blob: 5588fa594b61616af2daaea2852c471f97526c90 [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.
10 *
11 * Authors:
12 * Yaniv Kamay <yaniv@qumranet.com>
13 * Avi Kivity <avi@qumranet.com>
14 *
15 * This work is licensed under the terms of the GNU GPL, version 2. See
16 * the COPYING file in the top-level directory.
17 *
18 */
19
20/*
21 * We need the mmu code to access both 32-bit and 64-bit guest ptes,
22 * so the code in this file is compiled twice, once per pte size.
23 */
24
25#if PTTYPE == 64
26 #define pt_element_t u64
27 #define guest_walker guest_walker64
28 #define FNAME(name) paging##64_##name
29 #define PT_BASE_ADDR_MASK PT64_BASE_ADDR_MASK
30 #define PT_DIR_BASE_ADDR_MASK PT64_DIR_BASE_ADDR_MASK
31 #define PT_INDEX(addr, level) PT64_INDEX(addr, level)
32 #define SHADOW_PT_INDEX(addr, level) PT64_INDEX(addr, level)
33 #define PT_LEVEL_MASK(level) PT64_LEVEL_MASK(level)
Avi Kivityc7addb92007-09-16 18:58:32 +020034 #define PT_LEVEL_BITS PT64_LEVEL_BITS
Avi Kivitycea0f0e2007-01-05 16:36:43 -080035 #ifdef CONFIG_X86_64
36 #define PT_MAX_FULL_LEVELS 4
Marcelo Tosattib3e4e632007-12-07 07:56:58 -050037 #define CMPXCHG cmpxchg
Avi Kivitycea0f0e2007-01-05 16:36:43 -080038 #else
Marcelo Tosattib3e4e632007-12-07 07:56:58 -050039 #define CMPXCHG cmpxchg64
Avi Kivitycea0f0e2007-01-05 16:36:43 -080040 #define PT_MAX_FULL_LEVELS 2
41 #endif
Avi Kivity6aa8b732006-12-10 02:21:36 -080042#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
47 #define PT_DIR_BASE_ADDR_MASK PT32_DIR_BASE_ADDR_MASK
48 #define PT_INDEX(addr, level) PT32_INDEX(addr, level)
49 #define SHADOW_PT_INDEX(addr, level) PT64_INDEX(addr, level)
50 #define PT_LEVEL_MASK(level) PT32_LEVEL_MASK(level)
Avi Kivityc7addb92007-09-16 18:58:32 +020051 #define PT_LEVEL_BITS PT32_LEVEL_BITS
Avi Kivitycea0f0e2007-01-05 16:36:43 -080052 #define PT_MAX_FULL_LEVELS 2
Marcelo Tosattib3e4e632007-12-07 07:56:58 -050053 #define CMPXCHG cmpxchg
Avi Kivity6aa8b732006-12-10 02:21:36 -080054#else
55 #error Invalid PTTYPE value
56#endif
57
Avi Kivity5fb07dd2007-11-21 12:35:07 +020058#define gpte_to_gfn FNAME(gpte_to_gfn)
59#define gpte_to_gfn_pde FNAME(gpte_to_gfn_pde)
60
Avi Kivity6aa8b732006-12-10 02:21:36 -080061/*
62 * The guest_walker structure emulates the behavior of the hardware page
63 * table walker.
64 */
65struct guest_walker {
66 int level;
Avi Kivitycea0f0e2007-01-05 16:36:43 -080067 gfn_t table_gfn[PT_MAX_FULL_LEVELS];
Marcelo Tosatti78190262007-12-11 19:12:27 -050068 pt_element_t ptes[PT_MAX_FULL_LEVELS];
69 gpa_t pte_gpa[PT_MAX_FULL_LEVELS];
Avi Kivityfe135d22007-12-09 16:15:46 +020070 unsigned pt_access;
71 unsigned pte_access;
Avi Kivity815af8d2007-01-05 16:36:44 -080072 gfn_t gfn;
Avi Kivity7993ba42007-01-26 00:56:41 -080073 u32 error_code;
Avi Kivity6aa8b732006-12-10 02:21:36 -080074};
75
Avi Kivity5fb07dd2007-11-21 12:35:07 +020076static gfn_t gpte_to_gfn(pt_element_t gpte)
77{
78 return (gpte & PT_BASE_ADDR_MASK) >> PAGE_SHIFT;
79}
80
81static gfn_t gpte_to_gfn_pde(pt_element_t gpte)
82{
83 return (gpte & PT_DIR_BASE_ADDR_MASK) >> PAGE_SHIFT;
84}
85
Marcelo Tosattib3e4e632007-12-07 07:56:58 -050086static bool FNAME(cmpxchg_gpte)(struct kvm *kvm,
87 gfn_t table_gfn, unsigned index,
88 pt_element_t orig_pte, pt_element_t new_pte)
89{
90 pt_element_t ret;
91 pt_element_t *table;
92 struct page *page;
93
Izik Eidus72dc67a2008-02-10 18:04:15 +020094 down_read(&current->mm->mmap_sem);
Marcelo Tosattib3e4e632007-12-07 07:56:58 -050095 page = gfn_to_page(kvm, table_gfn);
Izik Eidus72dc67a2008-02-10 18:04:15 +020096 up_read(&current->mm->mmap_sem);
97
Marcelo Tosattib3e4e632007-12-07 07:56:58 -050098 table = kmap_atomic(page, KM_USER0);
99
100 ret = CMPXCHG(&table[index], orig_pte, new_pte);
101
102 kunmap_atomic(table, KM_USER0);
103
104 kvm_release_page_dirty(page);
105
106 return (ret != orig_pte);
107}
108
Avi Kivitybedbe4e2007-12-09 16:52:56 +0200109static unsigned FNAME(gpte_access)(struct kvm_vcpu *vcpu, pt_element_t gpte)
110{
111 unsigned access;
112
113 access = (gpte & (PT_WRITABLE_MASK | PT_USER_MASK)) | ACC_EXEC_MASK;
114#if PTTYPE == 64
115 if (is_nx(vcpu))
116 access &= ~(gpte >> PT64_NX_SHIFT);
117#endif
118 return access;
119}
120
Avi Kivityac79c972007-01-05 16:36:40 -0800121/*
122 * Fetch a guest pte for a guest virtual address
123 */
Avi Kivity7993ba42007-01-26 00:56:41 -0800124static int FNAME(walk_addr)(struct guest_walker *walker,
125 struct kvm_vcpu *vcpu, gva_t addr,
Avi Kivity73b10872007-01-26 00:56:41 -0800126 int write_fault, int user_fault, int fetch_fault)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800127{
Avi Kivity42bf3f02007-10-17 12:18:47 +0200128 pt_element_t pte;
Avi Kivitycea0f0e2007-01-05 16:36:43 -0800129 gfn_t table_gfn;
Avi Kivityfe135d22007-12-09 16:15:46 +0200130 unsigned index, pt_access, pte_access;
Avi Kivity42bf3f02007-10-17 12:18:47 +0200131 gpa_t pte_gpa;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800132
Avi Kivitycea0f0e2007-01-05 16:36:43 -0800133 pgprintk("%s: addr %lx\n", __FUNCTION__, addr);
Marcelo Tosattib3e4e632007-12-07 07:56:58 -0500134walk:
Zhang Xiantaoad312c72007-12-13 23:50:52 +0800135 walker->level = vcpu->arch.mmu.root_level;
136 pte = vcpu->arch.cr3;
Avi Kivity1b0973b2007-01-05 16:36:41 -0800137#if PTTYPE == 64
138 if (!is_long_mode(vcpu)) {
Zhang Xiantaoad312c72007-12-13 23:50:52 +0800139 pte = vcpu->arch.pdptrs[(addr >> 30) & 3];
Avi Kivity42bf3f02007-10-17 12:18:47 +0200140 if (!is_present_pte(pte))
Avi Kivity7993ba42007-01-26 00:56:41 -0800141 goto not_present;
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)) ||
Marcelo Tosatti24993d52008-02-14 21:25:39 -0200146 (vcpu->arch.cr3 & 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 Kivity42bf3f02007-10-17 12:18:47 +0200158 pgprintk("%s: table_gfn[%d] %lx\n", __FUNCTION__,
159 walker->level - 1, table_gfn);
Avi Kivityac79c972007-01-05 16:36:40 -0800160
Izik Eidusec8d4ea2007-11-19 11:28:19 +0200161 kvm_read_guest(vcpu->kvm, pte_gpa, &pte, sizeof(pte));
Avi Kivity42bf3f02007-10-17 12:18:47 +0200162
163 if (!is_present_pte(pte))
Avi Kivity7993ba42007-01-26 00:56:41 -0800164 goto not_present;
165
Avi Kivity42bf3f02007-10-17 12:18:47 +0200166 if (write_fault && !is_writeble_pte(pte))
Avi Kivity7993ba42007-01-26 00:56:41 -0800167 if (user_fault || is_write_protection(vcpu))
168 goto access_error;
169
Avi Kivity42bf3f02007-10-17 12:18:47 +0200170 if (user_fault && !(pte & PT_USER_MASK))
Avi Kivity7993ba42007-01-26 00:56:41 -0800171 goto access_error;
172
Avi Kivity73b10872007-01-26 00:56:41 -0800173#if PTTYPE == 64
Avi Kivity42bf3f02007-10-17 12:18:47 +0200174 if (fetch_fault && is_nx(vcpu) && (pte & PT64_NX_MASK))
Avi Kivity73b10872007-01-26 00:56:41 -0800175 goto access_error;
176#endif
177
Avi Kivity42bf3f02007-10-17 12:18:47 +0200178 if (!(pte & PT_ACCESSED_MASK)) {
Avi Kivitybf3f8e82007-02-19 14:37:46 +0200179 mark_page_dirty(vcpu->kvm, table_gfn);
Marcelo Tosattib3e4e632007-12-07 07:56:58 -0500180 if (FNAME(cmpxchg_gpte)(vcpu->kvm, table_gfn,
181 index, pte, pte|PT_ACCESSED_MASK))
182 goto walk;
Avi Kivity42bf3f02007-10-17 12:18:47 +0200183 pte |= PT_ACCESSED_MASK;
Avi Kivitybf3f8e82007-02-19 14:37:46 +0200184 }
Avi Kivityac79c972007-01-05 16:36:40 -0800185
Avi Kivitybedbe4e2007-12-09 16:52:56 +0200186 pte_access = pt_access & FNAME(gpte_access)(vcpu, pte);
Avi Kivityfe135d22007-12-09 16:15:46 +0200187
Marcelo Tosatti78190262007-12-11 19:12:27 -0500188 walker->ptes[walker->level - 1] = pte;
189
Avi Kivity815af8d2007-01-05 16:36:44 -0800190 if (walker->level == PT_PAGE_TABLE_LEVEL) {
Avi Kivity5fb07dd2007-11-21 12:35:07 +0200191 walker->gfn = gpte_to_gfn(pte);
Avi Kivity815af8d2007-01-05 16:36:44 -0800192 break;
193 }
194
195 if (walker->level == PT_DIRECTORY_LEVEL
Avi Kivity42bf3f02007-10-17 12:18:47 +0200196 && (pte & PT_PAGE_SIZE_MASK)
Avi Kivity815af8d2007-01-05 16:36:44 -0800197 && (PTTYPE == 64 || is_pse(vcpu))) {
Avi Kivity5fb07dd2007-11-21 12:35:07 +0200198 walker->gfn = gpte_to_gfn_pde(pte);
Avi Kivity815af8d2007-01-05 16:36:44 -0800199 walker->gfn += PT_INDEX(addr, PT_PAGE_TABLE_LEVEL);
Avi Kivityda928522007-11-21 13:54:47 +0200200 if (PTTYPE == 32 && is_cpuid_PSE36())
201 walker->gfn += pse36_gfn_delta(pte);
Avi Kivity815af8d2007-01-05 16:36:44 -0800202 break;
203 }
204
Avi Kivityfe135d22007-12-09 16:15:46 +0200205 pt_access = pte_access;
Avi Kivityac79c972007-01-05 16:36:40 -0800206 --walker->level;
207 }
Avi Kivity42bf3f02007-10-17 12:18:47 +0200208
209 if (write_fault && !is_dirty_pte(pte)) {
Marcelo Tosattib3e4e632007-12-07 07:56:58 -0500210 bool ret;
211
Avi Kivity42bf3f02007-10-17 12:18:47 +0200212 mark_page_dirty(vcpu->kvm, table_gfn);
Marcelo Tosattib3e4e632007-12-07 07:56:58 -0500213 ret = FNAME(cmpxchg_gpte)(vcpu->kvm, table_gfn, index, pte,
214 pte|PT_DIRTY_MASK);
215 if (ret)
216 goto walk;
Avi Kivity42bf3f02007-10-17 12:18:47 +0200217 pte |= PT_DIRTY_MASK;
Avi Kivity42bf3f02007-10-17 12:18:47 +0200218 kvm_mmu_pte_write(vcpu, pte_gpa, (u8 *)&pte, sizeof(pte));
Marcelo Tosatti78190262007-12-11 19:12:27 -0500219 walker->ptes[walker->level - 1] = pte;
Avi Kivity42bf3f02007-10-17 12:18:47 +0200220 }
221
Avi Kivityfe135d22007-12-09 16:15:46 +0200222 walker->pt_access = pt_access;
223 walker->pte_access = pte_access;
224 pgprintk("%s: pte %llx pte_access %x pt_access %x\n",
225 __FUNCTION__, (u64)pte, pt_access, pte_access);
Avi Kivity7993ba42007-01-26 00:56:41 -0800226 return 1;
227
228not_present:
229 walker->error_code = 0;
230 goto err;
231
232access_error:
233 walker->error_code = PFERR_PRESENT_MASK;
234
235err:
236 if (write_fault)
237 walker->error_code |= PFERR_WRITE_MASK;
238 if (user_fault)
239 walker->error_code |= PFERR_USER_MASK;
Avi Kivity73b10872007-01-26 00:56:41 -0800240 if (fetch_fault)
241 walker->error_code |= PFERR_FETCH_MASK;
Shaohua Life551882007-07-23 14:51:39 +0800242 return 0;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800243}
244
Avi Kivity00284252007-05-01 16:53:31 +0300245static void FNAME(update_pte)(struct kvm_vcpu *vcpu, struct kvm_mmu_page *page,
Avi Kivityc7addb92007-09-16 18:58:32 +0200246 u64 *spte, const void *pte, int bytes,
247 int offset_in_pte)
Avi Kivity00284252007-05-01 16:53:31 +0300248{
249 pt_element_t gpte;
Avi Kivity41074d02007-12-09 17:00:02 +0200250 unsigned pte_access;
Avi Kivityd7824ff2007-12-30 12:29:05 +0200251 struct page *npage;
Avi Kivity00284252007-05-01 16:53:31 +0300252
Avi Kivity00284252007-05-01 16:53:31 +0300253 gpte = *(const pt_element_t *)pte;
Avi Kivityc7addb92007-09-16 18:58:32 +0200254 if (~gpte & (PT_PRESENT_MASK | PT_ACCESSED_MASK)) {
255 if (!offset_in_pte && !is_present_pte(gpte))
256 set_shadow_pte(spte, shadow_notrap_nonpresent_pte);
257 return;
258 }
259 if (bytes < sizeof(pt_element_t))
Avi Kivity00284252007-05-01 16:53:31 +0300260 return;
261 pgprintk("%s: gpte %llx spte %p\n", __FUNCTION__, (u64)gpte, spte);
Avi Kivity41074d02007-12-09 17:00:02 +0200262 pte_access = page->role.access & FNAME(gpte_access)(vcpu, gpte);
Avi Kivityd7824ff2007-12-30 12:29:05 +0200263 if (gpte_to_gfn(gpte) != vcpu->arch.update_pte.gfn)
264 return;
265 npage = vcpu->arch.update_pte.page;
266 if (!npage)
267 return;
268 get_page(npage);
Avi Kivity1c4f1fd2007-12-09 17:40:31 +0200269 mmu_set_spte(vcpu, spte, page->role.access, pte_access, 0, 0,
Avi Kivityd7824ff2007-12-30 12:29:05 +0200270 gpte & PT_DIRTY_MASK, NULL, gpte_to_gfn(gpte), npage);
Avi Kivity00284252007-05-01 16:53:31 +0300271}
272
Avi Kivity6aa8b732006-12-10 02:21:36 -0800273/*
Avi Kivity6aa8b732006-12-10 02:21:36 -0800274 * Fetch a shadow pte for a specific level in the paging hierarchy.
275 */
276static u64 *FNAME(fetch)(struct kvm_vcpu *vcpu, gva_t addr,
Avi Kivity97a0a012007-05-31 15:08:29 +0300277 struct guest_walker *walker,
Avi Kivityd7824ff2007-12-30 12:29:05 +0200278 int user_fault, int write_fault, int *ptwrite,
279 struct page *page)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800280{
281 hpa_t shadow_addr;
282 int level;
Avi Kivityef0197e2007-05-30 14:21:51 +0300283 u64 *shadow_ent;
Avi Kivityfe135d22007-12-09 16:15:46 +0200284 unsigned access = walker->pt_access;
Avi Kivityac79c972007-01-05 16:36:40 -0800285
Marcelo Tosatti78190262007-12-11 19:12:27 -0500286 if (!is_present_pte(walker->ptes[walker->level - 1]))
Avi Kivityac79c972007-01-05 16:36:40 -0800287 return NULL;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800288
Zhang Xiantaoad312c72007-12-13 23:50:52 +0800289 shadow_addr = vcpu->arch.mmu.root_hpa;
290 level = vcpu->arch.mmu.shadow_root_level;
Avi Kivityaef3d3f2007-01-05 16:36:41 -0800291 if (level == PT32E_ROOT_LEVEL) {
Zhang Xiantaoad312c72007-12-13 23:50:52 +0800292 shadow_addr = vcpu->arch.mmu.pae_root[(addr >> 30) & 3];
Avi Kivityaef3d3f2007-01-05 16:36:41 -0800293 shadow_addr &= PT64_BASE_ADDR_MASK;
294 --level;
295 }
Avi Kivity6aa8b732006-12-10 02:21:36 -0800296
297 for (; ; level--) {
298 u32 index = SHADOW_PT_INDEX(addr, level);
Avi Kivity25c0de22007-01-05 16:36:42 -0800299 struct kvm_mmu_page *shadow_page;
Avi Kivity8c7bb722006-12-13 00:34:02 -0800300 u64 shadow_pte;
Avi Kivitycea0f0e2007-01-05 16:36:43 -0800301 int metaphysical;
302 gfn_t table_gfn;
Marcelo Tosatti78190262007-12-11 19:12:27 -0500303 bool new_page = 0;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800304
Avi Kivityef0197e2007-05-30 14:21:51 +0300305 shadow_ent = ((u64 *)__va(shadow_addr)) + index;
Dong, Eddie58828422008-01-02 14:29:08 +0800306 if (level == PT_PAGE_TABLE_LEVEL)
307 break;
Avi Kivityc7addb92007-09-16 18:58:32 +0200308 if (is_shadow_present_pte(*shadow_ent)) {
Avi Kivity6aa8b732006-12-10 02:21:36 -0800309 shadow_addr = *shadow_ent & PT64_BASE_ADDR_MASK;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800310 continue;
311 }
312
Avi Kivitycea0f0e2007-01-05 16:36:43 -0800313 if (level - 1 == PT_PAGE_TABLE_LEVEL
314 && walker->level == PT_DIRECTORY_LEVEL) {
315 metaphysical = 1;
Marcelo Tosatti78190262007-12-11 19:12:27 -0500316 if (!is_dirty_pte(walker->ptes[level - 1]))
Avi Kivityfe135d22007-12-09 16:15:46 +0200317 access &= ~ACC_WRITE_MASK;
Marcelo Tosatti78190262007-12-11 19:12:27 -0500318 table_gfn = gpte_to_gfn(walker->ptes[level - 1]);
Avi Kivitycea0f0e2007-01-05 16:36:43 -0800319 } else {
320 metaphysical = 0;
321 table_gfn = walker->table_gfn[level - 2];
322 }
323 shadow_page = kvm_mmu_get_page(vcpu, table_gfn, addr, level-1,
Avi Kivityfe135d22007-12-09 16:15:46 +0200324 metaphysical, access,
Marcelo Tosatti78190262007-12-11 19:12:27 -0500325 shadow_ent, &new_page);
326 if (new_page && !metaphysical) {
Marcelo Tosatti7ec54582007-12-20 19:18:23 -0500327 int r;
Marcelo Tosatti78190262007-12-11 19:12:27 -0500328 pt_element_t curr_pte;
Marcelo Tosatti7ec54582007-12-20 19:18:23 -0500329 r = kvm_read_guest_atomic(vcpu->kvm,
330 walker->pte_gpa[level - 2],
331 &curr_pte, sizeof(curr_pte));
Avi Kivityd7824ff2007-12-30 12:29:05 +0200332 if (r || curr_pte != walker->ptes[level - 2]) {
333 kvm_release_page_clean(page);
Marcelo Tosatti78190262007-12-11 19:12:27 -0500334 return NULL;
Avi Kivityd7824ff2007-12-30 12:29:05 +0200335 }
Marcelo Tosatti78190262007-12-11 19:12:27 -0500336 }
Avi Kivity47ad8e62007-05-06 15:50:58 +0300337 shadow_addr = __pa(shadow_page->spt);
Avi Kivityaef3d3f2007-01-05 16:36:41 -0800338 shadow_pte = shadow_addr | PT_PRESENT_MASK | PT_ACCESSED_MASK
339 | PT_WRITABLE_MASK | PT_USER_MASK;
Avi Kivity8c7bb722006-12-13 00:34:02 -0800340 *shadow_ent = shadow_pte;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800341 }
Avi Kivityef0197e2007-05-30 14:21:51 +0300342
Avi Kivity1c4f1fd2007-12-09 17:40:31 +0200343 mmu_set_spte(vcpu, shadow_ent, access, walker->pte_access & access,
Marcelo Tosatti78190262007-12-11 19:12:27 -0500344 user_fault, write_fault,
345 walker->ptes[walker->level-1] & PT_DIRTY_MASK,
Avi Kivityd7824ff2007-12-30 12:29:05 +0200346 ptwrite, walker->gfn, page);
Avi Kivity050e6492007-11-21 14:11:49 +0200347
Avi Kivityef0197e2007-05-30 14:21:51 +0300348 return shadow_ent;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800349}
350
351/*
Avi Kivity6aa8b732006-12-10 02:21:36 -0800352 * Page fault handler. There are several causes for a page fault:
353 * - there is no shadow pte for the guest pte
354 * - write access through a shadow pte marked read only so that we can set
355 * the dirty bit
356 * - write access to a shadow pte marked read only so we can update the page
357 * dirty bitmap, when userspace requests it
358 * - mmio access; in this case we will never install a present shadow pte
359 * - normal guest page fault due to the guest pte marked not present, not
360 * writable, or not executable
361 *
Avi Kivitye2dec932007-01-05 16:36:54 -0800362 * Returns: 1 if we need to emulate the instruction, 0 otherwise, or
363 * a negative value on error.
Avi Kivity6aa8b732006-12-10 02:21:36 -0800364 */
365static int FNAME(page_fault)(struct kvm_vcpu *vcpu, gva_t addr,
366 u32 error_code)
367{
368 int write_fault = error_code & PFERR_WRITE_MASK;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800369 int user_fault = error_code & PFERR_USER_MASK;
Avi Kivity73b10872007-01-26 00:56:41 -0800370 int fetch_fault = error_code & PFERR_FETCH_MASK;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800371 struct guest_walker walker;
372 u64 *shadow_pte;
Avi Kivitycea0f0e2007-01-05 16:36:43 -0800373 int write_pt = 0;
Avi Kivitye2dec932007-01-05 16:36:54 -0800374 int r;
Avi Kivityd7824ff2007-12-30 12:29:05 +0200375 struct page *page;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800376
Avi Kivitycea0f0e2007-01-05 16:36:43 -0800377 pgprintk("%s: addr %lx err %x\n", __FUNCTION__, addr, error_code);
Avi Kivity37a7d8b2007-01-05 16:36:56 -0800378 kvm_mmu_audit(vcpu, "pre page fault");
Avi Kivity714b93d2007-01-05 16:36:53 -0800379
Avi Kivitye2dec932007-01-05 16:36:54 -0800380 r = mmu_topup_memory_caches(vcpu);
381 if (r)
382 return r;
Avi Kivity714b93d2007-01-05 16:36:53 -0800383
Izik Eidus72dc67a2008-02-10 18:04:15 +0200384 down_read(&vcpu->kvm->slots_lock);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800385 /*
386 * Look up the shadow pte for the faulting address.
387 */
Avi Kivity73b10872007-01-26 00:56:41 -0800388 r = FNAME(walk_addr)(&walker, vcpu, addr, write_fault, user_fault,
389 fetch_fault);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800390
391 /*
392 * The page is not mapped by the guest. Let the guest handle it.
393 */
Avi Kivity7993ba42007-01-26 00:56:41 -0800394 if (!r) {
395 pgprintk("%s: guest page fault\n", __FUNCTION__);
396 inject_page_fault(vcpu, addr, walker.error_code);
Zhang Xiantaoad312c72007-12-13 23:50:52 +0800397 vcpu->arch.last_pt_write_count = 0; /* reset fork detector */
Izik Eidus72dc67a2008-02-10 18:04:15 +0200398 up_read(&vcpu->kvm->slots_lock);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800399 return 0;
400 }
401
Izik Eidus72dc67a2008-02-10 18:04:15 +0200402 down_read(&current->mm->mmap_sem);
Avi Kivityd7824ff2007-12-30 12:29:05 +0200403 page = gfn_to_page(vcpu->kvm, walker.gfn);
Izik Eidus72dc67a2008-02-10 18:04:15 +0200404 up_read(&current->mm->mmap_sem);
Avi Kivityd7824ff2007-12-30 12:29:05 +0200405
Marcelo Tosattiaaee2c92007-12-20 19:18:26 -0500406 spin_lock(&vcpu->kvm->mmu_lock);
Avi Kivityeb787d12007-12-31 15:27:49 +0200407 kvm_mmu_free_some_pages(vcpu);
Avi Kivity97a0a012007-05-31 15:08:29 +0300408 shadow_pte = FNAME(fetch)(vcpu, addr, &walker, user_fault, write_fault,
Avi Kivityd7824ff2007-12-30 12:29:05 +0200409 &write_pt, page);
Avi Kivity97a0a012007-05-31 15:08:29 +0300410 pgprintk("%s: shadow pte %p %llx ptwrite %d\n", __FUNCTION__,
411 shadow_pte, *shadow_pte, write_pt);
Avi Kivitycea0f0e2007-01-05 16:36:43 -0800412
Avi Kivitya25f7e12007-04-30 17:05:38 +0300413 if (!write_pt)
Zhang Xiantaoad312c72007-12-13 23:50:52 +0800414 vcpu->arch.last_pt_write_count = 0; /* reset fork detector */
Avi Kivitya25f7e12007-04-30 17:05:38 +0300415
Avi Kivity6aa8b732006-12-10 02:21:36 -0800416 /*
417 * mmio: emulate if accessible, otherwise its a guest fault.
418 */
Marcelo Tosatti10589a42007-12-20 19:18:22 -0500419 if (shadow_pte && is_io_pte(*shadow_pte)) {
Marcelo Tosattiaaee2c92007-12-20 19:18:26 -0500420 spin_unlock(&vcpu->kvm->mmu_lock);
Izik Eidus72dc67a2008-02-10 18:04:15 +0200421 up_read(&vcpu->kvm->slots_lock);
Avi Kivity7993ba42007-01-26 00:56:41 -0800422 return 1;
Marcelo Tosatti10589a42007-12-20 19:18:22 -0500423 }
Avi Kivity6aa8b732006-12-10 02:21:36 -0800424
Avi Kivity1165f5f2007-04-19 17:27:43 +0300425 ++vcpu->stat.pf_fixed;
Avi Kivity37a7d8b2007-01-05 16:36:56 -0800426 kvm_mmu_audit(vcpu, "post page fault (fixed)");
Marcelo Tosattiaaee2c92007-12-20 19:18:26 -0500427 spin_unlock(&vcpu->kvm->mmu_lock);
Izik Eidus72dc67a2008-02-10 18:04:15 +0200428 up_read(&vcpu->kvm->slots_lock);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800429
Avi Kivitycea0f0e2007-01-05 16:36:43 -0800430 return write_pt;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800431}
432
433static gpa_t FNAME(gva_to_gpa)(struct kvm_vcpu *vcpu, gva_t vaddr)
434{
435 struct guest_walker walker;
Avi Kivitye119d112007-02-12 00:54:36 -0800436 gpa_t gpa = UNMAPPED_GVA;
437 int r;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800438
Avi Kivitye119d112007-02-12 00:54:36 -0800439 r = FNAME(walk_addr)(&walker, vcpu, vaddr, 0, 0, 0);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800440
Avi Kivitye119d112007-02-12 00:54:36 -0800441 if (r) {
Avi Kivity1755fbc2007-11-21 14:44:45 +0200442 gpa = gfn_to_gpa(walker.gfn);
Avi Kivitye119d112007-02-12 00:54:36 -0800443 gpa |= vaddr & ~PAGE_MASK;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800444 }
445
446 return gpa;
447}
448
Avi Kivityc7addb92007-09-16 18:58:32 +0200449static void FNAME(prefetch_page)(struct kvm_vcpu *vcpu,
450 struct kvm_mmu_page *sp)
451{
Marcelo Tosatti7ec54582007-12-20 19:18:23 -0500452 int i, offset = 0, r = 0;
453 pt_element_t pt;
Avi Kivityc7addb92007-09-16 18:58:32 +0200454
Avi Kivitye5a4c8c2007-11-20 21:39:54 +0200455 if (sp->role.metaphysical
456 || (PTTYPE == 32 && sp->role.level > PT_PAGE_TABLE_LEVEL)) {
Avi Kivityc7addb92007-09-16 18:58:32 +0200457 nonpaging_prefetch_page(vcpu, sp);
458 return;
459 }
460
Avi Kivitye5a4c8c2007-11-20 21:39:54 +0200461 if (PTTYPE == 32)
462 offset = sp->role.quadrant << PT64_LEVEL_BITS;
Marcelo Tosatti7ec54582007-12-20 19:18:23 -0500463
464 for (i = 0; i < PT64_ENT_PER_PAGE; ++i) {
465 gpa_t pte_gpa = gfn_to_gpa(sp->gfn);
466 pte_gpa += (i+offset) * sizeof(pt_element_t);
467
468 r = kvm_read_guest_atomic(vcpu->kvm, pte_gpa, &pt,
469 sizeof(pt_element_t));
470 if (r || is_present_pte(pt))
Avi Kivityc7addb92007-09-16 18:58:32 +0200471 sp->spt[i] = shadow_trap_nonpresent_pte;
472 else
473 sp->spt[i] = shadow_notrap_nonpresent_pte;
Marcelo Tosatti7ec54582007-12-20 19:18:23 -0500474 }
Avi Kivityc7addb92007-09-16 18:58:32 +0200475}
476
Avi Kivity6aa8b732006-12-10 02:21:36 -0800477#undef pt_element_t
478#undef guest_walker
479#undef FNAME
480#undef PT_BASE_ADDR_MASK
481#undef PT_INDEX
482#undef SHADOW_PT_INDEX
483#undef PT_LEVEL_MASK
Avi Kivity6aa8b732006-12-10 02:21:36 -0800484#undef PT_DIR_BASE_ADDR_MASK
Avi Kivityc7addb92007-09-16 18:58:32 +0200485#undef PT_LEVEL_BITS
Avi Kivitycea0f0e2007-01-05 16:36:43 -0800486#undef PT_MAX_FULL_LEVELS
Avi Kivity5fb07dd2007-11-21 12:35:07 +0200487#undef gpte_to_gfn
488#undef gpte_to_gfn_pde
Marcelo Tosattib3e4e632007-12-07 07:56:58 -0500489#undef CMPXCHG