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 u64 val;
322 unsigned int len;
323};
324
325struct kvm_vcpu {
326 struct kvm *kvm;
327#ifdef CONFIG_PREEMPT_NOTIFIERS
328 struct preempt_notifier preempt_notifier;
329#endif
330 int cpu;
331 int vcpu_id; /* id given by userspace at creation */
332 int vcpu_idx; /* index into kvm->vcpu_array */
333 int ____srcu_idx; /* Don't use this directly. You've been warned. */
334#ifdef CONFIG_PROVE_RCU
335 int srcu_depth;
336#endif
337 int mode;
338 u64 requests;
339 unsigned long guest_debug;
340
341 struct mutex mutex;
342 struct kvm_run *run;
343
344#ifndef __KVM_HAVE_ARCH_WQP
345 struct rcuwait wait;
346#endif
347 struct pid *pid;
348 rwlock_t pid_lock;
349 int sigset_active;
350 sigset_t sigset;
351 unsigned int halt_poll_ns;
352 bool valid_wakeup;
353
354#ifdef CONFIG_HAS_IOMEM
355 int mmio_needed;
356 int mmio_read_completed;
357 int mmio_is_write;
358 int mmio_cur_fragment;
359 int mmio_nr_fragments;
360 struct kvm_mmio_fragment mmio_fragments[KVM_MAX_MMIO_FRAGMENTS];
361#endif
362
363#ifdef CONFIG_KVM_ASYNC_PF
364 struct {
365 u32 queued;
366 struct list_head queue;
367 struct list_head done;
368 spinlock_t lock;
369 } async_pf;
370#endif
371
372#ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
373 /*
374 * Cpu relax intercept or pause loop exit optimization
375 * in_spin_loop: set when a vcpu does a pause loop exit
376 * or cpu relax intercepted.
377 * dy_eligible: indicates whether vcpu is eligible for directed yield.
378 */
379 struct {
380 bool in_spin_loop;
381 bool dy_eligible;
382 } spin_loop;
383#endif
384 bool wants_to_run;
385 bool preempted;
386 bool ready;
387 bool scheduled_out;
388 struct kvm_vcpu_arch arch;
389 struct kvm_vcpu_stat stat;
390 char stats_id[KVM_STATS_NAME_SIZE];
391 struct kvm_dirty_ring dirty_ring;
392
393 /*
394 * The most recently used memslot by this vCPU and the slots generation
395 * for which it is valid.
396 * No wraparound protection is needed since generations won't overflow in
397 * thousands of years, even assuming 1M memslot operations per second.
398 */
399 struct kvm_memory_slot *last_used_slot;
400 u64 last_used_slot_gen;
401};
402
403/*
404 * Start accounting time towards a guest.
405 * Must be called before entering guest context.
406 */
407static __always_inline void guest_timing_enter_irqoff(void)
408{
409 /*
410 * This is running in ioctl context so its safe to assume that it's the
411 * stime pending cputime to flush.
412 */
413 instrumentation_begin();
414 vtime_account_guest_enter();
415 instrumentation_end();
416}
417
418/*
419 * Enter guest context and enter an RCU extended quiescent state.
420 *
421 * Between guest_context_enter_irqoff() and guest_context_exit_irqoff() it is
422 * unsafe to use any code which may directly or indirectly use RCU, tracing
423 * (including IRQ flag tracing), or lockdep. All code in this period must be
424 * non-instrumentable.
425 */
426static __always_inline void guest_context_enter_irqoff(void)
427{
428 /*
429 * KVM does not hold any references to rcu protected data when it
430 * switches CPU into a guest mode. In fact switching to a guest mode
431 * is very similar to exiting to userspace from rcu point of view. In
432 * addition CPU may stay in a guest mode for quite a long time (up to
433 * one time slice). Lets treat guest mode as quiescent state, just like
434 * we do with user-mode execution.
435 */
436 if (!context_tracking_guest_enter()) {
437 instrumentation_begin();
438 rcu_virt_note_context_switch();
439 instrumentation_end();
440 }
441}
442
443/*
444 * Deprecated. Architectures should move to guest_timing_enter_irqoff() and
445 * guest_state_enter_irqoff().
446 */
447static __always_inline void guest_enter_irqoff(void)
448{
449 guest_timing_enter_irqoff();
450 guest_context_enter_irqoff();
451}
452
453/**
454 * guest_state_enter_irqoff - Fixup state when entering a guest
455 *
456 * Entry to a guest will enable interrupts, but the kernel state is interrupts
457 * disabled when this is invoked. Also tell RCU about it.
458 *
459 * 1) Trace interrupts on state
460 * 2) Invoke context tracking if enabled to adjust RCU state
461 * 3) Tell lockdep that interrupts are enabled
462 *
463 * Invoked from architecture specific code before entering a guest.
464 * Must be called with interrupts disabled and the caller must be
465 * non-instrumentable.
466 * The caller has to invoke guest_timing_enter_irqoff() before this.
467 *
468 * Note: this is analogous to exit_to_user_mode().
469 */
470static __always_inline void guest_state_enter_irqoff(void)
471{
472 instrumentation_begin();
473 trace_hardirqs_on_prepare();
474 lockdep_hardirqs_on_prepare();
475 instrumentation_end();
476
477 guest_context_enter_irqoff();
478 lockdep_hardirqs_on(CALLER_ADDR0);
479}
480
481/*
482 * Exit guest context and exit an RCU extended quiescent state.
483 *
484 * Between guest_context_enter_irqoff() and guest_context_exit_irqoff() it is
485 * unsafe to use any code which may directly or indirectly use RCU, tracing
486 * (including IRQ flag tracing), or lockdep. All code in this period must be
487 * non-instrumentable.
488 */
489static __always_inline void guest_context_exit_irqoff(void)
490{
491 /*
492 * Guest mode is treated as a quiescent state, see
493 * guest_context_enter_irqoff() for more details.
494 */
495 if (!context_tracking_guest_exit()) {
496 instrumentation_begin();
497 rcu_virt_note_context_switch();
498 instrumentation_end();
499 }
500}
501
502/*
503 * Stop accounting time towards a guest.
504 * Must be called after exiting guest context.
505 */
506static __always_inline void guest_timing_exit_irqoff(void)
507{
508 instrumentation_begin();
509 /* Flush the guest cputime we spent on the guest */
510 vtime_account_guest_exit();
511 instrumentation_end();
512}
513
514/*
515 * Deprecated. Architectures should move to guest_state_exit_irqoff() and
516 * guest_timing_exit_irqoff().
517 */
518static __always_inline void guest_exit_irqoff(void)
519{
520 guest_context_exit_irqoff();
521 guest_timing_exit_irqoff();
522}
523
524static inline void guest_exit(void)
525{
526 unsigned long flags;
527
528 local_irq_save(flags);
529 guest_exit_irqoff();
530 local_irq_restore(flags);
531}
532
533/**
534 * guest_state_exit_irqoff - Establish state when returning from guest mode
535 *
536 * Entry from a guest disables interrupts, but guest mode is traced as
537 * interrupts enabled. Also with NO_HZ_FULL RCU might be idle.
538 *
539 * 1) Tell lockdep that interrupts are disabled
540 * 2) Invoke context tracking if enabled to reactivate RCU
541 * 3) Trace interrupts off state
542 *
543 * Invoked from architecture specific code after exiting a guest.
544 * Must be invoked with interrupts disabled and the caller must be
545 * non-instrumentable.
546 * The caller has to invoke guest_timing_exit_irqoff() after this.
547 *
548 * Note: this is analogous to enter_from_user_mode().
549 */
550static __always_inline void guest_state_exit_irqoff(void)
551{
552 lockdep_hardirqs_off(CALLER_ADDR0);
553 guest_context_exit_irqoff();
554
555 instrumentation_begin();
556 trace_hardirqs_off_finish();
557 instrumentation_end();
558}
559
560static inline int kvm_vcpu_exiting_guest_mode(struct kvm_vcpu *vcpu)
561{
562 /*
563 * The memory barrier ensures a previous write to vcpu->requests cannot
564 * be reordered with the read of vcpu->mode. It pairs with the general
565 * memory barrier following the write of vcpu->mode in VCPU RUN.
566 */
567 smp_mb__before_atomic();
568 return cmpxchg(&vcpu->mode, IN_GUEST_MODE, EXITING_GUEST_MODE);
569}
570
571/*
572 * Some of the bitops functions do not support too long bitmaps.
573 * This number must be determined not to exceed such limits.
574 */
575#define KVM_MEM_MAX_NR_PAGES ((1UL << 31) - 1)
576
577/*
578 * Since at idle each memslot belongs to two memslot sets it has to contain
579 * two embedded nodes for each data structure that it forms a part of.
580 *
581 * Two memslot sets (one active and one inactive) are necessary so the VM
582 * continues to run on one memslot set while the other is being modified.
583 *
584 * These two memslot sets normally point to the same set of memslots.
585 * They can, however, be desynchronized when performing a memslot management
586 * operation by replacing the memslot to be modified by its copy.
587 * After the operation is complete, both memslot sets once again point to
588 * the same, common set of memslot data.
589 *
590 * The memslots themselves are independent of each other so they can be
591 * individually added or deleted.
592 */
593struct kvm_memory_slot {
594 struct hlist_node id_node[2];
595 struct interval_tree_node hva_node[2];
596 struct rb_node gfn_node[2];
597 gfn_t base_gfn;
598 unsigned long npages;
599 unsigned long *dirty_bitmap;
600 struct kvm_arch_memory_slot arch;
601 unsigned long userspace_addr;
602 u32 flags;
603 short id;
604 u16 as_id;
605
606#ifdef CONFIG_KVM_GUEST_MEMFD
607 struct {
608 /*
609 * Writes protected by kvm->slots_lock. Acquiring a
610 * reference via kvm_gmem_get_file() is protected by
611 * either kvm->slots_lock or kvm->srcu.
612 */
613 struct file *file;
614 pgoff_t pgoff;
615 } gmem;
616#endif
617};
618
619static inline bool kvm_slot_has_gmem(const struct kvm_memory_slot *slot)
620{
621 return slot && (slot->flags & KVM_MEM_GUEST_MEMFD);
622}
623
624static inline bool kvm_slot_dirty_track_enabled(const struct kvm_memory_slot *slot)
625{
626 return slot->flags & KVM_MEM_LOG_DIRTY_PAGES;
627}
628
629static inline unsigned long kvm_dirty_bitmap_bytes(struct kvm_memory_slot *memslot)
630{
631 return ALIGN(memslot->npages, BITS_PER_LONG) / 8;
632}
633
634static inline unsigned long *kvm_second_dirty_bitmap(struct kvm_memory_slot *memslot)
635{
636 unsigned long len = kvm_dirty_bitmap_bytes(memslot);
637
638 return memslot->dirty_bitmap + len / sizeof(*memslot->dirty_bitmap);
639}
640
641#ifndef KVM_DIRTY_LOG_MANUAL_CAPS
642#define KVM_DIRTY_LOG_MANUAL_CAPS KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE
643#endif
644
645struct kvm_s390_adapter_int {
646 u64 ind_addr;
647 u64 ind_gaddr;
648 u64 summary_addr;
649 u64 summary_gaddr;
650 u64 ind_offset;
651 u32 summary_offset;
652 u32 adapter_id;
653};
654
655struct kvm_hv_sint {
656 u32 vcpu;
657 u32 sint;
658};
659
660struct kvm_xen_evtchn {
661 u32 port;
662 u32 vcpu_id;
663 int vcpu_idx;
664 u32 priority;
665};
666
667struct kvm_kernel_irq_routing_entry {
668 u32 gsi;
669 u32 type;
670 int (*set)(struct kvm_kernel_irq_routing_entry *e,
671 struct kvm *kvm, int irq_source_id, int level,
672 bool line_status);
673 union {
674 struct {
675 unsigned irqchip;
676 unsigned pin;
677 } irqchip;
678 struct {
679 u32 address_lo;
680 u32 address_hi;
681 u32 data;
682 u32 flags;
683 u32 devid;
684 } msi;
685 struct kvm_s390_adapter_int adapter;
686 struct kvm_hv_sint hv_sint;
687 struct kvm_xen_evtchn xen_evtchn;
688 };
689 struct hlist_node link;
690};
691
692#ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
693struct kvm_irq_routing_table {
694 int chip[KVM_NR_IRQCHIPS][KVM_IRQCHIP_NUM_PINS];
695 u32 nr_rt_entries;
696 /*
697 * Array indexed by gsi. Each entry contains list of irq chips
698 * the gsi is connected to.
699 */
700 struct hlist_head map[] __counted_by(nr_rt_entries);
701};
702#endif
703
704bool kvm_arch_irqchip_in_kernel(struct kvm *kvm);
705
706#ifndef KVM_INTERNAL_MEM_SLOTS
707#define KVM_INTERNAL_MEM_SLOTS 0
708#endif
709
710#define KVM_MEM_SLOTS_NUM SHRT_MAX
711#define KVM_USER_MEM_SLOTS (KVM_MEM_SLOTS_NUM - KVM_INTERNAL_MEM_SLOTS)
712
713#if KVM_MAX_NR_ADDRESS_SPACES == 1
714static inline int kvm_arch_nr_memslot_as_ids(struct kvm *kvm)
715{
716 return KVM_MAX_NR_ADDRESS_SPACES;
717}
718
719static inline int kvm_arch_vcpu_memslots_id(struct kvm_vcpu *vcpu)
720{
721 return 0;
722}
723#endif
724
725#ifndef CONFIG_KVM_GENERIC_MEMORY_ATTRIBUTES
726static inline bool kvm_arch_has_private_mem(struct kvm *kvm)
727{
728 return false;
729}
730#endif
731
732#ifdef CONFIG_KVM_GUEST_MEMFD
733bool kvm_arch_supports_gmem_init_shared(struct kvm *kvm);
734
735static inline u64 kvm_gmem_get_supported_flags(struct kvm *kvm)
736{
737 u64 flags = GUEST_MEMFD_FLAG_MMAP;
738
739 if (!kvm || kvm_arch_supports_gmem_init_shared(kvm))
740 flags |= GUEST_MEMFD_FLAG_INIT_SHARED;
741
742 return flags;
743}
744#endif
745
746#ifndef kvm_arch_has_readonly_mem
747static inline bool kvm_arch_has_readonly_mem(struct kvm *kvm)
748{
749 return IS_ENABLED(CONFIG_HAVE_KVM_READONLY_MEM);
750}
751#endif
752
753struct kvm_memslots {
754 u64 generation;
755 atomic_long_t last_used_slot;
756 struct rb_root_cached hva_tree;
757 struct rb_root gfn_tree;
758 /*
759 * The mapping table from slot id to memslot.
760 *
761 * 7-bit bucket count matches the size of the old id to index array for
762 * 512 slots, while giving good performance with this slot count.
763 * Higher bucket counts bring only small performance improvements but
764 * always result in higher memory usage (even for lower memslot counts).
765 */
766 DECLARE_HASHTABLE(id_hash, 7);
767 int node_idx;
768};
769
770struct kvm {
771#ifdef KVM_HAVE_MMU_RWLOCK
772 rwlock_t mmu_lock;
773#else
774 spinlock_t mmu_lock;
775#endif /* KVM_HAVE_MMU_RWLOCK */
776
777 struct mutex slots_lock;
778
779 /*
780 * Protects the arch-specific fields of struct kvm_memory_slots in
781 * use by the VM. To be used under the slots_lock (above) or in a
782 * kvm->srcu critical section where acquiring the slots_lock would
783 * lead to deadlock with the synchronize_srcu in
784 * kvm_swap_active_memslots().
785 */
786 struct mutex slots_arch_lock;
787 struct mm_struct *mm; /* userspace tied to this vm */
788 unsigned long nr_memslot_pages;
789 /* The two memslot sets - active and inactive (per address space) */
790 struct kvm_memslots __memslots[KVM_MAX_NR_ADDRESS_SPACES][2];
791 /* The current active memslot set for each address space */
792 struct kvm_memslots __rcu *memslots[KVM_MAX_NR_ADDRESS_SPACES];
793 struct xarray vcpu_array;
794 /*
795 * Protected by slots_lock, but can be read outside if an
796 * incorrect answer is acceptable.
797 */
798 atomic_t nr_memslots_dirty_logging;
799
800 /* Used to wait for completion of MMU notifiers. */
801 spinlock_t mn_invalidate_lock;
802 unsigned long mn_active_invalidate_count;
803 struct rcuwait mn_memslots_update_rcuwait;
804
805 /* For management / invalidation of gfn_to_pfn_caches */
806 spinlock_t gpc_lock;
807 struct list_head gpc_list;
808
809 /*
810 * created_vcpus is protected by kvm->lock, and is incremented
811 * at the beginning of KVM_CREATE_VCPU. online_vcpus is only
812 * incremented after storing the kvm_vcpu pointer in vcpus,
813 * and is accessed atomically.
814 */
815 atomic_t online_vcpus;
816 int max_vcpus;
817 int created_vcpus;
818 int last_boosted_vcpu;
819 struct list_head vm_list;
820 struct mutex lock;
821 struct kvm_io_bus __rcu *buses[KVM_NR_BUSES];
822#ifdef CONFIG_HAVE_KVM_IRQCHIP
823 struct {
824 spinlock_t lock;
825 struct list_head items;
826 /* resampler_list update side is protected by resampler_lock. */
827 struct list_head resampler_list;
828 struct mutex resampler_lock;
829 } irqfds;
830#endif
831 struct list_head ioeventfds;
832 struct kvm_vm_stat stat;
833 struct kvm_arch arch;
834 refcount_t users_count;
835#ifdef CONFIG_KVM_MMIO
836 struct kvm_coalesced_mmio_ring *coalesced_mmio_ring;
837 spinlock_t ring_lock;
838 struct list_head coalesced_zones;
839#endif
840
841 struct mutex irq_lock;
842#ifdef CONFIG_HAVE_KVM_IRQCHIP
843 /*
844 * Update side is protected by irq_lock.
845 */
846 struct kvm_irq_routing_table __rcu *irq_routing;
847
848 struct hlist_head irq_ack_notifier_list;
849#endif
850
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
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
907static 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
913static 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
957static 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
966static 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
976static 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 */
986static 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
992static 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
1016static 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
1033static inline bool kvm_is_vcpu_creation_in_progress(struct kvm *kvm)
1034{
1035 lockdep_assert_held(&kvm->lock);
1036
1037 return kvm->created_vcpus != atomic_read(&kvm->online_vcpus);
1038}
1039
1040void kvm_destroy_vcpus(struct kvm *kvm);
1041
1042int kvm_trylock_all_vcpus(struct kvm *kvm);
1043int kvm_lock_all_vcpus(struct kvm *kvm);
1044void kvm_unlock_all_vcpus(struct kvm *kvm);
1045
1046void vcpu_load(struct kvm_vcpu *vcpu);
1047void vcpu_put(struct kvm_vcpu *vcpu);
1048
1049#ifdef CONFIG_KVM_IOAPIC
1050void kvm_arch_post_irq_ack_notifier_list_update(struct kvm *kvm);
1051#else
1052static inline void kvm_arch_post_irq_ack_notifier_list_update(struct kvm *kvm)
1053{
1054}
1055#endif
1056
1057#ifdef CONFIG_HAVE_KVM_IRQCHIP
1058int kvm_irqfd_init(void);
1059void kvm_irqfd_exit(void);
1060#else
1061static inline int kvm_irqfd_init(void)
1062{
1063 return 0;
1064}
1065
1066static inline void kvm_irqfd_exit(void)
1067{
1068}
1069#endif
1070int kvm_init(unsigned vcpu_size, unsigned vcpu_align, struct module *module);
1071void kvm_exit(void);
1072
1073void kvm_get_kvm(struct kvm *kvm);
1074bool kvm_get_kvm_safe(struct kvm *kvm);
1075void kvm_put_kvm(struct kvm *kvm);
1076bool file_is_kvm(struct file *file);
1077void kvm_put_kvm_no_destroy(struct kvm *kvm);
1078
1079static inline struct kvm_memslots *__kvm_memslots(struct kvm *kvm, int as_id)
1080{
1081 as_id = array_index_nospec(as_id, KVM_MAX_NR_ADDRESS_SPACES);
1082 return srcu_dereference_check(kvm->memslots[as_id], &kvm->srcu,
1083 lockdep_is_held(&kvm->slots_lock) ||
1084 !refcount_read(&kvm->users_count));
1085}
1086
1087static inline struct kvm_memslots *kvm_memslots(struct kvm *kvm)
1088{
1089 return __kvm_memslots(kvm, 0);
1090}
1091
1092static inline struct kvm_memslots *kvm_vcpu_memslots(struct kvm_vcpu *vcpu)
1093{
1094 int as_id = kvm_arch_vcpu_memslots_id(vcpu);
1095
1096 return __kvm_memslots(vcpu->kvm, as_id);
1097}
1098
1099static inline bool kvm_memslots_empty(struct kvm_memslots *slots)
1100{
1101 return RB_EMPTY_ROOT(&slots->gfn_tree);
1102}
1103
1104bool kvm_are_all_memslots_empty(struct kvm *kvm);
1105
1106#define kvm_for_each_memslot(memslot, bkt, slots) \
1107 hash_for_each(slots->id_hash, bkt, memslot, id_node[slots->node_idx]) \
1108 if (WARN_ON_ONCE(!memslot->npages)) { \
1109 } else
1110
1111static inline
1112struct kvm_memory_slot *id_to_memslot(struct kvm_memslots *slots, int id)
1113{
1114 struct kvm_memory_slot *slot;
1115 int idx = slots->node_idx;
1116
1117 hash_for_each_possible(slots->id_hash, slot, id_node[idx], id) {
1118 if (slot->id == id)
1119 return slot;
1120 }
1121
1122 return NULL;
1123}
1124
1125/* Iterator used for walking memslots that overlap a gfn range. */
1126struct kvm_memslot_iter {
1127 struct kvm_memslots *slots;
1128 struct rb_node *node;
1129 struct kvm_memory_slot *slot;
1130};
1131
1132static inline void kvm_memslot_iter_next(struct kvm_memslot_iter *iter)
1133{
1134 iter->node = rb_next(iter->node);
1135 if (!iter->node)
1136 return;
1137
1138 iter->slot = container_of(iter->node, struct kvm_memory_slot, gfn_node[iter->slots->node_idx]);
1139}
1140
1141static inline void kvm_memslot_iter_start(struct kvm_memslot_iter *iter,
1142 struct kvm_memslots *slots,
1143 gfn_t start)
1144{
1145 int idx = slots->node_idx;
1146 struct rb_node *tmp;
1147 struct kvm_memory_slot *slot;
1148
1149 iter->slots = slots;
1150
1151 /*
1152 * Find the so called "upper bound" of a key - the first node that has
1153 * its key strictly greater than the searched one (the start gfn in our case).
1154 */
1155 iter->node = NULL;
1156 for (tmp = slots->gfn_tree.rb_node; tmp; ) {
1157 slot = container_of(tmp, struct kvm_memory_slot, gfn_node[idx]);
1158 if (start < slot->base_gfn) {
1159 iter->node = tmp;
1160 tmp = tmp->rb_left;
1161 } else {
1162 tmp = tmp->rb_right;
1163 }
1164 }
1165
1166 /*
1167 * Find the slot with the lowest gfn that can possibly intersect with
1168 * the range, so we'll ideally have slot start <= range start
1169 */
1170 if (iter->node) {
1171 /*
1172 * A NULL previous node means that the very first slot
1173 * already has a higher start gfn.
1174 * In this case slot start > range start.
1175 */
1176 tmp = rb_prev(iter->node);
1177 if (tmp)
1178 iter->node = tmp;
1179 } else {
1180 /* a NULL node below means no slots */
1181 iter->node = rb_last(&slots->gfn_tree);
1182 }
1183
1184 if (iter->node) {
1185 iter->slot = container_of(iter->node, struct kvm_memory_slot, gfn_node[idx]);
1186
1187 /*
1188 * It is possible in the slot start < range start case that the
1189 * found slot ends before or at range start (slot end <= range start)
1190 * and so it does not overlap the requested range.
1191 *
1192 * In such non-overlapping case the next slot (if it exists) will
1193 * already have slot start > range start, otherwise the logic above
1194 * would have found it instead of the current slot.
1195 */
1196 if (iter->slot->base_gfn + iter->slot->npages <= start)
1197 kvm_memslot_iter_next(iter);
1198 }
1199}
1200
1201static inline bool kvm_memslot_iter_is_valid(struct kvm_memslot_iter *iter, gfn_t end)
1202{
1203 if (!iter->node)
1204 return false;
1205
1206 /*
1207 * If this slot starts beyond or at the end of the range so does
1208 * every next one
1209 */
1210 return iter->slot->base_gfn < end;
1211}
1212
1213/* Iterate over each memslot at least partially intersecting [start, end) range */
1214#define kvm_for_each_memslot_in_gfn_range(iter, slots, start, end) \
1215 for (kvm_memslot_iter_start(iter, slots, start); \
1216 kvm_memslot_iter_is_valid(iter, end); \
1217 kvm_memslot_iter_next(iter))
1218
1219struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn);
1220struct kvm_memslots *kvm_vcpu_memslots(struct kvm_vcpu *vcpu);
1221struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn);
1222
1223/*
1224 * KVM_SET_USER_MEMORY_REGION ioctl allows the following operations:
1225 * - create a new memory slot
1226 * - delete an existing memory slot
1227 * - modify an existing memory slot
1228 * -- move it in the guest physical memory space
1229 * -- just change its flags
1230 *
1231 * Since flags can be changed by some of these operations, the following
1232 * differentiation is the best we can do for kvm_set_memory_region():
1233 */
1234enum kvm_mr_change {
1235 KVM_MR_CREATE,
1236 KVM_MR_DELETE,
1237 KVM_MR_MOVE,
1238 KVM_MR_FLAGS_ONLY,
1239};
1240
1241int kvm_set_internal_memslot(struct kvm *kvm,
1242 const struct kvm_userspace_memory_region2 *mem);
1243void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot);
1244void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen);
1245int kvm_arch_prepare_memory_region(struct kvm *kvm,
1246 const struct kvm_memory_slot *old,
1247 struct kvm_memory_slot *new,
1248 enum kvm_mr_change change);
1249void kvm_arch_commit_memory_region(struct kvm *kvm,
1250 struct kvm_memory_slot *old,
1251 const struct kvm_memory_slot *new,
1252 enum kvm_mr_change change);
1253/* flush all memory translations */
1254void kvm_arch_flush_shadow_all(struct kvm *kvm);
1255/* flush memory translations pointing to 'slot' */
1256void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
1257 struct kvm_memory_slot *slot);
1258
1259int kvm_prefetch_pages(struct kvm_memory_slot *slot, gfn_t gfn,
1260 struct page **pages, int nr_pages);
1261
1262struct page *__gfn_to_page(struct kvm *kvm, gfn_t gfn, bool write);
1263static inline struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1264{
1265 return __gfn_to_page(kvm, gfn, true);
1266}
1267
1268unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn);
1269unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable);
1270unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot, gfn_t gfn);
1271unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot, gfn_t gfn,
1272 bool *writable);
1273
1274static inline void kvm_release_page_unused(struct page *page)
1275{
1276 if (!page)
1277 return;
1278
1279 put_page(page);
1280}
1281
1282void kvm_release_page_clean(struct page *page);
1283void kvm_release_page_dirty(struct page *page);
1284
1285static inline void kvm_release_faultin_page(struct kvm *kvm, struct page *page,
1286 bool unused, bool dirty)
1287{
1288 lockdep_assert_once(lockdep_is_held(&kvm->mmu_lock) || unused);
1289
1290 if (!page)
1291 return;
1292
1293 /*
1294 * If the page that KVM got from the *primary MMU* is writable, and KVM
1295 * installed or reused a SPTE, mark the page/folio dirty. Note, this
1296 * may mark a folio dirty even if KVM created a read-only SPTE, e.g. if
1297 * the GFN is write-protected. Folios can't be safely marked dirty
1298 * outside of mmu_lock as doing so could race with writeback on the
1299 * folio. As a result, KVM can't mark folios dirty in the fast page
1300 * fault handler, and so KVM must (somewhat) speculatively mark the
1301 * folio dirty if KVM could locklessly make the SPTE writable.
1302 */
1303 if (unused)
1304 kvm_release_page_unused(page);
1305 else if (dirty)
1306 kvm_release_page_dirty(page);
1307 else
1308 kvm_release_page_clean(page);
1309}
1310
1311kvm_pfn_t __kvm_faultin_pfn(const struct kvm_memory_slot *slot, gfn_t gfn,
1312 unsigned int foll, bool *writable,
1313 struct page **refcounted_page);
1314
1315static inline kvm_pfn_t kvm_faultin_pfn(struct kvm_vcpu *vcpu, gfn_t gfn,
1316 bool write, bool *writable,
1317 struct page **refcounted_page)
1318{
1319 return __kvm_faultin_pfn(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn,
1320 write ? FOLL_WRITE : 0, writable, refcounted_page);
1321}
1322
1323int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1324 int len);
1325int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len);
1326int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1327 void *data, unsigned long len);
1328int kvm_read_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1329 void *data, unsigned int offset,
1330 unsigned long len);
1331int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1332 int offset, int len);
1333int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1334 unsigned long len);
1335int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1336 void *data, unsigned long len);
1337int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1338 void *data, unsigned int offset,
1339 unsigned long len);
1340int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1341 gpa_t gpa, unsigned long len);
1342
1343#define __kvm_get_guest(kvm, gfn, offset, v) \
1344({ \
1345 unsigned long __addr = gfn_to_hva(kvm, gfn); \
1346 typeof(v) __user *__uaddr = (typeof(__uaddr))(__addr + offset); \
1347 int __ret = -EFAULT; \
1348 \
1349 if (!kvm_is_error_hva(__addr)) \
1350 __ret = get_user(v, __uaddr); \
1351 __ret; \
1352})
1353
1354#define kvm_get_guest(kvm, gpa, v) \
1355({ \
1356 gpa_t __gpa = gpa; \
1357 struct kvm *__kvm = kvm; \
1358 \
1359 __kvm_get_guest(__kvm, __gpa >> PAGE_SHIFT, \
1360 offset_in_page(__gpa), v); \
1361})
1362
1363#define __kvm_put_guest(kvm, gfn, offset, v) \
1364({ \
1365 unsigned long __addr = gfn_to_hva(kvm, gfn); \
1366 typeof(v) __user *__uaddr = (typeof(__uaddr))(__addr + offset); \
1367 int __ret = -EFAULT; \
1368 \
1369 if (!kvm_is_error_hva(__addr)) \
1370 __ret = put_user(v, __uaddr); \
1371 if (!__ret) \
1372 mark_page_dirty(kvm, gfn); \
1373 __ret; \
1374})
1375
1376#define kvm_put_guest(kvm, gpa, v) \
1377({ \
1378 gpa_t __gpa = gpa; \
1379 struct kvm *__kvm = kvm; \
1380 \
1381 __kvm_put_guest(__kvm, __gpa >> PAGE_SHIFT, \
1382 offset_in_page(__gpa), v); \
1383})
1384
1385int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len);
1386bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn);
1387bool kvm_vcpu_is_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn);
1388unsigned long kvm_host_page_size(struct kvm_vcpu *vcpu, gfn_t gfn);
1389void mark_page_dirty_in_slot(struct kvm *kvm, const struct kvm_memory_slot *memslot, gfn_t gfn);
1390void mark_page_dirty(struct kvm *kvm, gfn_t gfn);
1391void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn);
1392
1393int __kvm_vcpu_map(struct kvm_vcpu *vcpu, gpa_t gpa, struct kvm_host_map *map,
1394 bool writable);
1395void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map);
1396
1397static inline int kvm_vcpu_map(struct kvm_vcpu *vcpu, gpa_t gpa,
1398 struct kvm_host_map *map)
1399{
1400 return __kvm_vcpu_map(vcpu, gpa, map, true);
1401}
1402
1403static inline int kvm_vcpu_map_readonly(struct kvm_vcpu *vcpu, gpa_t gpa,
1404 struct kvm_host_map *map)
1405{
1406 return __kvm_vcpu_map(vcpu, gpa, map, false);
1407}
1408
1409static inline void kvm_vcpu_map_mark_dirty(struct kvm_vcpu *vcpu,
1410 struct kvm_host_map *map)
1411{
1412 if (kvm_vcpu_mapped(map))
1413 kvm_vcpu_mark_page_dirty(vcpu, map->gfn);
1414}
1415
1416unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn);
1417unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable);
1418int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data, int offset,
1419 int len);
1420int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa, void *data,
1421 unsigned long len);
1422int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data,
1423 unsigned long len);
1424int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, const void *data,
1425 int offset, int len);
1426int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
1427 unsigned long len);
1428
1429/**
1430 * kvm_gpc_init - initialize gfn_to_pfn_cache.
1431 *
1432 * @gpc: struct gfn_to_pfn_cache object.
1433 * @kvm: pointer to kvm instance.
1434 *
1435 * This sets up a gfn_to_pfn_cache by initializing locks and assigning the
1436 * immutable attributes. Note, the cache must be zero-allocated (or zeroed by
1437 * the caller before init).
1438 */
1439void kvm_gpc_init(struct gfn_to_pfn_cache *gpc, struct kvm *kvm);
1440
1441/**
1442 * kvm_gpc_activate - prepare a cached kernel mapping and HPA for a given guest
1443 * physical address.
1444 *
1445 * @gpc: struct gfn_to_pfn_cache object.
1446 * @gpa: guest physical address to map.
1447 * @len: sanity check; the range being access must fit a single page.
1448 *
1449 * @return: 0 for success.
1450 * -EINVAL for a mapping which would cross a page boundary.
1451 * -EFAULT for an untranslatable guest physical address.
1452 *
1453 * This primes a gfn_to_pfn_cache and links it into the @gpc->kvm's list for
1454 * invalidations to be processed. Callers are required to use kvm_gpc_check()
1455 * to ensure that the cache is valid before accessing the target page.
1456 */
1457int kvm_gpc_activate(struct gfn_to_pfn_cache *gpc, gpa_t gpa, unsigned long len);
1458
1459/**
1460 * kvm_gpc_activate_hva - prepare a cached kernel mapping and HPA for a given HVA.
1461 *
1462 * @gpc: struct gfn_to_pfn_cache object.
1463 * @hva: userspace virtual address to map.
1464 * @len: sanity check; the range being access must fit a single page.
1465 *
1466 * @return: 0 for success.
1467 * -EINVAL for a mapping which would cross a page boundary.
1468 * -EFAULT for an untranslatable guest physical address.
1469 *
1470 * The semantics of this function are the same as those of kvm_gpc_activate(). It
1471 * merely bypasses a layer of address translation.
1472 */
1473int kvm_gpc_activate_hva(struct gfn_to_pfn_cache *gpc, unsigned long hva, unsigned long len);
1474
1475/**
1476 * kvm_gpc_check - check validity of a gfn_to_pfn_cache.
1477 *
1478 * @gpc: struct gfn_to_pfn_cache object.
1479 * @len: sanity check; the range being access must fit a single page.
1480 *
1481 * @return: %true if the cache is still valid and the address matches.
1482 * %false if the cache is not valid.
1483 *
1484 * Callers outside IN_GUEST_MODE context should hold a read lock on @gpc->lock
1485 * while calling this function, and then continue to hold the lock until the
1486 * access is complete.
1487 *
1488 * Callers in IN_GUEST_MODE may do so without locking, although they should
1489 * still hold a read lock on kvm->scru for the memslot checks.
1490 */
1491bool kvm_gpc_check(struct gfn_to_pfn_cache *gpc, unsigned long len);
1492
1493/**
1494 * kvm_gpc_refresh - update a previously initialized cache.
1495 *
1496 * @gpc: struct gfn_to_pfn_cache object.
1497 * @len: sanity check; the range being access must fit a single page.
1498 *
1499 * @return: 0 for success.
1500 * -EINVAL for a mapping which would cross a page boundary.
1501 * -EFAULT for an untranslatable guest physical address.
1502 *
1503 * This will attempt to refresh a gfn_to_pfn_cache. Note that a successful
1504 * return from this function does not mean the page can be immediately
1505 * accessed because it may have raced with an invalidation. Callers must
1506 * still lock and check the cache status, as this function does not return
1507 * with the lock still held to permit access.
1508 */
1509int kvm_gpc_refresh(struct gfn_to_pfn_cache *gpc, unsigned long len);
1510
1511/**
1512 * kvm_gpc_deactivate - deactivate and unlink a gfn_to_pfn_cache.
1513 *
1514 * @gpc: struct gfn_to_pfn_cache object.
1515 *
1516 * This removes a cache from the VM's list to be processed on MMU notifier
1517 * invocation.
1518 */
1519void kvm_gpc_deactivate(struct gfn_to_pfn_cache *gpc);
1520
1521static inline bool kvm_gpc_is_gpa_active(struct gfn_to_pfn_cache *gpc)
1522{
1523 return gpc->active && !kvm_is_error_gpa(gpc->gpa);
1524}
1525
1526static inline bool kvm_gpc_is_hva_active(struct gfn_to_pfn_cache *gpc)
1527{
1528 return gpc->active && kvm_is_error_gpa(gpc->gpa);
1529}
1530
1531void kvm_sigset_activate(struct kvm_vcpu *vcpu);
1532void kvm_sigset_deactivate(struct kvm_vcpu *vcpu);
1533
1534void kvm_vcpu_halt(struct kvm_vcpu *vcpu);
1535bool kvm_vcpu_block(struct kvm_vcpu *vcpu);
1536void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu);
1537void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu);
1538bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu);
1539
1540#ifndef CONFIG_S390
1541void __kvm_vcpu_kick(struct kvm_vcpu *vcpu, bool wait);
1542
1543static inline void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1544{
1545 __kvm_vcpu_kick(vcpu, false);
1546}
1547#endif
1548
1549int kvm_vcpu_yield_to(struct kvm_vcpu *target);
1550void kvm_vcpu_on_spin(struct kvm_vcpu *vcpu, bool yield_to_kernel_mode);
1551
1552void kvm_flush_remote_tlbs(struct kvm *kvm);
1553void kvm_flush_remote_tlbs_range(struct kvm *kvm, gfn_t gfn, u64 nr_pages);
1554void kvm_flush_remote_tlbs_memslot(struct kvm *kvm,
1555 const struct kvm_memory_slot *memslot);
1556
1557#ifdef KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE
1558int kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache *mc, int min);
1559int __kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache *mc, int capacity, int min);
1560int kvm_mmu_memory_cache_nr_free_objects(struct kvm_mmu_memory_cache *mc);
1561void kvm_mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc);
1562void *kvm_mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc);
1563#endif
1564
1565void kvm_mmu_invalidate_begin(struct kvm *kvm);
1566void kvm_mmu_invalidate_range_add(struct kvm *kvm, gfn_t start, gfn_t end);
1567void kvm_mmu_invalidate_end(struct kvm *kvm);
1568bool kvm_mmu_unmap_gfn_range(struct kvm *kvm, struct kvm_gfn_range *range);
1569
1570long kvm_arch_dev_ioctl(struct file *filp,
1571 unsigned int ioctl, unsigned long arg);
1572long kvm_arch_vcpu_ioctl(struct file *filp,
1573 unsigned int ioctl, unsigned long arg);
1574long kvm_arch_vcpu_unlocked_ioctl(struct file *filp,
1575 unsigned int ioctl, unsigned long arg);
1576vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf);
1577
1578int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext);
1579
1580void kvm_arch_mmu_enable_log_dirty_pt_masked(struct kvm *kvm,
1581 struct kvm_memory_slot *slot,
1582 gfn_t gfn_offset,
1583 unsigned long mask);
1584void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot);
1585
1586#ifndef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1587int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log);
1588int kvm_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log,
1589 int *is_dirty, struct kvm_memory_slot **memslot);
1590#endif
1591
1592int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
1593 bool line_status);
1594int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
1595 struct kvm_enable_cap *cap);
1596int kvm_arch_vm_ioctl(struct file *filp, unsigned int ioctl, unsigned long arg);
1597long kvm_arch_vm_compat_ioctl(struct file *filp, unsigned int ioctl,
1598 unsigned long arg);
1599
1600int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu);
1601int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu);
1602
1603int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
1604 struct kvm_translation *tr);
1605
1606int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs);
1607int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs);
1608int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
1609 struct kvm_sregs *sregs);
1610int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
1611 struct kvm_sregs *sregs);
1612int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
1613 struct kvm_mp_state *mp_state);
1614int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
1615 struct kvm_mp_state *mp_state);
1616int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
1617 struct kvm_guest_debug *dbg);
1618int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu);
1619
1620void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu);
1621void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu);
1622int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id);
1623int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu);
1624void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu);
1625void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu);
1626
1627#ifdef CONFIG_HAVE_KVM_PM_NOTIFIER
1628int kvm_arch_pm_notifier(struct kvm *kvm, unsigned long state);
1629#endif
1630
1631#ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS
1632void kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu, struct dentry *debugfs_dentry);
1633#else
1634static inline void kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu) {}
1635#endif
1636
1637#ifdef CONFIG_KVM_GENERIC_HARDWARE_ENABLING
1638/*
1639 * kvm_arch_shutdown() is invoked immediately prior to forcefully disabling
1640 * hardware virtualization on all CPUs via IPI function calls (in preparation
1641 * for shutdown or reboot), e.g. to allow arch code to prepare for disabling
1642 * virtualization while KVM may be actively running vCPUs.
1643 */
1644void kvm_arch_shutdown(void);
1645/*
1646 * kvm_arch_{enable,disable}_virtualization() are called on one CPU, under
1647 * kvm_usage_lock, immediately after/before 0=>1 and 1=>0 transitions of
1648 * kvm_usage_count, i.e. at the beginning of the generic hardware enabling
1649 * sequence, and at the end of the generic hardware disabling sequence.
1650 */
1651void kvm_arch_enable_virtualization(void);
1652void kvm_arch_disable_virtualization(void);
1653/*
1654 * kvm_arch_{enable,disable}_virtualization_cpu() are called on "every" CPU to
1655 * do the actual twiddling of hardware bits. The hooks are called on all
1656 * online CPUs when KVM enables/disabled virtualization, and on a single CPU
1657 * when that CPU is onlined/offlined (including for Resume/Suspend).
1658 */
1659int kvm_arch_enable_virtualization_cpu(void);
1660void kvm_arch_disable_virtualization_cpu(void);
1661#endif
1662bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu);
1663int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu);
1664bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu);
1665int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu);
1666bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu);
1667bool kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu);
1668bool kvm_arch_vcpu_preempted_in_kernel(struct kvm_vcpu *vcpu);
1669void kvm_arch_pre_destroy_vm(struct kvm *kvm);
1670void kvm_arch_create_vm_debugfs(struct kvm *kvm);
1671
1672#ifndef __KVM_HAVE_ARCH_VM_ALLOC
1673/*
1674 * All architectures that want to use vzalloc currently also
1675 * need their own kvm_arch_alloc_vm implementation.
1676 */
1677static inline struct kvm *kvm_arch_alloc_vm(void)
1678{
1679 return kzalloc_obj(struct kvm, GFP_KERNEL_ACCOUNT);
1680}
1681#endif
1682
1683static inline void __kvm_arch_free_vm(struct kvm *kvm)
1684{
1685 kvfree(kvm);
1686}
1687
1688#ifndef __KVM_HAVE_ARCH_VM_FREE
1689static inline void kvm_arch_free_vm(struct kvm *kvm)
1690{
1691 __kvm_arch_free_vm(kvm);
1692}
1693#endif
1694
1695#ifndef __KVM_HAVE_ARCH_FLUSH_REMOTE_TLBS
1696static inline int kvm_arch_flush_remote_tlbs(struct kvm *kvm)
1697{
1698 return -ENOTSUPP;
1699}
1700#else
1701int kvm_arch_flush_remote_tlbs(struct kvm *kvm);
1702#endif
1703
1704#ifndef __KVM_HAVE_ARCH_FLUSH_REMOTE_TLBS_RANGE
1705static inline int kvm_arch_flush_remote_tlbs_range(struct kvm *kvm,
1706 gfn_t gfn, u64 nr_pages)
1707{
1708 return -EOPNOTSUPP;
1709}
1710#else
1711int kvm_arch_flush_remote_tlbs_range(struct kvm *kvm, gfn_t gfn, u64 nr_pages);
1712#endif
1713
1714#ifdef __KVM_HAVE_ARCH_NONCOHERENT_DMA
1715void kvm_arch_register_noncoherent_dma(struct kvm *kvm);
1716void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm);
1717bool kvm_arch_has_noncoherent_dma(struct kvm *kvm);
1718#else
1719static inline void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
1720{
1721}
1722
1723static inline void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
1724{
1725}
1726
1727static inline bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
1728{
1729 return false;
1730}
1731#endif
1732
1733static inline struct rcuwait *kvm_arch_vcpu_get_wait(struct kvm_vcpu *vcpu)
1734{
1735#ifdef __KVM_HAVE_ARCH_WQP
1736 return vcpu->arch.waitp;
1737#else
1738 return &vcpu->wait;
1739#endif
1740}
1741
1742/*
1743 * Wake a vCPU if necessary, but don't do any stats/metadata updates. Returns
1744 * true if the vCPU was blocking and was awakened, false otherwise.
1745 */
1746static inline bool __kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
1747{
1748 return !!rcuwait_wake_up(kvm_arch_vcpu_get_wait(vcpu));
1749}
1750
1751static inline bool kvm_vcpu_is_blocking(struct kvm_vcpu *vcpu)
1752{
1753 return rcuwait_active(kvm_arch_vcpu_get_wait(vcpu));
1754}
1755
1756#ifdef __KVM_HAVE_ARCH_INTC_INITIALIZED
1757/*
1758 * returns true if the virtual interrupt controller is initialized and
1759 * ready to accept virtual IRQ. On some architectures the virtual interrupt
1760 * controller is dynamically instantiated and this is not always true.
1761 */
1762bool kvm_arch_intc_initialized(struct kvm *kvm);
1763#else
1764static inline bool kvm_arch_intc_initialized(struct kvm *kvm)
1765{
1766 return true;
1767}
1768#endif
1769
1770#ifdef CONFIG_GUEST_PERF_EVENTS
1771unsigned long kvm_arch_vcpu_get_ip(struct kvm_vcpu *vcpu);
1772
1773void __kvm_register_perf_callbacks(unsigned int (*pt_intr_handler)(void),
1774 void (*mediated_pmi_handler)(void));
1775
1776static inline void kvm_register_perf_callbacks(void)
1777{
1778 __kvm_register_perf_callbacks(NULL, NULL);
1779}
1780
1781void kvm_unregister_perf_callbacks(void);
1782#else
1783static inline void kvm_register_perf_callbacks(void) {}
1784static inline void kvm_unregister_perf_callbacks(void) {}
1785#endif /* CONFIG_GUEST_PERF_EVENTS */
1786
1787int kvm_arch_init_vm(struct kvm *kvm, unsigned long type);
1788void kvm_arch_destroy_vm(struct kvm *kvm);
1789
1790int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu);
1791
1792struct kvm_irq_ack_notifier {
1793 struct hlist_node link;
1794 unsigned gsi;
1795 void (*irq_acked)(struct kvm_irq_ack_notifier *kian);
1796};
1797
1798int kvm_irq_map_gsi(struct kvm *kvm,
1799 struct kvm_kernel_irq_routing_entry *entries, int gsi);
1800int kvm_irq_map_chip_pin(struct kvm *kvm, unsigned irqchip, unsigned pin);
1801
1802int kvm_set_irq(struct kvm *kvm, int irq_source_id, u32 irq, int level,
1803 bool line_status);
1804int kvm_set_msi(struct kvm_kernel_irq_routing_entry *irq_entry, struct kvm *kvm,
1805 int irq_source_id, int level, bool line_status);
1806int kvm_arch_set_irq_inatomic(struct kvm_kernel_irq_routing_entry *e,
1807 struct kvm *kvm, int irq_source_id,
1808 int level, bool line_status);
1809bool kvm_irq_has_notifier(struct kvm *kvm, unsigned irqchip, unsigned pin);
1810void kvm_notify_acked_gsi(struct kvm *kvm, int gsi);
1811void kvm_notify_acked_irq(struct kvm *kvm, unsigned irqchip, unsigned pin);
1812void kvm_register_irq_ack_notifier(struct kvm *kvm,
1813 struct kvm_irq_ack_notifier *kian);
1814void kvm_unregister_irq_ack_notifier(struct kvm *kvm,
1815 struct kvm_irq_ack_notifier *kian);
1816bool kvm_arch_irqfd_allowed(struct kvm *kvm, struct kvm_irqfd *args);
1817
1818/*
1819 * Returns a pointer to the memslot if it contains gfn.
1820 * Otherwise returns NULL.
1821 */
1822static inline struct kvm_memory_slot *
1823try_get_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1824{
1825 if (!slot)
1826 return NULL;
1827
1828 if (gfn >= slot->base_gfn && gfn < slot->base_gfn + slot->npages)
1829 return slot;
1830 else
1831 return NULL;
1832}
1833
1834/*
1835 * Returns a pointer to the memslot that contains gfn. Otherwise returns NULL.
1836 *
1837 * With "approx" set returns the memslot also when the address falls
1838 * in a hole. In that case one of the memslots bordering the hole is
1839 * returned.
1840 */
1841static inline struct kvm_memory_slot *
1842search_memslots(struct kvm_memslots *slots, gfn_t gfn, bool approx)
1843{
1844 struct kvm_memory_slot *slot;
1845 struct rb_node *node;
1846 int idx = slots->node_idx;
1847
1848 slot = NULL;
1849 for (node = slots->gfn_tree.rb_node; node; ) {
1850 slot = container_of(node, struct kvm_memory_slot, gfn_node[idx]);
1851 if (gfn >= slot->base_gfn) {
1852 if (gfn < slot->base_gfn + slot->npages)
1853 return slot;
1854 node = node->rb_right;
1855 } else
1856 node = node->rb_left;
1857 }
1858
1859 return approx ? slot : NULL;
1860}
1861
1862static inline struct kvm_memory_slot *
1863____gfn_to_memslot(struct kvm_memslots *slots, gfn_t gfn, bool approx)
1864{
1865 struct kvm_memory_slot *slot;
1866
1867 slot = (struct kvm_memory_slot *)atomic_long_read(&slots->last_used_slot);
1868 slot = try_get_memslot(slot, gfn);
1869 if (slot)
1870 return slot;
1871
1872 slot = search_memslots(slots, gfn, approx);
1873 if (slot) {
1874 atomic_long_set(&slots->last_used_slot, (unsigned long)slot);
1875 return slot;
1876 }
1877
1878 return NULL;
1879}
1880
1881/*
1882 * __gfn_to_memslot() and its descendants are here to allow arch code to inline
1883 * the lookups in hot paths. gfn_to_memslot() itself isn't here as an inline
1884 * because that would bloat other code too much.
1885 */
1886static inline struct kvm_memory_slot *
1887__gfn_to_memslot(struct kvm_memslots *slots, gfn_t gfn)
1888{
1889 return ____gfn_to_memslot(slots, gfn, false);
1890}
1891
1892static inline unsigned long
1893__gfn_to_hva_memslot(const struct kvm_memory_slot *slot, gfn_t gfn)
1894{
1895 /*
1896 * The index was checked originally in search_memslots. To avoid
1897 * that a malicious guest builds a Spectre gadget out of e.g. page
1898 * table walks, do not let the processor speculate loads outside
1899 * the guest's registered memslots.
1900 */
1901 unsigned long offset = gfn - slot->base_gfn;
1902 offset = array_index_nospec(offset, slot->npages);
1903 return slot->userspace_addr + offset * PAGE_SIZE;
1904}
1905
1906static inline int memslot_id(struct kvm *kvm, gfn_t gfn)
1907{
1908 return gfn_to_memslot(kvm, gfn)->id;
1909}
1910
1911static inline gfn_t
1912hva_to_gfn_memslot(unsigned long hva, struct kvm_memory_slot *slot)
1913{
1914 gfn_t gfn_offset = (hva - slot->userspace_addr) >> PAGE_SHIFT;
1915
1916 return slot->base_gfn + gfn_offset;
1917}
1918
1919static inline gpa_t gfn_to_gpa(gfn_t gfn)
1920{
1921 return (gpa_t)gfn << PAGE_SHIFT;
1922}
1923
1924static inline gfn_t gpa_to_gfn(gpa_t gpa)
1925{
1926 return (gfn_t)(gpa >> PAGE_SHIFT);
1927}
1928
1929static inline hpa_t pfn_to_hpa(kvm_pfn_t pfn)
1930{
1931 return (hpa_t)pfn << PAGE_SHIFT;
1932}
1933
1934static inline bool kvm_is_gpa_in_memslot(struct kvm *kvm, gpa_t gpa)
1935{
1936 unsigned long hva = gfn_to_hva(kvm, gpa_to_gfn(gpa));
1937
1938 return !kvm_is_error_hva(hva);
1939}
1940
1941static inline void kvm_gpc_mark_dirty_in_slot(struct gfn_to_pfn_cache *gpc)
1942{
1943 lockdep_assert_held(&gpc->lock);
1944
1945 if (!gpc->memslot)
1946 return;
1947
1948 mark_page_dirty_in_slot(gpc->kvm, gpc->memslot, gpa_to_gfn(gpc->gpa));
1949}
1950
1951enum kvm_stat_kind {
1952 KVM_STAT_VM,
1953 KVM_STAT_VCPU,
1954};
1955
1956struct kvm_stat_data {
1957 struct kvm *kvm;
1958 const struct kvm_stats_desc *desc;
1959 enum kvm_stat_kind kind;
1960};
1961
1962#define STATS_DESC_COMMON(type, unit, base, exp, sz, bsz) \
1963 .flags = type | unit | base | \
1964 BUILD_BUG_ON_ZERO(type & ~KVM_STATS_TYPE_MASK) | \
1965 BUILD_BUG_ON_ZERO(unit & ~KVM_STATS_UNIT_MASK) | \
1966 BUILD_BUG_ON_ZERO(base & ~KVM_STATS_BASE_MASK), \
1967 .exponent = exp, \
1968 .size = sz, \
1969 .bucket_size = bsz
1970
1971#define VM_GENERIC_STATS_DESC(stat, type, unit, base, exp, sz, bsz) \
1972{ \
1973 STATS_DESC_COMMON(type, unit, base, exp, sz, bsz), \
1974 .offset = offsetof(struct kvm_vm_stat, generic.stat), \
1975 .name = #stat, \
1976}
1977#define VCPU_GENERIC_STATS_DESC(stat, type, unit, base, exp, sz, bsz) \
1978{ \
1979 STATS_DESC_COMMON(type, unit, base, exp, sz, bsz), \
1980 .offset = offsetof(struct kvm_vcpu_stat, generic.stat), \
1981 .name = #stat, \
1982}
1983#define VM_STATS_DESC(stat, type, unit, base, exp, sz, bsz) \
1984{ \
1985 STATS_DESC_COMMON(type, unit, base, exp, sz, bsz), \
1986 .offset = offsetof(struct kvm_vm_stat, stat), \
1987 .name = #stat, \
1988}
1989#define VCPU_STATS_DESC(stat, type, unit, base, exp, sz, bsz) \
1990{ \
1991 STATS_DESC_COMMON(type, unit, base, exp, sz, bsz), \
1992 .offset = offsetof(struct kvm_vcpu_stat, stat), \
1993 .name = #stat, \
1994}
1995/* SCOPE: VM, VM_GENERIC, VCPU, VCPU_GENERIC */
1996#define STATS_DESC(SCOPE, stat, type, unit, base, exp, sz, bsz) \
1997 SCOPE##_STATS_DESC(stat, type, unit, base, exp, sz, bsz)
1998
1999#define STATS_DESC_CUMULATIVE(SCOPE, name, unit, base, exponent) \
2000 STATS_DESC(SCOPE, name, KVM_STATS_TYPE_CUMULATIVE, \
2001 unit, base, exponent, 1, 0)
2002#define STATS_DESC_INSTANT(SCOPE, name, unit, base, exponent) \
2003 STATS_DESC(SCOPE, name, KVM_STATS_TYPE_INSTANT, \
2004 unit, base, exponent, 1, 0)
2005#define STATS_DESC_PEAK(SCOPE, name, unit, base, exponent) \
2006 STATS_DESC(SCOPE, name, KVM_STATS_TYPE_PEAK, \
2007 unit, base, exponent, 1, 0)
2008#define STATS_DESC_LINEAR_HIST(SCOPE, name, unit, base, exponent, sz, bsz) \
2009 STATS_DESC(SCOPE, name, KVM_STATS_TYPE_LINEAR_HIST, \
2010 unit, base, exponent, sz, bsz)
2011#define STATS_DESC_LOG_HIST(SCOPE, name, unit, base, exponent, sz) \
2012 STATS_DESC(SCOPE, name, KVM_STATS_TYPE_LOG_HIST, \
2013 unit, base, exponent, sz, 0)
2014
2015/* Cumulative counter, read/write */
2016#define STATS_DESC_COUNTER(SCOPE, name) \
2017 STATS_DESC_CUMULATIVE(SCOPE, name, KVM_STATS_UNIT_NONE, \
2018 KVM_STATS_BASE_POW10, 0)
2019/* Instantaneous counter, read only */
2020#define STATS_DESC_ICOUNTER(SCOPE, name) \
2021 STATS_DESC_INSTANT(SCOPE, name, KVM_STATS_UNIT_NONE, \
2022 KVM_STATS_BASE_POW10, 0)
2023/* Peak counter, read/write */
2024#define STATS_DESC_PCOUNTER(SCOPE, name) \
2025 STATS_DESC_PEAK(SCOPE, name, KVM_STATS_UNIT_NONE, \
2026 KVM_STATS_BASE_POW10, 0)
2027
2028/* Instantaneous boolean value, read only */
2029#define STATS_DESC_IBOOLEAN(SCOPE, name) \
2030 STATS_DESC_INSTANT(SCOPE, name, KVM_STATS_UNIT_BOOLEAN, \
2031 KVM_STATS_BASE_POW10, 0)
2032/* Peak (sticky) boolean value, read/write */
2033#define STATS_DESC_PBOOLEAN(SCOPE, name) \
2034 STATS_DESC_PEAK(SCOPE, name, KVM_STATS_UNIT_BOOLEAN, \
2035 KVM_STATS_BASE_POW10, 0)
2036
2037/* Cumulative time in nanosecond */
2038#define STATS_DESC_TIME_NSEC(SCOPE, name) \
2039 STATS_DESC_CUMULATIVE(SCOPE, name, KVM_STATS_UNIT_SECONDS, \
2040 KVM_STATS_BASE_POW10, -9)
2041/* Linear histogram for time in nanosecond */
2042#define STATS_DESC_LINHIST_TIME_NSEC(SCOPE, name, sz, bsz) \
2043 STATS_DESC_LINEAR_HIST(SCOPE, name, KVM_STATS_UNIT_SECONDS, \
2044 KVM_STATS_BASE_POW10, -9, sz, bsz)
2045/* Logarithmic histogram for time in nanosecond */
2046#define STATS_DESC_LOGHIST_TIME_NSEC(SCOPE, name, sz) \
2047 STATS_DESC_LOG_HIST(SCOPE, name, KVM_STATS_UNIT_SECONDS, \
2048 KVM_STATS_BASE_POW10, -9, sz)
2049
2050#define KVM_GENERIC_VM_STATS() \
2051 STATS_DESC_COUNTER(VM_GENERIC, remote_tlb_flush), \
2052 STATS_DESC_COUNTER(VM_GENERIC, remote_tlb_flush_requests)
2053
2054#define KVM_GENERIC_VCPU_STATS() \
2055 STATS_DESC_COUNTER(VCPU_GENERIC, halt_successful_poll), \
2056 STATS_DESC_COUNTER(VCPU_GENERIC, halt_attempted_poll), \
2057 STATS_DESC_COUNTER(VCPU_GENERIC, halt_poll_invalid), \
2058 STATS_DESC_COUNTER(VCPU_GENERIC, halt_wakeup), \
2059 STATS_DESC_TIME_NSEC(VCPU_GENERIC, halt_poll_success_ns), \
2060 STATS_DESC_TIME_NSEC(VCPU_GENERIC, halt_poll_fail_ns), \
2061 STATS_DESC_TIME_NSEC(VCPU_GENERIC, halt_wait_ns), \
2062 STATS_DESC_LOGHIST_TIME_NSEC(VCPU_GENERIC, halt_poll_success_hist, \
2063 HALT_POLL_HIST_COUNT), \
2064 STATS_DESC_LOGHIST_TIME_NSEC(VCPU_GENERIC, halt_poll_fail_hist, \
2065 HALT_POLL_HIST_COUNT), \
2066 STATS_DESC_LOGHIST_TIME_NSEC(VCPU_GENERIC, halt_wait_hist, \
2067 HALT_POLL_HIST_COUNT), \
2068 STATS_DESC_IBOOLEAN(VCPU_GENERIC, blocking)
2069
2070ssize_t kvm_stats_read(char *id, const struct kvm_stats_header *header,
2071 const struct kvm_stats_desc *desc,
2072 void *stats, size_t size_stats,
2073 char __user *user_buffer, size_t size, loff_t *offset);
2074
2075/**
2076 * kvm_stats_linear_hist_update() - Update bucket value for linear histogram
2077 * statistics data.
2078 *
2079 * @data: start address of the stats data
2080 * @size: the number of bucket of the stats data
2081 * @value: the new value used to update the linear histogram's bucket
2082 * @bucket_size: the size (width) of a bucket
2083 */
2084static inline void kvm_stats_linear_hist_update(u64 *data, size_t size,
2085 u64 value, size_t bucket_size)
2086{
2087 size_t index = div64_u64(value, bucket_size);
2088
2089 index = min(index, size - 1);
2090 ++data[index];
2091}
2092
2093/**
2094 * kvm_stats_log_hist_update() - Update bucket value for logarithmic histogram
2095 * statistics data.
2096 *
2097 * @data: start address of the stats data
2098 * @size: the number of bucket of the stats data
2099 * @value: the new value used to update the logarithmic histogram's bucket
2100 */
2101static inline void kvm_stats_log_hist_update(u64 *data, size_t size, u64 value)
2102{
2103 size_t index = fls64(value);
2104
2105 index = min(index, size - 1);
2106 ++data[index];
2107}
2108
2109#define KVM_STATS_LINEAR_HIST_UPDATE(array, value, bsize) \
2110 kvm_stats_linear_hist_update(array, ARRAY_SIZE(array), value, bsize)
2111#define KVM_STATS_LOG_HIST_UPDATE(array, value) \
2112 kvm_stats_log_hist_update(array, ARRAY_SIZE(array), value)
2113
2114
2115extern const struct kvm_stats_header kvm_vm_stats_header;
2116extern const struct kvm_stats_desc kvm_vm_stats_desc[];
2117extern const struct kvm_stats_header kvm_vcpu_stats_header;
2118extern const struct kvm_stats_desc kvm_vcpu_stats_desc[];
2119
2120static inline int mmu_invalidate_retry(struct kvm *kvm, unsigned long mmu_seq)
2121{
2122 if (unlikely(kvm->mmu_invalidate_in_progress))
2123 return 1;
2124 /*
2125 * Ensure the read of mmu_invalidate_in_progress happens before
2126 * the read of mmu_invalidate_seq. This interacts with the
2127 * smp_wmb() in mmu_notifier_invalidate_range_end to make sure
2128 * that the caller either sees the old (non-zero) value of
2129 * mmu_invalidate_in_progress or the new (incremented) value of
2130 * mmu_invalidate_seq.
2131 *
2132 * PowerPC Book3s HV KVM calls this under a per-page lock rather
2133 * than under kvm->mmu_lock, for scalability, so can't rely on
2134 * kvm->mmu_lock to keep things ordered.
2135 */
2136 smp_rmb();
2137 if (kvm->mmu_invalidate_seq != mmu_seq)
2138 return 1;
2139 return 0;
2140}
2141
2142static inline int mmu_invalidate_retry_gfn(struct kvm *kvm,
2143 unsigned long mmu_seq,
2144 gfn_t gfn)
2145{
2146 lockdep_assert_held(&kvm->mmu_lock);
2147 /*
2148 * If mmu_invalidate_in_progress is non-zero, then the range maintained
2149 * by kvm_mmu_notifier_invalidate_range_start contains all addresses
2150 * that might be being invalidated. Note that it may include some false
2151 * positives, due to shortcuts when handing concurrent invalidations.
2152 */
2153 if (unlikely(kvm->mmu_invalidate_in_progress)) {
2154 /*
2155 * Dropping mmu_lock after bumping mmu_invalidate_in_progress
2156 * but before updating the range is a KVM bug.
2157 */
2158 if (WARN_ON_ONCE(kvm->mmu_invalidate_range_start == INVALID_GPA ||
2159 kvm->mmu_invalidate_range_end == INVALID_GPA))
2160 return 1;
2161
2162 if (gfn >= kvm->mmu_invalidate_range_start &&
2163 gfn < kvm->mmu_invalidate_range_end)
2164 return 1;
2165 }
2166
2167 if (kvm->mmu_invalidate_seq != mmu_seq)
2168 return 1;
2169 return 0;
2170}
2171
2172/*
2173 * This lockless version of the range-based retry check *must* be paired with a
2174 * call to the locked version after acquiring mmu_lock, i.e. this is safe to
2175 * use only as a pre-check to avoid contending mmu_lock. This version *will*
2176 * get false negatives and false positives.
2177 */
2178static inline bool mmu_invalidate_retry_gfn_unsafe(struct kvm *kvm,
2179 unsigned long mmu_seq,
2180 gfn_t gfn)
2181{
2182 /*
2183 * Use READ_ONCE() to ensure the in-progress flag and sequence counter
2184 * are always read from memory, e.g. so that checking for retry in a
2185 * loop won't result in an infinite retry loop. Don't force loads for
2186 * start+end, as the key to avoiding infinite retry loops is observing
2187 * the 1=>0 transition of in-progress, i.e. getting false negatives
2188 * due to stale start+end values is acceptable.
2189 */
2190 if (unlikely(READ_ONCE(kvm->mmu_invalidate_in_progress)) &&
2191 gfn >= kvm->mmu_invalidate_range_start &&
2192 gfn < kvm->mmu_invalidate_range_end)
2193 return true;
2194
2195 return READ_ONCE(kvm->mmu_invalidate_seq) != mmu_seq;
2196}
2197
2198#ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2199
2200#define KVM_MAX_IRQ_ROUTES 4096 /* might need extension/rework in the future */
2201
2202bool kvm_arch_can_set_irq_routing(struct kvm *kvm);
2203int kvm_set_irq_routing(struct kvm *kvm,
2204 const struct kvm_irq_routing_entry *entries,
2205 unsigned nr,
2206 unsigned flags);
2207int kvm_init_irq_routing(struct kvm *kvm);
2208int kvm_set_routing_entry(struct kvm *kvm,
2209 struct kvm_kernel_irq_routing_entry *e,
2210 const struct kvm_irq_routing_entry *ue);
2211void kvm_free_irq_routing(struct kvm *kvm);
2212
2213#else
2214
2215static inline void kvm_free_irq_routing(struct kvm *kvm) {}
2216
2217static inline int kvm_init_irq_routing(struct kvm *kvm)
2218{
2219 return 0;
2220}
2221
2222#endif
2223
2224int kvm_send_userspace_msi(struct kvm *kvm, struct kvm_msi *msi);
2225
2226void kvm_eventfd_init(struct kvm *kvm);
2227int kvm_ioeventfd(struct kvm *kvm, struct kvm_ioeventfd *args);
2228
2229#ifdef CONFIG_HAVE_KVM_IRQCHIP
2230int kvm_irqfd(struct kvm *kvm, struct kvm_irqfd *args);
2231void kvm_irqfd_release(struct kvm *kvm);
2232bool kvm_notify_irqfd_resampler(struct kvm *kvm,
2233 unsigned int irqchip,
2234 unsigned int pin);
2235void kvm_irq_routing_update(struct kvm *);
2236#else
2237static inline int kvm_irqfd(struct kvm *kvm, struct kvm_irqfd *args)
2238{
2239 return -EINVAL;
2240}
2241
2242static inline void kvm_irqfd_release(struct kvm *kvm) {}
2243
2244static inline bool kvm_notify_irqfd_resampler(struct kvm *kvm,
2245 unsigned int irqchip,
2246 unsigned int pin)
2247{
2248 return false;
2249}
2250#endif /* CONFIG_HAVE_KVM_IRQCHIP */
2251
2252void kvm_arch_irq_routing_update(struct kvm *kvm);
2253
2254static inline void __kvm_make_request(int req, struct kvm_vcpu *vcpu)
2255{
2256 /*
2257 * Ensure the rest of the request is published to kvm_check_request's
2258 * caller. Paired with the smp_mb__after_atomic in kvm_check_request.
2259 */
2260 smp_wmb();
2261 set_bit(req & KVM_REQUEST_MASK, (void *)&vcpu->requests);
2262}
2263
2264static __always_inline void kvm_make_request(int req, struct kvm_vcpu *vcpu)
2265{
2266 /*
2267 * Request that don't require vCPU action should never be logged in
2268 * vcpu->requests. The vCPU won't clear the request, so it will stay
2269 * logged indefinitely and prevent the vCPU from entering the guest.
2270 */
2271 BUILD_BUG_ON(!__builtin_constant_p(req) ||
2272 (req & KVM_REQUEST_NO_ACTION));
2273
2274 __kvm_make_request(req, vcpu);
2275}
2276
2277#ifndef CONFIG_S390
2278static inline void kvm_make_request_and_kick(int req, struct kvm_vcpu *vcpu)
2279{
2280 kvm_make_request(req, vcpu);
2281 __kvm_vcpu_kick(vcpu, req & KVM_REQUEST_WAIT);
2282}
2283#endif
2284
2285static inline bool kvm_request_pending(struct kvm_vcpu *vcpu)
2286{
2287 return READ_ONCE(vcpu->requests);
2288}
2289
2290static inline bool kvm_test_request(int req, struct kvm_vcpu *vcpu)
2291{
2292 return test_bit(req & KVM_REQUEST_MASK, (void *)&vcpu->requests);
2293}
2294
2295static inline void kvm_clear_request(int req, struct kvm_vcpu *vcpu)
2296{
2297 clear_bit(req & KVM_REQUEST_MASK, (void *)&vcpu->requests);
2298}
2299
2300static inline bool kvm_check_request(int req, struct kvm_vcpu *vcpu)
2301{
2302 if (kvm_test_request(req, vcpu)) {
2303 kvm_clear_request(req, vcpu);
2304
2305 /*
2306 * Ensure the rest of the request is visible to kvm_check_request's
2307 * caller. Paired with the smp_wmb in kvm_make_request.
2308 */
2309 smp_mb__after_atomic();
2310 return true;
2311 } else {
2312 return false;
2313 }
2314}
2315
2316#ifdef CONFIG_KVM_GENERIC_HARDWARE_ENABLING
2317extern bool enable_virt_at_load;
2318#endif
2319
2320extern unsigned int halt_poll_ns;
2321extern unsigned int halt_poll_ns_grow;
2322extern unsigned int halt_poll_ns_grow_start;
2323extern unsigned int halt_poll_ns_shrink;
2324
2325struct kvm_device {
2326 const struct kvm_device_ops *ops;
2327 struct kvm *kvm;
2328 void *private;
2329 struct list_head vm_node;
2330};
2331
2332/* create, destroy, and name are mandatory */
2333struct kvm_device_ops {
2334 const char *name;
2335
2336 /*
2337 * create is called holding kvm->lock and any operations not suitable
2338 * to do while holding the lock should be deferred to init (see
2339 * below).
2340 */
2341 int (*create)(struct kvm_device *dev, u32 type);
2342
2343 /*
2344 * init is called after create if create is successful and is called
2345 * outside of holding kvm->lock.
2346 */
2347 void (*init)(struct kvm_device *dev);
2348
2349 /*
2350 * Destroy is responsible for freeing dev.
2351 *
2352 * Destroy may be called before or after destructors are called
2353 * on emulated I/O regions, depending on whether a reference is
2354 * held by a vcpu or other kvm component that gets destroyed
2355 * after the emulated I/O.
2356 */
2357 void (*destroy)(struct kvm_device *dev);
2358
2359 /*
2360 * Release is an alternative method to free the device. It is
2361 * called when the device file descriptor is closed. Once
2362 * release is called, the destroy method will not be called
2363 * anymore as the device is removed from the device list of
2364 * the VM. kvm->lock is held.
2365 */
2366 void (*release)(struct kvm_device *dev);
2367
2368 int (*set_attr)(struct kvm_device *dev, struct kvm_device_attr *attr);
2369 int (*get_attr)(struct kvm_device *dev, struct kvm_device_attr *attr);
2370 int (*has_attr)(struct kvm_device *dev, struct kvm_device_attr *attr);
2371 long (*ioctl)(struct kvm_device *dev, unsigned int ioctl,
2372 unsigned long arg);
2373 int (*mmap)(struct kvm_device *dev, struct vm_area_struct *vma);
2374};
2375
2376struct kvm_device *kvm_device_from_filp(struct file *filp);
2377int kvm_register_device_ops(const struct kvm_device_ops *ops, u32 type);
2378void kvm_unregister_device_ops(u32 type);
2379
2380extern struct kvm_device_ops kvm_mpic_ops;
2381extern struct kvm_device_ops kvm_arm_vgic_v2_ops;
2382extern struct kvm_device_ops kvm_arm_vgic_v3_ops;
2383extern struct kvm_device_ops kvm_arm_vgic_v5_ops;
2384
2385#ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2386
2387static inline void kvm_vcpu_set_in_spin_loop(struct kvm_vcpu *vcpu, bool val)
2388{
2389 vcpu->spin_loop.in_spin_loop = val;
2390}
2391static inline void kvm_vcpu_set_dy_eligible(struct kvm_vcpu *vcpu, bool val)
2392{
2393 vcpu->spin_loop.dy_eligible = val;
2394}
2395
2396#else /* !CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT */
2397
2398static inline void kvm_vcpu_set_in_spin_loop(struct kvm_vcpu *vcpu, bool val)
2399{
2400}
2401
2402static inline void kvm_vcpu_set_dy_eligible(struct kvm_vcpu *vcpu, bool val)
2403{
2404}
2405#endif /* CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT */
2406
2407static inline bool kvm_is_visible_memslot(struct kvm_memory_slot *memslot)
2408{
2409 return (memslot && memslot->id < KVM_USER_MEM_SLOTS &&
2410 !(memslot->flags & KVM_MEMSLOT_INVALID));
2411}
2412
2413struct kvm_vcpu *kvm_get_running_vcpu(void);
2414struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void);
2415
2416#if IS_ENABLED(CONFIG_HAVE_KVM_IRQ_BYPASS)
2417struct kvm_kernel_irqfd;
2418
2419bool kvm_arch_has_irq_bypass(void);
2420int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *,
2421 struct irq_bypass_producer *);
2422void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *,
2423 struct irq_bypass_producer *);
2424void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *);
2425void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *);
2426void kvm_arch_update_irqfd_routing(struct kvm_kernel_irqfd *irqfd,
2427 struct kvm_kernel_irq_routing_entry *old,
2428 struct kvm_kernel_irq_routing_entry *new);
2429#endif /* CONFIG_HAVE_KVM_IRQ_BYPASS */
2430
2431#ifdef CONFIG_HAVE_KVM_INVALID_WAKEUPS
2432/* If we wakeup during the poll time, was it a sucessful poll? */
2433static inline bool vcpu_valid_wakeup(struct kvm_vcpu *vcpu)
2434{
2435 return vcpu->valid_wakeup;
2436}
2437
2438#else
2439static inline bool vcpu_valid_wakeup(struct kvm_vcpu *vcpu)
2440{
2441 return true;
2442}
2443#endif /* CONFIG_HAVE_KVM_INVALID_WAKEUPS */
2444
2445#ifdef CONFIG_HAVE_KVM_NO_POLL
2446/* Callback that tells if we must not poll */
2447bool kvm_arch_no_poll(struct kvm_vcpu *vcpu);
2448#else
2449static inline bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
2450{
2451 return false;
2452}
2453#endif /* CONFIG_HAVE_KVM_NO_POLL */
2454
2455void kvm_arch_guest_memory_reclaimed(struct kvm *kvm);
2456
2457#ifdef CONFIG_HAVE_KVM_VCPU_RUN_PID_CHANGE
2458int kvm_arch_vcpu_run_pid_change(struct kvm_vcpu *vcpu);
2459#else
2460static inline int kvm_arch_vcpu_run_pid_change(struct kvm_vcpu *vcpu)
2461{
2462 return 0;
2463}
2464#endif /* CONFIG_HAVE_KVM_VCPU_RUN_PID_CHANGE */
2465
2466#ifdef CONFIG_VIRT_XFER_TO_GUEST_WORK
2467static inline void kvm_handle_signal_exit(struct kvm_vcpu *vcpu)
2468{
2469 vcpu->run->exit_reason = KVM_EXIT_INTR;
2470 vcpu->stat.signal_exits++;
2471}
2472
2473static inline int kvm_xfer_to_guest_mode_handle_work(struct kvm_vcpu *vcpu)
2474{
2475 int r = xfer_to_guest_mode_handle_work();
2476
2477 if (r) {
2478 WARN_ON_ONCE(r != -EINTR);
2479 kvm_handle_signal_exit(vcpu);
2480 }
2481 return r;
2482}
2483#endif /* CONFIG_VIRT_XFER_TO_GUEST_WORK */
2484
2485/*
2486 * If more than one page is being (un)accounted, @virt must be the address of
2487 * the first page of a block of pages what were allocated together (i.e
2488 * accounted together).
2489 *
2490 * kvm_account_pgtable_pages() is thread-safe because mod_lruvec_page_state()
2491 * is thread-safe.
2492 */
2493static inline void kvm_account_pgtable_pages(void *virt, int nr)
2494{
2495 mod_lruvec_page_state(virt_to_page(virt), NR_SECONDARY_PAGETABLE, nr);
2496}
2497
2498/*
2499 * This defines how many reserved entries we want to keep before we
2500 * kick the vcpu to the userspace to avoid dirty ring full. This
2501 * value can be tuned to higher if e.g. PML is enabled on the host.
2502 */
2503#define KVM_DIRTY_RING_RSVD_ENTRIES 64
2504
2505/* Max number of entries allowed for each kvm dirty ring */
2506#define KVM_DIRTY_RING_MAX_ENTRIES 65536
2507
2508static inline void kvm_prepare_memory_fault_exit(struct kvm_vcpu *vcpu,
2509 gpa_t gpa, gpa_t size,
2510 bool is_write, bool is_exec,
2511 bool is_private)
2512{
2513 vcpu->run->exit_reason = KVM_EXIT_MEMORY_FAULT;
2514 vcpu->run->memory_fault.gpa = gpa;
2515 vcpu->run->memory_fault.size = size;
2516
2517 /* RWX flags are not (yet) defined or communicated to userspace. */
2518 vcpu->run->memory_fault.flags = 0;
2519 if (is_private)
2520 vcpu->run->memory_fault.flags |= KVM_MEMORY_EXIT_FLAG_PRIVATE;
2521}
2522
2523static inline bool kvm_memslot_is_gmem_only(const struct kvm_memory_slot *slot)
2524{
2525 if (!IS_ENABLED(CONFIG_KVM_GUEST_MEMFD))
2526 return false;
2527
2528 return slot->flags & KVM_MEMSLOT_GMEM_ONLY;
2529}
2530
2531#ifdef CONFIG_KVM_GENERIC_MEMORY_ATTRIBUTES
2532static inline unsigned long kvm_get_memory_attributes(struct kvm *kvm, gfn_t gfn)
2533{
2534 return xa_to_value(xa_load(&kvm->mem_attr_array, gfn));
2535}
2536
2537bool kvm_range_has_memory_attributes(struct kvm *kvm, gfn_t start, gfn_t end,
2538 unsigned long mask, unsigned long attrs);
2539bool kvm_arch_pre_set_memory_attributes(struct kvm *kvm,
2540 struct kvm_gfn_range *range);
2541bool kvm_arch_post_set_memory_attributes(struct kvm *kvm,
2542 struct kvm_gfn_range *range);
2543
2544static inline bool kvm_mem_is_private(struct kvm *kvm, gfn_t gfn)
2545{
2546 return kvm_get_memory_attributes(kvm, gfn) & KVM_MEMORY_ATTRIBUTE_PRIVATE;
2547}
2548#else
2549static inline bool kvm_mem_is_private(struct kvm *kvm, gfn_t gfn)
2550{
2551 return false;
2552}
2553#endif /* CONFIG_KVM_GENERIC_MEMORY_ATTRIBUTES */
2554
2555#ifdef CONFIG_KVM_GUEST_MEMFD
2556int kvm_gmem_get_pfn(struct kvm *kvm, struct kvm_memory_slot *slot,
2557 gfn_t gfn, kvm_pfn_t *pfn, struct page **page,
2558 int *max_order);
2559#else
2560static inline int kvm_gmem_get_pfn(struct kvm *kvm,
2561 struct kvm_memory_slot *slot, gfn_t gfn,
2562 kvm_pfn_t *pfn, struct page **page,
2563 int *max_order)
2564{
2565 KVM_BUG_ON(1, kvm);
2566 return -EIO;
2567}
2568#endif /* CONFIG_KVM_GUEST_MEMFD */
2569
2570#ifdef CONFIG_HAVE_KVM_ARCH_GMEM_PREPARE
2571int kvm_arch_gmem_prepare(struct kvm *kvm, gfn_t gfn, kvm_pfn_t pfn, int max_order);
2572#endif
2573
2574#ifdef CONFIG_HAVE_KVM_ARCH_GMEM_POPULATE
2575/**
2576 * kvm_gmem_populate() - Populate/prepare a GPA range with guest data
2577 *
2578 * @kvm: KVM instance
2579 * @gfn: starting GFN to be populated
2580 * @src: userspace-provided buffer containing data to copy into GFN range
2581 * (passed to @post_populate, and incremented on each iteration
2582 * if not NULL). Must be page-aligned.
2583 * @npages: number of pages to copy from userspace-buffer
2584 * @post_populate: callback to issue for each gmem page that backs the GPA
2585 * range
2586 * @opaque: opaque data to pass to @post_populate callback
2587 *
2588 * This is primarily intended for cases where a gmem-backed GPA range needs
2589 * to be initialized with userspace-provided data prior to being mapped into
2590 * the guest as a private page. This should be called with the slots->lock
2591 * held so that caller-enforced invariants regarding the expected memory
2592 * attributes of the GPA range do not race with KVM_SET_MEMORY_ATTRIBUTES.
2593 *
2594 * Returns the number of pages that were populated.
2595 */
2596typedef int (*kvm_gmem_populate_cb)(struct kvm *kvm, gfn_t gfn, kvm_pfn_t pfn,
2597 struct page *page, void *opaque);
2598
2599long kvm_gmem_populate(struct kvm *kvm, gfn_t gfn, void __user *src, long npages,
2600 kvm_gmem_populate_cb post_populate, void *opaque);
2601#endif
2602
2603#ifdef CONFIG_HAVE_KVM_ARCH_GMEM_INVALIDATE
2604void kvm_arch_gmem_invalidate(kvm_pfn_t start, kvm_pfn_t end);
2605#endif
2606
2607#ifdef CONFIG_KVM_GENERIC_PRE_FAULT_MEMORY
2608long kvm_arch_vcpu_pre_fault_memory(struct kvm_vcpu *vcpu,
2609 struct kvm_pre_fault_memory *range);
2610#endif
2611
2612#endif