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 <linux/ftrace.h>
33 #include <linux/hashtable.h>
34 #include <linux/instrumentation.h>
35 #include <linux/interval_tree.h>
36 #include <linux/rbtree.h>
37 #include <linux/xarray.h>
38 #include <asm/signal.h>
40 #include <linux/kvm.h>
41 #include <linux/kvm_para.h>
43 #include <linux/kvm_types.h>
45 #include <asm/kvm_host.h>
46 #include <linux/kvm_dirty_ring.h>
48 #ifndef KVM_MAX_VCPU_IDS
49 #define KVM_MAX_VCPU_IDS KVM_MAX_VCPUS
53 * The bit 16 ~ bit 31 of kvm_userspace_memory_region::flags are internally
54 * used in kvm, other bits are visible for userspace which are defined in
55 * include/linux/kvm_h.
57 #define KVM_MEMSLOT_INVALID (1UL << 16)
60 * Bit 63 of the memslot generation number is an "update in-progress flag",
61 * e.g. is temporarily set for the duration of kvm_swap_active_memslots().
62 * This flag effectively creates a unique generation number that is used to
63 * mark cached memslot data, e.g. MMIO accesses, as potentially being stale,
64 * i.e. may (or may not) have come from the previous memslots generation.
66 * This is necessary because the actual memslots update is not atomic with
67 * respect to the generation number update. Updating the generation number
68 * first would allow a vCPU to cache a spte from the old memslots using the
69 * new generation number, and updating the generation number after switching
70 * to the new memslots would allow cache hits using the old generation number
71 * to reference the defunct memslots.
73 * This mechanism is used to prevent getting hits in KVM's caches while a
74 * memslot update is in-progress, and to prevent cache hits *after* updating
75 * the actual generation number against accesses that were inserted into the
76 * cache *before* the memslots were updated.
78 #define KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS BIT_ULL(63)
80 /* Two fragments for cross MMIO pages. */
81 #define KVM_MAX_MMIO_FRAGMENTS 2
83 #ifndef KVM_MAX_NR_ADDRESS_SPACES
84 #define KVM_MAX_NR_ADDRESS_SPACES 1
88 * For the normal pfn, the highest 12 bits should be zero,
89 * so we can mask bit 62 ~ bit 52 to indicate the error pfn,
90 * mask bit 63 to indicate the noslot pfn.
92 #define KVM_PFN_ERR_MASK (0x7ffULL << 52)
93 #define KVM_PFN_ERR_NOSLOT_MASK (0xfffULL << 52)
94 #define KVM_PFN_NOSLOT (0x1ULL << 63)
96 #define KVM_PFN_ERR_FAULT (KVM_PFN_ERR_MASK)
97 #define KVM_PFN_ERR_HWPOISON (KVM_PFN_ERR_MASK + 1)
98 #define KVM_PFN_ERR_RO_FAULT (KVM_PFN_ERR_MASK + 2)
99 #define KVM_PFN_ERR_SIGPENDING (KVM_PFN_ERR_MASK + 3)
102 * error pfns indicate that the gfn is in slot but faild to
103 * translate it to pfn on host.
105 static inline bool is_error_pfn(kvm_pfn_t pfn)
107 return !!(pfn & KVM_PFN_ERR_MASK);
111 * KVM_PFN_ERR_SIGPENDING indicates that fetching the PFN was interrupted
112 * by a pending signal. Note, the signal may or may not be fatal.
114 static inline bool is_sigpending_pfn(kvm_pfn_t pfn)
116 return pfn == KVM_PFN_ERR_SIGPENDING;
120 * error_noslot pfns indicate that the gfn can not be
121 * translated to pfn - it is not in slot or failed to
122 * translate it to pfn.
124 static inline bool is_error_noslot_pfn(kvm_pfn_t pfn)
126 return !!(pfn & KVM_PFN_ERR_NOSLOT_MASK);
129 /* noslot pfn indicates that the gfn is not in slot. */
130 static inline bool is_noslot_pfn(kvm_pfn_t pfn)
132 return pfn == KVM_PFN_NOSLOT;
136 * architectures with KVM_HVA_ERR_BAD other than PAGE_OFFSET (e.g. s390)
137 * provide own defines and kvm_is_error_hva
139 #ifndef KVM_HVA_ERR_BAD
141 #define KVM_HVA_ERR_BAD (PAGE_OFFSET)
142 #define KVM_HVA_ERR_RO_BAD (PAGE_OFFSET + PAGE_SIZE)
144 static inline bool kvm_is_error_hva(unsigned long addr)
146 return addr >= PAGE_OFFSET;
151 static inline bool kvm_is_error_gpa(gpa_t gpa)
153 return gpa == INVALID_GPA;
156 #define KVM_ERR_PTR_BAD_PAGE (ERR_PTR(-ENOENT))
158 static inline bool is_error_page(struct page *page)
163 #define KVM_REQUEST_MASK GENMASK(7,0)
164 #define KVM_REQUEST_NO_WAKEUP BIT(8)
165 #define KVM_REQUEST_WAIT BIT(9)
166 #define KVM_REQUEST_NO_ACTION BIT(10)
168 * Architecture-independent vcpu->requests bit members
169 * Bits 3-7 are reserved for more arch-independent bits.
171 #define KVM_REQ_TLB_FLUSH (0 | KVM_REQUEST_WAIT | KVM_REQUEST_NO_WAKEUP)
172 #define KVM_REQ_VM_DEAD (1 | KVM_REQUEST_WAIT | KVM_REQUEST_NO_WAKEUP)
173 #define KVM_REQ_UNBLOCK 2
174 #define KVM_REQ_DIRTY_RING_SOFT_FULL 3
175 #define KVM_REQUEST_ARCH_BASE 8
178 * KVM_REQ_OUTSIDE_GUEST_MODE exists is purely as way to force the vCPU to
179 * OUTSIDE_GUEST_MODE. KVM_REQ_OUTSIDE_GUEST_MODE differs from a vCPU "kick"
180 * in that it ensures the vCPU has reached OUTSIDE_GUEST_MODE before continuing
181 * on. A kick only guarantees that the vCPU is on its way out, e.g. a previous
182 * kick may have set vcpu->mode to EXITING_GUEST_MODE, and so there's no
183 * guarantee the vCPU received an IPI and has actually exited guest mode.
185 #define KVM_REQ_OUTSIDE_GUEST_MODE (KVM_REQUEST_NO_ACTION | KVM_REQUEST_WAIT | KVM_REQUEST_NO_WAKEUP)
187 #define KVM_ARCH_REQ_FLAGS(nr, flags) ({ \
188 BUILD_BUG_ON((unsigned)(nr) >= (sizeof_field(struct kvm_vcpu, requests) * 8) - KVM_REQUEST_ARCH_BASE); \
189 (unsigned)(((nr) + KVM_REQUEST_ARCH_BASE) | (flags)); \
191 #define KVM_ARCH_REQ(nr) KVM_ARCH_REQ_FLAGS(nr, 0)
193 bool kvm_make_vcpus_request_mask(struct kvm *kvm, unsigned int req,
194 unsigned long *vcpu_bitmap);
195 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req);
197 #define KVM_USERSPACE_IRQ_SOURCE_ID 0
198 #define KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID 1
200 extern struct mutex kvm_lock;
201 extern struct list_head vm_list;
203 struct kvm_io_range {
206 struct kvm_io_device *dev;
209 #define NR_IOBUS_DEVS 1000
214 struct kvm_io_range range[];
220 KVM_VIRTIO_CCW_NOTIFY_BUS,
225 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
226 int len, const void *val);
227 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
228 gpa_t addr, int len, const void *val, long cookie);
229 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
231 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
232 int len, struct kvm_io_device *dev);
233 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
234 struct kvm_io_device *dev);
235 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
238 #ifdef CONFIG_KVM_ASYNC_PF
239 struct kvm_async_pf {
240 struct work_struct work;
241 struct list_head link;
242 struct list_head queue;
243 struct kvm_vcpu *vcpu;
246 struct kvm_arch_async_pf arch;
248 bool notpresent_injected;
251 void kvm_clear_async_pf_completion_queue(struct kvm_vcpu *vcpu);
252 void kvm_check_async_pf_completion(struct kvm_vcpu *vcpu);
253 bool kvm_setup_async_pf(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
254 unsigned long hva, struct kvm_arch_async_pf *arch);
255 int kvm_async_pf_wakeup_all(struct kvm_vcpu *vcpu);
258 #ifdef CONFIG_KVM_GENERIC_MMU_NOTIFIER
259 union kvm_mmu_notifier_arg {
260 unsigned long attributes;
263 struct kvm_gfn_range {
264 struct kvm_memory_slot *slot;
267 union kvm_mmu_notifier_arg arg;
270 bool kvm_unmap_gfn_range(struct kvm *kvm, struct kvm_gfn_range *range);
271 bool kvm_age_gfn(struct kvm *kvm, struct kvm_gfn_range *range);
272 bool kvm_test_age_gfn(struct kvm *kvm, struct kvm_gfn_range *range);
279 READING_SHADOW_PAGE_TABLES,
282 #define KVM_UNMAPPED_PAGE ((void *) 0x500 + POISON_POINTER_DELTA)
284 struct kvm_host_map {
286 * Only valid if the 'pfn' is managed by the host kernel (i.e. There is
287 * a 'struct page' for it. When using mem= kernel parameter some memory
288 * can be used as guest memory but they are not managed by host
290 * If 'pfn' is not managed by the host kernel, this field is
291 * initialized to KVM_UNMAPPED_PAGE.
300 * Used to check if the mapping is valid or not. Never use 'kvm_host_map'
301 * directly to check for that.
303 static inline bool kvm_vcpu_mapped(struct kvm_host_map *map)
308 static inline bool kvm_vcpu_can_poll(ktime_t cur, ktime_t stop)
310 return single_task_running() && !need_resched() && ktime_before(cur, stop);
314 * Sometimes a large or cross-page mmio needs to be broken up into separate
315 * exits for userspace servicing.
317 struct kvm_mmio_fragment {
325 #ifdef CONFIG_PREEMPT_NOTIFIERS
326 struct preempt_notifier preempt_notifier;
329 int vcpu_id; /* id given by userspace at creation */
330 int vcpu_idx; /* index into kvm->vcpu_array */
331 int ____srcu_idx; /* Don't use this directly. You've been warned. */
332 #ifdef CONFIG_PROVE_RCU
337 unsigned long guest_debug;
342 #ifndef __KVM_HAVE_ARCH_WQP
345 struct pid __rcu *pid;
348 unsigned int halt_poll_ns;
351 #ifdef CONFIG_HAS_IOMEM
353 int mmio_read_completed;
355 int mmio_cur_fragment;
356 int mmio_nr_fragments;
357 struct kvm_mmio_fragment mmio_fragments[KVM_MAX_MMIO_FRAGMENTS];
360 #ifdef CONFIG_KVM_ASYNC_PF
363 struct list_head queue;
364 struct list_head done;
369 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
371 * Cpu relax intercept or pause loop exit optimization
372 * in_spin_loop: set when a vcpu does a pause loop exit
373 * or cpu relax intercepted.
374 * dy_eligible: indicates whether vcpu is eligible for directed yield.
383 struct kvm_vcpu_arch arch;
384 struct kvm_vcpu_stat stat;
385 char stats_id[KVM_STATS_NAME_SIZE];
386 struct kvm_dirty_ring dirty_ring;
389 * The most recently used memslot by this vCPU and the slots generation
390 * for which it is valid.
391 * No wraparound protection is needed since generations won't overflow in
392 * thousands of years, even assuming 1M memslot operations per second.
394 struct kvm_memory_slot *last_used_slot;
395 u64 last_used_slot_gen;
399 * Start accounting time towards a guest.
400 * Must be called before entering guest context.
402 static __always_inline void guest_timing_enter_irqoff(void)
405 * This is running in ioctl context so its safe to assume that it's the
406 * stime pending cputime to flush.
408 instrumentation_begin();
409 vtime_account_guest_enter();
410 instrumentation_end();
414 * Enter guest context and enter an RCU extended quiescent state.
416 * Between guest_context_enter_irqoff() and guest_context_exit_irqoff() it is
417 * unsafe to use any code which may directly or indirectly use RCU, tracing
418 * (including IRQ flag tracing), or lockdep. All code in this period must be
419 * non-instrumentable.
421 static __always_inline void guest_context_enter_irqoff(void)
424 * KVM does not hold any references to rcu protected data when it
425 * switches CPU into a guest mode. In fact switching to a guest mode
426 * is very similar to exiting to userspace from rcu point of view. In
427 * addition CPU may stay in a guest mode for quite a long time (up to
428 * one time slice). Lets treat guest mode as quiescent state, just like
429 * we do with user-mode execution.
431 if (!context_tracking_guest_enter()) {
432 instrumentation_begin();
433 rcu_virt_note_context_switch();
434 instrumentation_end();
439 * Deprecated. Architectures should move to guest_timing_enter_irqoff() and
440 * guest_state_enter_irqoff().
442 static __always_inline void guest_enter_irqoff(void)
444 guest_timing_enter_irqoff();
445 guest_context_enter_irqoff();
449 * guest_state_enter_irqoff - Fixup state when entering a guest
451 * Entry to a guest will enable interrupts, but the kernel state is interrupts
452 * disabled when this is invoked. Also tell RCU about it.
454 * 1) Trace interrupts on state
455 * 2) Invoke context tracking if enabled to adjust RCU state
456 * 3) Tell lockdep that interrupts are enabled
458 * Invoked from architecture specific code before entering a guest.
459 * Must be called with interrupts disabled and the caller must be
460 * non-instrumentable.
461 * The caller has to invoke guest_timing_enter_irqoff() before this.
463 * Note: this is analogous to exit_to_user_mode().
465 static __always_inline void guest_state_enter_irqoff(void)
467 instrumentation_begin();
468 trace_hardirqs_on_prepare();
469 lockdep_hardirqs_on_prepare();
470 instrumentation_end();
472 guest_context_enter_irqoff();
473 lockdep_hardirqs_on(CALLER_ADDR0);
477 * Exit guest context and exit an RCU extended quiescent state.
479 * Between guest_context_enter_irqoff() and guest_context_exit_irqoff() it is
480 * unsafe to use any code which may directly or indirectly use RCU, tracing
481 * (including IRQ flag tracing), or lockdep. All code in this period must be
482 * non-instrumentable.
484 static __always_inline void guest_context_exit_irqoff(void)
486 context_tracking_guest_exit();
490 * Stop accounting time towards a guest.
491 * Must be called after exiting guest context.
493 static __always_inline void guest_timing_exit_irqoff(void)
495 instrumentation_begin();
496 /* Flush the guest cputime we spent on the guest */
497 vtime_account_guest_exit();
498 instrumentation_end();
502 * Deprecated. Architectures should move to guest_state_exit_irqoff() and
503 * guest_timing_exit_irqoff().
505 static __always_inline void guest_exit_irqoff(void)
507 guest_context_exit_irqoff();
508 guest_timing_exit_irqoff();
511 static inline void guest_exit(void)
515 local_irq_save(flags);
517 local_irq_restore(flags);
521 * guest_state_exit_irqoff - Establish state when returning from guest mode
523 * Entry from a guest disables interrupts, but guest mode is traced as
524 * interrupts enabled. Also with NO_HZ_FULL RCU might be idle.
526 * 1) Tell lockdep that interrupts are disabled
527 * 2) Invoke context tracking if enabled to reactivate RCU
528 * 3) Trace interrupts off state
530 * Invoked from architecture specific code after exiting a guest.
531 * Must be invoked with interrupts disabled and the caller must be
532 * non-instrumentable.
533 * The caller has to invoke guest_timing_exit_irqoff() after this.
535 * Note: this is analogous to enter_from_user_mode().
537 static __always_inline void guest_state_exit_irqoff(void)
539 lockdep_hardirqs_off(CALLER_ADDR0);
540 guest_context_exit_irqoff();
542 instrumentation_begin();
543 trace_hardirqs_off_finish();
544 instrumentation_end();
547 static inline int kvm_vcpu_exiting_guest_mode(struct kvm_vcpu *vcpu)
550 * The memory barrier ensures a previous write to vcpu->requests cannot
551 * be reordered with the read of vcpu->mode. It pairs with the general
552 * memory barrier following the write of vcpu->mode in VCPU RUN.
554 smp_mb__before_atomic();
555 return cmpxchg(&vcpu->mode, IN_GUEST_MODE, EXITING_GUEST_MODE);
559 * Some of the bitops functions do not support too long bitmaps.
560 * This number must be determined not to exceed such limits.
562 #define KVM_MEM_MAX_NR_PAGES ((1UL << 31) - 1)
565 * Since at idle each memslot belongs to two memslot sets it has to contain
566 * two embedded nodes for each data structure that it forms a part of.
568 * Two memslot sets (one active and one inactive) are necessary so the VM
569 * continues to run on one memslot set while the other is being modified.
571 * These two memslot sets normally point to the same set of memslots.
572 * They can, however, be desynchronized when performing a memslot management
573 * operation by replacing the memslot to be modified by its copy.
574 * After the operation is complete, both memslot sets once again point to
575 * the same, common set of memslot data.
577 * The memslots themselves are independent of each other so they can be
578 * individually added or deleted.
580 struct kvm_memory_slot {
581 struct hlist_node id_node[2];
582 struct interval_tree_node hva_node[2];
583 struct rb_node gfn_node[2];
585 unsigned long npages;
586 unsigned long *dirty_bitmap;
587 struct kvm_arch_memory_slot arch;
588 unsigned long userspace_addr;
593 #ifdef CONFIG_KVM_PRIVATE_MEM
595 struct file __rcu *file;
601 static inline bool kvm_slot_can_be_private(const struct kvm_memory_slot *slot)
603 return slot && (slot->flags & KVM_MEM_GUEST_MEMFD);
606 static inline bool kvm_slot_dirty_track_enabled(const struct kvm_memory_slot *slot)
608 return slot->flags & KVM_MEM_LOG_DIRTY_PAGES;
611 static inline unsigned long kvm_dirty_bitmap_bytes(struct kvm_memory_slot *memslot)
613 return ALIGN(memslot->npages, BITS_PER_LONG) / 8;
616 static inline unsigned long *kvm_second_dirty_bitmap(struct kvm_memory_slot *memslot)
618 unsigned long len = kvm_dirty_bitmap_bytes(memslot);
620 return memslot->dirty_bitmap + len / sizeof(*memslot->dirty_bitmap);
623 #ifndef KVM_DIRTY_LOG_MANUAL_CAPS
624 #define KVM_DIRTY_LOG_MANUAL_CAPS KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE
627 struct kvm_s390_adapter_int {
640 struct kvm_xen_evtchn {
647 struct kvm_kernel_irq_routing_entry {
650 int (*set)(struct kvm_kernel_irq_routing_entry *e,
651 struct kvm *kvm, int irq_source_id, int level,
665 struct kvm_s390_adapter_int adapter;
666 struct kvm_hv_sint hv_sint;
667 struct kvm_xen_evtchn xen_evtchn;
669 struct hlist_node link;
672 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
673 struct kvm_irq_routing_table {
674 int chip[KVM_NR_IRQCHIPS][KVM_IRQCHIP_NUM_PINS];
677 * Array indexed by gsi. Each entry contains list of irq chips
678 * the gsi is connected to.
680 struct hlist_head map[] __counted_by(nr_rt_entries);
684 bool kvm_arch_irqchip_in_kernel(struct kvm *kvm);
686 #ifndef KVM_INTERNAL_MEM_SLOTS
687 #define KVM_INTERNAL_MEM_SLOTS 0
690 #define KVM_MEM_SLOTS_NUM SHRT_MAX
691 #define KVM_USER_MEM_SLOTS (KVM_MEM_SLOTS_NUM - KVM_INTERNAL_MEM_SLOTS)
693 #if KVM_MAX_NR_ADDRESS_SPACES == 1
694 static inline int kvm_arch_nr_memslot_as_ids(struct kvm *kvm)
696 return KVM_MAX_NR_ADDRESS_SPACES;
699 static inline int kvm_arch_vcpu_memslots_id(struct kvm_vcpu *vcpu)
706 * Arch code must define kvm_arch_has_private_mem if support for private memory
709 #if !defined(kvm_arch_has_private_mem) && !IS_ENABLED(CONFIG_KVM_PRIVATE_MEM)
710 static inline bool kvm_arch_has_private_mem(struct kvm *kvm)
716 struct kvm_memslots {
718 atomic_long_t last_used_slot;
719 struct rb_root_cached hva_tree;
720 struct rb_root gfn_tree;
722 * The mapping table from slot id to memslot.
724 * 7-bit bucket count matches the size of the old id to index array for
725 * 512 slots, while giving good performance with this slot count.
726 * Higher bucket counts bring only small performance improvements but
727 * always result in higher memory usage (even for lower memslot counts).
729 DECLARE_HASHTABLE(id_hash, 7);
734 #ifdef KVM_HAVE_MMU_RWLOCK
738 #endif /* KVM_HAVE_MMU_RWLOCK */
740 struct mutex slots_lock;
743 * Protects the arch-specific fields of struct kvm_memory_slots in
744 * use by the VM. To be used under the slots_lock (above) or in a
745 * kvm->srcu critical section where acquiring the slots_lock would
746 * lead to deadlock with the synchronize_srcu in
747 * kvm_swap_active_memslots().
749 struct mutex slots_arch_lock;
750 struct mm_struct *mm; /* userspace tied to this vm */
751 unsigned long nr_memslot_pages;
752 /* The two memslot sets - active and inactive (per address space) */
753 struct kvm_memslots __memslots[KVM_MAX_NR_ADDRESS_SPACES][2];
754 /* The current active memslot set for each address space */
755 struct kvm_memslots __rcu *memslots[KVM_MAX_NR_ADDRESS_SPACES];
756 struct xarray vcpu_array;
758 * Protected by slots_lock, but can be read outside if an
759 * incorrect answer is acceptable.
761 atomic_t nr_memslots_dirty_logging;
763 /* Used to wait for completion of MMU notifiers. */
764 spinlock_t mn_invalidate_lock;
765 unsigned long mn_active_invalidate_count;
766 struct rcuwait mn_memslots_update_rcuwait;
768 /* For management / invalidation of gfn_to_pfn_caches */
770 struct list_head gpc_list;
773 * created_vcpus is protected by kvm->lock, and is incremented
774 * at the beginning of KVM_CREATE_VCPU. online_vcpus is only
775 * incremented after storing the kvm_vcpu pointer in vcpus,
776 * and is accessed atomically.
778 atomic_t online_vcpus;
781 int last_boosted_vcpu;
782 struct list_head vm_list;
784 struct kvm_io_bus __rcu *buses[KVM_NR_BUSES];
785 #ifdef CONFIG_HAVE_KVM_IRQCHIP
788 struct list_head items;
789 /* resampler_list update side is protected by resampler_lock. */
790 struct list_head resampler_list;
791 struct mutex resampler_lock;
794 struct list_head ioeventfds;
795 struct kvm_vm_stat stat;
796 struct kvm_arch arch;
797 refcount_t users_count;
798 #ifdef CONFIG_KVM_MMIO
799 struct kvm_coalesced_mmio_ring *coalesced_mmio_ring;
800 spinlock_t ring_lock;
801 struct list_head coalesced_zones;
804 struct mutex irq_lock;
805 #ifdef CONFIG_HAVE_KVM_IRQCHIP
807 * Update side is protected by irq_lock.
809 struct kvm_irq_routing_table __rcu *irq_routing;
811 struct hlist_head irq_ack_notifier_list;
814 #ifdef CONFIG_KVM_GENERIC_MMU_NOTIFIER
815 struct mmu_notifier mmu_notifier;
816 unsigned long mmu_invalidate_seq;
817 long mmu_invalidate_in_progress;
818 gfn_t mmu_invalidate_range_start;
819 gfn_t mmu_invalidate_range_end;
821 struct list_head devices;
822 u64 manual_dirty_log_protect;
823 struct dentry *debugfs_dentry;
824 struct kvm_stat_data **debugfs_stat_data;
825 struct srcu_struct srcu;
826 struct srcu_struct irq_srcu;
828 bool override_halt_poll_ns;
829 unsigned int max_halt_poll_ns;
831 bool dirty_ring_with_bitmap;
835 #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER
836 struct notifier_block pm_notifier;
838 #ifdef CONFIG_KVM_GENERIC_MEMORY_ATTRIBUTES
839 /* Protected by slots_locks (for writes) and RCU (for reads) */
840 struct xarray mem_attr_array;
842 char stats_id[KVM_STATS_NAME_SIZE];
845 #define kvm_err(fmt, ...) \
846 pr_err("kvm [%i]: " fmt, task_pid_nr(current), ## __VA_ARGS__)
847 #define kvm_info(fmt, ...) \
848 pr_info("kvm [%i]: " fmt, task_pid_nr(current), ## __VA_ARGS__)
849 #define kvm_debug(fmt, ...) \
850 pr_debug("kvm [%i]: " fmt, task_pid_nr(current), ## __VA_ARGS__)
851 #define kvm_debug_ratelimited(fmt, ...) \
852 pr_debug_ratelimited("kvm [%i]: " fmt, task_pid_nr(current), \
854 #define kvm_pr_unimpl(fmt, ...) \
855 pr_err_ratelimited("kvm [%i]: " fmt, \
856 task_tgid_nr(current), ## __VA_ARGS__)
858 /* The guest did something we don't support. */
859 #define vcpu_unimpl(vcpu, fmt, ...) \
860 kvm_pr_unimpl("vcpu%i, guest rIP: 0x%lx " fmt, \
861 (vcpu)->vcpu_id, kvm_rip_read(vcpu), ## __VA_ARGS__)
863 #define vcpu_debug(vcpu, fmt, ...) \
864 kvm_debug("vcpu%i " fmt, (vcpu)->vcpu_id, ## __VA_ARGS__)
865 #define vcpu_debug_ratelimited(vcpu, fmt, ...) \
866 kvm_debug_ratelimited("vcpu%i " fmt, (vcpu)->vcpu_id, \
868 #define vcpu_err(vcpu, fmt, ...) \
869 kvm_err("vcpu%i " fmt, (vcpu)->vcpu_id, ## __VA_ARGS__)
871 static inline void kvm_vm_dead(struct kvm *kvm)
874 kvm_make_all_cpus_request(kvm, KVM_REQ_VM_DEAD);
877 static inline void kvm_vm_bugged(struct kvm *kvm)
879 kvm->vm_bugged = true;
884 #define KVM_BUG(cond, kvm, fmt...) \
886 bool __ret = !!(cond); \
888 if (WARN_ONCE(__ret && !(kvm)->vm_bugged, fmt)) \
889 kvm_vm_bugged(kvm); \
893 #define KVM_BUG_ON(cond, kvm) \
895 bool __ret = !!(cond); \
897 if (WARN_ON_ONCE(__ret && !(kvm)->vm_bugged)) \
898 kvm_vm_bugged(kvm); \
903 * Note, "data corruption" refers to corruption of host kernel data structures,
904 * not guest data. Guest data corruption, suspected or confirmed, that is tied
905 * and contained to a single VM should *never* BUG() and potentially panic the
906 * host, i.e. use this variant of KVM_BUG() if and only if a KVM data structure
907 * is corrupted and that corruption can have a cascading effect to other parts
908 * of the hosts and/or to other VMs.
910 #define KVM_BUG_ON_DATA_CORRUPTION(cond, kvm) \
912 bool __ret = !!(cond); \
914 if (IS_ENABLED(CONFIG_BUG_ON_DATA_CORRUPTION)) \
916 else if (WARN_ON_ONCE(__ret && !(kvm)->vm_bugged)) \
917 kvm_vm_bugged(kvm); \
921 static inline void kvm_vcpu_srcu_read_lock(struct kvm_vcpu *vcpu)
923 #ifdef CONFIG_PROVE_RCU
924 WARN_ONCE(vcpu->srcu_depth++,
925 "KVM: Illegal vCPU srcu_idx LOCK, depth=%d", vcpu->srcu_depth - 1);
927 vcpu->____srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
930 static inline void kvm_vcpu_srcu_read_unlock(struct kvm_vcpu *vcpu)
932 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->____srcu_idx);
934 #ifdef CONFIG_PROVE_RCU
935 WARN_ONCE(--vcpu->srcu_depth,
936 "KVM: Illegal vCPU srcu_idx UNLOCK, depth=%d", vcpu->srcu_depth);
940 static inline bool kvm_dirty_log_manual_protect_and_init_set(struct kvm *kvm)
942 return !!(kvm->manual_dirty_log_protect & KVM_DIRTY_LOG_INITIALLY_SET);
945 static inline struct kvm_io_bus *kvm_get_bus(struct kvm *kvm, enum kvm_bus idx)
947 return srcu_dereference_check(kvm->buses[idx], &kvm->srcu,
948 lockdep_is_held(&kvm->slots_lock) ||
949 !refcount_read(&kvm->users_count));
952 static inline struct kvm_vcpu *kvm_get_vcpu(struct kvm *kvm, int i)
954 int num_vcpus = atomic_read(&kvm->online_vcpus);
955 i = array_index_nospec(i, num_vcpus);
957 /* Pairs with smp_wmb() in kvm_vm_ioctl_create_vcpu. */
959 return xa_load(&kvm->vcpu_array, i);
962 #define kvm_for_each_vcpu(idx, vcpup, kvm) \
963 xa_for_each_range(&kvm->vcpu_array, idx, vcpup, 0, \
964 (atomic_read(&kvm->online_vcpus) - 1))
966 static inline struct kvm_vcpu *kvm_get_vcpu_by_id(struct kvm *kvm, int id)
968 struct kvm_vcpu *vcpu = NULL;
973 if (id < KVM_MAX_VCPUS)
974 vcpu = kvm_get_vcpu(kvm, id);
975 if (vcpu && vcpu->vcpu_id == id)
977 kvm_for_each_vcpu(i, vcpu, kvm)
978 if (vcpu->vcpu_id == id)
983 void kvm_destroy_vcpus(struct kvm *kvm);
985 void vcpu_load(struct kvm_vcpu *vcpu);
986 void vcpu_put(struct kvm_vcpu *vcpu);
988 #ifdef __KVM_HAVE_IOAPIC
989 void kvm_arch_post_irq_ack_notifier_list_update(struct kvm *kvm);
990 void kvm_arch_post_irq_routing_update(struct kvm *kvm);
992 static inline void kvm_arch_post_irq_ack_notifier_list_update(struct kvm *kvm)
995 static inline void kvm_arch_post_irq_routing_update(struct kvm *kvm)
1000 #ifdef CONFIG_HAVE_KVM_IRQCHIP
1001 int kvm_irqfd_init(void);
1002 void kvm_irqfd_exit(void);
1004 static inline int kvm_irqfd_init(void)
1009 static inline void kvm_irqfd_exit(void)
1013 int kvm_init(unsigned vcpu_size, unsigned vcpu_align, struct module *module);
1014 void kvm_exit(void);
1016 void kvm_get_kvm(struct kvm *kvm);
1017 bool kvm_get_kvm_safe(struct kvm *kvm);
1018 void kvm_put_kvm(struct kvm *kvm);
1019 bool file_is_kvm(struct file *file);
1020 void kvm_put_kvm_no_destroy(struct kvm *kvm);
1022 static inline struct kvm_memslots *__kvm_memslots(struct kvm *kvm, int as_id)
1024 as_id = array_index_nospec(as_id, KVM_MAX_NR_ADDRESS_SPACES);
1025 return srcu_dereference_check(kvm->memslots[as_id], &kvm->srcu,
1026 lockdep_is_held(&kvm->slots_lock) ||
1027 !refcount_read(&kvm->users_count));
1030 static inline struct kvm_memslots *kvm_memslots(struct kvm *kvm)
1032 return __kvm_memslots(kvm, 0);
1035 static inline struct kvm_memslots *kvm_vcpu_memslots(struct kvm_vcpu *vcpu)
1037 int as_id = kvm_arch_vcpu_memslots_id(vcpu);
1039 return __kvm_memslots(vcpu->kvm, as_id);
1042 static inline bool kvm_memslots_empty(struct kvm_memslots *slots)
1044 return RB_EMPTY_ROOT(&slots->gfn_tree);
1047 bool kvm_are_all_memslots_empty(struct kvm *kvm);
1049 #define kvm_for_each_memslot(memslot, bkt, slots) \
1050 hash_for_each(slots->id_hash, bkt, memslot, id_node[slots->node_idx]) \
1051 if (WARN_ON_ONCE(!memslot->npages)) { \
1055 struct kvm_memory_slot *id_to_memslot(struct kvm_memslots *slots, int id)
1057 struct kvm_memory_slot *slot;
1058 int idx = slots->node_idx;
1060 hash_for_each_possible(slots->id_hash, slot, id_node[idx], id) {
1068 /* Iterator used for walking memslots that overlap a gfn range. */
1069 struct kvm_memslot_iter {
1070 struct kvm_memslots *slots;
1071 struct rb_node *node;
1072 struct kvm_memory_slot *slot;
1075 static inline void kvm_memslot_iter_next(struct kvm_memslot_iter *iter)
1077 iter->node = rb_next(iter->node);
1081 iter->slot = container_of(iter->node, struct kvm_memory_slot, gfn_node[iter->slots->node_idx]);
1084 static inline void kvm_memslot_iter_start(struct kvm_memslot_iter *iter,
1085 struct kvm_memslots *slots,
1088 int idx = slots->node_idx;
1089 struct rb_node *tmp;
1090 struct kvm_memory_slot *slot;
1092 iter->slots = slots;
1095 * Find the so called "upper bound" of a key - the first node that has
1096 * its key strictly greater than the searched one (the start gfn in our case).
1099 for (tmp = slots->gfn_tree.rb_node; tmp; ) {
1100 slot = container_of(tmp, struct kvm_memory_slot, gfn_node[idx]);
1101 if (start < slot->base_gfn) {
1105 tmp = tmp->rb_right;
1110 * Find the slot with the lowest gfn that can possibly intersect with
1111 * the range, so we'll ideally have slot start <= range start
1115 * A NULL previous node means that the very first slot
1116 * already has a higher start gfn.
1117 * In this case slot start > range start.
1119 tmp = rb_prev(iter->node);
1123 /* a NULL node below means no slots */
1124 iter->node = rb_last(&slots->gfn_tree);
1128 iter->slot = container_of(iter->node, struct kvm_memory_slot, gfn_node[idx]);
1131 * It is possible in the slot start < range start case that the
1132 * found slot ends before or at range start (slot end <= range start)
1133 * and so it does not overlap the requested range.
1135 * In such non-overlapping case the next slot (if it exists) will
1136 * already have slot start > range start, otherwise the logic above
1137 * would have found it instead of the current slot.
1139 if (iter->slot->base_gfn + iter->slot->npages <= start)
1140 kvm_memslot_iter_next(iter);
1144 static inline bool kvm_memslot_iter_is_valid(struct kvm_memslot_iter *iter, gfn_t end)
1150 * If this slot starts beyond or at the end of the range so does
1153 return iter->slot->base_gfn < end;
1156 /* Iterate over each memslot at least partially intersecting [start, end) range */
1157 #define kvm_for_each_memslot_in_gfn_range(iter, slots, start, end) \
1158 for (kvm_memslot_iter_start(iter, slots, start); \
1159 kvm_memslot_iter_is_valid(iter, end); \
1160 kvm_memslot_iter_next(iter))
1163 * KVM_SET_USER_MEMORY_REGION ioctl allows the following operations:
1164 * - create a new memory slot
1165 * - delete an existing memory slot
1166 * - modify an existing memory slot
1167 * -- move it in the guest physical memory space
1168 * -- just change its flags
1170 * Since flags can be changed by some of these operations, the following
1171 * differentiation is the best we can do for __kvm_set_memory_region():
1173 enum kvm_mr_change {
1180 int kvm_set_memory_region(struct kvm *kvm,
1181 const struct kvm_userspace_memory_region2 *mem);
1182 int __kvm_set_memory_region(struct kvm *kvm,
1183 const struct kvm_userspace_memory_region2 *mem);
1184 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot);
1185 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen);
1186 int kvm_arch_prepare_memory_region(struct kvm *kvm,
1187 const struct kvm_memory_slot *old,
1188 struct kvm_memory_slot *new,
1189 enum kvm_mr_change change);
1190 void kvm_arch_commit_memory_region(struct kvm *kvm,
1191 struct kvm_memory_slot *old,
1192 const struct kvm_memory_slot *new,
1193 enum kvm_mr_change change);
1194 /* flush all memory translations */
1195 void kvm_arch_flush_shadow_all(struct kvm *kvm);
1196 /* flush memory translations pointing to 'slot' */
1197 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
1198 struct kvm_memory_slot *slot);
1200 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1201 struct page **pages, int nr_pages);
1203 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn);
1204 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn);
1205 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable);
1206 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot, gfn_t gfn);
1207 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot, gfn_t gfn,
1209 void kvm_release_page_clean(struct page *page);
1210 void kvm_release_page_dirty(struct page *page);
1212 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn);
1213 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1215 kvm_pfn_t gfn_to_pfn_memslot(const struct kvm_memory_slot *slot, gfn_t gfn);
1216 kvm_pfn_t gfn_to_pfn_memslot_atomic(const struct kvm_memory_slot *slot, gfn_t gfn);
1217 kvm_pfn_t __gfn_to_pfn_memslot(const struct kvm_memory_slot *slot, gfn_t gfn,
1218 bool atomic, bool interruptible, bool *async,
1219 bool write_fault, bool *writable, hva_t *hva);
1221 void kvm_release_pfn_clean(kvm_pfn_t pfn);
1222 void kvm_release_pfn_dirty(kvm_pfn_t pfn);
1223 void kvm_set_pfn_dirty(kvm_pfn_t pfn);
1224 void kvm_set_pfn_accessed(kvm_pfn_t pfn);
1226 void kvm_release_pfn(kvm_pfn_t pfn, bool dirty);
1227 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1229 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len);
1230 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1231 void *data, unsigned long len);
1232 int kvm_read_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1233 void *data, unsigned int offset,
1235 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1236 int offset, int len);
1237 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1239 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1240 void *data, unsigned long len);
1241 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1242 void *data, unsigned int offset,
1244 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1245 gpa_t gpa, unsigned long len);
1247 #define __kvm_get_guest(kvm, gfn, offset, v) \
1249 unsigned long __addr = gfn_to_hva(kvm, gfn); \
1250 typeof(v) __user *__uaddr = (typeof(__uaddr))(__addr + offset); \
1251 int __ret = -EFAULT; \
1253 if (!kvm_is_error_hva(__addr)) \
1254 __ret = get_user(v, __uaddr); \
1258 #define kvm_get_guest(kvm, gpa, v) \
1260 gpa_t __gpa = gpa; \
1261 struct kvm *__kvm = kvm; \
1263 __kvm_get_guest(__kvm, __gpa >> PAGE_SHIFT, \
1264 offset_in_page(__gpa), v); \
1267 #define __kvm_put_guest(kvm, gfn, offset, v) \
1269 unsigned long __addr = gfn_to_hva(kvm, gfn); \
1270 typeof(v) __user *__uaddr = (typeof(__uaddr))(__addr + offset); \
1271 int __ret = -EFAULT; \
1273 if (!kvm_is_error_hva(__addr)) \
1274 __ret = put_user(v, __uaddr); \
1276 mark_page_dirty(kvm, gfn); \
1280 #define kvm_put_guest(kvm, gpa, v) \
1282 gpa_t __gpa = gpa; \
1283 struct kvm *__kvm = kvm; \
1285 __kvm_put_guest(__kvm, __gpa >> PAGE_SHIFT, \
1286 offset_in_page(__gpa), v); \
1289 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len);
1290 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn);
1291 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn);
1292 bool kvm_vcpu_is_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn);
1293 unsigned long kvm_host_page_size(struct kvm_vcpu *vcpu, gfn_t gfn);
1294 void mark_page_dirty_in_slot(struct kvm *kvm, const struct kvm_memory_slot *memslot, gfn_t gfn);
1295 void mark_page_dirty(struct kvm *kvm, gfn_t gfn);
1297 struct kvm_memslots *kvm_vcpu_memslots(struct kvm_vcpu *vcpu);
1298 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn);
1299 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn);
1300 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn);
1301 int kvm_vcpu_map(struct kvm_vcpu *vcpu, gpa_t gpa, struct kvm_host_map *map);
1302 void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map, bool dirty);
1303 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn);
1304 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable);
1305 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data, int offset,
1307 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa, void *data,
1309 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data,
1311 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, const void *data,
1312 int offset, int len);
1313 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
1315 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn);
1318 * kvm_gpc_init - initialize gfn_to_pfn_cache.
1320 * @gpc: struct gfn_to_pfn_cache object.
1321 * @kvm: pointer to kvm instance.
1323 * This sets up a gfn_to_pfn_cache by initializing locks and assigning the
1324 * immutable attributes. Note, the cache must be zero-allocated (or zeroed by
1325 * the caller before init).
1327 void kvm_gpc_init(struct gfn_to_pfn_cache *gpc, struct kvm *kvm);
1330 * kvm_gpc_activate - prepare a cached kernel mapping and HPA for a given guest
1333 * @gpc: struct gfn_to_pfn_cache object.
1334 * @gpa: guest physical address to map.
1335 * @len: sanity check; the range being access must fit a single page.
1337 * @return: 0 for success.
1338 * -EINVAL for a mapping which would cross a page boundary.
1339 * -EFAULT for an untranslatable guest physical address.
1341 * This primes a gfn_to_pfn_cache and links it into the @gpc->kvm's list for
1342 * invalidations to be processed. Callers are required to use kvm_gpc_check()
1343 * to ensure that the cache is valid before accessing the target page.
1345 int kvm_gpc_activate(struct gfn_to_pfn_cache *gpc, gpa_t gpa, unsigned long len);
1348 * kvm_gpc_activate_hva - prepare a cached kernel mapping and HPA for a given HVA.
1350 * @gpc: struct gfn_to_pfn_cache object.
1351 * @hva: userspace virtual address to map.
1352 * @len: sanity check; the range being access must fit a single page.
1354 * @return: 0 for success.
1355 * -EINVAL for a mapping which would cross a page boundary.
1356 * -EFAULT for an untranslatable guest physical address.
1358 * The semantics of this function are the same as those of kvm_gpc_activate(). It
1359 * merely bypasses a layer of address translation.
1361 int kvm_gpc_activate_hva(struct gfn_to_pfn_cache *gpc, unsigned long hva, unsigned long len);
1364 * kvm_gpc_check - check validity of a gfn_to_pfn_cache.
1366 * @gpc: struct gfn_to_pfn_cache object.
1367 * @len: sanity check; the range being access must fit a single page.
1369 * @return: %true if the cache is still valid and the address matches.
1370 * %false if the cache is not valid.
1372 * Callers outside IN_GUEST_MODE context should hold a read lock on @gpc->lock
1373 * while calling this function, and then continue to hold the lock until the
1374 * access is complete.
1376 * Callers in IN_GUEST_MODE may do so without locking, although they should
1377 * still hold a read lock on kvm->scru for the memslot checks.
1379 bool kvm_gpc_check(struct gfn_to_pfn_cache *gpc, unsigned long len);
1382 * kvm_gpc_refresh - update a previously initialized cache.
1384 * @gpc: struct gfn_to_pfn_cache object.
1385 * @len: sanity check; the range being access must fit a single page.
1387 * @return: 0 for success.
1388 * -EINVAL for a mapping which would cross a page boundary.
1389 * -EFAULT for an untranslatable guest physical address.
1391 * This will attempt to refresh a gfn_to_pfn_cache. Note that a successful
1392 * return from this function does not mean the page can be immediately
1393 * accessed because it may have raced with an invalidation. Callers must
1394 * still lock and check the cache status, as this function does not return
1395 * with the lock still held to permit access.
1397 int kvm_gpc_refresh(struct gfn_to_pfn_cache *gpc, unsigned long len);
1400 * kvm_gpc_deactivate - deactivate and unlink a gfn_to_pfn_cache.
1402 * @gpc: struct gfn_to_pfn_cache object.
1404 * This removes a cache from the VM's list to be processed on MMU notifier
1407 void kvm_gpc_deactivate(struct gfn_to_pfn_cache *gpc);
1409 static inline bool kvm_gpc_is_gpa_active(struct gfn_to_pfn_cache *gpc)
1411 return gpc->active && !kvm_is_error_gpa(gpc->gpa);
1414 static inline bool kvm_gpc_is_hva_active(struct gfn_to_pfn_cache *gpc)
1416 return gpc->active && kvm_is_error_gpa(gpc->gpa);
1419 void kvm_sigset_activate(struct kvm_vcpu *vcpu);
1420 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu);
1422 void kvm_vcpu_halt(struct kvm_vcpu *vcpu);
1423 bool kvm_vcpu_block(struct kvm_vcpu *vcpu);
1424 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu);
1425 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu);
1426 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu);
1427 void kvm_vcpu_kick(struct kvm_vcpu *vcpu);
1428 int kvm_vcpu_yield_to(struct kvm_vcpu *target);
1429 void kvm_vcpu_on_spin(struct kvm_vcpu *vcpu, bool yield_to_kernel_mode);
1431 void kvm_flush_remote_tlbs(struct kvm *kvm);
1432 void kvm_flush_remote_tlbs_range(struct kvm *kvm, gfn_t gfn, u64 nr_pages);
1433 void kvm_flush_remote_tlbs_memslot(struct kvm *kvm,
1434 const struct kvm_memory_slot *memslot);
1436 #ifdef KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE
1437 int kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache *mc, int min);
1438 int __kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache *mc, int capacity, int min);
1439 int kvm_mmu_memory_cache_nr_free_objects(struct kvm_mmu_memory_cache *mc);
1440 void kvm_mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc);
1441 void *kvm_mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc);
1444 void kvm_mmu_invalidate_begin(struct kvm *kvm);
1445 void kvm_mmu_invalidate_range_add(struct kvm *kvm, gfn_t start, gfn_t end);
1446 void kvm_mmu_invalidate_end(struct kvm *kvm);
1447 bool kvm_mmu_unmap_gfn_range(struct kvm *kvm, struct kvm_gfn_range *range);
1449 long kvm_arch_dev_ioctl(struct file *filp,
1450 unsigned int ioctl, unsigned long arg);
1451 long kvm_arch_vcpu_ioctl(struct file *filp,
1452 unsigned int ioctl, unsigned long arg);
1453 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf);
1455 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext);
1457 void kvm_arch_mmu_enable_log_dirty_pt_masked(struct kvm *kvm,
1458 struct kvm_memory_slot *slot,
1460 unsigned long mask);
1461 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot);
1463 #ifndef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1464 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log);
1465 int kvm_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log,
1466 int *is_dirty, struct kvm_memory_slot **memslot);
1469 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
1471 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
1472 struct kvm_enable_cap *cap);
1473 int kvm_arch_vm_ioctl(struct file *filp, unsigned int ioctl, unsigned long arg);
1474 long kvm_arch_vm_compat_ioctl(struct file *filp, unsigned int ioctl,
1477 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu);
1478 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu);
1480 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
1481 struct kvm_translation *tr);
1483 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs);
1484 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs);
1485 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
1486 struct kvm_sregs *sregs);
1487 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
1488 struct kvm_sregs *sregs);
1489 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
1490 struct kvm_mp_state *mp_state);
1491 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
1492 struct kvm_mp_state *mp_state);
1493 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
1494 struct kvm_guest_debug *dbg);
1495 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu);
1497 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu);
1499 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu);
1500 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu);
1501 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id);
1502 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu);
1503 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu);
1504 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu);
1506 #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER
1507 int kvm_arch_pm_notifier(struct kvm *kvm, unsigned long state);
1510 #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS
1511 void kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu, struct dentry *debugfs_dentry);
1513 static inline void kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu) {}
1516 #ifdef CONFIG_KVM_GENERIC_HARDWARE_ENABLING
1517 int kvm_arch_hardware_enable(void);
1518 void kvm_arch_hardware_disable(void);
1520 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu);
1521 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu);
1522 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu);
1523 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu);
1524 bool kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu);
1525 bool kvm_arch_vcpu_preempted_in_kernel(struct kvm_vcpu *vcpu);
1526 int kvm_arch_post_init_vm(struct kvm *kvm);
1527 void kvm_arch_pre_destroy_vm(struct kvm *kvm);
1528 void kvm_arch_create_vm_debugfs(struct kvm *kvm);
1530 #ifndef __KVM_HAVE_ARCH_VM_ALLOC
1532 * All architectures that want to use vzalloc currently also
1533 * need their own kvm_arch_alloc_vm implementation.
1535 static inline struct kvm *kvm_arch_alloc_vm(void)
1537 return kzalloc(sizeof(struct kvm), GFP_KERNEL_ACCOUNT);
1541 static inline void __kvm_arch_free_vm(struct kvm *kvm)
1546 #ifndef __KVM_HAVE_ARCH_VM_FREE
1547 static inline void kvm_arch_free_vm(struct kvm *kvm)
1549 __kvm_arch_free_vm(kvm);
1553 #ifndef __KVM_HAVE_ARCH_FLUSH_REMOTE_TLBS
1554 static inline int kvm_arch_flush_remote_tlbs(struct kvm *kvm)
1559 int kvm_arch_flush_remote_tlbs(struct kvm *kvm);
1562 #ifndef __KVM_HAVE_ARCH_FLUSH_REMOTE_TLBS_RANGE
1563 static inline int kvm_arch_flush_remote_tlbs_range(struct kvm *kvm,
1564 gfn_t gfn, u64 nr_pages)
1569 int kvm_arch_flush_remote_tlbs_range(struct kvm *kvm, gfn_t gfn, u64 nr_pages);
1572 #ifdef __KVM_HAVE_ARCH_NONCOHERENT_DMA
1573 void kvm_arch_register_noncoherent_dma(struct kvm *kvm);
1574 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm);
1575 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm);
1577 static inline void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
1581 static inline void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
1585 static inline bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
1590 #ifdef __KVM_HAVE_ARCH_ASSIGNED_DEVICE
1591 void kvm_arch_start_assignment(struct kvm *kvm);
1592 void kvm_arch_end_assignment(struct kvm *kvm);
1593 bool kvm_arch_has_assigned_device(struct kvm *kvm);
1595 static inline void kvm_arch_start_assignment(struct kvm *kvm)
1599 static inline void kvm_arch_end_assignment(struct kvm *kvm)
1603 static __always_inline bool kvm_arch_has_assigned_device(struct kvm *kvm)
1609 static inline struct rcuwait *kvm_arch_vcpu_get_wait(struct kvm_vcpu *vcpu)
1611 #ifdef __KVM_HAVE_ARCH_WQP
1612 return vcpu->arch.waitp;
1619 * Wake a vCPU if necessary, but don't do any stats/metadata updates. Returns
1620 * true if the vCPU was blocking and was awakened, false otherwise.
1622 static inline bool __kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
1624 return !!rcuwait_wake_up(kvm_arch_vcpu_get_wait(vcpu));
1627 static inline bool kvm_vcpu_is_blocking(struct kvm_vcpu *vcpu)
1629 return rcuwait_active(kvm_arch_vcpu_get_wait(vcpu));
1632 #ifdef __KVM_HAVE_ARCH_INTC_INITIALIZED
1634 * returns true if the virtual interrupt controller is initialized and
1635 * ready to accept virtual IRQ. On some architectures the virtual interrupt
1636 * controller is dynamically instantiated and this is not always true.
1638 bool kvm_arch_intc_initialized(struct kvm *kvm);
1640 static inline bool kvm_arch_intc_initialized(struct kvm *kvm)
1646 #ifdef CONFIG_GUEST_PERF_EVENTS
1647 unsigned long kvm_arch_vcpu_get_ip(struct kvm_vcpu *vcpu);
1649 void kvm_register_perf_callbacks(unsigned int (*pt_intr_handler)(void));
1650 void kvm_unregister_perf_callbacks(void);
1652 static inline void kvm_register_perf_callbacks(void *ign) {}
1653 static inline void kvm_unregister_perf_callbacks(void) {}
1654 #endif /* CONFIG_GUEST_PERF_EVENTS */
1656 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type);
1657 void kvm_arch_destroy_vm(struct kvm *kvm);
1658 void kvm_arch_sync_events(struct kvm *kvm);
1660 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu);
1662 struct page *kvm_pfn_to_refcounted_page(kvm_pfn_t pfn);
1663 bool kvm_is_zone_device_page(struct page *page);
1665 struct kvm_irq_ack_notifier {
1666 struct hlist_node link;
1668 void (*irq_acked)(struct kvm_irq_ack_notifier *kian);
1671 int kvm_irq_map_gsi(struct kvm *kvm,
1672 struct kvm_kernel_irq_routing_entry *entries, int gsi);
1673 int kvm_irq_map_chip_pin(struct kvm *kvm, unsigned irqchip, unsigned pin);
1675 int kvm_set_irq(struct kvm *kvm, int irq_source_id, u32 irq, int level,
1677 int kvm_set_msi(struct kvm_kernel_irq_routing_entry *irq_entry, struct kvm *kvm,
1678 int irq_source_id, int level, bool line_status);
1679 int kvm_arch_set_irq_inatomic(struct kvm_kernel_irq_routing_entry *e,
1680 struct kvm *kvm, int irq_source_id,
1681 int level, bool line_status);
1682 bool kvm_irq_has_notifier(struct kvm *kvm, unsigned irqchip, unsigned pin);
1683 void kvm_notify_acked_gsi(struct kvm *kvm, int gsi);
1684 void kvm_notify_acked_irq(struct kvm *kvm, unsigned irqchip, unsigned pin);
1685 void kvm_register_irq_ack_notifier(struct kvm *kvm,
1686 struct kvm_irq_ack_notifier *kian);
1687 void kvm_unregister_irq_ack_notifier(struct kvm *kvm,
1688 struct kvm_irq_ack_notifier *kian);
1689 int kvm_request_irq_source_id(struct kvm *kvm);
1690 void kvm_free_irq_source_id(struct kvm *kvm, int irq_source_id);
1691 bool kvm_arch_irqfd_allowed(struct kvm *kvm, struct kvm_irqfd *args);
1694 * Returns a pointer to the memslot if it contains gfn.
1695 * Otherwise returns NULL.
1697 static inline struct kvm_memory_slot *
1698 try_get_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1703 if (gfn >= slot->base_gfn && gfn < slot->base_gfn + slot->npages)
1710 * Returns a pointer to the memslot that contains gfn. Otherwise returns NULL.
1712 * With "approx" set returns the memslot also when the address falls
1713 * in a hole. In that case one of the memslots bordering the hole is
1716 static inline struct kvm_memory_slot *
1717 search_memslots(struct kvm_memslots *slots, gfn_t gfn, bool approx)
1719 struct kvm_memory_slot *slot;
1720 struct rb_node *node;
1721 int idx = slots->node_idx;
1724 for (node = slots->gfn_tree.rb_node; node; ) {
1725 slot = container_of(node, struct kvm_memory_slot, gfn_node[idx]);
1726 if (gfn >= slot->base_gfn) {
1727 if (gfn < slot->base_gfn + slot->npages)
1729 node = node->rb_right;
1731 node = node->rb_left;
1734 return approx ? slot : NULL;
1737 static inline struct kvm_memory_slot *
1738 ____gfn_to_memslot(struct kvm_memslots *slots, gfn_t gfn, bool approx)
1740 struct kvm_memory_slot *slot;
1742 slot = (struct kvm_memory_slot *)atomic_long_read(&slots->last_used_slot);
1743 slot = try_get_memslot(slot, gfn);
1747 slot = search_memslots(slots, gfn, approx);
1749 atomic_long_set(&slots->last_used_slot, (unsigned long)slot);
1757 * __gfn_to_memslot() and its descendants are here to allow arch code to inline
1758 * the lookups in hot paths. gfn_to_memslot() itself isn't here as an inline
1759 * because that would bloat other code too much.
1761 static inline struct kvm_memory_slot *
1762 __gfn_to_memslot(struct kvm_memslots *slots, gfn_t gfn)
1764 return ____gfn_to_memslot(slots, gfn, false);
1767 static inline unsigned long
1768 __gfn_to_hva_memslot(const struct kvm_memory_slot *slot, gfn_t gfn)
1771 * The index was checked originally in search_memslots. To avoid
1772 * that a malicious guest builds a Spectre gadget out of e.g. page
1773 * table walks, do not let the processor speculate loads outside
1774 * the guest's registered memslots.
1776 unsigned long offset = gfn - slot->base_gfn;
1777 offset = array_index_nospec(offset, slot->npages);
1778 return slot->userspace_addr + offset * PAGE_SIZE;
1781 static inline int memslot_id(struct kvm *kvm, gfn_t gfn)
1783 return gfn_to_memslot(kvm, gfn)->id;
1787 hva_to_gfn_memslot(unsigned long hva, struct kvm_memory_slot *slot)
1789 gfn_t gfn_offset = (hva - slot->userspace_addr) >> PAGE_SHIFT;
1791 return slot->base_gfn + gfn_offset;
1794 static inline gpa_t gfn_to_gpa(gfn_t gfn)
1796 return (gpa_t)gfn << PAGE_SHIFT;
1799 static inline gfn_t gpa_to_gfn(gpa_t gpa)
1801 return (gfn_t)(gpa >> PAGE_SHIFT);
1804 static inline hpa_t pfn_to_hpa(kvm_pfn_t pfn)
1806 return (hpa_t)pfn << PAGE_SHIFT;
1809 static inline bool kvm_is_gpa_in_memslot(struct kvm *kvm, gpa_t gpa)
1811 unsigned long hva = gfn_to_hva(kvm, gpa_to_gfn(gpa));
1813 return !kvm_is_error_hva(hva);
1816 static inline void kvm_gpc_mark_dirty_in_slot(struct gfn_to_pfn_cache *gpc)
1818 lockdep_assert_held(&gpc->lock);
1823 mark_page_dirty_in_slot(gpc->kvm, gpc->memslot, gpa_to_gfn(gpc->gpa));
1826 enum kvm_stat_kind {
1831 struct kvm_stat_data {
1833 const struct _kvm_stats_desc *desc;
1834 enum kvm_stat_kind kind;
1837 struct _kvm_stats_desc {
1838 struct kvm_stats_desc desc;
1839 char name[KVM_STATS_NAME_SIZE];
1842 #define STATS_DESC_COMMON(type, unit, base, exp, sz, bsz) \
1843 .flags = type | unit | base | \
1844 BUILD_BUG_ON_ZERO(type & ~KVM_STATS_TYPE_MASK) | \
1845 BUILD_BUG_ON_ZERO(unit & ~KVM_STATS_UNIT_MASK) | \
1846 BUILD_BUG_ON_ZERO(base & ~KVM_STATS_BASE_MASK), \
1851 #define VM_GENERIC_STATS_DESC(stat, type, unit, base, exp, sz, bsz) \
1854 STATS_DESC_COMMON(type, unit, base, exp, sz, bsz), \
1855 .offset = offsetof(struct kvm_vm_stat, generic.stat) \
1859 #define VCPU_GENERIC_STATS_DESC(stat, type, unit, base, exp, sz, bsz) \
1862 STATS_DESC_COMMON(type, unit, base, exp, sz, bsz), \
1863 .offset = offsetof(struct kvm_vcpu_stat, generic.stat) \
1867 #define VM_STATS_DESC(stat, type, unit, base, exp, sz, bsz) \
1870 STATS_DESC_COMMON(type, unit, base, exp, sz, bsz), \
1871 .offset = offsetof(struct kvm_vm_stat, stat) \
1875 #define VCPU_STATS_DESC(stat, type, unit, base, exp, sz, bsz) \
1878 STATS_DESC_COMMON(type, unit, base, exp, sz, bsz), \
1879 .offset = offsetof(struct kvm_vcpu_stat, stat) \
1883 /* SCOPE: VM, VM_GENERIC, VCPU, VCPU_GENERIC */
1884 #define STATS_DESC(SCOPE, stat, type, unit, base, exp, sz, bsz) \
1885 SCOPE##_STATS_DESC(stat, type, unit, base, exp, sz, bsz)
1887 #define STATS_DESC_CUMULATIVE(SCOPE, name, unit, base, exponent) \
1888 STATS_DESC(SCOPE, name, KVM_STATS_TYPE_CUMULATIVE, \
1889 unit, base, exponent, 1, 0)
1890 #define STATS_DESC_INSTANT(SCOPE, name, unit, base, exponent) \
1891 STATS_DESC(SCOPE, name, KVM_STATS_TYPE_INSTANT, \
1892 unit, base, exponent, 1, 0)
1893 #define STATS_DESC_PEAK(SCOPE, name, unit, base, exponent) \
1894 STATS_DESC(SCOPE, name, KVM_STATS_TYPE_PEAK, \
1895 unit, base, exponent, 1, 0)
1896 #define STATS_DESC_LINEAR_HIST(SCOPE, name, unit, base, exponent, sz, bsz) \
1897 STATS_DESC(SCOPE, name, KVM_STATS_TYPE_LINEAR_HIST, \
1898 unit, base, exponent, sz, bsz)
1899 #define STATS_DESC_LOG_HIST(SCOPE, name, unit, base, exponent, sz) \
1900 STATS_DESC(SCOPE, name, KVM_STATS_TYPE_LOG_HIST, \
1901 unit, base, exponent, sz, 0)
1903 /* Cumulative counter, read/write */
1904 #define STATS_DESC_COUNTER(SCOPE, name) \
1905 STATS_DESC_CUMULATIVE(SCOPE, name, KVM_STATS_UNIT_NONE, \
1906 KVM_STATS_BASE_POW10, 0)
1907 /* Instantaneous counter, read only */
1908 #define STATS_DESC_ICOUNTER(SCOPE, name) \
1909 STATS_DESC_INSTANT(SCOPE, name, KVM_STATS_UNIT_NONE, \
1910 KVM_STATS_BASE_POW10, 0)
1911 /* Peak counter, read/write */
1912 #define STATS_DESC_PCOUNTER(SCOPE, name) \
1913 STATS_DESC_PEAK(SCOPE, name, KVM_STATS_UNIT_NONE, \
1914 KVM_STATS_BASE_POW10, 0)
1916 /* Instantaneous boolean value, read only */
1917 #define STATS_DESC_IBOOLEAN(SCOPE, name) \
1918 STATS_DESC_INSTANT(SCOPE, name, KVM_STATS_UNIT_BOOLEAN, \
1919 KVM_STATS_BASE_POW10, 0)
1920 /* Peak (sticky) boolean value, read/write */
1921 #define STATS_DESC_PBOOLEAN(SCOPE, name) \
1922 STATS_DESC_PEAK(SCOPE, name, KVM_STATS_UNIT_BOOLEAN, \
1923 KVM_STATS_BASE_POW10, 0)
1925 /* Cumulative time in nanosecond */
1926 #define STATS_DESC_TIME_NSEC(SCOPE, name) \
1927 STATS_DESC_CUMULATIVE(SCOPE, name, KVM_STATS_UNIT_SECONDS, \
1928 KVM_STATS_BASE_POW10, -9)
1929 /* Linear histogram for time in nanosecond */
1930 #define STATS_DESC_LINHIST_TIME_NSEC(SCOPE, name, sz, bsz) \
1931 STATS_DESC_LINEAR_HIST(SCOPE, name, KVM_STATS_UNIT_SECONDS, \
1932 KVM_STATS_BASE_POW10, -9, sz, bsz)
1933 /* Logarithmic histogram for time in nanosecond */
1934 #define STATS_DESC_LOGHIST_TIME_NSEC(SCOPE, name, sz) \
1935 STATS_DESC_LOG_HIST(SCOPE, name, KVM_STATS_UNIT_SECONDS, \
1936 KVM_STATS_BASE_POW10, -9, sz)
1938 #define KVM_GENERIC_VM_STATS() \
1939 STATS_DESC_COUNTER(VM_GENERIC, remote_tlb_flush), \
1940 STATS_DESC_COUNTER(VM_GENERIC, remote_tlb_flush_requests)
1942 #define KVM_GENERIC_VCPU_STATS() \
1943 STATS_DESC_COUNTER(VCPU_GENERIC, halt_successful_poll), \
1944 STATS_DESC_COUNTER(VCPU_GENERIC, halt_attempted_poll), \
1945 STATS_DESC_COUNTER(VCPU_GENERIC, halt_poll_invalid), \
1946 STATS_DESC_COUNTER(VCPU_GENERIC, halt_wakeup), \
1947 STATS_DESC_TIME_NSEC(VCPU_GENERIC, halt_poll_success_ns), \
1948 STATS_DESC_TIME_NSEC(VCPU_GENERIC, halt_poll_fail_ns), \
1949 STATS_DESC_TIME_NSEC(VCPU_GENERIC, halt_wait_ns), \
1950 STATS_DESC_LOGHIST_TIME_NSEC(VCPU_GENERIC, halt_poll_success_hist, \
1951 HALT_POLL_HIST_COUNT), \
1952 STATS_DESC_LOGHIST_TIME_NSEC(VCPU_GENERIC, halt_poll_fail_hist, \
1953 HALT_POLL_HIST_COUNT), \
1954 STATS_DESC_LOGHIST_TIME_NSEC(VCPU_GENERIC, halt_wait_hist, \
1955 HALT_POLL_HIST_COUNT), \
1956 STATS_DESC_IBOOLEAN(VCPU_GENERIC, blocking)
1958 extern struct dentry *kvm_debugfs_dir;
1960 ssize_t kvm_stats_read(char *id, const struct kvm_stats_header *header,
1961 const struct _kvm_stats_desc *desc,
1962 void *stats, size_t size_stats,
1963 char __user *user_buffer, size_t size, loff_t *offset);
1966 * kvm_stats_linear_hist_update() - Update bucket value for linear histogram
1969 * @data: start address of the stats data
1970 * @size: the number of bucket of the stats data
1971 * @value: the new value used to update the linear histogram's bucket
1972 * @bucket_size: the size (width) of a bucket
1974 static inline void kvm_stats_linear_hist_update(u64 *data, size_t size,
1975 u64 value, size_t bucket_size)
1977 size_t index = div64_u64(value, bucket_size);
1979 index = min(index, size - 1);
1984 * kvm_stats_log_hist_update() - Update bucket value for logarithmic histogram
1987 * @data: start address of the stats data
1988 * @size: the number of bucket of the stats data
1989 * @value: the new value used to update the logarithmic histogram's bucket
1991 static inline void kvm_stats_log_hist_update(u64 *data, size_t size, u64 value)
1993 size_t index = fls64(value);
1995 index = min(index, size - 1);
1999 #define KVM_STATS_LINEAR_HIST_UPDATE(array, value, bsize) \
2000 kvm_stats_linear_hist_update(array, ARRAY_SIZE(array), value, bsize)
2001 #define KVM_STATS_LOG_HIST_UPDATE(array, value) \
2002 kvm_stats_log_hist_update(array, ARRAY_SIZE(array), value)
2005 extern const struct kvm_stats_header kvm_vm_stats_header;
2006 extern const struct _kvm_stats_desc kvm_vm_stats_desc[];
2007 extern const struct kvm_stats_header kvm_vcpu_stats_header;
2008 extern const struct _kvm_stats_desc kvm_vcpu_stats_desc[];
2010 #ifdef CONFIG_KVM_GENERIC_MMU_NOTIFIER
2011 static inline int mmu_invalidate_retry(struct kvm *kvm, unsigned long mmu_seq)
2013 if (unlikely(kvm->mmu_invalidate_in_progress))
2016 * Ensure the read of mmu_invalidate_in_progress happens before
2017 * the read of mmu_invalidate_seq. This interacts with the
2018 * smp_wmb() in mmu_notifier_invalidate_range_end to make sure
2019 * that the caller either sees the old (non-zero) value of
2020 * mmu_invalidate_in_progress or the new (incremented) value of
2021 * mmu_invalidate_seq.
2023 * PowerPC Book3s HV KVM calls this under a per-page lock rather
2024 * than under kvm->mmu_lock, for scalability, so can't rely on
2025 * kvm->mmu_lock to keep things ordered.
2028 if (kvm->mmu_invalidate_seq != mmu_seq)
2033 static inline int mmu_invalidate_retry_gfn(struct kvm *kvm,
2034 unsigned long mmu_seq,
2037 lockdep_assert_held(&kvm->mmu_lock);
2039 * If mmu_invalidate_in_progress is non-zero, then the range maintained
2040 * by kvm_mmu_notifier_invalidate_range_start contains all addresses
2041 * that might be being invalidated. Note that it may include some false
2042 * positives, due to shortcuts when handing concurrent invalidations.
2044 if (unlikely(kvm->mmu_invalidate_in_progress)) {
2046 * Dropping mmu_lock after bumping mmu_invalidate_in_progress
2047 * but before updating the range is a KVM bug.
2049 if (WARN_ON_ONCE(kvm->mmu_invalidate_range_start == INVALID_GPA ||
2050 kvm->mmu_invalidate_range_end == INVALID_GPA))
2053 if (gfn >= kvm->mmu_invalidate_range_start &&
2054 gfn < kvm->mmu_invalidate_range_end)
2058 if (kvm->mmu_invalidate_seq != mmu_seq)
2064 * This lockless version of the range-based retry check *must* be paired with a
2065 * call to the locked version after acquiring mmu_lock, i.e. this is safe to
2066 * use only as a pre-check to avoid contending mmu_lock. This version *will*
2067 * get false negatives and false positives.
2069 static inline bool mmu_invalidate_retry_gfn_unsafe(struct kvm *kvm,
2070 unsigned long mmu_seq,
2074 * Use READ_ONCE() to ensure the in-progress flag and sequence counter
2075 * are always read from memory, e.g. so that checking for retry in a
2076 * loop won't result in an infinite retry loop. Don't force loads for
2077 * start+end, as the key to avoiding infinite retry loops is observing
2078 * the 1=>0 transition of in-progress, i.e. getting false negatives
2079 * due to stale start+end values is acceptable.
2081 if (unlikely(READ_ONCE(kvm->mmu_invalidate_in_progress)) &&
2082 gfn >= kvm->mmu_invalidate_range_start &&
2083 gfn < kvm->mmu_invalidate_range_end)
2086 return READ_ONCE(kvm->mmu_invalidate_seq) != mmu_seq;
2090 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2092 #define KVM_MAX_IRQ_ROUTES 4096 /* might need extension/rework in the future */
2094 bool kvm_arch_can_set_irq_routing(struct kvm *kvm);
2095 int kvm_set_irq_routing(struct kvm *kvm,
2096 const struct kvm_irq_routing_entry *entries,
2099 int kvm_set_routing_entry(struct kvm *kvm,
2100 struct kvm_kernel_irq_routing_entry *e,
2101 const struct kvm_irq_routing_entry *ue);
2102 void kvm_free_irq_routing(struct kvm *kvm);
2106 static inline void kvm_free_irq_routing(struct kvm *kvm) {}
2110 int kvm_send_userspace_msi(struct kvm *kvm, struct kvm_msi *msi);
2112 void kvm_eventfd_init(struct kvm *kvm);
2113 int kvm_ioeventfd(struct kvm *kvm, struct kvm_ioeventfd *args);
2115 #ifdef CONFIG_HAVE_KVM_IRQCHIP
2116 int kvm_irqfd(struct kvm *kvm, struct kvm_irqfd *args);
2117 void kvm_irqfd_release(struct kvm *kvm);
2118 bool kvm_notify_irqfd_resampler(struct kvm *kvm,
2119 unsigned int irqchip,
2121 void kvm_irq_routing_update(struct kvm *);
2123 static inline int kvm_irqfd(struct kvm *kvm, struct kvm_irqfd *args)
2128 static inline void kvm_irqfd_release(struct kvm *kvm) {}
2130 static inline bool kvm_notify_irqfd_resampler(struct kvm *kvm,
2131 unsigned int irqchip,
2136 #endif /* CONFIG_HAVE_KVM_IRQCHIP */
2138 void kvm_arch_irq_routing_update(struct kvm *kvm);
2140 static inline void __kvm_make_request(int req, struct kvm_vcpu *vcpu)
2143 * Ensure the rest of the request is published to kvm_check_request's
2144 * caller. Paired with the smp_mb__after_atomic in kvm_check_request.
2147 set_bit(req & KVM_REQUEST_MASK, (void *)&vcpu->requests);
2150 static __always_inline void kvm_make_request(int req, struct kvm_vcpu *vcpu)
2153 * Request that don't require vCPU action should never be logged in
2154 * vcpu->requests. The vCPU won't clear the request, so it will stay
2155 * logged indefinitely and prevent the vCPU from entering the guest.
2157 BUILD_BUG_ON(!__builtin_constant_p(req) ||
2158 (req & KVM_REQUEST_NO_ACTION));
2160 __kvm_make_request(req, vcpu);
2163 static inline bool kvm_request_pending(struct kvm_vcpu *vcpu)
2165 return READ_ONCE(vcpu->requests);
2168 static inline bool kvm_test_request(int req, struct kvm_vcpu *vcpu)
2170 return test_bit(req & KVM_REQUEST_MASK, (void *)&vcpu->requests);
2173 static inline void kvm_clear_request(int req, struct kvm_vcpu *vcpu)
2175 clear_bit(req & KVM_REQUEST_MASK, (void *)&vcpu->requests);
2178 static inline bool kvm_check_request(int req, struct kvm_vcpu *vcpu)
2180 if (kvm_test_request(req, vcpu)) {
2181 kvm_clear_request(req, vcpu);
2184 * Ensure the rest of the request is visible to kvm_check_request's
2185 * caller. Paired with the smp_wmb in kvm_make_request.
2187 smp_mb__after_atomic();
2194 #ifdef CONFIG_KVM_GENERIC_HARDWARE_ENABLING
2195 extern bool kvm_rebooting;
2198 extern unsigned int halt_poll_ns;
2199 extern unsigned int halt_poll_ns_grow;
2200 extern unsigned int halt_poll_ns_grow_start;
2201 extern unsigned int halt_poll_ns_shrink;
2204 const struct kvm_device_ops *ops;
2207 struct list_head vm_node;
2210 /* create, destroy, and name are mandatory */
2211 struct kvm_device_ops {
2215 * create is called holding kvm->lock and any operations not suitable
2216 * to do while holding the lock should be deferred to init (see
2219 int (*create)(struct kvm_device *dev, u32 type);
2222 * init is called after create if create is successful and is called
2223 * outside of holding kvm->lock.
2225 void (*init)(struct kvm_device *dev);
2228 * Destroy is responsible for freeing dev.
2230 * Destroy may be called before or after destructors are called
2231 * on emulated I/O regions, depending on whether a reference is
2232 * held by a vcpu or other kvm component that gets destroyed
2233 * after the emulated I/O.
2235 void (*destroy)(struct kvm_device *dev);
2238 * Release is an alternative method to free the device. It is
2239 * called when the device file descriptor is closed. Once
2240 * release is called, the destroy method will not be called
2241 * anymore as the device is removed from the device list of
2242 * the VM. kvm->lock is held.
2244 void (*release)(struct kvm_device *dev);
2246 int (*set_attr)(struct kvm_device *dev, struct kvm_device_attr *attr);
2247 int (*get_attr)(struct kvm_device *dev, struct kvm_device_attr *attr);
2248 int (*has_attr)(struct kvm_device *dev, struct kvm_device_attr *attr);
2249 long (*ioctl)(struct kvm_device *dev, unsigned int ioctl,
2251 int (*mmap)(struct kvm_device *dev, struct vm_area_struct *vma);
2254 struct kvm_device *kvm_device_from_filp(struct file *filp);
2255 int kvm_register_device_ops(const struct kvm_device_ops *ops, u32 type);
2256 void kvm_unregister_device_ops(u32 type);
2258 extern struct kvm_device_ops kvm_mpic_ops;
2259 extern struct kvm_device_ops kvm_arm_vgic_v2_ops;
2260 extern struct kvm_device_ops kvm_arm_vgic_v3_ops;
2262 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2264 static inline void kvm_vcpu_set_in_spin_loop(struct kvm_vcpu *vcpu, bool val)
2266 vcpu->spin_loop.in_spin_loop = val;
2268 static inline void kvm_vcpu_set_dy_eligible(struct kvm_vcpu *vcpu, bool val)
2270 vcpu->spin_loop.dy_eligible = val;
2273 #else /* !CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT */
2275 static inline void kvm_vcpu_set_in_spin_loop(struct kvm_vcpu *vcpu, bool val)
2279 static inline void kvm_vcpu_set_dy_eligible(struct kvm_vcpu *vcpu, bool val)
2282 #endif /* CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT */
2284 static inline bool kvm_is_visible_memslot(struct kvm_memory_slot *memslot)
2286 return (memslot && memslot->id < KVM_USER_MEM_SLOTS &&
2287 !(memslot->flags & KVM_MEMSLOT_INVALID));
2290 struct kvm_vcpu *kvm_get_running_vcpu(void);
2291 struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void);
2293 #ifdef CONFIG_HAVE_KVM_IRQ_BYPASS
2294 bool kvm_arch_has_irq_bypass(void);
2295 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *,
2296 struct irq_bypass_producer *);
2297 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *,
2298 struct irq_bypass_producer *);
2299 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *);
2300 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *);
2301 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
2302 uint32_t guest_irq, bool set);
2303 bool kvm_arch_irqfd_route_changed(struct kvm_kernel_irq_routing_entry *,
2304 struct kvm_kernel_irq_routing_entry *);
2305 #endif /* CONFIG_HAVE_KVM_IRQ_BYPASS */
2307 #ifdef CONFIG_HAVE_KVM_INVALID_WAKEUPS
2308 /* If we wakeup during the poll time, was it a sucessful poll? */
2309 static inline bool vcpu_valid_wakeup(struct kvm_vcpu *vcpu)
2311 return vcpu->valid_wakeup;
2315 static inline bool vcpu_valid_wakeup(struct kvm_vcpu *vcpu)
2319 #endif /* CONFIG_HAVE_KVM_INVALID_WAKEUPS */
2321 #ifdef CONFIG_HAVE_KVM_NO_POLL
2322 /* Callback that tells if we must not poll */
2323 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu);
2325 static inline bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
2329 #endif /* CONFIG_HAVE_KVM_NO_POLL */
2331 #ifdef CONFIG_HAVE_KVM_VCPU_ASYNC_IOCTL
2332 long kvm_arch_vcpu_async_ioctl(struct file *filp,
2333 unsigned int ioctl, unsigned long arg);
2335 static inline long kvm_arch_vcpu_async_ioctl(struct file *filp,
2339 return -ENOIOCTLCMD;
2341 #endif /* CONFIG_HAVE_KVM_VCPU_ASYNC_IOCTL */
2343 void kvm_arch_guest_memory_reclaimed(struct kvm *kvm);
2345 #ifdef CONFIG_HAVE_KVM_VCPU_RUN_PID_CHANGE
2346 int kvm_arch_vcpu_run_pid_change(struct kvm_vcpu *vcpu);
2348 static inline int kvm_arch_vcpu_run_pid_change(struct kvm_vcpu *vcpu)
2352 #endif /* CONFIG_HAVE_KVM_VCPU_RUN_PID_CHANGE */
2354 typedef int (*kvm_vm_thread_fn_t)(struct kvm *kvm, uintptr_t data);
2356 int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn,
2357 uintptr_t data, const char *name,
2358 struct task_struct **thread_ptr);
2360 #ifdef CONFIG_KVM_XFER_TO_GUEST_WORK
2361 static inline void kvm_handle_signal_exit(struct kvm_vcpu *vcpu)
2363 vcpu->run->exit_reason = KVM_EXIT_INTR;
2364 vcpu->stat.signal_exits++;
2366 #endif /* CONFIG_KVM_XFER_TO_GUEST_WORK */
2369 * If more than one page is being (un)accounted, @virt must be the address of
2370 * the first page of a block of pages what were allocated together (i.e
2371 * accounted together).
2373 * kvm_account_pgtable_pages() is thread-safe because mod_lruvec_page_state()
2376 static inline void kvm_account_pgtable_pages(void *virt, int nr)
2378 mod_lruvec_page_state(virt_to_page(virt), NR_SECONDARY_PAGETABLE, nr);
2382 * This defines how many reserved entries we want to keep before we
2383 * kick the vcpu to the userspace to avoid dirty ring full. This
2384 * value can be tuned to higher if e.g. PML is enabled on the host.
2386 #define KVM_DIRTY_RING_RSVD_ENTRIES 64
2388 /* Max number of entries allowed for each kvm dirty ring */
2389 #define KVM_DIRTY_RING_MAX_ENTRIES 65536
2391 static inline void kvm_prepare_memory_fault_exit(struct kvm_vcpu *vcpu,
2392 gpa_t gpa, gpa_t size,
2393 bool is_write, bool is_exec,
2396 vcpu->run->exit_reason = KVM_EXIT_MEMORY_FAULT;
2397 vcpu->run->memory_fault.gpa = gpa;
2398 vcpu->run->memory_fault.size = size;
2400 /* RWX flags are not (yet) defined or communicated to userspace. */
2401 vcpu->run->memory_fault.flags = 0;
2403 vcpu->run->memory_fault.flags |= KVM_MEMORY_EXIT_FLAG_PRIVATE;
2406 #ifdef CONFIG_KVM_GENERIC_MEMORY_ATTRIBUTES
2407 static inline unsigned long kvm_get_memory_attributes(struct kvm *kvm, gfn_t gfn)
2409 return xa_to_value(xa_load(&kvm->mem_attr_array, gfn));
2412 bool kvm_range_has_memory_attributes(struct kvm *kvm, gfn_t start, gfn_t end,
2413 unsigned long attrs);
2414 bool kvm_arch_pre_set_memory_attributes(struct kvm *kvm,
2415 struct kvm_gfn_range *range);
2416 bool kvm_arch_post_set_memory_attributes(struct kvm *kvm,
2417 struct kvm_gfn_range *range);
2419 static inline bool kvm_mem_is_private(struct kvm *kvm, gfn_t gfn)
2421 return IS_ENABLED(CONFIG_KVM_PRIVATE_MEM) &&
2422 kvm_get_memory_attributes(kvm, gfn) & KVM_MEMORY_ATTRIBUTE_PRIVATE;
2425 static inline bool kvm_mem_is_private(struct kvm *kvm, gfn_t gfn)
2429 #endif /* CONFIG_KVM_GENERIC_MEMORY_ATTRIBUTES */
2431 #ifdef CONFIG_KVM_PRIVATE_MEM
2432 int kvm_gmem_get_pfn(struct kvm *kvm, struct kvm_memory_slot *slot,
2433 gfn_t gfn, kvm_pfn_t *pfn, int *max_order);
2435 static inline int kvm_gmem_get_pfn(struct kvm *kvm,
2436 struct kvm_memory_slot *slot, gfn_t gfn,
2437 kvm_pfn_t *pfn, int *max_order)
2442 #endif /* CONFIG_KVM_PRIVATE_MEM */