xref: /openbmc/linux/include/linux/kvm_host.h (revision 78eee7b5f110c9884c8ffd1dfcdd9c29296f3e43)
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 <asm/signal.h>
33 
34 #include <linux/kvm.h>
35 #include <linux/kvm_para.h>
36 
37 #include <linux/kvm_types.h>
38 
39 #include <asm/kvm_host.h>
40 #include <linux/kvm_dirty_ring.h>
41 
42 #ifndef KVM_MAX_VCPU_ID
43 #define KVM_MAX_VCPU_ID KVM_MAX_VCPUS
44 #endif
45 
46 /*
47  * The bit 16 ~ bit 31 of kvm_memory_region::flags are internally used
48  * in kvm, other bits are visible for userspace which are defined in
49  * include/linux/kvm_h.
50  */
51 #define KVM_MEMSLOT_INVALID	(1UL << 16)
52 
53 /*
54  * Bit 63 of the memslot generation number is an "update in-progress flag",
55  * e.g. is temporarily set for the duration of install_new_memslots().
56  * This flag effectively creates a unique generation number that is used to
57  * mark cached memslot data, e.g. MMIO accesses, as potentially being stale,
58  * i.e. may (or may not) have come from the previous memslots generation.
59  *
60  * This is necessary because the actual memslots update is not atomic with
61  * respect to the generation number update.  Updating the generation number
62  * first would allow a vCPU to cache a spte from the old memslots using the
63  * new generation number, and updating the generation number after switching
64  * to the new memslots would allow cache hits using the old generation number
65  * to reference the defunct memslots.
66  *
67  * This mechanism is used to prevent getting hits in KVM's caches while a
68  * memslot update is in-progress, and to prevent cache hits *after* updating
69  * the actual generation number against accesses that were inserted into the
70  * cache *before* the memslots were updated.
71  */
72 #define KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS	BIT_ULL(63)
73 
74 /* Two fragments for cross MMIO pages. */
75 #define KVM_MAX_MMIO_FRAGMENTS	2
76 
77 #ifndef KVM_ADDRESS_SPACE_NUM
78 #define KVM_ADDRESS_SPACE_NUM	1
79 #endif
80 
81 /*
82  * For the normal pfn, the highest 12 bits should be zero,
83  * so we can mask bit 62 ~ bit 52  to indicate the error pfn,
84  * mask bit 63 to indicate the noslot pfn.
85  */
86 #define KVM_PFN_ERR_MASK	(0x7ffULL << 52)
87 #define KVM_PFN_ERR_NOSLOT_MASK	(0xfffULL << 52)
88 #define KVM_PFN_NOSLOT		(0x1ULL << 63)
89 
90 #define KVM_PFN_ERR_FAULT	(KVM_PFN_ERR_MASK)
91 #define KVM_PFN_ERR_HWPOISON	(KVM_PFN_ERR_MASK + 1)
92 #define KVM_PFN_ERR_RO_FAULT	(KVM_PFN_ERR_MASK + 2)
93 
94 /*
95  * error pfns indicate that the gfn is in slot but faild to
96  * translate it to pfn on host.
97  */
98 static inline bool is_error_pfn(kvm_pfn_t pfn)
99 {
100 	return !!(pfn & KVM_PFN_ERR_MASK);
101 }
102 
103 /*
104  * error_noslot pfns indicate that the gfn can not be
105  * translated to pfn - it is not in slot or failed to
106  * translate it to pfn.
107  */
108 static inline bool is_error_noslot_pfn(kvm_pfn_t pfn)
109 {
110 	return !!(pfn & KVM_PFN_ERR_NOSLOT_MASK);
111 }
112 
113 /* noslot pfn indicates that the gfn is not in slot. */
114 static inline bool is_noslot_pfn(kvm_pfn_t pfn)
115 {
116 	return pfn == KVM_PFN_NOSLOT;
117 }
118 
119 /*
120  * architectures with KVM_HVA_ERR_BAD other than PAGE_OFFSET (e.g. s390)
121  * provide own defines and kvm_is_error_hva
122  */
123 #ifndef KVM_HVA_ERR_BAD
124 
125 #define KVM_HVA_ERR_BAD		(PAGE_OFFSET)
126 #define KVM_HVA_ERR_RO_BAD	(PAGE_OFFSET + PAGE_SIZE)
127 
128 static inline bool kvm_is_error_hva(unsigned long addr)
129 {
130 	return addr >= PAGE_OFFSET;
131 }
132 
133 #endif
134 
135 #define KVM_ERR_PTR_BAD_PAGE	(ERR_PTR(-ENOENT))
136 
137 static inline bool is_error_page(struct page *page)
138 {
139 	return IS_ERR(page);
140 }
141 
142 #define KVM_REQUEST_MASK           GENMASK(7,0)
143 #define KVM_REQUEST_NO_WAKEUP      BIT(8)
144 #define KVM_REQUEST_WAIT           BIT(9)
145 /*
146  * Architecture-independent vcpu->requests bit members
147  * Bits 4-7 are reserved for more arch-independent bits.
148  */
149 #define KVM_REQ_TLB_FLUSH         (0 | KVM_REQUEST_WAIT | KVM_REQUEST_NO_WAKEUP)
150 #define KVM_REQ_MMU_RELOAD        (1 | KVM_REQUEST_WAIT | KVM_REQUEST_NO_WAKEUP)
151 #define KVM_REQ_UNBLOCK           2
152 #define KVM_REQ_UNHALT            3
153 #define KVM_REQ_VM_BUGGED         (4 | KVM_REQUEST_WAIT | KVM_REQUEST_NO_WAKEUP)
154 #define KVM_REQUEST_ARCH_BASE     8
155 
156 #define KVM_ARCH_REQ_FLAGS(nr, flags) ({ \
157 	BUILD_BUG_ON((unsigned)(nr) >= (sizeof_field(struct kvm_vcpu, requests) * 8) - KVM_REQUEST_ARCH_BASE); \
158 	(unsigned)(((nr) + KVM_REQUEST_ARCH_BASE) | (flags)); \
159 })
160 #define KVM_ARCH_REQ(nr)           KVM_ARCH_REQ_FLAGS(nr, 0)
161 
162 bool kvm_make_vcpus_request_mask(struct kvm *kvm, unsigned int req,
163 				 struct kvm_vcpu *except,
164 				 unsigned long *vcpu_bitmap, cpumask_var_t tmp);
165 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req);
166 bool kvm_make_all_cpus_request_except(struct kvm *kvm, unsigned int req,
167 				      struct kvm_vcpu *except);
168 bool kvm_make_cpus_request_mask(struct kvm *kvm, unsigned int req,
169 				unsigned long *vcpu_bitmap);
170 
171 #define KVM_USERSPACE_IRQ_SOURCE_ID		0
172 #define KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID	1
173 
174 extern struct mutex kvm_lock;
175 extern struct list_head vm_list;
176 
177 struct kvm_io_range {
178 	gpa_t addr;
179 	int len;
180 	struct kvm_io_device *dev;
181 };
182 
183 #define NR_IOBUS_DEVS 1000
184 
185 struct kvm_io_bus {
186 	int dev_count;
187 	int ioeventfd_count;
188 	struct kvm_io_range range[];
189 };
190 
191 enum kvm_bus {
192 	KVM_MMIO_BUS,
193 	KVM_PIO_BUS,
194 	KVM_VIRTIO_CCW_NOTIFY_BUS,
195 	KVM_FAST_MMIO_BUS,
196 	KVM_NR_BUSES
197 };
198 
199 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
200 		     int len, const void *val);
201 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
202 			    gpa_t addr, int len, const void *val, long cookie);
203 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
204 		    int len, void *val);
205 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
206 			    int len, struct kvm_io_device *dev);
207 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
208 			      struct kvm_io_device *dev);
209 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
210 					 gpa_t addr);
211 
212 #ifdef CONFIG_KVM_ASYNC_PF
213 struct kvm_async_pf {
214 	struct work_struct work;
215 	struct list_head link;
216 	struct list_head queue;
217 	struct kvm_vcpu *vcpu;
218 	struct mm_struct *mm;
219 	gpa_t cr2_or_gpa;
220 	unsigned long addr;
221 	struct kvm_arch_async_pf arch;
222 	bool   wakeup_all;
223 	bool notpresent_injected;
224 };
225 
226 void kvm_clear_async_pf_completion_queue(struct kvm_vcpu *vcpu);
227 void kvm_check_async_pf_completion(struct kvm_vcpu *vcpu);
228 bool kvm_setup_async_pf(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
229 			unsigned long hva, struct kvm_arch_async_pf *arch);
230 int kvm_async_pf_wakeup_all(struct kvm_vcpu *vcpu);
231 #endif
232 
233 #ifdef KVM_ARCH_WANT_MMU_NOTIFIER
234 struct kvm_gfn_range {
235 	struct kvm_memory_slot *slot;
236 	gfn_t start;
237 	gfn_t end;
238 	pte_t pte;
239 	bool may_block;
240 };
241 bool kvm_unmap_gfn_range(struct kvm *kvm, struct kvm_gfn_range *range);
242 bool kvm_age_gfn(struct kvm *kvm, struct kvm_gfn_range *range);
243 bool kvm_test_age_gfn(struct kvm *kvm, struct kvm_gfn_range *range);
244 bool kvm_set_spte_gfn(struct kvm *kvm, struct kvm_gfn_range *range);
245 #endif
246 
247 enum {
248 	OUTSIDE_GUEST_MODE,
249 	IN_GUEST_MODE,
250 	EXITING_GUEST_MODE,
251 	READING_SHADOW_PAGE_TABLES,
252 };
253 
254 #define KVM_UNMAPPED_PAGE	((void *) 0x500 + POISON_POINTER_DELTA)
255 
256 struct kvm_host_map {
257 	/*
258 	 * Only valid if the 'pfn' is managed by the host kernel (i.e. There is
259 	 * a 'struct page' for it. When using mem= kernel parameter some memory
260 	 * can be used as guest memory but they are not managed by host
261 	 * kernel).
262 	 * If 'pfn' is not managed by the host kernel, this field is
263 	 * initialized to KVM_UNMAPPED_PAGE.
264 	 */
265 	struct page *page;
266 	void *hva;
267 	kvm_pfn_t pfn;
268 	kvm_pfn_t gfn;
269 };
270 
271 /*
272  * Used to check if the mapping is valid or not. Never use 'kvm_host_map'
273  * directly to check for that.
274  */
275 static inline bool kvm_vcpu_mapped(struct kvm_host_map *map)
276 {
277 	return !!map->hva;
278 }
279 
280 static inline bool kvm_vcpu_can_poll(ktime_t cur, ktime_t stop)
281 {
282 	return single_task_running() && !need_resched() && ktime_before(cur, stop);
283 }
284 
285 /*
286  * Sometimes a large or cross-page mmio needs to be broken up into separate
287  * exits for userspace servicing.
288  */
289 struct kvm_mmio_fragment {
290 	gpa_t gpa;
291 	void *data;
292 	unsigned len;
293 };
294 
295 struct kvm_vcpu {
296 	struct kvm *kvm;
297 #ifdef CONFIG_PREEMPT_NOTIFIERS
298 	struct preempt_notifier preempt_notifier;
299 #endif
300 	int cpu;
301 	int vcpu_id; /* id given by userspace at creation */
302 	int vcpu_idx; /* index in kvm->vcpus array */
303 	int srcu_idx;
304 	int mode;
305 	u64 requests;
306 	unsigned long guest_debug;
307 
308 	int pre_pcpu;
309 	struct list_head blocked_vcpu_list;
310 
311 	struct mutex mutex;
312 	struct kvm_run *run;
313 
314 	struct rcuwait wait;
315 	struct pid __rcu *pid;
316 	int sigset_active;
317 	sigset_t sigset;
318 	unsigned int halt_poll_ns;
319 	bool valid_wakeup;
320 
321 #ifdef CONFIG_HAS_IOMEM
322 	int mmio_needed;
323 	int mmio_read_completed;
324 	int mmio_is_write;
325 	int mmio_cur_fragment;
326 	int mmio_nr_fragments;
327 	struct kvm_mmio_fragment mmio_fragments[KVM_MAX_MMIO_FRAGMENTS];
328 #endif
329 
330 #ifdef CONFIG_KVM_ASYNC_PF
331 	struct {
332 		u32 queued;
333 		struct list_head queue;
334 		struct list_head done;
335 		spinlock_t lock;
336 	} async_pf;
337 #endif
338 
339 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
340 	/*
341 	 * Cpu relax intercept or pause loop exit optimization
342 	 * in_spin_loop: set when a vcpu does a pause loop exit
343 	 *  or cpu relax intercepted.
344 	 * dy_eligible: indicates whether vcpu is eligible for directed yield.
345 	 */
346 	struct {
347 		bool in_spin_loop;
348 		bool dy_eligible;
349 	} spin_loop;
350 #endif
351 	bool preempted;
352 	bool ready;
353 	struct kvm_vcpu_arch arch;
354 	struct kvm_vcpu_stat stat;
355 	char stats_id[KVM_STATS_NAME_SIZE];
356 	struct kvm_dirty_ring dirty_ring;
357 
358 	/*
359 	 * The index of the most recently used memslot by this vCPU. It's ok
360 	 * if this becomes stale due to memslot changes since we always check
361 	 * it is a valid slot.
362 	 */
363 	int last_used_slot;
364 };
365 
366 /* must be called with irqs disabled */
367 static __always_inline void guest_enter_irqoff(void)
368 {
369 	/*
370 	 * This is running in ioctl context so its safe to assume that it's the
371 	 * stime pending cputime to flush.
372 	 */
373 	instrumentation_begin();
374 	vtime_account_guest_enter();
375 	instrumentation_end();
376 
377 	/*
378 	 * KVM does not hold any references to rcu protected data when it
379 	 * switches CPU into a guest mode. In fact switching to a guest mode
380 	 * is very similar to exiting to userspace from rcu point of view. In
381 	 * addition CPU may stay in a guest mode for quite a long time (up to
382 	 * one time slice). Lets treat guest mode as quiescent state, just like
383 	 * we do with user-mode execution.
384 	 */
385 	if (!context_tracking_guest_enter()) {
386 		instrumentation_begin();
387 		rcu_virt_note_context_switch(smp_processor_id());
388 		instrumentation_end();
389 	}
390 }
391 
392 static __always_inline void guest_exit_irqoff(void)
393 {
394 	context_tracking_guest_exit();
395 
396 	instrumentation_begin();
397 	/* Flush the guest cputime we spent on the guest */
398 	vtime_account_guest_exit();
399 	instrumentation_end();
400 }
401 
402 static inline void guest_exit(void)
403 {
404 	unsigned long flags;
405 
406 	local_irq_save(flags);
407 	guest_exit_irqoff();
408 	local_irq_restore(flags);
409 }
410 
411 static inline int kvm_vcpu_exiting_guest_mode(struct kvm_vcpu *vcpu)
412 {
413 	/*
414 	 * The memory barrier ensures a previous write to vcpu->requests cannot
415 	 * be reordered with the read of vcpu->mode.  It pairs with the general
416 	 * memory barrier following the write of vcpu->mode in VCPU RUN.
417 	 */
418 	smp_mb__before_atomic();
419 	return cmpxchg(&vcpu->mode, IN_GUEST_MODE, EXITING_GUEST_MODE);
420 }
421 
422 /*
423  * Some of the bitops functions do not support too long bitmaps.
424  * This number must be determined not to exceed such limits.
425  */
426 #define KVM_MEM_MAX_NR_PAGES ((1UL << 31) - 1)
427 
428 struct kvm_memory_slot {
429 	gfn_t base_gfn;
430 	unsigned long npages;
431 	unsigned long *dirty_bitmap;
432 	struct kvm_arch_memory_slot arch;
433 	unsigned long userspace_addr;
434 	u32 flags;
435 	short id;
436 	u16 as_id;
437 };
438 
439 static inline bool kvm_slot_dirty_track_enabled(struct kvm_memory_slot *slot)
440 {
441 	return slot->flags & KVM_MEM_LOG_DIRTY_PAGES;
442 }
443 
444 static inline unsigned long kvm_dirty_bitmap_bytes(struct kvm_memory_slot *memslot)
445 {
446 	return ALIGN(memslot->npages, BITS_PER_LONG) / 8;
447 }
448 
449 static inline unsigned long *kvm_second_dirty_bitmap(struct kvm_memory_slot *memslot)
450 {
451 	unsigned long len = kvm_dirty_bitmap_bytes(memslot);
452 
453 	return memslot->dirty_bitmap + len / sizeof(*memslot->dirty_bitmap);
454 }
455 
456 #ifndef KVM_DIRTY_LOG_MANUAL_CAPS
457 #define KVM_DIRTY_LOG_MANUAL_CAPS KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE
458 #endif
459 
460 struct kvm_s390_adapter_int {
461 	u64 ind_addr;
462 	u64 summary_addr;
463 	u64 ind_offset;
464 	u32 summary_offset;
465 	u32 adapter_id;
466 };
467 
468 struct kvm_hv_sint {
469 	u32 vcpu;
470 	u32 sint;
471 };
472 
473 struct kvm_kernel_irq_routing_entry {
474 	u32 gsi;
475 	u32 type;
476 	int (*set)(struct kvm_kernel_irq_routing_entry *e,
477 		   struct kvm *kvm, int irq_source_id, int level,
478 		   bool line_status);
479 	union {
480 		struct {
481 			unsigned irqchip;
482 			unsigned pin;
483 		} irqchip;
484 		struct {
485 			u32 address_lo;
486 			u32 address_hi;
487 			u32 data;
488 			u32 flags;
489 			u32 devid;
490 		} msi;
491 		struct kvm_s390_adapter_int adapter;
492 		struct kvm_hv_sint hv_sint;
493 	};
494 	struct hlist_node link;
495 };
496 
497 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
498 struct kvm_irq_routing_table {
499 	int chip[KVM_NR_IRQCHIPS][KVM_IRQCHIP_NUM_PINS];
500 	u32 nr_rt_entries;
501 	/*
502 	 * Array indexed by gsi. Each entry contains list of irq chips
503 	 * the gsi is connected to.
504 	 */
505 	struct hlist_head map[];
506 };
507 #endif
508 
509 #ifndef KVM_PRIVATE_MEM_SLOTS
510 #define KVM_PRIVATE_MEM_SLOTS 0
511 #endif
512 
513 #define KVM_MEM_SLOTS_NUM SHRT_MAX
514 #define KVM_USER_MEM_SLOTS (KVM_MEM_SLOTS_NUM - KVM_PRIVATE_MEM_SLOTS)
515 
516 #ifndef __KVM_VCPU_MULTIPLE_ADDRESS_SPACE
517 static inline int kvm_arch_vcpu_memslots_id(struct kvm_vcpu *vcpu)
518 {
519 	return 0;
520 }
521 #endif
522 
523 /*
524  * Note:
525  * memslots are not sorted by id anymore, please use id_to_memslot()
526  * to get the memslot by its id.
527  */
528 struct kvm_memslots {
529 	u64 generation;
530 	/* The mapping table from slot id to the index in memslots[]. */
531 	short id_to_index[KVM_MEM_SLOTS_NUM];
532 	atomic_t last_used_slot;
533 	int used_slots;
534 	struct kvm_memory_slot memslots[];
535 };
536 
537 struct kvm {
538 #ifdef KVM_HAVE_MMU_RWLOCK
539 	rwlock_t mmu_lock;
540 #else
541 	spinlock_t mmu_lock;
542 #endif /* KVM_HAVE_MMU_RWLOCK */
543 
544 	struct mutex slots_lock;
545 
546 	/*
547 	 * Protects the arch-specific fields of struct kvm_memory_slots in
548 	 * use by the VM. To be used under the slots_lock (above) or in a
549 	 * kvm->srcu critical section where acquiring the slots_lock would
550 	 * lead to deadlock with the synchronize_srcu in
551 	 * install_new_memslots.
552 	 */
553 	struct mutex slots_arch_lock;
554 	struct mm_struct *mm; /* userspace tied to this vm */
555 	struct kvm_memslots __rcu *memslots[KVM_ADDRESS_SPACE_NUM];
556 	struct kvm_vcpu *vcpus[KVM_MAX_VCPUS];
557 
558 	/* Used to wait for completion of MMU notifiers.  */
559 	spinlock_t mn_invalidate_lock;
560 	unsigned long mn_active_invalidate_count;
561 	struct rcuwait mn_memslots_update_rcuwait;
562 
563 	/*
564 	 * created_vcpus is protected by kvm->lock, and is incremented
565 	 * at the beginning of KVM_CREATE_VCPU.  online_vcpus is only
566 	 * incremented after storing the kvm_vcpu pointer in vcpus,
567 	 * and is accessed atomically.
568 	 */
569 	atomic_t online_vcpus;
570 	int created_vcpus;
571 	int last_boosted_vcpu;
572 	struct list_head vm_list;
573 	struct mutex lock;
574 	struct kvm_io_bus __rcu *buses[KVM_NR_BUSES];
575 #ifdef CONFIG_HAVE_KVM_EVENTFD
576 	struct {
577 		spinlock_t        lock;
578 		struct list_head  items;
579 		struct list_head  resampler_list;
580 		struct mutex      resampler_lock;
581 	} irqfds;
582 	struct list_head ioeventfds;
583 #endif
584 	struct kvm_vm_stat stat;
585 	struct kvm_arch arch;
586 	refcount_t users_count;
587 #ifdef CONFIG_KVM_MMIO
588 	struct kvm_coalesced_mmio_ring *coalesced_mmio_ring;
589 	spinlock_t ring_lock;
590 	struct list_head coalesced_zones;
591 #endif
592 
593 	struct mutex irq_lock;
594 #ifdef CONFIG_HAVE_KVM_IRQCHIP
595 	/*
596 	 * Update side is protected by irq_lock.
597 	 */
598 	struct kvm_irq_routing_table __rcu *irq_routing;
599 #endif
600 #ifdef CONFIG_HAVE_KVM_IRQFD
601 	struct hlist_head irq_ack_notifier_list;
602 #endif
603 
604 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
605 	struct mmu_notifier mmu_notifier;
606 	unsigned long mmu_notifier_seq;
607 	long mmu_notifier_count;
608 	unsigned long mmu_notifier_range_start;
609 	unsigned long mmu_notifier_range_end;
610 #endif
611 	long tlbs_dirty;
612 	struct list_head devices;
613 	u64 manual_dirty_log_protect;
614 	struct dentry *debugfs_dentry;
615 	struct kvm_stat_data **debugfs_stat_data;
616 	struct srcu_struct srcu;
617 	struct srcu_struct irq_srcu;
618 	pid_t userspace_pid;
619 	unsigned int max_halt_poll_ns;
620 	u32 dirty_ring_size;
621 	bool vm_bugged;
622 
623 #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER
624 	struct notifier_block pm_notifier;
625 #endif
626 	char stats_id[KVM_STATS_NAME_SIZE];
627 };
628 
629 #define kvm_err(fmt, ...) \
630 	pr_err("kvm [%i]: " fmt, task_pid_nr(current), ## __VA_ARGS__)
631 #define kvm_info(fmt, ...) \
632 	pr_info("kvm [%i]: " fmt, task_pid_nr(current), ## __VA_ARGS__)
633 #define kvm_debug(fmt, ...) \
634 	pr_debug("kvm [%i]: " fmt, task_pid_nr(current), ## __VA_ARGS__)
635 #define kvm_debug_ratelimited(fmt, ...) \
636 	pr_debug_ratelimited("kvm [%i]: " fmt, task_pid_nr(current), \
637 			     ## __VA_ARGS__)
638 #define kvm_pr_unimpl(fmt, ...) \
639 	pr_err_ratelimited("kvm [%i]: " fmt, \
640 			   task_tgid_nr(current), ## __VA_ARGS__)
641 
642 /* The guest did something we don't support. */
643 #define vcpu_unimpl(vcpu, fmt, ...)					\
644 	kvm_pr_unimpl("vcpu%i, guest rIP: 0x%lx " fmt,			\
645 			(vcpu)->vcpu_id, kvm_rip_read(vcpu), ## __VA_ARGS__)
646 
647 #define vcpu_debug(vcpu, fmt, ...)					\
648 	kvm_debug("vcpu%i " fmt, (vcpu)->vcpu_id, ## __VA_ARGS__)
649 #define vcpu_debug_ratelimited(vcpu, fmt, ...)				\
650 	kvm_debug_ratelimited("vcpu%i " fmt, (vcpu)->vcpu_id,           \
651 			      ## __VA_ARGS__)
652 #define vcpu_err(vcpu, fmt, ...)					\
653 	kvm_err("vcpu%i " fmt, (vcpu)->vcpu_id, ## __VA_ARGS__)
654 
655 static inline void kvm_vm_bugged(struct kvm *kvm)
656 {
657 	kvm->vm_bugged = true;
658 	kvm_make_all_cpus_request(kvm, KVM_REQ_VM_BUGGED);
659 }
660 
661 #define KVM_BUG(cond, kvm, fmt...)				\
662 ({								\
663 	int __ret = (cond);					\
664 								\
665 	if (WARN_ONCE(__ret && !(kvm)->vm_bugged, fmt))		\
666 		kvm_vm_bugged(kvm);				\
667 	unlikely(__ret);					\
668 })
669 
670 #define KVM_BUG_ON(cond, kvm)					\
671 ({								\
672 	int __ret = (cond);					\
673 								\
674 	if (WARN_ON_ONCE(__ret && !(kvm)->vm_bugged))		\
675 		kvm_vm_bugged(kvm);				\
676 	unlikely(__ret);					\
677 })
678 
679 static inline bool kvm_dirty_log_manual_protect_and_init_set(struct kvm *kvm)
680 {
681 	return !!(kvm->manual_dirty_log_protect & KVM_DIRTY_LOG_INITIALLY_SET);
682 }
683 
684 static inline struct kvm_io_bus *kvm_get_bus(struct kvm *kvm, enum kvm_bus idx)
685 {
686 	return srcu_dereference_check(kvm->buses[idx], &kvm->srcu,
687 				      lockdep_is_held(&kvm->slots_lock) ||
688 				      !refcount_read(&kvm->users_count));
689 }
690 
691 static inline struct kvm_vcpu *kvm_get_vcpu(struct kvm *kvm, int i)
692 {
693 	int num_vcpus = atomic_read(&kvm->online_vcpus);
694 	i = array_index_nospec(i, num_vcpus);
695 
696 	/* Pairs with smp_wmb() in kvm_vm_ioctl_create_vcpu.  */
697 	smp_rmb();
698 	return kvm->vcpus[i];
699 }
700 
701 #define kvm_for_each_vcpu(idx, vcpup, kvm) \
702 	for (idx = 0; \
703 	     idx < atomic_read(&kvm->online_vcpus) && \
704 	     (vcpup = kvm_get_vcpu(kvm, idx)) != NULL; \
705 	     idx++)
706 
707 static inline struct kvm_vcpu *kvm_get_vcpu_by_id(struct kvm *kvm, int id)
708 {
709 	struct kvm_vcpu *vcpu = NULL;
710 	int i;
711 
712 	if (id < 0)
713 		return NULL;
714 	if (id < KVM_MAX_VCPUS)
715 		vcpu = kvm_get_vcpu(kvm, id);
716 	if (vcpu && vcpu->vcpu_id == id)
717 		return vcpu;
718 	kvm_for_each_vcpu(i, vcpu, kvm)
719 		if (vcpu->vcpu_id == id)
720 			return vcpu;
721 	return NULL;
722 }
723 
724 static inline int kvm_vcpu_get_idx(struct kvm_vcpu *vcpu)
725 {
726 	return vcpu->vcpu_idx;
727 }
728 
729 #define kvm_for_each_memslot(memslot, slots)				\
730 	for (memslot = &slots->memslots[0];				\
731 	     memslot < slots->memslots + slots->used_slots; memslot++)	\
732 		if (WARN_ON_ONCE(!memslot->npages)) {			\
733 		} else
734 
735 void kvm_vcpu_destroy(struct kvm_vcpu *vcpu);
736 
737 void vcpu_load(struct kvm_vcpu *vcpu);
738 void vcpu_put(struct kvm_vcpu *vcpu);
739 
740 #ifdef __KVM_HAVE_IOAPIC
741 void kvm_arch_post_irq_ack_notifier_list_update(struct kvm *kvm);
742 void kvm_arch_post_irq_routing_update(struct kvm *kvm);
743 #else
744 static inline void kvm_arch_post_irq_ack_notifier_list_update(struct kvm *kvm)
745 {
746 }
747 static inline void kvm_arch_post_irq_routing_update(struct kvm *kvm)
748 {
749 }
750 #endif
751 
752 #ifdef CONFIG_HAVE_KVM_IRQFD
753 int kvm_irqfd_init(void);
754 void kvm_irqfd_exit(void);
755 #else
756 static inline int kvm_irqfd_init(void)
757 {
758 	return 0;
759 }
760 
761 static inline void kvm_irqfd_exit(void)
762 {
763 }
764 #endif
765 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
766 		  struct module *module);
767 void kvm_exit(void);
768 
769 void kvm_get_kvm(struct kvm *kvm);
770 bool kvm_get_kvm_safe(struct kvm *kvm);
771 void kvm_put_kvm(struct kvm *kvm);
772 bool file_is_kvm(struct file *file);
773 void kvm_put_kvm_no_destroy(struct kvm *kvm);
774 
775 static inline struct kvm_memslots *__kvm_memslots(struct kvm *kvm, int as_id)
776 {
777 	as_id = array_index_nospec(as_id, KVM_ADDRESS_SPACE_NUM);
778 	return srcu_dereference_check(kvm->memslots[as_id], &kvm->srcu,
779 			lockdep_is_held(&kvm->slots_lock) ||
780 			!refcount_read(&kvm->users_count));
781 }
782 
783 static inline struct kvm_memslots *kvm_memslots(struct kvm *kvm)
784 {
785 	return __kvm_memslots(kvm, 0);
786 }
787 
788 static inline struct kvm_memslots *kvm_vcpu_memslots(struct kvm_vcpu *vcpu)
789 {
790 	int as_id = kvm_arch_vcpu_memslots_id(vcpu);
791 
792 	return __kvm_memslots(vcpu->kvm, as_id);
793 }
794 
795 static inline
796 struct kvm_memory_slot *id_to_memslot(struct kvm_memslots *slots, int id)
797 {
798 	int index = slots->id_to_index[id];
799 	struct kvm_memory_slot *slot;
800 
801 	if (index < 0)
802 		return NULL;
803 
804 	slot = &slots->memslots[index];
805 
806 	WARN_ON(slot->id != id);
807 	return slot;
808 }
809 
810 /*
811  * KVM_SET_USER_MEMORY_REGION ioctl allows the following operations:
812  * - create a new memory slot
813  * - delete an existing memory slot
814  * - modify an existing memory slot
815  *   -- move it in the guest physical memory space
816  *   -- just change its flags
817  *
818  * Since flags can be changed by some of these operations, the following
819  * differentiation is the best we can do for __kvm_set_memory_region():
820  */
821 enum kvm_mr_change {
822 	KVM_MR_CREATE,
823 	KVM_MR_DELETE,
824 	KVM_MR_MOVE,
825 	KVM_MR_FLAGS_ONLY,
826 };
827 
828 int kvm_set_memory_region(struct kvm *kvm,
829 			  const struct kvm_userspace_memory_region *mem);
830 int __kvm_set_memory_region(struct kvm *kvm,
831 			    const struct kvm_userspace_memory_region *mem);
832 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot);
833 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen);
834 int kvm_arch_prepare_memory_region(struct kvm *kvm,
835 				struct kvm_memory_slot *memslot,
836 				const struct kvm_userspace_memory_region *mem,
837 				enum kvm_mr_change change);
838 void kvm_arch_commit_memory_region(struct kvm *kvm,
839 				const struct kvm_userspace_memory_region *mem,
840 				struct kvm_memory_slot *old,
841 				const struct kvm_memory_slot *new,
842 				enum kvm_mr_change change);
843 /* flush all memory translations */
844 void kvm_arch_flush_shadow_all(struct kvm *kvm);
845 /* flush memory translations pointing to 'slot' */
846 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
847 				   struct kvm_memory_slot *slot);
848 
849 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
850 			    struct page **pages, int nr_pages);
851 
852 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn);
853 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn);
854 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable);
855 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot, gfn_t gfn);
856 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot, gfn_t gfn,
857 				      bool *writable);
858 void kvm_release_page_clean(struct page *page);
859 void kvm_release_page_dirty(struct page *page);
860 void kvm_set_page_accessed(struct page *page);
861 
862 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn);
863 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
864 		      bool *writable);
865 kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn);
866 kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn);
867 kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
868 			       bool atomic, bool *async, bool write_fault,
869 			       bool *writable, hva_t *hva);
870 
871 void kvm_release_pfn_clean(kvm_pfn_t pfn);
872 void kvm_release_pfn_dirty(kvm_pfn_t pfn);
873 void kvm_set_pfn_dirty(kvm_pfn_t pfn);
874 void kvm_set_pfn_accessed(kvm_pfn_t pfn);
875 
876 void kvm_release_pfn(kvm_pfn_t pfn, bool dirty, struct gfn_to_pfn_cache *cache);
877 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
878 			int len);
879 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len);
880 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
881 			   void *data, unsigned long len);
882 int kvm_read_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
883 				 void *data, unsigned int offset,
884 				 unsigned long len);
885 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
886 			 int offset, int len);
887 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
888 		    unsigned long len);
889 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
890 			   void *data, unsigned long len);
891 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
892 				  void *data, unsigned int offset,
893 				  unsigned long len);
894 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
895 			      gpa_t gpa, unsigned long len);
896 
897 #define __kvm_get_guest(kvm, gfn, offset, v)				\
898 ({									\
899 	unsigned long __addr = gfn_to_hva(kvm, gfn);			\
900 	typeof(v) __user *__uaddr = (typeof(__uaddr))(__addr + offset);	\
901 	int __ret = -EFAULT;						\
902 									\
903 	if (!kvm_is_error_hva(__addr))					\
904 		__ret = get_user(v, __uaddr);				\
905 	__ret;								\
906 })
907 
908 #define kvm_get_guest(kvm, gpa, v)					\
909 ({									\
910 	gpa_t __gpa = gpa;						\
911 	struct kvm *__kvm = kvm;					\
912 									\
913 	__kvm_get_guest(__kvm, __gpa >> PAGE_SHIFT,			\
914 			offset_in_page(__gpa), v);			\
915 })
916 
917 #define __kvm_put_guest(kvm, gfn, offset, v)				\
918 ({									\
919 	unsigned long __addr = gfn_to_hva(kvm, gfn);			\
920 	typeof(v) __user *__uaddr = (typeof(__uaddr))(__addr + offset);	\
921 	int __ret = -EFAULT;						\
922 									\
923 	if (!kvm_is_error_hva(__addr))					\
924 		__ret = put_user(v, __uaddr);				\
925 	if (!__ret)							\
926 		mark_page_dirty(kvm, gfn);				\
927 	__ret;								\
928 })
929 
930 #define kvm_put_guest(kvm, gpa, v)					\
931 ({									\
932 	gpa_t __gpa = gpa;						\
933 	struct kvm *__kvm = kvm;					\
934 									\
935 	__kvm_put_guest(__kvm, __gpa >> PAGE_SHIFT,			\
936 			offset_in_page(__gpa), v);			\
937 })
938 
939 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len);
940 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn);
941 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn);
942 bool kvm_vcpu_is_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn);
943 unsigned long kvm_host_page_size(struct kvm_vcpu *vcpu, gfn_t gfn);
944 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot, gfn_t gfn);
945 void mark_page_dirty(struct kvm *kvm, gfn_t gfn);
946 
947 struct kvm_memslots *kvm_vcpu_memslots(struct kvm_vcpu *vcpu);
948 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn);
949 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn);
950 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn);
951 int kvm_vcpu_map(struct kvm_vcpu *vcpu, gpa_t gpa, struct kvm_host_map *map);
952 int kvm_map_gfn(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map,
953 		struct gfn_to_pfn_cache *cache, bool atomic);
954 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn);
955 void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map, bool dirty);
956 int kvm_unmap_gfn(struct kvm_vcpu *vcpu, struct kvm_host_map *map,
957 		  struct gfn_to_pfn_cache *cache, bool dirty, bool atomic);
958 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn);
959 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable);
960 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data, int offset,
961 			     int len);
962 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa, void *data,
963 			       unsigned long len);
964 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data,
965 			unsigned long len);
966 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, const void *data,
967 			      int offset, int len);
968 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
969 			 unsigned long len);
970 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn);
971 
972 void kvm_sigset_activate(struct kvm_vcpu *vcpu);
973 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu);
974 
975 void kvm_vcpu_block(struct kvm_vcpu *vcpu);
976 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu);
977 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu);
978 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu);
979 void kvm_vcpu_kick(struct kvm_vcpu *vcpu);
980 int kvm_vcpu_yield_to(struct kvm_vcpu *target);
981 void kvm_vcpu_on_spin(struct kvm_vcpu *vcpu, bool usermode_vcpu_not_eligible);
982 
983 void kvm_flush_remote_tlbs(struct kvm *kvm);
984 void kvm_reload_remote_mmus(struct kvm *kvm);
985 
986 #ifdef KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE
987 int kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache *mc, int min);
988 int kvm_mmu_memory_cache_nr_free_objects(struct kvm_mmu_memory_cache *mc);
989 void kvm_mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc);
990 void *kvm_mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc);
991 #endif
992 
993 void kvm_inc_notifier_count(struct kvm *kvm, unsigned long start,
994 				   unsigned long end);
995 void kvm_dec_notifier_count(struct kvm *kvm, unsigned long start,
996 				   unsigned long end);
997 
998 long kvm_arch_dev_ioctl(struct file *filp,
999 			unsigned int ioctl, unsigned long arg);
1000 long kvm_arch_vcpu_ioctl(struct file *filp,
1001 			 unsigned int ioctl, unsigned long arg);
1002 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf);
1003 
1004 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext);
1005 
1006 void kvm_arch_mmu_enable_log_dirty_pt_masked(struct kvm *kvm,
1007 					struct kvm_memory_slot *slot,
1008 					gfn_t gfn_offset,
1009 					unsigned long mask);
1010 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot);
1011 
1012 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1013 void kvm_arch_flush_remote_tlbs_memslot(struct kvm *kvm,
1014 					const struct kvm_memory_slot *memslot);
1015 #else /* !CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
1016 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log);
1017 int kvm_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log,
1018 		      int *is_dirty, struct kvm_memory_slot **memslot);
1019 #endif
1020 
1021 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
1022 			bool line_status);
1023 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
1024 			    struct kvm_enable_cap *cap);
1025 long kvm_arch_vm_ioctl(struct file *filp,
1026 		       unsigned int ioctl, unsigned long arg);
1027 
1028 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu);
1029 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu);
1030 
1031 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
1032 				    struct kvm_translation *tr);
1033 
1034 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs);
1035 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs);
1036 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
1037 				  struct kvm_sregs *sregs);
1038 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
1039 				  struct kvm_sregs *sregs);
1040 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
1041 				    struct kvm_mp_state *mp_state);
1042 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
1043 				    struct kvm_mp_state *mp_state);
1044 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
1045 					struct kvm_guest_debug *dbg);
1046 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu);
1047 
1048 int kvm_arch_init(void *opaque);
1049 void kvm_arch_exit(void);
1050 
1051 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu);
1052 
1053 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu);
1054 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu);
1055 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id);
1056 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu);
1057 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu);
1058 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu);
1059 
1060 #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER
1061 int kvm_arch_pm_notifier(struct kvm *kvm, unsigned long state);
1062 #endif
1063 
1064 #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS
1065 void kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu, struct dentry *debugfs_dentry);
1066 #endif
1067 
1068 int kvm_arch_hardware_enable(void);
1069 void kvm_arch_hardware_disable(void);
1070 int kvm_arch_hardware_setup(void *opaque);
1071 void kvm_arch_hardware_unsetup(void);
1072 int kvm_arch_check_processor_compat(void *opaque);
1073 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu);
1074 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu);
1075 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu);
1076 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu);
1077 bool kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu);
1078 int kvm_arch_post_init_vm(struct kvm *kvm);
1079 void kvm_arch_pre_destroy_vm(struct kvm *kvm);
1080 int kvm_arch_create_vm_debugfs(struct kvm *kvm);
1081 
1082 #ifndef __KVM_HAVE_ARCH_VM_ALLOC
1083 /*
1084  * All architectures that want to use vzalloc currently also
1085  * need their own kvm_arch_alloc_vm implementation.
1086  */
1087 static inline struct kvm *kvm_arch_alloc_vm(void)
1088 {
1089 	return kzalloc(sizeof(struct kvm), GFP_KERNEL);
1090 }
1091 
1092 static inline void kvm_arch_free_vm(struct kvm *kvm)
1093 {
1094 	kfree(kvm);
1095 }
1096 #endif
1097 
1098 #ifndef __KVM_HAVE_ARCH_FLUSH_REMOTE_TLB
1099 static inline int kvm_arch_flush_remote_tlb(struct kvm *kvm)
1100 {
1101 	return -ENOTSUPP;
1102 }
1103 #endif
1104 
1105 #ifdef __KVM_HAVE_ARCH_NONCOHERENT_DMA
1106 void kvm_arch_register_noncoherent_dma(struct kvm *kvm);
1107 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm);
1108 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm);
1109 #else
1110 static inline void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
1111 {
1112 }
1113 
1114 static inline void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
1115 {
1116 }
1117 
1118 static inline bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
1119 {
1120 	return false;
1121 }
1122 #endif
1123 #ifdef __KVM_HAVE_ARCH_ASSIGNED_DEVICE
1124 void kvm_arch_start_assignment(struct kvm *kvm);
1125 void kvm_arch_end_assignment(struct kvm *kvm);
1126 bool kvm_arch_has_assigned_device(struct kvm *kvm);
1127 #else
1128 static inline void kvm_arch_start_assignment(struct kvm *kvm)
1129 {
1130 }
1131 
1132 static inline void kvm_arch_end_assignment(struct kvm *kvm)
1133 {
1134 }
1135 
1136 static inline bool kvm_arch_has_assigned_device(struct kvm *kvm)
1137 {
1138 	return false;
1139 }
1140 #endif
1141 
1142 static inline struct rcuwait *kvm_arch_vcpu_get_wait(struct kvm_vcpu *vcpu)
1143 {
1144 #ifdef __KVM_HAVE_ARCH_WQP
1145 	return vcpu->arch.waitp;
1146 #else
1147 	return &vcpu->wait;
1148 #endif
1149 }
1150 
1151 #ifdef __KVM_HAVE_ARCH_INTC_INITIALIZED
1152 /*
1153  * returns true if the virtual interrupt controller is initialized and
1154  * ready to accept virtual IRQ. On some architectures the virtual interrupt
1155  * controller is dynamically instantiated and this is not always true.
1156  */
1157 bool kvm_arch_intc_initialized(struct kvm *kvm);
1158 #else
1159 static inline bool kvm_arch_intc_initialized(struct kvm *kvm)
1160 {
1161 	return true;
1162 }
1163 #endif
1164 
1165 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type);
1166 void kvm_arch_destroy_vm(struct kvm *kvm);
1167 void kvm_arch_sync_events(struct kvm *kvm);
1168 
1169 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu);
1170 
1171 bool kvm_is_reserved_pfn(kvm_pfn_t pfn);
1172 bool kvm_is_zone_device_pfn(kvm_pfn_t pfn);
1173 bool kvm_is_transparent_hugepage(kvm_pfn_t pfn);
1174 
1175 struct kvm_irq_ack_notifier {
1176 	struct hlist_node link;
1177 	unsigned gsi;
1178 	void (*irq_acked)(struct kvm_irq_ack_notifier *kian);
1179 };
1180 
1181 int kvm_irq_map_gsi(struct kvm *kvm,
1182 		    struct kvm_kernel_irq_routing_entry *entries, int gsi);
1183 int kvm_irq_map_chip_pin(struct kvm *kvm, unsigned irqchip, unsigned pin);
1184 
1185 int kvm_set_irq(struct kvm *kvm, int irq_source_id, u32 irq, int level,
1186 		bool line_status);
1187 int kvm_set_msi(struct kvm_kernel_irq_routing_entry *irq_entry, struct kvm *kvm,
1188 		int irq_source_id, int level, bool line_status);
1189 int kvm_arch_set_irq_inatomic(struct kvm_kernel_irq_routing_entry *e,
1190 			       struct kvm *kvm, int irq_source_id,
1191 			       int level, bool line_status);
1192 bool kvm_irq_has_notifier(struct kvm *kvm, unsigned irqchip, unsigned pin);
1193 void kvm_notify_acked_gsi(struct kvm *kvm, int gsi);
1194 void kvm_notify_acked_irq(struct kvm *kvm, unsigned irqchip, unsigned pin);
1195 void kvm_register_irq_ack_notifier(struct kvm *kvm,
1196 				   struct kvm_irq_ack_notifier *kian);
1197 void kvm_unregister_irq_ack_notifier(struct kvm *kvm,
1198 				   struct kvm_irq_ack_notifier *kian);
1199 int kvm_request_irq_source_id(struct kvm *kvm);
1200 void kvm_free_irq_source_id(struct kvm *kvm, int irq_source_id);
1201 bool kvm_arch_irqfd_allowed(struct kvm *kvm, struct kvm_irqfd *args);
1202 
1203 /*
1204  * Returns a pointer to the memslot at slot_index if it contains gfn.
1205  * Otherwise returns NULL.
1206  */
1207 static inline struct kvm_memory_slot *
1208 try_get_memslot(struct kvm_memslots *slots, int slot_index, gfn_t gfn)
1209 {
1210 	struct kvm_memory_slot *slot;
1211 
1212 	if (slot_index < 0 || slot_index >= slots->used_slots)
1213 		return NULL;
1214 
1215 	/*
1216 	 * slot_index can come from vcpu->last_used_slot which is not kept
1217 	 * in sync with userspace-controllable memslot deletion. So use nospec
1218 	 * to prevent the CPU from speculating past the end of memslots[].
1219 	 */
1220 	slot_index = array_index_nospec(slot_index, slots->used_slots);
1221 	slot = &slots->memslots[slot_index];
1222 
1223 	if (gfn >= slot->base_gfn && gfn < slot->base_gfn + slot->npages)
1224 		return slot;
1225 	else
1226 		return NULL;
1227 }
1228 
1229 /*
1230  * Returns a pointer to the memslot that contains gfn and records the index of
1231  * the slot in index. Otherwise returns NULL.
1232  *
1233  * IMPORTANT: Slots are sorted from highest GFN to lowest GFN!
1234  */
1235 static inline struct kvm_memory_slot *
1236 search_memslots(struct kvm_memslots *slots, gfn_t gfn, int *index)
1237 {
1238 	int start = 0, end = slots->used_slots;
1239 	struct kvm_memory_slot *memslots = slots->memslots;
1240 	struct kvm_memory_slot *slot;
1241 
1242 	if (unlikely(!slots->used_slots))
1243 		return NULL;
1244 
1245 	while (start < end) {
1246 		int slot = start + (end - start) / 2;
1247 
1248 		if (gfn >= memslots[slot].base_gfn)
1249 			end = slot;
1250 		else
1251 			start = slot + 1;
1252 	}
1253 
1254 	slot = try_get_memslot(slots, start, gfn);
1255 	if (slot) {
1256 		*index = start;
1257 		return slot;
1258 	}
1259 
1260 	return NULL;
1261 }
1262 
1263 /*
1264  * __gfn_to_memslot() and its descendants are here because it is called from
1265  * non-modular code in arch/powerpc/kvm/book3s_64_vio{,_hv}.c. gfn_to_memslot()
1266  * itself isn't here as an inline because that would bloat other code too much.
1267  */
1268 static inline struct kvm_memory_slot *
1269 __gfn_to_memslot(struct kvm_memslots *slots, gfn_t gfn)
1270 {
1271 	struct kvm_memory_slot *slot;
1272 	int slot_index = atomic_read(&slots->last_used_slot);
1273 
1274 	slot = try_get_memslot(slots, slot_index, gfn);
1275 	if (slot)
1276 		return slot;
1277 
1278 	slot = search_memslots(slots, gfn, &slot_index);
1279 	if (slot) {
1280 		atomic_set(&slots->last_used_slot, slot_index);
1281 		return slot;
1282 	}
1283 
1284 	return NULL;
1285 }
1286 
1287 static inline unsigned long
1288 __gfn_to_hva_memslot(const struct kvm_memory_slot *slot, gfn_t gfn)
1289 {
1290 	/*
1291 	 * The index was checked originally in search_memslots.  To avoid
1292 	 * that a malicious guest builds a Spectre gadget out of e.g. page
1293 	 * table walks, do not let the processor speculate loads outside
1294 	 * the guest's registered memslots.
1295 	 */
1296 	unsigned long offset = gfn - slot->base_gfn;
1297 	offset = array_index_nospec(offset, slot->npages);
1298 	return slot->userspace_addr + offset * PAGE_SIZE;
1299 }
1300 
1301 static inline int memslot_id(struct kvm *kvm, gfn_t gfn)
1302 {
1303 	return gfn_to_memslot(kvm, gfn)->id;
1304 }
1305 
1306 static inline gfn_t
1307 hva_to_gfn_memslot(unsigned long hva, struct kvm_memory_slot *slot)
1308 {
1309 	gfn_t gfn_offset = (hva - slot->userspace_addr) >> PAGE_SHIFT;
1310 
1311 	return slot->base_gfn + gfn_offset;
1312 }
1313 
1314 static inline gpa_t gfn_to_gpa(gfn_t gfn)
1315 {
1316 	return (gpa_t)gfn << PAGE_SHIFT;
1317 }
1318 
1319 static inline gfn_t gpa_to_gfn(gpa_t gpa)
1320 {
1321 	return (gfn_t)(gpa >> PAGE_SHIFT);
1322 }
1323 
1324 static inline hpa_t pfn_to_hpa(kvm_pfn_t pfn)
1325 {
1326 	return (hpa_t)pfn << PAGE_SHIFT;
1327 }
1328 
1329 static inline struct page *kvm_vcpu_gpa_to_page(struct kvm_vcpu *vcpu,
1330 						gpa_t gpa)
1331 {
1332 	return kvm_vcpu_gfn_to_page(vcpu, gpa_to_gfn(gpa));
1333 }
1334 
1335 static inline bool kvm_is_error_gpa(struct kvm *kvm, gpa_t gpa)
1336 {
1337 	unsigned long hva = gfn_to_hva(kvm, gpa_to_gfn(gpa));
1338 
1339 	return kvm_is_error_hva(hva);
1340 }
1341 
1342 enum kvm_stat_kind {
1343 	KVM_STAT_VM,
1344 	KVM_STAT_VCPU,
1345 };
1346 
1347 struct kvm_stat_data {
1348 	struct kvm *kvm;
1349 	const struct _kvm_stats_desc *desc;
1350 	enum kvm_stat_kind kind;
1351 };
1352 
1353 struct _kvm_stats_desc {
1354 	struct kvm_stats_desc desc;
1355 	char name[KVM_STATS_NAME_SIZE];
1356 };
1357 
1358 #define STATS_DESC_COMMON(type, unit, base, exp, sz, bsz)		       \
1359 	.flags = type | unit | base |					       \
1360 		 BUILD_BUG_ON_ZERO(type & ~KVM_STATS_TYPE_MASK) |	       \
1361 		 BUILD_BUG_ON_ZERO(unit & ~KVM_STATS_UNIT_MASK) |	       \
1362 		 BUILD_BUG_ON_ZERO(base & ~KVM_STATS_BASE_MASK),	       \
1363 	.exponent = exp,						       \
1364 	.size = sz,							       \
1365 	.bucket_size = bsz
1366 
1367 #define VM_GENERIC_STATS_DESC(stat, type, unit, base, exp, sz, bsz)	       \
1368 	{								       \
1369 		{							       \
1370 			STATS_DESC_COMMON(type, unit, base, exp, sz, bsz),     \
1371 			.offset = offsetof(struct kvm_vm_stat, generic.stat)   \
1372 		},							       \
1373 		.name = #stat,						       \
1374 	}
1375 #define VCPU_GENERIC_STATS_DESC(stat, type, unit, base, exp, sz, bsz)	       \
1376 	{								       \
1377 		{							       \
1378 			STATS_DESC_COMMON(type, unit, base, exp, sz, bsz),     \
1379 			.offset = offsetof(struct kvm_vcpu_stat, generic.stat) \
1380 		},							       \
1381 		.name = #stat,						       \
1382 	}
1383 #define VM_STATS_DESC(stat, type, unit, base, exp, sz, bsz)		       \
1384 	{								       \
1385 		{							       \
1386 			STATS_DESC_COMMON(type, unit, base, exp, sz, bsz),     \
1387 			.offset = offsetof(struct kvm_vm_stat, stat)	       \
1388 		},							       \
1389 		.name = #stat,						       \
1390 	}
1391 #define VCPU_STATS_DESC(stat, type, unit, base, exp, sz, bsz)		       \
1392 	{								       \
1393 		{							       \
1394 			STATS_DESC_COMMON(type, unit, base, exp, sz, bsz),     \
1395 			.offset = offsetof(struct kvm_vcpu_stat, stat)	       \
1396 		},							       \
1397 		.name = #stat,						       \
1398 	}
1399 /* SCOPE: VM, VM_GENERIC, VCPU, VCPU_GENERIC */
1400 #define STATS_DESC(SCOPE, stat, type, unit, base, exp, sz, bsz)		       \
1401 	SCOPE##_STATS_DESC(stat, type, unit, base, exp, sz, bsz)
1402 
1403 #define STATS_DESC_CUMULATIVE(SCOPE, name, unit, base, exponent)	       \
1404 	STATS_DESC(SCOPE, name, KVM_STATS_TYPE_CUMULATIVE,		       \
1405 		unit, base, exponent, 1, 0)
1406 #define STATS_DESC_INSTANT(SCOPE, name, unit, base, exponent)		       \
1407 	STATS_DESC(SCOPE, name, KVM_STATS_TYPE_INSTANT,			       \
1408 		unit, base, exponent, 1, 0)
1409 #define STATS_DESC_PEAK(SCOPE, name, unit, base, exponent)		       \
1410 	STATS_DESC(SCOPE, name, KVM_STATS_TYPE_PEAK,			       \
1411 		unit, base, exponent, 1, 0)
1412 #define STATS_DESC_LINEAR_HIST(SCOPE, name, unit, base, exponent, sz, bsz)     \
1413 	STATS_DESC(SCOPE, name, KVM_STATS_TYPE_LINEAR_HIST,		       \
1414 		unit, base, exponent, sz, bsz)
1415 #define STATS_DESC_LOG_HIST(SCOPE, name, unit, base, exponent, sz)	       \
1416 	STATS_DESC(SCOPE, name, KVM_STATS_TYPE_LOG_HIST,		       \
1417 		unit, base, exponent, sz, 0)
1418 
1419 /* Cumulative counter, read/write */
1420 #define STATS_DESC_COUNTER(SCOPE, name)					       \
1421 	STATS_DESC_CUMULATIVE(SCOPE, name, KVM_STATS_UNIT_NONE,		       \
1422 		KVM_STATS_BASE_POW10, 0)
1423 /* Instantaneous counter, read only */
1424 #define STATS_DESC_ICOUNTER(SCOPE, name)				       \
1425 	STATS_DESC_INSTANT(SCOPE, name, KVM_STATS_UNIT_NONE,		       \
1426 		KVM_STATS_BASE_POW10, 0)
1427 /* Peak counter, read/write */
1428 #define STATS_DESC_PCOUNTER(SCOPE, name)				       \
1429 	STATS_DESC_PEAK(SCOPE, name, KVM_STATS_UNIT_NONE,		       \
1430 		KVM_STATS_BASE_POW10, 0)
1431 
1432 /* Cumulative time in nanosecond */
1433 #define STATS_DESC_TIME_NSEC(SCOPE, name)				       \
1434 	STATS_DESC_CUMULATIVE(SCOPE, name, KVM_STATS_UNIT_SECONDS,	       \
1435 		KVM_STATS_BASE_POW10, -9)
1436 /* Linear histogram for time in nanosecond */
1437 #define STATS_DESC_LINHIST_TIME_NSEC(SCOPE, name, sz, bsz)		       \
1438 	STATS_DESC_LINEAR_HIST(SCOPE, name, KVM_STATS_UNIT_SECONDS,	       \
1439 		KVM_STATS_BASE_POW10, -9, sz, bsz)
1440 /* Logarithmic histogram for time in nanosecond */
1441 #define STATS_DESC_LOGHIST_TIME_NSEC(SCOPE, name, sz)			       \
1442 	STATS_DESC_LOG_HIST(SCOPE, name, KVM_STATS_UNIT_SECONDS,	       \
1443 		KVM_STATS_BASE_POW10, -9, sz)
1444 
1445 #define KVM_GENERIC_VM_STATS()						       \
1446 	STATS_DESC_COUNTER(VM_GENERIC, remote_tlb_flush),		       \
1447 	STATS_DESC_COUNTER(VM_GENERIC, remote_tlb_flush_requests)
1448 
1449 #define KVM_GENERIC_VCPU_STATS()					       \
1450 	STATS_DESC_COUNTER(VCPU_GENERIC, halt_successful_poll),		       \
1451 	STATS_DESC_COUNTER(VCPU_GENERIC, halt_attempted_poll),		       \
1452 	STATS_DESC_COUNTER(VCPU_GENERIC, halt_poll_invalid),		       \
1453 	STATS_DESC_COUNTER(VCPU_GENERIC, halt_wakeup),			       \
1454 	STATS_DESC_TIME_NSEC(VCPU_GENERIC, halt_poll_success_ns),	       \
1455 	STATS_DESC_TIME_NSEC(VCPU_GENERIC, halt_poll_fail_ns),		       \
1456 	STATS_DESC_TIME_NSEC(VCPU_GENERIC, halt_wait_ns),		       \
1457 	STATS_DESC_LOGHIST_TIME_NSEC(VCPU_GENERIC, halt_poll_success_hist,     \
1458 			HALT_POLL_HIST_COUNT),				       \
1459 	STATS_DESC_LOGHIST_TIME_NSEC(VCPU_GENERIC, halt_poll_fail_hist,	       \
1460 			HALT_POLL_HIST_COUNT),				       \
1461 	STATS_DESC_LOGHIST_TIME_NSEC(VCPU_GENERIC, halt_wait_hist,	       \
1462 			HALT_POLL_HIST_COUNT)
1463 
1464 extern struct dentry *kvm_debugfs_dir;
1465 
1466 ssize_t kvm_stats_read(char *id, const struct kvm_stats_header *header,
1467 		       const struct _kvm_stats_desc *desc,
1468 		       void *stats, size_t size_stats,
1469 		       char __user *user_buffer, size_t size, loff_t *offset);
1470 
1471 /**
1472  * kvm_stats_linear_hist_update() - Update bucket value for linear histogram
1473  * statistics data.
1474  *
1475  * @data: start address of the stats data
1476  * @size: the number of bucket of the stats data
1477  * @value: the new value used to update the linear histogram's bucket
1478  * @bucket_size: the size (width) of a bucket
1479  */
1480 static inline void kvm_stats_linear_hist_update(u64 *data, size_t size,
1481 						u64 value, size_t bucket_size)
1482 {
1483 	size_t index = div64_u64(value, bucket_size);
1484 
1485 	index = min(index, size - 1);
1486 	++data[index];
1487 }
1488 
1489 /**
1490  * kvm_stats_log_hist_update() - Update bucket value for logarithmic histogram
1491  * statistics data.
1492  *
1493  * @data: start address of the stats data
1494  * @size: the number of bucket of the stats data
1495  * @value: the new value used to update the logarithmic histogram's bucket
1496  */
1497 static inline void kvm_stats_log_hist_update(u64 *data, size_t size, u64 value)
1498 {
1499 	size_t index = fls64(value);
1500 
1501 	index = min(index, size - 1);
1502 	++data[index];
1503 }
1504 
1505 #define KVM_STATS_LINEAR_HIST_UPDATE(array, value, bsize)		       \
1506 	kvm_stats_linear_hist_update(array, ARRAY_SIZE(array), value, bsize)
1507 #define KVM_STATS_LOG_HIST_UPDATE(array, value)				       \
1508 	kvm_stats_log_hist_update(array, ARRAY_SIZE(array), value)
1509 
1510 
1511 extern const struct kvm_stats_header kvm_vm_stats_header;
1512 extern const struct _kvm_stats_desc kvm_vm_stats_desc[];
1513 extern const struct kvm_stats_header kvm_vcpu_stats_header;
1514 extern const struct _kvm_stats_desc kvm_vcpu_stats_desc[];
1515 
1516 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1517 static inline int mmu_notifier_retry(struct kvm *kvm, unsigned long mmu_seq)
1518 {
1519 	if (unlikely(kvm->mmu_notifier_count))
1520 		return 1;
1521 	/*
1522 	 * Ensure the read of mmu_notifier_count happens before the read
1523 	 * of mmu_notifier_seq.  This interacts with the smp_wmb() in
1524 	 * mmu_notifier_invalidate_range_end to make sure that the caller
1525 	 * either sees the old (non-zero) value of mmu_notifier_count or
1526 	 * the new (incremented) value of mmu_notifier_seq.
1527 	 * PowerPC Book3s HV KVM calls this under a per-page lock
1528 	 * rather than under kvm->mmu_lock, for scalability, so
1529 	 * can't rely on kvm->mmu_lock to keep things ordered.
1530 	 */
1531 	smp_rmb();
1532 	if (kvm->mmu_notifier_seq != mmu_seq)
1533 		return 1;
1534 	return 0;
1535 }
1536 
1537 static inline int mmu_notifier_retry_hva(struct kvm *kvm,
1538 					 unsigned long mmu_seq,
1539 					 unsigned long hva)
1540 {
1541 	lockdep_assert_held(&kvm->mmu_lock);
1542 	/*
1543 	 * If mmu_notifier_count is non-zero, then the range maintained by
1544 	 * kvm_mmu_notifier_invalidate_range_start contains all addresses that
1545 	 * might be being invalidated. Note that it may include some false
1546 	 * positives, due to shortcuts when handing concurrent invalidations.
1547 	 */
1548 	if (unlikely(kvm->mmu_notifier_count) &&
1549 	    hva >= kvm->mmu_notifier_range_start &&
1550 	    hva < kvm->mmu_notifier_range_end)
1551 		return 1;
1552 	if (kvm->mmu_notifier_seq != mmu_seq)
1553 		return 1;
1554 	return 0;
1555 }
1556 #endif
1557 
1558 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
1559 
1560 #define KVM_MAX_IRQ_ROUTES 4096 /* might need extension/rework in the future */
1561 
1562 bool kvm_arch_can_set_irq_routing(struct kvm *kvm);
1563 int kvm_set_irq_routing(struct kvm *kvm,
1564 			const struct kvm_irq_routing_entry *entries,
1565 			unsigned nr,
1566 			unsigned flags);
1567 int kvm_set_routing_entry(struct kvm *kvm,
1568 			  struct kvm_kernel_irq_routing_entry *e,
1569 			  const struct kvm_irq_routing_entry *ue);
1570 void kvm_free_irq_routing(struct kvm *kvm);
1571 
1572 #else
1573 
1574 static inline void kvm_free_irq_routing(struct kvm *kvm) {}
1575 
1576 #endif
1577 
1578 int kvm_send_userspace_msi(struct kvm *kvm, struct kvm_msi *msi);
1579 
1580 #ifdef CONFIG_HAVE_KVM_EVENTFD
1581 
1582 void kvm_eventfd_init(struct kvm *kvm);
1583 int kvm_ioeventfd(struct kvm *kvm, struct kvm_ioeventfd *args);
1584 
1585 #ifdef CONFIG_HAVE_KVM_IRQFD
1586 int kvm_irqfd(struct kvm *kvm, struct kvm_irqfd *args);
1587 void kvm_irqfd_release(struct kvm *kvm);
1588 void kvm_irq_routing_update(struct kvm *);
1589 #else
1590 static inline int kvm_irqfd(struct kvm *kvm, struct kvm_irqfd *args)
1591 {
1592 	return -EINVAL;
1593 }
1594 
1595 static inline void kvm_irqfd_release(struct kvm *kvm) {}
1596 #endif
1597 
1598 #else
1599 
1600 static inline void kvm_eventfd_init(struct kvm *kvm) {}
1601 
1602 static inline int kvm_irqfd(struct kvm *kvm, struct kvm_irqfd *args)
1603 {
1604 	return -EINVAL;
1605 }
1606 
1607 static inline void kvm_irqfd_release(struct kvm *kvm) {}
1608 
1609 #ifdef CONFIG_HAVE_KVM_IRQCHIP
1610 static inline void kvm_irq_routing_update(struct kvm *kvm)
1611 {
1612 }
1613 #endif
1614 
1615 static inline int kvm_ioeventfd(struct kvm *kvm, struct kvm_ioeventfd *args)
1616 {
1617 	return -ENOSYS;
1618 }
1619 
1620 #endif /* CONFIG_HAVE_KVM_EVENTFD */
1621 
1622 void kvm_arch_irq_routing_update(struct kvm *kvm);
1623 
1624 static inline void kvm_make_request(int req, struct kvm_vcpu *vcpu)
1625 {
1626 	/*
1627 	 * Ensure the rest of the request is published to kvm_check_request's
1628 	 * caller.  Paired with the smp_mb__after_atomic in kvm_check_request.
1629 	 */
1630 	smp_wmb();
1631 	set_bit(req & KVM_REQUEST_MASK, (void *)&vcpu->requests);
1632 }
1633 
1634 static inline bool kvm_request_pending(struct kvm_vcpu *vcpu)
1635 {
1636 	return READ_ONCE(vcpu->requests);
1637 }
1638 
1639 static inline bool kvm_test_request(int req, struct kvm_vcpu *vcpu)
1640 {
1641 	return test_bit(req & KVM_REQUEST_MASK, (void *)&vcpu->requests);
1642 }
1643 
1644 static inline void kvm_clear_request(int req, struct kvm_vcpu *vcpu)
1645 {
1646 	clear_bit(req & KVM_REQUEST_MASK, (void *)&vcpu->requests);
1647 }
1648 
1649 static inline bool kvm_check_request(int req, struct kvm_vcpu *vcpu)
1650 {
1651 	if (kvm_test_request(req, vcpu)) {
1652 		kvm_clear_request(req, vcpu);
1653 
1654 		/*
1655 		 * Ensure the rest of the request is visible to kvm_check_request's
1656 		 * caller.  Paired with the smp_wmb in kvm_make_request.
1657 		 */
1658 		smp_mb__after_atomic();
1659 		return true;
1660 	} else {
1661 		return false;
1662 	}
1663 }
1664 
1665 extern bool kvm_rebooting;
1666 
1667 extern unsigned int halt_poll_ns;
1668 extern unsigned int halt_poll_ns_grow;
1669 extern unsigned int halt_poll_ns_grow_start;
1670 extern unsigned int halt_poll_ns_shrink;
1671 
1672 struct kvm_device {
1673 	const struct kvm_device_ops *ops;
1674 	struct kvm *kvm;
1675 	void *private;
1676 	struct list_head vm_node;
1677 };
1678 
1679 /* create, destroy, and name are mandatory */
1680 struct kvm_device_ops {
1681 	const char *name;
1682 
1683 	/*
1684 	 * create is called holding kvm->lock and any operations not suitable
1685 	 * to do while holding the lock should be deferred to init (see
1686 	 * below).
1687 	 */
1688 	int (*create)(struct kvm_device *dev, u32 type);
1689 
1690 	/*
1691 	 * init is called after create if create is successful and is called
1692 	 * outside of holding kvm->lock.
1693 	 */
1694 	void (*init)(struct kvm_device *dev);
1695 
1696 	/*
1697 	 * Destroy is responsible for freeing dev.
1698 	 *
1699 	 * Destroy may be called before or after destructors are called
1700 	 * on emulated I/O regions, depending on whether a reference is
1701 	 * held by a vcpu or other kvm component that gets destroyed
1702 	 * after the emulated I/O.
1703 	 */
1704 	void (*destroy)(struct kvm_device *dev);
1705 
1706 	/*
1707 	 * Release is an alternative method to free the device. It is
1708 	 * called when the device file descriptor is closed. Once
1709 	 * release is called, the destroy method will not be called
1710 	 * anymore as the device is removed from the device list of
1711 	 * the VM. kvm->lock is held.
1712 	 */
1713 	void (*release)(struct kvm_device *dev);
1714 
1715 	int (*set_attr)(struct kvm_device *dev, struct kvm_device_attr *attr);
1716 	int (*get_attr)(struct kvm_device *dev, struct kvm_device_attr *attr);
1717 	int (*has_attr)(struct kvm_device *dev, struct kvm_device_attr *attr);
1718 	long (*ioctl)(struct kvm_device *dev, unsigned int ioctl,
1719 		      unsigned long arg);
1720 	int (*mmap)(struct kvm_device *dev, struct vm_area_struct *vma);
1721 };
1722 
1723 void kvm_device_get(struct kvm_device *dev);
1724 void kvm_device_put(struct kvm_device *dev);
1725 struct kvm_device *kvm_device_from_filp(struct file *filp);
1726 int kvm_register_device_ops(const struct kvm_device_ops *ops, u32 type);
1727 void kvm_unregister_device_ops(u32 type);
1728 
1729 extern struct kvm_device_ops kvm_mpic_ops;
1730 extern struct kvm_device_ops kvm_arm_vgic_v2_ops;
1731 extern struct kvm_device_ops kvm_arm_vgic_v3_ops;
1732 
1733 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1734 
1735 static inline void kvm_vcpu_set_in_spin_loop(struct kvm_vcpu *vcpu, bool val)
1736 {
1737 	vcpu->spin_loop.in_spin_loop = val;
1738 }
1739 static inline void kvm_vcpu_set_dy_eligible(struct kvm_vcpu *vcpu, bool val)
1740 {
1741 	vcpu->spin_loop.dy_eligible = val;
1742 }
1743 
1744 #else /* !CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT */
1745 
1746 static inline void kvm_vcpu_set_in_spin_loop(struct kvm_vcpu *vcpu, bool val)
1747 {
1748 }
1749 
1750 static inline void kvm_vcpu_set_dy_eligible(struct kvm_vcpu *vcpu, bool val)
1751 {
1752 }
1753 #endif /* CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT */
1754 
1755 static inline bool kvm_is_visible_memslot(struct kvm_memory_slot *memslot)
1756 {
1757 	return (memslot && memslot->id < KVM_USER_MEM_SLOTS &&
1758 		!(memslot->flags & KVM_MEMSLOT_INVALID));
1759 }
1760 
1761 struct kvm_vcpu *kvm_get_running_vcpu(void);
1762 struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void);
1763 
1764 #ifdef CONFIG_HAVE_KVM_IRQ_BYPASS
1765 bool kvm_arch_has_irq_bypass(void);
1766 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *,
1767 			   struct irq_bypass_producer *);
1768 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *,
1769 			   struct irq_bypass_producer *);
1770 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *);
1771 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *);
1772 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
1773 				  uint32_t guest_irq, bool set);
1774 #endif /* CONFIG_HAVE_KVM_IRQ_BYPASS */
1775 
1776 #ifdef CONFIG_HAVE_KVM_INVALID_WAKEUPS
1777 /* If we wakeup during the poll time, was it a sucessful poll? */
1778 static inline bool vcpu_valid_wakeup(struct kvm_vcpu *vcpu)
1779 {
1780 	return vcpu->valid_wakeup;
1781 }
1782 
1783 #else
1784 static inline bool vcpu_valid_wakeup(struct kvm_vcpu *vcpu)
1785 {
1786 	return true;
1787 }
1788 #endif /* CONFIG_HAVE_KVM_INVALID_WAKEUPS */
1789 
1790 #ifdef CONFIG_HAVE_KVM_NO_POLL
1791 /* Callback that tells if we must not poll */
1792 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu);
1793 #else
1794 static inline bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
1795 {
1796 	return false;
1797 }
1798 #endif /* CONFIG_HAVE_KVM_NO_POLL */
1799 
1800 #ifdef CONFIG_HAVE_KVM_VCPU_ASYNC_IOCTL
1801 long kvm_arch_vcpu_async_ioctl(struct file *filp,
1802 			       unsigned int ioctl, unsigned long arg);
1803 #else
1804 static inline long kvm_arch_vcpu_async_ioctl(struct file *filp,
1805 					     unsigned int ioctl,
1806 					     unsigned long arg)
1807 {
1808 	return -ENOIOCTLCMD;
1809 }
1810 #endif /* CONFIG_HAVE_KVM_VCPU_ASYNC_IOCTL */
1811 
1812 void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
1813 					    unsigned long start, unsigned long end);
1814 
1815 #ifdef CONFIG_HAVE_KVM_VCPU_RUN_PID_CHANGE
1816 int kvm_arch_vcpu_run_pid_change(struct kvm_vcpu *vcpu);
1817 #else
1818 static inline int kvm_arch_vcpu_run_pid_change(struct kvm_vcpu *vcpu)
1819 {
1820 	return 0;
1821 }
1822 #endif /* CONFIG_HAVE_KVM_VCPU_RUN_PID_CHANGE */
1823 
1824 typedef int (*kvm_vm_thread_fn_t)(struct kvm *kvm, uintptr_t data);
1825 
1826 int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn,
1827 				uintptr_t data, const char *name,
1828 				struct task_struct **thread_ptr);
1829 
1830 #ifdef CONFIG_KVM_XFER_TO_GUEST_WORK
1831 static inline void kvm_handle_signal_exit(struct kvm_vcpu *vcpu)
1832 {
1833 	vcpu->run->exit_reason = KVM_EXIT_INTR;
1834 	vcpu->stat.signal_exits++;
1835 }
1836 #endif /* CONFIG_KVM_XFER_TO_GUEST_WORK */
1837 
1838 /*
1839  * This defines how many reserved entries we want to keep before we
1840  * kick the vcpu to the userspace to avoid dirty ring full.  This
1841  * value can be tuned to higher if e.g. PML is enabled on the host.
1842  */
1843 #define  KVM_DIRTY_RING_RSVD_ENTRIES  64
1844 
1845 /* Max number of entries allowed for each kvm dirty ring */
1846 #define  KVM_DIRTY_RING_MAX_ENTRIES  65536
1847 
1848 #endif
1849