1 /* SPDX-License-Identifier: GPL-2.0-only */
2 #ifndef __KVM_HOST_H
3 #define __KVM_HOST_H
4
5 #include <linux/entry-virt.h>
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/uapi/linux/kvm.h.
56 */
57 #define KVM_MEMSLOT_INVALID (1UL << 16)
58 #define KVM_MEMSLOT_GMEM_ONLY (1UL << 17)
59
60 /*
61 * Bit 63 of the memslot generation number is an "update in-progress flag",
62 * e.g. is temporarily set for the duration of kvm_swap_active_memslots().
63 * This flag effectively creates a unique generation number that is used to
64 * mark cached memslot data, e.g. MMIO accesses, as potentially being stale,
65 * i.e. may (or may not) have come from the previous memslots generation.
66 *
67 * This is necessary because the actual memslots update is not atomic with
68 * respect to the generation number update. Updating the generation number
69 * first would allow a vCPU to cache a spte from the old memslots using the
70 * new generation number, and updating the generation number after switching
71 * to the new memslots would allow cache hits using the old generation number
72 * to reference the defunct memslots.
73 *
74 * This mechanism is used to prevent getting hits in KVM's caches while a
75 * memslot update is in-progress, and to prevent cache hits *after* updating
76 * the actual generation number against accesses that were inserted into the
77 * cache *before* the memslots were updated.
78 */
79 #define KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS BIT_ULL(63)
80
81 /* Two fragments for cross MMIO pages. */
82 #define KVM_MAX_MMIO_FRAGMENTS 2
83
84 #ifndef KVM_MAX_NR_ADDRESS_SPACES
85 #define KVM_MAX_NR_ADDRESS_SPACES 1
86 #endif
87
88 /*
89 * For the normal pfn, the highest 12 bits should be zero,
90 * so we can mask bit 62 ~ bit 52 to indicate the error pfn,
91 * mask bit 63 to indicate the noslot pfn.
92 */
93 #define KVM_PFN_ERR_MASK (0x7ffULL << 52)
94 #define KVM_PFN_ERR_NOSLOT_MASK (0xfffULL << 52)
95 #define KVM_PFN_NOSLOT (0x1ULL << 63)
96
97 #define KVM_PFN_ERR_FAULT (KVM_PFN_ERR_MASK)
98 #define KVM_PFN_ERR_HWPOISON (KVM_PFN_ERR_MASK + 1)
99 #define KVM_PFN_ERR_RO_FAULT (KVM_PFN_ERR_MASK + 2)
100 #define KVM_PFN_ERR_SIGPENDING (KVM_PFN_ERR_MASK + 3)
101 #define KVM_PFN_ERR_NEEDS_IO (KVM_PFN_ERR_MASK + 4)
102
103 /*
104 * error pfns indicate that the gfn is in slot but faild to
105 * translate it to pfn on host.
106 */
is_error_pfn(kvm_pfn_t pfn)107 static inline bool is_error_pfn(kvm_pfn_t pfn)
108 {
109 return !!(pfn & KVM_PFN_ERR_MASK);
110 }
111
112 /*
113 * KVM_PFN_ERR_SIGPENDING indicates that fetching the PFN was interrupted
114 * by a pending signal. Note, the signal may or may not be fatal.
115 */
is_sigpending_pfn(kvm_pfn_t pfn)116 static inline bool is_sigpending_pfn(kvm_pfn_t pfn)
117 {
118 return pfn == KVM_PFN_ERR_SIGPENDING;
119 }
120
121 /*
122 * error_noslot pfns indicate that the gfn can not be
123 * translated to pfn - it is not in slot or failed to
124 * translate it to pfn.
125 */
is_error_noslot_pfn(kvm_pfn_t pfn)126 static inline bool is_error_noslot_pfn(kvm_pfn_t pfn)
127 {
128 return !!(pfn & KVM_PFN_ERR_NOSLOT_MASK);
129 }
130
131 /* noslot pfn indicates that the gfn is not in slot. */
is_noslot_pfn(kvm_pfn_t pfn)132 static inline bool is_noslot_pfn(kvm_pfn_t pfn)
133 {
134 return pfn == KVM_PFN_NOSLOT;
135 }
136
137 /*
138 * architectures with KVM_HVA_ERR_BAD other than PAGE_OFFSET (e.g. s390)
139 * provide own defines and kvm_is_error_hva
140 */
141 #ifndef KVM_HVA_ERR_BAD
142
143 #define KVM_HVA_ERR_BAD (PAGE_OFFSET)
144 #define KVM_HVA_ERR_RO_BAD (PAGE_OFFSET + PAGE_SIZE)
145
kvm_is_error_hva(unsigned long addr)146 static inline bool kvm_is_error_hva(unsigned long addr)
147 {
148 return addr >= PAGE_OFFSET;
149 }
150
151 #endif
152
kvm_is_error_gpa(gpa_t gpa)153 static inline bool kvm_is_error_gpa(gpa_t gpa)
154 {
155 return gpa == INVALID_GPA;
156 }
157
158 #define KVM_REQUEST_MASK GENMASK(7,0)
159 #define KVM_REQUEST_NO_WAKEUP BIT(8)
160 #define KVM_REQUEST_WAIT BIT(9)
161 #define KVM_REQUEST_NO_ACTION BIT(10)
162 /*
163 * Architecture-independent vcpu->requests bit members
164 * Bits 3-7 are reserved for more arch-independent bits.
165 */
166 #define KVM_REQ_TLB_FLUSH (0 | KVM_REQUEST_WAIT | KVM_REQUEST_NO_WAKEUP)
167 #define KVM_REQ_VM_DEAD (1 | KVM_REQUEST_WAIT | KVM_REQUEST_NO_WAKEUP)
168 #define KVM_REQ_UNBLOCK 2
169 #define KVM_REQ_DIRTY_RING_SOFT_FULL 3
170 #define KVM_REQUEST_ARCH_BASE 8
171
172 /*
173 * KVM_REQ_OUTSIDE_GUEST_MODE exists is purely as way to force the vCPU to
174 * OUTSIDE_GUEST_MODE. KVM_REQ_OUTSIDE_GUEST_MODE differs from a vCPU "kick"
175 * in that it ensures the vCPU has reached OUTSIDE_GUEST_MODE before continuing
176 * on. A kick only guarantees that the vCPU is on its way out, e.g. a previous
177 * kick may have set vcpu->mode to EXITING_GUEST_MODE, and so there's no
178 * guarantee the vCPU received an IPI and has actually exited guest mode.
179 */
180 #define KVM_REQ_OUTSIDE_GUEST_MODE (KVM_REQUEST_NO_ACTION | KVM_REQUEST_WAIT | KVM_REQUEST_NO_WAKEUP)
181
182 #define KVM_ARCH_REQ_FLAGS(nr, flags) ({ \
183 BUILD_BUG_ON((unsigned)(nr) >= (sizeof_field(struct kvm_vcpu, requests) * 8) - KVM_REQUEST_ARCH_BASE); \
184 (unsigned)(((nr) + KVM_REQUEST_ARCH_BASE) | (flags)); \
185 })
186 #define KVM_ARCH_REQ(nr) KVM_ARCH_REQ_FLAGS(nr, 0)
187
188 bool kvm_make_vcpus_request_mask(struct kvm *kvm, unsigned int req,
189 unsigned long *vcpu_bitmap);
190 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req);
191
192 #define KVM_USERSPACE_IRQ_SOURCE_ID 0
193 #define KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID 1
194 #define KVM_PIT_IRQ_SOURCE_ID 2
195
196 extern struct mutex kvm_lock;
197 extern struct list_head vm_list;
198
199 struct kvm_io_range {
200 gpa_t addr;
201 int len;
202 struct kvm_io_device *dev;
203 };
204
205 #define NR_IOBUS_DEVS 1000
206
207 struct kvm_io_bus {
208 int dev_count;
209 int ioeventfd_count;
210 struct rcu_head rcu;
211 struct kvm_io_range range[];
212 };
213
214 enum kvm_bus {
215 KVM_MMIO_BUS,
216 KVM_PIO_BUS,
217 KVM_VIRTIO_CCW_NOTIFY_BUS,
218 KVM_FAST_MMIO_BUS,
219 KVM_IOCSR_BUS,
220 KVM_NR_BUSES
221 };
222
223 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
224 int len, const void *val);
225 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
226 gpa_t addr, int len, const void *val, long cookie);
227 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
228 int len, void *val);
229 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
230 int len, struct kvm_io_device *dev);
231 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
232 struct kvm_io_device *dev);
233 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
234 gpa_t addr);
235
236 #ifdef CONFIG_KVM_ASYNC_PF
237 struct kvm_async_pf {
238 struct work_struct work;
239 struct list_head link;
240 struct list_head queue;
241 struct kvm_vcpu *vcpu;
242 gpa_t cr2_or_gpa;
243 unsigned long addr;
244 struct kvm_arch_async_pf arch;
245 bool wakeup_all;
246 bool notpresent_injected;
247 };
248
249 void kvm_clear_async_pf_completion_queue(struct kvm_vcpu *vcpu);
250 void kvm_check_async_pf_completion(struct kvm_vcpu *vcpu);
251 bool kvm_setup_async_pf(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
252 unsigned long hva, struct kvm_arch_async_pf *arch);
253 int kvm_async_pf_wakeup_all(struct kvm_vcpu *vcpu);
254 #endif
255
256 #ifdef CONFIG_KVM_GENERIC_MMU_NOTIFIER
257 union kvm_mmu_notifier_arg {
258 unsigned long attributes;
259 };
260
261 enum kvm_gfn_range_filter {
262 KVM_FILTER_SHARED = BIT(0),
263 KVM_FILTER_PRIVATE = BIT(1),
264 };
265
266 struct kvm_gfn_range {
267 struct kvm_memory_slot *slot;
268 gfn_t start;
269 gfn_t end;
270 union kvm_mmu_notifier_arg arg;
271 enum kvm_gfn_range_filter attr_filter;
272 bool may_block;
273 bool lockless;
274 };
275 bool kvm_unmap_gfn_range(struct kvm *kvm, struct kvm_gfn_range *range);
276 bool kvm_age_gfn(struct kvm *kvm, struct kvm_gfn_range *range);
277 bool kvm_test_age_gfn(struct kvm *kvm, struct kvm_gfn_range *range);
278 #endif
279
280 enum {
281 OUTSIDE_GUEST_MODE,
282 IN_GUEST_MODE,
283 EXITING_GUEST_MODE,
284 READING_SHADOW_PAGE_TABLES,
285 };
286
287 struct kvm_host_map {
288 /*
289 * Only valid if the 'pfn' is managed by the host kernel (i.e. There is
290 * a 'struct page' for it. When using mem= kernel parameter some memory
291 * can be used as guest memory but they are not managed by host
292 * kernel).
293 */
294 struct page *pinned_page;
295 struct page *page;
296 void *hva;
297 kvm_pfn_t pfn;
298 kvm_pfn_t gfn;
299 bool writable;
300 };
301
302 /*
303 * Used to check if the mapping is valid or not. Never use 'kvm_host_map'
304 * directly to check for that.
305 */
kvm_vcpu_mapped(struct kvm_host_map * map)306 static inline bool kvm_vcpu_mapped(struct kvm_host_map *map)
307 {
308 return !!map->hva;
309 }
310
kvm_vcpu_can_poll(ktime_t cur,ktime_t stop)311 static inline bool kvm_vcpu_can_poll(ktime_t cur, ktime_t stop)
312 {
313 return single_task_running() && !need_resched() && ktime_before(cur, stop);
314 }
315
316 /*
317 * Sometimes a large or cross-page mmio needs to be broken up into separate
318 * exits for userspace servicing.
319 */
320 struct kvm_mmio_fragment {
321 gpa_t gpa;
322 void *data;
323 unsigned len;
324 };
325
326 struct kvm_vcpu {
327 struct kvm *kvm;
328 #ifdef CONFIG_PREEMPT_NOTIFIERS
329 struct preempt_notifier preempt_notifier;
330 #endif
331 int cpu;
332 int vcpu_id; /* id given by userspace at creation */
333 int vcpu_idx; /* index into kvm->vcpu_array */
334 int ____srcu_idx; /* Don't use this directly. You've been warned. */
335 #ifdef CONFIG_PROVE_RCU
336 int srcu_depth;
337 #endif
338 int mode;
339 u64 requests;
340 unsigned long guest_debug;
341
342 struct mutex mutex;
343 struct kvm_run *run;
344
345 #ifndef __KVM_HAVE_ARCH_WQP
346 struct rcuwait wait;
347 #endif
348 struct pid *pid;
349 rwlock_t pid_lock;
350 int sigset_active;
351 sigset_t sigset;
352 unsigned int halt_poll_ns;
353 bool valid_wakeup;
354
355 #ifdef CONFIG_HAS_IOMEM
356 int mmio_needed;
357 int mmio_read_completed;
358 int mmio_is_write;
359 int mmio_cur_fragment;
360 int mmio_nr_fragments;
361 struct kvm_mmio_fragment mmio_fragments[KVM_MAX_MMIO_FRAGMENTS];
362 #endif
363
364 #ifdef CONFIG_KVM_ASYNC_PF
365 struct {
366 u32 queued;
367 struct list_head queue;
368 struct list_head done;
369 spinlock_t lock;
370 } async_pf;
371 #endif
372
373 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
374 /*
375 * Cpu relax intercept or pause loop exit optimization
376 * in_spin_loop: set when a vcpu does a pause loop exit
377 * or cpu relax intercepted.
378 * dy_eligible: indicates whether vcpu is eligible for directed yield.
379 */
380 struct {
381 bool in_spin_loop;
382 bool dy_eligible;
383 } spin_loop;
384 #endif
385 bool wants_to_run;
386 bool preempted;
387 bool ready;
388 bool scheduled_out;
389 struct kvm_vcpu_arch arch;
390 struct kvm_vcpu_stat stat;
391 char stats_id[KVM_STATS_NAME_SIZE];
392 struct kvm_dirty_ring dirty_ring;
393
394 /*
395 * The most recently used memslot by this vCPU and the slots generation
396 * for which it is valid.
397 * No wraparound protection is needed since generations won't overflow in
398 * thousands of years, even assuming 1M memslot operations per second.
399 */
400 struct kvm_memory_slot *last_used_slot;
401 u64 last_used_slot_gen;
402 };
403
404 /*
405 * Start accounting time towards a guest.
406 * Must be called before entering guest context.
407 */
guest_timing_enter_irqoff(void)408 static __always_inline void guest_timing_enter_irqoff(void)
409 {
410 /*
411 * This is running in ioctl context so its safe to assume that it's the
412 * stime pending cputime to flush.
413 */
414 instrumentation_begin();
415 vtime_account_guest_enter();
416 instrumentation_end();
417 }
418
419 /*
420 * Enter guest context and enter an RCU extended quiescent state.
421 *
422 * Between guest_context_enter_irqoff() and guest_context_exit_irqoff() it is
423 * unsafe to use any code which may directly or indirectly use RCU, tracing
424 * (including IRQ flag tracing), or lockdep. All code in this period must be
425 * non-instrumentable.
426 */
guest_context_enter_irqoff(void)427 static __always_inline void guest_context_enter_irqoff(void)
428 {
429 /*
430 * KVM does not hold any references to rcu protected data when it
431 * switches CPU into a guest mode. In fact switching to a guest mode
432 * is very similar to exiting to userspace from rcu point of view. In
433 * addition CPU may stay in a guest mode for quite a long time (up to
434 * one time slice). Lets treat guest mode as quiescent state, just like
435 * we do with user-mode execution.
436 */
437 if (!context_tracking_guest_enter()) {
438 instrumentation_begin();
439 rcu_virt_note_context_switch();
440 instrumentation_end();
441 }
442 }
443
444 /*
445 * Deprecated. Architectures should move to guest_timing_enter_irqoff() and
446 * guest_state_enter_irqoff().
447 */
guest_enter_irqoff(void)448 static __always_inline void guest_enter_irqoff(void)
449 {
450 guest_timing_enter_irqoff();
451 guest_context_enter_irqoff();
452 }
453
454 /**
455 * guest_state_enter_irqoff - Fixup state when entering a guest
456 *
457 * Entry to a guest will enable interrupts, but the kernel state is interrupts
458 * disabled when this is invoked. Also tell RCU about it.
459 *
460 * 1) Trace interrupts on state
461 * 2) Invoke context tracking if enabled to adjust RCU state
462 * 3) Tell lockdep that interrupts are enabled
463 *
464 * Invoked from architecture specific code before entering a guest.
465 * Must be called with interrupts disabled and the caller must be
466 * non-instrumentable.
467 * The caller has to invoke guest_timing_enter_irqoff() before this.
468 *
469 * Note: this is analogous to exit_to_user_mode().
470 */
guest_state_enter_irqoff(void)471 static __always_inline void guest_state_enter_irqoff(void)
472 {
473 instrumentation_begin();
474 trace_hardirqs_on_prepare();
475 lockdep_hardirqs_on_prepare();
476 instrumentation_end();
477
478 guest_context_enter_irqoff();
479 lockdep_hardirqs_on(CALLER_ADDR0);
480 }
481
482 /*
483 * Exit guest context and exit an RCU extended quiescent state.
484 *
485 * Between guest_context_enter_irqoff() and guest_context_exit_irqoff() it is
486 * unsafe to use any code which may directly or indirectly use RCU, tracing
487 * (including IRQ flag tracing), or lockdep. All code in this period must be
488 * non-instrumentable.
489 */
guest_context_exit_irqoff(void)490 static __always_inline void guest_context_exit_irqoff(void)
491 {
492 /*
493 * Guest mode is treated as a quiescent state, see
494 * guest_context_enter_irqoff() for more details.
495 */
496 if (!context_tracking_guest_exit()) {
497 instrumentation_begin();
498 rcu_virt_note_context_switch();
499 instrumentation_end();
500 }
501 }
502
503 /*
504 * Stop accounting time towards a guest.
505 * Must be called after exiting guest context.
506 */
guest_timing_exit_irqoff(void)507 static __always_inline void guest_timing_exit_irqoff(void)
508 {
509 instrumentation_begin();
510 /* Flush the guest cputime we spent on the guest */
511 vtime_account_guest_exit();
512 instrumentation_end();
513 }
514
515 /*
516 * Deprecated. Architectures should move to guest_state_exit_irqoff() and
517 * guest_timing_exit_irqoff().
518 */
guest_exit_irqoff(void)519 static __always_inline void guest_exit_irqoff(void)
520 {
521 guest_context_exit_irqoff();
522 guest_timing_exit_irqoff();
523 }
524
guest_exit(void)525 static inline void guest_exit(void)
526 {
527 unsigned long flags;
528
529 local_irq_save(flags);
530 guest_exit_irqoff();
531 local_irq_restore(flags);
532 }
533
534 /**
535 * guest_state_exit_irqoff - Establish state when returning from guest mode
536 *
537 * Entry from a guest disables interrupts, but guest mode is traced as
538 * interrupts enabled. Also with NO_HZ_FULL RCU might be idle.
539 *
540 * 1) Tell lockdep that interrupts are disabled
541 * 2) Invoke context tracking if enabled to reactivate RCU
542 * 3) Trace interrupts off state
543 *
544 * Invoked from architecture specific code after exiting a guest.
545 * Must be invoked with interrupts disabled and the caller must be
546 * non-instrumentable.
547 * The caller has to invoke guest_timing_exit_irqoff() after this.
548 *
549 * Note: this is analogous to enter_from_user_mode().
550 */
guest_state_exit_irqoff(void)551 static __always_inline void guest_state_exit_irqoff(void)
552 {
553 lockdep_hardirqs_off(CALLER_ADDR0);
554 guest_context_exit_irqoff();
555
556 instrumentation_begin();
557 trace_hardirqs_off_finish();
558 instrumentation_end();
559 }
560
kvm_vcpu_exiting_guest_mode(struct kvm_vcpu * vcpu)561 static inline int kvm_vcpu_exiting_guest_mode(struct kvm_vcpu *vcpu)
562 {
563 /*
564 * The memory barrier ensures a previous write to vcpu->requests cannot
565 * be reordered with the read of vcpu->mode. It pairs with the general
566 * memory barrier following the write of vcpu->mode in VCPU RUN.
567 */
568 smp_mb__before_atomic();
569 return cmpxchg(&vcpu->mode, IN_GUEST_MODE, EXITING_GUEST_MODE);
570 }
571
572 /*
573 * Some of the bitops functions do not support too long bitmaps.
574 * This number must be determined not to exceed such limits.
575 */
576 #define KVM_MEM_MAX_NR_PAGES ((1UL << 31) - 1)
577
578 /*
579 * Since at idle each memslot belongs to two memslot sets it has to contain
580 * two embedded nodes for each data structure that it forms a part of.
581 *
582 * Two memslot sets (one active and one inactive) are necessary so the VM
583 * continues to run on one memslot set while the other is being modified.
584 *
585 * These two memslot sets normally point to the same set of memslots.
586 * They can, however, be desynchronized when performing a memslot management
587 * operation by replacing the memslot to be modified by its copy.
588 * After the operation is complete, both memslot sets once again point to
589 * the same, common set of memslot data.
590 *
591 * The memslots themselves are independent of each other so they can be
592 * individually added or deleted.
593 */
594 struct kvm_memory_slot {
595 struct hlist_node id_node[2];
596 struct interval_tree_node hva_node[2];
597 struct rb_node gfn_node[2];
598 gfn_t base_gfn;
599 unsigned long npages;
600 unsigned long *dirty_bitmap;
601 struct kvm_arch_memory_slot arch;
602 unsigned long userspace_addr;
603 u32 flags;
604 short id;
605 u16 as_id;
606
607 #ifdef CONFIG_KVM_GUEST_MEMFD
608 struct {
609 /*
610 * Writes protected by kvm->slots_lock. Acquiring a
611 * reference via kvm_gmem_get_file() is protected by
612 * either kvm->slots_lock or kvm->srcu.
613 */
614 struct file *file;
615 pgoff_t pgoff;
616 } gmem;
617 #endif
618 };
619
kvm_slot_has_gmem(const struct kvm_memory_slot * slot)620 static inline bool kvm_slot_has_gmem(const struct kvm_memory_slot *slot)
621 {
622 return slot && (slot->flags & KVM_MEM_GUEST_MEMFD);
623 }
624
kvm_slot_dirty_track_enabled(const struct kvm_memory_slot * slot)625 static inline bool kvm_slot_dirty_track_enabled(const struct kvm_memory_slot *slot)
626 {
627 return slot->flags & KVM_MEM_LOG_DIRTY_PAGES;
628 }
629
kvm_dirty_bitmap_bytes(struct kvm_memory_slot * memslot)630 static inline unsigned long kvm_dirty_bitmap_bytes(struct kvm_memory_slot *memslot)
631 {
632 return ALIGN(memslot->npages, BITS_PER_LONG) / 8;
633 }
634
kvm_second_dirty_bitmap(struct kvm_memory_slot * memslot)635 static inline unsigned long *kvm_second_dirty_bitmap(struct kvm_memory_slot *memslot)
636 {
637 unsigned long len = kvm_dirty_bitmap_bytes(memslot);
638
639 return memslot->dirty_bitmap + len / sizeof(*memslot->dirty_bitmap);
640 }
641
642 #ifndef KVM_DIRTY_LOG_MANUAL_CAPS
643 #define KVM_DIRTY_LOG_MANUAL_CAPS KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE
644 #endif
645
646 struct kvm_s390_adapter_int {
647 u64 ind_addr;
648 u64 summary_addr;
649 u64 ind_offset;
650 u32 summary_offset;
651 u32 adapter_id;
652 };
653
654 struct kvm_hv_sint {
655 u32 vcpu;
656 u32 sint;
657 };
658
659 struct kvm_xen_evtchn {
660 u32 port;
661 u32 vcpu_id;
662 int vcpu_idx;
663 u32 priority;
664 };
665
666 struct kvm_kernel_irq_routing_entry {
667 u32 gsi;
668 u32 type;
669 int (*set)(struct kvm_kernel_irq_routing_entry *e,
670 struct kvm *kvm, int irq_source_id, int level,
671 bool line_status);
672 union {
673 struct {
674 unsigned irqchip;
675 unsigned pin;
676 } irqchip;
677 struct {
678 u32 address_lo;
679 u32 address_hi;
680 u32 data;
681 u32 flags;
682 u32 devid;
683 } msi;
684 struct kvm_s390_adapter_int adapter;
685 struct kvm_hv_sint hv_sint;
686 struct kvm_xen_evtchn xen_evtchn;
687 };
688 struct hlist_node link;
689 };
690
691 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
692 struct kvm_irq_routing_table {
693 int chip[KVM_NR_IRQCHIPS][KVM_IRQCHIP_NUM_PINS];
694 u32 nr_rt_entries;
695 /*
696 * Array indexed by gsi. Each entry contains list of irq chips
697 * the gsi is connected to.
698 */
699 struct hlist_head map[] __counted_by(nr_rt_entries);
700 };
701 #endif
702
703 bool kvm_arch_irqchip_in_kernel(struct kvm *kvm);
704
705 #ifndef KVM_INTERNAL_MEM_SLOTS
706 #define KVM_INTERNAL_MEM_SLOTS 0
707 #endif
708
709 #define KVM_MEM_SLOTS_NUM SHRT_MAX
710 #define KVM_USER_MEM_SLOTS (KVM_MEM_SLOTS_NUM - KVM_INTERNAL_MEM_SLOTS)
711
712 #if KVM_MAX_NR_ADDRESS_SPACES == 1
kvm_arch_nr_memslot_as_ids(struct kvm * kvm)713 static inline int kvm_arch_nr_memslot_as_ids(struct kvm *kvm)
714 {
715 return KVM_MAX_NR_ADDRESS_SPACES;
716 }
717
kvm_arch_vcpu_memslots_id(struct kvm_vcpu * vcpu)718 static inline int kvm_arch_vcpu_memslots_id(struct kvm_vcpu *vcpu)
719 {
720 return 0;
721 }
722 #endif
723
724 #ifndef CONFIG_KVM_GENERIC_MEMORY_ATTRIBUTES
kvm_arch_has_private_mem(struct kvm * kvm)725 static inline bool kvm_arch_has_private_mem(struct kvm *kvm)
726 {
727 return false;
728 }
729 #endif
730
731 #ifdef CONFIG_KVM_GUEST_MEMFD
732 bool kvm_arch_supports_gmem_init_shared(struct kvm *kvm);
733
kvm_gmem_get_supported_flags(struct kvm * kvm)734 static inline u64 kvm_gmem_get_supported_flags(struct kvm *kvm)
735 {
736 u64 flags = GUEST_MEMFD_FLAG_MMAP;
737
738 if (!kvm || kvm_arch_supports_gmem_init_shared(kvm))
739 flags |= GUEST_MEMFD_FLAG_INIT_SHARED;
740
741 return flags;
742 }
743 #endif
744
745 #ifndef kvm_arch_has_readonly_mem
kvm_arch_has_readonly_mem(struct kvm * kvm)746 static inline bool kvm_arch_has_readonly_mem(struct kvm *kvm)
747 {
748 return IS_ENABLED(CONFIG_HAVE_KVM_READONLY_MEM);
749 }
750 #endif
751
752 struct kvm_memslots {
753 u64 generation;
754 atomic_long_t last_used_slot;
755 struct rb_root_cached hva_tree;
756 struct rb_root gfn_tree;
757 /*
758 * The mapping table from slot id to memslot.
759 *
760 * 7-bit bucket count matches the size of the old id to index array for
761 * 512 slots, while giving good performance with this slot count.
762 * Higher bucket counts bring only small performance improvements but
763 * always result in higher memory usage (even for lower memslot counts).
764 */
765 DECLARE_HASHTABLE(id_hash, 7);
766 int node_idx;
767 };
768
769 struct kvm {
770 #ifdef KVM_HAVE_MMU_RWLOCK
771 rwlock_t mmu_lock;
772 #else
773 spinlock_t mmu_lock;
774 #endif /* KVM_HAVE_MMU_RWLOCK */
775
776 struct mutex slots_lock;
777
778 /*
779 * Protects the arch-specific fields of struct kvm_memory_slots in
780 * use by the VM. To be used under the slots_lock (above) or in a
781 * kvm->srcu critical section where acquiring the slots_lock would
782 * lead to deadlock with the synchronize_srcu in
783 * kvm_swap_active_memslots().
784 */
785 struct mutex slots_arch_lock;
786 struct mm_struct *mm; /* userspace tied to this vm */
787 unsigned long nr_memslot_pages;
788 /* The two memslot sets - active and inactive (per address space) */
789 struct kvm_memslots __memslots[KVM_MAX_NR_ADDRESS_SPACES][2];
790 /* The current active memslot set for each address space */
791 struct kvm_memslots __rcu *memslots[KVM_MAX_NR_ADDRESS_SPACES];
792 struct xarray vcpu_array;
793 /*
794 * Protected by slots_lock, but can be read outside if an
795 * incorrect answer is acceptable.
796 */
797 atomic_t nr_memslots_dirty_logging;
798
799 /* Used to wait for completion of MMU notifiers. */
800 spinlock_t mn_invalidate_lock;
801 unsigned long mn_active_invalidate_count;
802 struct rcuwait mn_memslots_update_rcuwait;
803
804 /* For management / invalidation of gfn_to_pfn_caches */
805 spinlock_t gpc_lock;
806 struct list_head gpc_list;
807
808 /*
809 * created_vcpus is protected by kvm->lock, and is incremented
810 * at the beginning of KVM_CREATE_VCPU. online_vcpus is only
811 * incremented after storing the kvm_vcpu pointer in vcpus,
812 * and is accessed atomically.
813 */
814 atomic_t online_vcpus;
815 int max_vcpus;
816 int created_vcpus;
817 int last_boosted_vcpu;
818 struct list_head vm_list;
819 struct mutex lock;
820 struct kvm_io_bus __rcu *buses[KVM_NR_BUSES];
821 #ifdef CONFIG_HAVE_KVM_IRQCHIP
822 struct {
823 spinlock_t lock;
824 struct list_head items;
825 /* resampler_list update side is protected by resampler_lock. */
826 struct list_head resampler_list;
827 struct mutex resampler_lock;
828 } irqfds;
829 #endif
830 struct list_head ioeventfds;
831 struct kvm_vm_stat stat;
832 struct kvm_arch arch;
833 refcount_t users_count;
834 #ifdef CONFIG_KVM_MMIO
835 struct kvm_coalesced_mmio_ring *coalesced_mmio_ring;
836 spinlock_t ring_lock;
837 struct list_head coalesced_zones;
838 #endif
839
840 struct mutex irq_lock;
841 #ifdef CONFIG_HAVE_KVM_IRQCHIP
842 /*
843 * Update side is protected by irq_lock.
844 */
845 struct kvm_irq_routing_table __rcu *irq_routing;
846
847 struct hlist_head irq_ack_notifier_list;
848 #endif
849
850 #ifdef CONFIG_KVM_GENERIC_MMU_NOTIFIER
851 struct mmu_notifier mmu_notifier;
852 unsigned long mmu_invalidate_seq;
853 long mmu_invalidate_in_progress;
854 gfn_t mmu_invalidate_range_start;
855 gfn_t mmu_invalidate_range_end;
856 #endif
857 struct list_head devices;
858 u64 manual_dirty_log_protect;
859 struct dentry *debugfs_dentry;
860 struct kvm_stat_data **debugfs_stat_data;
861 struct srcu_struct srcu;
862 struct srcu_struct irq_srcu;
863 pid_t userspace_pid;
864 bool override_halt_poll_ns;
865 unsigned int max_halt_poll_ns;
866 u32 dirty_ring_size;
867 bool dirty_ring_with_bitmap;
868 bool vm_bugged;
869 bool vm_dead;
870
871 #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER
872 struct notifier_block pm_notifier;
873 #endif
874 #ifdef CONFIG_KVM_GENERIC_MEMORY_ATTRIBUTES
875 /* Protected by slots_lock (for writes) and RCU (for reads) */
876 struct xarray mem_attr_array;
877 #endif
878 char stats_id[KVM_STATS_NAME_SIZE];
879 };
880
881 #define kvm_err(fmt, ...) \
882 pr_err("kvm [%i]: " fmt, task_pid_nr(current), ## __VA_ARGS__)
883 #define kvm_info(fmt, ...) \
884 pr_info("kvm [%i]: " fmt, task_pid_nr(current), ## __VA_ARGS__)
885 #define kvm_debug(fmt, ...) \
886 pr_debug("kvm [%i]: " fmt, task_pid_nr(current), ## __VA_ARGS__)
887 #define kvm_debug_ratelimited(fmt, ...) \
888 pr_debug_ratelimited("kvm [%i]: " fmt, task_pid_nr(current), \
889 ## __VA_ARGS__)
890 #define kvm_pr_unimpl(fmt, ...) \
891 pr_err_ratelimited("kvm [%i]: " fmt, \
892 task_tgid_nr(current), ## __VA_ARGS__)
893
894 /* The guest did something we don't support. */
895 #define vcpu_unimpl(vcpu, fmt, ...) \
896 kvm_pr_unimpl("vcpu%i, guest rIP: 0x%lx " fmt, \
897 (vcpu)->vcpu_id, kvm_rip_read(vcpu), ## __VA_ARGS__)
898
899 #define vcpu_debug(vcpu, fmt, ...) \
900 kvm_debug("vcpu%i " fmt, (vcpu)->vcpu_id, ## __VA_ARGS__)
901 #define vcpu_debug_ratelimited(vcpu, fmt, ...) \
902 kvm_debug_ratelimited("vcpu%i " fmt, (vcpu)->vcpu_id, \
903 ## __VA_ARGS__)
904 #define vcpu_err(vcpu, fmt, ...) \
905 kvm_err("vcpu%i " fmt, (vcpu)->vcpu_id, ## __VA_ARGS__)
906
kvm_vm_dead(struct kvm * kvm)907 static inline void kvm_vm_dead(struct kvm *kvm)
908 {
909 kvm->vm_dead = true;
910 kvm_make_all_cpus_request(kvm, KVM_REQ_VM_DEAD);
911 }
912
kvm_vm_bugged(struct kvm * kvm)913 static inline void kvm_vm_bugged(struct kvm *kvm)
914 {
915 kvm->vm_bugged = true;
916 kvm_vm_dead(kvm);
917 }
918
919
920 #define KVM_BUG(cond, kvm, fmt...) \
921 ({ \
922 bool __ret = !!(cond); \
923 \
924 if (WARN_ONCE(__ret && !(kvm)->vm_bugged, fmt)) \
925 kvm_vm_bugged(kvm); \
926 unlikely(__ret); \
927 })
928
929 #define KVM_BUG_ON(cond, kvm) \
930 ({ \
931 bool __ret = !!(cond); \
932 \
933 if (WARN_ON_ONCE(__ret && !(kvm)->vm_bugged)) \
934 kvm_vm_bugged(kvm); \
935 unlikely(__ret); \
936 })
937
938 /*
939 * Note, "data corruption" refers to corruption of host kernel data structures,
940 * not guest data. Guest data corruption, suspected or confirmed, that is tied
941 * and contained to a single VM should *never* BUG() and potentially panic the
942 * host, i.e. use this variant of KVM_BUG() if and only if a KVM data structure
943 * is corrupted and that corruption can have a cascading effect to other parts
944 * of the hosts and/or to other VMs.
945 */
946 #define KVM_BUG_ON_DATA_CORRUPTION(cond, kvm) \
947 ({ \
948 bool __ret = !!(cond); \
949 \
950 if (IS_ENABLED(CONFIG_BUG_ON_DATA_CORRUPTION)) \
951 BUG_ON(__ret); \
952 else if (WARN_ON_ONCE(__ret && !(kvm)->vm_bugged)) \
953 kvm_vm_bugged(kvm); \
954 unlikely(__ret); \
955 })
956
kvm_vcpu_srcu_read_lock(struct kvm_vcpu * vcpu)957 static inline void kvm_vcpu_srcu_read_lock(struct kvm_vcpu *vcpu)
958 {
959 #ifdef CONFIG_PROVE_RCU
960 WARN_ONCE(vcpu->srcu_depth++,
961 "KVM: Illegal vCPU srcu_idx LOCK, depth=%d", vcpu->srcu_depth - 1);
962 #endif
963 vcpu->____srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
964 }
965
kvm_vcpu_srcu_read_unlock(struct kvm_vcpu * vcpu)966 static inline void kvm_vcpu_srcu_read_unlock(struct kvm_vcpu *vcpu)
967 {
968 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->____srcu_idx);
969
970 #ifdef CONFIG_PROVE_RCU
971 WARN_ONCE(--vcpu->srcu_depth,
972 "KVM: Illegal vCPU srcu_idx UNLOCK, depth=%d", vcpu->srcu_depth);
973 #endif
974 }
975
kvm_dirty_log_manual_protect_and_init_set(struct kvm * kvm)976 static inline bool kvm_dirty_log_manual_protect_and_init_set(struct kvm *kvm)
977 {
978 return !!(kvm->manual_dirty_log_protect & KVM_DIRTY_LOG_INITIALLY_SET);
979 }
980
981 /*
982 * Get a bus reference under the update-side lock. No long-term SRCU reader
983 * references are permitted, to avoid stale reads vs concurrent IO
984 * registrations.
985 */
kvm_get_bus(struct kvm * kvm,enum kvm_bus idx)986 static inline struct kvm_io_bus *kvm_get_bus(struct kvm *kvm, enum kvm_bus idx)
987 {
988 return rcu_dereference_protected(kvm->buses[idx],
989 lockdep_is_held(&kvm->slots_lock));
990 }
991
kvm_get_vcpu(struct kvm * kvm,int i)992 static inline struct kvm_vcpu *kvm_get_vcpu(struct kvm *kvm, int i)
993 {
994 int num_vcpus = atomic_read(&kvm->online_vcpus);
995
996 /*
997 * Explicitly verify the target vCPU is online, as the anti-speculation
998 * logic only limits the CPU's ability to speculate, e.g. given a "bad"
999 * index, clamping the index to 0 would return vCPU0, not NULL.
1000 */
1001 if (i >= num_vcpus)
1002 return NULL;
1003
1004 i = array_index_nospec(i, num_vcpus);
1005
1006 /* Pairs with smp_wmb() in kvm_vm_ioctl_create_vcpu. */
1007 smp_rmb();
1008 return xa_load(&kvm->vcpu_array, i);
1009 }
1010
1011 #define kvm_for_each_vcpu(idx, vcpup, kvm) \
1012 if (atomic_read(&kvm->online_vcpus)) \
1013 xa_for_each_range(&kvm->vcpu_array, idx, vcpup, 0, \
1014 (atomic_read(&kvm->online_vcpus) - 1))
1015
kvm_get_vcpu_by_id(struct kvm * kvm,int id)1016 static inline struct kvm_vcpu *kvm_get_vcpu_by_id(struct kvm *kvm, int id)
1017 {
1018 struct kvm_vcpu *vcpu = NULL;
1019 unsigned long i;
1020
1021 if (id < 0)
1022 return NULL;
1023 if (id < KVM_MAX_VCPUS)
1024 vcpu = kvm_get_vcpu(kvm, id);
1025 if (vcpu && vcpu->vcpu_id == id)
1026 return vcpu;
1027 kvm_for_each_vcpu(i, vcpu, kvm)
1028 if (vcpu->vcpu_id == id)
1029 return vcpu;
1030 return NULL;
1031 }
1032
1033 void kvm_destroy_vcpus(struct kvm *kvm);
1034
1035 int kvm_trylock_all_vcpus(struct kvm *kvm);
1036 int kvm_lock_all_vcpus(struct kvm *kvm);
1037 void kvm_unlock_all_vcpus(struct kvm *kvm);
1038
1039 void vcpu_load(struct kvm_vcpu *vcpu);
1040 void vcpu_put(struct kvm_vcpu *vcpu);
1041
1042 #ifdef CONFIG_KVM_IOAPIC
1043 void kvm_arch_post_irq_ack_notifier_list_update(struct kvm *kvm);
1044 #else
kvm_arch_post_irq_ack_notifier_list_update(struct kvm * kvm)1045 static inline void kvm_arch_post_irq_ack_notifier_list_update(struct kvm *kvm)
1046 {
1047 }
1048 #endif
1049
1050 #ifdef CONFIG_HAVE_KVM_IRQCHIP
1051 int kvm_irqfd_init(void);
1052 void kvm_irqfd_exit(void);
1053 #else
kvm_irqfd_init(void)1054 static inline int kvm_irqfd_init(void)
1055 {
1056 return 0;
1057 }
1058
kvm_irqfd_exit(void)1059 static inline void kvm_irqfd_exit(void)
1060 {
1061 }
1062 #endif
1063 int kvm_init(unsigned vcpu_size, unsigned vcpu_align, struct module *module);
1064 void kvm_exit(void);
1065
1066 void kvm_get_kvm(struct kvm *kvm);
1067 bool kvm_get_kvm_safe(struct kvm *kvm);
1068 void kvm_put_kvm(struct kvm *kvm);
1069 bool file_is_kvm(struct file *file);
1070 void kvm_put_kvm_no_destroy(struct kvm *kvm);
1071
__kvm_memslots(struct kvm * kvm,int as_id)1072 static inline struct kvm_memslots *__kvm_memslots(struct kvm *kvm, int as_id)
1073 {
1074 as_id = array_index_nospec(as_id, KVM_MAX_NR_ADDRESS_SPACES);
1075 return srcu_dereference_check(kvm->memslots[as_id], &kvm->srcu,
1076 lockdep_is_held(&kvm->slots_lock) ||
1077 !refcount_read(&kvm->users_count));
1078 }
1079
kvm_memslots(struct kvm * kvm)1080 static inline struct kvm_memslots *kvm_memslots(struct kvm *kvm)
1081 {
1082 return __kvm_memslots(kvm, 0);
1083 }
1084
kvm_vcpu_memslots(struct kvm_vcpu * vcpu)1085 static inline struct kvm_memslots *kvm_vcpu_memslots(struct kvm_vcpu *vcpu)
1086 {
1087 int as_id = kvm_arch_vcpu_memslots_id(vcpu);
1088
1089 return __kvm_memslots(vcpu->kvm, as_id);
1090 }
1091
kvm_memslots_empty(struct kvm_memslots * slots)1092 static inline bool kvm_memslots_empty(struct kvm_memslots *slots)
1093 {
1094 return RB_EMPTY_ROOT(&slots->gfn_tree);
1095 }
1096
1097 bool kvm_are_all_memslots_empty(struct kvm *kvm);
1098
1099 #define kvm_for_each_memslot(memslot, bkt, slots) \
1100 hash_for_each(slots->id_hash, bkt, memslot, id_node[slots->node_idx]) \
1101 if (WARN_ON_ONCE(!memslot->npages)) { \
1102 } else
1103
1104 static inline
id_to_memslot(struct kvm_memslots * slots,int id)1105 struct kvm_memory_slot *id_to_memslot(struct kvm_memslots *slots, int id)
1106 {
1107 struct kvm_memory_slot *slot;
1108 int idx = slots->node_idx;
1109
1110 hash_for_each_possible(slots->id_hash, slot, id_node[idx], id) {
1111 if (slot->id == id)
1112 return slot;
1113 }
1114
1115 return NULL;
1116 }
1117
1118 /* Iterator used for walking memslots that overlap a gfn range. */
1119 struct kvm_memslot_iter {
1120 struct kvm_memslots *slots;
1121 struct rb_node *node;
1122 struct kvm_memory_slot *slot;
1123 };
1124
kvm_memslot_iter_next(struct kvm_memslot_iter * iter)1125 static inline void kvm_memslot_iter_next(struct kvm_memslot_iter *iter)
1126 {
1127 iter->node = rb_next(iter->node);
1128 if (!iter->node)
1129 return;
1130
1131 iter->slot = container_of(iter->node, struct kvm_memory_slot, gfn_node[iter->slots->node_idx]);
1132 }
1133
kvm_memslot_iter_start(struct kvm_memslot_iter * iter,struct kvm_memslots * slots,gfn_t start)1134 static inline void kvm_memslot_iter_start(struct kvm_memslot_iter *iter,
1135 struct kvm_memslots *slots,
1136 gfn_t start)
1137 {
1138 int idx = slots->node_idx;
1139 struct rb_node *tmp;
1140 struct kvm_memory_slot *slot;
1141
1142 iter->slots = slots;
1143
1144 /*
1145 * Find the so called "upper bound" of a key - the first node that has
1146 * its key strictly greater than the searched one (the start gfn in our case).
1147 */
1148 iter->node = NULL;
1149 for (tmp = slots->gfn_tree.rb_node; tmp; ) {
1150 slot = container_of(tmp, struct kvm_memory_slot, gfn_node[idx]);
1151 if (start < slot->base_gfn) {
1152 iter->node = tmp;
1153 tmp = tmp->rb_left;
1154 } else {
1155 tmp = tmp->rb_right;
1156 }
1157 }
1158
1159 /*
1160 * Find the slot with the lowest gfn that can possibly intersect with
1161 * the range, so we'll ideally have slot start <= range start
1162 */
1163 if (iter->node) {
1164 /*
1165 * A NULL previous node means that the very first slot
1166 * already has a higher start gfn.
1167 * In this case slot start > range start.
1168 */
1169 tmp = rb_prev(iter->node);
1170 if (tmp)
1171 iter->node = tmp;
1172 } else {
1173 /* a NULL node below means no slots */
1174 iter->node = rb_last(&slots->gfn_tree);
1175 }
1176
1177 if (iter->node) {
1178 iter->slot = container_of(iter->node, struct kvm_memory_slot, gfn_node[idx]);
1179
1180 /*
1181 * It is possible in the slot start < range start case that the
1182 * found slot ends before or at range start (slot end <= range start)
1183 * and so it does not overlap the requested range.
1184 *
1185 * In such non-overlapping case the next slot (if it exists) will
1186 * already have slot start > range start, otherwise the logic above
1187 * would have found it instead of the current slot.
1188 */
1189 if (iter->slot->base_gfn + iter->slot->npages <= start)
1190 kvm_memslot_iter_next(iter);
1191 }
1192 }
1193
kvm_memslot_iter_is_valid(struct kvm_memslot_iter * iter,gfn_t end)1194 static inline bool kvm_memslot_iter_is_valid(struct kvm_memslot_iter *iter, gfn_t end)
1195 {
1196 if (!iter->node)
1197 return false;
1198
1199 /*
1200 * If this slot starts beyond or at the end of the range so does
1201 * every next one
1202 */
1203 return iter->slot->base_gfn < end;
1204 }
1205
1206 /* Iterate over each memslot at least partially intersecting [start, end) range */
1207 #define kvm_for_each_memslot_in_gfn_range(iter, slots, start, end) \
1208 for (kvm_memslot_iter_start(iter, slots, start); \
1209 kvm_memslot_iter_is_valid(iter, end); \
1210 kvm_memslot_iter_next(iter))
1211
1212 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn);
1213 struct kvm_memslots *kvm_vcpu_memslots(struct kvm_vcpu *vcpu);
1214 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn);
1215
1216 /*
1217 * KVM_SET_USER_MEMORY_REGION ioctl allows the following operations:
1218 * - create a new memory slot
1219 * - delete an existing memory slot
1220 * - modify an existing memory slot
1221 * -- move it in the guest physical memory space
1222 * -- just change its flags
1223 *
1224 * Since flags can be changed by some of these operations, the following
1225 * differentiation is the best we can do for kvm_set_memory_region():
1226 */
1227 enum kvm_mr_change {
1228 KVM_MR_CREATE,
1229 KVM_MR_DELETE,
1230 KVM_MR_MOVE,
1231 KVM_MR_FLAGS_ONLY,
1232 };
1233
1234 int kvm_set_internal_memslot(struct kvm *kvm,
1235 const struct kvm_userspace_memory_region2 *mem);
1236 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot);
1237 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen);
1238 int kvm_arch_prepare_memory_region(struct kvm *kvm,
1239 const struct kvm_memory_slot *old,
1240 struct kvm_memory_slot *new,
1241 enum kvm_mr_change change);
1242 void kvm_arch_commit_memory_region(struct kvm *kvm,
1243 struct kvm_memory_slot *old,
1244 const struct kvm_memory_slot *new,
1245 enum kvm_mr_change change);
1246 /* flush all memory translations */
1247 void kvm_arch_flush_shadow_all(struct kvm *kvm);
1248 /* flush memory translations pointing to 'slot' */
1249 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
1250 struct kvm_memory_slot *slot);
1251
1252 int kvm_prefetch_pages(struct kvm_memory_slot *slot, gfn_t gfn,
1253 struct page **pages, int nr_pages);
1254
1255 struct page *__gfn_to_page(struct kvm *kvm, gfn_t gfn, bool write);
gfn_to_page(struct kvm * kvm,gfn_t gfn)1256 static inline struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1257 {
1258 return __gfn_to_page(kvm, gfn, true);
1259 }
1260
1261 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn);
1262 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable);
1263 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot, gfn_t gfn);
1264 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot, gfn_t gfn,
1265 bool *writable);
1266
kvm_release_page_unused(struct page * page)1267 static inline void kvm_release_page_unused(struct page *page)
1268 {
1269 if (!page)
1270 return;
1271
1272 put_page(page);
1273 }
1274
1275 void kvm_release_page_clean(struct page *page);
1276 void kvm_release_page_dirty(struct page *page);
1277
kvm_release_faultin_page(struct kvm * kvm,struct page * page,bool unused,bool dirty)1278 static inline void kvm_release_faultin_page(struct kvm *kvm, struct page *page,
1279 bool unused, bool dirty)
1280 {
1281 lockdep_assert_once(lockdep_is_held(&kvm->mmu_lock) || unused);
1282
1283 if (!page)
1284 return;
1285
1286 /*
1287 * If the page that KVM got from the *primary MMU* is writable, and KVM
1288 * installed or reused a SPTE, mark the page/folio dirty. Note, this
1289 * may mark a folio dirty even if KVM created a read-only SPTE, e.g. if
1290 * the GFN is write-protected. Folios can't be safely marked dirty
1291 * outside of mmu_lock as doing so could race with writeback on the
1292 * folio. As a result, KVM can't mark folios dirty in the fast page
1293 * fault handler, and so KVM must (somewhat) speculatively mark the
1294 * folio dirty if KVM could locklessly make the SPTE writable.
1295 */
1296 if (unused)
1297 kvm_release_page_unused(page);
1298 else if (dirty)
1299 kvm_release_page_dirty(page);
1300 else
1301 kvm_release_page_clean(page);
1302 }
1303
1304 kvm_pfn_t __kvm_faultin_pfn(const struct kvm_memory_slot *slot, gfn_t gfn,
1305 unsigned int foll, bool *writable,
1306 struct page **refcounted_page);
1307
kvm_faultin_pfn(struct kvm_vcpu * vcpu,gfn_t gfn,bool write,bool * writable,struct page ** refcounted_page)1308 static inline kvm_pfn_t kvm_faultin_pfn(struct kvm_vcpu *vcpu, gfn_t gfn,
1309 bool write, bool *writable,
1310 struct page **refcounted_page)
1311 {
1312 return __kvm_faultin_pfn(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn,
1313 write ? FOLL_WRITE : 0, writable, refcounted_page);
1314 }
1315
1316 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1317 int len);
1318 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len);
1319 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1320 void *data, unsigned long len);
1321 int kvm_read_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1322 void *data, unsigned int offset,
1323 unsigned long len);
1324 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1325 int offset, int len);
1326 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1327 unsigned long len);
1328 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1329 void *data, unsigned long len);
1330 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1331 void *data, unsigned int offset,
1332 unsigned long len);
1333 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1334 gpa_t gpa, unsigned long len);
1335
1336 #define __kvm_get_guest(kvm, gfn, offset, v) \
1337 ({ \
1338 unsigned long __addr = gfn_to_hva(kvm, gfn); \
1339 typeof(v) __user *__uaddr = (typeof(__uaddr))(__addr + offset); \
1340 int __ret = -EFAULT; \
1341 \
1342 if (!kvm_is_error_hva(__addr)) \
1343 __ret = get_user(v, __uaddr); \
1344 __ret; \
1345 })
1346
1347 #define kvm_get_guest(kvm, gpa, v) \
1348 ({ \
1349 gpa_t __gpa = gpa; \
1350 struct kvm *__kvm = kvm; \
1351 \
1352 __kvm_get_guest(__kvm, __gpa >> PAGE_SHIFT, \
1353 offset_in_page(__gpa), v); \
1354 })
1355
1356 #define __kvm_put_guest(kvm, gfn, offset, v) \
1357 ({ \
1358 unsigned long __addr = gfn_to_hva(kvm, gfn); \
1359 typeof(v) __user *__uaddr = (typeof(__uaddr))(__addr + offset); \
1360 int __ret = -EFAULT; \
1361 \
1362 if (!kvm_is_error_hva(__addr)) \
1363 __ret = put_user(v, __uaddr); \
1364 if (!__ret) \
1365 mark_page_dirty(kvm, gfn); \
1366 __ret; \
1367 })
1368
1369 #define kvm_put_guest(kvm, gpa, v) \
1370 ({ \
1371 gpa_t __gpa = gpa; \
1372 struct kvm *__kvm = kvm; \
1373 \
1374 __kvm_put_guest(__kvm, __gpa >> PAGE_SHIFT, \
1375 offset_in_page(__gpa), v); \
1376 })
1377
1378 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len);
1379 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn);
1380 bool kvm_vcpu_is_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn);
1381 unsigned long kvm_host_page_size(struct kvm_vcpu *vcpu, gfn_t gfn);
1382 void mark_page_dirty_in_slot(struct kvm *kvm, const struct kvm_memory_slot *memslot, gfn_t gfn);
1383 void mark_page_dirty(struct kvm *kvm, gfn_t gfn);
1384
1385 int __kvm_vcpu_map(struct kvm_vcpu *vcpu, gpa_t gpa, struct kvm_host_map *map,
1386 bool writable);
1387 void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map);
1388
kvm_vcpu_map(struct kvm_vcpu * vcpu,gpa_t gpa,struct kvm_host_map * map)1389 static inline int kvm_vcpu_map(struct kvm_vcpu *vcpu, gpa_t gpa,
1390 struct kvm_host_map *map)
1391 {
1392 return __kvm_vcpu_map(vcpu, gpa, map, true);
1393 }
1394
kvm_vcpu_map_readonly(struct kvm_vcpu * vcpu,gpa_t gpa,struct kvm_host_map * map)1395 static inline int kvm_vcpu_map_readonly(struct kvm_vcpu *vcpu, gpa_t gpa,
1396 struct kvm_host_map *map)
1397 {
1398 return __kvm_vcpu_map(vcpu, gpa, map, false);
1399 }
1400
1401 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn);
1402 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable);
1403 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data, int offset,
1404 int len);
1405 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa, void *data,
1406 unsigned long len);
1407 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data,
1408 unsigned long len);
1409 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, const void *data,
1410 int offset, int len);
1411 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
1412 unsigned long len);
1413 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn);
1414
1415 /**
1416 * kvm_gpc_init - initialize gfn_to_pfn_cache.
1417 *
1418 * @gpc: struct gfn_to_pfn_cache object.
1419 * @kvm: pointer to kvm instance.
1420 *
1421 * This sets up a gfn_to_pfn_cache by initializing locks and assigning the
1422 * immutable attributes. Note, the cache must be zero-allocated (or zeroed by
1423 * the caller before init).
1424 */
1425 void kvm_gpc_init(struct gfn_to_pfn_cache *gpc, struct kvm *kvm);
1426
1427 /**
1428 * kvm_gpc_activate - prepare a cached kernel mapping and HPA for a given guest
1429 * physical address.
1430 *
1431 * @gpc: struct gfn_to_pfn_cache object.
1432 * @gpa: guest physical address to map.
1433 * @len: sanity check; the range being access must fit a single page.
1434 *
1435 * @return: 0 for success.
1436 * -EINVAL for a mapping which would cross a page boundary.
1437 * -EFAULT for an untranslatable guest physical address.
1438 *
1439 * This primes a gfn_to_pfn_cache and links it into the @gpc->kvm's list for
1440 * invalidations to be processed. Callers are required to use kvm_gpc_check()
1441 * to ensure that the cache is valid before accessing the target page.
1442 */
1443 int kvm_gpc_activate(struct gfn_to_pfn_cache *gpc, gpa_t gpa, unsigned long len);
1444
1445 /**
1446 * kvm_gpc_activate_hva - prepare a cached kernel mapping and HPA for a given HVA.
1447 *
1448 * @gpc: struct gfn_to_pfn_cache object.
1449 * @hva: userspace virtual address to map.
1450 * @len: sanity check; the range being access must fit a single page.
1451 *
1452 * @return: 0 for success.
1453 * -EINVAL for a mapping which would cross a page boundary.
1454 * -EFAULT for an untranslatable guest physical address.
1455 *
1456 * The semantics of this function are the same as those of kvm_gpc_activate(). It
1457 * merely bypasses a layer of address translation.
1458 */
1459 int kvm_gpc_activate_hva(struct gfn_to_pfn_cache *gpc, unsigned long hva, unsigned long len);
1460
1461 /**
1462 * kvm_gpc_check - check validity of a gfn_to_pfn_cache.
1463 *
1464 * @gpc: struct gfn_to_pfn_cache object.
1465 * @len: sanity check; the range being access must fit a single page.
1466 *
1467 * @return: %true if the cache is still valid and the address matches.
1468 * %false if the cache is not valid.
1469 *
1470 * Callers outside IN_GUEST_MODE context should hold a read lock on @gpc->lock
1471 * while calling this function, and then continue to hold the lock until the
1472 * access is complete.
1473 *
1474 * Callers in IN_GUEST_MODE may do so without locking, although they should
1475 * still hold a read lock on kvm->scru for the memslot checks.
1476 */
1477 bool kvm_gpc_check(struct gfn_to_pfn_cache *gpc, unsigned long len);
1478
1479 /**
1480 * kvm_gpc_refresh - update a previously initialized cache.
1481 *
1482 * @gpc: struct gfn_to_pfn_cache object.
1483 * @len: sanity check; the range being access must fit a single page.
1484 *
1485 * @return: 0 for success.
1486 * -EINVAL for a mapping which would cross a page boundary.
1487 * -EFAULT for an untranslatable guest physical address.
1488 *
1489 * This will attempt to refresh a gfn_to_pfn_cache. Note that a successful
1490 * return from this function does not mean the page can be immediately
1491 * accessed because it may have raced with an invalidation. Callers must
1492 * still lock and check the cache status, as this function does not return
1493 * with the lock still held to permit access.
1494 */
1495 int kvm_gpc_refresh(struct gfn_to_pfn_cache *gpc, unsigned long len);
1496
1497 /**
1498 * kvm_gpc_deactivate - deactivate and unlink a gfn_to_pfn_cache.
1499 *
1500 * @gpc: struct gfn_to_pfn_cache object.
1501 *
1502 * This removes a cache from the VM's list to be processed on MMU notifier
1503 * invocation.
1504 */
1505 void kvm_gpc_deactivate(struct gfn_to_pfn_cache *gpc);
1506
kvm_gpc_is_gpa_active(struct gfn_to_pfn_cache * gpc)1507 static inline bool kvm_gpc_is_gpa_active(struct gfn_to_pfn_cache *gpc)
1508 {
1509 return gpc->active && !kvm_is_error_gpa(gpc->gpa);
1510 }
1511
kvm_gpc_is_hva_active(struct gfn_to_pfn_cache * gpc)1512 static inline bool kvm_gpc_is_hva_active(struct gfn_to_pfn_cache *gpc)
1513 {
1514 return gpc->active && kvm_is_error_gpa(gpc->gpa);
1515 }
1516
1517 void kvm_sigset_activate(struct kvm_vcpu *vcpu);
1518 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu);
1519
1520 void kvm_vcpu_halt(struct kvm_vcpu *vcpu);
1521 bool kvm_vcpu_block(struct kvm_vcpu *vcpu);
1522 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu);
1523 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu);
1524 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu);
1525
1526 #ifndef CONFIG_S390
1527 void __kvm_vcpu_kick(struct kvm_vcpu *vcpu, bool wait);
1528
kvm_vcpu_kick(struct kvm_vcpu * vcpu)1529 static inline void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1530 {
1531 __kvm_vcpu_kick(vcpu, false);
1532 }
1533 #endif
1534
1535 int kvm_vcpu_yield_to(struct kvm_vcpu *target);
1536 void kvm_vcpu_on_spin(struct kvm_vcpu *vcpu, bool yield_to_kernel_mode);
1537
1538 void kvm_flush_remote_tlbs(struct kvm *kvm);
1539 void kvm_flush_remote_tlbs_range(struct kvm *kvm, gfn_t gfn, u64 nr_pages);
1540 void kvm_flush_remote_tlbs_memslot(struct kvm *kvm,
1541 const struct kvm_memory_slot *memslot);
1542
1543 #ifdef KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE
1544 int kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache *mc, int min);
1545 int __kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache *mc, int capacity, int min);
1546 int kvm_mmu_memory_cache_nr_free_objects(struct kvm_mmu_memory_cache *mc);
1547 void kvm_mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc);
1548 void *kvm_mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc);
1549 #endif
1550
1551 void kvm_mmu_invalidate_begin(struct kvm *kvm);
1552 void kvm_mmu_invalidate_range_add(struct kvm *kvm, gfn_t start, gfn_t end);
1553 void kvm_mmu_invalidate_end(struct kvm *kvm);
1554 bool kvm_mmu_unmap_gfn_range(struct kvm *kvm, struct kvm_gfn_range *range);
1555
1556 long kvm_arch_dev_ioctl(struct file *filp,
1557 unsigned int ioctl, unsigned long arg);
1558 long kvm_arch_vcpu_ioctl(struct file *filp,
1559 unsigned int ioctl, unsigned long arg);
1560 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf);
1561
1562 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext);
1563
1564 void kvm_arch_mmu_enable_log_dirty_pt_masked(struct kvm *kvm,
1565 struct kvm_memory_slot *slot,
1566 gfn_t gfn_offset,
1567 unsigned long mask);
1568 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot);
1569
1570 #ifndef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1571 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log);
1572 int kvm_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log,
1573 int *is_dirty, struct kvm_memory_slot **memslot);
1574 #endif
1575
1576 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
1577 bool line_status);
1578 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
1579 struct kvm_enable_cap *cap);
1580 int kvm_arch_vm_ioctl(struct file *filp, unsigned int ioctl, unsigned long arg);
1581 long kvm_arch_vm_compat_ioctl(struct file *filp, unsigned int ioctl,
1582 unsigned long arg);
1583
1584 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu);
1585 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu);
1586
1587 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
1588 struct kvm_translation *tr);
1589
1590 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs);
1591 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs);
1592 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
1593 struct kvm_sregs *sregs);
1594 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
1595 struct kvm_sregs *sregs);
1596 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
1597 struct kvm_mp_state *mp_state);
1598 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
1599 struct kvm_mp_state *mp_state);
1600 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
1601 struct kvm_guest_debug *dbg);
1602 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu);
1603
1604 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu);
1605 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu);
1606 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id);
1607 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu);
1608 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu);
1609 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu);
1610
1611 #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER
1612 int kvm_arch_pm_notifier(struct kvm *kvm, unsigned long state);
1613 #endif
1614
1615 #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS
1616 void kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu, struct dentry *debugfs_dentry);
1617 #else
kvm_create_vcpu_debugfs(struct kvm_vcpu * vcpu)1618 static inline void kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu) {}
1619 #endif
1620
1621 #ifdef CONFIG_KVM_GENERIC_HARDWARE_ENABLING
1622 /*
1623 * kvm_arch_{enable,disable}_virtualization() are called on one CPU, under
1624 * kvm_usage_lock, immediately after/before 0=>1 and 1=>0 transitions of
1625 * kvm_usage_count, i.e. at the beginning of the generic hardware enabling
1626 * sequence, and at the end of the generic hardware disabling sequence.
1627 */
1628 void kvm_arch_enable_virtualization(void);
1629 void kvm_arch_disable_virtualization(void);
1630 /*
1631 * kvm_arch_{enable,disable}_virtualization_cpu() are called on "every" CPU to
1632 * do the actual twiddling of hardware bits. The hooks are called on all
1633 * online CPUs when KVM enables/disabled virtualization, and on a single CPU
1634 * when that CPU is onlined/offlined (including for Resume/Suspend).
1635 */
1636 int kvm_arch_enable_virtualization_cpu(void);
1637 void kvm_arch_disable_virtualization_cpu(void);
1638 #endif
1639 bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu);
1640 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu);
1641 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu);
1642 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu);
1643 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu);
1644 bool kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu);
1645 bool kvm_arch_vcpu_preempted_in_kernel(struct kvm_vcpu *vcpu);
1646 void kvm_arch_pre_destroy_vm(struct kvm *kvm);
1647 void kvm_arch_create_vm_debugfs(struct kvm *kvm);
1648
1649 #ifndef __KVM_HAVE_ARCH_VM_ALLOC
1650 /*
1651 * All architectures that want to use vzalloc currently also
1652 * need their own kvm_arch_alloc_vm implementation.
1653 */
kvm_arch_alloc_vm(void)1654 static inline struct kvm *kvm_arch_alloc_vm(void)
1655 {
1656 return kzalloc(sizeof(struct kvm), GFP_KERNEL_ACCOUNT);
1657 }
1658 #endif
1659
__kvm_arch_free_vm(struct kvm * kvm)1660 static inline void __kvm_arch_free_vm(struct kvm *kvm)
1661 {
1662 kvfree(kvm);
1663 }
1664
1665 #ifndef __KVM_HAVE_ARCH_VM_FREE
kvm_arch_free_vm(struct kvm * kvm)1666 static inline void kvm_arch_free_vm(struct kvm *kvm)
1667 {
1668 __kvm_arch_free_vm(kvm);
1669 }
1670 #endif
1671
1672 #ifndef __KVM_HAVE_ARCH_FLUSH_REMOTE_TLBS
kvm_arch_flush_remote_tlbs(struct kvm * kvm)1673 static inline int kvm_arch_flush_remote_tlbs(struct kvm *kvm)
1674 {
1675 return -ENOTSUPP;
1676 }
1677 #else
1678 int kvm_arch_flush_remote_tlbs(struct kvm *kvm);
1679 #endif
1680
1681 #ifndef __KVM_HAVE_ARCH_FLUSH_REMOTE_TLBS_RANGE
kvm_arch_flush_remote_tlbs_range(struct kvm * kvm,gfn_t gfn,u64 nr_pages)1682 static inline int kvm_arch_flush_remote_tlbs_range(struct kvm *kvm,
1683 gfn_t gfn, u64 nr_pages)
1684 {
1685 return -EOPNOTSUPP;
1686 }
1687 #else
1688 int kvm_arch_flush_remote_tlbs_range(struct kvm *kvm, gfn_t gfn, u64 nr_pages);
1689 #endif
1690
1691 #ifdef __KVM_HAVE_ARCH_NONCOHERENT_DMA
1692 void kvm_arch_register_noncoherent_dma(struct kvm *kvm);
1693 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm);
1694 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm);
1695 #else
kvm_arch_register_noncoherent_dma(struct kvm * kvm)1696 static inline void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
1697 {
1698 }
1699
kvm_arch_unregister_noncoherent_dma(struct kvm * kvm)1700 static inline void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
1701 {
1702 }
1703
kvm_arch_has_noncoherent_dma(struct kvm * kvm)1704 static inline bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
1705 {
1706 return false;
1707 }
1708 #endif
1709
kvm_arch_vcpu_get_wait(struct kvm_vcpu * vcpu)1710 static inline struct rcuwait *kvm_arch_vcpu_get_wait(struct kvm_vcpu *vcpu)
1711 {
1712 #ifdef __KVM_HAVE_ARCH_WQP
1713 return vcpu->arch.waitp;
1714 #else
1715 return &vcpu->wait;
1716 #endif
1717 }
1718
1719 /*
1720 * Wake a vCPU if necessary, but don't do any stats/metadata updates. Returns
1721 * true if the vCPU was blocking and was awakened, false otherwise.
1722 */
__kvm_vcpu_wake_up(struct kvm_vcpu * vcpu)1723 static inline bool __kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
1724 {
1725 return !!rcuwait_wake_up(kvm_arch_vcpu_get_wait(vcpu));
1726 }
1727
kvm_vcpu_is_blocking(struct kvm_vcpu * vcpu)1728 static inline bool kvm_vcpu_is_blocking(struct kvm_vcpu *vcpu)
1729 {
1730 return rcuwait_active(kvm_arch_vcpu_get_wait(vcpu));
1731 }
1732
1733 #ifdef __KVM_HAVE_ARCH_INTC_INITIALIZED
1734 /*
1735 * returns true if the virtual interrupt controller is initialized and
1736 * ready to accept virtual IRQ. On some architectures the virtual interrupt
1737 * controller is dynamically instantiated and this is not always true.
1738 */
1739 bool kvm_arch_intc_initialized(struct kvm *kvm);
1740 #else
kvm_arch_intc_initialized(struct kvm * kvm)1741 static inline bool kvm_arch_intc_initialized(struct kvm *kvm)
1742 {
1743 return true;
1744 }
1745 #endif
1746
1747 #ifdef CONFIG_GUEST_PERF_EVENTS
1748 unsigned long kvm_arch_vcpu_get_ip(struct kvm_vcpu *vcpu);
1749
1750 void kvm_register_perf_callbacks(unsigned int (*pt_intr_handler)(void));
1751 void kvm_unregister_perf_callbacks(void);
1752 #else
kvm_register_perf_callbacks(void * ign)1753 static inline void kvm_register_perf_callbacks(void *ign) {}
kvm_unregister_perf_callbacks(void)1754 static inline void kvm_unregister_perf_callbacks(void) {}
1755 #endif /* CONFIG_GUEST_PERF_EVENTS */
1756
1757 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type);
1758 void kvm_arch_destroy_vm(struct kvm *kvm);
1759
1760 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu);
1761
1762 struct kvm_irq_ack_notifier {
1763 struct hlist_node link;
1764 unsigned gsi;
1765 void (*irq_acked)(struct kvm_irq_ack_notifier *kian);
1766 };
1767
1768 int kvm_irq_map_gsi(struct kvm *kvm,
1769 struct kvm_kernel_irq_routing_entry *entries, int gsi);
1770 int kvm_irq_map_chip_pin(struct kvm *kvm, unsigned irqchip, unsigned pin);
1771
1772 int kvm_set_irq(struct kvm *kvm, int irq_source_id, u32 irq, int level,
1773 bool line_status);
1774 int kvm_set_msi(struct kvm_kernel_irq_routing_entry *irq_entry, struct kvm *kvm,
1775 int irq_source_id, int level, bool line_status);
1776 int kvm_arch_set_irq_inatomic(struct kvm_kernel_irq_routing_entry *e,
1777 struct kvm *kvm, int irq_source_id,
1778 int level, bool line_status);
1779 bool kvm_irq_has_notifier(struct kvm *kvm, unsigned irqchip, unsigned pin);
1780 void kvm_notify_acked_gsi(struct kvm *kvm, int gsi);
1781 void kvm_notify_acked_irq(struct kvm *kvm, unsigned irqchip, unsigned pin);
1782 void kvm_register_irq_ack_notifier(struct kvm *kvm,
1783 struct kvm_irq_ack_notifier *kian);
1784 void kvm_unregister_irq_ack_notifier(struct kvm *kvm,
1785 struct kvm_irq_ack_notifier *kian);
1786 bool kvm_arch_irqfd_allowed(struct kvm *kvm, struct kvm_irqfd *args);
1787
1788 /*
1789 * Returns a pointer to the memslot if it contains gfn.
1790 * Otherwise returns NULL.
1791 */
1792 static inline struct kvm_memory_slot *
try_get_memslot(struct kvm_memory_slot * slot,gfn_t gfn)1793 try_get_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1794 {
1795 if (!slot)
1796 return NULL;
1797
1798 if (gfn >= slot->base_gfn && gfn < slot->base_gfn + slot->npages)
1799 return slot;
1800 else
1801 return NULL;
1802 }
1803
1804 /*
1805 * Returns a pointer to the memslot that contains gfn. Otherwise returns NULL.
1806 *
1807 * With "approx" set returns the memslot also when the address falls
1808 * in a hole. In that case one of the memslots bordering the hole is
1809 * returned.
1810 */
1811 static inline struct kvm_memory_slot *
search_memslots(struct kvm_memslots * slots,gfn_t gfn,bool approx)1812 search_memslots(struct kvm_memslots *slots, gfn_t gfn, bool approx)
1813 {
1814 struct kvm_memory_slot *slot;
1815 struct rb_node *node;
1816 int idx = slots->node_idx;
1817
1818 slot = NULL;
1819 for (node = slots->gfn_tree.rb_node; node; ) {
1820 slot = container_of(node, struct kvm_memory_slot, gfn_node[idx]);
1821 if (gfn >= slot->base_gfn) {
1822 if (gfn < slot->base_gfn + slot->npages)
1823 return slot;
1824 node = node->rb_right;
1825 } else
1826 node = node->rb_left;
1827 }
1828
1829 return approx ? slot : NULL;
1830 }
1831
1832 static inline struct kvm_memory_slot *
____gfn_to_memslot(struct kvm_memslots * slots,gfn_t gfn,bool approx)1833 ____gfn_to_memslot(struct kvm_memslots *slots, gfn_t gfn, bool approx)
1834 {
1835 struct kvm_memory_slot *slot;
1836
1837 slot = (struct kvm_memory_slot *)atomic_long_read(&slots->last_used_slot);
1838 slot = try_get_memslot(slot, gfn);
1839 if (slot)
1840 return slot;
1841
1842 slot = search_memslots(slots, gfn, approx);
1843 if (slot) {
1844 atomic_long_set(&slots->last_used_slot, (unsigned long)slot);
1845 return slot;
1846 }
1847
1848 return NULL;
1849 }
1850
1851 /*
1852 * __gfn_to_memslot() and its descendants are here to allow arch code to inline
1853 * the lookups in hot paths. gfn_to_memslot() itself isn't here as an inline
1854 * because that would bloat other code too much.
1855 */
1856 static inline struct kvm_memory_slot *
__gfn_to_memslot(struct kvm_memslots * slots,gfn_t gfn)1857 __gfn_to_memslot(struct kvm_memslots *slots, gfn_t gfn)
1858 {
1859 return ____gfn_to_memslot(slots, gfn, false);
1860 }
1861
1862 static inline unsigned long
__gfn_to_hva_memslot(const struct kvm_memory_slot * slot,gfn_t gfn)1863 __gfn_to_hva_memslot(const struct kvm_memory_slot *slot, gfn_t gfn)
1864 {
1865 /*
1866 * The index was checked originally in search_memslots. To avoid
1867 * that a malicious guest builds a Spectre gadget out of e.g. page
1868 * table walks, do not let the processor speculate loads outside
1869 * the guest's registered memslots.
1870 */
1871 unsigned long offset = gfn - slot->base_gfn;
1872 offset = array_index_nospec(offset, slot->npages);
1873 return slot->userspace_addr + offset * PAGE_SIZE;
1874 }
1875
memslot_id(struct kvm * kvm,gfn_t gfn)1876 static inline int memslot_id(struct kvm *kvm, gfn_t gfn)
1877 {
1878 return gfn_to_memslot(kvm, gfn)->id;
1879 }
1880
1881 static inline gfn_t
hva_to_gfn_memslot(unsigned long hva,struct kvm_memory_slot * slot)1882 hva_to_gfn_memslot(unsigned long hva, struct kvm_memory_slot *slot)
1883 {
1884 gfn_t gfn_offset = (hva - slot->userspace_addr) >> PAGE_SHIFT;
1885
1886 return slot->base_gfn + gfn_offset;
1887 }
1888
gfn_to_gpa(gfn_t gfn)1889 static inline gpa_t gfn_to_gpa(gfn_t gfn)
1890 {
1891 return (gpa_t)gfn << PAGE_SHIFT;
1892 }
1893
gpa_to_gfn(gpa_t gpa)1894 static inline gfn_t gpa_to_gfn(gpa_t gpa)
1895 {
1896 return (gfn_t)(gpa >> PAGE_SHIFT);
1897 }
1898
pfn_to_hpa(kvm_pfn_t pfn)1899 static inline hpa_t pfn_to_hpa(kvm_pfn_t pfn)
1900 {
1901 return (hpa_t)pfn << PAGE_SHIFT;
1902 }
1903
kvm_is_gpa_in_memslot(struct kvm * kvm,gpa_t gpa)1904 static inline bool kvm_is_gpa_in_memslot(struct kvm *kvm, gpa_t gpa)
1905 {
1906 unsigned long hva = gfn_to_hva(kvm, gpa_to_gfn(gpa));
1907
1908 return !kvm_is_error_hva(hva);
1909 }
1910
kvm_gpc_mark_dirty_in_slot(struct gfn_to_pfn_cache * gpc)1911 static inline void kvm_gpc_mark_dirty_in_slot(struct gfn_to_pfn_cache *gpc)
1912 {
1913 lockdep_assert_held(&gpc->lock);
1914
1915 if (!gpc->memslot)
1916 return;
1917
1918 mark_page_dirty_in_slot(gpc->kvm, gpc->memslot, gpa_to_gfn(gpc->gpa));
1919 }
1920
1921 enum kvm_stat_kind {
1922 KVM_STAT_VM,
1923 KVM_STAT_VCPU,
1924 };
1925
1926 struct kvm_stat_data {
1927 struct kvm *kvm;
1928 const struct _kvm_stats_desc *desc;
1929 enum kvm_stat_kind kind;
1930 };
1931
1932 struct _kvm_stats_desc {
1933 struct kvm_stats_desc desc;
1934 char name[KVM_STATS_NAME_SIZE];
1935 };
1936
1937 #define STATS_DESC_COMMON(type, unit, base, exp, sz, bsz) \
1938 .flags = type | unit | base | \
1939 BUILD_BUG_ON_ZERO(type & ~KVM_STATS_TYPE_MASK) | \
1940 BUILD_BUG_ON_ZERO(unit & ~KVM_STATS_UNIT_MASK) | \
1941 BUILD_BUG_ON_ZERO(base & ~KVM_STATS_BASE_MASK), \
1942 .exponent = exp, \
1943 .size = sz, \
1944 .bucket_size = bsz
1945
1946 #define VM_GENERIC_STATS_DESC(stat, type, unit, base, exp, sz, bsz) \
1947 { \
1948 { \
1949 STATS_DESC_COMMON(type, unit, base, exp, sz, bsz), \
1950 .offset = offsetof(struct kvm_vm_stat, generic.stat) \
1951 }, \
1952 .name = #stat, \
1953 }
1954 #define VCPU_GENERIC_STATS_DESC(stat, type, unit, base, exp, sz, bsz) \
1955 { \
1956 { \
1957 STATS_DESC_COMMON(type, unit, base, exp, sz, bsz), \
1958 .offset = offsetof(struct kvm_vcpu_stat, generic.stat) \
1959 }, \
1960 .name = #stat, \
1961 }
1962 #define VM_STATS_DESC(stat, type, unit, base, exp, sz, bsz) \
1963 { \
1964 { \
1965 STATS_DESC_COMMON(type, unit, base, exp, sz, bsz), \
1966 .offset = offsetof(struct kvm_vm_stat, stat) \
1967 }, \
1968 .name = #stat, \
1969 }
1970 #define VCPU_STATS_DESC(stat, type, unit, base, exp, sz, bsz) \
1971 { \
1972 { \
1973 STATS_DESC_COMMON(type, unit, base, exp, sz, bsz), \
1974 .offset = offsetof(struct kvm_vcpu_stat, stat) \
1975 }, \
1976 .name = #stat, \
1977 }
1978 /* SCOPE: VM, VM_GENERIC, VCPU, VCPU_GENERIC */
1979 #define STATS_DESC(SCOPE, stat, type, unit, base, exp, sz, bsz) \
1980 SCOPE##_STATS_DESC(stat, type, unit, base, exp, sz, bsz)
1981
1982 #define STATS_DESC_CUMULATIVE(SCOPE, name, unit, base, exponent) \
1983 STATS_DESC(SCOPE, name, KVM_STATS_TYPE_CUMULATIVE, \
1984 unit, base, exponent, 1, 0)
1985 #define STATS_DESC_INSTANT(SCOPE, name, unit, base, exponent) \
1986 STATS_DESC(SCOPE, name, KVM_STATS_TYPE_INSTANT, \
1987 unit, base, exponent, 1, 0)
1988 #define STATS_DESC_PEAK(SCOPE, name, unit, base, exponent) \
1989 STATS_DESC(SCOPE, name, KVM_STATS_TYPE_PEAK, \
1990 unit, base, exponent, 1, 0)
1991 #define STATS_DESC_LINEAR_HIST(SCOPE, name, unit, base, exponent, sz, bsz) \
1992 STATS_DESC(SCOPE, name, KVM_STATS_TYPE_LINEAR_HIST, \
1993 unit, base, exponent, sz, bsz)
1994 #define STATS_DESC_LOG_HIST(SCOPE, name, unit, base, exponent, sz) \
1995 STATS_DESC(SCOPE, name, KVM_STATS_TYPE_LOG_HIST, \
1996 unit, base, exponent, sz, 0)
1997
1998 /* Cumulative counter, read/write */
1999 #define STATS_DESC_COUNTER(SCOPE, name) \
2000 STATS_DESC_CUMULATIVE(SCOPE, name, KVM_STATS_UNIT_NONE, \
2001 KVM_STATS_BASE_POW10, 0)
2002 /* Instantaneous counter, read only */
2003 #define STATS_DESC_ICOUNTER(SCOPE, name) \
2004 STATS_DESC_INSTANT(SCOPE, name, KVM_STATS_UNIT_NONE, \
2005 KVM_STATS_BASE_POW10, 0)
2006 /* Peak counter, read/write */
2007 #define STATS_DESC_PCOUNTER(SCOPE, name) \
2008 STATS_DESC_PEAK(SCOPE, name, KVM_STATS_UNIT_NONE, \
2009 KVM_STATS_BASE_POW10, 0)
2010
2011 /* Instantaneous boolean value, read only */
2012 #define STATS_DESC_IBOOLEAN(SCOPE, name) \
2013 STATS_DESC_INSTANT(SCOPE, name, KVM_STATS_UNIT_BOOLEAN, \
2014 KVM_STATS_BASE_POW10, 0)
2015 /* Peak (sticky) boolean value, read/write */
2016 #define STATS_DESC_PBOOLEAN(SCOPE, name) \
2017 STATS_DESC_PEAK(SCOPE, name, KVM_STATS_UNIT_BOOLEAN, \
2018 KVM_STATS_BASE_POW10, 0)
2019
2020 /* Cumulative time in nanosecond */
2021 #define STATS_DESC_TIME_NSEC(SCOPE, name) \
2022 STATS_DESC_CUMULATIVE(SCOPE, name, KVM_STATS_UNIT_SECONDS, \
2023 KVM_STATS_BASE_POW10, -9)
2024 /* Linear histogram for time in nanosecond */
2025 #define STATS_DESC_LINHIST_TIME_NSEC(SCOPE, name, sz, bsz) \
2026 STATS_DESC_LINEAR_HIST(SCOPE, name, KVM_STATS_UNIT_SECONDS, \
2027 KVM_STATS_BASE_POW10, -9, sz, bsz)
2028 /* Logarithmic histogram for time in nanosecond */
2029 #define STATS_DESC_LOGHIST_TIME_NSEC(SCOPE, name, sz) \
2030 STATS_DESC_LOG_HIST(SCOPE, name, KVM_STATS_UNIT_SECONDS, \
2031 KVM_STATS_BASE_POW10, -9, sz)
2032
2033 #define KVM_GENERIC_VM_STATS() \
2034 STATS_DESC_COUNTER(VM_GENERIC, remote_tlb_flush), \
2035 STATS_DESC_COUNTER(VM_GENERIC, remote_tlb_flush_requests)
2036
2037 #define KVM_GENERIC_VCPU_STATS() \
2038 STATS_DESC_COUNTER(VCPU_GENERIC, halt_successful_poll), \
2039 STATS_DESC_COUNTER(VCPU_GENERIC, halt_attempted_poll), \
2040 STATS_DESC_COUNTER(VCPU_GENERIC, halt_poll_invalid), \
2041 STATS_DESC_COUNTER(VCPU_GENERIC, halt_wakeup), \
2042 STATS_DESC_TIME_NSEC(VCPU_GENERIC, halt_poll_success_ns), \
2043 STATS_DESC_TIME_NSEC(VCPU_GENERIC, halt_poll_fail_ns), \
2044 STATS_DESC_TIME_NSEC(VCPU_GENERIC, halt_wait_ns), \
2045 STATS_DESC_LOGHIST_TIME_NSEC(VCPU_GENERIC, halt_poll_success_hist, \
2046 HALT_POLL_HIST_COUNT), \
2047 STATS_DESC_LOGHIST_TIME_NSEC(VCPU_GENERIC, halt_poll_fail_hist, \
2048 HALT_POLL_HIST_COUNT), \
2049 STATS_DESC_LOGHIST_TIME_NSEC(VCPU_GENERIC, halt_wait_hist, \
2050 HALT_POLL_HIST_COUNT), \
2051 STATS_DESC_IBOOLEAN(VCPU_GENERIC, blocking)
2052
2053 ssize_t kvm_stats_read(char *id, const struct kvm_stats_header *header,
2054 const struct _kvm_stats_desc *desc,
2055 void *stats, size_t size_stats,
2056 char __user *user_buffer, size_t size, loff_t *offset);
2057
2058 /**
2059 * kvm_stats_linear_hist_update() - Update bucket value for linear histogram
2060 * statistics data.
2061 *
2062 * @data: start address of the stats data
2063 * @size: the number of bucket of the stats data
2064 * @value: the new value used to update the linear histogram's bucket
2065 * @bucket_size: the size (width) of a bucket
2066 */
kvm_stats_linear_hist_update(u64 * data,size_t size,u64 value,size_t bucket_size)2067 static inline void kvm_stats_linear_hist_update(u64 *data, size_t size,
2068 u64 value, size_t bucket_size)
2069 {
2070 size_t index = div64_u64(value, bucket_size);
2071
2072 index = min(index, size - 1);
2073 ++data[index];
2074 }
2075
2076 /**
2077 * kvm_stats_log_hist_update() - Update bucket value for logarithmic histogram
2078 * statistics data.
2079 *
2080 * @data: start address of the stats data
2081 * @size: the number of bucket of the stats data
2082 * @value: the new value used to update the logarithmic histogram's bucket
2083 */
kvm_stats_log_hist_update(u64 * data,size_t size,u64 value)2084 static inline void kvm_stats_log_hist_update(u64 *data, size_t size, u64 value)
2085 {
2086 size_t index = fls64(value);
2087
2088 index = min(index, size - 1);
2089 ++data[index];
2090 }
2091
2092 #define KVM_STATS_LINEAR_HIST_UPDATE(array, value, bsize) \
2093 kvm_stats_linear_hist_update(array, ARRAY_SIZE(array), value, bsize)
2094 #define KVM_STATS_LOG_HIST_UPDATE(array, value) \
2095 kvm_stats_log_hist_update(array, ARRAY_SIZE(array), value)
2096
2097
2098 extern const struct kvm_stats_header kvm_vm_stats_header;
2099 extern const struct _kvm_stats_desc kvm_vm_stats_desc[];
2100 extern const struct kvm_stats_header kvm_vcpu_stats_header;
2101 extern const struct _kvm_stats_desc kvm_vcpu_stats_desc[];
2102
2103 #ifdef CONFIG_KVM_GENERIC_MMU_NOTIFIER
mmu_invalidate_retry(struct kvm * kvm,unsigned long mmu_seq)2104 static inline int mmu_invalidate_retry(struct kvm *kvm, unsigned long mmu_seq)
2105 {
2106 if (unlikely(kvm->mmu_invalidate_in_progress))
2107 return 1;
2108 /*
2109 * Ensure the read of mmu_invalidate_in_progress happens before
2110 * the read of mmu_invalidate_seq. This interacts with the
2111 * smp_wmb() in mmu_notifier_invalidate_range_end to make sure
2112 * that the caller either sees the old (non-zero) value of
2113 * mmu_invalidate_in_progress or the new (incremented) value of
2114 * mmu_invalidate_seq.
2115 *
2116 * PowerPC Book3s HV KVM calls this under a per-page lock rather
2117 * than under kvm->mmu_lock, for scalability, so can't rely on
2118 * kvm->mmu_lock to keep things ordered.
2119 */
2120 smp_rmb();
2121 if (kvm->mmu_invalidate_seq != mmu_seq)
2122 return 1;
2123 return 0;
2124 }
2125
mmu_invalidate_retry_gfn(struct kvm * kvm,unsigned long mmu_seq,gfn_t gfn)2126 static inline int mmu_invalidate_retry_gfn(struct kvm *kvm,
2127 unsigned long mmu_seq,
2128 gfn_t gfn)
2129 {
2130 lockdep_assert_held(&kvm->mmu_lock);
2131 /*
2132 * If mmu_invalidate_in_progress is non-zero, then the range maintained
2133 * by kvm_mmu_notifier_invalidate_range_start contains all addresses
2134 * that might be being invalidated. Note that it may include some false
2135 * positives, due to shortcuts when handing concurrent invalidations.
2136 */
2137 if (unlikely(kvm->mmu_invalidate_in_progress)) {
2138 /*
2139 * Dropping mmu_lock after bumping mmu_invalidate_in_progress
2140 * but before updating the range is a KVM bug.
2141 */
2142 if (WARN_ON_ONCE(kvm->mmu_invalidate_range_start == INVALID_GPA ||
2143 kvm->mmu_invalidate_range_end == INVALID_GPA))
2144 return 1;
2145
2146 if (gfn >= kvm->mmu_invalidate_range_start &&
2147 gfn < kvm->mmu_invalidate_range_end)
2148 return 1;
2149 }
2150
2151 if (kvm->mmu_invalidate_seq != mmu_seq)
2152 return 1;
2153 return 0;
2154 }
2155
2156 /*
2157 * This lockless version of the range-based retry check *must* be paired with a
2158 * call to the locked version after acquiring mmu_lock, i.e. this is safe to
2159 * use only as a pre-check to avoid contending mmu_lock. This version *will*
2160 * get false negatives and false positives.
2161 */
mmu_invalidate_retry_gfn_unsafe(struct kvm * kvm,unsigned long mmu_seq,gfn_t gfn)2162 static inline bool mmu_invalidate_retry_gfn_unsafe(struct kvm *kvm,
2163 unsigned long mmu_seq,
2164 gfn_t gfn)
2165 {
2166 /*
2167 * Use READ_ONCE() to ensure the in-progress flag and sequence counter
2168 * are always read from memory, e.g. so that checking for retry in a
2169 * loop won't result in an infinite retry loop. Don't force loads for
2170 * start+end, as the key to avoiding infinite retry loops is observing
2171 * the 1=>0 transition of in-progress, i.e. getting false negatives
2172 * due to stale start+end values is acceptable.
2173 */
2174 if (unlikely(READ_ONCE(kvm->mmu_invalidate_in_progress)) &&
2175 gfn >= kvm->mmu_invalidate_range_start &&
2176 gfn < kvm->mmu_invalidate_range_end)
2177 return true;
2178
2179 return READ_ONCE(kvm->mmu_invalidate_seq) != mmu_seq;
2180 }
2181 #endif
2182
2183 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2184
2185 #define KVM_MAX_IRQ_ROUTES 4096 /* might need extension/rework in the future */
2186
2187 bool kvm_arch_can_set_irq_routing(struct kvm *kvm);
2188 int kvm_set_irq_routing(struct kvm *kvm,
2189 const struct kvm_irq_routing_entry *entries,
2190 unsigned nr,
2191 unsigned flags);
2192 int kvm_init_irq_routing(struct kvm *kvm);
2193 int kvm_set_routing_entry(struct kvm *kvm,
2194 struct kvm_kernel_irq_routing_entry *e,
2195 const struct kvm_irq_routing_entry *ue);
2196 void kvm_free_irq_routing(struct kvm *kvm);
2197
2198 #else
2199
kvm_free_irq_routing(struct kvm * kvm)2200 static inline void kvm_free_irq_routing(struct kvm *kvm) {}
2201
kvm_init_irq_routing(struct kvm * kvm)2202 static inline int kvm_init_irq_routing(struct kvm *kvm)
2203 {
2204 return 0;
2205 }
2206
2207 #endif
2208
2209 int kvm_send_userspace_msi(struct kvm *kvm, struct kvm_msi *msi);
2210
2211 void kvm_eventfd_init(struct kvm *kvm);
2212 int kvm_ioeventfd(struct kvm *kvm, struct kvm_ioeventfd *args);
2213
2214 #ifdef CONFIG_HAVE_KVM_IRQCHIP
2215 int kvm_irqfd(struct kvm *kvm, struct kvm_irqfd *args);
2216 void kvm_irqfd_release(struct kvm *kvm);
2217 bool kvm_notify_irqfd_resampler(struct kvm *kvm,
2218 unsigned int irqchip,
2219 unsigned int pin);
2220 void kvm_irq_routing_update(struct kvm *);
2221 #else
kvm_irqfd(struct kvm * kvm,struct kvm_irqfd * args)2222 static inline int kvm_irqfd(struct kvm *kvm, struct kvm_irqfd *args)
2223 {
2224 return -EINVAL;
2225 }
2226
kvm_irqfd_release(struct kvm * kvm)2227 static inline void kvm_irqfd_release(struct kvm *kvm) {}
2228
kvm_notify_irqfd_resampler(struct kvm * kvm,unsigned int irqchip,unsigned int pin)2229 static inline bool kvm_notify_irqfd_resampler(struct kvm *kvm,
2230 unsigned int irqchip,
2231 unsigned int pin)
2232 {
2233 return false;
2234 }
2235 #endif /* CONFIG_HAVE_KVM_IRQCHIP */
2236
2237 void kvm_arch_irq_routing_update(struct kvm *kvm);
2238
__kvm_make_request(int req,struct kvm_vcpu * vcpu)2239 static inline void __kvm_make_request(int req, struct kvm_vcpu *vcpu)
2240 {
2241 /*
2242 * Ensure the rest of the request is published to kvm_check_request's
2243 * caller. Paired with the smp_mb__after_atomic in kvm_check_request.
2244 */
2245 smp_wmb();
2246 set_bit(req & KVM_REQUEST_MASK, (void *)&vcpu->requests);
2247 }
2248
kvm_make_request(int req,struct kvm_vcpu * vcpu)2249 static __always_inline void kvm_make_request(int req, struct kvm_vcpu *vcpu)
2250 {
2251 /*
2252 * Request that don't require vCPU action should never be logged in
2253 * vcpu->requests. The vCPU won't clear the request, so it will stay
2254 * logged indefinitely and prevent the vCPU from entering the guest.
2255 */
2256 BUILD_BUG_ON(!__builtin_constant_p(req) ||
2257 (req & KVM_REQUEST_NO_ACTION));
2258
2259 __kvm_make_request(req, vcpu);
2260 }
2261
2262 #ifndef CONFIG_S390
kvm_make_request_and_kick(int req,struct kvm_vcpu * vcpu)2263 static inline void kvm_make_request_and_kick(int req, struct kvm_vcpu *vcpu)
2264 {
2265 kvm_make_request(req, vcpu);
2266 __kvm_vcpu_kick(vcpu, req & KVM_REQUEST_WAIT);
2267 }
2268 #endif
2269
kvm_request_pending(struct kvm_vcpu * vcpu)2270 static inline bool kvm_request_pending(struct kvm_vcpu *vcpu)
2271 {
2272 return READ_ONCE(vcpu->requests);
2273 }
2274
kvm_test_request(int req,struct kvm_vcpu * vcpu)2275 static inline bool kvm_test_request(int req, struct kvm_vcpu *vcpu)
2276 {
2277 return test_bit(req & KVM_REQUEST_MASK, (void *)&vcpu->requests);
2278 }
2279
kvm_clear_request(int req,struct kvm_vcpu * vcpu)2280 static inline void kvm_clear_request(int req, struct kvm_vcpu *vcpu)
2281 {
2282 clear_bit(req & KVM_REQUEST_MASK, (void *)&vcpu->requests);
2283 }
2284
kvm_check_request(int req,struct kvm_vcpu * vcpu)2285 static inline bool kvm_check_request(int req, struct kvm_vcpu *vcpu)
2286 {
2287 if (kvm_test_request(req, vcpu)) {
2288 kvm_clear_request(req, vcpu);
2289
2290 /*
2291 * Ensure the rest of the request is visible to kvm_check_request's
2292 * caller. Paired with the smp_wmb in kvm_make_request.
2293 */
2294 smp_mb__after_atomic();
2295 return true;
2296 } else {
2297 return false;
2298 }
2299 }
2300
2301 #ifdef CONFIG_KVM_GENERIC_HARDWARE_ENABLING
2302 extern bool enable_virt_at_load;
2303 extern bool kvm_rebooting;
2304 #endif
2305
2306 extern unsigned int halt_poll_ns;
2307 extern unsigned int halt_poll_ns_grow;
2308 extern unsigned int halt_poll_ns_grow_start;
2309 extern unsigned int halt_poll_ns_shrink;
2310
2311 struct kvm_device {
2312 const struct kvm_device_ops *ops;
2313 struct kvm *kvm;
2314 void *private;
2315 struct list_head vm_node;
2316 };
2317
2318 /* create, destroy, and name are mandatory */
2319 struct kvm_device_ops {
2320 const char *name;
2321
2322 /*
2323 * create is called holding kvm->lock and any operations not suitable
2324 * to do while holding the lock should be deferred to init (see
2325 * below).
2326 */
2327 int (*create)(struct kvm_device *dev, u32 type);
2328
2329 /*
2330 * init is called after create if create is successful and is called
2331 * outside of holding kvm->lock.
2332 */
2333 void (*init)(struct kvm_device *dev);
2334
2335 /*
2336 * Destroy is responsible for freeing dev.
2337 *
2338 * Destroy may be called before or after destructors are called
2339 * on emulated I/O regions, depending on whether a reference is
2340 * held by a vcpu or other kvm component that gets destroyed
2341 * after the emulated I/O.
2342 */
2343 void (*destroy)(struct kvm_device *dev);
2344
2345 /*
2346 * Release is an alternative method to free the device. It is
2347 * called when the device file descriptor is closed. Once
2348 * release is called, the destroy method will not be called
2349 * anymore as the device is removed from the device list of
2350 * the VM. kvm->lock is held.
2351 */
2352 void (*release)(struct kvm_device *dev);
2353
2354 int (*set_attr)(struct kvm_device *dev, struct kvm_device_attr *attr);
2355 int (*get_attr)(struct kvm_device *dev, struct kvm_device_attr *attr);
2356 int (*has_attr)(struct kvm_device *dev, struct kvm_device_attr *attr);
2357 long (*ioctl)(struct kvm_device *dev, unsigned int ioctl,
2358 unsigned long arg);
2359 int (*mmap)(struct kvm_device *dev, struct vm_area_struct *vma);
2360 };
2361
2362 struct kvm_device *kvm_device_from_filp(struct file *filp);
2363 int kvm_register_device_ops(const struct kvm_device_ops *ops, u32 type);
2364 void kvm_unregister_device_ops(u32 type);
2365
2366 extern struct kvm_device_ops kvm_mpic_ops;
2367 extern struct kvm_device_ops kvm_arm_vgic_v2_ops;
2368 extern struct kvm_device_ops kvm_arm_vgic_v3_ops;
2369
2370 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2371
kvm_vcpu_set_in_spin_loop(struct kvm_vcpu * vcpu,bool val)2372 static inline void kvm_vcpu_set_in_spin_loop(struct kvm_vcpu *vcpu, bool val)
2373 {
2374 vcpu->spin_loop.in_spin_loop = val;
2375 }
kvm_vcpu_set_dy_eligible(struct kvm_vcpu * vcpu,bool val)2376 static inline void kvm_vcpu_set_dy_eligible(struct kvm_vcpu *vcpu, bool val)
2377 {
2378 vcpu->spin_loop.dy_eligible = val;
2379 }
2380
2381 #else /* !CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT */
2382
kvm_vcpu_set_in_spin_loop(struct kvm_vcpu * vcpu,bool val)2383 static inline void kvm_vcpu_set_in_spin_loop(struct kvm_vcpu *vcpu, bool val)
2384 {
2385 }
2386
kvm_vcpu_set_dy_eligible(struct kvm_vcpu * vcpu,bool val)2387 static inline void kvm_vcpu_set_dy_eligible(struct kvm_vcpu *vcpu, bool val)
2388 {
2389 }
2390 #endif /* CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT */
2391
kvm_is_visible_memslot(struct kvm_memory_slot * memslot)2392 static inline bool kvm_is_visible_memslot(struct kvm_memory_slot *memslot)
2393 {
2394 return (memslot && memslot->id < KVM_USER_MEM_SLOTS &&
2395 !(memslot->flags & KVM_MEMSLOT_INVALID));
2396 }
2397
2398 struct kvm_vcpu *kvm_get_running_vcpu(void);
2399 struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void);
2400
2401 #if IS_ENABLED(CONFIG_HAVE_KVM_IRQ_BYPASS)
2402 struct kvm_kernel_irqfd;
2403
2404 bool kvm_arch_has_irq_bypass(void);
2405 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *,
2406 struct irq_bypass_producer *);
2407 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *,
2408 struct irq_bypass_producer *);
2409 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *);
2410 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *);
2411 void kvm_arch_update_irqfd_routing(struct kvm_kernel_irqfd *irqfd,
2412 struct kvm_kernel_irq_routing_entry *old,
2413 struct kvm_kernel_irq_routing_entry *new);
2414 #endif /* CONFIG_HAVE_KVM_IRQ_BYPASS */
2415
2416 #ifdef CONFIG_HAVE_KVM_INVALID_WAKEUPS
2417 /* If we wakeup during the poll time, was it a sucessful poll? */
vcpu_valid_wakeup(struct kvm_vcpu * vcpu)2418 static inline bool vcpu_valid_wakeup(struct kvm_vcpu *vcpu)
2419 {
2420 return vcpu->valid_wakeup;
2421 }
2422
2423 #else
vcpu_valid_wakeup(struct kvm_vcpu * vcpu)2424 static inline bool vcpu_valid_wakeup(struct kvm_vcpu *vcpu)
2425 {
2426 return true;
2427 }
2428 #endif /* CONFIG_HAVE_KVM_INVALID_WAKEUPS */
2429
2430 #ifdef CONFIG_HAVE_KVM_NO_POLL
2431 /* Callback that tells if we must not poll */
2432 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu);
2433 #else
kvm_arch_no_poll(struct kvm_vcpu * vcpu)2434 static inline bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
2435 {
2436 return false;
2437 }
2438 #endif /* CONFIG_HAVE_KVM_NO_POLL */
2439
2440 #ifdef CONFIG_HAVE_KVM_VCPU_ASYNC_IOCTL
2441 long kvm_arch_vcpu_async_ioctl(struct file *filp,
2442 unsigned int ioctl, unsigned long arg);
2443 #else
kvm_arch_vcpu_async_ioctl(struct file * filp,unsigned int ioctl,unsigned long arg)2444 static inline long kvm_arch_vcpu_async_ioctl(struct file *filp,
2445 unsigned int ioctl,
2446 unsigned long arg)
2447 {
2448 return -ENOIOCTLCMD;
2449 }
2450 #endif /* CONFIG_HAVE_KVM_VCPU_ASYNC_IOCTL */
2451
2452 void kvm_arch_guest_memory_reclaimed(struct kvm *kvm);
2453
2454 #ifdef CONFIG_HAVE_KVM_VCPU_RUN_PID_CHANGE
2455 int kvm_arch_vcpu_run_pid_change(struct kvm_vcpu *vcpu);
2456 #else
kvm_arch_vcpu_run_pid_change(struct kvm_vcpu * vcpu)2457 static inline int kvm_arch_vcpu_run_pid_change(struct kvm_vcpu *vcpu)
2458 {
2459 return 0;
2460 }
2461 #endif /* CONFIG_HAVE_KVM_VCPU_RUN_PID_CHANGE */
2462
2463 #ifdef CONFIG_VIRT_XFER_TO_GUEST_WORK
kvm_handle_signal_exit(struct kvm_vcpu * vcpu)2464 static inline void kvm_handle_signal_exit(struct kvm_vcpu *vcpu)
2465 {
2466 vcpu->run->exit_reason = KVM_EXIT_INTR;
2467 vcpu->stat.signal_exits++;
2468 }
2469
kvm_xfer_to_guest_mode_handle_work(struct kvm_vcpu * vcpu)2470 static inline int kvm_xfer_to_guest_mode_handle_work(struct kvm_vcpu *vcpu)
2471 {
2472 int r = xfer_to_guest_mode_handle_work();
2473
2474 if (r) {
2475 WARN_ON_ONCE(r != -EINTR);
2476 kvm_handle_signal_exit(vcpu);
2477 }
2478 return r;
2479 }
2480 #endif /* CONFIG_VIRT_XFER_TO_GUEST_WORK */
2481
2482 /*
2483 * If more than one page is being (un)accounted, @virt must be the address of
2484 * the first page of a block of pages what were allocated together (i.e
2485 * accounted together).
2486 *
2487 * kvm_account_pgtable_pages() is thread-safe because mod_lruvec_page_state()
2488 * is thread-safe.
2489 */
kvm_account_pgtable_pages(void * virt,int nr)2490 static inline void kvm_account_pgtable_pages(void *virt, int nr)
2491 {
2492 mod_lruvec_page_state(virt_to_page(virt), NR_SECONDARY_PAGETABLE, nr);
2493 }
2494
2495 /*
2496 * This defines how many reserved entries we want to keep before we
2497 * kick the vcpu to the userspace to avoid dirty ring full. This
2498 * value can be tuned to higher if e.g. PML is enabled on the host.
2499 */
2500 #define KVM_DIRTY_RING_RSVD_ENTRIES 64
2501
2502 /* Max number of entries allowed for each kvm dirty ring */
2503 #define KVM_DIRTY_RING_MAX_ENTRIES 65536
2504
kvm_prepare_memory_fault_exit(struct kvm_vcpu * vcpu,gpa_t gpa,gpa_t size,bool is_write,bool is_exec,bool is_private)2505 static inline void kvm_prepare_memory_fault_exit(struct kvm_vcpu *vcpu,
2506 gpa_t gpa, gpa_t size,
2507 bool is_write, bool is_exec,
2508 bool is_private)
2509 {
2510 vcpu->run->exit_reason = KVM_EXIT_MEMORY_FAULT;
2511 vcpu->run->memory_fault.gpa = gpa;
2512 vcpu->run->memory_fault.size = size;
2513
2514 /* RWX flags are not (yet) defined or communicated to userspace. */
2515 vcpu->run->memory_fault.flags = 0;
2516 if (is_private)
2517 vcpu->run->memory_fault.flags |= KVM_MEMORY_EXIT_FLAG_PRIVATE;
2518 }
2519
kvm_memslot_is_gmem_only(const struct kvm_memory_slot * slot)2520 static inline bool kvm_memslot_is_gmem_only(const struct kvm_memory_slot *slot)
2521 {
2522 if (!IS_ENABLED(CONFIG_KVM_GUEST_MEMFD))
2523 return false;
2524
2525 return slot->flags & KVM_MEMSLOT_GMEM_ONLY;
2526 }
2527
2528 #ifdef CONFIG_KVM_GENERIC_MEMORY_ATTRIBUTES
kvm_get_memory_attributes(struct kvm * kvm,gfn_t gfn)2529 static inline unsigned long kvm_get_memory_attributes(struct kvm *kvm, gfn_t gfn)
2530 {
2531 return xa_to_value(xa_load(&kvm->mem_attr_array, gfn));
2532 }
2533
2534 bool kvm_range_has_memory_attributes(struct kvm *kvm, gfn_t start, gfn_t end,
2535 unsigned long mask, unsigned long attrs);
2536 bool kvm_arch_pre_set_memory_attributes(struct kvm *kvm,
2537 struct kvm_gfn_range *range);
2538 bool kvm_arch_post_set_memory_attributes(struct kvm *kvm,
2539 struct kvm_gfn_range *range);
2540
kvm_mem_is_private(struct kvm * kvm,gfn_t gfn)2541 static inline bool kvm_mem_is_private(struct kvm *kvm, gfn_t gfn)
2542 {
2543 return kvm_get_memory_attributes(kvm, gfn) & KVM_MEMORY_ATTRIBUTE_PRIVATE;
2544 }
2545 #else
kvm_mem_is_private(struct kvm * kvm,gfn_t gfn)2546 static inline bool kvm_mem_is_private(struct kvm *kvm, gfn_t gfn)
2547 {
2548 return false;
2549 }
2550 #endif /* CONFIG_KVM_GENERIC_MEMORY_ATTRIBUTES */
2551
2552 #ifdef CONFIG_KVM_GUEST_MEMFD
2553 int kvm_gmem_get_pfn(struct kvm *kvm, struct kvm_memory_slot *slot,
2554 gfn_t gfn, kvm_pfn_t *pfn, struct page **page,
2555 int *max_order);
2556 #else
kvm_gmem_get_pfn(struct kvm * kvm,struct kvm_memory_slot * slot,gfn_t gfn,kvm_pfn_t * pfn,struct page ** page,int * max_order)2557 static inline int kvm_gmem_get_pfn(struct kvm *kvm,
2558 struct kvm_memory_slot *slot, gfn_t gfn,
2559 kvm_pfn_t *pfn, struct page **page,
2560 int *max_order)
2561 {
2562 KVM_BUG_ON(1, kvm);
2563 return -EIO;
2564 }
2565 #endif /* CONFIG_KVM_GUEST_MEMFD */
2566
2567 #ifdef CONFIG_HAVE_KVM_ARCH_GMEM_PREPARE
2568 int kvm_arch_gmem_prepare(struct kvm *kvm, gfn_t gfn, kvm_pfn_t pfn, int max_order);
2569 #endif
2570
2571 #ifdef CONFIG_HAVE_KVM_ARCH_GMEM_POPULATE
2572 /**
2573 * kvm_gmem_populate() - Populate/prepare a GPA range with guest data
2574 *
2575 * @kvm: KVM instance
2576 * @gfn: starting GFN to be populated
2577 * @src: userspace-provided buffer containing data to copy into GFN range
2578 * (passed to @post_populate, and incremented on each iteration
2579 * if not NULL)
2580 * @npages: number of pages to copy from userspace-buffer
2581 * @post_populate: callback to issue for each gmem page that backs the GPA
2582 * range
2583 * @opaque: opaque data to pass to @post_populate callback
2584 *
2585 * This is primarily intended for cases where a gmem-backed GPA range needs
2586 * to be initialized with userspace-provided data prior to being mapped into
2587 * the guest as a private page. This should be called with the slots->lock
2588 * held so that caller-enforced invariants regarding the expected memory
2589 * attributes of the GPA range do not race with KVM_SET_MEMORY_ATTRIBUTES.
2590 *
2591 * Returns the number of pages that were populated.
2592 */
2593 typedef int (*kvm_gmem_populate_cb)(struct kvm *kvm, gfn_t gfn, kvm_pfn_t pfn,
2594 void __user *src, int order, void *opaque);
2595
2596 long kvm_gmem_populate(struct kvm *kvm, gfn_t gfn, void __user *src, long npages,
2597 kvm_gmem_populate_cb post_populate, void *opaque);
2598 #endif
2599
2600 #ifdef CONFIG_HAVE_KVM_ARCH_GMEM_INVALIDATE
2601 void kvm_arch_gmem_invalidate(kvm_pfn_t start, kvm_pfn_t end);
2602 #endif
2603
2604 #ifdef CONFIG_KVM_GENERIC_PRE_FAULT_MEMORY
2605 long kvm_arch_vcpu_pre_fault_memory(struct kvm_vcpu *vcpu,
2606 struct kvm_pre_fault_memory *range);
2607 #endif
2608
2609 #ifdef CONFIG_KVM_GENERIC_HARDWARE_ENABLING
2610 int kvm_enable_virtualization(void);
2611 void kvm_disable_virtualization(void);
2612 #else
kvm_enable_virtualization(void)2613 static inline int kvm_enable_virtualization(void) { return 0; }
kvm_disable_virtualization(void)2614 static inline void kvm_disable_virtualization(void) { }
2615 #endif
2616
2617 #endif
2618