1 // SPDX-License-Identifier: GPL-2.0-only
4 * Copyright 2016 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
7 #include <linux/types.h>
8 #include <linux/string.h>
10 #include <linux/kvm_host.h>
11 #include <linux/anon_inodes.h>
12 #include <linux/file.h>
13 #include <linux/debugfs.h>
15 #include <asm/kvm_ppc.h>
16 #include <asm/kvm_book3s.h>
19 #include <asm/pgtable.h>
20 #include <asm/pgalloc.h>
21 #include <asm/pte-walk.h>
24 * Supported radix tree geometry.
25 * Like p9, we support either 5 or 9 bits at the first (lowest) level,
26 * for a page size of 64k or 4k.
28 static int p9_supported_radix_bits[4] = { 5, 9, 9, 13 };
30 unsigned long __kvmhv_copy_tofrom_guest_radix(int lpid, int pid,
31 gva_t eaddr, void *to, void *from,
34 int uninitialized_var(old_pid), old_lpid;
35 unsigned long quadrant, ret = n;
38 /* Can't access quadrants 1 or 2 in non-HV mode, call the HV to do it */
39 if (kvmhv_on_pseries())
40 return plpar_hcall_norets(H_COPY_TOFROM_GUEST, lpid, pid, eaddr,
41 __pa(to), __pa(from), n);
47 from = (void *) (eaddr | (quadrant << 62));
49 to = (void *) (eaddr | (quadrant << 62));
53 /* switch the lpid first to avoid running host with unallocated pid */
54 old_lpid = mfspr(SPRN_LPID);
56 mtspr(SPRN_LPID, lpid);
58 old_pid = mfspr(SPRN_PID);
66 ret = raw_copy_from_user(to, from, n);
68 ret = raw_copy_to_user(to, from, n);
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);
75 mtspr(SPRN_LPID, old_lpid);
82 EXPORT_SYMBOL_GPL(__kvmhv_copy_tofrom_guest_radix);
84 static long kvmhv_copy_tofrom_guest_radix(struct kvm_vcpu *vcpu, gva_t eaddr,
85 void *to, void *from, unsigned long n)
87 int lpid = vcpu->kvm->arch.lpid;
88 int pid = vcpu->arch.pid;
90 /* This would cause a data segment intr so don't allow the access */
91 if (eaddr & (0x3FFUL << 52))
94 /* Should we be using the nested lpid */
95 if (vcpu->arch.nested)
96 lpid = vcpu->arch.nested->shadow_lpid;
98 /* If accessing quadrant 3 then pid is expected to be 0 */
99 if (((eaddr >> 62) & 0x3) == 0x3)
102 eaddr &= ~(0xFFFUL << 52);
104 return __kvmhv_copy_tofrom_guest_radix(lpid, pid, eaddr, to, from, n);
107 long kvmhv_copy_from_guest_radix(struct kvm_vcpu *vcpu, gva_t eaddr, void *to,
112 ret = kvmhv_copy_tofrom_guest_radix(vcpu, eaddr, to, NULL, n);
114 memset(to + (n - ret), 0, ret);
118 EXPORT_SYMBOL_GPL(kvmhv_copy_from_guest_radix);
120 long kvmhv_copy_to_guest_radix(struct kvm_vcpu *vcpu, gva_t eaddr, void *from,
123 return kvmhv_copy_tofrom_guest_radix(vcpu, eaddr, NULL, from, n);
125 EXPORT_SYMBOL_GPL(kvmhv_copy_to_guest_radix);
127 int kvmppc_mmu_walk_radix_tree(struct kvm_vcpu *vcpu, gva_t eaddr,
128 struct kvmppc_pte *gpte, u64 root,
131 struct kvm *kvm = vcpu->kvm;
133 unsigned long rts, bits, offset, index;
137 rts = ((root & RTS1_MASK) >> (RTS1_SHIFT - 3)) |
138 ((root & RTS2_MASK) >> RTS2_SHIFT);
139 bits = root & RPDS_MASK;
140 base = root & RPDB_MASK;
144 /* Current implementations only support 52-bit space */
148 /* Walk each level of the radix tree */
149 for (level = 3; level >= 0; --level) {
151 /* Check a valid size */
152 if (level && bits != p9_supported_radix_bits[level])
154 if (level == 0 && !(bits == 5 || bits == 9))
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))
161 /* Read the entry from guest memory */
162 addr = base + (index * sizeof(rpte));
163 ret = kvm_read_guest(kvm, addr, &rpte, sizeof(rpte));
169 pte = __be64_to_cpu(rpte);
170 if (!(pte & _PAGE_PRESENT))
172 /* Check if a leaf entry */
175 /* Get ready to walk the next level */
176 base = pte & RPDB_MASK;
177 bits = pte & RPDS_MASK;
180 /* Need a leaf at lowest level; 512GB pages not supported */
181 if (level < 0 || level == 3)
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))
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)
193 gpte->page_size = ps;
194 gpte->page_shift = offset;
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);
204 gpte->rc = pte & (_PAGE_ACCESSED | _PAGE_DIRTY);
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
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)
224 struct kvm *kvm = vcpu->kvm;
226 unsigned long size, ptbl, root;
227 struct prtb_entry entry;
229 if ((table & PRTS_MASK) > 24)
231 size = 1ul << ((table & PRTS_MASK) + 12);
233 /* Is the table big enough to contain this entry? */
234 if ((table_index * sizeof(entry)) >= size)
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));
243 /* Root is stored in the first double word */
244 root = be64_to_cpu(entry.prtb0);
246 return kvmppc_mmu_walk_radix_tree(vcpu, eaddr, gpte, root, pte_ret_p);
249 int kvmppc_mmu_radix_xlate(struct kvm_vcpu *vcpu, gva_t eaddr,
250 struct kvmppc_pte *gpte, bool data, bool iswrite)
256 /* Work out effective PID */
257 switch (eaddr >> 62) {
259 pid = vcpu->arch.pid;
268 ret = kvmppc_mmu_radix_translate_table(vcpu, eaddr, gpte,
269 vcpu->kvm->arch.process_table, pid, &pte);
273 /* Check privilege (applies only to process scoped translations) */
274 if (kvmppc_get_msr(vcpu) & MSR_PR) {
275 if (pte & _PAGE_PRIVILEGED) {
278 gpte->may_execute = 0;
281 if (!(pte & _PAGE_PRIVILEGED)) {
282 /* Check AMR/IAMR to see if strict mode is in force */
283 if (vcpu->arch.amr & (1ul << 62))
285 if (vcpu->arch.amr & (1ul << 63))
287 if (vcpu->arch.iamr & (1ul << 62))
288 gpte->may_execute = 0;
295 void kvmppc_radix_tlbie_page(struct kvm *kvm, unsigned long addr,
296 unsigned int pshift, unsigned int lpid)
298 unsigned long psize = PAGE_SIZE;
304 psize = 1UL << pshift;
308 addr &= ~(psize - 1);
310 if (!kvmhv_on_pseries()) {
311 radix__flush_tlb_lpid_page(lpid, addr, psize);
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),
320 pr_err("KVM: TLB page invalidation hcall failed, rc=%ld\n", rc);
323 static void kvmppc_radix_flush_pwc(struct kvm *kvm, unsigned int lpid)
327 if (!kvmhv_on_pseries()) {
328 radix__flush_pwc_lpid(lpid);
332 rc = plpar_hcall_norets(H_TLB_INVALIDATE, H_TLBIE_P1_ENC(1, 0, 1),
333 lpid, TLBIEL_INVAL_SET_LPID);
335 pr_err("KVM: TLB PWC invalidation hcall failed, rc=%ld\n", rc);
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)
342 return __radix_pte_update(ptep, clr, set);
345 void kvmppc_radix_set_pte_at(struct kvm *kvm, unsigned long addr,
346 pte_t *ptep, pte_t pte)
348 radix__set_pte_at(kvm->mm, addr, ptep, pte, 0);
351 static struct kmem_cache *kvm_pte_cache;
352 static struct kmem_cache *kvm_pmd_cache;
354 static pte_t *kvmppc_pte_alloc(void)
356 return kmem_cache_alloc(kvm_pte_cache, GFP_KERNEL);
359 static void kvmppc_pte_free(pte_t *ptep)
361 kmem_cache_free(kvm_pte_cache, ptep);
364 /* Like pmd_huge() and pmd_large(), but works regardless of config options */
365 static inline int pmd_is_leaf(pmd_t pmd)
367 return !!(pmd_val(pmd) & _PAGE_PTE);
370 static pmd_t *kvmppc_pmd_alloc(void)
372 return kmem_cache_alloc(kvm_pmd_cache, GFP_KERNEL);
375 static void kvmppc_pmd_free(pmd_t *pmdp)
377 kmem_cache_free(kvm_pmd_cache, pmdp);
380 /* Called with kvm->mmu_lock held */
381 void kvmppc_unmap_pte(struct kvm *kvm, pte_t *pte, unsigned long gpa,
383 const struct kvm_memory_slot *memslot,
388 unsigned long gfn = gpa >> PAGE_SHIFT;
389 unsigned long page_size = PAGE_SIZE;
392 old = kvmppc_radix_update_pte(kvm, pte, ~0UL, 0, gpa, shift);
393 kvmppc_radix_tlbie_page(kvm, gpa, shift, lpid);
395 /* The following only applies to L1 entries */
396 if (lpid != kvm->arch.lpid)
400 memslot = gfn_to_memslot(kvm, gfn);
404 if (shift) { /* 1GB or 2MB page */
405 page_size = 1ul << shift;
406 if (shift == PMD_SHIFT)
407 kvm->stat.num_2M_pages--;
408 else if (shift == PUD_SHIFT)
409 kvm->stat.num_1G_pages--;
412 gpa &= ~(page_size - 1);
413 hpa = old & PTE_RPN_MASK;
414 kvmhv_remove_nest_rmap_range(kvm, memslot, gpa, hpa, page_size);
416 if ((old & _PAGE_DIRTY) && memslot->dirty_bitmap)
417 kvmppc_update_dirty_map(memslot, gfn, page_size);
421 * kvmppc_free_p?d are used to free existing page tables, and recursively
422 * descend and clear and free children.
423 * Callers are responsible for flushing the PWC.
425 * When page tables are being unmapped/freed as part of page fault path
426 * (full == false), ptes are not expected. There is code to unmap them
427 * and emit a warning if encountered, but there may already be data
428 * corruption due to the unexpected mappings.
430 static void kvmppc_unmap_free_pte(struct kvm *kvm, pte_t *pte, bool full,
434 memset(pte, 0, sizeof(long) << PTE_INDEX_SIZE);
439 for (it = 0; it < PTRS_PER_PTE; ++it, ++p) {
440 if (pte_val(*p) == 0)
443 kvmppc_unmap_pte(kvm, p,
444 pte_pfn(*p) << PAGE_SHIFT,
445 PAGE_SHIFT, NULL, lpid);
449 kvmppc_pte_free(pte);
452 static void kvmppc_unmap_free_pmd(struct kvm *kvm, pmd_t *pmd, bool full,
458 for (im = 0; im < PTRS_PER_PMD; ++im, ++p) {
459 if (!pmd_present(*p))
461 if (pmd_is_leaf(*p)) {
466 kvmppc_unmap_pte(kvm, (pte_t *)p,
467 pte_pfn(*(pte_t *)p) << PAGE_SHIFT,
468 PMD_SHIFT, NULL, lpid);
473 pte = pte_offset_map(p, 0);
474 kvmppc_unmap_free_pte(kvm, pte, full, lpid);
478 kvmppc_pmd_free(pmd);
481 static void kvmppc_unmap_free_pud(struct kvm *kvm, pud_t *pud,
487 for (iu = 0; iu < PTRS_PER_PUD; ++iu, ++p) {
488 if (!pud_present(*p))
495 pmd = pmd_offset(p, 0);
496 kvmppc_unmap_free_pmd(kvm, pmd, true, lpid);
500 pud_free(kvm->mm, pud);
503 void kvmppc_free_pgtable_radix(struct kvm *kvm, pgd_t *pgd, unsigned int lpid)
507 for (ig = 0; ig < PTRS_PER_PGD; ++ig, ++pgd) {
510 if (!pgd_present(*pgd))
512 pud = pud_offset(pgd, 0);
513 kvmppc_unmap_free_pud(kvm, pud, lpid);
518 void kvmppc_free_radix(struct kvm *kvm)
520 if (kvm->arch.pgtable) {
521 kvmppc_free_pgtable_radix(kvm, kvm->arch.pgtable,
523 pgd_free(kvm->mm, kvm->arch.pgtable);
524 kvm->arch.pgtable = NULL;
528 static void kvmppc_unmap_free_pmd_entry_table(struct kvm *kvm, pmd_t *pmd,
529 unsigned long gpa, unsigned int lpid)
531 pte_t *pte = pte_offset_kernel(pmd, 0);
534 * Clearing the pmd entry then flushing the PWC ensures that the pte
535 * page no longer be cached by the MMU, so can be freed without
536 * flushing the PWC again.
539 kvmppc_radix_flush_pwc(kvm, lpid);
541 kvmppc_unmap_free_pte(kvm, pte, false, lpid);
544 static void kvmppc_unmap_free_pud_entry_table(struct kvm *kvm, pud_t *pud,
545 unsigned long gpa, unsigned int lpid)
547 pmd_t *pmd = pmd_offset(pud, 0);
550 * Clearing the pud entry then flushing the PWC ensures that the pmd
551 * page and any children pte pages will no longer be cached by the MMU,
552 * so can be freed without flushing the PWC again.
555 kvmppc_radix_flush_pwc(kvm, lpid);
557 kvmppc_unmap_free_pmd(kvm, pmd, false, lpid);
561 * There are a number of bits which may differ between different faults to
562 * the same partition scope entry. RC bits, in the course of cleaning and
563 * aging. And the write bit can change, either the access could have been
564 * upgraded, or a read fault could happen concurrently with a write fault
565 * that sets those bits first.
567 #define PTE_BITS_MUST_MATCH (~(_PAGE_WRITE | _PAGE_DIRTY | _PAGE_ACCESSED))
569 int kvmppc_create_pte(struct kvm *kvm, pgd_t *pgtable, pte_t pte,
570 unsigned long gpa, unsigned int level,
571 unsigned long mmu_seq, unsigned int lpid,
572 unsigned long *rmapp, struct rmap_nested **n_rmap)
575 pud_t *pud, *new_pud = NULL;
576 pmd_t *pmd, *new_pmd = NULL;
577 pte_t *ptep, *new_ptep = NULL;
580 /* Traverse the guest's 2nd-level tree, allocate new levels needed */
581 pgd = pgtable + pgd_index(gpa);
583 if (pgd_present(*pgd))
584 pud = pud_offset(pgd, gpa);
586 new_pud = pud_alloc_one(kvm->mm, gpa);
589 if (pud && pud_present(*pud) && !pud_huge(*pud))
590 pmd = pmd_offset(pud, gpa);
592 new_pmd = kvmppc_pmd_alloc();
594 if (level == 0 && !(pmd && pmd_present(*pmd) && !pmd_is_leaf(*pmd)))
595 new_ptep = kvmppc_pte_alloc();
597 /* Check if we might have been invalidated; let the guest retry if so */
598 spin_lock(&kvm->mmu_lock);
600 if (mmu_notifier_retry(kvm, mmu_seq))
603 /* Now traverse again under the lock and change the tree */
605 if (pgd_none(*pgd)) {
608 pgd_populate(kvm->mm, pgd, new_pud);
611 pud = pud_offset(pgd, gpa);
612 if (pud_huge(*pud)) {
613 unsigned long hgpa = gpa & PUD_MASK;
615 /* Check if we raced and someone else has set the same thing */
617 if (pud_raw(*pud) == pte_raw(pte)) {
621 /* Valid 1GB page here already, add our extra bits */
622 WARN_ON_ONCE((pud_val(*pud) ^ pte_val(pte)) &
623 PTE_BITS_MUST_MATCH);
624 kvmppc_radix_update_pte(kvm, (pte_t *)pud,
625 0, pte_val(pte), hgpa, PUD_SHIFT);
630 * If we raced with another CPU which has just put
631 * a 1GB pte in after we saw a pmd page, try again.
637 /* Valid 1GB page here already, remove it */
638 kvmppc_unmap_pte(kvm, (pte_t *)pud, hgpa, PUD_SHIFT, NULL,
642 if (!pud_none(*pud)) {
644 * There's a page table page here, but we wanted to
645 * install a large page, so remove and free the page
648 kvmppc_unmap_free_pud_entry_table(kvm, pud, gpa, lpid);
650 kvmppc_radix_set_pte_at(kvm, gpa, (pte_t *)pud, pte);
652 kvmhv_insert_nest_rmap(kvm, rmapp, n_rmap);
656 if (pud_none(*pud)) {
659 pud_populate(kvm->mm, pud, new_pmd);
662 pmd = pmd_offset(pud, gpa);
663 if (pmd_is_leaf(*pmd)) {
664 unsigned long lgpa = gpa & PMD_MASK;
666 /* Check if we raced and someone else has set the same thing */
668 if (pmd_raw(*pmd) == pte_raw(pte)) {
672 /* Valid 2MB page here already, add our extra bits */
673 WARN_ON_ONCE((pmd_val(*pmd) ^ pte_val(pte)) &
674 PTE_BITS_MUST_MATCH);
675 kvmppc_radix_update_pte(kvm, pmdp_ptep(pmd),
676 0, pte_val(pte), lgpa, PMD_SHIFT);
682 * If we raced with another CPU which has just put
683 * a 2MB pte in after we saw a pte page, try again.
689 /* Valid 2MB page here already, remove it */
690 kvmppc_unmap_pte(kvm, pmdp_ptep(pmd), lgpa, PMD_SHIFT, NULL,
694 if (!pmd_none(*pmd)) {
696 * There's a page table page here, but we wanted to
697 * install a large page, so remove and free the page
700 kvmppc_unmap_free_pmd_entry_table(kvm, pmd, gpa, lpid);
702 kvmppc_radix_set_pte_at(kvm, gpa, pmdp_ptep(pmd), pte);
704 kvmhv_insert_nest_rmap(kvm, rmapp, n_rmap);
708 if (pmd_none(*pmd)) {
711 pmd_populate(kvm->mm, pmd, new_ptep);
714 ptep = pte_offset_kernel(pmd, gpa);
715 if (pte_present(*ptep)) {
716 /* Check if someone else set the same thing */
717 if (pte_raw(*ptep) == pte_raw(pte)) {
721 /* Valid page here already, add our extra bits */
722 WARN_ON_ONCE((pte_val(*ptep) ^ pte_val(pte)) &
723 PTE_BITS_MUST_MATCH);
724 kvmppc_radix_update_pte(kvm, ptep, 0, pte_val(pte), gpa, 0);
728 kvmppc_radix_set_pte_at(kvm, gpa, ptep, pte);
730 kvmhv_insert_nest_rmap(kvm, rmapp, n_rmap);
734 spin_unlock(&kvm->mmu_lock);
736 pud_free(kvm->mm, new_pud);
738 kvmppc_pmd_free(new_pmd);
740 kvmppc_pte_free(new_ptep);
744 bool kvmppc_hv_handle_set_rc(struct kvm *kvm, pgd_t *pgtable, bool writing,
745 unsigned long gpa, unsigned int lpid)
747 unsigned long pgflags;
752 * Need to set an R or C bit in the 2nd-level tables;
753 * since we are just helping out the hardware here,
754 * it is sufficient to do what the hardware does.
756 pgflags = _PAGE_ACCESSED;
758 pgflags |= _PAGE_DIRTY;
760 * We are walking the secondary (partition-scoped) page table here.
761 * We can do this without disabling irq because the Linux MM
762 * subsystem doesn't do THP splits and collapses on this tree.
764 ptep = __find_linux_pte(pgtable, gpa, NULL, &shift);
765 if (ptep && pte_present(*ptep) && (!writing || pte_write(*ptep))) {
766 kvmppc_radix_update_pte(kvm, ptep, 0, pgflags, gpa, shift);
772 int kvmppc_book3s_instantiate_page(struct kvm_vcpu *vcpu,
774 struct kvm_memory_slot *memslot,
775 bool writing, bool kvm_ro,
776 pte_t *inserted_pte, unsigned int *levelp)
778 struct kvm *kvm = vcpu->kvm;
779 struct page *page = NULL;
780 unsigned long mmu_seq;
781 unsigned long hva, gfn = gpa >> PAGE_SHIFT;
782 bool upgrade_write = false;
783 bool *upgrade_p = &upgrade_write;
785 unsigned int shift, level;
789 /* used to check for invalidations in progress */
790 mmu_seq = kvm->mmu_notifier_seq;
794 * Do a fast check first, since __gfn_to_pfn_memslot doesn't
795 * do it with !atomic && !async, which is how we call it.
796 * We always ask for write permission since the common case
797 * is that the page is writable.
799 hva = gfn_to_hva_memslot(memslot, gfn);
800 if (!kvm_ro && __get_user_pages_fast(hva, 1, 1, &page) == 1) {
801 upgrade_write = true;
805 /* Call KVM generic code to do the slow-path check */
806 pfn = __gfn_to_pfn_memslot(memslot, gfn, false, NULL,
808 if (is_error_noslot_pfn(pfn))
811 if (pfn_valid(pfn)) {
812 page = pfn_to_page(pfn);
813 if (PageReserved(page))
819 * Read the PTE from the process' radix tree and use that
820 * so we get the shift and attribute bits.
823 ptep = __find_linux_pte(vcpu->arch.pgdir, hva, NULL, &shift);
825 * If the PTE disappeared temporarily due to a THP
826 * collapse, just return and let the guest try again.
837 /* If we're logging dirty pages, always map single pages */
838 large_enable = !(memslot->flags & KVM_MEM_LOG_DIRTY_PAGES);
840 /* Get pte level from shift/size */
841 if (large_enable && shift == PUD_SHIFT &&
842 (gpa & (PUD_SIZE - PAGE_SIZE)) ==
843 (hva & (PUD_SIZE - PAGE_SIZE))) {
845 } else if (large_enable && shift == PMD_SHIFT &&
846 (gpa & (PMD_SIZE - PAGE_SIZE)) ==
847 (hva & (PMD_SIZE - PAGE_SIZE))) {
851 if (shift > PAGE_SHIFT) {
853 * If the pte maps more than one page, bring over
854 * bits from the virtual address to get the real
855 * address of the specific single page we want.
857 unsigned long rpnmask = (1ul << shift) - PAGE_SIZE;
858 pte = __pte(pte_val(pte) | (hva & rpnmask));
862 pte = __pte(pte_val(pte) | _PAGE_EXEC | _PAGE_ACCESSED);
863 if (writing || upgrade_write) {
864 if (pte_val(pte) & _PAGE_WRITE)
865 pte = __pte(pte_val(pte) | _PAGE_DIRTY);
867 pte = __pte(pte_val(pte) & ~(_PAGE_WRITE | _PAGE_DIRTY));
870 /* Allocate space in the tree and write the PTE */
871 ret = kvmppc_create_pte(kvm, kvm->arch.pgtable, pte, gpa, level,
872 mmu_seq, kvm->arch.lpid, NULL, NULL);
879 if (!ret && (pte_val(pte) & _PAGE_WRITE))
880 set_page_dirty_lock(page);
884 /* Increment number of large pages if we (successfully) inserted one */
887 kvm->stat.num_2M_pages++;
889 kvm->stat.num_1G_pages++;
895 int kvmppc_book3s_radix_page_fault(struct kvm_run *run, struct kvm_vcpu *vcpu,
896 unsigned long ea, unsigned long dsisr)
898 struct kvm *kvm = vcpu->kvm;
899 unsigned long gpa, gfn;
900 struct kvm_memory_slot *memslot;
902 bool writing = !!(dsisr & DSISR_ISSTORE);
905 /* Check for unusual errors */
906 if (dsisr & DSISR_UNSUPP_MMU) {
907 pr_err("KVM: Got unsupported MMU fault\n");
910 if (dsisr & DSISR_BADACCESS) {
911 /* Reflect to the guest as DSI */
912 pr_err("KVM: Got radix HV page fault with DSISR=%lx\n", dsisr);
913 kvmppc_core_queue_data_storage(vcpu, ea, dsisr);
917 /* Translate the logical address */
918 gpa = vcpu->arch.fault_gpa & ~0xfffUL;
919 gpa &= ~0xF000000000000000ul;
920 gfn = gpa >> PAGE_SHIFT;
921 if (!(dsisr & DSISR_PRTABLE_FAULT))
924 /* Get the corresponding memslot */
925 memslot = gfn_to_memslot(kvm, gfn);
927 /* No memslot means it's an emulated MMIO region */
928 if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID)) {
929 if (dsisr & (DSISR_PRTABLE_FAULT | DSISR_BADACCESS |
932 * Bad address in guest page table tree, or other
933 * unusual error - reflect it to the guest as DSI.
935 kvmppc_core_queue_data_storage(vcpu, ea, dsisr);
938 return kvmppc_hv_emulate_mmio(run, vcpu, gpa, ea, writing);
941 if (memslot->flags & KVM_MEM_READONLY) {
943 /* give the guest a DSI */
944 kvmppc_core_queue_data_storage(vcpu, ea, DSISR_ISSTORE |
951 /* Failed to set the reference/change bits */
952 if (dsisr & DSISR_SET_RC) {
953 spin_lock(&kvm->mmu_lock);
954 if (kvmppc_hv_handle_set_rc(kvm, kvm->arch.pgtable,
955 writing, gpa, kvm->arch.lpid))
956 dsisr &= ~DSISR_SET_RC;
957 spin_unlock(&kvm->mmu_lock);
959 if (!(dsisr & (DSISR_BAD_FAULT_64S | DSISR_NOHPTE |
960 DSISR_PROTFAULT | DSISR_SET_RC)))
964 /* Try to insert a pte */
965 ret = kvmppc_book3s_instantiate_page(vcpu, gpa, memslot, writing,
968 if (ret == 0 || ret == -EAGAIN)
973 /* Called with kvm->mmu_lock held */
974 int kvm_unmap_radix(struct kvm *kvm, struct kvm_memory_slot *memslot,
978 unsigned long gpa = gfn << PAGE_SHIFT;
981 ptep = __find_linux_pte(kvm->arch.pgtable, gpa, NULL, &shift);
982 if (ptep && pte_present(*ptep))
983 kvmppc_unmap_pte(kvm, ptep, gpa, shift, memslot,
988 /* Called with kvm->mmu_lock held */
989 int kvm_age_radix(struct kvm *kvm, struct kvm_memory_slot *memslot,
993 unsigned long gpa = gfn << PAGE_SHIFT;
996 unsigned long old, *rmapp;
998 ptep = __find_linux_pte(kvm->arch.pgtable, gpa, NULL, &shift);
999 if (ptep && pte_present(*ptep) && pte_young(*ptep)) {
1000 old = kvmppc_radix_update_pte(kvm, ptep, _PAGE_ACCESSED, 0,
1002 /* XXX need to flush tlb here? */
1003 /* Also clear bit in ptes in shadow pgtable for nested guests */
1004 rmapp = &memslot->arch.rmap[gfn - memslot->base_gfn];
1005 kvmhv_update_nest_rmap_rc_list(kvm, rmapp, _PAGE_ACCESSED, 0,
1013 /* Called with kvm->mmu_lock held */
1014 int kvm_test_age_radix(struct kvm *kvm, struct kvm_memory_slot *memslot,
1018 unsigned long gpa = gfn << PAGE_SHIFT;
1022 ptep = __find_linux_pte(kvm->arch.pgtable, gpa, NULL, &shift);
1023 if (ptep && pte_present(*ptep) && pte_young(*ptep))
1028 /* Returns the number of PAGE_SIZE pages that are dirty */
1029 static int kvm_radix_test_clear_dirty(struct kvm *kvm,
1030 struct kvm_memory_slot *memslot, int pagenum)
1032 unsigned long gfn = memslot->base_gfn + pagenum;
1033 unsigned long gpa = gfn << PAGE_SHIFT;
1037 unsigned long old, *rmapp;
1039 ptep = __find_linux_pte(kvm->arch.pgtable, gpa, NULL, &shift);
1040 if (ptep && pte_present(*ptep) && pte_dirty(*ptep)) {
1043 ret = 1 << (shift - PAGE_SHIFT);
1044 spin_lock(&kvm->mmu_lock);
1045 old = kvmppc_radix_update_pte(kvm, ptep, _PAGE_DIRTY, 0,
1047 kvmppc_radix_tlbie_page(kvm, gpa, shift, kvm->arch.lpid);
1048 /* Also clear bit in ptes in shadow pgtable for nested guests */
1049 rmapp = &memslot->arch.rmap[gfn - memslot->base_gfn];
1050 kvmhv_update_nest_rmap_rc_list(kvm, rmapp, _PAGE_DIRTY, 0,
1053 spin_unlock(&kvm->mmu_lock);
1058 long kvmppc_hv_get_dirty_log_radix(struct kvm *kvm,
1059 struct kvm_memory_slot *memslot, unsigned long *map)
1064 for (i = 0; i < memslot->npages; i = j) {
1065 npages = kvm_radix_test_clear_dirty(kvm, memslot, i);
1068 * Note that if npages > 0 then i must be a multiple of npages,
1069 * since huge pages are only used to back the guest at guest
1070 * real addresses that are a multiple of their size.
1071 * Since we have at most one PTE covering any given guest
1072 * real address, if npages > 1 we can skip to i + npages.
1076 set_dirty_bits(map, i, npages);
1083 void kvmppc_radix_flush_memslot(struct kvm *kvm,
1084 const struct kvm_memory_slot *memslot)
1091 gpa = memslot->base_gfn << PAGE_SHIFT;
1092 spin_lock(&kvm->mmu_lock);
1093 for (n = memslot->npages; n; --n) {
1094 ptep = __find_linux_pte(kvm->arch.pgtable, gpa, NULL, &shift);
1095 if (ptep && pte_present(*ptep))
1096 kvmppc_unmap_pte(kvm, ptep, gpa, shift, memslot,
1100 spin_unlock(&kvm->mmu_lock);
1103 static void add_rmmu_ap_encoding(struct kvm_ppc_rmmu_info *info,
1104 int psize, int *indexp)
1106 if (!mmu_psize_defs[psize].shift)
1108 info->ap_encodings[*indexp] = mmu_psize_defs[psize].shift |
1109 (mmu_psize_defs[psize].ap << 29);
1113 int kvmhv_get_rmmu_info(struct kvm *kvm, struct kvm_ppc_rmmu_info *info)
1117 if (!radix_enabled())
1119 memset(info, 0, sizeof(*info));
1122 info->geometries[0].page_shift = 12;
1123 info->geometries[0].level_bits[0] = 9;
1124 for (i = 1; i < 4; ++i)
1125 info->geometries[0].level_bits[i] = p9_supported_radix_bits[i];
1127 info->geometries[1].page_shift = 16;
1128 for (i = 0; i < 4; ++i)
1129 info->geometries[1].level_bits[i] = p9_supported_radix_bits[i];
1132 add_rmmu_ap_encoding(info, MMU_PAGE_4K, &i);
1133 add_rmmu_ap_encoding(info, MMU_PAGE_64K, &i);
1134 add_rmmu_ap_encoding(info, MMU_PAGE_2M, &i);
1135 add_rmmu_ap_encoding(info, MMU_PAGE_1G, &i);
1140 int kvmppc_init_vm_radix(struct kvm *kvm)
1142 kvm->arch.pgtable = pgd_alloc(kvm->mm);
1143 if (!kvm->arch.pgtable)
1148 static void pte_ctor(void *addr)
1150 memset(addr, 0, RADIX_PTE_TABLE_SIZE);
1153 static void pmd_ctor(void *addr)
1155 memset(addr, 0, RADIX_PMD_TABLE_SIZE);
1158 struct debugfs_radix_state {
1169 static int debugfs_radix_open(struct inode *inode, struct file *file)
1171 struct kvm *kvm = inode->i_private;
1172 struct debugfs_radix_state *p;
1174 p = kzalloc(sizeof(*p), GFP_KERNEL);
1180 mutex_init(&p->mutex);
1181 file->private_data = p;
1183 return nonseekable_open(inode, file);
1186 static int debugfs_radix_release(struct inode *inode, struct file *file)
1188 struct debugfs_radix_state *p = file->private_data;
1190 kvm_put_kvm(p->kvm);
1195 static ssize_t debugfs_radix_read(struct file *file, char __user *buf,
1196 size_t len, loff_t *ppos)
1198 struct debugfs_radix_state *p = file->private_data;
1204 struct kvm_nested_guest *nested;
1213 if (!kvm_is_radix(kvm))
1216 ret = mutex_lock_interruptible(&p->mutex);
1220 if (p->chars_left) {
1224 r = copy_to_user(buf, p->buf + p->buf_index, n);
1241 while (len != 0 && p->lpid >= 0) {
1242 if (gpa >= RADIX_PGTABLE_RANGE) {
1246 kvmhv_put_nested(nested);
1249 p->lpid = kvmhv_nested_next_lpid(kvm, p->lpid);
1256 pgt = kvm->arch.pgtable;
1258 nested = kvmhv_get_nested(kvm, p->lpid, false);
1260 gpa = RADIX_PGTABLE_RANGE;
1263 pgt = nested->shadow_pgtable;
1269 n = scnprintf(p->buf, sizeof(p->buf),
1270 "\nNested LPID %d: ", p->lpid);
1271 n += scnprintf(p->buf + n, sizeof(p->buf) - n,
1272 "pgdir: %lx\n", (unsigned long)pgt);
1277 pgdp = pgt + pgd_index(gpa);
1278 pgd = READ_ONCE(*pgdp);
1279 if (!(pgd_val(pgd) & _PAGE_PRESENT)) {
1280 gpa = (gpa & PGDIR_MASK) + PGDIR_SIZE;
1284 pudp = pud_offset(&pgd, gpa);
1285 pud = READ_ONCE(*pudp);
1286 if (!(pud_val(pud) & _PAGE_PRESENT)) {
1287 gpa = (gpa & PUD_MASK) + PUD_SIZE;
1290 if (pud_val(pud) & _PAGE_PTE) {
1296 pmdp = pmd_offset(&pud, gpa);
1297 pmd = READ_ONCE(*pmdp);
1298 if (!(pmd_val(pmd) & _PAGE_PRESENT)) {
1299 gpa = (gpa & PMD_MASK) + PMD_SIZE;
1302 if (pmd_val(pmd) & _PAGE_PTE) {
1308 ptep = pte_offset_kernel(&pmd, gpa);
1309 pte = pte_val(READ_ONCE(*ptep));
1310 if (!(pte & _PAGE_PRESENT)) {
1316 n = scnprintf(p->buf, sizeof(p->buf),
1317 " %lx: %lx %d\n", gpa, pte, shift);
1318 gpa += 1ul << shift;
1323 r = copy_to_user(buf, p->buf, n);
1338 kvmhv_put_nested(nested);
1341 mutex_unlock(&p->mutex);
1345 static ssize_t debugfs_radix_write(struct file *file, const char __user *buf,
1346 size_t len, loff_t *ppos)
1351 static const struct file_operations debugfs_radix_fops = {
1352 .owner = THIS_MODULE,
1353 .open = debugfs_radix_open,
1354 .release = debugfs_radix_release,
1355 .read = debugfs_radix_read,
1356 .write = debugfs_radix_write,
1357 .llseek = generic_file_llseek,
1360 void kvmhv_radix_debugfs_init(struct kvm *kvm)
1362 kvm->arch.radix_dentry = debugfs_create_file("radix", 0400,
1363 kvm->arch.debugfs_dir, kvm,
1364 &debugfs_radix_fops);
1367 int kvmppc_radix_init(void)
1369 unsigned long size = sizeof(void *) << RADIX_PTE_INDEX_SIZE;
1371 kvm_pte_cache = kmem_cache_create("kvm-pte", size, size, 0, pte_ctor);
1375 size = sizeof(void *) << RADIX_PMD_INDEX_SIZE;
1377 kvm_pmd_cache = kmem_cache_create("kvm-pmd", size, size, 0, pmd_ctor);
1378 if (!kvm_pmd_cache) {
1379 kmem_cache_destroy(kvm_pte_cache);
1386 void kvmppc_radix_exit(void)
1388 kmem_cache_destroy(kvm_pte_cache);
1389 kmem_cache_destroy(kvm_pmd_cache);