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 <trace/events/kvm.h>
24 #include <asm/pgalloc.h>
25 #include <asm/cacheflush.h>
26 #include <asm/kvm_arm.h>
27 #include <asm/kvm_mmu.h>
28 #include <asm/kvm_mmio.h>
29 #include <asm/kvm_asm.h>
30 #include <asm/kvm_emulate.h>
35 static pgd_t *boot_hyp_pgd;
36 static pgd_t *hyp_pgd;
37 static pgd_t *merged_hyp_pgd;
38 static DEFINE_MUTEX(kvm_hyp_pgd_mutex);
40 static unsigned long hyp_idmap_start;
41 static unsigned long hyp_idmap_end;
42 static phys_addr_t hyp_idmap_vector;
44 #define S2_PGD_SIZE (PTRS_PER_S2_PGD * sizeof(pgd_t))
45 #define hyp_pgd_order get_order(PTRS_PER_PGD * sizeof(pgd_t))
47 #define KVM_S2PTE_FLAG_IS_IOMAP (1UL << 0)
48 #define KVM_S2_FLAG_LOGGING_ACTIVE (1UL << 1)
50 static bool memslot_is_logging(struct kvm_memory_slot *memslot)
52 return memslot->dirty_bitmap && !(memslot->flags & KVM_MEM_READONLY);
56 * kvm_flush_remote_tlbs() - flush all VM TLB entries for v7/8
57 * @kvm: pointer to kvm structure.
59 * Interface to HYP function to flush all VM TLB entries
61 void kvm_flush_remote_tlbs(struct kvm *kvm)
63 kvm_call_hyp(__kvm_tlb_flush_vmid, kvm);
66 static void kvm_tlb_flush_vmid_ipa(struct kvm *kvm, phys_addr_t ipa)
68 kvm_call_hyp(__kvm_tlb_flush_vmid_ipa, kvm, ipa);
72 * D-Cache management functions. They take the page table entries by
73 * value, as they are flushing the cache using the kernel mapping (or
76 static void kvm_flush_dcache_pte(pte_t pte)
78 __kvm_flush_dcache_pte(pte);
81 static void kvm_flush_dcache_pmd(pmd_t pmd)
83 __kvm_flush_dcache_pmd(pmd);
86 static void kvm_flush_dcache_pud(pud_t pud)
88 __kvm_flush_dcache_pud(pud);
91 static bool kvm_is_device_pfn(unsigned long pfn)
93 return !pfn_valid(pfn);
97 * stage2_dissolve_pmd() - clear and flush huge PMD entry
98 * @kvm: pointer to kvm structure.
100 * @pmd: pmd pointer for IPA
102 * Function clears a PMD entry, flushes addr 1st and 2nd stage TLBs. Marks all
103 * pages in the range dirty.
105 static void stage2_dissolve_pmd(struct kvm *kvm, phys_addr_t addr, pmd_t *pmd)
107 if (!pmd_thp_or_huge(*pmd))
111 kvm_tlb_flush_vmid_ipa(kvm, addr);
112 put_page(virt_to_page(pmd));
115 static int mmu_topup_memory_cache(struct kvm_mmu_memory_cache *cache,
120 BUG_ON(max > KVM_NR_MEM_OBJS);
121 if (cache->nobjs >= min)
123 while (cache->nobjs < max) {
124 page = (void *)__get_free_page(PGALLOC_GFP);
127 cache->objects[cache->nobjs++] = page;
132 static void mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc)
135 free_page((unsigned long)mc->objects[--mc->nobjs]);
138 static void *mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc)
142 BUG_ON(!mc || !mc->nobjs);
143 p = mc->objects[--mc->nobjs];
147 static void clear_stage2_pgd_entry(struct kvm *kvm, pgd_t *pgd, phys_addr_t addr)
149 pud_t *pud_table __maybe_unused = stage2_pud_offset(pgd, 0UL);
150 stage2_pgd_clear(pgd);
151 kvm_tlb_flush_vmid_ipa(kvm, addr);
152 stage2_pud_free(pud_table);
153 put_page(virt_to_page(pgd));
156 static void clear_stage2_pud_entry(struct kvm *kvm, pud_t *pud, phys_addr_t addr)
158 pmd_t *pmd_table __maybe_unused = stage2_pmd_offset(pud, 0);
159 VM_BUG_ON(stage2_pud_huge(*pud));
160 stage2_pud_clear(pud);
161 kvm_tlb_flush_vmid_ipa(kvm, addr);
162 stage2_pmd_free(pmd_table);
163 put_page(virt_to_page(pud));
166 static void clear_stage2_pmd_entry(struct kvm *kvm, pmd_t *pmd, phys_addr_t addr)
168 pte_t *pte_table = pte_offset_kernel(pmd, 0);
169 VM_BUG_ON(pmd_thp_or_huge(*pmd));
171 kvm_tlb_flush_vmid_ipa(kvm, addr);
172 pte_free_kernel(NULL, pte_table);
173 put_page(virt_to_page(pmd));
177 * Unmapping vs dcache management:
179 * If a guest maps certain memory pages as uncached, all writes will
180 * bypass the data cache and go directly to RAM. However, the CPUs
181 * can still speculate reads (not writes) and fill cache lines with
184 * Those cache lines will be *clean* cache lines though, so a
185 * clean+invalidate operation is equivalent to an invalidate
186 * operation, because no cache lines are marked dirty.
188 * Those clean cache lines could be filled prior to an uncached write
189 * by the guest, and the cache coherent IO subsystem would therefore
190 * end up writing old data to disk.
192 * This is why right after unmapping a page/section and invalidating
193 * the corresponding TLBs, we call kvm_flush_dcache_p*() to make sure
194 * the IO subsystem will never hit in the cache.
196 static void unmap_stage2_ptes(struct kvm *kvm, pmd_t *pmd,
197 phys_addr_t addr, phys_addr_t end)
199 phys_addr_t start_addr = addr;
200 pte_t *pte, *start_pte;
202 start_pte = pte = pte_offset_kernel(pmd, addr);
204 if (!pte_none(*pte)) {
205 pte_t old_pte = *pte;
207 kvm_set_pte(pte, __pte(0));
208 kvm_tlb_flush_vmid_ipa(kvm, addr);
210 /* No need to invalidate the cache for device mappings */
211 if (!kvm_is_device_pfn(pte_pfn(old_pte)))
212 kvm_flush_dcache_pte(old_pte);
214 put_page(virt_to_page(pte));
216 } while (pte++, addr += PAGE_SIZE, addr != end);
218 if (stage2_pte_table_empty(start_pte))
219 clear_stage2_pmd_entry(kvm, pmd, start_addr);
222 static void unmap_stage2_pmds(struct kvm *kvm, pud_t *pud,
223 phys_addr_t addr, phys_addr_t end)
225 phys_addr_t next, start_addr = addr;
226 pmd_t *pmd, *start_pmd;
228 start_pmd = pmd = stage2_pmd_offset(pud, addr);
230 next = stage2_pmd_addr_end(addr, end);
231 if (!pmd_none(*pmd)) {
232 if (pmd_thp_or_huge(*pmd)) {
233 pmd_t old_pmd = *pmd;
236 kvm_tlb_flush_vmid_ipa(kvm, addr);
238 kvm_flush_dcache_pmd(old_pmd);
240 put_page(virt_to_page(pmd));
242 unmap_stage2_ptes(kvm, pmd, addr, next);
245 } while (pmd++, addr = next, addr != end);
247 if (stage2_pmd_table_empty(start_pmd))
248 clear_stage2_pud_entry(kvm, pud, start_addr);
251 static void unmap_stage2_puds(struct kvm *kvm, pgd_t *pgd,
252 phys_addr_t addr, phys_addr_t end)
254 phys_addr_t next, start_addr = addr;
255 pud_t *pud, *start_pud;
257 start_pud = pud = stage2_pud_offset(pgd, addr);
259 next = stage2_pud_addr_end(addr, end);
260 if (!stage2_pud_none(*pud)) {
261 if (stage2_pud_huge(*pud)) {
262 pud_t old_pud = *pud;
264 stage2_pud_clear(pud);
265 kvm_tlb_flush_vmid_ipa(kvm, addr);
266 kvm_flush_dcache_pud(old_pud);
267 put_page(virt_to_page(pud));
269 unmap_stage2_pmds(kvm, pud, addr, next);
272 } while (pud++, addr = next, addr != end);
274 if (stage2_pud_table_empty(start_pud))
275 clear_stage2_pgd_entry(kvm, pgd, start_addr);
279 * unmap_stage2_range -- Clear stage2 page table entries to unmap a range
280 * @kvm: The VM pointer
281 * @start: The intermediate physical base address of the range to unmap
282 * @size: The size of the area to unmap
284 * Clear a range of stage-2 mappings, lowering the various ref-counts. Must
285 * be called while holding mmu_lock (unless for freeing the stage2 pgd before
286 * destroying the VM), otherwise another faulting VCPU may come in and mess
287 * with things behind our backs.
289 static void unmap_stage2_range(struct kvm *kvm, phys_addr_t start, u64 size)
292 phys_addr_t addr = start, end = start + size;
295 assert_spin_locked(&kvm->mmu_lock);
296 pgd = kvm->arch.pgd + stage2_pgd_index(addr);
298 next = stage2_pgd_addr_end(addr, end);
299 if (!stage2_pgd_none(*pgd))
300 unmap_stage2_puds(kvm, pgd, addr, next);
302 * If the range is too large, release the kvm->mmu_lock
303 * to prevent starvation and lockup detector warnings.
306 cond_resched_lock(&kvm->mmu_lock);
307 } while (pgd++, addr = next, addr != end);
310 static void stage2_flush_ptes(struct kvm *kvm, pmd_t *pmd,
311 phys_addr_t addr, phys_addr_t end)
315 pte = pte_offset_kernel(pmd, addr);
317 if (!pte_none(*pte) && !kvm_is_device_pfn(pte_pfn(*pte)))
318 kvm_flush_dcache_pte(*pte);
319 } while (pte++, addr += PAGE_SIZE, addr != end);
322 static void stage2_flush_pmds(struct kvm *kvm, pud_t *pud,
323 phys_addr_t addr, phys_addr_t end)
328 pmd = stage2_pmd_offset(pud, addr);
330 next = stage2_pmd_addr_end(addr, end);
331 if (!pmd_none(*pmd)) {
332 if (pmd_thp_or_huge(*pmd))
333 kvm_flush_dcache_pmd(*pmd);
335 stage2_flush_ptes(kvm, pmd, addr, next);
337 } while (pmd++, addr = next, addr != end);
340 static void stage2_flush_puds(struct kvm *kvm, pgd_t *pgd,
341 phys_addr_t addr, phys_addr_t end)
346 pud = stage2_pud_offset(pgd, addr);
348 next = stage2_pud_addr_end(addr, end);
349 if (!stage2_pud_none(*pud)) {
350 if (stage2_pud_huge(*pud))
351 kvm_flush_dcache_pud(*pud);
353 stage2_flush_pmds(kvm, pud, addr, next);
355 } while (pud++, addr = next, addr != end);
358 static void stage2_flush_memslot(struct kvm *kvm,
359 struct kvm_memory_slot *memslot)
361 phys_addr_t addr = memslot->base_gfn << PAGE_SHIFT;
362 phys_addr_t end = addr + PAGE_SIZE * memslot->npages;
366 pgd = kvm->arch.pgd + stage2_pgd_index(addr);
368 next = stage2_pgd_addr_end(addr, end);
369 stage2_flush_puds(kvm, pgd, addr, next);
370 } while (pgd++, addr = next, addr != end);
374 * stage2_flush_vm - Invalidate cache for pages mapped in stage 2
375 * @kvm: The struct kvm pointer
377 * Go through the stage 2 page tables and invalidate any cache lines
378 * backing memory already mapped to the VM.
380 static void stage2_flush_vm(struct kvm *kvm)
382 struct kvm_memslots *slots;
383 struct kvm_memory_slot *memslot;
386 idx = srcu_read_lock(&kvm->srcu);
387 spin_lock(&kvm->mmu_lock);
389 slots = kvm_memslots(kvm);
390 kvm_for_each_memslot(memslot, slots)
391 stage2_flush_memslot(kvm, memslot);
393 spin_unlock(&kvm->mmu_lock);
394 srcu_read_unlock(&kvm->srcu, idx);
397 static void clear_hyp_pgd_entry(pgd_t *pgd)
399 pud_t *pud_table __maybe_unused = pud_offset(pgd, 0UL);
401 pud_free(NULL, pud_table);
402 put_page(virt_to_page(pgd));
405 static void clear_hyp_pud_entry(pud_t *pud)
407 pmd_t *pmd_table __maybe_unused = pmd_offset(pud, 0);
408 VM_BUG_ON(pud_huge(*pud));
410 pmd_free(NULL, pmd_table);
411 put_page(virt_to_page(pud));
414 static void clear_hyp_pmd_entry(pmd_t *pmd)
416 pte_t *pte_table = pte_offset_kernel(pmd, 0);
417 VM_BUG_ON(pmd_thp_or_huge(*pmd));
419 pte_free_kernel(NULL, pte_table);
420 put_page(virt_to_page(pmd));
423 static void unmap_hyp_ptes(pmd_t *pmd, phys_addr_t addr, phys_addr_t end)
425 pte_t *pte, *start_pte;
427 start_pte = pte = pte_offset_kernel(pmd, addr);
429 if (!pte_none(*pte)) {
430 kvm_set_pte(pte, __pte(0));
431 put_page(virt_to_page(pte));
433 } while (pte++, addr += PAGE_SIZE, addr != end);
435 if (hyp_pte_table_empty(start_pte))
436 clear_hyp_pmd_entry(pmd);
439 static void unmap_hyp_pmds(pud_t *pud, phys_addr_t addr, phys_addr_t end)
442 pmd_t *pmd, *start_pmd;
444 start_pmd = pmd = pmd_offset(pud, addr);
446 next = pmd_addr_end(addr, end);
447 /* Hyp doesn't use huge pmds */
449 unmap_hyp_ptes(pmd, addr, next);
450 } while (pmd++, addr = next, addr != end);
452 if (hyp_pmd_table_empty(start_pmd))
453 clear_hyp_pud_entry(pud);
456 static void unmap_hyp_puds(pgd_t *pgd, phys_addr_t addr, phys_addr_t end)
459 pud_t *pud, *start_pud;
461 start_pud = pud = pud_offset(pgd, addr);
463 next = pud_addr_end(addr, end);
464 /* Hyp doesn't use huge puds */
466 unmap_hyp_pmds(pud, addr, next);
467 } while (pud++, addr = next, addr != end);
469 if (hyp_pud_table_empty(start_pud))
470 clear_hyp_pgd_entry(pgd);
473 static void unmap_hyp_range(pgd_t *pgdp, phys_addr_t start, u64 size)
476 phys_addr_t addr = start, end = start + size;
480 * We don't unmap anything from HYP, except at the hyp tear down.
481 * Hence, we don't have to invalidate the TLBs here.
483 pgd = pgdp + pgd_index(addr);
485 next = pgd_addr_end(addr, end);
487 unmap_hyp_puds(pgd, addr, next);
488 } while (pgd++, addr = next, addr != end);
492 * free_hyp_pgds - free Hyp-mode page tables
494 * Assumes hyp_pgd is a page table used strictly in Hyp-mode and
495 * therefore contains either mappings in the kernel memory area (above
496 * PAGE_OFFSET), or device mappings in the vmalloc range (from
497 * VMALLOC_START to VMALLOC_END).
499 * boot_hyp_pgd should only map two pages for the init code.
501 void free_hyp_pgds(void)
505 mutex_lock(&kvm_hyp_pgd_mutex);
508 unmap_hyp_range(boot_hyp_pgd, hyp_idmap_start, PAGE_SIZE);
509 free_pages((unsigned long)boot_hyp_pgd, hyp_pgd_order);
514 unmap_hyp_range(hyp_pgd, hyp_idmap_start, PAGE_SIZE);
515 for (addr = PAGE_OFFSET; virt_addr_valid(addr); addr += PGDIR_SIZE)
516 unmap_hyp_range(hyp_pgd, kern_hyp_va(addr), PGDIR_SIZE);
517 for (addr = VMALLOC_START; is_vmalloc_addr((void*)addr); addr += PGDIR_SIZE)
518 unmap_hyp_range(hyp_pgd, kern_hyp_va(addr), PGDIR_SIZE);
520 free_pages((unsigned long)hyp_pgd, hyp_pgd_order);
523 if (merged_hyp_pgd) {
524 clear_page(merged_hyp_pgd);
525 free_page((unsigned long)merged_hyp_pgd);
526 merged_hyp_pgd = NULL;
529 mutex_unlock(&kvm_hyp_pgd_mutex);
532 static void create_hyp_pte_mappings(pmd_t *pmd, unsigned long start,
533 unsigned long end, unsigned long pfn,
541 pte = pte_offset_kernel(pmd, addr);
542 kvm_set_pte(pte, pfn_pte(pfn, prot));
543 get_page(virt_to_page(pte));
544 kvm_flush_dcache_to_poc(pte, sizeof(*pte));
546 } while (addr += PAGE_SIZE, addr != end);
549 static int create_hyp_pmd_mappings(pud_t *pud, unsigned long start,
550 unsigned long end, unsigned long pfn,
555 unsigned long addr, next;
559 pmd = pmd_offset(pud, addr);
561 BUG_ON(pmd_sect(*pmd));
563 if (pmd_none(*pmd)) {
564 pte = pte_alloc_one_kernel(NULL, addr);
566 kvm_err("Cannot allocate Hyp pte\n");
569 pmd_populate_kernel(NULL, pmd, pte);
570 get_page(virt_to_page(pmd));
571 kvm_flush_dcache_to_poc(pmd, sizeof(*pmd));
574 next = pmd_addr_end(addr, end);
576 create_hyp_pte_mappings(pmd, addr, next, pfn, prot);
577 pfn += (next - addr) >> PAGE_SHIFT;
578 } while (addr = next, addr != end);
583 static int create_hyp_pud_mappings(pgd_t *pgd, unsigned long start,
584 unsigned long end, unsigned long pfn,
589 unsigned long addr, next;
594 pud = pud_offset(pgd, addr);
596 if (pud_none_or_clear_bad(pud)) {
597 pmd = pmd_alloc_one(NULL, addr);
599 kvm_err("Cannot allocate Hyp pmd\n");
602 pud_populate(NULL, pud, pmd);
603 get_page(virt_to_page(pud));
604 kvm_flush_dcache_to_poc(pud, sizeof(*pud));
607 next = pud_addr_end(addr, end);
608 ret = create_hyp_pmd_mappings(pud, addr, next, pfn, prot);
611 pfn += (next - addr) >> PAGE_SHIFT;
612 } while (addr = next, addr != end);
617 static int __create_hyp_mappings(pgd_t *pgdp,
618 unsigned long start, unsigned long end,
619 unsigned long pfn, pgprot_t prot)
623 unsigned long addr, next;
626 mutex_lock(&kvm_hyp_pgd_mutex);
627 addr = start & PAGE_MASK;
628 end = PAGE_ALIGN(end);
630 pgd = pgdp + pgd_index(addr);
632 if (pgd_none(*pgd)) {
633 pud = pud_alloc_one(NULL, addr);
635 kvm_err("Cannot allocate Hyp pud\n");
639 pgd_populate(NULL, pgd, pud);
640 get_page(virt_to_page(pgd));
641 kvm_flush_dcache_to_poc(pgd, sizeof(*pgd));
644 next = pgd_addr_end(addr, end);
645 err = create_hyp_pud_mappings(pgd, addr, next, pfn, prot);
648 pfn += (next - addr) >> PAGE_SHIFT;
649 } while (addr = next, addr != end);
651 mutex_unlock(&kvm_hyp_pgd_mutex);
655 static phys_addr_t kvm_kaddr_to_phys(void *kaddr)
657 if (!is_vmalloc_addr(kaddr)) {
658 BUG_ON(!virt_addr_valid(kaddr));
661 return page_to_phys(vmalloc_to_page(kaddr)) +
662 offset_in_page(kaddr);
667 * create_hyp_mappings - duplicate a kernel virtual address range in Hyp mode
668 * @from: The virtual kernel start address of the range
669 * @to: The virtual kernel end address of the range (exclusive)
670 * @prot: The protection to be applied to this range
672 * The same virtual address as the kernel virtual address is also used
673 * in Hyp-mode mapping (modulo HYP_PAGE_OFFSET) to the same underlying
676 int create_hyp_mappings(void *from, void *to, pgprot_t prot)
678 phys_addr_t phys_addr;
679 unsigned long virt_addr;
680 unsigned long start = kern_hyp_va((unsigned long)from);
681 unsigned long end = kern_hyp_va((unsigned long)to);
683 if (is_kernel_in_hyp_mode())
686 start = start & PAGE_MASK;
687 end = PAGE_ALIGN(end);
689 for (virt_addr = start; virt_addr < end; virt_addr += PAGE_SIZE) {
692 phys_addr = kvm_kaddr_to_phys(from + virt_addr - start);
693 err = __create_hyp_mappings(hyp_pgd, virt_addr,
694 virt_addr + PAGE_SIZE,
695 __phys_to_pfn(phys_addr),
705 * create_hyp_io_mappings - duplicate a kernel IO mapping into Hyp mode
706 * @from: The kernel start VA of the range
707 * @to: The kernel end VA of the range (exclusive)
708 * @phys_addr: The physical start address which gets mapped
710 * The resulting HYP VA is the same as the kernel VA, modulo
713 int create_hyp_io_mappings(void *from, void *to, phys_addr_t phys_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 /* Check for a valid kernel IO mapping */
722 if (!is_vmalloc_addr(from) || !is_vmalloc_addr(to - 1))
725 return __create_hyp_mappings(hyp_pgd, start, end,
726 __phys_to_pfn(phys_addr), PAGE_HYP_DEVICE);
730 * kvm_alloc_stage2_pgd - allocate level-1 table for stage-2 translation.
731 * @kvm: The KVM struct pointer for the VM.
733 * Allocates only the stage-2 HW PGD level table(s) (can support either full
734 * 40-bit input addresses or limited to 32-bit input addresses). Clears the
737 * Note we don't need locking here as this is only called when the VM is
738 * created, which can only be done once.
740 int kvm_alloc_stage2_pgd(struct kvm *kvm)
744 if (kvm->arch.pgd != NULL) {
745 kvm_err("kvm_arch already initialized?\n");
749 /* Allocate the HW PGD, making sure that each page gets its own refcount */
750 pgd = alloc_pages_exact(S2_PGD_SIZE, GFP_KERNEL | __GFP_ZERO);
758 static void stage2_unmap_memslot(struct kvm *kvm,
759 struct kvm_memory_slot *memslot)
761 hva_t hva = memslot->userspace_addr;
762 phys_addr_t addr = memslot->base_gfn << PAGE_SHIFT;
763 phys_addr_t size = PAGE_SIZE * memslot->npages;
764 hva_t reg_end = hva + size;
767 * A memory region could potentially cover multiple VMAs, and any holes
768 * between them, so iterate over all of them to find out if we should
771 * +--------------------------------------------+
772 * +---------------+----------------+ +----------------+
773 * | : VMA 1 | VMA 2 | | VMA 3 : |
774 * +---------------+----------------+ +----------------+
776 * +--------------------------------------------+
779 struct vm_area_struct *vma = find_vma(current->mm, hva);
780 hva_t vm_start, vm_end;
782 if (!vma || vma->vm_start >= reg_end)
786 * Take the intersection of this VMA with the memory region
788 vm_start = max(hva, vma->vm_start);
789 vm_end = min(reg_end, vma->vm_end);
791 if (!(vma->vm_flags & VM_PFNMAP)) {
792 gpa_t gpa = addr + (vm_start - memslot->userspace_addr);
793 unmap_stage2_range(kvm, gpa, vm_end - vm_start);
796 } while (hva < reg_end);
800 * stage2_unmap_vm - Unmap Stage-2 RAM mappings
801 * @kvm: The struct kvm pointer
803 * Go through the memregions and unmap any reguler RAM
804 * backing memory already mapped to the VM.
806 void stage2_unmap_vm(struct kvm *kvm)
808 struct kvm_memslots *slots;
809 struct kvm_memory_slot *memslot;
812 idx = srcu_read_lock(&kvm->srcu);
813 down_read(¤t->mm->mmap_sem);
814 spin_lock(&kvm->mmu_lock);
816 slots = kvm_memslots(kvm);
817 kvm_for_each_memslot(memslot, slots)
818 stage2_unmap_memslot(kvm, memslot);
820 spin_unlock(&kvm->mmu_lock);
821 up_read(¤t->mm->mmap_sem);
822 srcu_read_unlock(&kvm->srcu, idx);
826 * kvm_free_stage2_pgd - free all stage-2 tables
827 * @kvm: The KVM struct pointer for the VM.
829 * Walks the level-1 page table pointed to by kvm->arch.pgd and frees all
830 * underlying level-2 and level-3 tables before freeing the actual level-1 table
831 * and setting the struct pointer to NULL.
833 * Note we don't need locking here as this is only called when the VM is
834 * destroyed, which can only be done once.
836 void kvm_free_stage2_pgd(struct kvm *kvm)
838 if (kvm->arch.pgd == NULL)
841 spin_lock(&kvm->mmu_lock);
842 unmap_stage2_range(kvm, 0, KVM_PHYS_SIZE);
843 spin_unlock(&kvm->mmu_lock);
845 /* Free the HW pgd, one page at a time */
846 free_pages_exact(kvm->arch.pgd, S2_PGD_SIZE);
847 kvm->arch.pgd = NULL;
850 static pud_t *stage2_get_pud(struct kvm *kvm, struct kvm_mmu_memory_cache *cache,
856 pgd = kvm->arch.pgd + stage2_pgd_index(addr);
857 if (WARN_ON(stage2_pgd_none(*pgd))) {
860 pud = mmu_memory_cache_alloc(cache);
861 stage2_pgd_populate(pgd, pud);
862 get_page(virt_to_page(pgd));
865 return stage2_pud_offset(pgd, addr);
868 static pmd_t *stage2_get_pmd(struct kvm *kvm, struct kvm_mmu_memory_cache *cache,
874 pud = stage2_get_pud(kvm, cache, addr);
875 if (stage2_pud_none(*pud)) {
878 pmd = mmu_memory_cache_alloc(cache);
879 stage2_pud_populate(pud, pmd);
880 get_page(virt_to_page(pud));
883 return stage2_pmd_offset(pud, addr);
886 static int stage2_set_pmd_huge(struct kvm *kvm, struct kvm_mmu_memory_cache
887 *cache, phys_addr_t addr, const pmd_t *new_pmd)
891 pmd = stage2_get_pmd(kvm, cache, addr);
895 * Mapping in huge pages should only happen through a fault. If a
896 * page is merged into a transparent huge page, the individual
897 * subpages of that huge page should be unmapped through MMU
898 * notifiers before we get here.
900 * Merging of CompoundPages is not supported; they should become
901 * splitting first, unmapped, merged, and mapped back in on-demand.
903 VM_BUG_ON(pmd_present(*pmd) && pmd_pfn(*pmd) != pmd_pfn(*new_pmd));
906 if (pmd_present(old_pmd)) {
908 kvm_tlb_flush_vmid_ipa(kvm, addr);
910 get_page(virt_to_page(pmd));
913 kvm_set_pmd(pmd, *new_pmd);
917 static int stage2_set_pte(struct kvm *kvm, struct kvm_mmu_memory_cache *cache,
918 phys_addr_t addr, const pte_t *new_pte,
923 bool iomap = flags & KVM_S2PTE_FLAG_IS_IOMAP;
924 bool logging_active = flags & KVM_S2_FLAG_LOGGING_ACTIVE;
926 VM_BUG_ON(logging_active && !cache);
928 /* Create stage-2 page table mapping - Levels 0 and 1 */
929 pmd = stage2_get_pmd(kvm, cache, addr);
932 * Ignore calls from kvm_set_spte_hva for unallocated
939 * While dirty page logging - dissolve huge PMD, then continue on to
943 stage2_dissolve_pmd(kvm, addr, pmd);
945 /* Create stage-2 page mappings - Level 2 */
946 if (pmd_none(*pmd)) {
948 return 0; /* ignore calls from kvm_set_spte_hva */
949 pte = mmu_memory_cache_alloc(cache);
950 pmd_populate_kernel(NULL, pmd, pte);
951 get_page(virt_to_page(pmd));
954 pte = pte_offset_kernel(pmd, addr);
956 if (iomap && pte_present(*pte))
959 /* Create 2nd stage page table mapping - Level 3 */
961 if (pte_present(old_pte)) {
962 kvm_set_pte(pte, __pte(0));
963 kvm_tlb_flush_vmid_ipa(kvm, addr);
965 get_page(virt_to_page(pte));
968 kvm_set_pte(pte, *new_pte);
972 #ifndef __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
973 static int stage2_ptep_test_and_clear_young(pte_t *pte)
975 if (pte_young(*pte)) {
976 *pte = pte_mkold(*pte);
982 static int stage2_ptep_test_and_clear_young(pte_t *pte)
984 return __ptep_test_and_clear_young(pte);
988 static int stage2_pmdp_test_and_clear_young(pmd_t *pmd)
990 return stage2_ptep_test_and_clear_young((pte_t *)pmd);
994 * kvm_phys_addr_ioremap - map a device range to guest IPA
996 * @kvm: The KVM pointer
997 * @guest_ipa: The IPA at which to insert the mapping
998 * @pa: The physical address of the device
999 * @size: The size of the mapping
1001 int kvm_phys_addr_ioremap(struct kvm *kvm, phys_addr_t guest_ipa,
1002 phys_addr_t pa, unsigned long size, bool writable)
1004 phys_addr_t addr, end;
1007 struct kvm_mmu_memory_cache cache = { 0, };
1009 end = (guest_ipa + size + PAGE_SIZE - 1) & PAGE_MASK;
1010 pfn = __phys_to_pfn(pa);
1012 for (addr = guest_ipa; addr < end; addr += PAGE_SIZE) {
1013 pte_t pte = pfn_pte(pfn, PAGE_S2_DEVICE);
1016 pte = kvm_s2pte_mkwrite(pte);
1018 ret = mmu_topup_memory_cache(&cache, KVM_MMU_CACHE_MIN_PAGES,
1022 spin_lock(&kvm->mmu_lock);
1023 ret = stage2_set_pte(kvm, &cache, addr, &pte,
1024 KVM_S2PTE_FLAG_IS_IOMAP);
1025 spin_unlock(&kvm->mmu_lock);
1033 mmu_free_memory_cache(&cache);
1037 static bool transparent_hugepage_adjust(kvm_pfn_t *pfnp, phys_addr_t *ipap)
1039 kvm_pfn_t pfn = *pfnp;
1040 gfn_t gfn = *ipap >> PAGE_SHIFT;
1042 if (PageTransCompoundMap(pfn_to_page(pfn))) {
1045 * The address we faulted on is backed by a transparent huge
1046 * page. However, because we map the compound huge page and
1047 * not the individual tail page, we need to transfer the
1048 * refcount to the head page. We have to be careful that the
1049 * THP doesn't start to split while we are adjusting the
1052 * We are sure this doesn't happen, because mmu_notifier_retry
1053 * was successful and we are holding the mmu_lock, so if this
1054 * THP is trying to split, it will be blocked in the mmu
1055 * notifier before touching any of the pages, specifically
1056 * before being able to call __split_huge_page_refcount().
1058 * We can therefore safely transfer the refcount from PG_tail
1059 * to PG_head and switch the pfn from a tail page to the head
1062 mask = PTRS_PER_PMD - 1;
1063 VM_BUG_ON((gfn & mask) != (pfn & mask));
1066 kvm_release_pfn_clean(pfn);
1078 static bool kvm_is_write_fault(struct kvm_vcpu *vcpu)
1080 if (kvm_vcpu_trap_is_iabt(vcpu))
1083 return kvm_vcpu_dabt_iswrite(vcpu);
1087 * stage2_wp_ptes - write protect PMD range
1088 * @pmd: pointer to pmd entry
1089 * @addr: range start address
1090 * @end: range end address
1092 static void stage2_wp_ptes(pmd_t *pmd, phys_addr_t addr, phys_addr_t end)
1096 pte = pte_offset_kernel(pmd, addr);
1098 if (!pte_none(*pte)) {
1099 if (!kvm_s2pte_readonly(pte))
1100 kvm_set_s2pte_readonly(pte);
1102 } while (pte++, addr += PAGE_SIZE, addr != end);
1106 * stage2_wp_pmds - write protect PUD range
1107 * @pud: pointer to pud entry
1108 * @addr: range start address
1109 * @end: range end address
1111 static void stage2_wp_pmds(pud_t *pud, phys_addr_t addr, phys_addr_t end)
1116 pmd = stage2_pmd_offset(pud, addr);
1119 next = stage2_pmd_addr_end(addr, end);
1120 if (!pmd_none(*pmd)) {
1121 if (pmd_thp_or_huge(*pmd)) {
1122 if (!kvm_s2pmd_readonly(pmd))
1123 kvm_set_s2pmd_readonly(pmd);
1125 stage2_wp_ptes(pmd, addr, next);
1128 } while (pmd++, addr = next, addr != end);
1132 * stage2_wp_puds - write protect PGD range
1133 * @pgd: pointer to pgd entry
1134 * @addr: range start address
1135 * @end: range end address
1137 * Process PUD entries, for a huge PUD we cause a panic.
1139 static void stage2_wp_puds(pgd_t *pgd, phys_addr_t addr, phys_addr_t end)
1144 pud = stage2_pud_offset(pgd, addr);
1146 next = stage2_pud_addr_end(addr, end);
1147 if (!stage2_pud_none(*pud)) {
1148 /* TODO:PUD not supported, revisit later if supported */
1149 BUG_ON(stage2_pud_huge(*pud));
1150 stage2_wp_pmds(pud, addr, next);
1152 } while (pud++, addr = next, addr != end);
1156 * stage2_wp_range() - write protect stage2 memory region range
1157 * @kvm: The KVM pointer
1158 * @addr: Start address of range
1159 * @end: End address of range
1161 static void stage2_wp_range(struct kvm *kvm, phys_addr_t addr, phys_addr_t end)
1166 pgd = kvm->arch.pgd + stage2_pgd_index(addr);
1169 * Release kvm_mmu_lock periodically if the memory region is
1170 * large. Otherwise, we may see kernel panics with
1171 * CONFIG_DETECT_HUNG_TASK, CONFIG_LOCKUP_DETECTOR,
1172 * CONFIG_LOCKDEP. Additionally, holding the lock too long
1173 * will also starve other vCPUs.
1175 if (need_resched() || spin_needbreak(&kvm->mmu_lock))
1176 cond_resched_lock(&kvm->mmu_lock);
1178 next = stage2_pgd_addr_end(addr, end);
1179 if (stage2_pgd_present(*pgd))
1180 stage2_wp_puds(pgd, addr, next);
1181 } while (pgd++, addr = next, addr != end);
1185 * kvm_mmu_wp_memory_region() - write protect stage 2 entries for memory slot
1186 * @kvm: The KVM pointer
1187 * @slot: The memory slot to write protect
1189 * Called to start logging dirty pages after memory region
1190 * KVM_MEM_LOG_DIRTY_PAGES operation is called. After this function returns
1191 * all present PMD and PTEs are write protected in the memory region.
1192 * Afterwards read of dirty page log can be called.
1194 * Acquires kvm_mmu_lock. Called with kvm->slots_lock mutex acquired,
1195 * serializing operations for VM memory regions.
1197 void kvm_mmu_wp_memory_region(struct kvm *kvm, int slot)
1199 struct kvm_memslots *slots = kvm_memslots(kvm);
1200 struct kvm_memory_slot *memslot = id_to_memslot(slots, slot);
1201 phys_addr_t start = memslot->base_gfn << PAGE_SHIFT;
1202 phys_addr_t end = (memslot->base_gfn + memslot->npages) << PAGE_SHIFT;
1204 spin_lock(&kvm->mmu_lock);
1205 stage2_wp_range(kvm, start, end);
1206 spin_unlock(&kvm->mmu_lock);
1207 kvm_flush_remote_tlbs(kvm);
1211 * kvm_mmu_write_protect_pt_masked() - write protect dirty pages
1212 * @kvm: The KVM pointer
1213 * @slot: The memory slot associated with mask
1214 * @gfn_offset: The gfn offset in memory slot
1215 * @mask: The mask of dirty pages at offset 'gfn_offset' in this memory
1216 * slot to be write protected
1218 * Walks bits set in mask write protects the associated pte's. Caller must
1219 * acquire kvm_mmu_lock.
1221 static void kvm_mmu_write_protect_pt_masked(struct kvm *kvm,
1222 struct kvm_memory_slot *slot,
1223 gfn_t gfn_offset, unsigned long mask)
1225 phys_addr_t base_gfn = slot->base_gfn + gfn_offset;
1226 phys_addr_t start = (base_gfn + __ffs(mask)) << PAGE_SHIFT;
1227 phys_addr_t end = (base_gfn + __fls(mask) + 1) << PAGE_SHIFT;
1229 stage2_wp_range(kvm, start, end);
1233 * kvm_arch_mmu_enable_log_dirty_pt_masked - enable dirty logging for selected
1236 * It calls kvm_mmu_write_protect_pt_masked to write protect selected pages to
1237 * enable dirty logging for them.
1239 void kvm_arch_mmu_enable_log_dirty_pt_masked(struct kvm *kvm,
1240 struct kvm_memory_slot *slot,
1241 gfn_t gfn_offset, unsigned long mask)
1243 kvm_mmu_write_protect_pt_masked(kvm, slot, gfn_offset, mask);
1246 static void coherent_cache_guest_page(struct kvm_vcpu *vcpu, kvm_pfn_t pfn,
1249 __coherent_cache_guest_page(vcpu, pfn, size);
1252 static int user_mem_abort(struct kvm_vcpu *vcpu, phys_addr_t fault_ipa,
1253 struct kvm_memory_slot *memslot, unsigned long hva,
1254 unsigned long fault_status)
1257 bool write_fault, writable, hugetlb = false, force_pte = false;
1258 unsigned long mmu_seq;
1259 gfn_t gfn = fault_ipa >> PAGE_SHIFT;
1260 struct kvm *kvm = vcpu->kvm;
1261 struct kvm_mmu_memory_cache *memcache = &vcpu->arch.mmu_page_cache;
1262 struct vm_area_struct *vma;
1264 pgprot_t mem_type = PAGE_S2;
1265 bool logging_active = memslot_is_logging(memslot);
1266 unsigned long flags = 0;
1268 write_fault = kvm_is_write_fault(vcpu);
1269 if (fault_status == FSC_PERM && !write_fault) {
1270 kvm_err("Unexpected L2 read permission error\n");
1274 /* Let's check if we will get back a huge page backed by hugetlbfs */
1275 down_read(¤t->mm->mmap_sem);
1276 vma = find_vma_intersection(current->mm, hva, hva + 1);
1277 if (unlikely(!vma)) {
1278 kvm_err("Failed to find VMA for hva 0x%lx\n", hva);
1279 up_read(¤t->mm->mmap_sem);
1283 if (is_vm_hugetlb_page(vma) && !logging_active) {
1285 gfn = (fault_ipa & PMD_MASK) >> PAGE_SHIFT;
1288 * Pages belonging to memslots that don't have the same
1289 * alignment for userspace and IPA cannot be mapped using
1290 * block descriptors even if the pages belong to a THP for
1291 * the process, because the stage-2 block descriptor will
1292 * cover more than a single THP and we loose atomicity for
1293 * unmapping, updates, and splits of the THP or other pages
1294 * in the stage-2 block range.
1296 if ((memslot->userspace_addr & ~PMD_MASK) !=
1297 ((memslot->base_gfn << PAGE_SHIFT) & ~PMD_MASK))
1300 up_read(¤t->mm->mmap_sem);
1302 /* We need minimum second+third level pages */
1303 ret = mmu_topup_memory_cache(memcache, KVM_MMU_CACHE_MIN_PAGES,
1308 mmu_seq = vcpu->kvm->mmu_notifier_seq;
1310 * Ensure the read of mmu_notifier_seq happens before we call
1311 * gfn_to_pfn_prot (which calls get_user_pages), so that we don't risk
1312 * the page we just got a reference to gets unmapped before we have a
1313 * chance to grab the mmu_lock, which ensure that if the page gets
1314 * unmapped afterwards, the call to kvm_unmap_hva will take it away
1315 * from us again properly. This smp_rmb() interacts with the smp_wmb()
1316 * in kvm_mmu_notifier_invalidate_<page|range_end>.
1320 pfn = gfn_to_pfn_prot(kvm, gfn, write_fault, &writable);
1321 if (is_error_noslot_pfn(pfn))
1324 if (kvm_is_device_pfn(pfn)) {
1325 mem_type = PAGE_S2_DEVICE;
1326 flags |= KVM_S2PTE_FLAG_IS_IOMAP;
1327 } else if (logging_active) {
1329 * Faults on pages in a memslot with logging enabled
1330 * should not be mapped with huge pages (it introduces churn
1331 * and performance degradation), so force a pte mapping.
1334 flags |= KVM_S2_FLAG_LOGGING_ACTIVE;
1337 * Only actually map the page as writable if this was a write
1344 spin_lock(&kvm->mmu_lock);
1345 if (mmu_notifier_retry(kvm, mmu_seq))
1348 if (!hugetlb && !force_pte)
1349 hugetlb = transparent_hugepage_adjust(&pfn, &fault_ipa);
1352 pmd_t new_pmd = pfn_pmd(pfn, mem_type);
1353 new_pmd = pmd_mkhuge(new_pmd);
1355 new_pmd = kvm_s2pmd_mkwrite(new_pmd);
1356 kvm_set_pfn_dirty(pfn);
1358 coherent_cache_guest_page(vcpu, pfn, PMD_SIZE);
1359 ret = stage2_set_pmd_huge(kvm, memcache, fault_ipa, &new_pmd);
1361 pte_t new_pte = pfn_pte(pfn, mem_type);
1364 new_pte = kvm_s2pte_mkwrite(new_pte);
1365 kvm_set_pfn_dirty(pfn);
1366 mark_page_dirty(kvm, gfn);
1368 coherent_cache_guest_page(vcpu, pfn, PAGE_SIZE);
1369 ret = stage2_set_pte(kvm, memcache, fault_ipa, &new_pte, flags);
1373 spin_unlock(&kvm->mmu_lock);
1374 kvm_set_pfn_accessed(pfn);
1375 kvm_release_pfn_clean(pfn);
1380 * Resolve the access fault by making the page young again.
1381 * Note that because the faulting entry is guaranteed not to be
1382 * cached in the TLB, we don't need to invalidate anything.
1383 * Only the HW Access Flag updates are supported for Stage 2 (no DBM),
1384 * so there is no need for atomic (pte|pmd)_mkyoung operations.
1386 static void handle_access_fault(struct kvm_vcpu *vcpu, phys_addr_t fault_ipa)
1391 bool pfn_valid = false;
1393 trace_kvm_access_fault(fault_ipa);
1395 spin_lock(&vcpu->kvm->mmu_lock);
1397 pmd = stage2_get_pmd(vcpu->kvm, NULL, fault_ipa);
1398 if (!pmd || pmd_none(*pmd)) /* Nothing there */
1401 if (pmd_thp_or_huge(*pmd)) { /* THP, HugeTLB */
1402 *pmd = pmd_mkyoung(*pmd);
1403 pfn = pmd_pfn(*pmd);
1408 pte = pte_offset_kernel(pmd, fault_ipa);
1409 if (pte_none(*pte)) /* Nothing there either */
1412 *pte = pte_mkyoung(*pte); /* Just a page... */
1413 pfn = pte_pfn(*pte);
1416 spin_unlock(&vcpu->kvm->mmu_lock);
1418 kvm_set_pfn_accessed(pfn);
1422 * kvm_handle_guest_abort - handles all 2nd stage aborts
1423 * @vcpu: the VCPU pointer
1424 * @run: the kvm_run structure
1426 * Any abort that gets to the host is almost guaranteed to be caused by a
1427 * missing second stage translation table entry, which can mean that either the
1428 * guest simply needs more memory and we must allocate an appropriate page or it
1429 * can mean that the guest tried to access I/O memory, which is emulated by user
1430 * space. The distinction is based on the IPA causing the fault and whether this
1431 * memory region has been registered as standard RAM by user space.
1433 int kvm_handle_guest_abort(struct kvm_vcpu *vcpu, struct kvm_run *run)
1435 unsigned long fault_status;
1436 phys_addr_t fault_ipa;
1437 struct kvm_memory_slot *memslot;
1439 bool is_iabt, write_fault, writable;
1443 is_iabt = kvm_vcpu_trap_is_iabt(vcpu);
1444 if (unlikely(!is_iabt && kvm_vcpu_dabt_isextabt(vcpu))) {
1445 kvm_inject_vabt(vcpu);
1449 fault_ipa = kvm_vcpu_get_fault_ipa(vcpu);
1451 trace_kvm_guest_fault(*vcpu_pc(vcpu), kvm_vcpu_get_hsr(vcpu),
1452 kvm_vcpu_get_hfar(vcpu), fault_ipa);
1454 /* Check the stage-2 fault is trans. fault or write fault */
1455 fault_status = kvm_vcpu_trap_get_fault_type(vcpu);
1456 if (fault_status != FSC_FAULT && fault_status != FSC_PERM &&
1457 fault_status != FSC_ACCESS) {
1458 kvm_err("Unsupported FSC: EC=%#x xFSC=%#lx ESR_EL2=%#lx\n",
1459 kvm_vcpu_trap_get_class(vcpu),
1460 (unsigned long)kvm_vcpu_trap_get_fault(vcpu),
1461 (unsigned long)kvm_vcpu_get_hsr(vcpu));
1465 idx = srcu_read_lock(&vcpu->kvm->srcu);
1467 gfn = fault_ipa >> PAGE_SHIFT;
1468 memslot = gfn_to_memslot(vcpu->kvm, gfn);
1469 hva = gfn_to_hva_memslot_prot(memslot, gfn, &writable);
1470 write_fault = kvm_is_write_fault(vcpu);
1471 if (kvm_is_error_hva(hva) || (write_fault && !writable)) {
1473 /* Prefetch Abort on I/O address */
1474 kvm_inject_pabt(vcpu, kvm_vcpu_get_hfar(vcpu));
1480 * Check for a cache maintenance operation. Since we
1481 * ended-up here, we know it is outside of any memory
1482 * slot. But we can't find out if that is for a device,
1483 * or if the guest is just being stupid. The only thing
1484 * we know for sure is that this range cannot be cached.
1486 * So let's assume that the guest is just being
1487 * cautious, and skip the instruction.
1489 if (kvm_vcpu_dabt_is_cm(vcpu)) {
1490 kvm_skip_instr(vcpu, kvm_vcpu_trap_il_is32bit(vcpu));
1496 * The IPA is reported as [MAX:12], so we need to
1497 * complement it with the bottom 12 bits from the
1498 * faulting VA. This is always 12 bits, irrespective
1501 fault_ipa |= kvm_vcpu_get_hfar(vcpu) & ((1 << 12) - 1);
1502 ret = io_mem_abort(vcpu, run, fault_ipa);
1506 /* Userspace should not be able to register out-of-bounds IPAs */
1507 VM_BUG_ON(fault_ipa >= KVM_PHYS_SIZE);
1509 if (fault_status == FSC_ACCESS) {
1510 handle_access_fault(vcpu, fault_ipa);
1515 ret = user_mem_abort(vcpu, fault_ipa, memslot, hva, fault_status);
1519 srcu_read_unlock(&vcpu->kvm->srcu, idx);
1523 static int handle_hva_to_gpa(struct kvm *kvm,
1524 unsigned long start,
1526 int (*handler)(struct kvm *kvm,
1527 gpa_t gpa, u64 size,
1531 struct kvm_memslots *slots;
1532 struct kvm_memory_slot *memslot;
1535 slots = kvm_memslots(kvm);
1537 /* we only care about the pages that the guest sees */
1538 kvm_for_each_memslot(memslot, slots) {
1539 unsigned long hva_start, hva_end;
1542 hva_start = max(start, memslot->userspace_addr);
1543 hva_end = min(end, memslot->userspace_addr +
1544 (memslot->npages << PAGE_SHIFT));
1545 if (hva_start >= hva_end)
1548 gpa = hva_to_gfn_memslot(hva_start, memslot) << PAGE_SHIFT;
1549 ret |= handler(kvm, gpa, (u64)(hva_end - hva_start), data);
1555 static int kvm_unmap_hva_handler(struct kvm *kvm, gpa_t gpa, u64 size, void *data)
1557 unmap_stage2_range(kvm, gpa, size);
1561 int kvm_unmap_hva(struct kvm *kvm, unsigned long hva)
1563 unsigned long end = hva + PAGE_SIZE;
1568 trace_kvm_unmap_hva(hva);
1569 handle_hva_to_gpa(kvm, hva, end, &kvm_unmap_hva_handler, NULL);
1573 int kvm_unmap_hva_range(struct kvm *kvm,
1574 unsigned long start, unsigned long end)
1579 trace_kvm_unmap_hva_range(start, end);
1580 handle_hva_to_gpa(kvm, start, end, &kvm_unmap_hva_handler, NULL);
1584 static int kvm_set_spte_handler(struct kvm *kvm, gpa_t gpa, u64 size, void *data)
1586 pte_t *pte = (pte_t *)data;
1588 WARN_ON(size != PAGE_SIZE);
1590 * We can always call stage2_set_pte with KVM_S2PTE_FLAG_LOGGING_ACTIVE
1591 * flag clear because MMU notifiers will have unmapped a huge PMD before
1592 * calling ->change_pte() (which in turn calls kvm_set_spte_hva()) and
1593 * therefore stage2_set_pte() never needs to clear out a huge PMD
1594 * through this calling path.
1596 stage2_set_pte(kvm, NULL, gpa, pte, 0);
1601 void kvm_set_spte_hva(struct kvm *kvm, unsigned long hva, pte_t pte)
1603 unsigned long end = hva + PAGE_SIZE;
1609 trace_kvm_set_spte_hva(hva);
1610 stage2_pte = pfn_pte(pte_pfn(pte), PAGE_S2);
1611 handle_hva_to_gpa(kvm, hva, end, &kvm_set_spte_handler, &stage2_pte);
1614 static int kvm_age_hva_handler(struct kvm *kvm, gpa_t gpa, u64 size, void *data)
1619 WARN_ON(size != PAGE_SIZE && size != PMD_SIZE);
1620 pmd = stage2_get_pmd(kvm, NULL, gpa);
1621 if (!pmd || pmd_none(*pmd)) /* Nothing there */
1624 if (pmd_thp_or_huge(*pmd)) /* THP, HugeTLB */
1625 return stage2_pmdp_test_and_clear_young(pmd);
1627 pte = pte_offset_kernel(pmd, gpa);
1631 return stage2_ptep_test_and_clear_young(pte);
1634 static int kvm_test_age_hva_handler(struct kvm *kvm, gpa_t gpa, u64 size, void *data)
1639 WARN_ON(size != PAGE_SIZE && size != PMD_SIZE);
1640 pmd = stage2_get_pmd(kvm, NULL, gpa);
1641 if (!pmd || pmd_none(*pmd)) /* Nothing there */
1644 if (pmd_thp_or_huge(*pmd)) /* THP, HugeTLB */
1645 return pmd_young(*pmd);
1647 pte = pte_offset_kernel(pmd, gpa);
1648 if (!pte_none(*pte)) /* Just a page... */
1649 return pte_young(*pte);
1654 int kvm_age_hva(struct kvm *kvm, unsigned long start, unsigned long end)
1656 trace_kvm_age_hva(start, end);
1657 return handle_hva_to_gpa(kvm, start, end, kvm_age_hva_handler, NULL);
1660 int kvm_test_age_hva(struct kvm *kvm, unsigned long hva)
1662 trace_kvm_test_age_hva(hva);
1663 return handle_hva_to_gpa(kvm, hva, hva, kvm_test_age_hva_handler, NULL);
1666 void kvm_mmu_free_memory_caches(struct kvm_vcpu *vcpu)
1668 mmu_free_memory_cache(&vcpu->arch.mmu_page_cache);
1671 phys_addr_t kvm_mmu_get_httbr(void)
1673 if (__kvm_cpu_uses_extended_idmap())
1674 return virt_to_phys(merged_hyp_pgd);
1676 return virt_to_phys(hyp_pgd);
1679 phys_addr_t kvm_get_idmap_vector(void)
1681 return hyp_idmap_vector;
1684 static int kvm_map_idmap_text(pgd_t *pgd)
1688 /* Create the idmap in the boot page tables */
1689 err = __create_hyp_mappings(pgd,
1690 hyp_idmap_start, hyp_idmap_end,
1691 __phys_to_pfn(hyp_idmap_start),
1694 kvm_err("Failed to idmap %lx-%lx\n",
1695 hyp_idmap_start, hyp_idmap_end);
1700 int kvm_mmu_init(void)
1704 hyp_idmap_start = kvm_virt_to_phys(__hyp_idmap_text_start);
1705 hyp_idmap_end = kvm_virt_to_phys(__hyp_idmap_text_end);
1706 hyp_idmap_vector = kvm_virt_to_phys(__kvm_hyp_init);
1709 * We rely on the linker script to ensure at build time that the HYP
1710 * init code does not cross a page boundary.
1712 BUG_ON((hyp_idmap_start ^ (hyp_idmap_end - 1)) & PAGE_MASK);
1714 kvm_info("IDMAP page: %lx\n", hyp_idmap_start);
1715 kvm_info("HYP VA range: %lx:%lx\n",
1716 kern_hyp_va(PAGE_OFFSET), kern_hyp_va(~0UL));
1718 if (hyp_idmap_start >= kern_hyp_va(PAGE_OFFSET) &&
1719 hyp_idmap_start < kern_hyp_va(~0UL) &&
1720 hyp_idmap_start != (unsigned long)__hyp_idmap_text_start) {
1722 * The idmap page is intersecting with the VA space,
1723 * it is not safe to continue further.
1725 kvm_err("IDMAP intersecting with HYP VA, unable to continue\n");
1730 hyp_pgd = (pgd_t *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, hyp_pgd_order);
1732 kvm_err("Hyp mode PGD not allocated\n");
1737 if (__kvm_cpu_uses_extended_idmap()) {
1738 boot_hyp_pgd = (pgd_t *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
1740 if (!boot_hyp_pgd) {
1741 kvm_err("Hyp boot PGD not allocated\n");
1746 err = kvm_map_idmap_text(boot_hyp_pgd);
1750 merged_hyp_pgd = (pgd_t *)__get_free_page(GFP_KERNEL | __GFP_ZERO);
1751 if (!merged_hyp_pgd) {
1752 kvm_err("Failed to allocate extra HYP pgd\n");
1755 __kvm_extend_hypmap(boot_hyp_pgd, hyp_pgd, merged_hyp_pgd,
1758 err = kvm_map_idmap_text(hyp_pgd);
1769 void kvm_arch_commit_memory_region(struct kvm *kvm,
1770 const struct kvm_userspace_memory_region *mem,
1771 const struct kvm_memory_slot *old,
1772 const struct kvm_memory_slot *new,
1773 enum kvm_mr_change change)
1776 * At this point memslot has been committed and there is an
1777 * allocated dirty_bitmap[], dirty pages will be be tracked while the
1778 * memory slot is write protected.
1780 if (change != KVM_MR_DELETE && mem->flags & KVM_MEM_LOG_DIRTY_PAGES)
1781 kvm_mmu_wp_memory_region(kvm, mem->slot);
1784 int kvm_arch_prepare_memory_region(struct kvm *kvm,
1785 struct kvm_memory_slot *memslot,
1786 const struct kvm_userspace_memory_region *mem,
1787 enum kvm_mr_change change)
1789 hva_t hva = mem->userspace_addr;
1790 hva_t reg_end = hva + mem->memory_size;
1791 bool writable = !(mem->flags & KVM_MEM_READONLY);
1794 if (change != KVM_MR_CREATE && change != KVM_MR_MOVE &&
1795 change != KVM_MR_FLAGS_ONLY)
1799 * Prevent userspace from creating a memory region outside of the IPA
1800 * space addressable by the KVM guest IPA space.
1802 if (memslot->base_gfn + memslot->npages >=
1803 (KVM_PHYS_SIZE >> PAGE_SHIFT))
1806 down_read(¤t->mm->mmap_sem);
1808 * A memory region could potentially cover multiple VMAs, and any holes
1809 * between them, so iterate over all of them to find out if we can map
1810 * any of them right now.
1812 * +--------------------------------------------+
1813 * +---------------+----------------+ +----------------+
1814 * | : VMA 1 | VMA 2 | | VMA 3 : |
1815 * +---------------+----------------+ +----------------+
1817 * +--------------------------------------------+
1820 struct vm_area_struct *vma = find_vma(current->mm, hva);
1821 hva_t vm_start, vm_end;
1823 if (!vma || vma->vm_start >= reg_end)
1827 * Mapping a read-only VMA is only allowed if the
1828 * memory region is configured as read-only.
1830 if (writable && !(vma->vm_flags & VM_WRITE)) {
1836 * Take the intersection of this VMA with the memory region
1838 vm_start = max(hva, vma->vm_start);
1839 vm_end = min(reg_end, vma->vm_end);
1841 if (vma->vm_flags & VM_PFNMAP) {
1842 gpa_t gpa = mem->guest_phys_addr +
1843 (vm_start - mem->userspace_addr);
1846 pa = (phys_addr_t)vma->vm_pgoff << PAGE_SHIFT;
1847 pa += vm_start - vma->vm_start;
1849 /* IO region dirty page logging not allowed */
1850 if (memslot->flags & KVM_MEM_LOG_DIRTY_PAGES) {
1855 ret = kvm_phys_addr_ioremap(kvm, gpa, pa,
1862 } while (hva < reg_end);
1864 if (change == KVM_MR_FLAGS_ONLY)
1867 spin_lock(&kvm->mmu_lock);
1869 unmap_stage2_range(kvm, mem->guest_phys_addr, mem->memory_size);
1871 stage2_flush_memslot(kvm, memslot);
1872 spin_unlock(&kvm->mmu_lock);
1874 up_read(¤t->mm->mmap_sem);
1878 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
1879 struct kvm_memory_slot *dont)
1883 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
1884 unsigned long npages)
1889 void kvm_arch_memslots_updated(struct kvm *kvm, struct kvm_memslots *slots)
1893 void kvm_arch_flush_shadow_all(struct kvm *kvm)
1895 kvm_free_stage2_pgd(kvm);
1898 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
1899 struct kvm_memory_slot *slot)
1901 gpa_t gpa = slot->base_gfn << PAGE_SHIFT;
1902 phys_addr_t size = slot->npages << PAGE_SHIFT;
1904 spin_lock(&kvm->mmu_lock);
1905 unmap_stage2_range(kvm, gpa, size);
1906 spin_unlock(&kvm->mmu_lock);
1910 * See note at ARMv7 ARM B1.14.4 (TL;DR: S/W ops are not easily virtualized).
1913 * - S/W ops are local to a CPU (not broadcast)
1914 * - We have line migration behind our back (speculation)
1915 * - System caches don't support S/W at all (damn!)
1917 * In the face of the above, the best we can do is to try and convert
1918 * S/W ops to VA ops. Because the guest is not allowed to infer the
1919 * S/W to PA mapping, it can only use S/W to nuke the whole cache,
1920 * which is a rather good thing for us.
1922 * Also, it is only used when turning caches on/off ("The expected
1923 * usage of the cache maintenance instructions that operate by set/way
1924 * is associated with the cache maintenance instructions associated
1925 * with the powerdown and powerup of caches, if this is required by
1926 * the implementation.").
1928 * We use the following policy:
1930 * - If we trap a S/W operation, we enable VM trapping to detect
1931 * caches being turned on/off, and do a full clean.
1933 * - We flush the caches on both caches being turned on and off.
1935 * - Once the caches are enabled, we stop trapping VM ops.
1937 void kvm_set_way_flush(struct kvm_vcpu *vcpu)
1939 unsigned long hcr = vcpu_get_hcr(vcpu);
1942 * If this is the first time we do a S/W operation
1943 * (i.e. HCR_TVM not set) flush the whole memory, and set the
1946 * Otherwise, rely on the VM trapping to wait for the MMU +
1947 * Caches to be turned off. At that point, we'll be able to
1948 * clean the caches again.
1950 if (!(hcr & HCR_TVM)) {
1951 trace_kvm_set_way_flush(*vcpu_pc(vcpu),
1952 vcpu_has_cache_enabled(vcpu));
1953 stage2_flush_vm(vcpu->kvm);
1954 vcpu_set_hcr(vcpu, hcr | HCR_TVM);
1958 void kvm_toggle_cache(struct kvm_vcpu *vcpu, bool was_enabled)
1960 bool now_enabled = vcpu_has_cache_enabled(vcpu);
1963 * If switching the MMU+caches on, need to invalidate the caches.
1964 * If switching it off, need to clean the caches.
1965 * Clean + invalidate does the trick always.
1967 if (now_enabled != was_enabled)
1968 stage2_flush_vm(vcpu->kvm);
1970 /* Caches are now on, stop trapping VM ops (until a S/W op) */
1972 vcpu_set_hcr(vcpu, vcpu_get_hcr(vcpu) & ~HCR_TVM);
1974 trace_kvm_toggle_cache(*vcpu_pc(vcpu), was_enabled, now_enabled);