xref: /linux-6.15/include/linux/kvm_host.h (revision 8c3a2b4c)
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/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 <asm/signal.h>
28 
29 #include <linux/kvm.h>
30 #include <linux/kvm_para.h>
31 
32 #include <linux/kvm_types.h>
33 
34 #include <asm/kvm_host.h>
35 
36 #ifndef KVM_MMIO_SIZE
37 #define KVM_MMIO_SIZE 8
38 #endif
39 
40 /*
41  * The bit 16 ~ bit 31 of kvm_memory_region::flags are internally used
42  * in kvm, other bits are visible for userspace which are defined in
43  * include/linux/kvm_h.
44  */
45 #define KVM_MEMSLOT_INVALID	(1UL << 16)
46 
47 /* Two fragments for cross MMIO pages. */
48 #define KVM_MAX_MMIO_FRAGMENTS	2
49 
50 /*
51  * For the normal pfn, the highest 12 bits should be zero,
52  * so we can mask bit 62 ~ bit 52  to indicate the error pfn,
53  * mask bit 63 to indicate the noslot pfn.
54  */
55 #define KVM_PFN_ERR_MASK	(0x7ffULL << 52)
56 #define KVM_PFN_ERR_NOSLOT_MASK	(0xfffULL << 52)
57 #define KVM_PFN_NOSLOT		(0x1ULL << 63)
58 
59 #define KVM_PFN_ERR_FAULT	(KVM_PFN_ERR_MASK)
60 #define KVM_PFN_ERR_HWPOISON	(KVM_PFN_ERR_MASK + 1)
61 #define KVM_PFN_ERR_RO_FAULT	(KVM_PFN_ERR_MASK + 2)
62 
63 /*
64  * error pfns indicate that the gfn is in slot but faild to
65  * translate it to pfn on host.
66  */
67 static inline bool is_error_pfn(pfn_t pfn)
68 {
69 	return !!(pfn & KVM_PFN_ERR_MASK);
70 }
71 
72 /*
73  * error_noslot pfns indicate that the gfn can not be
74  * translated to pfn - it is not in slot or failed to
75  * translate it to pfn.
76  */
77 static inline bool is_error_noslot_pfn(pfn_t pfn)
78 {
79 	return !!(pfn & KVM_PFN_ERR_NOSLOT_MASK);
80 }
81 
82 /* noslot pfn indicates that the gfn is not in slot. */
83 static inline bool is_noslot_pfn(pfn_t pfn)
84 {
85 	return pfn == KVM_PFN_NOSLOT;
86 }
87 
88 /*
89  * architectures with KVM_HVA_ERR_BAD other than PAGE_OFFSET (e.g. s390)
90  * provide own defines and kvm_is_error_hva
91  */
92 #ifndef KVM_HVA_ERR_BAD
93 
94 #define KVM_HVA_ERR_BAD		(PAGE_OFFSET)
95 #define KVM_HVA_ERR_RO_BAD	(PAGE_OFFSET + PAGE_SIZE)
96 
97 static inline bool kvm_is_error_hva(unsigned long addr)
98 {
99 	return addr >= PAGE_OFFSET;
100 }
101 
102 #endif
103 
104 #define KVM_ERR_PTR_BAD_PAGE	(ERR_PTR(-ENOENT))
105 
106 static inline bool is_error_page(struct page *page)
107 {
108 	return IS_ERR(page);
109 }
110 
111 /*
112  * vcpu->requests bit members
113  */
114 #define KVM_REQ_TLB_FLUSH          0
115 #define KVM_REQ_MIGRATE_TIMER      1
116 #define KVM_REQ_REPORT_TPR_ACCESS  2
117 #define KVM_REQ_MMU_RELOAD         3
118 #define KVM_REQ_TRIPLE_FAULT       4
119 #define KVM_REQ_PENDING_TIMER      5
120 #define KVM_REQ_UNHALT             6
121 #define KVM_REQ_MMU_SYNC           7
122 #define KVM_REQ_CLOCK_UPDATE       8
123 #define KVM_REQ_KICK               9
124 #define KVM_REQ_DEACTIVATE_FPU    10
125 #define KVM_REQ_EVENT             11
126 #define KVM_REQ_APF_HALT          12
127 #define KVM_REQ_STEAL_UPDATE      13
128 #define KVM_REQ_NMI               14
129 #define KVM_REQ_PMU               15
130 #define KVM_REQ_PMI               16
131 #define KVM_REQ_WATCHDOG          17
132 #define KVM_REQ_MASTERCLOCK_UPDATE 18
133 #define KVM_REQ_MCLOCK_INPROGRESS 19
134 #define KVM_REQ_EPR_EXIT          20
135 #define KVM_REQ_SCAN_IOAPIC       21
136 #define KVM_REQ_GLOBAL_CLOCK_UPDATE 22
137 
138 #define KVM_USERSPACE_IRQ_SOURCE_ID		0
139 #define KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID	1
140 
141 struct kvm;
142 struct kvm_vcpu;
143 extern struct kmem_cache *kvm_vcpu_cache;
144 
145 extern raw_spinlock_t kvm_lock;
146 extern struct list_head vm_list;
147 
148 struct kvm_io_range {
149 	gpa_t addr;
150 	int len;
151 	struct kvm_io_device *dev;
152 };
153 
154 #define NR_IOBUS_DEVS 1000
155 
156 struct kvm_io_bus {
157 	int dev_count;
158 	int ioeventfd_count;
159 	struct kvm_io_range range[];
160 };
161 
162 enum kvm_bus {
163 	KVM_MMIO_BUS,
164 	KVM_PIO_BUS,
165 	KVM_VIRTIO_CCW_NOTIFY_BUS,
166 	KVM_NR_BUSES
167 };
168 
169 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
170 		     int len, const void *val);
171 int kvm_io_bus_write_cookie(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
172 			    int len, const void *val, long cookie);
173 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr, int len,
174 		    void *val);
175 int kvm_io_bus_read_cookie(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
176 			   int len, void *val, long cookie);
177 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
178 			    int len, struct kvm_io_device *dev);
179 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
180 			      struct kvm_io_device *dev);
181 
182 #ifdef CONFIG_KVM_ASYNC_PF
183 struct kvm_async_pf {
184 	struct work_struct work;
185 	struct list_head link;
186 	struct list_head queue;
187 	struct kvm_vcpu *vcpu;
188 	struct mm_struct *mm;
189 	gva_t gva;
190 	unsigned long addr;
191 	struct kvm_arch_async_pf arch;
192 	struct page *page;
193 	bool done;
194 };
195 
196 void kvm_clear_async_pf_completion_queue(struct kvm_vcpu *vcpu);
197 void kvm_check_async_pf_completion(struct kvm_vcpu *vcpu);
198 int kvm_setup_async_pf(struct kvm_vcpu *vcpu, gva_t gva, gfn_t gfn,
199 		       struct kvm_arch_async_pf *arch);
200 int kvm_async_pf_wakeup_all(struct kvm_vcpu *vcpu);
201 #endif
202 
203 enum {
204 	OUTSIDE_GUEST_MODE,
205 	IN_GUEST_MODE,
206 	EXITING_GUEST_MODE,
207 	READING_SHADOW_PAGE_TABLES,
208 };
209 
210 /*
211  * Sometimes a large or cross-page mmio needs to be broken up into separate
212  * exits for userspace servicing.
213  */
214 struct kvm_mmio_fragment {
215 	gpa_t gpa;
216 	void *data;
217 	unsigned len;
218 };
219 
220 struct kvm_vcpu {
221 	struct kvm *kvm;
222 #ifdef CONFIG_PREEMPT_NOTIFIERS
223 	struct preempt_notifier preempt_notifier;
224 #endif
225 	int cpu;
226 	int vcpu_id;
227 	int srcu_idx;
228 	int mode;
229 	unsigned long requests;
230 	unsigned long guest_debug;
231 
232 	struct mutex mutex;
233 	struct kvm_run *run;
234 
235 	int fpu_active;
236 	int guest_fpu_loaded, guest_xcr0_loaded;
237 	wait_queue_head_t wq;
238 	struct pid *pid;
239 	int sigset_active;
240 	sigset_t sigset;
241 	struct kvm_vcpu_stat stat;
242 
243 #ifdef CONFIG_HAS_IOMEM
244 	int mmio_needed;
245 	int mmio_read_completed;
246 	int mmio_is_write;
247 	int mmio_cur_fragment;
248 	int mmio_nr_fragments;
249 	struct kvm_mmio_fragment mmio_fragments[KVM_MAX_MMIO_FRAGMENTS];
250 #endif
251 
252 #ifdef CONFIG_KVM_ASYNC_PF
253 	struct {
254 		u32 queued;
255 		struct list_head queue;
256 		struct list_head done;
257 		spinlock_t lock;
258 	} async_pf;
259 #endif
260 
261 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
262 	/*
263 	 * Cpu relax intercept or pause loop exit optimization
264 	 * in_spin_loop: set when a vcpu does a pause loop exit
265 	 *  or cpu relax intercepted.
266 	 * dy_eligible: indicates whether vcpu is eligible for directed yield.
267 	 */
268 	struct {
269 		bool in_spin_loop;
270 		bool dy_eligible;
271 	} spin_loop;
272 #endif
273 	bool preempted;
274 	struct kvm_vcpu_arch arch;
275 };
276 
277 static inline int kvm_vcpu_exiting_guest_mode(struct kvm_vcpu *vcpu)
278 {
279 	return cmpxchg(&vcpu->mode, IN_GUEST_MODE, EXITING_GUEST_MODE);
280 }
281 
282 /*
283  * Some of the bitops functions do not support too long bitmaps.
284  * This number must be determined not to exceed such limits.
285  */
286 #define KVM_MEM_MAX_NR_PAGES ((1UL << 31) - 1)
287 
288 struct kvm_memory_slot {
289 	gfn_t base_gfn;
290 	unsigned long npages;
291 	unsigned long *dirty_bitmap;
292 	struct kvm_arch_memory_slot arch;
293 	unsigned long userspace_addr;
294 	u32 flags;
295 	short id;
296 };
297 
298 static inline unsigned long kvm_dirty_bitmap_bytes(struct kvm_memory_slot *memslot)
299 {
300 	return ALIGN(memslot->npages, BITS_PER_LONG) / 8;
301 }
302 
303 struct kvm_kernel_irq_routing_entry {
304 	u32 gsi;
305 	u32 type;
306 	int (*set)(struct kvm_kernel_irq_routing_entry *e,
307 		   struct kvm *kvm, int irq_source_id, int level,
308 		   bool line_status);
309 	union {
310 		struct {
311 			unsigned irqchip;
312 			unsigned pin;
313 		} irqchip;
314 		struct msi_msg msi;
315 	};
316 	struct hlist_node link;
317 };
318 
319 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
320 
321 struct kvm_irq_routing_table {
322 	int chip[KVM_NR_IRQCHIPS][KVM_IRQCHIP_NUM_PINS];
323 	struct kvm_kernel_irq_routing_entry *rt_entries;
324 	u32 nr_rt_entries;
325 	/*
326 	 * Array indexed by gsi. Each entry contains list of irq chips
327 	 * the gsi is connected to.
328 	 */
329 	struct hlist_head map[0];
330 };
331 
332 #else
333 
334 struct kvm_irq_routing_table {};
335 
336 #endif
337 
338 #ifndef KVM_PRIVATE_MEM_SLOTS
339 #define KVM_PRIVATE_MEM_SLOTS 0
340 #endif
341 
342 #ifndef KVM_MEM_SLOTS_NUM
343 #define KVM_MEM_SLOTS_NUM (KVM_USER_MEM_SLOTS + KVM_PRIVATE_MEM_SLOTS)
344 #endif
345 
346 /*
347  * Note:
348  * memslots are not sorted by id anymore, please use id_to_memslot()
349  * to get the memslot by its id.
350  */
351 struct kvm_memslots {
352 	u64 generation;
353 	struct kvm_memory_slot memslots[KVM_MEM_SLOTS_NUM];
354 	/* The mapping table from slot id to the index in memslots[]. */
355 	short id_to_index[KVM_MEM_SLOTS_NUM];
356 };
357 
358 struct kvm {
359 	spinlock_t mmu_lock;
360 	struct mutex slots_lock;
361 	struct mm_struct *mm; /* userspace tied to this vm */
362 	struct kvm_memslots *memslots;
363 	struct srcu_struct srcu;
364 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
365 	u32 bsp_vcpu_id;
366 #endif
367 	struct kvm_vcpu *vcpus[KVM_MAX_VCPUS];
368 	atomic_t online_vcpus;
369 	int last_boosted_vcpu;
370 	struct list_head vm_list;
371 	struct mutex lock;
372 	struct kvm_io_bus *buses[KVM_NR_BUSES];
373 #ifdef CONFIG_HAVE_KVM_EVENTFD
374 	struct {
375 		spinlock_t        lock;
376 		struct list_head  items;
377 		struct list_head  resampler_list;
378 		struct mutex      resampler_lock;
379 	} irqfds;
380 	struct list_head ioeventfds;
381 #endif
382 	struct kvm_vm_stat stat;
383 	struct kvm_arch arch;
384 	atomic_t users_count;
385 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
386 	struct kvm_coalesced_mmio_ring *coalesced_mmio_ring;
387 	spinlock_t ring_lock;
388 	struct list_head coalesced_zones;
389 #endif
390 
391 	struct mutex irq_lock;
392 #ifdef CONFIG_HAVE_KVM_IRQCHIP
393 	/*
394 	 * Update side is protected by irq_lock and,
395 	 * if configured, irqfds.lock.
396 	 */
397 	struct kvm_irq_routing_table __rcu *irq_routing;
398 	struct hlist_head mask_notifier_list;
399 	struct hlist_head irq_ack_notifier_list;
400 #endif
401 
402 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
403 	struct mmu_notifier mmu_notifier;
404 	unsigned long mmu_notifier_seq;
405 	long mmu_notifier_count;
406 #endif
407 	long tlbs_dirty;
408 	struct list_head devices;
409 };
410 
411 #define kvm_err(fmt, ...) \
412 	pr_err("kvm [%i]: " fmt, task_pid_nr(current), ## __VA_ARGS__)
413 #define kvm_info(fmt, ...) \
414 	pr_info("kvm [%i]: " fmt, task_pid_nr(current), ## __VA_ARGS__)
415 #define kvm_debug(fmt, ...) \
416 	pr_debug("kvm [%i]: " fmt, task_pid_nr(current), ## __VA_ARGS__)
417 #define kvm_pr_unimpl(fmt, ...) \
418 	pr_err_ratelimited("kvm [%i]: " fmt, \
419 			   task_tgid_nr(current), ## __VA_ARGS__)
420 
421 /* The guest did something we don't support. */
422 #define vcpu_unimpl(vcpu, fmt, ...)					\
423 	kvm_pr_unimpl("vcpu%i " fmt, (vcpu)->vcpu_id, ## __VA_ARGS__)
424 
425 static inline struct kvm_vcpu *kvm_get_vcpu(struct kvm *kvm, int i)
426 {
427 	smp_rmb();
428 	return kvm->vcpus[i];
429 }
430 
431 #define kvm_for_each_vcpu(idx, vcpup, kvm) \
432 	for (idx = 0; \
433 	     idx < atomic_read(&kvm->online_vcpus) && \
434 	     (vcpup = kvm_get_vcpu(kvm, idx)) != NULL; \
435 	     idx++)
436 
437 #define kvm_for_each_memslot(memslot, slots)	\
438 	for (memslot = &slots->memslots[0];	\
439 	      memslot < slots->memslots + KVM_MEM_SLOTS_NUM && memslot->npages;\
440 		memslot++)
441 
442 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id);
443 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu);
444 
445 int __must_check vcpu_load(struct kvm_vcpu *vcpu);
446 void vcpu_put(struct kvm_vcpu *vcpu);
447 
448 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
449 int kvm_irqfd_init(void);
450 void kvm_irqfd_exit(void);
451 #else
452 static inline int kvm_irqfd_init(void)
453 {
454 	return 0;
455 }
456 
457 static inline void kvm_irqfd_exit(void)
458 {
459 }
460 #endif
461 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
462 		  struct module *module);
463 void kvm_exit(void);
464 
465 void kvm_get_kvm(struct kvm *kvm);
466 void kvm_put_kvm(struct kvm *kvm);
467 void update_memslots(struct kvm_memslots *slots, struct kvm_memory_slot *new,
468 		     u64 last_generation);
469 
470 static inline struct kvm_memslots *kvm_memslots(struct kvm *kvm)
471 {
472 	return rcu_dereference_check(kvm->memslots,
473 			srcu_read_lock_held(&kvm->srcu)
474 			|| lockdep_is_held(&kvm->slots_lock));
475 }
476 
477 static inline struct kvm_memory_slot *
478 id_to_memslot(struct kvm_memslots *slots, int id)
479 {
480 	int index = slots->id_to_index[id];
481 	struct kvm_memory_slot *slot;
482 
483 	slot = &slots->memslots[index];
484 
485 	WARN_ON(slot->id != id);
486 	return slot;
487 }
488 
489 /*
490  * KVM_SET_USER_MEMORY_REGION ioctl allows the following operations:
491  * - create a new memory slot
492  * - delete an existing memory slot
493  * - modify an existing memory slot
494  *   -- move it in the guest physical memory space
495  *   -- just change its flags
496  *
497  * Since flags can be changed by some of these operations, the following
498  * differentiation is the best we can do for __kvm_set_memory_region():
499  */
500 enum kvm_mr_change {
501 	KVM_MR_CREATE,
502 	KVM_MR_DELETE,
503 	KVM_MR_MOVE,
504 	KVM_MR_FLAGS_ONLY,
505 };
506 
507 int kvm_set_memory_region(struct kvm *kvm,
508 			  struct kvm_userspace_memory_region *mem);
509 int __kvm_set_memory_region(struct kvm *kvm,
510 			    struct kvm_userspace_memory_region *mem);
511 void kvm_arch_free_memslot(struct kvm_memory_slot *free,
512 			   struct kvm_memory_slot *dont);
513 int kvm_arch_create_memslot(struct kvm_memory_slot *slot, unsigned long npages);
514 void kvm_arch_memslots_updated(struct kvm *kvm);
515 int kvm_arch_prepare_memory_region(struct kvm *kvm,
516 				struct kvm_memory_slot *memslot,
517 				struct kvm_userspace_memory_region *mem,
518 				enum kvm_mr_change change);
519 void kvm_arch_commit_memory_region(struct kvm *kvm,
520 				struct kvm_userspace_memory_region *mem,
521 				const struct kvm_memory_slot *old,
522 				enum kvm_mr_change change);
523 bool kvm_largepages_enabled(void);
524 void kvm_disable_largepages(void);
525 /* flush all memory translations */
526 void kvm_arch_flush_shadow_all(struct kvm *kvm);
527 /* flush memory translations pointing to 'slot' */
528 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
529 				   struct kvm_memory_slot *slot);
530 
531 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
532 			    int nr_pages);
533 
534 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn);
535 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn);
536 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot, gfn_t gfn);
537 void kvm_release_page_clean(struct page *page);
538 void kvm_release_page_dirty(struct page *page);
539 void kvm_set_page_dirty(struct page *page);
540 void kvm_set_page_accessed(struct page *page);
541 
542 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn);
543 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
544 		       bool write_fault, bool *writable);
545 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn);
546 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
547 		      bool *writable);
548 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn);
549 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn);
550 
551 void kvm_release_pfn_dirty(pfn_t pfn);
552 void kvm_release_pfn_clean(pfn_t pfn);
553 void kvm_set_pfn_dirty(pfn_t pfn);
554 void kvm_set_pfn_accessed(pfn_t pfn);
555 void kvm_get_pfn(pfn_t pfn);
556 
557 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
558 			int len);
559 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
560 			  unsigned long len);
561 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len);
562 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
563 			   void *data, unsigned long len);
564 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
565 			 int offset, int len);
566 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
567 		    unsigned long len);
568 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
569 			   void *data, unsigned long len);
570 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
571 			      gpa_t gpa, unsigned long len);
572 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len);
573 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len);
574 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn);
575 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn);
576 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn);
577 void mark_page_dirty(struct kvm *kvm, gfn_t gfn);
578 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
579 			     gfn_t gfn);
580 
581 void kvm_vcpu_block(struct kvm_vcpu *vcpu);
582 void kvm_vcpu_kick(struct kvm_vcpu *vcpu);
583 bool kvm_vcpu_yield_to(struct kvm_vcpu *target);
584 void kvm_vcpu_on_spin(struct kvm_vcpu *vcpu);
585 void kvm_resched(struct kvm_vcpu *vcpu);
586 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu);
587 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu);
588 
589 void kvm_flush_remote_tlbs(struct kvm *kvm);
590 void kvm_reload_remote_mmus(struct kvm *kvm);
591 void kvm_make_mclock_inprogress_request(struct kvm *kvm);
592 void kvm_make_scan_ioapic_request(struct kvm *kvm);
593 
594 long kvm_arch_dev_ioctl(struct file *filp,
595 			unsigned int ioctl, unsigned long arg);
596 long kvm_arch_vcpu_ioctl(struct file *filp,
597 			 unsigned int ioctl, unsigned long arg);
598 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf);
599 
600 int kvm_dev_ioctl_check_extension(long ext);
601 
602 int kvm_get_dirty_log(struct kvm *kvm,
603 			struct kvm_dirty_log *log, int *is_dirty);
604 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
605 				struct kvm_dirty_log *log);
606 
607 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
608 				   struct kvm_userspace_memory_region *mem);
609 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
610 			bool line_status);
611 long kvm_arch_vm_ioctl(struct file *filp,
612 		       unsigned int ioctl, unsigned long arg);
613 
614 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu);
615 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu);
616 
617 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
618 				    struct kvm_translation *tr);
619 
620 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs);
621 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs);
622 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
623 				  struct kvm_sregs *sregs);
624 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
625 				  struct kvm_sregs *sregs);
626 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
627 				    struct kvm_mp_state *mp_state);
628 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
629 				    struct kvm_mp_state *mp_state);
630 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
631 					struct kvm_guest_debug *dbg);
632 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run);
633 
634 int kvm_arch_init(void *opaque);
635 void kvm_arch_exit(void);
636 
637 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu);
638 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu);
639 
640 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu);
641 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu);
642 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu);
643 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id);
644 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu);
645 int kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu);
646 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu);
647 
648 int kvm_arch_hardware_enable(void *garbage);
649 void kvm_arch_hardware_disable(void *garbage);
650 int kvm_arch_hardware_setup(void);
651 void kvm_arch_hardware_unsetup(void);
652 void kvm_arch_check_processor_compat(void *rtn);
653 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu);
654 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu);
655 
656 void kvm_free_physmem(struct kvm *kvm);
657 
658 void *kvm_kvzalloc(unsigned long size);
659 void kvm_kvfree(const void *addr);
660 
661 #ifndef __KVM_HAVE_ARCH_VM_ALLOC
662 static inline struct kvm *kvm_arch_alloc_vm(void)
663 {
664 	return kzalloc(sizeof(struct kvm), GFP_KERNEL);
665 }
666 
667 static inline void kvm_arch_free_vm(struct kvm *kvm)
668 {
669 	kfree(kvm);
670 }
671 #endif
672 
673 static inline wait_queue_head_t *kvm_arch_vcpu_wq(struct kvm_vcpu *vcpu)
674 {
675 #ifdef __KVM_HAVE_ARCH_WQP
676 	return vcpu->arch.wqp;
677 #else
678 	return &vcpu->wq;
679 #endif
680 }
681 
682 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type);
683 void kvm_arch_destroy_vm(struct kvm *kvm);
684 void kvm_arch_sync_events(struct kvm *kvm);
685 
686 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu);
687 void kvm_vcpu_kick(struct kvm_vcpu *vcpu);
688 
689 bool kvm_is_mmio_pfn(pfn_t pfn);
690 
691 struct kvm_irq_ack_notifier {
692 	struct hlist_node link;
693 	unsigned gsi;
694 	void (*irq_acked)(struct kvm_irq_ack_notifier *kian);
695 };
696 
697 struct kvm_assigned_dev_kernel {
698 	struct kvm_irq_ack_notifier ack_notifier;
699 	struct list_head list;
700 	int assigned_dev_id;
701 	int host_segnr;
702 	int host_busnr;
703 	int host_devfn;
704 	unsigned int entries_nr;
705 	int host_irq;
706 	bool host_irq_disabled;
707 	bool pci_2_3;
708 	struct msix_entry *host_msix_entries;
709 	int guest_irq;
710 	struct msix_entry *guest_msix_entries;
711 	unsigned long irq_requested_type;
712 	int irq_source_id;
713 	int flags;
714 	struct pci_dev *dev;
715 	struct kvm *kvm;
716 	spinlock_t intx_lock;
717 	spinlock_t intx_mask_lock;
718 	char irq_name[32];
719 	struct pci_saved_state *pci_saved_state;
720 };
721 
722 struct kvm_irq_mask_notifier {
723 	void (*func)(struct kvm_irq_mask_notifier *kimn, bool masked);
724 	int irq;
725 	struct hlist_node link;
726 };
727 
728 void kvm_register_irq_mask_notifier(struct kvm *kvm, int irq,
729 				    struct kvm_irq_mask_notifier *kimn);
730 void kvm_unregister_irq_mask_notifier(struct kvm *kvm, int irq,
731 				      struct kvm_irq_mask_notifier *kimn);
732 void kvm_fire_mask_notifiers(struct kvm *kvm, unsigned irqchip, unsigned pin,
733 			     bool mask);
734 
735 int kvm_set_irq(struct kvm *kvm, int irq_source_id, u32 irq, int level,
736 		bool line_status);
737 int kvm_set_irq_inatomic(struct kvm *kvm, int irq_source_id, u32 irq, int level);
738 int kvm_set_msi(struct kvm_kernel_irq_routing_entry *irq_entry, struct kvm *kvm,
739 		int irq_source_id, int level, bool line_status);
740 bool kvm_irq_has_notifier(struct kvm *kvm, unsigned irqchip, unsigned pin);
741 void kvm_notify_acked_irq(struct kvm *kvm, unsigned irqchip, unsigned pin);
742 void kvm_register_irq_ack_notifier(struct kvm *kvm,
743 				   struct kvm_irq_ack_notifier *kian);
744 void kvm_unregister_irq_ack_notifier(struct kvm *kvm,
745 				   struct kvm_irq_ack_notifier *kian);
746 int kvm_request_irq_source_id(struct kvm *kvm);
747 void kvm_free_irq_source_id(struct kvm *kvm, int irq_source_id);
748 
749 /* For vcpu->arch.iommu_flags */
750 #define KVM_IOMMU_CACHE_COHERENCY	0x1
751 
752 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
753 int kvm_iommu_map_pages(struct kvm *kvm, struct kvm_memory_slot *slot);
754 void kvm_iommu_unmap_pages(struct kvm *kvm, struct kvm_memory_slot *slot);
755 int kvm_iommu_map_guest(struct kvm *kvm);
756 int kvm_iommu_unmap_guest(struct kvm *kvm);
757 int kvm_assign_device(struct kvm *kvm,
758 		      struct kvm_assigned_dev_kernel *assigned_dev);
759 int kvm_deassign_device(struct kvm *kvm,
760 			struct kvm_assigned_dev_kernel *assigned_dev);
761 #else
762 static inline int kvm_iommu_map_pages(struct kvm *kvm,
763 				      struct kvm_memory_slot *slot)
764 {
765 	return 0;
766 }
767 
768 static inline void kvm_iommu_unmap_pages(struct kvm *kvm,
769 					 struct kvm_memory_slot *slot)
770 {
771 }
772 
773 static inline int kvm_iommu_unmap_guest(struct kvm *kvm)
774 {
775 	return 0;
776 }
777 #endif
778 
779 static inline void kvm_guest_enter(void)
780 {
781 	unsigned long flags;
782 
783 	BUG_ON(preemptible());
784 
785 	local_irq_save(flags);
786 	guest_enter();
787 	local_irq_restore(flags);
788 
789 	/* KVM does not hold any references to rcu protected data when it
790 	 * switches CPU into a guest mode. In fact switching to a guest mode
791 	 * is very similar to exiting to userspase from rcu point of view. In
792 	 * addition CPU may stay in a guest mode for quite a long time (up to
793 	 * one time slice). Lets treat guest mode as quiescent state, just like
794 	 * we do with user-mode execution.
795 	 */
796 	rcu_virt_note_context_switch(smp_processor_id());
797 }
798 
799 static inline void kvm_guest_exit(void)
800 {
801 	unsigned long flags;
802 
803 	local_irq_save(flags);
804 	guest_exit();
805 	local_irq_restore(flags);
806 }
807 
808 /*
809  * search_memslots() and __gfn_to_memslot() are here because they are
810  * used in non-modular code in arch/powerpc/kvm/book3s_hv_rm_mmu.c.
811  * gfn_to_memslot() itself isn't here as an inline because that would
812  * bloat other code too much.
813  */
814 static inline struct kvm_memory_slot *
815 search_memslots(struct kvm_memslots *slots, gfn_t gfn)
816 {
817 	struct kvm_memory_slot *memslot;
818 
819 	kvm_for_each_memslot(memslot, slots)
820 		if (gfn >= memslot->base_gfn &&
821 		      gfn < memslot->base_gfn + memslot->npages)
822 			return memslot;
823 
824 	return NULL;
825 }
826 
827 static inline struct kvm_memory_slot *
828 __gfn_to_memslot(struct kvm_memslots *slots, gfn_t gfn)
829 {
830 	return search_memslots(slots, gfn);
831 }
832 
833 static inline unsigned long
834 __gfn_to_hva_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
835 {
836 	return slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE;
837 }
838 
839 static inline int memslot_id(struct kvm *kvm, gfn_t gfn)
840 {
841 	return gfn_to_memslot(kvm, gfn)->id;
842 }
843 
844 static inline gfn_t gfn_to_index(gfn_t gfn, gfn_t base_gfn, int level)
845 {
846 	/* KVM_HPAGE_GFN_SHIFT(PT_PAGE_TABLE_LEVEL) must be 0. */
847 	return (gfn >> KVM_HPAGE_GFN_SHIFT(level)) -
848 		(base_gfn >> KVM_HPAGE_GFN_SHIFT(level));
849 }
850 
851 static inline gfn_t
852 hva_to_gfn_memslot(unsigned long hva, struct kvm_memory_slot *slot)
853 {
854 	gfn_t gfn_offset = (hva - slot->userspace_addr) >> PAGE_SHIFT;
855 
856 	return slot->base_gfn + gfn_offset;
857 }
858 
859 static inline gpa_t gfn_to_gpa(gfn_t gfn)
860 {
861 	return (gpa_t)gfn << PAGE_SHIFT;
862 }
863 
864 static inline gfn_t gpa_to_gfn(gpa_t gpa)
865 {
866 	return (gfn_t)(gpa >> PAGE_SHIFT);
867 }
868 
869 static inline hpa_t pfn_to_hpa(pfn_t pfn)
870 {
871 	return (hpa_t)pfn << PAGE_SHIFT;
872 }
873 
874 static inline void kvm_migrate_timers(struct kvm_vcpu *vcpu)
875 {
876 	set_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests);
877 }
878 
879 enum kvm_stat_kind {
880 	KVM_STAT_VM,
881 	KVM_STAT_VCPU,
882 };
883 
884 struct kvm_stats_debugfs_item {
885 	const char *name;
886 	int offset;
887 	enum kvm_stat_kind kind;
888 	struct dentry *dentry;
889 };
890 extern struct kvm_stats_debugfs_item debugfs_entries[];
891 extern struct dentry *kvm_debugfs_dir;
892 
893 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
894 static inline int mmu_notifier_retry(struct kvm *kvm, unsigned long mmu_seq)
895 {
896 	if (unlikely(kvm->mmu_notifier_count))
897 		return 1;
898 	/*
899 	 * Ensure the read of mmu_notifier_count happens before the read
900 	 * of mmu_notifier_seq.  This interacts with the smp_wmb() in
901 	 * mmu_notifier_invalidate_range_end to make sure that the caller
902 	 * either sees the old (non-zero) value of mmu_notifier_count or
903 	 * the new (incremented) value of mmu_notifier_seq.
904 	 * PowerPC Book3s HV KVM calls this under a per-page lock
905 	 * rather than under kvm->mmu_lock, for scalability, so
906 	 * can't rely on kvm->mmu_lock to keep things ordered.
907 	 */
908 	smp_rmb();
909 	if (kvm->mmu_notifier_seq != mmu_seq)
910 		return 1;
911 	return 0;
912 }
913 #endif
914 
915 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
916 
917 #define KVM_MAX_IRQ_ROUTES 1024
918 
919 int kvm_setup_default_irq_routing(struct kvm *kvm);
920 int kvm_set_irq_routing(struct kvm *kvm,
921 			const struct kvm_irq_routing_entry *entries,
922 			unsigned nr,
923 			unsigned flags);
924 int kvm_set_routing_entry(struct kvm_irq_routing_table *rt,
925 			  struct kvm_kernel_irq_routing_entry *e,
926 			  const struct kvm_irq_routing_entry *ue);
927 void kvm_free_irq_routing(struct kvm *kvm);
928 
929 int kvm_send_userspace_msi(struct kvm *kvm, struct kvm_msi *msi);
930 
931 #else
932 
933 static inline void kvm_free_irq_routing(struct kvm *kvm) {}
934 
935 #endif
936 
937 #ifdef CONFIG_HAVE_KVM_EVENTFD
938 
939 void kvm_eventfd_init(struct kvm *kvm);
940 int kvm_ioeventfd(struct kvm *kvm, struct kvm_ioeventfd *args);
941 
942 #ifdef CONFIG_HAVE_KVM_IRQCHIP
943 int kvm_irqfd(struct kvm *kvm, struct kvm_irqfd *args);
944 void kvm_irqfd_release(struct kvm *kvm);
945 void kvm_irq_routing_update(struct kvm *, struct kvm_irq_routing_table *);
946 #else
947 static inline int kvm_irqfd(struct kvm *kvm, struct kvm_irqfd *args)
948 {
949 	return -EINVAL;
950 }
951 
952 static inline void kvm_irqfd_release(struct kvm *kvm) {}
953 #endif
954 
955 #else
956 
957 static inline void kvm_eventfd_init(struct kvm *kvm) {}
958 
959 static inline int kvm_irqfd(struct kvm *kvm, struct kvm_irqfd *args)
960 {
961 	return -EINVAL;
962 }
963 
964 static inline void kvm_irqfd_release(struct kvm *kvm) {}
965 
966 #ifdef CONFIG_HAVE_KVM_IRQCHIP
967 static inline void kvm_irq_routing_update(struct kvm *kvm,
968 					  struct kvm_irq_routing_table *irq_rt)
969 {
970 	rcu_assign_pointer(kvm->irq_routing, irq_rt);
971 }
972 #endif
973 
974 static inline int kvm_ioeventfd(struct kvm *kvm, struct kvm_ioeventfd *args)
975 {
976 	return -ENOSYS;
977 }
978 
979 #endif /* CONFIG_HAVE_KVM_EVENTFD */
980 
981 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
982 static inline bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
983 {
984 	return vcpu->kvm->bsp_vcpu_id == vcpu->vcpu_id;
985 }
986 
987 bool kvm_vcpu_compatible(struct kvm_vcpu *vcpu);
988 
989 #else
990 
991 static inline bool kvm_vcpu_compatible(struct kvm_vcpu *vcpu) { return true; }
992 
993 #endif
994 
995 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
996 
997 long kvm_vm_ioctl_assigned_device(struct kvm *kvm, unsigned ioctl,
998 				  unsigned long arg);
999 
1000 void kvm_free_all_assigned_devices(struct kvm *kvm);
1001 
1002 #else
1003 
1004 static inline long kvm_vm_ioctl_assigned_device(struct kvm *kvm, unsigned ioctl,
1005 						unsigned long arg)
1006 {
1007 	return -ENOTTY;
1008 }
1009 
1010 static inline void kvm_free_all_assigned_devices(struct kvm *kvm) {}
1011 
1012 #endif
1013 
1014 static inline void kvm_make_request(int req, struct kvm_vcpu *vcpu)
1015 {
1016 	set_bit(req, &vcpu->requests);
1017 }
1018 
1019 static inline bool kvm_check_request(int req, struct kvm_vcpu *vcpu)
1020 {
1021 	if (test_bit(req, &vcpu->requests)) {
1022 		clear_bit(req, &vcpu->requests);
1023 		return true;
1024 	} else {
1025 		return false;
1026 	}
1027 }
1028 
1029 extern bool kvm_rebooting;
1030 
1031 struct kvm_device_ops;
1032 
1033 struct kvm_device {
1034 	struct kvm_device_ops *ops;
1035 	struct kvm *kvm;
1036 	void *private;
1037 	struct list_head vm_node;
1038 };
1039 
1040 /* create, destroy, and name are mandatory */
1041 struct kvm_device_ops {
1042 	const char *name;
1043 	int (*create)(struct kvm_device *dev, u32 type);
1044 
1045 	/*
1046 	 * Destroy is responsible for freeing dev.
1047 	 *
1048 	 * Destroy may be called before or after destructors are called
1049 	 * on emulated I/O regions, depending on whether a reference is
1050 	 * held by a vcpu or other kvm component that gets destroyed
1051 	 * after the emulated I/O.
1052 	 */
1053 	void (*destroy)(struct kvm_device *dev);
1054 
1055 	int (*set_attr)(struct kvm_device *dev, struct kvm_device_attr *attr);
1056 	int (*get_attr)(struct kvm_device *dev, struct kvm_device_attr *attr);
1057 	int (*has_attr)(struct kvm_device *dev, struct kvm_device_attr *attr);
1058 	long (*ioctl)(struct kvm_device *dev, unsigned int ioctl,
1059 		      unsigned long arg);
1060 };
1061 
1062 void kvm_device_get(struct kvm_device *dev);
1063 void kvm_device_put(struct kvm_device *dev);
1064 struct kvm_device *kvm_device_from_filp(struct file *filp);
1065 
1066 extern struct kvm_device_ops kvm_mpic_ops;
1067 extern struct kvm_device_ops kvm_xics_ops;
1068 
1069 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1070 
1071 static inline void kvm_vcpu_set_in_spin_loop(struct kvm_vcpu *vcpu, bool val)
1072 {
1073 	vcpu->spin_loop.in_spin_loop = val;
1074 }
1075 static inline void kvm_vcpu_set_dy_eligible(struct kvm_vcpu *vcpu, bool val)
1076 {
1077 	vcpu->spin_loop.dy_eligible = val;
1078 }
1079 
1080 #else /* !CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT */
1081 
1082 static inline void kvm_vcpu_set_in_spin_loop(struct kvm_vcpu *vcpu, bool val)
1083 {
1084 }
1085 
1086 static inline void kvm_vcpu_set_dy_eligible(struct kvm_vcpu *vcpu, bool val)
1087 {
1088 }
1089 
1090 static inline bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1091 {
1092 	return true;
1093 }
1094 
1095 #endif /* CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT */
1096 #endif
1097 
1098