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/sysctl.h> 35 #include <sys/malloc.h> 36 #include <sys/pcpu.h> 37 #include <sys/lock.h> 38 #include <sys/mutex.h> 39 #include <sys/proc.h> 40 #include <sys/rwlock.h> 41 #include <sys/sched.h> 42 #include <sys/smp.h> 43 #include <sys/sx.h> 44 #include <sys/vnode.h> 45 46 #include <vm/vm.h> 47 #include <vm/vm_param.h> 48 #include <vm/vm_extern.h> 49 #include <vm/vm_object.h> 50 #include <vm/vm_page.h> 51 #include <vm/pmap.h> 52 #include <vm/vm_map.h> 53 #include <vm/vm_pager.h> 54 #include <vm/vm_kern.h> 55 #include <vm/vnode_pager.h> 56 #include <vm/swap_pager.h> 57 #include <vm/uma.h> 58 59 #include <machine/cpu.h> 60 #include <machine/pcb.h> 61 #include <machine/smp.h> 62 #include <machine/md_var.h> 63 #include <x86/psl.h> 64 #include <x86/apicreg.h> 65 #include <x86/ifunc.h> 66 67 #include <machine/vmm.h> 68 #include <machine/vmm_instruction_emul.h> 69 #include <machine/vmm_snapshot.h> 70 71 #include <dev/vmm/vmm_dev.h> 72 #include <dev/vmm/vmm_ktr.h> 73 #include <dev/vmm/vmm_mem.h> 74 75 #include "vmm_ioport.h" 76 #include "vmm_host.h" 77 #include "vmm_mem.h" 78 #include "vmm_util.h" 79 #include "vatpic.h" 80 #include "vatpit.h" 81 #include "vhpet.h" 82 #include "vioapic.h" 83 #include "vlapic.h" 84 #include "vpmtmr.h" 85 #include "vrtc.h" 86 #include "vmm_stat.h" 87 #include "vmm_lapic.h" 88 89 #include "io/ppt.h" 90 #include "io/iommu.h" 91 92 struct vlapic; 93 94 /* 95 * Initialization: 96 * (a) allocated when vcpu is created 97 * (i) initialized when vcpu is created and when it is reinitialized 98 * (o) initialized the first time the vcpu is created 99 * (x) initialized before use 100 */ 101 struct vcpu { 102 struct mtx mtx; /* (o) protects 'state' and 'hostcpu' */ 103 enum vcpu_state state; /* (o) vcpu state */ 104 int vcpuid; /* (o) */ 105 int hostcpu; /* (o) vcpu's host cpu */ 106 int reqidle; /* (i) request vcpu to idle */ 107 struct vm *vm; /* (o) */ 108 void *cookie; /* (i) cpu-specific data */ 109 struct vlapic *vlapic; /* (i) APIC device model */ 110 enum x2apic_state x2apic_state; /* (i) APIC mode */ 111 uint64_t exitintinfo; /* (i) events pending at VM exit */ 112 int nmi_pending; /* (i) NMI pending */ 113 int extint_pending; /* (i) INTR pending */ 114 int exception_pending; /* (i) exception pending */ 115 int exc_vector; /* (x) exception collateral */ 116 int exc_errcode_valid; 117 uint32_t exc_errcode; 118 struct savefpu *guestfpu; /* (a,i) guest fpu state */ 119 uint64_t guest_xcr0; /* (i) guest %xcr0 register */ 120 void *stats; /* (a,i) statistics */ 121 struct vm_exit exitinfo; /* (x) exit reason and collateral */ 122 cpuset_t exitinfo_cpuset; /* (x) storage for vmexit handlers */ 123 uint64_t nextrip; /* (x) next instruction to execute */ 124 uint64_t tsc_offset; /* (o) TSC offsetting */ 125 }; 126 127 #define vcpu_lock_init(v) mtx_init(&((v)->mtx), "vcpu lock", 0, MTX_SPIN) 128 #define vcpu_lock_destroy(v) mtx_destroy(&((v)->mtx)) 129 #define vcpu_lock(v) mtx_lock_spin(&((v)->mtx)) 130 #define vcpu_unlock(v) mtx_unlock_spin(&((v)->mtx)) 131 #define vcpu_assert_locked(v) mtx_assert(&((v)->mtx), MA_OWNED) 132 133 /* 134 * Initialization: 135 * (o) initialized the first time the VM is created 136 * (i) initialized when VM is created and when it is reinitialized 137 * (x) initialized before use 138 * 139 * Locking: 140 * [m] mem_segs_lock 141 * [r] rendezvous_mtx 142 * [v] reads require one frozen vcpu, writes require freezing all vcpus 143 */ 144 struct vm { 145 void *cookie; /* (i) cpu-specific data */ 146 void *iommu; /* (x) iommu-specific data */ 147 struct vhpet *vhpet; /* (i) virtual HPET */ 148 struct vioapic *vioapic; /* (i) virtual ioapic */ 149 struct vatpic *vatpic; /* (i) virtual atpic */ 150 struct vatpit *vatpit; /* (i) virtual atpit */ 151 struct vpmtmr *vpmtmr; /* (i) virtual ACPI PM timer */ 152 struct vrtc *vrtc; /* (o) virtual RTC */ 153 volatile cpuset_t active_cpus; /* (i) active vcpus */ 154 volatile cpuset_t debug_cpus; /* (i) vcpus stopped for debug */ 155 cpuset_t startup_cpus; /* (i) [r] waiting for startup */ 156 int suspend; /* (i) stop VM execution */ 157 bool dying; /* (o) is dying */ 158 volatile cpuset_t suspended_cpus; /* (i) suspended vcpus */ 159 volatile cpuset_t halted_cpus; /* (x) cpus in a hard halt */ 160 cpuset_t rendezvous_req_cpus; /* (x) [r] rendezvous requested */ 161 cpuset_t rendezvous_done_cpus; /* (x) [r] rendezvous finished */ 162 void *rendezvous_arg; /* (x) [r] rendezvous func/arg */ 163 vm_rendezvous_func_t rendezvous_func; 164 struct mtx rendezvous_mtx; /* (o) rendezvous lock */ 165 struct vm_mem mem; /* (i) [m+v] guest memory */ 166 char name[VM_MAX_NAMELEN+1]; /* (o) virtual machine name */ 167 struct vcpu **vcpu; /* (o) guest vcpus */ 168 /* The following describe the vm cpu topology */ 169 uint16_t sockets; /* (o) num of sockets */ 170 uint16_t cores; /* (o) num of cores/socket */ 171 uint16_t threads; /* (o) num of threads/core */ 172 uint16_t maxcpus; /* (o) max pluggable cpus */ 173 struct sx vcpus_init_lock; /* (o) */ 174 }; 175 176 #define VMM_CTR0(vcpu, format) \ 177 VCPU_CTR0((vcpu)->vm, (vcpu)->vcpuid, format) 178 179 #define VMM_CTR1(vcpu, format, p1) \ 180 VCPU_CTR1((vcpu)->vm, (vcpu)->vcpuid, format, p1) 181 182 #define VMM_CTR2(vcpu, format, p1, p2) \ 183 VCPU_CTR2((vcpu)->vm, (vcpu)->vcpuid, format, p1, p2) 184 185 #define VMM_CTR3(vcpu, format, p1, p2, p3) \ 186 VCPU_CTR3((vcpu)->vm, (vcpu)->vcpuid, format, p1, p2, p3) 187 188 #define VMM_CTR4(vcpu, format, p1, p2, p3, p4) \ 189 VCPU_CTR4((vcpu)->vm, (vcpu)->vcpuid, format, p1, p2, p3, p4) 190 191 static void vmmops_panic(void); 192 193 static void 194 vmmops_panic(void) 195 { 196 panic("vmm_ops func called when !vmm_is_intel() && !vmm_is_svm()"); 197 } 198 199 #define DEFINE_VMMOPS_IFUNC(ret_type, opname, args) \ 200 DEFINE_IFUNC(, ret_type, vmmops_##opname, args) \ 201 { \ 202 if (vmm_is_intel()) \ 203 return (vmm_ops_intel.opname); \ 204 else if (vmm_is_svm()) \ 205 return (vmm_ops_amd.opname); \ 206 else \ 207 return ((ret_type (*)args)vmmops_panic); \ 208 } 209 210 DEFINE_VMMOPS_IFUNC(int, modinit, (int ipinum)) 211 DEFINE_VMMOPS_IFUNC(int, modcleanup, (void)) 212 DEFINE_VMMOPS_IFUNC(void, modsuspend, (void)) 213 DEFINE_VMMOPS_IFUNC(void, modresume, (void)) 214 DEFINE_VMMOPS_IFUNC(void *, init, (struct vm *vm, struct pmap *pmap)) 215 DEFINE_VMMOPS_IFUNC(int, run, (void *vcpui, register_t rip, struct pmap *pmap, 216 struct vm_eventinfo *info)) 217 DEFINE_VMMOPS_IFUNC(void, cleanup, (void *vmi)) 218 DEFINE_VMMOPS_IFUNC(void *, vcpu_init, (void *vmi, struct vcpu *vcpu, 219 int vcpu_id)) 220 DEFINE_VMMOPS_IFUNC(void, vcpu_cleanup, (void *vcpui)) 221 DEFINE_VMMOPS_IFUNC(int, getreg, (void *vcpui, int num, uint64_t *retval)) 222 DEFINE_VMMOPS_IFUNC(int, setreg, (void *vcpui, int num, uint64_t val)) 223 DEFINE_VMMOPS_IFUNC(int, getdesc, (void *vcpui, int num, struct seg_desc *desc)) 224 DEFINE_VMMOPS_IFUNC(int, setdesc, (void *vcpui, int num, struct seg_desc *desc)) 225 DEFINE_VMMOPS_IFUNC(int, getcap, (void *vcpui, int num, int *retval)) 226 DEFINE_VMMOPS_IFUNC(int, setcap, (void *vcpui, int num, int val)) 227 DEFINE_VMMOPS_IFUNC(struct vmspace *, vmspace_alloc, (vm_offset_t min, 228 vm_offset_t max)) 229 DEFINE_VMMOPS_IFUNC(void, vmspace_free, (struct vmspace *vmspace)) 230 DEFINE_VMMOPS_IFUNC(struct vlapic *, vlapic_init, (void *vcpui)) 231 DEFINE_VMMOPS_IFUNC(void, vlapic_cleanup, (struct vlapic *vlapic)) 232 #ifdef BHYVE_SNAPSHOT 233 DEFINE_VMMOPS_IFUNC(int, vcpu_snapshot, (void *vcpui, 234 struct vm_snapshot_meta *meta)) 235 DEFINE_VMMOPS_IFUNC(int, restore_tsc, (void *vcpui, uint64_t now)) 236 #endif 237 238 SDT_PROVIDER_DEFINE(vmm); 239 240 static MALLOC_DEFINE(M_VM, "vm", "vm"); 241 242 /* statistics */ 243 static VMM_STAT(VCPU_TOTAL_RUNTIME, "vcpu total runtime"); 244 245 SYSCTL_NODE(_hw, OID_AUTO, vmm, CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, 246 NULL); 247 248 /* 249 * Halt the guest if all vcpus are executing a HLT instruction with 250 * interrupts disabled. 251 */ 252 static int halt_detection_enabled = 1; 253 SYSCTL_INT(_hw_vmm, OID_AUTO, halt_detection, CTLFLAG_RDTUN, 254 &halt_detection_enabled, 0, 255 "Halt VM if all vcpus execute HLT with interrupts disabled"); 256 257 static int vmm_ipinum; 258 SYSCTL_INT(_hw_vmm, OID_AUTO, ipinum, CTLFLAG_RD, &vmm_ipinum, 0, 259 "IPI vector used for vcpu notifications"); 260 261 static int trace_guest_exceptions; 262 SYSCTL_INT(_hw_vmm, OID_AUTO, trace_guest_exceptions, CTLFLAG_RDTUN, 263 &trace_guest_exceptions, 0, 264 "Trap into hypervisor on all guest exceptions and reflect them back"); 265 266 static int trap_wbinvd; 267 SYSCTL_INT(_hw_vmm, OID_AUTO, trap_wbinvd, CTLFLAG_RDTUN, &trap_wbinvd, 0, 268 "WBINVD triggers a VM-exit"); 269 270 static void vcpu_notify_event_locked(struct vcpu *vcpu); 271 272 /* global statistics */ 273 VMM_STAT(VCPU_MIGRATIONS, "vcpu migration across host cpus"); 274 VMM_STAT(VMEXIT_COUNT, "total number of vm exits"); 275 VMM_STAT(VMEXIT_EXTINT, "vm exits due to external interrupt"); 276 VMM_STAT(VMEXIT_HLT, "number of times hlt was intercepted"); 277 VMM_STAT(VMEXIT_CR_ACCESS, "number of times %cr access was intercepted"); 278 VMM_STAT(VMEXIT_RDMSR, "number of times rdmsr was intercepted"); 279 VMM_STAT(VMEXIT_WRMSR, "number of times wrmsr was intercepted"); 280 VMM_STAT(VMEXIT_MTRAP, "number of monitor trap exits"); 281 VMM_STAT(VMEXIT_PAUSE, "number of times pause was intercepted"); 282 VMM_STAT(VMEXIT_INTR_WINDOW, "vm exits due to interrupt window opening"); 283 VMM_STAT(VMEXIT_NMI_WINDOW, "vm exits due to nmi window opening"); 284 VMM_STAT(VMEXIT_INOUT, "number of times in/out was intercepted"); 285 VMM_STAT(VMEXIT_CPUID, "number of times cpuid was intercepted"); 286 VMM_STAT(VMEXIT_NESTED_FAULT, "vm exits due to nested page fault"); 287 VMM_STAT(VMEXIT_INST_EMUL, "vm exits for instruction emulation"); 288 VMM_STAT(VMEXIT_UNKNOWN, "number of vm exits for unknown reason"); 289 VMM_STAT(VMEXIT_ASTPENDING, "number of times astpending at exit"); 290 VMM_STAT(VMEXIT_REQIDLE, "number of times idle requested at exit"); 291 VMM_STAT(VMEXIT_USERSPACE, "number of vm exits handled in userspace"); 292 VMM_STAT(VMEXIT_RENDEZVOUS, "number of times rendezvous pending at exit"); 293 VMM_STAT(VMEXIT_EXCEPTION, "number of vm exits due to exceptions"); 294 295 #ifdef KTR 296 static const char * 297 vcpu_state2str(enum vcpu_state state) 298 { 299 300 switch (state) { 301 case VCPU_IDLE: 302 return ("idle"); 303 case VCPU_FROZEN: 304 return ("frozen"); 305 case VCPU_RUNNING: 306 return ("running"); 307 case VCPU_SLEEPING: 308 return ("sleeping"); 309 default: 310 return ("unknown"); 311 } 312 } 313 #endif 314 315 static void 316 vcpu_cleanup(struct vcpu *vcpu, bool destroy) 317 { 318 vmmops_vlapic_cleanup(vcpu->vlapic); 319 vmmops_vcpu_cleanup(vcpu->cookie); 320 vcpu->cookie = NULL; 321 if (destroy) { 322 vmm_stat_free(vcpu->stats); 323 fpu_save_area_free(vcpu->guestfpu); 324 vcpu_lock_destroy(vcpu); 325 free(vcpu, M_VM); 326 } 327 } 328 329 static struct vcpu * 330 vcpu_alloc(struct vm *vm, int vcpu_id) 331 { 332 struct vcpu *vcpu; 333 334 KASSERT(vcpu_id >= 0 && vcpu_id < vm->maxcpus, 335 ("vcpu_init: invalid vcpu %d", vcpu_id)); 336 337 vcpu = malloc(sizeof(*vcpu), M_VM, M_WAITOK | M_ZERO); 338 vcpu_lock_init(vcpu); 339 vcpu->state = VCPU_IDLE; 340 vcpu->hostcpu = NOCPU; 341 vcpu->vcpuid = vcpu_id; 342 vcpu->vm = vm; 343 vcpu->guestfpu = fpu_save_area_alloc(); 344 vcpu->stats = vmm_stat_alloc(); 345 vcpu->tsc_offset = 0; 346 return (vcpu); 347 } 348 349 static void 350 vcpu_init(struct vcpu *vcpu) 351 { 352 vcpu->cookie = vmmops_vcpu_init(vcpu->vm->cookie, vcpu, vcpu->vcpuid); 353 vcpu->vlapic = vmmops_vlapic_init(vcpu->cookie); 354 vm_set_x2apic_state(vcpu, X2APIC_DISABLED); 355 vcpu->reqidle = 0; 356 vcpu->exitintinfo = 0; 357 vcpu->nmi_pending = 0; 358 vcpu->extint_pending = 0; 359 vcpu->exception_pending = 0; 360 vcpu->guest_xcr0 = XFEATURE_ENABLED_X87; 361 fpu_save_area_reset(vcpu->guestfpu); 362 vmm_stat_init(vcpu->stats); 363 } 364 365 int 366 vcpu_trace_exceptions(struct vcpu *vcpu) 367 { 368 369 return (trace_guest_exceptions); 370 } 371 372 int 373 vcpu_trap_wbinvd(struct vcpu *vcpu) 374 { 375 return (trap_wbinvd); 376 } 377 378 struct vm_exit * 379 vm_exitinfo(struct vcpu *vcpu) 380 { 381 return (&vcpu->exitinfo); 382 } 383 384 cpuset_t * 385 vm_exitinfo_cpuset(struct vcpu *vcpu) 386 { 387 return (&vcpu->exitinfo_cpuset); 388 } 389 390 int 391 vmm_modinit(void) 392 { 393 if (!vmm_is_hw_supported()) 394 return (ENXIO); 395 396 vmm_host_state_init(); 397 398 vmm_ipinum = lapic_ipi_alloc(pti ? &IDTVEC(justreturn1_pti) : 399 &IDTVEC(justreturn)); 400 if (vmm_ipinum < 0) 401 vmm_ipinum = IPI_AST; 402 403 vmm_suspend_p = vmmops_modsuspend; 404 vmm_resume_p = vmmops_modresume; 405 406 return (vmmops_modinit(vmm_ipinum)); 407 } 408 409 int 410 vmm_modcleanup(void) 411 { 412 vmm_suspend_p = NULL; 413 vmm_resume_p = NULL; 414 iommu_cleanup(); 415 if (vmm_ipinum != IPI_AST) 416 lapic_ipi_free(vmm_ipinum); 417 return (vmmops_modcleanup()); 418 } 419 420 static void 421 vm_init(struct vm *vm, bool create) 422 { 423 vm->cookie = vmmops_init(vm, vmspace_pmap(vm_vmspace(vm))); 424 vm->iommu = NULL; 425 vm->vioapic = vioapic_init(vm); 426 vm->vhpet = vhpet_init(vm); 427 vm->vatpic = vatpic_init(vm); 428 vm->vatpit = vatpit_init(vm); 429 vm->vpmtmr = vpmtmr_init(vm); 430 if (create) 431 vm->vrtc = vrtc_init(vm); 432 433 CPU_ZERO(&vm->active_cpus); 434 CPU_ZERO(&vm->debug_cpus); 435 CPU_ZERO(&vm->startup_cpus); 436 437 vm->suspend = 0; 438 CPU_ZERO(&vm->suspended_cpus); 439 440 if (!create) { 441 for (int i = 0; i < vm->maxcpus; i++) { 442 if (vm->vcpu[i] != NULL) 443 vcpu_init(vm->vcpu[i]); 444 } 445 } 446 } 447 448 void 449 vm_disable_vcpu_creation(struct vm *vm) 450 { 451 sx_xlock(&vm->vcpus_init_lock); 452 vm->dying = true; 453 sx_xunlock(&vm->vcpus_init_lock); 454 } 455 456 struct vcpu * 457 vm_alloc_vcpu(struct vm *vm, int vcpuid) 458 { 459 struct vcpu *vcpu; 460 461 if (vcpuid < 0 || vcpuid >= vm_get_maxcpus(vm)) 462 return (NULL); 463 464 vcpu = (struct vcpu *) 465 atomic_load_acq_ptr((uintptr_t *)&vm->vcpu[vcpuid]); 466 if (__predict_true(vcpu != NULL)) 467 return (vcpu); 468 469 sx_xlock(&vm->vcpus_init_lock); 470 vcpu = vm->vcpu[vcpuid]; 471 if (vcpu == NULL && !vm->dying) { 472 vcpu = vcpu_alloc(vm, vcpuid); 473 vcpu_init(vcpu); 474 475 /* 476 * Ensure vCPU is fully created before updating pointer 477 * to permit unlocked reads above. 478 */ 479 atomic_store_rel_ptr((uintptr_t *)&vm->vcpu[vcpuid], 480 (uintptr_t)vcpu); 481 } 482 sx_xunlock(&vm->vcpus_init_lock); 483 return (vcpu); 484 } 485 486 void 487 vm_lock_vcpus(struct vm *vm) 488 { 489 sx_xlock(&vm->vcpus_init_lock); 490 } 491 492 void 493 vm_unlock_vcpus(struct vm *vm) 494 { 495 sx_unlock(&vm->vcpus_init_lock); 496 } 497 498 int 499 vm_create(const char *name, struct vm **retvm) 500 { 501 struct vm *vm; 502 int error; 503 504 vm = malloc(sizeof(struct vm), M_VM, M_WAITOK | M_ZERO); 505 error = vm_mem_init(&vm->mem, 0, VM_MAXUSER_ADDRESS_LA48); 506 if (error != 0) { 507 free(vm, M_VM); 508 return (error); 509 } 510 strcpy(vm->name, name); 511 mtx_init(&vm->rendezvous_mtx, "vm rendezvous lock", 0, MTX_DEF); 512 sx_init(&vm->vcpus_init_lock, "vm vcpus"); 513 vm->vcpu = malloc(sizeof(*vm->vcpu) * vm_maxcpu, M_VM, M_WAITOK | 514 M_ZERO); 515 516 vm->sockets = 1; 517 vm->cores = 1; /* XXX backwards compatibility */ 518 vm->threads = 1; /* XXX backwards compatibility */ 519 vm->maxcpus = vm_maxcpu; 520 521 vm_init(vm, true); 522 523 *retvm = vm; 524 return (0); 525 } 526 527 void 528 vm_get_topology(struct vm *vm, uint16_t *sockets, uint16_t *cores, 529 uint16_t *threads, uint16_t *maxcpus) 530 { 531 *sockets = vm->sockets; 532 *cores = vm->cores; 533 *threads = vm->threads; 534 *maxcpus = vm->maxcpus; 535 } 536 537 uint16_t 538 vm_get_maxcpus(struct vm *vm) 539 { 540 return (vm->maxcpus); 541 } 542 543 int 544 vm_set_topology(struct vm *vm, uint16_t sockets, uint16_t cores, 545 uint16_t threads, uint16_t maxcpus __unused) 546 { 547 /* Ignore maxcpus. */ 548 if ((sockets * cores * threads) > vm->maxcpus) 549 return (EINVAL); 550 vm->sockets = sockets; 551 vm->cores = cores; 552 vm->threads = threads; 553 return(0); 554 } 555 556 static void 557 vm_cleanup(struct vm *vm, bool destroy) 558 { 559 if (destroy) 560 vm_xlock_memsegs(vm); 561 else 562 vm_assert_memseg_xlocked(vm); 563 564 ppt_unassign_all(vm); 565 566 if (vm->iommu != NULL) 567 iommu_destroy_domain(vm->iommu); 568 569 if (destroy) 570 vrtc_cleanup(vm->vrtc); 571 else 572 vrtc_reset(vm->vrtc); 573 vpmtmr_cleanup(vm->vpmtmr); 574 vatpit_cleanup(vm->vatpit); 575 vhpet_cleanup(vm->vhpet); 576 vatpic_cleanup(vm->vatpic); 577 vioapic_cleanup(vm->vioapic); 578 579 for (int i = 0; i < vm->maxcpus; i++) { 580 if (vm->vcpu[i] != NULL) 581 vcpu_cleanup(vm->vcpu[i], destroy); 582 } 583 584 vmmops_cleanup(vm->cookie); 585 586 vm_mem_cleanup(vm); 587 588 if (destroy) { 589 vm_mem_destroy(vm); 590 591 free(vm->vcpu, M_VM); 592 sx_destroy(&vm->vcpus_init_lock); 593 mtx_destroy(&vm->rendezvous_mtx); 594 } 595 } 596 597 void 598 vm_destroy(struct vm *vm) 599 { 600 vm_cleanup(vm, true); 601 free(vm, M_VM); 602 } 603 604 int 605 vm_reinit(struct vm *vm) 606 { 607 int error; 608 609 /* 610 * A virtual machine can be reset only if all vcpus are suspended. 611 */ 612 if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) { 613 vm_cleanup(vm, false); 614 vm_init(vm, false); 615 error = 0; 616 } else { 617 error = EBUSY; 618 } 619 620 return (error); 621 } 622 623 const char * 624 vm_name(struct vm *vm) 625 { 626 return (vm->name); 627 } 628 629 int 630 vm_map_mmio(struct vm *vm, vm_paddr_t gpa, size_t len, vm_paddr_t hpa) 631 { 632 return (vmm_mmio_alloc(vm_vmspace(vm), gpa, len, hpa)); 633 } 634 635 int 636 vm_unmap_mmio(struct vm *vm, vm_paddr_t gpa, size_t len) 637 { 638 639 vmm_mmio_free(vm_vmspace(vm), gpa, len); 640 return (0); 641 } 642 643 static int 644 vm_iommu_map(struct vm *vm) 645 { 646 pmap_t pmap; 647 vm_paddr_t gpa, hpa; 648 struct vm_mem_map *mm; 649 int error, i; 650 651 sx_assert(&vm->mem.mem_segs_lock, SX_LOCKED); 652 653 pmap = vmspace_pmap(vm_vmspace(vm)); 654 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 655 if (!vm_memseg_sysmem(vm, i)) 656 continue; 657 658 mm = &vm->mem.mem_maps[i]; 659 KASSERT((mm->flags & VM_MEMMAP_F_IOMMU) == 0, 660 ("iommu map found invalid memmap %#lx/%#lx/%#x", 661 mm->gpa, mm->len, mm->flags)); 662 if ((mm->flags & VM_MEMMAP_F_WIRED) == 0) 663 continue; 664 mm->flags |= VM_MEMMAP_F_IOMMU; 665 666 for (gpa = mm->gpa; gpa < mm->gpa + mm->len; gpa += PAGE_SIZE) { 667 hpa = pmap_extract(pmap, gpa); 668 669 /* 670 * All mappings in the vmm vmspace must be 671 * present since they are managed by vmm in this way. 672 * Because we are in pass-through mode, the 673 * mappings must also be wired. This implies 674 * that all pages must be mapped and wired, 675 * allowing to use pmap_extract() and avoiding the 676 * need to use vm_gpa_hold_global(). 677 * 678 * This could change if/when we start 679 * supporting page faults on IOMMU maps. 680 */ 681 KASSERT(vm_page_wired(PHYS_TO_VM_PAGE(hpa)), 682 ("vm_iommu_map: vm %p gpa %jx hpa %jx not wired", 683 vm, (uintmax_t)gpa, (uintmax_t)hpa)); 684 685 iommu_create_mapping(vm->iommu, gpa, hpa, PAGE_SIZE); 686 } 687 } 688 689 error = iommu_invalidate_tlb(iommu_host_domain()); 690 return (error); 691 } 692 693 static int 694 vm_iommu_unmap(struct vm *vm) 695 { 696 vm_paddr_t gpa; 697 struct vm_mem_map *mm; 698 int error, i; 699 700 sx_assert(&vm->mem.mem_segs_lock, SX_LOCKED); 701 702 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 703 if (!vm_memseg_sysmem(vm, i)) 704 continue; 705 706 mm = &vm->mem.mem_maps[i]; 707 if ((mm->flags & VM_MEMMAP_F_IOMMU) == 0) 708 continue; 709 mm->flags &= ~VM_MEMMAP_F_IOMMU; 710 KASSERT((mm->flags & VM_MEMMAP_F_WIRED) != 0, 711 ("iommu unmap found invalid memmap %#lx/%#lx/%#x", 712 mm->gpa, mm->len, mm->flags)); 713 714 for (gpa = mm->gpa; gpa < mm->gpa + mm->len; gpa += PAGE_SIZE) { 715 KASSERT(vm_page_wired(PHYS_TO_VM_PAGE(pmap_extract( 716 vmspace_pmap(vm_vmspace(vm)), gpa))), 717 ("vm_iommu_unmap: vm %p gpa %jx not wired", 718 vm, (uintmax_t)gpa)); 719 iommu_remove_mapping(vm->iommu, gpa, PAGE_SIZE); 720 } 721 } 722 723 /* 724 * Invalidate the cached translations associated with the domain 725 * from which pages were removed. 726 */ 727 error = iommu_invalidate_tlb(vm->iommu); 728 return (error); 729 } 730 731 int 732 vm_unassign_pptdev(struct vm *vm, int bus, int slot, int func) 733 { 734 int error; 735 736 error = ppt_unassign_device(vm, bus, slot, func); 737 if (error) 738 return (error); 739 740 if (ppt_assigned_devices(vm) == 0) 741 error = vm_iommu_unmap(vm); 742 743 return (error); 744 } 745 746 int 747 vm_assign_pptdev(struct vm *vm, int bus, int slot, int func) 748 { 749 int error; 750 vm_paddr_t maxaddr; 751 bool map = false; 752 753 /* Set up the IOMMU to do the 'gpa' to 'hpa' translation */ 754 if (ppt_assigned_devices(vm) == 0) { 755 KASSERT(vm->iommu == NULL, 756 ("vm_assign_pptdev: iommu must be NULL")); 757 maxaddr = vmm_sysmem_maxaddr(vm); 758 vm->iommu = iommu_create_domain(maxaddr); 759 if (vm->iommu == NULL) 760 return (ENXIO); 761 map = true; 762 } 763 764 error = ppt_assign_device(vm, bus, slot, func); 765 if (error == 0 && map) 766 error = vm_iommu_map(vm); 767 return (error); 768 } 769 770 int 771 vm_get_register(struct vcpu *vcpu, int reg, uint64_t *retval) 772 { 773 /* Negative values represent VM control structure fields. */ 774 if (reg >= VM_REG_LAST) 775 return (EINVAL); 776 777 return (vmmops_getreg(vcpu->cookie, reg, retval)); 778 } 779 780 int 781 vm_set_register(struct vcpu *vcpu, int reg, uint64_t val) 782 { 783 int error; 784 785 /* Negative values represent VM control structure fields. */ 786 if (reg >= VM_REG_LAST) 787 return (EINVAL); 788 789 error = vmmops_setreg(vcpu->cookie, reg, val); 790 if (error || reg != VM_REG_GUEST_RIP) 791 return (error); 792 793 /* Set 'nextrip' to match the value of %rip */ 794 VMM_CTR1(vcpu, "Setting nextrip to %#lx", val); 795 vcpu->nextrip = val; 796 return (0); 797 } 798 799 static bool 800 is_descriptor_table(int reg) 801 { 802 803 switch (reg) { 804 case VM_REG_GUEST_IDTR: 805 case VM_REG_GUEST_GDTR: 806 return (true); 807 default: 808 return (false); 809 } 810 } 811 812 static bool 813 is_segment_register(int reg) 814 { 815 816 switch (reg) { 817 case VM_REG_GUEST_ES: 818 case VM_REG_GUEST_CS: 819 case VM_REG_GUEST_SS: 820 case VM_REG_GUEST_DS: 821 case VM_REG_GUEST_FS: 822 case VM_REG_GUEST_GS: 823 case VM_REG_GUEST_TR: 824 case VM_REG_GUEST_LDTR: 825 return (true); 826 default: 827 return (false); 828 } 829 } 830 831 int 832 vm_get_seg_desc(struct vcpu *vcpu, int reg, struct seg_desc *desc) 833 { 834 835 if (!is_segment_register(reg) && !is_descriptor_table(reg)) 836 return (EINVAL); 837 838 return (vmmops_getdesc(vcpu->cookie, reg, desc)); 839 } 840 841 int 842 vm_set_seg_desc(struct vcpu *vcpu, int reg, struct seg_desc *desc) 843 { 844 845 if (!is_segment_register(reg) && !is_descriptor_table(reg)) 846 return (EINVAL); 847 848 return (vmmops_setdesc(vcpu->cookie, reg, desc)); 849 } 850 851 static void 852 restore_guest_fpustate(struct vcpu *vcpu) 853 { 854 855 /* flush host state to the pcb */ 856 fpuexit(curthread); 857 858 /* restore guest FPU state */ 859 fpu_enable(); 860 fpurestore(vcpu->guestfpu); 861 862 /* restore guest XCR0 if XSAVE is enabled in the host */ 863 if (rcr4() & CR4_XSAVE) 864 load_xcr(0, vcpu->guest_xcr0); 865 866 /* 867 * The FPU is now "dirty" with the guest's state so disable 868 * the FPU to trap any access by the host. 869 */ 870 fpu_disable(); 871 } 872 873 static void 874 save_guest_fpustate(struct vcpu *vcpu) 875 { 876 877 if ((rcr0() & CR0_TS) == 0) 878 panic("fpu emulation not enabled in host!"); 879 880 /* save guest XCR0 and restore host XCR0 */ 881 if (rcr4() & CR4_XSAVE) { 882 vcpu->guest_xcr0 = rxcr(0); 883 load_xcr(0, vmm_get_host_xcr0()); 884 } 885 886 /* save guest FPU state */ 887 fpu_enable(); 888 fpusave(vcpu->guestfpu); 889 fpu_disable(); 890 } 891 892 static VMM_STAT(VCPU_IDLE_TICKS, "number of ticks vcpu was idle"); 893 894 /* 895 * Invoke the rendezvous function on the specified vcpu if applicable. Return 896 * true if the rendezvous is finished, false otherwise. 897 */ 898 static bool 899 vm_rendezvous(struct vcpu *vcpu) 900 { 901 struct vm *vm = vcpu->vm; 902 int vcpuid; 903 904 mtx_assert(&vcpu->vm->rendezvous_mtx, MA_OWNED); 905 KASSERT(vcpu->vm->rendezvous_func != NULL, 906 ("vm_rendezvous: no rendezvous pending")); 907 908 /* 'rendezvous_req_cpus' must be a subset of 'active_cpus' */ 909 CPU_AND(&vm->rendezvous_req_cpus, &vm->rendezvous_req_cpus, 910 &vm->active_cpus); 911 912 vcpuid = vcpu->vcpuid; 913 if (CPU_ISSET(vcpuid, &vm->rendezvous_req_cpus) && 914 !CPU_ISSET(vcpuid, &vm->rendezvous_done_cpus)) { 915 VMM_CTR0(vcpu, "Calling rendezvous func"); 916 (*vm->rendezvous_func)(vcpu, vm->rendezvous_arg); 917 CPU_SET(vcpuid, &vm->rendezvous_done_cpus); 918 } 919 if (CPU_CMP(&vm->rendezvous_req_cpus, 920 &vm->rendezvous_done_cpus) == 0) { 921 VMM_CTR0(vcpu, "Rendezvous completed"); 922 CPU_ZERO(&vm->rendezvous_req_cpus); 923 vm->rendezvous_func = NULL; 924 wakeup(&vm->rendezvous_func); 925 return (true); 926 } 927 return (false); 928 } 929 930 static void 931 vcpu_wait_idle(struct vcpu *vcpu) 932 { 933 KASSERT(vcpu->state != VCPU_IDLE, ("vcpu already idle")); 934 935 vcpu->reqidle = 1; 936 vcpu_notify_event_locked(vcpu); 937 VMM_CTR1(vcpu, "vcpu state change from %s to " 938 "idle requested", vcpu_state2str(vcpu->state)); 939 msleep_spin(&vcpu->state, &vcpu->mtx, "vmstat", hz); 940 } 941 942 static int 943 vcpu_set_state_locked(struct vcpu *vcpu, enum vcpu_state newstate, 944 bool from_idle) 945 { 946 int error; 947 948 vcpu_assert_locked(vcpu); 949 950 /* 951 * State transitions from the vmmdev_ioctl() must always begin from 952 * the VCPU_IDLE state. This guarantees that there is only a single 953 * ioctl() operating on a vcpu at any point. 954 */ 955 if (from_idle) { 956 while (vcpu->state != VCPU_IDLE) 957 vcpu_wait_idle(vcpu); 958 } else { 959 KASSERT(vcpu->state != VCPU_IDLE, ("invalid transition from " 960 "vcpu idle state")); 961 } 962 963 if (vcpu->state == VCPU_RUNNING) { 964 KASSERT(vcpu->hostcpu == curcpu, ("curcpu %d and hostcpu %d " 965 "mismatch for running vcpu", curcpu, vcpu->hostcpu)); 966 } else { 967 KASSERT(vcpu->hostcpu == NOCPU, ("Invalid hostcpu %d for a " 968 "vcpu that is not running", vcpu->hostcpu)); 969 } 970 971 /* 972 * The following state transitions are allowed: 973 * IDLE -> FROZEN -> IDLE 974 * FROZEN -> RUNNING -> FROZEN 975 * FROZEN -> SLEEPING -> FROZEN 976 */ 977 switch (vcpu->state) { 978 case VCPU_IDLE: 979 case VCPU_RUNNING: 980 case VCPU_SLEEPING: 981 error = (newstate != VCPU_FROZEN); 982 break; 983 case VCPU_FROZEN: 984 error = (newstate == VCPU_FROZEN); 985 break; 986 default: 987 error = 1; 988 break; 989 } 990 991 if (error) 992 return (EBUSY); 993 994 VMM_CTR2(vcpu, "vcpu state changed from %s to %s", 995 vcpu_state2str(vcpu->state), vcpu_state2str(newstate)); 996 997 vcpu->state = newstate; 998 if (newstate == VCPU_RUNNING) 999 vcpu->hostcpu = curcpu; 1000 else 1001 vcpu->hostcpu = NOCPU; 1002 1003 if (newstate == VCPU_IDLE) 1004 wakeup(&vcpu->state); 1005 1006 return (0); 1007 } 1008 1009 /* 1010 * Try to lock all of the vCPUs in the VM while taking care to avoid deadlocks 1011 * with vm_smp_rendezvous(). 1012 * 1013 * The complexity here suggests that the rendezvous mechanism needs a rethink. 1014 */ 1015 int 1016 vcpu_set_state_all(struct vm *vm, enum vcpu_state newstate) 1017 { 1018 cpuset_t locked; 1019 struct vcpu *vcpu; 1020 int error, i; 1021 uint16_t maxcpus; 1022 1023 KASSERT(newstate != VCPU_IDLE, 1024 ("vcpu_set_state_all: invalid target state %d", newstate)); 1025 1026 error = 0; 1027 CPU_ZERO(&locked); 1028 maxcpus = vm->maxcpus; 1029 1030 mtx_lock(&vm->rendezvous_mtx); 1031 restart: 1032 if (vm->rendezvous_func != NULL) { 1033 /* 1034 * If we have a pending rendezvous, then the initiator may be 1035 * blocked waiting for other vCPUs to execute the callback. The 1036 * current thread may be a vCPU thread so we must not block 1037 * waiting for the initiator, otherwise we get a deadlock. 1038 * Thus, execute the callback on behalf of any idle vCPUs. 1039 */ 1040 for (i = 0; i < maxcpus; i++) { 1041 vcpu = vm_vcpu(vm, i); 1042 if (vcpu == NULL) 1043 continue; 1044 vcpu_lock(vcpu); 1045 if (vcpu->state == VCPU_IDLE) { 1046 (void)vcpu_set_state_locked(vcpu, VCPU_FROZEN, 1047 true); 1048 CPU_SET(i, &locked); 1049 } 1050 if (CPU_ISSET(i, &locked)) { 1051 /* 1052 * We can safely execute the callback on this 1053 * vCPU's behalf. 1054 */ 1055 vcpu_unlock(vcpu); 1056 (void)vm_rendezvous(vcpu); 1057 vcpu_lock(vcpu); 1058 } 1059 vcpu_unlock(vcpu); 1060 } 1061 } 1062 1063 /* 1064 * Now wait for remaining vCPUs to become idle. This may include the 1065 * initiator of a rendezvous that is currently blocked on the rendezvous 1066 * mutex. 1067 */ 1068 CPU_FOREACH_ISCLR(i, &locked) { 1069 if (i >= maxcpus) 1070 break; 1071 vcpu = vm_vcpu(vm, i); 1072 if (vcpu == NULL) 1073 continue; 1074 vcpu_lock(vcpu); 1075 while (vcpu->state != VCPU_IDLE) { 1076 mtx_unlock(&vm->rendezvous_mtx); 1077 vcpu_wait_idle(vcpu); 1078 vcpu_unlock(vcpu); 1079 mtx_lock(&vm->rendezvous_mtx); 1080 if (vm->rendezvous_func != NULL) 1081 goto restart; 1082 vcpu_lock(vcpu); 1083 } 1084 error = vcpu_set_state_locked(vcpu, newstate, true); 1085 vcpu_unlock(vcpu); 1086 if (error != 0) { 1087 /* Roll back state changes. */ 1088 CPU_FOREACH_ISSET(i, &locked) 1089 (void)vcpu_set_state(vcpu, VCPU_IDLE, false); 1090 break; 1091 } 1092 CPU_SET(i, &locked); 1093 } 1094 mtx_unlock(&vm->rendezvous_mtx); 1095 return (error); 1096 } 1097 1098 static void 1099 vcpu_require_state(struct vcpu *vcpu, enum vcpu_state newstate) 1100 { 1101 int error; 1102 1103 if ((error = vcpu_set_state(vcpu, newstate, false)) != 0) 1104 panic("Error %d setting state to %d\n", error, newstate); 1105 } 1106 1107 static void 1108 vcpu_require_state_locked(struct vcpu *vcpu, enum vcpu_state newstate) 1109 { 1110 int error; 1111 1112 if ((error = vcpu_set_state_locked(vcpu, newstate, false)) != 0) 1113 panic("Error %d setting state to %d", error, newstate); 1114 } 1115 1116 static int 1117 vm_handle_rendezvous(struct vcpu *vcpu) 1118 { 1119 struct vm *vm; 1120 struct thread *td; 1121 1122 td = curthread; 1123 vm = vcpu->vm; 1124 1125 mtx_lock(&vm->rendezvous_mtx); 1126 while (vm->rendezvous_func != NULL) { 1127 if (vm_rendezvous(vcpu)) 1128 break; 1129 1130 VMM_CTR0(vcpu, "Wait for rendezvous completion"); 1131 mtx_sleep(&vm->rendezvous_func, &vm->rendezvous_mtx, 0, 1132 "vmrndv", hz); 1133 if (td_ast_pending(td, TDA_SUSPEND)) { 1134 int error; 1135 1136 mtx_unlock(&vm->rendezvous_mtx); 1137 error = thread_check_susp(td, true); 1138 if (error != 0) 1139 return (error); 1140 mtx_lock(&vm->rendezvous_mtx); 1141 } 1142 } 1143 mtx_unlock(&vm->rendezvous_mtx); 1144 return (0); 1145 } 1146 1147 /* 1148 * Emulate a guest 'hlt' by sleeping until the vcpu is ready to run. 1149 */ 1150 static int 1151 vm_handle_hlt(struct vcpu *vcpu, bool intr_disabled, bool *retu) 1152 { 1153 struct vm *vm = vcpu->vm; 1154 const char *wmesg; 1155 struct thread *td; 1156 int error, t, vcpuid, vcpu_halted, vm_halted; 1157 1158 vcpuid = vcpu->vcpuid; 1159 vcpu_halted = 0; 1160 vm_halted = 0; 1161 error = 0; 1162 td = curthread; 1163 1164 KASSERT(!CPU_ISSET(vcpuid, &vm->halted_cpus), ("vcpu already halted")); 1165 1166 vcpu_lock(vcpu); 1167 while (1) { 1168 /* 1169 * Do a final check for pending NMI or interrupts before 1170 * really putting this thread to sleep. Also check for 1171 * software events that would cause this vcpu to wakeup. 1172 * 1173 * These interrupts/events could have happened after the 1174 * vcpu returned from vmmops_run() and before it acquired the 1175 * vcpu lock above. 1176 */ 1177 if (vm->rendezvous_func != NULL || vm->suspend || vcpu->reqidle) 1178 break; 1179 if (vm_nmi_pending(vcpu)) 1180 break; 1181 if (!intr_disabled) { 1182 if (vm_extint_pending(vcpu) || 1183 vlapic_pending_intr(vcpu->vlapic, NULL)) { 1184 break; 1185 } 1186 } 1187 1188 /* Don't go to sleep if the vcpu thread needs to yield */ 1189 if (vcpu_should_yield(vcpu)) 1190 break; 1191 1192 if (vcpu_debugged(vcpu)) 1193 break; 1194 1195 /* 1196 * Some Linux guests implement "halt" by having all vcpus 1197 * execute HLT with interrupts disabled. 'halted_cpus' keeps 1198 * track of the vcpus that have entered this state. When all 1199 * vcpus enter the halted state the virtual machine is halted. 1200 */ 1201 if (intr_disabled) { 1202 wmesg = "vmhalt"; 1203 VMM_CTR0(vcpu, "Halted"); 1204 if (!vcpu_halted && halt_detection_enabled) { 1205 vcpu_halted = 1; 1206 CPU_SET_ATOMIC(vcpuid, &vm->halted_cpus); 1207 } 1208 if (CPU_CMP(&vm->halted_cpus, &vm->active_cpus) == 0) { 1209 vm_halted = 1; 1210 break; 1211 } 1212 } else { 1213 wmesg = "vmidle"; 1214 } 1215 1216 t = ticks; 1217 vcpu_require_state_locked(vcpu, VCPU_SLEEPING); 1218 /* 1219 * XXX msleep_spin() cannot be interrupted by signals so 1220 * wake up periodically to check pending signals. 1221 */ 1222 msleep_spin(vcpu, &vcpu->mtx, wmesg, hz); 1223 vcpu_require_state_locked(vcpu, VCPU_FROZEN); 1224 vmm_stat_incr(vcpu, VCPU_IDLE_TICKS, ticks - t); 1225 if (td_ast_pending(td, TDA_SUSPEND)) { 1226 vcpu_unlock(vcpu); 1227 error = thread_check_susp(td, false); 1228 if (error != 0) { 1229 if (vcpu_halted) { 1230 CPU_CLR_ATOMIC(vcpuid, 1231 &vm->halted_cpus); 1232 } 1233 return (error); 1234 } 1235 vcpu_lock(vcpu); 1236 } 1237 } 1238 1239 if (vcpu_halted) 1240 CPU_CLR_ATOMIC(vcpuid, &vm->halted_cpus); 1241 1242 vcpu_unlock(vcpu); 1243 1244 if (vm_halted) 1245 vm_suspend(vm, VM_SUSPEND_HALT); 1246 1247 return (0); 1248 } 1249 1250 static int 1251 vm_handle_paging(struct vcpu *vcpu, bool *retu) 1252 { 1253 struct vm *vm = vcpu->vm; 1254 int rv, ftype; 1255 struct vm_map *map; 1256 struct vm_exit *vme; 1257 1258 vme = &vcpu->exitinfo; 1259 1260 KASSERT(vme->inst_length == 0, ("%s: invalid inst_length %d", 1261 __func__, vme->inst_length)); 1262 1263 ftype = vme->u.paging.fault_type; 1264 KASSERT(ftype == VM_PROT_READ || 1265 ftype == VM_PROT_WRITE || ftype == VM_PROT_EXECUTE, 1266 ("vm_handle_paging: invalid fault_type %d", ftype)); 1267 1268 if (ftype == VM_PROT_READ || ftype == VM_PROT_WRITE) { 1269 rv = pmap_emulate_accessed_dirty(vmspace_pmap(vm_vmspace(vm)), 1270 vme->u.paging.gpa, ftype); 1271 if (rv == 0) { 1272 VMM_CTR2(vcpu, "%s bit emulation for gpa %#lx", 1273 ftype == VM_PROT_READ ? "accessed" : "dirty", 1274 vme->u.paging.gpa); 1275 goto done; 1276 } 1277 } 1278 1279 map = &vm_vmspace(vm)->vm_map; 1280 rv = vm_fault(map, vme->u.paging.gpa, ftype, VM_FAULT_NORMAL, NULL); 1281 1282 VMM_CTR3(vcpu, "vm_handle_paging rv = %d, gpa = %#lx, " 1283 "ftype = %d", rv, vme->u.paging.gpa, ftype); 1284 1285 if (rv != KERN_SUCCESS) 1286 return (EFAULT); 1287 done: 1288 return (0); 1289 } 1290 1291 static int 1292 vm_handle_inst_emul(struct vcpu *vcpu, bool *retu) 1293 { 1294 struct vie *vie; 1295 struct vm_exit *vme; 1296 uint64_t gla, gpa, cs_base; 1297 struct vm_guest_paging *paging; 1298 mem_region_read_t mread; 1299 mem_region_write_t mwrite; 1300 enum vm_cpu_mode cpu_mode; 1301 int cs_d, error, fault; 1302 1303 vme = &vcpu->exitinfo; 1304 1305 KASSERT(vme->inst_length == 0, ("%s: invalid inst_length %d", 1306 __func__, vme->inst_length)); 1307 1308 gla = vme->u.inst_emul.gla; 1309 gpa = vme->u.inst_emul.gpa; 1310 cs_base = vme->u.inst_emul.cs_base; 1311 cs_d = vme->u.inst_emul.cs_d; 1312 vie = &vme->u.inst_emul.vie; 1313 paging = &vme->u.inst_emul.paging; 1314 cpu_mode = paging->cpu_mode; 1315 1316 VMM_CTR1(vcpu, "inst_emul fault accessing gpa %#lx", gpa); 1317 1318 /* Fetch, decode and emulate the faulting instruction */ 1319 if (vie->num_valid == 0) { 1320 error = vmm_fetch_instruction(vcpu, paging, vme->rip + cs_base, 1321 VIE_INST_SIZE, vie, &fault); 1322 } else { 1323 /* 1324 * The instruction bytes have already been copied into 'vie' 1325 */ 1326 error = fault = 0; 1327 } 1328 if (error || fault) 1329 return (error); 1330 1331 if (vmm_decode_instruction(vcpu, gla, cpu_mode, cs_d, vie) != 0) { 1332 VMM_CTR1(vcpu, "Error decoding instruction at %#lx", 1333 vme->rip + cs_base); 1334 *retu = true; /* dump instruction bytes in userspace */ 1335 return (0); 1336 } 1337 1338 /* 1339 * Update 'nextrip' based on the length of the emulated instruction. 1340 */ 1341 vme->inst_length = vie->num_processed; 1342 vcpu->nextrip += vie->num_processed; 1343 VMM_CTR1(vcpu, "nextrip updated to %#lx after instruction decoding", 1344 vcpu->nextrip); 1345 1346 /* return to userland unless this is an in-kernel emulated device */ 1347 if (gpa >= DEFAULT_APIC_BASE && gpa < DEFAULT_APIC_BASE + PAGE_SIZE) { 1348 mread = lapic_mmio_read; 1349 mwrite = lapic_mmio_write; 1350 } else if (gpa >= VIOAPIC_BASE && gpa < VIOAPIC_BASE + VIOAPIC_SIZE) { 1351 mread = vioapic_mmio_read; 1352 mwrite = vioapic_mmio_write; 1353 } else if (gpa >= VHPET_BASE && gpa < VHPET_BASE + VHPET_SIZE) { 1354 mread = vhpet_mmio_read; 1355 mwrite = vhpet_mmio_write; 1356 } else { 1357 *retu = true; 1358 return (0); 1359 } 1360 1361 error = vmm_emulate_instruction(vcpu, gpa, vie, paging, mread, mwrite, 1362 retu); 1363 1364 return (error); 1365 } 1366 1367 static int 1368 vm_handle_suspend(struct vcpu *vcpu, bool *retu) 1369 { 1370 struct vm *vm = vcpu->vm; 1371 int error, i; 1372 struct thread *td; 1373 1374 error = 0; 1375 td = curthread; 1376 1377 CPU_SET_ATOMIC(vcpu->vcpuid, &vm->suspended_cpus); 1378 1379 /* 1380 * Wait until all 'active_cpus' have suspended themselves. 1381 * 1382 * Since a VM may be suspended at any time including when one or 1383 * more vcpus are doing a rendezvous we need to call the rendezvous 1384 * handler while we are waiting to prevent a deadlock. 1385 */ 1386 vcpu_lock(vcpu); 1387 while (error == 0) { 1388 if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) { 1389 VMM_CTR0(vcpu, "All vcpus suspended"); 1390 break; 1391 } 1392 1393 if (vm->rendezvous_func == NULL) { 1394 VMM_CTR0(vcpu, "Sleeping during suspend"); 1395 vcpu_require_state_locked(vcpu, VCPU_SLEEPING); 1396 msleep_spin(vcpu, &vcpu->mtx, "vmsusp", hz); 1397 vcpu_require_state_locked(vcpu, VCPU_FROZEN); 1398 if (td_ast_pending(td, TDA_SUSPEND)) { 1399 vcpu_unlock(vcpu); 1400 error = thread_check_susp(td, false); 1401 vcpu_lock(vcpu); 1402 } 1403 } else { 1404 VMM_CTR0(vcpu, "Rendezvous during suspend"); 1405 vcpu_unlock(vcpu); 1406 error = vm_handle_rendezvous(vcpu); 1407 vcpu_lock(vcpu); 1408 } 1409 } 1410 vcpu_unlock(vcpu); 1411 1412 /* 1413 * Wakeup the other sleeping vcpus and return to userspace. 1414 */ 1415 for (i = 0; i < vm->maxcpus; i++) { 1416 if (CPU_ISSET(i, &vm->suspended_cpus)) { 1417 vcpu_notify_event(vm_vcpu(vm, i)); 1418 } 1419 } 1420 1421 *retu = true; 1422 return (error); 1423 } 1424 1425 static int 1426 vm_handle_reqidle(struct vcpu *vcpu, bool *retu) 1427 { 1428 vcpu_lock(vcpu); 1429 KASSERT(vcpu->reqidle, ("invalid vcpu reqidle %d", vcpu->reqidle)); 1430 vcpu->reqidle = 0; 1431 vcpu_unlock(vcpu); 1432 *retu = true; 1433 return (0); 1434 } 1435 1436 static int 1437 vm_handle_db(struct vcpu *vcpu, struct vm_exit *vme, bool *retu) 1438 { 1439 int error, fault; 1440 uint64_t rsp; 1441 uint64_t rflags; 1442 struct vm_copyinfo copyinfo[2]; 1443 1444 *retu = true; 1445 if (!vme->u.dbg.pushf_intercept || vme->u.dbg.tf_shadow_val != 0) { 1446 return (0); 1447 } 1448 1449 vm_get_register(vcpu, VM_REG_GUEST_RSP, &rsp); 1450 error = vm_copy_setup(vcpu, &vme->u.dbg.paging, rsp, sizeof(uint64_t), 1451 VM_PROT_RW, copyinfo, nitems(copyinfo), &fault); 1452 if (error != 0 || fault != 0) { 1453 *retu = false; 1454 return (EINVAL); 1455 } 1456 1457 /* Read pushed rflags value from top of stack. */ 1458 vm_copyin(copyinfo, &rflags, sizeof(uint64_t)); 1459 1460 /* Clear TF bit. */ 1461 rflags &= ~(PSL_T); 1462 1463 /* Write updated value back to memory. */ 1464 vm_copyout(&rflags, copyinfo, sizeof(uint64_t)); 1465 vm_copy_teardown(copyinfo, nitems(copyinfo)); 1466 1467 return (0); 1468 } 1469 1470 int 1471 vm_suspend(struct vm *vm, enum vm_suspend_how how) 1472 { 1473 int i; 1474 1475 if (how <= VM_SUSPEND_NONE || how >= VM_SUSPEND_LAST) 1476 return (EINVAL); 1477 1478 if (atomic_cmpset_int(&vm->suspend, 0, how) == 0) { 1479 VM_CTR2(vm, "virtual machine already suspended %d/%d", 1480 vm->suspend, how); 1481 return (EALREADY); 1482 } 1483 1484 VM_CTR1(vm, "virtual machine successfully suspended %d", how); 1485 1486 /* 1487 * Notify all active vcpus that they are now suspended. 1488 */ 1489 for (i = 0; i < vm->maxcpus; i++) { 1490 if (CPU_ISSET(i, &vm->active_cpus)) 1491 vcpu_notify_event(vm_vcpu(vm, i)); 1492 } 1493 1494 return (0); 1495 } 1496 1497 void 1498 vm_exit_suspended(struct vcpu *vcpu, uint64_t rip) 1499 { 1500 struct vm *vm = vcpu->vm; 1501 struct vm_exit *vmexit; 1502 1503 KASSERT(vm->suspend > VM_SUSPEND_NONE && vm->suspend < VM_SUSPEND_LAST, 1504 ("vm_exit_suspended: invalid suspend type %d", vm->suspend)); 1505 1506 vmexit = vm_exitinfo(vcpu); 1507 vmexit->rip = rip; 1508 vmexit->inst_length = 0; 1509 vmexit->exitcode = VM_EXITCODE_SUSPENDED; 1510 vmexit->u.suspended.how = vm->suspend; 1511 } 1512 1513 void 1514 vm_exit_debug(struct vcpu *vcpu, uint64_t rip) 1515 { 1516 struct vm_exit *vmexit; 1517 1518 vmexit = vm_exitinfo(vcpu); 1519 vmexit->rip = rip; 1520 vmexit->inst_length = 0; 1521 vmexit->exitcode = VM_EXITCODE_DEBUG; 1522 } 1523 1524 void 1525 vm_exit_rendezvous(struct vcpu *vcpu, uint64_t rip) 1526 { 1527 struct vm_exit *vmexit; 1528 1529 vmexit = vm_exitinfo(vcpu); 1530 vmexit->rip = rip; 1531 vmexit->inst_length = 0; 1532 vmexit->exitcode = VM_EXITCODE_RENDEZVOUS; 1533 vmm_stat_incr(vcpu, VMEXIT_RENDEZVOUS, 1); 1534 } 1535 1536 void 1537 vm_exit_reqidle(struct vcpu *vcpu, uint64_t rip) 1538 { 1539 struct vm_exit *vmexit; 1540 1541 vmexit = vm_exitinfo(vcpu); 1542 vmexit->rip = rip; 1543 vmexit->inst_length = 0; 1544 vmexit->exitcode = VM_EXITCODE_REQIDLE; 1545 vmm_stat_incr(vcpu, VMEXIT_REQIDLE, 1); 1546 } 1547 1548 void 1549 vm_exit_astpending(struct vcpu *vcpu, uint64_t rip) 1550 { 1551 struct vm_exit *vmexit; 1552 1553 vmexit = vm_exitinfo(vcpu); 1554 vmexit->rip = rip; 1555 vmexit->inst_length = 0; 1556 vmexit->exitcode = VM_EXITCODE_BOGUS; 1557 vmm_stat_incr(vcpu, VMEXIT_ASTPENDING, 1); 1558 } 1559 1560 int 1561 vm_run(struct vcpu *vcpu) 1562 { 1563 struct vm *vm = vcpu->vm; 1564 struct vm_eventinfo evinfo; 1565 int error, vcpuid; 1566 struct pcb *pcb; 1567 uint64_t tscval; 1568 struct vm_exit *vme; 1569 bool retu, intr_disabled; 1570 pmap_t pmap; 1571 1572 vcpuid = vcpu->vcpuid; 1573 1574 if (!CPU_ISSET(vcpuid, &vm->active_cpus)) 1575 return (EINVAL); 1576 1577 if (CPU_ISSET(vcpuid, &vm->suspended_cpus)) 1578 return (EINVAL); 1579 1580 pmap = vmspace_pmap(vm_vmspace(vm)); 1581 vme = &vcpu->exitinfo; 1582 evinfo.rptr = &vm->rendezvous_req_cpus; 1583 evinfo.sptr = &vm->suspend; 1584 evinfo.iptr = &vcpu->reqidle; 1585 restart: 1586 critical_enter(); 1587 1588 KASSERT(!CPU_ISSET(curcpu, &pmap->pm_active), 1589 ("vm_run: absurd pm_active")); 1590 1591 tscval = rdtsc(); 1592 1593 pcb = PCPU_GET(curpcb); 1594 set_pcb_flags(pcb, PCB_FULL_IRET); 1595 1596 restore_guest_fpustate(vcpu); 1597 1598 vcpu_require_state(vcpu, VCPU_RUNNING); 1599 error = vmmops_run(vcpu->cookie, vcpu->nextrip, pmap, &evinfo); 1600 vcpu_require_state(vcpu, VCPU_FROZEN); 1601 1602 save_guest_fpustate(vcpu); 1603 1604 vmm_stat_incr(vcpu, VCPU_TOTAL_RUNTIME, rdtsc() - tscval); 1605 1606 critical_exit(); 1607 1608 if (error == 0) { 1609 retu = false; 1610 vcpu->nextrip = vme->rip + vme->inst_length; 1611 switch (vme->exitcode) { 1612 case VM_EXITCODE_REQIDLE: 1613 error = vm_handle_reqidle(vcpu, &retu); 1614 break; 1615 case VM_EXITCODE_SUSPENDED: 1616 error = vm_handle_suspend(vcpu, &retu); 1617 break; 1618 case VM_EXITCODE_IOAPIC_EOI: 1619 vioapic_process_eoi(vm, vme->u.ioapic_eoi.vector); 1620 break; 1621 case VM_EXITCODE_RENDEZVOUS: 1622 error = vm_handle_rendezvous(vcpu); 1623 break; 1624 case VM_EXITCODE_HLT: 1625 intr_disabled = ((vme->u.hlt.rflags & PSL_I) == 0); 1626 error = vm_handle_hlt(vcpu, intr_disabled, &retu); 1627 break; 1628 case VM_EXITCODE_PAGING: 1629 error = vm_handle_paging(vcpu, &retu); 1630 break; 1631 case VM_EXITCODE_INST_EMUL: 1632 error = vm_handle_inst_emul(vcpu, &retu); 1633 break; 1634 case VM_EXITCODE_INOUT: 1635 case VM_EXITCODE_INOUT_STR: 1636 error = vm_handle_inout(vcpu, vme, &retu); 1637 break; 1638 case VM_EXITCODE_DB: 1639 error = vm_handle_db(vcpu, vme, &retu); 1640 break; 1641 case VM_EXITCODE_MONITOR: 1642 case VM_EXITCODE_MWAIT: 1643 case VM_EXITCODE_VMINSN: 1644 vm_inject_ud(vcpu); 1645 break; 1646 default: 1647 retu = true; /* handled in userland */ 1648 break; 1649 } 1650 } 1651 1652 /* 1653 * VM_EXITCODE_INST_EMUL could access the apic which could transform the 1654 * exit code into VM_EXITCODE_IPI. 1655 */ 1656 if (error == 0 && vme->exitcode == VM_EXITCODE_IPI) 1657 error = vm_handle_ipi(vcpu, vme, &retu); 1658 1659 if (error == 0 && retu == false) 1660 goto restart; 1661 1662 vmm_stat_incr(vcpu, VMEXIT_USERSPACE, 1); 1663 VMM_CTR2(vcpu, "retu %d/%d", error, vme->exitcode); 1664 1665 return (error); 1666 } 1667 1668 int 1669 vm_restart_instruction(struct vcpu *vcpu) 1670 { 1671 enum vcpu_state state; 1672 uint64_t rip; 1673 int error __diagused; 1674 1675 state = vcpu_get_state(vcpu, NULL); 1676 if (state == VCPU_RUNNING) { 1677 /* 1678 * When a vcpu is "running" the next instruction is determined 1679 * by adding 'rip' and 'inst_length' in the vcpu's 'exitinfo'. 1680 * Thus setting 'inst_length' to zero will cause the current 1681 * instruction to be restarted. 1682 */ 1683 vcpu->exitinfo.inst_length = 0; 1684 VMM_CTR1(vcpu, "restarting instruction at %#lx by " 1685 "setting inst_length to zero", vcpu->exitinfo.rip); 1686 } else if (state == VCPU_FROZEN) { 1687 /* 1688 * When a vcpu is "frozen" it is outside the critical section 1689 * around vmmops_run() and 'nextrip' points to the next 1690 * instruction. Thus instruction restart is achieved by setting 1691 * 'nextrip' to the vcpu's %rip. 1692 */ 1693 error = vm_get_register(vcpu, VM_REG_GUEST_RIP, &rip); 1694 KASSERT(!error, ("%s: error %d getting rip", __func__, error)); 1695 VMM_CTR2(vcpu, "restarting instruction by updating " 1696 "nextrip from %#lx to %#lx", vcpu->nextrip, rip); 1697 vcpu->nextrip = rip; 1698 } else { 1699 panic("%s: invalid state %d", __func__, state); 1700 } 1701 return (0); 1702 } 1703 1704 int 1705 vm_exit_intinfo(struct vcpu *vcpu, uint64_t info) 1706 { 1707 int type, vector; 1708 1709 if (info & VM_INTINFO_VALID) { 1710 type = info & VM_INTINFO_TYPE; 1711 vector = info & 0xff; 1712 if (type == VM_INTINFO_NMI && vector != IDT_NMI) 1713 return (EINVAL); 1714 if (type == VM_INTINFO_HWEXCEPTION && vector >= 32) 1715 return (EINVAL); 1716 if (info & VM_INTINFO_RSVD) 1717 return (EINVAL); 1718 } else { 1719 info = 0; 1720 } 1721 VMM_CTR2(vcpu, "%s: info1(%#lx)", __func__, info); 1722 vcpu->exitintinfo = info; 1723 return (0); 1724 } 1725 1726 enum exc_class { 1727 EXC_BENIGN, 1728 EXC_CONTRIBUTORY, 1729 EXC_PAGEFAULT 1730 }; 1731 1732 #define IDT_VE 20 /* Virtualization Exception (Intel specific) */ 1733 1734 static enum exc_class 1735 exception_class(uint64_t info) 1736 { 1737 int type, vector; 1738 1739 KASSERT(info & VM_INTINFO_VALID, ("intinfo must be valid: %#lx", info)); 1740 type = info & VM_INTINFO_TYPE; 1741 vector = info & 0xff; 1742 1743 /* Table 6-4, "Interrupt and Exception Classes", Intel SDM, Vol 3 */ 1744 switch (type) { 1745 case VM_INTINFO_HWINTR: 1746 case VM_INTINFO_SWINTR: 1747 case VM_INTINFO_NMI: 1748 return (EXC_BENIGN); 1749 default: 1750 /* 1751 * Hardware exception. 1752 * 1753 * SVM and VT-x use identical type values to represent NMI, 1754 * hardware interrupt and software interrupt. 1755 * 1756 * SVM uses type '3' for all exceptions. VT-x uses type '3' 1757 * for exceptions except #BP and #OF. #BP and #OF use a type 1758 * value of '5' or '6'. Therefore we don't check for explicit 1759 * values of 'type' to classify 'intinfo' into a hardware 1760 * exception. 1761 */ 1762 break; 1763 } 1764 1765 switch (vector) { 1766 case IDT_PF: 1767 case IDT_VE: 1768 return (EXC_PAGEFAULT); 1769 case IDT_DE: 1770 case IDT_TS: 1771 case IDT_NP: 1772 case IDT_SS: 1773 case IDT_GP: 1774 return (EXC_CONTRIBUTORY); 1775 default: 1776 return (EXC_BENIGN); 1777 } 1778 } 1779 1780 static int 1781 nested_fault(struct vcpu *vcpu, uint64_t info1, uint64_t info2, 1782 uint64_t *retinfo) 1783 { 1784 enum exc_class exc1, exc2; 1785 int type1, vector1; 1786 1787 KASSERT(info1 & VM_INTINFO_VALID, ("info1 %#lx is not valid", info1)); 1788 KASSERT(info2 & VM_INTINFO_VALID, ("info2 %#lx is not valid", info2)); 1789 1790 /* 1791 * If an exception occurs while attempting to call the double-fault 1792 * handler the processor enters shutdown mode (aka triple fault). 1793 */ 1794 type1 = info1 & VM_INTINFO_TYPE; 1795 vector1 = info1 & 0xff; 1796 if (type1 == VM_INTINFO_HWEXCEPTION && vector1 == IDT_DF) { 1797 VMM_CTR2(vcpu, "triple fault: info1(%#lx), info2(%#lx)", 1798 info1, info2); 1799 vm_suspend(vcpu->vm, VM_SUSPEND_TRIPLEFAULT); 1800 *retinfo = 0; 1801 return (0); 1802 } 1803 1804 /* 1805 * Table 6-5 "Conditions for Generating a Double Fault", Intel SDM, Vol3 1806 */ 1807 exc1 = exception_class(info1); 1808 exc2 = exception_class(info2); 1809 if ((exc1 == EXC_CONTRIBUTORY && exc2 == EXC_CONTRIBUTORY) || 1810 (exc1 == EXC_PAGEFAULT && exc2 != EXC_BENIGN)) { 1811 /* Convert nested fault into a double fault. */ 1812 *retinfo = IDT_DF; 1813 *retinfo |= VM_INTINFO_VALID | VM_INTINFO_HWEXCEPTION; 1814 *retinfo |= VM_INTINFO_DEL_ERRCODE; 1815 } else { 1816 /* Handle exceptions serially */ 1817 *retinfo = info2; 1818 } 1819 return (1); 1820 } 1821 1822 static uint64_t 1823 vcpu_exception_intinfo(struct vcpu *vcpu) 1824 { 1825 uint64_t info = 0; 1826 1827 if (vcpu->exception_pending) { 1828 info = vcpu->exc_vector & 0xff; 1829 info |= VM_INTINFO_VALID | VM_INTINFO_HWEXCEPTION; 1830 if (vcpu->exc_errcode_valid) { 1831 info |= VM_INTINFO_DEL_ERRCODE; 1832 info |= (uint64_t)vcpu->exc_errcode << 32; 1833 } 1834 } 1835 return (info); 1836 } 1837 1838 int 1839 vm_entry_intinfo(struct vcpu *vcpu, uint64_t *retinfo) 1840 { 1841 uint64_t info1, info2; 1842 int valid; 1843 1844 info1 = vcpu->exitintinfo; 1845 vcpu->exitintinfo = 0; 1846 1847 info2 = 0; 1848 if (vcpu->exception_pending) { 1849 info2 = vcpu_exception_intinfo(vcpu); 1850 vcpu->exception_pending = 0; 1851 VMM_CTR2(vcpu, "Exception %d delivered: %#lx", 1852 vcpu->exc_vector, info2); 1853 } 1854 1855 if ((info1 & VM_INTINFO_VALID) && (info2 & VM_INTINFO_VALID)) { 1856 valid = nested_fault(vcpu, info1, info2, retinfo); 1857 } else if (info1 & VM_INTINFO_VALID) { 1858 *retinfo = info1; 1859 valid = 1; 1860 } else if (info2 & VM_INTINFO_VALID) { 1861 *retinfo = info2; 1862 valid = 1; 1863 } else { 1864 valid = 0; 1865 } 1866 1867 if (valid) { 1868 VMM_CTR4(vcpu, "%s: info1(%#lx), info2(%#lx), " 1869 "retinfo(%#lx)", __func__, info1, info2, *retinfo); 1870 } 1871 1872 return (valid); 1873 } 1874 1875 int 1876 vm_get_intinfo(struct vcpu *vcpu, uint64_t *info1, uint64_t *info2) 1877 { 1878 *info1 = vcpu->exitintinfo; 1879 *info2 = vcpu_exception_intinfo(vcpu); 1880 return (0); 1881 } 1882 1883 int 1884 vm_inject_exception(struct vcpu *vcpu, int vector, int errcode_valid, 1885 uint32_t errcode, int restart_instruction) 1886 { 1887 uint64_t regval; 1888 int error __diagused; 1889 1890 if (vector < 0 || vector >= 32) 1891 return (EINVAL); 1892 1893 /* 1894 * A double fault exception should never be injected directly into 1895 * the guest. It is a derived exception that results from specific 1896 * combinations of nested faults. 1897 */ 1898 if (vector == IDT_DF) 1899 return (EINVAL); 1900 1901 if (vcpu->exception_pending) { 1902 VMM_CTR2(vcpu, "Unable to inject exception %d due to " 1903 "pending exception %d", vector, vcpu->exc_vector); 1904 return (EBUSY); 1905 } 1906 1907 if (errcode_valid) { 1908 /* 1909 * Exceptions don't deliver an error code in real mode. 1910 */ 1911 error = vm_get_register(vcpu, VM_REG_GUEST_CR0, ®val); 1912 KASSERT(!error, ("%s: error %d getting CR0", __func__, error)); 1913 if (!(regval & CR0_PE)) 1914 errcode_valid = 0; 1915 } 1916 1917 /* 1918 * From section 26.6.1 "Interruptibility State" in Intel SDM: 1919 * 1920 * Event blocking by "STI" or "MOV SS" is cleared after guest executes 1921 * one instruction or incurs an exception. 1922 */ 1923 error = vm_set_register(vcpu, VM_REG_GUEST_INTR_SHADOW, 0); 1924 KASSERT(error == 0, ("%s: error %d clearing interrupt shadow", 1925 __func__, error)); 1926 1927 if (restart_instruction) 1928 vm_restart_instruction(vcpu); 1929 1930 vcpu->exception_pending = 1; 1931 vcpu->exc_vector = vector; 1932 vcpu->exc_errcode = errcode; 1933 vcpu->exc_errcode_valid = errcode_valid; 1934 VMM_CTR1(vcpu, "Exception %d pending", vector); 1935 return (0); 1936 } 1937 1938 void 1939 vm_inject_fault(struct vcpu *vcpu, int vector, int errcode_valid, int errcode) 1940 { 1941 int error __diagused, restart_instruction; 1942 1943 restart_instruction = 1; 1944 1945 error = vm_inject_exception(vcpu, vector, errcode_valid, 1946 errcode, restart_instruction); 1947 KASSERT(error == 0, ("vm_inject_exception error %d", error)); 1948 } 1949 1950 void 1951 vm_inject_pf(struct vcpu *vcpu, int error_code, uint64_t cr2) 1952 { 1953 int error __diagused; 1954 1955 VMM_CTR2(vcpu, "Injecting page fault: error_code %#x, cr2 %#lx", 1956 error_code, cr2); 1957 1958 error = vm_set_register(vcpu, VM_REG_GUEST_CR2, cr2); 1959 KASSERT(error == 0, ("vm_set_register(cr2) error %d", error)); 1960 1961 vm_inject_fault(vcpu, IDT_PF, 1, error_code); 1962 } 1963 1964 static VMM_STAT(VCPU_NMI_COUNT, "number of NMIs delivered to vcpu"); 1965 1966 int 1967 vm_inject_nmi(struct vcpu *vcpu) 1968 { 1969 1970 vcpu->nmi_pending = 1; 1971 vcpu_notify_event(vcpu); 1972 return (0); 1973 } 1974 1975 int 1976 vm_nmi_pending(struct vcpu *vcpu) 1977 { 1978 return (vcpu->nmi_pending); 1979 } 1980 1981 void 1982 vm_nmi_clear(struct vcpu *vcpu) 1983 { 1984 if (vcpu->nmi_pending == 0) 1985 panic("vm_nmi_clear: inconsistent nmi_pending state"); 1986 1987 vcpu->nmi_pending = 0; 1988 vmm_stat_incr(vcpu, VCPU_NMI_COUNT, 1); 1989 } 1990 1991 static VMM_STAT(VCPU_EXTINT_COUNT, "number of ExtINTs delivered to vcpu"); 1992 1993 int 1994 vm_inject_extint(struct vcpu *vcpu) 1995 { 1996 1997 vcpu->extint_pending = 1; 1998 vcpu_notify_event(vcpu); 1999 return (0); 2000 } 2001 2002 int 2003 vm_extint_pending(struct vcpu *vcpu) 2004 { 2005 return (vcpu->extint_pending); 2006 } 2007 2008 void 2009 vm_extint_clear(struct vcpu *vcpu) 2010 { 2011 if (vcpu->extint_pending == 0) 2012 panic("vm_extint_clear: inconsistent extint_pending state"); 2013 2014 vcpu->extint_pending = 0; 2015 vmm_stat_incr(vcpu, VCPU_EXTINT_COUNT, 1); 2016 } 2017 2018 int 2019 vm_get_capability(struct vcpu *vcpu, int type, int *retval) 2020 { 2021 if (type < 0 || type >= VM_CAP_MAX) 2022 return (EINVAL); 2023 2024 return (vmmops_getcap(vcpu->cookie, type, retval)); 2025 } 2026 2027 int 2028 vm_set_capability(struct vcpu *vcpu, int type, int val) 2029 { 2030 if (type < 0 || type >= VM_CAP_MAX) 2031 return (EINVAL); 2032 2033 return (vmmops_setcap(vcpu->cookie, type, val)); 2034 } 2035 2036 struct vm * 2037 vcpu_vm(struct vcpu *vcpu) 2038 { 2039 return (vcpu->vm); 2040 } 2041 2042 int 2043 vcpu_vcpuid(struct vcpu *vcpu) 2044 { 2045 return (vcpu->vcpuid); 2046 } 2047 2048 struct vcpu * 2049 vm_vcpu(struct vm *vm, int vcpuid) 2050 { 2051 return (vm->vcpu[vcpuid]); 2052 } 2053 2054 struct vlapic * 2055 vm_lapic(struct vcpu *vcpu) 2056 { 2057 return (vcpu->vlapic); 2058 } 2059 2060 struct vioapic * 2061 vm_ioapic(struct vm *vm) 2062 { 2063 2064 return (vm->vioapic); 2065 } 2066 2067 struct vhpet * 2068 vm_hpet(struct vm *vm) 2069 { 2070 2071 return (vm->vhpet); 2072 } 2073 2074 bool 2075 vmm_is_pptdev(int bus, int slot, int func) 2076 { 2077 int b, f, i, n, s; 2078 char *val, *cp, *cp2; 2079 bool found; 2080 2081 /* 2082 * XXX 2083 * The length of an environment variable is limited to 128 bytes which 2084 * puts an upper limit on the number of passthru devices that may be 2085 * specified using a single environment variable. 2086 * 2087 * Work around this by scanning multiple environment variable 2088 * names instead of a single one - yuck! 2089 */ 2090 const char *names[] = { "pptdevs", "pptdevs2", "pptdevs3", NULL }; 2091 2092 /* set pptdevs="1/2/3 4/5/6 7/8/9 10/11/12" */ 2093 found = false; 2094 for (i = 0; names[i] != NULL && !found; i++) { 2095 cp = val = kern_getenv(names[i]); 2096 while (cp != NULL && *cp != '\0') { 2097 if ((cp2 = strchr(cp, ' ')) != NULL) 2098 *cp2 = '\0'; 2099 2100 n = sscanf(cp, "%d/%d/%d", &b, &s, &f); 2101 if (n == 3 && bus == b && slot == s && func == f) { 2102 found = true; 2103 break; 2104 } 2105 2106 if (cp2 != NULL) 2107 *cp2++ = ' '; 2108 2109 cp = cp2; 2110 } 2111 freeenv(val); 2112 } 2113 return (found); 2114 } 2115 2116 void * 2117 vm_iommu_domain(struct vm *vm) 2118 { 2119 2120 return (vm->iommu); 2121 } 2122 2123 int 2124 vcpu_set_state(struct vcpu *vcpu, enum vcpu_state newstate, bool from_idle) 2125 { 2126 int error; 2127 2128 vcpu_lock(vcpu); 2129 error = vcpu_set_state_locked(vcpu, newstate, from_idle); 2130 vcpu_unlock(vcpu); 2131 2132 return (error); 2133 } 2134 2135 enum vcpu_state 2136 vcpu_get_state(struct vcpu *vcpu, int *hostcpu) 2137 { 2138 enum vcpu_state state; 2139 2140 vcpu_lock(vcpu); 2141 state = vcpu->state; 2142 if (hostcpu != NULL) 2143 *hostcpu = vcpu->hostcpu; 2144 vcpu_unlock(vcpu); 2145 2146 return (state); 2147 } 2148 2149 int 2150 vm_activate_cpu(struct vcpu *vcpu) 2151 { 2152 struct vm *vm = vcpu->vm; 2153 2154 if (CPU_ISSET(vcpu->vcpuid, &vm->active_cpus)) 2155 return (EBUSY); 2156 2157 VMM_CTR0(vcpu, "activated"); 2158 CPU_SET_ATOMIC(vcpu->vcpuid, &vm->active_cpus); 2159 return (0); 2160 } 2161 2162 int 2163 vm_suspend_cpu(struct vm *vm, struct vcpu *vcpu) 2164 { 2165 if (vcpu == NULL) { 2166 vm->debug_cpus = vm->active_cpus; 2167 for (int i = 0; i < vm->maxcpus; i++) { 2168 if (CPU_ISSET(i, &vm->active_cpus)) 2169 vcpu_notify_event(vm_vcpu(vm, i)); 2170 } 2171 } else { 2172 if (!CPU_ISSET(vcpu->vcpuid, &vm->active_cpus)) 2173 return (EINVAL); 2174 2175 CPU_SET_ATOMIC(vcpu->vcpuid, &vm->debug_cpus); 2176 vcpu_notify_event(vcpu); 2177 } 2178 return (0); 2179 } 2180 2181 int 2182 vm_resume_cpu(struct vm *vm, struct vcpu *vcpu) 2183 { 2184 2185 if (vcpu == NULL) { 2186 CPU_ZERO(&vm->debug_cpus); 2187 } else { 2188 if (!CPU_ISSET(vcpu->vcpuid, &vm->debug_cpus)) 2189 return (EINVAL); 2190 2191 CPU_CLR_ATOMIC(vcpu->vcpuid, &vm->debug_cpus); 2192 } 2193 return (0); 2194 } 2195 2196 int 2197 vcpu_debugged(struct vcpu *vcpu) 2198 { 2199 2200 return (CPU_ISSET(vcpu->vcpuid, &vcpu->vm->debug_cpus)); 2201 } 2202 2203 cpuset_t 2204 vm_active_cpus(struct vm *vm) 2205 { 2206 2207 return (vm->active_cpus); 2208 } 2209 2210 cpuset_t 2211 vm_debug_cpus(struct vm *vm) 2212 { 2213 2214 return (vm->debug_cpus); 2215 } 2216 2217 cpuset_t 2218 vm_suspended_cpus(struct vm *vm) 2219 { 2220 2221 return (vm->suspended_cpus); 2222 } 2223 2224 /* 2225 * Returns the subset of vCPUs in tostart that are awaiting startup. 2226 * These vCPUs are also marked as no longer awaiting startup. 2227 */ 2228 cpuset_t 2229 vm_start_cpus(struct vm *vm, const cpuset_t *tostart) 2230 { 2231 cpuset_t set; 2232 2233 mtx_lock(&vm->rendezvous_mtx); 2234 CPU_AND(&set, &vm->startup_cpus, tostart); 2235 CPU_ANDNOT(&vm->startup_cpus, &vm->startup_cpus, &set); 2236 mtx_unlock(&vm->rendezvous_mtx); 2237 return (set); 2238 } 2239 2240 void 2241 vm_await_start(struct vm *vm, const cpuset_t *waiting) 2242 { 2243 mtx_lock(&vm->rendezvous_mtx); 2244 CPU_OR(&vm->startup_cpus, &vm->startup_cpus, waiting); 2245 mtx_unlock(&vm->rendezvous_mtx); 2246 } 2247 2248 void * 2249 vcpu_stats(struct vcpu *vcpu) 2250 { 2251 2252 return (vcpu->stats); 2253 } 2254 2255 int 2256 vm_get_x2apic_state(struct vcpu *vcpu, enum x2apic_state *state) 2257 { 2258 *state = vcpu->x2apic_state; 2259 2260 return (0); 2261 } 2262 2263 int 2264 vm_set_x2apic_state(struct vcpu *vcpu, enum x2apic_state state) 2265 { 2266 if (state >= X2APIC_STATE_LAST) 2267 return (EINVAL); 2268 2269 vcpu->x2apic_state = state; 2270 2271 vlapic_set_x2apic_state(vcpu, state); 2272 2273 return (0); 2274 } 2275 2276 /* 2277 * This function is called to ensure that a vcpu "sees" a pending event 2278 * as soon as possible: 2279 * - If the vcpu thread is sleeping then it is woken up. 2280 * - If the vcpu is running on a different host_cpu then an IPI will be directed 2281 * to the host_cpu to cause the vcpu to trap into the hypervisor. 2282 */ 2283 static void 2284 vcpu_notify_event_locked(struct vcpu *vcpu) 2285 { 2286 int hostcpu; 2287 2288 hostcpu = vcpu->hostcpu; 2289 if (vcpu->state == VCPU_RUNNING) { 2290 KASSERT(hostcpu != NOCPU, ("vcpu running on invalid hostcpu")); 2291 if (hostcpu != curcpu) { 2292 ipi_cpu(hostcpu, vmm_ipinum); 2293 } else { 2294 /* 2295 * If the 'vcpu' is running on 'curcpu' then it must 2296 * be sending a notification to itself (e.g. SELF_IPI). 2297 * The pending event will be picked up when the vcpu 2298 * transitions back to guest context. 2299 */ 2300 } 2301 } else { 2302 KASSERT(hostcpu == NOCPU, ("vcpu state %d not consistent " 2303 "with hostcpu %d", vcpu->state, hostcpu)); 2304 if (vcpu->state == VCPU_SLEEPING) 2305 wakeup_one(vcpu); 2306 } 2307 } 2308 2309 void 2310 vcpu_notify_event(struct vcpu *vcpu) 2311 { 2312 vcpu_lock(vcpu); 2313 vcpu_notify_event_locked(vcpu); 2314 vcpu_unlock(vcpu); 2315 } 2316 2317 void 2318 vcpu_notify_lapic(struct vcpu *vcpu) 2319 { 2320 vcpu_lock(vcpu); 2321 if (vcpu->state == VCPU_RUNNING && vcpu->hostcpu != curcpu) 2322 vlapic_post_intr(vcpu->vlapic, vcpu->hostcpu, vmm_ipinum); 2323 else 2324 vcpu_notify_event_locked(vcpu); 2325 vcpu_unlock(vcpu); 2326 } 2327 2328 struct vm_mem * 2329 vm_mem(struct vm *vm) 2330 { 2331 return (&vm->mem); 2332 } 2333 2334 int 2335 vm_apicid2vcpuid(struct vm *vm, int apicid) 2336 { 2337 /* 2338 * XXX apic id is assumed to be numerically identical to vcpu id 2339 */ 2340 return (apicid); 2341 } 2342 2343 int 2344 vm_smp_rendezvous(struct vcpu *vcpu, cpuset_t dest, 2345 vm_rendezvous_func_t func, void *arg) 2346 { 2347 struct vm *vm = vcpu->vm; 2348 int error, i; 2349 2350 /* 2351 * Enforce that this function is called without any locks 2352 */ 2353 WITNESS_WARN(WARN_PANIC, NULL, "vm_smp_rendezvous"); 2354 2355 restart: 2356 mtx_lock(&vm->rendezvous_mtx); 2357 if (vm->rendezvous_func != NULL) { 2358 /* 2359 * If a rendezvous is already in progress then we need to 2360 * call the rendezvous handler in case this 'vcpu' is one 2361 * of the targets of the rendezvous. 2362 */ 2363 VMM_CTR0(vcpu, "Rendezvous already in progress"); 2364 mtx_unlock(&vm->rendezvous_mtx); 2365 error = vm_handle_rendezvous(vcpu); 2366 if (error != 0) 2367 return (error); 2368 goto restart; 2369 } 2370 KASSERT(vm->rendezvous_func == NULL, ("vm_smp_rendezvous: previous " 2371 "rendezvous is still in progress")); 2372 2373 VMM_CTR0(vcpu, "Initiating rendezvous"); 2374 vm->rendezvous_req_cpus = dest; 2375 CPU_ZERO(&vm->rendezvous_done_cpus); 2376 vm->rendezvous_arg = arg; 2377 vm->rendezvous_func = func; 2378 mtx_unlock(&vm->rendezvous_mtx); 2379 2380 /* 2381 * Wake up any sleeping vcpus and trigger a VM-exit in any running 2382 * vcpus so they handle the rendezvous as soon as possible. 2383 */ 2384 for (i = 0; i < vm->maxcpus; i++) { 2385 if (CPU_ISSET(i, &dest)) 2386 vcpu_notify_event(vm_vcpu(vm, i)); 2387 } 2388 2389 return (vm_handle_rendezvous(vcpu)); 2390 } 2391 2392 struct vatpic * 2393 vm_atpic(struct vm *vm) 2394 { 2395 return (vm->vatpic); 2396 } 2397 2398 struct vatpit * 2399 vm_atpit(struct vm *vm) 2400 { 2401 return (vm->vatpit); 2402 } 2403 2404 struct vpmtmr * 2405 vm_pmtmr(struct vm *vm) 2406 { 2407 2408 return (vm->vpmtmr); 2409 } 2410 2411 struct vrtc * 2412 vm_rtc(struct vm *vm) 2413 { 2414 2415 return (vm->vrtc); 2416 } 2417 2418 enum vm_reg_name 2419 vm_segment_name(int seg) 2420 { 2421 static enum vm_reg_name seg_names[] = { 2422 VM_REG_GUEST_ES, 2423 VM_REG_GUEST_CS, 2424 VM_REG_GUEST_SS, 2425 VM_REG_GUEST_DS, 2426 VM_REG_GUEST_FS, 2427 VM_REG_GUEST_GS 2428 }; 2429 2430 KASSERT(seg >= 0 && seg < nitems(seg_names), 2431 ("%s: invalid segment encoding %d", __func__, seg)); 2432 return (seg_names[seg]); 2433 } 2434 2435 void 2436 vm_copy_teardown(struct vm_copyinfo *copyinfo, int num_copyinfo) 2437 { 2438 int idx; 2439 2440 for (idx = 0; idx < num_copyinfo; idx++) { 2441 if (copyinfo[idx].cookie != NULL) 2442 vm_gpa_release(copyinfo[idx].cookie); 2443 } 2444 bzero(copyinfo, num_copyinfo * sizeof(struct vm_copyinfo)); 2445 } 2446 2447 int 2448 vm_copy_setup(struct vcpu *vcpu, struct vm_guest_paging *paging, 2449 uint64_t gla, size_t len, int prot, struct vm_copyinfo *copyinfo, 2450 int num_copyinfo, int *fault) 2451 { 2452 int error, idx, nused; 2453 size_t n, off, remaining; 2454 void *hva, *cookie; 2455 uint64_t gpa; 2456 2457 bzero(copyinfo, sizeof(struct vm_copyinfo) * num_copyinfo); 2458 2459 nused = 0; 2460 remaining = len; 2461 while (remaining > 0) { 2462 if (nused >= num_copyinfo) 2463 return (EFAULT); 2464 error = vm_gla2gpa(vcpu, paging, gla, prot, &gpa, fault); 2465 if (error || *fault) 2466 return (error); 2467 off = gpa & PAGE_MASK; 2468 n = min(remaining, PAGE_SIZE - off); 2469 copyinfo[nused].gpa = gpa; 2470 copyinfo[nused].len = n; 2471 remaining -= n; 2472 gla += n; 2473 nused++; 2474 } 2475 2476 for (idx = 0; idx < nused; idx++) { 2477 hva = vm_gpa_hold(vcpu, copyinfo[idx].gpa, 2478 copyinfo[idx].len, prot, &cookie); 2479 if (hva == NULL) 2480 break; 2481 copyinfo[idx].hva = hva; 2482 copyinfo[idx].cookie = cookie; 2483 } 2484 2485 if (idx != nused) { 2486 vm_copy_teardown(copyinfo, num_copyinfo); 2487 return (EFAULT); 2488 } else { 2489 *fault = 0; 2490 return (0); 2491 } 2492 } 2493 2494 void 2495 vm_copyin(struct vm_copyinfo *copyinfo, void *kaddr, size_t len) 2496 { 2497 char *dst; 2498 int idx; 2499 2500 dst = kaddr; 2501 idx = 0; 2502 while (len > 0) { 2503 bcopy(copyinfo[idx].hva, dst, copyinfo[idx].len); 2504 len -= copyinfo[idx].len; 2505 dst += copyinfo[idx].len; 2506 idx++; 2507 } 2508 } 2509 2510 void 2511 vm_copyout(const void *kaddr, struct vm_copyinfo *copyinfo, size_t len) 2512 { 2513 const char *src; 2514 int idx; 2515 2516 src = kaddr; 2517 idx = 0; 2518 while (len > 0) { 2519 bcopy(src, copyinfo[idx].hva, copyinfo[idx].len); 2520 len -= copyinfo[idx].len; 2521 src += copyinfo[idx].len; 2522 idx++; 2523 } 2524 } 2525 2526 /* 2527 * Return the amount of in-use and wired memory for the VM. Since 2528 * these are global stats, only return the values with for vCPU 0 2529 */ 2530 VMM_STAT_DECLARE(VMM_MEM_RESIDENT); 2531 VMM_STAT_DECLARE(VMM_MEM_WIRED); 2532 2533 static void 2534 vm_get_rescnt(struct vcpu *vcpu, struct vmm_stat_type *stat) 2535 { 2536 2537 if (vcpu->vcpuid == 0) { 2538 vmm_stat_set(vcpu, VMM_MEM_RESIDENT, PAGE_SIZE * 2539 vmspace_resident_count(vm_vmspace(vcpu->vm))); 2540 } 2541 } 2542 2543 static void 2544 vm_get_wiredcnt(struct vcpu *vcpu, struct vmm_stat_type *stat) 2545 { 2546 2547 if (vcpu->vcpuid == 0) { 2548 vmm_stat_set(vcpu, VMM_MEM_WIRED, PAGE_SIZE * 2549 pmap_wired_count(vmspace_pmap(vm_vmspace(vcpu->vm)))); 2550 } 2551 } 2552 2553 VMM_STAT_FUNC(VMM_MEM_RESIDENT, "Resident memory", vm_get_rescnt); 2554 VMM_STAT_FUNC(VMM_MEM_WIRED, "Wired memory", vm_get_wiredcnt); 2555 2556 #ifdef BHYVE_SNAPSHOT 2557 static int 2558 vm_snapshot_vcpus(struct vm *vm, struct vm_snapshot_meta *meta) 2559 { 2560 uint64_t tsc, now; 2561 int ret; 2562 struct vcpu *vcpu; 2563 uint16_t i, maxcpus; 2564 2565 now = rdtsc(); 2566 maxcpus = vm_get_maxcpus(vm); 2567 for (i = 0; i < maxcpus; i++) { 2568 vcpu = vm->vcpu[i]; 2569 if (vcpu == NULL) 2570 continue; 2571 2572 SNAPSHOT_VAR_OR_LEAVE(vcpu->x2apic_state, meta, ret, done); 2573 SNAPSHOT_VAR_OR_LEAVE(vcpu->exitintinfo, meta, ret, done); 2574 SNAPSHOT_VAR_OR_LEAVE(vcpu->exc_vector, meta, ret, done); 2575 SNAPSHOT_VAR_OR_LEAVE(vcpu->exc_errcode_valid, meta, ret, done); 2576 SNAPSHOT_VAR_OR_LEAVE(vcpu->exc_errcode, meta, ret, done); 2577 SNAPSHOT_VAR_OR_LEAVE(vcpu->guest_xcr0, meta, ret, done); 2578 SNAPSHOT_VAR_OR_LEAVE(vcpu->exitinfo, meta, ret, done); 2579 SNAPSHOT_VAR_OR_LEAVE(vcpu->nextrip, meta, ret, done); 2580 2581 /* 2582 * Save the absolute TSC value by adding now to tsc_offset. 2583 * 2584 * It will be turned turned back into an actual offset when the 2585 * TSC restore function is called 2586 */ 2587 tsc = now + vcpu->tsc_offset; 2588 SNAPSHOT_VAR_OR_LEAVE(tsc, meta, ret, done); 2589 if (meta->op == VM_SNAPSHOT_RESTORE) 2590 vcpu->tsc_offset = tsc; 2591 } 2592 2593 done: 2594 return (ret); 2595 } 2596 2597 static int 2598 vm_snapshot_vm(struct vm *vm, struct vm_snapshot_meta *meta) 2599 { 2600 int ret; 2601 2602 ret = vm_snapshot_vcpus(vm, meta); 2603 if (ret != 0) 2604 goto done; 2605 2606 SNAPSHOT_VAR_OR_LEAVE(vm->startup_cpus, meta, ret, done); 2607 done: 2608 return (ret); 2609 } 2610 2611 static int 2612 vm_snapshot_vcpu(struct vm *vm, struct vm_snapshot_meta *meta) 2613 { 2614 int error; 2615 struct vcpu *vcpu; 2616 uint16_t i, maxcpus; 2617 2618 error = 0; 2619 2620 maxcpus = vm_get_maxcpus(vm); 2621 for (i = 0; i < maxcpus; i++) { 2622 vcpu = vm->vcpu[i]; 2623 if (vcpu == NULL) 2624 continue; 2625 2626 error = vmmops_vcpu_snapshot(vcpu->cookie, meta); 2627 if (error != 0) { 2628 printf("%s: failed to snapshot vmcs/vmcb data for " 2629 "vCPU: %d; error: %d\n", __func__, i, error); 2630 goto done; 2631 } 2632 } 2633 2634 done: 2635 return (error); 2636 } 2637 2638 /* 2639 * Save kernel-side structures to user-space for snapshotting. 2640 */ 2641 int 2642 vm_snapshot_req(struct vm *vm, struct vm_snapshot_meta *meta) 2643 { 2644 int ret = 0; 2645 2646 switch (meta->dev_req) { 2647 case STRUCT_VMCX: 2648 ret = vm_snapshot_vcpu(vm, meta); 2649 break; 2650 case STRUCT_VM: 2651 ret = vm_snapshot_vm(vm, meta); 2652 break; 2653 case STRUCT_VIOAPIC: 2654 ret = vioapic_snapshot(vm_ioapic(vm), meta); 2655 break; 2656 case STRUCT_VLAPIC: 2657 ret = vlapic_snapshot(vm, meta); 2658 break; 2659 case STRUCT_VHPET: 2660 ret = vhpet_snapshot(vm_hpet(vm), meta); 2661 break; 2662 case STRUCT_VATPIC: 2663 ret = vatpic_snapshot(vm_atpic(vm), meta); 2664 break; 2665 case STRUCT_VATPIT: 2666 ret = vatpit_snapshot(vm_atpit(vm), meta); 2667 break; 2668 case STRUCT_VPMTMR: 2669 ret = vpmtmr_snapshot(vm_pmtmr(vm), meta); 2670 break; 2671 case STRUCT_VRTC: 2672 ret = vrtc_snapshot(vm_rtc(vm), meta); 2673 break; 2674 default: 2675 printf("%s: failed to find the requested type %#x\n", 2676 __func__, meta->dev_req); 2677 ret = (EINVAL); 2678 } 2679 return (ret); 2680 } 2681 2682 void 2683 vm_set_tsc_offset(struct vcpu *vcpu, uint64_t offset) 2684 { 2685 vcpu->tsc_offset = offset; 2686 } 2687 2688 int 2689 vm_restore_time(struct vm *vm) 2690 { 2691 int error; 2692 uint64_t now; 2693 struct vcpu *vcpu; 2694 uint16_t i, maxcpus; 2695 2696 now = rdtsc(); 2697 2698 error = vhpet_restore_time(vm_hpet(vm)); 2699 if (error) 2700 return (error); 2701 2702 maxcpus = vm_get_maxcpus(vm); 2703 for (i = 0; i < maxcpus; i++) { 2704 vcpu = vm->vcpu[i]; 2705 if (vcpu == NULL) 2706 continue; 2707 2708 error = vmmops_restore_tsc(vcpu->cookie, 2709 vcpu->tsc_offset - now); 2710 if (error) 2711 return (error); 2712 } 2713 2714 return (0); 2715 } 2716 #endif 2717