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>
34 extern char __hyp_idmap_text_start[], __hyp_idmap_text_end[];
36 static pgd_t *boot_hyp_pgd;
37 static pgd_t *hyp_pgd;
38 static DEFINE_MUTEX(kvm_hyp_pgd_mutex);
40 static void *init_bounce_page;
41 static unsigned long hyp_idmap_start;
42 static unsigned long hyp_idmap_end;
43 static phys_addr_t hyp_idmap_vector;
45 #define kvm_pmd_huge(_x) (pmd_huge(_x) || pmd_trans_huge(_x))
47 static void kvm_tlb_flush_vmid_ipa(struct kvm *kvm, phys_addr_t ipa)
50 * This function also gets called when dealing with HYP page
51 * tables. As HYP doesn't have an associated struct kvm (and
52 * the HYP page tables are fairly static), we don't do
56 kvm_call_hyp(__kvm_tlb_flush_vmid_ipa, kvm, ipa);
59 static int mmu_topup_memory_cache(struct kvm_mmu_memory_cache *cache,
64 BUG_ON(max > KVM_NR_MEM_OBJS);
65 if (cache->nobjs >= min)
67 while (cache->nobjs < max) {
68 page = (void *)__get_free_page(PGALLOC_GFP);
71 cache->objects[cache->nobjs++] = page;
76 static void mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc)
79 free_page((unsigned long)mc->objects[--mc->nobjs]);
82 static void *mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc)
86 BUG_ON(!mc || !mc->nobjs);
87 p = mc->objects[--mc->nobjs];
91 static bool page_empty(void *ptr)
93 struct page *ptr_page = virt_to_page(ptr);
94 return page_count(ptr_page) == 1;
97 static void clear_pud_entry(struct kvm *kvm, pud_t *pud, phys_addr_t addr)
101 kvm_tlb_flush_vmid_ipa(kvm, addr);
103 pmd_t *pmd_table = pmd_offset(pud, 0);
105 kvm_tlb_flush_vmid_ipa(kvm, addr);
106 pmd_free(NULL, pmd_table);
108 put_page(virt_to_page(pud));
111 static void clear_pmd_entry(struct kvm *kvm, pmd_t *pmd, phys_addr_t addr)
113 if (kvm_pmd_huge(*pmd)) {
115 kvm_tlb_flush_vmid_ipa(kvm, addr);
117 pte_t *pte_table = pte_offset_kernel(pmd, 0);
119 kvm_tlb_flush_vmid_ipa(kvm, addr);
120 pte_free_kernel(NULL, pte_table);
122 put_page(virt_to_page(pmd));
125 static void clear_pte_entry(struct kvm *kvm, pte_t *pte, phys_addr_t addr)
127 if (pte_present(*pte)) {
128 kvm_set_pte(pte, __pte(0));
129 put_page(virt_to_page(pte));
130 kvm_tlb_flush_vmid_ipa(kvm, addr);
134 static void unmap_range(struct kvm *kvm, pgd_t *pgdp,
135 unsigned long long start, u64 size)
141 unsigned long long addr = start, end = start + size;
145 pgd = pgdp + pgd_index(addr);
146 pud = pud_offset(pgd, addr);
147 if (pud_none(*pud)) {
148 addr = kvm_pud_addr_end(addr, end);
152 if (pud_huge(*pud)) {
154 * If we are dealing with a huge pud, just clear it and
157 clear_pud_entry(kvm, pud, addr);
158 addr = kvm_pud_addr_end(addr, end);
162 pmd = pmd_offset(pud, addr);
163 if (pmd_none(*pmd)) {
164 addr = kvm_pmd_addr_end(addr, end);
168 if (!kvm_pmd_huge(*pmd)) {
169 pte = pte_offset_kernel(pmd, addr);
170 clear_pte_entry(kvm, pte, addr);
171 next = addr + PAGE_SIZE;
175 * If the pmd entry is to be cleared, walk back up the ladder
177 if (kvm_pmd_huge(*pmd) || page_empty(pte)) {
178 clear_pmd_entry(kvm, pmd, addr);
179 next = kvm_pmd_addr_end(addr, end);
180 if (page_empty(pmd) && !page_empty(pud)) {
181 clear_pud_entry(kvm, pud, addr);
182 next = kvm_pud_addr_end(addr, end);
190 static void stage2_flush_ptes(struct kvm *kvm, pmd_t *pmd,
191 phys_addr_t addr, phys_addr_t end)
195 pte = pte_offset_kernel(pmd, addr);
197 if (!pte_none(*pte)) {
198 hva_t hva = gfn_to_hva(kvm, addr >> PAGE_SHIFT);
199 kvm_flush_dcache_to_poc((void*)hva, PAGE_SIZE);
201 } while (pte++, addr += PAGE_SIZE, addr != end);
204 static void stage2_flush_pmds(struct kvm *kvm, pud_t *pud,
205 phys_addr_t addr, phys_addr_t end)
210 pmd = pmd_offset(pud, addr);
212 next = kvm_pmd_addr_end(addr, end);
213 if (!pmd_none(*pmd)) {
214 if (kvm_pmd_huge(*pmd)) {
215 hva_t hva = gfn_to_hva(kvm, addr >> PAGE_SHIFT);
216 kvm_flush_dcache_to_poc((void*)hva, PMD_SIZE);
218 stage2_flush_ptes(kvm, pmd, addr, next);
221 } while (pmd++, addr = next, addr != end);
224 static void stage2_flush_puds(struct kvm *kvm, pgd_t *pgd,
225 phys_addr_t addr, phys_addr_t end)
230 pud = pud_offset(pgd, addr);
232 next = kvm_pud_addr_end(addr, end);
233 if (!pud_none(*pud)) {
234 if (pud_huge(*pud)) {
235 hva_t hva = gfn_to_hva(kvm, addr >> PAGE_SHIFT);
236 kvm_flush_dcache_to_poc((void*)hva, PUD_SIZE);
238 stage2_flush_pmds(kvm, pud, addr, next);
241 } while (pud++, addr = next, addr != end);
244 static void stage2_flush_memslot(struct kvm *kvm,
245 struct kvm_memory_slot *memslot)
247 phys_addr_t addr = memslot->base_gfn << PAGE_SHIFT;
248 phys_addr_t end = addr + PAGE_SIZE * memslot->npages;
252 pgd = kvm->arch.pgd + pgd_index(addr);
254 next = kvm_pgd_addr_end(addr, end);
255 stage2_flush_puds(kvm, pgd, addr, next);
256 } while (pgd++, addr = next, addr != end);
260 * stage2_flush_vm - Invalidate cache for pages mapped in stage 2
261 * @kvm: The struct kvm pointer
263 * Go through the stage 2 page tables and invalidate any cache lines
264 * backing memory already mapped to the VM.
266 void stage2_flush_vm(struct kvm *kvm)
268 struct kvm_memslots *slots;
269 struct kvm_memory_slot *memslot;
272 idx = srcu_read_lock(&kvm->srcu);
273 spin_lock(&kvm->mmu_lock);
275 slots = kvm_memslots(kvm);
276 kvm_for_each_memslot(memslot, slots)
277 stage2_flush_memslot(kvm, memslot);
279 spin_unlock(&kvm->mmu_lock);
280 srcu_read_unlock(&kvm->srcu, idx);
284 * free_boot_hyp_pgd - free HYP boot page tables
286 * Free the HYP boot page tables. The bounce page is also freed.
288 void free_boot_hyp_pgd(void)
290 mutex_lock(&kvm_hyp_pgd_mutex);
293 unmap_range(NULL, boot_hyp_pgd, hyp_idmap_start, PAGE_SIZE);
294 unmap_range(NULL, boot_hyp_pgd, TRAMPOLINE_VA, PAGE_SIZE);
300 unmap_range(NULL, hyp_pgd, TRAMPOLINE_VA, PAGE_SIZE);
302 kfree(init_bounce_page);
303 init_bounce_page = NULL;
305 mutex_unlock(&kvm_hyp_pgd_mutex);
309 * free_hyp_pgds - free Hyp-mode page tables
311 * Assumes hyp_pgd is a page table used strictly in Hyp-mode and
312 * therefore contains either mappings in the kernel memory area (above
313 * PAGE_OFFSET), or device mappings in the vmalloc range (from
314 * VMALLOC_START to VMALLOC_END).
316 * boot_hyp_pgd should only map two pages for the init code.
318 void free_hyp_pgds(void)
324 mutex_lock(&kvm_hyp_pgd_mutex);
327 for (addr = PAGE_OFFSET; virt_addr_valid(addr); addr += PGDIR_SIZE)
328 unmap_range(NULL, hyp_pgd, KERN_TO_HYP(addr), PGDIR_SIZE);
329 for (addr = VMALLOC_START; is_vmalloc_addr((void*)addr); addr += PGDIR_SIZE)
330 unmap_range(NULL, hyp_pgd, KERN_TO_HYP(addr), PGDIR_SIZE);
336 mutex_unlock(&kvm_hyp_pgd_mutex);
339 static void create_hyp_pte_mappings(pmd_t *pmd, unsigned long start,
340 unsigned long end, unsigned long pfn,
348 pte = pte_offset_kernel(pmd, addr);
349 kvm_set_pte(pte, pfn_pte(pfn, prot));
350 get_page(virt_to_page(pte));
351 kvm_flush_dcache_to_poc(pte, sizeof(*pte));
353 } while (addr += PAGE_SIZE, addr != end);
356 static int create_hyp_pmd_mappings(pud_t *pud, unsigned long start,
357 unsigned long end, unsigned long pfn,
362 unsigned long addr, next;
366 pmd = pmd_offset(pud, addr);
368 BUG_ON(pmd_sect(*pmd));
370 if (pmd_none(*pmd)) {
371 pte = pte_alloc_one_kernel(NULL, addr);
373 kvm_err("Cannot allocate Hyp pte\n");
376 pmd_populate_kernel(NULL, pmd, pte);
377 get_page(virt_to_page(pmd));
378 kvm_flush_dcache_to_poc(pmd, sizeof(*pmd));
381 next = pmd_addr_end(addr, end);
383 create_hyp_pte_mappings(pmd, addr, next, pfn, prot);
384 pfn += (next - addr) >> PAGE_SHIFT;
385 } while (addr = next, addr != end);
390 static int __create_hyp_mappings(pgd_t *pgdp,
391 unsigned long start, unsigned long end,
392 unsigned long pfn, pgprot_t prot)
397 unsigned long addr, next;
400 mutex_lock(&kvm_hyp_pgd_mutex);
401 addr = start & PAGE_MASK;
402 end = PAGE_ALIGN(end);
404 pgd = pgdp + pgd_index(addr);
405 pud = pud_offset(pgd, addr);
407 if (pud_none_or_clear_bad(pud)) {
408 pmd = pmd_alloc_one(NULL, addr);
410 kvm_err("Cannot allocate Hyp pmd\n");
414 pud_populate(NULL, pud, pmd);
415 get_page(virt_to_page(pud));
416 kvm_flush_dcache_to_poc(pud, sizeof(*pud));
419 next = pgd_addr_end(addr, end);
420 err = create_hyp_pmd_mappings(pud, addr, next, pfn, prot);
423 pfn += (next - addr) >> PAGE_SHIFT;
424 } while (addr = next, addr != end);
426 mutex_unlock(&kvm_hyp_pgd_mutex);
430 static phys_addr_t kvm_kaddr_to_phys(void *kaddr)
432 if (!is_vmalloc_addr(kaddr)) {
433 BUG_ON(!virt_addr_valid(kaddr));
436 return page_to_phys(vmalloc_to_page(kaddr)) +
437 offset_in_page(kaddr);
442 * create_hyp_mappings - duplicate a kernel virtual address range in Hyp mode
443 * @from: The virtual kernel start address of the range
444 * @to: The virtual kernel end address of the range (exclusive)
446 * The same virtual address as the kernel virtual address is also used
447 * in Hyp-mode mapping (modulo HYP_PAGE_OFFSET) to the same underlying
450 int create_hyp_mappings(void *from, void *to)
452 phys_addr_t phys_addr;
453 unsigned long virt_addr;
454 unsigned long start = KERN_TO_HYP((unsigned long)from);
455 unsigned long end = KERN_TO_HYP((unsigned long)to);
457 start = start & PAGE_MASK;
458 end = PAGE_ALIGN(end);
460 for (virt_addr = start; virt_addr < end; virt_addr += PAGE_SIZE) {
463 phys_addr = kvm_kaddr_to_phys(from + virt_addr - start);
464 err = __create_hyp_mappings(hyp_pgd, virt_addr,
465 virt_addr + PAGE_SIZE,
466 __phys_to_pfn(phys_addr),
476 * create_hyp_io_mappings - duplicate a kernel IO mapping into Hyp mode
477 * @from: The kernel start VA of the range
478 * @to: The kernel end VA of the range (exclusive)
479 * @phys_addr: The physical start address which gets mapped
481 * The resulting HYP VA is the same as the kernel VA, modulo
484 int create_hyp_io_mappings(void *from, void *to, phys_addr_t phys_addr)
486 unsigned long start = KERN_TO_HYP((unsigned long)from);
487 unsigned long end = KERN_TO_HYP((unsigned long)to);
489 /* Check for a valid kernel IO mapping */
490 if (!is_vmalloc_addr(from) || !is_vmalloc_addr(to - 1))
493 return __create_hyp_mappings(hyp_pgd, start, end,
494 __phys_to_pfn(phys_addr), PAGE_HYP_DEVICE);
498 * kvm_alloc_stage2_pgd - allocate level-1 table for stage-2 translation.
499 * @kvm: The KVM struct pointer for the VM.
501 * Allocates the 1st level table only of size defined by S2_PGD_ORDER (can
502 * support either full 40-bit input addresses or limited to 32-bit input
503 * addresses). Clears the allocated pages.
505 * Note we don't need locking here as this is only called when the VM is
506 * created, which can only be done once.
508 int kvm_alloc_stage2_pgd(struct kvm *kvm)
512 if (kvm->arch.pgd != NULL) {
513 kvm_err("kvm_arch already initialized?\n");
517 pgd = (pgd_t *)__get_free_pages(GFP_KERNEL, S2_PGD_ORDER);
521 memset(pgd, 0, PTRS_PER_S2_PGD * sizeof(pgd_t));
529 * unmap_stage2_range -- Clear stage2 page table entries to unmap a range
530 * @kvm: The VM pointer
531 * @start: The intermediate physical base address of the range to unmap
532 * @size: The size of the area to unmap
534 * Clear a range of stage-2 mappings, lowering the various ref-counts. Must
535 * be called while holding mmu_lock (unless for freeing the stage2 pgd before
536 * destroying the VM), otherwise another faulting VCPU may come in and mess
537 * with things behind our backs.
539 static void unmap_stage2_range(struct kvm *kvm, phys_addr_t start, u64 size)
541 unmap_range(kvm, kvm->arch.pgd, start, size);
545 * kvm_free_stage2_pgd - free all stage-2 tables
546 * @kvm: The KVM struct pointer for the VM.
548 * Walks the level-1 page table pointed to by kvm->arch.pgd and frees all
549 * underlying level-2 and level-3 tables before freeing the actual level-1 table
550 * and setting the struct pointer to NULL.
552 * Note we don't need locking here as this is only called when the VM is
553 * destroyed, which can only be done once.
555 void kvm_free_stage2_pgd(struct kvm *kvm)
557 if (kvm->arch.pgd == NULL)
560 unmap_stage2_range(kvm, 0, KVM_PHYS_SIZE);
561 free_pages((unsigned long)kvm->arch.pgd, S2_PGD_ORDER);
562 kvm->arch.pgd = NULL;
565 static pmd_t *stage2_get_pmd(struct kvm *kvm, struct kvm_mmu_memory_cache *cache,
572 pgd = kvm->arch.pgd + pgd_index(addr);
573 pud = pud_offset(pgd, addr);
574 if (pud_none(*pud)) {
577 pmd = mmu_memory_cache_alloc(cache);
578 pud_populate(NULL, pud, pmd);
579 get_page(virt_to_page(pud));
582 return pmd_offset(pud, addr);
585 static int stage2_set_pmd_huge(struct kvm *kvm, struct kvm_mmu_memory_cache
586 *cache, phys_addr_t addr, const pmd_t *new_pmd)
590 pmd = stage2_get_pmd(kvm, cache, addr);
594 * Mapping in huge pages should only happen through a fault. If a
595 * page is merged into a transparent huge page, the individual
596 * subpages of that huge page should be unmapped through MMU
597 * notifiers before we get here.
599 * Merging of CompoundPages is not supported; they should become
600 * splitting first, unmapped, merged, and mapped back in on-demand.
602 VM_BUG_ON(pmd_present(*pmd) && pmd_pfn(*pmd) != pmd_pfn(*new_pmd));
605 kvm_set_pmd(pmd, *new_pmd);
606 if (pmd_present(old_pmd))
607 kvm_tlb_flush_vmid_ipa(kvm, addr);
609 get_page(virt_to_page(pmd));
613 static int stage2_set_pte(struct kvm *kvm, struct kvm_mmu_memory_cache *cache,
614 phys_addr_t addr, const pte_t *new_pte, bool iomap)
619 /* Create stage-2 page table mapping - Level 1 */
620 pmd = stage2_get_pmd(kvm, cache, addr);
623 * Ignore calls from kvm_set_spte_hva for unallocated
629 /* Create stage-2 page mappings - Level 2 */
630 if (pmd_none(*pmd)) {
632 return 0; /* ignore calls from kvm_set_spte_hva */
633 pte = mmu_memory_cache_alloc(cache);
635 pmd_populate_kernel(NULL, pmd, pte);
636 get_page(virt_to_page(pmd));
639 pte = pte_offset_kernel(pmd, addr);
641 if (iomap && pte_present(*pte))
644 /* Create 2nd stage page table mapping - Level 3 */
646 kvm_set_pte(pte, *new_pte);
647 if (pte_present(old_pte))
648 kvm_tlb_flush_vmid_ipa(kvm, addr);
650 get_page(virt_to_page(pte));
656 * kvm_phys_addr_ioremap - map a device range to guest IPA
658 * @kvm: The KVM pointer
659 * @guest_ipa: The IPA at which to insert the mapping
660 * @pa: The physical address of the device
661 * @size: The size of the mapping
663 int kvm_phys_addr_ioremap(struct kvm *kvm, phys_addr_t guest_ipa,
664 phys_addr_t pa, unsigned long size)
666 phys_addr_t addr, end;
669 struct kvm_mmu_memory_cache cache = { 0, };
671 end = (guest_ipa + size + PAGE_SIZE - 1) & PAGE_MASK;
672 pfn = __phys_to_pfn(pa);
674 for (addr = guest_ipa; addr < end; addr += PAGE_SIZE) {
675 pte_t pte = pfn_pte(pfn, PAGE_S2_DEVICE);
677 ret = mmu_topup_memory_cache(&cache, 2, 2);
680 spin_lock(&kvm->mmu_lock);
681 ret = stage2_set_pte(kvm, &cache, addr, &pte, true);
682 spin_unlock(&kvm->mmu_lock);
690 mmu_free_memory_cache(&cache);
694 static bool transparent_hugepage_adjust(pfn_t *pfnp, phys_addr_t *ipap)
697 gfn_t gfn = *ipap >> PAGE_SHIFT;
699 if (PageTransCompound(pfn_to_page(pfn))) {
702 * The address we faulted on is backed by a transparent huge
703 * page. However, because we map the compound huge page and
704 * not the individual tail page, we need to transfer the
705 * refcount to the head page. We have to be careful that the
706 * THP doesn't start to split while we are adjusting the
709 * We are sure this doesn't happen, because mmu_notifier_retry
710 * was successful and we are holding the mmu_lock, so if this
711 * THP is trying to split, it will be blocked in the mmu
712 * notifier before touching any of the pages, specifically
713 * before being able to call __split_huge_page_refcount().
715 * We can therefore safely transfer the refcount from PG_tail
716 * to PG_head and switch the pfn from a tail page to the head
719 mask = PTRS_PER_PMD - 1;
720 VM_BUG_ON((gfn & mask) != (pfn & mask));
723 kvm_release_pfn_clean(pfn);
735 static int user_mem_abort(struct kvm_vcpu *vcpu, phys_addr_t fault_ipa,
736 struct kvm_memory_slot *memslot,
737 unsigned long fault_status)
740 bool write_fault, writable, hugetlb = false, force_pte = false;
741 unsigned long mmu_seq;
742 gfn_t gfn = fault_ipa >> PAGE_SHIFT;
743 unsigned long hva = gfn_to_hva(vcpu->kvm, gfn);
744 struct kvm *kvm = vcpu->kvm;
745 struct kvm_mmu_memory_cache *memcache = &vcpu->arch.mmu_page_cache;
746 struct vm_area_struct *vma;
749 write_fault = kvm_is_write_fault(kvm_vcpu_get_hsr(vcpu));
750 if (fault_status == FSC_PERM && !write_fault) {
751 kvm_err("Unexpected L2 read permission error\n");
755 /* Let's check if we will get back a huge page backed by hugetlbfs */
756 down_read(¤t->mm->mmap_sem);
757 vma = find_vma_intersection(current->mm, hva, hva + 1);
758 if (is_vm_hugetlb_page(vma)) {
760 gfn = (fault_ipa & PMD_MASK) >> PAGE_SHIFT;
763 * Pages belonging to memslots that don't have the same
764 * alignment for userspace and IPA cannot be mapped using
765 * block descriptors even if the pages belong to a THP for
766 * the process, because the stage-2 block descriptor will
767 * cover more than a single THP and we loose atomicity for
768 * unmapping, updates, and splits of the THP or other pages
769 * in the stage-2 block range.
771 if ((memslot->userspace_addr & ~PMD_MASK) !=
772 ((memslot->base_gfn << PAGE_SHIFT) & ~PMD_MASK))
775 up_read(¤t->mm->mmap_sem);
777 /* We need minimum second+third level pages */
778 ret = mmu_topup_memory_cache(memcache, 2, KVM_NR_MEM_OBJS);
782 mmu_seq = vcpu->kvm->mmu_notifier_seq;
784 * Ensure the read of mmu_notifier_seq happens before we call
785 * gfn_to_pfn_prot (which calls get_user_pages), so that we don't risk
786 * the page we just got a reference to gets unmapped before we have a
787 * chance to grab the mmu_lock, which ensure that if the page gets
788 * unmapped afterwards, the call to kvm_unmap_hva will take it away
789 * from us again properly. This smp_rmb() interacts with the smp_wmb()
790 * in kvm_mmu_notifier_invalidate_<page|range_end>.
794 pfn = gfn_to_pfn_prot(kvm, gfn, write_fault, &writable);
795 if (is_error_pfn(pfn))
798 spin_lock(&kvm->mmu_lock);
799 if (mmu_notifier_retry(kvm, mmu_seq))
801 if (!hugetlb && !force_pte)
802 hugetlb = transparent_hugepage_adjust(&pfn, &fault_ipa);
805 pmd_t new_pmd = pfn_pmd(pfn, PAGE_S2);
806 new_pmd = pmd_mkhuge(new_pmd);
808 kvm_set_s2pmd_writable(&new_pmd);
809 kvm_set_pfn_dirty(pfn);
811 coherent_cache_guest_page(vcpu, hva & PMD_MASK, PMD_SIZE);
812 ret = stage2_set_pmd_huge(kvm, memcache, fault_ipa, &new_pmd);
814 pte_t new_pte = pfn_pte(pfn, PAGE_S2);
816 kvm_set_s2pte_writable(&new_pte);
817 kvm_set_pfn_dirty(pfn);
819 coherent_cache_guest_page(vcpu, hva, PAGE_SIZE);
820 ret = stage2_set_pte(kvm, memcache, fault_ipa, &new_pte, false);
825 spin_unlock(&kvm->mmu_lock);
826 kvm_release_pfn_clean(pfn);
831 * kvm_handle_guest_abort - handles all 2nd stage aborts
832 * @vcpu: the VCPU pointer
833 * @run: the kvm_run structure
835 * Any abort that gets to the host is almost guaranteed to be caused by a
836 * missing second stage translation table entry, which can mean that either the
837 * guest simply needs more memory and we must allocate an appropriate page or it
838 * can mean that the guest tried to access I/O memory, which is emulated by user
839 * space. The distinction is based on the IPA causing the fault and whether this
840 * memory region has been registered as standard RAM by user space.
842 int kvm_handle_guest_abort(struct kvm_vcpu *vcpu, struct kvm_run *run)
844 unsigned long fault_status;
845 phys_addr_t fault_ipa;
846 struct kvm_memory_slot *memslot;
851 is_iabt = kvm_vcpu_trap_is_iabt(vcpu);
852 fault_ipa = kvm_vcpu_get_fault_ipa(vcpu);
854 trace_kvm_guest_fault(*vcpu_pc(vcpu), kvm_vcpu_get_hsr(vcpu),
855 kvm_vcpu_get_hfar(vcpu), fault_ipa);
857 /* Check the stage-2 fault is trans. fault or write fault */
858 fault_status = kvm_vcpu_trap_get_fault(vcpu);
859 if (fault_status != FSC_FAULT && fault_status != FSC_PERM) {
860 kvm_err("Unsupported fault status: EC=%#x DFCS=%#lx\n",
861 kvm_vcpu_trap_get_class(vcpu), fault_status);
865 idx = srcu_read_lock(&vcpu->kvm->srcu);
867 gfn = fault_ipa >> PAGE_SHIFT;
868 if (!kvm_is_visible_gfn(vcpu->kvm, gfn)) {
870 /* Prefetch Abort on I/O address */
871 kvm_inject_pabt(vcpu, kvm_vcpu_get_hfar(vcpu));
876 if (fault_status != FSC_FAULT) {
877 kvm_err("Unsupported fault status on io memory: %#lx\n",
884 * The IPA is reported as [MAX:12], so we need to
885 * complement it with the bottom 12 bits from the
886 * faulting VA. This is always 12 bits, irrespective
889 fault_ipa |= kvm_vcpu_get_hfar(vcpu) & ((1 << 12) - 1);
890 ret = io_mem_abort(vcpu, run, fault_ipa);
894 memslot = gfn_to_memslot(vcpu->kvm, gfn);
896 ret = user_mem_abort(vcpu, fault_ipa, memslot, fault_status);
900 srcu_read_unlock(&vcpu->kvm->srcu, idx);
904 static void handle_hva_to_gpa(struct kvm *kvm,
907 void (*handler)(struct kvm *kvm,
908 gpa_t gpa, void *data),
911 struct kvm_memslots *slots;
912 struct kvm_memory_slot *memslot;
914 slots = kvm_memslots(kvm);
916 /* we only care about the pages that the guest sees */
917 kvm_for_each_memslot(memslot, slots) {
918 unsigned long hva_start, hva_end;
921 hva_start = max(start, memslot->userspace_addr);
922 hva_end = min(end, memslot->userspace_addr +
923 (memslot->npages << PAGE_SHIFT));
924 if (hva_start >= hva_end)
928 * {gfn(page) | page intersects with [hva_start, hva_end)} =
929 * {gfn_start, gfn_start+1, ..., gfn_end-1}.
931 gfn = hva_to_gfn_memslot(hva_start, memslot);
932 gfn_end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, memslot);
934 for (; gfn < gfn_end; ++gfn) {
935 gpa_t gpa = gfn << PAGE_SHIFT;
936 handler(kvm, gpa, data);
941 static void kvm_unmap_hva_handler(struct kvm *kvm, gpa_t gpa, void *data)
943 unmap_stage2_range(kvm, gpa, PAGE_SIZE);
946 int kvm_unmap_hva(struct kvm *kvm, unsigned long hva)
948 unsigned long end = hva + PAGE_SIZE;
953 trace_kvm_unmap_hva(hva);
954 handle_hva_to_gpa(kvm, hva, end, &kvm_unmap_hva_handler, NULL);
958 int kvm_unmap_hva_range(struct kvm *kvm,
959 unsigned long start, unsigned long end)
964 trace_kvm_unmap_hva_range(start, end);
965 handle_hva_to_gpa(kvm, start, end, &kvm_unmap_hva_handler, NULL);
969 static void kvm_set_spte_handler(struct kvm *kvm, gpa_t gpa, void *data)
971 pte_t *pte = (pte_t *)data;
973 stage2_set_pte(kvm, NULL, gpa, pte, false);
977 void kvm_set_spte_hva(struct kvm *kvm, unsigned long hva, pte_t pte)
979 unsigned long end = hva + PAGE_SIZE;
985 trace_kvm_set_spte_hva(hva);
986 stage2_pte = pfn_pte(pte_pfn(pte), PAGE_S2);
987 handle_hva_to_gpa(kvm, hva, end, &kvm_set_spte_handler, &stage2_pte);
990 void kvm_mmu_free_memory_caches(struct kvm_vcpu *vcpu)
992 mmu_free_memory_cache(&vcpu->arch.mmu_page_cache);
995 phys_addr_t kvm_mmu_get_httbr(void)
997 return virt_to_phys(hyp_pgd);
1000 phys_addr_t kvm_mmu_get_boot_httbr(void)
1002 return virt_to_phys(boot_hyp_pgd);
1005 phys_addr_t kvm_get_idmap_vector(void)
1007 return hyp_idmap_vector;
1010 int kvm_mmu_init(void)
1014 hyp_idmap_start = kvm_virt_to_phys(__hyp_idmap_text_start);
1015 hyp_idmap_end = kvm_virt_to_phys(__hyp_idmap_text_end);
1016 hyp_idmap_vector = kvm_virt_to_phys(__kvm_hyp_init);
1018 if ((hyp_idmap_start ^ hyp_idmap_end) & PAGE_MASK) {
1020 * Our init code is crossing a page boundary. Allocate
1021 * a bounce page, copy the code over and use that.
1023 size_t len = __hyp_idmap_text_end - __hyp_idmap_text_start;
1024 phys_addr_t phys_base;
1026 init_bounce_page = kmalloc(PAGE_SIZE, GFP_KERNEL);
1027 if (!init_bounce_page) {
1028 kvm_err("Couldn't allocate HYP init bounce page\n");
1033 memcpy(init_bounce_page, __hyp_idmap_text_start, len);
1035 * Warning: the code we just copied to the bounce page
1036 * must be flushed to the point of coherency.
1037 * Otherwise, the data may be sitting in L2, and HYP
1038 * mode won't be able to observe it as it runs with
1039 * caches off at that point.
1041 kvm_flush_dcache_to_poc(init_bounce_page, len);
1043 phys_base = kvm_virt_to_phys(init_bounce_page);
1044 hyp_idmap_vector += phys_base - hyp_idmap_start;
1045 hyp_idmap_start = phys_base;
1046 hyp_idmap_end = phys_base + len;
1048 kvm_info("Using HYP init bounce page @%lx\n",
1049 (unsigned long)phys_base);
1052 hyp_pgd = kzalloc(PTRS_PER_PGD * sizeof(pgd_t), GFP_KERNEL);
1053 boot_hyp_pgd = kzalloc(PTRS_PER_PGD * sizeof(pgd_t), GFP_KERNEL);
1054 if (!hyp_pgd || !boot_hyp_pgd) {
1055 kvm_err("Hyp mode PGD not allocated\n");
1060 /* Create the idmap in the boot page tables */
1061 err = __create_hyp_mappings(boot_hyp_pgd,
1062 hyp_idmap_start, hyp_idmap_end,
1063 __phys_to_pfn(hyp_idmap_start),
1067 kvm_err("Failed to idmap %lx-%lx\n",
1068 hyp_idmap_start, hyp_idmap_end);
1072 /* Map the very same page at the trampoline VA */
1073 err = __create_hyp_mappings(boot_hyp_pgd,
1074 TRAMPOLINE_VA, TRAMPOLINE_VA + PAGE_SIZE,
1075 __phys_to_pfn(hyp_idmap_start),
1078 kvm_err("Failed to map trampoline @%lx into boot HYP pgd\n",
1083 /* Map the same page again into the runtime page tables */
1084 err = __create_hyp_mappings(hyp_pgd,
1085 TRAMPOLINE_VA, TRAMPOLINE_VA + PAGE_SIZE,
1086 __phys_to_pfn(hyp_idmap_start),
1089 kvm_err("Failed to map trampoline @%lx into runtime HYP pgd\n",