source: src/linux/universal/linux-4.4/arch/arm/kvm/mmu.c @ 31859

Last change on this file since 31859 was 31859, checked in by brainslayer, 6 weeks ago

kernel update

File size: 50.5 KB
Line 
1/*
2 * Copyright (C) 2012 - Virtual Open Systems and Columbia University
3 * Author: Christoffer Dall <c.dall@virtualopensystems.com>
4 *
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.
8 *
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.
13 *
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.
17 */
18
19#include <linux/mman.h>
20#include <linux/kvm_host.h>
21#include <linux/io.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>
31
32#include "trace.h"
33
34extern char  __hyp_idmap_text_start[], __hyp_idmap_text_end[];
35
36static pgd_t *boot_hyp_pgd;
37static pgd_t *hyp_pgd;
38static pgd_t *merged_hyp_pgd;
39static DEFINE_MUTEX(kvm_hyp_pgd_mutex);
40
41static unsigned long hyp_idmap_start;
42static unsigned long hyp_idmap_end;
43static phys_addr_t hyp_idmap_vector;
44
45#define hyp_pgd_order get_order(PTRS_PER_PGD * sizeof(pgd_t))
46
47#define kvm_pmd_huge(_x)        (pmd_huge(_x) || pmd_trans_huge(_x))
48#define kvm_pud_huge(_x)        pud_huge(_x)
49
50#define KVM_S2PTE_FLAG_IS_IOMAP         (1UL << 0)
51#define KVM_S2_FLAG_LOGGING_ACTIVE      (1UL << 1)
52
53static bool memslot_is_logging(struct kvm_memory_slot *memslot)
54{
55        return memslot->dirty_bitmap && !(memslot->flags & KVM_MEM_READONLY);
56}
57
58/**
59 * kvm_flush_remote_tlbs() - flush all VM TLB entries for v7/8
60 * @kvm:        pointer to kvm structure.
61 *
62 * Interface to HYP function to flush all VM TLB entries
63 */
64void kvm_flush_remote_tlbs(struct kvm *kvm)
65{
66        kvm_call_hyp(__kvm_tlb_flush_vmid, kvm);
67}
68
69static void kvm_tlb_flush_vmid_ipa(struct kvm *kvm, phys_addr_t ipa)
70{
71        /*
72         * This function also gets called when dealing with HYP page
73         * tables. As HYP doesn't have an associated struct kvm (and
74         * the HYP page tables are fairly static), we don't do
75         * anything there.
76         */
77        if (kvm)
78                kvm_call_hyp(__kvm_tlb_flush_vmid_ipa, kvm, ipa);
79}
80
81/*
82 * D-Cache management functions. They take the page table entries by
83 * value, as they are flushing the cache using the kernel mapping (or
84 * kmap on 32bit).
85 */
86static void kvm_flush_dcache_pte(pte_t pte)
87{
88        __kvm_flush_dcache_pte(pte);
89}
90
91static void kvm_flush_dcache_pmd(pmd_t pmd)
92{
93        __kvm_flush_dcache_pmd(pmd);
94}
95
96static void kvm_flush_dcache_pud(pud_t pud)
97{
98        __kvm_flush_dcache_pud(pud);
99}
100
101static bool kvm_is_device_pfn(unsigned long pfn)
102{
103        return !pfn_valid(pfn);
104}
105
106/**
107 * stage2_dissolve_pmd() - clear and flush huge PMD entry
108 * @kvm:        pointer to kvm structure.
109 * @addr:       IPA
110 * @pmd:        pmd pointer for IPA
111 *
112 * Function clears a PMD entry, flushes addr 1st and 2nd stage TLBs. Marks all
113 * pages in the range dirty.
114 */
115static void stage2_dissolve_pmd(struct kvm *kvm, phys_addr_t addr, pmd_t *pmd)
116{
117        if (!kvm_pmd_huge(*pmd))
118                return;
119
120        pmd_clear(pmd);
121        kvm_tlb_flush_vmid_ipa(kvm, addr);
122        put_page(virt_to_page(pmd));
123}
124
125static int mmu_topup_memory_cache(struct kvm_mmu_memory_cache *cache,
126                                  int min, int max)
127{
128        void *page;
129
130        BUG_ON(max > KVM_NR_MEM_OBJS);
131        if (cache->nobjs >= min)
132                return 0;
133        while (cache->nobjs < max) {
134                page = (void *)__get_free_page(PGALLOC_GFP);
135                if (!page)
136                        return -ENOMEM;
137                cache->objects[cache->nobjs++] = page;
138        }
139        return 0;
140}
141
142static void mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc)
143{
144        while (mc->nobjs)
145                free_page((unsigned long)mc->objects[--mc->nobjs]);
146}
147
148static void *mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc)
149{
150        void *p;
151
152        BUG_ON(!mc || !mc->nobjs);
153        p = mc->objects[--mc->nobjs];
154        return p;
155}
156
157static void clear_pgd_entry(struct kvm *kvm, pgd_t *pgd, phys_addr_t addr)
158{
159        pud_t *pud_table __maybe_unused = pud_offset(pgd, 0);
160        pgd_clear(pgd);
161        kvm_tlb_flush_vmid_ipa(kvm, addr);
162        pud_free(NULL, pud_table);
163        put_page(virt_to_page(pgd));
164}
165
166static void clear_pud_entry(struct kvm *kvm, pud_t *pud, phys_addr_t addr)
167{
168        pmd_t *pmd_table = pmd_offset(pud, 0);
169        VM_BUG_ON(pud_huge(*pud));
170        pud_clear(pud);
171        kvm_tlb_flush_vmid_ipa(kvm, addr);
172        pmd_free(NULL, pmd_table);
173        put_page(virt_to_page(pud));
174}
175
176static void clear_pmd_entry(struct kvm *kvm, pmd_t *pmd, phys_addr_t addr)
177{
178        pte_t *pte_table = pte_offset_kernel(pmd, 0);
179        VM_BUG_ON(kvm_pmd_huge(*pmd));
180        pmd_clear(pmd);
181        kvm_tlb_flush_vmid_ipa(kvm, addr);
182        pte_free_kernel(NULL, pte_table);
183        put_page(virt_to_page(pmd));
184}
185
186/*
187 * Unmapping vs dcache management:
188 *
189 * If a guest maps certain memory pages as uncached, all writes will
190 * bypass the data cache and go directly to RAM.  However, the CPUs
191 * can still speculate reads (not writes) and fill cache lines with
192 * data.
193 *
194 * Those cache lines will be *clean* cache lines though, so a
195 * clean+invalidate operation is equivalent to an invalidate
196 * operation, because no cache lines are marked dirty.
197 *
198 * Those clean cache lines could be filled prior to an uncached write
199 * by the guest, and the cache coherent IO subsystem would therefore
200 * end up writing old data to disk.
201 *
202 * This is why right after unmapping a page/section and invalidating
203 * the corresponding TLBs, we call kvm_flush_dcache_p*() to make sure
204 * the IO subsystem will never hit in the cache.
205 */
206static void unmap_ptes(struct kvm *kvm, pmd_t *pmd,
207                       phys_addr_t addr, phys_addr_t end)
208{
209        phys_addr_t start_addr = addr;
210        pte_t *pte, *start_pte;
211
212        start_pte = pte = pte_offset_kernel(pmd, addr);
213        do {
214                if (!pte_none(*pte)) {
215                        pte_t old_pte = *pte;
216
217                        kvm_set_pte(pte, __pte(0));
218                        kvm_tlb_flush_vmid_ipa(kvm, addr);
219
220                        /* No need to invalidate the cache for device mappings */
221                        if (!kvm_is_device_pfn(pte_pfn(old_pte)))
222                                kvm_flush_dcache_pte(old_pte);
223
224                        put_page(virt_to_page(pte));
225                }
226        } while (pte++, addr += PAGE_SIZE, addr != end);
227
228        if (kvm_pte_table_empty(kvm, start_pte))
229                clear_pmd_entry(kvm, pmd, start_addr);
230}
231
232static void unmap_pmds(struct kvm *kvm, pud_t *pud,
233                       phys_addr_t addr, phys_addr_t end)
234{
235        phys_addr_t next, start_addr = addr;
236        pmd_t *pmd, *start_pmd;
237
238        start_pmd = pmd = pmd_offset(pud, addr);
239        do {
240                next = kvm_pmd_addr_end(addr, end);
241                if (!pmd_none(*pmd)) {
242                        if (kvm_pmd_huge(*pmd)) {
243                                pmd_t old_pmd = *pmd;
244
245                                pmd_clear(pmd);
246                                kvm_tlb_flush_vmid_ipa(kvm, addr);
247
248                                kvm_flush_dcache_pmd(old_pmd);
249
250                                put_page(virt_to_page(pmd));
251                        } else {
252                                unmap_ptes(kvm, pmd, addr, next);
253                        }
254                }
255        } while (pmd++, addr = next, addr != end);
256
257        if (kvm_pmd_table_empty(kvm, start_pmd))
258                clear_pud_entry(kvm, pud, start_addr);
259}
260
261static void unmap_puds(struct kvm *kvm, pgd_t *pgd,
262                       phys_addr_t addr, phys_addr_t end)
263{
264        phys_addr_t next, start_addr = addr;
265        pud_t *pud, *start_pud;
266
267        start_pud = pud = pud_offset(pgd, addr);
268        do {
269                next = kvm_pud_addr_end(addr, end);
270                if (!pud_none(*pud)) {
271                        if (pud_huge(*pud)) {
272                                pud_t old_pud = *pud;
273
274                                pud_clear(pud);
275                                kvm_tlb_flush_vmid_ipa(kvm, addr);
276
277                                kvm_flush_dcache_pud(old_pud);
278
279                                put_page(virt_to_page(pud));
280                        } else {
281                                unmap_pmds(kvm, pud, addr, next);
282                        }
283                }
284        } while (pud++, addr = next, addr != end);
285
286        if (kvm_pud_table_empty(kvm, start_pud))
287                clear_pgd_entry(kvm, pgd, start_addr);
288}
289
290
291static void unmap_range(struct kvm *kvm, pgd_t *pgdp,
292                        phys_addr_t start, u64 size)
293{
294        pgd_t *pgd;
295        phys_addr_t addr = start, end = start + size;
296        phys_addr_t next;
297
298        pgd = pgdp + kvm_pgd_index(addr);
299        do {
300                next = kvm_pgd_addr_end(addr, end);
301                if (!pgd_none(*pgd))
302                        unmap_puds(kvm, pgd, addr, next);
303        } while (pgd++, addr = next, addr != end);
304}
305
306static void stage2_flush_ptes(struct kvm *kvm, pmd_t *pmd,
307                              phys_addr_t addr, phys_addr_t end)
308{
309        pte_t *pte;
310
311        pte = pte_offset_kernel(pmd, addr);
312        do {
313                if (!pte_none(*pte) && !kvm_is_device_pfn(pte_pfn(*pte)))
314                        kvm_flush_dcache_pte(*pte);
315        } while (pte++, addr += PAGE_SIZE, addr != end);
316}
317
318static void stage2_flush_pmds(struct kvm *kvm, pud_t *pud,
319                              phys_addr_t addr, phys_addr_t end)
320{
321        pmd_t *pmd;
322        phys_addr_t next;
323
324        pmd = pmd_offset(pud, addr);
325        do {
326                next = kvm_pmd_addr_end(addr, end);
327                if (!pmd_none(*pmd)) {
328                        if (kvm_pmd_huge(*pmd))
329                                kvm_flush_dcache_pmd(*pmd);
330                        else
331                                stage2_flush_ptes(kvm, pmd, addr, next);
332                }
333        } while (pmd++, addr = next, addr != end);
334}
335
336static void stage2_flush_puds(struct kvm *kvm, pgd_t *pgd,
337                              phys_addr_t addr, phys_addr_t end)
338{
339        pud_t *pud;
340        phys_addr_t next;
341
342        pud = pud_offset(pgd, addr);
343        do {
344                next = kvm_pud_addr_end(addr, end);
345                if (!pud_none(*pud)) {
346                        if (pud_huge(*pud))
347                                kvm_flush_dcache_pud(*pud);
348                        else
349                                stage2_flush_pmds(kvm, pud, addr, next);
350                }
351        } while (pud++, addr = next, addr != end);
352}
353
354static void stage2_flush_memslot(struct kvm *kvm,
355                                 struct kvm_memory_slot *memslot)
356{
357        phys_addr_t addr = memslot->base_gfn << PAGE_SHIFT;
358        phys_addr_t end = addr + PAGE_SIZE * memslot->npages;
359        phys_addr_t next;
360        pgd_t *pgd;
361
362        pgd = kvm->arch.pgd + kvm_pgd_index(addr);
363        do {
364                next = kvm_pgd_addr_end(addr, end);
365                stage2_flush_puds(kvm, pgd, addr, next);
366        } while (pgd++, addr = next, addr != end);
367}
368
369/**
370 * stage2_flush_vm - Invalidate cache for pages mapped in stage 2
371 * @kvm: The struct kvm pointer
372 *
373 * Go through the stage 2 page tables and invalidate any cache lines
374 * backing memory already mapped to the VM.
375 */
376static void stage2_flush_vm(struct kvm *kvm)
377{
378        struct kvm_memslots *slots;
379        struct kvm_memory_slot *memslot;
380        int idx;
381
382        idx = srcu_read_lock(&kvm->srcu);
383        spin_lock(&kvm->mmu_lock);
384
385        slots = kvm_memslots(kvm);
386        kvm_for_each_memslot(memslot, slots)
387                stage2_flush_memslot(kvm, memslot);
388
389        spin_unlock(&kvm->mmu_lock);
390        srcu_read_unlock(&kvm->srcu, idx);
391}
392
393/**
394 * free_boot_hyp_pgd - free HYP boot page tables
395 *
396 * Free the HYP boot page tables. The bounce page is also freed.
397 */
398void free_boot_hyp_pgd(void)
399{
400        mutex_lock(&kvm_hyp_pgd_mutex);
401
402        if (boot_hyp_pgd) {
403                unmap_range(NULL, boot_hyp_pgd, hyp_idmap_start, PAGE_SIZE);
404                unmap_range(NULL, boot_hyp_pgd, TRAMPOLINE_VA, PAGE_SIZE);
405                free_pages((unsigned long)boot_hyp_pgd, hyp_pgd_order);
406                boot_hyp_pgd = NULL;
407        }
408
409        if (hyp_pgd)
410                unmap_range(NULL, hyp_pgd, TRAMPOLINE_VA, PAGE_SIZE);
411
412        mutex_unlock(&kvm_hyp_pgd_mutex);
413}
414
415/**
416 * free_hyp_pgds - free Hyp-mode page tables
417 *
418 * Assumes hyp_pgd is a page table used strictly in Hyp-mode and
419 * therefore contains either mappings in the kernel memory area (above
420 * PAGE_OFFSET), or device mappings in the vmalloc range (from
421 * VMALLOC_START to VMALLOC_END).
422 *
423 * boot_hyp_pgd should only map two pages for the init code.
424 */
425void free_hyp_pgds(void)
426{
427        unsigned long addr;
428
429        free_boot_hyp_pgd();
430
431        mutex_lock(&kvm_hyp_pgd_mutex);
432
433        if (hyp_pgd) {
434                for (addr = PAGE_OFFSET; virt_addr_valid(addr); addr += PGDIR_SIZE)
435                        unmap_range(NULL, hyp_pgd, KERN_TO_HYP(addr), PGDIR_SIZE);
436                for (addr = VMALLOC_START; is_vmalloc_addr((void*)addr); addr += PGDIR_SIZE)
437                        unmap_range(NULL, hyp_pgd, KERN_TO_HYP(addr), PGDIR_SIZE);
438
439                free_pages((unsigned long)hyp_pgd, hyp_pgd_order);
440                hyp_pgd = NULL;
441        }
442        if (merged_hyp_pgd) {
443                clear_page(merged_hyp_pgd);
444                free_page((unsigned long)merged_hyp_pgd);
445                merged_hyp_pgd = NULL;
446        }
447
448        mutex_unlock(&kvm_hyp_pgd_mutex);
449}
450
451static void create_hyp_pte_mappings(pmd_t *pmd, unsigned long start,
452                                    unsigned long end, unsigned long pfn,
453                                    pgprot_t prot)
454{
455        pte_t *pte;
456        unsigned long addr;
457
458        addr = start;
459        do {
460                pte = pte_offset_kernel(pmd, addr);
461                kvm_set_pte(pte, pfn_pte(pfn, prot));
462                get_page(virt_to_page(pte));
463                kvm_flush_dcache_to_poc(pte, sizeof(*pte));
464                pfn++;
465        } while (addr += PAGE_SIZE, addr != end);
466}
467
468static int create_hyp_pmd_mappings(pud_t *pud, unsigned long start,
469                                   unsigned long end, unsigned long pfn,
470                                   pgprot_t prot)
471{
472        pmd_t *pmd;
473        pte_t *pte;
474        unsigned long addr, next;
475
476        addr = start;
477        do {
478                pmd = pmd_offset(pud, addr);
479
480                BUG_ON(pmd_sect(*pmd));
481
482                if (pmd_none(*pmd)) {
483                        pte = pte_alloc_one_kernel(NULL, addr);
484                        if (!pte) {
485                                kvm_err("Cannot allocate Hyp pte\n");
486                                return -ENOMEM;
487                        }
488                        pmd_populate_kernel(NULL, pmd, pte);
489                        get_page(virt_to_page(pmd));
490                        kvm_flush_dcache_to_poc(pmd, sizeof(*pmd));
491                }
492
493                next = pmd_addr_end(addr, end);
494
495                create_hyp_pte_mappings(pmd, addr, next, pfn, prot);
496                pfn += (next - addr) >> PAGE_SHIFT;
497        } while (addr = next, addr != end);
498
499        return 0;
500}
501
502static int create_hyp_pud_mappings(pgd_t *pgd, unsigned long start,
503                                   unsigned long end, unsigned long pfn,
504                                   pgprot_t prot)
505{
506        pud_t *pud;
507        pmd_t *pmd;
508        unsigned long addr, next;
509        int ret;
510
511        addr = start;
512        do {
513                pud = pud_offset(pgd, addr);
514
515                if (pud_none_or_clear_bad(pud)) {
516                        pmd = pmd_alloc_one(NULL, addr);
517                        if (!pmd) {
518                                kvm_err("Cannot allocate Hyp pmd\n");
519                                return -ENOMEM;
520                        }
521                        pud_populate(NULL, pud, pmd);
522                        get_page(virt_to_page(pud));
523                        kvm_flush_dcache_to_poc(pud, sizeof(*pud));
524                }
525
526                next = pud_addr_end(addr, end);
527                ret = create_hyp_pmd_mappings(pud, addr, next, pfn, prot);
528                if (ret)
529                        return ret;
530                pfn += (next - addr) >> PAGE_SHIFT;
531        } while (addr = next, addr != end);
532
533        return 0;
534}
535
536static int __create_hyp_mappings(pgd_t *pgdp,
537                                 unsigned long start, unsigned long end,
538                                 unsigned long pfn, pgprot_t prot)
539{
540        pgd_t *pgd;
541        pud_t *pud;
542        unsigned long addr, next;
543        int err = 0;
544
545        mutex_lock(&kvm_hyp_pgd_mutex);
546        addr = start & PAGE_MASK;
547        end = PAGE_ALIGN(end);
548        do {
549                pgd = pgdp + pgd_index(addr);
550
551                if (pgd_none(*pgd)) {
552                        pud = pud_alloc_one(NULL, addr);
553                        if (!pud) {
554                                kvm_err("Cannot allocate Hyp pud\n");
555                                err = -ENOMEM;
556                                goto out;
557                        }
558                        pgd_populate(NULL, pgd, pud);
559                        get_page(virt_to_page(pgd));
560                        kvm_flush_dcache_to_poc(pgd, sizeof(*pgd));
561                }
562
563                next = pgd_addr_end(addr, end);
564                err = create_hyp_pud_mappings(pgd, addr, next, pfn, prot);
565                if (err)
566                        goto out;
567                pfn += (next - addr) >> PAGE_SHIFT;
568        } while (addr = next, addr != end);
569out:
570        mutex_unlock(&kvm_hyp_pgd_mutex);
571        return err;
572}
573
574static phys_addr_t kvm_kaddr_to_phys(void *kaddr)
575{
576        if (!is_vmalloc_addr(kaddr)) {
577                BUG_ON(!virt_addr_valid(kaddr));
578                return __pa(kaddr);
579        } else {
580                return page_to_phys(vmalloc_to_page(kaddr)) +
581                       offset_in_page(kaddr);
582        }
583}
584
585/**
586 * create_hyp_mappings - duplicate a kernel virtual address range in Hyp mode
587 * @from:       The virtual kernel start address of the range
588 * @to:         The virtual kernel end address of the range (exclusive)
589 *
590 * The same virtual address as the kernel virtual address is also used
591 * in Hyp-mode mapping (modulo HYP_PAGE_OFFSET) to the same underlying
592 * physical pages.
593 */
594int create_hyp_mappings(void *from, void *to)
595{
596        phys_addr_t phys_addr;
597        unsigned long virt_addr;
598        unsigned long start = KERN_TO_HYP((unsigned long)from);
599        unsigned long end = KERN_TO_HYP((unsigned long)to);
600
601        start = start & PAGE_MASK;
602        end = PAGE_ALIGN(end);
603
604        for (virt_addr = start; virt_addr < end; virt_addr += PAGE_SIZE) {
605                int err;
606
607                phys_addr = kvm_kaddr_to_phys(from + virt_addr - start);
608                err = __create_hyp_mappings(hyp_pgd, virt_addr,
609                                            virt_addr + PAGE_SIZE,
610                                            __phys_to_pfn(phys_addr),
611                                            PAGE_HYP);
612                if (err)
613                        return err;
614        }
615
616        return 0;
617}
618
619/**
620 * create_hyp_io_mappings - duplicate a kernel IO mapping into Hyp mode
621 * @from:       The kernel start VA of the range
622 * @to:         The kernel end VA of the range (exclusive)
623 * @phys_addr:  The physical start address which gets mapped
624 *
625 * The resulting HYP VA is the same as the kernel VA, modulo
626 * HYP_PAGE_OFFSET.
627 */
628int create_hyp_io_mappings(void *from, void *to, phys_addr_t phys_addr)
629{
630        unsigned long start = KERN_TO_HYP((unsigned long)from);
631        unsigned long end = KERN_TO_HYP((unsigned long)to);
632
633        /* Check for a valid kernel IO mapping */
634        if (!is_vmalloc_addr(from) || !is_vmalloc_addr(to - 1))
635                return -EINVAL;
636
637        return __create_hyp_mappings(hyp_pgd, start, end,
638                                     __phys_to_pfn(phys_addr), PAGE_HYP_DEVICE);
639}
640
641/* Free the HW pgd, one page at a time */
642static void kvm_free_hwpgd(void *hwpgd)
643{
644        free_pages_exact(hwpgd, kvm_get_hwpgd_size());
645}
646
647/* Allocate the HW PGD, making sure that each page gets its own refcount */
648static void *kvm_alloc_hwpgd(void)
649{
650        unsigned int size = kvm_get_hwpgd_size();
651
652        return alloc_pages_exact(size, GFP_KERNEL | __GFP_ZERO);
653}
654
655/**
656 * kvm_alloc_stage2_pgd - allocate level-1 table for stage-2 translation.
657 * @kvm:        The KVM struct pointer for the VM.
658 *
659 * Allocates the 1st level table only of size defined by S2_PGD_ORDER (can
660 * support either full 40-bit input addresses or limited to 32-bit input
661 * addresses). Clears the allocated pages.
662 *
663 * Note we don't need locking here as this is only called when the VM is
664 * created, which can only be done once.
665 */
666int kvm_alloc_stage2_pgd(struct kvm *kvm)
667{
668        pgd_t *pgd;
669        void *hwpgd;
670
671        if (kvm->arch.pgd != NULL) {
672                kvm_err("kvm_arch already initialized?\n");
673                return -EINVAL;
674        }
675
676        hwpgd = kvm_alloc_hwpgd();
677        if (!hwpgd)
678                return -ENOMEM;
679
680        /* When the kernel uses more levels of page tables than the
681         * guest, we allocate a fake PGD and pre-populate it to point
682         * to the next-level page table, which will be the real
683         * initial page table pointed to by the VTTBR.
684         *
685         * When KVM_PREALLOC_LEVEL==2, we allocate a single page for
686         * the PMD and the kernel will use folded pud.
687         * When KVM_PREALLOC_LEVEL==1, we allocate 2 consecutive PUD
688         * pages.
689         */
690        if (KVM_PREALLOC_LEVEL > 0) {
691                int i;
692
693                /*
694                 * Allocate fake pgd for the page table manipulation macros to
695                 * work.  This is not used by the hardware and we have no
696                 * alignment requirement for this allocation.
697                 */
698                pgd = kmalloc(PTRS_PER_S2_PGD * sizeof(pgd_t),
699                                GFP_KERNEL | __GFP_ZERO);
700
701                if (!pgd) {
702                        kvm_free_hwpgd(hwpgd);
703                        return -ENOMEM;
704                }
705
706                /* Plug the HW PGD into the fake one. */
707                for (i = 0; i < PTRS_PER_S2_PGD; i++) {
708                        if (KVM_PREALLOC_LEVEL == 1)
709                                pgd_populate(NULL, pgd + i,
710                                             (pud_t *)hwpgd + i * PTRS_PER_PUD);
711                        else if (KVM_PREALLOC_LEVEL == 2)
712                                pud_populate(NULL, pud_offset(pgd, 0) + i,
713                                             (pmd_t *)hwpgd + i * PTRS_PER_PMD);
714                }
715        } else {
716                /*
717                 * Allocate actual first-level Stage-2 page table used by the
718                 * hardware for Stage-2 page table walks.
719                 */
720                pgd = (pgd_t *)hwpgd;
721        }
722
723        kvm_clean_pgd(pgd);
724        kvm->arch.pgd = pgd;
725        return 0;
726}
727
728/**
729 * unmap_stage2_range -- Clear stage2 page table entries to unmap a range
730 * @kvm:   The VM pointer
731 * @start: The intermediate physical base address of the range to unmap
732 * @size:  The size of the area to unmap
733 *
734 * Clear a range of stage-2 mappings, lowering the various ref-counts.  Must
735 * be called while holding mmu_lock (unless for freeing the stage2 pgd before
736 * destroying the VM), otherwise another faulting VCPU may come in and mess
737 * with things behind our backs.
738 */
739static void unmap_stage2_range(struct kvm *kvm, phys_addr_t start, u64 size)
740{
741        unmap_range(kvm, kvm->arch.pgd, start, size);
742}
743
744static void stage2_unmap_memslot(struct kvm *kvm,
745                                 struct kvm_memory_slot *memslot)
746{
747        hva_t hva = memslot->userspace_addr;
748        phys_addr_t addr = memslot->base_gfn << PAGE_SHIFT;
749        phys_addr_t size = PAGE_SIZE * memslot->npages;
750        hva_t reg_end = hva + size;
751
752        /*
753         * A memory region could potentially cover multiple VMAs, and any holes
754         * between them, so iterate over all of them to find out if we should
755         * unmap any of them.
756         *
757         *     +--------------------------------------------+
758         * +---------------+----------------+   +----------------+
759         * |   : VMA 1     |      VMA 2     |   |    VMA 3  :    |
760         * +---------------+----------------+   +----------------+
761         *     |               memory region                |
762         *     +--------------------------------------------+
763         */
764        do {
765                struct vm_area_struct *vma = find_vma(current->mm, hva);
766                hva_t vm_start, vm_end;
767
768                if (!vma || vma->vm_start >= reg_end)
769                        break;
770
771                /*
772                 * Take the intersection of this VMA with the memory region
773                 */
774                vm_start = max(hva, vma->vm_start);
775                vm_end = min(reg_end, vma->vm_end);
776
777                if (!(vma->vm_flags & VM_PFNMAP)) {
778                        gpa_t gpa = addr + (vm_start - memslot->userspace_addr);
779                        unmap_stage2_range(kvm, gpa, vm_end - vm_start);
780                }
781                hva = vm_end;
782        } while (hva < reg_end);
783}
784
785/**
786 * stage2_unmap_vm - Unmap Stage-2 RAM mappings
787 * @kvm: The struct kvm pointer
788 *
789 * Go through the memregions and unmap any reguler RAM
790 * backing memory already mapped to the VM.
791 */
792void stage2_unmap_vm(struct kvm *kvm)
793{
794        struct kvm_memslots *slots;
795        struct kvm_memory_slot *memslot;
796        int idx;
797
798        idx = srcu_read_lock(&kvm->srcu);
799        down_read(&current->mm->mmap_sem);
800        spin_lock(&kvm->mmu_lock);
801
802        slots = kvm_memslots(kvm);
803        kvm_for_each_memslot(memslot, slots)
804                stage2_unmap_memslot(kvm, memslot);
805
806        spin_unlock(&kvm->mmu_lock);
807        up_read(&current->mm->mmap_sem);
808        srcu_read_unlock(&kvm->srcu, idx);
809}
810
811/**
812 * kvm_free_stage2_pgd - free all stage-2 tables
813 * @kvm:        The KVM struct pointer for the VM.
814 *
815 * Walks the level-1 page table pointed to by kvm->arch.pgd and frees all
816 * underlying level-2 and level-3 tables before freeing the actual level-1 table
817 * and setting the struct pointer to NULL.
818 *
819 * Note we don't need locking here as this is only called when the VM is
820 * destroyed, which can only be done once.
821 */
822void kvm_free_stage2_pgd(struct kvm *kvm)
823{
824        if (kvm->arch.pgd == NULL)
825                return;
826
827        unmap_stage2_range(kvm, 0, KVM_PHYS_SIZE);
828        kvm_free_hwpgd(kvm_get_hwpgd(kvm));
829        if (KVM_PREALLOC_LEVEL > 0)
830                kfree(kvm->arch.pgd);
831
832        kvm->arch.pgd = NULL;
833}
834
835static pud_t *stage2_get_pud(struct kvm *kvm, struct kvm_mmu_memory_cache *cache,
836                             phys_addr_t addr)
837{
838        pgd_t *pgd;
839        pud_t *pud;
840
841        pgd = kvm->arch.pgd + kvm_pgd_index(addr);
842        if (WARN_ON(pgd_none(*pgd))) {
843                if (!cache)
844                        return NULL;
845                pud = mmu_memory_cache_alloc(cache);
846                pgd_populate(NULL, pgd, pud);
847                get_page(virt_to_page(pgd));
848        }
849
850        return pud_offset(pgd, addr);
851}
852
853static pmd_t *stage2_get_pmd(struct kvm *kvm, struct kvm_mmu_memory_cache *cache,
854                             phys_addr_t addr)
855{
856        pud_t *pud;
857        pmd_t *pmd;
858
859        pud = stage2_get_pud(kvm, cache, addr);
860        if (pud_none(*pud)) {
861                if (!cache)
862                        return NULL;
863                pmd = mmu_memory_cache_alloc(cache);
864                pud_populate(NULL, pud, pmd);
865                get_page(virt_to_page(pud));
866        }
867
868        return pmd_offset(pud, addr);
869}
870
871static int stage2_set_pmd_huge(struct kvm *kvm, struct kvm_mmu_memory_cache
872                               *cache, phys_addr_t addr, const pmd_t *new_pmd)
873{
874        pmd_t *pmd, old_pmd;
875
876        pmd = stage2_get_pmd(kvm, cache, addr);
877        VM_BUG_ON(!pmd);
878
879        /*
880         * Mapping in huge pages should only happen through a fault.  If a
881         * page is merged into a transparent huge page, the individual
882         * subpages of that huge page should be unmapped through MMU
883         * notifiers before we get here.
884         *
885         * Merging of CompoundPages is not supported; they should become
886         * splitting first, unmapped, merged, and mapped back in on-demand.
887         */
888        VM_BUG_ON(pmd_present(*pmd) && pmd_pfn(*pmd) != pmd_pfn(*new_pmd));
889
890        old_pmd = *pmd;
891        if (pmd_present(old_pmd)) {
892                pmd_clear(pmd);
893                kvm_tlb_flush_vmid_ipa(kvm, addr);
894        } else {
895                get_page(virt_to_page(pmd));
896        }
897
898        kvm_set_pmd(pmd, *new_pmd);
899        return 0;
900}
901
902static int stage2_set_pte(struct kvm *kvm, struct kvm_mmu_memory_cache *cache,
903                          phys_addr_t addr, const pte_t *new_pte,
904                          unsigned long flags)
905{
906        pmd_t *pmd;
907        pte_t *pte, old_pte;
908        bool iomap = flags & KVM_S2PTE_FLAG_IS_IOMAP;
909        bool logging_active = flags & KVM_S2_FLAG_LOGGING_ACTIVE;
910
911        VM_BUG_ON(logging_active && !cache);
912
913        /* Create stage-2 page table mapping - Levels 0 and 1 */
914        pmd = stage2_get_pmd(kvm, cache, addr);
915        if (!pmd) {
916                /*
917                 * Ignore calls from kvm_set_spte_hva for unallocated
918                 * address ranges.
919                 */
920                return 0;
921        }
922
923        /*
924         * While dirty page logging - dissolve huge PMD, then continue on to
925         * allocate page.
926         */
927        if (logging_active)
928                stage2_dissolve_pmd(kvm, addr, pmd);
929
930        /* Create stage-2 page mappings - Level 2 */
931        if (pmd_none(*pmd)) {
932                if (!cache)
933                        return 0; /* ignore calls from kvm_set_spte_hva */
934                pte = mmu_memory_cache_alloc(cache);
935                kvm_clean_pte(pte);
936                pmd_populate_kernel(NULL, pmd, pte);
937                get_page(virt_to_page(pmd));
938        }
939
940        pte = pte_offset_kernel(pmd, addr);
941
942        if (iomap && pte_present(*pte))
943                return -EFAULT;
944
945        /* Create 2nd stage page table mapping - Level 3 */
946        old_pte = *pte;
947        if (pte_present(old_pte)) {
948                kvm_set_pte(pte, __pte(0));
949                kvm_tlb_flush_vmid_ipa(kvm, addr);
950        } else {
951                get_page(virt_to_page(pte));
952        }
953
954        kvm_set_pte(pte, *new_pte);
955        return 0;
956}
957
958/**
959 * kvm_phys_addr_ioremap - map a device range to guest IPA
960 *
961 * @kvm:        The KVM pointer
962 * @guest_ipa:  The IPA at which to insert the mapping
963 * @pa:         The physical address of the device
964 * @size:       The size of the mapping
965 */
966int kvm_phys_addr_ioremap(struct kvm *kvm, phys_addr_t guest_ipa,
967                          phys_addr_t pa, unsigned long size, bool writable)
968{
969        phys_addr_t addr, end;
970        int ret = 0;
971        unsigned long pfn;
972        struct kvm_mmu_memory_cache cache = { 0, };
973
974        end = (guest_ipa + size + PAGE_SIZE - 1) & PAGE_MASK;
975        pfn = __phys_to_pfn(pa);
976
977        for (addr = guest_ipa; addr < end; addr += PAGE_SIZE) {
978                pte_t pte = pfn_pte(pfn, PAGE_S2_DEVICE);
979
980                if (writable)
981                        kvm_set_s2pte_writable(&pte);
982
983                ret = mmu_topup_memory_cache(&cache, KVM_MMU_CACHE_MIN_PAGES,
984                                                KVM_NR_MEM_OBJS);
985                if (ret)
986                        goto out;
987                spin_lock(&kvm->mmu_lock);
988                ret = stage2_set_pte(kvm, &cache, addr, &pte,
989                                                KVM_S2PTE_FLAG_IS_IOMAP);
990                spin_unlock(&kvm->mmu_lock);
991                if (ret)
992                        goto out;
993
994                pfn++;
995        }
996
997out:
998        mmu_free_memory_cache(&cache);
999        return ret;
1000}
1001
1002static bool transparent_hugepage_adjust(pfn_t *pfnp, phys_addr_t *ipap)
1003{
1004        pfn_t pfn = *pfnp;
1005        gfn_t gfn = *ipap >> PAGE_SHIFT;
1006
1007        if (PageTransCompound(pfn_to_page(pfn))) {
1008                unsigned long mask;
1009                /*
1010                 * The address we faulted on is backed by a transparent huge
1011                 * page.  However, because we map the compound huge page and
1012                 * not the individual tail page, we need to transfer the
1013                 * refcount to the head page.  We have to be careful that the
1014                 * THP doesn't start to split while we are adjusting the
1015                 * refcounts.
1016                 *
1017                 * We are sure this doesn't happen, because mmu_notifier_retry
1018                 * was successful and we are holding the mmu_lock, so if this
1019                 * THP is trying to split, it will be blocked in the mmu
1020                 * notifier before touching any of the pages, specifically
1021                 * before being able to call __split_huge_page_refcount().
1022                 *
1023                 * We can therefore safely transfer the refcount from PG_tail
1024                 * to PG_head and switch the pfn from a tail page to the head
1025                 * page accordingly.
1026                 */
1027                mask = PTRS_PER_PMD - 1;
1028                VM_BUG_ON((gfn & mask) != (pfn & mask));
1029                if (pfn & mask) {
1030                        *ipap &= PMD_MASK;
1031                        kvm_release_pfn_clean(pfn);
1032                        pfn &= ~mask;
1033                        kvm_get_pfn(pfn);
1034                        *pfnp = pfn;
1035                }
1036
1037                return true;
1038        }
1039
1040        return false;
1041}
1042
1043static bool kvm_is_write_fault(struct kvm_vcpu *vcpu)
1044{
1045        if (kvm_vcpu_trap_is_iabt(vcpu))
1046                return false;
1047
1048        return kvm_vcpu_dabt_iswrite(vcpu);
1049}
1050
1051/**
1052 * stage2_wp_ptes - write protect PMD range
1053 * @pmd:        pointer to pmd entry
1054 * @addr:       range start address
1055 * @end:        range end address
1056 */
1057static void stage2_wp_ptes(pmd_t *pmd, phys_addr_t addr, phys_addr_t end)
1058{
1059        pte_t *pte;
1060
1061        pte = pte_offset_kernel(pmd, addr);
1062        do {
1063                if (!pte_none(*pte)) {
1064                        if (!kvm_s2pte_readonly(pte))
1065                                kvm_set_s2pte_readonly(pte);
1066                }
1067        } while (pte++, addr += PAGE_SIZE, addr != end);
1068}
1069
1070/**
1071 * stage2_wp_pmds - write protect PUD range
1072 * @pud:        pointer to pud entry
1073 * @addr:       range start address
1074 * @end:        range end address
1075 */
1076static void stage2_wp_pmds(pud_t *pud, phys_addr_t addr, phys_addr_t end)
1077{
1078        pmd_t *pmd;
1079        phys_addr_t next;
1080
1081        pmd = pmd_offset(pud, addr);
1082
1083        do {
1084                next = kvm_pmd_addr_end(addr, end);
1085                if (!pmd_none(*pmd)) {
1086                        if (kvm_pmd_huge(*pmd)) {
1087                                if (!kvm_s2pmd_readonly(pmd))
1088                                        kvm_set_s2pmd_readonly(pmd);
1089                        } else {
1090                                stage2_wp_ptes(pmd, addr, next);
1091                        }
1092                }
1093        } while (pmd++, addr = next, addr != end);
1094}
1095
1096/**
1097  * stage2_wp_puds - write protect PGD range
1098  * @pgd:       pointer to pgd entry
1099  * @addr:      range start address
1100  * @end:       range end address
1101  *
1102  * Process PUD entries, for a huge PUD we cause a panic.
1103  */
1104static void  stage2_wp_puds(pgd_t *pgd, phys_addr_t addr, phys_addr_t end)
1105{
1106        pud_t *pud;
1107        phys_addr_t next;
1108
1109        pud = pud_offset(pgd, addr);
1110        do {
1111                next = kvm_pud_addr_end(addr, end);
1112                if (!pud_none(*pud)) {
1113                        /* TODO:PUD not supported, revisit later if supported */
1114                        BUG_ON(kvm_pud_huge(*pud));
1115                        stage2_wp_pmds(pud, addr, next);
1116                }
1117        } while (pud++, addr = next, addr != end);
1118}
1119
1120/**
1121 * stage2_wp_range() - write protect stage2 memory region range
1122 * @kvm:        The KVM pointer
1123 * @addr:       Start address of range
1124 * @end:        End address of range
1125 */
1126static void stage2_wp_range(struct kvm *kvm, phys_addr_t addr, phys_addr_t end)
1127{
1128        pgd_t *pgd;
1129        phys_addr_t next;
1130
1131        pgd = kvm->arch.pgd + kvm_pgd_index(addr);
1132        do {
1133                /*
1134                 * Release kvm_mmu_lock periodically if the memory region is
1135                 * large. Otherwise, we may see kernel panics with
1136                 * CONFIG_DETECT_HUNG_TASK, CONFIG_LOCKUP_DETECTOR,
1137                 * CONFIG_LOCKDEP. Additionally, holding the lock too long
1138                 * will also starve other vCPUs.
1139                 */
1140                if (need_resched() || spin_needbreak(&kvm->mmu_lock))
1141                        cond_resched_lock(&kvm->mmu_lock);
1142
1143                next = kvm_pgd_addr_end(addr, end);
1144                if (pgd_present(*pgd))
1145                        stage2_wp_puds(pgd, addr, next);
1146        } while (pgd++, addr = next, addr != end);
1147}
1148
1149/**
1150 * kvm_mmu_wp_memory_region() - write protect stage 2 entries for memory slot
1151 * @kvm:        The KVM pointer
1152 * @slot:       The memory slot to write protect
1153 *
1154 * Called to start logging dirty pages after memory region
1155 * KVM_MEM_LOG_DIRTY_PAGES operation is called. After this function returns
1156 * all present PMD and PTEs are write protected in the memory region.
1157 * Afterwards read of dirty page log can be called.
1158 *
1159 * Acquires kvm_mmu_lock. Called with kvm->slots_lock mutex acquired,
1160 * serializing operations for VM memory regions.
1161 */
1162void kvm_mmu_wp_memory_region(struct kvm *kvm, int slot)
1163{
1164        struct kvm_memslots *slots = kvm_memslots(kvm);
1165        struct kvm_memory_slot *memslot = id_to_memslot(slots, slot);
1166        phys_addr_t start = memslot->base_gfn << PAGE_SHIFT;
1167        phys_addr_t end = (memslot->base_gfn + memslot->npages) << PAGE_SHIFT;
1168
1169        spin_lock(&kvm->mmu_lock);
1170        stage2_wp_range(kvm, start, end);
1171        spin_unlock(&kvm->mmu_lock);
1172        kvm_flush_remote_tlbs(kvm);
1173}
1174
1175/**
1176 * kvm_mmu_write_protect_pt_masked() - write protect dirty pages
1177 * @kvm:        The KVM pointer
1178 * @slot:       The memory slot associated with mask
1179 * @gfn_offset: The gfn offset in memory slot
1180 * @mask:       The mask of dirty pages at offset 'gfn_offset' in this memory
1181 *              slot to be write protected
1182 *
1183 * Walks bits set in mask write protects the associated pte's. Caller must
1184 * acquire kvm_mmu_lock.
1185 */
1186static void kvm_mmu_write_protect_pt_masked(struct kvm *kvm,
1187                struct kvm_memory_slot *slot,
1188                gfn_t gfn_offset, unsigned long mask)
1189{
1190        phys_addr_t base_gfn = slot->base_gfn + gfn_offset;
1191        phys_addr_t start = (base_gfn +  __ffs(mask)) << PAGE_SHIFT;
1192        phys_addr_t end = (base_gfn + __fls(mask) + 1) << PAGE_SHIFT;
1193
1194        stage2_wp_range(kvm, start, end);
1195}
1196
1197/*
1198 * kvm_arch_mmu_enable_log_dirty_pt_masked - enable dirty logging for selected
1199 * dirty pages.
1200 *
1201 * It calls kvm_mmu_write_protect_pt_masked to write protect selected pages to
1202 * enable dirty logging for them.
1203 */
1204void kvm_arch_mmu_enable_log_dirty_pt_masked(struct kvm *kvm,
1205                struct kvm_memory_slot *slot,
1206                gfn_t gfn_offset, unsigned long mask)
1207{
1208        kvm_mmu_write_protect_pt_masked(kvm, slot, gfn_offset, mask);
1209}
1210
1211static void coherent_cache_guest_page(struct kvm_vcpu *vcpu, pfn_t pfn,
1212                                      unsigned long size, bool uncached)
1213{
1214        __coherent_cache_guest_page(vcpu, pfn, size, uncached);
1215}
1216
1217static int user_mem_abort(struct kvm_vcpu *vcpu, phys_addr_t fault_ipa,
1218                          struct kvm_memory_slot *memslot, unsigned long hva,
1219                          unsigned long fault_status)
1220{
1221        int ret;
1222        bool write_fault, writable, hugetlb = false, force_pte = false;
1223        unsigned long mmu_seq;
1224        gfn_t gfn = fault_ipa >> PAGE_SHIFT;
1225        struct kvm *kvm = vcpu->kvm;
1226        struct kvm_mmu_memory_cache *memcache = &vcpu->arch.mmu_page_cache;
1227        struct vm_area_struct *vma;
1228        pfn_t pfn;
1229        pgprot_t mem_type = PAGE_S2;
1230        bool fault_ipa_uncached;
1231        bool logging_active = memslot_is_logging(memslot);
1232        unsigned long flags = 0;
1233
1234        write_fault = kvm_is_write_fault(vcpu);
1235        if (fault_status == FSC_PERM && !write_fault) {
1236                kvm_err("Unexpected L2 read permission error\n");
1237                return -EFAULT;
1238        }
1239
1240        /* Let's check if we will get back a huge page backed by hugetlbfs */
1241        down_read(&current->mm->mmap_sem);
1242        vma = find_vma_intersection(current->mm, hva, hva + 1);
1243        if (unlikely(!vma)) {
1244                kvm_err("Failed to find VMA for hva 0x%lx\n", hva);
1245                up_read(&current->mm->mmap_sem);
1246                return -EFAULT;
1247        }
1248
1249        if (is_vm_hugetlb_page(vma) && !logging_active) {
1250                hugetlb = true;
1251                gfn = (fault_ipa & PMD_MASK) >> PAGE_SHIFT;
1252        } else {
1253                /*
1254                 * Pages belonging to memslots that don't have the same
1255                 * alignment for userspace and IPA cannot be mapped using
1256                 * block descriptors even if the pages belong to a THP for
1257                 * the process, because the stage-2 block descriptor will
1258                 * cover more than a single THP and we loose atomicity for
1259                 * unmapping, updates, and splits of the THP or other pages
1260                 * in the stage-2 block range.
1261                 */
1262                if ((memslot->userspace_addr & ~PMD_MASK) !=
1263                    ((memslot->base_gfn << PAGE_SHIFT) & ~PMD_MASK))
1264                        force_pte = true;
1265        }
1266        up_read(&current->mm->mmap_sem);
1267
1268        /* We need minimum second+third level pages */
1269        ret = mmu_topup_memory_cache(memcache, KVM_MMU_CACHE_MIN_PAGES,
1270                                     KVM_NR_MEM_OBJS);
1271        if (ret)
1272                return ret;
1273
1274        mmu_seq = vcpu->kvm->mmu_notifier_seq;
1275        /*
1276         * Ensure the read of mmu_notifier_seq happens before we call
1277         * gfn_to_pfn_prot (which calls get_user_pages), so that we don't risk
1278         * the page we just got a reference to gets unmapped before we have a
1279         * chance to grab the mmu_lock, which ensure that if the page gets
1280         * unmapped afterwards, the call to kvm_unmap_hva will take it away
1281         * from us again properly. This smp_rmb() interacts with the smp_wmb()
1282         * in kvm_mmu_notifier_invalidate_<page|range_end>.
1283         */
1284        smp_rmb();
1285
1286        pfn = gfn_to_pfn_prot(kvm, gfn, write_fault, &writable);
1287        if (is_error_pfn(pfn))
1288                return -EFAULT;
1289
1290        if (kvm_is_device_pfn(pfn)) {
1291                mem_type = PAGE_S2_DEVICE;
1292                flags |= KVM_S2PTE_FLAG_IS_IOMAP;
1293        } else if (logging_active) {
1294                /*
1295                 * Faults on pages in a memslot with logging enabled
1296                 * should not be mapped with huge pages (it introduces churn
1297                 * and performance degradation), so force a pte mapping.
1298                 */
1299                force_pte = true;
1300                flags |= KVM_S2_FLAG_LOGGING_ACTIVE;
1301
1302                /*
1303                 * Only actually map the page as writable if this was a write
1304                 * fault.
1305                 */
1306                if (!write_fault)
1307                        writable = false;
1308        }
1309
1310        spin_lock(&kvm->mmu_lock);
1311        if (mmu_notifier_retry(kvm, mmu_seq))
1312                goto out_unlock;
1313
1314        if (!hugetlb && !force_pte)
1315                hugetlb = transparent_hugepage_adjust(&pfn, &fault_ipa);
1316
1317        fault_ipa_uncached = memslot->flags & KVM_MEMSLOT_INCOHERENT;
1318
1319        if (hugetlb) {
1320                pmd_t new_pmd = pfn_pmd(pfn, mem_type);
1321                new_pmd = pmd_mkhuge(new_pmd);
1322                if (writable) {
1323                        kvm_set_s2pmd_writable(&new_pmd);
1324                        kvm_set_pfn_dirty(pfn);
1325                }
1326                coherent_cache_guest_page(vcpu, pfn, PMD_SIZE, fault_ipa_uncached);
1327                ret = stage2_set_pmd_huge(kvm, memcache, fault_ipa, &new_pmd);
1328        } else {
1329                pte_t new_pte = pfn_pte(pfn, mem_type);
1330
1331                if (writable) {
1332                        kvm_set_s2pte_writable(&new_pte);
1333                        kvm_set_pfn_dirty(pfn);
1334                        mark_page_dirty(kvm, gfn);
1335                }
1336                coherent_cache_guest_page(vcpu, pfn, PAGE_SIZE, fault_ipa_uncached);
1337                ret = stage2_set_pte(kvm, memcache, fault_ipa, &new_pte, flags);
1338        }
1339
1340out_unlock:
1341        spin_unlock(&kvm->mmu_lock);
1342        kvm_set_pfn_accessed(pfn);
1343        kvm_release_pfn_clean(pfn);
1344        return ret;
1345}
1346
1347/*
1348 * Resolve the access fault by making the page young again.
1349 * Note that because the faulting entry is guaranteed not to be
1350 * cached in the TLB, we don't need to invalidate anything.
1351 */
1352static void handle_access_fault(struct kvm_vcpu *vcpu, phys_addr_t fault_ipa)
1353{
1354        pmd_t *pmd;
1355        pte_t *pte;
1356        pfn_t pfn;
1357        bool pfn_valid = false;
1358
1359        trace_kvm_access_fault(fault_ipa);
1360
1361        spin_lock(&vcpu->kvm->mmu_lock);
1362
1363        pmd = stage2_get_pmd(vcpu->kvm, NULL, fault_ipa);
1364        if (!pmd || pmd_none(*pmd))     /* Nothing there */
1365                goto out;
1366
1367        if (kvm_pmd_huge(*pmd)) {       /* THP, HugeTLB */
1368                *pmd = pmd_mkyoung(*pmd);
1369                pfn = pmd_pfn(*pmd);
1370                pfn_valid = true;
1371                goto out;
1372        }
1373
1374        pte = pte_offset_kernel(pmd, fault_ipa);
1375        if (pte_none(*pte))             /* Nothing there either */
1376                goto out;
1377
1378        *pte = pte_mkyoung(*pte);       /* Just a page... */
1379        pfn = pte_pfn(*pte);
1380        pfn_valid = true;
1381out:
1382        spin_unlock(&vcpu->kvm->mmu_lock);
1383        if (pfn_valid)
1384                kvm_set_pfn_accessed(pfn);
1385}
1386
1387/**
1388 * kvm_handle_guest_abort - handles all 2nd stage aborts
1389 * @vcpu:       the VCPU pointer
1390 * @run:        the kvm_run structure
1391 *
1392 * Any abort that gets to the host is almost guaranteed to be caused by a
1393 * missing second stage translation table entry, which can mean that either the
1394 * guest simply needs more memory and we must allocate an appropriate page or it
1395 * can mean that the guest tried to access I/O memory, which is emulated by user
1396 * space. The distinction is based on the IPA causing the fault and whether this
1397 * memory region has been registered as standard RAM by user space.
1398 */
1399int kvm_handle_guest_abort(struct kvm_vcpu *vcpu, struct kvm_run *run)
1400{
1401        unsigned long fault_status;
1402        phys_addr_t fault_ipa;
1403        struct kvm_memory_slot *memslot;
1404        unsigned long hva;
1405        bool is_iabt, write_fault, writable;
1406        gfn_t gfn;
1407        int ret, idx;
1408
1409        is_iabt = kvm_vcpu_trap_is_iabt(vcpu);
1410        fault_ipa = kvm_vcpu_get_fault_ipa(vcpu);
1411
1412        trace_kvm_guest_fault(*vcpu_pc(vcpu), kvm_vcpu_get_hsr(vcpu),
1413                              kvm_vcpu_get_hfar(vcpu), fault_ipa);
1414
1415        /* Check the stage-2 fault is trans. fault or write fault */
1416        fault_status = kvm_vcpu_trap_get_fault_type(vcpu);
1417        if (fault_status != FSC_FAULT && fault_status != FSC_PERM &&
1418            fault_status != FSC_ACCESS) {
1419                kvm_err("Unsupported FSC: EC=%#x xFSC=%#lx ESR_EL2=%#lx\n",
1420                        kvm_vcpu_trap_get_class(vcpu),
1421                        (unsigned long)kvm_vcpu_trap_get_fault(vcpu),
1422                        (unsigned long)kvm_vcpu_get_hsr(vcpu));
1423                return -EFAULT;
1424        }
1425
1426        idx = srcu_read_lock(&vcpu->kvm->srcu);
1427
1428        gfn = fault_ipa >> PAGE_SHIFT;
1429        memslot = gfn_to_memslot(vcpu->kvm, gfn);
1430        hva = gfn_to_hva_memslot_prot(memslot, gfn, &writable);
1431        write_fault = kvm_is_write_fault(vcpu);
1432        if (kvm_is_error_hva(hva) || (write_fault && !writable)) {
1433                if (is_iabt) {
1434                        /* Prefetch Abort on I/O address */
1435                        kvm_inject_pabt(vcpu, kvm_vcpu_get_hfar(vcpu));
1436                        ret = 1;
1437                        goto out_unlock;
1438                }
1439
1440                /*
1441                 * The IPA is reported as [MAX:12], so we need to
1442                 * complement it with the bottom 12 bits from the
1443                 * faulting VA. This is always 12 bits, irrespective
1444                 * of the page size.
1445                 */
1446                fault_ipa |= kvm_vcpu_get_hfar(vcpu) & ((1 << 12) - 1);
1447                ret = io_mem_abort(vcpu, run, fault_ipa);
1448                goto out_unlock;
1449        }
1450
1451        /* Userspace should not be able to register out-of-bounds IPAs */
1452        VM_BUG_ON(fault_ipa >= KVM_PHYS_SIZE);
1453
1454        if (fault_status == FSC_ACCESS) {
1455                handle_access_fault(vcpu, fault_ipa);
1456                ret = 1;
1457                goto out_unlock;
1458        }
1459
1460        ret = user_mem_abort(vcpu, fault_ipa, memslot, hva, fault_status);
1461        if (ret == 0)
1462                ret = 1;
1463out_unlock:
1464        srcu_read_unlock(&vcpu->kvm->srcu, idx);
1465        return ret;
1466}
1467
1468static int handle_hva_to_gpa(struct kvm *kvm,
1469                             unsigned long start,
1470                             unsigned long end,
1471                             int (*handler)(struct kvm *kvm,
1472                                            gpa_t gpa, void *data),
1473                             void *data)
1474{
1475        struct kvm_memslots *slots;
1476        struct kvm_memory_slot *memslot;
1477        int ret = 0;
1478
1479        slots = kvm_memslots(kvm);
1480
1481        /* we only care about the pages that the guest sees */
1482        kvm_for_each_memslot(memslot, slots) {
1483                unsigned long hva_start, hva_end;
1484                gfn_t gfn, gfn_end;
1485
1486                hva_start = max(start, memslot->userspace_addr);
1487                hva_end = min(end, memslot->userspace_addr +
1488                                        (memslot->npages << PAGE_SHIFT));
1489                if (hva_start >= hva_end)
1490                        continue;
1491
1492                /*
1493                 * {gfn(page) | page intersects with [hva_start, hva_end)} =
1494                 * {gfn_start, gfn_start+1, ..., gfn_end-1}.
1495                 */
1496                gfn = hva_to_gfn_memslot(hva_start, memslot);
1497                gfn_end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, memslot);
1498
1499                for (; gfn < gfn_end; ++gfn) {
1500                        gpa_t gpa = gfn << PAGE_SHIFT;
1501                        ret |= handler(kvm, gpa, data);
1502                }
1503        }
1504
1505        return ret;
1506}
1507
1508static int kvm_unmap_hva_handler(struct kvm *kvm, gpa_t gpa, void *data)
1509{
1510        unmap_stage2_range(kvm, gpa, PAGE_SIZE);
1511        return 0;
1512}
1513
1514int kvm_unmap_hva(struct kvm *kvm, unsigned long hva)
1515{
1516        unsigned long end = hva + PAGE_SIZE;
1517
1518        if (!kvm->arch.pgd)
1519                return 0;
1520
1521        trace_kvm_unmap_hva(hva);
1522        handle_hva_to_gpa(kvm, hva, end, &kvm_unmap_hva_handler, NULL);
1523        return 0;
1524}
1525
1526int kvm_unmap_hva_range(struct kvm *kvm,
1527                        unsigned long start, unsigned long end)
1528{
1529        if (!kvm->arch.pgd)
1530                return 0;
1531
1532        trace_kvm_unmap_hva_range(start, end);
1533        handle_hva_to_gpa(kvm, start, end, &kvm_unmap_hva_handler, NULL);
1534        return 0;
1535}
1536
1537static int kvm_set_spte_handler(struct kvm *kvm, gpa_t gpa, void *data)
1538{
1539        pte_t *pte = (pte_t *)data;
1540
1541        /*
1542         * We can always call stage2_set_pte with KVM_S2PTE_FLAG_LOGGING_ACTIVE
1543         * flag clear because MMU notifiers will have unmapped a huge PMD before
1544         * calling ->change_pte() (which in turn calls kvm_set_spte_hva()) and
1545         * therefore stage2_set_pte() never needs to clear out a huge PMD
1546         * through this calling path.
1547         */
1548        stage2_set_pte(kvm, NULL, gpa, pte, 0);
1549        return 0;
1550}
1551
1552
1553void kvm_set_spte_hva(struct kvm *kvm, unsigned long hva, pte_t pte)
1554{
1555        unsigned long end = hva + PAGE_SIZE;
1556        pte_t stage2_pte;
1557
1558        if (!kvm->arch.pgd)
1559                return;
1560
1561        trace_kvm_set_spte_hva(hva);
1562        stage2_pte = pfn_pte(pte_pfn(pte), PAGE_S2);
1563        handle_hva_to_gpa(kvm, hva, end, &kvm_set_spte_handler, &stage2_pte);
1564}
1565
1566static int kvm_age_hva_handler(struct kvm *kvm, gpa_t gpa, void *data)
1567{
1568        pmd_t *pmd;
1569        pte_t *pte;
1570
1571        pmd = stage2_get_pmd(kvm, NULL, gpa);
1572        if (!pmd || pmd_none(*pmd))     /* Nothing there */
1573                return 0;
1574
1575        if (kvm_pmd_huge(*pmd)) {       /* THP, HugeTLB */
1576                if (pmd_young(*pmd)) {
1577                        *pmd = pmd_mkold(*pmd);
1578                        return 1;
1579                }
1580
1581                return 0;
1582        }
1583
1584        pte = pte_offset_kernel(pmd, gpa);
1585        if (pte_none(*pte))
1586                return 0;
1587
1588        if (pte_young(*pte)) {
1589                *pte = pte_mkold(*pte); /* Just a page... */
1590                return 1;
1591        }
1592
1593        return 0;
1594}
1595
1596static int kvm_test_age_hva_handler(struct kvm *kvm, gpa_t gpa, void *data)
1597{
1598        pmd_t *pmd;
1599        pte_t *pte;
1600
1601        pmd = stage2_get_pmd(kvm, NULL, gpa);
1602        if (!pmd || pmd_none(*pmd))     /* Nothing there */
1603                return 0;
1604
1605        if (kvm_pmd_huge(*pmd))         /* THP, HugeTLB */
1606                return pmd_young(*pmd);
1607
1608        pte = pte_offset_kernel(pmd, gpa);
1609        if (!pte_none(*pte))            /* Just a page... */
1610                return pte_young(*pte);
1611
1612        return 0;
1613}
1614
1615int kvm_age_hva(struct kvm *kvm, unsigned long start, unsigned long end)
1616{
1617        trace_kvm_age_hva(start, end);
1618        return handle_hva_to_gpa(kvm, start, end, kvm_age_hva_handler, NULL);
1619}
1620
1621int kvm_test_age_hva(struct kvm *kvm, unsigned long hva)
1622{
1623        trace_kvm_test_age_hva(hva);
1624        return handle_hva_to_gpa(kvm, hva, hva, kvm_test_age_hva_handler, NULL);
1625}
1626
1627void kvm_mmu_free_memory_caches(struct kvm_vcpu *vcpu)
1628{
1629        mmu_free_memory_cache(&vcpu->arch.mmu_page_cache);
1630}
1631
1632phys_addr_t kvm_mmu_get_httbr(void)
1633{
1634        if (__kvm_cpu_uses_extended_idmap())
1635                return virt_to_phys(merged_hyp_pgd);
1636        else
1637                return virt_to_phys(hyp_pgd);
1638}
1639
1640phys_addr_t kvm_mmu_get_boot_httbr(void)
1641{
1642        if (__kvm_cpu_uses_extended_idmap())
1643                return virt_to_phys(merged_hyp_pgd);
1644        else
1645                return virt_to_phys(boot_hyp_pgd);
1646}
1647
1648phys_addr_t kvm_get_idmap_vector(void)
1649{
1650        return hyp_idmap_vector;
1651}
1652
1653int kvm_mmu_init(void)
1654{
1655        int err;
1656
1657        hyp_idmap_start = kvm_virt_to_phys(__hyp_idmap_text_start);
1658        hyp_idmap_end = kvm_virt_to_phys(__hyp_idmap_text_end);
1659        hyp_idmap_vector = kvm_virt_to_phys(__kvm_hyp_init);
1660
1661        /*
1662         * We rely on the linker script to ensure at build time that the HYP
1663         * init code does not cross a page boundary.
1664         */
1665        BUG_ON((hyp_idmap_start ^ (hyp_idmap_end - 1)) & PAGE_MASK);
1666
1667        hyp_pgd = (pgd_t *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, hyp_pgd_order);
1668        boot_hyp_pgd = (pgd_t *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, hyp_pgd_order);
1669
1670        if (!hyp_pgd || !boot_hyp_pgd) {
1671                kvm_err("Hyp mode PGD not allocated\n");
1672                err = -ENOMEM;
1673                goto out;
1674        }
1675
1676        /* Create the idmap in the boot page tables */
1677        err =   __create_hyp_mappings(boot_hyp_pgd,
1678                                      hyp_idmap_start, hyp_idmap_end,
1679                                      __phys_to_pfn(hyp_idmap_start),
1680                                      PAGE_HYP);
1681
1682        if (err) {
1683                kvm_err("Failed to idmap %lx-%lx\n",
1684                        hyp_idmap_start, hyp_idmap_end);
1685                goto out;
1686        }
1687
1688        if (__kvm_cpu_uses_extended_idmap()) {
1689                merged_hyp_pgd = (pgd_t *)__get_free_page(GFP_KERNEL | __GFP_ZERO);
1690                if (!merged_hyp_pgd) {
1691                        kvm_err("Failed to allocate extra HYP pgd\n");
1692                        goto out;
1693                }
1694                __kvm_extend_hypmap(boot_hyp_pgd, hyp_pgd, merged_hyp_pgd,
1695                                    hyp_idmap_start);
1696                return 0;
1697        }
1698
1699        /* Map the very same page at the trampoline VA */
1700        err =   __create_hyp_mappings(boot_hyp_pgd,
1701                                      TRAMPOLINE_VA, TRAMPOLINE_VA + PAGE_SIZE,
1702                                      __phys_to_pfn(hyp_idmap_start),
1703                                      PAGE_HYP);
1704        if (err) {
1705                kvm_err("Failed to map trampoline @%lx into boot HYP pgd\n",
1706                        TRAMPOLINE_VA);
1707                goto out;
1708        }
1709
1710        /* Map the same page again into the runtime page tables */
1711        err =   __create_hyp_mappings(hyp_pgd,
1712                                      TRAMPOLINE_VA, TRAMPOLINE_VA + PAGE_SIZE,
1713                                      __phys_to_pfn(hyp_idmap_start),
1714                                      PAGE_HYP);
1715        if (err) {
1716                kvm_err("Failed to map trampoline @%lx into runtime HYP pgd\n",
1717                        TRAMPOLINE_VA);
1718                goto out;
1719        }
1720
1721        return 0;
1722out:
1723        free_hyp_pgds();
1724        return err;
1725}
1726
1727void kvm_arch_commit_memory_region(struct kvm *kvm,
1728                                   const struct kvm_userspace_memory_region *mem,
1729                                   const struct kvm_memory_slot *old,
1730                                   const struct kvm_memory_slot *new,
1731                                   enum kvm_mr_change change)
1732{
1733        /*
1734         * At this point memslot has been committed and there is an
1735         * allocated dirty_bitmap[], dirty pages will be be tracked while the
1736         * memory slot is write protected.
1737         */
1738        if (change != KVM_MR_DELETE && mem->flags & KVM_MEM_LOG_DIRTY_PAGES)
1739                kvm_mmu_wp_memory_region(kvm, mem->slot);
1740}
1741
1742int kvm_arch_prepare_memory_region(struct kvm *kvm,
1743                                   struct kvm_memory_slot *memslot,
1744                                   const struct kvm_userspace_memory_region *mem,
1745                                   enum kvm_mr_change change)
1746{
1747        hva_t hva = mem->userspace_addr;
1748        hva_t reg_end = hva + mem->memory_size;
1749        bool writable = !(mem->flags & KVM_MEM_READONLY);
1750        int ret = 0;
1751
1752        if (change != KVM_MR_CREATE && change != KVM_MR_MOVE &&
1753                        change != KVM_MR_FLAGS_ONLY)
1754                return 0;
1755
1756        /*
1757         * Prevent userspace from creating a memory region outside of the IPA
1758         * space addressable by the KVM guest IPA space.
1759         */
1760        if (memslot->base_gfn + memslot->npages >=
1761            (KVM_PHYS_SIZE >> PAGE_SHIFT))
1762                return -EFAULT;
1763
1764        down_read(&current->mm->mmap_sem);
1765        /*
1766         * A memory region could potentially cover multiple VMAs, and any holes
1767         * between them, so iterate over all of them to find out if we can map
1768         * any of them right now.
1769         *
1770         *     +--------------------------------------------+
1771         * +---------------+----------------+   +----------------+
1772         * |   : VMA 1     |      VMA 2     |   |    VMA 3  :    |
1773         * +---------------+----------------+   +----------------+
1774         *     |               memory region                |
1775         *     +--------------------------------------------+
1776         */
1777        do {
1778                struct vm_area_struct *vma = find_vma(current->mm, hva);
1779                hva_t vm_start, vm_end;
1780
1781                if (!vma || vma->vm_start >= reg_end)
1782                        break;
1783
1784                /*
1785                 * Mapping a read-only VMA is only allowed if the
1786                 * memory region is configured as read-only.
1787                 */
1788                if (writable && !(vma->vm_flags & VM_WRITE)) {
1789                        ret = -EPERM;
1790                        break;
1791                }
1792
1793                /*
1794                 * Take the intersection of this VMA with the memory region
1795                 */
1796                vm_start = max(hva, vma->vm_start);
1797                vm_end = min(reg_end, vma->vm_end);
1798
1799                if (vma->vm_flags & VM_PFNMAP) {
1800                        gpa_t gpa = mem->guest_phys_addr +
1801                                    (vm_start - mem->userspace_addr);
1802                        phys_addr_t pa;
1803
1804                        pa = (phys_addr_t)vma->vm_pgoff << PAGE_SHIFT;
1805                        pa += vm_start - vma->vm_start;
1806
1807                        /* IO region dirty page logging not allowed */
1808                        if (memslot->flags & KVM_MEM_LOG_DIRTY_PAGES) {
1809                                ret = -EINVAL;
1810                                goto out;
1811                        }
1812
1813                        ret = kvm_phys_addr_ioremap(kvm, gpa, pa,
1814                                                    vm_end - vm_start,
1815                                                    writable);
1816                        if (ret)
1817                                break;
1818                }
1819                hva = vm_end;
1820        } while (hva < reg_end);
1821
1822        if (change == KVM_MR_FLAGS_ONLY)
1823                goto out;
1824
1825        spin_lock(&kvm->mmu_lock);
1826        if (ret)
1827                unmap_stage2_range(kvm, mem->guest_phys_addr, mem->memory_size);
1828        else
1829                stage2_flush_memslot(kvm, memslot);
1830        spin_unlock(&kvm->mmu_lock);
1831out:
1832        up_read(&current->mm->mmap_sem);
1833        return ret;
1834}
1835
1836void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
1837                           struct kvm_memory_slot *dont)
1838{
1839}
1840
1841int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
1842                            unsigned long npages)
1843{
1844        /*
1845         * Readonly memslots are not incoherent with the caches by definition,
1846         * but in practice, they are used mostly to emulate ROMs or NOR flashes
1847         * that the guest may consider devices and hence map as uncached.
1848         * To prevent incoherency issues in these cases, tag all readonly
1849         * regions as incoherent.
1850         */
1851        if (slot->flags & KVM_MEM_READONLY)
1852                slot->flags |= KVM_MEMSLOT_INCOHERENT;
1853        return 0;
1854}
1855
1856void kvm_arch_memslots_updated(struct kvm *kvm, struct kvm_memslots *slots)
1857{
1858}
1859
1860void kvm_arch_flush_shadow_all(struct kvm *kvm)
1861{
1862        kvm_free_stage2_pgd(kvm);
1863}
1864
1865void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
1866                                   struct kvm_memory_slot *slot)
1867{
1868        gpa_t gpa = slot->base_gfn << PAGE_SHIFT;
1869        phys_addr_t size = slot->npages << PAGE_SHIFT;
1870
1871        spin_lock(&kvm->mmu_lock);
1872        unmap_stage2_range(kvm, gpa, size);
1873        spin_unlock(&kvm->mmu_lock);
1874}
1875
1876/*
1877 * See note at ARMv7 ARM B1.14.4 (TL;DR: S/W ops are not easily virtualized).
1878 *
1879 * Main problems:
1880 * - S/W ops are local to a CPU (not broadcast)
1881 * - We have line migration behind our back (speculation)
1882 * - System caches don't support S/W at all (damn!)
1883 *
1884 * In the face of the above, the best we can do is to try and convert
1885 * S/W ops to VA ops. Because the guest is not allowed to infer the
1886 * S/W to PA mapping, it can only use S/W to nuke the whole cache,
1887 * which is a rather good thing for us.
1888 *
1889 * Also, it is only used when turning caches on/off ("The expected
1890 * usage of the cache maintenance instructions that operate by set/way
1891 * is associated with the cache maintenance instructions associated
1892 * with the powerdown and powerup of caches, if this is required by
1893 * the implementation.").
1894 *
1895 * We use the following policy:
1896 *
1897 * - If we trap a S/W operation, we enable VM trapping to detect
1898 *   caches being turned on/off, and do a full clean.
1899 *
1900 * - We flush the caches on both caches being turned on and off.
1901 *
1902 * - Once the caches are enabled, we stop trapping VM ops.
1903 */
1904void kvm_set_way_flush(struct kvm_vcpu *vcpu)
1905{
1906        unsigned long hcr = vcpu_get_hcr(vcpu);
1907
1908        /*
1909         * If this is the first time we do a S/W operation
1910         * (i.e. HCR_TVM not set) flush the whole memory, and set the
1911         * VM trapping.
1912         *
1913         * Otherwise, rely on the VM trapping to wait for the MMU +
1914         * Caches to be turned off. At that point, we'll be able to
1915         * clean the caches again.
1916         */
1917        if (!(hcr & HCR_TVM)) {
1918                trace_kvm_set_way_flush(*vcpu_pc(vcpu),
1919                                        vcpu_has_cache_enabled(vcpu));
1920                stage2_flush_vm(vcpu->kvm);
1921                vcpu_set_hcr(vcpu, hcr | HCR_TVM);
1922        }
1923}
1924
1925void kvm_toggle_cache(struct kvm_vcpu *vcpu, bool was_enabled)
1926{
1927        bool now_enabled = vcpu_has_cache_enabled(vcpu);
1928
1929        /*
1930         * If switching the MMU+caches on, need to invalidate the caches.
1931         * If switching it off, need to clean the caches.
1932         * Clean + invalidate does the trick always.
1933         */
1934        if (now_enabled != was_enabled)
1935                stage2_flush_vm(vcpu->kvm);
1936
1937        /* Caches are now on, stop trapping VM ops (until a S/W op) */
1938        if (now_enabled)
1939                vcpu_set_hcr(vcpu, vcpu_get_hcr(vcpu) & ~HCR_TVM);
1940
1941        trace_kvm_toggle_cache(*vcpu_pc(vcpu), was_enabled, now_enabled);
1942}
Note: See TracBrowser for help on using the repository browser.