blob: d07f48a06f099e44d24a3b7e4d88d81fe8402b94 [file] [log] [blame]
Avi Kivity6aa8b732006-12-10 02:21:36 -08001/*
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.
Avi Kivity221d0592010-05-23 18:37:00 +030010 * Copyright 2010 Red Hat, Inc. and/or its affilates.
Avi Kivity6aa8b732006-12-10 02:21:36 -080011 *
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 Roedele04da982009-07-27 16:30:45 +020031 #define PT_LVL_ADDR_MASK(lvl) PT64_LVL_ADDR_MASK(lvl)
32 #define PT_LVL_OFFSET_MASK(lvl) PT64_LVL_OFFSET_MASK(lvl)
Avi Kivity6aa8b732006-12-10 02:21:36 -080033 #define PT_INDEX(addr, level) PT64_INDEX(addr, level)
Avi Kivity6aa8b732006-12-10 02:21:36 -080034 #define PT_LEVEL_MASK(level) PT64_LEVEL_MASK(level)
Avi Kivityc7addb92007-09-16 18:58:32 +020035 #define PT_LEVEL_BITS PT64_LEVEL_BITS
Avi Kivitycea0f0e2007-01-05 16:36:43 -080036 #ifdef CONFIG_X86_64
37 #define PT_MAX_FULL_LEVELS 4
Marcelo Tosattib3e4e632007-12-07 07:56:58 -050038 #define CMPXCHG cmpxchg
Avi Kivitycea0f0e2007-01-05 16:36:43 -080039 #else
Marcelo Tosattib3e4e632007-12-07 07:56:58 -050040 #define CMPXCHG cmpxchg64
Avi Kivitycea0f0e2007-01-05 16:36:43 -080041 #define PT_MAX_FULL_LEVELS 2
42 #endif
Avi Kivity6aa8b732006-12-10 02:21:36 -080043#elif PTTYPE == 32
44 #define pt_element_t u32
45 #define guest_walker guest_walker32
46 #define FNAME(name) paging##32_##name
47 #define PT_BASE_ADDR_MASK PT32_BASE_ADDR_MASK
Joerg Roedele04da982009-07-27 16:30:45 +020048 #define PT_LVL_ADDR_MASK(lvl) PT32_LVL_ADDR_MASK(lvl)
49 #define PT_LVL_OFFSET_MASK(lvl) PT32_LVL_OFFSET_MASK(lvl)
Avi Kivity6aa8b732006-12-10 02:21:36 -080050 #define PT_INDEX(addr, level) PT32_INDEX(addr, level)
Avi Kivity6aa8b732006-12-10 02:21:36 -080051 #define PT_LEVEL_MASK(level) PT32_LEVEL_MASK(level)
Avi Kivityc7addb92007-09-16 18:58:32 +020052 #define PT_LEVEL_BITS PT32_LEVEL_BITS
Avi Kivitycea0f0e2007-01-05 16:36:43 -080053 #define PT_MAX_FULL_LEVELS 2
Marcelo Tosattib3e4e632007-12-07 07:56:58 -050054 #define CMPXCHG cmpxchg
Avi Kivity6aa8b732006-12-10 02:21:36 -080055#else
56 #error Invalid PTTYPE value
57#endif
58
Joerg Roedele04da982009-07-27 16:30:45 +020059#define gpte_to_gfn_lvl FNAME(gpte_to_gfn_lvl)
60#define gpte_to_gfn(pte) gpte_to_gfn_lvl((pte), PT_PAGE_TABLE_LEVEL)
Avi Kivity5fb07dd2007-11-21 12:35:07 +020061
Avi Kivity6aa8b732006-12-10 02:21:36 -080062/*
63 * The guest_walker structure emulates the behavior of the hardware page
64 * table walker.
65 */
66struct guest_walker {
67 int level;
Avi Kivitycea0f0e2007-01-05 16:36:43 -080068 gfn_t table_gfn[PT_MAX_FULL_LEVELS];
Marcelo Tosatti78190262007-12-11 19:12:27 -050069 pt_element_t ptes[PT_MAX_FULL_LEVELS];
Xiao Guangrong189be382010-08-22 19:13:33 +080070 pt_element_t prefetch_ptes[PTE_PREFETCH_NUM];
Marcelo Tosatti78190262007-12-11 19:12:27 -050071 gpa_t pte_gpa[PT_MAX_FULL_LEVELS];
Avi Kivityfe135d22007-12-09 16:15:46 +020072 unsigned pt_access;
73 unsigned pte_access;
Avi Kivity815af8d2007-01-05 16:36:44 -080074 gfn_t gfn;
Avi Kivity7993ba42007-01-26 00:56:41 -080075 u32 error_code;
Avi Kivity6aa8b732006-12-10 02:21:36 -080076};
77
Joerg Roedele04da982009-07-27 16:30:45 +020078static gfn_t gpte_to_gfn_lvl(pt_element_t gpte, int lvl)
Avi Kivity5fb07dd2007-11-21 12:35:07 +020079{
Joerg Roedele04da982009-07-27 16:30:45 +020080 return (gpte & PT_LVL_ADDR_MASK(lvl)) >> PAGE_SHIFT;
Avi Kivity5fb07dd2007-11-21 12:35:07 +020081}
82
Marcelo Tosattib3e4e632007-12-07 07:56:58 -050083static bool FNAME(cmpxchg_gpte)(struct kvm *kvm,
84 gfn_t table_gfn, unsigned index,
85 pt_element_t orig_pte, pt_element_t new_pte)
86{
87 pt_element_t ret;
88 pt_element_t *table;
89 struct page *page;
90
91 page = gfn_to_page(kvm, table_gfn);
Izik Eidus72dc67a2008-02-10 18:04:15 +020092
Marcelo Tosattib3e4e632007-12-07 07:56:58 -050093 table = kmap_atomic(page, KM_USER0);
Marcelo Tosattib3e4e632007-12-07 07:56:58 -050094 ret = CMPXCHG(&table[index], orig_pte, new_pte);
Marcelo Tosattib3e4e632007-12-07 07:56:58 -050095 kunmap_atomic(table, KM_USER0);
96
97 kvm_release_page_dirty(page);
98
99 return (ret != orig_pte);
100}
101
Avi Kivitybedbe4e2007-12-09 16:52:56 +0200102static unsigned FNAME(gpte_access)(struct kvm_vcpu *vcpu, pt_element_t gpte)
103{
104 unsigned access;
105
106 access = (gpte & (PT_WRITABLE_MASK | PT_USER_MASK)) | ACC_EXEC_MASK;
107#if PTTYPE == 64
108 if (is_nx(vcpu))
109 access &= ~(gpte >> PT64_NX_SHIFT);
110#endif
111 return access;
112}
113
Avi Kivityac79c972007-01-05 16:36:40 -0800114/*
115 * Fetch a guest pte for a guest virtual address
116 */
Avi Kivity7993ba42007-01-26 00:56:41 -0800117static int FNAME(walk_addr)(struct guest_walker *walker,
118 struct kvm_vcpu *vcpu, gva_t addr,
Avi Kivity73b10872007-01-26 00:56:41 -0800119 int write_fault, int user_fault, int fetch_fault)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800120{
Avi Kivity42bf3f02007-10-17 12:18:47 +0200121 pt_element_t pte;
Avi Kivitycea0f0e2007-01-05 16:36:43 -0800122 gfn_t table_gfn;
Avi Kivityf59c1d22010-07-06 16:20:43 +0300123 unsigned index, pt_access, uninitialized_var(pte_access);
Avi Kivity42bf3f02007-10-17 12:18:47 +0200124 gpa_t pte_gpa;
Avi Kivityf59c1d22010-07-06 16:20:43 +0300125 bool eperm, present, rsvd_fault;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800126
Avi Kivity07420172009-07-06 12:21:32 +0300127 trace_kvm_mmu_pagetable_walk(addr, write_fault, user_fault,
128 fetch_fault);
Marcelo Tosattib3e4e632007-12-07 07:56:58 -0500129walk:
Avi Kivityf59c1d22010-07-06 16:20:43 +0300130 present = true;
131 eperm = rsvd_fault = false;
Zhang Xiantaoad312c72007-12-13 23:50:52 +0800132 walker->level = vcpu->arch.mmu.root_level;
Joerg Roedel5777ed32010-09-10 17:30:42 +0200133 pte = vcpu->arch.mmu.get_cr3(vcpu);
Avi Kivity1b0973b2007-01-05 16:36:41 -0800134#if PTTYPE == 64
Joerg Roedel957446a2010-09-10 17:30:38 +0200135 if (vcpu->arch.mmu.root_level == PT32E_ROOT_LEVEL) {
Avi Kivity6de4f3a2009-05-31 22:58:47 +0300136 pte = kvm_pdptr_read(vcpu, (addr >> 30) & 3);
Avi Kivity07420172009-07-06 12:21:32 +0300137 trace_kvm_mmu_paging_element(pte, walker->level);
Avi Kivityf59c1d22010-07-06 16:20:43 +0300138 if (!is_present_gpte(pte)) {
139 present = false;
140 goto error;
141 }
Avi Kivity1b0973b2007-01-05 16:36:41 -0800142 --walker->level;
143 }
144#endif
Avi Kivitya9058ec2006-12-29 16:49:37 -0800145 ASSERT((!is_long_mode(vcpu) && is_pae(vcpu)) ||
Joerg Roedel5777ed32010-09-10 17:30:42 +0200146 (vcpu->arch.mmu.get_cr3(vcpu) & CR3_NONPAE_RESERVED_BITS) == 0);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800147
Avi Kivityfe135d22007-12-09 16:15:46 +0200148 pt_access = ACC_ALL;
Avi Kivityac79c972007-01-05 16:36:40 -0800149
150 for (;;) {
Avi Kivity42bf3f02007-10-17 12:18:47 +0200151 index = PT_INDEX(addr, walker->level);
Avi Kivityac79c972007-01-05 16:36:40 -0800152
Avi Kivity5fb07dd2007-11-21 12:35:07 +0200153 table_gfn = gpte_to_gfn(pte);
Avi Kivity1755fbc2007-11-21 14:44:45 +0200154 pte_gpa = gfn_to_gpa(table_gfn);
Izik Eidusec8d4ea2007-11-19 11:28:19 +0200155 pte_gpa += index * sizeof(pt_element_t);
Avi Kivity42bf3f02007-10-17 12:18:47 +0200156 walker->table_gfn[walker->level - 1] = table_gfn;
Marcelo Tosatti78190262007-12-11 19:12:27 -0500157 walker->pte_gpa[walker->level - 1] = pte_gpa;
Avi Kivityac79c972007-01-05 16:36:40 -0800158
Avi Kivityf59c1d22010-07-06 16:20:43 +0300159 if (kvm_read_guest(vcpu->kvm, pte_gpa, &pte, sizeof(pte))) {
160 present = false;
161 break;
162 }
Marcelo Tosattia6085fba2010-01-14 17:41:27 -0200163
Avi Kivity07420172009-07-06 12:21:32 +0300164 trace_kvm_mmu_paging_element(pte, walker->level);
Avi Kivity42bf3f02007-10-17 12:18:47 +0200165
Avi Kivityf59c1d22010-07-06 16:20:43 +0300166 if (!is_present_gpte(pte)) {
167 present = false;
168 break;
169 }
Avi Kivity7993ba42007-01-26 00:56:41 -0800170
Joerg Roedel3241f222010-09-10 17:30:45 +0200171 if (is_rsvd_bits_set(&vcpu->arch.mmu, pte, walker->level)) {
Avi Kivityf59c1d22010-07-06 16:20:43 +0300172 rsvd_fault = true;
173 break;
174 }
Dong, Eddie82725b22009-03-30 16:21:08 +0800175
Takuya Yoshikawa8dae4442010-01-18 18:45:10 +0900176 if (write_fault && !is_writable_pte(pte))
Avi Kivity7993ba42007-01-26 00:56:41 -0800177 if (user_fault || is_write_protection(vcpu))
Avi Kivityf59c1d22010-07-06 16:20:43 +0300178 eperm = true;
Avi Kivity7993ba42007-01-26 00:56:41 -0800179
Avi Kivity42bf3f02007-10-17 12:18:47 +0200180 if (user_fault && !(pte & PT_USER_MASK))
Avi Kivityf59c1d22010-07-06 16:20:43 +0300181 eperm = true;
Avi Kivity7993ba42007-01-26 00:56:41 -0800182
Avi Kivity73b10872007-01-26 00:56:41 -0800183#if PTTYPE == 64
Xiao Guangrong24222c22010-04-06 18:31:13 +0800184 if (fetch_fault && (pte & PT64_NX_MASK))
Avi Kivityf59c1d22010-07-06 16:20:43 +0300185 eperm = true;
Avi Kivity73b10872007-01-26 00:56:41 -0800186#endif
187
Avi Kivityf59c1d22010-07-06 16:20:43 +0300188 if (!eperm && !rsvd_fault && !(pte & PT_ACCESSED_MASK)) {
Avi Kivity07420172009-07-06 12:21:32 +0300189 trace_kvm_mmu_set_accessed_bit(table_gfn, index,
190 sizeof(pte));
Marcelo Tosattib3e4e632007-12-07 07:56:58 -0500191 if (FNAME(cmpxchg_gpte)(vcpu->kvm, table_gfn,
192 index, pte, pte|PT_ACCESSED_MASK))
193 goto walk;
Gui Jianfengf3b8c962010-05-05 09:09:21 +0800194 mark_page_dirty(vcpu->kvm, table_gfn);
Avi Kivity42bf3f02007-10-17 12:18:47 +0200195 pte |= PT_ACCESSED_MASK;
Avi Kivitybf3f8e82007-02-19 14:37:46 +0200196 }
Avi Kivityac79c972007-01-05 16:36:40 -0800197
Avi Kivitybedbe4e2007-12-09 16:52:56 +0200198 pte_access = pt_access & FNAME(gpte_access)(vcpu, pte);
Avi Kivityfe135d22007-12-09 16:15:46 +0200199
Marcelo Tosatti78190262007-12-11 19:12:27 -0500200 walker->ptes[walker->level - 1] = pte;
201
Joerg Roedele04da982009-07-27 16:30:45 +0200202 if ((walker->level == PT_PAGE_TABLE_LEVEL) ||
203 ((walker->level == PT_DIRECTORY_LEVEL) &&
Gui Jianfeng814a59d2010-04-16 17:18:01 +0800204 is_large_pte(pte) &&
Joerg Roedele04da982009-07-27 16:30:45 +0200205 (PTTYPE == 64 || is_pse(vcpu))) ||
206 ((walker->level == PT_PDPE_LEVEL) &&
Gui Jianfeng814a59d2010-04-16 17:18:01 +0800207 is_large_pte(pte) &&
Joerg Roedel957446a2010-09-10 17:30:38 +0200208 vcpu->arch.mmu.root_level == PT64_ROOT_LEVEL)) {
Joerg Roedele04da982009-07-27 16:30:45 +0200209 int lvl = walker->level;
Avi Kivity815af8d2007-01-05 16:36:44 -0800210
Joerg Roedele04da982009-07-27 16:30:45 +0200211 walker->gfn = gpte_to_gfn_lvl(pte, lvl);
212 walker->gfn += (addr & PT_LVL_OFFSET_MASK(lvl))
213 >> PAGE_SHIFT;
214
215 if (PTTYPE == 32 &&
216 walker->level == PT_DIRECTORY_LEVEL &&
217 is_cpuid_PSE36())
Avi Kivityda9285212007-11-21 13:54:47 +0200218 walker->gfn += pse36_gfn_delta(pte);
Joerg Roedele04da982009-07-27 16:30:45 +0200219
Avi Kivity815af8d2007-01-05 16:36:44 -0800220 break;
221 }
222
Avi Kivityfe135d22007-12-09 16:15:46 +0200223 pt_access = pte_access;
Avi Kivityac79c972007-01-05 16:36:40 -0800224 --walker->level;
225 }
Avi Kivity42bf3f02007-10-17 12:18:47 +0200226
Avi Kivityf59c1d22010-07-06 16:20:43 +0300227 if (!present || eperm || rsvd_fault)
228 goto error;
229
Avi Kivity43a37952009-06-10 14:12:05 +0300230 if (write_fault && !is_dirty_gpte(pte)) {
Marcelo Tosattib3e4e632007-12-07 07:56:58 -0500231 bool ret;
232
Avi Kivity07420172009-07-06 12:21:32 +0300233 trace_kvm_mmu_set_dirty_bit(table_gfn, index, sizeof(pte));
Marcelo Tosattib3e4e632007-12-07 07:56:58 -0500234 ret = FNAME(cmpxchg_gpte)(vcpu->kvm, table_gfn, index, pte,
235 pte|PT_DIRTY_MASK);
236 if (ret)
237 goto walk;
Gui Jianfengf3b8c962010-05-05 09:09:21 +0800238 mark_page_dirty(vcpu->kvm, table_gfn);
Avi Kivity42bf3f02007-10-17 12:18:47 +0200239 pte |= PT_DIRTY_MASK;
Marcelo Tosatti78190262007-12-11 19:12:27 -0500240 walker->ptes[walker->level - 1] = pte;
Avi Kivity42bf3f02007-10-17 12:18:47 +0200241 }
242
Avi Kivityfe135d22007-12-09 16:15:46 +0200243 walker->pt_access = pt_access;
244 walker->pte_access = pte_access;
245 pgprintk("%s: pte %llx pte_access %x pt_access %x\n",
Gui Jianfeng518c5a02010-05-05 09:58:33 +0800246 __func__, (u64)pte, pte_access, pt_access);
Avi Kivity7993ba42007-01-26 00:56:41 -0800247 return 1;
248
Avi Kivityf59c1d22010-07-06 16:20:43 +0300249error:
Avi Kivity7993ba42007-01-26 00:56:41 -0800250 walker->error_code = 0;
Avi Kivityf59c1d22010-07-06 16:20:43 +0300251 if (present)
252 walker->error_code |= PFERR_PRESENT_MASK;
Avi Kivity7993ba42007-01-26 00:56:41 -0800253 if (write_fault)
254 walker->error_code |= PFERR_WRITE_MASK;
255 if (user_fault)
256 walker->error_code |= PFERR_USER_MASK;
Avi Kivityb0eeec22010-07-06 15:40:18 +0300257 if (fetch_fault && is_nx(vcpu))
Avi Kivity73b10872007-01-26 00:56:41 -0800258 walker->error_code |= PFERR_FETCH_MASK;
Dong, Eddie82725b22009-03-30 16:21:08 +0800259 if (rsvd_fault)
260 walker->error_code |= PFERR_RSVD_MASK;
Joerg Roedel8df25a32010-09-10 17:30:46 +0200261
262 vcpu->arch.fault.address = addr;
263 vcpu->arch.fault.error_code = walker->error_code;
264
Avi Kivity07420172009-07-06 12:21:32 +0300265 trace_kvm_mmu_walker_error(walker->error_code);
Shaohua Life5518812007-07-23 14:51:39 +0800266 return 0;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800267}
268
Xiao Guangrongac3cd032010-06-11 21:28:14 +0800269static void FNAME(update_pte)(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp,
Dong, Eddie489f1d62008-01-07 11:14:20 +0200270 u64 *spte, const void *pte)
Avi Kivity00284252007-05-01 16:53:31 +0300271{
272 pt_element_t gpte;
Avi Kivity41074d02007-12-09 17:00:02 +0200273 unsigned pte_access;
Anthony Liguori35149e22008-04-02 14:46:56 -0500274 pfn_t pfn;
Avi Kivityfbc5d132010-03-15 13:59:56 +0200275 u64 new_spte;
Avi Kivity00284252007-05-01 16:53:31 +0300276
Avi Kivity00284252007-05-01 16:53:31 +0300277 gpte = *(const pt_element_t *)pte;
Avi Kivityc7addb92007-09-16 18:58:32 +0200278 if (~gpte & (PT_PRESENT_MASK | PT_ACCESSED_MASK)) {
Avi Kivityfbc5d132010-03-15 13:59:56 +0200279 if (!is_present_gpte(gpte)) {
Xiao Guangrongac3cd032010-06-11 21:28:14 +0800280 if (sp->unsync)
Avi Kivityfbc5d132010-03-15 13:59:56 +0200281 new_spte = shadow_trap_nonpresent_pte;
282 else
283 new_spte = shadow_notrap_nonpresent_pte;
284 __set_spte(spte, new_spte);
285 }
Avi Kivityc7addb92007-09-16 18:58:32 +0200286 return;
287 }
Harvey Harrisonb8688d52008-03-03 12:59:56 -0800288 pgprintk("%s: gpte %llx spte %p\n", __func__, (u64)gpte, spte);
Xiao Guangrongac3cd032010-06-11 21:28:14 +0800289 pte_access = sp->role.access & FNAME(gpte_access)(vcpu, gpte);
Avi Kivityd7824ff2007-12-30 12:29:05 +0200290 if (gpte_to_gfn(gpte) != vcpu->arch.update_pte.gfn)
291 return;
Anthony Liguori35149e22008-04-02 14:46:56 -0500292 pfn = vcpu->arch.update_pte.pfn;
293 if (is_error_pfn(pfn))
Avi Kivityd7824ff2007-12-30 12:29:05 +0200294 return;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200295 if (mmu_notifier_retry(vcpu, vcpu->arch.update_pte.mmu_seq))
296 return;
Anthony Liguori35149e22008-04-02 14:46:56 -0500297 kvm_get_pfn(pfn);
Izik Eidus14032832009-09-23 21:47:17 +0300298 /*
299 * we call mmu_set_spte() with reset_host_protection = true beacuse that
300 * vcpu->arch.update_pte.pfn was fetched from get_user_pages(write = 1).
301 */
Xiao Guangrongac3cd032010-06-11 21:28:14 +0800302 mmu_set_spte(vcpu, spte, sp->role.access, pte_access, 0, 0,
Xiao Guangrongcb83cad2010-06-11 21:29:42 +0800303 is_dirty_gpte(gpte), NULL, PT_PAGE_TABLE_LEVEL,
Izik Eidus14032832009-09-23 21:47:17 +0300304 gpte_to_gfn(gpte), pfn, true, true);
Avi Kivity00284252007-05-01 16:53:31 +0300305}
306
Avi Kivity39c8c672010-07-13 14:27:08 +0300307static bool FNAME(gpte_changed)(struct kvm_vcpu *vcpu,
308 struct guest_walker *gw, int level)
309{
Avi Kivity39c8c672010-07-13 14:27:08 +0300310 pt_element_t curr_pte;
Xiao Guangrong189be382010-08-22 19:13:33 +0800311 gpa_t base_gpa, pte_gpa = gw->pte_gpa[level - 1];
312 u64 mask;
313 int r, index;
Avi Kivity39c8c672010-07-13 14:27:08 +0300314
Xiao Guangrong189be382010-08-22 19:13:33 +0800315 if (level == PT_PAGE_TABLE_LEVEL) {
316 mask = PTE_PREFETCH_NUM * sizeof(pt_element_t) - 1;
317 base_gpa = pte_gpa & ~mask;
318 index = (pte_gpa - base_gpa) / sizeof(pt_element_t);
319
320 r = kvm_read_guest_atomic(vcpu->kvm, base_gpa,
321 gw->prefetch_ptes, sizeof(gw->prefetch_ptes));
322 curr_pte = gw->prefetch_ptes[index];
323 } else
324 r = kvm_read_guest_atomic(vcpu->kvm, pte_gpa,
Avi Kivity39c8c672010-07-13 14:27:08 +0300325 &curr_pte, sizeof(curr_pte));
Xiao Guangrong189be382010-08-22 19:13:33 +0800326
Avi Kivity39c8c672010-07-13 14:27:08 +0300327 return r || curr_pte != gw->ptes[level - 1];
328}
329
Xiao Guangrong189be382010-08-22 19:13:33 +0800330static void FNAME(pte_prefetch)(struct kvm_vcpu *vcpu, struct guest_walker *gw,
331 u64 *sptep)
Xiao Guangrong957ed9e2010-08-22 19:12:48 +0800332{
333 struct kvm_mmu_page *sp;
Joerg Roedel3241f222010-09-10 17:30:45 +0200334 struct kvm_mmu *mmu = &vcpu->arch.mmu;
Xiao Guangrong189be382010-08-22 19:13:33 +0800335 pt_element_t *gptep = gw->prefetch_ptes;
Xiao Guangrong957ed9e2010-08-22 19:12:48 +0800336 u64 *spte;
Xiao Guangrong189be382010-08-22 19:13:33 +0800337 int i;
Xiao Guangrong957ed9e2010-08-22 19:12:48 +0800338
339 sp = page_header(__pa(sptep));
340
341 if (sp->role.level > PT_PAGE_TABLE_LEVEL)
342 return;
343
344 if (sp->role.direct)
345 return __direct_pte_prefetch(vcpu, sp, sptep);
346
347 i = (sptep - sp->spt) & ~(PTE_PREFETCH_NUM - 1);
Xiao Guangrong957ed9e2010-08-22 19:12:48 +0800348 spte = sp->spt + i;
349
350 for (i = 0; i < PTE_PREFETCH_NUM; i++, spte++) {
351 pt_element_t gpte;
352 unsigned pte_access;
353 gfn_t gfn;
354 pfn_t pfn;
355 bool dirty;
356
357 if (spte == sptep)
358 continue;
359
360 if (*spte != shadow_trap_nonpresent_pte)
361 continue;
362
363 gpte = gptep[i];
364
365 if (!is_present_gpte(gpte) ||
Joerg Roedel3241f222010-09-10 17:30:45 +0200366 is_rsvd_bits_set(mmu, gpte, PT_PAGE_TABLE_LEVEL)) {
Xiao Guangrong957ed9e2010-08-22 19:12:48 +0800367 if (!sp->unsync)
368 __set_spte(spte, shadow_notrap_nonpresent_pte);
369 continue;
370 }
371
372 if (!(gpte & PT_ACCESSED_MASK))
373 continue;
374
375 pte_access = sp->role.access & FNAME(gpte_access)(vcpu, gpte);
376 gfn = gpte_to_gfn(gpte);
377 dirty = is_dirty_gpte(gpte);
378 pfn = pte_prefetch_gfn_to_pfn(vcpu, gfn,
379 (pte_access & ACC_WRITE_MASK) && dirty);
380 if (is_error_pfn(pfn)) {
381 kvm_release_pfn_clean(pfn);
382 break;
383 }
384
385 mmu_set_spte(vcpu, spte, sp->role.access, pte_access, 0, 0,
386 dirty, NULL, PT_PAGE_TABLE_LEVEL, gfn,
387 pfn, true, true);
388 }
389}
390
Avi Kivity6aa8b732006-12-10 02:21:36 -0800391/*
Avi Kivity6aa8b732006-12-10 02:21:36 -0800392 * Fetch a shadow pte for a specific level in the paging hierarchy.
393 */
394static u64 *FNAME(fetch)(struct kvm_vcpu *vcpu, gva_t addr,
Avi Kivitye7a04c92008-12-25 15:10:50 +0200395 struct guest_walker *gw,
Joerg Roedel7e4e4052009-07-27 16:30:46 +0200396 int user_fault, int write_fault, int hlevel,
Anthony Liguori35149e22008-04-02 14:46:56 -0500397 int *ptwrite, pfn_t pfn)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800398{
Avi Kivitye7a04c92008-12-25 15:10:50 +0200399 unsigned access = gw->pt_access;
Avi Kivity5991b332010-07-13 14:27:10 +0300400 struct kvm_mmu_page *sp = NULL;
Xiao Guangrong84754cd2010-06-30 16:05:00 +0800401 bool dirty = is_dirty_gpte(gw->ptes[gw->level - 1]);
Avi Kivity5991b332010-07-13 14:27:10 +0300402 int top_level;
Xiao Guangrong84754cd2010-06-30 16:05:00 +0800403 unsigned direct_access;
Avi Kivity24157aa2010-07-13 14:27:11 +0300404 struct kvm_shadow_walk_iterator it;
Avi Kivityac79c972007-01-05 16:36:40 -0800405
Avi Kivity43a37952009-06-10 14:12:05 +0300406 if (!is_present_gpte(gw->ptes[gw->level - 1]))
Avi Kivityac79c972007-01-05 16:36:40 -0800407 return NULL;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800408
Xiao Guangrong84754cd2010-06-30 16:05:00 +0800409 direct_access = gw->pt_access & gw->pte_access;
410 if (!dirty)
411 direct_access &= ~ACC_WRITE_MASK;
412
Avi Kivity5991b332010-07-13 14:27:10 +0300413 top_level = vcpu->arch.mmu.root_level;
414 if (top_level == PT32E_ROOT_LEVEL)
415 top_level = PT32_ROOT_LEVEL;
416 /*
417 * Verify that the top-level gpte is still there. Since the page
418 * is a root page, it is either write protected (and cannot be
419 * changed from now on) or it is invalid (in which case, we don't
420 * really care if it changes underneath us after this point).
421 */
422 if (FNAME(gpte_changed)(vcpu, gw, top_level))
423 goto out_gpte_changed;
424
Avi Kivity24157aa2010-07-13 14:27:11 +0300425 for (shadow_walk_init(&it, vcpu, addr);
426 shadow_walk_okay(&it) && it.level > gw->level;
427 shadow_walk_next(&it)) {
Avi Kivity0b3c9332010-07-13 14:27:09 +0300428 gfn_t table_gfn;
429
Avi Kivity24157aa2010-07-13 14:27:11 +0300430 drop_large_spte(vcpu, it.sptep);
Avi Kivitye7a04c92008-12-25 15:10:50 +0200431
Avi Kivity5991b332010-07-13 14:27:10 +0300432 sp = NULL;
Avi Kivity24157aa2010-07-13 14:27:11 +0300433 if (!is_shadow_present_pte(*it.sptep)) {
434 table_gfn = gw->table_gfn[it.level - 2];
435 sp = kvm_mmu_get_page(vcpu, table_gfn, addr, it.level-1,
436 false, access, it.sptep);
Avi Kivity5991b332010-07-13 14:27:10 +0300437 }
Avi Kivity0b3c9332010-07-13 14:27:09 +0300438
439 /*
440 * Verify that the gpte in the page we've just write
441 * protected is still there.
442 */
Avi Kivity24157aa2010-07-13 14:27:11 +0300443 if (FNAME(gpte_changed)(vcpu, gw, it.level - 1))
Avi Kivity0b3c9332010-07-13 14:27:09 +0300444 goto out_gpte_changed;
Avi Kivitye7a04c92008-12-25 15:10:50 +0200445
Avi Kivity5991b332010-07-13 14:27:10 +0300446 if (sp)
Avi Kivity24157aa2010-07-13 14:27:11 +0300447 link_shadow_page(it.sptep, sp);
Avi Kivitye7a04c92008-12-25 15:10:50 +0200448 }
449
Avi Kivity0b3c9332010-07-13 14:27:09 +0300450 for (;
Avi Kivity24157aa2010-07-13 14:27:11 +0300451 shadow_walk_okay(&it) && it.level > hlevel;
452 shadow_walk_next(&it)) {
Avi Kivity0b3c9332010-07-13 14:27:09 +0300453 gfn_t direct_gfn;
454
Avi Kivity24157aa2010-07-13 14:27:11 +0300455 validate_direct_spte(vcpu, it.sptep, direct_access);
Avi Kivity0b3c9332010-07-13 14:27:09 +0300456
Avi Kivity24157aa2010-07-13 14:27:11 +0300457 drop_large_spte(vcpu, it.sptep);
Avi Kivity0b3c9332010-07-13 14:27:09 +0300458
Avi Kivity24157aa2010-07-13 14:27:11 +0300459 if (is_shadow_present_pte(*it.sptep))
Avi Kivity0b3c9332010-07-13 14:27:09 +0300460 continue;
461
Avi Kivity24157aa2010-07-13 14:27:11 +0300462 direct_gfn = gw->gfn & ~(KVM_PAGES_PER_HPAGE(it.level) - 1);
Avi Kivity0b3c9332010-07-13 14:27:09 +0300463
Avi Kivity24157aa2010-07-13 14:27:11 +0300464 sp = kvm_mmu_get_page(vcpu, direct_gfn, addr, it.level-1,
465 true, direct_access, it.sptep);
466 link_shadow_page(it.sptep, sp);
Avi Kivity0b3c9332010-07-13 14:27:09 +0300467 }
468
Avi Kivity24157aa2010-07-13 14:27:11 +0300469 mmu_set_spte(vcpu, it.sptep, access, gw->pte_access & access,
470 user_fault, write_fault, dirty, ptwrite, it.level,
Avi Kivity0b3c9332010-07-13 14:27:09 +0300471 gw->gfn, pfn, false, true);
Xiao Guangrong189be382010-08-22 19:13:33 +0800472 FNAME(pte_prefetch)(vcpu, gw, it.sptep);
Avi Kivity0b3c9332010-07-13 14:27:09 +0300473
Avi Kivity24157aa2010-07-13 14:27:11 +0300474 return it.sptep;
Avi Kivity0b3c9332010-07-13 14:27:09 +0300475
476out_gpte_changed:
Avi Kivity5991b332010-07-13 14:27:10 +0300477 if (sp)
Avi Kivity24157aa2010-07-13 14:27:11 +0300478 kvm_mmu_put_page(sp, it.sptep);
Avi Kivity0b3c9332010-07-13 14:27:09 +0300479 kvm_release_pfn_clean(pfn);
480 return NULL;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800481}
482
483/*
Avi Kivity6aa8b732006-12-10 02:21:36 -0800484 * Page fault handler. There are several causes for a page fault:
485 * - there is no shadow pte for the guest pte
486 * - write access through a shadow pte marked read only so that we can set
487 * the dirty bit
488 * - write access to a shadow pte marked read only so we can update the page
489 * dirty bitmap, when userspace requests it
490 * - mmio access; in this case we will never install a present shadow pte
491 * - normal guest page fault due to the guest pte marked not present, not
492 * writable, or not executable
493 *
Avi Kivitye2dec932007-01-05 16:36:54 -0800494 * Returns: 1 if we need to emulate the instruction, 0 otherwise, or
495 * a negative value on error.
Avi Kivity6aa8b732006-12-10 02:21:36 -0800496 */
497static int FNAME(page_fault)(struct kvm_vcpu *vcpu, gva_t addr,
498 u32 error_code)
499{
500 int write_fault = error_code & PFERR_WRITE_MASK;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800501 int user_fault = error_code & PFERR_USER_MASK;
Avi Kivity73b10872007-01-26 00:56:41 -0800502 int fetch_fault = error_code & PFERR_FETCH_MASK;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800503 struct guest_walker walker;
Avi Kivityd555c332009-06-10 14:24:23 +0300504 u64 *sptep;
Avi Kivitycea0f0e2007-01-05 16:36:43 -0800505 int write_pt = 0;
Avi Kivitye2dec932007-01-05 16:36:54 -0800506 int r;
Anthony Liguori35149e22008-04-02 14:46:56 -0500507 pfn_t pfn;
Joerg Roedel7e4e4052009-07-27 16:30:46 +0200508 int level = PT_PAGE_TABLE_LEVEL;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200509 unsigned long mmu_seq;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800510
Harvey Harrisonb8688d52008-03-03 12:59:56 -0800511 pgprintk("%s: addr %lx err %x\n", __func__, addr, error_code);
Avi Kivity714b93d2007-01-05 16:36:53 -0800512
Avi Kivitye2dec932007-01-05 16:36:54 -0800513 r = mmu_topup_memory_caches(vcpu);
514 if (r)
515 return r;
Avi Kivity714b93d2007-01-05 16:36:53 -0800516
Avi Kivity6aa8b732006-12-10 02:21:36 -0800517 /*
Eddie Donga8b876b2009-03-26 15:28:40 +0800518 * Look up the guest pte for the faulting address.
Avi Kivity6aa8b732006-12-10 02:21:36 -0800519 */
Avi Kivity73b10872007-01-26 00:56:41 -0800520 r = FNAME(walk_addr)(&walker, vcpu, addr, write_fault, user_fault,
521 fetch_fault);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800522
523 /*
524 * The page is not mapped by the guest. Let the guest handle it.
525 */
Avi Kivity7993ba42007-01-26 00:56:41 -0800526 if (!r) {
Harvey Harrisonb8688d52008-03-03 12:59:56 -0800527 pgprintk("%s: guest page fault\n", __func__);
Joerg Roedel8df25a32010-09-10 17:30:46 +0200528 inject_page_fault(vcpu);
Zhang Xiantaoad312c72007-12-13 23:50:52 +0800529 vcpu->arch.last_pt_write_count = 0; /* reset fork detector */
Avi Kivity6aa8b732006-12-10 02:21:36 -0800530 return 0;
531 }
532
Joerg Roedel7e4e4052009-07-27 16:30:46 +0200533 if (walker.level >= PT_DIRECTORY_LEVEL) {
534 level = min(walker.level, mapping_level(vcpu, walker.gfn));
535 walker.gfn = walker.gfn & ~(KVM_PAGES_PER_HPAGE(level) - 1);
Marcelo Tosatti05da4552008-02-23 11:44:30 -0300536 }
Joerg Roedel7e4e4052009-07-27 16:30:46 +0200537
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200538 mmu_seq = vcpu->kvm->mmu_notifier_seq;
Marcelo Tosatti4c2155c2008-09-16 20:54:47 -0300539 smp_rmb();
Anthony Liguori35149e22008-04-02 14:46:56 -0500540 pfn = gfn_to_pfn(vcpu->kvm, walker.gfn);
Avi Kivityd7824ff2007-12-30 12:29:05 +0200541
Avi Kivityd196e342008-01-24 11:44:11 +0200542 /* mmio */
Huang Yingbf998152010-05-31 14:28:19 +0800543 if (is_error_pfn(pfn))
544 return kvm_handle_bad_page(vcpu->kvm, walker.gfn, pfn);
Avi Kivityd196e342008-01-24 11:44:11 +0200545
Marcelo Tosattiaaee2c92007-12-20 19:18:26 -0500546 spin_lock(&vcpu->kvm->mmu_lock);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200547 if (mmu_notifier_retry(vcpu, mmu_seq))
548 goto out_unlock;
Xiao Guangrongbc32ce22010-08-28 19:22:46 +0800549
Xiao Guangrong8b1fe172010-08-30 18:22:53 +0800550 trace_kvm_mmu_audit(vcpu, AUDIT_PRE_PAGE_FAULT);
Avi Kivityeb787d12007-12-31 15:27:49 +0200551 kvm_mmu_free_some_pages(vcpu);
Avi Kivityd555c332009-06-10 14:24:23 +0300552 sptep = FNAME(fetch)(vcpu, addr, &walker, user_fault, write_fault,
Joerg Roedel7e4e4052009-07-27 16:30:46 +0200553 level, &write_pt, pfn);
Andi Kleena24e8092010-06-10 13:10:55 +0200554 (void)sptep;
Harvey Harrisonb8688d52008-03-03 12:59:56 -0800555 pgprintk("%s: shadow pte %p %llx ptwrite %d\n", __func__,
Avi Kivityd555c332009-06-10 14:24:23 +0300556 sptep, *sptep, write_pt);
Avi Kivitycea0f0e2007-01-05 16:36:43 -0800557
Avi Kivitya25f7e12007-04-30 17:05:38 +0300558 if (!write_pt)
Zhang Xiantaoad312c72007-12-13 23:50:52 +0800559 vcpu->arch.last_pt_write_count = 0; /* reset fork detector */
Avi Kivitya25f7e12007-04-30 17:05:38 +0300560
Avi Kivity1165f5f2007-04-19 17:27:43 +0300561 ++vcpu->stat.pf_fixed;
Xiao Guangrong8b1fe172010-08-30 18:22:53 +0800562 trace_kvm_mmu_audit(vcpu, AUDIT_POST_PAGE_FAULT);
Marcelo Tosattiaaee2c92007-12-20 19:18:26 -0500563 spin_unlock(&vcpu->kvm->mmu_lock);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800564
Avi Kivitycea0f0e2007-01-05 16:36:43 -0800565 return write_pt;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200566
567out_unlock:
568 spin_unlock(&vcpu->kvm->mmu_lock);
569 kvm_release_pfn_clean(pfn);
570 return 0;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800571}
572
Marcelo Tosattia7052892008-09-23 13:18:35 -0300573static void FNAME(invlpg)(struct kvm_vcpu *vcpu, gva_t gva)
574{
Avi Kivitya4619302008-12-25 15:19:00 +0200575 struct kvm_shadow_walk_iterator iterator;
Xiao Guangrongf78978a2010-05-15 18:53:35 +0800576 struct kvm_mmu_page *sp;
Avi Kivity08e850c2010-03-15 13:59:57 +0200577 gpa_t pte_gpa = -1;
Avi Kivitya4619302008-12-25 15:19:00 +0200578 int level;
579 u64 *sptep;
Andrea Arcangeli4539b352009-03-12 18:18:43 +0100580 int need_flush = 0;
Marcelo Tosattia7052892008-09-23 13:18:35 -0300581
Marcelo Tosattiad218f82008-12-01 22:32:05 -0200582 spin_lock(&vcpu->kvm->mmu_lock);
Avi Kivitya4619302008-12-25 15:19:00 +0200583
584 for_each_shadow_entry(vcpu, gva, iterator) {
585 level = iterator.level;
586 sptep = iterator.sptep;
587
Xiao Guangrongf78978a2010-05-15 18:53:35 +0800588 sp = page_header(__pa(sptep));
Xiao Guangrong884a0ff2010-04-28 11:55:15 +0800589 if (is_last_spte(*sptep, level)) {
Xiao Guangrong22c9b2d2010-04-28 11:54:44 +0800590 int offset, shift;
Avi Kivity08e850c2010-03-15 13:59:57 +0200591
Xiao Guangrongf78978a2010-05-15 18:53:35 +0800592 if (!sp->unsync)
593 break;
594
Xiao Guangrong22c9b2d2010-04-28 11:54:44 +0800595 shift = PAGE_SHIFT -
596 (PT_LEVEL_BITS - PT64_LEVEL_BITS) * level;
597 offset = sp->role.quadrant << shift;
598
599 pte_gpa = (sp->gfn << PAGE_SHIFT) + offset;
Avi Kivity08e850c2010-03-15 13:59:57 +0200600 pte_gpa += (sptep - sp->spt) * sizeof(pt_element_t);
Avi Kivitya4619302008-12-25 15:19:00 +0200601
602 if (is_shadow_present_pte(*sptep)) {
Avi Kivitya4619302008-12-25 15:19:00 +0200603 if (is_large_pte(*sptep))
604 --vcpu->kvm->stat.lpages;
Avi Kivitybe38d272010-06-06 14:31:27 +0300605 drop_spte(vcpu->kvm, sptep,
606 shadow_trap_nonpresent_pte);
Andrea Arcangeli4539b352009-03-12 18:18:43 +0100607 need_flush = 1;
Avi Kivitybe38d272010-06-06 14:31:27 +0300608 } else
609 __set_spte(sptep, shadow_trap_nonpresent_pte);
Avi Kivitya4619302008-12-25 15:19:00 +0200610 break;
611 }
612
Xiao Guangrongf78978a2010-05-15 18:53:35 +0800613 if (!is_shadow_present_pte(*sptep) || !sp->unsync_children)
Avi Kivitya4619302008-12-25 15:19:00 +0200614 break;
615 }
616
Andrea Arcangeli4539b352009-03-12 18:18:43 +0100617 if (need_flush)
618 kvm_flush_remote_tlbs(vcpu->kvm);
Avi Kivity08e850c2010-03-15 13:59:57 +0200619
620 atomic_inc(&vcpu->kvm->arch.invlpg_counter);
621
Marcelo Tosattiad218f82008-12-01 22:32:05 -0200622 spin_unlock(&vcpu->kvm->mmu_lock);
Avi Kivity08e850c2010-03-15 13:59:57 +0200623
624 if (pte_gpa == -1)
625 return;
626
627 if (mmu_topup_memory_caches(vcpu))
628 return;
629 kvm_mmu_pte_write(vcpu, pte_gpa, NULL, sizeof(pt_element_t), 0);
Marcelo Tosattia7052892008-09-23 13:18:35 -0300630}
631
Gleb Natapov1871c602010-02-10 14:21:32 +0200632static gpa_t FNAME(gva_to_gpa)(struct kvm_vcpu *vcpu, gva_t vaddr, u32 access,
633 u32 *error)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800634{
635 struct guest_walker walker;
Avi Kivitye119d112007-02-12 00:54:36 -0800636 gpa_t gpa = UNMAPPED_GVA;
637 int r;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800638
Gleb Natapov1871c602010-02-10 14:21:32 +0200639 r = FNAME(walk_addr)(&walker, vcpu, vaddr,
640 !!(access & PFERR_WRITE_MASK),
641 !!(access & PFERR_USER_MASK),
642 !!(access & PFERR_FETCH_MASK));
Avi Kivity6aa8b732006-12-10 02:21:36 -0800643
Avi Kivitye119d112007-02-12 00:54:36 -0800644 if (r) {
Avi Kivity1755fbc2007-11-21 14:44:45 +0200645 gpa = gfn_to_gpa(walker.gfn);
Avi Kivitye119d112007-02-12 00:54:36 -0800646 gpa |= vaddr & ~PAGE_MASK;
Gleb Natapov1871c602010-02-10 14:21:32 +0200647 } else if (error)
648 *error = walker.error_code;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800649
650 return gpa;
651}
652
Avi Kivityc7addb92007-09-16 18:58:32 +0200653static void FNAME(prefetch_page)(struct kvm_vcpu *vcpu,
654 struct kvm_mmu_page *sp)
655{
Avi Kivityeab9f712008-05-29 14:20:16 +0300656 int i, j, offset, r;
657 pt_element_t pt[256 / sizeof(pt_element_t)];
658 gpa_t pte_gpa;
Avi Kivityc7addb92007-09-16 18:58:32 +0200659
Avi Kivityf6e2c02b2009-01-11 13:02:10 +0200660 if (sp->role.direct
Avi Kivitye5a4c8c2007-11-20 21:39:54 +0200661 || (PTTYPE == 32 && sp->role.level > PT_PAGE_TABLE_LEVEL)) {
Avi Kivityc7addb92007-09-16 18:58:32 +0200662 nonpaging_prefetch_page(vcpu, sp);
663 return;
664 }
665
Avi Kivityeab9f712008-05-29 14:20:16 +0300666 pte_gpa = gfn_to_gpa(sp->gfn);
667 if (PTTYPE == 32) {
Avi Kivitye5a4c8c2007-11-20 21:39:54 +0200668 offset = sp->role.quadrant << PT64_LEVEL_BITS;
Avi Kivityeab9f712008-05-29 14:20:16 +0300669 pte_gpa += offset * sizeof(pt_element_t);
670 }
Marcelo Tosatti7ec54582007-12-20 19:18:23 -0500671
Avi Kivityeab9f712008-05-29 14:20:16 +0300672 for (i = 0; i < PT64_ENT_PER_PAGE; i += ARRAY_SIZE(pt)) {
673 r = kvm_read_guest_atomic(vcpu->kvm, pte_gpa, pt, sizeof pt);
674 pte_gpa += ARRAY_SIZE(pt) * sizeof(pt_element_t);
675 for (j = 0; j < ARRAY_SIZE(pt); ++j)
Avi Kivity43a37952009-06-10 14:12:05 +0300676 if (r || is_present_gpte(pt[j]))
Avi Kivityeab9f712008-05-29 14:20:16 +0300677 sp->spt[i+j] = shadow_trap_nonpresent_pte;
678 else
679 sp->spt[i+j] = shadow_notrap_nonpresent_pte;
Marcelo Tosatti7ec54582007-12-20 19:18:23 -0500680 }
Avi Kivityc7addb92007-09-16 18:58:32 +0200681}
682
Marcelo Tosattie8bc2172008-09-23 13:18:33 -0300683/*
684 * Using the cached information from sp->gfns is safe because:
685 * - The spte has a reference to the struct page, so the pfn for a given gfn
686 * can't change unless all sptes pointing to it are nuked first.
Marcelo Tosattie8bc2172008-09-23 13:18:33 -0300687 */
Xiao Guangrongbe71e062010-06-11 21:31:38 +0800688static int FNAME(sync_page)(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp,
689 bool clear_unsync)
Marcelo Tosattie8bc2172008-09-23 13:18:33 -0300690{
691 int i, offset, nr_present;
Izik Eidus14032832009-09-23 21:47:17 +0300692 bool reset_host_protection;
Gui Jianfeng51fb60d2010-04-16 17:16:40 +0800693 gpa_t first_pte_gpa;
Marcelo Tosattie8bc2172008-09-23 13:18:33 -0300694
695 offset = nr_present = 0;
696
Lai Jiangshan2032a932010-05-26 16:49:59 +0800697 /* direct kvm_mmu_page can not be unsync. */
698 BUG_ON(sp->role.direct);
699
Marcelo Tosattie8bc2172008-09-23 13:18:33 -0300700 if (PTTYPE == 32)
701 offset = sp->role.quadrant << PT64_LEVEL_BITS;
702
Gui Jianfeng51fb60d2010-04-16 17:16:40 +0800703 first_pte_gpa = gfn_to_gpa(sp->gfn) + offset * sizeof(pt_element_t);
704
Marcelo Tosattie8bc2172008-09-23 13:18:33 -0300705 for (i = 0; i < PT64_ENT_PER_PAGE; i++) {
706 unsigned pte_access;
707 pt_element_t gpte;
708 gpa_t pte_gpa;
Xiao Guangrongf55c3f42010-05-13 10:08:08 +0800709 gfn_t gfn;
Marcelo Tosattie8bc2172008-09-23 13:18:33 -0300710
711 if (!is_shadow_present_pte(sp->spt[i]))
712 continue;
713
Gui Jianfeng51fb60d2010-04-16 17:16:40 +0800714 pte_gpa = first_pte_gpa + i * sizeof(pt_element_t);
Marcelo Tosattie8bc2172008-09-23 13:18:33 -0300715
716 if (kvm_read_guest_atomic(vcpu->kvm, pte_gpa, &gpte,
717 sizeof(pt_element_t)))
718 return -EINVAL;
719
Xiao Guangrongf55c3f42010-05-13 10:08:08 +0800720 gfn = gpte_to_gfn(gpte);
Joerg Roedel3241f222010-09-10 17:30:45 +0200721 if (is_rsvd_bits_set(&vcpu->arch.mmu, gpte, PT_PAGE_TABLE_LEVEL)
Xiao Guangrongfa1de2b2010-07-16 11:19:51 +0800722 || gfn != sp->gfns[i] || !is_present_gpte(gpte)
723 || !(gpte & PT_ACCESSED_MASK)) {
Marcelo Tosattie8bc2172008-09-23 13:18:33 -0300724 u64 nonpresent;
725
Xiao Guangrongbe71e062010-06-11 21:31:38 +0800726 if (is_present_gpte(gpte) || !clear_unsync)
Marcelo Tosattie8bc2172008-09-23 13:18:33 -0300727 nonpresent = shadow_trap_nonpresent_pte;
728 else
729 nonpresent = shadow_notrap_nonpresent_pte;
Avi Kivitybe38d272010-06-06 14:31:27 +0300730 drop_spte(vcpu->kvm, &sp->spt[i], nonpresent);
Marcelo Tosattie8bc2172008-09-23 13:18:33 -0300731 continue;
732 }
733
734 nr_present++;
735 pte_access = sp->role.access & FNAME(gpte_access)(vcpu, gpte);
Izik Eidus14032832009-09-23 21:47:17 +0300736 if (!(sp->spt[i] & SPTE_HOST_WRITEABLE)) {
737 pte_access &= ~ACC_WRITE_MASK;
738 reset_host_protection = 0;
739 } else {
740 reset_host_protection = 1;
741 }
Marcelo Tosattie8bc2172008-09-23 13:18:33 -0300742 set_spte(vcpu, &sp->spt[i], pte_access, 0, 0,
Joerg Roedel7e4e4052009-07-27 16:30:46 +0200743 is_dirty_gpte(gpte), PT_PAGE_TABLE_LEVEL, gfn,
Izik Eidus14032832009-09-23 21:47:17 +0300744 spte_to_pfn(sp->spt[i]), true, false,
745 reset_host_protection);
Marcelo Tosattie8bc2172008-09-23 13:18:33 -0300746 }
747
748 return !nr_present;
749}
750
Avi Kivity6aa8b732006-12-10 02:21:36 -0800751#undef pt_element_t
752#undef guest_walker
753#undef FNAME
754#undef PT_BASE_ADDR_MASK
755#undef PT_INDEX
Avi Kivity6aa8b732006-12-10 02:21:36 -0800756#undef PT_LEVEL_MASK
Joerg Roedele04da982009-07-27 16:30:45 +0200757#undef PT_LVL_ADDR_MASK
758#undef PT_LVL_OFFSET_MASK
Avi Kivityc7addb92007-09-16 18:58:32 +0200759#undef PT_LEVEL_BITS
Avi Kivitycea0f0e2007-01-05 16:36:43 -0800760#undef PT_MAX_FULL_LEVELS
Avi Kivity5fb07dd2007-11-21 12:35:07 +0200761#undef gpte_to_gfn
Joerg Roedele04da982009-07-27 16:30:45 +0200762#undef gpte_to_gfn_lvl
Marcelo Tosattib3e4e632007-12-07 07:56:58 -0500763#undef CMPXCHG