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)
34 #define PT_LEVEL_BITS PT64_LEVEL_BITS
36 #define PT_MAX_FULL_LEVELS 4
37 #define CMPXCHG cmpxchg
39 #define CMPXCHG cmpxchg64
40 #define PT_MAX_FULL_LEVELS 2
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)
51 #define PT_LEVEL_BITS PT32_LEVEL_BITS
52 #define PT_MAX_FULL_LEVELS 2
53 #define CMPXCHG cmpxchg
55 #error Invalid PTTYPE value
58 #define gpte_to_gfn FNAME(gpte_to_gfn)
59 #define gpte_to_gfn_pde FNAME(gpte_to_gfn_pde)
62 * The guest_walker structure emulates the behavior of the hardware page
67 gfn_t table_gfn[PT_MAX_FULL_LEVELS];
68 pt_element_t ptes[PT_MAX_FULL_LEVELS];
69 gpa_t pte_gpa[PT_MAX_FULL_LEVELS];
76 static gfn_t gpte_to_gfn(pt_element_t gpte)
78 return (gpte & PT_BASE_ADDR_MASK) >> PAGE_SHIFT;
81 static gfn_t gpte_to_gfn_pde(pt_element_t gpte)
83 return (gpte & PT_DIR_BASE_ADDR_MASK) >> PAGE_SHIFT;
86 static 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)
94 page = gfn_to_page(kvm, table_gfn);
95 table = kmap_atomic(page, KM_USER0);
97 ret = CMPXCHG(&table[index], orig_pte, new_pte);
99 kunmap_atomic(table, KM_USER0);
101 kvm_release_page_dirty(page);
103 return (ret != orig_pte);
106 static unsigned FNAME(gpte_access)(struct kvm_vcpu *vcpu, pt_element_t gpte)
110 access = (gpte & (PT_WRITABLE_MASK | PT_USER_MASK)) | ACC_EXEC_MASK;
113 access &= ~(gpte >> PT64_NX_SHIFT);
119 * Fetch a guest pte for a guest virtual address
121 static int FNAME(walk_addr)(struct guest_walker *walker,
122 struct kvm_vcpu *vcpu, gva_t addr,
123 int write_fault, int user_fault, int fetch_fault)
127 unsigned index, pt_access, pte_access;
130 pgprintk("%s: addr %lx\n", __FUNCTION__, addr);
132 walker->level = vcpu->arch.mmu.root_level;
133 pte = vcpu->arch.cr3;
135 if (!is_long_mode(vcpu)) {
136 pte = vcpu->arch.pdptrs[(addr >> 30) & 3];
137 if (!is_present_pte(pte))
142 ASSERT((!is_long_mode(vcpu) && is_pae(vcpu)) ||
143 (vcpu->cr3 & CR3_NONPAE_RESERVED_BITS) == 0);
148 index = PT_INDEX(addr, walker->level);
150 table_gfn = gpte_to_gfn(pte);
151 pte_gpa = gfn_to_gpa(table_gfn);
152 pte_gpa += index * sizeof(pt_element_t);
153 walker->table_gfn[walker->level - 1] = table_gfn;
154 walker->pte_gpa[walker->level - 1] = pte_gpa;
155 pgprintk("%s: table_gfn[%d] %lx\n", __FUNCTION__,
156 walker->level - 1, table_gfn);
158 kvm_read_guest(vcpu->kvm, pte_gpa, &pte, sizeof(pte));
160 if (!is_present_pte(pte))
163 if (write_fault && !is_writeble_pte(pte))
164 if (user_fault || is_write_protection(vcpu))
167 if (user_fault && !(pte & PT_USER_MASK))
171 if (fetch_fault && is_nx(vcpu) && (pte & PT64_NX_MASK))
175 if (!(pte & PT_ACCESSED_MASK)) {
176 mark_page_dirty(vcpu->kvm, table_gfn);
177 if (FNAME(cmpxchg_gpte)(vcpu->kvm, table_gfn,
178 index, pte, pte|PT_ACCESSED_MASK))
180 pte |= PT_ACCESSED_MASK;
183 pte_access = pt_access & FNAME(gpte_access)(vcpu, pte);
185 walker->ptes[walker->level - 1] = pte;
187 if (walker->level == PT_PAGE_TABLE_LEVEL) {
188 walker->gfn = gpte_to_gfn(pte);
192 if (walker->level == PT_DIRECTORY_LEVEL
193 && (pte & PT_PAGE_SIZE_MASK)
194 && (PTTYPE == 64 || is_pse(vcpu))) {
195 walker->gfn = gpte_to_gfn_pde(pte);
196 walker->gfn += PT_INDEX(addr, PT_PAGE_TABLE_LEVEL);
197 if (PTTYPE == 32 && is_cpuid_PSE36())
198 walker->gfn += pse36_gfn_delta(pte);
202 pt_access = pte_access;
206 if (write_fault && !is_dirty_pte(pte)) {
209 mark_page_dirty(vcpu->kvm, table_gfn);
210 ret = FNAME(cmpxchg_gpte)(vcpu->kvm, table_gfn, index, pte,
214 pte |= PT_DIRTY_MASK;
215 kvm_mmu_pte_write(vcpu, pte_gpa, (u8 *)&pte, sizeof(pte));
216 walker->ptes[walker->level - 1] = pte;
219 walker->pt_access = pt_access;
220 walker->pte_access = pte_access;
221 pgprintk("%s: pte %llx pte_access %x pt_access %x\n",
222 __FUNCTION__, (u64)pte, pt_access, pte_access);
226 walker->error_code = 0;
230 walker->error_code = PFERR_PRESENT_MASK;
234 walker->error_code |= PFERR_WRITE_MASK;
236 walker->error_code |= PFERR_USER_MASK;
238 walker->error_code |= PFERR_FETCH_MASK;
242 static void FNAME(update_pte)(struct kvm_vcpu *vcpu, struct kvm_mmu_page *page,
243 u64 *spte, const void *pte, int bytes,
250 gpte = *(const pt_element_t *)pte;
251 if (~gpte & (PT_PRESENT_MASK | PT_ACCESSED_MASK)) {
252 if (!offset_in_pte && !is_present_pte(gpte))
253 set_shadow_pte(spte, shadow_notrap_nonpresent_pte);
256 if (bytes < sizeof(pt_element_t))
258 pgprintk("%s: gpte %llx spte %p\n", __FUNCTION__, (u64)gpte, spte);
259 pte_access = page->role.access & FNAME(gpte_access)(vcpu, gpte);
260 if (gpte_to_gfn(gpte) != vcpu->arch.update_pte.gfn)
262 npage = vcpu->arch.update_pte.page;
266 mmu_set_spte(vcpu, spte, page->role.access, pte_access, 0, 0,
267 gpte & PT_DIRTY_MASK, NULL, gpte_to_gfn(gpte), npage);
271 * Fetch a shadow pte for a specific level in the paging hierarchy.
273 static u64 *FNAME(fetch)(struct kvm_vcpu *vcpu, gva_t addr,
274 struct guest_walker *walker,
275 int user_fault, int write_fault, int *ptwrite,
281 unsigned access = walker->pt_access;
283 if (!is_present_pte(walker->ptes[walker->level - 1]))
286 shadow_addr = vcpu->arch.mmu.root_hpa;
287 level = vcpu->arch.mmu.shadow_root_level;
288 if (level == PT32E_ROOT_LEVEL) {
289 shadow_addr = vcpu->arch.mmu.pae_root[(addr >> 30) & 3];
290 shadow_addr &= PT64_BASE_ADDR_MASK;
295 u32 index = SHADOW_PT_INDEX(addr, level);
296 struct kvm_mmu_page *shadow_page;
302 shadow_ent = ((u64 *)__va(shadow_addr)) + index;
303 if (is_shadow_present_pte(*shadow_ent)) {
304 if (level == PT_PAGE_TABLE_LEVEL)
306 shadow_addr = *shadow_ent & PT64_BASE_ADDR_MASK;
310 if (level == PT_PAGE_TABLE_LEVEL)
313 if (level - 1 == PT_PAGE_TABLE_LEVEL
314 && walker->level == PT_DIRECTORY_LEVEL) {
316 if (!is_dirty_pte(walker->ptes[level - 1]))
317 access &= ~ACC_WRITE_MASK;
318 table_gfn = gpte_to_gfn(walker->ptes[level - 1]);
321 table_gfn = walker->table_gfn[level - 2];
323 shadow_page = kvm_mmu_get_page(vcpu, table_gfn, addr, level-1,
324 metaphysical, access,
325 shadow_ent, &new_page);
326 if (new_page && !metaphysical) {
328 pt_element_t curr_pte;
329 r = kvm_read_guest_atomic(vcpu->kvm,
330 walker->pte_gpa[level - 2],
331 &curr_pte, sizeof(curr_pte));
332 if (r || curr_pte != walker->ptes[level - 2]) {
333 kvm_release_page_clean(page);
337 shadow_addr = __pa(shadow_page->spt);
338 shadow_pte = shadow_addr | PT_PRESENT_MASK | PT_ACCESSED_MASK
339 | PT_WRITABLE_MASK | PT_USER_MASK;
340 *shadow_ent = shadow_pte;
343 mmu_set_spte(vcpu, shadow_ent, access, walker->pte_access & access,
344 user_fault, write_fault,
345 walker->ptes[walker->level-1] & PT_DIRTY_MASK,
346 ptwrite, walker->gfn, page);
352 * 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
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
362 * Returns: 1 if we need to emulate the instruction, 0 otherwise, or
363 * a negative value on error.
365 static int FNAME(page_fault)(struct kvm_vcpu *vcpu, gva_t addr,
368 int write_fault = error_code & PFERR_WRITE_MASK;
369 int user_fault = error_code & PFERR_USER_MASK;
370 int fetch_fault = error_code & PFERR_FETCH_MASK;
371 struct guest_walker walker;
377 pgprintk("%s: addr %lx err %x\n", __FUNCTION__, addr, error_code);
378 kvm_mmu_audit(vcpu, "pre page fault");
380 r = mmu_topup_memory_caches(vcpu);
384 down_read(¤t->mm->mmap_sem);
386 * Look up the shadow pte for the faulting address.
388 r = FNAME(walk_addr)(&walker, vcpu, addr, write_fault, user_fault,
392 * The page is not mapped by the guest. Let the guest handle it.
395 pgprintk("%s: guest page fault\n", __FUNCTION__);
396 inject_page_fault(vcpu, addr, walker.error_code);
397 vcpu->arch.last_pt_write_count = 0; /* reset fork detector */
398 up_read(¤t->mm->mmap_sem);
402 page = gfn_to_page(vcpu->kvm, walker.gfn);
404 spin_lock(&vcpu->kvm->mmu_lock);
405 kvm_mmu_free_some_pages(vcpu);
406 shadow_pte = FNAME(fetch)(vcpu, addr, &walker, user_fault, write_fault,
408 pgprintk("%s: shadow pte %p %llx ptwrite %d\n", __FUNCTION__,
409 shadow_pte, *shadow_pte, write_pt);
412 vcpu->arch.last_pt_write_count = 0; /* reset fork detector */
415 * mmio: emulate if accessible, otherwise its a guest fault.
417 if (shadow_pte && is_io_pte(*shadow_pte)) {
418 spin_unlock(&vcpu->kvm->mmu_lock);
419 up_read(¤t->mm->mmap_sem);
423 ++vcpu->stat.pf_fixed;
424 kvm_mmu_audit(vcpu, "post page fault (fixed)");
425 spin_unlock(&vcpu->kvm->mmu_lock);
426 up_read(¤t->mm->mmap_sem);
431 static gpa_t FNAME(gva_to_gpa)(struct kvm_vcpu *vcpu, gva_t vaddr)
433 struct guest_walker walker;
434 gpa_t gpa = UNMAPPED_GVA;
437 r = FNAME(walk_addr)(&walker, vcpu, vaddr, 0, 0, 0);
440 gpa = gfn_to_gpa(walker.gfn);
441 gpa |= vaddr & ~PAGE_MASK;
447 static void FNAME(prefetch_page)(struct kvm_vcpu *vcpu,
448 struct kvm_mmu_page *sp)
450 int i, offset = 0, r = 0;
453 if (sp->role.metaphysical
454 || (PTTYPE == 32 && sp->role.level > PT_PAGE_TABLE_LEVEL)) {
455 nonpaging_prefetch_page(vcpu, sp);
460 offset = sp->role.quadrant << PT64_LEVEL_BITS;
462 for (i = 0; i < PT64_ENT_PER_PAGE; ++i) {
463 gpa_t pte_gpa = gfn_to_gpa(sp->gfn);
464 pte_gpa += (i+offset) * sizeof(pt_element_t);
466 r = kvm_read_guest_atomic(vcpu->kvm, pte_gpa, &pt,
467 sizeof(pt_element_t));
468 if (r || is_present_pte(pt))
469 sp->spt[i] = shadow_trap_nonpresent_pte;
471 sp->spt[i] = shadow_notrap_nonpresent_pte;
478 #undef PT_BASE_ADDR_MASK
480 #undef SHADOW_PT_INDEX
482 #undef PT_DIR_BASE_ADDR_MASK
484 #undef PT_MAX_FULL_LEVELS
486 #undef gpte_to_gfn_pde