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