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 pgd
= kvm
->arch
.pgd
+ stage2_pgd_index(addr
);
297 next
= stage2_pgd_addr_end(addr
, end
);
298 if (!stage2_pgd_none(*pgd
))
299 unmap_stage2_puds(kvm
, pgd
, addr
, next
);
300 } while (pgd
++, addr
= next
, addr
!= end
);
303 static void stage2_flush_ptes(struct kvm
*kvm
, pmd_t
*pmd
,
304 phys_addr_t addr
, phys_addr_t end
)
308 pte
= pte_offset_kernel(pmd
, addr
);
310 if (!pte_none(*pte
) && !kvm_is_device_pfn(pte_pfn(*pte
)))
311 kvm_flush_dcache_pte(*pte
);
312 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
315 static void stage2_flush_pmds(struct kvm
*kvm
, pud_t
*pud
,
316 phys_addr_t addr
, phys_addr_t end
)
321 pmd
= stage2_pmd_offset(pud
, addr
);
323 next
= stage2_pmd_addr_end(addr
, end
);
324 if (!pmd_none(*pmd
)) {
325 if (pmd_thp_or_huge(*pmd
))
326 kvm_flush_dcache_pmd(*pmd
);
328 stage2_flush_ptes(kvm
, pmd
, addr
, next
);
330 } while (pmd
++, addr
= next
, addr
!= end
);
333 static void stage2_flush_puds(struct kvm
*kvm
, pgd_t
*pgd
,
334 phys_addr_t addr
, phys_addr_t end
)
339 pud
= stage2_pud_offset(pgd
, addr
);
341 next
= stage2_pud_addr_end(addr
, end
);
342 if (!stage2_pud_none(*pud
)) {
343 if (stage2_pud_huge(*pud
))
344 kvm_flush_dcache_pud(*pud
);
346 stage2_flush_pmds(kvm
, pud
, addr
, next
);
348 } while (pud
++, addr
= next
, addr
!= end
);
351 static void stage2_flush_memslot(struct kvm
*kvm
,
352 struct kvm_memory_slot
*memslot
)
354 phys_addr_t addr
= memslot
->base_gfn
<< PAGE_SHIFT
;
355 phys_addr_t end
= addr
+ PAGE_SIZE
* memslot
->npages
;
359 pgd
= kvm
->arch
.pgd
+ stage2_pgd_index(addr
);
361 next
= stage2_pgd_addr_end(addr
, end
);
362 stage2_flush_puds(kvm
, pgd
, addr
, next
);
363 } while (pgd
++, addr
= next
, addr
!= end
);
367 * stage2_flush_vm - Invalidate cache for pages mapped in stage 2
368 * @kvm: The struct kvm pointer
370 * Go through the stage 2 page tables and invalidate any cache lines
371 * backing memory already mapped to the VM.
373 static void stage2_flush_vm(struct kvm
*kvm
)
375 struct kvm_memslots
*slots
;
376 struct kvm_memory_slot
*memslot
;
379 idx
= srcu_read_lock(&kvm
->srcu
);
380 spin_lock(&kvm
->mmu_lock
);
382 slots
= kvm_memslots(kvm
);
383 kvm_for_each_memslot(memslot
, slots
)
384 stage2_flush_memslot(kvm
, memslot
);
386 spin_unlock(&kvm
->mmu_lock
);
387 srcu_read_unlock(&kvm
->srcu
, idx
);
390 static void clear_hyp_pgd_entry(pgd_t
*pgd
)
392 pud_t
*pud_table __maybe_unused
= pud_offset(pgd
, 0UL);
394 pud_free(NULL
, pud_table
);
395 put_page(virt_to_page(pgd
));
398 static void clear_hyp_pud_entry(pud_t
*pud
)
400 pmd_t
*pmd_table __maybe_unused
= pmd_offset(pud
, 0);
401 VM_BUG_ON(pud_huge(*pud
));
403 pmd_free(NULL
, pmd_table
);
404 put_page(virt_to_page(pud
));
407 static void clear_hyp_pmd_entry(pmd_t
*pmd
)
409 pte_t
*pte_table
= pte_offset_kernel(pmd
, 0);
410 VM_BUG_ON(pmd_thp_or_huge(*pmd
));
412 pte_free_kernel(NULL
, pte_table
);
413 put_page(virt_to_page(pmd
));
416 static void unmap_hyp_ptes(pmd_t
*pmd
, phys_addr_t addr
, phys_addr_t end
)
418 pte_t
*pte
, *start_pte
;
420 start_pte
= pte
= pte_offset_kernel(pmd
, addr
);
422 if (!pte_none(*pte
)) {
423 kvm_set_pte(pte
, __pte(0));
424 put_page(virt_to_page(pte
));
426 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
428 if (hyp_pte_table_empty(start_pte
))
429 clear_hyp_pmd_entry(pmd
);
432 static void unmap_hyp_pmds(pud_t
*pud
, phys_addr_t addr
, phys_addr_t end
)
435 pmd_t
*pmd
, *start_pmd
;
437 start_pmd
= pmd
= pmd_offset(pud
, addr
);
439 next
= pmd_addr_end(addr
, end
);
440 /* Hyp doesn't use huge pmds */
442 unmap_hyp_ptes(pmd
, addr
, next
);
443 } while (pmd
++, addr
= next
, addr
!= end
);
445 if (hyp_pmd_table_empty(start_pmd
))
446 clear_hyp_pud_entry(pud
);
449 static void unmap_hyp_puds(pgd_t
*pgd
, phys_addr_t addr
, phys_addr_t end
)
452 pud_t
*pud
, *start_pud
;
454 start_pud
= pud
= pud_offset(pgd
, addr
);
456 next
= pud_addr_end(addr
, end
);
457 /* Hyp doesn't use huge puds */
459 unmap_hyp_pmds(pud
, addr
, next
);
460 } while (pud
++, addr
= next
, addr
!= end
);
462 if (hyp_pud_table_empty(start_pud
))
463 clear_hyp_pgd_entry(pgd
);
466 static void unmap_hyp_range(pgd_t
*pgdp
, phys_addr_t start
, u64 size
)
469 phys_addr_t addr
= start
, end
= start
+ size
;
473 * We don't unmap anything from HYP, except at the hyp tear down.
474 * Hence, we don't have to invalidate the TLBs here.
476 pgd
= pgdp
+ pgd_index(addr
);
478 next
= pgd_addr_end(addr
, end
);
480 unmap_hyp_puds(pgd
, addr
, next
);
481 } while (pgd
++, addr
= next
, addr
!= end
);
485 * free_hyp_pgds - free Hyp-mode page tables
487 * Assumes hyp_pgd is a page table used strictly in Hyp-mode and
488 * therefore contains either mappings in the kernel memory area (above
489 * PAGE_OFFSET), or device mappings in the vmalloc range (from
490 * VMALLOC_START to VMALLOC_END).
492 * boot_hyp_pgd should only map two pages for the init code.
494 void free_hyp_pgds(void)
498 mutex_lock(&kvm_hyp_pgd_mutex
);
501 unmap_hyp_range(boot_hyp_pgd
, hyp_idmap_start
, PAGE_SIZE
);
502 free_pages((unsigned long)boot_hyp_pgd
, hyp_pgd_order
);
507 unmap_hyp_range(hyp_pgd
, hyp_idmap_start
, PAGE_SIZE
);
508 for (addr
= PAGE_OFFSET
; virt_addr_valid(addr
); addr
+= PGDIR_SIZE
)
509 unmap_hyp_range(hyp_pgd
, kern_hyp_va(addr
), PGDIR_SIZE
);
510 for (addr
= VMALLOC_START
; is_vmalloc_addr((void*)addr
); addr
+= PGDIR_SIZE
)
511 unmap_hyp_range(hyp_pgd
, kern_hyp_va(addr
), PGDIR_SIZE
);
513 free_pages((unsigned long)hyp_pgd
, hyp_pgd_order
);
516 if (merged_hyp_pgd
) {
517 clear_page(merged_hyp_pgd
);
518 free_page((unsigned long)merged_hyp_pgd
);
519 merged_hyp_pgd
= NULL
;
522 mutex_unlock(&kvm_hyp_pgd_mutex
);
525 static void create_hyp_pte_mappings(pmd_t
*pmd
, unsigned long start
,
526 unsigned long end
, unsigned long pfn
,
534 pte
= pte_offset_kernel(pmd
, addr
);
535 kvm_set_pte(pte
, pfn_pte(pfn
, prot
));
536 get_page(virt_to_page(pte
));
537 kvm_flush_dcache_to_poc(pte
, sizeof(*pte
));
539 } while (addr
+= PAGE_SIZE
, addr
!= end
);
542 static int create_hyp_pmd_mappings(pud_t
*pud
, unsigned long start
,
543 unsigned long end
, unsigned long pfn
,
548 unsigned long addr
, next
;
552 pmd
= pmd_offset(pud
, addr
);
554 BUG_ON(pmd_sect(*pmd
));
556 if (pmd_none(*pmd
)) {
557 pte
= pte_alloc_one_kernel(NULL
, addr
);
559 kvm_err("Cannot allocate Hyp pte\n");
562 pmd_populate_kernel(NULL
, pmd
, pte
);
563 get_page(virt_to_page(pmd
));
564 kvm_flush_dcache_to_poc(pmd
, sizeof(*pmd
));
567 next
= pmd_addr_end(addr
, end
);
569 create_hyp_pte_mappings(pmd
, addr
, next
, pfn
, prot
);
570 pfn
+= (next
- addr
) >> PAGE_SHIFT
;
571 } while (addr
= next
, addr
!= end
);
576 static int create_hyp_pud_mappings(pgd_t
*pgd
, unsigned long start
,
577 unsigned long end
, unsigned long pfn
,
582 unsigned long addr
, next
;
587 pud
= pud_offset(pgd
, addr
);
589 if (pud_none_or_clear_bad(pud
)) {
590 pmd
= pmd_alloc_one(NULL
, addr
);
592 kvm_err("Cannot allocate Hyp pmd\n");
595 pud_populate(NULL
, pud
, pmd
);
596 get_page(virt_to_page(pud
));
597 kvm_flush_dcache_to_poc(pud
, sizeof(*pud
));
600 next
= pud_addr_end(addr
, end
);
601 ret
= create_hyp_pmd_mappings(pud
, addr
, next
, pfn
, prot
);
604 pfn
+= (next
- addr
) >> PAGE_SHIFT
;
605 } while (addr
= next
, addr
!= end
);
610 static int __create_hyp_mappings(pgd_t
*pgdp
,
611 unsigned long start
, unsigned long end
,
612 unsigned long pfn
, pgprot_t prot
)
616 unsigned long addr
, next
;
619 mutex_lock(&kvm_hyp_pgd_mutex
);
620 addr
= start
& PAGE_MASK
;
621 end
= PAGE_ALIGN(end
);
623 pgd
= pgdp
+ pgd_index(addr
);
625 if (pgd_none(*pgd
)) {
626 pud
= pud_alloc_one(NULL
, addr
);
628 kvm_err("Cannot allocate Hyp pud\n");
632 pgd_populate(NULL
, pgd
, pud
);
633 get_page(virt_to_page(pgd
));
634 kvm_flush_dcache_to_poc(pgd
, sizeof(*pgd
));
637 next
= pgd_addr_end(addr
, end
);
638 err
= create_hyp_pud_mappings(pgd
, addr
, next
, pfn
, prot
);
641 pfn
+= (next
- addr
) >> PAGE_SHIFT
;
642 } while (addr
= next
, addr
!= end
);
644 mutex_unlock(&kvm_hyp_pgd_mutex
);
648 static phys_addr_t
kvm_kaddr_to_phys(void *kaddr
)
650 if (!is_vmalloc_addr(kaddr
)) {
651 BUG_ON(!virt_addr_valid(kaddr
));
654 return page_to_phys(vmalloc_to_page(kaddr
)) +
655 offset_in_page(kaddr
);
660 * create_hyp_mappings - duplicate a kernel virtual address range in Hyp mode
661 * @from: The virtual kernel start address of the range
662 * @to: The virtual kernel end address of the range (exclusive)
663 * @prot: The protection to be applied to this range
665 * The same virtual address as the kernel virtual address is also used
666 * in Hyp-mode mapping (modulo HYP_PAGE_OFFSET) to the same underlying
669 int create_hyp_mappings(void *from
, void *to
, pgprot_t prot
)
671 phys_addr_t phys_addr
;
672 unsigned long virt_addr
;
673 unsigned long start
= kern_hyp_va((unsigned long)from
);
674 unsigned long end
= kern_hyp_va((unsigned long)to
);
676 if (is_kernel_in_hyp_mode())
679 start
= start
& PAGE_MASK
;
680 end
= PAGE_ALIGN(end
);
682 for (virt_addr
= start
; virt_addr
< end
; virt_addr
+= PAGE_SIZE
) {
685 phys_addr
= kvm_kaddr_to_phys(from
+ virt_addr
- start
);
686 err
= __create_hyp_mappings(hyp_pgd
, virt_addr
,
687 virt_addr
+ PAGE_SIZE
,
688 __phys_to_pfn(phys_addr
),
698 * create_hyp_io_mappings - duplicate a kernel IO mapping into Hyp mode
699 * @from: The kernel start VA of the range
700 * @to: The kernel end VA of the range (exclusive)
701 * @phys_addr: The physical start address which gets mapped
703 * The resulting HYP VA is the same as the kernel VA, modulo
706 int create_hyp_io_mappings(void *from
, void *to
, phys_addr_t phys_addr
)
708 unsigned long start
= kern_hyp_va((unsigned long)from
);
709 unsigned long end
= kern_hyp_va((unsigned long)to
);
711 if (is_kernel_in_hyp_mode())
714 /* Check for a valid kernel IO mapping */
715 if (!is_vmalloc_addr(from
) || !is_vmalloc_addr(to
- 1))
718 return __create_hyp_mappings(hyp_pgd
, start
, end
,
719 __phys_to_pfn(phys_addr
), PAGE_HYP_DEVICE
);
723 * kvm_alloc_stage2_pgd - allocate level-1 table for stage-2 translation.
724 * @kvm: The KVM struct pointer for the VM.
726 * Allocates only the stage-2 HW PGD level table(s) (can support either full
727 * 40-bit input addresses or limited to 32-bit input addresses). Clears the
730 * Note we don't need locking here as this is only called when the VM is
731 * created, which can only be done once.
733 int kvm_alloc_stage2_pgd(struct kvm
*kvm
)
737 if (kvm
->arch
.pgd
!= NULL
) {
738 kvm_err("kvm_arch already initialized?\n");
742 /* Allocate the HW PGD, making sure that each page gets its own refcount */
743 pgd
= alloc_pages_exact(S2_PGD_SIZE
, GFP_KERNEL
| __GFP_ZERO
);
752 static void stage2_unmap_memslot(struct kvm
*kvm
,
753 struct kvm_memory_slot
*memslot
)
755 hva_t hva
= memslot
->userspace_addr
;
756 phys_addr_t addr
= memslot
->base_gfn
<< PAGE_SHIFT
;
757 phys_addr_t size
= PAGE_SIZE
* memslot
->npages
;
758 hva_t reg_end
= hva
+ size
;
761 * A memory region could potentially cover multiple VMAs, and any holes
762 * between them, so iterate over all of them to find out if we should
765 * +--------------------------------------------+
766 * +---------------+----------------+ +----------------+
767 * | : VMA 1 | VMA 2 | | VMA 3 : |
768 * +---------------+----------------+ +----------------+
770 * +--------------------------------------------+
773 struct vm_area_struct
*vma
= find_vma(current
->mm
, hva
);
774 hva_t vm_start
, vm_end
;
776 if (!vma
|| vma
->vm_start
>= reg_end
)
780 * Take the intersection of this VMA with the memory region
782 vm_start
= max(hva
, vma
->vm_start
);
783 vm_end
= min(reg_end
, vma
->vm_end
);
785 if (!(vma
->vm_flags
& VM_PFNMAP
)) {
786 gpa_t gpa
= addr
+ (vm_start
- memslot
->userspace_addr
);
787 unmap_stage2_range(kvm
, gpa
, vm_end
- vm_start
);
790 } while (hva
< reg_end
);
794 * stage2_unmap_vm - Unmap Stage-2 RAM mappings
795 * @kvm: The struct kvm pointer
797 * Go through the memregions and unmap any reguler RAM
798 * backing memory already mapped to the VM.
800 void stage2_unmap_vm(struct kvm
*kvm
)
802 struct kvm_memslots
*slots
;
803 struct kvm_memory_slot
*memslot
;
806 idx
= srcu_read_lock(&kvm
->srcu
);
807 spin_lock(&kvm
->mmu_lock
);
809 slots
= kvm_memslots(kvm
);
810 kvm_for_each_memslot(memslot
, slots
)
811 stage2_unmap_memslot(kvm
, memslot
);
813 spin_unlock(&kvm
->mmu_lock
);
814 srcu_read_unlock(&kvm
->srcu
, idx
);
818 * kvm_free_stage2_pgd - free all stage-2 tables
819 * @kvm: The KVM struct pointer for the VM.
821 * Walks the level-1 page table pointed to by kvm->arch.pgd and frees all
822 * underlying level-2 and level-3 tables before freeing the actual level-1 table
823 * and setting the struct pointer to NULL.
825 * Note we don't need locking here as this is only called when the VM is
826 * destroyed, which can only be done once.
828 void kvm_free_stage2_pgd(struct kvm
*kvm
)
830 if (kvm
->arch
.pgd
== NULL
)
833 unmap_stage2_range(kvm
, 0, KVM_PHYS_SIZE
);
834 /* Free the HW pgd, one page at a time */
835 free_pages_exact(kvm
->arch
.pgd
, S2_PGD_SIZE
);
836 kvm
->arch
.pgd
= NULL
;
839 static pud_t
*stage2_get_pud(struct kvm
*kvm
, struct kvm_mmu_memory_cache
*cache
,
845 pgd
= kvm
->arch
.pgd
+ stage2_pgd_index(addr
);
846 if (WARN_ON(stage2_pgd_none(*pgd
))) {
849 pud
= mmu_memory_cache_alloc(cache
);
850 stage2_pgd_populate(pgd
, pud
);
851 get_page(virt_to_page(pgd
));
854 return stage2_pud_offset(pgd
, addr
);
857 static pmd_t
*stage2_get_pmd(struct kvm
*kvm
, struct kvm_mmu_memory_cache
*cache
,
863 pud
= stage2_get_pud(kvm
, cache
, addr
);
864 if (stage2_pud_none(*pud
)) {
867 pmd
= mmu_memory_cache_alloc(cache
);
868 stage2_pud_populate(pud
, pmd
);
869 get_page(virt_to_page(pud
));
872 return stage2_pmd_offset(pud
, addr
);
875 static int stage2_set_pmd_huge(struct kvm
*kvm
, struct kvm_mmu_memory_cache
876 *cache
, phys_addr_t addr
, const pmd_t
*new_pmd
)
880 pmd
= stage2_get_pmd(kvm
, cache
, addr
);
884 * Mapping in huge pages should only happen through a fault. If a
885 * page is merged into a transparent huge page, the individual
886 * subpages of that huge page should be unmapped through MMU
887 * notifiers before we get here.
889 * Merging of CompoundPages is not supported; they should become
890 * splitting first, unmapped, merged, and mapped back in on-demand.
892 VM_BUG_ON(pmd_present(*pmd
) && pmd_pfn(*pmd
) != pmd_pfn(*new_pmd
));
895 if (pmd_present(old_pmd
)) {
897 kvm_tlb_flush_vmid_ipa(kvm
, addr
);
899 get_page(virt_to_page(pmd
));
902 kvm_set_pmd(pmd
, *new_pmd
);
906 static int stage2_set_pte(struct kvm
*kvm
, struct kvm_mmu_memory_cache
*cache
,
907 phys_addr_t addr
, const pte_t
*new_pte
,
912 bool iomap
= flags
& KVM_S2PTE_FLAG_IS_IOMAP
;
913 bool logging_active
= flags
& KVM_S2_FLAG_LOGGING_ACTIVE
;
915 VM_BUG_ON(logging_active
&& !cache
);
917 /* Create stage-2 page table mapping - Levels 0 and 1 */
918 pmd
= stage2_get_pmd(kvm
, cache
, addr
);
921 * Ignore calls from kvm_set_spte_hva for unallocated
928 * While dirty page logging - dissolve huge PMD, then continue on to
932 stage2_dissolve_pmd(kvm
, addr
, pmd
);
934 /* Create stage-2 page mappings - Level 2 */
935 if (pmd_none(*pmd
)) {
937 return 0; /* ignore calls from kvm_set_spte_hva */
938 pte
= mmu_memory_cache_alloc(cache
);
940 pmd_populate_kernel(NULL
, pmd
, pte
);
941 get_page(virt_to_page(pmd
));
944 pte
= pte_offset_kernel(pmd
, addr
);
946 if (iomap
&& pte_present(*pte
))
949 /* Create 2nd stage page table mapping - Level 3 */
951 if (pte_present(old_pte
)) {
952 kvm_set_pte(pte
, __pte(0));
953 kvm_tlb_flush_vmid_ipa(kvm
, addr
);
955 get_page(virt_to_page(pte
));
958 kvm_set_pte(pte
, *new_pte
);
962 #ifndef __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
963 static int stage2_ptep_test_and_clear_young(pte_t
*pte
)
965 if (pte_young(*pte
)) {
966 *pte
= pte_mkold(*pte
);
972 static int stage2_ptep_test_and_clear_young(pte_t
*pte
)
974 return __ptep_test_and_clear_young(pte
);
978 static int stage2_pmdp_test_and_clear_young(pmd_t
*pmd
)
980 return stage2_ptep_test_and_clear_young((pte_t
*)pmd
);
984 * kvm_phys_addr_ioremap - map a device range to guest IPA
986 * @kvm: The KVM pointer
987 * @guest_ipa: The IPA at which to insert the mapping
988 * @pa: The physical address of the device
989 * @size: The size of the mapping
991 int kvm_phys_addr_ioremap(struct kvm
*kvm
, phys_addr_t guest_ipa
,
992 phys_addr_t pa
, unsigned long size
, bool writable
)
994 phys_addr_t addr
, end
;
997 struct kvm_mmu_memory_cache cache
= { 0, };
999 end
= (guest_ipa
+ size
+ PAGE_SIZE
- 1) & PAGE_MASK
;
1000 pfn
= __phys_to_pfn(pa
);
1002 for (addr
= guest_ipa
; addr
< end
; addr
+= PAGE_SIZE
) {
1003 pte_t pte
= pfn_pte(pfn
, PAGE_S2_DEVICE
);
1006 pte
= kvm_s2pte_mkwrite(pte
);
1008 ret
= mmu_topup_memory_cache(&cache
, KVM_MMU_CACHE_MIN_PAGES
,
1012 spin_lock(&kvm
->mmu_lock
);
1013 ret
= stage2_set_pte(kvm
, &cache
, addr
, &pte
,
1014 KVM_S2PTE_FLAG_IS_IOMAP
);
1015 spin_unlock(&kvm
->mmu_lock
);
1023 mmu_free_memory_cache(&cache
);
1027 static bool transparent_hugepage_adjust(kvm_pfn_t
*pfnp
, phys_addr_t
*ipap
)
1029 kvm_pfn_t pfn
= *pfnp
;
1030 gfn_t gfn
= *ipap
>> PAGE_SHIFT
;
1032 if (PageTransCompoundMap(pfn_to_page(pfn
))) {
1035 * The address we faulted on is backed by a transparent huge
1036 * page. However, because we map the compound huge page and
1037 * not the individual tail page, we need to transfer the
1038 * refcount to the head page. We have to be careful that the
1039 * THP doesn't start to split while we are adjusting the
1042 * We are sure this doesn't happen, because mmu_notifier_retry
1043 * was successful and we are holding the mmu_lock, so if this
1044 * THP is trying to split, it will be blocked in the mmu
1045 * notifier before touching any of the pages, specifically
1046 * before being able to call __split_huge_page_refcount().
1048 * We can therefore safely transfer the refcount from PG_tail
1049 * to PG_head and switch the pfn from a tail page to the head
1052 mask
= PTRS_PER_PMD
- 1;
1053 VM_BUG_ON((gfn
& mask
) != (pfn
& mask
));
1056 kvm_release_pfn_clean(pfn
);
1068 static bool kvm_is_write_fault(struct kvm_vcpu
*vcpu
)
1070 if (kvm_vcpu_trap_is_iabt(vcpu
))
1073 return kvm_vcpu_dabt_iswrite(vcpu
);
1077 * stage2_wp_ptes - write protect PMD range
1078 * @pmd: pointer to pmd entry
1079 * @addr: range start address
1080 * @end: range end address
1082 static void stage2_wp_ptes(pmd_t
*pmd
, phys_addr_t addr
, phys_addr_t end
)
1086 pte
= pte_offset_kernel(pmd
, addr
);
1088 if (!pte_none(*pte
)) {
1089 if (!kvm_s2pte_readonly(pte
))
1090 kvm_set_s2pte_readonly(pte
);
1092 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
1096 * stage2_wp_pmds - write protect PUD range
1097 * @pud: pointer to pud entry
1098 * @addr: range start address
1099 * @end: range end address
1101 static void stage2_wp_pmds(pud_t
*pud
, phys_addr_t addr
, phys_addr_t end
)
1106 pmd
= stage2_pmd_offset(pud
, addr
);
1109 next
= stage2_pmd_addr_end(addr
, end
);
1110 if (!pmd_none(*pmd
)) {
1111 if (pmd_thp_or_huge(*pmd
)) {
1112 if (!kvm_s2pmd_readonly(pmd
))
1113 kvm_set_s2pmd_readonly(pmd
);
1115 stage2_wp_ptes(pmd
, addr
, next
);
1118 } while (pmd
++, addr
= next
, addr
!= end
);
1122 * stage2_wp_puds - write protect PGD range
1123 * @pgd: pointer to pgd entry
1124 * @addr: range start address
1125 * @end: range end address
1127 * Process PUD entries, for a huge PUD we cause a panic.
1129 static void stage2_wp_puds(pgd_t
*pgd
, phys_addr_t addr
, phys_addr_t end
)
1134 pud
= stage2_pud_offset(pgd
, addr
);
1136 next
= stage2_pud_addr_end(addr
, end
);
1137 if (!stage2_pud_none(*pud
)) {
1138 /* TODO:PUD not supported, revisit later if supported */
1139 BUG_ON(stage2_pud_huge(*pud
));
1140 stage2_wp_pmds(pud
, addr
, next
);
1142 } while (pud
++, addr
= next
, addr
!= end
);
1146 * stage2_wp_range() - write protect stage2 memory region range
1147 * @kvm: The KVM pointer
1148 * @addr: Start address of range
1149 * @end: End address of range
1151 static void stage2_wp_range(struct kvm
*kvm
, phys_addr_t addr
, phys_addr_t end
)
1156 pgd
= kvm
->arch
.pgd
+ stage2_pgd_index(addr
);
1159 * Release kvm_mmu_lock periodically if the memory region is
1160 * large. Otherwise, we may see kernel panics with
1161 * CONFIG_DETECT_HUNG_TASK, CONFIG_LOCKUP_DETECTOR,
1162 * CONFIG_LOCKDEP. Additionally, holding the lock too long
1163 * will also starve other vCPUs.
1165 if (need_resched() || spin_needbreak(&kvm
->mmu_lock
))
1166 cond_resched_lock(&kvm
->mmu_lock
);
1168 next
= stage2_pgd_addr_end(addr
, end
);
1169 if (stage2_pgd_present(*pgd
))
1170 stage2_wp_puds(pgd
, addr
, next
);
1171 } while (pgd
++, addr
= next
, addr
!= end
);
1175 * kvm_mmu_wp_memory_region() - write protect stage 2 entries for memory slot
1176 * @kvm: The KVM pointer
1177 * @slot: The memory slot to write protect
1179 * Called to start logging dirty pages after memory region
1180 * KVM_MEM_LOG_DIRTY_PAGES operation is called. After this function returns
1181 * all present PMD and PTEs are write protected in the memory region.
1182 * Afterwards read of dirty page log can be called.
1184 * Acquires kvm_mmu_lock. Called with kvm->slots_lock mutex acquired,
1185 * serializing operations for VM memory regions.
1187 void kvm_mmu_wp_memory_region(struct kvm
*kvm
, int slot
)
1189 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
1190 struct kvm_memory_slot
*memslot
= id_to_memslot(slots
, slot
);
1191 phys_addr_t start
= memslot
->base_gfn
<< PAGE_SHIFT
;
1192 phys_addr_t end
= (memslot
->base_gfn
+ memslot
->npages
) << PAGE_SHIFT
;
1194 spin_lock(&kvm
->mmu_lock
);
1195 stage2_wp_range(kvm
, start
, end
);
1196 spin_unlock(&kvm
->mmu_lock
);
1197 kvm_flush_remote_tlbs(kvm
);
1201 * kvm_mmu_write_protect_pt_masked() - write protect dirty pages
1202 * @kvm: The KVM pointer
1203 * @slot: The memory slot associated with mask
1204 * @gfn_offset: The gfn offset in memory slot
1205 * @mask: The mask of dirty pages at offset 'gfn_offset' in this memory
1206 * slot to be write protected
1208 * Walks bits set in mask write protects the associated pte's. Caller must
1209 * acquire kvm_mmu_lock.
1211 static void kvm_mmu_write_protect_pt_masked(struct kvm
*kvm
,
1212 struct kvm_memory_slot
*slot
,
1213 gfn_t gfn_offset
, unsigned long mask
)
1215 phys_addr_t base_gfn
= slot
->base_gfn
+ gfn_offset
;
1216 phys_addr_t start
= (base_gfn
+ __ffs(mask
)) << PAGE_SHIFT
;
1217 phys_addr_t end
= (base_gfn
+ __fls(mask
) + 1) << PAGE_SHIFT
;
1219 stage2_wp_range(kvm
, start
, end
);
1223 * kvm_arch_mmu_enable_log_dirty_pt_masked - enable dirty logging for selected
1226 * It calls kvm_mmu_write_protect_pt_masked to write protect selected pages to
1227 * enable dirty logging for them.
1229 void kvm_arch_mmu_enable_log_dirty_pt_masked(struct kvm
*kvm
,
1230 struct kvm_memory_slot
*slot
,
1231 gfn_t gfn_offset
, unsigned long mask
)
1233 kvm_mmu_write_protect_pt_masked(kvm
, slot
, gfn_offset
, mask
);
1236 static void coherent_cache_guest_page(struct kvm_vcpu
*vcpu
, kvm_pfn_t pfn
,
1237 unsigned long size
, bool uncached
)
1239 __coherent_cache_guest_page(vcpu
, pfn
, size
, uncached
);
1242 static int user_mem_abort(struct kvm_vcpu
*vcpu
, phys_addr_t fault_ipa
,
1243 struct kvm_memory_slot
*memslot
, unsigned long hva
,
1244 unsigned long fault_status
)
1247 bool write_fault
, writable
, hugetlb
= false, force_pte
= false;
1248 unsigned long mmu_seq
;
1249 gfn_t gfn
= fault_ipa
>> PAGE_SHIFT
;
1250 struct kvm
*kvm
= vcpu
->kvm
;
1251 struct kvm_mmu_memory_cache
*memcache
= &vcpu
->arch
.mmu_page_cache
;
1252 struct vm_area_struct
*vma
;
1254 pgprot_t mem_type
= PAGE_S2
;
1255 bool fault_ipa_uncached
;
1256 bool logging_active
= memslot_is_logging(memslot
);
1257 unsigned long flags
= 0;
1259 write_fault
= kvm_is_write_fault(vcpu
);
1260 if (fault_status
== FSC_PERM
&& !write_fault
) {
1261 kvm_err("Unexpected L2 read permission error\n");
1265 /* Let's check if we will get back a huge page backed by hugetlbfs */
1266 down_read(¤t
->mm
->mmap_sem
);
1267 vma
= find_vma_intersection(current
->mm
, hva
, hva
+ 1);
1268 if (unlikely(!vma
)) {
1269 kvm_err("Failed to find VMA for hva 0x%lx\n", hva
);
1270 up_read(¤t
->mm
->mmap_sem
);
1274 if (is_vm_hugetlb_page(vma
) && !logging_active
) {
1276 gfn
= (fault_ipa
& PMD_MASK
) >> PAGE_SHIFT
;
1279 * Pages belonging to memslots that don't have the same
1280 * alignment for userspace and IPA cannot be mapped using
1281 * block descriptors even if the pages belong to a THP for
1282 * the process, because the stage-2 block descriptor will
1283 * cover more than a single THP and we loose atomicity for
1284 * unmapping, updates, and splits of the THP or other pages
1285 * in the stage-2 block range.
1287 if ((memslot
->userspace_addr
& ~PMD_MASK
) !=
1288 ((memslot
->base_gfn
<< PAGE_SHIFT
) & ~PMD_MASK
))
1291 up_read(¤t
->mm
->mmap_sem
);
1293 /* We need minimum second+third level pages */
1294 ret
= mmu_topup_memory_cache(memcache
, KVM_MMU_CACHE_MIN_PAGES
,
1299 mmu_seq
= vcpu
->kvm
->mmu_notifier_seq
;
1301 * Ensure the read of mmu_notifier_seq happens before we call
1302 * gfn_to_pfn_prot (which calls get_user_pages), so that we don't risk
1303 * the page we just got a reference to gets unmapped before we have a
1304 * chance to grab the mmu_lock, which ensure that if the page gets
1305 * unmapped afterwards, the call to kvm_unmap_hva will take it away
1306 * from us again properly. This smp_rmb() interacts with the smp_wmb()
1307 * in kvm_mmu_notifier_invalidate_<page|range_end>.
1311 pfn
= gfn_to_pfn_prot(kvm
, gfn
, write_fault
, &writable
);
1312 if (is_error_noslot_pfn(pfn
))
1315 if (kvm_is_device_pfn(pfn
)) {
1316 mem_type
= PAGE_S2_DEVICE
;
1317 flags
|= KVM_S2PTE_FLAG_IS_IOMAP
;
1318 } else if (logging_active
) {
1320 * Faults on pages in a memslot with logging enabled
1321 * should not be mapped with huge pages (it introduces churn
1322 * and performance degradation), so force a pte mapping.
1325 flags
|= KVM_S2_FLAG_LOGGING_ACTIVE
;
1328 * Only actually map the page as writable if this was a write
1335 spin_lock(&kvm
->mmu_lock
);
1336 if (mmu_notifier_retry(kvm
, mmu_seq
))
1339 if (!hugetlb
&& !force_pte
)
1340 hugetlb
= transparent_hugepage_adjust(&pfn
, &fault_ipa
);
1342 fault_ipa_uncached
= memslot
->flags
& KVM_MEMSLOT_INCOHERENT
;
1345 pmd_t new_pmd
= pfn_pmd(pfn
, mem_type
);
1346 new_pmd
= pmd_mkhuge(new_pmd
);
1348 new_pmd
= kvm_s2pmd_mkwrite(new_pmd
);
1349 kvm_set_pfn_dirty(pfn
);
1351 coherent_cache_guest_page(vcpu
, pfn
, PMD_SIZE
, fault_ipa_uncached
);
1352 ret
= stage2_set_pmd_huge(kvm
, memcache
, fault_ipa
, &new_pmd
);
1354 pte_t new_pte
= pfn_pte(pfn
, mem_type
);
1357 new_pte
= kvm_s2pte_mkwrite(new_pte
);
1358 kvm_set_pfn_dirty(pfn
);
1359 mark_page_dirty(kvm
, gfn
);
1361 coherent_cache_guest_page(vcpu
, pfn
, PAGE_SIZE
, fault_ipa_uncached
);
1362 ret
= stage2_set_pte(kvm
, memcache
, fault_ipa
, &new_pte
, flags
);
1366 spin_unlock(&kvm
->mmu_lock
);
1367 kvm_set_pfn_accessed(pfn
);
1368 kvm_release_pfn_clean(pfn
);
1373 * Resolve the access fault by making the page young again.
1374 * Note that because the faulting entry is guaranteed not to be
1375 * cached in the TLB, we don't need to invalidate anything.
1376 * Only the HW Access Flag updates are supported for Stage 2 (no DBM),
1377 * so there is no need for atomic (pte|pmd)_mkyoung operations.
1379 static void handle_access_fault(struct kvm_vcpu
*vcpu
, phys_addr_t fault_ipa
)
1384 bool pfn_valid
= false;
1386 trace_kvm_access_fault(fault_ipa
);
1388 spin_lock(&vcpu
->kvm
->mmu_lock
);
1390 pmd
= stage2_get_pmd(vcpu
->kvm
, NULL
, fault_ipa
);
1391 if (!pmd
|| pmd_none(*pmd
)) /* Nothing there */
1394 if (pmd_thp_or_huge(*pmd
)) { /* THP, HugeTLB */
1395 *pmd
= pmd_mkyoung(*pmd
);
1396 pfn
= pmd_pfn(*pmd
);
1401 pte
= pte_offset_kernel(pmd
, fault_ipa
);
1402 if (pte_none(*pte
)) /* Nothing there either */
1405 *pte
= pte_mkyoung(*pte
); /* Just a page... */
1406 pfn
= pte_pfn(*pte
);
1409 spin_unlock(&vcpu
->kvm
->mmu_lock
);
1411 kvm_set_pfn_accessed(pfn
);
1415 * kvm_handle_guest_abort - handles all 2nd stage aborts
1416 * @vcpu: the VCPU pointer
1417 * @run: the kvm_run structure
1419 * Any abort that gets to the host is almost guaranteed to be caused by a
1420 * missing second stage translation table entry, which can mean that either the
1421 * guest simply needs more memory and we must allocate an appropriate page or it
1422 * can mean that the guest tried to access I/O memory, which is emulated by user
1423 * space. The distinction is based on the IPA causing the fault and whether this
1424 * memory region has been registered as standard RAM by user space.
1426 int kvm_handle_guest_abort(struct kvm_vcpu
*vcpu
, struct kvm_run
*run
)
1428 unsigned long fault_status
;
1429 phys_addr_t fault_ipa
;
1430 struct kvm_memory_slot
*memslot
;
1432 bool is_iabt
, write_fault
, writable
;
1436 is_iabt
= kvm_vcpu_trap_is_iabt(vcpu
);
1437 fault_ipa
= kvm_vcpu_get_fault_ipa(vcpu
);
1439 trace_kvm_guest_fault(*vcpu_pc(vcpu
), kvm_vcpu_get_hsr(vcpu
),
1440 kvm_vcpu_get_hfar(vcpu
), fault_ipa
);
1442 /* Check the stage-2 fault is trans. fault or write fault */
1443 fault_status
= kvm_vcpu_trap_get_fault_type(vcpu
);
1444 if (fault_status
!= FSC_FAULT
&& fault_status
!= FSC_PERM
&&
1445 fault_status
!= FSC_ACCESS
) {
1446 kvm_err("Unsupported FSC: EC=%#x xFSC=%#lx ESR_EL2=%#lx\n",
1447 kvm_vcpu_trap_get_class(vcpu
),
1448 (unsigned long)kvm_vcpu_trap_get_fault(vcpu
),
1449 (unsigned long)kvm_vcpu_get_hsr(vcpu
));
1453 idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
1455 gfn
= fault_ipa
>> PAGE_SHIFT
;
1456 memslot
= gfn_to_memslot(vcpu
->kvm
, gfn
);
1457 hva
= gfn_to_hva_memslot_prot(memslot
, gfn
, &writable
);
1458 write_fault
= kvm_is_write_fault(vcpu
);
1459 if (kvm_is_error_hva(hva
) || (write_fault
&& !writable
)) {
1461 /* Prefetch Abort on I/O address */
1462 kvm_inject_pabt(vcpu
, kvm_vcpu_get_hfar(vcpu
));
1468 * Check for a cache maintenance operation. Since we
1469 * ended-up here, we know it is outside of any memory
1470 * slot. But we can't find out if that is for a device,
1471 * or if the guest is just being stupid. The only thing
1472 * we know for sure is that this range cannot be cached.
1474 * So let's assume that the guest is just being
1475 * cautious, and skip the instruction.
1477 if (kvm_vcpu_dabt_is_cm(vcpu
)) {
1478 kvm_skip_instr(vcpu
, kvm_vcpu_trap_il_is32bit(vcpu
));
1484 * The IPA is reported as [MAX:12], so we need to
1485 * complement it with the bottom 12 bits from the
1486 * faulting VA. This is always 12 bits, irrespective
1489 fault_ipa
|= kvm_vcpu_get_hfar(vcpu
) & ((1 << 12) - 1);
1490 ret
= io_mem_abort(vcpu
, run
, fault_ipa
);
1494 /* Userspace should not be able to register out-of-bounds IPAs */
1495 VM_BUG_ON(fault_ipa
>= KVM_PHYS_SIZE
);
1497 if (fault_status
== FSC_ACCESS
) {
1498 handle_access_fault(vcpu
, fault_ipa
);
1503 ret
= user_mem_abort(vcpu
, fault_ipa
, memslot
, hva
, fault_status
);
1507 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
1511 static int handle_hva_to_gpa(struct kvm
*kvm
,
1512 unsigned long start
,
1514 int (*handler
)(struct kvm
*kvm
,
1515 gpa_t gpa
, void *data
),
1518 struct kvm_memslots
*slots
;
1519 struct kvm_memory_slot
*memslot
;
1522 slots
= kvm_memslots(kvm
);
1524 /* we only care about the pages that the guest sees */
1525 kvm_for_each_memslot(memslot
, slots
) {
1526 unsigned long hva_start
, hva_end
;
1529 hva_start
= max(start
, memslot
->userspace_addr
);
1530 hva_end
= min(end
, memslot
->userspace_addr
+
1531 (memslot
->npages
<< PAGE_SHIFT
));
1532 if (hva_start
>= hva_end
)
1536 * {gfn(page) | page intersects with [hva_start, hva_end)} =
1537 * {gfn_start, gfn_start+1, ..., gfn_end-1}.
1539 gfn
= hva_to_gfn_memslot(hva_start
, memslot
);
1540 gfn_end
= hva_to_gfn_memslot(hva_end
+ PAGE_SIZE
- 1, memslot
);
1542 for (; gfn
< gfn_end
; ++gfn
) {
1543 gpa_t gpa
= gfn
<< PAGE_SHIFT
;
1544 ret
|= handler(kvm
, gpa
, data
);
1551 static int kvm_unmap_hva_handler(struct kvm
*kvm
, gpa_t gpa
, void *data
)
1553 unmap_stage2_range(kvm
, gpa
, PAGE_SIZE
);
1557 int kvm_unmap_hva(struct kvm
*kvm
, unsigned long hva
)
1559 unsigned long end
= hva
+ PAGE_SIZE
;
1564 trace_kvm_unmap_hva(hva
);
1565 handle_hva_to_gpa(kvm
, hva
, end
, &kvm_unmap_hva_handler
, NULL
);
1569 int kvm_unmap_hva_range(struct kvm
*kvm
,
1570 unsigned long start
, unsigned long end
)
1575 trace_kvm_unmap_hva_range(start
, end
);
1576 handle_hva_to_gpa(kvm
, start
, end
, &kvm_unmap_hva_handler
, NULL
);
1580 static int kvm_set_spte_handler(struct kvm
*kvm
, gpa_t gpa
, void *data
)
1582 pte_t
*pte
= (pte_t
*)data
;
1585 * We can always call stage2_set_pte with KVM_S2PTE_FLAG_LOGGING_ACTIVE
1586 * flag clear because MMU notifiers will have unmapped a huge PMD before
1587 * calling ->change_pte() (which in turn calls kvm_set_spte_hva()) and
1588 * therefore stage2_set_pte() never needs to clear out a huge PMD
1589 * through this calling path.
1591 stage2_set_pte(kvm
, NULL
, gpa
, pte
, 0);
1596 void kvm_set_spte_hva(struct kvm
*kvm
, unsigned long hva
, pte_t pte
)
1598 unsigned long end
= hva
+ PAGE_SIZE
;
1604 trace_kvm_set_spte_hva(hva
);
1605 stage2_pte
= pfn_pte(pte_pfn(pte
), PAGE_S2
);
1606 handle_hva_to_gpa(kvm
, hva
, end
, &kvm_set_spte_handler
, &stage2_pte
);
1609 static int kvm_age_hva_handler(struct kvm
*kvm
, gpa_t gpa
, void *data
)
1614 pmd
= stage2_get_pmd(kvm
, NULL
, gpa
);
1615 if (!pmd
|| pmd_none(*pmd
)) /* Nothing there */
1618 if (pmd_thp_or_huge(*pmd
)) /* THP, HugeTLB */
1619 return stage2_pmdp_test_and_clear_young(pmd
);
1621 pte
= pte_offset_kernel(pmd
, gpa
);
1625 return stage2_ptep_test_and_clear_young(pte
);
1628 static int kvm_test_age_hva_handler(struct kvm
*kvm
, gpa_t gpa
, void *data
)
1633 pmd
= stage2_get_pmd(kvm
, NULL
, gpa
);
1634 if (!pmd
|| pmd_none(*pmd
)) /* Nothing there */
1637 if (pmd_thp_or_huge(*pmd
)) /* THP, HugeTLB */
1638 return pmd_young(*pmd
);
1640 pte
= pte_offset_kernel(pmd
, gpa
);
1641 if (!pte_none(*pte
)) /* Just a page... */
1642 return pte_young(*pte
);
1647 int kvm_age_hva(struct kvm
*kvm
, unsigned long start
, unsigned long end
)
1649 trace_kvm_age_hva(start
, end
);
1650 return handle_hva_to_gpa(kvm
, start
, end
, kvm_age_hva_handler
, NULL
);
1653 int kvm_test_age_hva(struct kvm
*kvm
, unsigned long hva
)
1655 trace_kvm_test_age_hva(hva
);
1656 return handle_hva_to_gpa(kvm
, hva
, hva
, kvm_test_age_hva_handler
, NULL
);
1659 void kvm_mmu_free_memory_caches(struct kvm_vcpu
*vcpu
)
1661 mmu_free_memory_cache(&vcpu
->arch
.mmu_page_cache
);
1664 phys_addr_t
kvm_mmu_get_httbr(void)
1666 if (__kvm_cpu_uses_extended_idmap())
1667 return virt_to_phys(merged_hyp_pgd
);
1669 return virt_to_phys(hyp_pgd
);
1672 phys_addr_t
kvm_get_idmap_vector(void)
1674 return hyp_idmap_vector
;
1677 phys_addr_t
kvm_get_idmap_start(void)
1679 return hyp_idmap_start
;
1682 static int kvm_map_idmap_text(pgd_t
*pgd
)
1686 /* Create the idmap in the boot page tables */
1687 err
= __create_hyp_mappings(pgd
,
1688 hyp_idmap_start
, hyp_idmap_end
,
1689 __phys_to_pfn(hyp_idmap_start
),
1692 kvm_err("Failed to idmap %lx-%lx\n",
1693 hyp_idmap_start
, hyp_idmap_end
);
1698 int kvm_mmu_init(void)
1702 hyp_idmap_start
= kvm_virt_to_phys(__hyp_idmap_text_start
);
1703 hyp_idmap_end
= kvm_virt_to_phys(__hyp_idmap_text_end
);
1704 hyp_idmap_vector
= kvm_virt_to_phys(__kvm_hyp_init
);
1707 * We rely on the linker script to ensure at build time that the HYP
1708 * init code does not cross a page boundary.
1710 BUG_ON((hyp_idmap_start
^ (hyp_idmap_end
- 1)) & PAGE_MASK
);
1712 kvm_info("IDMAP page: %lx\n", hyp_idmap_start
);
1713 kvm_info("HYP VA range: %lx:%lx\n",
1714 kern_hyp_va(PAGE_OFFSET
), kern_hyp_va(~0UL));
1716 if (hyp_idmap_start
>= kern_hyp_va(PAGE_OFFSET
) &&
1717 hyp_idmap_start
< kern_hyp_va(~0UL) &&
1718 hyp_idmap_start
!= (unsigned long)__hyp_idmap_text_start
) {
1720 * The idmap page is intersecting with the VA space,
1721 * it is not safe to continue further.
1723 kvm_err("IDMAP intersecting with HYP VA, unable to continue\n");
1728 hyp_pgd
= (pgd_t
*)__get_free_pages(GFP_KERNEL
| __GFP_ZERO
, hyp_pgd_order
);
1730 kvm_err("Hyp mode PGD not allocated\n");
1735 if (__kvm_cpu_uses_extended_idmap()) {
1736 boot_hyp_pgd
= (pgd_t
*)__get_free_pages(GFP_KERNEL
| __GFP_ZERO
,
1738 if (!boot_hyp_pgd
) {
1739 kvm_err("Hyp boot PGD not allocated\n");
1744 err
= kvm_map_idmap_text(boot_hyp_pgd
);
1748 merged_hyp_pgd
= (pgd_t
*)__get_free_page(GFP_KERNEL
| __GFP_ZERO
);
1749 if (!merged_hyp_pgd
) {
1750 kvm_err("Failed to allocate extra HYP pgd\n");
1753 __kvm_extend_hypmap(boot_hyp_pgd
, hyp_pgd
, merged_hyp_pgd
,
1756 err
= kvm_map_idmap_text(hyp_pgd
);
1767 void kvm_arch_commit_memory_region(struct kvm
*kvm
,
1768 const struct kvm_userspace_memory_region
*mem
,
1769 const struct kvm_memory_slot
*old
,
1770 const struct kvm_memory_slot
*new,
1771 enum kvm_mr_change change
)
1774 * At this point memslot has been committed and there is an
1775 * allocated dirty_bitmap[], dirty pages will be be tracked while the
1776 * memory slot is write protected.
1778 if (change
!= KVM_MR_DELETE
&& mem
->flags
& KVM_MEM_LOG_DIRTY_PAGES
)
1779 kvm_mmu_wp_memory_region(kvm
, mem
->slot
);
1782 int kvm_arch_prepare_memory_region(struct kvm
*kvm
,
1783 struct kvm_memory_slot
*memslot
,
1784 const struct kvm_userspace_memory_region
*mem
,
1785 enum kvm_mr_change change
)
1787 hva_t hva
= mem
->userspace_addr
;
1788 hva_t reg_end
= hva
+ mem
->memory_size
;
1789 bool writable
= !(mem
->flags
& KVM_MEM_READONLY
);
1792 if (change
!= KVM_MR_CREATE
&& change
!= KVM_MR_MOVE
&&
1793 change
!= KVM_MR_FLAGS_ONLY
)
1797 * Prevent userspace from creating a memory region outside of the IPA
1798 * space addressable by the KVM guest IPA space.
1800 if (memslot
->base_gfn
+ memslot
->npages
>=
1801 (KVM_PHYS_SIZE
>> PAGE_SHIFT
))
1805 * A memory region could potentially cover multiple VMAs, and any holes
1806 * between them, so iterate over all of them to find out if we can map
1807 * any of them right now.
1809 * +--------------------------------------------+
1810 * +---------------+----------------+ +----------------+
1811 * | : VMA 1 | VMA 2 | | VMA 3 : |
1812 * +---------------+----------------+ +----------------+
1814 * +--------------------------------------------+
1817 struct vm_area_struct
*vma
= find_vma(current
->mm
, hva
);
1818 hva_t vm_start
, vm_end
;
1820 if (!vma
|| vma
->vm_start
>= reg_end
)
1824 * Mapping a read-only VMA is only allowed if the
1825 * memory region is configured as read-only.
1827 if (writable
&& !(vma
->vm_flags
& VM_WRITE
)) {
1833 * Take the intersection of this VMA with the memory region
1835 vm_start
= max(hva
, vma
->vm_start
);
1836 vm_end
= min(reg_end
, vma
->vm_end
);
1838 if (vma
->vm_flags
& VM_PFNMAP
) {
1839 gpa_t gpa
= mem
->guest_phys_addr
+
1840 (vm_start
- mem
->userspace_addr
);
1843 pa
= (phys_addr_t
)vma
->vm_pgoff
<< PAGE_SHIFT
;
1844 pa
+= vm_start
- vma
->vm_start
;
1846 /* IO region dirty page logging not allowed */
1847 if (memslot
->flags
& KVM_MEM_LOG_DIRTY_PAGES
)
1850 ret
= kvm_phys_addr_ioremap(kvm
, gpa
, pa
,
1857 } while (hva
< reg_end
);
1859 if (change
== KVM_MR_FLAGS_ONLY
)
1862 spin_lock(&kvm
->mmu_lock
);
1864 unmap_stage2_range(kvm
, mem
->guest_phys_addr
, mem
->memory_size
);
1866 stage2_flush_memslot(kvm
, memslot
);
1867 spin_unlock(&kvm
->mmu_lock
);
1871 void kvm_arch_free_memslot(struct kvm
*kvm
, struct kvm_memory_slot
*free
,
1872 struct kvm_memory_slot
*dont
)
1876 int kvm_arch_create_memslot(struct kvm
*kvm
, struct kvm_memory_slot
*slot
,
1877 unsigned long npages
)
1880 * Readonly memslots are not incoherent with the caches by definition,
1881 * but in practice, they are used mostly to emulate ROMs or NOR flashes
1882 * that the guest may consider devices and hence map as uncached.
1883 * To prevent incoherency issues in these cases, tag all readonly
1884 * regions as incoherent.
1886 if (slot
->flags
& KVM_MEM_READONLY
)
1887 slot
->flags
|= KVM_MEMSLOT_INCOHERENT
;
1891 void kvm_arch_memslots_updated(struct kvm
*kvm
, struct kvm_memslots
*slots
)
1895 void kvm_arch_flush_shadow_all(struct kvm
*kvm
)
1897 kvm_free_stage2_pgd(kvm
);
1900 void kvm_arch_flush_shadow_memslot(struct kvm
*kvm
,
1901 struct kvm_memory_slot
*slot
)
1903 gpa_t gpa
= slot
->base_gfn
<< PAGE_SHIFT
;
1904 phys_addr_t size
= slot
->npages
<< PAGE_SHIFT
;
1906 spin_lock(&kvm
->mmu_lock
);
1907 unmap_stage2_range(kvm
, gpa
, size
);
1908 spin_unlock(&kvm
->mmu_lock
);
1912 * See note at ARMv7 ARM B1.14.4 (TL;DR: S/W ops are not easily virtualized).
1915 * - S/W ops are local to a CPU (not broadcast)
1916 * - We have line migration behind our back (speculation)
1917 * - System caches don't support S/W at all (damn!)
1919 * In the face of the above, the best we can do is to try and convert
1920 * S/W ops to VA ops. Because the guest is not allowed to infer the
1921 * S/W to PA mapping, it can only use S/W to nuke the whole cache,
1922 * which is a rather good thing for us.
1924 * Also, it is only used when turning caches on/off ("The expected
1925 * usage of the cache maintenance instructions that operate by set/way
1926 * is associated with the cache maintenance instructions associated
1927 * with the powerdown and powerup of caches, if this is required by
1928 * the implementation.").
1930 * We use the following policy:
1932 * - If we trap a S/W operation, we enable VM trapping to detect
1933 * caches being turned on/off, and do a full clean.
1935 * - We flush the caches on both caches being turned on and off.
1937 * - Once the caches are enabled, we stop trapping VM ops.
1939 void kvm_set_way_flush(struct kvm_vcpu
*vcpu
)
1941 unsigned long hcr
= vcpu_get_hcr(vcpu
);
1944 * If this is the first time we do a S/W operation
1945 * (i.e. HCR_TVM not set) flush the whole memory, and set the
1948 * Otherwise, rely on the VM trapping to wait for the MMU +
1949 * Caches to be turned off. At that point, we'll be able to
1950 * clean the caches again.
1952 if (!(hcr
& HCR_TVM
)) {
1953 trace_kvm_set_way_flush(*vcpu_pc(vcpu
),
1954 vcpu_has_cache_enabled(vcpu
));
1955 stage2_flush_vm(vcpu
->kvm
);
1956 vcpu_set_hcr(vcpu
, hcr
| HCR_TVM
);
1960 void kvm_toggle_cache(struct kvm_vcpu
*vcpu
, bool was_enabled
)
1962 bool now_enabled
= vcpu_has_cache_enabled(vcpu
);
1965 * If switching the MMU+caches on, need to invalidate the caches.
1966 * If switching it off, need to clean the caches.
1967 * Clean + invalidate does the trick always.
1969 if (now_enabled
!= was_enabled
)
1970 stage2_flush_vm(vcpu
->kvm
);
1972 /* Caches are now on, stop trapping VM ops (until a S/W op) */
1974 vcpu_set_hcr(vcpu
, vcpu_get_hcr(vcpu
) & ~HCR_TVM
);
1976 trace_kvm_toggle_cache(*vcpu_pc(vcpu
), was_enabled
, now_enabled
);