1 /* SPDX-License-Identifier: GPL-2.0-only */ 2 #ifndef __KVM_HOST_H 3 #define __KVM_HOST_H 4 5 6 #include <linux/types.h> 7 #include <linux/hardirq.h> 8 #include <linux/list.h> 9 #include <linux/mutex.h> 10 #include <linux/spinlock.h> 11 #include <linux/signal.h> 12 #include <linux/sched.h> 13 #include <linux/sched/stat.h> 14 #include <linux/bug.h> 15 #include <linux/minmax.h> 16 #include <linux/mm.h> 17 #include <linux/mmu_notifier.h> 18 #include <linux/preempt.h> 19 #include <linux/msi.h> 20 #include <linux/slab.h> 21 #include <linux/vmalloc.h> 22 #include <linux/rcupdate.h> 23 #include <linux/ratelimit.h> 24 #include <linux/err.h> 25 #include <linux/irqflags.h> 26 #include <linux/context_tracking.h> 27 #include <linux/irqbypass.h> 28 #include <linux/rcuwait.h> 29 #include <linux/refcount.h> 30 #include <linux/nospec.h> 31 #include <asm/signal.h> 32 33 #include <linux/kvm.h> 34 #include <linux/kvm_para.h> 35 36 #include <linux/kvm_types.h> 37 38 #include <asm/kvm_host.h> 39 #include <linux/kvm_dirty_ring.h> 40 41 #ifndef KVM_MAX_VCPU_ID 42 #define KVM_MAX_VCPU_ID KVM_MAX_VCPUS 43 #endif 44 45 /* 46 * The bit 16 ~ bit 31 of kvm_memory_region::flags are internally used 47 * in kvm, other bits are visible for userspace which are defined in 48 * include/linux/kvm_h. 49 */ 50 #define KVM_MEMSLOT_INVALID (1UL << 16) 51 52 /* 53 * Bit 63 of the memslot generation number is an "update in-progress flag", 54 * e.g. is temporarily set for the duration of install_new_memslots(). 55 * This flag effectively creates a unique generation number that is used to 56 * mark cached memslot data, e.g. MMIO accesses, as potentially being stale, 57 * i.e. may (or may not) have come from the previous memslots generation. 58 * 59 * This is necessary because the actual memslots update is not atomic with 60 * respect to the generation number update. Updating the generation number 61 * first would allow a vCPU to cache a spte from the old memslots using the 62 * new generation number, and updating the generation number after switching 63 * to the new memslots would allow cache hits using the old generation number 64 * to reference the defunct memslots. 65 * 66 * This mechanism is used to prevent getting hits in KVM's caches while a 67 * memslot update is in-progress, and to prevent cache hits *after* updating 68 * the actual generation number against accesses that were inserted into the 69 * cache *before* the memslots were updated. 70 */ 71 #define KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS BIT_ULL(63) 72 73 /* Two fragments for cross MMIO pages. */ 74 #define KVM_MAX_MMIO_FRAGMENTS 2 75 76 #ifndef KVM_ADDRESS_SPACE_NUM 77 #define KVM_ADDRESS_SPACE_NUM 1 78 #endif 79 80 /* 81 * For the normal pfn, the highest 12 bits should be zero, 82 * so we can mask bit 62 ~ bit 52 to indicate the error pfn, 83 * mask bit 63 to indicate the noslot pfn. 84 */ 85 #define KVM_PFN_ERR_MASK (0x7ffULL << 52) 86 #define KVM_PFN_ERR_NOSLOT_MASK (0xfffULL << 52) 87 #define KVM_PFN_NOSLOT (0x1ULL << 63) 88 89 #define KVM_PFN_ERR_FAULT (KVM_PFN_ERR_MASK) 90 #define KVM_PFN_ERR_HWPOISON (KVM_PFN_ERR_MASK + 1) 91 #define KVM_PFN_ERR_RO_FAULT (KVM_PFN_ERR_MASK + 2) 92 93 /* 94 * error pfns indicate that the gfn is in slot but faild to 95 * translate it to pfn on host. 96 */ 97 static inline bool is_error_pfn(kvm_pfn_t pfn) 98 { 99 return !!(pfn & KVM_PFN_ERR_MASK); 100 } 101 102 /* 103 * error_noslot pfns indicate that the gfn can not be 104 * translated to pfn - it is not in slot or failed to 105 * translate it to pfn. 106 */ 107 static inline bool is_error_noslot_pfn(kvm_pfn_t pfn) 108 { 109 return !!(pfn & KVM_PFN_ERR_NOSLOT_MASK); 110 } 111 112 /* noslot pfn indicates that the gfn is not in slot. */ 113 static inline bool is_noslot_pfn(kvm_pfn_t pfn) 114 { 115 return pfn == KVM_PFN_NOSLOT; 116 } 117 118 /* 119 * architectures with KVM_HVA_ERR_BAD other than PAGE_OFFSET (e.g. s390) 120 * provide own defines and kvm_is_error_hva 121 */ 122 #ifndef KVM_HVA_ERR_BAD 123 124 #define KVM_HVA_ERR_BAD (PAGE_OFFSET) 125 #define KVM_HVA_ERR_RO_BAD (PAGE_OFFSET + PAGE_SIZE) 126 127 static inline bool kvm_is_error_hva(unsigned long addr) 128 { 129 return addr >= PAGE_OFFSET; 130 } 131 132 #endif 133 134 #define KVM_ERR_PTR_BAD_PAGE (ERR_PTR(-ENOENT)) 135 136 static inline bool is_error_page(struct page *page) 137 { 138 return IS_ERR(page); 139 } 140 141 #define KVM_REQUEST_MASK GENMASK(7,0) 142 #define KVM_REQUEST_NO_WAKEUP BIT(8) 143 #define KVM_REQUEST_WAIT BIT(9) 144 /* 145 * Architecture-independent vcpu->requests bit members 146 * Bits 4-7 are reserved for more arch-independent bits. 147 */ 148 #define KVM_REQ_TLB_FLUSH (0 | KVM_REQUEST_WAIT | KVM_REQUEST_NO_WAKEUP) 149 #define KVM_REQ_MMU_RELOAD (1 | KVM_REQUEST_WAIT | KVM_REQUEST_NO_WAKEUP) 150 #define KVM_REQ_UNBLOCK 2 151 #define KVM_REQ_UNHALT 3 152 #define KVM_REQUEST_ARCH_BASE 8 153 154 #define KVM_ARCH_REQ_FLAGS(nr, flags) ({ \ 155 BUILD_BUG_ON((unsigned)(nr) >= (sizeof_field(struct kvm_vcpu, requests) * 8) - KVM_REQUEST_ARCH_BASE); \ 156 (unsigned)(((nr) + KVM_REQUEST_ARCH_BASE) | (flags)); \ 157 }) 158 #define KVM_ARCH_REQ(nr) KVM_ARCH_REQ_FLAGS(nr, 0) 159 160 #define KVM_USERSPACE_IRQ_SOURCE_ID 0 161 #define KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID 1 162 163 extern struct mutex kvm_lock; 164 extern struct list_head vm_list; 165 166 struct kvm_io_range { 167 gpa_t addr; 168 int len; 169 struct kvm_io_device *dev; 170 }; 171 172 #define NR_IOBUS_DEVS 1000 173 174 struct kvm_io_bus { 175 int dev_count; 176 int ioeventfd_count; 177 struct kvm_io_range range[]; 178 }; 179 180 enum kvm_bus { 181 KVM_MMIO_BUS, 182 KVM_PIO_BUS, 183 KVM_VIRTIO_CCW_NOTIFY_BUS, 184 KVM_FAST_MMIO_BUS, 185 KVM_NR_BUSES 186 }; 187 188 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr, 189 int len, const void *val); 190 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, 191 gpa_t addr, int len, const void *val, long cookie); 192 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr, 193 int len, void *val); 194 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr, 195 int len, struct kvm_io_device *dev); 196 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx, 197 struct kvm_io_device *dev); 198 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx, 199 gpa_t addr); 200 201 #ifdef CONFIG_KVM_ASYNC_PF 202 struct kvm_async_pf { 203 struct work_struct work; 204 struct list_head link; 205 struct list_head queue; 206 struct kvm_vcpu *vcpu; 207 struct mm_struct *mm; 208 gpa_t cr2_or_gpa; 209 unsigned long addr; 210 struct kvm_arch_async_pf arch; 211 bool wakeup_all; 212 bool notpresent_injected; 213 }; 214 215 void kvm_clear_async_pf_completion_queue(struct kvm_vcpu *vcpu); 216 void kvm_check_async_pf_completion(struct kvm_vcpu *vcpu); 217 bool kvm_setup_async_pf(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa, 218 unsigned long hva, struct kvm_arch_async_pf *arch); 219 int kvm_async_pf_wakeup_all(struct kvm_vcpu *vcpu); 220 #endif 221 222 #ifdef KVM_ARCH_WANT_MMU_NOTIFIER 223 struct kvm_gfn_range { 224 struct kvm_memory_slot *slot; 225 gfn_t start; 226 gfn_t end; 227 pte_t pte; 228 bool may_block; 229 }; 230 bool kvm_unmap_gfn_range(struct kvm *kvm, struct kvm_gfn_range *range); 231 bool kvm_age_gfn(struct kvm *kvm, struct kvm_gfn_range *range); 232 bool kvm_test_age_gfn(struct kvm *kvm, struct kvm_gfn_range *range); 233 bool kvm_set_spte_gfn(struct kvm *kvm, struct kvm_gfn_range *range); 234 #endif 235 236 enum { 237 OUTSIDE_GUEST_MODE, 238 IN_GUEST_MODE, 239 EXITING_GUEST_MODE, 240 READING_SHADOW_PAGE_TABLES, 241 }; 242 243 #define KVM_UNMAPPED_PAGE ((void *) 0x500 + POISON_POINTER_DELTA) 244 245 struct kvm_host_map { 246 /* 247 * Only valid if the 'pfn' is managed by the host kernel (i.e. There is 248 * a 'struct page' for it. When using mem= kernel parameter some memory 249 * can be used as guest memory but they are not managed by host 250 * kernel). 251 * If 'pfn' is not managed by the host kernel, this field is 252 * initialized to KVM_UNMAPPED_PAGE. 253 */ 254 struct page *page; 255 void *hva; 256 kvm_pfn_t pfn; 257 kvm_pfn_t gfn; 258 }; 259 260 /* 261 * Used to check if the mapping is valid or not. Never use 'kvm_host_map' 262 * directly to check for that. 263 */ 264 static inline bool kvm_vcpu_mapped(struct kvm_host_map *map) 265 { 266 return !!map->hva; 267 } 268 269 static inline bool kvm_vcpu_can_poll(ktime_t cur, ktime_t stop) 270 { 271 return single_task_running() && !need_resched() && ktime_before(cur, stop); 272 } 273 274 /* 275 * Sometimes a large or cross-page mmio needs to be broken up into separate 276 * exits for userspace servicing. 277 */ 278 struct kvm_mmio_fragment { 279 gpa_t gpa; 280 void *data; 281 unsigned len; 282 }; 283 284 struct kvm_vcpu { 285 struct kvm *kvm; 286 #ifdef CONFIG_PREEMPT_NOTIFIERS 287 struct preempt_notifier preempt_notifier; 288 #endif 289 int cpu; 290 int vcpu_id; /* id given by userspace at creation */ 291 int vcpu_idx; /* index in kvm->vcpus array */ 292 int srcu_idx; 293 int mode; 294 u64 requests; 295 unsigned long guest_debug; 296 297 int pre_pcpu; 298 struct list_head blocked_vcpu_list; 299 300 struct mutex mutex; 301 struct kvm_run *run; 302 303 struct rcuwait wait; 304 struct pid __rcu *pid; 305 int sigset_active; 306 sigset_t sigset; 307 struct kvm_vcpu_stat stat; 308 unsigned int halt_poll_ns; 309 bool valid_wakeup; 310 311 #ifdef CONFIG_HAS_IOMEM 312 int mmio_needed; 313 int mmio_read_completed; 314 int mmio_is_write; 315 int mmio_cur_fragment; 316 int mmio_nr_fragments; 317 struct kvm_mmio_fragment mmio_fragments[KVM_MAX_MMIO_FRAGMENTS]; 318 #endif 319 320 #ifdef CONFIG_KVM_ASYNC_PF 321 struct { 322 u32 queued; 323 struct list_head queue; 324 struct list_head done; 325 spinlock_t lock; 326 } async_pf; 327 #endif 328 329 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT 330 /* 331 * Cpu relax intercept or pause loop exit optimization 332 * in_spin_loop: set when a vcpu does a pause loop exit 333 * or cpu relax intercepted. 334 * dy_eligible: indicates whether vcpu is eligible for directed yield. 335 */ 336 struct { 337 bool in_spin_loop; 338 bool dy_eligible; 339 } spin_loop; 340 #endif 341 bool preempted; 342 bool ready; 343 struct kvm_vcpu_arch arch; 344 struct kvm_dirty_ring dirty_ring; 345 }; 346 347 /* must be called with irqs disabled */ 348 static __always_inline void guest_enter_irqoff(void) 349 { 350 /* 351 * This is running in ioctl context so its safe to assume that it's the 352 * stime pending cputime to flush. 353 */ 354 instrumentation_begin(); 355 vtime_account_guest_enter(); 356 instrumentation_end(); 357 358 /* 359 * KVM does not hold any references to rcu protected data when it 360 * switches CPU into a guest mode. In fact switching to a guest mode 361 * is very similar to exiting to userspace from rcu point of view. In 362 * addition CPU may stay in a guest mode for quite a long time (up to 363 * one time slice). Lets treat guest mode as quiescent state, just like 364 * we do with user-mode execution. 365 */ 366 if (!context_tracking_guest_enter()) { 367 instrumentation_begin(); 368 rcu_virt_note_context_switch(smp_processor_id()); 369 instrumentation_end(); 370 } 371 } 372 373 static __always_inline void guest_exit_irqoff(void) 374 { 375 context_tracking_guest_exit(); 376 377 instrumentation_begin(); 378 /* Flush the guest cputime we spent on the guest */ 379 vtime_account_guest_exit(); 380 instrumentation_end(); 381 } 382 383 static inline void guest_exit(void) 384 { 385 unsigned long flags; 386 387 local_irq_save(flags); 388 guest_exit_irqoff(); 389 local_irq_restore(flags); 390 } 391 392 static inline int kvm_vcpu_exiting_guest_mode(struct kvm_vcpu *vcpu) 393 { 394 /* 395 * The memory barrier ensures a previous write to vcpu->requests cannot 396 * be reordered with the read of vcpu->mode. It pairs with the general 397 * memory barrier following the write of vcpu->mode in VCPU RUN. 398 */ 399 smp_mb__before_atomic(); 400 return cmpxchg(&vcpu->mode, IN_GUEST_MODE, EXITING_GUEST_MODE); 401 } 402 403 /* 404 * Some of the bitops functions do not support too long bitmaps. 405 * This number must be determined not to exceed such limits. 406 */ 407 #define KVM_MEM_MAX_NR_PAGES ((1UL << 31) - 1) 408 409 struct kvm_memory_slot { 410 gfn_t base_gfn; 411 unsigned long npages; 412 unsigned long *dirty_bitmap; 413 struct kvm_arch_memory_slot arch; 414 unsigned long userspace_addr; 415 u32 flags; 416 short id; 417 u16 as_id; 418 }; 419 420 static inline bool kvm_slot_dirty_track_enabled(struct kvm_memory_slot *slot) 421 { 422 return slot->flags & KVM_MEM_LOG_DIRTY_PAGES; 423 } 424 425 static inline unsigned long kvm_dirty_bitmap_bytes(struct kvm_memory_slot *memslot) 426 { 427 return ALIGN(memslot->npages, BITS_PER_LONG) / 8; 428 } 429 430 static inline unsigned long *kvm_second_dirty_bitmap(struct kvm_memory_slot *memslot) 431 { 432 unsigned long len = kvm_dirty_bitmap_bytes(memslot); 433 434 return memslot->dirty_bitmap + len / sizeof(*memslot->dirty_bitmap); 435 } 436 437 #ifndef KVM_DIRTY_LOG_MANUAL_CAPS 438 #define KVM_DIRTY_LOG_MANUAL_CAPS KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE 439 #endif 440 441 struct kvm_s390_adapter_int { 442 u64 ind_addr; 443 u64 summary_addr; 444 u64 ind_offset; 445 u32 summary_offset; 446 u32 adapter_id; 447 }; 448 449 struct kvm_hv_sint { 450 u32 vcpu; 451 u32 sint; 452 }; 453 454 struct kvm_kernel_irq_routing_entry { 455 u32 gsi; 456 u32 type; 457 int (*set)(struct kvm_kernel_irq_routing_entry *e, 458 struct kvm *kvm, int irq_source_id, int level, 459 bool line_status); 460 union { 461 struct { 462 unsigned irqchip; 463 unsigned pin; 464 } irqchip; 465 struct { 466 u32 address_lo; 467 u32 address_hi; 468 u32 data; 469 u32 flags; 470 u32 devid; 471 } msi; 472 struct kvm_s390_adapter_int adapter; 473 struct kvm_hv_sint hv_sint; 474 }; 475 struct hlist_node link; 476 }; 477 478 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING 479 struct kvm_irq_routing_table { 480 int chip[KVM_NR_IRQCHIPS][KVM_IRQCHIP_NUM_PINS]; 481 u32 nr_rt_entries; 482 /* 483 * Array indexed by gsi. Each entry contains list of irq chips 484 * the gsi is connected to. 485 */ 486 struct hlist_head map[]; 487 }; 488 #endif 489 490 #ifndef KVM_PRIVATE_MEM_SLOTS 491 #define KVM_PRIVATE_MEM_SLOTS 0 492 #endif 493 494 #define KVM_MEM_SLOTS_NUM SHRT_MAX 495 #define KVM_USER_MEM_SLOTS (KVM_MEM_SLOTS_NUM - KVM_PRIVATE_MEM_SLOTS) 496 497 #ifndef __KVM_VCPU_MULTIPLE_ADDRESS_SPACE 498 static inline int kvm_arch_vcpu_memslots_id(struct kvm_vcpu *vcpu) 499 { 500 return 0; 501 } 502 #endif 503 504 /* 505 * Note: 506 * memslots are not sorted by id anymore, please use id_to_memslot() 507 * to get the memslot by its id. 508 */ 509 struct kvm_memslots { 510 u64 generation; 511 /* The mapping table from slot id to the index in memslots[]. */ 512 short id_to_index[KVM_MEM_SLOTS_NUM]; 513 atomic_t lru_slot; 514 int used_slots; 515 struct kvm_memory_slot memslots[]; 516 }; 517 518 struct kvm { 519 #ifdef KVM_HAVE_MMU_RWLOCK 520 rwlock_t mmu_lock; 521 #else 522 spinlock_t mmu_lock; 523 #endif /* KVM_HAVE_MMU_RWLOCK */ 524 525 struct mutex slots_lock; 526 struct mm_struct *mm; /* userspace tied to this vm */ 527 struct kvm_memslots __rcu *memslots[KVM_ADDRESS_SPACE_NUM]; 528 struct kvm_vcpu *vcpus[KVM_MAX_VCPUS]; 529 530 /* 531 * created_vcpus is protected by kvm->lock, and is incremented 532 * at the beginning of KVM_CREATE_VCPU. online_vcpus is only 533 * incremented after storing the kvm_vcpu pointer in vcpus, 534 * and is accessed atomically. 535 */ 536 atomic_t online_vcpus; 537 int created_vcpus; 538 int last_boosted_vcpu; 539 struct list_head vm_list; 540 struct mutex lock; 541 struct kvm_io_bus __rcu *buses[KVM_NR_BUSES]; 542 #ifdef CONFIG_HAVE_KVM_EVENTFD 543 struct { 544 spinlock_t lock; 545 struct list_head items; 546 struct list_head resampler_list; 547 struct mutex resampler_lock; 548 } irqfds; 549 struct list_head ioeventfds; 550 #endif 551 struct kvm_vm_stat stat; 552 struct kvm_arch arch; 553 refcount_t users_count; 554 #ifdef CONFIG_KVM_MMIO 555 struct kvm_coalesced_mmio_ring *coalesced_mmio_ring; 556 spinlock_t ring_lock; 557 struct list_head coalesced_zones; 558 #endif 559 560 struct mutex irq_lock; 561 #ifdef CONFIG_HAVE_KVM_IRQCHIP 562 /* 563 * Update side is protected by irq_lock. 564 */ 565 struct kvm_irq_routing_table __rcu *irq_routing; 566 #endif 567 #ifdef CONFIG_HAVE_KVM_IRQFD 568 struct hlist_head irq_ack_notifier_list; 569 #endif 570 571 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) 572 struct mmu_notifier mmu_notifier; 573 unsigned long mmu_notifier_seq; 574 long mmu_notifier_count; 575 unsigned long mmu_notifier_range_start; 576 unsigned long mmu_notifier_range_end; 577 #endif 578 long tlbs_dirty; 579 struct list_head devices; 580 u64 manual_dirty_log_protect; 581 struct dentry *debugfs_dentry; 582 struct kvm_stat_data **debugfs_stat_data; 583 struct srcu_struct srcu; 584 struct srcu_struct irq_srcu; 585 pid_t userspace_pid; 586 unsigned int max_halt_poll_ns; 587 u32 dirty_ring_size; 588 }; 589 590 #define kvm_err(fmt, ...) \ 591 pr_err("kvm [%i]: " fmt, task_pid_nr(current), ## __VA_ARGS__) 592 #define kvm_info(fmt, ...) \ 593 pr_info("kvm [%i]: " fmt, task_pid_nr(current), ## __VA_ARGS__) 594 #define kvm_debug(fmt, ...) \ 595 pr_debug("kvm [%i]: " fmt, task_pid_nr(current), ## __VA_ARGS__) 596 #define kvm_debug_ratelimited(fmt, ...) \ 597 pr_debug_ratelimited("kvm [%i]: " fmt, task_pid_nr(current), \ 598 ## __VA_ARGS__) 599 #define kvm_pr_unimpl(fmt, ...) \ 600 pr_err_ratelimited("kvm [%i]: " fmt, \ 601 task_tgid_nr(current), ## __VA_ARGS__) 602 603 /* The guest did something we don't support. */ 604 #define vcpu_unimpl(vcpu, fmt, ...) \ 605 kvm_pr_unimpl("vcpu%i, guest rIP: 0x%lx " fmt, \ 606 (vcpu)->vcpu_id, kvm_rip_read(vcpu), ## __VA_ARGS__) 607 608 #define vcpu_debug(vcpu, fmt, ...) \ 609 kvm_debug("vcpu%i " fmt, (vcpu)->vcpu_id, ## __VA_ARGS__) 610 #define vcpu_debug_ratelimited(vcpu, fmt, ...) \ 611 kvm_debug_ratelimited("vcpu%i " fmt, (vcpu)->vcpu_id, \ 612 ## __VA_ARGS__) 613 #define vcpu_err(vcpu, fmt, ...) \ 614 kvm_err("vcpu%i " fmt, (vcpu)->vcpu_id, ## __VA_ARGS__) 615 616 static inline bool kvm_dirty_log_manual_protect_and_init_set(struct kvm *kvm) 617 { 618 return !!(kvm->manual_dirty_log_protect & KVM_DIRTY_LOG_INITIALLY_SET); 619 } 620 621 static inline struct kvm_io_bus *kvm_get_bus(struct kvm *kvm, enum kvm_bus idx) 622 { 623 return srcu_dereference_check(kvm->buses[idx], &kvm->srcu, 624 lockdep_is_held(&kvm->slots_lock) || 625 !refcount_read(&kvm->users_count)); 626 } 627 628 static inline struct kvm_vcpu *kvm_get_vcpu(struct kvm *kvm, int i) 629 { 630 int num_vcpus = atomic_read(&kvm->online_vcpus); 631 i = array_index_nospec(i, num_vcpus); 632 633 /* Pairs with smp_wmb() in kvm_vm_ioctl_create_vcpu. */ 634 smp_rmb(); 635 return kvm->vcpus[i]; 636 } 637 638 #define kvm_for_each_vcpu(idx, vcpup, kvm) \ 639 for (idx = 0; \ 640 idx < atomic_read(&kvm->online_vcpus) && \ 641 (vcpup = kvm_get_vcpu(kvm, idx)) != NULL; \ 642 idx++) 643 644 static inline struct kvm_vcpu *kvm_get_vcpu_by_id(struct kvm *kvm, int id) 645 { 646 struct kvm_vcpu *vcpu = NULL; 647 int i; 648 649 if (id < 0) 650 return NULL; 651 if (id < KVM_MAX_VCPUS) 652 vcpu = kvm_get_vcpu(kvm, id); 653 if (vcpu && vcpu->vcpu_id == id) 654 return vcpu; 655 kvm_for_each_vcpu(i, vcpu, kvm) 656 if (vcpu->vcpu_id == id) 657 return vcpu; 658 return NULL; 659 } 660 661 static inline int kvm_vcpu_get_idx(struct kvm_vcpu *vcpu) 662 { 663 return vcpu->vcpu_idx; 664 } 665 666 #define kvm_for_each_memslot(memslot, slots) \ 667 for (memslot = &slots->memslots[0]; \ 668 memslot < slots->memslots + slots->used_slots; memslot++) \ 669 if (WARN_ON_ONCE(!memslot->npages)) { \ 670 } else 671 672 void kvm_vcpu_destroy(struct kvm_vcpu *vcpu); 673 674 void vcpu_load(struct kvm_vcpu *vcpu); 675 void vcpu_put(struct kvm_vcpu *vcpu); 676 677 #ifdef __KVM_HAVE_IOAPIC 678 void kvm_arch_post_irq_ack_notifier_list_update(struct kvm *kvm); 679 void kvm_arch_post_irq_routing_update(struct kvm *kvm); 680 #else 681 static inline void kvm_arch_post_irq_ack_notifier_list_update(struct kvm *kvm) 682 { 683 } 684 static inline void kvm_arch_post_irq_routing_update(struct kvm *kvm) 685 { 686 } 687 #endif 688 689 #ifdef CONFIG_HAVE_KVM_IRQFD 690 int kvm_irqfd_init(void); 691 void kvm_irqfd_exit(void); 692 #else 693 static inline int kvm_irqfd_init(void) 694 { 695 return 0; 696 } 697 698 static inline void kvm_irqfd_exit(void) 699 { 700 } 701 #endif 702 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align, 703 struct module *module); 704 void kvm_exit(void); 705 706 void kvm_get_kvm(struct kvm *kvm); 707 void kvm_put_kvm(struct kvm *kvm); 708 bool file_is_kvm(struct file *file); 709 void kvm_put_kvm_no_destroy(struct kvm *kvm); 710 711 static inline struct kvm_memslots *__kvm_memslots(struct kvm *kvm, int as_id) 712 { 713 as_id = array_index_nospec(as_id, KVM_ADDRESS_SPACE_NUM); 714 return srcu_dereference_check(kvm->memslots[as_id], &kvm->srcu, 715 lockdep_is_held(&kvm->slots_lock) || 716 !refcount_read(&kvm->users_count)); 717 } 718 719 static inline struct kvm_memslots *kvm_memslots(struct kvm *kvm) 720 { 721 return __kvm_memslots(kvm, 0); 722 } 723 724 static inline struct kvm_memslots *kvm_vcpu_memslots(struct kvm_vcpu *vcpu) 725 { 726 int as_id = kvm_arch_vcpu_memslots_id(vcpu); 727 728 return __kvm_memslots(vcpu->kvm, as_id); 729 } 730 731 static inline 732 struct kvm_memory_slot *id_to_memslot(struct kvm_memslots *slots, int id) 733 { 734 int index = slots->id_to_index[id]; 735 struct kvm_memory_slot *slot; 736 737 if (index < 0) 738 return NULL; 739 740 slot = &slots->memslots[index]; 741 742 WARN_ON(slot->id != id); 743 return slot; 744 } 745 746 /* 747 * KVM_SET_USER_MEMORY_REGION ioctl allows the following operations: 748 * - create a new memory slot 749 * - delete an existing memory slot 750 * - modify an existing memory slot 751 * -- move it in the guest physical memory space 752 * -- just change its flags 753 * 754 * Since flags can be changed by some of these operations, the following 755 * differentiation is the best we can do for __kvm_set_memory_region(): 756 */ 757 enum kvm_mr_change { 758 KVM_MR_CREATE, 759 KVM_MR_DELETE, 760 KVM_MR_MOVE, 761 KVM_MR_FLAGS_ONLY, 762 }; 763 764 int kvm_set_memory_region(struct kvm *kvm, 765 const struct kvm_userspace_memory_region *mem); 766 int __kvm_set_memory_region(struct kvm *kvm, 767 const struct kvm_userspace_memory_region *mem); 768 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot); 769 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen); 770 int kvm_arch_prepare_memory_region(struct kvm *kvm, 771 struct kvm_memory_slot *memslot, 772 const struct kvm_userspace_memory_region *mem, 773 enum kvm_mr_change change); 774 void kvm_arch_commit_memory_region(struct kvm *kvm, 775 const struct kvm_userspace_memory_region *mem, 776 struct kvm_memory_slot *old, 777 const struct kvm_memory_slot *new, 778 enum kvm_mr_change change); 779 /* flush all memory translations */ 780 void kvm_arch_flush_shadow_all(struct kvm *kvm); 781 /* flush memory translations pointing to 'slot' */ 782 void kvm_arch_flush_shadow_memslot(struct kvm *kvm, 783 struct kvm_memory_slot *slot); 784 785 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn, 786 struct page **pages, int nr_pages); 787 788 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn); 789 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn); 790 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable); 791 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot, gfn_t gfn); 792 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot, gfn_t gfn, 793 bool *writable); 794 void kvm_release_page_clean(struct page *page); 795 void kvm_release_page_dirty(struct page *page); 796 void kvm_set_page_accessed(struct page *page); 797 798 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn); 799 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault, 800 bool *writable); 801 kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn); 802 kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn); 803 kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, 804 bool atomic, bool *async, bool write_fault, 805 bool *writable, hva_t *hva); 806 807 void kvm_release_pfn_clean(kvm_pfn_t pfn); 808 void kvm_release_pfn_dirty(kvm_pfn_t pfn); 809 void kvm_set_pfn_dirty(kvm_pfn_t pfn); 810 void kvm_set_pfn_accessed(kvm_pfn_t pfn); 811 void kvm_get_pfn(kvm_pfn_t pfn); 812 813 void kvm_release_pfn(kvm_pfn_t pfn, bool dirty, struct gfn_to_pfn_cache *cache); 814 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset, 815 int len); 816 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len); 817 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, 818 void *data, unsigned long len); 819 int kvm_read_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, 820 void *data, unsigned int offset, 821 unsigned long len); 822 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data, 823 int offset, int len); 824 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data, 825 unsigned long len); 826 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, 827 void *data, unsigned long len); 828 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, 829 void *data, unsigned int offset, 830 unsigned long len); 831 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc, 832 gpa_t gpa, unsigned long len); 833 834 #define __kvm_get_guest(kvm, gfn, offset, v) \ 835 ({ \ 836 unsigned long __addr = gfn_to_hva(kvm, gfn); \ 837 typeof(v) __user *__uaddr = (typeof(__uaddr))(__addr + offset); \ 838 int __ret = -EFAULT; \ 839 \ 840 if (!kvm_is_error_hva(__addr)) \ 841 __ret = get_user(v, __uaddr); \ 842 __ret; \ 843 }) 844 845 #define kvm_get_guest(kvm, gpa, v) \ 846 ({ \ 847 gpa_t __gpa = gpa; \ 848 struct kvm *__kvm = kvm; \ 849 \ 850 __kvm_get_guest(__kvm, __gpa >> PAGE_SHIFT, \ 851 offset_in_page(__gpa), v); \ 852 }) 853 854 #define __kvm_put_guest(kvm, gfn, offset, v) \ 855 ({ \ 856 unsigned long __addr = gfn_to_hva(kvm, gfn); \ 857 typeof(v) __user *__uaddr = (typeof(__uaddr))(__addr + offset); \ 858 int __ret = -EFAULT; \ 859 \ 860 if (!kvm_is_error_hva(__addr)) \ 861 __ret = put_user(v, __uaddr); \ 862 if (!__ret) \ 863 mark_page_dirty(kvm, gfn); \ 864 __ret; \ 865 }) 866 867 #define kvm_put_guest(kvm, gpa, v) \ 868 ({ \ 869 gpa_t __gpa = gpa; \ 870 struct kvm *__kvm = kvm; \ 871 \ 872 __kvm_put_guest(__kvm, __gpa >> PAGE_SHIFT, \ 873 offset_in_page(__gpa), v); \ 874 }) 875 876 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len); 877 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn); 878 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn); 879 bool kvm_vcpu_is_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn); 880 unsigned long kvm_host_page_size(struct kvm_vcpu *vcpu, gfn_t gfn); 881 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot, gfn_t gfn); 882 void mark_page_dirty(struct kvm *kvm, gfn_t gfn); 883 884 struct kvm_memslots *kvm_vcpu_memslots(struct kvm_vcpu *vcpu); 885 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn); 886 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn); 887 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn); 888 int kvm_vcpu_map(struct kvm_vcpu *vcpu, gpa_t gpa, struct kvm_host_map *map); 889 int kvm_map_gfn(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map, 890 struct gfn_to_pfn_cache *cache, bool atomic); 891 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn); 892 void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map, bool dirty); 893 int kvm_unmap_gfn(struct kvm_vcpu *vcpu, struct kvm_host_map *map, 894 struct gfn_to_pfn_cache *cache, bool dirty, bool atomic); 895 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn); 896 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable); 897 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data, int offset, 898 int len); 899 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, 900 unsigned long len); 901 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, 902 unsigned long len); 903 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, const void *data, 904 int offset, int len); 905 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data, 906 unsigned long len); 907 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn); 908 909 void kvm_sigset_activate(struct kvm_vcpu *vcpu); 910 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu); 911 912 void kvm_vcpu_block(struct kvm_vcpu *vcpu); 913 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu); 914 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu); 915 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu); 916 void kvm_vcpu_kick(struct kvm_vcpu *vcpu); 917 int kvm_vcpu_yield_to(struct kvm_vcpu *target); 918 void kvm_vcpu_on_spin(struct kvm_vcpu *vcpu, bool usermode_vcpu_not_eligible); 919 920 void kvm_flush_remote_tlbs(struct kvm *kvm); 921 void kvm_reload_remote_mmus(struct kvm *kvm); 922 923 #ifdef KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE 924 int kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache *mc, int min); 925 int kvm_mmu_memory_cache_nr_free_objects(struct kvm_mmu_memory_cache *mc); 926 void kvm_mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc); 927 void *kvm_mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc); 928 #endif 929 930 bool kvm_make_vcpus_request_mask(struct kvm *kvm, unsigned int req, 931 struct kvm_vcpu *except, 932 unsigned long *vcpu_bitmap, cpumask_var_t tmp); 933 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req); 934 bool kvm_make_all_cpus_request_except(struct kvm *kvm, unsigned int req, 935 struct kvm_vcpu *except); 936 bool kvm_make_cpus_request_mask(struct kvm *kvm, unsigned int req, 937 unsigned long *vcpu_bitmap); 938 939 long kvm_arch_dev_ioctl(struct file *filp, 940 unsigned int ioctl, unsigned long arg); 941 long kvm_arch_vcpu_ioctl(struct file *filp, 942 unsigned int ioctl, unsigned long arg); 943 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf); 944 945 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext); 946 947 void kvm_arch_mmu_enable_log_dirty_pt_masked(struct kvm *kvm, 948 struct kvm_memory_slot *slot, 949 gfn_t gfn_offset, 950 unsigned long mask); 951 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot); 952 953 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT 954 void kvm_arch_flush_remote_tlbs_memslot(struct kvm *kvm, 955 const struct kvm_memory_slot *memslot); 956 #else /* !CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */ 957 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log); 958 int kvm_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log, 959 int *is_dirty, struct kvm_memory_slot **memslot); 960 #endif 961 962 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level, 963 bool line_status); 964 int kvm_vm_ioctl_enable_cap(struct kvm *kvm, 965 struct kvm_enable_cap *cap); 966 long kvm_arch_vm_ioctl(struct file *filp, 967 unsigned int ioctl, unsigned long arg); 968 969 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu); 970 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu); 971 972 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu, 973 struct kvm_translation *tr); 974 975 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs); 976 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs); 977 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu, 978 struct kvm_sregs *sregs); 979 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu, 980 struct kvm_sregs *sregs); 981 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu, 982 struct kvm_mp_state *mp_state); 983 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu, 984 struct kvm_mp_state *mp_state); 985 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu, 986 struct kvm_guest_debug *dbg); 987 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu); 988 989 int kvm_arch_init(void *opaque); 990 void kvm_arch_exit(void); 991 992 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu); 993 994 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu); 995 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu); 996 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id); 997 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu); 998 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu); 999 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu); 1000 1001 #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS 1002 void kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu, struct dentry *debugfs_dentry); 1003 #endif 1004 1005 int kvm_arch_hardware_enable(void); 1006 void kvm_arch_hardware_disable(void); 1007 int kvm_arch_hardware_setup(void *opaque); 1008 void kvm_arch_hardware_unsetup(void); 1009 int kvm_arch_check_processor_compat(void *opaque); 1010 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu); 1011 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu); 1012 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu); 1013 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu); 1014 bool kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu); 1015 int kvm_arch_post_init_vm(struct kvm *kvm); 1016 void kvm_arch_pre_destroy_vm(struct kvm *kvm); 1017 1018 #ifndef __KVM_HAVE_ARCH_VM_ALLOC 1019 /* 1020 * All architectures that want to use vzalloc currently also 1021 * need their own kvm_arch_alloc_vm implementation. 1022 */ 1023 static inline struct kvm *kvm_arch_alloc_vm(void) 1024 { 1025 return kzalloc(sizeof(struct kvm), GFP_KERNEL); 1026 } 1027 1028 static inline void kvm_arch_free_vm(struct kvm *kvm) 1029 { 1030 kfree(kvm); 1031 } 1032 #endif 1033 1034 #ifndef __KVM_HAVE_ARCH_FLUSH_REMOTE_TLB 1035 static inline int kvm_arch_flush_remote_tlb(struct kvm *kvm) 1036 { 1037 return -ENOTSUPP; 1038 } 1039 #endif 1040 1041 #ifdef __KVM_HAVE_ARCH_NONCOHERENT_DMA 1042 void kvm_arch_register_noncoherent_dma(struct kvm *kvm); 1043 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm); 1044 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm); 1045 #else 1046 static inline void kvm_arch_register_noncoherent_dma(struct kvm *kvm) 1047 { 1048 } 1049 1050 static inline void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm) 1051 { 1052 } 1053 1054 static inline bool kvm_arch_has_noncoherent_dma(struct kvm *kvm) 1055 { 1056 return false; 1057 } 1058 #endif 1059 #ifdef __KVM_HAVE_ARCH_ASSIGNED_DEVICE 1060 void kvm_arch_start_assignment(struct kvm *kvm); 1061 void kvm_arch_end_assignment(struct kvm *kvm); 1062 bool kvm_arch_has_assigned_device(struct kvm *kvm); 1063 #else 1064 static inline void kvm_arch_start_assignment(struct kvm *kvm) 1065 { 1066 } 1067 1068 static inline void kvm_arch_end_assignment(struct kvm *kvm) 1069 { 1070 } 1071 1072 static inline bool kvm_arch_has_assigned_device(struct kvm *kvm) 1073 { 1074 return false; 1075 } 1076 #endif 1077 1078 static inline struct rcuwait *kvm_arch_vcpu_get_wait(struct kvm_vcpu *vcpu) 1079 { 1080 #ifdef __KVM_HAVE_ARCH_WQP 1081 return vcpu->arch.waitp; 1082 #else 1083 return &vcpu->wait; 1084 #endif 1085 } 1086 1087 #ifdef __KVM_HAVE_ARCH_INTC_INITIALIZED 1088 /* 1089 * returns true if the virtual interrupt controller is initialized and 1090 * ready to accept virtual IRQ. On some architectures the virtual interrupt 1091 * controller is dynamically instantiated and this is not always true. 1092 */ 1093 bool kvm_arch_intc_initialized(struct kvm *kvm); 1094 #else 1095 static inline bool kvm_arch_intc_initialized(struct kvm *kvm) 1096 { 1097 return true; 1098 } 1099 #endif 1100 1101 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type); 1102 void kvm_arch_destroy_vm(struct kvm *kvm); 1103 void kvm_arch_sync_events(struct kvm *kvm); 1104 1105 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu); 1106 1107 bool kvm_is_reserved_pfn(kvm_pfn_t pfn); 1108 bool kvm_is_zone_device_pfn(kvm_pfn_t pfn); 1109 bool kvm_is_transparent_hugepage(kvm_pfn_t pfn); 1110 1111 struct kvm_irq_ack_notifier { 1112 struct hlist_node link; 1113 unsigned gsi; 1114 void (*irq_acked)(struct kvm_irq_ack_notifier *kian); 1115 }; 1116 1117 int kvm_irq_map_gsi(struct kvm *kvm, 1118 struct kvm_kernel_irq_routing_entry *entries, int gsi); 1119 int kvm_irq_map_chip_pin(struct kvm *kvm, unsigned irqchip, unsigned pin); 1120 1121 int kvm_set_irq(struct kvm *kvm, int irq_source_id, u32 irq, int level, 1122 bool line_status); 1123 int kvm_set_msi(struct kvm_kernel_irq_routing_entry *irq_entry, struct kvm *kvm, 1124 int irq_source_id, int level, bool line_status); 1125 int kvm_arch_set_irq_inatomic(struct kvm_kernel_irq_routing_entry *e, 1126 struct kvm *kvm, int irq_source_id, 1127 int level, bool line_status); 1128 bool kvm_irq_has_notifier(struct kvm *kvm, unsigned irqchip, unsigned pin); 1129 void kvm_notify_acked_gsi(struct kvm *kvm, int gsi); 1130 void kvm_notify_acked_irq(struct kvm *kvm, unsigned irqchip, unsigned pin); 1131 void kvm_register_irq_ack_notifier(struct kvm *kvm, 1132 struct kvm_irq_ack_notifier *kian); 1133 void kvm_unregister_irq_ack_notifier(struct kvm *kvm, 1134 struct kvm_irq_ack_notifier *kian); 1135 int kvm_request_irq_source_id(struct kvm *kvm); 1136 void kvm_free_irq_source_id(struct kvm *kvm, int irq_source_id); 1137 bool kvm_arch_irqfd_allowed(struct kvm *kvm, struct kvm_irqfd *args); 1138 1139 /* 1140 * search_memslots() and __gfn_to_memslot() are here because they are 1141 * used in non-modular code in arch/powerpc/kvm/book3s_hv_rm_mmu.c. 1142 * gfn_to_memslot() itself isn't here as an inline because that would 1143 * bloat other code too much. 1144 * 1145 * IMPORTANT: Slots are sorted from highest GFN to lowest GFN! 1146 */ 1147 static inline struct kvm_memory_slot * 1148 search_memslots(struct kvm_memslots *slots, gfn_t gfn) 1149 { 1150 int start = 0, end = slots->used_slots; 1151 int slot = atomic_read(&slots->lru_slot); 1152 struct kvm_memory_slot *memslots = slots->memslots; 1153 1154 if (unlikely(!slots->used_slots)) 1155 return NULL; 1156 1157 if (gfn >= memslots[slot].base_gfn && 1158 gfn < memslots[slot].base_gfn + memslots[slot].npages) 1159 return &memslots[slot]; 1160 1161 while (start < end) { 1162 slot = start + (end - start) / 2; 1163 1164 if (gfn >= memslots[slot].base_gfn) 1165 end = slot; 1166 else 1167 start = slot + 1; 1168 } 1169 1170 if (start < slots->used_slots && gfn >= memslots[start].base_gfn && 1171 gfn < memslots[start].base_gfn + memslots[start].npages) { 1172 atomic_set(&slots->lru_slot, start); 1173 return &memslots[start]; 1174 } 1175 1176 return NULL; 1177 } 1178 1179 static inline struct kvm_memory_slot * 1180 __gfn_to_memslot(struct kvm_memslots *slots, gfn_t gfn) 1181 { 1182 return search_memslots(slots, gfn); 1183 } 1184 1185 static inline unsigned long 1186 __gfn_to_hva_memslot(const struct kvm_memory_slot *slot, gfn_t gfn) 1187 { 1188 return slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE; 1189 } 1190 1191 static inline int memslot_id(struct kvm *kvm, gfn_t gfn) 1192 { 1193 return gfn_to_memslot(kvm, gfn)->id; 1194 } 1195 1196 static inline gfn_t 1197 hva_to_gfn_memslot(unsigned long hva, struct kvm_memory_slot *slot) 1198 { 1199 gfn_t gfn_offset = (hva - slot->userspace_addr) >> PAGE_SHIFT; 1200 1201 return slot->base_gfn + gfn_offset; 1202 } 1203 1204 static inline gpa_t gfn_to_gpa(gfn_t gfn) 1205 { 1206 return (gpa_t)gfn << PAGE_SHIFT; 1207 } 1208 1209 static inline gfn_t gpa_to_gfn(gpa_t gpa) 1210 { 1211 return (gfn_t)(gpa >> PAGE_SHIFT); 1212 } 1213 1214 static inline hpa_t pfn_to_hpa(kvm_pfn_t pfn) 1215 { 1216 return (hpa_t)pfn << PAGE_SHIFT; 1217 } 1218 1219 static inline struct page *kvm_vcpu_gpa_to_page(struct kvm_vcpu *vcpu, 1220 gpa_t gpa) 1221 { 1222 return kvm_vcpu_gfn_to_page(vcpu, gpa_to_gfn(gpa)); 1223 } 1224 1225 static inline bool kvm_is_error_gpa(struct kvm *kvm, gpa_t gpa) 1226 { 1227 unsigned long hva = gfn_to_hva(kvm, gpa_to_gfn(gpa)); 1228 1229 return kvm_is_error_hva(hva); 1230 } 1231 1232 enum kvm_stat_kind { 1233 KVM_STAT_VM, 1234 KVM_STAT_VCPU, 1235 }; 1236 1237 struct kvm_stat_data { 1238 struct kvm *kvm; 1239 struct kvm_stats_debugfs_item *dbgfs_item; 1240 }; 1241 1242 struct kvm_stats_debugfs_item { 1243 const char *name; 1244 int offset; 1245 enum kvm_stat_kind kind; 1246 int mode; 1247 }; 1248 1249 #define KVM_DBGFS_GET_MODE(dbgfs_item) \ 1250 ((dbgfs_item)->mode ? (dbgfs_item)->mode : 0644) 1251 1252 #define VM_STAT(n, x, ...) \ 1253 { n, offsetof(struct kvm, stat.x), KVM_STAT_VM, ## __VA_ARGS__ } 1254 #define VCPU_STAT(n, x, ...) \ 1255 { n, offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU, ## __VA_ARGS__ } 1256 1257 extern struct kvm_stats_debugfs_item debugfs_entries[]; 1258 extern struct dentry *kvm_debugfs_dir; 1259 1260 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) 1261 static inline int mmu_notifier_retry(struct kvm *kvm, unsigned long mmu_seq) 1262 { 1263 if (unlikely(kvm->mmu_notifier_count)) 1264 return 1; 1265 /* 1266 * Ensure the read of mmu_notifier_count happens before the read 1267 * of mmu_notifier_seq. This interacts with the smp_wmb() in 1268 * mmu_notifier_invalidate_range_end to make sure that the caller 1269 * either sees the old (non-zero) value of mmu_notifier_count or 1270 * the new (incremented) value of mmu_notifier_seq. 1271 * PowerPC Book3s HV KVM calls this under a per-page lock 1272 * rather than under kvm->mmu_lock, for scalability, so 1273 * can't rely on kvm->mmu_lock to keep things ordered. 1274 */ 1275 smp_rmb(); 1276 if (kvm->mmu_notifier_seq != mmu_seq) 1277 return 1; 1278 return 0; 1279 } 1280 1281 static inline int mmu_notifier_retry_hva(struct kvm *kvm, 1282 unsigned long mmu_seq, 1283 unsigned long hva) 1284 { 1285 lockdep_assert_held(&kvm->mmu_lock); 1286 /* 1287 * If mmu_notifier_count is non-zero, then the range maintained by 1288 * kvm_mmu_notifier_invalidate_range_start contains all addresses that 1289 * might be being invalidated. Note that it may include some false 1290 * positives, due to shortcuts when handing concurrent invalidations. 1291 */ 1292 if (unlikely(kvm->mmu_notifier_count) && 1293 hva >= kvm->mmu_notifier_range_start && 1294 hva < kvm->mmu_notifier_range_end) 1295 return 1; 1296 if (kvm->mmu_notifier_seq != mmu_seq) 1297 return 1; 1298 return 0; 1299 } 1300 #endif 1301 1302 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING 1303 1304 #define KVM_MAX_IRQ_ROUTES 4096 /* might need extension/rework in the future */ 1305 1306 bool kvm_arch_can_set_irq_routing(struct kvm *kvm); 1307 int kvm_set_irq_routing(struct kvm *kvm, 1308 const struct kvm_irq_routing_entry *entries, 1309 unsigned nr, 1310 unsigned flags); 1311 int kvm_set_routing_entry(struct kvm *kvm, 1312 struct kvm_kernel_irq_routing_entry *e, 1313 const struct kvm_irq_routing_entry *ue); 1314 void kvm_free_irq_routing(struct kvm *kvm); 1315 1316 #else 1317 1318 static inline void kvm_free_irq_routing(struct kvm *kvm) {} 1319 1320 #endif 1321 1322 int kvm_send_userspace_msi(struct kvm *kvm, struct kvm_msi *msi); 1323 1324 #ifdef CONFIG_HAVE_KVM_EVENTFD 1325 1326 void kvm_eventfd_init(struct kvm *kvm); 1327 int kvm_ioeventfd(struct kvm *kvm, struct kvm_ioeventfd *args); 1328 1329 #ifdef CONFIG_HAVE_KVM_IRQFD 1330 int kvm_irqfd(struct kvm *kvm, struct kvm_irqfd *args); 1331 void kvm_irqfd_release(struct kvm *kvm); 1332 void kvm_irq_routing_update(struct kvm *); 1333 #else 1334 static inline int kvm_irqfd(struct kvm *kvm, struct kvm_irqfd *args) 1335 { 1336 return -EINVAL; 1337 } 1338 1339 static inline void kvm_irqfd_release(struct kvm *kvm) {} 1340 #endif 1341 1342 #else 1343 1344 static inline void kvm_eventfd_init(struct kvm *kvm) {} 1345 1346 static inline int kvm_irqfd(struct kvm *kvm, struct kvm_irqfd *args) 1347 { 1348 return -EINVAL; 1349 } 1350 1351 static inline void kvm_irqfd_release(struct kvm *kvm) {} 1352 1353 #ifdef CONFIG_HAVE_KVM_IRQCHIP 1354 static inline void kvm_irq_routing_update(struct kvm *kvm) 1355 { 1356 } 1357 #endif 1358 1359 static inline int kvm_ioeventfd(struct kvm *kvm, struct kvm_ioeventfd *args) 1360 { 1361 return -ENOSYS; 1362 } 1363 1364 #endif /* CONFIG_HAVE_KVM_EVENTFD */ 1365 1366 void kvm_arch_irq_routing_update(struct kvm *kvm); 1367 1368 static inline void kvm_make_request(int req, struct kvm_vcpu *vcpu) 1369 { 1370 /* 1371 * Ensure the rest of the request is published to kvm_check_request's 1372 * caller. Paired with the smp_mb__after_atomic in kvm_check_request. 1373 */ 1374 smp_wmb(); 1375 set_bit(req & KVM_REQUEST_MASK, (void *)&vcpu->requests); 1376 } 1377 1378 static inline bool kvm_request_pending(struct kvm_vcpu *vcpu) 1379 { 1380 return READ_ONCE(vcpu->requests); 1381 } 1382 1383 static inline bool kvm_test_request(int req, struct kvm_vcpu *vcpu) 1384 { 1385 return test_bit(req & KVM_REQUEST_MASK, (void *)&vcpu->requests); 1386 } 1387 1388 static inline void kvm_clear_request(int req, struct kvm_vcpu *vcpu) 1389 { 1390 clear_bit(req & KVM_REQUEST_MASK, (void *)&vcpu->requests); 1391 } 1392 1393 static inline bool kvm_check_request(int req, struct kvm_vcpu *vcpu) 1394 { 1395 if (kvm_test_request(req, vcpu)) { 1396 kvm_clear_request(req, vcpu); 1397 1398 /* 1399 * Ensure the rest of the request is visible to kvm_check_request's 1400 * caller. Paired with the smp_wmb in kvm_make_request. 1401 */ 1402 smp_mb__after_atomic(); 1403 return true; 1404 } else { 1405 return false; 1406 } 1407 } 1408 1409 extern bool kvm_rebooting; 1410 1411 extern unsigned int halt_poll_ns; 1412 extern unsigned int halt_poll_ns_grow; 1413 extern unsigned int halt_poll_ns_grow_start; 1414 extern unsigned int halt_poll_ns_shrink; 1415 1416 struct kvm_device { 1417 const struct kvm_device_ops *ops; 1418 struct kvm *kvm; 1419 void *private; 1420 struct list_head vm_node; 1421 }; 1422 1423 /* create, destroy, and name are mandatory */ 1424 struct kvm_device_ops { 1425 const char *name; 1426 1427 /* 1428 * create is called holding kvm->lock and any operations not suitable 1429 * to do while holding the lock should be deferred to init (see 1430 * below). 1431 */ 1432 int (*create)(struct kvm_device *dev, u32 type); 1433 1434 /* 1435 * init is called after create if create is successful and is called 1436 * outside of holding kvm->lock. 1437 */ 1438 void (*init)(struct kvm_device *dev); 1439 1440 /* 1441 * Destroy is responsible for freeing dev. 1442 * 1443 * Destroy may be called before or after destructors are called 1444 * on emulated I/O regions, depending on whether a reference is 1445 * held by a vcpu or other kvm component that gets destroyed 1446 * after the emulated I/O. 1447 */ 1448 void (*destroy)(struct kvm_device *dev); 1449 1450 /* 1451 * Release is an alternative method to free the device. It is 1452 * called when the device file descriptor is closed. Once 1453 * release is called, the destroy method will not be called 1454 * anymore as the device is removed from the device list of 1455 * the VM. kvm->lock is held. 1456 */ 1457 void (*release)(struct kvm_device *dev); 1458 1459 int (*set_attr)(struct kvm_device *dev, struct kvm_device_attr *attr); 1460 int (*get_attr)(struct kvm_device *dev, struct kvm_device_attr *attr); 1461 int (*has_attr)(struct kvm_device *dev, struct kvm_device_attr *attr); 1462 long (*ioctl)(struct kvm_device *dev, unsigned int ioctl, 1463 unsigned long arg); 1464 int (*mmap)(struct kvm_device *dev, struct vm_area_struct *vma); 1465 }; 1466 1467 void kvm_device_get(struct kvm_device *dev); 1468 void kvm_device_put(struct kvm_device *dev); 1469 struct kvm_device *kvm_device_from_filp(struct file *filp); 1470 int kvm_register_device_ops(const struct kvm_device_ops *ops, u32 type); 1471 void kvm_unregister_device_ops(u32 type); 1472 1473 extern struct kvm_device_ops kvm_mpic_ops; 1474 extern struct kvm_device_ops kvm_arm_vgic_v2_ops; 1475 extern struct kvm_device_ops kvm_arm_vgic_v3_ops; 1476 1477 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT 1478 1479 static inline void kvm_vcpu_set_in_spin_loop(struct kvm_vcpu *vcpu, bool val) 1480 { 1481 vcpu->spin_loop.in_spin_loop = val; 1482 } 1483 static inline void kvm_vcpu_set_dy_eligible(struct kvm_vcpu *vcpu, bool val) 1484 { 1485 vcpu->spin_loop.dy_eligible = val; 1486 } 1487 1488 #else /* !CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT */ 1489 1490 static inline void kvm_vcpu_set_in_spin_loop(struct kvm_vcpu *vcpu, bool val) 1491 { 1492 } 1493 1494 static inline void kvm_vcpu_set_dy_eligible(struct kvm_vcpu *vcpu, bool val) 1495 { 1496 } 1497 #endif /* CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT */ 1498 1499 static inline bool kvm_is_visible_memslot(struct kvm_memory_slot *memslot) 1500 { 1501 return (memslot && memslot->id < KVM_USER_MEM_SLOTS && 1502 !(memslot->flags & KVM_MEMSLOT_INVALID)); 1503 } 1504 1505 struct kvm_vcpu *kvm_get_running_vcpu(void); 1506 struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void); 1507 1508 #ifdef CONFIG_HAVE_KVM_IRQ_BYPASS 1509 bool kvm_arch_has_irq_bypass(void); 1510 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *, 1511 struct irq_bypass_producer *); 1512 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *, 1513 struct irq_bypass_producer *); 1514 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *); 1515 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *); 1516 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq, 1517 uint32_t guest_irq, bool set); 1518 #endif /* CONFIG_HAVE_KVM_IRQ_BYPASS */ 1519 1520 #ifdef CONFIG_HAVE_KVM_INVALID_WAKEUPS 1521 /* If we wakeup during the poll time, was it a sucessful poll? */ 1522 static inline bool vcpu_valid_wakeup(struct kvm_vcpu *vcpu) 1523 { 1524 return vcpu->valid_wakeup; 1525 } 1526 1527 #else 1528 static inline bool vcpu_valid_wakeup(struct kvm_vcpu *vcpu) 1529 { 1530 return true; 1531 } 1532 #endif /* CONFIG_HAVE_KVM_INVALID_WAKEUPS */ 1533 1534 #ifdef CONFIG_HAVE_KVM_NO_POLL 1535 /* Callback that tells if we must not poll */ 1536 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu); 1537 #else 1538 static inline bool kvm_arch_no_poll(struct kvm_vcpu *vcpu) 1539 { 1540 return false; 1541 } 1542 #endif /* CONFIG_HAVE_KVM_NO_POLL */ 1543 1544 #ifdef CONFIG_HAVE_KVM_VCPU_ASYNC_IOCTL 1545 long kvm_arch_vcpu_async_ioctl(struct file *filp, 1546 unsigned int ioctl, unsigned long arg); 1547 #else 1548 static inline long kvm_arch_vcpu_async_ioctl(struct file *filp, 1549 unsigned int ioctl, 1550 unsigned long arg) 1551 { 1552 return -ENOIOCTLCMD; 1553 } 1554 #endif /* CONFIG_HAVE_KVM_VCPU_ASYNC_IOCTL */ 1555 1556 void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm, 1557 unsigned long start, unsigned long end); 1558 1559 #ifdef CONFIG_HAVE_KVM_VCPU_RUN_PID_CHANGE 1560 int kvm_arch_vcpu_run_pid_change(struct kvm_vcpu *vcpu); 1561 #else 1562 static inline int kvm_arch_vcpu_run_pid_change(struct kvm_vcpu *vcpu) 1563 { 1564 return 0; 1565 } 1566 #endif /* CONFIG_HAVE_KVM_VCPU_RUN_PID_CHANGE */ 1567 1568 typedef int (*kvm_vm_thread_fn_t)(struct kvm *kvm, uintptr_t data); 1569 1570 int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn, 1571 uintptr_t data, const char *name, 1572 struct task_struct **thread_ptr); 1573 1574 #ifdef CONFIG_KVM_XFER_TO_GUEST_WORK 1575 static inline void kvm_handle_signal_exit(struct kvm_vcpu *vcpu) 1576 { 1577 vcpu->run->exit_reason = KVM_EXIT_INTR; 1578 vcpu->stat.signal_exits++; 1579 } 1580 #endif /* CONFIG_KVM_XFER_TO_GUEST_WORK */ 1581 1582 /* 1583 * This defines how many reserved entries we want to keep before we 1584 * kick the vcpu to the userspace to avoid dirty ring full. This 1585 * value can be tuned to higher if e.g. PML is enabled on the host. 1586 */ 1587 #define KVM_DIRTY_RING_RSVD_ENTRIES 64 1588 1589 /* Max number of entries allowed for each kvm dirty ring */ 1590 #define KVM_DIRTY_RING_MAX_ENTRIES 65536 1591 1592 #endif 1593