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 */ 25static void *real_vmalloc_addr(void *x) 26{ 27 unsigned long addr = (unsigned long) x; 28 pte_t *p; 29 /* 30 * assume we don't have huge pages in vmalloc space... 31 * So don't worry about THP collapse/split. Called 32 * Only in realmode, hence won't need irq_save/restore. 33 */ 34 p = __find_linux_pte_or_hugepte(swapper_pg_dir, addr, NULL); 35 if (!p || !pte_present(*p)) 36 return NULL; 37 addr = (pte_pfn(*p) << PAGE_SHIFT) | (addr & ~PAGE_MASK); 38 return __va(addr); 39} 40 41/* Return 1 if we need to do a global tlbie, 0 if we can use tlbiel */ 42static int global_invalidates(struct kvm *kvm, unsigned long flags) 43{ 44 int global; 45 46 /* 47 * If there is only one vcore, and it's currently running, 48 * as indicated by local_paca->kvm_hstate.kvm_vcpu being set, 49 * we can use tlbiel as long as we mark all other physical 50 * cores as potentially having stale TLB entries for this lpid. 51 * Otherwise, don't use tlbiel. 52 */ 53 if (kvm->arch.online_vcores == 1 && local_paca->kvm_hstate.kvm_vcpu) 54 global = 0; 55 else 56 global = 1; 57 58 if (!global) { 59 /* any other core might now have stale TLB entries... */ 60 smp_wmb(); 61 cpumask_setall(&kvm->arch.need_tlb_flush); 62 cpumask_clear_cpu(local_paca->kvm_hstate.kvm_vcore->pcpu, 63 &kvm->arch.need_tlb_flush); 64 } 65 66 return global; 67} 68 69/* 70 * Add this HPTE into the chain for the real page. 71 * Must be called with the chain locked; it unlocks the chain. 72 */ 73void kvmppc_add_revmap_chain(struct kvm *kvm, struct revmap_entry *rev, 74 unsigned long *rmap, long pte_index, int realmode) 75{ 76 struct revmap_entry *head, *tail; 77 unsigned long i; 78 79 if (*rmap & KVMPPC_RMAP_PRESENT) { 80 i = *rmap & KVMPPC_RMAP_INDEX; 81 head = &kvm->arch.revmap[i]; 82 if (realmode) 83 head = real_vmalloc_addr(head); 84 tail = &kvm->arch.revmap[head->back]; 85 if (realmode) 86 tail = real_vmalloc_addr(tail); 87 rev->forw = i; 88 rev->back = head->back; 89 tail->forw = pte_index; 90 head->back = pte_index; 91 } else { 92 rev->forw = rev->back = pte_index; 93 *rmap = (*rmap & ~KVMPPC_RMAP_INDEX) | 94 pte_index | KVMPPC_RMAP_PRESENT; 95 } 96 unlock_rmap(rmap); 97} 98EXPORT_SYMBOL_GPL(kvmppc_add_revmap_chain); 99 100/* Remove this HPTE from the chain for a real page */ 101static void remove_revmap_chain(struct kvm *kvm, long pte_index, 102 struct revmap_entry *rev, 103 unsigned long hpte_v, unsigned long hpte_r) 104{ 105 struct revmap_entry *next, *prev; 106 unsigned long gfn, ptel, head; 107 struct kvm_memory_slot *memslot; 108 unsigned long *rmap; 109 unsigned long rcbits; 110 111 rcbits = hpte_r & (HPTE_R_R | HPTE_R_C); 112 ptel = rev->guest_rpte |= rcbits; 113 gfn = hpte_rpn(ptel, hpte_page_size(hpte_v, ptel)); 114 memslot = __gfn_to_memslot(kvm_memslots_raw(kvm), gfn); 115 if (!memslot) 116 return; 117 118 rmap = real_vmalloc_addr(&memslot->arch.rmap[gfn - memslot->base_gfn]); 119 lock_rmap(rmap); 120 121 head = *rmap & KVMPPC_RMAP_INDEX; 122 next = real_vmalloc_addr(&kvm->arch.revmap[rev->forw]); 123 prev = real_vmalloc_addr(&kvm->arch.revmap[rev->back]); 124 next->back = rev->back; 125 prev->forw = rev->forw; 126 if (head == pte_index) { 127 head = rev->forw; 128 if (head == pte_index) 129 *rmap &= ~(KVMPPC_RMAP_PRESENT | KVMPPC_RMAP_INDEX); 130 else 131 *rmap = (*rmap & ~KVMPPC_RMAP_INDEX) | head; 132 } 133 *rmap |= rcbits << KVMPPC_RMAP_RC_SHIFT; 134 unlock_rmap(rmap); 135} 136 137long kvmppc_do_h_enter(struct kvm *kvm, unsigned long flags, 138 long pte_index, unsigned long pteh, unsigned long ptel, 139 pgd_t *pgdir, bool realmode, unsigned long *pte_idx_ret) 140{ 141 unsigned long i, pa, gpa, gfn, psize; 142 unsigned long slot_fn, hva; 143 __be64 *hpte; 144 struct revmap_entry *rev; 145 unsigned long g_ptel; 146 struct kvm_memory_slot *memslot; 147 unsigned hpage_shift; 148 unsigned long is_io; 149 unsigned long *rmap; 150 pte_t *ptep; 151 unsigned int writing; 152 unsigned long mmu_seq; 153 unsigned long rcbits, irq_flags = 0; 154 155 psize = hpte_page_size(pteh, ptel); 156 if (!psize) 157 return H_PARAMETER; 158 writing = hpte_is_writable(ptel); 159 pteh &= ~(HPTE_V_HVLOCK | HPTE_V_ABSENT | HPTE_V_VALID); 160 ptel &= ~HPTE_GR_RESERVED; 161 g_ptel = ptel; 162 163 /* used later to detect if we might have been invalidated */ 164 mmu_seq = kvm->mmu_notifier_seq; 165 smp_rmb(); 166 167 /* Find the memslot (if any) for this address */ 168 gpa = (ptel & HPTE_R_RPN) & ~(psize - 1); 169 gfn = gpa >> PAGE_SHIFT; 170 memslot = __gfn_to_memslot(kvm_memslots_raw(kvm), gfn); 171 pa = 0; 172 is_io = ~0ul; 173 rmap = NULL; 174 if (!(memslot && !(memslot->flags & KVM_MEMSLOT_INVALID))) { 175 /* Emulated MMIO - mark this with key=31 */ 176 pteh |= HPTE_V_ABSENT; 177 ptel |= HPTE_R_KEY_HI | HPTE_R_KEY_LO; 178 goto do_insert; 179 } 180 181 /* Check if the requested page fits entirely in the memslot. */ 182 if (!slot_is_aligned(memslot, psize)) 183 return H_PARAMETER; 184 slot_fn = gfn - memslot->base_gfn; 185 rmap = &memslot->arch.rmap[slot_fn]; 186 187 /* Translate to host virtual address */ 188 hva = __gfn_to_hva_memslot(memslot, gfn); 189 /* 190 * If we had a page table table change after lookup, we would 191 * retry via mmu_notifier_retry. 192 */ 193 if (realmode) 194 ptep = __find_linux_pte_or_hugepte(pgdir, hva, &hpage_shift); 195 else { 196 local_irq_save(irq_flags); 197 ptep = find_linux_pte_or_hugepte(pgdir, hva, &hpage_shift); 198 } 199 if (ptep) { 200 pte_t pte; 201 unsigned int host_pte_size; 202 203 if (hpage_shift) 204 host_pte_size = 1ul << hpage_shift; 205 else 206 host_pte_size = PAGE_SIZE; 207 /* 208 * We should always find the guest page size 209 * to <= host page size, if host is using hugepage 210 */ 211 if (host_pte_size < psize) { 212 if (!realmode) 213 local_irq_restore(flags); 214 return H_PARAMETER; 215 } 216 pte = kvmppc_read_update_linux_pte(ptep, writing); 217 if (pte_present(pte) && !pte_protnone(pte)) { 218 if (writing && !pte_write(pte)) 219 /* make the actual HPTE be read-only */ 220 ptel = hpte_make_readonly(ptel); 221 is_io = hpte_cache_bits(pte_val(pte)); 222 pa = pte_pfn(pte) << PAGE_SHIFT; 223 pa |= hva & (host_pte_size - 1); 224 pa |= gpa & ~PAGE_MASK; 225 } 226 } 227 if (!realmode) 228 local_irq_restore(irq_flags); 229 230 ptel &= ~(HPTE_R_PP0 - psize); 231 ptel |= pa; 232 233 if (pa) 234 pteh |= HPTE_V_VALID; 235 else 236 pteh |= HPTE_V_ABSENT; 237 238 /* Check WIMG */ 239 if (is_io != ~0ul && !hpte_cache_flags_ok(ptel, is_io)) { 240 if (is_io) 241 return H_PARAMETER; 242 /* 243 * Allow guest to map emulated device memory as 244 * uncacheable, but actually make it cacheable. 245 */ 246 ptel &= ~(HPTE_R_W|HPTE_R_I|HPTE_R_G); 247 ptel |= HPTE_R_M; 248 } 249 250 /* Find and lock the HPTEG slot to use */ 251 do_insert: 252 if (pte_index >= kvm->arch.hpt_npte) 253 return H_PARAMETER; 254 if (likely((flags & H_EXACT) == 0)) { 255 pte_index &= ~7UL; 256 hpte = (__be64 *)(kvm->arch.hpt_virt + (pte_index << 4)); 257 for (i = 0; i < 8; ++i) { 258 if ((be64_to_cpu(*hpte) & HPTE_V_VALID) == 0 && 259 try_lock_hpte(hpte, HPTE_V_HVLOCK | HPTE_V_VALID | 260 HPTE_V_ABSENT)) 261 break; 262 hpte += 2; 263 } 264 if (i == 8) { 265 /* 266 * Since try_lock_hpte doesn't retry (not even stdcx. 267 * failures), it could be that there is a free slot 268 * but we transiently failed to lock it. Try again, 269 * actually locking each slot and checking it. 270 */ 271 hpte -= 16; 272 for (i = 0; i < 8; ++i) { 273 u64 pte; 274 while (!try_lock_hpte(hpte, HPTE_V_HVLOCK)) 275 cpu_relax(); 276 pte = be64_to_cpu(hpte[0]); 277 if (!(pte & (HPTE_V_VALID | HPTE_V_ABSENT))) 278 break; 279 __unlock_hpte(hpte, pte); 280 hpte += 2; 281 } 282 if (i == 8) 283 return H_PTEG_FULL; 284 } 285 pte_index += i; 286 } else { 287 hpte = (__be64 *)(kvm->arch.hpt_virt + (pte_index << 4)); 288 if (!try_lock_hpte(hpte, HPTE_V_HVLOCK | HPTE_V_VALID | 289 HPTE_V_ABSENT)) { 290 /* Lock the slot and check again */ 291 u64 pte; 292 293 while (!try_lock_hpte(hpte, HPTE_V_HVLOCK)) 294 cpu_relax(); 295 pte = be64_to_cpu(hpte[0]); 296 if (pte & (HPTE_V_VALID | HPTE_V_ABSENT)) { 297 __unlock_hpte(hpte, pte); 298 return H_PTEG_FULL; 299 } 300 } 301 } 302 303 /* Save away the guest's idea of the second HPTE dword */ 304 rev = &kvm->arch.revmap[pte_index]; 305 if (realmode) 306 rev = real_vmalloc_addr(rev); 307 if (rev) { 308 rev->guest_rpte = g_ptel; 309 note_hpte_modification(kvm, rev); 310 } 311 312 /* Link HPTE into reverse-map chain */ 313 if (pteh & HPTE_V_VALID) { 314 if (realmode) 315 rmap = real_vmalloc_addr(rmap); 316 lock_rmap(rmap); 317 /* Check for pending invalidations under the rmap chain lock */ 318 if (mmu_notifier_retry(kvm, mmu_seq)) { 319 /* inval in progress, write a non-present HPTE */ 320 pteh |= HPTE_V_ABSENT; 321 pteh &= ~HPTE_V_VALID; 322 unlock_rmap(rmap); 323 } else { 324 kvmppc_add_revmap_chain(kvm, rev, rmap, pte_index, 325 realmode); 326 /* Only set R/C in real HPTE if already set in *rmap */ 327 rcbits = *rmap >> KVMPPC_RMAP_RC_SHIFT; 328 ptel &= rcbits | ~(HPTE_R_R | HPTE_R_C); 329 } 330 } 331 332 hpte[1] = cpu_to_be64(ptel); 333 334 /* Write the first HPTE dword, unlocking the HPTE and making it valid */ 335 eieio(); 336 __unlock_hpte(hpte, pteh); 337 asm volatile("ptesync" : : : "memory"); 338 339 *pte_idx_ret = pte_index; 340 return H_SUCCESS; 341} 342EXPORT_SYMBOL_GPL(kvmppc_do_h_enter); 343 344long kvmppc_h_enter(struct kvm_vcpu *vcpu, unsigned long flags, 345 long pte_index, unsigned long pteh, unsigned long ptel) 346{ 347 return kvmppc_do_h_enter(vcpu->kvm, flags, pte_index, pteh, ptel, 348 vcpu->arch.pgdir, true, &vcpu->arch.gpr[4]); 349} 350 351#ifdef __BIG_ENDIAN__ 352#define LOCK_TOKEN (*(u32 *)(&get_paca()->lock_token)) 353#else 354#define LOCK_TOKEN (*(u32 *)(&get_paca()->paca_index)) 355#endif 356 357static inline int try_lock_tlbie(unsigned int *lock) 358{ 359 unsigned int tmp, old; 360 unsigned int token = LOCK_TOKEN; 361 362 asm volatile("1:lwarx %1,0,%2\n" 363 " cmpwi cr0,%1,0\n" 364 " bne 2f\n" 365 " stwcx. %3,0,%2\n" 366 " bne- 1b\n" 367 " isync\n" 368 "2:" 369 : "=&r" (tmp), "=&r" (old) 370 : "r" (lock), "r" (token) 371 : "cc", "memory"); 372 return old == 0; 373} 374 375static void do_tlbies(struct kvm *kvm, unsigned long *rbvalues, 376 long npages, int global, bool need_sync) 377{ 378 long i; 379 380 if (global) { 381 while (!try_lock_tlbie(&kvm->arch.tlbie_lock)) 382 cpu_relax(); 383 if (need_sync) 384 asm volatile("ptesync" : : : "memory"); 385 for (i = 0; i < npages; ++i) 386 asm volatile(PPC_TLBIE(%1,%0) : : 387 "r" (rbvalues[i]), "r" (kvm->arch.lpid)); 388 asm volatile("eieio; tlbsync; ptesync" : : : "memory"); 389 kvm->arch.tlbie_lock = 0; 390 } else { 391 if (need_sync) 392 asm volatile("ptesync" : : : "memory"); 393 for (i = 0; i < npages; ++i) 394 asm volatile("tlbiel %0" : : "r" (rbvalues[i])); 395 asm volatile("ptesync" : : : "memory"); 396 } 397} 398 399long kvmppc_do_h_remove(struct kvm *kvm, unsigned long flags, 400 unsigned long pte_index, unsigned long avpn, 401 unsigned long *hpret) 402{ 403 __be64 *hpte; 404 unsigned long v, r, rb; 405 struct revmap_entry *rev; 406 u64 pte; 407 408 if (pte_index >= kvm->arch.hpt_npte) 409 return H_PARAMETER; 410 hpte = (__be64 *)(kvm->arch.hpt_virt + (pte_index << 4)); 411 while (!try_lock_hpte(hpte, HPTE_V_HVLOCK)) 412 cpu_relax(); 413 pte = be64_to_cpu(hpte[0]); 414 if ((pte & (HPTE_V_ABSENT | HPTE_V_VALID)) == 0 || 415 ((flags & H_AVPN) && (pte & ~0x7fUL) != avpn) || 416 ((flags & H_ANDCOND) && (pte & avpn) != 0)) { 417 __unlock_hpte(hpte, pte); 418 return H_NOT_FOUND; 419 } 420 421 rev = real_vmalloc_addr(&kvm->arch.revmap[pte_index]); 422 v = pte & ~HPTE_V_HVLOCK; 423 if (v & HPTE_V_VALID) { 424 hpte[0] &= ~cpu_to_be64(HPTE_V_VALID); 425 rb = compute_tlbie_rb(v, be64_to_cpu(hpte[1]), pte_index); 426 do_tlbies(kvm, &rb, 1, global_invalidates(kvm, flags), true); 427 /* 428 * The reference (R) and change (C) bits in a HPT 429 * entry can be set by hardware at any time up until 430 * the HPTE is invalidated and the TLB invalidation 431 * sequence has completed. This means that when 432 * removing a HPTE, we need to re-read the HPTE after 433 * the invalidation sequence has completed in order to 434 * obtain reliable values of R and C. 435 */ 436 remove_revmap_chain(kvm, pte_index, rev, v, 437 be64_to_cpu(hpte[1])); 438 } 439 r = rev->guest_rpte & ~HPTE_GR_RESERVED; 440 note_hpte_modification(kvm, rev); 441 unlock_hpte(hpte, 0); 442 443 hpret[0] = v; 444 hpret[1] = r; 445 return H_SUCCESS; 446} 447EXPORT_SYMBOL_GPL(kvmppc_do_h_remove); 448 449long kvmppc_h_remove(struct kvm_vcpu *vcpu, unsigned long flags, 450 unsigned long pte_index, unsigned long avpn) 451{ 452 return kvmppc_do_h_remove(vcpu->kvm, flags, pte_index, avpn, 453 &vcpu->arch.gpr[4]); 454} 455 456long kvmppc_h_bulk_remove(struct kvm_vcpu *vcpu) 457{ 458 struct kvm *kvm = vcpu->kvm; 459 unsigned long *args = &vcpu->arch.gpr[4]; 460 __be64 *hp, *hptes[4]; 461 unsigned long tlbrb[4]; 462 long int i, j, k, n, found, indexes[4]; 463 unsigned long flags, req, pte_index, rcbits; 464 int global; 465 long int ret = H_SUCCESS; 466 struct revmap_entry *rev, *revs[4]; 467 u64 hp0; 468 469 global = global_invalidates(kvm, 0); 470 for (i = 0; i < 4 && ret == H_SUCCESS; ) { 471 n = 0; 472 for (; i < 4; ++i) { 473 j = i * 2; 474 pte_index = args[j]; 475 flags = pte_index >> 56; 476 pte_index &= ((1ul << 56) - 1); 477 req = flags >> 6; 478 flags &= 3; 479 if (req == 3) { /* no more requests */ 480 i = 4; 481 break; 482 } 483 if (req != 1 || flags == 3 || 484 pte_index >= kvm->arch.hpt_npte) { 485 /* parameter error */ 486 args[j] = ((0xa0 | flags) << 56) + pte_index; 487 ret = H_PARAMETER; 488 break; 489 } 490 hp = (__be64 *) (kvm->arch.hpt_virt + (pte_index << 4)); 491 /* to avoid deadlock, don't spin except for first */ 492 if (!try_lock_hpte(hp, HPTE_V_HVLOCK)) { 493 if (n) 494 break; 495 while (!try_lock_hpte(hp, HPTE_V_HVLOCK)) 496 cpu_relax(); 497 } 498 found = 0; 499 hp0 = be64_to_cpu(hp[0]); 500 if (hp0 & (HPTE_V_ABSENT | HPTE_V_VALID)) { 501 switch (flags & 3) { 502 case 0: /* absolute */ 503 found = 1; 504 break; 505 case 1: /* andcond */ 506 if (!(hp0 & args[j + 1])) 507 found = 1; 508 break; 509 case 2: /* AVPN */ 510 if ((hp0 & ~0x7fUL) == args[j + 1]) 511 found = 1; 512 break; 513 } 514 } 515 if (!found) { 516 hp[0] &= ~cpu_to_be64(HPTE_V_HVLOCK); 517 args[j] = ((0x90 | flags) << 56) + pte_index; 518 continue; 519 } 520 521 args[j] = ((0x80 | flags) << 56) + pte_index; 522 rev = real_vmalloc_addr(&kvm->arch.revmap[pte_index]); 523 note_hpte_modification(kvm, rev); 524 525 if (!(hp0 & HPTE_V_VALID)) { 526 /* insert R and C bits from PTE */ 527 rcbits = rev->guest_rpte & (HPTE_R_R|HPTE_R_C); 528 args[j] |= rcbits << (56 - 5); 529 hp[0] = 0; 530 continue; 531 } 532 533 /* leave it locked */ 534 hp[0] &= ~cpu_to_be64(HPTE_V_VALID); 535 tlbrb[n] = compute_tlbie_rb(be64_to_cpu(hp[0]), 536 be64_to_cpu(hp[1]), pte_index); 537 indexes[n] = j; 538 hptes[n] = hp; 539 revs[n] = rev; 540 ++n; 541 } 542 543 if (!n) 544 break; 545 546 /* Now that we've collected a batch, do the tlbies */ 547 do_tlbies(kvm, tlbrb, n, global, true); 548 549 /* Read PTE low words after tlbie to get final R/C values */ 550 for (k = 0; k < n; ++k) { 551 j = indexes[k]; 552 pte_index = args[j] & ((1ul << 56) - 1); 553 hp = hptes[k]; 554 rev = revs[k]; 555 remove_revmap_chain(kvm, pte_index, rev, 556 be64_to_cpu(hp[0]), be64_to_cpu(hp[1])); 557 rcbits = rev->guest_rpte & (HPTE_R_R|HPTE_R_C); 558 args[j] |= rcbits << (56 - 5); 559 __unlock_hpte(hp, 0); 560 } 561 } 562 563 return ret; 564} 565 566long kvmppc_h_protect(struct kvm_vcpu *vcpu, unsigned long flags, 567 unsigned long pte_index, unsigned long avpn, 568 unsigned long va) 569{ 570 struct kvm *kvm = vcpu->kvm; 571 __be64 *hpte; 572 struct revmap_entry *rev; 573 unsigned long v, r, rb, mask, bits; 574 u64 pte; 575 576 if (pte_index >= kvm->arch.hpt_npte) 577 return H_PARAMETER; 578 579 hpte = (__be64 *)(kvm->arch.hpt_virt + (pte_index << 4)); 580 while (!try_lock_hpte(hpte, HPTE_V_HVLOCK)) 581 cpu_relax(); 582 pte = be64_to_cpu(hpte[0]); 583 if ((pte & (HPTE_V_ABSENT | HPTE_V_VALID)) == 0 || 584 ((flags & H_AVPN) && (pte & ~0x7fUL) != avpn)) { 585 __unlock_hpte(hpte, pte); 586 return H_NOT_FOUND; 587 } 588 589 v = pte; 590 bits = (flags << 55) & HPTE_R_PP0; 591 bits |= (flags << 48) & HPTE_R_KEY_HI; 592 bits |= flags & (HPTE_R_PP | HPTE_R_N | HPTE_R_KEY_LO); 593 594 /* Update guest view of 2nd HPTE dword */ 595 mask = HPTE_R_PP0 | HPTE_R_PP | HPTE_R_N | 596 HPTE_R_KEY_HI | HPTE_R_KEY_LO; 597 rev = real_vmalloc_addr(&kvm->arch.revmap[pte_index]); 598 if (rev) { 599 r = (rev->guest_rpte & ~mask) | bits; 600 rev->guest_rpte = r; 601 note_hpte_modification(kvm, rev); 602 } 603 604 /* Update HPTE */ 605 if (v & HPTE_V_VALID) { 606 /* 607 * If the page is valid, don't let it transition from 608 * readonly to writable. If it should be writable, we'll 609 * take a trap and let the page fault code sort it out. 610 */ 611 pte = be64_to_cpu(hpte[1]); 612 r = (pte & ~mask) | bits; 613 if (hpte_is_writable(r) && !hpte_is_writable(pte)) 614 r = hpte_make_readonly(r); 615 /* If the PTE is changing, invalidate it first */ 616 if (r != pte) { 617 rb = compute_tlbie_rb(v, r, pte_index); 618 hpte[0] = cpu_to_be64((v & ~HPTE_V_VALID) | 619 HPTE_V_ABSENT); 620 do_tlbies(kvm, &rb, 1, global_invalidates(kvm, flags), 621 true); 622 hpte[1] = cpu_to_be64(r); 623 } 624 } 625 unlock_hpte(hpte, v & ~HPTE_V_HVLOCK); 626 asm volatile("ptesync" : : : "memory"); 627 return H_SUCCESS; 628} 629 630long kvmppc_h_read(struct kvm_vcpu *vcpu, unsigned long flags, 631 unsigned long pte_index) 632{ 633 struct kvm *kvm = vcpu->kvm; 634 __be64 *hpte; 635 unsigned long v, r; 636 int i, n = 1; 637 struct revmap_entry *rev = NULL; 638 639 if (pte_index >= kvm->arch.hpt_npte) 640 return H_PARAMETER; 641 if (flags & H_READ_4) { 642 pte_index &= ~3; 643 n = 4; 644 } 645 rev = real_vmalloc_addr(&kvm->arch.revmap[pte_index]); 646 for (i = 0; i < n; ++i, ++pte_index) { 647 hpte = (__be64 *)(kvm->arch.hpt_virt + (pte_index << 4)); 648 v = be64_to_cpu(hpte[0]) & ~HPTE_V_HVLOCK; 649 r = be64_to_cpu(hpte[1]); 650 if (v & HPTE_V_ABSENT) { 651 v &= ~HPTE_V_ABSENT; 652 v |= HPTE_V_VALID; 653 } 654 if (v & HPTE_V_VALID) { 655 r = rev[i].guest_rpte | (r & (HPTE_R_R | HPTE_R_C)); 656 r &= ~HPTE_GR_RESERVED; 657 } 658 vcpu->arch.gpr[4 + i * 2] = v; 659 vcpu->arch.gpr[5 + i * 2] = r; 660 } 661 return H_SUCCESS; 662} 663 664void kvmppc_invalidate_hpte(struct kvm *kvm, __be64 *hptep, 665 unsigned long pte_index) 666{ 667 unsigned long rb; 668 669 hptep[0] &= ~cpu_to_be64(HPTE_V_VALID); 670 rb = compute_tlbie_rb(be64_to_cpu(hptep[0]), be64_to_cpu(hptep[1]), 671 pte_index); 672 do_tlbies(kvm, &rb, 1, 1, true); 673} 674EXPORT_SYMBOL_GPL(kvmppc_invalidate_hpte); 675 676void kvmppc_clear_ref_hpte(struct kvm *kvm, __be64 *hptep, 677 unsigned long pte_index) 678{ 679 unsigned long rb; 680 unsigned char rbyte; 681 682 rb = compute_tlbie_rb(be64_to_cpu(hptep[0]), be64_to_cpu(hptep[1]), 683 pte_index); 684 rbyte = (be64_to_cpu(hptep[1]) & ~HPTE_R_R) >> 8; 685 /* modify only the second-last byte, which contains the ref bit */ 686 *((char *)hptep + 14) = rbyte; 687 do_tlbies(kvm, &rb, 1, 1, false); 688} 689EXPORT_SYMBOL_GPL(kvmppc_clear_ref_hpte); 690 691static int slb_base_page_shift[4] = { 692 24, /* 16M */ 693 16, /* 64k */ 694 34, /* 16G */ 695 20, /* 1M, unsupported */ 696}; 697 698/* When called from virtmode, this func should be protected by 699 * preempt_disable(), otherwise, the holding of HPTE_V_HVLOCK 700 * can trigger deadlock issue. 701 */ 702long kvmppc_hv_find_lock_hpte(struct kvm *kvm, gva_t eaddr, unsigned long slb_v, 703 unsigned long valid) 704{ 705 unsigned int i; 706 unsigned int pshift; 707 unsigned long somask; 708 unsigned long vsid, hash; 709 unsigned long avpn; 710 __be64 *hpte; 711 unsigned long mask, val; 712 unsigned long v, r; 713 714 /* Get page shift, work out hash and AVPN etc. */ 715 mask = SLB_VSID_B | HPTE_V_AVPN | HPTE_V_SECONDARY; 716 val = 0; 717 pshift = 12; 718 if (slb_v & SLB_VSID_L) { 719 mask |= HPTE_V_LARGE; 720 val |= HPTE_V_LARGE; 721 pshift = slb_base_page_shift[(slb_v & SLB_VSID_LP) >> 4]; 722 } 723 if (slb_v & SLB_VSID_B_1T) { 724 somask = (1UL << 40) - 1; 725 vsid = (slb_v & ~SLB_VSID_B) >> SLB_VSID_SHIFT_1T; 726 vsid ^= vsid << 25; 727 } else { 728 somask = (1UL << 28) - 1; 729 vsid = (slb_v & ~SLB_VSID_B) >> SLB_VSID_SHIFT; 730 } 731 hash = (vsid ^ ((eaddr & somask) >> pshift)) & kvm->arch.hpt_mask; 732 avpn = slb_v & ~(somask >> 16); /* also includes B */ 733 avpn |= (eaddr & somask) >> 16; 734 735 if (pshift >= 24) 736 avpn &= ~((1UL << (pshift - 16)) - 1); 737 else 738 avpn &= ~0x7fUL; 739 val |= avpn; 740 741 for (;;) { 742 hpte = (__be64 *)(kvm->arch.hpt_virt + (hash << 7)); 743 744 for (i = 0; i < 16; i += 2) { 745 /* Read the PTE racily */ 746 v = be64_to_cpu(hpte[i]) & ~HPTE_V_HVLOCK; 747 748 /* Check valid/absent, hash, segment size and AVPN */ 749 if (!(v & valid) || (v & mask) != val) 750 continue; 751 752 /* Lock the PTE and read it under the lock */ 753 while (!try_lock_hpte(&hpte[i], HPTE_V_HVLOCK)) 754 cpu_relax(); 755 v = be64_to_cpu(hpte[i]) & ~HPTE_V_HVLOCK; 756 r = be64_to_cpu(hpte[i+1]); 757 758 /* 759 * Check the HPTE again, including base page size 760 */ 761 if ((v & valid) && (v & mask) == val && 762 hpte_base_page_size(v, r) == (1ul << pshift)) 763 /* Return with the HPTE still locked */ 764 return (hash << 3) + (i >> 1); 765 766 __unlock_hpte(&hpte[i], v); 767 } 768 769 if (val & HPTE_V_SECONDARY) 770 break; 771 val |= HPTE_V_SECONDARY; 772 hash = hash ^ kvm->arch.hpt_mask; 773 } 774 return -1; 775} 776EXPORT_SYMBOL(kvmppc_hv_find_lock_hpte); 777 778/* 779 * Called in real mode to check whether an HPTE not found fault 780 * is due to accessing a paged-out page or an emulated MMIO page, 781 * or if a protection fault is due to accessing a page that the 782 * guest wanted read/write access to but which we made read-only. 783 * Returns a possibly modified status (DSISR) value if not 784 * (i.e. pass the interrupt to the guest), 785 * -1 to pass the fault up to host kernel mode code, -2 to do that 786 * and also load the instruction word (for MMIO emulation), 787 * or 0 if we should make the guest retry the access. 788 */ 789long kvmppc_hpte_hv_fault(struct kvm_vcpu *vcpu, unsigned long addr, 790 unsigned long slb_v, unsigned int status, bool data) 791{ 792 struct kvm *kvm = vcpu->kvm; 793 long int index; 794 unsigned long v, r, gr; 795 __be64 *hpte; 796 unsigned long valid; 797 struct revmap_entry *rev; 798 unsigned long pp, key; 799 800 /* For protection fault, expect to find a valid HPTE */ 801 valid = HPTE_V_VALID; 802 if (status & DSISR_NOHPTE) 803 valid |= HPTE_V_ABSENT; 804 805 index = kvmppc_hv_find_lock_hpte(kvm, addr, slb_v, valid); 806 if (index < 0) { 807 if (status & DSISR_NOHPTE) 808 return status; /* there really was no HPTE */ 809 return 0; /* for prot fault, HPTE disappeared */ 810 } 811 hpte = (__be64 *)(kvm->arch.hpt_virt + (index << 4)); 812 v = be64_to_cpu(hpte[0]) & ~HPTE_V_HVLOCK; 813 r = be64_to_cpu(hpte[1]); 814 rev = real_vmalloc_addr(&kvm->arch.revmap[index]); 815 gr = rev->guest_rpte; 816 817 unlock_hpte(hpte, v); 818 819 /* For not found, if the HPTE is valid by now, retry the instruction */ 820 if ((status & DSISR_NOHPTE) && (v & HPTE_V_VALID)) 821 return 0; 822 823 /* Check access permissions to the page */ 824 pp = gr & (HPTE_R_PP0 | HPTE_R_PP); 825 key = (vcpu->arch.shregs.msr & MSR_PR) ? SLB_VSID_KP : SLB_VSID_KS; 826 status &= ~DSISR_NOHPTE; /* DSISR_NOHPTE == SRR1_ISI_NOPT */ 827 if (!data) { 828 if (gr & (HPTE_R_N | HPTE_R_G)) 829 return status | SRR1_ISI_N_OR_G; 830 if (!hpte_read_permission(pp, slb_v & key)) 831 return status | SRR1_ISI_PROT; 832 } else if (status & DSISR_ISSTORE) { 833 /* check write permission */ 834 if (!hpte_write_permission(pp, slb_v & key)) 835 return status | DSISR_PROTFAULT; 836 } else { 837 if (!hpte_read_permission(pp, slb_v & key)) 838 return status | DSISR_PROTFAULT; 839 } 840 841 /* Check storage key, if applicable */ 842 if (data && (vcpu->arch.shregs.msr & MSR_DR)) { 843 unsigned int perm = hpte_get_skey_perm(gr, vcpu->arch.amr); 844 if (status & DSISR_ISSTORE) 845 perm >>= 1; 846 if (perm & 1) 847 return status | DSISR_KEYFAULT; 848 } 849 850 /* Save HPTE info for virtual-mode handler */ 851 vcpu->arch.pgfault_addr = addr; 852 vcpu->arch.pgfault_index = index; 853 vcpu->arch.pgfault_hpte[0] = v; 854 vcpu->arch.pgfault_hpte[1] = r; 855 856 /* Check the storage key to see if it is possibly emulated MMIO */ 857 if (data && (vcpu->arch.shregs.msr & MSR_IR) && 858 (r & (HPTE_R_KEY_HI | HPTE_R_KEY_LO)) == 859 (HPTE_R_KEY_HI | HPTE_R_KEY_LO)) 860 return -2; /* MMIO emulation - load instr word */ 861 862 return -1; /* send fault up to host kernel mode */ 863} 864