1#ifndef __KVM_HOST_H 2#define __KVM_HOST_H 3 4/* 5 * This work is licensed under the terms of the GNU GPL, version 2. See 6 * the COPYING file in the top-level directory. 7 */ 8 9#include <linux/types.h> 10#include <linux/hardirq.h> 11#include <linux/list.h> 12#include <linux/mutex.h> 13#include <linux/spinlock.h> 14#include <linux/signal.h> 15#include <linux/sched.h> 16#include <linux/bug.h> 17#include <linux/mm.h> 18#include <linux/mmu_notifier.h> 19#include <linux/preempt.h> 20#include <linux/msi.h> 21#include <linux/slab.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 <asm/signal.h> 28 29#include <linux/kvm.h> 30#include <linux/kvm_para.h> 31 32#include <linux/kvm_types.h> 33 34#include <asm/kvm_host.h> 35 36/* 37 * The bit 16 ~ bit 31 of kvm_memory_region::flags are internally used 38 * in kvm, other bits are visible for userspace which are defined in 39 * include/linux/kvm_h. 40 */ 41#define KVM_MEMSLOT_INVALID (1UL << 16) 42#define KVM_MEMSLOT_INCOHERENT (1UL << 17) 43 44/* Two fragments for cross MMIO pages. */ 45#define KVM_MAX_MMIO_FRAGMENTS 2 46 47/* 48 * For the normal pfn, the highest 12 bits should be zero, 49 * so we can mask bit 62 ~ bit 52 to indicate the error pfn, 50 * mask bit 63 to indicate the noslot pfn. 51 */ 52#define KVM_PFN_ERR_MASK (0x7ffULL << 52) 53#define KVM_PFN_ERR_NOSLOT_MASK (0xfffULL << 52) 54#define KVM_PFN_NOSLOT (0x1ULL << 63) 55 56#define KVM_PFN_ERR_FAULT (KVM_PFN_ERR_MASK) 57#define KVM_PFN_ERR_HWPOISON (KVM_PFN_ERR_MASK + 1) 58#define KVM_PFN_ERR_RO_FAULT (KVM_PFN_ERR_MASK + 2) 59 60/* 61 * error pfns indicate that the gfn is in slot but faild to 62 * translate it to pfn on host. 63 */ 64static inline bool is_error_pfn(pfn_t pfn) 65{ 66 return !!(pfn & KVM_PFN_ERR_MASK); 67} 68 69/* 70 * error_noslot pfns indicate that the gfn can not be 71 * translated to pfn - it is not in slot or failed to 72 * translate it to pfn. 73 */ 74static inline bool is_error_noslot_pfn(pfn_t pfn) 75{ 76 return !!(pfn & KVM_PFN_ERR_NOSLOT_MASK); 77} 78 79/* noslot pfn indicates that the gfn is not in slot. */ 80static inline bool is_noslot_pfn(pfn_t pfn) 81{ 82 return pfn == KVM_PFN_NOSLOT; 83} 84 85/* 86 * architectures with KVM_HVA_ERR_BAD other than PAGE_OFFSET (e.g. s390) 87 * provide own defines and kvm_is_error_hva 88 */ 89#ifndef KVM_HVA_ERR_BAD 90 91#define KVM_HVA_ERR_BAD (PAGE_OFFSET) 92#define KVM_HVA_ERR_RO_BAD (PAGE_OFFSET + PAGE_SIZE) 93 94static inline bool kvm_is_error_hva(unsigned long addr) 95{ 96 return addr >= PAGE_OFFSET; 97} 98 99#endif 100 101#define KVM_ERR_PTR_BAD_PAGE (ERR_PTR(-ENOENT)) 102 103static inline bool is_error_page(struct page *page) 104{ 105 return IS_ERR(page); 106} 107 108/* 109 * vcpu->requests bit members 110 */ 111#define KVM_REQ_TLB_FLUSH 0 112#define KVM_REQ_MIGRATE_TIMER 1 113#define KVM_REQ_REPORT_TPR_ACCESS 2 114#define KVM_REQ_MMU_RELOAD 3 115#define KVM_REQ_TRIPLE_FAULT 4 116#define KVM_REQ_PENDING_TIMER 5 117#define KVM_REQ_UNHALT 6 118#define KVM_REQ_MMU_SYNC 7 119#define KVM_REQ_CLOCK_UPDATE 8 120#define KVM_REQ_KICK 9 121#define KVM_REQ_DEACTIVATE_FPU 10 122#define KVM_REQ_EVENT 11 123#define KVM_REQ_APF_HALT 12 124#define KVM_REQ_STEAL_UPDATE 13 125#define KVM_REQ_NMI 14 126#define KVM_REQ_PMU 15 127#define KVM_REQ_PMI 16 128#define KVM_REQ_WATCHDOG 17 129#define KVM_REQ_MASTERCLOCK_UPDATE 18 130#define KVM_REQ_MCLOCK_INPROGRESS 19 131#define KVM_REQ_EPR_EXIT 20 132#define KVM_REQ_SCAN_IOAPIC 21 133#define KVM_REQ_GLOBAL_CLOCK_UPDATE 22 134#define KVM_REQ_ENABLE_IBS 23 135#define KVM_REQ_DISABLE_IBS 24 136#define KVM_REQ_APIC_PAGE_RELOAD 25 137 138#define KVM_USERSPACE_IRQ_SOURCE_ID 0 139#define KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID 1 140 141extern struct kmem_cache *kvm_vcpu_cache; 142 143extern spinlock_t kvm_lock; 144extern struct list_head vm_list; 145 146struct kvm_io_range { 147 gpa_t addr; 148 int len; 149 struct kvm_io_device *dev; 150}; 151 152#define NR_IOBUS_DEVS 1000 153 154struct kvm_io_bus { 155 int dev_count; 156 int ioeventfd_count; 157 struct kvm_io_range range[]; 158}; 159 160enum kvm_bus { 161 KVM_MMIO_BUS, 162 KVM_PIO_BUS, 163 KVM_VIRTIO_CCW_NOTIFY_BUS, 164 KVM_FAST_MMIO_BUS, 165 KVM_NR_BUSES 166}; 167 168int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr, 169 int len, const void *val); 170int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, 171 gpa_t addr, int len, const void *val, long cookie); 172int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr, 173 int len, void *val); 174int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr, 175 int len, struct kvm_io_device *dev); 176int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx, 177 struct kvm_io_device *dev); 178 179#ifdef CONFIG_KVM_ASYNC_PF 180struct kvm_async_pf { 181 struct work_struct work; 182 struct list_head link; 183 struct list_head queue; 184 struct kvm_vcpu *vcpu; 185 struct mm_struct *mm; 186 gva_t gva; 187 unsigned long addr; 188 struct kvm_arch_async_pf arch; 189 bool wakeup_all; 190}; 191 192void kvm_clear_async_pf_completion_queue(struct kvm_vcpu *vcpu); 193void kvm_check_async_pf_completion(struct kvm_vcpu *vcpu); 194int kvm_setup_async_pf(struct kvm_vcpu *vcpu, gva_t gva, unsigned long hva, 195 struct kvm_arch_async_pf *arch); 196int kvm_async_pf_wakeup_all(struct kvm_vcpu *vcpu); 197#endif 198 199enum { 200 OUTSIDE_GUEST_MODE, 201 IN_GUEST_MODE, 202 EXITING_GUEST_MODE, 203 READING_SHADOW_PAGE_TABLES, 204}; 205 206/* 207 * Sometimes a large or cross-page mmio needs to be broken up into separate 208 * exits for userspace servicing. 209 */ 210struct kvm_mmio_fragment { 211 gpa_t gpa; 212 void *data; 213 unsigned len; 214}; 215 216struct kvm_vcpu { 217 struct kvm *kvm; 218#ifdef CONFIG_PREEMPT_NOTIFIERS 219 struct preempt_notifier preempt_notifier; 220#endif 221 int cpu; 222 int vcpu_id; 223 int srcu_idx; 224 int mode; 225 unsigned long requests; 226 unsigned long guest_debug; 227 228 struct mutex mutex; 229 struct kvm_run *run; 230 231 int fpu_active; 232 int guest_fpu_loaded, guest_xcr0_loaded; 233 wait_queue_head_t wq; 234 struct pid *pid; 235 int sigset_active; 236 sigset_t sigset; 237 struct kvm_vcpu_stat stat; 238 239#ifdef CONFIG_HAS_IOMEM 240 int mmio_needed; 241 int mmio_read_completed; 242 int mmio_is_write; 243 int mmio_cur_fragment; 244 int mmio_nr_fragments; 245 struct kvm_mmio_fragment mmio_fragments[KVM_MAX_MMIO_FRAGMENTS]; 246#endif 247 248#ifdef CONFIG_KVM_ASYNC_PF 249 struct { 250 u32 queued; 251 struct list_head queue; 252 struct list_head done; 253 spinlock_t lock; 254 } async_pf; 255#endif 256 257#ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT 258 /* 259 * Cpu relax intercept or pause loop exit optimization 260 * in_spin_loop: set when a vcpu does a pause loop exit 261 * or cpu relax intercepted. 262 * dy_eligible: indicates whether vcpu is eligible for directed yield. 263 */ 264 struct { 265 bool in_spin_loop; 266 bool dy_eligible; 267 } spin_loop; 268#endif 269 bool preempted; 270 struct kvm_vcpu_arch arch; 271}; 272 273static inline int kvm_vcpu_exiting_guest_mode(struct kvm_vcpu *vcpu) 274{ 275 return cmpxchg(&vcpu->mode, IN_GUEST_MODE, EXITING_GUEST_MODE); 276} 277 278/* 279 * Some of the bitops functions do not support too long bitmaps. 280 * This number must be determined not to exceed such limits. 281 */ 282#define KVM_MEM_MAX_NR_PAGES ((1UL << 31) - 1) 283 284struct kvm_memory_slot { 285 gfn_t base_gfn; 286 unsigned long npages; 287 unsigned long *dirty_bitmap; 288 struct kvm_arch_memory_slot arch; 289 unsigned long userspace_addr; 290 u32 flags; 291 short id; 292}; 293 294static inline unsigned long kvm_dirty_bitmap_bytes(struct kvm_memory_slot *memslot) 295{ 296 return ALIGN(memslot->npages, BITS_PER_LONG) / 8; 297} 298 299struct kvm_s390_adapter_int { 300 u64 ind_addr; 301 u64 summary_addr; 302 u64 ind_offset; 303 u32 summary_offset; 304 u32 adapter_id; 305}; 306 307struct kvm_kernel_irq_routing_entry { 308 u32 gsi; 309 u32 type; 310 int (*set)(struct kvm_kernel_irq_routing_entry *e, 311 struct kvm *kvm, int irq_source_id, int level, 312 bool line_status); 313 union { 314 struct { 315 unsigned irqchip; 316 unsigned pin; 317 } irqchip; 318 struct msi_msg msi; 319 struct kvm_s390_adapter_int adapter; 320 }; 321 struct hlist_node link; 322}; 323 324#ifndef KVM_PRIVATE_MEM_SLOTS 325#define KVM_PRIVATE_MEM_SLOTS 0 326#endif 327 328#ifndef KVM_MEM_SLOTS_NUM 329#define KVM_MEM_SLOTS_NUM (KVM_USER_MEM_SLOTS + KVM_PRIVATE_MEM_SLOTS) 330#endif 331 332/* 333 * Note: 334 * memslots are not sorted by id anymore, please use id_to_memslot() 335 * to get the memslot by its id. 336 */ 337struct kvm_memslots { 338 u64 generation; 339 struct kvm_memory_slot memslots[KVM_MEM_SLOTS_NUM]; 340 /* The mapping table from slot id to the index in memslots[]. */ 341 short id_to_index[KVM_MEM_SLOTS_NUM]; 342 atomic_t lru_slot; 343 int used_slots; 344}; 345 346struct kvm { 347 spinlock_t mmu_lock; 348 struct mutex slots_lock; 349 struct mm_struct *mm; /* userspace tied to this vm */ 350 struct kvm_memslots *memslots; 351 struct srcu_struct srcu; 352 struct srcu_struct irq_srcu; 353#ifdef CONFIG_KVM_APIC_ARCHITECTURE 354 u32 bsp_vcpu_id; 355#endif 356 struct kvm_vcpu *vcpus[KVM_MAX_VCPUS]; 357 atomic_t online_vcpus; 358 int last_boosted_vcpu; 359 struct list_head vm_list; 360 struct mutex lock; 361 struct kvm_io_bus *buses[KVM_NR_BUSES]; 362#ifdef CONFIG_HAVE_KVM_EVENTFD 363 struct { 364 spinlock_t lock; 365 struct list_head items; 366 struct list_head resampler_list; 367 struct mutex resampler_lock; 368 } irqfds; 369 struct list_head ioeventfds; 370#endif 371 struct kvm_vm_stat stat; 372 struct kvm_arch arch; 373 atomic_t users_count; 374#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET 375 struct kvm_coalesced_mmio_ring *coalesced_mmio_ring; 376 spinlock_t ring_lock; 377 struct list_head coalesced_zones; 378#endif 379 380 struct mutex irq_lock; 381#ifdef CONFIG_HAVE_KVM_IRQCHIP 382 /* 383 * Update side is protected by irq_lock. 384 */ 385 struct kvm_irq_routing_table __rcu *irq_routing; 386#endif 387#ifdef CONFIG_HAVE_KVM_IRQFD 388 struct hlist_head irq_ack_notifier_list; 389#endif 390 391#if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) 392 struct mmu_notifier mmu_notifier; 393 unsigned long mmu_notifier_seq; 394 long mmu_notifier_count; 395#endif 396 long tlbs_dirty; 397 struct list_head devices; 398}; 399 400#define kvm_err(fmt, ...) \ 401 pr_err("kvm [%i]: " fmt, task_pid_nr(current), ## __VA_ARGS__) 402#define kvm_info(fmt, ...) \ 403 pr_info("kvm [%i]: " fmt, task_pid_nr(current), ## __VA_ARGS__) 404#define kvm_debug(fmt, ...) \ 405 pr_debug("kvm [%i]: " fmt, task_pid_nr(current), ## __VA_ARGS__) 406#define kvm_pr_unimpl(fmt, ...) \ 407 pr_err_ratelimited("kvm [%i]: " fmt, \ 408 task_tgid_nr(current), ## __VA_ARGS__) 409 410/* The guest did something we don't support. */ 411#define vcpu_unimpl(vcpu, fmt, ...) \ 412 kvm_pr_unimpl("vcpu%i " fmt, (vcpu)->vcpu_id, ## __VA_ARGS__) 413 414static inline struct kvm_vcpu *kvm_get_vcpu(struct kvm *kvm, int i) 415{ 416 smp_rmb(); 417 return kvm->vcpus[i]; 418} 419 420#define kvm_for_each_vcpu(idx, vcpup, kvm) \ 421 for (idx = 0; \ 422 idx < atomic_read(&kvm->online_vcpus) && \ 423 (vcpup = kvm_get_vcpu(kvm, idx)) != NULL; \ 424 idx++) 425 426static inline struct kvm_vcpu *kvm_get_vcpu_by_id(struct kvm *kvm, int id) 427{ 428 struct kvm_vcpu *vcpu; 429 int i; 430 431 kvm_for_each_vcpu(i, vcpu, kvm) 432 if (vcpu->vcpu_id == id) 433 return vcpu; 434 return NULL; 435} 436 437#define kvm_for_each_memslot(memslot, slots) \ 438 for (memslot = &slots->memslots[0]; \ 439 memslot < slots->memslots + KVM_MEM_SLOTS_NUM && memslot->npages;\ 440 memslot++) 441 442int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id); 443void kvm_vcpu_uninit(struct kvm_vcpu *vcpu); 444 445int __must_check vcpu_load(struct kvm_vcpu *vcpu); 446void vcpu_put(struct kvm_vcpu *vcpu); 447 448#ifdef __KVM_HAVE_IOAPIC 449void kvm_vcpu_request_scan_ioapic(struct kvm *kvm); 450#else 451static inline void kvm_vcpu_request_scan_ioapic(struct kvm *kvm) 452{ 453} 454#endif 455 456#ifdef CONFIG_HAVE_KVM_IRQFD 457int kvm_irqfd_init(void); 458void kvm_irqfd_exit(void); 459#else 460static inline int kvm_irqfd_init(void) 461{ 462 return 0; 463} 464 465static inline void kvm_irqfd_exit(void) 466{ 467} 468#endif 469int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align, 470 struct module *module); 471void kvm_exit(void); 472 473void kvm_get_kvm(struct kvm *kvm); 474void kvm_put_kvm(struct kvm *kvm); 475 476static inline struct kvm_memslots *kvm_memslots(struct kvm *kvm) 477{ 478 return rcu_dereference_check(kvm->memslots, 479 srcu_read_lock_held(&kvm->srcu) 480 || lockdep_is_held(&kvm->slots_lock)); 481} 482 483static inline struct kvm_memory_slot * 484id_to_memslot(struct kvm_memslots *slots, int id) 485{ 486 int index = slots->id_to_index[id]; 487 struct kvm_memory_slot *slot; 488 489 slot = &slots->memslots[index]; 490 491 WARN_ON(slot->id != id); 492 return slot; 493} 494 495/* 496 * KVM_SET_USER_MEMORY_REGION ioctl allows the following operations: 497 * - create a new memory slot 498 * - delete an existing memory slot 499 * - modify an existing memory slot 500 * -- move it in the guest physical memory space 501 * -- just change its flags 502 * 503 * Since flags can be changed by some of these operations, the following 504 * differentiation is the best we can do for __kvm_set_memory_region(): 505 */ 506enum kvm_mr_change { 507 KVM_MR_CREATE, 508 KVM_MR_DELETE, 509 KVM_MR_MOVE, 510 KVM_MR_FLAGS_ONLY, 511}; 512 513int kvm_set_memory_region(struct kvm *kvm, 514 struct kvm_userspace_memory_region *mem); 515int __kvm_set_memory_region(struct kvm *kvm, 516 struct kvm_userspace_memory_region *mem); 517void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free, 518 struct kvm_memory_slot *dont); 519int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot, 520 unsigned long npages); 521void kvm_arch_memslots_updated(struct kvm *kvm); 522int kvm_arch_prepare_memory_region(struct kvm *kvm, 523 struct kvm_memory_slot *memslot, 524 struct kvm_userspace_memory_region *mem, 525 enum kvm_mr_change change); 526void kvm_arch_commit_memory_region(struct kvm *kvm, 527 struct kvm_userspace_memory_region *mem, 528 const struct kvm_memory_slot *old, 529 enum kvm_mr_change change); 530bool kvm_largepages_enabled(void); 531void kvm_disable_largepages(void); 532/* flush all memory translations */ 533void kvm_arch_flush_shadow_all(struct kvm *kvm); 534/* flush memory translations pointing to 'slot' */ 535void kvm_arch_flush_shadow_memslot(struct kvm *kvm, 536 struct kvm_memory_slot *slot); 537 538int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages, 539 int nr_pages); 540 541struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn); 542unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn); 543unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable); 544unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot, gfn_t gfn); 545unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot, gfn_t gfn, 546 bool *writable); 547void kvm_release_page_clean(struct page *page); 548void kvm_release_page_dirty(struct page *page); 549void kvm_set_page_accessed(struct page *page); 550 551pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn); 552pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async, 553 bool write_fault, bool *writable); 554pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn); 555pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault, 556 bool *writable); 557pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn); 558pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn); 559 560void kvm_release_pfn_clean(pfn_t pfn); 561void kvm_set_pfn_dirty(pfn_t pfn); 562void kvm_set_pfn_accessed(pfn_t pfn); 563void kvm_get_pfn(pfn_t pfn); 564 565int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset, 566 int len); 567int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data, 568 unsigned long len); 569int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len); 570int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, 571 void *data, unsigned long len); 572int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data, 573 int offset, int len); 574int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data, 575 unsigned long len); 576int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, 577 void *data, unsigned long len); 578int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc, 579 gpa_t gpa, unsigned long len); 580int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len); 581int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len); 582struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn); 583int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn); 584unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn); 585void mark_page_dirty(struct kvm *kvm, gfn_t gfn); 586 587void kvm_vcpu_block(struct kvm_vcpu *vcpu); 588void kvm_vcpu_kick(struct kvm_vcpu *vcpu); 589int kvm_vcpu_yield_to(struct kvm_vcpu *target); 590void kvm_vcpu_on_spin(struct kvm_vcpu *vcpu); 591void kvm_load_guest_fpu(struct kvm_vcpu *vcpu); 592void kvm_put_guest_fpu(struct kvm_vcpu *vcpu); 593 594void kvm_flush_remote_tlbs(struct kvm *kvm); 595void kvm_reload_remote_mmus(struct kvm *kvm); 596void kvm_make_mclock_inprogress_request(struct kvm *kvm); 597void kvm_make_scan_ioapic_request(struct kvm *kvm); 598bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req); 599 600long kvm_arch_dev_ioctl(struct file *filp, 601 unsigned int ioctl, unsigned long arg); 602long kvm_arch_vcpu_ioctl(struct file *filp, 603 unsigned int ioctl, unsigned long arg); 604int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf); 605 606int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext); 607 608int kvm_get_dirty_log(struct kvm *kvm, 609 struct kvm_dirty_log *log, int *is_dirty); 610 611int kvm_get_dirty_log_protect(struct kvm *kvm, 612 struct kvm_dirty_log *log, bool *is_dirty); 613 614void kvm_arch_mmu_enable_log_dirty_pt_masked(struct kvm *kvm, 615 struct kvm_memory_slot *slot, 616 gfn_t gfn_offset, 617 unsigned long mask); 618 619int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, 620 struct kvm_dirty_log *log); 621 622int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level, 623 bool line_status); 624long kvm_arch_vm_ioctl(struct file *filp, 625 unsigned int ioctl, unsigned long arg); 626 627int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu); 628int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu); 629 630int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu, 631 struct kvm_translation *tr); 632 633int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs); 634int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs); 635int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu, 636 struct kvm_sregs *sregs); 637int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu, 638 struct kvm_sregs *sregs); 639int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu, 640 struct kvm_mp_state *mp_state); 641int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu, 642 struct kvm_mp_state *mp_state); 643int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu, 644 struct kvm_guest_debug *dbg); 645int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run); 646 647int kvm_arch_init(void *opaque); 648void kvm_arch_exit(void); 649 650int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu); 651void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu); 652 653void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu); 654 655void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu); 656void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu); 657void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu); 658struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id); 659int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu); 660void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu); 661void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu); 662 663int kvm_arch_hardware_enable(void); 664void kvm_arch_hardware_disable(void); 665int kvm_arch_hardware_setup(void); 666void kvm_arch_hardware_unsetup(void); 667void kvm_arch_check_processor_compat(void *rtn); 668int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu); 669int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu); 670 671void *kvm_kvzalloc(unsigned long size); 672 673#ifndef __KVM_HAVE_ARCH_VM_ALLOC 674static inline struct kvm *kvm_arch_alloc_vm(void) 675{ 676 return kzalloc(sizeof(struct kvm), GFP_KERNEL); 677} 678 679static inline void kvm_arch_free_vm(struct kvm *kvm) 680{ 681 kfree(kvm); 682} 683#endif 684 685#ifdef __KVM_HAVE_ARCH_NONCOHERENT_DMA 686void kvm_arch_register_noncoherent_dma(struct kvm *kvm); 687void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm); 688bool kvm_arch_has_noncoherent_dma(struct kvm *kvm); 689#else 690static inline void kvm_arch_register_noncoherent_dma(struct kvm *kvm) 691{ 692} 693 694static inline void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm) 695{ 696} 697 698static inline bool kvm_arch_has_noncoherent_dma(struct kvm *kvm) 699{ 700 return false; 701} 702#endif 703 704static inline wait_queue_head_t *kvm_arch_vcpu_wq(struct kvm_vcpu *vcpu) 705{ 706#ifdef __KVM_HAVE_ARCH_WQP 707 return vcpu->arch.wqp; 708#else 709 return &vcpu->wq; 710#endif 711} 712 713#ifdef __KVM_HAVE_ARCH_INTC_INITIALIZED 714/* 715 * returns true if the virtual interrupt controller is initialized and 716 * ready to accept virtual IRQ. On some architectures the virtual interrupt 717 * controller is dynamically instantiated and this is not always true. 718 */ 719bool kvm_arch_intc_initialized(struct kvm *kvm); 720#else 721static inline bool kvm_arch_intc_initialized(struct kvm *kvm) 722{ 723 return true; 724} 725#endif 726 727int kvm_arch_init_vm(struct kvm *kvm, unsigned long type); 728void kvm_arch_destroy_vm(struct kvm *kvm); 729void kvm_arch_sync_events(struct kvm *kvm); 730 731int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu); 732void kvm_vcpu_kick(struct kvm_vcpu *vcpu); 733 734bool kvm_is_reserved_pfn(pfn_t pfn); 735 736struct kvm_irq_ack_notifier { 737 struct hlist_node link; 738 unsigned gsi; 739 void (*irq_acked)(struct kvm_irq_ack_notifier *kian); 740}; 741 742int kvm_irq_map_gsi(struct kvm *kvm, 743 struct kvm_kernel_irq_routing_entry *entries, int gsi); 744int kvm_irq_map_chip_pin(struct kvm *kvm, unsigned irqchip, unsigned pin); 745 746int kvm_set_irq(struct kvm *kvm, int irq_source_id, u32 irq, int level, 747 bool line_status); 748int kvm_set_irq_inatomic(struct kvm *kvm, int irq_source_id, u32 irq, int level); 749int kvm_set_msi(struct kvm_kernel_irq_routing_entry *irq_entry, struct kvm *kvm, 750 int irq_source_id, int level, bool line_status); 751bool kvm_irq_has_notifier(struct kvm *kvm, unsigned irqchip, unsigned pin); 752void kvm_notify_acked_irq(struct kvm *kvm, unsigned irqchip, unsigned pin); 753void kvm_register_irq_ack_notifier(struct kvm *kvm, 754 struct kvm_irq_ack_notifier *kian); 755void kvm_unregister_irq_ack_notifier(struct kvm *kvm, 756 struct kvm_irq_ack_notifier *kian); 757int kvm_request_irq_source_id(struct kvm *kvm); 758void kvm_free_irq_source_id(struct kvm *kvm, int irq_source_id); 759 760#ifdef CONFIG_KVM_DEVICE_ASSIGNMENT 761int kvm_iommu_map_pages(struct kvm *kvm, struct kvm_memory_slot *slot); 762void kvm_iommu_unmap_pages(struct kvm *kvm, struct kvm_memory_slot *slot); 763#else 764static inline int kvm_iommu_map_pages(struct kvm *kvm, 765 struct kvm_memory_slot *slot) 766{ 767 return 0; 768} 769 770static inline void kvm_iommu_unmap_pages(struct kvm *kvm, 771 struct kvm_memory_slot *slot) 772{ 773} 774#endif 775 776static inline void kvm_guest_enter(void) 777{ 778 unsigned long flags; 779 780 BUG_ON(preemptible()); 781 782 local_irq_save(flags); 783 guest_enter(); 784 local_irq_restore(flags); 785 786 /* KVM does not hold any references to rcu protected data when it 787 * switches CPU into a guest mode. In fact switching to a guest mode 788 * is very similar to exiting to userspace from rcu point of view. In 789 * addition CPU may stay in a guest mode for quite a long time (up to 790 * one time slice). Lets treat guest mode as quiescent state, just like 791 * we do with user-mode execution. 792 */ 793 if (!context_tracking_cpu_is_enabled()) 794 rcu_virt_note_context_switch(smp_processor_id()); 795} 796 797static inline void kvm_guest_exit(void) 798{ 799 unsigned long flags; 800 801 local_irq_save(flags); 802 guest_exit(); 803 local_irq_restore(flags); 804} 805 806/* 807 * search_memslots() and __gfn_to_memslot() are here because they are 808 * used in non-modular code in arch/powerpc/kvm/book3s_hv_rm_mmu.c. 809 * gfn_to_memslot() itself isn't here as an inline because that would 810 * bloat other code too much. 811 */ 812static inline struct kvm_memory_slot * 813search_memslots(struct kvm_memslots *slots, gfn_t gfn) 814{ 815 int start = 0, end = slots->used_slots; 816 int slot = atomic_read(&slots->lru_slot); 817 struct kvm_memory_slot *memslots = slots->memslots; 818 819 if (gfn >= memslots[slot].base_gfn && 820 gfn < memslots[slot].base_gfn + memslots[slot].npages) 821 return &memslots[slot]; 822 823 while (start < end) { 824 slot = start + (end - start) / 2; 825 826 if (gfn >= memslots[slot].base_gfn) 827 end = slot; 828 else 829 start = slot + 1; 830 } 831 832 if (gfn >= memslots[start].base_gfn && 833 gfn < memslots[start].base_gfn + memslots[start].npages) { 834 atomic_set(&slots->lru_slot, start); 835 return &memslots[start]; 836 } 837 838 return NULL; 839} 840 841static inline struct kvm_memory_slot * 842__gfn_to_memslot(struct kvm_memslots *slots, gfn_t gfn) 843{ 844 return search_memslots(slots, gfn); 845} 846 847static inline unsigned long 848__gfn_to_hva_memslot(struct kvm_memory_slot *slot, gfn_t gfn) 849{ 850 return slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE; 851} 852 853static inline int memslot_id(struct kvm *kvm, gfn_t gfn) 854{ 855 return gfn_to_memslot(kvm, gfn)->id; 856} 857 858static inline gfn_t 859hva_to_gfn_memslot(unsigned long hva, struct kvm_memory_slot *slot) 860{ 861 gfn_t gfn_offset = (hva - slot->userspace_addr) >> PAGE_SHIFT; 862 863 return slot->base_gfn + gfn_offset; 864} 865 866static inline gpa_t gfn_to_gpa(gfn_t gfn) 867{ 868 return (gpa_t)gfn << PAGE_SHIFT; 869} 870 871static inline gfn_t gpa_to_gfn(gpa_t gpa) 872{ 873 return (gfn_t)(gpa >> PAGE_SHIFT); 874} 875 876static inline hpa_t pfn_to_hpa(pfn_t pfn) 877{ 878 return (hpa_t)pfn << PAGE_SHIFT; 879} 880 881static inline bool kvm_is_error_gpa(struct kvm *kvm, gpa_t gpa) 882{ 883 unsigned long hva = gfn_to_hva(kvm, gpa_to_gfn(gpa)); 884 885 return kvm_is_error_hva(hva); 886} 887 888static inline void kvm_migrate_timers(struct kvm_vcpu *vcpu) 889{ 890 set_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests); 891} 892 893enum kvm_stat_kind { 894 KVM_STAT_VM, 895 KVM_STAT_VCPU, 896}; 897 898struct kvm_stats_debugfs_item { 899 const char *name; 900 int offset; 901 enum kvm_stat_kind kind; 902 struct dentry *dentry; 903}; 904extern struct kvm_stats_debugfs_item debugfs_entries[]; 905extern struct dentry *kvm_debugfs_dir; 906 907#if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) 908static inline int mmu_notifier_retry(struct kvm *kvm, unsigned long mmu_seq) 909{ 910 if (unlikely(kvm->mmu_notifier_count)) 911 return 1; 912 /* 913 * Ensure the read of mmu_notifier_count happens before the read 914 * of mmu_notifier_seq. This interacts with the smp_wmb() in 915 * mmu_notifier_invalidate_range_end to make sure that the caller 916 * either sees the old (non-zero) value of mmu_notifier_count or 917 * the new (incremented) value of mmu_notifier_seq. 918 * PowerPC Book3s HV KVM calls this under a per-page lock 919 * rather than under kvm->mmu_lock, for scalability, so 920 * can't rely on kvm->mmu_lock to keep things ordered. 921 */ 922 smp_rmb(); 923 if (kvm->mmu_notifier_seq != mmu_seq) 924 return 1; 925 return 0; 926} 927#endif 928 929#ifdef CONFIG_HAVE_KVM_IRQ_ROUTING 930 931#ifdef CONFIG_S390 932#define KVM_MAX_IRQ_ROUTES 4096 //FIXME: we can have more than that... 933#else 934#define KVM_MAX_IRQ_ROUTES 1024 935#endif 936 937int kvm_setup_default_irq_routing(struct kvm *kvm); 938int kvm_set_irq_routing(struct kvm *kvm, 939 const struct kvm_irq_routing_entry *entries, 940 unsigned nr, 941 unsigned flags); 942int kvm_set_routing_entry(struct kvm_kernel_irq_routing_entry *e, 943 const struct kvm_irq_routing_entry *ue); 944void kvm_free_irq_routing(struct kvm *kvm); 945 946#else 947 948static inline void kvm_free_irq_routing(struct kvm *kvm) {} 949 950#endif 951 952int kvm_send_userspace_msi(struct kvm *kvm, struct kvm_msi *msi); 953 954#ifdef CONFIG_HAVE_KVM_EVENTFD 955 956void kvm_eventfd_init(struct kvm *kvm); 957int kvm_ioeventfd(struct kvm *kvm, struct kvm_ioeventfd *args); 958 959#ifdef CONFIG_HAVE_KVM_IRQFD 960int kvm_irqfd(struct kvm *kvm, struct kvm_irqfd *args); 961void kvm_irqfd_release(struct kvm *kvm); 962void kvm_irq_routing_update(struct kvm *); 963#else 964static inline int kvm_irqfd(struct kvm *kvm, struct kvm_irqfd *args) 965{ 966 return -EINVAL; 967} 968 969static inline void kvm_irqfd_release(struct kvm *kvm) {} 970#endif 971 972#else 973 974static inline void kvm_eventfd_init(struct kvm *kvm) {} 975 976static inline int kvm_irqfd(struct kvm *kvm, struct kvm_irqfd *args) 977{ 978 return -EINVAL; 979} 980 981static inline void kvm_irqfd_release(struct kvm *kvm) {} 982 983#ifdef CONFIG_HAVE_KVM_IRQCHIP 984static inline void kvm_irq_routing_update(struct kvm *kvm) 985{ 986} 987#endif 988 989static inline int kvm_ioeventfd(struct kvm *kvm, struct kvm_ioeventfd *args) 990{ 991 return -ENOSYS; 992} 993 994#endif /* CONFIG_HAVE_KVM_EVENTFD */ 995 996#ifdef CONFIG_KVM_APIC_ARCHITECTURE 997static inline bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu) 998{ 999 return vcpu->kvm->bsp_vcpu_id == vcpu->vcpu_id; 1000} 1001 1002static inline bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu) 1003{ 1004 return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0; 1005} 1006 1007bool kvm_vcpu_compatible(struct kvm_vcpu *vcpu); 1008 1009#else 1010 1011static inline bool kvm_vcpu_compatible(struct kvm_vcpu *vcpu) { return true; } 1012 1013#endif 1014 1015static inline void kvm_make_request(int req, struct kvm_vcpu *vcpu) 1016{ 1017 set_bit(req, &vcpu->requests); 1018} 1019 1020static inline bool kvm_check_request(int req, struct kvm_vcpu *vcpu) 1021{ 1022 if (test_bit(req, &vcpu->requests)) { 1023 clear_bit(req, &vcpu->requests); 1024 return true; 1025 } else { 1026 return false; 1027 } 1028} 1029 1030extern bool kvm_rebooting; 1031 1032struct kvm_device { 1033 struct kvm_device_ops *ops; 1034 struct kvm *kvm; 1035 void *private; 1036 struct list_head vm_node; 1037}; 1038 1039/* create, destroy, and name are mandatory */ 1040struct kvm_device_ops { 1041 const char *name; 1042 int (*create)(struct kvm_device *dev, u32 type); 1043 1044 /* 1045 * Destroy is responsible for freeing dev. 1046 * 1047 * Destroy may be called before or after destructors are called 1048 * on emulated I/O regions, depending on whether a reference is 1049 * held by a vcpu or other kvm component that gets destroyed 1050 * after the emulated I/O. 1051 */ 1052 void (*destroy)(struct kvm_device *dev); 1053 1054 int (*set_attr)(struct kvm_device *dev, struct kvm_device_attr *attr); 1055 int (*get_attr)(struct kvm_device *dev, struct kvm_device_attr *attr); 1056 int (*has_attr)(struct kvm_device *dev, struct kvm_device_attr *attr); 1057 long (*ioctl)(struct kvm_device *dev, unsigned int ioctl, 1058 unsigned long arg); 1059}; 1060 1061void kvm_device_get(struct kvm_device *dev); 1062void kvm_device_put(struct kvm_device *dev); 1063struct kvm_device *kvm_device_from_filp(struct file *filp); 1064int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type); 1065void kvm_unregister_device_ops(u32 type); 1066 1067extern struct kvm_device_ops kvm_mpic_ops; 1068extern struct kvm_device_ops kvm_xics_ops; 1069extern struct kvm_device_ops kvm_arm_vgic_v2_ops; 1070extern struct kvm_device_ops kvm_arm_vgic_v3_ops; 1071 1072#ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT 1073 1074static inline void kvm_vcpu_set_in_spin_loop(struct kvm_vcpu *vcpu, bool val) 1075{ 1076 vcpu->spin_loop.in_spin_loop = val; 1077} 1078static inline void kvm_vcpu_set_dy_eligible(struct kvm_vcpu *vcpu, bool val) 1079{ 1080 vcpu->spin_loop.dy_eligible = val; 1081} 1082 1083#else /* !CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT */ 1084 1085static inline void kvm_vcpu_set_in_spin_loop(struct kvm_vcpu *vcpu, bool val) 1086{ 1087} 1088 1089static inline void kvm_vcpu_set_dy_eligible(struct kvm_vcpu *vcpu, bool val) 1090{ 1091} 1092#endif /* CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT */ 1093#endif 1094 1095