xref: /openbmc/linux/arch/mips/kvm/mmu.c (revision 0e407915)
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.
5  *
6  * KVM/MIPS MMU handling in the KVM module.
7  *
8  * Copyright (C) 2012  MIPS Technologies, Inc.  All rights reserved.
9  * Authors: Sanjay Lal <sanjayl@kymasys.com>
10  */
11 
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>
17 
18 /*
19  * KVM_MMU_CACHE_MIN_PAGES is the number of GPA page table translation levels
20  * for which pages need to be cached.
21  */
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
27 
28 void kvm_mmu_free_memory_caches(struct kvm_vcpu *vcpu)
29 {
30 	kvm_mmu_free_memory_cache(&vcpu->arch.mmu_page_cache);
31 }
32 
33 /**
34  * kvm_pgd_init() - Initialise KVM GPA page directory.
35  * @page:	Pointer to page directory (PGD) for KVM GPA.
36  *
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).
42  */
43 static void kvm_pgd_init(void *page)
44 {
45 	unsigned long *p, *end;
46 	unsigned long entry;
47 
48 #ifdef __PAGETABLE_PMD_FOLDED
49 	entry = (unsigned long)invalid_pte_table;
50 #else
51 	entry = (unsigned long)invalid_pmd_table;
52 #endif
53 
54 	p = (unsigned long *)page;
55 	end = p + PTRS_PER_PGD;
56 
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);
68 }
69 
70 /**
71  * kvm_pgd_alloc() - Allocate and initialise a KVM GPA page directory.
72  *
73  * Allocate a blank KVM GPA page directory (PGD) for representing guest physical
74  * to host physical page mappings.
75  *
76  * Returns:	Pointer to new KVM GPA page directory.
77  *		NULL on allocation failure.
78  */
79 pgd_t *kvm_pgd_alloc(void)
80 {
81 	pgd_t *ret;
82 
83 	ret = (pgd_t *)__get_free_pages(GFP_KERNEL, PGD_TABLE_ORDER);
84 	if (ret)
85 		kvm_pgd_init(ret);
86 
87 	return ret;
88 }
89 
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.
95  *
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.
99  *
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.
103  */
104 static pte_t *kvm_mips_walk_pgd(pgd_t *pgd, struct kvm_mmu_memory_cache *cache,
105 				unsigned long addr)
106 {
107 	p4d_t *p4d;
108 	pud_t *pud;
109 	pmd_t *pmd;
110 
111 	pgd += pgd_index(addr);
112 	if (pgd_none(*pgd)) {
113 		/* Not used on MIPS yet */
114 		BUG();
115 		return NULL;
116 	}
117 	p4d = p4d_offset(pgd, addr);
118 	pud = pud_offset(p4d, addr);
119 	if (pud_none(*pud)) {
120 		pmd_t *new_pmd;
121 
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);
128 	}
129 	pmd = pmd_offset(pud, addr);
130 	if (pmd_none(*pmd)) {
131 		pte_t *new_pte;
132 
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);
138 	}
139 	return pte_offset_kernel(pmd, addr);
140 }
141 
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)
146 {
147 	return kvm_mips_walk_pgd(kvm->arch.gpa_mm.pgd, cache, addr);
148 }
149 
150 /*
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.
153  */
154 
155 static bool kvm_mips_flush_gpa_pte(pte_t *pte, unsigned long start_gpa,
156 				   unsigned long end_gpa)
157 {
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;
162 
163 	for (i = i_min; i <= i_max; ++i) {
164 		if (!pte_present(pte[i]))
165 			continue;
166 
167 		set_pte(pte + i, __pte(0));
168 	}
169 	return safe_to_remove;
170 }
171 
172 static bool kvm_mips_flush_gpa_pmd(pmd_t *pmd, unsigned long start_gpa,
173 				   unsigned long end_gpa)
174 {
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;
181 
182 	for (i = i_min; i <= i_max; ++i, start_gpa = 0) {
183 		if (!pmd_present(pmd[i]))
184 			continue;
185 
186 		pte = pte_offset_kernel(pmd + i, 0);
187 		if (i == i_max)
188 			end = end_gpa;
189 
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;
195 		}
196 	}
197 	return safe_to_remove;
198 }
199 
200 static bool kvm_mips_flush_gpa_pud(pud_t *pud, unsigned long start_gpa,
201 				   unsigned long end_gpa)
202 {
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;
209 
210 	for (i = i_min; i <= i_max; ++i, start_gpa = 0) {
211 		if (!pud_present(pud[i]))
212 			continue;
213 
214 		pmd = pmd_offset(pud + i, 0);
215 		if (i == i_max)
216 			end = end_gpa;
217 
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;
223 		}
224 	}
225 	return safe_to_remove;
226 }
227 
228 static bool kvm_mips_flush_gpa_pgd(pgd_t *pgd, unsigned long start_gpa,
229 				   unsigned long end_gpa)
230 {
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;
238 
239 	for (i = i_min; i <= i_max; ++i, start_gpa = 0) {
240 		if (!pgd_present(pgd[i]))
241 			continue;
242 
243 		p4d = p4d_offset(pgd, 0);
244 		pud = pud_offset(p4d + i, 0);
245 		if (i == i_max)
246 			end = end_gpa;
247 
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;
253 		}
254 	}
255 	return safe_to_remove;
256 }
257 
258 /**
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.
263  *
264  * Flushes a range of GPA mappings from the GPA page tables.
265  *
266  * The caller must hold the @kvm->mmu_lock spinlock.
267  *
268  * Returns:	Whether its safe to remove the top level page directory because
269  *		all lower levels have been removed.
270  */
271 bool kvm_mips_flush_gpa_pt(struct kvm *kvm, gfn_t start_gfn, gfn_t end_gfn)
272 {
273 	return kvm_mips_flush_gpa_pgd(kvm->arch.gpa_mm.pgd,
274 				      start_gfn << PAGE_SHIFT,
275 				      end_gfn << PAGE_SHIFT);
276 }
277 
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)			\
281 {									\
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;							\
287 									\
288 	for (i = i_min; i <= i_max; ++i) {				\
289 		if (!pte_present(pte[i]))				\
290 			continue;					\
291 									\
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;						\
298 	}								\
299 	return ret;							\
300 }									\
301 									\
302 /* returns true if anything was done */					\
303 static int kvm_mips_##name##_pmd(pmd_t *pmd, unsigned long start,	\
304 				 unsigned long end)			\
305 {									\
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;								\
312 									\
313 	for (i = i_min; i <= i_max; ++i, start = 0) {			\
314 		if (!pmd_present(pmd[i]))				\
315 			continue;					\
316 									\
317 		pte = pte_offset_kernel(pmd + i, 0);				\
318 		if (i == i_max)						\
319 			cur_end = end;					\
320 									\
321 		ret |= kvm_mips_##name##_pte(pte, start, cur_end);	\
322 	}								\
323 	return ret;							\
324 }									\
325 									\
326 static int kvm_mips_##name##_pud(pud_t *pud, unsigned long start,	\
327 				 unsigned long end)			\
328 {									\
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;								\
335 									\
336 	for (i = i_min; i <= i_max; ++i, start = 0) {			\
337 		if (!pud_present(pud[i]))				\
338 			continue;					\
339 									\
340 		pmd = pmd_offset(pud + i, 0);				\
341 		if (i == i_max)						\
342 			cur_end = end;					\
343 									\
344 		ret |= kvm_mips_##name##_pmd(pmd, start, cur_end);	\
345 	}								\
346 	return ret;							\
347 }									\
348 									\
349 static int kvm_mips_##name##_pgd(pgd_t *pgd, unsigned long start,	\
350 				 unsigned long end)			\
351 {									\
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;								\
359 									\
360 	for (i = i_min; i <= i_max; ++i, start = 0) {			\
361 		if (!pgd_present(pgd[i]))				\
362 			continue;					\
363 									\
364 		p4d = p4d_offset(pgd, 0);				\
365 		pud = pud_offset(p4d + i, 0);				\
366 		if (i == i_max)						\
367 			cur_end = end;					\
368 									\
369 		ret |= kvm_mips_##name##_pud(pud, start, cur_end);	\
370 	}								\
371 	return ret;							\
372 }
373 
374 /*
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.
378  */
379 
380 BUILD_PTE_RANGE_OP(mkclean, pte_mkclean)
381 
382 /**
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.
387  *
388  * Make a range of GPA mappings clean so that guest writes will fault and
389  * trigger dirty page logging.
390  *
391  * The caller must hold the @kvm->mmu_lock spinlock.
392  *
393  * Returns:	Whether any GPA mappings were modified, which would require
394  *		derived mappings (GVA page tables & TLB enties) to be
395  *		invalidated.
396  */
397 int kvm_mips_mkclean_gpa_pt(struct kvm *kvm, gfn_t start_gfn, gfn_t end_gfn)
398 {
399 	return kvm_mips_mkclean_pgd(kvm->arch.gpa_mm.pgd,
400 				    start_gfn << PAGE_SHIFT,
401 				    end_gfn << PAGE_SHIFT);
402 }
403 
404 /**
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
411  *
412  * Walks bits set in mask write protects the associated pte's. Caller must
413  * acquire @kvm->mmu_lock.
414  */
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)
418 {
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);
422 
423 	kvm_mips_mkclean_gpa_pt(kvm, start, end);
424 }
425 
426 /*
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.
430  */
431 
432 BUILD_PTE_RANGE_OP(mkold, pte_mkold)
433 
434 static int kvm_mips_mkold_gpa_pt(struct kvm *kvm, gfn_t start_gfn,
435 				 gfn_t end_gfn)
436 {
437 	return kvm_mips_mkold_pgd(kvm->arch.gpa_mm.pgd,
438 				  start_gfn << PAGE_SHIFT,
439 				  end_gfn << PAGE_SHIFT);
440 }
441 
442 bool kvm_unmap_gfn_range(struct kvm *kvm, struct kvm_gfn_range *range)
443 {
444 	kvm_mips_flush_gpa_pt(kvm, range->start, range->end);
445 	return true;
446 }
447 
448 bool kvm_set_spte_gfn(struct kvm *kvm, struct kvm_gfn_range *range)
449 {
450 	gpa_t gpa = range->start << PAGE_SHIFT;
451 	pte_t hva_pte = range->pte;
452 	pte_t *gpa_pte = kvm_mips_pte_for_gpa(kvm, NULL, gpa);
453 	pte_t old_pte;
454 
455 	if (!gpa_pte)
456 		return false;
457 
458 	/* Mapping may need adjusting depending on memslot flags */
459 	old_pte = *gpa_pte;
460 	if (range->slot->flags & KVM_MEM_LOG_DIRTY_PAGES && !pte_dirty(old_pte))
461 		hva_pte = pte_mkclean(hva_pte);
462 	else if (range->slot->flags & KVM_MEM_READONLY)
463 		hva_pte = pte_wrprotect(hva_pte);
464 
465 	set_pte(gpa_pte, hva_pte);
466 
467 	/* Replacing an absent or old page doesn't need flushes */
468 	if (!pte_present(old_pte) || !pte_young(old_pte))
469 		return false;
470 
471 	/* Pages swapped, aged, moved, or cleaned require flushes */
472 	return !pte_present(hva_pte) ||
473 	       !pte_young(hva_pte) ||
474 	       pte_pfn(old_pte) != pte_pfn(hva_pte) ||
475 	       (pte_dirty(old_pte) && !pte_dirty(hva_pte));
476 }
477 
478 bool kvm_age_gfn(struct kvm *kvm, struct kvm_gfn_range *range)
479 {
480 	return kvm_mips_mkold_gpa_pt(kvm, range->start, range->end);
481 }
482 
483 bool kvm_test_age_gfn(struct kvm *kvm, struct kvm_gfn_range *range)
484 {
485 	gpa_t gpa = range->start << PAGE_SHIFT;
486 	pte_t *gpa_pte = kvm_mips_pte_for_gpa(kvm, NULL, gpa);
487 
488 	if (!gpa_pte)
489 		return false;
490 	return pte_young(*gpa_pte);
491 }
492 
493 /**
494  * _kvm_mips_map_page_fast() - Fast path GPA fault handler.
495  * @vcpu:		VCPU pointer.
496  * @gpa:		Guest physical address of fault.
497  * @write_fault:	Whether the fault was due to a write.
498  * @out_entry:		New PTE for @gpa (written on success unless NULL).
499  * @out_buddy:		New PTE for @gpa's buddy (written on success unless
500  *			NULL).
501  *
502  * Perform fast path GPA fault handling, doing all that can be done without
503  * calling into KVM. This handles marking old pages young (for idle page
504  * tracking), and dirtying of clean pages (for dirty page logging).
505  *
506  * Returns:	0 on success, in which case we can update derived mappings and
507  *		resume guest execution.
508  *		-EFAULT on failure due to absent GPA mapping or write to
509  *		read-only page, in which case KVM must be consulted.
510  */
511 static int _kvm_mips_map_page_fast(struct kvm_vcpu *vcpu, unsigned long gpa,
512 				   bool write_fault,
513 				   pte_t *out_entry, pte_t *out_buddy)
514 {
515 	struct kvm *kvm = vcpu->kvm;
516 	gfn_t gfn = gpa >> PAGE_SHIFT;
517 	pte_t *ptep;
518 	kvm_pfn_t pfn = 0;	/* silence bogus GCC warning */
519 	bool pfn_valid = false;
520 	int ret = 0;
521 
522 	spin_lock(&kvm->mmu_lock);
523 
524 	/* Fast path - just check GPA page table for an existing entry */
525 	ptep = kvm_mips_pte_for_gpa(kvm, NULL, gpa);
526 	if (!ptep || !pte_present(*ptep)) {
527 		ret = -EFAULT;
528 		goto out;
529 	}
530 
531 	/* Track access to pages marked old */
532 	if (!pte_young(*ptep)) {
533 		set_pte(ptep, pte_mkyoung(*ptep));
534 		pfn = pte_pfn(*ptep);
535 		pfn_valid = true;
536 		/* call kvm_set_pfn_accessed() after unlock */
537 	}
538 	if (write_fault && !pte_dirty(*ptep)) {
539 		if (!pte_write(*ptep)) {
540 			ret = -EFAULT;
541 			goto out;
542 		}
543 
544 		/* Track dirtying of writeable pages */
545 		set_pte(ptep, pte_mkdirty(*ptep));
546 		pfn = pte_pfn(*ptep);
547 		mark_page_dirty(kvm, gfn);
548 		kvm_set_pfn_dirty(pfn);
549 	}
550 
551 	if (out_entry)
552 		*out_entry = *ptep;
553 	if (out_buddy)
554 		*out_buddy = *ptep_buddy(ptep);
555 
556 out:
557 	spin_unlock(&kvm->mmu_lock);
558 	if (pfn_valid)
559 		kvm_set_pfn_accessed(pfn);
560 	return ret;
561 }
562 
563 /**
564  * kvm_mips_map_page() - Map a guest physical page.
565  * @vcpu:		VCPU pointer.
566  * @gpa:		Guest physical address of fault.
567  * @write_fault:	Whether the fault was due to a write.
568  * @out_entry:		New PTE for @gpa (written on success unless NULL).
569  * @out_buddy:		New PTE for @gpa's buddy (written on success unless
570  *			NULL).
571  *
572  * Handle GPA faults by creating a new GPA mapping (or updating an existing
573  * one).
574  *
575  * This takes care of marking pages young or dirty (idle/dirty page tracking),
576  * asking KVM for the corresponding PFN, and creating a mapping in the GPA page
577  * tables. Derived mappings (GVA page tables and TLBs) must be handled by the
578  * caller.
579  *
580  * Returns:	0 on success, in which case the caller may use the @out_entry
581  *		and @out_buddy PTEs to update derived mappings and resume guest
582  *		execution.
583  *		-EFAULT if there is no memory region at @gpa or a write was
584  *		attempted to a read-only memory region. This is usually handled
585  *		as an MMIO access.
586  */
587 static int kvm_mips_map_page(struct kvm_vcpu *vcpu, unsigned long gpa,
588 			     bool write_fault,
589 			     pte_t *out_entry, pte_t *out_buddy)
590 {
591 	struct kvm *kvm = vcpu->kvm;
592 	struct kvm_mmu_memory_cache *memcache = &vcpu->arch.mmu_page_cache;
593 	gfn_t gfn = gpa >> PAGE_SHIFT;
594 	int srcu_idx, err;
595 	kvm_pfn_t pfn;
596 	pte_t *ptep, entry, old_pte;
597 	bool writeable;
598 	unsigned long prot_bits;
599 	unsigned long mmu_seq;
600 
601 	/* Try the fast path to handle old / clean pages */
602 	srcu_idx = srcu_read_lock(&kvm->srcu);
603 	err = _kvm_mips_map_page_fast(vcpu, gpa, write_fault, out_entry,
604 				      out_buddy);
605 	if (!err)
606 		goto out;
607 
608 	/* We need a minimum of cached pages ready for page table creation */
609 	err = kvm_mmu_topup_memory_cache(memcache, KVM_MMU_CACHE_MIN_PAGES);
610 	if (err)
611 		goto out;
612 
613 retry:
614 	/*
615 	 * Used to check for invalidations in progress, of the pfn that is
616 	 * returned by pfn_to_pfn_prot below.
617 	 */
618 	mmu_seq = kvm->mmu_invalidate_seq;
619 	/*
620 	 * Ensure the read of mmu_invalidate_seq isn't reordered with PTE reads
621 	 * in gfn_to_pfn_prot() (which calls get_user_pages()), so that we don't
622 	 * risk the page we get a reference to getting unmapped before we have a
623 	 * chance to grab the mmu_lock without mmu_invalidate_retry() noticing.
624 	 *
625 	 * This smp_rmb() pairs with the effective smp_wmb() of the combination
626 	 * of the pte_unmap_unlock() after the PTE is zapped, and the
627 	 * spin_lock() in kvm_mmu_notifier_invalidate_<page|range_end>() before
628 	 * mmu_invalidate_seq is incremented.
629 	 */
630 	smp_rmb();
631 
632 	/* Slow path - ask KVM core whether we can access this GPA */
633 	pfn = gfn_to_pfn_prot(kvm, gfn, write_fault, &writeable);
634 	if (is_error_noslot_pfn(pfn)) {
635 		err = -EFAULT;
636 		goto out;
637 	}
638 
639 	spin_lock(&kvm->mmu_lock);
640 	/* Check if an invalidation has taken place since we got pfn */
641 	if (mmu_invalidate_retry(kvm, mmu_seq)) {
642 		/*
643 		 * This can happen when mappings are changed asynchronously, but
644 		 * also synchronously if a COW is triggered by
645 		 * gfn_to_pfn_prot().
646 		 */
647 		spin_unlock(&kvm->mmu_lock);
648 		kvm_release_pfn_clean(pfn);
649 		goto retry;
650 	}
651 
652 	/* Ensure page tables are allocated */
653 	ptep = kvm_mips_pte_for_gpa(kvm, memcache, gpa);
654 
655 	/* Set up the PTE */
656 	prot_bits = _PAGE_PRESENT | __READABLE | _page_cachable_default;
657 	if (writeable) {
658 		prot_bits |= _PAGE_WRITE;
659 		if (write_fault) {
660 			prot_bits |= __WRITEABLE;
661 			mark_page_dirty(kvm, gfn);
662 			kvm_set_pfn_dirty(pfn);
663 		}
664 	}
665 	entry = pfn_pte(pfn, __pgprot(prot_bits));
666 
667 	/* Write the PTE */
668 	old_pte = *ptep;
669 	set_pte(ptep, entry);
670 
671 	err = 0;
672 	if (out_entry)
673 		*out_entry = *ptep;
674 	if (out_buddy)
675 		*out_buddy = *ptep_buddy(ptep);
676 
677 	spin_unlock(&kvm->mmu_lock);
678 	kvm_release_pfn_clean(pfn);
679 	kvm_set_pfn_accessed(pfn);
680 out:
681 	srcu_read_unlock(&kvm->srcu, srcu_idx);
682 	return err;
683 }
684 
685 int kvm_mips_handle_vz_root_tlb_fault(unsigned long badvaddr,
686 				      struct kvm_vcpu *vcpu,
687 				      bool write_fault)
688 {
689 	int ret;
690 
691 	ret = kvm_mips_map_page(vcpu, badvaddr, write_fault, NULL, NULL);
692 	if (ret)
693 		return ret;
694 
695 	/* Invalidate this entry in the TLB */
696 	return kvm_vz_host_tlb_inv(vcpu, badvaddr);
697 }
698 
699 /**
700  * kvm_mips_migrate_count() - Migrate timer.
701  * @vcpu:	Virtual CPU.
702  *
703  * Migrate CP0_Count hrtimer to the current CPU by cancelling and restarting it
704  * if it was running prior to being cancelled.
705  *
706  * Must be called when the VCPU is migrated to a different CPU to ensure that
707  * timer expiry during guest execution interrupts the guest and causes the
708  * interrupt to be delivered in a timely manner.
709  */
710 static void kvm_mips_migrate_count(struct kvm_vcpu *vcpu)
711 {
712 	if (hrtimer_cancel(&vcpu->arch.comparecount_timer))
713 		hrtimer_restart(&vcpu->arch.comparecount_timer);
714 }
715 
716 /* Restore ASID once we are scheduled back after preemption */
717 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
718 {
719 	unsigned long flags;
720 
721 	kvm_debug("%s: vcpu %p, cpu: %d\n", __func__, vcpu, cpu);
722 
723 	local_irq_save(flags);
724 
725 	vcpu->cpu = cpu;
726 	if (vcpu->arch.last_sched_cpu != cpu) {
727 		kvm_debug("[%d->%d]KVM VCPU[%d] switch\n",
728 			  vcpu->arch.last_sched_cpu, cpu, vcpu->vcpu_id);
729 		/*
730 		 * Migrate the timer interrupt to the current CPU so that it
731 		 * always interrupts the guest and synchronously triggers a
732 		 * guest timer interrupt.
733 		 */
734 		kvm_mips_migrate_count(vcpu);
735 	}
736 
737 	/* restore guest state to registers */
738 	kvm_mips_callbacks->vcpu_load(vcpu, cpu);
739 
740 	local_irq_restore(flags);
741 }
742 
743 /* ASID can change if another task is scheduled during preemption */
744 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
745 {
746 	unsigned long flags;
747 	int cpu;
748 
749 	local_irq_save(flags);
750 
751 	cpu = smp_processor_id();
752 	vcpu->arch.last_sched_cpu = cpu;
753 	vcpu->cpu = -1;
754 
755 	/* save guest state in registers */
756 	kvm_mips_callbacks->vcpu_put(vcpu, cpu);
757 
758 	local_irq_restore(flags);
759 }
760