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_userspace_memory_region::flags are internally
54 * used 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 kvm_swap_active_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_MAX_NR_ADDRESS_SPACES
84 #define KVM_MAX_NR_ADDRESS_SPACES 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 #define KVM_PFN_ERR_SIGPENDING (KVM_PFN_ERR_MASK + 3)
100
101 /*
102 * error pfns indicate that the gfn is in slot but faild to
103 * translate it to pfn on host.
104 */
is_error_pfn(kvm_pfn_t pfn)105 static inline bool is_error_pfn(kvm_pfn_t pfn)
106 {
107 return !!(pfn & KVM_PFN_ERR_MASK);
108 }
109
110 /*
111 * KVM_PFN_ERR_SIGPENDING indicates that fetching the PFN was interrupted
112 * by a pending signal. Note, the signal may or may not be fatal.
113 */
is_sigpending_pfn(kvm_pfn_t pfn)114 static inline bool is_sigpending_pfn(kvm_pfn_t pfn)
115 {
116 return pfn == KVM_PFN_ERR_SIGPENDING;
117 }
118
119 /*
120 * error_noslot pfns indicate that the gfn can not be
121 * translated to pfn - it is not in slot or failed to
122 * translate it to pfn.
123 */
is_error_noslot_pfn(kvm_pfn_t pfn)124 static inline bool is_error_noslot_pfn(kvm_pfn_t pfn)
125 {
126 return !!(pfn & KVM_PFN_ERR_NOSLOT_MASK);
127 }
128
129 /* noslot pfn indicates that the gfn is not in slot. */
is_noslot_pfn(kvm_pfn_t pfn)130 static inline bool is_noslot_pfn(kvm_pfn_t pfn)
131 {
132 return pfn == KVM_PFN_NOSLOT;
133 }
134
135 /*
136 * architectures with KVM_HVA_ERR_BAD other than PAGE_OFFSET (e.g. s390)
137 * provide own defines and kvm_is_error_hva
138 */
139 #ifndef KVM_HVA_ERR_BAD
140
141 #define KVM_HVA_ERR_BAD (PAGE_OFFSET)
142 #define KVM_HVA_ERR_RO_BAD (PAGE_OFFSET + PAGE_SIZE)
143
kvm_is_error_hva(unsigned long addr)144 static inline bool kvm_is_error_hva(unsigned long addr)
145 {
146 return addr >= PAGE_OFFSET;
147 }
148
149 #endif
150
kvm_is_error_gpa(gpa_t gpa)151 static inline bool kvm_is_error_gpa(gpa_t gpa)
152 {
153 return gpa == INVALID_GPA;
154 }
155
156 #define KVM_ERR_PTR_BAD_PAGE (ERR_PTR(-ENOENT))
157
is_error_page(struct page * page)158 static inline bool is_error_page(struct page *page)
159 {
160 return IS_ERR(page);
161 }
162
163 #define KVM_REQUEST_MASK GENMASK(7,0)
164 #define KVM_REQUEST_NO_WAKEUP BIT(8)
165 #define KVM_REQUEST_WAIT BIT(9)
166 #define KVM_REQUEST_NO_ACTION BIT(10)
167 /*
168 * Architecture-independent vcpu->requests bit members
169 * Bits 3-7 are reserved for more arch-independent bits.
170 */
171 #define KVM_REQ_TLB_FLUSH (0 | KVM_REQUEST_WAIT | KVM_REQUEST_NO_WAKEUP)
172 #define KVM_REQ_VM_DEAD (1 | KVM_REQUEST_WAIT | KVM_REQUEST_NO_WAKEUP)
173 #define KVM_REQ_UNBLOCK 2
174 #define KVM_REQ_DIRTY_RING_SOFT_FULL 3
175 #define KVM_REQUEST_ARCH_BASE 8
176
177 /*
178 * KVM_REQ_OUTSIDE_GUEST_MODE exists is purely as way to force the vCPU to
179 * OUTSIDE_GUEST_MODE. KVM_REQ_OUTSIDE_GUEST_MODE differs from a vCPU "kick"
180 * in that it ensures the vCPU has reached OUTSIDE_GUEST_MODE before continuing
181 * on. A kick only guarantees that the vCPU is on its way out, e.g. a previous
182 * kick may have set vcpu->mode to EXITING_GUEST_MODE, and so there's no
183 * guarantee the vCPU received an IPI and has actually exited guest mode.
184 */
185 #define KVM_REQ_OUTSIDE_GUEST_MODE (KVM_REQUEST_NO_ACTION | KVM_REQUEST_WAIT | KVM_REQUEST_NO_WAKEUP)
186
187 #define KVM_ARCH_REQ_FLAGS(nr, flags) ({ \
188 BUILD_BUG_ON((unsigned)(nr) >= (sizeof_field(struct kvm_vcpu, requests) * 8) - KVM_REQUEST_ARCH_BASE); \
189 (unsigned)(((nr) + KVM_REQUEST_ARCH_BASE) | (flags)); \
190 })
191 #define KVM_ARCH_REQ(nr) KVM_ARCH_REQ_FLAGS(nr, 0)
192
193 bool kvm_make_vcpus_request_mask(struct kvm *kvm, unsigned int req,
194 unsigned long *vcpu_bitmap);
195 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req);
196
197 #define KVM_USERSPACE_IRQ_SOURCE_ID 0
198 #define KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID 1
199
200 extern struct mutex kvm_lock;
201 extern struct list_head vm_list;
202
203 struct kvm_io_range {
204 gpa_t addr;
205 int len;
206 struct kvm_io_device *dev;
207 };
208
209 #define NR_IOBUS_DEVS 1000
210
211 struct kvm_io_bus {
212 int dev_count;
213 int ioeventfd_count;
214 struct kvm_io_range range[];
215 };
216
217 enum kvm_bus {
218 KVM_MMIO_BUS,
219 KVM_PIO_BUS,
220 KVM_VIRTIO_CCW_NOTIFY_BUS,
221 KVM_FAST_MMIO_BUS,
222 KVM_NR_BUSES
223 };
224
225 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
226 int len, const void *val);
227 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
228 gpa_t addr, int len, const void *val, long cookie);
229 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
230 int len, void *val);
231 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
232 int len, struct kvm_io_device *dev);
233 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
234 struct kvm_io_device *dev);
235 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
236 gpa_t addr);
237
238 #ifdef CONFIG_KVM_ASYNC_PF
239 struct kvm_async_pf {
240 struct work_struct work;
241 struct list_head link;
242 struct list_head queue;
243 struct kvm_vcpu *vcpu;
244 gpa_t cr2_or_gpa;
245 unsigned long addr;
246 struct kvm_arch_async_pf arch;
247 bool wakeup_all;
248 bool notpresent_injected;
249 };
250
251 void kvm_clear_async_pf_completion_queue(struct kvm_vcpu *vcpu);
252 void kvm_check_async_pf_completion(struct kvm_vcpu *vcpu);
253 bool kvm_setup_async_pf(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
254 unsigned long hva, struct kvm_arch_async_pf *arch);
255 int kvm_async_pf_wakeup_all(struct kvm_vcpu *vcpu);
256 #endif
257
258 #ifdef CONFIG_KVM_GENERIC_MMU_NOTIFIER
259 union kvm_mmu_notifier_arg {
260 unsigned long attributes;
261 };
262
263 struct kvm_gfn_range {
264 struct kvm_memory_slot *slot;
265 gfn_t start;
266 gfn_t end;
267 union kvm_mmu_notifier_arg arg;
268 bool may_block;
269 };
270 bool kvm_unmap_gfn_range(struct kvm *kvm, struct kvm_gfn_range *range);
271 bool kvm_age_gfn(struct kvm *kvm, struct kvm_gfn_range *range);
272 bool kvm_test_age_gfn(struct kvm *kvm, struct kvm_gfn_range *range);
273 #endif
274
275 enum {
276 OUTSIDE_GUEST_MODE,
277 IN_GUEST_MODE,
278 EXITING_GUEST_MODE,
279 READING_SHADOW_PAGE_TABLES,
280 };
281
282 #define KVM_UNMAPPED_PAGE ((void *) 0x500 + POISON_POINTER_DELTA)
283
284 struct kvm_host_map {
285 /*
286 * Only valid if the 'pfn' is managed by the host kernel (i.e. There is
287 * a 'struct page' for it. When using mem= kernel parameter some memory
288 * can be used as guest memory but they are not managed by host
289 * kernel).
290 * If 'pfn' is not managed by the host kernel, this field is
291 * initialized to KVM_UNMAPPED_PAGE.
292 */
293 struct page *page;
294 void *hva;
295 kvm_pfn_t pfn;
296 kvm_pfn_t gfn;
297 };
298
299 /*
300 * Used to check if the mapping is valid or not. Never use 'kvm_host_map'
301 * directly to check for that.
302 */
kvm_vcpu_mapped(struct kvm_host_map * map)303 static inline bool kvm_vcpu_mapped(struct kvm_host_map *map)
304 {
305 return !!map->hva;
306 }
307
kvm_vcpu_can_poll(ktime_t cur,ktime_t stop)308 static inline bool kvm_vcpu_can_poll(ktime_t cur, ktime_t stop)
309 {
310 return single_task_running() && !need_resched() && ktime_before(cur, stop);
311 }
312
313 /*
314 * Sometimes a large or cross-page mmio needs to be broken up into separate
315 * exits for userspace servicing.
316 */
317 struct kvm_mmio_fragment {
318 gpa_t gpa;
319 void *data;
320 unsigned len;
321 };
322
323 struct kvm_vcpu {
324 struct kvm *kvm;
325 #ifdef CONFIG_PREEMPT_NOTIFIERS
326 struct preempt_notifier preempt_notifier;
327 #endif
328 int cpu;
329 int vcpu_id; /* id given by userspace at creation */
330 int vcpu_idx; /* index into kvm->vcpu_array */
331 int ____srcu_idx; /* Don't use this directly. You've been warned. */
332 #ifdef CONFIG_PROVE_RCU
333 int srcu_depth;
334 #endif
335 int mode;
336 u64 requests;
337 unsigned long guest_debug;
338
339 struct mutex mutex;
340 struct kvm_run *run;
341
342 #ifndef __KVM_HAVE_ARCH_WQP
343 struct rcuwait wait;
344 #endif
345 struct pid __rcu *pid;
346 int sigset_active;
347 sigset_t sigset;
348 unsigned int halt_poll_ns;
349 bool valid_wakeup;
350
351 #ifdef CONFIG_HAS_IOMEM
352 int mmio_needed;
353 int mmio_read_completed;
354 int mmio_is_write;
355 int mmio_cur_fragment;
356 int mmio_nr_fragments;
357 struct kvm_mmio_fragment mmio_fragments[KVM_MAX_MMIO_FRAGMENTS];
358 #endif
359
360 #ifdef CONFIG_KVM_ASYNC_PF
361 struct {
362 u32 queued;
363 struct list_head queue;
364 struct list_head done;
365 spinlock_t lock;
366 } async_pf;
367 #endif
368
369 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
370 /*
371 * Cpu relax intercept or pause loop exit optimization
372 * in_spin_loop: set when a vcpu does a pause loop exit
373 * or cpu relax intercepted.
374 * dy_eligible: indicates whether vcpu is eligible for directed yield.
375 */
376 struct {
377 bool in_spin_loop;
378 bool dy_eligible;
379 } spin_loop;
380 #endif
381 bool wants_to_run;
382 bool preempted;
383 bool ready;
384 bool scheduled_out;
385 struct kvm_vcpu_arch arch;
386 struct kvm_vcpu_stat stat;
387 char stats_id[KVM_STATS_NAME_SIZE];
388 struct kvm_dirty_ring dirty_ring;
389
390 /*
391 * The most recently used memslot by this vCPU and the slots generation
392 * for which it is valid.
393 * No wraparound protection is needed since generations won't overflow in
394 * thousands of years, even assuming 1M memslot operations per second.
395 */
396 struct kvm_memory_slot *last_used_slot;
397 u64 last_used_slot_gen;
398 };
399
400 /*
401 * Start accounting time towards a guest.
402 * Must be called before entering guest context.
403 */
guest_timing_enter_irqoff(void)404 static __always_inline void guest_timing_enter_irqoff(void)
405 {
406 /*
407 * This is running in ioctl context so its safe to assume that it's the
408 * stime pending cputime to flush.
409 */
410 instrumentation_begin();
411 vtime_account_guest_enter();
412 instrumentation_end();
413 }
414
415 /*
416 * Enter guest context and enter an RCU extended quiescent state.
417 *
418 * Between guest_context_enter_irqoff() and guest_context_exit_irqoff() it is
419 * unsafe to use any code which may directly or indirectly use RCU, tracing
420 * (including IRQ flag tracing), or lockdep. All code in this period must be
421 * non-instrumentable.
422 */
guest_context_enter_irqoff(void)423 static __always_inline void guest_context_enter_irqoff(void)
424 {
425 /*
426 * KVM does not hold any references to rcu protected data when it
427 * switches CPU into a guest mode. In fact switching to a guest mode
428 * is very similar to exiting to userspace from rcu point of view. In
429 * addition CPU may stay in a guest mode for quite a long time (up to
430 * one time slice). Lets treat guest mode as quiescent state, just like
431 * we do with user-mode execution.
432 */
433 if (!context_tracking_guest_enter()) {
434 instrumentation_begin();
435 rcu_virt_note_context_switch();
436 instrumentation_end();
437 }
438 }
439
440 /*
441 * Deprecated. Architectures should move to guest_timing_enter_irqoff() and
442 * guest_state_enter_irqoff().
443 */
guest_enter_irqoff(void)444 static __always_inline void guest_enter_irqoff(void)
445 {
446 guest_timing_enter_irqoff();
447 guest_context_enter_irqoff();
448 }
449
450 /**
451 * guest_state_enter_irqoff - Fixup state when entering a guest
452 *
453 * Entry to a guest will enable interrupts, but the kernel state is interrupts
454 * disabled when this is invoked. Also tell RCU about it.
455 *
456 * 1) Trace interrupts on state
457 * 2) Invoke context tracking if enabled to adjust RCU state
458 * 3) Tell lockdep that interrupts are enabled
459 *
460 * Invoked from architecture specific code before entering a guest.
461 * Must be called with interrupts disabled and the caller must be
462 * non-instrumentable.
463 * The caller has to invoke guest_timing_enter_irqoff() before this.
464 *
465 * Note: this is analogous to exit_to_user_mode().
466 */
guest_state_enter_irqoff(void)467 static __always_inline void guest_state_enter_irqoff(void)
468 {
469 instrumentation_begin();
470 trace_hardirqs_on_prepare();
471 lockdep_hardirqs_on_prepare();
472 instrumentation_end();
473
474 guest_context_enter_irqoff();
475 lockdep_hardirqs_on(CALLER_ADDR0);
476 }
477
478 /*
479 * Exit guest context and exit an RCU extended quiescent state.
480 *
481 * Between guest_context_enter_irqoff() and guest_context_exit_irqoff() it is
482 * unsafe to use any code which may directly or indirectly use RCU, tracing
483 * (including IRQ flag tracing), or lockdep. All code in this period must be
484 * non-instrumentable.
485 */
guest_context_exit_irqoff(void)486 static __always_inline void guest_context_exit_irqoff(void)
487 {
488 /*
489 * Guest mode is treated as a quiescent state, see
490 * guest_context_enter_irqoff() for more details.
491 */
492 if (!context_tracking_guest_exit()) {
493 instrumentation_begin();
494 rcu_virt_note_context_switch();
495 instrumentation_end();
496 }
497 }
498
499 /*
500 * Stop accounting time towards a guest.
501 * Must be called after exiting guest context.
502 */
guest_timing_exit_irqoff(void)503 static __always_inline void guest_timing_exit_irqoff(void)
504 {
505 instrumentation_begin();
506 /* Flush the guest cputime we spent on the guest */
507 vtime_account_guest_exit();
508 instrumentation_end();
509 }
510
511 /*
512 * Deprecated. Architectures should move to guest_state_exit_irqoff() and
513 * guest_timing_exit_irqoff().
514 */
guest_exit_irqoff(void)515 static __always_inline void guest_exit_irqoff(void)
516 {
517 guest_context_exit_irqoff();
518 guest_timing_exit_irqoff();
519 }
520
guest_exit(void)521 static inline void guest_exit(void)
522 {
523 unsigned long flags;
524
525 local_irq_save(flags);
526 guest_exit_irqoff();
527 local_irq_restore(flags);
528 }
529
530 /**
531 * guest_state_exit_irqoff - Establish state when returning from guest mode
532 *
533 * Entry from a guest disables interrupts, but guest mode is traced as
534 * interrupts enabled. Also with NO_HZ_FULL RCU might be idle.
535 *
536 * 1) Tell lockdep that interrupts are disabled
537 * 2) Invoke context tracking if enabled to reactivate RCU
538 * 3) Trace interrupts off state
539 *
540 * Invoked from architecture specific code after exiting a guest.
541 * Must be invoked with interrupts disabled and the caller must be
542 * non-instrumentable.
543 * The caller has to invoke guest_timing_exit_irqoff() after this.
544 *
545 * Note: this is analogous to enter_from_user_mode().
546 */
guest_state_exit_irqoff(void)547 static __always_inline void guest_state_exit_irqoff(void)
548 {
549 lockdep_hardirqs_off(CALLER_ADDR0);
550 guest_context_exit_irqoff();
551
552 instrumentation_begin();
553 trace_hardirqs_off_finish();
554 instrumentation_end();
555 }
556
kvm_vcpu_exiting_guest_mode(struct kvm_vcpu * vcpu)557 static inline int kvm_vcpu_exiting_guest_mode(struct kvm_vcpu *vcpu)
558 {
559 /*
560 * The memory barrier ensures a previous write to vcpu->requests cannot
561 * be reordered with the read of vcpu->mode. It pairs with the general
562 * memory barrier following the write of vcpu->mode in VCPU RUN.
563 */
564 smp_mb__before_atomic();
565 return cmpxchg(&vcpu->mode, IN_GUEST_MODE, EXITING_GUEST_MODE);
566 }
567
568 /*
569 * Some of the bitops functions do not support too long bitmaps.
570 * This number must be determined not to exceed such limits.
571 */
572 #define KVM_MEM_MAX_NR_PAGES ((1UL << 31) - 1)
573
574 /*
575 * Since at idle each memslot belongs to two memslot sets it has to contain
576 * two embedded nodes for each data structure that it forms a part of.
577 *
578 * Two memslot sets (one active and one inactive) are necessary so the VM
579 * continues to run on one memslot set while the other is being modified.
580 *
581 * These two memslot sets normally point to the same set of memslots.
582 * They can, however, be desynchronized when performing a memslot management
583 * operation by replacing the memslot to be modified by its copy.
584 * After the operation is complete, both memslot sets once again point to
585 * the same, common set of memslot data.
586 *
587 * The memslots themselves are independent of each other so they can be
588 * individually added or deleted.
589 */
590 struct kvm_memory_slot {
591 struct hlist_node id_node[2];
592 struct interval_tree_node hva_node[2];
593 struct rb_node gfn_node[2];
594 gfn_t base_gfn;
595 unsigned long npages;
596 unsigned long *dirty_bitmap;
597 struct kvm_arch_memory_slot arch;
598 unsigned long userspace_addr;
599 u32 flags;
600 short id;
601 u16 as_id;
602
603 #ifdef CONFIG_KVM_PRIVATE_MEM
604 struct {
605 struct file __rcu *file;
606 pgoff_t pgoff;
607 } gmem;
608 #endif
609 };
610
kvm_slot_can_be_private(const struct kvm_memory_slot * slot)611 static inline bool kvm_slot_can_be_private(const struct kvm_memory_slot *slot)
612 {
613 return slot && (slot->flags & KVM_MEM_GUEST_MEMFD);
614 }
615
kvm_slot_dirty_track_enabled(const struct kvm_memory_slot * slot)616 static inline bool kvm_slot_dirty_track_enabled(const struct kvm_memory_slot *slot)
617 {
618 return slot->flags & KVM_MEM_LOG_DIRTY_PAGES;
619 }
620
kvm_dirty_bitmap_bytes(struct kvm_memory_slot * memslot)621 static inline unsigned long kvm_dirty_bitmap_bytes(struct kvm_memory_slot *memslot)
622 {
623 return ALIGN(memslot->npages, BITS_PER_LONG) / 8;
624 }
625
kvm_second_dirty_bitmap(struct kvm_memory_slot * memslot)626 static inline unsigned long *kvm_second_dirty_bitmap(struct kvm_memory_slot *memslot)
627 {
628 unsigned long len = kvm_dirty_bitmap_bytes(memslot);
629
630 return memslot->dirty_bitmap + len / sizeof(*memslot->dirty_bitmap);
631 }
632
633 #ifndef KVM_DIRTY_LOG_MANUAL_CAPS
634 #define KVM_DIRTY_LOG_MANUAL_CAPS KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE
635 #endif
636
637 struct kvm_s390_adapter_int {
638 u64 ind_addr;
639 u64 summary_addr;
640 u64 ind_offset;
641 u32 summary_offset;
642 u32 adapter_id;
643 };
644
645 struct kvm_hv_sint {
646 u32 vcpu;
647 u32 sint;
648 };
649
650 struct kvm_xen_evtchn {
651 u32 port;
652 u32 vcpu_id;
653 int vcpu_idx;
654 u32 priority;
655 };
656
657 struct kvm_kernel_irq_routing_entry {
658 u32 gsi;
659 u32 type;
660 int (*set)(struct kvm_kernel_irq_routing_entry *e,
661 struct kvm *kvm, int irq_source_id, int level,
662 bool line_status);
663 union {
664 struct {
665 unsigned irqchip;
666 unsigned pin;
667 } irqchip;
668 struct {
669 u32 address_lo;
670 u32 address_hi;
671 u32 data;
672 u32 flags;
673 u32 devid;
674 } msi;
675 struct kvm_s390_adapter_int adapter;
676 struct kvm_hv_sint hv_sint;
677 struct kvm_xen_evtchn xen_evtchn;
678 };
679 struct hlist_node link;
680 };
681
682 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
683 struct kvm_irq_routing_table {
684 int chip[KVM_NR_IRQCHIPS][KVM_IRQCHIP_NUM_PINS];
685 u32 nr_rt_entries;
686 /*
687 * Array indexed by gsi. Each entry contains list of irq chips
688 * the gsi is connected to.
689 */
690 struct hlist_head map[] __counted_by(nr_rt_entries);
691 };
692 #endif
693
694 bool kvm_arch_irqchip_in_kernel(struct kvm *kvm);
695
696 #ifndef KVM_INTERNAL_MEM_SLOTS
697 #define KVM_INTERNAL_MEM_SLOTS 0
698 #endif
699
700 #define KVM_MEM_SLOTS_NUM SHRT_MAX
701 #define KVM_USER_MEM_SLOTS (KVM_MEM_SLOTS_NUM - KVM_INTERNAL_MEM_SLOTS)
702
703 #if KVM_MAX_NR_ADDRESS_SPACES == 1
kvm_arch_nr_memslot_as_ids(struct kvm * kvm)704 static inline int kvm_arch_nr_memslot_as_ids(struct kvm *kvm)
705 {
706 return KVM_MAX_NR_ADDRESS_SPACES;
707 }
708
kvm_arch_vcpu_memslots_id(struct kvm_vcpu * vcpu)709 static inline int kvm_arch_vcpu_memslots_id(struct kvm_vcpu *vcpu)
710 {
711 return 0;
712 }
713 #endif
714
715 /*
716 * Arch code must define kvm_arch_has_private_mem if support for private memory
717 * is enabled.
718 */
719 #if !defined(kvm_arch_has_private_mem) && !IS_ENABLED(CONFIG_KVM_PRIVATE_MEM)
kvm_arch_has_private_mem(struct kvm * kvm)720 static inline bool kvm_arch_has_private_mem(struct kvm *kvm)
721 {
722 return false;
723 }
724 #endif
725
726 #ifndef kvm_arch_has_readonly_mem
kvm_arch_has_readonly_mem(struct kvm * kvm)727 static inline bool kvm_arch_has_readonly_mem(struct kvm *kvm)
728 {
729 return IS_ENABLED(CONFIG_HAVE_KVM_READONLY_MEM);
730 }
731 #endif
732
733 struct kvm_memslots {
734 u64 generation;
735 atomic_long_t last_used_slot;
736 struct rb_root_cached hva_tree;
737 struct rb_root gfn_tree;
738 /*
739 * The mapping table from slot id to memslot.
740 *
741 * 7-bit bucket count matches the size of the old id to index array for
742 * 512 slots, while giving good performance with this slot count.
743 * Higher bucket counts bring only small performance improvements but
744 * always result in higher memory usage (even for lower memslot counts).
745 */
746 DECLARE_HASHTABLE(id_hash, 7);
747 int node_idx;
748 };
749
750 struct kvm {
751 #ifdef KVM_HAVE_MMU_RWLOCK
752 rwlock_t mmu_lock;
753 #else
754 spinlock_t mmu_lock;
755 #endif /* KVM_HAVE_MMU_RWLOCK */
756
757 struct mutex slots_lock;
758
759 /*
760 * Protects the arch-specific fields of struct kvm_memory_slots in
761 * use by the VM. To be used under the slots_lock (above) or in a
762 * kvm->srcu critical section where acquiring the slots_lock would
763 * lead to deadlock with the synchronize_srcu in
764 * kvm_swap_active_memslots().
765 */
766 struct mutex slots_arch_lock;
767 struct mm_struct *mm; /* userspace tied to this vm */
768 unsigned long nr_memslot_pages;
769 /* The two memslot sets - active and inactive (per address space) */
770 struct kvm_memslots __memslots[KVM_MAX_NR_ADDRESS_SPACES][2];
771 /* The current active memslot set for each address space */
772 struct kvm_memslots __rcu *memslots[KVM_MAX_NR_ADDRESS_SPACES];
773 struct xarray vcpu_array;
774 /*
775 * Protected by slots_lock, but can be read outside if an
776 * incorrect answer is acceptable.
777 */
778 atomic_t nr_memslots_dirty_logging;
779
780 /* Used to wait for completion of MMU notifiers. */
781 spinlock_t mn_invalidate_lock;
782 unsigned long mn_active_invalidate_count;
783 struct rcuwait mn_memslots_update_rcuwait;
784
785 /* For management / invalidation of gfn_to_pfn_caches */
786 spinlock_t gpc_lock;
787 struct list_head gpc_list;
788
789 /*
790 * created_vcpus is protected by kvm->lock, and is incremented
791 * at the beginning of KVM_CREATE_VCPU. online_vcpus is only
792 * incremented after storing the kvm_vcpu pointer in vcpus,
793 * and is accessed atomically.
794 */
795 atomic_t online_vcpus;
796 int max_vcpus;
797 int created_vcpus;
798 int last_boosted_vcpu;
799 struct list_head vm_list;
800 struct mutex lock;
801 struct kvm_io_bus __rcu *buses[KVM_NR_BUSES];
802 #ifdef CONFIG_HAVE_KVM_IRQCHIP
803 struct {
804 spinlock_t lock;
805 struct list_head items;
806 /* resampler_list update side is protected by resampler_lock. */
807 struct list_head resampler_list;
808 struct mutex resampler_lock;
809 } irqfds;
810 #endif
811 struct list_head ioeventfds;
812 struct kvm_vm_stat stat;
813 struct kvm_arch arch;
814 refcount_t users_count;
815 #ifdef CONFIG_KVM_MMIO
816 struct kvm_coalesced_mmio_ring *coalesced_mmio_ring;
817 spinlock_t ring_lock;
818 struct list_head coalesced_zones;
819 #endif
820
821 struct mutex irq_lock;
822 #ifdef CONFIG_HAVE_KVM_IRQCHIP
823 /*
824 * Update side is protected by irq_lock.
825 */
826 struct kvm_irq_routing_table __rcu *irq_routing;
827
828 struct hlist_head irq_ack_notifier_list;
829 #endif
830
831 #ifdef CONFIG_KVM_GENERIC_MMU_NOTIFIER
832 struct mmu_notifier mmu_notifier;
833 unsigned long mmu_invalidate_seq;
834 long mmu_invalidate_in_progress;
835 gfn_t mmu_invalidate_range_start;
836 gfn_t mmu_invalidate_range_end;
837 #endif
838 struct list_head devices;
839 u64 manual_dirty_log_protect;
840 struct dentry *debugfs_dentry;
841 struct kvm_stat_data **debugfs_stat_data;
842 struct srcu_struct srcu;
843 struct srcu_struct irq_srcu;
844 pid_t userspace_pid;
845 bool override_halt_poll_ns;
846 unsigned int max_halt_poll_ns;
847 u32 dirty_ring_size;
848 bool dirty_ring_with_bitmap;
849 bool vm_bugged;
850 bool vm_dead;
851
852 #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER
853 struct notifier_block pm_notifier;
854 #endif
855 #ifdef CONFIG_KVM_GENERIC_MEMORY_ATTRIBUTES
856 /* Protected by slots_locks (for writes) and RCU (for reads) */
857 struct xarray mem_attr_array;
858 #endif
859 char stats_id[KVM_STATS_NAME_SIZE];
860 };
861
862 #define kvm_err(fmt, ...) \
863 pr_err("kvm [%i]: " fmt, task_pid_nr(current), ## __VA_ARGS__)
864 #define kvm_info(fmt, ...) \
865 pr_info("kvm [%i]: " fmt, task_pid_nr(current), ## __VA_ARGS__)
866 #define kvm_debug(fmt, ...) \
867 pr_debug("kvm [%i]: " fmt, task_pid_nr(current), ## __VA_ARGS__)
868 #define kvm_debug_ratelimited(fmt, ...) \
869 pr_debug_ratelimited("kvm [%i]: " fmt, task_pid_nr(current), \
870 ## __VA_ARGS__)
871 #define kvm_pr_unimpl(fmt, ...) \
872 pr_err_ratelimited("kvm [%i]: " fmt, \
873 task_tgid_nr(current), ## __VA_ARGS__)
874
875 /* The guest did something we don't support. */
876 #define vcpu_unimpl(vcpu, fmt, ...) \
877 kvm_pr_unimpl("vcpu%i, guest rIP: 0x%lx " fmt, \
878 (vcpu)->vcpu_id, kvm_rip_read(vcpu), ## __VA_ARGS__)
879
880 #define vcpu_debug(vcpu, fmt, ...) \
881 kvm_debug("vcpu%i " fmt, (vcpu)->vcpu_id, ## __VA_ARGS__)
882 #define vcpu_debug_ratelimited(vcpu, fmt, ...) \
883 kvm_debug_ratelimited("vcpu%i " fmt, (vcpu)->vcpu_id, \
884 ## __VA_ARGS__)
885 #define vcpu_err(vcpu, fmt, ...) \
886 kvm_err("vcpu%i " fmt, (vcpu)->vcpu_id, ## __VA_ARGS__)
887
kvm_vm_dead(struct kvm * kvm)888 static inline void kvm_vm_dead(struct kvm *kvm)
889 {
890 kvm->vm_dead = true;
891 kvm_make_all_cpus_request(kvm, KVM_REQ_VM_DEAD);
892 }
893
kvm_vm_bugged(struct kvm * kvm)894 static inline void kvm_vm_bugged(struct kvm *kvm)
895 {
896 kvm->vm_bugged = true;
897 kvm_vm_dead(kvm);
898 }
899
900
901 #define KVM_BUG(cond, kvm, fmt...) \
902 ({ \
903 bool __ret = !!(cond); \
904 \
905 if (WARN_ONCE(__ret && !(kvm)->vm_bugged, fmt)) \
906 kvm_vm_bugged(kvm); \
907 unlikely(__ret); \
908 })
909
910 #define KVM_BUG_ON(cond, kvm) \
911 ({ \
912 bool __ret = !!(cond); \
913 \
914 if (WARN_ON_ONCE(__ret && !(kvm)->vm_bugged)) \
915 kvm_vm_bugged(kvm); \
916 unlikely(__ret); \
917 })
918
919 /*
920 * Note, "data corruption" refers to corruption of host kernel data structures,
921 * not guest data. Guest data corruption, suspected or confirmed, that is tied
922 * and contained to a single VM should *never* BUG() and potentially panic the
923 * host, i.e. use this variant of KVM_BUG() if and only if a KVM data structure
924 * is corrupted and that corruption can have a cascading effect to other parts
925 * of the hosts and/or to other VMs.
926 */
927 #define KVM_BUG_ON_DATA_CORRUPTION(cond, kvm) \
928 ({ \
929 bool __ret = !!(cond); \
930 \
931 if (IS_ENABLED(CONFIG_BUG_ON_DATA_CORRUPTION)) \
932 BUG_ON(__ret); \
933 else if (WARN_ON_ONCE(__ret && !(kvm)->vm_bugged)) \
934 kvm_vm_bugged(kvm); \
935 unlikely(__ret); \
936 })
937
kvm_vcpu_srcu_read_lock(struct kvm_vcpu * vcpu)938 static inline void kvm_vcpu_srcu_read_lock(struct kvm_vcpu *vcpu)
939 {
940 #ifdef CONFIG_PROVE_RCU
941 WARN_ONCE(vcpu->srcu_depth++,
942 "KVM: Illegal vCPU srcu_idx LOCK, depth=%d", vcpu->srcu_depth - 1);
943 #endif
944 vcpu->____srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
945 }
946
kvm_vcpu_srcu_read_unlock(struct kvm_vcpu * vcpu)947 static inline void kvm_vcpu_srcu_read_unlock(struct kvm_vcpu *vcpu)
948 {
949 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->____srcu_idx);
950
951 #ifdef CONFIG_PROVE_RCU
952 WARN_ONCE(--vcpu->srcu_depth,
953 "KVM: Illegal vCPU srcu_idx UNLOCK, depth=%d", vcpu->srcu_depth);
954 #endif
955 }
956
kvm_dirty_log_manual_protect_and_init_set(struct kvm * kvm)957 static inline bool kvm_dirty_log_manual_protect_and_init_set(struct kvm *kvm)
958 {
959 return !!(kvm->manual_dirty_log_protect & KVM_DIRTY_LOG_INITIALLY_SET);
960 }
961
kvm_get_bus(struct kvm * kvm,enum kvm_bus idx)962 static inline struct kvm_io_bus *kvm_get_bus(struct kvm *kvm, enum kvm_bus idx)
963 {
964 return srcu_dereference_check(kvm->buses[idx], &kvm->srcu,
965 lockdep_is_held(&kvm->slots_lock) ||
966 !refcount_read(&kvm->users_count));
967 }
968
kvm_get_vcpu(struct kvm * kvm,int i)969 static inline struct kvm_vcpu *kvm_get_vcpu(struct kvm *kvm, int i)
970 {
971 int num_vcpus = atomic_read(&kvm->online_vcpus);
972 i = array_index_nospec(i, num_vcpus);
973
974 /* Pairs with smp_wmb() in kvm_vm_ioctl_create_vcpu. */
975 smp_rmb();
976 return xa_load(&kvm->vcpu_array, i);
977 }
978
979 #define kvm_for_each_vcpu(idx, vcpup, kvm) \
980 xa_for_each_range(&kvm->vcpu_array, idx, vcpup, 0, \
981 (atomic_read(&kvm->online_vcpus) - 1))
982
kvm_get_vcpu_by_id(struct kvm * kvm,int id)983 static inline struct kvm_vcpu *kvm_get_vcpu_by_id(struct kvm *kvm, int id)
984 {
985 struct kvm_vcpu *vcpu = NULL;
986 unsigned long i;
987
988 if (id < 0)
989 return NULL;
990 if (id < KVM_MAX_VCPUS)
991 vcpu = kvm_get_vcpu(kvm, id);
992 if (vcpu && vcpu->vcpu_id == id)
993 return vcpu;
994 kvm_for_each_vcpu(i, vcpu, kvm)
995 if (vcpu->vcpu_id == id)
996 return vcpu;
997 return NULL;
998 }
999
1000 void kvm_destroy_vcpus(struct kvm *kvm);
1001
1002 void vcpu_load(struct kvm_vcpu *vcpu);
1003 void vcpu_put(struct kvm_vcpu *vcpu);
1004
1005 #ifdef __KVM_HAVE_IOAPIC
1006 void kvm_arch_post_irq_ack_notifier_list_update(struct kvm *kvm);
1007 void kvm_arch_post_irq_routing_update(struct kvm *kvm);
1008 #else
kvm_arch_post_irq_ack_notifier_list_update(struct kvm * kvm)1009 static inline void kvm_arch_post_irq_ack_notifier_list_update(struct kvm *kvm)
1010 {
1011 }
kvm_arch_post_irq_routing_update(struct kvm * kvm)1012 static inline void kvm_arch_post_irq_routing_update(struct kvm *kvm)
1013 {
1014 }
1015 #endif
1016
1017 #ifdef CONFIG_HAVE_KVM_IRQCHIP
1018 int kvm_irqfd_init(void);
1019 void kvm_irqfd_exit(void);
1020 #else
kvm_irqfd_init(void)1021 static inline int kvm_irqfd_init(void)
1022 {
1023 return 0;
1024 }
1025
kvm_irqfd_exit(void)1026 static inline void kvm_irqfd_exit(void)
1027 {
1028 }
1029 #endif
1030 int kvm_init(unsigned vcpu_size, unsigned vcpu_align, struct module *module);
1031 void kvm_exit(void);
1032
1033 void kvm_get_kvm(struct kvm *kvm);
1034 bool kvm_get_kvm_safe(struct kvm *kvm);
1035 void kvm_put_kvm(struct kvm *kvm);
1036 bool file_is_kvm(struct file *file);
1037 void kvm_put_kvm_no_destroy(struct kvm *kvm);
1038
__kvm_memslots(struct kvm * kvm,int as_id)1039 static inline struct kvm_memslots *__kvm_memslots(struct kvm *kvm, int as_id)
1040 {
1041 as_id = array_index_nospec(as_id, KVM_MAX_NR_ADDRESS_SPACES);
1042 return srcu_dereference_check(kvm->memslots[as_id], &kvm->srcu,
1043 lockdep_is_held(&kvm->slots_lock) ||
1044 !refcount_read(&kvm->users_count));
1045 }
1046
kvm_memslots(struct kvm * kvm)1047 static inline struct kvm_memslots *kvm_memslots(struct kvm *kvm)
1048 {
1049 return __kvm_memslots(kvm, 0);
1050 }
1051
kvm_vcpu_memslots(struct kvm_vcpu * vcpu)1052 static inline struct kvm_memslots *kvm_vcpu_memslots(struct kvm_vcpu *vcpu)
1053 {
1054 int as_id = kvm_arch_vcpu_memslots_id(vcpu);
1055
1056 return __kvm_memslots(vcpu->kvm, as_id);
1057 }
1058
kvm_memslots_empty(struct kvm_memslots * slots)1059 static inline bool kvm_memslots_empty(struct kvm_memslots *slots)
1060 {
1061 return RB_EMPTY_ROOT(&slots->gfn_tree);
1062 }
1063
1064 bool kvm_are_all_memslots_empty(struct kvm *kvm);
1065
1066 #define kvm_for_each_memslot(memslot, bkt, slots) \
1067 hash_for_each(slots->id_hash, bkt, memslot, id_node[slots->node_idx]) \
1068 if (WARN_ON_ONCE(!memslot->npages)) { \
1069 } else
1070
1071 static inline
id_to_memslot(struct kvm_memslots * slots,int id)1072 struct kvm_memory_slot *id_to_memslot(struct kvm_memslots *slots, int id)
1073 {
1074 struct kvm_memory_slot *slot;
1075 int idx = slots->node_idx;
1076
1077 hash_for_each_possible(slots->id_hash, slot, id_node[idx], id) {
1078 if (slot->id == id)
1079 return slot;
1080 }
1081
1082 return NULL;
1083 }
1084
1085 /* Iterator used for walking memslots that overlap a gfn range. */
1086 struct kvm_memslot_iter {
1087 struct kvm_memslots *slots;
1088 struct rb_node *node;
1089 struct kvm_memory_slot *slot;
1090 };
1091
kvm_memslot_iter_next(struct kvm_memslot_iter * iter)1092 static inline void kvm_memslot_iter_next(struct kvm_memslot_iter *iter)
1093 {
1094 iter->node = rb_next(iter->node);
1095 if (!iter->node)
1096 return;
1097
1098 iter->slot = container_of(iter->node, struct kvm_memory_slot, gfn_node[iter->slots->node_idx]);
1099 }
1100
kvm_memslot_iter_start(struct kvm_memslot_iter * iter,struct kvm_memslots * slots,gfn_t start)1101 static inline void kvm_memslot_iter_start(struct kvm_memslot_iter *iter,
1102 struct kvm_memslots *slots,
1103 gfn_t start)
1104 {
1105 int idx = slots->node_idx;
1106 struct rb_node *tmp;
1107 struct kvm_memory_slot *slot;
1108
1109 iter->slots = slots;
1110
1111 /*
1112 * Find the so called "upper bound" of a key - the first node that has
1113 * its key strictly greater than the searched one (the start gfn in our case).
1114 */
1115 iter->node = NULL;
1116 for (tmp = slots->gfn_tree.rb_node; tmp; ) {
1117 slot = container_of(tmp, struct kvm_memory_slot, gfn_node[idx]);
1118 if (start < slot->base_gfn) {
1119 iter->node = tmp;
1120 tmp = tmp->rb_left;
1121 } else {
1122 tmp = tmp->rb_right;
1123 }
1124 }
1125
1126 /*
1127 * Find the slot with the lowest gfn that can possibly intersect with
1128 * the range, so we'll ideally have slot start <= range start
1129 */
1130 if (iter->node) {
1131 /*
1132 * A NULL previous node means that the very first slot
1133 * already has a higher start gfn.
1134 * In this case slot start > range start.
1135 */
1136 tmp = rb_prev(iter->node);
1137 if (tmp)
1138 iter->node = tmp;
1139 } else {
1140 /* a NULL node below means no slots */
1141 iter->node = rb_last(&slots->gfn_tree);
1142 }
1143
1144 if (iter->node) {
1145 iter->slot = container_of(iter->node, struct kvm_memory_slot, gfn_node[idx]);
1146
1147 /*
1148 * It is possible in the slot start < range start case that the
1149 * found slot ends before or at range start (slot end <= range start)
1150 * and so it does not overlap the requested range.
1151 *
1152 * In such non-overlapping case the next slot (if it exists) will
1153 * already have slot start > range start, otherwise the logic above
1154 * would have found it instead of the current slot.
1155 */
1156 if (iter->slot->base_gfn + iter->slot->npages <= start)
1157 kvm_memslot_iter_next(iter);
1158 }
1159 }
1160
kvm_memslot_iter_is_valid(struct kvm_memslot_iter * iter,gfn_t end)1161 static inline bool kvm_memslot_iter_is_valid(struct kvm_memslot_iter *iter, gfn_t end)
1162 {
1163 if (!iter->node)
1164 return false;
1165
1166 /*
1167 * If this slot starts beyond or at the end of the range so does
1168 * every next one
1169 */
1170 return iter->slot->base_gfn < end;
1171 }
1172
1173 /* Iterate over each memslot at least partially intersecting [start, end) range */
1174 #define kvm_for_each_memslot_in_gfn_range(iter, slots, start, end) \
1175 for (kvm_memslot_iter_start(iter, slots, start); \
1176 kvm_memslot_iter_is_valid(iter, end); \
1177 kvm_memslot_iter_next(iter))
1178
1179 /*
1180 * KVM_SET_USER_MEMORY_REGION ioctl allows the following operations:
1181 * - create a new memory slot
1182 * - delete an existing memory slot
1183 * - modify an existing memory slot
1184 * -- move it in the guest physical memory space
1185 * -- just change its flags
1186 *
1187 * Since flags can be changed by some of these operations, the following
1188 * differentiation is the best we can do for __kvm_set_memory_region():
1189 */
1190 enum kvm_mr_change {
1191 KVM_MR_CREATE,
1192 KVM_MR_DELETE,
1193 KVM_MR_MOVE,
1194 KVM_MR_FLAGS_ONLY,
1195 };
1196
1197 int kvm_set_memory_region(struct kvm *kvm,
1198 const struct kvm_userspace_memory_region2 *mem);
1199 int __kvm_set_memory_region(struct kvm *kvm,
1200 const struct kvm_userspace_memory_region2 *mem);
1201 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot);
1202 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen);
1203 int kvm_arch_prepare_memory_region(struct kvm *kvm,
1204 const struct kvm_memory_slot *old,
1205 struct kvm_memory_slot *new,
1206 enum kvm_mr_change change);
1207 void kvm_arch_commit_memory_region(struct kvm *kvm,
1208 struct kvm_memory_slot *old,
1209 const struct kvm_memory_slot *new,
1210 enum kvm_mr_change change);
1211 /* flush all memory translations */
1212 void kvm_arch_flush_shadow_all(struct kvm *kvm);
1213 /* flush memory translations pointing to 'slot' */
1214 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
1215 struct kvm_memory_slot *slot);
1216
1217 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1218 struct page **pages, int nr_pages);
1219
1220 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn);
1221 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn);
1222 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable);
1223 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot, gfn_t gfn);
1224 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot, gfn_t gfn,
1225 bool *writable);
1226 void kvm_release_page_clean(struct page *page);
1227 void kvm_release_page_dirty(struct page *page);
1228
1229 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn);
1230 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1231 bool *writable);
1232 kvm_pfn_t gfn_to_pfn_memslot(const struct kvm_memory_slot *slot, gfn_t gfn);
1233 kvm_pfn_t gfn_to_pfn_memslot_atomic(const struct kvm_memory_slot *slot, gfn_t gfn);
1234 kvm_pfn_t __gfn_to_pfn_memslot(const struct kvm_memory_slot *slot, gfn_t gfn,
1235 bool atomic, bool interruptible, bool *async,
1236 bool write_fault, bool *writable, hva_t *hva);
1237
1238 void kvm_release_pfn_clean(kvm_pfn_t pfn);
1239 void kvm_release_pfn_dirty(kvm_pfn_t pfn);
1240 void kvm_set_pfn_dirty(kvm_pfn_t pfn);
1241 void kvm_set_pfn_accessed(kvm_pfn_t pfn);
1242
1243 void kvm_release_pfn(kvm_pfn_t pfn, bool dirty);
1244 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1245 int len);
1246 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len);
1247 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1248 void *data, unsigned long len);
1249 int kvm_read_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1250 void *data, unsigned int offset,
1251 unsigned long len);
1252 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1253 int offset, int len);
1254 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1255 unsigned long len);
1256 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1257 void *data, unsigned long len);
1258 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1259 void *data, unsigned int offset,
1260 unsigned long len);
1261 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1262 gpa_t gpa, unsigned long len);
1263
1264 #define __kvm_get_guest(kvm, gfn, offset, v) \
1265 ({ \
1266 unsigned long __addr = gfn_to_hva(kvm, gfn); \
1267 typeof(v) __user *__uaddr = (typeof(__uaddr))(__addr + offset); \
1268 int __ret = -EFAULT; \
1269 \
1270 if (!kvm_is_error_hva(__addr)) \
1271 __ret = get_user(v, __uaddr); \
1272 __ret; \
1273 })
1274
1275 #define kvm_get_guest(kvm, gpa, v) \
1276 ({ \
1277 gpa_t __gpa = gpa; \
1278 struct kvm *__kvm = kvm; \
1279 \
1280 __kvm_get_guest(__kvm, __gpa >> PAGE_SHIFT, \
1281 offset_in_page(__gpa), v); \
1282 })
1283
1284 #define __kvm_put_guest(kvm, gfn, offset, v) \
1285 ({ \
1286 unsigned long __addr = gfn_to_hva(kvm, gfn); \
1287 typeof(v) __user *__uaddr = (typeof(__uaddr))(__addr + offset); \
1288 int __ret = -EFAULT; \
1289 \
1290 if (!kvm_is_error_hva(__addr)) \
1291 __ret = put_user(v, __uaddr); \
1292 if (!__ret) \
1293 mark_page_dirty(kvm, gfn); \
1294 __ret; \
1295 })
1296
1297 #define kvm_put_guest(kvm, gpa, v) \
1298 ({ \
1299 gpa_t __gpa = gpa; \
1300 struct kvm *__kvm = kvm; \
1301 \
1302 __kvm_put_guest(__kvm, __gpa >> PAGE_SHIFT, \
1303 offset_in_page(__gpa), v); \
1304 })
1305
1306 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len);
1307 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn);
1308 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn);
1309 bool kvm_vcpu_is_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn);
1310 unsigned long kvm_host_page_size(struct kvm_vcpu *vcpu, gfn_t gfn);
1311 void mark_page_dirty_in_slot(struct kvm *kvm, const struct kvm_memory_slot *memslot, gfn_t gfn);
1312 void mark_page_dirty(struct kvm *kvm, gfn_t gfn);
1313
1314 struct kvm_memslots *kvm_vcpu_memslots(struct kvm_vcpu *vcpu);
1315 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn);
1316 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn);
1317 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn);
1318 int kvm_vcpu_map(struct kvm_vcpu *vcpu, gpa_t gpa, struct kvm_host_map *map);
1319 void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map, bool dirty);
1320 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn);
1321 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable);
1322 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data, int offset,
1323 int len);
1324 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa, void *data,
1325 unsigned long len);
1326 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data,
1327 unsigned long len);
1328 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, const void *data,
1329 int offset, int len);
1330 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
1331 unsigned long len);
1332 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn);
1333
1334 /**
1335 * kvm_gpc_init - initialize gfn_to_pfn_cache.
1336 *
1337 * @gpc: struct gfn_to_pfn_cache object.
1338 * @kvm: pointer to kvm instance.
1339 *
1340 * This sets up a gfn_to_pfn_cache by initializing locks and assigning the
1341 * immutable attributes. Note, the cache must be zero-allocated (or zeroed by
1342 * the caller before init).
1343 */
1344 void kvm_gpc_init(struct gfn_to_pfn_cache *gpc, struct kvm *kvm);
1345
1346 /**
1347 * kvm_gpc_activate - prepare a cached kernel mapping and HPA for a given guest
1348 * physical address.
1349 *
1350 * @gpc: struct gfn_to_pfn_cache object.
1351 * @gpa: guest physical address to map.
1352 * @len: sanity check; the range being access must fit a single page.
1353 *
1354 * @return: 0 for success.
1355 * -EINVAL for a mapping which would cross a page boundary.
1356 * -EFAULT for an untranslatable guest physical address.
1357 *
1358 * This primes a gfn_to_pfn_cache and links it into the @gpc->kvm's list for
1359 * invalidations to be processed. Callers are required to use kvm_gpc_check()
1360 * to ensure that the cache is valid before accessing the target page.
1361 */
1362 int kvm_gpc_activate(struct gfn_to_pfn_cache *gpc, gpa_t gpa, unsigned long len);
1363
1364 /**
1365 * kvm_gpc_activate_hva - prepare a cached kernel mapping and HPA for a given HVA.
1366 *
1367 * @gpc: struct gfn_to_pfn_cache object.
1368 * @hva: userspace virtual address to map.
1369 * @len: sanity check; the range being access must fit a single page.
1370 *
1371 * @return: 0 for success.
1372 * -EINVAL for a mapping which would cross a page boundary.
1373 * -EFAULT for an untranslatable guest physical address.
1374 *
1375 * The semantics of this function are the same as those of kvm_gpc_activate(). It
1376 * merely bypasses a layer of address translation.
1377 */
1378 int kvm_gpc_activate_hva(struct gfn_to_pfn_cache *gpc, unsigned long hva, unsigned long len);
1379
1380 /**
1381 * kvm_gpc_check - check validity of a gfn_to_pfn_cache.
1382 *
1383 * @gpc: struct gfn_to_pfn_cache object.
1384 * @len: sanity check; the range being access must fit a single page.
1385 *
1386 * @return: %true if the cache is still valid and the address matches.
1387 * %false if the cache is not valid.
1388 *
1389 * Callers outside IN_GUEST_MODE context should hold a read lock on @gpc->lock
1390 * while calling this function, and then continue to hold the lock until the
1391 * access is complete.
1392 *
1393 * Callers in IN_GUEST_MODE may do so without locking, although they should
1394 * still hold a read lock on kvm->scru for the memslot checks.
1395 */
1396 bool kvm_gpc_check(struct gfn_to_pfn_cache *gpc, unsigned long len);
1397
1398 /**
1399 * kvm_gpc_refresh - update a previously initialized cache.
1400 *
1401 * @gpc: struct gfn_to_pfn_cache object.
1402 * @len: sanity check; the range being access must fit a single page.
1403 *
1404 * @return: 0 for success.
1405 * -EINVAL for a mapping which would cross a page boundary.
1406 * -EFAULT for an untranslatable guest physical address.
1407 *
1408 * This will attempt to refresh a gfn_to_pfn_cache. Note that a successful
1409 * return from this function does not mean the page can be immediately
1410 * accessed because it may have raced with an invalidation. Callers must
1411 * still lock and check the cache status, as this function does not return
1412 * with the lock still held to permit access.
1413 */
1414 int kvm_gpc_refresh(struct gfn_to_pfn_cache *gpc, unsigned long len);
1415
1416 /**
1417 * kvm_gpc_deactivate - deactivate and unlink a gfn_to_pfn_cache.
1418 *
1419 * @gpc: struct gfn_to_pfn_cache object.
1420 *
1421 * This removes a cache from the VM's list to be processed on MMU notifier
1422 * invocation.
1423 */
1424 void kvm_gpc_deactivate(struct gfn_to_pfn_cache *gpc);
1425
kvm_gpc_is_gpa_active(struct gfn_to_pfn_cache * gpc)1426 static inline bool kvm_gpc_is_gpa_active(struct gfn_to_pfn_cache *gpc)
1427 {
1428 return gpc->active && !kvm_is_error_gpa(gpc->gpa);
1429 }
1430
kvm_gpc_is_hva_active(struct gfn_to_pfn_cache * gpc)1431 static inline bool kvm_gpc_is_hva_active(struct gfn_to_pfn_cache *gpc)
1432 {
1433 return gpc->active && kvm_is_error_gpa(gpc->gpa);
1434 }
1435
1436 void kvm_sigset_activate(struct kvm_vcpu *vcpu);
1437 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu);
1438
1439 void kvm_vcpu_halt(struct kvm_vcpu *vcpu);
1440 bool kvm_vcpu_block(struct kvm_vcpu *vcpu);
1441 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu);
1442 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu);
1443 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu);
1444 void kvm_vcpu_kick(struct kvm_vcpu *vcpu);
1445 int kvm_vcpu_yield_to(struct kvm_vcpu *target);
1446 void kvm_vcpu_on_spin(struct kvm_vcpu *vcpu, bool yield_to_kernel_mode);
1447
1448 void kvm_flush_remote_tlbs(struct kvm *kvm);
1449 void kvm_flush_remote_tlbs_range(struct kvm *kvm, gfn_t gfn, u64 nr_pages);
1450 void kvm_flush_remote_tlbs_memslot(struct kvm *kvm,
1451 const struct kvm_memory_slot *memslot);
1452
1453 #ifdef KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE
1454 int kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache *mc, int min);
1455 int __kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache *mc, int capacity, int min);
1456 int kvm_mmu_memory_cache_nr_free_objects(struct kvm_mmu_memory_cache *mc);
1457 void kvm_mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc);
1458 void *kvm_mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc);
1459 #endif
1460
1461 void kvm_mmu_invalidate_begin(struct kvm *kvm);
1462 void kvm_mmu_invalidate_range_add(struct kvm *kvm, gfn_t start, gfn_t end);
1463 void kvm_mmu_invalidate_end(struct kvm *kvm);
1464 bool kvm_mmu_unmap_gfn_range(struct kvm *kvm, struct kvm_gfn_range *range);
1465
1466 long kvm_arch_dev_ioctl(struct file *filp,
1467 unsigned int ioctl, unsigned long arg);
1468 long kvm_arch_vcpu_ioctl(struct file *filp,
1469 unsigned int ioctl, unsigned long arg);
1470 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf);
1471
1472 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext);
1473
1474 void kvm_arch_mmu_enable_log_dirty_pt_masked(struct kvm *kvm,
1475 struct kvm_memory_slot *slot,
1476 gfn_t gfn_offset,
1477 unsigned long mask);
1478 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot);
1479
1480 #ifndef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1481 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log);
1482 int kvm_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log,
1483 int *is_dirty, struct kvm_memory_slot **memslot);
1484 #endif
1485
1486 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
1487 bool line_status);
1488 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
1489 struct kvm_enable_cap *cap);
1490 int kvm_arch_vm_ioctl(struct file *filp, unsigned int ioctl, unsigned long arg);
1491 long kvm_arch_vm_compat_ioctl(struct file *filp, unsigned int ioctl,
1492 unsigned long arg);
1493
1494 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu);
1495 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu);
1496
1497 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
1498 struct kvm_translation *tr);
1499
1500 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs);
1501 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs);
1502 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
1503 struct kvm_sregs *sregs);
1504 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
1505 struct kvm_sregs *sregs);
1506 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
1507 struct kvm_mp_state *mp_state);
1508 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
1509 struct kvm_mp_state *mp_state);
1510 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
1511 struct kvm_guest_debug *dbg);
1512 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu);
1513
1514 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu);
1515 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu);
1516 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id);
1517 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu);
1518 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu);
1519 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu);
1520
1521 #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER
1522 int kvm_arch_pm_notifier(struct kvm *kvm, unsigned long state);
1523 #endif
1524
1525 #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS
1526 void kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu, struct dentry *debugfs_dentry);
1527 #else
kvm_create_vcpu_debugfs(struct kvm_vcpu * vcpu)1528 static inline void kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu) {}
1529 #endif
1530
1531 #ifdef CONFIG_KVM_GENERIC_HARDWARE_ENABLING
1532 /*
1533 * kvm_arch_{enable,disable}_virtualization() are called on one CPU, under
1534 * kvm_usage_lock, immediately after/before 0=>1 and 1=>0 transitions of
1535 * kvm_usage_count, i.e. at the beginning of the generic hardware enabling
1536 * sequence, and at the end of the generic hardware disabling sequence.
1537 */
1538 void kvm_arch_enable_virtualization(void);
1539 void kvm_arch_disable_virtualization(void);
1540 /*
1541 * kvm_arch_{enable,disable}_virtualization_cpu() are called on "every" CPU to
1542 * do the actual twiddling of hardware bits. The hooks are called on all
1543 * online CPUs when KVM enables/disabled virtualization, and on a single CPU
1544 * when that CPU is onlined/offlined (including for Resume/Suspend).
1545 */
1546 int kvm_arch_enable_virtualization_cpu(void);
1547 void kvm_arch_disable_virtualization_cpu(void);
1548 #endif
1549 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu);
1550 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu);
1551 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu);
1552 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu);
1553 bool kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu);
1554 bool kvm_arch_vcpu_preempted_in_kernel(struct kvm_vcpu *vcpu);
1555 int kvm_arch_post_init_vm(struct kvm *kvm);
1556 void kvm_arch_pre_destroy_vm(struct kvm *kvm);
1557 void kvm_arch_create_vm_debugfs(struct kvm *kvm);
1558
1559 #ifndef __KVM_HAVE_ARCH_VM_ALLOC
1560 /*
1561 * All architectures that want to use vzalloc currently also
1562 * need their own kvm_arch_alloc_vm implementation.
1563 */
kvm_arch_alloc_vm(void)1564 static inline struct kvm *kvm_arch_alloc_vm(void)
1565 {
1566 return kzalloc(sizeof(struct kvm), GFP_KERNEL_ACCOUNT);
1567 }
1568 #endif
1569
__kvm_arch_free_vm(struct kvm * kvm)1570 static inline void __kvm_arch_free_vm(struct kvm *kvm)
1571 {
1572 kvfree(kvm);
1573 }
1574
1575 #ifndef __KVM_HAVE_ARCH_VM_FREE
kvm_arch_free_vm(struct kvm * kvm)1576 static inline void kvm_arch_free_vm(struct kvm *kvm)
1577 {
1578 __kvm_arch_free_vm(kvm);
1579 }
1580 #endif
1581
1582 #ifndef __KVM_HAVE_ARCH_FLUSH_REMOTE_TLBS
kvm_arch_flush_remote_tlbs(struct kvm * kvm)1583 static inline int kvm_arch_flush_remote_tlbs(struct kvm *kvm)
1584 {
1585 return -ENOTSUPP;
1586 }
1587 #else
1588 int kvm_arch_flush_remote_tlbs(struct kvm *kvm);
1589 #endif
1590
1591 #ifndef __KVM_HAVE_ARCH_FLUSH_REMOTE_TLBS_RANGE
kvm_arch_flush_remote_tlbs_range(struct kvm * kvm,gfn_t gfn,u64 nr_pages)1592 static inline int kvm_arch_flush_remote_tlbs_range(struct kvm *kvm,
1593 gfn_t gfn, u64 nr_pages)
1594 {
1595 return -EOPNOTSUPP;
1596 }
1597 #else
1598 int kvm_arch_flush_remote_tlbs_range(struct kvm *kvm, gfn_t gfn, u64 nr_pages);
1599 #endif
1600
1601 #ifdef __KVM_HAVE_ARCH_NONCOHERENT_DMA
1602 void kvm_arch_register_noncoherent_dma(struct kvm *kvm);
1603 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm);
1604 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm);
1605 #else
kvm_arch_register_noncoherent_dma(struct kvm * kvm)1606 static inline void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
1607 {
1608 }
1609
kvm_arch_unregister_noncoherent_dma(struct kvm * kvm)1610 static inline void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
1611 {
1612 }
1613
kvm_arch_has_noncoherent_dma(struct kvm * kvm)1614 static inline bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
1615 {
1616 return false;
1617 }
1618 #endif
1619 #ifdef __KVM_HAVE_ARCH_ASSIGNED_DEVICE
1620 void kvm_arch_start_assignment(struct kvm *kvm);
1621 void kvm_arch_end_assignment(struct kvm *kvm);
1622 bool kvm_arch_has_assigned_device(struct kvm *kvm);
1623 #else
kvm_arch_start_assignment(struct kvm * kvm)1624 static inline void kvm_arch_start_assignment(struct kvm *kvm)
1625 {
1626 }
1627
kvm_arch_end_assignment(struct kvm * kvm)1628 static inline void kvm_arch_end_assignment(struct kvm *kvm)
1629 {
1630 }
1631
kvm_arch_has_assigned_device(struct kvm * kvm)1632 static __always_inline bool kvm_arch_has_assigned_device(struct kvm *kvm)
1633 {
1634 return false;
1635 }
1636 #endif
1637
kvm_arch_vcpu_get_wait(struct kvm_vcpu * vcpu)1638 static inline struct rcuwait *kvm_arch_vcpu_get_wait(struct kvm_vcpu *vcpu)
1639 {
1640 #ifdef __KVM_HAVE_ARCH_WQP
1641 return vcpu->arch.waitp;
1642 #else
1643 return &vcpu->wait;
1644 #endif
1645 }
1646
1647 /*
1648 * Wake a vCPU if necessary, but don't do any stats/metadata updates. Returns
1649 * true if the vCPU was blocking and was awakened, false otherwise.
1650 */
__kvm_vcpu_wake_up(struct kvm_vcpu * vcpu)1651 static inline bool __kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
1652 {
1653 return !!rcuwait_wake_up(kvm_arch_vcpu_get_wait(vcpu));
1654 }
1655
kvm_vcpu_is_blocking(struct kvm_vcpu * vcpu)1656 static inline bool kvm_vcpu_is_blocking(struct kvm_vcpu *vcpu)
1657 {
1658 return rcuwait_active(kvm_arch_vcpu_get_wait(vcpu));
1659 }
1660
1661 #ifdef __KVM_HAVE_ARCH_INTC_INITIALIZED
1662 /*
1663 * returns true if the virtual interrupt controller is initialized and
1664 * ready to accept virtual IRQ. On some architectures the virtual interrupt
1665 * controller is dynamically instantiated and this is not always true.
1666 */
1667 bool kvm_arch_intc_initialized(struct kvm *kvm);
1668 #else
kvm_arch_intc_initialized(struct kvm * kvm)1669 static inline bool kvm_arch_intc_initialized(struct kvm *kvm)
1670 {
1671 return true;
1672 }
1673 #endif
1674
1675 #ifdef CONFIG_GUEST_PERF_EVENTS
1676 unsigned long kvm_arch_vcpu_get_ip(struct kvm_vcpu *vcpu);
1677
1678 void kvm_register_perf_callbacks(unsigned int (*pt_intr_handler)(void));
1679 void kvm_unregister_perf_callbacks(void);
1680 #else
kvm_register_perf_callbacks(void * ign)1681 static inline void kvm_register_perf_callbacks(void *ign) {}
kvm_unregister_perf_callbacks(void)1682 static inline void kvm_unregister_perf_callbacks(void) {}
1683 #endif /* CONFIG_GUEST_PERF_EVENTS */
1684
1685 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type);
1686 void kvm_arch_destroy_vm(struct kvm *kvm);
1687 void kvm_arch_sync_events(struct kvm *kvm);
1688
1689 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu);
1690
1691 struct page *kvm_pfn_to_refcounted_page(kvm_pfn_t pfn);
1692 bool kvm_is_zone_device_page(struct page *page);
1693
1694 struct kvm_irq_ack_notifier {
1695 struct hlist_node link;
1696 unsigned gsi;
1697 void (*irq_acked)(struct kvm_irq_ack_notifier *kian);
1698 };
1699
1700 int kvm_irq_map_gsi(struct kvm *kvm,
1701 struct kvm_kernel_irq_routing_entry *entries, int gsi);
1702 int kvm_irq_map_chip_pin(struct kvm *kvm, unsigned irqchip, unsigned pin);
1703
1704 int kvm_set_irq(struct kvm *kvm, int irq_source_id, u32 irq, int level,
1705 bool line_status);
1706 int kvm_set_msi(struct kvm_kernel_irq_routing_entry *irq_entry, struct kvm *kvm,
1707 int irq_source_id, int level, bool line_status);
1708 int kvm_arch_set_irq_inatomic(struct kvm_kernel_irq_routing_entry *e,
1709 struct kvm *kvm, int irq_source_id,
1710 int level, bool line_status);
1711 bool kvm_irq_has_notifier(struct kvm *kvm, unsigned irqchip, unsigned pin);
1712 void kvm_notify_acked_gsi(struct kvm *kvm, int gsi);
1713 void kvm_notify_acked_irq(struct kvm *kvm, unsigned irqchip, unsigned pin);
1714 void kvm_register_irq_ack_notifier(struct kvm *kvm,
1715 struct kvm_irq_ack_notifier *kian);
1716 void kvm_unregister_irq_ack_notifier(struct kvm *kvm,
1717 struct kvm_irq_ack_notifier *kian);
1718 int kvm_request_irq_source_id(struct kvm *kvm);
1719 void kvm_free_irq_source_id(struct kvm *kvm, int irq_source_id);
1720 bool kvm_arch_irqfd_allowed(struct kvm *kvm, struct kvm_irqfd *args);
1721
1722 /*
1723 * Returns a pointer to the memslot if it contains gfn.
1724 * Otherwise returns NULL.
1725 */
1726 static inline struct kvm_memory_slot *
try_get_memslot(struct kvm_memory_slot * slot,gfn_t gfn)1727 try_get_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1728 {
1729 if (!slot)
1730 return NULL;
1731
1732 if (gfn >= slot->base_gfn && gfn < slot->base_gfn + slot->npages)
1733 return slot;
1734 else
1735 return NULL;
1736 }
1737
1738 /*
1739 * Returns a pointer to the memslot that contains gfn. Otherwise returns NULL.
1740 *
1741 * With "approx" set returns the memslot also when the address falls
1742 * in a hole. In that case one of the memslots bordering the hole is
1743 * returned.
1744 */
1745 static inline struct kvm_memory_slot *
search_memslots(struct kvm_memslots * slots,gfn_t gfn,bool approx)1746 search_memslots(struct kvm_memslots *slots, gfn_t gfn, bool approx)
1747 {
1748 struct kvm_memory_slot *slot;
1749 struct rb_node *node;
1750 int idx = slots->node_idx;
1751
1752 slot = NULL;
1753 for (node = slots->gfn_tree.rb_node; node; ) {
1754 slot = container_of(node, struct kvm_memory_slot, gfn_node[idx]);
1755 if (gfn >= slot->base_gfn) {
1756 if (gfn < slot->base_gfn + slot->npages)
1757 return slot;
1758 node = node->rb_right;
1759 } else
1760 node = node->rb_left;
1761 }
1762
1763 return approx ? slot : NULL;
1764 }
1765
1766 static inline struct kvm_memory_slot *
____gfn_to_memslot(struct kvm_memslots * slots,gfn_t gfn,bool approx)1767 ____gfn_to_memslot(struct kvm_memslots *slots, gfn_t gfn, bool approx)
1768 {
1769 struct kvm_memory_slot *slot;
1770
1771 slot = (struct kvm_memory_slot *)atomic_long_read(&slots->last_used_slot);
1772 slot = try_get_memslot(slot, gfn);
1773 if (slot)
1774 return slot;
1775
1776 slot = search_memslots(slots, gfn, approx);
1777 if (slot) {
1778 atomic_long_set(&slots->last_used_slot, (unsigned long)slot);
1779 return slot;
1780 }
1781
1782 return NULL;
1783 }
1784
1785 /*
1786 * __gfn_to_memslot() and its descendants are here to allow arch code to inline
1787 * the lookups in hot paths. gfn_to_memslot() itself isn't here as an inline
1788 * because that would bloat other code too much.
1789 */
1790 static inline struct kvm_memory_slot *
__gfn_to_memslot(struct kvm_memslots * slots,gfn_t gfn)1791 __gfn_to_memslot(struct kvm_memslots *slots, gfn_t gfn)
1792 {
1793 return ____gfn_to_memslot(slots, gfn, false);
1794 }
1795
1796 static inline unsigned long
__gfn_to_hva_memslot(const struct kvm_memory_slot * slot,gfn_t gfn)1797 __gfn_to_hva_memslot(const struct kvm_memory_slot *slot, gfn_t gfn)
1798 {
1799 /*
1800 * The index was checked originally in search_memslots. To avoid
1801 * that a malicious guest builds a Spectre gadget out of e.g. page
1802 * table walks, do not let the processor speculate loads outside
1803 * the guest's registered memslots.
1804 */
1805 unsigned long offset = gfn - slot->base_gfn;
1806 offset = array_index_nospec(offset, slot->npages);
1807 return slot->userspace_addr + offset * PAGE_SIZE;
1808 }
1809
memslot_id(struct kvm * kvm,gfn_t gfn)1810 static inline int memslot_id(struct kvm *kvm, gfn_t gfn)
1811 {
1812 return gfn_to_memslot(kvm, gfn)->id;
1813 }
1814
1815 static inline gfn_t
hva_to_gfn_memslot(unsigned long hva,struct kvm_memory_slot * slot)1816 hva_to_gfn_memslot(unsigned long hva, struct kvm_memory_slot *slot)
1817 {
1818 gfn_t gfn_offset = (hva - slot->userspace_addr) >> PAGE_SHIFT;
1819
1820 return slot->base_gfn + gfn_offset;
1821 }
1822
gfn_to_gpa(gfn_t gfn)1823 static inline gpa_t gfn_to_gpa(gfn_t gfn)
1824 {
1825 return (gpa_t)gfn << PAGE_SHIFT;
1826 }
1827
gpa_to_gfn(gpa_t gpa)1828 static inline gfn_t gpa_to_gfn(gpa_t gpa)
1829 {
1830 return (gfn_t)(gpa >> PAGE_SHIFT);
1831 }
1832
pfn_to_hpa(kvm_pfn_t pfn)1833 static inline hpa_t pfn_to_hpa(kvm_pfn_t pfn)
1834 {
1835 return (hpa_t)pfn << PAGE_SHIFT;
1836 }
1837
kvm_is_gpa_in_memslot(struct kvm * kvm,gpa_t gpa)1838 static inline bool kvm_is_gpa_in_memslot(struct kvm *kvm, gpa_t gpa)
1839 {
1840 unsigned long hva = gfn_to_hva(kvm, gpa_to_gfn(gpa));
1841
1842 return !kvm_is_error_hva(hva);
1843 }
1844
kvm_gpc_mark_dirty_in_slot(struct gfn_to_pfn_cache * gpc)1845 static inline void kvm_gpc_mark_dirty_in_slot(struct gfn_to_pfn_cache *gpc)
1846 {
1847 lockdep_assert_held(&gpc->lock);
1848
1849 if (!gpc->memslot)
1850 return;
1851
1852 mark_page_dirty_in_slot(gpc->kvm, gpc->memslot, gpa_to_gfn(gpc->gpa));
1853 }
1854
1855 enum kvm_stat_kind {
1856 KVM_STAT_VM,
1857 KVM_STAT_VCPU,
1858 };
1859
1860 struct kvm_stat_data {
1861 struct kvm *kvm;
1862 const struct _kvm_stats_desc *desc;
1863 enum kvm_stat_kind kind;
1864 };
1865
1866 struct _kvm_stats_desc {
1867 struct kvm_stats_desc desc;
1868 char name[KVM_STATS_NAME_SIZE];
1869 };
1870
1871 #define STATS_DESC_COMMON(type, unit, base, exp, sz, bsz) \
1872 .flags = type | unit | base | \
1873 BUILD_BUG_ON_ZERO(type & ~KVM_STATS_TYPE_MASK) | \
1874 BUILD_BUG_ON_ZERO(unit & ~KVM_STATS_UNIT_MASK) | \
1875 BUILD_BUG_ON_ZERO(base & ~KVM_STATS_BASE_MASK), \
1876 .exponent = exp, \
1877 .size = sz, \
1878 .bucket_size = bsz
1879
1880 #define VM_GENERIC_STATS_DESC(stat, type, unit, base, exp, sz, bsz) \
1881 { \
1882 { \
1883 STATS_DESC_COMMON(type, unit, base, exp, sz, bsz), \
1884 .offset = offsetof(struct kvm_vm_stat, generic.stat) \
1885 }, \
1886 .name = #stat, \
1887 }
1888 #define VCPU_GENERIC_STATS_DESC(stat, type, unit, base, exp, sz, bsz) \
1889 { \
1890 { \
1891 STATS_DESC_COMMON(type, unit, base, exp, sz, bsz), \
1892 .offset = offsetof(struct kvm_vcpu_stat, generic.stat) \
1893 }, \
1894 .name = #stat, \
1895 }
1896 #define VM_STATS_DESC(stat, type, unit, base, exp, sz, bsz) \
1897 { \
1898 { \
1899 STATS_DESC_COMMON(type, unit, base, exp, sz, bsz), \
1900 .offset = offsetof(struct kvm_vm_stat, stat) \
1901 }, \
1902 .name = #stat, \
1903 }
1904 #define VCPU_STATS_DESC(stat, type, unit, base, exp, sz, bsz) \
1905 { \
1906 { \
1907 STATS_DESC_COMMON(type, unit, base, exp, sz, bsz), \
1908 .offset = offsetof(struct kvm_vcpu_stat, stat) \
1909 }, \
1910 .name = #stat, \
1911 }
1912 /* SCOPE: VM, VM_GENERIC, VCPU, VCPU_GENERIC */
1913 #define STATS_DESC(SCOPE, stat, type, unit, base, exp, sz, bsz) \
1914 SCOPE##_STATS_DESC(stat, type, unit, base, exp, sz, bsz)
1915
1916 #define STATS_DESC_CUMULATIVE(SCOPE, name, unit, base, exponent) \
1917 STATS_DESC(SCOPE, name, KVM_STATS_TYPE_CUMULATIVE, \
1918 unit, base, exponent, 1, 0)
1919 #define STATS_DESC_INSTANT(SCOPE, name, unit, base, exponent) \
1920 STATS_DESC(SCOPE, name, KVM_STATS_TYPE_INSTANT, \
1921 unit, base, exponent, 1, 0)
1922 #define STATS_DESC_PEAK(SCOPE, name, unit, base, exponent) \
1923 STATS_DESC(SCOPE, name, KVM_STATS_TYPE_PEAK, \
1924 unit, base, exponent, 1, 0)
1925 #define STATS_DESC_LINEAR_HIST(SCOPE, name, unit, base, exponent, sz, bsz) \
1926 STATS_DESC(SCOPE, name, KVM_STATS_TYPE_LINEAR_HIST, \
1927 unit, base, exponent, sz, bsz)
1928 #define STATS_DESC_LOG_HIST(SCOPE, name, unit, base, exponent, sz) \
1929 STATS_DESC(SCOPE, name, KVM_STATS_TYPE_LOG_HIST, \
1930 unit, base, exponent, sz, 0)
1931
1932 /* Cumulative counter, read/write */
1933 #define STATS_DESC_COUNTER(SCOPE, name) \
1934 STATS_DESC_CUMULATIVE(SCOPE, name, KVM_STATS_UNIT_NONE, \
1935 KVM_STATS_BASE_POW10, 0)
1936 /* Instantaneous counter, read only */
1937 #define STATS_DESC_ICOUNTER(SCOPE, name) \
1938 STATS_DESC_INSTANT(SCOPE, name, KVM_STATS_UNIT_NONE, \
1939 KVM_STATS_BASE_POW10, 0)
1940 /* Peak counter, read/write */
1941 #define STATS_DESC_PCOUNTER(SCOPE, name) \
1942 STATS_DESC_PEAK(SCOPE, name, KVM_STATS_UNIT_NONE, \
1943 KVM_STATS_BASE_POW10, 0)
1944
1945 /* Instantaneous boolean value, read only */
1946 #define STATS_DESC_IBOOLEAN(SCOPE, name) \
1947 STATS_DESC_INSTANT(SCOPE, name, KVM_STATS_UNIT_BOOLEAN, \
1948 KVM_STATS_BASE_POW10, 0)
1949 /* Peak (sticky) boolean value, read/write */
1950 #define STATS_DESC_PBOOLEAN(SCOPE, name) \
1951 STATS_DESC_PEAK(SCOPE, name, KVM_STATS_UNIT_BOOLEAN, \
1952 KVM_STATS_BASE_POW10, 0)
1953
1954 /* Cumulative time in nanosecond */
1955 #define STATS_DESC_TIME_NSEC(SCOPE, name) \
1956 STATS_DESC_CUMULATIVE(SCOPE, name, KVM_STATS_UNIT_SECONDS, \
1957 KVM_STATS_BASE_POW10, -9)
1958 /* Linear histogram for time in nanosecond */
1959 #define STATS_DESC_LINHIST_TIME_NSEC(SCOPE, name, sz, bsz) \
1960 STATS_DESC_LINEAR_HIST(SCOPE, name, KVM_STATS_UNIT_SECONDS, \
1961 KVM_STATS_BASE_POW10, -9, sz, bsz)
1962 /* Logarithmic histogram for time in nanosecond */
1963 #define STATS_DESC_LOGHIST_TIME_NSEC(SCOPE, name, sz) \
1964 STATS_DESC_LOG_HIST(SCOPE, name, KVM_STATS_UNIT_SECONDS, \
1965 KVM_STATS_BASE_POW10, -9, sz)
1966
1967 #define KVM_GENERIC_VM_STATS() \
1968 STATS_DESC_COUNTER(VM_GENERIC, remote_tlb_flush), \
1969 STATS_DESC_COUNTER(VM_GENERIC, remote_tlb_flush_requests)
1970
1971 #define KVM_GENERIC_VCPU_STATS() \
1972 STATS_DESC_COUNTER(VCPU_GENERIC, halt_successful_poll), \
1973 STATS_DESC_COUNTER(VCPU_GENERIC, halt_attempted_poll), \
1974 STATS_DESC_COUNTER(VCPU_GENERIC, halt_poll_invalid), \
1975 STATS_DESC_COUNTER(VCPU_GENERIC, halt_wakeup), \
1976 STATS_DESC_TIME_NSEC(VCPU_GENERIC, halt_poll_success_ns), \
1977 STATS_DESC_TIME_NSEC(VCPU_GENERIC, halt_poll_fail_ns), \
1978 STATS_DESC_TIME_NSEC(VCPU_GENERIC, halt_wait_ns), \
1979 STATS_DESC_LOGHIST_TIME_NSEC(VCPU_GENERIC, halt_poll_success_hist, \
1980 HALT_POLL_HIST_COUNT), \
1981 STATS_DESC_LOGHIST_TIME_NSEC(VCPU_GENERIC, halt_poll_fail_hist, \
1982 HALT_POLL_HIST_COUNT), \
1983 STATS_DESC_LOGHIST_TIME_NSEC(VCPU_GENERIC, halt_wait_hist, \
1984 HALT_POLL_HIST_COUNT), \
1985 STATS_DESC_IBOOLEAN(VCPU_GENERIC, blocking)
1986
1987 ssize_t kvm_stats_read(char *id, const struct kvm_stats_header *header,
1988 const struct _kvm_stats_desc *desc,
1989 void *stats, size_t size_stats,
1990 char __user *user_buffer, size_t size, loff_t *offset);
1991
1992 /**
1993 * kvm_stats_linear_hist_update() - Update bucket value for linear histogram
1994 * statistics data.
1995 *
1996 * @data: start address of the stats data
1997 * @size: the number of bucket of the stats data
1998 * @value: the new value used to update the linear histogram's bucket
1999 * @bucket_size: the size (width) of a bucket
2000 */
kvm_stats_linear_hist_update(u64 * data,size_t size,u64 value,size_t bucket_size)2001 static inline void kvm_stats_linear_hist_update(u64 *data, size_t size,
2002 u64 value, size_t bucket_size)
2003 {
2004 size_t index = div64_u64(value, bucket_size);
2005
2006 index = min(index, size - 1);
2007 ++data[index];
2008 }
2009
2010 /**
2011 * kvm_stats_log_hist_update() - Update bucket value for logarithmic histogram
2012 * statistics data.
2013 *
2014 * @data: start address of the stats data
2015 * @size: the number of bucket of the stats data
2016 * @value: the new value used to update the logarithmic histogram's bucket
2017 */
kvm_stats_log_hist_update(u64 * data,size_t size,u64 value)2018 static inline void kvm_stats_log_hist_update(u64 *data, size_t size, u64 value)
2019 {
2020 size_t index = fls64(value);
2021
2022 index = min(index, size - 1);
2023 ++data[index];
2024 }
2025
2026 #define KVM_STATS_LINEAR_HIST_UPDATE(array, value, bsize) \
2027 kvm_stats_linear_hist_update(array, ARRAY_SIZE(array), value, bsize)
2028 #define KVM_STATS_LOG_HIST_UPDATE(array, value) \
2029 kvm_stats_log_hist_update(array, ARRAY_SIZE(array), value)
2030
2031
2032 extern const struct kvm_stats_header kvm_vm_stats_header;
2033 extern const struct _kvm_stats_desc kvm_vm_stats_desc[];
2034 extern const struct kvm_stats_header kvm_vcpu_stats_header;
2035 extern const struct _kvm_stats_desc kvm_vcpu_stats_desc[];
2036
2037 #ifdef CONFIG_KVM_GENERIC_MMU_NOTIFIER
mmu_invalidate_retry(struct kvm * kvm,unsigned long mmu_seq)2038 static inline int mmu_invalidate_retry(struct kvm *kvm, unsigned long mmu_seq)
2039 {
2040 if (unlikely(kvm->mmu_invalidate_in_progress))
2041 return 1;
2042 /*
2043 * Ensure the read of mmu_invalidate_in_progress happens before
2044 * the read of mmu_invalidate_seq. This interacts with the
2045 * smp_wmb() in mmu_notifier_invalidate_range_end to make sure
2046 * that the caller either sees the old (non-zero) value of
2047 * mmu_invalidate_in_progress or the new (incremented) value of
2048 * mmu_invalidate_seq.
2049 *
2050 * PowerPC Book3s HV KVM calls this under a per-page lock rather
2051 * than under kvm->mmu_lock, for scalability, so can't rely on
2052 * kvm->mmu_lock to keep things ordered.
2053 */
2054 smp_rmb();
2055 if (kvm->mmu_invalidate_seq != mmu_seq)
2056 return 1;
2057 return 0;
2058 }
2059
mmu_invalidate_retry_gfn(struct kvm * kvm,unsigned long mmu_seq,gfn_t gfn)2060 static inline int mmu_invalidate_retry_gfn(struct kvm *kvm,
2061 unsigned long mmu_seq,
2062 gfn_t gfn)
2063 {
2064 lockdep_assert_held(&kvm->mmu_lock);
2065 /*
2066 * If mmu_invalidate_in_progress is non-zero, then the range maintained
2067 * by kvm_mmu_notifier_invalidate_range_start contains all addresses
2068 * that might be being invalidated. Note that it may include some false
2069 * positives, due to shortcuts when handing concurrent invalidations.
2070 */
2071 if (unlikely(kvm->mmu_invalidate_in_progress)) {
2072 /*
2073 * Dropping mmu_lock after bumping mmu_invalidate_in_progress
2074 * but before updating the range is a KVM bug.
2075 */
2076 if (WARN_ON_ONCE(kvm->mmu_invalidate_range_start == INVALID_GPA ||
2077 kvm->mmu_invalidate_range_end == INVALID_GPA))
2078 return 1;
2079
2080 if (gfn >= kvm->mmu_invalidate_range_start &&
2081 gfn < kvm->mmu_invalidate_range_end)
2082 return 1;
2083 }
2084
2085 if (kvm->mmu_invalidate_seq != mmu_seq)
2086 return 1;
2087 return 0;
2088 }
2089
2090 /*
2091 * This lockless version of the range-based retry check *must* be paired with a
2092 * call to the locked version after acquiring mmu_lock, i.e. this is safe to
2093 * use only as a pre-check to avoid contending mmu_lock. This version *will*
2094 * get false negatives and false positives.
2095 */
mmu_invalidate_retry_gfn_unsafe(struct kvm * kvm,unsigned long mmu_seq,gfn_t gfn)2096 static inline bool mmu_invalidate_retry_gfn_unsafe(struct kvm *kvm,
2097 unsigned long mmu_seq,
2098 gfn_t gfn)
2099 {
2100 /*
2101 * Use READ_ONCE() to ensure the in-progress flag and sequence counter
2102 * are always read from memory, e.g. so that checking for retry in a
2103 * loop won't result in an infinite retry loop. Don't force loads for
2104 * start+end, as the key to avoiding infinite retry loops is observing
2105 * the 1=>0 transition of in-progress, i.e. getting false negatives
2106 * due to stale start+end values is acceptable.
2107 */
2108 if (unlikely(READ_ONCE(kvm->mmu_invalidate_in_progress)) &&
2109 gfn >= kvm->mmu_invalidate_range_start &&
2110 gfn < kvm->mmu_invalidate_range_end)
2111 return true;
2112
2113 return READ_ONCE(kvm->mmu_invalidate_seq) != mmu_seq;
2114 }
2115 #endif
2116
2117 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2118
2119 #define KVM_MAX_IRQ_ROUTES 4096 /* might need extension/rework in the future */
2120
2121 bool kvm_arch_can_set_irq_routing(struct kvm *kvm);
2122 int kvm_set_irq_routing(struct kvm *kvm,
2123 const struct kvm_irq_routing_entry *entries,
2124 unsigned nr,
2125 unsigned flags);
2126 int kvm_init_irq_routing(struct kvm *kvm);
2127 int kvm_set_routing_entry(struct kvm *kvm,
2128 struct kvm_kernel_irq_routing_entry *e,
2129 const struct kvm_irq_routing_entry *ue);
2130 void kvm_free_irq_routing(struct kvm *kvm);
2131
2132 #else
2133
kvm_free_irq_routing(struct kvm * kvm)2134 static inline void kvm_free_irq_routing(struct kvm *kvm) {}
2135
kvm_init_irq_routing(struct kvm * kvm)2136 static inline int kvm_init_irq_routing(struct kvm *kvm)
2137 {
2138 return 0;
2139 }
2140
2141 #endif
2142
2143 int kvm_send_userspace_msi(struct kvm *kvm, struct kvm_msi *msi);
2144
2145 void kvm_eventfd_init(struct kvm *kvm);
2146 int kvm_ioeventfd(struct kvm *kvm, struct kvm_ioeventfd *args);
2147
2148 #ifdef CONFIG_HAVE_KVM_IRQCHIP
2149 int kvm_irqfd(struct kvm *kvm, struct kvm_irqfd *args);
2150 void kvm_irqfd_release(struct kvm *kvm);
2151 bool kvm_notify_irqfd_resampler(struct kvm *kvm,
2152 unsigned int irqchip,
2153 unsigned int pin);
2154 void kvm_irq_routing_update(struct kvm *);
2155 #else
kvm_irqfd(struct kvm * kvm,struct kvm_irqfd * args)2156 static inline int kvm_irqfd(struct kvm *kvm, struct kvm_irqfd *args)
2157 {
2158 return -EINVAL;
2159 }
2160
kvm_irqfd_release(struct kvm * kvm)2161 static inline void kvm_irqfd_release(struct kvm *kvm) {}
2162
kvm_notify_irqfd_resampler(struct kvm * kvm,unsigned int irqchip,unsigned int pin)2163 static inline bool kvm_notify_irqfd_resampler(struct kvm *kvm,
2164 unsigned int irqchip,
2165 unsigned int pin)
2166 {
2167 return false;
2168 }
2169 #endif /* CONFIG_HAVE_KVM_IRQCHIP */
2170
2171 void kvm_arch_irq_routing_update(struct kvm *kvm);
2172
__kvm_make_request(int req,struct kvm_vcpu * vcpu)2173 static inline void __kvm_make_request(int req, struct kvm_vcpu *vcpu)
2174 {
2175 /*
2176 * Ensure the rest of the request is published to kvm_check_request's
2177 * caller. Paired with the smp_mb__after_atomic in kvm_check_request.
2178 */
2179 smp_wmb();
2180 set_bit(req & KVM_REQUEST_MASK, (void *)&vcpu->requests);
2181 }
2182
kvm_make_request(int req,struct kvm_vcpu * vcpu)2183 static __always_inline void kvm_make_request(int req, struct kvm_vcpu *vcpu)
2184 {
2185 /*
2186 * Request that don't require vCPU action should never be logged in
2187 * vcpu->requests. The vCPU won't clear the request, so it will stay
2188 * logged indefinitely and prevent the vCPU from entering the guest.
2189 */
2190 BUILD_BUG_ON(!__builtin_constant_p(req) ||
2191 (req & KVM_REQUEST_NO_ACTION));
2192
2193 __kvm_make_request(req, vcpu);
2194 }
2195
kvm_request_pending(struct kvm_vcpu * vcpu)2196 static inline bool kvm_request_pending(struct kvm_vcpu *vcpu)
2197 {
2198 return READ_ONCE(vcpu->requests);
2199 }
2200
kvm_test_request(int req,struct kvm_vcpu * vcpu)2201 static inline bool kvm_test_request(int req, struct kvm_vcpu *vcpu)
2202 {
2203 return test_bit(req & KVM_REQUEST_MASK, (void *)&vcpu->requests);
2204 }
2205
kvm_clear_request(int req,struct kvm_vcpu * vcpu)2206 static inline void kvm_clear_request(int req, struct kvm_vcpu *vcpu)
2207 {
2208 clear_bit(req & KVM_REQUEST_MASK, (void *)&vcpu->requests);
2209 }
2210
kvm_check_request(int req,struct kvm_vcpu * vcpu)2211 static inline bool kvm_check_request(int req, struct kvm_vcpu *vcpu)
2212 {
2213 if (kvm_test_request(req, vcpu)) {
2214 kvm_clear_request(req, vcpu);
2215
2216 /*
2217 * Ensure the rest of the request is visible to kvm_check_request's
2218 * caller. Paired with the smp_wmb in kvm_make_request.
2219 */
2220 smp_mb__after_atomic();
2221 return true;
2222 } else {
2223 return false;
2224 }
2225 }
2226
2227 #ifdef CONFIG_KVM_GENERIC_HARDWARE_ENABLING
2228 extern bool kvm_rebooting;
2229 #endif
2230
2231 extern unsigned int halt_poll_ns;
2232 extern unsigned int halt_poll_ns_grow;
2233 extern unsigned int halt_poll_ns_grow_start;
2234 extern unsigned int halt_poll_ns_shrink;
2235
2236 struct kvm_device {
2237 const struct kvm_device_ops *ops;
2238 struct kvm *kvm;
2239 void *private;
2240 struct list_head vm_node;
2241 };
2242
2243 /* create, destroy, and name are mandatory */
2244 struct kvm_device_ops {
2245 const char *name;
2246
2247 /*
2248 * create is called holding kvm->lock and any operations not suitable
2249 * to do while holding the lock should be deferred to init (see
2250 * below).
2251 */
2252 int (*create)(struct kvm_device *dev, u32 type);
2253
2254 /*
2255 * init is called after create if create is successful and is called
2256 * outside of holding kvm->lock.
2257 */
2258 void (*init)(struct kvm_device *dev);
2259
2260 /*
2261 * Destroy is responsible for freeing dev.
2262 *
2263 * Destroy may be called before or after destructors are called
2264 * on emulated I/O regions, depending on whether a reference is
2265 * held by a vcpu or other kvm component that gets destroyed
2266 * after the emulated I/O.
2267 */
2268 void (*destroy)(struct kvm_device *dev);
2269
2270 /*
2271 * Release is an alternative method to free the device. It is
2272 * called when the device file descriptor is closed. Once
2273 * release is called, the destroy method will not be called
2274 * anymore as the device is removed from the device list of
2275 * the VM. kvm->lock is held.
2276 */
2277 void (*release)(struct kvm_device *dev);
2278
2279 int (*set_attr)(struct kvm_device *dev, struct kvm_device_attr *attr);
2280 int (*get_attr)(struct kvm_device *dev, struct kvm_device_attr *attr);
2281 int (*has_attr)(struct kvm_device *dev, struct kvm_device_attr *attr);
2282 long (*ioctl)(struct kvm_device *dev, unsigned int ioctl,
2283 unsigned long arg);
2284 int (*mmap)(struct kvm_device *dev, struct vm_area_struct *vma);
2285 };
2286
2287 struct kvm_device *kvm_device_from_filp(struct file *filp);
2288 int kvm_register_device_ops(const struct kvm_device_ops *ops, u32 type);
2289 void kvm_unregister_device_ops(u32 type);
2290
2291 extern struct kvm_device_ops kvm_mpic_ops;
2292 extern struct kvm_device_ops kvm_arm_vgic_v2_ops;
2293 extern struct kvm_device_ops kvm_arm_vgic_v3_ops;
2294
2295 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2296
kvm_vcpu_set_in_spin_loop(struct kvm_vcpu * vcpu,bool val)2297 static inline void kvm_vcpu_set_in_spin_loop(struct kvm_vcpu *vcpu, bool val)
2298 {
2299 vcpu->spin_loop.in_spin_loop = val;
2300 }
kvm_vcpu_set_dy_eligible(struct kvm_vcpu * vcpu,bool val)2301 static inline void kvm_vcpu_set_dy_eligible(struct kvm_vcpu *vcpu, bool val)
2302 {
2303 vcpu->spin_loop.dy_eligible = val;
2304 }
2305
2306 #else /* !CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT */
2307
kvm_vcpu_set_in_spin_loop(struct kvm_vcpu * vcpu,bool val)2308 static inline void kvm_vcpu_set_in_spin_loop(struct kvm_vcpu *vcpu, bool val)
2309 {
2310 }
2311
kvm_vcpu_set_dy_eligible(struct kvm_vcpu * vcpu,bool val)2312 static inline void kvm_vcpu_set_dy_eligible(struct kvm_vcpu *vcpu, bool val)
2313 {
2314 }
2315 #endif /* CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT */
2316
kvm_is_visible_memslot(struct kvm_memory_slot * memslot)2317 static inline bool kvm_is_visible_memslot(struct kvm_memory_slot *memslot)
2318 {
2319 return (memslot && memslot->id < KVM_USER_MEM_SLOTS &&
2320 !(memslot->flags & KVM_MEMSLOT_INVALID));
2321 }
2322
2323 struct kvm_vcpu *kvm_get_running_vcpu(void);
2324 struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void);
2325
2326 #ifdef CONFIG_HAVE_KVM_IRQ_BYPASS
2327 bool kvm_arch_has_irq_bypass(void);
2328 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *,
2329 struct irq_bypass_producer *);
2330 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *,
2331 struct irq_bypass_producer *);
2332 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *);
2333 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *);
2334 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
2335 uint32_t guest_irq, bool set);
2336 bool kvm_arch_irqfd_route_changed(struct kvm_kernel_irq_routing_entry *,
2337 struct kvm_kernel_irq_routing_entry *);
2338 #endif /* CONFIG_HAVE_KVM_IRQ_BYPASS */
2339
2340 #ifdef CONFIG_HAVE_KVM_INVALID_WAKEUPS
2341 /* If we wakeup during the poll time, was it a sucessful poll? */
vcpu_valid_wakeup(struct kvm_vcpu * vcpu)2342 static inline bool vcpu_valid_wakeup(struct kvm_vcpu *vcpu)
2343 {
2344 return vcpu->valid_wakeup;
2345 }
2346
2347 #else
vcpu_valid_wakeup(struct kvm_vcpu * vcpu)2348 static inline bool vcpu_valid_wakeup(struct kvm_vcpu *vcpu)
2349 {
2350 return true;
2351 }
2352 #endif /* CONFIG_HAVE_KVM_INVALID_WAKEUPS */
2353
2354 #ifdef CONFIG_HAVE_KVM_NO_POLL
2355 /* Callback that tells if we must not poll */
2356 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu);
2357 #else
kvm_arch_no_poll(struct kvm_vcpu * vcpu)2358 static inline bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
2359 {
2360 return false;
2361 }
2362 #endif /* CONFIG_HAVE_KVM_NO_POLL */
2363
2364 #ifdef CONFIG_HAVE_KVM_VCPU_ASYNC_IOCTL
2365 long kvm_arch_vcpu_async_ioctl(struct file *filp,
2366 unsigned int ioctl, unsigned long arg);
2367 #else
kvm_arch_vcpu_async_ioctl(struct file * filp,unsigned int ioctl,unsigned long arg)2368 static inline long kvm_arch_vcpu_async_ioctl(struct file *filp,
2369 unsigned int ioctl,
2370 unsigned long arg)
2371 {
2372 return -ENOIOCTLCMD;
2373 }
2374 #endif /* CONFIG_HAVE_KVM_VCPU_ASYNC_IOCTL */
2375
2376 void kvm_arch_guest_memory_reclaimed(struct kvm *kvm);
2377
2378 #ifdef CONFIG_HAVE_KVM_VCPU_RUN_PID_CHANGE
2379 int kvm_arch_vcpu_run_pid_change(struct kvm_vcpu *vcpu);
2380 #else
kvm_arch_vcpu_run_pid_change(struct kvm_vcpu * vcpu)2381 static inline int kvm_arch_vcpu_run_pid_change(struct kvm_vcpu *vcpu)
2382 {
2383 return 0;
2384 }
2385 #endif /* CONFIG_HAVE_KVM_VCPU_RUN_PID_CHANGE */
2386
2387 typedef int (*kvm_vm_thread_fn_t)(struct kvm *kvm, uintptr_t data);
2388
2389 int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn,
2390 uintptr_t data, const char *name,
2391 struct task_struct **thread_ptr);
2392
2393 #ifdef CONFIG_KVM_XFER_TO_GUEST_WORK
kvm_handle_signal_exit(struct kvm_vcpu * vcpu)2394 static inline void kvm_handle_signal_exit(struct kvm_vcpu *vcpu)
2395 {
2396 vcpu->run->exit_reason = KVM_EXIT_INTR;
2397 vcpu->stat.signal_exits++;
2398 }
2399 #endif /* CONFIG_KVM_XFER_TO_GUEST_WORK */
2400
2401 /*
2402 * If more than one page is being (un)accounted, @virt must be the address of
2403 * the first page of a block of pages what were allocated together (i.e
2404 * accounted together).
2405 *
2406 * kvm_account_pgtable_pages() is thread-safe because mod_lruvec_page_state()
2407 * is thread-safe.
2408 */
kvm_account_pgtable_pages(void * virt,int nr)2409 static inline void kvm_account_pgtable_pages(void *virt, int nr)
2410 {
2411 mod_lruvec_page_state(virt_to_page(virt), NR_SECONDARY_PAGETABLE, nr);
2412 }
2413
2414 /*
2415 * This defines how many reserved entries we want to keep before we
2416 * kick the vcpu to the userspace to avoid dirty ring full. This
2417 * value can be tuned to higher if e.g. PML is enabled on the host.
2418 */
2419 #define KVM_DIRTY_RING_RSVD_ENTRIES 64
2420
2421 /* Max number of entries allowed for each kvm dirty ring */
2422 #define KVM_DIRTY_RING_MAX_ENTRIES 65536
2423
kvm_prepare_memory_fault_exit(struct kvm_vcpu * vcpu,gpa_t gpa,gpa_t size,bool is_write,bool is_exec,bool is_private)2424 static inline void kvm_prepare_memory_fault_exit(struct kvm_vcpu *vcpu,
2425 gpa_t gpa, gpa_t size,
2426 bool is_write, bool is_exec,
2427 bool is_private)
2428 {
2429 vcpu->run->exit_reason = KVM_EXIT_MEMORY_FAULT;
2430 vcpu->run->memory_fault.gpa = gpa;
2431 vcpu->run->memory_fault.size = size;
2432
2433 /* RWX flags are not (yet) defined or communicated to userspace. */
2434 vcpu->run->memory_fault.flags = 0;
2435 if (is_private)
2436 vcpu->run->memory_fault.flags |= KVM_MEMORY_EXIT_FLAG_PRIVATE;
2437 }
2438
2439 #ifdef CONFIG_KVM_GENERIC_MEMORY_ATTRIBUTES
kvm_get_memory_attributes(struct kvm * kvm,gfn_t gfn)2440 static inline unsigned long kvm_get_memory_attributes(struct kvm *kvm, gfn_t gfn)
2441 {
2442 return xa_to_value(xa_load(&kvm->mem_attr_array, gfn));
2443 }
2444
2445 bool kvm_range_has_memory_attributes(struct kvm *kvm, gfn_t start, gfn_t end,
2446 unsigned long mask, unsigned long attrs);
2447 bool kvm_arch_pre_set_memory_attributes(struct kvm *kvm,
2448 struct kvm_gfn_range *range);
2449 bool kvm_arch_post_set_memory_attributes(struct kvm *kvm,
2450 struct kvm_gfn_range *range);
2451
kvm_mem_is_private(struct kvm * kvm,gfn_t gfn)2452 static inline bool kvm_mem_is_private(struct kvm *kvm, gfn_t gfn)
2453 {
2454 return IS_ENABLED(CONFIG_KVM_PRIVATE_MEM) &&
2455 kvm_get_memory_attributes(kvm, gfn) & KVM_MEMORY_ATTRIBUTE_PRIVATE;
2456 }
2457 #else
kvm_mem_is_private(struct kvm * kvm,gfn_t gfn)2458 static inline bool kvm_mem_is_private(struct kvm *kvm, gfn_t gfn)
2459 {
2460 return false;
2461 }
2462 #endif /* CONFIG_KVM_GENERIC_MEMORY_ATTRIBUTES */
2463
2464 #ifdef CONFIG_KVM_PRIVATE_MEM
2465 int kvm_gmem_get_pfn(struct kvm *kvm, struct kvm_memory_slot *slot,
2466 gfn_t gfn, kvm_pfn_t *pfn, int *max_order);
2467 #else
kvm_gmem_get_pfn(struct kvm * kvm,struct kvm_memory_slot * slot,gfn_t gfn,kvm_pfn_t * pfn,int * max_order)2468 static inline int kvm_gmem_get_pfn(struct kvm *kvm,
2469 struct kvm_memory_slot *slot, gfn_t gfn,
2470 kvm_pfn_t *pfn, int *max_order)
2471 {
2472 KVM_BUG_ON(1, kvm);
2473 return -EIO;
2474 }
2475 #endif /* CONFIG_KVM_PRIVATE_MEM */
2476
2477 #ifdef CONFIG_HAVE_KVM_ARCH_GMEM_PREPARE
2478 int kvm_arch_gmem_prepare(struct kvm *kvm, gfn_t gfn, kvm_pfn_t pfn, int max_order);
2479 #endif
2480
2481 #ifdef CONFIG_KVM_GENERIC_PRIVATE_MEM
2482 /**
2483 * kvm_gmem_populate() - Populate/prepare a GPA range with guest data
2484 *
2485 * @kvm: KVM instance
2486 * @gfn: starting GFN to be populated
2487 * @src: userspace-provided buffer containing data to copy into GFN range
2488 * (passed to @post_populate, and incremented on each iteration
2489 * if not NULL)
2490 * @npages: number of pages to copy from userspace-buffer
2491 * @post_populate: callback to issue for each gmem page that backs the GPA
2492 * range
2493 * @opaque: opaque data to pass to @post_populate callback
2494 *
2495 * This is primarily intended for cases where a gmem-backed GPA range needs
2496 * to be initialized with userspace-provided data prior to being mapped into
2497 * the guest as a private page. This should be called with the slots->lock
2498 * held so that caller-enforced invariants regarding the expected memory
2499 * attributes of the GPA range do not race with KVM_SET_MEMORY_ATTRIBUTES.
2500 *
2501 * Returns the number of pages that were populated.
2502 */
2503 typedef int (*kvm_gmem_populate_cb)(struct kvm *kvm, gfn_t gfn, kvm_pfn_t pfn,
2504 void __user *src, int order, void *opaque);
2505
2506 long kvm_gmem_populate(struct kvm *kvm, gfn_t gfn, void __user *src, long npages,
2507 kvm_gmem_populate_cb post_populate, void *opaque);
2508 #endif
2509
2510 #ifdef CONFIG_HAVE_KVM_ARCH_GMEM_INVALIDATE
2511 void kvm_arch_gmem_invalidate(kvm_pfn_t start, kvm_pfn_t end);
2512 #endif
2513
2514 #ifdef CONFIG_KVM_GENERIC_PRE_FAULT_MEMORY
2515 long kvm_arch_vcpu_pre_fault_memory(struct kvm_vcpu *vcpu,
2516 struct kvm_pre_fault_memory *range);
2517 #endif
2518
2519 #endif
2520