blob: ad75d77999d085037469539d42a5df140533971d [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.
Nicolas Kaiser9611c182010-10-06 14:23:22 +020010 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
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
Nadav Har'El37406aa2013-08-05 11:07:12 +030026/*
27 * This is used to catch non optimized PT_GUEST_(DIRTY|ACCESS)_SHIFT macro
28 * uses for EPT without A/D paging type.
29 */
30extern u64 __pure __using_nonexistent_pte_bit(void)
31 __compiletime_error("wrong use of PT_GUEST_(DIRTY|ACCESS)_SHIFT");
32
Avi Kivity6aa8b732006-12-10 02:21:36 -080033#if PTTYPE == 64
34 #define pt_element_t u64
35 #define guest_walker guest_walker64
36 #define FNAME(name) paging##64_##name
37 #define PT_BASE_ADDR_MASK PT64_BASE_ADDR_MASK
Joerg Roedele04da982009-07-27 16:30:45 +020038 #define PT_LVL_ADDR_MASK(lvl) PT64_LVL_ADDR_MASK(lvl)
39 #define PT_LVL_OFFSET_MASK(lvl) PT64_LVL_OFFSET_MASK(lvl)
Avi Kivity6aa8b732006-12-10 02:21:36 -080040 #define PT_INDEX(addr, level) PT64_INDEX(addr, level)
Avi Kivityc7addb92007-09-16 18:58:32 +020041 #define PT_LEVEL_BITS PT64_LEVEL_BITS
Gleb Natapovd8089ba2013-08-05 11:07:10 +030042 #define PT_GUEST_ACCESSED_MASK PT_ACCESSED_MASK
43 #define PT_GUEST_DIRTY_MASK PT_DIRTY_MASK
44 #define PT_GUEST_DIRTY_SHIFT PT_DIRTY_SHIFT
45 #define PT_GUEST_ACCESSED_SHIFT PT_ACCESSED_SHIFT
Avi Kivitycea0f0e2007-01-05 16:36:43 -080046 #ifdef CONFIG_X86_64
47 #define PT_MAX_FULL_LEVELS 4
Marcelo Tosattib3e4e632007-12-07 07:56:58 -050048 #define CMPXCHG cmpxchg
Avi Kivitycea0f0e2007-01-05 16:36:43 -080049 #else
Marcelo Tosattib3e4e632007-12-07 07:56:58 -050050 #define CMPXCHG cmpxchg64
Avi Kivitycea0f0e2007-01-05 16:36:43 -080051 #define PT_MAX_FULL_LEVELS 2
52 #endif
Avi Kivity6aa8b732006-12-10 02:21:36 -080053#elif PTTYPE == 32
54 #define pt_element_t u32
55 #define guest_walker guest_walker32
56 #define FNAME(name) paging##32_##name
57 #define PT_BASE_ADDR_MASK PT32_BASE_ADDR_MASK
Joerg Roedele04da982009-07-27 16:30:45 +020058 #define PT_LVL_ADDR_MASK(lvl) PT32_LVL_ADDR_MASK(lvl)
59 #define PT_LVL_OFFSET_MASK(lvl) PT32_LVL_OFFSET_MASK(lvl)
Avi Kivity6aa8b732006-12-10 02:21:36 -080060 #define PT_INDEX(addr, level) PT32_INDEX(addr, level)
Avi Kivityc7addb92007-09-16 18:58:32 +020061 #define PT_LEVEL_BITS PT32_LEVEL_BITS
Avi Kivitycea0f0e2007-01-05 16:36:43 -080062 #define PT_MAX_FULL_LEVELS 2
Gleb Natapovd8089ba2013-08-05 11:07:10 +030063 #define PT_GUEST_ACCESSED_MASK PT_ACCESSED_MASK
64 #define PT_GUEST_DIRTY_MASK PT_DIRTY_MASK
65 #define PT_GUEST_DIRTY_SHIFT PT_DIRTY_SHIFT
66 #define PT_GUEST_ACCESSED_SHIFT PT_ACCESSED_SHIFT
Marcelo Tosattib3e4e632007-12-07 07:56:58 -050067 #define CMPXCHG cmpxchg
Nadav Har'El37406aa2013-08-05 11:07:12 +030068#elif PTTYPE == PTTYPE_EPT
69 #define pt_element_t u64
70 #define guest_walker guest_walkerEPT
71 #define FNAME(name) ept_##name
72 #define PT_BASE_ADDR_MASK PT64_BASE_ADDR_MASK
73 #define PT_LVL_ADDR_MASK(lvl) PT64_LVL_ADDR_MASK(lvl)
74 #define PT_LVL_OFFSET_MASK(lvl) PT64_LVL_OFFSET_MASK(lvl)
75 #define PT_INDEX(addr, level) PT64_INDEX(addr, level)
76 #define PT_LEVEL_BITS PT64_LEVEL_BITS
77 #define PT_GUEST_ACCESSED_MASK 0
78 #define PT_GUEST_DIRTY_MASK 0
79 #define PT_GUEST_DIRTY_SHIFT __using_nonexistent_pte_bit()
80 #define PT_GUEST_ACCESSED_SHIFT __using_nonexistent_pte_bit()
81 #define CMPXCHG cmpxchg64
82 #define PT_MAX_FULL_LEVELS 4
Avi Kivity6aa8b732006-12-10 02:21:36 -080083#else
84 #error Invalid PTTYPE value
85#endif
86
Joerg Roedele04da982009-07-27 16:30:45 +020087#define gpte_to_gfn_lvl FNAME(gpte_to_gfn_lvl)
88#define gpte_to_gfn(pte) gpte_to_gfn_lvl((pte), PT_PAGE_TABLE_LEVEL)
Avi Kivity5fb07dd2007-11-21 12:35:07 +020089
Avi Kivity6aa8b732006-12-10 02:21:36 -080090/*
91 * The guest_walker structure emulates the behavior of the hardware page
92 * table walker.
93 */
94struct guest_walker {
95 int level;
Avi Kivity8cbc7062012-09-16 14:18:51 +030096 unsigned max_level;
Avi Kivitycea0f0e2007-01-05 16:36:43 -080097 gfn_t table_gfn[PT_MAX_FULL_LEVELS];
Marcelo Tosatti78190262007-12-11 19:12:27 -050098 pt_element_t ptes[PT_MAX_FULL_LEVELS];
Xiao Guangrong189be382010-08-22 19:13:33 +080099 pt_element_t prefetch_ptes[PTE_PREFETCH_NUM];
Marcelo Tosatti78190262007-12-11 19:12:27 -0500100 gpa_t pte_gpa[PT_MAX_FULL_LEVELS];
Avi Kivity8cbc7062012-09-16 14:18:51 +0300101 pt_element_t __user *ptep_user[PT_MAX_FULL_LEVELS];
Paolo Bonziniba6a3542013-09-09 13:52:33 +0200102 bool pte_writable[PT_MAX_FULL_LEVELS];
Avi Kivityfe135d22007-12-09 16:15:46 +0200103 unsigned pt_access;
104 unsigned pte_access;
Avi Kivity815af8d2007-01-05 16:36:44 -0800105 gfn_t gfn;
Avi Kivity8c28d032010-11-22 17:53:27 +0200106 struct x86_exception fault;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800107};
108
Joerg Roedele04da982009-07-27 16:30:45 +0200109static gfn_t gpte_to_gfn_lvl(pt_element_t gpte, int lvl)
Avi Kivity5fb07dd2007-11-21 12:35:07 +0200110{
Joerg Roedele04da982009-07-27 16:30:45 +0200111 return (gpte & PT_LVL_ADDR_MASK(lvl)) >> PAGE_SHIFT;
Avi Kivity5fb07dd2007-11-21 12:35:07 +0200112}
113
Nadav Har'El0ad805a2013-08-05 11:07:09 +0300114static inline void FNAME(protect_clean_gpte)(unsigned *access, unsigned gpte)
115{
116 unsigned mask;
117
Gleb Natapov61719a82013-08-05 11:07:11 +0300118 /* dirty bit is not supported, so no need to track it */
119 if (!PT_GUEST_DIRTY_MASK)
120 return;
121
Nadav Har'El0ad805a2013-08-05 11:07:09 +0300122 BUILD_BUG_ON(PT_WRITABLE_MASK != ACC_WRITE_MASK);
123
124 mask = (unsigned)~ACC_WRITE_MASK;
125 /* Allow write access to dirty gptes */
Gleb Natapovd8089ba2013-08-05 11:07:10 +0300126 mask |= (gpte >> (PT_GUEST_DIRTY_SHIFT - PT_WRITABLE_SHIFT)) &
127 PT_WRITABLE_MASK;
Nadav Har'El0ad805a2013-08-05 11:07:09 +0300128 *access &= mask;
129}
130
131static bool FNAME(is_rsvd_bits_set)(struct kvm_mmu *mmu, u64 gpte, int level)
132{
Yang Zhang25d92082013-08-06 12:00:32 +0300133 int bit7 = (gpte >> 7) & 1, low6 = gpte & 0x3f;
Nadav Har'El0ad805a2013-08-05 11:07:09 +0300134
Yang Zhang25d92082013-08-06 12:00:32 +0300135 return (gpte & mmu->rsvd_bits_mask[bit7][level-1]) |
136 ((mmu->bad_mt_xwr & (1ull << low6)) != 0);
Nadav Har'El0ad805a2013-08-05 11:07:09 +0300137}
138
139static inline int FNAME(is_present_gpte)(unsigned long pte)
140{
Nadav Har'El37406aa2013-08-05 11:07:12 +0300141#if PTTYPE != PTTYPE_EPT
Nadav Har'El0ad805a2013-08-05 11:07:09 +0300142 return is_present_gpte(pte);
Nadav Har'El37406aa2013-08-05 11:07:12 +0300143#else
144 return pte & 7;
145#endif
Nadav Har'El0ad805a2013-08-05 11:07:09 +0300146}
147
Roedel, Joerga78484c2011-04-20 15:33:16 +0200148static int FNAME(cmpxchg_gpte)(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
Takuya Yoshikawac8cfbb52011-05-01 14:33:07 +0900149 pt_element_t __user *ptep_user, unsigned index,
150 pt_element_t orig_pte, pt_element_t new_pte)
Marcelo Tosattib3e4e632007-12-07 07:56:58 -0500151{
Takuya Yoshikawac8cfbb52011-05-01 14:33:07 +0900152 int npages;
Marcelo Tosattib3e4e632007-12-07 07:56:58 -0500153 pt_element_t ret;
154 pt_element_t *table;
155 struct page *page;
156
Takuya Yoshikawac8cfbb52011-05-01 14:33:07 +0900157 npages = get_user_pages_fast((unsigned long)ptep_user, 1, 1, &page);
158 /* Check if the user is doing something meaningless. */
159 if (unlikely(npages != 1))
Roedel, Joerga78484c2011-04-20 15:33:16 +0200160 return -EFAULT;
161
Cong Wang8fd75e12011-11-25 23:14:17 +0800162 table = kmap_atomic(page);
Marcelo Tosattib3e4e632007-12-07 07:56:58 -0500163 ret = CMPXCHG(&table[index], orig_pte, new_pte);
Cong Wang8fd75e12011-11-25 23:14:17 +0800164 kunmap_atomic(table);
Marcelo Tosattib3e4e632007-12-07 07:56:58 -0500165
166 kvm_release_page_dirty(page);
167
168 return (ret != orig_pte);
169}
170
Nadav Har'El0ad805a2013-08-05 11:07:09 +0300171static bool FNAME(prefetch_invalid_gpte)(struct kvm_vcpu *vcpu,
172 struct kvm_mmu_page *sp, u64 *spte,
173 u64 gpte)
174{
175 if (FNAME(is_rsvd_bits_set)(&vcpu->arch.mmu, gpte, PT_PAGE_TABLE_LEVEL))
176 goto no_present;
177
178 if (!FNAME(is_present_gpte)(gpte))
179 goto no_present;
180
Gleb Natapov61719a82013-08-05 11:07:11 +0300181 /* if accessed bit is not supported prefetch non accessed gpte */
182 if (PT_GUEST_ACCESSED_MASK && !(gpte & PT_GUEST_ACCESSED_MASK))
Nadav Har'El0ad805a2013-08-05 11:07:09 +0300183 goto no_present;
184
185 return false;
186
187no_present:
188 drop_spte(vcpu->kvm, spte);
189 return true;
190}
191
192static inline unsigned FNAME(gpte_access)(struct kvm_vcpu *vcpu, u64 gpte)
193{
194 unsigned access;
Nadav Har'El37406aa2013-08-05 11:07:12 +0300195#if PTTYPE == PTTYPE_EPT
196 access = ((gpte & VMX_EPT_WRITABLE_MASK) ? ACC_WRITE_MASK : 0) |
197 ((gpte & VMX_EPT_EXECUTABLE_MASK) ? ACC_EXEC_MASK : 0) |
198 ACC_USER_MASK;
199#else
Nadav Har'El0ad805a2013-08-05 11:07:09 +0300200 access = (gpte & (PT_WRITABLE_MASK | PT_USER_MASK)) | ACC_EXEC_MASK;
201 access &= ~(gpte >> PT64_NX_SHIFT);
Nadav Har'El37406aa2013-08-05 11:07:12 +0300202#endif
Nadav Har'El0ad805a2013-08-05 11:07:09 +0300203
204 return access;
205}
206
Avi Kivity8cbc7062012-09-16 14:18:51 +0300207static int FNAME(update_accessed_dirty_bits)(struct kvm_vcpu *vcpu,
208 struct kvm_mmu *mmu,
209 struct guest_walker *walker,
210 int write_fault)
211{
212 unsigned level, index;
213 pt_element_t pte, orig_pte;
214 pt_element_t __user *ptep_user;
215 gfn_t table_gfn;
216 int ret;
217
Gleb Natapov61719a82013-08-05 11:07:11 +0300218 /* dirty/accessed bits are not supported, so no need to update them */
219 if (!PT_GUEST_DIRTY_MASK)
220 return 0;
221
Avi Kivity8cbc7062012-09-16 14:18:51 +0300222 for (level = walker->max_level; level >= walker->level; --level) {
223 pte = orig_pte = walker->ptes[level - 1];
224 table_gfn = walker->table_gfn[level - 1];
225 ptep_user = walker->ptep_user[level - 1];
226 index = offset_in_page(ptep_user) / sizeof(pt_element_t);
Gleb Natapovd8089ba2013-08-05 11:07:10 +0300227 if (!(pte & PT_GUEST_ACCESSED_MASK)) {
Avi Kivity8cbc7062012-09-16 14:18:51 +0300228 trace_kvm_mmu_set_accessed_bit(table_gfn, index, sizeof(pte));
Gleb Natapovd8089ba2013-08-05 11:07:10 +0300229 pte |= PT_GUEST_ACCESSED_MASK;
Avi Kivity8cbc7062012-09-16 14:18:51 +0300230 }
Nadav Har'El0ad805a2013-08-05 11:07:09 +0300231 if (level == walker->level && write_fault &&
Gleb Natapovd8089ba2013-08-05 11:07:10 +0300232 !(pte & PT_GUEST_DIRTY_MASK)) {
Avi Kivity8cbc7062012-09-16 14:18:51 +0300233 trace_kvm_mmu_set_dirty_bit(table_gfn, index, sizeof(pte));
Gleb Natapovd8089ba2013-08-05 11:07:10 +0300234 pte |= PT_GUEST_DIRTY_MASK;
Avi Kivity8cbc7062012-09-16 14:18:51 +0300235 }
236 if (pte == orig_pte)
237 continue;
238
Paolo Bonziniba6a3542013-09-09 13:52:33 +0200239 /*
240 * If the slot is read-only, simply do not process the accessed
241 * and dirty bits. This is the correct thing to do if the slot
242 * is ROM, and page tables in read-as-ROM/write-as-MMIO slots
243 * are only supported if the accessed and dirty bits are already
244 * set in the ROM (so that MMIO writes are never needed).
245 *
246 * Note that NPT does not allow this at all and faults, since
247 * it always wants nested page table entries for the guest
248 * page tables to be writable. And EPT works but will simply
249 * overwrite the read-only memory to set the accessed and dirty
250 * bits.
251 */
252 if (unlikely(!walker->pte_writable[level - 1]))
253 continue;
254
Avi Kivity8cbc7062012-09-16 14:18:51 +0300255 ret = FNAME(cmpxchg_gpte)(vcpu, mmu, ptep_user, index, orig_pte, pte);
256 if (ret)
257 return ret;
258
259 mark_page_dirty(vcpu->kvm, table_gfn);
260 walker->ptes[level] = pte;
261 }
262 return 0;
263}
264
Avi Kivityac79c972007-01-05 16:36:40 -0800265/*
266 * Fetch a guest pte for a guest virtual address
267 */
Joerg Roedel1e301fe2010-09-10 17:30:47 +0200268static int FNAME(walk_addr_generic)(struct guest_walker *walker,
269 struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
Xiao Guangrong33770782010-09-28 17:03:14 +0800270 gva_t addr, u32 access)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800271{
Avi Kivity8cbc7062012-09-16 14:18:51 +0300272 int ret;
Avi Kivity42bf3f02007-10-17 12:18:47 +0200273 pt_element_t pte;
Borislav Petkovb7233632011-05-30 22:11:17 +0200274 pt_element_t __user *uninitialized_var(ptep_user);
Avi Kivitycea0f0e2007-01-05 16:36:43 -0800275 gfn_t table_gfn;
Gleb Natapovb0cfeb52013-01-08 10:49:00 +0200276 unsigned index, pt_access, pte_access, accessed_dirty;
Avi Kivity42bf3f02007-10-17 12:18:47 +0200277 gpa_t pte_gpa;
Takuya Yoshikawa134291b2011-07-01 01:34:56 +0900278 int offset;
279 const int write_fault = access & PFERR_WRITE_MASK;
280 const int user_fault = access & PFERR_USER_MASK;
281 const int fetch_fault = access & PFERR_FETCH_MASK;
282 u16 errcode = 0;
Avi Kivity13d22b62012-09-12 15:12:09 +0300283 gpa_t real_gpa;
284 gfn_t gfn;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800285
Xiao Guangrong6fbc2772012-06-20 16:00:00 +0800286 trace_kvm_mmu_pagetable_walk(addr, access);
Takuya Yoshikawa92c1c1e2011-07-01 01:36:07 +0900287retry_walk:
Joerg Roedel1e301fe2010-09-10 17:30:47 +0200288 walker->level = mmu->root_level;
289 pte = mmu->get_cr3(vcpu);
290
Avi Kivity1b0973b2007-01-05 16:36:41 -0800291#if PTTYPE == 64
Joerg Roedel1e301fe2010-09-10 17:30:47 +0200292 if (walker->level == PT32E_ROOT_LEVEL) {
Avi Kivitye4e517b2011-07-28 11:36:17 +0300293 pte = mmu->get_pdptr(vcpu, (addr >> 30) & 3);
Avi Kivity07420172009-07-06 12:21:32 +0300294 trace_kvm_mmu_paging_element(pte, walker->level);
Nadav Har'El0ad805a2013-08-05 11:07:09 +0300295 if (!FNAME(is_present_gpte)(pte))
Avi Kivityf59c1d22010-07-06 16:20:43 +0300296 goto error;
Avi Kivity1b0973b2007-01-05 16:36:41 -0800297 --walker->level;
298 }
299#endif
Avi Kivity8cbc7062012-09-16 14:18:51 +0300300 walker->max_level = walker->level;
Avi Kivitya9058ec2006-12-29 16:49:37 -0800301 ASSERT((!is_long_mode(vcpu) && is_pae(vcpu)) ||
Joerg Roedel1e301fe2010-09-10 17:30:47 +0200302 (mmu->get_cr3(vcpu) & CR3_NONPAE_RESERVED_BITS) == 0);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800303
Gleb Natapovd8089ba2013-08-05 11:07:10 +0300304 accessed_dirty = PT_GUEST_ACCESSED_MASK;
Avi Kivity13d22b62012-09-12 15:12:09 +0300305 pt_access = pte_access = ACC_ALL;
306 ++walker->level;
Avi Kivityac79c972007-01-05 16:36:40 -0800307
Avi Kivity13d22b62012-09-12 15:12:09 +0300308 do {
Takuya Yoshikawa6e2ca7d2011-04-22 00:34:44 +0900309 gfn_t real_gfn;
310 unsigned long host_addr;
311
Avi Kivity13d22b62012-09-12 15:12:09 +0300312 pt_access &= pte_access;
313 --walker->level;
314
Avi Kivity42bf3f02007-10-17 12:18:47 +0200315 index = PT_INDEX(addr, walker->level);
Avi Kivityac79c972007-01-05 16:36:40 -0800316
Avi Kivity5fb07dd2007-11-21 12:35:07 +0200317 table_gfn = gpte_to_gfn(pte);
Joerg Roedel2329d462010-09-10 17:30:52 +0200318 offset = index * sizeof(pt_element_t);
319 pte_gpa = gfn_to_gpa(table_gfn) + offset;
Avi Kivity42bf3f02007-10-17 12:18:47 +0200320 walker->table_gfn[walker->level - 1] = table_gfn;
Marcelo Tosatti78190262007-12-11 19:12:27 -0500321 walker->pte_gpa[walker->level - 1] = pte_gpa;
Avi Kivityac79c972007-01-05 16:36:40 -0800322
Takuya Yoshikawa6e2ca7d2011-04-22 00:34:44 +0900323 real_gfn = mmu->translate_gpa(vcpu, gfn_to_gpa(table_gfn),
324 PFERR_USER_MASK|PFERR_WRITE_MASK);
Takuya Yoshikawa134291b2011-07-01 01:34:56 +0900325 if (unlikely(real_gfn == UNMAPPED_GVA))
326 goto error;
Takuya Yoshikawa6e2ca7d2011-04-22 00:34:44 +0900327 real_gfn = gpa_to_gfn(real_gfn);
328
Paolo Bonziniba6a3542013-09-09 13:52:33 +0200329 host_addr = gfn_to_hva_prot(vcpu->kvm, real_gfn,
330 &walker->pte_writable[walker->level - 1]);
Takuya Yoshikawa134291b2011-07-01 01:34:56 +0900331 if (unlikely(kvm_is_error_hva(host_addr)))
332 goto error;
Takuya Yoshikawa6e2ca7d2011-04-22 00:34:44 +0900333
334 ptep_user = (pt_element_t __user *)((void *)host_addr + offset);
Takuya Yoshikawa134291b2011-07-01 01:34:56 +0900335 if (unlikely(__copy_from_user(&pte, ptep_user, sizeof(pte))))
336 goto error;
Avi Kivity8cbc7062012-09-16 14:18:51 +0300337 walker->ptep_user[walker->level - 1] = ptep_user;
Marcelo Tosattia6085fb2010-01-14 17:41:27 -0200338
Avi Kivity07420172009-07-06 12:21:32 +0300339 trace_kvm_mmu_paging_element(pte, walker->level);
Avi Kivity42bf3f02007-10-17 12:18:47 +0200340
Nadav Har'El0ad805a2013-08-05 11:07:09 +0300341 if (unlikely(!FNAME(is_present_gpte)(pte)))
Takuya Yoshikawa134291b2011-07-01 01:34:56 +0900342 goto error;
Avi Kivity7993ba42007-01-26 00:56:41 -0800343
Nadav Har'El0ad805a2013-08-05 11:07:09 +0300344 if (unlikely(FNAME(is_rsvd_bits_set)(mmu, pte,
345 walker->level))) {
Takuya Yoshikawa134291b2011-07-01 01:34:56 +0900346 errcode |= PFERR_RSVD_MASK | PFERR_PRESENT_MASK;
347 goto error;
Avi Kivityf59c1d22010-07-06 16:20:43 +0300348 }
Dong, Eddie82725b22009-03-30 16:21:08 +0800349
Avi Kivityb514c302012-09-16 15:03:02 +0300350 accessed_dirty &= pte;
Nadav Har'El0ad805a2013-08-05 11:07:09 +0300351 pte_access = pt_access & FNAME(gpte_access)(vcpu, pte);
Avi Kivity73b10872007-01-26 00:56:41 -0800352
Marcelo Tosatti78190262007-12-11 19:12:27 -0500353 walker->ptes[walker->level - 1] = pte;
Avi Kivity6fd01b72012-09-12 20:46:56 +0300354 } while (!is_last_gpte(mmu, walker->level, pte));
Avi Kivity42bf3f02007-10-17 12:18:47 +0200355
Avi Kivity71331a12012-09-16 14:49:15 +0300356 if (unlikely(permission_fault(mmu, pte_access, access))) {
Takuya Yoshikawa134291b2011-07-01 01:34:56 +0900357 errcode |= PFERR_PRESENT_MASK;
Avi Kivityf59c1d22010-07-06 16:20:43 +0300358 goto error;
Takuya Yoshikawa134291b2011-07-01 01:34:56 +0900359 }
Avi Kivityf59c1d22010-07-06 16:20:43 +0300360
Avi Kivity13d22b62012-09-12 15:12:09 +0300361 gfn = gpte_to_gfn_lvl(pte, walker->level);
362 gfn += (addr & PT_LVL_OFFSET_MASK(walker->level)) >> PAGE_SHIFT;
363
364 if (PTTYPE == 32 && walker->level == PT_DIRECTORY_LEVEL && is_cpuid_PSE36())
365 gfn += pse36_gfn_delta(pte);
366
Avi Kivityc5421512012-09-19 19:33:48 +0300367 real_gpa = mmu->translate_gpa(vcpu, gfn_to_gpa(gfn), access);
Avi Kivity13d22b62012-09-12 15:12:09 +0300368 if (real_gpa == UNMAPPED_GVA)
369 return 0;
370
371 walker->gfn = real_gpa >> PAGE_SHIFT;
372
Avi Kivity8ea667f2012-09-12 13:44:53 +0300373 if (!write_fault)
Nadav Har'El0ad805a2013-08-05 11:07:09 +0300374 FNAME(protect_clean_gpte)(&pte_access, pte);
Gleb Natapov908e7d72012-12-27 14:44:58 +0200375 else
376 /*
Gleb Natapov61719a82013-08-05 11:07:11 +0300377 * On a write fault, fold the dirty bit into accessed_dirty.
378 * For modes without A/D bits support accessed_dirty will be
379 * always clear.
Gleb Natapov908e7d72012-12-27 14:44:58 +0200380 */
Gleb Natapovd8089ba2013-08-05 11:07:10 +0300381 accessed_dirty &= pte >>
382 (PT_GUEST_DIRTY_SHIFT - PT_GUEST_ACCESSED_SHIFT);
Avi Kivityb514c302012-09-16 15:03:02 +0300383
384 if (unlikely(!accessed_dirty)) {
385 ret = FNAME(update_accessed_dirty_bits)(vcpu, mmu, walker, write_fault);
386 if (unlikely(ret < 0))
387 goto error;
388 else if (ret)
389 goto retry_walk;
390 }
Avi Kivity42bf3f02007-10-17 12:18:47 +0200391
Avi Kivityfe135d22007-12-09 16:15:46 +0200392 walker->pt_access = pt_access;
393 walker->pte_access = pte_access;
394 pgprintk("%s: pte %llx pte_access %x pt_access %x\n",
Gui Jianfeng518c5a02010-05-05 09:58:33 +0800395 __func__, (u64)pte, pte_access, pt_access);
Avi Kivity7993ba42007-01-26 00:56:41 -0800396 return 1;
397
Avi Kivityf59c1d22010-07-06 16:20:43 +0300398error:
Takuya Yoshikawa134291b2011-07-01 01:34:56 +0900399 errcode |= write_fault | user_fault;
Yang, Wei Ye57d4a32011-06-03 11:14:16 +0800400 if (fetch_fault && (mmu->nx ||
401 kvm_read_cr4_bits(vcpu, X86_CR4_SMEP)))
Takuya Yoshikawa134291b2011-07-01 01:34:56 +0900402 errcode |= PFERR_FETCH_MASK;
Joerg Roedel8df25a32010-09-10 17:30:46 +0200403
Takuya Yoshikawa134291b2011-07-01 01:34:56 +0900404 walker->fault.vector = PF_VECTOR;
405 walker->fault.error_code_valid = true;
406 walker->fault.error_code = errcode;
Yang Zhang25d92082013-08-06 12:00:32 +0300407
408#if PTTYPE == PTTYPE_EPT
409 /*
410 * Use PFERR_RSVD_MASK in error_code to to tell if EPT
411 * misconfiguration requires to be injected. The detection is
412 * done by is_rsvd_bits_set() above.
413 *
414 * We set up the value of exit_qualification to inject:
415 * [2:0] - Derive from [2:0] of real exit_qualification at EPT violation
416 * [5:3] - Calculated by the page walk of the guest EPT page tables
417 * [7:8] - Derived from [7:8] of real exit_qualification
418 *
419 * The other bits are set to 0.
420 */
421 if (!(errcode & PFERR_RSVD_MASK)) {
422 vcpu->arch.exit_qualification &= 0x187;
423 vcpu->arch.exit_qualification |= ((pt_access & pte) & 0x7) << 3;
424 }
425#endif
Avi Kivity6389ee92010-11-29 16:12:30 +0200426 walker->fault.address = addr;
427 walker->fault.nested_page_fault = mmu != vcpu->arch.walk_mmu;
Joerg Roedel8df25a32010-09-10 17:30:46 +0200428
Avi Kivity8c28d032010-11-22 17:53:27 +0200429 trace_kvm_mmu_walker_error(walker->fault.error_code);
Shaohua Life551882007-07-23 14:51:39 +0800430 return 0;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800431}
432
Joerg Roedel1e301fe2010-09-10 17:30:47 +0200433static int FNAME(walk_addr)(struct guest_walker *walker,
Xiao Guangrong33770782010-09-28 17:03:14 +0800434 struct kvm_vcpu *vcpu, gva_t addr, u32 access)
Joerg Roedel1e301fe2010-09-10 17:30:47 +0200435{
436 return FNAME(walk_addr_generic)(walker, vcpu, &vcpu->arch.mmu, addr,
Xiao Guangrong33770782010-09-28 17:03:14 +0800437 access);
Joerg Roedel1e301fe2010-09-10 17:30:47 +0200438}
439
Nadav Har'El37406aa2013-08-05 11:07:12 +0300440#if PTTYPE != PTTYPE_EPT
Joerg Roedel6539e732010-09-10 17:30:50 +0200441static int FNAME(walk_addr_nested)(struct guest_walker *walker,
442 struct kvm_vcpu *vcpu, gva_t addr,
Xiao Guangrong33770782010-09-28 17:03:14 +0800443 u32 access)
Joerg Roedel6539e732010-09-10 17:30:50 +0200444{
445 return FNAME(walk_addr_generic)(walker, vcpu, &vcpu->arch.nested_mmu,
Xiao Guangrong33770782010-09-28 17:03:14 +0800446 addr, access);
Joerg Roedel6539e732010-09-10 17:30:50 +0200447}
Nadav Har'El37406aa2013-08-05 11:07:12 +0300448#endif
Joerg Roedel6539e732010-09-10 17:30:50 +0200449
Xiao Guangrongbd6360c2012-10-16 20:10:12 +0800450static bool
451FNAME(prefetch_gpte)(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp,
452 u64 *spte, pt_element_t gpte, bool no_dirty_log)
453{
454 unsigned pte_access;
455 gfn_t gfn;
456 pfn_t pfn;
457
Nadav Har'El0ad805a2013-08-05 11:07:09 +0300458 if (FNAME(prefetch_invalid_gpte)(vcpu, sp, spte, gpte))
Xiao Guangrongbd6360c2012-10-16 20:10:12 +0800459 return false;
460
461 pgprintk("%s: gpte %llx spte %p\n", __func__, (u64)gpte, spte);
462
463 gfn = gpte_to_gfn(gpte);
Nadav Har'El0ad805a2013-08-05 11:07:09 +0300464 pte_access = sp->role.access & FNAME(gpte_access)(vcpu, gpte);
465 FNAME(protect_clean_gpte)(&pte_access, gpte);
Xiao Guangrongbd6360c2012-10-16 20:10:12 +0800466 pfn = pte_prefetch_gfn_to_pfn(vcpu, gfn,
467 no_dirty_log && (pte_access & ACC_WRITE_MASK));
Xiao Guangrong81c52c52012-10-16 20:10:59 +0800468 if (is_error_pfn(pfn))
Xiao Guangrongbd6360c2012-10-16 20:10:12 +0800469 return false;
470
471 /*
472 * we call mmu_set_spte() with host_writable = true because
473 * pte_prefetch_gfn_to_pfn always gets a writable pfn.
474 */
Xiao Guangrongf7616202013-02-05 15:27:27 +0800475 mmu_set_spte(vcpu, spte, pte_access, 0, NULL, PT_PAGE_TABLE_LEVEL,
476 gfn, pfn, true, true);
Xiao Guangrongbd6360c2012-10-16 20:10:12 +0800477
478 return true;
479}
480
Xiao Guangrongac3cd032010-06-11 21:28:14 +0800481static void FNAME(update_pte)(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp,
Xiao Guangrong7c562522011-03-28 10:29:27 +0800482 u64 *spte, const void *pte)
Avi Kivity00284252007-05-01 16:53:31 +0300483{
Xiao Guangrongbd6360c2012-10-16 20:10:12 +0800484 pt_element_t gpte = *(const pt_element_t *)pte;
Avi Kivity00284252007-05-01 16:53:31 +0300485
Xiao Guangrongbd6360c2012-10-16 20:10:12 +0800486 FNAME(prefetch_gpte)(vcpu, sp, spte, gpte, false);
Avi Kivity00284252007-05-01 16:53:31 +0300487}
488
Avi Kivity39c8c672010-07-13 14:27:08 +0300489static bool FNAME(gpte_changed)(struct kvm_vcpu *vcpu,
490 struct guest_walker *gw, int level)
491{
Avi Kivity39c8c672010-07-13 14:27:08 +0300492 pt_element_t curr_pte;
Xiao Guangrong189be382010-08-22 19:13:33 +0800493 gpa_t base_gpa, pte_gpa = gw->pte_gpa[level - 1];
494 u64 mask;
495 int r, index;
Avi Kivity39c8c672010-07-13 14:27:08 +0300496
Xiao Guangrong189be382010-08-22 19:13:33 +0800497 if (level == PT_PAGE_TABLE_LEVEL) {
498 mask = PTE_PREFETCH_NUM * sizeof(pt_element_t) - 1;
499 base_gpa = pte_gpa & ~mask;
500 index = (pte_gpa - base_gpa) / sizeof(pt_element_t);
501
502 r = kvm_read_guest_atomic(vcpu->kvm, base_gpa,
503 gw->prefetch_ptes, sizeof(gw->prefetch_ptes));
504 curr_pte = gw->prefetch_ptes[index];
505 } else
506 r = kvm_read_guest_atomic(vcpu->kvm, pte_gpa,
Avi Kivity39c8c672010-07-13 14:27:08 +0300507 &curr_pte, sizeof(curr_pte));
Xiao Guangrong189be382010-08-22 19:13:33 +0800508
Avi Kivity39c8c672010-07-13 14:27:08 +0300509 return r || curr_pte != gw->ptes[level - 1];
510}
511
Xiao Guangrong189be382010-08-22 19:13:33 +0800512static void FNAME(pte_prefetch)(struct kvm_vcpu *vcpu, struct guest_walker *gw,
513 u64 *sptep)
Xiao Guangrong957ed9e2010-08-22 19:12:48 +0800514{
515 struct kvm_mmu_page *sp;
Xiao Guangrong189be382010-08-22 19:13:33 +0800516 pt_element_t *gptep = gw->prefetch_ptes;
Xiao Guangrong957ed9e2010-08-22 19:12:48 +0800517 u64 *spte;
Xiao Guangrong189be382010-08-22 19:13:33 +0800518 int i;
Xiao Guangrong957ed9e2010-08-22 19:12:48 +0800519
520 sp = page_header(__pa(sptep));
521
522 if (sp->role.level > PT_PAGE_TABLE_LEVEL)
523 return;
524
525 if (sp->role.direct)
526 return __direct_pte_prefetch(vcpu, sp, sptep);
527
528 i = (sptep - sp->spt) & ~(PTE_PREFETCH_NUM - 1);
Xiao Guangrong957ed9e2010-08-22 19:12:48 +0800529 spte = sp->spt + i;
530
531 for (i = 0; i < PTE_PREFETCH_NUM; i++, spte++) {
Xiao Guangrong957ed9e2010-08-22 19:12:48 +0800532 if (spte == sptep)
533 continue;
534
Xiao Guangrongc3707952011-07-12 03:28:04 +0800535 if (is_shadow_present_pte(*spte))
Xiao Guangrong957ed9e2010-08-22 19:12:48 +0800536 continue;
537
Xiao Guangrongbd6360c2012-10-16 20:10:12 +0800538 if (!FNAME(prefetch_gpte)(vcpu, sp, spte, gptep[i], true))
Xiao Guangrong957ed9e2010-08-22 19:12:48 +0800539 break;
Xiao Guangrong957ed9e2010-08-22 19:12:48 +0800540 }
541}
542
Avi Kivity6aa8b732006-12-10 02:21:36 -0800543/*
Avi Kivity6aa8b732006-12-10 02:21:36 -0800544 * Fetch a shadow pte for a specific level in the paging hierarchy.
Xiao Guangrongd4878f22012-10-16 20:08:43 +0800545 * If the guest tries to write a write-protected page, we need to
546 * emulate this operation, return 1 to indicate this case.
Avi Kivity6aa8b732006-12-10 02:21:36 -0800547 */
Xiao Guangrongd4878f22012-10-16 20:08:43 +0800548static int FNAME(fetch)(struct kvm_vcpu *vcpu, gva_t addr,
Avi Kivitye7a04c92008-12-25 15:10:50 +0200549 struct guest_walker *gw,
Xiao Guangrongc22885052013-01-08 14:36:04 +0800550 int write_fault, int hlevel,
Xiao Guangrongd4878f22012-10-16 20:08:43 +0800551 pfn_t pfn, bool map_writable, bool prefault)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800552{
Avi Kivity5991b332010-07-13 14:27:10 +0300553 struct kvm_mmu_page *sp = NULL;
Avi Kivity24157aa2010-07-13 14:27:11 +0300554 struct kvm_shadow_walk_iterator it;
Xiao Guangrongd4878f22012-10-16 20:08:43 +0800555 unsigned direct_access, access = gw->pt_access;
556 int top_level, emulate = 0;
Avi Kivityac79c972007-01-05 16:36:40 -0800557
Xiao Guangrongb36c7a72011-07-12 03:25:19 +0800558 direct_access = gw->pte_access;
Xiao Guangrong84754cd2010-06-30 16:05:00 +0800559
Avi Kivity5991b332010-07-13 14:27:10 +0300560 top_level = vcpu->arch.mmu.root_level;
561 if (top_level == PT32E_ROOT_LEVEL)
562 top_level = PT32_ROOT_LEVEL;
563 /*
564 * Verify that the top-level gpte is still there. Since the page
565 * is a root page, it is either write protected (and cannot be
566 * changed from now on) or it is invalid (in which case, we don't
567 * really care if it changes underneath us after this point).
568 */
569 if (FNAME(gpte_changed)(vcpu, gw, top_level))
570 goto out_gpte_changed;
571
Avi Kivity24157aa2010-07-13 14:27:11 +0300572 for (shadow_walk_init(&it, vcpu, addr);
573 shadow_walk_okay(&it) && it.level > gw->level;
574 shadow_walk_next(&it)) {
Avi Kivity0b3c9332010-07-13 14:27:09 +0300575 gfn_t table_gfn;
576
Xiao Guangronga30f47c2011-09-22 16:58:36 +0800577 clear_sp_write_flooding_count(it.sptep);
Avi Kivity24157aa2010-07-13 14:27:11 +0300578 drop_large_spte(vcpu, it.sptep);
Avi Kivitye7a04c92008-12-25 15:10:50 +0200579
Avi Kivity5991b332010-07-13 14:27:10 +0300580 sp = NULL;
Avi Kivity24157aa2010-07-13 14:27:11 +0300581 if (!is_shadow_present_pte(*it.sptep)) {
582 table_gfn = gw->table_gfn[it.level - 2];
583 sp = kvm_mmu_get_page(vcpu, table_gfn, addr, it.level-1,
584 false, access, it.sptep);
Avi Kivity5991b332010-07-13 14:27:10 +0300585 }
Avi Kivity0b3c9332010-07-13 14:27:09 +0300586
587 /*
588 * Verify that the gpte in the page we've just write
589 * protected is still there.
590 */
Avi Kivity24157aa2010-07-13 14:27:11 +0300591 if (FNAME(gpte_changed)(vcpu, gw, it.level - 1))
Avi Kivity0b3c9332010-07-13 14:27:09 +0300592 goto out_gpte_changed;
Avi Kivitye7a04c92008-12-25 15:10:50 +0200593
Avi Kivity5991b332010-07-13 14:27:10 +0300594 if (sp)
Yang Zhang7a1638c2013-08-05 11:07:13 +0300595 link_shadow_page(it.sptep, sp, PT_GUEST_ACCESSED_MASK);
Avi Kivitye7a04c92008-12-25 15:10:50 +0200596 }
597
Avi Kivity0b3c9332010-07-13 14:27:09 +0300598 for (;
Avi Kivity24157aa2010-07-13 14:27:11 +0300599 shadow_walk_okay(&it) && it.level > hlevel;
600 shadow_walk_next(&it)) {
Avi Kivity0b3c9332010-07-13 14:27:09 +0300601 gfn_t direct_gfn;
602
Xiao Guangronga30f47c2011-09-22 16:58:36 +0800603 clear_sp_write_flooding_count(it.sptep);
Avi Kivity24157aa2010-07-13 14:27:11 +0300604 validate_direct_spte(vcpu, it.sptep, direct_access);
Avi Kivity0b3c9332010-07-13 14:27:09 +0300605
Avi Kivity24157aa2010-07-13 14:27:11 +0300606 drop_large_spte(vcpu, it.sptep);
Avi Kivity0b3c9332010-07-13 14:27:09 +0300607
Avi Kivity24157aa2010-07-13 14:27:11 +0300608 if (is_shadow_present_pte(*it.sptep))
Avi Kivity0b3c9332010-07-13 14:27:09 +0300609 continue;
610
Avi Kivity24157aa2010-07-13 14:27:11 +0300611 direct_gfn = gw->gfn & ~(KVM_PAGES_PER_HPAGE(it.level) - 1);
Avi Kivity0b3c9332010-07-13 14:27:09 +0300612
Avi Kivity24157aa2010-07-13 14:27:11 +0300613 sp = kvm_mmu_get_page(vcpu, direct_gfn, addr, it.level-1,
614 true, direct_access, it.sptep);
Yang Zhang7a1638c2013-08-05 11:07:13 +0300615 link_shadow_page(it.sptep, sp, PT_GUEST_ACCESSED_MASK);
Avi Kivity0b3c9332010-07-13 14:27:09 +0300616 }
617
Xiao Guangronga30f47c2011-09-22 16:58:36 +0800618 clear_sp_write_flooding_count(it.sptep);
Xiao Guangrongf7616202013-02-05 15:27:27 +0800619 mmu_set_spte(vcpu, it.sptep, gw->pte_access, write_fault, &emulate,
620 it.level, gw->gfn, pfn, prefault, map_writable);
Xiao Guangrong189be382010-08-22 19:13:33 +0800621 FNAME(pte_prefetch)(vcpu, gw, it.sptep);
Avi Kivity0b3c9332010-07-13 14:27:09 +0300622
Xiao Guangrongd4878f22012-10-16 20:08:43 +0800623 return emulate;
Avi Kivity0b3c9332010-07-13 14:27:09 +0300624
625out_gpte_changed:
Avi Kivity5991b332010-07-13 14:27:10 +0300626 if (sp)
Avi Kivity24157aa2010-07-13 14:27:11 +0300627 kvm_mmu_put_page(sp, it.sptep);
Avi Kivity0b3c9332010-07-13 14:27:09 +0300628 kvm_release_pfn_clean(pfn);
Xiao Guangrongd4878f22012-10-16 20:08:43 +0800629 return 0;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800630}
631
Xiao Guangrong7751bab2013-01-08 14:36:51 +0800632 /*
633 * To see whether the mapped gfn can write its page table in the current
634 * mapping.
635 *
636 * It is the helper function of FNAME(page_fault). When guest uses large page
637 * size to map the writable gfn which is used as current page table, we should
638 * force kvm to use small page size to map it because new shadow page will be
639 * created when kvm establishes shadow page table that stop kvm using large
640 * page size. Do it early can avoid unnecessary #PF and emulation.
641 *
Xiao Guangrong93c05d32013-01-13 23:49:07 +0800642 * @write_fault_to_shadow_pgtable will return true if the fault gfn is
643 * currently used as its page table.
644 *
Xiao Guangrong7751bab2013-01-08 14:36:51 +0800645 * Note: the PDPT page table is not checked for PAE-32 bit guest. It is ok
646 * since the PDPT is always shadowed, that means, we can not use large page
647 * size to map the gfn which is used as PDPT.
648 */
649static bool
650FNAME(is_self_change_mapping)(struct kvm_vcpu *vcpu,
Xiao Guangrong93c05d32013-01-13 23:49:07 +0800651 struct guest_walker *walker, int user_fault,
652 bool *write_fault_to_shadow_pgtable)
Xiao Guangrong7751bab2013-01-08 14:36:51 +0800653{
654 int level;
655 gfn_t mask = ~(KVM_PAGES_PER_HPAGE(walker->level) - 1);
Xiao Guangrong93c05d32013-01-13 23:49:07 +0800656 bool self_changed = false;
Xiao Guangrong7751bab2013-01-08 14:36:51 +0800657
658 if (!(walker->pte_access & ACC_WRITE_MASK ||
659 (!is_write_protection(vcpu) && !user_fault)))
660 return false;
661
Xiao Guangrong93c05d32013-01-13 23:49:07 +0800662 for (level = walker->level; level <= walker->max_level; level++) {
663 gfn_t gfn = walker->gfn ^ walker->table_gfn[level - 1];
Xiao Guangrong7751bab2013-01-08 14:36:51 +0800664
Xiao Guangrong93c05d32013-01-13 23:49:07 +0800665 self_changed |= !(gfn & mask);
666 *write_fault_to_shadow_pgtable |= !gfn;
667 }
668
669 return self_changed;
Xiao Guangrong7751bab2013-01-08 14:36:51 +0800670}
671
Avi Kivity6aa8b732006-12-10 02:21:36 -0800672/*
Avi Kivity6aa8b732006-12-10 02:21:36 -0800673 * Page fault handler. There are several causes for a page fault:
674 * - there is no shadow pte for the guest pte
675 * - write access through a shadow pte marked read only so that we can set
676 * the dirty bit
677 * - write access to a shadow pte marked read only so we can update the page
678 * dirty bitmap, when userspace requests it
679 * - mmio access; in this case we will never install a present shadow pte
680 * - normal guest page fault due to the guest pte marked not present, not
681 * writable, or not executable
682 *
Avi Kivitye2dec932007-01-05 16:36:54 -0800683 * Returns: 1 if we need to emulate the instruction, 0 otherwise, or
684 * a negative value on error.
Avi Kivity6aa8b732006-12-10 02:21:36 -0800685 */
Gleb Natapov56028d02010-10-17 18:13:42 +0200686static int FNAME(page_fault)(struct kvm_vcpu *vcpu, gva_t addr, u32 error_code,
Xiao Guangrong78b2c542010-12-07 10:48:06 +0800687 bool prefault)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800688{
689 int write_fault = error_code & PFERR_WRITE_MASK;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800690 int user_fault = error_code & PFERR_USER_MASK;
691 struct guest_walker walker;
Avi Kivitye2dec932007-01-05 16:36:54 -0800692 int r;
Anthony Liguori35149e22008-04-02 14:46:56 -0500693 pfn_t pfn;
Joerg Roedel7e4e4052009-07-27 16:30:46 +0200694 int level = PT_PAGE_TABLE_LEVEL;
Andrea Arcangeli936a5fe2011-01-13 15:46:48 -0800695 int force_pt_level;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200696 unsigned long mmu_seq;
Xiao Guangrong93c05d32013-01-13 23:49:07 +0800697 bool map_writable, is_self_change_mapping;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800698
Harvey Harrisonb8688d52008-03-03 12:59:56 -0800699 pgprintk("%s: addr %lx err %x\n", __func__, addr, error_code);
Avi Kivity714b93d2007-01-05 16:36:53 -0800700
Xiao Guangrongf8f55942013-06-07 16:51:26 +0800701 if (unlikely(error_code & PFERR_RSVD_MASK)) {
702 r = handle_mmio_page_fault(vcpu, addr, error_code,
Xiao Guangrongce88dec2011-07-12 03:33:44 +0800703 mmu_is_nested(vcpu));
Xiao Guangrongf8f55942013-06-07 16:51:26 +0800704 if (likely(r != RET_MMIO_PF_INVALID))
705 return r;
706 };
Xiao Guangrongce88dec2011-07-12 03:33:44 +0800707
Avi Kivitye2dec932007-01-05 16:36:54 -0800708 r = mmu_topup_memory_caches(vcpu);
709 if (r)
710 return r;
Avi Kivity714b93d2007-01-05 16:36:53 -0800711
Avi Kivity6aa8b732006-12-10 02:21:36 -0800712 /*
Eddie Donga8b876b2009-03-26 15:28:40 +0800713 * Look up the guest pte for the faulting address.
Avi Kivity6aa8b732006-12-10 02:21:36 -0800714 */
Xiao Guangrong33770782010-09-28 17:03:14 +0800715 r = FNAME(walk_addr)(&walker, vcpu, addr, error_code);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800716
717 /*
718 * The page is not mapped by the guest. Let the guest handle it.
719 */
Avi Kivity7993ba42007-01-26 00:56:41 -0800720 if (!r) {
Harvey Harrisonb8688d52008-03-03 12:59:56 -0800721 pgprintk("%s: guest page fault\n", __func__);
Xiao Guangronga30f47c2011-09-22 16:58:36 +0800722 if (!prefault)
Xiao Guangrongfb67e142010-12-07 10:35:25 +0800723 inject_page_fault(vcpu, &walker.fault);
Xiao Guangronga30f47c2011-09-22 16:58:36 +0800724
Avi Kivity6aa8b732006-12-10 02:21:36 -0800725 return 0;
726 }
727
Xiao Guangrong93c05d32013-01-13 23:49:07 +0800728 vcpu->arch.write_fault_to_shadow_pgtable = false;
729
730 is_self_change_mapping = FNAME(is_self_change_mapping)(vcpu,
731 &walker, user_fault, &vcpu->arch.write_fault_to_shadow_pgtable);
732
Andrea Arcangeli936a5fe2011-01-13 15:46:48 -0800733 if (walker.level >= PT_DIRECTORY_LEVEL)
Xiao Guangrong7751bab2013-01-08 14:36:51 +0800734 force_pt_level = mapping_level_dirty_bitmap(vcpu, walker.gfn)
Xiao Guangrong93c05d32013-01-13 23:49:07 +0800735 || is_self_change_mapping;
Andrea Arcangeli936a5fe2011-01-13 15:46:48 -0800736 else
737 force_pt_level = 1;
738 if (!force_pt_level) {
Joerg Roedel7e4e4052009-07-27 16:30:46 +0200739 level = min(walker.level, mapping_level(vcpu, walker.gfn));
740 walker.gfn = walker.gfn & ~(KVM_PAGES_PER_HPAGE(level) - 1);
Marcelo Tosatti05da4552008-02-23 11:44:30 -0300741 }
Joerg Roedel7e4e4052009-07-27 16:30:46 +0200742
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200743 mmu_seq = vcpu->kvm->mmu_notifier_seq;
Marcelo Tosatti4c2155c2008-09-16 20:54:47 -0300744 smp_rmb();
Gleb Natapovaf585b92010-10-14 11:22:46 +0200745
Xiao Guangrong78b2c542010-12-07 10:48:06 +0800746 if (try_async_pf(vcpu, prefault, walker.gfn, addr, &pfn, write_fault,
Marcelo Tosatti612819c2010-10-22 14:18:18 -0200747 &map_writable))
Gleb Natapovaf585b92010-10-14 11:22:46 +0200748 return 0;
Avi Kivityd7824ff2007-12-30 12:29:05 +0200749
Xiao Guangrongd7c55202011-07-12 03:29:38 +0800750 if (handle_abnormal_pfn(vcpu, mmu_is_nested(vcpu) ? 0 : addr,
751 walker.gfn, pfn, walker.pte_access, &r))
752 return r;
753
Xiao Guangrongc22885052013-01-08 14:36:04 +0800754 /*
755 * Do not change pte_access if the pfn is a mmio page, otherwise
756 * we will cache the incorrect access into mmio spte.
757 */
758 if (write_fault && !(walker.pte_access & ACC_WRITE_MASK) &&
759 !is_write_protection(vcpu) && !user_fault &&
760 !is_noslot_pfn(pfn)) {
761 walker.pte_access |= ACC_WRITE_MASK;
762 walker.pte_access &= ~ACC_USER_MASK;
763
764 /*
765 * If we converted a user page to a kernel page,
766 * so that the kernel can write to it when cr0.wp=0,
767 * then we should prevent the kernel from executing it
768 * if SMEP is enabled.
769 */
770 if (kvm_read_cr4_bits(vcpu, X86_CR4_SMEP))
771 walker.pte_access &= ~ACC_EXEC_MASK;
772 }
773
Marcelo Tosattiaaee2c92007-12-20 19:18:26 -0500774 spin_lock(&vcpu->kvm->mmu_lock);
Christoffer Dall8ca40a72012-10-14 23:10:18 -0400775 if (mmu_notifier_retry(vcpu->kvm, mmu_seq))
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200776 goto out_unlock;
Xiao Guangrongbc32ce22010-08-28 19:22:46 +0800777
Xiao Guangrong0375f7f2011-11-28 20:41:00 +0800778 kvm_mmu_audit(vcpu, AUDIT_PRE_PAGE_FAULT);
Takuya Yoshikawa450e0b42013-03-29 14:05:26 +0900779 make_mmu_pages_available(vcpu);
Andrea Arcangeli936a5fe2011-01-13 15:46:48 -0800780 if (!force_pt_level)
781 transparent_hugepage_adjust(vcpu, &walker.gfn, &pfn, &level);
Xiao Guangrongc22885052013-01-08 14:36:04 +0800782 r = FNAME(fetch)(vcpu, addr, &walker, write_fault,
Xiao Guangrongd4878f22012-10-16 20:08:43 +0800783 level, pfn, map_writable, prefault);
Avi Kivity1165f5f2007-04-19 17:27:43 +0300784 ++vcpu->stat.pf_fixed;
Xiao Guangrong0375f7f2011-11-28 20:41:00 +0800785 kvm_mmu_audit(vcpu, AUDIT_POST_PAGE_FAULT);
Marcelo Tosattiaaee2c92007-12-20 19:18:26 -0500786 spin_unlock(&vcpu->kvm->mmu_lock);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800787
Xiao Guangrongd4878f22012-10-16 20:08:43 +0800788 return r;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200789
790out_unlock:
791 spin_unlock(&vcpu->kvm->mmu_lock);
792 kvm_release_pfn_clean(pfn);
793 return 0;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800794}
795
Xiao Guangrong505aef82011-09-22 16:56:06 +0800796static gpa_t FNAME(get_level1_sp_gpa)(struct kvm_mmu_page *sp)
797{
798 int offset = 0;
799
Davidlohr Buesof71fa312012-04-18 12:24:29 +0200800 WARN_ON(sp->role.level != PT_PAGE_TABLE_LEVEL);
Xiao Guangrong505aef82011-09-22 16:56:06 +0800801
802 if (PTTYPE == 32)
803 offset = sp->role.quadrant << PT64_LEVEL_BITS;
804
805 return gfn_to_gpa(sp->gfn) + offset * sizeof(pt_element_t);
806}
807
Marcelo Tosattia7052892008-09-23 13:18:35 -0300808static void FNAME(invlpg)(struct kvm_vcpu *vcpu, gva_t gva)
809{
Avi Kivitya4619302008-12-25 15:19:00 +0200810 struct kvm_shadow_walk_iterator iterator;
Xiao Guangrongf78978a2010-05-15 18:53:35 +0800811 struct kvm_mmu_page *sp;
Avi Kivitya4619302008-12-25 15:19:00 +0200812 int level;
813 u64 *sptep;
Marcelo Tosattia7052892008-09-23 13:18:35 -0300814
Xiao Guangrongbebb1062011-07-12 03:23:20 +0800815 vcpu_clear_mmio_info(vcpu, gva);
816
Xiao Guangrongf57f2ef2011-09-22 16:56:39 +0800817 /*
818 * No need to check return value here, rmap_can_add() can
819 * help us to skip pte prefetch later.
820 */
821 mmu_topup_memory_caches(vcpu);
Avi Kivitya4619302008-12-25 15:19:00 +0200822
Xiao Guangrongf57f2ef2011-09-22 16:56:39 +0800823 spin_lock(&vcpu->kvm->mmu_lock);
Avi Kivitya4619302008-12-25 15:19:00 +0200824 for_each_shadow_entry(vcpu, gva, iterator) {
825 level = iterator.level;
826 sptep = iterator.sptep;
827
Xiao Guangrongf78978a2010-05-15 18:53:35 +0800828 sp = page_header(__pa(sptep));
Xiao Guangrong884a0ff2010-04-28 11:55:15 +0800829 if (is_last_spte(*sptep, level)) {
Xiao Guangrongf57f2ef2011-09-22 16:56:39 +0800830 pt_element_t gpte;
831 gpa_t pte_gpa;
832
Xiao Guangrongf78978a2010-05-15 18:53:35 +0800833 if (!sp->unsync)
834 break;
835
Xiao Guangrong505aef82011-09-22 16:56:06 +0800836 pte_gpa = FNAME(get_level1_sp_gpa)(sp);
Avi Kivity08e850c2010-03-15 13:59:57 +0200837 pte_gpa += (sptep - sp->spt) * sizeof(pt_element_t);
Avi Kivitya4619302008-12-25 15:19:00 +0200838
Xiao Guangrong505aef82011-09-22 16:56:06 +0800839 if (mmu_page_zap_pte(vcpu->kvm, sp, sptep))
840 kvm_flush_remote_tlbs(vcpu->kvm);
Xiao Guangrongf57f2ef2011-09-22 16:56:39 +0800841
842 if (!rmap_can_add(vcpu))
843 break;
844
845 if (kvm_read_guest_atomic(vcpu->kvm, pte_gpa, &gpte,
846 sizeof(pt_element_t)))
847 break;
848
849 FNAME(update_pte)(vcpu, sp, sptep, &gpte);
Avi Kivitya4619302008-12-25 15:19:00 +0200850 }
851
Xiao Guangrongf78978a2010-05-15 18:53:35 +0800852 if (!is_shadow_present_pte(*sptep) || !sp->unsync_children)
Avi Kivitya4619302008-12-25 15:19:00 +0200853 break;
854 }
Marcelo Tosattiad218f82008-12-01 22:32:05 -0200855 spin_unlock(&vcpu->kvm->mmu_lock);
Marcelo Tosattia7052892008-09-23 13:18:35 -0300856}
857
Gleb Natapov1871c602010-02-10 14:21:32 +0200858static gpa_t FNAME(gva_to_gpa)(struct kvm_vcpu *vcpu, gva_t vaddr, u32 access,
Avi Kivityab9ae312010-11-22 17:53:26 +0200859 struct x86_exception *exception)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800860{
861 struct guest_walker walker;
Avi Kivitye119d112007-02-12 00:54:36 -0800862 gpa_t gpa = UNMAPPED_GVA;
863 int r;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800864
Xiao Guangrong33770782010-09-28 17:03:14 +0800865 r = FNAME(walk_addr)(&walker, vcpu, vaddr, access);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800866
Avi Kivitye119d112007-02-12 00:54:36 -0800867 if (r) {
Avi Kivity1755fbc2007-11-21 14:44:45 +0200868 gpa = gfn_to_gpa(walker.gfn);
Avi Kivitye119d112007-02-12 00:54:36 -0800869 gpa |= vaddr & ~PAGE_MASK;
Avi Kivity8c28d032010-11-22 17:53:27 +0200870 } else if (exception)
871 *exception = walker.fault;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800872
873 return gpa;
874}
875
Nadav Har'El37406aa2013-08-05 11:07:12 +0300876#if PTTYPE != PTTYPE_EPT
Joerg Roedel6539e732010-09-10 17:30:50 +0200877static gpa_t FNAME(gva_to_gpa_nested)(struct kvm_vcpu *vcpu, gva_t vaddr,
Avi Kivityab9ae312010-11-22 17:53:26 +0200878 u32 access,
879 struct x86_exception *exception)
Joerg Roedel6539e732010-09-10 17:30:50 +0200880{
881 struct guest_walker walker;
882 gpa_t gpa = UNMAPPED_GVA;
883 int r;
884
Xiao Guangrong33770782010-09-28 17:03:14 +0800885 r = FNAME(walk_addr_nested)(&walker, vcpu, vaddr, access);
Joerg Roedel6539e732010-09-10 17:30:50 +0200886
887 if (r) {
888 gpa = gfn_to_gpa(walker.gfn);
889 gpa |= vaddr & ~PAGE_MASK;
Avi Kivity8c28d032010-11-22 17:53:27 +0200890 } else if (exception)
891 *exception = walker.fault;
Joerg Roedel6539e732010-09-10 17:30:50 +0200892
893 return gpa;
894}
Nadav Har'El37406aa2013-08-05 11:07:12 +0300895#endif
Joerg Roedel6539e732010-09-10 17:30:50 +0200896
Marcelo Tosattie8bc2172008-09-23 13:18:33 -0300897/*
898 * Using the cached information from sp->gfns is safe because:
899 * - The spte has a reference to the struct page, so the pfn for a given gfn
900 * can't change unless all sptes pointing to it are nuked first.
Xiao Guangronga4ee1ca2010-11-23 11:13:00 +0800901 *
902 * Note:
903 * We should flush all tlbs if spte is dropped even though guest is
904 * responsible for it. Since if we don't, kvm_mmu_notifier_invalidate_page
905 * and kvm_mmu_notifier_invalidate_range_start detect the mapping page isn't
906 * used by guest then tlbs are not flushed, so guest is allowed to access the
907 * freed pages.
908 * And we increase kvm->tlbs_dirty to delay tlbs flush in this case.
Marcelo Tosattie8bc2172008-09-23 13:18:33 -0300909 */
Xiao Guangronga4a8e6f2010-11-19 17:04:03 +0800910static int FNAME(sync_page)(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp)
Marcelo Tosattie8bc2172008-09-23 13:18:33 -0300911{
Xiao Guangrong505aef82011-09-22 16:56:06 +0800912 int i, nr_present = 0;
Lai Jiangshan9bdbba12010-11-19 17:03:22 +0800913 bool host_writable;
Gui Jianfeng51fb60d2010-04-16 17:16:40 +0800914 gpa_t first_pte_gpa;
Marcelo Tosattie8bc2172008-09-23 13:18:33 -0300915
Lai Jiangshan2032a932010-05-26 16:49:59 +0800916 /* direct kvm_mmu_page can not be unsync. */
917 BUG_ON(sp->role.direct);
918
Xiao Guangrong505aef82011-09-22 16:56:06 +0800919 first_pte_gpa = FNAME(get_level1_sp_gpa)(sp);
Gui Jianfeng51fb60d2010-04-16 17:16:40 +0800920
Marcelo Tosattie8bc2172008-09-23 13:18:33 -0300921 for (i = 0; i < PT64_ENT_PER_PAGE; i++) {
922 unsigned pte_access;
923 pt_element_t gpte;
924 gpa_t pte_gpa;
Xiao Guangrongf55c3f42010-05-13 10:08:08 +0800925 gfn_t gfn;
Marcelo Tosattie8bc2172008-09-23 13:18:33 -0300926
Xiao Guangrongce88dec2011-07-12 03:33:44 +0800927 if (!sp->spt[i])
Marcelo Tosattie8bc2172008-09-23 13:18:33 -0300928 continue;
929
Gui Jianfeng51fb60d2010-04-16 17:16:40 +0800930 pte_gpa = first_pte_gpa + i * sizeof(pt_element_t);
Marcelo Tosattie8bc2172008-09-23 13:18:33 -0300931
932 if (kvm_read_guest_atomic(vcpu->kvm, pte_gpa, &gpte,
933 sizeof(pt_element_t)))
934 return -EINVAL;
935
Nadav Har'El0ad805a2013-08-05 11:07:09 +0300936 if (FNAME(prefetch_invalid_gpte)(vcpu, sp, &sp->spt[i], gpte)) {
Xiao Guangronga4ee1ca2010-11-23 11:13:00 +0800937 vcpu->kvm->tlbs_dirty++;
Xiao Guangrong407c61c2010-11-23 11:08:42 +0800938 continue;
939 }
940
Xiao Guangrongce88dec2011-07-12 03:33:44 +0800941 gfn = gpte_to_gfn(gpte);
942 pte_access = sp->role.access;
Nadav Har'El0ad805a2013-08-05 11:07:09 +0300943 pte_access &= FNAME(gpte_access)(vcpu, gpte);
944 FNAME(protect_clean_gpte)(&pte_access, gpte);
Xiao Guangrongce88dec2011-07-12 03:33:44 +0800945
Xiao Guangrongf2fd1252013-06-07 16:51:24 +0800946 if (sync_mmio_spte(vcpu->kvm, &sp->spt[i], gfn, pte_access,
947 &nr_present))
Xiao Guangrongce88dec2011-07-12 03:33:44 +0800948 continue;
949
Xiao Guangrong407c61c2010-11-23 11:08:42 +0800950 if (gfn != sp->gfns[i]) {
Xiao Guangrongc3707952011-07-12 03:28:04 +0800951 drop_spte(vcpu->kvm, &sp->spt[i]);
Xiao Guangronga4ee1ca2010-11-23 11:13:00 +0800952 vcpu->kvm->tlbs_dirty++;
Marcelo Tosattie8bc2172008-09-23 13:18:33 -0300953 continue;
954 }
955
956 nr_present++;
Xiao Guangrongce88dec2011-07-12 03:33:44 +0800957
Xiao Guangrongf8e453b2010-12-23 16:09:29 +0800958 host_writable = sp->spt[i] & SPTE_HOST_WRITEABLE;
959
Xiao Guangrongc22885052013-01-08 14:36:04 +0800960 set_spte(vcpu, &sp->spt[i], pte_access,
Xiao Guangrong640d9b02011-07-12 03:24:39 +0800961 PT_PAGE_TABLE_LEVEL, gfn,
Izik Eidus14032832009-09-23 21:47:17 +0300962 spte_to_pfn(sp->spt[i]), true, false,
Lai Jiangshan9bdbba12010-11-19 17:03:22 +0800963 host_writable);
Marcelo Tosattie8bc2172008-09-23 13:18:33 -0300964 }
965
966 return !nr_present;
967}
968
Avi Kivity6aa8b732006-12-10 02:21:36 -0800969#undef pt_element_t
970#undef guest_walker
971#undef FNAME
972#undef PT_BASE_ADDR_MASK
973#undef PT_INDEX
Joerg Roedele04da982009-07-27 16:30:45 +0200974#undef PT_LVL_ADDR_MASK
975#undef PT_LVL_OFFSET_MASK
Avi Kivityc7addb92007-09-16 18:58:32 +0200976#undef PT_LEVEL_BITS
Avi Kivitycea0f0e2007-01-05 16:36:43 -0800977#undef PT_MAX_FULL_LEVELS
Avi Kivity5fb07dd2007-11-21 12:35:07 +0200978#undef gpte_to_gfn
Joerg Roedele04da982009-07-27 16:30:45 +0200979#undef gpte_to_gfn_lvl
Marcelo Tosattib3e4e632007-12-07 07:56:58 -0500980#undef CMPXCHG
Gleb Natapovd8089ba2013-08-05 11:07:10 +0300981#undef PT_GUEST_ACCESSED_MASK
982#undef PT_GUEST_DIRTY_MASK
983#undef PT_GUEST_DIRTY_SHIFT
984#undef PT_GUEST_ACCESSED_SHIFT