]>
Commit | Line | Data |
---|---|---|
6aa8b732 AK |
1 | /* |
2 | * Kernel-based Virtual Machine driver for Linux | |
3 | * | |
4 | * This module enables machines with Intel VT-x extensions to run virtual | |
5 | * machines without emulation or binary translation. | |
6 | * | |
7 | * MMU support | |
8 | * | |
9 | * Copyright (C) 2006 Qumranet, Inc. | |
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) | |
34 | #define PT_PTE_COPY_MASK PT64_PTE_COPY_MASK | |
6aa8b732 AK |
35 | #elif PTTYPE == 32 |
36 | #define pt_element_t u32 | |
37 | #define guest_walker guest_walker32 | |
38 | #define FNAME(name) paging##32_##name | |
39 | #define PT_BASE_ADDR_MASK PT32_BASE_ADDR_MASK | |
40 | #define PT_DIR_BASE_ADDR_MASK PT32_DIR_BASE_ADDR_MASK | |
41 | #define PT_INDEX(addr, level) PT32_INDEX(addr, level) | |
42 | #define SHADOW_PT_INDEX(addr, level) PT64_INDEX(addr, level) | |
43 | #define PT_LEVEL_MASK(level) PT32_LEVEL_MASK(level) | |
44 | #define PT_PTE_COPY_MASK PT32_PTE_COPY_MASK | |
6aa8b732 AK |
45 | #else |
46 | #error Invalid PTTYPE value | |
47 | #endif | |
48 | ||
49 | /* | |
50 | * The guest_walker structure emulates the behavior of the hardware page | |
51 | * table walker. | |
52 | */ | |
53 | struct guest_walker { | |
54 | int level; | |
6bcbd6ab | 55 | gfn_t table_gfn; |
6aa8b732 AK |
56 | pt_element_t *table; |
57 | pt_element_t inherited_ar; | |
58 | }; | |
59 | ||
60 | static void FNAME(init_walker)(struct guest_walker *walker, | |
61 | struct kvm_vcpu *vcpu) | |
62 | { | |
63 | hpa_t hpa; | |
64 | struct kvm_memory_slot *slot; | |
65 | ||
66 | walker->level = vcpu->mmu.root_level; | |
6bcbd6ab AK |
67 | walker->table_gfn = (vcpu->cr3 & PT64_BASE_ADDR_MASK) >> PAGE_SHIFT; |
68 | slot = gfn_to_memslot(vcpu->kvm, walker->table_gfn); | |
6aa8b732 AK |
69 | hpa = safe_gpa_to_hpa(vcpu, vcpu->cr3 & PT64_BASE_ADDR_MASK); |
70 | walker->table = kmap_atomic(pfn_to_page(hpa >> PAGE_SHIFT), KM_USER0); | |
71 | ||
a9058ecd | 72 | ASSERT((!is_long_mode(vcpu) && is_pae(vcpu)) || |
6aa8b732 AK |
73 | (vcpu->cr3 & ~(PAGE_MASK | CR3_FLAGS_MASK)) == 0); |
74 | ||
75 | walker->table = (pt_element_t *)( (unsigned long)walker->table | | |
76 | (unsigned long)(vcpu->cr3 & ~(PAGE_MASK | CR3_FLAGS_MASK)) ); | |
77 | walker->inherited_ar = PT_USER_MASK | PT_WRITABLE_MASK; | |
78 | } | |
79 | ||
80 | static void FNAME(release_walker)(struct guest_walker *walker) | |
81 | { | |
82 | kunmap_atomic(walker->table, KM_USER0); | |
83 | } | |
84 | ||
85 | static void FNAME(set_pte)(struct kvm_vcpu *vcpu, u64 guest_pte, | |
86 | u64 *shadow_pte, u64 access_bits) | |
87 | { | |
88 | ASSERT(*shadow_pte == 0); | |
89 | access_bits &= guest_pte; | |
90 | *shadow_pte = (guest_pte & PT_PTE_COPY_MASK); | |
91 | set_pte_common(vcpu, shadow_pte, guest_pte & PT_BASE_ADDR_MASK, | |
92 | guest_pte & PT_DIRTY_MASK, access_bits); | |
93 | } | |
94 | ||
95 | static void FNAME(set_pde)(struct kvm_vcpu *vcpu, u64 guest_pde, | |
96 | u64 *shadow_pte, u64 access_bits, | |
97 | int index) | |
98 | { | |
99 | gpa_t gaddr; | |
100 | ||
101 | ASSERT(*shadow_pte == 0); | |
102 | access_bits &= guest_pde; | |
103 | gaddr = (guest_pde & PT_DIR_BASE_ADDR_MASK) + PAGE_SIZE * index; | |
104 | if (PTTYPE == 32 && is_cpuid_PSE36()) | |
105 | gaddr |= (guest_pde & PT32_DIR_PSE36_MASK) << | |
106 | (32 - PT32_DIR_PSE36_SHIFT); | |
8c7bb723 | 107 | *shadow_pte = guest_pde & PT_PTE_COPY_MASK; |
6aa8b732 AK |
108 | set_pte_common(vcpu, shadow_pte, gaddr, |
109 | guest_pde & PT_DIRTY_MASK, access_bits); | |
110 | } | |
111 | ||
112 | /* | |
113 | * Fetch a guest pte from a specific level in the paging hierarchy. | |
114 | */ | |
115 | static pt_element_t *FNAME(fetch_guest)(struct kvm_vcpu *vcpu, | |
116 | struct guest_walker *walker, | |
117 | int level, | |
118 | gva_t addr) | |
119 | { | |
120 | ||
121 | ASSERT(level > 0 && level <= walker->level); | |
122 | ||
123 | for (;;) { | |
124 | int index = PT_INDEX(addr, walker->level); | |
125 | hpa_t paddr; | |
126 | ||
127 | ASSERT(((unsigned long)walker->table & PAGE_MASK) == | |
128 | ((unsigned long)&walker->table[index] & PAGE_MASK)); | |
129 | if (level == walker->level || | |
130 | !is_present_pte(walker->table[index]) || | |
131 | (walker->level == PT_DIRECTORY_LEVEL && | |
132 | (walker->table[index] & PT_PAGE_SIZE_MASK) && | |
133 | (PTTYPE == 64 || is_pse(vcpu)))) | |
134 | return &walker->table[index]; | |
a9058ecd | 135 | if (walker->level != 3 || is_long_mode(vcpu)) |
6aa8b732 | 136 | walker->inherited_ar &= walker->table[index]; |
6bcbd6ab AK |
137 | walker->table_gfn = (walker->table[index] & PT_BASE_ADDR_MASK) |
138 | >> PAGE_SHIFT; | |
6aa8b732 AK |
139 | paddr = safe_gpa_to_hpa(vcpu, walker->table[index] & PT_BASE_ADDR_MASK); |
140 | kunmap_atomic(walker->table, KM_USER0); | |
141 | walker->table = kmap_atomic(pfn_to_page(paddr >> PAGE_SHIFT), | |
142 | KM_USER0); | |
143 | --walker->level; | |
144 | } | |
145 | } | |
146 | ||
147 | /* | |
148 | * Fetch a shadow pte for a specific level in the paging hierarchy. | |
149 | */ | |
150 | static u64 *FNAME(fetch)(struct kvm_vcpu *vcpu, gva_t addr, | |
151 | struct guest_walker *walker) | |
152 | { | |
153 | hpa_t shadow_addr; | |
154 | int level; | |
155 | u64 *prev_shadow_ent = NULL; | |
156 | ||
157 | shadow_addr = vcpu->mmu.root_hpa; | |
158 | level = vcpu->mmu.shadow_root_level; | |
159 | ||
160 | for (; ; level--) { | |
161 | u32 index = SHADOW_PT_INDEX(addr, level); | |
162 | u64 *shadow_ent = ((u64 *)__va(shadow_addr)) + index; | |
163 | pt_element_t *guest_ent; | |
8c7bb723 | 164 | u64 shadow_pte; |
6aa8b732 AK |
165 | |
166 | if (is_present_pte(*shadow_ent) || is_io_pte(*shadow_ent)) { | |
167 | if (level == PT_PAGE_TABLE_LEVEL) | |
168 | return shadow_ent; | |
169 | shadow_addr = *shadow_ent & PT64_BASE_ADDR_MASK; | |
170 | prev_shadow_ent = shadow_ent; | |
171 | continue; | |
172 | } | |
173 | ||
174 | if (PTTYPE == 32 && level > PT32_ROOT_LEVEL) { | |
175 | ASSERT(level == PT32E_ROOT_LEVEL); | |
176 | guest_ent = FNAME(fetch_guest)(vcpu, walker, | |
177 | PT32_ROOT_LEVEL, addr); | |
178 | } else | |
179 | guest_ent = FNAME(fetch_guest)(vcpu, walker, | |
180 | level, addr); | |
181 | ||
182 | if (!is_present_pte(*guest_ent)) | |
183 | return NULL; | |
184 | ||
185 | /* Don't set accessed bit on PAE PDPTRs */ | |
186 | if (vcpu->mmu.root_level != 3 || walker->level != 3) | |
187 | *guest_ent |= PT_ACCESSED_MASK; | |
188 | ||
189 | if (level == PT_PAGE_TABLE_LEVEL) { | |
190 | ||
191 | if (walker->level == PT_DIRECTORY_LEVEL) { | |
192 | if (prev_shadow_ent) | |
193 | *prev_shadow_ent |= PT_SHADOW_PS_MARK; | |
194 | FNAME(set_pde)(vcpu, *guest_ent, shadow_ent, | |
195 | walker->inherited_ar, | |
196 | PT_INDEX(addr, PT_PAGE_TABLE_LEVEL)); | |
197 | } else { | |
198 | ASSERT(walker->level == PT_PAGE_TABLE_LEVEL); | |
199 | FNAME(set_pte)(vcpu, *guest_ent, shadow_ent, walker->inherited_ar); | |
200 | } | |
201 | return shadow_ent; | |
202 | } | |
203 | ||
204 | shadow_addr = kvm_mmu_alloc_page(vcpu, shadow_ent); | |
205 | if (!VALID_PAGE(shadow_addr)) | |
206 | return ERR_PTR(-ENOMEM); | |
8c7bb723 AK |
207 | shadow_pte = shadow_addr | PT_PRESENT_MASK; |
208 | if (vcpu->mmu.root_level > 3 || level != 3) | |
209 | shadow_pte |= PT_ACCESSED_MASK | |
210 | | PT_WRITABLE_MASK | PT_USER_MASK; | |
211 | *shadow_ent = shadow_pte; | |
6aa8b732 AK |
212 | prev_shadow_ent = shadow_ent; |
213 | } | |
214 | } | |
215 | ||
216 | /* | |
217 | * The guest faulted for write. We need to | |
218 | * | |
219 | * - check write permissions | |
220 | * - update the guest pte dirty bit | |
221 | * - update our own dirty page tracking structures | |
222 | */ | |
223 | static int FNAME(fix_write_pf)(struct kvm_vcpu *vcpu, | |
224 | u64 *shadow_ent, | |
225 | struct guest_walker *walker, | |
226 | gva_t addr, | |
227 | int user) | |
228 | { | |
229 | pt_element_t *guest_ent; | |
230 | int writable_shadow; | |
231 | gfn_t gfn; | |
232 | ||
233 | if (is_writeble_pte(*shadow_ent)) | |
234 | return 0; | |
235 | ||
236 | writable_shadow = *shadow_ent & PT_SHADOW_WRITABLE_MASK; | |
237 | if (user) { | |
238 | /* | |
239 | * User mode access. Fail if it's a kernel page or a read-only | |
240 | * page. | |
241 | */ | |
242 | if (!(*shadow_ent & PT_SHADOW_USER_MASK) || !writable_shadow) | |
243 | return 0; | |
244 | ASSERT(*shadow_ent & PT_USER_MASK); | |
245 | } else | |
246 | /* | |
247 | * Kernel mode access. Fail if it's a read-only page and | |
248 | * supervisor write protection is enabled. | |
249 | */ | |
250 | if (!writable_shadow) { | |
251 | if (is_write_protection(vcpu)) | |
252 | return 0; | |
253 | *shadow_ent &= ~PT_USER_MASK; | |
254 | } | |
255 | ||
256 | guest_ent = FNAME(fetch_guest)(vcpu, walker, PT_PAGE_TABLE_LEVEL, addr); | |
257 | ||
258 | if (!is_present_pte(*guest_ent)) { | |
259 | *shadow_ent = 0; | |
260 | return 0; | |
261 | } | |
262 | ||
263 | gfn = (*guest_ent & PT64_BASE_ADDR_MASK) >> PAGE_SHIFT; | |
264 | mark_page_dirty(vcpu->kvm, gfn); | |
265 | *shadow_ent |= PT_WRITABLE_MASK; | |
266 | *guest_ent |= PT_DIRTY_MASK; | |
cd4a4e53 | 267 | rmap_add(vcpu->kvm, shadow_ent); |
6aa8b732 AK |
268 | |
269 | return 1; | |
270 | } | |
271 | ||
272 | /* | |
273 | * Page fault handler. There are several causes for a page fault: | |
274 | * - there is no shadow pte for the guest pte | |
275 | * - write access through a shadow pte marked read only so that we can set | |
276 | * the dirty bit | |
277 | * - write access to a shadow pte marked read only so we can update the page | |
278 | * dirty bitmap, when userspace requests it | |
279 | * - mmio access; in this case we will never install a present shadow pte | |
280 | * - normal guest page fault due to the guest pte marked not present, not | |
281 | * writable, or not executable | |
282 | * | |
283 | * Returns: 1 if we need to emulate the instruction, 0 otherwise | |
284 | */ | |
285 | static int FNAME(page_fault)(struct kvm_vcpu *vcpu, gva_t addr, | |
286 | u32 error_code) | |
287 | { | |
288 | int write_fault = error_code & PFERR_WRITE_MASK; | |
289 | int pte_present = error_code & PFERR_PRESENT_MASK; | |
290 | int user_fault = error_code & PFERR_USER_MASK; | |
291 | struct guest_walker walker; | |
292 | u64 *shadow_pte; | |
293 | int fixed; | |
294 | ||
295 | /* | |
296 | * Look up the shadow pte for the faulting address. | |
297 | */ | |
298 | for (;;) { | |
299 | FNAME(init_walker)(&walker, vcpu); | |
300 | shadow_pte = FNAME(fetch)(vcpu, addr, &walker); | |
301 | if (IS_ERR(shadow_pte)) { /* must be -ENOMEM */ | |
302 | nonpaging_flush(vcpu); | |
303 | FNAME(release_walker)(&walker); | |
304 | continue; | |
305 | } | |
306 | break; | |
307 | } | |
308 | ||
309 | /* | |
310 | * The page is not mapped by the guest. Let the guest handle it. | |
311 | */ | |
312 | if (!shadow_pte) { | |
313 | inject_page_fault(vcpu, addr, error_code); | |
314 | FNAME(release_walker)(&walker); | |
315 | return 0; | |
316 | } | |
317 | ||
318 | /* | |
319 | * Update the shadow pte. | |
320 | */ | |
321 | if (write_fault) | |
322 | fixed = FNAME(fix_write_pf)(vcpu, shadow_pte, &walker, addr, | |
323 | user_fault); | |
324 | else | |
325 | fixed = fix_read_pf(shadow_pte); | |
326 | ||
327 | FNAME(release_walker)(&walker); | |
328 | ||
329 | /* | |
330 | * mmio: emulate if accessible, otherwise its a guest fault. | |
331 | */ | |
332 | if (is_io_pte(*shadow_pte)) { | |
333 | if (may_access(*shadow_pte, write_fault, user_fault)) | |
334 | return 1; | |
335 | pgprintk("%s: io work, no access\n", __FUNCTION__); | |
336 | inject_page_fault(vcpu, addr, | |
337 | error_code | PFERR_PRESENT_MASK); | |
338 | return 0; | |
339 | } | |
340 | ||
341 | /* | |
342 | * pte not present, guest page fault. | |
343 | */ | |
344 | if (pte_present && !fixed) { | |
345 | inject_page_fault(vcpu, addr, error_code); | |
346 | return 0; | |
347 | } | |
348 | ||
349 | ++kvm_stat.pf_fixed; | |
350 | ||
351 | return 0; | |
352 | } | |
353 | ||
354 | static gpa_t FNAME(gva_to_gpa)(struct kvm_vcpu *vcpu, gva_t vaddr) | |
355 | { | |
356 | struct guest_walker walker; | |
357 | pt_element_t guest_pte; | |
358 | gpa_t gpa; | |
359 | ||
360 | FNAME(init_walker)(&walker, vcpu); | |
361 | guest_pte = *FNAME(fetch_guest)(vcpu, &walker, PT_PAGE_TABLE_LEVEL, | |
362 | vaddr); | |
363 | FNAME(release_walker)(&walker); | |
364 | ||
365 | if (!is_present_pte(guest_pte)) | |
366 | return UNMAPPED_GVA; | |
367 | ||
368 | if (walker.level == PT_DIRECTORY_LEVEL) { | |
369 | ASSERT((guest_pte & PT_PAGE_SIZE_MASK)); | |
370 | ASSERT(PTTYPE == 64 || is_pse(vcpu)); | |
371 | ||
372 | gpa = (guest_pte & PT_DIR_BASE_ADDR_MASK) | (vaddr & | |
373 | (PT_LEVEL_MASK(PT_PAGE_TABLE_LEVEL) | ~PAGE_MASK)); | |
374 | ||
375 | if (PTTYPE == 32 && is_cpuid_PSE36()) | |
376 | gpa |= (guest_pte & PT32_DIR_PSE36_MASK) << | |
377 | (32 - PT32_DIR_PSE36_SHIFT); | |
378 | } else { | |
379 | gpa = (guest_pte & PT_BASE_ADDR_MASK); | |
380 | gpa |= (vaddr & ~PAGE_MASK); | |
381 | } | |
382 | ||
383 | return gpa; | |
384 | } | |
385 | ||
386 | #undef pt_element_t | |
387 | #undef guest_walker | |
388 | #undef FNAME | |
389 | #undef PT_BASE_ADDR_MASK | |
390 | #undef PT_INDEX | |
391 | #undef SHADOW_PT_INDEX | |
392 | #undef PT_LEVEL_MASK | |
393 | #undef PT_PTE_COPY_MASK | |
394 | #undef PT_NON_PTE_COPY_MASK | |
395 | #undef PT_DIR_BASE_ADDR_MASK |