xref: /openbmc/linux/drivers/iommu/intel/pasid.c (revision f0838790)
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 "intel-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, unsigned int *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, unsigned int 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 	int 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(int, 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, int 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 < 0 ||
255 		    pasid >= intel_pasid_get_dev_max_id(dev)))
256 		return NULL;
257 
258 	dir = pasid_table->table;
259 	info = get_domain_info(dev);
260 	dir_index = pasid >> PASID_PDE_SHIFT;
261 	index = pasid & PASID_PTE_MASK;
262 
263 	spin_lock(&pasid_lock);
264 	entries = get_pasid_table_from_pde(&dir[dir_index]);
265 	if (!entries) {
266 		entries = alloc_pgtable_page(info->iommu->node);
267 		if (!entries) {
268 			spin_unlock(&pasid_lock);
269 			return NULL;
270 		}
271 
272 		WRITE_ONCE(dir[dir_index].val,
273 			   (u64)virt_to_phys(entries) | PASID_PTE_PRESENT);
274 	}
275 	spin_unlock(&pasid_lock);
276 
277 	return &entries[index];
278 }
279 
280 /*
281  * Interfaces for PASID table entry manipulation:
282  */
283 static inline void pasid_clear_entry(struct pasid_entry *pe)
284 {
285 	WRITE_ONCE(pe->val[0], 0);
286 	WRITE_ONCE(pe->val[1], 0);
287 	WRITE_ONCE(pe->val[2], 0);
288 	WRITE_ONCE(pe->val[3], 0);
289 	WRITE_ONCE(pe->val[4], 0);
290 	WRITE_ONCE(pe->val[5], 0);
291 	WRITE_ONCE(pe->val[6], 0);
292 	WRITE_ONCE(pe->val[7], 0);
293 }
294 
295 static inline void pasid_clear_entry_with_fpd(struct pasid_entry *pe)
296 {
297 	WRITE_ONCE(pe->val[0], PASID_PTE_FPD);
298 	WRITE_ONCE(pe->val[1], 0);
299 	WRITE_ONCE(pe->val[2], 0);
300 	WRITE_ONCE(pe->val[3], 0);
301 	WRITE_ONCE(pe->val[4], 0);
302 	WRITE_ONCE(pe->val[5], 0);
303 	WRITE_ONCE(pe->val[6], 0);
304 	WRITE_ONCE(pe->val[7], 0);
305 }
306 
307 static void
308 intel_pasid_clear_entry(struct device *dev, int pasid, bool fault_ignore)
309 {
310 	struct pasid_entry *pe;
311 
312 	pe = intel_pasid_get_entry(dev, pasid);
313 	if (WARN_ON(!pe))
314 		return;
315 
316 	if (fault_ignore && pasid_pte_is_present(pe))
317 		pasid_clear_entry_with_fpd(pe);
318 	else
319 		pasid_clear_entry(pe);
320 }
321 
322 static inline void pasid_set_bits(u64 *ptr, u64 mask, u64 bits)
323 {
324 	u64 old;
325 
326 	old = READ_ONCE(*ptr);
327 	WRITE_ONCE(*ptr, (old & ~mask) | bits);
328 }
329 
330 /*
331  * Setup the DID(Domain Identifier) field (Bit 64~79) of scalable mode
332  * PASID entry.
333  */
334 static inline void
335 pasid_set_domain_id(struct pasid_entry *pe, u64 value)
336 {
337 	pasid_set_bits(&pe->val[1], GENMASK_ULL(15, 0), value);
338 }
339 
340 /*
341  * Get domain ID value of a scalable mode PASID entry.
342  */
343 static inline u16
344 pasid_get_domain_id(struct pasid_entry *pe)
345 {
346 	return (u16)(READ_ONCE(pe->val[1]) & GENMASK_ULL(15, 0));
347 }
348 
349 /*
350  * Setup the SLPTPTR(Second Level Page Table Pointer) field (Bit 12~63)
351  * of a scalable mode PASID entry.
352  */
353 static inline void
354 pasid_set_slptr(struct pasid_entry *pe, u64 value)
355 {
356 	pasid_set_bits(&pe->val[0], VTD_PAGE_MASK, value);
357 }
358 
359 /*
360  * Setup the AW(Address Width) field (Bit 2~4) of a scalable mode PASID
361  * entry.
362  */
363 static inline void
364 pasid_set_address_width(struct pasid_entry *pe, u64 value)
365 {
366 	pasid_set_bits(&pe->val[0], GENMASK_ULL(4, 2), value << 2);
367 }
368 
369 /*
370  * Setup the PGTT(PASID Granular Translation Type) field (Bit 6~8)
371  * of a scalable mode PASID entry.
372  */
373 static inline void
374 pasid_set_translation_type(struct pasid_entry *pe, u64 value)
375 {
376 	pasid_set_bits(&pe->val[0], GENMASK_ULL(8, 6), value << 6);
377 }
378 
379 /*
380  * Enable fault processing by clearing the FPD(Fault Processing
381  * Disable) field (Bit 1) of a scalable mode PASID entry.
382  */
383 static inline void pasid_set_fault_enable(struct pasid_entry *pe)
384 {
385 	pasid_set_bits(&pe->val[0], 1 << 1, 0);
386 }
387 
388 /*
389  * Setup the SRE(Supervisor Request Enable) field (Bit 128) of a
390  * scalable mode PASID entry.
391  */
392 static inline void pasid_set_sre(struct pasid_entry *pe)
393 {
394 	pasid_set_bits(&pe->val[2], 1 << 0, 1);
395 }
396 
397 /*
398  * Setup the P(Present) field (Bit 0) of a scalable mode PASID
399  * entry.
400  */
401 static inline void pasid_set_present(struct pasid_entry *pe)
402 {
403 	pasid_set_bits(&pe->val[0], 1 << 0, 1);
404 }
405 
406 /*
407  * Setup Page Walk Snoop bit (Bit 87) of a scalable mode PASID
408  * entry.
409  */
410 static inline void pasid_set_page_snoop(struct pasid_entry *pe, bool value)
411 {
412 	pasid_set_bits(&pe->val[1], 1 << 23, value << 23);
413 }
414 
415 /*
416  * Setup the First Level Page table Pointer field (Bit 140~191)
417  * of a scalable mode PASID entry.
418  */
419 static inline void
420 pasid_set_flptr(struct pasid_entry *pe, u64 value)
421 {
422 	pasid_set_bits(&pe->val[2], VTD_PAGE_MASK, value);
423 }
424 
425 /*
426  * Setup the First Level Paging Mode field (Bit 130~131) of a
427  * scalable mode PASID entry.
428  */
429 static inline void
430 pasid_set_flpm(struct pasid_entry *pe, u64 value)
431 {
432 	pasid_set_bits(&pe->val[2], GENMASK_ULL(3, 2), value << 2);
433 }
434 
435 /*
436  * Setup the Extended Access Flag Enable (EAFE) field (Bit 135)
437  * of a scalable mode PASID entry.
438  */
439 static inline void
440 pasid_set_eafe(struct pasid_entry *pe)
441 {
442 	pasid_set_bits(&pe->val[2], 1 << 7, 1 << 7);
443 }
444 
445 static void
446 pasid_cache_invalidation_with_pasid(struct intel_iommu *iommu,
447 				    u16 did, int pasid)
448 {
449 	struct qi_desc desc;
450 
451 	desc.qw0 = QI_PC_DID(did) | QI_PC_GRAN(QI_PC_PASID_SEL) |
452 		QI_PC_PASID(pasid) | QI_PC_TYPE;
453 	desc.qw1 = 0;
454 	desc.qw2 = 0;
455 	desc.qw3 = 0;
456 
457 	qi_submit_sync(iommu, &desc, 1, 0);
458 }
459 
460 static void
461 iotlb_invalidation_with_pasid(struct intel_iommu *iommu, u16 did, u32 pasid)
462 {
463 	struct qi_desc desc;
464 
465 	desc.qw0 = QI_EIOTLB_PASID(pasid) | QI_EIOTLB_DID(did) |
466 			QI_EIOTLB_GRAN(QI_GRAN_NONG_PASID) | QI_EIOTLB_TYPE;
467 	desc.qw1 = 0;
468 	desc.qw2 = 0;
469 	desc.qw3 = 0;
470 
471 	qi_submit_sync(iommu, &desc, 1, 0);
472 }
473 
474 static void
475 devtlb_invalidation_with_pasid(struct intel_iommu *iommu,
476 			       struct device *dev, int pasid)
477 {
478 	struct device_domain_info *info;
479 	u16 sid, qdep, pfsid;
480 
481 	info = get_domain_info(dev);
482 	if (!info || !info->ats_enabled)
483 		return;
484 
485 	sid = info->bus << 8 | info->devfn;
486 	qdep = info->ats_qdep;
487 	pfsid = info->pfsid;
488 
489 	qi_flush_dev_iotlb(iommu, sid, pfsid, qdep, 0, 64 - VTD_PAGE_SHIFT);
490 }
491 
492 void intel_pasid_tear_down_entry(struct intel_iommu *iommu, struct device *dev,
493 				 int pasid, bool fault_ignore)
494 {
495 	struct pasid_entry *pte;
496 	u16 did;
497 
498 	pte = intel_pasid_get_entry(dev, pasid);
499 	if (WARN_ON(!pte))
500 		return;
501 
502 	did = pasid_get_domain_id(pte);
503 	intel_pasid_clear_entry(dev, pasid, fault_ignore);
504 
505 	if (!ecap_coherent(iommu->ecap))
506 		clflush_cache_range(pte, sizeof(*pte));
507 
508 	pasid_cache_invalidation_with_pasid(iommu, did, pasid);
509 	iotlb_invalidation_with_pasid(iommu, did, pasid);
510 
511 	/* Device IOTLB doesn't need to be flushed in caching mode. */
512 	if (!cap_caching_mode(iommu->cap))
513 		devtlb_invalidation_with_pasid(iommu, dev, pasid);
514 }
515 
516 static void pasid_flush_caches(struct intel_iommu *iommu,
517 				struct pasid_entry *pte,
518 				int pasid, u16 did)
519 {
520 	if (!ecap_coherent(iommu->ecap))
521 		clflush_cache_range(pte, sizeof(*pte));
522 
523 	if (cap_caching_mode(iommu->cap)) {
524 		pasid_cache_invalidation_with_pasid(iommu, did, pasid);
525 		iotlb_invalidation_with_pasid(iommu, did, pasid);
526 	} else {
527 		iommu_flush_write_buffer(iommu);
528 	}
529 }
530 
531 /*
532  * Set up the scalable mode pasid table entry for first only
533  * translation type.
534  */
535 int intel_pasid_setup_first_level(struct intel_iommu *iommu,
536 				  struct device *dev, pgd_t *pgd,
537 				  int pasid, u16 did, int flags)
538 {
539 	struct pasid_entry *pte;
540 
541 	if (!ecap_flts(iommu->ecap)) {
542 		pr_err("No first level translation support on %s\n",
543 		       iommu->name);
544 		return -EINVAL;
545 	}
546 
547 	pte = intel_pasid_get_entry(dev, pasid);
548 	if (WARN_ON(!pte))
549 		return -EINVAL;
550 
551 	pasid_clear_entry(pte);
552 
553 	/* Setup the first level page table pointer: */
554 	pasid_set_flptr(pte, (u64)__pa(pgd));
555 	if (flags & PASID_FLAG_SUPERVISOR_MODE) {
556 		if (!ecap_srs(iommu->ecap)) {
557 			pr_err("No supervisor request support on %s\n",
558 			       iommu->name);
559 			return -EINVAL;
560 		}
561 		pasid_set_sre(pte);
562 	}
563 
564 	if (flags & PASID_FLAG_FL5LP) {
565 		if (cap_5lp_support(iommu->cap)) {
566 			pasid_set_flpm(pte, 1);
567 		} else {
568 			pr_err("No 5-level paging support for first-level\n");
569 			pasid_clear_entry(pte);
570 			return -EINVAL;
571 		}
572 	}
573 
574 	pasid_set_domain_id(pte, did);
575 	pasid_set_address_width(pte, iommu->agaw);
576 	pasid_set_page_snoop(pte, !!ecap_smpwc(iommu->ecap));
577 
578 	/* Setup Present and PASID Granular Transfer Type: */
579 	pasid_set_translation_type(pte, PASID_ENTRY_PGTT_FL_ONLY);
580 	pasid_set_present(pte);
581 	pasid_flush_caches(iommu, pte, pasid, did);
582 
583 	return 0;
584 }
585 
586 /*
587  * Skip top levels of page tables for iommu which has less agaw
588  * than default. Unnecessary for PT mode.
589  */
590 static inline int iommu_skip_agaw(struct dmar_domain *domain,
591 				  struct intel_iommu *iommu,
592 				  struct dma_pte **pgd)
593 {
594 	int agaw;
595 
596 	for (agaw = domain->agaw; agaw > iommu->agaw; agaw--) {
597 		*pgd = phys_to_virt(dma_pte_addr(*pgd));
598 		if (!dma_pte_present(*pgd))
599 			return -EINVAL;
600 	}
601 
602 	return agaw;
603 }
604 
605 /*
606  * Set up the scalable mode pasid entry for second only translation type.
607  */
608 int intel_pasid_setup_second_level(struct intel_iommu *iommu,
609 				   struct dmar_domain *domain,
610 				   struct device *dev, int pasid)
611 {
612 	struct pasid_entry *pte;
613 	struct dma_pte *pgd;
614 	u64 pgd_val;
615 	int agaw;
616 	u16 did;
617 
618 	/*
619 	 * If hardware advertises no support for second level
620 	 * translation, return directly.
621 	 */
622 	if (!ecap_slts(iommu->ecap)) {
623 		pr_err("No second level translation support on %s\n",
624 		       iommu->name);
625 		return -EINVAL;
626 	}
627 
628 	pgd = domain->pgd;
629 	agaw = iommu_skip_agaw(domain, iommu, &pgd);
630 	if (agaw < 0) {
631 		dev_err(dev, "Invalid domain page table\n");
632 		return -EINVAL;
633 	}
634 
635 	pgd_val = virt_to_phys(pgd);
636 	did = domain->iommu_did[iommu->seq_id];
637 
638 	pte = intel_pasid_get_entry(dev, pasid);
639 	if (!pte) {
640 		dev_err(dev, "Failed to get pasid entry of PASID %d\n", pasid);
641 		return -ENODEV;
642 	}
643 
644 	pasid_clear_entry(pte);
645 	pasid_set_domain_id(pte, did);
646 	pasid_set_slptr(pte, pgd_val);
647 	pasid_set_address_width(pte, agaw);
648 	pasid_set_translation_type(pte, PASID_ENTRY_PGTT_SL_ONLY);
649 	pasid_set_fault_enable(pte);
650 	pasid_set_page_snoop(pte, !!ecap_smpwc(iommu->ecap));
651 
652 	/*
653 	 * Since it is a second level only translation setup, we should
654 	 * set SRE bit as well (addresses are expected to be GPAs).
655 	 */
656 	pasid_set_sre(pte);
657 	pasid_set_present(pte);
658 	pasid_flush_caches(iommu, pte, pasid, did);
659 
660 	return 0;
661 }
662 
663 /*
664  * Set up the scalable mode pasid entry for passthrough translation type.
665  */
666 int intel_pasid_setup_pass_through(struct intel_iommu *iommu,
667 				   struct dmar_domain *domain,
668 				   struct device *dev, int pasid)
669 {
670 	u16 did = FLPT_DEFAULT_DID;
671 	struct pasid_entry *pte;
672 
673 	pte = intel_pasid_get_entry(dev, pasid);
674 	if (!pte) {
675 		dev_err(dev, "Failed to get pasid entry of PASID %d\n", pasid);
676 		return -ENODEV;
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 	pasid_set_sre(pte);
691 	pasid_set_present(pte);
692 	pasid_flush_caches(iommu, pte, pasid, did);
693 
694 	return 0;
695 }
696 
697 static int
698 intel_pasid_setup_bind_data(struct intel_iommu *iommu, struct pasid_entry *pte,
699 			    struct iommu_gpasid_bind_data_vtd *pasid_data)
700 {
701 	/*
702 	 * Not all guest PASID table entry fields are passed down during bind,
703 	 * here we only set up the ones that are dependent on guest settings.
704 	 * Execution related bits such as NXE, SMEP are not supported.
705 	 * Other fields, such as snoop related, are set based on host needs
706 	 * regardless of guest settings.
707 	 */
708 	if (pasid_data->flags & IOMMU_SVA_VTD_GPASID_SRE) {
709 		if (!ecap_srs(iommu->ecap)) {
710 			pr_err_ratelimited("No supervisor request support on %s\n",
711 					   iommu->name);
712 			return -EINVAL;
713 		}
714 		pasid_set_sre(pte);
715 	}
716 
717 	if (pasid_data->flags & IOMMU_SVA_VTD_GPASID_EAFE) {
718 		if (!ecap_eafs(iommu->ecap)) {
719 			pr_err_ratelimited("No extended access flag support on %s\n",
720 					   iommu->name);
721 			return -EINVAL;
722 		}
723 		pasid_set_eafe(pte);
724 	}
725 
726 	/*
727 	 * Memory type is only applicable to devices inside processor coherent
728 	 * domain. Will add MTS support once coherent devices are available.
729 	 */
730 	if (pasid_data->flags & IOMMU_SVA_VTD_GPASID_MTS_MASK) {
731 		pr_warn_ratelimited("No memory type support %s\n",
732 				    iommu->name);
733 		return -EINVAL;
734 	}
735 
736 	return 0;
737 }
738 
739 /**
740  * intel_pasid_setup_nested() - Set up PASID entry for nested translation.
741  * This could be used for guest shared virtual address. In this case, the
742  * first level page tables are used for GVA-GPA translation in the guest,
743  * second level page tables are used for GPA-HPA translation.
744  *
745  * @iommu:      IOMMU which the device belong to
746  * @dev:        Device to be set up for translation
747  * @gpgd:       FLPTPTR: First Level Page translation pointer in GPA
748  * @pasid:      PASID to be programmed in the device PASID table
749  * @pasid_data: Additional PASID info from the guest bind request
750  * @domain:     Domain info for setting up second level page tables
751  * @addr_width: Address width of the first level (guest)
752  */
753 int intel_pasid_setup_nested(struct intel_iommu *iommu, struct device *dev,
754 			     pgd_t *gpgd, int pasid,
755 			     struct iommu_gpasid_bind_data_vtd *pasid_data,
756 			     struct dmar_domain *domain, int addr_width)
757 {
758 	struct pasid_entry *pte;
759 	struct dma_pte *pgd;
760 	int ret = 0;
761 	u64 pgd_val;
762 	int agaw;
763 	u16 did;
764 
765 	if (!ecap_nest(iommu->ecap)) {
766 		pr_err_ratelimited("IOMMU: %s: No nested translation support\n",
767 				   iommu->name);
768 		return -EINVAL;
769 	}
770 
771 	if (!(domain->flags & DOMAIN_FLAG_NESTING_MODE)) {
772 		pr_err_ratelimited("Domain is not in nesting mode, %x\n",
773 				   domain->flags);
774 		return -EINVAL;
775 	}
776 
777 	pte = intel_pasid_get_entry(dev, pasid);
778 	if (WARN_ON(!pte))
779 		return -EINVAL;
780 
781 	/*
782 	 * Caller must ensure PASID entry is not in use, i.e. not bind the
783 	 * same PASID to the same device twice.
784 	 */
785 	if (pasid_pte_is_present(pte))
786 		return -EBUSY;
787 
788 	pasid_clear_entry(pte);
789 
790 	/* Sanity checking performed by caller to make sure address
791 	 * width matching in two dimensions:
792 	 * 1. CPU vs. IOMMU
793 	 * 2. Guest vs. Host.
794 	 */
795 	switch (addr_width) {
796 #ifdef CONFIG_X86
797 	case ADDR_WIDTH_5LEVEL:
798 		if (!cpu_feature_enabled(X86_FEATURE_LA57) ||
799 		    !cap_5lp_support(iommu->cap)) {
800 			dev_err_ratelimited(dev,
801 					    "5-level paging not supported\n");
802 			return -EINVAL;
803 		}
804 
805 		pasid_set_flpm(pte, 1);
806 		break;
807 #endif
808 	case ADDR_WIDTH_4LEVEL:
809 		pasid_set_flpm(pte, 0);
810 		break;
811 	default:
812 		dev_err_ratelimited(dev, "Invalid guest address width %d\n",
813 				    addr_width);
814 		return -EINVAL;
815 	}
816 
817 	/* First level PGD is in GPA, must be supported by the second level */
818 	if ((uintptr_t)gpgd > domain->max_addr) {
819 		dev_err_ratelimited(dev,
820 				    "Guest PGD %lx not supported, max %llx\n",
821 				    (uintptr_t)gpgd, domain->max_addr);
822 		return -EINVAL;
823 	}
824 	pasid_set_flptr(pte, (uintptr_t)gpgd);
825 
826 	ret = intel_pasid_setup_bind_data(iommu, pte, pasid_data);
827 	if (ret)
828 		return ret;
829 
830 	/* Setup the second level based on the given domain */
831 	pgd = domain->pgd;
832 
833 	agaw = iommu_skip_agaw(domain, iommu, &pgd);
834 	if (agaw < 0) {
835 		dev_err_ratelimited(dev, "Invalid domain page table\n");
836 		return -EINVAL;
837 	}
838 	pgd_val = virt_to_phys(pgd);
839 	pasid_set_slptr(pte, pgd_val);
840 	pasid_set_fault_enable(pte);
841 
842 	did = domain->iommu_did[iommu->seq_id];
843 	pasid_set_domain_id(pte, did);
844 
845 	pasid_set_address_width(pte, agaw);
846 	pasid_set_page_snoop(pte, !!ecap_smpwc(iommu->ecap));
847 
848 	pasid_set_translation_type(pte, PASID_ENTRY_PGTT_NESTED);
849 	pasid_set_present(pte);
850 	pasid_flush_caches(iommu, pte, pasid, did);
851 
852 	return ret;
853 }
854