1 /* SPDX-License-Identifier: GPL-2.0-only */
2 #ifndef __KVM_HOST_H
3 #define __KVM_HOST_H
4 
5 
6 #include <linux/types.h>
7 #include <linux/hardirq.h>
8 #include <linux/list.h>
9 #include <linux/mutex.h>
10 #include <linux/spinlock.h>
11 #include <linux/signal.h>
12 #include <linux/sched.h>
13 #include <linux/sched/stat.h>
14 #include <linux/bug.h>
15 #include <linux/minmax.h>
16 #include <linux/mm.h>
17 #include <linux/mmu_notifier.h>
18 #include <linux/preempt.h>
19 #include <linux/msi.h>
20 #include <linux/slab.h>
21 #include <linux/vmalloc.h>
22 #include <linux/rcupdate.h>
23 #include <linux/ratelimit.h>
24 #include <linux/err.h>
25 #include <linux/irqflags.h>
26 #include <linux/context_tracking.h>
27 #include <linux/irqbypass.h>
28 #include <linux/rcuwait.h>
29 #include <linux/refcount.h>
30 #include <linux/nospec.h>
31 #include <linux/notifier.h>
32 #include <linux/ftrace.h>
33 #include <linux/hashtable.h>
34 #include <linux/instrumentation.h>
35 #include <linux/interval_tree.h>
36 #include <linux/rbtree.h>
37 #include <linux/xarray.h>
38 #include <asm/signal.h>
39 
40 #include <linux/kvm.h>
41 #include <linux/kvm_para.h>
42 
43 #include <linux/kvm_types.h>
44 
45 #include <asm/kvm_host.h>
46 #include <linux/kvm_dirty_ring.h>
47 
48 #ifndef KVM_MAX_VCPU_IDS
49 #define KVM_MAX_VCPU_IDS KVM_MAX_VCPUS
50 #endif
51 
52 /*
53  * The bit 16 ~ bit 31 of kvm_memory_region::flags are internally used
54  * in kvm, other bits are visible for userspace which are defined in
55  * include/linux/kvm_h.
56  */
57 #define KVM_MEMSLOT_INVALID	(1UL << 16)
58 
59 /*
60  * Bit 63 of the memslot generation number is an "update in-progress flag",
61  * e.g. is temporarily set for the duration of install_new_memslots().
62  * This flag effectively creates a unique generation number that is used to
63  * mark cached memslot data, e.g. MMIO accesses, as potentially being stale,
64  * i.e. may (or may not) have come from the previous memslots generation.
65  *
66  * This is necessary because the actual memslots update is not atomic with
67  * respect to the generation number update.  Updating the generation number
68  * first would allow a vCPU to cache a spte from the old memslots using the
69  * new generation number, and updating the generation number after switching
70  * to the new memslots would allow cache hits using the old generation number
71  * to reference the defunct memslots.
72  *
73  * This mechanism is used to prevent getting hits in KVM's caches while a
74  * memslot update is in-progress, and to prevent cache hits *after* updating
75  * the actual generation number against accesses that were inserted into the
76  * cache *before* the memslots were updated.
77  */
78 #define KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS	BIT_ULL(63)
79 
80 /* Two fragments for cross MMIO pages. */
81 #define KVM_MAX_MMIO_FRAGMENTS	2
82 
83 #ifndef KVM_ADDRESS_SPACE_NUM
84 #define KVM_ADDRESS_SPACE_NUM	1
85 #endif
86 
87 /*
88  * For the normal pfn, the highest 12 bits should be zero,
89  * so we can mask bit 62 ~ bit 52  to indicate the error pfn,
90  * mask bit 63 to indicate the noslot pfn.
91  */
92 #define KVM_PFN_ERR_MASK	(0x7ffULL << 52)
93 #define KVM_PFN_ERR_NOSLOT_MASK	(0xfffULL << 52)
94 #define KVM_PFN_NOSLOT		(0x1ULL << 63)
95 
96 #define KVM_PFN_ERR_FAULT	(KVM_PFN_ERR_MASK)
97 #define KVM_PFN_ERR_HWPOISON	(KVM_PFN_ERR_MASK + 1)
98 #define KVM_PFN_ERR_RO_FAULT	(KVM_PFN_ERR_MASK + 2)
99 
100 /*
101  * error pfns indicate that the gfn is in slot but faild to
102  * translate it to pfn on host.
103  */
is_error_pfn(kvm_pfn_t pfn)104 static inline bool is_error_pfn(kvm_pfn_t pfn)
105 {
106 	return !!(pfn & KVM_PFN_ERR_MASK);
107 }
108 
109 /*
110  * error_noslot pfns indicate that the gfn can not be
111  * translated to pfn - it is not in slot or failed to
112  * translate it to pfn.
113  */
is_error_noslot_pfn(kvm_pfn_t pfn)114 static inline bool is_error_noslot_pfn(kvm_pfn_t pfn)
115 {
116 	return !!(pfn & KVM_PFN_ERR_NOSLOT_MASK);
117 }
118 
119 /* noslot pfn indicates that the gfn is not in slot. */
is_noslot_pfn(kvm_pfn_t pfn)120 static inline bool is_noslot_pfn(kvm_pfn_t pfn)
121 {
122 	return pfn == KVM_PFN_NOSLOT;
123 }
124 
125 /*
126  * architectures with KVM_HVA_ERR_BAD other than PAGE_OFFSET (e.g. s390)
127  * provide own defines and kvm_is_error_hva
128  */
129 #ifndef KVM_HVA_ERR_BAD
130 
131 #define KVM_HVA_ERR_BAD		(PAGE_OFFSET)
132 #define KVM_HVA_ERR_RO_BAD	(PAGE_OFFSET + PAGE_SIZE)
133 
kvm_is_error_hva(unsigned long addr)134 static inline bool kvm_is_error_hva(unsigned long addr)
135 {
136 	return addr >= PAGE_OFFSET;
137 }
138 
139 #endif
140 
141 #define KVM_ERR_PTR_BAD_PAGE	(ERR_PTR(-ENOENT))
142 
is_error_page(struct page * page)143 static inline bool is_error_page(struct page *page)
144 {
145 	return IS_ERR(page);
146 }
147 
148 #define KVM_REQUEST_MASK           GENMASK(7,0)
149 #define KVM_REQUEST_NO_WAKEUP      BIT(8)
150 #define KVM_REQUEST_WAIT           BIT(9)
151 #define KVM_REQUEST_NO_ACTION      BIT(10)
152 /*
153  * Architecture-independent vcpu->requests bit members
154  * Bits 3-7 are reserved for more arch-independent bits.
155  */
156 #define KVM_REQ_TLB_FLUSH         (0 | KVM_REQUEST_WAIT | KVM_REQUEST_NO_WAKEUP)
157 #define KVM_REQ_VM_DEAD           (1 | KVM_REQUEST_WAIT | KVM_REQUEST_NO_WAKEUP)
158 #define KVM_REQ_UNBLOCK           2
159 #define KVM_REQUEST_ARCH_BASE     8
160 
161 /*
162  * KVM_REQ_OUTSIDE_GUEST_MODE exists is purely as way to force the vCPU to
163  * OUTSIDE_GUEST_MODE.  KVM_REQ_OUTSIDE_GUEST_MODE differs from a vCPU "kick"
164  * in that it ensures the vCPU has reached OUTSIDE_GUEST_MODE before continuing
165  * on.  A kick only guarantees that the vCPU is on its way out, e.g. a previous
166  * kick may have set vcpu->mode to EXITING_GUEST_MODE, and so there's no
167  * guarantee the vCPU received an IPI and has actually exited guest mode.
168  */
169 #define KVM_REQ_OUTSIDE_GUEST_MODE	(KVM_REQUEST_NO_ACTION | KVM_REQUEST_WAIT | KVM_REQUEST_NO_WAKEUP)
170 
171 #define KVM_ARCH_REQ_FLAGS(nr, flags) ({ \
172 	BUILD_BUG_ON((unsigned)(nr) >= (sizeof_field(struct kvm_vcpu, requests) * 8) - KVM_REQUEST_ARCH_BASE); \
173 	(unsigned)(((nr) + KVM_REQUEST_ARCH_BASE) | (flags)); \
174 })
175 #define KVM_ARCH_REQ(nr)           KVM_ARCH_REQ_FLAGS(nr, 0)
176 
177 bool kvm_make_vcpus_request_mask(struct kvm *kvm, unsigned int req,
178 				 unsigned long *vcpu_bitmap);
179 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req);
180 bool kvm_make_all_cpus_request_except(struct kvm *kvm, unsigned int req,
181 				      struct kvm_vcpu *except);
182 bool kvm_make_cpus_request_mask(struct kvm *kvm, unsigned int req,
183 				unsigned long *vcpu_bitmap);
184 
185 #define KVM_USERSPACE_IRQ_SOURCE_ID		0
186 #define KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID	1
187 
188 extern struct mutex kvm_lock;
189 extern struct list_head vm_list;
190 
191 struct kvm_io_range {
192 	gpa_t addr;
193 	int len;
194 	struct kvm_io_device *dev;
195 };
196 
197 #define NR_IOBUS_DEVS 1000
198 
199 struct kvm_io_bus {
200 	int dev_count;
201 	int ioeventfd_count;
202 	struct kvm_io_range range[];
203 };
204 
205 enum kvm_bus {
206 	KVM_MMIO_BUS,
207 	KVM_PIO_BUS,
208 	KVM_VIRTIO_CCW_NOTIFY_BUS,
209 	KVM_FAST_MMIO_BUS,
210 	KVM_NR_BUSES
211 };
212 
213 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
214 		     int len, const void *val);
215 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
216 			    gpa_t addr, int len, const void *val, long cookie);
217 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
218 		    int len, void *val);
219 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
220 			    int len, struct kvm_io_device *dev);
221 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
222 			      struct kvm_io_device *dev);
223 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
224 					 gpa_t addr);
225 
226 #ifdef CONFIG_KVM_ASYNC_PF
227 struct kvm_async_pf {
228 	struct work_struct work;
229 	struct list_head link;
230 	struct list_head queue;
231 	struct kvm_vcpu *vcpu;
232 	struct mm_struct *mm;
233 	gpa_t cr2_or_gpa;
234 	unsigned long addr;
235 	struct kvm_arch_async_pf arch;
236 	bool   wakeup_all;
237 	bool notpresent_injected;
238 };
239 
240 void kvm_clear_async_pf_completion_queue(struct kvm_vcpu *vcpu);
241 void kvm_check_async_pf_completion(struct kvm_vcpu *vcpu);
242 bool kvm_setup_async_pf(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
243 			unsigned long hva, struct kvm_arch_async_pf *arch);
244 int kvm_async_pf_wakeup_all(struct kvm_vcpu *vcpu);
245 #endif
246 
247 #ifdef KVM_ARCH_WANT_MMU_NOTIFIER
248 struct kvm_gfn_range {
249 	struct kvm_memory_slot *slot;
250 	gfn_t start;
251 	gfn_t end;
252 	pte_t pte;
253 	bool may_block;
254 };
255 bool kvm_unmap_gfn_range(struct kvm *kvm, struct kvm_gfn_range *range);
256 bool kvm_age_gfn(struct kvm *kvm, struct kvm_gfn_range *range);
257 bool kvm_test_age_gfn(struct kvm *kvm, struct kvm_gfn_range *range);
258 bool kvm_set_spte_gfn(struct kvm *kvm, struct kvm_gfn_range *range);
259 #endif
260 
261 enum {
262 	OUTSIDE_GUEST_MODE,
263 	IN_GUEST_MODE,
264 	EXITING_GUEST_MODE,
265 	READING_SHADOW_PAGE_TABLES,
266 };
267 
268 #define KVM_UNMAPPED_PAGE	((void *) 0x500 + POISON_POINTER_DELTA)
269 
270 struct kvm_host_map {
271 	/*
272 	 * Only valid if the 'pfn' is managed by the host kernel (i.e. There is
273 	 * a 'struct page' for it. When using mem= kernel parameter some memory
274 	 * can be used as guest memory but they are not managed by host
275 	 * kernel).
276 	 * If 'pfn' is not managed by the host kernel, this field is
277 	 * initialized to KVM_UNMAPPED_PAGE.
278 	 */
279 	struct page *page;
280 	void *hva;
281 	kvm_pfn_t pfn;
282 	kvm_pfn_t gfn;
283 };
284 
285 /*
286  * Used to check if the mapping is valid or not. Never use 'kvm_host_map'
287  * directly to check for that.
288  */
kvm_vcpu_mapped(struct kvm_host_map * map)289 static inline bool kvm_vcpu_mapped(struct kvm_host_map *map)
290 {
291 	return !!map->hva;
292 }
293 
kvm_vcpu_can_poll(ktime_t cur,ktime_t stop)294 static inline bool kvm_vcpu_can_poll(ktime_t cur, ktime_t stop)
295 {
296 	return single_task_running() && !need_resched() && ktime_before(cur, stop);
297 }
298 
299 /*
300  * Sometimes a large or cross-page mmio needs to be broken up into separate
301  * exits for userspace servicing.
302  */
303 struct kvm_mmio_fragment {
304 	gpa_t gpa;
305 	void *data;
306 	unsigned len;
307 };
308 
309 struct kvm_vcpu {
310 	struct kvm *kvm;
311 #ifdef CONFIG_PREEMPT_NOTIFIERS
312 	struct preempt_notifier preempt_notifier;
313 #endif
314 	int cpu;
315 	int vcpu_id; /* id given by userspace at creation */
316 	int vcpu_idx; /* index in kvm->vcpus array */
317 	int ____srcu_idx; /* Don't use this directly.  You've been warned. */
318 #ifdef CONFIG_PROVE_RCU
319 	int srcu_depth;
320 #endif
321 	int mode;
322 	u64 requests;
323 	unsigned long guest_debug;
324 
325 	struct mutex mutex;
326 	struct kvm_run *run;
327 
328 #ifndef __KVM_HAVE_ARCH_WQP
329 	struct rcuwait wait;
330 #endif
331 	struct pid __rcu *pid;
332 	int sigset_active;
333 	sigset_t sigset;
334 	unsigned int halt_poll_ns;
335 	bool valid_wakeup;
336 
337 #ifdef CONFIG_HAS_IOMEM
338 	int mmio_needed;
339 	int mmio_read_completed;
340 	int mmio_is_write;
341 	int mmio_cur_fragment;
342 	int mmio_nr_fragments;
343 	struct kvm_mmio_fragment mmio_fragments[KVM_MAX_MMIO_FRAGMENTS];
344 #endif
345 
346 #ifdef CONFIG_KVM_ASYNC_PF
347 	struct {
348 		u32 queued;
349 		struct list_head queue;
350 		struct list_head done;
351 		spinlock_t lock;
352 	} async_pf;
353 #endif
354 
355 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
356 	/*
357 	 * Cpu relax intercept or pause loop exit optimization
358 	 * in_spin_loop: set when a vcpu does a pause loop exit
359 	 *  or cpu relax intercepted.
360 	 * dy_eligible: indicates whether vcpu is eligible for directed yield.
361 	 */
362 	struct {
363 		bool in_spin_loop;
364 		bool dy_eligible;
365 	} spin_loop;
366 #endif
367 	bool preempted;
368 	bool ready;
369 	struct kvm_vcpu_arch arch;
370 	struct kvm_vcpu_stat stat;
371 	char stats_id[KVM_STATS_NAME_SIZE];
372 	struct kvm_dirty_ring dirty_ring;
373 
374 	/*
375 	 * The most recently used memslot by this vCPU and the slots generation
376 	 * for which it is valid.
377 	 * No wraparound protection is needed since generations won't overflow in
378 	 * thousands of years, even assuming 1M memslot operations per second.
379 	 */
380 	struct kvm_memory_slot *last_used_slot;
381 	u64 last_used_slot_gen;
382 };
383 
384 /*
385  * Start accounting time towards a guest.
386  * Must be called before entering guest context.
387  */
guest_timing_enter_irqoff(void)388 static __always_inline void guest_timing_enter_irqoff(void)
389 {
390 	/*
391 	 * This is running in ioctl context so its safe to assume that it's the
392 	 * stime pending cputime to flush.
393 	 */
394 	instrumentation_begin();
395 	vtime_account_guest_enter();
396 	instrumentation_end();
397 }
398 
399 /*
400  * Enter guest context and enter an RCU extended quiescent state.
401  *
402  * Between guest_context_enter_irqoff() and guest_context_exit_irqoff() it is
403  * unsafe to use any code which may directly or indirectly use RCU, tracing
404  * (including IRQ flag tracing), or lockdep. All code in this period must be
405  * non-instrumentable.
406  */
guest_context_enter_irqoff(void)407 static __always_inline void guest_context_enter_irqoff(void)
408 {
409 	/*
410 	 * KVM does not hold any references to rcu protected data when it
411 	 * switches CPU into a guest mode. In fact switching to a guest mode
412 	 * is very similar to exiting to userspace from rcu point of view. In
413 	 * addition CPU may stay in a guest mode for quite a long time (up to
414 	 * one time slice). Lets treat guest mode as quiescent state, just like
415 	 * we do with user-mode execution.
416 	 */
417 	if (!context_tracking_guest_enter()) {
418 		instrumentation_begin();
419 		rcu_virt_note_context_switch(smp_processor_id());
420 		instrumentation_end();
421 	}
422 }
423 
424 /*
425  * Deprecated. Architectures should move to guest_timing_enter_irqoff() and
426  * guest_state_enter_irqoff().
427  */
guest_enter_irqoff(void)428 static __always_inline void guest_enter_irqoff(void)
429 {
430 	guest_timing_enter_irqoff();
431 	guest_context_enter_irqoff();
432 }
433 
434 /**
435  * guest_state_enter_irqoff - Fixup state when entering a guest
436  *
437  * Entry to a guest will enable interrupts, but the kernel state is interrupts
438  * disabled when this is invoked. Also tell RCU about it.
439  *
440  * 1) Trace interrupts on state
441  * 2) Invoke context tracking if enabled to adjust RCU state
442  * 3) Tell lockdep that interrupts are enabled
443  *
444  * Invoked from architecture specific code before entering a guest.
445  * Must be called with interrupts disabled and the caller must be
446  * non-instrumentable.
447  * The caller has to invoke guest_timing_enter_irqoff() before this.
448  *
449  * Note: this is analogous to exit_to_user_mode().
450  */
guest_state_enter_irqoff(void)451 static __always_inline void guest_state_enter_irqoff(void)
452 {
453 	instrumentation_begin();
454 	trace_hardirqs_on_prepare();
455 	lockdep_hardirqs_on_prepare();
456 	instrumentation_end();
457 
458 	guest_context_enter_irqoff();
459 	lockdep_hardirqs_on(CALLER_ADDR0);
460 }
461 
462 /*
463  * Exit guest context and exit an RCU extended quiescent state.
464  *
465  * Between guest_context_enter_irqoff() and guest_context_exit_irqoff() it is
466  * unsafe to use any code which may directly or indirectly use RCU, tracing
467  * (including IRQ flag tracing), or lockdep. All code in this period must be
468  * non-instrumentable.
469  */
guest_context_exit_irqoff(void)470 static __always_inline void guest_context_exit_irqoff(void)
471 {
472 	context_tracking_guest_exit();
473 }
474 
475 /*
476  * Stop accounting time towards a guest.
477  * Must be called after exiting guest context.
478  */
guest_timing_exit_irqoff(void)479 static __always_inline void guest_timing_exit_irqoff(void)
480 {
481 	instrumentation_begin();
482 	/* Flush the guest cputime we spent on the guest */
483 	vtime_account_guest_exit();
484 	instrumentation_end();
485 }
486 
487 /*
488  * Deprecated. Architectures should move to guest_state_exit_irqoff() and
489  * guest_timing_exit_irqoff().
490  */
guest_exit_irqoff(void)491 static __always_inline void guest_exit_irqoff(void)
492 {
493 	guest_context_exit_irqoff();
494 	guest_timing_exit_irqoff();
495 }
496 
guest_exit(void)497 static inline void guest_exit(void)
498 {
499 	unsigned long flags;
500 
501 	local_irq_save(flags);
502 	guest_exit_irqoff();
503 	local_irq_restore(flags);
504 }
505 
506 /**
507  * guest_state_exit_irqoff - Establish state when returning from guest mode
508  *
509  * Entry from a guest disables interrupts, but guest mode is traced as
510  * interrupts enabled. Also with NO_HZ_FULL RCU might be idle.
511  *
512  * 1) Tell lockdep that interrupts are disabled
513  * 2) Invoke context tracking if enabled to reactivate RCU
514  * 3) Trace interrupts off state
515  *
516  * Invoked from architecture specific code after exiting a guest.
517  * Must be invoked with interrupts disabled and the caller must be
518  * non-instrumentable.
519  * The caller has to invoke guest_timing_exit_irqoff() after this.
520  *
521  * Note: this is analogous to enter_from_user_mode().
522  */
guest_state_exit_irqoff(void)523 static __always_inline void guest_state_exit_irqoff(void)
524 {
525 	lockdep_hardirqs_off(CALLER_ADDR0);
526 	guest_context_exit_irqoff();
527 
528 	instrumentation_begin();
529 	trace_hardirqs_off_finish();
530 	instrumentation_end();
531 }
532 
kvm_vcpu_exiting_guest_mode(struct kvm_vcpu * vcpu)533 static inline int kvm_vcpu_exiting_guest_mode(struct kvm_vcpu *vcpu)
534 {
535 	/*
536 	 * The memory barrier ensures a previous write to vcpu->requests cannot
537 	 * be reordered with the read of vcpu->mode.  It pairs with the general
538 	 * memory barrier following the write of vcpu->mode in VCPU RUN.
539 	 */
540 	smp_mb__before_atomic();
541 	return cmpxchg(&vcpu->mode, IN_GUEST_MODE, EXITING_GUEST_MODE);
542 }
543 
544 /*
545  * Some of the bitops functions do not support too long bitmaps.
546  * This number must be determined not to exceed such limits.
547  */
548 #define KVM_MEM_MAX_NR_PAGES ((1UL << 31) - 1)
549 
550 /*
551  * Since at idle each memslot belongs to two memslot sets it has to contain
552  * two embedded nodes for each data structure that it forms a part of.
553  *
554  * Two memslot sets (one active and one inactive) are necessary so the VM
555  * continues to run on one memslot set while the other is being modified.
556  *
557  * These two memslot sets normally point to the same set of memslots.
558  * They can, however, be desynchronized when performing a memslot management
559  * operation by replacing the memslot to be modified by its copy.
560  * After the operation is complete, both memslot sets once again point to
561  * the same, common set of memslot data.
562  *
563  * The memslots themselves are independent of each other so they can be
564  * individually added or deleted.
565  */
566 struct kvm_memory_slot {
567 	struct hlist_node id_node[2];
568 	struct interval_tree_node hva_node[2];
569 	struct rb_node gfn_node[2];
570 	gfn_t base_gfn;
571 	unsigned long npages;
572 	unsigned long *dirty_bitmap;
573 	struct kvm_arch_memory_slot arch;
574 	unsigned long userspace_addr;
575 	u32 flags;
576 	short id;
577 	u16 as_id;
578 };
579 
kvm_slot_dirty_track_enabled(const struct kvm_memory_slot * slot)580 static inline bool kvm_slot_dirty_track_enabled(const struct kvm_memory_slot *slot)
581 {
582 	return slot->flags & KVM_MEM_LOG_DIRTY_PAGES;
583 }
584 
kvm_dirty_bitmap_bytes(struct kvm_memory_slot * memslot)585 static inline unsigned long kvm_dirty_bitmap_bytes(struct kvm_memory_slot *memslot)
586 {
587 	return ALIGN(memslot->npages, BITS_PER_LONG) / 8;
588 }
589 
kvm_second_dirty_bitmap(struct kvm_memory_slot * memslot)590 static inline unsigned long *kvm_second_dirty_bitmap(struct kvm_memory_slot *memslot)
591 {
592 	unsigned long len = kvm_dirty_bitmap_bytes(memslot);
593 
594 	return memslot->dirty_bitmap + len / sizeof(*memslot->dirty_bitmap);
595 }
596 
597 #ifndef KVM_DIRTY_LOG_MANUAL_CAPS
598 #define KVM_DIRTY_LOG_MANUAL_CAPS KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE
599 #endif
600 
601 struct kvm_s390_adapter_int {
602 	u64 ind_addr;
603 	u64 summary_addr;
604 	u64 ind_offset;
605 	u32 summary_offset;
606 	u32 adapter_id;
607 };
608 
609 struct kvm_hv_sint {
610 	u32 vcpu;
611 	u32 sint;
612 };
613 
614 struct kvm_xen_evtchn {
615 	u32 port;
616 	u32 vcpu_id;
617 	int vcpu_idx;
618 	u32 priority;
619 };
620 
621 struct kvm_kernel_irq_routing_entry {
622 	u32 gsi;
623 	u32 type;
624 	int (*set)(struct kvm_kernel_irq_routing_entry *e,
625 		   struct kvm *kvm, int irq_source_id, int level,
626 		   bool line_status);
627 	union {
628 		struct {
629 			unsigned irqchip;
630 			unsigned pin;
631 		} irqchip;
632 		struct {
633 			u32 address_lo;
634 			u32 address_hi;
635 			u32 data;
636 			u32 flags;
637 			u32 devid;
638 		} msi;
639 		struct kvm_s390_adapter_int adapter;
640 		struct kvm_hv_sint hv_sint;
641 		struct kvm_xen_evtchn xen_evtchn;
642 	};
643 	struct hlist_node link;
644 };
645 
646 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
647 struct kvm_irq_routing_table {
648 	int chip[KVM_NR_IRQCHIPS][KVM_IRQCHIP_NUM_PINS];
649 	u32 nr_rt_entries;
650 	/*
651 	 * Array indexed by gsi. Each entry contains list of irq chips
652 	 * the gsi is connected to.
653 	 */
654 	struct hlist_head map[];
655 };
656 #endif
657 
658 #ifndef KVM_INTERNAL_MEM_SLOTS
659 #define KVM_INTERNAL_MEM_SLOTS 0
660 #endif
661 
662 #define KVM_MEM_SLOTS_NUM SHRT_MAX
663 #define KVM_USER_MEM_SLOTS (KVM_MEM_SLOTS_NUM - KVM_INTERNAL_MEM_SLOTS)
664 
665 #ifndef __KVM_VCPU_MULTIPLE_ADDRESS_SPACE
kvm_arch_vcpu_memslots_id(struct kvm_vcpu * vcpu)666 static inline int kvm_arch_vcpu_memslots_id(struct kvm_vcpu *vcpu)
667 {
668 	return 0;
669 }
670 #endif
671 
672 struct kvm_memslots {
673 	u64 generation;
674 	atomic_long_t last_used_slot;
675 	struct rb_root_cached hva_tree;
676 	struct rb_root gfn_tree;
677 	/*
678 	 * The mapping table from slot id to memslot.
679 	 *
680 	 * 7-bit bucket count matches the size of the old id to index array for
681 	 * 512 slots, while giving good performance with this slot count.
682 	 * Higher bucket counts bring only small performance improvements but
683 	 * always result in higher memory usage (even for lower memslot counts).
684 	 */
685 	DECLARE_HASHTABLE(id_hash, 7);
686 	int node_idx;
687 };
688 
689 struct kvm {
690 #ifdef KVM_HAVE_MMU_RWLOCK
691 	rwlock_t mmu_lock;
692 #else
693 	spinlock_t mmu_lock;
694 #endif /* KVM_HAVE_MMU_RWLOCK */
695 
696 	struct mutex slots_lock;
697 
698 	/*
699 	 * Protects the arch-specific fields of struct kvm_memory_slots in
700 	 * use by the VM. To be used under the slots_lock (above) or in a
701 	 * kvm->srcu critical section where acquiring the slots_lock would
702 	 * lead to deadlock with the synchronize_srcu in
703 	 * install_new_memslots.
704 	 */
705 	struct mutex slots_arch_lock;
706 	struct mm_struct *mm; /* userspace tied to this vm */
707 	unsigned long nr_memslot_pages;
708 	/* The two memslot sets - active and inactive (per address space) */
709 	struct kvm_memslots __memslots[KVM_ADDRESS_SPACE_NUM][2];
710 	/* The current active memslot set for each address space */
711 	struct kvm_memslots __rcu *memslots[KVM_ADDRESS_SPACE_NUM];
712 	struct xarray vcpu_array;
713 
714 	/* Used to wait for completion of MMU notifiers.  */
715 	spinlock_t mn_invalidate_lock;
716 	unsigned long mn_active_invalidate_count;
717 	struct rcuwait mn_memslots_update_rcuwait;
718 
719 	/* For management / invalidation of gfn_to_pfn_caches */
720 	spinlock_t gpc_lock;
721 	struct list_head gpc_list;
722 
723 	/*
724 	 * created_vcpus is protected by kvm->lock, and is incremented
725 	 * at the beginning of KVM_CREATE_VCPU.  online_vcpus is only
726 	 * incremented after storing the kvm_vcpu pointer in vcpus,
727 	 * and is accessed atomically.
728 	 */
729 	atomic_t online_vcpus;
730 	int max_vcpus;
731 	int created_vcpus;
732 	int last_boosted_vcpu;
733 	struct list_head vm_list;
734 	struct mutex lock;
735 	struct kvm_io_bus __rcu *buses[KVM_NR_BUSES];
736 #ifdef CONFIG_HAVE_KVM_EVENTFD
737 	struct {
738 		spinlock_t        lock;
739 		struct list_head  items;
740 		struct list_head  resampler_list;
741 		struct mutex      resampler_lock;
742 	} irqfds;
743 	struct list_head ioeventfds;
744 #endif
745 	struct kvm_vm_stat stat;
746 	struct kvm_arch arch;
747 	refcount_t users_count;
748 #ifdef CONFIG_KVM_MMIO
749 	struct kvm_coalesced_mmio_ring *coalesced_mmio_ring;
750 	spinlock_t ring_lock;
751 	struct list_head coalesced_zones;
752 #endif
753 
754 	struct mutex irq_lock;
755 #ifdef CONFIG_HAVE_KVM_IRQCHIP
756 	/*
757 	 * Update side is protected by irq_lock.
758 	 */
759 	struct kvm_irq_routing_table __rcu *irq_routing;
760 #endif
761 #ifdef CONFIG_HAVE_KVM_IRQFD
762 	struct hlist_head irq_ack_notifier_list;
763 #endif
764 
765 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
766 	struct mmu_notifier mmu_notifier;
767 	unsigned long mmu_invalidate_seq;
768 	long mmu_invalidate_in_progress;
769 	unsigned long mmu_invalidate_range_start;
770 	unsigned long mmu_invalidate_range_end;
771 #endif
772 	struct list_head devices;
773 	u64 manual_dirty_log_protect;
774 	struct dentry *debugfs_dentry;
775 	struct kvm_stat_data **debugfs_stat_data;
776 	struct srcu_struct srcu;
777 	struct srcu_struct irq_srcu;
778 	pid_t userspace_pid;
779 	bool override_halt_poll_ns;
780 	unsigned int max_halt_poll_ns;
781 	u32 dirty_ring_size;
782 	bool vm_bugged;
783 	bool vm_dead;
784 
785 #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER
786 	struct notifier_block pm_notifier;
787 #endif
788 	char stats_id[KVM_STATS_NAME_SIZE];
789 };
790 
791 #define kvm_err(fmt, ...) \
792 	pr_err("kvm [%i]: " fmt, task_pid_nr(current), ## __VA_ARGS__)
793 #define kvm_info(fmt, ...) \
794 	pr_info("kvm [%i]: " fmt, task_pid_nr(current), ## __VA_ARGS__)
795 #define kvm_debug(fmt, ...) \
796 	pr_debug("kvm [%i]: " fmt, task_pid_nr(current), ## __VA_ARGS__)
797 #define kvm_debug_ratelimited(fmt, ...) \
798 	pr_debug_ratelimited("kvm [%i]: " fmt, task_pid_nr(current), \
799 			     ## __VA_ARGS__)
800 #define kvm_pr_unimpl(fmt, ...) \
801 	pr_err_ratelimited("kvm [%i]: " fmt, \
802 			   task_tgid_nr(current), ## __VA_ARGS__)
803 
804 /* The guest did something we don't support. */
805 #define vcpu_unimpl(vcpu, fmt, ...)					\
806 	kvm_pr_unimpl("vcpu%i, guest rIP: 0x%lx " fmt,			\
807 			(vcpu)->vcpu_id, kvm_rip_read(vcpu), ## __VA_ARGS__)
808 
809 #define vcpu_debug(vcpu, fmt, ...)					\
810 	kvm_debug("vcpu%i " fmt, (vcpu)->vcpu_id, ## __VA_ARGS__)
811 #define vcpu_debug_ratelimited(vcpu, fmt, ...)				\
812 	kvm_debug_ratelimited("vcpu%i " fmt, (vcpu)->vcpu_id,           \
813 			      ## __VA_ARGS__)
814 #define vcpu_err(vcpu, fmt, ...)					\
815 	kvm_err("vcpu%i " fmt, (vcpu)->vcpu_id, ## __VA_ARGS__)
816 
kvm_vm_dead(struct kvm * kvm)817 static inline void kvm_vm_dead(struct kvm *kvm)
818 {
819 	kvm->vm_dead = true;
820 	kvm_make_all_cpus_request(kvm, KVM_REQ_VM_DEAD);
821 }
822 
kvm_vm_bugged(struct kvm * kvm)823 static inline void kvm_vm_bugged(struct kvm *kvm)
824 {
825 	kvm->vm_bugged = true;
826 	kvm_vm_dead(kvm);
827 }
828 
829 
830 #define KVM_BUG(cond, kvm, fmt...)				\
831 ({								\
832 	int __ret = (cond);					\
833 								\
834 	if (WARN_ONCE(__ret && !(kvm)->vm_bugged, fmt))		\
835 		kvm_vm_bugged(kvm);				\
836 	unlikely(__ret);					\
837 })
838 
839 #define KVM_BUG_ON(cond, kvm)					\
840 ({								\
841 	int __ret = (cond);					\
842 								\
843 	if (WARN_ON_ONCE(__ret && !(kvm)->vm_bugged))		\
844 		kvm_vm_bugged(kvm);				\
845 	unlikely(__ret);					\
846 })
847 
kvm_vcpu_srcu_read_lock(struct kvm_vcpu * vcpu)848 static inline void kvm_vcpu_srcu_read_lock(struct kvm_vcpu *vcpu)
849 {
850 #ifdef CONFIG_PROVE_RCU
851 	WARN_ONCE(vcpu->srcu_depth++,
852 		  "KVM: Illegal vCPU srcu_idx LOCK, depth=%d", vcpu->srcu_depth - 1);
853 #endif
854 	vcpu->____srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
855 }
856 
kvm_vcpu_srcu_read_unlock(struct kvm_vcpu * vcpu)857 static inline void kvm_vcpu_srcu_read_unlock(struct kvm_vcpu *vcpu)
858 {
859 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->____srcu_idx);
860 
861 #ifdef CONFIG_PROVE_RCU
862 	WARN_ONCE(--vcpu->srcu_depth,
863 		  "KVM: Illegal vCPU srcu_idx UNLOCK, depth=%d", vcpu->srcu_depth);
864 #endif
865 }
866 
kvm_dirty_log_manual_protect_and_init_set(struct kvm * kvm)867 static inline bool kvm_dirty_log_manual_protect_and_init_set(struct kvm *kvm)
868 {
869 	return !!(kvm->manual_dirty_log_protect & KVM_DIRTY_LOG_INITIALLY_SET);
870 }
871 
kvm_get_bus(struct kvm * kvm,enum kvm_bus idx)872 static inline struct kvm_io_bus *kvm_get_bus(struct kvm *kvm, enum kvm_bus idx)
873 {
874 	return srcu_dereference_check(kvm->buses[idx], &kvm->srcu,
875 				      lockdep_is_held(&kvm->slots_lock) ||
876 				      !refcount_read(&kvm->users_count));
877 }
878 
kvm_get_vcpu(struct kvm * kvm,int i)879 static inline struct kvm_vcpu *kvm_get_vcpu(struct kvm *kvm, int i)
880 {
881 	int num_vcpus = atomic_read(&kvm->online_vcpus);
882 	i = array_index_nospec(i, num_vcpus);
883 
884 	/* Pairs with smp_wmb() in kvm_vm_ioctl_create_vcpu.  */
885 	smp_rmb();
886 	return xa_load(&kvm->vcpu_array, i);
887 }
888 
889 #define kvm_for_each_vcpu(idx, vcpup, kvm)		   \
890 	xa_for_each_range(&kvm->vcpu_array, idx, vcpup, 0, \
891 			  (atomic_read(&kvm->online_vcpus) - 1))
892 
kvm_get_vcpu_by_id(struct kvm * kvm,int id)893 static inline struct kvm_vcpu *kvm_get_vcpu_by_id(struct kvm *kvm, int id)
894 {
895 	struct kvm_vcpu *vcpu = NULL;
896 	unsigned long i;
897 
898 	if (id < 0)
899 		return NULL;
900 	if (id < KVM_MAX_VCPUS)
901 		vcpu = kvm_get_vcpu(kvm, id);
902 	if (vcpu && vcpu->vcpu_id == id)
903 		return vcpu;
904 	kvm_for_each_vcpu(i, vcpu, kvm)
905 		if (vcpu->vcpu_id == id)
906 			return vcpu;
907 	return NULL;
908 }
909 
910 void kvm_destroy_vcpus(struct kvm *kvm);
911 
912 void vcpu_load(struct kvm_vcpu *vcpu);
913 void vcpu_put(struct kvm_vcpu *vcpu);
914 
915 #ifdef __KVM_HAVE_IOAPIC
916 void kvm_arch_post_irq_ack_notifier_list_update(struct kvm *kvm);
917 void kvm_arch_post_irq_routing_update(struct kvm *kvm);
918 #else
kvm_arch_post_irq_ack_notifier_list_update(struct kvm * kvm)919 static inline void kvm_arch_post_irq_ack_notifier_list_update(struct kvm *kvm)
920 {
921 }
kvm_arch_post_irq_routing_update(struct kvm * kvm)922 static inline void kvm_arch_post_irq_routing_update(struct kvm *kvm)
923 {
924 }
925 #endif
926 
927 #ifdef CONFIG_HAVE_KVM_IRQFD
928 int kvm_irqfd_init(void);
929 void kvm_irqfd_exit(void);
930 #else
kvm_irqfd_init(void)931 static inline int kvm_irqfd_init(void)
932 {
933 	return 0;
934 }
935 
kvm_irqfd_exit(void)936 static inline void kvm_irqfd_exit(void)
937 {
938 }
939 #endif
940 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
941 		  struct module *module);
942 void kvm_exit(void);
943 
944 void kvm_get_kvm(struct kvm *kvm);
945 bool kvm_get_kvm_safe(struct kvm *kvm);
946 void kvm_put_kvm(struct kvm *kvm);
947 bool file_is_kvm(struct file *file);
948 void kvm_put_kvm_no_destroy(struct kvm *kvm);
949 
__kvm_memslots(struct kvm * kvm,int as_id)950 static inline struct kvm_memslots *__kvm_memslots(struct kvm *kvm, int as_id)
951 {
952 	as_id = array_index_nospec(as_id, KVM_ADDRESS_SPACE_NUM);
953 	return srcu_dereference_check(kvm->memslots[as_id], &kvm->srcu,
954 			lockdep_is_held(&kvm->slots_lock) ||
955 			!refcount_read(&kvm->users_count));
956 }
957 
kvm_memslots(struct kvm * kvm)958 static inline struct kvm_memslots *kvm_memslots(struct kvm *kvm)
959 {
960 	return __kvm_memslots(kvm, 0);
961 }
962 
kvm_vcpu_memslots(struct kvm_vcpu * vcpu)963 static inline struct kvm_memslots *kvm_vcpu_memslots(struct kvm_vcpu *vcpu)
964 {
965 	int as_id = kvm_arch_vcpu_memslots_id(vcpu);
966 
967 	return __kvm_memslots(vcpu->kvm, as_id);
968 }
969 
kvm_memslots_empty(struct kvm_memslots * slots)970 static inline bool kvm_memslots_empty(struct kvm_memslots *slots)
971 {
972 	return RB_EMPTY_ROOT(&slots->gfn_tree);
973 }
974 
975 #define kvm_for_each_memslot(memslot, bkt, slots)			      \
976 	hash_for_each(slots->id_hash, bkt, memslot, id_node[slots->node_idx]) \
977 		if (WARN_ON_ONCE(!memslot->npages)) {			      \
978 		} else
979 
980 static inline
id_to_memslot(struct kvm_memslots * slots,int id)981 struct kvm_memory_slot *id_to_memslot(struct kvm_memslots *slots, int id)
982 {
983 	struct kvm_memory_slot *slot;
984 	int idx = slots->node_idx;
985 
986 	hash_for_each_possible(slots->id_hash, slot, id_node[idx], id) {
987 		if (slot->id == id)
988 			return slot;
989 	}
990 
991 	return NULL;
992 }
993 
994 /* Iterator used for walking memslots that overlap a gfn range. */
995 struct kvm_memslot_iter {
996 	struct kvm_memslots *slots;
997 	struct rb_node *node;
998 	struct kvm_memory_slot *slot;
999 };
1000 
kvm_memslot_iter_next(struct kvm_memslot_iter * iter)1001 static inline void kvm_memslot_iter_next(struct kvm_memslot_iter *iter)
1002 {
1003 	iter->node = rb_next(iter->node);
1004 	if (!iter->node)
1005 		return;
1006 
1007 	iter->slot = container_of(iter->node, struct kvm_memory_slot, gfn_node[iter->slots->node_idx]);
1008 }
1009 
kvm_memslot_iter_start(struct kvm_memslot_iter * iter,struct kvm_memslots * slots,gfn_t start)1010 static inline void kvm_memslot_iter_start(struct kvm_memslot_iter *iter,
1011 					  struct kvm_memslots *slots,
1012 					  gfn_t start)
1013 {
1014 	int idx = slots->node_idx;
1015 	struct rb_node *tmp;
1016 	struct kvm_memory_slot *slot;
1017 
1018 	iter->slots = slots;
1019 
1020 	/*
1021 	 * Find the so called "upper bound" of a key - the first node that has
1022 	 * its key strictly greater than the searched one (the start gfn in our case).
1023 	 */
1024 	iter->node = NULL;
1025 	for (tmp = slots->gfn_tree.rb_node; tmp; ) {
1026 		slot = container_of(tmp, struct kvm_memory_slot, gfn_node[idx]);
1027 		if (start < slot->base_gfn) {
1028 			iter->node = tmp;
1029 			tmp = tmp->rb_left;
1030 		} else {
1031 			tmp = tmp->rb_right;
1032 		}
1033 	}
1034 
1035 	/*
1036 	 * Find the slot with the lowest gfn that can possibly intersect with
1037 	 * the range, so we'll ideally have slot start <= range start
1038 	 */
1039 	if (iter->node) {
1040 		/*
1041 		 * A NULL previous node means that the very first slot
1042 		 * already has a higher start gfn.
1043 		 * In this case slot start > range start.
1044 		 */
1045 		tmp = rb_prev(iter->node);
1046 		if (tmp)
1047 			iter->node = tmp;
1048 	} else {
1049 		/* a NULL node below means no slots */
1050 		iter->node = rb_last(&slots->gfn_tree);
1051 	}
1052 
1053 	if (iter->node) {
1054 		iter->slot = container_of(iter->node, struct kvm_memory_slot, gfn_node[idx]);
1055 
1056 		/*
1057 		 * It is possible in the slot start < range start case that the
1058 		 * found slot ends before or at range start (slot end <= range start)
1059 		 * and so it does not overlap the requested range.
1060 		 *
1061 		 * In such non-overlapping case the next slot (if it exists) will
1062 		 * already have slot start > range start, otherwise the logic above
1063 		 * would have found it instead of the current slot.
1064 		 */
1065 		if (iter->slot->base_gfn + iter->slot->npages <= start)
1066 			kvm_memslot_iter_next(iter);
1067 	}
1068 }
1069 
kvm_memslot_iter_is_valid(struct kvm_memslot_iter * iter,gfn_t end)1070 static inline bool kvm_memslot_iter_is_valid(struct kvm_memslot_iter *iter, gfn_t end)
1071 {
1072 	if (!iter->node)
1073 		return false;
1074 
1075 	/*
1076 	 * If this slot starts beyond or at the end of the range so does
1077 	 * every next one
1078 	 */
1079 	return iter->slot->base_gfn < end;
1080 }
1081 
1082 /* Iterate over each memslot at least partially intersecting [start, end) range */
1083 #define kvm_for_each_memslot_in_gfn_range(iter, slots, start, end)	\
1084 	for (kvm_memslot_iter_start(iter, slots, start);		\
1085 	     kvm_memslot_iter_is_valid(iter, end);			\
1086 	     kvm_memslot_iter_next(iter))
1087 
1088 /*
1089  * KVM_SET_USER_MEMORY_REGION ioctl allows the following operations:
1090  * - create a new memory slot
1091  * - delete an existing memory slot
1092  * - modify an existing memory slot
1093  *   -- move it in the guest physical memory space
1094  *   -- just change its flags
1095  *
1096  * Since flags can be changed by some of these operations, the following
1097  * differentiation is the best we can do for __kvm_set_memory_region():
1098  */
1099 enum kvm_mr_change {
1100 	KVM_MR_CREATE,
1101 	KVM_MR_DELETE,
1102 	KVM_MR_MOVE,
1103 	KVM_MR_FLAGS_ONLY,
1104 };
1105 
1106 int kvm_set_memory_region(struct kvm *kvm,
1107 			  const struct kvm_userspace_memory_region *mem);
1108 int __kvm_set_memory_region(struct kvm *kvm,
1109 			    const struct kvm_userspace_memory_region *mem);
1110 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot);
1111 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen);
1112 int kvm_arch_prepare_memory_region(struct kvm *kvm,
1113 				const struct kvm_memory_slot *old,
1114 				struct kvm_memory_slot *new,
1115 				enum kvm_mr_change change);
1116 void kvm_arch_commit_memory_region(struct kvm *kvm,
1117 				struct kvm_memory_slot *old,
1118 				const struct kvm_memory_slot *new,
1119 				enum kvm_mr_change change);
1120 /* flush all memory translations */
1121 void kvm_arch_flush_shadow_all(struct kvm *kvm);
1122 /* flush memory translations pointing to 'slot' */
1123 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
1124 				   struct kvm_memory_slot *slot);
1125 
1126 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1127 			    struct page **pages, int nr_pages);
1128 
1129 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn);
1130 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn);
1131 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable);
1132 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot, gfn_t gfn);
1133 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot, gfn_t gfn,
1134 				      bool *writable);
1135 void kvm_release_page_clean(struct page *page);
1136 void kvm_release_page_dirty(struct page *page);
1137 
1138 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn);
1139 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1140 		      bool *writable);
1141 kvm_pfn_t gfn_to_pfn_memslot(const struct kvm_memory_slot *slot, gfn_t gfn);
1142 kvm_pfn_t gfn_to_pfn_memslot_atomic(const struct kvm_memory_slot *slot, gfn_t gfn);
1143 kvm_pfn_t __gfn_to_pfn_memslot(const struct kvm_memory_slot *slot, gfn_t gfn,
1144 			       bool atomic, bool *async, bool write_fault,
1145 			       bool *writable, hva_t *hva);
1146 
1147 void kvm_release_pfn_clean(kvm_pfn_t pfn);
1148 void kvm_release_pfn_dirty(kvm_pfn_t pfn);
1149 void kvm_set_pfn_dirty(kvm_pfn_t pfn);
1150 void kvm_set_pfn_accessed(kvm_pfn_t pfn);
1151 
1152 void kvm_release_pfn(kvm_pfn_t pfn, bool dirty);
1153 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1154 			int len);
1155 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len);
1156 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1157 			   void *data, unsigned long len);
1158 int kvm_read_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1159 				 void *data, unsigned int offset,
1160 				 unsigned long len);
1161 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1162 			 int offset, int len);
1163 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1164 		    unsigned long len);
1165 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1166 			   void *data, unsigned long len);
1167 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1168 				  void *data, unsigned int offset,
1169 				  unsigned long len);
1170 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1171 			      gpa_t gpa, unsigned long len);
1172 
1173 #define __kvm_get_guest(kvm, gfn, offset, v)				\
1174 ({									\
1175 	unsigned long __addr = gfn_to_hva(kvm, gfn);			\
1176 	typeof(v) __user *__uaddr = (typeof(__uaddr))(__addr + offset);	\
1177 	int __ret = -EFAULT;						\
1178 									\
1179 	if (!kvm_is_error_hva(__addr))					\
1180 		__ret = get_user(v, __uaddr);				\
1181 	__ret;								\
1182 })
1183 
1184 #define kvm_get_guest(kvm, gpa, v)					\
1185 ({									\
1186 	gpa_t __gpa = gpa;						\
1187 	struct kvm *__kvm = kvm;					\
1188 									\
1189 	__kvm_get_guest(__kvm, __gpa >> PAGE_SHIFT,			\
1190 			offset_in_page(__gpa), v);			\
1191 })
1192 
1193 #define __kvm_put_guest(kvm, gfn, offset, v)				\
1194 ({									\
1195 	unsigned long __addr = gfn_to_hva(kvm, gfn);			\
1196 	typeof(v) __user *__uaddr = (typeof(__uaddr))(__addr + offset);	\
1197 	int __ret = -EFAULT;						\
1198 									\
1199 	if (!kvm_is_error_hva(__addr))					\
1200 		__ret = put_user(v, __uaddr);				\
1201 	if (!__ret)							\
1202 		mark_page_dirty(kvm, gfn);				\
1203 	__ret;								\
1204 })
1205 
1206 #define kvm_put_guest(kvm, gpa, v)					\
1207 ({									\
1208 	gpa_t __gpa = gpa;						\
1209 	struct kvm *__kvm = kvm;					\
1210 									\
1211 	__kvm_put_guest(__kvm, __gpa >> PAGE_SHIFT,			\
1212 			offset_in_page(__gpa), v);			\
1213 })
1214 
1215 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len);
1216 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn);
1217 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn);
1218 bool kvm_vcpu_is_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn);
1219 unsigned long kvm_host_page_size(struct kvm_vcpu *vcpu, gfn_t gfn);
1220 void mark_page_dirty_in_slot(struct kvm *kvm, const struct kvm_memory_slot *memslot, gfn_t gfn);
1221 void mark_page_dirty(struct kvm *kvm, gfn_t gfn);
1222 
1223 struct kvm_memslots *kvm_vcpu_memslots(struct kvm_vcpu *vcpu);
1224 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn);
1225 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn);
1226 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn);
1227 int kvm_vcpu_map(struct kvm_vcpu *vcpu, gpa_t gpa, struct kvm_host_map *map);
1228 void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map, bool dirty);
1229 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn);
1230 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable);
1231 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data, int offset,
1232 			     int len);
1233 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa, void *data,
1234 			       unsigned long len);
1235 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data,
1236 			unsigned long len);
1237 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, const void *data,
1238 			      int offset, int len);
1239 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
1240 			 unsigned long len);
1241 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn);
1242 
1243 /**
1244  * kvm_gpc_init - initialize gfn_to_pfn_cache.
1245  *
1246  * @gpc:	   struct gfn_to_pfn_cache object.
1247  *
1248  * This sets up a gfn_to_pfn_cache by initializing locks.  Note, the cache must
1249  * be zero-allocated (or zeroed by the caller before init).
1250  */
1251 void kvm_gpc_init(struct gfn_to_pfn_cache *gpc);
1252 
1253 /**
1254  * kvm_gpc_activate - prepare a cached kernel mapping and HPA for a given guest
1255  *                    physical address.
1256  *
1257  * @kvm:	   pointer to kvm instance.
1258  * @gpc:	   struct gfn_to_pfn_cache object.
1259  * @vcpu:	   vCPU to be used for marking pages dirty and to be woken on
1260  *		   invalidation.
1261  * @usage:	   indicates if the resulting host physical PFN is used while
1262  *		   the @vcpu is IN_GUEST_MODE (in which case invalidation of
1263  *		   the cache from MMU notifiers---but not for KVM memslot
1264  *		   changes!---will also force @vcpu to exit the guest and
1265  *		   refresh the cache); and/or if the PFN used directly
1266  *		   by KVM (and thus needs a kernel virtual mapping).
1267  * @gpa:	   guest physical address to map.
1268  * @len:	   sanity check; the range being access must fit a single page.
1269  *
1270  * @return:	   0 for success.
1271  *		   -EINVAL for a mapping which would cross a page boundary.
1272  *                 -EFAULT for an untranslatable guest physical address.
1273  *
1274  * This primes a gfn_to_pfn_cache and links it into the @kvm's list for
1275  * invalidations to be processed.  Callers are required to use
1276  * kvm_gfn_to_pfn_cache_check() to ensure that the cache is valid before
1277  * accessing the target page.
1278  */
1279 int kvm_gpc_activate(struct kvm *kvm, struct gfn_to_pfn_cache *gpc,
1280 		     struct kvm_vcpu *vcpu, enum pfn_cache_usage usage,
1281 		     gpa_t gpa, unsigned long len);
1282 
1283 /**
1284  * kvm_gfn_to_pfn_cache_check - check validity of a gfn_to_pfn_cache.
1285  *
1286  * @kvm:	   pointer to kvm instance.
1287  * @gpc:	   struct gfn_to_pfn_cache object.
1288  * @gpa:	   current guest physical address to map.
1289  * @len:	   sanity check; the range being access must fit a single page.
1290  *
1291  * @return:	   %true if the cache is still valid and the address matches.
1292  *		   %false if the cache is not valid.
1293  *
1294  * Callers outside IN_GUEST_MODE context should hold a read lock on @gpc->lock
1295  * while calling this function, and then continue to hold the lock until the
1296  * access is complete.
1297  *
1298  * Callers in IN_GUEST_MODE may do so without locking, although they should
1299  * still hold a read lock on kvm->scru for the memslot checks.
1300  */
1301 bool kvm_gfn_to_pfn_cache_check(struct kvm *kvm, struct gfn_to_pfn_cache *gpc,
1302 				gpa_t gpa, unsigned long len);
1303 
1304 /**
1305  * kvm_gfn_to_pfn_cache_refresh - update a previously initialized cache.
1306  *
1307  * @kvm:	   pointer to kvm instance.
1308  * @gpc:	   struct gfn_to_pfn_cache object.
1309  * @gpa:	   updated guest physical address to map.
1310  * @len:	   sanity check; the range being access must fit a single page.
1311  *
1312  * @return:	   0 for success.
1313  *		   -EINVAL for a mapping which would cross a page boundary.
1314  *                 -EFAULT for an untranslatable guest physical address.
1315  *
1316  * This will attempt to refresh a gfn_to_pfn_cache. Note that a successful
1317  * returm from this function does not mean the page can be immediately
1318  * accessed because it may have raced with an invalidation. Callers must
1319  * still lock and check the cache status, as this function does not return
1320  * with the lock still held to permit access.
1321  */
1322 int kvm_gfn_to_pfn_cache_refresh(struct kvm *kvm, struct gfn_to_pfn_cache *gpc,
1323 				 gpa_t gpa, unsigned long len);
1324 
1325 /**
1326  * kvm_gfn_to_pfn_cache_unmap - temporarily unmap a gfn_to_pfn_cache.
1327  *
1328  * @kvm:	   pointer to kvm instance.
1329  * @gpc:	   struct gfn_to_pfn_cache object.
1330  *
1331  * This unmaps the referenced page. The cache is left in the invalid state
1332  * but at least the mapping from GPA to userspace HVA will remain cached
1333  * and can be reused on a subsequent refresh.
1334  */
1335 void kvm_gfn_to_pfn_cache_unmap(struct kvm *kvm, struct gfn_to_pfn_cache *gpc);
1336 
1337 /**
1338  * kvm_gpc_deactivate - deactivate and unlink a gfn_to_pfn_cache.
1339  *
1340  * @kvm:	   pointer to kvm instance.
1341  * @gpc:	   struct gfn_to_pfn_cache object.
1342  *
1343  * This removes a cache from the @kvm's list to be processed on MMU notifier
1344  * invocation.
1345  */
1346 void kvm_gpc_deactivate(struct kvm *kvm, struct gfn_to_pfn_cache *gpc);
1347 
1348 void kvm_sigset_activate(struct kvm_vcpu *vcpu);
1349 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu);
1350 
1351 void kvm_vcpu_halt(struct kvm_vcpu *vcpu);
1352 bool kvm_vcpu_block(struct kvm_vcpu *vcpu);
1353 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu);
1354 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu);
1355 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu);
1356 void kvm_vcpu_kick(struct kvm_vcpu *vcpu);
1357 int kvm_vcpu_yield_to(struct kvm_vcpu *target);
1358 void kvm_vcpu_on_spin(struct kvm_vcpu *vcpu, bool usermode_vcpu_not_eligible);
1359 
1360 void kvm_flush_remote_tlbs(struct kvm *kvm);
1361 
1362 #ifdef KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE
1363 int kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache *mc, int min);
1364 int __kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache *mc, int capacity, int min);
1365 int kvm_mmu_memory_cache_nr_free_objects(struct kvm_mmu_memory_cache *mc);
1366 void kvm_mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc);
1367 void *kvm_mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc);
1368 #endif
1369 
1370 void kvm_mmu_invalidate_begin(struct kvm *kvm, unsigned long start,
1371 			      unsigned long end);
1372 void kvm_mmu_invalidate_end(struct kvm *kvm, unsigned long start,
1373 			    unsigned long end);
1374 
1375 long kvm_arch_dev_ioctl(struct file *filp,
1376 			unsigned int ioctl, unsigned long arg);
1377 long kvm_arch_vcpu_ioctl(struct file *filp,
1378 			 unsigned int ioctl, unsigned long arg);
1379 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf);
1380 
1381 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext);
1382 
1383 void kvm_arch_mmu_enable_log_dirty_pt_masked(struct kvm *kvm,
1384 					struct kvm_memory_slot *slot,
1385 					gfn_t gfn_offset,
1386 					unsigned long mask);
1387 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot);
1388 
1389 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1390 void kvm_arch_flush_remote_tlbs_memslot(struct kvm *kvm,
1391 					const struct kvm_memory_slot *memslot);
1392 #else /* !CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
1393 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log);
1394 int kvm_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log,
1395 		      int *is_dirty, struct kvm_memory_slot **memslot);
1396 #endif
1397 
1398 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
1399 			bool line_status);
1400 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
1401 			    struct kvm_enable_cap *cap);
1402 long kvm_arch_vm_ioctl(struct file *filp,
1403 		       unsigned int ioctl, unsigned long arg);
1404 long kvm_arch_vm_compat_ioctl(struct file *filp, unsigned int ioctl,
1405 			      unsigned long arg);
1406 
1407 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu);
1408 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu);
1409 
1410 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
1411 				    struct kvm_translation *tr);
1412 
1413 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs);
1414 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs);
1415 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
1416 				  struct kvm_sregs *sregs);
1417 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
1418 				  struct kvm_sregs *sregs);
1419 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
1420 				    struct kvm_mp_state *mp_state);
1421 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
1422 				    struct kvm_mp_state *mp_state);
1423 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
1424 					struct kvm_guest_debug *dbg);
1425 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu);
1426 
1427 int kvm_arch_init(void *opaque);
1428 void kvm_arch_exit(void);
1429 
1430 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu);
1431 
1432 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu);
1433 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu);
1434 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id);
1435 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu);
1436 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu);
1437 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu);
1438 
1439 #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER
1440 int kvm_arch_pm_notifier(struct kvm *kvm, unsigned long state);
1441 #endif
1442 
1443 #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS
1444 void kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu, struct dentry *debugfs_dentry);
1445 #else
kvm_create_vcpu_debugfs(struct kvm_vcpu * vcpu)1446 static inline void kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu) {}
1447 #endif
1448 
1449 int kvm_arch_hardware_enable(void);
1450 void kvm_arch_hardware_disable(void);
1451 int kvm_arch_hardware_setup(void *opaque);
1452 void kvm_arch_hardware_unsetup(void);
1453 int kvm_arch_check_processor_compat(void *opaque);
1454 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu);
1455 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu);
1456 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu);
1457 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu);
1458 bool kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu);
1459 int kvm_arch_post_init_vm(struct kvm *kvm);
1460 void kvm_arch_pre_destroy_vm(struct kvm *kvm);
1461 int kvm_arch_create_vm_debugfs(struct kvm *kvm);
1462 
1463 #ifndef __KVM_HAVE_ARCH_VM_ALLOC
1464 /*
1465  * All architectures that want to use vzalloc currently also
1466  * need their own kvm_arch_alloc_vm implementation.
1467  */
kvm_arch_alloc_vm(void)1468 static inline struct kvm *kvm_arch_alloc_vm(void)
1469 {
1470 	return kzalloc(sizeof(struct kvm), GFP_KERNEL);
1471 }
1472 #endif
1473 
__kvm_arch_free_vm(struct kvm * kvm)1474 static inline void __kvm_arch_free_vm(struct kvm *kvm)
1475 {
1476 	kvfree(kvm);
1477 }
1478 
1479 #ifndef __KVM_HAVE_ARCH_VM_FREE
kvm_arch_free_vm(struct kvm * kvm)1480 static inline void kvm_arch_free_vm(struct kvm *kvm)
1481 {
1482 	__kvm_arch_free_vm(kvm);
1483 }
1484 #endif
1485 
1486 #ifndef __KVM_HAVE_ARCH_FLUSH_REMOTE_TLB
kvm_arch_flush_remote_tlb(struct kvm * kvm)1487 static inline int kvm_arch_flush_remote_tlb(struct kvm *kvm)
1488 {
1489 	return -ENOTSUPP;
1490 }
1491 #endif
1492 
1493 #ifdef __KVM_HAVE_ARCH_NONCOHERENT_DMA
1494 void kvm_arch_register_noncoherent_dma(struct kvm *kvm);
1495 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm);
1496 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm);
1497 #else
kvm_arch_register_noncoherent_dma(struct kvm * kvm)1498 static inline void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
1499 {
1500 }
1501 
kvm_arch_unregister_noncoherent_dma(struct kvm * kvm)1502 static inline void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
1503 {
1504 }
1505 
kvm_arch_has_noncoherent_dma(struct kvm * kvm)1506 static inline bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
1507 {
1508 	return false;
1509 }
1510 #endif
1511 #ifdef __KVM_HAVE_ARCH_ASSIGNED_DEVICE
1512 void kvm_arch_start_assignment(struct kvm *kvm);
1513 void kvm_arch_end_assignment(struct kvm *kvm);
1514 bool kvm_arch_has_assigned_device(struct kvm *kvm);
1515 #else
kvm_arch_start_assignment(struct kvm * kvm)1516 static inline void kvm_arch_start_assignment(struct kvm *kvm)
1517 {
1518 }
1519 
kvm_arch_end_assignment(struct kvm * kvm)1520 static inline void kvm_arch_end_assignment(struct kvm *kvm)
1521 {
1522 }
1523 
kvm_arch_has_assigned_device(struct kvm * kvm)1524 static __always_inline bool kvm_arch_has_assigned_device(struct kvm *kvm)
1525 {
1526 	return false;
1527 }
1528 #endif
1529 
kvm_arch_vcpu_get_wait(struct kvm_vcpu * vcpu)1530 static inline struct rcuwait *kvm_arch_vcpu_get_wait(struct kvm_vcpu *vcpu)
1531 {
1532 #ifdef __KVM_HAVE_ARCH_WQP
1533 	return vcpu->arch.waitp;
1534 #else
1535 	return &vcpu->wait;
1536 #endif
1537 }
1538 
1539 /*
1540  * Wake a vCPU if necessary, but don't do any stats/metadata updates.  Returns
1541  * true if the vCPU was blocking and was awakened, false otherwise.
1542  */
__kvm_vcpu_wake_up(struct kvm_vcpu * vcpu)1543 static inline bool __kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
1544 {
1545 	return !!rcuwait_wake_up(kvm_arch_vcpu_get_wait(vcpu));
1546 }
1547 
kvm_vcpu_is_blocking(struct kvm_vcpu * vcpu)1548 static inline bool kvm_vcpu_is_blocking(struct kvm_vcpu *vcpu)
1549 {
1550 	return rcuwait_active(kvm_arch_vcpu_get_wait(vcpu));
1551 }
1552 
1553 #ifdef __KVM_HAVE_ARCH_INTC_INITIALIZED
1554 /*
1555  * returns true if the virtual interrupt controller is initialized and
1556  * ready to accept virtual IRQ. On some architectures the virtual interrupt
1557  * controller is dynamically instantiated and this is not always true.
1558  */
1559 bool kvm_arch_intc_initialized(struct kvm *kvm);
1560 #else
kvm_arch_intc_initialized(struct kvm * kvm)1561 static inline bool kvm_arch_intc_initialized(struct kvm *kvm)
1562 {
1563 	return true;
1564 }
1565 #endif
1566 
1567 #ifdef CONFIG_GUEST_PERF_EVENTS
1568 unsigned long kvm_arch_vcpu_get_ip(struct kvm_vcpu *vcpu);
1569 
1570 void kvm_register_perf_callbacks(unsigned int (*pt_intr_handler)(void));
1571 void kvm_unregister_perf_callbacks(void);
1572 #else
kvm_register_perf_callbacks(void * ign)1573 static inline void kvm_register_perf_callbacks(void *ign) {}
kvm_unregister_perf_callbacks(void)1574 static inline void kvm_unregister_perf_callbacks(void) {}
1575 #endif /* CONFIG_GUEST_PERF_EVENTS */
1576 
1577 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type);
1578 void kvm_arch_destroy_vm(struct kvm *kvm);
1579 void kvm_arch_sync_events(struct kvm *kvm);
1580 
1581 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu);
1582 
1583 struct page *kvm_pfn_to_refcounted_page(kvm_pfn_t pfn);
1584 bool kvm_is_zone_device_page(struct page *page);
1585 
1586 struct kvm_irq_ack_notifier {
1587 	struct hlist_node link;
1588 	unsigned gsi;
1589 	void (*irq_acked)(struct kvm_irq_ack_notifier *kian);
1590 };
1591 
1592 int kvm_irq_map_gsi(struct kvm *kvm,
1593 		    struct kvm_kernel_irq_routing_entry *entries, int gsi);
1594 int kvm_irq_map_chip_pin(struct kvm *kvm, unsigned irqchip, unsigned pin);
1595 
1596 int kvm_set_irq(struct kvm *kvm, int irq_source_id, u32 irq, int level,
1597 		bool line_status);
1598 int kvm_set_msi(struct kvm_kernel_irq_routing_entry *irq_entry, struct kvm *kvm,
1599 		int irq_source_id, int level, bool line_status);
1600 int kvm_arch_set_irq_inatomic(struct kvm_kernel_irq_routing_entry *e,
1601 			       struct kvm *kvm, int irq_source_id,
1602 			       int level, bool line_status);
1603 bool kvm_irq_has_notifier(struct kvm *kvm, unsigned irqchip, unsigned pin);
1604 void kvm_notify_acked_gsi(struct kvm *kvm, int gsi);
1605 void kvm_notify_acked_irq(struct kvm *kvm, unsigned irqchip, unsigned pin);
1606 void kvm_register_irq_ack_notifier(struct kvm *kvm,
1607 				   struct kvm_irq_ack_notifier *kian);
1608 void kvm_unregister_irq_ack_notifier(struct kvm *kvm,
1609 				   struct kvm_irq_ack_notifier *kian);
1610 int kvm_request_irq_source_id(struct kvm *kvm);
1611 void kvm_free_irq_source_id(struct kvm *kvm, int irq_source_id);
1612 bool kvm_arch_irqfd_allowed(struct kvm *kvm, struct kvm_irqfd *args);
1613 
1614 /*
1615  * Returns a pointer to the memslot if it contains gfn.
1616  * Otherwise returns NULL.
1617  */
1618 static inline struct kvm_memory_slot *
try_get_memslot(struct kvm_memory_slot * slot,gfn_t gfn)1619 try_get_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1620 {
1621 	if (!slot)
1622 		return NULL;
1623 
1624 	if (gfn >= slot->base_gfn && gfn < slot->base_gfn + slot->npages)
1625 		return slot;
1626 	else
1627 		return NULL;
1628 }
1629 
1630 /*
1631  * Returns a pointer to the memslot that contains gfn. Otherwise returns NULL.
1632  *
1633  * With "approx" set returns the memslot also when the address falls
1634  * in a hole. In that case one of the memslots bordering the hole is
1635  * returned.
1636  */
1637 static inline struct kvm_memory_slot *
search_memslots(struct kvm_memslots * slots,gfn_t gfn,bool approx)1638 search_memslots(struct kvm_memslots *slots, gfn_t gfn, bool approx)
1639 {
1640 	struct kvm_memory_slot *slot;
1641 	struct rb_node *node;
1642 	int idx = slots->node_idx;
1643 
1644 	slot = NULL;
1645 	for (node = slots->gfn_tree.rb_node; node; ) {
1646 		slot = container_of(node, struct kvm_memory_slot, gfn_node[idx]);
1647 		if (gfn >= slot->base_gfn) {
1648 			if (gfn < slot->base_gfn + slot->npages)
1649 				return slot;
1650 			node = node->rb_right;
1651 		} else
1652 			node = node->rb_left;
1653 	}
1654 
1655 	return approx ? slot : NULL;
1656 }
1657 
1658 static inline struct kvm_memory_slot *
____gfn_to_memslot(struct kvm_memslots * slots,gfn_t gfn,bool approx)1659 ____gfn_to_memslot(struct kvm_memslots *slots, gfn_t gfn, bool approx)
1660 {
1661 	struct kvm_memory_slot *slot;
1662 
1663 	slot = (struct kvm_memory_slot *)atomic_long_read(&slots->last_used_slot);
1664 	slot = try_get_memslot(slot, gfn);
1665 	if (slot)
1666 		return slot;
1667 
1668 	slot = search_memslots(slots, gfn, approx);
1669 	if (slot) {
1670 		atomic_long_set(&slots->last_used_slot, (unsigned long)slot);
1671 		return slot;
1672 	}
1673 
1674 	return NULL;
1675 }
1676 
1677 /*
1678  * __gfn_to_memslot() and its descendants are here to allow arch code to inline
1679  * the lookups in hot paths.  gfn_to_memslot() itself isn't here as an inline
1680  * because that would bloat other code too much.
1681  */
1682 static inline struct kvm_memory_slot *
__gfn_to_memslot(struct kvm_memslots * slots,gfn_t gfn)1683 __gfn_to_memslot(struct kvm_memslots *slots, gfn_t gfn)
1684 {
1685 	return ____gfn_to_memslot(slots, gfn, false);
1686 }
1687 
1688 static inline unsigned long
__gfn_to_hva_memslot(const struct kvm_memory_slot * slot,gfn_t gfn)1689 __gfn_to_hva_memslot(const struct kvm_memory_slot *slot, gfn_t gfn)
1690 {
1691 	/*
1692 	 * The index was checked originally in search_memslots.  To avoid
1693 	 * that a malicious guest builds a Spectre gadget out of e.g. page
1694 	 * table walks, do not let the processor speculate loads outside
1695 	 * the guest's registered memslots.
1696 	 */
1697 	unsigned long offset = gfn - slot->base_gfn;
1698 	offset = array_index_nospec(offset, slot->npages);
1699 	return slot->userspace_addr + offset * PAGE_SIZE;
1700 }
1701 
memslot_id(struct kvm * kvm,gfn_t gfn)1702 static inline int memslot_id(struct kvm *kvm, gfn_t gfn)
1703 {
1704 	return gfn_to_memslot(kvm, gfn)->id;
1705 }
1706 
1707 static inline gfn_t
hva_to_gfn_memslot(unsigned long hva,struct kvm_memory_slot * slot)1708 hva_to_gfn_memslot(unsigned long hva, struct kvm_memory_slot *slot)
1709 {
1710 	gfn_t gfn_offset = (hva - slot->userspace_addr) >> PAGE_SHIFT;
1711 
1712 	return slot->base_gfn + gfn_offset;
1713 }
1714 
gfn_to_gpa(gfn_t gfn)1715 static inline gpa_t gfn_to_gpa(gfn_t gfn)
1716 {
1717 	return (gpa_t)gfn << PAGE_SHIFT;
1718 }
1719 
gpa_to_gfn(gpa_t gpa)1720 static inline gfn_t gpa_to_gfn(gpa_t gpa)
1721 {
1722 	return (gfn_t)(gpa >> PAGE_SHIFT);
1723 }
1724 
pfn_to_hpa(kvm_pfn_t pfn)1725 static inline hpa_t pfn_to_hpa(kvm_pfn_t pfn)
1726 {
1727 	return (hpa_t)pfn << PAGE_SHIFT;
1728 }
1729 
kvm_is_error_gpa(struct kvm * kvm,gpa_t gpa)1730 static inline bool kvm_is_error_gpa(struct kvm *kvm, gpa_t gpa)
1731 {
1732 	unsigned long hva = gfn_to_hva(kvm, gpa_to_gfn(gpa));
1733 
1734 	return kvm_is_error_hva(hva);
1735 }
1736 
1737 enum kvm_stat_kind {
1738 	KVM_STAT_VM,
1739 	KVM_STAT_VCPU,
1740 };
1741 
1742 struct kvm_stat_data {
1743 	struct kvm *kvm;
1744 	const struct _kvm_stats_desc *desc;
1745 	enum kvm_stat_kind kind;
1746 };
1747 
1748 struct _kvm_stats_desc {
1749 	struct kvm_stats_desc desc;
1750 	char name[KVM_STATS_NAME_SIZE];
1751 };
1752 
1753 #define STATS_DESC_COMMON(type, unit, base, exp, sz, bsz)		       \
1754 	.flags = type | unit | base |					       \
1755 		 BUILD_BUG_ON_ZERO(type & ~KVM_STATS_TYPE_MASK) |	       \
1756 		 BUILD_BUG_ON_ZERO(unit & ~KVM_STATS_UNIT_MASK) |	       \
1757 		 BUILD_BUG_ON_ZERO(base & ~KVM_STATS_BASE_MASK),	       \
1758 	.exponent = exp,						       \
1759 	.size = sz,							       \
1760 	.bucket_size = bsz
1761 
1762 #define VM_GENERIC_STATS_DESC(stat, type, unit, base, exp, sz, bsz)	       \
1763 	{								       \
1764 		{							       \
1765 			STATS_DESC_COMMON(type, unit, base, exp, sz, bsz),     \
1766 			.offset = offsetof(struct kvm_vm_stat, generic.stat)   \
1767 		},							       \
1768 		.name = #stat,						       \
1769 	}
1770 #define VCPU_GENERIC_STATS_DESC(stat, type, unit, base, exp, sz, bsz)	       \
1771 	{								       \
1772 		{							       \
1773 			STATS_DESC_COMMON(type, unit, base, exp, sz, bsz),     \
1774 			.offset = offsetof(struct kvm_vcpu_stat, generic.stat) \
1775 		},							       \
1776 		.name = #stat,						       \
1777 	}
1778 #define VM_STATS_DESC(stat, type, unit, base, exp, sz, bsz)		       \
1779 	{								       \
1780 		{							       \
1781 			STATS_DESC_COMMON(type, unit, base, exp, sz, bsz),     \
1782 			.offset = offsetof(struct kvm_vm_stat, stat)	       \
1783 		},							       \
1784 		.name = #stat,						       \
1785 	}
1786 #define VCPU_STATS_DESC(stat, type, unit, base, exp, sz, bsz)		       \
1787 	{								       \
1788 		{							       \
1789 			STATS_DESC_COMMON(type, unit, base, exp, sz, bsz),     \
1790 			.offset = offsetof(struct kvm_vcpu_stat, stat)	       \
1791 		},							       \
1792 		.name = #stat,						       \
1793 	}
1794 /* SCOPE: VM, VM_GENERIC, VCPU, VCPU_GENERIC */
1795 #define STATS_DESC(SCOPE, stat, type, unit, base, exp, sz, bsz)		       \
1796 	SCOPE##_STATS_DESC(stat, type, unit, base, exp, sz, bsz)
1797 
1798 #define STATS_DESC_CUMULATIVE(SCOPE, name, unit, base, exponent)	       \
1799 	STATS_DESC(SCOPE, name, KVM_STATS_TYPE_CUMULATIVE,		       \
1800 		unit, base, exponent, 1, 0)
1801 #define STATS_DESC_INSTANT(SCOPE, name, unit, base, exponent)		       \
1802 	STATS_DESC(SCOPE, name, KVM_STATS_TYPE_INSTANT,			       \
1803 		unit, base, exponent, 1, 0)
1804 #define STATS_DESC_PEAK(SCOPE, name, unit, base, exponent)		       \
1805 	STATS_DESC(SCOPE, name, KVM_STATS_TYPE_PEAK,			       \
1806 		unit, base, exponent, 1, 0)
1807 #define STATS_DESC_LINEAR_HIST(SCOPE, name, unit, base, exponent, sz, bsz)     \
1808 	STATS_DESC(SCOPE, name, KVM_STATS_TYPE_LINEAR_HIST,		       \
1809 		unit, base, exponent, sz, bsz)
1810 #define STATS_DESC_LOG_HIST(SCOPE, name, unit, base, exponent, sz)	       \
1811 	STATS_DESC(SCOPE, name, KVM_STATS_TYPE_LOG_HIST,		       \
1812 		unit, base, exponent, sz, 0)
1813 
1814 /* Cumulative counter, read/write */
1815 #define STATS_DESC_COUNTER(SCOPE, name)					       \
1816 	STATS_DESC_CUMULATIVE(SCOPE, name, KVM_STATS_UNIT_NONE,		       \
1817 		KVM_STATS_BASE_POW10, 0)
1818 /* Instantaneous counter, read only */
1819 #define STATS_DESC_ICOUNTER(SCOPE, name)				       \
1820 	STATS_DESC_INSTANT(SCOPE, name, KVM_STATS_UNIT_NONE,		       \
1821 		KVM_STATS_BASE_POW10, 0)
1822 /* Peak counter, read/write */
1823 #define STATS_DESC_PCOUNTER(SCOPE, name)				       \
1824 	STATS_DESC_PEAK(SCOPE, name, KVM_STATS_UNIT_NONE,		       \
1825 		KVM_STATS_BASE_POW10, 0)
1826 
1827 /* Instantaneous boolean value, read only */
1828 #define STATS_DESC_IBOOLEAN(SCOPE, name)				       \
1829 	STATS_DESC_INSTANT(SCOPE, name, KVM_STATS_UNIT_BOOLEAN,		       \
1830 		KVM_STATS_BASE_POW10, 0)
1831 /* Peak (sticky) boolean value, read/write */
1832 #define STATS_DESC_PBOOLEAN(SCOPE, name)				       \
1833 	STATS_DESC_PEAK(SCOPE, name, KVM_STATS_UNIT_BOOLEAN,		       \
1834 		KVM_STATS_BASE_POW10, 0)
1835 
1836 /* Cumulative time in nanosecond */
1837 #define STATS_DESC_TIME_NSEC(SCOPE, name)				       \
1838 	STATS_DESC_CUMULATIVE(SCOPE, name, KVM_STATS_UNIT_SECONDS,	       \
1839 		KVM_STATS_BASE_POW10, -9)
1840 /* Linear histogram for time in nanosecond */
1841 #define STATS_DESC_LINHIST_TIME_NSEC(SCOPE, name, sz, bsz)		       \
1842 	STATS_DESC_LINEAR_HIST(SCOPE, name, KVM_STATS_UNIT_SECONDS,	       \
1843 		KVM_STATS_BASE_POW10, -9, sz, bsz)
1844 /* Logarithmic histogram for time in nanosecond */
1845 #define STATS_DESC_LOGHIST_TIME_NSEC(SCOPE, name, sz)			       \
1846 	STATS_DESC_LOG_HIST(SCOPE, name, KVM_STATS_UNIT_SECONDS,	       \
1847 		KVM_STATS_BASE_POW10, -9, sz)
1848 
1849 #define KVM_GENERIC_VM_STATS()						       \
1850 	STATS_DESC_COUNTER(VM_GENERIC, remote_tlb_flush),		       \
1851 	STATS_DESC_COUNTER(VM_GENERIC, remote_tlb_flush_requests)
1852 
1853 #define KVM_GENERIC_VCPU_STATS()					       \
1854 	STATS_DESC_COUNTER(VCPU_GENERIC, halt_successful_poll),		       \
1855 	STATS_DESC_COUNTER(VCPU_GENERIC, halt_attempted_poll),		       \
1856 	STATS_DESC_COUNTER(VCPU_GENERIC, halt_poll_invalid),		       \
1857 	STATS_DESC_COUNTER(VCPU_GENERIC, halt_wakeup),			       \
1858 	STATS_DESC_TIME_NSEC(VCPU_GENERIC, halt_poll_success_ns),	       \
1859 	STATS_DESC_TIME_NSEC(VCPU_GENERIC, halt_poll_fail_ns),		       \
1860 	STATS_DESC_TIME_NSEC(VCPU_GENERIC, halt_wait_ns),		       \
1861 	STATS_DESC_LOGHIST_TIME_NSEC(VCPU_GENERIC, halt_poll_success_hist,     \
1862 			HALT_POLL_HIST_COUNT),				       \
1863 	STATS_DESC_LOGHIST_TIME_NSEC(VCPU_GENERIC, halt_poll_fail_hist,	       \
1864 			HALT_POLL_HIST_COUNT),				       \
1865 	STATS_DESC_LOGHIST_TIME_NSEC(VCPU_GENERIC, halt_wait_hist,	       \
1866 			HALT_POLL_HIST_COUNT),				       \
1867 	STATS_DESC_IBOOLEAN(VCPU_GENERIC, blocking)
1868 
1869 extern struct dentry *kvm_debugfs_dir;
1870 
1871 ssize_t kvm_stats_read(char *id, const struct kvm_stats_header *header,
1872 		       const struct _kvm_stats_desc *desc,
1873 		       void *stats, size_t size_stats,
1874 		       char __user *user_buffer, size_t size, loff_t *offset);
1875 
1876 /**
1877  * kvm_stats_linear_hist_update() - Update bucket value for linear histogram
1878  * statistics data.
1879  *
1880  * @data: start address of the stats data
1881  * @size: the number of bucket of the stats data
1882  * @value: the new value used to update the linear histogram's bucket
1883  * @bucket_size: the size (width) of a bucket
1884  */
kvm_stats_linear_hist_update(u64 * data,size_t size,u64 value,size_t bucket_size)1885 static inline void kvm_stats_linear_hist_update(u64 *data, size_t size,
1886 						u64 value, size_t bucket_size)
1887 {
1888 	size_t index = div64_u64(value, bucket_size);
1889 
1890 	index = min(index, size - 1);
1891 	++data[index];
1892 }
1893 
1894 /**
1895  * kvm_stats_log_hist_update() - Update bucket value for logarithmic histogram
1896  * statistics data.
1897  *
1898  * @data: start address of the stats data
1899  * @size: the number of bucket of the stats data
1900  * @value: the new value used to update the logarithmic histogram's bucket
1901  */
kvm_stats_log_hist_update(u64 * data,size_t size,u64 value)1902 static inline void kvm_stats_log_hist_update(u64 *data, size_t size, u64 value)
1903 {
1904 	size_t index = fls64(value);
1905 
1906 	index = min(index, size - 1);
1907 	++data[index];
1908 }
1909 
1910 #define KVM_STATS_LINEAR_HIST_UPDATE(array, value, bsize)		       \
1911 	kvm_stats_linear_hist_update(array, ARRAY_SIZE(array), value, bsize)
1912 #define KVM_STATS_LOG_HIST_UPDATE(array, value)				       \
1913 	kvm_stats_log_hist_update(array, ARRAY_SIZE(array), value)
1914 
1915 
1916 extern const struct kvm_stats_header kvm_vm_stats_header;
1917 extern const struct _kvm_stats_desc kvm_vm_stats_desc[];
1918 extern const struct kvm_stats_header kvm_vcpu_stats_header;
1919 extern const struct _kvm_stats_desc kvm_vcpu_stats_desc[];
1920 
1921 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
mmu_invalidate_retry(struct kvm * kvm,unsigned long mmu_seq)1922 static inline int mmu_invalidate_retry(struct kvm *kvm, unsigned long mmu_seq)
1923 {
1924 	if (unlikely(kvm->mmu_invalidate_in_progress))
1925 		return 1;
1926 	/*
1927 	 * Ensure the read of mmu_invalidate_in_progress happens before
1928 	 * the read of mmu_invalidate_seq.  This interacts with the
1929 	 * smp_wmb() in mmu_notifier_invalidate_range_end to make sure
1930 	 * that the caller either sees the old (non-zero) value of
1931 	 * mmu_invalidate_in_progress or the new (incremented) value of
1932 	 * mmu_invalidate_seq.
1933 	 *
1934 	 * PowerPC Book3s HV KVM calls this under a per-page lock rather
1935 	 * than under kvm->mmu_lock, for scalability, so can't rely on
1936 	 * kvm->mmu_lock to keep things ordered.
1937 	 */
1938 	smp_rmb();
1939 	if (kvm->mmu_invalidate_seq != mmu_seq)
1940 		return 1;
1941 	return 0;
1942 }
1943 
mmu_invalidate_retry_hva(struct kvm * kvm,unsigned long mmu_seq,unsigned long hva)1944 static inline int mmu_invalidate_retry_hva(struct kvm *kvm,
1945 					   unsigned long mmu_seq,
1946 					   unsigned long hva)
1947 {
1948 	lockdep_assert_held(&kvm->mmu_lock);
1949 	/*
1950 	 * If mmu_invalidate_in_progress is non-zero, then the range maintained
1951 	 * by kvm_mmu_notifier_invalidate_range_start contains all addresses
1952 	 * that might be being invalidated. Note that it may include some false
1953 	 * positives, due to shortcuts when handing concurrent invalidations.
1954 	 */
1955 	if (unlikely(kvm->mmu_invalidate_in_progress) &&
1956 	    hva >= kvm->mmu_invalidate_range_start &&
1957 	    hva < kvm->mmu_invalidate_range_end)
1958 		return 1;
1959 	if (kvm->mmu_invalidate_seq != mmu_seq)
1960 		return 1;
1961 	return 0;
1962 }
1963 #endif
1964 
1965 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
1966 
1967 #define KVM_MAX_IRQ_ROUTES 4096 /* might need extension/rework in the future */
1968 
1969 bool kvm_arch_can_set_irq_routing(struct kvm *kvm);
1970 int kvm_set_irq_routing(struct kvm *kvm,
1971 			const struct kvm_irq_routing_entry *entries,
1972 			unsigned nr,
1973 			unsigned flags);
1974 int kvm_set_routing_entry(struct kvm *kvm,
1975 			  struct kvm_kernel_irq_routing_entry *e,
1976 			  const struct kvm_irq_routing_entry *ue);
1977 void kvm_free_irq_routing(struct kvm *kvm);
1978 
1979 #else
1980 
kvm_free_irq_routing(struct kvm * kvm)1981 static inline void kvm_free_irq_routing(struct kvm *kvm) {}
1982 
1983 #endif
1984 
1985 int kvm_send_userspace_msi(struct kvm *kvm, struct kvm_msi *msi);
1986 
1987 #ifdef CONFIG_HAVE_KVM_EVENTFD
1988 
1989 void kvm_eventfd_init(struct kvm *kvm);
1990 int kvm_ioeventfd(struct kvm *kvm, struct kvm_ioeventfd *args);
1991 
1992 #ifdef CONFIG_HAVE_KVM_IRQFD
1993 int kvm_irqfd(struct kvm *kvm, struct kvm_irqfd *args);
1994 void kvm_irqfd_release(struct kvm *kvm);
1995 void kvm_irq_routing_update(struct kvm *);
1996 #else
kvm_irqfd(struct kvm * kvm,struct kvm_irqfd * args)1997 static inline int kvm_irqfd(struct kvm *kvm, struct kvm_irqfd *args)
1998 {
1999 	return -EINVAL;
2000 }
2001 
kvm_irqfd_release(struct kvm * kvm)2002 static inline void kvm_irqfd_release(struct kvm *kvm) {}
2003 #endif
2004 
2005 #else
2006 
kvm_eventfd_init(struct kvm * kvm)2007 static inline void kvm_eventfd_init(struct kvm *kvm) {}
2008 
kvm_irqfd(struct kvm * kvm,struct kvm_irqfd * args)2009 static inline int kvm_irqfd(struct kvm *kvm, struct kvm_irqfd *args)
2010 {
2011 	return -EINVAL;
2012 }
2013 
kvm_irqfd_release(struct kvm * kvm)2014 static inline void kvm_irqfd_release(struct kvm *kvm) {}
2015 
2016 #ifdef CONFIG_HAVE_KVM_IRQCHIP
kvm_irq_routing_update(struct kvm * kvm)2017 static inline void kvm_irq_routing_update(struct kvm *kvm)
2018 {
2019 }
2020 #endif
2021 
kvm_ioeventfd(struct kvm * kvm,struct kvm_ioeventfd * args)2022 static inline int kvm_ioeventfd(struct kvm *kvm, struct kvm_ioeventfd *args)
2023 {
2024 	return -ENOSYS;
2025 }
2026 
2027 #endif /* CONFIG_HAVE_KVM_EVENTFD */
2028 
2029 void kvm_arch_irq_routing_update(struct kvm *kvm);
2030 
__kvm_make_request(int req,struct kvm_vcpu * vcpu)2031 static inline void __kvm_make_request(int req, struct kvm_vcpu *vcpu)
2032 {
2033 	/*
2034 	 * Ensure the rest of the request is published to kvm_check_request's
2035 	 * caller.  Paired with the smp_mb__after_atomic in kvm_check_request.
2036 	 */
2037 	smp_wmb();
2038 	set_bit(req & KVM_REQUEST_MASK, (void *)&vcpu->requests);
2039 }
2040 
kvm_make_request(int req,struct kvm_vcpu * vcpu)2041 static __always_inline void kvm_make_request(int req, struct kvm_vcpu *vcpu)
2042 {
2043 	/*
2044 	 * Request that don't require vCPU action should never be logged in
2045 	 * vcpu->requests.  The vCPU won't clear the request, so it will stay
2046 	 * logged indefinitely and prevent the vCPU from entering the guest.
2047 	 */
2048 	BUILD_BUG_ON(!__builtin_constant_p(req) ||
2049 		     (req & KVM_REQUEST_NO_ACTION));
2050 
2051 	__kvm_make_request(req, vcpu);
2052 }
2053 
kvm_request_pending(struct kvm_vcpu * vcpu)2054 static inline bool kvm_request_pending(struct kvm_vcpu *vcpu)
2055 {
2056 	return READ_ONCE(vcpu->requests);
2057 }
2058 
kvm_test_request(int req,struct kvm_vcpu * vcpu)2059 static inline bool kvm_test_request(int req, struct kvm_vcpu *vcpu)
2060 {
2061 	return test_bit(req & KVM_REQUEST_MASK, (void *)&vcpu->requests);
2062 }
2063 
kvm_clear_request(int req,struct kvm_vcpu * vcpu)2064 static inline void kvm_clear_request(int req, struct kvm_vcpu *vcpu)
2065 {
2066 	clear_bit(req & KVM_REQUEST_MASK, (void *)&vcpu->requests);
2067 }
2068 
kvm_check_request(int req,struct kvm_vcpu * vcpu)2069 static inline bool kvm_check_request(int req, struct kvm_vcpu *vcpu)
2070 {
2071 	if (kvm_test_request(req, vcpu)) {
2072 		kvm_clear_request(req, vcpu);
2073 
2074 		/*
2075 		 * Ensure the rest of the request is visible to kvm_check_request's
2076 		 * caller.  Paired with the smp_wmb in kvm_make_request.
2077 		 */
2078 		smp_mb__after_atomic();
2079 		return true;
2080 	} else {
2081 		return false;
2082 	}
2083 }
2084 
2085 extern bool kvm_rebooting;
2086 
2087 extern unsigned int halt_poll_ns;
2088 extern unsigned int halt_poll_ns_grow;
2089 extern unsigned int halt_poll_ns_grow_start;
2090 extern unsigned int halt_poll_ns_shrink;
2091 
2092 struct kvm_device {
2093 	const struct kvm_device_ops *ops;
2094 	struct kvm *kvm;
2095 	void *private;
2096 	struct list_head vm_node;
2097 };
2098 
2099 /* create, destroy, and name are mandatory */
2100 struct kvm_device_ops {
2101 	const char *name;
2102 
2103 	/*
2104 	 * create is called holding kvm->lock and any operations not suitable
2105 	 * to do while holding the lock should be deferred to init (see
2106 	 * below).
2107 	 */
2108 	int (*create)(struct kvm_device *dev, u32 type);
2109 
2110 	/*
2111 	 * init is called after create if create is successful and is called
2112 	 * outside of holding kvm->lock.
2113 	 */
2114 	void (*init)(struct kvm_device *dev);
2115 
2116 	/*
2117 	 * Destroy is responsible for freeing dev.
2118 	 *
2119 	 * Destroy may be called before or after destructors are called
2120 	 * on emulated I/O regions, depending on whether a reference is
2121 	 * held by a vcpu or other kvm component that gets destroyed
2122 	 * after the emulated I/O.
2123 	 */
2124 	void (*destroy)(struct kvm_device *dev);
2125 
2126 	/*
2127 	 * Release is an alternative method to free the device. It is
2128 	 * called when the device file descriptor is closed. Once
2129 	 * release is called, the destroy method will not be called
2130 	 * anymore as the device is removed from the device list of
2131 	 * the VM. kvm->lock is held.
2132 	 */
2133 	void (*release)(struct kvm_device *dev);
2134 
2135 	int (*set_attr)(struct kvm_device *dev, struct kvm_device_attr *attr);
2136 	int (*get_attr)(struct kvm_device *dev, struct kvm_device_attr *attr);
2137 	int (*has_attr)(struct kvm_device *dev, struct kvm_device_attr *attr);
2138 	long (*ioctl)(struct kvm_device *dev, unsigned int ioctl,
2139 		      unsigned long arg);
2140 	int (*mmap)(struct kvm_device *dev, struct vm_area_struct *vma);
2141 };
2142 
2143 void kvm_device_get(struct kvm_device *dev);
2144 void kvm_device_put(struct kvm_device *dev);
2145 struct kvm_device *kvm_device_from_filp(struct file *filp);
2146 int kvm_register_device_ops(const struct kvm_device_ops *ops, u32 type);
2147 void kvm_unregister_device_ops(u32 type);
2148 
2149 extern struct kvm_device_ops kvm_mpic_ops;
2150 extern struct kvm_device_ops kvm_arm_vgic_v2_ops;
2151 extern struct kvm_device_ops kvm_arm_vgic_v3_ops;
2152 
2153 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2154 
kvm_vcpu_set_in_spin_loop(struct kvm_vcpu * vcpu,bool val)2155 static inline void kvm_vcpu_set_in_spin_loop(struct kvm_vcpu *vcpu, bool val)
2156 {
2157 	vcpu->spin_loop.in_spin_loop = val;
2158 }
kvm_vcpu_set_dy_eligible(struct kvm_vcpu * vcpu,bool val)2159 static inline void kvm_vcpu_set_dy_eligible(struct kvm_vcpu *vcpu, bool val)
2160 {
2161 	vcpu->spin_loop.dy_eligible = val;
2162 }
2163 
2164 #else /* !CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT */
2165 
kvm_vcpu_set_in_spin_loop(struct kvm_vcpu * vcpu,bool val)2166 static inline void kvm_vcpu_set_in_spin_loop(struct kvm_vcpu *vcpu, bool val)
2167 {
2168 }
2169 
kvm_vcpu_set_dy_eligible(struct kvm_vcpu * vcpu,bool val)2170 static inline void kvm_vcpu_set_dy_eligible(struct kvm_vcpu *vcpu, bool val)
2171 {
2172 }
2173 #endif /* CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT */
2174 
kvm_is_visible_memslot(struct kvm_memory_slot * memslot)2175 static inline bool kvm_is_visible_memslot(struct kvm_memory_slot *memslot)
2176 {
2177 	return (memslot && memslot->id < KVM_USER_MEM_SLOTS &&
2178 		!(memslot->flags & KVM_MEMSLOT_INVALID));
2179 }
2180 
2181 struct kvm_vcpu *kvm_get_running_vcpu(void);
2182 struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void);
2183 
2184 #ifdef CONFIG_HAVE_KVM_IRQ_BYPASS
2185 bool kvm_arch_has_irq_bypass(void);
2186 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *,
2187 			   struct irq_bypass_producer *);
2188 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *,
2189 			   struct irq_bypass_producer *);
2190 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *);
2191 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *);
2192 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
2193 				  uint32_t guest_irq, bool set);
2194 bool kvm_arch_irqfd_route_changed(struct kvm_kernel_irq_routing_entry *,
2195 				  struct kvm_kernel_irq_routing_entry *);
2196 #endif /* CONFIG_HAVE_KVM_IRQ_BYPASS */
2197 
2198 #ifdef CONFIG_HAVE_KVM_INVALID_WAKEUPS
2199 /* If we wakeup during the poll time, was it a sucessful poll? */
vcpu_valid_wakeup(struct kvm_vcpu * vcpu)2200 static inline bool vcpu_valid_wakeup(struct kvm_vcpu *vcpu)
2201 {
2202 	return vcpu->valid_wakeup;
2203 }
2204 
2205 #else
vcpu_valid_wakeup(struct kvm_vcpu * vcpu)2206 static inline bool vcpu_valid_wakeup(struct kvm_vcpu *vcpu)
2207 {
2208 	return true;
2209 }
2210 #endif /* CONFIG_HAVE_KVM_INVALID_WAKEUPS */
2211 
2212 #ifdef CONFIG_HAVE_KVM_NO_POLL
2213 /* Callback that tells if we must not poll */
2214 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu);
2215 #else
kvm_arch_no_poll(struct kvm_vcpu * vcpu)2216 static inline bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
2217 {
2218 	return false;
2219 }
2220 #endif /* CONFIG_HAVE_KVM_NO_POLL */
2221 
2222 #ifdef CONFIG_HAVE_KVM_VCPU_ASYNC_IOCTL
2223 long kvm_arch_vcpu_async_ioctl(struct file *filp,
2224 			       unsigned int ioctl, unsigned long arg);
2225 #else
kvm_arch_vcpu_async_ioctl(struct file * filp,unsigned int ioctl,unsigned long arg)2226 static inline long kvm_arch_vcpu_async_ioctl(struct file *filp,
2227 					     unsigned int ioctl,
2228 					     unsigned long arg)
2229 {
2230 	return -ENOIOCTLCMD;
2231 }
2232 #endif /* CONFIG_HAVE_KVM_VCPU_ASYNC_IOCTL */
2233 
2234 void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
2235 					    unsigned long start, unsigned long end);
2236 
2237 void kvm_arch_guest_memory_reclaimed(struct kvm *kvm);
2238 
2239 #ifdef CONFIG_HAVE_KVM_VCPU_RUN_PID_CHANGE
2240 int kvm_arch_vcpu_run_pid_change(struct kvm_vcpu *vcpu);
2241 #else
kvm_arch_vcpu_run_pid_change(struct kvm_vcpu * vcpu)2242 static inline int kvm_arch_vcpu_run_pid_change(struct kvm_vcpu *vcpu)
2243 {
2244 	return 0;
2245 }
2246 #endif /* CONFIG_HAVE_KVM_VCPU_RUN_PID_CHANGE */
2247 
2248 typedef int (*kvm_vm_thread_fn_t)(struct kvm *kvm, uintptr_t data);
2249 
2250 int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn,
2251 				uintptr_t data, const char *name,
2252 				struct task_struct **thread_ptr);
2253 
2254 #ifdef CONFIG_KVM_XFER_TO_GUEST_WORK
kvm_handle_signal_exit(struct kvm_vcpu * vcpu)2255 static inline void kvm_handle_signal_exit(struct kvm_vcpu *vcpu)
2256 {
2257 	vcpu->run->exit_reason = KVM_EXIT_INTR;
2258 	vcpu->stat.signal_exits++;
2259 }
2260 #endif /* CONFIG_KVM_XFER_TO_GUEST_WORK */
2261 
2262 /*
2263  * If more than one page is being (un)accounted, @virt must be the address of
2264  * the first page of a block of pages what were allocated together (i.e
2265  * accounted together).
2266  *
2267  * kvm_account_pgtable_pages() is thread-safe because mod_lruvec_page_state()
2268  * is thread-safe.
2269  */
kvm_account_pgtable_pages(void * virt,int nr)2270 static inline void kvm_account_pgtable_pages(void *virt, int nr)
2271 {
2272 	mod_lruvec_page_state(virt_to_page(virt), NR_SECONDARY_PAGETABLE, nr);
2273 }
2274 
2275 /*
2276  * This defines how many reserved entries we want to keep before we
2277  * kick the vcpu to the userspace to avoid dirty ring full.  This
2278  * value can be tuned to higher if e.g. PML is enabled on the host.
2279  */
2280 #define  KVM_DIRTY_RING_RSVD_ENTRIES  64
2281 
2282 /* Max number of entries allowed for each kvm dirty ring */
2283 #define  KVM_DIRTY_RING_MAX_ENTRIES  65536
2284 
2285 #endif
2286