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(swapper_pg_dir, addr); 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 /* 39 * Add this HPTE into the chain for the real page. 40 * Must be called with the chain locked; it unlocks the chain. 41 */ 42 void kvmppc_add_revmap_chain(struct kvm *kvm, struct revmap_entry *rev, 43 unsigned long *rmap, long pte_index, int realmode) 44 { 45 struct revmap_entry *head, *tail; 46 unsigned long i; 47 48 if (*rmap & KVMPPC_RMAP_PRESENT) { 49 i = *rmap & KVMPPC_RMAP_INDEX; 50 head = &kvm->arch.revmap[i]; 51 if (realmode) 52 head = real_vmalloc_addr(head); 53 tail = &kvm->arch.revmap[head->back]; 54 if (realmode) 55 tail = real_vmalloc_addr(tail); 56 rev->forw = i; 57 rev->back = head->back; 58 tail->forw = pte_index; 59 head->back = pte_index; 60 } else { 61 rev->forw = rev->back = pte_index; 62 i = pte_index; 63 } 64 smp_wmb(); 65 *rmap = i | KVMPPC_RMAP_REFERENCED | KVMPPC_RMAP_PRESENT; /* unlock */ 66 } 67 EXPORT_SYMBOL_GPL(kvmppc_add_revmap_chain); 68 69 /* Remove this HPTE from the chain for a real page */ 70 static void remove_revmap_chain(struct kvm *kvm, long pte_index, 71 struct revmap_entry *rev, 72 unsigned long hpte_v, unsigned long hpte_r) 73 { 74 struct revmap_entry *next, *prev; 75 unsigned long gfn, ptel, head; 76 struct kvm_memory_slot *memslot; 77 unsigned long *rmap; 78 unsigned long rcbits; 79 80 rcbits = hpte_r & (HPTE_R_R | HPTE_R_C); 81 ptel = rev->guest_rpte |= rcbits; 82 gfn = hpte_rpn(ptel, hpte_page_size(hpte_v, ptel)); 83 memslot = __gfn_to_memslot(kvm_memslots(kvm), gfn); 84 if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID)) 85 return; 86 87 rmap = real_vmalloc_addr(&memslot->rmap[gfn - memslot->base_gfn]); 88 lock_rmap(rmap); 89 90 head = *rmap & KVMPPC_RMAP_INDEX; 91 next = real_vmalloc_addr(&kvm->arch.revmap[rev->forw]); 92 prev = real_vmalloc_addr(&kvm->arch.revmap[rev->back]); 93 next->back = rev->back; 94 prev->forw = rev->forw; 95 if (head == pte_index) { 96 head = rev->forw; 97 if (head == pte_index) 98 *rmap &= ~(KVMPPC_RMAP_PRESENT | KVMPPC_RMAP_INDEX); 99 else 100 *rmap = (*rmap & ~KVMPPC_RMAP_INDEX) | head; 101 } 102 *rmap |= rcbits << KVMPPC_RMAP_RC_SHIFT; 103 unlock_rmap(rmap); 104 } 105 106 static pte_t lookup_linux_pte(struct kvm_vcpu *vcpu, unsigned long hva, 107 int writing, unsigned long *pte_sizep) 108 { 109 pte_t *ptep; 110 unsigned long ps = *pte_sizep; 111 unsigned int shift; 112 113 ptep = find_linux_pte_or_hugepte(vcpu->arch.pgdir, hva, &shift); 114 if (!ptep) 115 return __pte(0); 116 if (shift) 117 *pte_sizep = 1ul << shift; 118 else 119 *pte_sizep = PAGE_SIZE; 120 if (ps > *pte_sizep) 121 return __pte(0); 122 if (!pte_present(*ptep)) 123 return __pte(0); 124 return kvmppc_read_update_linux_pte(ptep, writing); 125 } 126 127 static inline void unlock_hpte(unsigned long *hpte, unsigned long hpte_v) 128 { 129 asm volatile(PPC_RELEASE_BARRIER "" : : : "memory"); 130 hpte[0] = hpte_v; 131 } 132 133 long kvmppc_h_enter(struct kvm_vcpu *vcpu, unsigned long flags, 134 long pte_index, unsigned long pteh, unsigned long ptel) 135 { 136 struct kvm *kvm = vcpu->kvm; 137 unsigned long i, pa, gpa, gfn, psize; 138 unsigned long slot_fn, hva; 139 unsigned long *hpte; 140 struct revmap_entry *rev; 141 unsigned long g_ptel = ptel; 142 struct kvm_memory_slot *memslot; 143 unsigned long *physp, pte_size; 144 unsigned long is_io; 145 unsigned long *rmap; 146 pte_t pte; 147 unsigned int writing; 148 unsigned long mmu_seq; 149 unsigned long rcbits; 150 bool realmode = vcpu->arch.vcore->vcore_state == VCORE_RUNNING; 151 152 psize = hpte_page_size(pteh, ptel); 153 if (!psize) 154 return H_PARAMETER; 155 writing = hpte_is_writable(ptel); 156 pteh &= ~(HPTE_V_HVLOCK | HPTE_V_ABSENT | HPTE_V_VALID); 157 158 /* used later to detect if we might have been invalidated */ 159 mmu_seq = kvm->mmu_notifier_seq; 160 smp_rmb(); 161 162 /* Find the memslot (if any) for this address */ 163 gpa = (ptel & HPTE_R_RPN) & ~(psize - 1); 164 gfn = gpa >> PAGE_SHIFT; 165 memslot = __gfn_to_memslot(kvm_memslots(kvm), gfn); 166 pa = 0; 167 is_io = ~0ul; 168 rmap = NULL; 169 if (!(memslot && !(memslot->flags & KVM_MEMSLOT_INVALID))) { 170 /* PPC970 can't do emulated MMIO */ 171 if (!cpu_has_feature(CPU_FTR_ARCH_206)) 172 return H_PARAMETER; 173 /* Emulated MMIO - mark this with key=31 */ 174 pteh |= HPTE_V_ABSENT; 175 ptel |= HPTE_R_KEY_HI | HPTE_R_KEY_LO; 176 goto do_insert; 177 } 178 179 /* Check if the requested page fits entirely in the memslot. */ 180 if (!slot_is_aligned(memslot, psize)) 181 return H_PARAMETER; 182 slot_fn = gfn - memslot->base_gfn; 183 rmap = &memslot->rmap[slot_fn]; 184 185 if (!kvm->arch.using_mmu_notifiers) { 186 physp = kvm->arch.slot_phys[memslot->id]; 187 if (!physp) 188 return H_PARAMETER; 189 physp += slot_fn; 190 if (realmode) 191 physp = real_vmalloc_addr(physp); 192 pa = *physp; 193 if (!pa) 194 return H_TOO_HARD; 195 is_io = pa & (HPTE_R_I | HPTE_R_W); 196 pte_size = PAGE_SIZE << (pa & KVMPPC_PAGE_ORDER_MASK); 197 pa &= PAGE_MASK; 198 } else { 199 /* Translate to host virtual address */ 200 hva = gfn_to_hva_memslot(memslot, gfn); 201 202 /* Look up the Linux PTE for the backing page */ 203 pte_size = psize; 204 pte = lookup_linux_pte(vcpu, hva, writing, &pte_size); 205 if (pte_present(pte)) { 206 if (writing && !pte_write(pte)) 207 /* make the actual HPTE be read-only */ 208 ptel = hpte_make_readonly(ptel); 209 is_io = hpte_cache_bits(pte_val(pte)); 210 pa = pte_pfn(pte) << PAGE_SHIFT; 211 } 212 } 213 if (pte_size < psize) 214 return H_PARAMETER; 215 if (pa && pte_size > psize) 216 pa |= gpa & (pte_size - 1); 217 218 ptel &= ~(HPTE_R_PP0 - psize); 219 ptel |= pa; 220 221 if (pa) 222 pteh |= HPTE_V_VALID; 223 else 224 pteh |= HPTE_V_ABSENT; 225 226 /* Check WIMG */ 227 if (is_io != ~0ul && !hpte_cache_flags_ok(ptel, is_io)) { 228 if (is_io) 229 return H_PARAMETER; 230 /* 231 * Allow guest to map emulated device memory as 232 * uncacheable, but actually make it cacheable. 233 */ 234 ptel &= ~(HPTE_R_W|HPTE_R_I|HPTE_R_G); 235 ptel |= HPTE_R_M; 236 } 237 238 /* Find and lock the HPTEG slot to use */ 239 do_insert: 240 if (pte_index >= HPT_NPTE) 241 return H_PARAMETER; 242 if (likely((flags & H_EXACT) == 0)) { 243 pte_index &= ~7UL; 244 hpte = (unsigned long *)(kvm->arch.hpt_virt + (pte_index << 4)); 245 for (i = 0; i < 8; ++i) { 246 if ((*hpte & HPTE_V_VALID) == 0 && 247 try_lock_hpte(hpte, HPTE_V_HVLOCK | HPTE_V_VALID | 248 HPTE_V_ABSENT)) 249 break; 250 hpte += 2; 251 } 252 if (i == 8) { 253 /* 254 * Since try_lock_hpte doesn't retry (not even stdcx. 255 * failures), it could be that there is a free slot 256 * but we transiently failed to lock it. Try again, 257 * actually locking each slot and checking it. 258 */ 259 hpte -= 16; 260 for (i = 0; i < 8; ++i) { 261 while (!try_lock_hpte(hpte, HPTE_V_HVLOCK)) 262 cpu_relax(); 263 if (!(*hpte & (HPTE_V_VALID | HPTE_V_ABSENT))) 264 break; 265 *hpte &= ~HPTE_V_HVLOCK; 266 hpte += 2; 267 } 268 if (i == 8) 269 return H_PTEG_FULL; 270 } 271 pte_index += i; 272 } else { 273 hpte = (unsigned long *)(kvm->arch.hpt_virt + (pte_index << 4)); 274 if (!try_lock_hpte(hpte, HPTE_V_HVLOCK | HPTE_V_VALID | 275 HPTE_V_ABSENT)) { 276 /* Lock the slot and check again */ 277 while (!try_lock_hpte(hpte, HPTE_V_HVLOCK)) 278 cpu_relax(); 279 if (*hpte & (HPTE_V_VALID | HPTE_V_ABSENT)) { 280 *hpte &= ~HPTE_V_HVLOCK; 281 return H_PTEG_FULL; 282 } 283 } 284 } 285 286 /* Save away the guest's idea of the second HPTE dword */ 287 rev = &kvm->arch.revmap[pte_index]; 288 if (realmode) 289 rev = real_vmalloc_addr(rev); 290 if (rev) 291 rev->guest_rpte = g_ptel; 292 293 /* Link HPTE into reverse-map chain */ 294 if (pteh & HPTE_V_VALID) { 295 if (realmode) 296 rmap = real_vmalloc_addr(rmap); 297 lock_rmap(rmap); 298 /* Check for pending invalidations under the rmap chain lock */ 299 if (kvm->arch.using_mmu_notifiers && 300 mmu_notifier_retry(vcpu, mmu_seq)) { 301 /* inval in progress, write a non-present HPTE */ 302 pteh |= HPTE_V_ABSENT; 303 pteh &= ~HPTE_V_VALID; 304 unlock_rmap(rmap); 305 } else { 306 kvmppc_add_revmap_chain(kvm, rev, rmap, pte_index, 307 realmode); 308 /* Only set R/C in real HPTE if already set in *rmap */ 309 rcbits = *rmap >> KVMPPC_RMAP_RC_SHIFT; 310 ptel &= rcbits | ~(HPTE_R_R | HPTE_R_C); 311 } 312 } 313 314 hpte[1] = ptel; 315 316 /* Write the first HPTE dword, unlocking the HPTE and making it valid */ 317 eieio(); 318 hpte[0] = pteh; 319 asm volatile("ptesync" : : : "memory"); 320 321 vcpu->arch.gpr[4] = pte_index; 322 return H_SUCCESS; 323 } 324 EXPORT_SYMBOL_GPL(kvmppc_h_enter); 325 326 #define LOCK_TOKEN (*(u32 *)(&get_paca()->lock_token)) 327 328 static inline int try_lock_tlbie(unsigned int *lock) 329 { 330 unsigned int tmp, old; 331 unsigned int token = LOCK_TOKEN; 332 333 asm volatile("1:lwarx %1,0,%2\n" 334 " cmpwi cr0,%1,0\n" 335 " bne 2f\n" 336 " stwcx. %3,0,%2\n" 337 " bne- 1b\n" 338 " isync\n" 339 "2:" 340 : "=&r" (tmp), "=&r" (old) 341 : "r" (lock), "r" (token) 342 : "cc", "memory"); 343 return old == 0; 344 } 345 346 long kvmppc_h_remove(struct kvm_vcpu *vcpu, unsigned long flags, 347 unsigned long pte_index, unsigned long avpn, 348 unsigned long va) 349 { 350 struct kvm *kvm = vcpu->kvm; 351 unsigned long *hpte; 352 unsigned long v, r, rb; 353 struct revmap_entry *rev; 354 355 if (pte_index >= HPT_NPTE) 356 return H_PARAMETER; 357 hpte = (unsigned long *)(kvm->arch.hpt_virt + (pte_index << 4)); 358 while (!try_lock_hpte(hpte, HPTE_V_HVLOCK)) 359 cpu_relax(); 360 if ((hpte[0] & (HPTE_V_ABSENT | HPTE_V_VALID)) == 0 || 361 ((flags & H_AVPN) && (hpte[0] & ~0x7fUL) != avpn) || 362 ((flags & H_ANDCOND) && (hpte[0] & avpn) != 0)) { 363 hpte[0] &= ~HPTE_V_HVLOCK; 364 return H_NOT_FOUND; 365 } 366 367 rev = real_vmalloc_addr(&kvm->arch.revmap[pte_index]); 368 v = hpte[0] & ~HPTE_V_HVLOCK; 369 if (v & HPTE_V_VALID) { 370 hpte[0] &= ~HPTE_V_VALID; 371 rb = compute_tlbie_rb(v, hpte[1], pte_index); 372 if (!(flags & H_LOCAL) && atomic_read(&kvm->online_vcpus) > 1) { 373 while (!try_lock_tlbie(&kvm->arch.tlbie_lock)) 374 cpu_relax(); 375 asm volatile("ptesync" : : : "memory"); 376 asm volatile(PPC_TLBIE(%1,%0)"; eieio; tlbsync" 377 : : "r" (rb), "r" (kvm->arch.lpid)); 378 asm volatile("ptesync" : : : "memory"); 379 kvm->arch.tlbie_lock = 0; 380 } else { 381 asm volatile("ptesync" : : : "memory"); 382 asm volatile("tlbiel %0" : : "r" (rb)); 383 asm volatile("ptesync" : : : "memory"); 384 } 385 /* Read PTE low word after tlbie to get final R/C values */ 386 remove_revmap_chain(kvm, pte_index, rev, v, hpte[1]); 387 } 388 r = rev->guest_rpte; 389 unlock_hpte(hpte, 0); 390 391 vcpu->arch.gpr[4] = v; 392 vcpu->arch.gpr[5] = r; 393 return H_SUCCESS; 394 } 395 396 long kvmppc_h_bulk_remove(struct kvm_vcpu *vcpu) 397 { 398 struct kvm *kvm = vcpu->kvm; 399 unsigned long *args = &vcpu->arch.gpr[4]; 400 unsigned long *hp, *hptes[4], tlbrb[4]; 401 long int i, j, k, n, found, indexes[4]; 402 unsigned long flags, req, pte_index, rcbits; 403 long int local = 0; 404 long int ret = H_SUCCESS; 405 struct revmap_entry *rev, *revs[4]; 406 407 if (atomic_read(&kvm->online_vcpus) == 1) 408 local = 1; 409 for (i = 0; i < 4 && ret == H_SUCCESS; ) { 410 n = 0; 411 for (; i < 4; ++i) { 412 j = i * 2; 413 pte_index = args[j]; 414 flags = pte_index >> 56; 415 pte_index &= ((1ul << 56) - 1); 416 req = flags >> 6; 417 flags &= 3; 418 if (req == 3) { /* no more requests */ 419 i = 4; 420 break; 421 } 422 if (req != 1 || flags == 3 || pte_index >= HPT_NPTE) { 423 /* parameter error */ 424 args[j] = ((0xa0 | flags) << 56) + pte_index; 425 ret = H_PARAMETER; 426 break; 427 } 428 hp = (unsigned long *) 429 (kvm->arch.hpt_virt + (pte_index << 4)); 430 /* to avoid deadlock, don't spin except for first */ 431 if (!try_lock_hpte(hp, HPTE_V_HVLOCK)) { 432 if (n) 433 break; 434 while (!try_lock_hpte(hp, HPTE_V_HVLOCK)) 435 cpu_relax(); 436 } 437 found = 0; 438 if (hp[0] & (HPTE_V_ABSENT | HPTE_V_VALID)) { 439 switch (flags & 3) { 440 case 0: /* absolute */ 441 found = 1; 442 break; 443 case 1: /* andcond */ 444 if (!(hp[0] & args[j + 1])) 445 found = 1; 446 break; 447 case 2: /* AVPN */ 448 if ((hp[0] & ~0x7fUL) == args[j + 1]) 449 found = 1; 450 break; 451 } 452 } 453 if (!found) { 454 hp[0] &= ~HPTE_V_HVLOCK; 455 args[j] = ((0x90 | flags) << 56) + pte_index; 456 continue; 457 } 458 459 args[j] = ((0x80 | flags) << 56) + pte_index; 460 rev = real_vmalloc_addr(&kvm->arch.revmap[pte_index]); 461 462 if (!(hp[0] & HPTE_V_VALID)) { 463 /* insert R and C bits from PTE */ 464 rcbits = rev->guest_rpte & (HPTE_R_R|HPTE_R_C); 465 args[j] |= rcbits << (56 - 5); 466 hp[0] = 0; 467 continue; 468 } 469 470 hp[0] &= ~HPTE_V_VALID; /* leave it locked */ 471 tlbrb[n] = compute_tlbie_rb(hp[0], hp[1], pte_index); 472 indexes[n] = j; 473 hptes[n] = hp; 474 revs[n] = rev; 475 ++n; 476 } 477 478 if (!n) 479 break; 480 481 /* Now that we've collected a batch, do the tlbies */ 482 if (!local) { 483 while(!try_lock_tlbie(&kvm->arch.tlbie_lock)) 484 cpu_relax(); 485 asm volatile("ptesync" : : : "memory"); 486 for (k = 0; k < n; ++k) 487 asm volatile(PPC_TLBIE(%1,%0) : : 488 "r" (tlbrb[k]), 489 "r" (kvm->arch.lpid)); 490 asm volatile("eieio; tlbsync; ptesync" : : : "memory"); 491 kvm->arch.tlbie_lock = 0; 492 } else { 493 asm volatile("ptesync" : : : "memory"); 494 for (k = 0; k < n; ++k) 495 asm volatile("tlbiel %0" : : "r" (tlbrb[k])); 496 asm volatile("ptesync" : : : "memory"); 497 } 498 499 /* Read PTE low words after tlbie to get final R/C values */ 500 for (k = 0; k < n; ++k) { 501 j = indexes[k]; 502 pte_index = args[j] & ((1ul << 56) - 1); 503 hp = hptes[k]; 504 rev = revs[k]; 505 remove_revmap_chain(kvm, pte_index, rev, hp[0], hp[1]); 506 rcbits = rev->guest_rpte & (HPTE_R_R|HPTE_R_C); 507 args[j] |= rcbits << (56 - 5); 508 hp[0] = 0; 509 } 510 } 511 512 return ret; 513 } 514 515 long kvmppc_h_protect(struct kvm_vcpu *vcpu, unsigned long flags, 516 unsigned long pte_index, unsigned long avpn, 517 unsigned long va) 518 { 519 struct kvm *kvm = vcpu->kvm; 520 unsigned long *hpte; 521 struct revmap_entry *rev; 522 unsigned long v, r, rb, mask, bits; 523 524 if (pte_index >= HPT_NPTE) 525 return H_PARAMETER; 526 527 hpte = (unsigned long *)(kvm->arch.hpt_virt + (pte_index << 4)); 528 while (!try_lock_hpte(hpte, HPTE_V_HVLOCK)) 529 cpu_relax(); 530 if ((hpte[0] & (HPTE_V_ABSENT | HPTE_V_VALID)) == 0 || 531 ((flags & H_AVPN) && (hpte[0] & ~0x7fUL) != avpn)) { 532 hpte[0] &= ~HPTE_V_HVLOCK; 533 return H_NOT_FOUND; 534 } 535 536 if (atomic_read(&kvm->online_vcpus) == 1) 537 flags |= H_LOCAL; 538 v = hpte[0]; 539 bits = (flags << 55) & HPTE_R_PP0; 540 bits |= (flags << 48) & HPTE_R_KEY_HI; 541 bits |= flags & (HPTE_R_PP | HPTE_R_N | HPTE_R_KEY_LO); 542 543 /* Update guest view of 2nd HPTE dword */ 544 mask = HPTE_R_PP0 | HPTE_R_PP | HPTE_R_N | 545 HPTE_R_KEY_HI | HPTE_R_KEY_LO; 546 rev = real_vmalloc_addr(&kvm->arch.revmap[pte_index]); 547 if (rev) { 548 r = (rev->guest_rpte & ~mask) | bits; 549 rev->guest_rpte = r; 550 } 551 r = (hpte[1] & ~mask) | bits; 552 553 /* Update HPTE */ 554 if (v & HPTE_V_VALID) { 555 rb = compute_tlbie_rb(v, r, pte_index); 556 hpte[0] = v & ~HPTE_V_VALID; 557 if (!(flags & H_LOCAL)) { 558 while(!try_lock_tlbie(&kvm->arch.tlbie_lock)) 559 cpu_relax(); 560 asm volatile("ptesync" : : : "memory"); 561 asm volatile(PPC_TLBIE(%1,%0)"; eieio; tlbsync" 562 : : "r" (rb), "r" (kvm->arch.lpid)); 563 asm volatile("ptesync" : : : "memory"); 564 kvm->arch.tlbie_lock = 0; 565 } else { 566 asm volatile("ptesync" : : : "memory"); 567 asm volatile("tlbiel %0" : : "r" (rb)); 568 asm volatile("ptesync" : : : "memory"); 569 } 570 } 571 hpte[1] = r; 572 eieio(); 573 hpte[0] = v & ~HPTE_V_HVLOCK; 574 asm volatile("ptesync" : : : "memory"); 575 return H_SUCCESS; 576 } 577 578 long kvmppc_h_read(struct kvm_vcpu *vcpu, unsigned long flags, 579 unsigned long pte_index) 580 { 581 struct kvm *kvm = vcpu->kvm; 582 unsigned long *hpte, v, r; 583 int i, n = 1; 584 struct revmap_entry *rev = NULL; 585 586 if (pte_index >= HPT_NPTE) 587 return H_PARAMETER; 588 if (flags & H_READ_4) { 589 pte_index &= ~3; 590 n = 4; 591 } 592 rev = real_vmalloc_addr(&kvm->arch.revmap[pte_index]); 593 for (i = 0; i < n; ++i, ++pte_index) { 594 hpte = (unsigned long *)(kvm->arch.hpt_virt + (pte_index << 4)); 595 v = hpte[0] & ~HPTE_V_HVLOCK; 596 r = hpte[1]; 597 if (v & HPTE_V_ABSENT) { 598 v &= ~HPTE_V_ABSENT; 599 v |= HPTE_V_VALID; 600 } 601 if (v & HPTE_V_VALID) 602 r = rev[i].guest_rpte | (r & (HPTE_R_R | HPTE_R_C)); 603 vcpu->arch.gpr[4 + i * 2] = v; 604 vcpu->arch.gpr[5 + i * 2] = r; 605 } 606 return H_SUCCESS; 607 } 608 609 void kvmppc_invalidate_hpte(struct kvm *kvm, unsigned long *hptep, 610 unsigned long pte_index) 611 { 612 unsigned long rb; 613 614 hptep[0] &= ~HPTE_V_VALID; 615 rb = compute_tlbie_rb(hptep[0], hptep[1], pte_index); 616 while (!try_lock_tlbie(&kvm->arch.tlbie_lock)) 617 cpu_relax(); 618 asm volatile("ptesync" : : : "memory"); 619 asm volatile(PPC_TLBIE(%1,%0)"; eieio; tlbsync" 620 : : "r" (rb), "r" (kvm->arch.lpid)); 621 asm volatile("ptesync" : : : "memory"); 622 kvm->arch.tlbie_lock = 0; 623 } 624 EXPORT_SYMBOL_GPL(kvmppc_invalidate_hpte); 625 626 void kvmppc_clear_ref_hpte(struct kvm *kvm, unsigned long *hptep, 627 unsigned long pte_index) 628 { 629 unsigned long rb; 630 unsigned char rbyte; 631 632 rb = compute_tlbie_rb(hptep[0], hptep[1], pte_index); 633 rbyte = (hptep[1] & ~HPTE_R_R) >> 8; 634 /* modify only the second-last byte, which contains the ref bit */ 635 *((char *)hptep + 14) = rbyte; 636 while (!try_lock_tlbie(&kvm->arch.tlbie_lock)) 637 cpu_relax(); 638 asm volatile(PPC_TLBIE(%1,%0)"; eieio; tlbsync" 639 : : "r" (rb), "r" (kvm->arch.lpid)); 640 asm volatile("ptesync" : : : "memory"); 641 kvm->arch.tlbie_lock = 0; 642 } 643 EXPORT_SYMBOL_GPL(kvmppc_clear_ref_hpte); 644 645 static int slb_base_page_shift[4] = { 646 24, /* 16M */ 647 16, /* 64k */ 648 34, /* 16G */ 649 20, /* 1M, unsupported */ 650 }; 651 652 long kvmppc_hv_find_lock_hpte(struct kvm *kvm, gva_t eaddr, unsigned long slb_v, 653 unsigned long valid) 654 { 655 unsigned int i; 656 unsigned int pshift; 657 unsigned long somask; 658 unsigned long vsid, hash; 659 unsigned long avpn; 660 unsigned long *hpte; 661 unsigned long mask, val; 662 unsigned long v, r; 663 664 /* Get page shift, work out hash and AVPN etc. */ 665 mask = SLB_VSID_B | HPTE_V_AVPN | HPTE_V_SECONDARY; 666 val = 0; 667 pshift = 12; 668 if (slb_v & SLB_VSID_L) { 669 mask |= HPTE_V_LARGE; 670 val |= HPTE_V_LARGE; 671 pshift = slb_base_page_shift[(slb_v & SLB_VSID_LP) >> 4]; 672 } 673 if (slb_v & SLB_VSID_B_1T) { 674 somask = (1UL << 40) - 1; 675 vsid = (slb_v & ~SLB_VSID_B) >> SLB_VSID_SHIFT_1T; 676 vsid ^= vsid << 25; 677 } else { 678 somask = (1UL << 28) - 1; 679 vsid = (slb_v & ~SLB_VSID_B) >> SLB_VSID_SHIFT; 680 } 681 hash = (vsid ^ ((eaddr & somask) >> pshift)) & HPT_HASH_MASK; 682 avpn = slb_v & ~(somask >> 16); /* also includes B */ 683 avpn |= (eaddr & somask) >> 16; 684 685 if (pshift >= 24) 686 avpn &= ~((1UL << (pshift - 16)) - 1); 687 else 688 avpn &= ~0x7fUL; 689 val |= avpn; 690 691 for (;;) { 692 hpte = (unsigned long *)(kvm->arch.hpt_virt + (hash << 7)); 693 694 for (i = 0; i < 16; i += 2) { 695 /* Read the PTE racily */ 696 v = hpte[i] & ~HPTE_V_HVLOCK; 697 698 /* Check valid/absent, hash, segment size and AVPN */ 699 if (!(v & valid) || (v & mask) != val) 700 continue; 701 702 /* Lock the PTE and read it under the lock */ 703 while (!try_lock_hpte(&hpte[i], HPTE_V_HVLOCK)) 704 cpu_relax(); 705 v = hpte[i] & ~HPTE_V_HVLOCK; 706 r = hpte[i+1]; 707 708 /* 709 * Check the HPTE again, including large page size 710 * Since we don't currently allow any MPSS (mixed 711 * page-size segment) page sizes, it is sufficient 712 * to check against the actual page size. 713 */ 714 if ((v & valid) && (v & mask) == val && 715 hpte_page_size(v, r) == (1ul << pshift)) 716 /* Return with the HPTE still locked */ 717 return (hash << 3) + (i >> 1); 718 719 /* Unlock and move on */ 720 hpte[i] = v; 721 } 722 723 if (val & HPTE_V_SECONDARY) 724 break; 725 val |= HPTE_V_SECONDARY; 726 hash = hash ^ HPT_HASH_MASK; 727 } 728 return -1; 729 } 730 EXPORT_SYMBOL(kvmppc_hv_find_lock_hpte); 731 732 /* 733 * Called in real mode to check whether an HPTE not found fault 734 * is due to accessing a paged-out page or an emulated MMIO page, 735 * or if a protection fault is due to accessing a page that the 736 * guest wanted read/write access to but which we made read-only. 737 * Returns a possibly modified status (DSISR) value if not 738 * (i.e. pass the interrupt to the guest), 739 * -1 to pass the fault up to host kernel mode code, -2 to do that 740 * and also load the instruction word (for MMIO emulation), 741 * or 0 if we should make the guest retry the access. 742 */ 743 long kvmppc_hpte_hv_fault(struct kvm_vcpu *vcpu, unsigned long addr, 744 unsigned long slb_v, unsigned int status, bool data) 745 { 746 struct kvm *kvm = vcpu->kvm; 747 long int index; 748 unsigned long v, r, gr; 749 unsigned long *hpte; 750 unsigned long valid; 751 struct revmap_entry *rev; 752 unsigned long pp, key; 753 754 /* For protection fault, expect to find a valid HPTE */ 755 valid = HPTE_V_VALID; 756 if (status & DSISR_NOHPTE) 757 valid |= HPTE_V_ABSENT; 758 759 index = kvmppc_hv_find_lock_hpte(kvm, addr, slb_v, valid); 760 if (index < 0) { 761 if (status & DSISR_NOHPTE) 762 return status; /* there really was no HPTE */ 763 return 0; /* for prot fault, HPTE disappeared */ 764 } 765 hpte = (unsigned long *)(kvm->arch.hpt_virt + (index << 4)); 766 v = hpte[0] & ~HPTE_V_HVLOCK; 767 r = hpte[1]; 768 rev = real_vmalloc_addr(&kvm->arch.revmap[index]); 769 gr = rev->guest_rpte; 770 771 unlock_hpte(hpte, v); 772 773 /* For not found, if the HPTE is valid by now, retry the instruction */ 774 if ((status & DSISR_NOHPTE) && (v & HPTE_V_VALID)) 775 return 0; 776 777 /* Check access permissions to the page */ 778 pp = gr & (HPTE_R_PP0 | HPTE_R_PP); 779 key = (vcpu->arch.shregs.msr & MSR_PR) ? SLB_VSID_KP : SLB_VSID_KS; 780 status &= ~DSISR_NOHPTE; /* DSISR_NOHPTE == SRR1_ISI_NOPT */ 781 if (!data) { 782 if (gr & (HPTE_R_N | HPTE_R_G)) 783 return status | SRR1_ISI_N_OR_G; 784 if (!hpte_read_permission(pp, slb_v & key)) 785 return status | SRR1_ISI_PROT; 786 } else if (status & DSISR_ISSTORE) { 787 /* check write permission */ 788 if (!hpte_write_permission(pp, slb_v & key)) 789 return status | DSISR_PROTFAULT; 790 } else { 791 if (!hpte_read_permission(pp, slb_v & key)) 792 return status | DSISR_PROTFAULT; 793 } 794 795 /* Check storage key, if applicable */ 796 if (data && (vcpu->arch.shregs.msr & MSR_DR)) { 797 unsigned int perm = hpte_get_skey_perm(gr, vcpu->arch.amr); 798 if (status & DSISR_ISSTORE) 799 perm >>= 1; 800 if (perm & 1) 801 return status | DSISR_KEYFAULT; 802 } 803 804 /* Save HPTE info for virtual-mode handler */ 805 vcpu->arch.pgfault_addr = addr; 806 vcpu->arch.pgfault_index = index; 807 vcpu->arch.pgfault_hpte[0] = v; 808 vcpu->arch.pgfault_hpte[1] = r; 809 810 /* Check the storage key to see if it is possibly emulated MMIO */ 811 if (data && (vcpu->arch.shregs.msr & MSR_IR) && 812 (r & (HPTE_R_KEY_HI | HPTE_R_KEY_LO)) == 813 (HPTE_R_KEY_HI | HPTE_R_KEY_LO)) 814 return -2; /* MMIO emulation - load instr word */ 815 816 return -1; /* send fault up to host kernel mode */ 817 } 818