arch/mips/kvm/mmu.c

   1 /*
   2  * This file is subject to the terms and conditions of the GNU General Public
   3  * License.  See the file "COPYING" in the main directory of this archive
   4  * for more details.
   5  *
   6  * KVM/MIPS MMU handling in the KVM module.
   7  *
   8  * Copyright (C) 2012  MIPS Technologies, Inc.  All rights reserved.
   9  * Authors: Sanjay Lal <sanjayl@kymasys.com>
  10  */
  11
  12 #include <linux/highmem.h>
  13 #include <linux/kvm_host.h>
  14 #include <linux/uaccess.h>
  15 #include <asm/mmu_context.h>
  16 #include <asm/pgalloc.h>
  17
  18 /*
  19  * KVM_MMU_CACHE_MIN_PAGES is the number of GPA page table translation levels
  20  * for which pages need to be cached.
  21  */
  22 #if defined(__PAGETABLE_PMD_FOLDED)
  23 #define KVM_MMU_CACHE_MIN_PAGES 1
  24 #else
  25 #define KVM_MMU_CACHE_MIN_PAGES 2
  26 #endif
  27
  28 void kvm_mmu_free_memory_caches(struct kvm_vcpu *vcpu)
  29 {
  30         kvm_mmu_free_memory_cache(&vcpu->arch.mmu_page_cache);
  31 }
  32
  33 /**
  34  * kvm_pgd_init() - Initialise KVM GPA page directory.
  35  * @page:       Pointer to page directory (PGD) for KVM GPA.
  36  *
  37  * Initialise a KVM GPA page directory with pointers to the invalid table, i.e.
  38  * representing no mappings. This is similar to pgd_init(), however it
  39  * initialises all the page directory pointers, not just the ones corresponding
  40  * to the userland address space (since it is for the guest physical address
  41  * space rather than a virtual address space).
  42  */
  43 static void kvm_pgd_init(void *page)
  44 {
  45         unsigned long *p, *end;
  46         unsigned long entry;
  47
  48 #ifdef __PAGETABLE_PMD_FOLDED
  49         entry = (unsigned long)invalid_pte_table;
  50 #else
  51         entry = (unsigned long)invalid_pmd_table;
  52 #endif
  53
  54         p = (unsigned long *)page;
  55         end = p + PTRS_PER_PGD;
  56
  57         do {
  58                 p[0] = entry;
  59                 p[1] = entry;
  60                 p[2] = entry;
  61                 p[3] = entry;
  62                 p[4] = entry;
  63                 p += 8;
  64                 p[-3] = entry;
  65                 p[-2] = entry;
  66                 p[-1] = entry;
  67         } while (p != end);
  68 }
  69
  70 /**
  71  * kvm_pgd_alloc() - Allocate and initialise a KVM GPA page directory.
  72  *
  73  * Allocate a blank KVM GPA page directory (PGD) for representing guest physical
  74  * to host physical page mappings.
  75  *
  76  * Returns:     Pointer to new KVM GPA page directory.
  77  *              NULL on allocation failure.
  78  */
  79 pgd_t *kvm_pgd_alloc(void)
  80 {
  81         pgd_t *ret;
  82
  83         ret = (pgd_t *)__get_free_pages(GFP_KERNEL, PGD_ORDER);
  84         if (ret)
  85                 kvm_pgd_init(ret);
  86
  87         return ret;
  88 }
  89
  90 /**
  91  * kvm_mips_walk_pgd() - Walk page table with optional allocation.
  92  * @pgd:        Page directory pointer.
  93  * @addr:       Address to index page table using.
  94  * @cache:      MMU page cache to allocate new page tables from, or NULL.
  95  *
  96  * Walk the page tables pointed to by @pgd to find the PTE corresponding to the
  97  * address @addr. If page tables don't exist for @addr, they will be created
  98  * from the MMU cache if @cache is not NULL.
  99  *
 100  * Returns:     Pointer to pte_t corresponding to @addr.
 101  *              NULL if a page table doesn't exist for @addr and !@cache.
 102  *              NULL if a page table allocation failed.
 103  */
 104 static pte_t *kvm_mips_walk_pgd(pgd_t *pgd, struct kvm_mmu_memory_cache *cache,
 105                                 unsigned long addr)
 106 {
 107         p4d_t *p4d;
 108         pud_t *pud;
 109         pmd_t *pmd;
 110
 111         pgd += pgd_index(addr);
 112         if (pgd_none(*pgd)) {
 113                 /* Not used on MIPS yet */
 114                 BUG();
 115                 return NULL;
 116         }
 117         p4d = p4d_offset(pgd, addr);
 118         pud = pud_offset(p4d, addr);
 119         if (pud_none(*pud)) {
 120                 pmd_t *new_pmd;
 121
 122                 if (!cache)
 123                         return NULL;
 124                 new_pmd = kvm_mmu_memory_cache_alloc(cache);
 125                 pmd_init((unsigned long)new_pmd,
 126                          (unsigned long)invalid_pte_table);
 127                 pud_populate(NULL, pud, new_pmd);
 128         }
 129         pmd = pmd_offset(pud, addr);
 130         if (pmd_none(*pmd)) {
 131                 pte_t *new_pte;
 132
 133                 if (!cache)
 134                         return NULL;
 135                 new_pte = kvm_mmu_memory_cache_alloc(cache);
 136                 clear_page(new_pte);
 137                 pmd_populate_kernel(NULL, pmd, new_pte);
 138         }
 139         return pte_offset_kernel(pmd, addr);
 140 }
 141
 142 /* Caller must hold kvm->mm_lock */
 143 static pte_t *kvm_mips_pte_for_gpa(struct kvm *kvm,
 144                                    struct kvm_mmu_memory_cache *cache,
 145                                    unsigned long addr)
 146 {
 147         return kvm_mips_walk_pgd(kvm->arch.gpa_mm.pgd, cache, addr);
 148 }
 149
 150 /*
 151  * kvm_mips_flush_gpa_{pte,pmd,pud,pgd,pt}.
 152  * Flush a range of guest physical address space from the VM's GPA page tables.
 153  */
 154
 155 static bool kvm_mips_flush_gpa_pte(pte_t *pte, unsigned long start_gpa,
 156                                    unsigned long end_gpa)
 157 {
 158         int i_min = pte_index(start_gpa);
 159         int i_max = pte_index(end_gpa);
 160         bool safe_to_remove = (i_min == 0 && i_max == PTRS_PER_PTE - 1);
 161         int i;
 162
 163         for (i = i_min; i <= i_max; ++i) {
 164                 if (!pte_present(pte[i]))
 165                         continue;
 166
 167                 set_pte(pte + i, __pte(0));
 168         }
 169         return safe_to_remove;
 170 }
 171
 172 static bool kvm_mips_flush_gpa_pmd(pmd_t *pmd, unsigned long start_gpa,
 173                                    unsigned long end_gpa)
 174 {
 175         pte_t *pte;
 176         unsigned long end = ~0ul;
 177         int i_min = pmd_index(start_gpa);
 178         int i_max = pmd_index(end_gpa);
 179         bool safe_to_remove = (i_min == 0 && i_max == PTRS_PER_PMD - 1);
 180         int i;
 181
 182         for (i = i_min; i <= i_max; ++i, start_gpa = 0) {
 183                 if (!pmd_present(pmd[i]))
 184                         continue;
 185
 186                 pte = pte_offset_kernel(pmd + i, 0);
 187                 if (i == i_max)
 188                         end = end_gpa;
 189
 190                 if (kvm_mips_flush_gpa_pte(pte, start_gpa, end)) {
 191                         pmd_clear(pmd + i);
 192                         pte_free_kernel(NULL, pte);
 193                 } else {
 194                         safe_to_remove = false;
 195                 }
 196         }
 197         return safe_to_remove;
 198 }
 199
 200 static bool kvm_mips_flush_gpa_pud(pud_t *pud, unsigned long start_gpa,
 201                                    unsigned long end_gpa)
 202 {
 203         pmd_t *pmd;
 204         unsigned long end = ~0ul;
 205         int i_min = pud_index(start_gpa);
 206         int i_max = pud_index(end_gpa);
 207         bool safe_to_remove = (i_min == 0 && i_max == PTRS_PER_PUD - 1);
 208         int i;
 209
 210         for (i = i_min; i <= i_max; ++i, start_gpa = 0) {
 211                 if (!pud_present(pud[i]))
 212                         continue;
 213
 214                 pmd = pmd_offset(pud + i, 0);
 215                 if (i == i_max)
 216                         end = end_gpa;
 217
 218                 if (kvm_mips_flush_gpa_pmd(pmd, start_gpa, end)) {
 219                         pud_clear(pud + i);
 220                         pmd_free(NULL, pmd);
 221                 } else {
 222                         safe_to_remove = false;
 223                 }
 224         }
 225         return safe_to_remove;
 226 }
 227
 228 static bool kvm_mips_flush_gpa_pgd(pgd_t *pgd, unsigned long start_gpa,
 229                                    unsigned long end_gpa)
 230 {
 231         p4d_t *p4d;
 232         pud_t *pud;
 233         unsigned long end = ~0ul;
 234         int i_min = pgd_index(start_gpa);
 235         int i_max = pgd_index(end_gpa);
 236         bool safe_to_remove = (i_min == 0 && i_max == PTRS_PER_PGD - 1);
 237         int i;
 238
 239         for (i = i_min; i <= i_max; ++i, start_gpa = 0) {
 240                 if (!pgd_present(pgd[i]))
 241                         continue;
 242
 243                 p4d = p4d_offset(pgd, 0);
 244                 pud = pud_offset(p4d + i, 0);
 245                 if (i == i_max)
 246                         end = end_gpa;
 247
 248                 if (kvm_mips_flush_gpa_pud(pud, start_gpa, end)) {
 249                         pgd_clear(pgd + i);
 250                         pud_free(NULL, pud);
 251                 } else {
 252                         safe_to_remove = false;
 253                 }
 254         }
 255         return safe_to_remove;
 256 }
 257
 258 /**
 259  * kvm_mips_flush_gpa_pt() - Flush a range of guest physical addresses.
 260  * @kvm:        KVM pointer.
 261  * @start_gfn:  Guest frame number of first page in GPA range to flush.
 262  * @end_gfn:    Guest frame number of last page in GPA range to flush.
 263  *
 264  * Flushes a range of GPA mappings from the GPA page tables.
 265  *
 266  * The caller must hold the @kvm->mmu_lock spinlock.
 267  *
 268  * Returns:     Whether its safe to remove the top level page directory because
 269  *              all lower levels have been removed.
 270  */
 271 bool kvm_mips_flush_gpa_pt(struct kvm *kvm, gfn_t start_gfn, gfn_t end_gfn)
 272 {
 273         return kvm_mips_flush_gpa_pgd(kvm->arch.gpa_mm.pgd,
 274                                       start_gfn << PAGE_SHIFT,
 275                                       end_gfn << PAGE_SHIFT);
 276 }
 277
 278 #define BUILD_PTE_RANGE_OP(name, op)                                    \
 279 static int kvm_mips_##name##_pte(pte_t *pte, unsigned long start,       \
 280                                  unsigned long end)                     \
 281 {                                                                       \
 282         int ret = 0;                                                    \
 283         int i_min = pte_index(start);                           \
 284         int i_max = pte_index(end);                                     \
 285         int i;                                                          \
 286         pte_t old, new;                                                 \
 287                                                                         \
 288         for (i = i_min; i <= i_max; ++i) {                              \
 289                 if (!pte_present(pte[i]))                               \
 290                         continue;                                       \
 291                                                                         \
 292                 old = pte[i];                                           \
 293                 new = op(old);                                          \
 294                 if (pte_val(new) == pte_val(old))                       \
 295                         continue;                                       \
 296                 set_pte(pte + i, new);                                  \
 297                 ret = 1;                                                \
 298         }                                                               \
 299         return ret;                                                     \
 300 }                                                                       \
 301                                                                         \
 302 /* returns true if anything was done */                                 \
 303 static int kvm_mips_##name##_pmd(pmd_t *pmd, unsigned long start,       \
 304                                  unsigned long end)                     \
 305 {                                                                       \
 306         int ret = 0;                                                    \
 307         pte_t *pte;                                                     \
 308         unsigned long cur_end = ~0ul;                                   \
 309         int i_min = pmd_index(start);                           \
 310         int i_max = pmd_index(end);                                     \
 311         int i;                                                          \
 312                                                                         \
 313         for (i = i_min; i <= i_max; ++i, start = 0) {                   \
 314                 if (!pmd_present(pmd[i]))                               \
 315                         continue;                                       \
 316                                                                         \
 317                 pte = pte_offset_kernel(pmd + i, 0);                            \
 318                 if (i == i_max)                                         \
 319                         cur_end = end;                                  \
 320                                                                         \
 321                 ret |= kvm_mips_##name##_pte(pte, start, cur_end);      \
 322         }                                                               \
 323         return ret;                                                     \
 324 }                                                                       \
 325                                                                         \
 326 static int kvm_mips_##name##_pud(pud_t *pud, unsigned long start,       \
 327                                  unsigned long end)                     \
 328 {                                                                       \
 329         int ret = 0;                                                    \
 330         pmd_t *pmd;                                                     \
 331         unsigned long cur_end = ~0ul;                                   \
 332         int i_min = pud_index(start);                           \
 333         int i_max = pud_index(end);                                     \
 334         int i;                                                          \
 335                                                                         \
 336         for (i = i_min; i <= i_max; ++i, start = 0) {                   \
 337                 if (!pud_present(pud[i]))                               \
 338                         continue;                                       \
 339                                                                         \
 340                 pmd = pmd_offset(pud + i, 0);                           \
 341                 if (i == i_max)                                         \
 342                         cur_end = end;                                  \
 343                                                                         \
 344                 ret |= kvm_mips_##name##_pmd(pmd, start, cur_end);      \
 345         }                                                               \
 346         return ret;                                                     \
 347 }                                                                       \
 348                                                                         \
 349 static int kvm_mips_##name##_pgd(pgd_t *pgd, unsigned long start,       \
 350                                  unsigned long end)                     \
 351 {                                                                       \
 352         int ret = 0;                                                    \
 353         p4d_t *p4d;                                                     \
 354         pud_t *pud;                                                     \
 355         unsigned long cur_end = ~0ul;                                   \
 356         int i_min = pgd_index(start);                                   \
 357         int i_max = pgd_index(end);                                     \
 358         int i;                                                          \
 359                                                                         \
 360         for (i = i_min; i <= i_max; ++i, start = 0) {                   \
 361                 if (!pgd_present(pgd[i]))                               \
 362                         continue;                                       \
 363                                                                         \
 364                 p4d = p4d_offset(pgd, 0);                               \
 365                 pud = pud_offset(p4d + i, 0);                           \
 366                 if (i == i_max)                                         \
 367                         cur_end = end;                                  \
 368                                                                         \
 369                 ret |= kvm_mips_##name##_pud(pud, start, cur_end);      \
 370         }                                                               \
 371         return ret;                                                     \
 372 }
 373
 374 /*
 375  * kvm_mips_mkclean_gpa_pt.
 376  * Mark a range of guest physical address space clean (writes fault) in the VM's
 377  * GPA page table to allow dirty page tracking.
 378  */
 379
 380 BUILD_PTE_RANGE_OP(mkclean, pte_mkclean)
 381
 382 /**
 383  * kvm_mips_mkclean_gpa_pt() - Make a range of guest physical addresses clean.
 384  * @kvm:        KVM pointer.
 385  * @start_gfn:  Guest frame number of first page in GPA range to flush.
 386  * @end_gfn:    Guest frame number of last page in GPA range to flush.
 387  *
 388  * Make a range of GPA mappings clean so that guest writes will fault and
 389  * trigger dirty page logging.
 390  *
 391  * The caller must hold the @kvm->mmu_lock spinlock.
 392  *
 393  * Returns:     Whether any GPA mappings were modified, which would require
 394  *              derived mappings (GVA page tables & TLB enties) to be
 395  *              invalidated.
 396  */
 397 int kvm_mips_mkclean_gpa_pt(struct kvm *kvm, gfn_t start_gfn, gfn_t end_gfn)
 398 {
 399         return kvm_mips_mkclean_pgd(kvm->arch.gpa_mm.pgd,
 400                                     start_gfn << PAGE_SHIFT,
 401                                     end_gfn << PAGE_SHIFT);
 402 }
 403
 404 /**
 405  * kvm_arch_mmu_enable_log_dirty_pt_masked() - write protect dirty pages
 406  * @kvm:        The KVM pointer
 407  * @slot:       The memory slot associated with mask
 408  * @gfn_offset: The gfn offset in memory slot
 409  * @mask:       The mask of dirty pages at offset 'gfn_offset' in this memory
 410  *              slot to be write protected
 411  *
 412  * Walks bits set in mask write protects the associated pte's. Caller must
 413  * acquire @kvm->mmu_lock.
 414  */
 415 void kvm_arch_mmu_enable_log_dirty_pt_masked(struct kvm *kvm,
 416                 struct kvm_memory_slot *slot,
 417                 gfn_t gfn_offset, unsigned long mask)
 418 {
 419         gfn_t base_gfn = slot->base_gfn + gfn_offset;
 420         gfn_t start = base_gfn +  __ffs(mask);
 421         gfn_t end = base_gfn + __fls(mask);
 422
 423         kvm_mips_mkclean_gpa_pt(kvm, start, end);
 424 }
 425
 426 /*
 427  * kvm_mips_mkold_gpa_pt.
 428  * Mark a range of guest physical address space old (all accesses fault) in the
 429  * VM's GPA page table to allow detection of commonly used pages.
 430  */
 431
 432 BUILD_PTE_RANGE_OP(mkold, pte_mkold)
 433
 434 static int kvm_mips_mkold_gpa_pt(struct kvm *kvm, gfn_t start_gfn,
 435                                  gfn_t end_gfn)
 436 {
 437         return kvm_mips_mkold_pgd(kvm->arch.gpa_mm.pgd,
 438                                   start_gfn << PAGE_SHIFT,
 439                                   end_gfn << PAGE_SHIFT);
 440 }
 441
 442 static int handle_hva_to_gpa(struct kvm *kvm,
 443                              unsigned long start,
 444                              unsigned long end,
 445                              int (*handler)(struct kvm *kvm, gfn_t gfn,
 446                                             gpa_t gfn_end,
 447                                             struct kvm_memory_slot *memslot,
 448                                             void *data),
 449                              void *data)
 450 {
 451         struct kvm_memslots *slots;
 452         struct kvm_memory_slot *memslot;
 453         int ret = 0;
 454
 455         slots = kvm_memslots(kvm);
 456
 457         /* we only care about the pages that the guest sees */
 458         kvm_for_each_memslot(memslot, slots) {
 459                 unsigned long hva_start, hva_end;
 460                 gfn_t gfn, gfn_end;
 461
 462                 hva_start = max(start, memslot->userspace_addr);
 463                 hva_end = min(end, memslot->userspace_addr +
 464                                         (memslot->npages << PAGE_SHIFT));
 465                 if (hva_start >= hva_end)
 466                         continue;
 467
 468                 /*
 469                  * {gfn(page) | page intersects with [hva_start, hva_end)} =
 470                  * {gfn_start, gfn_start+1, ..., gfn_end-1}.
 471                  */
 472                 gfn = hva_to_gfn_memslot(hva_start, memslot);
 473                 gfn_end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, memslot);
 474
 475                 ret |= handler(kvm, gfn, gfn_end, memslot, data);
 476         }
 477
 478         return ret;
 479 }
 480
 481
 482 static int kvm_unmap_hva_handler(struct kvm *kvm, gfn_t gfn, gfn_t gfn_end,
 483                                  struct kvm_memory_slot *memslot, void *data)
 484 {
 485         kvm_mips_flush_gpa_pt(kvm, gfn, gfn_end);
 486         return 1;
 487 }
 488
 489 int kvm_unmap_hva_range(struct kvm *kvm, unsigned long start, unsigned long end,
 490                         unsigned flags)
 491 {
 492         handle_hva_to_gpa(kvm, start, end, &kvm_unmap_hva_handler, NULL);
 493
 494         kvm_mips_callbacks->prepare_flush_shadow(kvm);
 495         kvm_flush_remote_tlbs(kvm);
 496         return 0;
 497 }
 498
 499 static int kvm_set_spte_handler(struct kvm *kvm, gfn_t gfn, gfn_t gfn_end,
 500                                 struct kvm_memory_slot *memslot, void *data)
 501 {
 502         gpa_t gpa = gfn << PAGE_SHIFT;
 503         pte_t hva_pte = *(pte_t *)data;
 504         pte_t *gpa_pte = kvm_mips_pte_for_gpa(kvm, NULL, gpa);
 505         pte_t old_pte;
 506
 507         if (!gpa_pte)
 508                 return 0;
 509
 510         /* Mapping may need adjusting depending on memslot flags */
 511         old_pte = *gpa_pte;
 512         if (memslot->flags & KVM_MEM_LOG_DIRTY_PAGES && !pte_dirty(old_pte))
 513                 hva_pte = pte_mkclean(hva_pte);
 514         else if (memslot->flags & KVM_MEM_READONLY)
 515                 hva_pte = pte_wrprotect(hva_pte);
 516
 517         set_pte(gpa_pte, hva_pte);
 518
 519         /* Replacing an absent or old page doesn't need flushes */
 520         if (!pte_present(old_pte) || !pte_young(old_pte))
 521                 return 0;
 522
 523         /* Pages swapped, aged, moved, or cleaned require flushes */
 524         return !pte_present(hva_pte) ||
 525                !pte_young(hva_pte) ||
 526                pte_pfn(old_pte) != pte_pfn(hva_pte) ||
 527                (pte_dirty(old_pte) && !pte_dirty(hva_pte));
 528 }
 529
 530 int kvm_set_spte_hva(struct kvm *kvm, unsigned long hva, pte_t pte)
 531 {
 532         unsigned long end = hva + PAGE_SIZE;
 533         int ret;
 534
 535         ret = handle_hva_to_gpa(kvm, hva, end, &kvm_set_spte_handler, &pte);
 536         if (ret) {
 537                 kvm_mips_callbacks->prepare_flush_shadow(kvm);
 538                 kvm_flush_remote_tlbs(kvm);
 539         }
 540         return 0;
 541 }
 542
 543 static int kvm_age_hva_handler(struct kvm *kvm, gfn_t gfn, gfn_t gfn_end,
 544                                struct kvm_memory_slot *memslot, void *data)
 545 {
 546         return kvm_mips_mkold_gpa_pt(kvm, gfn, gfn_end);
 547 }
 548
 549 static int kvm_test_age_hva_handler(struct kvm *kvm, gfn_t gfn, gfn_t gfn_end,
 550                                     struct kvm_memory_slot *memslot, void *data)
 551 {
 552         gpa_t gpa = gfn << PAGE_SHIFT;
 553         pte_t *gpa_pte = kvm_mips_pte_for_gpa(kvm, NULL, gpa);
 554
 555         if (!gpa_pte)
 556                 return 0;
 557         return pte_young(*gpa_pte);
 558 }
 559
 560 int kvm_age_hva(struct kvm *kvm, unsigned long start, unsigned long end)
 561 {
 562         return handle_hva_to_gpa(kvm, start, end, kvm_age_hva_handler, NULL);
 563 }
 564
 565 int kvm_test_age_hva(struct kvm *kvm, unsigned long hva)
 566 {
 567         return handle_hva_to_gpa(kvm, hva, hva, kvm_test_age_hva_handler, NULL);
 568 }
 569
 570 /**
 571  * _kvm_mips_map_page_fast() - Fast path GPA fault handler.
 572  * @vcpu:               VCPU pointer.
 573  * @gpa:                Guest physical address of fault.
 574  * @write_fault:        Whether the fault was due to a write.
 575  * @out_entry:          New PTE for @gpa (written on success unless NULL).
 576  * @out_buddy:          New PTE for @gpa's buddy (written on success unless
 577  *                      NULL).
 578  *
 579  * Perform fast path GPA fault handling, doing all that can be done without
 580  * calling into KVM. This handles marking old pages young (for idle page
 581  * tracking), and dirtying of clean pages (for dirty page logging).
 582  *
 583  * Returns:     0 on success, in which case we can update derived mappings and
 584  *              resume guest execution.
 585  *              -EFAULT on failure due to absent GPA mapping or write to
 586  *              read-only page, in which case KVM must be consulted.
 587  */
 588 static int _kvm_mips_map_page_fast(struct kvm_vcpu *vcpu, unsigned long gpa,
 589                                    bool write_fault,
 590                                    pte_t *out_entry, pte_t *out_buddy)
 591 {
 592         struct kvm *kvm = vcpu->kvm;
 593         gfn_t gfn = gpa >> PAGE_SHIFT;
 594         pte_t *ptep;
 595         kvm_pfn_t pfn = 0;      /* silence bogus GCC warning */
 596         bool pfn_valid = false;
 597         int ret = 0;
 598
 599         spin_lock(&kvm->mmu_lock);
 600
 601         /* Fast path - just check GPA page table for an existing entry */
 602         ptep = kvm_mips_pte_for_gpa(kvm, NULL, gpa);
 603         if (!ptep || !pte_present(*ptep)) {
 604                 ret = -EFAULT;
 605                 goto out;
 606         }
 607
 608         /* Track access to pages marked old */
 609         if (!pte_young(*ptep)) {
 610                 set_pte(ptep, pte_mkyoung(*ptep));
 611                 pfn = pte_pfn(*ptep);
 612                 pfn_valid = true;
 613                 /* call kvm_set_pfn_accessed() after unlock */
 614         }
 615         if (write_fault && !pte_dirty(*ptep)) {
 616                 if (!pte_write(*ptep)) {
 617                         ret = -EFAULT;
 618                         goto out;
 619                 }
 620
 621                 /* Track dirtying of writeable pages */
 622                 set_pte(ptep, pte_mkdirty(*ptep));
 623                 pfn = pte_pfn(*ptep);
 624                 mark_page_dirty(kvm, gfn);
 625                 kvm_set_pfn_dirty(pfn);
 626         }
 627
 628         if (out_entry)
 629                 *out_entry = *ptep;
 630         if (out_buddy)
 631                 *out_buddy = *ptep_buddy(ptep);
 632
 633 out:
 634         spin_unlock(&kvm->mmu_lock);
 635         if (pfn_valid)
 636                 kvm_set_pfn_accessed(pfn);
 637         return ret;
 638 }
 639
 640 /**
 641  * kvm_mips_map_page() - Map a guest physical page.
 642  * @vcpu:               VCPU pointer.
 643  * @gpa:                Guest physical address of fault.
 644  * @write_fault:        Whether the fault was due to a write.
 645  * @out_entry:          New PTE for @gpa (written on success unless NULL).
 646  * @out_buddy:          New PTE for @gpa's buddy (written on success unless
 647  *                      NULL).
 648  *
 649  * Handle GPA faults by creating a new GPA mapping (or updating an existing
 650  * one).
 651  *
 652  * This takes care of marking pages young or dirty (idle/dirty page tracking),
 653  * asking KVM for the corresponding PFN, and creating a mapping in the GPA page
 654  * tables. Derived mappings (GVA page tables and TLBs) must be handled by the
 655  * caller.
 656  *
 657  * Returns:     0 on success, in which case the caller may use the @out_entry
 658  *              and @out_buddy PTEs to update derived mappings and resume guest
 659  *              execution.
 660  *              -EFAULT if there is no memory region at @gpa or a write was
 661  *              attempted to a read-only memory region. This is usually handled
 662  *              as an MMIO access.
 663  */
 664 static int kvm_mips_map_page(struct kvm_vcpu *vcpu, unsigned long gpa,
 665                              bool write_fault,
 666                              pte_t *out_entry, pte_t *out_buddy)
 667 {
 668         struct kvm *kvm = vcpu->kvm;
 669         struct kvm_mmu_memory_cache *memcache = &vcpu->arch.mmu_page_cache;
 670         gfn_t gfn = gpa >> PAGE_SHIFT;
 671         int srcu_idx, err;
 672         kvm_pfn_t pfn;
 673         pte_t *ptep, entry, old_pte;
 674         bool writeable;
 675         unsigned long prot_bits;
 676         unsigned long mmu_seq;
 677
 678         /* Try the fast path to handle old / clean pages */
 679         srcu_idx = srcu_read_lock(&kvm->srcu);
 680         err = _kvm_mips_map_page_fast(vcpu, gpa, write_fault, out_entry,
 681                                       out_buddy);
 682         if (!err)
 683                 goto out;
 684
 685         /* We need a minimum of cached pages ready for page table creation */
 686         err = kvm_mmu_topup_memory_cache(memcache, KVM_MMU_CACHE_MIN_PAGES);
 687         if (err)
 688                 goto out;
 689
 690 retry:
 691         /*
 692          * Used to check for invalidations in progress, of the pfn that is
 693          * returned by pfn_to_pfn_prot below.
 694          */
 695         mmu_seq = kvm->mmu_notifier_seq;
 696         /*
 697          * Ensure the read of mmu_notifier_seq isn't reordered with PTE reads in
 698          * gfn_to_pfn_prot() (which calls get_user_pages()), so that we don't
 699          * risk the page we get a reference to getting unmapped before we have a
 700          * chance to grab the mmu_lock without mmu_notifier_retry() noticing.
 701          *
 702          * This smp_rmb() pairs with the effective smp_wmb() of the combination
 703          * of the pte_unmap_unlock() after the PTE is zapped, and the
 704          * spin_lock() in kvm_mmu_notifier_invalidate_<page|range_end>() before
 705          * mmu_notifier_seq is incremented.
 706          */
 707         smp_rmb();
 708
 709         /* Slow path - ask KVM core whether we can access this GPA */
 710         pfn = gfn_to_pfn_prot(kvm, gfn, write_fault, &writeable);
 711         if (is_error_noslot_pfn(pfn)) {
 712                 err = -EFAULT;
 713                 goto out;
 714         }
 715
 716         spin_lock(&kvm->mmu_lock);
 717         /* Check if an invalidation has taken place since we got pfn */
 718         if (mmu_notifier_retry(kvm, mmu_seq)) {
 719                 /*
 720                  * This can happen when mappings are changed asynchronously, but
 721                  * also synchronously if a COW is triggered by
 722                  * gfn_to_pfn_prot().
 723                  */
 724                 spin_unlock(&kvm->mmu_lock);
 725                 kvm_release_pfn_clean(pfn);
 726                 goto retry;
 727         }
 728
 729         /* Ensure page tables are allocated */
 730         ptep = kvm_mips_pte_for_gpa(kvm, memcache, gpa);
 731
 732         /* Set up the PTE */
 733         prot_bits = _PAGE_PRESENT | __READABLE | _page_cachable_default;
 734         if (writeable) {
 735                 prot_bits |= _PAGE_WRITE;
 736                 if (write_fault) {
 737                         prot_bits |= __WRITEABLE;
 738                         mark_page_dirty(kvm, gfn);
 739                         kvm_set_pfn_dirty(pfn);
 740                 }
 741         }
 742         entry = pfn_pte(pfn, __pgprot(prot_bits));
 743
 744         /* Write the PTE */
 745         old_pte = *ptep;
 746         set_pte(ptep, entry);
 747
 748         err = 0;
 749         if (out_entry)
 750                 *out_entry = *ptep;
 751         if (out_buddy)
 752                 *out_buddy = *ptep_buddy(ptep);
 753
 754         spin_unlock(&kvm->mmu_lock);
 755         kvm_release_pfn_clean(pfn);
 756         kvm_set_pfn_accessed(pfn);
 757 out:
 758         srcu_read_unlock(&kvm->srcu, srcu_idx);
 759         return err;
 760 }
 761
 762 static pte_t *kvm_trap_emul_pte_for_gva(struct kvm_vcpu *vcpu,
 763                                         unsigned long addr)
 764 {
 765         struct kvm_mmu_memory_cache *memcache = &vcpu->arch.mmu_page_cache;
 766         pgd_t *pgdp;
 767         int ret;
 768
 769         /* We need a minimum of cached pages ready for page table creation */
 770         ret = kvm_mmu_topup_memory_cache(memcache, KVM_MMU_CACHE_MIN_PAGES);
 771         if (ret)
 772                 return NULL;
 773
 774         if (KVM_GUEST_KERNEL_MODE(vcpu))
 775                 pgdp = vcpu->arch.guest_kernel_mm.pgd;
 776         else
 777                 pgdp = vcpu->arch.guest_user_mm.pgd;
 778
 779         return kvm_mips_walk_pgd(pgdp, memcache, addr);
 780 }
 781
 782 void kvm_trap_emul_invalidate_gva(struct kvm_vcpu *vcpu, unsigned long addr,
 783                                   bool user)
 784 {
 785         pgd_t *pgdp;
 786         pte_t *ptep;
 787
 788         addr &= PAGE_MASK << 1;
 789
 790         pgdp = vcpu->arch.guest_kernel_mm.pgd;
 791         ptep = kvm_mips_walk_pgd(pgdp, NULL, addr);
 792         if (ptep) {
 793                 ptep[0] = pfn_pte(0, __pgprot(0));
 794                 ptep[1] = pfn_pte(0, __pgprot(0));
 795         }
 796
 797         if (user) {
 798                 pgdp = vcpu->arch.guest_user_mm.pgd;
 799                 ptep = kvm_mips_walk_pgd(pgdp, NULL, addr);
 800                 if (ptep) {
 801                         ptep[0] = pfn_pte(0, __pgprot(0));
 802                         ptep[1] = pfn_pte(0, __pgprot(0));
 803                 }
 804         }
 805 }
 806
 807 /*
 808  * kvm_mips_flush_gva_{pte,pmd,pud,pgd,pt}.
 809  * Flush a range of guest physical address space from the VM's GPA page tables.
 810  */
 811
 812 static bool kvm_mips_flush_gva_pte(pte_t *pte, unsigned long start_gva,
 813                                    unsigned long end_gva)
 814 {
 815         int i_min = pte_index(start_gva);
 816         int i_max = pte_index(end_gva);
 817         bool safe_to_remove = (i_min == 0 && i_max == PTRS_PER_PTE - 1);
 818         int i;
 819
 820         /*
 821          * There's no freeing to do, so there's no point clearing individual
 822          * entries unless only part of the last level page table needs flushing.
 823          */
 824         if (safe_to_remove)
 825                 return true;
 826
 827         for (i = i_min; i <= i_max; ++i) {
 828                 if (!pte_present(pte[i]))
 829                         continue;
 830
 831                 set_pte(pte + i, __pte(0));
 832         }
 833         return false;
 834 }
 835
 836 static bool kvm_mips_flush_gva_pmd(pmd_t *pmd, unsigned long start_gva,
 837                                    unsigned long end_gva)
 838 {
 839         pte_t *pte;
 840         unsigned long end = ~0ul;
 841         int i_min = pmd_index(start_gva);
 842         int i_max = pmd_index(end_gva);
 843         bool safe_to_remove = (i_min == 0 && i_max == PTRS_PER_PMD - 1);
 844         int i;
 845
 846         for (i = i_min; i <= i_max; ++i, start_gva = 0) {
 847                 if (!pmd_present(pmd[i]))
 848                         continue;
 849
 850                 pte = pte_offset_kernel(pmd + i, 0);
 851                 if (i == i_max)
 852                         end = end_gva;
 853
 854                 if (kvm_mips_flush_gva_pte(pte, start_gva, end)) {
 855                         pmd_clear(pmd + i);
 856                         pte_free_kernel(NULL, pte);
 857                 } else {
 858                         safe_to_remove = false;
 859                 }
 860         }
 861         return safe_to_remove;
 862 }
 863
 864 static bool kvm_mips_flush_gva_pud(pud_t *pud, unsigned long start_gva,
 865                                    unsigned long end_gva)
 866 {
 867         pmd_t *pmd;
 868         unsigned long end = ~0ul;
 869         int i_min = pud_index(start_gva);
 870         int i_max = pud_index(end_gva);
 871         bool safe_to_remove = (i_min == 0 && i_max == PTRS_PER_PUD - 1);
 872         int i;
 873
 874         for (i = i_min; i <= i_max; ++i, start_gva = 0) {
 875                 if (!pud_present(pud[i]))
 876                         continue;
 877
 878                 pmd = pmd_offset(pud + i, 0);
 879                 if (i == i_max)
 880                         end = end_gva;
 881
 882                 if (kvm_mips_flush_gva_pmd(pmd, start_gva, end)) {
 883                         pud_clear(pud + i);
 884                         pmd_free(NULL, pmd);
 885                 } else {
 886                         safe_to_remove = false;
 887                 }
 888         }
 889         return safe_to_remove;
 890 }
 891
 892 static bool kvm_mips_flush_gva_pgd(pgd_t *pgd, unsigned long start_gva,
 893                                    unsigned long end_gva)
 894 {
 895         p4d_t *p4d;
 896         pud_t *pud;
 897         unsigned long end = ~0ul;
 898         int i_min = pgd_index(start_gva);
 899         int i_max = pgd_index(end_gva);
 900         bool safe_to_remove = (i_min == 0 && i_max == PTRS_PER_PGD - 1);
 901         int i;
 902
 903         for (i = i_min; i <= i_max; ++i, start_gva = 0) {
 904                 if (!pgd_present(pgd[i]))
 905                         continue;
 906
 907                 p4d = p4d_offset(pgd, 0);
 908                 pud = pud_offset(p4d + i, 0);
 909                 if (i == i_max)
 910                         end = end_gva;
 911
 912                 if (kvm_mips_flush_gva_pud(pud, start_gva, end)) {
 913                         pgd_clear(pgd + i);
 914                         pud_free(NULL, pud);
 915                 } else {
 916                         safe_to_remove = false;
 917                 }
 918         }
 919         return safe_to_remove;
 920 }
 921
 922 void kvm_mips_flush_gva_pt(pgd_t *pgd, enum kvm_mips_flush flags)
 923 {
 924         if (flags & KMF_GPA) {
 925                 /* all of guest virtual address space could be affected */
 926                 if (flags & KMF_KERN)
 927                         /* useg, kseg0, seg2/3 */
 928                         kvm_mips_flush_gva_pgd(pgd, 0, 0x7fffffff);
 929                 else
 930                         /* useg */
 931                         kvm_mips_flush_gva_pgd(pgd, 0, 0x3fffffff);
 932         } else {
 933                 /* useg */
 934                 kvm_mips_flush_gva_pgd(pgd, 0, 0x3fffffff);
 935
 936                 /* kseg2/3 */
 937                 if (flags & KMF_KERN)
 938                         kvm_mips_flush_gva_pgd(pgd, 0x60000000, 0x7fffffff);
 939         }
 940 }
 941
 942 static pte_t kvm_mips_gpa_pte_to_gva_unmapped(pte_t pte)
 943 {
 944         /*
 945          * Don't leak writeable but clean entries from GPA page tables. We don't
 946          * want the normal Linux tlbmod handler to handle dirtying when KVM
 947          * accesses guest memory.
 948          */
 949         if (!pte_dirty(pte))
 950                 pte = pte_wrprotect(pte);
 951
 952         return pte;
 953 }
 954
 955 static pte_t kvm_mips_gpa_pte_to_gva_mapped(pte_t pte, long entrylo)
 956 {
 957         /* Guest EntryLo overrides host EntryLo */
 958         if (!(entrylo & ENTRYLO_D))
 959                 pte = pte_mkclean(pte);
 960
 961         return kvm_mips_gpa_pte_to_gva_unmapped(pte);
 962 }
 963
 964 #ifdef CONFIG_KVM_MIPS_VZ
 965 int kvm_mips_handle_vz_root_tlb_fault(unsigned long badvaddr,
 966                                       struct kvm_vcpu *vcpu,
 967                                       bool write_fault)
 968 {
 969         int ret;
 970
 971         ret = kvm_mips_map_page(vcpu, badvaddr, write_fault, NULL, NULL);
 972         if (ret)
 973                 return ret;
 974
 975         /* Invalidate this entry in the TLB */
 976         return kvm_vz_host_tlb_inv(vcpu, badvaddr);
 977 }
 978 #endif
 979
 980 /* XXXKYMA: Must be called with interrupts disabled */
 981 int kvm_mips_handle_kseg0_tlb_fault(unsigned long badvaddr,
 982                                     struct kvm_vcpu *vcpu,
 983                                     bool write_fault)
 984 {
 985         unsigned long gpa;
 986         pte_t pte_gpa[2], *ptep_gva;
 987         int idx;
 988
 989         if (KVM_GUEST_KSEGX(badvaddr) != KVM_GUEST_KSEG0) {
 990                 kvm_err("%s: Invalid BadVaddr: %#lx\n", __func__, badvaddr);
 991                 kvm_mips_dump_host_tlbs();
 992                 return -1;
 993         }
 994
 995         /* Get the GPA page table entry */
 996         gpa = KVM_GUEST_CPHYSADDR(badvaddr);
 997         idx = (badvaddr >> PAGE_SHIFT) & 1;
 998         if (kvm_mips_map_page(vcpu, gpa, write_fault, &pte_gpa[idx],
 999                               &pte_gpa[!idx]) < 0)
1000                 return -1;
1001
1002         /* Get the GVA page table entry */
1003         ptep_gva = kvm_trap_emul_pte_for_gva(vcpu, badvaddr & ~PAGE_SIZE);
1004         if (!ptep_gva) {
1005                 kvm_err("No ptep for gva %lx\n", badvaddr);
1006                 return -1;
1007         }
1008
1009         /* Copy a pair of entries from GPA page table to GVA page table */
1010         ptep_gva[0] = kvm_mips_gpa_pte_to_gva_unmapped(pte_gpa[0]);
1011         ptep_gva[1] = kvm_mips_gpa_pte_to_gva_unmapped(pte_gpa[1]);
1012
1013         /* Invalidate this entry in the TLB, guest kernel ASID only */
1014         kvm_mips_host_tlb_inv(vcpu, badvaddr, false, true);
1015         return 0;
1016 }
1017
1018 int kvm_mips_handle_mapped_seg_tlb_fault(struct kvm_vcpu *vcpu,
1019                                          struct kvm_mips_tlb *tlb,
1020                                          unsigned long gva,
1021                                          bool write_fault)
1022 {
1023         struct kvm *kvm = vcpu->kvm;
1024         long tlb_lo[2];
1025         pte_t pte_gpa[2], *ptep_buddy, *ptep_gva;
1026         unsigned int idx = TLB_LO_IDX(*tlb, gva);
1027         bool kernel = KVM_GUEST_KERNEL_MODE(vcpu);
1028
1029         tlb_lo[0] = tlb->tlb_lo[0];
1030         tlb_lo[1] = tlb->tlb_lo[1];
1031
1032         /*
1033          * The commpage address must not be mapped to anything else if the guest
1034          * TLB contains entries nearby, or commpage accesses will break.
1035          */
1036         if (!((gva ^ KVM_GUEST_COMMPAGE_ADDR) & VPN2_MASK & (PAGE_MASK << 1)))
1037                 tlb_lo[TLB_LO_IDX(*tlb, KVM_GUEST_COMMPAGE_ADDR)] = 0;
1038
1039         /* Get the GPA page table entry */
1040         if (kvm_mips_map_page(vcpu, mips3_tlbpfn_to_paddr(tlb_lo[idx]),
1041                               write_fault, &pte_gpa[idx], NULL) < 0)
1042                 return -1;
1043
1044         /* And its GVA buddy's GPA page table entry if it also exists */
1045         pte_gpa[!idx] = pfn_pte(0, __pgprot(0));
1046         if (tlb_lo[!idx] & ENTRYLO_V) {
1047                 spin_lock(&kvm->mmu_lock);
1048                 ptep_buddy = kvm_mips_pte_for_gpa(kvm, NULL,
1049                                         mips3_tlbpfn_to_paddr(tlb_lo[!idx]));
1050                 if (ptep_buddy)
1051                         pte_gpa[!idx] = *ptep_buddy;
1052                 spin_unlock(&kvm->mmu_lock);
1053         }
1054
1055         /* Get the GVA page table entry pair */
1056         ptep_gva = kvm_trap_emul_pte_for_gva(vcpu, gva & ~PAGE_SIZE);
1057         if (!ptep_gva) {
1058                 kvm_err("No ptep for gva %lx\n", gva);
1059                 return -1;
1060         }
1061
1062         /* Copy a pair of entries from GPA page table to GVA page table */
1063         ptep_gva[0] = kvm_mips_gpa_pte_to_gva_mapped(pte_gpa[0], tlb_lo[0]);
1064         ptep_gva[1] = kvm_mips_gpa_pte_to_gva_mapped(pte_gpa[1], tlb_lo[1]);
1065
1066         /* Invalidate this entry in the TLB, current guest mode ASID only */
1067         kvm_mips_host_tlb_inv(vcpu, gva, !kernel, kernel);
1068
1069         kvm_debug("@ %#lx tlb_lo0: 0x%08lx tlb_lo1: 0x%08lx\n", vcpu->arch.pc,
1070                   tlb->tlb_lo[0], tlb->tlb_lo[1]);
1071
1072         return 0;
1073 }
1074
1075 int kvm_mips_handle_commpage_tlb_fault(unsigned long badvaddr,
1076                                        struct kvm_vcpu *vcpu)
1077 {
1078         kvm_pfn_t pfn;
1079         pte_t *ptep;
1080         pgprot_t prot;
1081
1082         ptep = kvm_trap_emul_pte_for_gva(vcpu, badvaddr);
1083         if (!ptep) {
1084                 kvm_err("No ptep for commpage %lx\n", badvaddr);
1085                 return -1;
1086         }
1087
1088         pfn = PFN_DOWN(virt_to_phys(vcpu->arch.kseg0_commpage));
1089         /* Also set valid and dirty, so refill handler doesn't have to */
1090         prot = vm_get_page_prot(VM_READ|VM_WRITE|VM_SHARED);
1091         *ptep = pte_mkyoung(pte_mkdirty(pfn_pte(pfn, prot)));
1092
1093         /* Invalidate this entry in the TLB, guest kernel ASID only */
1094         kvm_mips_host_tlb_inv(vcpu, badvaddr, false, true);
1095         return 0;
1096 }
1097
1098 /**
1099  * kvm_mips_migrate_count() - Migrate timer.
1100  * @vcpu:       Virtual CPU.
1101  *
1102  * Migrate CP0_Count hrtimer to the current CPU by cancelling and restarting it
1103  * if it was running prior to being cancelled.
1104  *
1105  * Must be called when the VCPU is migrated to a different CPU to ensure that
1106  * timer expiry during guest execution interrupts the guest and causes the
1107  * interrupt to be delivered in a timely manner.
1108  */
1109 static void kvm_mips_migrate_count(struct kvm_vcpu *vcpu)
1110 {
1111         if (hrtimer_cancel(&vcpu->arch.comparecount_timer))
1112                 hrtimer_restart(&vcpu->arch.comparecount_timer);
1113 }
1114
1115 /* Restore ASID once we are scheduled back after preemption */
1116 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1117 {
1118         unsigned long flags;
1119
1120         kvm_debug("%s: vcpu %p, cpu: %d\n", __func__, vcpu, cpu);
1121
1122         local_irq_save(flags);
1123
1124         vcpu->cpu = cpu;
1125         if (vcpu->arch.last_sched_cpu != cpu) {
1126                 kvm_debug("[%d->%d]KVM VCPU[%d] switch\n",
1127                           vcpu->arch.last_sched_cpu, cpu, vcpu->vcpu_id);
1128                 /*
1129                  * Migrate the timer interrupt to the current CPU so that it
1130                  * always interrupts the guest and synchronously triggers a
1131                  * guest timer interrupt.
1132                  */
1133                 kvm_mips_migrate_count(vcpu);
1134         }
1135
1136         /* restore guest state to registers */
1137         kvm_mips_callbacks->vcpu_load(vcpu, cpu);
1138
1139         local_irq_restore(flags);
1140 }
1141
1142 /* ASID can change if another task is scheduled during preemption */
1143 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1144 {
1145         unsigned long flags;
1146         int cpu;
1147
1148         local_irq_save(flags);
1149
1150         cpu = smp_processor_id();
1151         vcpu->arch.last_sched_cpu = cpu;
1152         vcpu->cpu = -1;
1153
1154         /* save guest state in registers */
1155         kvm_mips_callbacks->vcpu_put(vcpu, cpu);
1156
1157         local_irq_restore(flags);
1158 }
1159
1160 /**
1161  * kvm_trap_emul_gva_fault() - Safely attempt to handle a GVA access fault.
1162  * @vcpu:       Virtual CPU.
1163  * @gva:        Guest virtual address to be accessed.
1164  * @write:      True if write attempted (must be dirtied and made writable).
1165  *
1166  * Safely attempt to handle a GVA fault, mapping GVA pages if necessary, and
1167  * dirtying the page if @write so that guest instructions can be modified.
1168  *
1169  * Returns:     KVM_MIPS_MAPPED on success.
1170  *              KVM_MIPS_GVA if bad guest virtual address.
1171  *              KVM_MIPS_GPA if bad guest physical address.
1172  *              KVM_MIPS_TLB if guest TLB not present.
1173  *              KVM_MIPS_TLBINV if guest TLB present but not valid.
1174  *              KVM_MIPS_TLBMOD if guest TLB read only.
1175  */
1176 enum kvm_mips_fault_result kvm_trap_emul_gva_fault(struct kvm_vcpu *vcpu,
1177                                                    unsigned long gva,
1178                                                    bool write)
1179 {
1180         struct mips_coproc *cop0 = vcpu->arch.cop0;
1181         struct kvm_mips_tlb *tlb;
1182         int index;
1183
1184         if (KVM_GUEST_KSEGX(gva) == KVM_GUEST_KSEG0) {
1185                 if (kvm_mips_handle_kseg0_tlb_fault(gva, vcpu, write) < 0)
1186                         return KVM_MIPS_GPA;
1187         } else if ((KVM_GUEST_KSEGX(gva) < KVM_GUEST_KSEG0) ||
1188                    KVM_GUEST_KSEGX(gva) == KVM_GUEST_KSEG23) {
1189                 /* Address should be in the guest TLB */
1190                 index = kvm_mips_guest_tlb_lookup(vcpu, (gva & VPN2_MASK) |
1191                           (kvm_read_c0_guest_entryhi(cop0) & KVM_ENTRYHI_ASID));
1192                 if (index < 0)
1193                         return KVM_MIPS_TLB;
1194                 tlb = &vcpu->arch.guest_tlb[index];
1195
1196                 /* Entry should be valid, and dirty for writes */
1197                 if (!TLB_IS_VALID(*tlb, gva))
1198                         return KVM_MIPS_TLBINV;
1199                 if (write && !TLB_IS_DIRTY(*tlb, gva))
1200                         return KVM_MIPS_TLBMOD;
1201
1202                 if (kvm_mips_handle_mapped_seg_tlb_fault(vcpu, tlb, gva, write))
1203                         return KVM_MIPS_GPA;
1204         } else {
1205                 return KVM_MIPS_GVA;
1206         }
1207
1208         return KVM_MIPS_MAPPED;
1209 }
1210
1211 int kvm_get_inst(u32 *opc, struct kvm_vcpu *vcpu, u32 *out)
1212 {
1213         int err;
1214
1215         if (WARN(IS_ENABLED(CONFIG_KVM_MIPS_VZ),
1216                  "Expect BadInstr/BadInstrP registers to be used with VZ\n"))
1217                 return -EINVAL;
1218
1219 retry:
1220         kvm_trap_emul_gva_lockless_begin(vcpu);
1221         err = get_user(*out, opc);
1222         kvm_trap_emul_gva_lockless_end(vcpu);
1223
1224         if (unlikely(err)) {
1225                 /*
1226                  * Try to handle the fault, maybe we just raced with a GVA
1227                  * invalidation.
1228                  */
1229                 err = kvm_trap_emul_gva_fault(vcpu, (unsigned long)opc,
1230                                               false);
1231                 if (unlikely(err)) {
1232                         kvm_err("%s: illegal address: %p\n",
1233                                 __func__, opc);
1234                         return -EFAULT;
1235                 }
1236
1237                 /* Hopefully it'll work now */
1238                 goto retry;
1239         }
1240         return 0;
1241 }