
414 /*
415  * Setup the First Level Page table Pointer field (Bit 140~191)
416  * of a scalable mode PASID entry.
417  */
418 static inline void
419 pasid_set_flptr(struct pasid_entry *pe, u64 value)
420 {
421 	pasid_set_bits(&pe->val[2], VTD_PAGE_MASK, value);
422 }
423 
424 /*
425  * Setup the First Level Paging Mode field (Bit 130~131) of a
426  * scalable mode PASID entry.
427  */
428 static inline void
429 pasid_set_flpm(struct pasid_entry *pe, u64 value)
430 {
431 	pasid_set_bits(&pe->val[2], GENMASK_ULL(3, 2), value << 2);
432 }
433 
434 /*
435  * Setup the Extended Access Flag Enable (EAFE) field (Bit 135)
436  * of a scalable mode PASID entry.
437  */
438 static inline void
439 pasid_set_eafe(struct pasid_entry *pe)
440 {
441 	pasid_set_bits(&pe->val[2], 1 << 7, 1 << 7);
442 }
443 
444 static void
445 pasid_cache_invalidation_with_pasid(struct intel_iommu *iommu,
446 				    u16 did, u32 pasid)
447 {
448 	struct qi_desc desc;
449 
450 	desc.qw0 = QI_PC_DID(did) | QI_PC_GRAN(QI_PC_PASID_SEL) |
451 		QI_PC_PASID(pasid) | QI_PC_TYPE;
452 	desc.qw1 = 0;
453 	desc.qw2 = 0;
454 	desc.qw3 = 0;
455 
456 	qi_submit_sync(iommu, &desc, 1, 0);
457 }
458 
459 static void
460 devtlb_invalidation_with_pasid(struct intel_iommu *iommu,
461 			       struct device *dev, u32 pasid)
462 {
463 	struct device_domain_info *info;
464 	u16 sid, qdep, pfsid;
465 
466 	info = get_domain_info(dev);
467 	if (!info || !info->ats_enabled)
468 		return;
469 
470 	sid = info->bus << 8 | info->devfn;
471 	qdep = info->ats_qdep;
472 	pfsid = info->pfsid;
473 
474 	/*
475 	 * When PASID 0 is used, it indicates RID2PASID(DMA request w/o PASID),
476 	 * devTLB flush w/o PASID should be used. For non-zero PASID under
477 	 * SVA usage, device could do DMA with multiple PASIDs. It is more
478 	 * efficient to flush devTLB specific to the PASID.
479 	 */
480 	if (pasid == PASID_RID2PASID)
481 		qi_flush_dev_iotlb(iommu, sid, pfsid, qdep, 0, 64 - VTD_PAGE_SHIFT);
482 	else
483 		qi_flush_dev_iotlb_pasid(iommu, sid, pfsid, pasid, qdep, 0, 64 - VTD_PAGE_SHIFT);
484 }
485 
486 void intel_pasid_tear_down_entry(struct intel_iommu *iommu, struct device *dev,
487 				 u32 pasid, bool fault_ignore)
488 {
489 	struct pasid_entry *pte;
490 	u16 did;
491 
492 	pte = intel_pasid_get_entry(dev, pasid);
493 	if (WARN_ON(!pte))
494 		return;
495 
496 	did = pasid_get_domain_id(pte);
497 	intel_pasid_clear_entry(dev, pasid, fault_ignore);
498 
499 	if (!ecap_coherent(iommu->ecap))
500 		clflush_cache_range(pte, sizeof(*pte));
501 
502 	pasid_cache_invalidation_with_pasid(iommu, did, pasid);
503 	qi_flush_piotlb(iommu, did, pasid, 0, -1, 0);
504 
505 	/* Device IOTLB doesn't need to be flushed in caching mode. */
506 	if (!cap_caching_mode(iommu->cap))
507 		devtlb_invalidation_with_pasid(iommu, dev, pasid);
508 }
509 
510 static void pasid_flush_caches(struct intel_iommu *iommu,
511 				struct pasid_entry *pte,
512 			       u32 pasid, u16 did)
513 {
514 	if (!ecap_coherent(iommu->ecap))
515 		clflush_cache_range(pte, sizeof(*pte));
516 
517 	if (cap_caching_mode(iommu->cap)) {
518 		pasid_cache_invalidation_with_pasid(iommu, did, pasid);
519 		qi_flush_piotlb(iommu, did, pasid, 0, -1, 0);
520 	} else {
521 		iommu_flush_write_buffer(iommu);
522 	}
523 }
524 
525 /*
526  * Set up the scalable mode pasid table entry for first only
527  * translation type.
528  */
529 int intel_pasid_setup_first_level(struct intel_iommu *iommu,
530 				  struct device *dev, pgd_t *pgd,
531 				  u32 pasid, u16 did, int flags)
532 {
533 	struct pasid_entry *pte;
534 
535 	if (!ecap_flts(iommu->ecap)) {
536 		pr_err("No first level translation support on %s\n",
537 		       iommu->name);
538 		return -EINVAL;
539 	}
540 
541 	pte = intel_pasid_get_entry(dev, pasid);
542 	if (WARN_ON(!pte))
543 		return -EINVAL;
544 
545 	pasid_clear_entry(pte);
546 
547 	/* Setup the first level page table pointer: */
548 	pasid_set_flptr(pte, (u64)__pa(pgd));
549 	if (flags & PASID_FLAG_SUPERVISOR_MODE) {
550 		if (!ecap_srs(iommu->ecap)) {
551 			pr_err("No supervisor request support on %s\n",
552 			       iommu->name);
553 			return -EINVAL;
554 		}
555 		pasid_set_sre(pte);
556 	}
557 
558 	if (flags & PASID_FLAG_FL5LP) {
559 		if (cap_5lp_support(iommu->cap)) {
560 			pasid_set_flpm(pte, 1);
561 		} else {
562 			pr_err("No 5-level paging support for first-level\n");
563 			pasid_clear_entry(pte);
564 			return -EINVAL;
565 		}
566 	}
567 
568 	pasid_set_domain_id(pte, did);
569 	pasid_set_address_width(pte, iommu->agaw);
570 	pasid_set_page_snoop(pte, !!ecap_smpwc(iommu->ecap));
571 
572 	/* Setup Present and PASID Granular Transfer Type: */
573 	pasid_set_translation_type(pte, PASID_ENTRY_PGTT_FL_ONLY);
574 	pasid_set_present(pte);
575 	pasid_flush_caches(iommu, pte, pasid, did);
576 
577 	return 0;
578 }
579 
580 /*
581  * Skip top levels of page tables for iommu which has less agaw
582  * than default. Unnecessary for PT mode.
583  */
584 static inline int iommu_skip_agaw(struct dmar_domain *domain,
585 				  struct intel_iommu *iommu,
586 				  struct dma_pte **pgd)
587 {
588 	int agaw;
589 
590 	for (agaw = domain->agaw; agaw > iommu->agaw; agaw--) {
591 		*pgd = phys_to_virt(dma_pte_addr(*pgd));
592 		if (!dma_pte_present(*pgd))
593 			return -EINVAL;
594 	}
595 
596 	return agaw;
597 }
598 
599 /*
600  * Set up the scalable mode pasid entry for second only translation type.
601  */
602 int intel_pasid_setup_second_level(struct intel_iommu *iommu,
603 				   struct dmar_domain *domain,
604 				   struct device *dev, u32 pasid)
605 {
606 	struct pasid_entry *pte;
607 	struct dma_pte *pgd;
608 	u64 pgd_val;
609 	int agaw;
610 	u16 did;
611 
612 	/*
613 	 * If hardware advertises no support for second level
614 	 * translation, return directly.
615 	 */
616 	if (!ecap_slts(iommu->ecap)) {
617 		pr_err("No second level translation support on %s\n",
618 		       iommu->name);
619 		return -EINVAL;
620 	}
621 
622 	pgd = domain->pgd;
623 	agaw = iommu_skip_agaw(domain, iommu, &pgd);
624 	if (agaw < 0) {
625 		dev_err(dev, "Invalid domain page table\n");
626 		return -EINVAL;
627 	}
628 
629 	pgd_val = virt_to_phys(pgd);
630 	did = domain->iommu_did[iommu->seq_id];
631 
632 	pte = intel_pasid_get_entry(dev, pasid);
633 	if (!pte) {
634 		dev_err(dev, "Failed to get pasid entry of PASID %d\n", pasid);
635 		return -ENODEV;
636 	}
637 
638 	pasid_clear_entry(pte);
639 	pasid_set_domain_id(pte, did);
640 	pasid_set_slptr(pte, pgd_val);
641 	pasid_set_address_width(pte, agaw);
642 	pasid_set_translation_type(pte, PASID_ENTRY_PGTT_SL_ONLY);
643 	pasid_set_fault_enable(pte);
644 	pasid_set_page_snoop(pte, !!ecap_smpwc(iommu->ecap));
645 
646 	/*
647 	 * Since it is a second level only translation setup, we should
648 	 * set SRE bit as well (addresses are expected to be GPAs).
649 	 */
650 	pasid_set_sre(pte);
651 	pasid_set_present(pte);
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 	pte = intel_pasid_get_entry(dev, pasid);
668 	if (!pte) {
669 		dev_err(dev, "Failed to get pasid entry of PASID %d\n", pasid);
670 		return -ENODEV;
671 	}
672 
673 	pasid_clear_entry(pte);
674 	pasid_set_domain_id(pte, did);
675 	pasid_set_address_width(pte, iommu->agaw);
676 	pasid_set_translation_type(pte, PASID_ENTRY_PGTT_PT);
677 	pasid_set_fault_enable(pte);
678 	pasid_set_page_snoop(pte, !!ecap_smpwc(iommu->ecap));
679 
680 	/*
681 	 * We should set SRE bit as well since the addresses are expected
682 	 * to be GPAs.
683 	 */
684 	pasid_set_sre(pte);
685 	pasid_set_present(pte);
686 	pasid_flush_caches(iommu, pte, pasid, did);
687 
688 	return 0;
689 }
690 
691 static int
692 intel_pasid_setup_bind_data(struct intel_iommu *iommu, struct pasid_entry *pte,
693 			    struct iommu_gpasid_bind_data_vtd *pasid_data)
694 {
695 	/*
696 	 * Not all guest PASID table entry fields are passed down during bind,
697 	 * here we only set up the ones that are dependent on guest settings.
698 	 * Execution related bits such as NXE, SMEP are not supported.
699 	 * Other fields, such as snoop related, are set based on host needs
700 	 * regardless of guest settings.
701 	 */
702 	if (pasid_data->flags & IOMMU_SVA_VTD_GPASID_SRE) {
703 		if (!ecap_srs(iommu->ecap)) {
704 			pr_err_ratelimited("No supervisor request support on %s\n",
705 					   iommu->name);
706 			return -EINVAL;
707 		}
708 		pasid_set_sre(pte);
709 	}
710 
711 	if (pasid_data->flags & IOMMU_SVA_VTD_GPASID_EAFE) {
712 		if (!ecap_eafs(iommu->ecap)) {
713 			pr_err_ratelimited("No extended access flag support on %s\n",
714 					   iommu->name);
715 			return -EINVAL;
716 		}
717 		pasid_set_eafe(pte);
718 	}
719 
720 	/*
721 	 * Memory type is only applicable to devices inside processor coherent
722 	 * domain. Will add MTS support once coherent devices are available.
723 	 */
724 	if (pasid_data->flags & IOMMU_SVA_VTD_GPASID_MTS_MASK) {
725 		pr_warn_ratelimited("No memory type support %s\n",
726 				    iommu->name);
727 		return -EINVAL;
728 	}
729 
730 	return 0;
731 }
732 
733 /**
734  * intel_pasid_setup_nested() - Set up PASID entry for nested translation.
735  * This could be used for guest shared virtual address. In this case, the
736  * first level page tables are used for GVA-GPA translation in the guest,
737  * second level page tables are used for GPA-HPA translation.
738  *
739  * @iommu:      IOMMU which the device belong to
740  * @dev:        Device to be set up for translation
741  * @gpgd:       FLPTPTR: First Level Page translation pointer in GPA
742  * @pasid:      PASID to be programmed in the device PASID table
743  * @pasid_data: Additional PASID info from the guest bind request
744  * @domain:     Domain info for setting up second level page tables
745  * @addr_width: Address width of the first level (guest)
746  */
747 int intel_pasid_setup_nested(struct intel_iommu *iommu, struct device *dev,
748 			     pgd_t *gpgd, u32 pasid,
749 			     struct iommu_gpasid_bind_data_vtd *pasid_data,
750 			     struct dmar_domain *domain, int addr_width)
751 {
752 	struct pasid_entry *pte;
753 	struct dma_pte *pgd;
754 	int ret = 0;
755 	u64 pgd_val;
756 	int agaw;
757 	u16 did;
758 
759 	if (!ecap_nest(iommu->ecap)) {
760 		pr_err_ratelimited("IOMMU: %s: No nested translation support\n",
761 				   iommu->name);
762 		return -EINVAL;
763 	}
764 
765 	if (!(domain->flags & DOMAIN_FLAG_NESTING_MODE)) {
766 		pr_err_ratelimited("Domain is not in nesting mode, %x\n",
767 				   domain->flags);
768 		return -EINVAL;
769 	}
770 
771 	pte = intel_pasid_get_entry(dev, pasid);
772 	if (WARN_ON(!pte))
773 		return -EINVAL;
774 
775 	/*
776 	 * Caller must ensure PASID entry is not in use, i.e. not bind the
777 	 * same PASID to the same device twice.
778 	 */
779 	if (pasid_pte_is_present(pte))
780 		return -EBUSY;
781 
782 	pasid_clear_entry(pte);
783 
784 	/* Sanity checking performed by caller to make sure address
785 	 * width matching in two dimensions:
786 	 * 1. CPU vs. IOMMU
787 	 * 2. Guest vs. Host.
788 	 */
789 	switch (addr_width) {
790 #ifdef CONFIG_X86
791 	case ADDR_WIDTH_5LEVEL:
792 		if (!cpu_feature_enabled(X86_FEATURE_LA57) ||
793 		    !cap_5lp_support(iommu->cap)) {
794 			dev_err_ratelimited(dev,
795 					    "5-level paging not supported\n");
796 			return -EINVAL;
797 		}
798 
799 		pasid_set_flpm(pte, 1);
800 		break;
801 #endif
802 	case ADDR_WIDTH_4LEVEL:
803 		pasid_set_flpm(pte, 0);
804 		break;
805 	default:
806 		dev_err_ratelimited(dev, "Invalid guest address width %d\n",
807 				    addr_width);
808 		return -EINVAL;
809 	}
810 
811 	/* First level PGD is in GPA, must be supported by the second level */
812 	if ((uintptr_t)gpgd > domain->max_addr) {
813 		dev_err_ratelimited(dev,
814 				    "Guest PGD %lx not supported, max %llx\n",
815 				    (uintptr_t)gpgd, domain->max_addr);
816 		return -EINVAL;
817 	}
818 	pasid_set_flptr(pte, (uintptr_t)gpgd);
819 
820 	ret = intel_pasid_setup_bind_data(iommu, pte, pasid_data);
821 	if (ret)
822 		return ret;
823 
824 	/* Setup the second level based on the given domain */
825 	pgd = domain->pgd;
826 
827 	agaw = iommu_skip_agaw(domain, iommu, &pgd);
828 	if (agaw < 0) {
829 		dev_err_ratelimited(dev, "Invalid domain page table\n");
830 		return -EINVAL;
831 	}
832 	pgd_val = virt_to_phys(pgd);
833 	pasid_set_slptr(pte, pgd_val);
834 	pasid_set_fault_enable(pte);
835 
836 	did = domain->iommu_did[iommu->seq_id];
837 	pasid_set_domain_id(pte, did);
838 
839 	pasid_set_address_width(pte, agaw);
840 	pasid_set_page_snoop(pte, !!ecap_smpwc(iommu->ecap));
841 
842 	pasid_set_translation_type(pte, PASID_ENTRY_PGTT_NESTED);
843 	pasid_set_present(pte);
844 	pasid_flush_caches(iommu, pte, pasid, did);
845 
846 	return ret;
847 }
848