Merge tag 'locking-urgent-2020-12-27' of git://git.kernel.org/pub/scm/linux/kernel...
[linux/fpc-iii.git] / arch / mips / kvm / mmu.c
blob3dabeda82458678c8995d3ad84ad4c7270c0f346
1 /*
2 * This file is subject to the terms and conditions of the GNU General Public
3 * License. See the file "COPYING" in the main directory of this archive
4 * for more details.
6 * KVM/MIPS MMU handling in the KVM module.
8 * Copyright (C) 2012 MIPS Technologies, Inc. All rights reserved.
9 * Authors: Sanjay Lal <sanjayl@kymasys.com>
12 #include <linux/highmem.h>
13 #include <linux/kvm_host.h>
14 #include <linux/uaccess.h>
15 #include <asm/mmu_context.h>
16 #include <asm/pgalloc.h>
19 * KVM_MMU_CACHE_MIN_PAGES is the number of GPA page table translation levels
20 * for which pages need to be cached.
22 #if defined(__PAGETABLE_PMD_FOLDED)
23 #define KVM_MMU_CACHE_MIN_PAGES 1
24 #else
25 #define KVM_MMU_CACHE_MIN_PAGES 2
26 #endif
28 void kvm_mmu_free_memory_caches(struct kvm_vcpu *vcpu)
30 kvm_mmu_free_memory_cache(&vcpu->arch.mmu_page_cache);
33 /**
34 * kvm_pgd_init() - Initialise KVM GPA page directory.
35 * @page: Pointer to page directory (PGD) for KVM GPA.
37 * Initialise a KVM GPA page directory with pointers to the invalid table, i.e.
38 * representing no mappings. This is similar to pgd_init(), however it
39 * initialises all the page directory pointers, not just the ones corresponding
40 * to the userland address space (since it is for the guest physical address
41 * space rather than a virtual address space).
43 static void kvm_pgd_init(void *page)
45 unsigned long *p, *end;
46 unsigned long entry;
48 #ifdef __PAGETABLE_PMD_FOLDED
49 entry = (unsigned long)invalid_pte_table;
50 #else
51 entry = (unsigned long)invalid_pmd_table;
52 #endif
54 p = (unsigned long *)page;
55 end = p + PTRS_PER_PGD;
57 do {
58 p[0] = entry;
59 p[1] = entry;
60 p[2] = entry;
61 p[3] = entry;
62 p[4] = entry;
63 p += 8;
64 p[-3] = entry;
65 p[-2] = entry;
66 p[-1] = entry;
67 } while (p != end);
70 /**
71 * kvm_pgd_alloc() - Allocate and initialise a KVM GPA page directory.
73 * Allocate a blank KVM GPA page directory (PGD) for representing guest physical
74 * to host physical page mappings.
76 * Returns: Pointer to new KVM GPA page directory.
77 * NULL on allocation failure.
79 pgd_t *kvm_pgd_alloc(void)
81 pgd_t *ret;
83 ret = (pgd_t *)__get_free_pages(GFP_KERNEL, PGD_ORDER);
84 if (ret)
85 kvm_pgd_init(ret);
87 return ret;
90 /**
91 * kvm_mips_walk_pgd() - Walk page table with optional allocation.
92 * @pgd: Page directory pointer.
93 * @addr: Address to index page table using.
94 * @cache: MMU page cache to allocate new page tables from, or NULL.
96 * Walk the page tables pointed to by @pgd to find the PTE corresponding to the
97 * address @addr. If page tables don't exist for @addr, they will be created
98 * from the MMU cache if @cache is not NULL.
100 * Returns: Pointer to pte_t corresponding to @addr.
101 * NULL if a page table doesn't exist for @addr and !@cache.
102 * NULL if a page table allocation failed.
104 static pte_t *kvm_mips_walk_pgd(pgd_t *pgd, struct kvm_mmu_memory_cache *cache,
105 unsigned long addr)
107 p4d_t *p4d;
108 pud_t *pud;
109 pmd_t *pmd;
111 pgd += pgd_index(addr);
112 if (pgd_none(*pgd)) {
113 /* Not used on MIPS yet */
114 BUG();
115 return NULL;
117 p4d = p4d_offset(pgd, addr);
118 pud = pud_offset(p4d, addr);
119 if (pud_none(*pud)) {
120 pmd_t *new_pmd;
122 if (!cache)
123 return NULL;
124 new_pmd = kvm_mmu_memory_cache_alloc(cache);
125 pmd_init((unsigned long)new_pmd,
126 (unsigned long)invalid_pte_table);
127 pud_populate(NULL, pud, new_pmd);
129 pmd = pmd_offset(pud, addr);
130 if (pmd_none(*pmd)) {
131 pte_t *new_pte;
133 if (!cache)
134 return NULL;
135 new_pte = kvm_mmu_memory_cache_alloc(cache);
136 clear_page(new_pte);
137 pmd_populate_kernel(NULL, pmd, new_pte);
139 return pte_offset_kernel(pmd, addr);
142 /* Caller must hold kvm->mm_lock */
143 static pte_t *kvm_mips_pte_for_gpa(struct kvm *kvm,
144 struct kvm_mmu_memory_cache *cache,
145 unsigned long addr)
147 return kvm_mips_walk_pgd(kvm->arch.gpa_mm.pgd, cache, addr);
151 * kvm_mips_flush_gpa_{pte,pmd,pud,pgd,pt}.
152 * Flush a range of guest physical address space from the VM's GPA page tables.
155 static bool kvm_mips_flush_gpa_pte(pte_t *pte, unsigned long start_gpa,
156 unsigned long end_gpa)
158 int i_min = pte_index(start_gpa);
159 int i_max = pte_index(end_gpa);
160 bool safe_to_remove = (i_min == 0 && i_max == PTRS_PER_PTE - 1);
161 int i;
163 for (i = i_min; i <= i_max; ++i) {
164 if (!pte_present(pte[i]))
165 continue;
167 set_pte(pte + i, __pte(0));
169 return safe_to_remove;
172 static bool kvm_mips_flush_gpa_pmd(pmd_t *pmd, unsigned long start_gpa,
173 unsigned long end_gpa)
175 pte_t *pte;
176 unsigned long end = ~0ul;
177 int i_min = pmd_index(start_gpa);
178 int i_max = pmd_index(end_gpa);
179 bool safe_to_remove = (i_min == 0 && i_max == PTRS_PER_PMD - 1);
180 int i;
182 for (i = i_min; i <= i_max; ++i, start_gpa = 0) {
183 if (!pmd_present(pmd[i]))
184 continue;
186 pte = pte_offset_kernel(pmd + i, 0);
187 if (i == i_max)
188 end = end_gpa;
190 if (kvm_mips_flush_gpa_pte(pte, start_gpa, end)) {
191 pmd_clear(pmd + i);
192 pte_free_kernel(NULL, pte);
193 } else {
194 safe_to_remove = false;
197 return safe_to_remove;
200 static bool kvm_mips_flush_gpa_pud(pud_t *pud, unsigned long start_gpa,
201 unsigned long end_gpa)
203 pmd_t *pmd;
204 unsigned long end = ~0ul;
205 int i_min = pud_index(start_gpa);
206 int i_max = pud_index(end_gpa);
207 bool safe_to_remove = (i_min == 0 && i_max == PTRS_PER_PUD - 1);
208 int i;
210 for (i = i_min; i <= i_max; ++i, start_gpa = 0) {
211 if (!pud_present(pud[i]))
212 continue;
214 pmd = pmd_offset(pud + i, 0);
215 if (i == i_max)
216 end = end_gpa;
218 if (kvm_mips_flush_gpa_pmd(pmd, start_gpa, end)) {
219 pud_clear(pud + i);
220 pmd_free(NULL, pmd);
221 } else {
222 safe_to_remove = false;
225 return safe_to_remove;
228 static bool kvm_mips_flush_gpa_pgd(pgd_t *pgd, unsigned long start_gpa,
229 unsigned long end_gpa)
231 p4d_t *p4d;
232 pud_t *pud;
233 unsigned long end = ~0ul;
234 int i_min = pgd_index(start_gpa);
235 int i_max = pgd_index(end_gpa);
236 bool safe_to_remove = (i_min == 0 && i_max == PTRS_PER_PGD - 1);
237 int i;
239 for (i = i_min; i <= i_max; ++i, start_gpa = 0) {
240 if (!pgd_present(pgd[i]))
241 continue;
243 p4d = p4d_offset(pgd, 0);
244 pud = pud_offset(p4d + i, 0);
245 if (i == i_max)
246 end = end_gpa;
248 if (kvm_mips_flush_gpa_pud(pud, start_gpa, end)) {
249 pgd_clear(pgd + i);
250 pud_free(NULL, pud);
251 } else {
252 safe_to_remove = false;
255 return safe_to_remove;
259 * kvm_mips_flush_gpa_pt() - Flush a range of guest physical addresses.
260 * @kvm: KVM pointer.
261 * @start_gfn: Guest frame number of first page in GPA range to flush.
262 * @end_gfn: Guest frame number of last page in GPA range to flush.
264 * Flushes a range of GPA mappings from the GPA page tables.
266 * The caller must hold the @kvm->mmu_lock spinlock.
268 * Returns: Whether its safe to remove the top level page directory because
269 * all lower levels have been removed.
271 bool kvm_mips_flush_gpa_pt(struct kvm *kvm, gfn_t start_gfn, gfn_t end_gfn)
273 return kvm_mips_flush_gpa_pgd(kvm->arch.gpa_mm.pgd,
274 start_gfn << PAGE_SHIFT,
275 end_gfn << PAGE_SHIFT);
278 #define BUILD_PTE_RANGE_OP(name, op) \
279 static int kvm_mips_##name##_pte(pte_t *pte, unsigned long start, \
280 unsigned long end) \
282 int ret = 0; \
283 int i_min = pte_index(start); \
284 int i_max = pte_index(end); \
285 int i; \
286 pte_t old, new; \
288 for (i = i_min; i <= i_max; ++i) { \
289 if (!pte_present(pte[i])) \
290 continue; \
292 old = pte[i]; \
293 new = op(old); \
294 if (pte_val(new) == pte_val(old)) \
295 continue; \
296 set_pte(pte + i, new); \
297 ret = 1; \
299 return ret; \
302 /* returns true if anything was done */ \
303 static int kvm_mips_##name##_pmd(pmd_t *pmd, unsigned long start, \
304 unsigned long end) \
306 int ret = 0; \
307 pte_t *pte; \
308 unsigned long cur_end = ~0ul; \
309 int i_min = pmd_index(start); \
310 int i_max = pmd_index(end); \
311 int i; \
313 for (i = i_min; i <= i_max; ++i, start = 0) { \
314 if (!pmd_present(pmd[i])) \
315 continue; \
317 pte = pte_offset_kernel(pmd + i, 0); \
318 if (i == i_max) \
319 cur_end = end; \
321 ret |= kvm_mips_##name##_pte(pte, start, cur_end); \
323 return ret; \
326 static int kvm_mips_##name##_pud(pud_t *pud, unsigned long start, \
327 unsigned long end) \
329 int ret = 0; \
330 pmd_t *pmd; \
331 unsigned long cur_end = ~0ul; \
332 int i_min = pud_index(start); \
333 int i_max = pud_index(end); \
334 int i; \
336 for (i = i_min; i <= i_max; ++i, start = 0) { \
337 if (!pud_present(pud[i])) \
338 continue; \
340 pmd = pmd_offset(pud + i, 0); \
341 if (i == i_max) \
342 cur_end = end; \
344 ret |= kvm_mips_##name##_pmd(pmd, start, cur_end); \
346 return ret; \
349 static int kvm_mips_##name##_pgd(pgd_t *pgd, unsigned long start, \
350 unsigned long end) \
352 int ret = 0; \
353 p4d_t *p4d; \
354 pud_t *pud; \
355 unsigned long cur_end = ~0ul; \
356 int i_min = pgd_index(start); \
357 int i_max = pgd_index(end); \
358 int i; \
360 for (i = i_min; i <= i_max; ++i, start = 0) { \
361 if (!pgd_present(pgd[i])) \
362 continue; \
364 p4d = p4d_offset(pgd, 0); \
365 pud = pud_offset(p4d + i, 0); \
366 if (i == i_max) \
367 cur_end = end; \
369 ret |= kvm_mips_##name##_pud(pud, start, cur_end); \
371 return ret; \
375 * kvm_mips_mkclean_gpa_pt.
376 * Mark a range of guest physical address space clean (writes fault) in the VM's
377 * GPA page table to allow dirty page tracking.
380 BUILD_PTE_RANGE_OP(mkclean, pte_mkclean)
383 * kvm_mips_mkclean_gpa_pt() - Make a range of guest physical addresses clean.
384 * @kvm: KVM pointer.
385 * @start_gfn: Guest frame number of first page in GPA range to flush.
386 * @end_gfn: Guest frame number of last page in GPA range to flush.
388 * Make a range of GPA mappings clean so that guest writes will fault and
389 * trigger dirty page logging.
391 * The caller must hold the @kvm->mmu_lock spinlock.
393 * Returns: Whether any GPA mappings were modified, which would require
394 * derived mappings (GVA page tables & TLB enties) to be
395 * invalidated.
397 int kvm_mips_mkclean_gpa_pt(struct kvm *kvm, gfn_t start_gfn, gfn_t end_gfn)
399 return kvm_mips_mkclean_pgd(kvm->arch.gpa_mm.pgd,
400 start_gfn << PAGE_SHIFT,
401 end_gfn << PAGE_SHIFT);
405 * kvm_arch_mmu_enable_log_dirty_pt_masked() - write protect dirty pages
406 * @kvm: The KVM pointer
407 * @slot: The memory slot associated with mask
408 * @gfn_offset: The gfn offset in memory slot
409 * @mask: The mask of dirty pages at offset 'gfn_offset' in this memory
410 * slot to be write protected
412 * Walks bits set in mask write protects the associated pte's. Caller must
413 * acquire @kvm->mmu_lock.
415 void kvm_arch_mmu_enable_log_dirty_pt_masked(struct kvm *kvm,
416 struct kvm_memory_slot *slot,
417 gfn_t gfn_offset, unsigned long mask)
419 gfn_t base_gfn = slot->base_gfn + gfn_offset;
420 gfn_t start = base_gfn + __ffs(mask);
421 gfn_t end = base_gfn + __fls(mask);
423 kvm_mips_mkclean_gpa_pt(kvm, start, end);
427 * kvm_mips_mkold_gpa_pt.
428 * Mark a range of guest physical address space old (all accesses fault) in the
429 * VM's GPA page table to allow detection of commonly used pages.
432 BUILD_PTE_RANGE_OP(mkold, pte_mkold)
434 static int kvm_mips_mkold_gpa_pt(struct kvm *kvm, gfn_t start_gfn,
435 gfn_t end_gfn)
437 return kvm_mips_mkold_pgd(kvm->arch.gpa_mm.pgd,
438 start_gfn << PAGE_SHIFT,
439 end_gfn << PAGE_SHIFT);
442 static int handle_hva_to_gpa(struct kvm *kvm,
443 unsigned long start,
444 unsigned long end,
445 int (*handler)(struct kvm *kvm, gfn_t gfn,
446 gpa_t gfn_end,
447 struct kvm_memory_slot *memslot,
448 void *data),
449 void *data)
451 struct kvm_memslots *slots;
452 struct kvm_memory_slot *memslot;
453 int ret = 0;
455 slots = kvm_memslots(kvm);
457 /* we only care about the pages that the guest sees */
458 kvm_for_each_memslot(memslot, slots) {
459 unsigned long hva_start, hva_end;
460 gfn_t gfn, gfn_end;
462 hva_start = max(start, memslot->userspace_addr);
463 hva_end = min(end, memslot->userspace_addr +
464 (memslot->npages << PAGE_SHIFT));
465 if (hva_start >= hva_end)
466 continue;
469 * {gfn(page) | page intersects with [hva_start, hva_end)} =
470 * {gfn_start, gfn_start+1, ..., gfn_end-1}.
472 gfn = hva_to_gfn_memslot(hva_start, memslot);
473 gfn_end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, memslot);
475 ret |= handler(kvm, gfn, gfn_end, memslot, data);
478 return ret;
482 static int kvm_unmap_hva_handler(struct kvm *kvm, gfn_t gfn, gfn_t gfn_end,
483 struct kvm_memory_slot *memslot, void *data)
485 kvm_mips_flush_gpa_pt(kvm, gfn, gfn_end);
486 return 1;
489 int kvm_unmap_hva_range(struct kvm *kvm, unsigned long start, unsigned long end,
490 unsigned flags)
492 handle_hva_to_gpa(kvm, start, end, &kvm_unmap_hva_handler, NULL);
494 kvm_mips_callbacks->flush_shadow_all(kvm);
495 return 0;
498 static int kvm_set_spte_handler(struct kvm *kvm, gfn_t gfn, gfn_t gfn_end,
499 struct kvm_memory_slot *memslot, void *data)
501 gpa_t gpa = gfn << PAGE_SHIFT;
502 pte_t hva_pte = *(pte_t *)data;
503 pte_t *gpa_pte = kvm_mips_pte_for_gpa(kvm, NULL, gpa);
504 pte_t old_pte;
506 if (!gpa_pte)
507 return 0;
509 /* Mapping may need adjusting depending on memslot flags */
510 old_pte = *gpa_pte;
511 if (memslot->flags & KVM_MEM_LOG_DIRTY_PAGES && !pte_dirty(old_pte))
512 hva_pte = pte_mkclean(hva_pte);
513 else if (memslot->flags & KVM_MEM_READONLY)
514 hva_pte = pte_wrprotect(hva_pte);
516 set_pte(gpa_pte, hva_pte);
518 /* Replacing an absent or old page doesn't need flushes */
519 if (!pte_present(old_pte) || !pte_young(old_pte))
520 return 0;
522 /* Pages swapped, aged, moved, or cleaned require flushes */
523 return !pte_present(hva_pte) ||
524 !pte_young(hva_pte) ||
525 pte_pfn(old_pte) != pte_pfn(hva_pte) ||
526 (pte_dirty(old_pte) && !pte_dirty(hva_pte));
529 int kvm_set_spte_hva(struct kvm *kvm, unsigned long hva, pte_t pte)
531 unsigned long end = hva + PAGE_SIZE;
532 int ret;
534 ret = handle_hva_to_gpa(kvm, hva, end, &kvm_set_spte_handler, &pte);
535 if (ret)
536 kvm_mips_callbacks->flush_shadow_all(kvm);
537 return 0;
540 static int kvm_age_hva_handler(struct kvm *kvm, gfn_t gfn, gfn_t gfn_end,
541 struct kvm_memory_slot *memslot, void *data)
543 return kvm_mips_mkold_gpa_pt(kvm, gfn, gfn_end);
546 static int kvm_test_age_hva_handler(struct kvm *kvm, gfn_t gfn, gfn_t gfn_end,
547 struct kvm_memory_slot *memslot, void *data)
549 gpa_t gpa = gfn << PAGE_SHIFT;
550 pte_t *gpa_pte = kvm_mips_pte_for_gpa(kvm, NULL, gpa);
552 if (!gpa_pte)
553 return 0;
554 return pte_young(*gpa_pte);
557 int kvm_age_hva(struct kvm *kvm, unsigned long start, unsigned long end)
559 return handle_hva_to_gpa(kvm, start, end, kvm_age_hva_handler, NULL);
562 int kvm_test_age_hva(struct kvm *kvm, unsigned long hva)
564 return handle_hva_to_gpa(kvm, hva, hva, kvm_test_age_hva_handler, NULL);
568 * _kvm_mips_map_page_fast() - Fast path GPA fault handler.
569 * @vcpu: VCPU pointer.
570 * @gpa: Guest physical address of fault.
571 * @write_fault: Whether the fault was due to a write.
572 * @out_entry: New PTE for @gpa (written on success unless NULL).
573 * @out_buddy: New PTE for @gpa's buddy (written on success unless
574 * NULL).
576 * Perform fast path GPA fault handling, doing all that can be done without
577 * calling into KVM. This handles marking old pages young (for idle page
578 * tracking), and dirtying of clean pages (for dirty page logging).
580 * Returns: 0 on success, in which case we can update derived mappings and
581 * resume guest execution.
582 * -EFAULT on failure due to absent GPA mapping or write to
583 * read-only page, in which case KVM must be consulted.
585 static int _kvm_mips_map_page_fast(struct kvm_vcpu *vcpu, unsigned long gpa,
586 bool write_fault,
587 pte_t *out_entry, pte_t *out_buddy)
589 struct kvm *kvm = vcpu->kvm;
590 gfn_t gfn = gpa >> PAGE_SHIFT;
591 pte_t *ptep;
592 kvm_pfn_t pfn = 0; /* silence bogus GCC warning */
593 bool pfn_valid = false;
594 int ret = 0;
596 spin_lock(&kvm->mmu_lock);
598 /* Fast path - just check GPA page table for an existing entry */
599 ptep = kvm_mips_pte_for_gpa(kvm, NULL, gpa);
600 if (!ptep || !pte_present(*ptep)) {
601 ret = -EFAULT;
602 goto out;
605 /* Track access to pages marked old */
606 if (!pte_young(*ptep)) {
607 set_pte(ptep, pte_mkyoung(*ptep));
608 pfn = pte_pfn(*ptep);
609 pfn_valid = true;
610 /* call kvm_set_pfn_accessed() after unlock */
612 if (write_fault && !pte_dirty(*ptep)) {
613 if (!pte_write(*ptep)) {
614 ret = -EFAULT;
615 goto out;
618 /* Track dirtying of writeable pages */
619 set_pte(ptep, pte_mkdirty(*ptep));
620 pfn = pte_pfn(*ptep);
621 mark_page_dirty(kvm, gfn);
622 kvm_set_pfn_dirty(pfn);
625 if (out_entry)
626 *out_entry = *ptep;
627 if (out_buddy)
628 *out_buddy = *ptep_buddy(ptep);
630 out:
631 spin_unlock(&kvm->mmu_lock);
632 if (pfn_valid)
633 kvm_set_pfn_accessed(pfn);
634 return ret;
638 * kvm_mips_map_page() - Map a guest physical page.
639 * @vcpu: VCPU pointer.
640 * @gpa: Guest physical address of fault.
641 * @write_fault: Whether the fault was due to a write.
642 * @out_entry: New PTE for @gpa (written on success unless NULL).
643 * @out_buddy: New PTE for @gpa's buddy (written on success unless
644 * NULL).
646 * Handle GPA faults by creating a new GPA mapping (or updating an existing
647 * one).
649 * This takes care of marking pages young or dirty (idle/dirty page tracking),
650 * asking KVM for the corresponding PFN, and creating a mapping in the GPA page
651 * tables. Derived mappings (GVA page tables and TLBs) must be handled by the
652 * caller.
654 * Returns: 0 on success, in which case the caller may use the @out_entry
655 * and @out_buddy PTEs to update derived mappings and resume guest
656 * execution.
657 * -EFAULT if there is no memory region at @gpa or a write was
658 * attempted to a read-only memory region. This is usually handled
659 * as an MMIO access.
661 static int kvm_mips_map_page(struct kvm_vcpu *vcpu, unsigned long gpa,
662 bool write_fault,
663 pte_t *out_entry, pte_t *out_buddy)
665 struct kvm *kvm = vcpu->kvm;
666 struct kvm_mmu_memory_cache *memcache = &vcpu->arch.mmu_page_cache;
667 gfn_t gfn = gpa >> PAGE_SHIFT;
668 int srcu_idx, err;
669 kvm_pfn_t pfn;
670 pte_t *ptep, entry, old_pte;
671 bool writeable;
672 unsigned long prot_bits;
673 unsigned long mmu_seq;
675 /* Try the fast path to handle old / clean pages */
676 srcu_idx = srcu_read_lock(&kvm->srcu);
677 err = _kvm_mips_map_page_fast(vcpu, gpa, write_fault, out_entry,
678 out_buddy);
679 if (!err)
680 goto out;
682 /* We need a minimum of cached pages ready for page table creation */
683 err = kvm_mmu_topup_memory_cache(memcache, KVM_MMU_CACHE_MIN_PAGES);
684 if (err)
685 goto out;
687 retry:
689 * Used to check for invalidations in progress, of the pfn that is
690 * returned by pfn_to_pfn_prot below.
692 mmu_seq = kvm->mmu_notifier_seq;
694 * Ensure the read of mmu_notifier_seq isn't reordered with PTE reads in
695 * gfn_to_pfn_prot() (which calls get_user_pages()), so that we don't
696 * risk the page we get a reference to getting unmapped before we have a
697 * chance to grab the mmu_lock without mmu_notifier_retry() noticing.
699 * This smp_rmb() pairs with the effective smp_wmb() of the combination
700 * of the pte_unmap_unlock() after the PTE is zapped, and the
701 * spin_lock() in kvm_mmu_notifier_invalidate_<page|range_end>() before
702 * mmu_notifier_seq is incremented.
704 smp_rmb();
706 /* Slow path - ask KVM core whether we can access this GPA */
707 pfn = gfn_to_pfn_prot(kvm, gfn, write_fault, &writeable);
708 if (is_error_noslot_pfn(pfn)) {
709 err = -EFAULT;
710 goto out;
713 spin_lock(&kvm->mmu_lock);
714 /* Check if an invalidation has taken place since we got pfn */
715 if (mmu_notifier_retry(kvm, mmu_seq)) {
717 * This can happen when mappings are changed asynchronously, but
718 * also synchronously if a COW is triggered by
719 * gfn_to_pfn_prot().
721 spin_unlock(&kvm->mmu_lock);
722 kvm_release_pfn_clean(pfn);
723 goto retry;
726 /* Ensure page tables are allocated */
727 ptep = kvm_mips_pte_for_gpa(kvm, memcache, gpa);
729 /* Set up the PTE */
730 prot_bits = _PAGE_PRESENT | __READABLE | _page_cachable_default;
731 if (writeable) {
732 prot_bits |= _PAGE_WRITE;
733 if (write_fault) {
734 prot_bits |= __WRITEABLE;
735 mark_page_dirty(kvm, gfn);
736 kvm_set_pfn_dirty(pfn);
739 entry = pfn_pte(pfn, __pgprot(prot_bits));
741 /* Write the PTE */
742 old_pte = *ptep;
743 set_pte(ptep, entry);
745 err = 0;
746 if (out_entry)
747 *out_entry = *ptep;
748 if (out_buddy)
749 *out_buddy = *ptep_buddy(ptep);
751 spin_unlock(&kvm->mmu_lock);
752 kvm_release_pfn_clean(pfn);
753 kvm_set_pfn_accessed(pfn);
754 out:
755 srcu_read_unlock(&kvm->srcu, srcu_idx);
756 return err;
759 static pte_t *kvm_trap_emul_pte_for_gva(struct kvm_vcpu *vcpu,
760 unsigned long addr)
762 struct kvm_mmu_memory_cache *memcache = &vcpu->arch.mmu_page_cache;
763 pgd_t *pgdp;
764 int ret;
766 /* We need a minimum of cached pages ready for page table creation */
767 ret = kvm_mmu_topup_memory_cache(memcache, KVM_MMU_CACHE_MIN_PAGES);
768 if (ret)
769 return NULL;
771 if (KVM_GUEST_KERNEL_MODE(vcpu))
772 pgdp = vcpu->arch.guest_kernel_mm.pgd;
773 else
774 pgdp = vcpu->arch.guest_user_mm.pgd;
776 return kvm_mips_walk_pgd(pgdp, memcache, addr);
779 void kvm_trap_emul_invalidate_gva(struct kvm_vcpu *vcpu, unsigned long addr,
780 bool user)
782 pgd_t *pgdp;
783 pte_t *ptep;
785 addr &= PAGE_MASK << 1;
787 pgdp = vcpu->arch.guest_kernel_mm.pgd;
788 ptep = kvm_mips_walk_pgd(pgdp, NULL, addr);
789 if (ptep) {
790 ptep[0] = pfn_pte(0, __pgprot(0));
791 ptep[1] = pfn_pte(0, __pgprot(0));
794 if (user) {
795 pgdp = vcpu->arch.guest_user_mm.pgd;
796 ptep = kvm_mips_walk_pgd(pgdp, NULL, addr);
797 if (ptep) {
798 ptep[0] = pfn_pte(0, __pgprot(0));
799 ptep[1] = pfn_pte(0, __pgprot(0));
805 * kvm_mips_flush_gva_{pte,pmd,pud,pgd,pt}.
806 * Flush a range of guest physical address space from the VM's GPA page tables.
809 static bool kvm_mips_flush_gva_pte(pte_t *pte, unsigned long start_gva,
810 unsigned long end_gva)
812 int i_min = pte_index(start_gva);
813 int i_max = pte_index(end_gva);
814 bool safe_to_remove = (i_min == 0 && i_max == PTRS_PER_PTE - 1);
815 int i;
818 * There's no freeing to do, so there's no point clearing individual
819 * entries unless only part of the last level page table needs flushing.
821 if (safe_to_remove)
822 return true;
824 for (i = i_min; i <= i_max; ++i) {
825 if (!pte_present(pte[i]))
826 continue;
828 set_pte(pte + i, __pte(0));
830 return false;
833 static bool kvm_mips_flush_gva_pmd(pmd_t *pmd, unsigned long start_gva,
834 unsigned long end_gva)
836 pte_t *pte;
837 unsigned long end = ~0ul;
838 int i_min = pmd_index(start_gva);
839 int i_max = pmd_index(end_gva);
840 bool safe_to_remove = (i_min == 0 && i_max == PTRS_PER_PMD - 1);
841 int i;
843 for (i = i_min; i <= i_max; ++i, start_gva = 0) {
844 if (!pmd_present(pmd[i]))
845 continue;
847 pte = pte_offset_kernel(pmd + i, 0);
848 if (i == i_max)
849 end = end_gva;
851 if (kvm_mips_flush_gva_pte(pte, start_gva, end)) {
852 pmd_clear(pmd + i);
853 pte_free_kernel(NULL, pte);
854 } else {
855 safe_to_remove = false;
858 return safe_to_remove;
861 static bool kvm_mips_flush_gva_pud(pud_t *pud, unsigned long start_gva,
862 unsigned long end_gva)
864 pmd_t *pmd;
865 unsigned long end = ~0ul;
866 int i_min = pud_index(start_gva);
867 int i_max = pud_index(end_gva);
868 bool safe_to_remove = (i_min == 0 && i_max == PTRS_PER_PUD - 1);
869 int i;
871 for (i = i_min; i <= i_max; ++i, start_gva = 0) {
872 if (!pud_present(pud[i]))
873 continue;
875 pmd = pmd_offset(pud + i, 0);
876 if (i == i_max)
877 end = end_gva;
879 if (kvm_mips_flush_gva_pmd(pmd, start_gva, end)) {
880 pud_clear(pud + i);
881 pmd_free(NULL, pmd);
882 } else {
883 safe_to_remove = false;
886 return safe_to_remove;
889 static bool kvm_mips_flush_gva_pgd(pgd_t *pgd, unsigned long start_gva,
890 unsigned long end_gva)
892 p4d_t *p4d;
893 pud_t *pud;
894 unsigned long end = ~0ul;
895 int i_min = pgd_index(start_gva);
896 int i_max = pgd_index(end_gva);
897 bool safe_to_remove = (i_min == 0 && i_max == PTRS_PER_PGD - 1);
898 int i;
900 for (i = i_min; i <= i_max; ++i, start_gva = 0) {
901 if (!pgd_present(pgd[i]))
902 continue;
904 p4d = p4d_offset(pgd, 0);
905 pud = pud_offset(p4d + i, 0);
906 if (i == i_max)
907 end = end_gva;
909 if (kvm_mips_flush_gva_pud(pud, start_gva, end)) {
910 pgd_clear(pgd + i);
911 pud_free(NULL, pud);
912 } else {
913 safe_to_remove = false;
916 return safe_to_remove;
919 void kvm_mips_flush_gva_pt(pgd_t *pgd, enum kvm_mips_flush flags)
921 if (flags & KMF_GPA) {
922 /* all of guest virtual address space could be affected */
923 if (flags & KMF_KERN)
924 /* useg, kseg0, seg2/3 */
925 kvm_mips_flush_gva_pgd(pgd, 0, 0x7fffffff);
926 else
927 /* useg */
928 kvm_mips_flush_gva_pgd(pgd, 0, 0x3fffffff);
929 } else {
930 /* useg */
931 kvm_mips_flush_gva_pgd(pgd, 0, 0x3fffffff);
933 /* kseg2/3 */
934 if (flags & KMF_KERN)
935 kvm_mips_flush_gva_pgd(pgd, 0x60000000, 0x7fffffff);
939 static pte_t kvm_mips_gpa_pte_to_gva_unmapped(pte_t pte)
942 * Don't leak writeable but clean entries from GPA page tables. We don't
943 * want the normal Linux tlbmod handler to handle dirtying when KVM
944 * accesses guest memory.
946 if (!pte_dirty(pte))
947 pte = pte_wrprotect(pte);
949 return pte;
952 static pte_t kvm_mips_gpa_pte_to_gva_mapped(pte_t pte, long entrylo)
954 /* Guest EntryLo overrides host EntryLo */
955 if (!(entrylo & ENTRYLO_D))
956 pte = pte_mkclean(pte);
958 return kvm_mips_gpa_pte_to_gva_unmapped(pte);
961 #ifdef CONFIG_KVM_MIPS_VZ
962 int kvm_mips_handle_vz_root_tlb_fault(unsigned long badvaddr,
963 struct kvm_vcpu *vcpu,
964 bool write_fault)
966 int ret;
968 ret = kvm_mips_map_page(vcpu, badvaddr, write_fault, NULL, NULL);
969 if (ret)
970 return ret;
972 /* Invalidate this entry in the TLB */
973 return kvm_vz_host_tlb_inv(vcpu, badvaddr);
975 #endif
977 /* XXXKYMA: Must be called with interrupts disabled */
978 int kvm_mips_handle_kseg0_tlb_fault(unsigned long badvaddr,
979 struct kvm_vcpu *vcpu,
980 bool write_fault)
982 unsigned long gpa;
983 pte_t pte_gpa[2], *ptep_gva;
984 int idx;
986 if (KVM_GUEST_KSEGX(badvaddr) != KVM_GUEST_KSEG0) {
987 kvm_err("%s: Invalid BadVaddr: %#lx\n", __func__, badvaddr);
988 kvm_mips_dump_host_tlbs();
989 return -1;
992 /* Get the GPA page table entry */
993 gpa = KVM_GUEST_CPHYSADDR(badvaddr);
994 idx = (badvaddr >> PAGE_SHIFT) & 1;
995 if (kvm_mips_map_page(vcpu, gpa, write_fault, &pte_gpa[idx],
996 &pte_gpa[!idx]) < 0)
997 return -1;
999 /* Get the GVA page table entry */
1000 ptep_gva = kvm_trap_emul_pte_for_gva(vcpu, badvaddr & ~PAGE_SIZE);
1001 if (!ptep_gva) {
1002 kvm_err("No ptep for gva %lx\n", badvaddr);
1003 return -1;
1006 /* Copy a pair of entries from GPA page table to GVA page table */
1007 ptep_gva[0] = kvm_mips_gpa_pte_to_gva_unmapped(pte_gpa[0]);
1008 ptep_gva[1] = kvm_mips_gpa_pte_to_gva_unmapped(pte_gpa[1]);
1010 /* Invalidate this entry in the TLB, guest kernel ASID only */
1011 kvm_mips_host_tlb_inv(vcpu, badvaddr, false, true);
1012 return 0;
1015 int kvm_mips_handle_mapped_seg_tlb_fault(struct kvm_vcpu *vcpu,
1016 struct kvm_mips_tlb *tlb,
1017 unsigned long gva,
1018 bool write_fault)
1020 struct kvm *kvm = vcpu->kvm;
1021 long tlb_lo[2];
1022 pte_t pte_gpa[2], *ptep_buddy, *ptep_gva;
1023 unsigned int idx = TLB_LO_IDX(*tlb, gva);
1024 bool kernel = KVM_GUEST_KERNEL_MODE(vcpu);
1026 tlb_lo[0] = tlb->tlb_lo[0];
1027 tlb_lo[1] = tlb->tlb_lo[1];
1030 * The commpage address must not be mapped to anything else if the guest
1031 * TLB contains entries nearby, or commpage accesses will break.
1033 if (!((gva ^ KVM_GUEST_COMMPAGE_ADDR) & VPN2_MASK & (PAGE_MASK << 1)))
1034 tlb_lo[TLB_LO_IDX(*tlb, KVM_GUEST_COMMPAGE_ADDR)] = 0;
1036 /* Get the GPA page table entry */
1037 if (kvm_mips_map_page(vcpu, mips3_tlbpfn_to_paddr(tlb_lo[idx]),
1038 write_fault, &pte_gpa[idx], NULL) < 0)
1039 return -1;
1041 /* And its GVA buddy's GPA page table entry if it also exists */
1042 pte_gpa[!idx] = pfn_pte(0, __pgprot(0));
1043 if (tlb_lo[!idx] & ENTRYLO_V) {
1044 spin_lock(&kvm->mmu_lock);
1045 ptep_buddy = kvm_mips_pte_for_gpa(kvm, NULL,
1046 mips3_tlbpfn_to_paddr(tlb_lo[!idx]));
1047 if (ptep_buddy)
1048 pte_gpa[!idx] = *ptep_buddy;
1049 spin_unlock(&kvm->mmu_lock);
1052 /* Get the GVA page table entry pair */
1053 ptep_gva = kvm_trap_emul_pte_for_gva(vcpu, gva & ~PAGE_SIZE);
1054 if (!ptep_gva) {
1055 kvm_err("No ptep for gva %lx\n", gva);
1056 return -1;
1059 /* Copy a pair of entries from GPA page table to GVA page table */
1060 ptep_gva[0] = kvm_mips_gpa_pte_to_gva_mapped(pte_gpa[0], tlb_lo[0]);
1061 ptep_gva[1] = kvm_mips_gpa_pte_to_gva_mapped(pte_gpa[1], tlb_lo[1]);
1063 /* Invalidate this entry in the TLB, current guest mode ASID only */
1064 kvm_mips_host_tlb_inv(vcpu, gva, !kernel, kernel);
1066 kvm_debug("@ %#lx tlb_lo0: 0x%08lx tlb_lo1: 0x%08lx\n", vcpu->arch.pc,
1067 tlb->tlb_lo[0], tlb->tlb_lo[1]);
1069 return 0;
1072 int kvm_mips_handle_commpage_tlb_fault(unsigned long badvaddr,
1073 struct kvm_vcpu *vcpu)
1075 kvm_pfn_t pfn;
1076 pte_t *ptep;
1077 pgprot_t prot;
1079 ptep = kvm_trap_emul_pte_for_gva(vcpu, badvaddr);
1080 if (!ptep) {
1081 kvm_err("No ptep for commpage %lx\n", badvaddr);
1082 return -1;
1085 pfn = PFN_DOWN(virt_to_phys(vcpu->arch.kseg0_commpage));
1086 /* Also set valid and dirty, so refill handler doesn't have to */
1087 prot = vm_get_page_prot(VM_READ|VM_WRITE|VM_SHARED);
1088 *ptep = pte_mkyoung(pte_mkdirty(pfn_pte(pfn, prot)));
1090 /* Invalidate this entry in the TLB, guest kernel ASID only */
1091 kvm_mips_host_tlb_inv(vcpu, badvaddr, false, true);
1092 return 0;
1096 * kvm_mips_migrate_count() - Migrate timer.
1097 * @vcpu: Virtual CPU.
1099 * Migrate CP0_Count hrtimer to the current CPU by cancelling and restarting it
1100 * if it was running prior to being cancelled.
1102 * Must be called when the VCPU is migrated to a different CPU to ensure that
1103 * timer expiry during guest execution interrupts the guest and causes the
1104 * interrupt to be delivered in a timely manner.
1106 static void kvm_mips_migrate_count(struct kvm_vcpu *vcpu)
1108 if (hrtimer_cancel(&vcpu->arch.comparecount_timer))
1109 hrtimer_restart(&vcpu->arch.comparecount_timer);
1112 /* Restore ASID once we are scheduled back after preemption */
1113 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1115 unsigned long flags;
1117 kvm_debug("%s: vcpu %p, cpu: %d\n", __func__, vcpu, cpu);
1119 local_irq_save(flags);
1121 vcpu->cpu = cpu;
1122 if (vcpu->arch.last_sched_cpu != cpu) {
1123 kvm_debug("[%d->%d]KVM VCPU[%d] switch\n",
1124 vcpu->arch.last_sched_cpu, cpu, vcpu->vcpu_id);
1126 * Migrate the timer interrupt to the current CPU so that it
1127 * always interrupts the guest and synchronously triggers a
1128 * guest timer interrupt.
1130 kvm_mips_migrate_count(vcpu);
1133 /* restore guest state to registers */
1134 kvm_mips_callbacks->vcpu_load(vcpu, cpu);
1136 local_irq_restore(flags);
1139 /* ASID can change if another task is scheduled during preemption */
1140 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1142 unsigned long flags;
1143 int cpu;
1145 local_irq_save(flags);
1147 cpu = smp_processor_id();
1148 vcpu->arch.last_sched_cpu = cpu;
1149 vcpu->cpu = -1;
1151 /* save guest state in registers */
1152 kvm_mips_callbacks->vcpu_put(vcpu, cpu);
1154 local_irq_restore(flags);
1158 * kvm_trap_emul_gva_fault() - Safely attempt to handle a GVA access fault.
1159 * @vcpu: Virtual CPU.
1160 * @gva: Guest virtual address to be accessed.
1161 * @write: True if write attempted (must be dirtied and made writable).
1163 * Safely attempt to handle a GVA fault, mapping GVA pages if necessary, and
1164 * dirtying the page if @write so that guest instructions can be modified.
1166 * Returns: KVM_MIPS_MAPPED on success.
1167 * KVM_MIPS_GVA if bad guest virtual address.
1168 * KVM_MIPS_GPA if bad guest physical address.
1169 * KVM_MIPS_TLB if guest TLB not present.
1170 * KVM_MIPS_TLBINV if guest TLB present but not valid.
1171 * KVM_MIPS_TLBMOD if guest TLB read only.
1173 enum kvm_mips_fault_result kvm_trap_emul_gva_fault(struct kvm_vcpu *vcpu,
1174 unsigned long gva,
1175 bool write)
1177 struct mips_coproc *cop0 = vcpu->arch.cop0;
1178 struct kvm_mips_tlb *tlb;
1179 int index;
1181 if (KVM_GUEST_KSEGX(gva) == KVM_GUEST_KSEG0) {
1182 if (kvm_mips_handle_kseg0_tlb_fault(gva, vcpu, write) < 0)
1183 return KVM_MIPS_GPA;
1184 } else if ((KVM_GUEST_KSEGX(gva) < KVM_GUEST_KSEG0) ||
1185 KVM_GUEST_KSEGX(gva) == KVM_GUEST_KSEG23) {
1186 /* Address should be in the guest TLB */
1187 index = kvm_mips_guest_tlb_lookup(vcpu, (gva & VPN2_MASK) |
1188 (kvm_read_c0_guest_entryhi(cop0) & KVM_ENTRYHI_ASID));
1189 if (index < 0)
1190 return KVM_MIPS_TLB;
1191 tlb = &vcpu->arch.guest_tlb[index];
1193 /* Entry should be valid, and dirty for writes */
1194 if (!TLB_IS_VALID(*tlb, gva))
1195 return KVM_MIPS_TLBINV;
1196 if (write && !TLB_IS_DIRTY(*tlb, gva))
1197 return KVM_MIPS_TLBMOD;
1199 if (kvm_mips_handle_mapped_seg_tlb_fault(vcpu, tlb, gva, write))
1200 return KVM_MIPS_GPA;
1201 } else {
1202 return KVM_MIPS_GVA;
1205 return KVM_MIPS_MAPPED;
1208 int kvm_get_inst(u32 *opc, struct kvm_vcpu *vcpu, u32 *out)
1210 int err;
1212 if (WARN(IS_ENABLED(CONFIG_KVM_MIPS_VZ),
1213 "Expect BadInstr/BadInstrP registers to be used with VZ\n"))
1214 return -EINVAL;
1216 retry:
1217 kvm_trap_emul_gva_lockless_begin(vcpu);
1218 err = get_user(*out, opc);
1219 kvm_trap_emul_gva_lockless_end(vcpu);
1221 if (unlikely(err)) {
1223 * Try to handle the fault, maybe we just raced with a GVA
1224 * invalidation.
1226 err = kvm_trap_emul_gva_fault(vcpu, (unsigned long)opc,
1227 false);
1228 if (unlikely(err)) {
1229 kvm_err("%s: illegal address: %p\n",
1230 __func__, opc);
1231 return -EFAULT;
1234 /* Hopefully it'll work now */
1235 goto retry;
1237 return 0;