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 #include <asm/pte-walk.h> 21 22 /* 23 * Supported radix tree geometry. 24 * Like p9, we support either 5 or 9 bits at the first (lowest) level, 25 * for a page size of 64k or 4k. 26 */ 27 static int p9_supported_radix_bits[4] = { 5, 9, 9, 13 }; 28 29 int kvmppc_mmu_radix_xlate(struct kvm_vcpu *vcpu, gva_t eaddr, 30 struct kvmppc_pte *gpte, bool data, bool iswrite) 31 { 32 struct kvm *kvm = vcpu->kvm; 33 u32 pid; 34 int ret, level, ps; 35 __be64 prte, rpte; 36 unsigned long ptbl; 37 unsigned long root, pte, index; 38 unsigned long rts, bits, offset; 39 unsigned long gpa; 40 unsigned long proc_tbl_size; 41 42 /* Work out effective PID */ 43 switch (eaddr >> 62) { 44 case 0: 45 pid = vcpu->arch.pid; 46 break; 47 case 3: 48 pid = 0; 49 break; 50 default: 51 return -EINVAL; 52 } 53 proc_tbl_size = 1 << ((kvm->arch.process_table & PRTS_MASK) + 12); 54 if (pid * 16 >= proc_tbl_size) 55 return -EINVAL; 56 57 /* Read partition table to find root of tree for effective PID */ 58 ptbl = (kvm->arch.process_table & PRTB_MASK) + (pid * 16); 59 ret = kvm_read_guest(kvm, ptbl, &prte, sizeof(prte)); 60 if (ret) 61 return ret; 62 63 root = be64_to_cpu(prte); 64 rts = ((root & RTS1_MASK) >> (RTS1_SHIFT - 3)) | 65 ((root & RTS2_MASK) >> RTS2_SHIFT); 66 bits = root & RPDS_MASK; 67 root = root & RPDB_MASK; 68 69 /* P9 DD1 interprets RTS (radix tree size) differently */ 70 offset = rts + 31; 71 if (cpu_has_feature(CPU_FTR_POWER9_DD1)) 72 offset -= 3; 73 74 /* current implementations only support 52-bit space */ 75 if (offset != 52) 76 return -EINVAL; 77 78 for (level = 3; level >= 0; --level) { 79 if (level && bits != p9_supported_radix_bits[level]) 80 return -EINVAL; 81 if (level == 0 && !(bits == 5 || bits == 9)) 82 return -EINVAL; 83 offset -= bits; 84 index = (eaddr >> offset) & ((1UL << bits) - 1); 85 /* check that low bits of page table base are zero */ 86 if (root & ((1UL << (bits + 3)) - 1)) 87 return -EINVAL; 88 ret = kvm_read_guest(kvm, root + index * 8, 89 &rpte, sizeof(rpte)); 90 if (ret) 91 return ret; 92 pte = __be64_to_cpu(rpte); 93 if (!(pte & _PAGE_PRESENT)) 94 return -ENOENT; 95 if (pte & _PAGE_PTE) 96 break; 97 bits = pte & 0x1f; 98 root = pte & 0x0fffffffffffff00ul; 99 } 100 /* need a leaf at lowest level; 512GB pages not supported */ 101 if (level < 0 || level == 3) 102 return -EINVAL; 103 104 /* offset is now log base 2 of the page size */ 105 gpa = pte & 0x01fffffffffff000ul; 106 if (gpa & ((1ul << offset) - 1)) 107 return -EINVAL; 108 gpa += eaddr & ((1ul << offset) - 1); 109 for (ps = MMU_PAGE_4K; ps < MMU_PAGE_COUNT; ++ps) 110 if (offset == mmu_psize_defs[ps].shift) 111 break; 112 gpte->page_size = ps; 113 114 gpte->eaddr = eaddr; 115 gpte->raddr = gpa; 116 117 /* Work out permissions */ 118 gpte->may_read = !!(pte & _PAGE_READ); 119 gpte->may_write = !!(pte & _PAGE_WRITE); 120 gpte->may_execute = !!(pte & _PAGE_EXEC); 121 if (kvmppc_get_msr(vcpu) & MSR_PR) { 122 if (pte & _PAGE_PRIVILEGED) { 123 gpte->may_read = 0; 124 gpte->may_write = 0; 125 gpte->may_execute = 0; 126 } 127 } else { 128 if (!(pte & _PAGE_PRIVILEGED)) { 129 /* Check AMR/IAMR to see if strict mode is in force */ 130 if (vcpu->arch.amr & (1ul << 62)) 131 gpte->may_read = 0; 132 if (vcpu->arch.amr & (1ul << 63)) 133 gpte->may_write = 0; 134 if (vcpu->arch.iamr & (1ul << 62)) 135 gpte->may_execute = 0; 136 } 137 } 138 139 return 0; 140 } 141 142 #ifdef CONFIG_PPC_64K_PAGES 143 #define MMU_BASE_PSIZE MMU_PAGE_64K 144 #else 145 #define MMU_BASE_PSIZE MMU_PAGE_4K 146 #endif 147 148 static void kvmppc_radix_tlbie_page(struct kvm *kvm, unsigned long addr, 149 unsigned int pshift) 150 { 151 int psize = MMU_BASE_PSIZE; 152 153 if (pshift >= PMD_SHIFT) 154 psize = MMU_PAGE_2M; 155 addr &= ~0xfffUL; 156 addr |= mmu_psize_defs[psize].ap << 5; 157 asm volatile("ptesync": : :"memory"); 158 asm volatile(PPC_TLBIE_5(%0, %1, 0, 0, 1) 159 : : "r" (addr), "r" (kvm->arch.lpid) : "memory"); 160 if (cpu_has_feature(CPU_FTR_P9_TLBIE_BUG)) 161 asm volatile(PPC_TLBIE_5(%0, %1, 0, 0, 1) 162 : : "r" (addr), "r" (kvm->arch.lpid) : "memory"); 163 asm volatile("ptesync": : :"memory"); 164 } 165 166 unsigned long kvmppc_radix_update_pte(struct kvm *kvm, pte_t *ptep, 167 unsigned long clr, unsigned long set, 168 unsigned long addr, unsigned int shift) 169 { 170 unsigned long old = 0; 171 172 if (!(clr & _PAGE_PRESENT) && cpu_has_feature(CPU_FTR_POWER9_DD1) && 173 pte_present(*ptep)) { 174 /* have to invalidate it first */ 175 old = __radix_pte_update(ptep, _PAGE_PRESENT, 0); 176 kvmppc_radix_tlbie_page(kvm, addr, shift); 177 set |= _PAGE_PRESENT; 178 old &= _PAGE_PRESENT; 179 } 180 return __radix_pte_update(ptep, clr, set) | old; 181 } 182 183 void kvmppc_radix_set_pte_at(struct kvm *kvm, unsigned long addr, 184 pte_t *ptep, pte_t pte) 185 { 186 radix__set_pte_at(kvm->mm, addr, ptep, pte, 0); 187 } 188 189 static struct kmem_cache *kvm_pte_cache; 190 191 static pte_t *kvmppc_pte_alloc(void) 192 { 193 return kmem_cache_alloc(kvm_pte_cache, GFP_KERNEL); 194 } 195 196 static void kvmppc_pte_free(pte_t *ptep) 197 { 198 kmem_cache_free(kvm_pte_cache, ptep); 199 } 200 201 /* Like pmd_huge() and pmd_large(), but works regardless of config options */ 202 static inline int pmd_is_leaf(pmd_t pmd) 203 { 204 return !!(pmd_val(pmd) & _PAGE_PTE); 205 } 206 207 static int kvmppc_create_pte(struct kvm *kvm, pte_t pte, unsigned long gpa, 208 unsigned int level, unsigned long mmu_seq) 209 { 210 pgd_t *pgd; 211 pud_t *pud, *new_pud = NULL; 212 pmd_t *pmd, *new_pmd = NULL; 213 pte_t *ptep, *new_ptep = NULL; 214 unsigned long old; 215 int ret; 216 217 /* Traverse the guest's 2nd-level tree, allocate new levels needed */ 218 pgd = kvm->arch.pgtable + pgd_index(gpa); 219 pud = NULL; 220 if (pgd_present(*pgd)) 221 pud = pud_offset(pgd, gpa); 222 else 223 new_pud = pud_alloc_one(kvm->mm, gpa); 224 225 pmd = NULL; 226 if (pud && pud_present(*pud)) 227 pmd = pmd_offset(pud, gpa); 228 else 229 new_pmd = pmd_alloc_one(kvm->mm, gpa); 230 231 if (level == 0 && !(pmd && pmd_present(*pmd) && !pmd_is_leaf(*pmd))) 232 new_ptep = kvmppc_pte_alloc(); 233 234 /* Check if we might have been invalidated; let the guest retry if so */ 235 spin_lock(&kvm->mmu_lock); 236 ret = -EAGAIN; 237 if (mmu_notifier_retry(kvm, mmu_seq)) 238 goto out_unlock; 239 240 /* Now traverse again under the lock and change the tree */ 241 ret = -ENOMEM; 242 if (pgd_none(*pgd)) { 243 if (!new_pud) 244 goto out_unlock; 245 pgd_populate(kvm->mm, pgd, new_pud); 246 new_pud = NULL; 247 } 248 pud = pud_offset(pgd, gpa); 249 if (pud_none(*pud)) { 250 if (!new_pmd) 251 goto out_unlock; 252 pud_populate(kvm->mm, pud, new_pmd); 253 new_pmd = NULL; 254 } 255 pmd = pmd_offset(pud, gpa); 256 if (pmd_is_leaf(*pmd)) { 257 unsigned long lgpa = gpa & PMD_MASK; 258 259 /* 260 * If we raced with another CPU which has just put 261 * a 2MB pte in after we saw a pte page, try again. 262 */ 263 if (level == 0 && !new_ptep) { 264 ret = -EAGAIN; 265 goto out_unlock; 266 } 267 /* Valid 2MB page here already, remove it */ 268 old = kvmppc_radix_update_pte(kvm, pmdp_ptep(pmd), 269 ~0UL, 0, lgpa, PMD_SHIFT); 270 kvmppc_radix_tlbie_page(kvm, lgpa, PMD_SHIFT); 271 if (old & _PAGE_DIRTY) { 272 unsigned long gfn = lgpa >> PAGE_SHIFT; 273 struct kvm_memory_slot *memslot; 274 memslot = gfn_to_memslot(kvm, gfn); 275 if (memslot && memslot->dirty_bitmap) 276 kvmppc_update_dirty_map(memslot, 277 gfn, PMD_SIZE); 278 } 279 } else if (level == 1 && !pmd_none(*pmd)) { 280 /* 281 * There's a page table page here, but we wanted 282 * to install a large page. Tell the caller and let 283 * it try installing a normal page if it wants. 284 */ 285 ret = -EBUSY; 286 goto out_unlock; 287 } 288 if (level == 0) { 289 if (pmd_none(*pmd)) { 290 if (!new_ptep) 291 goto out_unlock; 292 pmd_populate(kvm->mm, pmd, new_ptep); 293 new_ptep = NULL; 294 } 295 ptep = pte_offset_kernel(pmd, gpa); 296 if (pte_present(*ptep)) { 297 /* PTE was previously valid, so invalidate it */ 298 old = kvmppc_radix_update_pte(kvm, ptep, _PAGE_PRESENT, 299 0, gpa, 0); 300 kvmppc_radix_tlbie_page(kvm, gpa, 0); 301 if (old & _PAGE_DIRTY) 302 mark_page_dirty(kvm, gpa >> PAGE_SHIFT); 303 } 304 kvmppc_radix_set_pte_at(kvm, gpa, ptep, pte); 305 } else { 306 kvmppc_radix_set_pte_at(kvm, gpa, pmdp_ptep(pmd), pte); 307 } 308 ret = 0; 309 310 out_unlock: 311 spin_unlock(&kvm->mmu_lock); 312 if (new_pud) 313 pud_free(kvm->mm, new_pud); 314 if (new_pmd) 315 pmd_free(kvm->mm, new_pmd); 316 if (new_ptep) 317 kvmppc_pte_free(new_ptep); 318 return ret; 319 } 320 321 int kvmppc_book3s_radix_page_fault(struct kvm_run *run, struct kvm_vcpu *vcpu, 322 unsigned long ea, unsigned long dsisr) 323 { 324 struct kvm *kvm = vcpu->kvm; 325 unsigned long mmu_seq, pte_size; 326 unsigned long gpa, gfn, hva, pfn; 327 struct kvm_memory_slot *memslot; 328 struct page *page = NULL, *pages[1]; 329 long ret, npages, ok; 330 unsigned int writing; 331 struct vm_area_struct *vma; 332 unsigned long flags; 333 pte_t pte, *ptep; 334 unsigned long pgflags; 335 unsigned int shift, level; 336 337 /* Check for unusual errors */ 338 if (dsisr & DSISR_UNSUPP_MMU) { 339 pr_err("KVM: Got unsupported MMU fault\n"); 340 return -EFAULT; 341 } 342 if (dsisr & DSISR_BADACCESS) { 343 /* Reflect to the guest as DSI */ 344 pr_err("KVM: Got radix HV page fault with DSISR=%lx\n", dsisr); 345 kvmppc_core_queue_data_storage(vcpu, ea, dsisr); 346 return RESUME_GUEST; 347 } 348 349 /* Translate the logical address and get the page */ 350 gpa = vcpu->arch.fault_gpa & ~0xfffUL; 351 gpa &= ~0xF000000000000000ul; 352 gfn = gpa >> PAGE_SHIFT; 353 if (!(dsisr & DSISR_PRTABLE_FAULT)) 354 gpa |= ea & 0xfff; 355 memslot = gfn_to_memslot(kvm, gfn); 356 357 /* No memslot means it's an emulated MMIO region */ 358 if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID)) { 359 if (dsisr & (DSISR_PRTABLE_FAULT | DSISR_BADACCESS | 360 DSISR_SET_RC)) { 361 /* 362 * Bad address in guest page table tree, or other 363 * unusual error - reflect it to the guest as DSI. 364 */ 365 kvmppc_core_queue_data_storage(vcpu, ea, dsisr); 366 return RESUME_GUEST; 367 } 368 return kvmppc_hv_emulate_mmio(run, vcpu, gpa, ea, 369 dsisr & DSISR_ISSTORE); 370 } 371 372 /* used to check for invalidations in progress */ 373 mmu_seq = kvm->mmu_notifier_seq; 374 smp_rmb(); 375 376 writing = (dsisr & DSISR_ISSTORE) != 0; 377 hva = gfn_to_hva_memslot(memslot, gfn); 378 if (dsisr & DSISR_SET_RC) { 379 /* 380 * Need to set an R or C bit in the 2nd-level tables; 381 * if the relevant bits aren't already set in the linux 382 * page tables, fall through to do the gup_fast to 383 * set them in the linux page tables too. 384 */ 385 ok = 0; 386 pgflags = _PAGE_ACCESSED; 387 if (writing) 388 pgflags |= _PAGE_DIRTY; 389 local_irq_save(flags); 390 ptep = find_current_mm_pte(current->mm->pgd, hva, NULL, NULL); 391 if (ptep) { 392 pte = READ_ONCE(*ptep); 393 if (pte_present(pte) && 394 (pte_val(pte) & pgflags) == pgflags) 395 ok = 1; 396 } 397 local_irq_restore(flags); 398 if (ok) { 399 spin_lock(&kvm->mmu_lock); 400 if (mmu_notifier_retry(vcpu->kvm, mmu_seq)) { 401 spin_unlock(&kvm->mmu_lock); 402 return RESUME_GUEST; 403 } 404 /* 405 * We are walking the secondary page table here. We can do this 406 * without disabling irq. 407 */ 408 ptep = __find_linux_pte(kvm->arch.pgtable, 409 gpa, NULL, &shift); 410 if (ptep && pte_present(*ptep)) { 411 kvmppc_radix_update_pte(kvm, ptep, 0, pgflags, 412 gpa, shift); 413 spin_unlock(&kvm->mmu_lock); 414 return RESUME_GUEST; 415 } 416 spin_unlock(&kvm->mmu_lock); 417 } 418 } 419 420 ret = -EFAULT; 421 pfn = 0; 422 pte_size = PAGE_SIZE; 423 pgflags = _PAGE_READ | _PAGE_EXEC; 424 level = 0; 425 npages = get_user_pages_fast(hva, 1, writing, pages); 426 if (npages < 1) { 427 /* Check if it's an I/O mapping */ 428 down_read(¤t->mm->mmap_sem); 429 vma = find_vma(current->mm, hva); 430 if (vma && vma->vm_start <= hva && hva < vma->vm_end && 431 (vma->vm_flags & VM_PFNMAP)) { 432 pfn = vma->vm_pgoff + 433 ((hva - vma->vm_start) >> PAGE_SHIFT); 434 pgflags = pgprot_val(vma->vm_page_prot); 435 } 436 up_read(¤t->mm->mmap_sem); 437 if (!pfn) 438 return -EFAULT; 439 } else { 440 page = pages[0]; 441 pfn = page_to_pfn(page); 442 if (PageCompound(page)) { 443 pte_size <<= compound_order(compound_head(page)); 444 /* See if we can insert a 2MB large-page PTE here */ 445 if (pte_size >= PMD_SIZE && 446 (gpa & (PMD_SIZE - PAGE_SIZE)) == 447 (hva & (PMD_SIZE - PAGE_SIZE))) { 448 level = 1; 449 pfn &= ~((PMD_SIZE >> PAGE_SHIFT) - 1); 450 } 451 } 452 /* See if we can provide write access */ 453 if (writing) { 454 pgflags |= _PAGE_WRITE; 455 } else { 456 local_irq_save(flags); 457 ptep = find_current_mm_pte(current->mm->pgd, 458 hva, NULL, NULL); 459 if (ptep && pte_write(*ptep)) 460 pgflags |= _PAGE_WRITE; 461 local_irq_restore(flags); 462 } 463 } 464 465 /* 466 * Compute the PTE value that we need to insert. 467 */ 468 pgflags |= _PAGE_PRESENT | _PAGE_PTE | _PAGE_ACCESSED; 469 if (pgflags & _PAGE_WRITE) 470 pgflags |= _PAGE_DIRTY; 471 pte = pfn_pte(pfn, __pgprot(pgflags)); 472 473 /* Allocate space in the tree and write the PTE */ 474 ret = kvmppc_create_pte(kvm, pte, gpa, level, mmu_seq); 475 if (ret == -EBUSY) { 476 /* 477 * There's already a PMD where wanted to install a large page; 478 * for now, fall back to installing a small page. 479 */ 480 level = 0; 481 pfn |= gfn & ((PMD_SIZE >> PAGE_SHIFT) - 1); 482 pte = pfn_pte(pfn, __pgprot(pgflags)); 483 ret = kvmppc_create_pte(kvm, pte, gpa, level, mmu_seq); 484 } 485 486 if (page) { 487 if (!ret && (pgflags & _PAGE_WRITE)) 488 set_page_dirty_lock(page); 489 put_page(page); 490 } 491 492 if (ret == 0 || ret == -EAGAIN) 493 ret = RESUME_GUEST; 494 return ret; 495 } 496 497 /* Called with kvm->lock held */ 498 int kvm_unmap_radix(struct kvm *kvm, struct kvm_memory_slot *memslot, 499 unsigned long gfn) 500 { 501 pte_t *ptep; 502 unsigned long gpa = gfn << PAGE_SHIFT; 503 unsigned int shift; 504 unsigned long old; 505 506 ptep = __find_linux_pte(kvm->arch.pgtable, gpa, NULL, &shift); 507 if (ptep && pte_present(*ptep)) { 508 old = kvmppc_radix_update_pte(kvm, ptep, _PAGE_PRESENT, 0, 509 gpa, shift); 510 kvmppc_radix_tlbie_page(kvm, gpa, shift); 511 if ((old & _PAGE_DIRTY) && memslot->dirty_bitmap) { 512 unsigned long npages = 1; 513 if (shift) 514 npages = 1ul << (shift - PAGE_SHIFT); 515 kvmppc_update_dirty_map(memslot, gfn, npages); 516 } 517 } 518 return 0; 519 } 520 521 /* Called with kvm->lock held */ 522 int kvm_age_radix(struct kvm *kvm, struct kvm_memory_slot *memslot, 523 unsigned long gfn) 524 { 525 pte_t *ptep; 526 unsigned long gpa = gfn << PAGE_SHIFT; 527 unsigned int shift; 528 int ref = 0; 529 530 ptep = __find_linux_pte(kvm->arch.pgtable, gpa, NULL, &shift); 531 if (ptep && pte_present(*ptep) && pte_young(*ptep)) { 532 kvmppc_radix_update_pte(kvm, ptep, _PAGE_ACCESSED, 0, 533 gpa, shift); 534 /* XXX need to flush tlb here? */ 535 ref = 1; 536 } 537 return ref; 538 } 539 540 /* Called with kvm->lock held */ 541 int kvm_test_age_radix(struct kvm *kvm, struct kvm_memory_slot *memslot, 542 unsigned long gfn) 543 { 544 pte_t *ptep; 545 unsigned long gpa = gfn << PAGE_SHIFT; 546 unsigned int shift; 547 int ref = 0; 548 549 ptep = __find_linux_pte(kvm->arch.pgtable, gpa, NULL, &shift); 550 if (ptep && pte_present(*ptep) && pte_young(*ptep)) 551 ref = 1; 552 return ref; 553 } 554 555 /* Returns the number of PAGE_SIZE pages that are dirty */ 556 static int kvm_radix_test_clear_dirty(struct kvm *kvm, 557 struct kvm_memory_slot *memslot, int pagenum) 558 { 559 unsigned long gfn = memslot->base_gfn + pagenum; 560 unsigned long gpa = gfn << PAGE_SHIFT; 561 pte_t *ptep; 562 unsigned int shift; 563 int ret = 0; 564 565 ptep = __find_linux_pte(kvm->arch.pgtable, gpa, 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 int npages; 582 583 for (i = 0; i < memslot->npages; i = j) { 584 npages = kvm_radix_test_clear_dirty(kvm, memslot, i); 585 586 /* 587 * Note that if npages > 0 then i must be a multiple of npages, 588 * since huge pages are only used to back the guest at guest 589 * real addresses that are a multiple of their size. 590 * Since we have at most one PTE covering any given guest 591 * real address, if npages > 1 we can skip to i + npages. 592 */ 593 j = i + 1; 594 if (npages) { 595 set_dirty_bits(map, i, npages); 596 j = i + npages; 597 } 598 } 599 return 0; 600 } 601 602 static void add_rmmu_ap_encoding(struct kvm_ppc_rmmu_info *info, 603 int psize, int *indexp) 604 { 605 if (!mmu_psize_defs[psize].shift) 606 return; 607 info->ap_encodings[*indexp] = mmu_psize_defs[psize].shift | 608 (mmu_psize_defs[psize].ap << 29); 609 ++(*indexp); 610 } 611 612 int kvmhv_get_rmmu_info(struct kvm *kvm, struct kvm_ppc_rmmu_info *info) 613 { 614 int i; 615 616 if (!radix_enabled()) 617 return -EINVAL; 618 memset(info, 0, sizeof(*info)); 619 620 /* 4k page size */ 621 info->geometries[0].page_shift = 12; 622 info->geometries[0].level_bits[0] = 9; 623 for (i = 1; i < 4; ++i) 624 info->geometries[0].level_bits[i] = p9_supported_radix_bits[i]; 625 /* 64k page size */ 626 info->geometries[1].page_shift = 16; 627 for (i = 0; i < 4; ++i) 628 info->geometries[1].level_bits[i] = p9_supported_radix_bits[i]; 629 630 i = 0; 631 add_rmmu_ap_encoding(info, MMU_PAGE_4K, &i); 632 add_rmmu_ap_encoding(info, MMU_PAGE_64K, &i); 633 add_rmmu_ap_encoding(info, MMU_PAGE_2M, &i); 634 add_rmmu_ap_encoding(info, MMU_PAGE_1G, &i); 635 636 return 0; 637 } 638 639 int kvmppc_init_vm_radix(struct kvm *kvm) 640 { 641 kvm->arch.pgtable = pgd_alloc(kvm->mm); 642 if (!kvm->arch.pgtable) 643 return -ENOMEM; 644 return 0; 645 } 646 647 void kvmppc_free_radix(struct kvm *kvm) 648 { 649 unsigned long ig, iu, im; 650 pte_t *pte; 651 pmd_t *pmd; 652 pud_t *pud; 653 pgd_t *pgd; 654 655 if (!kvm->arch.pgtable) 656 return; 657 pgd = kvm->arch.pgtable; 658 for (ig = 0; ig < PTRS_PER_PGD; ++ig, ++pgd) { 659 if (!pgd_present(*pgd)) 660 continue; 661 pud = pud_offset(pgd, 0); 662 for (iu = 0; iu < PTRS_PER_PUD; ++iu, ++pud) { 663 if (!pud_present(*pud)) 664 continue; 665 pmd = pmd_offset(pud, 0); 666 for (im = 0; im < PTRS_PER_PMD; ++im, ++pmd) { 667 if (pmd_is_leaf(*pmd)) { 668 pmd_clear(pmd); 669 continue; 670 } 671 if (!pmd_present(*pmd)) 672 continue; 673 pte = pte_offset_map(pmd, 0); 674 memset(pte, 0, sizeof(long) << PTE_INDEX_SIZE); 675 kvmppc_pte_free(pte); 676 pmd_clear(pmd); 677 } 678 pmd_free(kvm->mm, pmd_offset(pud, 0)); 679 pud_clear(pud); 680 } 681 pud_free(kvm->mm, pud_offset(pgd, 0)); 682 pgd_clear(pgd); 683 } 684 pgd_free(kvm->mm, kvm->arch.pgtable); 685 kvm->arch.pgtable = NULL; 686 } 687 688 static void pte_ctor(void *addr) 689 { 690 memset(addr, 0, PTE_TABLE_SIZE); 691 } 692 693 int kvmppc_radix_init(void) 694 { 695 unsigned long size = sizeof(void *) << PTE_INDEX_SIZE; 696 697 kvm_pte_cache = kmem_cache_create("kvm-pte", size, size, 0, pte_ctor); 698 if (!kvm_pte_cache) 699 return -ENOMEM; 700 return 0; 701 } 702 703 void kvmppc_radix_exit(void) 704 { 705 kmem_cache_destroy(kvm_pte_cache); 706 } 707