xref: /openbmc/linux/include/linux/kvm_host.h (revision 17bc4815de586d001c82d0ddf75247283c3f002a)
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 /* Two fragments for cross MMIO pages. */
52 #define KVM_MAX_MMIO_FRAGMENTS	2
53 
54 #ifndef KVM_ADDRESS_SPACE_NUM
55 #define KVM_ADDRESS_SPACE_NUM	1
56 #endif
57 
58 /*
59  * For the normal pfn, the highest 12 bits should be zero,
60  * so we can mask bit 62 ~ bit 52  to indicate the error pfn,
61  * mask bit 63 to indicate the noslot pfn.
62  */
63 #define KVM_PFN_ERR_MASK	(0x7ffULL << 52)
64 #define KVM_PFN_ERR_NOSLOT_MASK	(0xfffULL << 52)
65 #define KVM_PFN_NOSLOT		(0x1ULL << 63)
66 
67 #define KVM_PFN_ERR_FAULT	(KVM_PFN_ERR_MASK)
68 #define KVM_PFN_ERR_HWPOISON	(KVM_PFN_ERR_MASK + 1)
69 #define KVM_PFN_ERR_RO_FAULT	(KVM_PFN_ERR_MASK + 2)
70 
71 /*
72  * error pfns indicate that the gfn is in slot but faild to
73  * translate it to pfn on host.
74  */
75 static inline bool is_error_pfn(kvm_pfn_t pfn)
76 {
77 	return !!(pfn & KVM_PFN_ERR_MASK);
78 }
79 
80 /*
81  * error_noslot pfns indicate that the gfn can not be
82  * translated to pfn - it is not in slot or failed to
83  * translate it to pfn.
84  */
85 static inline bool is_error_noslot_pfn(kvm_pfn_t pfn)
86 {
87 	return !!(pfn & KVM_PFN_ERR_NOSLOT_MASK);
88 }
89 
90 /* noslot pfn indicates that the gfn is not in slot. */
91 static inline bool is_noslot_pfn(kvm_pfn_t pfn)
92 {
93 	return pfn == KVM_PFN_NOSLOT;
94 }
95 
96 /*
97  * architectures with KVM_HVA_ERR_BAD other than PAGE_OFFSET (e.g. s390)
98  * provide own defines and kvm_is_error_hva
99  */
100 #ifndef KVM_HVA_ERR_BAD
101 
102 #define KVM_HVA_ERR_BAD		(PAGE_OFFSET)
103 #define KVM_HVA_ERR_RO_BAD	(PAGE_OFFSET + PAGE_SIZE)
104 
105 static inline bool kvm_is_error_hva(unsigned long addr)
106 {
107 	return addr >= PAGE_OFFSET;
108 }
109 
110 #endif
111 
112 #define KVM_ERR_PTR_BAD_PAGE	(ERR_PTR(-ENOENT))
113 
114 static inline bool is_error_page(struct page *page)
115 {
116 	return IS_ERR(page);
117 }
118 
119 #define KVM_REQUEST_MASK           GENMASK(7,0)
120 #define KVM_REQUEST_NO_WAKEUP      BIT(8)
121 #define KVM_REQUEST_WAIT           BIT(9)
122 /*
123  * Architecture-independent vcpu->requests bit members
124  * Bits 4-7 are reserved for more arch-independent bits.
125  */
126 #define KVM_REQ_TLB_FLUSH         (0 | KVM_REQUEST_WAIT | KVM_REQUEST_NO_WAKEUP)
127 #define KVM_REQ_MMU_RELOAD        (1 | KVM_REQUEST_WAIT | KVM_REQUEST_NO_WAKEUP)
128 #define KVM_REQ_PENDING_TIMER     2
129 #define KVM_REQ_UNHALT            3
130 #define KVM_REQUEST_ARCH_BASE     8
131 
132 #define KVM_ARCH_REQ_FLAGS(nr, flags) ({ \
133 	BUILD_BUG_ON((unsigned)(nr) >= (FIELD_SIZEOF(struct kvm_vcpu, requests) * 8) - KVM_REQUEST_ARCH_BASE); \
134 	(unsigned)(((nr) + KVM_REQUEST_ARCH_BASE) | (flags)); \
135 })
136 #define KVM_ARCH_REQ(nr)           KVM_ARCH_REQ_FLAGS(nr, 0)
137 
138 #define KVM_USERSPACE_IRQ_SOURCE_ID		0
139 #define KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID	1
140 
141 extern struct kmem_cache *kvm_vcpu_cache;
142 
143 extern spinlock_t kvm_lock;
144 extern struct list_head vm_list;
145 
146 struct kvm_io_range {
147 	gpa_t addr;
148 	int len;
149 	struct kvm_io_device *dev;
150 };
151 
152 #define NR_IOBUS_DEVS 1000
153 
154 struct kvm_io_bus {
155 	int dev_count;
156 	int ioeventfd_count;
157 	struct kvm_io_range range[];
158 };
159 
160 enum kvm_bus {
161 	KVM_MMIO_BUS,
162 	KVM_PIO_BUS,
163 	KVM_VIRTIO_CCW_NOTIFY_BUS,
164 	KVM_FAST_MMIO_BUS,
165 	KVM_NR_BUSES
166 };
167 
168 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
169 		     int len, const void *val);
170 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
171 			    gpa_t addr, int len, const void *val, long cookie);
172 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
173 		    int len, void *val);
174 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
175 			    int len, struct kvm_io_device *dev);
176 void kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
177 			       struct kvm_io_device *dev);
178 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
179 					 gpa_t addr);
180 
181 #ifdef CONFIG_KVM_ASYNC_PF
182 struct kvm_async_pf {
183 	struct work_struct work;
184 	struct list_head link;
185 	struct list_head queue;
186 	struct kvm_vcpu *vcpu;
187 	struct mm_struct *mm;
188 	gva_t gva;
189 	unsigned long addr;
190 	struct kvm_arch_async_pf arch;
191 	bool   wakeup_all;
192 };
193 
194 void kvm_clear_async_pf_completion_queue(struct kvm_vcpu *vcpu);
195 void kvm_check_async_pf_completion(struct kvm_vcpu *vcpu);
196 int kvm_setup_async_pf(struct kvm_vcpu *vcpu, gva_t gva, unsigned long hva,
197 		       struct kvm_arch_async_pf *arch);
198 int kvm_async_pf_wakeup_all(struct kvm_vcpu *vcpu);
199 #endif
200 
201 enum {
202 	OUTSIDE_GUEST_MODE,
203 	IN_GUEST_MODE,
204 	EXITING_GUEST_MODE,
205 	READING_SHADOW_PAGE_TABLES,
206 };
207 
208 /*
209  * Sometimes a large or cross-page mmio needs to be broken up into separate
210  * exits for userspace servicing.
211  */
212 struct kvm_mmio_fragment {
213 	gpa_t gpa;
214 	void *data;
215 	unsigned len;
216 };
217 
218 struct kvm_vcpu {
219 	struct kvm *kvm;
220 #ifdef CONFIG_PREEMPT_NOTIFIERS
221 	struct preempt_notifier preempt_notifier;
222 #endif
223 	int cpu;
224 	int vcpu_id;
225 	int srcu_idx;
226 	int mode;
227 	u64 requests;
228 	unsigned long guest_debug;
229 
230 	int pre_pcpu;
231 	struct list_head blocked_vcpu_list;
232 
233 	struct mutex mutex;
234 	struct kvm_run *run;
235 
236 	int guest_xcr0_loaded;
237 	struct swait_queue_head wq;
238 	struct pid __rcu *pid;
239 	int sigset_active;
240 	sigset_t sigset;
241 	struct kvm_vcpu_stat stat;
242 	unsigned int halt_poll_ns;
243 	bool valid_wakeup;
244 
245 #ifdef CONFIG_HAS_IOMEM
246 	int mmio_needed;
247 	int mmio_read_completed;
248 	int mmio_is_write;
249 	int mmio_cur_fragment;
250 	int mmio_nr_fragments;
251 	struct kvm_mmio_fragment mmio_fragments[KVM_MAX_MMIO_FRAGMENTS];
252 #endif
253 
254 #ifdef CONFIG_KVM_ASYNC_PF
255 	struct {
256 		u32 queued;
257 		struct list_head queue;
258 		struct list_head done;
259 		spinlock_t lock;
260 	} async_pf;
261 #endif
262 
263 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
264 	/*
265 	 * Cpu relax intercept or pause loop exit optimization
266 	 * in_spin_loop: set when a vcpu does a pause loop exit
267 	 *  or cpu relax intercepted.
268 	 * dy_eligible: indicates whether vcpu is eligible for directed yield.
269 	 */
270 	struct {
271 		bool in_spin_loop;
272 		bool dy_eligible;
273 	} spin_loop;
274 #endif
275 	bool preempted;
276 	struct kvm_vcpu_arch arch;
277 	struct dentry *debugfs_dentry;
278 };
279 
280 static inline int kvm_vcpu_exiting_guest_mode(struct kvm_vcpu *vcpu)
281 {
282 	/*
283 	 * The memory barrier ensures a previous write to vcpu->requests cannot
284 	 * be reordered with the read of vcpu->mode.  It pairs with the general
285 	 * memory barrier following the write of vcpu->mode in VCPU RUN.
286 	 */
287 	smp_mb__before_atomic();
288 	return cmpxchg(&vcpu->mode, IN_GUEST_MODE, EXITING_GUEST_MODE);
289 }
290 
291 /*
292  * Some of the bitops functions do not support too long bitmaps.
293  * This number must be determined not to exceed such limits.
294  */
295 #define KVM_MEM_MAX_NR_PAGES ((1UL << 31) - 1)
296 
297 struct kvm_memory_slot {
298 	gfn_t base_gfn;
299 	unsigned long npages;
300 	unsigned long *dirty_bitmap;
301 	struct kvm_arch_memory_slot arch;
302 	unsigned long userspace_addr;
303 	u32 flags;
304 	short id;
305 };
306 
307 static inline unsigned long kvm_dirty_bitmap_bytes(struct kvm_memory_slot *memslot)
308 {
309 	return ALIGN(memslot->npages, BITS_PER_LONG) / 8;
310 }
311 
312 static inline unsigned long *kvm_second_dirty_bitmap(struct kvm_memory_slot *memslot)
313 {
314 	unsigned long len = kvm_dirty_bitmap_bytes(memslot);
315 
316 	return memslot->dirty_bitmap + len / sizeof(*memslot->dirty_bitmap);
317 }
318 
319 struct kvm_s390_adapter_int {
320 	u64 ind_addr;
321 	u64 summary_addr;
322 	u64 ind_offset;
323 	u32 summary_offset;
324 	u32 adapter_id;
325 };
326 
327 struct kvm_hv_sint {
328 	u32 vcpu;
329 	u32 sint;
330 };
331 
332 struct kvm_kernel_irq_routing_entry {
333 	u32 gsi;
334 	u32 type;
335 	int (*set)(struct kvm_kernel_irq_routing_entry *e,
336 		   struct kvm *kvm, int irq_source_id, int level,
337 		   bool line_status);
338 	union {
339 		struct {
340 			unsigned irqchip;
341 			unsigned pin;
342 		} irqchip;
343 		struct {
344 			u32 address_lo;
345 			u32 address_hi;
346 			u32 data;
347 			u32 flags;
348 			u32 devid;
349 		} msi;
350 		struct kvm_s390_adapter_int adapter;
351 		struct kvm_hv_sint hv_sint;
352 	};
353 	struct hlist_node link;
354 };
355 
356 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
357 struct kvm_irq_routing_table {
358 	int chip[KVM_NR_IRQCHIPS][KVM_IRQCHIP_NUM_PINS];
359 	u32 nr_rt_entries;
360 	/*
361 	 * Array indexed by gsi. Each entry contains list of irq chips
362 	 * the gsi is connected to.
363 	 */
364 	struct hlist_head map[0];
365 };
366 #endif
367 
368 #ifndef KVM_PRIVATE_MEM_SLOTS
369 #define KVM_PRIVATE_MEM_SLOTS 0
370 #endif
371 
372 #ifndef KVM_MEM_SLOTS_NUM
373 #define KVM_MEM_SLOTS_NUM (KVM_USER_MEM_SLOTS + KVM_PRIVATE_MEM_SLOTS)
374 #endif
375 
376 #ifndef __KVM_VCPU_MULTIPLE_ADDRESS_SPACE
377 static inline int kvm_arch_vcpu_memslots_id(struct kvm_vcpu *vcpu)
378 {
379 	return 0;
380 }
381 #endif
382 
383 /*
384  * Note:
385  * memslots are not sorted by id anymore, please use id_to_memslot()
386  * to get the memslot by its id.
387  */
388 struct kvm_memslots {
389 	u64 generation;
390 	struct kvm_memory_slot memslots[KVM_MEM_SLOTS_NUM];
391 	/* The mapping table from slot id to the index in memslots[]. */
392 	short id_to_index[KVM_MEM_SLOTS_NUM];
393 	atomic_t lru_slot;
394 	int used_slots;
395 };
396 
397 struct kvm {
398 	spinlock_t mmu_lock;
399 	struct mutex slots_lock;
400 	struct mm_struct *mm; /* userspace tied to this vm */
401 	struct kvm_memslots __rcu *memslots[KVM_ADDRESS_SPACE_NUM];
402 	struct kvm_vcpu *vcpus[KVM_MAX_VCPUS];
403 
404 	/*
405 	 * created_vcpus is protected by kvm->lock, and is incremented
406 	 * at the beginning of KVM_CREATE_VCPU.  online_vcpus is only
407 	 * incremented after storing the kvm_vcpu pointer in vcpus,
408 	 * and is accessed atomically.
409 	 */
410 	atomic_t online_vcpus;
411 	int created_vcpus;
412 	int last_boosted_vcpu;
413 	struct list_head vm_list;
414 	struct mutex lock;
415 	struct kvm_io_bus __rcu *buses[KVM_NR_BUSES];
416 #ifdef CONFIG_HAVE_KVM_EVENTFD
417 	struct {
418 		spinlock_t        lock;
419 		struct list_head  items;
420 		struct list_head  resampler_list;
421 		struct mutex      resampler_lock;
422 	} irqfds;
423 	struct list_head ioeventfds;
424 #endif
425 	struct kvm_vm_stat stat;
426 	struct kvm_arch arch;
427 	refcount_t users_count;
428 #ifdef CONFIG_KVM_MMIO
429 	struct kvm_coalesced_mmio_ring *coalesced_mmio_ring;
430 	spinlock_t ring_lock;
431 	struct list_head coalesced_zones;
432 #endif
433 
434 	struct mutex irq_lock;
435 #ifdef CONFIG_HAVE_KVM_IRQCHIP
436 	/*
437 	 * Update side is protected by irq_lock.
438 	 */
439 	struct kvm_irq_routing_table __rcu *irq_routing;
440 #endif
441 #ifdef CONFIG_HAVE_KVM_IRQFD
442 	struct hlist_head irq_ack_notifier_list;
443 #endif
444 
445 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
446 	struct mmu_notifier mmu_notifier;
447 	unsigned long mmu_notifier_seq;
448 	long mmu_notifier_count;
449 #endif
450 	long tlbs_dirty;
451 	struct list_head devices;
452 	bool manual_dirty_log_protect;
453 	struct dentry *debugfs_dentry;
454 	struct kvm_stat_data **debugfs_stat_data;
455 	struct srcu_struct srcu;
456 	struct srcu_struct irq_srcu;
457 	pid_t userspace_pid;
458 };
459 
460 #define kvm_err(fmt, ...) \
461 	pr_err("kvm [%i]: " fmt, task_pid_nr(current), ## __VA_ARGS__)
462 #define kvm_info(fmt, ...) \
463 	pr_info("kvm [%i]: " fmt, task_pid_nr(current), ## __VA_ARGS__)
464 #define kvm_debug(fmt, ...) \
465 	pr_debug("kvm [%i]: " fmt, task_pid_nr(current), ## __VA_ARGS__)
466 #define kvm_debug_ratelimited(fmt, ...) \
467 	pr_debug_ratelimited("kvm [%i]: " fmt, task_pid_nr(current), \
468 			     ## __VA_ARGS__)
469 #define kvm_pr_unimpl(fmt, ...) \
470 	pr_err_ratelimited("kvm [%i]: " fmt, \
471 			   task_tgid_nr(current), ## __VA_ARGS__)
472 
473 /* The guest did something we don't support. */
474 #define vcpu_unimpl(vcpu, fmt, ...)					\
475 	kvm_pr_unimpl("vcpu%i, guest rIP: 0x%lx " fmt,			\
476 			(vcpu)->vcpu_id, kvm_rip_read(vcpu), ## __VA_ARGS__)
477 
478 #define vcpu_debug(vcpu, fmt, ...)					\
479 	kvm_debug("vcpu%i " fmt, (vcpu)->vcpu_id, ## __VA_ARGS__)
480 #define vcpu_debug_ratelimited(vcpu, fmt, ...)				\
481 	kvm_debug_ratelimited("vcpu%i " fmt, (vcpu)->vcpu_id,           \
482 			      ## __VA_ARGS__)
483 #define vcpu_err(vcpu, fmt, ...)					\
484 	kvm_err("vcpu%i " fmt, (vcpu)->vcpu_id, ## __VA_ARGS__)
485 
486 static inline struct kvm_io_bus *kvm_get_bus(struct kvm *kvm, enum kvm_bus idx)
487 {
488 	return srcu_dereference_check(kvm->buses[idx], &kvm->srcu,
489 				      lockdep_is_held(&kvm->slots_lock) ||
490 				      !refcount_read(&kvm->users_count));
491 }
492 
493 static inline struct kvm_vcpu *kvm_get_vcpu(struct kvm *kvm, int i)
494 {
495 	/* Pairs with smp_wmb() in kvm_vm_ioctl_create_vcpu, in case
496 	 * the caller has read kvm->online_vcpus before (as is the case
497 	 * for kvm_for_each_vcpu, for example).
498 	 */
499 	smp_rmb();
500 	return kvm->vcpus[i];
501 }
502 
503 #define kvm_for_each_vcpu(idx, vcpup, kvm) \
504 	for (idx = 0; \
505 	     idx < atomic_read(&kvm->online_vcpus) && \
506 	     (vcpup = kvm_get_vcpu(kvm, idx)) != NULL; \
507 	     idx++)
508 
509 static inline struct kvm_vcpu *kvm_get_vcpu_by_id(struct kvm *kvm, int id)
510 {
511 	struct kvm_vcpu *vcpu = NULL;
512 	int i;
513 
514 	if (id < 0)
515 		return NULL;
516 	if (id < KVM_MAX_VCPUS)
517 		vcpu = kvm_get_vcpu(kvm, id);
518 	if (vcpu && vcpu->vcpu_id == id)
519 		return vcpu;
520 	kvm_for_each_vcpu(i, vcpu, kvm)
521 		if (vcpu->vcpu_id == id)
522 			return vcpu;
523 	return NULL;
524 }
525 
526 static inline int kvm_vcpu_get_idx(struct kvm_vcpu *vcpu)
527 {
528 	struct kvm_vcpu *tmp;
529 	int idx;
530 
531 	kvm_for_each_vcpu(idx, tmp, vcpu->kvm)
532 		if (tmp == vcpu)
533 			return idx;
534 	BUG();
535 }
536 
537 #define kvm_for_each_memslot(memslot, slots)	\
538 	for (memslot = &slots->memslots[0];	\
539 	      memslot < slots->memslots + KVM_MEM_SLOTS_NUM && memslot->npages;\
540 		memslot++)
541 
542 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id);
543 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu);
544 
545 void vcpu_load(struct kvm_vcpu *vcpu);
546 void vcpu_put(struct kvm_vcpu *vcpu);
547 
548 #ifdef __KVM_HAVE_IOAPIC
549 void kvm_arch_post_irq_ack_notifier_list_update(struct kvm *kvm);
550 void kvm_arch_post_irq_routing_update(struct kvm *kvm);
551 #else
552 static inline void kvm_arch_post_irq_ack_notifier_list_update(struct kvm *kvm)
553 {
554 }
555 static inline void kvm_arch_post_irq_routing_update(struct kvm *kvm)
556 {
557 }
558 #endif
559 
560 #ifdef CONFIG_HAVE_KVM_IRQFD
561 int kvm_irqfd_init(void);
562 void kvm_irqfd_exit(void);
563 #else
564 static inline int kvm_irqfd_init(void)
565 {
566 	return 0;
567 }
568 
569 static inline void kvm_irqfd_exit(void)
570 {
571 }
572 #endif
573 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
574 		  struct module *module);
575 void kvm_exit(void);
576 
577 void kvm_get_kvm(struct kvm *kvm);
578 void kvm_put_kvm(struct kvm *kvm);
579 
580 static inline struct kvm_memslots *__kvm_memslots(struct kvm *kvm, int as_id)
581 {
582 	return srcu_dereference_check(kvm->memslots[as_id], &kvm->srcu,
583 			lockdep_is_held(&kvm->slots_lock) ||
584 			!refcount_read(&kvm->users_count));
585 }
586 
587 static inline struct kvm_memslots *kvm_memslots(struct kvm *kvm)
588 {
589 	return __kvm_memslots(kvm, 0);
590 }
591 
592 static inline struct kvm_memslots *kvm_vcpu_memslots(struct kvm_vcpu *vcpu)
593 {
594 	int as_id = kvm_arch_vcpu_memslots_id(vcpu);
595 
596 	return __kvm_memslots(vcpu->kvm, as_id);
597 }
598 
599 static inline struct kvm_memory_slot *
600 id_to_memslot(struct kvm_memslots *slots, int id)
601 {
602 	int index = slots->id_to_index[id];
603 	struct kvm_memory_slot *slot;
604 
605 	slot = &slots->memslots[index];
606 
607 	WARN_ON(slot->id != id);
608 	return slot;
609 }
610 
611 /*
612  * KVM_SET_USER_MEMORY_REGION ioctl allows the following operations:
613  * - create a new memory slot
614  * - delete an existing memory slot
615  * - modify an existing memory slot
616  *   -- move it in the guest physical memory space
617  *   -- just change its flags
618  *
619  * Since flags can be changed by some of these operations, the following
620  * differentiation is the best we can do for __kvm_set_memory_region():
621  */
622 enum kvm_mr_change {
623 	KVM_MR_CREATE,
624 	KVM_MR_DELETE,
625 	KVM_MR_MOVE,
626 	KVM_MR_FLAGS_ONLY,
627 };
628 
629 int kvm_set_memory_region(struct kvm *kvm,
630 			  const struct kvm_userspace_memory_region *mem);
631 int __kvm_set_memory_region(struct kvm *kvm,
632 			    const struct kvm_userspace_memory_region *mem);
633 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
634 			   struct kvm_memory_slot *dont);
635 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
636 			    unsigned long npages);
637 void kvm_arch_memslots_updated(struct kvm *kvm, struct kvm_memslots *slots);
638 int kvm_arch_prepare_memory_region(struct kvm *kvm,
639 				struct kvm_memory_slot *memslot,
640 				const struct kvm_userspace_memory_region *mem,
641 				enum kvm_mr_change change);
642 void kvm_arch_commit_memory_region(struct kvm *kvm,
643 				const struct kvm_userspace_memory_region *mem,
644 				const struct kvm_memory_slot *old,
645 				const struct kvm_memory_slot *new,
646 				enum kvm_mr_change change);
647 bool kvm_largepages_enabled(void);
648 void kvm_disable_largepages(void);
649 /* flush all memory translations */
650 void kvm_arch_flush_shadow_all(struct kvm *kvm);
651 /* flush memory translations pointing to 'slot' */
652 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
653 				   struct kvm_memory_slot *slot);
654 
655 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
656 			    struct page **pages, int nr_pages);
657 
658 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn);
659 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn);
660 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable);
661 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot, gfn_t gfn);
662 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot, gfn_t gfn,
663 				      bool *writable);
664 void kvm_release_page_clean(struct page *page);
665 void kvm_release_page_dirty(struct page *page);
666 void kvm_set_page_accessed(struct page *page);
667 
668 kvm_pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn);
669 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn);
670 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
671 		      bool *writable);
672 kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn);
673 kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn);
674 kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
675 			       bool atomic, bool *async, bool write_fault,
676 			       bool *writable);
677 
678 void kvm_release_pfn_clean(kvm_pfn_t pfn);
679 void kvm_release_pfn_dirty(kvm_pfn_t pfn);
680 void kvm_set_pfn_dirty(kvm_pfn_t pfn);
681 void kvm_set_pfn_accessed(kvm_pfn_t pfn);
682 void kvm_get_pfn(kvm_pfn_t pfn);
683 
684 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
685 			int len);
686 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
687 			  unsigned long len);
688 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len);
689 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
690 			   void *data, unsigned long len);
691 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
692 			 int offset, int len);
693 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
694 		    unsigned long len);
695 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
696 			   void *data, unsigned long len);
697 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
698 				  void *data, unsigned int offset,
699 				  unsigned long len);
700 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
701 			      gpa_t gpa, unsigned long len);
702 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len);
703 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len);
704 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn);
705 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn);
706 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn);
707 void mark_page_dirty(struct kvm *kvm, gfn_t gfn);
708 
709 struct kvm_memslots *kvm_vcpu_memslots(struct kvm_vcpu *vcpu);
710 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn);
711 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn);
712 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn);
713 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn);
714 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn);
715 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable);
716 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data, int offset,
717 			     int len);
718 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa, void *data,
719 			       unsigned long len);
720 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data,
721 			unsigned long len);
722 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, const void *data,
723 			      int offset, int len);
724 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
725 			 unsigned long len);
726 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn);
727 
728 void kvm_sigset_activate(struct kvm_vcpu *vcpu);
729 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu);
730 
731 void kvm_vcpu_block(struct kvm_vcpu *vcpu);
732 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu);
733 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu);
734 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu);
735 void kvm_vcpu_kick(struct kvm_vcpu *vcpu);
736 int kvm_vcpu_yield_to(struct kvm_vcpu *target);
737 void kvm_vcpu_on_spin(struct kvm_vcpu *vcpu, bool usermode_vcpu_not_eligible);
738 
739 void kvm_flush_remote_tlbs(struct kvm *kvm);
740 void kvm_reload_remote_mmus(struct kvm *kvm);
741 
742 bool kvm_make_vcpus_request_mask(struct kvm *kvm, unsigned int req,
743 				 unsigned long *vcpu_bitmap, cpumask_var_t tmp);
744 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req);
745 
746 long kvm_arch_dev_ioctl(struct file *filp,
747 			unsigned int ioctl, unsigned long arg);
748 long kvm_arch_vcpu_ioctl(struct file *filp,
749 			 unsigned int ioctl, unsigned long arg);
750 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf);
751 
752 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext);
753 
754 int kvm_get_dirty_log(struct kvm *kvm,
755 			struct kvm_dirty_log *log, int *is_dirty);
756 
757 int kvm_get_dirty_log_protect(struct kvm *kvm,
758 			      struct kvm_dirty_log *log, bool *flush);
759 int kvm_clear_dirty_log_protect(struct kvm *kvm,
760 				struct kvm_clear_dirty_log *log, bool *flush);
761 
762 void kvm_arch_mmu_enable_log_dirty_pt_masked(struct kvm *kvm,
763 					struct kvm_memory_slot *slot,
764 					gfn_t gfn_offset,
765 					unsigned long mask);
766 
767 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
768 				struct kvm_dirty_log *log);
769 int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm,
770 				  struct kvm_clear_dirty_log *log);
771 
772 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
773 			bool line_status);
774 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
775 			    struct kvm_enable_cap *cap);
776 long kvm_arch_vm_ioctl(struct file *filp,
777 		       unsigned int ioctl, unsigned long arg);
778 
779 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu);
780 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu);
781 
782 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
783 				    struct kvm_translation *tr);
784 
785 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs);
786 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs);
787 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
788 				  struct kvm_sregs *sregs);
789 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
790 				  struct kvm_sregs *sregs);
791 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
792 				    struct kvm_mp_state *mp_state);
793 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
794 				    struct kvm_mp_state *mp_state);
795 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
796 					struct kvm_guest_debug *dbg);
797 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run);
798 
799 int kvm_arch_init(void *opaque);
800 void kvm_arch_exit(void);
801 
802 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu);
803 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu);
804 
805 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu);
806 
807 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu);
808 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu);
809 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu);
810 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id);
811 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu);
812 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu);
813 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu);
814 
815 bool kvm_arch_has_vcpu_debugfs(void);
816 int kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu);
817 
818 int kvm_arch_hardware_enable(void);
819 void kvm_arch_hardware_disable(void);
820 int kvm_arch_hardware_setup(void);
821 void kvm_arch_hardware_unsetup(void);
822 void kvm_arch_check_processor_compat(void *rtn);
823 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu);
824 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu);
825 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu);
826 
827 #ifndef __KVM_HAVE_ARCH_VM_ALLOC
828 /*
829  * All architectures that want to use vzalloc currently also
830  * need their own kvm_arch_alloc_vm implementation.
831  */
832 static inline struct kvm *kvm_arch_alloc_vm(void)
833 {
834 	return kzalloc(sizeof(struct kvm), GFP_KERNEL);
835 }
836 
837 static inline void kvm_arch_free_vm(struct kvm *kvm)
838 {
839 	kfree(kvm);
840 }
841 #endif
842 
843 #ifndef __KVM_HAVE_ARCH_FLUSH_REMOTE_TLB
844 static inline int kvm_arch_flush_remote_tlb(struct kvm *kvm)
845 {
846 	return -ENOTSUPP;
847 }
848 #endif
849 
850 #ifdef __KVM_HAVE_ARCH_NONCOHERENT_DMA
851 void kvm_arch_register_noncoherent_dma(struct kvm *kvm);
852 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm);
853 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm);
854 #else
855 static inline void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
856 {
857 }
858 
859 static inline void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
860 {
861 }
862 
863 static inline bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
864 {
865 	return false;
866 }
867 #endif
868 #ifdef __KVM_HAVE_ARCH_ASSIGNED_DEVICE
869 void kvm_arch_start_assignment(struct kvm *kvm);
870 void kvm_arch_end_assignment(struct kvm *kvm);
871 bool kvm_arch_has_assigned_device(struct kvm *kvm);
872 #else
873 static inline void kvm_arch_start_assignment(struct kvm *kvm)
874 {
875 }
876 
877 static inline void kvm_arch_end_assignment(struct kvm *kvm)
878 {
879 }
880 
881 static inline bool kvm_arch_has_assigned_device(struct kvm *kvm)
882 {
883 	return false;
884 }
885 #endif
886 
887 static inline struct swait_queue_head *kvm_arch_vcpu_wq(struct kvm_vcpu *vcpu)
888 {
889 #ifdef __KVM_HAVE_ARCH_WQP
890 	return vcpu->arch.wqp;
891 #else
892 	return &vcpu->wq;
893 #endif
894 }
895 
896 #ifdef __KVM_HAVE_ARCH_INTC_INITIALIZED
897 /*
898  * returns true if the virtual interrupt controller is initialized and
899  * ready to accept virtual IRQ. On some architectures the virtual interrupt
900  * controller is dynamically instantiated and this is not always true.
901  */
902 bool kvm_arch_intc_initialized(struct kvm *kvm);
903 #else
904 static inline bool kvm_arch_intc_initialized(struct kvm *kvm)
905 {
906 	return true;
907 }
908 #endif
909 
910 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type);
911 void kvm_arch_destroy_vm(struct kvm *kvm);
912 void kvm_arch_sync_events(struct kvm *kvm);
913 
914 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu);
915 void kvm_vcpu_kick(struct kvm_vcpu *vcpu);
916 
917 bool kvm_is_reserved_pfn(kvm_pfn_t pfn);
918 
919 struct kvm_irq_ack_notifier {
920 	struct hlist_node link;
921 	unsigned gsi;
922 	void (*irq_acked)(struct kvm_irq_ack_notifier *kian);
923 };
924 
925 int kvm_irq_map_gsi(struct kvm *kvm,
926 		    struct kvm_kernel_irq_routing_entry *entries, int gsi);
927 int kvm_irq_map_chip_pin(struct kvm *kvm, unsigned irqchip, unsigned pin);
928 
929 int kvm_set_irq(struct kvm *kvm, int irq_source_id, u32 irq, int level,
930 		bool line_status);
931 int kvm_set_msi(struct kvm_kernel_irq_routing_entry *irq_entry, struct kvm *kvm,
932 		int irq_source_id, int level, bool line_status);
933 int kvm_arch_set_irq_inatomic(struct kvm_kernel_irq_routing_entry *e,
934 			       struct kvm *kvm, int irq_source_id,
935 			       int level, bool line_status);
936 bool kvm_irq_has_notifier(struct kvm *kvm, unsigned irqchip, unsigned pin);
937 void kvm_notify_acked_gsi(struct kvm *kvm, int gsi);
938 void kvm_notify_acked_irq(struct kvm *kvm, unsigned irqchip, unsigned pin);
939 void kvm_register_irq_ack_notifier(struct kvm *kvm,
940 				   struct kvm_irq_ack_notifier *kian);
941 void kvm_unregister_irq_ack_notifier(struct kvm *kvm,
942 				   struct kvm_irq_ack_notifier *kian);
943 int kvm_request_irq_source_id(struct kvm *kvm);
944 void kvm_free_irq_source_id(struct kvm *kvm, int irq_source_id);
945 
946 /*
947  * search_memslots() and __gfn_to_memslot() are here because they are
948  * used in non-modular code in arch/powerpc/kvm/book3s_hv_rm_mmu.c.
949  * gfn_to_memslot() itself isn't here as an inline because that would
950  * bloat other code too much.
951  */
952 static inline struct kvm_memory_slot *
953 search_memslots(struct kvm_memslots *slots, gfn_t gfn)
954 {
955 	int start = 0, end = slots->used_slots;
956 	int slot = atomic_read(&slots->lru_slot);
957 	struct kvm_memory_slot *memslots = slots->memslots;
958 
959 	if (gfn >= memslots[slot].base_gfn &&
960 	    gfn < memslots[slot].base_gfn + memslots[slot].npages)
961 		return &memslots[slot];
962 
963 	while (start < end) {
964 		slot = start + (end - start) / 2;
965 
966 		if (gfn >= memslots[slot].base_gfn)
967 			end = slot;
968 		else
969 			start = slot + 1;
970 	}
971 
972 	if (gfn >= memslots[start].base_gfn &&
973 	    gfn < memslots[start].base_gfn + memslots[start].npages) {
974 		atomic_set(&slots->lru_slot, start);
975 		return &memslots[start];
976 	}
977 
978 	return NULL;
979 }
980 
981 static inline struct kvm_memory_slot *
982 __gfn_to_memslot(struct kvm_memslots *slots, gfn_t gfn)
983 {
984 	return search_memslots(slots, gfn);
985 }
986 
987 static inline unsigned long
988 __gfn_to_hva_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
989 {
990 	return slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE;
991 }
992 
993 static inline int memslot_id(struct kvm *kvm, gfn_t gfn)
994 {
995 	return gfn_to_memslot(kvm, gfn)->id;
996 }
997 
998 static inline gfn_t
999 hva_to_gfn_memslot(unsigned long hva, struct kvm_memory_slot *slot)
1000 {
1001 	gfn_t gfn_offset = (hva - slot->userspace_addr) >> PAGE_SHIFT;
1002 
1003 	return slot->base_gfn + gfn_offset;
1004 }
1005 
1006 static inline gpa_t gfn_to_gpa(gfn_t gfn)
1007 {
1008 	return (gpa_t)gfn << PAGE_SHIFT;
1009 }
1010 
1011 static inline gfn_t gpa_to_gfn(gpa_t gpa)
1012 {
1013 	return (gfn_t)(gpa >> PAGE_SHIFT);
1014 }
1015 
1016 static inline hpa_t pfn_to_hpa(kvm_pfn_t pfn)
1017 {
1018 	return (hpa_t)pfn << PAGE_SHIFT;
1019 }
1020 
1021 static inline struct page *kvm_vcpu_gpa_to_page(struct kvm_vcpu *vcpu,
1022 						gpa_t gpa)
1023 {
1024 	return kvm_vcpu_gfn_to_page(vcpu, gpa_to_gfn(gpa));
1025 }
1026 
1027 static inline bool kvm_is_error_gpa(struct kvm *kvm, gpa_t gpa)
1028 {
1029 	unsigned long hva = gfn_to_hva(kvm, gpa_to_gfn(gpa));
1030 
1031 	return kvm_is_error_hva(hva);
1032 }
1033 
1034 enum kvm_stat_kind {
1035 	KVM_STAT_VM,
1036 	KVM_STAT_VCPU,
1037 };
1038 
1039 struct kvm_stat_data {
1040 	int offset;
1041 	struct kvm *kvm;
1042 };
1043 
1044 struct kvm_stats_debugfs_item {
1045 	const char *name;
1046 	int offset;
1047 	enum kvm_stat_kind kind;
1048 };
1049 extern struct kvm_stats_debugfs_item debugfs_entries[];
1050 extern struct dentry *kvm_debugfs_dir;
1051 
1052 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1053 static inline int mmu_notifier_retry(struct kvm *kvm, unsigned long mmu_seq)
1054 {
1055 	if (unlikely(kvm->mmu_notifier_count))
1056 		return 1;
1057 	/*
1058 	 * Ensure the read of mmu_notifier_count happens before the read
1059 	 * of mmu_notifier_seq.  This interacts with the smp_wmb() in
1060 	 * mmu_notifier_invalidate_range_end to make sure that the caller
1061 	 * either sees the old (non-zero) value of mmu_notifier_count or
1062 	 * the new (incremented) value of mmu_notifier_seq.
1063 	 * PowerPC Book3s HV KVM calls this under a per-page lock
1064 	 * rather than under kvm->mmu_lock, for scalability, so
1065 	 * can't rely on kvm->mmu_lock to keep things ordered.
1066 	 */
1067 	smp_rmb();
1068 	if (kvm->mmu_notifier_seq != mmu_seq)
1069 		return 1;
1070 	return 0;
1071 }
1072 #endif
1073 
1074 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
1075 
1076 #define KVM_MAX_IRQ_ROUTES 4096 /* might need extension/rework in the future */
1077 
1078 bool kvm_arch_can_set_irq_routing(struct kvm *kvm);
1079 int kvm_set_irq_routing(struct kvm *kvm,
1080 			const struct kvm_irq_routing_entry *entries,
1081 			unsigned nr,
1082 			unsigned flags);
1083 int kvm_set_routing_entry(struct kvm *kvm,
1084 			  struct kvm_kernel_irq_routing_entry *e,
1085 			  const struct kvm_irq_routing_entry *ue);
1086 void kvm_free_irq_routing(struct kvm *kvm);
1087 
1088 #else
1089 
1090 static inline void kvm_free_irq_routing(struct kvm *kvm) {}
1091 
1092 #endif
1093 
1094 int kvm_send_userspace_msi(struct kvm *kvm, struct kvm_msi *msi);
1095 
1096 #ifdef CONFIG_HAVE_KVM_EVENTFD
1097 
1098 void kvm_eventfd_init(struct kvm *kvm);
1099 int kvm_ioeventfd(struct kvm *kvm, struct kvm_ioeventfd *args);
1100 
1101 #ifdef CONFIG_HAVE_KVM_IRQFD
1102 int kvm_irqfd(struct kvm *kvm, struct kvm_irqfd *args);
1103 void kvm_irqfd_release(struct kvm *kvm);
1104 void kvm_irq_routing_update(struct kvm *);
1105 #else
1106 static inline int kvm_irqfd(struct kvm *kvm, struct kvm_irqfd *args)
1107 {
1108 	return -EINVAL;
1109 }
1110 
1111 static inline void kvm_irqfd_release(struct kvm *kvm) {}
1112 #endif
1113 
1114 #else
1115 
1116 static inline void kvm_eventfd_init(struct kvm *kvm) {}
1117 
1118 static inline int kvm_irqfd(struct kvm *kvm, struct kvm_irqfd *args)
1119 {
1120 	return -EINVAL;
1121 }
1122 
1123 static inline void kvm_irqfd_release(struct kvm *kvm) {}
1124 
1125 #ifdef CONFIG_HAVE_KVM_IRQCHIP
1126 static inline void kvm_irq_routing_update(struct kvm *kvm)
1127 {
1128 }
1129 #endif
1130 
1131 static inline int kvm_ioeventfd(struct kvm *kvm, struct kvm_ioeventfd *args)
1132 {
1133 	return -ENOSYS;
1134 }
1135 
1136 #endif /* CONFIG_HAVE_KVM_EVENTFD */
1137 
1138 void kvm_arch_irq_routing_update(struct kvm *kvm);
1139 
1140 static inline void kvm_make_request(int req, struct kvm_vcpu *vcpu)
1141 {
1142 	/*
1143 	 * Ensure the rest of the request is published to kvm_check_request's
1144 	 * caller.  Paired with the smp_mb__after_atomic in kvm_check_request.
1145 	 */
1146 	smp_wmb();
1147 	set_bit(req & KVM_REQUEST_MASK, (void *)&vcpu->requests);
1148 }
1149 
1150 static inline bool kvm_request_pending(struct kvm_vcpu *vcpu)
1151 {
1152 	return READ_ONCE(vcpu->requests);
1153 }
1154 
1155 static inline bool kvm_test_request(int req, struct kvm_vcpu *vcpu)
1156 {
1157 	return test_bit(req & KVM_REQUEST_MASK, (void *)&vcpu->requests);
1158 }
1159 
1160 static inline void kvm_clear_request(int req, struct kvm_vcpu *vcpu)
1161 {
1162 	clear_bit(req & KVM_REQUEST_MASK, (void *)&vcpu->requests);
1163 }
1164 
1165 static inline bool kvm_check_request(int req, struct kvm_vcpu *vcpu)
1166 {
1167 	if (kvm_test_request(req, vcpu)) {
1168 		kvm_clear_request(req, vcpu);
1169 
1170 		/*
1171 		 * Ensure the rest of the request is visible to kvm_check_request's
1172 		 * caller.  Paired with the smp_wmb in kvm_make_request.
1173 		 */
1174 		smp_mb__after_atomic();
1175 		return true;
1176 	} else {
1177 		return false;
1178 	}
1179 }
1180 
1181 extern bool kvm_rebooting;
1182 
1183 extern unsigned int halt_poll_ns;
1184 extern unsigned int halt_poll_ns_grow;
1185 extern unsigned int halt_poll_ns_shrink;
1186 
1187 struct kvm_device {
1188 	struct kvm_device_ops *ops;
1189 	struct kvm *kvm;
1190 	void *private;
1191 	struct list_head vm_node;
1192 };
1193 
1194 /* create, destroy, and name are mandatory */
1195 struct kvm_device_ops {
1196 	const char *name;
1197 
1198 	/*
1199 	 * create is called holding kvm->lock and any operations not suitable
1200 	 * to do while holding the lock should be deferred to init (see
1201 	 * below).
1202 	 */
1203 	int (*create)(struct kvm_device *dev, u32 type);
1204 
1205 	/*
1206 	 * init is called after create if create is successful and is called
1207 	 * outside of holding kvm->lock.
1208 	 */
1209 	void (*init)(struct kvm_device *dev);
1210 
1211 	/*
1212 	 * Destroy is responsible for freeing dev.
1213 	 *
1214 	 * Destroy may be called before or after destructors are called
1215 	 * on emulated I/O regions, depending on whether a reference is
1216 	 * held by a vcpu or other kvm component that gets destroyed
1217 	 * after the emulated I/O.
1218 	 */
1219 	void (*destroy)(struct kvm_device *dev);
1220 
1221 	int (*set_attr)(struct kvm_device *dev, struct kvm_device_attr *attr);
1222 	int (*get_attr)(struct kvm_device *dev, struct kvm_device_attr *attr);
1223 	int (*has_attr)(struct kvm_device *dev, struct kvm_device_attr *attr);
1224 	long (*ioctl)(struct kvm_device *dev, unsigned int ioctl,
1225 		      unsigned long arg);
1226 };
1227 
1228 void kvm_device_get(struct kvm_device *dev);
1229 void kvm_device_put(struct kvm_device *dev);
1230 struct kvm_device *kvm_device_from_filp(struct file *filp);
1231 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type);
1232 void kvm_unregister_device_ops(u32 type);
1233 
1234 extern struct kvm_device_ops kvm_mpic_ops;
1235 extern struct kvm_device_ops kvm_arm_vgic_v2_ops;
1236 extern struct kvm_device_ops kvm_arm_vgic_v3_ops;
1237 
1238 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1239 
1240 static inline void kvm_vcpu_set_in_spin_loop(struct kvm_vcpu *vcpu, bool val)
1241 {
1242 	vcpu->spin_loop.in_spin_loop = val;
1243 }
1244 static inline void kvm_vcpu_set_dy_eligible(struct kvm_vcpu *vcpu, bool val)
1245 {
1246 	vcpu->spin_loop.dy_eligible = val;
1247 }
1248 
1249 #else /* !CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT */
1250 
1251 static inline void kvm_vcpu_set_in_spin_loop(struct kvm_vcpu *vcpu, bool val)
1252 {
1253 }
1254 
1255 static inline void kvm_vcpu_set_dy_eligible(struct kvm_vcpu *vcpu, bool val)
1256 {
1257 }
1258 #endif /* CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT */
1259 
1260 #ifdef CONFIG_HAVE_KVM_IRQ_BYPASS
1261 bool kvm_arch_has_irq_bypass(void);
1262 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *,
1263 			   struct irq_bypass_producer *);
1264 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *,
1265 			   struct irq_bypass_producer *);
1266 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *);
1267 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *);
1268 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
1269 				  uint32_t guest_irq, bool set);
1270 #endif /* CONFIG_HAVE_KVM_IRQ_BYPASS */
1271 
1272 #ifdef CONFIG_HAVE_KVM_INVALID_WAKEUPS
1273 /* If we wakeup during the poll time, was it a sucessful poll? */
1274 static inline bool vcpu_valid_wakeup(struct kvm_vcpu *vcpu)
1275 {
1276 	return vcpu->valid_wakeup;
1277 }
1278 
1279 #else
1280 static inline bool vcpu_valid_wakeup(struct kvm_vcpu *vcpu)
1281 {
1282 	return true;
1283 }
1284 #endif /* CONFIG_HAVE_KVM_INVALID_WAKEUPS */
1285 
1286 #ifdef CONFIG_HAVE_KVM_VCPU_ASYNC_IOCTL
1287 long kvm_arch_vcpu_async_ioctl(struct file *filp,
1288 			       unsigned int ioctl, unsigned long arg);
1289 #else
1290 static inline long kvm_arch_vcpu_async_ioctl(struct file *filp,
1291 					     unsigned int ioctl,
1292 					     unsigned long arg)
1293 {
1294 	return -ENOIOCTLCMD;
1295 }
1296 #endif /* CONFIG_HAVE_KVM_VCPU_ASYNC_IOCTL */
1297 
1298 int kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
1299 		unsigned long start, unsigned long end, bool blockable);
1300 
1301 #ifdef CONFIG_HAVE_KVM_VCPU_RUN_PID_CHANGE
1302 int kvm_arch_vcpu_run_pid_change(struct kvm_vcpu *vcpu);
1303 #else
1304 static inline int kvm_arch_vcpu_run_pid_change(struct kvm_vcpu *vcpu)
1305 {
1306 	return 0;
1307 }
1308 #endif /* CONFIG_HAVE_KVM_VCPU_RUN_PID_CHANGE */
1309 
1310 #endif
1311