1 /* SPDX-License-Identifier: GPL-2.0-only */
3 * Kernel-based Virtual Machine driver for Linux
5 * This module enables machines with Intel VT-x extensions to run virtual
6 * machines without emulation or binary translation.
10 * Copyright (C) 2006 Qumranet, Inc.
11 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
19 * The MMU needs to be able to access/walk 32-bit and 64-bit guest page tables,
20 * as well as guest EPT tables, so the code in this file is compiled thrice,
21 * once per guest PTE type. The per-type defines are #undef'd at the end.
25 #define pt_element_t u64
26 #define guest_walker guest_walker64
27 #define FNAME(name) paging##64_##name
28 #define PT_LEVEL_BITS 9
29 #define PT_GUEST_DIRTY_SHIFT PT_DIRTY_SHIFT
30 #define PT_GUEST_ACCESSED_SHIFT PT_ACCESSED_SHIFT
31 #define PT_HAVE_ACCESSED_DIRTY(mmu) true
33 #define PT_MAX_FULL_LEVELS PT64_ROOT_MAX_LEVEL
35 #define PT_MAX_FULL_LEVELS 2
38 #define pt_element_t u32
39 #define guest_walker guest_walker32
40 #define FNAME(name) paging##32_##name
41 #define PT_LEVEL_BITS 10
42 #define PT_MAX_FULL_LEVELS 2
43 #define PT_GUEST_DIRTY_SHIFT PT_DIRTY_SHIFT
44 #define PT_GUEST_ACCESSED_SHIFT PT_ACCESSED_SHIFT
45 #define PT_HAVE_ACCESSED_DIRTY(mmu) true
47 #define PT32_DIR_PSE36_SIZE 4
48 #define PT32_DIR_PSE36_SHIFT 13
49 #define PT32_DIR_PSE36_MASK \
50 (((1ULL << PT32_DIR_PSE36_SIZE) - 1) << PT32_DIR_PSE36_SHIFT)
51 #elif PTTYPE == PTTYPE_EPT
52 #define pt_element_t u64
53 #define guest_walker guest_walkerEPT
54 #define FNAME(name) ept_##name
55 #define PT_LEVEL_BITS 9
56 #define PT_GUEST_DIRTY_SHIFT 9
57 #define PT_GUEST_ACCESSED_SHIFT 8
58 #define PT_HAVE_ACCESSED_DIRTY(mmu) (!(mmu)->cpu_role.base.ad_disabled)
59 #define PT_MAX_FULL_LEVELS PT64_ROOT_MAX_LEVEL
61 #error Invalid PTTYPE value
64 /* Common logic, but per-type values. These also need to be undefined. */
65 #define PT_BASE_ADDR_MASK ((pt_element_t)(((1ULL << 52) - 1) & ~(u64)(PAGE_SIZE-1)))
66 #define PT_LVL_ADDR_MASK(lvl) __PT_LVL_ADDR_MASK(PT_BASE_ADDR_MASK, lvl, PT_LEVEL_BITS)
67 #define PT_LVL_OFFSET_MASK(lvl) __PT_LVL_OFFSET_MASK(PT_BASE_ADDR_MASK, lvl, PT_LEVEL_BITS)
68 #define PT_INDEX(addr, lvl) __PT_INDEX(addr, lvl, PT_LEVEL_BITS)
70 #define PT_GUEST_DIRTY_MASK (1 << PT_GUEST_DIRTY_SHIFT)
71 #define PT_GUEST_ACCESSED_MASK (1 << PT_GUEST_ACCESSED_SHIFT)
73 #define gpte_to_gfn_lvl FNAME(gpte_to_gfn_lvl)
74 #define gpte_to_gfn(pte) gpte_to_gfn_lvl((pte), PG_LEVEL_4K)
77 * The guest_walker structure emulates the behavior of the hardware page
83 gfn_t table_gfn[PT_MAX_FULL_LEVELS];
84 pt_element_t ptes[PT_MAX_FULL_LEVELS];
85 pt_element_t prefetch_ptes[PTE_PREFETCH_NUM];
86 gpa_t pte_gpa[PT_MAX_FULL_LEVELS];
87 pt_element_t __user *ptep_user[PT_MAX_FULL_LEVELS];
88 bool pte_writable[PT_MAX_FULL_LEVELS];
89 unsigned int pt_access[PT_MAX_FULL_LEVELS];
90 unsigned int pte_access;
92 struct x86_exception fault;
96 static inline gfn_t pse36_gfn_delta(u32 gpte)
98 int shift = 32 - PT32_DIR_PSE36_SHIFT - PAGE_SHIFT;
100 return (gpte & PT32_DIR_PSE36_MASK) << shift;
104 static gfn_t gpte_to_gfn_lvl(pt_element_t gpte, int lvl)
106 return (gpte & PT_LVL_ADDR_MASK(lvl)) >> PAGE_SHIFT;
109 static inline void FNAME(protect_clean_gpte)(struct kvm_mmu *mmu, unsigned *access,
114 /* dirty bit is not supported, so no need to track it */
115 if (!PT_HAVE_ACCESSED_DIRTY(mmu))
118 BUILD_BUG_ON(PT_WRITABLE_MASK != ACC_WRITE_MASK);
120 mask = (unsigned)~ACC_WRITE_MASK;
121 /* Allow write access to dirty gptes */
122 mask |= (gpte >> (PT_GUEST_DIRTY_SHIFT - PT_WRITABLE_SHIFT)) &
127 static inline int FNAME(is_present_gpte)(unsigned long pte)
129 #if PTTYPE != PTTYPE_EPT
130 return pte & PT_PRESENT_MASK;
136 static bool FNAME(is_bad_mt_xwr)(struct rsvd_bits_validate *rsvd_check, u64 gpte)
138 #if PTTYPE != PTTYPE_EPT
141 return __is_bad_mt_xwr(rsvd_check, gpte);
145 static bool FNAME(is_rsvd_bits_set)(struct kvm_mmu *mmu, u64 gpte, int level)
147 return __is_rsvd_bits_set(&mmu->guest_rsvd_check, gpte, level) ||
148 FNAME(is_bad_mt_xwr)(&mmu->guest_rsvd_check, gpte);
151 static bool FNAME(prefetch_invalid_gpte)(struct kvm_vcpu *vcpu,
152 struct kvm_mmu_page *sp, u64 *spte,
155 if (!FNAME(is_present_gpte)(gpte))
158 /* Prefetch only accessed entries (unless A/D bits are disabled). */
159 if (PT_HAVE_ACCESSED_DIRTY(vcpu->arch.mmu) &&
160 !(gpte & PT_GUEST_ACCESSED_MASK))
163 if (FNAME(is_rsvd_bits_set)(vcpu->arch.mmu, gpte, PG_LEVEL_4K))
169 drop_spte(vcpu->kvm, spte);
174 * For PTTYPE_EPT, a page table can be executable but not readable
175 * on supported processors. Therefore, set_spte does not automatically
176 * set bit 0 if execute only is supported. Here, we repurpose ACC_USER_MASK
177 * to signify readability since it isn't used in the EPT case
179 static inline unsigned FNAME(gpte_access)(u64 gpte)
182 #if PTTYPE == PTTYPE_EPT
183 access = ((gpte & VMX_EPT_WRITABLE_MASK) ? ACC_WRITE_MASK : 0) |
184 ((gpte & VMX_EPT_EXECUTABLE_MASK) ? ACC_EXEC_MASK : 0) |
185 ((gpte & VMX_EPT_READABLE_MASK) ? ACC_USER_MASK : 0);
187 BUILD_BUG_ON(ACC_EXEC_MASK != PT_PRESENT_MASK);
188 BUILD_BUG_ON(ACC_EXEC_MASK != 1);
189 access = gpte & (PT_WRITABLE_MASK | PT_USER_MASK | PT_PRESENT_MASK);
190 /* Combine NX with P (which is set here) to get ACC_EXEC_MASK. */
191 access ^= (gpte >> PT64_NX_SHIFT);
197 static int FNAME(update_accessed_dirty_bits)(struct kvm_vcpu *vcpu,
199 struct guest_walker *walker,
200 gpa_t addr, int write_fault)
202 unsigned level, index;
203 pt_element_t pte, orig_pte;
204 pt_element_t __user *ptep_user;
208 /* dirty/accessed bits are not supported, so no need to update them */
209 if (!PT_HAVE_ACCESSED_DIRTY(mmu))
212 for (level = walker->max_level; level >= walker->level; --level) {
213 pte = orig_pte = walker->ptes[level - 1];
214 table_gfn = walker->table_gfn[level - 1];
215 ptep_user = walker->ptep_user[level - 1];
216 index = offset_in_page(ptep_user) / sizeof(pt_element_t);
217 if (!(pte & PT_GUEST_ACCESSED_MASK)) {
218 trace_kvm_mmu_set_accessed_bit(table_gfn, index, sizeof(pte));
219 pte |= PT_GUEST_ACCESSED_MASK;
221 if (level == walker->level && write_fault &&
222 !(pte & PT_GUEST_DIRTY_MASK)) {
223 trace_kvm_mmu_set_dirty_bit(table_gfn, index, sizeof(pte));
224 #if PTTYPE == PTTYPE_EPT
225 if (kvm_x86_ops.nested_ops->write_log_dirty(vcpu, addr))
228 pte |= PT_GUEST_DIRTY_MASK;
234 * If the slot is read-only, simply do not process the accessed
235 * and dirty bits. This is the correct thing to do if the slot
236 * is ROM, and page tables in read-as-ROM/write-as-MMIO slots
237 * are only supported if the accessed and dirty bits are already
238 * set in the ROM (so that MMIO writes are never needed).
240 * Note that NPT does not allow this at all and faults, since
241 * it always wants nested page table entries for the guest
242 * page tables to be writable. And EPT works but will simply
243 * overwrite the read-only memory to set the accessed and dirty
246 if (unlikely(!walker->pte_writable[level - 1]))
249 ret = __try_cmpxchg_user(ptep_user, &orig_pte, pte, fault);
253 kvm_vcpu_mark_page_dirty(vcpu, table_gfn);
254 walker->ptes[level - 1] = pte;
259 static inline unsigned FNAME(gpte_pkeys)(struct kvm_vcpu *vcpu, u64 gpte)
263 pte_t pte = {.pte = gpte};
265 pkeys = pte_flags_pkey(pte_flags(pte));
270 static inline bool FNAME(is_last_gpte)(struct kvm_mmu *mmu,
271 unsigned int level, unsigned int gpte)
274 * For EPT and PAE paging (both variants), bit 7 is either reserved at
275 * all level or indicates a huge page (ignoring CR3/EPTP). In either
276 * case, bit 7 being set terminates the walk.
280 * 32-bit paging requires special handling because bit 7 is ignored if
281 * CR4.PSE=0, not reserved. Clear bit 7 in the gpte if the level is
282 * greater than the last level for which bit 7 is the PAGE_SIZE bit.
284 * The RHS has bit 7 set iff level < (2 + PSE). If it is clear, bit 7
285 * is not reserved and does not indicate a large page at this level,
286 * so clear PT_PAGE_SIZE_MASK in gpte if that is the case.
288 gpte &= level - (PT32_ROOT_LEVEL + mmu->cpu_role.ext.cr4_pse);
291 * PG_LEVEL_4K always terminates. The RHS has bit 7 set
292 * iff level <= PG_LEVEL_4K, which for our purpose means
293 * level == PG_LEVEL_4K; set PT_PAGE_SIZE_MASK in gpte then.
295 gpte |= level - PG_LEVEL_4K - 1;
297 return gpte & PT_PAGE_SIZE_MASK;
300 * Fetch a guest pte for a guest virtual address, or for an L2's GPA.
302 static int FNAME(walk_addr_generic)(struct guest_walker *walker,
303 struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
304 gpa_t addr, u64 access)
308 pt_element_t __user *ptep_user;
310 u64 pt_access, pte_access;
311 unsigned index, accessed_dirty, pte_pkey;
316 u64 walk_nx_mask = 0;
317 const int write_fault = access & PFERR_WRITE_MASK;
318 const int user_fault = access & PFERR_USER_MASK;
319 const int fetch_fault = access & PFERR_FETCH_MASK;
324 trace_kvm_mmu_pagetable_walk(addr, access);
326 walker->level = mmu->cpu_role.base.level;
327 pte = kvm_mmu_get_guest_pgd(vcpu, mmu);
328 have_ad = PT_HAVE_ACCESSED_DIRTY(mmu);
331 walk_nx_mask = 1ULL << PT64_NX_SHIFT;
332 if (walker->level == PT32E_ROOT_LEVEL) {
333 pte = mmu->get_pdptr(vcpu, (addr >> 30) & 3);
334 trace_kvm_mmu_paging_element(pte, walker->level);
335 if (!FNAME(is_present_gpte)(pte))
340 walker->max_level = walker->level;
341 ASSERT(!(is_long_mode(vcpu) && !is_pae(vcpu)));
344 * FIXME: on Intel processors, loads of the PDPTE registers for PAE paging
345 * by the MOV to CR instruction are treated as reads and do not cause the
346 * processor to set the dirty flag in any EPT paging-structure entry.
348 nested_access = (have_ad ? PFERR_WRITE_MASK : 0) | PFERR_USER_MASK;
354 unsigned long host_addr;
356 pt_access = pte_access;
359 index = PT_INDEX(addr, walker->level);
360 table_gfn = gpte_to_gfn(pte);
361 offset = index * sizeof(pt_element_t);
362 pte_gpa = gfn_to_gpa(table_gfn) + offset;
364 BUG_ON(walker->level < 1);
365 walker->table_gfn[walker->level - 1] = table_gfn;
366 walker->pte_gpa[walker->level - 1] = pte_gpa;
368 real_gpa = kvm_translate_gpa(vcpu, mmu, gfn_to_gpa(table_gfn),
369 nested_access, &walker->fault);
372 * FIXME: This can happen if emulation (for of an INS/OUTS
373 * instruction) triggers a nested page fault. The exit
374 * qualification / exit info field will incorrectly have
375 * "guest page access" as the nested page fault's cause,
376 * instead of "guest page structure access". To fix this,
377 * the x86_exception struct should be augmented with enough
378 * information to fix the exit_qualification or exit_info_1
381 if (unlikely(real_gpa == INVALID_GPA))
384 host_addr = kvm_vcpu_gfn_to_hva_prot(vcpu, gpa_to_gfn(real_gpa),
385 &walker->pte_writable[walker->level - 1]);
386 if (unlikely(kvm_is_error_hva(host_addr)))
389 ptep_user = (pt_element_t __user *)((void *)host_addr + offset);
390 if (unlikely(__get_user(pte, ptep_user)))
392 walker->ptep_user[walker->level - 1] = ptep_user;
394 trace_kvm_mmu_paging_element(pte, walker->level);
397 * Inverting the NX it lets us AND it like other
400 pte_access = pt_access & (pte ^ walk_nx_mask);
402 if (unlikely(!FNAME(is_present_gpte)(pte)))
405 if (unlikely(FNAME(is_rsvd_bits_set)(mmu, pte, walker->level))) {
406 errcode = PFERR_RSVD_MASK | PFERR_PRESENT_MASK;
410 walker->ptes[walker->level - 1] = pte;
412 /* Convert to ACC_*_MASK flags for struct guest_walker. */
413 walker->pt_access[walker->level - 1] = FNAME(gpte_access)(pt_access ^ walk_nx_mask);
414 } while (!FNAME(is_last_gpte)(mmu, walker->level, pte));
416 pte_pkey = FNAME(gpte_pkeys)(vcpu, pte);
417 accessed_dirty = have_ad ? pte_access & PT_GUEST_ACCESSED_MASK : 0;
419 /* Convert to ACC_*_MASK flags for struct guest_walker. */
420 walker->pte_access = FNAME(gpte_access)(pte_access ^ walk_nx_mask);
421 errcode = permission_fault(vcpu, mmu, walker->pte_access, pte_pkey, access);
422 if (unlikely(errcode))
425 gfn = gpte_to_gfn_lvl(pte, walker->level);
426 gfn += (addr & PT_LVL_OFFSET_MASK(walker->level)) >> PAGE_SHIFT;
429 if (walker->level > PG_LEVEL_4K && is_cpuid_PSE36())
430 gfn += pse36_gfn_delta(pte);
433 real_gpa = kvm_translate_gpa(vcpu, mmu, gfn_to_gpa(gfn), access, &walker->fault);
434 if (real_gpa == INVALID_GPA)
437 walker->gfn = real_gpa >> PAGE_SHIFT;
440 FNAME(protect_clean_gpte)(mmu, &walker->pte_access, pte);
443 * On a write fault, fold the dirty bit into accessed_dirty.
444 * For modes without A/D bits support accessed_dirty will be
447 accessed_dirty &= pte >>
448 (PT_GUEST_DIRTY_SHIFT - PT_GUEST_ACCESSED_SHIFT);
450 if (unlikely(!accessed_dirty)) {
451 ret = FNAME(update_accessed_dirty_bits)(vcpu, mmu, walker,
453 if (unlikely(ret < 0))
459 pgprintk("%s: pte %llx pte_access %x pt_access %x\n",
460 __func__, (u64)pte, walker->pte_access,
461 walker->pt_access[walker->level - 1]);
465 errcode |= write_fault | user_fault;
466 if (fetch_fault && (is_efer_nx(mmu) || is_cr4_smep(mmu)))
467 errcode |= PFERR_FETCH_MASK;
469 walker->fault.vector = PF_VECTOR;
470 walker->fault.error_code_valid = true;
471 walker->fault.error_code = errcode;
473 #if PTTYPE == PTTYPE_EPT
475 * Use PFERR_RSVD_MASK in error_code to tell if EPT
476 * misconfiguration requires to be injected. The detection is
477 * done by is_rsvd_bits_set() above.
479 * We set up the value of exit_qualification to inject:
480 * [2:0] - Derive from the access bits. The exit_qualification might be
481 * out of date if it is serving an EPT misconfiguration.
482 * [5:3] - Calculated by the page walk of the guest EPT page tables
483 * [7:8] - Derived from [7:8] of real exit_qualification
485 * The other bits are set to 0.
487 if (!(errcode & PFERR_RSVD_MASK)) {
488 vcpu->arch.exit_qualification &= (EPT_VIOLATION_GVA_IS_VALID |
489 EPT_VIOLATION_GVA_TRANSLATED);
491 vcpu->arch.exit_qualification |= EPT_VIOLATION_ACC_WRITE;
493 vcpu->arch.exit_qualification |= EPT_VIOLATION_ACC_READ;
495 vcpu->arch.exit_qualification |= EPT_VIOLATION_ACC_INSTR;
498 * Note, pte_access holds the raw RWX bits from the EPTE, not
501 vcpu->arch.exit_qualification |= (pte_access & VMX_EPT_RWX_MASK) <<
502 EPT_VIOLATION_RWX_SHIFT;
505 walker->fault.address = addr;
506 walker->fault.nested_page_fault = mmu != vcpu->arch.walk_mmu;
507 walker->fault.async_page_fault = false;
509 trace_kvm_mmu_walker_error(walker->fault.error_code);
513 static int FNAME(walk_addr)(struct guest_walker *walker,
514 struct kvm_vcpu *vcpu, gpa_t addr, u64 access)
516 return FNAME(walk_addr_generic)(walker, vcpu, vcpu->arch.mmu, addr,
521 FNAME(prefetch_gpte)(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp,
522 u64 *spte, pt_element_t gpte)
524 struct kvm_memory_slot *slot;
529 if (FNAME(prefetch_invalid_gpte)(vcpu, sp, spte, gpte))
532 pgprintk("%s: gpte %llx spte %p\n", __func__, (u64)gpte, spte);
534 gfn = gpte_to_gfn(gpte);
535 pte_access = sp->role.access & FNAME(gpte_access)(gpte);
536 FNAME(protect_clean_gpte)(vcpu->arch.mmu, &pte_access, gpte);
538 slot = gfn_to_memslot_dirty_bitmap(vcpu, gfn, pte_access & ACC_WRITE_MASK);
542 pfn = gfn_to_pfn_memslot_atomic(slot, gfn);
543 if (is_error_pfn(pfn))
546 mmu_set_spte(vcpu, slot, spte, pte_access, gfn, pfn, NULL);
547 kvm_release_pfn_clean(pfn);
551 static bool FNAME(gpte_changed)(struct kvm_vcpu *vcpu,
552 struct guest_walker *gw, int level)
554 pt_element_t curr_pte;
555 gpa_t base_gpa, pte_gpa = gw->pte_gpa[level - 1];
559 if (level == PG_LEVEL_4K) {
560 mask = PTE_PREFETCH_NUM * sizeof(pt_element_t) - 1;
561 base_gpa = pte_gpa & ~mask;
562 index = (pte_gpa - base_gpa) / sizeof(pt_element_t);
564 r = kvm_vcpu_read_guest_atomic(vcpu, base_gpa,
565 gw->prefetch_ptes, sizeof(gw->prefetch_ptes));
566 curr_pte = gw->prefetch_ptes[index];
568 r = kvm_vcpu_read_guest_atomic(vcpu, pte_gpa,
569 &curr_pte, sizeof(curr_pte));
571 return r || curr_pte != gw->ptes[level - 1];
574 static void FNAME(pte_prefetch)(struct kvm_vcpu *vcpu, struct guest_walker *gw,
577 struct kvm_mmu_page *sp;
578 pt_element_t *gptep = gw->prefetch_ptes;
582 sp = sptep_to_sp(sptep);
584 if (sp->role.level > PG_LEVEL_4K)
588 * If addresses are being invalidated, skip prefetching to avoid
589 * accidentally prefetching those addresses.
591 if (unlikely(vcpu->kvm->mmu_invalidate_in_progress))
595 return __direct_pte_prefetch(vcpu, sp, sptep);
597 i = spte_index(sptep) & ~(PTE_PREFETCH_NUM - 1);
600 for (i = 0; i < PTE_PREFETCH_NUM; i++, spte++) {
604 if (is_shadow_present_pte(*spte))
607 if (!FNAME(prefetch_gpte)(vcpu, sp, spte, gptep[i]))
613 * Fetch a shadow pte for a specific level in the paging hierarchy.
614 * If the guest tries to write a write-protected page, we need to
615 * emulate this operation, return 1 to indicate this case.
617 static int FNAME(fetch)(struct kvm_vcpu *vcpu, struct kvm_page_fault *fault,
618 struct guest_walker *gw)
620 struct kvm_mmu_page *sp = NULL;
621 struct kvm_shadow_walk_iterator it;
622 unsigned int direct_access, access;
624 gfn_t base_gfn = fault->gfn;
626 WARN_ON_ONCE(gw->gfn != base_gfn);
627 direct_access = gw->pte_access;
629 top_level = vcpu->arch.mmu->cpu_role.base.level;
630 if (top_level == PT32E_ROOT_LEVEL)
631 top_level = PT32_ROOT_LEVEL;
633 * Verify that the top-level gpte is still there. Since the page
634 * is a root page, it is either write protected (and cannot be
635 * changed from now on) or it is invalid (in which case, we don't
636 * really care if it changes underneath us after this point).
638 if (FNAME(gpte_changed)(vcpu, gw, top_level))
639 goto out_gpte_changed;
641 if (WARN_ON(!VALID_PAGE(vcpu->arch.mmu->root.hpa)))
642 goto out_gpte_changed;
644 for_each_shadow_entry(vcpu, fault->addr, it) {
647 clear_sp_write_flooding_count(it.sptep);
648 if (it.level == gw->level)
651 table_gfn = gw->table_gfn[it.level - 2];
652 access = gw->pt_access[it.level - 2];
653 sp = kvm_mmu_get_child_sp(vcpu, it.sptep, table_gfn,
656 if (sp != ERR_PTR(-EEXIST)) {
658 * We must synchronize the pagetable before linking it
659 * because the guest doesn't need to flush tlb when
660 * the gpte is changed from non-present to present.
661 * Otherwise, the guest may use the wrong mapping.
663 * For PG_LEVEL_4K, kvm_mmu_get_page() has already
664 * synchronized it transiently via kvm_sync_page().
666 * For higher level pagetable, we synchronize it via
667 * the slower mmu_sync_children(). If it needs to
668 * break, some progress has been made; return
669 * RET_PF_RETRY and retry on the next #PF.
670 * KVM_REQ_MMU_SYNC is not necessary but it
671 * expedites the process.
673 if (sp->unsync_children &&
674 mmu_sync_children(vcpu, sp, false))
679 * Verify that the gpte in the page we've just write
680 * protected is still there.
682 if (FNAME(gpte_changed)(vcpu, gw, it.level - 1))
683 goto out_gpte_changed;
685 if (sp != ERR_PTR(-EEXIST))
686 link_shadow_page(vcpu, it.sptep, sp);
688 if (fault->write && table_gfn == fault->gfn)
689 fault->write_fault_to_shadow_pgtable = true;
693 * Adjust the hugepage size _after_ resolving indirect shadow pages.
694 * KVM doesn't support mapping hugepages into the guest for gfns that
695 * are being shadowed by KVM, i.e. allocating a new shadow page may
696 * affect the allowed hugepage size.
698 kvm_mmu_hugepage_adjust(vcpu, fault);
700 trace_kvm_mmu_spte_requested(fault);
702 for (; shadow_walk_okay(&it); shadow_walk_next(&it)) {
704 * We cannot overwrite existing page tables with an NX
705 * large page, as the leaf could be executable.
707 if (fault->nx_huge_page_workaround_enabled)
708 disallowed_hugepage_adjust(fault, *it.sptep, it.level);
710 base_gfn = gfn_round_for_level(fault->gfn, it.level);
711 if (it.level == fault->goal_level)
714 validate_direct_spte(vcpu, it.sptep, direct_access);
716 sp = kvm_mmu_get_child_sp(vcpu, it.sptep, base_gfn,
717 true, direct_access);
718 if (sp == ERR_PTR(-EEXIST))
721 link_shadow_page(vcpu, it.sptep, sp);
722 if (fault->huge_page_disallowed)
723 account_nx_huge_page(vcpu->kvm, sp,
724 fault->req_level >= it.level);
727 if (WARN_ON_ONCE(it.level != fault->goal_level))
730 ret = mmu_set_spte(vcpu, fault->slot, it.sptep, gw->pte_access,
731 base_gfn, fault->pfn, fault);
732 if (ret == RET_PF_SPURIOUS)
735 FNAME(pte_prefetch)(vcpu, gw, it.sptep);
743 * Page fault handler. There are several causes for a page fault:
744 * - there is no shadow pte for the guest pte
745 * - write access through a shadow pte marked read only so that we can set
747 * - write access to a shadow pte marked read only so we can update the page
748 * dirty bitmap, when userspace requests it
749 * - mmio access; in this case we will never install a present shadow pte
750 * - normal guest page fault due to the guest pte marked not present, not
751 * writable, or not executable
753 * Returns: 1 if we need to emulate the instruction, 0 otherwise, or
754 * a negative value on error.
756 static int FNAME(page_fault)(struct kvm_vcpu *vcpu, struct kvm_page_fault *fault)
758 struct guest_walker walker;
761 pgprintk("%s: addr %lx err %x\n", __func__, fault->addr, fault->error_code);
762 WARN_ON_ONCE(fault->is_tdp);
765 * Look up the guest pte for the faulting address.
766 * If PFEC.RSVD is set, this is a shadow page fault.
767 * The bit needs to be cleared before walking guest page tables.
769 r = FNAME(walk_addr)(&walker, vcpu, fault->addr,
770 fault->error_code & ~PFERR_RSVD_MASK);
773 * The page is not mapped by the guest. Let the guest handle it.
776 pgprintk("%s: guest page fault\n", __func__);
777 if (!fault->prefetch)
778 kvm_inject_emulated_page_fault(vcpu, &walker.fault);
783 fault->gfn = walker.gfn;
784 fault->max_level = walker.level;
785 fault->slot = kvm_vcpu_gfn_to_memslot(vcpu, fault->gfn);
787 if (page_fault_handle_page_track(vcpu, fault)) {
788 shadow_page_table_clear_flood(vcpu, fault->addr);
789 return RET_PF_EMULATE;
792 r = mmu_topup_memory_caches(vcpu, true);
796 r = kvm_faultin_pfn(vcpu, fault, walker.pte_access);
797 if (r != RET_PF_CONTINUE)
801 * Do not change pte_access if the pfn is a mmio page, otherwise
802 * we will cache the incorrect access into mmio spte.
804 if (fault->write && !(walker.pte_access & ACC_WRITE_MASK) &&
805 !is_cr0_wp(vcpu->arch.mmu) && !fault->user && fault->slot) {
806 walker.pte_access |= ACC_WRITE_MASK;
807 walker.pte_access &= ~ACC_USER_MASK;
810 * If we converted a user page to a kernel page,
811 * so that the kernel can write to it when cr0.wp=0,
812 * then we should prevent the kernel from executing it
813 * if SMEP is enabled.
815 if (is_cr4_smep(vcpu->arch.mmu))
816 walker.pte_access &= ~ACC_EXEC_MASK;
820 write_lock(&vcpu->kvm->mmu_lock);
822 if (is_page_fault_stale(vcpu, fault))
825 r = make_mmu_pages_available(vcpu);
828 r = FNAME(fetch)(vcpu, fault, &walker);
831 write_unlock(&vcpu->kvm->mmu_lock);
832 kvm_release_pfn_clean(fault->pfn);
836 static gpa_t FNAME(get_level1_sp_gpa)(struct kvm_mmu_page *sp)
840 WARN_ON(sp->role.level != PG_LEVEL_4K);
843 offset = sp->role.quadrant << SPTE_LEVEL_BITS;
845 return gfn_to_gpa(sp->gfn) + offset * sizeof(pt_element_t);
848 /* Note, @addr is a GPA when gva_to_gpa() translates an L2 GPA to an L1 GPA. */
849 static gpa_t FNAME(gva_to_gpa)(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
850 gpa_t addr, u64 access,
851 struct x86_exception *exception)
853 struct guest_walker walker;
854 gpa_t gpa = INVALID_GPA;
857 #ifndef CONFIG_X86_64
858 /* A 64-bit GVA should be impossible on 32-bit KVM. */
859 WARN_ON_ONCE((addr >> 32) && mmu == vcpu->arch.walk_mmu);
862 r = FNAME(walk_addr_generic)(&walker, vcpu, mmu, addr, access);
865 gpa = gfn_to_gpa(walker.gfn);
866 gpa |= addr & ~PAGE_MASK;
867 } else if (exception)
868 *exception = walker.fault;
874 * Using the information in sp->shadowed_translation (kvm_mmu_page_get_gfn()) is
876 * - The spte has a reference to the struct page, so the pfn for a given gfn
877 * can't change unless all sptes pointing to it are nuked first.
880 * < 0: failed to sync spte
881 * 0: the spte is synced and no tlb flushing is required
882 * > 0: the spte is synced and tlb flushing is required
884 static int FNAME(sync_spte)(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp, int i)
889 struct kvm_memory_slot *slot;
895 if (WARN_ON_ONCE(!sp->spt[i]))
898 first_pte_gpa = FNAME(get_level1_sp_gpa)(sp);
899 pte_gpa = first_pte_gpa + i * sizeof(pt_element_t);
901 if (kvm_vcpu_read_guest_atomic(vcpu, pte_gpa, &gpte,
902 sizeof(pt_element_t)))
905 if (FNAME(prefetch_invalid_gpte)(vcpu, sp, &sp->spt[i], gpte))
908 gfn = gpte_to_gfn(gpte);
909 pte_access = sp->role.access;
910 pte_access &= FNAME(gpte_access)(gpte);
911 FNAME(protect_clean_gpte)(vcpu->arch.mmu, &pte_access, gpte);
913 if (sync_mmio_spte(vcpu, &sp->spt[i], gfn, pte_access))
917 * Drop the SPTE if the new protections would result in a RWX=0
918 * SPTE or if the gfn is changing. The RWX=0 case only affects
919 * EPT with execute-only support, i.e. EPT without an effective
920 * "present" bit, as all other paging modes will create a
921 * read-only SPTE if pte_access is zero.
923 if ((!pte_access && !shadow_present_mask) ||
924 gfn != kvm_mmu_page_get_gfn(sp, i)) {
925 drop_spte(vcpu->kvm, &sp->spt[i]);
929 * Do nothing if the permissions are unchanged. The existing SPTE is
930 * still, and prefetch_invalid_gpte() has verified that the A/D bits
931 * are set in the "new" gPTE, i.e. there is no danger of missing an A/D
932 * update due to A/D bits being set in the SPTE but not the gPTE.
934 if (kvm_mmu_page_get_access(sp, i) == pte_access)
937 /* Update the shadowed access bits in case they changed. */
938 kvm_mmu_page_set_access(sp, i, pte_access);
942 host_writable = spte & shadow_host_writable_mask;
943 slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
944 make_spte(vcpu, sp, slot, pte_access, gfn,
945 spte_to_pfn(spte), spte, true, false,
946 host_writable, &spte);
948 return mmu_spte_update(sptep, spte);
954 #undef PT_BASE_ADDR_MASK
956 #undef PT_LVL_ADDR_MASK
957 #undef PT_LVL_OFFSET_MASK
959 #undef PT_MAX_FULL_LEVELS
961 #undef gpte_to_gfn_lvl
962 #undef PT_GUEST_ACCESSED_MASK
963 #undef PT_GUEST_DIRTY_MASK
964 #undef PT_GUEST_DIRTY_SHIFT
965 #undef PT_GUEST_ACCESSED_SHIFT
966 #undef PT_HAVE_ACCESSED_DIRTY