1/* SPDX-License-Identifier: GPL-2.0-only */
2#ifndef __KVM_HOST_H
3#define __KVM_HOST_H
4
5
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>
16#include <linux/mm.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/hashtable.h>
33#include <linux/interval_tree.h>
34#include <linux/rbtree.h>
35#include <linux/xarray.h>
36#include <asm/signal.h>
37
38#include <linux/kvm.h>
39#include <linux/kvm_para.h>
40
41#include <linux/kvm_types.h>
42
43#include <asm/kvm_host.h>
44#include <linux/kvm_dirty_ring.h>
45
46#ifndef KVM_MAX_VCPU_IDS
47#define KVM_MAX_VCPU_IDS KVM_MAX_VCPUS
48#endif
49
50/*
51 * The bit 16 ~ bit 31 of kvm_memory_region::flags are internally used
52 * in kvm, other bits are visible for userspace which are defined in
53 * include/linux/kvm_h.
54 */
55#define KVM_MEMSLOT_INVALID	(1UL << 16)
56
57/*
58 * Bit 63 of the memslot generation number is an "update in-progress flag",
59 * e.g. is temporarily set for the duration of install_new_memslots().
60 * This flag effectively creates a unique generation number that is used to
61 * mark cached memslot data, e.g. MMIO accesses, as potentially being stale,
62 * i.e. may (or may not) have come from the previous memslots generation.
63 *
64 * This is necessary because the actual memslots update is not atomic with
65 * respect to the generation number update.  Updating the generation number
66 * first would allow a vCPU to cache a spte from the old memslots using the
67 * new generation number, and updating the generation number after switching
68 * to the new memslots would allow cache hits using the old generation number
69 * to reference the defunct memslots.
70 *
71 * This mechanism is used to prevent getting hits in KVM's caches while a
72 * memslot update is in-progress, and to prevent cache hits *after* updating
73 * the actual generation number against accesses that were inserted into the
74 * cache *before* the memslots were updated.
75 */
76#define KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS	BIT_ULL(63)
77
78/* Two fragments for cross MMIO pages. */
79#define KVM_MAX_MMIO_FRAGMENTS	2
80
81#ifndef KVM_ADDRESS_SPACE_NUM
82#define KVM_ADDRESS_SPACE_NUM	1
83#endif
84
85/*
86 * For the normal pfn, the highest 12 bits should be zero,
87 * so we can mask bit 62 ~ bit 52  to indicate the error pfn,
88 * mask bit 63 to indicate the noslot pfn.
89 */
90#define KVM_PFN_ERR_MASK	(0x7ffULL << 52)
91#define KVM_PFN_ERR_NOSLOT_MASK	(0xfffULL << 52)
92#define KVM_PFN_NOSLOT		(0x1ULL << 63)
93
94#define KVM_PFN_ERR_FAULT	(KVM_PFN_ERR_MASK)
95#define KVM_PFN_ERR_HWPOISON	(KVM_PFN_ERR_MASK + 1)
96#define KVM_PFN_ERR_RO_FAULT	(KVM_PFN_ERR_MASK + 2)
97
98/*
99 * error pfns indicate that the gfn is in slot but faild to
100 * translate it to pfn on host.
101 */
102static inline bool is_error_pfn(kvm_pfn_t pfn)
103{
104	return !!(pfn & KVM_PFN_ERR_MASK);
105}
106
107/*
108 * error_noslot pfns indicate that the gfn can not be
109 * translated to pfn - it is not in slot or failed to
110 * translate it to pfn.
111 */
112static inline bool is_error_noslot_pfn(kvm_pfn_t pfn)
113{
114	return !!(pfn & KVM_PFN_ERR_NOSLOT_MASK);
115}
116
117/* noslot pfn indicates that the gfn is not in slot. */
118static inline bool is_noslot_pfn(kvm_pfn_t pfn)
119{
120	return pfn == KVM_PFN_NOSLOT;
121}
122
123/*
124 * architectures with KVM_HVA_ERR_BAD other than PAGE_OFFSET (e.g. s390)
125 * provide own defines and kvm_is_error_hva
126 */
127#ifndef KVM_HVA_ERR_BAD
128
129#define KVM_HVA_ERR_BAD		(PAGE_OFFSET)
130#define KVM_HVA_ERR_RO_BAD	(PAGE_OFFSET + PAGE_SIZE)
131
132static inline bool kvm_is_error_hva(unsigned long addr)
133{
134	return addr >= PAGE_OFFSET;
135}
136
137#endif
138
139#define KVM_ERR_PTR_BAD_PAGE	(ERR_PTR(-ENOENT))
140
141static inline bool is_error_page(struct page *page)
142{
143	return IS_ERR(page);
144}
145
146#define KVM_REQUEST_MASK           GENMASK(7,0)
147#define KVM_REQUEST_NO_WAKEUP      BIT(8)
148#define KVM_REQUEST_WAIT           BIT(9)
149/*
150 * Architecture-independent vcpu->requests bit members
151 * Bits 4-7 are reserved for more arch-independent bits.
152 */
153#define KVM_REQ_TLB_FLUSH         (0 | KVM_REQUEST_WAIT | KVM_REQUEST_NO_WAKEUP)
154#define KVM_REQ_MMU_RELOAD        (1 | KVM_REQUEST_WAIT | KVM_REQUEST_NO_WAKEUP)
155#define KVM_REQ_UNBLOCK           2
156#define KVM_REQ_UNHALT            3
157#define KVM_REQ_VM_DEAD           (4 | KVM_REQUEST_WAIT | KVM_REQUEST_NO_WAKEUP)
158#define KVM_REQ_GPC_INVALIDATE    (5 | KVM_REQUEST_WAIT | KVM_REQUEST_NO_WAKEUP)
159#define KVM_REQUEST_ARCH_BASE     8
160
161#define KVM_ARCH_REQ_FLAGS(nr, flags) ({ \
162	BUILD_BUG_ON((unsigned)(nr) >= (sizeof_field(struct kvm_vcpu, requests) * 8) - KVM_REQUEST_ARCH_BASE); \
163	(unsigned)(((nr) + KVM_REQUEST_ARCH_BASE) | (flags)); \
164})
165#define KVM_ARCH_REQ(nr)           KVM_ARCH_REQ_FLAGS(nr, 0)
166
167bool kvm_make_vcpus_request_mask(struct kvm *kvm, unsigned int req,
168				 unsigned long *vcpu_bitmap);
169bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req);
170bool kvm_make_all_cpus_request_except(struct kvm *kvm, unsigned int req,
171				      struct kvm_vcpu *except);
172bool kvm_make_cpus_request_mask(struct kvm *kvm, unsigned int req,
173				unsigned long *vcpu_bitmap);
174
175#define KVM_USERSPACE_IRQ_SOURCE_ID		0
176#define KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID	1
177
178extern struct mutex kvm_lock;
179extern struct list_head vm_list;
180
181struct kvm_io_range {
182	gpa_t addr;
183	int len;
184	struct kvm_io_device *dev;
185};
186
187#define NR_IOBUS_DEVS 1000
188
189struct kvm_io_bus {
190	int dev_count;
191	int ioeventfd_count;
192	struct kvm_io_range range[];
193};
194
195enum kvm_bus {
196	KVM_MMIO_BUS,
197	KVM_PIO_BUS,
198	KVM_VIRTIO_CCW_NOTIFY_BUS,
199	KVM_FAST_MMIO_BUS,
200	KVM_NR_BUSES
201};
202
203int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
204		     int len, const void *val);
205int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
206			    gpa_t addr, int len, const void *val, long cookie);
207int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
208		    int len, void *val);
209int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
210			    int len, struct kvm_io_device *dev);
211int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
212			      struct kvm_io_device *dev);
213struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
214					 gpa_t addr);
215
216#ifdef CONFIG_KVM_ASYNC_PF
217struct kvm_async_pf {
218	struct work_struct work;
219	struct list_head link;
220	struct list_head queue;
221	struct kvm_vcpu *vcpu;
222	struct mm_struct *mm;
223	gpa_t cr2_or_gpa;
224	unsigned long addr;
225	struct kvm_arch_async_pf arch;
226	bool   wakeup_all;
227	bool notpresent_injected;
228};
229
230void kvm_clear_async_pf_completion_queue(struct kvm_vcpu *vcpu);
231void kvm_check_async_pf_completion(struct kvm_vcpu *vcpu);
232bool kvm_setup_async_pf(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
233			unsigned long hva, struct kvm_arch_async_pf *arch);
234int kvm_async_pf_wakeup_all(struct kvm_vcpu *vcpu);
235#endif
236
237#ifdef KVM_ARCH_WANT_MMU_NOTIFIER
238struct kvm_gfn_range {
239	struct kvm_memory_slot *slot;
240	gfn_t start;
241	gfn_t end;
242	pte_t pte;
243	bool may_block;
244};
245bool kvm_unmap_gfn_range(struct kvm *kvm, struct kvm_gfn_range *range);
246bool kvm_age_gfn(struct kvm *kvm, struct kvm_gfn_range *range);
247bool kvm_test_age_gfn(struct kvm *kvm, struct kvm_gfn_range *range);
248bool kvm_set_spte_gfn(struct kvm *kvm, struct kvm_gfn_range *range);
249#endif
250
251enum {
252	OUTSIDE_GUEST_MODE,
253	IN_GUEST_MODE,
254	EXITING_GUEST_MODE,
255	READING_SHADOW_PAGE_TABLES,
256};
257
258#define KVM_UNMAPPED_PAGE	((void *) 0x500 + POISON_POINTER_DELTA)
259
260struct kvm_host_map {
261	/*
262	 * Only valid if the 'pfn' is managed by the host kernel (i.e. There is
263	 * a 'struct page' for it. When using mem= kernel parameter some memory
264	 * can be used as guest memory but they are not managed by host
265	 * kernel).
266	 * If 'pfn' is not managed by the host kernel, this field is
267	 * initialized to KVM_UNMAPPED_PAGE.
268	 */
269	struct page *page;
270	void *hva;
271	kvm_pfn_t pfn;
272	kvm_pfn_t gfn;
273};
274
275/*
276 * Used to check if the mapping is valid or not. Never use 'kvm_host_map'
277 * directly to check for that.
278 */
279static inline bool kvm_vcpu_mapped(struct kvm_host_map *map)
280{
281	return !!map->hva;
282}
283
284static inline bool kvm_vcpu_can_poll(ktime_t cur, ktime_t stop)
285{
286	return single_task_running() && !need_resched() && ktime_before(cur, stop);
287}
288
289/*
290 * Sometimes a large or cross-page mmio needs to be broken up into separate
291 * exits for userspace servicing.
292 */
293struct kvm_mmio_fragment {
294	gpa_t gpa;
295	void *data;
296	unsigned len;
297};
298
299struct kvm_vcpu {
300	struct kvm *kvm;
301#ifdef CONFIG_PREEMPT_NOTIFIERS
302	struct preempt_notifier preempt_notifier;
303#endif
304	int cpu;
305	int vcpu_id; /* id given by userspace at creation */
306	int vcpu_idx; /* index in kvm->vcpus array */
307	int srcu_idx;
308	int mode;
309	u64 requests;
310	unsigned long guest_debug;
311
312	int pre_pcpu;
313	struct list_head blocked_vcpu_list;
314
315	struct mutex mutex;
316	struct kvm_run *run;
317
318#ifndef __KVM_HAVE_ARCH_WQP
319	struct rcuwait wait;
320#endif
321	struct pid __rcu *pid;
322	int sigset_active;
323	sigset_t sigset;
324	unsigned int halt_poll_ns;
325	bool valid_wakeup;
326
327#ifdef CONFIG_HAS_IOMEM
328	int mmio_needed;
329	int mmio_read_completed;
330	int mmio_is_write;
331	int mmio_cur_fragment;
332	int mmio_nr_fragments;
333	struct kvm_mmio_fragment mmio_fragments[KVM_MAX_MMIO_FRAGMENTS];
334#endif
335
336#ifdef CONFIG_KVM_ASYNC_PF
337	struct {
338		u32 queued;
339		struct list_head queue;
340		struct list_head done;
341		spinlock_t lock;
342	} async_pf;
343#endif
344
345#ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
346	/*
347	 * Cpu relax intercept or pause loop exit optimization
348	 * in_spin_loop: set when a vcpu does a pause loop exit
349	 *  or cpu relax intercepted.
350	 * dy_eligible: indicates whether vcpu is eligible for directed yield.
351	 */
352	struct {
353		bool in_spin_loop;
354		bool dy_eligible;
355	} spin_loop;
356#endif
357	bool preempted;
358	bool ready;
359	struct kvm_vcpu_arch arch;
360	struct kvm_vcpu_stat stat;
361	char stats_id[KVM_STATS_NAME_SIZE];
362	struct kvm_dirty_ring dirty_ring;
363
364	/*
365	 * The most recently used memslot by this vCPU and the slots generation
366	 * for which it is valid.
367	 * No wraparound protection is needed since generations won't overflow in
368	 * thousands of years, even assuming 1M memslot operations per second.
369	 */
370	struct kvm_memory_slot *last_used_slot;
371	u64 last_used_slot_gen;
372};
373
374/* must be called with irqs disabled */
375static __always_inline void guest_enter_irqoff(void)
376{
377	/*
378	 * This is running in ioctl context so its safe to assume that it's the
379	 * stime pending cputime to flush.
380	 */
381	instrumentation_begin();
382	vtime_account_guest_enter();
383	instrumentation_end();
384
385	/*
386	 * KVM does not hold any references to rcu protected data when it
387	 * switches CPU into a guest mode. In fact switching to a guest mode
388	 * is very similar to exiting to userspace from rcu point of view. In
389	 * addition CPU may stay in a guest mode for quite a long time (up to
390	 * one time slice). Lets treat guest mode as quiescent state, just like
391	 * we do with user-mode execution.
392	 */
393	if (!context_tracking_guest_enter()) {
394		instrumentation_begin();
395		rcu_virt_note_context_switch(smp_processor_id());
396		instrumentation_end();
397	}
398}
399
400static __always_inline void guest_exit_irqoff(void)
401{
402	context_tracking_guest_exit();
403
404	instrumentation_begin();
405	/* Flush the guest cputime we spent on the guest */
406	vtime_account_guest_exit();
407	instrumentation_end();
408}
409
410static inline void guest_exit(void)
411{
412	unsigned long flags;
413
414	local_irq_save(flags);
415	guest_exit_irqoff();
416	local_irq_restore(flags);
417}
418
419static inline int kvm_vcpu_exiting_guest_mode(struct kvm_vcpu *vcpu)
420{
421	/*
422	 * The memory barrier ensures a previous write to vcpu->requests cannot
423	 * be reordered with the read of vcpu->mode.  It pairs with the general
424	 * memory barrier following the write of vcpu->mode in VCPU RUN.
425	 */
426	smp_mb__before_atomic();
427	return cmpxchg(&vcpu->mode, IN_GUEST_MODE, EXITING_GUEST_MODE);
428}
429
430/*
431 * Some of the bitops functions do not support too long bitmaps.
432 * This number must be determined not to exceed such limits.
433 */
434#define KVM_MEM_MAX_NR_PAGES ((1UL << 31) - 1)
435
436/*
437 * Since at idle each memslot belongs to two memslot sets it has to contain
438 * two embedded nodes for each data structure that it forms a part of.
439 *
440 * Two memslot sets (one active and one inactive) are necessary so the VM
441 * continues to run on one memslot set while the other is being modified.
442 *
443 * These two memslot sets normally point to the same set of memslots.
444 * They can, however, be desynchronized when performing a memslot management
445 * operation by replacing the memslot to be modified by its copy.
446 * After the operation is complete, both memslot sets once again point to
447 * the same, common set of memslot data.
448 *
449 * The memslots themselves are independent of each other so they can be
450 * individually added or deleted.
451 */
452struct kvm_memory_slot {
453	struct hlist_node id_node[2];
454	struct interval_tree_node hva_node[2];
455	struct rb_node gfn_node[2];
456	gfn_t base_gfn;
457	unsigned long npages;
458	unsigned long *dirty_bitmap;
459	struct kvm_arch_memory_slot arch;
460	unsigned long userspace_addr;
461	u32 flags;
462	short id;
463	u16 as_id;
464};
465
466static inline bool kvm_slot_dirty_track_enabled(const struct kvm_memory_slot *slot)
467{
468	return slot->flags & KVM_MEM_LOG_DIRTY_PAGES;
469}
470
471static inline unsigned long kvm_dirty_bitmap_bytes(struct kvm_memory_slot *memslot)
472{
473	return ALIGN(memslot->npages, BITS_PER_LONG) / 8;
474}
475
476static inline unsigned long *kvm_second_dirty_bitmap(struct kvm_memory_slot *memslot)
477{
478	unsigned long len = kvm_dirty_bitmap_bytes(memslot);
479
480	return memslot->dirty_bitmap + len / sizeof(*memslot->dirty_bitmap);
481}
482
483#ifndef KVM_DIRTY_LOG_MANUAL_CAPS
484#define KVM_DIRTY_LOG_MANUAL_CAPS KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE
485#endif
486
487struct kvm_s390_adapter_int {
488	u64 ind_addr;
489	u64 summary_addr;
490	u64 ind_offset;
491	u32 summary_offset;
492	u32 adapter_id;
493};
494
495struct kvm_hv_sint {
496	u32 vcpu;
497	u32 sint;
498};
499
500struct kvm_xen_evtchn {
501	u32 port;
502	u32 vcpu;
503	u32 priority;
504};
505
506struct kvm_kernel_irq_routing_entry {
507	u32 gsi;
508	u32 type;
509	int (*set)(struct kvm_kernel_irq_routing_entry *e,
510		   struct kvm *kvm, int irq_source_id, int level,
511		   bool line_status);
512	union {
513		struct {
514			unsigned irqchip;
515			unsigned pin;
516		} irqchip;
517		struct {
518			u32 address_lo;
519			u32 address_hi;
520			u32 data;
521			u32 flags;
522			u32 devid;
523		} msi;
524		struct kvm_s390_adapter_int adapter;
525		struct kvm_hv_sint hv_sint;
526		struct kvm_xen_evtchn xen_evtchn;
527	};
528	struct hlist_node link;
529};
530
531#ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
532struct kvm_irq_routing_table {
533	int chip[KVM_NR_IRQCHIPS][KVM_IRQCHIP_NUM_PINS];
534	u32 nr_rt_entries;
535	/*
536	 * Array indexed by gsi. Each entry contains list of irq chips
537	 * the gsi is connected to.
538	 */
539	struct hlist_head map[];
540};
541#endif
542
543#ifndef KVM_PRIVATE_MEM_SLOTS
544#define KVM_PRIVATE_MEM_SLOTS 0
545#endif
546
547#define KVM_MEM_SLOTS_NUM SHRT_MAX
548#define KVM_USER_MEM_SLOTS (KVM_MEM_SLOTS_NUM - KVM_PRIVATE_MEM_SLOTS)
549
550#ifndef __KVM_VCPU_MULTIPLE_ADDRESS_SPACE
551static inline int kvm_arch_vcpu_memslots_id(struct kvm_vcpu *vcpu)
552{
553	return 0;
554}
555#endif
556
557struct kvm_memslots {
558	u64 generation;
559	atomic_long_t last_used_slot;
560	struct rb_root_cached hva_tree;
561	struct rb_root gfn_tree;
562	/*
563	 * The mapping table from slot id to memslot.
564	 *
565	 * 7-bit bucket count matches the size of the old id to index array for
566	 * 512 slots, while giving good performance with this slot count.
567	 * Higher bucket counts bring only small performance improvements but
568	 * always result in higher memory usage (even for lower memslot counts).
569	 */
570	DECLARE_HASHTABLE(id_hash, 7);
571	int node_idx;
572};
573
574struct kvm {
575#ifdef KVM_HAVE_MMU_RWLOCK
576	rwlock_t mmu_lock;
577#else
578	spinlock_t mmu_lock;
579#endif /* KVM_HAVE_MMU_RWLOCK */
580
581	struct mutex slots_lock;
582
583	/*
584	 * Protects the arch-specific fields of struct kvm_memory_slots in
585	 * use by the VM. To be used under the slots_lock (above) or in a
586	 * kvm->srcu critical section where acquiring the slots_lock would
587	 * lead to deadlock with the synchronize_srcu in
588	 * install_new_memslots.
589	 */
590	struct mutex slots_arch_lock;
591	struct mm_struct *mm; /* userspace tied to this vm */
592	unsigned long nr_memslot_pages;
593	/* The two memslot sets - active and inactive (per address space) */
594	struct kvm_memslots __memslots[KVM_ADDRESS_SPACE_NUM][2];
595	/* The current active memslot set for each address space */
596	struct kvm_memslots __rcu *memslots[KVM_ADDRESS_SPACE_NUM];
597	struct xarray vcpu_array;
598
599	/* Used to wait for completion of MMU notifiers.  */
600	spinlock_t mn_invalidate_lock;
601	unsigned long mn_active_invalidate_count;
602	struct rcuwait mn_memslots_update_rcuwait;
603
604	/* For management / invalidation of gfn_to_pfn_caches */
605	spinlock_t gpc_lock;
606	struct list_head gpc_list;
607
608	/*
609	 * created_vcpus is protected by kvm->lock, and is incremented
610	 * at the beginning of KVM_CREATE_VCPU.  online_vcpus is only
611	 * incremented after storing the kvm_vcpu pointer in vcpus,
612	 * and is accessed atomically.
613	 */
614	atomic_t online_vcpus;
615	int created_vcpus;
616	int last_boosted_vcpu;
617	struct list_head vm_list;
618	struct mutex lock;
619	struct kvm_io_bus __rcu *buses[KVM_NR_BUSES];
620#ifdef CONFIG_HAVE_KVM_EVENTFD
621	struct {
622		spinlock_t        lock;
623		struct list_head  items;
624		struct list_head  resampler_list;
625		struct mutex      resampler_lock;
626	} irqfds;
627	struct list_head ioeventfds;
628#endif
629	struct kvm_vm_stat stat;
630	struct kvm_arch arch;
631	refcount_t users_count;
632#ifdef CONFIG_KVM_MMIO
633	struct kvm_coalesced_mmio_ring *coalesced_mmio_ring;
634	spinlock_t ring_lock;
635	struct list_head coalesced_zones;
636#endif
637
638	struct mutex irq_lock;
639#ifdef CONFIG_HAVE_KVM_IRQCHIP
640	/*
641	 * Update side is protected by irq_lock.
642	 */
643	struct kvm_irq_routing_table __rcu *irq_routing;
644#endif
645#ifdef CONFIG_HAVE_KVM_IRQFD
646	struct hlist_head irq_ack_notifier_list;
647#endif
648
649#if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
650	struct mmu_notifier mmu_notifier;
651	unsigned long mmu_notifier_seq;
652	long mmu_notifier_count;
653	unsigned long mmu_notifier_range_start;
654	unsigned long mmu_notifier_range_end;
655#endif
656	struct list_head devices;
657	u64 manual_dirty_log_protect;
658	struct dentry *debugfs_dentry;
659	struct kvm_stat_data **debugfs_stat_data;
660	struct srcu_struct srcu;
661	struct srcu_struct irq_srcu;
662	pid_t userspace_pid;
663	unsigned int max_halt_poll_ns;
664	u32 dirty_ring_size;
665	bool vm_bugged;
666	bool vm_dead;
667
668#ifdef CONFIG_HAVE_KVM_PM_NOTIFIER
669	struct notifier_block pm_notifier;
670#endif
671	char stats_id[KVM_STATS_NAME_SIZE];
672};
673
674#define kvm_err(fmt, ...) \
675	pr_err("kvm [%i]: " fmt, task_pid_nr(current), ## __VA_ARGS__)
676#define kvm_info(fmt, ...) \
677	pr_info("kvm [%i]: " fmt, task_pid_nr(current), ## __VA_ARGS__)
678#define kvm_debug(fmt, ...) \
679	pr_debug("kvm [%i]: " fmt, task_pid_nr(current), ## __VA_ARGS__)
680#define kvm_debug_ratelimited(fmt, ...) \
681	pr_debug_ratelimited("kvm [%i]: " fmt, task_pid_nr(current), \
682			     ## __VA_ARGS__)
683#define kvm_pr_unimpl(fmt, ...) \
684	pr_err_ratelimited("kvm [%i]: " fmt, \
685			   task_tgid_nr(current), ## __VA_ARGS__)
686
687/* The guest did something we don't support. */
688#define vcpu_unimpl(vcpu, fmt, ...)					\
689	kvm_pr_unimpl("vcpu%i, guest rIP: 0x%lx " fmt,			\
690			(vcpu)->vcpu_id, kvm_rip_read(vcpu), ## __VA_ARGS__)
691
692#define vcpu_debug(vcpu, fmt, ...)					\
693	kvm_debug("vcpu%i " fmt, (vcpu)->vcpu_id, ## __VA_ARGS__)
694#define vcpu_debug_ratelimited(vcpu, fmt, ...)				\
695	kvm_debug_ratelimited("vcpu%i " fmt, (vcpu)->vcpu_id,           \
696			      ## __VA_ARGS__)
697#define vcpu_err(vcpu, fmt, ...)					\
698	kvm_err("vcpu%i " fmt, (vcpu)->vcpu_id, ## __VA_ARGS__)
699
700static inline void kvm_vm_dead(struct kvm *kvm)
701{
702	kvm->vm_dead = true;
703	kvm_make_all_cpus_request(kvm, KVM_REQ_VM_DEAD);
704}
705
706static inline void kvm_vm_bugged(struct kvm *kvm)
707{
708	kvm->vm_bugged = true;
709	kvm_vm_dead(kvm);
710}
711
712
713#define KVM_BUG(cond, kvm, fmt...)				\
714({								\
715	int __ret = (cond);					\
716								\
717	if (WARN_ONCE(__ret && !(kvm)->vm_bugged, fmt))		\
718		kvm_vm_bugged(kvm);				\
719	unlikely(__ret);					\
720})
721
722#define KVM_BUG_ON(cond, kvm)					\
723({								\
724	int __ret = (cond);					\
725								\
726	if (WARN_ON_ONCE(__ret && !(kvm)->vm_bugged))		\
727		kvm_vm_bugged(kvm);				\
728	unlikely(__ret);					\
729})
730
731static inline bool kvm_dirty_log_manual_protect_and_init_set(struct kvm *kvm)
732{
733	return !!(kvm->manual_dirty_log_protect & KVM_DIRTY_LOG_INITIALLY_SET);
734}
735
736static inline struct kvm_io_bus *kvm_get_bus(struct kvm *kvm, enum kvm_bus idx)
737{
738	return srcu_dereference_check(kvm->buses[idx], &kvm->srcu,
739				      lockdep_is_held(&kvm->slots_lock) ||
740				      !refcount_read(&kvm->users_count));
741}
742
743static inline struct kvm_vcpu *kvm_get_vcpu(struct kvm *kvm, int i)
744{
745	int num_vcpus = atomic_read(&kvm->online_vcpus);
746	i = array_index_nospec(i, num_vcpus);
747
748	/* Pairs with smp_wmb() in kvm_vm_ioctl_create_vcpu.  */
749	smp_rmb();
750	return xa_load(&kvm->vcpu_array, i);
751}
752
753#define kvm_for_each_vcpu(idx, vcpup, kvm)		   \
754	xa_for_each_range(&kvm->vcpu_array, idx, vcpup, 0, \
755			  (atomic_read(&kvm->online_vcpus) - 1))
756
757static inline struct kvm_vcpu *kvm_get_vcpu_by_id(struct kvm *kvm, int id)
758{
759	struct kvm_vcpu *vcpu = NULL;
760	unsigned long i;
761
762	if (id < 0)
763		return NULL;
764	if (id < KVM_MAX_VCPUS)
765		vcpu = kvm_get_vcpu(kvm, id);
766	if (vcpu && vcpu->vcpu_id == id)
767		return vcpu;
768	kvm_for_each_vcpu(i, vcpu, kvm)
769		if (vcpu->vcpu_id == id)
770			return vcpu;
771	return NULL;
772}
773
774static inline int kvm_vcpu_get_idx(struct kvm_vcpu *vcpu)
775{
776	return vcpu->vcpu_idx;
777}
778
779void kvm_destroy_vcpus(struct kvm *kvm);
780
781void vcpu_load(struct kvm_vcpu *vcpu);
782void vcpu_put(struct kvm_vcpu *vcpu);
783
784#ifdef __KVM_HAVE_IOAPIC
785void kvm_arch_post_irq_ack_notifier_list_update(struct kvm *kvm);
786void kvm_arch_post_irq_routing_update(struct kvm *kvm);
787#else
788static inline void kvm_arch_post_irq_ack_notifier_list_update(struct kvm *kvm)
789{
790}
791static inline void kvm_arch_post_irq_routing_update(struct kvm *kvm)
792{
793}
794#endif
795
796#ifdef CONFIG_HAVE_KVM_IRQFD
797int kvm_irqfd_init(void);
798void kvm_irqfd_exit(void);
799#else
800static inline int kvm_irqfd_init(void)
801{
802	return 0;
803}
804
805static inline void kvm_irqfd_exit(void)
806{
807}
808#endif
809int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
810		  struct module *module);
811void kvm_exit(void);
812
813void kvm_get_kvm(struct kvm *kvm);
814bool kvm_get_kvm_safe(struct kvm *kvm);
815void kvm_put_kvm(struct kvm *kvm);
816bool file_is_kvm(struct file *file);
817void kvm_put_kvm_no_destroy(struct kvm *kvm);
818
819static inline struct kvm_memslots *__kvm_memslots(struct kvm *kvm, int as_id)
820{
821	as_id = array_index_nospec(as_id, KVM_ADDRESS_SPACE_NUM);
822	return srcu_dereference_check(kvm->memslots[as_id], &kvm->srcu,
823			lockdep_is_held(&kvm->slots_lock) ||
824			!refcount_read(&kvm->users_count));
825}
826
827static inline struct kvm_memslots *kvm_memslots(struct kvm *kvm)
828{
829	return __kvm_memslots(kvm, 0);
830}
831
832static inline struct kvm_memslots *kvm_vcpu_memslots(struct kvm_vcpu *vcpu)
833{
834	int as_id = kvm_arch_vcpu_memslots_id(vcpu);
835
836	return __kvm_memslots(vcpu->kvm, as_id);
837}
838
839static inline bool kvm_memslots_empty(struct kvm_memslots *slots)
840{
841	return RB_EMPTY_ROOT(&slots->gfn_tree);
842}
843
844#define kvm_for_each_memslot(memslot, bkt, slots)			      \
845	hash_for_each(slots->id_hash, bkt, memslot, id_node[slots->node_idx]) \
846		if (WARN_ON_ONCE(!memslot->npages)) {			      \
847		} else
848
849static inline
850struct kvm_memory_slot *id_to_memslot(struct kvm_memslots *slots, int id)
851{
852	struct kvm_memory_slot *slot;
853	int idx = slots->node_idx;
854
855	hash_for_each_possible(slots->id_hash, slot, id_node[idx], id) {
856		if (slot->id == id)
857			return slot;
858	}
859
860	return NULL;
861}
862
863/* Iterator used for walking memslots that overlap a gfn range. */
864struct kvm_memslot_iter {
865	struct kvm_memslots *slots;
866	struct rb_node *node;
867	struct kvm_memory_slot *slot;
868};
869
870static inline void kvm_memslot_iter_next(struct kvm_memslot_iter *iter)
871{
872	iter->node = rb_next(iter->node);
873	if (!iter->node)
874		return;
875
876	iter->slot = container_of(iter->node, struct kvm_memory_slot, gfn_node[iter->slots->node_idx]);
877}
878
879static inline void kvm_memslot_iter_start(struct kvm_memslot_iter *iter,
880					  struct kvm_memslots *slots,
881					  gfn_t start)
882{
883	int idx = slots->node_idx;
884	struct rb_node *tmp;
885	struct kvm_memory_slot *slot;
886
887	iter->slots = slots;
888
889	/*
890	 * Find the so called "upper bound" of a key - the first node that has
891	 * its key strictly greater than the searched one (the start gfn in our case).
892	 */
893	iter->node = NULL;
894	for (tmp = slots->gfn_tree.rb_node; tmp; ) {
895		slot = container_of(tmp, struct kvm_memory_slot, gfn_node[idx]);
896		if (start < slot->base_gfn) {
897			iter->node = tmp;
898			tmp = tmp->rb_left;
899		} else {
900			tmp = tmp->rb_right;
901		}
902	}
903
904	/*
905	 * Find the slot with the lowest gfn that can possibly intersect with
906	 * the range, so we'll ideally have slot start <= range start
907	 */
908	if (iter->node) {
909		/*
910		 * A NULL previous node means that the very first slot
911		 * already has a higher start gfn.
912		 * In this case slot start > range start.
913		 */
914		tmp = rb_prev(iter->node);
915		if (tmp)
916			iter->node = tmp;
917	} else {
918		/* a NULL node below means no slots */
919		iter->node = rb_last(&slots->gfn_tree);
920	}
921
922	if (iter->node) {
923		iter->slot = container_of(iter->node, struct kvm_memory_slot, gfn_node[idx]);
924
925		/*
926		 * It is possible in the slot start < range start case that the
927		 * found slot ends before or at range start (slot end <= range start)
928		 * and so it does not overlap the requested range.
929		 *
930		 * In such non-overlapping case the next slot (if it exists) will
931		 * already have slot start > range start, otherwise the logic above
932		 * would have found it instead of the current slot.
933		 */
934		if (iter->slot->base_gfn + iter->slot->npages <= start)
935			kvm_memslot_iter_next(iter);
936	}
937}
938
939static inline bool kvm_memslot_iter_is_valid(struct kvm_memslot_iter *iter, gfn_t end)
940{
941	if (!iter->node)
942		return false;
943
944	/*
945	 * If this slot starts beyond or at the end of the range so does
946	 * every next one
947	 */
948	return iter->slot->base_gfn < end;
949}
950
951/* Iterate over each memslot at least partially intersecting [start, end) range */
952#define kvm_for_each_memslot_in_gfn_range(iter, slots, start, end)	\
953	for (kvm_memslot_iter_start(iter, slots, start);		\
954	     kvm_memslot_iter_is_valid(iter, end);			\
955	     kvm_memslot_iter_next(iter))
956
957/*
958 * KVM_SET_USER_MEMORY_REGION ioctl allows the following operations:
959 * - create a new memory slot
960 * - delete an existing memory slot
961 * - modify an existing memory slot
962 *   -- move it in the guest physical memory space
963 *   -- just change its flags
964 *
965 * Since flags can be changed by some of these operations, the following
966 * differentiation is the best we can do for __kvm_set_memory_region():
967 */
968enum kvm_mr_change {
969	KVM_MR_CREATE,
970	KVM_MR_DELETE,
971	KVM_MR_MOVE,
972	KVM_MR_FLAGS_ONLY,
973};
974
975int kvm_set_memory_region(struct kvm *kvm,
976			  const struct kvm_userspace_memory_region *mem);
977int __kvm_set_memory_region(struct kvm *kvm,
978			    const struct kvm_userspace_memory_region *mem);
979void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot);
980void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen);
981int kvm_arch_prepare_memory_region(struct kvm *kvm,
982				const struct kvm_memory_slot *old,
983				struct kvm_memory_slot *new,
984				enum kvm_mr_change change);
985void kvm_arch_commit_memory_region(struct kvm *kvm,
986				struct kvm_memory_slot *old,
987				const struct kvm_memory_slot *new,
988				enum kvm_mr_change change);
989/* flush all memory translations */
990void kvm_arch_flush_shadow_all(struct kvm *kvm);
991/* flush memory translations pointing to 'slot' */
992void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
993				   struct kvm_memory_slot *slot);
994
995int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
996			    struct page **pages, int nr_pages);
997
998struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn);
999unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn);
1000unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable);
1001unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot, gfn_t gfn);
1002unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot, gfn_t gfn,
1003				      bool *writable);
1004void kvm_release_page_clean(struct page *page);
1005void kvm_release_page_dirty(struct page *page);
1006void kvm_set_page_accessed(struct page *page);
1007
1008kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn);
1009kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1010		      bool *writable);
1011kvm_pfn_t gfn_to_pfn_memslot(const struct kvm_memory_slot *slot, gfn_t gfn);
1012kvm_pfn_t gfn_to_pfn_memslot_atomic(const struct kvm_memory_slot *slot, gfn_t gfn);
1013kvm_pfn_t __gfn_to_pfn_memslot(const struct kvm_memory_slot *slot, gfn_t gfn,
1014			       bool atomic, bool *async, bool write_fault,
1015			       bool *writable, hva_t *hva);
1016
1017void kvm_release_pfn_clean(kvm_pfn_t pfn);
1018void kvm_release_pfn_dirty(kvm_pfn_t pfn);
1019void kvm_set_pfn_dirty(kvm_pfn_t pfn);
1020void kvm_set_pfn_accessed(kvm_pfn_t pfn);
1021
1022void kvm_release_pfn(kvm_pfn_t pfn, bool dirty);
1023int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1024			int len);
1025int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len);
1026int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1027			   void *data, unsigned long len);
1028int kvm_read_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1029				 void *data, unsigned int offset,
1030				 unsigned long len);
1031int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1032			 int offset, int len);
1033int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1034		    unsigned long len);
1035int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1036			   void *data, unsigned long len);
1037int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1038				  void *data, unsigned int offset,
1039				  unsigned long len);
1040int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1041			      gpa_t gpa, unsigned long len);
1042
1043#define __kvm_get_guest(kvm, gfn, offset, v)				\
1044({									\
1045	unsigned long __addr = gfn_to_hva(kvm, gfn);			\
1046	typeof(v) __user *__uaddr = (typeof(__uaddr))(__addr + offset);	\
1047	int __ret = -EFAULT;						\
1048									\
1049	if (!kvm_is_error_hva(__addr))					\
1050		__ret = get_user(v, __uaddr);				\
1051	__ret;								\
1052})
1053
1054#define kvm_get_guest(kvm, gpa, v)					\
1055({									\
1056	gpa_t __gpa = gpa;						\
1057	struct kvm *__kvm = kvm;					\
1058									\
1059	__kvm_get_guest(__kvm, __gpa >> PAGE_SHIFT,			\
1060			offset_in_page(__gpa), v);			\
1061})
1062
1063#define __kvm_put_guest(kvm, gfn, offset, v)				\
1064({									\
1065	unsigned long __addr = gfn_to_hva(kvm, gfn);			\
1066	typeof(v) __user *__uaddr = (typeof(__uaddr))(__addr + offset);	\
1067	int __ret = -EFAULT;						\
1068									\
1069	if (!kvm_is_error_hva(__addr))					\
1070		__ret = put_user(v, __uaddr);				\
1071	if (!__ret)							\
1072		mark_page_dirty(kvm, gfn);				\
1073	__ret;								\
1074})
1075
1076#define kvm_put_guest(kvm, gpa, v)					\
1077({									\
1078	gpa_t __gpa = gpa;						\
1079	struct kvm *__kvm = kvm;					\
1080									\
1081	__kvm_put_guest(__kvm, __gpa >> PAGE_SHIFT,			\
1082			offset_in_page(__gpa), v);			\
1083})
1084
1085int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len);
1086struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn);
1087bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn);
1088bool kvm_vcpu_is_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn);
1089unsigned long kvm_host_page_size(struct kvm_vcpu *vcpu, gfn_t gfn);
1090void mark_page_dirty_in_slot(struct kvm *kvm, const struct kvm_memory_slot *memslot, gfn_t gfn);
1091void mark_page_dirty(struct kvm *kvm, gfn_t gfn);
1092
1093struct kvm_memslots *kvm_vcpu_memslots(struct kvm_vcpu *vcpu);
1094struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn);
1095kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn);
1096kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn);
1097int kvm_vcpu_map(struct kvm_vcpu *vcpu, gpa_t gpa, struct kvm_host_map *map);
1098struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn);
1099void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map, bool dirty);
1100unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn);
1101unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable);
1102int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data, int offset,
1103			     int len);
1104int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa, void *data,
1105			       unsigned long len);
1106int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data,
1107			unsigned long len);
1108int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, const void *data,
1109			      int offset, int len);
1110int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
1111			 unsigned long len);
1112void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn);
1113
1114/**
1115 * kvm_gfn_to_pfn_cache_init - prepare a cached kernel mapping and HPA for a
1116 *                             given guest physical address.
1117 *
1118 * @kvm:	   pointer to kvm instance.
1119 * @gpc:	   struct gfn_to_pfn_cache object.
1120 * @vcpu:	   vCPU to be used for marking pages dirty and to be woken on
1121 *		   invalidation.
1122 * @guest_uses_pa: indicates that the resulting host physical PFN is used while
1123 *		   @vcpu is IN_GUEST_MODE so invalidations should wake it.
1124 * @kernel_map:    requests a kernel virtual mapping (kmap / memremap).
1125 * @gpa:	   guest physical address to map.
1126 * @len:	   sanity check; the range being access must fit a single page.
1127 * @dirty:         mark the cache dirty immediately.
1128 *
1129 * @return:	   0 for success.
1130 *		   -EINVAL for a mapping which would cross a page boundary.
1131 *                 -EFAULT for an untranslatable guest physical address.
1132 *
1133 * This primes a gfn_to_pfn_cache and links it into the @kvm's list for
1134 * invalidations to be processed. Invalidation callbacks to @vcpu using
1135 * %KVM_REQ_GPC_INVALIDATE will occur only for MMU notifiers, not for KVM
1136 * memslot changes. Callers are required to use kvm_gfn_to_pfn_cache_check()
1137 * to ensure that the cache is valid before accessing the target page.
1138 */
1139int kvm_gfn_to_pfn_cache_init(struct kvm *kvm, struct gfn_to_pfn_cache *gpc,
1140			      struct kvm_vcpu *vcpu, bool guest_uses_pa,
1141			      bool kernel_map, gpa_t gpa, unsigned long len,
1142			      bool dirty);
1143
1144/**
1145 * kvm_gfn_to_pfn_cache_check - check validity of a gfn_to_pfn_cache.
1146 *
1147 * @kvm:	   pointer to kvm instance.
1148 * @gpc:	   struct gfn_to_pfn_cache object.
1149 * @gpa:	   current guest physical address to map.
1150 * @len:	   sanity check; the range being access must fit a single page.
1151 * @dirty:         mark the cache dirty immediately.
1152 *
1153 * @return:	   %true if the cache is still valid and the address matches.
1154 *		   %false if the cache is not valid.
1155 *
1156 * Callers outside IN_GUEST_MODE context should hold a read lock on @gpc->lock
1157 * while calling this function, and then continue to hold the lock until the
1158 * access is complete.
1159 *
1160 * Callers in IN_GUEST_MODE may do so without locking, although they should
1161 * still hold a read lock on kvm->scru for the memslot checks.
1162 */
1163bool kvm_gfn_to_pfn_cache_check(struct kvm *kvm, struct gfn_to_pfn_cache *gpc,
1164				gpa_t gpa, unsigned long len);
1165
1166/**
1167 * kvm_gfn_to_pfn_cache_refresh - update a previously initialized cache.
1168 *
1169 * @kvm:	   pointer to kvm instance.
1170 * @gpc:	   struct gfn_to_pfn_cache object.
1171 * @gpa:	   updated guest physical address to map.
1172 * @len:	   sanity check; the range being access must fit a single page.
1173 * @dirty:         mark the cache dirty immediately.
1174 *
1175 * @return:	   0 for success.
1176 *		   -EINVAL for a mapping which would cross a page boundary.
1177 *                 -EFAULT for an untranslatable guest physical address.
1178 *
1179 * This will attempt to refresh a gfn_to_pfn_cache. Note that a successful
1180 * returm from this function does not mean the page can be immediately
1181 * accessed because it may have raced with an invalidation. Callers must
1182 * still lock and check the cache status, as this function does not return
1183 * with the lock still held to permit access.
1184 */
1185int kvm_gfn_to_pfn_cache_refresh(struct kvm *kvm, struct gfn_to_pfn_cache *gpc,
1186				 gpa_t gpa, unsigned long len, bool dirty);
1187
1188/**
1189 * kvm_gfn_to_pfn_cache_unmap - temporarily unmap a gfn_to_pfn_cache.
1190 *
1191 * @kvm:	   pointer to kvm instance.
1192 * @gpc:	   struct gfn_to_pfn_cache object.
1193 *
1194 * This unmaps the referenced page and marks it dirty, if appropriate. The
1195 * cache is left in the invalid state but at least the mapping from GPA to
1196 * userspace HVA will remain cached and can be reused on a subsequent
1197 * refresh.
1198 */
1199void kvm_gfn_to_pfn_cache_unmap(struct kvm *kvm, struct gfn_to_pfn_cache *gpc);
1200
1201/**
1202 * kvm_gfn_to_pfn_cache_destroy - destroy and unlink a gfn_to_pfn_cache.
1203 *
1204 * @kvm:	   pointer to kvm instance.
1205 * @gpc:	   struct gfn_to_pfn_cache object.
1206 *
1207 * This removes a cache from the @kvm's list to be processed on MMU notifier
1208 * invocation.
1209 */
1210void kvm_gfn_to_pfn_cache_destroy(struct kvm *kvm, struct gfn_to_pfn_cache *gpc);
1211
1212void kvm_sigset_activate(struct kvm_vcpu *vcpu);
1213void kvm_sigset_deactivate(struct kvm_vcpu *vcpu);
1214
1215void kvm_vcpu_halt(struct kvm_vcpu *vcpu);
1216bool kvm_vcpu_block(struct kvm_vcpu *vcpu);
1217void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu);
1218void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu);
1219bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu);
1220void kvm_vcpu_kick(struct kvm_vcpu *vcpu);
1221int kvm_vcpu_yield_to(struct kvm_vcpu *target);
1222void kvm_vcpu_on_spin(struct kvm_vcpu *vcpu, bool usermode_vcpu_not_eligible);
1223
1224void kvm_flush_remote_tlbs(struct kvm *kvm);
1225void kvm_reload_remote_mmus(struct kvm *kvm);
1226
1227#ifdef KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE
1228int kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache *mc, int min);
1229int kvm_mmu_memory_cache_nr_free_objects(struct kvm_mmu_memory_cache *mc);
1230void kvm_mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc);
1231void *kvm_mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc);
1232#endif
1233
1234void kvm_inc_notifier_count(struct kvm *kvm, unsigned long start,
1235				   unsigned long end);
1236void kvm_dec_notifier_count(struct kvm *kvm, unsigned long start,
1237				   unsigned long end);
1238
1239long kvm_arch_dev_ioctl(struct file *filp,
1240			unsigned int ioctl, unsigned long arg);
1241long kvm_arch_vcpu_ioctl(struct file *filp,
1242			 unsigned int ioctl, unsigned long arg);
1243vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf);
1244
1245int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext);
1246
1247void kvm_arch_mmu_enable_log_dirty_pt_masked(struct kvm *kvm,
1248					struct kvm_memory_slot *slot,
1249					gfn_t gfn_offset,
1250					unsigned long mask);
1251void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot);
1252
1253#ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1254void kvm_arch_flush_remote_tlbs_memslot(struct kvm *kvm,
1255					const struct kvm_memory_slot *memslot);
1256#else /* !CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
1257int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log);
1258int kvm_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log,
1259		      int *is_dirty, struct kvm_memory_slot **memslot);
1260#endif
1261
1262int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
1263			bool line_status);
1264int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
1265			    struct kvm_enable_cap *cap);
1266long kvm_arch_vm_ioctl(struct file *filp,
1267		       unsigned int ioctl, unsigned long arg);
1268
1269int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu);
1270int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu);
1271
1272int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
1273				    struct kvm_translation *tr);
1274
1275int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs);
1276int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs);
1277int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
1278				  struct kvm_sregs *sregs);
1279int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
1280				  struct kvm_sregs *sregs);
1281int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
1282				    struct kvm_mp_state *mp_state);
1283int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
1284				    struct kvm_mp_state *mp_state);
1285int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
1286					struct kvm_guest_debug *dbg);
1287int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu);
1288
1289int kvm_arch_init(void *opaque);
1290void kvm_arch_exit(void);
1291
1292void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu);
1293
1294void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu);
1295void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu);
1296int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id);
1297int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu);
1298void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu);
1299void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu);
1300
1301#ifdef CONFIG_HAVE_KVM_PM_NOTIFIER
1302int kvm_arch_pm_notifier(struct kvm *kvm, unsigned long state);
1303#endif
1304
1305#ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS
1306void kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu, struct dentry *debugfs_dentry);
1307#endif
1308
1309int kvm_arch_hardware_enable(void);
1310void kvm_arch_hardware_disable(void);
1311int kvm_arch_hardware_setup(void *opaque);
1312void kvm_arch_hardware_unsetup(void);
1313int kvm_arch_check_processor_compat(void *opaque);
1314int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu);
1315bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu);
1316int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu);
1317bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu);
1318bool kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu);
1319int kvm_arch_post_init_vm(struct kvm *kvm);
1320void kvm_arch_pre_destroy_vm(struct kvm *kvm);
1321int kvm_arch_create_vm_debugfs(struct kvm *kvm);
1322
1323#ifndef __KVM_HAVE_ARCH_VM_ALLOC
1324/*
1325 * All architectures that want to use vzalloc currently also
1326 * need their own kvm_arch_alloc_vm implementation.
1327 */
1328static inline struct kvm *kvm_arch_alloc_vm(void)
1329{
1330	return kzalloc(sizeof(struct kvm), GFP_KERNEL);
1331}
1332#endif
1333
1334static inline void __kvm_arch_free_vm(struct kvm *kvm)
1335{
1336	kvfree(kvm);
1337}
1338
1339#ifndef __KVM_HAVE_ARCH_VM_FREE
1340static inline void kvm_arch_free_vm(struct kvm *kvm)
1341{
1342	__kvm_arch_free_vm(kvm);
1343}
1344#endif
1345
1346#ifndef __KVM_HAVE_ARCH_FLUSH_REMOTE_TLB
1347static inline int kvm_arch_flush_remote_tlb(struct kvm *kvm)
1348{
1349	return -ENOTSUPP;
1350}
1351#endif
1352
1353#ifdef __KVM_HAVE_ARCH_NONCOHERENT_DMA
1354void kvm_arch_register_noncoherent_dma(struct kvm *kvm);
1355void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm);
1356bool kvm_arch_has_noncoherent_dma(struct kvm *kvm);
1357#else
1358static inline void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
1359{
1360}
1361
1362static inline void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
1363{
1364}
1365
1366static inline bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
1367{
1368	return false;
1369}
1370#endif
1371#ifdef __KVM_HAVE_ARCH_ASSIGNED_DEVICE
1372void kvm_arch_start_assignment(struct kvm *kvm);
1373void kvm_arch_end_assignment(struct kvm *kvm);
1374bool kvm_arch_has_assigned_device(struct kvm *kvm);
1375#else
1376static inline void kvm_arch_start_assignment(struct kvm *kvm)
1377{
1378}
1379
1380static inline void kvm_arch_end_assignment(struct kvm *kvm)
1381{
1382}
1383
1384static inline bool kvm_arch_has_assigned_device(struct kvm *kvm)
1385{
1386	return false;
1387}
1388#endif
1389
1390static inline struct rcuwait *kvm_arch_vcpu_get_wait(struct kvm_vcpu *vcpu)
1391{
1392#ifdef __KVM_HAVE_ARCH_WQP
1393	return vcpu->arch.waitp;
1394#else
1395	return &vcpu->wait;
1396#endif
1397}
1398
1399/*
1400 * Wake a vCPU if necessary, but don't do any stats/metadata updates.  Returns
1401 * true if the vCPU was blocking and was awakened, false otherwise.
1402 */
1403static inline bool __kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
1404{
1405	return !!rcuwait_wake_up(kvm_arch_vcpu_get_wait(vcpu));
1406}
1407
1408static inline bool kvm_vcpu_is_blocking(struct kvm_vcpu *vcpu)
1409{
1410	return rcuwait_active(kvm_arch_vcpu_get_wait(vcpu));
1411}
1412
1413#ifdef __KVM_HAVE_ARCH_INTC_INITIALIZED
1414/*
1415 * returns true if the virtual interrupt controller is initialized and
1416 * ready to accept virtual IRQ. On some architectures the virtual interrupt
1417 * controller is dynamically instantiated and this is not always true.
1418 */
1419bool kvm_arch_intc_initialized(struct kvm *kvm);
1420#else
1421static inline bool kvm_arch_intc_initialized(struct kvm *kvm)
1422{
1423	return true;
1424}
1425#endif
1426
1427#ifdef CONFIG_GUEST_PERF_EVENTS
1428unsigned long kvm_arch_vcpu_get_ip(struct kvm_vcpu *vcpu);
1429
1430void kvm_register_perf_callbacks(unsigned int (*pt_intr_handler)(void));
1431void kvm_unregister_perf_callbacks(void);
1432#else
1433static inline void kvm_register_perf_callbacks(void *ign) {}
1434static inline void kvm_unregister_perf_callbacks(void) {}
1435#endif /* CONFIG_GUEST_PERF_EVENTS */
1436
1437int kvm_arch_init_vm(struct kvm *kvm, unsigned long type);
1438void kvm_arch_destroy_vm(struct kvm *kvm);
1439void kvm_arch_sync_events(struct kvm *kvm);
1440
1441int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu);
1442
1443bool kvm_is_reserved_pfn(kvm_pfn_t pfn);
1444bool kvm_is_zone_device_pfn(kvm_pfn_t pfn);
1445
1446struct kvm_irq_ack_notifier {
1447	struct hlist_node link;
1448	unsigned gsi;
1449	void (*irq_acked)(struct kvm_irq_ack_notifier *kian);
1450};
1451
1452int kvm_irq_map_gsi(struct kvm *kvm,
1453		    struct kvm_kernel_irq_routing_entry *entries, int gsi);
1454int kvm_irq_map_chip_pin(struct kvm *kvm, unsigned irqchip, unsigned pin);
1455
1456int kvm_set_irq(struct kvm *kvm, int irq_source_id, u32 irq, int level,
1457		bool line_status);
1458int kvm_set_msi(struct kvm_kernel_irq_routing_entry *irq_entry, struct kvm *kvm,
1459		int irq_source_id, int level, bool line_status);
1460int kvm_arch_set_irq_inatomic(struct kvm_kernel_irq_routing_entry *e,
1461			       struct kvm *kvm, int irq_source_id,
1462			       int level, bool line_status);
1463bool kvm_irq_has_notifier(struct kvm *kvm, unsigned irqchip, unsigned pin);
1464void kvm_notify_acked_gsi(struct kvm *kvm, int gsi);
1465void kvm_notify_acked_irq(struct kvm *kvm, unsigned irqchip, unsigned pin);
1466void kvm_register_irq_ack_notifier(struct kvm *kvm,
1467				   struct kvm_irq_ack_notifier *kian);
1468void kvm_unregister_irq_ack_notifier(struct kvm *kvm,
1469				   struct kvm_irq_ack_notifier *kian);
1470int kvm_request_irq_source_id(struct kvm *kvm);
1471void kvm_free_irq_source_id(struct kvm *kvm, int irq_source_id);
1472bool kvm_arch_irqfd_allowed(struct kvm *kvm, struct kvm_irqfd *args);
1473
1474/*
1475 * Returns a pointer to the memslot if it contains gfn.
1476 * Otherwise returns NULL.
1477 */
1478static inline struct kvm_memory_slot *
1479try_get_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1480{
1481	if (!slot)
1482		return NULL;
1483
1484	if (gfn >= slot->base_gfn && gfn < slot->base_gfn + slot->npages)
1485		return slot;
1486	else
1487		return NULL;
1488}
1489
1490/*
1491 * Returns a pointer to the memslot that contains gfn. Otherwise returns NULL.
1492 *
1493 * With "approx" set returns the memslot also when the address falls
1494 * in a hole. In that case one of the memslots bordering the hole is
1495 * returned.
1496 */
1497static inline struct kvm_memory_slot *
1498search_memslots(struct kvm_memslots *slots, gfn_t gfn, bool approx)
1499{
1500	struct kvm_memory_slot *slot;
1501	struct rb_node *node;
1502	int idx = slots->node_idx;
1503
1504	slot = NULL;
1505	for (node = slots->gfn_tree.rb_node; node; ) {
1506		slot = container_of(node, struct kvm_memory_slot, gfn_node[idx]);
1507		if (gfn >= slot->base_gfn) {
1508			if (gfn < slot->base_gfn + slot->npages)
1509				return slot;
1510			node = node->rb_right;
1511		} else
1512			node = node->rb_left;
1513	}
1514
1515	return approx ? slot : NULL;
1516}
1517
1518static inline struct kvm_memory_slot *
1519____gfn_to_memslot(struct kvm_memslots *slots, gfn_t gfn, bool approx)
1520{
1521	struct kvm_memory_slot *slot;
1522
1523	slot = (struct kvm_memory_slot *)atomic_long_read(&slots->last_used_slot);
1524	slot = try_get_memslot(slot, gfn);
1525	if (slot)
1526		return slot;
1527
1528	slot = search_memslots(slots, gfn, approx);
1529	if (slot) {
1530		atomic_long_set(&slots->last_used_slot, (unsigned long)slot);
1531		return slot;
1532	}
1533
1534	return NULL;
1535}
1536
1537/*
1538 * __gfn_to_memslot() and its descendants are here to allow arch code to inline
1539 * the lookups in hot paths.  gfn_to_memslot() itself isn't here as an inline
1540 * because that would bloat other code too much.
1541 */
1542static inline struct kvm_memory_slot *
1543__gfn_to_memslot(struct kvm_memslots *slots, gfn_t gfn)
1544{
1545	return ____gfn_to_memslot(slots, gfn, false);
1546}
1547
1548static inline unsigned long
1549__gfn_to_hva_memslot(const struct kvm_memory_slot *slot, gfn_t gfn)
1550{
1551	/*
1552	 * The index was checked originally in search_memslots.  To avoid
1553	 * that a malicious guest builds a Spectre gadget out of e.g. page
1554	 * table walks, do not let the processor speculate loads outside
1555	 * the guest's registered memslots.
1556	 */
1557	unsigned long offset = gfn - slot->base_gfn;
1558	offset = array_index_nospec(offset, slot->npages);
1559	return slot->userspace_addr + offset * PAGE_SIZE;
1560}
1561
1562static inline int memslot_id(struct kvm *kvm, gfn_t gfn)
1563{
1564	return gfn_to_memslot(kvm, gfn)->id;
1565}
1566
1567static inline gfn_t
1568hva_to_gfn_memslot(unsigned long hva, struct kvm_memory_slot *slot)
1569{
1570	gfn_t gfn_offset = (hva - slot->userspace_addr) >> PAGE_SHIFT;
1571
1572	return slot->base_gfn + gfn_offset;
1573}
1574
1575static inline gpa_t gfn_to_gpa(gfn_t gfn)
1576{
1577	return (gpa_t)gfn << PAGE_SHIFT;
1578}
1579
1580static inline gfn_t gpa_to_gfn(gpa_t gpa)
1581{
1582	return (gfn_t)(gpa >> PAGE_SHIFT);
1583}
1584
1585static inline hpa_t pfn_to_hpa(kvm_pfn_t pfn)
1586{
1587	return (hpa_t)pfn << PAGE_SHIFT;
1588}
1589
1590static inline struct page *kvm_vcpu_gpa_to_page(struct kvm_vcpu *vcpu,
1591						gpa_t gpa)
1592{
1593	return kvm_vcpu_gfn_to_page(vcpu, gpa_to_gfn(gpa));
1594}
1595
1596static inline bool kvm_is_error_gpa(struct kvm *kvm, gpa_t gpa)
1597{
1598	unsigned long hva = gfn_to_hva(kvm, gpa_to_gfn(gpa));
1599
1600	return kvm_is_error_hva(hva);
1601}
1602
1603enum kvm_stat_kind {
1604	KVM_STAT_VM,
1605	KVM_STAT_VCPU,
1606};
1607
1608struct kvm_stat_data {
1609	struct kvm *kvm;
1610	const struct _kvm_stats_desc *desc;
1611	enum kvm_stat_kind kind;
1612};
1613
1614struct _kvm_stats_desc {
1615	struct kvm_stats_desc desc;
1616	char name[KVM_STATS_NAME_SIZE];
1617};
1618
1619#define STATS_DESC_COMMON(type, unit, base, exp, sz, bsz)		       \
1620	.flags = type | unit | base |					       \
1621		 BUILD_BUG_ON_ZERO(type & ~KVM_STATS_TYPE_MASK) |	       \
1622		 BUILD_BUG_ON_ZERO(unit & ~KVM_STATS_UNIT_MASK) |	       \
1623		 BUILD_BUG_ON_ZERO(base & ~KVM_STATS_BASE_MASK),	       \
1624	.exponent = exp,						       \
1625	.size = sz,							       \
1626	.bucket_size = bsz
1627
1628#define VM_GENERIC_STATS_DESC(stat, type, unit, base, exp, sz, bsz)	       \
1629	{								       \
1630		{							       \
1631			STATS_DESC_COMMON(type, unit, base, exp, sz, bsz),     \
1632			.offset = offsetof(struct kvm_vm_stat, generic.stat)   \
1633		},							       \
1634		.name = #stat,						       \
1635	}
1636#define VCPU_GENERIC_STATS_DESC(stat, type, unit, base, exp, sz, bsz)	       \
1637	{								       \
1638		{							       \
1639			STATS_DESC_COMMON(type, unit, base, exp, sz, bsz),     \
1640			.offset = offsetof(struct kvm_vcpu_stat, generic.stat) \
1641		},							       \
1642		.name = #stat,						       \
1643	}
1644#define VM_STATS_DESC(stat, type, unit, base, exp, sz, bsz)		       \
1645	{								       \
1646		{							       \
1647			STATS_DESC_COMMON(type, unit, base, exp, sz, bsz),     \
1648			.offset = offsetof(struct kvm_vm_stat, stat)	       \
1649		},							       \
1650		.name = #stat,						       \
1651	}
1652#define VCPU_STATS_DESC(stat, type, unit, base, exp, sz, bsz)		       \
1653	{								       \
1654		{							       \
1655			STATS_DESC_COMMON(type, unit, base, exp, sz, bsz),     \
1656			.offset = offsetof(struct kvm_vcpu_stat, stat)	       \
1657		},							       \
1658		.name = #stat,						       \
1659	}
1660/* SCOPE: VM, VM_GENERIC, VCPU, VCPU_GENERIC */
1661#define STATS_DESC(SCOPE, stat, type, unit, base, exp, sz, bsz)		       \
1662	SCOPE##_STATS_DESC(stat, type, unit, base, exp, sz, bsz)
1663
1664#define STATS_DESC_CUMULATIVE(SCOPE, name, unit, base, exponent)	       \
1665	STATS_DESC(SCOPE, name, KVM_STATS_TYPE_CUMULATIVE,		       \
1666		unit, base, exponent, 1, 0)
1667#define STATS_DESC_INSTANT(SCOPE, name, unit, base, exponent)		       \
1668	STATS_DESC(SCOPE, name, KVM_STATS_TYPE_INSTANT,			       \
1669		unit, base, exponent, 1, 0)
1670#define STATS_DESC_PEAK(SCOPE, name, unit, base, exponent)		       \
1671	STATS_DESC(SCOPE, name, KVM_STATS_TYPE_PEAK,			       \
1672		unit, base, exponent, 1, 0)
1673#define STATS_DESC_LINEAR_HIST(SCOPE, name, unit, base, exponent, sz, bsz)     \
1674	STATS_DESC(SCOPE, name, KVM_STATS_TYPE_LINEAR_HIST,		       \
1675		unit, base, exponent, sz, bsz)
1676#define STATS_DESC_LOG_HIST(SCOPE, name, unit, base, exponent, sz)	       \
1677	STATS_DESC(SCOPE, name, KVM_STATS_TYPE_LOG_HIST,		       \
1678		unit, base, exponent, sz, 0)
1679
1680/* Cumulative counter, read/write */
1681#define STATS_DESC_COUNTER(SCOPE, name)					       \
1682	STATS_DESC_CUMULATIVE(SCOPE, name, KVM_STATS_UNIT_NONE,		       \
1683		KVM_STATS_BASE_POW10, 0)
1684/* Instantaneous counter, read only */
1685#define STATS_DESC_ICOUNTER(SCOPE, name)				       \
1686	STATS_DESC_INSTANT(SCOPE, name, KVM_STATS_UNIT_NONE,		       \
1687		KVM_STATS_BASE_POW10, 0)
1688/* Peak counter, read/write */
1689#define STATS_DESC_PCOUNTER(SCOPE, name)				       \
1690	STATS_DESC_PEAK(SCOPE, name, KVM_STATS_UNIT_NONE,		       \
1691		KVM_STATS_BASE_POW10, 0)
1692
1693/* Cumulative time in nanosecond */
1694#define STATS_DESC_TIME_NSEC(SCOPE, name)				       \
1695	STATS_DESC_CUMULATIVE(SCOPE, name, KVM_STATS_UNIT_SECONDS,	       \
1696		KVM_STATS_BASE_POW10, -9)
1697/* Linear histogram for time in nanosecond */
1698#define STATS_DESC_LINHIST_TIME_NSEC(SCOPE, name, sz, bsz)		       \
1699	STATS_DESC_LINEAR_HIST(SCOPE, name, KVM_STATS_UNIT_SECONDS,	       \
1700		KVM_STATS_BASE_POW10, -9, sz, bsz)
1701/* Logarithmic histogram for time in nanosecond */
1702#define STATS_DESC_LOGHIST_TIME_NSEC(SCOPE, name, sz)			       \
1703	STATS_DESC_LOG_HIST(SCOPE, name, KVM_STATS_UNIT_SECONDS,	       \
1704		KVM_STATS_BASE_POW10, -9, sz)
1705
1706#define KVM_GENERIC_VM_STATS()						       \
1707	STATS_DESC_COUNTER(VM_GENERIC, remote_tlb_flush),		       \
1708	STATS_DESC_COUNTER(VM_GENERIC, remote_tlb_flush_requests)
1709
1710#define KVM_GENERIC_VCPU_STATS()					       \
1711	STATS_DESC_COUNTER(VCPU_GENERIC, halt_successful_poll),		       \
1712	STATS_DESC_COUNTER(VCPU_GENERIC, halt_attempted_poll),		       \
1713	STATS_DESC_COUNTER(VCPU_GENERIC, halt_poll_invalid),		       \
1714	STATS_DESC_COUNTER(VCPU_GENERIC, halt_wakeup),			       \
1715	STATS_DESC_TIME_NSEC(VCPU_GENERIC, halt_poll_success_ns),	       \
1716	STATS_DESC_TIME_NSEC(VCPU_GENERIC, halt_poll_fail_ns),		       \
1717	STATS_DESC_TIME_NSEC(VCPU_GENERIC, halt_wait_ns),		       \
1718	STATS_DESC_LOGHIST_TIME_NSEC(VCPU_GENERIC, halt_poll_success_hist,     \
1719			HALT_POLL_HIST_COUNT),				       \
1720	STATS_DESC_LOGHIST_TIME_NSEC(VCPU_GENERIC, halt_poll_fail_hist,	       \
1721			HALT_POLL_HIST_COUNT),				       \
1722	STATS_DESC_LOGHIST_TIME_NSEC(VCPU_GENERIC, halt_wait_hist,	       \
1723			HALT_POLL_HIST_COUNT),				       \
1724	STATS_DESC_ICOUNTER(VCPU_GENERIC, blocking)
1725
1726extern struct dentry *kvm_debugfs_dir;
1727
1728ssize_t kvm_stats_read(char *id, const struct kvm_stats_header *header,
1729		       const struct _kvm_stats_desc *desc,
1730		       void *stats, size_t size_stats,
1731		       char __user *user_buffer, size_t size, loff_t *offset);
1732
1733/**
1734 * kvm_stats_linear_hist_update() - Update bucket value for linear histogram
1735 * statistics data.
1736 *
1737 * @data: start address of the stats data
1738 * @size: the number of bucket of the stats data
1739 * @value: the new value used to update the linear histogram's bucket
1740 * @bucket_size: the size (width) of a bucket
1741 */
1742static inline void kvm_stats_linear_hist_update(u64 *data, size_t size,
1743						u64 value, size_t bucket_size)
1744{
1745	size_t index = div64_u64(value, bucket_size);
1746
1747	index = min(index, size - 1);
1748	++data[index];
1749}
1750
1751/**
1752 * kvm_stats_log_hist_update() - Update bucket value for logarithmic histogram
1753 * statistics data.
1754 *
1755 * @data: start address of the stats data
1756 * @size: the number of bucket of the stats data
1757 * @value: the new value used to update the logarithmic histogram's bucket
1758 */
1759static inline void kvm_stats_log_hist_update(u64 *data, size_t size, u64 value)
1760{
1761	size_t index = fls64(value);
1762
1763	index = min(index, size - 1);
1764	++data[index];
1765}
1766
1767#define KVM_STATS_LINEAR_HIST_UPDATE(array, value, bsize)		       \
1768	kvm_stats_linear_hist_update(array, ARRAY_SIZE(array), value, bsize)
1769#define KVM_STATS_LOG_HIST_UPDATE(array, value)				       \
1770	kvm_stats_log_hist_update(array, ARRAY_SIZE(array), value)
1771
1772
1773extern const struct kvm_stats_header kvm_vm_stats_header;
1774extern const struct _kvm_stats_desc kvm_vm_stats_desc[];
1775extern const struct kvm_stats_header kvm_vcpu_stats_header;
1776extern const struct _kvm_stats_desc kvm_vcpu_stats_desc[];
1777
1778#if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1779static inline int mmu_notifier_retry(struct kvm *kvm, unsigned long mmu_seq)
1780{
1781	if (unlikely(kvm->mmu_notifier_count))
1782		return 1;
1783	/*
1784	 * Ensure the read of mmu_notifier_count happens before the read
1785	 * of mmu_notifier_seq.  This interacts with the smp_wmb() in
1786	 * mmu_notifier_invalidate_range_end to make sure that the caller
1787	 * either sees the old (non-zero) value of mmu_notifier_count or
1788	 * the new (incremented) value of mmu_notifier_seq.
1789	 * PowerPC Book3s HV KVM calls this under a per-page lock
1790	 * rather than under kvm->mmu_lock, for scalability, so
1791	 * can't rely on kvm->mmu_lock to keep things ordered.
1792	 */
1793	smp_rmb();
1794	if (kvm->mmu_notifier_seq != mmu_seq)
1795		return 1;
1796	return 0;
1797}
1798
1799static inline int mmu_notifier_retry_hva(struct kvm *kvm,
1800					 unsigned long mmu_seq,
1801					 unsigned long hva)
1802{
1803	lockdep_assert_held(&kvm->mmu_lock);
1804	/*
1805	 * If mmu_notifier_count is non-zero, then the range maintained by
1806	 * kvm_mmu_notifier_invalidate_range_start contains all addresses that
1807	 * might be being invalidated. Note that it may include some false
1808	 * positives, due to shortcuts when handing concurrent invalidations.
1809	 */
1810	if (unlikely(kvm->mmu_notifier_count) &&
1811	    hva >= kvm->mmu_notifier_range_start &&
1812	    hva < kvm->mmu_notifier_range_end)
1813		return 1;
1814	if (kvm->mmu_notifier_seq != mmu_seq)
1815		return 1;
1816	return 0;
1817}
1818#endif
1819
1820#ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
1821
1822#define KVM_MAX_IRQ_ROUTES 4096 /* might need extension/rework in the future */
1823
1824bool kvm_arch_can_set_irq_routing(struct kvm *kvm);
1825int kvm_set_irq_routing(struct kvm *kvm,
1826			const struct kvm_irq_routing_entry *entries,
1827			unsigned nr,
1828			unsigned flags);
1829int kvm_set_routing_entry(struct kvm *kvm,
1830			  struct kvm_kernel_irq_routing_entry *e,
1831			  const struct kvm_irq_routing_entry *ue);
1832void kvm_free_irq_routing(struct kvm *kvm);
1833
1834#else
1835
1836static inline void kvm_free_irq_routing(struct kvm *kvm) {}
1837
1838#endif
1839
1840int kvm_send_userspace_msi(struct kvm *kvm, struct kvm_msi *msi);
1841
1842#ifdef CONFIG_HAVE_KVM_EVENTFD
1843
1844void kvm_eventfd_init(struct kvm *kvm);
1845int kvm_ioeventfd(struct kvm *kvm, struct kvm_ioeventfd *args);
1846
1847#ifdef CONFIG_HAVE_KVM_IRQFD
1848int kvm_irqfd(struct kvm *kvm, struct kvm_irqfd *args);
1849void kvm_irqfd_release(struct kvm *kvm);
1850void kvm_irq_routing_update(struct kvm *);
1851#else
1852static inline int kvm_irqfd(struct kvm *kvm, struct kvm_irqfd *args)
1853{
1854	return -EINVAL;
1855}
1856
1857static inline void kvm_irqfd_release(struct kvm *kvm) {}
1858#endif
1859
1860#else
1861
1862static inline void kvm_eventfd_init(struct kvm *kvm) {}
1863
1864static inline int kvm_irqfd(struct kvm *kvm, struct kvm_irqfd *args)
1865{
1866	return -EINVAL;
1867}
1868
1869static inline void kvm_irqfd_release(struct kvm *kvm) {}
1870
1871#ifdef CONFIG_HAVE_KVM_IRQCHIP
1872static inline void kvm_irq_routing_update(struct kvm *kvm)
1873{
1874}
1875#endif
1876
1877static inline int kvm_ioeventfd(struct kvm *kvm, struct kvm_ioeventfd *args)
1878{
1879	return -ENOSYS;
1880}
1881
1882#endif /* CONFIG_HAVE_KVM_EVENTFD */
1883
1884void kvm_arch_irq_routing_update(struct kvm *kvm);
1885
1886static inline void kvm_make_request(int req, struct kvm_vcpu *vcpu)
1887{
1888	/*
1889	 * Ensure the rest of the request is published to kvm_check_request's
1890	 * caller.  Paired with the smp_mb__after_atomic in kvm_check_request.
1891	 */
1892	smp_wmb();
1893	set_bit(req & KVM_REQUEST_MASK, (void *)&vcpu->requests);
1894}
1895
1896static inline bool kvm_request_pending(struct kvm_vcpu *vcpu)
1897{
1898	return READ_ONCE(vcpu->requests);
1899}
1900
1901static inline bool kvm_test_request(int req, struct kvm_vcpu *vcpu)
1902{
1903	return test_bit(req & KVM_REQUEST_MASK, (void *)&vcpu->requests);
1904}
1905
1906static inline void kvm_clear_request(int req, struct kvm_vcpu *vcpu)
1907{
1908	clear_bit(req & KVM_REQUEST_MASK, (void *)&vcpu->requests);
1909}
1910
1911static inline bool kvm_check_request(int req, struct kvm_vcpu *vcpu)
1912{
1913	if (kvm_test_request(req, vcpu)) {
1914		kvm_clear_request(req, vcpu);
1915
1916		/*
1917		 * Ensure the rest of the request is visible to kvm_check_request's
1918		 * caller.  Paired with the smp_wmb in kvm_make_request.
1919		 */
1920		smp_mb__after_atomic();
1921		return true;
1922	} else {
1923		return false;
1924	}
1925}
1926
1927extern bool kvm_rebooting;
1928
1929extern unsigned int halt_poll_ns;
1930extern unsigned int halt_poll_ns_grow;
1931extern unsigned int halt_poll_ns_grow_start;
1932extern unsigned int halt_poll_ns_shrink;
1933
1934struct kvm_device {
1935	const struct kvm_device_ops *ops;
1936	struct kvm *kvm;
1937	void *private;
1938	struct list_head vm_node;
1939};
1940
1941/* create, destroy, and name are mandatory */
1942struct kvm_device_ops {
1943	const char *name;
1944
1945	/*
1946	 * create is called holding kvm->lock and any operations not suitable
1947	 * to do while holding the lock should be deferred to init (see
1948	 * below).
1949	 */
1950	int (*create)(struct kvm_device *dev, u32 type);
1951
1952	/*
1953	 * init is called after create if create is successful and is called
1954	 * outside of holding kvm->lock.
1955	 */
1956	void (*init)(struct kvm_device *dev);
1957
1958	/*
1959	 * Destroy is responsible for freeing dev.
1960	 *
1961	 * Destroy may be called before or after destructors are called
1962	 * on emulated I/O regions, depending on whether a reference is
1963	 * held by a vcpu or other kvm component that gets destroyed
1964	 * after the emulated I/O.
1965	 */
1966	void (*destroy)(struct kvm_device *dev);
1967
1968	/*
1969	 * Release is an alternative method to free the device. It is
1970	 * called when the device file descriptor is closed. Once
1971	 * release is called, the destroy method will not be called
1972	 * anymore as the device is removed from the device list of
1973	 * the VM. kvm->lock is held.
1974	 */
1975	void (*release)(struct kvm_device *dev);
1976
1977	int (*set_attr)(struct kvm_device *dev, struct kvm_device_attr *attr);
1978	int (*get_attr)(struct kvm_device *dev, struct kvm_device_attr *attr);
1979	int (*has_attr)(struct kvm_device *dev, struct kvm_device_attr *attr);
1980	long (*ioctl)(struct kvm_device *dev, unsigned int ioctl,
1981		      unsigned long arg);
1982	int (*mmap)(struct kvm_device *dev, struct vm_area_struct *vma);
1983};
1984
1985void kvm_device_get(struct kvm_device *dev);
1986void kvm_device_put(struct kvm_device *dev);
1987struct kvm_device *kvm_device_from_filp(struct file *filp);
1988int kvm_register_device_ops(const struct kvm_device_ops *ops, u32 type);
1989void kvm_unregister_device_ops(u32 type);
1990
1991extern struct kvm_device_ops kvm_mpic_ops;
1992extern struct kvm_device_ops kvm_arm_vgic_v2_ops;
1993extern struct kvm_device_ops kvm_arm_vgic_v3_ops;
1994
1995#ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1996
1997static inline void kvm_vcpu_set_in_spin_loop(struct kvm_vcpu *vcpu, bool val)
1998{
1999	vcpu->spin_loop.in_spin_loop = val;
2000}
2001static inline void kvm_vcpu_set_dy_eligible(struct kvm_vcpu *vcpu, bool val)
2002{
2003	vcpu->spin_loop.dy_eligible = val;
2004}
2005
2006#else /* !CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT */
2007
2008static inline void kvm_vcpu_set_in_spin_loop(struct kvm_vcpu *vcpu, bool val)
2009{
2010}
2011
2012static inline void kvm_vcpu_set_dy_eligible(struct kvm_vcpu *vcpu, bool val)
2013{
2014}
2015#endif /* CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT */
2016
2017static inline bool kvm_is_visible_memslot(struct kvm_memory_slot *memslot)
2018{
2019	return (memslot && memslot->id < KVM_USER_MEM_SLOTS &&
2020		!(memslot->flags & KVM_MEMSLOT_INVALID));
2021}
2022
2023struct kvm_vcpu *kvm_get_running_vcpu(void);
2024struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void);
2025
2026#ifdef CONFIG_HAVE_KVM_IRQ_BYPASS
2027bool kvm_arch_has_irq_bypass(void);
2028int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *,
2029			   struct irq_bypass_producer *);
2030void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *,
2031			   struct irq_bypass_producer *);
2032void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *);
2033void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *);
2034int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
2035				  uint32_t guest_irq, bool set);
2036bool kvm_arch_irqfd_route_changed(struct kvm_kernel_irq_routing_entry *,
2037				  struct kvm_kernel_irq_routing_entry *);
2038#endif /* CONFIG_HAVE_KVM_IRQ_BYPASS */
2039
2040#ifdef CONFIG_HAVE_KVM_INVALID_WAKEUPS
2041/* If we wakeup during the poll time, was it a sucessful poll? */
2042static inline bool vcpu_valid_wakeup(struct kvm_vcpu *vcpu)
2043{
2044	return vcpu->valid_wakeup;
2045}
2046
2047#else
2048static inline bool vcpu_valid_wakeup(struct kvm_vcpu *vcpu)
2049{
2050	return true;
2051}
2052#endif /* CONFIG_HAVE_KVM_INVALID_WAKEUPS */
2053
2054#ifdef CONFIG_HAVE_KVM_NO_POLL
2055/* Callback that tells if we must not poll */
2056bool kvm_arch_no_poll(struct kvm_vcpu *vcpu);
2057#else
2058static inline bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
2059{
2060	return false;
2061}
2062#endif /* CONFIG_HAVE_KVM_NO_POLL */
2063
2064#ifdef CONFIG_HAVE_KVM_VCPU_ASYNC_IOCTL
2065long kvm_arch_vcpu_async_ioctl(struct file *filp,
2066			       unsigned int ioctl, unsigned long arg);
2067#else
2068static inline long kvm_arch_vcpu_async_ioctl(struct file *filp,
2069					     unsigned int ioctl,
2070					     unsigned long arg)
2071{
2072	return -ENOIOCTLCMD;
2073}
2074#endif /* CONFIG_HAVE_KVM_VCPU_ASYNC_IOCTL */
2075
2076void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
2077					    unsigned long start, unsigned long end);
2078
2079#ifdef CONFIG_HAVE_KVM_VCPU_RUN_PID_CHANGE
2080int kvm_arch_vcpu_run_pid_change(struct kvm_vcpu *vcpu);
2081#else
2082static inline int kvm_arch_vcpu_run_pid_change(struct kvm_vcpu *vcpu)
2083{
2084	return 0;
2085}
2086#endif /* CONFIG_HAVE_KVM_VCPU_RUN_PID_CHANGE */
2087
2088typedef int (*kvm_vm_thread_fn_t)(struct kvm *kvm, uintptr_t data);
2089
2090int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn,
2091				uintptr_t data, const char *name,
2092				struct task_struct **thread_ptr);
2093
2094#ifdef CONFIG_KVM_XFER_TO_GUEST_WORK
2095static inline void kvm_handle_signal_exit(struct kvm_vcpu *vcpu)
2096{
2097	vcpu->run->exit_reason = KVM_EXIT_INTR;
2098	vcpu->stat.signal_exits++;
2099}
2100#endif /* CONFIG_KVM_XFER_TO_GUEST_WORK */
2101
2102/*
2103 * This defines how many reserved entries we want to keep before we
2104 * kick the vcpu to the userspace to avoid dirty ring full.  This
2105 * value can be tuned to higher if e.g. PML is enabled on the host.
2106 */
2107#define  KVM_DIRTY_RING_RSVD_ENTRIES  64
2108
2109/* Max number of entries allowed for each kvm dirty ring */
2110#define  KVM_DIRTY_RING_MAX_ENTRIES  65536
2111
2112#endif
2113