1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * 4 * Copyright 2016 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com> 5 */ 6 7 #include <linux/types.h> 8 #include <linux/string.h> 9 #include <linux/kvm.h> 10 #include <linux/kvm_host.h> 11 #include <linux/anon_inodes.h> 12 #include <linux/file.h> 13 #include <linux/debugfs.h> 14 15 #include <asm/kvm_ppc.h> 16 #include <asm/kvm_book3s.h> 17 #include <asm/page.h> 18 #include <asm/mmu.h> 19 #include <asm/pgtable.h> 20 #include <asm/pgalloc.h> 21 #include <asm/pte-walk.h> 22 #include <asm/ultravisor.h> 23 #include <asm/kvm_book3s_uvmem.h> 24 25 /* 26 * Supported radix tree geometry. 27 * Like p9, we support either 5 or 9 bits at the first (lowest) level, 28 * for a page size of 64k or 4k. 29 */ 30 static int p9_supported_radix_bits[4] = { 5, 9, 9, 13 }; 31 32 unsigned long __kvmhv_copy_tofrom_guest_radix(int lpid, int pid, 33 gva_t eaddr, void *to, void *from, 34 unsigned long n) 35 { 36 int uninitialized_var(old_pid), old_lpid; 37 unsigned long quadrant, ret = n; 38 bool is_load = !!to; 39 40 /* Can't access quadrants 1 or 2 in non-HV mode, call the HV to do it */ 41 if (kvmhv_on_pseries()) 42 return plpar_hcall_norets(H_COPY_TOFROM_GUEST, lpid, pid, eaddr, 43 __pa(to), __pa(from), n); 44 45 quadrant = 1; 46 if (!pid) 47 quadrant = 2; 48 if (is_load) 49 from = (void *) (eaddr | (quadrant << 62)); 50 else 51 to = (void *) (eaddr | (quadrant << 62)); 52 53 preempt_disable(); 54 55 /* switch the lpid first to avoid running host with unallocated pid */ 56 old_lpid = mfspr(SPRN_LPID); 57 if (old_lpid != lpid) 58 mtspr(SPRN_LPID, lpid); 59 if (quadrant == 1) { 60 old_pid = mfspr(SPRN_PID); 61 if (old_pid != pid) 62 mtspr(SPRN_PID, pid); 63 } 64 isync(); 65 66 pagefault_disable(); 67 if (is_load) 68 ret = raw_copy_from_user(to, from, n); 69 else 70 ret = raw_copy_to_user(to, from, n); 71 pagefault_enable(); 72 73 /* switch the pid first to avoid running host with unallocated pid */ 74 if (quadrant == 1 && pid != old_pid) 75 mtspr(SPRN_PID, old_pid); 76 if (lpid != old_lpid) 77 mtspr(SPRN_LPID, old_lpid); 78 isync(); 79 80 preempt_enable(); 81 82 return ret; 83 } 84 EXPORT_SYMBOL_GPL(__kvmhv_copy_tofrom_guest_radix); 85 86 static long kvmhv_copy_tofrom_guest_radix(struct kvm_vcpu *vcpu, gva_t eaddr, 87 void *to, void *from, unsigned long n) 88 { 89 int lpid = vcpu->kvm->arch.lpid; 90 int pid = vcpu->arch.pid; 91 92 /* This would cause a data segment intr so don't allow the access */ 93 if (eaddr & (0x3FFUL << 52)) 94 return -EINVAL; 95 96 /* Should we be using the nested lpid */ 97 if (vcpu->arch.nested) 98 lpid = vcpu->arch.nested->shadow_lpid; 99 100 /* If accessing quadrant 3 then pid is expected to be 0 */ 101 if (((eaddr >> 62) & 0x3) == 0x3) 102 pid = 0; 103 104 eaddr &= ~(0xFFFUL << 52); 105 106 return __kvmhv_copy_tofrom_guest_radix(lpid, pid, eaddr, to, from, n); 107 } 108 109 long kvmhv_copy_from_guest_radix(struct kvm_vcpu *vcpu, gva_t eaddr, void *to, 110 unsigned long n) 111 { 112 long ret; 113 114 ret = kvmhv_copy_tofrom_guest_radix(vcpu, eaddr, to, NULL, n); 115 if (ret > 0) 116 memset(to + (n - ret), 0, ret); 117 118 return ret; 119 } 120 EXPORT_SYMBOL_GPL(kvmhv_copy_from_guest_radix); 121 122 long kvmhv_copy_to_guest_radix(struct kvm_vcpu *vcpu, gva_t eaddr, void *from, 123 unsigned long n) 124 { 125 return kvmhv_copy_tofrom_guest_radix(vcpu, eaddr, NULL, from, n); 126 } 127 EXPORT_SYMBOL_GPL(kvmhv_copy_to_guest_radix); 128 129 int kvmppc_mmu_walk_radix_tree(struct kvm_vcpu *vcpu, gva_t eaddr, 130 struct kvmppc_pte *gpte, u64 root, 131 u64 *pte_ret_p) 132 { 133 struct kvm *kvm = vcpu->kvm; 134 int ret, level, ps; 135 unsigned long rts, bits, offset, index; 136 u64 pte, base, gpa; 137 __be64 rpte; 138 139 rts = ((root & RTS1_MASK) >> (RTS1_SHIFT - 3)) | 140 ((root & RTS2_MASK) >> RTS2_SHIFT); 141 bits = root & RPDS_MASK; 142 base = root & RPDB_MASK; 143 144 offset = rts + 31; 145 146 /* Current implementations only support 52-bit space */ 147 if (offset != 52) 148 return -EINVAL; 149 150 /* Walk each level of the radix tree */ 151 for (level = 3; level >= 0; --level) { 152 u64 addr; 153 /* Check a valid size */ 154 if (level && bits != p9_supported_radix_bits[level]) 155 return -EINVAL; 156 if (level == 0 && !(bits == 5 || bits == 9)) 157 return -EINVAL; 158 offset -= bits; 159 index = (eaddr >> offset) & ((1UL << bits) - 1); 160 /* Check that low bits of page table base are zero */ 161 if (base & ((1UL << (bits + 3)) - 1)) 162 return -EINVAL; 163 /* Read the entry from guest memory */ 164 addr = base + (index * sizeof(rpte)); 165 ret = kvm_read_guest(kvm, addr, &rpte, sizeof(rpte)); 166 if (ret) { 167 if (pte_ret_p) 168 *pte_ret_p = addr; 169 return ret; 170 } 171 pte = __be64_to_cpu(rpte); 172 if (!(pte & _PAGE_PRESENT)) 173 return -ENOENT; 174 /* Check if a leaf entry */ 175 if (pte & _PAGE_PTE) 176 break; 177 /* Get ready to walk the next level */ 178 base = pte & RPDB_MASK; 179 bits = pte & RPDS_MASK; 180 } 181 182 /* Need a leaf at lowest level; 512GB pages not supported */ 183 if (level < 0 || level == 3) 184 return -EINVAL; 185 186 /* We found a valid leaf PTE */ 187 /* Offset is now log base 2 of the page size */ 188 gpa = pte & 0x01fffffffffff000ul; 189 if (gpa & ((1ul << offset) - 1)) 190 return -EINVAL; 191 gpa |= eaddr & ((1ul << offset) - 1); 192 for (ps = MMU_PAGE_4K; ps < MMU_PAGE_COUNT; ++ps) 193 if (offset == mmu_psize_defs[ps].shift) 194 break; 195 gpte->page_size = ps; 196 gpte->page_shift = offset; 197 198 gpte->eaddr = eaddr; 199 gpte->raddr = gpa; 200 201 /* Work out permissions */ 202 gpte->may_read = !!(pte & _PAGE_READ); 203 gpte->may_write = !!(pte & _PAGE_WRITE); 204 gpte->may_execute = !!(pte & _PAGE_EXEC); 205 206 gpte->rc = pte & (_PAGE_ACCESSED | _PAGE_DIRTY); 207 208 if (pte_ret_p) 209 *pte_ret_p = pte; 210 211 return 0; 212 } 213 214 /* 215 * Used to walk a partition or process table radix tree in guest memory 216 * Note: We exploit the fact that a partition table and a process 217 * table have the same layout, a partition-scoped page table and a 218 * process-scoped page table have the same layout, and the 2nd 219 * doubleword of a partition table entry has the same layout as 220 * the PTCR register. 221 */ 222 int kvmppc_mmu_radix_translate_table(struct kvm_vcpu *vcpu, gva_t eaddr, 223 struct kvmppc_pte *gpte, u64 table, 224 int table_index, u64 *pte_ret_p) 225 { 226 struct kvm *kvm = vcpu->kvm; 227 int ret; 228 unsigned long size, ptbl, root; 229 struct prtb_entry entry; 230 231 if ((table & PRTS_MASK) > 24) 232 return -EINVAL; 233 size = 1ul << ((table & PRTS_MASK) + 12); 234 235 /* Is the table big enough to contain this entry? */ 236 if ((table_index * sizeof(entry)) >= size) 237 return -EINVAL; 238 239 /* Read the table to find the root of the radix tree */ 240 ptbl = (table & PRTB_MASK) + (table_index * sizeof(entry)); 241 ret = kvm_read_guest(kvm, ptbl, &entry, sizeof(entry)); 242 if (ret) 243 return ret; 244 245 /* Root is stored in the first double word */ 246 root = be64_to_cpu(entry.prtb0); 247 248 return kvmppc_mmu_walk_radix_tree(vcpu, eaddr, gpte, root, pte_ret_p); 249 } 250 251 int kvmppc_mmu_radix_xlate(struct kvm_vcpu *vcpu, gva_t eaddr, 252 struct kvmppc_pte *gpte, bool data, bool iswrite) 253 { 254 u32 pid; 255 u64 pte; 256 int ret; 257 258 /* Work out effective PID */ 259 switch (eaddr >> 62) { 260 case 0: 261 pid = vcpu->arch.pid; 262 break; 263 case 3: 264 pid = 0; 265 break; 266 default: 267 return -EINVAL; 268 } 269 270 ret = kvmppc_mmu_radix_translate_table(vcpu, eaddr, gpte, 271 vcpu->kvm->arch.process_table, pid, &pte); 272 if (ret) 273 return ret; 274 275 /* Check privilege (applies only to process scoped translations) */ 276 if (kvmppc_get_msr(vcpu) & MSR_PR) { 277 if (pte & _PAGE_PRIVILEGED) { 278 gpte->may_read = 0; 279 gpte->may_write = 0; 280 gpte->may_execute = 0; 281 } 282 } else { 283 if (!(pte & _PAGE_PRIVILEGED)) { 284 /* Check AMR/IAMR to see if strict mode is in force */ 285 if (vcpu->arch.amr & (1ul << 62)) 286 gpte->may_read = 0; 287 if (vcpu->arch.amr & (1ul << 63)) 288 gpte->may_write = 0; 289 if (vcpu->arch.iamr & (1ul << 62)) 290 gpte->may_execute = 0; 291 } 292 } 293 294 return 0; 295 } 296 297 void kvmppc_radix_tlbie_page(struct kvm *kvm, unsigned long addr, 298 unsigned int pshift, unsigned int lpid) 299 { 300 unsigned long psize = PAGE_SIZE; 301 int psi; 302 long rc; 303 unsigned long rb; 304 305 if (pshift) 306 psize = 1UL << pshift; 307 else 308 pshift = PAGE_SHIFT; 309 310 addr &= ~(psize - 1); 311 312 if (!kvmhv_on_pseries()) { 313 radix__flush_tlb_lpid_page(lpid, addr, psize); 314 return; 315 } 316 317 psi = shift_to_mmu_psize(pshift); 318 rb = addr | (mmu_get_ap(psi) << PPC_BITLSHIFT(58)); 319 rc = plpar_hcall_norets(H_TLB_INVALIDATE, H_TLBIE_P1_ENC(0, 0, 1), 320 lpid, rb); 321 if (rc) 322 pr_err("KVM: TLB page invalidation hcall failed, rc=%ld\n", rc); 323 } 324 325 static void kvmppc_radix_flush_pwc(struct kvm *kvm, unsigned int lpid) 326 { 327 long rc; 328 329 if (!kvmhv_on_pseries()) { 330 radix__flush_pwc_lpid(lpid); 331 return; 332 } 333 334 rc = plpar_hcall_norets(H_TLB_INVALIDATE, H_TLBIE_P1_ENC(1, 0, 1), 335 lpid, TLBIEL_INVAL_SET_LPID); 336 if (rc) 337 pr_err("KVM: TLB PWC invalidation hcall failed, rc=%ld\n", rc); 338 } 339 340 static unsigned long kvmppc_radix_update_pte(struct kvm *kvm, pte_t *ptep, 341 unsigned long clr, unsigned long set, 342 unsigned long addr, unsigned int shift) 343 { 344 return __radix_pte_update(ptep, clr, set); 345 } 346 347 void kvmppc_radix_set_pte_at(struct kvm *kvm, unsigned long addr, 348 pte_t *ptep, pte_t pte) 349 { 350 radix__set_pte_at(kvm->mm, addr, ptep, pte, 0); 351 } 352 353 static struct kmem_cache *kvm_pte_cache; 354 static struct kmem_cache *kvm_pmd_cache; 355 356 static pte_t *kvmppc_pte_alloc(void) 357 { 358 return kmem_cache_alloc(kvm_pte_cache, GFP_KERNEL); 359 } 360 361 static void kvmppc_pte_free(pte_t *ptep) 362 { 363 kmem_cache_free(kvm_pte_cache, ptep); 364 } 365 366 static pmd_t *kvmppc_pmd_alloc(void) 367 { 368 return kmem_cache_alloc(kvm_pmd_cache, GFP_KERNEL); 369 } 370 371 static void kvmppc_pmd_free(pmd_t *pmdp) 372 { 373 kmem_cache_free(kvm_pmd_cache, pmdp); 374 } 375 376 /* Called with kvm->mmu_lock held */ 377 void kvmppc_unmap_pte(struct kvm *kvm, pte_t *pte, unsigned long gpa, 378 unsigned int shift, 379 const struct kvm_memory_slot *memslot, 380 unsigned int lpid) 381 382 { 383 unsigned long old; 384 unsigned long gfn = gpa >> PAGE_SHIFT; 385 unsigned long page_size = PAGE_SIZE; 386 unsigned long hpa; 387 388 old = kvmppc_radix_update_pte(kvm, pte, ~0UL, 0, gpa, shift); 389 kvmppc_radix_tlbie_page(kvm, gpa, shift, lpid); 390 391 /* The following only applies to L1 entries */ 392 if (lpid != kvm->arch.lpid) 393 return; 394 395 if (!memslot) { 396 memslot = gfn_to_memslot(kvm, gfn); 397 if (!memslot) 398 return; 399 } 400 if (shift) { /* 1GB or 2MB page */ 401 page_size = 1ul << shift; 402 if (shift == PMD_SHIFT) 403 kvm->stat.num_2M_pages--; 404 else if (shift == PUD_SHIFT) 405 kvm->stat.num_1G_pages--; 406 } 407 408 gpa &= ~(page_size - 1); 409 hpa = old & PTE_RPN_MASK; 410 kvmhv_remove_nest_rmap_range(kvm, memslot, gpa, hpa, page_size); 411 412 if ((old & _PAGE_DIRTY) && memslot->dirty_bitmap) 413 kvmppc_update_dirty_map(memslot, gfn, page_size); 414 } 415 416 /* 417 * kvmppc_free_p?d are used to free existing page tables, and recursively 418 * descend and clear and free children. 419 * Callers are responsible for flushing the PWC. 420 * 421 * When page tables are being unmapped/freed as part of page fault path 422 * (full == false), ptes are not expected. There is code to unmap them 423 * and emit a warning if encountered, but there may already be data 424 * corruption due to the unexpected mappings. 425 */ 426 static void kvmppc_unmap_free_pte(struct kvm *kvm, pte_t *pte, bool full, 427 unsigned int lpid) 428 { 429 if (full) { 430 memset(pte, 0, sizeof(long) << PTE_INDEX_SIZE); 431 } else { 432 pte_t *p = pte; 433 unsigned long it; 434 435 for (it = 0; it < PTRS_PER_PTE; ++it, ++p) { 436 if (pte_val(*p) == 0) 437 continue; 438 WARN_ON_ONCE(1); 439 kvmppc_unmap_pte(kvm, p, 440 pte_pfn(*p) << PAGE_SHIFT, 441 PAGE_SHIFT, NULL, lpid); 442 } 443 } 444 445 kvmppc_pte_free(pte); 446 } 447 448 static void kvmppc_unmap_free_pmd(struct kvm *kvm, pmd_t *pmd, bool full, 449 unsigned int lpid) 450 { 451 unsigned long im; 452 pmd_t *p = pmd; 453 454 for (im = 0; im < PTRS_PER_PMD; ++im, ++p) { 455 if (!pmd_present(*p)) 456 continue; 457 if (pmd_is_leaf(*p)) { 458 if (full) { 459 pmd_clear(p); 460 } else { 461 WARN_ON_ONCE(1); 462 kvmppc_unmap_pte(kvm, (pte_t *)p, 463 pte_pfn(*(pte_t *)p) << PAGE_SHIFT, 464 PMD_SHIFT, NULL, lpid); 465 } 466 } else { 467 pte_t *pte; 468 469 pte = pte_offset_map(p, 0); 470 kvmppc_unmap_free_pte(kvm, pte, full, lpid); 471 pmd_clear(p); 472 } 473 } 474 kvmppc_pmd_free(pmd); 475 } 476 477 static void kvmppc_unmap_free_pud(struct kvm *kvm, pud_t *pud, 478 unsigned int lpid) 479 { 480 unsigned long iu; 481 pud_t *p = pud; 482 483 for (iu = 0; iu < PTRS_PER_PUD; ++iu, ++p) { 484 if (!pud_present(*p)) 485 continue; 486 if (pud_is_leaf(*p)) { 487 pud_clear(p); 488 } else { 489 pmd_t *pmd; 490 491 pmd = pmd_offset(p, 0); 492 kvmppc_unmap_free_pmd(kvm, pmd, true, lpid); 493 pud_clear(p); 494 } 495 } 496 pud_free(kvm->mm, pud); 497 } 498 499 void kvmppc_free_pgtable_radix(struct kvm *kvm, pgd_t *pgd, unsigned int lpid) 500 { 501 unsigned long ig; 502 503 for (ig = 0; ig < PTRS_PER_PGD; ++ig, ++pgd) { 504 pud_t *pud; 505 506 if (!pgd_present(*pgd)) 507 continue; 508 pud = pud_offset(pgd, 0); 509 kvmppc_unmap_free_pud(kvm, pud, lpid); 510 pgd_clear(pgd); 511 } 512 } 513 514 void kvmppc_free_radix(struct kvm *kvm) 515 { 516 if (kvm->arch.pgtable) { 517 kvmppc_free_pgtable_radix(kvm, kvm->arch.pgtable, 518 kvm->arch.lpid); 519 pgd_free(kvm->mm, kvm->arch.pgtable); 520 kvm->arch.pgtable = NULL; 521 } 522 } 523 524 static void kvmppc_unmap_free_pmd_entry_table(struct kvm *kvm, pmd_t *pmd, 525 unsigned long gpa, unsigned int lpid) 526 { 527 pte_t *pte = pte_offset_kernel(pmd, 0); 528 529 /* 530 * Clearing the pmd entry then flushing the PWC ensures that the pte 531 * page no longer be cached by the MMU, so can be freed without 532 * flushing the PWC again. 533 */ 534 pmd_clear(pmd); 535 kvmppc_radix_flush_pwc(kvm, lpid); 536 537 kvmppc_unmap_free_pte(kvm, pte, false, lpid); 538 } 539 540 static void kvmppc_unmap_free_pud_entry_table(struct kvm *kvm, pud_t *pud, 541 unsigned long gpa, unsigned int lpid) 542 { 543 pmd_t *pmd = pmd_offset(pud, 0); 544 545 /* 546 * Clearing the pud entry then flushing the PWC ensures that the pmd 547 * page and any children pte pages will no longer be cached by the MMU, 548 * so can be freed without flushing the PWC again. 549 */ 550 pud_clear(pud); 551 kvmppc_radix_flush_pwc(kvm, lpid); 552 553 kvmppc_unmap_free_pmd(kvm, pmd, false, lpid); 554 } 555 556 /* 557 * There are a number of bits which may differ between different faults to 558 * the same partition scope entry. RC bits, in the course of cleaning and 559 * aging. And the write bit can change, either the access could have been 560 * upgraded, or a read fault could happen concurrently with a write fault 561 * that sets those bits first. 562 */ 563 #define PTE_BITS_MUST_MATCH (~(_PAGE_WRITE | _PAGE_DIRTY | _PAGE_ACCESSED)) 564 565 int kvmppc_create_pte(struct kvm *kvm, pgd_t *pgtable, pte_t pte, 566 unsigned long gpa, unsigned int level, 567 unsigned long mmu_seq, unsigned int lpid, 568 unsigned long *rmapp, struct rmap_nested **n_rmap) 569 { 570 pgd_t *pgd; 571 pud_t *pud, *new_pud = NULL; 572 pmd_t *pmd, *new_pmd = NULL; 573 pte_t *ptep, *new_ptep = NULL; 574 int ret; 575 576 /* Traverse the guest's 2nd-level tree, allocate new levels needed */ 577 pgd = pgtable + pgd_index(gpa); 578 pud = NULL; 579 if (pgd_present(*pgd)) 580 pud = pud_offset(pgd, gpa); 581 else 582 new_pud = pud_alloc_one(kvm->mm, gpa); 583 584 pmd = NULL; 585 if (pud && pud_present(*pud) && !pud_is_leaf(*pud)) 586 pmd = pmd_offset(pud, gpa); 587 else if (level <= 1) 588 new_pmd = kvmppc_pmd_alloc(); 589 590 if (level == 0 && !(pmd && pmd_present(*pmd) && !pmd_is_leaf(*pmd))) 591 new_ptep = kvmppc_pte_alloc(); 592 593 /* Check if we might have been invalidated; let the guest retry if so */ 594 spin_lock(&kvm->mmu_lock); 595 ret = -EAGAIN; 596 if (mmu_notifier_retry(kvm, mmu_seq)) 597 goto out_unlock; 598 599 /* Now traverse again under the lock and change the tree */ 600 ret = -ENOMEM; 601 if (pgd_none(*pgd)) { 602 if (!new_pud) 603 goto out_unlock; 604 pgd_populate(kvm->mm, pgd, new_pud); 605 new_pud = NULL; 606 } 607 pud = pud_offset(pgd, gpa); 608 if (pud_is_leaf(*pud)) { 609 unsigned long hgpa = gpa & PUD_MASK; 610 611 /* Check if we raced and someone else has set the same thing */ 612 if (level == 2) { 613 if (pud_raw(*pud) == pte_raw(pte)) { 614 ret = 0; 615 goto out_unlock; 616 } 617 /* Valid 1GB page here already, add our extra bits */ 618 WARN_ON_ONCE((pud_val(*pud) ^ pte_val(pte)) & 619 PTE_BITS_MUST_MATCH); 620 kvmppc_radix_update_pte(kvm, (pte_t *)pud, 621 0, pte_val(pte), hgpa, PUD_SHIFT); 622 ret = 0; 623 goto out_unlock; 624 } 625 /* 626 * If we raced with another CPU which has just put 627 * a 1GB pte in after we saw a pmd page, try again. 628 */ 629 if (!new_pmd) { 630 ret = -EAGAIN; 631 goto out_unlock; 632 } 633 /* Valid 1GB page here already, remove it */ 634 kvmppc_unmap_pte(kvm, (pte_t *)pud, hgpa, PUD_SHIFT, NULL, 635 lpid); 636 } 637 if (level == 2) { 638 if (!pud_none(*pud)) { 639 /* 640 * There's a page table page here, but we wanted to 641 * install a large page, so remove and free the page 642 * table page. 643 */ 644 kvmppc_unmap_free_pud_entry_table(kvm, pud, gpa, lpid); 645 } 646 kvmppc_radix_set_pte_at(kvm, gpa, (pte_t *)pud, pte); 647 if (rmapp && n_rmap) 648 kvmhv_insert_nest_rmap(kvm, rmapp, n_rmap); 649 ret = 0; 650 goto out_unlock; 651 } 652 if (pud_none(*pud)) { 653 if (!new_pmd) 654 goto out_unlock; 655 pud_populate(kvm->mm, pud, new_pmd); 656 new_pmd = NULL; 657 } 658 pmd = pmd_offset(pud, gpa); 659 if (pmd_is_leaf(*pmd)) { 660 unsigned long lgpa = gpa & PMD_MASK; 661 662 /* Check if we raced and someone else has set the same thing */ 663 if (level == 1) { 664 if (pmd_raw(*pmd) == pte_raw(pte)) { 665 ret = 0; 666 goto out_unlock; 667 } 668 /* Valid 2MB page here already, add our extra bits */ 669 WARN_ON_ONCE((pmd_val(*pmd) ^ pte_val(pte)) & 670 PTE_BITS_MUST_MATCH); 671 kvmppc_radix_update_pte(kvm, pmdp_ptep(pmd), 672 0, pte_val(pte), lgpa, PMD_SHIFT); 673 ret = 0; 674 goto out_unlock; 675 } 676 677 /* 678 * If we raced with another CPU which has just put 679 * a 2MB pte in after we saw a pte page, try again. 680 */ 681 if (!new_ptep) { 682 ret = -EAGAIN; 683 goto out_unlock; 684 } 685 /* Valid 2MB page here already, remove it */ 686 kvmppc_unmap_pte(kvm, pmdp_ptep(pmd), lgpa, PMD_SHIFT, NULL, 687 lpid); 688 } 689 if (level == 1) { 690 if (!pmd_none(*pmd)) { 691 /* 692 * There's a page table page here, but we wanted to 693 * install a large page, so remove and free the page 694 * table page. 695 */ 696 kvmppc_unmap_free_pmd_entry_table(kvm, pmd, gpa, lpid); 697 } 698 kvmppc_radix_set_pte_at(kvm, gpa, pmdp_ptep(pmd), pte); 699 if (rmapp && n_rmap) 700 kvmhv_insert_nest_rmap(kvm, rmapp, n_rmap); 701 ret = 0; 702 goto out_unlock; 703 } 704 if (pmd_none(*pmd)) { 705 if (!new_ptep) 706 goto out_unlock; 707 pmd_populate(kvm->mm, pmd, new_ptep); 708 new_ptep = NULL; 709 } 710 ptep = pte_offset_kernel(pmd, gpa); 711 if (pte_present(*ptep)) { 712 /* Check if someone else set the same thing */ 713 if (pte_raw(*ptep) == pte_raw(pte)) { 714 ret = 0; 715 goto out_unlock; 716 } 717 /* Valid page here already, add our extra bits */ 718 WARN_ON_ONCE((pte_val(*ptep) ^ pte_val(pte)) & 719 PTE_BITS_MUST_MATCH); 720 kvmppc_radix_update_pte(kvm, ptep, 0, pte_val(pte), gpa, 0); 721 ret = 0; 722 goto out_unlock; 723 } 724 kvmppc_radix_set_pte_at(kvm, gpa, ptep, pte); 725 if (rmapp && n_rmap) 726 kvmhv_insert_nest_rmap(kvm, rmapp, n_rmap); 727 ret = 0; 728 729 out_unlock: 730 spin_unlock(&kvm->mmu_lock); 731 if (new_pud) 732 pud_free(kvm->mm, new_pud); 733 if (new_pmd) 734 kvmppc_pmd_free(new_pmd); 735 if (new_ptep) 736 kvmppc_pte_free(new_ptep); 737 return ret; 738 } 739 740 bool kvmppc_hv_handle_set_rc(struct kvm *kvm, pgd_t *pgtable, bool writing, 741 unsigned long gpa, unsigned int lpid) 742 { 743 unsigned long pgflags; 744 unsigned int shift; 745 pte_t *ptep; 746 747 /* 748 * Need to set an R or C bit in the 2nd-level tables; 749 * since we are just helping out the hardware here, 750 * it is sufficient to do what the hardware does. 751 */ 752 pgflags = _PAGE_ACCESSED; 753 if (writing) 754 pgflags |= _PAGE_DIRTY; 755 /* 756 * We are walking the secondary (partition-scoped) page table here. 757 * We can do this without disabling irq because the Linux MM 758 * subsystem doesn't do THP splits and collapses on this tree. 759 */ 760 ptep = __find_linux_pte(pgtable, gpa, NULL, &shift); 761 if (ptep && pte_present(*ptep) && (!writing || pte_write(*ptep))) { 762 kvmppc_radix_update_pte(kvm, ptep, 0, pgflags, gpa, shift); 763 return true; 764 } 765 return false; 766 } 767 768 int kvmppc_book3s_instantiate_page(struct kvm_vcpu *vcpu, 769 unsigned long gpa, 770 struct kvm_memory_slot *memslot, 771 bool writing, bool kvm_ro, 772 pte_t *inserted_pte, unsigned int *levelp) 773 { 774 struct kvm *kvm = vcpu->kvm; 775 struct page *page = NULL; 776 unsigned long mmu_seq; 777 unsigned long hva, gfn = gpa >> PAGE_SHIFT; 778 bool upgrade_write = false; 779 bool *upgrade_p = &upgrade_write; 780 pte_t pte, *ptep; 781 unsigned int shift, level; 782 int ret; 783 bool large_enable; 784 785 /* used to check for invalidations in progress */ 786 mmu_seq = kvm->mmu_notifier_seq; 787 smp_rmb(); 788 789 /* 790 * Do a fast check first, since __gfn_to_pfn_memslot doesn't 791 * do it with !atomic && !async, which is how we call it. 792 * We always ask for write permission since the common case 793 * is that the page is writable. 794 */ 795 hva = gfn_to_hva_memslot(memslot, gfn); 796 if (!kvm_ro && __get_user_pages_fast(hva, 1, 1, &page) == 1) { 797 upgrade_write = true; 798 } else { 799 unsigned long pfn; 800 801 /* Call KVM generic code to do the slow-path check */ 802 pfn = __gfn_to_pfn_memslot(memslot, gfn, false, NULL, 803 writing, upgrade_p); 804 if (is_error_noslot_pfn(pfn)) 805 return -EFAULT; 806 page = NULL; 807 if (pfn_valid(pfn)) { 808 page = pfn_to_page(pfn); 809 if (PageReserved(page)) 810 page = NULL; 811 } 812 } 813 814 /* 815 * Read the PTE from the process' radix tree and use that 816 * so we get the shift and attribute bits. 817 */ 818 local_irq_disable(); 819 ptep = __find_linux_pte(vcpu->arch.pgdir, hva, NULL, &shift); 820 /* 821 * If the PTE disappeared temporarily due to a THP 822 * collapse, just return and let the guest try again. 823 */ 824 if (!ptep) { 825 local_irq_enable(); 826 if (page) 827 put_page(page); 828 return RESUME_GUEST; 829 } 830 pte = *ptep; 831 local_irq_enable(); 832 833 /* If we're logging dirty pages, always map single pages */ 834 large_enable = !(memslot->flags & KVM_MEM_LOG_DIRTY_PAGES); 835 836 /* Get pte level from shift/size */ 837 if (large_enable && shift == PUD_SHIFT && 838 (gpa & (PUD_SIZE - PAGE_SIZE)) == 839 (hva & (PUD_SIZE - PAGE_SIZE))) { 840 level = 2; 841 } else if (large_enable && shift == PMD_SHIFT && 842 (gpa & (PMD_SIZE - PAGE_SIZE)) == 843 (hva & (PMD_SIZE - PAGE_SIZE))) { 844 level = 1; 845 } else { 846 level = 0; 847 if (shift > PAGE_SHIFT) { 848 /* 849 * If the pte maps more than one page, bring over 850 * bits from the virtual address to get the real 851 * address of the specific single page we want. 852 */ 853 unsigned long rpnmask = (1ul << shift) - PAGE_SIZE; 854 pte = __pte(pte_val(pte) | (hva & rpnmask)); 855 } 856 } 857 858 pte = __pte(pte_val(pte) | _PAGE_EXEC | _PAGE_ACCESSED); 859 if (writing || upgrade_write) { 860 if (pte_val(pte) & _PAGE_WRITE) 861 pte = __pte(pte_val(pte) | _PAGE_DIRTY); 862 } else { 863 pte = __pte(pte_val(pte) & ~(_PAGE_WRITE | _PAGE_DIRTY)); 864 } 865 866 /* Allocate space in the tree and write the PTE */ 867 ret = kvmppc_create_pte(kvm, kvm->arch.pgtable, pte, gpa, level, 868 mmu_seq, kvm->arch.lpid, NULL, NULL); 869 if (inserted_pte) 870 *inserted_pte = pte; 871 if (levelp) 872 *levelp = level; 873 874 if (page) { 875 if (!ret && (pte_val(pte) & _PAGE_WRITE)) 876 set_page_dirty_lock(page); 877 put_page(page); 878 } 879 880 /* Increment number of large pages if we (successfully) inserted one */ 881 if (!ret) { 882 if (level == 1) 883 kvm->stat.num_2M_pages++; 884 else if (level == 2) 885 kvm->stat.num_1G_pages++; 886 } 887 888 return ret; 889 } 890 891 int kvmppc_book3s_radix_page_fault(struct kvm_run *run, struct kvm_vcpu *vcpu, 892 unsigned long ea, unsigned long dsisr) 893 { 894 struct kvm *kvm = vcpu->kvm; 895 unsigned long gpa, gfn; 896 struct kvm_memory_slot *memslot; 897 long ret; 898 bool writing = !!(dsisr & DSISR_ISSTORE); 899 bool kvm_ro = false; 900 901 /* Check for unusual errors */ 902 if (dsisr & DSISR_UNSUPP_MMU) { 903 pr_err("KVM: Got unsupported MMU fault\n"); 904 return -EFAULT; 905 } 906 if (dsisr & DSISR_BADACCESS) { 907 /* Reflect to the guest as DSI */ 908 pr_err("KVM: Got radix HV page fault with DSISR=%lx\n", dsisr); 909 kvmppc_core_queue_data_storage(vcpu, ea, dsisr); 910 return RESUME_GUEST; 911 } 912 913 /* Translate the logical address */ 914 gpa = vcpu->arch.fault_gpa & ~0xfffUL; 915 gpa &= ~0xF000000000000000ul; 916 gfn = gpa >> PAGE_SHIFT; 917 if (!(dsisr & DSISR_PRTABLE_FAULT)) 918 gpa |= ea & 0xfff; 919 920 if (kvm->arch.secure_guest & KVMPPC_SECURE_INIT_DONE) 921 return kvmppc_send_page_to_uv(kvm, gfn); 922 923 /* Get the corresponding memslot */ 924 memslot = gfn_to_memslot(kvm, gfn); 925 926 /* No memslot means it's an emulated MMIO region */ 927 if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID)) { 928 if (dsisr & (DSISR_PRTABLE_FAULT | DSISR_BADACCESS | 929 DSISR_SET_RC)) { 930 /* 931 * Bad address in guest page table tree, or other 932 * unusual error - reflect it to the guest as DSI. 933 */ 934 kvmppc_core_queue_data_storage(vcpu, ea, dsisr); 935 return RESUME_GUEST; 936 } 937 return kvmppc_hv_emulate_mmio(run, vcpu, gpa, ea, writing); 938 } 939 940 if (memslot->flags & KVM_MEM_READONLY) { 941 if (writing) { 942 /* give the guest a DSI */ 943 kvmppc_core_queue_data_storage(vcpu, ea, DSISR_ISSTORE | 944 DSISR_PROTFAULT); 945 return RESUME_GUEST; 946 } 947 kvm_ro = true; 948 } 949 950 /* Failed to set the reference/change bits */ 951 if (dsisr & DSISR_SET_RC) { 952 spin_lock(&kvm->mmu_lock); 953 if (kvmppc_hv_handle_set_rc(kvm, kvm->arch.pgtable, 954 writing, gpa, kvm->arch.lpid)) 955 dsisr &= ~DSISR_SET_RC; 956 spin_unlock(&kvm->mmu_lock); 957 958 if (!(dsisr & (DSISR_BAD_FAULT_64S | DSISR_NOHPTE | 959 DSISR_PROTFAULT | DSISR_SET_RC))) 960 return RESUME_GUEST; 961 } 962 963 /* Try to insert a pte */ 964 ret = kvmppc_book3s_instantiate_page(vcpu, gpa, memslot, writing, 965 kvm_ro, NULL, NULL); 966 967 if (ret == 0 || ret == -EAGAIN) 968 ret = RESUME_GUEST; 969 return ret; 970 } 971 972 /* Called with kvm->mmu_lock held */ 973 int kvm_unmap_radix(struct kvm *kvm, struct kvm_memory_slot *memslot, 974 unsigned long gfn) 975 { 976 pte_t *ptep; 977 unsigned long gpa = gfn << PAGE_SHIFT; 978 unsigned int shift; 979 980 if (kvm->arch.secure_guest & KVMPPC_SECURE_INIT_DONE) { 981 uv_page_inval(kvm->arch.lpid, gpa, PAGE_SHIFT); 982 return 0; 983 } 984 985 ptep = __find_linux_pte(kvm->arch.pgtable, gpa, NULL, &shift); 986 if (ptep && pte_present(*ptep)) 987 kvmppc_unmap_pte(kvm, ptep, gpa, shift, memslot, 988 kvm->arch.lpid); 989 return 0; 990 } 991 992 /* Called with kvm->mmu_lock held */ 993 int kvm_age_radix(struct kvm *kvm, struct kvm_memory_slot *memslot, 994 unsigned long gfn) 995 { 996 pte_t *ptep; 997 unsigned long gpa = gfn << PAGE_SHIFT; 998 unsigned int shift; 999 int ref = 0; 1000 unsigned long old, *rmapp; 1001 1002 if (kvm->arch.secure_guest & KVMPPC_SECURE_INIT_DONE) 1003 return ref; 1004 1005 ptep = __find_linux_pte(kvm->arch.pgtable, gpa, NULL, &shift); 1006 if (ptep && pte_present(*ptep) && pte_young(*ptep)) { 1007 old = kvmppc_radix_update_pte(kvm, ptep, _PAGE_ACCESSED, 0, 1008 gpa, shift); 1009 /* XXX need to flush tlb here? */ 1010 /* Also clear bit in ptes in shadow pgtable for nested guests */ 1011 rmapp = &memslot->arch.rmap[gfn - memslot->base_gfn]; 1012 kvmhv_update_nest_rmap_rc_list(kvm, rmapp, _PAGE_ACCESSED, 0, 1013 old & PTE_RPN_MASK, 1014 1UL << shift); 1015 ref = 1; 1016 } 1017 return ref; 1018 } 1019 1020 /* Called with kvm->mmu_lock held */ 1021 int kvm_test_age_radix(struct kvm *kvm, struct kvm_memory_slot *memslot, 1022 unsigned long gfn) 1023 { 1024 pte_t *ptep; 1025 unsigned long gpa = gfn << PAGE_SHIFT; 1026 unsigned int shift; 1027 int ref = 0; 1028 1029 if (kvm->arch.secure_guest & KVMPPC_SECURE_INIT_DONE) 1030 return ref; 1031 1032 ptep = __find_linux_pte(kvm->arch.pgtable, gpa, NULL, &shift); 1033 if (ptep && pte_present(*ptep) && pte_young(*ptep)) 1034 ref = 1; 1035 return ref; 1036 } 1037 1038 /* Returns the number of PAGE_SIZE pages that are dirty */ 1039 static int kvm_radix_test_clear_dirty(struct kvm *kvm, 1040 struct kvm_memory_slot *memslot, int pagenum) 1041 { 1042 unsigned long gfn = memslot->base_gfn + pagenum; 1043 unsigned long gpa = gfn << PAGE_SHIFT; 1044 pte_t *ptep; 1045 unsigned int shift; 1046 int ret = 0; 1047 unsigned long old, *rmapp; 1048 1049 if (kvm->arch.secure_guest & KVMPPC_SECURE_INIT_DONE) 1050 return ret; 1051 1052 ptep = __find_linux_pte(kvm->arch.pgtable, gpa, NULL, &shift); 1053 if (ptep && pte_present(*ptep) && pte_dirty(*ptep)) { 1054 ret = 1; 1055 if (shift) 1056 ret = 1 << (shift - PAGE_SHIFT); 1057 spin_lock(&kvm->mmu_lock); 1058 old = kvmppc_radix_update_pte(kvm, ptep, _PAGE_DIRTY, 0, 1059 gpa, shift); 1060 kvmppc_radix_tlbie_page(kvm, gpa, shift, kvm->arch.lpid); 1061 /* Also clear bit in ptes in shadow pgtable for nested guests */ 1062 rmapp = &memslot->arch.rmap[gfn - memslot->base_gfn]; 1063 kvmhv_update_nest_rmap_rc_list(kvm, rmapp, _PAGE_DIRTY, 0, 1064 old & PTE_RPN_MASK, 1065 1UL << shift); 1066 spin_unlock(&kvm->mmu_lock); 1067 } 1068 return ret; 1069 } 1070 1071 long kvmppc_hv_get_dirty_log_radix(struct kvm *kvm, 1072 struct kvm_memory_slot *memslot, unsigned long *map) 1073 { 1074 unsigned long i, j; 1075 int npages; 1076 1077 for (i = 0; i < memslot->npages; i = j) { 1078 npages = kvm_radix_test_clear_dirty(kvm, memslot, i); 1079 1080 /* 1081 * Note that if npages > 0 then i must be a multiple of npages, 1082 * since huge pages are only used to back the guest at guest 1083 * real addresses that are a multiple of their size. 1084 * Since we have at most one PTE covering any given guest 1085 * real address, if npages > 1 we can skip to i + npages. 1086 */ 1087 j = i + 1; 1088 if (npages) { 1089 set_dirty_bits(map, i, npages); 1090 j = i + npages; 1091 } 1092 } 1093 return 0; 1094 } 1095 1096 void kvmppc_radix_flush_memslot(struct kvm *kvm, 1097 const struct kvm_memory_slot *memslot) 1098 { 1099 unsigned long n; 1100 pte_t *ptep; 1101 unsigned long gpa; 1102 unsigned int shift; 1103 1104 if (kvm->arch.secure_guest & KVMPPC_SECURE_INIT_START) 1105 kvmppc_uvmem_drop_pages(memslot, kvm); 1106 1107 if (kvm->arch.secure_guest & KVMPPC_SECURE_INIT_DONE) 1108 return; 1109 1110 gpa = memslot->base_gfn << PAGE_SHIFT; 1111 spin_lock(&kvm->mmu_lock); 1112 for (n = memslot->npages; n; --n) { 1113 ptep = __find_linux_pte(kvm->arch.pgtable, gpa, NULL, &shift); 1114 if (ptep && pte_present(*ptep)) 1115 kvmppc_unmap_pte(kvm, ptep, gpa, shift, memslot, 1116 kvm->arch.lpid); 1117 gpa += PAGE_SIZE; 1118 } 1119 spin_unlock(&kvm->mmu_lock); 1120 } 1121 1122 static void add_rmmu_ap_encoding(struct kvm_ppc_rmmu_info *info, 1123 int psize, int *indexp) 1124 { 1125 if (!mmu_psize_defs[psize].shift) 1126 return; 1127 info->ap_encodings[*indexp] = mmu_psize_defs[psize].shift | 1128 (mmu_psize_defs[psize].ap << 29); 1129 ++(*indexp); 1130 } 1131 1132 int kvmhv_get_rmmu_info(struct kvm *kvm, struct kvm_ppc_rmmu_info *info) 1133 { 1134 int i; 1135 1136 if (!radix_enabled()) 1137 return -EINVAL; 1138 memset(info, 0, sizeof(*info)); 1139 1140 /* 4k page size */ 1141 info->geometries[0].page_shift = 12; 1142 info->geometries[0].level_bits[0] = 9; 1143 for (i = 1; i < 4; ++i) 1144 info->geometries[0].level_bits[i] = p9_supported_radix_bits[i]; 1145 /* 64k page size */ 1146 info->geometries[1].page_shift = 16; 1147 for (i = 0; i < 4; ++i) 1148 info->geometries[1].level_bits[i] = p9_supported_radix_bits[i]; 1149 1150 i = 0; 1151 add_rmmu_ap_encoding(info, MMU_PAGE_4K, &i); 1152 add_rmmu_ap_encoding(info, MMU_PAGE_64K, &i); 1153 add_rmmu_ap_encoding(info, MMU_PAGE_2M, &i); 1154 add_rmmu_ap_encoding(info, MMU_PAGE_1G, &i); 1155 1156 return 0; 1157 } 1158 1159 int kvmppc_init_vm_radix(struct kvm *kvm) 1160 { 1161 kvm->arch.pgtable = pgd_alloc(kvm->mm); 1162 if (!kvm->arch.pgtable) 1163 return -ENOMEM; 1164 return 0; 1165 } 1166 1167 static void pte_ctor(void *addr) 1168 { 1169 memset(addr, 0, RADIX_PTE_TABLE_SIZE); 1170 } 1171 1172 static void pmd_ctor(void *addr) 1173 { 1174 memset(addr, 0, RADIX_PMD_TABLE_SIZE); 1175 } 1176 1177 struct debugfs_radix_state { 1178 struct kvm *kvm; 1179 struct mutex mutex; 1180 unsigned long gpa; 1181 int lpid; 1182 int chars_left; 1183 int buf_index; 1184 char buf[128]; 1185 u8 hdr; 1186 }; 1187 1188 static int debugfs_radix_open(struct inode *inode, struct file *file) 1189 { 1190 struct kvm *kvm = inode->i_private; 1191 struct debugfs_radix_state *p; 1192 1193 p = kzalloc(sizeof(*p), GFP_KERNEL); 1194 if (!p) 1195 return -ENOMEM; 1196 1197 kvm_get_kvm(kvm); 1198 p->kvm = kvm; 1199 mutex_init(&p->mutex); 1200 file->private_data = p; 1201 1202 return nonseekable_open(inode, file); 1203 } 1204 1205 static int debugfs_radix_release(struct inode *inode, struct file *file) 1206 { 1207 struct debugfs_radix_state *p = file->private_data; 1208 1209 kvm_put_kvm(p->kvm); 1210 kfree(p); 1211 return 0; 1212 } 1213 1214 static ssize_t debugfs_radix_read(struct file *file, char __user *buf, 1215 size_t len, loff_t *ppos) 1216 { 1217 struct debugfs_radix_state *p = file->private_data; 1218 ssize_t ret, r; 1219 unsigned long n; 1220 struct kvm *kvm; 1221 unsigned long gpa; 1222 pgd_t *pgt; 1223 struct kvm_nested_guest *nested; 1224 pgd_t pgd, *pgdp; 1225 pud_t pud, *pudp; 1226 pmd_t pmd, *pmdp; 1227 pte_t *ptep; 1228 int shift; 1229 unsigned long pte; 1230 1231 kvm = p->kvm; 1232 if (!kvm_is_radix(kvm)) 1233 return 0; 1234 1235 ret = mutex_lock_interruptible(&p->mutex); 1236 if (ret) 1237 return ret; 1238 1239 if (p->chars_left) { 1240 n = p->chars_left; 1241 if (n > len) 1242 n = len; 1243 r = copy_to_user(buf, p->buf + p->buf_index, n); 1244 n -= r; 1245 p->chars_left -= n; 1246 p->buf_index += n; 1247 buf += n; 1248 len -= n; 1249 ret = n; 1250 if (r) { 1251 if (!n) 1252 ret = -EFAULT; 1253 goto out; 1254 } 1255 } 1256 1257 gpa = p->gpa; 1258 nested = NULL; 1259 pgt = NULL; 1260 while (len != 0 && p->lpid >= 0) { 1261 if (gpa >= RADIX_PGTABLE_RANGE) { 1262 gpa = 0; 1263 pgt = NULL; 1264 if (nested) { 1265 kvmhv_put_nested(nested); 1266 nested = NULL; 1267 } 1268 p->lpid = kvmhv_nested_next_lpid(kvm, p->lpid); 1269 p->hdr = 0; 1270 if (p->lpid < 0) 1271 break; 1272 } 1273 if (!pgt) { 1274 if (p->lpid == 0) { 1275 pgt = kvm->arch.pgtable; 1276 } else { 1277 nested = kvmhv_get_nested(kvm, p->lpid, false); 1278 if (!nested) { 1279 gpa = RADIX_PGTABLE_RANGE; 1280 continue; 1281 } 1282 pgt = nested->shadow_pgtable; 1283 } 1284 } 1285 n = 0; 1286 if (!p->hdr) { 1287 if (p->lpid > 0) 1288 n = scnprintf(p->buf, sizeof(p->buf), 1289 "\nNested LPID %d: ", p->lpid); 1290 n += scnprintf(p->buf + n, sizeof(p->buf) - n, 1291 "pgdir: %lx\n", (unsigned long)pgt); 1292 p->hdr = 1; 1293 goto copy; 1294 } 1295 1296 pgdp = pgt + pgd_index(gpa); 1297 pgd = READ_ONCE(*pgdp); 1298 if (!(pgd_val(pgd) & _PAGE_PRESENT)) { 1299 gpa = (gpa & PGDIR_MASK) + PGDIR_SIZE; 1300 continue; 1301 } 1302 1303 pudp = pud_offset(&pgd, gpa); 1304 pud = READ_ONCE(*pudp); 1305 if (!(pud_val(pud) & _PAGE_PRESENT)) { 1306 gpa = (gpa & PUD_MASK) + PUD_SIZE; 1307 continue; 1308 } 1309 if (pud_val(pud) & _PAGE_PTE) { 1310 pte = pud_val(pud); 1311 shift = PUD_SHIFT; 1312 goto leaf; 1313 } 1314 1315 pmdp = pmd_offset(&pud, gpa); 1316 pmd = READ_ONCE(*pmdp); 1317 if (!(pmd_val(pmd) & _PAGE_PRESENT)) { 1318 gpa = (gpa & PMD_MASK) + PMD_SIZE; 1319 continue; 1320 } 1321 if (pmd_val(pmd) & _PAGE_PTE) { 1322 pte = pmd_val(pmd); 1323 shift = PMD_SHIFT; 1324 goto leaf; 1325 } 1326 1327 ptep = pte_offset_kernel(&pmd, gpa); 1328 pte = pte_val(READ_ONCE(*ptep)); 1329 if (!(pte & _PAGE_PRESENT)) { 1330 gpa += PAGE_SIZE; 1331 continue; 1332 } 1333 shift = PAGE_SHIFT; 1334 leaf: 1335 n = scnprintf(p->buf, sizeof(p->buf), 1336 " %lx: %lx %d\n", gpa, pte, shift); 1337 gpa += 1ul << shift; 1338 copy: 1339 p->chars_left = n; 1340 if (n > len) 1341 n = len; 1342 r = copy_to_user(buf, p->buf, n); 1343 n -= r; 1344 p->chars_left -= n; 1345 p->buf_index = n; 1346 buf += n; 1347 len -= n; 1348 ret += n; 1349 if (r) { 1350 if (!ret) 1351 ret = -EFAULT; 1352 break; 1353 } 1354 } 1355 p->gpa = gpa; 1356 if (nested) 1357 kvmhv_put_nested(nested); 1358 1359 out: 1360 mutex_unlock(&p->mutex); 1361 return ret; 1362 } 1363 1364 static ssize_t debugfs_radix_write(struct file *file, const char __user *buf, 1365 size_t len, loff_t *ppos) 1366 { 1367 return -EACCES; 1368 } 1369 1370 static const struct file_operations debugfs_radix_fops = { 1371 .owner = THIS_MODULE, 1372 .open = debugfs_radix_open, 1373 .release = debugfs_radix_release, 1374 .read = debugfs_radix_read, 1375 .write = debugfs_radix_write, 1376 .llseek = generic_file_llseek, 1377 }; 1378 1379 void kvmhv_radix_debugfs_init(struct kvm *kvm) 1380 { 1381 kvm->arch.radix_dentry = debugfs_create_file("radix", 0400, 1382 kvm->arch.debugfs_dir, kvm, 1383 &debugfs_radix_fops); 1384 } 1385 1386 int kvmppc_radix_init(void) 1387 { 1388 unsigned long size = sizeof(void *) << RADIX_PTE_INDEX_SIZE; 1389 1390 kvm_pte_cache = kmem_cache_create("kvm-pte", size, size, 0, pte_ctor); 1391 if (!kvm_pte_cache) 1392 return -ENOMEM; 1393 1394 size = sizeof(void *) << RADIX_PMD_INDEX_SIZE; 1395 1396 kvm_pmd_cache = kmem_cache_create("kvm-pmd", size, size, 0, pmd_ctor); 1397 if (!kvm_pmd_cache) { 1398 kmem_cache_destroy(kvm_pte_cache); 1399 return -ENOMEM; 1400 } 1401 1402 return 0; 1403 } 1404 1405 void kvmppc_radix_exit(void) 1406 { 1407 kmem_cache_destroy(kvm_pte_cache); 1408 kmem_cache_destroy(kvm_pmd_cache); 1409 } 1410