xref: /linux-6.15/include/linux/kvm_host.h (revision b60a5b8d)
1 #ifndef __KVM_HOST_H
2 #define __KVM_HOST_H
3 
4 /*
5  * This work is licensed under the terms of the GNU GPL, version 2.  See
6  * the COPYING file in the top-level directory.
7  */
8 
9 #include <linux/types.h>
10 #include <linux/hardirq.h>
11 #include <linux/list.h>
12 #include <linux/mutex.h>
13 #include <linux/spinlock.h>
14 #include <linux/signal.h>
15 #include <linux/sched.h>
16 #include <linux/bug.h>
17 #include <linux/mm.h>
18 #include <linux/mmu_notifier.h>
19 #include <linux/preempt.h>
20 #include <linux/msi.h>
21 #include <linux/slab.h>
22 #include <linux/vmalloc.h>
23 #include <linux/rcupdate.h>
24 #include <linux/ratelimit.h>
25 #include <linux/err.h>
26 #include <linux/irqflags.h>
27 #include <linux/context_tracking.h>
28 #include <linux/irqbypass.h>
29 #include <linux/swait.h>
30 #include <linux/refcount.h>
31 #include <asm/signal.h>
32 
33 #include <linux/kvm.h>
34 #include <linux/kvm_para.h>
35 
36 #include <linux/kvm_types.h>
37 
38 #include <asm/kvm_host.h>
39 
40 #ifndef KVM_MAX_VCPU_ID
41 #define KVM_MAX_VCPU_ID KVM_MAX_VCPUS
42 #endif
43 
44 /*
45  * The bit 16 ~ bit 31 of kvm_memory_region::flags are internally used
46  * in kvm, other bits are visible for userspace which are defined in
47  * include/linux/kvm_h.
48  */
49 #define KVM_MEMSLOT_INVALID	(1UL << 16)
50 
51 /*
52  * Bit 63 of the memslot generation number is an "update in-progress flag",
53  * e.g. is temporarily set for the duration of install_new_memslots().
54  * This flag effectively creates a unique generation number that is used to
55  * mark cached memslot data, e.g. MMIO accesses, as potentially being stale,
56  * i.e. may (or may not) have come from the previous memslots generation.
57  *
58  * This is necessary because the actual memslots update is not atomic with
59  * respect to the generation number update.  Updating the generation number
60  * first would allow a vCPU to cache a spte from the old memslots using the
61  * new generation number, and updating the generation number after switching
62  * to the new memslots would allow cache hits using the old generation number
63  * to reference the defunct memslots.
64  *
65  * This mechanism is used to prevent getting hits in KVM's caches while a
66  * memslot update is in-progress, and to prevent cache hits *after* updating
67  * the actual generation number against accesses that were inserted into the
68  * cache *before* the memslots were updated.
69  */
70 #define KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS	BIT_ULL(63)
71 
72 /* Two fragments for cross MMIO pages. */
73 #define KVM_MAX_MMIO_FRAGMENTS	2
74 
75 #ifndef KVM_ADDRESS_SPACE_NUM
76 #define KVM_ADDRESS_SPACE_NUM	1
77 #endif
78 
79 /*
80  * For the normal pfn, the highest 12 bits should be zero,
81  * so we can mask bit 62 ~ bit 52  to indicate the error pfn,
82  * mask bit 63 to indicate the noslot pfn.
83  */
84 #define KVM_PFN_ERR_MASK	(0x7ffULL << 52)
85 #define KVM_PFN_ERR_NOSLOT_MASK	(0xfffULL << 52)
86 #define KVM_PFN_NOSLOT		(0x1ULL << 63)
87 
88 #define KVM_PFN_ERR_FAULT	(KVM_PFN_ERR_MASK)
89 #define KVM_PFN_ERR_HWPOISON	(KVM_PFN_ERR_MASK + 1)
90 #define KVM_PFN_ERR_RO_FAULT	(KVM_PFN_ERR_MASK + 2)
91 
92 /*
93  * error pfns indicate that the gfn is in slot but faild to
94  * translate it to pfn on host.
95  */
96 static inline bool is_error_pfn(kvm_pfn_t pfn)
97 {
98 	return !!(pfn & KVM_PFN_ERR_MASK);
99 }
100 
101 /*
102  * error_noslot pfns indicate that the gfn can not be
103  * translated to pfn - it is not in slot or failed to
104  * translate it to pfn.
105  */
106 static inline bool is_error_noslot_pfn(kvm_pfn_t pfn)
107 {
108 	return !!(pfn & KVM_PFN_ERR_NOSLOT_MASK);
109 }
110 
111 /* noslot pfn indicates that the gfn is not in slot. */
112 static inline bool is_noslot_pfn(kvm_pfn_t pfn)
113 {
114 	return pfn == KVM_PFN_NOSLOT;
115 }
116 
117 /*
118  * architectures with KVM_HVA_ERR_BAD other than PAGE_OFFSET (e.g. s390)
119  * provide own defines and kvm_is_error_hva
120  */
121 #ifndef KVM_HVA_ERR_BAD
122 
123 #define KVM_HVA_ERR_BAD		(PAGE_OFFSET)
124 #define KVM_HVA_ERR_RO_BAD	(PAGE_OFFSET + PAGE_SIZE)
125 
126 static inline bool kvm_is_error_hva(unsigned long addr)
127 {
128 	return addr >= PAGE_OFFSET;
129 }
130 
131 #endif
132 
133 #define KVM_ERR_PTR_BAD_PAGE	(ERR_PTR(-ENOENT))
134 
135 static inline bool is_error_page(struct page *page)
136 {
137 	return IS_ERR(page);
138 }
139 
140 #define KVM_REQUEST_MASK           GENMASK(7,0)
141 #define KVM_REQUEST_NO_WAKEUP      BIT(8)
142 #define KVM_REQUEST_WAIT           BIT(9)
143 /*
144  * Architecture-independent vcpu->requests bit members
145  * Bits 4-7 are reserved for more arch-independent bits.
146  */
147 #define KVM_REQ_TLB_FLUSH         (0 | KVM_REQUEST_WAIT | KVM_REQUEST_NO_WAKEUP)
148 #define KVM_REQ_MMU_RELOAD        (1 | KVM_REQUEST_WAIT | KVM_REQUEST_NO_WAKEUP)
149 #define KVM_REQ_PENDING_TIMER     2
150 #define KVM_REQ_UNHALT            3
151 #define KVM_REQUEST_ARCH_BASE     8
152 
153 #define KVM_ARCH_REQ_FLAGS(nr, flags) ({ \
154 	BUILD_BUG_ON((unsigned)(nr) >= (FIELD_SIZEOF(struct kvm_vcpu, requests) * 8) - KVM_REQUEST_ARCH_BASE); \
155 	(unsigned)(((nr) + KVM_REQUEST_ARCH_BASE) | (flags)); \
156 })
157 #define KVM_ARCH_REQ(nr)           KVM_ARCH_REQ_FLAGS(nr, 0)
158 
159 #define KVM_USERSPACE_IRQ_SOURCE_ID		0
160 #define KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID	1
161 
162 extern struct kmem_cache *kvm_vcpu_cache;
163 
164 extern spinlock_t kvm_lock;
165 extern struct list_head vm_list;
166 
167 struct kvm_io_range {
168 	gpa_t addr;
169 	int len;
170 	struct kvm_io_device *dev;
171 };
172 
173 #define NR_IOBUS_DEVS 1000
174 
175 struct kvm_io_bus {
176 	int dev_count;
177 	int ioeventfd_count;
178 	struct kvm_io_range range[];
179 };
180 
181 enum kvm_bus {
182 	KVM_MMIO_BUS,
183 	KVM_PIO_BUS,
184 	KVM_VIRTIO_CCW_NOTIFY_BUS,
185 	KVM_FAST_MMIO_BUS,
186 	KVM_NR_BUSES
187 };
188 
189 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
190 		     int len, const void *val);
191 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
192 			    gpa_t addr, int len, const void *val, long cookie);
193 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
194 		    int len, void *val);
195 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
196 			    int len, struct kvm_io_device *dev);
197 void kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
198 			       struct kvm_io_device *dev);
199 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
200 					 gpa_t addr);
201 
202 #ifdef CONFIG_KVM_ASYNC_PF
203 struct kvm_async_pf {
204 	struct work_struct work;
205 	struct list_head link;
206 	struct list_head queue;
207 	struct kvm_vcpu *vcpu;
208 	struct mm_struct *mm;
209 	gva_t gva;
210 	unsigned long addr;
211 	struct kvm_arch_async_pf arch;
212 	bool   wakeup_all;
213 };
214 
215 void kvm_clear_async_pf_completion_queue(struct kvm_vcpu *vcpu);
216 void kvm_check_async_pf_completion(struct kvm_vcpu *vcpu);
217 int kvm_setup_async_pf(struct kvm_vcpu *vcpu, gva_t gva, unsigned long hva,
218 		       struct kvm_arch_async_pf *arch);
219 int kvm_async_pf_wakeup_all(struct kvm_vcpu *vcpu);
220 #endif
221 
222 enum {
223 	OUTSIDE_GUEST_MODE,
224 	IN_GUEST_MODE,
225 	EXITING_GUEST_MODE,
226 	READING_SHADOW_PAGE_TABLES,
227 };
228 
229 /*
230  * Sometimes a large or cross-page mmio needs to be broken up into separate
231  * exits for userspace servicing.
232  */
233 struct kvm_mmio_fragment {
234 	gpa_t gpa;
235 	void *data;
236 	unsigned len;
237 };
238 
239 struct kvm_vcpu {
240 	struct kvm *kvm;
241 #ifdef CONFIG_PREEMPT_NOTIFIERS
242 	struct preempt_notifier preempt_notifier;
243 #endif
244 	int cpu;
245 	int vcpu_id;
246 	int srcu_idx;
247 	int mode;
248 	u64 requests;
249 	unsigned long guest_debug;
250 
251 	int pre_pcpu;
252 	struct list_head blocked_vcpu_list;
253 
254 	struct mutex mutex;
255 	struct kvm_run *run;
256 
257 	int guest_xcr0_loaded;
258 	struct swait_queue_head wq;
259 	struct pid __rcu *pid;
260 	int sigset_active;
261 	sigset_t sigset;
262 	struct kvm_vcpu_stat stat;
263 	unsigned int halt_poll_ns;
264 	bool valid_wakeup;
265 
266 #ifdef CONFIG_HAS_IOMEM
267 	int mmio_needed;
268 	int mmio_read_completed;
269 	int mmio_is_write;
270 	int mmio_cur_fragment;
271 	int mmio_nr_fragments;
272 	struct kvm_mmio_fragment mmio_fragments[KVM_MAX_MMIO_FRAGMENTS];
273 #endif
274 
275 #ifdef CONFIG_KVM_ASYNC_PF
276 	struct {
277 		u32 queued;
278 		struct list_head queue;
279 		struct list_head done;
280 		spinlock_t lock;
281 	} async_pf;
282 #endif
283 
284 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
285 	/*
286 	 * Cpu relax intercept or pause loop exit optimization
287 	 * in_spin_loop: set when a vcpu does a pause loop exit
288 	 *  or cpu relax intercepted.
289 	 * dy_eligible: indicates whether vcpu is eligible for directed yield.
290 	 */
291 	struct {
292 		bool in_spin_loop;
293 		bool dy_eligible;
294 	} spin_loop;
295 #endif
296 	bool preempted;
297 	struct kvm_vcpu_arch arch;
298 	struct dentry *debugfs_dentry;
299 };
300 
301 static inline int kvm_vcpu_exiting_guest_mode(struct kvm_vcpu *vcpu)
302 {
303 	/*
304 	 * The memory barrier ensures a previous write to vcpu->requests cannot
305 	 * be reordered with the read of vcpu->mode.  It pairs with the general
306 	 * memory barrier following the write of vcpu->mode in VCPU RUN.
307 	 */
308 	smp_mb__before_atomic();
309 	return cmpxchg(&vcpu->mode, IN_GUEST_MODE, EXITING_GUEST_MODE);
310 }
311 
312 /*
313  * Some of the bitops functions do not support too long bitmaps.
314  * This number must be determined not to exceed such limits.
315  */
316 #define KVM_MEM_MAX_NR_PAGES ((1UL << 31) - 1)
317 
318 struct kvm_memory_slot {
319 	gfn_t base_gfn;
320 	unsigned long npages;
321 	unsigned long *dirty_bitmap;
322 	struct kvm_arch_memory_slot arch;
323 	unsigned long userspace_addr;
324 	u32 flags;
325 	short id;
326 };
327 
328 static inline unsigned long kvm_dirty_bitmap_bytes(struct kvm_memory_slot *memslot)
329 {
330 	return ALIGN(memslot->npages, BITS_PER_LONG) / 8;
331 }
332 
333 static inline unsigned long *kvm_second_dirty_bitmap(struct kvm_memory_slot *memslot)
334 {
335 	unsigned long len = kvm_dirty_bitmap_bytes(memslot);
336 
337 	return memslot->dirty_bitmap + len / sizeof(*memslot->dirty_bitmap);
338 }
339 
340 struct kvm_s390_adapter_int {
341 	u64 ind_addr;
342 	u64 summary_addr;
343 	u64 ind_offset;
344 	u32 summary_offset;
345 	u32 adapter_id;
346 };
347 
348 struct kvm_hv_sint {
349 	u32 vcpu;
350 	u32 sint;
351 };
352 
353 struct kvm_kernel_irq_routing_entry {
354 	u32 gsi;
355 	u32 type;
356 	int (*set)(struct kvm_kernel_irq_routing_entry *e,
357 		   struct kvm *kvm, int irq_source_id, int level,
358 		   bool line_status);
359 	union {
360 		struct {
361 			unsigned irqchip;
362 			unsigned pin;
363 		} irqchip;
364 		struct {
365 			u32 address_lo;
366 			u32 address_hi;
367 			u32 data;
368 			u32 flags;
369 			u32 devid;
370 		} msi;
371 		struct kvm_s390_adapter_int adapter;
372 		struct kvm_hv_sint hv_sint;
373 	};
374 	struct hlist_node link;
375 };
376 
377 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
378 struct kvm_irq_routing_table {
379 	int chip[KVM_NR_IRQCHIPS][KVM_IRQCHIP_NUM_PINS];
380 	u32 nr_rt_entries;
381 	/*
382 	 * Array indexed by gsi. Each entry contains list of irq chips
383 	 * the gsi is connected to.
384 	 */
385 	struct hlist_head map[0];
386 };
387 #endif
388 
389 #ifndef KVM_PRIVATE_MEM_SLOTS
390 #define KVM_PRIVATE_MEM_SLOTS 0
391 #endif
392 
393 #ifndef KVM_MEM_SLOTS_NUM
394 #define KVM_MEM_SLOTS_NUM (KVM_USER_MEM_SLOTS + KVM_PRIVATE_MEM_SLOTS)
395 #endif
396 
397 #ifndef __KVM_VCPU_MULTIPLE_ADDRESS_SPACE
398 static inline int kvm_arch_vcpu_memslots_id(struct kvm_vcpu *vcpu)
399 {
400 	return 0;
401 }
402 #endif
403 
404 /*
405  * Note:
406  * memslots are not sorted by id anymore, please use id_to_memslot()
407  * to get the memslot by its id.
408  */
409 struct kvm_memslots {
410 	u64 generation;
411 	struct kvm_memory_slot memslots[KVM_MEM_SLOTS_NUM];
412 	/* The mapping table from slot id to the index in memslots[]. */
413 	short id_to_index[KVM_MEM_SLOTS_NUM];
414 	atomic_t lru_slot;
415 	int used_slots;
416 };
417 
418 struct kvm {
419 	spinlock_t mmu_lock;
420 	struct mutex slots_lock;
421 	struct mm_struct *mm; /* userspace tied to this vm */
422 	struct kvm_memslots __rcu *memslots[KVM_ADDRESS_SPACE_NUM];
423 	struct kvm_vcpu *vcpus[KVM_MAX_VCPUS];
424 
425 	/*
426 	 * created_vcpus is protected by kvm->lock, and is incremented
427 	 * at the beginning of KVM_CREATE_VCPU.  online_vcpus is only
428 	 * incremented after storing the kvm_vcpu pointer in vcpus,
429 	 * and is accessed atomically.
430 	 */
431 	atomic_t online_vcpus;
432 	int created_vcpus;
433 	int last_boosted_vcpu;
434 	struct list_head vm_list;
435 	struct mutex lock;
436 	struct kvm_io_bus __rcu *buses[KVM_NR_BUSES];
437 #ifdef CONFIG_HAVE_KVM_EVENTFD
438 	struct {
439 		spinlock_t        lock;
440 		struct list_head  items;
441 		struct list_head  resampler_list;
442 		struct mutex      resampler_lock;
443 	} irqfds;
444 	struct list_head ioeventfds;
445 #endif
446 	struct kvm_vm_stat stat;
447 	struct kvm_arch arch;
448 	refcount_t users_count;
449 #ifdef CONFIG_KVM_MMIO
450 	struct kvm_coalesced_mmio_ring *coalesced_mmio_ring;
451 	spinlock_t ring_lock;
452 	struct list_head coalesced_zones;
453 #endif
454 
455 	struct mutex irq_lock;
456 #ifdef CONFIG_HAVE_KVM_IRQCHIP
457 	/*
458 	 * Update side is protected by irq_lock.
459 	 */
460 	struct kvm_irq_routing_table __rcu *irq_routing;
461 #endif
462 #ifdef CONFIG_HAVE_KVM_IRQFD
463 	struct hlist_head irq_ack_notifier_list;
464 #endif
465 
466 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
467 	struct mmu_notifier mmu_notifier;
468 	unsigned long mmu_notifier_seq;
469 	long mmu_notifier_count;
470 #endif
471 	long tlbs_dirty;
472 	struct list_head devices;
473 	bool manual_dirty_log_protect;
474 	struct dentry *debugfs_dentry;
475 	struct kvm_stat_data **debugfs_stat_data;
476 	struct srcu_struct srcu;
477 	struct srcu_struct irq_srcu;
478 	pid_t userspace_pid;
479 };
480 
481 #define kvm_err(fmt, ...) \
482 	pr_err("kvm [%i]: " fmt, task_pid_nr(current), ## __VA_ARGS__)
483 #define kvm_info(fmt, ...) \
484 	pr_info("kvm [%i]: " fmt, task_pid_nr(current), ## __VA_ARGS__)
485 #define kvm_debug(fmt, ...) \
486 	pr_debug("kvm [%i]: " fmt, task_pid_nr(current), ## __VA_ARGS__)
487 #define kvm_debug_ratelimited(fmt, ...) \
488 	pr_debug_ratelimited("kvm [%i]: " fmt, task_pid_nr(current), \
489 			     ## __VA_ARGS__)
490 #define kvm_pr_unimpl(fmt, ...) \
491 	pr_err_ratelimited("kvm [%i]: " fmt, \
492 			   task_tgid_nr(current), ## __VA_ARGS__)
493 
494 /* The guest did something we don't support. */
495 #define vcpu_unimpl(vcpu, fmt, ...)					\
496 	kvm_pr_unimpl("vcpu%i, guest rIP: 0x%lx " fmt,			\
497 			(vcpu)->vcpu_id, kvm_rip_read(vcpu), ## __VA_ARGS__)
498 
499 #define vcpu_debug(vcpu, fmt, ...)					\
500 	kvm_debug("vcpu%i " fmt, (vcpu)->vcpu_id, ## __VA_ARGS__)
501 #define vcpu_debug_ratelimited(vcpu, fmt, ...)				\
502 	kvm_debug_ratelimited("vcpu%i " fmt, (vcpu)->vcpu_id,           \
503 			      ## __VA_ARGS__)
504 #define vcpu_err(vcpu, fmt, ...)					\
505 	kvm_err("vcpu%i " fmt, (vcpu)->vcpu_id, ## __VA_ARGS__)
506 
507 static inline struct kvm_io_bus *kvm_get_bus(struct kvm *kvm, enum kvm_bus idx)
508 {
509 	return srcu_dereference_check(kvm->buses[idx], &kvm->srcu,
510 				      lockdep_is_held(&kvm->slots_lock) ||
511 				      !refcount_read(&kvm->users_count));
512 }
513 
514 static inline struct kvm_vcpu *kvm_get_vcpu(struct kvm *kvm, int i)
515 {
516 	/* Pairs with smp_wmb() in kvm_vm_ioctl_create_vcpu, in case
517 	 * the caller has read kvm->online_vcpus before (as is the case
518 	 * for kvm_for_each_vcpu, for example).
519 	 */
520 	smp_rmb();
521 	return kvm->vcpus[i];
522 }
523 
524 #define kvm_for_each_vcpu(idx, vcpup, kvm) \
525 	for (idx = 0; \
526 	     idx < atomic_read(&kvm->online_vcpus) && \
527 	     (vcpup = kvm_get_vcpu(kvm, idx)) != NULL; \
528 	     idx++)
529 
530 static inline struct kvm_vcpu *kvm_get_vcpu_by_id(struct kvm *kvm, int id)
531 {
532 	struct kvm_vcpu *vcpu = NULL;
533 	int i;
534 
535 	if (id < 0)
536 		return NULL;
537 	if (id < KVM_MAX_VCPUS)
538 		vcpu = kvm_get_vcpu(kvm, id);
539 	if (vcpu && vcpu->vcpu_id == id)
540 		return vcpu;
541 	kvm_for_each_vcpu(i, vcpu, kvm)
542 		if (vcpu->vcpu_id == id)
543 			return vcpu;
544 	return NULL;
545 }
546 
547 static inline int kvm_vcpu_get_idx(struct kvm_vcpu *vcpu)
548 {
549 	struct kvm_vcpu *tmp;
550 	int idx;
551 
552 	kvm_for_each_vcpu(idx, tmp, vcpu->kvm)
553 		if (tmp == vcpu)
554 			return idx;
555 	BUG();
556 }
557 
558 #define kvm_for_each_memslot(memslot, slots)	\
559 	for (memslot = &slots->memslots[0];	\
560 	      memslot < slots->memslots + KVM_MEM_SLOTS_NUM && memslot->npages;\
561 		memslot++)
562 
563 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id);
564 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu);
565 
566 void vcpu_load(struct kvm_vcpu *vcpu);
567 void vcpu_put(struct kvm_vcpu *vcpu);
568 
569 #ifdef __KVM_HAVE_IOAPIC
570 void kvm_arch_post_irq_ack_notifier_list_update(struct kvm *kvm);
571 void kvm_arch_post_irq_routing_update(struct kvm *kvm);
572 #else
573 static inline void kvm_arch_post_irq_ack_notifier_list_update(struct kvm *kvm)
574 {
575 }
576 static inline void kvm_arch_post_irq_routing_update(struct kvm *kvm)
577 {
578 }
579 #endif
580 
581 #ifdef CONFIG_HAVE_KVM_IRQFD
582 int kvm_irqfd_init(void);
583 void kvm_irqfd_exit(void);
584 #else
585 static inline int kvm_irqfd_init(void)
586 {
587 	return 0;
588 }
589 
590 static inline void kvm_irqfd_exit(void)
591 {
592 }
593 #endif
594 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
595 		  struct module *module);
596 void kvm_exit(void);
597 
598 void kvm_get_kvm(struct kvm *kvm);
599 void kvm_put_kvm(struct kvm *kvm);
600 
601 static inline struct kvm_memslots *__kvm_memslots(struct kvm *kvm, int as_id)
602 {
603 	return srcu_dereference_check(kvm->memslots[as_id], &kvm->srcu,
604 			lockdep_is_held(&kvm->slots_lock) ||
605 			!refcount_read(&kvm->users_count));
606 }
607 
608 static inline struct kvm_memslots *kvm_memslots(struct kvm *kvm)
609 {
610 	return __kvm_memslots(kvm, 0);
611 }
612 
613 static inline struct kvm_memslots *kvm_vcpu_memslots(struct kvm_vcpu *vcpu)
614 {
615 	int as_id = kvm_arch_vcpu_memslots_id(vcpu);
616 
617 	return __kvm_memslots(vcpu->kvm, as_id);
618 }
619 
620 static inline struct kvm_memory_slot *
621 id_to_memslot(struct kvm_memslots *slots, int id)
622 {
623 	int index = slots->id_to_index[id];
624 	struct kvm_memory_slot *slot;
625 
626 	slot = &slots->memslots[index];
627 
628 	WARN_ON(slot->id != id);
629 	return slot;
630 }
631 
632 /*
633  * KVM_SET_USER_MEMORY_REGION ioctl allows the following operations:
634  * - create a new memory slot
635  * - delete an existing memory slot
636  * - modify an existing memory slot
637  *   -- move it in the guest physical memory space
638  *   -- just change its flags
639  *
640  * Since flags can be changed by some of these operations, the following
641  * differentiation is the best we can do for __kvm_set_memory_region():
642  */
643 enum kvm_mr_change {
644 	KVM_MR_CREATE,
645 	KVM_MR_DELETE,
646 	KVM_MR_MOVE,
647 	KVM_MR_FLAGS_ONLY,
648 };
649 
650 int kvm_set_memory_region(struct kvm *kvm,
651 			  const struct kvm_userspace_memory_region *mem);
652 int __kvm_set_memory_region(struct kvm *kvm,
653 			    const struct kvm_userspace_memory_region *mem);
654 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
655 			   struct kvm_memory_slot *dont);
656 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
657 			    unsigned long npages);
658 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen);
659 int kvm_arch_prepare_memory_region(struct kvm *kvm,
660 				struct kvm_memory_slot *memslot,
661 				const struct kvm_userspace_memory_region *mem,
662 				enum kvm_mr_change change);
663 void kvm_arch_commit_memory_region(struct kvm *kvm,
664 				const struct kvm_userspace_memory_region *mem,
665 				const struct kvm_memory_slot *old,
666 				const struct kvm_memory_slot *new,
667 				enum kvm_mr_change change);
668 bool kvm_largepages_enabled(void);
669 void kvm_disable_largepages(void);
670 /* flush all memory translations */
671 void kvm_arch_flush_shadow_all(struct kvm *kvm);
672 /* flush memory translations pointing to 'slot' */
673 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
674 				   struct kvm_memory_slot *slot);
675 
676 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
677 			    struct page **pages, int nr_pages);
678 
679 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn);
680 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn);
681 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable);
682 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot, gfn_t gfn);
683 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot, gfn_t gfn,
684 				      bool *writable);
685 void kvm_release_page_clean(struct page *page);
686 void kvm_release_page_dirty(struct page *page);
687 void kvm_set_page_accessed(struct page *page);
688 
689 kvm_pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn);
690 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn);
691 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
692 		      bool *writable);
693 kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn);
694 kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn);
695 kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
696 			       bool atomic, bool *async, bool write_fault,
697 			       bool *writable);
698 
699 void kvm_release_pfn_clean(kvm_pfn_t pfn);
700 void kvm_release_pfn_dirty(kvm_pfn_t pfn);
701 void kvm_set_pfn_dirty(kvm_pfn_t pfn);
702 void kvm_set_pfn_accessed(kvm_pfn_t pfn);
703 void kvm_get_pfn(kvm_pfn_t pfn);
704 
705 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
706 			int len);
707 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
708 			  unsigned long len);
709 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len);
710 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
711 			   void *data, unsigned long len);
712 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
713 			 int offset, int len);
714 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
715 		    unsigned long len);
716 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
717 			   void *data, unsigned long len);
718 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
719 				  void *data, unsigned int offset,
720 				  unsigned long len);
721 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
722 			      gpa_t gpa, unsigned long len);
723 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len);
724 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len);
725 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn);
726 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn);
727 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn);
728 void mark_page_dirty(struct kvm *kvm, gfn_t gfn);
729 
730 struct kvm_memslots *kvm_vcpu_memslots(struct kvm_vcpu *vcpu);
731 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn);
732 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn);
733 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn);
734 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn);
735 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn);
736 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable);
737 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data, int offset,
738 			     int len);
739 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa, void *data,
740 			       unsigned long len);
741 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data,
742 			unsigned long len);
743 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, const void *data,
744 			      int offset, int len);
745 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
746 			 unsigned long len);
747 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn);
748 
749 void kvm_sigset_activate(struct kvm_vcpu *vcpu);
750 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu);
751 
752 void kvm_vcpu_block(struct kvm_vcpu *vcpu);
753 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu);
754 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu);
755 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu);
756 void kvm_vcpu_kick(struct kvm_vcpu *vcpu);
757 int kvm_vcpu_yield_to(struct kvm_vcpu *target);
758 void kvm_vcpu_on_spin(struct kvm_vcpu *vcpu, bool usermode_vcpu_not_eligible);
759 
760 void kvm_flush_remote_tlbs(struct kvm *kvm);
761 void kvm_reload_remote_mmus(struct kvm *kvm);
762 
763 bool kvm_make_vcpus_request_mask(struct kvm *kvm, unsigned int req,
764 				 unsigned long *vcpu_bitmap, cpumask_var_t tmp);
765 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req);
766 
767 long kvm_arch_dev_ioctl(struct file *filp,
768 			unsigned int ioctl, unsigned long arg);
769 long kvm_arch_vcpu_ioctl(struct file *filp,
770 			 unsigned int ioctl, unsigned long arg);
771 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf);
772 
773 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext);
774 
775 int kvm_get_dirty_log(struct kvm *kvm,
776 			struct kvm_dirty_log *log, int *is_dirty);
777 
778 int kvm_get_dirty_log_protect(struct kvm *kvm,
779 			      struct kvm_dirty_log *log, bool *flush);
780 int kvm_clear_dirty_log_protect(struct kvm *kvm,
781 				struct kvm_clear_dirty_log *log, bool *flush);
782 
783 void kvm_arch_mmu_enable_log_dirty_pt_masked(struct kvm *kvm,
784 					struct kvm_memory_slot *slot,
785 					gfn_t gfn_offset,
786 					unsigned long mask);
787 
788 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
789 				struct kvm_dirty_log *log);
790 int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm,
791 				  struct kvm_clear_dirty_log *log);
792 
793 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
794 			bool line_status);
795 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
796 			    struct kvm_enable_cap *cap);
797 long kvm_arch_vm_ioctl(struct file *filp,
798 		       unsigned int ioctl, unsigned long arg);
799 
800 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu);
801 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu);
802 
803 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
804 				    struct kvm_translation *tr);
805 
806 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs);
807 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs);
808 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
809 				  struct kvm_sregs *sregs);
810 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
811 				  struct kvm_sregs *sregs);
812 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
813 				    struct kvm_mp_state *mp_state);
814 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
815 				    struct kvm_mp_state *mp_state);
816 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
817 					struct kvm_guest_debug *dbg);
818 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run);
819 
820 int kvm_arch_init(void *opaque);
821 void kvm_arch_exit(void);
822 
823 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu);
824 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu);
825 
826 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu);
827 
828 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu);
829 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu);
830 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu);
831 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id);
832 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu);
833 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu);
834 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu);
835 
836 bool kvm_arch_has_vcpu_debugfs(void);
837 int kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu);
838 
839 int kvm_arch_hardware_enable(void);
840 void kvm_arch_hardware_disable(void);
841 int kvm_arch_hardware_setup(void);
842 void kvm_arch_hardware_unsetup(void);
843 void kvm_arch_check_processor_compat(void *rtn);
844 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu);
845 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu);
846 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu);
847 
848 #ifndef __KVM_HAVE_ARCH_VM_ALLOC
849 /*
850  * All architectures that want to use vzalloc currently also
851  * need their own kvm_arch_alloc_vm implementation.
852  */
853 static inline struct kvm *kvm_arch_alloc_vm(void)
854 {
855 	return kzalloc(sizeof(struct kvm), GFP_KERNEL);
856 }
857 
858 static inline void kvm_arch_free_vm(struct kvm *kvm)
859 {
860 	kfree(kvm);
861 }
862 #endif
863 
864 #ifndef __KVM_HAVE_ARCH_FLUSH_REMOTE_TLB
865 static inline int kvm_arch_flush_remote_tlb(struct kvm *kvm)
866 {
867 	return -ENOTSUPP;
868 }
869 #endif
870 
871 #ifdef __KVM_HAVE_ARCH_NONCOHERENT_DMA
872 void kvm_arch_register_noncoherent_dma(struct kvm *kvm);
873 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm);
874 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm);
875 #else
876 static inline void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
877 {
878 }
879 
880 static inline void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
881 {
882 }
883 
884 static inline bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
885 {
886 	return false;
887 }
888 #endif
889 #ifdef __KVM_HAVE_ARCH_ASSIGNED_DEVICE
890 void kvm_arch_start_assignment(struct kvm *kvm);
891 void kvm_arch_end_assignment(struct kvm *kvm);
892 bool kvm_arch_has_assigned_device(struct kvm *kvm);
893 #else
894 static inline void kvm_arch_start_assignment(struct kvm *kvm)
895 {
896 }
897 
898 static inline void kvm_arch_end_assignment(struct kvm *kvm)
899 {
900 }
901 
902 static inline bool kvm_arch_has_assigned_device(struct kvm *kvm)
903 {
904 	return false;
905 }
906 #endif
907 
908 static inline struct swait_queue_head *kvm_arch_vcpu_wq(struct kvm_vcpu *vcpu)
909 {
910 #ifdef __KVM_HAVE_ARCH_WQP
911 	return vcpu->arch.wqp;
912 #else
913 	return &vcpu->wq;
914 #endif
915 }
916 
917 #ifdef __KVM_HAVE_ARCH_INTC_INITIALIZED
918 /*
919  * returns true if the virtual interrupt controller is initialized and
920  * ready to accept virtual IRQ. On some architectures the virtual interrupt
921  * controller is dynamically instantiated and this is not always true.
922  */
923 bool kvm_arch_intc_initialized(struct kvm *kvm);
924 #else
925 static inline bool kvm_arch_intc_initialized(struct kvm *kvm)
926 {
927 	return true;
928 }
929 #endif
930 
931 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type);
932 void kvm_arch_destroy_vm(struct kvm *kvm);
933 void kvm_arch_sync_events(struct kvm *kvm);
934 
935 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu);
936 void kvm_vcpu_kick(struct kvm_vcpu *vcpu);
937 
938 bool kvm_is_reserved_pfn(kvm_pfn_t pfn);
939 
940 struct kvm_irq_ack_notifier {
941 	struct hlist_node link;
942 	unsigned gsi;
943 	void (*irq_acked)(struct kvm_irq_ack_notifier *kian);
944 };
945 
946 int kvm_irq_map_gsi(struct kvm *kvm,
947 		    struct kvm_kernel_irq_routing_entry *entries, int gsi);
948 int kvm_irq_map_chip_pin(struct kvm *kvm, unsigned irqchip, unsigned pin);
949 
950 int kvm_set_irq(struct kvm *kvm, int irq_source_id, u32 irq, int level,
951 		bool line_status);
952 int kvm_set_msi(struct kvm_kernel_irq_routing_entry *irq_entry, struct kvm *kvm,
953 		int irq_source_id, int level, bool line_status);
954 int kvm_arch_set_irq_inatomic(struct kvm_kernel_irq_routing_entry *e,
955 			       struct kvm *kvm, int irq_source_id,
956 			       int level, bool line_status);
957 bool kvm_irq_has_notifier(struct kvm *kvm, unsigned irqchip, unsigned pin);
958 void kvm_notify_acked_gsi(struct kvm *kvm, int gsi);
959 void kvm_notify_acked_irq(struct kvm *kvm, unsigned irqchip, unsigned pin);
960 void kvm_register_irq_ack_notifier(struct kvm *kvm,
961 				   struct kvm_irq_ack_notifier *kian);
962 void kvm_unregister_irq_ack_notifier(struct kvm *kvm,
963 				   struct kvm_irq_ack_notifier *kian);
964 int kvm_request_irq_source_id(struct kvm *kvm);
965 void kvm_free_irq_source_id(struct kvm *kvm, int irq_source_id);
966 
967 /*
968  * search_memslots() and __gfn_to_memslot() are here because they are
969  * used in non-modular code in arch/powerpc/kvm/book3s_hv_rm_mmu.c.
970  * gfn_to_memslot() itself isn't here as an inline because that would
971  * bloat other code too much.
972  */
973 static inline struct kvm_memory_slot *
974 search_memslots(struct kvm_memslots *slots, gfn_t gfn)
975 {
976 	int start = 0, end = slots->used_slots;
977 	int slot = atomic_read(&slots->lru_slot);
978 	struct kvm_memory_slot *memslots = slots->memslots;
979 
980 	if (gfn >= memslots[slot].base_gfn &&
981 	    gfn < memslots[slot].base_gfn + memslots[slot].npages)
982 		return &memslots[slot];
983 
984 	while (start < end) {
985 		slot = start + (end - start) / 2;
986 
987 		if (gfn >= memslots[slot].base_gfn)
988 			end = slot;
989 		else
990 			start = slot + 1;
991 	}
992 
993 	if (gfn >= memslots[start].base_gfn &&
994 	    gfn < memslots[start].base_gfn + memslots[start].npages) {
995 		atomic_set(&slots->lru_slot, start);
996 		return &memslots[start];
997 	}
998 
999 	return NULL;
1000 }
1001 
1002 static inline struct kvm_memory_slot *
1003 __gfn_to_memslot(struct kvm_memslots *slots, gfn_t gfn)
1004 {
1005 	return search_memslots(slots, gfn);
1006 }
1007 
1008 static inline unsigned long
1009 __gfn_to_hva_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1010 {
1011 	return slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE;
1012 }
1013 
1014 static inline int memslot_id(struct kvm *kvm, gfn_t gfn)
1015 {
1016 	return gfn_to_memslot(kvm, gfn)->id;
1017 }
1018 
1019 static inline gfn_t
1020 hva_to_gfn_memslot(unsigned long hva, struct kvm_memory_slot *slot)
1021 {
1022 	gfn_t gfn_offset = (hva - slot->userspace_addr) >> PAGE_SHIFT;
1023 
1024 	return slot->base_gfn + gfn_offset;
1025 }
1026 
1027 static inline gpa_t gfn_to_gpa(gfn_t gfn)
1028 {
1029 	return (gpa_t)gfn << PAGE_SHIFT;
1030 }
1031 
1032 static inline gfn_t gpa_to_gfn(gpa_t gpa)
1033 {
1034 	return (gfn_t)(gpa >> PAGE_SHIFT);
1035 }
1036 
1037 static inline hpa_t pfn_to_hpa(kvm_pfn_t pfn)
1038 {
1039 	return (hpa_t)pfn << PAGE_SHIFT;
1040 }
1041 
1042 static inline struct page *kvm_vcpu_gpa_to_page(struct kvm_vcpu *vcpu,
1043 						gpa_t gpa)
1044 {
1045 	return kvm_vcpu_gfn_to_page(vcpu, gpa_to_gfn(gpa));
1046 }
1047 
1048 static inline bool kvm_is_error_gpa(struct kvm *kvm, gpa_t gpa)
1049 {
1050 	unsigned long hva = gfn_to_hva(kvm, gpa_to_gfn(gpa));
1051 
1052 	return kvm_is_error_hva(hva);
1053 }
1054 
1055 enum kvm_stat_kind {
1056 	KVM_STAT_VM,
1057 	KVM_STAT_VCPU,
1058 };
1059 
1060 struct kvm_stat_data {
1061 	int offset;
1062 	struct kvm *kvm;
1063 };
1064 
1065 struct kvm_stats_debugfs_item {
1066 	const char *name;
1067 	int offset;
1068 	enum kvm_stat_kind kind;
1069 };
1070 extern struct kvm_stats_debugfs_item debugfs_entries[];
1071 extern struct dentry *kvm_debugfs_dir;
1072 
1073 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1074 static inline int mmu_notifier_retry(struct kvm *kvm, unsigned long mmu_seq)
1075 {
1076 	if (unlikely(kvm->mmu_notifier_count))
1077 		return 1;
1078 	/*
1079 	 * Ensure the read of mmu_notifier_count happens before the read
1080 	 * of mmu_notifier_seq.  This interacts with the smp_wmb() in
1081 	 * mmu_notifier_invalidate_range_end to make sure that the caller
1082 	 * either sees the old (non-zero) value of mmu_notifier_count or
1083 	 * the new (incremented) value of mmu_notifier_seq.
1084 	 * PowerPC Book3s HV KVM calls this under a per-page lock
1085 	 * rather than under kvm->mmu_lock, for scalability, so
1086 	 * can't rely on kvm->mmu_lock to keep things ordered.
1087 	 */
1088 	smp_rmb();
1089 	if (kvm->mmu_notifier_seq != mmu_seq)
1090 		return 1;
1091 	return 0;
1092 }
1093 #endif
1094 
1095 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
1096 
1097 #define KVM_MAX_IRQ_ROUTES 4096 /* might need extension/rework in the future */
1098 
1099 bool kvm_arch_can_set_irq_routing(struct kvm *kvm);
1100 int kvm_set_irq_routing(struct kvm *kvm,
1101 			const struct kvm_irq_routing_entry *entries,
1102 			unsigned nr,
1103 			unsigned flags);
1104 int kvm_set_routing_entry(struct kvm *kvm,
1105 			  struct kvm_kernel_irq_routing_entry *e,
1106 			  const struct kvm_irq_routing_entry *ue);
1107 void kvm_free_irq_routing(struct kvm *kvm);
1108 
1109 #else
1110 
1111 static inline void kvm_free_irq_routing(struct kvm *kvm) {}
1112 
1113 #endif
1114 
1115 int kvm_send_userspace_msi(struct kvm *kvm, struct kvm_msi *msi);
1116 
1117 #ifdef CONFIG_HAVE_KVM_EVENTFD
1118 
1119 void kvm_eventfd_init(struct kvm *kvm);
1120 int kvm_ioeventfd(struct kvm *kvm, struct kvm_ioeventfd *args);
1121 
1122 #ifdef CONFIG_HAVE_KVM_IRQFD
1123 int kvm_irqfd(struct kvm *kvm, struct kvm_irqfd *args);
1124 void kvm_irqfd_release(struct kvm *kvm);
1125 void kvm_irq_routing_update(struct kvm *);
1126 #else
1127 static inline int kvm_irqfd(struct kvm *kvm, struct kvm_irqfd *args)
1128 {
1129 	return -EINVAL;
1130 }
1131 
1132 static inline void kvm_irqfd_release(struct kvm *kvm) {}
1133 #endif
1134 
1135 #else
1136 
1137 static inline void kvm_eventfd_init(struct kvm *kvm) {}
1138 
1139 static inline int kvm_irqfd(struct kvm *kvm, struct kvm_irqfd *args)
1140 {
1141 	return -EINVAL;
1142 }
1143 
1144 static inline void kvm_irqfd_release(struct kvm *kvm) {}
1145 
1146 #ifdef CONFIG_HAVE_KVM_IRQCHIP
1147 static inline void kvm_irq_routing_update(struct kvm *kvm)
1148 {
1149 }
1150 #endif
1151 
1152 static inline int kvm_ioeventfd(struct kvm *kvm, struct kvm_ioeventfd *args)
1153 {
1154 	return -ENOSYS;
1155 }
1156 
1157 #endif /* CONFIG_HAVE_KVM_EVENTFD */
1158 
1159 void kvm_arch_irq_routing_update(struct kvm *kvm);
1160 
1161 static inline void kvm_make_request(int req, struct kvm_vcpu *vcpu)
1162 {
1163 	/*
1164 	 * Ensure the rest of the request is published to kvm_check_request's
1165 	 * caller.  Paired with the smp_mb__after_atomic in kvm_check_request.
1166 	 */
1167 	smp_wmb();
1168 	set_bit(req & KVM_REQUEST_MASK, (void *)&vcpu->requests);
1169 }
1170 
1171 static inline bool kvm_request_pending(struct kvm_vcpu *vcpu)
1172 {
1173 	return READ_ONCE(vcpu->requests);
1174 }
1175 
1176 static inline bool kvm_test_request(int req, struct kvm_vcpu *vcpu)
1177 {
1178 	return test_bit(req & KVM_REQUEST_MASK, (void *)&vcpu->requests);
1179 }
1180 
1181 static inline void kvm_clear_request(int req, struct kvm_vcpu *vcpu)
1182 {
1183 	clear_bit(req & KVM_REQUEST_MASK, (void *)&vcpu->requests);
1184 }
1185 
1186 static inline bool kvm_check_request(int req, struct kvm_vcpu *vcpu)
1187 {
1188 	if (kvm_test_request(req, vcpu)) {
1189 		kvm_clear_request(req, vcpu);
1190 
1191 		/*
1192 		 * Ensure the rest of the request is visible to kvm_check_request's
1193 		 * caller.  Paired with the smp_wmb in kvm_make_request.
1194 		 */
1195 		smp_mb__after_atomic();
1196 		return true;
1197 	} else {
1198 		return false;
1199 	}
1200 }
1201 
1202 extern bool kvm_rebooting;
1203 
1204 extern unsigned int halt_poll_ns;
1205 extern unsigned int halt_poll_ns_grow;
1206 extern unsigned int halt_poll_ns_grow_start;
1207 extern unsigned int halt_poll_ns_shrink;
1208 
1209 struct kvm_device {
1210 	struct kvm_device_ops *ops;
1211 	struct kvm *kvm;
1212 	void *private;
1213 	struct list_head vm_node;
1214 };
1215 
1216 /* create, destroy, and name are mandatory */
1217 struct kvm_device_ops {
1218 	const char *name;
1219 
1220 	/*
1221 	 * create is called holding kvm->lock and any operations not suitable
1222 	 * to do while holding the lock should be deferred to init (see
1223 	 * below).
1224 	 */
1225 	int (*create)(struct kvm_device *dev, u32 type);
1226 
1227 	/*
1228 	 * init is called after create if create is successful and is called
1229 	 * outside of holding kvm->lock.
1230 	 */
1231 	void (*init)(struct kvm_device *dev);
1232 
1233 	/*
1234 	 * Destroy is responsible for freeing dev.
1235 	 *
1236 	 * Destroy may be called before or after destructors are called
1237 	 * on emulated I/O regions, depending on whether a reference is
1238 	 * held by a vcpu or other kvm component that gets destroyed
1239 	 * after the emulated I/O.
1240 	 */
1241 	void (*destroy)(struct kvm_device *dev);
1242 
1243 	int (*set_attr)(struct kvm_device *dev, struct kvm_device_attr *attr);
1244 	int (*get_attr)(struct kvm_device *dev, struct kvm_device_attr *attr);
1245 	int (*has_attr)(struct kvm_device *dev, struct kvm_device_attr *attr);
1246 	long (*ioctl)(struct kvm_device *dev, unsigned int ioctl,
1247 		      unsigned long arg);
1248 };
1249 
1250 void kvm_device_get(struct kvm_device *dev);
1251 void kvm_device_put(struct kvm_device *dev);
1252 struct kvm_device *kvm_device_from_filp(struct file *filp);
1253 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type);
1254 void kvm_unregister_device_ops(u32 type);
1255 
1256 extern struct kvm_device_ops kvm_mpic_ops;
1257 extern struct kvm_device_ops kvm_arm_vgic_v2_ops;
1258 extern struct kvm_device_ops kvm_arm_vgic_v3_ops;
1259 
1260 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1261 
1262 static inline void kvm_vcpu_set_in_spin_loop(struct kvm_vcpu *vcpu, bool val)
1263 {
1264 	vcpu->spin_loop.in_spin_loop = val;
1265 }
1266 static inline void kvm_vcpu_set_dy_eligible(struct kvm_vcpu *vcpu, bool val)
1267 {
1268 	vcpu->spin_loop.dy_eligible = val;
1269 }
1270 
1271 #else /* !CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT */
1272 
1273 static inline void kvm_vcpu_set_in_spin_loop(struct kvm_vcpu *vcpu, bool val)
1274 {
1275 }
1276 
1277 static inline void kvm_vcpu_set_dy_eligible(struct kvm_vcpu *vcpu, bool val)
1278 {
1279 }
1280 #endif /* CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT */
1281 
1282 #ifdef CONFIG_HAVE_KVM_IRQ_BYPASS
1283 bool kvm_arch_has_irq_bypass(void);
1284 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *,
1285 			   struct irq_bypass_producer *);
1286 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *,
1287 			   struct irq_bypass_producer *);
1288 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *);
1289 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *);
1290 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
1291 				  uint32_t guest_irq, bool set);
1292 #endif /* CONFIG_HAVE_KVM_IRQ_BYPASS */
1293 
1294 #ifdef CONFIG_HAVE_KVM_INVALID_WAKEUPS
1295 /* If we wakeup during the poll time, was it a sucessful poll? */
1296 static inline bool vcpu_valid_wakeup(struct kvm_vcpu *vcpu)
1297 {
1298 	return vcpu->valid_wakeup;
1299 }
1300 
1301 #else
1302 static inline bool vcpu_valid_wakeup(struct kvm_vcpu *vcpu)
1303 {
1304 	return true;
1305 }
1306 #endif /* CONFIG_HAVE_KVM_INVALID_WAKEUPS */
1307 
1308 #ifdef CONFIG_HAVE_KVM_VCPU_ASYNC_IOCTL
1309 long kvm_arch_vcpu_async_ioctl(struct file *filp,
1310 			       unsigned int ioctl, unsigned long arg);
1311 #else
1312 static inline long kvm_arch_vcpu_async_ioctl(struct file *filp,
1313 					     unsigned int ioctl,
1314 					     unsigned long arg)
1315 {
1316 	return -ENOIOCTLCMD;
1317 }
1318 #endif /* CONFIG_HAVE_KVM_VCPU_ASYNC_IOCTL */
1319 
1320 int kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
1321 		unsigned long start, unsigned long end, bool blockable);
1322 
1323 #ifdef CONFIG_HAVE_KVM_VCPU_RUN_PID_CHANGE
1324 int kvm_arch_vcpu_run_pid_change(struct kvm_vcpu *vcpu);
1325 #else
1326 static inline int kvm_arch_vcpu_run_pid_change(struct kvm_vcpu *vcpu)
1327 {
1328 	return 0;
1329 }
1330 #endif /* CONFIG_HAVE_KVM_VCPU_RUN_PID_CHANGE */
1331 
1332 #endif
1333