| 1 | /* |
| 2 | * Kernel-based Virtual Machine driver for Linux |
| 3 | * |
| 4 | * This module enables machines with Intel VT-x extensions to run virtual |
| 5 | * machines without emulation or binary translation. |
| 6 | * |
| 7 | * Copyright (C) 2006 Qumranet, Inc. |
| 8 | * Copyright 2010 Red Hat, Inc. and/or its affiliates. |
| 9 | * |
| 10 | * Authors: |
| 11 | * Avi Kivity <avi@qumranet.com> |
| 12 | * Yaniv Kamay <yaniv@qumranet.com> |
| 13 | * |
| 14 | * This work is licensed under the terms of the GNU GPL, version 2. See |
| 15 | * the COPYING file in the top-level directory. |
| 16 | * |
| 17 | */ |
| 18 | |
| 19 | #include <kvm/iodev.h> |
| 20 | |
| 21 | #include <linux/kvm_host.h> |
| 22 | #include <linux/kvm.h> |
| 23 | #include <linux/module.h> |
| 24 | #include <linux/errno.h> |
| 25 | #include <linux/percpu.h> |
| 26 | #include <linux/mm.h> |
| 27 | #include <linux/miscdevice.h> |
| 28 | #include <linux/vmalloc.h> |
| 29 | #include <linux/reboot.h> |
| 30 | #include <linux/debugfs.h> |
| 31 | #include <linux/highmem.h> |
| 32 | #include <linux/file.h> |
| 33 | #include <linux/syscore_ops.h> |
| 34 | #include <linux/cpu.h> |
| 35 | #include <linux/sched.h> |
| 36 | #include <linux/cpumask.h> |
| 37 | #include <linux/smp.h> |
| 38 | #include <linux/anon_inodes.h> |
| 39 | #include <linux/profile.h> |
| 40 | #include <linux/kvm_para.h> |
| 41 | #include <linux/pagemap.h> |
| 42 | #include <linux/mman.h> |
| 43 | #include <linux/swap.h> |
| 44 | #include <linux/bitops.h> |
| 45 | #include <linux/spinlock.h> |
| 46 | #include <linux/compat.h> |
| 47 | #include <linux/srcu.h> |
| 48 | #include <linux/hugetlb.h> |
| 49 | #include <linux/slab.h> |
| 50 | #include <linux/sort.h> |
| 51 | #include <linux/bsearch.h> |
| 52 | |
| 53 | #include <asm/processor.h> |
| 54 | #include <asm/io.h> |
| 55 | #include <asm/ioctl.h> |
| 56 | #include <asm/uaccess.h> |
| 57 | #include <asm/pgtable.h> |
| 58 | |
| 59 | #include "coalesced_mmio.h" |
| 60 | #include "async_pf.h" |
| 61 | #include "vfio.h" |
| 62 | |
| 63 | #define CREATE_TRACE_POINTS |
| 64 | #include <trace/events/kvm.h> |
| 65 | |
| 66 | MODULE_AUTHOR("Qumranet"); |
| 67 | MODULE_LICENSE("GPL"); |
| 68 | |
| 69 | static unsigned int halt_poll_ns; |
| 70 | module_param(halt_poll_ns, uint, S_IRUGO | S_IWUSR); |
| 71 | |
| 72 | /* |
| 73 | * Ordering of locks: |
| 74 | * |
| 75 | * kvm->lock --> kvm->slots_lock --> kvm->irq_lock |
| 76 | */ |
| 77 | |
| 78 | DEFINE_SPINLOCK(kvm_lock); |
| 79 | static DEFINE_RAW_SPINLOCK(kvm_count_lock); |
| 80 | LIST_HEAD(vm_list); |
| 81 | |
| 82 | static cpumask_var_t cpus_hardware_enabled; |
| 83 | static int kvm_usage_count; |
| 84 | static atomic_t hardware_enable_failed; |
| 85 | |
| 86 | struct kmem_cache *kvm_vcpu_cache; |
| 87 | EXPORT_SYMBOL_GPL(kvm_vcpu_cache); |
| 88 | |
| 89 | static __read_mostly struct preempt_ops kvm_preempt_ops; |
| 90 | |
| 91 | struct dentry *kvm_debugfs_dir; |
| 92 | EXPORT_SYMBOL_GPL(kvm_debugfs_dir); |
| 93 | |
| 94 | static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl, |
| 95 | unsigned long arg); |
| 96 | #ifdef CONFIG_KVM_COMPAT |
| 97 | static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl, |
| 98 | unsigned long arg); |
| 99 | #endif |
| 100 | static int hardware_enable_all(void); |
| 101 | static void hardware_disable_all(void); |
| 102 | |
| 103 | static void kvm_io_bus_destroy(struct kvm_io_bus *bus); |
| 104 | |
| 105 | static void kvm_release_pfn_dirty(pfn_t pfn); |
| 106 | static void mark_page_dirty_in_slot(struct kvm *kvm, |
| 107 | struct kvm_memory_slot *memslot, gfn_t gfn); |
| 108 | |
| 109 | __visible bool kvm_rebooting; |
| 110 | EXPORT_SYMBOL_GPL(kvm_rebooting); |
| 111 | |
| 112 | static bool largepages_enabled = true; |
| 113 | |
| 114 | bool kvm_is_reserved_pfn(pfn_t pfn) |
| 115 | { |
| 116 | if (pfn_valid(pfn)) |
| 117 | return PageReserved(pfn_to_page(pfn)); |
| 118 | |
| 119 | return true; |
| 120 | } |
| 121 | |
| 122 | /* |
| 123 | * Switches to specified vcpu, until a matching vcpu_put() |
| 124 | */ |
| 125 | int vcpu_load(struct kvm_vcpu *vcpu) |
| 126 | { |
| 127 | int cpu; |
| 128 | |
| 129 | if (mutex_lock_killable(&vcpu->mutex)) |
| 130 | return -EINTR; |
| 131 | cpu = get_cpu(); |
| 132 | preempt_notifier_register(&vcpu->preempt_notifier); |
| 133 | kvm_arch_vcpu_load(vcpu, cpu); |
| 134 | put_cpu(); |
| 135 | return 0; |
| 136 | } |
| 137 | |
| 138 | void vcpu_put(struct kvm_vcpu *vcpu) |
| 139 | { |
| 140 | preempt_disable(); |
| 141 | kvm_arch_vcpu_put(vcpu); |
| 142 | preempt_notifier_unregister(&vcpu->preempt_notifier); |
| 143 | preempt_enable(); |
| 144 | mutex_unlock(&vcpu->mutex); |
| 145 | } |
| 146 | |
| 147 | static void ack_flush(void *_completed) |
| 148 | { |
| 149 | } |
| 150 | |
| 151 | bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req) |
| 152 | { |
| 153 | int i, cpu, me; |
| 154 | cpumask_var_t cpus; |
| 155 | bool called = true; |
| 156 | struct kvm_vcpu *vcpu; |
| 157 | |
| 158 | zalloc_cpumask_var(&cpus, GFP_ATOMIC); |
| 159 | |
| 160 | me = get_cpu(); |
| 161 | kvm_for_each_vcpu(i, vcpu, kvm) { |
| 162 | kvm_make_request(req, vcpu); |
| 163 | cpu = vcpu->cpu; |
| 164 | |
| 165 | /* Set ->requests bit before we read ->mode */ |
| 166 | smp_mb(); |
| 167 | |
| 168 | if (cpus != NULL && cpu != -1 && cpu != me && |
| 169 | kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE) |
| 170 | cpumask_set_cpu(cpu, cpus); |
| 171 | } |
| 172 | if (unlikely(cpus == NULL)) |
| 173 | smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1); |
| 174 | else if (!cpumask_empty(cpus)) |
| 175 | smp_call_function_many(cpus, ack_flush, NULL, 1); |
| 176 | else |
| 177 | called = false; |
| 178 | put_cpu(); |
| 179 | free_cpumask_var(cpus); |
| 180 | return called; |
| 181 | } |
| 182 | |
| 183 | #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL |
| 184 | void kvm_flush_remote_tlbs(struct kvm *kvm) |
| 185 | { |
| 186 | long dirty_count = kvm->tlbs_dirty; |
| 187 | |
| 188 | smp_mb(); |
| 189 | if (kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH)) |
| 190 | ++kvm->stat.remote_tlb_flush; |
| 191 | cmpxchg(&kvm->tlbs_dirty, dirty_count, 0); |
| 192 | } |
| 193 | EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs); |
| 194 | #endif |
| 195 | |
| 196 | void kvm_reload_remote_mmus(struct kvm *kvm) |
| 197 | { |
| 198 | kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD); |
| 199 | } |
| 200 | |
| 201 | void kvm_make_mclock_inprogress_request(struct kvm *kvm) |
| 202 | { |
| 203 | kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS); |
| 204 | } |
| 205 | |
| 206 | void kvm_make_scan_ioapic_request(struct kvm *kvm) |
| 207 | { |
| 208 | kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC); |
| 209 | } |
| 210 | |
| 211 | int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id) |
| 212 | { |
| 213 | struct page *page; |
| 214 | int r; |
| 215 | |
| 216 | mutex_init(&vcpu->mutex); |
| 217 | vcpu->cpu = -1; |
| 218 | vcpu->kvm = kvm; |
| 219 | vcpu->vcpu_id = id; |
| 220 | vcpu->pid = NULL; |
| 221 | init_waitqueue_head(&vcpu->wq); |
| 222 | kvm_async_pf_vcpu_init(vcpu); |
| 223 | |
| 224 | page = alloc_page(GFP_KERNEL | __GFP_ZERO); |
| 225 | if (!page) { |
| 226 | r = -ENOMEM; |
| 227 | goto fail; |
| 228 | } |
| 229 | vcpu->run = page_address(page); |
| 230 | |
| 231 | kvm_vcpu_set_in_spin_loop(vcpu, false); |
| 232 | kvm_vcpu_set_dy_eligible(vcpu, false); |
| 233 | vcpu->preempted = false; |
| 234 | |
| 235 | r = kvm_arch_vcpu_init(vcpu); |
| 236 | if (r < 0) |
| 237 | goto fail_free_run; |
| 238 | return 0; |
| 239 | |
| 240 | fail_free_run: |
| 241 | free_page((unsigned long)vcpu->run); |
| 242 | fail: |
| 243 | return r; |
| 244 | } |
| 245 | EXPORT_SYMBOL_GPL(kvm_vcpu_init); |
| 246 | |
| 247 | void kvm_vcpu_uninit(struct kvm_vcpu *vcpu) |
| 248 | { |
| 249 | put_pid(vcpu->pid); |
| 250 | kvm_arch_vcpu_uninit(vcpu); |
| 251 | free_page((unsigned long)vcpu->run); |
| 252 | } |
| 253 | EXPORT_SYMBOL_GPL(kvm_vcpu_uninit); |
| 254 | |
| 255 | #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) |
| 256 | static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn) |
| 257 | { |
| 258 | return container_of(mn, struct kvm, mmu_notifier); |
| 259 | } |
| 260 | |
| 261 | static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn, |
| 262 | struct mm_struct *mm, |
| 263 | unsigned long address) |
| 264 | { |
| 265 | struct kvm *kvm = mmu_notifier_to_kvm(mn); |
| 266 | int need_tlb_flush, idx; |
| 267 | |
| 268 | /* |
| 269 | * When ->invalidate_page runs, the linux pte has been zapped |
| 270 | * already but the page is still allocated until |
| 271 | * ->invalidate_page returns. So if we increase the sequence |
| 272 | * here the kvm page fault will notice if the spte can't be |
| 273 | * established because the page is going to be freed. If |
| 274 | * instead the kvm page fault establishes the spte before |
| 275 | * ->invalidate_page runs, kvm_unmap_hva will release it |
| 276 | * before returning. |
| 277 | * |
| 278 | * The sequence increase only need to be seen at spin_unlock |
| 279 | * time, and not at spin_lock time. |
| 280 | * |
| 281 | * Increasing the sequence after the spin_unlock would be |
| 282 | * unsafe because the kvm page fault could then establish the |
| 283 | * pte after kvm_unmap_hva returned, without noticing the page |
| 284 | * is going to be freed. |
| 285 | */ |
| 286 | idx = srcu_read_lock(&kvm->srcu); |
| 287 | spin_lock(&kvm->mmu_lock); |
| 288 | |
| 289 | kvm->mmu_notifier_seq++; |
| 290 | need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty; |
| 291 | /* we've to flush the tlb before the pages can be freed */ |
| 292 | if (need_tlb_flush) |
| 293 | kvm_flush_remote_tlbs(kvm); |
| 294 | |
| 295 | spin_unlock(&kvm->mmu_lock); |
| 296 | |
| 297 | kvm_arch_mmu_notifier_invalidate_page(kvm, address); |
| 298 | |
| 299 | srcu_read_unlock(&kvm->srcu, idx); |
| 300 | } |
| 301 | |
| 302 | static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn, |
| 303 | struct mm_struct *mm, |
| 304 | unsigned long address, |
| 305 | pte_t pte) |
| 306 | { |
| 307 | struct kvm *kvm = mmu_notifier_to_kvm(mn); |
| 308 | int idx; |
| 309 | |
| 310 | idx = srcu_read_lock(&kvm->srcu); |
| 311 | spin_lock(&kvm->mmu_lock); |
| 312 | kvm->mmu_notifier_seq++; |
| 313 | kvm_set_spte_hva(kvm, address, pte); |
| 314 | spin_unlock(&kvm->mmu_lock); |
| 315 | srcu_read_unlock(&kvm->srcu, idx); |
| 316 | } |
| 317 | |
| 318 | static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn, |
| 319 | struct mm_struct *mm, |
| 320 | unsigned long start, |
| 321 | unsigned long end) |
| 322 | { |
| 323 | struct kvm *kvm = mmu_notifier_to_kvm(mn); |
| 324 | int need_tlb_flush = 0, idx; |
| 325 | |
| 326 | idx = srcu_read_lock(&kvm->srcu); |
| 327 | spin_lock(&kvm->mmu_lock); |
| 328 | /* |
| 329 | * The count increase must become visible at unlock time as no |
| 330 | * spte can be established without taking the mmu_lock and |
| 331 | * count is also read inside the mmu_lock critical section. |
| 332 | */ |
| 333 | kvm->mmu_notifier_count++; |
| 334 | need_tlb_flush = kvm_unmap_hva_range(kvm, start, end); |
| 335 | need_tlb_flush |= kvm->tlbs_dirty; |
| 336 | /* we've to flush the tlb before the pages can be freed */ |
| 337 | if (need_tlb_flush) |
| 338 | kvm_flush_remote_tlbs(kvm); |
| 339 | |
| 340 | spin_unlock(&kvm->mmu_lock); |
| 341 | srcu_read_unlock(&kvm->srcu, idx); |
| 342 | } |
| 343 | |
| 344 | static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn, |
| 345 | struct mm_struct *mm, |
| 346 | unsigned long start, |
| 347 | unsigned long end) |
| 348 | { |
| 349 | struct kvm *kvm = mmu_notifier_to_kvm(mn); |
| 350 | |
| 351 | spin_lock(&kvm->mmu_lock); |
| 352 | /* |
| 353 | * This sequence increase will notify the kvm page fault that |
| 354 | * the page that is going to be mapped in the spte could have |
| 355 | * been freed. |
| 356 | */ |
| 357 | kvm->mmu_notifier_seq++; |
| 358 | smp_wmb(); |
| 359 | /* |
| 360 | * The above sequence increase must be visible before the |
| 361 | * below count decrease, which is ensured by the smp_wmb above |
| 362 | * in conjunction with the smp_rmb in mmu_notifier_retry(). |
| 363 | */ |
| 364 | kvm->mmu_notifier_count--; |
| 365 | spin_unlock(&kvm->mmu_lock); |
| 366 | |
| 367 | BUG_ON(kvm->mmu_notifier_count < 0); |
| 368 | } |
| 369 | |
| 370 | static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn, |
| 371 | struct mm_struct *mm, |
| 372 | unsigned long start, |
| 373 | unsigned long end) |
| 374 | { |
| 375 | struct kvm *kvm = mmu_notifier_to_kvm(mn); |
| 376 | int young, idx; |
| 377 | |
| 378 | idx = srcu_read_lock(&kvm->srcu); |
| 379 | spin_lock(&kvm->mmu_lock); |
| 380 | |
| 381 | young = kvm_age_hva(kvm, start, end); |
| 382 | if (young) |
| 383 | kvm_flush_remote_tlbs(kvm); |
| 384 | |
| 385 | spin_unlock(&kvm->mmu_lock); |
| 386 | srcu_read_unlock(&kvm->srcu, idx); |
| 387 | |
| 388 | return young; |
| 389 | } |
| 390 | |
| 391 | static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn, |
| 392 | struct mm_struct *mm, |
| 393 | unsigned long address) |
| 394 | { |
| 395 | struct kvm *kvm = mmu_notifier_to_kvm(mn); |
| 396 | int young, idx; |
| 397 | |
| 398 | idx = srcu_read_lock(&kvm->srcu); |
| 399 | spin_lock(&kvm->mmu_lock); |
| 400 | young = kvm_test_age_hva(kvm, address); |
| 401 | spin_unlock(&kvm->mmu_lock); |
| 402 | srcu_read_unlock(&kvm->srcu, idx); |
| 403 | |
| 404 | return young; |
| 405 | } |
| 406 | |
| 407 | static void kvm_mmu_notifier_release(struct mmu_notifier *mn, |
| 408 | struct mm_struct *mm) |
| 409 | { |
| 410 | struct kvm *kvm = mmu_notifier_to_kvm(mn); |
| 411 | int idx; |
| 412 | |
| 413 | idx = srcu_read_lock(&kvm->srcu); |
| 414 | kvm_arch_flush_shadow_all(kvm); |
| 415 | srcu_read_unlock(&kvm->srcu, idx); |
| 416 | } |
| 417 | |
| 418 | static const struct mmu_notifier_ops kvm_mmu_notifier_ops = { |
| 419 | .invalidate_page = kvm_mmu_notifier_invalidate_page, |
| 420 | .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start, |
| 421 | .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end, |
| 422 | .clear_flush_young = kvm_mmu_notifier_clear_flush_young, |
| 423 | .test_young = kvm_mmu_notifier_test_young, |
| 424 | .change_pte = kvm_mmu_notifier_change_pte, |
| 425 | .release = kvm_mmu_notifier_release, |
| 426 | }; |
| 427 | |
| 428 | static int kvm_init_mmu_notifier(struct kvm *kvm) |
| 429 | { |
| 430 | kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops; |
| 431 | return mmu_notifier_register(&kvm->mmu_notifier, current->mm); |
| 432 | } |
| 433 | |
| 434 | #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */ |
| 435 | |
| 436 | static int kvm_init_mmu_notifier(struct kvm *kvm) |
| 437 | { |
| 438 | return 0; |
| 439 | } |
| 440 | |
| 441 | #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */ |
| 442 | |
| 443 | static struct kvm_memslots *kvm_alloc_memslots(void) |
| 444 | { |
| 445 | int i; |
| 446 | struct kvm_memslots *slots; |
| 447 | |
| 448 | slots = kvm_kvzalloc(sizeof(struct kvm_memslots)); |
| 449 | if (!slots) |
| 450 | return NULL; |
| 451 | |
| 452 | /* |
| 453 | * Init kvm generation close to the maximum to easily test the |
| 454 | * code of handling generation number wrap-around. |
| 455 | */ |
| 456 | slots->generation = -150; |
| 457 | for (i = 0; i < KVM_MEM_SLOTS_NUM; i++) |
| 458 | slots->id_to_index[i] = slots->memslots[i].id = i; |
| 459 | |
| 460 | return slots; |
| 461 | } |
| 462 | |
| 463 | static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot) |
| 464 | { |
| 465 | if (!memslot->dirty_bitmap) |
| 466 | return; |
| 467 | |
| 468 | kvfree(memslot->dirty_bitmap); |
| 469 | memslot->dirty_bitmap = NULL; |
| 470 | } |
| 471 | |
| 472 | /* |
| 473 | * Free any memory in @free but not in @dont. |
| 474 | */ |
| 475 | static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free, |
| 476 | struct kvm_memory_slot *dont) |
| 477 | { |
| 478 | if (!dont || free->dirty_bitmap != dont->dirty_bitmap) |
| 479 | kvm_destroy_dirty_bitmap(free); |
| 480 | |
| 481 | kvm_arch_free_memslot(kvm, free, dont); |
| 482 | |
| 483 | free->npages = 0; |
| 484 | } |
| 485 | |
| 486 | static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots) |
| 487 | { |
| 488 | struct kvm_memory_slot *memslot; |
| 489 | |
| 490 | if (!slots) |
| 491 | return; |
| 492 | |
| 493 | kvm_for_each_memslot(memslot, slots) |
| 494 | kvm_free_memslot(kvm, memslot, NULL); |
| 495 | |
| 496 | kvfree(slots); |
| 497 | } |
| 498 | |
| 499 | static struct kvm *kvm_create_vm(unsigned long type) |
| 500 | { |
| 501 | int r, i; |
| 502 | struct kvm *kvm = kvm_arch_alloc_vm(); |
| 503 | |
| 504 | if (!kvm) |
| 505 | return ERR_PTR(-ENOMEM); |
| 506 | |
| 507 | r = kvm_arch_init_vm(kvm, type); |
| 508 | if (r) |
| 509 | goto out_err_no_disable; |
| 510 | |
| 511 | r = hardware_enable_all(); |
| 512 | if (r) |
| 513 | goto out_err_no_disable; |
| 514 | |
| 515 | #ifdef CONFIG_HAVE_KVM_IRQFD |
| 516 | INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list); |
| 517 | #endif |
| 518 | |
| 519 | BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX); |
| 520 | |
| 521 | r = -ENOMEM; |
| 522 | kvm->memslots = kvm_alloc_memslots(); |
| 523 | if (!kvm->memslots) |
| 524 | goto out_err_no_srcu; |
| 525 | |
| 526 | if (init_srcu_struct(&kvm->srcu)) |
| 527 | goto out_err_no_srcu; |
| 528 | if (init_srcu_struct(&kvm->irq_srcu)) |
| 529 | goto out_err_no_irq_srcu; |
| 530 | for (i = 0; i < KVM_NR_BUSES; i++) { |
| 531 | kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus), |
| 532 | GFP_KERNEL); |
| 533 | if (!kvm->buses[i]) |
| 534 | goto out_err; |
| 535 | } |
| 536 | |
| 537 | spin_lock_init(&kvm->mmu_lock); |
| 538 | kvm->mm = current->mm; |
| 539 | atomic_inc(&kvm->mm->mm_count); |
| 540 | kvm_eventfd_init(kvm); |
| 541 | mutex_init(&kvm->lock); |
| 542 | mutex_init(&kvm->irq_lock); |
| 543 | mutex_init(&kvm->slots_lock); |
| 544 | atomic_set(&kvm->users_count, 1); |
| 545 | INIT_LIST_HEAD(&kvm->devices); |
| 546 | |
| 547 | r = kvm_init_mmu_notifier(kvm); |
| 548 | if (r) |
| 549 | goto out_err; |
| 550 | |
| 551 | spin_lock(&kvm_lock); |
| 552 | list_add(&kvm->vm_list, &vm_list); |
| 553 | spin_unlock(&kvm_lock); |
| 554 | |
| 555 | return kvm; |
| 556 | |
| 557 | out_err: |
| 558 | cleanup_srcu_struct(&kvm->irq_srcu); |
| 559 | out_err_no_irq_srcu: |
| 560 | cleanup_srcu_struct(&kvm->srcu); |
| 561 | out_err_no_srcu: |
| 562 | hardware_disable_all(); |
| 563 | out_err_no_disable: |
| 564 | for (i = 0; i < KVM_NR_BUSES; i++) |
| 565 | kfree(kvm->buses[i]); |
| 566 | kvm_free_memslots(kvm, kvm->memslots); |
| 567 | kvm_arch_free_vm(kvm); |
| 568 | return ERR_PTR(r); |
| 569 | } |
| 570 | |
| 571 | /* |
| 572 | * Avoid using vmalloc for a small buffer. |
| 573 | * Should not be used when the size is statically known. |
| 574 | */ |
| 575 | void *kvm_kvzalloc(unsigned long size) |
| 576 | { |
| 577 | if (size > PAGE_SIZE) |
| 578 | return vzalloc(size); |
| 579 | else |
| 580 | return kzalloc(size, GFP_KERNEL); |
| 581 | } |
| 582 | |
| 583 | static void kvm_destroy_devices(struct kvm *kvm) |
| 584 | { |
| 585 | struct list_head *node, *tmp; |
| 586 | |
| 587 | list_for_each_safe(node, tmp, &kvm->devices) { |
| 588 | struct kvm_device *dev = |
| 589 | list_entry(node, struct kvm_device, vm_node); |
| 590 | |
| 591 | list_del(node); |
| 592 | dev->ops->destroy(dev); |
| 593 | } |
| 594 | } |
| 595 | |
| 596 | static void kvm_destroy_vm(struct kvm *kvm) |
| 597 | { |
| 598 | int i; |
| 599 | struct mm_struct *mm = kvm->mm; |
| 600 | |
| 601 | kvm_arch_sync_events(kvm); |
| 602 | spin_lock(&kvm_lock); |
| 603 | list_del(&kvm->vm_list); |
| 604 | spin_unlock(&kvm_lock); |
| 605 | kvm_free_irq_routing(kvm); |
| 606 | for (i = 0; i < KVM_NR_BUSES; i++) |
| 607 | kvm_io_bus_destroy(kvm->buses[i]); |
| 608 | kvm_coalesced_mmio_free(kvm); |
| 609 | #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) |
| 610 | mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm); |
| 611 | #else |
| 612 | kvm_arch_flush_shadow_all(kvm); |
| 613 | #endif |
| 614 | kvm_arch_destroy_vm(kvm); |
| 615 | kvm_destroy_devices(kvm); |
| 616 | kvm_free_memslots(kvm, kvm->memslots); |
| 617 | cleanup_srcu_struct(&kvm->irq_srcu); |
| 618 | cleanup_srcu_struct(&kvm->srcu); |
| 619 | kvm_arch_free_vm(kvm); |
| 620 | hardware_disable_all(); |
| 621 | mmdrop(mm); |
| 622 | } |
| 623 | |
| 624 | void kvm_get_kvm(struct kvm *kvm) |
| 625 | { |
| 626 | atomic_inc(&kvm->users_count); |
| 627 | } |
| 628 | EXPORT_SYMBOL_GPL(kvm_get_kvm); |
| 629 | |
| 630 | void kvm_put_kvm(struct kvm *kvm) |
| 631 | { |
| 632 | if (atomic_dec_and_test(&kvm->users_count)) |
| 633 | kvm_destroy_vm(kvm); |
| 634 | } |
| 635 | EXPORT_SYMBOL_GPL(kvm_put_kvm); |
| 636 | |
| 637 | |
| 638 | static int kvm_vm_release(struct inode *inode, struct file *filp) |
| 639 | { |
| 640 | struct kvm *kvm = filp->private_data; |
| 641 | |
| 642 | kvm_irqfd_release(kvm); |
| 643 | |
| 644 | kvm_put_kvm(kvm); |
| 645 | return 0; |
| 646 | } |
| 647 | |
| 648 | /* |
| 649 | * Allocation size is twice as large as the actual dirty bitmap size. |
| 650 | * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed. |
| 651 | */ |
| 652 | static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot) |
| 653 | { |
| 654 | unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot); |
| 655 | |
| 656 | memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes); |
| 657 | if (!memslot->dirty_bitmap) |
| 658 | return -ENOMEM; |
| 659 | |
| 660 | return 0; |
| 661 | } |
| 662 | |
| 663 | /* |
| 664 | * Insert memslot and re-sort memslots based on their GFN, |
| 665 | * so binary search could be used to lookup GFN. |
| 666 | * Sorting algorithm takes advantage of having initially |
| 667 | * sorted array and known changed memslot position. |
| 668 | */ |
| 669 | static void update_memslots(struct kvm_memslots *slots, |
| 670 | struct kvm_memory_slot *new) |
| 671 | { |
| 672 | int id = new->id; |
| 673 | int i = slots->id_to_index[id]; |
| 674 | struct kvm_memory_slot *mslots = slots->memslots; |
| 675 | |
| 676 | WARN_ON(mslots[i].id != id); |
| 677 | if (!new->npages) { |
| 678 | WARN_ON(!mslots[i].npages); |
| 679 | new->base_gfn = 0; |
| 680 | new->flags = 0; |
| 681 | if (mslots[i].npages) |
| 682 | slots->used_slots--; |
| 683 | } else { |
| 684 | if (!mslots[i].npages) |
| 685 | slots->used_slots++; |
| 686 | } |
| 687 | |
| 688 | while (i < KVM_MEM_SLOTS_NUM - 1 && |
| 689 | new->base_gfn <= mslots[i + 1].base_gfn) { |
| 690 | if (!mslots[i + 1].npages) |
| 691 | break; |
| 692 | mslots[i] = mslots[i + 1]; |
| 693 | slots->id_to_index[mslots[i].id] = i; |
| 694 | i++; |
| 695 | } |
| 696 | |
| 697 | /* |
| 698 | * The ">=" is needed when creating a slot with base_gfn == 0, |
| 699 | * so that it moves before all those with base_gfn == npages == 0. |
| 700 | * |
| 701 | * On the other hand, if new->npages is zero, the above loop has |
| 702 | * already left i pointing to the beginning of the empty part of |
| 703 | * mslots, and the ">=" would move the hole backwards in this |
| 704 | * case---which is wrong. So skip the loop when deleting a slot. |
| 705 | */ |
| 706 | if (new->npages) { |
| 707 | while (i > 0 && |
| 708 | new->base_gfn >= mslots[i - 1].base_gfn) { |
| 709 | mslots[i] = mslots[i - 1]; |
| 710 | slots->id_to_index[mslots[i].id] = i; |
| 711 | i--; |
| 712 | } |
| 713 | } else |
| 714 | WARN_ON_ONCE(i != slots->used_slots); |
| 715 | |
| 716 | mslots[i] = *new; |
| 717 | slots->id_to_index[mslots[i].id] = i; |
| 718 | } |
| 719 | |
| 720 | static int check_memory_region_flags(struct kvm_userspace_memory_region *mem) |
| 721 | { |
| 722 | u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES; |
| 723 | |
| 724 | #ifdef __KVM_HAVE_READONLY_MEM |
| 725 | valid_flags |= KVM_MEM_READONLY; |
| 726 | #endif |
| 727 | |
| 728 | if (mem->flags & ~valid_flags) |
| 729 | return -EINVAL; |
| 730 | |
| 731 | return 0; |
| 732 | } |
| 733 | |
| 734 | static struct kvm_memslots *install_new_memslots(struct kvm *kvm, |
| 735 | struct kvm_memslots *slots) |
| 736 | { |
| 737 | struct kvm_memslots *old_memslots = kvm->memslots; |
| 738 | |
| 739 | /* |
| 740 | * Set the low bit in the generation, which disables SPTE caching |
| 741 | * until the end of synchronize_srcu_expedited. |
| 742 | */ |
| 743 | WARN_ON(old_memslots->generation & 1); |
| 744 | slots->generation = old_memslots->generation + 1; |
| 745 | |
| 746 | rcu_assign_pointer(kvm->memslots, slots); |
| 747 | synchronize_srcu_expedited(&kvm->srcu); |
| 748 | |
| 749 | /* |
| 750 | * Increment the new memslot generation a second time. This prevents |
| 751 | * vm exits that race with memslot updates from caching a memslot |
| 752 | * generation that will (potentially) be valid forever. |
| 753 | */ |
| 754 | slots->generation++; |
| 755 | |
| 756 | kvm_arch_memslots_updated(kvm); |
| 757 | |
| 758 | return old_memslots; |
| 759 | } |
| 760 | |
| 761 | /* |
| 762 | * Allocate some memory and give it an address in the guest physical address |
| 763 | * space. |
| 764 | * |
| 765 | * Discontiguous memory is allowed, mostly for framebuffers. |
| 766 | * |
| 767 | * Must be called holding kvm->slots_lock for write. |
| 768 | */ |
| 769 | int __kvm_set_memory_region(struct kvm *kvm, |
| 770 | struct kvm_userspace_memory_region *mem) |
| 771 | { |
| 772 | int r; |
| 773 | gfn_t base_gfn; |
| 774 | unsigned long npages; |
| 775 | struct kvm_memory_slot *slot; |
| 776 | struct kvm_memory_slot old, new; |
| 777 | struct kvm_memslots *slots = NULL, *old_memslots; |
| 778 | enum kvm_mr_change change; |
| 779 | |
| 780 | r = check_memory_region_flags(mem); |
| 781 | if (r) |
| 782 | goto out; |
| 783 | |
| 784 | r = -EINVAL; |
| 785 | /* General sanity checks */ |
| 786 | if (mem->memory_size & (PAGE_SIZE - 1)) |
| 787 | goto out; |
| 788 | if (mem->guest_phys_addr & (PAGE_SIZE - 1)) |
| 789 | goto out; |
| 790 | /* We can read the guest memory with __xxx_user() later on. */ |
| 791 | if ((mem->slot < KVM_USER_MEM_SLOTS) && |
| 792 | ((mem->userspace_addr & (PAGE_SIZE - 1)) || |
| 793 | !access_ok(VERIFY_WRITE, |
| 794 | (void __user *)(unsigned long)mem->userspace_addr, |
| 795 | mem->memory_size))) |
| 796 | goto out; |
| 797 | if (mem->slot >= KVM_MEM_SLOTS_NUM) |
| 798 | goto out; |
| 799 | if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr) |
| 800 | goto out; |
| 801 | |
| 802 | slot = id_to_memslot(kvm->memslots, mem->slot); |
| 803 | base_gfn = mem->guest_phys_addr >> PAGE_SHIFT; |
| 804 | npages = mem->memory_size >> PAGE_SHIFT; |
| 805 | |
| 806 | if (npages > KVM_MEM_MAX_NR_PAGES) |
| 807 | goto out; |
| 808 | |
| 809 | if (!npages) |
| 810 | mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES; |
| 811 | |
| 812 | new = old = *slot; |
| 813 | |
| 814 | new.id = mem->slot; |
| 815 | new.base_gfn = base_gfn; |
| 816 | new.npages = npages; |
| 817 | new.flags = mem->flags; |
| 818 | |
| 819 | if (npages) { |
| 820 | if (!old.npages) |
| 821 | change = KVM_MR_CREATE; |
| 822 | else { /* Modify an existing slot. */ |
| 823 | if ((mem->userspace_addr != old.userspace_addr) || |
| 824 | (npages != old.npages) || |
| 825 | ((new.flags ^ old.flags) & KVM_MEM_READONLY)) |
| 826 | goto out; |
| 827 | |
| 828 | if (base_gfn != old.base_gfn) |
| 829 | change = KVM_MR_MOVE; |
| 830 | else if (new.flags != old.flags) |
| 831 | change = KVM_MR_FLAGS_ONLY; |
| 832 | else { /* Nothing to change. */ |
| 833 | r = 0; |
| 834 | goto out; |
| 835 | } |
| 836 | } |
| 837 | } else if (old.npages) { |
| 838 | change = KVM_MR_DELETE; |
| 839 | } else /* Modify a non-existent slot: disallowed. */ |
| 840 | goto out; |
| 841 | |
| 842 | if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) { |
| 843 | /* Check for overlaps */ |
| 844 | r = -EEXIST; |
| 845 | kvm_for_each_memslot(slot, kvm->memslots) { |
| 846 | if ((slot->id >= KVM_USER_MEM_SLOTS) || |
| 847 | (slot->id == mem->slot)) |
| 848 | continue; |
| 849 | if (!((base_gfn + npages <= slot->base_gfn) || |
| 850 | (base_gfn >= slot->base_gfn + slot->npages))) |
| 851 | goto out; |
| 852 | } |
| 853 | } |
| 854 | |
| 855 | /* Free page dirty bitmap if unneeded */ |
| 856 | if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES)) |
| 857 | new.dirty_bitmap = NULL; |
| 858 | |
| 859 | r = -ENOMEM; |
| 860 | if (change == KVM_MR_CREATE) { |
| 861 | new.userspace_addr = mem->userspace_addr; |
| 862 | |
| 863 | if (kvm_arch_create_memslot(kvm, &new, npages)) |
| 864 | goto out_free; |
| 865 | } |
| 866 | |
| 867 | /* Allocate page dirty bitmap if needed */ |
| 868 | if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) { |
| 869 | if (kvm_create_dirty_bitmap(&new) < 0) |
| 870 | goto out_free; |
| 871 | } |
| 872 | |
| 873 | slots = kvm_kvzalloc(sizeof(struct kvm_memslots)); |
| 874 | if (!slots) |
| 875 | goto out_free; |
| 876 | memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots)); |
| 877 | |
| 878 | if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) { |
| 879 | slot = id_to_memslot(slots, mem->slot); |
| 880 | slot->flags |= KVM_MEMSLOT_INVALID; |
| 881 | |
| 882 | old_memslots = install_new_memslots(kvm, slots); |
| 883 | |
| 884 | /* slot was deleted or moved, clear iommu mapping */ |
| 885 | kvm_iommu_unmap_pages(kvm, &old); |
| 886 | /* From this point no new shadow pages pointing to a deleted, |
| 887 | * or moved, memslot will be created. |
| 888 | * |
| 889 | * validation of sp->gfn happens in: |
| 890 | * - gfn_to_hva (kvm_read_guest, gfn_to_pfn) |
| 891 | * - kvm_is_visible_gfn (mmu_check_roots) |
| 892 | */ |
| 893 | kvm_arch_flush_shadow_memslot(kvm, slot); |
| 894 | |
| 895 | /* |
| 896 | * We can re-use the old_memslots from above, the only difference |
| 897 | * from the currently installed memslots is the invalid flag. This |
| 898 | * will get overwritten by update_memslots anyway. |
| 899 | */ |
| 900 | slots = old_memslots; |
| 901 | } |
| 902 | |
| 903 | r = kvm_arch_prepare_memory_region(kvm, &new, mem, change); |
| 904 | if (r) |
| 905 | goto out_slots; |
| 906 | |
| 907 | /* actual memory is freed via old in kvm_free_memslot below */ |
| 908 | if (change == KVM_MR_DELETE) { |
| 909 | new.dirty_bitmap = NULL; |
| 910 | memset(&new.arch, 0, sizeof(new.arch)); |
| 911 | } |
| 912 | |
| 913 | update_memslots(slots, &new); |
| 914 | old_memslots = install_new_memslots(kvm, slots); |
| 915 | |
| 916 | kvm_arch_commit_memory_region(kvm, mem, &old, change); |
| 917 | |
| 918 | kvm_free_memslot(kvm, &old, &new); |
| 919 | kvfree(old_memslots); |
| 920 | |
| 921 | /* |
| 922 | * IOMMU mapping: New slots need to be mapped. Old slots need to be |
| 923 | * un-mapped and re-mapped if their base changes. Since base change |
| 924 | * unmapping is handled above with slot deletion, mapping alone is |
| 925 | * needed here. Anything else the iommu might care about for existing |
| 926 | * slots (size changes, userspace addr changes and read-only flag |
| 927 | * changes) is disallowed above, so any other attribute changes getting |
| 928 | * here can be skipped. |
| 929 | */ |
| 930 | if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) { |
| 931 | r = kvm_iommu_map_pages(kvm, &new); |
| 932 | return r; |
| 933 | } |
| 934 | |
| 935 | return 0; |
| 936 | |
| 937 | out_slots: |
| 938 | kvfree(slots); |
| 939 | out_free: |
| 940 | kvm_free_memslot(kvm, &new, &old); |
| 941 | out: |
| 942 | return r; |
| 943 | } |
| 944 | EXPORT_SYMBOL_GPL(__kvm_set_memory_region); |
| 945 | |
| 946 | int kvm_set_memory_region(struct kvm *kvm, |
| 947 | struct kvm_userspace_memory_region *mem) |
| 948 | { |
| 949 | int r; |
| 950 | |
| 951 | mutex_lock(&kvm->slots_lock); |
| 952 | r = __kvm_set_memory_region(kvm, mem); |
| 953 | mutex_unlock(&kvm->slots_lock); |
| 954 | return r; |
| 955 | } |
| 956 | EXPORT_SYMBOL_GPL(kvm_set_memory_region); |
| 957 | |
| 958 | static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm, |
| 959 | struct kvm_userspace_memory_region *mem) |
| 960 | { |
| 961 | if (mem->slot >= KVM_USER_MEM_SLOTS) |
| 962 | return -EINVAL; |
| 963 | return kvm_set_memory_region(kvm, mem); |
| 964 | } |
| 965 | |
| 966 | int kvm_get_dirty_log(struct kvm *kvm, |
| 967 | struct kvm_dirty_log *log, int *is_dirty) |
| 968 | { |
| 969 | struct kvm_memory_slot *memslot; |
| 970 | int r, i; |
| 971 | unsigned long n; |
| 972 | unsigned long any = 0; |
| 973 | |
| 974 | r = -EINVAL; |
| 975 | if (log->slot >= KVM_USER_MEM_SLOTS) |
| 976 | goto out; |
| 977 | |
| 978 | memslot = id_to_memslot(kvm->memslots, log->slot); |
| 979 | r = -ENOENT; |
| 980 | if (!memslot->dirty_bitmap) |
| 981 | goto out; |
| 982 | |
| 983 | n = kvm_dirty_bitmap_bytes(memslot); |
| 984 | |
| 985 | for (i = 0; !any && i < n/sizeof(long); ++i) |
| 986 | any = memslot->dirty_bitmap[i]; |
| 987 | |
| 988 | r = -EFAULT; |
| 989 | if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n)) |
| 990 | goto out; |
| 991 | |
| 992 | if (any) |
| 993 | *is_dirty = 1; |
| 994 | |
| 995 | r = 0; |
| 996 | out: |
| 997 | return r; |
| 998 | } |
| 999 | EXPORT_SYMBOL_GPL(kvm_get_dirty_log); |
| 1000 | |
| 1001 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
| 1002 | /** |
| 1003 | * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages |
| 1004 | * are dirty write protect them for next write. |
| 1005 | * @kvm: pointer to kvm instance |
| 1006 | * @log: slot id and address to which we copy the log |
| 1007 | * @is_dirty: flag set if any page is dirty |
| 1008 | * |
| 1009 | * We need to keep it in mind that VCPU threads can write to the bitmap |
| 1010 | * concurrently. So, to avoid losing track of dirty pages we keep the |
| 1011 | * following order: |
| 1012 | * |
| 1013 | * 1. Take a snapshot of the bit and clear it if needed. |
| 1014 | * 2. Write protect the corresponding page. |
| 1015 | * 3. Copy the snapshot to the userspace. |
| 1016 | * 4. Upon return caller flushes TLB's if needed. |
| 1017 | * |
| 1018 | * Between 2 and 4, the guest may write to the page using the remaining TLB |
| 1019 | * entry. This is not a problem because the page is reported dirty using |
| 1020 | * the snapshot taken before and step 4 ensures that writes done after |
| 1021 | * exiting to userspace will be logged for the next call. |
| 1022 | * |
| 1023 | */ |
| 1024 | int kvm_get_dirty_log_protect(struct kvm *kvm, |
| 1025 | struct kvm_dirty_log *log, bool *is_dirty) |
| 1026 | { |
| 1027 | struct kvm_memory_slot *memslot; |
| 1028 | int r, i; |
| 1029 | unsigned long n; |
| 1030 | unsigned long *dirty_bitmap; |
| 1031 | unsigned long *dirty_bitmap_buffer; |
| 1032 | |
| 1033 | r = -EINVAL; |
| 1034 | if (log->slot >= KVM_USER_MEM_SLOTS) |
| 1035 | goto out; |
| 1036 | |
| 1037 | memslot = id_to_memslot(kvm->memslots, log->slot); |
| 1038 | |
| 1039 | dirty_bitmap = memslot->dirty_bitmap; |
| 1040 | r = -ENOENT; |
| 1041 | if (!dirty_bitmap) |
| 1042 | goto out; |
| 1043 | |
| 1044 | n = kvm_dirty_bitmap_bytes(memslot); |
| 1045 | |
| 1046 | dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long); |
| 1047 | memset(dirty_bitmap_buffer, 0, n); |
| 1048 | |
| 1049 | spin_lock(&kvm->mmu_lock); |
| 1050 | *is_dirty = false; |
| 1051 | for (i = 0; i < n / sizeof(long); i++) { |
| 1052 | unsigned long mask; |
| 1053 | gfn_t offset; |
| 1054 | |
| 1055 | if (!dirty_bitmap[i]) |
| 1056 | continue; |
| 1057 | |
| 1058 | *is_dirty = true; |
| 1059 | |
| 1060 | mask = xchg(&dirty_bitmap[i], 0); |
| 1061 | dirty_bitmap_buffer[i] = mask; |
| 1062 | |
| 1063 | if (mask) { |
| 1064 | offset = i * BITS_PER_LONG; |
| 1065 | kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot, |
| 1066 | offset, mask); |
| 1067 | } |
| 1068 | } |
| 1069 | |
| 1070 | spin_unlock(&kvm->mmu_lock); |
| 1071 | |
| 1072 | r = -EFAULT; |
| 1073 | if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n)) |
| 1074 | goto out; |
| 1075 | |
| 1076 | r = 0; |
| 1077 | out: |
| 1078 | return r; |
| 1079 | } |
| 1080 | EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect); |
| 1081 | #endif |
| 1082 | |
| 1083 | bool kvm_largepages_enabled(void) |
| 1084 | { |
| 1085 | return largepages_enabled; |
| 1086 | } |
| 1087 | |
| 1088 | void kvm_disable_largepages(void) |
| 1089 | { |
| 1090 | largepages_enabled = false; |
| 1091 | } |
| 1092 | EXPORT_SYMBOL_GPL(kvm_disable_largepages); |
| 1093 | |
| 1094 | struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn) |
| 1095 | { |
| 1096 | return __gfn_to_memslot(kvm_memslots(kvm), gfn); |
| 1097 | } |
| 1098 | EXPORT_SYMBOL_GPL(gfn_to_memslot); |
| 1099 | |
| 1100 | int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn) |
| 1101 | { |
| 1102 | struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn); |
| 1103 | |
| 1104 | if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS || |
| 1105 | memslot->flags & KVM_MEMSLOT_INVALID) |
| 1106 | return 0; |
| 1107 | |
| 1108 | return 1; |
| 1109 | } |
| 1110 | EXPORT_SYMBOL_GPL(kvm_is_visible_gfn); |
| 1111 | |
| 1112 | unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn) |
| 1113 | { |
| 1114 | struct vm_area_struct *vma; |
| 1115 | unsigned long addr, size; |
| 1116 | |
| 1117 | size = PAGE_SIZE; |
| 1118 | |
| 1119 | addr = gfn_to_hva(kvm, gfn); |
| 1120 | if (kvm_is_error_hva(addr)) |
| 1121 | return PAGE_SIZE; |
| 1122 | |
| 1123 | down_read(¤t->mm->mmap_sem); |
| 1124 | vma = find_vma(current->mm, addr); |
| 1125 | if (!vma) |
| 1126 | goto out; |
| 1127 | |
| 1128 | size = vma_kernel_pagesize(vma); |
| 1129 | |
| 1130 | out: |
| 1131 | up_read(¤t->mm->mmap_sem); |
| 1132 | |
| 1133 | return size; |
| 1134 | } |
| 1135 | |
| 1136 | static bool memslot_is_readonly(struct kvm_memory_slot *slot) |
| 1137 | { |
| 1138 | return slot->flags & KVM_MEM_READONLY; |
| 1139 | } |
| 1140 | |
| 1141 | static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn, |
| 1142 | gfn_t *nr_pages, bool write) |
| 1143 | { |
| 1144 | if (!slot || slot->flags & KVM_MEMSLOT_INVALID) |
| 1145 | return KVM_HVA_ERR_BAD; |
| 1146 | |
| 1147 | if (memslot_is_readonly(slot) && write) |
| 1148 | return KVM_HVA_ERR_RO_BAD; |
| 1149 | |
| 1150 | if (nr_pages) |
| 1151 | *nr_pages = slot->npages - (gfn - slot->base_gfn); |
| 1152 | |
| 1153 | return __gfn_to_hva_memslot(slot, gfn); |
| 1154 | } |
| 1155 | |
| 1156 | static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn, |
| 1157 | gfn_t *nr_pages) |
| 1158 | { |
| 1159 | return __gfn_to_hva_many(slot, gfn, nr_pages, true); |
| 1160 | } |
| 1161 | |
| 1162 | unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot, |
| 1163 | gfn_t gfn) |
| 1164 | { |
| 1165 | return gfn_to_hva_many(slot, gfn, NULL); |
| 1166 | } |
| 1167 | EXPORT_SYMBOL_GPL(gfn_to_hva_memslot); |
| 1168 | |
| 1169 | unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn) |
| 1170 | { |
| 1171 | return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL); |
| 1172 | } |
| 1173 | EXPORT_SYMBOL_GPL(gfn_to_hva); |
| 1174 | |
| 1175 | /* |
| 1176 | * If writable is set to false, the hva returned by this function is only |
| 1177 | * allowed to be read. |
| 1178 | */ |
| 1179 | unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot, |
| 1180 | gfn_t gfn, bool *writable) |
| 1181 | { |
| 1182 | unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false); |
| 1183 | |
| 1184 | if (!kvm_is_error_hva(hva) && writable) |
| 1185 | *writable = !memslot_is_readonly(slot); |
| 1186 | |
| 1187 | return hva; |
| 1188 | } |
| 1189 | |
| 1190 | unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable) |
| 1191 | { |
| 1192 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); |
| 1193 | |
| 1194 | return gfn_to_hva_memslot_prot(slot, gfn, writable); |
| 1195 | } |
| 1196 | |
| 1197 | static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm, |
| 1198 | unsigned long start, int write, struct page **page) |
| 1199 | { |
| 1200 | int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET; |
| 1201 | |
| 1202 | if (write) |
| 1203 | flags |= FOLL_WRITE; |
| 1204 | |
| 1205 | return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL); |
| 1206 | } |
| 1207 | |
| 1208 | static inline int check_user_page_hwpoison(unsigned long addr) |
| 1209 | { |
| 1210 | int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE; |
| 1211 | |
| 1212 | rc = __get_user_pages(current, current->mm, addr, 1, |
| 1213 | flags, NULL, NULL, NULL); |
| 1214 | return rc == -EHWPOISON; |
| 1215 | } |
| 1216 | |
| 1217 | /* |
| 1218 | * The atomic path to get the writable pfn which will be stored in @pfn, |
| 1219 | * true indicates success, otherwise false is returned. |
| 1220 | */ |
| 1221 | static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async, |
| 1222 | bool write_fault, bool *writable, pfn_t *pfn) |
| 1223 | { |
| 1224 | struct page *page[1]; |
| 1225 | int npages; |
| 1226 | |
| 1227 | if (!(async || atomic)) |
| 1228 | return false; |
| 1229 | |
| 1230 | /* |
| 1231 | * Fast pin a writable pfn only if it is a write fault request |
| 1232 | * or the caller allows to map a writable pfn for a read fault |
| 1233 | * request. |
| 1234 | */ |
| 1235 | if (!(write_fault || writable)) |
| 1236 | return false; |
| 1237 | |
| 1238 | npages = __get_user_pages_fast(addr, 1, 1, page); |
| 1239 | if (npages == 1) { |
| 1240 | *pfn = page_to_pfn(page[0]); |
| 1241 | |
| 1242 | if (writable) |
| 1243 | *writable = true; |
| 1244 | return true; |
| 1245 | } |
| 1246 | |
| 1247 | return false; |
| 1248 | } |
| 1249 | |
| 1250 | /* |
| 1251 | * The slow path to get the pfn of the specified host virtual address, |
| 1252 | * 1 indicates success, -errno is returned if error is detected. |
| 1253 | */ |
| 1254 | static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault, |
| 1255 | bool *writable, pfn_t *pfn) |
| 1256 | { |
| 1257 | struct page *page[1]; |
| 1258 | int npages = 0; |
| 1259 | |
| 1260 | might_sleep(); |
| 1261 | |
| 1262 | if (writable) |
| 1263 | *writable = write_fault; |
| 1264 | |
| 1265 | if (async) { |
| 1266 | down_read(¤t->mm->mmap_sem); |
| 1267 | npages = get_user_page_nowait(current, current->mm, |
| 1268 | addr, write_fault, page); |
| 1269 | up_read(¤t->mm->mmap_sem); |
| 1270 | } else |
| 1271 | npages = __get_user_pages_unlocked(current, current->mm, addr, 1, |
| 1272 | write_fault, 0, page, |
| 1273 | FOLL_TOUCH|FOLL_HWPOISON); |
| 1274 | if (npages != 1) |
| 1275 | return npages; |
| 1276 | |
| 1277 | /* map read fault as writable if possible */ |
| 1278 | if (unlikely(!write_fault) && writable) { |
| 1279 | struct page *wpage[1]; |
| 1280 | |
| 1281 | npages = __get_user_pages_fast(addr, 1, 1, wpage); |
| 1282 | if (npages == 1) { |
| 1283 | *writable = true; |
| 1284 | put_page(page[0]); |
| 1285 | page[0] = wpage[0]; |
| 1286 | } |
| 1287 | |
| 1288 | npages = 1; |
| 1289 | } |
| 1290 | *pfn = page_to_pfn(page[0]); |
| 1291 | return npages; |
| 1292 | } |
| 1293 | |
| 1294 | static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault) |
| 1295 | { |
| 1296 | if (unlikely(!(vma->vm_flags & VM_READ))) |
| 1297 | return false; |
| 1298 | |
| 1299 | if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE)))) |
| 1300 | return false; |
| 1301 | |
| 1302 | return true; |
| 1303 | } |
| 1304 | |
| 1305 | /* |
| 1306 | * Pin guest page in memory and return its pfn. |
| 1307 | * @addr: host virtual address which maps memory to the guest |
| 1308 | * @atomic: whether this function can sleep |
| 1309 | * @async: whether this function need to wait IO complete if the |
| 1310 | * host page is not in the memory |
| 1311 | * @write_fault: whether we should get a writable host page |
| 1312 | * @writable: whether it allows to map a writable host page for !@write_fault |
| 1313 | * |
| 1314 | * The function will map a writable host page for these two cases: |
| 1315 | * 1): @write_fault = true |
| 1316 | * 2): @write_fault = false && @writable, @writable will tell the caller |
| 1317 | * whether the mapping is writable. |
| 1318 | */ |
| 1319 | static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async, |
| 1320 | bool write_fault, bool *writable) |
| 1321 | { |
| 1322 | struct vm_area_struct *vma; |
| 1323 | pfn_t pfn = 0; |
| 1324 | int npages; |
| 1325 | |
| 1326 | /* we can do it either atomically or asynchronously, not both */ |
| 1327 | BUG_ON(atomic && async); |
| 1328 | |
| 1329 | if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn)) |
| 1330 | return pfn; |
| 1331 | |
| 1332 | if (atomic) |
| 1333 | return KVM_PFN_ERR_FAULT; |
| 1334 | |
| 1335 | npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn); |
| 1336 | if (npages == 1) |
| 1337 | return pfn; |
| 1338 | |
| 1339 | down_read(¤t->mm->mmap_sem); |
| 1340 | if (npages == -EHWPOISON || |
| 1341 | (!async && check_user_page_hwpoison(addr))) { |
| 1342 | pfn = KVM_PFN_ERR_HWPOISON; |
| 1343 | goto exit; |
| 1344 | } |
| 1345 | |
| 1346 | vma = find_vma_intersection(current->mm, addr, addr + 1); |
| 1347 | |
| 1348 | if (vma == NULL) |
| 1349 | pfn = KVM_PFN_ERR_FAULT; |
| 1350 | else if ((vma->vm_flags & VM_PFNMAP)) { |
| 1351 | pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + |
| 1352 | vma->vm_pgoff; |
| 1353 | BUG_ON(!kvm_is_reserved_pfn(pfn)); |
| 1354 | } else { |
| 1355 | if (async && vma_is_valid(vma, write_fault)) |
| 1356 | *async = true; |
| 1357 | pfn = KVM_PFN_ERR_FAULT; |
| 1358 | } |
| 1359 | exit: |
| 1360 | up_read(¤t->mm->mmap_sem); |
| 1361 | return pfn; |
| 1362 | } |
| 1363 | |
| 1364 | pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic, |
| 1365 | bool *async, bool write_fault, bool *writable) |
| 1366 | { |
| 1367 | unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault); |
| 1368 | |
| 1369 | if (addr == KVM_HVA_ERR_RO_BAD) |
| 1370 | return KVM_PFN_ERR_RO_FAULT; |
| 1371 | |
| 1372 | if (kvm_is_error_hva(addr)) |
| 1373 | return KVM_PFN_NOSLOT; |
| 1374 | |
| 1375 | /* Do not map writable pfn in the readonly memslot. */ |
| 1376 | if (writable && memslot_is_readonly(slot)) { |
| 1377 | *writable = false; |
| 1378 | writable = NULL; |
| 1379 | } |
| 1380 | |
| 1381 | return hva_to_pfn(addr, atomic, async, write_fault, |
| 1382 | writable); |
| 1383 | } |
| 1384 | EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot); |
| 1385 | |
| 1386 | static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, |
| 1387 | bool write_fault, bool *writable) |
| 1388 | { |
| 1389 | struct kvm_memory_slot *slot; |
| 1390 | |
| 1391 | slot = gfn_to_memslot(kvm, gfn); |
| 1392 | |
| 1393 | return __gfn_to_pfn_memslot(slot, gfn, atomic, NULL, write_fault, |
| 1394 | writable); |
| 1395 | } |
| 1396 | |
| 1397 | pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn) |
| 1398 | { |
| 1399 | return __gfn_to_pfn(kvm, gfn, true, true, NULL); |
| 1400 | } |
| 1401 | EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic); |
| 1402 | |
| 1403 | pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn) |
| 1404 | { |
| 1405 | return __gfn_to_pfn(kvm, gfn, false, true, NULL); |
| 1406 | } |
| 1407 | EXPORT_SYMBOL_GPL(gfn_to_pfn); |
| 1408 | |
| 1409 | pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault, |
| 1410 | bool *writable) |
| 1411 | { |
| 1412 | return __gfn_to_pfn(kvm, gfn, false, write_fault, writable); |
| 1413 | } |
| 1414 | EXPORT_SYMBOL_GPL(gfn_to_pfn_prot); |
| 1415 | |
| 1416 | pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn) |
| 1417 | { |
| 1418 | return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL); |
| 1419 | } |
| 1420 | |
| 1421 | pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn) |
| 1422 | { |
| 1423 | return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL); |
| 1424 | } |
| 1425 | EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic); |
| 1426 | |
| 1427 | int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages, |
| 1428 | int nr_pages) |
| 1429 | { |
| 1430 | unsigned long addr; |
| 1431 | gfn_t entry; |
| 1432 | |
| 1433 | addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry); |
| 1434 | if (kvm_is_error_hva(addr)) |
| 1435 | return -1; |
| 1436 | |
| 1437 | if (entry < nr_pages) |
| 1438 | return 0; |
| 1439 | |
| 1440 | return __get_user_pages_fast(addr, nr_pages, 1, pages); |
| 1441 | } |
| 1442 | EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic); |
| 1443 | |
| 1444 | static struct page *kvm_pfn_to_page(pfn_t pfn) |
| 1445 | { |
| 1446 | if (is_error_noslot_pfn(pfn)) |
| 1447 | return KVM_ERR_PTR_BAD_PAGE; |
| 1448 | |
| 1449 | if (kvm_is_reserved_pfn(pfn)) { |
| 1450 | WARN_ON(1); |
| 1451 | return KVM_ERR_PTR_BAD_PAGE; |
| 1452 | } |
| 1453 | |
| 1454 | return pfn_to_page(pfn); |
| 1455 | } |
| 1456 | |
| 1457 | struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn) |
| 1458 | { |
| 1459 | pfn_t pfn; |
| 1460 | |
| 1461 | pfn = gfn_to_pfn(kvm, gfn); |
| 1462 | |
| 1463 | return kvm_pfn_to_page(pfn); |
| 1464 | } |
| 1465 | EXPORT_SYMBOL_GPL(gfn_to_page); |
| 1466 | |
| 1467 | void kvm_release_page_clean(struct page *page) |
| 1468 | { |
| 1469 | WARN_ON(is_error_page(page)); |
| 1470 | |
| 1471 | kvm_release_pfn_clean(page_to_pfn(page)); |
| 1472 | } |
| 1473 | EXPORT_SYMBOL_GPL(kvm_release_page_clean); |
| 1474 | |
| 1475 | void kvm_release_pfn_clean(pfn_t pfn) |
| 1476 | { |
| 1477 | if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn)) |
| 1478 | put_page(pfn_to_page(pfn)); |
| 1479 | } |
| 1480 | EXPORT_SYMBOL_GPL(kvm_release_pfn_clean); |
| 1481 | |
| 1482 | void kvm_release_page_dirty(struct page *page) |
| 1483 | { |
| 1484 | WARN_ON(is_error_page(page)); |
| 1485 | |
| 1486 | kvm_release_pfn_dirty(page_to_pfn(page)); |
| 1487 | } |
| 1488 | EXPORT_SYMBOL_GPL(kvm_release_page_dirty); |
| 1489 | |
| 1490 | static void kvm_release_pfn_dirty(pfn_t pfn) |
| 1491 | { |
| 1492 | kvm_set_pfn_dirty(pfn); |
| 1493 | kvm_release_pfn_clean(pfn); |
| 1494 | } |
| 1495 | |
| 1496 | void kvm_set_pfn_dirty(pfn_t pfn) |
| 1497 | { |
| 1498 | if (!kvm_is_reserved_pfn(pfn)) { |
| 1499 | struct page *page = pfn_to_page(pfn); |
| 1500 | |
| 1501 | if (!PageReserved(page)) |
| 1502 | SetPageDirty(page); |
| 1503 | } |
| 1504 | } |
| 1505 | EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty); |
| 1506 | |
| 1507 | void kvm_set_pfn_accessed(pfn_t pfn) |
| 1508 | { |
| 1509 | if (!kvm_is_reserved_pfn(pfn)) |
| 1510 | mark_page_accessed(pfn_to_page(pfn)); |
| 1511 | } |
| 1512 | EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed); |
| 1513 | |
| 1514 | void kvm_get_pfn(pfn_t pfn) |
| 1515 | { |
| 1516 | if (!kvm_is_reserved_pfn(pfn)) |
| 1517 | get_page(pfn_to_page(pfn)); |
| 1518 | } |
| 1519 | EXPORT_SYMBOL_GPL(kvm_get_pfn); |
| 1520 | |
| 1521 | static int next_segment(unsigned long len, int offset) |
| 1522 | { |
| 1523 | if (len > PAGE_SIZE - offset) |
| 1524 | return PAGE_SIZE - offset; |
| 1525 | else |
| 1526 | return len; |
| 1527 | } |
| 1528 | |
| 1529 | int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset, |
| 1530 | int len) |
| 1531 | { |
| 1532 | int r; |
| 1533 | unsigned long addr; |
| 1534 | |
| 1535 | addr = gfn_to_hva_prot(kvm, gfn, NULL); |
| 1536 | if (kvm_is_error_hva(addr)) |
| 1537 | return -EFAULT; |
| 1538 | r = __copy_from_user(data, (void __user *)addr + offset, len); |
| 1539 | if (r) |
| 1540 | return -EFAULT; |
| 1541 | return 0; |
| 1542 | } |
| 1543 | EXPORT_SYMBOL_GPL(kvm_read_guest_page); |
| 1544 | |
| 1545 | int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len) |
| 1546 | { |
| 1547 | gfn_t gfn = gpa >> PAGE_SHIFT; |
| 1548 | int seg; |
| 1549 | int offset = offset_in_page(gpa); |
| 1550 | int ret; |
| 1551 | |
| 1552 | while ((seg = next_segment(len, offset)) != 0) { |
| 1553 | ret = kvm_read_guest_page(kvm, gfn, data, offset, seg); |
| 1554 | if (ret < 0) |
| 1555 | return ret; |
| 1556 | offset = 0; |
| 1557 | len -= seg; |
| 1558 | data += seg; |
| 1559 | ++gfn; |
| 1560 | } |
| 1561 | return 0; |
| 1562 | } |
| 1563 | EXPORT_SYMBOL_GPL(kvm_read_guest); |
| 1564 | |
| 1565 | int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data, |
| 1566 | unsigned long len) |
| 1567 | { |
| 1568 | int r; |
| 1569 | unsigned long addr; |
| 1570 | gfn_t gfn = gpa >> PAGE_SHIFT; |
| 1571 | int offset = offset_in_page(gpa); |
| 1572 | |
| 1573 | addr = gfn_to_hva_prot(kvm, gfn, NULL); |
| 1574 | if (kvm_is_error_hva(addr)) |
| 1575 | return -EFAULT; |
| 1576 | pagefault_disable(); |
| 1577 | r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len); |
| 1578 | pagefault_enable(); |
| 1579 | if (r) |
| 1580 | return -EFAULT; |
| 1581 | return 0; |
| 1582 | } |
| 1583 | EXPORT_SYMBOL(kvm_read_guest_atomic); |
| 1584 | |
| 1585 | int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data, |
| 1586 | int offset, int len) |
| 1587 | { |
| 1588 | int r; |
| 1589 | struct kvm_memory_slot *memslot; |
| 1590 | unsigned long addr; |
| 1591 | |
| 1592 | memslot = gfn_to_memslot(kvm, gfn); |
| 1593 | addr = gfn_to_hva_memslot(memslot, gfn); |
| 1594 | if (kvm_is_error_hva(addr)) |
| 1595 | return -EFAULT; |
| 1596 | r = __copy_to_user((void __user *)addr + offset, data, len); |
| 1597 | if (r) |
| 1598 | return -EFAULT; |
| 1599 | mark_page_dirty_in_slot(kvm, memslot, gfn); |
| 1600 | return 0; |
| 1601 | } |
| 1602 | EXPORT_SYMBOL_GPL(kvm_write_guest_page); |
| 1603 | |
| 1604 | int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data, |
| 1605 | unsigned long len) |
| 1606 | { |
| 1607 | gfn_t gfn = gpa >> PAGE_SHIFT; |
| 1608 | int seg; |
| 1609 | int offset = offset_in_page(gpa); |
| 1610 | int ret; |
| 1611 | |
| 1612 | while ((seg = next_segment(len, offset)) != 0) { |
| 1613 | ret = kvm_write_guest_page(kvm, gfn, data, offset, seg); |
| 1614 | if (ret < 0) |
| 1615 | return ret; |
| 1616 | offset = 0; |
| 1617 | len -= seg; |
| 1618 | data += seg; |
| 1619 | ++gfn; |
| 1620 | } |
| 1621 | return 0; |
| 1622 | } |
| 1623 | EXPORT_SYMBOL_GPL(kvm_write_guest); |
| 1624 | |
| 1625 | int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
| 1626 | gpa_t gpa, unsigned long len) |
| 1627 | { |
| 1628 | struct kvm_memslots *slots = kvm_memslots(kvm); |
| 1629 | int offset = offset_in_page(gpa); |
| 1630 | gfn_t start_gfn = gpa >> PAGE_SHIFT; |
| 1631 | gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT; |
| 1632 | gfn_t nr_pages_needed = end_gfn - start_gfn + 1; |
| 1633 | gfn_t nr_pages_avail; |
| 1634 | |
| 1635 | ghc->gpa = gpa; |
| 1636 | ghc->generation = slots->generation; |
| 1637 | ghc->len = len; |
| 1638 | ghc->memslot = gfn_to_memslot(kvm, start_gfn); |
| 1639 | ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, NULL); |
| 1640 | if (!kvm_is_error_hva(ghc->hva) && nr_pages_needed <= 1) { |
| 1641 | ghc->hva += offset; |
| 1642 | } else { |
| 1643 | /* |
| 1644 | * If the requested region crosses two memslots, we still |
| 1645 | * verify that the entire region is valid here. |
| 1646 | */ |
| 1647 | while (start_gfn <= end_gfn) { |
| 1648 | ghc->memslot = gfn_to_memslot(kvm, start_gfn); |
| 1649 | ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, |
| 1650 | &nr_pages_avail); |
| 1651 | if (kvm_is_error_hva(ghc->hva)) |
| 1652 | return -EFAULT; |
| 1653 | start_gfn += nr_pages_avail; |
| 1654 | } |
| 1655 | /* Use the slow path for cross page reads and writes. */ |
| 1656 | ghc->memslot = NULL; |
| 1657 | } |
| 1658 | return 0; |
| 1659 | } |
| 1660 | EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init); |
| 1661 | |
| 1662 | int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
| 1663 | void *data, unsigned long len) |
| 1664 | { |
| 1665 | struct kvm_memslots *slots = kvm_memslots(kvm); |
| 1666 | int r; |
| 1667 | |
| 1668 | BUG_ON(len > ghc->len); |
| 1669 | |
| 1670 | if (slots->generation != ghc->generation) |
| 1671 | kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len); |
| 1672 | |
| 1673 | if (unlikely(!ghc->memslot)) |
| 1674 | return kvm_write_guest(kvm, ghc->gpa, data, len); |
| 1675 | |
| 1676 | if (kvm_is_error_hva(ghc->hva)) |
| 1677 | return -EFAULT; |
| 1678 | |
| 1679 | r = __copy_to_user((void __user *)ghc->hva, data, len); |
| 1680 | if (r) |
| 1681 | return -EFAULT; |
| 1682 | mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT); |
| 1683 | |
| 1684 | return 0; |
| 1685 | } |
| 1686 | EXPORT_SYMBOL_GPL(kvm_write_guest_cached); |
| 1687 | |
| 1688 | int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
| 1689 | void *data, unsigned long len) |
| 1690 | { |
| 1691 | struct kvm_memslots *slots = kvm_memslots(kvm); |
| 1692 | int r; |
| 1693 | |
| 1694 | BUG_ON(len > ghc->len); |
| 1695 | |
| 1696 | if (slots->generation != ghc->generation) |
| 1697 | kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len); |
| 1698 | |
| 1699 | if (unlikely(!ghc->memslot)) |
| 1700 | return kvm_read_guest(kvm, ghc->gpa, data, len); |
| 1701 | |
| 1702 | if (kvm_is_error_hva(ghc->hva)) |
| 1703 | return -EFAULT; |
| 1704 | |
| 1705 | r = __copy_from_user(data, (void __user *)ghc->hva, len); |
| 1706 | if (r) |
| 1707 | return -EFAULT; |
| 1708 | |
| 1709 | return 0; |
| 1710 | } |
| 1711 | EXPORT_SYMBOL_GPL(kvm_read_guest_cached); |
| 1712 | |
| 1713 | int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len) |
| 1714 | { |
| 1715 | const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0))); |
| 1716 | |
| 1717 | return kvm_write_guest_page(kvm, gfn, zero_page, offset, len); |
| 1718 | } |
| 1719 | EXPORT_SYMBOL_GPL(kvm_clear_guest_page); |
| 1720 | |
| 1721 | int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len) |
| 1722 | { |
| 1723 | gfn_t gfn = gpa >> PAGE_SHIFT; |
| 1724 | int seg; |
| 1725 | int offset = offset_in_page(gpa); |
| 1726 | int ret; |
| 1727 | |
| 1728 | while ((seg = next_segment(len, offset)) != 0) { |
| 1729 | ret = kvm_clear_guest_page(kvm, gfn, offset, seg); |
| 1730 | if (ret < 0) |
| 1731 | return ret; |
| 1732 | offset = 0; |
| 1733 | len -= seg; |
| 1734 | ++gfn; |
| 1735 | } |
| 1736 | return 0; |
| 1737 | } |
| 1738 | EXPORT_SYMBOL_GPL(kvm_clear_guest); |
| 1739 | |
| 1740 | static void mark_page_dirty_in_slot(struct kvm *kvm, |
| 1741 | struct kvm_memory_slot *memslot, |
| 1742 | gfn_t gfn) |
| 1743 | { |
| 1744 | if (memslot && memslot->dirty_bitmap) { |
| 1745 | unsigned long rel_gfn = gfn - memslot->base_gfn; |
| 1746 | |
| 1747 | set_bit_le(rel_gfn, memslot->dirty_bitmap); |
| 1748 | } |
| 1749 | } |
| 1750 | |
| 1751 | void mark_page_dirty(struct kvm *kvm, gfn_t gfn) |
| 1752 | { |
| 1753 | struct kvm_memory_slot *memslot; |
| 1754 | |
| 1755 | memslot = gfn_to_memslot(kvm, gfn); |
| 1756 | mark_page_dirty_in_slot(kvm, memslot, gfn); |
| 1757 | } |
| 1758 | EXPORT_SYMBOL_GPL(mark_page_dirty); |
| 1759 | |
| 1760 | static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu) |
| 1761 | { |
| 1762 | if (kvm_arch_vcpu_runnable(vcpu)) { |
| 1763 | kvm_make_request(KVM_REQ_UNHALT, vcpu); |
| 1764 | return -EINTR; |
| 1765 | } |
| 1766 | if (kvm_cpu_has_pending_timer(vcpu)) |
| 1767 | return -EINTR; |
| 1768 | if (signal_pending(current)) |
| 1769 | return -EINTR; |
| 1770 | |
| 1771 | return 0; |
| 1772 | } |
| 1773 | |
| 1774 | /* |
| 1775 | * The vCPU has executed a HLT instruction with in-kernel mode enabled. |
| 1776 | */ |
| 1777 | void kvm_vcpu_block(struct kvm_vcpu *vcpu) |
| 1778 | { |
| 1779 | ktime_t start, cur; |
| 1780 | DEFINE_WAIT(wait); |
| 1781 | bool waited = false; |
| 1782 | |
| 1783 | start = cur = ktime_get(); |
| 1784 | if (halt_poll_ns) { |
| 1785 | ktime_t stop = ktime_add_ns(ktime_get(), halt_poll_ns); |
| 1786 | |
| 1787 | do { |
| 1788 | /* |
| 1789 | * This sets KVM_REQ_UNHALT if an interrupt |
| 1790 | * arrives. |
| 1791 | */ |
| 1792 | if (kvm_vcpu_check_block(vcpu) < 0) { |
| 1793 | ++vcpu->stat.halt_successful_poll; |
| 1794 | goto out; |
| 1795 | } |
| 1796 | cur = ktime_get(); |
| 1797 | } while (single_task_running() && ktime_before(cur, stop)); |
| 1798 | } |
| 1799 | |
| 1800 | for (;;) { |
| 1801 | prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE); |
| 1802 | |
| 1803 | if (kvm_vcpu_check_block(vcpu) < 0) |
| 1804 | break; |
| 1805 | |
| 1806 | waited = true; |
| 1807 | schedule(); |
| 1808 | } |
| 1809 | |
| 1810 | finish_wait(&vcpu->wq, &wait); |
| 1811 | cur = ktime_get(); |
| 1812 | |
| 1813 | out: |
| 1814 | trace_kvm_vcpu_wakeup(ktime_to_ns(cur) - ktime_to_ns(start), waited); |
| 1815 | } |
| 1816 | EXPORT_SYMBOL_GPL(kvm_vcpu_block); |
| 1817 | |
| 1818 | #ifndef CONFIG_S390 |
| 1819 | /* |
| 1820 | * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode. |
| 1821 | */ |
| 1822 | void kvm_vcpu_kick(struct kvm_vcpu *vcpu) |
| 1823 | { |
| 1824 | int me; |
| 1825 | int cpu = vcpu->cpu; |
| 1826 | wait_queue_head_t *wqp; |
| 1827 | |
| 1828 | wqp = kvm_arch_vcpu_wq(vcpu); |
| 1829 | if (waitqueue_active(wqp)) { |
| 1830 | wake_up_interruptible(wqp); |
| 1831 | ++vcpu->stat.halt_wakeup; |
| 1832 | } |
| 1833 | |
| 1834 | me = get_cpu(); |
| 1835 | if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu)) |
| 1836 | if (kvm_arch_vcpu_should_kick(vcpu)) |
| 1837 | smp_send_reschedule(cpu); |
| 1838 | put_cpu(); |
| 1839 | } |
| 1840 | EXPORT_SYMBOL_GPL(kvm_vcpu_kick); |
| 1841 | #endif /* !CONFIG_S390 */ |
| 1842 | |
| 1843 | int kvm_vcpu_yield_to(struct kvm_vcpu *target) |
| 1844 | { |
| 1845 | struct pid *pid; |
| 1846 | struct task_struct *task = NULL; |
| 1847 | int ret = 0; |
| 1848 | |
| 1849 | rcu_read_lock(); |
| 1850 | pid = rcu_dereference(target->pid); |
| 1851 | if (pid) |
| 1852 | task = get_pid_task(pid, PIDTYPE_PID); |
| 1853 | rcu_read_unlock(); |
| 1854 | if (!task) |
| 1855 | return ret; |
| 1856 | ret = yield_to(task, 1); |
| 1857 | put_task_struct(task); |
| 1858 | |
| 1859 | return ret; |
| 1860 | } |
| 1861 | EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to); |
| 1862 | |
| 1863 | /* |
| 1864 | * Helper that checks whether a VCPU is eligible for directed yield. |
| 1865 | * Most eligible candidate to yield is decided by following heuristics: |
| 1866 | * |
| 1867 | * (a) VCPU which has not done pl-exit or cpu relax intercepted recently |
| 1868 | * (preempted lock holder), indicated by @in_spin_loop. |
| 1869 | * Set at the beiginning and cleared at the end of interception/PLE handler. |
| 1870 | * |
| 1871 | * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get |
| 1872 | * chance last time (mostly it has become eligible now since we have probably |
| 1873 | * yielded to lockholder in last iteration. This is done by toggling |
| 1874 | * @dy_eligible each time a VCPU checked for eligibility.) |
| 1875 | * |
| 1876 | * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding |
| 1877 | * to preempted lock-holder could result in wrong VCPU selection and CPU |
| 1878 | * burning. Giving priority for a potential lock-holder increases lock |
| 1879 | * progress. |
| 1880 | * |
| 1881 | * Since algorithm is based on heuristics, accessing another VCPU data without |
| 1882 | * locking does not harm. It may result in trying to yield to same VCPU, fail |
| 1883 | * and continue with next VCPU and so on. |
| 1884 | */ |
| 1885 | static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu) |
| 1886 | { |
| 1887 | #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT |
| 1888 | bool eligible; |
| 1889 | |
| 1890 | eligible = !vcpu->spin_loop.in_spin_loop || |
| 1891 | vcpu->spin_loop.dy_eligible; |
| 1892 | |
| 1893 | if (vcpu->spin_loop.in_spin_loop) |
| 1894 | kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible); |
| 1895 | |
| 1896 | return eligible; |
| 1897 | #else |
| 1898 | return true; |
| 1899 | #endif |
| 1900 | } |
| 1901 | |
| 1902 | void kvm_vcpu_on_spin(struct kvm_vcpu *me) |
| 1903 | { |
| 1904 | struct kvm *kvm = me->kvm; |
| 1905 | struct kvm_vcpu *vcpu; |
| 1906 | int last_boosted_vcpu = me->kvm->last_boosted_vcpu; |
| 1907 | int yielded = 0; |
| 1908 | int try = 3; |
| 1909 | int pass; |
| 1910 | int i; |
| 1911 | |
| 1912 | kvm_vcpu_set_in_spin_loop(me, true); |
| 1913 | /* |
| 1914 | * We boost the priority of a VCPU that is runnable but not |
| 1915 | * currently running, because it got preempted by something |
| 1916 | * else and called schedule in __vcpu_run. Hopefully that |
| 1917 | * VCPU is holding the lock that we need and will release it. |
| 1918 | * We approximate round-robin by starting at the last boosted VCPU. |
| 1919 | */ |
| 1920 | for (pass = 0; pass < 2 && !yielded && try; pass++) { |
| 1921 | kvm_for_each_vcpu(i, vcpu, kvm) { |
| 1922 | if (!pass && i <= last_boosted_vcpu) { |
| 1923 | i = last_boosted_vcpu; |
| 1924 | continue; |
| 1925 | } else if (pass && i > last_boosted_vcpu) |
| 1926 | break; |
| 1927 | if (!ACCESS_ONCE(vcpu->preempted)) |
| 1928 | continue; |
| 1929 | if (vcpu == me) |
| 1930 | continue; |
| 1931 | if (waitqueue_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu)) |
| 1932 | continue; |
| 1933 | if (!kvm_vcpu_eligible_for_directed_yield(vcpu)) |
| 1934 | continue; |
| 1935 | |
| 1936 | yielded = kvm_vcpu_yield_to(vcpu); |
| 1937 | if (yielded > 0) { |
| 1938 | kvm->last_boosted_vcpu = i; |
| 1939 | break; |
| 1940 | } else if (yielded < 0) { |
| 1941 | try--; |
| 1942 | if (!try) |
| 1943 | break; |
| 1944 | } |
| 1945 | } |
| 1946 | } |
| 1947 | kvm_vcpu_set_in_spin_loop(me, false); |
| 1948 | |
| 1949 | /* Ensure vcpu is not eligible during next spinloop */ |
| 1950 | kvm_vcpu_set_dy_eligible(me, false); |
| 1951 | } |
| 1952 | EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin); |
| 1953 | |
| 1954 | static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf) |
| 1955 | { |
| 1956 | struct kvm_vcpu *vcpu = vma->vm_file->private_data; |
| 1957 | struct page *page; |
| 1958 | |
| 1959 | if (vmf->pgoff == 0) |
| 1960 | page = virt_to_page(vcpu->run); |
| 1961 | #ifdef CONFIG_X86 |
| 1962 | else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET) |
| 1963 | page = virt_to_page(vcpu->arch.pio_data); |
| 1964 | #endif |
| 1965 | #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET |
| 1966 | else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET) |
| 1967 | page = virt_to_page(vcpu->kvm->coalesced_mmio_ring); |
| 1968 | #endif |
| 1969 | else |
| 1970 | return kvm_arch_vcpu_fault(vcpu, vmf); |
| 1971 | get_page(page); |
| 1972 | vmf->page = page; |
| 1973 | return 0; |
| 1974 | } |
| 1975 | |
| 1976 | static const struct vm_operations_struct kvm_vcpu_vm_ops = { |
| 1977 | .fault = kvm_vcpu_fault, |
| 1978 | }; |
| 1979 | |
| 1980 | static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma) |
| 1981 | { |
| 1982 | vma->vm_ops = &kvm_vcpu_vm_ops; |
| 1983 | return 0; |
| 1984 | } |
| 1985 | |
| 1986 | static int kvm_vcpu_release(struct inode *inode, struct file *filp) |
| 1987 | { |
| 1988 | struct kvm_vcpu *vcpu = filp->private_data; |
| 1989 | |
| 1990 | kvm_put_kvm(vcpu->kvm); |
| 1991 | return 0; |
| 1992 | } |
| 1993 | |
| 1994 | static struct file_operations kvm_vcpu_fops = { |
| 1995 | .release = kvm_vcpu_release, |
| 1996 | .unlocked_ioctl = kvm_vcpu_ioctl, |
| 1997 | #ifdef CONFIG_KVM_COMPAT |
| 1998 | .compat_ioctl = kvm_vcpu_compat_ioctl, |
| 1999 | #endif |
| 2000 | .mmap = kvm_vcpu_mmap, |
| 2001 | .llseek = noop_llseek, |
| 2002 | }; |
| 2003 | |
| 2004 | /* |
| 2005 | * Allocates an inode for the vcpu. |
| 2006 | */ |
| 2007 | static int create_vcpu_fd(struct kvm_vcpu *vcpu) |
| 2008 | { |
| 2009 | return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC); |
| 2010 | } |
| 2011 | |
| 2012 | /* |
| 2013 | * Creates some virtual cpus. Good luck creating more than one. |
| 2014 | */ |
| 2015 | static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id) |
| 2016 | { |
| 2017 | int r; |
| 2018 | struct kvm_vcpu *vcpu, *v; |
| 2019 | |
| 2020 | if (id >= KVM_MAX_VCPUS) |
| 2021 | return -EINVAL; |
| 2022 | |
| 2023 | vcpu = kvm_arch_vcpu_create(kvm, id); |
| 2024 | if (IS_ERR(vcpu)) |
| 2025 | return PTR_ERR(vcpu); |
| 2026 | |
| 2027 | preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops); |
| 2028 | |
| 2029 | r = kvm_arch_vcpu_setup(vcpu); |
| 2030 | if (r) |
| 2031 | goto vcpu_destroy; |
| 2032 | |
| 2033 | mutex_lock(&kvm->lock); |
| 2034 | if (!kvm_vcpu_compatible(vcpu)) { |
| 2035 | r = -EINVAL; |
| 2036 | goto unlock_vcpu_destroy; |
| 2037 | } |
| 2038 | if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) { |
| 2039 | r = -EINVAL; |
| 2040 | goto unlock_vcpu_destroy; |
| 2041 | } |
| 2042 | |
| 2043 | kvm_for_each_vcpu(r, v, kvm) |
| 2044 | if (v->vcpu_id == id) { |
| 2045 | r = -EEXIST; |
| 2046 | goto unlock_vcpu_destroy; |
| 2047 | } |
| 2048 | |
| 2049 | BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]); |
| 2050 | |
| 2051 | /* Now it's all set up, let userspace reach it */ |
| 2052 | kvm_get_kvm(kvm); |
| 2053 | r = create_vcpu_fd(vcpu); |
| 2054 | if (r < 0) { |
| 2055 | kvm_put_kvm(kvm); |
| 2056 | goto unlock_vcpu_destroy; |
| 2057 | } |
| 2058 | |
| 2059 | kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu; |
| 2060 | smp_wmb(); |
| 2061 | atomic_inc(&kvm->online_vcpus); |
| 2062 | |
| 2063 | mutex_unlock(&kvm->lock); |
| 2064 | kvm_arch_vcpu_postcreate(vcpu); |
| 2065 | return r; |
| 2066 | |
| 2067 | unlock_vcpu_destroy: |
| 2068 | mutex_unlock(&kvm->lock); |
| 2069 | vcpu_destroy: |
| 2070 | kvm_arch_vcpu_destroy(vcpu); |
| 2071 | return r; |
| 2072 | } |
| 2073 | |
| 2074 | static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset) |
| 2075 | { |
| 2076 | if (sigset) { |
| 2077 | sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP)); |
| 2078 | vcpu->sigset_active = 1; |
| 2079 | vcpu->sigset = *sigset; |
| 2080 | } else |
| 2081 | vcpu->sigset_active = 0; |
| 2082 | return 0; |
| 2083 | } |
| 2084 | |
| 2085 | static long kvm_vcpu_ioctl(struct file *filp, |
| 2086 | unsigned int ioctl, unsigned long arg) |
| 2087 | { |
| 2088 | struct kvm_vcpu *vcpu = filp->private_data; |
| 2089 | void __user *argp = (void __user *)arg; |
| 2090 | int r; |
| 2091 | struct kvm_fpu *fpu = NULL; |
| 2092 | struct kvm_sregs *kvm_sregs = NULL; |
| 2093 | |
| 2094 | if (vcpu->kvm->mm != current->mm) |
| 2095 | return -EIO; |
| 2096 | |
| 2097 | if (unlikely(_IOC_TYPE(ioctl) != KVMIO)) |
| 2098 | return -EINVAL; |
| 2099 | |
| 2100 | #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS) |
| 2101 | /* |
| 2102 | * Special cases: vcpu ioctls that are asynchronous to vcpu execution, |
| 2103 | * so vcpu_load() would break it. |
| 2104 | */ |
| 2105 | if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_S390_IRQ || ioctl == KVM_INTERRUPT) |
| 2106 | return kvm_arch_vcpu_ioctl(filp, ioctl, arg); |
| 2107 | #endif |
| 2108 | |
| 2109 | |
| 2110 | r = vcpu_load(vcpu); |
| 2111 | if (r) |
| 2112 | return r; |
| 2113 | switch (ioctl) { |
| 2114 | case KVM_RUN: |
| 2115 | r = -EINVAL; |
| 2116 | if (arg) |
| 2117 | goto out; |
| 2118 | if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) { |
| 2119 | /* The thread running this VCPU changed. */ |
| 2120 | struct pid *oldpid = vcpu->pid; |
| 2121 | struct pid *newpid = get_task_pid(current, PIDTYPE_PID); |
| 2122 | |
| 2123 | rcu_assign_pointer(vcpu->pid, newpid); |
| 2124 | if (oldpid) |
| 2125 | synchronize_rcu(); |
| 2126 | put_pid(oldpid); |
| 2127 | } |
| 2128 | r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run); |
| 2129 | trace_kvm_userspace_exit(vcpu->run->exit_reason, r); |
| 2130 | break; |
| 2131 | case KVM_GET_REGS: { |
| 2132 | struct kvm_regs *kvm_regs; |
| 2133 | |
| 2134 | r = -ENOMEM; |
| 2135 | kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL); |
| 2136 | if (!kvm_regs) |
| 2137 | goto out; |
| 2138 | r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs); |
| 2139 | if (r) |
| 2140 | goto out_free1; |
| 2141 | r = -EFAULT; |
| 2142 | if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs))) |
| 2143 | goto out_free1; |
| 2144 | r = 0; |
| 2145 | out_free1: |
| 2146 | kfree(kvm_regs); |
| 2147 | break; |
| 2148 | } |
| 2149 | case KVM_SET_REGS: { |
| 2150 | struct kvm_regs *kvm_regs; |
| 2151 | |
| 2152 | r = -ENOMEM; |
| 2153 | kvm_regs = memdup_user(argp, sizeof(*kvm_regs)); |
| 2154 | if (IS_ERR(kvm_regs)) { |
| 2155 | r = PTR_ERR(kvm_regs); |
| 2156 | goto out; |
| 2157 | } |
| 2158 | r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs); |
| 2159 | kfree(kvm_regs); |
| 2160 | break; |
| 2161 | } |
| 2162 | case KVM_GET_SREGS: { |
| 2163 | kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL); |
| 2164 | r = -ENOMEM; |
| 2165 | if (!kvm_sregs) |
| 2166 | goto out; |
| 2167 | r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs); |
| 2168 | if (r) |
| 2169 | goto out; |
| 2170 | r = -EFAULT; |
| 2171 | if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs))) |
| 2172 | goto out; |
| 2173 | r = 0; |
| 2174 | break; |
| 2175 | } |
| 2176 | case KVM_SET_SREGS: { |
| 2177 | kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs)); |
| 2178 | if (IS_ERR(kvm_sregs)) { |
| 2179 | r = PTR_ERR(kvm_sregs); |
| 2180 | kvm_sregs = NULL; |
| 2181 | goto out; |
| 2182 | } |
| 2183 | r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs); |
| 2184 | break; |
| 2185 | } |
| 2186 | case KVM_GET_MP_STATE: { |
| 2187 | struct kvm_mp_state mp_state; |
| 2188 | |
| 2189 | r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state); |
| 2190 | if (r) |
| 2191 | goto out; |
| 2192 | r = -EFAULT; |
| 2193 | if (copy_to_user(argp, &mp_state, sizeof(mp_state))) |
| 2194 | goto out; |
| 2195 | r = 0; |
| 2196 | break; |
| 2197 | } |
| 2198 | case KVM_SET_MP_STATE: { |
| 2199 | struct kvm_mp_state mp_state; |
| 2200 | |
| 2201 | r = -EFAULT; |
| 2202 | if (copy_from_user(&mp_state, argp, sizeof(mp_state))) |
| 2203 | goto out; |
| 2204 | r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state); |
| 2205 | break; |
| 2206 | } |
| 2207 | case KVM_TRANSLATE: { |
| 2208 | struct kvm_translation tr; |
| 2209 | |
| 2210 | r = -EFAULT; |
| 2211 | if (copy_from_user(&tr, argp, sizeof(tr))) |
| 2212 | goto out; |
| 2213 | r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr); |
| 2214 | if (r) |
| 2215 | goto out; |
| 2216 | r = -EFAULT; |
| 2217 | if (copy_to_user(argp, &tr, sizeof(tr))) |
| 2218 | goto out; |
| 2219 | r = 0; |
| 2220 | break; |
| 2221 | } |
| 2222 | case KVM_SET_GUEST_DEBUG: { |
| 2223 | struct kvm_guest_debug dbg; |
| 2224 | |
| 2225 | r = -EFAULT; |
| 2226 | if (copy_from_user(&dbg, argp, sizeof(dbg))) |
| 2227 | goto out; |
| 2228 | r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg); |
| 2229 | break; |
| 2230 | } |
| 2231 | case KVM_SET_SIGNAL_MASK: { |
| 2232 | struct kvm_signal_mask __user *sigmask_arg = argp; |
| 2233 | struct kvm_signal_mask kvm_sigmask; |
| 2234 | sigset_t sigset, *p; |
| 2235 | |
| 2236 | p = NULL; |
| 2237 | if (argp) { |
| 2238 | r = -EFAULT; |
| 2239 | if (copy_from_user(&kvm_sigmask, argp, |
| 2240 | sizeof(kvm_sigmask))) |
| 2241 | goto out; |
| 2242 | r = -EINVAL; |
| 2243 | if (kvm_sigmask.len != sizeof(sigset)) |
| 2244 | goto out; |
| 2245 | r = -EFAULT; |
| 2246 | if (copy_from_user(&sigset, sigmask_arg->sigset, |
| 2247 | sizeof(sigset))) |
| 2248 | goto out; |
| 2249 | p = &sigset; |
| 2250 | } |
| 2251 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, p); |
| 2252 | break; |
| 2253 | } |
| 2254 | case KVM_GET_FPU: { |
| 2255 | fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL); |
| 2256 | r = -ENOMEM; |
| 2257 | if (!fpu) |
| 2258 | goto out; |
| 2259 | r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu); |
| 2260 | if (r) |
| 2261 | goto out; |
| 2262 | r = -EFAULT; |
| 2263 | if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu))) |
| 2264 | goto out; |
| 2265 | r = 0; |
| 2266 | break; |
| 2267 | } |
| 2268 | case KVM_SET_FPU: { |
| 2269 | fpu = memdup_user(argp, sizeof(*fpu)); |
| 2270 | if (IS_ERR(fpu)) { |
| 2271 | r = PTR_ERR(fpu); |
| 2272 | fpu = NULL; |
| 2273 | goto out; |
| 2274 | } |
| 2275 | r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu); |
| 2276 | break; |
| 2277 | } |
| 2278 | default: |
| 2279 | r = kvm_arch_vcpu_ioctl(filp, ioctl, arg); |
| 2280 | } |
| 2281 | out: |
| 2282 | vcpu_put(vcpu); |
| 2283 | kfree(fpu); |
| 2284 | kfree(kvm_sregs); |
| 2285 | return r; |
| 2286 | } |
| 2287 | |
| 2288 | #ifdef CONFIG_KVM_COMPAT |
| 2289 | static long kvm_vcpu_compat_ioctl(struct file *filp, |
| 2290 | unsigned int ioctl, unsigned long arg) |
| 2291 | { |
| 2292 | struct kvm_vcpu *vcpu = filp->private_data; |
| 2293 | void __user *argp = compat_ptr(arg); |
| 2294 | int r; |
| 2295 | |
| 2296 | if (vcpu->kvm->mm != current->mm) |
| 2297 | return -EIO; |
| 2298 | |
| 2299 | switch (ioctl) { |
| 2300 | case KVM_SET_SIGNAL_MASK: { |
| 2301 | struct kvm_signal_mask __user *sigmask_arg = argp; |
| 2302 | struct kvm_signal_mask kvm_sigmask; |
| 2303 | compat_sigset_t csigset; |
| 2304 | sigset_t sigset; |
| 2305 | |
| 2306 | if (argp) { |
| 2307 | r = -EFAULT; |
| 2308 | if (copy_from_user(&kvm_sigmask, argp, |
| 2309 | sizeof(kvm_sigmask))) |
| 2310 | goto out; |
| 2311 | r = -EINVAL; |
| 2312 | if (kvm_sigmask.len != sizeof(csigset)) |
| 2313 | goto out; |
| 2314 | r = -EFAULT; |
| 2315 | if (copy_from_user(&csigset, sigmask_arg->sigset, |
| 2316 | sizeof(csigset))) |
| 2317 | goto out; |
| 2318 | sigset_from_compat(&sigset, &csigset); |
| 2319 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset); |
| 2320 | } else |
| 2321 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL); |
| 2322 | break; |
| 2323 | } |
| 2324 | default: |
| 2325 | r = kvm_vcpu_ioctl(filp, ioctl, arg); |
| 2326 | } |
| 2327 | |
| 2328 | out: |
| 2329 | return r; |
| 2330 | } |
| 2331 | #endif |
| 2332 | |
| 2333 | static int kvm_device_ioctl_attr(struct kvm_device *dev, |
| 2334 | int (*accessor)(struct kvm_device *dev, |
| 2335 | struct kvm_device_attr *attr), |
| 2336 | unsigned long arg) |
| 2337 | { |
| 2338 | struct kvm_device_attr attr; |
| 2339 | |
| 2340 | if (!accessor) |
| 2341 | return -EPERM; |
| 2342 | |
| 2343 | if (copy_from_user(&attr, (void __user *)arg, sizeof(attr))) |
| 2344 | return -EFAULT; |
| 2345 | |
| 2346 | return accessor(dev, &attr); |
| 2347 | } |
| 2348 | |
| 2349 | static long kvm_device_ioctl(struct file *filp, unsigned int ioctl, |
| 2350 | unsigned long arg) |
| 2351 | { |
| 2352 | struct kvm_device *dev = filp->private_data; |
| 2353 | |
| 2354 | switch (ioctl) { |
| 2355 | case KVM_SET_DEVICE_ATTR: |
| 2356 | return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg); |
| 2357 | case KVM_GET_DEVICE_ATTR: |
| 2358 | return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg); |
| 2359 | case KVM_HAS_DEVICE_ATTR: |
| 2360 | return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg); |
| 2361 | default: |
| 2362 | if (dev->ops->ioctl) |
| 2363 | return dev->ops->ioctl(dev, ioctl, arg); |
| 2364 | |
| 2365 | return -ENOTTY; |
| 2366 | } |
| 2367 | } |
| 2368 | |
| 2369 | static int kvm_device_release(struct inode *inode, struct file *filp) |
| 2370 | { |
| 2371 | struct kvm_device *dev = filp->private_data; |
| 2372 | struct kvm *kvm = dev->kvm; |
| 2373 | |
| 2374 | kvm_put_kvm(kvm); |
| 2375 | return 0; |
| 2376 | } |
| 2377 | |
| 2378 | static const struct file_operations kvm_device_fops = { |
| 2379 | .unlocked_ioctl = kvm_device_ioctl, |
| 2380 | #ifdef CONFIG_KVM_COMPAT |
| 2381 | .compat_ioctl = kvm_device_ioctl, |
| 2382 | #endif |
| 2383 | .release = kvm_device_release, |
| 2384 | }; |
| 2385 | |
| 2386 | struct kvm_device *kvm_device_from_filp(struct file *filp) |
| 2387 | { |
| 2388 | if (filp->f_op != &kvm_device_fops) |
| 2389 | return NULL; |
| 2390 | |
| 2391 | return filp->private_data; |
| 2392 | } |
| 2393 | |
| 2394 | static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = { |
| 2395 | #ifdef CONFIG_KVM_MPIC |
| 2396 | [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops, |
| 2397 | [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops, |
| 2398 | #endif |
| 2399 | |
| 2400 | #ifdef CONFIG_KVM_XICS |
| 2401 | [KVM_DEV_TYPE_XICS] = &kvm_xics_ops, |
| 2402 | #endif |
| 2403 | }; |
| 2404 | |
| 2405 | int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type) |
| 2406 | { |
| 2407 | if (type >= ARRAY_SIZE(kvm_device_ops_table)) |
| 2408 | return -ENOSPC; |
| 2409 | |
| 2410 | if (kvm_device_ops_table[type] != NULL) |
| 2411 | return -EEXIST; |
| 2412 | |
| 2413 | kvm_device_ops_table[type] = ops; |
| 2414 | return 0; |
| 2415 | } |
| 2416 | |
| 2417 | void kvm_unregister_device_ops(u32 type) |
| 2418 | { |
| 2419 | if (kvm_device_ops_table[type] != NULL) |
| 2420 | kvm_device_ops_table[type] = NULL; |
| 2421 | } |
| 2422 | |
| 2423 | static int kvm_ioctl_create_device(struct kvm *kvm, |
| 2424 | struct kvm_create_device *cd) |
| 2425 | { |
| 2426 | struct kvm_device_ops *ops = NULL; |
| 2427 | struct kvm_device *dev; |
| 2428 | bool test = cd->flags & KVM_CREATE_DEVICE_TEST; |
| 2429 | int ret; |
| 2430 | |
| 2431 | if (cd->type >= ARRAY_SIZE(kvm_device_ops_table)) |
| 2432 | return -ENODEV; |
| 2433 | |
| 2434 | ops = kvm_device_ops_table[cd->type]; |
| 2435 | if (ops == NULL) |
| 2436 | return -ENODEV; |
| 2437 | |
| 2438 | if (test) |
| 2439 | return 0; |
| 2440 | |
| 2441 | dev = kzalloc(sizeof(*dev), GFP_KERNEL); |
| 2442 | if (!dev) |
| 2443 | return -ENOMEM; |
| 2444 | |
| 2445 | dev->ops = ops; |
| 2446 | dev->kvm = kvm; |
| 2447 | |
| 2448 | ret = ops->create(dev, cd->type); |
| 2449 | if (ret < 0) { |
| 2450 | kfree(dev); |
| 2451 | return ret; |
| 2452 | } |
| 2453 | |
| 2454 | ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC); |
| 2455 | if (ret < 0) { |
| 2456 | ops->destroy(dev); |
| 2457 | return ret; |
| 2458 | } |
| 2459 | |
| 2460 | list_add(&dev->vm_node, &kvm->devices); |
| 2461 | kvm_get_kvm(kvm); |
| 2462 | cd->fd = ret; |
| 2463 | return 0; |
| 2464 | } |
| 2465 | |
| 2466 | static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg) |
| 2467 | { |
| 2468 | switch (arg) { |
| 2469 | case KVM_CAP_USER_MEMORY: |
| 2470 | case KVM_CAP_DESTROY_MEMORY_REGION_WORKS: |
| 2471 | case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS: |
| 2472 | #ifdef CONFIG_KVM_APIC_ARCHITECTURE |
| 2473 | case KVM_CAP_SET_BOOT_CPU_ID: |
| 2474 | #endif |
| 2475 | case KVM_CAP_INTERNAL_ERROR_DATA: |
| 2476 | #ifdef CONFIG_HAVE_KVM_MSI |
| 2477 | case KVM_CAP_SIGNAL_MSI: |
| 2478 | #endif |
| 2479 | #ifdef CONFIG_HAVE_KVM_IRQFD |
| 2480 | case KVM_CAP_IRQFD: |
| 2481 | case KVM_CAP_IRQFD_RESAMPLE: |
| 2482 | #endif |
| 2483 | case KVM_CAP_CHECK_EXTENSION_VM: |
| 2484 | return 1; |
| 2485 | #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING |
| 2486 | case KVM_CAP_IRQ_ROUTING: |
| 2487 | return KVM_MAX_IRQ_ROUTES; |
| 2488 | #endif |
| 2489 | default: |
| 2490 | break; |
| 2491 | } |
| 2492 | return kvm_vm_ioctl_check_extension(kvm, arg); |
| 2493 | } |
| 2494 | |
| 2495 | static long kvm_vm_ioctl(struct file *filp, |
| 2496 | unsigned int ioctl, unsigned long arg) |
| 2497 | { |
| 2498 | struct kvm *kvm = filp->private_data; |
| 2499 | void __user *argp = (void __user *)arg; |
| 2500 | int r; |
| 2501 | |
| 2502 | if (kvm->mm != current->mm) |
| 2503 | return -EIO; |
| 2504 | switch (ioctl) { |
| 2505 | case KVM_CREATE_VCPU: |
| 2506 | r = kvm_vm_ioctl_create_vcpu(kvm, arg); |
| 2507 | break; |
| 2508 | case KVM_SET_USER_MEMORY_REGION: { |
| 2509 | struct kvm_userspace_memory_region kvm_userspace_mem; |
| 2510 | |
| 2511 | r = -EFAULT; |
| 2512 | if (copy_from_user(&kvm_userspace_mem, argp, |
| 2513 | sizeof(kvm_userspace_mem))) |
| 2514 | goto out; |
| 2515 | |
| 2516 | r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem); |
| 2517 | break; |
| 2518 | } |
| 2519 | case KVM_GET_DIRTY_LOG: { |
| 2520 | struct kvm_dirty_log log; |
| 2521 | |
| 2522 | r = -EFAULT; |
| 2523 | if (copy_from_user(&log, argp, sizeof(log))) |
| 2524 | goto out; |
| 2525 | r = kvm_vm_ioctl_get_dirty_log(kvm, &log); |
| 2526 | break; |
| 2527 | } |
| 2528 | #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET |
| 2529 | case KVM_REGISTER_COALESCED_MMIO: { |
| 2530 | struct kvm_coalesced_mmio_zone zone; |
| 2531 | |
| 2532 | r = -EFAULT; |
| 2533 | if (copy_from_user(&zone, argp, sizeof(zone))) |
| 2534 | goto out; |
| 2535 | r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone); |
| 2536 | break; |
| 2537 | } |
| 2538 | case KVM_UNREGISTER_COALESCED_MMIO: { |
| 2539 | struct kvm_coalesced_mmio_zone zone; |
| 2540 | |
| 2541 | r = -EFAULT; |
| 2542 | if (copy_from_user(&zone, argp, sizeof(zone))) |
| 2543 | goto out; |
| 2544 | r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone); |
| 2545 | break; |
| 2546 | } |
| 2547 | #endif |
| 2548 | case KVM_IRQFD: { |
| 2549 | struct kvm_irqfd data; |
| 2550 | |
| 2551 | r = -EFAULT; |
| 2552 | if (copy_from_user(&data, argp, sizeof(data))) |
| 2553 | goto out; |
| 2554 | r = kvm_irqfd(kvm, &data); |
| 2555 | break; |
| 2556 | } |
| 2557 | case KVM_IOEVENTFD: { |
| 2558 | struct kvm_ioeventfd data; |
| 2559 | |
| 2560 | r = -EFAULT; |
| 2561 | if (copy_from_user(&data, argp, sizeof(data))) |
| 2562 | goto out; |
| 2563 | r = kvm_ioeventfd(kvm, &data); |
| 2564 | break; |
| 2565 | } |
| 2566 | #ifdef CONFIG_KVM_APIC_ARCHITECTURE |
| 2567 | case KVM_SET_BOOT_CPU_ID: |
| 2568 | r = 0; |
| 2569 | mutex_lock(&kvm->lock); |
| 2570 | if (atomic_read(&kvm->online_vcpus) != 0) |
| 2571 | r = -EBUSY; |
| 2572 | else |
| 2573 | kvm->bsp_vcpu_id = arg; |
| 2574 | mutex_unlock(&kvm->lock); |
| 2575 | break; |
| 2576 | #endif |
| 2577 | #ifdef CONFIG_HAVE_KVM_MSI |
| 2578 | case KVM_SIGNAL_MSI: { |
| 2579 | struct kvm_msi msi; |
| 2580 | |
| 2581 | r = -EFAULT; |
| 2582 | if (copy_from_user(&msi, argp, sizeof(msi))) |
| 2583 | goto out; |
| 2584 | r = kvm_send_userspace_msi(kvm, &msi); |
| 2585 | break; |
| 2586 | } |
| 2587 | #endif |
| 2588 | #ifdef __KVM_HAVE_IRQ_LINE |
| 2589 | case KVM_IRQ_LINE_STATUS: |
| 2590 | case KVM_IRQ_LINE: { |
| 2591 | struct kvm_irq_level irq_event; |
| 2592 | |
| 2593 | r = -EFAULT; |
| 2594 | if (copy_from_user(&irq_event, argp, sizeof(irq_event))) |
| 2595 | goto out; |
| 2596 | |
| 2597 | r = kvm_vm_ioctl_irq_line(kvm, &irq_event, |
| 2598 | ioctl == KVM_IRQ_LINE_STATUS); |
| 2599 | if (r) |
| 2600 | goto out; |
| 2601 | |
| 2602 | r = -EFAULT; |
| 2603 | if (ioctl == KVM_IRQ_LINE_STATUS) { |
| 2604 | if (copy_to_user(argp, &irq_event, sizeof(irq_event))) |
| 2605 | goto out; |
| 2606 | } |
| 2607 | |
| 2608 | r = 0; |
| 2609 | break; |
| 2610 | } |
| 2611 | #endif |
| 2612 | #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING |
| 2613 | case KVM_SET_GSI_ROUTING: { |
| 2614 | struct kvm_irq_routing routing; |
| 2615 | struct kvm_irq_routing __user *urouting; |
| 2616 | struct kvm_irq_routing_entry *entries; |
| 2617 | |
| 2618 | r = -EFAULT; |
| 2619 | if (copy_from_user(&routing, argp, sizeof(routing))) |
| 2620 | goto out; |
| 2621 | r = -EINVAL; |
| 2622 | if (routing.nr >= KVM_MAX_IRQ_ROUTES) |
| 2623 | goto out; |
| 2624 | if (routing.flags) |
| 2625 | goto out; |
| 2626 | r = -ENOMEM; |
| 2627 | entries = vmalloc(routing.nr * sizeof(*entries)); |
| 2628 | if (!entries) |
| 2629 | goto out; |
| 2630 | r = -EFAULT; |
| 2631 | urouting = argp; |
| 2632 | if (copy_from_user(entries, urouting->entries, |
| 2633 | routing.nr * sizeof(*entries))) |
| 2634 | goto out_free_irq_routing; |
| 2635 | r = kvm_set_irq_routing(kvm, entries, routing.nr, |
| 2636 | routing.flags); |
| 2637 | out_free_irq_routing: |
| 2638 | vfree(entries); |
| 2639 | break; |
| 2640 | } |
| 2641 | #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */ |
| 2642 | case KVM_CREATE_DEVICE: { |
| 2643 | struct kvm_create_device cd; |
| 2644 | |
| 2645 | r = -EFAULT; |
| 2646 | if (copy_from_user(&cd, argp, sizeof(cd))) |
| 2647 | goto out; |
| 2648 | |
| 2649 | r = kvm_ioctl_create_device(kvm, &cd); |
| 2650 | if (r) |
| 2651 | goto out; |
| 2652 | |
| 2653 | r = -EFAULT; |
| 2654 | if (copy_to_user(argp, &cd, sizeof(cd))) |
| 2655 | goto out; |
| 2656 | |
| 2657 | r = 0; |
| 2658 | break; |
| 2659 | } |
| 2660 | case KVM_CHECK_EXTENSION: |
| 2661 | r = kvm_vm_ioctl_check_extension_generic(kvm, arg); |
| 2662 | break; |
| 2663 | default: |
| 2664 | r = kvm_arch_vm_ioctl(filp, ioctl, arg); |
| 2665 | } |
| 2666 | out: |
| 2667 | return r; |
| 2668 | } |
| 2669 | |
| 2670 | #ifdef CONFIG_KVM_COMPAT |
| 2671 | struct compat_kvm_dirty_log { |
| 2672 | __u32 slot; |
| 2673 | __u32 padding1; |
| 2674 | union { |
| 2675 | compat_uptr_t dirty_bitmap; /* one bit per page */ |
| 2676 | __u64 padding2; |
| 2677 | }; |
| 2678 | }; |
| 2679 | |
| 2680 | static long kvm_vm_compat_ioctl(struct file *filp, |
| 2681 | unsigned int ioctl, unsigned long arg) |
| 2682 | { |
| 2683 | struct kvm *kvm = filp->private_data; |
| 2684 | int r; |
| 2685 | |
| 2686 | if (kvm->mm != current->mm) |
| 2687 | return -EIO; |
| 2688 | switch (ioctl) { |
| 2689 | case KVM_GET_DIRTY_LOG: { |
| 2690 | struct compat_kvm_dirty_log compat_log; |
| 2691 | struct kvm_dirty_log log; |
| 2692 | |
| 2693 | r = -EFAULT; |
| 2694 | if (copy_from_user(&compat_log, (void __user *)arg, |
| 2695 | sizeof(compat_log))) |
| 2696 | goto out; |
| 2697 | log.slot = compat_log.slot; |
| 2698 | log.padding1 = compat_log.padding1; |
| 2699 | log.padding2 = compat_log.padding2; |
| 2700 | log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap); |
| 2701 | |
| 2702 | r = kvm_vm_ioctl_get_dirty_log(kvm, &log); |
| 2703 | break; |
| 2704 | } |
| 2705 | default: |
| 2706 | r = kvm_vm_ioctl(filp, ioctl, arg); |
| 2707 | } |
| 2708 | |
| 2709 | out: |
| 2710 | return r; |
| 2711 | } |
| 2712 | #endif |
| 2713 | |
| 2714 | static struct file_operations kvm_vm_fops = { |
| 2715 | .release = kvm_vm_release, |
| 2716 | .unlocked_ioctl = kvm_vm_ioctl, |
| 2717 | #ifdef CONFIG_KVM_COMPAT |
| 2718 | .compat_ioctl = kvm_vm_compat_ioctl, |
| 2719 | #endif |
| 2720 | .llseek = noop_llseek, |
| 2721 | }; |
| 2722 | |
| 2723 | static int kvm_dev_ioctl_create_vm(unsigned long type) |
| 2724 | { |
| 2725 | int r; |
| 2726 | struct kvm *kvm; |
| 2727 | |
| 2728 | kvm = kvm_create_vm(type); |
| 2729 | if (IS_ERR(kvm)) |
| 2730 | return PTR_ERR(kvm); |
| 2731 | #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET |
| 2732 | r = kvm_coalesced_mmio_init(kvm); |
| 2733 | if (r < 0) { |
| 2734 | kvm_put_kvm(kvm); |
| 2735 | return r; |
| 2736 | } |
| 2737 | #endif |
| 2738 | r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC); |
| 2739 | if (r < 0) |
| 2740 | kvm_put_kvm(kvm); |
| 2741 | |
| 2742 | return r; |
| 2743 | } |
| 2744 | |
| 2745 | static long kvm_dev_ioctl(struct file *filp, |
| 2746 | unsigned int ioctl, unsigned long arg) |
| 2747 | { |
| 2748 | long r = -EINVAL; |
| 2749 | |
| 2750 | switch (ioctl) { |
| 2751 | case KVM_GET_API_VERSION: |
| 2752 | if (arg) |
| 2753 | goto out; |
| 2754 | r = KVM_API_VERSION; |
| 2755 | break; |
| 2756 | case KVM_CREATE_VM: |
| 2757 | r = kvm_dev_ioctl_create_vm(arg); |
| 2758 | break; |
| 2759 | case KVM_CHECK_EXTENSION: |
| 2760 | r = kvm_vm_ioctl_check_extension_generic(NULL, arg); |
| 2761 | break; |
| 2762 | case KVM_GET_VCPU_MMAP_SIZE: |
| 2763 | if (arg) |
| 2764 | goto out; |
| 2765 | r = PAGE_SIZE; /* struct kvm_run */ |
| 2766 | #ifdef CONFIG_X86 |
| 2767 | r += PAGE_SIZE; /* pio data page */ |
| 2768 | #endif |
| 2769 | #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET |
| 2770 | r += PAGE_SIZE; /* coalesced mmio ring page */ |
| 2771 | #endif |
| 2772 | break; |
| 2773 | case KVM_TRACE_ENABLE: |
| 2774 | case KVM_TRACE_PAUSE: |
| 2775 | case KVM_TRACE_DISABLE: |
| 2776 | r = -EOPNOTSUPP; |
| 2777 | break; |
| 2778 | default: |
| 2779 | return kvm_arch_dev_ioctl(filp, ioctl, arg); |
| 2780 | } |
| 2781 | out: |
| 2782 | return r; |
| 2783 | } |
| 2784 | |
| 2785 | static struct file_operations kvm_chardev_ops = { |
| 2786 | .unlocked_ioctl = kvm_dev_ioctl, |
| 2787 | .compat_ioctl = kvm_dev_ioctl, |
| 2788 | .llseek = noop_llseek, |
| 2789 | }; |
| 2790 | |
| 2791 | static struct miscdevice kvm_dev = { |
| 2792 | KVM_MINOR, |
| 2793 | "kvm", |
| 2794 | &kvm_chardev_ops, |
| 2795 | }; |
| 2796 | |
| 2797 | static void hardware_enable_nolock(void *junk) |
| 2798 | { |
| 2799 | int cpu = raw_smp_processor_id(); |
| 2800 | int r; |
| 2801 | |
| 2802 | if (cpumask_test_cpu(cpu, cpus_hardware_enabled)) |
| 2803 | return; |
| 2804 | |
| 2805 | cpumask_set_cpu(cpu, cpus_hardware_enabled); |
| 2806 | |
| 2807 | r = kvm_arch_hardware_enable(); |
| 2808 | |
| 2809 | if (r) { |
| 2810 | cpumask_clear_cpu(cpu, cpus_hardware_enabled); |
| 2811 | atomic_inc(&hardware_enable_failed); |
| 2812 | pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu); |
| 2813 | } |
| 2814 | } |
| 2815 | |
| 2816 | static void hardware_enable(void) |
| 2817 | { |
| 2818 | raw_spin_lock(&kvm_count_lock); |
| 2819 | if (kvm_usage_count) |
| 2820 | hardware_enable_nolock(NULL); |
| 2821 | raw_spin_unlock(&kvm_count_lock); |
| 2822 | } |
| 2823 | |
| 2824 | static void hardware_disable_nolock(void *junk) |
| 2825 | { |
| 2826 | int cpu = raw_smp_processor_id(); |
| 2827 | |
| 2828 | if (!cpumask_test_cpu(cpu, cpus_hardware_enabled)) |
| 2829 | return; |
| 2830 | cpumask_clear_cpu(cpu, cpus_hardware_enabled); |
| 2831 | kvm_arch_hardware_disable(); |
| 2832 | } |
| 2833 | |
| 2834 | static void hardware_disable(void) |
| 2835 | { |
| 2836 | raw_spin_lock(&kvm_count_lock); |
| 2837 | if (kvm_usage_count) |
| 2838 | hardware_disable_nolock(NULL); |
| 2839 | raw_spin_unlock(&kvm_count_lock); |
| 2840 | } |
| 2841 | |
| 2842 | static void hardware_disable_all_nolock(void) |
| 2843 | { |
| 2844 | BUG_ON(!kvm_usage_count); |
| 2845 | |
| 2846 | kvm_usage_count--; |
| 2847 | if (!kvm_usage_count) |
| 2848 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
| 2849 | } |
| 2850 | |
| 2851 | static void hardware_disable_all(void) |
| 2852 | { |
| 2853 | raw_spin_lock(&kvm_count_lock); |
| 2854 | hardware_disable_all_nolock(); |
| 2855 | raw_spin_unlock(&kvm_count_lock); |
| 2856 | } |
| 2857 | |
| 2858 | static int hardware_enable_all(void) |
| 2859 | { |
| 2860 | int r = 0; |
| 2861 | |
| 2862 | raw_spin_lock(&kvm_count_lock); |
| 2863 | |
| 2864 | kvm_usage_count++; |
| 2865 | if (kvm_usage_count == 1) { |
| 2866 | atomic_set(&hardware_enable_failed, 0); |
| 2867 | on_each_cpu(hardware_enable_nolock, NULL, 1); |
| 2868 | |
| 2869 | if (atomic_read(&hardware_enable_failed)) { |
| 2870 | hardware_disable_all_nolock(); |
| 2871 | r = -EBUSY; |
| 2872 | } |
| 2873 | } |
| 2874 | |
| 2875 | raw_spin_unlock(&kvm_count_lock); |
| 2876 | |
| 2877 | return r; |
| 2878 | } |
| 2879 | |
| 2880 | static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val, |
| 2881 | void *v) |
| 2882 | { |
| 2883 | val &= ~CPU_TASKS_FROZEN; |
| 2884 | switch (val) { |
| 2885 | case CPU_DYING: |
| 2886 | hardware_disable(); |
| 2887 | break; |
| 2888 | case CPU_STARTING: |
| 2889 | hardware_enable(); |
| 2890 | break; |
| 2891 | } |
| 2892 | return NOTIFY_OK; |
| 2893 | } |
| 2894 | |
| 2895 | static int kvm_reboot(struct notifier_block *notifier, unsigned long val, |
| 2896 | void *v) |
| 2897 | { |
| 2898 | /* |
| 2899 | * Some (well, at least mine) BIOSes hang on reboot if |
| 2900 | * in vmx root mode. |
| 2901 | * |
| 2902 | * And Intel TXT required VMX off for all cpu when system shutdown. |
| 2903 | */ |
| 2904 | pr_info("kvm: exiting hardware virtualization\n"); |
| 2905 | kvm_rebooting = true; |
| 2906 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
| 2907 | return NOTIFY_OK; |
| 2908 | } |
| 2909 | |
| 2910 | static struct notifier_block kvm_reboot_notifier = { |
| 2911 | .notifier_call = kvm_reboot, |
| 2912 | .priority = 0, |
| 2913 | }; |
| 2914 | |
| 2915 | static void kvm_io_bus_destroy(struct kvm_io_bus *bus) |
| 2916 | { |
| 2917 | int i; |
| 2918 | |
| 2919 | for (i = 0; i < bus->dev_count; i++) { |
| 2920 | struct kvm_io_device *pos = bus->range[i].dev; |
| 2921 | |
| 2922 | kvm_iodevice_destructor(pos); |
| 2923 | } |
| 2924 | kfree(bus); |
| 2925 | } |
| 2926 | |
| 2927 | static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1, |
| 2928 | const struct kvm_io_range *r2) |
| 2929 | { |
| 2930 | if (r1->addr < r2->addr) |
| 2931 | return -1; |
| 2932 | if (r1->addr + r1->len > r2->addr + r2->len) |
| 2933 | return 1; |
| 2934 | return 0; |
| 2935 | } |
| 2936 | |
| 2937 | static int kvm_io_bus_sort_cmp(const void *p1, const void *p2) |
| 2938 | { |
| 2939 | return kvm_io_bus_cmp(p1, p2); |
| 2940 | } |
| 2941 | |
| 2942 | static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev, |
| 2943 | gpa_t addr, int len) |
| 2944 | { |
| 2945 | bus->range[bus->dev_count++] = (struct kvm_io_range) { |
| 2946 | .addr = addr, |
| 2947 | .len = len, |
| 2948 | .dev = dev, |
| 2949 | }; |
| 2950 | |
| 2951 | sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range), |
| 2952 | kvm_io_bus_sort_cmp, NULL); |
| 2953 | |
| 2954 | return 0; |
| 2955 | } |
| 2956 | |
| 2957 | static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus, |
| 2958 | gpa_t addr, int len) |
| 2959 | { |
| 2960 | struct kvm_io_range *range, key; |
| 2961 | int off; |
| 2962 | |
| 2963 | key = (struct kvm_io_range) { |
| 2964 | .addr = addr, |
| 2965 | .len = len, |
| 2966 | }; |
| 2967 | |
| 2968 | range = bsearch(&key, bus->range, bus->dev_count, |
| 2969 | sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp); |
| 2970 | if (range == NULL) |
| 2971 | return -ENOENT; |
| 2972 | |
| 2973 | off = range - bus->range; |
| 2974 | |
| 2975 | while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0) |
| 2976 | off--; |
| 2977 | |
| 2978 | return off; |
| 2979 | } |
| 2980 | |
| 2981 | static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus, |
| 2982 | struct kvm_io_range *range, const void *val) |
| 2983 | { |
| 2984 | int idx; |
| 2985 | |
| 2986 | idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len); |
| 2987 | if (idx < 0) |
| 2988 | return -EOPNOTSUPP; |
| 2989 | |
| 2990 | while (idx < bus->dev_count && |
| 2991 | kvm_io_bus_cmp(range, &bus->range[idx]) == 0) { |
| 2992 | if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr, |
| 2993 | range->len, val)) |
| 2994 | return idx; |
| 2995 | idx++; |
| 2996 | } |
| 2997 | |
| 2998 | return -EOPNOTSUPP; |
| 2999 | } |
| 3000 | |
| 3001 | /* kvm_io_bus_write - called under kvm->slots_lock */ |
| 3002 | int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr, |
| 3003 | int len, const void *val) |
| 3004 | { |
| 3005 | struct kvm_io_bus *bus; |
| 3006 | struct kvm_io_range range; |
| 3007 | int r; |
| 3008 | |
| 3009 | range = (struct kvm_io_range) { |
| 3010 | .addr = addr, |
| 3011 | .len = len, |
| 3012 | }; |
| 3013 | |
| 3014 | bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu); |
| 3015 | r = __kvm_io_bus_write(vcpu, bus, &range, val); |
| 3016 | return r < 0 ? r : 0; |
| 3017 | } |
| 3018 | |
| 3019 | /* kvm_io_bus_write_cookie - called under kvm->slots_lock */ |
| 3020 | int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, |
| 3021 | gpa_t addr, int len, const void *val, long cookie) |
| 3022 | { |
| 3023 | struct kvm_io_bus *bus; |
| 3024 | struct kvm_io_range range; |
| 3025 | |
| 3026 | range = (struct kvm_io_range) { |
| 3027 | .addr = addr, |
| 3028 | .len = len, |
| 3029 | }; |
| 3030 | |
| 3031 | bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu); |
| 3032 | |
| 3033 | /* First try the device referenced by cookie. */ |
| 3034 | if ((cookie >= 0) && (cookie < bus->dev_count) && |
| 3035 | (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0)) |
| 3036 | if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len, |
| 3037 | val)) |
| 3038 | return cookie; |
| 3039 | |
| 3040 | /* |
| 3041 | * cookie contained garbage; fall back to search and return the |
| 3042 | * correct cookie value. |
| 3043 | */ |
| 3044 | return __kvm_io_bus_write(vcpu, bus, &range, val); |
| 3045 | } |
| 3046 | |
| 3047 | static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus, |
| 3048 | struct kvm_io_range *range, void *val) |
| 3049 | { |
| 3050 | int idx; |
| 3051 | |
| 3052 | idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len); |
| 3053 | if (idx < 0) |
| 3054 | return -EOPNOTSUPP; |
| 3055 | |
| 3056 | while (idx < bus->dev_count && |
| 3057 | kvm_io_bus_cmp(range, &bus->range[idx]) == 0) { |
| 3058 | if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr, |
| 3059 | range->len, val)) |
| 3060 | return idx; |
| 3061 | idx++; |
| 3062 | } |
| 3063 | |
| 3064 | return -EOPNOTSUPP; |
| 3065 | } |
| 3066 | EXPORT_SYMBOL_GPL(kvm_io_bus_write); |
| 3067 | |
| 3068 | /* kvm_io_bus_read - called under kvm->slots_lock */ |
| 3069 | int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr, |
| 3070 | int len, void *val) |
| 3071 | { |
| 3072 | struct kvm_io_bus *bus; |
| 3073 | struct kvm_io_range range; |
| 3074 | int r; |
| 3075 | |
| 3076 | range = (struct kvm_io_range) { |
| 3077 | .addr = addr, |
| 3078 | .len = len, |
| 3079 | }; |
| 3080 | |
| 3081 | bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu); |
| 3082 | r = __kvm_io_bus_read(vcpu, bus, &range, val); |
| 3083 | return r < 0 ? r : 0; |
| 3084 | } |
| 3085 | |
| 3086 | |
| 3087 | /* Caller must hold slots_lock. */ |
| 3088 | int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr, |
| 3089 | int len, struct kvm_io_device *dev) |
| 3090 | { |
| 3091 | struct kvm_io_bus *new_bus, *bus; |
| 3092 | |
| 3093 | bus = kvm->buses[bus_idx]; |
| 3094 | /* exclude ioeventfd which is limited by maximum fd */ |
| 3095 | if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1) |
| 3096 | return -ENOSPC; |
| 3097 | |
| 3098 | new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) * |
| 3099 | sizeof(struct kvm_io_range)), GFP_KERNEL); |
| 3100 | if (!new_bus) |
| 3101 | return -ENOMEM; |
| 3102 | memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count * |
| 3103 | sizeof(struct kvm_io_range))); |
| 3104 | kvm_io_bus_insert_dev(new_bus, dev, addr, len); |
| 3105 | rcu_assign_pointer(kvm->buses[bus_idx], new_bus); |
| 3106 | synchronize_srcu_expedited(&kvm->srcu); |
| 3107 | kfree(bus); |
| 3108 | |
| 3109 | return 0; |
| 3110 | } |
| 3111 | |
| 3112 | /* Caller must hold slots_lock. */ |
| 3113 | int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx, |
| 3114 | struct kvm_io_device *dev) |
| 3115 | { |
| 3116 | int i, r; |
| 3117 | struct kvm_io_bus *new_bus, *bus; |
| 3118 | |
| 3119 | bus = kvm->buses[bus_idx]; |
| 3120 | r = -ENOENT; |
| 3121 | for (i = 0; i < bus->dev_count; i++) |
| 3122 | if (bus->range[i].dev == dev) { |
| 3123 | r = 0; |
| 3124 | break; |
| 3125 | } |
| 3126 | |
| 3127 | if (r) |
| 3128 | return r; |
| 3129 | |
| 3130 | new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) * |
| 3131 | sizeof(struct kvm_io_range)), GFP_KERNEL); |
| 3132 | if (!new_bus) |
| 3133 | return -ENOMEM; |
| 3134 | |
| 3135 | memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range)); |
| 3136 | new_bus->dev_count--; |
| 3137 | memcpy(new_bus->range + i, bus->range + i + 1, |
| 3138 | (new_bus->dev_count - i) * sizeof(struct kvm_io_range)); |
| 3139 | |
| 3140 | rcu_assign_pointer(kvm->buses[bus_idx], new_bus); |
| 3141 | synchronize_srcu_expedited(&kvm->srcu); |
| 3142 | kfree(bus); |
| 3143 | return r; |
| 3144 | } |
| 3145 | |
| 3146 | static struct notifier_block kvm_cpu_notifier = { |
| 3147 | .notifier_call = kvm_cpu_hotplug, |
| 3148 | }; |
| 3149 | |
| 3150 | static int vm_stat_get(void *_offset, u64 *val) |
| 3151 | { |
| 3152 | unsigned offset = (long)_offset; |
| 3153 | struct kvm *kvm; |
| 3154 | |
| 3155 | *val = 0; |
| 3156 | spin_lock(&kvm_lock); |
| 3157 | list_for_each_entry(kvm, &vm_list, vm_list) |
| 3158 | *val += *(u32 *)((void *)kvm + offset); |
| 3159 | spin_unlock(&kvm_lock); |
| 3160 | return 0; |
| 3161 | } |
| 3162 | |
| 3163 | DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n"); |
| 3164 | |
| 3165 | static int vcpu_stat_get(void *_offset, u64 *val) |
| 3166 | { |
| 3167 | unsigned offset = (long)_offset; |
| 3168 | struct kvm *kvm; |
| 3169 | struct kvm_vcpu *vcpu; |
| 3170 | int i; |
| 3171 | |
| 3172 | *val = 0; |
| 3173 | spin_lock(&kvm_lock); |
| 3174 | list_for_each_entry(kvm, &vm_list, vm_list) |
| 3175 | kvm_for_each_vcpu(i, vcpu, kvm) |
| 3176 | *val += *(u32 *)((void *)vcpu + offset); |
| 3177 | |
| 3178 | spin_unlock(&kvm_lock); |
| 3179 | return 0; |
| 3180 | } |
| 3181 | |
| 3182 | DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n"); |
| 3183 | |
| 3184 | static const struct file_operations *stat_fops[] = { |
| 3185 | [KVM_STAT_VCPU] = &vcpu_stat_fops, |
| 3186 | [KVM_STAT_VM] = &vm_stat_fops, |
| 3187 | }; |
| 3188 | |
| 3189 | static int kvm_init_debug(void) |
| 3190 | { |
| 3191 | int r = -EEXIST; |
| 3192 | struct kvm_stats_debugfs_item *p; |
| 3193 | |
| 3194 | kvm_debugfs_dir = debugfs_create_dir("kvm", NULL); |
| 3195 | if (kvm_debugfs_dir == NULL) |
| 3196 | goto out; |
| 3197 | |
| 3198 | for (p = debugfs_entries; p->name; ++p) { |
| 3199 | p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir, |
| 3200 | (void *)(long)p->offset, |
| 3201 | stat_fops[p->kind]); |
| 3202 | if (p->dentry == NULL) |
| 3203 | goto out_dir; |
| 3204 | } |
| 3205 | |
| 3206 | return 0; |
| 3207 | |
| 3208 | out_dir: |
| 3209 | debugfs_remove_recursive(kvm_debugfs_dir); |
| 3210 | out: |
| 3211 | return r; |
| 3212 | } |
| 3213 | |
| 3214 | static void kvm_exit_debug(void) |
| 3215 | { |
| 3216 | struct kvm_stats_debugfs_item *p; |
| 3217 | |
| 3218 | for (p = debugfs_entries; p->name; ++p) |
| 3219 | debugfs_remove(p->dentry); |
| 3220 | debugfs_remove(kvm_debugfs_dir); |
| 3221 | } |
| 3222 | |
| 3223 | static int kvm_suspend(void) |
| 3224 | { |
| 3225 | if (kvm_usage_count) |
| 3226 | hardware_disable_nolock(NULL); |
| 3227 | return 0; |
| 3228 | } |
| 3229 | |
| 3230 | static void kvm_resume(void) |
| 3231 | { |
| 3232 | if (kvm_usage_count) { |
| 3233 | WARN_ON(raw_spin_is_locked(&kvm_count_lock)); |
| 3234 | hardware_enable_nolock(NULL); |
| 3235 | } |
| 3236 | } |
| 3237 | |
| 3238 | static struct syscore_ops kvm_syscore_ops = { |
| 3239 | .suspend = kvm_suspend, |
| 3240 | .resume = kvm_resume, |
| 3241 | }; |
| 3242 | |
| 3243 | static inline |
| 3244 | struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn) |
| 3245 | { |
| 3246 | return container_of(pn, struct kvm_vcpu, preempt_notifier); |
| 3247 | } |
| 3248 | |
| 3249 | static void kvm_sched_in(struct preempt_notifier *pn, int cpu) |
| 3250 | { |
| 3251 | struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn); |
| 3252 | |
| 3253 | if (vcpu->preempted) |
| 3254 | vcpu->preempted = false; |
| 3255 | |
| 3256 | kvm_arch_sched_in(vcpu, cpu); |
| 3257 | |
| 3258 | kvm_arch_vcpu_load(vcpu, cpu); |
| 3259 | } |
| 3260 | |
| 3261 | static void kvm_sched_out(struct preempt_notifier *pn, |
| 3262 | struct task_struct *next) |
| 3263 | { |
| 3264 | struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn); |
| 3265 | |
| 3266 | if (current->state == TASK_RUNNING) |
| 3267 | vcpu->preempted = true; |
| 3268 | kvm_arch_vcpu_put(vcpu); |
| 3269 | } |
| 3270 | |
| 3271 | int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align, |
| 3272 | struct module *module) |
| 3273 | { |
| 3274 | int r; |
| 3275 | int cpu; |
| 3276 | |
| 3277 | r = kvm_arch_init(opaque); |
| 3278 | if (r) |
| 3279 | goto out_fail; |
| 3280 | |
| 3281 | /* |
| 3282 | * kvm_arch_init makes sure there's at most one caller |
| 3283 | * for architectures that support multiple implementations, |
| 3284 | * like intel and amd on x86. |
| 3285 | * kvm_arch_init must be called before kvm_irqfd_init to avoid creating |
| 3286 | * conflicts in case kvm is already setup for another implementation. |
| 3287 | */ |
| 3288 | r = kvm_irqfd_init(); |
| 3289 | if (r) |
| 3290 | goto out_irqfd; |
| 3291 | |
| 3292 | if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) { |
| 3293 | r = -ENOMEM; |
| 3294 | goto out_free_0; |
| 3295 | } |
| 3296 | |
| 3297 | r = kvm_arch_hardware_setup(); |
| 3298 | if (r < 0) |
| 3299 | goto out_free_0a; |
| 3300 | |
| 3301 | for_each_online_cpu(cpu) { |
| 3302 | smp_call_function_single(cpu, |
| 3303 | kvm_arch_check_processor_compat, |
| 3304 | &r, 1); |
| 3305 | if (r < 0) |
| 3306 | goto out_free_1; |
| 3307 | } |
| 3308 | |
| 3309 | r = register_cpu_notifier(&kvm_cpu_notifier); |
| 3310 | if (r) |
| 3311 | goto out_free_2; |
| 3312 | register_reboot_notifier(&kvm_reboot_notifier); |
| 3313 | |
| 3314 | /* A kmem cache lets us meet the alignment requirements of fx_save. */ |
| 3315 | if (!vcpu_align) |
| 3316 | vcpu_align = __alignof__(struct kvm_vcpu); |
| 3317 | kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align, |
| 3318 | 0, NULL); |
| 3319 | if (!kvm_vcpu_cache) { |
| 3320 | r = -ENOMEM; |
| 3321 | goto out_free_3; |
| 3322 | } |
| 3323 | |
| 3324 | r = kvm_async_pf_init(); |
| 3325 | if (r) |
| 3326 | goto out_free; |
| 3327 | |
| 3328 | kvm_chardev_ops.owner = module; |
| 3329 | kvm_vm_fops.owner = module; |
| 3330 | kvm_vcpu_fops.owner = module; |
| 3331 | |
| 3332 | r = misc_register(&kvm_dev); |
| 3333 | if (r) { |
| 3334 | pr_err("kvm: misc device register failed\n"); |
| 3335 | goto out_unreg; |
| 3336 | } |
| 3337 | |
| 3338 | register_syscore_ops(&kvm_syscore_ops); |
| 3339 | |
| 3340 | kvm_preempt_ops.sched_in = kvm_sched_in; |
| 3341 | kvm_preempt_ops.sched_out = kvm_sched_out; |
| 3342 | |
| 3343 | r = kvm_init_debug(); |
| 3344 | if (r) { |
| 3345 | pr_err("kvm: create debugfs files failed\n"); |
| 3346 | goto out_undebugfs; |
| 3347 | } |
| 3348 | |
| 3349 | r = kvm_vfio_ops_init(); |
| 3350 | WARN_ON(r); |
| 3351 | |
| 3352 | return 0; |
| 3353 | |
| 3354 | out_undebugfs: |
| 3355 | unregister_syscore_ops(&kvm_syscore_ops); |
| 3356 | misc_deregister(&kvm_dev); |
| 3357 | out_unreg: |
| 3358 | kvm_async_pf_deinit(); |
| 3359 | out_free: |
| 3360 | kmem_cache_destroy(kvm_vcpu_cache); |
| 3361 | out_free_3: |
| 3362 | unregister_reboot_notifier(&kvm_reboot_notifier); |
| 3363 | unregister_cpu_notifier(&kvm_cpu_notifier); |
| 3364 | out_free_2: |
| 3365 | out_free_1: |
| 3366 | kvm_arch_hardware_unsetup(); |
| 3367 | out_free_0a: |
| 3368 | free_cpumask_var(cpus_hardware_enabled); |
| 3369 | out_free_0: |
| 3370 | kvm_irqfd_exit(); |
| 3371 | out_irqfd: |
| 3372 | kvm_arch_exit(); |
| 3373 | out_fail: |
| 3374 | return r; |
| 3375 | } |
| 3376 | EXPORT_SYMBOL_GPL(kvm_init); |
| 3377 | |
| 3378 | void kvm_exit(void) |
| 3379 | { |
| 3380 | kvm_exit_debug(); |
| 3381 | misc_deregister(&kvm_dev); |
| 3382 | kmem_cache_destroy(kvm_vcpu_cache); |
| 3383 | kvm_async_pf_deinit(); |
| 3384 | unregister_syscore_ops(&kvm_syscore_ops); |
| 3385 | unregister_reboot_notifier(&kvm_reboot_notifier); |
| 3386 | unregister_cpu_notifier(&kvm_cpu_notifier); |
| 3387 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
| 3388 | kvm_arch_hardware_unsetup(); |
| 3389 | kvm_arch_exit(); |
| 3390 | kvm_irqfd_exit(); |
| 3391 | free_cpumask_var(cpus_hardware_enabled); |
| 3392 | kvm_vfio_ops_exit(); |
| 3393 | } |
| 3394 | EXPORT_SYMBOL_GPL(kvm_exit); |