2 * Kernel-based Virtual Machine driver for Linux
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
9 * Copyright (C) 2006 Qumranet, Inc.
12 * Yaniv Kamay <yaniv@qumranet.com>
13 * Avi Kivity <avi@qumranet.com>
15 * This work is licensed under the terms of the GNU GPL, version 2. See
16 * the COPYING file in the top-level directory.
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.
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)
35 #define PT_MAX_FULL_LEVELS 4
37 #define PT_MAX_FULL_LEVELS 2
40 #define pt_element_t u32
41 #define guest_walker guest_walker32
42 #define FNAME(name) paging##32_##name
43 #define PT_BASE_ADDR_MASK PT32_BASE_ADDR_MASK
44 #define PT_DIR_BASE_ADDR_MASK PT32_DIR_BASE_ADDR_MASK
45 #define PT_INDEX(addr, level) PT32_INDEX(addr, level)
46 #define SHADOW_PT_INDEX(addr, level) PT64_INDEX(addr, level)
47 #define PT_LEVEL_MASK(level) PT32_LEVEL_MASK(level)
48 #define PT_MAX_FULL_LEVELS 2
50 #error Invalid PTTYPE value
54 * The guest_walker structure emulates the behavior of the hardware page
59 gfn_t table_gfn[PT_MAX_FULL_LEVELS];
62 pt_element_t inherited_ar;
68 * Fetch a guest pte for a guest virtual address
70 static int FNAME(walk_addr)(struct guest_walker *walker,
71 struct kvm_vcpu *vcpu, gva_t addr,
72 int write_fault, int user_fault, int fetch_fault)
75 struct kvm_memory_slot *slot;
80 pgprintk("%s: addr %lx\n", __FUNCTION__, addr);
81 walker->level = vcpu->mmu.root_level;
85 if (!is_long_mode(vcpu)) {
86 walker->ptep = &vcpu->pdptrs[(addr >> 30) & 3];
88 if (!(root & PT_PRESENT_MASK))
93 table_gfn = (root & PT64_BASE_ADDR_MASK) >> PAGE_SHIFT;
94 walker->table_gfn[walker->level - 1] = table_gfn;
95 pgprintk("%s: table_gfn[%d] %lx\n", __FUNCTION__,
96 walker->level - 1, table_gfn);
97 slot = gfn_to_memslot(vcpu->kvm, table_gfn);
98 hpa = safe_gpa_to_hpa(vcpu, root & PT64_BASE_ADDR_MASK);
99 walker->table = kmap_atomic(pfn_to_page(hpa >> PAGE_SHIFT), KM_USER0);
101 ASSERT((!is_long_mode(vcpu) && is_pae(vcpu)) ||
102 (vcpu->cr3 & ~(PAGE_MASK | CR3_FLAGS_MASK)) == 0);
104 walker->inherited_ar = PT_USER_MASK | PT_WRITABLE_MASK;
107 int index = PT_INDEX(addr, walker->level);
110 ptep = &walker->table[index];
111 ASSERT(((unsigned long)walker->table & PAGE_MASK) ==
112 ((unsigned long)ptep & PAGE_MASK));
114 if (!is_present_pte(*ptep))
117 if (write_fault && !is_writeble_pte(*ptep))
118 if (user_fault || is_write_protection(vcpu))
121 if (user_fault && !(*ptep & PT_USER_MASK))
125 if (fetch_fault && is_nx(vcpu) && (*ptep & PT64_NX_MASK))
129 if (!(*ptep & PT_ACCESSED_MASK)) {
130 mark_page_dirty(vcpu->kvm, table_gfn);
131 *ptep |= PT_ACCESSED_MASK;
134 if (walker->level == PT_PAGE_TABLE_LEVEL) {
135 walker->gfn = (*ptep & PT_BASE_ADDR_MASK)
140 if (walker->level == PT_DIRECTORY_LEVEL
141 && (*ptep & PT_PAGE_SIZE_MASK)
142 && (PTTYPE == 64 || is_pse(vcpu))) {
143 walker->gfn = (*ptep & PT_DIR_BASE_ADDR_MASK)
145 walker->gfn += PT_INDEX(addr, PT_PAGE_TABLE_LEVEL);
149 walker->inherited_ar &= walker->table[index];
150 table_gfn = (*ptep & PT_BASE_ADDR_MASK) >> PAGE_SHIFT;
151 paddr = safe_gpa_to_hpa(vcpu, *ptep & PT_BASE_ADDR_MASK);
152 kunmap_atomic(walker->table, KM_USER0);
153 walker->table = kmap_atomic(pfn_to_page(paddr >> PAGE_SHIFT),
156 walker->table_gfn[walker->level - 1 ] = table_gfn;
157 pgprintk("%s: table_gfn[%d] %lx\n", __FUNCTION__,
158 walker->level - 1, table_gfn);
161 pgprintk("%s: pte %llx\n", __FUNCTION__, (u64)*ptep);
165 walker->error_code = 0;
169 walker->error_code = PFERR_PRESENT_MASK;
173 walker->error_code |= PFERR_WRITE_MASK;
175 walker->error_code |= PFERR_USER_MASK;
177 walker->error_code |= PFERR_FETCH_MASK;
181 static void FNAME(release_walker)(struct guest_walker *walker)
184 kunmap_atomic(walker->table, KM_USER0);
187 static void FNAME(mark_pagetable_dirty)(struct kvm *kvm,
188 struct guest_walker *walker)
190 mark_page_dirty(kvm, walker->table_gfn[walker->level - 1]);
193 static void FNAME(set_pte_common)(struct kvm_vcpu *vcpu,
201 struct guest_walker *walker,
205 int dirty = *gpte & PT_DIRTY_MASK;
206 u64 spte = *shadow_pte;
207 int was_rmapped = is_rmap_pte(spte);
209 pgprintk("%s: spte %llx gpte %llx access %llx write_fault %d"
210 " user_fault %d gfn %lx\n",
211 __FUNCTION__, spte, (u64)*gpte, access_bits,
212 write_fault, user_fault, gfn);
214 if (write_fault && !dirty) {
215 *gpte |= PT_DIRTY_MASK;
217 FNAME(mark_pagetable_dirty)(vcpu->kvm, walker);
220 spte |= PT_PRESENT_MASK | PT_ACCESSED_MASK | PT_DIRTY_MASK;
221 spte |= *gpte & PT64_NX_MASK;
222 spte |= access_bits << PT_SHADOW_BITS_OFFSET;
224 access_bits &= ~PT_WRITABLE_MASK;
226 paddr = gpa_to_hpa(vcpu, gaddr & PT64_BASE_ADDR_MASK);
228 spte |= PT_PRESENT_MASK;
229 if (access_bits & PT_USER_MASK)
230 spte |= PT_USER_MASK;
232 if (is_error_hpa(paddr)) {
234 spte |= PT_SHADOW_IO_MARK;
235 spte &= ~PT_PRESENT_MASK;
236 set_shadow_pte(shadow_pte, spte);
242 if ((access_bits & PT_WRITABLE_MASK)
243 || (write_fault && !is_write_protection(vcpu) && !user_fault)) {
244 struct kvm_mmu_page *shadow;
246 spte |= PT_WRITABLE_MASK;
248 mmu_unshadow(vcpu, gfn);
252 shadow = kvm_mmu_lookup_page(vcpu, gfn);
254 pgprintk("%s: found shadow page for %lx, marking ro\n",
256 access_bits &= ~PT_WRITABLE_MASK;
257 if (is_writeble_pte(spte)) {
258 spte &= ~PT_WRITABLE_MASK;
259 kvm_arch_ops->tlb_flush(vcpu);
268 if (access_bits & PT_WRITABLE_MASK)
269 mark_page_dirty(vcpu->kvm, gaddr >> PAGE_SHIFT);
271 set_shadow_pte(shadow_pte, spte);
272 page_header_update_slot(vcpu->kvm, shadow_pte, gaddr);
274 rmap_add(vcpu, shadow_pte);
277 static void FNAME(set_pte)(struct kvm_vcpu *vcpu, pt_element_t *gpte,
278 u64 *shadow_pte, u64 access_bits,
279 int user_fault, int write_fault, int *ptwrite,
280 struct guest_walker *walker, gfn_t gfn)
282 access_bits &= *gpte;
283 FNAME(set_pte_common)(vcpu, shadow_pte, *gpte & PT_BASE_ADDR_MASK,
284 gpte, access_bits, user_fault, write_fault,
285 ptwrite, walker, gfn);
288 static void FNAME(update_pte)(struct kvm_vcpu *vcpu, struct kvm_mmu_page *page,
289 u64 *spte, const void *pte, int bytes)
293 if (bytes < sizeof(pt_element_t))
295 gpte = *(const pt_element_t *)pte;
296 if (~gpte & (PT_PRESENT_MASK | PT_ACCESSED_MASK))
298 pgprintk("%s: gpte %llx spte %p\n", __FUNCTION__, (u64)gpte, spte);
299 FNAME(set_pte)(vcpu, &gpte, spte, PT_USER_MASK | PT_WRITABLE_MASK, 0,
301 (gpte & PT_BASE_ADDR_MASK) >> PAGE_SHIFT);
304 static void FNAME(set_pde)(struct kvm_vcpu *vcpu, pt_element_t *gpde,
305 u64 *shadow_pte, u64 access_bits,
306 int user_fault, int write_fault, int *ptwrite,
307 struct guest_walker *walker, gfn_t gfn)
311 access_bits &= *gpde;
312 gaddr = (gpa_t)gfn << PAGE_SHIFT;
313 if (PTTYPE == 32 && is_cpuid_PSE36())
314 gaddr |= (*gpde & PT32_DIR_PSE36_MASK) <<
315 (32 - PT32_DIR_PSE36_SHIFT);
316 FNAME(set_pte_common)(vcpu, shadow_pte, gaddr,
317 gpde, access_bits, user_fault, write_fault,
318 ptwrite, walker, gfn);
322 * Fetch a shadow pte for a specific level in the paging hierarchy.
324 static u64 *FNAME(fetch)(struct kvm_vcpu *vcpu, gva_t addr,
325 struct guest_walker *walker,
326 int user_fault, int write_fault, int *ptwrite)
331 u64 *prev_shadow_ent = NULL;
332 pt_element_t *guest_ent = walker->ptep;
334 if (!is_present_pte(*guest_ent))
337 shadow_addr = vcpu->mmu.root_hpa;
338 level = vcpu->mmu.shadow_root_level;
339 if (level == PT32E_ROOT_LEVEL) {
340 shadow_addr = vcpu->mmu.pae_root[(addr >> 30) & 3];
341 shadow_addr &= PT64_BASE_ADDR_MASK;
346 u32 index = SHADOW_PT_INDEX(addr, level);
347 struct kvm_mmu_page *shadow_page;
351 unsigned hugepage_access = 0;
353 shadow_ent = ((u64 *)__va(shadow_addr)) + index;
354 if (is_present_pte(*shadow_ent) || is_io_pte(*shadow_ent)) {
355 if (level == PT_PAGE_TABLE_LEVEL)
357 shadow_addr = *shadow_ent & PT64_BASE_ADDR_MASK;
358 prev_shadow_ent = shadow_ent;
362 if (level == PT_PAGE_TABLE_LEVEL)
365 if (level - 1 == PT_PAGE_TABLE_LEVEL
366 && walker->level == PT_DIRECTORY_LEVEL) {
368 hugepage_access = *guest_ent;
369 hugepage_access &= PT_USER_MASK | PT_WRITABLE_MASK;
370 hugepage_access >>= PT_WRITABLE_SHIFT;
371 table_gfn = (*guest_ent & PT_BASE_ADDR_MASK)
375 table_gfn = walker->table_gfn[level - 2];
377 shadow_page = kvm_mmu_get_page(vcpu, table_gfn, addr, level-1,
378 metaphysical, hugepage_access,
380 shadow_addr = __pa(shadow_page->spt);
381 shadow_pte = shadow_addr | PT_PRESENT_MASK | PT_ACCESSED_MASK
382 | PT_WRITABLE_MASK | PT_USER_MASK;
383 *shadow_ent = shadow_pte;
384 prev_shadow_ent = shadow_ent;
387 if (walker->level == PT_DIRECTORY_LEVEL) {
389 *prev_shadow_ent |= PT_SHADOW_PS_MARK;
390 FNAME(set_pde)(vcpu, guest_ent, shadow_ent,
391 walker->inherited_ar, user_fault, write_fault,
392 ptwrite, walker, walker->gfn);
394 ASSERT(walker->level == PT_PAGE_TABLE_LEVEL);
395 FNAME(set_pte)(vcpu, guest_ent, shadow_ent,
396 walker->inherited_ar, user_fault, write_fault,
397 ptwrite, walker, walker->gfn);
403 * Page fault handler. There are several causes for a page fault:
404 * - there is no shadow pte for the guest pte
405 * - write access through a shadow pte marked read only so that we can set
407 * - write access to a shadow pte marked read only so we can update the page
408 * dirty bitmap, when userspace requests it
409 * - mmio access; in this case we will never install a present shadow pte
410 * - normal guest page fault due to the guest pte marked not present, not
411 * writable, or not executable
413 * Returns: 1 if we need to emulate the instruction, 0 otherwise, or
414 * a negative value on error.
416 static int FNAME(page_fault)(struct kvm_vcpu *vcpu, gva_t addr,
419 int write_fault = error_code & PFERR_WRITE_MASK;
420 int user_fault = error_code & PFERR_USER_MASK;
421 int fetch_fault = error_code & PFERR_FETCH_MASK;
422 struct guest_walker walker;
427 pgprintk("%s: addr %lx err %x\n", __FUNCTION__, addr, error_code);
428 kvm_mmu_audit(vcpu, "pre page fault");
430 r = mmu_topup_memory_caches(vcpu);
435 * Look up the shadow pte for the faulting address.
437 r = FNAME(walk_addr)(&walker, vcpu, addr, write_fault, user_fault,
441 * The page is not mapped by the guest. Let the guest handle it.
444 pgprintk("%s: guest page fault\n", __FUNCTION__);
445 inject_page_fault(vcpu, addr, walker.error_code);
446 FNAME(release_walker)(&walker);
447 vcpu->last_pt_write_count = 0; /* reset fork detector */
451 shadow_pte = FNAME(fetch)(vcpu, addr, &walker, user_fault, write_fault,
453 pgprintk("%s: shadow pte %p %llx ptwrite %d\n", __FUNCTION__,
454 shadow_pte, *shadow_pte, write_pt);
456 FNAME(release_walker)(&walker);
459 vcpu->last_pt_write_count = 0; /* reset fork detector */
462 * mmio: emulate if accessible, otherwise its a guest fault.
464 if (is_io_pte(*shadow_pte))
467 ++vcpu->stat.pf_fixed;
468 kvm_mmu_audit(vcpu, "post page fault (fixed)");
473 static gpa_t FNAME(gva_to_gpa)(struct kvm_vcpu *vcpu, gva_t vaddr)
475 struct guest_walker walker;
476 gpa_t gpa = UNMAPPED_GVA;
479 r = FNAME(walk_addr)(&walker, vcpu, vaddr, 0, 0, 0);
482 gpa = (gpa_t)walker.gfn << PAGE_SHIFT;
483 gpa |= vaddr & ~PAGE_MASK;
486 FNAME(release_walker)(&walker);
493 #undef PT_BASE_ADDR_MASK
495 #undef SHADOW_PT_INDEX
497 #undef PT_DIR_BASE_ADDR_MASK
498 #undef PT_MAX_FULL_LEVELS