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 if (is_load) 67 ret = probe_user_read(to, (const void __user *)from, n); 68 else 69 ret = probe_user_write((void __user *)to, from, n); 70 71 /* switch the pid first to avoid running host with unallocated pid */ 72 if (quadrant == 1 && pid != old_pid) 73 mtspr(SPRN_PID, old_pid); 74 if (lpid != old_lpid) 75 mtspr(SPRN_LPID, old_lpid); 76 isync(); 77 78 preempt_enable(); 79 80 return ret; 81 } 82 EXPORT_SYMBOL_GPL(__kvmhv_copy_tofrom_guest_radix); 83 84 static long kvmhv_copy_tofrom_guest_radix(struct kvm_vcpu *vcpu, gva_t eaddr, 85 void *to, void *from, unsigned long n) 86 { 87 int lpid = vcpu->kvm->arch.lpid; 88 int pid = vcpu->arch.pid; 89 90 /* This would cause a data segment intr so don't allow the access */ 91 if (eaddr & (0x3FFUL << 52)) 92 return -EINVAL; 93 94 /* Should we be using the nested lpid */ 95 if (vcpu->arch.nested) 96 lpid = vcpu->arch.nested->shadow_lpid; 97 98 /* If accessing quadrant 3 then pid is expected to be 0 */ 99 if (((eaddr >> 62) & 0x3) == 0x3) 100 pid = 0; 101 102 eaddr &= ~(0xFFFUL << 52); 103 104 return __kvmhv_copy_tofrom_guest_radix(lpid, pid, eaddr, to, from, n); 105 } 106 107 long kvmhv_copy_from_guest_radix(struct kvm_vcpu *vcpu, gva_t eaddr, void *to, 108 unsigned long n) 109 { 110 long ret; 111 112 ret = kvmhv_copy_tofrom_guest_radix(vcpu, eaddr, to, NULL, n); 113 if (ret > 0) 114 memset(to + (n - ret), 0, ret); 115 116 return ret; 117 } 118 EXPORT_SYMBOL_GPL(kvmhv_copy_from_guest_radix); 119 120 long kvmhv_copy_to_guest_radix(struct kvm_vcpu *vcpu, gva_t eaddr, void *from, 121 unsigned long n) 122 { 123 return kvmhv_copy_tofrom_guest_radix(vcpu, eaddr, NULL, from, n); 124 } 125 EXPORT_SYMBOL_GPL(kvmhv_copy_to_guest_radix); 126 127 int kvmppc_mmu_walk_radix_tree(struct kvm_vcpu *vcpu, gva_t eaddr, 128 struct kvmppc_pte *gpte, u64 root, 129 u64 *pte_ret_p) 130 { 131 struct kvm *kvm = vcpu->kvm; 132 int ret, level, ps; 133 unsigned long rts, bits, offset, index; 134 u64 pte, base, gpa; 135 __be64 rpte; 136 137 rts = ((root & RTS1_MASK) >> (RTS1_SHIFT - 3)) | 138 ((root & RTS2_MASK) >> RTS2_SHIFT); 139 bits = root & RPDS_MASK; 140 base = root & RPDB_MASK; 141 142 offset = rts + 31; 143 144 /* Current implementations only support 52-bit space */ 145 if (offset != 52) 146 return -EINVAL; 147 148 /* Walk each level of the radix tree */ 149 for (level = 3; level >= 0; --level) { 150 u64 addr; 151 /* Check a valid size */ 152 if (level && bits != p9_supported_radix_bits[level]) 153 return -EINVAL; 154 if (level == 0 && !(bits == 5 || bits == 9)) 155 return -EINVAL; 156 offset -= bits; 157 index = (eaddr >> offset) & ((1UL << bits) - 1); 158 /* Check that low bits of page table base are zero */ 159 if (base & ((1UL << (bits + 3)) - 1)) 160 return -EINVAL; 161 /* Read the entry from guest memory */ 162 addr = base + (index * sizeof(rpte)); 163 ret = kvm_read_guest(kvm, addr, &rpte, sizeof(rpte)); 164 if (ret) { 165 if (pte_ret_p) 166 *pte_ret_p = addr; 167 return ret; 168 } 169 pte = __be64_to_cpu(rpte); 170 if (!(pte & _PAGE_PRESENT)) 171 return -ENOENT; 172 /* Check if a leaf entry */ 173 if (pte & _PAGE_PTE) 174 break; 175 /* Get ready to walk the next level */ 176 base = pte & RPDB_MASK; 177 bits = pte & RPDS_MASK; 178 } 179 180 /* Need a leaf at lowest level; 512GB pages not supported */ 181 if (level < 0 || level == 3) 182 return -EINVAL; 183 184 /* We found a valid leaf PTE */ 185 /* Offset is now log base 2 of the page size */ 186 gpa = pte & 0x01fffffffffff000ul; 187 if (gpa & ((1ul << offset) - 1)) 188 return -EINVAL; 189 gpa |= eaddr & ((1ul << offset) - 1); 190 for (ps = MMU_PAGE_4K; ps < MMU_PAGE_COUNT; ++ps) 191 if (offset == mmu_psize_defs[ps].shift) 192 break; 193 gpte->page_size = ps; 194 gpte->page_shift = offset; 195 196 gpte->eaddr = eaddr; 197 gpte->raddr = gpa; 198 199 /* Work out permissions */ 200 gpte->may_read = !!(pte & _PAGE_READ); 201 gpte->may_write = !!(pte & _PAGE_WRITE); 202 gpte->may_execute = !!(pte & _PAGE_EXEC); 203 204 gpte->rc = pte & (_PAGE_ACCESSED | _PAGE_DIRTY); 205 206 if (pte_ret_p) 207 *pte_ret_p = pte; 208 209 return 0; 210 } 211 212 /* 213 * Used to walk a partition or process table radix tree in guest memory 214 * Note: We exploit the fact that a partition table and a process 215 * table have the same layout, a partition-scoped page table and a 216 * process-scoped page table have the same layout, and the 2nd 217 * doubleword of a partition table entry has the same layout as 218 * the PTCR register. 219 */ 220 int kvmppc_mmu_radix_translate_table(struct kvm_vcpu *vcpu, gva_t eaddr, 221 struct kvmppc_pte *gpte, u64 table, 222 int table_index, u64 *pte_ret_p) 223 { 224 struct kvm *kvm = vcpu->kvm; 225 int ret; 226 unsigned long size, ptbl, root; 227 struct prtb_entry entry; 228 229 if ((table & PRTS_MASK) > 24) 230 return -EINVAL; 231 size = 1ul << ((table & PRTS_MASK) + 12); 232 233 /* Is the table big enough to contain this entry? */ 234 if ((table_index * sizeof(entry)) >= size) 235 return -EINVAL; 236 237 /* Read the table to find the root of the radix tree */ 238 ptbl = (table & PRTB_MASK) + (table_index * sizeof(entry)); 239 ret = kvm_read_guest(kvm, ptbl, &entry, sizeof(entry)); 240 if (ret) 241 return ret; 242 243 /* Root is stored in the first double word */ 244 root = be64_to_cpu(entry.prtb0); 245 246 return kvmppc_mmu_walk_radix_tree(vcpu, eaddr, gpte, root, pte_ret_p); 247 } 248 249 int kvmppc_mmu_radix_xlate(struct kvm_vcpu *vcpu, gva_t eaddr, 250 struct kvmppc_pte *gpte, bool data, bool iswrite) 251 { 252 u32 pid; 253 u64 pte; 254 int ret; 255 256 /* Work out effective PID */ 257 switch (eaddr >> 62) { 258 case 0: 259 pid = vcpu->arch.pid; 260 break; 261 case 3: 262 pid = 0; 263 break; 264 default: 265 return -EINVAL; 266 } 267 268 ret = kvmppc_mmu_radix_translate_table(vcpu, eaddr, gpte, 269 vcpu->kvm->arch.process_table, pid, &pte); 270 if (ret) 271 return ret; 272 273 /* Check privilege (applies only to process scoped translations) */ 274 if (kvmppc_get_msr(vcpu) & MSR_PR) { 275 if (pte & _PAGE_PRIVILEGED) { 276 gpte->may_read = 0; 277 gpte->may_write = 0; 278 gpte->may_execute = 0; 279 } 280 } else { 281 if (!(pte & _PAGE_PRIVILEGED)) { 282 /* Check AMR/IAMR to see if strict mode is in force */ 283 if (vcpu->arch.amr & (1ul << 62)) 284 gpte->may_read = 0; 285 if (vcpu->arch.amr & (1ul << 63)) 286 gpte->may_write = 0; 287 if (vcpu->arch.iamr & (1ul << 62)) 288 gpte->may_execute = 0; 289 } 290 } 291 292 return 0; 293 } 294 295 void kvmppc_radix_tlbie_page(struct kvm *kvm, unsigned long addr, 296 unsigned int pshift, unsigned int lpid) 297 { 298 unsigned long psize = PAGE_SIZE; 299 int psi; 300 long rc; 301 unsigned long rb; 302 303 if (pshift) 304 psize = 1UL << pshift; 305 else 306 pshift = PAGE_SHIFT; 307 308 addr &= ~(psize - 1); 309 310 if (!kvmhv_on_pseries()) { 311 radix__flush_tlb_lpid_page(lpid, addr, psize); 312 return; 313 } 314 315 psi = shift_to_mmu_psize(pshift); 316 rb = addr | (mmu_get_ap(psi) << PPC_BITLSHIFT(58)); 317 rc = plpar_hcall_norets(H_TLB_INVALIDATE, H_TLBIE_P1_ENC(0, 0, 1), 318 lpid, rb); 319 if (rc) 320 pr_err("KVM: TLB page invalidation hcall failed, rc=%ld\n", rc); 321 } 322 323 static void kvmppc_radix_flush_pwc(struct kvm *kvm, unsigned int lpid) 324 { 325 long rc; 326 327 if (!kvmhv_on_pseries()) { 328 radix__flush_pwc_lpid(lpid); 329 return; 330 } 331 332 rc = plpar_hcall_norets(H_TLB_INVALIDATE, H_TLBIE_P1_ENC(1, 0, 1), 333 lpid, TLBIEL_INVAL_SET_LPID); 334 if (rc) 335 pr_err("KVM: TLB PWC invalidation hcall failed, rc=%ld\n", rc); 336 } 337 338 static unsigned long kvmppc_radix_update_pte(struct kvm *kvm, pte_t *ptep, 339 unsigned long clr, unsigned long set, 340 unsigned long addr, unsigned int shift) 341 { 342 return __radix_pte_update(ptep, clr, set); 343 } 344 345 void kvmppc_radix_set_pte_at(struct kvm *kvm, unsigned long addr, 346 pte_t *ptep, pte_t pte) 347 { 348 radix__set_pte_at(kvm->mm, addr, ptep, pte, 0); 349 } 350 351 static struct kmem_cache *kvm_pte_cache; 352 static struct kmem_cache *kvm_pmd_cache; 353 354 static pte_t *kvmppc_pte_alloc(void) 355 { 356 return kmem_cache_alloc(kvm_pte_cache, GFP_KERNEL); 357 } 358 359 static void kvmppc_pte_free(pte_t *ptep) 360 { 361 kmem_cache_free(kvm_pte_cache, ptep); 362 } 363 364 static pmd_t *kvmppc_pmd_alloc(void) 365 { 366 return kmem_cache_alloc(kvm_pmd_cache, GFP_KERNEL); 367 } 368 369 static void kvmppc_pmd_free(pmd_t *pmdp) 370 { 371 kmem_cache_free(kvm_pmd_cache, pmdp); 372 } 373 374 /* Called with kvm->mmu_lock held */ 375 void kvmppc_unmap_pte(struct kvm *kvm, pte_t *pte, unsigned long gpa, 376 unsigned int shift, 377 const struct kvm_memory_slot *memslot, 378 unsigned int lpid) 379 380 { 381 unsigned long old; 382 unsigned long gfn = gpa >> PAGE_SHIFT; 383 unsigned long page_size = PAGE_SIZE; 384 unsigned long hpa; 385 386 old = kvmppc_radix_update_pte(kvm, pte, ~0UL, 0, gpa, shift); 387 kvmppc_radix_tlbie_page(kvm, gpa, shift, lpid); 388 389 /* The following only applies to L1 entries */ 390 if (lpid != kvm->arch.lpid) 391 return; 392 393 if (!memslot) { 394 memslot = gfn_to_memslot(kvm, gfn); 395 if (!memslot) 396 return; 397 } 398 if (shift) { /* 1GB or 2MB page */ 399 page_size = 1ul << shift; 400 if (shift == PMD_SHIFT) 401 kvm->stat.num_2M_pages--; 402 else if (shift == PUD_SHIFT) 403 kvm->stat.num_1G_pages--; 404 } 405 406 gpa &= ~(page_size - 1); 407 hpa = old & PTE_RPN_MASK; 408 kvmhv_remove_nest_rmap_range(kvm, memslot, gpa, hpa, page_size); 409 410 if ((old & _PAGE_DIRTY) && memslot->dirty_bitmap) 411 kvmppc_update_dirty_map(memslot, gfn, page_size); 412 } 413 414 /* 415 * kvmppc_free_p?d are used to free existing page tables, and recursively 416 * descend and clear and free children. 417 * Callers are responsible for flushing the PWC. 418 * 419 * When page tables are being unmapped/freed as part of page fault path 420 * (full == false), ptes are not expected. There is code to unmap them 421 * and emit a warning if encountered, but there may already be data 422 * corruption due to the unexpected mappings. 423 */ 424 static void kvmppc_unmap_free_pte(struct kvm *kvm, pte_t *pte, bool full, 425 unsigned int lpid) 426 { 427 if (full) { 428 memset(pte, 0, sizeof(long) << RADIX_PTE_INDEX_SIZE); 429 } else { 430 pte_t *p = pte; 431 unsigned long it; 432 433 for (it = 0; it < PTRS_PER_PTE; ++it, ++p) { 434 if (pte_val(*p) == 0) 435 continue; 436 WARN_ON_ONCE(1); 437 kvmppc_unmap_pte(kvm, p, 438 pte_pfn(*p) << PAGE_SHIFT, 439 PAGE_SHIFT, NULL, lpid); 440 } 441 } 442 443 kvmppc_pte_free(pte); 444 } 445 446 static void kvmppc_unmap_free_pmd(struct kvm *kvm, pmd_t *pmd, bool full, 447 unsigned int lpid) 448 { 449 unsigned long im; 450 pmd_t *p = pmd; 451 452 for (im = 0; im < PTRS_PER_PMD; ++im, ++p) { 453 if (!pmd_present(*p)) 454 continue; 455 if (pmd_is_leaf(*p)) { 456 if (full) { 457 pmd_clear(p); 458 } else { 459 WARN_ON_ONCE(1); 460 kvmppc_unmap_pte(kvm, (pte_t *)p, 461 pte_pfn(*(pte_t *)p) << PAGE_SHIFT, 462 PMD_SHIFT, NULL, lpid); 463 } 464 } else { 465 pte_t *pte; 466 467 pte = pte_offset_map(p, 0); 468 kvmppc_unmap_free_pte(kvm, pte, full, lpid); 469 pmd_clear(p); 470 } 471 } 472 kvmppc_pmd_free(pmd); 473 } 474 475 static void kvmppc_unmap_free_pud(struct kvm *kvm, pud_t *pud, 476 unsigned int lpid) 477 { 478 unsigned long iu; 479 pud_t *p = pud; 480 481 for (iu = 0; iu < PTRS_PER_PUD; ++iu, ++p) { 482 if (!pud_present(*p)) 483 continue; 484 if (pud_is_leaf(*p)) { 485 pud_clear(p); 486 } else { 487 pmd_t *pmd; 488 489 pmd = pmd_offset(p, 0); 490 kvmppc_unmap_free_pmd(kvm, pmd, true, lpid); 491 pud_clear(p); 492 } 493 } 494 pud_free(kvm->mm, pud); 495 } 496 497 void kvmppc_free_pgtable_radix(struct kvm *kvm, pgd_t *pgd, unsigned int lpid) 498 { 499 unsigned long ig; 500 501 for (ig = 0; ig < PTRS_PER_PGD; ++ig, ++pgd) { 502 p4d_t *p4d = p4d_offset(pgd, 0); 503 pud_t *pud; 504 505 if (!p4d_present(*p4d)) 506 continue; 507 pud = pud_offset(p4d, 0); 508 kvmppc_unmap_free_pud(kvm, pud, lpid); 509 p4d_clear(p4d); 510 } 511 } 512 513 void kvmppc_free_radix(struct kvm *kvm) 514 { 515 if (kvm->arch.pgtable) { 516 kvmppc_free_pgtable_radix(kvm, kvm->arch.pgtable, 517 kvm->arch.lpid); 518 pgd_free(kvm->mm, kvm->arch.pgtable); 519 kvm->arch.pgtable = NULL; 520 } 521 } 522 523 static void kvmppc_unmap_free_pmd_entry_table(struct kvm *kvm, pmd_t *pmd, 524 unsigned long gpa, unsigned int lpid) 525 { 526 pte_t *pte = pte_offset_kernel(pmd, 0); 527 528 /* 529 * Clearing the pmd entry then flushing the PWC ensures that the pte 530 * page no longer be cached by the MMU, so can be freed without 531 * flushing the PWC again. 532 */ 533 pmd_clear(pmd); 534 kvmppc_radix_flush_pwc(kvm, lpid); 535 536 kvmppc_unmap_free_pte(kvm, pte, false, lpid); 537 } 538 539 static void kvmppc_unmap_free_pud_entry_table(struct kvm *kvm, pud_t *pud, 540 unsigned long gpa, unsigned int lpid) 541 { 542 pmd_t *pmd = pmd_offset(pud, 0); 543 544 /* 545 * Clearing the pud entry then flushing the PWC ensures that the pmd 546 * page and any children pte pages will no longer be cached by the MMU, 547 * so can be freed without flushing the PWC again. 548 */ 549 pud_clear(pud); 550 kvmppc_radix_flush_pwc(kvm, lpid); 551 552 kvmppc_unmap_free_pmd(kvm, pmd, false, lpid); 553 } 554 555 /* 556 * There are a number of bits which may differ between different faults to 557 * the same partition scope entry. RC bits, in the course of cleaning and 558 * aging. And the write bit can change, either the access could have been 559 * upgraded, or a read fault could happen concurrently with a write fault 560 * that sets those bits first. 561 */ 562 #define PTE_BITS_MUST_MATCH (~(_PAGE_WRITE | _PAGE_DIRTY | _PAGE_ACCESSED)) 563 564 int kvmppc_create_pte(struct kvm *kvm, pgd_t *pgtable, pte_t pte, 565 unsigned long gpa, unsigned int level, 566 unsigned long mmu_seq, unsigned int lpid, 567 unsigned long *rmapp, struct rmap_nested **n_rmap) 568 { 569 pgd_t *pgd; 570 p4d_t *p4d; 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 p4d = p4d_offset(pgd, gpa); 579 580 pud = NULL; 581 if (p4d_present(*p4d)) 582 pud = pud_offset(p4d, gpa); 583 else 584 new_pud = pud_alloc_one(kvm->mm, gpa); 585 586 pmd = NULL; 587 if (pud && pud_present(*pud) && !pud_is_leaf(*pud)) 588 pmd = pmd_offset(pud, gpa); 589 else if (level <= 1) 590 new_pmd = kvmppc_pmd_alloc(); 591 592 if (level == 0 && !(pmd && pmd_present(*pmd) && !pmd_is_leaf(*pmd))) 593 new_ptep = kvmppc_pte_alloc(); 594 595 /* Check if we might have been invalidated; let the guest retry if so */ 596 spin_lock(&kvm->mmu_lock); 597 ret = -EAGAIN; 598 if (mmu_notifier_retry(kvm, mmu_seq)) 599 goto out_unlock; 600 601 /* Now traverse again under the lock and change the tree */ 602 ret = -ENOMEM; 603 if (p4d_none(*p4d)) { 604 if (!new_pud) 605 goto out_unlock; 606 p4d_populate(kvm->mm, p4d, new_pud); 607 new_pud = NULL; 608 } 609 pud = pud_offset(p4d, gpa); 610 if (pud_is_leaf(*pud)) { 611 unsigned long hgpa = gpa & PUD_MASK; 612 613 /* Check if we raced and someone else has set the same thing */ 614 if (level == 2) { 615 if (pud_raw(*pud) == pte_raw(pte)) { 616 ret = 0; 617 goto out_unlock; 618 } 619 /* Valid 1GB page here already, add our extra bits */ 620 WARN_ON_ONCE((pud_val(*pud) ^ pte_val(pte)) & 621 PTE_BITS_MUST_MATCH); 622 kvmppc_radix_update_pte(kvm, (pte_t *)pud, 623 0, pte_val(pte), hgpa, PUD_SHIFT); 624 ret = 0; 625 goto out_unlock; 626 } 627 /* 628 * If we raced with another CPU which has just put 629 * a 1GB pte in after we saw a pmd page, try again. 630 */ 631 if (!new_pmd) { 632 ret = -EAGAIN; 633 goto out_unlock; 634 } 635 /* Valid 1GB page here already, remove it */ 636 kvmppc_unmap_pte(kvm, (pte_t *)pud, hgpa, PUD_SHIFT, NULL, 637 lpid); 638 } 639 if (level == 2) { 640 if (!pud_none(*pud)) { 641 /* 642 * There's a page table page here, but we wanted to 643 * install a large page, so remove and free the page 644 * table page. 645 */ 646 kvmppc_unmap_free_pud_entry_table(kvm, pud, gpa, lpid); 647 } 648 kvmppc_radix_set_pte_at(kvm, gpa, (pte_t *)pud, pte); 649 if (rmapp && n_rmap) 650 kvmhv_insert_nest_rmap(kvm, rmapp, n_rmap); 651 ret = 0; 652 goto out_unlock; 653 } 654 if (pud_none(*pud)) { 655 if (!new_pmd) 656 goto out_unlock; 657 pud_populate(kvm->mm, pud, new_pmd); 658 new_pmd = NULL; 659 } 660 pmd = pmd_offset(pud, gpa); 661 if (pmd_is_leaf(*pmd)) { 662 unsigned long lgpa = gpa & PMD_MASK; 663 664 /* Check if we raced and someone else has set the same thing */ 665 if (level == 1) { 666 if (pmd_raw(*pmd) == pte_raw(pte)) { 667 ret = 0; 668 goto out_unlock; 669 } 670 /* Valid 2MB page here already, add our extra bits */ 671 WARN_ON_ONCE((pmd_val(*pmd) ^ pte_val(pte)) & 672 PTE_BITS_MUST_MATCH); 673 kvmppc_radix_update_pte(kvm, pmdp_ptep(pmd), 674 0, pte_val(pte), lgpa, PMD_SHIFT); 675 ret = 0; 676 goto out_unlock; 677 } 678 679 /* 680 * If we raced with another CPU which has just put 681 * a 2MB pte in after we saw a pte page, try again. 682 */ 683 if (!new_ptep) { 684 ret = -EAGAIN; 685 goto out_unlock; 686 } 687 /* Valid 2MB page here already, remove it */ 688 kvmppc_unmap_pte(kvm, pmdp_ptep(pmd), lgpa, PMD_SHIFT, NULL, 689 lpid); 690 } 691 if (level == 1) { 692 if (!pmd_none(*pmd)) { 693 /* 694 * There's a page table page here, but we wanted to 695 * install a large page, so remove and free the page 696 * table page. 697 */ 698 kvmppc_unmap_free_pmd_entry_table(kvm, pmd, gpa, lpid); 699 } 700 kvmppc_radix_set_pte_at(kvm, gpa, pmdp_ptep(pmd), pte); 701 if (rmapp && n_rmap) 702 kvmhv_insert_nest_rmap(kvm, rmapp, n_rmap); 703 ret = 0; 704 goto out_unlock; 705 } 706 if (pmd_none(*pmd)) { 707 if (!new_ptep) 708 goto out_unlock; 709 pmd_populate(kvm->mm, pmd, new_ptep); 710 new_ptep = NULL; 711 } 712 ptep = pte_offset_kernel(pmd, gpa); 713 if (pte_present(*ptep)) { 714 /* Check if someone else set the same thing */ 715 if (pte_raw(*ptep) == pte_raw(pte)) { 716 ret = 0; 717 goto out_unlock; 718 } 719 /* Valid page here already, add our extra bits */ 720 WARN_ON_ONCE((pte_val(*ptep) ^ pte_val(pte)) & 721 PTE_BITS_MUST_MATCH); 722 kvmppc_radix_update_pte(kvm, ptep, 0, pte_val(pte), gpa, 0); 723 ret = 0; 724 goto out_unlock; 725 } 726 kvmppc_radix_set_pte_at(kvm, gpa, ptep, pte); 727 if (rmapp && n_rmap) 728 kvmhv_insert_nest_rmap(kvm, rmapp, n_rmap); 729 ret = 0; 730 731 out_unlock: 732 spin_unlock(&kvm->mmu_lock); 733 if (new_pud) 734 pud_free(kvm->mm, new_pud); 735 if (new_pmd) 736 kvmppc_pmd_free(new_pmd); 737 if (new_ptep) 738 kvmppc_pte_free(new_ptep); 739 return ret; 740 } 741 742 bool kvmppc_hv_handle_set_rc(struct kvm *kvm, pgd_t *pgtable, bool writing, 743 unsigned long gpa, unsigned int lpid) 744 { 745 unsigned long pgflags; 746 unsigned int shift; 747 pte_t *ptep; 748 749 /* 750 * Need to set an R or C bit in the 2nd-level tables; 751 * since we are just helping out the hardware here, 752 * it is sufficient to do what the hardware does. 753 */ 754 pgflags = _PAGE_ACCESSED; 755 if (writing) 756 pgflags |= _PAGE_DIRTY; 757 /* 758 * We are walking the secondary (partition-scoped) page table here. 759 * We can do this without disabling irq because the Linux MM 760 * subsystem doesn't do THP splits and collapses on this tree. 761 */ 762 ptep = __find_linux_pte(pgtable, gpa, NULL, &shift); 763 if (ptep && pte_present(*ptep) && (!writing || pte_write(*ptep))) { 764 kvmppc_radix_update_pte(kvm, ptep, 0, pgflags, gpa, shift); 765 return true; 766 } 767 return false; 768 } 769 770 int kvmppc_book3s_instantiate_page(struct kvm_vcpu *vcpu, 771 unsigned long gpa, 772 struct kvm_memory_slot *memslot, 773 bool writing, bool kvm_ro, 774 pte_t *inserted_pte, unsigned int *levelp) 775 { 776 struct kvm *kvm = vcpu->kvm; 777 struct page *page = NULL; 778 unsigned long mmu_seq; 779 unsigned long hva, gfn = gpa >> PAGE_SHIFT; 780 bool upgrade_write = false; 781 bool *upgrade_p = &upgrade_write; 782 pte_t pte, *ptep; 783 unsigned int shift, level; 784 int ret; 785 bool large_enable; 786 787 /* used to check for invalidations in progress */ 788 mmu_seq = kvm->mmu_notifier_seq; 789 smp_rmb(); 790 791 /* 792 * Do a fast check first, since __gfn_to_pfn_memslot doesn't 793 * do it with !atomic && !async, which is how we call it. 794 * We always ask for write permission since the common case 795 * is that the page is writable. 796 */ 797 hva = gfn_to_hva_memslot(memslot, gfn); 798 if (!kvm_ro && __get_user_pages_fast(hva, 1, 1, &page) == 1) { 799 upgrade_write = true; 800 } else { 801 unsigned long pfn; 802 803 /* Call KVM generic code to do the slow-path check */ 804 pfn = __gfn_to_pfn_memslot(memslot, gfn, false, NULL, 805 writing, upgrade_p); 806 if (is_error_noslot_pfn(pfn)) 807 return -EFAULT; 808 page = NULL; 809 if (pfn_valid(pfn)) { 810 page = pfn_to_page(pfn); 811 if (PageReserved(page)) 812 page = NULL; 813 } 814 } 815 816 /* 817 * Read the PTE from the process' radix tree and use that 818 * so we get the shift and attribute bits. 819 */ 820 local_irq_disable(); 821 ptep = __find_linux_pte(vcpu->arch.pgdir, hva, NULL, &shift); 822 pte = __pte(0); 823 if (ptep) 824 pte = *ptep; 825 local_irq_enable(); 826 /* 827 * If the PTE disappeared temporarily due to a THP 828 * collapse, just return and let the guest try again. 829 */ 830 if (!pte_present(pte)) { 831 if (page) 832 put_page(page); 833 return RESUME_GUEST; 834 } 835 836 /* If we're logging dirty pages, always map single pages */ 837 large_enable = !(memslot->flags & KVM_MEM_LOG_DIRTY_PAGES); 838 839 /* Get pte level from shift/size */ 840 if (large_enable && shift == PUD_SHIFT && 841 (gpa & (PUD_SIZE - PAGE_SIZE)) == 842 (hva & (PUD_SIZE - PAGE_SIZE))) { 843 level = 2; 844 } else if (large_enable && shift == PMD_SHIFT && 845 (gpa & (PMD_SIZE - PAGE_SIZE)) == 846 (hva & (PMD_SIZE - PAGE_SIZE))) { 847 level = 1; 848 } else { 849 level = 0; 850 if (shift > PAGE_SHIFT) { 851 /* 852 * If the pte maps more than one page, bring over 853 * bits from the virtual address to get the real 854 * address of the specific single page we want. 855 */ 856 unsigned long rpnmask = (1ul << shift) - PAGE_SIZE; 857 pte = __pte(pte_val(pte) | (hva & rpnmask)); 858 } 859 } 860 861 pte = __pte(pte_val(pte) | _PAGE_EXEC | _PAGE_ACCESSED); 862 if (writing || upgrade_write) { 863 if (pte_val(pte) & _PAGE_WRITE) 864 pte = __pte(pte_val(pte) | _PAGE_DIRTY); 865 } else { 866 pte = __pte(pte_val(pte) & ~(_PAGE_WRITE | _PAGE_DIRTY)); 867 } 868 869 /* Allocate space in the tree and write the PTE */ 870 ret = kvmppc_create_pte(kvm, kvm->arch.pgtable, pte, gpa, level, 871 mmu_seq, kvm->arch.lpid, NULL, NULL); 872 if (inserted_pte) 873 *inserted_pte = pte; 874 if (levelp) 875 *levelp = level; 876 877 if (page) { 878 if (!ret && (pte_val(pte) & _PAGE_WRITE)) 879 set_page_dirty_lock(page); 880 put_page(page); 881 } 882 883 /* Increment number of large pages if we (successfully) inserted one */ 884 if (!ret) { 885 if (level == 1) 886 kvm->stat.num_2M_pages++; 887 else if (level == 2) 888 kvm->stat.num_1G_pages++; 889 } 890 891 return ret; 892 } 893 894 int kvmppc_book3s_radix_page_fault(struct kvm_run *run, struct kvm_vcpu *vcpu, 895 unsigned long ea, unsigned long dsisr) 896 { 897 struct kvm *kvm = vcpu->kvm; 898 unsigned long gpa, gfn; 899 struct kvm_memory_slot *memslot; 900 long ret; 901 bool writing = !!(dsisr & DSISR_ISSTORE); 902 bool kvm_ro = false; 903 904 /* Check for unusual errors */ 905 if (dsisr & DSISR_UNSUPP_MMU) { 906 pr_err("KVM: Got unsupported MMU fault\n"); 907 return -EFAULT; 908 } 909 if (dsisr & DSISR_BADACCESS) { 910 /* Reflect to the guest as DSI */ 911 pr_err("KVM: Got radix HV page fault with DSISR=%lx\n", dsisr); 912 kvmppc_core_queue_data_storage(vcpu, ea, dsisr); 913 return RESUME_GUEST; 914 } 915 916 /* Translate the logical address */ 917 gpa = vcpu->arch.fault_gpa & ~0xfffUL; 918 gpa &= ~0xF000000000000000ul; 919 gfn = gpa >> PAGE_SHIFT; 920 if (!(dsisr & DSISR_PRTABLE_FAULT)) 921 gpa |= ea & 0xfff; 922 923 if (kvm->arch.secure_guest & KVMPPC_SECURE_INIT_DONE) 924 return kvmppc_send_page_to_uv(kvm, gfn); 925 926 /* Get the corresponding memslot */ 927 memslot = gfn_to_memslot(kvm, gfn); 928 929 /* No memslot means it's an emulated MMIO region */ 930 if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID)) { 931 if (dsisr & (DSISR_PRTABLE_FAULT | DSISR_BADACCESS | 932 DSISR_SET_RC)) { 933 /* 934 * Bad address in guest page table tree, or other 935 * unusual error - reflect it to the guest as DSI. 936 */ 937 kvmppc_core_queue_data_storage(vcpu, ea, dsisr); 938 return RESUME_GUEST; 939 } 940 return kvmppc_hv_emulate_mmio(run, vcpu, gpa, ea, writing); 941 } 942 943 if (memslot->flags & KVM_MEM_READONLY) { 944 if (writing) { 945 /* give the guest a DSI */ 946 kvmppc_core_queue_data_storage(vcpu, ea, DSISR_ISSTORE | 947 DSISR_PROTFAULT); 948 return RESUME_GUEST; 949 } 950 kvm_ro = true; 951 } 952 953 /* Failed to set the reference/change bits */ 954 if (dsisr & DSISR_SET_RC) { 955 spin_lock(&kvm->mmu_lock); 956 if (kvmppc_hv_handle_set_rc(kvm, kvm->arch.pgtable, 957 writing, gpa, kvm->arch.lpid)) 958 dsisr &= ~DSISR_SET_RC; 959 spin_unlock(&kvm->mmu_lock); 960 961 if (!(dsisr & (DSISR_BAD_FAULT_64S | DSISR_NOHPTE | 962 DSISR_PROTFAULT | DSISR_SET_RC))) 963 return RESUME_GUEST; 964 } 965 966 /* Try to insert a pte */ 967 ret = kvmppc_book3s_instantiate_page(vcpu, gpa, memslot, writing, 968 kvm_ro, NULL, NULL); 969 970 if (ret == 0 || ret == -EAGAIN) 971 ret = RESUME_GUEST; 972 return ret; 973 } 974 975 /* Called with kvm->mmu_lock held */ 976 int kvm_unmap_radix(struct kvm *kvm, struct kvm_memory_slot *memslot, 977 unsigned long gfn) 978 { 979 pte_t *ptep; 980 unsigned long gpa = gfn << PAGE_SHIFT; 981 unsigned int shift; 982 983 if (kvm->arch.secure_guest & KVMPPC_SECURE_INIT_DONE) { 984 uv_page_inval(kvm->arch.lpid, gpa, PAGE_SHIFT); 985 return 0; 986 } 987 988 ptep = __find_linux_pte(kvm->arch.pgtable, gpa, NULL, &shift); 989 if (ptep && pte_present(*ptep)) 990 kvmppc_unmap_pte(kvm, ptep, gpa, shift, memslot, 991 kvm->arch.lpid); 992 return 0; 993 } 994 995 /* Called with kvm->mmu_lock held */ 996 int kvm_age_radix(struct kvm *kvm, struct kvm_memory_slot *memslot, 997 unsigned long gfn) 998 { 999 pte_t *ptep; 1000 unsigned long gpa = gfn << PAGE_SHIFT; 1001 unsigned int shift; 1002 int ref = 0; 1003 unsigned long old, *rmapp; 1004 1005 if (kvm->arch.secure_guest & KVMPPC_SECURE_INIT_DONE) 1006 return ref; 1007 1008 ptep = __find_linux_pte(kvm->arch.pgtable, gpa, NULL, &shift); 1009 if (ptep && pte_present(*ptep) && pte_young(*ptep)) { 1010 old = kvmppc_radix_update_pte(kvm, ptep, _PAGE_ACCESSED, 0, 1011 gpa, shift); 1012 /* XXX need to flush tlb here? */ 1013 /* Also clear bit in ptes in shadow pgtable for nested guests */ 1014 rmapp = &memslot->arch.rmap[gfn - memslot->base_gfn]; 1015 kvmhv_update_nest_rmap_rc_list(kvm, rmapp, _PAGE_ACCESSED, 0, 1016 old & PTE_RPN_MASK, 1017 1UL << shift); 1018 ref = 1; 1019 } 1020 return ref; 1021 } 1022 1023 /* Called with kvm->mmu_lock held */ 1024 int kvm_test_age_radix(struct kvm *kvm, struct kvm_memory_slot *memslot, 1025 unsigned long gfn) 1026 { 1027 pte_t *ptep; 1028 unsigned long gpa = gfn << PAGE_SHIFT; 1029 unsigned int shift; 1030 int ref = 0; 1031 1032 if (kvm->arch.secure_guest & KVMPPC_SECURE_INIT_DONE) 1033 return ref; 1034 1035 ptep = __find_linux_pte(kvm->arch.pgtable, gpa, NULL, &shift); 1036 if (ptep && pte_present(*ptep) && pte_young(*ptep)) 1037 ref = 1; 1038 return ref; 1039 } 1040 1041 /* Returns the number of PAGE_SIZE pages that are dirty */ 1042 static int kvm_radix_test_clear_dirty(struct kvm *kvm, 1043 struct kvm_memory_slot *memslot, int pagenum) 1044 { 1045 unsigned long gfn = memslot->base_gfn + pagenum; 1046 unsigned long gpa = gfn << PAGE_SHIFT; 1047 pte_t *ptep; 1048 unsigned int shift; 1049 int ret = 0; 1050 unsigned long old, *rmapp; 1051 1052 if (kvm->arch.secure_guest & KVMPPC_SECURE_INIT_DONE) 1053 return ret; 1054 1055 ptep = __find_linux_pte(kvm->arch.pgtable, gpa, NULL, &shift); 1056 if (ptep && pte_present(*ptep) && pte_dirty(*ptep)) { 1057 ret = 1; 1058 if (shift) 1059 ret = 1 << (shift - PAGE_SHIFT); 1060 spin_lock(&kvm->mmu_lock); 1061 old = kvmppc_radix_update_pte(kvm, ptep, _PAGE_DIRTY, 0, 1062 gpa, shift); 1063 kvmppc_radix_tlbie_page(kvm, gpa, shift, kvm->arch.lpid); 1064 /* Also clear bit in ptes in shadow pgtable for nested guests */ 1065 rmapp = &memslot->arch.rmap[gfn - memslot->base_gfn]; 1066 kvmhv_update_nest_rmap_rc_list(kvm, rmapp, _PAGE_DIRTY, 0, 1067 old & PTE_RPN_MASK, 1068 1UL << shift); 1069 spin_unlock(&kvm->mmu_lock); 1070 } 1071 return ret; 1072 } 1073 1074 long kvmppc_hv_get_dirty_log_radix(struct kvm *kvm, 1075 struct kvm_memory_slot *memslot, unsigned long *map) 1076 { 1077 unsigned long i, j; 1078 int npages; 1079 1080 for (i = 0; i < memslot->npages; i = j) { 1081 npages = kvm_radix_test_clear_dirty(kvm, memslot, i); 1082 1083 /* 1084 * Note that if npages > 0 then i must be a multiple of npages, 1085 * since huge pages are only used to back the guest at guest 1086 * real addresses that are a multiple of their size. 1087 * Since we have at most one PTE covering any given guest 1088 * real address, if npages > 1 we can skip to i + npages. 1089 */ 1090 j = i + 1; 1091 if (npages) { 1092 set_dirty_bits(map, i, npages); 1093 j = i + npages; 1094 } 1095 } 1096 return 0; 1097 } 1098 1099 void kvmppc_radix_flush_memslot(struct kvm *kvm, 1100 const struct kvm_memory_slot *memslot) 1101 { 1102 unsigned long n; 1103 pte_t *ptep; 1104 unsigned long gpa; 1105 unsigned int shift; 1106 1107 if (kvm->arch.secure_guest & KVMPPC_SECURE_INIT_START) 1108 kvmppc_uvmem_drop_pages(memslot, kvm, true); 1109 1110 if (kvm->arch.secure_guest & KVMPPC_SECURE_INIT_DONE) 1111 return; 1112 1113 gpa = memslot->base_gfn << PAGE_SHIFT; 1114 spin_lock(&kvm->mmu_lock); 1115 for (n = memslot->npages; n; --n) { 1116 ptep = __find_linux_pte(kvm->arch.pgtable, gpa, NULL, &shift); 1117 if (ptep && pte_present(*ptep)) 1118 kvmppc_unmap_pte(kvm, ptep, gpa, shift, memslot, 1119 kvm->arch.lpid); 1120 gpa += PAGE_SIZE; 1121 } 1122 spin_unlock(&kvm->mmu_lock); 1123 } 1124 1125 static void add_rmmu_ap_encoding(struct kvm_ppc_rmmu_info *info, 1126 int psize, int *indexp) 1127 { 1128 if (!mmu_psize_defs[psize].shift) 1129 return; 1130 info->ap_encodings[*indexp] = mmu_psize_defs[psize].shift | 1131 (mmu_psize_defs[psize].ap << 29); 1132 ++(*indexp); 1133 } 1134 1135 int kvmhv_get_rmmu_info(struct kvm *kvm, struct kvm_ppc_rmmu_info *info) 1136 { 1137 int i; 1138 1139 if (!radix_enabled()) 1140 return -EINVAL; 1141 memset(info, 0, sizeof(*info)); 1142 1143 /* 4k page size */ 1144 info->geometries[0].page_shift = 12; 1145 info->geometries[0].level_bits[0] = 9; 1146 for (i = 1; i < 4; ++i) 1147 info->geometries[0].level_bits[i] = p9_supported_radix_bits[i]; 1148 /* 64k page size */ 1149 info->geometries[1].page_shift = 16; 1150 for (i = 0; i < 4; ++i) 1151 info->geometries[1].level_bits[i] = p9_supported_radix_bits[i]; 1152 1153 i = 0; 1154 add_rmmu_ap_encoding(info, MMU_PAGE_4K, &i); 1155 add_rmmu_ap_encoding(info, MMU_PAGE_64K, &i); 1156 add_rmmu_ap_encoding(info, MMU_PAGE_2M, &i); 1157 add_rmmu_ap_encoding(info, MMU_PAGE_1G, &i); 1158 1159 return 0; 1160 } 1161 1162 int kvmppc_init_vm_radix(struct kvm *kvm) 1163 { 1164 kvm->arch.pgtable = pgd_alloc(kvm->mm); 1165 if (!kvm->arch.pgtable) 1166 return -ENOMEM; 1167 return 0; 1168 } 1169 1170 static void pte_ctor(void *addr) 1171 { 1172 memset(addr, 0, RADIX_PTE_TABLE_SIZE); 1173 } 1174 1175 static void pmd_ctor(void *addr) 1176 { 1177 memset(addr, 0, RADIX_PMD_TABLE_SIZE); 1178 } 1179 1180 struct debugfs_radix_state { 1181 struct kvm *kvm; 1182 struct mutex mutex; 1183 unsigned long gpa; 1184 int lpid; 1185 int chars_left; 1186 int buf_index; 1187 char buf[128]; 1188 u8 hdr; 1189 }; 1190 1191 static int debugfs_radix_open(struct inode *inode, struct file *file) 1192 { 1193 struct kvm *kvm = inode->i_private; 1194 struct debugfs_radix_state *p; 1195 1196 p = kzalloc(sizeof(*p), GFP_KERNEL); 1197 if (!p) 1198 return -ENOMEM; 1199 1200 kvm_get_kvm(kvm); 1201 p->kvm = kvm; 1202 mutex_init(&p->mutex); 1203 file->private_data = p; 1204 1205 return nonseekable_open(inode, file); 1206 } 1207 1208 static int debugfs_radix_release(struct inode *inode, struct file *file) 1209 { 1210 struct debugfs_radix_state *p = file->private_data; 1211 1212 kvm_put_kvm(p->kvm); 1213 kfree(p); 1214 return 0; 1215 } 1216 1217 static ssize_t debugfs_radix_read(struct file *file, char __user *buf, 1218 size_t len, loff_t *ppos) 1219 { 1220 struct debugfs_radix_state *p = file->private_data; 1221 ssize_t ret, r; 1222 unsigned long n; 1223 struct kvm *kvm; 1224 unsigned long gpa; 1225 pgd_t *pgt; 1226 struct kvm_nested_guest *nested; 1227 pgd_t *pgdp; 1228 p4d_t p4d, *p4dp; 1229 pud_t pud, *pudp; 1230 pmd_t pmd, *pmdp; 1231 pte_t *ptep; 1232 int shift; 1233 unsigned long pte; 1234 1235 kvm = p->kvm; 1236 if (!kvm_is_radix(kvm)) 1237 return 0; 1238 1239 ret = mutex_lock_interruptible(&p->mutex); 1240 if (ret) 1241 return ret; 1242 1243 if (p->chars_left) { 1244 n = p->chars_left; 1245 if (n > len) 1246 n = len; 1247 r = copy_to_user(buf, p->buf + p->buf_index, n); 1248 n -= r; 1249 p->chars_left -= n; 1250 p->buf_index += n; 1251 buf += n; 1252 len -= n; 1253 ret = n; 1254 if (r) { 1255 if (!n) 1256 ret = -EFAULT; 1257 goto out; 1258 } 1259 } 1260 1261 gpa = p->gpa; 1262 nested = NULL; 1263 pgt = NULL; 1264 while (len != 0 && p->lpid >= 0) { 1265 if (gpa >= RADIX_PGTABLE_RANGE) { 1266 gpa = 0; 1267 pgt = NULL; 1268 if (nested) { 1269 kvmhv_put_nested(nested); 1270 nested = NULL; 1271 } 1272 p->lpid = kvmhv_nested_next_lpid(kvm, p->lpid); 1273 p->hdr = 0; 1274 if (p->lpid < 0) 1275 break; 1276 } 1277 if (!pgt) { 1278 if (p->lpid == 0) { 1279 pgt = kvm->arch.pgtable; 1280 } else { 1281 nested = kvmhv_get_nested(kvm, p->lpid, false); 1282 if (!nested) { 1283 gpa = RADIX_PGTABLE_RANGE; 1284 continue; 1285 } 1286 pgt = nested->shadow_pgtable; 1287 } 1288 } 1289 n = 0; 1290 if (!p->hdr) { 1291 if (p->lpid > 0) 1292 n = scnprintf(p->buf, sizeof(p->buf), 1293 "\nNested LPID %d: ", p->lpid); 1294 n += scnprintf(p->buf + n, sizeof(p->buf) - n, 1295 "pgdir: %lx\n", (unsigned long)pgt); 1296 p->hdr = 1; 1297 goto copy; 1298 } 1299 1300 pgdp = pgt + pgd_index(gpa); 1301 p4dp = p4d_offset(pgdp, gpa); 1302 p4d = READ_ONCE(*p4dp); 1303 if (!(p4d_val(p4d) & _PAGE_PRESENT)) { 1304 gpa = (gpa & P4D_MASK) + P4D_SIZE; 1305 continue; 1306 } 1307 1308 pudp = pud_offset(&p4d, gpa); 1309 pud = READ_ONCE(*pudp); 1310 if (!(pud_val(pud) & _PAGE_PRESENT)) { 1311 gpa = (gpa & PUD_MASK) + PUD_SIZE; 1312 continue; 1313 } 1314 if (pud_val(pud) & _PAGE_PTE) { 1315 pte = pud_val(pud); 1316 shift = PUD_SHIFT; 1317 goto leaf; 1318 } 1319 1320 pmdp = pmd_offset(&pud, gpa); 1321 pmd = READ_ONCE(*pmdp); 1322 if (!(pmd_val(pmd) & _PAGE_PRESENT)) { 1323 gpa = (gpa & PMD_MASK) + PMD_SIZE; 1324 continue; 1325 } 1326 if (pmd_val(pmd) & _PAGE_PTE) { 1327 pte = pmd_val(pmd); 1328 shift = PMD_SHIFT; 1329 goto leaf; 1330 } 1331 1332 ptep = pte_offset_kernel(&pmd, gpa); 1333 pte = pte_val(READ_ONCE(*ptep)); 1334 if (!(pte & _PAGE_PRESENT)) { 1335 gpa += PAGE_SIZE; 1336 continue; 1337 } 1338 shift = PAGE_SHIFT; 1339 leaf: 1340 n = scnprintf(p->buf, sizeof(p->buf), 1341 " %lx: %lx %d\n", gpa, pte, shift); 1342 gpa += 1ul << shift; 1343 copy: 1344 p->chars_left = n; 1345 if (n > len) 1346 n = len; 1347 r = copy_to_user(buf, p->buf, n); 1348 n -= r; 1349 p->chars_left -= n; 1350 p->buf_index = n; 1351 buf += n; 1352 len -= n; 1353 ret += n; 1354 if (r) { 1355 if (!ret) 1356 ret = -EFAULT; 1357 break; 1358 } 1359 } 1360 p->gpa = gpa; 1361 if (nested) 1362 kvmhv_put_nested(nested); 1363 1364 out: 1365 mutex_unlock(&p->mutex); 1366 return ret; 1367 } 1368 1369 static ssize_t debugfs_radix_write(struct file *file, const char __user *buf, 1370 size_t len, loff_t *ppos) 1371 { 1372 return -EACCES; 1373 } 1374 1375 static const struct file_operations debugfs_radix_fops = { 1376 .owner = THIS_MODULE, 1377 .open = debugfs_radix_open, 1378 .release = debugfs_radix_release, 1379 .read = debugfs_radix_read, 1380 .write = debugfs_radix_write, 1381 .llseek = generic_file_llseek, 1382 }; 1383 1384 void kvmhv_radix_debugfs_init(struct kvm *kvm) 1385 { 1386 debugfs_create_file("radix", 0400, kvm->arch.debugfs_dir, kvm, 1387 &debugfs_radix_fops); 1388 } 1389 1390 int kvmppc_radix_init(void) 1391 { 1392 unsigned long size = sizeof(void *) << RADIX_PTE_INDEX_SIZE; 1393 1394 kvm_pte_cache = kmem_cache_create("kvm-pte", size, size, 0, pte_ctor); 1395 if (!kvm_pte_cache) 1396 return -ENOMEM; 1397 1398 size = sizeof(void *) << RADIX_PMD_INDEX_SIZE; 1399 1400 kvm_pmd_cache = kmem_cache_create("kvm-pmd", size, size, 0, pmd_ctor); 1401 if (!kvm_pmd_cache) { 1402 kmem_cache_destroy(kvm_pte_cache); 1403 return -ENOMEM; 1404 } 1405 1406 return 0; 1407 } 1408 1409 void kvmppc_radix_exit(void) 1410 { 1411 kmem_cache_destroy(kvm_pte_cache); 1412 kmem_cache_destroy(kvm_pmd_cache); 1413 } 1414