1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause 3 * 4 * Copyright (c) 2011 NetApp, Inc. 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 16 * THIS SOFTWARE IS PROVIDED BY NETAPP, INC ``AS IS'' AND 17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 19 * ARE DISCLAIMED. IN NO EVENT SHALL NETAPP, INC OR CONTRIBUTORS BE LIABLE 20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 26 * SUCH DAMAGE. 27 */ 28 29 #include "opt_bhyve_snapshot.h" 30 31 #include <sys/param.h> 32 #include <sys/systm.h> 33 #include <sys/kernel.h> 34 #include <sys/module.h> 35 #include <sys/sysctl.h> 36 #include <sys/malloc.h> 37 #include <sys/pcpu.h> 38 #include <sys/lock.h> 39 #include <sys/mutex.h> 40 #include <sys/proc.h> 41 #include <sys/rwlock.h> 42 #include <sys/sched.h> 43 #include <sys/smp.h> 44 #include <sys/sx.h> 45 #include <sys/vnode.h> 46 47 #include <vm/vm.h> 48 #include <vm/vm_param.h> 49 #include <vm/vm_extern.h> 50 #include <vm/vm_object.h> 51 #include <vm/vm_page.h> 52 #include <vm/pmap.h> 53 #include <vm/vm_map.h> 54 #include <vm/vm_pager.h> 55 #include <vm/vm_kern.h> 56 #include <vm/vnode_pager.h> 57 #include <vm/swap_pager.h> 58 #include <vm/uma.h> 59 60 #include <machine/cpu.h> 61 #include <machine/pcb.h> 62 #include <machine/smp.h> 63 #include <machine/md_var.h> 64 #include <x86/psl.h> 65 #include <x86/apicreg.h> 66 #include <x86/ifunc.h> 67 68 #include <machine/vmm.h> 69 #include <machine/vmm_instruction_emul.h> 70 #include <machine/vmm_snapshot.h> 71 72 #include <dev/vmm/vmm_dev.h> 73 #include <dev/vmm/vmm_ktr.h> 74 #include <dev/vmm/vmm_mem.h> 75 76 #include "vmm_ioport.h" 77 #include "vmm_host.h" 78 #include "vmm_mem.h" 79 #include "vmm_util.h" 80 #include "vatpic.h" 81 #include "vatpit.h" 82 #include "vhpet.h" 83 #include "vioapic.h" 84 #include "vlapic.h" 85 #include "vpmtmr.h" 86 #include "vrtc.h" 87 #include "vmm_stat.h" 88 #include "vmm_lapic.h" 89 90 #include "io/ppt.h" 91 #include "io/iommu.h" 92 93 struct vlapic; 94 95 /* 96 * Initialization: 97 * (a) allocated when vcpu is created 98 * (i) initialized when vcpu is created and when it is reinitialized 99 * (o) initialized the first time the vcpu is created 100 * (x) initialized before use 101 */ 102 struct vcpu { 103 struct mtx mtx; /* (o) protects 'state' and 'hostcpu' */ 104 enum vcpu_state state; /* (o) vcpu state */ 105 int vcpuid; /* (o) */ 106 int hostcpu; /* (o) vcpu's host cpu */ 107 int reqidle; /* (i) request vcpu to idle */ 108 struct vm *vm; /* (o) */ 109 void *cookie; /* (i) cpu-specific data */ 110 struct vlapic *vlapic; /* (i) APIC device model */ 111 enum x2apic_state x2apic_state; /* (i) APIC mode */ 112 uint64_t exitintinfo; /* (i) events pending at VM exit */ 113 int nmi_pending; /* (i) NMI pending */ 114 int extint_pending; /* (i) INTR pending */ 115 int exception_pending; /* (i) exception pending */ 116 int exc_vector; /* (x) exception collateral */ 117 int exc_errcode_valid; 118 uint32_t exc_errcode; 119 struct savefpu *guestfpu; /* (a,i) guest fpu state */ 120 uint64_t guest_xcr0; /* (i) guest %xcr0 register */ 121 void *stats; /* (a,i) statistics */ 122 struct vm_exit exitinfo; /* (x) exit reason and collateral */ 123 cpuset_t exitinfo_cpuset; /* (x) storage for vmexit handlers */ 124 uint64_t nextrip; /* (x) next instruction to execute */ 125 uint64_t tsc_offset; /* (o) TSC offsetting */ 126 }; 127 128 #define vcpu_lock_init(v) mtx_init(&((v)->mtx), "vcpu lock", 0, MTX_SPIN) 129 #define vcpu_lock_destroy(v) mtx_destroy(&((v)->mtx)) 130 #define vcpu_lock(v) mtx_lock_spin(&((v)->mtx)) 131 #define vcpu_unlock(v) mtx_unlock_spin(&((v)->mtx)) 132 #define vcpu_assert_locked(v) mtx_assert(&((v)->mtx), MA_OWNED) 133 134 /* 135 * Initialization: 136 * (o) initialized the first time the VM is created 137 * (i) initialized when VM is created and when it is reinitialized 138 * (x) initialized before use 139 * 140 * Locking: 141 * [m] mem_segs_lock 142 * [r] rendezvous_mtx 143 * [v] reads require one frozen vcpu, writes require freezing all vcpus 144 */ 145 struct vm { 146 void *cookie; /* (i) cpu-specific data */ 147 void *iommu; /* (x) iommu-specific data */ 148 struct vhpet *vhpet; /* (i) virtual HPET */ 149 struct vioapic *vioapic; /* (i) virtual ioapic */ 150 struct vatpic *vatpic; /* (i) virtual atpic */ 151 struct vatpit *vatpit; /* (i) virtual atpit */ 152 struct vpmtmr *vpmtmr; /* (i) virtual ACPI PM timer */ 153 struct vrtc *vrtc; /* (o) virtual RTC */ 154 volatile cpuset_t active_cpus; /* (i) active vcpus */ 155 volatile cpuset_t debug_cpus; /* (i) vcpus stopped for debug */ 156 cpuset_t startup_cpus; /* (i) [r] waiting for startup */ 157 int suspend; /* (i) stop VM execution */ 158 bool dying; /* (o) is dying */ 159 volatile cpuset_t suspended_cpus; /* (i) suspended vcpus */ 160 volatile cpuset_t halted_cpus; /* (x) cpus in a hard halt */ 161 cpuset_t rendezvous_req_cpus; /* (x) [r] rendezvous requested */ 162 cpuset_t rendezvous_done_cpus; /* (x) [r] rendezvous finished */ 163 void *rendezvous_arg; /* (x) [r] rendezvous func/arg */ 164 vm_rendezvous_func_t rendezvous_func; 165 struct mtx rendezvous_mtx; /* (o) rendezvous lock */ 166 struct vm_mem mem; /* (i) [m+v] guest memory */ 167 char name[VM_MAX_NAMELEN+1]; /* (o) virtual machine name */ 168 struct vcpu **vcpu; /* (o) guest vcpus */ 169 /* The following describe the vm cpu topology */ 170 uint16_t sockets; /* (o) num of sockets */ 171 uint16_t cores; /* (o) num of cores/socket */ 172 uint16_t threads; /* (o) num of threads/core */ 173 uint16_t maxcpus; /* (o) max pluggable cpus */ 174 struct sx vcpus_init_lock; /* (o) */ 175 }; 176 177 #define VMM_CTR0(vcpu, format) \ 178 VCPU_CTR0((vcpu)->vm, (vcpu)->vcpuid, format) 179 180 #define VMM_CTR1(vcpu, format, p1) \ 181 VCPU_CTR1((vcpu)->vm, (vcpu)->vcpuid, format, p1) 182 183 #define VMM_CTR2(vcpu, format, p1, p2) \ 184 VCPU_CTR2((vcpu)->vm, (vcpu)->vcpuid, format, p1, p2) 185 186 #define VMM_CTR3(vcpu, format, p1, p2, p3) \ 187 VCPU_CTR3((vcpu)->vm, (vcpu)->vcpuid, format, p1, p2, p3) 188 189 #define VMM_CTR4(vcpu, format, p1, p2, p3, p4) \ 190 VCPU_CTR4((vcpu)->vm, (vcpu)->vcpuid, format, p1, p2, p3, p4) 191 192 static int vmm_initialized; 193 194 static void vmmops_panic(void); 195 196 static void 197 vmmops_panic(void) 198 { 199 panic("vmm_ops func called when !vmm_is_intel() && !vmm_is_svm()"); 200 } 201 202 #define DEFINE_VMMOPS_IFUNC(ret_type, opname, args) \ 203 DEFINE_IFUNC(, ret_type, vmmops_##opname, args) \ 204 { \ 205 if (vmm_is_intel()) \ 206 return (vmm_ops_intel.opname); \ 207 else if (vmm_is_svm()) \ 208 return (vmm_ops_amd.opname); \ 209 else \ 210 return ((ret_type (*)args)vmmops_panic); \ 211 } 212 213 DEFINE_VMMOPS_IFUNC(int, modinit, (int ipinum)) 214 DEFINE_VMMOPS_IFUNC(int, modcleanup, (void)) 215 DEFINE_VMMOPS_IFUNC(void, modsuspend, (void)) 216 DEFINE_VMMOPS_IFUNC(void, modresume, (void)) 217 DEFINE_VMMOPS_IFUNC(void *, init, (struct vm *vm, struct pmap *pmap)) 218 DEFINE_VMMOPS_IFUNC(int, run, (void *vcpui, register_t rip, struct pmap *pmap, 219 struct vm_eventinfo *info)) 220 DEFINE_VMMOPS_IFUNC(void, cleanup, (void *vmi)) 221 DEFINE_VMMOPS_IFUNC(void *, vcpu_init, (void *vmi, struct vcpu *vcpu, 222 int vcpu_id)) 223 DEFINE_VMMOPS_IFUNC(void, vcpu_cleanup, (void *vcpui)) 224 DEFINE_VMMOPS_IFUNC(int, getreg, (void *vcpui, int num, uint64_t *retval)) 225 DEFINE_VMMOPS_IFUNC(int, setreg, (void *vcpui, int num, uint64_t val)) 226 DEFINE_VMMOPS_IFUNC(int, getdesc, (void *vcpui, int num, struct seg_desc *desc)) 227 DEFINE_VMMOPS_IFUNC(int, setdesc, (void *vcpui, int num, struct seg_desc *desc)) 228 DEFINE_VMMOPS_IFUNC(int, getcap, (void *vcpui, int num, int *retval)) 229 DEFINE_VMMOPS_IFUNC(int, setcap, (void *vcpui, int num, int val)) 230 DEFINE_VMMOPS_IFUNC(struct vmspace *, vmspace_alloc, (vm_offset_t min, 231 vm_offset_t max)) 232 DEFINE_VMMOPS_IFUNC(void, vmspace_free, (struct vmspace *vmspace)) 233 DEFINE_VMMOPS_IFUNC(struct vlapic *, vlapic_init, (void *vcpui)) 234 DEFINE_VMMOPS_IFUNC(void, vlapic_cleanup, (struct vlapic *vlapic)) 235 #ifdef BHYVE_SNAPSHOT 236 DEFINE_VMMOPS_IFUNC(int, vcpu_snapshot, (void *vcpui, 237 struct vm_snapshot_meta *meta)) 238 DEFINE_VMMOPS_IFUNC(int, restore_tsc, (void *vcpui, uint64_t now)) 239 #endif 240 241 SDT_PROVIDER_DEFINE(vmm); 242 243 static MALLOC_DEFINE(M_VM, "vm", "vm"); 244 245 /* statistics */ 246 static VMM_STAT(VCPU_TOTAL_RUNTIME, "vcpu total runtime"); 247 248 SYSCTL_NODE(_hw, OID_AUTO, vmm, CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, 249 NULL); 250 251 /* 252 * Halt the guest if all vcpus are executing a HLT instruction with 253 * interrupts disabled. 254 */ 255 static int halt_detection_enabled = 1; 256 SYSCTL_INT(_hw_vmm, OID_AUTO, halt_detection, CTLFLAG_RDTUN, 257 &halt_detection_enabled, 0, 258 "Halt VM if all vcpus execute HLT with interrupts disabled"); 259 260 static int vmm_ipinum; 261 SYSCTL_INT(_hw_vmm, OID_AUTO, ipinum, CTLFLAG_RD, &vmm_ipinum, 0, 262 "IPI vector used for vcpu notifications"); 263 264 static int trace_guest_exceptions; 265 SYSCTL_INT(_hw_vmm, OID_AUTO, trace_guest_exceptions, CTLFLAG_RDTUN, 266 &trace_guest_exceptions, 0, 267 "Trap into hypervisor on all guest exceptions and reflect them back"); 268 269 static int trap_wbinvd; 270 SYSCTL_INT(_hw_vmm, OID_AUTO, trap_wbinvd, CTLFLAG_RDTUN, &trap_wbinvd, 0, 271 "WBINVD triggers a VM-exit"); 272 273 u_int vm_maxcpu; 274 SYSCTL_UINT(_hw_vmm, OID_AUTO, maxcpu, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, 275 &vm_maxcpu, 0, "Maximum number of vCPUs"); 276 277 static void vcpu_notify_event_locked(struct vcpu *vcpu, bool lapic_intr); 278 279 /* global statistics */ 280 VMM_STAT(VCPU_MIGRATIONS, "vcpu migration across host cpus"); 281 VMM_STAT(VMEXIT_COUNT, "total number of vm exits"); 282 VMM_STAT(VMEXIT_EXTINT, "vm exits due to external interrupt"); 283 VMM_STAT(VMEXIT_HLT, "number of times hlt was intercepted"); 284 VMM_STAT(VMEXIT_CR_ACCESS, "number of times %cr access was intercepted"); 285 VMM_STAT(VMEXIT_RDMSR, "number of times rdmsr was intercepted"); 286 VMM_STAT(VMEXIT_WRMSR, "number of times wrmsr was intercepted"); 287 VMM_STAT(VMEXIT_MTRAP, "number of monitor trap exits"); 288 VMM_STAT(VMEXIT_PAUSE, "number of times pause was intercepted"); 289 VMM_STAT(VMEXIT_INTR_WINDOW, "vm exits due to interrupt window opening"); 290 VMM_STAT(VMEXIT_NMI_WINDOW, "vm exits due to nmi window opening"); 291 VMM_STAT(VMEXIT_INOUT, "number of times in/out was intercepted"); 292 VMM_STAT(VMEXIT_CPUID, "number of times cpuid was intercepted"); 293 VMM_STAT(VMEXIT_NESTED_FAULT, "vm exits due to nested page fault"); 294 VMM_STAT(VMEXIT_INST_EMUL, "vm exits for instruction emulation"); 295 VMM_STAT(VMEXIT_UNKNOWN, "number of vm exits for unknown reason"); 296 VMM_STAT(VMEXIT_ASTPENDING, "number of times astpending at exit"); 297 VMM_STAT(VMEXIT_REQIDLE, "number of times idle requested at exit"); 298 VMM_STAT(VMEXIT_USERSPACE, "number of vm exits handled in userspace"); 299 VMM_STAT(VMEXIT_RENDEZVOUS, "number of times rendezvous pending at exit"); 300 VMM_STAT(VMEXIT_EXCEPTION, "number of vm exits due to exceptions"); 301 302 /* 303 * Upper limit on vm_maxcpu. Limited by use of uint16_t types for CPU 304 * counts as well as range of vpid values for VT-x and by the capacity 305 * of cpuset_t masks. The call to new_unrhdr() in vpid_init() in 306 * vmx.c requires 'vm_maxcpu + 1 <= 0xffff', hence the '- 1' below. 307 */ 308 #define VM_MAXCPU MIN(0xffff - 1, CPU_SETSIZE) 309 310 #ifdef KTR 311 static const char * 312 vcpu_state2str(enum vcpu_state state) 313 { 314 315 switch (state) { 316 case VCPU_IDLE: 317 return ("idle"); 318 case VCPU_FROZEN: 319 return ("frozen"); 320 case VCPU_RUNNING: 321 return ("running"); 322 case VCPU_SLEEPING: 323 return ("sleeping"); 324 default: 325 return ("unknown"); 326 } 327 } 328 #endif 329 330 static void 331 vcpu_cleanup(struct vcpu *vcpu, bool destroy) 332 { 333 vmmops_vlapic_cleanup(vcpu->vlapic); 334 vmmops_vcpu_cleanup(vcpu->cookie); 335 vcpu->cookie = NULL; 336 if (destroy) { 337 vmm_stat_free(vcpu->stats); 338 fpu_save_area_free(vcpu->guestfpu); 339 vcpu_lock_destroy(vcpu); 340 free(vcpu, M_VM); 341 } 342 } 343 344 static struct vcpu * 345 vcpu_alloc(struct vm *vm, int vcpu_id) 346 { 347 struct vcpu *vcpu; 348 349 KASSERT(vcpu_id >= 0 && vcpu_id < vm->maxcpus, 350 ("vcpu_init: invalid vcpu %d", vcpu_id)); 351 352 vcpu = malloc(sizeof(*vcpu), M_VM, M_WAITOK | M_ZERO); 353 vcpu_lock_init(vcpu); 354 vcpu->state = VCPU_IDLE; 355 vcpu->hostcpu = NOCPU; 356 vcpu->vcpuid = vcpu_id; 357 vcpu->vm = vm; 358 vcpu->guestfpu = fpu_save_area_alloc(); 359 vcpu->stats = vmm_stat_alloc(); 360 vcpu->tsc_offset = 0; 361 return (vcpu); 362 } 363 364 static void 365 vcpu_init(struct vcpu *vcpu) 366 { 367 vcpu->cookie = vmmops_vcpu_init(vcpu->vm->cookie, vcpu, vcpu->vcpuid); 368 vcpu->vlapic = vmmops_vlapic_init(vcpu->cookie); 369 vm_set_x2apic_state(vcpu, X2APIC_DISABLED); 370 vcpu->reqidle = 0; 371 vcpu->exitintinfo = 0; 372 vcpu->nmi_pending = 0; 373 vcpu->extint_pending = 0; 374 vcpu->exception_pending = 0; 375 vcpu->guest_xcr0 = XFEATURE_ENABLED_X87; 376 fpu_save_area_reset(vcpu->guestfpu); 377 vmm_stat_init(vcpu->stats); 378 } 379 380 int 381 vcpu_trace_exceptions(struct vcpu *vcpu) 382 { 383 384 return (trace_guest_exceptions); 385 } 386 387 int 388 vcpu_trap_wbinvd(struct vcpu *vcpu) 389 { 390 return (trap_wbinvd); 391 } 392 393 struct vm_exit * 394 vm_exitinfo(struct vcpu *vcpu) 395 { 396 return (&vcpu->exitinfo); 397 } 398 399 cpuset_t * 400 vm_exitinfo_cpuset(struct vcpu *vcpu) 401 { 402 return (&vcpu->exitinfo_cpuset); 403 } 404 405 static int 406 vmm_init(void) 407 { 408 if (!vmm_is_hw_supported()) 409 return (ENXIO); 410 411 vm_maxcpu = mp_ncpus; 412 TUNABLE_INT_FETCH("hw.vmm.maxcpu", &vm_maxcpu); 413 414 if (vm_maxcpu > VM_MAXCPU) { 415 printf("vmm: vm_maxcpu clamped to %u\n", VM_MAXCPU); 416 vm_maxcpu = VM_MAXCPU; 417 } 418 if (vm_maxcpu == 0) 419 vm_maxcpu = 1; 420 421 vmm_host_state_init(); 422 423 vmm_ipinum = lapic_ipi_alloc(pti ? &IDTVEC(justreturn1_pti) : 424 &IDTVEC(justreturn)); 425 if (vmm_ipinum < 0) 426 vmm_ipinum = IPI_AST; 427 428 vmm_suspend_p = vmmops_modsuspend; 429 vmm_resume_p = vmmops_modresume; 430 431 return (vmmops_modinit(vmm_ipinum)); 432 } 433 434 static int 435 vmm_handler(module_t mod, int what, void *arg) 436 { 437 int error; 438 439 switch (what) { 440 case MOD_LOAD: 441 if (vmm_is_hw_supported()) { 442 error = vmmdev_init(); 443 if (error != 0) 444 break; 445 error = vmm_init(); 446 if (error == 0) 447 vmm_initialized = 1; 448 else 449 (void)vmmdev_cleanup(); 450 } else { 451 error = ENXIO; 452 } 453 break; 454 case MOD_UNLOAD: 455 if (vmm_is_hw_supported()) { 456 error = vmmdev_cleanup(); 457 if (error == 0) { 458 vmm_suspend_p = NULL; 459 vmm_resume_p = NULL; 460 iommu_cleanup(); 461 if (vmm_ipinum != IPI_AST) 462 lapic_ipi_free(vmm_ipinum); 463 error = vmmops_modcleanup(); 464 /* 465 * Something bad happened - prevent new 466 * VMs from being created 467 */ 468 if (error) 469 vmm_initialized = 0; 470 } 471 } else { 472 error = 0; 473 } 474 break; 475 default: 476 error = 0; 477 break; 478 } 479 return (error); 480 } 481 482 static moduledata_t vmm_kmod = { 483 "vmm", 484 vmm_handler, 485 NULL 486 }; 487 488 /* 489 * vmm initialization has the following dependencies: 490 * 491 * - VT-x initialization requires smp_rendezvous() and therefore must happen 492 * after SMP is fully functional (after SI_SUB_SMP). 493 * - vmm device initialization requires an initialized devfs. 494 */ 495 DECLARE_MODULE(vmm, vmm_kmod, MAX(SI_SUB_SMP, SI_SUB_DEVFS) + 1, SI_ORDER_ANY); 496 MODULE_VERSION(vmm, 1); 497 498 static void 499 vm_init(struct vm *vm, bool create) 500 { 501 vm->cookie = vmmops_init(vm, vmspace_pmap(vm_vmspace(vm))); 502 vm->iommu = NULL; 503 vm->vioapic = vioapic_init(vm); 504 vm->vhpet = vhpet_init(vm); 505 vm->vatpic = vatpic_init(vm); 506 vm->vatpit = vatpit_init(vm); 507 vm->vpmtmr = vpmtmr_init(vm); 508 if (create) 509 vm->vrtc = vrtc_init(vm); 510 511 CPU_ZERO(&vm->active_cpus); 512 CPU_ZERO(&vm->debug_cpus); 513 CPU_ZERO(&vm->startup_cpus); 514 515 vm->suspend = 0; 516 CPU_ZERO(&vm->suspended_cpus); 517 518 if (!create) { 519 for (int i = 0; i < vm->maxcpus; i++) { 520 if (vm->vcpu[i] != NULL) 521 vcpu_init(vm->vcpu[i]); 522 } 523 } 524 } 525 526 void 527 vm_disable_vcpu_creation(struct vm *vm) 528 { 529 sx_xlock(&vm->vcpus_init_lock); 530 vm->dying = true; 531 sx_xunlock(&vm->vcpus_init_lock); 532 } 533 534 struct vcpu * 535 vm_alloc_vcpu(struct vm *vm, int vcpuid) 536 { 537 struct vcpu *vcpu; 538 539 if (vcpuid < 0 || vcpuid >= vm_get_maxcpus(vm)) 540 return (NULL); 541 542 vcpu = (struct vcpu *) 543 atomic_load_acq_ptr((uintptr_t *)&vm->vcpu[vcpuid]); 544 if (__predict_true(vcpu != NULL)) 545 return (vcpu); 546 547 sx_xlock(&vm->vcpus_init_lock); 548 vcpu = vm->vcpu[vcpuid]; 549 if (vcpu == NULL && !vm->dying) { 550 vcpu = vcpu_alloc(vm, vcpuid); 551 vcpu_init(vcpu); 552 553 /* 554 * Ensure vCPU is fully created before updating pointer 555 * to permit unlocked reads above. 556 */ 557 atomic_store_rel_ptr((uintptr_t *)&vm->vcpu[vcpuid], 558 (uintptr_t)vcpu); 559 } 560 sx_xunlock(&vm->vcpus_init_lock); 561 return (vcpu); 562 } 563 564 void 565 vm_slock_vcpus(struct vm *vm) 566 { 567 sx_slock(&vm->vcpus_init_lock); 568 } 569 570 void 571 vm_unlock_vcpus(struct vm *vm) 572 { 573 sx_unlock(&vm->vcpus_init_lock); 574 } 575 576 /* 577 * The default CPU topology is a single thread per package. 578 */ 579 u_int cores_per_package = 1; 580 u_int threads_per_core = 1; 581 582 int 583 vm_create(const char *name, struct vm **retvm) 584 { 585 struct vm *vm; 586 int error; 587 588 /* 589 * If vmm.ko could not be successfully initialized then don't attempt 590 * to create the virtual machine. 591 */ 592 if (!vmm_initialized) 593 return (ENXIO); 594 595 if (name == NULL || strnlen(name, VM_MAX_NAMELEN + 1) == 596 VM_MAX_NAMELEN + 1) 597 return (EINVAL); 598 599 vm = malloc(sizeof(struct vm), M_VM, M_WAITOK | M_ZERO); 600 error = vm_mem_init(&vm->mem, 0, VM_MAXUSER_ADDRESS_LA48); 601 if (error != 0) { 602 free(vm, M_VM); 603 return (error); 604 } 605 strcpy(vm->name, name); 606 mtx_init(&vm->rendezvous_mtx, "vm rendezvous lock", 0, MTX_DEF); 607 sx_init(&vm->vcpus_init_lock, "vm vcpus"); 608 vm->vcpu = malloc(sizeof(*vm->vcpu) * vm_maxcpu, M_VM, M_WAITOK | 609 M_ZERO); 610 611 vm->sockets = 1; 612 vm->cores = cores_per_package; /* XXX backwards compatibility */ 613 vm->threads = threads_per_core; /* XXX backwards compatibility */ 614 vm->maxcpus = vm_maxcpu; 615 616 vm_init(vm, true); 617 618 *retvm = vm; 619 return (0); 620 } 621 622 void 623 vm_get_topology(struct vm *vm, uint16_t *sockets, uint16_t *cores, 624 uint16_t *threads, uint16_t *maxcpus) 625 { 626 *sockets = vm->sockets; 627 *cores = vm->cores; 628 *threads = vm->threads; 629 *maxcpus = vm->maxcpus; 630 } 631 632 uint16_t 633 vm_get_maxcpus(struct vm *vm) 634 { 635 return (vm->maxcpus); 636 } 637 638 int 639 vm_set_topology(struct vm *vm, uint16_t sockets, uint16_t cores, 640 uint16_t threads, uint16_t maxcpus __unused) 641 { 642 /* Ignore maxcpus. */ 643 if ((sockets * cores * threads) > vm->maxcpus) 644 return (EINVAL); 645 vm->sockets = sockets; 646 vm->cores = cores; 647 vm->threads = threads; 648 return(0); 649 } 650 651 static void 652 vm_cleanup(struct vm *vm, bool destroy) 653 { 654 if (destroy) 655 vm_xlock_memsegs(vm); 656 else 657 vm_assert_memseg_xlocked(vm); 658 659 ppt_unassign_all(vm); 660 661 if (vm->iommu != NULL) 662 iommu_destroy_domain(vm->iommu); 663 664 if (destroy) 665 vrtc_cleanup(vm->vrtc); 666 else 667 vrtc_reset(vm->vrtc); 668 vpmtmr_cleanup(vm->vpmtmr); 669 vatpit_cleanup(vm->vatpit); 670 vhpet_cleanup(vm->vhpet); 671 vatpic_cleanup(vm->vatpic); 672 vioapic_cleanup(vm->vioapic); 673 674 for (int i = 0; i < vm->maxcpus; i++) { 675 if (vm->vcpu[i] != NULL) 676 vcpu_cleanup(vm->vcpu[i], destroy); 677 } 678 679 vmmops_cleanup(vm->cookie); 680 681 vm_mem_cleanup(vm); 682 683 if (destroy) { 684 vm_mem_destroy(vm); 685 686 free(vm->vcpu, M_VM); 687 sx_destroy(&vm->vcpus_init_lock); 688 mtx_destroy(&vm->rendezvous_mtx); 689 } 690 } 691 692 void 693 vm_destroy(struct vm *vm) 694 { 695 vm_cleanup(vm, true); 696 free(vm, M_VM); 697 } 698 699 int 700 vm_reinit(struct vm *vm) 701 { 702 int error; 703 704 /* 705 * A virtual machine can be reset only if all vcpus are suspended. 706 */ 707 if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) { 708 vm_cleanup(vm, false); 709 vm_init(vm, false); 710 error = 0; 711 } else { 712 error = EBUSY; 713 } 714 715 return (error); 716 } 717 718 const char * 719 vm_name(struct vm *vm) 720 { 721 return (vm->name); 722 } 723 724 int 725 vm_map_mmio(struct vm *vm, vm_paddr_t gpa, size_t len, vm_paddr_t hpa) 726 { 727 vm_object_t obj; 728 729 if ((obj = vmm_mmio_alloc(vm_vmspace(vm), gpa, len, hpa)) == NULL) 730 return (ENOMEM); 731 else 732 return (0); 733 } 734 735 int 736 vm_unmap_mmio(struct vm *vm, vm_paddr_t gpa, size_t len) 737 { 738 739 vmm_mmio_free(vm_vmspace(vm), gpa, len); 740 return (0); 741 } 742 743 static int 744 vm_iommu_map(struct vm *vm) 745 { 746 pmap_t pmap; 747 vm_paddr_t gpa, hpa; 748 struct vm_mem_map *mm; 749 int error, i; 750 751 sx_assert(&vm->mem.mem_segs_lock, SX_LOCKED); 752 753 pmap = vmspace_pmap(vm_vmspace(vm)); 754 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 755 if (!vm_memseg_sysmem(vm, i)) 756 continue; 757 758 mm = &vm->mem.mem_maps[i]; 759 KASSERT((mm->flags & VM_MEMMAP_F_IOMMU) == 0, 760 ("iommu map found invalid memmap %#lx/%#lx/%#x", 761 mm->gpa, mm->len, mm->flags)); 762 if ((mm->flags & VM_MEMMAP_F_WIRED) == 0) 763 continue; 764 mm->flags |= VM_MEMMAP_F_IOMMU; 765 766 for (gpa = mm->gpa; gpa < mm->gpa + mm->len; gpa += PAGE_SIZE) { 767 hpa = pmap_extract(pmap, gpa); 768 769 /* 770 * All mappings in the vmm vmspace must be 771 * present since they are managed by vmm in this way. 772 * Because we are in pass-through mode, the 773 * mappings must also be wired. This implies 774 * that all pages must be mapped and wired, 775 * allowing to use pmap_extract() and avoiding the 776 * need to use vm_gpa_hold_global(). 777 * 778 * This could change if/when we start 779 * supporting page faults on IOMMU maps. 780 */ 781 KASSERT(vm_page_wired(PHYS_TO_VM_PAGE(hpa)), 782 ("vm_iommu_map: vm %p gpa %jx hpa %jx not wired", 783 vm, (uintmax_t)gpa, (uintmax_t)hpa)); 784 785 iommu_create_mapping(vm->iommu, gpa, hpa, PAGE_SIZE); 786 } 787 } 788 789 error = iommu_invalidate_tlb(iommu_host_domain()); 790 return (error); 791 } 792 793 static int 794 vm_iommu_unmap(struct vm *vm) 795 { 796 vm_paddr_t gpa; 797 struct vm_mem_map *mm; 798 int error, i; 799 800 sx_assert(&vm->mem.mem_segs_lock, SX_LOCKED); 801 802 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 803 if (!vm_memseg_sysmem(vm, i)) 804 continue; 805 806 mm = &vm->mem.mem_maps[i]; 807 if ((mm->flags & VM_MEMMAP_F_IOMMU) == 0) 808 continue; 809 mm->flags &= ~VM_MEMMAP_F_IOMMU; 810 KASSERT((mm->flags & VM_MEMMAP_F_WIRED) != 0, 811 ("iommu unmap found invalid memmap %#lx/%#lx/%#x", 812 mm->gpa, mm->len, mm->flags)); 813 814 for (gpa = mm->gpa; gpa < mm->gpa + mm->len; gpa += PAGE_SIZE) { 815 KASSERT(vm_page_wired(PHYS_TO_VM_PAGE(pmap_extract( 816 vmspace_pmap(vm_vmspace(vm)), gpa))), 817 ("vm_iommu_unmap: vm %p gpa %jx not wired", 818 vm, (uintmax_t)gpa)); 819 iommu_remove_mapping(vm->iommu, gpa, PAGE_SIZE); 820 } 821 } 822 823 /* 824 * Invalidate the cached translations associated with the domain 825 * from which pages were removed. 826 */ 827 error = iommu_invalidate_tlb(vm->iommu); 828 return (error); 829 } 830 831 int 832 vm_unassign_pptdev(struct vm *vm, int bus, int slot, int func) 833 { 834 int error; 835 836 error = ppt_unassign_device(vm, bus, slot, func); 837 if (error) 838 return (error); 839 840 if (ppt_assigned_devices(vm) == 0) 841 error = vm_iommu_unmap(vm); 842 843 return (error); 844 } 845 846 int 847 vm_assign_pptdev(struct vm *vm, int bus, int slot, int func) 848 { 849 int error; 850 vm_paddr_t maxaddr; 851 bool map = false; 852 853 /* Set up the IOMMU to do the 'gpa' to 'hpa' translation */ 854 if (ppt_assigned_devices(vm) == 0) { 855 KASSERT(vm->iommu == NULL, 856 ("vm_assign_pptdev: iommu must be NULL")); 857 maxaddr = vmm_sysmem_maxaddr(vm); 858 vm->iommu = iommu_create_domain(maxaddr); 859 if (vm->iommu == NULL) 860 return (ENXIO); 861 map = true; 862 } 863 864 error = ppt_assign_device(vm, bus, slot, func); 865 if (error == 0 && map) 866 error = vm_iommu_map(vm); 867 return (error); 868 } 869 870 int 871 vm_get_register(struct vcpu *vcpu, int reg, uint64_t *retval) 872 { 873 874 if (reg >= VM_REG_LAST) 875 return (EINVAL); 876 877 return (vmmops_getreg(vcpu->cookie, reg, retval)); 878 } 879 880 int 881 vm_set_register(struct vcpu *vcpu, int reg, uint64_t val) 882 { 883 int error; 884 885 if (reg >= VM_REG_LAST) 886 return (EINVAL); 887 888 error = vmmops_setreg(vcpu->cookie, reg, val); 889 if (error || reg != VM_REG_GUEST_RIP) 890 return (error); 891 892 /* Set 'nextrip' to match the value of %rip */ 893 VMM_CTR1(vcpu, "Setting nextrip to %#lx", val); 894 vcpu->nextrip = val; 895 return (0); 896 } 897 898 static bool 899 is_descriptor_table(int reg) 900 { 901 902 switch (reg) { 903 case VM_REG_GUEST_IDTR: 904 case VM_REG_GUEST_GDTR: 905 return (true); 906 default: 907 return (false); 908 } 909 } 910 911 static bool 912 is_segment_register(int reg) 913 { 914 915 switch (reg) { 916 case VM_REG_GUEST_ES: 917 case VM_REG_GUEST_CS: 918 case VM_REG_GUEST_SS: 919 case VM_REG_GUEST_DS: 920 case VM_REG_GUEST_FS: 921 case VM_REG_GUEST_GS: 922 case VM_REG_GUEST_TR: 923 case VM_REG_GUEST_LDTR: 924 return (true); 925 default: 926 return (false); 927 } 928 } 929 930 int 931 vm_get_seg_desc(struct vcpu *vcpu, int reg, struct seg_desc *desc) 932 { 933 934 if (!is_segment_register(reg) && !is_descriptor_table(reg)) 935 return (EINVAL); 936 937 return (vmmops_getdesc(vcpu->cookie, reg, desc)); 938 } 939 940 int 941 vm_set_seg_desc(struct vcpu *vcpu, int reg, struct seg_desc *desc) 942 { 943 944 if (!is_segment_register(reg) && !is_descriptor_table(reg)) 945 return (EINVAL); 946 947 return (vmmops_setdesc(vcpu->cookie, reg, desc)); 948 } 949 950 static void 951 restore_guest_fpustate(struct vcpu *vcpu) 952 { 953 954 /* flush host state to the pcb */ 955 fpuexit(curthread); 956 957 /* restore guest FPU state */ 958 fpu_enable(); 959 fpurestore(vcpu->guestfpu); 960 961 /* restore guest XCR0 if XSAVE is enabled in the host */ 962 if (rcr4() & CR4_XSAVE) 963 load_xcr(0, vcpu->guest_xcr0); 964 965 /* 966 * The FPU is now "dirty" with the guest's state so disable 967 * the FPU to trap any access by the host. 968 */ 969 fpu_disable(); 970 } 971 972 static void 973 save_guest_fpustate(struct vcpu *vcpu) 974 { 975 976 if ((rcr0() & CR0_TS) == 0) 977 panic("fpu emulation not enabled in host!"); 978 979 /* save guest XCR0 and restore host XCR0 */ 980 if (rcr4() & CR4_XSAVE) { 981 vcpu->guest_xcr0 = rxcr(0); 982 load_xcr(0, vmm_get_host_xcr0()); 983 } 984 985 /* save guest FPU state */ 986 fpu_enable(); 987 fpusave(vcpu->guestfpu); 988 fpu_disable(); 989 } 990 991 static VMM_STAT(VCPU_IDLE_TICKS, "number of ticks vcpu was idle"); 992 993 static int 994 vcpu_set_state_locked(struct vcpu *vcpu, enum vcpu_state newstate, 995 bool from_idle) 996 { 997 int error; 998 999 vcpu_assert_locked(vcpu); 1000 1001 /* 1002 * State transitions from the vmmdev_ioctl() must always begin from 1003 * the VCPU_IDLE state. This guarantees that there is only a single 1004 * ioctl() operating on a vcpu at any point. 1005 */ 1006 if (from_idle) { 1007 while (vcpu->state != VCPU_IDLE) { 1008 vcpu->reqidle = 1; 1009 vcpu_notify_event_locked(vcpu, false); 1010 VMM_CTR1(vcpu, "vcpu state change from %s to " 1011 "idle requested", vcpu_state2str(vcpu->state)); 1012 msleep_spin(&vcpu->state, &vcpu->mtx, "vmstat", hz); 1013 } 1014 } else { 1015 KASSERT(vcpu->state != VCPU_IDLE, ("invalid transition from " 1016 "vcpu idle state")); 1017 } 1018 1019 if (vcpu->state == VCPU_RUNNING) { 1020 KASSERT(vcpu->hostcpu == curcpu, ("curcpu %d and hostcpu %d " 1021 "mismatch for running vcpu", curcpu, vcpu->hostcpu)); 1022 } else { 1023 KASSERT(vcpu->hostcpu == NOCPU, ("Invalid hostcpu %d for a " 1024 "vcpu that is not running", vcpu->hostcpu)); 1025 } 1026 1027 /* 1028 * The following state transitions are allowed: 1029 * IDLE -> FROZEN -> IDLE 1030 * FROZEN -> RUNNING -> FROZEN 1031 * FROZEN -> SLEEPING -> FROZEN 1032 */ 1033 switch (vcpu->state) { 1034 case VCPU_IDLE: 1035 case VCPU_RUNNING: 1036 case VCPU_SLEEPING: 1037 error = (newstate != VCPU_FROZEN); 1038 break; 1039 case VCPU_FROZEN: 1040 error = (newstate == VCPU_FROZEN); 1041 break; 1042 default: 1043 error = 1; 1044 break; 1045 } 1046 1047 if (error) 1048 return (EBUSY); 1049 1050 VMM_CTR2(vcpu, "vcpu state changed from %s to %s", 1051 vcpu_state2str(vcpu->state), vcpu_state2str(newstate)); 1052 1053 vcpu->state = newstate; 1054 if (newstate == VCPU_RUNNING) 1055 vcpu->hostcpu = curcpu; 1056 else 1057 vcpu->hostcpu = NOCPU; 1058 1059 if (newstate == VCPU_IDLE) 1060 wakeup(&vcpu->state); 1061 1062 return (0); 1063 } 1064 1065 static void 1066 vcpu_require_state(struct vcpu *vcpu, enum vcpu_state newstate) 1067 { 1068 int error; 1069 1070 if ((error = vcpu_set_state(vcpu, newstate, false)) != 0) 1071 panic("Error %d setting state to %d\n", error, newstate); 1072 } 1073 1074 static void 1075 vcpu_require_state_locked(struct vcpu *vcpu, enum vcpu_state newstate) 1076 { 1077 int error; 1078 1079 if ((error = vcpu_set_state_locked(vcpu, newstate, false)) != 0) 1080 panic("Error %d setting state to %d", error, newstate); 1081 } 1082 1083 static int 1084 vm_handle_rendezvous(struct vcpu *vcpu) 1085 { 1086 struct vm *vm = vcpu->vm; 1087 struct thread *td; 1088 int error, vcpuid; 1089 1090 error = 0; 1091 vcpuid = vcpu->vcpuid; 1092 td = curthread; 1093 mtx_lock(&vm->rendezvous_mtx); 1094 while (vm->rendezvous_func != NULL) { 1095 /* 'rendezvous_req_cpus' must be a subset of 'active_cpus' */ 1096 CPU_AND(&vm->rendezvous_req_cpus, &vm->rendezvous_req_cpus, &vm->active_cpus); 1097 1098 if (CPU_ISSET(vcpuid, &vm->rendezvous_req_cpus) && 1099 !CPU_ISSET(vcpuid, &vm->rendezvous_done_cpus)) { 1100 VMM_CTR0(vcpu, "Calling rendezvous func"); 1101 (*vm->rendezvous_func)(vcpu, vm->rendezvous_arg); 1102 CPU_SET(vcpuid, &vm->rendezvous_done_cpus); 1103 } 1104 if (CPU_CMP(&vm->rendezvous_req_cpus, 1105 &vm->rendezvous_done_cpus) == 0) { 1106 VMM_CTR0(vcpu, "Rendezvous completed"); 1107 CPU_ZERO(&vm->rendezvous_req_cpus); 1108 vm->rendezvous_func = NULL; 1109 wakeup(&vm->rendezvous_func); 1110 break; 1111 } 1112 VMM_CTR0(vcpu, "Wait for rendezvous completion"); 1113 mtx_sleep(&vm->rendezvous_func, &vm->rendezvous_mtx, 0, 1114 "vmrndv", hz); 1115 if (td_ast_pending(td, TDA_SUSPEND)) { 1116 mtx_unlock(&vm->rendezvous_mtx); 1117 error = thread_check_susp(td, true); 1118 if (error != 0) 1119 return (error); 1120 mtx_lock(&vm->rendezvous_mtx); 1121 } 1122 } 1123 mtx_unlock(&vm->rendezvous_mtx); 1124 return (0); 1125 } 1126 1127 /* 1128 * Emulate a guest 'hlt' by sleeping until the vcpu is ready to run. 1129 */ 1130 static int 1131 vm_handle_hlt(struct vcpu *vcpu, bool intr_disabled, bool *retu) 1132 { 1133 struct vm *vm = vcpu->vm; 1134 const char *wmesg; 1135 struct thread *td; 1136 int error, t, vcpuid, vcpu_halted, vm_halted; 1137 1138 vcpuid = vcpu->vcpuid; 1139 vcpu_halted = 0; 1140 vm_halted = 0; 1141 error = 0; 1142 td = curthread; 1143 1144 KASSERT(!CPU_ISSET(vcpuid, &vm->halted_cpus), ("vcpu already halted")); 1145 1146 vcpu_lock(vcpu); 1147 while (1) { 1148 /* 1149 * Do a final check for pending NMI or interrupts before 1150 * really putting this thread to sleep. Also check for 1151 * software events that would cause this vcpu to wakeup. 1152 * 1153 * These interrupts/events could have happened after the 1154 * vcpu returned from vmmops_run() and before it acquired the 1155 * vcpu lock above. 1156 */ 1157 if (vm->rendezvous_func != NULL || vm->suspend || vcpu->reqidle) 1158 break; 1159 if (vm_nmi_pending(vcpu)) 1160 break; 1161 if (!intr_disabled) { 1162 if (vm_extint_pending(vcpu) || 1163 vlapic_pending_intr(vcpu->vlapic, NULL)) { 1164 break; 1165 } 1166 } 1167 1168 /* Don't go to sleep if the vcpu thread needs to yield */ 1169 if (vcpu_should_yield(vcpu)) 1170 break; 1171 1172 if (vcpu_debugged(vcpu)) 1173 break; 1174 1175 /* 1176 * Some Linux guests implement "halt" by having all vcpus 1177 * execute HLT with interrupts disabled. 'halted_cpus' keeps 1178 * track of the vcpus that have entered this state. When all 1179 * vcpus enter the halted state the virtual machine is halted. 1180 */ 1181 if (intr_disabled) { 1182 wmesg = "vmhalt"; 1183 VMM_CTR0(vcpu, "Halted"); 1184 if (!vcpu_halted && halt_detection_enabled) { 1185 vcpu_halted = 1; 1186 CPU_SET_ATOMIC(vcpuid, &vm->halted_cpus); 1187 } 1188 if (CPU_CMP(&vm->halted_cpus, &vm->active_cpus) == 0) { 1189 vm_halted = 1; 1190 break; 1191 } 1192 } else { 1193 wmesg = "vmidle"; 1194 } 1195 1196 t = ticks; 1197 vcpu_require_state_locked(vcpu, VCPU_SLEEPING); 1198 /* 1199 * XXX msleep_spin() cannot be interrupted by signals so 1200 * wake up periodically to check pending signals. 1201 */ 1202 msleep_spin(vcpu, &vcpu->mtx, wmesg, hz); 1203 vcpu_require_state_locked(vcpu, VCPU_FROZEN); 1204 vmm_stat_incr(vcpu, VCPU_IDLE_TICKS, ticks - t); 1205 if (td_ast_pending(td, TDA_SUSPEND)) { 1206 vcpu_unlock(vcpu); 1207 error = thread_check_susp(td, false); 1208 if (error != 0) { 1209 if (vcpu_halted) { 1210 CPU_CLR_ATOMIC(vcpuid, 1211 &vm->halted_cpus); 1212 } 1213 return (error); 1214 } 1215 vcpu_lock(vcpu); 1216 } 1217 } 1218 1219 if (vcpu_halted) 1220 CPU_CLR_ATOMIC(vcpuid, &vm->halted_cpus); 1221 1222 vcpu_unlock(vcpu); 1223 1224 if (vm_halted) 1225 vm_suspend(vm, VM_SUSPEND_HALT); 1226 1227 return (0); 1228 } 1229 1230 static int 1231 vm_handle_paging(struct vcpu *vcpu, bool *retu) 1232 { 1233 struct vm *vm = vcpu->vm; 1234 int rv, ftype; 1235 struct vm_map *map; 1236 struct vm_exit *vme; 1237 1238 vme = &vcpu->exitinfo; 1239 1240 KASSERT(vme->inst_length == 0, ("%s: invalid inst_length %d", 1241 __func__, vme->inst_length)); 1242 1243 ftype = vme->u.paging.fault_type; 1244 KASSERT(ftype == VM_PROT_READ || 1245 ftype == VM_PROT_WRITE || ftype == VM_PROT_EXECUTE, 1246 ("vm_handle_paging: invalid fault_type %d", ftype)); 1247 1248 if (ftype == VM_PROT_READ || ftype == VM_PROT_WRITE) { 1249 rv = pmap_emulate_accessed_dirty(vmspace_pmap(vm_vmspace(vm)), 1250 vme->u.paging.gpa, ftype); 1251 if (rv == 0) { 1252 VMM_CTR2(vcpu, "%s bit emulation for gpa %#lx", 1253 ftype == VM_PROT_READ ? "accessed" : "dirty", 1254 vme->u.paging.gpa); 1255 goto done; 1256 } 1257 } 1258 1259 map = &vm_vmspace(vm)->vm_map; 1260 rv = vm_fault(map, vme->u.paging.gpa, ftype, VM_FAULT_NORMAL, NULL); 1261 1262 VMM_CTR3(vcpu, "vm_handle_paging rv = %d, gpa = %#lx, " 1263 "ftype = %d", rv, vme->u.paging.gpa, ftype); 1264 1265 if (rv != KERN_SUCCESS) 1266 return (EFAULT); 1267 done: 1268 return (0); 1269 } 1270 1271 static int 1272 vm_handle_inst_emul(struct vcpu *vcpu, bool *retu) 1273 { 1274 struct vie *vie; 1275 struct vm_exit *vme; 1276 uint64_t gla, gpa, cs_base; 1277 struct vm_guest_paging *paging; 1278 mem_region_read_t mread; 1279 mem_region_write_t mwrite; 1280 enum vm_cpu_mode cpu_mode; 1281 int cs_d, error, fault; 1282 1283 vme = &vcpu->exitinfo; 1284 1285 KASSERT(vme->inst_length == 0, ("%s: invalid inst_length %d", 1286 __func__, vme->inst_length)); 1287 1288 gla = vme->u.inst_emul.gla; 1289 gpa = vme->u.inst_emul.gpa; 1290 cs_base = vme->u.inst_emul.cs_base; 1291 cs_d = vme->u.inst_emul.cs_d; 1292 vie = &vme->u.inst_emul.vie; 1293 paging = &vme->u.inst_emul.paging; 1294 cpu_mode = paging->cpu_mode; 1295 1296 VMM_CTR1(vcpu, "inst_emul fault accessing gpa %#lx", gpa); 1297 1298 /* Fetch, decode and emulate the faulting instruction */ 1299 if (vie->num_valid == 0) { 1300 error = vmm_fetch_instruction(vcpu, paging, vme->rip + cs_base, 1301 VIE_INST_SIZE, vie, &fault); 1302 } else { 1303 /* 1304 * The instruction bytes have already been copied into 'vie' 1305 */ 1306 error = fault = 0; 1307 } 1308 if (error || fault) 1309 return (error); 1310 1311 if (vmm_decode_instruction(vcpu, gla, cpu_mode, cs_d, vie) != 0) { 1312 VMM_CTR1(vcpu, "Error decoding instruction at %#lx", 1313 vme->rip + cs_base); 1314 *retu = true; /* dump instruction bytes in userspace */ 1315 return (0); 1316 } 1317 1318 /* 1319 * Update 'nextrip' based on the length of the emulated instruction. 1320 */ 1321 vme->inst_length = vie->num_processed; 1322 vcpu->nextrip += vie->num_processed; 1323 VMM_CTR1(vcpu, "nextrip updated to %#lx after instruction decoding", 1324 vcpu->nextrip); 1325 1326 /* return to userland unless this is an in-kernel emulated device */ 1327 if (gpa >= DEFAULT_APIC_BASE && gpa < DEFAULT_APIC_BASE + PAGE_SIZE) { 1328 mread = lapic_mmio_read; 1329 mwrite = lapic_mmio_write; 1330 } else if (gpa >= VIOAPIC_BASE && gpa < VIOAPIC_BASE + VIOAPIC_SIZE) { 1331 mread = vioapic_mmio_read; 1332 mwrite = vioapic_mmio_write; 1333 } else if (gpa >= VHPET_BASE && gpa < VHPET_BASE + VHPET_SIZE) { 1334 mread = vhpet_mmio_read; 1335 mwrite = vhpet_mmio_write; 1336 } else { 1337 *retu = true; 1338 return (0); 1339 } 1340 1341 error = vmm_emulate_instruction(vcpu, gpa, vie, paging, mread, mwrite, 1342 retu); 1343 1344 return (error); 1345 } 1346 1347 static int 1348 vm_handle_suspend(struct vcpu *vcpu, bool *retu) 1349 { 1350 struct vm *vm = vcpu->vm; 1351 int error, i; 1352 struct thread *td; 1353 1354 error = 0; 1355 td = curthread; 1356 1357 CPU_SET_ATOMIC(vcpu->vcpuid, &vm->suspended_cpus); 1358 1359 /* 1360 * Wait until all 'active_cpus' have suspended themselves. 1361 * 1362 * Since a VM may be suspended at any time including when one or 1363 * more vcpus are doing a rendezvous we need to call the rendezvous 1364 * handler while we are waiting to prevent a deadlock. 1365 */ 1366 vcpu_lock(vcpu); 1367 while (error == 0) { 1368 if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) { 1369 VMM_CTR0(vcpu, "All vcpus suspended"); 1370 break; 1371 } 1372 1373 if (vm->rendezvous_func == NULL) { 1374 VMM_CTR0(vcpu, "Sleeping during suspend"); 1375 vcpu_require_state_locked(vcpu, VCPU_SLEEPING); 1376 msleep_spin(vcpu, &vcpu->mtx, "vmsusp", hz); 1377 vcpu_require_state_locked(vcpu, VCPU_FROZEN); 1378 if (td_ast_pending(td, TDA_SUSPEND)) { 1379 vcpu_unlock(vcpu); 1380 error = thread_check_susp(td, false); 1381 vcpu_lock(vcpu); 1382 } 1383 } else { 1384 VMM_CTR0(vcpu, "Rendezvous during suspend"); 1385 vcpu_unlock(vcpu); 1386 error = vm_handle_rendezvous(vcpu); 1387 vcpu_lock(vcpu); 1388 } 1389 } 1390 vcpu_unlock(vcpu); 1391 1392 /* 1393 * Wakeup the other sleeping vcpus and return to userspace. 1394 */ 1395 for (i = 0; i < vm->maxcpus; i++) { 1396 if (CPU_ISSET(i, &vm->suspended_cpus)) { 1397 vcpu_notify_event(vm_vcpu(vm, i), false); 1398 } 1399 } 1400 1401 *retu = true; 1402 return (error); 1403 } 1404 1405 static int 1406 vm_handle_reqidle(struct vcpu *vcpu, bool *retu) 1407 { 1408 vcpu_lock(vcpu); 1409 KASSERT(vcpu->reqidle, ("invalid vcpu reqidle %d", vcpu->reqidle)); 1410 vcpu->reqidle = 0; 1411 vcpu_unlock(vcpu); 1412 *retu = true; 1413 return (0); 1414 } 1415 1416 static int 1417 vm_handle_db(struct vcpu *vcpu, struct vm_exit *vme, bool *retu) 1418 { 1419 int error, fault; 1420 uint64_t rsp; 1421 uint64_t rflags; 1422 struct vm_copyinfo copyinfo[2]; 1423 1424 *retu = true; 1425 if (!vme->u.dbg.pushf_intercept || vme->u.dbg.tf_shadow_val != 0) { 1426 return (0); 1427 } 1428 1429 vm_get_register(vcpu, VM_REG_GUEST_RSP, &rsp); 1430 error = vm_copy_setup(vcpu, &vme->u.dbg.paging, rsp, sizeof(uint64_t), 1431 VM_PROT_RW, copyinfo, nitems(copyinfo), &fault); 1432 if (error != 0 || fault != 0) { 1433 *retu = false; 1434 return (EINVAL); 1435 } 1436 1437 /* Read pushed rflags value from top of stack. */ 1438 vm_copyin(copyinfo, &rflags, sizeof(uint64_t)); 1439 1440 /* Clear TF bit. */ 1441 rflags &= ~(PSL_T); 1442 1443 /* Write updated value back to memory. */ 1444 vm_copyout(&rflags, copyinfo, sizeof(uint64_t)); 1445 vm_copy_teardown(copyinfo, nitems(copyinfo)); 1446 1447 return (0); 1448 } 1449 1450 int 1451 vm_suspend(struct vm *vm, enum vm_suspend_how how) 1452 { 1453 int i; 1454 1455 if (how <= VM_SUSPEND_NONE || how >= VM_SUSPEND_LAST) 1456 return (EINVAL); 1457 1458 if (atomic_cmpset_int(&vm->suspend, 0, how) == 0) { 1459 VM_CTR2(vm, "virtual machine already suspended %d/%d", 1460 vm->suspend, how); 1461 return (EALREADY); 1462 } 1463 1464 VM_CTR1(vm, "virtual machine successfully suspended %d", how); 1465 1466 /* 1467 * Notify all active vcpus that they are now suspended. 1468 */ 1469 for (i = 0; i < vm->maxcpus; i++) { 1470 if (CPU_ISSET(i, &vm->active_cpus)) 1471 vcpu_notify_event(vm_vcpu(vm, i), false); 1472 } 1473 1474 return (0); 1475 } 1476 1477 void 1478 vm_exit_suspended(struct vcpu *vcpu, uint64_t rip) 1479 { 1480 struct vm *vm = vcpu->vm; 1481 struct vm_exit *vmexit; 1482 1483 KASSERT(vm->suspend > VM_SUSPEND_NONE && vm->suspend < VM_SUSPEND_LAST, 1484 ("vm_exit_suspended: invalid suspend type %d", vm->suspend)); 1485 1486 vmexit = vm_exitinfo(vcpu); 1487 vmexit->rip = rip; 1488 vmexit->inst_length = 0; 1489 vmexit->exitcode = VM_EXITCODE_SUSPENDED; 1490 vmexit->u.suspended.how = vm->suspend; 1491 } 1492 1493 void 1494 vm_exit_debug(struct vcpu *vcpu, uint64_t rip) 1495 { 1496 struct vm_exit *vmexit; 1497 1498 vmexit = vm_exitinfo(vcpu); 1499 vmexit->rip = rip; 1500 vmexit->inst_length = 0; 1501 vmexit->exitcode = VM_EXITCODE_DEBUG; 1502 } 1503 1504 void 1505 vm_exit_rendezvous(struct vcpu *vcpu, uint64_t rip) 1506 { 1507 struct vm_exit *vmexit; 1508 1509 vmexit = vm_exitinfo(vcpu); 1510 vmexit->rip = rip; 1511 vmexit->inst_length = 0; 1512 vmexit->exitcode = VM_EXITCODE_RENDEZVOUS; 1513 vmm_stat_incr(vcpu, VMEXIT_RENDEZVOUS, 1); 1514 } 1515 1516 void 1517 vm_exit_reqidle(struct vcpu *vcpu, uint64_t rip) 1518 { 1519 struct vm_exit *vmexit; 1520 1521 vmexit = vm_exitinfo(vcpu); 1522 vmexit->rip = rip; 1523 vmexit->inst_length = 0; 1524 vmexit->exitcode = VM_EXITCODE_REQIDLE; 1525 vmm_stat_incr(vcpu, VMEXIT_REQIDLE, 1); 1526 } 1527 1528 void 1529 vm_exit_astpending(struct vcpu *vcpu, uint64_t rip) 1530 { 1531 struct vm_exit *vmexit; 1532 1533 vmexit = vm_exitinfo(vcpu); 1534 vmexit->rip = rip; 1535 vmexit->inst_length = 0; 1536 vmexit->exitcode = VM_EXITCODE_BOGUS; 1537 vmm_stat_incr(vcpu, VMEXIT_ASTPENDING, 1); 1538 } 1539 1540 int 1541 vm_run(struct vcpu *vcpu) 1542 { 1543 struct vm *vm = vcpu->vm; 1544 struct vm_eventinfo evinfo; 1545 int error, vcpuid; 1546 struct pcb *pcb; 1547 uint64_t tscval; 1548 struct vm_exit *vme; 1549 bool retu, intr_disabled; 1550 pmap_t pmap; 1551 1552 vcpuid = vcpu->vcpuid; 1553 1554 if (!CPU_ISSET(vcpuid, &vm->active_cpus)) 1555 return (EINVAL); 1556 1557 if (CPU_ISSET(vcpuid, &vm->suspended_cpus)) 1558 return (EINVAL); 1559 1560 pmap = vmspace_pmap(vm_vmspace(vm)); 1561 vme = &vcpu->exitinfo; 1562 evinfo.rptr = &vm->rendezvous_req_cpus; 1563 evinfo.sptr = &vm->suspend; 1564 evinfo.iptr = &vcpu->reqidle; 1565 restart: 1566 critical_enter(); 1567 1568 KASSERT(!CPU_ISSET(curcpu, &pmap->pm_active), 1569 ("vm_run: absurd pm_active")); 1570 1571 tscval = rdtsc(); 1572 1573 pcb = PCPU_GET(curpcb); 1574 set_pcb_flags(pcb, PCB_FULL_IRET); 1575 1576 restore_guest_fpustate(vcpu); 1577 1578 vcpu_require_state(vcpu, VCPU_RUNNING); 1579 error = vmmops_run(vcpu->cookie, vcpu->nextrip, pmap, &evinfo); 1580 vcpu_require_state(vcpu, VCPU_FROZEN); 1581 1582 save_guest_fpustate(vcpu); 1583 1584 vmm_stat_incr(vcpu, VCPU_TOTAL_RUNTIME, rdtsc() - tscval); 1585 1586 critical_exit(); 1587 1588 if (error == 0) { 1589 retu = false; 1590 vcpu->nextrip = vme->rip + vme->inst_length; 1591 switch (vme->exitcode) { 1592 case VM_EXITCODE_REQIDLE: 1593 error = vm_handle_reqidle(vcpu, &retu); 1594 break; 1595 case VM_EXITCODE_SUSPENDED: 1596 error = vm_handle_suspend(vcpu, &retu); 1597 break; 1598 case VM_EXITCODE_IOAPIC_EOI: 1599 vioapic_process_eoi(vm, vme->u.ioapic_eoi.vector); 1600 break; 1601 case VM_EXITCODE_RENDEZVOUS: 1602 error = vm_handle_rendezvous(vcpu); 1603 break; 1604 case VM_EXITCODE_HLT: 1605 intr_disabled = ((vme->u.hlt.rflags & PSL_I) == 0); 1606 error = vm_handle_hlt(vcpu, intr_disabled, &retu); 1607 break; 1608 case VM_EXITCODE_PAGING: 1609 error = vm_handle_paging(vcpu, &retu); 1610 break; 1611 case VM_EXITCODE_INST_EMUL: 1612 error = vm_handle_inst_emul(vcpu, &retu); 1613 break; 1614 case VM_EXITCODE_INOUT: 1615 case VM_EXITCODE_INOUT_STR: 1616 error = vm_handle_inout(vcpu, vme, &retu); 1617 break; 1618 case VM_EXITCODE_DB: 1619 error = vm_handle_db(vcpu, vme, &retu); 1620 break; 1621 case VM_EXITCODE_MONITOR: 1622 case VM_EXITCODE_MWAIT: 1623 case VM_EXITCODE_VMINSN: 1624 vm_inject_ud(vcpu); 1625 break; 1626 default: 1627 retu = true; /* handled in userland */ 1628 break; 1629 } 1630 } 1631 1632 /* 1633 * VM_EXITCODE_INST_EMUL could access the apic which could transform the 1634 * exit code into VM_EXITCODE_IPI. 1635 */ 1636 if (error == 0 && vme->exitcode == VM_EXITCODE_IPI) 1637 error = vm_handle_ipi(vcpu, vme, &retu); 1638 1639 if (error == 0 && retu == false) 1640 goto restart; 1641 1642 vmm_stat_incr(vcpu, VMEXIT_USERSPACE, 1); 1643 VMM_CTR2(vcpu, "retu %d/%d", error, vme->exitcode); 1644 1645 return (error); 1646 } 1647 1648 int 1649 vm_restart_instruction(struct vcpu *vcpu) 1650 { 1651 enum vcpu_state state; 1652 uint64_t rip; 1653 int error __diagused; 1654 1655 state = vcpu_get_state(vcpu, NULL); 1656 if (state == VCPU_RUNNING) { 1657 /* 1658 * When a vcpu is "running" the next instruction is determined 1659 * by adding 'rip' and 'inst_length' in the vcpu's 'exitinfo'. 1660 * Thus setting 'inst_length' to zero will cause the current 1661 * instruction to be restarted. 1662 */ 1663 vcpu->exitinfo.inst_length = 0; 1664 VMM_CTR1(vcpu, "restarting instruction at %#lx by " 1665 "setting inst_length to zero", vcpu->exitinfo.rip); 1666 } else if (state == VCPU_FROZEN) { 1667 /* 1668 * When a vcpu is "frozen" it is outside the critical section 1669 * around vmmops_run() and 'nextrip' points to the next 1670 * instruction. Thus instruction restart is achieved by setting 1671 * 'nextrip' to the vcpu's %rip. 1672 */ 1673 error = vm_get_register(vcpu, VM_REG_GUEST_RIP, &rip); 1674 KASSERT(!error, ("%s: error %d getting rip", __func__, error)); 1675 VMM_CTR2(vcpu, "restarting instruction by updating " 1676 "nextrip from %#lx to %#lx", vcpu->nextrip, rip); 1677 vcpu->nextrip = rip; 1678 } else { 1679 panic("%s: invalid state %d", __func__, state); 1680 } 1681 return (0); 1682 } 1683 1684 int 1685 vm_exit_intinfo(struct vcpu *vcpu, uint64_t info) 1686 { 1687 int type, vector; 1688 1689 if (info & VM_INTINFO_VALID) { 1690 type = info & VM_INTINFO_TYPE; 1691 vector = info & 0xff; 1692 if (type == VM_INTINFO_NMI && vector != IDT_NMI) 1693 return (EINVAL); 1694 if (type == VM_INTINFO_HWEXCEPTION && vector >= 32) 1695 return (EINVAL); 1696 if (info & VM_INTINFO_RSVD) 1697 return (EINVAL); 1698 } else { 1699 info = 0; 1700 } 1701 VMM_CTR2(vcpu, "%s: info1(%#lx)", __func__, info); 1702 vcpu->exitintinfo = info; 1703 return (0); 1704 } 1705 1706 enum exc_class { 1707 EXC_BENIGN, 1708 EXC_CONTRIBUTORY, 1709 EXC_PAGEFAULT 1710 }; 1711 1712 #define IDT_VE 20 /* Virtualization Exception (Intel specific) */ 1713 1714 static enum exc_class 1715 exception_class(uint64_t info) 1716 { 1717 int type, vector; 1718 1719 KASSERT(info & VM_INTINFO_VALID, ("intinfo must be valid: %#lx", info)); 1720 type = info & VM_INTINFO_TYPE; 1721 vector = info & 0xff; 1722 1723 /* Table 6-4, "Interrupt and Exception Classes", Intel SDM, Vol 3 */ 1724 switch (type) { 1725 case VM_INTINFO_HWINTR: 1726 case VM_INTINFO_SWINTR: 1727 case VM_INTINFO_NMI: 1728 return (EXC_BENIGN); 1729 default: 1730 /* 1731 * Hardware exception. 1732 * 1733 * SVM and VT-x use identical type values to represent NMI, 1734 * hardware interrupt and software interrupt. 1735 * 1736 * SVM uses type '3' for all exceptions. VT-x uses type '3' 1737 * for exceptions except #BP and #OF. #BP and #OF use a type 1738 * value of '5' or '6'. Therefore we don't check for explicit 1739 * values of 'type' to classify 'intinfo' into a hardware 1740 * exception. 1741 */ 1742 break; 1743 } 1744 1745 switch (vector) { 1746 case IDT_PF: 1747 case IDT_VE: 1748 return (EXC_PAGEFAULT); 1749 case IDT_DE: 1750 case IDT_TS: 1751 case IDT_NP: 1752 case IDT_SS: 1753 case IDT_GP: 1754 return (EXC_CONTRIBUTORY); 1755 default: 1756 return (EXC_BENIGN); 1757 } 1758 } 1759 1760 static int 1761 nested_fault(struct vcpu *vcpu, uint64_t info1, uint64_t info2, 1762 uint64_t *retinfo) 1763 { 1764 enum exc_class exc1, exc2; 1765 int type1, vector1; 1766 1767 KASSERT(info1 & VM_INTINFO_VALID, ("info1 %#lx is not valid", info1)); 1768 KASSERT(info2 & VM_INTINFO_VALID, ("info2 %#lx is not valid", info2)); 1769 1770 /* 1771 * If an exception occurs while attempting to call the double-fault 1772 * handler the processor enters shutdown mode (aka triple fault). 1773 */ 1774 type1 = info1 & VM_INTINFO_TYPE; 1775 vector1 = info1 & 0xff; 1776 if (type1 == VM_INTINFO_HWEXCEPTION && vector1 == IDT_DF) { 1777 VMM_CTR2(vcpu, "triple fault: info1(%#lx), info2(%#lx)", 1778 info1, info2); 1779 vm_suspend(vcpu->vm, VM_SUSPEND_TRIPLEFAULT); 1780 *retinfo = 0; 1781 return (0); 1782 } 1783 1784 /* 1785 * Table 6-5 "Conditions for Generating a Double Fault", Intel SDM, Vol3 1786 */ 1787 exc1 = exception_class(info1); 1788 exc2 = exception_class(info2); 1789 if ((exc1 == EXC_CONTRIBUTORY && exc2 == EXC_CONTRIBUTORY) || 1790 (exc1 == EXC_PAGEFAULT && exc2 != EXC_BENIGN)) { 1791 /* Convert nested fault into a double fault. */ 1792 *retinfo = IDT_DF; 1793 *retinfo |= VM_INTINFO_VALID | VM_INTINFO_HWEXCEPTION; 1794 *retinfo |= VM_INTINFO_DEL_ERRCODE; 1795 } else { 1796 /* Handle exceptions serially */ 1797 *retinfo = info2; 1798 } 1799 return (1); 1800 } 1801 1802 static uint64_t 1803 vcpu_exception_intinfo(struct vcpu *vcpu) 1804 { 1805 uint64_t info = 0; 1806 1807 if (vcpu->exception_pending) { 1808 info = vcpu->exc_vector & 0xff; 1809 info |= VM_INTINFO_VALID | VM_INTINFO_HWEXCEPTION; 1810 if (vcpu->exc_errcode_valid) { 1811 info |= VM_INTINFO_DEL_ERRCODE; 1812 info |= (uint64_t)vcpu->exc_errcode << 32; 1813 } 1814 } 1815 return (info); 1816 } 1817 1818 int 1819 vm_entry_intinfo(struct vcpu *vcpu, uint64_t *retinfo) 1820 { 1821 uint64_t info1, info2; 1822 int valid; 1823 1824 info1 = vcpu->exitintinfo; 1825 vcpu->exitintinfo = 0; 1826 1827 info2 = 0; 1828 if (vcpu->exception_pending) { 1829 info2 = vcpu_exception_intinfo(vcpu); 1830 vcpu->exception_pending = 0; 1831 VMM_CTR2(vcpu, "Exception %d delivered: %#lx", 1832 vcpu->exc_vector, info2); 1833 } 1834 1835 if ((info1 & VM_INTINFO_VALID) && (info2 & VM_INTINFO_VALID)) { 1836 valid = nested_fault(vcpu, info1, info2, retinfo); 1837 } else if (info1 & VM_INTINFO_VALID) { 1838 *retinfo = info1; 1839 valid = 1; 1840 } else if (info2 & VM_INTINFO_VALID) { 1841 *retinfo = info2; 1842 valid = 1; 1843 } else { 1844 valid = 0; 1845 } 1846 1847 if (valid) { 1848 VMM_CTR4(vcpu, "%s: info1(%#lx), info2(%#lx), " 1849 "retinfo(%#lx)", __func__, info1, info2, *retinfo); 1850 } 1851 1852 return (valid); 1853 } 1854 1855 int 1856 vm_get_intinfo(struct vcpu *vcpu, uint64_t *info1, uint64_t *info2) 1857 { 1858 *info1 = vcpu->exitintinfo; 1859 *info2 = vcpu_exception_intinfo(vcpu); 1860 return (0); 1861 } 1862 1863 int 1864 vm_inject_exception(struct vcpu *vcpu, int vector, int errcode_valid, 1865 uint32_t errcode, int restart_instruction) 1866 { 1867 uint64_t regval; 1868 int error __diagused; 1869 1870 if (vector < 0 || vector >= 32) 1871 return (EINVAL); 1872 1873 /* 1874 * A double fault exception should never be injected directly into 1875 * the guest. It is a derived exception that results from specific 1876 * combinations of nested faults. 1877 */ 1878 if (vector == IDT_DF) 1879 return (EINVAL); 1880 1881 if (vcpu->exception_pending) { 1882 VMM_CTR2(vcpu, "Unable to inject exception %d due to " 1883 "pending exception %d", vector, vcpu->exc_vector); 1884 return (EBUSY); 1885 } 1886 1887 if (errcode_valid) { 1888 /* 1889 * Exceptions don't deliver an error code in real mode. 1890 */ 1891 error = vm_get_register(vcpu, VM_REG_GUEST_CR0, ®val); 1892 KASSERT(!error, ("%s: error %d getting CR0", __func__, error)); 1893 if (!(regval & CR0_PE)) 1894 errcode_valid = 0; 1895 } 1896 1897 /* 1898 * From section 26.6.1 "Interruptibility State" in Intel SDM: 1899 * 1900 * Event blocking by "STI" or "MOV SS" is cleared after guest executes 1901 * one instruction or incurs an exception. 1902 */ 1903 error = vm_set_register(vcpu, VM_REG_GUEST_INTR_SHADOW, 0); 1904 KASSERT(error == 0, ("%s: error %d clearing interrupt shadow", 1905 __func__, error)); 1906 1907 if (restart_instruction) 1908 vm_restart_instruction(vcpu); 1909 1910 vcpu->exception_pending = 1; 1911 vcpu->exc_vector = vector; 1912 vcpu->exc_errcode = errcode; 1913 vcpu->exc_errcode_valid = errcode_valid; 1914 VMM_CTR1(vcpu, "Exception %d pending", vector); 1915 return (0); 1916 } 1917 1918 void 1919 vm_inject_fault(struct vcpu *vcpu, int vector, int errcode_valid, int errcode) 1920 { 1921 int error __diagused, restart_instruction; 1922 1923 restart_instruction = 1; 1924 1925 error = vm_inject_exception(vcpu, vector, errcode_valid, 1926 errcode, restart_instruction); 1927 KASSERT(error == 0, ("vm_inject_exception error %d", error)); 1928 } 1929 1930 void 1931 vm_inject_pf(struct vcpu *vcpu, int error_code, uint64_t cr2) 1932 { 1933 int error __diagused; 1934 1935 VMM_CTR2(vcpu, "Injecting page fault: error_code %#x, cr2 %#lx", 1936 error_code, cr2); 1937 1938 error = vm_set_register(vcpu, VM_REG_GUEST_CR2, cr2); 1939 KASSERT(error == 0, ("vm_set_register(cr2) error %d", error)); 1940 1941 vm_inject_fault(vcpu, IDT_PF, 1, error_code); 1942 } 1943 1944 static VMM_STAT(VCPU_NMI_COUNT, "number of NMIs delivered to vcpu"); 1945 1946 int 1947 vm_inject_nmi(struct vcpu *vcpu) 1948 { 1949 1950 vcpu->nmi_pending = 1; 1951 vcpu_notify_event(vcpu, false); 1952 return (0); 1953 } 1954 1955 int 1956 vm_nmi_pending(struct vcpu *vcpu) 1957 { 1958 return (vcpu->nmi_pending); 1959 } 1960 1961 void 1962 vm_nmi_clear(struct vcpu *vcpu) 1963 { 1964 if (vcpu->nmi_pending == 0) 1965 panic("vm_nmi_clear: inconsistent nmi_pending state"); 1966 1967 vcpu->nmi_pending = 0; 1968 vmm_stat_incr(vcpu, VCPU_NMI_COUNT, 1); 1969 } 1970 1971 static VMM_STAT(VCPU_EXTINT_COUNT, "number of ExtINTs delivered to vcpu"); 1972 1973 int 1974 vm_inject_extint(struct vcpu *vcpu) 1975 { 1976 1977 vcpu->extint_pending = 1; 1978 vcpu_notify_event(vcpu, false); 1979 return (0); 1980 } 1981 1982 int 1983 vm_extint_pending(struct vcpu *vcpu) 1984 { 1985 return (vcpu->extint_pending); 1986 } 1987 1988 void 1989 vm_extint_clear(struct vcpu *vcpu) 1990 { 1991 if (vcpu->extint_pending == 0) 1992 panic("vm_extint_clear: inconsistent extint_pending state"); 1993 1994 vcpu->extint_pending = 0; 1995 vmm_stat_incr(vcpu, VCPU_EXTINT_COUNT, 1); 1996 } 1997 1998 int 1999 vm_get_capability(struct vcpu *vcpu, int type, int *retval) 2000 { 2001 if (type < 0 || type >= VM_CAP_MAX) 2002 return (EINVAL); 2003 2004 return (vmmops_getcap(vcpu->cookie, type, retval)); 2005 } 2006 2007 int 2008 vm_set_capability(struct vcpu *vcpu, int type, int val) 2009 { 2010 if (type < 0 || type >= VM_CAP_MAX) 2011 return (EINVAL); 2012 2013 return (vmmops_setcap(vcpu->cookie, type, val)); 2014 } 2015 2016 struct vm * 2017 vcpu_vm(struct vcpu *vcpu) 2018 { 2019 return (vcpu->vm); 2020 } 2021 2022 int 2023 vcpu_vcpuid(struct vcpu *vcpu) 2024 { 2025 return (vcpu->vcpuid); 2026 } 2027 2028 struct vcpu * 2029 vm_vcpu(struct vm *vm, int vcpuid) 2030 { 2031 return (vm->vcpu[vcpuid]); 2032 } 2033 2034 struct vlapic * 2035 vm_lapic(struct vcpu *vcpu) 2036 { 2037 return (vcpu->vlapic); 2038 } 2039 2040 struct vioapic * 2041 vm_ioapic(struct vm *vm) 2042 { 2043 2044 return (vm->vioapic); 2045 } 2046 2047 struct vhpet * 2048 vm_hpet(struct vm *vm) 2049 { 2050 2051 return (vm->vhpet); 2052 } 2053 2054 bool 2055 vmm_is_pptdev(int bus, int slot, int func) 2056 { 2057 int b, f, i, n, s; 2058 char *val, *cp, *cp2; 2059 bool found; 2060 2061 /* 2062 * XXX 2063 * The length of an environment variable is limited to 128 bytes which 2064 * puts an upper limit on the number of passthru devices that may be 2065 * specified using a single environment variable. 2066 * 2067 * Work around this by scanning multiple environment variable 2068 * names instead of a single one - yuck! 2069 */ 2070 const char *names[] = { "pptdevs", "pptdevs2", "pptdevs3", NULL }; 2071 2072 /* set pptdevs="1/2/3 4/5/6 7/8/9 10/11/12" */ 2073 found = false; 2074 for (i = 0; names[i] != NULL && !found; i++) { 2075 cp = val = kern_getenv(names[i]); 2076 while (cp != NULL && *cp != '\0') { 2077 if ((cp2 = strchr(cp, ' ')) != NULL) 2078 *cp2 = '\0'; 2079 2080 n = sscanf(cp, "%d/%d/%d", &b, &s, &f); 2081 if (n == 3 && bus == b && slot == s && func == f) { 2082 found = true; 2083 break; 2084 } 2085 2086 if (cp2 != NULL) 2087 *cp2++ = ' '; 2088 2089 cp = cp2; 2090 } 2091 freeenv(val); 2092 } 2093 return (found); 2094 } 2095 2096 void * 2097 vm_iommu_domain(struct vm *vm) 2098 { 2099 2100 return (vm->iommu); 2101 } 2102 2103 int 2104 vcpu_set_state(struct vcpu *vcpu, enum vcpu_state newstate, bool from_idle) 2105 { 2106 int error; 2107 2108 vcpu_lock(vcpu); 2109 error = vcpu_set_state_locked(vcpu, newstate, from_idle); 2110 vcpu_unlock(vcpu); 2111 2112 return (error); 2113 } 2114 2115 enum vcpu_state 2116 vcpu_get_state(struct vcpu *vcpu, int *hostcpu) 2117 { 2118 enum vcpu_state state; 2119 2120 vcpu_lock(vcpu); 2121 state = vcpu->state; 2122 if (hostcpu != NULL) 2123 *hostcpu = vcpu->hostcpu; 2124 vcpu_unlock(vcpu); 2125 2126 return (state); 2127 } 2128 2129 int 2130 vm_activate_cpu(struct vcpu *vcpu) 2131 { 2132 struct vm *vm = vcpu->vm; 2133 2134 if (CPU_ISSET(vcpu->vcpuid, &vm->active_cpus)) 2135 return (EBUSY); 2136 2137 VMM_CTR0(vcpu, "activated"); 2138 CPU_SET_ATOMIC(vcpu->vcpuid, &vm->active_cpus); 2139 return (0); 2140 } 2141 2142 int 2143 vm_suspend_cpu(struct vm *vm, struct vcpu *vcpu) 2144 { 2145 if (vcpu == NULL) { 2146 vm->debug_cpus = vm->active_cpus; 2147 for (int i = 0; i < vm->maxcpus; i++) { 2148 if (CPU_ISSET(i, &vm->active_cpus)) 2149 vcpu_notify_event(vm_vcpu(vm, i), false); 2150 } 2151 } else { 2152 if (!CPU_ISSET(vcpu->vcpuid, &vm->active_cpus)) 2153 return (EINVAL); 2154 2155 CPU_SET_ATOMIC(vcpu->vcpuid, &vm->debug_cpus); 2156 vcpu_notify_event(vcpu, false); 2157 } 2158 return (0); 2159 } 2160 2161 int 2162 vm_resume_cpu(struct vm *vm, struct vcpu *vcpu) 2163 { 2164 2165 if (vcpu == NULL) { 2166 CPU_ZERO(&vm->debug_cpus); 2167 } else { 2168 if (!CPU_ISSET(vcpu->vcpuid, &vm->debug_cpus)) 2169 return (EINVAL); 2170 2171 CPU_CLR_ATOMIC(vcpu->vcpuid, &vm->debug_cpus); 2172 } 2173 return (0); 2174 } 2175 2176 int 2177 vcpu_debugged(struct vcpu *vcpu) 2178 { 2179 2180 return (CPU_ISSET(vcpu->vcpuid, &vcpu->vm->debug_cpus)); 2181 } 2182 2183 cpuset_t 2184 vm_active_cpus(struct vm *vm) 2185 { 2186 2187 return (vm->active_cpus); 2188 } 2189 2190 cpuset_t 2191 vm_debug_cpus(struct vm *vm) 2192 { 2193 2194 return (vm->debug_cpus); 2195 } 2196 2197 cpuset_t 2198 vm_suspended_cpus(struct vm *vm) 2199 { 2200 2201 return (vm->suspended_cpus); 2202 } 2203 2204 /* 2205 * Returns the subset of vCPUs in tostart that are awaiting startup. 2206 * These vCPUs are also marked as no longer awaiting startup. 2207 */ 2208 cpuset_t 2209 vm_start_cpus(struct vm *vm, const cpuset_t *tostart) 2210 { 2211 cpuset_t set; 2212 2213 mtx_lock(&vm->rendezvous_mtx); 2214 CPU_AND(&set, &vm->startup_cpus, tostart); 2215 CPU_ANDNOT(&vm->startup_cpus, &vm->startup_cpus, &set); 2216 mtx_unlock(&vm->rendezvous_mtx); 2217 return (set); 2218 } 2219 2220 void 2221 vm_await_start(struct vm *vm, const cpuset_t *waiting) 2222 { 2223 mtx_lock(&vm->rendezvous_mtx); 2224 CPU_OR(&vm->startup_cpus, &vm->startup_cpus, waiting); 2225 mtx_unlock(&vm->rendezvous_mtx); 2226 } 2227 2228 void * 2229 vcpu_stats(struct vcpu *vcpu) 2230 { 2231 2232 return (vcpu->stats); 2233 } 2234 2235 int 2236 vm_get_x2apic_state(struct vcpu *vcpu, enum x2apic_state *state) 2237 { 2238 *state = vcpu->x2apic_state; 2239 2240 return (0); 2241 } 2242 2243 int 2244 vm_set_x2apic_state(struct vcpu *vcpu, enum x2apic_state state) 2245 { 2246 if (state >= X2APIC_STATE_LAST) 2247 return (EINVAL); 2248 2249 vcpu->x2apic_state = state; 2250 2251 vlapic_set_x2apic_state(vcpu, state); 2252 2253 return (0); 2254 } 2255 2256 /* 2257 * This function is called to ensure that a vcpu "sees" a pending event 2258 * as soon as possible: 2259 * - If the vcpu thread is sleeping then it is woken up. 2260 * - If the vcpu is running on a different host_cpu then an IPI will be directed 2261 * to the host_cpu to cause the vcpu to trap into the hypervisor. 2262 */ 2263 static void 2264 vcpu_notify_event_locked(struct vcpu *vcpu, bool lapic_intr) 2265 { 2266 int hostcpu; 2267 2268 hostcpu = vcpu->hostcpu; 2269 if (vcpu->state == VCPU_RUNNING) { 2270 KASSERT(hostcpu != NOCPU, ("vcpu running on invalid hostcpu")); 2271 if (hostcpu != curcpu) { 2272 if (lapic_intr) { 2273 vlapic_post_intr(vcpu->vlapic, hostcpu, 2274 vmm_ipinum); 2275 } else { 2276 ipi_cpu(hostcpu, vmm_ipinum); 2277 } 2278 } else { 2279 /* 2280 * If the 'vcpu' is running on 'curcpu' then it must 2281 * be sending a notification to itself (e.g. SELF_IPI). 2282 * The pending event will be picked up when the vcpu 2283 * transitions back to guest context. 2284 */ 2285 } 2286 } else { 2287 KASSERT(hostcpu == NOCPU, ("vcpu state %d not consistent " 2288 "with hostcpu %d", vcpu->state, hostcpu)); 2289 if (vcpu->state == VCPU_SLEEPING) 2290 wakeup_one(vcpu); 2291 } 2292 } 2293 2294 void 2295 vcpu_notify_event(struct vcpu *vcpu, bool lapic_intr) 2296 { 2297 vcpu_lock(vcpu); 2298 vcpu_notify_event_locked(vcpu, lapic_intr); 2299 vcpu_unlock(vcpu); 2300 } 2301 2302 struct vm_mem * 2303 vm_mem(struct vm *vm) 2304 { 2305 return (&vm->mem); 2306 } 2307 2308 int 2309 vm_apicid2vcpuid(struct vm *vm, int apicid) 2310 { 2311 /* 2312 * XXX apic id is assumed to be numerically identical to vcpu id 2313 */ 2314 return (apicid); 2315 } 2316 2317 int 2318 vm_smp_rendezvous(struct vcpu *vcpu, cpuset_t dest, 2319 vm_rendezvous_func_t func, void *arg) 2320 { 2321 struct vm *vm = vcpu->vm; 2322 int error, i; 2323 2324 /* 2325 * Enforce that this function is called without any locks 2326 */ 2327 WITNESS_WARN(WARN_PANIC, NULL, "vm_smp_rendezvous"); 2328 2329 restart: 2330 mtx_lock(&vm->rendezvous_mtx); 2331 if (vm->rendezvous_func != NULL) { 2332 /* 2333 * If a rendezvous is already in progress then we need to 2334 * call the rendezvous handler in case this 'vcpu' is one 2335 * of the targets of the rendezvous. 2336 */ 2337 VMM_CTR0(vcpu, "Rendezvous already in progress"); 2338 mtx_unlock(&vm->rendezvous_mtx); 2339 error = vm_handle_rendezvous(vcpu); 2340 if (error != 0) 2341 return (error); 2342 goto restart; 2343 } 2344 KASSERT(vm->rendezvous_func == NULL, ("vm_smp_rendezvous: previous " 2345 "rendezvous is still in progress")); 2346 2347 VMM_CTR0(vcpu, "Initiating rendezvous"); 2348 vm->rendezvous_req_cpus = dest; 2349 CPU_ZERO(&vm->rendezvous_done_cpus); 2350 vm->rendezvous_arg = arg; 2351 vm->rendezvous_func = func; 2352 mtx_unlock(&vm->rendezvous_mtx); 2353 2354 /* 2355 * Wake up any sleeping vcpus and trigger a VM-exit in any running 2356 * vcpus so they handle the rendezvous as soon as possible. 2357 */ 2358 for (i = 0; i < vm->maxcpus; i++) { 2359 if (CPU_ISSET(i, &dest)) 2360 vcpu_notify_event(vm_vcpu(vm, i), false); 2361 } 2362 2363 return (vm_handle_rendezvous(vcpu)); 2364 } 2365 2366 struct vatpic * 2367 vm_atpic(struct vm *vm) 2368 { 2369 return (vm->vatpic); 2370 } 2371 2372 struct vatpit * 2373 vm_atpit(struct vm *vm) 2374 { 2375 return (vm->vatpit); 2376 } 2377 2378 struct vpmtmr * 2379 vm_pmtmr(struct vm *vm) 2380 { 2381 2382 return (vm->vpmtmr); 2383 } 2384 2385 struct vrtc * 2386 vm_rtc(struct vm *vm) 2387 { 2388 2389 return (vm->vrtc); 2390 } 2391 2392 enum vm_reg_name 2393 vm_segment_name(int seg) 2394 { 2395 static enum vm_reg_name seg_names[] = { 2396 VM_REG_GUEST_ES, 2397 VM_REG_GUEST_CS, 2398 VM_REG_GUEST_SS, 2399 VM_REG_GUEST_DS, 2400 VM_REG_GUEST_FS, 2401 VM_REG_GUEST_GS 2402 }; 2403 2404 KASSERT(seg >= 0 && seg < nitems(seg_names), 2405 ("%s: invalid segment encoding %d", __func__, seg)); 2406 return (seg_names[seg]); 2407 } 2408 2409 void 2410 vm_copy_teardown(struct vm_copyinfo *copyinfo, int num_copyinfo) 2411 { 2412 int idx; 2413 2414 for (idx = 0; idx < num_copyinfo; idx++) { 2415 if (copyinfo[idx].cookie != NULL) 2416 vm_gpa_release(copyinfo[idx].cookie); 2417 } 2418 bzero(copyinfo, num_copyinfo * sizeof(struct vm_copyinfo)); 2419 } 2420 2421 int 2422 vm_copy_setup(struct vcpu *vcpu, struct vm_guest_paging *paging, 2423 uint64_t gla, size_t len, int prot, struct vm_copyinfo *copyinfo, 2424 int num_copyinfo, int *fault) 2425 { 2426 int error, idx, nused; 2427 size_t n, off, remaining; 2428 void *hva, *cookie; 2429 uint64_t gpa; 2430 2431 bzero(copyinfo, sizeof(struct vm_copyinfo) * num_copyinfo); 2432 2433 nused = 0; 2434 remaining = len; 2435 while (remaining > 0) { 2436 if (nused >= num_copyinfo) 2437 return (EFAULT); 2438 error = vm_gla2gpa(vcpu, paging, gla, prot, &gpa, fault); 2439 if (error || *fault) 2440 return (error); 2441 off = gpa & PAGE_MASK; 2442 n = min(remaining, PAGE_SIZE - off); 2443 copyinfo[nused].gpa = gpa; 2444 copyinfo[nused].len = n; 2445 remaining -= n; 2446 gla += n; 2447 nused++; 2448 } 2449 2450 for (idx = 0; idx < nused; idx++) { 2451 hva = vm_gpa_hold(vcpu, copyinfo[idx].gpa, 2452 copyinfo[idx].len, prot, &cookie); 2453 if (hva == NULL) 2454 break; 2455 copyinfo[idx].hva = hva; 2456 copyinfo[idx].cookie = cookie; 2457 } 2458 2459 if (idx != nused) { 2460 vm_copy_teardown(copyinfo, num_copyinfo); 2461 return (EFAULT); 2462 } else { 2463 *fault = 0; 2464 return (0); 2465 } 2466 } 2467 2468 void 2469 vm_copyin(struct vm_copyinfo *copyinfo, void *kaddr, size_t len) 2470 { 2471 char *dst; 2472 int idx; 2473 2474 dst = kaddr; 2475 idx = 0; 2476 while (len > 0) { 2477 bcopy(copyinfo[idx].hva, dst, copyinfo[idx].len); 2478 len -= copyinfo[idx].len; 2479 dst += copyinfo[idx].len; 2480 idx++; 2481 } 2482 } 2483 2484 void 2485 vm_copyout(const void *kaddr, struct vm_copyinfo *copyinfo, size_t len) 2486 { 2487 const char *src; 2488 int idx; 2489 2490 src = kaddr; 2491 idx = 0; 2492 while (len > 0) { 2493 bcopy(src, copyinfo[idx].hva, copyinfo[idx].len); 2494 len -= copyinfo[idx].len; 2495 src += copyinfo[idx].len; 2496 idx++; 2497 } 2498 } 2499 2500 /* 2501 * Return the amount of in-use and wired memory for the VM. Since 2502 * these are global stats, only return the values with for vCPU 0 2503 */ 2504 VMM_STAT_DECLARE(VMM_MEM_RESIDENT); 2505 VMM_STAT_DECLARE(VMM_MEM_WIRED); 2506 2507 static void 2508 vm_get_rescnt(struct vcpu *vcpu, struct vmm_stat_type *stat) 2509 { 2510 2511 if (vcpu->vcpuid == 0) { 2512 vmm_stat_set(vcpu, VMM_MEM_RESIDENT, PAGE_SIZE * 2513 vmspace_resident_count(vm_vmspace(vcpu->vm))); 2514 } 2515 } 2516 2517 static void 2518 vm_get_wiredcnt(struct vcpu *vcpu, struct vmm_stat_type *stat) 2519 { 2520 2521 if (vcpu->vcpuid == 0) { 2522 vmm_stat_set(vcpu, VMM_MEM_WIRED, PAGE_SIZE * 2523 pmap_wired_count(vmspace_pmap(vm_vmspace(vcpu->vm)))); 2524 } 2525 } 2526 2527 VMM_STAT_FUNC(VMM_MEM_RESIDENT, "Resident memory", vm_get_rescnt); 2528 VMM_STAT_FUNC(VMM_MEM_WIRED, "Wired memory", vm_get_wiredcnt); 2529 2530 #ifdef BHYVE_SNAPSHOT 2531 static int 2532 vm_snapshot_vcpus(struct vm *vm, struct vm_snapshot_meta *meta) 2533 { 2534 uint64_t tsc, now; 2535 int ret; 2536 struct vcpu *vcpu; 2537 uint16_t i, maxcpus; 2538 2539 now = rdtsc(); 2540 maxcpus = vm_get_maxcpus(vm); 2541 for (i = 0; i < maxcpus; i++) { 2542 vcpu = vm->vcpu[i]; 2543 if (vcpu == NULL) 2544 continue; 2545 2546 SNAPSHOT_VAR_OR_LEAVE(vcpu->x2apic_state, meta, ret, done); 2547 SNAPSHOT_VAR_OR_LEAVE(vcpu->exitintinfo, meta, ret, done); 2548 SNAPSHOT_VAR_OR_LEAVE(vcpu->exc_vector, meta, ret, done); 2549 SNAPSHOT_VAR_OR_LEAVE(vcpu->exc_errcode_valid, meta, ret, done); 2550 SNAPSHOT_VAR_OR_LEAVE(vcpu->exc_errcode, meta, ret, done); 2551 SNAPSHOT_VAR_OR_LEAVE(vcpu->guest_xcr0, meta, ret, done); 2552 SNAPSHOT_VAR_OR_LEAVE(vcpu->exitinfo, meta, ret, done); 2553 SNAPSHOT_VAR_OR_LEAVE(vcpu->nextrip, meta, ret, done); 2554 2555 /* 2556 * Save the absolute TSC value by adding now to tsc_offset. 2557 * 2558 * It will be turned turned back into an actual offset when the 2559 * TSC restore function is called 2560 */ 2561 tsc = now + vcpu->tsc_offset; 2562 SNAPSHOT_VAR_OR_LEAVE(tsc, meta, ret, done); 2563 if (meta->op == VM_SNAPSHOT_RESTORE) 2564 vcpu->tsc_offset = tsc; 2565 } 2566 2567 done: 2568 return (ret); 2569 } 2570 2571 static int 2572 vm_snapshot_vm(struct vm *vm, struct vm_snapshot_meta *meta) 2573 { 2574 int ret; 2575 2576 ret = vm_snapshot_vcpus(vm, meta); 2577 if (ret != 0) 2578 goto done; 2579 2580 SNAPSHOT_VAR_OR_LEAVE(vm->startup_cpus, meta, ret, done); 2581 done: 2582 return (ret); 2583 } 2584 2585 static int 2586 vm_snapshot_vcpu(struct vm *vm, struct vm_snapshot_meta *meta) 2587 { 2588 int error; 2589 struct vcpu *vcpu; 2590 uint16_t i, maxcpus; 2591 2592 error = 0; 2593 2594 maxcpus = vm_get_maxcpus(vm); 2595 for (i = 0; i < maxcpus; i++) { 2596 vcpu = vm->vcpu[i]; 2597 if (vcpu == NULL) 2598 continue; 2599 2600 error = vmmops_vcpu_snapshot(vcpu->cookie, meta); 2601 if (error != 0) { 2602 printf("%s: failed to snapshot vmcs/vmcb data for " 2603 "vCPU: %d; error: %d\n", __func__, i, error); 2604 goto done; 2605 } 2606 } 2607 2608 done: 2609 return (error); 2610 } 2611 2612 /* 2613 * Save kernel-side structures to user-space for snapshotting. 2614 */ 2615 int 2616 vm_snapshot_req(struct vm *vm, struct vm_snapshot_meta *meta) 2617 { 2618 int ret = 0; 2619 2620 switch (meta->dev_req) { 2621 case STRUCT_VMCX: 2622 ret = vm_snapshot_vcpu(vm, meta); 2623 break; 2624 case STRUCT_VM: 2625 ret = vm_snapshot_vm(vm, meta); 2626 break; 2627 case STRUCT_VIOAPIC: 2628 ret = vioapic_snapshot(vm_ioapic(vm), meta); 2629 break; 2630 case STRUCT_VLAPIC: 2631 ret = vlapic_snapshot(vm, meta); 2632 break; 2633 case STRUCT_VHPET: 2634 ret = vhpet_snapshot(vm_hpet(vm), meta); 2635 break; 2636 case STRUCT_VATPIC: 2637 ret = vatpic_snapshot(vm_atpic(vm), meta); 2638 break; 2639 case STRUCT_VATPIT: 2640 ret = vatpit_snapshot(vm_atpit(vm), meta); 2641 break; 2642 case STRUCT_VPMTMR: 2643 ret = vpmtmr_snapshot(vm_pmtmr(vm), meta); 2644 break; 2645 case STRUCT_VRTC: 2646 ret = vrtc_snapshot(vm_rtc(vm), meta); 2647 break; 2648 default: 2649 printf("%s: failed to find the requested type %#x\n", 2650 __func__, meta->dev_req); 2651 ret = (EINVAL); 2652 } 2653 return (ret); 2654 } 2655 2656 void 2657 vm_set_tsc_offset(struct vcpu *vcpu, uint64_t offset) 2658 { 2659 vcpu->tsc_offset = offset; 2660 } 2661 2662 int 2663 vm_restore_time(struct vm *vm) 2664 { 2665 int error; 2666 uint64_t now; 2667 struct vcpu *vcpu; 2668 uint16_t i, maxcpus; 2669 2670 now = rdtsc(); 2671 2672 error = vhpet_restore_time(vm_hpet(vm)); 2673 if (error) 2674 return (error); 2675 2676 maxcpus = vm_get_maxcpus(vm); 2677 for (i = 0; i < maxcpus; i++) { 2678 vcpu = vm->vcpu[i]; 2679 if (vcpu == NULL) 2680 continue; 2681 2682 error = vmmops_restore_tsc(vcpu->cookie, 2683 vcpu->tsc_offset - now); 2684 if (error) 2685 return (error); 2686 } 2687 2688 return (0); 2689 } 2690 #endif 2691