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.
14 * Yaniv Kamay <yaniv@qumranet.com>
15 * Avi Kivity <avi@qumranet.com>
19 * We need the mmu code to access both 32-bit and 64-bit guest ptes,
20 * so the code in this file is compiled twice, once per pte size.
24 #define pt_element_t u64
25 #define guest_walker guest_walker64
26 #define FNAME(name) paging##64_##name
27 #define PT_BASE_ADDR_MASK GUEST_PT64_BASE_ADDR_MASK
28 #define PT_LVL_ADDR_MASK(lvl) PT64_LVL_ADDR_MASK(lvl)
29 #define PT_LVL_OFFSET_MASK(lvl) PT64_LVL_OFFSET_MASK(lvl)
30 #define PT_INDEX(addr, level) PT64_INDEX(addr, level)
31 #define PT_LEVEL_BITS PT64_LEVEL_BITS
32 #define PT_GUEST_DIRTY_SHIFT PT_DIRTY_SHIFT
33 #define PT_GUEST_ACCESSED_SHIFT PT_ACCESSED_SHIFT
34 #define PT_HAVE_ACCESSED_DIRTY(mmu) true
36 #define PT_MAX_FULL_LEVELS PT64_ROOT_MAX_LEVEL
37 #define CMPXCHG "cmpxchgq"
39 #define PT_MAX_FULL_LEVELS 2
42 #define pt_element_t u32
43 #define guest_walker guest_walker32
44 #define FNAME(name) paging##32_##name
45 #define PT_BASE_ADDR_MASK PT32_BASE_ADDR_MASK
46 #define PT_LVL_ADDR_MASK(lvl) PT32_LVL_ADDR_MASK(lvl)
47 #define PT_LVL_OFFSET_MASK(lvl) PT32_LVL_OFFSET_MASK(lvl)
48 #define PT_INDEX(addr, level) PT32_INDEX(addr, level)
49 #define PT_LEVEL_BITS PT32_LEVEL_BITS
50 #define PT_MAX_FULL_LEVELS 2
51 #define PT_GUEST_DIRTY_SHIFT PT_DIRTY_SHIFT
52 #define PT_GUEST_ACCESSED_SHIFT PT_ACCESSED_SHIFT
53 #define PT_HAVE_ACCESSED_DIRTY(mmu) true
54 #define CMPXCHG "cmpxchgl"
55 #elif PTTYPE == PTTYPE_EPT
56 #define pt_element_t u64
57 #define guest_walker guest_walkerEPT
58 #define FNAME(name) ept_##name
59 #define PT_BASE_ADDR_MASK GUEST_PT64_BASE_ADDR_MASK
60 #define PT_LVL_ADDR_MASK(lvl) PT64_LVL_ADDR_MASK(lvl)
61 #define PT_LVL_OFFSET_MASK(lvl) PT64_LVL_OFFSET_MASK(lvl)
62 #define PT_INDEX(addr, level) PT64_INDEX(addr, level)
63 #define PT_LEVEL_BITS PT64_LEVEL_BITS
64 #define PT_GUEST_DIRTY_SHIFT 9
65 #define PT_GUEST_ACCESSED_SHIFT 8
66 #define PT_HAVE_ACCESSED_DIRTY(mmu) ((mmu)->ept_ad)
68 #define CMPXCHG "cmpxchgq"
70 #define PT_MAX_FULL_LEVELS PT64_ROOT_MAX_LEVEL
72 #error Invalid PTTYPE value
75 #define PT_GUEST_DIRTY_MASK (1 << PT_GUEST_DIRTY_SHIFT)
76 #define PT_GUEST_ACCESSED_MASK (1 << PT_GUEST_ACCESSED_SHIFT)
78 #define gpte_to_gfn_lvl FNAME(gpte_to_gfn_lvl)
79 #define gpte_to_gfn(pte) gpte_to_gfn_lvl((pte), PG_LEVEL_4K)
82 * The guest_walker structure emulates the behavior of the hardware page
88 gfn_t table_gfn[PT_MAX_FULL_LEVELS];
89 pt_element_t ptes[PT_MAX_FULL_LEVELS];
90 pt_element_t prefetch_ptes[PTE_PREFETCH_NUM];
91 gpa_t pte_gpa[PT_MAX_FULL_LEVELS];
92 pt_element_t __user *ptep_user[PT_MAX_FULL_LEVELS];
93 bool pte_writable[PT_MAX_FULL_LEVELS];
94 unsigned int pt_access[PT_MAX_FULL_LEVELS];
95 unsigned int pte_access;
97 struct x86_exception fault;
100 static gfn_t gpte_to_gfn_lvl(pt_element_t gpte, int lvl)
102 return (gpte & PT_LVL_ADDR_MASK(lvl)) >> PAGE_SHIFT;
105 static inline void FNAME(protect_clean_gpte)(struct kvm_mmu *mmu, unsigned *access,
110 /* dirty bit is not supported, so no need to track it */
111 if (!PT_HAVE_ACCESSED_DIRTY(mmu))
114 BUILD_BUG_ON(PT_WRITABLE_MASK != ACC_WRITE_MASK);
116 mask = (unsigned)~ACC_WRITE_MASK;
117 /* Allow write access to dirty gptes */
118 mask |= (gpte >> (PT_GUEST_DIRTY_SHIFT - PT_WRITABLE_SHIFT)) &
123 static inline int FNAME(is_present_gpte)(unsigned long pte)
125 #if PTTYPE != PTTYPE_EPT
126 return pte & PT_PRESENT_MASK;
132 static bool FNAME(is_bad_mt_xwr)(struct rsvd_bits_validate *rsvd_check, u64 gpte)
134 #if PTTYPE != PTTYPE_EPT
137 return __is_bad_mt_xwr(rsvd_check, gpte);
141 static bool FNAME(is_rsvd_bits_set)(struct kvm_mmu *mmu, u64 gpte, int level)
143 return __is_rsvd_bits_set(&mmu->guest_rsvd_check, gpte, level) ||
144 FNAME(is_bad_mt_xwr)(&mmu->guest_rsvd_check, gpte);
147 static int FNAME(cmpxchg_gpte)(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
148 pt_element_t __user *ptep_user, unsigned index,
149 pt_element_t orig_pte, pt_element_t new_pte)
153 if (!user_access_begin(ptep_user, sizeof(pt_element_t)))
157 asm volatile("1:" LOCK_PREFIX CMPXCHG " %[new], %[ptr]\n"
161 _ASM_EXTABLE_UA(1b, 2b)
162 : [ptr] "+m" (*ptep_user),
163 [old] "+a" (orig_pte),
165 : [new] "r" (new_pte)
168 asm volatile("1:" LOCK_PREFIX "cmpxchg8b %[ptr]\n"
173 _ASM_EXTABLE_UA(1b, 2b)
174 : [ptr] "+m" (*ptep_user),
175 [old] "+A" (orig_pte),
177 : [new_lo] "b" ((u32)new_pte),
178 [new_hi] "c" ((u32)(new_pte >> 32))
186 static bool FNAME(prefetch_invalid_gpte)(struct kvm_vcpu *vcpu,
187 struct kvm_mmu_page *sp, u64 *spte,
190 if (!FNAME(is_present_gpte)(gpte))
193 /* if accessed bit is not supported prefetch non accessed gpte */
194 if (PT_HAVE_ACCESSED_DIRTY(vcpu->arch.mmu) &&
195 !(gpte & PT_GUEST_ACCESSED_MASK))
198 if (FNAME(is_rsvd_bits_set)(vcpu->arch.mmu, gpte, PG_LEVEL_4K))
204 drop_spte(vcpu->kvm, spte);
209 * For PTTYPE_EPT, a page table can be executable but not readable
210 * on supported processors. Therefore, set_spte does not automatically
211 * set bit 0 if execute only is supported. Here, we repurpose ACC_USER_MASK
212 * to signify readability since it isn't used in the EPT case
214 static inline unsigned FNAME(gpte_access)(u64 gpte)
217 #if PTTYPE == PTTYPE_EPT
218 access = ((gpte & VMX_EPT_WRITABLE_MASK) ? ACC_WRITE_MASK : 0) |
219 ((gpte & VMX_EPT_EXECUTABLE_MASK) ? ACC_EXEC_MASK : 0) |
220 ((gpte & VMX_EPT_READABLE_MASK) ? ACC_USER_MASK : 0);
222 BUILD_BUG_ON(ACC_EXEC_MASK != PT_PRESENT_MASK);
223 BUILD_BUG_ON(ACC_EXEC_MASK != 1);
224 access = gpte & (PT_WRITABLE_MASK | PT_USER_MASK | PT_PRESENT_MASK);
225 /* Combine NX with P (which is set here) to get ACC_EXEC_MASK. */
226 access ^= (gpte >> PT64_NX_SHIFT);
232 static int FNAME(update_accessed_dirty_bits)(struct kvm_vcpu *vcpu,
234 struct guest_walker *walker,
235 gpa_t addr, int write_fault)
237 unsigned level, index;
238 pt_element_t pte, orig_pte;
239 pt_element_t __user *ptep_user;
243 /* dirty/accessed bits are not supported, so no need to update them */
244 if (!PT_HAVE_ACCESSED_DIRTY(mmu))
247 for (level = walker->max_level; level >= walker->level; --level) {
248 pte = orig_pte = walker->ptes[level - 1];
249 table_gfn = walker->table_gfn[level - 1];
250 ptep_user = walker->ptep_user[level - 1];
251 index = offset_in_page(ptep_user) / sizeof(pt_element_t);
252 if (!(pte & PT_GUEST_ACCESSED_MASK)) {
253 trace_kvm_mmu_set_accessed_bit(table_gfn, index, sizeof(pte));
254 pte |= PT_GUEST_ACCESSED_MASK;
256 if (level == walker->level && write_fault &&
257 !(pte & PT_GUEST_DIRTY_MASK)) {
258 trace_kvm_mmu_set_dirty_bit(table_gfn, index, sizeof(pte));
259 #if PTTYPE == PTTYPE_EPT
260 if (kvm_x86_ops.nested_ops->write_log_dirty(vcpu, addr))
263 pte |= PT_GUEST_DIRTY_MASK;
269 * If the slot is read-only, simply do not process the accessed
270 * and dirty bits. This is the correct thing to do if the slot
271 * is ROM, and page tables in read-as-ROM/write-as-MMIO slots
272 * are only supported if the accessed and dirty bits are already
273 * set in the ROM (so that MMIO writes are never needed).
275 * Note that NPT does not allow this at all and faults, since
276 * it always wants nested page table entries for the guest
277 * page tables to be writable. And EPT works but will simply
278 * overwrite the read-only memory to set the accessed and dirty
281 if (unlikely(!walker->pte_writable[level - 1]))
284 ret = FNAME(cmpxchg_gpte)(vcpu, mmu, ptep_user, index, orig_pte, pte);
288 kvm_vcpu_mark_page_dirty(vcpu, table_gfn);
289 walker->ptes[level - 1] = pte;
294 static inline unsigned FNAME(gpte_pkeys)(struct kvm_vcpu *vcpu, u64 gpte)
298 pte_t pte = {.pte = gpte};
300 pkeys = pte_flags_pkey(pte_flags(pte));
305 static inline bool FNAME(is_last_gpte)(struct kvm_mmu *mmu,
306 unsigned int level, unsigned int gpte)
309 * For EPT and PAE paging (both variants), bit 7 is either reserved at
310 * all level or indicates a huge page (ignoring CR3/EPTP). In either
311 * case, bit 7 being set terminates the walk.
315 * 32-bit paging requires special handling because bit 7 is ignored if
316 * CR4.PSE=0, not reserved. Clear bit 7 in the gpte if the level is
317 * greater than the last level for which bit 7 is the PAGE_SIZE bit.
319 * The RHS has bit 7 set iff level < (2 + PSE). If it is clear, bit 7
320 * is not reserved and does not indicate a large page at this level,
321 * so clear PT_PAGE_SIZE_MASK in gpte if that is the case.
323 gpte &= level - (PT32_ROOT_LEVEL + mmu->mmu_role.ext.cr4_pse);
326 * PG_LEVEL_4K always terminates. The RHS has bit 7 set
327 * iff level <= PG_LEVEL_4K, which for our purpose means
328 * level == PG_LEVEL_4K; set PT_PAGE_SIZE_MASK in gpte then.
330 gpte |= level - PG_LEVEL_4K - 1;
332 return gpte & PT_PAGE_SIZE_MASK;
335 * Fetch a guest pte for a guest virtual address, or for an L2's GPA.
337 static int FNAME(walk_addr_generic)(struct guest_walker *walker,
338 struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
339 gpa_t addr, u32 access)
343 pt_element_t __user *ptep_user;
345 u64 pt_access, pte_access;
346 unsigned index, accessed_dirty, pte_pkey;
347 unsigned nested_access;
351 u64 walk_nx_mask = 0;
352 const int write_fault = access & PFERR_WRITE_MASK;
353 const int user_fault = access & PFERR_USER_MASK;
354 const int fetch_fault = access & PFERR_FETCH_MASK;
359 trace_kvm_mmu_pagetable_walk(addr, access);
361 walker->level = mmu->root_level;
362 pte = mmu->get_guest_pgd(vcpu);
363 have_ad = PT_HAVE_ACCESSED_DIRTY(mmu);
366 walk_nx_mask = 1ULL << PT64_NX_SHIFT;
367 if (walker->level == PT32E_ROOT_LEVEL) {
368 pte = mmu->get_pdptr(vcpu, (addr >> 30) & 3);
369 trace_kvm_mmu_paging_element(pte, walker->level);
370 if (!FNAME(is_present_gpte)(pte))
375 walker->max_level = walker->level;
376 ASSERT(!(is_long_mode(vcpu) && !is_pae(vcpu)));
379 * FIXME: on Intel processors, loads of the PDPTE registers for PAE paging
380 * by the MOV to CR instruction are treated as reads and do not cause the
381 * processor to set the dirty flag in any EPT paging-structure entry.
383 nested_access = (have_ad ? PFERR_WRITE_MASK : 0) | PFERR_USER_MASK;
389 unsigned long host_addr;
391 pt_access = pte_access;
394 index = PT_INDEX(addr, walker->level);
395 table_gfn = gpte_to_gfn(pte);
396 offset = index * sizeof(pt_element_t);
397 pte_gpa = gfn_to_gpa(table_gfn) + offset;
399 BUG_ON(walker->level < 1);
400 walker->table_gfn[walker->level - 1] = table_gfn;
401 walker->pte_gpa[walker->level - 1] = pte_gpa;
403 real_gpa = mmu->translate_gpa(vcpu, gfn_to_gpa(table_gfn),
408 * FIXME: This can happen if emulation (for of an INS/OUTS
409 * instruction) triggers a nested page fault. The exit
410 * qualification / exit info field will incorrectly have
411 * "guest page access" as the nested page fault's cause,
412 * instead of "guest page structure access". To fix this,
413 * the x86_exception struct should be augmented with enough
414 * information to fix the exit_qualification or exit_info_1
417 if (unlikely(real_gpa == UNMAPPED_GVA))
420 host_addr = kvm_vcpu_gfn_to_hva_prot(vcpu, gpa_to_gfn(real_gpa),
421 &walker->pte_writable[walker->level - 1]);
422 if (unlikely(kvm_is_error_hva(host_addr)))
425 ptep_user = (pt_element_t __user *)((void *)host_addr + offset);
426 if (unlikely(__get_user(pte, ptep_user)))
428 walker->ptep_user[walker->level - 1] = ptep_user;
430 trace_kvm_mmu_paging_element(pte, walker->level);
433 * Inverting the NX it lets us AND it like other
436 pte_access = pt_access & (pte ^ walk_nx_mask);
438 if (unlikely(!FNAME(is_present_gpte)(pte)))
441 if (unlikely(FNAME(is_rsvd_bits_set)(mmu, pte, walker->level))) {
442 errcode = PFERR_RSVD_MASK | PFERR_PRESENT_MASK;
446 walker->ptes[walker->level - 1] = pte;
448 /* Convert to ACC_*_MASK flags for struct guest_walker. */
449 walker->pt_access[walker->level - 1] = FNAME(gpte_access)(pt_access ^ walk_nx_mask);
450 } while (!FNAME(is_last_gpte)(mmu, walker->level, pte));
452 pte_pkey = FNAME(gpte_pkeys)(vcpu, pte);
453 accessed_dirty = have_ad ? pte_access & PT_GUEST_ACCESSED_MASK : 0;
455 /* Convert to ACC_*_MASK flags for struct guest_walker. */
456 walker->pte_access = FNAME(gpte_access)(pte_access ^ walk_nx_mask);
457 errcode = permission_fault(vcpu, mmu, walker->pte_access, pte_pkey, access);
458 if (unlikely(errcode))
461 gfn = gpte_to_gfn_lvl(pte, walker->level);
462 gfn += (addr & PT_LVL_OFFSET_MASK(walker->level)) >> PAGE_SHIFT;
464 if (PTTYPE == 32 && walker->level > PG_LEVEL_4K && is_cpuid_PSE36())
465 gfn += pse36_gfn_delta(pte);
467 real_gpa = mmu->translate_gpa(vcpu, gfn_to_gpa(gfn), access, &walker->fault);
468 if (real_gpa == UNMAPPED_GVA)
471 walker->gfn = real_gpa >> PAGE_SHIFT;
474 FNAME(protect_clean_gpte)(mmu, &walker->pte_access, pte);
477 * On a write fault, fold the dirty bit into accessed_dirty.
478 * For modes without A/D bits support accessed_dirty will be
481 accessed_dirty &= pte >>
482 (PT_GUEST_DIRTY_SHIFT - PT_GUEST_ACCESSED_SHIFT);
484 if (unlikely(!accessed_dirty)) {
485 ret = FNAME(update_accessed_dirty_bits)(vcpu, mmu, walker,
487 if (unlikely(ret < 0))
493 pgprintk("%s: pte %llx pte_access %x pt_access %x\n",
494 __func__, (u64)pte, walker->pte_access,
495 walker->pt_access[walker->level - 1]);
499 errcode |= write_fault | user_fault;
500 if (fetch_fault && (is_efer_nx(mmu) || is_cr4_smep(mmu)))
501 errcode |= PFERR_FETCH_MASK;
503 walker->fault.vector = PF_VECTOR;
504 walker->fault.error_code_valid = true;
505 walker->fault.error_code = errcode;
507 #if PTTYPE == PTTYPE_EPT
509 * Use PFERR_RSVD_MASK in error_code to to tell if EPT
510 * misconfiguration requires to be injected. The detection is
511 * done by is_rsvd_bits_set() above.
513 * We set up the value of exit_qualification to inject:
514 * [2:0] - Derive from the access bits. The exit_qualification might be
515 * out of date if it is serving an EPT misconfiguration.
516 * [5:3] - Calculated by the page walk of the guest EPT page tables
517 * [7:8] - Derived from [7:8] of real exit_qualification
519 * The other bits are set to 0.
521 if (!(errcode & PFERR_RSVD_MASK)) {
522 vcpu->arch.exit_qualification &= 0x180;
524 vcpu->arch.exit_qualification |= EPT_VIOLATION_ACC_WRITE;
526 vcpu->arch.exit_qualification |= EPT_VIOLATION_ACC_READ;
528 vcpu->arch.exit_qualification |= EPT_VIOLATION_ACC_INSTR;
529 vcpu->arch.exit_qualification |= (pte_access & 0x7) << 3;
532 walker->fault.address = addr;
533 walker->fault.nested_page_fault = mmu != vcpu->arch.walk_mmu;
534 walker->fault.async_page_fault = false;
536 trace_kvm_mmu_walker_error(walker->fault.error_code);
540 static int FNAME(walk_addr)(struct guest_walker *walker,
541 struct kvm_vcpu *vcpu, gpa_t addr, u32 access)
543 return FNAME(walk_addr_generic)(walker, vcpu, vcpu->arch.mmu, addr,
547 #if PTTYPE != PTTYPE_EPT
548 static int FNAME(walk_addr_nested)(struct guest_walker *walker,
549 struct kvm_vcpu *vcpu, gva_t addr,
552 return FNAME(walk_addr_generic)(walker, vcpu, &vcpu->arch.nested_mmu,
558 FNAME(prefetch_gpte)(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp,
559 u64 *spte, pt_element_t gpte, bool no_dirty_log)
565 if (FNAME(prefetch_invalid_gpte)(vcpu, sp, spte, gpte))
568 pgprintk("%s: gpte %llx spte %p\n", __func__, (u64)gpte, spte);
570 gfn = gpte_to_gfn(gpte);
571 pte_access = sp->role.access & FNAME(gpte_access)(gpte);
572 FNAME(protect_clean_gpte)(vcpu->arch.mmu, &pte_access, gpte);
573 pfn = pte_prefetch_gfn_to_pfn(vcpu, gfn,
574 no_dirty_log && (pte_access & ACC_WRITE_MASK));
575 if (is_error_pfn(pfn))
579 * we call mmu_set_spte() with host_writable = true because
580 * pte_prefetch_gfn_to_pfn always gets a writable pfn.
582 mmu_set_spte(vcpu, spte, pte_access, false, PG_LEVEL_4K, gfn, pfn,
585 kvm_release_pfn_clean(pfn);
589 static void FNAME(update_pte)(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp,
590 u64 *spte, const void *pte)
592 pt_element_t gpte = *(const pt_element_t *)pte;
594 FNAME(prefetch_gpte)(vcpu, sp, spte, gpte, false);
597 static bool FNAME(gpte_changed)(struct kvm_vcpu *vcpu,
598 struct guest_walker *gw, int level)
600 pt_element_t curr_pte;
601 gpa_t base_gpa, pte_gpa = gw->pte_gpa[level - 1];
605 if (level == PG_LEVEL_4K) {
606 mask = PTE_PREFETCH_NUM * sizeof(pt_element_t) - 1;
607 base_gpa = pte_gpa & ~mask;
608 index = (pte_gpa - base_gpa) / sizeof(pt_element_t);
610 r = kvm_vcpu_read_guest_atomic(vcpu, base_gpa,
611 gw->prefetch_ptes, sizeof(gw->prefetch_ptes));
612 curr_pte = gw->prefetch_ptes[index];
614 r = kvm_vcpu_read_guest_atomic(vcpu, pte_gpa,
615 &curr_pte, sizeof(curr_pte));
617 return r || curr_pte != gw->ptes[level - 1];
620 static void FNAME(pte_prefetch)(struct kvm_vcpu *vcpu, struct guest_walker *gw,
623 struct kvm_mmu_page *sp;
624 pt_element_t *gptep = gw->prefetch_ptes;
628 sp = sptep_to_sp(sptep);
630 if (sp->role.level > PG_LEVEL_4K)
634 * If addresses are being invalidated, skip prefetching to avoid
635 * accidentally prefetching those addresses.
637 if (unlikely(vcpu->kvm->mmu_notifier_count))
641 return __direct_pte_prefetch(vcpu, sp, sptep);
643 i = (sptep - sp->spt) & ~(PTE_PREFETCH_NUM - 1);
646 for (i = 0; i < PTE_PREFETCH_NUM; i++, spte++) {
650 if (is_shadow_present_pte(*spte))
653 if (!FNAME(prefetch_gpte)(vcpu, sp, spte, gptep[i], true))
659 * Fetch a shadow pte for a specific level in the paging hierarchy.
660 * If the guest tries to write a write-protected page, we need to
661 * emulate this operation, return 1 to indicate this case.
663 static int FNAME(fetch)(struct kvm_vcpu *vcpu, gpa_t addr,
664 struct guest_walker *gw, u32 error_code,
665 int max_level, kvm_pfn_t pfn, bool map_writable,
668 bool nx_huge_page_workaround_enabled = is_nx_huge_page_enabled();
669 bool write_fault = error_code & PFERR_WRITE_MASK;
670 bool exec = error_code & PFERR_FETCH_MASK;
671 bool huge_page_disallowed = exec && nx_huge_page_workaround_enabled;
672 struct kvm_mmu_page *sp = NULL;
673 struct kvm_shadow_walk_iterator it;
674 unsigned int direct_access, access;
675 int top_level, level, req_level, ret;
676 gfn_t base_gfn = gw->gfn;
678 direct_access = gw->pte_access;
680 top_level = vcpu->arch.mmu->root_level;
681 if (top_level == PT32E_ROOT_LEVEL)
682 top_level = PT32_ROOT_LEVEL;
684 * Verify that the top-level gpte is still there. Since the page
685 * is a root page, it is either write protected (and cannot be
686 * changed from now on) or it is invalid (in which case, we don't
687 * really care if it changes underneath us after this point).
689 if (FNAME(gpte_changed)(vcpu, gw, top_level))
690 goto out_gpte_changed;
692 if (WARN_ON(!VALID_PAGE(vcpu->arch.mmu->root_hpa)))
693 goto out_gpte_changed;
695 for (shadow_walk_init(&it, vcpu, addr);
696 shadow_walk_okay(&it) && it.level > gw->level;
697 shadow_walk_next(&it)) {
700 clear_sp_write_flooding_count(it.sptep);
701 drop_large_spte(vcpu, it.sptep);
704 if (!is_shadow_present_pte(*it.sptep)) {
705 table_gfn = gw->table_gfn[it.level - 2];
706 access = gw->pt_access[it.level - 2];
707 sp = kvm_mmu_get_page(vcpu, table_gfn, addr,
708 it.level-1, false, access);
710 * We must synchronize the pagetable before linking it
711 * because the guest doesn't need to flush tlb when
712 * the gpte is changed from non-present to present.
713 * Otherwise, the guest may use the wrong mapping.
715 * For PG_LEVEL_4K, kvm_mmu_get_page() has already
716 * synchronized it transiently via kvm_sync_page().
718 * For higher level pagetable, we synchronize it via
719 * the slower mmu_sync_children(). If it needs to
720 * break, some progress has been made; return
721 * RET_PF_RETRY and retry on the next #PF.
722 * KVM_REQ_MMU_SYNC is not necessary but it
723 * expedites the process.
725 if (sp->unsync_children &&
726 mmu_sync_children(vcpu, sp, false))
731 * Verify that the gpte in the page we've just write
732 * protected is still there.
734 if (FNAME(gpte_changed)(vcpu, gw, it.level - 1))
735 goto out_gpte_changed;
738 link_shadow_page(vcpu, it.sptep, sp);
741 level = kvm_mmu_hugepage_adjust(vcpu, gw->gfn, max_level, &pfn,
742 huge_page_disallowed, &req_level);
744 trace_kvm_mmu_spte_requested(addr, gw->level, pfn);
746 for (; shadow_walk_okay(&it); shadow_walk_next(&it)) {
747 clear_sp_write_flooding_count(it.sptep);
750 * We cannot overwrite existing page tables with an NX
751 * large page, as the leaf could be executable.
753 if (nx_huge_page_workaround_enabled)
754 disallowed_hugepage_adjust(*it.sptep, gw->gfn, it.level,
757 base_gfn = gw->gfn & ~(KVM_PAGES_PER_HPAGE(it.level) - 1);
758 if (it.level == level)
761 validate_direct_spte(vcpu, it.sptep, direct_access);
763 drop_large_spte(vcpu, it.sptep);
765 if (!is_shadow_present_pte(*it.sptep)) {
766 sp = kvm_mmu_get_page(vcpu, base_gfn, addr,
767 it.level - 1, true, direct_access);
768 link_shadow_page(vcpu, it.sptep, sp);
769 if (huge_page_disallowed && req_level >= it.level)
770 account_huge_nx_page(vcpu->kvm, sp);
774 ret = mmu_set_spte(vcpu, it.sptep, gw->pte_access, write_fault,
775 it.level, base_gfn, pfn, prefault, map_writable);
776 if (ret == RET_PF_SPURIOUS)
779 FNAME(pte_prefetch)(vcpu, gw, it.sptep);
780 ++vcpu->stat.pf_fixed;
788 * To see whether the mapped gfn can write its page table in the current
791 * It is the helper function of FNAME(page_fault). When guest uses large page
792 * size to map the writable gfn which is used as current page table, we should
793 * force kvm to use small page size to map it because new shadow page will be
794 * created when kvm establishes shadow page table that stop kvm using large
795 * page size. Do it early can avoid unnecessary #PF and emulation.
797 * @write_fault_to_shadow_pgtable will return true if the fault gfn is
798 * currently used as its page table.
800 * Note: the PDPT page table is not checked for PAE-32 bit guest. It is ok
801 * since the PDPT is always shadowed, that means, we can not use large page
802 * size to map the gfn which is used as PDPT.
805 FNAME(is_self_change_mapping)(struct kvm_vcpu *vcpu,
806 struct guest_walker *walker, bool user_fault,
807 bool *write_fault_to_shadow_pgtable)
810 gfn_t mask = ~(KVM_PAGES_PER_HPAGE(walker->level) - 1);
811 bool self_changed = false;
813 if (!(walker->pte_access & ACC_WRITE_MASK ||
814 (!is_cr0_wp(vcpu->arch.mmu) && !user_fault)))
817 for (level = walker->level; level <= walker->max_level; level++) {
818 gfn_t gfn = walker->gfn ^ walker->table_gfn[level - 1];
820 self_changed |= !(gfn & mask);
821 *write_fault_to_shadow_pgtable |= !gfn;
828 * Page fault handler. There are several causes for a page fault:
829 * - there is no shadow pte for the guest pte
830 * - write access through a shadow pte marked read only so that we can set
832 * - write access to a shadow pte marked read only so we can update the page
833 * dirty bitmap, when userspace requests it
834 * - mmio access; in this case we will never install a present shadow pte
835 * - normal guest page fault due to the guest pte marked not present, not
836 * writable, or not executable
838 * Returns: 1 if we need to emulate the instruction, 0 otherwise, or
839 * a negative value on error.
841 static int FNAME(page_fault)(struct kvm_vcpu *vcpu, gpa_t addr, u32 error_code,
844 bool write_fault = error_code & PFERR_WRITE_MASK;
845 bool user_fault = error_code & PFERR_USER_MASK;
846 struct guest_walker walker;
850 unsigned long mmu_seq;
851 bool map_writable, is_self_change_mapping;
854 pgprintk("%s: addr %lx err %x\n", __func__, addr, error_code);
857 * If PFEC.RSVD is set, this is a shadow page fault.
858 * The bit needs to be cleared before walking guest page tables.
860 error_code &= ~PFERR_RSVD_MASK;
863 * Look up the guest pte for the faulting address.
865 r = FNAME(walk_addr)(&walker, vcpu, addr, error_code);
868 * The page is not mapped by the guest. Let the guest handle it.
871 pgprintk("%s: guest page fault\n", __func__);
873 kvm_inject_emulated_page_fault(vcpu, &walker.fault);
878 if (page_fault_handle_page_track(vcpu, error_code, walker.gfn)) {
879 shadow_page_table_clear_flood(vcpu, addr);
880 return RET_PF_EMULATE;
883 r = mmu_topup_memory_caches(vcpu, true);
887 vcpu->arch.write_fault_to_shadow_pgtable = false;
889 is_self_change_mapping = FNAME(is_self_change_mapping)(vcpu,
890 &walker, user_fault, &vcpu->arch.write_fault_to_shadow_pgtable);
892 if (is_self_change_mapping)
893 max_level = PG_LEVEL_4K;
895 max_level = walker.level;
897 mmu_seq = vcpu->kvm->mmu_notifier_seq;
900 if (kvm_faultin_pfn(vcpu, prefault, walker.gfn, addr, &pfn, &hva,
901 write_fault, &map_writable, &r))
904 if (handle_abnormal_pfn(vcpu, addr, walker.gfn, pfn, walker.pte_access, &r))
908 * Do not change pte_access if the pfn is a mmio page, otherwise
909 * we will cache the incorrect access into mmio spte.
911 if (write_fault && !(walker.pte_access & ACC_WRITE_MASK) &&
912 !is_cr0_wp(vcpu->arch.mmu) && !user_fault && !is_noslot_pfn(pfn)) {
913 walker.pte_access |= ACC_WRITE_MASK;
914 walker.pte_access &= ~ACC_USER_MASK;
917 * If we converted a user page to a kernel page,
918 * so that the kernel can write to it when cr0.wp=0,
919 * then we should prevent the kernel from executing it
920 * if SMEP is enabled.
922 if (is_cr4_smep(vcpu->arch.mmu))
923 walker.pte_access &= ~ACC_EXEC_MASK;
927 write_lock(&vcpu->kvm->mmu_lock);
928 if (!is_noslot_pfn(pfn) && mmu_notifier_retry_hva(vcpu->kvm, mmu_seq, hva))
931 kvm_mmu_audit(vcpu, AUDIT_PRE_PAGE_FAULT);
932 r = make_mmu_pages_available(vcpu);
935 r = FNAME(fetch)(vcpu, addr, &walker, error_code, max_level, pfn,
936 map_writable, prefault);
937 kvm_mmu_audit(vcpu, AUDIT_POST_PAGE_FAULT);
940 write_unlock(&vcpu->kvm->mmu_lock);
941 kvm_release_pfn_clean(pfn);
945 static gpa_t FNAME(get_level1_sp_gpa)(struct kvm_mmu_page *sp)
949 WARN_ON(sp->role.level != PG_LEVEL_4K);
952 offset = sp->role.quadrant << PT64_LEVEL_BITS;
954 return gfn_to_gpa(sp->gfn) + offset * sizeof(pt_element_t);
957 static void FNAME(invlpg)(struct kvm_vcpu *vcpu, gva_t gva, hpa_t root_hpa)
959 struct kvm_shadow_walk_iterator iterator;
960 struct kvm_mmu_page *sp;
965 vcpu_clear_mmio_info(vcpu, gva);
968 * No need to check return value here, rmap_can_add() can
969 * help us to skip pte prefetch later.
971 mmu_topup_memory_caches(vcpu, true);
973 if (!VALID_PAGE(root_hpa)) {
978 write_lock(&vcpu->kvm->mmu_lock);
979 for_each_shadow_entry_using_root(vcpu, root_hpa, gva, iterator) {
980 level = iterator.level;
981 sptep = iterator.sptep;
983 sp = sptep_to_sp(sptep);
985 if (is_last_spte(old_spte, level)) {
992 pte_gpa = FNAME(get_level1_sp_gpa)(sp);
993 pte_gpa += (sptep - sp->spt) * sizeof(pt_element_t);
995 mmu_page_zap_pte(vcpu->kvm, sp, sptep, NULL);
996 if (is_shadow_present_pte(old_spte))
997 kvm_flush_remote_tlbs_with_address(vcpu->kvm,
998 sp->gfn, KVM_PAGES_PER_HPAGE(sp->role.level));
1000 if (!rmap_can_add(vcpu))
1003 if (kvm_vcpu_read_guest_atomic(vcpu, pte_gpa, &gpte,
1004 sizeof(pt_element_t)))
1007 FNAME(update_pte)(vcpu, sp, sptep, &gpte);
1010 if (!is_shadow_present_pte(*sptep) || !sp->unsync_children)
1013 write_unlock(&vcpu->kvm->mmu_lock);
1016 /* Note, @addr is a GPA when gva_to_gpa() translates an L2 GPA to an L1 GPA. */
1017 static gpa_t FNAME(gva_to_gpa)(struct kvm_vcpu *vcpu, gpa_t addr, u32 access,
1018 struct x86_exception *exception)
1020 struct guest_walker walker;
1021 gpa_t gpa = UNMAPPED_GVA;
1024 r = FNAME(walk_addr)(&walker, vcpu, addr, access);
1027 gpa = gfn_to_gpa(walker.gfn);
1028 gpa |= addr & ~PAGE_MASK;
1029 } else if (exception)
1030 *exception = walker.fault;
1035 #if PTTYPE != PTTYPE_EPT
1036 /* Note, gva_to_gpa_nested() is only used to translate L2 GVAs. */
1037 static gpa_t FNAME(gva_to_gpa_nested)(struct kvm_vcpu *vcpu, gpa_t vaddr,
1039 struct x86_exception *exception)
1041 struct guest_walker walker;
1042 gpa_t gpa = UNMAPPED_GVA;
1045 #ifndef CONFIG_X86_64
1046 /* A 64-bit GVA should be impossible on 32-bit KVM. */
1047 WARN_ON_ONCE(vaddr >> 32);
1050 r = FNAME(walk_addr_nested)(&walker, vcpu, vaddr, access);
1053 gpa = gfn_to_gpa(walker.gfn);
1054 gpa |= vaddr & ~PAGE_MASK;
1055 } else if (exception)
1056 *exception = walker.fault;
1063 * Using the cached information from sp->gfns is safe because:
1064 * - The spte has a reference to the struct page, so the pfn for a given gfn
1065 * can't change unless all sptes pointing to it are nuked first.
1067 static int FNAME(sync_page)(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp)
1069 union kvm_mmu_page_role mmu_role = vcpu->arch.mmu->mmu_role.base;
1070 int i, nr_present = 0;
1072 gpa_t first_pte_gpa;
1073 int set_spte_ret = 0;
1076 * Ignore various flags when verifying that it's safe to sync a shadow
1077 * page using the current MMU context.
1079 * - level: not part of the overall MMU role and will never match as the MMU's
1080 * level tracks the root level
1081 * - access: updated based on the new guest PTE
1082 * - quadrant: not part of the overall MMU role (similar to level)
1084 const union kvm_mmu_page_role sync_role_ign = {
1091 * Direct pages can never be unsync, and KVM should never attempt to
1092 * sync a shadow page for a different MMU context, e.g. if the role
1093 * differs then the memslot lookup (SMM vs. non-SMM) will be bogus, the
1094 * reserved bits checks will be wrong, etc...
1096 if (WARN_ON_ONCE(sp->role.direct ||
1097 (sp->role.word ^ mmu_role.word) & ~sync_role_ign.word))
1100 first_pte_gpa = FNAME(get_level1_sp_gpa)(sp);
1102 for (i = 0; i < PT64_ENT_PER_PAGE; i++) {
1103 unsigned pte_access;
1111 pte_gpa = first_pte_gpa + i * sizeof(pt_element_t);
1113 if (kvm_vcpu_read_guest_atomic(vcpu, pte_gpa, &gpte,
1114 sizeof(pt_element_t)))
1117 if (FNAME(prefetch_invalid_gpte)(vcpu, sp, &sp->spt[i], gpte)) {
1118 set_spte_ret |= SET_SPTE_NEED_REMOTE_TLB_FLUSH;
1122 gfn = gpte_to_gfn(gpte);
1123 pte_access = sp->role.access;
1124 pte_access &= FNAME(gpte_access)(gpte);
1125 FNAME(protect_clean_gpte)(vcpu->arch.mmu, &pte_access, gpte);
1127 if (sync_mmio_spte(vcpu, &sp->spt[i], gfn, pte_access,
1131 if (gfn != sp->gfns[i]) {
1132 drop_spte(vcpu->kvm, &sp->spt[i]);
1133 set_spte_ret |= SET_SPTE_NEED_REMOTE_TLB_FLUSH;
1139 host_writable = sp->spt[i] & shadow_host_writable_mask;
1141 set_spte_ret |= set_spte(vcpu, &sp->spt[i],
1142 pte_access, PG_LEVEL_4K,
1143 gfn, spte_to_pfn(sp->spt[i]),
1144 true, false, host_writable);
1147 if (set_spte_ret & SET_SPTE_NEED_REMOTE_TLB_FLUSH)
1148 kvm_flush_remote_tlbs(vcpu->kvm);
1156 #undef PT_BASE_ADDR_MASK
1158 #undef PT_LVL_ADDR_MASK
1159 #undef PT_LVL_OFFSET_MASK
1160 #undef PT_LEVEL_BITS
1161 #undef PT_MAX_FULL_LEVELS
1163 #undef gpte_to_gfn_lvl
1165 #undef PT_GUEST_ACCESSED_MASK
1166 #undef PT_GUEST_DIRTY_MASK
1167 #undef PT_GUEST_DIRTY_SHIFT
1168 #undef PT_GUEST_ACCESSED_SHIFT
1169 #undef PT_HAVE_ACCESSED_DIRTY