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