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
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
25 #define KVM_MMU_CACHE_MIN_PAGES 2
28 void kvm_mmu_free_memory_caches(struct kvm_vcpu
*vcpu
)
30 kvm_mmu_free_memory_cache(&vcpu
->arch
.mmu_page_cache
);
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
;
48 #ifdef __PAGETABLE_PMD_FOLDED
49 entry
= (unsigned long)invalid_pte_table
;
51 entry
= (unsigned long)invalid_pmd_table
;
54 p
= (unsigned long *)page
;
55 end
= p
+ PTRS_PER_PGD
;
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)
83 ret
= (pgd_t
*)__get_free_pages(GFP_KERNEL
, PGD_TABLE_ORDER
);
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
,
111 pgd
+= pgd_index(addr
);
112 if (pgd_none(*pgd
)) {
113 /* Not used on MIPS yet */
117 p4d
= p4d_offset(pgd
, addr
);
118 pud
= pud_offset(p4d
, addr
);
119 if (pud_none(*pud
)) {
124 new_pmd
= kvm_mmu_memory_cache_alloc(cache
);
126 pud_populate(NULL
, pud
, new_pmd
);
128 pmd
= pmd_offset(pud
, addr
);
129 if (pmd_none(*pmd
)) {
134 new_pte
= kvm_mmu_memory_cache_alloc(cache
);
136 pmd_populate_kernel(NULL
, pmd
, new_pte
);
138 return pte_offset_kernel(pmd
, addr
);
141 /* Caller must hold kvm->mm_lock */
142 static pte_t
*kvm_mips_pte_for_gpa(struct kvm
*kvm
,
143 struct kvm_mmu_memory_cache
*cache
,
146 return kvm_mips_walk_pgd(kvm
->arch
.gpa_mm
.pgd
, cache
, addr
);
150 * kvm_mips_flush_gpa_{pte,pmd,pud,pgd,pt}.
151 * Flush a range of guest physical address space from the VM's GPA page tables.
154 static bool kvm_mips_flush_gpa_pte(pte_t
*pte
, unsigned long start_gpa
,
155 unsigned long end_gpa
)
157 int i_min
= pte_index(start_gpa
);
158 int i_max
= pte_index(end_gpa
);
159 bool safe_to_remove
= (i_min
== 0 && i_max
== PTRS_PER_PTE
- 1);
162 for (i
= i_min
; i
<= i_max
; ++i
) {
163 if (!pte_present(pte
[i
]))
166 set_pte(pte
+ i
, __pte(0));
168 return safe_to_remove
;
171 static bool kvm_mips_flush_gpa_pmd(pmd_t
*pmd
, unsigned long start_gpa
,
172 unsigned long end_gpa
)
175 unsigned long end
= ~0ul;
176 int i_min
= pmd_index(start_gpa
);
177 int i_max
= pmd_index(end_gpa
);
178 bool safe_to_remove
= (i_min
== 0 && i_max
== PTRS_PER_PMD
- 1);
181 for (i
= i_min
; i
<= i_max
; ++i
, start_gpa
= 0) {
182 if (!pmd_present(pmd
[i
]))
185 pte
= pte_offset_kernel(pmd
+ i
, 0);
189 if (kvm_mips_flush_gpa_pte(pte
, start_gpa
, end
)) {
191 pte_free_kernel(NULL
, pte
);
193 safe_to_remove
= false;
196 return safe_to_remove
;
199 static bool kvm_mips_flush_gpa_pud(pud_t
*pud
, unsigned long start_gpa
,
200 unsigned long end_gpa
)
203 unsigned long end
= ~0ul;
204 int i_min
= pud_index(start_gpa
);
205 int i_max
= pud_index(end_gpa
);
206 bool safe_to_remove
= (i_min
== 0 && i_max
== PTRS_PER_PUD
- 1);
209 for (i
= i_min
; i
<= i_max
; ++i
, start_gpa
= 0) {
210 if (!pud_present(pud
[i
]))
213 pmd
= pmd_offset(pud
+ i
, 0);
217 if (kvm_mips_flush_gpa_pmd(pmd
, start_gpa
, end
)) {
221 safe_to_remove
= false;
224 return safe_to_remove
;
227 static bool kvm_mips_flush_gpa_pgd(pgd_t
*pgd
, unsigned long start_gpa
,
228 unsigned long end_gpa
)
232 unsigned long end
= ~0ul;
233 int i_min
= pgd_index(start_gpa
);
234 int i_max
= pgd_index(end_gpa
);
235 bool safe_to_remove
= (i_min
== 0 && i_max
== PTRS_PER_PGD
- 1);
238 for (i
= i_min
; i
<= i_max
; ++i
, start_gpa
= 0) {
239 if (!pgd_present(pgd
[i
]))
242 p4d
= p4d_offset(pgd
, 0);
243 pud
= pud_offset(p4d
+ i
, 0);
247 if (kvm_mips_flush_gpa_pud(pud
, start_gpa
, end
)) {
251 safe_to_remove
= false;
254 return safe_to_remove
;
258 * kvm_mips_flush_gpa_pt() - Flush a range of guest physical addresses.
260 * @start_gfn: Guest frame number of first page in GPA range to flush.
261 * @end_gfn: Guest frame number of last page in GPA range to flush.
263 * Flushes a range of GPA mappings from the GPA page tables.
265 * The caller must hold the @kvm->mmu_lock spinlock.
267 * Returns: Whether its safe to remove the top level page directory because
268 * all lower levels have been removed.
270 bool kvm_mips_flush_gpa_pt(struct kvm
*kvm
, gfn_t start_gfn
, gfn_t end_gfn
)
272 return kvm_mips_flush_gpa_pgd(kvm
->arch
.gpa_mm
.pgd
,
273 start_gfn
<< PAGE_SHIFT
,
274 end_gfn
<< PAGE_SHIFT
);
277 #define BUILD_PTE_RANGE_OP(name, op) \
278 static int kvm_mips_##name##_pte(pte_t *pte, unsigned long start, \
282 int i_min = pte_index(start); \
283 int i_max = pte_index(end); \
287 for (i = i_min; i <= i_max; ++i) { \
288 if (!pte_present(pte[i])) \
293 if (pte_val(new) == pte_val(old)) \
295 set_pte(pte + i, new); \
301 /* returns true if anything was done */ \
302 static int kvm_mips_##name##_pmd(pmd_t *pmd, unsigned long start, \
307 unsigned long cur_end = ~0ul; \
308 int i_min = pmd_index(start); \
309 int i_max = pmd_index(end); \
312 for (i = i_min; i <= i_max; ++i, start = 0) { \
313 if (!pmd_present(pmd[i])) \
316 pte = pte_offset_kernel(pmd + i, 0); \
320 ret |= kvm_mips_##name##_pte(pte, start, cur_end); \
325 static int kvm_mips_##name##_pud(pud_t *pud, unsigned long start, \
330 unsigned long cur_end = ~0ul; \
331 int i_min = pud_index(start); \
332 int i_max = pud_index(end); \
335 for (i = i_min; i <= i_max; ++i, start = 0) { \
336 if (!pud_present(pud[i])) \
339 pmd = pmd_offset(pud + i, 0); \
343 ret |= kvm_mips_##name##_pmd(pmd, start, cur_end); \
348 static int kvm_mips_##name##_pgd(pgd_t *pgd, unsigned long start, \
354 unsigned long cur_end = ~0ul; \
355 int i_min = pgd_index(start); \
356 int i_max = pgd_index(end); \
359 for (i = i_min; i <= i_max; ++i, start = 0) { \
360 if (!pgd_present(pgd[i])) \
363 p4d = p4d_offset(pgd, 0); \
364 pud = pud_offset(p4d + i, 0); \
368 ret |= kvm_mips_##name##_pud(pud, start, cur_end); \
374 * kvm_mips_mkclean_gpa_pt.
375 * Mark a range of guest physical address space clean (writes fault) in the VM's
376 * GPA page table to allow dirty page tracking.
379 BUILD_PTE_RANGE_OP(mkclean
, pte_mkclean
)
382 * kvm_mips_mkclean_gpa_pt() - Make a range of guest physical addresses clean.
384 * @start_gfn: Guest frame number of first page in GPA range to flush.
385 * @end_gfn: Guest frame number of last page in GPA range to flush.
387 * Make a range of GPA mappings clean so that guest writes will fault and
388 * trigger dirty page logging.
390 * The caller must hold the @kvm->mmu_lock spinlock.
392 * Returns: Whether any GPA mappings were modified, which would require
393 * derived mappings (GVA page tables & TLB enties) to be
396 int kvm_mips_mkclean_gpa_pt(struct kvm
*kvm
, gfn_t start_gfn
, gfn_t end_gfn
)
398 return kvm_mips_mkclean_pgd(kvm
->arch
.gpa_mm
.pgd
,
399 start_gfn
<< PAGE_SHIFT
,
400 end_gfn
<< PAGE_SHIFT
);
404 * kvm_arch_mmu_enable_log_dirty_pt_masked() - write protect dirty pages
405 * @kvm: The KVM pointer
406 * @slot: The memory slot associated with mask
407 * @gfn_offset: The gfn offset in memory slot
408 * @mask: The mask of dirty pages at offset 'gfn_offset' in this memory
409 * slot to be write protected
411 * Walks bits set in mask write protects the associated pte's. Caller must
412 * acquire @kvm->mmu_lock.
414 void kvm_arch_mmu_enable_log_dirty_pt_masked(struct kvm
*kvm
,
415 struct kvm_memory_slot
*slot
,
416 gfn_t gfn_offset
, unsigned long mask
)
418 gfn_t base_gfn
= slot
->base_gfn
+ gfn_offset
;
419 gfn_t start
= base_gfn
+ __ffs(mask
);
420 gfn_t end
= base_gfn
+ __fls(mask
);
422 kvm_mips_mkclean_gpa_pt(kvm
, start
, end
);
426 * kvm_mips_mkold_gpa_pt.
427 * Mark a range of guest physical address space old (all accesses fault) in the
428 * VM's GPA page table to allow detection of commonly used pages.
431 BUILD_PTE_RANGE_OP(mkold
, pte_mkold
)
433 static int kvm_mips_mkold_gpa_pt(struct kvm
*kvm
, gfn_t start_gfn
,
436 return kvm_mips_mkold_pgd(kvm
->arch
.gpa_mm
.pgd
,
437 start_gfn
<< PAGE_SHIFT
,
438 end_gfn
<< PAGE_SHIFT
);
441 bool kvm_unmap_gfn_range(struct kvm
*kvm
, struct kvm_gfn_range
*range
)
443 kvm_mips_flush_gpa_pt(kvm
, range
->start
, range
->end
);
447 bool kvm_age_gfn(struct kvm
*kvm
, struct kvm_gfn_range
*range
)
449 return kvm_mips_mkold_gpa_pt(kvm
, range
->start
, range
->end
);
452 bool kvm_test_age_gfn(struct kvm
*kvm
, struct kvm_gfn_range
*range
)
454 gpa_t gpa
= range
->start
<< PAGE_SHIFT
;
455 pte_t
*gpa_pte
= kvm_mips_pte_for_gpa(kvm
, NULL
, gpa
);
459 return pte_young(*gpa_pte
);
463 * _kvm_mips_map_page_fast() - Fast path GPA fault handler.
464 * @vcpu: VCPU pointer.
465 * @gpa: Guest physical address of fault.
466 * @write_fault: Whether the fault was due to a write.
467 * @out_entry: New PTE for @gpa (written on success unless NULL).
468 * @out_buddy: New PTE for @gpa's buddy (written on success unless
471 * Perform fast path GPA fault handling, doing all that can be done without
472 * calling into KVM. This handles marking old pages young (for idle page
473 * tracking), and dirtying of clean pages (for dirty page logging).
475 * Returns: 0 on success, in which case we can update derived mappings and
476 * resume guest execution.
477 * -EFAULT on failure due to absent GPA mapping or write to
478 * read-only page, in which case KVM must be consulted.
480 static int _kvm_mips_map_page_fast(struct kvm_vcpu
*vcpu
, unsigned long gpa
,
482 pte_t
*out_entry
, pte_t
*out_buddy
)
484 struct kvm
*kvm
= vcpu
->kvm
;
485 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
489 spin_lock(&kvm
->mmu_lock
);
491 /* Fast path - just check GPA page table for an existing entry */
492 ptep
= kvm_mips_pte_for_gpa(kvm
, NULL
, gpa
);
493 if (!ptep
|| !pte_present(*ptep
)) {
498 /* Track access to pages marked old */
499 if (!pte_young(*ptep
))
500 set_pte(ptep
, pte_mkyoung(*ptep
));
502 if (write_fault
&& !pte_dirty(*ptep
)) {
503 if (!pte_write(*ptep
)) {
508 /* Track dirtying of writeable pages */
509 set_pte(ptep
, pte_mkdirty(*ptep
));
510 mark_page_dirty(kvm
, gfn
);
516 *out_buddy
= *ptep_buddy(ptep
);
519 spin_unlock(&kvm
->mmu_lock
);
524 * kvm_mips_map_page() - Map a guest physical page.
525 * @vcpu: VCPU pointer.
526 * @gpa: Guest physical address of fault.
527 * @write_fault: Whether the fault was due to a write.
528 * @out_entry: New PTE for @gpa (written on success unless NULL).
529 * @out_buddy: New PTE for @gpa's buddy (written on success unless
532 * Handle GPA faults by creating a new GPA mapping (or updating an existing
535 * This takes care of marking pages young or dirty (idle/dirty page tracking),
536 * asking KVM for the corresponding PFN, and creating a mapping in the GPA page
537 * tables. Derived mappings (GVA page tables and TLBs) must be handled by the
540 * Returns: 0 on success, in which case the caller may use the @out_entry
541 * and @out_buddy PTEs to update derived mappings and resume guest
543 * -EFAULT if there is no memory region at @gpa or a write was
544 * attempted to a read-only memory region. This is usually handled
547 static int kvm_mips_map_page(struct kvm_vcpu
*vcpu
, unsigned long gpa
,
549 pte_t
*out_entry
, pte_t
*out_buddy
)
551 struct kvm
*kvm
= vcpu
->kvm
;
552 struct kvm_mmu_memory_cache
*memcache
= &vcpu
->arch
.mmu_page_cache
;
553 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
558 unsigned long prot_bits
;
559 unsigned long mmu_seq
;
562 /* Try the fast path to handle old / clean pages */
563 srcu_idx
= srcu_read_lock(&kvm
->srcu
);
564 err
= _kvm_mips_map_page_fast(vcpu
, gpa
, write_fault
, out_entry
,
569 /* We need a minimum of cached pages ready for page table creation */
570 err
= kvm_mmu_topup_memory_cache(memcache
, KVM_MMU_CACHE_MIN_PAGES
);
576 * Used to check for invalidations in progress, of the pfn that is
577 * returned by pfn_to_pfn_prot below.
579 mmu_seq
= kvm
->mmu_invalidate_seq
;
581 * Ensure the read of mmu_invalidate_seq isn't reordered with PTE reads
582 * in kvm_faultin_pfn() (which calls get_user_pages()), so that we don't
583 * risk the page we get a reference to getting unmapped before we have a
584 * chance to grab the mmu_lock without mmu_invalidate_retry() noticing.
586 * This smp_rmb() pairs with the effective smp_wmb() of the combination
587 * of the pte_unmap_unlock() after the PTE is zapped, and the
588 * spin_lock() in kvm_mmu_notifier_invalidate_<page|range_end>() before
589 * mmu_invalidate_seq is incremented.
593 /* Slow path - ask KVM core whether we can access this GPA */
594 pfn
= kvm_faultin_pfn(vcpu
, gfn
, write_fault
, &writeable
, &page
);
595 if (is_error_noslot_pfn(pfn
)) {
600 spin_lock(&kvm
->mmu_lock
);
601 /* Check if an invalidation has taken place since we got pfn */
602 if (mmu_invalidate_retry(kvm
, mmu_seq
)) {
604 * This can happen when mappings are changed asynchronously, but
605 * also synchronously if a COW is triggered by
608 spin_unlock(&kvm
->mmu_lock
);
609 kvm_release_page_unused(page
);
613 /* Ensure page tables are allocated */
614 ptep
= kvm_mips_pte_for_gpa(kvm
, memcache
, gpa
);
617 prot_bits
= _PAGE_PRESENT
| __READABLE
| _page_cachable_default
;
619 prot_bits
|= _PAGE_WRITE
;
621 prot_bits
|= __WRITEABLE
;
622 mark_page_dirty(kvm
, gfn
);
625 entry
= pfn_pte(pfn
, __pgprot(prot_bits
));
628 set_pte(ptep
, entry
);
634 *out_buddy
= *ptep_buddy(ptep
);
636 kvm_release_faultin_page(kvm
, page
, false, writeable
);
637 spin_unlock(&kvm
->mmu_lock
);
639 srcu_read_unlock(&kvm
->srcu
, srcu_idx
);
643 int kvm_mips_handle_vz_root_tlb_fault(unsigned long badvaddr
,
644 struct kvm_vcpu
*vcpu
,
649 ret
= kvm_mips_map_page(vcpu
, badvaddr
, write_fault
, NULL
, NULL
);
653 /* Invalidate this entry in the TLB */
654 return kvm_vz_host_tlb_inv(vcpu
, badvaddr
);
658 * kvm_mips_migrate_count() - Migrate timer.
659 * @vcpu: Virtual CPU.
661 * Migrate CP0_Count hrtimer to the current CPU by cancelling and restarting it
662 * if it was running prior to being cancelled.
664 * Must be called when the VCPU is migrated to a different CPU to ensure that
665 * timer expiry during guest execution interrupts the guest and causes the
666 * interrupt to be delivered in a timely manner.
668 static void kvm_mips_migrate_count(struct kvm_vcpu
*vcpu
)
670 if (hrtimer_cancel(&vcpu
->arch
.comparecount_timer
))
671 hrtimer_restart(&vcpu
->arch
.comparecount_timer
);
674 /* Restore ASID once we are scheduled back after preemption */
675 void kvm_arch_vcpu_load(struct kvm_vcpu
*vcpu
, int cpu
)
679 kvm_debug("%s: vcpu %p, cpu: %d\n", __func__
, vcpu
, cpu
);
681 local_irq_save(flags
);
684 if (vcpu
->arch
.last_sched_cpu
!= cpu
) {
685 kvm_debug("[%d->%d]KVM VCPU[%d] switch\n",
686 vcpu
->arch
.last_sched_cpu
, cpu
, vcpu
->vcpu_id
);
688 * Migrate the timer interrupt to the current CPU so that it
689 * always interrupts the guest and synchronously triggers a
690 * guest timer interrupt.
692 kvm_mips_migrate_count(vcpu
);
695 /* restore guest state to registers */
696 kvm_mips_callbacks
->vcpu_load(vcpu
, cpu
);
698 local_irq_restore(flags
);
701 /* ASID can change if another task is scheduled during preemption */
702 void kvm_arch_vcpu_put(struct kvm_vcpu
*vcpu
)
707 local_irq_save(flags
);
709 cpu
= smp_processor_id();
710 vcpu
->arch
.last_sched_cpu
= cpu
;
713 /* save guest state in registers */
714 kvm_mips_callbacks
->vcpu_put(vcpu
, cpu
);
716 local_irq_restore(flags
);