book3s_64_mmu_radix.c

   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#include <linux/pgtable.h>
  15
  16#include <asm/kvm_ppc.h>
  17#include <asm/kvm_book3s.h>
  18#include "book3s_hv.h"
  19#include <asm/page.h>
  20#include <asm/mmu.h>
  21#include <asm/pgalloc.h>
  22#include <asm/pte-walk.h>
  23#include <asm/ultravisor.h>
  24#include <asm/kvm_book3s_uvmem.h>
  25#include <asm/plpar_wrappers.h>
  26#include <asm/firmware.h>
  27
  28/*
  29 * Supported radix tree geometry.
  30 * Like p9, we support either 5 or 9 bits at the first (lowest) level,
  31 * for a page size of 64k or 4k.
  32 */
  33static int p9_supported_radix_bits[4] = { 5, 9, 9, 13 };
  34
  35unsigned long __kvmhv_copy_tofrom_guest_radix(int lpid, int pid,
  36					      gva_t eaddr, void *to, void *from,
  37					      unsigned long n)
  38{
  39	int old_pid, old_lpid;
  40	unsigned long quadrant, ret = n;
  41	bool is_load = !!to;
  42
  43	if (kvmhv_is_nestedv2())
  44		return H_UNSUPPORTED;
  45
  46	/* Can't access quadrants 1 or 2 in non-HV mode, call the HV to do it */
  47	if (kvmhv_on_pseries())
  48		return plpar_hcall_norets(H_COPY_TOFROM_GUEST, lpid, pid, eaddr,
  49					  (to != NULL) ? __pa(to): 0,
  50					  (from != NULL) ? __pa(from): 0, n);
  51
  52	if (eaddr & (0xFFFUL << 52))
  53		return ret;
  54
  55	quadrant = 1;
  56	if (!pid)
  57		quadrant = 2;
  58	if (is_load)
  59		from = (void *) (eaddr | (quadrant << 62));
  60	else
  61		to = (void *) (eaddr | (quadrant << 62));
  62
  63	preempt_disable();
  64
  65	asm volatile("hwsync" ::: "memory");
  66	isync();
  67	/* switch the lpid first to avoid running host with unallocated pid */
  68	old_lpid = mfspr(SPRN_LPID);
  69	if (old_lpid != lpid)
  70		mtspr(SPRN_LPID, lpid);
  71	if (quadrant == 1) {
  72		old_pid = mfspr(SPRN_PID);
  73		if (old_pid != pid)
  74			mtspr(SPRN_PID, pid);
  75	}
  76	isync();
  77
  78	pagefault_disable();
  79	if (is_load)
  80		ret = __copy_from_user_inatomic(to, (const void __user *)from, n);
  81	else
  82		ret = __copy_to_user_inatomic((void __user *)to, from, n);
  83	pagefault_enable();
  84
  85	asm volatile("hwsync" ::: "memory");
  86	isync();
  87	/* switch the pid first to avoid running host with unallocated pid */
  88	if (quadrant == 1 && pid != old_pid)
  89		mtspr(SPRN_PID, old_pid);
  90	if (lpid != old_lpid)
  91		mtspr(SPRN_LPID, old_lpid);
  92	isync();
  93
  94	preempt_enable();
  95
  96	return ret;
  97}
  98
  99static long kvmhv_copy_tofrom_guest_radix(struct kvm_vcpu *vcpu, gva_t eaddr,
 100					  void *to, void *from, unsigned long n)
 101{
 102	int lpid = vcpu->kvm->arch.lpid;
 103	int pid;
 104
 105	/* This would cause a data segment intr so don't allow the access */
 106	if (eaddr & (0x3FFUL << 52))
 107		return -EINVAL;
 108
 109	/* Should we be using the nested lpid */
 110	if (vcpu->arch.nested)
 111		lpid = vcpu->arch.nested->shadow_lpid;
 112
 113	/* If accessing quadrant 3 then pid is expected to be 0 */
 114	if (((eaddr >> 62) & 0x3) == 0x3)
 115		pid = 0;
 116	else
 117		pid = kvmppc_get_pid(vcpu);
 118
 119	eaddr &= ~(0xFFFUL << 52);
 120
 121	return __kvmhv_copy_tofrom_guest_radix(lpid, pid, eaddr, to, from, n);
 122}
 123
 124long kvmhv_copy_from_guest_radix(struct kvm_vcpu *vcpu, gva_t eaddr, void *to,
 125				 unsigned long n)
 126{
 127	long ret;
 128
 129	ret = kvmhv_copy_tofrom_guest_radix(vcpu, eaddr, to, NULL, n);
 130	if (ret > 0)
 131		memset(to + (n - ret), 0, ret);
 132
 133	return ret;
 134}
 135
 136long kvmhv_copy_to_guest_radix(struct kvm_vcpu *vcpu, gva_t eaddr, void *from,
 137			       unsigned long n)
 138{
 139	return kvmhv_copy_tofrom_guest_radix(vcpu, eaddr, NULL, from, n);
 140}
 141
 142int kvmppc_mmu_walk_radix_tree(struct kvm_vcpu *vcpu, gva_t eaddr,
 143			       struct kvmppc_pte *gpte, u64 root,
 144			       u64 *pte_ret_p)
 145{
 146	struct kvm *kvm = vcpu->kvm;
 147	int ret, level, ps;
 148	unsigned long rts, bits, offset, index;
 149	u64 pte, base, gpa;
 150	__be64 rpte;
 151
 152	rts = ((root & RTS1_MASK) >> (RTS1_SHIFT - 3)) |
 153		((root & RTS2_MASK) >> RTS2_SHIFT);
 154	bits = root & RPDS_MASK;
 155	base = root & RPDB_MASK;
 156
 157	offset = rts + 31;
 158
 159	/* Current implementations only support 52-bit space */
 160	if (offset != 52)
 161		return -EINVAL;
 162
 163	/* Walk each level of the radix tree */
 164	for (level = 3; level >= 0; --level) {
 165		u64 addr;
 166		/* Check a valid size */
 167		if (level && bits != p9_supported_radix_bits[level])
 168			return -EINVAL;
 169		if (level == 0 && !(bits == 5 || bits == 9))
 170			return -EINVAL;
 171		offset -= bits;
 172		index = (eaddr >> offset) & ((1UL << bits) - 1);
 173		/* Check that low bits of page table base are zero */
 174		if (base & ((1UL << (bits + 3)) - 1))
 175			return -EINVAL;
 176		/* Read the entry from guest memory */
 177		addr = base + (index * sizeof(rpte));
 178
 179		kvm_vcpu_srcu_read_lock(vcpu);
 180		ret = kvm_read_guest(kvm, addr, &rpte, sizeof(rpte));
 181		kvm_vcpu_srcu_read_unlock(vcpu);
 182		if (ret) {
 183			if (pte_ret_p)
 184				*pte_ret_p = addr;
 185			return ret;
 186		}
 187		pte = __be64_to_cpu(rpte);
 188		if (!(pte & _PAGE_PRESENT))
 189			return -ENOENT;
 190		/* Check if a leaf entry */
 191		if (pte & _PAGE_PTE)
 192			break;
 193		/* Get ready to walk the next level */
 194		base = pte & RPDB_MASK;
 195		bits = pte & RPDS_MASK;
 196	}
 197
 198	/* Need a leaf at lowest level; 512GB pages not supported */
 199	if (level < 0 || level == 3)
 200		return -EINVAL;
 201
 202	/* We found a valid leaf PTE */
 203	/* Offset is now log base 2 of the page size */
 204	gpa = pte & 0x01fffffffffff000ul;
 205	if (gpa & ((1ul << offset) - 1))
 206		return -EINVAL;
 207	gpa |= eaddr & ((1ul << offset) - 1);
 208	for (ps = MMU_PAGE_4K; ps < MMU_PAGE_COUNT; ++ps)
 209		if (offset == mmu_psize_defs[ps].shift)
 210			break;
 211	gpte->page_size = ps;
 212	gpte->page_shift = offset;
 213
 214	gpte->eaddr = eaddr;
 215	gpte->raddr = gpa;
 216
 217	/* Work out permissions */
 218	gpte->may_read = !!(pte & _PAGE_READ);
 219	gpte->may_write = !!(pte & _PAGE_WRITE);
 220	gpte->may_execute = !!(pte & _PAGE_EXEC);
 221
 222	gpte->rc = pte & (_PAGE_ACCESSED | _PAGE_DIRTY);
 223
 224	if (pte_ret_p)
 225		*pte_ret_p = pte;
 226
 227	return 0;
 228}
 229
 230/*
 231 * Used to walk a partition or process table radix tree in guest memory
 232 * Note: We exploit the fact that a partition table and a process
 233 * table have the same layout, a partition-scoped page table and a
 234 * process-scoped page table have the same layout, and the 2nd
 235 * doubleword of a partition table entry has the same layout as
 236 * the PTCR register.
 237 */
 238int kvmppc_mmu_radix_translate_table(struct kvm_vcpu *vcpu, gva_t eaddr,
 239				     struct kvmppc_pte *gpte, u64 table,
 240				     int table_index, u64 *pte_ret_p)
 241{
 242	struct kvm *kvm = vcpu->kvm;
 243	int ret;
 244	unsigned long size, ptbl, root;
 245	struct prtb_entry entry;
 246
 247	if ((table & PRTS_MASK) > 24)
 248		return -EINVAL;
 249	size = 1ul << ((table & PRTS_MASK) + 12);
 250
 251	/* Is the table big enough to contain this entry? */
 252	if ((table_index * sizeof(entry)) >= size)
 253		return -EINVAL;
 254
 255	/* Read the table to find the root of the radix tree */
 256	ptbl = (table & PRTB_MASK) + (table_index * sizeof(entry));
 257	kvm_vcpu_srcu_read_lock(vcpu);
 258	ret = kvm_read_guest(kvm, ptbl, &entry, sizeof(entry));
 259	kvm_vcpu_srcu_read_unlock(vcpu);
 260	if (ret)
 261		return ret;
 262
 263	/* Root is stored in the first double word */
 264	root = be64_to_cpu(entry.prtb0);
 265
 266	return kvmppc_mmu_walk_radix_tree(vcpu, eaddr, gpte, root, pte_ret_p);
 267}
 268
 269int kvmppc_mmu_radix_xlate(struct kvm_vcpu *vcpu, gva_t eaddr,
 270			   struct kvmppc_pte *gpte, bool data, bool iswrite)
 271{
 272	u32 pid;
 273	u64 pte;
 274	int ret;
 275
 276	/* Work out effective PID */
 277	switch (eaddr >> 62) {
 278	case 0:
 279		pid = kvmppc_get_pid(vcpu);
 280		break;
 281	case 3:
 282		pid = 0;
 283		break;
 284	default:
 285		return -EINVAL;
 286	}
 287
 288	ret = kvmppc_mmu_radix_translate_table(vcpu, eaddr, gpte,
 289				vcpu->kvm->arch.process_table, pid, &pte);
 290	if (ret)
 291		return ret;
 292
 293	/* Check privilege (applies only to process scoped translations) */
 294	if (kvmppc_get_msr(vcpu) & MSR_PR) {
 295		if (pte & _PAGE_PRIVILEGED) {
 296			gpte->may_read = 0;
 297			gpte->may_write = 0;
 298			gpte->may_execute = 0;
 299		}
 300	} else {
 301		if (!(pte & _PAGE_PRIVILEGED)) {
 302			/* Check AMR/IAMR to see if strict mode is in force */
 303			if (kvmppc_get_amr_hv(vcpu) & (1ul << 62))
 304				gpte->may_read = 0;
 305			if (kvmppc_get_amr_hv(vcpu) & (1ul << 63))
 306				gpte->may_write = 0;
 307			if (vcpu->arch.iamr & (1ul << 62))
 308				gpte->may_execute = 0;
 309		}
 310	}
 311
 312	return 0;
 313}
 314
 315void kvmppc_radix_tlbie_page(struct kvm *kvm, unsigned long addr,
 316			     unsigned int pshift, u64 lpid)
 317{
 318	unsigned long psize = PAGE_SIZE;
 319	int psi;
 320	long rc;
 321	unsigned long rb;
 322
 323	if (pshift)
 324		psize = 1UL << pshift;
 325	else
 326		pshift = PAGE_SHIFT;
 327
 328	addr &= ~(psize - 1);
 329
 330	if (!kvmhv_on_pseries()) {
 331		radix__flush_tlb_lpid_page(lpid, addr, psize);
 332		return;
 333	}
 334
 335	psi = shift_to_mmu_psize(pshift);
 336
 337	if (!firmware_has_feature(FW_FEATURE_RPT_INVALIDATE)) {
 338		rb = addr | (mmu_get_ap(psi) << PPC_BITLSHIFT(58));
 339		rc = plpar_hcall_norets(H_TLB_INVALIDATE, H_TLBIE_P1_ENC(0, 0, 1),
 340					lpid, rb);
 341	} else {
 342		rc = pseries_rpt_invalidate(lpid, H_RPTI_TARGET_CMMU,
 343					    H_RPTI_TYPE_NESTED |
 344					    H_RPTI_TYPE_TLB,
 345					    psize_to_rpti_pgsize(psi),
 346					    addr, addr + psize);
 347	}
 348
 349	if (rc)
 350		pr_err("KVM: TLB page invalidation hcall failed, rc=%ld\n", rc);
 351}
 352
 353static void kvmppc_radix_flush_pwc(struct kvm *kvm, u64 lpid)
 354{
 355	long rc;
 356
 357	if (!kvmhv_on_pseries()) {
 358		radix__flush_pwc_lpid(lpid);
 359		return;
 360	}
 361
 362	if (!firmware_has_feature(FW_FEATURE_RPT_INVALIDATE))
 363		rc = plpar_hcall_norets(H_TLB_INVALIDATE, H_TLBIE_P1_ENC(1, 0, 1),
 364					lpid, TLBIEL_INVAL_SET_LPID);
 365	else
 366		rc = pseries_rpt_invalidate(lpid, H_RPTI_TARGET_CMMU,
 367					    H_RPTI_TYPE_NESTED |
 368					    H_RPTI_TYPE_PWC, H_RPTI_PAGE_ALL,
 369					    0, -1UL);
 370	if (rc)
 371		pr_err("KVM: TLB PWC invalidation hcall failed, rc=%ld\n", rc);
 372}
 373
 374static unsigned long kvmppc_radix_update_pte(struct kvm *kvm, pte_t *ptep,
 375				      unsigned long clr, unsigned long set,
 376				      unsigned long addr, unsigned int shift)
 377{
 378	return __radix_pte_update(ptep, clr, set);
 379}
 380
 381static void kvmppc_radix_set_pte_at(struct kvm *kvm, unsigned long addr,
 382			     pte_t *ptep, pte_t pte)
 383{
 384	radix__set_pte_at(kvm->mm, addr, ptep, pte, 0);
 385}
 386
 387static struct kmem_cache *kvm_pte_cache;
 388static struct kmem_cache *kvm_pmd_cache;
 389
 390static pte_t *kvmppc_pte_alloc(void)
 391{
 392	pte_t *pte;
 393
 394	pte = kmem_cache_alloc(kvm_pte_cache, GFP_KERNEL);
 395	/* pmd_populate() will only reference _pa(pte). */
 396	kmemleak_ignore(pte);
 397
 398	return pte;
 399}
 400
 401static void kvmppc_pte_free(pte_t *ptep)
 402{
 403	kmem_cache_free(kvm_pte_cache, ptep);
 404}
 405
 406static pmd_t *kvmppc_pmd_alloc(void)
 407{
 408	pmd_t *pmd;
 409
 410	pmd = kmem_cache_alloc(kvm_pmd_cache, GFP_KERNEL);
 411	/* pud_populate() will only reference _pa(pmd). */
 412	kmemleak_ignore(pmd);
 413
 414	return pmd;
 415}
 416
 417static void kvmppc_pmd_free(pmd_t *pmdp)
 418{
 419	kmem_cache_free(kvm_pmd_cache, pmdp);
 420}
 421
 422/* Called with kvm->mmu_lock held */
 423void kvmppc_unmap_pte(struct kvm *kvm, pte_t *pte, unsigned long gpa,
 424		      unsigned int shift,
 425		      const struct kvm_memory_slot *memslot,
 426		      u64 lpid)
 427
 428{
 429	unsigned long old;
 430	unsigned long gfn = gpa >> PAGE_SHIFT;
 431	unsigned long page_size = PAGE_SIZE;
 432	unsigned long hpa;
 433
 434	old = kvmppc_radix_update_pte(kvm, pte, ~0UL, 0, gpa, shift);
 435	kvmppc_radix_tlbie_page(kvm, gpa, shift, lpid);
 436
 437	/* The following only applies to L1 entries */
 438	if (lpid != kvm->arch.lpid)
 439		return;
 440
 441	if (!memslot) {
 442		memslot = gfn_to_memslot(kvm, gfn);
 443		if (!memslot)
 444			return;
 445	}
 446	if (shift) { /* 1GB or 2MB page */
 447		page_size = 1ul << shift;
 448		if (shift == PMD_SHIFT)
 449			kvm->stat.num_2M_pages--;
 450		else if (shift == PUD_SHIFT)
 451			kvm->stat.num_1G_pages--;
 452	}
 453
 454	gpa &= ~(page_size - 1);
 455	hpa = old & PTE_RPN_MASK;
 456	kvmhv_remove_nest_rmap_range(kvm, memslot, gpa, hpa, page_size);
 457
 458	if ((old & _PAGE_DIRTY) && memslot->dirty_bitmap)
 459		kvmppc_update_dirty_map(memslot, gfn, page_size);
 460}
 461
 462/*
 463 * kvmppc_free_p?d are used to free existing page tables, and recursively
 464 * descend and clear and free children.
 465 * Callers are responsible for flushing the PWC.
 466 *
 467 * When page tables are being unmapped/freed as part of page fault path
 468 * (full == false), valid ptes are generally not expected; however, there
 469 * is one situation where they arise, which is when dirty page logging is
 470 * turned off for a memslot while the VM is running.  The new memslot
 471 * becomes visible to page faults before the memslot commit function
 472 * gets to flush the memslot, which can lead to a 2MB page mapping being
 473 * installed for a guest physical address where there are already 64kB
 474 * (or 4kB) mappings (of sub-pages of the same 2MB page).
 475 */
 476static void kvmppc_unmap_free_pte(struct kvm *kvm, pte_t *pte, bool full,
 477				  u64 lpid)
 478{
 479	if (full) {
 480		memset(pte, 0, sizeof(long) << RADIX_PTE_INDEX_SIZE);
 481	} else {
 482		pte_t *p = pte;
 483		unsigned long it;
 484
 485		for (it = 0; it < PTRS_PER_PTE; ++it, ++p) {
 486			if (pte_val(*p) == 0)
 487				continue;
 488			kvmppc_unmap_pte(kvm, p,
 489					 pte_pfn(*p) << PAGE_SHIFT,
 490					 PAGE_SHIFT, NULL, lpid);
 491		}
 492	}
 493
 494	kvmppc_pte_free(pte);
 495}
 496
 497static void kvmppc_unmap_free_pmd(struct kvm *kvm, pmd_t *pmd, bool full,
 498				  u64 lpid)
 499{
 500	unsigned long im;
 501	pmd_t *p = pmd;
 502
 503	for (im = 0; im < PTRS_PER_PMD; ++im, ++p) {
 504		if (!pmd_present(*p))
 505			continue;
 506		if (pmd_leaf(*p)) {
 507			if (full) {
 508				pmd_clear(p);
 509			} else {
 510				WARN_ON_ONCE(1);
 511				kvmppc_unmap_pte(kvm, (pte_t *)p,
 512					 pte_pfn(*(pte_t *)p) << PAGE_SHIFT,
 513					 PMD_SHIFT, NULL, lpid);
 514			}
 515		} else {
 516			pte_t *pte;
 517
 518			pte = pte_offset_kernel(p, 0);
 519			kvmppc_unmap_free_pte(kvm, pte, full, lpid);
 520			pmd_clear(p);
 521		}
 522	}
 523	kvmppc_pmd_free(pmd);
 524}
 525
 526static void kvmppc_unmap_free_pud(struct kvm *kvm, pud_t *pud,
 527				  u64 lpid)
 528{
 529	unsigned long iu;
 530	pud_t *p = pud;
 531
 532	for (iu = 0; iu < PTRS_PER_PUD; ++iu, ++p) {
 533		if (!pud_present(*p))
 534			continue;
 535		if (pud_leaf(*p)) {
 536			pud_clear(p);
 537		} else {
 538			pmd_t *pmd;
 539
 540			pmd = pmd_offset(p, 0);
 541			kvmppc_unmap_free_pmd(kvm, pmd, true, lpid);
 542			pud_clear(p);
 543		}
 544	}
 545	pud_free(kvm->mm, pud);
 546}
 547
 548void kvmppc_free_pgtable_radix(struct kvm *kvm, pgd_t *pgd, u64 lpid)
 549{
 550	unsigned long ig;
 551
 552	for (ig = 0; ig < PTRS_PER_PGD; ++ig, ++pgd) {
 553		p4d_t *p4d = p4d_offset(pgd, 0);
 554		pud_t *pud;
 555
 556		if (!p4d_present(*p4d))
 557			continue;
 558		pud = pud_offset(p4d, 0);
 559		kvmppc_unmap_free_pud(kvm, pud, lpid);
 560		p4d_clear(p4d);
 561	}
 562}
 563
 564void kvmppc_free_radix(struct kvm *kvm)
 565{
 566	if (kvm->arch.pgtable) {
 567		kvmppc_free_pgtable_radix(kvm, kvm->arch.pgtable,
 568					  kvm->arch.lpid);
 569		pgd_free(kvm->mm, kvm->arch.pgtable);
 570		kvm->arch.pgtable = NULL;
 571	}
 572}
 573
 574static void kvmppc_unmap_free_pmd_entry_table(struct kvm *kvm, pmd_t *pmd,
 575					unsigned long gpa, u64 lpid)
 576{
 577	pte_t *pte = pte_offset_kernel(pmd, 0);
 578
 579	/*
 580	 * Clearing the pmd entry then flushing the PWC ensures that the pte
 581	 * page no longer be cached by the MMU, so can be freed without
 582	 * flushing the PWC again.
 583	 */
 584	pmd_clear(pmd);
 585	kvmppc_radix_flush_pwc(kvm, lpid);
 586
 587	kvmppc_unmap_free_pte(kvm, pte, false, lpid);
 588}
 589
 590static void kvmppc_unmap_free_pud_entry_table(struct kvm *kvm, pud_t *pud,
 591					unsigned long gpa, u64 lpid)
 592{
 593	pmd_t *pmd = pmd_offset(pud, 0);
 594
 595	/*
 596	 * Clearing the pud entry then flushing the PWC ensures that the pmd
 597	 * page and any children pte pages will no longer be cached by the MMU,
 598	 * so can be freed without flushing the PWC again.
 599	 */
 600	pud_clear(pud);
 601	kvmppc_radix_flush_pwc(kvm, lpid);
 602
 603	kvmppc_unmap_free_pmd(kvm, pmd, false, lpid);
 604}
 605
 606/*
 607 * There are a number of bits which may differ between different faults to
 608 * the same partition scope entry. RC bits, in the course of cleaning and
 609 * aging. And the write bit can change, either the access could have been
 610 * upgraded, or a read fault could happen concurrently with a write fault
 611 * that sets those bits first.
 612 */
 613#define PTE_BITS_MUST_MATCH (~(_PAGE_WRITE | _PAGE_DIRTY | _PAGE_ACCESSED))
 614
 615int kvmppc_create_pte(struct kvm *kvm, pgd_t *pgtable, pte_t pte,
 616		      unsigned long gpa, unsigned int level,
 617		      unsigned long mmu_seq, u64 lpid,
 618		      unsigned long *rmapp, struct rmap_nested **n_rmap)
 619{
 620	pgd_t *pgd;
 621	p4d_t *p4d;
 622	pud_t *pud, *new_pud = NULL;
 623	pmd_t *pmd, *new_pmd = NULL;
 624	pte_t *ptep, *new_ptep = NULL;
 625	int ret;
 626
 627	/* Traverse the guest's 2nd-level tree, allocate new levels needed */
 628	pgd = pgtable + pgd_index(gpa);
 629	p4d = p4d_offset(pgd, gpa);
 630
 631	pud = NULL;
 632	if (p4d_present(*p4d))
 633		pud = pud_offset(p4d, gpa);
 634	else
 635		new_pud = pud_alloc_one(kvm->mm, gpa);
 636
 637	pmd = NULL;
 638	if (pud && pud_present(*pud) && !pud_leaf(*pud))
 639		pmd = pmd_offset(pud, gpa);
 640	else if (level <= 1)
 641		new_pmd = kvmppc_pmd_alloc();
 642
 643	if (level == 0 && !(pmd && pmd_present(*pmd) && !pmd_leaf(*pmd)))
 644		new_ptep = kvmppc_pte_alloc();
 645
 646	/* Check if we might have been invalidated; let the guest retry if so */
 647	spin_lock(&kvm->mmu_lock);
 648	ret = -EAGAIN;
 649	if (mmu_invalidate_retry(kvm, mmu_seq))
 650		goto out_unlock;
 651
 652	/* Now traverse again under the lock and change the tree */
 653	ret = -ENOMEM;
 654	if (p4d_none(*p4d)) {
 655		if (!new_pud)
 656			goto out_unlock;
 657		p4d_populate(kvm->mm, p4d, new_pud);
 658		new_pud = NULL;
 659	}
 660	pud = pud_offset(p4d, gpa);
 661	if (pud_leaf(*pud)) {
 662		unsigned long hgpa = gpa & PUD_MASK;
 663
 664		/* Check if we raced and someone else has set the same thing */
 665		if (level == 2) {
 666			if (pud_raw(*pud) == pte_raw(pte)) {
 667				ret = 0;
 668				goto out_unlock;
 669			}
 670			/* Valid 1GB page here already, add our extra bits */
 671			WARN_ON_ONCE((pud_val(*pud) ^ pte_val(pte)) &
 672							PTE_BITS_MUST_MATCH);
 673			kvmppc_radix_update_pte(kvm, (pte_t *)pud,
 674					      0, pte_val(pte), hgpa, PUD_SHIFT);
 675			ret = 0;
 676			goto out_unlock;
 677		}
 678		/*
 679		 * If we raced with another CPU which has just put
 680		 * a 1GB pte in after we saw a pmd page, try again.
 681		 */
 682		if (!new_pmd) {
 683			ret = -EAGAIN;
 684			goto out_unlock;
 685		}
 686		/* Valid 1GB page here already, remove it */
 687		kvmppc_unmap_pte(kvm, (pte_t *)pud, hgpa, PUD_SHIFT, NULL,
 688				 lpid);
 689	}
 690	if (level == 2) {
 691		if (!pud_none(*pud)) {
 692			/*
 693			 * There's a page table page here, but we wanted to
 694			 * install a large page, so remove and free the page
 695			 * table page.
 696			 */
 697			kvmppc_unmap_free_pud_entry_table(kvm, pud, gpa, lpid);
 698		}
 699		kvmppc_radix_set_pte_at(kvm, gpa, (pte_t *)pud, pte);
 700		if (rmapp && n_rmap)
 701			kvmhv_insert_nest_rmap(kvm, rmapp, n_rmap);
 702		ret = 0;
 703		goto out_unlock;
 704	}
 705	if (pud_none(*pud)) {
 706		if (!new_pmd)
 707			goto out_unlock;
 708		pud_populate(kvm->mm, pud, new_pmd);
 709		new_pmd = NULL;
 710	}
 711	pmd = pmd_offset(pud, gpa);
 712	if (pmd_leaf(*pmd)) {
 713		unsigned long lgpa = gpa & PMD_MASK;
 714
 715		/* Check if we raced and someone else has set the same thing */
 716		if (level == 1) {
 717			if (pmd_raw(*pmd) == pte_raw(pte)) {
 718				ret = 0;
 719				goto out_unlock;
 720			}
 721			/* Valid 2MB page here already, add our extra bits */
 722			WARN_ON_ONCE((pmd_val(*pmd) ^ pte_val(pte)) &
 723							PTE_BITS_MUST_MATCH);
 724			kvmppc_radix_update_pte(kvm, pmdp_ptep(pmd),
 725					0, pte_val(pte), lgpa, PMD_SHIFT);
 726			ret = 0;
 727			goto out_unlock;
 728		}
 729
 730		/*
 731		 * If we raced with another CPU which has just put
 732		 * a 2MB pte in after we saw a pte page, try again.
 733		 */
 734		if (!new_ptep) {
 735			ret = -EAGAIN;
 736			goto out_unlock;
 737		}
 738		/* Valid 2MB page here already, remove it */
 739		kvmppc_unmap_pte(kvm, pmdp_ptep(pmd), lgpa, PMD_SHIFT, NULL,
 740				 lpid);
 741	}
 742	if (level == 1) {
 743		if (!pmd_none(*pmd)) {
 744			/*
 745			 * There's a page table page here, but we wanted to
 746			 * install a large page, so remove and free the page
 747			 * table page.
 748			 */
 749			kvmppc_unmap_free_pmd_entry_table(kvm, pmd, gpa, lpid);
 750		}
 751		kvmppc_radix_set_pte_at(kvm, gpa, pmdp_ptep(pmd), pte);
 752		if (rmapp && n_rmap)
 753			kvmhv_insert_nest_rmap(kvm, rmapp, n_rmap);
 754		ret = 0;
 755		goto out_unlock;
 756	}
 757	if (pmd_none(*pmd)) {
 758		if (!new_ptep)
 759			goto out_unlock;
 760		pmd_populate(kvm->mm, pmd, new_ptep);
 761		new_ptep = NULL;
 762	}
 763	ptep = pte_offset_kernel(pmd, gpa);
 764	if (pte_present(*ptep)) {
 765		/* Check if someone else set the same thing */
 766		if (pte_raw(*ptep) == pte_raw(pte)) {
 767			ret = 0;
 768			goto out_unlock;
 769		}
 770		/* Valid page here already, add our extra bits */
 771		WARN_ON_ONCE((pte_val(*ptep) ^ pte_val(pte)) &
 772							PTE_BITS_MUST_MATCH);
 773		kvmppc_radix_update_pte(kvm, ptep, 0, pte_val(pte), gpa, 0);
 774		ret = 0;
 775		goto out_unlock;
 776	}
 777	kvmppc_radix_set_pte_at(kvm, gpa, ptep, pte);
 778	if (rmapp && n_rmap)
 779		kvmhv_insert_nest_rmap(kvm, rmapp, n_rmap);
 780	ret = 0;
 781
 782 out_unlock:
 783	spin_unlock(&kvm->mmu_lock);
 784	if (new_pud)
 785		pud_free(kvm->mm, new_pud);
 786	if (new_pmd)
 787		kvmppc_pmd_free(new_pmd);
 788	if (new_ptep)
 789		kvmppc_pte_free(new_ptep);
 790	return ret;
 791}
 792
 793bool kvmppc_hv_handle_set_rc(struct kvm *kvm, bool nested, bool writing,
 794			     unsigned long gpa, u64 lpid)
 795{
 796	unsigned long pgflags;
 797	unsigned int shift;
 798	pte_t *ptep;
 799
 800	/*
 801	 * Need to set an R or C bit in the 2nd-level tables;
 802	 * since we are just helping out the hardware here,
 803	 * it is sufficient to do what the hardware does.
 804	 */
 805	pgflags = _PAGE_ACCESSED;
 806	if (writing)
 807		pgflags |= _PAGE_DIRTY;
 808
 809	if (nested)
 810		ptep = find_kvm_nested_guest_pte(kvm, lpid, gpa, &shift);
 811	else
 812		ptep = find_kvm_secondary_pte(kvm, gpa, &shift);
 813
 814	if (ptep && pte_present(*ptep) && (!writing || pte_write(*ptep))) {
 815		kvmppc_radix_update_pte(kvm, ptep, 0, pgflags, gpa, shift);
 816		return true;
 817	}
 818	return false;
 819}
 820
 821int kvmppc_book3s_instantiate_page(struct kvm_vcpu *vcpu,
 822				   unsigned long gpa,
 823				   struct kvm_memory_slot *memslot,
 824				   bool writing,
 825				   pte_t *inserted_pte, unsigned int *levelp)
 826{
 827	struct kvm *kvm = vcpu->kvm;
 828	struct page *page = NULL;
 829	unsigned long mmu_seq;
 830	unsigned long hva, gfn = gpa >> PAGE_SHIFT;
 831	bool upgrade_write = false;
 832	pte_t pte, *ptep;
 833	unsigned int shift, level;
 834	int ret;
 835	bool large_enable;
 836	kvm_pfn_t pfn;
 837
 838	/* used to check for invalidations in progress */
 839	mmu_seq = kvm->mmu_invalidate_seq;
 840	smp_rmb();
 841
 842	hva = gfn_to_hva_memslot(memslot, gfn);
 843	pfn = __kvm_faultin_pfn(memslot, gfn, writing ? FOLL_WRITE : 0,
 844				&upgrade_write, &page);
 845	if (is_error_noslot_pfn(pfn))
 846		return -EFAULT;
 847
 848	/*
 849	 * Read the PTE from the process' radix tree and use that
 850	 * so we get the shift and attribute bits.
 851	 */
 852	spin_lock(&kvm->mmu_lock);
 853	ptep = find_kvm_host_pte(kvm, mmu_seq, hva, &shift);
 854	pte = __pte(0);
 855	if (ptep)
 856		pte = READ_ONCE(*ptep);
 857	spin_unlock(&kvm->mmu_lock);
 858	/*
 859	 * If the PTE disappeared temporarily due to a THP
 860	 * collapse, just return and let the guest try again.
 861	 */
 862	if (!pte_present(pte)) {
 863		if (page)
 864			put_page(page);
 865		return RESUME_GUEST;
 866	}
 867
 868	/* If we're logging dirty pages, always map single pages */
 869	large_enable = !(memslot->flags & KVM_MEM_LOG_DIRTY_PAGES);
 870
 871	/* Get pte level from shift/size */
 872	if (large_enable && shift == PUD_SHIFT &&
 873	    (gpa & (PUD_SIZE - PAGE_SIZE)) ==
 874	    (hva & (PUD_SIZE - PAGE_SIZE))) {
 875		level = 2;
 876	} else if (large_enable && shift == PMD_SHIFT &&
 877		   (gpa & (PMD_SIZE - PAGE_SIZE)) ==
 878		   (hva & (PMD_SIZE - PAGE_SIZE))) {
 879		level = 1;
 880	} else {
 881		level = 0;
 882		if (shift > PAGE_SHIFT) {
 883			/*
 884			 * If the pte maps more than one page, bring over
 885			 * bits from the virtual address to get the real
 886			 * address of the specific single page we want.
 887			 */
 888			unsigned long rpnmask = (1ul << shift) - PAGE_SIZE;
 889			pte = __pte(pte_val(pte) | (hva & rpnmask));
 890		}
 891	}
 892
 893	pte = __pte(pte_val(pte) | _PAGE_EXEC | _PAGE_ACCESSED);
 894	if (writing || upgrade_write) {
 895		if (pte_val(pte) & _PAGE_WRITE)
 896			pte = __pte(pte_val(pte) | _PAGE_DIRTY);
 897	} else {
 898		pte = __pte(pte_val(pte) & ~(_PAGE_WRITE | _PAGE_DIRTY));
 899	}
 900
 901	/* Allocate space in the tree and write the PTE */
 902	ret = kvmppc_create_pte(kvm, kvm->arch.pgtable, pte, gpa, level,
 903				mmu_seq, kvm->arch.lpid, NULL, NULL);
 904	if (inserted_pte)
 905		*inserted_pte = pte;
 906	if (levelp)
 907		*levelp = level;
 908
 909	if (page) {
 910		if (!ret && (pte_val(pte) & _PAGE_WRITE))
 911			set_page_dirty_lock(page);
 912		put_page(page);
 913	}
 914
 915	/* Increment number of large pages if we (successfully) inserted one */
 916	if (!ret) {
 917		if (level == 1)
 918			kvm->stat.num_2M_pages++;
 919		else if (level == 2)
 920			kvm->stat.num_1G_pages++;
 921	}
 922
 923	return ret;
 924}
 925
 926int kvmppc_book3s_radix_page_fault(struct kvm_vcpu *vcpu,
 927				   unsigned long ea, unsigned long dsisr)
 928{
 929	struct kvm *kvm = vcpu->kvm;
 930	unsigned long gpa, gfn;
 931	struct kvm_memory_slot *memslot;
 932	long ret;
 933	bool writing = !!(dsisr & DSISR_ISSTORE);
 934
 935	/* Check for unusual errors */
 936	if (dsisr & DSISR_UNSUPP_MMU) {
 937		pr_err("KVM: Got unsupported MMU fault\n");
 938		return -EFAULT;
 939	}
 940	if (dsisr & DSISR_BADACCESS) {
 941		/* Reflect to the guest as DSI */
 942		pr_err("KVM: Got radix HV page fault with DSISR=%lx\n", dsisr);
 943		kvmppc_core_queue_data_storage(vcpu,
 944				kvmppc_get_msr(vcpu) & SRR1_PREFIXED,
 945				ea, dsisr);
 946		return RESUME_GUEST;
 947	}
 948
 949	/* Translate the logical address */
 950	gpa = vcpu->arch.fault_gpa & ~0xfffUL;
 951	gpa &= ~0xF000000000000000ul;
 952	gfn = gpa >> PAGE_SHIFT;
 953	if (!(dsisr & DSISR_PRTABLE_FAULT))
 954		gpa |= ea & 0xfff;
 955
 956	if (kvm->arch.secure_guest & KVMPPC_SECURE_INIT_DONE)
 957		return kvmppc_send_page_to_uv(kvm, gfn);
 958
 959	/* Get the corresponding memslot */
 960	memslot = gfn_to_memslot(kvm, gfn);
 961
 962	/* No memslot means it's an emulated MMIO region */
 963	if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID)) {
 964		if (dsisr & (DSISR_PRTABLE_FAULT | DSISR_BADACCESS |
 965			     DSISR_SET_RC)) {
 966			/*
 967			 * Bad address in guest page table tree, or other
 968			 * unusual error - reflect it to the guest as DSI.
 969			 */
 970			kvmppc_core_queue_data_storage(vcpu,
 971					kvmppc_get_msr(vcpu) & SRR1_PREFIXED,
 972					ea, dsisr);
 973			return RESUME_GUEST;
 974		}
 975		return kvmppc_hv_emulate_mmio(vcpu, gpa, ea, writing);
 976	}
 977
 978	if (memslot->flags & KVM_MEM_READONLY) {
 979		if (writing) {
 980			/* give the guest a DSI */
 981			kvmppc_core_queue_data_storage(vcpu,
 982					kvmppc_get_msr(vcpu) & SRR1_PREFIXED,
 983					ea, DSISR_ISSTORE | DSISR_PROTFAULT);
 984			return RESUME_GUEST;
 985		}
 986	}
 987
 988	/* Failed to set the reference/change bits */
 989	if (dsisr & DSISR_SET_RC) {
 990		spin_lock(&kvm->mmu_lock);
 991		if (kvmppc_hv_handle_set_rc(kvm, false, writing,
 992					    gpa, kvm->arch.lpid))
 993			dsisr &= ~DSISR_SET_RC;
 994		spin_unlock(&kvm->mmu_lock);
 995
 996		if (!(dsisr & (DSISR_BAD_FAULT_64S | DSISR_NOHPTE |
 997			       DSISR_PROTFAULT | DSISR_SET_RC)))
 998			return RESUME_GUEST;
 999	}
1000
1001	/* Try to insert a pte */
1002	ret = kvmppc_book3s_instantiate_page(vcpu, gpa, memslot, writing,
1003					     NULL, NULL);
1004
1005	if (ret == 0 || ret == -EAGAIN)
1006		ret = RESUME_GUEST;
1007	return ret;
1008}
1009
1010/* Called with kvm->mmu_lock held */
1011void kvm_unmap_radix(struct kvm *kvm, struct kvm_memory_slot *memslot,
1012		     unsigned long gfn)
1013{
1014	pte_t *ptep;
1015	unsigned long gpa = gfn << PAGE_SHIFT;
1016	unsigned int shift;
1017
1018	if (kvm->arch.secure_guest & KVMPPC_SECURE_INIT_DONE) {
1019		uv_page_inval(kvm->arch.lpid, gpa, PAGE_SHIFT);
1020		return;
1021	}
1022
1023	ptep = find_kvm_secondary_pte(kvm, gpa, &shift);
1024	if (ptep && pte_present(*ptep))
1025		kvmppc_unmap_pte(kvm, ptep, gpa, shift, memslot,
1026				 kvm->arch.lpid);
1027}
1028
1029/* Called with kvm->mmu_lock held */
1030bool kvm_age_radix(struct kvm *kvm, struct kvm_memory_slot *memslot,
1031		   unsigned long gfn)
1032{
1033	pte_t *ptep;
1034	unsigned long gpa = gfn << PAGE_SHIFT;
1035	unsigned int shift;
1036	bool ref = false;
1037	unsigned long old, *rmapp;
1038
1039	if (kvm->arch.secure_guest & KVMPPC_SECURE_INIT_DONE)
1040		return ref;
1041
1042	ptep = find_kvm_secondary_pte(kvm, gpa, &shift);
1043	if (ptep && pte_present(*ptep) && pte_young(*ptep)) {
1044		old = kvmppc_radix_update_pte(kvm, ptep, _PAGE_ACCESSED, 0,
1045					      gpa, shift);
1046		/* XXX need to flush tlb here? */
1047		/* Also clear bit in ptes in shadow pgtable for nested guests */
1048		rmapp = &memslot->arch.rmap[gfn - memslot->base_gfn];
1049		kvmhv_update_nest_rmap_rc_list(kvm, rmapp, _PAGE_ACCESSED, 0,
1050					       old & PTE_RPN_MASK,
1051					       1UL << shift);
1052		ref = true;
1053	}
1054	return ref;
1055}
1056
1057/* Called with kvm->mmu_lock held */
1058bool kvm_test_age_radix(struct kvm *kvm, struct kvm_memory_slot *memslot,
1059			unsigned long gfn)
1060
1061{
1062	pte_t *ptep;
1063	unsigned long gpa = gfn << PAGE_SHIFT;
1064	unsigned int shift;
1065	bool ref = false;
1066
1067	if (kvm->arch.secure_guest & KVMPPC_SECURE_INIT_DONE)
1068		return ref;
1069
1070	ptep = find_kvm_secondary_pte(kvm, gpa, &shift);
1071	if (ptep && pte_present(*ptep) && pte_young(*ptep))
1072		ref = true;
1073	return ref;
1074}
1075
1076/* Returns the number of PAGE_SIZE pages that are dirty */
1077static int kvm_radix_test_clear_dirty(struct kvm *kvm,
1078				struct kvm_memory_slot *memslot, int pagenum)
1079{
1080	unsigned long gfn = memslot->base_gfn + pagenum;
1081	unsigned long gpa = gfn << PAGE_SHIFT;
1082	pte_t *ptep, pte;
1083	unsigned int shift;
1084	int ret = 0;
1085	unsigned long old, *rmapp;
1086
1087	if (kvm->arch.secure_guest & KVMPPC_SECURE_INIT_DONE)
1088		return ret;
1089
1090	/*
1091	 * For performance reasons we don't hold kvm->mmu_lock while walking the
1092	 * partition scoped table.
1093	 */
1094	ptep = find_kvm_secondary_pte_unlocked(kvm, gpa, &shift);
1095	if (!ptep)
1096		return 0;
1097
1098	pte = READ_ONCE(*ptep);
1099	if (pte_present(pte) && pte_dirty(pte)) {
1100		spin_lock(&kvm->mmu_lock);
1101		/*
1102		 * Recheck the pte again
1103		 */
1104		if (pte_val(pte) != pte_val(*ptep)) {
1105			/*
1106			 * We have KVM_MEM_LOG_DIRTY_PAGES enabled. Hence we can
1107			 * only find PAGE_SIZE pte entries here. We can continue
1108			 * to use the pte addr returned by above page table
1109			 * walk.
1110			 */
1111			if (!pte_present(*ptep) || !pte_dirty(*ptep)) {
1112				spin_unlock(&kvm->mmu_lock);
1113				return 0;
1114			}
1115		}
1116
1117		ret = 1;
1118		VM_BUG_ON(shift);
1119		old = kvmppc_radix_update_pte(kvm, ptep, _PAGE_DIRTY, 0,
1120					      gpa, shift);
1121		kvmppc_radix_tlbie_page(kvm, gpa, shift, kvm->arch.lpid);
1122		/* Also clear bit in ptes in shadow pgtable for nested guests */
1123		rmapp = &memslot->arch.rmap[gfn - memslot->base_gfn];
1124		kvmhv_update_nest_rmap_rc_list(kvm, rmapp, _PAGE_DIRTY, 0,
1125					       old & PTE_RPN_MASK,
1126					       1UL << shift);
1127		spin_unlock(&kvm->mmu_lock);
1128	}
1129	return ret;
1130}
1131
1132long kvmppc_hv_get_dirty_log_radix(struct kvm *kvm,
1133			struct kvm_memory_slot *memslot, unsigned long *map)
1134{
1135	unsigned long i, j;
1136	int npages;
1137
1138	for (i = 0; i < memslot->npages; i = j) {
1139		npages = kvm_radix_test_clear_dirty(kvm, memslot, i);
1140
1141		/*
1142		 * Note that if npages > 0 then i must be a multiple of npages,
1143		 * since huge pages are only used to back the guest at guest
1144		 * real addresses that are a multiple of their size.
1145		 * Since we have at most one PTE covering any given guest
1146		 * real address, if npages > 1 we can skip to i + npages.
1147		 */
1148		j = i + 1;
1149		if (npages) {
1150			set_dirty_bits(map, i, npages);
1151			j = i + npages;
1152		}
1153	}
1154	return 0;
1155}
1156
1157void kvmppc_radix_flush_memslot(struct kvm *kvm,
1158				const struct kvm_memory_slot *memslot)
1159{
1160	unsigned long n;
1161	pte_t *ptep;
1162	unsigned long gpa;
1163	unsigned int shift;
1164
1165	if (kvm->arch.secure_guest & KVMPPC_SECURE_INIT_START)
1166		kvmppc_uvmem_drop_pages(memslot, kvm, true);
1167
1168	if (kvm->arch.secure_guest & KVMPPC_SECURE_INIT_DONE)
1169		return;
1170
1171	gpa = memslot->base_gfn << PAGE_SHIFT;
1172	spin_lock(&kvm->mmu_lock);
1173	for (n = memslot->npages; n; --n) {
1174		ptep = find_kvm_secondary_pte(kvm, gpa, &shift);
1175		if (ptep && pte_present(*ptep))
1176			kvmppc_unmap_pte(kvm, ptep, gpa, shift, memslot,
1177					 kvm->arch.lpid);
1178		gpa += PAGE_SIZE;
1179	}
1180	/*
1181	 * Increase the mmu notifier sequence number to prevent any page
1182	 * fault that read the memslot earlier from writing a PTE.
1183	 */
1184	kvm->mmu_invalidate_seq++;
1185	spin_unlock(&kvm->mmu_lock);
1186}
1187
1188static void add_rmmu_ap_encoding(struct kvm_ppc_rmmu_info *info,
1189				 int psize, int *indexp)
1190{
1191	if (!mmu_psize_defs[psize].shift)
1192		return;
1193	info->ap_encodings[*indexp] = mmu_psize_defs[psize].shift |
1194		(mmu_psize_defs[psize].ap << 29);
1195	++(*indexp);
1196}
1197
1198int kvmhv_get_rmmu_info(struct kvm *kvm, struct kvm_ppc_rmmu_info *info)
1199{
1200	int i;
1201
1202	if (!radix_enabled())
1203		return -EINVAL;
1204	memset(info, 0, sizeof(*info));
1205
1206	/* 4k page size */
1207	info->geometries[0].page_shift = 12;
1208	info->geometries[0].level_bits[0] = 9;
1209	for (i = 1; i < 4; ++i)
1210		info->geometries[0].level_bits[i] = p9_supported_radix_bits[i];
1211	/* 64k page size */
1212	info->geometries[1].page_shift = 16;
1213	for (i = 0; i < 4; ++i)
1214		info->geometries[1].level_bits[i] = p9_supported_radix_bits[i];
1215
1216	i = 0;
1217	add_rmmu_ap_encoding(info, MMU_PAGE_4K, &i);
1218	add_rmmu_ap_encoding(info, MMU_PAGE_64K, &i);
1219	add_rmmu_ap_encoding(info, MMU_PAGE_2M, &i);
1220	add_rmmu_ap_encoding(info, MMU_PAGE_1G, &i);
1221
1222	return 0;
1223}
1224
1225int kvmppc_init_vm_radix(struct kvm *kvm)
1226{
1227	kvm->arch.pgtable = pgd_alloc(kvm->mm);
1228	if (!kvm->arch.pgtable)
1229		return -ENOMEM;
1230	return 0;
1231}
1232
1233static void pte_ctor(void *addr)
1234{
1235	memset(addr, 0, RADIX_PTE_TABLE_SIZE);
1236}
1237
1238static void pmd_ctor(void *addr)
1239{
1240	memset(addr, 0, RADIX_PMD_TABLE_SIZE);
1241}
1242
1243struct debugfs_radix_state {
1244	struct kvm	*kvm;
1245	struct mutex	mutex;
1246	unsigned long	gpa;
1247	int		lpid;
1248	int		chars_left;
1249	int		buf_index;
1250	char		buf[128];
1251	u8		hdr;
1252};
1253
1254static int debugfs_radix_open(struct inode *inode, struct file *file)
1255{
1256	struct kvm *kvm = inode->i_private;
1257	struct debugfs_radix_state *p;
1258
1259	p = kzalloc(sizeof(*p), GFP_KERNEL);
1260	if (!p)
1261		return -ENOMEM;
1262
1263	kvm_get_kvm(kvm);
1264	p->kvm = kvm;
1265	mutex_init(&p->mutex);
1266	file->private_data = p;
1267
1268	return nonseekable_open(inode, file);
1269}
1270
1271static int debugfs_radix_release(struct inode *inode, struct file *file)
1272{
1273	struct debugfs_radix_state *p = file->private_data;
1274
1275	kvm_put_kvm(p->kvm);
1276	kfree(p);
1277	return 0;
1278}
1279
1280static ssize_t debugfs_radix_read(struct file *file, char __user *buf,
1281				 size_t len, loff_t *ppos)
1282{
1283	struct debugfs_radix_state *p = file->private_data;
1284	ssize_t ret, r;
1285	unsigned long n;
1286	struct kvm *kvm;
1287	unsigned long gpa;
1288	pgd_t *pgt;
1289	struct kvm_nested_guest *nested;
1290	pgd_t *pgdp;
1291	p4d_t p4d, *p4dp;
1292	pud_t pud, *pudp;
1293	pmd_t pmd, *pmdp;
1294	pte_t *ptep;
1295	int shift;
1296	unsigned long pte;
1297
1298	kvm = p->kvm;
1299	if (!kvm_is_radix(kvm))
1300		return 0;
1301
1302	ret = mutex_lock_interruptible(&p->mutex);
1303	if (ret)
1304		return ret;
1305
1306	if (p->chars_left) {
1307		n = p->chars_left;
1308		if (n > len)
1309			n = len;
1310		r = copy_to_user(buf, p->buf + p->buf_index, n);
1311		n -= r;
1312		p->chars_left -= n;
1313		p->buf_index += n;
1314		buf += n;
1315		len -= n;
1316		ret = n;
1317		if (r) {
1318			if (!n)
1319				ret = -EFAULT;
1320			goto out;
1321		}
1322	}
1323
1324	gpa = p->gpa;
1325	nested = NULL;
1326	pgt = NULL;
1327	while (len != 0 && p->lpid >= 0) {
1328		if (gpa >= RADIX_PGTABLE_RANGE) {
1329			gpa = 0;
1330			pgt = NULL;
1331			if (nested) {
1332				kvmhv_put_nested(nested);
1333				nested = NULL;
1334			}
1335			p->lpid = kvmhv_nested_next_lpid(kvm, p->lpid);
1336			p->hdr = 0;
1337			if (p->lpid < 0)
1338				break;
1339		}
1340		if (!pgt) {
1341			if (p->lpid == 0) {
1342				pgt = kvm->arch.pgtable;
1343			} else {
1344				nested = kvmhv_get_nested(kvm, p->lpid, false);
1345				if (!nested) {
1346					gpa = RADIX_PGTABLE_RANGE;
1347					continue;
1348				}
1349				pgt = nested->shadow_pgtable;
1350			}
1351		}
1352		n = 0;
1353		if (!p->hdr) {
1354			if (p->lpid > 0)
1355				n = scnprintf(p->buf, sizeof(p->buf),
1356					      "\nNested LPID %d: ", p->lpid);
1357			n += scnprintf(p->buf + n, sizeof(p->buf) - n,
1358				      "pgdir: %lx\n", (unsigned long)pgt);
1359			p->hdr = 1;
1360			goto copy;
1361		}
1362
1363		pgdp = pgt + pgd_index(gpa);
1364		p4dp = p4d_offset(pgdp, gpa);
1365		p4d = READ_ONCE(*p4dp);
1366		if (!(p4d_val(p4d) & _PAGE_PRESENT)) {
1367			gpa = (gpa & P4D_MASK) + P4D_SIZE;
1368			continue;
1369		}
1370
1371		pudp = pud_offset(&p4d, gpa);
1372		pud = READ_ONCE(*pudp);
1373		if (!(pud_val(pud) & _PAGE_PRESENT)) {
1374			gpa = (gpa & PUD_MASK) + PUD_SIZE;
1375			continue;
1376		}
1377		if (pud_val(pud) & _PAGE_PTE) {
1378			pte = pud_val(pud);
1379			shift = PUD_SHIFT;
1380			goto leaf;
1381		}
1382
1383		pmdp = pmd_offset(&pud, gpa);
1384		pmd = READ_ONCE(*pmdp);
1385		if (!(pmd_val(pmd) & _PAGE_PRESENT)) {
1386			gpa = (gpa & PMD_MASK) + PMD_SIZE;
1387			continue;
1388		}
1389		if (pmd_val(pmd) & _PAGE_PTE) {
1390			pte = pmd_val(pmd);
1391			shift = PMD_SHIFT;
1392			goto leaf;
1393		}
1394
1395		ptep = pte_offset_kernel(&pmd, gpa);
1396		pte = pte_val(READ_ONCE(*ptep));
1397		if (!(pte & _PAGE_PRESENT)) {
1398			gpa += PAGE_SIZE;
1399			continue;
1400		}
1401		shift = PAGE_SHIFT;
1402	leaf:
1403		n = scnprintf(p->buf, sizeof(p->buf),
1404			      " %lx: %lx %d\n", gpa, pte, shift);
1405		gpa += 1ul << shift;
1406	copy:
1407		p->chars_left = n;
1408		if (n > len)
1409			n = len;
1410		r = copy_to_user(buf, p->buf, n);
1411		n -= r;
1412		p->chars_left -= n;
1413		p->buf_index = n;
1414		buf += n;
1415		len -= n;
1416		ret += n;
1417		if (r) {
1418			if (!ret)
1419				ret = -EFAULT;
1420			break;
1421		}
1422	}
1423	p->gpa = gpa;
1424	if (nested)
1425		kvmhv_put_nested(nested);
1426
1427 out:
1428	mutex_unlock(&p->mutex);
1429	return ret;
1430}
1431
1432static ssize_t debugfs_radix_write(struct file *file, const char __user *buf,
1433			   size_t len, loff_t *ppos)
1434{
1435	return -EACCES;
1436}
1437
1438static const struct file_operations debugfs_radix_fops = {
1439	.owner	 = THIS_MODULE,
1440	.open	 = debugfs_radix_open,
1441	.release = debugfs_radix_release,
1442	.read	 = debugfs_radix_read,
1443	.write	 = debugfs_radix_write,
1444	.llseek	 = generic_file_llseek,
1445};
1446
1447void kvmhv_radix_debugfs_init(struct kvm *kvm)
1448{
1449	debugfs_create_file("radix", 0400, kvm->debugfs_dentry, kvm,
1450			    &debugfs_radix_fops);
1451}
1452
1453int kvmppc_radix_init(void)
1454{
1455	unsigned long size = sizeof(void *) << RADIX_PTE_INDEX_SIZE;
1456
1457	kvm_pte_cache = kmem_cache_create("kvm-pte", size, size, 0, pte_ctor);
1458	if (!kvm_pte_cache)
1459		return -ENOMEM;
1460
1461	size = sizeof(void *) << RADIX_PMD_INDEX_SIZE;
1462
1463	kvm_pmd_cache = kmem_cache_create("kvm-pmd", size, size, 0, pmd_ctor);
1464	if (!kvm_pmd_cache) {
1465		kmem_cache_destroy(kvm_pte_cache);
1466		return -ENOMEM;
1467	}
1468
1469	return 0;
1470}
1471
1472void kvmppc_radix_exit(void)
1473{
1474	kmem_cache_destroy(kvm_pte_cache);
1475	kmem_cache_destroy(kvm_pmd_cache);
1476}