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 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)
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 #define KVM_ERR_PTR_BAD_PAGE (ERR_PTR(-ENOENT))
153 static inline bool is_error_page(struct page *page)
158 #define KVM_REQUEST_MASK GENMASK(7,0)
159 #define KVM_REQUEST_NO_WAKEUP BIT(8)
160 #define KVM_REQUEST_WAIT BIT(9)
161 #define KVM_REQUEST_NO_ACTION BIT(10)
163 * Architecture-independent vcpu->requests bit members
164 * Bits 3-7 are reserved for more arch-independent bits.
166 #define KVM_REQ_TLB_FLUSH (0 | KVM_REQUEST_WAIT | KVM_REQUEST_NO_WAKEUP)
167 #define KVM_REQ_VM_DEAD (1 | KVM_REQUEST_WAIT | KVM_REQUEST_NO_WAKEUP)
168 #define KVM_REQ_UNBLOCK 2
169 #define KVM_REQ_DIRTY_RING_SOFT_FULL 3
170 #define KVM_REQUEST_ARCH_BASE 8
173 * KVM_REQ_OUTSIDE_GUEST_MODE exists is purely as way to force the vCPU to
174 * OUTSIDE_GUEST_MODE. KVM_REQ_OUTSIDE_GUEST_MODE differs from a vCPU "kick"
175 * in that it ensures the vCPU has reached OUTSIDE_GUEST_MODE before continuing
176 * on. A kick only guarantees that the vCPU is on its way out, e.g. a previous
177 * kick may have set vcpu->mode to EXITING_GUEST_MODE, and so there's no
178 * guarantee the vCPU received an IPI and has actually exited guest mode.
180 #define KVM_REQ_OUTSIDE_GUEST_MODE (KVM_REQUEST_NO_ACTION | KVM_REQUEST_WAIT | KVM_REQUEST_NO_WAKEUP)
182 #define KVM_ARCH_REQ_FLAGS(nr, flags) ({ \
183 BUILD_BUG_ON((unsigned)(nr) >= (sizeof_field(struct kvm_vcpu, requests) * 8) - KVM_REQUEST_ARCH_BASE); \
184 (unsigned)(((nr) + KVM_REQUEST_ARCH_BASE) | (flags)); \
186 #define KVM_ARCH_REQ(nr) KVM_ARCH_REQ_FLAGS(nr, 0)
188 bool kvm_make_vcpus_request_mask(struct kvm *kvm, unsigned int req,
189 unsigned long *vcpu_bitmap);
190 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req);
191 bool kvm_make_all_cpus_request_except(struct kvm *kvm, unsigned int req,
192 struct kvm_vcpu *except);
193 bool kvm_make_cpus_request_mask(struct kvm *kvm, unsigned int req,
194 unsigned long *vcpu_bitmap);
196 #define KVM_USERSPACE_IRQ_SOURCE_ID 0
197 #define KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID 1
199 extern struct mutex kvm_lock;
200 extern struct list_head vm_list;
202 struct kvm_io_range {
205 struct kvm_io_device *dev;
208 #define NR_IOBUS_DEVS 1000
213 struct kvm_io_range range[];
219 KVM_VIRTIO_CCW_NOTIFY_BUS,
224 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
225 int len, const void *val);
226 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
227 gpa_t addr, int len, const void *val, long cookie);
228 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
230 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
231 int len, struct kvm_io_device *dev);
232 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
233 struct kvm_io_device *dev);
234 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
237 #ifdef CONFIG_KVM_ASYNC_PF
238 struct kvm_async_pf {
239 struct work_struct work;
240 struct list_head link;
241 struct list_head queue;
242 struct kvm_vcpu *vcpu;
243 struct mm_struct *mm;
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 KVM_ARCH_WANT_MMU_NOTIFIER
259 struct kvm_gfn_range {
260 struct kvm_memory_slot *slot;
266 bool kvm_unmap_gfn_range(struct kvm *kvm, struct kvm_gfn_range *range);
267 bool kvm_age_gfn(struct kvm *kvm, struct kvm_gfn_range *range);
268 bool kvm_test_age_gfn(struct kvm *kvm, struct kvm_gfn_range *range);
269 bool kvm_set_spte_gfn(struct kvm *kvm, struct kvm_gfn_range *range);
276 READING_SHADOW_PAGE_TABLES,
279 #define KVM_UNMAPPED_PAGE ((void *) 0x500 + POISON_POINTER_DELTA)
281 struct kvm_host_map {
283 * Only valid if the 'pfn' is managed by the host kernel (i.e. There is
284 * a 'struct page' for it. When using mem= kernel parameter some memory
285 * can be used as guest memory but they are not managed by host
287 * If 'pfn' is not managed by the host kernel, this field is
288 * initialized to KVM_UNMAPPED_PAGE.
297 * Used to check if the mapping is valid or not. Never use 'kvm_host_map'
298 * directly to check for that.
300 static inline bool kvm_vcpu_mapped(struct kvm_host_map *map)
305 static inline bool kvm_vcpu_can_poll(ktime_t cur, ktime_t stop)
307 return single_task_running() && !need_resched() && ktime_before(cur, stop);
311 * Sometimes a large or cross-page mmio needs to be broken up into separate
312 * exits for userspace servicing.
314 struct kvm_mmio_fragment {
322 #ifdef CONFIG_PREEMPT_NOTIFIERS
323 struct preempt_notifier preempt_notifier;
326 int vcpu_id; /* id given by userspace at creation */
327 int vcpu_idx; /* index into kvm->vcpu_array */
328 int ____srcu_idx; /* Don't use this directly. You've been warned. */
329 #ifdef CONFIG_PROVE_RCU
334 unsigned long guest_debug;
339 #ifndef __KVM_HAVE_ARCH_WQP
342 struct pid __rcu *pid;
345 unsigned int halt_poll_ns;
348 #ifdef CONFIG_HAS_IOMEM
350 int mmio_read_completed;
352 int mmio_cur_fragment;
353 int mmio_nr_fragments;
354 struct kvm_mmio_fragment mmio_fragments[KVM_MAX_MMIO_FRAGMENTS];
357 #ifdef CONFIG_KVM_ASYNC_PF
360 struct list_head queue;
361 struct list_head done;
366 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
368 * Cpu relax intercept or pause loop exit optimization
369 * in_spin_loop: set when a vcpu does a pause loop exit
370 * or cpu relax intercepted.
371 * dy_eligible: indicates whether vcpu is eligible for directed yield.
380 struct kvm_vcpu_arch arch;
381 struct kvm_vcpu_stat stat;
382 char stats_id[KVM_STATS_NAME_SIZE];
383 struct kvm_dirty_ring dirty_ring;
386 * The most recently used memslot by this vCPU and the slots generation
387 * for which it is valid.
388 * No wraparound protection is needed since generations won't overflow in
389 * thousands of years, even assuming 1M memslot operations per second.
391 struct kvm_memory_slot *last_used_slot;
392 u64 last_used_slot_gen;
396 * Start accounting time towards a guest.
397 * Must be called before entering guest context.
399 static __always_inline void guest_timing_enter_irqoff(void)
402 * This is running in ioctl context so its safe to assume that it's the
403 * stime pending cputime to flush.
405 instrumentation_begin();
406 vtime_account_guest_enter();
407 instrumentation_end();
411 * Enter guest context and enter an RCU extended quiescent state.
413 * Between guest_context_enter_irqoff() and guest_context_exit_irqoff() it is
414 * unsafe to use any code which may directly or indirectly use RCU, tracing
415 * (including IRQ flag tracing), or lockdep. All code in this period must be
416 * non-instrumentable.
418 static __always_inline void guest_context_enter_irqoff(void)
421 * KVM does not hold any references to rcu protected data when it
422 * switches CPU into a guest mode. In fact switching to a guest mode
423 * is very similar to exiting to userspace from rcu point of view. In
424 * addition CPU may stay in a guest mode for quite a long time (up to
425 * one time slice). Lets treat guest mode as quiescent state, just like
426 * we do with user-mode execution.
428 if (!context_tracking_guest_enter()) {
429 instrumentation_begin();
430 rcu_virt_note_context_switch();
431 instrumentation_end();
436 * Deprecated. Architectures should move to guest_timing_enter_irqoff() and
437 * guest_state_enter_irqoff().
439 static __always_inline void guest_enter_irqoff(void)
441 guest_timing_enter_irqoff();
442 guest_context_enter_irqoff();
446 * guest_state_enter_irqoff - Fixup state when entering a guest
448 * Entry to a guest will enable interrupts, but the kernel state is interrupts
449 * disabled when this is invoked. Also tell RCU about it.
451 * 1) Trace interrupts on state
452 * 2) Invoke context tracking if enabled to adjust RCU state
453 * 3) Tell lockdep that interrupts are enabled
455 * Invoked from architecture specific code before entering a guest.
456 * Must be called with interrupts disabled and the caller must be
457 * non-instrumentable.
458 * The caller has to invoke guest_timing_enter_irqoff() before this.
460 * Note: this is analogous to exit_to_user_mode().
462 static __always_inline void guest_state_enter_irqoff(void)
464 instrumentation_begin();
465 trace_hardirqs_on_prepare();
466 lockdep_hardirqs_on_prepare();
467 instrumentation_end();
469 guest_context_enter_irqoff();
470 lockdep_hardirqs_on(CALLER_ADDR0);
474 * Exit guest context and exit an RCU extended quiescent state.
476 * Between guest_context_enter_irqoff() and guest_context_exit_irqoff() it is
477 * unsafe to use any code which may directly or indirectly use RCU, tracing
478 * (including IRQ flag tracing), or lockdep. All code in this period must be
479 * non-instrumentable.
481 static __always_inline void guest_context_exit_irqoff(void)
483 context_tracking_guest_exit();
487 * Stop accounting time towards a guest.
488 * Must be called after exiting guest context.
490 static __always_inline void guest_timing_exit_irqoff(void)
492 instrumentation_begin();
493 /* Flush the guest cputime we spent on the guest */
494 vtime_account_guest_exit();
495 instrumentation_end();
499 * Deprecated. Architectures should move to guest_state_exit_irqoff() and
500 * guest_timing_exit_irqoff().
502 static __always_inline void guest_exit_irqoff(void)
504 guest_context_exit_irqoff();
505 guest_timing_exit_irqoff();
508 static inline void guest_exit(void)
512 local_irq_save(flags);
514 local_irq_restore(flags);
518 * guest_state_exit_irqoff - Establish state when returning from guest mode
520 * Entry from a guest disables interrupts, but guest mode is traced as
521 * interrupts enabled. Also with NO_HZ_FULL RCU might be idle.
523 * 1) Tell lockdep that interrupts are disabled
524 * 2) Invoke context tracking if enabled to reactivate RCU
525 * 3) Trace interrupts off state
527 * Invoked from architecture specific code after exiting a guest.
528 * Must be invoked with interrupts disabled and the caller must be
529 * non-instrumentable.
530 * The caller has to invoke guest_timing_exit_irqoff() after this.
532 * Note: this is analogous to enter_from_user_mode().
534 static __always_inline void guest_state_exit_irqoff(void)
536 lockdep_hardirqs_off(CALLER_ADDR0);
537 guest_context_exit_irqoff();
539 instrumentation_begin();
540 trace_hardirqs_off_finish();
541 instrumentation_end();
544 static inline int kvm_vcpu_exiting_guest_mode(struct kvm_vcpu *vcpu)
547 * The memory barrier ensures a previous write to vcpu->requests cannot
548 * be reordered with the read of vcpu->mode. It pairs with the general
549 * memory barrier following the write of vcpu->mode in VCPU RUN.
551 smp_mb__before_atomic();
552 return cmpxchg(&vcpu->mode, IN_GUEST_MODE, EXITING_GUEST_MODE);
556 * Some of the bitops functions do not support too long bitmaps.
557 * This number must be determined not to exceed such limits.
559 #define KVM_MEM_MAX_NR_PAGES ((1UL << 31) - 1)
562 * Since at idle each memslot belongs to two memslot sets it has to contain
563 * two embedded nodes for each data structure that it forms a part of.
565 * Two memslot sets (one active and one inactive) are necessary so the VM
566 * continues to run on one memslot set while the other is being modified.
568 * These two memslot sets normally point to the same set of memslots.
569 * They can, however, be desynchronized when performing a memslot management
570 * operation by replacing the memslot to be modified by its copy.
571 * After the operation is complete, both memslot sets once again point to
572 * the same, common set of memslot data.
574 * The memslots themselves are independent of each other so they can be
575 * individually added or deleted.
577 struct kvm_memory_slot {
578 struct hlist_node id_node[2];
579 struct interval_tree_node hva_node[2];
580 struct rb_node gfn_node[2];
582 unsigned long npages;
583 unsigned long *dirty_bitmap;
584 struct kvm_arch_memory_slot arch;
585 unsigned long userspace_addr;
591 static inline bool kvm_slot_dirty_track_enabled(const struct kvm_memory_slot *slot)
593 return slot->flags & KVM_MEM_LOG_DIRTY_PAGES;
596 static inline unsigned long kvm_dirty_bitmap_bytes(struct kvm_memory_slot *memslot)
598 return ALIGN(memslot->npages, BITS_PER_LONG) / 8;
601 static inline unsigned long *kvm_second_dirty_bitmap(struct kvm_memory_slot *memslot)
603 unsigned long len = kvm_dirty_bitmap_bytes(memslot);
605 return memslot->dirty_bitmap + len / sizeof(*memslot->dirty_bitmap);
608 #ifndef KVM_DIRTY_LOG_MANUAL_CAPS
609 #define KVM_DIRTY_LOG_MANUAL_CAPS KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE
612 struct kvm_s390_adapter_int {
625 struct kvm_xen_evtchn {
632 struct kvm_kernel_irq_routing_entry {
635 int (*set)(struct kvm_kernel_irq_routing_entry *e,
636 struct kvm *kvm, int irq_source_id, int level,
650 struct kvm_s390_adapter_int adapter;
651 struct kvm_hv_sint hv_sint;
652 struct kvm_xen_evtchn xen_evtchn;
654 struct hlist_node link;
657 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
658 struct kvm_irq_routing_table {
659 int chip[KVM_NR_IRQCHIPS][KVM_IRQCHIP_NUM_PINS];
662 * Array indexed by gsi. Each entry contains list of irq chips
663 * the gsi is connected to.
665 struct hlist_head map[];
669 bool kvm_arch_irqchip_in_kernel(struct kvm *kvm);
671 #ifndef KVM_INTERNAL_MEM_SLOTS
672 #define KVM_INTERNAL_MEM_SLOTS 0
675 #define KVM_MEM_SLOTS_NUM SHRT_MAX
676 #define KVM_USER_MEM_SLOTS (KVM_MEM_SLOTS_NUM - KVM_INTERNAL_MEM_SLOTS)
678 #ifndef __KVM_VCPU_MULTIPLE_ADDRESS_SPACE
679 static inline int kvm_arch_vcpu_memslots_id(struct kvm_vcpu *vcpu)
685 struct kvm_memslots {
687 atomic_long_t last_used_slot;
688 struct rb_root_cached hva_tree;
689 struct rb_root gfn_tree;
691 * The mapping table from slot id to memslot.
693 * 7-bit bucket count matches the size of the old id to index array for
694 * 512 slots, while giving good performance with this slot count.
695 * Higher bucket counts bring only small performance improvements but
696 * always result in higher memory usage (even for lower memslot counts).
698 DECLARE_HASHTABLE(id_hash, 7);
703 #ifdef KVM_HAVE_MMU_RWLOCK
707 #endif /* KVM_HAVE_MMU_RWLOCK */
709 struct mutex slots_lock;
712 * Protects the arch-specific fields of struct kvm_memory_slots in
713 * use by the VM. To be used under the slots_lock (above) or in a
714 * kvm->srcu critical section where acquiring the slots_lock would
715 * lead to deadlock with the synchronize_srcu in
716 * install_new_memslots.
718 struct mutex slots_arch_lock;
719 struct mm_struct *mm; /* userspace tied to this vm */
720 unsigned long nr_memslot_pages;
721 /* The two memslot sets - active and inactive (per address space) */
722 struct kvm_memslots __memslots[KVM_ADDRESS_SPACE_NUM][2];
723 /* The current active memslot set for each address space */
724 struct kvm_memslots __rcu *memslots[KVM_ADDRESS_SPACE_NUM];
725 struct xarray vcpu_array;
727 * Protected by slots_lock, but can be read outside if an
728 * incorrect answer is acceptable.
730 atomic_t nr_memslots_dirty_logging;
732 /* Used to wait for completion of MMU notifiers. */
733 spinlock_t mn_invalidate_lock;
734 unsigned long mn_active_invalidate_count;
735 struct rcuwait mn_memslots_update_rcuwait;
737 /* For management / invalidation of gfn_to_pfn_caches */
739 struct list_head gpc_list;
742 * created_vcpus is protected by kvm->lock, and is incremented
743 * at the beginning of KVM_CREATE_VCPU. online_vcpus is only
744 * incremented after storing the kvm_vcpu pointer in vcpus,
745 * and is accessed atomically.
747 atomic_t online_vcpus;
750 int last_boosted_vcpu;
751 struct list_head vm_list;
753 struct kvm_io_bus __rcu *buses[KVM_NR_BUSES];
754 #ifdef CONFIG_HAVE_KVM_EVENTFD
757 struct list_head items;
758 struct list_head resampler_list;
759 struct mutex resampler_lock;
761 struct list_head ioeventfds;
763 struct kvm_vm_stat stat;
764 struct kvm_arch arch;
765 refcount_t users_count;
766 #ifdef CONFIG_KVM_MMIO
767 struct kvm_coalesced_mmio_ring *coalesced_mmio_ring;
768 spinlock_t ring_lock;
769 struct list_head coalesced_zones;
772 struct mutex irq_lock;
773 #ifdef CONFIG_HAVE_KVM_IRQCHIP
775 * Update side is protected by irq_lock.
777 struct kvm_irq_routing_table __rcu *irq_routing;
779 #ifdef CONFIG_HAVE_KVM_IRQFD
780 struct hlist_head irq_ack_notifier_list;
783 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
784 struct mmu_notifier mmu_notifier;
785 unsigned long mmu_invalidate_seq;
786 long mmu_invalidate_in_progress;
787 unsigned long mmu_invalidate_range_start;
788 unsigned long mmu_invalidate_range_end;
790 struct list_head devices;
791 u64 manual_dirty_log_protect;
792 struct dentry *debugfs_dentry;
793 struct kvm_stat_data **debugfs_stat_data;
794 struct srcu_struct srcu;
795 struct srcu_struct irq_srcu;
797 bool override_halt_poll_ns;
798 unsigned int max_halt_poll_ns;
800 bool dirty_ring_with_bitmap;
804 #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER
805 struct notifier_block pm_notifier;
807 char stats_id[KVM_STATS_NAME_SIZE];
810 #define kvm_err(fmt, ...) \
811 pr_err("kvm [%i]: " fmt, task_pid_nr(current), ## __VA_ARGS__)
812 #define kvm_info(fmt, ...) \
813 pr_info("kvm [%i]: " fmt, task_pid_nr(current), ## __VA_ARGS__)
814 #define kvm_debug(fmt, ...) \
815 pr_debug("kvm [%i]: " fmt, task_pid_nr(current), ## __VA_ARGS__)
816 #define kvm_debug_ratelimited(fmt, ...) \
817 pr_debug_ratelimited("kvm [%i]: " fmt, task_pid_nr(current), \
819 #define kvm_pr_unimpl(fmt, ...) \
820 pr_err_ratelimited("kvm [%i]: " fmt, \
821 task_tgid_nr(current), ## __VA_ARGS__)
823 /* The guest did something we don't support. */
824 #define vcpu_unimpl(vcpu, fmt, ...) \
825 kvm_pr_unimpl("vcpu%i, guest rIP: 0x%lx " fmt, \
826 (vcpu)->vcpu_id, kvm_rip_read(vcpu), ## __VA_ARGS__)
828 #define vcpu_debug(vcpu, fmt, ...) \
829 kvm_debug("vcpu%i " fmt, (vcpu)->vcpu_id, ## __VA_ARGS__)
830 #define vcpu_debug_ratelimited(vcpu, fmt, ...) \
831 kvm_debug_ratelimited("vcpu%i " fmt, (vcpu)->vcpu_id, \
833 #define vcpu_err(vcpu, fmt, ...) \
834 kvm_err("vcpu%i " fmt, (vcpu)->vcpu_id, ## __VA_ARGS__)
836 static inline void kvm_vm_dead(struct kvm *kvm)
839 kvm_make_all_cpus_request(kvm, KVM_REQ_VM_DEAD);
842 static inline void kvm_vm_bugged(struct kvm *kvm)
844 kvm->vm_bugged = true;
849 #define KVM_BUG(cond, kvm, fmt...) \
851 int __ret = (cond); \
853 if (WARN_ONCE(__ret && !(kvm)->vm_bugged, fmt)) \
854 kvm_vm_bugged(kvm); \
858 #define KVM_BUG_ON(cond, kvm) \
860 int __ret = (cond); \
862 if (WARN_ON_ONCE(__ret && !(kvm)->vm_bugged)) \
863 kvm_vm_bugged(kvm); \
867 static inline void kvm_vcpu_srcu_read_lock(struct kvm_vcpu *vcpu)
869 #ifdef CONFIG_PROVE_RCU
870 WARN_ONCE(vcpu->srcu_depth++,
871 "KVM: Illegal vCPU srcu_idx LOCK, depth=%d", vcpu->srcu_depth - 1);
873 vcpu->____srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
876 static inline void kvm_vcpu_srcu_read_unlock(struct kvm_vcpu *vcpu)
878 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->____srcu_idx);
880 #ifdef CONFIG_PROVE_RCU
881 WARN_ONCE(--vcpu->srcu_depth,
882 "KVM: Illegal vCPU srcu_idx UNLOCK, depth=%d", vcpu->srcu_depth);
886 static inline bool kvm_dirty_log_manual_protect_and_init_set(struct kvm *kvm)
888 return !!(kvm->manual_dirty_log_protect & KVM_DIRTY_LOG_INITIALLY_SET);
891 static inline struct kvm_io_bus *kvm_get_bus(struct kvm *kvm, enum kvm_bus idx)
893 return srcu_dereference_check(kvm->buses[idx], &kvm->srcu,
894 lockdep_is_held(&kvm->slots_lock) ||
895 !refcount_read(&kvm->users_count));
898 static inline struct kvm_vcpu *kvm_get_vcpu(struct kvm *kvm, int i)
900 int num_vcpus = atomic_read(&kvm->online_vcpus);
901 i = array_index_nospec(i, num_vcpus);
903 /* Pairs with smp_wmb() in kvm_vm_ioctl_create_vcpu. */
905 return xa_load(&kvm->vcpu_array, i);
908 #define kvm_for_each_vcpu(idx, vcpup, kvm) \
909 xa_for_each_range(&kvm->vcpu_array, idx, vcpup, 0, \
910 (atomic_read(&kvm->online_vcpus) - 1))
912 static inline struct kvm_vcpu *kvm_get_vcpu_by_id(struct kvm *kvm, int id)
914 struct kvm_vcpu *vcpu = NULL;
919 if (id < KVM_MAX_VCPUS)
920 vcpu = kvm_get_vcpu(kvm, id);
921 if (vcpu && vcpu->vcpu_id == id)
923 kvm_for_each_vcpu(i, vcpu, kvm)
924 if (vcpu->vcpu_id == id)
929 void kvm_destroy_vcpus(struct kvm *kvm);
931 void vcpu_load(struct kvm_vcpu *vcpu);
932 void vcpu_put(struct kvm_vcpu *vcpu);
934 #ifdef __KVM_HAVE_IOAPIC
935 void kvm_arch_post_irq_ack_notifier_list_update(struct kvm *kvm);
936 void kvm_arch_post_irq_routing_update(struct kvm *kvm);
938 static inline void kvm_arch_post_irq_ack_notifier_list_update(struct kvm *kvm)
941 static inline void kvm_arch_post_irq_routing_update(struct kvm *kvm)
946 #ifdef CONFIG_HAVE_KVM_IRQFD
947 int kvm_irqfd_init(void);
948 void kvm_irqfd_exit(void);
950 static inline int kvm_irqfd_init(void)
955 static inline void kvm_irqfd_exit(void)
959 int kvm_init(unsigned vcpu_size, unsigned vcpu_align, struct module *module);
962 void kvm_get_kvm(struct kvm *kvm);
963 bool kvm_get_kvm_safe(struct kvm *kvm);
964 void kvm_put_kvm(struct kvm *kvm);
965 bool file_is_kvm(struct file *file);
966 void kvm_put_kvm_no_destroy(struct kvm *kvm);
968 static inline struct kvm_memslots *__kvm_memslots(struct kvm *kvm, int as_id)
970 as_id = array_index_nospec(as_id, KVM_ADDRESS_SPACE_NUM);
971 return srcu_dereference_check(kvm->memslots[as_id], &kvm->srcu,
972 lockdep_is_held(&kvm->slots_lock) ||
973 !refcount_read(&kvm->users_count));
976 static inline struct kvm_memslots *kvm_memslots(struct kvm *kvm)
978 return __kvm_memslots(kvm, 0);
981 static inline struct kvm_memslots *kvm_vcpu_memslots(struct kvm_vcpu *vcpu)
983 int as_id = kvm_arch_vcpu_memslots_id(vcpu);
985 return __kvm_memslots(vcpu->kvm, as_id);
988 static inline bool kvm_memslots_empty(struct kvm_memslots *slots)
990 return RB_EMPTY_ROOT(&slots->gfn_tree);
993 #define kvm_for_each_memslot(memslot, bkt, slots) \
994 hash_for_each(slots->id_hash, bkt, memslot, id_node[slots->node_idx]) \
995 if (WARN_ON_ONCE(!memslot->npages)) { \
999 struct kvm_memory_slot *id_to_memslot(struct kvm_memslots *slots, int id)
1001 struct kvm_memory_slot *slot;
1002 int idx = slots->node_idx;
1004 hash_for_each_possible(slots->id_hash, slot, id_node[idx], id) {
1012 /* Iterator used for walking memslots that overlap a gfn range. */
1013 struct kvm_memslot_iter {
1014 struct kvm_memslots *slots;
1015 struct rb_node *node;
1016 struct kvm_memory_slot *slot;
1019 static inline void kvm_memslot_iter_next(struct kvm_memslot_iter *iter)
1021 iter->node = rb_next(iter->node);
1025 iter->slot = container_of(iter->node, struct kvm_memory_slot, gfn_node[iter->slots->node_idx]);
1028 static inline void kvm_memslot_iter_start(struct kvm_memslot_iter *iter,
1029 struct kvm_memslots *slots,
1032 int idx = slots->node_idx;
1033 struct rb_node *tmp;
1034 struct kvm_memory_slot *slot;
1036 iter->slots = slots;
1039 * Find the so called "upper bound" of a key - the first node that has
1040 * its key strictly greater than the searched one (the start gfn in our case).
1043 for (tmp = slots->gfn_tree.rb_node; tmp; ) {
1044 slot = container_of(tmp, struct kvm_memory_slot, gfn_node[idx]);
1045 if (start < slot->base_gfn) {
1049 tmp = tmp->rb_right;
1054 * Find the slot with the lowest gfn that can possibly intersect with
1055 * the range, so we'll ideally have slot start <= range start
1059 * A NULL previous node means that the very first slot
1060 * already has a higher start gfn.
1061 * In this case slot start > range start.
1063 tmp = rb_prev(iter->node);
1067 /* a NULL node below means no slots */
1068 iter->node = rb_last(&slots->gfn_tree);
1072 iter->slot = container_of(iter->node, struct kvm_memory_slot, gfn_node[idx]);
1075 * It is possible in the slot start < range start case that the
1076 * found slot ends before or at range start (slot end <= range start)
1077 * and so it does not overlap the requested range.
1079 * In such non-overlapping case the next slot (if it exists) will
1080 * already have slot start > range start, otherwise the logic above
1081 * would have found it instead of the current slot.
1083 if (iter->slot->base_gfn + iter->slot->npages <= start)
1084 kvm_memslot_iter_next(iter);
1088 static inline bool kvm_memslot_iter_is_valid(struct kvm_memslot_iter *iter, gfn_t end)
1094 * If this slot starts beyond or at the end of the range so does
1097 return iter->slot->base_gfn < end;
1100 /* Iterate over each memslot at least partially intersecting [start, end) range */
1101 #define kvm_for_each_memslot_in_gfn_range(iter, slots, start, end) \
1102 for (kvm_memslot_iter_start(iter, slots, start); \
1103 kvm_memslot_iter_is_valid(iter, end); \
1104 kvm_memslot_iter_next(iter))
1107 * KVM_SET_USER_MEMORY_REGION ioctl allows the following operations:
1108 * - create a new memory slot
1109 * - delete an existing memory slot
1110 * - modify an existing memory slot
1111 * -- move it in the guest physical memory space
1112 * -- just change its flags
1114 * Since flags can be changed by some of these operations, the following
1115 * differentiation is the best we can do for __kvm_set_memory_region():
1117 enum kvm_mr_change {
1124 int kvm_set_memory_region(struct kvm *kvm,
1125 const struct kvm_userspace_memory_region *mem);
1126 int __kvm_set_memory_region(struct kvm *kvm,
1127 const struct kvm_userspace_memory_region *mem);
1128 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot);
1129 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen);
1130 int kvm_arch_prepare_memory_region(struct kvm *kvm,
1131 const struct kvm_memory_slot *old,
1132 struct kvm_memory_slot *new,
1133 enum kvm_mr_change change);
1134 void kvm_arch_commit_memory_region(struct kvm *kvm,
1135 struct kvm_memory_slot *old,
1136 const struct kvm_memory_slot *new,
1137 enum kvm_mr_change change);
1138 /* flush all memory translations */
1139 void kvm_arch_flush_shadow_all(struct kvm *kvm);
1140 /* flush memory translations pointing to 'slot' */
1141 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
1142 struct kvm_memory_slot *slot);
1144 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1145 struct page **pages, int nr_pages);
1147 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn);
1148 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn);
1149 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable);
1150 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot, gfn_t gfn);
1151 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot, gfn_t gfn,
1153 void kvm_release_page_clean(struct page *page);
1154 void kvm_release_page_dirty(struct page *page);
1156 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn);
1157 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1159 kvm_pfn_t gfn_to_pfn_memslot(const struct kvm_memory_slot *slot, gfn_t gfn);
1160 kvm_pfn_t gfn_to_pfn_memslot_atomic(const struct kvm_memory_slot *slot, gfn_t gfn);
1161 kvm_pfn_t __gfn_to_pfn_memslot(const struct kvm_memory_slot *slot, gfn_t gfn,
1162 bool atomic, bool interruptible, bool *async,
1163 bool write_fault, bool *writable, hva_t *hva);
1165 void kvm_release_pfn_clean(kvm_pfn_t pfn);
1166 void kvm_release_pfn_dirty(kvm_pfn_t pfn);
1167 void kvm_set_pfn_dirty(kvm_pfn_t pfn);
1168 void kvm_set_pfn_accessed(kvm_pfn_t pfn);
1170 void kvm_release_pfn(kvm_pfn_t pfn, bool dirty);
1171 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1173 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len);
1174 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1175 void *data, unsigned long len);
1176 int kvm_read_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1177 void *data, unsigned int offset,
1179 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1180 int offset, int len);
1181 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1183 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1184 void *data, unsigned long len);
1185 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1186 void *data, unsigned int offset,
1188 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1189 gpa_t gpa, unsigned long len);
1191 #define __kvm_get_guest(kvm, gfn, offset, v) \
1193 unsigned long __addr = gfn_to_hva(kvm, gfn); \
1194 typeof(v) __user *__uaddr = (typeof(__uaddr))(__addr + offset); \
1195 int __ret = -EFAULT; \
1197 if (!kvm_is_error_hva(__addr)) \
1198 __ret = get_user(v, __uaddr); \
1202 #define kvm_get_guest(kvm, gpa, v) \
1204 gpa_t __gpa = gpa; \
1205 struct kvm *__kvm = kvm; \
1207 __kvm_get_guest(__kvm, __gpa >> PAGE_SHIFT, \
1208 offset_in_page(__gpa), v); \
1211 #define __kvm_put_guest(kvm, gfn, offset, v) \
1213 unsigned long __addr = gfn_to_hva(kvm, gfn); \
1214 typeof(v) __user *__uaddr = (typeof(__uaddr))(__addr + offset); \
1215 int __ret = -EFAULT; \
1217 if (!kvm_is_error_hva(__addr)) \
1218 __ret = put_user(v, __uaddr); \
1220 mark_page_dirty(kvm, gfn); \
1224 #define kvm_put_guest(kvm, gpa, v) \
1226 gpa_t __gpa = gpa; \
1227 struct kvm *__kvm = kvm; \
1229 __kvm_put_guest(__kvm, __gpa >> PAGE_SHIFT, \
1230 offset_in_page(__gpa), v); \
1233 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len);
1234 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn);
1235 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn);
1236 bool kvm_vcpu_is_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn);
1237 unsigned long kvm_host_page_size(struct kvm_vcpu *vcpu, gfn_t gfn);
1238 void mark_page_dirty_in_slot(struct kvm *kvm, const struct kvm_memory_slot *memslot, gfn_t gfn);
1239 void mark_page_dirty(struct kvm *kvm, gfn_t gfn);
1241 struct kvm_memslots *kvm_vcpu_memslots(struct kvm_vcpu *vcpu);
1242 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn);
1243 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn);
1244 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn);
1245 int kvm_vcpu_map(struct kvm_vcpu *vcpu, gpa_t gpa, struct kvm_host_map *map);
1246 void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map, bool dirty);
1247 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn);
1248 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable);
1249 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data, int offset,
1251 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa, void *data,
1253 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data,
1255 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, const void *data,
1256 int offset, int len);
1257 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
1259 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn);
1262 * kvm_gpc_init - initialize gfn_to_pfn_cache.
1264 * @gpc: struct gfn_to_pfn_cache object.
1265 * @kvm: pointer to kvm instance.
1266 * @vcpu: vCPU to be used for marking pages dirty and to be woken on
1268 * @usage: indicates if the resulting host physical PFN is used while
1269 * the @vcpu is IN_GUEST_MODE (in which case invalidation of
1270 * the cache from MMU notifiers---but not for KVM memslot
1271 * changes!---will also force @vcpu to exit the guest and
1272 * refresh the cache); and/or if the PFN used directly
1273 * by KVM (and thus needs a kernel virtual mapping).
1275 * This sets up a gfn_to_pfn_cache by initializing locks and assigning the
1276 * immutable attributes. Note, the cache must be zero-allocated (or zeroed by
1277 * the caller before init).
1279 void kvm_gpc_init(struct gfn_to_pfn_cache *gpc, struct kvm *kvm,
1280 struct kvm_vcpu *vcpu, enum pfn_cache_usage usage);
1283 * kvm_gpc_activate - prepare a cached kernel mapping and HPA for a given guest
1286 * @gpc: struct gfn_to_pfn_cache object.
1287 * @gpa: guest physical address to map.
1288 * @len: sanity check; the range being access must fit a single page.
1290 * @return: 0 for success.
1291 * -EINVAL for a mapping which would cross a page boundary.
1292 * -EFAULT for an untranslatable guest physical address.
1294 * This primes a gfn_to_pfn_cache and links it into the @gpc->kvm's list for
1295 * invalidations to be processed. Callers are required to use kvm_gpc_check()
1296 * to ensure that the cache is valid before accessing the target page.
1298 int kvm_gpc_activate(struct gfn_to_pfn_cache *gpc, gpa_t gpa, unsigned long len);
1301 * kvm_gpc_check - check validity of a gfn_to_pfn_cache.
1303 * @gpc: struct gfn_to_pfn_cache object.
1304 * @len: sanity check; the range being access must fit a single page.
1306 * @return: %true if the cache is still valid and the address matches.
1307 * %false if the cache is not valid.
1309 * Callers outside IN_GUEST_MODE context should hold a read lock on @gpc->lock
1310 * while calling this function, and then continue to hold the lock until the
1311 * access is complete.
1313 * Callers in IN_GUEST_MODE may do so without locking, although they should
1314 * still hold a read lock on kvm->scru for the memslot checks.
1316 bool kvm_gpc_check(struct gfn_to_pfn_cache *gpc, unsigned long len);
1319 * kvm_gpc_refresh - update a previously initialized cache.
1321 * @gpc: struct gfn_to_pfn_cache object.
1322 * @len: sanity check; the range being access must fit a single page.
1324 * @return: 0 for success.
1325 * -EINVAL for a mapping which would cross a page boundary.
1326 * -EFAULT for an untranslatable guest physical address.
1328 * This will attempt to refresh a gfn_to_pfn_cache. Note that a successful
1329 * return from this function does not mean the page can be immediately
1330 * accessed because it may have raced with an invalidation. Callers must
1331 * still lock and check the cache status, as this function does not return
1332 * with the lock still held to permit access.
1334 int kvm_gpc_refresh(struct gfn_to_pfn_cache *gpc, unsigned long len);
1337 * kvm_gpc_deactivate - deactivate and unlink a gfn_to_pfn_cache.
1339 * @gpc: struct gfn_to_pfn_cache object.
1341 * This removes a cache from the VM's list to be processed on MMU notifier
1344 void kvm_gpc_deactivate(struct gfn_to_pfn_cache *gpc);
1346 void kvm_sigset_activate(struct kvm_vcpu *vcpu);
1347 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu);
1349 void kvm_vcpu_halt(struct kvm_vcpu *vcpu);
1350 bool kvm_vcpu_block(struct kvm_vcpu *vcpu);
1351 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu);
1352 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu);
1353 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu);
1354 void kvm_vcpu_kick(struct kvm_vcpu *vcpu);
1355 int kvm_vcpu_yield_to(struct kvm_vcpu *target);
1356 void kvm_vcpu_on_spin(struct kvm_vcpu *vcpu, bool yield_to_kernel_mode);
1358 void kvm_flush_remote_tlbs(struct kvm *kvm);
1360 #ifdef KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE
1361 int kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache *mc, int min);
1362 int __kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache *mc, int capacity, int min);
1363 int kvm_mmu_memory_cache_nr_free_objects(struct kvm_mmu_memory_cache *mc);
1364 void kvm_mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc);
1365 void *kvm_mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc);
1368 void kvm_mmu_invalidate_begin(struct kvm *kvm, unsigned long start,
1370 void kvm_mmu_invalidate_end(struct kvm *kvm, unsigned long start,
1373 long kvm_arch_dev_ioctl(struct file *filp,
1374 unsigned int ioctl, unsigned long arg);
1375 long kvm_arch_vcpu_ioctl(struct file *filp,
1376 unsigned int ioctl, unsigned long arg);
1377 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf);
1379 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext);
1381 void kvm_arch_mmu_enable_log_dirty_pt_masked(struct kvm *kvm,
1382 struct kvm_memory_slot *slot,
1384 unsigned long mask);
1385 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot);
1387 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1388 void kvm_arch_flush_remote_tlbs_memslot(struct kvm *kvm,
1389 const struct kvm_memory_slot *memslot);
1390 #else /* !CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
1391 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log);
1392 int kvm_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log,
1393 int *is_dirty, struct kvm_memory_slot **memslot);
1396 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
1398 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
1399 struct kvm_enable_cap *cap);
1400 long kvm_arch_vm_ioctl(struct file *filp,
1401 unsigned int ioctl, unsigned long arg);
1402 long kvm_arch_vm_compat_ioctl(struct file *filp, unsigned int ioctl,
1405 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu);
1406 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu);
1408 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
1409 struct kvm_translation *tr);
1411 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs);
1412 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs);
1413 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
1414 struct kvm_sregs *sregs);
1415 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
1416 struct kvm_sregs *sregs);
1417 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
1418 struct kvm_mp_state *mp_state);
1419 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
1420 struct kvm_mp_state *mp_state);
1421 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
1422 struct kvm_guest_debug *dbg);
1423 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu);
1425 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu);
1427 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu);
1428 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu);
1429 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id);
1430 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu);
1431 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu);
1432 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu);
1434 #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER
1435 int kvm_arch_pm_notifier(struct kvm *kvm, unsigned long state);
1438 #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS
1439 void kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu, struct dentry *debugfs_dentry);
1441 static inline void kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu) {}
1444 #ifdef CONFIG_KVM_GENERIC_HARDWARE_ENABLING
1445 int kvm_arch_hardware_enable(void);
1446 void kvm_arch_hardware_disable(void);
1448 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu);
1449 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu);
1450 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu);
1451 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu);
1452 bool kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu);
1453 int kvm_arch_post_init_vm(struct kvm *kvm);
1454 void kvm_arch_pre_destroy_vm(struct kvm *kvm);
1455 int kvm_arch_create_vm_debugfs(struct kvm *kvm);
1457 #ifndef __KVM_HAVE_ARCH_VM_ALLOC
1459 * All architectures that want to use vzalloc currently also
1460 * need their own kvm_arch_alloc_vm implementation.
1462 static inline struct kvm *kvm_arch_alloc_vm(void)
1464 return kzalloc(sizeof(struct kvm), GFP_KERNEL_ACCOUNT);
1468 static inline void __kvm_arch_free_vm(struct kvm *kvm)
1473 #ifndef __KVM_HAVE_ARCH_VM_FREE
1474 static inline void kvm_arch_free_vm(struct kvm *kvm)
1476 __kvm_arch_free_vm(kvm);
1480 #ifndef __KVM_HAVE_ARCH_FLUSH_REMOTE_TLB
1481 static inline int kvm_arch_flush_remote_tlb(struct kvm *kvm)
1487 #ifdef __KVM_HAVE_ARCH_NONCOHERENT_DMA
1488 void kvm_arch_register_noncoherent_dma(struct kvm *kvm);
1489 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm);
1490 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm);
1492 static inline void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
1496 static inline void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
1500 static inline bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
1505 #ifdef __KVM_HAVE_ARCH_ASSIGNED_DEVICE
1506 void kvm_arch_start_assignment(struct kvm *kvm);
1507 void kvm_arch_end_assignment(struct kvm *kvm);
1508 bool kvm_arch_has_assigned_device(struct kvm *kvm);
1510 static inline void kvm_arch_start_assignment(struct kvm *kvm)
1514 static inline void kvm_arch_end_assignment(struct kvm *kvm)
1518 static __always_inline bool kvm_arch_has_assigned_device(struct kvm *kvm)
1524 static inline struct rcuwait *kvm_arch_vcpu_get_wait(struct kvm_vcpu *vcpu)
1526 #ifdef __KVM_HAVE_ARCH_WQP
1527 return vcpu->arch.waitp;
1534 * Wake a vCPU if necessary, but don't do any stats/metadata updates. Returns
1535 * true if the vCPU was blocking and was awakened, false otherwise.
1537 static inline bool __kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
1539 return !!rcuwait_wake_up(kvm_arch_vcpu_get_wait(vcpu));
1542 static inline bool kvm_vcpu_is_blocking(struct kvm_vcpu *vcpu)
1544 return rcuwait_active(kvm_arch_vcpu_get_wait(vcpu));
1547 #ifdef __KVM_HAVE_ARCH_INTC_INITIALIZED
1549 * returns true if the virtual interrupt controller is initialized and
1550 * ready to accept virtual IRQ. On some architectures the virtual interrupt
1551 * controller is dynamically instantiated and this is not always true.
1553 bool kvm_arch_intc_initialized(struct kvm *kvm);
1555 static inline bool kvm_arch_intc_initialized(struct kvm *kvm)
1561 #ifdef CONFIG_GUEST_PERF_EVENTS
1562 unsigned long kvm_arch_vcpu_get_ip(struct kvm_vcpu *vcpu);
1564 void kvm_register_perf_callbacks(unsigned int (*pt_intr_handler)(void));
1565 void kvm_unregister_perf_callbacks(void);
1567 static inline void kvm_register_perf_callbacks(void *ign) {}
1568 static inline void kvm_unregister_perf_callbacks(void) {}
1569 #endif /* CONFIG_GUEST_PERF_EVENTS */
1571 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type);
1572 void kvm_arch_destroy_vm(struct kvm *kvm);
1573 void kvm_arch_sync_events(struct kvm *kvm);
1575 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu);
1577 struct page *kvm_pfn_to_refcounted_page(kvm_pfn_t pfn);
1578 bool kvm_is_zone_device_page(struct page *page);
1580 struct kvm_irq_ack_notifier {
1581 struct hlist_node link;
1583 void (*irq_acked)(struct kvm_irq_ack_notifier *kian);
1586 int kvm_irq_map_gsi(struct kvm *kvm,
1587 struct kvm_kernel_irq_routing_entry *entries, int gsi);
1588 int kvm_irq_map_chip_pin(struct kvm *kvm, unsigned irqchip, unsigned pin);
1590 int kvm_set_irq(struct kvm *kvm, int irq_source_id, u32 irq, int level,
1592 int kvm_set_msi(struct kvm_kernel_irq_routing_entry *irq_entry, struct kvm *kvm,
1593 int irq_source_id, int level, bool line_status);
1594 int kvm_arch_set_irq_inatomic(struct kvm_kernel_irq_routing_entry *e,
1595 struct kvm *kvm, int irq_source_id,
1596 int level, bool line_status);
1597 bool kvm_irq_has_notifier(struct kvm *kvm, unsigned irqchip, unsigned pin);
1598 void kvm_notify_acked_gsi(struct kvm *kvm, int gsi);
1599 void kvm_notify_acked_irq(struct kvm *kvm, unsigned irqchip, unsigned pin);
1600 void kvm_register_irq_ack_notifier(struct kvm *kvm,
1601 struct kvm_irq_ack_notifier *kian);
1602 void kvm_unregister_irq_ack_notifier(struct kvm *kvm,
1603 struct kvm_irq_ack_notifier *kian);
1604 int kvm_request_irq_source_id(struct kvm *kvm);
1605 void kvm_free_irq_source_id(struct kvm *kvm, int irq_source_id);
1606 bool kvm_arch_irqfd_allowed(struct kvm *kvm, struct kvm_irqfd *args);
1609 * Returns a pointer to the memslot if it contains gfn.
1610 * Otherwise returns NULL.
1612 static inline struct kvm_memory_slot *
1613 try_get_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1618 if (gfn >= slot->base_gfn && gfn < slot->base_gfn + slot->npages)
1625 * Returns a pointer to the memslot that contains gfn. Otherwise returns NULL.
1627 * With "approx" set returns the memslot also when the address falls
1628 * in a hole. In that case one of the memslots bordering the hole is
1631 static inline struct kvm_memory_slot *
1632 search_memslots(struct kvm_memslots *slots, gfn_t gfn, bool approx)
1634 struct kvm_memory_slot *slot;
1635 struct rb_node *node;
1636 int idx = slots->node_idx;
1639 for (node = slots->gfn_tree.rb_node; node; ) {
1640 slot = container_of(node, struct kvm_memory_slot, gfn_node[idx]);
1641 if (gfn >= slot->base_gfn) {
1642 if (gfn < slot->base_gfn + slot->npages)
1644 node = node->rb_right;
1646 node = node->rb_left;
1649 return approx ? slot : NULL;
1652 static inline struct kvm_memory_slot *
1653 ____gfn_to_memslot(struct kvm_memslots *slots, gfn_t gfn, bool approx)
1655 struct kvm_memory_slot *slot;
1657 slot = (struct kvm_memory_slot *)atomic_long_read(&slots->last_used_slot);
1658 slot = try_get_memslot(slot, gfn);
1662 slot = search_memslots(slots, gfn, approx);
1664 atomic_long_set(&slots->last_used_slot, (unsigned long)slot);
1672 * __gfn_to_memslot() and its descendants are here to allow arch code to inline
1673 * the lookups in hot paths. gfn_to_memslot() itself isn't here as an inline
1674 * because that would bloat other code too much.
1676 static inline struct kvm_memory_slot *
1677 __gfn_to_memslot(struct kvm_memslots *slots, gfn_t gfn)
1679 return ____gfn_to_memslot(slots, gfn, false);
1682 static inline unsigned long
1683 __gfn_to_hva_memslot(const struct kvm_memory_slot *slot, gfn_t gfn)
1686 * The index was checked originally in search_memslots. To avoid
1687 * that a malicious guest builds a Spectre gadget out of e.g. page
1688 * table walks, do not let the processor speculate loads outside
1689 * the guest's registered memslots.
1691 unsigned long offset = gfn - slot->base_gfn;
1692 offset = array_index_nospec(offset, slot->npages);
1693 return slot->userspace_addr + offset * PAGE_SIZE;
1696 static inline int memslot_id(struct kvm *kvm, gfn_t gfn)
1698 return gfn_to_memslot(kvm, gfn)->id;
1702 hva_to_gfn_memslot(unsigned long hva, struct kvm_memory_slot *slot)
1704 gfn_t gfn_offset = (hva - slot->userspace_addr) >> PAGE_SHIFT;
1706 return slot->base_gfn + gfn_offset;
1709 static inline gpa_t gfn_to_gpa(gfn_t gfn)
1711 return (gpa_t)gfn << PAGE_SHIFT;
1714 static inline gfn_t gpa_to_gfn(gpa_t gpa)
1716 return (gfn_t)(gpa >> PAGE_SHIFT);
1719 static inline hpa_t pfn_to_hpa(kvm_pfn_t pfn)
1721 return (hpa_t)pfn << PAGE_SHIFT;
1724 static inline bool kvm_is_error_gpa(struct kvm *kvm, gpa_t gpa)
1726 unsigned long hva = gfn_to_hva(kvm, gpa_to_gfn(gpa));
1728 return kvm_is_error_hva(hva);
1731 enum kvm_stat_kind {
1736 struct kvm_stat_data {
1738 const struct _kvm_stats_desc *desc;
1739 enum kvm_stat_kind kind;
1742 struct _kvm_stats_desc {
1743 struct kvm_stats_desc desc;
1744 char name[KVM_STATS_NAME_SIZE];
1747 #define STATS_DESC_COMMON(type, unit, base, exp, sz, bsz) \
1748 .flags = type | unit | base | \
1749 BUILD_BUG_ON_ZERO(type & ~KVM_STATS_TYPE_MASK) | \
1750 BUILD_BUG_ON_ZERO(unit & ~KVM_STATS_UNIT_MASK) | \
1751 BUILD_BUG_ON_ZERO(base & ~KVM_STATS_BASE_MASK), \
1756 #define VM_GENERIC_STATS_DESC(stat, type, unit, base, exp, sz, bsz) \
1759 STATS_DESC_COMMON(type, unit, base, exp, sz, bsz), \
1760 .offset = offsetof(struct kvm_vm_stat, generic.stat) \
1764 #define VCPU_GENERIC_STATS_DESC(stat, type, unit, base, exp, sz, bsz) \
1767 STATS_DESC_COMMON(type, unit, base, exp, sz, bsz), \
1768 .offset = offsetof(struct kvm_vcpu_stat, generic.stat) \
1772 #define VM_STATS_DESC(stat, type, unit, base, exp, sz, bsz) \
1775 STATS_DESC_COMMON(type, unit, base, exp, sz, bsz), \
1776 .offset = offsetof(struct kvm_vm_stat, stat) \
1780 #define VCPU_STATS_DESC(stat, type, unit, base, exp, sz, bsz) \
1783 STATS_DESC_COMMON(type, unit, base, exp, sz, bsz), \
1784 .offset = offsetof(struct kvm_vcpu_stat, stat) \
1788 /* SCOPE: VM, VM_GENERIC, VCPU, VCPU_GENERIC */
1789 #define STATS_DESC(SCOPE, stat, type, unit, base, exp, sz, bsz) \
1790 SCOPE##_STATS_DESC(stat, type, unit, base, exp, sz, bsz)
1792 #define STATS_DESC_CUMULATIVE(SCOPE, name, unit, base, exponent) \
1793 STATS_DESC(SCOPE, name, KVM_STATS_TYPE_CUMULATIVE, \
1794 unit, base, exponent, 1, 0)
1795 #define STATS_DESC_INSTANT(SCOPE, name, unit, base, exponent) \
1796 STATS_DESC(SCOPE, name, KVM_STATS_TYPE_INSTANT, \
1797 unit, base, exponent, 1, 0)
1798 #define STATS_DESC_PEAK(SCOPE, name, unit, base, exponent) \
1799 STATS_DESC(SCOPE, name, KVM_STATS_TYPE_PEAK, \
1800 unit, base, exponent, 1, 0)
1801 #define STATS_DESC_LINEAR_HIST(SCOPE, name, unit, base, exponent, sz, bsz) \
1802 STATS_DESC(SCOPE, name, KVM_STATS_TYPE_LINEAR_HIST, \
1803 unit, base, exponent, sz, bsz)
1804 #define STATS_DESC_LOG_HIST(SCOPE, name, unit, base, exponent, sz) \
1805 STATS_DESC(SCOPE, name, KVM_STATS_TYPE_LOG_HIST, \
1806 unit, base, exponent, sz, 0)
1808 /* Cumulative counter, read/write */
1809 #define STATS_DESC_COUNTER(SCOPE, name) \
1810 STATS_DESC_CUMULATIVE(SCOPE, name, KVM_STATS_UNIT_NONE, \
1811 KVM_STATS_BASE_POW10, 0)
1812 /* Instantaneous counter, read only */
1813 #define STATS_DESC_ICOUNTER(SCOPE, name) \
1814 STATS_DESC_INSTANT(SCOPE, name, KVM_STATS_UNIT_NONE, \
1815 KVM_STATS_BASE_POW10, 0)
1816 /* Peak counter, read/write */
1817 #define STATS_DESC_PCOUNTER(SCOPE, name) \
1818 STATS_DESC_PEAK(SCOPE, name, KVM_STATS_UNIT_NONE, \
1819 KVM_STATS_BASE_POW10, 0)
1821 /* Instantaneous boolean value, read only */
1822 #define STATS_DESC_IBOOLEAN(SCOPE, name) \
1823 STATS_DESC_INSTANT(SCOPE, name, KVM_STATS_UNIT_BOOLEAN, \
1824 KVM_STATS_BASE_POW10, 0)
1825 /* Peak (sticky) boolean value, read/write */
1826 #define STATS_DESC_PBOOLEAN(SCOPE, name) \
1827 STATS_DESC_PEAK(SCOPE, name, KVM_STATS_UNIT_BOOLEAN, \
1828 KVM_STATS_BASE_POW10, 0)
1830 /* Cumulative time in nanosecond */
1831 #define STATS_DESC_TIME_NSEC(SCOPE, name) \
1832 STATS_DESC_CUMULATIVE(SCOPE, name, KVM_STATS_UNIT_SECONDS, \
1833 KVM_STATS_BASE_POW10, -9)
1834 /* Linear histogram for time in nanosecond */
1835 #define STATS_DESC_LINHIST_TIME_NSEC(SCOPE, name, sz, bsz) \
1836 STATS_DESC_LINEAR_HIST(SCOPE, name, KVM_STATS_UNIT_SECONDS, \
1837 KVM_STATS_BASE_POW10, -9, sz, bsz)
1838 /* Logarithmic histogram for time in nanosecond */
1839 #define STATS_DESC_LOGHIST_TIME_NSEC(SCOPE, name, sz) \
1840 STATS_DESC_LOG_HIST(SCOPE, name, KVM_STATS_UNIT_SECONDS, \
1841 KVM_STATS_BASE_POW10, -9, sz)
1843 #define KVM_GENERIC_VM_STATS() \
1844 STATS_DESC_COUNTER(VM_GENERIC, remote_tlb_flush), \
1845 STATS_DESC_COUNTER(VM_GENERIC, remote_tlb_flush_requests)
1847 #define KVM_GENERIC_VCPU_STATS() \
1848 STATS_DESC_COUNTER(VCPU_GENERIC, halt_successful_poll), \
1849 STATS_DESC_COUNTER(VCPU_GENERIC, halt_attempted_poll), \
1850 STATS_DESC_COUNTER(VCPU_GENERIC, halt_poll_invalid), \
1851 STATS_DESC_COUNTER(VCPU_GENERIC, halt_wakeup), \
1852 STATS_DESC_TIME_NSEC(VCPU_GENERIC, halt_poll_success_ns), \
1853 STATS_DESC_TIME_NSEC(VCPU_GENERIC, halt_poll_fail_ns), \
1854 STATS_DESC_TIME_NSEC(VCPU_GENERIC, halt_wait_ns), \
1855 STATS_DESC_LOGHIST_TIME_NSEC(VCPU_GENERIC, halt_poll_success_hist, \
1856 HALT_POLL_HIST_COUNT), \
1857 STATS_DESC_LOGHIST_TIME_NSEC(VCPU_GENERIC, halt_poll_fail_hist, \
1858 HALT_POLL_HIST_COUNT), \
1859 STATS_DESC_LOGHIST_TIME_NSEC(VCPU_GENERIC, halt_wait_hist, \
1860 HALT_POLL_HIST_COUNT), \
1861 STATS_DESC_IBOOLEAN(VCPU_GENERIC, blocking)
1863 extern struct dentry *kvm_debugfs_dir;
1865 ssize_t kvm_stats_read(char *id, const struct kvm_stats_header *header,
1866 const struct _kvm_stats_desc *desc,
1867 void *stats, size_t size_stats,
1868 char __user *user_buffer, size_t size, loff_t *offset);
1871 * kvm_stats_linear_hist_update() - Update bucket value for linear histogram
1874 * @data: start address of the stats data
1875 * @size: the number of bucket of the stats data
1876 * @value: the new value used to update the linear histogram's bucket
1877 * @bucket_size: the size (width) of a bucket
1879 static inline void kvm_stats_linear_hist_update(u64 *data, size_t size,
1880 u64 value, size_t bucket_size)
1882 size_t index = div64_u64(value, bucket_size);
1884 index = min(index, size - 1);
1889 * kvm_stats_log_hist_update() - Update bucket value for logarithmic histogram
1892 * @data: start address of the stats data
1893 * @size: the number of bucket of the stats data
1894 * @value: the new value used to update the logarithmic histogram's bucket
1896 static inline void kvm_stats_log_hist_update(u64 *data, size_t size, u64 value)
1898 size_t index = fls64(value);
1900 index = min(index, size - 1);
1904 #define KVM_STATS_LINEAR_HIST_UPDATE(array, value, bsize) \
1905 kvm_stats_linear_hist_update(array, ARRAY_SIZE(array), value, bsize)
1906 #define KVM_STATS_LOG_HIST_UPDATE(array, value) \
1907 kvm_stats_log_hist_update(array, ARRAY_SIZE(array), value)
1910 extern const struct kvm_stats_header kvm_vm_stats_header;
1911 extern const struct _kvm_stats_desc kvm_vm_stats_desc[];
1912 extern const struct kvm_stats_header kvm_vcpu_stats_header;
1913 extern const struct _kvm_stats_desc kvm_vcpu_stats_desc[];
1915 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1916 static inline int mmu_invalidate_retry(struct kvm *kvm, unsigned long mmu_seq)
1918 if (unlikely(kvm->mmu_invalidate_in_progress))
1921 * Ensure the read of mmu_invalidate_in_progress happens before
1922 * the read of mmu_invalidate_seq. This interacts with the
1923 * smp_wmb() in mmu_notifier_invalidate_range_end to make sure
1924 * that the caller either sees the old (non-zero) value of
1925 * mmu_invalidate_in_progress or the new (incremented) value of
1926 * mmu_invalidate_seq.
1928 * PowerPC Book3s HV KVM calls this under a per-page lock rather
1929 * than under kvm->mmu_lock, for scalability, so can't rely on
1930 * kvm->mmu_lock to keep things ordered.
1933 if (kvm->mmu_invalidate_seq != mmu_seq)
1938 static inline int mmu_invalidate_retry_hva(struct kvm *kvm,
1939 unsigned long mmu_seq,
1942 lockdep_assert_held(&kvm->mmu_lock);
1944 * If mmu_invalidate_in_progress is non-zero, then the range maintained
1945 * by kvm_mmu_notifier_invalidate_range_start contains all addresses
1946 * that might be being invalidated. Note that it may include some false
1947 * positives, due to shortcuts when handing concurrent invalidations.
1949 if (unlikely(kvm->mmu_invalidate_in_progress) &&
1950 hva >= kvm->mmu_invalidate_range_start &&
1951 hva < kvm->mmu_invalidate_range_end)
1953 if (kvm->mmu_invalidate_seq != mmu_seq)
1959 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
1961 #define KVM_MAX_IRQ_ROUTES 4096 /* might need extension/rework in the future */
1963 bool kvm_arch_can_set_irq_routing(struct kvm *kvm);
1964 int kvm_set_irq_routing(struct kvm *kvm,
1965 const struct kvm_irq_routing_entry *entries,
1968 int kvm_set_routing_entry(struct kvm *kvm,
1969 struct kvm_kernel_irq_routing_entry *e,
1970 const struct kvm_irq_routing_entry *ue);
1971 void kvm_free_irq_routing(struct kvm *kvm);
1975 static inline void kvm_free_irq_routing(struct kvm *kvm) {}
1979 int kvm_send_userspace_msi(struct kvm *kvm, struct kvm_msi *msi);
1981 #ifdef CONFIG_HAVE_KVM_EVENTFD
1983 void kvm_eventfd_init(struct kvm *kvm);
1984 int kvm_ioeventfd(struct kvm *kvm, struct kvm_ioeventfd *args);
1986 #ifdef CONFIG_HAVE_KVM_IRQFD
1987 int kvm_irqfd(struct kvm *kvm, struct kvm_irqfd *args);
1988 void kvm_irqfd_release(struct kvm *kvm);
1989 void kvm_irq_routing_update(struct kvm *);
1991 static inline int kvm_irqfd(struct kvm *kvm, struct kvm_irqfd *args)
1996 static inline void kvm_irqfd_release(struct kvm *kvm) {}
2001 static inline void kvm_eventfd_init(struct kvm *kvm) {}
2003 static inline int kvm_irqfd(struct kvm *kvm, struct kvm_irqfd *args)
2008 static inline void kvm_irqfd_release(struct kvm *kvm) {}
2010 #ifdef CONFIG_HAVE_KVM_IRQCHIP
2011 static inline void kvm_irq_routing_update(struct kvm *kvm)
2016 static inline int kvm_ioeventfd(struct kvm *kvm, struct kvm_ioeventfd *args)
2021 #endif /* CONFIG_HAVE_KVM_EVENTFD */
2023 void kvm_arch_irq_routing_update(struct kvm *kvm);
2025 static inline void __kvm_make_request(int req, struct kvm_vcpu *vcpu)
2028 * Ensure the rest of the request is published to kvm_check_request's
2029 * caller. Paired with the smp_mb__after_atomic in kvm_check_request.
2032 set_bit(req & KVM_REQUEST_MASK, (void *)&vcpu->requests);
2035 static __always_inline void kvm_make_request(int req, struct kvm_vcpu *vcpu)
2038 * Request that don't require vCPU action should never be logged in
2039 * vcpu->requests. The vCPU won't clear the request, so it will stay
2040 * logged indefinitely and prevent the vCPU from entering the guest.
2042 BUILD_BUG_ON(!__builtin_constant_p(req) ||
2043 (req & KVM_REQUEST_NO_ACTION));
2045 __kvm_make_request(req, vcpu);
2048 static inline bool kvm_request_pending(struct kvm_vcpu *vcpu)
2050 return READ_ONCE(vcpu->requests);
2053 static inline bool kvm_test_request(int req, struct kvm_vcpu *vcpu)
2055 return test_bit(req & KVM_REQUEST_MASK, (void *)&vcpu->requests);
2058 static inline void kvm_clear_request(int req, struct kvm_vcpu *vcpu)
2060 clear_bit(req & KVM_REQUEST_MASK, (void *)&vcpu->requests);
2063 static inline bool kvm_check_request(int req, struct kvm_vcpu *vcpu)
2065 if (kvm_test_request(req, vcpu)) {
2066 kvm_clear_request(req, vcpu);
2069 * Ensure the rest of the request is visible to kvm_check_request's
2070 * caller. Paired with the smp_wmb in kvm_make_request.
2072 smp_mb__after_atomic();
2079 #ifdef CONFIG_KVM_GENERIC_HARDWARE_ENABLING
2080 extern bool kvm_rebooting;
2083 extern unsigned int halt_poll_ns;
2084 extern unsigned int halt_poll_ns_grow;
2085 extern unsigned int halt_poll_ns_grow_start;
2086 extern unsigned int halt_poll_ns_shrink;
2089 const struct kvm_device_ops *ops;
2092 struct list_head vm_node;
2095 /* create, destroy, and name are mandatory */
2096 struct kvm_device_ops {
2100 * create is called holding kvm->lock and any operations not suitable
2101 * to do while holding the lock should be deferred to init (see
2104 int (*create)(struct kvm_device *dev, u32 type);
2107 * init is called after create if create is successful and is called
2108 * outside of holding kvm->lock.
2110 void (*init)(struct kvm_device *dev);
2113 * Destroy is responsible for freeing dev.
2115 * Destroy may be called before or after destructors are called
2116 * on emulated I/O regions, depending on whether a reference is
2117 * held by a vcpu or other kvm component that gets destroyed
2118 * after the emulated I/O.
2120 void (*destroy)(struct kvm_device *dev);
2123 * Release is an alternative method to free the device. It is
2124 * called when the device file descriptor is closed. Once
2125 * release is called, the destroy method will not be called
2126 * anymore as the device is removed from the device list of
2127 * the VM. kvm->lock is held.
2129 void (*release)(struct kvm_device *dev);
2131 int (*set_attr)(struct kvm_device *dev, struct kvm_device_attr *attr);
2132 int (*get_attr)(struct kvm_device *dev, struct kvm_device_attr *attr);
2133 int (*has_attr)(struct kvm_device *dev, struct kvm_device_attr *attr);
2134 long (*ioctl)(struct kvm_device *dev, unsigned int ioctl,
2136 int (*mmap)(struct kvm_device *dev, struct vm_area_struct *vma);
2139 void kvm_device_get(struct kvm_device *dev);
2140 void kvm_device_put(struct kvm_device *dev);
2141 struct kvm_device *kvm_device_from_filp(struct file *filp);
2142 int kvm_register_device_ops(const struct kvm_device_ops *ops, u32 type);
2143 void kvm_unregister_device_ops(u32 type);
2145 extern struct kvm_device_ops kvm_mpic_ops;
2146 extern struct kvm_device_ops kvm_arm_vgic_v2_ops;
2147 extern struct kvm_device_ops kvm_arm_vgic_v3_ops;
2149 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2151 static inline void kvm_vcpu_set_in_spin_loop(struct kvm_vcpu *vcpu, bool val)
2153 vcpu->spin_loop.in_spin_loop = val;
2155 static inline void kvm_vcpu_set_dy_eligible(struct kvm_vcpu *vcpu, bool val)
2157 vcpu->spin_loop.dy_eligible = val;
2160 #else /* !CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT */
2162 static inline void kvm_vcpu_set_in_spin_loop(struct kvm_vcpu *vcpu, bool val)
2166 static inline void kvm_vcpu_set_dy_eligible(struct kvm_vcpu *vcpu, bool val)
2169 #endif /* CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT */
2171 static inline bool kvm_is_visible_memslot(struct kvm_memory_slot *memslot)
2173 return (memslot && memslot->id < KVM_USER_MEM_SLOTS &&
2174 !(memslot->flags & KVM_MEMSLOT_INVALID));
2177 struct kvm_vcpu *kvm_get_running_vcpu(void);
2178 struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void);
2180 #ifdef CONFIG_HAVE_KVM_IRQ_BYPASS
2181 bool kvm_arch_has_irq_bypass(void);
2182 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *,
2183 struct irq_bypass_producer *);
2184 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *,
2185 struct irq_bypass_producer *);
2186 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *);
2187 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *);
2188 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
2189 uint32_t guest_irq, bool set);
2190 bool kvm_arch_irqfd_route_changed(struct kvm_kernel_irq_routing_entry *,
2191 struct kvm_kernel_irq_routing_entry *);
2192 #endif /* CONFIG_HAVE_KVM_IRQ_BYPASS */
2194 #ifdef CONFIG_HAVE_KVM_INVALID_WAKEUPS
2195 /* If we wakeup during the poll time, was it a sucessful poll? */
2196 static inline bool vcpu_valid_wakeup(struct kvm_vcpu *vcpu)
2198 return vcpu->valid_wakeup;
2202 static inline bool vcpu_valid_wakeup(struct kvm_vcpu *vcpu)
2206 #endif /* CONFIG_HAVE_KVM_INVALID_WAKEUPS */
2208 #ifdef CONFIG_HAVE_KVM_NO_POLL
2209 /* Callback that tells if we must not poll */
2210 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu);
2212 static inline bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
2216 #endif /* CONFIG_HAVE_KVM_NO_POLL */
2218 #ifdef CONFIG_HAVE_KVM_VCPU_ASYNC_IOCTL
2219 long kvm_arch_vcpu_async_ioctl(struct file *filp,
2220 unsigned int ioctl, unsigned long arg);
2222 static inline long kvm_arch_vcpu_async_ioctl(struct file *filp,
2226 return -ENOIOCTLCMD;
2228 #endif /* CONFIG_HAVE_KVM_VCPU_ASYNC_IOCTL */
2230 void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
2231 unsigned long start, unsigned long end);
2233 void kvm_arch_guest_memory_reclaimed(struct kvm *kvm);
2235 #ifdef CONFIG_HAVE_KVM_VCPU_RUN_PID_CHANGE
2236 int kvm_arch_vcpu_run_pid_change(struct kvm_vcpu *vcpu);
2238 static inline int kvm_arch_vcpu_run_pid_change(struct kvm_vcpu *vcpu)
2242 #endif /* CONFIG_HAVE_KVM_VCPU_RUN_PID_CHANGE */
2244 typedef int (*kvm_vm_thread_fn_t)(struct kvm *kvm, uintptr_t data);
2246 int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn,
2247 uintptr_t data, const char *name,
2248 struct task_struct **thread_ptr);
2250 #ifdef CONFIG_KVM_XFER_TO_GUEST_WORK
2251 static inline void kvm_handle_signal_exit(struct kvm_vcpu *vcpu)
2253 vcpu->run->exit_reason = KVM_EXIT_INTR;
2254 vcpu->stat.signal_exits++;
2256 #endif /* CONFIG_KVM_XFER_TO_GUEST_WORK */
2259 * If more than one page is being (un)accounted, @virt must be the address of
2260 * the first page of a block of pages what were allocated together (i.e
2261 * accounted together).
2263 * kvm_account_pgtable_pages() is thread-safe because mod_lruvec_page_state()
2266 static inline void kvm_account_pgtable_pages(void *virt, int nr)
2268 mod_lruvec_page_state(virt_to_page(virt), NR_SECONDARY_PAGETABLE, nr);
2272 * This defines how many reserved entries we want to keep before we
2273 * kick the vcpu to the userspace to avoid dirty ring full. This
2274 * value can be tuned to higher if e.g. PML is enabled on the host.
2276 #define KVM_DIRTY_RING_RSVD_ENTRIES 64
2278 /* Max number of entries allowed for each kvm dirty ring */
2279 #define KVM_DIRTY_RING_MAX_ENTRIES 65536