arch/powerpc/kvm/book3s_hv_rm_mmu.c

   1 /*
   2  * This program is free software; you can redistribute it and/or modify
   3  * it under the terms of the GNU General Public License, version 2, as
   4  * published by the Free Software Foundation.
   5  *
   6  * Copyright 2010-2011 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
   7  */
   8
   9 #include <linux/types.h>
  10 #include <linux/string.h>
  11 #include <linux/kvm.h>
  12 #include <linux/kvm_host.h>
  13 #include <linux/hugetlb.h>
  14 #include <linux/module.h>
  15
  16 #include <asm/tlbflush.h>
  17 #include <asm/kvm_ppc.h>
  18 #include <asm/kvm_book3s.h>
  19 #include <asm/mmu-hash64.h>
  20 #include <asm/hvcall.h>
  21 #include <asm/synch.h>
  22 #include <asm/ppc-opcode.h>
  23
  24 /* Translate address of a vmalloc'd thing to a linear map address */
  25 static void *real_vmalloc_addr(void *x)
  26 {
  27         unsigned long addr = (unsigned long) x;
  28         pte_t *p;
  29
  30         p = find_linux_pte_or_hugepte(swapper_pg_dir, addr, NULL);
  31         if (!p || !pte_present(*p))
  32                 return NULL;
  33         /* assume we don't have huge pages in vmalloc space... */
  34         addr = (pte_pfn(*p) << PAGE_SHIFT) | (addr & ~PAGE_MASK);
  35         return __va(addr);
  36 }
  37
  38 /* Return 1 if we need to do a global tlbie, 0 if we can use tlbiel */
  39 static int global_invalidates(struct kvm *kvm, unsigned long flags)
  40 {
  41         int global;
  42
  43         /*
  44          * If there is only one vcore, and it's currently running,
  45          * as indicated by local_paca->kvm_hstate.kvm_vcpu being set,
  46          * we can use tlbiel as long as we mark all other physical
  47          * cores as potentially having stale TLB entries for this lpid.
  48          * Otherwise, don't use tlbiel.
  49          */
  50         if (kvm->arch.online_vcores == 1 && local_paca->kvm_hstate.kvm_vcpu)
  51                 global = 0;
  52         else
  53                 global = 1;
  54
  55         if (!global) {
  56                 /* any other core might now have stale TLB entries... */
  57                 smp_wmb();
  58                 cpumask_setall(&kvm->arch.need_tlb_flush);
  59                 cpumask_clear_cpu(local_paca->kvm_hstate.kvm_vcore->pcpu,
  60                                   &kvm->arch.need_tlb_flush);
  61         }
  62
  63         return global;
  64 }
  65
  66 /*
  67  * Add this HPTE into the chain for the real page.
  68  * Must be called with the chain locked; it unlocks the chain.
  69  */
  70 void kvmppc_add_revmap_chain(struct kvm *kvm, struct revmap_entry *rev,
  71                              unsigned long *rmap, long pte_index, int realmode)
  72 {
  73         struct revmap_entry *head, *tail;
  74         unsigned long i;
  75
  76         if (*rmap & KVMPPC_RMAP_PRESENT) {
  77                 i = *rmap & KVMPPC_RMAP_INDEX;
  78                 head = &kvm->arch.revmap[i];
  79                 if (realmode)
  80                         head = real_vmalloc_addr(head);
  81                 tail = &kvm->arch.revmap[head->back];
  82                 if (realmode)
  83                         tail = real_vmalloc_addr(tail);
  84                 rev->forw = i;
  85                 rev->back = head->back;
  86                 tail->forw = pte_index;
  87                 head->back = pte_index;
  88         } else {
  89                 rev->forw = rev->back = pte_index;
  90                 *rmap = (*rmap & ~KVMPPC_RMAP_INDEX) |
  91                         pte_index | KVMPPC_RMAP_PRESENT;
  92         }
  93         unlock_rmap(rmap);
  94 }
  95 EXPORT_SYMBOL_GPL(kvmppc_add_revmap_chain);
  96
  97 /* Remove this HPTE from the chain for a real page */
  98 static void remove_revmap_chain(struct kvm *kvm, long pte_index,
  99                                 struct revmap_entry *rev,
 100                                 unsigned long hpte_v, unsigned long hpte_r)
 101 {
 102         struct revmap_entry *next, *prev;
 103         unsigned long gfn, ptel, head;
 104         struct kvm_memory_slot *memslot;
 105         unsigned long *rmap;
 106         unsigned long rcbits;
 107
 108         rcbits = hpte_r & (HPTE_R_R | HPTE_R_C);
 109         ptel = rev->guest_rpte |= rcbits;
 110         gfn = hpte_rpn(ptel, hpte_page_size(hpte_v, ptel));
 111         memslot = __gfn_to_memslot(kvm_memslots_raw(kvm), gfn);
 112         if (!memslot)
 113                 return;
 114
 115         rmap = real_vmalloc_addr(&memslot->arch.rmap[gfn - memslot->base_gfn]);
 116         lock_rmap(rmap);
 117
 118         head = *rmap & KVMPPC_RMAP_INDEX;
 119         next = real_vmalloc_addr(&kvm->arch.revmap[rev->forw]);
 120         prev = real_vmalloc_addr(&kvm->arch.revmap[rev->back]);
 121         next->back = rev->back;
 122         prev->forw = rev->forw;
 123         if (head == pte_index) {
 124                 head = rev->forw;
 125                 if (head == pte_index)
 126                         *rmap &= ~(KVMPPC_RMAP_PRESENT | KVMPPC_RMAP_INDEX);
 127                 else
 128                         *rmap = (*rmap & ~KVMPPC_RMAP_INDEX) | head;
 129         }
 130         *rmap |= rcbits << KVMPPC_RMAP_RC_SHIFT;
 131         unlock_rmap(rmap);
 132 }
 133
 134 static pte_t lookup_linux_pte_and_update(pgd_t *pgdir, unsigned long hva,
 135                               int writing, unsigned long *pte_sizep)
 136 {
 137         pte_t *ptep;
 138         unsigned long ps = *pte_sizep;
 139         unsigned int hugepage_shift;
 140
 141         ptep = find_linux_pte_or_hugepte(pgdir, hva, &hugepage_shift);
 142         if (!ptep)
 143                 return __pte(0);
 144         if (hugepage_shift)
 145                 *pte_sizep = 1ul << hugepage_shift;
 146         else
 147                 *pte_sizep = PAGE_SIZE;
 148         if (ps > *pte_sizep)
 149                 return __pte(0);
 150         return kvmppc_read_update_linux_pte(ptep, writing, hugepage_shift);
 151 }
 152
 153 long kvmppc_do_h_enter(struct kvm *kvm, unsigned long flags,
 154                        long pte_index, unsigned long pteh, unsigned long ptel,
 155                        pgd_t *pgdir, bool realmode, unsigned long *pte_idx_ret)
 156 {
 157         unsigned long i, pa, gpa, gfn, psize;
 158         unsigned long slot_fn, hva;
 159         __be64 *hpte;
 160         struct revmap_entry *rev;
 161         unsigned long g_ptel;
 162         struct kvm_memory_slot *memslot;
 163         unsigned long pte_size;
 164         unsigned long is_io;
 165         unsigned long *rmap;
 166         pte_t pte;
 167         unsigned int writing;
 168         unsigned long mmu_seq;
 169         unsigned long rcbits;
 170
 171         psize = hpte_page_size(pteh, ptel);
 172         if (!psize)
 173                 return H_PARAMETER;
 174         writing = hpte_is_writable(ptel);
 175         pteh &= ~(HPTE_V_HVLOCK | HPTE_V_ABSENT | HPTE_V_VALID);
 176         ptel &= ~HPTE_GR_RESERVED;
 177         g_ptel = ptel;
 178
 179         /* used later to detect if we might have been invalidated */
 180         mmu_seq = kvm->mmu_notifier_seq;
 181         smp_rmb();
 182
 183         /* Find the memslot (if any) for this address */
 184         gpa = (ptel & HPTE_R_RPN) & ~(psize - 1);
 185         gfn = gpa >> PAGE_SHIFT;
 186         memslot = __gfn_to_memslot(kvm_memslots_raw(kvm), gfn);
 187         pa = 0;
 188         is_io = ~0ul;
 189         rmap = NULL;
 190         if (!(memslot && !(memslot->flags & KVM_MEMSLOT_INVALID))) {
 191                 /* Emulated MMIO - mark this with key=31 */
 192                 pteh |= HPTE_V_ABSENT;
 193                 ptel |= HPTE_R_KEY_HI | HPTE_R_KEY_LO;
 194                 goto do_insert;
 195         }
 196
 197         /* Check if the requested page fits entirely in the memslot. */
 198         if (!slot_is_aligned(memslot, psize))
 199                 return H_PARAMETER;
 200         slot_fn = gfn - memslot->base_gfn;
 201         rmap = &memslot->arch.rmap[slot_fn];
 202
 203         /* Translate to host virtual address */
 204         hva = __gfn_to_hva_memslot(memslot, gfn);
 205
 206         /* Look up the Linux PTE for the backing page */
 207         pte_size = psize;
 208         pte = lookup_linux_pte_and_update(pgdir, hva, writing, &pte_size);
 209         if (pte_present(pte) && !pte_protnone(pte)) {
 210                 if (writing && !pte_write(pte))
 211                         /* make the actual HPTE be read-only */
 212                         ptel = hpte_make_readonly(ptel);
 213                 is_io = hpte_cache_bits(pte_val(pte));
 214                 pa = pte_pfn(pte) << PAGE_SHIFT;
 215                 pa |= hva & (pte_size - 1);
 216                 pa |= gpa & ~PAGE_MASK;
 217         }
 218
 219         if (pte_size < psize)
 220                 return H_PARAMETER;
 221
 222         ptel &= ~(HPTE_R_PP0 - psize);
 223         ptel |= pa;
 224
 225         if (pa)
 226                 pteh |= HPTE_V_VALID;
 227         else
 228                 pteh |= HPTE_V_ABSENT;
 229
 230         /* Check WIMG */
 231         if (is_io != ~0ul && !hpte_cache_flags_ok(ptel, is_io)) {
 232                 if (is_io)
 233                         return H_PARAMETER;
 234                 /*
 235                  * Allow guest to map emulated device memory as
 236                  * uncacheable, but actually make it cacheable.
 237                  */
 238                 ptel &= ~(HPTE_R_W|HPTE_R_I|HPTE_R_G);
 239                 ptel |= HPTE_R_M;
 240         }
 241
 242         /* Find and lock the HPTEG slot to use */
 243  do_insert:
 244         if (pte_index >= kvm->arch.hpt_npte)
 245                 return H_PARAMETER;
 246         if (likely((flags & H_EXACT) == 0)) {
 247                 pte_index &= ~7UL;
 248                 hpte = (__be64 *)(kvm->arch.hpt_virt + (pte_index << 4));
 249                 for (i = 0; i < 8; ++i) {
 250                         if ((be64_to_cpu(*hpte) & HPTE_V_VALID) == 0 &&
 251                             try_lock_hpte(hpte, HPTE_V_HVLOCK | HPTE_V_VALID |
 252                                           HPTE_V_ABSENT))
 253                                 break;
 254                         hpte += 2;
 255                 }
 256                 if (i == 8) {
 257                         /*
 258                          * Since try_lock_hpte doesn't retry (not even stdcx.
 259                          * failures), it could be that there is a free slot
 260                          * but we transiently failed to lock it.  Try again,
 261                          * actually locking each slot and checking it.
 262                          */
 263                         hpte -= 16;
 264                         for (i = 0; i < 8; ++i) {
 265                                 u64 pte;
 266                                 while (!try_lock_hpte(hpte, HPTE_V_HVLOCK))
 267                                         cpu_relax();
 268                                 pte = be64_to_cpu(hpte[0]);
 269                                 if (!(pte & (HPTE_V_VALID | HPTE_V_ABSENT)))
 270                                         break;
 271                                 __unlock_hpte(hpte, pte);
 272                                 hpte += 2;
 273                         }
 274                         if (i == 8)
 275                                 return H_PTEG_FULL;
 276                 }
 277                 pte_index += i;
 278         } else {
 279                 hpte = (__be64 *)(kvm->arch.hpt_virt + (pte_index << 4));
 280                 if (!try_lock_hpte(hpte, HPTE_V_HVLOCK | HPTE_V_VALID |
 281                                    HPTE_V_ABSENT)) {
 282                         /* Lock the slot and check again */
 283                         u64 pte;
 284
 285                         while (!try_lock_hpte(hpte, HPTE_V_HVLOCK))
 286                                 cpu_relax();
 287                         pte = be64_to_cpu(hpte[0]);
 288                         if (pte & (HPTE_V_VALID | HPTE_V_ABSENT)) {
 289                                 __unlock_hpte(hpte, pte);
 290                                 return H_PTEG_FULL;
 291                         }
 292                 }
 293         }
 294
 295         /* Save away the guest's idea of the second HPTE dword */
 296         rev = &kvm->arch.revmap[pte_index];
 297         if (realmode)
 298                 rev = real_vmalloc_addr(rev);
 299         if (rev) {
 300                 rev->guest_rpte = g_ptel;
 301                 note_hpte_modification(kvm, rev);
 302         }
 303
 304         /* Link HPTE into reverse-map chain */
 305         if (pteh & HPTE_V_VALID) {
 306                 if (realmode)
 307                         rmap = real_vmalloc_addr(rmap);
 308                 lock_rmap(rmap);
 309                 /* Check for pending invalidations under the rmap chain lock */
 310                 if (mmu_notifier_retry(kvm, mmu_seq)) {
 311                         /* inval in progress, write a non-present HPTE */
 312                         pteh |= HPTE_V_ABSENT;
 313                         pteh &= ~HPTE_V_VALID;
 314                         unlock_rmap(rmap);
 315                 } else {
 316                         kvmppc_add_revmap_chain(kvm, rev, rmap, pte_index,
 317                                                 realmode);
 318                         /* Only set R/C in real HPTE if already set in *rmap */
 319                         rcbits = *rmap >> KVMPPC_RMAP_RC_SHIFT;
 320                         ptel &= rcbits | ~(HPTE_R_R | HPTE_R_C);
 321                 }
 322         }
 323
 324         hpte[1] = cpu_to_be64(ptel);
 325
 326         /* Write the first HPTE dword, unlocking the HPTE and making it valid */
 327         eieio();
 328         __unlock_hpte(hpte, pteh);
 329         asm volatile("ptesync" : : : "memory");
 330
 331         *pte_idx_ret = pte_index;
 332         return H_SUCCESS;
 333 }
 334 EXPORT_SYMBOL_GPL(kvmppc_do_h_enter);
 335
 336 long kvmppc_h_enter(struct kvm_vcpu *vcpu, unsigned long flags,
 337                     long pte_index, unsigned long pteh, unsigned long ptel)
 338 {
 339         return kvmppc_do_h_enter(vcpu->kvm, flags, pte_index, pteh, ptel,
 340                                  vcpu->arch.pgdir, true, &vcpu->arch.gpr[4]);
 341 }
 342
 343 #ifdef __BIG_ENDIAN__
 344 #define LOCK_TOKEN      (*(u32 *)(&get_paca()->lock_token))
 345 #else
 346 #define LOCK_TOKEN      (*(u32 *)(&get_paca()->paca_index))
 347 #endif
 348
 349 static inline int try_lock_tlbie(unsigned int *lock)
 350 {
 351         unsigned int tmp, old;
 352         unsigned int token = LOCK_TOKEN;
 353
 354         asm volatile("1:lwarx   %1,0,%2\n"
 355                      "  cmpwi   cr0,%1,0\n"
 356                      "  bne     2f\n"
 357                      "  stwcx.  %3,0,%2\n"
 358                      "  bne-    1b\n"
 359                      "  isync\n"
 360                      "2:"
 361                      : "=&r" (tmp), "=&r" (old)
 362                      : "r" (lock), "r" (token)
 363                      : "cc", "memory");
 364         return old == 0;
 365 }
 366
 367 static void do_tlbies(struct kvm *kvm, unsigned long *rbvalues,
 368                       long npages, int global, bool need_sync)
 369 {
 370         long i;
 371
 372         if (global) {
 373                 while (!try_lock_tlbie(&kvm->arch.tlbie_lock))
 374                         cpu_relax();
 375                 if (need_sync)
 376                         asm volatile("ptesync" : : : "memory");
 377                 for (i = 0; i < npages; ++i)
 378                         asm volatile(PPC_TLBIE(%1,%0) : :
 379                                      "r" (rbvalues[i]), "r" (kvm->arch.lpid));
 380                 asm volatile("eieio; tlbsync; ptesync" : : : "memory");
 381                 kvm->arch.tlbie_lock = 0;
 382         } else {
 383                 if (need_sync)
 384                         asm volatile("ptesync" : : : "memory");
 385                 for (i = 0; i < npages; ++i)
 386                         asm volatile("tlbiel %0" : : "r" (rbvalues[i]));
 387                 asm volatile("ptesync" : : : "memory");
 388         }
 389 }
 390
 391 long kvmppc_do_h_remove(struct kvm *kvm, unsigned long flags,
 392                         unsigned long pte_index, unsigned long avpn,
 393                         unsigned long *hpret)
 394 {
 395         __be64 *hpte;
 396         unsigned long v, r, rb;
 397         struct revmap_entry *rev;
 398         u64 pte;
 399
 400         if (pte_index >= kvm->arch.hpt_npte)
 401                 return H_PARAMETER;
 402         hpte = (__be64 *)(kvm->arch.hpt_virt + (pte_index << 4));
 403         while (!try_lock_hpte(hpte, HPTE_V_HVLOCK))
 404                 cpu_relax();
 405         pte = be64_to_cpu(hpte[0]);
 406         if ((pte & (HPTE_V_ABSENT | HPTE_V_VALID)) == 0 ||
 407             ((flags & H_AVPN) && (pte & ~0x7fUL) != avpn) ||
 408             ((flags & H_ANDCOND) && (pte & avpn) != 0)) {
 409                 __unlock_hpte(hpte, pte);
 410                 return H_NOT_FOUND;
 411         }
 412
 413         rev = real_vmalloc_addr(&kvm->arch.revmap[pte_index]);
 414         v = pte & ~HPTE_V_HVLOCK;
 415         if (v & HPTE_V_VALID) {
 416                 u64 pte1;
 417
 418                 pte1 = be64_to_cpu(hpte[1]);
 419                 hpte[0] &= ~cpu_to_be64(HPTE_V_VALID);
 420                 rb = compute_tlbie_rb(v, pte1, pte_index);
 421                 do_tlbies(kvm, &rb, 1, global_invalidates(kvm, flags), true);
 422                 /* Read PTE low word after tlbie to get final R/C values */
 423                 remove_revmap_chain(kvm, pte_index, rev, v, pte1);
 424         }
 425         r = rev->guest_rpte & ~HPTE_GR_RESERVED;
 426         note_hpte_modification(kvm, rev);
 427         unlock_hpte(hpte, 0);
 428
 429         hpret[0] = v;
 430         hpret[1] = r;
 431         return H_SUCCESS;
 432 }
 433 EXPORT_SYMBOL_GPL(kvmppc_do_h_remove);
 434
 435 long kvmppc_h_remove(struct kvm_vcpu *vcpu, unsigned long flags,
 436                      unsigned long pte_index, unsigned long avpn)
 437 {
 438         return kvmppc_do_h_remove(vcpu->kvm, flags, pte_index, avpn,
 439                                   &vcpu->arch.gpr[4]);
 440 }
 441
 442 long kvmppc_h_bulk_remove(struct kvm_vcpu *vcpu)
 443 {
 444         struct kvm *kvm = vcpu->kvm;
 445         unsigned long *args = &vcpu->arch.gpr[4];
 446         __be64 *hp, *hptes[4];
 447         unsigned long tlbrb[4];
 448         long int i, j, k, n, found, indexes[4];
 449         unsigned long flags, req, pte_index, rcbits;
 450         int global;
 451         long int ret = H_SUCCESS;
 452         struct revmap_entry *rev, *revs[4];
 453         u64 hp0;
 454
 455         global = global_invalidates(kvm, 0);
 456         for (i = 0; i < 4 && ret == H_SUCCESS; ) {
 457                 n = 0;
 458                 for (; i < 4; ++i) {
 459                         j = i * 2;
 460                         pte_index = args[j];
 461                         flags = pte_index >> 56;
 462                         pte_index &= ((1ul << 56) - 1);
 463                         req = flags >> 6;
 464                         flags &= 3;
 465                         if (req == 3) {         /* no more requests */
 466                                 i = 4;
 467                                 break;
 468                         }
 469                         if (req != 1 || flags == 3 ||
 470                             pte_index >= kvm->arch.hpt_npte) {
 471                                 /* parameter error */
 472                                 args[j] = ((0xa0 | flags) << 56) + pte_index;
 473                                 ret = H_PARAMETER;
 474                                 break;
 475                         }
 476                         hp = (__be64 *) (kvm->arch.hpt_virt + (pte_index << 4));
 477                         /* to avoid deadlock, don't spin except for first */
 478                         if (!try_lock_hpte(hp, HPTE_V_HVLOCK)) {
 479                                 if (n)
 480                                         break;
 481                                 while (!try_lock_hpte(hp, HPTE_V_HVLOCK))
 482                                         cpu_relax();
 483                         }
 484                         found = 0;
 485                         hp0 = be64_to_cpu(hp[0]);
 486                         if (hp0 & (HPTE_V_ABSENT | HPTE_V_VALID)) {
 487                                 switch (flags & 3) {
 488                                 case 0:         /* absolute */
 489                                         found = 1;
 490                                         break;
 491                                 case 1:         /* andcond */
 492                                         if (!(hp0 & args[j + 1]))
 493                                                 found = 1;
 494                                         break;
 495                                 case 2:         /* AVPN */
 496                                         if ((hp0 & ~0x7fUL) == args[j + 1])
 497                                                 found = 1;
 498                                         break;
 499                                 }
 500                         }
 501                         if (!found) {
 502                                 hp[0] &= ~cpu_to_be64(HPTE_V_HVLOCK);
 503                                 args[j] = ((0x90 | flags) << 56) + pte_index;
 504                                 continue;
 505                         }
 506
 507                         args[j] = ((0x80 | flags) << 56) + pte_index;
 508                         rev = real_vmalloc_addr(&kvm->arch.revmap[pte_index]);
 509                         note_hpte_modification(kvm, rev);
 510
 511                         if (!(hp0 & HPTE_V_VALID)) {
 512                                 /* insert R and C bits from PTE */
 513                                 rcbits = rev->guest_rpte & (HPTE_R_R|HPTE_R_C);
 514                                 args[j] |= rcbits << (56 - 5);
 515                                 hp[0] = 0;
 516                                 continue;
 517                         }
 518
 519                         /* leave it locked */
 520                         hp[0] &= ~cpu_to_be64(HPTE_V_VALID);
 521                         tlbrb[n] = compute_tlbie_rb(be64_to_cpu(hp[0]),
 522                                 be64_to_cpu(hp[1]), pte_index);
 523                         indexes[n] = j;
 524                         hptes[n] = hp;
 525                         revs[n] = rev;
 526                         ++n;
 527                 }
 528
 529                 if (!n)
 530                         break;
 531
 532                 /* Now that we've collected a batch, do the tlbies */
 533                 do_tlbies(kvm, tlbrb, n, global, true);
 534
 535                 /* Read PTE low words after tlbie to get final R/C values */
 536                 for (k = 0; k < n; ++k) {
 537                         j = indexes[k];
 538                         pte_index = args[j] & ((1ul << 56) - 1);
 539                         hp = hptes[k];
 540                         rev = revs[k];
 541                         remove_revmap_chain(kvm, pte_index, rev,
 542                                 be64_to_cpu(hp[0]), be64_to_cpu(hp[1]));
 543                         rcbits = rev->guest_rpte & (HPTE_R_R|HPTE_R_C);
 544                         args[j] |= rcbits << (56 - 5);
 545                         __unlock_hpte(hp, 0);
 546                 }
 547         }
 548
 549         return ret;
 550 }
 551
 552 long kvmppc_h_protect(struct kvm_vcpu *vcpu, unsigned long flags,
 553                       unsigned long pte_index, unsigned long avpn,
 554                       unsigned long va)
 555 {
 556         struct kvm *kvm = vcpu->kvm;
 557         __be64 *hpte;
 558         struct revmap_entry *rev;
 559         unsigned long v, r, rb, mask, bits;
 560         u64 pte;
 561
 562         if (pte_index >= kvm->arch.hpt_npte)
 563                 return H_PARAMETER;
 564
 565         hpte = (__be64 *)(kvm->arch.hpt_virt + (pte_index << 4));
 566         while (!try_lock_hpte(hpte, HPTE_V_HVLOCK))
 567                 cpu_relax();
 568         pte = be64_to_cpu(hpte[0]);
 569         if ((pte & (HPTE_V_ABSENT | HPTE_V_VALID)) == 0 ||
 570             ((flags & H_AVPN) && (pte & ~0x7fUL) != avpn)) {
 571                 __unlock_hpte(hpte, pte);
 572                 return H_NOT_FOUND;
 573         }
 574
 575         v = pte;
 576         bits = (flags << 55) & HPTE_R_PP0;
 577         bits |= (flags << 48) & HPTE_R_KEY_HI;
 578         bits |= flags & (HPTE_R_PP | HPTE_R_N | HPTE_R_KEY_LO);
 579
 580         /* Update guest view of 2nd HPTE dword */
 581         mask = HPTE_R_PP0 | HPTE_R_PP | HPTE_R_N |
 582                 HPTE_R_KEY_HI | HPTE_R_KEY_LO;
 583         rev = real_vmalloc_addr(&kvm->arch.revmap[pte_index]);
 584         if (rev) {
 585                 r = (rev->guest_rpte & ~mask) | bits;
 586                 rev->guest_rpte = r;
 587                 note_hpte_modification(kvm, rev);
 588         }
 589
 590         /* Update HPTE */
 591         if (v & HPTE_V_VALID) {
 592                 /*
 593                  * If the page is valid, don't let it transition from
 594                  * readonly to writable.  If it should be writable, we'll
 595                  * take a trap and let the page fault code sort it out.
 596                  */
 597                 pte = be64_to_cpu(hpte[1]);
 598                 r = (pte & ~mask) | bits;
 599                 if (hpte_is_writable(r) && !hpte_is_writable(pte))
 600                         r = hpte_make_readonly(r);
 601                 /* If the PTE is changing, invalidate it first */
 602                 if (r != pte) {
 603                         rb = compute_tlbie_rb(v, r, pte_index);
 604                         hpte[0] = cpu_to_be64((v & ~HPTE_V_VALID) |
 605                                               HPTE_V_ABSENT);
 606                         do_tlbies(kvm, &rb, 1, global_invalidates(kvm, flags),
 607                                   true);
 608                         hpte[1] = cpu_to_be64(r);
 609                 }
 610         }
 611         unlock_hpte(hpte, v & ~HPTE_V_HVLOCK);
 612         asm volatile("ptesync" : : : "memory");
 613         return H_SUCCESS;
 614 }
 615
 616 long kvmppc_h_read(struct kvm_vcpu *vcpu, unsigned long flags,
 617                    unsigned long pte_index)
 618 {
 619         struct kvm *kvm = vcpu->kvm;
 620         __be64 *hpte;
 621         unsigned long v, r;
 622         int i, n = 1;
 623         struct revmap_entry *rev = NULL;
 624
 625         if (pte_index >= kvm->arch.hpt_npte)
 626                 return H_PARAMETER;
 627         if (flags & H_READ_4) {
 628                 pte_index &= ~3;
 629                 n = 4;
 630         }
 631         rev = real_vmalloc_addr(&kvm->arch.revmap[pte_index]);
 632         for (i = 0; i < n; ++i, ++pte_index) {
 633                 hpte = (__be64 *)(kvm->arch.hpt_virt + (pte_index << 4));
 634                 v = be64_to_cpu(hpte[0]) & ~HPTE_V_HVLOCK;
 635                 r = be64_to_cpu(hpte[1]);
 636                 if (v & HPTE_V_ABSENT) {
 637                         v &= ~HPTE_V_ABSENT;
 638                         v |= HPTE_V_VALID;
 639                 }
 640                 if (v & HPTE_V_VALID) {
 641                         r = rev[i].guest_rpte | (r & (HPTE_R_R | HPTE_R_C));
 642                         r &= ~HPTE_GR_RESERVED;
 643                 }
 644                 vcpu->arch.gpr[4 + i * 2] = v;
 645                 vcpu->arch.gpr[5 + i * 2] = r;
 646         }
 647         return H_SUCCESS;
 648 }
 649
 650 void kvmppc_invalidate_hpte(struct kvm *kvm, __be64 *hptep,
 651                         unsigned long pte_index)
 652 {
 653         unsigned long rb;
 654
 655         hptep[0] &= ~cpu_to_be64(HPTE_V_VALID);
 656         rb = compute_tlbie_rb(be64_to_cpu(hptep[0]), be64_to_cpu(hptep[1]),
 657                               pte_index);
 658         do_tlbies(kvm, &rb, 1, 1, true);
 659 }
 660 EXPORT_SYMBOL_GPL(kvmppc_invalidate_hpte);
 661
 662 void kvmppc_clear_ref_hpte(struct kvm *kvm, __be64 *hptep,
 663                            unsigned long pte_index)
 664 {
 665         unsigned long rb;
 666         unsigned char rbyte;
 667
 668         rb = compute_tlbie_rb(be64_to_cpu(hptep[0]), be64_to_cpu(hptep[1]),
 669                               pte_index);
 670         rbyte = (be64_to_cpu(hptep[1]) & ~HPTE_R_R) >> 8;
 671         /* modify only the second-last byte, which contains the ref bit */
 672         *((char *)hptep + 14) = rbyte;
 673         do_tlbies(kvm, &rb, 1, 1, false);
 674 }
 675 EXPORT_SYMBOL_GPL(kvmppc_clear_ref_hpte);
 676
 677 static int slb_base_page_shift[4] = {
 678         24,     /* 16M */
 679         16,     /* 64k */
 680         34,     /* 16G */
 681         20,     /* 1M, unsupported */
 682 };
 683
 684 /* When called from virtmode, this func should be protected by
 685  * preempt_disable(), otherwise, the holding of HPTE_V_HVLOCK
 686  * can trigger deadlock issue.
 687  */
 688 long kvmppc_hv_find_lock_hpte(struct kvm *kvm, gva_t eaddr, unsigned long slb_v,
 689                               unsigned long valid)
 690 {
 691         unsigned int i;
 692         unsigned int pshift;
 693         unsigned long somask;
 694         unsigned long vsid, hash;
 695         unsigned long avpn;
 696         __be64 *hpte;
 697         unsigned long mask, val;
 698         unsigned long v, r;
 699
 700         /* Get page shift, work out hash and AVPN etc. */
 701         mask = SLB_VSID_B | HPTE_V_AVPN | HPTE_V_SECONDARY;
 702         val = 0;
 703         pshift = 12;
 704         if (slb_v & SLB_VSID_L) {
 705                 mask |= HPTE_V_LARGE;
 706                 val |= HPTE_V_LARGE;
 707                 pshift = slb_base_page_shift[(slb_v & SLB_VSID_LP) >> 4];
 708         }
 709         if (slb_v & SLB_VSID_B_1T) {
 710                 somask = (1UL << 40) - 1;
 711                 vsid = (slb_v & ~SLB_VSID_B) >> SLB_VSID_SHIFT_1T;
 712                 vsid ^= vsid << 25;
 713         } else {
 714                 somask = (1UL << 28) - 1;
 715                 vsid = (slb_v & ~SLB_VSID_B) >> SLB_VSID_SHIFT;
 716         }
 717         hash = (vsid ^ ((eaddr & somask) >> pshift)) & kvm->arch.hpt_mask;
 718         avpn = slb_v & ~(somask >> 16); /* also includes B */
 719         avpn |= (eaddr & somask) >> 16;
 720
 721         if (pshift >= 24)
 722                 avpn &= ~((1UL << (pshift - 16)) - 1);
 723         else
 724                 avpn &= ~0x7fUL;
 725         val |= avpn;
 726
 727         for (;;) {
 728                 hpte = (__be64 *)(kvm->arch.hpt_virt + (hash << 7));
 729
 730                 for (i = 0; i < 16; i += 2) {
 731                         /* Read the PTE racily */
 732                         v = be64_to_cpu(hpte[i]) & ~HPTE_V_HVLOCK;
 733
 734                         /* Check valid/absent, hash, segment size and AVPN */
 735                         if (!(v & valid) || (v & mask) != val)
 736                                 continue;
 737
 738                         /* Lock the PTE and read it under the lock */
 739                         while (!try_lock_hpte(&hpte[i], HPTE_V_HVLOCK))
 740                                 cpu_relax();
 741                         v = be64_to_cpu(hpte[i]) & ~HPTE_V_HVLOCK;
 742                         r = be64_to_cpu(hpte[i+1]);
 743
 744                         /*
 745                          * Check the HPTE again, including base page size
 746                          */
 747                         if ((v & valid) && (v & mask) == val &&
 748                             hpte_base_page_size(v, r) == (1ul << pshift))
 749                                 /* Return with the HPTE still locked */
 750                                 return (hash << 3) + (i >> 1);
 751
 752                         __unlock_hpte(&hpte[i], v);
 753                 }
 754
 755                 if (val & HPTE_V_SECONDARY)
 756                         break;
 757                 val |= HPTE_V_SECONDARY;
 758                 hash = hash ^ kvm->arch.hpt_mask;
 759         }
 760         return -1;
 761 }
 762 EXPORT_SYMBOL(kvmppc_hv_find_lock_hpte);
 763
 764 /*
 765  * Called in real mode to check whether an HPTE not found fault
 766  * is due to accessing a paged-out page or an emulated MMIO page,
 767  * or if a protection fault is due to accessing a page that the
 768  * guest wanted read/write access to but which we made read-only.
 769  * Returns a possibly modified status (DSISR) value if not
 770  * (i.e. pass the interrupt to the guest),
 771  * -1 to pass the fault up to host kernel mode code, -2 to do that
 772  * and also load the instruction word (for MMIO emulation),
 773  * or 0 if we should make the guest retry the access.
 774  */
 775 long kvmppc_hpte_hv_fault(struct kvm_vcpu *vcpu, unsigned long addr,
 776                           unsigned long slb_v, unsigned int status, bool data)
 777 {
 778         struct kvm *kvm = vcpu->kvm;
 779         long int index;
 780         unsigned long v, r, gr;
 781         __be64 *hpte;
 782         unsigned long valid;
 783         struct revmap_entry *rev;
 784         unsigned long pp, key;
 785
 786         /* For protection fault, expect to find a valid HPTE */
 787         valid = HPTE_V_VALID;
 788         if (status & DSISR_NOHPTE)
 789                 valid |= HPTE_V_ABSENT;
 790
 791         index = kvmppc_hv_find_lock_hpte(kvm, addr, slb_v, valid);
 792         if (index < 0) {
 793                 if (status & DSISR_NOHPTE)
 794                         return status;  /* there really was no HPTE */
 795                 return 0;               /* for prot fault, HPTE disappeared */
 796         }
 797         hpte = (__be64 *)(kvm->arch.hpt_virt + (index << 4));
 798         v = be64_to_cpu(hpte[0]) & ~HPTE_V_HVLOCK;
 799         r = be64_to_cpu(hpte[1]);
 800         rev = real_vmalloc_addr(&kvm->arch.revmap[index]);
 801         gr = rev->guest_rpte;
 802
 803         unlock_hpte(hpte, v);
 804
 805         /* For not found, if the HPTE is valid by now, retry the instruction */
 806         if ((status & DSISR_NOHPTE) && (v & HPTE_V_VALID))
 807                 return 0;
 808
 809         /* Check access permissions to the page */
 810         pp = gr & (HPTE_R_PP0 | HPTE_R_PP);
 811         key = (vcpu->arch.shregs.msr & MSR_PR) ? SLB_VSID_KP : SLB_VSID_KS;
 812         status &= ~DSISR_NOHPTE;        /* DSISR_NOHPTE == SRR1_ISI_NOPT */
 813         if (!data) {
 814                 if (gr & (HPTE_R_N | HPTE_R_G))
 815                         return status | SRR1_ISI_N_OR_G;
 816                 if (!hpte_read_permission(pp, slb_v & key))
 817                         return status | SRR1_ISI_PROT;
 818         } else if (status & DSISR_ISSTORE) {
 819                 /* check write permission */
 820                 if (!hpte_write_permission(pp, slb_v & key))
 821                         return status | DSISR_PROTFAULT;
 822         } else {
 823                 if (!hpte_read_permission(pp, slb_v & key))
 824                         return status | DSISR_PROTFAULT;
 825         }
 826
 827         /* Check storage key, if applicable */
 828         if (data && (vcpu->arch.shregs.msr & MSR_DR)) {
 829                 unsigned int perm = hpte_get_skey_perm(gr, vcpu->arch.amr);
 830                 if (status & DSISR_ISSTORE)
 831                         perm >>= 1;
 832                 if (perm & 1)
 833                         return status | DSISR_KEYFAULT;
 834         }
 835
 836         /* Save HPTE info for virtual-mode handler */
 837         vcpu->arch.pgfault_addr = addr;
 838         vcpu->arch.pgfault_index = index;
 839         vcpu->arch.pgfault_hpte[0] = v;
 840         vcpu->arch.pgfault_hpte[1] = r;
 841
 842         /* Check the storage key to see if it is possibly emulated MMIO */
 843         if (data && (vcpu->arch.shregs.msr & MSR_IR) &&
 844             (r & (HPTE_R_KEY_HI | HPTE_R_KEY_LO)) ==
 845             (HPTE_R_KEY_HI | HPTE_R_KEY_LO))
 846                 return -2;      /* MMIO emulation - load instr word */
 847
 848         return -1;              /* send fault up to host kernel mode */
 849 }