Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1 | /* |
| 2 | * Kernel-based Virtual Machine driver for Linux |
| 3 | * |
| 4 | * This module enables machines with Intel VT-x extensions to run virtual |
| 5 | * machines without emulation or binary translation. |
| 6 | * |
| 7 | * MMU support |
| 8 | * |
| 9 | * Copyright (C) 2006 Qumranet, Inc. |
Nicolas Kaiser | 9611c18 | 2010-10-06 14:23:22 +0200 | [diff] [blame] | 10 | * Copyright 2010 Red Hat, Inc. and/or its affiliates. |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 11 | * |
| 12 | * Authors: |
| 13 | * Yaniv Kamay <yaniv@qumranet.com> |
| 14 | * Avi Kivity <avi@qumranet.com> |
| 15 | * |
| 16 | * This work is licensed under the terms of the GNU GPL, version 2. See |
| 17 | * the COPYING file in the top-level directory. |
| 18 | * |
| 19 | */ |
| 20 | |
| 21 | /* |
| 22 | * We need the mmu code to access both 32-bit and 64-bit guest ptes, |
| 23 | * so the code in this file is compiled twice, once per pte size. |
| 24 | */ |
| 25 | |
| 26 | #if PTTYPE == 64 |
| 27 | #define pt_element_t u64 |
| 28 | #define guest_walker guest_walker64 |
| 29 | #define FNAME(name) paging##64_##name |
| 30 | #define PT_BASE_ADDR_MASK PT64_BASE_ADDR_MASK |
Joerg Roedel | e04da98 | 2009-07-27 16:30:45 +0200 | [diff] [blame] | 31 | #define PT_LVL_ADDR_MASK(lvl) PT64_LVL_ADDR_MASK(lvl) |
| 32 | #define PT_LVL_OFFSET_MASK(lvl) PT64_LVL_OFFSET_MASK(lvl) |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 33 | #define PT_INDEX(addr, level) PT64_INDEX(addr, level) |
Avi Kivity | c7addb9 | 2007-09-16 18:58:32 +0200 | [diff] [blame] | 34 | #define PT_LEVEL_BITS PT64_LEVEL_BITS |
Avi Kivity | cea0f0e | 2007-01-05 16:36:43 -0800 | [diff] [blame] | 35 | #ifdef CONFIG_X86_64 |
| 36 | #define PT_MAX_FULL_LEVELS 4 |
Marcelo Tosatti | b3e4e63 | 2007-12-07 07:56:58 -0500 | [diff] [blame] | 37 | #define CMPXCHG cmpxchg |
Avi Kivity | cea0f0e | 2007-01-05 16:36:43 -0800 | [diff] [blame] | 38 | #else |
Marcelo Tosatti | b3e4e63 | 2007-12-07 07:56:58 -0500 | [diff] [blame] | 39 | #define CMPXCHG cmpxchg64 |
Avi Kivity | cea0f0e | 2007-01-05 16:36:43 -0800 | [diff] [blame] | 40 | #define PT_MAX_FULL_LEVELS 2 |
| 41 | #endif |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 42 | #elif PTTYPE == 32 |
| 43 | #define pt_element_t u32 |
| 44 | #define guest_walker guest_walker32 |
| 45 | #define FNAME(name) paging##32_##name |
| 46 | #define PT_BASE_ADDR_MASK PT32_BASE_ADDR_MASK |
Joerg Roedel | e04da98 | 2009-07-27 16:30:45 +0200 | [diff] [blame] | 47 | #define PT_LVL_ADDR_MASK(lvl) PT32_LVL_ADDR_MASK(lvl) |
| 48 | #define PT_LVL_OFFSET_MASK(lvl) PT32_LVL_OFFSET_MASK(lvl) |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 49 | #define PT_INDEX(addr, level) PT32_INDEX(addr, level) |
Avi Kivity | c7addb9 | 2007-09-16 18:58:32 +0200 | [diff] [blame] | 50 | #define PT_LEVEL_BITS PT32_LEVEL_BITS |
Avi Kivity | cea0f0e | 2007-01-05 16:36:43 -0800 | [diff] [blame] | 51 | #define PT_MAX_FULL_LEVELS 2 |
Marcelo Tosatti | b3e4e63 | 2007-12-07 07:56:58 -0500 | [diff] [blame] | 52 | #define CMPXCHG cmpxchg |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 53 | #else |
| 54 | #error Invalid PTTYPE value |
| 55 | #endif |
| 56 | |
Joerg Roedel | e04da98 | 2009-07-27 16:30:45 +0200 | [diff] [blame] | 57 | #define gpte_to_gfn_lvl FNAME(gpte_to_gfn_lvl) |
| 58 | #define gpte_to_gfn(pte) gpte_to_gfn_lvl((pte), PT_PAGE_TABLE_LEVEL) |
Avi Kivity | 5fb07dd | 2007-11-21 12:35:07 +0200 | [diff] [blame] | 59 | |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 60 | /* |
| 61 | * The guest_walker structure emulates the behavior of the hardware page |
| 62 | * table walker. |
| 63 | */ |
| 64 | struct guest_walker { |
| 65 | int level; |
Avi Kivity | cea0f0e | 2007-01-05 16:36:43 -0800 | [diff] [blame] | 66 | gfn_t table_gfn[PT_MAX_FULL_LEVELS]; |
Marcelo Tosatti | 7819026 | 2007-12-11 19:12:27 -0500 | [diff] [blame] | 67 | pt_element_t ptes[PT_MAX_FULL_LEVELS]; |
Xiao Guangrong | 189be38 | 2010-08-22 19:13:33 +0800 | [diff] [blame] | 68 | pt_element_t prefetch_ptes[PTE_PREFETCH_NUM]; |
Marcelo Tosatti | 7819026 | 2007-12-11 19:12:27 -0500 | [diff] [blame] | 69 | gpa_t pte_gpa[PT_MAX_FULL_LEVELS]; |
Avi Kivity | fe135d2 | 2007-12-09 16:15:46 +0200 | [diff] [blame] | 70 | unsigned pt_access; |
| 71 | unsigned pte_access; |
Avi Kivity | 815af8d | 2007-01-05 16:36:44 -0800 | [diff] [blame] | 72 | gfn_t gfn; |
Avi Kivity | 8c28d03 | 2010-11-22 17:53:27 +0200 | [diff] [blame] | 73 | struct x86_exception fault; |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 74 | }; |
| 75 | |
Joerg Roedel | e04da98 | 2009-07-27 16:30:45 +0200 | [diff] [blame] | 76 | static gfn_t gpte_to_gfn_lvl(pt_element_t gpte, int lvl) |
Avi Kivity | 5fb07dd | 2007-11-21 12:35:07 +0200 | [diff] [blame] | 77 | { |
Joerg Roedel | e04da98 | 2009-07-27 16:30:45 +0200 | [diff] [blame] | 78 | return (gpte & PT_LVL_ADDR_MASK(lvl)) >> PAGE_SHIFT; |
Avi Kivity | 5fb07dd | 2007-11-21 12:35:07 +0200 | [diff] [blame] | 79 | } |
| 80 | |
Roedel, Joerg | a78484c | 2011-04-20 15:33:16 +0200 | [diff] [blame] | 81 | static int FNAME(cmpxchg_gpte)(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, |
Takuya Yoshikawa | c8cfbb5 | 2011-05-01 14:33:07 +0900 | [diff] [blame] | 82 | pt_element_t __user *ptep_user, unsigned index, |
| 83 | pt_element_t orig_pte, pt_element_t new_pte) |
Marcelo Tosatti | b3e4e63 | 2007-12-07 07:56:58 -0500 | [diff] [blame] | 84 | { |
Takuya Yoshikawa | c8cfbb5 | 2011-05-01 14:33:07 +0900 | [diff] [blame] | 85 | int npages; |
Marcelo Tosatti | b3e4e63 | 2007-12-07 07:56:58 -0500 | [diff] [blame] | 86 | pt_element_t ret; |
| 87 | pt_element_t *table; |
| 88 | struct page *page; |
| 89 | |
Takuya Yoshikawa | c8cfbb5 | 2011-05-01 14:33:07 +0900 | [diff] [blame] | 90 | npages = get_user_pages_fast((unsigned long)ptep_user, 1, 1, &page); |
| 91 | /* Check if the user is doing something meaningless. */ |
| 92 | if (unlikely(npages != 1)) |
Roedel, Joerg | a78484c | 2011-04-20 15:33:16 +0200 | [diff] [blame] | 93 | return -EFAULT; |
| 94 | |
Cong Wang | 8fd75e1 | 2011-11-25 23:14:17 +0800 | [diff] [blame] | 95 | table = kmap_atomic(page); |
Marcelo Tosatti | b3e4e63 | 2007-12-07 07:56:58 -0500 | [diff] [blame] | 96 | ret = CMPXCHG(&table[index], orig_pte, new_pte); |
Cong Wang | 8fd75e1 | 2011-11-25 23:14:17 +0800 | [diff] [blame] | 97 | kunmap_atomic(table); |
Marcelo Tosatti | b3e4e63 | 2007-12-07 07:56:58 -0500 | [diff] [blame] | 98 | |
| 99 | kvm_release_page_dirty(page); |
| 100 | |
| 101 | return (ret != orig_pte); |
| 102 | } |
| 103 | |
Xiao Guangrong | 640d9b0 | 2011-07-12 03:24:39 +0800 | [diff] [blame] | 104 | static unsigned FNAME(gpte_access)(struct kvm_vcpu *vcpu, pt_element_t gpte, |
| 105 | bool last) |
Avi Kivity | bedbe4e | 2007-12-09 16:52:56 +0200 | [diff] [blame] | 106 | { |
| 107 | unsigned access; |
| 108 | |
| 109 | access = (gpte & (PT_WRITABLE_MASK | PT_USER_MASK)) | ACC_EXEC_MASK; |
Xiao Guangrong | 640d9b0 | 2011-07-12 03:24:39 +0800 | [diff] [blame] | 110 | if (last && !is_dirty_gpte(gpte)) |
| 111 | access &= ~ACC_WRITE_MASK; |
| 112 | |
Avi Kivity | bedbe4e | 2007-12-09 16:52:56 +0200 | [diff] [blame] | 113 | #if PTTYPE == 64 |
Joerg Roedel | 2d48a98 | 2010-09-10 17:31:01 +0200 | [diff] [blame] | 114 | if (vcpu->arch.mmu.nx) |
Avi Kivity | bedbe4e | 2007-12-09 16:52:56 +0200 | [diff] [blame] | 115 | access &= ~(gpte >> PT64_NX_SHIFT); |
| 116 | #endif |
| 117 | return access; |
| 118 | } |
| 119 | |
Takuya Yoshikawa | 3c8c652 | 2011-07-01 01:37:24 +0900 | [diff] [blame] | 120 | static bool FNAME(is_last_gpte)(struct guest_walker *walker, |
| 121 | struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, |
| 122 | pt_element_t gpte) |
| 123 | { |
| 124 | if (walker->level == PT_PAGE_TABLE_LEVEL) |
| 125 | return true; |
| 126 | |
| 127 | if ((walker->level == PT_DIRECTORY_LEVEL) && is_large_pte(gpte) && |
| 128 | (PTTYPE == 64 || is_pse(vcpu))) |
| 129 | return true; |
| 130 | |
| 131 | if ((walker->level == PT_PDPE_LEVEL) && is_large_pte(gpte) && |
| 132 | (mmu->root_level == PT64_ROOT_LEVEL)) |
| 133 | return true; |
| 134 | |
| 135 | return false; |
| 136 | } |
| 137 | |
Avi Kivity | ac79c97 | 2007-01-05 16:36:40 -0800 | [diff] [blame] | 138 | /* |
| 139 | * Fetch a guest pte for a guest virtual address |
| 140 | */ |
Joerg Roedel | 1e301fe | 2010-09-10 17:30:47 +0200 | [diff] [blame] | 141 | static int FNAME(walk_addr_generic)(struct guest_walker *walker, |
| 142 | struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, |
Xiao Guangrong | 3377078 | 2010-09-28 17:03:14 +0800 | [diff] [blame] | 143 | gva_t addr, u32 access) |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 144 | { |
Avi Kivity | 42bf3f0 | 2007-10-17 12:18:47 +0200 | [diff] [blame] | 145 | pt_element_t pte; |
Borislav Petkov | b723363 | 2011-05-30 22:11:17 +0200 | [diff] [blame] | 146 | pt_element_t __user *uninitialized_var(ptep_user); |
Avi Kivity | cea0f0e | 2007-01-05 16:36:43 -0800 | [diff] [blame] | 147 | gfn_t table_gfn; |
Avi Kivity | f59c1d2 | 2010-07-06 16:20:43 +0300 | [diff] [blame] | 148 | unsigned index, pt_access, uninitialized_var(pte_access); |
Avi Kivity | 42bf3f0 | 2007-10-17 12:18:47 +0200 | [diff] [blame] | 149 | gpa_t pte_gpa; |
Yang, Wei Y | cd46868 | 2011-08-09 18:14:01 +0800 | [diff] [blame] | 150 | bool eperm, last_gpte; |
Takuya Yoshikawa | 134291b | 2011-07-01 01:34:56 +0900 | [diff] [blame] | 151 | int offset; |
| 152 | const int write_fault = access & PFERR_WRITE_MASK; |
| 153 | const int user_fault = access & PFERR_USER_MASK; |
| 154 | const int fetch_fault = access & PFERR_FETCH_MASK; |
| 155 | u16 errcode = 0; |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 156 | |
Avi Kivity | 0742017 | 2009-07-06 12:21:32 +0300 | [diff] [blame] | 157 | trace_kvm_mmu_pagetable_walk(addr, write_fault, user_fault, |
| 158 | fetch_fault); |
Takuya Yoshikawa | 92c1c1e | 2011-07-01 01:36:07 +0900 | [diff] [blame] | 159 | retry_walk: |
Takuya Yoshikawa | 134291b | 2011-07-01 01:34:56 +0900 | [diff] [blame] | 160 | eperm = false; |
Joerg Roedel | 1e301fe | 2010-09-10 17:30:47 +0200 | [diff] [blame] | 161 | walker->level = mmu->root_level; |
| 162 | pte = mmu->get_cr3(vcpu); |
| 163 | |
Avi Kivity | 1b0973b | 2007-01-05 16:36:41 -0800 | [diff] [blame] | 164 | #if PTTYPE == 64 |
Joerg Roedel | 1e301fe | 2010-09-10 17:30:47 +0200 | [diff] [blame] | 165 | if (walker->level == PT32E_ROOT_LEVEL) { |
Avi Kivity | e4e517b | 2011-07-28 11:36:17 +0300 | [diff] [blame] | 166 | pte = mmu->get_pdptr(vcpu, (addr >> 30) & 3); |
Avi Kivity | 0742017 | 2009-07-06 12:21:32 +0300 | [diff] [blame] | 167 | trace_kvm_mmu_paging_element(pte, walker->level); |
Takuya Yoshikawa | 134291b | 2011-07-01 01:34:56 +0900 | [diff] [blame] | 168 | if (!is_present_gpte(pte)) |
Avi Kivity | f59c1d2 | 2010-07-06 16:20:43 +0300 | [diff] [blame] | 169 | goto error; |
Avi Kivity | 1b0973b | 2007-01-05 16:36:41 -0800 | [diff] [blame] | 170 | --walker->level; |
| 171 | } |
| 172 | #endif |
Avi Kivity | a9058ec | 2006-12-29 16:49:37 -0800 | [diff] [blame] | 173 | ASSERT((!is_long_mode(vcpu) && is_pae(vcpu)) || |
Joerg Roedel | 1e301fe | 2010-09-10 17:30:47 +0200 | [diff] [blame] | 174 | (mmu->get_cr3(vcpu) & CR3_NONPAE_RESERVED_BITS) == 0); |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 175 | |
Avi Kivity | fe135d2 | 2007-12-09 16:15:46 +0200 | [diff] [blame] | 176 | pt_access = ACC_ALL; |
Avi Kivity | ac79c97 | 2007-01-05 16:36:40 -0800 | [diff] [blame] | 177 | |
| 178 | for (;;) { |
Takuya Yoshikawa | 6e2ca7d | 2011-04-22 00:34:44 +0900 | [diff] [blame] | 179 | gfn_t real_gfn; |
| 180 | unsigned long host_addr; |
| 181 | |
Avi Kivity | 42bf3f0 | 2007-10-17 12:18:47 +0200 | [diff] [blame] | 182 | index = PT_INDEX(addr, walker->level); |
Avi Kivity | ac79c97 | 2007-01-05 16:36:40 -0800 | [diff] [blame] | 183 | |
Avi Kivity | 5fb07dd | 2007-11-21 12:35:07 +0200 | [diff] [blame] | 184 | table_gfn = gpte_to_gfn(pte); |
Joerg Roedel | 2329d46 | 2010-09-10 17:30:52 +0200 | [diff] [blame] | 185 | offset = index * sizeof(pt_element_t); |
| 186 | pte_gpa = gfn_to_gpa(table_gfn) + offset; |
Avi Kivity | 42bf3f0 | 2007-10-17 12:18:47 +0200 | [diff] [blame] | 187 | walker->table_gfn[walker->level - 1] = table_gfn; |
Marcelo Tosatti | 7819026 | 2007-12-11 19:12:27 -0500 | [diff] [blame] | 188 | walker->pte_gpa[walker->level - 1] = pte_gpa; |
Avi Kivity | ac79c97 | 2007-01-05 16:36:40 -0800 | [diff] [blame] | 189 | |
Takuya Yoshikawa | 6e2ca7d | 2011-04-22 00:34:44 +0900 | [diff] [blame] | 190 | real_gfn = mmu->translate_gpa(vcpu, gfn_to_gpa(table_gfn), |
| 191 | PFERR_USER_MASK|PFERR_WRITE_MASK); |
Takuya Yoshikawa | 134291b | 2011-07-01 01:34:56 +0900 | [diff] [blame] | 192 | if (unlikely(real_gfn == UNMAPPED_GVA)) |
| 193 | goto error; |
Takuya Yoshikawa | 6e2ca7d | 2011-04-22 00:34:44 +0900 | [diff] [blame] | 194 | real_gfn = gpa_to_gfn(real_gfn); |
| 195 | |
| 196 | host_addr = gfn_to_hva(vcpu->kvm, real_gfn); |
Takuya Yoshikawa | 134291b | 2011-07-01 01:34:56 +0900 | [diff] [blame] | 197 | if (unlikely(kvm_is_error_hva(host_addr))) |
| 198 | goto error; |
Takuya Yoshikawa | 6e2ca7d | 2011-04-22 00:34:44 +0900 | [diff] [blame] | 199 | |
| 200 | ptep_user = (pt_element_t __user *)((void *)host_addr + offset); |
Takuya Yoshikawa | 134291b | 2011-07-01 01:34:56 +0900 | [diff] [blame] | 201 | if (unlikely(__copy_from_user(&pte, ptep_user, sizeof(pte)))) |
| 202 | goto error; |
Marcelo Tosatti | a6085fb | 2010-01-14 17:41:27 -0200 | [diff] [blame] | 203 | |
Avi Kivity | 0742017 | 2009-07-06 12:21:32 +0300 | [diff] [blame] | 204 | trace_kvm_mmu_paging_element(pte, walker->level); |
Avi Kivity | 42bf3f0 | 2007-10-17 12:18:47 +0200 | [diff] [blame] | 205 | |
Takuya Yoshikawa | 134291b | 2011-07-01 01:34:56 +0900 | [diff] [blame] | 206 | if (unlikely(!is_present_gpte(pte))) |
| 207 | goto error; |
Avi Kivity | 7993ba4 | 2007-01-26 00:56:41 -0800 | [diff] [blame] | 208 | |
Avi Kivity | 781e074 | 2011-04-24 12:25:50 +0300 | [diff] [blame] | 209 | if (unlikely(is_rsvd_bits_set(&vcpu->arch.mmu, pte, |
| 210 | walker->level))) { |
Takuya Yoshikawa | 134291b | 2011-07-01 01:34:56 +0900 | [diff] [blame] | 211 | errcode |= PFERR_RSVD_MASK | PFERR_PRESENT_MASK; |
| 212 | goto error; |
Avi Kivity | f59c1d2 | 2010-07-06 16:20:43 +0300 | [diff] [blame] | 213 | } |
Dong, Eddie | 82725b2 | 2009-03-30 16:21:08 +0800 | [diff] [blame] | 214 | |
Xiao Guangrong | bebb106 | 2011-07-12 03:23:20 +0800 | [diff] [blame] | 215 | if (!check_write_user_access(vcpu, write_fault, user_fault, |
| 216 | pte)) |
Avi Kivity | f59c1d2 | 2010-07-06 16:20:43 +0300 | [diff] [blame] | 217 | eperm = true; |
Avi Kivity | 7993ba4 | 2007-01-26 00:56:41 -0800 | [diff] [blame] | 218 | |
Avi Kivity | 73b1087 | 2007-01-26 00:56:41 -0800 | [diff] [blame] | 219 | #if PTTYPE == 64 |
Avi Kivity | 781e074 | 2011-04-24 12:25:50 +0300 | [diff] [blame] | 220 | if (unlikely(fetch_fault && (pte & PT64_NX_MASK))) |
Avi Kivity | f59c1d2 | 2010-07-06 16:20:43 +0300 | [diff] [blame] | 221 | eperm = true; |
Avi Kivity | 73b1087 | 2007-01-26 00:56:41 -0800 | [diff] [blame] | 222 | #endif |
| 223 | |
Yang, Wei Y | cd46868 | 2011-08-09 18:14:01 +0800 | [diff] [blame] | 224 | last_gpte = FNAME(is_last_gpte)(walker, vcpu, mmu, pte); |
| 225 | if (last_gpte) { |
| 226 | pte_access = pt_access & |
| 227 | FNAME(gpte_access)(vcpu, pte, true); |
| 228 | /* check if the kernel is fetching from user page */ |
| 229 | if (unlikely(pte_access & PT_USER_MASK) && |
| 230 | kvm_read_cr4_bits(vcpu, X86_CR4_SMEP)) |
| 231 | if (fetch_fault && !user_fault) |
| 232 | eperm = true; |
| 233 | } |
| 234 | |
Takuya Yoshikawa | 134291b | 2011-07-01 01:34:56 +0900 | [diff] [blame] | 235 | if (!eperm && unlikely(!(pte & PT_ACCESSED_MASK))) { |
Roedel, Joerg | a78484c | 2011-04-20 15:33:16 +0200 | [diff] [blame] | 236 | int ret; |
Avi Kivity | 0742017 | 2009-07-06 12:21:32 +0300 | [diff] [blame] | 237 | trace_kvm_mmu_set_accessed_bit(table_gfn, index, |
| 238 | sizeof(pte)); |
Takuya Yoshikawa | c8cfbb5 | 2011-05-01 14:33:07 +0900 | [diff] [blame] | 239 | ret = FNAME(cmpxchg_gpte)(vcpu, mmu, ptep_user, index, |
| 240 | pte, pte|PT_ACCESSED_MASK); |
Takuya Yoshikawa | 134291b | 2011-07-01 01:34:56 +0900 | [diff] [blame] | 241 | if (unlikely(ret < 0)) |
| 242 | goto error; |
| 243 | else if (ret) |
Takuya Yoshikawa | 92c1c1e | 2011-07-01 01:36:07 +0900 | [diff] [blame] | 244 | goto retry_walk; |
Roedel, Joerg | a78484c | 2011-04-20 15:33:16 +0200 | [diff] [blame] | 245 | |
Gui Jianfeng | f3b8c96 | 2010-05-05 09:09:21 +0800 | [diff] [blame] | 246 | mark_page_dirty(vcpu->kvm, table_gfn); |
Avi Kivity | 42bf3f0 | 2007-10-17 12:18:47 +0200 | [diff] [blame] | 247 | pte |= PT_ACCESSED_MASK; |
Avi Kivity | bf3f8e8 | 2007-02-19 14:37:46 +0200 | [diff] [blame] | 248 | } |
Avi Kivity | ac79c97 | 2007-01-05 16:36:40 -0800 | [diff] [blame] | 249 | |
Marcelo Tosatti | 7819026 | 2007-12-11 19:12:27 -0500 | [diff] [blame] | 250 | walker->ptes[walker->level - 1] = pte; |
| 251 | |
Yang, Wei Y | cd46868 | 2011-08-09 18:14:01 +0800 | [diff] [blame] | 252 | if (last_gpte) { |
Joerg Roedel | e04da98 | 2009-07-27 16:30:45 +0200 | [diff] [blame] | 253 | int lvl = walker->level; |
Joerg Roedel | 2329d46 | 2010-09-10 17:30:52 +0200 | [diff] [blame] | 254 | gpa_t real_gpa; |
| 255 | gfn_t gfn; |
Xiao Guangrong | 3377078 | 2010-09-28 17:03:14 +0800 | [diff] [blame] | 256 | u32 ac; |
Avi Kivity | 815af8d | 2007-01-05 16:36:44 -0800 | [diff] [blame] | 257 | |
Joerg Roedel | 2329d46 | 2010-09-10 17:30:52 +0200 | [diff] [blame] | 258 | gfn = gpte_to_gfn_lvl(pte, lvl); |
| 259 | gfn += (addr & PT_LVL_OFFSET_MASK(lvl)) >> PAGE_SHIFT; |
Joerg Roedel | e04da98 | 2009-07-27 16:30:45 +0200 | [diff] [blame] | 260 | |
| 261 | if (PTTYPE == 32 && |
| 262 | walker->level == PT_DIRECTORY_LEVEL && |
| 263 | is_cpuid_PSE36()) |
Joerg Roedel | 2329d46 | 2010-09-10 17:30:52 +0200 | [diff] [blame] | 264 | gfn += pse36_gfn_delta(pte); |
| 265 | |
Xiao Guangrong | 3377078 | 2010-09-28 17:03:14 +0800 | [diff] [blame] | 266 | ac = write_fault | fetch_fault | user_fault; |
Joerg Roedel | 2329d46 | 2010-09-10 17:30:52 +0200 | [diff] [blame] | 267 | |
| 268 | real_gpa = mmu->translate_gpa(vcpu, gfn_to_gpa(gfn), |
Xiao Guangrong | 3377078 | 2010-09-28 17:03:14 +0800 | [diff] [blame] | 269 | ac); |
Joerg Roedel | 2329d46 | 2010-09-10 17:30:52 +0200 | [diff] [blame] | 270 | if (real_gpa == UNMAPPED_GVA) |
| 271 | return 0; |
| 272 | |
| 273 | walker->gfn = real_gpa >> PAGE_SHIFT; |
Joerg Roedel | e04da98 | 2009-07-27 16:30:45 +0200 | [diff] [blame] | 274 | |
Avi Kivity | 815af8d | 2007-01-05 16:36:44 -0800 | [diff] [blame] | 275 | break; |
| 276 | } |
| 277 | |
Xiao Guangrong | 640d9b0 | 2011-07-12 03:24:39 +0800 | [diff] [blame] | 278 | pt_access &= FNAME(gpte_access)(vcpu, pte, false); |
Avi Kivity | ac79c97 | 2007-01-05 16:36:40 -0800 | [diff] [blame] | 279 | --walker->level; |
| 280 | } |
Avi Kivity | 42bf3f0 | 2007-10-17 12:18:47 +0200 | [diff] [blame] | 281 | |
Takuya Yoshikawa | 134291b | 2011-07-01 01:34:56 +0900 | [diff] [blame] | 282 | if (unlikely(eperm)) { |
| 283 | errcode |= PFERR_PRESENT_MASK; |
Avi Kivity | f59c1d2 | 2010-07-06 16:20:43 +0300 | [diff] [blame] | 284 | goto error; |
Takuya Yoshikawa | 134291b | 2011-07-01 01:34:56 +0900 | [diff] [blame] | 285 | } |
Avi Kivity | f59c1d2 | 2010-07-06 16:20:43 +0300 | [diff] [blame] | 286 | |
Avi Kivity | 781e074 | 2011-04-24 12:25:50 +0300 | [diff] [blame] | 287 | if (write_fault && unlikely(!is_dirty_gpte(pte))) { |
Roedel, Joerg | a78484c | 2011-04-20 15:33:16 +0200 | [diff] [blame] | 288 | int ret; |
Marcelo Tosatti | b3e4e63 | 2007-12-07 07:56:58 -0500 | [diff] [blame] | 289 | |
Avi Kivity | 0742017 | 2009-07-06 12:21:32 +0300 | [diff] [blame] | 290 | trace_kvm_mmu_set_dirty_bit(table_gfn, index, sizeof(pte)); |
Takuya Yoshikawa | c8cfbb5 | 2011-05-01 14:33:07 +0900 | [diff] [blame] | 291 | ret = FNAME(cmpxchg_gpte)(vcpu, mmu, ptep_user, index, |
| 292 | pte, pte|PT_DIRTY_MASK); |
Takuya Yoshikawa | 134291b | 2011-07-01 01:34:56 +0900 | [diff] [blame] | 293 | if (unlikely(ret < 0)) |
Roedel, Joerg | a78484c | 2011-04-20 15:33:16 +0200 | [diff] [blame] | 294 | goto error; |
Takuya Yoshikawa | 134291b | 2011-07-01 01:34:56 +0900 | [diff] [blame] | 295 | else if (ret) |
Takuya Yoshikawa | 92c1c1e | 2011-07-01 01:36:07 +0900 | [diff] [blame] | 296 | goto retry_walk; |
Roedel, Joerg | a78484c | 2011-04-20 15:33:16 +0200 | [diff] [blame] | 297 | |
Gui Jianfeng | f3b8c96 | 2010-05-05 09:09:21 +0800 | [diff] [blame] | 298 | mark_page_dirty(vcpu->kvm, table_gfn); |
Avi Kivity | 42bf3f0 | 2007-10-17 12:18:47 +0200 | [diff] [blame] | 299 | pte |= PT_DIRTY_MASK; |
Marcelo Tosatti | 7819026 | 2007-12-11 19:12:27 -0500 | [diff] [blame] | 300 | walker->ptes[walker->level - 1] = pte; |
Avi Kivity | 42bf3f0 | 2007-10-17 12:18:47 +0200 | [diff] [blame] | 301 | } |
| 302 | |
Avi Kivity | fe135d2 | 2007-12-09 16:15:46 +0200 | [diff] [blame] | 303 | walker->pt_access = pt_access; |
| 304 | walker->pte_access = pte_access; |
| 305 | pgprintk("%s: pte %llx pte_access %x pt_access %x\n", |
Gui Jianfeng | 518c5a0 | 2010-05-05 09:58:33 +0800 | [diff] [blame] | 306 | __func__, (u64)pte, pte_access, pt_access); |
Avi Kivity | 7993ba4 | 2007-01-26 00:56:41 -0800 | [diff] [blame] | 307 | return 1; |
| 308 | |
Avi Kivity | f59c1d2 | 2010-07-06 16:20:43 +0300 | [diff] [blame] | 309 | error: |
Takuya Yoshikawa | 134291b | 2011-07-01 01:34:56 +0900 | [diff] [blame] | 310 | errcode |= write_fault | user_fault; |
Yang, Wei Y | e57d4a3 | 2011-06-03 11:14:16 +0800 | [diff] [blame] | 311 | if (fetch_fault && (mmu->nx || |
| 312 | kvm_read_cr4_bits(vcpu, X86_CR4_SMEP))) |
Takuya Yoshikawa | 134291b | 2011-07-01 01:34:56 +0900 | [diff] [blame] | 313 | errcode |= PFERR_FETCH_MASK; |
Joerg Roedel | 8df25a3 | 2010-09-10 17:30:46 +0200 | [diff] [blame] | 314 | |
Takuya Yoshikawa | 134291b | 2011-07-01 01:34:56 +0900 | [diff] [blame] | 315 | walker->fault.vector = PF_VECTOR; |
| 316 | walker->fault.error_code_valid = true; |
| 317 | walker->fault.error_code = errcode; |
Avi Kivity | 6389ee9 | 2010-11-29 16:12:30 +0200 | [diff] [blame] | 318 | walker->fault.address = addr; |
| 319 | walker->fault.nested_page_fault = mmu != vcpu->arch.walk_mmu; |
Joerg Roedel | 8df25a3 | 2010-09-10 17:30:46 +0200 | [diff] [blame] | 320 | |
Avi Kivity | 8c28d03 | 2010-11-22 17:53:27 +0200 | [diff] [blame] | 321 | trace_kvm_mmu_walker_error(walker->fault.error_code); |
Shaohua Li | fe55188 | 2007-07-23 14:51:39 +0800 | [diff] [blame] | 322 | return 0; |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 323 | } |
| 324 | |
Joerg Roedel | 1e301fe | 2010-09-10 17:30:47 +0200 | [diff] [blame] | 325 | static int FNAME(walk_addr)(struct guest_walker *walker, |
Xiao Guangrong | 3377078 | 2010-09-28 17:03:14 +0800 | [diff] [blame] | 326 | struct kvm_vcpu *vcpu, gva_t addr, u32 access) |
Joerg Roedel | 1e301fe | 2010-09-10 17:30:47 +0200 | [diff] [blame] | 327 | { |
| 328 | return FNAME(walk_addr_generic)(walker, vcpu, &vcpu->arch.mmu, addr, |
Xiao Guangrong | 3377078 | 2010-09-28 17:03:14 +0800 | [diff] [blame] | 329 | access); |
Joerg Roedel | 1e301fe | 2010-09-10 17:30:47 +0200 | [diff] [blame] | 330 | } |
| 331 | |
Joerg Roedel | 6539e73 | 2010-09-10 17:30:50 +0200 | [diff] [blame] | 332 | static int FNAME(walk_addr_nested)(struct guest_walker *walker, |
| 333 | struct kvm_vcpu *vcpu, gva_t addr, |
Xiao Guangrong | 3377078 | 2010-09-28 17:03:14 +0800 | [diff] [blame] | 334 | u32 access) |
Joerg Roedel | 6539e73 | 2010-09-10 17:30:50 +0200 | [diff] [blame] | 335 | { |
| 336 | return FNAME(walk_addr_generic)(walker, vcpu, &vcpu->arch.nested_mmu, |
Xiao Guangrong | 3377078 | 2010-09-28 17:03:14 +0800 | [diff] [blame] | 337 | addr, access); |
Joerg Roedel | 6539e73 | 2010-09-10 17:30:50 +0200 | [diff] [blame] | 338 | } |
| 339 | |
Xiao Guangrong | 407c61c | 2010-11-23 11:08:42 +0800 | [diff] [blame] | 340 | static bool FNAME(prefetch_invalid_gpte)(struct kvm_vcpu *vcpu, |
| 341 | struct kvm_mmu_page *sp, u64 *spte, |
| 342 | pt_element_t gpte) |
| 343 | { |
Xiao Guangrong | 407c61c | 2010-11-23 11:08:42 +0800 | [diff] [blame] | 344 | if (is_rsvd_bits_set(&vcpu->arch.mmu, gpte, PT_PAGE_TABLE_LEVEL)) |
| 345 | goto no_present; |
| 346 | |
Xiao Guangrong | c370795 | 2011-07-12 03:28:04 +0800 | [diff] [blame] | 347 | if (!is_present_gpte(gpte)) |
Xiao Guangrong | 407c61c | 2010-11-23 11:08:42 +0800 | [diff] [blame] | 348 | goto no_present; |
Xiao Guangrong | 407c61c | 2010-11-23 11:08:42 +0800 | [diff] [blame] | 349 | |
| 350 | if (!(gpte & PT_ACCESSED_MASK)) |
| 351 | goto no_present; |
| 352 | |
| 353 | return false; |
| 354 | |
| 355 | no_present: |
Xiao Guangrong | c370795 | 2011-07-12 03:28:04 +0800 | [diff] [blame] | 356 | drop_spte(vcpu->kvm, spte); |
Xiao Guangrong | 407c61c | 2010-11-23 11:08:42 +0800 | [diff] [blame] | 357 | return true; |
| 358 | } |
| 359 | |
Xiao Guangrong | ac3cd03 | 2010-06-11 21:28:14 +0800 | [diff] [blame] | 360 | static void FNAME(update_pte)(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp, |
Xiao Guangrong | 7c56252 | 2011-03-28 10:29:27 +0800 | [diff] [blame] | 361 | u64 *spte, const void *pte) |
Avi Kivity | 0028425 | 2007-05-01 16:53:31 +0300 | [diff] [blame] | 362 | { |
| 363 | pt_element_t gpte; |
Avi Kivity | 41074d0 | 2007-12-09 17:00:02 +0200 | [diff] [blame] | 364 | unsigned pte_access; |
Anthony Liguori | 35149e2 | 2008-04-02 14:46:56 -0500 | [diff] [blame] | 365 | pfn_t pfn; |
Avi Kivity | 0028425 | 2007-05-01 16:53:31 +0300 | [diff] [blame] | 366 | |
Avi Kivity | 0028425 | 2007-05-01 16:53:31 +0300 | [diff] [blame] | 367 | gpte = *(const pt_element_t *)pte; |
Xiao Guangrong | 407c61c | 2010-11-23 11:08:42 +0800 | [diff] [blame] | 368 | if (FNAME(prefetch_invalid_gpte)(vcpu, sp, spte, gpte)) |
Avi Kivity | c7addb9 | 2007-09-16 18:58:32 +0200 | [diff] [blame] | 369 | return; |
Xiao Guangrong | 407c61c | 2010-11-23 11:08:42 +0800 | [diff] [blame] | 370 | |
Harvey Harrison | b8688d5 | 2008-03-03 12:59:56 -0800 | [diff] [blame] | 371 | pgprintk("%s: gpte %llx spte %p\n", __func__, (u64)gpte, spte); |
Xiao Guangrong | 640d9b0 | 2011-07-12 03:24:39 +0800 | [diff] [blame] | 372 | pte_access = sp->role.access & FNAME(gpte_access)(vcpu, gpte, true); |
Xiao Guangrong | 0f53b5b | 2011-03-09 15:43:51 +0800 | [diff] [blame] | 373 | pfn = gfn_to_pfn_atomic(vcpu->kvm, gpte_to_gfn(gpte)); |
Xiao Guangrong | d7c5520 | 2011-07-12 03:29:38 +0800 | [diff] [blame] | 374 | if (mmu_invalid_pfn(pfn)) { |
Xiao Guangrong | 0f53b5b | 2011-03-09 15:43:51 +0800 | [diff] [blame] | 375 | kvm_release_pfn_clean(pfn); |
Avi Kivity | d7824ff | 2007-12-30 12:29:05 +0200 | [diff] [blame] | 376 | return; |
Xiao Guangrong | 0f53b5b | 2011-03-09 15:43:51 +0800 | [diff] [blame] | 377 | } |
Xiao Guangrong | 0f53b5b | 2011-03-09 15:43:51 +0800 | [diff] [blame] | 378 | |
Izik Eidus | 1403283 | 2009-09-23 21:47:17 +0300 | [diff] [blame] | 379 | /* |
Lucas De Marchi | 0d2eb44 | 2011-03-17 16:24:16 -0300 | [diff] [blame] | 380 | * we call mmu_set_spte() with host_writable = true because that |
Izik Eidus | 1403283 | 2009-09-23 21:47:17 +0300 | [diff] [blame] | 381 | * vcpu->arch.update_pte.pfn was fetched from get_user_pages(write = 1). |
| 382 | */ |
Xiao Guangrong | ac3cd03 | 2010-06-11 21:28:14 +0800 | [diff] [blame] | 383 | mmu_set_spte(vcpu, spte, sp->role.access, pte_access, 0, 0, |
Xiao Guangrong | 640d9b0 | 2011-07-12 03:24:39 +0800 | [diff] [blame] | 384 | NULL, PT_PAGE_TABLE_LEVEL, |
Izik Eidus | 1403283 | 2009-09-23 21:47:17 +0300 | [diff] [blame] | 385 | gpte_to_gfn(gpte), pfn, true, true); |
Avi Kivity | 0028425 | 2007-05-01 16:53:31 +0300 | [diff] [blame] | 386 | } |
| 387 | |
Avi Kivity | 39c8c67 | 2010-07-13 14:27:08 +0300 | [diff] [blame] | 388 | static bool FNAME(gpte_changed)(struct kvm_vcpu *vcpu, |
| 389 | struct guest_walker *gw, int level) |
| 390 | { |
Avi Kivity | 39c8c67 | 2010-07-13 14:27:08 +0300 | [diff] [blame] | 391 | pt_element_t curr_pte; |
Xiao Guangrong | 189be38 | 2010-08-22 19:13:33 +0800 | [diff] [blame] | 392 | gpa_t base_gpa, pte_gpa = gw->pte_gpa[level - 1]; |
| 393 | u64 mask; |
| 394 | int r, index; |
Avi Kivity | 39c8c67 | 2010-07-13 14:27:08 +0300 | [diff] [blame] | 395 | |
Xiao Guangrong | 189be38 | 2010-08-22 19:13:33 +0800 | [diff] [blame] | 396 | if (level == PT_PAGE_TABLE_LEVEL) { |
| 397 | mask = PTE_PREFETCH_NUM * sizeof(pt_element_t) - 1; |
| 398 | base_gpa = pte_gpa & ~mask; |
| 399 | index = (pte_gpa - base_gpa) / sizeof(pt_element_t); |
| 400 | |
| 401 | r = kvm_read_guest_atomic(vcpu->kvm, base_gpa, |
| 402 | gw->prefetch_ptes, sizeof(gw->prefetch_ptes)); |
| 403 | curr_pte = gw->prefetch_ptes[index]; |
| 404 | } else |
| 405 | r = kvm_read_guest_atomic(vcpu->kvm, pte_gpa, |
Avi Kivity | 39c8c67 | 2010-07-13 14:27:08 +0300 | [diff] [blame] | 406 | &curr_pte, sizeof(curr_pte)); |
Xiao Guangrong | 189be38 | 2010-08-22 19:13:33 +0800 | [diff] [blame] | 407 | |
Avi Kivity | 39c8c67 | 2010-07-13 14:27:08 +0300 | [diff] [blame] | 408 | return r || curr_pte != gw->ptes[level - 1]; |
| 409 | } |
| 410 | |
Xiao Guangrong | 189be38 | 2010-08-22 19:13:33 +0800 | [diff] [blame] | 411 | static void FNAME(pte_prefetch)(struct kvm_vcpu *vcpu, struct guest_walker *gw, |
| 412 | u64 *sptep) |
Xiao Guangrong | 957ed9e | 2010-08-22 19:12:48 +0800 | [diff] [blame] | 413 | { |
| 414 | struct kvm_mmu_page *sp; |
Xiao Guangrong | 189be38 | 2010-08-22 19:13:33 +0800 | [diff] [blame] | 415 | pt_element_t *gptep = gw->prefetch_ptes; |
Xiao Guangrong | 957ed9e | 2010-08-22 19:12:48 +0800 | [diff] [blame] | 416 | u64 *spte; |
Xiao Guangrong | 189be38 | 2010-08-22 19:13:33 +0800 | [diff] [blame] | 417 | int i; |
Xiao Guangrong | 957ed9e | 2010-08-22 19:12:48 +0800 | [diff] [blame] | 418 | |
| 419 | sp = page_header(__pa(sptep)); |
| 420 | |
| 421 | if (sp->role.level > PT_PAGE_TABLE_LEVEL) |
| 422 | return; |
| 423 | |
| 424 | if (sp->role.direct) |
| 425 | return __direct_pte_prefetch(vcpu, sp, sptep); |
| 426 | |
| 427 | i = (sptep - sp->spt) & ~(PTE_PREFETCH_NUM - 1); |
Xiao Guangrong | 957ed9e | 2010-08-22 19:12:48 +0800 | [diff] [blame] | 428 | spte = sp->spt + i; |
| 429 | |
| 430 | for (i = 0; i < PTE_PREFETCH_NUM; i++, spte++) { |
| 431 | pt_element_t gpte; |
| 432 | unsigned pte_access; |
| 433 | gfn_t gfn; |
| 434 | pfn_t pfn; |
Xiao Guangrong | 957ed9e | 2010-08-22 19:12:48 +0800 | [diff] [blame] | 435 | |
| 436 | if (spte == sptep) |
| 437 | continue; |
| 438 | |
Xiao Guangrong | c370795 | 2011-07-12 03:28:04 +0800 | [diff] [blame] | 439 | if (is_shadow_present_pte(*spte)) |
Xiao Guangrong | 957ed9e | 2010-08-22 19:12:48 +0800 | [diff] [blame] | 440 | continue; |
| 441 | |
| 442 | gpte = gptep[i]; |
| 443 | |
Xiao Guangrong | 407c61c | 2010-11-23 11:08:42 +0800 | [diff] [blame] | 444 | if (FNAME(prefetch_invalid_gpte)(vcpu, sp, spte, gpte)) |
Xiao Guangrong | 957ed9e | 2010-08-22 19:12:48 +0800 | [diff] [blame] | 445 | continue; |
| 446 | |
Xiao Guangrong | 640d9b0 | 2011-07-12 03:24:39 +0800 | [diff] [blame] | 447 | pte_access = sp->role.access & FNAME(gpte_access)(vcpu, gpte, |
| 448 | true); |
Xiao Guangrong | 957ed9e | 2010-08-22 19:12:48 +0800 | [diff] [blame] | 449 | gfn = gpte_to_gfn(gpte); |
Xiao Guangrong | 957ed9e | 2010-08-22 19:12:48 +0800 | [diff] [blame] | 450 | pfn = pte_prefetch_gfn_to_pfn(vcpu, gfn, |
Xiao Guangrong | 640d9b0 | 2011-07-12 03:24:39 +0800 | [diff] [blame] | 451 | pte_access & ACC_WRITE_MASK); |
Xiao Guangrong | d7c5520 | 2011-07-12 03:29:38 +0800 | [diff] [blame] | 452 | if (mmu_invalid_pfn(pfn)) { |
Xiao Guangrong | 957ed9e | 2010-08-22 19:12:48 +0800 | [diff] [blame] | 453 | kvm_release_pfn_clean(pfn); |
| 454 | break; |
| 455 | } |
| 456 | |
| 457 | mmu_set_spte(vcpu, spte, sp->role.access, pte_access, 0, 0, |
Xiao Guangrong | 640d9b0 | 2011-07-12 03:24:39 +0800 | [diff] [blame] | 458 | NULL, PT_PAGE_TABLE_LEVEL, gfn, |
Xiao Guangrong | 957ed9e | 2010-08-22 19:12:48 +0800 | [diff] [blame] | 459 | pfn, true, true); |
| 460 | } |
| 461 | } |
| 462 | |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 463 | /* |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 464 | * Fetch a shadow pte for a specific level in the paging hierarchy. |
| 465 | */ |
| 466 | static u64 *FNAME(fetch)(struct kvm_vcpu *vcpu, gva_t addr, |
Avi Kivity | e7a04c9 | 2008-12-25 15:10:50 +0200 | [diff] [blame] | 467 | struct guest_walker *gw, |
Joerg Roedel | 7e4e405 | 2009-07-27 16:30:46 +0200 | [diff] [blame] | 468 | int user_fault, int write_fault, int hlevel, |
Xiao Guangrong | b90a0e6 | 2011-07-12 03:25:56 +0800 | [diff] [blame] | 469 | int *emulate, pfn_t pfn, bool map_writable, |
Xiao Guangrong | fb67e14 | 2010-12-07 10:35:25 +0800 | [diff] [blame] | 470 | bool prefault) |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 471 | { |
Avi Kivity | e7a04c9 | 2008-12-25 15:10:50 +0200 | [diff] [blame] | 472 | unsigned access = gw->pt_access; |
Avi Kivity | 5991b33 | 2010-07-13 14:27:10 +0300 | [diff] [blame] | 473 | struct kvm_mmu_page *sp = NULL; |
Avi Kivity | 5991b33 | 2010-07-13 14:27:10 +0300 | [diff] [blame] | 474 | int top_level; |
Xiao Guangrong | 84754cd | 2010-06-30 16:05:00 +0800 | [diff] [blame] | 475 | unsigned direct_access; |
Avi Kivity | 24157aa | 2010-07-13 14:27:11 +0300 | [diff] [blame] | 476 | struct kvm_shadow_walk_iterator it; |
Avi Kivity | ac79c97 | 2007-01-05 16:36:40 -0800 | [diff] [blame] | 477 | |
Avi Kivity | 43a3795 | 2009-06-10 14:12:05 +0300 | [diff] [blame] | 478 | if (!is_present_gpte(gw->ptes[gw->level - 1])) |
Avi Kivity | ac79c97 | 2007-01-05 16:36:40 -0800 | [diff] [blame] | 479 | return NULL; |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 480 | |
Xiao Guangrong | b36c7a7 | 2011-07-12 03:25:19 +0800 | [diff] [blame] | 481 | direct_access = gw->pte_access; |
Xiao Guangrong | 84754cd | 2010-06-30 16:05:00 +0800 | [diff] [blame] | 482 | |
Avi Kivity | 5991b33 | 2010-07-13 14:27:10 +0300 | [diff] [blame] | 483 | top_level = vcpu->arch.mmu.root_level; |
| 484 | if (top_level == PT32E_ROOT_LEVEL) |
| 485 | top_level = PT32_ROOT_LEVEL; |
| 486 | /* |
| 487 | * Verify that the top-level gpte is still there. Since the page |
| 488 | * is a root page, it is either write protected (and cannot be |
| 489 | * changed from now on) or it is invalid (in which case, we don't |
| 490 | * really care if it changes underneath us after this point). |
| 491 | */ |
| 492 | if (FNAME(gpte_changed)(vcpu, gw, top_level)) |
| 493 | goto out_gpte_changed; |
| 494 | |
Avi Kivity | 24157aa | 2010-07-13 14:27:11 +0300 | [diff] [blame] | 495 | for (shadow_walk_init(&it, vcpu, addr); |
| 496 | shadow_walk_okay(&it) && it.level > gw->level; |
| 497 | shadow_walk_next(&it)) { |
Avi Kivity | 0b3c933 | 2010-07-13 14:27:09 +0300 | [diff] [blame] | 498 | gfn_t table_gfn; |
| 499 | |
Xiao Guangrong | a30f47c | 2011-09-22 16:58:36 +0800 | [diff] [blame] | 500 | clear_sp_write_flooding_count(it.sptep); |
Avi Kivity | 24157aa | 2010-07-13 14:27:11 +0300 | [diff] [blame] | 501 | drop_large_spte(vcpu, it.sptep); |
Avi Kivity | e7a04c9 | 2008-12-25 15:10:50 +0200 | [diff] [blame] | 502 | |
Avi Kivity | 5991b33 | 2010-07-13 14:27:10 +0300 | [diff] [blame] | 503 | sp = NULL; |
Avi Kivity | 24157aa | 2010-07-13 14:27:11 +0300 | [diff] [blame] | 504 | if (!is_shadow_present_pte(*it.sptep)) { |
| 505 | table_gfn = gw->table_gfn[it.level - 2]; |
| 506 | sp = kvm_mmu_get_page(vcpu, table_gfn, addr, it.level-1, |
| 507 | false, access, it.sptep); |
Avi Kivity | 5991b33 | 2010-07-13 14:27:10 +0300 | [diff] [blame] | 508 | } |
Avi Kivity | 0b3c933 | 2010-07-13 14:27:09 +0300 | [diff] [blame] | 509 | |
| 510 | /* |
| 511 | * Verify that the gpte in the page we've just write |
| 512 | * protected is still there. |
| 513 | */ |
Avi Kivity | 24157aa | 2010-07-13 14:27:11 +0300 | [diff] [blame] | 514 | if (FNAME(gpte_changed)(vcpu, gw, it.level - 1)) |
Avi Kivity | 0b3c933 | 2010-07-13 14:27:09 +0300 | [diff] [blame] | 515 | goto out_gpte_changed; |
Avi Kivity | e7a04c9 | 2008-12-25 15:10:50 +0200 | [diff] [blame] | 516 | |
Avi Kivity | 5991b33 | 2010-07-13 14:27:10 +0300 | [diff] [blame] | 517 | if (sp) |
Avi Kivity | 24157aa | 2010-07-13 14:27:11 +0300 | [diff] [blame] | 518 | link_shadow_page(it.sptep, sp); |
Avi Kivity | e7a04c9 | 2008-12-25 15:10:50 +0200 | [diff] [blame] | 519 | } |
| 520 | |
Avi Kivity | 0b3c933 | 2010-07-13 14:27:09 +0300 | [diff] [blame] | 521 | for (; |
Avi Kivity | 24157aa | 2010-07-13 14:27:11 +0300 | [diff] [blame] | 522 | shadow_walk_okay(&it) && it.level > hlevel; |
| 523 | shadow_walk_next(&it)) { |
Avi Kivity | 0b3c933 | 2010-07-13 14:27:09 +0300 | [diff] [blame] | 524 | gfn_t direct_gfn; |
| 525 | |
Xiao Guangrong | a30f47c | 2011-09-22 16:58:36 +0800 | [diff] [blame] | 526 | clear_sp_write_flooding_count(it.sptep); |
Avi Kivity | 24157aa | 2010-07-13 14:27:11 +0300 | [diff] [blame] | 527 | validate_direct_spte(vcpu, it.sptep, direct_access); |
Avi Kivity | 0b3c933 | 2010-07-13 14:27:09 +0300 | [diff] [blame] | 528 | |
Avi Kivity | 24157aa | 2010-07-13 14:27:11 +0300 | [diff] [blame] | 529 | drop_large_spte(vcpu, it.sptep); |
Avi Kivity | 0b3c933 | 2010-07-13 14:27:09 +0300 | [diff] [blame] | 530 | |
Avi Kivity | 24157aa | 2010-07-13 14:27:11 +0300 | [diff] [blame] | 531 | if (is_shadow_present_pte(*it.sptep)) |
Avi Kivity | 0b3c933 | 2010-07-13 14:27:09 +0300 | [diff] [blame] | 532 | continue; |
| 533 | |
Avi Kivity | 24157aa | 2010-07-13 14:27:11 +0300 | [diff] [blame] | 534 | direct_gfn = gw->gfn & ~(KVM_PAGES_PER_HPAGE(it.level) - 1); |
Avi Kivity | 0b3c933 | 2010-07-13 14:27:09 +0300 | [diff] [blame] | 535 | |
Avi Kivity | 24157aa | 2010-07-13 14:27:11 +0300 | [diff] [blame] | 536 | sp = kvm_mmu_get_page(vcpu, direct_gfn, addr, it.level-1, |
| 537 | true, direct_access, it.sptep); |
| 538 | link_shadow_page(it.sptep, sp); |
Avi Kivity | 0b3c933 | 2010-07-13 14:27:09 +0300 | [diff] [blame] | 539 | } |
| 540 | |
Xiao Guangrong | a30f47c | 2011-09-22 16:58:36 +0800 | [diff] [blame] | 541 | clear_sp_write_flooding_count(it.sptep); |
Xiao Guangrong | b36c7a7 | 2011-07-12 03:25:19 +0800 | [diff] [blame] | 542 | mmu_set_spte(vcpu, it.sptep, access, gw->pte_access, |
Xiao Guangrong | b90a0e6 | 2011-07-12 03:25:56 +0800 | [diff] [blame] | 543 | user_fault, write_fault, emulate, it.level, |
Xiao Guangrong | fb67e14 | 2010-12-07 10:35:25 +0800 | [diff] [blame] | 544 | gw->gfn, pfn, prefault, map_writable); |
Xiao Guangrong | 189be38 | 2010-08-22 19:13:33 +0800 | [diff] [blame] | 545 | FNAME(pte_prefetch)(vcpu, gw, it.sptep); |
Avi Kivity | 0b3c933 | 2010-07-13 14:27:09 +0300 | [diff] [blame] | 546 | |
Avi Kivity | 24157aa | 2010-07-13 14:27:11 +0300 | [diff] [blame] | 547 | return it.sptep; |
Avi Kivity | 0b3c933 | 2010-07-13 14:27:09 +0300 | [diff] [blame] | 548 | |
| 549 | out_gpte_changed: |
Avi Kivity | 5991b33 | 2010-07-13 14:27:10 +0300 | [diff] [blame] | 550 | if (sp) |
Avi Kivity | 24157aa | 2010-07-13 14:27:11 +0300 | [diff] [blame] | 551 | kvm_mmu_put_page(sp, it.sptep); |
Avi Kivity | 0b3c933 | 2010-07-13 14:27:09 +0300 | [diff] [blame] | 552 | kvm_release_pfn_clean(pfn); |
| 553 | return NULL; |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 554 | } |
| 555 | |
| 556 | /* |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 557 | * Page fault handler. There are several causes for a page fault: |
| 558 | * - there is no shadow pte for the guest pte |
| 559 | * - write access through a shadow pte marked read only so that we can set |
| 560 | * the dirty bit |
| 561 | * - write access to a shadow pte marked read only so we can update the page |
| 562 | * dirty bitmap, when userspace requests it |
| 563 | * - mmio access; in this case we will never install a present shadow pte |
| 564 | * - normal guest page fault due to the guest pte marked not present, not |
| 565 | * writable, or not executable |
| 566 | * |
Avi Kivity | e2dec93 | 2007-01-05 16:36:54 -0800 | [diff] [blame] | 567 | * Returns: 1 if we need to emulate the instruction, 0 otherwise, or |
| 568 | * a negative value on error. |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 569 | */ |
Gleb Natapov | 56028d0 | 2010-10-17 18:13:42 +0200 | [diff] [blame] | 570 | static int FNAME(page_fault)(struct kvm_vcpu *vcpu, gva_t addr, u32 error_code, |
Xiao Guangrong | 78b2c54 | 2010-12-07 10:48:06 +0800 | [diff] [blame] | 571 | bool prefault) |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 572 | { |
| 573 | int write_fault = error_code & PFERR_WRITE_MASK; |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 574 | int user_fault = error_code & PFERR_USER_MASK; |
| 575 | struct guest_walker walker; |
Avi Kivity | d555c33 | 2009-06-10 14:24:23 +0300 | [diff] [blame] | 576 | u64 *sptep; |
Xiao Guangrong | b90a0e6 | 2011-07-12 03:25:56 +0800 | [diff] [blame] | 577 | int emulate = 0; |
Avi Kivity | e2dec93 | 2007-01-05 16:36:54 -0800 | [diff] [blame] | 578 | int r; |
Anthony Liguori | 35149e2 | 2008-04-02 14:46:56 -0500 | [diff] [blame] | 579 | pfn_t pfn; |
Joerg Roedel | 7e4e405 | 2009-07-27 16:30:46 +0200 | [diff] [blame] | 580 | int level = PT_PAGE_TABLE_LEVEL; |
Andrea Arcangeli | 936a5fe | 2011-01-13 15:46:48 -0800 | [diff] [blame] | 581 | int force_pt_level; |
Andrea Arcangeli | e930bff | 2008-07-25 16:24:52 +0200 | [diff] [blame] | 582 | unsigned long mmu_seq; |
Marcelo Tosatti | 612819c | 2010-10-22 14:18:18 -0200 | [diff] [blame] | 583 | bool map_writable; |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 584 | |
Harvey Harrison | b8688d5 | 2008-03-03 12:59:56 -0800 | [diff] [blame] | 585 | pgprintk("%s: addr %lx err %x\n", __func__, addr, error_code); |
Avi Kivity | 714b93d | 2007-01-05 16:36:53 -0800 | [diff] [blame] | 586 | |
Xiao Guangrong | ce88dec | 2011-07-12 03:33:44 +0800 | [diff] [blame] | 587 | if (unlikely(error_code & PFERR_RSVD_MASK)) |
| 588 | return handle_mmio_page_fault(vcpu, addr, error_code, |
| 589 | mmu_is_nested(vcpu)); |
| 590 | |
Avi Kivity | e2dec93 | 2007-01-05 16:36:54 -0800 | [diff] [blame] | 591 | r = mmu_topup_memory_caches(vcpu); |
| 592 | if (r) |
| 593 | return r; |
Avi Kivity | 714b93d | 2007-01-05 16:36:53 -0800 | [diff] [blame] | 594 | |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 595 | /* |
Eddie Dong | a8b876b | 2009-03-26 15:28:40 +0800 | [diff] [blame] | 596 | * Look up the guest pte for the faulting address. |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 597 | */ |
Xiao Guangrong | 3377078 | 2010-09-28 17:03:14 +0800 | [diff] [blame] | 598 | r = FNAME(walk_addr)(&walker, vcpu, addr, error_code); |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 599 | |
| 600 | /* |
| 601 | * The page is not mapped by the guest. Let the guest handle it. |
| 602 | */ |
Avi Kivity | 7993ba4 | 2007-01-26 00:56:41 -0800 | [diff] [blame] | 603 | if (!r) { |
Harvey Harrison | b8688d5 | 2008-03-03 12:59:56 -0800 | [diff] [blame] | 604 | pgprintk("%s: guest page fault\n", __func__); |
Xiao Guangrong | a30f47c | 2011-09-22 16:58:36 +0800 | [diff] [blame] | 605 | if (!prefault) |
Xiao Guangrong | fb67e14 | 2010-12-07 10:35:25 +0800 | [diff] [blame] | 606 | inject_page_fault(vcpu, &walker.fault); |
Xiao Guangrong | a30f47c | 2011-09-22 16:58:36 +0800 | [diff] [blame] | 607 | |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 608 | return 0; |
| 609 | } |
| 610 | |
Andrea Arcangeli | 936a5fe | 2011-01-13 15:46:48 -0800 | [diff] [blame] | 611 | if (walker.level >= PT_DIRECTORY_LEVEL) |
| 612 | force_pt_level = mapping_level_dirty_bitmap(vcpu, walker.gfn); |
| 613 | else |
| 614 | force_pt_level = 1; |
| 615 | if (!force_pt_level) { |
Joerg Roedel | 7e4e405 | 2009-07-27 16:30:46 +0200 | [diff] [blame] | 616 | level = min(walker.level, mapping_level(vcpu, walker.gfn)); |
| 617 | walker.gfn = walker.gfn & ~(KVM_PAGES_PER_HPAGE(level) - 1); |
Marcelo Tosatti | 05da455 | 2008-02-23 11:44:30 -0300 | [diff] [blame] | 618 | } |
Joerg Roedel | 7e4e405 | 2009-07-27 16:30:46 +0200 | [diff] [blame] | 619 | |
Andrea Arcangeli | e930bff | 2008-07-25 16:24:52 +0200 | [diff] [blame] | 620 | mmu_seq = vcpu->kvm->mmu_notifier_seq; |
Marcelo Tosatti | 4c2155c | 2008-09-16 20:54:47 -0300 | [diff] [blame] | 621 | smp_rmb(); |
Gleb Natapov | af585b9 | 2010-10-14 11:22:46 +0200 | [diff] [blame] | 622 | |
Xiao Guangrong | 78b2c54 | 2010-12-07 10:48:06 +0800 | [diff] [blame] | 623 | if (try_async_pf(vcpu, prefault, walker.gfn, addr, &pfn, write_fault, |
Marcelo Tosatti | 612819c | 2010-10-22 14:18:18 -0200 | [diff] [blame] | 624 | &map_writable)) |
Gleb Natapov | af585b9 | 2010-10-14 11:22:46 +0200 | [diff] [blame] | 625 | return 0; |
Avi Kivity | d7824ff | 2007-12-30 12:29:05 +0200 | [diff] [blame] | 626 | |
Xiao Guangrong | d7c5520 | 2011-07-12 03:29:38 +0800 | [diff] [blame] | 627 | if (handle_abnormal_pfn(vcpu, mmu_is_nested(vcpu) ? 0 : addr, |
| 628 | walker.gfn, pfn, walker.pte_access, &r)) |
| 629 | return r; |
| 630 | |
Marcelo Tosatti | aaee2c9 | 2007-12-20 19:18:26 -0500 | [diff] [blame] | 631 | spin_lock(&vcpu->kvm->mmu_lock); |
Andrea Arcangeli | e930bff | 2008-07-25 16:24:52 +0200 | [diff] [blame] | 632 | if (mmu_notifier_retry(vcpu, mmu_seq)) |
| 633 | goto out_unlock; |
Xiao Guangrong | bc32ce2 | 2010-08-28 19:22:46 +0800 | [diff] [blame] | 634 | |
Xiao Guangrong | 0375f7f | 2011-11-28 20:41:00 +0800 | [diff] [blame] | 635 | kvm_mmu_audit(vcpu, AUDIT_PRE_PAGE_FAULT); |
Avi Kivity | eb787d1 | 2007-12-31 15:27:49 +0200 | [diff] [blame] | 636 | kvm_mmu_free_some_pages(vcpu); |
Andrea Arcangeli | 936a5fe | 2011-01-13 15:46:48 -0800 | [diff] [blame] | 637 | if (!force_pt_level) |
| 638 | transparent_hugepage_adjust(vcpu, &walker.gfn, &pfn, &level); |
Avi Kivity | d555c33 | 2009-06-10 14:24:23 +0300 | [diff] [blame] | 639 | sptep = FNAME(fetch)(vcpu, addr, &walker, user_fault, write_fault, |
Xiao Guangrong | b90a0e6 | 2011-07-12 03:25:56 +0800 | [diff] [blame] | 640 | level, &emulate, pfn, map_writable, prefault); |
Andi Kleen | a24e809 | 2010-06-10 13:10:55 +0200 | [diff] [blame] | 641 | (void)sptep; |
Xiao Guangrong | b90a0e6 | 2011-07-12 03:25:56 +0800 | [diff] [blame] | 642 | pgprintk("%s: shadow pte %p %llx emulate %d\n", __func__, |
| 643 | sptep, *sptep, emulate); |
Avi Kivity | cea0f0e | 2007-01-05 16:36:43 -0800 | [diff] [blame] | 644 | |
Avi Kivity | 1165f5f | 2007-04-19 17:27:43 +0300 | [diff] [blame] | 645 | ++vcpu->stat.pf_fixed; |
Xiao Guangrong | 0375f7f | 2011-11-28 20:41:00 +0800 | [diff] [blame] | 646 | kvm_mmu_audit(vcpu, AUDIT_POST_PAGE_FAULT); |
Marcelo Tosatti | aaee2c9 | 2007-12-20 19:18:26 -0500 | [diff] [blame] | 647 | spin_unlock(&vcpu->kvm->mmu_lock); |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 648 | |
Xiao Guangrong | b90a0e6 | 2011-07-12 03:25:56 +0800 | [diff] [blame] | 649 | return emulate; |
Andrea Arcangeli | e930bff | 2008-07-25 16:24:52 +0200 | [diff] [blame] | 650 | |
| 651 | out_unlock: |
| 652 | spin_unlock(&vcpu->kvm->mmu_lock); |
| 653 | kvm_release_pfn_clean(pfn); |
| 654 | return 0; |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 655 | } |
| 656 | |
Xiao Guangrong | 505aef8 | 2011-09-22 16:56:06 +0800 | [diff] [blame] | 657 | static gpa_t FNAME(get_level1_sp_gpa)(struct kvm_mmu_page *sp) |
| 658 | { |
| 659 | int offset = 0; |
| 660 | |
| 661 | WARN_ON(sp->role.level != 1); |
| 662 | |
| 663 | if (PTTYPE == 32) |
| 664 | offset = sp->role.quadrant << PT64_LEVEL_BITS; |
| 665 | |
| 666 | return gfn_to_gpa(sp->gfn) + offset * sizeof(pt_element_t); |
| 667 | } |
| 668 | |
Marcelo Tosatti | a705289 | 2008-09-23 13:18:35 -0300 | [diff] [blame] | 669 | static void FNAME(invlpg)(struct kvm_vcpu *vcpu, gva_t gva) |
| 670 | { |
Avi Kivity | a461930 | 2008-12-25 15:19:00 +0200 | [diff] [blame] | 671 | struct kvm_shadow_walk_iterator iterator; |
Xiao Guangrong | f78978a | 2010-05-15 18:53:35 +0800 | [diff] [blame] | 672 | struct kvm_mmu_page *sp; |
Avi Kivity | a461930 | 2008-12-25 15:19:00 +0200 | [diff] [blame] | 673 | int level; |
| 674 | u64 *sptep; |
Marcelo Tosatti | a705289 | 2008-09-23 13:18:35 -0300 | [diff] [blame] | 675 | |
Xiao Guangrong | bebb106 | 2011-07-12 03:23:20 +0800 | [diff] [blame] | 676 | vcpu_clear_mmio_info(vcpu, gva); |
| 677 | |
Xiao Guangrong | f57f2ef | 2011-09-22 16:56:39 +0800 | [diff] [blame] | 678 | /* |
| 679 | * No need to check return value here, rmap_can_add() can |
| 680 | * help us to skip pte prefetch later. |
| 681 | */ |
| 682 | mmu_topup_memory_caches(vcpu); |
Avi Kivity | a461930 | 2008-12-25 15:19:00 +0200 | [diff] [blame] | 683 | |
Xiao Guangrong | f57f2ef | 2011-09-22 16:56:39 +0800 | [diff] [blame] | 684 | spin_lock(&vcpu->kvm->mmu_lock); |
Avi Kivity | a461930 | 2008-12-25 15:19:00 +0200 | [diff] [blame] | 685 | for_each_shadow_entry(vcpu, gva, iterator) { |
| 686 | level = iterator.level; |
| 687 | sptep = iterator.sptep; |
| 688 | |
Xiao Guangrong | f78978a | 2010-05-15 18:53:35 +0800 | [diff] [blame] | 689 | sp = page_header(__pa(sptep)); |
Xiao Guangrong | 884a0ff | 2010-04-28 11:55:15 +0800 | [diff] [blame] | 690 | if (is_last_spte(*sptep, level)) { |
Xiao Guangrong | f57f2ef | 2011-09-22 16:56:39 +0800 | [diff] [blame] | 691 | pt_element_t gpte; |
| 692 | gpa_t pte_gpa; |
| 693 | |
Xiao Guangrong | f78978a | 2010-05-15 18:53:35 +0800 | [diff] [blame] | 694 | if (!sp->unsync) |
| 695 | break; |
| 696 | |
Xiao Guangrong | 505aef8 | 2011-09-22 16:56:06 +0800 | [diff] [blame] | 697 | pte_gpa = FNAME(get_level1_sp_gpa)(sp); |
Avi Kivity | 08e850c | 2010-03-15 13:59:57 +0200 | [diff] [blame] | 698 | pte_gpa += (sptep - sp->spt) * sizeof(pt_element_t); |
Avi Kivity | a461930 | 2008-12-25 15:19:00 +0200 | [diff] [blame] | 699 | |
Xiao Guangrong | 505aef8 | 2011-09-22 16:56:06 +0800 | [diff] [blame] | 700 | if (mmu_page_zap_pte(vcpu->kvm, sp, sptep)) |
| 701 | kvm_flush_remote_tlbs(vcpu->kvm); |
Xiao Guangrong | f57f2ef | 2011-09-22 16:56:39 +0800 | [diff] [blame] | 702 | |
| 703 | if (!rmap_can_add(vcpu)) |
| 704 | break; |
| 705 | |
| 706 | if (kvm_read_guest_atomic(vcpu->kvm, pte_gpa, &gpte, |
| 707 | sizeof(pt_element_t))) |
| 708 | break; |
| 709 | |
| 710 | FNAME(update_pte)(vcpu, sp, sptep, &gpte); |
Avi Kivity | a461930 | 2008-12-25 15:19:00 +0200 | [diff] [blame] | 711 | } |
| 712 | |
Xiao Guangrong | f78978a | 2010-05-15 18:53:35 +0800 | [diff] [blame] | 713 | if (!is_shadow_present_pte(*sptep) || !sp->unsync_children) |
Avi Kivity | a461930 | 2008-12-25 15:19:00 +0200 | [diff] [blame] | 714 | break; |
| 715 | } |
Marcelo Tosatti | ad218f8 | 2008-12-01 22:32:05 -0200 | [diff] [blame] | 716 | spin_unlock(&vcpu->kvm->mmu_lock); |
Marcelo Tosatti | a705289 | 2008-09-23 13:18:35 -0300 | [diff] [blame] | 717 | } |
| 718 | |
Gleb Natapov | 1871c60 | 2010-02-10 14:21:32 +0200 | [diff] [blame] | 719 | static gpa_t FNAME(gva_to_gpa)(struct kvm_vcpu *vcpu, gva_t vaddr, u32 access, |
Avi Kivity | ab9ae31 | 2010-11-22 17:53:26 +0200 | [diff] [blame] | 720 | struct x86_exception *exception) |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 721 | { |
| 722 | struct guest_walker walker; |
Avi Kivity | e119d11 | 2007-02-12 00:54:36 -0800 | [diff] [blame] | 723 | gpa_t gpa = UNMAPPED_GVA; |
| 724 | int r; |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 725 | |
Xiao Guangrong | 3377078 | 2010-09-28 17:03:14 +0800 | [diff] [blame] | 726 | r = FNAME(walk_addr)(&walker, vcpu, vaddr, access); |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 727 | |
Avi Kivity | e119d11 | 2007-02-12 00:54:36 -0800 | [diff] [blame] | 728 | if (r) { |
Avi Kivity | 1755fbc | 2007-11-21 14:44:45 +0200 | [diff] [blame] | 729 | gpa = gfn_to_gpa(walker.gfn); |
Avi Kivity | e119d11 | 2007-02-12 00:54:36 -0800 | [diff] [blame] | 730 | gpa |= vaddr & ~PAGE_MASK; |
Avi Kivity | 8c28d03 | 2010-11-22 17:53:27 +0200 | [diff] [blame] | 731 | } else if (exception) |
| 732 | *exception = walker.fault; |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 733 | |
| 734 | return gpa; |
| 735 | } |
| 736 | |
Joerg Roedel | 6539e73 | 2010-09-10 17:30:50 +0200 | [diff] [blame] | 737 | static gpa_t FNAME(gva_to_gpa_nested)(struct kvm_vcpu *vcpu, gva_t vaddr, |
Avi Kivity | ab9ae31 | 2010-11-22 17:53:26 +0200 | [diff] [blame] | 738 | u32 access, |
| 739 | struct x86_exception *exception) |
Joerg Roedel | 6539e73 | 2010-09-10 17:30:50 +0200 | [diff] [blame] | 740 | { |
| 741 | struct guest_walker walker; |
| 742 | gpa_t gpa = UNMAPPED_GVA; |
| 743 | int r; |
| 744 | |
Xiao Guangrong | 3377078 | 2010-09-28 17:03:14 +0800 | [diff] [blame] | 745 | r = FNAME(walk_addr_nested)(&walker, vcpu, vaddr, access); |
Joerg Roedel | 6539e73 | 2010-09-10 17:30:50 +0200 | [diff] [blame] | 746 | |
| 747 | if (r) { |
| 748 | gpa = gfn_to_gpa(walker.gfn); |
| 749 | gpa |= vaddr & ~PAGE_MASK; |
Avi Kivity | 8c28d03 | 2010-11-22 17:53:27 +0200 | [diff] [blame] | 750 | } else if (exception) |
| 751 | *exception = walker.fault; |
Joerg Roedel | 6539e73 | 2010-09-10 17:30:50 +0200 | [diff] [blame] | 752 | |
| 753 | return gpa; |
| 754 | } |
| 755 | |
Marcelo Tosatti | e8bc217 | 2008-09-23 13:18:33 -0300 | [diff] [blame] | 756 | /* |
| 757 | * Using the cached information from sp->gfns is safe because: |
| 758 | * - The spte has a reference to the struct page, so the pfn for a given gfn |
| 759 | * can't change unless all sptes pointing to it are nuked first. |
Xiao Guangrong | a4ee1ca | 2010-11-23 11:13:00 +0800 | [diff] [blame] | 760 | * |
| 761 | * Note: |
| 762 | * We should flush all tlbs if spte is dropped even though guest is |
| 763 | * responsible for it. Since if we don't, kvm_mmu_notifier_invalidate_page |
| 764 | * and kvm_mmu_notifier_invalidate_range_start detect the mapping page isn't |
| 765 | * used by guest then tlbs are not flushed, so guest is allowed to access the |
| 766 | * freed pages. |
| 767 | * And we increase kvm->tlbs_dirty to delay tlbs flush in this case. |
Marcelo Tosatti | e8bc217 | 2008-09-23 13:18:33 -0300 | [diff] [blame] | 768 | */ |
Xiao Guangrong | a4a8e6f | 2010-11-19 17:04:03 +0800 | [diff] [blame] | 769 | static int FNAME(sync_page)(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp) |
Marcelo Tosatti | e8bc217 | 2008-09-23 13:18:33 -0300 | [diff] [blame] | 770 | { |
Xiao Guangrong | 505aef8 | 2011-09-22 16:56:06 +0800 | [diff] [blame] | 771 | int i, nr_present = 0; |
Lai Jiangshan | 9bdbba1 | 2010-11-19 17:03:22 +0800 | [diff] [blame] | 772 | bool host_writable; |
Gui Jianfeng | 51fb60d | 2010-04-16 17:16:40 +0800 | [diff] [blame] | 773 | gpa_t first_pte_gpa; |
Marcelo Tosatti | e8bc217 | 2008-09-23 13:18:33 -0300 | [diff] [blame] | 774 | |
Lai Jiangshan | 2032a93 | 2010-05-26 16:49:59 +0800 | [diff] [blame] | 775 | /* direct kvm_mmu_page can not be unsync. */ |
| 776 | BUG_ON(sp->role.direct); |
| 777 | |
Xiao Guangrong | 505aef8 | 2011-09-22 16:56:06 +0800 | [diff] [blame] | 778 | first_pte_gpa = FNAME(get_level1_sp_gpa)(sp); |
Gui Jianfeng | 51fb60d | 2010-04-16 17:16:40 +0800 | [diff] [blame] | 779 | |
Marcelo Tosatti | e8bc217 | 2008-09-23 13:18:33 -0300 | [diff] [blame] | 780 | for (i = 0; i < PT64_ENT_PER_PAGE; i++) { |
| 781 | unsigned pte_access; |
| 782 | pt_element_t gpte; |
| 783 | gpa_t pte_gpa; |
Xiao Guangrong | f55c3f4 | 2010-05-13 10:08:08 +0800 | [diff] [blame] | 784 | gfn_t gfn; |
Marcelo Tosatti | e8bc217 | 2008-09-23 13:18:33 -0300 | [diff] [blame] | 785 | |
Xiao Guangrong | ce88dec | 2011-07-12 03:33:44 +0800 | [diff] [blame] | 786 | if (!sp->spt[i]) |
Marcelo Tosatti | e8bc217 | 2008-09-23 13:18:33 -0300 | [diff] [blame] | 787 | continue; |
| 788 | |
Gui Jianfeng | 51fb60d | 2010-04-16 17:16:40 +0800 | [diff] [blame] | 789 | pte_gpa = first_pte_gpa + i * sizeof(pt_element_t); |
Marcelo Tosatti | e8bc217 | 2008-09-23 13:18:33 -0300 | [diff] [blame] | 790 | |
| 791 | if (kvm_read_guest_atomic(vcpu->kvm, pte_gpa, &gpte, |
| 792 | sizeof(pt_element_t))) |
| 793 | return -EINVAL; |
| 794 | |
Xiao Guangrong | 407c61c | 2010-11-23 11:08:42 +0800 | [diff] [blame] | 795 | if (FNAME(prefetch_invalid_gpte)(vcpu, sp, &sp->spt[i], gpte)) { |
Xiao Guangrong | a4ee1ca | 2010-11-23 11:13:00 +0800 | [diff] [blame] | 796 | vcpu->kvm->tlbs_dirty++; |
Xiao Guangrong | 407c61c | 2010-11-23 11:08:42 +0800 | [diff] [blame] | 797 | continue; |
| 798 | } |
| 799 | |
Xiao Guangrong | ce88dec | 2011-07-12 03:33:44 +0800 | [diff] [blame] | 800 | gfn = gpte_to_gfn(gpte); |
| 801 | pte_access = sp->role.access; |
| 802 | pte_access &= FNAME(gpte_access)(vcpu, gpte, true); |
| 803 | |
| 804 | if (sync_mmio_spte(&sp->spt[i], gfn, pte_access, &nr_present)) |
| 805 | continue; |
| 806 | |
Xiao Guangrong | 407c61c | 2010-11-23 11:08:42 +0800 | [diff] [blame] | 807 | if (gfn != sp->gfns[i]) { |
Xiao Guangrong | c370795 | 2011-07-12 03:28:04 +0800 | [diff] [blame] | 808 | drop_spte(vcpu->kvm, &sp->spt[i]); |
Xiao Guangrong | a4ee1ca | 2010-11-23 11:13:00 +0800 | [diff] [blame] | 809 | vcpu->kvm->tlbs_dirty++; |
Marcelo Tosatti | e8bc217 | 2008-09-23 13:18:33 -0300 | [diff] [blame] | 810 | continue; |
| 811 | } |
| 812 | |
| 813 | nr_present++; |
Xiao Guangrong | ce88dec | 2011-07-12 03:33:44 +0800 | [diff] [blame] | 814 | |
Xiao Guangrong | f8e453b | 2010-12-23 16:09:29 +0800 | [diff] [blame] | 815 | host_writable = sp->spt[i] & SPTE_HOST_WRITEABLE; |
| 816 | |
Marcelo Tosatti | e8bc217 | 2008-09-23 13:18:33 -0300 | [diff] [blame] | 817 | set_spte(vcpu, &sp->spt[i], pte_access, 0, 0, |
Xiao Guangrong | 640d9b0 | 2011-07-12 03:24:39 +0800 | [diff] [blame] | 818 | PT_PAGE_TABLE_LEVEL, gfn, |
Izik Eidus | 1403283 | 2009-09-23 21:47:17 +0300 | [diff] [blame] | 819 | spte_to_pfn(sp->spt[i]), true, false, |
Lai Jiangshan | 9bdbba1 | 2010-11-19 17:03:22 +0800 | [diff] [blame] | 820 | host_writable); |
Marcelo Tosatti | e8bc217 | 2008-09-23 13:18:33 -0300 | [diff] [blame] | 821 | } |
| 822 | |
| 823 | return !nr_present; |
| 824 | } |
| 825 | |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 826 | #undef pt_element_t |
| 827 | #undef guest_walker |
| 828 | #undef FNAME |
| 829 | #undef PT_BASE_ADDR_MASK |
| 830 | #undef PT_INDEX |
Joerg Roedel | e04da98 | 2009-07-27 16:30:45 +0200 | [diff] [blame] | 831 | #undef PT_LVL_ADDR_MASK |
| 832 | #undef PT_LVL_OFFSET_MASK |
Avi Kivity | c7addb9 | 2007-09-16 18:58:32 +0200 | [diff] [blame] | 833 | #undef PT_LEVEL_BITS |
Avi Kivity | cea0f0e | 2007-01-05 16:36:43 -0800 | [diff] [blame] | 834 | #undef PT_MAX_FULL_LEVELS |
Avi Kivity | 5fb07dd | 2007-11-21 12:35:07 +0200 | [diff] [blame] | 835 | #undef gpte_to_gfn |
Joerg Roedel | e04da98 | 2009-07-27 16:30:45 +0200 | [diff] [blame] | 836 | #undef gpte_to_gfn_lvl |
Marcelo Tosatti | b3e4e63 | 2007-12-07 07:56:58 -0500 | [diff] [blame] | 837 | #undef CMPXCHG |