1 // SPDX-License-Identifier: GPL-2.0
3 * intel-pasid.c - PASID idr, table and entry manipulation
5 * Copyright (C) 2018 Intel Corporation
10 #define pr_fmt(fmt) "DMAR: " fmt
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>
25 * Intel IOMMU system wide PASID name space:
27 u32 intel_pasid_max_id = PASID_MAX;
29 int vcmd_alloc_pasid(struct intel_iommu *iommu, u32 *pasid)
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);
42 status_code = VCMD_VRSP_SC(res);
43 switch (status_code) {
44 case VCMD_VRSP_SC_SUCCESS:
45 *pasid = VCMD_VRSP_RESULT_PASID(res);
47 case VCMD_VRSP_SC_NO_PASID_AVAIL:
48 pr_info("IOMMU: %s: No PASID available\n", iommu->name);
53 pr_warn("IOMMU: %s: Unexpected error code %d\n",
54 iommu->name, status_code);
60 void vcmd_free_pasid(struct intel_iommu *iommu, u32 pasid)
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);
73 status_code = VCMD_VRSP_SC(res);
74 switch (status_code) {
75 case VCMD_VRSP_SC_SUCCESS:
77 case VCMD_VRSP_SC_INVALID_PASID:
78 pr_info("IOMMU: %s: Invalid PASID\n", iommu->name);
81 pr_warn("IOMMU: %s: Unexpected error code %d\n",
82 iommu->name, status_code);
87 * Per device pasid table management:
91 * Allocate a pasid table for @dev. It should be called in a
92 * single-thread context.
94 int intel_pasid_alloc_table(struct device *dev)
96 struct device_domain_info *info;
97 struct pasid_table *pasid_table;
103 info = dev_iommu_priv_get(dev);
104 if (WARN_ON(!info || !dev_is_pci(dev) || info->pasid_table))
107 pasid_table = kzalloc(sizeof(*pasid_table), GFP_KERNEL);
111 if (info->pasid_supported)
112 max_pasid = min_t(u32, pci_max_pasids(to_pci_dev(dev)),
115 size = max_pasid >> (PASID_PDE_SHIFT - 3);
116 order = size ? get_order(size) : 0;
117 pages = alloc_pages_node(info->iommu->node,
118 GFP_KERNEL | __GFP_ZERO, order);
124 pasid_table->table = page_address(pages);
125 pasid_table->order = order;
126 pasid_table->max_pasid = 1 << (order + PAGE_SHIFT + 3);
127 info->pasid_table = pasid_table;
132 void intel_pasid_free_table(struct device *dev)
134 struct device_domain_info *info;
135 struct pasid_table *pasid_table;
136 struct pasid_dir_entry *dir;
137 struct pasid_entry *table;
140 info = dev_iommu_priv_get(dev);
141 if (!info || !dev_is_pci(dev) || !info->pasid_table)
144 pasid_table = info->pasid_table;
145 info->pasid_table = NULL;
147 /* Free scalable mode PASID directory tables: */
148 dir = pasid_table->table;
149 max_pde = pasid_table->max_pasid >> PASID_PDE_SHIFT;
150 for (i = 0; i < max_pde; i++) {
151 table = get_pasid_table_from_pde(&dir[i]);
152 free_pgtable_page(table);
155 free_pages((unsigned long)pasid_table->table, pasid_table->order);
159 struct pasid_table *intel_pasid_get_table(struct device *dev)
161 struct device_domain_info *info;
163 info = dev_iommu_priv_get(dev);
167 return info->pasid_table;
170 static int intel_pasid_get_dev_max_id(struct device *dev)
172 struct device_domain_info *info;
174 info = dev_iommu_priv_get(dev);
175 if (!info || !info->pasid_table)
178 return info->pasid_table->max_pasid;
181 static struct pasid_entry *intel_pasid_get_entry(struct device *dev, u32 pasid)
183 struct device_domain_info *info;
184 struct pasid_table *pasid_table;
185 struct pasid_dir_entry *dir;
186 struct pasid_entry *entries;
187 int dir_index, index;
189 pasid_table = intel_pasid_get_table(dev);
190 if (WARN_ON(!pasid_table || pasid >= intel_pasid_get_dev_max_id(dev)))
193 dir = pasid_table->table;
194 info = dev_iommu_priv_get(dev);
195 dir_index = pasid >> PASID_PDE_SHIFT;
196 index = pasid & PASID_PTE_MASK;
199 entries = get_pasid_table_from_pde(&dir[dir_index]);
201 entries = alloc_pgtable_page(info->iommu->node);
206 * The pasid directory table entry won't be freed after
207 * allocation. No worry about the race with free and
208 * clear. However, this entry might be populated by others
209 * while we are preparing it. Use theirs with a retry.
211 if (cmpxchg64(&dir[dir_index].val, 0ULL,
212 (u64)virt_to_phys(entries) | PASID_PTE_PRESENT)) {
213 free_pgtable_page(entries);
218 return &entries[index];
222 * Interfaces for PASID table entry manipulation:
224 static inline void pasid_clear_entry(struct pasid_entry *pe)
226 WRITE_ONCE(pe->val[0], 0);
227 WRITE_ONCE(pe->val[1], 0);
228 WRITE_ONCE(pe->val[2], 0);
229 WRITE_ONCE(pe->val[3], 0);
230 WRITE_ONCE(pe->val[4], 0);
231 WRITE_ONCE(pe->val[5], 0);
232 WRITE_ONCE(pe->val[6], 0);
233 WRITE_ONCE(pe->val[7], 0);
236 static inline void pasid_clear_entry_with_fpd(struct pasid_entry *pe)
238 WRITE_ONCE(pe->val[0], PASID_PTE_FPD);
239 WRITE_ONCE(pe->val[1], 0);
240 WRITE_ONCE(pe->val[2], 0);
241 WRITE_ONCE(pe->val[3], 0);
242 WRITE_ONCE(pe->val[4], 0);
243 WRITE_ONCE(pe->val[5], 0);
244 WRITE_ONCE(pe->val[6], 0);
245 WRITE_ONCE(pe->val[7], 0);
249 intel_pasid_clear_entry(struct device *dev, u32 pasid, bool fault_ignore)
251 struct pasid_entry *pe;
253 pe = intel_pasid_get_entry(dev, pasid);
257 if (fault_ignore && pasid_pte_is_present(pe))
258 pasid_clear_entry_with_fpd(pe);
260 pasid_clear_entry(pe);
263 static inline void pasid_set_bits(u64 *ptr, u64 mask, u64 bits)
267 old = READ_ONCE(*ptr);
268 WRITE_ONCE(*ptr, (old & ~mask) | bits);
272 * Setup the DID(Domain Identifier) field (Bit 64~79) of scalable mode
276 pasid_set_domain_id(struct pasid_entry *pe, u64 value)
278 pasid_set_bits(&pe->val[1], GENMASK_ULL(15, 0), value);
282 * Get domain ID value of a scalable mode PASID entry.
285 pasid_get_domain_id(struct pasid_entry *pe)
287 return (u16)(READ_ONCE(pe->val[1]) & GENMASK_ULL(15, 0));
291 * Setup the SLPTPTR(Second Level Page Table Pointer) field (Bit 12~63)
292 * of a scalable mode PASID entry.
295 pasid_set_slptr(struct pasid_entry *pe, u64 value)
297 pasid_set_bits(&pe->val[0], VTD_PAGE_MASK, value);
301 * Setup the AW(Address Width) field (Bit 2~4) of a scalable mode PASID
305 pasid_set_address_width(struct pasid_entry *pe, u64 value)
307 pasid_set_bits(&pe->val[0], GENMASK_ULL(4, 2), value << 2);
311 * Setup the PGTT(PASID Granular Translation Type) field (Bit 6~8)
312 * of a scalable mode PASID entry.
315 pasid_set_translation_type(struct pasid_entry *pe, u64 value)
317 pasid_set_bits(&pe->val[0], GENMASK_ULL(8, 6), value << 6);
321 * Enable fault processing by clearing the FPD(Fault Processing
322 * Disable) field (Bit 1) of a scalable mode PASID entry.
324 static inline void pasid_set_fault_enable(struct pasid_entry *pe)
326 pasid_set_bits(&pe->val[0], 1 << 1, 0);
330 * Setup the SRE(Supervisor Request Enable) field (Bit 128) of a
331 * scalable mode PASID entry.
333 static inline void pasid_set_sre(struct pasid_entry *pe)
335 pasid_set_bits(&pe->val[2], 1 << 0, 1);
339 * Setup the WPE(Write Protect Enable) field (Bit 132) of a
340 * scalable mode PASID entry.
342 static inline void pasid_set_wpe(struct pasid_entry *pe)
344 pasid_set_bits(&pe->val[2], 1 << 4, 1 << 4);
348 * Setup the P(Present) field (Bit 0) of a scalable mode PASID
351 static inline void pasid_set_present(struct pasid_entry *pe)
353 pasid_set_bits(&pe->val[0], 1 << 0, 1);
357 * Setup Page Walk Snoop bit (Bit 87) of a scalable mode PASID
360 static inline void pasid_set_page_snoop(struct pasid_entry *pe, bool value)
362 pasid_set_bits(&pe->val[1], 1 << 23, value << 23);
366 * Setup the Page Snoop (PGSNP) field (Bit 88) of a scalable mode
370 pasid_set_pgsnp(struct pasid_entry *pe)
372 pasid_set_bits(&pe->val[1], 1ULL << 24, 1ULL << 24);
376 * Setup the First Level Page table Pointer field (Bit 140~191)
377 * of a scalable mode PASID entry.
380 pasid_set_flptr(struct pasid_entry *pe, u64 value)
382 pasid_set_bits(&pe->val[2], VTD_PAGE_MASK, value);
386 * Setup the First Level Paging Mode field (Bit 130~131) of a
387 * scalable mode PASID entry.
390 pasid_set_flpm(struct pasid_entry *pe, u64 value)
392 pasid_set_bits(&pe->val[2], GENMASK_ULL(3, 2), value << 2);
396 * Setup the Extended Access Flag Enable (EAFE) field (Bit 135)
397 * of a scalable mode PASID entry.
400 pasid_set_eafe(struct pasid_entry *pe)
402 pasid_set_bits(&pe->val[2], 1 << 7, 1 << 7);
406 pasid_cache_invalidation_with_pasid(struct intel_iommu *iommu,
411 desc.qw0 = QI_PC_DID(did) | QI_PC_GRAN(QI_PC_PASID_SEL) |
412 QI_PC_PASID(pasid) | QI_PC_TYPE;
417 qi_submit_sync(iommu, &desc, 1, 0);
421 devtlb_invalidation_with_pasid(struct intel_iommu *iommu,
422 struct device *dev, u32 pasid)
424 struct device_domain_info *info;
425 u16 sid, qdep, pfsid;
427 info = dev_iommu_priv_get(dev);
428 if (!info || !info->ats_enabled)
431 sid = info->bus << 8 | info->devfn;
432 qdep = info->ats_qdep;
436 * When PASID 0 is used, it indicates RID2PASID(DMA request w/o PASID),
437 * devTLB flush w/o PASID should be used. For non-zero PASID under
438 * SVA usage, device could do DMA with multiple PASIDs. It is more
439 * efficient to flush devTLB specific to the PASID.
441 if (pasid == PASID_RID2PASID)
442 qi_flush_dev_iotlb(iommu, sid, pfsid, qdep, 0, 64 - VTD_PAGE_SHIFT);
444 qi_flush_dev_iotlb_pasid(iommu, sid, pfsid, pasid, qdep, 0, 64 - VTD_PAGE_SHIFT);
447 void intel_pasid_tear_down_entry(struct intel_iommu *iommu, struct device *dev,
448 u32 pasid, bool fault_ignore)
450 struct pasid_entry *pte;
453 pte = intel_pasid_get_entry(dev, pasid);
457 if (!pasid_pte_is_present(pte))
460 did = pasid_get_domain_id(pte);
461 pgtt = pasid_pte_get_pgtt(pte);
463 intel_pasid_clear_entry(dev, pasid, fault_ignore);
465 if (!ecap_coherent(iommu->ecap))
466 clflush_cache_range(pte, sizeof(*pte));
468 pasid_cache_invalidation_with_pasid(iommu, did, pasid);
470 if (pgtt == PASID_ENTRY_PGTT_PT || pgtt == PASID_ENTRY_PGTT_FL_ONLY)
471 qi_flush_piotlb(iommu, did, pasid, 0, -1, 0);
473 iommu->flush.flush_iotlb(iommu, did, 0, 0, DMA_TLB_DSI_FLUSH);
475 /* Device IOTLB doesn't need to be flushed in caching mode. */
476 if (!cap_caching_mode(iommu->cap))
477 devtlb_invalidation_with_pasid(iommu, dev, pasid);
481 * This function flushes cache for a newly setup pasid table entry.
482 * Caller of it should not modify the in-use pasid table entries.
484 static void pasid_flush_caches(struct intel_iommu *iommu,
485 struct pasid_entry *pte,
488 if (!ecap_coherent(iommu->ecap))
489 clflush_cache_range(pte, sizeof(*pte));
491 if (cap_caching_mode(iommu->cap)) {
492 pasid_cache_invalidation_with_pasid(iommu, did, pasid);
493 qi_flush_piotlb(iommu, did, pasid, 0, -1, 0);
495 iommu_flush_write_buffer(iommu);
499 static inline int pasid_enable_wpe(struct pasid_entry *pte)
502 unsigned long cr0 = read_cr0();
504 /* CR0.WP is normally set but just to be sure */
505 if (unlikely(!(cr0 & X86_CR0_WP))) {
506 pr_err_ratelimited("No CPU write protect!\n");
516 * Set up the scalable mode pasid table entry for first only
519 int intel_pasid_setup_first_level(struct intel_iommu *iommu,
520 struct device *dev, pgd_t *pgd,
521 u32 pasid, u16 did, int flags)
523 struct pasid_entry *pte;
525 if (!ecap_flts(iommu->ecap)) {
526 pr_err("No first level translation support on %s\n",
531 pte = intel_pasid_get_entry(dev, pasid);
535 /* Caller must ensure PASID entry is not in use. */
536 if (pasid_pte_is_present(pte))
539 pasid_clear_entry(pte);
541 /* Setup the first level page table pointer: */
542 pasid_set_flptr(pte, (u64)__pa(pgd));
543 if (flags & PASID_FLAG_SUPERVISOR_MODE) {
544 if (!ecap_srs(iommu->ecap)) {
545 pr_err("No supervisor request support on %s\n",
550 if (pasid_enable_wpe(pte))
555 if (flags & PASID_FLAG_FL5LP) {
556 if (cap_5lp_support(iommu->cap)) {
557 pasid_set_flpm(pte, 1);
559 pr_err("No 5-level paging support for first-level\n");
560 pasid_clear_entry(pte);
565 if (flags & PASID_FLAG_PAGE_SNOOP)
566 pasid_set_pgsnp(pte);
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));
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);
581 * Skip top levels of page tables for iommu which has less agaw
582 * than default. Unnecessary for PT mode.
584 static inline int iommu_skip_agaw(struct dmar_domain *domain,
585 struct intel_iommu *iommu,
586 struct dma_pte **pgd)
590 for (agaw = domain->agaw; agaw > iommu->agaw; agaw--) {
591 *pgd = phys_to_virt(dma_pte_addr(*pgd));
592 if (!dma_pte_present(*pgd))
600 * Set up the scalable mode pasid entry for second only translation type.
602 int intel_pasid_setup_second_level(struct intel_iommu *iommu,
603 struct dmar_domain *domain,
604 struct device *dev, u32 pasid)
606 struct pasid_entry *pte;
613 * If hardware advertises no support for second level
614 * translation, return directly.
616 if (!ecap_slts(iommu->ecap)) {
617 pr_err("No second level translation support on %s\n",
623 agaw = iommu_skip_agaw(domain, iommu, &pgd);
625 dev_err(dev, "Invalid domain page table\n");
629 pgd_val = virt_to_phys(pgd);
630 did = domain->iommu_did[iommu->seq_id];
632 pte = intel_pasid_get_entry(dev, pasid);
634 dev_err(dev, "Failed to get pasid entry of PASID %d\n", pasid);
638 /* Caller must ensure PASID entry is not in use. */
639 if (pasid_pte_is_present(pte))
642 pasid_clear_entry(pte);
643 pasid_set_domain_id(pte, did);
644 pasid_set_slptr(pte, pgd_val);
645 pasid_set_address_width(pte, agaw);
646 pasid_set_translation_type(pte, PASID_ENTRY_PGTT_SL_ONLY);
647 pasid_set_fault_enable(pte);
648 pasid_set_page_snoop(pte, !!ecap_smpwc(iommu->ecap));
651 * Since it is a second level only translation setup, we should
652 * set SRE bit as well (addresses are expected to be GPAs).
654 if (pasid != PASID_RID2PASID)
656 pasid_set_present(pte);
657 pasid_flush_caches(iommu, pte, pasid, did);
663 * Set up the scalable mode pasid entry for passthrough translation type.
665 int intel_pasid_setup_pass_through(struct intel_iommu *iommu,
666 struct dmar_domain *domain,
667 struct device *dev, u32 pasid)
669 u16 did = FLPT_DEFAULT_DID;
670 struct pasid_entry *pte;
672 pte = intel_pasid_get_entry(dev, pasid);
674 dev_err(dev, "Failed to get pasid entry of PASID %d\n", pasid);
678 /* Caller must ensure PASID entry is not in use. */
679 if (pasid_pte_is_present(pte))
682 pasid_clear_entry(pte);
683 pasid_set_domain_id(pte, did);
684 pasid_set_address_width(pte, iommu->agaw);
685 pasid_set_translation_type(pte, PASID_ENTRY_PGTT_PT);
686 pasid_set_fault_enable(pte);
687 pasid_set_page_snoop(pte, !!ecap_smpwc(iommu->ecap));
690 * We should set SRE bit as well since the addresses are expected
694 pasid_set_present(pte);
695 pasid_flush_caches(iommu, pte, pasid, did);
701 * Set the page snoop control for a pasid entry which has been set up.
703 void intel_pasid_setup_page_snoop_control(struct intel_iommu *iommu,
704 struct device *dev, u32 pasid)
706 struct pasid_entry *pte;
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);
716 pasid_set_pgsnp(pte);
717 did = pasid_get_domain_id(pte);
718 spin_unlock(&iommu->lock);
720 if (!ecap_coherent(iommu->ecap))
721 clflush_cache_range(pte, sizeof(*pte));
724 * VT-d spec 3.4 table23 states guides for cache invalidation:
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
731 * - PASID-based Device-TLB invalidation (with S=1 and
732 * Addr[63:12]=0x7FFFFFFF_FFFFF) to affected functions
734 pasid_cache_invalidation_with_pasid(iommu, did, pasid);
735 qi_flush_piotlb(iommu, did, pasid, 0, -1, 0);
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);