arch/x86/kvm/paging_tmpl.h

   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_LEVEL_BITS PT64_LEVEL_BITS
  35         #ifdef CONFIG_X86_64
  36         #define PT_MAX_FULL_LEVELS 4
  37         #define CMPXCHG cmpxchg
  38         #else
  39         #define CMPXCHG cmpxchg64
  40         #define PT_MAX_FULL_LEVELS 2
  41         #endif
  42 #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)
  51         #define PT_LEVEL_BITS PT32_LEVEL_BITS
  52         #define PT_MAX_FULL_LEVELS 2
  53         #define CMPXCHG cmpxchg
  54 #else
  55         #error Invalid PTTYPE value
  56 #endif
  57
  58 #define gpte_to_gfn FNAME(gpte_to_gfn)
  59 #define gpte_to_gfn_pde FNAME(gpte_to_gfn_pde)
  60
  61 /*
  62  * The guest_walker structure emulates the behavior of the hardware page
  63  * table walker.
  64  */
  65 struct guest_walker {
  66         int level;
  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];
  70         unsigned pt_access;
  71         unsigned pte_access;
  72         gfn_t gfn;
  73         u32 error_code;
  74 };
  75
  76 static gfn_t gpte_to_gfn(pt_element_t gpte)
  77 {
  78         return (gpte & PT_BASE_ADDR_MASK) >> PAGE_SHIFT;
  79 }
  80
  81 static gfn_t gpte_to_gfn_pde(pt_element_t gpte)
  82 {
  83         return (gpte & PT_DIR_BASE_ADDR_MASK) >> PAGE_SHIFT;
  84 }
  85
  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)
  89 {
  90         pt_element_t ret;
  91         pt_element_t *table;
  92         struct page *page;
  93
  94         page = gfn_to_page(kvm, table_gfn);
  95         table = kmap_atomic(page, KM_USER0);
  96
  97         ret = CMPXCHG(&table[index], orig_pte, new_pte);
  98
  99         kunmap_atomic(table, KM_USER0);
 100
 101         kvm_release_page_dirty(page);
 102
 103         return (ret != orig_pte);
 104 }
 105
 106 static unsigned FNAME(gpte_access)(struct kvm_vcpu *vcpu, pt_element_t gpte)
 107 {
 108         unsigned access;
 109
 110         access = (gpte & (PT_WRITABLE_MASK | PT_USER_MASK)) | ACC_EXEC_MASK;
 111 #if PTTYPE == 64
 112         if (is_nx(vcpu))
 113                 access &= ~(gpte >> PT64_NX_SHIFT);
 114 #endif
 115         return access;
 116 }
 117
 118 /*
 119  * Fetch a guest pte for a guest virtual address
 120  */
 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)
 124 {
 125         pt_element_t pte;
 126         gfn_t table_gfn;
 127         unsigned index, pt_access, pte_access;
 128         gpa_t pte_gpa;
 129
 130         pgprintk("%s: addr %lx\n", __FUNCTION__, addr);
 131 walk:
 132         walker->level = vcpu->arch.mmu.root_level;
 133         pte = vcpu->arch.cr3;
 134 #if PTTYPE == 64
 135         if (!is_long_mode(vcpu)) {
 136                 pte = vcpu->arch.pdptrs[(addr >> 30) & 3];
 137                 if (!is_present_pte(pte))
 138                         goto not_present;
 139                 --walker->level;
 140         }
 141 #endif
 142         ASSERT((!is_long_mode(vcpu) && is_pae(vcpu)) ||
 143                (vcpu->cr3 & CR3_NONPAE_RESERVED_BITS) == 0);
 144
 145         pt_access = ACC_ALL;
 146
 147         for (;;) {
 148                 index = PT_INDEX(addr, walker->level);
 149
 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);
 157
 158                 kvm_read_guest(vcpu->kvm, pte_gpa, &pte, sizeof(pte));
 159
 160                 if (!is_present_pte(pte))
 161                         goto not_present;
 162
 163                 if (write_fault && !is_writeble_pte(pte))
 164                         if (user_fault || is_write_protection(vcpu))
 165                                 goto access_error;
 166
 167                 if (user_fault && !(pte & PT_USER_MASK))
 168                         goto access_error;
 169
 170 #if PTTYPE == 64
 171                 if (fetch_fault && is_nx(vcpu) && (pte & PT64_NX_MASK))
 172                         goto access_error;
 173 #endif
 174
 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))
 179                                 goto walk;
 180                         pte |= PT_ACCESSED_MASK;
 181                 }
 182
 183                 pte_access = pt_access & FNAME(gpte_access)(vcpu, pte);
 184
 185                 walker->ptes[walker->level - 1] = pte;
 186
 187                 if (walker->level == PT_PAGE_TABLE_LEVEL) {
 188                         walker->gfn = gpte_to_gfn(pte);
 189                         break;
 190                 }
 191
 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);
 199                         break;
 200                 }
 201
 202                 pt_access = pte_access;
 203                 --walker->level;
 204         }
 205
 206         if (write_fault && !is_dirty_pte(pte)) {
 207                 bool ret;
 208
 209                 mark_page_dirty(vcpu->kvm, table_gfn);
 210                 ret = FNAME(cmpxchg_gpte)(vcpu->kvm, table_gfn, index, pte,
 211                             pte|PT_DIRTY_MASK);
 212                 if (ret)
 213                         goto walk;
 214                 pte |= PT_DIRTY_MASK;
 215                 kvm_mmu_pte_write(vcpu, pte_gpa, (u8 *)&pte, sizeof(pte));
 216                 walker->ptes[walker->level - 1] = pte;
 217         }
 218
 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);
 223         return 1;
 224
 225 not_present:
 226         walker->error_code = 0;
 227         goto err;
 228
 229 access_error:
 230         walker->error_code = PFERR_PRESENT_MASK;
 231
 232 err:
 233         if (write_fault)
 234                 walker->error_code |= PFERR_WRITE_MASK;
 235         if (user_fault)
 236                 walker->error_code |= PFERR_USER_MASK;
 237         if (fetch_fault)
 238                 walker->error_code |= PFERR_FETCH_MASK;
 239         return 0;
 240 }
 241
 242 static void FNAME(update_pte)(struct kvm_vcpu *vcpu, struct kvm_mmu_page *page,
 243                               u64 *spte, const void *pte, int bytes,
 244                               int offset_in_pte)
 245 {
 246         pt_element_t gpte;
 247         unsigned pte_access;
 248         struct page *npage;
 249
 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);
 254                 return;
 255         }
 256         if (bytes < sizeof(pt_element_t))
 257                 return;
 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)
 261                 return;
 262         npage = vcpu->arch.update_pte.page;
 263         if (!npage)
 264                 return;
 265         get_page(npage);
 266         mmu_set_spte(vcpu, spte, page->role.access, pte_access, 0, 0,
 267                      gpte & PT_DIRTY_MASK, NULL, gpte_to_gfn(gpte), npage);
 268 }
 269
 270 /*
 271  * Fetch a shadow pte for a specific level in the paging hierarchy.
 272  */
 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,
 276                          struct page *page)
 277 {
 278         hpa_t shadow_addr;
 279         int level;
 280         u64 *shadow_ent;
 281         unsigned access = walker->pt_access;
 282
 283         if (!is_present_pte(walker->ptes[walker->level - 1]))
 284                 return NULL;
 285
 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;
 291                 --level;
 292         }
 293
 294         for (; ; level--) {
 295                 u32 index = SHADOW_PT_INDEX(addr, level);
 296                 struct kvm_mmu_page *shadow_page;
 297                 u64 shadow_pte;
 298                 int metaphysical;
 299                 gfn_t table_gfn;
 300                 bool new_page = 0;
 301
 302                 shadow_ent = ((u64 *)__va(shadow_addr)) + index;
 303                 if (level == PT_PAGE_TABLE_LEVEL)
 304                         break;
 305                 if (is_shadow_present_pte(*shadow_ent)) {
 306                         shadow_addr = *shadow_ent & PT64_BASE_ADDR_MASK;
 307                         continue;
 308                 }
 309
 310                 if (level - 1 == PT_PAGE_TABLE_LEVEL
 311                     && walker->level == PT_DIRECTORY_LEVEL) {
 312                         metaphysical = 1;
 313                         if (!is_dirty_pte(walker->ptes[level - 1]))
 314                                 access &= ~ACC_WRITE_MASK;
 315                         table_gfn = gpte_to_gfn(walker->ptes[level - 1]);
 316                 } else {
 317                         metaphysical = 0;
 318                         table_gfn = walker->table_gfn[level - 2];
 319                 }
 320                 shadow_page = kvm_mmu_get_page(vcpu, table_gfn, addr, level-1,
 321                                                metaphysical, access,
 322                                                shadow_ent, &new_page);
 323                 if (new_page && !metaphysical) {
 324                         int r;
 325                         pt_element_t curr_pte;
 326                         r = kvm_read_guest_atomic(vcpu->kvm,
 327                                                   walker->pte_gpa[level - 2],
 328                                                   &curr_pte, sizeof(curr_pte));
 329                         if (r || curr_pte != walker->ptes[level - 2]) {
 330                                 kvm_release_page_clean(page);
 331                                 return NULL;
 332                         }
 333                 }
 334                 shadow_addr = __pa(shadow_page->spt);
 335                 shadow_pte = shadow_addr | PT_PRESENT_MASK | PT_ACCESSED_MASK
 336                         | PT_WRITABLE_MASK | PT_USER_MASK;
 337                 *shadow_ent = shadow_pte;
 338         }
 339
 340         mmu_set_spte(vcpu, shadow_ent, access, walker->pte_access & access,
 341                      user_fault, write_fault,
 342                      walker->ptes[walker->level-1] & PT_DIRTY_MASK,
 343                      ptwrite, walker->gfn, page);
 344
 345         return shadow_ent;
 346 }
 347
 348 /*
 349  * Page fault handler.  There are several causes for a page fault:
 350  *   - there is no shadow pte for the guest pte
 351  *   - write access through a shadow pte marked read only so that we can set
 352  *     the dirty bit
 353  *   - write access to a shadow pte marked read only so we can update the page
 354  *     dirty bitmap, when userspace requests it
 355  *   - mmio access; in this case we will never install a present shadow pte
 356  *   - normal guest page fault due to the guest pte marked not present, not
 357  *     writable, or not executable
 358  *
 359  *  Returns: 1 if we need to emulate the instruction, 0 otherwise, or
 360  *           a negative value on error.
 361  */
 362 static int FNAME(page_fault)(struct kvm_vcpu *vcpu, gva_t addr,
 363                                u32 error_code)
 364 {
 365         int write_fault = error_code & PFERR_WRITE_MASK;
 366         int user_fault = error_code & PFERR_USER_MASK;
 367         int fetch_fault = error_code & PFERR_FETCH_MASK;
 368         struct guest_walker walker;
 369         u64 *shadow_pte;
 370         int write_pt = 0;
 371         int r;
 372         struct page *page;
 373
 374         pgprintk("%s: addr %lx err %x\n", __FUNCTION__, addr, error_code);
 375         kvm_mmu_audit(vcpu, "pre page fault");
 376
 377         r = mmu_topup_memory_caches(vcpu);
 378         if (r)
 379                 return r;
 380
 381         down_read(&current->mm->mmap_sem);
 382         /*
 383          * Look up the shadow pte for the faulting address.
 384          */
 385         r = FNAME(walk_addr)(&walker, vcpu, addr, write_fault, user_fault,
 386                              fetch_fault);
 387
 388         /*
 389          * The page is not mapped by the guest.  Let the guest handle it.
 390          */
 391         if (!r) {
 392                 pgprintk("%s: guest page fault\n", __FUNCTION__);
 393                 inject_page_fault(vcpu, addr, walker.error_code);
 394                 vcpu->arch.last_pt_write_count = 0; /* reset fork detector */
 395                 up_read(&current->mm->mmap_sem);
 396                 return 0;
 397         }
 398
 399         page = gfn_to_page(vcpu->kvm, walker.gfn);
 400
 401         spin_lock(&vcpu->kvm->mmu_lock);
 402         kvm_mmu_free_some_pages(vcpu);
 403         shadow_pte = FNAME(fetch)(vcpu, addr, &walker, user_fault, write_fault,
 404                                   &write_pt, page);
 405         pgprintk("%s: shadow pte %p %llx ptwrite %d\n", __FUNCTION__,
 406                  shadow_pte, *shadow_pte, write_pt);
 407
 408         if (!write_pt)
 409                 vcpu->arch.last_pt_write_count = 0; /* reset fork detector */
 410
 411         /*
 412          * mmio: emulate if accessible, otherwise its a guest fault.
 413          */
 414         if (shadow_pte && is_io_pte(*shadow_pte)) {
 415                 spin_unlock(&vcpu->kvm->mmu_lock);
 416                 up_read(&current->mm->mmap_sem);
 417                 return 1;
 418         }
 419
 420         ++vcpu->stat.pf_fixed;
 421         kvm_mmu_audit(vcpu, "post page fault (fixed)");
 422         spin_unlock(&vcpu->kvm->mmu_lock);
 423         up_read(&current->mm->mmap_sem);
 424
 425         return write_pt;
 426 }
 427
 428 static gpa_t FNAME(gva_to_gpa)(struct kvm_vcpu *vcpu, gva_t vaddr)
 429 {
 430         struct guest_walker walker;
 431         gpa_t gpa = UNMAPPED_GVA;
 432         int r;
 433
 434         r = FNAME(walk_addr)(&walker, vcpu, vaddr, 0, 0, 0);
 435
 436         if (r) {
 437                 gpa = gfn_to_gpa(walker.gfn);
 438                 gpa |= vaddr & ~PAGE_MASK;
 439         }
 440
 441         return gpa;
 442 }
 443
 444 static void FNAME(prefetch_page)(struct kvm_vcpu *vcpu,
 445                                  struct kvm_mmu_page *sp)
 446 {
 447         int i, offset = 0, r = 0;
 448         pt_element_t pt;
 449
 450         if (sp->role.metaphysical
 451             || (PTTYPE == 32 && sp->role.level > PT_PAGE_TABLE_LEVEL)) {
 452                 nonpaging_prefetch_page(vcpu, sp);
 453                 return;
 454         }
 455
 456         if (PTTYPE == 32)
 457                 offset = sp->role.quadrant << PT64_LEVEL_BITS;
 458
 459         for (i = 0; i < PT64_ENT_PER_PAGE; ++i) {
 460                 gpa_t pte_gpa = gfn_to_gpa(sp->gfn);
 461                 pte_gpa += (i+offset) * sizeof(pt_element_t);
 462
 463                 r = kvm_read_guest_atomic(vcpu->kvm, pte_gpa, &pt,
 464                                           sizeof(pt_element_t));
 465                 if (r || is_present_pte(pt))
 466                         sp->spt[i] = shadow_trap_nonpresent_pte;
 467                 else
 468                         sp->spt[i] = shadow_notrap_nonpresent_pte;
 469         }
 470 }
 471
 472 #undef pt_element_t
 473 #undef guest_walker
 474 #undef FNAME
 475 #undef PT_BASE_ADDR_MASK
 476 #undef PT_INDEX
 477 #undef SHADOW_PT_INDEX
 478 #undef PT_LEVEL_MASK
 479 #undef PT_DIR_BASE_ADDR_MASK
 480 #undef PT_LEVEL_BITS
 481 #undef PT_MAX_FULL_LEVELS
 482 #undef gpte_to_gfn
 483 #undef gpte_to_gfn_pde
 484 #undef CMPXCHG