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_memory_region::flags are internally used
54 * 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 install_new_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_ADDRESS_SPACE_NUM
84 #define KVM_ADDRESS_SPACE_NUM 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)
101 * error pfns indicate that the gfn is in slot but faild to
102 * translate it to pfn on host.
104 static inline bool is_error_pfn(kvm_pfn_t pfn)
106 return !!(pfn & KVM_PFN_ERR_MASK);
110 * error_noslot pfns indicate that the gfn can not be
111 * translated to pfn - it is not in slot or failed to
112 * translate it to pfn.
114 static inline bool is_error_noslot_pfn(kvm_pfn_t pfn)
116 return !!(pfn & KVM_PFN_ERR_NOSLOT_MASK);
119 /* noslot pfn indicates that the gfn is not in slot. */
120 static inline bool is_noslot_pfn(kvm_pfn_t pfn)
122 return pfn == KVM_PFN_NOSLOT;
126 * architectures with KVM_HVA_ERR_BAD other than PAGE_OFFSET (e.g. s390)
127 * provide own defines and kvm_is_error_hva
129 #ifndef KVM_HVA_ERR_BAD
131 #define KVM_HVA_ERR_BAD (PAGE_OFFSET)
132 #define KVM_HVA_ERR_RO_BAD (PAGE_OFFSET + PAGE_SIZE)
134 static inline bool kvm_is_error_hva(unsigned long addr)
136 return addr >= PAGE_OFFSET;
141 #define KVM_ERR_PTR_BAD_PAGE (ERR_PTR(-ENOENT))
143 static inline bool is_error_page(struct page *page)
148 #define KVM_REQUEST_MASK GENMASK(7,0)
149 #define KVM_REQUEST_NO_WAKEUP BIT(8)
150 #define KVM_REQUEST_WAIT BIT(9)
151 #define KVM_REQUEST_NO_ACTION BIT(10)
153 * Architecture-independent vcpu->requests bit members
154 * Bits 3-7 are reserved for more arch-independent bits.
156 #define KVM_REQ_TLB_FLUSH (0 | KVM_REQUEST_WAIT | KVM_REQUEST_NO_WAKEUP)
157 #define KVM_REQ_VM_DEAD (1 | KVM_REQUEST_WAIT | KVM_REQUEST_NO_WAKEUP)
158 #define KVM_REQ_UNBLOCK 2
159 #define KVM_REQUEST_ARCH_BASE 8
162 * KVM_REQ_OUTSIDE_GUEST_MODE exists is purely as way to force the vCPU to
163 * OUTSIDE_GUEST_MODE. KVM_REQ_OUTSIDE_GUEST_MODE differs from a vCPU "kick"
164 * in that it ensures the vCPU has reached OUTSIDE_GUEST_MODE before continuing
165 * on. A kick only guarantees that the vCPU is on its way out, e.g. a previous
166 * kick may have set vcpu->mode to EXITING_GUEST_MODE, and so there's no
167 * guarantee the vCPU received an IPI and has actually exited guest mode.
169 #define KVM_REQ_OUTSIDE_GUEST_MODE (KVM_REQUEST_NO_ACTION | KVM_REQUEST_WAIT | KVM_REQUEST_NO_WAKEUP)
171 #define KVM_ARCH_REQ_FLAGS(nr, flags) ({ \
172 BUILD_BUG_ON((unsigned)(nr) >= (sizeof_field(struct kvm_vcpu, requests) * 8) - KVM_REQUEST_ARCH_BASE); \
173 (unsigned)(((nr) + KVM_REQUEST_ARCH_BASE) | (flags)); \
175 #define KVM_ARCH_REQ(nr) KVM_ARCH_REQ_FLAGS(nr, 0)
177 bool kvm_make_vcpus_request_mask(struct kvm *kvm, unsigned int req,
178 unsigned long *vcpu_bitmap);
179 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req);
180 bool kvm_make_all_cpus_request_except(struct kvm *kvm, unsigned int req,
181 struct kvm_vcpu *except);
182 bool kvm_make_cpus_request_mask(struct kvm *kvm, unsigned int req,
183 unsigned long *vcpu_bitmap);
185 #define KVM_USERSPACE_IRQ_SOURCE_ID 0
186 #define KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID 1
188 extern struct mutex kvm_lock;
189 extern struct list_head vm_list;
191 struct kvm_io_range {
194 struct kvm_io_device *dev;
197 #define NR_IOBUS_DEVS 1000
202 struct kvm_io_range range[];
208 KVM_VIRTIO_CCW_NOTIFY_BUS,
213 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
214 int len, const void *val);
215 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
216 gpa_t addr, int len, const void *val, long cookie);
217 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
219 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
220 int len, struct kvm_io_device *dev);
221 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
222 struct kvm_io_device *dev);
223 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
226 #ifdef CONFIG_KVM_ASYNC_PF
227 struct kvm_async_pf {
228 struct work_struct work;
229 struct list_head link;
230 struct list_head queue;
231 struct kvm_vcpu *vcpu;
232 struct mm_struct *mm;
235 struct kvm_arch_async_pf arch;
237 bool notpresent_injected;
240 void kvm_clear_async_pf_completion_queue(struct kvm_vcpu *vcpu);
241 void kvm_check_async_pf_completion(struct kvm_vcpu *vcpu);
242 bool kvm_setup_async_pf(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
243 unsigned long hva, struct kvm_arch_async_pf *arch);
244 int kvm_async_pf_wakeup_all(struct kvm_vcpu *vcpu);
247 #ifdef KVM_ARCH_WANT_MMU_NOTIFIER
248 struct kvm_gfn_range {
249 struct kvm_memory_slot *slot;
255 bool kvm_unmap_gfn_range(struct kvm *kvm, struct kvm_gfn_range *range);
256 bool kvm_age_gfn(struct kvm *kvm, struct kvm_gfn_range *range);
257 bool kvm_test_age_gfn(struct kvm *kvm, struct kvm_gfn_range *range);
258 bool kvm_set_spte_gfn(struct kvm *kvm, struct kvm_gfn_range *range);
265 READING_SHADOW_PAGE_TABLES,
268 #define KVM_UNMAPPED_PAGE ((void *) 0x500 + POISON_POINTER_DELTA)
270 struct kvm_host_map {
272 * Only valid if the 'pfn' is managed by the host kernel (i.e. There is
273 * a 'struct page' for it. When using mem= kernel parameter some memory
274 * can be used as guest memory but they are not managed by host
276 * If 'pfn' is not managed by the host kernel, this field is
277 * initialized to KVM_UNMAPPED_PAGE.
286 * Used to check if the mapping is valid or not. Never use 'kvm_host_map'
287 * directly to check for that.
289 static inline bool kvm_vcpu_mapped(struct kvm_host_map *map)
294 static inline bool kvm_vcpu_can_poll(ktime_t cur, ktime_t stop)
296 return single_task_running() && !need_resched() && ktime_before(cur, stop);
300 * Sometimes a large or cross-page mmio needs to be broken up into separate
301 * exits for userspace servicing.
303 struct kvm_mmio_fragment {
311 #ifdef CONFIG_PREEMPT_NOTIFIERS
312 struct preempt_notifier preempt_notifier;
315 int vcpu_id; /* id given by userspace at creation */
316 int vcpu_idx; /* index in kvm->vcpus array */
317 int ____srcu_idx; /* Don't use this directly. You've been warned. */
318 #ifdef CONFIG_PROVE_RCU
323 unsigned long guest_debug;
328 #ifndef __KVM_HAVE_ARCH_WQP
331 struct pid __rcu *pid;
334 unsigned int halt_poll_ns;
337 #ifdef CONFIG_HAS_IOMEM
339 int mmio_read_completed;
341 int mmio_cur_fragment;
342 int mmio_nr_fragments;
343 struct kvm_mmio_fragment mmio_fragments[KVM_MAX_MMIO_FRAGMENTS];
346 #ifdef CONFIG_KVM_ASYNC_PF
349 struct list_head queue;
350 struct list_head done;
355 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
357 * Cpu relax intercept or pause loop exit optimization
358 * in_spin_loop: set when a vcpu does a pause loop exit
359 * or cpu relax intercepted.
360 * dy_eligible: indicates whether vcpu is eligible for directed yield.
369 struct kvm_vcpu_arch arch;
370 struct kvm_vcpu_stat stat;
371 char stats_id[KVM_STATS_NAME_SIZE];
372 struct kvm_dirty_ring dirty_ring;
375 * The most recently used memslot by this vCPU and the slots generation
376 * for which it is valid.
377 * No wraparound protection is needed since generations won't overflow in
378 * thousands of years, even assuming 1M memslot operations per second.
380 struct kvm_memory_slot *last_used_slot;
381 u64 last_used_slot_gen;
385 * Start accounting time towards a guest.
386 * Must be called before entering guest context.
388 static __always_inline void guest_timing_enter_irqoff(void)
391 * This is running in ioctl context so its safe to assume that it's the
392 * stime pending cputime to flush.
394 instrumentation_begin();
395 vtime_account_guest_enter();
396 instrumentation_end();
400 * Enter guest context and enter an RCU extended quiescent state.
402 * Between guest_context_enter_irqoff() and guest_context_exit_irqoff() it is
403 * unsafe to use any code which may directly or indirectly use RCU, tracing
404 * (including IRQ flag tracing), or lockdep. All code in this period must be
405 * non-instrumentable.
407 static __always_inline void guest_context_enter_irqoff(void)
410 * KVM does not hold any references to rcu protected data when it
411 * switches CPU into a guest mode. In fact switching to a guest mode
412 * is very similar to exiting to userspace from rcu point of view. In
413 * addition CPU may stay in a guest mode for quite a long time (up to
414 * one time slice). Lets treat guest mode as quiescent state, just like
415 * we do with user-mode execution.
417 if (!context_tracking_guest_enter()) {
418 instrumentation_begin();
419 rcu_virt_note_context_switch(smp_processor_id());
420 instrumentation_end();
425 * Deprecated. Architectures should move to guest_timing_enter_irqoff() and
426 * guest_state_enter_irqoff().
428 static __always_inline void guest_enter_irqoff(void)
430 guest_timing_enter_irqoff();
431 guest_context_enter_irqoff();
435 * guest_state_enter_irqoff - Fixup state when entering a guest
437 * Entry to a guest will enable interrupts, but the kernel state is interrupts
438 * disabled when this is invoked. Also tell RCU about it.
440 * 1) Trace interrupts on state
441 * 2) Invoke context tracking if enabled to adjust RCU state
442 * 3) Tell lockdep that interrupts are enabled
444 * Invoked from architecture specific code before entering a guest.
445 * Must be called with interrupts disabled and the caller must be
446 * non-instrumentable.
447 * The caller has to invoke guest_timing_enter_irqoff() before this.
449 * Note: this is analogous to exit_to_user_mode().
451 static __always_inline void guest_state_enter_irqoff(void)
453 instrumentation_begin();
454 trace_hardirqs_on_prepare();
455 lockdep_hardirqs_on_prepare();
456 instrumentation_end();
458 guest_context_enter_irqoff();
459 lockdep_hardirqs_on(CALLER_ADDR0);
463 * Exit guest context and exit an RCU extended quiescent state.
465 * Between guest_context_enter_irqoff() and guest_context_exit_irqoff() it is
466 * unsafe to use any code which may directly or indirectly use RCU, tracing
467 * (including IRQ flag tracing), or lockdep. All code in this period must be
468 * non-instrumentable.
470 static __always_inline void guest_context_exit_irqoff(void)
472 context_tracking_guest_exit();
476 * Stop accounting time towards a guest.
477 * Must be called after exiting guest context.
479 static __always_inline void guest_timing_exit_irqoff(void)
481 instrumentation_begin();
482 /* Flush the guest cputime we spent on the guest */
483 vtime_account_guest_exit();
484 instrumentation_end();
488 * Deprecated. Architectures should move to guest_state_exit_irqoff() and
489 * guest_timing_exit_irqoff().
491 static __always_inline void guest_exit_irqoff(void)
493 guest_context_exit_irqoff();
494 guest_timing_exit_irqoff();
497 static inline void guest_exit(void)
501 local_irq_save(flags);
503 local_irq_restore(flags);
507 * guest_state_exit_irqoff - Establish state when returning from guest mode
509 * Entry from a guest disables interrupts, but guest mode is traced as
510 * interrupts enabled. Also with NO_HZ_FULL RCU might be idle.
512 * 1) Tell lockdep that interrupts are disabled
513 * 2) Invoke context tracking if enabled to reactivate RCU
514 * 3) Trace interrupts off state
516 * Invoked from architecture specific code after exiting a guest.
517 * Must be invoked with interrupts disabled and the caller must be
518 * non-instrumentable.
519 * The caller has to invoke guest_timing_exit_irqoff() after this.
521 * Note: this is analogous to enter_from_user_mode().
523 static __always_inline void guest_state_exit_irqoff(void)
525 lockdep_hardirqs_off(CALLER_ADDR0);
526 guest_context_exit_irqoff();
528 instrumentation_begin();
529 trace_hardirqs_off_finish();
530 instrumentation_end();
533 static inline int kvm_vcpu_exiting_guest_mode(struct kvm_vcpu *vcpu)
536 * The memory barrier ensures a previous write to vcpu->requests cannot
537 * be reordered with the read of vcpu->mode. It pairs with the general
538 * memory barrier following the write of vcpu->mode in VCPU RUN.
540 smp_mb__before_atomic();
541 return cmpxchg(&vcpu->mode, IN_GUEST_MODE, EXITING_GUEST_MODE);
545 * Some of the bitops functions do not support too long bitmaps.
546 * This number must be determined not to exceed such limits.
548 #define KVM_MEM_MAX_NR_PAGES ((1UL << 31) - 1)
551 * Since at idle each memslot belongs to two memslot sets it has to contain
552 * two embedded nodes for each data structure that it forms a part of.
554 * Two memslot sets (one active and one inactive) are necessary so the VM
555 * continues to run on one memslot set while the other is being modified.
557 * These two memslot sets normally point to the same set of memslots.
558 * They can, however, be desynchronized when performing a memslot management
559 * operation by replacing the memslot to be modified by its copy.
560 * After the operation is complete, both memslot sets once again point to
561 * the same, common set of memslot data.
563 * The memslots themselves are independent of each other so they can be
564 * individually added or deleted.
566 struct kvm_memory_slot {
567 struct hlist_node id_node[2];
568 struct interval_tree_node hva_node[2];
569 struct rb_node gfn_node[2];
571 unsigned long npages;
572 unsigned long *dirty_bitmap;
573 struct kvm_arch_memory_slot arch;
574 unsigned long userspace_addr;
580 static inline bool kvm_slot_dirty_track_enabled(const struct kvm_memory_slot *slot)
582 return slot->flags & KVM_MEM_LOG_DIRTY_PAGES;
585 static inline unsigned long kvm_dirty_bitmap_bytes(struct kvm_memory_slot *memslot)
587 return ALIGN(memslot->npages, BITS_PER_LONG) / 8;
590 static inline unsigned long *kvm_second_dirty_bitmap(struct kvm_memory_slot *memslot)
592 unsigned long len = kvm_dirty_bitmap_bytes(memslot);
594 return memslot->dirty_bitmap + len / sizeof(*memslot->dirty_bitmap);
597 #ifndef KVM_DIRTY_LOG_MANUAL_CAPS
598 #define KVM_DIRTY_LOG_MANUAL_CAPS KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE
601 struct kvm_s390_adapter_int {
614 struct kvm_xen_evtchn {
621 struct kvm_kernel_irq_routing_entry {
624 int (*set)(struct kvm_kernel_irq_routing_entry *e,
625 struct kvm *kvm, int irq_source_id, int level,
639 struct kvm_s390_adapter_int adapter;
640 struct kvm_hv_sint hv_sint;
641 struct kvm_xen_evtchn xen_evtchn;
643 struct hlist_node link;
646 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
647 struct kvm_irq_routing_table {
648 int chip[KVM_NR_IRQCHIPS][KVM_IRQCHIP_NUM_PINS];
651 * Array indexed by gsi. Each entry contains list of irq chips
652 * the gsi is connected to.
654 struct hlist_head map[];
658 #ifndef KVM_INTERNAL_MEM_SLOTS
659 #define KVM_INTERNAL_MEM_SLOTS 0
662 #define KVM_MEM_SLOTS_NUM SHRT_MAX
663 #define KVM_USER_MEM_SLOTS (KVM_MEM_SLOTS_NUM - KVM_INTERNAL_MEM_SLOTS)
665 #ifndef __KVM_VCPU_MULTIPLE_ADDRESS_SPACE
666 static inline int kvm_arch_vcpu_memslots_id(struct kvm_vcpu *vcpu)
672 struct kvm_memslots {
674 atomic_long_t last_used_slot;
675 struct rb_root_cached hva_tree;
676 struct rb_root gfn_tree;
678 * The mapping table from slot id to memslot.
680 * 7-bit bucket count matches the size of the old id to index array for
681 * 512 slots, while giving good performance with this slot count.
682 * Higher bucket counts bring only small performance improvements but
683 * always result in higher memory usage (even for lower memslot counts).
685 DECLARE_HASHTABLE(id_hash, 7);
690 #ifdef KVM_HAVE_MMU_RWLOCK
694 #endif /* KVM_HAVE_MMU_RWLOCK */
696 struct mutex slots_lock;
699 * Protects the arch-specific fields of struct kvm_memory_slots in
700 * use by the VM. To be used under the slots_lock (above) or in a
701 * kvm->srcu critical section where acquiring the slots_lock would
702 * lead to deadlock with the synchronize_srcu in
703 * install_new_memslots.
705 struct mutex slots_arch_lock;
706 struct mm_struct *mm; /* userspace tied to this vm */
707 unsigned long nr_memslot_pages;
708 /* The two memslot sets - active and inactive (per address space) */
709 struct kvm_memslots __memslots[KVM_ADDRESS_SPACE_NUM][2];
710 /* The current active memslot set for each address space */
711 struct kvm_memslots __rcu *memslots[KVM_ADDRESS_SPACE_NUM];
712 struct xarray vcpu_array;
714 /* Used to wait for completion of MMU notifiers. */
715 spinlock_t mn_invalidate_lock;
716 unsigned long mn_active_invalidate_count;
717 struct rcuwait mn_memslots_update_rcuwait;
719 /* For management / invalidation of gfn_to_pfn_caches */
721 struct list_head gpc_list;
724 * created_vcpus is protected by kvm->lock, and is incremented
725 * at the beginning of KVM_CREATE_VCPU. online_vcpus is only
726 * incremented after storing the kvm_vcpu pointer in vcpus,
727 * and is accessed atomically.
729 atomic_t online_vcpus;
732 int last_boosted_vcpu;
733 struct list_head vm_list;
735 struct kvm_io_bus __rcu *buses[KVM_NR_BUSES];
736 #ifdef CONFIG_HAVE_KVM_EVENTFD
739 struct list_head items;
740 struct list_head resampler_list;
741 struct mutex resampler_lock;
743 struct list_head ioeventfds;
745 struct kvm_vm_stat stat;
746 struct kvm_arch arch;
747 refcount_t users_count;
748 #ifdef CONFIG_KVM_MMIO
749 struct kvm_coalesced_mmio_ring *coalesced_mmio_ring;
750 spinlock_t ring_lock;
751 struct list_head coalesced_zones;
754 struct mutex irq_lock;
755 #ifdef CONFIG_HAVE_KVM_IRQCHIP
757 * Update side is protected by irq_lock.
759 struct kvm_irq_routing_table __rcu *irq_routing;
761 #ifdef CONFIG_HAVE_KVM_IRQFD
762 struct hlist_head irq_ack_notifier_list;
765 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
766 struct mmu_notifier mmu_notifier;
767 unsigned long mmu_invalidate_seq;
768 long mmu_invalidate_in_progress;
769 unsigned long mmu_invalidate_range_start;
770 unsigned long mmu_invalidate_range_end;
772 struct list_head devices;
773 u64 manual_dirty_log_protect;
774 struct dentry *debugfs_dentry;
775 struct kvm_stat_data **debugfs_stat_data;
776 struct srcu_struct srcu;
777 struct srcu_struct irq_srcu;
779 unsigned int max_halt_poll_ns;
784 #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER
785 struct notifier_block pm_notifier;
787 char stats_id[KVM_STATS_NAME_SIZE];
790 #define kvm_err(fmt, ...) \
791 pr_err("kvm [%i]: " fmt, task_pid_nr(current), ## __VA_ARGS__)
792 #define kvm_info(fmt, ...) \
793 pr_info("kvm [%i]: " fmt, task_pid_nr(current), ## __VA_ARGS__)
794 #define kvm_debug(fmt, ...) \
795 pr_debug("kvm [%i]: " fmt, task_pid_nr(current), ## __VA_ARGS__)
796 #define kvm_debug_ratelimited(fmt, ...) \
797 pr_debug_ratelimited("kvm [%i]: " fmt, task_pid_nr(current), \
799 #define kvm_pr_unimpl(fmt, ...) \
800 pr_err_ratelimited("kvm [%i]: " fmt, \
801 task_tgid_nr(current), ## __VA_ARGS__)
803 /* The guest did something we don't support. */
804 #define vcpu_unimpl(vcpu, fmt, ...) \
805 kvm_pr_unimpl("vcpu%i, guest rIP: 0x%lx " fmt, \
806 (vcpu)->vcpu_id, kvm_rip_read(vcpu), ## __VA_ARGS__)
808 #define vcpu_debug(vcpu, fmt, ...) \
809 kvm_debug("vcpu%i " fmt, (vcpu)->vcpu_id, ## __VA_ARGS__)
810 #define vcpu_debug_ratelimited(vcpu, fmt, ...) \
811 kvm_debug_ratelimited("vcpu%i " fmt, (vcpu)->vcpu_id, \
813 #define vcpu_err(vcpu, fmt, ...) \
814 kvm_err("vcpu%i " fmt, (vcpu)->vcpu_id, ## __VA_ARGS__)
816 static inline void kvm_vm_dead(struct kvm *kvm)
819 kvm_make_all_cpus_request(kvm, KVM_REQ_VM_DEAD);
822 static inline void kvm_vm_bugged(struct kvm *kvm)
824 kvm->vm_bugged = true;
829 #define KVM_BUG(cond, kvm, fmt...) \
831 int __ret = (cond); \
833 if (WARN_ONCE(__ret && !(kvm)->vm_bugged, fmt)) \
834 kvm_vm_bugged(kvm); \
838 #define KVM_BUG_ON(cond, kvm) \
840 int __ret = (cond); \
842 if (WARN_ON_ONCE(__ret && !(kvm)->vm_bugged)) \
843 kvm_vm_bugged(kvm); \
847 static inline void kvm_vcpu_srcu_read_lock(struct kvm_vcpu *vcpu)
849 #ifdef CONFIG_PROVE_RCU
850 WARN_ONCE(vcpu->srcu_depth++,
851 "KVM: Illegal vCPU srcu_idx LOCK, depth=%d", vcpu->srcu_depth - 1);
853 vcpu->____srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
856 static inline void kvm_vcpu_srcu_read_unlock(struct kvm_vcpu *vcpu)
858 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->____srcu_idx);
860 #ifdef CONFIG_PROVE_RCU
861 WARN_ONCE(--vcpu->srcu_depth,
862 "KVM: Illegal vCPU srcu_idx UNLOCK, depth=%d", vcpu->srcu_depth);
866 static inline bool kvm_dirty_log_manual_protect_and_init_set(struct kvm *kvm)
868 return !!(kvm->manual_dirty_log_protect & KVM_DIRTY_LOG_INITIALLY_SET);
871 static inline struct kvm_io_bus *kvm_get_bus(struct kvm *kvm, enum kvm_bus idx)
873 return srcu_dereference_check(kvm->buses[idx], &kvm->srcu,
874 lockdep_is_held(&kvm->slots_lock) ||
875 !refcount_read(&kvm->users_count));
878 static inline struct kvm_vcpu *kvm_get_vcpu(struct kvm *kvm, int i)
880 int num_vcpus = atomic_read(&kvm->online_vcpus);
881 i = array_index_nospec(i, num_vcpus);
883 /* Pairs with smp_wmb() in kvm_vm_ioctl_create_vcpu. */
885 return xa_load(&kvm->vcpu_array, i);
888 #define kvm_for_each_vcpu(idx, vcpup, kvm) \
889 xa_for_each_range(&kvm->vcpu_array, idx, vcpup, 0, \
890 (atomic_read(&kvm->online_vcpus) - 1))
892 static inline struct kvm_vcpu *kvm_get_vcpu_by_id(struct kvm *kvm, int id)
894 struct kvm_vcpu *vcpu = NULL;
899 if (id < KVM_MAX_VCPUS)
900 vcpu = kvm_get_vcpu(kvm, id);
901 if (vcpu && vcpu->vcpu_id == id)
903 kvm_for_each_vcpu(i, vcpu, kvm)
904 if (vcpu->vcpu_id == id)
909 void kvm_destroy_vcpus(struct kvm *kvm);
911 void vcpu_load(struct kvm_vcpu *vcpu);
912 void vcpu_put(struct kvm_vcpu *vcpu);
914 #ifdef __KVM_HAVE_IOAPIC
915 void kvm_arch_post_irq_ack_notifier_list_update(struct kvm *kvm);
916 void kvm_arch_post_irq_routing_update(struct kvm *kvm);
918 static inline void kvm_arch_post_irq_ack_notifier_list_update(struct kvm *kvm)
921 static inline void kvm_arch_post_irq_routing_update(struct kvm *kvm)
926 #ifdef CONFIG_HAVE_KVM_IRQFD
927 int kvm_irqfd_init(void);
928 void kvm_irqfd_exit(void);
930 static inline int kvm_irqfd_init(void)
935 static inline void kvm_irqfd_exit(void)
939 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
940 struct module *module);
943 void kvm_get_kvm(struct kvm *kvm);
944 bool kvm_get_kvm_safe(struct kvm *kvm);
945 void kvm_put_kvm(struct kvm *kvm);
946 bool file_is_kvm(struct file *file);
947 void kvm_put_kvm_no_destroy(struct kvm *kvm);
949 static inline struct kvm_memslots *__kvm_memslots(struct kvm *kvm, int as_id)
951 as_id = array_index_nospec(as_id, KVM_ADDRESS_SPACE_NUM);
952 return srcu_dereference_check(kvm->memslots[as_id], &kvm->srcu,
953 lockdep_is_held(&kvm->slots_lock) ||
954 !refcount_read(&kvm->users_count));
957 static inline struct kvm_memslots *kvm_memslots(struct kvm *kvm)
959 return __kvm_memslots(kvm, 0);
962 static inline struct kvm_memslots *kvm_vcpu_memslots(struct kvm_vcpu *vcpu)
964 int as_id = kvm_arch_vcpu_memslots_id(vcpu);
966 return __kvm_memslots(vcpu->kvm, as_id);
969 static inline bool kvm_memslots_empty(struct kvm_memslots *slots)
971 return RB_EMPTY_ROOT(&slots->gfn_tree);
974 #define kvm_for_each_memslot(memslot, bkt, slots) \
975 hash_for_each(slots->id_hash, bkt, memslot, id_node[slots->node_idx]) \
976 if (WARN_ON_ONCE(!memslot->npages)) { \
980 struct kvm_memory_slot *id_to_memslot(struct kvm_memslots *slots, int id)
982 struct kvm_memory_slot *slot;
983 int idx = slots->node_idx;
985 hash_for_each_possible(slots->id_hash, slot, id_node[idx], id) {
993 /* Iterator used for walking memslots that overlap a gfn range. */
994 struct kvm_memslot_iter {
995 struct kvm_memslots *slots;
996 struct rb_node *node;
997 struct kvm_memory_slot *slot;
1000 static inline void kvm_memslot_iter_next(struct kvm_memslot_iter *iter)
1002 iter->node = rb_next(iter->node);
1006 iter->slot = container_of(iter->node, struct kvm_memory_slot, gfn_node[iter->slots->node_idx]);
1009 static inline void kvm_memslot_iter_start(struct kvm_memslot_iter *iter,
1010 struct kvm_memslots *slots,
1013 int idx = slots->node_idx;
1014 struct rb_node *tmp;
1015 struct kvm_memory_slot *slot;
1017 iter->slots = slots;
1020 * Find the so called "upper bound" of a key - the first node that has
1021 * its key strictly greater than the searched one (the start gfn in our case).
1024 for (tmp = slots->gfn_tree.rb_node; tmp; ) {
1025 slot = container_of(tmp, struct kvm_memory_slot, gfn_node[idx]);
1026 if (start < slot->base_gfn) {
1030 tmp = tmp->rb_right;
1035 * Find the slot with the lowest gfn that can possibly intersect with
1036 * the range, so we'll ideally have slot start <= range start
1040 * A NULL previous node means that the very first slot
1041 * already has a higher start gfn.
1042 * In this case slot start > range start.
1044 tmp = rb_prev(iter->node);
1048 /* a NULL node below means no slots */
1049 iter->node = rb_last(&slots->gfn_tree);
1053 iter->slot = container_of(iter->node, struct kvm_memory_slot, gfn_node[idx]);
1056 * It is possible in the slot start < range start case that the
1057 * found slot ends before or at range start (slot end <= range start)
1058 * and so it does not overlap the requested range.
1060 * In such non-overlapping case the next slot (if it exists) will
1061 * already have slot start > range start, otherwise the logic above
1062 * would have found it instead of the current slot.
1064 if (iter->slot->base_gfn + iter->slot->npages <= start)
1065 kvm_memslot_iter_next(iter);
1069 static inline bool kvm_memslot_iter_is_valid(struct kvm_memslot_iter *iter, gfn_t end)
1075 * If this slot starts beyond or at the end of the range so does
1078 return iter->slot->base_gfn < end;
1081 /* Iterate over each memslot at least partially intersecting [start, end) range */
1082 #define kvm_for_each_memslot_in_gfn_range(iter, slots, start, end) \
1083 for (kvm_memslot_iter_start(iter, slots, start); \
1084 kvm_memslot_iter_is_valid(iter, end); \
1085 kvm_memslot_iter_next(iter))
1088 * KVM_SET_USER_MEMORY_REGION ioctl allows the following operations:
1089 * - create a new memory slot
1090 * - delete an existing memory slot
1091 * - modify an existing memory slot
1092 * -- move it in the guest physical memory space
1093 * -- just change its flags
1095 * Since flags can be changed by some of these operations, the following
1096 * differentiation is the best we can do for __kvm_set_memory_region():
1098 enum kvm_mr_change {
1105 int kvm_set_memory_region(struct kvm *kvm,
1106 const struct kvm_userspace_memory_region *mem);
1107 int __kvm_set_memory_region(struct kvm *kvm,
1108 const struct kvm_userspace_memory_region *mem);
1109 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot);
1110 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen);
1111 int kvm_arch_prepare_memory_region(struct kvm *kvm,
1112 const struct kvm_memory_slot *old,
1113 struct kvm_memory_slot *new,
1114 enum kvm_mr_change change);
1115 void kvm_arch_commit_memory_region(struct kvm *kvm,
1116 struct kvm_memory_slot *old,
1117 const struct kvm_memory_slot *new,
1118 enum kvm_mr_change change);
1119 /* flush all memory translations */
1120 void kvm_arch_flush_shadow_all(struct kvm *kvm);
1121 /* flush memory translations pointing to 'slot' */
1122 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
1123 struct kvm_memory_slot *slot);
1125 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1126 struct page **pages, int nr_pages);
1128 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn);
1129 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn);
1130 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable);
1131 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot, gfn_t gfn);
1132 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot, gfn_t gfn,
1134 void kvm_release_page_clean(struct page *page);
1135 void kvm_release_page_dirty(struct page *page);
1137 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn);
1138 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1140 kvm_pfn_t gfn_to_pfn_memslot(const struct kvm_memory_slot *slot, gfn_t gfn);
1141 kvm_pfn_t gfn_to_pfn_memslot_atomic(const struct kvm_memory_slot *slot, gfn_t gfn);
1142 kvm_pfn_t __gfn_to_pfn_memslot(const struct kvm_memory_slot *slot, gfn_t gfn,
1143 bool atomic, bool *async, bool write_fault,
1144 bool *writable, hva_t *hva);
1146 void kvm_release_pfn_clean(kvm_pfn_t pfn);
1147 void kvm_release_pfn_dirty(kvm_pfn_t pfn);
1148 void kvm_set_pfn_dirty(kvm_pfn_t pfn);
1149 void kvm_set_pfn_accessed(kvm_pfn_t pfn);
1151 void kvm_release_pfn(kvm_pfn_t pfn, bool dirty);
1152 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1154 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len);
1155 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1156 void *data, unsigned long len);
1157 int kvm_read_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1158 void *data, unsigned int offset,
1160 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1161 int offset, int len);
1162 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1164 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1165 void *data, unsigned long len);
1166 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1167 void *data, unsigned int offset,
1169 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1170 gpa_t gpa, unsigned long len);
1172 #define __kvm_get_guest(kvm, gfn, offset, v) \
1174 unsigned long __addr = gfn_to_hva(kvm, gfn); \
1175 typeof(v) __user *__uaddr = (typeof(__uaddr))(__addr + offset); \
1176 int __ret = -EFAULT; \
1178 if (!kvm_is_error_hva(__addr)) \
1179 __ret = get_user(v, __uaddr); \
1183 #define kvm_get_guest(kvm, gpa, v) \
1185 gpa_t __gpa = gpa; \
1186 struct kvm *__kvm = kvm; \
1188 __kvm_get_guest(__kvm, __gpa >> PAGE_SHIFT, \
1189 offset_in_page(__gpa), v); \
1192 #define __kvm_put_guest(kvm, gfn, offset, v) \
1194 unsigned long __addr = gfn_to_hva(kvm, gfn); \
1195 typeof(v) __user *__uaddr = (typeof(__uaddr))(__addr + offset); \
1196 int __ret = -EFAULT; \
1198 if (!kvm_is_error_hva(__addr)) \
1199 __ret = put_user(v, __uaddr); \
1201 mark_page_dirty(kvm, gfn); \
1205 #define kvm_put_guest(kvm, gpa, v) \
1207 gpa_t __gpa = gpa; \
1208 struct kvm *__kvm = kvm; \
1210 __kvm_put_guest(__kvm, __gpa >> PAGE_SHIFT, \
1211 offset_in_page(__gpa), v); \
1214 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len);
1215 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn);
1216 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn);
1217 bool kvm_vcpu_is_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn);
1218 unsigned long kvm_host_page_size(struct kvm_vcpu *vcpu, gfn_t gfn);
1219 void mark_page_dirty_in_slot(struct kvm *kvm, const struct kvm_memory_slot *memslot, gfn_t gfn);
1220 void mark_page_dirty(struct kvm *kvm, gfn_t gfn);
1222 struct kvm_memslots *kvm_vcpu_memslots(struct kvm_vcpu *vcpu);
1223 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn);
1224 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn);
1225 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn);
1226 int kvm_vcpu_map(struct kvm_vcpu *vcpu, gpa_t gpa, struct kvm_host_map *map);
1227 void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map, bool dirty);
1228 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn);
1229 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable);
1230 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data, int offset,
1232 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa, void *data,
1234 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data,
1236 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, const void *data,
1237 int offset, int len);
1238 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
1240 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn);
1243 * kvm_gfn_to_pfn_cache_init - prepare a cached kernel mapping and HPA for a
1244 * given guest physical address.
1246 * @kvm: pointer to kvm instance.
1247 * @gpc: struct gfn_to_pfn_cache object.
1248 * @vcpu: vCPU to be used for marking pages dirty and to be woken on
1250 * @usage: indicates if the resulting host physical PFN is used while
1251 * the @vcpu is IN_GUEST_MODE (in which case invalidation of
1252 * the cache from MMU notifiers---but not for KVM memslot
1253 * changes!---will also force @vcpu to exit the guest and
1254 * refresh the cache); and/or if the PFN used directly
1255 * by KVM (and thus needs a kernel virtual mapping).
1256 * @gpa: guest physical address to map.
1257 * @len: sanity check; the range being access must fit a single page.
1259 * @return: 0 for success.
1260 * -EINVAL for a mapping which would cross a page boundary.
1261 * -EFAULT for an untranslatable guest physical address.
1263 * This primes a gfn_to_pfn_cache and links it into the @kvm's list for
1264 * invalidations to be processed. Callers are required to use
1265 * kvm_gfn_to_pfn_cache_check() to ensure that the cache is valid before
1266 * accessing the target page.
1268 int kvm_gfn_to_pfn_cache_init(struct kvm *kvm, struct gfn_to_pfn_cache *gpc,
1269 struct kvm_vcpu *vcpu, enum pfn_cache_usage usage,
1270 gpa_t gpa, unsigned long len);
1273 * kvm_gfn_to_pfn_cache_check - check validity of a gfn_to_pfn_cache.
1275 * @kvm: pointer to kvm instance.
1276 * @gpc: struct gfn_to_pfn_cache object.
1277 * @gpa: current guest physical address to map.
1278 * @len: sanity check; the range being access must fit a single page.
1280 * @return: %true if the cache is still valid and the address matches.
1281 * %false if the cache is not valid.
1283 * Callers outside IN_GUEST_MODE context should hold a read lock on @gpc->lock
1284 * while calling this function, and then continue to hold the lock until the
1285 * access is complete.
1287 * Callers in IN_GUEST_MODE may do so without locking, although they should
1288 * still hold a read lock on kvm->scru for the memslot checks.
1290 bool kvm_gfn_to_pfn_cache_check(struct kvm *kvm, struct gfn_to_pfn_cache *gpc,
1291 gpa_t gpa, unsigned long len);
1294 * kvm_gfn_to_pfn_cache_refresh - update a previously initialized cache.
1296 * @kvm: pointer to kvm instance.
1297 * @gpc: struct gfn_to_pfn_cache object.
1298 * @gpa: updated guest physical address to map.
1299 * @len: sanity check; the range being access must fit a single page.
1301 * @return: 0 for success.
1302 * -EINVAL for a mapping which would cross a page boundary.
1303 * -EFAULT for an untranslatable guest physical address.
1305 * This will attempt to refresh a gfn_to_pfn_cache. Note that a successful
1306 * returm from this function does not mean the page can be immediately
1307 * accessed because it may have raced with an invalidation. Callers must
1308 * still lock and check the cache status, as this function does not return
1309 * with the lock still held to permit access.
1311 int kvm_gfn_to_pfn_cache_refresh(struct kvm *kvm, struct gfn_to_pfn_cache *gpc,
1312 gpa_t gpa, unsigned long len);
1315 * kvm_gfn_to_pfn_cache_unmap - temporarily unmap a gfn_to_pfn_cache.
1317 * @kvm: pointer to kvm instance.
1318 * @gpc: struct gfn_to_pfn_cache object.
1320 * This unmaps the referenced page. The cache is left in the invalid state
1321 * but at least the mapping from GPA to userspace HVA will remain cached
1322 * and can be reused on a subsequent refresh.
1324 void kvm_gfn_to_pfn_cache_unmap(struct kvm *kvm, struct gfn_to_pfn_cache *gpc);
1327 * kvm_gfn_to_pfn_cache_destroy - destroy and unlink a gfn_to_pfn_cache.
1329 * @kvm: pointer to kvm instance.
1330 * @gpc: struct gfn_to_pfn_cache object.
1332 * This removes a cache from the @kvm's list to be processed on MMU notifier
1335 void kvm_gfn_to_pfn_cache_destroy(struct kvm *kvm, struct gfn_to_pfn_cache *gpc);
1337 void kvm_sigset_activate(struct kvm_vcpu *vcpu);
1338 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu);
1340 void kvm_vcpu_halt(struct kvm_vcpu *vcpu);
1341 bool kvm_vcpu_block(struct kvm_vcpu *vcpu);
1342 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu);
1343 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu);
1344 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu);
1345 void kvm_vcpu_kick(struct kvm_vcpu *vcpu);
1346 int kvm_vcpu_yield_to(struct kvm_vcpu *target);
1347 void kvm_vcpu_on_spin(struct kvm_vcpu *vcpu, bool usermode_vcpu_not_eligible);
1349 void kvm_flush_remote_tlbs(struct kvm *kvm);
1351 #ifdef KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE
1352 int kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache *mc, int min);
1353 int __kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache *mc, int capacity, int min);
1354 int kvm_mmu_memory_cache_nr_free_objects(struct kvm_mmu_memory_cache *mc);
1355 void kvm_mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc);
1356 void *kvm_mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc);
1359 void kvm_mmu_invalidate_begin(struct kvm *kvm, unsigned long start,
1361 void kvm_mmu_invalidate_end(struct kvm *kvm, unsigned long start,
1364 long kvm_arch_dev_ioctl(struct file *filp,
1365 unsigned int ioctl, unsigned long arg);
1366 long kvm_arch_vcpu_ioctl(struct file *filp,
1367 unsigned int ioctl, unsigned long arg);
1368 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf);
1370 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext);
1372 void kvm_arch_mmu_enable_log_dirty_pt_masked(struct kvm *kvm,
1373 struct kvm_memory_slot *slot,
1375 unsigned long mask);
1376 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot);
1378 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1379 void kvm_arch_flush_remote_tlbs_memslot(struct kvm *kvm,
1380 const struct kvm_memory_slot *memslot);
1381 #else /* !CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
1382 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log);
1383 int kvm_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log,
1384 int *is_dirty, struct kvm_memory_slot **memslot);
1387 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
1389 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
1390 struct kvm_enable_cap *cap);
1391 long kvm_arch_vm_ioctl(struct file *filp,
1392 unsigned int ioctl, unsigned long arg);
1394 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu);
1395 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu);
1397 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
1398 struct kvm_translation *tr);
1400 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs);
1401 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs);
1402 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
1403 struct kvm_sregs *sregs);
1404 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
1405 struct kvm_sregs *sregs);
1406 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
1407 struct kvm_mp_state *mp_state);
1408 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
1409 struct kvm_mp_state *mp_state);
1410 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
1411 struct kvm_guest_debug *dbg);
1412 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu);
1414 int kvm_arch_init(void *opaque);
1415 void kvm_arch_exit(void);
1417 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu);
1419 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu);
1420 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu);
1421 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id);
1422 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu);
1423 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu);
1424 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu);
1426 #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER
1427 int kvm_arch_pm_notifier(struct kvm *kvm, unsigned long state);
1430 #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS
1431 void kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu, struct dentry *debugfs_dentry);
1433 static inline void kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu) {}
1436 int kvm_arch_hardware_enable(void);
1437 void kvm_arch_hardware_disable(void);
1438 int kvm_arch_hardware_setup(void *opaque);
1439 void kvm_arch_hardware_unsetup(void);
1440 int kvm_arch_check_processor_compat(void *opaque);
1441 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu);
1442 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu);
1443 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu);
1444 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu);
1445 bool kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu);
1446 int kvm_arch_post_init_vm(struct kvm *kvm);
1447 void kvm_arch_pre_destroy_vm(struct kvm *kvm);
1448 int kvm_arch_create_vm_debugfs(struct kvm *kvm);
1450 #ifndef __KVM_HAVE_ARCH_VM_ALLOC
1452 * All architectures that want to use vzalloc currently also
1453 * need their own kvm_arch_alloc_vm implementation.
1455 static inline struct kvm *kvm_arch_alloc_vm(void)
1457 return kzalloc(sizeof(struct kvm), GFP_KERNEL);
1461 static inline void __kvm_arch_free_vm(struct kvm *kvm)
1466 #ifndef __KVM_HAVE_ARCH_VM_FREE
1467 static inline void kvm_arch_free_vm(struct kvm *kvm)
1469 __kvm_arch_free_vm(kvm);
1473 #ifndef __KVM_HAVE_ARCH_FLUSH_REMOTE_TLB
1474 static inline int kvm_arch_flush_remote_tlb(struct kvm *kvm)
1480 #ifdef __KVM_HAVE_ARCH_NONCOHERENT_DMA
1481 void kvm_arch_register_noncoherent_dma(struct kvm *kvm);
1482 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm);
1483 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm);
1485 static inline void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
1489 static inline void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
1493 static inline bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
1498 #ifdef __KVM_HAVE_ARCH_ASSIGNED_DEVICE
1499 void kvm_arch_start_assignment(struct kvm *kvm);
1500 void kvm_arch_end_assignment(struct kvm *kvm);
1501 bool kvm_arch_has_assigned_device(struct kvm *kvm);
1503 static inline void kvm_arch_start_assignment(struct kvm *kvm)
1507 static inline void kvm_arch_end_assignment(struct kvm *kvm)
1511 static __always_inline bool kvm_arch_has_assigned_device(struct kvm *kvm)
1517 static inline struct rcuwait *kvm_arch_vcpu_get_wait(struct kvm_vcpu *vcpu)
1519 #ifdef __KVM_HAVE_ARCH_WQP
1520 return vcpu->arch.waitp;
1527 * Wake a vCPU if necessary, but don't do any stats/metadata updates. Returns
1528 * true if the vCPU was blocking and was awakened, false otherwise.
1530 static inline bool __kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
1532 return !!rcuwait_wake_up(kvm_arch_vcpu_get_wait(vcpu));
1535 static inline bool kvm_vcpu_is_blocking(struct kvm_vcpu *vcpu)
1537 return rcuwait_active(kvm_arch_vcpu_get_wait(vcpu));
1540 #ifdef __KVM_HAVE_ARCH_INTC_INITIALIZED
1542 * returns true if the virtual interrupt controller is initialized and
1543 * ready to accept virtual IRQ. On some architectures the virtual interrupt
1544 * controller is dynamically instantiated and this is not always true.
1546 bool kvm_arch_intc_initialized(struct kvm *kvm);
1548 static inline bool kvm_arch_intc_initialized(struct kvm *kvm)
1554 #ifdef CONFIG_GUEST_PERF_EVENTS
1555 unsigned long kvm_arch_vcpu_get_ip(struct kvm_vcpu *vcpu);
1557 void kvm_register_perf_callbacks(unsigned int (*pt_intr_handler)(void));
1558 void kvm_unregister_perf_callbacks(void);
1560 static inline void kvm_register_perf_callbacks(void *ign) {}
1561 static inline void kvm_unregister_perf_callbacks(void) {}
1562 #endif /* CONFIG_GUEST_PERF_EVENTS */
1564 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type);
1565 void kvm_arch_destroy_vm(struct kvm *kvm);
1566 void kvm_arch_sync_events(struct kvm *kvm);
1568 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu);
1570 struct page *kvm_pfn_to_refcounted_page(kvm_pfn_t pfn);
1571 bool kvm_is_zone_device_page(struct page *page);
1573 struct kvm_irq_ack_notifier {
1574 struct hlist_node link;
1576 void (*irq_acked)(struct kvm_irq_ack_notifier *kian);
1579 int kvm_irq_map_gsi(struct kvm *kvm,
1580 struct kvm_kernel_irq_routing_entry *entries, int gsi);
1581 int kvm_irq_map_chip_pin(struct kvm *kvm, unsigned irqchip, unsigned pin);
1583 int kvm_set_irq(struct kvm *kvm, int irq_source_id, u32 irq, int level,
1585 int kvm_set_msi(struct kvm_kernel_irq_routing_entry *irq_entry, struct kvm *kvm,
1586 int irq_source_id, int level, bool line_status);
1587 int kvm_arch_set_irq_inatomic(struct kvm_kernel_irq_routing_entry *e,
1588 struct kvm *kvm, int irq_source_id,
1589 int level, bool line_status);
1590 bool kvm_irq_has_notifier(struct kvm *kvm, unsigned irqchip, unsigned pin);
1591 void kvm_notify_acked_gsi(struct kvm *kvm, int gsi);
1592 void kvm_notify_acked_irq(struct kvm *kvm, unsigned irqchip, unsigned pin);
1593 void kvm_register_irq_ack_notifier(struct kvm *kvm,
1594 struct kvm_irq_ack_notifier *kian);
1595 void kvm_unregister_irq_ack_notifier(struct kvm *kvm,
1596 struct kvm_irq_ack_notifier *kian);
1597 int kvm_request_irq_source_id(struct kvm *kvm);
1598 void kvm_free_irq_source_id(struct kvm *kvm, int irq_source_id);
1599 bool kvm_arch_irqfd_allowed(struct kvm *kvm, struct kvm_irqfd *args);
1602 * Returns a pointer to the memslot if it contains gfn.
1603 * Otherwise returns NULL.
1605 static inline struct kvm_memory_slot *
1606 try_get_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1611 if (gfn >= slot->base_gfn && gfn < slot->base_gfn + slot->npages)
1618 * Returns a pointer to the memslot that contains gfn. Otherwise returns NULL.
1620 * With "approx" set returns the memslot also when the address falls
1621 * in a hole. In that case one of the memslots bordering the hole is
1624 static inline struct kvm_memory_slot *
1625 search_memslots(struct kvm_memslots *slots, gfn_t gfn, bool approx)
1627 struct kvm_memory_slot *slot;
1628 struct rb_node *node;
1629 int idx = slots->node_idx;
1632 for (node = slots->gfn_tree.rb_node; node; ) {
1633 slot = container_of(node, struct kvm_memory_slot, gfn_node[idx]);
1634 if (gfn >= slot->base_gfn) {
1635 if (gfn < slot->base_gfn + slot->npages)
1637 node = node->rb_right;
1639 node = node->rb_left;
1642 return approx ? slot : NULL;
1645 static inline struct kvm_memory_slot *
1646 ____gfn_to_memslot(struct kvm_memslots *slots, gfn_t gfn, bool approx)
1648 struct kvm_memory_slot *slot;
1650 slot = (struct kvm_memory_slot *)atomic_long_read(&slots->last_used_slot);
1651 slot = try_get_memslot(slot, gfn);
1655 slot = search_memslots(slots, gfn, approx);
1657 atomic_long_set(&slots->last_used_slot, (unsigned long)slot);
1665 * __gfn_to_memslot() and its descendants are here to allow arch code to inline
1666 * the lookups in hot paths. gfn_to_memslot() itself isn't here as an inline
1667 * because that would bloat other code too much.
1669 static inline struct kvm_memory_slot *
1670 __gfn_to_memslot(struct kvm_memslots *slots, gfn_t gfn)
1672 return ____gfn_to_memslot(slots, gfn, false);
1675 static inline unsigned long
1676 __gfn_to_hva_memslot(const struct kvm_memory_slot *slot, gfn_t gfn)
1679 * The index was checked originally in search_memslots. To avoid
1680 * that a malicious guest builds a Spectre gadget out of e.g. page
1681 * table walks, do not let the processor speculate loads outside
1682 * the guest's registered memslots.
1684 unsigned long offset = gfn - slot->base_gfn;
1685 offset = array_index_nospec(offset, slot->npages);
1686 return slot->userspace_addr + offset * PAGE_SIZE;
1689 static inline int memslot_id(struct kvm *kvm, gfn_t gfn)
1691 return gfn_to_memslot(kvm, gfn)->id;
1695 hva_to_gfn_memslot(unsigned long hva, struct kvm_memory_slot *slot)
1697 gfn_t gfn_offset = (hva - slot->userspace_addr) >> PAGE_SHIFT;
1699 return slot->base_gfn + gfn_offset;
1702 static inline gpa_t gfn_to_gpa(gfn_t gfn)
1704 return (gpa_t)gfn << PAGE_SHIFT;
1707 static inline gfn_t gpa_to_gfn(gpa_t gpa)
1709 return (gfn_t)(gpa >> PAGE_SHIFT);
1712 static inline hpa_t pfn_to_hpa(kvm_pfn_t pfn)
1714 return (hpa_t)pfn << PAGE_SHIFT;
1717 static inline bool kvm_is_error_gpa(struct kvm *kvm, gpa_t gpa)
1719 unsigned long hva = gfn_to_hva(kvm, gpa_to_gfn(gpa));
1721 return kvm_is_error_hva(hva);
1724 enum kvm_stat_kind {
1729 struct kvm_stat_data {
1731 const struct _kvm_stats_desc *desc;
1732 enum kvm_stat_kind kind;
1735 struct _kvm_stats_desc {
1736 struct kvm_stats_desc desc;
1737 char name[KVM_STATS_NAME_SIZE];
1740 #define STATS_DESC_COMMON(type, unit, base, exp, sz, bsz) \
1741 .flags = type | unit | base | \
1742 BUILD_BUG_ON_ZERO(type & ~KVM_STATS_TYPE_MASK) | \
1743 BUILD_BUG_ON_ZERO(unit & ~KVM_STATS_UNIT_MASK) | \
1744 BUILD_BUG_ON_ZERO(base & ~KVM_STATS_BASE_MASK), \
1749 #define VM_GENERIC_STATS_DESC(stat, type, unit, base, exp, sz, bsz) \
1752 STATS_DESC_COMMON(type, unit, base, exp, sz, bsz), \
1753 .offset = offsetof(struct kvm_vm_stat, generic.stat) \
1757 #define VCPU_GENERIC_STATS_DESC(stat, type, unit, base, exp, sz, bsz) \
1760 STATS_DESC_COMMON(type, unit, base, exp, sz, bsz), \
1761 .offset = offsetof(struct kvm_vcpu_stat, generic.stat) \
1765 #define VM_STATS_DESC(stat, type, unit, base, exp, sz, bsz) \
1768 STATS_DESC_COMMON(type, unit, base, exp, sz, bsz), \
1769 .offset = offsetof(struct kvm_vm_stat, stat) \
1773 #define VCPU_STATS_DESC(stat, type, unit, base, exp, sz, bsz) \
1776 STATS_DESC_COMMON(type, unit, base, exp, sz, bsz), \
1777 .offset = offsetof(struct kvm_vcpu_stat, stat) \
1781 /* SCOPE: VM, VM_GENERIC, VCPU, VCPU_GENERIC */
1782 #define STATS_DESC(SCOPE, stat, type, unit, base, exp, sz, bsz) \
1783 SCOPE##_STATS_DESC(stat, type, unit, base, exp, sz, bsz)
1785 #define STATS_DESC_CUMULATIVE(SCOPE, name, unit, base, exponent) \
1786 STATS_DESC(SCOPE, name, KVM_STATS_TYPE_CUMULATIVE, \
1787 unit, base, exponent, 1, 0)
1788 #define STATS_DESC_INSTANT(SCOPE, name, unit, base, exponent) \
1789 STATS_DESC(SCOPE, name, KVM_STATS_TYPE_INSTANT, \
1790 unit, base, exponent, 1, 0)
1791 #define STATS_DESC_PEAK(SCOPE, name, unit, base, exponent) \
1792 STATS_DESC(SCOPE, name, KVM_STATS_TYPE_PEAK, \
1793 unit, base, exponent, 1, 0)
1794 #define STATS_DESC_LINEAR_HIST(SCOPE, name, unit, base, exponent, sz, bsz) \
1795 STATS_DESC(SCOPE, name, KVM_STATS_TYPE_LINEAR_HIST, \
1796 unit, base, exponent, sz, bsz)
1797 #define STATS_DESC_LOG_HIST(SCOPE, name, unit, base, exponent, sz) \
1798 STATS_DESC(SCOPE, name, KVM_STATS_TYPE_LOG_HIST, \
1799 unit, base, exponent, sz, 0)
1801 /* Cumulative counter, read/write */
1802 #define STATS_DESC_COUNTER(SCOPE, name) \
1803 STATS_DESC_CUMULATIVE(SCOPE, name, KVM_STATS_UNIT_NONE, \
1804 KVM_STATS_BASE_POW10, 0)
1805 /* Instantaneous counter, read only */
1806 #define STATS_DESC_ICOUNTER(SCOPE, name) \
1807 STATS_DESC_INSTANT(SCOPE, name, KVM_STATS_UNIT_NONE, \
1808 KVM_STATS_BASE_POW10, 0)
1809 /* Peak counter, read/write */
1810 #define STATS_DESC_PCOUNTER(SCOPE, name) \
1811 STATS_DESC_PEAK(SCOPE, name, KVM_STATS_UNIT_NONE, \
1812 KVM_STATS_BASE_POW10, 0)
1814 /* Instantaneous boolean value, read only */
1815 #define STATS_DESC_IBOOLEAN(SCOPE, name) \
1816 STATS_DESC_INSTANT(SCOPE, name, KVM_STATS_UNIT_BOOLEAN, \
1817 KVM_STATS_BASE_POW10, 0)
1818 /* Peak (sticky) boolean value, read/write */
1819 #define STATS_DESC_PBOOLEAN(SCOPE, name) \
1820 STATS_DESC_PEAK(SCOPE, name, KVM_STATS_UNIT_BOOLEAN, \
1821 KVM_STATS_BASE_POW10, 0)
1823 /* Cumulative time in nanosecond */
1824 #define STATS_DESC_TIME_NSEC(SCOPE, name) \
1825 STATS_DESC_CUMULATIVE(SCOPE, name, KVM_STATS_UNIT_SECONDS, \
1826 KVM_STATS_BASE_POW10, -9)
1827 /* Linear histogram for time in nanosecond */
1828 #define STATS_DESC_LINHIST_TIME_NSEC(SCOPE, name, sz, bsz) \
1829 STATS_DESC_LINEAR_HIST(SCOPE, name, KVM_STATS_UNIT_SECONDS, \
1830 KVM_STATS_BASE_POW10, -9, sz, bsz)
1831 /* Logarithmic histogram for time in nanosecond */
1832 #define STATS_DESC_LOGHIST_TIME_NSEC(SCOPE, name, sz) \
1833 STATS_DESC_LOG_HIST(SCOPE, name, KVM_STATS_UNIT_SECONDS, \
1834 KVM_STATS_BASE_POW10, -9, sz)
1836 #define KVM_GENERIC_VM_STATS() \
1837 STATS_DESC_COUNTER(VM_GENERIC, remote_tlb_flush), \
1838 STATS_DESC_COUNTER(VM_GENERIC, remote_tlb_flush_requests)
1840 #define KVM_GENERIC_VCPU_STATS() \
1841 STATS_DESC_COUNTER(VCPU_GENERIC, halt_successful_poll), \
1842 STATS_DESC_COUNTER(VCPU_GENERIC, halt_attempted_poll), \
1843 STATS_DESC_COUNTER(VCPU_GENERIC, halt_poll_invalid), \
1844 STATS_DESC_COUNTER(VCPU_GENERIC, halt_wakeup), \
1845 STATS_DESC_TIME_NSEC(VCPU_GENERIC, halt_poll_success_ns), \
1846 STATS_DESC_TIME_NSEC(VCPU_GENERIC, halt_poll_fail_ns), \
1847 STATS_DESC_TIME_NSEC(VCPU_GENERIC, halt_wait_ns), \
1848 STATS_DESC_LOGHIST_TIME_NSEC(VCPU_GENERIC, halt_poll_success_hist, \
1849 HALT_POLL_HIST_COUNT), \
1850 STATS_DESC_LOGHIST_TIME_NSEC(VCPU_GENERIC, halt_poll_fail_hist, \
1851 HALT_POLL_HIST_COUNT), \
1852 STATS_DESC_LOGHIST_TIME_NSEC(VCPU_GENERIC, halt_wait_hist, \
1853 HALT_POLL_HIST_COUNT), \
1854 STATS_DESC_IBOOLEAN(VCPU_GENERIC, blocking)
1856 extern struct dentry *kvm_debugfs_dir;
1858 ssize_t kvm_stats_read(char *id, const struct kvm_stats_header *header,
1859 const struct _kvm_stats_desc *desc,
1860 void *stats, size_t size_stats,
1861 char __user *user_buffer, size_t size, loff_t *offset);
1864 * kvm_stats_linear_hist_update() - Update bucket value for linear histogram
1867 * @data: start address of the stats data
1868 * @size: the number of bucket of the stats data
1869 * @value: the new value used to update the linear histogram's bucket
1870 * @bucket_size: the size (width) of a bucket
1872 static inline void kvm_stats_linear_hist_update(u64 *data, size_t size,
1873 u64 value, size_t bucket_size)
1875 size_t index = div64_u64(value, bucket_size);
1877 index = min(index, size - 1);
1882 * kvm_stats_log_hist_update() - Update bucket value for logarithmic histogram
1885 * @data: start address of the stats data
1886 * @size: the number of bucket of the stats data
1887 * @value: the new value used to update the logarithmic histogram's bucket
1889 static inline void kvm_stats_log_hist_update(u64 *data, size_t size, u64 value)
1891 size_t index = fls64(value);
1893 index = min(index, size - 1);
1897 #define KVM_STATS_LINEAR_HIST_UPDATE(array, value, bsize) \
1898 kvm_stats_linear_hist_update(array, ARRAY_SIZE(array), value, bsize)
1899 #define KVM_STATS_LOG_HIST_UPDATE(array, value) \
1900 kvm_stats_log_hist_update(array, ARRAY_SIZE(array), value)
1903 extern const struct kvm_stats_header kvm_vm_stats_header;
1904 extern const struct _kvm_stats_desc kvm_vm_stats_desc[];
1905 extern const struct kvm_stats_header kvm_vcpu_stats_header;
1906 extern const struct _kvm_stats_desc kvm_vcpu_stats_desc[];
1908 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1909 static inline int mmu_invalidate_retry(struct kvm *kvm, unsigned long mmu_seq)
1911 if (unlikely(kvm->mmu_invalidate_in_progress))
1914 * Ensure the read of mmu_invalidate_in_progress happens before
1915 * the read of mmu_invalidate_seq. This interacts with the
1916 * smp_wmb() in mmu_notifier_invalidate_range_end to make sure
1917 * that the caller either sees the old (non-zero) value of
1918 * mmu_invalidate_in_progress or the new (incremented) value of
1919 * mmu_invalidate_seq.
1921 * PowerPC Book3s HV KVM calls this under a per-page lock rather
1922 * than under kvm->mmu_lock, for scalability, so can't rely on
1923 * kvm->mmu_lock to keep things ordered.
1926 if (kvm->mmu_invalidate_seq != mmu_seq)
1931 static inline int mmu_invalidate_retry_hva(struct kvm *kvm,
1932 unsigned long mmu_seq,
1935 lockdep_assert_held(&kvm->mmu_lock);
1937 * If mmu_invalidate_in_progress is non-zero, then the range maintained
1938 * by kvm_mmu_notifier_invalidate_range_start contains all addresses
1939 * that might be being invalidated. Note that it may include some false
1940 * positives, due to shortcuts when handing concurrent invalidations.
1942 if (unlikely(kvm->mmu_invalidate_in_progress) &&
1943 hva >= kvm->mmu_invalidate_range_start &&
1944 hva < kvm->mmu_invalidate_range_end)
1946 if (kvm->mmu_invalidate_seq != mmu_seq)
1952 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
1954 #define KVM_MAX_IRQ_ROUTES 4096 /* might need extension/rework in the future */
1956 bool kvm_arch_can_set_irq_routing(struct kvm *kvm);
1957 int kvm_set_irq_routing(struct kvm *kvm,
1958 const struct kvm_irq_routing_entry *entries,
1961 int kvm_set_routing_entry(struct kvm *kvm,
1962 struct kvm_kernel_irq_routing_entry *e,
1963 const struct kvm_irq_routing_entry *ue);
1964 void kvm_free_irq_routing(struct kvm *kvm);
1968 static inline void kvm_free_irq_routing(struct kvm *kvm) {}
1972 int kvm_send_userspace_msi(struct kvm *kvm, struct kvm_msi *msi);
1974 #ifdef CONFIG_HAVE_KVM_EVENTFD
1976 void kvm_eventfd_init(struct kvm *kvm);
1977 int kvm_ioeventfd(struct kvm *kvm, struct kvm_ioeventfd *args);
1979 #ifdef CONFIG_HAVE_KVM_IRQFD
1980 int kvm_irqfd(struct kvm *kvm, struct kvm_irqfd *args);
1981 void kvm_irqfd_release(struct kvm *kvm);
1982 void kvm_irq_routing_update(struct kvm *);
1984 static inline int kvm_irqfd(struct kvm *kvm, struct kvm_irqfd *args)
1989 static inline void kvm_irqfd_release(struct kvm *kvm) {}
1994 static inline void kvm_eventfd_init(struct kvm *kvm) {}
1996 static inline int kvm_irqfd(struct kvm *kvm, struct kvm_irqfd *args)
2001 static inline void kvm_irqfd_release(struct kvm *kvm) {}
2003 #ifdef CONFIG_HAVE_KVM_IRQCHIP
2004 static inline void kvm_irq_routing_update(struct kvm *kvm)
2009 static inline int kvm_ioeventfd(struct kvm *kvm, struct kvm_ioeventfd *args)
2014 #endif /* CONFIG_HAVE_KVM_EVENTFD */
2016 void kvm_arch_irq_routing_update(struct kvm *kvm);
2018 static inline void __kvm_make_request(int req, struct kvm_vcpu *vcpu)
2021 * Ensure the rest of the request is published to kvm_check_request's
2022 * caller. Paired with the smp_mb__after_atomic in kvm_check_request.
2025 set_bit(req & KVM_REQUEST_MASK, (void *)&vcpu->requests);
2028 static __always_inline void kvm_make_request(int req, struct kvm_vcpu *vcpu)
2031 * Request that don't require vCPU action should never be logged in
2032 * vcpu->requests. The vCPU won't clear the request, so it will stay
2033 * logged indefinitely and prevent the vCPU from entering the guest.
2035 BUILD_BUG_ON(!__builtin_constant_p(req) ||
2036 (req & KVM_REQUEST_NO_ACTION));
2038 __kvm_make_request(req, vcpu);
2041 static inline bool kvm_request_pending(struct kvm_vcpu *vcpu)
2043 return READ_ONCE(vcpu->requests);
2046 static inline bool kvm_test_request(int req, struct kvm_vcpu *vcpu)
2048 return test_bit(req & KVM_REQUEST_MASK, (void *)&vcpu->requests);
2051 static inline void kvm_clear_request(int req, struct kvm_vcpu *vcpu)
2053 clear_bit(req & KVM_REQUEST_MASK, (void *)&vcpu->requests);
2056 static inline bool kvm_check_request(int req, struct kvm_vcpu *vcpu)
2058 if (kvm_test_request(req, vcpu)) {
2059 kvm_clear_request(req, vcpu);
2062 * Ensure the rest of the request is visible to kvm_check_request's
2063 * caller. Paired with the smp_wmb in kvm_make_request.
2065 smp_mb__after_atomic();
2072 extern bool kvm_rebooting;
2074 extern unsigned int halt_poll_ns;
2075 extern unsigned int halt_poll_ns_grow;
2076 extern unsigned int halt_poll_ns_grow_start;
2077 extern unsigned int halt_poll_ns_shrink;
2080 const struct kvm_device_ops *ops;
2083 struct list_head vm_node;
2086 /* create, destroy, and name are mandatory */
2087 struct kvm_device_ops {
2091 * create is called holding kvm->lock and any operations not suitable
2092 * to do while holding the lock should be deferred to init (see
2095 int (*create)(struct kvm_device *dev, u32 type);
2098 * init is called after create if create is successful and is called
2099 * outside of holding kvm->lock.
2101 void (*init)(struct kvm_device *dev);
2104 * Destroy is responsible for freeing dev.
2106 * Destroy may be called before or after destructors are called
2107 * on emulated I/O regions, depending on whether a reference is
2108 * held by a vcpu or other kvm component that gets destroyed
2109 * after the emulated I/O.
2111 void (*destroy)(struct kvm_device *dev);
2114 * Release is an alternative method to free the device. It is
2115 * called when the device file descriptor is closed. Once
2116 * release is called, the destroy method will not be called
2117 * anymore as the device is removed from the device list of
2118 * the VM. kvm->lock is held.
2120 void (*release)(struct kvm_device *dev);
2122 int (*set_attr)(struct kvm_device *dev, struct kvm_device_attr *attr);
2123 int (*get_attr)(struct kvm_device *dev, struct kvm_device_attr *attr);
2124 int (*has_attr)(struct kvm_device *dev, struct kvm_device_attr *attr);
2125 long (*ioctl)(struct kvm_device *dev, unsigned int ioctl,
2127 int (*mmap)(struct kvm_device *dev, struct vm_area_struct *vma);
2130 void kvm_device_get(struct kvm_device *dev);
2131 void kvm_device_put(struct kvm_device *dev);
2132 struct kvm_device *kvm_device_from_filp(struct file *filp);
2133 int kvm_register_device_ops(const struct kvm_device_ops *ops, u32 type);
2134 void kvm_unregister_device_ops(u32 type);
2136 extern struct kvm_device_ops kvm_mpic_ops;
2137 extern struct kvm_device_ops kvm_arm_vgic_v2_ops;
2138 extern struct kvm_device_ops kvm_arm_vgic_v3_ops;
2140 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2142 static inline void kvm_vcpu_set_in_spin_loop(struct kvm_vcpu *vcpu, bool val)
2144 vcpu->spin_loop.in_spin_loop = val;
2146 static inline void kvm_vcpu_set_dy_eligible(struct kvm_vcpu *vcpu, bool val)
2148 vcpu->spin_loop.dy_eligible = val;
2151 #else /* !CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT */
2153 static inline void kvm_vcpu_set_in_spin_loop(struct kvm_vcpu *vcpu, bool val)
2157 static inline void kvm_vcpu_set_dy_eligible(struct kvm_vcpu *vcpu, bool val)
2160 #endif /* CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT */
2162 static inline bool kvm_is_visible_memslot(struct kvm_memory_slot *memslot)
2164 return (memslot && memslot->id < KVM_USER_MEM_SLOTS &&
2165 !(memslot->flags & KVM_MEMSLOT_INVALID));
2168 struct kvm_vcpu *kvm_get_running_vcpu(void);
2169 struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void);
2171 #ifdef CONFIG_HAVE_KVM_IRQ_BYPASS
2172 bool kvm_arch_has_irq_bypass(void);
2173 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *,
2174 struct irq_bypass_producer *);
2175 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *,
2176 struct irq_bypass_producer *);
2177 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *);
2178 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *);
2179 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
2180 uint32_t guest_irq, bool set);
2181 bool kvm_arch_irqfd_route_changed(struct kvm_kernel_irq_routing_entry *,
2182 struct kvm_kernel_irq_routing_entry *);
2183 #endif /* CONFIG_HAVE_KVM_IRQ_BYPASS */
2185 #ifdef CONFIG_HAVE_KVM_INVALID_WAKEUPS
2186 /* If we wakeup during the poll time, was it a sucessful poll? */
2187 static inline bool vcpu_valid_wakeup(struct kvm_vcpu *vcpu)
2189 return vcpu->valid_wakeup;
2193 static inline bool vcpu_valid_wakeup(struct kvm_vcpu *vcpu)
2197 #endif /* CONFIG_HAVE_KVM_INVALID_WAKEUPS */
2199 #ifdef CONFIG_HAVE_KVM_NO_POLL
2200 /* Callback that tells if we must not poll */
2201 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu);
2203 static inline bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
2207 #endif /* CONFIG_HAVE_KVM_NO_POLL */
2209 #ifdef CONFIG_HAVE_KVM_VCPU_ASYNC_IOCTL
2210 long kvm_arch_vcpu_async_ioctl(struct file *filp,
2211 unsigned int ioctl, unsigned long arg);
2213 static inline long kvm_arch_vcpu_async_ioctl(struct file *filp,
2217 return -ENOIOCTLCMD;
2219 #endif /* CONFIG_HAVE_KVM_VCPU_ASYNC_IOCTL */
2221 void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
2222 unsigned long start, unsigned long end);
2224 void kvm_arch_guest_memory_reclaimed(struct kvm *kvm);
2226 #ifdef CONFIG_HAVE_KVM_VCPU_RUN_PID_CHANGE
2227 int kvm_arch_vcpu_run_pid_change(struct kvm_vcpu *vcpu);
2229 static inline int kvm_arch_vcpu_run_pid_change(struct kvm_vcpu *vcpu)
2233 #endif /* CONFIG_HAVE_KVM_VCPU_RUN_PID_CHANGE */
2235 typedef int (*kvm_vm_thread_fn_t)(struct kvm *kvm, uintptr_t data);
2237 int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn,
2238 uintptr_t data, const char *name,
2239 struct task_struct **thread_ptr);
2241 #ifdef CONFIG_KVM_XFER_TO_GUEST_WORK
2242 static inline void kvm_handle_signal_exit(struct kvm_vcpu *vcpu)
2244 vcpu->run->exit_reason = KVM_EXIT_INTR;
2245 vcpu->stat.signal_exits++;
2247 #endif /* CONFIG_KVM_XFER_TO_GUEST_WORK */
2250 * If more than one page is being (un)accounted, @virt must be the address of
2251 * the first page of a block of pages what were allocated together (i.e
2252 * accounted together).
2254 * kvm_account_pgtable_pages() is thread-safe because mod_lruvec_page_state()
2257 static inline void kvm_account_pgtable_pages(void *virt, int nr)
2259 mod_lruvec_page_state(virt_to_page(virt), NR_SECONDARY_PAGETABLE, nr);
2263 * This defines how many reserved entries we want to keep before we
2264 * kick the vcpu to the userspace to avoid dirty ring full. This
2265 * value can be tuned to higher if e.g. PML is enabled on the host.
2267 #define KVM_DIRTY_RING_RSVD_ENTRIES 64
2269 /* Max number of entries allowed for each kvm dirty ring */
2270 #define KVM_DIRTY_RING_MAX_ENTRIES 65536