2 * Copyright (C) 2012 - Virtual Open Systems and Columbia University
3 * Author: Christoffer Dall <c.dall@virtualopensystems.com>
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License, version 2, as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write to the Free Software
16 * Foundation, 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
19 #include <linux/mman.h>
20 #include <linux/kvm_host.h>
22 #include <linux/hugetlb.h>
23 #include <linux/sched/signal.h>
24 #include <trace/events/kvm.h>
25 #include <asm/pgalloc.h>
26 #include <asm/cacheflush.h>
27 #include <asm/kvm_arm.h>
28 #include <asm/kvm_mmu.h>
29 #include <asm/kvm_mmio.h>
30 #include <asm/kvm_asm.h>
31 #include <asm/kvm_emulate.h>
33 #include <asm/system_misc.h>
37 static pgd_t *boot_hyp_pgd;
38 static pgd_t *hyp_pgd;
39 static pgd_t *merged_hyp_pgd;
40 static DEFINE_MUTEX(kvm_hyp_pgd_mutex);
42 static unsigned long hyp_idmap_start;
43 static unsigned long hyp_idmap_end;
44 static phys_addr_t hyp_idmap_vector;
46 static unsigned long io_map_base;
48 #define S2_PGD_SIZE (PTRS_PER_S2_PGD * sizeof(pgd_t))
49 #define hyp_pgd_order get_order(PTRS_PER_PGD * sizeof(pgd_t))
51 #define KVM_S2PTE_FLAG_IS_IOMAP (1UL << 0)
52 #define KVM_S2_FLAG_LOGGING_ACTIVE (1UL << 1)
54 static bool memslot_is_logging(struct kvm_memory_slot *memslot)
56 return memslot->dirty_bitmap && !(memslot->flags & KVM_MEM_READONLY);
60 * kvm_flush_remote_tlbs() - flush all VM TLB entries for v7/8
61 * @kvm: pointer to kvm structure.
63 * Interface to HYP function to flush all VM TLB entries
65 void kvm_flush_remote_tlbs(struct kvm *kvm)
67 kvm_call_hyp(__kvm_tlb_flush_vmid, kvm);
70 static void kvm_tlb_flush_vmid_ipa(struct kvm *kvm, phys_addr_t ipa)
72 kvm_call_hyp(__kvm_tlb_flush_vmid_ipa, kvm, ipa);
76 * D-Cache management functions. They take the page table entries by
77 * value, as they are flushing the cache using the kernel mapping (or
80 static void kvm_flush_dcache_pte(pte_t pte)
82 __kvm_flush_dcache_pte(pte);
85 static void kvm_flush_dcache_pmd(pmd_t pmd)
87 __kvm_flush_dcache_pmd(pmd);
90 static void kvm_flush_dcache_pud(pud_t pud)
92 __kvm_flush_dcache_pud(pud);
95 static bool kvm_is_device_pfn(unsigned long pfn)
97 return !pfn_valid(pfn);
101 * stage2_dissolve_pmd() - clear and flush huge PMD entry
102 * @kvm: pointer to kvm structure.
104 * @pmd: pmd pointer for IPA
106 * Function clears a PMD entry, flushes addr 1st and 2nd stage TLBs. Marks all
107 * pages in the range dirty.
109 static void stage2_dissolve_pmd(struct kvm *kvm, phys_addr_t addr, pmd_t *pmd)
111 if (!pmd_thp_or_huge(*pmd))
115 kvm_tlb_flush_vmid_ipa(kvm, addr);
116 put_page(virt_to_page(pmd));
119 static int mmu_topup_memory_cache(struct kvm_mmu_memory_cache *cache,
124 BUG_ON(max > KVM_NR_MEM_OBJS);
125 if (cache->nobjs >= min)
127 while (cache->nobjs < max) {
128 page = (void *)__get_free_page(PGALLOC_GFP);
131 cache->objects[cache->nobjs++] = page;
136 static void mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc)
139 free_page((unsigned long)mc->objects[--mc->nobjs]);
142 static void *mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc)
146 BUG_ON(!mc || !mc->nobjs);
147 p = mc->objects[--mc->nobjs];
151 static void clear_stage2_pgd_entry(struct kvm *kvm, pgd_t *pgd, phys_addr_t addr)
153 pud_t *pud_table __maybe_unused = stage2_pud_offset(pgd, 0UL);
154 stage2_pgd_clear(pgd);
155 kvm_tlb_flush_vmid_ipa(kvm, addr);
156 stage2_pud_free(pud_table);
157 put_page(virt_to_page(pgd));
160 static void clear_stage2_pud_entry(struct kvm *kvm, pud_t *pud, phys_addr_t addr)
162 pmd_t *pmd_table __maybe_unused = stage2_pmd_offset(pud, 0);
163 VM_BUG_ON(stage2_pud_huge(*pud));
164 stage2_pud_clear(pud);
165 kvm_tlb_flush_vmid_ipa(kvm, addr);
166 stage2_pmd_free(pmd_table);
167 put_page(virt_to_page(pud));
170 static void clear_stage2_pmd_entry(struct kvm *kvm, pmd_t *pmd, phys_addr_t addr)
172 pte_t *pte_table = pte_offset_kernel(pmd, 0);
173 VM_BUG_ON(pmd_thp_or_huge(*pmd));
175 kvm_tlb_flush_vmid_ipa(kvm, addr);
176 pte_free_kernel(NULL, pte_table);
177 put_page(virt_to_page(pmd));
181 * Unmapping vs dcache management:
183 * If a guest maps certain memory pages as uncached, all writes will
184 * bypass the data cache and go directly to RAM. However, the CPUs
185 * can still speculate reads (not writes) and fill cache lines with
188 * Those cache lines will be *clean* cache lines though, so a
189 * clean+invalidate operation is equivalent to an invalidate
190 * operation, because no cache lines are marked dirty.
192 * Those clean cache lines could be filled prior to an uncached write
193 * by the guest, and the cache coherent IO subsystem would therefore
194 * end up writing old data to disk.
196 * This is why right after unmapping a page/section and invalidating
197 * the corresponding TLBs, we call kvm_flush_dcache_p*() to make sure
198 * the IO subsystem will never hit in the cache.
200 static void unmap_stage2_ptes(struct kvm *kvm, pmd_t *pmd,
201 phys_addr_t addr, phys_addr_t end)
203 phys_addr_t start_addr = addr;
204 pte_t *pte, *start_pte;
206 start_pte = pte = pte_offset_kernel(pmd, addr);
208 if (!pte_none(*pte)) {
209 pte_t old_pte = *pte;
211 kvm_set_pte(pte, __pte(0));
212 kvm_tlb_flush_vmid_ipa(kvm, addr);
214 /* No need to invalidate the cache for device mappings */
215 if (!kvm_is_device_pfn(pte_pfn(old_pte)))
216 kvm_flush_dcache_pte(old_pte);
218 put_page(virt_to_page(pte));
220 } while (pte++, addr += PAGE_SIZE, addr != end);
222 if (stage2_pte_table_empty(start_pte))
223 clear_stage2_pmd_entry(kvm, pmd, start_addr);
226 static void unmap_stage2_pmds(struct kvm *kvm, pud_t *pud,
227 phys_addr_t addr, phys_addr_t end)
229 phys_addr_t next, start_addr = addr;
230 pmd_t *pmd, *start_pmd;
232 start_pmd = pmd = stage2_pmd_offset(pud, addr);
234 next = stage2_pmd_addr_end(addr, end);
235 if (!pmd_none(*pmd)) {
236 if (pmd_thp_or_huge(*pmd)) {
237 pmd_t old_pmd = *pmd;
240 kvm_tlb_flush_vmid_ipa(kvm, addr);
242 kvm_flush_dcache_pmd(old_pmd);
244 put_page(virt_to_page(pmd));
246 unmap_stage2_ptes(kvm, pmd, addr, next);
249 } while (pmd++, addr = next, addr != end);
251 if (stage2_pmd_table_empty(start_pmd))
252 clear_stage2_pud_entry(kvm, pud, start_addr);
255 static void unmap_stage2_puds(struct kvm *kvm, pgd_t *pgd,
256 phys_addr_t addr, phys_addr_t end)
258 phys_addr_t next, start_addr = addr;
259 pud_t *pud, *start_pud;
261 start_pud = pud = stage2_pud_offset(pgd, addr);
263 next = stage2_pud_addr_end(addr, end);
264 if (!stage2_pud_none(*pud)) {
265 if (stage2_pud_huge(*pud)) {
266 pud_t old_pud = *pud;
268 stage2_pud_clear(pud);
269 kvm_tlb_flush_vmid_ipa(kvm, addr);
270 kvm_flush_dcache_pud(old_pud);
271 put_page(virt_to_page(pud));
273 unmap_stage2_pmds(kvm, pud, addr, next);
276 } while (pud++, addr = next, addr != end);
278 if (stage2_pud_table_empty(start_pud))
279 clear_stage2_pgd_entry(kvm, pgd, start_addr);
283 * unmap_stage2_range -- Clear stage2 page table entries to unmap a range
284 * @kvm: The VM pointer
285 * @start: The intermediate physical base address of the range to unmap
286 * @size: The size of the area to unmap
288 * Clear a range of stage-2 mappings, lowering the various ref-counts. Must
289 * be called while holding mmu_lock (unless for freeing the stage2 pgd before
290 * destroying the VM), otherwise another faulting VCPU may come in and mess
291 * with things behind our backs.
293 static void unmap_stage2_range(struct kvm *kvm, phys_addr_t start, u64 size)
296 phys_addr_t addr = start, end = start + size;
299 assert_spin_locked(&kvm->mmu_lock);
300 WARN_ON(size & ~PAGE_MASK);
302 pgd = kvm->arch.pgd + stage2_pgd_index(addr);
305 * Make sure the page table is still active, as another thread
306 * could have possibly freed the page table, while we released
309 if (!READ_ONCE(kvm->arch.pgd))
311 next = stage2_pgd_addr_end(addr, end);
312 if (!stage2_pgd_none(*pgd))
313 unmap_stage2_puds(kvm, pgd, addr, next);
315 * If the range is too large, release the kvm->mmu_lock
316 * to prevent starvation and lockup detector warnings.
319 cond_resched_lock(&kvm->mmu_lock);
320 } while (pgd++, addr = next, addr != end);
323 static void stage2_flush_ptes(struct kvm *kvm, pmd_t *pmd,
324 phys_addr_t addr, phys_addr_t end)
328 pte = pte_offset_kernel(pmd, addr);
330 if (!pte_none(*pte) && !kvm_is_device_pfn(pte_pfn(*pte)))
331 kvm_flush_dcache_pte(*pte);
332 } while (pte++, addr += PAGE_SIZE, addr != end);
335 static void stage2_flush_pmds(struct kvm *kvm, pud_t *pud,
336 phys_addr_t addr, phys_addr_t end)
341 pmd = stage2_pmd_offset(pud, addr);
343 next = stage2_pmd_addr_end(addr, end);
344 if (!pmd_none(*pmd)) {
345 if (pmd_thp_or_huge(*pmd))
346 kvm_flush_dcache_pmd(*pmd);
348 stage2_flush_ptes(kvm, pmd, addr, next);
350 } while (pmd++, addr = next, addr != end);
353 static void stage2_flush_puds(struct kvm *kvm, pgd_t *pgd,
354 phys_addr_t addr, phys_addr_t end)
359 pud = stage2_pud_offset(pgd, addr);
361 next = stage2_pud_addr_end(addr, end);
362 if (!stage2_pud_none(*pud)) {
363 if (stage2_pud_huge(*pud))
364 kvm_flush_dcache_pud(*pud);
366 stage2_flush_pmds(kvm, pud, addr, next);
368 } while (pud++, addr = next, addr != end);
371 static void stage2_flush_memslot(struct kvm *kvm,
372 struct kvm_memory_slot *memslot)
374 phys_addr_t addr = memslot->base_gfn << PAGE_SHIFT;
375 phys_addr_t end = addr + PAGE_SIZE * memslot->npages;
379 pgd = kvm->arch.pgd + stage2_pgd_index(addr);
381 next = stage2_pgd_addr_end(addr, end);
382 stage2_flush_puds(kvm, pgd, addr, next);
383 } while (pgd++, addr = next, addr != end);
387 * stage2_flush_vm - Invalidate cache for pages mapped in stage 2
388 * @kvm: The struct kvm pointer
390 * Go through the stage 2 page tables and invalidate any cache lines
391 * backing memory already mapped to the VM.
393 static void stage2_flush_vm(struct kvm *kvm)
395 struct kvm_memslots *slots;
396 struct kvm_memory_slot *memslot;
399 idx = srcu_read_lock(&kvm->srcu);
400 spin_lock(&kvm->mmu_lock);
402 slots = kvm_memslots(kvm);
403 kvm_for_each_memslot(memslot, slots)
404 stage2_flush_memslot(kvm, memslot);
406 spin_unlock(&kvm->mmu_lock);
407 srcu_read_unlock(&kvm->srcu, idx);
410 static void clear_hyp_pgd_entry(pgd_t *pgd)
412 pud_t *pud_table __maybe_unused = pud_offset(pgd, 0UL);
414 pud_free(NULL, pud_table);
415 put_page(virt_to_page(pgd));
418 static void clear_hyp_pud_entry(pud_t *pud)
420 pmd_t *pmd_table __maybe_unused = pmd_offset(pud, 0);
421 VM_BUG_ON(pud_huge(*pud));
423 pmd_free(NULL, pmd_table);
424 put_page(virt_to_page(pud));
427 static void clear_hyp_pmd_entry(pmd_t *pmd)
429 pte_t *pte_table = pte_offset_kernel(pmd, 0);
430 VM_BUG_ON(pmd_thp_or_huge(*pmd));
432 pte_free_kernel(NULL, pte_table);
433 put_page(virt_to_page(pmd));
436 static void unmap_hyp_ptes(pmd_t *pmd, phys_addr_t addr, phys_addr_t end)
438 pte_t *pte, *start_pte;
440 start_pte = pte = pte_offset_kernel(pmd, addr);
442 if (!pte_none(*pte)) {
443 kvm_set_pte(pte, __pte(0));
444 put_page(virt_to_page(pte));
446 } while (pte++, addr += PAGE_SIZE, addr != end);
448 if (hyp_pte_table_empty(start_pte))
449 clear_hyp_pmd_entry(pmd);
452 static void unmap_hyp_pmds(pud_t *pud, phys_addr_t addr, phys_addr_t end)
455 pmd_t *pmd, *start_pmd;
457 start_pmd = pmd = pmd_offset(pud, addr);
459 next = pmd_addr_end(addr, end);
460 /* Hyp doesn't use huge pmds */
462 unmap_hyp_ptes(pmd, addr, next);
463 } while (pmd++, addr = next, addr != end);
465 if (hyp_pmd_table_empty(start_pmd))
466 clear_hyp_pud_entry(pud);
469 static void unmap_hyp_puds(pgd_t *pgd, phys_addr_t addr, phys_addr_t end)
472 pud_t *pud, *start_pud;
474 start_pud = pud = pud_offset(pgd, addr);
476 next = pud_addr_end(addr, end);
477 /* Hyp doesn't use huge puds */
479 unmap_hyp_pmds(pud, addr, next);
480 } while (pud++, addr = next, addr != end);
482 if (hyp_pud_table_empty(start_pud))
483 clear_hyp_pgd_entry(pgd);
486 static unsigned int kvm_pgd_index(unsigned long addr, unsigned int ptrs_per_pgd)
488 return (addr >> PGDIR_SHIFT) & (ptrs_per_pgd - 1);
491 static void __unmap_hyp_range(pgd_t *pgdp, unsigned long ptrs_per_pgd,
492 phys_addr_t start, u64 size)
495 phys_addr_t addr = start, end = start + size;
499 * We don't unmap anything from HYP, except at the hyp tear down.
500 * Hence, we don't have to invalidate the TLBs here.
502 pgd = pgdp + kvm_pgd_index(addr, ptrs_per_pgd);
504 next = pgd_addr_end(addr, end);
506 unmap_hyp_puds(pgd, addr, next);
507 } while (pgd++, addr = next, addr != end);
510 static void unmap_hyp_range(pgd_t *pgdp, phys_addr_t start, u64 size)
512 __unmap_hyp_range(pgdp, PTRS_PER_PGD, start, size);
515 static void unmap_hyp_idmap_range(pgd_t *pgdp, phys_addr_t start, u64 size)
517 __unmap_hyp_range(pgdp, __kvm_idmap_ptrs_per_pgd(), start, size);
521 * free_hyp_pgds - free Hyp-mode page tables
523 * Assumes hyp_pgd is a page table used strictly in Hyp-mode and
524 * therefore contains either mappings in the kernel memory area (above
525 * PAGE_OFFSET), or device mappings in the idmap range.
527 * boot_hyp_pgd should only map the idmap range, and is only used in
528 * the extended idmap case.
530 void free_hyp_pgds(void)
534 mutex_lock(&kvm_hyp_pgd_mutex);
536 id_pgd = boot_hyp_pgd ? boot_hyp_pgd : hyp_pgd;
539 /* In case we never called hyp_mmu_init() */
541 io_map_base = hyp_idmap_start;
542 unmap_hyp_idmap_range(id_pgd, io_map_base,
543 hyp_idmap_start + PAGE_SIZE - io_map_base);
547 free_pages((unsigned long)boot_hyp_pgd, hyp_pgd_order);
552 unmap_hyp_range(hyp_pgd, kern_hyp_va(PAGE_OFFSET),
553 (uintptr_t)high_memory - PAGE_OFFSET);
555 free_pages((unsigned long)hyp_pgd, hyp_pgd_order);
558 if (merged_hyp_pgd) {
559 clear_page(merged_hyp_pgd);
560 free_page((unsigned long)merged_hyp_pgd);
561 merged_hyp_pgd = NULL;
564 mutex_unlock(&kvm_hyp_pgd_mutex);
567 static void create_hyp_pte_mappings(pmd_t *pmd, unsigned long start,
568 unsigned long end, unsigned long pfn,
576 pte = pte_offset_kernel(pmd, addr);
577 kvm_set_pte(pte, pfn_pte(pfn, prot));
578 get_page(virt_to_page(pte));
579 kvm_flush_dcache_to_poc(pte, sizeof(*pte));
581 } while (addr += PAGE_SIZE, addr != end);
584 static int create_hyp_pmd_mappings(pud_t *pud, unsigned long start,
585 unsigned long end, unsigned long pfn,
590 unsigned long addr, next;
594 pmd = pmd_offset(pud, addr);
596 BUG_ON(pmd_sect(*pmd));
598 if (pmd_none(*pmd)) {
599 pte = pte_alloc_one_kernel(NULL, addr);
601 kvm_err("Cannot allocate Hyp pte\n");
604 pmd_populate_kernel(NULL, pmd, pte);
605 get_page(virt_to_page(pmd));
606 kvm_flush_dcache_to_poc(pmd, sizeof(*pmd));
609 next = pmd_addr_end(addr, end);
611 create_hyp_pte_mappings(pmd, addr, next, pfn, prot);
612 pfn += (next - addr) >> PAGE_SHIFT;
613 } while (addr = next, addr != end);
618 static int create_hyp_pud_mappings(pgd_t *pgd, unsigned long start,
619 unsigned long end, unsigned long pfn,
624 unsigned long addr, next;
629 pud = pud_offset(pgd, addr);
631 if (pud_none_or_clear_bad(pud)) {
632 pmd = pmd_alloc_one(NULL, addr);
634 kvm_err("Cannot allocate Hyp pmd\n");
637 pud_populate(NULL, pud, pmd);
638 get_page(virt_to_page(pud));
639 kvm_flush_dcache_to_poc(pud, sizeof(*pud));
642 next = pud_addr_end(addr, end);
643 ret = create_hyp_pmd_mappings(pud, addr, next, pfn, prot);
646 pfn += (next - addr) >> PAGE_SHIFT;
647 } while (addr = next, addr != end);
652 static int __create_hyp_mappings(pgd_t *pgdp, unsigned long ptrs_per_pgd,
653 unsigned long start, unsigned long end,
654 unsigned long pfn, pgprot_t prot)
658 unsigned long addr, next;
661 mutex_lock(&kvm_hyp_pgd_mutex);
662 addr = start & PAGE_MASK;
663 end = PAGE_ALIGN(end);
665 pgd = pgdp + kvm_pgd_index(addr, ptrs_per_pgd);
667 if (pgd_none(*pgd)) {
668 pud = pud_alloc_one(NULL, addr);
670 kvm_err("Cannot allocate Hyp pud\n");
674 pgd_populate(NULL, pgd, pud);
675 get_page(virt_to_page(pgd));
676 kvm_flush_dcache_to_poc(pgd, sizeof(*pgd));
679 next = pgd_addr_end(addr, end);
680 err = create_hyp_pud_mappings(pgd, addr, next, pfn, prot);
683 pfn += (next - addr) >> PAGE_SHIFT;
684 } while (addr = next, addr != end);
686 mutex_unlock(&kvm_hyp_pgd_mutex);
690 static phys_addr_t kvm_kaddr_to_phys(void *kaddr)
692 if (!is_vmalloc_addr(kaddr)) {
693 BUG_ON(!virt_addr_valid(kaddr));
696 return page_to_phys(vmalloc_to_page(kaddr)) +
697 offset_in_page(kaddr);
702 * create_hyp_mappings - duplicate a kernel virtual address range in Hyp mode
703 * @from: The virtual kernel start address of the range
704 * @to: The virtual kernel end address of the range (exclusive)
705 * @prot: The protection to be applied to this range
707 * The same virtual address as the kernel virtual address is also used
708 * in Hyp-mode mapping (modulo HYP_PAGE_OFFSET) to the same underlying
711 int create_hyp_mappings(void *from, void *to, pgprot_t prot)
713 phys_addr_t phys_addr;
714 unsigned long virt_addr;
715 unsigned long start = kern_hyp_va((unsigned long)from);
716 unsigned long end = kern_hyp_va((unsigned long)to);
718 if (is_kernel_in_hyp_mode())
721 start = start & PAGE_MASK;
722 end = PAGE_ALIGN(end);
724 for (virt_addr = start; virt_addr < end; virt_addr += PAGE_SIZE) {
727 phys_addr = kvm_kaddr_to_phys(from + virt_addr - start);
728 err = __create_hyp_mappings(hyp_pgd, PTRS_PER_PGD,
729 virt_addr, virt_addr + PAGE_SIZE,
730 __phys_to_pfn(phys_addr),
739 static int __create_hyp_private_mapping(phys_addr_t phys_addr, size_t size,
740 unsigned long *haddr, pgprot_t prot)
742 pgd_t *pgd = hyp_pgd;
746 mutex_lock(&kvm_hyp_pgd_mutex);
749 * This assumes that we we have enough space below the idmap
750 * page to allocate our VAs. If not, the check below will
751 * kick. A potential alternative would be to detect that
752 * overflow and switch to an allocation above the idmap.
754 * The allocated size is always a multiple of PAGE_SIZE.
756 size = PAGE_ALIGN(size + offset_in_page(phys_addr));
757 base = io_map_base - size;
760 * Verify that BIT(VA_BITS - 1) hasn't been flipped by
761 * allocating the new area, as it would indicate we've
762 * overflowed the idmap/IO address range.
764 if ((base ^ io_map_base) & BIT(VA_BITS - 1))
769 mutex_unlock(&kvm_hyp_pgd_mutex);
774 if (__kvm_cpu_uses_extended_idmap())
777 ret = __create_hyp_mappings(pgd, __kvm_idmap_ptrs_per_pgd(),
779 __phys_to_pfn(phys_addr), prot);
783 *haddr = base + offset_in_page(phys_addr);
790 * create_hyp_io_mappings - Map IO into both kernel and HYP
791 * @phys_addr: The physical start address which gets mapped
792 * @size: Size of the region being mapped
793 * @kaddr: Kernel VA for this mapping
794 * @haddr: HYP VA for this mapping
796 int create_hyp_io_mappings(phys_addr_t phys_addr, size_t size,
797 void __iomem **kaddr,
798 void __iomem **haddr)
803 *kaddr = ioremap(phys_addr, size);
807 if (is_kernel_in_hyp_mode()) {
812 ret = __create_hyp_private_mapping(phys_addr, size,
813 &addr, PAGE_HYP_DEVICE);
821 *haddr = (void __iomem *)addr;
826 * create_hyp_exec_mappings - Map an executable range into HYP
827 * @phys_addr: The physical start address which gets mapped
828 * @size: Size of the region being mapped
829 * @haddr: HYP VA for this mapping
831 int create_hyp_exec_mappings(phys_addr_t phys_addr, size_t size,
837 BUG_ON(is_kernel_in_hyp_mode());
839 ret = __create_hyp_private_mapping(phys_addr, size,
840 &addr, PAGE_HYP_EXEC);
846 *haddr = (void *)addr;
851 * kvm_alloc_stage2_pgd - allocate level-1 table for stage-2 translation.
852 * @kvm: The KVM struct pointer for the VM.
854 * Allocates only the stage-2 HW PGD level table(s) (can support either full
855 * 40-bit input addresses or limited to 32-bit input addresses). Clears the
858 * Note we don't need locking here as this is only called when the VM is
859 * created, which can only be done once.
861 int kvm_alloc_stage2_pgd(struct kvm *kvm)
865 if (kvm->arch.pgd != NULL) {
866 kvm_err("kvm_arch already initialized?\n");
870 /* Allocate the HW PGD, making sure that each page gets its own refcount */
871 pgd = alloc_pages_exact(S2_PGD_SIZE, GFP_KERNEL | __GFP_ZERO);
879 static void stage2_unmap_memslot(struct kvm *kvm,
880 struct kvm_memory_slot *memslot)
882 hva_t hva = memslot->userspace_addr;
883 phys_addr_t addr = memslot->base_gfn << PAGE_SHIFT;
884 phys_addr_t size = PAGE_SIZE * memslot->npages;
885 hva_t reg_end = hva + size;
888 * A memory region could potentially cover multiple VMAs, and any holes
889 * between them, so iterate over all of them to find out if we should
892 * +--------------------------------------------+
893 * +---------------+----------------+ +----------------+
894 * | : VMA 1 | VMA 2 | | VMA 3 : |
895 * +---------------+----------------+ +----------------+
897 * +--------------------------------------------+
900 struct vm_area_struct *vma = find_vma(current->mm, hva);
901 hva_t vm_start, vm_end;
903 if (!vma || vma->vm_start >= reg_end)
907 * Take the intersection of this VMA with the memory region
909 vm_start = max(hva, vma->vm_start);
910 vm_end = min(reg_end, vma->vm_end);
912 if (!(vma->vm_flags & VM_PFNMAP)) {
913 gpa_t gpa = addr + (vm_start - memslot->userspace_addr);
914 unmap_stage2_range(kvm, gpa, vm_end - vm_start);
917 } while (hva < reg_end);
921 * stage2_unmap_vm - Unmap Stage-2 RAM mappings
922 * @kvm: The struct kvm pointer
924 * Go through the memregions and unmap any reguler RAM
925 * backing memory already mapped to the VM.
927 void stage2_unmap_vm(struct kvm *kvm)
929 struct kvm_memslots *slots;
930 struct kvm_memory_slot *memslot;
933 idx = srcu_read_lock(&kvm->srcu);
934 down_read(¤t->mm->mmap_sem);
935 spin_lock(&kvm->mmu_lock);
937 slots = kvm_memslots(kvm);
938 kvm_for_each_memslot(memslot, slots)
939 stage2_unmap_memslot(kvm, memslot);
941 spin_unlock(&kvm->mmu_lock);
942 up_read(¤t->mm->mmap_sem);
943 srcu_read_unlock(&kvm->srcu, idx);
947 * kvm_free_stage2_pgd - free all stage-2 tables
948 * @kvm: The KVM struct pointer for the VM.
950 * Walks the level-1 page table pointed to by kvm->arch.pgd and frees all
951 * underlying level-2 and level-3 tables before freeing the actual level-1 table
952 * and setting the struct pointer to NULL.
954 void kvm_free_stage2_pgd(struct kvm *kvm)
958 spin_lock(&kvm->mmu_lock);
960 unmap_stage2_range(kvm, 0, KVM_PHYS_SIZE);
961 pgd = READ_ONCE(kvm->arch.pgd);
962 kvm->arch.pgd = NULL;
964 spin_unlock(&kvm->mmu_lock);
966 /* Free the HW pgd, one page at a time */
968 free_pages_exact(pgd, S2_PGD_SIZE);
971 static pud_t *stage2_get_pud(struct kvm *kvm, struct kvm_mmu_memory_cache *cache,
977 pgd = kvm->arch.pgd + stage2_pgd_index(addr);
978 if (WARN_ON(stage2_pgd_none(*pgd))) {
981 pud = mmu_memory_cache_alloc(cache);
982 stage2_pgd_populate(pgd, pud);
983 get_page(virt_to_page(pgd));
986 return stage2_pud_offset(pgd, addr);
989 static pmd_t *stage2_get_pmd(struct kvm *kvm, struct kvm_mmu_memory_cache *cache,
995 pud = stage2_get_pud(kvm, cache, addr);
999 if (stage2_pud_none(*pud)) {
1002 pmd = mmu_memory_cache_alloc(cache);
1003 stage2_pud_populate(pud, pmd);
1004 get_page(virt_to_page(pud));
1007 return stage2_pmd_offset(pud, addr);
1010 static int stage2_set_pmd_huge(struct kvm *kvm, struct kvm_mmu_memory_cache
1011 *cache, phys_addr_t addr, const pmd_t *new_pmd)
1013 pmd_t *pmd, old_pmd;
1015 pmd = stage2_get_pmd(kvm, cache, addr);
1019 * Mapping in huge pages should only happen through a fault. If a
1020 * page is merged into a transparent huge page, the individual
1021 * subpages of that huge page should be unmapped through MMU
1022 * notifiers before we get here.
1024 * Merging of CompoundPages is not supported; they should become
1025 * splitting first, unmapped, merged, and mapped back in on-demand.
1027 VM_BUG_ON(pmd_present(*pmd) && pmd_pfn(*pmd) != pmd_pfn(*new_pmd));
1030 if (pmd_present(old_pmd)) {
1032 kvm_tlb_flush_vmid_ipa(kvm, addr);
1034 get_page(virt_to_page(pmd));
1037 kvm_set_pmd(pmd, *new_pmd);
1041 static bool stage2_is_exec(struct kvm *kvm, phys_addr_t addr)
1046 pmdp = stage2_get_pmd(kvm, NULL, addr);
1047 if (!pmdp || pmd_none(*pmdp) || !pmd_present(*pmdp))
1050 if (pmd_thp_or_huge(*pmdp))
1051 return kvm_s2pmd_exec(pmdp);
1053 ptep = pte_offset_kernel(pmdp, addr);
1054 if (!ptep || pte_none(*ptep) || !pte_present(*ptep))
1057 return kvm_s2pte_exec(ptep);
1060 static int stage2_set_pte(struct kvm *kvm, struct kvm_mmu_memory_cache *cache,
1061 phys_addr_t addr, const pte_t *new_pte,
1062 unsigned long flags)
1065 pte_t *pte, old_pte;
1066 bool iomap = flags & KVM_S2PTE_FLAG_IS_IOMAP;
1067 bool logging_active = flags & KVM_S2_FLAG_LOGGING_ACTIVE;
1069 VM_BUG_ON(logging_active && !cache);
1071 /* Create stage-2 page table mapping - Levels 0 and 1 */
1072 pmd = stage2_get_pmd(kvm, cache, addr);
1075 * Ignore calls from kvm_set_spte_hva for unallocated
1082 * While dirty page logging - dissolve huge PMD, then continue on to
1086 stage2_dissolve_pmd(kvm, addr, pmd);
1088 /* Create stage-2 page mappings - Level 2 */
1089 if (pmd_none(*pmd)) {
1091 return 0; /* ignore calls from kvm_set_spte_hva */
1092 pte = mmu_memory_cache_alloc(cache);
1093 pmd_populate_kernel(NULL, pmd, pte);
1094 get_page(virt_to_page(pmd));
1097 pte = pte_offset_kernel(pmd, addr);
1099 if (iomap && pte_present(*pte))
1102 /* Create 2nd stage page table mapping - Level 3 */
1104 if (pte_present(old_pte)) {
1105 kvm_set_pte(pte, __pte(0));
1106 kvm_tlb_flush_vmid_ipa(kvm, addr);
1108 get_page(virt_to_page(pte));
1111 kvm_set_pte(pte, *new_pte);
1115 #ifndef __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
1116 static int stage2_ptep_test_and_clear_young(pte_t *pte)
1118 if (pte_young(*pte)) {
1119 *pte = pte_mkold(*pte);
1125 static int stage2_ptep_test_and_clear_young(pte_t *pte)
1127 return __ptep_test_and_clear_young(pte);
1131 static int stage2_pmdp_test_and_clear_young(pmd_t *pmd)
1133 return stage2_ptep_test_and_clear_young((pte_t *)pmd);
1137 * kvm_phys_addr_ioremap - map a device range to guest IPA
1139 * @kvm: The KVM pointer
1140 * @guest_ipa: The IPA at which to insert the mapping
1141 * @pa: The physical address of the device
1142 * @size: The size of the mapping
1144 int kvm_phys_addr_ioremap(struct kvm *kvm, phys_addr_t guest_ipa,
1145 phys_addr_t pa, unsigned long size, bool writable)
1147 phys_addr_t addr, end;
1150 struct kvm_mmu_memory_cache cache = { 0, };
1152 end = (guest_ipa + size + PAGE_SIZE - 1) & PAGE_MASK;
1153 pfn = __phys_to_pfn(pa);
1155 for (addr = guest_ipa; addr < end; addr += PAGE_SIZE) {
1156 pte_t pte = pfn_pte(pfn, PAGE_S2_DEVICE);
1159 pte = kvm_s2pte_mkwrite(pte);
1161 ret = mmu_topup_memory_cache(&cache, KVM_MMU_CACHE_MIN_PAGES,
1165 spin_lock(&kvm->mmu_lock);
1166 ret = stage2_set_pte(kvm, &cache, addr, &pte,
1167 KVM_S2PTE_FLAG_IS_IOMAP);
1168 spin_unlock(&kvm->mmu_lock);
1176 mmu_free_memory_cache(&cache);
1180 static bool transparent_hugepage_adjust(kvm_pfn_t *pfnp, phys_addr_t *ipap)
1182 kvm_pfn_t pfn = *pfnp;
1183 gfn_t gfn = *ipap >> PAGE_SHIFT;
1185 if (PageTransCompoundMap(pfn_to_page(pfn))) {
1188 * The address we faulted on is backed by a transparent huge
1189 * page. However, because we map the compound huge page and
1190 * not the individual tail page, we need to transfer the
1191 * refcount to the head page. We have to be careful that the
1192 * THP doesn't start to split while we are adjusting the
1195 * We are sure this doesn't happen, because mmu_notifier_retry
1196 * was successful and we are holding the mmu_lock, so if this
1197 * THP is trying to split, it will be blocked in the mmu
1198 * notifier before touching any of the pages, specifically
1199 * before being able to call __split_huge_page_refcount().
1201 * We can therefore safely transfer the refcount from PG_tail
1202 * to PG_head and switch the pfn from a tail page to the head
1205 mask = PTRS_PER_PMD - 1;
1206 VM_BUG_ON((gfn & mask) != (pfn & mask));
1209 kvm_release_pfn_clean(pfn);
1221 static bool kvm_is_write_fault(struct kvm_vcpu *vcpu)
1223 if (kvm_vcpu_trap_is_iabt(vcpu))
1226 return kvm_vcpu_dabt_iswrite(vcpu);
1230 * stage2_wp_ptes - write protect PMD range
1231 * @pmd: pointer to pmd entry
1232 * @addr: range start address
1233 * @end: range end address
1235 static void stage2_wp_ptes(pmd_t *pmd, phys_addr_t addr, phys_addr_t end)
1239 pte = pte_offset_kernel(pmd, addr);
1241 if (!pte_none(*pte)) {
1242 if (!kvm_s2pte_readonly(pte))
1243 kvm_set_s2pte_readonly(pte);
1245 } while (pte++, addr += PAGE_SIZE, addr != end);
1249 * stage2_wp_pmds - write protect PUD range
1250 * @pud: pointer to pud entry
1251 * @addr: range start address
1252 * @end: range end address
1254 static void stage2_wp_pmds(pud_t *pud, phys_addr_t addr, phys_addr_t end)
1259 pmd = stage2_pmd_offset(pud, addr);
1262 next = stage2_pmd_addr_end(addr, end);
1263 if (!pmd_none(*pmd)) {
1264 if (pmd_thp_or_huge(*pmd)) {
1265 if (!kvm_s2pmd_readonly(pmd))
1266 kvm_set_s2pmd_readonly(pmd);
1268 stage2_wp_ptes(pmd, addr, next);
1271 } while (pmd++, addr = next, addr != end);
1275 * stage2_wp_puds - write protect PGD range
1276 * @pgd: pointer to pgd entry
1277 * @addr: range start address
1278 * @end: range end address
1280 * Process PUD entries, for a huge PUD we cause a panic.
1282 static void stage2_wp_puds(pgd_t *pgd, phys_addr_t addr, phys_addr_t end)
1287 pud = stage2_pud_offset(pgd, addr);
1289 next = stage2_pud_addr_end(addr, end);
1290 if (!stage2_pud_none(*pud)) {
1291 /* TODO:PUD not supported, revisit later if supported */
1292 BUG_ON(stage2_pud_huge(*pud));
1293 stage2_wp_pmds(pud, addr, next);
1295 } while (pud++, addr = next, addr != end);
1299 * stage2_wp_range() - write protect stage2 memory region range
1300 * @kvm: The KVM pointer
1301 * @addr: Start address of range
1302 * @end: End address of range
1304 static void stage2_wp_range(struct kvm *kvm, phys_addr_t addr, phys_addr_t end)
1309 pgd = kvm->arch.pgd + stage2_pgd_index(addr);
1312 * Release kvm_mmu_lock periodically if the memory region is
1313 * large. Otherwise, we may see kernel panics with
1314 * CONFIG_DETECT_HUNG_TASK, CONFIG_LOCKUP_DETECTOR,
1315 * CONFIG_LOCKDEP. Additionally, holding the lock too long
1316 * will also starve other vCPUs. We have to also make sure
1317 * that the page tables are not freed while we released
1320 cond_resched_lock(&kvm->mmu_lock);
1321 if (!READ_ONCE(kvm->arch.pgd))
1323 next = stage2_pgd_addr_end(addr, end);
1324 if (stage2_pgd_present(*pgd))
1325 stage2_wp_puds(pgd, addr, next);
1326 } while (pgd++, addr = next, addr != end);
1330 * kvm_mmu_wp_memory_region() - write protect stage 2 entries for memory slot
1331 * @kvm: The KVM pointer
1332 * @slot: The memory slot to write protect
1334 * Called to start logging dirty pages after memory region
1335 * KVM_MEM_LOG_DIRTY_PAGES operation is called. After this function returns
1336 * all present PMD and PTEs are write protected in the memory region.
1337 * Afterwards read of dirty page log can be called.
1339 * Acquires kvm_mmu_lock. Called with kvm->slots_lock mutex acquired,
1340 * serializing operations for VM memory regions.
1342 void kvm_mmu_wp_memory_region(struct kvm *kvm, int slot)
1344 struct kvm_memslots *slots = kvm_memslots(kvm);
1345 struct kvm_memory_slot *memslot = id_to_memslot(slots, slot);
1346 phys_addr_t start = memslot->base_gfn << PAGE_SHIFT;
1347 phys_addr_t end = (memslot->base_gfn + memslot->npages) << PAGE_SHIFT;
1349 spin_lock(&kvm->mmu_lock);
1350 stage2_wp_range(kvm, start, end);
1351 spin_unlock(&kvm->mmu_lock);
1352 kvm_flush_remote_tlbs(kvm);
1356 * kvm_mmu_write_protect_pt_masked() - write protect dirty pages
1357 * @kvm: The KVM pointer
1358 * @slot: The memory slot associated with mask
1359 * @gfn_offset: The gfn offset in memory slot
1360 * @mask: The mask of dirty pages at offset 'gfn_offset' in this memory
1361 * slot to be write protected
1363 * Walks bits set in mask write protects the associated pte's. Caller must
1364 * acquire kvm_mmu_lock.
1366 static void kvm_mmu_write_protect_pt_masked(struct kvm *kvm,
1367 struct kvm_memory_slot *slot,
1368 gfn_t gfn_offset, unsigned long mask)
1370 phys_addr_t base_gfn = slot->base_gfn + gfn_offset;
1371 phys_addr_t start = (base_gfn + __ffs(mask)) << PAGE_SHIFT;
1372 phys_addr_t end = (base_gfn + __fls(mask) + 1) << PAGE_SHIFT;
1374 stage2_wp_range(kvm, start, end);
1378 * kvm_arch_mmu_enable_log_dirty_pt_masked - enable dirty logging for selected
1381 * It calls kvm_mmu_write_protect_pt_masked to write protect selected pages to
1382 * enable dirty logging for them.
1384 void kvm_arch_mmu_enable_log_dirty_pt_masked(struct kvm *kvm,
1385 struct kvm_memory_slot *slot,
1386 gfn_t gfn_offset, unsigned long mask)
1388 kvm_mmu_write_protect_pt_masked(kvm, slot, gfn_offset, mask);
1391 static void clean_dcache_guest_page(kvm_pfn_t pfn, unsigned long size)
1393 __clean_dcache_guest_page(pfn, size);
1396 static void invalidate_icache_guest_page(kvm_pfn_t pfn, unsigned long size)
1398 __invalidate_icache_guest_page(pfn, size);
1401 static void kvm_send_hwpoison_signal(unsigned long address,
1402 struct vm_area_struct *vma)
1406 clear_siginfo(&info);
1407 info.si_signo = SIGBUS;
1409 info.si_code = BUS_MCEERR_AR;
1410 info.si_addr = (void __user *)address;
1412 if (is_vm_hugetlb_page(vma))
1413 info.si_addr_lsb = huge_page_shift(hstate_vma(vma));
1415 info.si_addr_lsb = PAGE_SHIFT;
1417 send_sig_info(SIGBUS, &info, current);
1420 static int user_mem_abort(struct kvm_vcpu *vcpu, phys_addr_t fault_ipa,
1421 struct kvm_memory_slot *memslot, unsigned long hva,
1422 unsigned long fault_status)
1425 bool write_fault, exec_fault, writable, hugetlb = false, force_pte = false;
1426 unsigned long mmu_seq;
1427 gfn_t gfn = fault_ipa >> PAGE_SHIFT;
1428 struct kvm *kvm = vcpu->kvm;
1429 struct kvm_mmu_memory_cache *memcache = &vcpu->arch.mmu_page_cache;
1430 struct vm_area_struct *vma;
1432 pgprot_t mem_type = PAGE_S2;
1433 bool logging_active = memslot_is_logging(memslot);
1434 unsigned long flags = 0;
1436 write_fault = kvm_is_write_fault(vcpu);
1437 exec_fault = kvm_vcpu_trap_is_iabt(vcpu);
1438 VM_BUG_ON(write_fault && exec_fault);
1440 if (fault_status == FSC_PERM && !write_fault && !exec_fault) {
1441 kvm_err("Unexpected L2 read permission error\n");
1445 /* Let's check if we will get back a huge page backed by hugetlbfs */
1446 down_read(¤t->mm->mmap_sem);
1447 vma = find_vma_intersection(current->mm, hva, hva + 1);
1448 if (unlikely(!vma)) {
1449 kvm_err("Failed to find VMA for hva 0x%lx\n", hva);
1450 up_read(¤t->mm->mmap_sem);
1454 if (vma_kernel_pagesize(vma) == PMD_SIZE && !logging_active) {
1456 gfn = (fault_ipa & PMD_MASK) >> PAGE_SHIFT;
1459 * Pages belonging to memslots that don't have the same
1460 * alignment for userspace and IPA cannot be mapped using
1461 * block descriptors even if the pages belong to a THP for
1462 * the process, because the stage-2 block descriptor will
1463 * cover more than a single THP and we loose atomicity for
1464 * unmapping, updates, and splits of the THP or other pages
1465 * in the stage-2 block range.
1467 if ((memslot->userspace_addr & ~PMD_MASK) !=
1468 ((memslot->base_gfn << PAGE_SHIFT) & ~PMD_MASK))
1471 up_read(¤t->mm->mmap_sem);
1473 /* We need minimum second+third level pages */
1474 ret = mmu_topup_memory_cache(memcache, KVM_MMU_CACHE_MIN_PAGES,
1479 mmu_seq = vcpu->kvm->mmu_notifier_seq;
1481 * Ensure the read of mmu_notifier_seq happens before we call
1482 * gfn_to_pfn_prot (which calls get_user_pages), so that we don't risk
1483 * the page we just got a reference to gets unmapped before we have a
1484 * chance to grab the mmu_lock, which ensure that if the page gets
1485 * unmapped afterwards, the call to kvm_unmap_hva will take it away
1486 * from us again properly. This smp_rmb() interacts with the smp_wmb()
1487 * in kvm_mmu_notifier_invalidate_<page|range_end>.
1491 pfn = gfn_to_pfn_prot(kvm, gfn, write_fault, &writable);
1492 if (pfn == KVM_PFN_ERR_HWPOISON) {
1493 kvm_send_hwpoison_signal(hva, vma);
1496 if (is_error_noslot_pfn(pfn))
1499 if (kvm_is_device_pfn(pfn)) {
1500 mem_type = PAGE_S2_DEVICE;
1501 flags |= KVM_S2PTE_FLAG_IS_IOMAP;
1502 } else if (logging_active) {
1504 * Faults on pages in a memslot with logging enabled
1505 * should not be mapped with huge pages (it introduces churn
1506 * and performance degradation), so force a pte mapping.
1509 flags |= KVM_S2_FLAG_LOGGING_ACTIVE;
1512 * Only actually map the page as writable if this was a write
1519 spin_lock(&kvm->mmu_lock);
1520 if (mmu_notifier_retry(kvm, mmu_seq))
1523 if (!hugetlb && !force_pte)
1524 hugetlb = transparent_hugepage_adjust(&pfn, &fault_ipa);
1527 pmd_t new_pmd = pfn_pmd(pfn, mem_type);
1528 new_pmd = pmd_mkhuge(new_pmd);
1530 new_pmd = kvm_s2pmd_mkwrite(new_pmd);
1531 kvm_set_pfn_dirty(pfn);
1534 if (fault_status != FSC_PERM)
1535 clean_dcache_guest_page(pfn, PMD_SIZE);
1538 new_pmd = kvm_s2pmd_mkexec(new_pmd);
1539 invalidate_icache_guest_page(pfn, PMD_SIZE);
1540 } else if (fault_status == FSC_PERM) {
1541 /* Preserve execute if XN was already cleared */
1542 if (stage2_is_exec(kvm, fault_ipa))
1543 new_pmd = kvm_s2pmd_mkexec(new_pmd);
1546 ret = stage2_set_pmd_huge(kvm, memcache, fault_ipa, &new_pmd);
1548 pte_t new_pte = pfn_pte(pfn, mem_type);
1551 new_pte = kvm_s2pte_mkwrite(new_pte);
1552 kvm_set_pfn_dirty(pfn);
1553 mark_page_dirty(kvm, gfn);
1556 if (fault_status != FSC_PERM)
1557 clean_dcache_guest_page(pfn, PAGE_SIZE);
1560 new_pte = kvm_s2pte_mkexec(new_pte);
1561 invalidate_icache_guest_page(pfn, PAGE_SIZE);
1562 } else if (fault_status == FSC_PERM) {
1563 /* Preserve execute if XN was already cleared */
1564 if (stage2_is_exec(kvm, fault_ipa))
1565 new_pte = kvm_s2pte_mkexec(new_pte);
1568 ret = stage2_set_pte(kvm, memcache, fault_ipa, &new_pte, flags);
1572 spin_unlock(&kvm->mmu_lock);
1573 kvm_set_pfn_accessed(pfn);
1574 kvm_release_pfn_clean(pfn);
1579 * Resolve the access fault by making the page young again.
1580 * Note that because the faulting entry is guaranteed not to be
1581 * cached in the TLB, we don't need to invalidate anything.
1582 * Only the HW Access Flag updates are supported for Stage 2 (no DBM),
1583 * so there is no need for atomic (pte|pmd)_mkyoung operations.
1585 static void handle_access_fault(struct kvm_vcpu *vcpu, phys_addr_t fault_ipa)
1590 bool pfn_valid = false;
1592 trace_kvm_access_fault(fault_ipa);
1594 spin_lock(&vcpu->kvm->mmu_lock);
1596 pmd = stage2_get_pmd(vcpu->kvm, NULL, fault_ipa);
1597 if (!pmd || pmd_none(*pmd)) /* Nothing there */
1600 if (pmd_thp_or_huge(*pmd)) { /* THP, HugeTLB */
1601 *pmd = pmd_mkyoung(*pmd);
1602 pfn = pmd_pfn(*pmd);
1607 pte = pte_offset_kernel(pmd, fault_ipa);
1608 if (pte_none(*pte)) /* Nothing there either */
1611 *pte = pte_mkyoung(*pte); /* Just a page... */
1612 pfn = pte_pfn(*pte);
1615 spin_unlock(&vcpu->kvm->mmu_lock);
1617 kvm_set_pfn_accessed(pfn);
1621 * kvm_handle_guest_abort - handles all 2nd stage aborts
1622 * @vcpu: the VCPU pointer
1623 * @run: the kvm_run structure
1625 * Any abort that gets to the host is almost guaranteed to be caused by a
1626 * missing second stage translation table entry, which can mean that either the
1627 * guest simply needs more memory and we must allocate an appropriate page or it
1628 * can mean that the guest tried to access I/O memory, which is emulated by user
1629 * space. The distinction is based on the IPA causing the fault and whether this
1630 * memory region has been registered as standard RAM by user space.
1632 int kvm_handle_guest_abort(struct kvm_vcpu *vcpu, struct kvm_run *run)
1634 unsigned long fault_status;
1635 phys_addr_t fault_ipa;
1636 struct kvm_memory_slot *memslot;
1638 bool is_iabt, write_fault, writable;
1642 fault_status = kvm_vcpu_trap_get_fault_type(vcpu);
1644 fault_ipa = kvm_vcpu_get_fault_ipa(vcpu);
1645 is_iabt = kvm_vcpu_trap_is_iabt(vcpu);
1647 /* Synchronous External Abort? */
1648 if (kvm_vcpu_dabt_isextabt(vcpu)) {
1650 * For RAS the host kernel may handle this abort.
1651 * There is no need to pass the error into the guest.
1653 if (!handle_guest_sea(fault_ipa, kvm_vcpu_get_hsr(vcpu)))
1656 if (unlikely(!is_iabt)) {
1657 kvm_inject_vabt(vcpu);
1662 trace_kvm_guest_fault(*vcpu_pc(vcpu), kvm_vcpu_get_hsr(vcpu),
1663 kvm_vcpu_get_hfar(vcpu), fault_ipa);
1665 /* Check the stage-2 fault is trans. fault or write fault */
1666 if (fault_status != FSC_FAULT && fault_status != FSC_PERM &&
1667 fault_status != FSC_ACCESS) {
1668 kvm_err("Unsupported FSC: EC=%#x xFSC=%#lx ESR_EL2=%#lx\n",
1669 kvm_vcpu_trap_get_class(vcpu),
1670 (unsigned long)kvm_vcpu_trap_get_fault(vcpu),
1671 (unsigned long)kvm_vcpu_get_hsr(vcpu));
1675 idx = srcu_read_lock(&vcpu->kvm->srcu);
1677 gfn = fault_ipa >> PAGE_SHIFT;
1678 memslot = gfn_to_memslot(vcpu->kvm, gfn);
1679 hva = gfn_to_hva_memslot_prot(memslot, gfn, &writable);
1680 write_fault = kvm_is_write_fault(vcpu);
1681 if (kvm_is_error_hva(hva) || (write_fault && !writable)) {
1683 /* Prefetch Abort on I/O address */
1684 kvm_inject_pabt(vcpu, kvm_vcpu_get_hfar(vcpu));
1690 * Check for a cache maintenance operation. Since we
1691 * ended-up here, we know it is outside of any memory
1692 * slot. But we can't find out if that is for a device,
1693 * or if the guest is just being stupid. The only thing
1694 * we know for sure is that this range cannot be cached.
1696 * So let's assume that the guest is just being
1697 * cautious, and skip the instruction.
1699 if (kvm_vcpu_dabt_is_cm(vcpu)) {
1700 kvm_skip_instr(vcpu, kvm_vcpu_trap_il_is32bit(vcpu));
1706 * The IPA is reported as [MAX:12], so we need to
1707 * complement it with the bottom 12 bits from the
1708 * faulting VA. This is always 12 bits, irrespective
1711 fault_ipa |= kvm_vcpu_get_hfar(vcpu) & ((1 << 12) - 1);
1712 ret = io_mem_abort(vcpu, run, fault_ipa);
1716 /* Userspace should not be able to register out-of-bounds IPAs */
1717 VM_BUG_ON(fault_ipa >= KVM_PHYS_SIZE);
1719 if (fault_status == FSC_ACCESS) {
1720 handle_access_fault(vcpu, fault_ipa);
1725 ret = user_mem_abort(vcpu, fault_ipa, memslot, hva, fault_status);
1729 srcu_read_unlock(&vcpu->kvm->srcu, idx);
1733 static int handle_hva_to_gpa(struct kvm *kvm,
1734 unsigned long start,
1736 int (*handler)(struct kvm *kvm,
1737 gpa_t gpa, u64 size,
1741 struct kvm_memslots *slots;
1742 struct kvm_memory_slot *memslot;
1745 slots = kvm_memslots(kvm);
1747 /* we only care about the pages that the guest sees */
1748 kvm_for_each_memslot(memslot, slots) {
1749 unsigned long hva_start, hva_end;
1752 hva_start = max(start, memslot->userspace_addr);
1753 hva_end = min(end, memslot->userspace_addr +
1754 (memslot->npages << PAGE_SHIFT));
1755 if (hva_start >= hva_end)
1758 gpa = hva_to_gfn_memslot(hva_start, memslot) << PAGE_SHIFT;
1759 ret |= handler(kvm, gpa, (u64)(hva_end - hva_start), data);
1765 static int kvm_unmap_hva_handler(struct kvm *kvm, gpa_t gpa, u64 size, void *data)
1767 unmap_stage2_range(kvm, gpa, size);
1771 int kvm_unmap_hva(struct kvm *kvm, unsigned long hva)
1773 unsigned long end = hva + PAGE_SIZE;
1778 trace_kvm_unmap_hva(hva);
1779 handle_hva_to_gpa(kvm, hva, end, &kvm_unmap_hva_handler, NULL);
1783 int kvm_unmap_hva_range(struct kvm *kvm,
1784 unsigned long start, unsigned long end)
1789 trace_kvm_unmap_hva_range(start, end);
1790 handle_hva_to_gpa(kvm, start, end, &kvm_unmap_hva_handler, NULL);
1794 static int kvm_set_spte_handler(struct kvm *kvm, gpa_t gpa, u64 size, void *data)
1796 pte_t *pte = (pte_t *)data;
1798 WARN_ON(size != PAGE_SIZE);
1800 * We can always call stage2_set_pte with KVM_S2PTE_FLAG_LOGGING_ACTIVE
1801 * flag clear because MMU notifiers will have unmapped a huge PMD before
1802 * calling ->change_pte() (which in turn calls kvm_set_spte_hva()) and
1803 * therefore stage2_set_pte() never needs to clear out a huge PMD
1804 * through this calling path.
1806 stage2_set_pte(kvm, NULL, gpa, pte, 0);
1811 void kvm_set_spte_hva(struct kvm *kvm, unsigned long hva, pte_t pte)
1813 unsigned long end = hva + PAGE_SIZE;
1819 trace_kvm_set_spte_hva(hva);
1820 stage2_pte = pfn_pte(pte_pfn(pte), PAGE_S2);
1821 handle_hva_to_gpa(kvm, hva, end, &kvm_set_spte_handler, &stage2_pte);
1824 static int kvm_age_hva_handler(struct kvm *kvm, gpa_t gpa, u64 size, void *data)
1829 WARN_ON(size != PAGE_SIZE && size != PMD_SIZE);
1830 pmd = stage2_get_pmd(kvm, NULL, gpa);
1831 if (!pmd || pmd_none(*pmd)) /* Nothing there */
1834 if (pmd_thp_or_huge(*pmd)) /* THP, HugeTLB */
1835 return stage2_pmdp_test_and_clear_young(pmd);
1837 pte = pte_offset_kernel(pmd, gpa);
1841 return stage2_ptep_test_and_clear_young(pte);
1844 static int kvm_test_age_hva_handler(struct kvm *kvm, gpa_t gpa, u64 size, void *data)
1849 WARN_ON(size != PAGE_SIZE && size != PMD_SIZE);
1850 pmd = stage2_get_pmd(kvm, NULL, gpa);
1851 if (!pmd || pmd_none(*pmd)) /* Nothing there */
1854 if (pmd_thp_or_huge(*pmd)) /* THP, HugeTLB */
1855 return pmd_young(*pmd);
1857 pte = pte_offset_kernel(pmd, gpa);
1858 if (!pte_none(*pte)) /* Just a page... */
1859 return pte_young(*pte);
1864 int kvm_age_hva(struct kvm *kvm, unsigned long start, unsigned long end)
1868 trace_kvm_age_hva(start, end);
1869 return handle_hva_to_gpa(kvm, start, end, kvm_age_hva_handler, NULL);
1872 int kvm_test_age_hva(struct kvm *kvm, unsigned long hva)
1876 trace_kvm_test_age_hva(hva);
1877 return handle_hva_to_gpa(kvm, hva, hva, kvm_test_age_hva_handler, NULL);
1880 void kvm_mmu_free_memory_caches(struct kvm_vcpu *vcpu)
1882 mmu_free_memory_cache(&vcpu->arch.mmu_page_cache);
1885 phys_addr_t kvm_mmu_get_httbr(void)
1887 if (__kvm_cpu_uses_extended_idmap())
1888 return virt_to_phys(merged_hyp_pgd);
1890 return virt_to_phys(hyp_pgd);
1893 phys_addr_t kvm_get_idmap_vector(void)
1895 return hyp_idmap_vector;
1898 static int kvm_map_idmap_text(pgd_t *pgd)
1902 /* Create the idmap in the boot page tables */
1903 err = __create_hyp_mappings(pgd, __kvm_idmap_ptrs_per_pgd(),
1904 hyp_idmap_start, hyp_idmap_end,
1905 __phys_to_pfn(hyp_idmap_start),
1908 kvm_err("Failed to idmap %lx-%lx\n",
1909 hyp_idmap_start, hyp_idmap_end);
1914 int kvm_mmu_init(void)
1918 hyp_idmap_start = kvm_virt_to_phys(__hyp_idmap_text_start);
1919 hyp_idmap_start = ALIGN_DOWN(hyp_idmap_start, PAGE_SIZE);
1920 hyp_idmap_end = kvm_virt_to_phys(__hyp_idmap_text_end);
1921 hyp_idmap_end = ALIGN(hyp_idmap_end, PAGE_SIZE);
1922 hyp_idmap_vector = kvm_virt_to_phys(__kvm_hyp_init);
1925 * We rely on the linker script to ensure at build time that the HYP
1926 * init code does not cross a page boundary.
1928 BUG_ON((hyp_idmap_start ^ (hyp_idmap_end - 1)) & PAGE_MASK);
1930 kvm_debug("IDMAP page: %lx\n", hyp_idmap_start);
1931 kvm_debug("HYP VA range: %lx:%lx\n",
1932 kern_hyp_va(PAGE_OFFSET),
1933 kern_hyp_va((unsigned long)high_memory - 1));
1935 if (hyp_idmap_start >= kern_hyp_va(PAGE_OFFSET) &&
1936 hyp_idmap_start < kern_hyp_va((unsigned long)high_memory - 1) &&
1937 hyp_idmap_start != (unsigned long)__hyp_idmap_text_start) {
1939 * The idmap page is intersecting with the VA space,
1940 * it is not safe to continue further.
1942 kvm_err("IDMAP intersecting with HYP VA, unable to continue\n");
1947 hyp_pgd = (pgd_t *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, hyp_pgd_order);
1949 kvm_err("Hyp mode PGD not allocated\n");
1954 if (__kvm_cpu_uses_extended_idmap()) {
1955 boot_hyp_pgd = (pgd_t *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
1957 if (!boot_hyp_pgd) {
1958 kvm_err("Hyp boot PGD not allocated\n");
1963 err = kvm_map_idmap_text(boot_hyp_pgd);
1967 merged_hyp_pgd = (pgd_t *)__get_free_page(GFP_KERNEL | __GFP_ZERO);
1968 if (!merged_hyp_pgd) {
1969 kvm_err("Failed to allocate extra HYP pgd\n");
1972 __kvm_extend_hypmap(boot_hyp_pgd, hyp_pgd, merged_hyp_pgd,
1975 err = kvm_map_idmap_text(hyp_pgd);
1980 io_map_base = hyp_idmap_start;
1987 void kvm_arch_commit_memory_region(struct kvm *kvm,
1988 const struct kvm_userspace_memory_region *mem,
1989 const struct kvm_memory_slot *old,
1990 const struct kvm_memory_slot *new,
1991 enum kvm_mr_change change)
1994 * At this point memslot has been committed and there is an
1995 * allocated dirty_bitmap[], dirty pages will be be tracked while the
1996 * memory slot is write protected.
1998 if (change != KVM_MR_DELETE && mem->flags & KVM_MEM_LOG_DIRTY_PAGES)
1999 kvm_mmu_wp_memory_region(kvm, mem->slot);
2002 int kvm_arch_prepare_memory_region(struct kvm *kvm,
2003 struct kvm_memory_slot *memslot,
2004 const struct kvm_userspace_memory_region *mem,
2005 enum kvm_mr_change change)
2007 hva_t hva = mem->userspace_addr;
2008 hva_t reg_end = hva + mem->memory_size;
2009 bool writable = !(mem->flags & KVM_MEM_READONLY);
2012 if (change != KVM_MR_CREATE && change != KVM_MR_MOVE &&
2013 change != KVM_MR_FLAGS_ONLY)
2017 * Prevent userspace from creating a memory region outside of the IPA
2018 * space addressable by the KVM guest IPA space.
2020 if (memslot->base_gfn + memslot->npages >=
2021 (KVM_PHYS_SIZE >> PAGE_SHIFT))
2024 down_read(¤t->mm->mmap_sem);
2026 * A memory region could potentially cover multiple VMAs, and any holes
2027 * between them, so iterate over all of them to find out if we can map
2028 * any of them right now.
2030 * +--------------------------------------------+
2031 * +---------------+----------------+ +----------------+
2032 * | : VMA 1 | VMA 2 | | VMA 3 : |
2033 * +---------------+----------------+ +----------------+
2035 * +--------------------------------------------+
2038 struct vm_area_struct *vma = find_vma(current->mm, hva);
2039 hva_t vm_start, vm_end;
2041 if (!vma || vma->vm_start >= reg_end)
2045 * Mapping a read-only VMA is only allowed if the
2046 * memory region is configured as read-only.
2048 if (writable && !(vma->vm_flags & VM_WRITE)) {
2054 * Take the intersection of this VMA with the memory region
2056 vm_start = max(hva, vma->vm_start);
2057 vm_end = min(reg_end, vma->vm_end);
2059 if (vma->vm_flags & VM_PFNMAP) {
2060 gpa_t gpa = mem->guest_phys_addr +
2061 (vm_start - mem->userspace_addr);
2064 pa = (phys_addr_t)vma->vm_pgoff << PAGE_SHIFT;
2065 pa += vm_start - vma->vm_start;
2067 /* IO region dirty page logging not allowed */
2068 if (memslot->flags & KVM_MEM_LOG_DIRTY_PAGES) {
2073 ret = kvm_phys_addr_ioremap(kvm, gpa, pa,
2080 } while (hva < reg_end);
2082 if (change == KVM_MR_FLAGS_ONLY)
2085 spin_lock(&kvm->mmu_lock);
2087 unmap_stage2_range(kvm, mem->guest_phys_addr, mem->memory_size);
2089 stage2_flush_memslot(kvm, memslot);
2090 spin_unlock(&kvm->mmu_lock);
2092 up_read(¤t->mm->mmap_sem);
2096 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
2097 struct kvm_memory_slot *dont)
2101 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
2102 unsigned long npages)
2107 void kvm_arch_memslots_updated(struct kvm *kvm, struct kvm_memslots *slots)
2111 void kvm_arch_flush_shadow_all(struct kvm *kvm)
2113 kvm_free_stage2_pgd(kvm);
2116 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
2117 struct kvm_memory_slot *slot)
2119 gpa_t gpa = slot->base_gfn << PAGE_SHIFT;
2120 phys_addr_t size = slot->npages << PAGE_SHIFT;
2122 spin_lock(&kvm->mmu_lock);
2123 unmap_stage2_range(kvm, gpa, size);
2124 spin_unlock(&kvm->mmu_lock);
2128 * See note at ARMv7 ARM B1.14.4 (TL;DR: S/W ops are not easily virtualized).
2131 * - S/W ops are local to a CPU (not broadcast)
2132 * - We have line migration behind our back (speculation)
2133 * - System caches don't support S/W at all (damn!)
2135 * In the face of the above, the best we can do is to try and convert
2136 * S/W ops to VA ops. Because the guest is not allowed to infer the
2137 * S/W to PA mapping, it can only use S/W to nuke the whole cache,
2138 * which is a rather good thing for us.
2140 * Also, it is only used when turning caches on/off ("The expected
2141 * usage of the cache maintenance instructions that operate by set/way
2142 * is associated with the cache maintenance instructions associated
2143 * with the powerdown and powerup of caches, if this is required by
2144 * the implementation.").
2146 * We use the following policy:
2148 * - If we trap a S/W operation, we enable VM trapping to detect
2149 * caches being turned on/off, and do a full clean.
2151 * - We flush the caches on both caches being turned on and off.
2153 * - Once the caches are enabled, we stop trapping VM ops.
2155 void kvm_set_way_flush(struct kvm_vcpu *vcpu)
2157 unsigned long hcr = *vcpu_hcr(vcpu);
2160 * If this is the first time we do a S/W operation
2161 * (i.e. HCR_TVM not set) flush the whole memory, and set the
2164 * Otherwise, rely on the VM trapping to wait for the MMU +
2165 * Caches to be turned off. At that point, we'll be able to
2166 * clean the caches again.
2168 if (!(hcr & HCR_TVM)) {
2169 trace_kvm_set_way_flush(*vcpu_pc(vcpu),
2170 vcpu_has_cache_enabled(vcpu));
2171 stage2_flush_vm(vcpu->kvm);
2172 *vcpu_hcr(vcpu) = hcr | HCR_TVM;
2176 void kvm_toggle_cache(struct kvm_vcpu *vcpu, bool was_enabled)
2178 bool now_enabled = vcpu_has_cache_enabled(vcpu);
2181 * If switching the MMU+caches on, need to invalidate the caches.
2182 * If switching it off, need to clean the caches.
2183 * Clean + invalidate does the trick always.
2185 if (now_enabled != was_enabled)
2186 stage2_flush_vm(vcpu->kvm);
2188 /* Caches are now on, stop trapping VM ops (until a S/W op) */
2190 *vcpu_hcr(vcpu) &= ~HCR_TVM;
2192 trace_kvm_toggle_cache(*vcpu_pc(vcpu), was_enabled, now_enabled);