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 static int mmu_topup_memory_cache(struct kvm_mmu_memory_cache *cache,
  29                                   int min, int max)
  30 {
  31         void *page;
  32
  33         BUG_ON(max > KVM_NR_MEM_OBJS);
  34         if (cache->nobjs >= min)
  35                 return 0;
  36         while (cache->nobjs < max) {
  37                 page = (void *)__get_free_page(GFP_KERNEL);
  38                 if (!page)
  39                         return -ENOMEM;
  40                 cache->objects[cache->nobjs++] = page;
  41         }
  42         return 0;
  43 }
  44
  45 static void mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc)
  46 {
  47         while (mc->nobjs)
  48                 free_page((unsigned long)mc->objects[--mc->nobjs]);
  49 }
  50
  51 static void *mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc)
  52 {
  53         void *p;
  54
  55         BUG_ON(!mc || !mc->nobjs);
  56         p = mc->objects[--mc->nobjs];
  57         return p;
  58 }
  59
  60 void kvm_mmu_free_memory_caches(struct kvm_vcpu *vcpu)
  61 {
  62         mmu_free_memory_cache(&vcpu->arch.mmu_page_cache);
  63 }
  64
  65 /**
  66  * kvm_pgd_init() - Initialise KVM GPA page directory.
  67  * @page:       Pointer to page directory (PGD) for KVM GPA.
  68  *
  69  * Initialise a KVM GPA page directory with pointers to the invalid table, i.e.
  70  * representing no mappings. This is similar to pgd_init(), however it
  71  * initialises all the page directory pointers, not just the ones corresponding
  72  * to the userland address space (since it is for the guest physical address
  73  * space rather than a virtual address space).
  74  */
  75 static void kvm_pgd_init(void *page)
  76 {
  77         unsigned long *p, *end;
  78         unsigned long entry;
  79
  80 #ifdef __PAGETABLE_PMD_FOLDED
  81         entry = (unsigned long)invalid_pte_table;
  82 #else
  83         entry = (unsigned long)invalid_pmd_table;
  84 #endif
  85
  86         p = (unsigned long *)page;
  87         end = p + PTRS_PER_PGD;
  88
  89         do {
  90                 p[0] = entry;
  91                 p[1] = entry;
  92                 p[2] = entry;
  93                 p[3] = entry;
  94                 p[4] = entry;
  95                 p += 8;
  96                 p[-3] = entry;
  97                 p[-2] = entry;
  98                 p[-1] = entry;
  99         } while (p != end);
 100 }
 101
 102 /**
 103  * kvm_pgd_alloc() - Allocate and initialise a KVM GPA page directory.
 104  *
 105  * Allocate a blank KVM GPA page directory (PGD) for representing guest physical
 106  * to host physical page mappings.
 107  *
 108  * Returns:     Pointer to new KVM GPA page directory.
 109  *              NULL on allocation failure.
 110  */
 111 pgd_t *kvm_pgd_alloc(void)
 112 {
 113         pgd_t *ret;
 114
 115         ret = (pgd_t *)__get_free_pages(GFP_KERNEL, PGD_ORDER);
 116         if (ret)
 117                 kvm_pgd_init(ret);
 118
 119         return ret;
 120 }
 121
 122 /**
 123  * kvm_mips_walk_pgd() - Walk page table with optional allocation.
 124  * @pgd:        Page directory pointer.
 125  * @addr:       Address to index page table using.
 126  * @cache:      MMU page cache to allocate new page tables from, or NULL.
 127  *
 128  * Walk the page tables pointed to by @pgd to find the PTE corresponding to the
 129  * address @addr. If page tables don't exist for @addr, they will be created
 130  * from the MMU cache if @cache is not NULL.
 131  *
 132  * Returns:     Pointer to pte_t corresponding to @addr.
 133  *              NULL if a page table doesn't exist for @addr and !@cache.
 134  *              NULL if a page table allocation failed.
 135  */
 136 static pte_t *kvm_mips_walk_pgd(pgd_t *pgd, struct kvm_mmu_memory_cache *cache,
 137                                 unsigned long addr)
 138 {
 139         pud_t *pud;
 140         pmd_t *pmd;
 141
 142         pgd += pgd_index(addr);
 143         if (pgd_none(*pgd)) {
 144                 /* Not used on MIPS yet */
 145                 BUG();
 146                 return NULL;
 147         }
 148         pud = pud_offset(pgd, addr);
 149         if (pud_none(*pud)) {
 150                 pmd_t *new_pmd;
 151
 152                 if (!cache)
 153                         return NULL;
 154                 new_pmd = mmu_memory_cache_alloc(cache);
 155                 pmd_init((unsigned long)new_pmd,
 156                          (unsigned long)invalid_pte_table);
 157                 pud_populate(NULL, pud, new_pmd);
 158         }
 159         pmd = pmd_offset(pud, addr);
 160         if (pmd_none(*pmd)) {
 161                 pte_t *new_pte;
 162
 163                 if (!cache)
 164                         return NULL;
 165                 new_pte = mmu_memory_cache_alloc(cache);
 166                 clear_page(new_pte);
 167                 pmd_populate_kernel(NULL, pmd, new_pte);
 168         }
 169         return pte_offset(pmd, addr);
 170 }
 171
 172 /* Caller must hold kvm->mm_lock */
 173 static pte_t *kvm_mips_pte_for_gpa(struct kvm *kvm,
 174                                    struct kvm_mmu_memory_cache *cache,
 175                                    unsigned long addr)
 176 {
 177         return kvm_mips_walk_pgd(kvm->arch.gpa_mm.pgd, cache, addr);
 178 }
 179
 180 /*
 181  * kvm_mips_flush_gpa_{pte,pmd,pud,pgd,pt}.
 182  * Flush a range of guest physical address space from the VM's GPA page tables.
 183  */
 184
 185 static bool kvm_mips_flush_gpa_pte(pte_t *pte, unsigned long start_gpa,
 186                                    unsigned long end_gpa)
 187 {
 188         int i_min = __pte_offset(start_gpa);
 189         int i_max = __pte_offset(end_gpa);
 190         bool safe_to_remove = (i_min == 0 && i_max == PTRS_PER_PTE - 1);
 191         int i;
 192
 193         for (i = i_min; i <= i_max; ++i) {
 194                 if (!pte_present(pte[i]))
 195                         continue;
 196
 197                 set_pte(pte + i, __pte(0));
 198         }
 199         return safe_to_remove;
 200 }
 201
 202 static bool kvm_mips_flush_gpa_pmd(pmd_t *pmd, unsigned long start_gpa,
 203                                    unsigned long end_gpa)
 204 {
 205         pte_t *pte;
 206         unsigned long end = ~0ul;
 207         int i_min = __pmd_offset(start_gpa);
 208         int i_max = __pmd_offset(end_gpa);
 209         bool safe_to_remove = (i_min == 0 && i_max == PTRS_PER_PMD - 1);
 210         int i;
 211
 212         for (i = i_min; i <= i_max; ++i, start_gpa = 0) {
 213                 if (!pmd_present(pmd[i]))
 214                         continue;
 215
 216                 pte = pte_offset(pmd + i, 0);
 217                 if (i == i_max)
 218                         end = end_gpa;
 219
 220                 if (kvm_mips_flush_gpa_pte(pte, start_gpa, end)) {
 221                         pmd_clear(pmd + i);
 222                         pte_free_kernel(NULL, pte);
 223                 } else {
 224                         safe_to_remove = false;
 225                 }
 226         }
 227         return safe_to_remove;
 228 }
 229
 230 static bool kvm_mips_flush_gpa_pud(pud_t *pud, unsigned long start_gpa,
 231                                    unsigned long end_gpa)
 232 {
 233         pmd_t *pmd;
 234         unsigned long end = ~0ul;
 235         int i_min = __pud_offset(start_gpa);
 236         int i_max = __pud_offset(end_gpa);
 237         bool safe_to_remove = (i_min == 0 && i_max == PTRS_PER_PUD - 1);
 238         int i;
 239
 240         for (i = i_min; i <= i_max; ++i, start_gpa = 0) {
 241                 if (!pud_present(pud[i]))
 242                         continue;
 243
 244                 pmd = pmd_offset(pud + i, 0);
 245                 if (i == i_max)
 246                         end = end_gpa;
 247
 248                 if (kvm_mips_flush_gpa_pmd(pmd, start_gpa, end)) {
 249                         pud_clear(pud + i);
 250                         pmd_free(NULL, pmd);
 251                 } else {
 252                         safe_to_remove = false;
 253                 }
 254         }
 255         return safe_to_remove;
 256 }
 257
 258 static bool kvm_mips_flush_gpa_pgd(pgd_t *pgd, unsigned long start_gpa,
 259                                    unsigned long end_gpa)
 260 {
 261         pud_t *pud;
 262         unsigned long end = ~0ul;
 263         int i_min = pgd_index(start_gpa);
 264         int i_max = pgd_index(end_gpa);
 265         bool safe_to_remove = (i_min == 0 && i_max == PTRS_PER_PGD - 1);
 266         int i;
 267
 268         for (i = i_min; i <= i_max; ++i, start_gpa = 0) {
 269                 if (!pgd_present(pgd[i]))
 270                         continue;
 271
 272                 pud = pud_offset(pgd + i, 0);
 273                 if (i == i_max)
 274                         end = end_gpa;
 275
 276                 if (kvm_mips_flush_gpa_pud(pud, start_gpa, end)) {
 277                         pgd_clear(pgd + i);
 278                         pud_free(NULL, pud);
 279                 } else {
 280                         safe_to_remove = false;
 281                 }
 282         }
 283         return safe_to_remove;
 284 }
 285
 286 /**
 287  * kvm_mips_flush_gpa_pt() - Flush a range of guest physical addresses.
 288  * @kvm:        KVM pointer.
 289  * @start_gfn:  Guest frame number of first page in GPA range to flush.
 290  * @end_gfn:    Guest frame number of last page in GPA range to flush.
 291  *
 292  * Flushes a range of GPA mappings from the GPA page tables.
 293  *
 294  * The caller must hold the @kvm->mmu_lock spinlock.
 295  *
 296  * Returns:     Whether its safe to remove the top level page directory because
 297  *              all lower levels have been removed.
 298  */
 299 bool kvm_mips_flush_gpa_pt(struct kvm *kvm, gfn_t start_gfn, gfn_t end_gfn)
 300 {
 301         return kvm_mips_flush_gpa_pgd(kvm->arch.gpa_mm.pgd,
 302                                       start_gfn << PAGE_SHIFT,
 303                                       end_gfn << PAGE_SHIFT);
 304 }
 305
 306 #define BUILD_PTE_RANGE_OP(name, op)                                    \
 307 static int kvm_mips_##name##_pte(pte_t *pte, unsigned long start,       \
 308                                  unsigned long end)                     \
 309 {                                                                       \
 310         int ret = 0;                                                    \
 311         int i_min = __pte_offset(start);                                \
 312         int i_max = __pte_offset(end);                                  \
 313         int i;                                                          \
 314         pte_t old, new;                                                 \
 315                                                                         \
 316         for (i = i_min; i <= i_max; ++i) {                              \
 317                 if (!pte_present(pte[i]))                               \
 318                         continue;                                       \
 319                                                                         \
 320                 old = pte[i];                                           \
 321                 new = op(old);                                          \
 322                 if (pte_val(new) == pte_val(old))                       \
 323                         continue;                                       \
 324                 set_pte(pte + i, new);                                  \
 325                 ret = 1;                                                \
 326         }                                                               \
 327         return ret;                                                     \
 328 }                                                                       \
 329                                                                         \
 330 /* returns true if anything was done */                                 \
 331 static int kvm_mips_##name##_pmd(pmd_t *pmd, unsigned long start,       \
 332                                  unsigned long end)                     \
 333 {                                                                       \
 334         int ret = 0;                                                    \
 335         pte_t *pte;                                                     \
 336         unsigned long cur_end = ~0ul;                                   \
 337         int i_min = __pmd_offset(start);                                \
 338         int i_max = __pmd_offset(end);                                  \
 339         int i;                                                          \
 340                                                                         \
 341         for (i = i_min; i <= i_max; ++i, start = 0) {                   \
 342                 if (!pmd_present(pmd[i]))                               \
 343                         continue;                                       \
 344                                                                         \
 345                 pte = pte_offset(pmd + i, 0);                           \
 346                 if (i == i_max)                                         \
 347                         cur_end = end;                                  \
 348                                                                         \
 349                 ret |= kvm_mips_##name##_pte(pte, start, cur_end);      \
 350         }                                                               \
 351         return ret;                                                     \
 352 }                                                                       \
 353                                                                         \
 354 static int kvm_mips_##name##_pud(pud_t *pud, unsigned long start,       \
 355                                  unsigned long end)                     \
 356 {                                                                       \
 357         int ret = 0;                                                    \
 358         pmd_t *pmd;                                                     \
 359         unsigned long cur_end = ~0ul;                                   \
 360         int i_min = __pud_offset(start);                                \
 361         int i_max = __pud_offset(end);                                  \
 362         int i;                                                          \
 363                                                                         \
 364         for (i = i_min; i <= i_max; ++i, start = 0) {                   \
 365                 if (!pud_present(pud[i]))                               \
 366                         continue;                                       \
 367                                                                         \
 368                 pmd = pmd_offset(pud + i, 0);                           \
 369                 if (i == i_max)                                         \
 370                         cur_end = end;                                  \
 371                                                                         \
 372                 ret |= kvm_mips_##name##_pmd(pmd, start, cur_end);      \
 373         }                                                               \
 374         return ret;                                                     \
 375 }                                                                       \
 376                                                                         \
 377 static int kvm_mips_##name##_pgd(pgd_t *pgd, unsigned long start,       \
 378                                  unsigned long end)                     \
 379 {                                                                       \
 380         int ret = 0;                                                    \
 381         pud_t *pud;                                                     \
 382         unsigned long cur_end = ~0ul;                                   \
 383         int i_min = pgd_index(start);                                   \
 384         int i_max = pgd_index(end);                                     \
 385         int i;                                                          \
 386                                                                         \
 387         for (i = i_min; i <= i_max; ++i, start = 0) {                   \
 388                 if (!pgd_present(pgd[i]))                               \
 389                         continue;                                       \
 390                                                                         \
 391                 pud = pud_offset(pgd + i, 0);                           \
 392                 if (i == i_max)                                         \
 393                         cur_end = end;                                  \
 394                                                                         \
 395                 ret |= kvm_mips_##name##_pud(pud, start, cur_end);      \
 396         }                                                               \
 397         return ret;                                                     \
 398 }
 399
 400 /*
 401  * kvm_mips_mkclean_gpa_pt.
 402  * Mark a range of guest physical address space clean (writes fault) in the VM's
 403  * GPA page table to allow dirty page tracking.
 404  */
 405
 406 BUILD_PTE_RANGE_OP(mkclean, pte_mkclean)
 407
 408 /**
 409  * kvm_mips_mkclean_gpa_pt() - Make a range of guest physical addresses clean.
 410  * @kvm:        KVM pointer.
 411  * @start_gfn:  Guest frame number of first page in GPA range to flush.
 412  * @end_gfn:    Guest frame number of last page in GPA range to flush.
 413  *
 414  * Make a range of GPA mappings clean so that guest writes will fault and
 415  * trigger dirty page logging.
 416  *
 417  * The caller must hold the @kvm->mmu_lock spinlock.
 418  *
 419  * Returns:     Whether any GPA mappings were modified, which would require
 420  *              derived mappings (GVA page tables & TLB enties) to be
 421  *              invalidated.
 422  */
 423 int kvm_mips_mkclean_gpa_pt(struct kvm *kvm, gfn_t start_gfn, gfn_t end_gfn)
 424 {
 425         return kvm_mips_mkclean_pgd(kvm->arch.gpa_mm.pgd,
 426                                     start_gfn << PAGE_SHIFT,
 427                                     end_gfn << PAGE_SHIFT);
 428 }
 429
 430 /**
 431  * kvm_arch_mmu_enable_log_dirty_pt_masked() - write protect dirty pages
 432  * @kvm:        The KVM pointer
 433  * @slot:       The memory slot associated with mask
 434  * @gfn_offset: The gfn offset in memory slot
 435  * @mask:       The mask of dirty pages at offset 'gfn_offset' in this memory
 436  *              slot to be write protected
 437  *
 438  * Walks bits set in mask write protects the associated pte's. Caller must
 439  * acquire @kvm->mmu_lock.
 440  */
 441 void kvm_arch_mmu_enable_log_dirty_pt_masked(struct kvm *kvm,
 442                 struct kvm_memory_slot *slot,
 443                 gfn_t gfn_offset, unsigned long mask)
 444 {
 445         gfn_t base_gfn = slot->base_gfn + gfn_offset;
 446         gfn_t start = base_gfn +  __ffs(mask);
 447         gfn_t end = base_gfn + __fls(mask);
 448
 449         kvm_mips_mkclean_gpa_pt(kvm, start, end);
 450 }
 451
 452 /*
 453  * kvm_mips_mkold_gpa_pt.
 454  * Mark a range of guest physical address space old (all accesses fault) in the
 455  * VM's GPA page table to allow detection of commonly used pages.
 456  */
 457
 458 BUILD_PTE_RANGE_OP(mkold, pte_mkold)
 459
 460 static int kvm_mips_mkold_gpa_pt(struct kvm *kvm, gfn_t start_gfn,
 461                                  gfn_t end_gfn)
 462 {
 463         return kvm_mips_mkold_pgd(kvm->arch.gpa_mm.pgd,
 464                                   start_gfn << PAGE_SHIFT,
 465                                   end_gfn << PAGE_SHIFT);
 466 }
 467
 468 static int handle_hva_to_gpa(struct kvm *kvm,
 469                              unsigned long start,
 470                              unsigned long end,
 471                              int (*handler)(struct kvm *kvm, gfn_t gfn,
 472                                             gpa_t gfn_end,
 473                                             struct kvm_memory_slot *memslot,
 474                                             void *data),
 475                              void *data)
 476 {
 477         struct kvm_memslots *slots;
 478         struct kvm_memory_slot *memslot;
 479         int ret = 0;
 480
 481         slots = kvm_memslots(kvm);
 482
 483         /* we only care about the pages that the guest sees */
 484         kvm_for_each_memslot(memslot, slots) {
 485                 unsigned long hva_start, hva_end;
 486                 gfn_t gfn, gfn_end;
 487
 488                 hva_start = max(start, memslot->userspace_addr);
 489                 hva_end = min(end, memslot->userspace_addr +
 490                                         (memslot->npages << PAGE_SHIFT));
 491                 if (hva_start >= hva_end)
 492                         continue;
 493
 494                 /*
 495                  * {gfn(page) | page intersects with [hva_start, hva_end)} =
 496                  * {gfn_start, gfn_start+1, ..., gfn_end-1}.
 497                  */
 498                 gfn = hva_to_gfn_memslot(hva_start, memslot);
 499                 gfn_end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, memslot);
 500
 501                 ret |= handler(kvm, gfn, gfn_end, memslot, data);
 502         }
 503
 504         return ret;
 505 }
 506
 507
 508 static int kvm_unmap_hva_handler(struct kvm *kvm, gfn_t gfn, gfn_t gfn_end,
 509                                  struct kvm_memory_slot *memslot, void *data)
 510 {
 511         kvm_mips_flush_gpa_pt(kvm, gfn, gfn_end);
 512         return 1;
 513 }
 514
 515 int kvm_unmap_hva_range(struct kvm *kvm, unsigned long start, unsigned long end)
 516 {
 517         handle_hva_to_gpa(kvm, start, end, &kvm_unmap_hva_handler, NULL);
 518
 519         kvm_mips_callbacks->flush_shadow_all(kvm);
 520         return 0;
 521 }
 522
 523 static int kvm_set_spte_handler(struct kvm *kvm, gfn_t gfn, gfn_t gfn_end,
 524                                 struct kvm_memory_slot *memslot, void *data)
 525 {
 526         gpa_t gpa = gfn << PAGE_SHIFT;
 527         pte_t hva_pte = *(pte_t *)data;
 528         pte_t *gpa_pte = kvm_mips_pte_for_gpa(kvm, NULL, gpa);
 529         pte_t old_pte;
 530
 531         if (!gpa_pte)
 532                 return 0;
 533
 534         /* Mapping may need adjusting depending on memslot flags */
 535         old_pte = *gpa_pte;
 536         if (memslot->flags & KVM_MEM_LOG_DIRTY_PAGES && !pte_dirty(old_pte))
 537                 hva_pte = pte_mkclean(hva_pte);
 538         else if (memslot->flags & KVM_MEM_READONLY)
 539                 hva_pte = pte_wrprotect(hva_pte);
 540
 541         set_pte(gpa_pte, hva_pte);
 542
 543         /* Replacing an absent or old page doesn't need flushes */
 544         if (!pte_present(old_pte) || !pte_young(old_pte))
 545                 return 0;
 546
 547         /* Pages swapped, aged, moved, or cleaned require flushes */
 548         return !pte_present(hva_pte) ||
 549                !pte_young(hva_pte) ||
 550                pte_pfn(old_pte) != pte_pfn(hva_pte) ||
 551                (pte_dirty(old_pte) && !pte_dirty(hva_pte));
 552 }
 553
 554 int kvm_set_spte_hva(struct kvm *kvm, unsigned long hva, pte_t pte)
 555 {
 556         unsigned long end = hva + PAGE_SIZE;
 557         int ret;
 558
 559         ret = handle_hva_to_gpa(kvm, hva, end, &kvm_set_spte_handler, &pte);
 560         if (ret)
 561                 kvm_mips_callbacks->flush_shadow_all(kvm);
 562         return 0;
 563 }
 564
 565 static int kvm_age_hva_handler(struct kvm *kvm, gfn_t gfn, gfn_t gfn_end,
 566                                struct kvm_memory_slot *memslot, void *data)
 567 {
 568         return kvm_mips_mkold_gpa_pt(kvm, gfn, gfn_end);
 569 }
 570
 571 static int kvm_test_age_hva_handler(struct kvm *kvm, gfn_t gfn, gfn_t gfn_end,
 572                                     struct kvm_memory_slot *memslot, void *data)
 573 {
 574         gpa_t gpa = gfn << PAGE_SHIFT;
 575         pte_t *gpa_pte = kvm_mips_pte_for_gpa(kvm, NULL, gpa);
 576
 577         if (!gpa_pte)
 578                 return 0;
 579         return pte_young(*gpa_pte);
 580 }
 581
 582 int kvm_age_hva(struct kvm *kvm, unsigned long start, unsigned long end)
 583 {
 584         return handle_hva_to_gpa(kvm, start, end, kvm_age_hva_handler, NULL);
 585 }
 586
 587 int kvm_test_age_hva(struct kvm *kvm, unsigned long hva)
 588 {
 589         return handle_hva_to_gpa(kvm, hva, hva, kvm_test_age_hva_handler, NULL);
 590 }
 591
 592 /**
 593  * _kvm_mips_map_page_fast() - Fast path GPA fault handler.
 594  * @vcpu:               VCPU pointer.
 595  * @gpa:                Guest physical address of fault.
 596  * @write_fault:        Whether the fault was due to a write.
 597  * @out_entry:          New PTE for @gpa (written on success unless NULL).
 598  * @out_buddy:          New PTE for @gpa's buddy (written on success unless
 599  *                      NULL).
 600  *
 601  * Perform fast path GPA fault handling, doing all that can be done without
 602  * calling into KVM. This handles marking old pages young (for idle page
 603  * tracking), and dirtying of clean pages (for dirty page logging).
 604  *
 605  * Returns:     0 on success, in which case we can update derived mappings and
 606  *              resume guest execution.
 607  *              -EFAULT on failure due to absent GPA mapping or write to
 608  *              read-only page, in which case KVM must be consulted.
 609  */
 610 static int _kvm_mips_map_page_fast(struct kvm_vcpu *vcpu, unsigned long gpa,
 611                                    bool write_fault,
 612                                    pte_t *out_entry, pte_t *out_buddy)
 613 {
 614         struct kvm *kvm = vcpu->kvm;
 615         gfn_t gfn = gpa >> PAGE_SHIFT;
 616         pte_t *ptep;
 617         kvm_pfn_t pfn = 0;      /* silence bogus GCC warning */
 618         bool pfn_valid = false;
 619         int ret = 0;
 620
 621         spin_lock(&kvm->mmu_lock);
 622
 623         /* Fast path - just check GPA page table for an existing entry */
 624         ptep = kvm_mips_pte_for_gpa(kvm, NULL, gpa);
 625         if (!ptep || !pte_present(*ptep)) {
 626                 ret = -EFAULT;
 627                 goto out;
 628         }
 629
 630         /* Track access to pages marked old */
 631         if (!pte_young(*ptep)) {
 632                 set_pte(ptep, pte_mkyoung(*ptep));
 633                 pfn = pte_pfn(*ptep);
 634                 pfn_valid = true;
 635                 /* call kvm_set_pfn_accessed() after unlock */
 636         }
 637         if (write_fault && !pte_dirty(*ptep)) {
 638                 if (!pte_write(*ptep)) {
 639                         ret = -EFAULT;
 640                         goto out;
 641                 }
 642
 643                 /* Track dirtying of writeable pages */
 644                 set_pte(ptep, pte_mkdirty(*ptep));
 645                 pfn = pte_pfn(*ptep);
 646                 mark_page_dirty(kvm, gfn);
 647                 kvm_set_pfn_dirty(pfn);
 648         }
 649
 650         if (out_entry)
 651                 *out_entry = *ptep;
 652         if (out_buddy)
 653                 *out_buddy = *ptep_buddy(ptep);
 654
 655 out:
 656         spin_unlock(&kvm->mmu_lock);
 657         if (pfn_valid)
 658                 kvm_set_pfn_accessed(pfn);
 659         return ret;
 660 }
 661
 662 /**
 663  * kvm_mips_map_page() - Map a guest physical page.
 664  * @vcpu:               VCPU pointer.
 665  * @gpa:                Guest physical address of fault.
 666  * @write_fault:        Whether the fault was due to a write.
 667  * @out_entry:          New PTE for @gpa (written on success unless NULL).
 668  * @out_buddy:          New PTE for @gpa's buddy (written on success unless
 669  *                      NULL).
 670  *
 671  * Handle GPA faults by creating a new GPA mapping (or updating an existing
 672  * one).
 673  *
 674  * This takes care of marking pages young or dirty (idle/dirty page tracking),
 675  * asking KVM for the corresponding PFN, and creating a mapping in the GPA page
 676  * tables. Derived mappings (GVA page tables and TLBs) must be handled by the
 677  * caller.
 678  *
 679  * Returns:     0 on success, in which case the caller may use the @out_entry
 680  *              and @out_buddy PTEs to update derived mappings and resume guest
 681  *              execution.
 682  *              -EFAULT if there is no memory region at @gpa or a write was
 683  *              attempted to a read-only memory region. This is usually handled
 684  *              as an MMIO access.
 685  */
 686 static int kvm_mips_map_page(struct kvm_vcpu *vcpu, unsigned long gpa,
 687                              bool write_fault,
 688                              pte_t *out_entry, pte_t *out_buddy)
 689 {
 690         struct kvm *kvm = vcpu->kvm;
 691         struct kvm_mmu_memory_cache *memcache = &vcpu->arch.mmu_page_cache;
 692         gfn_t gfn = gpa >> PAGE_SHIFT;
 693         int srcu_idx, err;
 694         kvm_pfn_t pfn;
 695         pte_t *ptep, entry, old_pte;
 696         bool writeable;
 697         unsigned long prot_bits;
 698         unsigned long mmu_seq;
 699
 700         /* Try the fast path to handle old / clean pages */
 701         srcu_idx = srcu_read_lock(&kvm->srcu);
 702         err = _kvm_mips_map_page_fast(vcpu, gpa, write_fault, out_entry,
 703                                       out_buddy);
 704         if (!err)
 705                 goto out;
 706
 707         /* We need a minimum of cached pages ready for page table creation */
 708         err = mmu_topup_memory_cache(memcache, KVM_MMU_CACHE_MIN_PAGES,
 709                                      KVM_NR_MEM_OBJS);
 710         if (err)
 711                 goto out;
 712
 713 retry:
 714         /*
 715          * Used to check for invalidations in progress, of the pfn that is
 716          * returned by pfn_to_pfn_prot below.
 717          */
 718         mmu_seq = kvm->mmu_notifier_seq;
 719         /*
 720          * Ensure the read of mmu_notifier_seq isn't reordered with PTE reads in
 721          * gfn_to_pfn_prot() (which calls get_user_pages()), so that we don't
 722          * risk the page we get a reference to getting unmapped before we have a
 723          * chance to grab the mmu_lock without mmu_notifier_retry() noticing.
 724          *
 725          * This smp_rmb() pairs with the effective smp_wmb() of the combination
 726          * of the pte_unmap_unlock() after the PTE is zapped, and the
 727          * spin_lock() in kvm_mmu_notifier_invalidate_<page|range_end>() before
 728          * mmu_notifier_seq is incremented.
 729          */
 730         smp_rmb();
 731
 732         /* Slow path - ask KVM core whether we can access this GPA */
 733         pfn = gfn_to_pfn_prot(kvm, gfn, write_fault, &writeable);
 734         if (is_error_noslot_pfn(pfn)) {
 735                 err = -EFAULT;
 736                 goto out;
 737         }
 738
 739         spin_lock(&kvm->mmu_lock);
 740         /* Check if an invalidation has taken place since we got pfn */
 741         if (mmu_notifier_retry(kvm, mmu_seq)) {
 742                 /*
 743                  * This can happen when mappings are changed asynchronously, but
 744                  * also synchronously if a COW is triggered by
 745                  * gfn_to_pfn_prot().
 746                  */
 747                 spin_unlock(&kvm->mmu_lock);
 748                 kvm_release_pfn_clean(pfn);
 749                 goto retry;
 750         }
 751
 752         /* Ensure page tables are allocated */
 753         ptep = kvm_mips_pte_for_gpa(kvm, memcache, gpa);
 754
 755         /* Set up the PTE */
 756         prot_bits = _PAGE_PRESENT | __READABLE | _page_cachable_default;
 757         if (writeable) {
 758                 prot_bits |= _PAGE_WRITE;
 759                 if (write_fault) {
 760                         prot_bits |= __WRITEABLE;
 761                         mark_page_dirty(kvm, gfn);
 762                         kvm_set_pfn_dirty(pfn);
 763                 }
 764         }
 765         entry = pfn_pte(pfn, __pgprot(prot_bits));
 766
 767         /* Write the PTE */
 768         old_pte = *ptep;
 769         set_pte(ptep, entry);
 770
 771         err = 0;
 772         if (out_entry)
 773                 *out_entry = *ptep;
 774         if (out_buddy)
 775                 *out_buddy = *ptep_buddy(ptep);
 776
 777         spin_unlock(&kvm->mmu_lock);
 778         kvm_release_pfn_clean(pfn);
 779         kvm_set_pfn_accessed(pfn);
 780 out:
 781         srcu_read_unlock(&kvm->srcu, srcu_idx);
 782         return err;
 783 }
 784
 785 static pte_t *kvm_trap_emul_pte_for_gva(struct kvm_vcpu *vcpu,
 786                                         unsigned long addr)
 787 {
 788         struct kvm_mmu_memory_cache *memcache = &vcpu->arch.mmu_page_cache;
 789         pgd_t *pgdp;
 790         int ret;
 791
 792         /* We need a minimum of cached pages ready for page table creation */
 793         ret = mmu_topup_memory_cache(memcache, KVM_MMU_CACHE_MIN_PAGES,
 794                                      KVM_NR_MEM_OBJS);
 795         if (ret)
 796                 return NULL;
 797
 798         if (KVM_GUEST_KERNEL_MODE(vcpu))
 799                 pgdp = vcpu->arch.guest_kernel_mm.pgd;
 800         else
 801                 pgdp = vcpu->arch.guest_user_mm.pgd;
 802
 803         return kvm_mips_walk_pgd(pgdp, memcache, addr);
 804 }
 805
 806 void kvm_trap_emul_invalidate_gva(struct kvm_vcpu *vcpu, unsigned long addr,
 807                                   bool user)
 808 {
 809         pgd_t *pgdp;
 810         pte_t *ptep;
 811
 812         addr &= PAGE_MASK << 1;
 813
 814         pgdp = vcpu->arch.guest_kernel_mm.pgd;
 815         ptep = kvm_mips_walk_pgd(pgdp, NULL, addr);
 816         if (ptep) {
 817                 ptep[0] = pfn_pte(0, __pgprot(0));
 818                 ptep[1] = pfn_pte(0, __pgprot(0));
 819         }
 820
 821         if (user) {
 822                 pgdp = vcpu->arch.guest_user_mm.pgd;
 823                 ptep = kvm_mips_walk_pgd(pgdp, NULL, addr);
 824                 if (ptep) {
 825                         ptep[0] = pfn_pte(0, __pgprot(0));
 826                         ptep[1] = pfn_pte(0, __pgprot(0));
 827                 }
 828         }
 829 }
 830
 831 /*
 832  * kvm_mips_flush_gva_{pte,pmd,pud,pgd,pt}.
 833  * Flush a range of guest physical address space from the VM's GPA page tables.
 834  */
 835
 836 static bool kvm_mips_flush_gva_pte(pte_t *pte, unsigned long start_gva,
 837                                    unsigned long end_gva)
 838 {
 839         int i_min = __pte_offset(start_gva);
 840         int i_max = __pte_offset(end_gva);
 841         bool safe_to_remove = (i_min == 0 && i_max == PTRS_PER_PTE - 1);
 842         int i;
 843
 844         /*
 845          * There's no freeing to do, so there's no point clearing individual
 846          * entries unless only part of the last level page table needs flushing.
 847          */
 848         if (safe_to_remove)
 849                 return true;
 850
 851         for (i = i_min; i <= i_max; ++i) {
 852                 if (!pte_present(pte[i]))
 853                         continue;
 854
 855                 set_pte(pte + i, __pte(0));
 856         }
 857         return false;
 858 }
 859
 860 static bool kvm_mips_flush_gva_pmd(pmd_t *pmd, unsigned long start_gva,
 861                                    unsigned long end_gva)
 862 {
 863         pte_t *pte;
 864         unsigned long end = ~0ul;
 865         int i_min = __pmd_offset(start_gva);
 866         int i_max = __pmd_offset(end_gva);
 867         bool safe_to_remove = (i_min == 0 && i_max == PTRS_PER_PMD - 1);
 868         int i;
 869
 870         for (i = i_min; i <= i_max; ++i, start_gva = 0) {
 871                 if (!pmd_present(pmd[i]))
 872                         continue;
 873
 874                 pte = pte_offset(pmd + i, 0);
 875                 if (i == i_max)
 876                         end = end_gva;
 877
 878                 if (kvm_mips_flush_gva_pte(pte, start_gva, end)) {
 879                         pmd_clear(pmd + i);
 880                         pte_free_kernel(NULL, pte);
 881                 } else {
 882                         safe_to_remove = false;
 883                 }
 884         }
 885         return safe_to_remove;
 886 }
 887
 888 static bool kvm_mips_flush_gva_pud(pud_t *pud, unsigned long start_gva,
 889                                    unsigned long end_gva)
 890 {
 891         pmd_t *pmd;
 892         unsigned long end = ~0ul;
 893         int i_min = __pud_offset(start_gva);
 894         int i_max = __pud_offset(end_gva);
 895         bool safe_to_remove = (i_min == 0 && i_max == PTRS_PER_PUD - 1);
 896         int i;
 897
 898         for (i = i_min; i <= i_max; ++i, start_gva = 0) {
 899                 if (!pud_present(pud[i]))
 900                         continue;
 901
 902                 pmd = pmd_offset(pud + i, 0);
 903                 if (i == i_max)
 904                         end = end_gva;
 905
 906                 if (kvm_mips_flush_gva_pmd(pmd, start_gva, end)) {
 907                         pud_clear(pud + i);
 908                         pmd_free(NULL, pmd);
 909                 } else {
 910                         safe_to_remove = false;
 911                 }
 912         }
 913         return safe_to_remove;
 914 }
 915
 916 static bool kvm_mips_flush_gva_pgd(pgd_t *pgd, unsigned long start_gva,
 917                                    unsigned long end_gva)
 918 {
 919         pud_t *pud;
 920         unsigned long end = ~0ul;
 921         int i_min = pgd_index(start_gva);
 922         int i_max = pgd_index(end_gva);
 923         bool safe_to_remove = (i_min == 0 && i_max == PTRS_PER_PGD - 1);
 924         int i;
 925
 926         for (i = i_min; i <= i_max; ++i, start_gva = 0) {
 927                 if (!pgd_present(pgd[i]))
 928                         continue;
 929
 930                 pud = pud_offset(pgd + i, 0);
 931                 if (i == i_max)
 932                         end = end_gva;
 933
 934                 if (kvm_mips_flush_gva_pud(pud, start_gva, end)) {
 935                         pgd_clear(pgd + i);
 936                         pud_free(NULL, pud);
 937                 } else {
 938                         safe_to_remove = false;
 939                 }
 940         }
 941         return safe_to_remove;
 942 }
 943
 944 void kvm_mips_flush_gva_pt(pgd_t *pgd, enum kvm_mips_flush flags)
 945 {
 946         if (flags & KMF_GPA) {
 947                 /* all of guest virtual address space could be affected */
 948                 if (flags & KMF_KERN)
 949                         /* useg, kseg0, seg2/3 */
 950                         kvm_mips_flush_gva_pgd(pgd, 0, 0x7fffffff);
 951                 else
 952                         /* useg */
 953                         kvm_mips_flush_gva_pgd(pgd, 0, 0x3fffffff);
 954         } else {
 955                 /* useg */
 956                 kvm_mips_flush_gva_pgd(pgd, 0, 0x3fffffff);
 957
 958                 /* kseg2/3 */
 959                 if (flags & KMF_KERN)
 960                         kvm_mips_flush_gva_pgd(pgd, 0x60000000, 0x7fffffff);
 961         }
 962 }
 963
 964 static pte_t kvm_mips_gpa_pte_to_gva_unmapped(pte_t pte)
 965 {
 966         /*
 967          * Don't leak writeable but clean entries from GPA page tables. We don't
 968          * want the normal Linux tlbmod handler to handle dirtying when KVM
 969          * accesses guest memory.
 970          */
 971         if (!pte_dirty(pte))
 972                 pte = pte_wrprotect(pte);
 973
 974         return pte;
 975 }
 976
 977 static pte_t kvm_mips_gpa_pte_to_gva_mapped(pte_t pte, long entrylo)
 978 {
 979         /* Guest EntryLo overrides host EntryLo */
 980         if (!(entrylo & ENTRYLO_D))
 981                 pte = pte_mkclean(pte);
 982
 983         return kvm_mips_gpa_pte_to_gva_unmapped(pte);
 984 }
 985
 986 #ifdef CONFIG_KVM_MIPS_VZ
 987 int kvm_mips_handle_vz_root_tlb_fault(unsigned long badvaddr,
 988                                       struct kvm_vcpu *vcpu,
 989                                       bool write_fault)
 990 {
 991         int ret;
 992
 993         ret = kvm_mips_map_page(vcpu, badvaddr, write_fault, NULL, NULL);
 994         if (ret)
 995                 return ret;
 996
 997         /* Invalidate this entry in the TLB */
 998         return kvm_vz_host_tlb_inv(vcpu, badvaddr);
 999 }
1000 #endif
1001
1002 /* XXXKYMA: Must be called with interrupts disabled */
1003 int kvm_mips_handle_kseg0_tlb_fault(unsigned long badvaddr,
1004                                     struct kvm_vcpu *vcpu,
1005                                     bool write_fault)
1006 {
1007         unsigned long gpa;
1008         pte_t pte_gpa[2], *ptep_gva;
1009         int idx;
1010
1011         if (KVM_GUEST_KSEGX(badvaddr) != KVM_GUEST_KSEG0) {
1012                 kvm_err("%s: Invalid BadVaddr: %#lx\n", __func__, badvaddr);
1013                 kvm_mips_dump_host_tlbs();
1014                 return -1;
1015         }
1016
1017         /* Get the GPA page table entry */
1018         gpa = KVM_GUEST_CPHYSADDR(badvaddr);
1019         idx = (badvaddr >> PAGE_SHIFT) & 1;
1020         if (kvm_mips_map_page(vcpu, gpa, write_fault, &pte_gpa[idx],
1021                               &pte_gpa[!idx]) < 0)
1022                 return -1;
1023
1024         /* Get the GVA page table entry */
1025         ptep_gva = kvm_trap_emul_pte_for_gva(vcpu, badvaddr & ~PAGE_SIZE);
1026         if (!ptep_gva) {
1027                 kvm_err("No ptep for gva %lx\n", badvaddr);
1028                 return -1;
1029         }
1030
1031         /* Copy a pair of entries from GPA page table to GVA page table */
1032         ptep_gva[0] = kvm_mips_gpa_pte_to_gva_unmapped(pte_gpa[0]);
1033         ptep_gva[1] = kvm_mips_gpa_pte_to_gva_unmapped(pte_gpa[1]);
1034
1035         /* Invalidate this entry in the TLB, guest kernel ASID only */
1036         kvm_mips_host_tlb_inv(vcpu, badvaddr, false, true);
1037         return 0;
1038 }
1039
1040 int kvm_mips_handle_mapped_seg_tlb_fault(struct kvm_vcpu *vcpu,
1041                                          struct kvm_mips_tlb *tlb,
1042                                          unsigned long gva,
1043                                          bool write_fault)
1044 {
1045         struct kvm *kvm = vcpu->kvm;
1046         long tlb_lo[2];
1047         pte_t pte_gpa[2], *ptep_buddy, *ptep_gva;
1048         unsigned int idx = TLB_LO_IDX(*tlb, gva);
1049         bool kernel = KVM_GUEST_KERNEL_MODE(vcpu);
1050
1051         tlb_lo[0] = tlb->tlb_lo[0];
1052         tlb_lo[1] = tlb->tlb_lo[1];
1053
1054         /*
1055          * The commpage address must not be mapped to anything else if the guest
1056          * TLB contains entries nearby, or commpage accesses will break.
1057          */
1058         if (!((gva ^ KVM_GUEST_COMMPAGE_ADDR) & VPN2_MASK & (PAGE_MASK << 1)))
1059                 tlb_lo[TLB_LO_IDX(*tlb, KVM_GUEST_COMMPAGE_ADDR)] = 0;
1060
1061         /* Get the GPA page table entry */
1062         if (kvm_mips_map_page(vcpu, mips3_tlbpfn_to_paddr(tlb_lo[idx]),
1063                               write_fault, &pte_gpa[idx], NULL) < 0)
1064                 return -1;
1065
1066         /* And its GVA buddy's GPA page table entry if it also exists */
1067         pte_gpa[!idx] = pfn_pte(0, __pgprot(0));
1068         if (tlb_lo[!idx] & ENTRYLO_V) {
1069                 spin_lock(&kvm->mmu_lock);
1070                 ptep_buddy = kvm_mips_pte_for_gpa(kvm, NULL,
1071                                         mips3_tlbpfn_to_paddr(tlb_lo[!idx]));
1072                 if (ptep_buddy)
1073                         pte_gpa[!idx] = *ptep_buddy;
1074                 spin_unlock(&kvm->mmu_lock);
1075         }
1076
1077         /* Get the GVA page table entry pair */
1078         ptep_gva = kvm_trap_emul_pte_for_gva(vcpu, gva & ~PAGE_SIZE);
1079         if (!ptep_gva) {
1080                 kvm_err("No ptep for gva %lx\n", gva);
1081                 return -1;
1082         }
1083
1084         /* Copy a pair of entries from GPA page table to GVA page table */
1085         ptep_gva[0] = kvm_mips_gpa_pte_to_gva_mapped(pte_gpa[0], tlb_lo[0]);
1086         ptep_gva[1] = kvm_mips_gpa_pte_to_gva_mapped(pte_gpa[1], tlb_lo[1]);
1087
1088         /* Invalidate this entry in the TLB, current guest mode ASID only */
1089         kvm_mips_host_tlb_inv(vcpu, gva, !kernel, kernel);
1090
1091         kvm_debug("@ %#lx tlb_lo0: 0x%08lx tlb_lo1: 0x%08lx\n", vcpu->arch.pc,
1092                   tlb->tlb_lo[0], tlb->tlb_lo[1]);
1093
1094         return 0;
1095 }
1096
1097 int kvm_mips_handle_commpage_tlb_fault(unsigned long badvaddr,
1098                                        struct kvm_vcpu *vcpu)
1099 {
1100         kvm_pfn_t pfn;
1101         pte_t *ptep;
1102
1103         ptep = kvm_trap_emul_pte_for_gva(vcpu, badvaddr);
1104         if (!ptep) {
1105                 kvm_err("No ptep for commpage %lx\n", badvaddr);
1106                 return -1;
1107         }
1108
1109         pfn = PFN_DOWN(virt_to_phys(vcpu->arch.kseg0_commpage));
1110         /* Also set valid and dirty, so refill handler doesn't have to */
1111         *ptep = pte_mkyoung(pte_mkdirty(pfn_pte(pfn, PAGE_SHARED)));
1112
1113         /* Invalidate this entry in the TLB, guest kernel ASID only */
1114         kvm_mips_host_tlb_inv(vcpu, badvaddr, false, true);
1115         return 0;
1116 }
1117
1118 /**
1119  * kvm_mips_migrate_count() - Migrate timer.
1120  * @vcpu:       Virtual CPU.
1121  *
1122  * Migrate CP0_Count hrtimer to the current CPU by cancelling and restarting it
1123  * if it was running prior to being cancelled.
1124  *
1125  * Must be called when the VCPU is migrated to a different CPU to ensure that
1126  * timer expiry during guest execution interrupts the guest and causes the
1127  * interrupt to be delivered in a timely manner.
1128  */
1129 static void kvm_mips_migrate_count(struct kvm_vcpu *vcpu)
1130 {
1131         if (hrtimer_cancel(&vcpu->arch.comparecount_timer))
1132                 hrtimer_restart(&vcpu->arch.comparecount_timer);
1133 }
1134
1135 /* Restore ASID once we are scheduled back after preemption */
1136 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1137 {
1138         unsigned long flags;
1139
1140         kvm_debug("%s: vcpu %p, cpu: %d\n", __func__, vcpu, cpu);
1141
1142         local_irq_save(flags);
1143
1144         vcpu->cpu = cpu;
1145         if (vcpu->arch.last_sched_cpu != cpu) {
1146                 kvm_debug("[%d->%d]KVM VCPU[%d] switch\n",
1147                           vcpu->arch.last_sched_cpu, cpu, vcpu->vcpu_id);
1148                 /*
1149                  * Migrate the timer interrupt to the current CPU so that it
1150                  * always interrupts the guest and synchronously triggers a
1151                  * guest timer interrupt.
1152                  */
1153                 kvm_mips_migrate_count(vcpu);
1154         }
1155
1156         /* restore guest state to registers */
1157         kvm_mips_callbacks->vcpu_load(vcpu, cpu);
1158
1159         local_irq_restore(flags);
1160 }
1161
1162 /* ASID can change if another task is scheduled during preemption */
1163 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1164 {
1165         unsigned long flags;
1166         int cpu;
1167
1168         local_irq_save(flags);
1169
1170         cpu = smp_processor_id();
1171         vcpu->arch.last_sched_cpu = cpu;
1172         vcpu->cpu = -1;
1173
1174         /* save guest state in registers */
1175         kvm_mips_callbacks->vcpu_put(vcpu, cpu);
1176
1177         local_irq_restore(flags);
1178 }
1179
1180 /**
1181  * kvm_trap_emul_gva_fault() - Safely attempt to handle a GVA access fault.
1182  * @vcpu:       Virtual CPU.
1183  * @gva:        Guest virtual address to be accessed.
1184  * @write:      True if write attempted (must be dirtied and made writable).
1185  *
1186  * Safely attempt to handle a GVA fault, mapping GVA pages if necessary, and
1187  * dirtying the page if @write so that guest instructions can be modified.
1188  *
1189  * Returns:     KVM_MIPS_MAPPED on success.
1190  *              KVM_MIPS_GVA if bad guest virtual address.
1191  *              KVM_MIPS_GPA if bad guest physical address.
1192  *              KVM_MIPS_TLB if guest TLB not present.
1193  *              KVM_MIPS_TLBINV if guest TLB present but not valid.
1194  *              KVM_MIPS_TLBMOD if guest TLB read only.
1195  */
1196 enum kvm_mips_fault_result kvm_trap_emul_gva_fault(struct kvm_vcpu *vcpu,
1197                                                    unsigned long gva,
1198                                                    bool write)
1199 {
1200         struct mips_coproc *cop0 = vcpu->arch.cop0;
1201         struct kvm_mips_tlb *tlb;
1202         int index;
1203
1204         if (KVM_GUEST_KSEGX(gva) == KVM_GUEST_KSEG0) {
1205                 if (kvm_mips_handle_kseg0_tlb_fault(gva, vcpu, write) < 0)
1206                         return KVM_MIPS_GPA;
1207         } else if ((KVM_GUEST_KSEGX(gva) < KVM_GUEST_KSEG0) ||
1208                    KVM_GUEST_KSEGX(gva) == KVM_GUEST_KSEG23) {
1209                 /* Address should be in the guest TLB */
1210                 index = kvm_mips_guest_tlb_lookup(vcpu, (gva & VPN2_MASK) |
1211                           (kvm_read_c0_guest_entryhi(cop0) & KVM_ENTRYHI_ASID));
1212                 if (index < 0)
1213                         return KVM_MIPS_TLB;
1214                 tlb = &vcpu->arch.guest_tlb[index];
1215
1216                 /* Entry should be valid, and dirty for writes */
1217                 if (!TLB_IS_VALID(*tlb, gva))
1218                         return KVM_MIPS_TLBINV;
1219                 if (write && !TLB_IS_DIRTY(*tlb, gva))
1220                         return KVM_MIPS_TLBMOD;
1221
1222                 if (kvm_mips_handle_mapped_seg_tlb_fault(vcpu, tlb, gva, write))
1223                         return KVM_MIPS_GPA;
1224         } else {
1225                 return KVM_MIPS_GVA;
1226         }
1227
1228         return KVM_MIPS_MAPPED;
1229 }
1230
1231 int kvm_get_inst(u32 *opc, struct kvm_vcpu *vcpu, u32 *out)
1232 {
1233         int err;
1234
1235         if (WARN(IS_ENABLED(CONFIG_KVM_MIPS_VZ),
1236                  "Expect BadInstr/BadInstrP registers to be used with VZ\n"))
1237                 return -EINVAL;
1238
1239 retry:
1240         kvm_trap_emul_gva_lockless_begin(vcpu);
1241         err = get_user(*out, opc);
1242         kvm_trap_emul_gva_lockless_end(vcpu);
1243
1244         if (unlikely(err)) {
1245                 /*
1246                  * Try to handle the fault, maybe we just raced with a GVA
1247                  * invalidation.
1248                  */
1249                 err = kvm_trap_emul_gva_fault(vcpu, (unsigned long)opc,
1250                                               false);
1251                 if (unlikely(err)) {
1252                         kvm_err("%s: illegal address: %p\n",
1253                                 __func__, opc);
1254                         return -EFAULT;
1255                 }
1256
1257                 /* Hopefully it'll work now */
1258                 goto retry;
1259         }
1260         return 0;
1261 }