1 /* SPDX-License-Identifier: GPL-2.0-only */
6 #include <linux/types.h>
7 #include <linux/hardirq.h>
8 #include <linux/list.h>
9 #include <linux/mutex.h>
10 #include <linux/spinlock.h>
11 #include <linux/signal.h>
12 #include <linux/sched.h>
13 #include <linux/sched/stat.h>
14 #include <linux/bug.h>
15 #include <linux/minmax.h>
17 #include <linux/mmu_notifier.h>
18 #include <linux/preempt.h>
19 #include <linux/msi.h>
20 #include <linux/slab.h>
21 #include <linux/vmalloc.h>
22 #include <linux/rcupdate.h>
23 #include <linux/ratelimit.h>
24 #include <linux/err.h>
25 #include <linux/irqflags.h>
26 #include <linux/context_tracking.h>
27 #include <linux/irqbypass.h>
28 #include <linux/rcuwait.h>
29 #include <linux/refcount.h>
30 #include <linux/nospec.h>
31 #include <linux/notifier.h>
32 #include <asm/signal.h>
34 #include <linux/kvm.h>
35 #include <linux/kvm_para.h>
37 #include <linux/kvm_types.h>
39 #include <asm/kvm_host.h>
40 #include <linux/kvm_dirty_ring.h>
42 #ifndef KVM_MAX_VCPU_ID
43 #define KVM_MAX_VCPU_ID KVM_MAX_VCPUS
47 * The bit 16 ~ bit 31 of kvm_memory_region::flags are internally used
48 * in kvm, other bits are visible for userspace which are defined in
49 * include/linux/kvm_h.
51 #define KVM_MEMSLOT_INVALID (1UL << 16)
54 * Bit 63 of the memslot generation number is an "update in-progress flag",
55 * e.g. is temporarily set for the duration of install_new_memslots().
56 * This flag effectively creates a unique generation number that is used to
57 * mark cached memslot data, e.g. MMIO accesses, as potentially being stale,
58 * i.e. may (or may not) have come from the previous memslots generation.
60 * This is necessary because the actual memslots update is not atomic with
61 * respect to the generation number update. Updating the generation number
62 * first would allow a vCPU to cache a spte from the old memslots using the
63 * new generation number, and updating the generation number after switching
64 * to the new memslots would allow cache hits using the old generation number
65 * to reference the defunct memslots.
67 * This mechanism is used to prevent getting hits in KVM's caches while a
68 * memslot update is in-progress, and to prevent cache hits *after* updating
69 * the actual generation number against accesses that were inserted into the
70 * cache *before* the memslots were updated.
72 #define KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS BIT_ULL(63)
74 /* Two fragments for cross MMIO pages. */
75 #define KVM_MAX_MMIO_FRAGMENTS 2
77 #ifndef KVM_ADDRESS_SPACE_NUM
78 #define KVM_ADDRESS_SPACE_NUM 1
82 * For the normal pfn, the highest 12 bits should be zero,
83 * so we can mask bit 62 ~ bit 52 to indicate the error pfn,
84 * mask bit 63 to indicate the noslot pfn.
86 #define KVM_PFN_ERR_MASK (0x7ffULL << 52)
87 #define KVM_PFN_ERR_NOSLOT_MASK (0xfffULL << 52)
88 #define KVM_PFN_NOSLOT (0x1ULL << 63)
90 #define KVM_PFN_ERR_FAULT (KVM_PFN_ERR_MASK)
91 #define KVM_PFN_ERR_HWPOISON (KVM_PFN_ERR_MASK + 1)
92 #define KVM_PFN_ERR_RO_FAULT (KVM_PFN_ERR_MASK + 2)
95 * error pfns indicate that the gfn is in slot but faild to
96 * translate it to pfn on host.
98 static inline bool is_error_pfn(kvm_pfn_t pfn)
100 return !!(pfn & KVM_PFN_ERR_MASK);
104 * error_noslot pfns indicate that the gfn can not be
105 * translated to pfn - it is not in slot or failed to
106 * translate it to pfn.
108 static inline bool is_error_noslot_pfn(kvm_pfn_t pfn)
110 return !!(pfn & KVM_PFN_ERR_NOSLOT_MASK);
113 /* noslot pfn indicates that the gfn is not in slot. */
114 static inline bool is_noslot_pfn(kvm_pfn_t pfn)
116 return pfn == KVM_PFN_NOSLOT;
120 * architectures with KVM_HVA_ERR_BAD other than PAGE_OFFSET (e.g. s390)
121 * provide own defines and kvm_is_error_hva
123 #ifndef KVM_HVA_ERR_BAD
125 #define KVM_HVA_ERR_BAD (PAGE_OFFSET)
126 #define KVM_HVA_ERR_RO_BAD (PAGE_OFFSET + PAGE_SIZE)
128 static inline bool kvm_is_error_hva(unsigned long addr)
130 return addr >= PAGE_OFFSET;
135 #define KVM_ERR_PTR_BAD_PAGE (ERR_PTR(-ENOENT))
137 static inline bool is_error_page(struct page *page)
142 #define KVM_REQUEST_MASK GENMASK(7,0)
143 #define KVM_REQUEST_NO_WAKEUP BIT(8)
144 #define KVM_REQUEST_WAIT BIT(9)
146 * Architecture-independent vcpu->requests bit members
147 * Bits 4-7 are reserved for more arch-independent bits.
149 #define KVM_REQ_TLB_FLUSH (0 | KVM_REQUEST_WAIT | KVM_REQUEST_NO_WAKEUP)
150 #define KVM_REQ_MMU_RELOAD (1 | KVM_REQUEST_WAIT | KVM_REQUEST_NO_WAKEUP)
151 #define KVM_REQ_UNBLOCK 2
152 #define KVM_REQ_UNHALT 3
153 #define KVM_REQUEST_ARCH_BASE 8
155 #define KVM_ARCH_REQ_FLAGS(nr, flags) ({ \
156 BUILD_BUG_ON((unsigned)(nr) >= (sizeof_field(struct kvm_vcpu, requests) * 8) - KVM_REQUEST_ARCH_BASE); \
157 (unsigned)(((nr) + KVM_REQUEST_ARCH_BASE) | (flags)); \
159 #define KVM_ARCH_REQ(nr) KVM_ARCH_REQ_FLAGS(nr, 0)
161 #define KVM_USERSPACE_IRQ_SOURCE_ID 0
162 #define KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID 1
164 extern struct mutex kvm_lock;
165 extern struct list_head vm_list;
167 struct kvm_io_range {
170 struct kvm_io_device *dev;
173 #define NR_IOBUS_DEVS 1000
178 struct kvm_io_range range[];
184 KVM_VIRTIO_CCW_NOTIFY_BUS,
189 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
190 int len, const void *val);
191 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
192 gpa_t addr, int len, const void *val, long cookie);
193 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
195 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
196 int len, struct kvm_io_device *dev);
197 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
198 struct kvm_io_device *dev);
199 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
202 #ifdef CONFIG_KVM_ASYNC_PF
203 struct kvm_async_pf {
204 struct work_struct work;
205 struct list_head link;
206 struct list_head queue;
207 struct kvm_vcpu *vcpu;
208 struct mm_struct *mm;
211 struct kvm_arch_async_pf arch;
213 bool notpresent_injected;
216 void kvm_clear_async_pf_completion_queue(struct kvm_vcpu *vcpu);
217 void kvm_check_async_pf_completion(struct kvm_vcpu *vcpu);
218 bool kvm_setup_async_pf(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
219 unsigned long hva, struct kvm_arch_async_pf *arch);
220 int kvm_async_pf_wakeup_all(struct kvm_vcpu *vcpu);
223 #ifdef KVM_ARCH_WANT_MMU_NOTIFIER
224 struct kvm_gfn_range {
225 struct kvm_memory_slot *slot;
231 bool kvm_unmap_gfn_range(struct kvm *kvm, struct kvm_gfn_range *range);
232 bool kvm_age_gfn(struct kvm *kvm, struct kvm_gfn_range *range);
233 bool kvm_test_age_gfn(struct kvm *kvm, struct kvm_gfn_range *range);
234 bool kvm_set_spte_gfn(struct kvm *kvm, struct kvm_gfn_range *range);
241 READING_SHADOW_PAGE_TABLES,
244 #define KVM_UNMAPPED_PAGE ((void *) 0x500 + POISON_POINTER_DELTA)
246 struct kvm_host_map {
248 * Only valid if the 'pfn' is managed by the host kernel (i.e. There is
249 * a 'struct page' for it. When using mem= kernel parameter some memory
250 * can be used as guest memory but they are not managed by host
252 * If 'pfn' is not managed by the host kernel, this field is
253 * initialized to KVM_UNMAPPED_PAGE.
262 * Used to check if the mapping is valid or not. Never use 'kvm_host_map'
263 * directly to check for that.
265 static inline bool kvm_vcpu_mapped(struct kvm_host_map *map)
270 static inline bool kvm_vcpu_can_poll(ktime_t cur, ktime_t stop)
272 return single_task_running() && !need_resched() && ktime_before(cur, stop);
276 * Sometimes a large or cross-page mmio needs to be broken up into separate
277 * exits for userspace servicing.
279 struct kvm_mmio_fragment {
287 #ifdef CONFIG_PREEMPT_NOTIFIERS
288 struct preempt_notifier preempt_notifier;
291 int vcpu_id; /* id given by userspace at creation */
292 int vcpu_idx; /* index in kvm->vcpus array */
296 unsigned long guest_debug;
299 struct list_head blocked_vcpu_list;
305 struct pid __rcu *pid;
308 unsigned int halt_poll_ns;
311 #ifdef CONFIG_HAS_IOMEM
313 int mmio_read_completed;
315 int mmio_cur_fragment;
316 int mmio_nr_fragments;
317 struct kvm_mmio_fragment mmio_fragments[KVM_MAX_MMIO_FRAGMENTS];
320 #ifdef CONFIG_KVM_ASYNC_PF
323 struct list_head queue;
324 struct list_head done;
329 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
331 * Cpu relax intercept or pause loop exit optimization
332 * in_spin_loop: set when a vcpu does a pause loop exit
333 * or cpu relax intercepted.
334 * dy_eligible: indicates whether vcpu is eligible for directed yield.
343 struct kvm_vcpu_arch arch;
344 struct kvm_vcpu_stat stat;
345 char stats_id[KVM_STATS_NAME_SIZE];
346 struct kvm_dirty_ring dirty_ring;
349 /* must be called with irqs disabled */
350 static __always_inline void guest_enter_irqoff(void)
353 * This is running in ioctl context so its safe to assume that it's the
354 * stime pending cputime to flush.
356 instrumentation_begin();
357 vtime_account_guest_enter();
358 instrumentation_end();
361 * KVM does not hold any references to rcu protected data when it
362 * switches CPU into a guest mode. In fact switching to a guest mode
363 * is very similar to exiting to userspace from rcu point of view. In
364 * addition CPU may stay in a guest mode for quite a long time (up to
365 * one time slice). Lets treat guest mode as quiescent state, just like
366 * we do with user-mode execution.
368 if (!context_tracking_guest_enter()) {
369 instrumentation_begin();
370 rcu_virt_note_context_switch(smp_processor_id());
371 instrumentation_end();
375 static __always_inline void guest_exit_irqoff(void)
377 context_tracking_guest_exit();
379 instrumentation_begin();
380 /* Flush the guest cputime we spent on the guest */
381 vtime_account_guest_exit();
382 instrumentation_end();
385 static inline void guest_exit(void)
389 local_irq_save(flags);
391 local_irq_restore(flags);
394 static inline int kvm_vcpu_exiting_guest_mode(struct kvm_vcpu *vcpu)
397 * The memory barrier ensures a previous write to vcpu->requests cannot
398 * be reordered with the read of vcpu->mode. It pairs with the general
399 * memory barrier following the write of vcpu->mode in VCPU RUN.
401 smp_mb__before_atomic();
402 return cmpxchg(&vcpu->mode, IN_GUEST_MODE, EXITING_GUEST_MODE);
406 * Some of the bitops functions do not support too long bitmaps.
407 * This number must be determined not to exceed such limits.
409 #define KVM_MEM_MAX_NR_PAGES ((1UL << 31) - 1)
411 struct kvm_memory_slot {
413 unsigned long npages;
414 unsigned long *dirty_bitmap;
415 struct kvm_arch_memory_slot arch;
416 unsigned long userspace_addr;
422 static inline bool kvm_slot_dirty_track_enabled(struct kvm_memory_slot *slot)
424 return slot->flags & KVM_MEM_LOG_DIRTY_PAGES;
427 static inline unsigned long kvm_dirty_bitmap_bytes(struct kvm_memory_slot *memslot)
429 return ALIGN(memslot->npages, BITS_PER_LONG) / 8;
432 static inline unsigned long *kvm_second_dirty_bitmap(struct kvm_memory_slot *memslot)
434 unsigned long len = kvm_dirty_bitmap_bytes(memslot);
436 return memslot->dirty_bitmap + len / sizeof(*memslot->dirty_bitmap);
439 #ifndef KVM_DIRTY_LOG_MANUAL_CAPS
440 #define KVM_DIRTY_LOG_MANUAL_CAPS KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE
443 struct kvm_s390_adapter_int {
456 struct kvm_kernel_irq_routing_entry {
459 int (*set)(struct kvm_kernel_irq_routing_entry *e,
460 struct kvm *kvm, int irq_source_id, int level,
474 struct kvm_s390_adapter_int adapter;
475 struct kvm_hv_sint hv_sint;
477 struct hlist_node link;
480 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
481 struct kvm_irq_routing_table {
482 int chip[KVM_NR_IRQCHIPS][KVM_IRQCHIP_NUM_PINS];
485 * Array indexed by gsi. Each entry contains list of irq chips
486 * the gsi is connected to.
488 struct hlist_head map[];
492 #ifndef KVM_PRIVATE_MEM_SLOTS
493 #define KVM_PRIVATE_MEM_SLOTS 0
496 #define KVM_MEM_SLOTS_NUM SHRT_MAX
497 #define KVM_USER_MEM_SLOTS (KVM_MEM_SLOTS_NUM - KVM_PRIVATE_MEM_SLOTS)
499 #ifndef __KVM_VCPU_MULTIPLE_ADDRESS_SPACE
500 static inline int kvm_arch_vcpu_memslots_id(struct kvm_vcpu *vcpu)
508 * memslots are not sorted by id anymore, please use id_to_memslot()
509 * to get the memslot by its id.
511 struct kvm_memslots {
513 /* The mapping table from slot id to the index in memslots[]. */
514 short id_to_index[KVM_MEM_SLOTS_NUM];
517 struct kvm_memory_slot memslots[];
521 #ifdef KVM_HAVE_MMU_RWLOCK
525 #endif /* KVM_HAVE_MMU_RWLOCK */
527 struct mutex slots_lock;
530 * Protects the arch-specific fields of struct kvm_memory_slots in
531 * use by the VM. To be used under the slots_lock (above) or in a
532 * kvm->srcu critical section where acquiring the slots_lock would
533 * lead to deadlock with the synchronize_srcu in
534 * install_new_memslots.
536 struct mutex slots_arch_lock;
537 struct mm_struct *mm; /* userspace tied to this vm */
538 struct kvm_memslots __rcu *memslots[KVM_ADDRESS_SPACE_NUM];
539 struct kvm_vcpu *vcpus[KVM_MAX_VCPUS];
542 * created_vcpus is protected by kvm->lock, and is incremented
543 * at the beginning of KVM_CREATE_VCPU. online_vcpus is only
544 * incremented after storing the kvm_vcpu pointer in vcpus,
545 * and is accessed atomically.
547 atomic_t online_vcpus;
549 int last_boosted_vcpu;
550 struct list_head vm_list;
552 struct kvm_io_bus __rcu *buses[KVM_NR_BUSES];
553 #ifdef CONFIG_HAVE_KVM_EVENTFD
556 struct list_head items;
557 struct list_head resampler_list;
558 struct mutex resampler_lock;
560 struct list_head ioeventfds;
562 struct kvm_vm_stat stat;
563 struct kvm_arch arch;
564 refcount_t users_count;
565 #ifdef CONFIG_KVM_MMIO
566 struct kvm_coalesced_mmio_ring *coalesced_mmio_ring;
567 spinlock_t ring_lock;
568 struct list_head coalesced_zones;
571 struct mutex irq_lock;
572 #ifdef CONFIG_HAVE_KVM_IRQCHIP
574 * Update side is protected by irq_lock.
576 struct kvm_irq_routing_table __rcu *irq_routing;
578 #ifdef CONFIG_HAVE_KVM_IRQFD
579 struct hlist_head irq_ack_notifier_list;
582 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
583 struct mmu_notifier mmu_notifier;
584 unsigned long mmu_notifier_seq;
585 long mmu_notifier_count;
586 unsigned long mmu_notifier_range_start;
587 unsigned long mmu_notifier_range_end;
590 struct list_head devices;
591 u64 manual_dirty_log_protect;
592 struct dentry *debugfs_dentry;
593 struct kvm_stat_data **debugfs_stat_data;
594 struct srcu_struct srcu;
595 struct srcu_struct irq_srcu;
597 unsigned int max_halt_poll_ns;
600 #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER
601 struct notifier_block pm_notifier;
603 char stats_id[KVM_STATS_NAME_SIZE];
606 #define kvm_err(fmt, ...) \
607 pr_err("kvm [%i]: " fmt, task_pid_nr(current), ## __VA_ARGS__)
608 #define kvm_info(fmt, ...) \
609 pr_info("kvm [%i]: " fmt, task_pid_nr(current), ## __VA_ARGS__)
610 #define kvm_debug(fmt, ...) \
611 pr_debug("kvm [%i]: " fmt, task_pid_nr(current), ## __VA_ARGS__)
612 #define kvm_debug_ratelimited(fmt, ...) \
613 pr_debug_ratelimited("kvm [%i]: " fmt, task_pid_nr(current), \
615 #define kvm_pr_unimpl(fmt, ...) \
616 pr_err_ratelimited("kvm [%i]: " fmt, \
617 task_tgid_nr(current), ## __VA_ARGS__)
619 /* The guest did something we don't support. */
620 #define vcpu_unimpl(vcpu, fmt, ...) \
621 kvm_pr_unimpl("vcpu%i, guest rIP: 0x%lx " fmt, \
622 (vcpu)->vcpu_id, kvm_rip_read(vcpu), ## __VA_ARGS__)
624 #define vcpu_debug(vcpu, fmt, ...) \
625 kvm_debug("vcpu%i " fmt, (vcpu)->vcpu_id, ## __VA_ARGS__)
626 #define vcpu_debug_ratelimited(vcpu, fmt, ...) \
627 kvm_debug_ratelimited("vcpu%i " fmt, (vcpu)->vcpu_id, \
629 #define vcpu_err(vcpu, fmt, ...) \
630 kvm_err("vcpu%i " fmt, (vcpu)->vcpu_id, ## __VA_ARGS__)
632 static inline bool kvm_dirty_log_manual_protect_and_init_set(struct kvm *kvm)
634 return !!(kvm->manual_dirty_log_protect & KVM_DIRTY_LOG_INITIALLY_SET);
637 static inline struct kvm_io_bus *kvm_get_bus(struct kvm *kvm, enum kvm_bus idx)
639 return srcu_dereference_check(kvm->buses[idx], &kvm->srcu,
640 lockdep_is_held(&kvm->slots_lock) ||
641 !refcount_read(&kvm->users_count));
644 static inline struct kvm_vcpu *kvm_get_vcpu(struct kvm *kvm, int i)
646 int num_vcpus = atomic_read(&kvm->online_vcpus);
647 i = array_index_nospec(i, num_vcpus);
649 /* Pairs with smp_wmb() in kvm_vm_ioctl_create_vcpu. */
651 return kvm->vcpus[i];
654 #define kvm_for_each_vcpu(idx, vcpup, kvm) \
656 idx < atomic_read(&kvm->online_vcpus) && \
657 (vcpup = kvm_get_vcpu(kvm, idx)) != NULL; \
660 static inline struct kvm_vcpu *kvm_get_vcpu_by_id(struct kvm *kvm, int id)
662 struct kvm_vcpu *vcpu = NULL;
667 if (id < KVM_MAX_VCPUS)
668 vcpu = kvm_get_vcpu(kvm, id);
669 if (vcpu && vcpu->vcpu_id == id)
671 kvm_for_each_vcpu(i, vcpu, kvm)
672 if (vcpu->vcpu_id == id)
677 static inline int kvm_vcpu_get_idx(struct kvm_vcpu *vcpu)
679 return vcpu->vcpu_idx;
682 #define kvm_for_each_memslot(memslot, slots) \
683 for (memslot = &slots->memslots[0]; \
684 memslot < slots->memslots + slots->used_slots; memslot++) \
685 if (WARN_ON_ONCE(!memslot->npages)) { \
688 void kvm_vcpu_destroy(struct kvm_vcpu *vcpu);
690 void vcpu_load(struct kvm_vcpu *vcpu);
691 void vcpu_put(struct kvm_vcpu *vcpu);
693 #ifdef __KVM_HAVE_IOAPIC
694 void kvm_arch_post_irq_ack_notifier_list_update(struct kvm *kvm);
695 void kvm_arch_post_irq_routing_update(struct kvm *kvm);
697 static inline void kvm_arch_post_irq_ack_notifier_list_update(struct kvm *kvm)
700 static inline void kvm_arch_post_irq_routing_update(struct kvm *kvm)
705 #ifdef CONFIG_HAVE_KVM_IRQFD
706 int kvm_irqfd_init(void);
707 void kvm_irqfd_exit(void);
709 static inline int kvm_irqfd_init(void)
714 static inline void kvm_irqfd_exit(void)
718 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
719 struct module *module);
722 void kvm_get_kvm(struct kvm *kvm);
723 void kvm_put_kvm(struct kvm *kvm);
724 bool file_is_kvm(struct file *file);
725 void kvm_put_kvm_no_destroy(struct kvm *kvm);
727 static inline struct kvm_memslots *__kvm_memslots(struct kvm *kvm, int as_id)
729 as_id = array_index_nospec(as_id, KVM_ADDRESS_SPACE_NUM);
730 return srcu_dereference_check(kvm->memslots[as_id], &kvm->srcu,
731 lockdep_is_held(&kvm->slots_lock) ||
732 !refcount_read(&kvm->users_count));
735 static inline struct kvm_memslots *kvm_memslots(struct kvm *kvm)
737 return __kvm_memslots(kvm, 0);
740 static inline struct kvm_memslots *kvm_vcpu_memslots(struct kvm_vcpu *vcpu)
742 int as_id = kvm_arch_vcpu_memslots_id(vcpu);
744 return __kvm_memslots(vcpu->kvm, as_id);
748 struct kvm_memory_slot *id_to_memslot(struct kvm_memslots *slots, int id)
750 int index = slots->id_to_index[id];
751 struct kvm_memory_slot *slot;
756 slot = &slots->memslots[index];
758 WARN_ON(slot->id != id);
763 * KVM_SET_USER_MEMORY_REGION ioctl allows the following operations:
764 * - create a new memory slot
765 * - delete an existing memory slot
766 * - modify an existing memory slot
767 * -- move it in the guest physical memory space
768 * -- just change its flags
770 * Since flags can be changed by some of these operations, the following
771 * differentiation is the best we can do for __kvm_set_memory_region():
780 int kvm_set_memory_region(struct kvm *kvm,
781 const struct kvm_userspace_memory_region *mem);
782 int __kvm_set_memory_region(struct kvm *kvm,
783 const struct kvm_userspace_memory_region *mem);
784 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot);
785 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen);
786 int kvm_arch_prepare_memory_region(struct kvm *kvm,
787 struct kvm_memory_slot *memslot,
788 const struct kvm_userspace_memory_region *mem,
789 enum kvm_mr_change change);
790 void kvm_arch_commit_memory_region(struct kvm *kvm,
791 const struct kvm_userspace_memory_region *mem,
792 struct kvm_memory_slot *old,
793 const struct kvm_memory_slot *new,
794 enum kvm_mr_change change);
795 /* flush all memory translations */
796 void kvm_arch_flush_shadow_all(struct kvm *kvm);
797 /* flush memory translations pointing to 'slot' */
798 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
799 struct kvm_memory_slot *slot);
801 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
802 struct page **pages, int nr_pages);
804 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn);
805 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn);
806 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable);
807 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot, gfn_t gfn);
808 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot, gfn_t gfn,
810 void kvm_release_page_clean(struct page *page);
811 void kvm_release_page_dirty(struct page *page);
812 void kvm_set_page_accessed(struct page *page);
814 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn);
815 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
817 kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn);
818 kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn);
819 kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
820 bool atomic, bool *async, bool write_fault,
821 bool *writable, hva_t *hva);
823 void kvm_release_pfn_clean(kvm_pfn_t pfn);
824 void kvm_release_pfn_dirty(kvm_pfn_t pfn);
825 void kvm_set_pfn_dirty(kvm_pfn_t pfn);
826 void kvm_set_pfn_accessed(kvm_pfn_t pfn);
827 void kvm_get_pfn(kvm_pfn_t pfn);
829 void kvm_release_pfn(kvm_pfn_t pfn, bool dirty, struct gfn_to_pfn_cache *cache);
830 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
832 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len);
833 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
834 void *data, unsigned long len);
835 int kvm_read_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
836 void *data, unsigned int offset,
838 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
839 int offset, int len);
840 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
842 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
843 void *data, unsigned long len);
844 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
845 void *data, unsigned int offset,
847 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
848 gpa_t gpa, unsigned long len);
850 #define __kvm_get_guest(kvm, gfn, offset, v) \
852 unsigned long __addr = gfn_to_hva(kvm, gfn); \
853 typeof(v) __user *__uaddr = (typeof(__uaddr))(__addr + offset); \
854 int __ret = -EFAULT; \
856 if (!kvm_is_error_hva(__addr)) \
857 __ret = get_user(v, __uaddr); \
861 #define kvm_get_guest(kvm, gpa, v) \
864 struct kvm *__kvm = kvm; \
866 __kvm_get_guest(__kvm, __gpa >> PAGE_SHIFT, \
867 offset_in_page(__gpa), v); \
870 #define __kvm_put_guest(kvm, gfn, offset, v) \
872 unsigned long __addr = gfn_to_hva(kvm, gfn); \
873 typeof(v) __user *__uaddr = (typeof(__uaddr))(__addr + offset); \
874 int __ret = -EFAULT; \
876 if (!kvm_is_error_hva(__addr)) \
877 __ret = put_user(v, __uaddr); \
879 mark_page_dirty(kvm, gfn); \
883 #define kvm_put_guest(kvm, gpa, v) \
886 struct kvm *__kvm = kvm; \
888 __kvm_put_guest(__kvm, __gpa >> PAGE_SHIFT, \
889 offset_in_page(__gpa), v); \
892 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len);
893 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn);
894 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn);
895 bool kvm_vcpu_is_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn);
896 unsigned long kvm_host_page_size(struct kvm_vcpu *vcpu, gfn_t gfn);
897 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot, gfn_t gfn);
898 void mark_page_dirty(struct kvm *kvm, gfn_t gfn);
900 struct kvm_memslots *kvm_vcpu_memslots(struct kvm_vcpu *vcpu);
901 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn);
902 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn);
903 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn);
904 int kvm_vcpu_map(struct kvm_vcpu *vcpu, gpa_t gpa, struct kvm_host_map *map);
905 int kvm_map_gfn(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map,
906 struct gfn_to_pfn_cache *cache, bool atomic);
907 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn);
908 void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map, bool dirty);
909 int kvm_unmap_gfn(struct kvm_vcpu *vcpu, struct kvm_host_map *map,
910 struct gfn_to_pfn_cache *cache, bool dirty, bool atomic);
911 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn);
912 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable);
913 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data, int offset,
915 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa, void *data,
917 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data,
919 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, const void *data,
920 int offset, int len);
921 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
923 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn);
925 void kvm_sigset_activate(struct kvm_vcpu *vcpu);
926 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu);
928 void kvm_vcpu_block(struct kvm_vcpu *vcpu);
929 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu);
930 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu);
931 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu);
932 void kvm_vcpu_kick(struct kvm_vcpu *vcpu);
933 int kvm_vcpu_yield_to(struct kvm_vcpu *target);
934 void kvm_vcpu_on_spin(struct kvm_vcpu *vcpu, bool usermode_vcpu_not_eligible);
936 void kvm_flush_remote_tlbs(struct kvm *kvm);
937 void kvm_reload_remote_mmus(struct kvm *kvm);
939 #ifdef KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE
940 int kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache *mc, int min);
941 int kvm_mmu_memory_cache_nr_free_objects(struct kvm_mmu_memory_cache *mc);
942 void kvm_mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc);
943 void *kvm_mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc);
946 bool kvm_make_vcpus_request_mask(struct kvm *kvm, unsigned int req,
947 struct kvm_vcpu *except,
948 unsigned long *vcpu_bitmap, cpumask_var_t tmp);
949 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req);
950 bool kvm_make_all_cpus_request_except(struct kvm *kvm, unsigned int req,
951 struct kvm_vcpu *except);
952 bool kvm_make_cpus_request_mask(struct kvm *kvm, unsigned int req,
953 unsigned long *vcpu_bitmap);
955 long kvm_arch_dev_ioctl(struct file *filp,
956 unsigned int ioctl, unsigned long arg);
957 long kvm_arch_vcpu_ioctl(struct file *filp,
958 unsigned int ioctl, unsigned long arg);
959 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf);
961 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext);
963 void kvm_arch_mmu_enable_log_dirty_pt_masked(struct kvm *kvm,
964 struct kvm_memory_slot *slot,
967 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot);
969 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
970 void kvm_arch_flush_remote_tlbs_memslot(struct kvm *kvm,
971 const struct kvm_memory_slot *memslot);
972 #else /* !CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
973 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log);
974 int kvm_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log,
975 int *is_dirty, struct kvm_memory_slot **memslot);
978 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
980 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
981 struct kvm_enable_cap *cap);
982 long kvm_arch_vm_ioctl(struct file *filp,
983 unsigned int ioctl, unsigned long arg);
985 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu);
986 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu);
988 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
989 struct kvm_translation *tr);
991 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs);
992 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs);
993 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
994 struct kvm_sregs *sregs);
995 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
996 struct kvm_sregs *sregs);
997 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
998 struct kvm_mp_state *mp_state);
999 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
1000 struct kvm_mp_state *mp_state);
1001 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
1002 struct kvm_guest_debug *dbg);
1003 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu);
1005 int kvm_arch_init(void *opaque);
1006 void kvm_arch_exit(void);
1008 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu);
1010 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu);
1011 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu);
1012 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id);
1013 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu);
1014 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu);
1015 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu);
1017 #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER
1018 int kvm_arch_pm_notifier(struct kvm *kvm, unsigned long state);
1021 #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS
1022 void kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu, struct dentry *debugfs_dentry);
1025 int kvm_arch_hardware_enable(void);
1026 void kvm_arch_hardware_disable(void);
1027 int kvm_arch_hardware_setup(void *opaque);
1028 void kvm_arch_hardware_unsetup(void);
1029 int kvm_arch_check_processor_compat(void *opaque);
1030 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu);
1031 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu);
1032 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu);
1033 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu);
1034 bool kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu);
1035 int kvm_arch_post_init_vm(struct kvm *kvm);
1036 void kvm_arch_pre_destroy_vm(struct kvm *kvm);
1038 #ifndef __KVM_HAVE_ARCH_VM_ALLOC
1040 * All architectures that want to use vzalloc currently also
1041 * need their own kvm_arch_alloc_vm implementation.
1043 static inline struct kvm *kvm_arch_alloc_vm(void)
1045 return kzalloc(sizeof(struct kvm), GFP_KERNEL);
1048 static inline void kvm_arch_free_vm(struct kvm *kvm)
1054 #ifndef __KVM_HAVE_ARCH_FLUSH_REMOTE_TLB
1055 static inline int kvm_arch_flush_remote_tlb(struct kvm *kvm)
1061 #ifdef __KVM_HAVE_ARCH_NONCOHERENT_DMA
1062 void kvm_arch_register_noncoherent_dma(struct kvm *kvm);
1063 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm);
1064 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm);
1066 static inline void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
1070 static inline void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
1074 static inline bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
1079 #ifdef __KVM_HAVE_ARCH_ASSIGNED_DEVICE
1080 void kvm_arch_start_assignment(struct kvm *kvm);
1081 void kvm_arch_end_assignment(struct kvm *kvm);
1082 bool kvm_arch_has_assigned_device(struct kvm *kvm);
1084 static inline void kvm_arch_start_assignment(struct kvm *kvm)
1088 static inline void kvm_arch_end_assignment(struct kvm *kvm)
1092 static inline bool kvm_arch_has_assigned_device(struct kvm *kvm)
1098 static inline struct rcuwait *kvm_arch_vcpu_get_wait(struct kvm_vcpu *vcpu)
1100 #ifdef __KVM_HAVE_ARCH_WQP
1101 return vcpu->arch.waitp;
1107 #ifdef __KVM_HAVE_ARCH_INTC_INITIALIZED
1109 * returns true if the virtual interrupt controller is initialized and
1110 * ready to accept virtual IRQ. On some architectures the virtual interrupt
1111 * controller is dynamically instantiated and this is not always true.
1113 bool kvm_arch_intc_initialized(struct kvm *kvm);
1115 static inline bool kvm_arch_intc_initialized(struct kvm *kvm)
1121 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type);
1122 void kvm_arch_destroy_vm(struct kvm *kvm);
1123 void kvm_arch_sync_events(struct kvm *kvm);
1125 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu);
1127 bool kvm_is_reserved_pfn(kvm_pfn_t pfn);
1128 bool kvm_is_zone_device_pfn(kvm_pfn_t pfn);
1129 bool kvm_is_transparent_hugepage(kvm_pfn_t pfn);
1131 struct kvm_irq_ack_notifier {
1132 struct hlist_node link;
1134 void (*irq_acked)(struct kvm_irq_ack_notifier *kian);
1137 int kvm_irq_map_gsi(struct kvm *kvm,
1138 struct kvm_kernel_irq_routing_entry *entries, int gsi);
1139 int kvm_irq_map_chip_pin(struct kvm *kvm, unsigned irqchip, unsigned pin);
1141 int kvm_set_irq(struct kvm *kvm, int irq_source_id, u32 irq, int level,
1143 int kvm_set_msi(struct kvm_kernel_irq_routing_entry *irq_entry, struct kvm *kvm,
1144 int irq_source_id, int level, bool line_status);
1145 int kvm_arch_set_irq_inatomic(struct kvm_kernel_irq_routing_entry *e,
1146 struct kvm *kvm, int irq_source_id,
1147 int level, bool line_status);
1148 bool kvm_irq_has_notifier(struct kvm *kvm, unsigned irqchip, unsigned pin);
1149 void kvm_notify_acked_gsi(struct kvm *kvm, int gsi);
1150 void kvm_notify_acked_irq(struct kvm *kvm, unsigned irqchip, unsigned pin);
1151 void kvm_register_irq_ack_notifier(struct kvm *kvm,
1152 struct kvm_irq_ack_notifier *kian);
1153 void kvm_unregister_irq_ack_notifier(struct kvm *kvm,
1154 struct kvm_irq_ack_notifier *kian);
1155 int kvm_request_irq_source_id(struct kvm *kvm);
1156 void kvm_free_irq_source_id(struct kvm *kvm, int irq_source_id);
1157 bool kvm_arch_irqfd_allowed(struct kvm *kvm, struct kvm_irqfd *args);
1160 * search_memslots() and __gfn_to_memslot() are here because they are
1161 * used in non-modular code in arch/powerpc/kvm/book3s_hv_rm_mmu.c.
1162 * gfn_to_memslot() itself isn't here as an inline because that would
1163 * bloat other code too much.
1165 * IMPORTANT: Slots are sorted from highest GFN to lowest GFN!
1167 static inline struct kvm_memory_slot *
1168 search_memslots(struct kvm_memslots *slots, gfn_t gfn)
1170 int start = 0, end = slots->used_slots;
1171 int slot = atomic_read(&slots->lru_slot);
1172 struct kvm_memory_slot *memslots = slots->memslots;
1174 if (unlikely(!slots->used_slots))
1177 if (gfn >= memslots[slot].base_gfn &&
1178 gfn < memslots[slot].base_gfn + memslots[slot].npages)
1179 return &memslots[slot];
1181 while (start < end) {
1182 slot = start + (end - start) / 2;
1184 if (gfn >= memslots[slot].base_gfn)
1190 if (start < slots->used_slots && gfn >= memslots[start].base_gfn &&
1191 gfn < memslots[start].base_gfn + memslots[start].npages) {
1192 atomic_set(&slots->lru_slot, start);
1193 return &memslots[start];
1199 static inline struct kvm_memory_slot *
1200 __gfn_to_memslot(struct kvm_memslots *slots, gfn_t gfn)
1202 return search_memslots(slots, gfn);
1205 static inline unsigned long
1206 __gfn_to_hva_memslot(const struct kvm_memory_slot *slot, gfn_t gfn)
1209 * The index was checked originally in search_memslots. To avoid
1210 * that a malicious guest builds a Spectre gadget out of e.g. page
1211 * table walks, do not let the processor speculate loads outside
1212 * the guest's registered memslots.
1214 unsigned long offset = gfn - slot->base_gfn;
1215 offset = array_index_nospec(offset, slot->npages);
1216 return slot->userspace_addr + offset * PAGE_SIZE;
1219 static inline int memslot_id(struct kvm *kvm, gfn_t gfn)
1221 return gfn_to_memslot(kvm, gfn)->id;
1225 hva_to_gfn_memslot(unsigned long hva, struct kvm_memory_slot *slot)
1227 gfn_t gfn_offset = (hva - slot->userspace_addr) >> PAGE_SHIFT;
1229 return slot->base_gfn + gfn_offset;
1232 static inline gpa_t gfn_to_gpa(gfn_t gfn)
1234 return (gpa_t)gfn << PAGE_SHIFT;
1237 static inline gfn_t gpa_to_gfn(gpa_t gpa)
1239 return (gfn_t)(gpa >> PAGE_SHIFT);
1242 static inline hpa_t pfn_to_hpa(kvm_pfn_t pfn)
1244 return (hpa_t)pfn << PAGE_SHIFT;
1247 static inline struct page *kvm_vcpu_gpa_to_page(struct kvm_vcpu *vcpu,
1250 return kvm_vcpu_gfn_to_page(vcpu, gpa_to_gfn(gpa));
1253 static inline bool kvm_is_error_gpa(struct kvm *kvm, gpa_t gpa)
1255 unsigned long hva = gfn_to_hva(kvm, gpa_to_gfn(gpa));
1257 return kvm_is_error_hva(hva);
1260 enum kvm_stat_kind {
1265 struct kvm_stat_data {
1267 const struct _kvm_stats_desc *desc;
1268 enum kvm_stat_kind kind;
1271 struct _kvm_stats_desc {
1272 struct kvm_stats_desc desc;
1273 char name[KVM_STATS_NAME_SIZE];
1276 #define STATS_DESC_COMMON(type, unit, base, exp) \
1277 .flags = type | unit | base | \
1278 BUILD_BUG_ON_ZERO(type & ~KVM_STATS_TYPE_MASK) | \
1279 BUILD_BUG_ON_ZERO(unit & ~KVM_STATS_UNIT_MASK) | \
1280 BUILD_BUG_ON_ZERO(base & ~KVM_STATS_BASE_MASK), \
1284 #define VM_GENERIC_STATS_DESC(stat, type, unit, base, exp) \
1287 STATS_DESC_COMMON(type, unit, base, exp), \
1288 .offset = offsetof(struct kvm_vm_stat, generic.stat) \
1292 #define VCPU_GENERIC_STATS_DESC(stat, type, unit, base, exp) \
1295 STATS_DESC_COMMON(type, unit, base, exp), \
1296 .offset = offsetof(struct kvm_vcpu_stat, generic.stat) \
1300 #define VM_STATS_DESC(stat, type, unit, base, exp) \
1303 STATS_DESC_COMMON(type, unit, base, exp), \
1304 .offset = offsetof(struct kvm_vm_stat, stat) \
1308 #define VCPU_STATS_DESC(stat, type, unit, base, exp) \
1311 STATS_DESC_COMMON(type, unit, base, exp), \
1312 .offset = offsetof(struct kvm_vcpu_stat, stat) \
1316 /* SCOPE: VM, VM_GENERIC, VCPU, VCPU_GENERIC */
1317 #define STATS_DESC(SCOPE, stat, type, unit, base, exp) \
1318 SCOPE##_STATS_DESC(stat, type, unit, base, exp)
1320 #define STATS_DESC_CUMULATIVE(SCOPE, name, unit, base, exponent) \
1321 STATS_DESC(SCOPE, name, KVM_STATS_TYPE_CUMULATIVE, unit, base, exponent)
1322 #define STATS_DESC_INSTANT(SCOPE, name, unit, base, exponent) \
1323 STATS_DESC(SCOPE, name, KVM_STATS_TYPE_INSTANT, unit, base, exponent)
1324 #define STATS_DESC_PEAK(SCOPE, name, unit, base, exponent) \
1325 STATS_DESC(SCOPE, name, KVM_STATS_TYPE_PEAK, unit, base, exponent)
1327 /* Cumulative counter, read/write */
1328 #define STATS_DESC_COUNTER(SCOPE, name) \
1329 STATS_DESC_CUMULATIVE(SCOPE, name, KVM_STATS_UNIT_NONE, \
1330 KVM_STATS_BASE_POW10, 0)
1331 /* Instantaneous counter, read only */
1332 #define STATS_DESC_ICOUNTER(SCOPE, name) \
1333 STATS_DESC_INSTANT(SCOPE, name, KVM_STATS_UNIT_NONE, \
1334 KVM_STATS_BASE_POW10, 0)
1335 /* Peak counter, read/write */
1336 #define STATS_DESC_PCOUNTER(SCOPE, name) \
1337 STATS_DESC_PEAK(SCOPE, name, KVM_STATS_UNIT_NONE, \
1338 KVM_STATS_BASE_POW10, 0)
1340 /* Cumulative time in nanosecond */
1341 #define STATS_DESC_TIME_NSEC(SCOPE, name) \
1342 STATS_DESC_CUMULATIVE(SCOPE, name, KVM_STATS_UNIT_SECONDS, \
1343 KVM_STATS_BASE_POW10, -9)
1345 #define KVM_GENERIC_VM_STATS() \
1346 STATS_DESC_COUNTER(VM_GENERIC, remote_tlb_flush)
1348 #define KVM_GENERIC_VCPU_STATS() \
1349 STATS_DESC_COUNTER(VCPU_GENERIC, halt_successful_poll), \
1350 STATS_DESC_COUNTER(VCPU_GENERIC, halt_attempted_poll), \
1351 STATS_DESC_COUNTER(VCPU_GENERIC, halt_poll_invalid), \
1352 STATS_DESC_COUNTER(VCPU_GENERIC, halt_wakeup), \
1353 STATS_DESC_TIME_NSEC(VCPU_GENERIC, halt_poll_success_ns), \
1354 STATS_DESC_TIME_NSEC(VCPU_GENERIC, halt_poll_fail_ns)
1356 extern struct dentry *kvm_debugfs_dir;
1357 ssize_t kvm_stats_read(char *id, const struct kvm_stats_header *header,
1358 const struct _kvm_stats_desc *desc,
1359 void *stats, size_t size_stats,
1360 char __user *user_buffer, size_t size, loff_t *offset);
1361 extern const struct kvm_stats_header kvm_vm_stats_header;
1362 extern const struct _kvm_stats_desc kvm_vm_stats_desc[];
1363 extern const struct kvm_stats_header kvm_vcpu_stats_header;
1364 extern const struct _kvm_stats_desc kvm_vcpu_stats_desc[];
1366 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1367 static inline int mmu_notifier_retry(struct kvm *kvm, unsigned long mmu_seq)
1369 if (unlikely(kvm->mmu_notifier_count))
1372 * Ensure the read of mmu_notifier_count happens before the read
1373 * of mmu_notifier_seq. This interacts with the smp_wmb() in
1374 * mmu_notifier_invalidate_range_end to make sure that the caller
1375 * either sees the old (non-zero) value of mmu_notifier_count or
1376 * the new (incremented) value of mmu_notifier_seq.
1377 * PowerPC Book3s HV KVM calls this under a per-page lock
1378 * rather than under kvm->mmu_lock, for scalability, so
1379 * can't rely on kvm->mmu_lock to keep things ordered.
1382 if (kvm->mmu_notifier_seq != mmu_seq)
1387 static inline int mmu_notifier_retry_hva(struct kvm *kvm,
1388 unsigned long mmu_seq,
1391 lockdep_assert_held(&kvm->mmu_lock);
1393 * If mmu_notifier_count is non-zero, then the range maintained by
1394 * kvm_mmu_notifier_invalidate_range_start contains all addresses that
1395 * might be being invalidated. Note that it may include some false
1396 * positives, due to shortcuts when handing concurrent invalidations.
1398 if (unlikely(kvm->mmu_notifier_count) &&
1399 hva >= kvm->mmu_notifier_range_start &&
1400 hva < kvm->mmu_notifier_range_end)
1402 if (kvm->mmu_notifier_seq != mmu_seq)
1408 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
1410 #define KVM_MAX_IRQ_ROUTES 4096 /* might need extension/rework in the future */
1412 bool kvm_arch_can_set_irq_routing(struct kvm *kvm);
1413 int kvm_set_irq_routing(struct kvm *kvm,
1414 const struct kvm_irq_routing_entry *entries,
1417 int kvm_set_routing_entry(struct kvm *kvm,
1418 struct kvm_kernel_irq_routing_entry *e,
1419 const struct kvm_irq_routing_entry *ue);
1420 void kvm_free_irq_routing(struct kvm *kvm);
1424 static inline void kvm_free_irq_routing(struct kvm *kvm) {}
1428 int kvm_send_userspace_msi(struct kvm *kvm, struct kvm_msi *msi);
1430 #ifdef CONFIG_HAVE_KVM_EVENTFD
1432 void kvm_eventfd_init(struct kvm *kvm);
1433 int kvm_ioeventfd(struct kvm *kvm, struct kvm_ioeventfd *args);
1435 #ifdef CONFIG_HAVE_KVM_IRQFD
1436 int kvm_irqfd(struct kvm *kvm, struct kvm_irqfd *args);
1437 void kvm_irqfd_release(struct kvm *kvm);
1438 void kvm_irq_routing_update(struct kvm *);
1440 static inline int kvm_irqfd(struct kvm *kvm, struct kvm_irqfd *args)
1445 static inline void kvm_irqfd_release(struct kvm *kvm) {}
1450 static inline void kvm_eventfd_init(struct kvm *kvm) {}
1452 static inline int kvm_irqfd(struct kvm *kvm, struct kvm_irqfd *args)
1457 static inline void kvm_irqfd_release(struct kvm *kvm) {}
1459 #ifdef CONFIG_HAVE_KVM_IRQCHIP
1460 static inline void kvm_irq_routing_update(struct kvm *kvm)
1465 static inline int kvm_ioeventfd(struct kvm *kvm, struct kvm_ioeventfd *args)
1470 #endif /* CONFIG_HAVE_KVM_EVENTFD */
1472 void kvm_arch_irq_routing_update(struct kvm *kvm);
1474 static inline void kvm_make_request(int req, struct kvm_vcpu *vcpu)
1477 * Ensure the rest of the request is published to kvm_check_request's
1478 * caller. Paired with the smp_mb__after_atomic in kvm_check_request.
1481 set_bit(req & KVM_REQUEST_MASK, (void *)&vcpu->requests);
1484 static inline bool kvm_request_pending(struct kvm_vcpu *vcpu)
1486 return READ_ONCE(vcpu->requests);
1489 static inline bool kvm_test_request(int req, struct kvm_vcpu *vcpu)
1491 return test_bit(req & KVM_REQUEST_MASK, (void *)&vcpu->requests);
1494 static inline void kvm_clear_request(int req, struct kvm_vcpu *vcpu)
1496 clear_bit(req & KVM_REQUEST_MASK, (void *)&vcpu->requests);
1499 static inline bool kvm_check_request(int req, struct kvm_vcpu *vcpu)
1501 if (kvm_test_request(req, vcpu)) {
1502 kvm_clear_request(req, vcpu);
1505 * Ensure the rest of the request is visible to kvm_check_request's
1506 * caller. Paired with the smp_wmb in kvm_make_request.
1508 smp_mb__after_atomic();
1515 extern bool kvm_rebooting;
1517 extern unsigned int halt_poll_ns;
1518 extern unsigned int halt_poll_ns_grow;
1519 extern unsigned int halt_poll_ns_grow_start;
1520 extern unsigned int halt_poll_ns_shrink;
1523 const struct kvm_device_ops *ops;
1526 struct list_head vm_node;
1529 /* create, destroy, and name are mandatory */
1530 struct kvm_device_ops {
1534 * create is called holding kvm->lock and any operations not suitable
1535 * to do while holding the lock should be deferred to init (see
1538 int (*create)(struct kvm_device *dev, u32 type);
1541 * init is called after create if create is successful and is called
1542 * outside of holding kvm->lock.
1544 void (*init)(struct kvm_device *dev);
1547 * Destroy is responsible for freeing dev.
1549 * Destroy may be called before or after destructors are called
1550 * on emulated I/O regions, depending on whether a reference is
1551 * held by a vcpu or other kvm component that gets destroyed
1552 * after the emulated I/O.
1554 void (*destroy)(struct kvm_device *dev);
1557 * Release is an alternative method to free the device. It is
1558 * called when the device file descriptor is closed. Once
1559 * release is called, the destroy method will not be called
1560 * anymore as the device is removed from the device list of
1561 * the VM. kvm->lock is held.
1563 void (*release)(struct kvm_device *dev);
1565 int (*set_attr)(struct kvm_device *dev, struct kvm_device_attr *attr);
1566 int (*get_attr)(struct kvm_device *dev, struct kvm_device_attr *attr);
1567 int (*has_attr)(struct kvm_device *dev, struct kvm_device_attr *attr);
1568 long (*ioctl)(struct kvm_device *dev, unsigned int ioctl,
1570 int (*mmap)(struct kvm_device *dev, struct vm_area_struct *vma);
1573 void kvm_device_get(struct kvm_device *dev);
1574 void kvm_device_put(struct kvm_device *dev);
1575 struct kvm_device *kvm_device_from_filp(struct file *filp);
1576 int kvm_register_device_ops(const struct kvm_device_ops *ops, u32 type);
1577 void kvm_unregister_device_ops(u32 type);
1579 extern struct kvm_device_ops kvm_mpic_ops;
1580 extern struct kvm_device_ops kvm_arm_vgic_v2_ops;
1581 extern struct kvm_device_ops kvm_arm_vgic_v3_ops;
1583 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1585 static inline void kvm_vcpu_set_in_spin_loop(struct kvm_vcpu *vcpu, bool val)
1587 vcpu->spin_loop.in_spin_loop = val;
1589 static inline void kvm_vcpu_set_dy_eligible(struct kvm_vcpu *vcpu, bool val)
1591 vcpu->spin_loop.dy_eligible = val;
1594 #else /* !CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT */
1596 static inline void kvm_vcpu_set_in_spin_loop(struct kvm_vcpu *vcpu, bool val)
1600 static inline void kvm_vcpu_set_dy_eligible(struct kvm_vcpu *vcpu, bool val)
1603 #endif /* CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT */
1605 static inline bool kvm_is_visible_memslot(struct kvm_memory_slot *memslot)
1607 return (memslot && memslot->id < KVM_USER_MEM_SLOTS &&
1608 !(memslot->flags & KVM_MEMSLOT_INVALID));
1611 struct kvm_vcpu *kvm_get_running_vcpu(void);
1612 struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void);
1614 #ifdef CONFIG_HAVE_KVM_IRQ_BYPASS
1615 bool kvm_arch_has_irq_bypass(void);
1616 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *,
1617 struct irq_bypass_producer *);
1618 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *,
1619 struct irq_bypass_producer *);
1620 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *);
1621 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *);
1622 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
1623 uint32_t guest_irq, bool set);
1624 #endif /* CONFIG_HAVE_KVM_IRQ_BYPASS */
1626 #ifdef CONFIG_HAVE_KVM_INVALID_WAKEUPS
1627 /* If we wakeup during the poll time, was it a sucessful poll? */
1628 static inline bool vcpu_valid_wakeup(struct kvm_vcpu *vcpu)
1630 return vcpu->valid_wakeup;
1634 static inline bool vcpu_valid_wakeup(struct kvm_vcpu *vcpu)
1638 #endif /* CONFIG_HAVE_KVM_INVALID_WAKEUPS */
1640 #ifdef CONFIG_HAVE_KVM_NO_POLL
1641 /* Callback that tells if we must not poll */
1642 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu);
1644 static inline bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
1648 #endif /* CONFIG_HAVE_KVM_NO_POLL */
1650 #ifdef CONFIG_HAVE_KVM_VCPU_ASYNC_IOCTL
1651 long kvm_arch_vcpu_async_ioctl(struct file *filp,
1652 unsigned int ioctl, unsigned long arg);
1654 static inline long kvm_arch_vcpu_async_ioctl(struct file *filp,
1658 return -ENOIOCTLCMD;
1660 #endif /* CONFIG_HAVE_KVM_VCPU_ASYNC_IOCTL */
1662 void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
1663 unsigned long start, unsigned long end);
1665 #ifdef CONFIG_HAVE_KVM_VCPU_RUN_PID_CHANGE
1666 int kvm_arch_vcpu_run_pid_change(struct kvm_vcpu *vcpu);
1668 static inline int kvm_arch_vcpu_run_pid_change(struct kvm_vcpu *vcpu)
1672 #endif /* CONFIG_HAVE_KVM_VCPU_RUN_PID_CHANGE */
1674 typedef int (*kvm_vm_thread_fn_t)(struct kvm *kvm, uintptr_t data);
1676 int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn,
1677 uintptr_t data, const char *name,
1678 struct task_struct **thread_ptr);
1680 #ifdef CONFIG_KVM_XFER_TO_GUEST_WORK
1681 static inline void kvm_handle_signal_exit(struct kvm_vcpu *vcpu)
1683 vcpu->run->exit_reason = KVM_EXIT_INTR;
1684 vcpu->stat.signal_exits++;
1686 #endif /* CONFIG_KVM_XFER_TO_GUEST_WORK */
1689 * This defines how many reserved entries we want to keep before we
1690 * kick the vcpu to the userspace to avoid dirty ring full. This
1691 * value can be tuned to higher if e.g. PML is enabled on the host.
1693 #define KVM_DIRTY_RING_RSVD_ENTRIES 64
1695 /* Max number of entries allowed for each kvm dirty ring */
1696 #define KVM_DIRTY_RING_MAX_ENTRIES 65536