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 2016 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 14 #include <asm/kvm_ppc.h> 15 #include <asm/kvm_book3s.h> 16 #include <asm/page.h> 17 #include <asm/mmu.h> 18 #include <asm/pgtable.h> 19 #include <asm/pgalloc.h> 20 21 /* 22 * Supported radix tree geometry. 23 * Like p9, we support either 5 or 9 bits at the first (lowest) level, 24 * for a page size of 64k or 4k. 25 */ 26 static int p9_supported_radix_bits[4] = { 5, 9, 9, 13 }; 27 28 int kvmppc_mmu_radix_xlate(struct kvm_vcpu *vcpu, gva_t eaddr, 29 struct kvmppc_pte *gpte, bool data, bool iswrite) 30 { 31 struct kvm *kvm = vcpu->kvm; 32 u32 pid; 33 int ret, level, ps; 34 __be64 prte, rpte; 35 unsigned long root, pte, index; 36 unsigned long rts, bits, offset; 37 unsigned long gpa; 38 unsigned long proc_tbl_size; 39 40 /* Work out effective PID */ 41 switch (eaddr >> 62) { 42 case 0: 43 pid = vcpu->arch.pid; 44 break; 45 case 3: 46 pid = 0; 47 break; 48 default: 49 return -EINVAL; 50 } 51 proc_tbl_size = 1 << ((kvm->arch.process_table & PRTS_MASK) + 12); 52 if (pid * 16 >= proc_tbl_size) 53 return -EINVAL; 54 55 /* Read partition table to find root of tree for effective PID */ 56 ret = kvm_read_guest(kvm, kvm->arch.process_table + pid * 16, 57 &prte, sizeof(prte)); 58 if (ret) 59 return ret; 60 61 root = be64_to_cpu(prte); 62 rts = ((root & RTS1_MASK) >> (RTS1_SHIFT - 3)) | 63 ((root & RTS2_MASK) >> RTS2_SHIFT); 64 bits = root & RPDS_MASK; 65 root = root & RPDB_MASK; 66 67 /* P9 DD1 interprets RTS (radix tree size) differently */ 68 offset = rts + 31; 69 if (cpu_has_feature(CPU_FTR_POWER9_DD1)) 70 offset -= 3; 71 72 /* current implementations only support 52-bit space */ 73 if (offset != 52) 74 return -EINVAL; 75 76 for (level = 3; level >= 0; --level) { 77 if (level && bits != p9_supported_radix_bits[level]) 78 return -EINVAL; 79 if (level == 0 && !(bits == 5 || bits == 9)) 80 return -EINVAL; 81 offset -= bits; 82 index = (eaddr >> offset) & ((1UL << bits) - 1); 83 /* check that low bits of page table base are zero */ 84 if (root & ((1UL << (bits + 3)) - 1)) 85 return -EINVAL; 86 ret = kvm_read_guest(kvm, root + index * 8, 87 &rpte, sizeof(rpte)); 88 if (ret) 89 return ret; 90 pte = __be64_to_cpu(rpte); 91 if (!(pte & _PAGE_PRESENT)) 92 return -ENOENT; 93 if (pte & _PAGE_PTE) 94 break; 95 bits = pte & 0x1f; 96 root = pte & 0x0fffffffffffff00ul; 97 } 98 /* need a leaf at lowest level; 512GB pages not supported */ 99 if (level < 0 || level == 3) 100 return -EINVAL; 101 102 /* offset is now log base 2 of the page size */ 103 gpa = pte & 0x01fffffffffff000ul; 104 if (gpa & ((1ul << offset) - 1)) 105 return -EINVAL; 106 gpa += eaddr & ((1ul << offset) - 1); 107 for (ps = MMU_PAGE_4K; ps < MMU_PAGE_COUNT; ++ps) 108 if (offset == mmu_psize_defs[ps].shift) 109 break; 110 gpte->page_size = ps; 111 112 gpte->eaddr = eaddr; 113 gpte->raddr = gpa; 114 115 /* Work out permissions */ 116 gpte->may_read = !!(pte & _PAGE_READ); 117 gpte->may_write = !!(pte & _PAGE_WRITE); 118 gpte->may_execute = !!(pte & _PAGE_EXEC); 119 if (kvmppc_get_msr(vcpu) & MSR_PR) { 120 if (pte & _PAGE_PRIVILEGED) { 121 gpte->may_read = 0; 122 gpte->may_write = 0; 123 gpte->may_execute = 0; 124 } 125 } else { 126 if (!(pte & _PAGE_PRIVILEGED)) { 127 /* Check AMR/IAMR to see if strict mode is in force */ 128 if (vcpu->arch.amr & (1ul << 62)) 129 gpte->may_read = 0; 130 if (vcpu->arch.amr & (1ul << 63)) 131 gpte->may_write = 0; 132 if (vcpu->arch.iamr & (1ul << 62)) 133 gpte->may_execute = 0; 134 } 135 } 136 137 return 0; 138 } 139 140 #ifdef CONFIG_PPC_64K_PAGES 141 #define MMU_BASE_PSIZE MMU_PAGE_64K 142 #else 143 #define MMU_BASE_PSIZE MMU_PAGE_4K 144 #endif 145 146 static void kvmppc_radix_tlbie_page(struct kvm *kvm, unsigned long addr, 147 unsigned int pshift) 148 { 149 int psize = MMU_BASE_PSIZE; 150 151 if (pshift >= PMD_SHIFT) 152 psize = MMU_PAGE_2M; 153 addr &= ~0xfffUL; 154 addr |= mmu_psize_defs[psize].ap << 5; 155 asm volatile("ptesync": : :"memory"); 156 asm volatile(PPC_TLBIE_5(%0, %1, 0, 0, 1) 157 : : "r" (addr), "r" (kvm->arch.lpid) : "memory"); 158 asm volatile("ptesync": : :"memory"); 159 } 160 161 unsigned long kvmppc_radix_update_pte(struct kvm *kvm, pte_t *ptep, 162 unsigned long clr, unsigned long set, 163 unsigned long addr, unsigned int shift) 164 { 165 unsigned long old = 0; 166 167 if (!(clr & _PAGE_PRESENT) && cpu_has_feature(CPU_FTR_POWER9_DD1) && 168 pte_present(*ptep)) { 169 /* have to invalidate it first */ 170 old = __radix_pte_update(ptep, _PAGE_PRESENT, 0); 171 kvmppc_radix_tlbie_page(kvm, addr, shift); 172 set |= _PAGE_PRESENT; 173 old &= _PAGE_PRESENT; 174 } 175 return __radix_pte_update(ptep, clr, set) | old; 176 } 177 178 void kvmppc_radix_set_pte_at(struct kvm *kvm, unsigned long addr, 179 pte_t *ptep, pte_t pte) 180 { 181 radix__set_pte_at(kvm->mm, addr, ptep, pte, 0); 182 } 183 184 static struct kmem_cache *kvm_pte_cache; 185 186 static pte_t *kvmppc_pte_alloc(void) 187 { 188 return kmem_cache_alloc(kvm_pte_cache, GFP_KERNEL); 189 } 190 191 static void kvmppc_pte_free(pte_t *ptep) 192 { 193 kmem_cache_free(kvm_pte_cache, ptep); 194 } 195 196 static int kvmppc_create_pte(struct kvm *kvm, pte_t pte, unsigned long gpa, 197 unsigned int level, unsigned long mmu_seq) 198 { 199 pgd_t *pgd; 200 pud_t *pud, *new_pud = NULL; 201 pmd_t *pmd, *new_pmd = NULL; 202 pte_t *ptep, *new_ptep = NULL; 203 unsigned long old; 204 int ret; 205 206 /* Traverse the guest's 2nd-level tree, allocate new levels needed */ 207 pgd = kvm->arch.pgtable + pgd_index(gpa); 208 pud = NULL; 209 if (pgd_present(*pgd)) 210 pud = pud_offset(pgd, gpa); 211 else 212 new_pud = pud_alloc_one(kvm->mm, gpa); 213 214 pmd = NULL; 215 if (pud && pud_present(*pud)) 216 pmd = pmd_offset(pud, gpa); 217 else 218 new_pmd = pmd_alloc_one(kvm->mm, gpa); 219 220 if (level == 0 && !(pmd && pmd_present(*pmd))) 221 new_ptep = kvmppc_pte_alloc(); 222 223 /* Check if we might have been invalidated; let the guest retry if so */ 224 spin_lock(&kvm->mmu_lock); 225 ret = -EAGAIN; 226 if (mmu_notifier_retry(kvm, mmu_seq)) 227 goto out_unlock; 228 229 /* Now traverse again under the lock and change the tree */ 230 ret = -ENOMEM; 231 if (pgd_none(*pgd)) { 232 if (!new_pud) 233 goto out_unlock; 234 pgd_populate(kvm->mm, pgd, new_pud); 235 new_pud = NULL; 236 } 237 pud = pud_offset(pgd, gpa); 238 if (pud_none(*pud)) { 239 if (!new_pmd) 240 goto out_unlock; 241 pud_populate(kvm->mm, pud, new_pmd); 242 new_pmd = NULL; 243 } 244 pmd = pmd_offset(pud, gpa); 245 if (pmd_large(*pmd)) { 246 /* Someone else has instantiated a large page here; retry */ 247 ret = -EAGAIN; 248 goto out_unlock; 249 } 250 if (level == 1 && !pmd_none(*pmd)) { 251 /* 252 * There's a page table page here, but we wanted 253 * to install a large page. Tell the caller and let 254 * it try installing a normal page if it wants. 255 */ 256 ret = -EBUSY; 257 goto out_unlock; 258 } 259 if (level == 0) { 260 if (pmd_none(*pmd)) { 261 if (!new_ptep) 262 goto out_unlock; 263 pmd_populate(kvm->mm, pmd, new_ptep); 264 new_ptep = NULL; 265 } 266 ptep = pte_offset_kernel(pmd, gpa); 267 if (pte_present(*ptep)) { 268 /* PTE was previously valid, so invalidate it */ 269 old = kvmppc_radix_update_pte(kvm, ptep, _PAGE_PRESENT, 270 0, gpa, 0); 271 kvmppc_radix_tlbie_page(kvm, gpa, 0); 272 if (old & _PAGE_DIRTY) 273 mark_page_dirty(kvm, gpa >> PAGE_SHIFT); 274 } 275 kvmppc_radix_set_pte_at(kvm, gpa, ptep, pte); 276 } else { 277 kvmppc_radix_set_pte_at(kvm, gpa, pmdp_ptep(pmd), pte); 278 } 279 ret = 0; 280 281 out_unlock: 282 spin_unlock(&kvm->mmu_lock); 283 if (new_pud) 284 pud_free(kvm->mm, new_pud); 285 if (new_pmd) 286 pmd_free(kvm->mm, new_pmd); 287 if (new_ptep) 288 kvmppc_pte_free(new_ptep); 289 return ret; 290 } 291 292 int kvmppc_book3s_radix_page_fault(struct kvm_run *run, struct kvm_vcpu *vcpu, 293 unsigned long ea, unsigned long dsisr) 294 { 295 struct kvm *kvm = vcpu->kvm; 296 unsigned long mmu_seq, pte_size; 297 unsigned long gpa, gfn, hva, pfn; 298 struct kvm_memory_slot *memslot; 299 struct page *page = NULL, *pages[1]; 300 long ret, npages, ok; 301 unsigned int writing; 302 struct vm_area_struct *vma; 303 unsigned long flags; 304 pte_t pte, *ptep; 305 unsigned long pgflags; 306 unsigned int shift, level; 307 308 /* Check for unusual errors */ 309 if (dsisr & DSISR_UNSUPP_MMU) { 310 pr_err("KVM: Got unsupported MMU fault\n"); 311 return -EFAULT; 312 } 313 if (dsisr & DSISR_BADACCESS) { 314 /* Reflect to the guest as DSI */ 315 pr_err("KVM: Got radix HV page fault with DSISR=%lx\n", dsisr); 316 kvmppc_core_queue_data_storage(vcpu, ea, dsisr); 317 return RESUME_GUEST; 318 } 319 320 /* Translate the logical address and get the page */ 321 gpa = vcpu->arch.fault_gpa & ~0xfffUL; 322 gpa &= ~0xF000000000000000ul; 323 gfn = gpa >> PAGE_SHIFT; 324 if (!(dsisr & DSISR_PGDIRFAULT)) 325 gpa |= ea & 0xfff; 326 memslot = gfn_to_memslot(kvm, gfn); 327 328 /* No memslot means it's an emulated MMIO region */ 329 if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID)) { 330 if (dsisr & (DSISR_PGDIRFAULT | DSISR_BADACCESS | 331 DSISR_SET_RC)) { 332 /* 333 * Bad address in guest page table tree, or other 334 * unusual error - reflect it to the guest as DSI. 335 */ 336 kvmppc_core_queue_data_storage(vcpu, ea, dsisr); 337 return RESUME_GUEST; 338 } 339 return kvmppc_hv_emulate_mmio(run, vcpu, gpa, ea, 340 dsisr & DSISR_ISSTORE); 341 } 342 343 /* used to check for invalidations in progress */ 344 mmu_seq = kvm->mmu_notifier_seq; 345 smp_rmb(); 346 347 writing = (dsisr & DSISR_ISSTORE) != 0; 348 hva = gfn_to_hva_memslot(memslot, gfn); 349 if (dsisr & DSISR_SET_RC) { 350 /* 351 * Need to set an R or C bit in the 2nd-level tables; 352 * if the relevant bits aren't already set in the linux 353 * page tables, fall through to do the gup_fast to 354 * set them in the linux page tables too. 355 */ 356 ok = 0; 357 pgflags = _PAGE_ACCESSED; 358 if (writing) 359 pgflags |= _PAGE_DIRTY; 360 local_irq_save(flags); 361 ptep = __find_linux_pte_or_hugepte(current->mm->pgd, hva, 362 NULL, NULL); 363 if (ptep) { 364 pte = READ_ONCE(*ptep); 365 if (pte_present(pte) && 366 (pte_val(pte) & pgflags) == pgflags) 367 ok = 1; 368 } 369 local_irq_restore(flags); 370 if (ok) { 371 spin_lock(&kvm->mmu_lock); 372 if (mmu_notifier_retry(vcpu->kvm, mmu_seq)) { 373 spin_unlock(&kvm->mmu_lock); 374 return RESUME_GUEST; 375 } 376 ptep = __find_linux_pte_or_hugepte(kvm->arch.pgtable, 377 gpa, NULL, &shift); 378 if (ptep && pte_present(*ptep)) { 379 kvmppc_radix_update_pte(kvm, ptep, 0, pgflags, 380 gpa, shift); 381 spin_unlock(&kvm->mmu_lock); 382 return RESUME_GUEST; 383 } 384 spin_unlock(&kvm->mmu_lock); 385 } 386 } 387 388 ret = -EFAULT; 389 pfn = 0; 390 pte_size = PAGE_SIZE; 391 pgflags = _PAGE_READ | _PAGE_EXEC; 392 level = 0; 393 npages = get_user_pages_fast(hva, 1, writing, pages); 394 if (npages < 1) { 395 /* Check if it's an I/O mapping */ 396 down_read(¤t->mm->mmap_sem); 397 vma = find_vma(current->mm, hva); 398 if (vma && vma->vm_start <= hva && hva < vma->vm_end && 399 (vma->vm_flags & VM_PFNMAP)) { 400 pfn = vma->vm_pgoff + 401 ((hva - vma->vm_start) >> PAGE_SHIFT); 402 pgflags = pgprot_val(vma->vm_page_prot); 403 } 404 up_read(¤t->mm->mmap_sem); 405 if (!pfn) 406 return -EFAULT; 407 } else { 408 page = pages[0]; 409 pfn = page_to_pfn(page); 410 if (PageHuge(page)) { 411 page = compound_head(page); 412 pte_size <<= compound_order(page); 413 /* See if we can insert a 2MB large-page PTE here */ 414 if (pte_size >= PMD_SIZE && 415 (gpa & PMD_MASK & PAGE_MASK) == 416 (hva & PMD_MASK & PAGE_MASK)) { 417 level = 1; 418 pfn &= ~((PMD_SIZE >> PAGE_SHIFT) - 1); 419 } 420 } 421 /* See if we can provide write access */ 422 if (writing) { 423 /* 424 * We assume gup_fast has set dirty on the host PTE. 425 */ 426 pgflags |= _PAGE_WRITE; 427 } else { 428 local_irq_save(flags); 429 ptep = __find_linux_pte_or_hugepte(current->mm->pgd, 430 hva, NULL, NULL); 431 if (ptep && pte_write(*ptep) && pte_dirty(*ptep)) 432 pgflags |= _PAGE_WRITE; 433 local_irq_restore(flags); 434 } 435 } 436 437 /* 438 * Compute the PTE value that we need to insert. 439 */ 440 pgflags |= _PAGE_PRESENT | _PAGE_PTE | _PAGE_ACCESSED; 441 if (pgflags & _PAGE_WRITE) 442 pgflags |= _PAGE_DIRTY; 443 pte = pfn_pte(pfn, __pgprot(pgflags)); 444 445 /* Allocate space in the tree and write the PTE */ 446 ret = kvmppc_create_pte(kvm, pte, gpa, level, mmu_seq); 447 if (ret == -EBUSY) { 448 /* 449 * There's already a PMD where wanted to install a large page; 450 * for now, fall back to installing a small page. 451 */ 452 level = 0; 453 pfn |= gfn & ((PMD_SIZE >> PAGE_SHIFT) - 1); 454 pte = pfn_pte(pfn, __pgprot(pgflags)); 455 ret = kvmppc_create_pte(kvm, pte, gpa, level, mmu_seq); 456 } 457 if (ret == 0 || ret == -EAGAIN) 458 ret = RESUME_GUEST; 459 460 if (page) { 461 /* 462 * We drop pages[0] here, not page because page might 463 * have been set to the head page of a compound, but 464 * we have to drop the reference on the correct tail 465 * page to match the get inside gup() 466 */ 467 put_page(pages[0]); 468 } 469 return ret; 470 } 471 472 static void mark_pages_dirty(struct kvm *kvm, struct kvm_memory_slot *memslot, 473 unsigned long gfn, unsigned int order) 474 { 475 unsigned long i, limit; 476 unsigned long *dp; 477 478 if (!memslot->dirty_bitmap) 479 return; 480 limit = 1ul << order; 481 if (limit < BITS_PER_LONG) { 482 for (i = 0; i < limit; ++i) 483 mark_page_dirty(kvm, gfn + i); 484 return; 485 } 486 dp = memslot->dirty_bitmap + (gfn - memslot->base_gfn); 487 limit /= BITS_PER_LONG; 488 for (i = 0; i < limit; ++i) 489 *dp++ = ~0ul; 490 } 491 492 /* Called with kvm->lock held */ 493 int kvm_unmap_radix(struct kvm *kvm, struct kvm_memory_slot *memslot, 494 unsigned long gfn) 495 { 496 pte_t *ptep; 497 unsigned long gpa = gfn << PAGE_SHIFT; 498 unsigned int shift; 499 unsigned long old; 500 501 ptep = __find_linux_pte_or_hugepte(kvm->arch.pgtable, gpa, 502 NULL, &shift); 503 if (ptep && pte_present(*ptep)) { 504 old = kvmppc_radix_update_pte(kvm, ptep, _PAGE_PRESENT, 0, 505 gpa, shift); 506 kvmppc_radix_tlbie_page(kvm, gpa, shift); 507 if (old & _PAGE_DIRTY) { 508 if (!shift) 509 mark_page_dirty(kvm, gfn); 510 else 511 mark_pages_dirty(kvm, memslot, 512 gfn, shift - PAGE_SHIFT); 513 } 514 } 515 return 0; 516 } 517 518 /* Called with kvm->lock held */ 519 int kvm_age_radix(struct kvm *kvm, struct kvm_memory_slot *memslot, 520 unsigned long gfn) 521 { 522 pte_t *ptep; 523 unsigned long gpa = gfn << PAGE_SHIFT; 524 unsigned int shift; 525 int ref = 0; 526 527 ptep = __find_linux_pte_or_hugepte(kvm->arch.pgtable, gpa, 528 NULL, &shift); 529 if (ptep && pte_present(*ptep) && pte_young(*ptep)) { 530 kvmppc_radix_update_pte(kvm, ptep, _PAGE_ACCESSED, 0, 531 gpa, shift); 532 /* XXX need to flush tlb here? */ 533 ref = 1; 534 } 535 return ref; 536 } 537 538 /* Called with kvm->lock held */ 539 int kvm_test_age_radix(struct kvm *kvm, struct kvm_memory_slot *memslot, 540 unsigned long gfn) 541 { 542 pte_t *ptep; 543 unsigned long gpa = gfn << PAGE_SHIFT; 544 unsigned int shift; 545 int ref = 0; 546 547 ptep = __find_linux_pte_or_hugepte(kvm->arch.pgtable, gpa, 548 NULL, &shift); 549 if (ptep && pte_present(*ptep) && pte_young(*ptep)) 550 ref = 1; 551 return ref; 552 } 553 554 /* Returns the number of PAGE_SIZE pages that are dirty */ 555 static int kvm_radix_test_clear_dirty(struct kvm *kvm, 556 struct kvm_memory_slot *memslot, int pagenum) 557 { 558 unsigned long gfn = memslot->base_gfn + pagenum; 559 unsigned long gpa = gfn << PAGE_SHIFT; 560 pte_t *ptep; 561 unsigned int shift; 562 int ret = 0; 563 564 ptep = __find_linux_pte_or_hugepte(kvm->arch.pgtable, gpa, 565 NULL, &shift); 566 if (ptep && pte_present(*ptep) && pte_dirty(*ptep)) { 567 ret = 1; 568 if (shift) 569 ret = 1 << (shift - PAGE_SHIFT); 570 kvmppc_radix_update_pte(kvm, ptep, _PAGE_DIRTY, 0, 571 gpa, shift); 572 kvmppc_radix_tlbie_page(kvm, gpa, shift); 573 } 574 return ret; 575 } 576 577 long kvmppc_hv_get_dirty_log_radix(struct kvm *kvm, 578 struct kvm_memory_slot *memslot, unsigned long *map) 579 { 580 unsigned long i, j; 581 unsigned long n, *p; 582 int npages; 583 584 /* 585 * Radix accumulates dirty bits in the first half of the 586 * memslot's dirty_bitmap area, for when pages are paged 587 * out or modified by the host directly. Pick up these 588 * bits and add them to the map. 589 */ 590 n = kvm_dirty_bitmap_bytes(memslot) / sizeof(long); 591 p = memslot->dirty_bitmap; 592 for (i = 0; i < n; ++i) 593 map[i] |= xchg(&p[i], 0); 594 595 for (i = 0; i < memslot->npages; i = j) { 596 npages = kvm_radix_test_clear_dirty(kvm, memslot, i); 597 598 /* 599 * Note that if npages > 0 then i must be a multiple of npages, 600 * since huge pages are only used to back the guest at guest 601 * real addresses that are a multiple of their size. 602 * Since we have at most one PTE covering any given guest 603 * real address, if npages > 1 we can skip to i + npages. 604 */ 605 j = i + 1; 606 if (npages) 607 for (j = i; npages; ++j, --npages) 608 __set_bit_le(j, map); 609 } 610 return 0; 611 } 612 613 static void add_rmmu_ap_encoding(struct kvm_ppc_rmmu_info *info, 614 int psize, int *indexp) 615 { 616 if (!mmu_psize_defs[psize].shift) 617 return; 618 info->ap_encodings[*indexp] = mmu_psize_defs[psize].shift | 619 (mmu_psize_defs[psize].ap << 29); 620 ++(*indexp); 621 } 622 623 int kvmhv_get_rmmu_info(struct kvm *kvm, struct kvm_ppc_rmmu_info *info) 624 { 625 int i; 626 627 if (!radix_enabled()) 628 return -EINVAL; 629 memset(info, 0, sizeof(*info)); 630 631 /* 4k page size */ 632 info->geometries[0].page_shift = 12; 633 info->geometries[0].level_bits[0] = 9; 634 for (i = 1; i < 4; ++i) 635 info->geometries[0].level_bits[i] = p9_supported_radix_bits[i]; 636 /* 64k page size */ 637 info->geometries[1].page_shift = 16; 638 for (i = 0; i < 4; ++i) 639 info->geometries[1].level_bits[i] = p9_supported_radix_bits[i]; 640 641 i = 0; 642 add_rmmu_ap_encoding(info, MMU_PAGE_4K, &i); 643 add_rmmu_ap_encoding(info, MMU_PAGE_64K, &i); 644 add_rmmu_ap_encoding(info, MMU_PAGE_2M, &i); 645 add_rmmu_ap_encoding(info, MMU_PAGE_1G, &i); 646 647 return 0; 648 } 649 650 int kvmppc_init_vm_radix(struct kvm *kvm) 651 { 652 kvm->arch.pgtable = pgd_alloc(kvm->mm); 653 if (!kvm->arch.pgtable) 654 return -ENOMEM; 655 return 0; 656 } 657 658 void kvmppc_free_radix(struct kvm *kvm) 659 { 660 unsigned long ig, iu, im; 661 pte_t *pte; 662 pmd_t *pmd; 663 pud_t *pud; 664 pgd_t *pgd; 665 666 if (!kvm->arch.pgtable) 667 return; 668 pgd = kvm->arch.pgtable; 669 for (ig = 0; ig < PTRS_PER_PGD; ++ig, ++pgd) { 670 if (!pgd_present(*pgd)) 671 continue; 672 pud = pud_offset(pgd, 0); 673 for (iu = 0; iu < PTRS_PER_PUD; ++iu, ++pud) { 674 if (!pud_present(*pud)) 675 continue; 676 pmd = pmd_offset(pud, 0); 677 for (im = 0; im < PTRS_PER_PMD; ++im, ++pmd) { 678 if (pmd_huge(*pmd)) { 679 pmd_clear(pmd); 680 continue; 681 } 682 if (!pmd_present(*pmd)) 683 continue; 684 pte = pte_offset_map(pmd, 0); 685 memset(pte, 0, sizeof(long) << PTE_INDEX_SIZE); 686 kvmppc_pte_free(pte); 687 pmd_clear(pmd); 688 } 689 pmd_free(kvm->mm, pmd_offset(pud, 0)); 690 pud_clear(pud); 691 } 692 pud_free(kvm->mm, pud_offset(pgd, 0)); 693 pgd_clear(pgd); 694 } 695 pgd_free(kvm->mm, kvm->arch.pgtable); 696 } 697 698 static void pte_ctor(void *addr) 699 { 700 memset(addr, 0, PTE_TABLE_SIZE); 701 } 702 703 int kvmppc_radix_init(void) 704 { 705 unsigned long size = sizeof(void *) << PTE_INDEX_SIZE; 706 707 kvm_pte_cache = kmem_cache_create("kvm-pte", size, size, 0, pte_ctor); 708 if (!kvm_pte_cache) 709 return -ENOMEM; 710 return 0; 711 } 712 713 void kvmppc_radix_exit(void) 714 { 715 kmem_cache_destroy(kvm_pte_cache); 716 } 717