1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause 3 * 4 * Copyright (c) 2011 NetApp, Inc. 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 16 * THIS SOFTWARE IS PROVIDED BY NETAPP, INC ``AS IS'' AND 17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 19 * ARE DISCLAIMED. IN NO EVENT SHALL NETAPP, INC OR CONTRIBUTORS BE LIABLE 20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 26 * SUCH DAMAGE. 27 */ 28 29 #include "opt_bhyve_snapshot.h" 30 31 #include <sys/param.h> 32 #include <sys/systm.h> 33 #include <sys/kernel.h> 34 #include <sys/module.h> 35 #include <sys/sysctl.h> 36 #include <sys/malloc.h> 37 #include <sys/pcpu.h> 38 #include <sys/lock.h> 39 #include <sys/mutex.h> 40 #include <sys/proc.h> 41 #include <sys/rwlock.h> 42 #include <sys/sched.h> 43 #include <sys/smp.h> 44 #include <sys/sx.h> 45 #include <sys/vnode.h> 46 47 #include <vm/vm.h> 48 #include <vm/vm_param.h> 49 #include <vm/vm_extern.h> 50 #include <vm/vm_object.h> 51 #include <vm/vm_page.h> 52 #include <vm/pmap.h> 53 #include <vm/vm_map.h> 54 #include <vm/vm_pager.h> 55 #include <vm/vm_kern.h> 56 #include <vm/vnode_pager.h> 57 #include <vm/swap_pager.h> 58 #include <vm/uma.h> 59 60 #include <machine/cpu.h> 61 #include <machine/pcb.h> 62 #include <machine/smp.h> 63 #include <machine/md_var.h> 64 #include <x86/psl.h> 65 #include <x86/apicreg.h> 66 #include <x86/ifunc.h> 67 68 #include <machine/vmm.h> 69 #include <machine/vmm_instruction_emul.h> 70 #include <machine/vmm_snapshot.h> 71 72 #include <dev/vmm/vmm_dev.h> 73 #include <dev/vmm/vmm_ktr.h> 74 #include <dev/vmm/vmm_mem.h> 75 76 #include "vmm_ioport.h" 77 #include "vmm_host.h" 78 #include "vmm_mem.h" 79 #include "vmm_util.h" 80 #include "vatpic.h" 81 #include "vatpit.h" 82 #include "vhpet.h" 83 #include "vioapic.h" 84 #include "vlapic.h" 85 #include "vpmtmr.h" 86 #include "vrtc.h" 87 #include "vmm_stat.h" 88 #include "vmm_lapic.h" 89 90 #include "io/ppt.h" 91 #include "io/iommu.h" 92 93 struct vlapic; 94 95 /* 96 * Initialization: 97 * (a) allocated when vcpu is created 98 * (i) initialized when vcpu is created and when it is reinitialized 99 * (o) initialized the first time the vcpu is created 100 * (x) initialized before use 101 */ 102 struct vcpu { 103 struct mtx mtx; /* (o) protects 'state' and 'hostcpu' */ 104 enum vcpu_state state; /* (o) vcpu state */ 105 int vcpuid; /* (o) */ 106 int hostcpu; /* (o) vcpu's host cpu */ 107 int reqidle; /* (i) request vcpu to idle */ 108 struct vm *vm; /* (o) */ 109 void *cookie; /* (i) cpu-specific data */ 110 struct vlapic *vlapic; /* (i) APIC device model */ 111 enum x2apic_state x2apic_state; /* (i) APIC mode */ 112 uint64_t exitintinfo; /* (i) events pending at VM exit */ 113 int nmi_pending; /* (i) NMI pending */ 114 int extint_pending; /* (i) INTR pending */ 115 int exception_pending; /* (i) exception pending */ 116 int exc_vector; /* (x) exception collateral */ 117 int exc_errcode_valid; 118 uint32_t exc_errcode; 119 struct savefpu *guestfpu; /* (a,i) guest fpu state */ 120 uint64_t guest_xcr0; /* (i) guest %xcr0 register */ 121 void *stats; /* (a,i) statistics */ 122 struct vm_exit exitinfo; /* (x) exit reason and collateral */ 123 cpuset_t exitinfo_cpuset; /* (x) storage for vmexit handlers */ 124 uint64_t nextrip; /* (x) next instruction to execute */ 125 uint64_t tsc_offset; /* (o) TSC offsetting */ 126 }; 127 128 #define vcpu_lock_init(v) mtx_init(&((v)->mtx), "vcpu lock", 0, MTX_SPIN) 129 #define vcpu_lock_destroy(v) mtx_destroy(&((v)->mtx)) 130 #define vcpu_lock(v) mtx_lock_spin(&((v)->mtx)) 131 #define vcpu_unlock(v) mtx_unlock_spin(&((v)->mtx)) 132 #define vcpu_assert_locked(v) mtx_assert(&((v)->mtx), MA_OWNED) 133 134 /* 135 * Initialization: 136 * (o) initialized the first time the VM is created 137 * (i) initialized when VM is created and when it is reinitialized 138 * (x) initialized before use 139 * 140 * Locking: 141 * [m] mem_segs_lock 142 * [r] rendezvous_mtx 143 * [v] reads require one frozen vcpu, writes require freezing all vcpus 144 */ 145 struct vm { 146 void *cookie; /* (i) cpu-specific data */ 147 void *iommu; /* (x) iommu-specific data */ 148 struct vhpet *vhpet; /* (i) virtual HPET */ 149 struct vioapic *vioapic; /* (i) virtual ioapic */ 150 struct vatpic *vatpic; /* (i) virtual atpic */ 151 struct vatpit *vatpit; /* (i) virtual atpit */ 152 struct vpmtmr *vpmtmr; /* (i) virtual ACPI PM timer */ 153 struct vrtc *vrtc; /* (o) virtual RTC */ 154 volatile cpuset_t active_cpus; /* (i) active vcpus */ 155 volatile cpuset_t debug_cpus; /* (i) vcpus stopped for debug */ 156 cpuset_t startup_cpus; /* (i) [r] waiting for startup */ 157 int suspend; /* (i) stop VM execution */ 158 bool dying; /* (o) is dying */ 159 volatile cpuset_t suspended_cpus; /* (i) suspended vcpus */ 160 volatile cpuset_t halted_cpus; /* (x) cpus in a hard halt */ 161 cpuset_t rendezvous_req_cpus; /* (x) [r] rendezvous requested */ 162 cpuset_t rendezvous_done_cpus; /* (x) [r] rendezvous finished */ 163 void *rendezvous_arg; /* (x) [r] rendezvous func/arg */ 164 vm_rendezvous_func_t rendezvous_func; 165 struct mtx rendezvous_mtx; /* (o) rendezvous lock */ 166 struct vm_mem mem; /* (i) [m+v] guest memory */ 167 char name[VM_MAX_NAMELEN+1]; /* (o) virtual machine name */ 168 struct vcpu **vcpu; /* (o) guest vcpus */ 169 /* The following describe the vm cpu topology */ 170 uint16_t sockets; /* (o) num of sockets */ 171 uint16_t cores; /* (o) num of cores/socket */ 172 uint16_t threads; /* (o) num of threads/core */ 173 uint16_t maxcpus; /* (o) max pluggable cpus */ 174 struct sx vcpus_init_lock; /* (o) */ 175 }; 176 177 #define VMM_CTR0(vcpu, format) \ 178 VCPU_CTR0((vcpu)->vm, (vcpu)->vcpuid, format) 179 180 #define VMM_CTR1(vcpu, format, p1) \ 181 VCPU_CTR1((vcpu)->vm, (vcpu)->vcpuid, format, p1) 182 183 #define VMM_CTR2(vcpu, format, p1, p2) \ 184 VCPU_CTR2((vcpu)->vm, (vcpu)->vcpuid, format, p1, p2) 185 186 #define VMM_CTR3(vcpu, format, p1, p2, p3) \ 187 VCPU_CTR3((vcpu)->vm, (vcpu)->vcpuid, format, p1, p2, p3) 188 189 #define VMM_CTR4(vcpu, format, p1, p2, p3, p4) \ 190 VCPU_CTR4((vcpu)->vm, (vcpu)->vcpuid, format, p1, p2, p3, p4) 191 192 static int vmm_initialized; 193 194 static void vmmops_panic(void); 195 196 static void 197 vmmops_panic(void) 198 { 199 panic("vmm_ops func called when !vmm_is_intel() && !vmm_is_svm()"); 200 } 201 202 #define DEFINE_VMMOPS_IFUNC(ret_type, opname, args) \ 203 DEFINE_IFUNC(, ret_type, vmmops_##opname, args) \ 204 { \ 205 if (vmm_is_intel()) \ 206 return (vmm_ops_intel.opname); \ 207 else if (vmm_is_svm()) \ 208 return (vmm_ops_amd.opname); \ 209 else \ 210 return ((ret_type (*)args)vmmops_panic); \ 211 } 212 213 DEFINE_VMMOPS_IFUNC(int, modinit, (int ipinum)) 214 DEFINE_VMMOPS_IFUNC(int, modcleanup, (void)) 215 DEFINE_VMMOPS_IFUNC(void, modsuspend, (void)) 216 DEFINE_VMMOPS_IFUNC(void, modresume, (void)) 217 DEFINE_VMMOPS_IFUNC(void *, init, (struct vm *vm, struct pmap *pmap)) 218 DEFINE_VMMOPS_IFUNC(int, run, (void *vcpui, register_t rip, struct pmap *pmap, 219 struct vm_eventinfo *info)) 220 DEFINE_VMMOPS_IFUNC(void, cleanup, (void *vmi)) 221 DEFINE_VMMOPS_IFUNC(void *, vcpu_init, (void *vmi, struct vcpu *vcpu, 222 int vcpu_id)) 223 DEFINE_VMMOPS_IFUNC(void, vcpu_cleanup, (void *vcpui)) 224 DEFINE_VMMOPS_IFUNC(int, getreg, (void *vcpui, int num, uint64_t *retval)) 225 DEFINE_VMMOPS_IFUNC(int, setreg, (void *vcpui, int num, uint64_t val)) 226 DEFINE_VMMOPS_IFUNC(int, getdesc, (void *vcpui, int num, struct seg_desc *desc)) 227 DEFINE_VMMOPS_IFUNC(int, setdesc, (void *vcpui, int num, struct seg_desc *desc)) 228 DEFINE_VMMOPS_IFUNC(int, getcap, (void *vcpui, int num, int *retval)) 229 DEFINE_VMMOPS_IFUNC(int, setcap, (void *vcpui, int num, int val)) 230 DEFINE_VMMOPS_IFUNC(struct vmspace *, vmspace_alloc, (vm_offset_t min, 231 vm_offset_t max)) 232 DEFINE_VMMOPS_IFUNC(void, vmspace_free, (struct vmspace *vmspace)) 233 DEFINE_VMMOPS_IFUNC(struct vlapic *, vlapic_init, (void *vcpui)) 234 DEFINE_VMMOPS_IFUNC(void, vlapic_cleanup, (struct vlapic *vlapic)) 235 #ifdef BHYVE_SNAPSHOT 236 DEFINE_VMMOPS_IFUNC(int, vcpu_snapshot, (void *vcpui, 237 struct vm_snapshot_meta *meta)) 238 DEFINE_VMMOPS_IFUNC(int, restore_tsc, (void *vcpui, uint64_t now)) 239 #endif 240 241 SDT_PROVIDER_DEFINE(vmm); 242 243 static MALLOC_DEFINE(M_VM, "vm", "vm"); 244 245 /* statistics */ 246 static VMM_STAT(VCPU_TOTAL_RUNTIME, "vcpu total runtime"); 247 248 SYSCTL_NODE(_hw, OID_AUTO, vmm, CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, 249 NULL); 250 251 /* 252 * Halt the guest if all vcpus are executing a HLT instruction with 253 * interrupts disabled. 254 */ 255 static int halt_detection_enabled = 1; 256 SYSCTL_INT(_hw_vmm, OID_AUTO, halt_detection, CTLFLAG_RDTUN, 257 &halt_detection_enabled, 0, 258 "Halt VM if all vcpus execute HLT with interrupts disabled"); 259 260 static int vmm_ipinum; 261 SYSCTL_INT(_hw_vmm, OID_AUTO, ipinum, CTLFLAG_RD, &vmm_ipinum, 0, 262 "IPI vector used for vcpu notifications"); 263 264 static int trace_guest_exceptions; 265 SYSCTL_INT(_hw_vmm, OID_AUTO, trace_guest_exceptions, CTLFLAG_RDTUN, 266 &trace_guest_exceptions, 0, 267 "Trap into hypervisor on all guest exceptions and reflect them back"); 268 269 static int trap_wbinvd; 270 SYSCTL_INT(_hw_vmm, OID_AUTO, trap_wbinvd, CTLFLAG_RDTUN, &trap_wbinvd, 0, 271 "WBINVD triggers a VM-exit"); 272 273 u_int vm_maxcpu; 274 SYSCTL_UINT(_hw_vmm, OID_AUTO, maxcpu, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, 275 &vm_maxcpu, 0, "Maximum number of vCPUs"); 276 277 static void vcpu_notify_event_locked(struct vcpu *vcpu, bool lapic_intr); 278 279 /* global statistics */ 280 VMM_STAT(VCPU_MIGRATIONS, "vcpu migration across host cpus"); 281 VMM_STAT(VMEXIT_COUNT, "total number of vm exits"); 282 VMM_STAT(VMEXIT_EXTINT, "vm exits due to external interrupt"); 283 VMM_STAT(VMEXIT_HLT, "number of times hlt was intercepted"); 284 VMM_STAT(VMEXIT_CR_ACCESS, "number of times %cr access was intercepted"); 285 VMM_STAT(VMEXIT_RDMSR, "number of times rdmsr was intercepted"); 286 VMM_STAT(VMEXIT_WRMSR, "number of times wrmsr was intercepted"); 287 VMM_STAT(VMEXIT_MTRAP, "number of monitor trap exits"); 288 VMM_STAT(VMEXIT_PAUSE, "number of times pause was intercepted"); 289 VMM_STAT(VMEXIT_INTR_WINDOW, "vm exits due to interrupt window opening"); 290 VMM_STAT(VMEXIT_NMI_WINDOW, "vm exits due to nmi window opening"); 291 VMM_STAT(VMEXIT_INOUT, "number of times in/out was intercepted"); 292 VMM_STAT(VMEXIT_CPUID, "number of times cpuid was intercepted"); 293 VMM_STAT(VMEXIT_NESTED_FAULT, "vm exits due to nested page fault"); 294 VMM_STAT(VMEXIT_INST_EMUL, "vm exits for instruction emulation"); 295 VMM_STAT(VMEXIT_UNKNOWN, "number of vm exits for unknown reason"); 296 VMM_STAT(VMEXIT_ASTPENDING, "number of times astpending at exit"); 297 VMM_STAT(VMEXIT_REQIDLE, "number of times idle requested at exit"); 298 VMM_STAT(VMEXIT_USERSPACE, "number of vm exits handled in userspace"); 299 VMM_STAT(VMEXIT_RENDEZVOUS, "number of times rendezvous pending at exit"); 300 VMM_STAT(VMEXIT_EXCEPTION, "number of vm exits due to exceptions"); 301 302 /* 303 * Upper limit on vm_maxcpu. Limited by use of uint16_t types for CPU 304 * counts as well as range of vpid values for VT-x and by the capacity 305 * of cpuset_t masks. The call to new_unrhdr() in vpid_init() in 306 * vmx.c requires 'vm_maxcpu + 1 <= 0xffff', hence the '- 1' below. 307 */ 308 #define VM_MAXCPU MIN(0xffff - 1, CPU_SETSIZE) 309 310 #ifdef KTR 311 static const char * 312 vcpu_state2str(enum vcpu_state state) 313 { 314 315 switch (state) { 316 case VCPU_IDLE: 317 return ("idle"); 318 case VCPU_FROZEN: 319 return ("frozen"); 320 case VCPU_RUNNING: 321 return ("running"); 322 case VCPU_SLEEPING: 323 return ("sleeping"); 324 default: 325 return ("unknown"); 326 } 327 } 328 #endif 329 330 static void 331 vcpu_cleanup(struct vcpu *vcpu, bool destroy) 332 { 333 vmmops_vlapic_cleanup(vcpu->vlapic); 334 vmmops_vcpu_cleanup(vcpu->cookie); 335 vcpu->cookie = NULL; 336 if (destroy) { 337 vmm_stat_free(vcpu->stats); 338 fpu_save_area_free(vcpu->guestfpu); 339 vcpu_lock_destroy(vcpu); 340 free(vcpu, M_VM); 341 } 342 } 343 344 static struct vcpu * 345 vcpu_alloc(struct vm *vm, int vcpu_id) 346 { 347 struct vcpu *vcpu; 348 349 KASSERT(vcpu_id >= 0 && vcpu_id < vm->maxcpus, 350 ("vcpu_init: invalid vcpu %d", vcpu_id)); 351 352 vcpu = malloc(sizeof(*vcpu), M_VM, M_WAITOK | M_ZERO); 353 vcpu_lock_init(vcpu); 354 vcpu->state = VCPU_IDLE; 355 vcpu->hostcpu = NOCPU; 356 vcpu->vcpuid = vcpu_id; 357 vcpu->vm = vm; 358 vcpu->guestfpu = fpu_save_area_alloc(); 359 vcpu->stats = vmm_stat_alloc(); 360 vcpu->tsc_offset = 0; 361 return (vcpu); 362 } 363 364 static void 365 vcpu_init(struct vcpu *vcpu) 366 { 367 vcpu->cookie = vmmops_vcpu_init(vcpu->vm->cookie, vcpu, vcpu->vcpuid); 368 vcpu->vlapic = vmmops_vlapic_init(vcpu->cookie); 369 vm_set_x2apic_state(vcpu, X2APIC_DISABLED); 370 vcpu->reqidle = 0; 371 vcpu->exitintinfo = 0; 372 vcpu->nmi_pending = 0; 373 vcpu->extint_pending = 0; 374 vcpu->exception_pending = 0; 375 vcpu->guest_xcr0 = XFEATURE_ENABLED_X87; 376 fpu_save_area_reset(vcpu->guestfpu); 377 vmm_stat_init(vcpu->stats); 378 } 379 380 int 381 vcpu_trace_exceptions(struct vcpu *vcpu) 382 { 383 384 return (trace_guest_exceptions); 385 } 386 387 int 388 vcpu_trap_wbinvd(struct vcpu *vcpu) 389 { 390 return (trap_wbinvd); 391 } 392 393 struct vm_exit * 394 vm_exitinfo(struct vcpu *vcpu) 395 { 396 return (&vcpu->exitinfo); 397 } 398 399 cpuset_t * 400 vm_exitinfo_cpuset(struct vcpu *vcpu) 401 { 402 return (&vcpu->exitinfo_cpuset); 403 } 404 405 static int 406 vmm_init(void) 407 { 408 if (!vmm_is_hw_supported()) 409 return (ENXIO); 410 411 vm_maxcpu = mp_ncpus; 412 TUNABLE_INT_FETCH("hw.vmm.maxcpu", &vm_maxcpu); 413 414 if (vm_maxcpu > VM_MAXCPU) { 415 printf("vmm: vm_maxcpu clamped to %u\n", VM_MAXCPU); 416 vm_maxcpu = VM_MAXCPU; 417 } 418 if (vm_maxcpu == 0) 419 vm_maxcpu = 1; 420 421 vmm_host_state_init(); 422 423 vmm_ipinum = lapic_ipi_alloc(pti ? &IDTVEC(justreturn1_pti) : 424 &IDTVEC(justreturn)); 425 if (vmm_ipinum < 0) 426 vmm_ipinum = IPI_AST; 427 428 vmm_suspend_p = vmmops_modsuspend; 429 vmm_resume_p = vmmops_modresume; 430 431 return (vmmops_modinit(vmm_ipinum)); 432 } 433 434 static int 435 vmm_handler(module_t mod, int what, void *arg) 436 { 437 int error; 438 439 switch (what) { 440 case MOD_LOAD: 441 if (vmm_is_hw_supported()) { 442 error = vmmdev_init(); 443 if (error != 0) 444 break; 445 error = vmm_init(); 446 if (error == 0) 447 vmm_initialized = 1; 448 else 449 (void)vmmdev_cleanup(); 450 } else { 451 error = ENXIO; 452 } 453 break; 454 case MOD_UNLOAD: 455 if (vmm_is_hw_supported()) { 456 error = vmmdev_cleanup(); 457 if (error == 0) { 458 vmm_suspend_p = NULL; 459 vmm_resume_p = NULL; 460 iommu_cleanup(); 461 if (vmm_ipinum != IPI_AST) 462 lapic_ipi_free(vmm_ipinum); 463 error = vmmops_modcleanup(); 464 /* 465 * Something bad happened - prevent new 466 * VMs from being created 467 */ 468 if (error) 469 vmm_initialized = 0; 470 } 471 } else { 472 error = 0; 473 } 474 break; 475 default: 476 error = 0; 477 break; 478 } 479 return (error); 480 } 481 482 static moduledata_t vmm_kmod = { 483 "vmm", 484 vmm_handler, 485 NULL 486 }; 487 488 /* 489 * vmm initialization has the following dependencies: 490 * 491 * - VT-x initialization requires smp_rendezvous() and therefore must happen 492 * after SMP is fully functional (after SI_SUB_SMP). 493 * - vmm device initialization requires an initialized devfs. 494 */ 495 DECLARE_MODULE(vmm, vmm_kmod, MAX(SI_SUB_SMP, SI_SUB_DEVFS) + 1, SI_ORDER_ANY); 496 MODULE_VERSION(vmm, 1); 497 498 static void 499 vm_init(struct vm *vm, bool create) 500 { 501 vm->cookie = vmmops_init(vm, vmspace_pmap(vm_vmspace(vm))); 502 vm->iommu = NULL; 503 vm->vioapic = vioapic_init(vm); 504 vm->vhpet = vhpet_init(vm); 505 vm->vatpic = vatpic_init(vm); 506 vm->vatpit = vatpit_init(vm); 507 vm->vpmtmr = vpmtmr_init(vm); 508 if (create) 509 vm->vrtc = vrtc_init(vm); 510 511 CPU_ZERO(&vm->active_cpus); 512 CPU_ZERO(&vm->debug_cpus); 513 CPU_ZERO(&vm->startup_cpus); 514 515 vm->suspend = 0; 516 CPU_ZERO(&vm->suspended_cpus); 517 518 if (!create) { 519 for (int i = 0; i < vm->maxcpus; i++) { 520 if (vm->vcpu[i] != NULL) 521 vcpu_init(vm->vcpu[i]); 522 } 523 } 524 } 525 526 void 527 vm_disable_vcpu_creation(struct vm *vm) 528 { 529 sx_xlock(&vm->vcpus_init_lock); 530 vm->dying = true; 531 sx_xunlock(&vm->vcpus_init_lock); 532 } 533 534 struct vcpu * 535 vm_alloc_vcpu(struct vm *vm, int vcpuid) 536 { 537 struct vcpu *vcpu; 538 539 if (vcpuid < 0 || vcpuid >= vm_get_maxcpus(vm)) 540 return (NULL); 541 542 vcpu = (struct vcpu *) 543 atomic_load_acq_ptr((uintptr_t *)&vm->vcpu[vcpuid]); 544 if (__predict_true(vcpu != NULL)) 545 return (vcpu); 546 547 sx_xlock(&vm->vcpus_init_lock); 548 vcpu = vm->vcpu[vcpuid]; 549 if (vcpu == NULL && !vm->dying) { 550 vcpu = vcpu_alloc(vm, vcpuid); 551 vcpu_init(vcpu); 552 553 /* 554 * Ensure vCPU is fully created before updating pointer 555 * to permit unlocked reads above. 556 */ 557 atomic_store_rel_ptr((uintptr_t *)&vm->vcpu[vcpuid], 558 (uintptr_t)vcpu); 559 } 560 sx_xunlock(&vm->vcpus_init_lock); 561 return (vcpu); 562 } 563 564 void 565 vm_lock_vcpus(struct vm *vm) 566 { 567 sx_xlock(&vm->vcpus_init_lock); 568 } 569 570 void 571 vm_unlock_vcpus(struct vm *vm) 572 { 573 sx_unlock(&vm->vcpus_init_lock); 574 } 575 576 /* 577 * The default CPU topology is a single thread per package. 578 */ 579 u_int cores_per_package = 1; 580 u_int threads_per_core = 1; 581 582 int 583 vm_create(const char *name, struct vm **retvm) 584 { 585 struct vm *vm; 586 int error; 587 588 /* 589 * If vmm.ko could not be successfully initialized then don't attempt 590 * to create the virtual machine. 591 */ 592 if (!vmm_initialized) 593 return (ENXIO); 594 595 if (name == NULL || strnlen(name, VM_MAX_NAMELEN + 1) == 596 VM_MAX_NAMELEN + 1) 597 return (EINVAL); 598 599 vm = malloc(sizeof(struct vm), M_VM, M_WAITOK | M_ZERO); 600 error = vm_mem_init(&vm->mem, 0, VM_MAXUSER_ADDRESS_LA48); 601 if (error != 0) { 602 free(vm, M_VM); 603 return (error); 604 } 605 strcpy(vm->name, name); 606 mtx_init(&vm->rendezvous_mtx, "vm rendezvous lock", 0, MTX_DEF); 607 sx_init(&vm->vcpus_init_lock, "vm vcpus"); 608 vm->vcpu = malloc(sizeof(*vm->vcpu) * vm_maxcpu, M_VM, M_WAITOK | 609 M_ZERO); 610 611 vm->sockets = 1; 612 vm->cores = cores_per_package; /* XXX backwards compatibility */ 613 vm->threads = threads_per_core; /* XXX backwards compatibility */ 614 vm->maxcpus = vm_maxcpu; 615 616 vm_init(vm, true); 617 618 *retvm = vm; 619 return (0); 620 } 621 622 void 623 vm_get_topology(struct vm *vm, uint16_t *sockets, uint16_t *cores, 624 uint16_t *threads, uint16_t *maxcpus) 625 { 626 *sockets = vm->sockets; 627 *cores = vm->cores; 628 *threads = vm->threads; 629 *maxcpus = vm->maxcpus; 630 } 631 632 uint16_t 633 vm_get_maxcpus(struct vm *vm) 634 { 635 return (vm->maxcpus); 636 } 637 638 int 639 vm_set_topology(struct vm *vm, uint16_t sockets, uint16_t cores, 640 uint16_t threads, uint16_t maxcpus __unused) 641 { 642 /* Ignore maxcpus. */ 643 if ((sockets * cores * threads) > vm->maxcpus) 644 return (EINVAL); 645 vm->sockets = sockets; 646 vm->cores = cores; 647 vm->threads = threads; 648 return(0); 649 } 650 651 static void 652 vm_cleanup(struct vm *vm, bool destroy) 653 { 654 if (destroy) 655 vm_xlock_memsegs(vm); 656 else 657 vm_assert_memseg_xlocked(vm); 658 659 ppt_unassign_all(vm); 660 661 if (vm->iommu != NULL) 662 iommu_destroy_domain(vm->iommu); 663 664 if (destroy) 665 vrtc_cleanup(vm->vrtc); 666 else 667 vrtc_reset(vm->vrtc); 668 vpmtmr_cleanup(vm->vpmtmr); 669 vatpit_cleanup(vm->vatpit); 670 vhpet_cleanup(vm->vhpet); 671 vatpic_cleanup(vm->vatpic); 672 vioapic_cleanup(vm->vioapic); 673 674 for (int i = 0; i < vm->maxcpus; i++) { 675 if (vm->vcpu[i] != NULL) 676 vcpu_cleanup(vm->vcpu[i], destroy); 677 } 678 679 vmmops_cleanup(vm->cookie); 680 681 vm_mem_cleanup(vm); 682 683 if (destroy) { 684 vm_mem_destroy(vm); 685 686 free(vm->vcpu, M_VM); 687 sx_destroy(&vm->vcpus_init_lock); 688 mtx_destroy(&vm->rendezvous_mtx); 689 } 690 } 691 692 void 693 vm_destroy(struct vm *vm) 694 { 695 vm_cleanup(vm, true); 696 free(vm, M_VM); 697 } 698 699 int 700 vm_reinit(struct vm *vm) 701 { 702 int error; 703 704 /* 705 * A virtual machine can be reset only if all vcpus are suspended. 706 */ 707 if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) { 708 vm_cleanup(vm, false); 709 vm_init(vm, false); 710 error = 0; 711 } else { 712 error = EBUSY; 713 } 714 715 return (error); 716 } 717 718 const char * 719 vm_name(struct vm *vm) 720 { 721 return (vm->name); 722 } 723 724 int 725 vm_map_mmio(struct vm *vm, vm_paddr_t gpa, size_t len, vm_paddr_t hpa) 726 { 727 vm_object_t obj; 728 729 if ((obj = vmm_mmio_alloc(vm_vmspace(vm), gpa, len, hpa)) == NULL) 730 return (ENOMEM); 731 else 732 return (0); 733 } 734 735 int 736 vm_unmap_mmio(struct vm *vm, vm_paddr_t gpa, size_t len) 737 { 738 739 vmm_mmio_free(vm_vmspace(vm), gpa, len); 740 return (0); 741 } 742 743 static int 744 vm_iommu_map(struct vm *vm) 745 { 746 pmap_t pmap; 747 vm_paddr_t gpa, hpa; 748 struct vm_mem_map *mm; 749 int error, i; 750 751 sx_assert(&vm->mem.mem_segs_lock, SX_LOCKED); 752 753 pmap = vmspace_pmap(vm_vmspace(vm)); 754 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 755 if (!vm_memseg_sysmem(vm, i)) 756 continue; 757 758 mm = &vm->mem.mem_maps[i]; 759 KASSERT((mm->flags & VM_MEMMAP_F_IOMMU) == 0, 760 ("iommu map found invalid memmap %#lx/%#lx/%#x", 761 mm->gpa, mm->len, mm->flags)); 762 if ((mm->flags & VM_MEMMAP_F_WIRED) == 0) 763 continue; 764 mm->flags |= VM_MEMMAP_F_IOMMU; 765 766 for (gpa = mm->gpa; gpa < mm->gpa + mm->len; gpa += PAGE_SIZE) { 767 hpa = pmap_extract(pmap, gpa); 768 769 /* 770 * All mappings in the vmm vmspace must be 771 * present since they are managed by vmm in this way. 772 * Because we are in pass-through mode, the 773 * mappings must also be wired. This implies 774 * that all pages must be mapped and wired, 775 * allowing to use pmap_extract() and avoiding the 776 * need to use vm_gpa_hold_global(). 777 * 778 * This could change if/when we start 779 * supporting page faults on IOMMU maps. 780 */ 781 KASSERT(vm_page_wired(PHYS_TO_VM_PAGE(hpa)), 782 ("vm_iommu_map: vm %p gpa %jx hpa %jx not wired", 783 vm, (uintmax_t)gpa, (uintmax_t)hpa)); 784 785 iommu_create_mapping(vm->iommu, gpa, hpa, PAGE_SIZE); 786 } 787 } 788 789 error = iommu_invalidate_tlb(iommu_host_domain()); 790 return (error); 791 } 792 793 static int 794 vm_iommu_unmap(struct vm *vm) 795 { 796 vm_paddr_t gpa; 797 struct vm_mem_map *mm; 798 int error, i; 799 800 sx_assert(&vm->mem.mem_segs_lock, SX_LOCKED); 801 802 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 803 if (!vm_memseg_sysmem(vm, i)) 804 continue; 805 806 mm = &vm->mem.mem_maps[i]; 807 if ((mm->flags & VM_MEMMAP_F_IOMMU) == 0) 808 continue; 809 mm->flags &= ~VM_MEMMAP_F_IOMMU; 810 KASSERT((mm->flags & VM_MEMMAP_F_WIRED) != 0, 811 ("iommu unmap found invalid memmap %#lx/%#lx/%#x", 812 mm->gpa, mm->len, mm->flags)); 813 814 for (gpa = mm->gpa; gpa < mm->gpa + mm->len; gpa += PAGE_SIZE) { 815 KASSERT(vm_page_wired(PHYS_TO_VM_PAGE(pmap_extract( 816 vmspace_pmap(vm_vmspace(vm)), gpa))), 817 ("vm_iommu_unmap: vm %p gpa %jx not wired", 818 vm, (uintmax_t)gpa)); 819 iommu_remove_mapping(vm->iommu, gpa, PAGE_SIZE); 820 } 821 } 822 823 /* 824 * Invalidate the cached translations associated with the domain 825 * from which pages were removed. 826 */ 827 error = iommu_invalidate_tlb(vm->iommu); 828 return (error); 829 } 830 831 int 832 vm_unassign_pptdev(struct vm *vm, int bus, int slot, int func) 833 { 834 int error; 835 836 error = ppt_unassign_device(vm, bus, slot, func); 837 if (error) 838 return (error); 839 840 if (ppt_assigned_devices(vm) == 0) 841 error = vm_iommu_unmap(vm); 842 843 return (error); 844 } 845 846 int 847 vm_assign_pptdev(struct vm *vm, int bus, int slot, int func) 848 { 849 int error; 850 vm_paddr_t maxaddr; 851 bool map = false; 852 853 /* Set up the IOMMU to do the 'gpa' to 'hpa' translation */ 854 if (ppt_assigned_devices(vm) == 0) { 855 KASSERT(vm->iommu == NULL, 856 ("vm_assign_pptdev: iommu must be NULL")); 857 maxaddr = vmm_sysmem_maxaddr(vm); 858 vm->iommu = iommu_create_domain(maxaddr); 859 if (vm->iommu == NULL) 860 return (ENXIO); 861 map = true; 862 } 863 864 error = ppt_assign_device(vm, bus, slot, func); 865 if (error == 0 && map) 866 error = vm_iommu_map(vm); 867 return (error); 868 } 869 870 int 871 vm_get_register(struct vcpu *vcpu, int reg, uint64_t *retval) 872 { 873 874 if (reg >= VM_REG_LAST) 875 return (EINVAL); 876 877 return (vmmops_getreg(vcpu->cookie, reg, retval)); 878 } 879 880 int 881 vm_set_register(struct vcpu *vcpu, int reg, uint64_t val) 882 { 883 int error; 884 885 if (reg >= VM_REG_LAST) 886 return (EINVAL); 887 888 error = vmmops_setreg(vcpu->cookie, reg, val); 889 if (error || reg != VM_REG_GUEST_RIP) 890 return (error); 891 892 /* Set 'nextrip' to match the value of %rip */ 893 VMM_CTR1(vcpu, "Setting nextrip to %#lx", val); 894 vcpu->nextrip = val; 895 return (0); 896 } 897 898 static bool 899 is_descriptor_table(int reg) 900 { 901 902 switch (reg) { 903 case VM_REG_GUEST_IDTR: 904 case VM_REG_GUEST_GDTR: 905 return (true); 906 default: 907 return (false); 908 } 909 } 910 911 static bool 912 is_segment_register(int reg) 913 { 914 915 switch (reg) { 916 case VM_REG_GUEST_ES: 917 case VM_REG_GUEST_CS: 918 case VM_REG_GUEST_SS: 919 case VM_REG_GUEST_DS: 920 case VM_REG_GUEST_FS: 921 case VM_REG_GUEST_GS: 922 case VM_REG_GUEST_TR: 923 case VM_REG_GUEST_LDTR: 924 return (true); 925 default: 926 return (false); 927 } 928 } 929 930 int 931 vm_get_seg_desc(struct vcpu *vcpu, int reg, struct seg_desc *desc) 932 { 933 934 if (!is_segment_register(reg) && !is_descriptor_table(reg)) 935 return (EINVAL); 936 937 return (vmmops_getdesc(vcpu->cookie, reg, desc)); 938 } 939 940 int 941 vm_set_seg_desc(struct vcpu *vcpu, int reg, struct seg_desc *desc) 942 { 943 944 if (!is_segment_register(reg) && !is_descriptor_table(reg)) 945 return (EINVAL); 946 947 return (vmmops_setdesc(vcpu->cookie, reg, desc)); 948 } 949 950 static void 951 restore_guest_fpustate(struct vcpu *vcpu) 952 { 953 954 /* flush host state to the pcb */ 955 fpuexit(curthread); 956 957 /* restore guest FPU state */ 958 fpu_enable(); 959 fpurestore(vcpu->guestfpu); 960 961 /* restore guest XCR0 if XSAVE is enabled in the host */ 962 if (rcr4() & CR4_XSAVE) 963 load_xcr(0, vcpu->guest_xcr0); 964 965 /* 966 * The FPU is now "dirty" with the guest's state so disable 967 * the FPU to trap any access by the host. 968 */ 969 fpu_disable(); 970 } 971 972 static void 973 save_guest_fpustate(struct vcpu *vcpu) 974 { 975 976 if ((rcr0() & CR0_TS) == 0) 977 panic("fpu emulation not enabled in host!"); 978 979 /* save guest XCR0 and restore host XCR0 */ 980 if (rcr4() & CR4_XSAVE) { 981 vcpu->guest_xcr0 = rxcr(0); 982 load_xcr(0, vmm_get_host_xcr0()); 983 } 984 985 /* save guest FPU state */ 986 fpu_enable(); 987 fpusave(vcpu->guestfpu); 988 fpu_disable(); 989 } 990 991 static VMM_STAT(VCPU_IDLE_TICKS, "number of ticks vcpu was idle"); 992 993 /* 994 * Invoke the rendezvous function on the specified vcpu if applicable. Return 995 * true if the rendezvous is finished, false otherwise. 996 */ 997 static bool 998 vm_rendezvous(struct vcpu *vcpu) 999 { 1000 struct vm *vm = vcpu->vm; 1001 int vcpuid; 1002 1003 mtx_assert(&vcpu->vm->rendezvous_mtx, MA_OWNED); 1004 KASSERT(vcpu->vm->rendezvous_func != NULL, 1005 ("vm_rendezvous: no rendezvous pending")); 1006 1007 /* 'rendezvous_req_cpus' must be a subset of 'active_cpus' */ 1008 CPU_AND(&vm->rendezvous_req_cpus, &vm->rendezvous_req_cpus, 1009 &vm->active_cpus); 1010 1011 vcpuid = vcpu->vcpuid; 1012 if (CPU_ISSET(vcpuid, &vm->rendezvous_req_cpus) && 1013 !CPU_ISSET(vcpuid, &vm->rendezvous_done_cpus)) { 1014 VMM_CTR0(vcpu, "Calling rendezvous func"); 1015 (*vm->rendezvous_func)(vcpu, vm->rendezvous_arg); 1016 CPU_SET(vcpuid, &vm->rendezvous_done_cpus); 1017 } 1018 if (CPU_CMP(&vm->rendezvous_req_cpus, 1019 &vm->rendezvous_done_cpus) == 0) { 1020 VMM_CTR0(vcpu, "Rendezvous completed"); 1021 CPU_ZERO(&vm->rendezvous_req_cpus); 1022 vm->rendezvous_func = NULL; 1023 wakeup(&vm->rendezvous_func); 1024 return (true); 1025 } 1026 return (false); 1027 } 1028 1029 static void 1030 vcpu_wait_idle(struct vcpu *vcpu) 1031 { 1032 KASSERT(vcpu->state != VCPU_IDLE, ("vcpu already idle")); 1033 1034 vcpu->reqidle = 1; 1035 vcpu_notify_event_locked(vcpu, false); 1036 VMM_CTR1(vcpu, "vcpu state change from %s to " 1037 "idle requested", vcpu_state2str(vcpu->state)); 1038 msleep_spin(&vcpu->state, &vcpu->mtx, "vmstat", hz); 1039 } 1040 1041 static int 1042 vcpu_set_state_locked(struct vcpu *vcpu, enum vcpu_state newstate, 1043 bool from_idle) 1044 { 1045 int error; 1046 1047 vcpu_assert_locked(vcpu); 1048 1049 /* 1050 * State transitions from the vmmdev_ioctl() must always begin from 1051 * the VCPU_IDLE state. This guarantees that there is only a single 1052 * ioctl() operating on a vcpu at any point. 1053 */ 1054 if (from_idle) { 1055 while (vcpu->state != VCPU_IDLE) 1056 vcpu_wait_idle(vcpu); 1057 } else { 1058 KASSERT(vcpu->state != VCPU_IDLE, ("invalid transition from " 1059 "vcpu idle state")); 1060 } 1061 1062 if (vcpu->state == VCPU_RUNNING) { 1063 KASSERT(vcpu->hostcpu == curcpu, ("curcpu %d and hostcpu %d " 1064 "mismatch for running vcpu", curcpu, vcpu->hostcpu)); 1065 } else { 1066 KASSERT(vcpu->hostcpu == NOCPU, ("Invalid hostcpu %d for a " 1067 "vcpu that is not running", vcpu->hostcpu)); 1068 } 1069 1070 /* 1071 * The following state transitions are allowed: 1072 * IDLE -> FROZEN -> IDLE 1073 * FROZEN -> RUNNING -> FROZEN 1074 * FROZEN -> SLEEPING -> FROZEN 1075 */ 1076 switch (vcpu->state) { 1077 case VCPU_IDLE: 1078 case VCPU_RUNNING: 1079 case VCPU_SLEEPING: 1080 error = (newstate != VCPU_FROZEN); 1081 break; 1082 case VCPU_FROZEN: 1083 error = (newstate == VCPU_FROZEN); 1084 break; 1085 default: 1086 error = 1; 1087 break; 1088 } 1089 1090 if (error) 1091 return (EBUSY); 1092 1093 VMM_CTR2(vcpu, "vcpu state changed from %s to %s", 1094 vcpu_state2str(vcpu->state), vcpu_state2str(newstate)); 1095 1096 vcpu->state = newstate; 1097 if (newstate == VCPU_RUNNING) 1098 vcpu->hostcpu = curcpu; 1099 else 1100 vcpu->hostcpu = NOCPU; 1101 1102 if (newstate == VCPU_IDLE) 1103 wakeup(&vcpu->state); 1104 1105 return (0); 1106 } 1107 1108 /* 1109 * Try to lock all of the vCPUs in the VM while taking care to avoid deadlocks 1110 * with vm_smp_rendezvous(). 1111 * 1112 * The complexity here suggests that the rendezvous mechanism needs a rethink. 1113 */ 1114 int 1115 vcpu_set_state_all(struct vm *vm, enum vcpu_state newstate) 1116 { 1117 cpuset_t locked; 1118 struct vcpu *vcpu; 1119 int error, i; 1120 uint16_t maxcpus; 1121 1122 KASSERT(newstate != VCPU_IDLE, 1123 ("vcpu_set_state_all: invalid target state %d", newstate)); 1124 1125 error = 0; 1126 CPU_ZERO(&locked); 1127 maxcpus = vm->maxcpus; 1128 1129 mtx_lock(&vm->rendezvous_mtx); 1130 restart: 1131 if (vm->rendezvous_func != NULL) { 1132 /* 1133 * If we have a pending rendezvous, then the initiator may be 1134 * blocked waiting for other vCPUs to execute the callback. The 1135 * current thread may be a vCPU thread so we must not block 1136 * waiting for the initiator, otherwise we get a deadlock. 1137 * Thus, execute the callback on behalf of any idle vCPUs. 1138 */ 1139 for (i = 0; i < maxcpus; i++) { 1140 vcpu = vm_vcpu(vm, i); 1141 if (vcpu == NULL) 1142 continue; 1143 vcpu_lock(vcpu); 1144 if (vcpu->state == VCPU_IDLE) { 1145 (void)vcpu_set_state_locked(vcpu, VCPU_FROZEN, 1146 true); 1147 CPU_SET(i, &locked); 1148 } 1149 if (CPU_ISSET(i, &locked)) { 1150 /* 1151 * We can safely execute the callback on this 1152 * vCPU's behalf. 1153 */ 1154 vcpu_unlock(vcpu); 1155 (void)vm_rendezvous(vcpu); 1156 vcpu_lock(vcpu); 1157 } 1158 vcpu_unlock(vcpu); 1159 } 1160 } 1161 1162 /* 1163 * Now wait for remaining vCPUs to become idle. This may include the 1164 * initiator of a rendezvous that is currently blocked on the rendezvous 1165 * mutex. 1166 */ 1167 CPU_FOREACH_ISCLR(i, &locked) { 1168 if (i >= maxcpus) 1169 break; 1170 vcpu = vm_vcpu(vm, i); 1171 if (vcpu == NULL) 1172 continue; 1173 vcpu_lock(vcpu); 1174 while (vcpu->state != VCPU_IDLE) { 1175 mtx_unlock(&vm->rendezvous_mtx); 1176 vcpu_wait_idle(vcpu); 1177 vcpu_unlock(vcpu); 1178 mtx_lock(&vm->rendezvous_mtx); 1179 if (vm->rendezvous_func != NULL) 1180 goto restart; 1181 vcpu_lock(vcpu); 1182 } 1183 error = vcpu_set_state_locked(vcpu, newstate, true); 1184 vcpu_unlock(vcpu); 1185 if (error != 0) { 1186 /* Roll back state changes. */ 1187 CPU_FOREACH_ISSET(i, &locked) 1188 (void)vcpu_set_state(vcpu, VCPU_IDLE, false); 1189 break; 1190 } 1191 CPU_SET(i, &locked); 1192 } 1193 mtx_unlock(&vm->rendezvous_mtx); 1194 return (error); 1195 } 1196 1197 static void 1198 vcpu_require_state(struct vcpu *vcpu, enum vcpu_state newstate) 1199 { 1200 int error; 1201 1202 if ((error = vcpu_set_state(vcpu, newstate, false)) != 0) 1203 panic("Error %d setting state to %d\n", error, newstate); 1204 } 1205 1206 static void 1207 vcpu_require_state_locked(struct vcpu *vcpu, enum vcpu_state newstate) 1208 { 1209 int error; 1210 1211 if ((error = vcpu_set_state_locked(vcpu, newstate, false)) != 0) 1212 panic("Error %d setting state to %d", error, newstate); 1213 } 1214 1215 static int 1216 vm_handle_rendezvous(struct vcpu *vcpu) 1217 { 1218 struct vm *vm; 1219 struct thread *td; 1220 1221 td = curthread; 1222 vm = vcpu->vm; 1223 1224 mtx_lock(&vm->rendezvous_mtx); 1225 while (vm->rendezvous_func != NULL) { 1226 if (vm_rendezvous(vcpu)) 1227 break; 1228 1229 VMM_CTR0(vcpu, "Wait for rendezvous completion"); 1230 mtx_sleep(&vm->rendezvous_func, &vm->rendezvous_mtx, 0, 1231 "vmrndv", hz); 1232 if (td_ast_pending(td, TDA_SUSPEND)) { 1233 int error; 1234 1235 mtx_unlock(&vm->rendezvous_mtx); 1236 error = thread_check_susp(td, true); 1237 if (error != 0) 1238 return (error); 1239 mtx_lock(&vm->rendezvous_mtx); 1240 } 1241 } 1242 mtx_unlock(&vm->rendezvous_mtx); 1243 return (0); 1244 } 1245 1246 /* 1247 * Emulate a guest 'hlt' by sleeping until the vcpu is ready to run. 1248 */ 1249 static int 1250 vm_handle_hlt(struct vcpu *vcpu, bool intr_disabled, bool *retu) 1251 { 1252 struct vm *vm = vcpu->vm; 1253 const char *wmesg; 1254 struct thread *td; 1255 int error, t, vcpuid, vcpu_halted, vm_halted; 1256 1257 vcpuid = vcpu->vcpuid; 1258 vcpu_halted = 0; 1259 vm_halted = 0; 1260 error = 0; 1261 td = curthread; 1262 1263 KASSERT(!CPU_ISSET(vcpuid, &vm->halted_cpus), ("vcpu already halted")); 1264 1265 vcpu_lock(vcpu); 1266 while (1) { 1267 /* 1268 * Do a final check for pending NMI or interrupts before 1269 * really putting this thread to sleep. Also check for 1270 * software events that would cause this vcpu to wakeup. 1271 * 1272 * These interrupts/events could have happened after the 1273 * vcpu returned from vmmops_run() and before it acquired the 1274 * vcpu lock above. 1275 */ 1276 if (vm->rendezvous_func != NULL || vm->suspend || vcpu->reqidle) 1277 break; 1278 if (vm_nmi_pending(vcpu)) 1279 break; 1280 if (!intr_disabled) { 1281 if (vm_extint_pending(vcpu) || 1282 vlapic_pending_intr(vcpu->vlapic, NULL)) { 1283 break; 1284 } 1285 } 1286 1287 /* Don't go to sleep if the vcpu thread needs to yield */ 1288 if (vcpu_should_yield(vcpu)) 1289 break; 1290 1291 if (vcpu_debugged(vcpu)) 1292 break; 1293 1294 /* 1295 * Some Linux guests implement "halt" by having all vcpus 1296 * execute HLT with interrupts disabled. 'halted_cpus' keeps 1297 * track of the vcpus that have entered this state. When all 1298 * vcpus enter the halted state the virtual machine is halted. 1299 */ 1300 if (intr_disabled) { 1301 wmesg = "vmhalt"; 1302 VMM_CTR0(vcpu, "Halted"); 1303 if (!vcpu_halted && halt_detection_enabled) { 1304 vcpu_halted = 1; 1305 CPU_SET_ATOMIC(vcpuid, &vm->halted_cpus); 1306 } 1307 if (CPU_CMP(&vm->halted_cpus, &vm->active_cpus) == 0) { 1308 vm_halted = 1; 1309 break; 1310 } 1311 } else { 1312 wmesg = "vmidle"; 1313 } 1314 1315 t = ticks; 1316 vcpu_require_state_locked(vcpu, VCPU_SLEEPING); 1317 /* 1318 * XXX msleep_spin() cannot be interrupted by signals so 1319 * wake up periodically to check pending signals. 1320 */ 1321 msleep_spin(vcpu, &vcpu->mtx, wmesg, hz); 1322 vcpu_require_state_locked(vcpu, VCPU_FROZEN); 1323 vmm_stat_incr(vcpu, VCPU_IDLE_TICKS, ticks - t); 1324 if (td_ast_pending(td, TDA_SUSPEND)) { 1325 vcpu_unlock(vcpu); 1326 error = thread_check_susp(td, false); 1327 if (error != 0) { 1328 if (vcpu_halted) { 1329 CPU_CLR_ATOMIC(vcpuid, 1330 &vm->halted_cpus); 1331 } 1332 return (error); 1333 } 1334 vcpu_lock(vcpu); 1335 } 1336 } 1337 1338 if (vcpu_halted) 1339 CPU_CLR_ATOMIC(vcpuid, &vm->halted_cpus); 1340 1341 vcpu_unlock(vcpu); 1342 1343 if (vm_halted) 1344 vm_suspend(vm, VM_SUSPEND_HALT); 1345 1346 return (0); 1347 } 1348 1349 static int 1350 vm_handle_paging(struct vcpu *vcpu, bool *retu) 1351 { 1352 struct vm *vm = vcpu->vm; 1353 int rv, ftype; 1354 struct vm_map *map; 1355 struct vm_exit *vme; 1356 1357 vme = &vcpu->exitinfo; 1358 1359 KASSERT(vme->inst_length == 0, ("%s: invalid inst_length %d", 1360 __func__, vme->inst_length)); 1361 1362 ftype = vme->u.paging.fault_type; 1363 KASSERT(ftype == VM_PROT_READ || 1364 ftype == VM_PROT_WRITE || ftype == VM_PROT_EXECUTE, 1365 ("vm_handle_paging: invalid fault_type %d", ftype)); 1366 1367 if (ftype == VM_PROT_READ || ftype == VM_PROT_WRITE) { 1368 rv = pmap_emulate_accessed_dirty(vmspace_pmap(vm_vmspace(vm)), 1369 vme->u.paging.gpa, ftype); 1370 if (rv == 0) { 1371 VMM_CTR2(vcpu, "%s bit emulation for gpa %#lx", 1372 ftype == VM_PROT_READ ? "accessed" : "dirty", 1373 vme->u.paging.gpa); 1374 goto done; 1375 } 1376 } 1377 1378 map = &vm_vmspace(vm)->vm_map; 1379 rv = vm_fault(map, vme->u.paging.gpa, ftype, VM_FAULT_NORMAL, NULL); 1380 1381 VMM_CTR3(vcpu, "vm_handle_paging rv = %d, gpa = %#lx, " 1382 "ftype = %d", rv, vme->u.paging.gpa, ftype); 1383 1384 if (rv != KERN_SUCCESS) 1385 return (EFAULT); 1386 done: 1387 return (0); 1388 } 1389 1390 static int 1391 vm_handle_inst_emul(struct vcpu *vcpu, bool *retu) 1392 { 1393 struct vie *vie; 1394 struct vm_exit *vme; 1395 uint64_t gla, gpa, cs_base; 1396 struct vm_guest_paging *paging; 1397 mem_region_read_t mread; 1398 mem_region_write_t mwrite; 1399 enum vm_cpu_mode cpu_mode; 1400 int cs_d, error, fault; 1401 1402 vme = &vcpu->exitinfo; 1403 1404 KASSERT(vme->inst_length == 0, ("%s: invalid inst_length %d", 1405 __func__, vme->inst_length)); 1406 1407 gla = vme->u.inst_emul.gla; 1408 gpa = vme->u.inst_emul.gpa; 1409 cs_base = vme->u.inst_emul.cs_base; 1410 cs_d = vme->u.inst_emul.cs_d; 1411 vie = &vme->u.inst_emul.vie; 1412 paging = &vme->u.inst_emul.paging; 1413 cpu_mode = paging->cpu_mode; 1414 1415 VMM_CTR1(vcpu, "inst_emul fault accessing gpa %#lx", gpa); 1416 1417 /* Fetch, decode and emulate the faulting instruction */ 1418 if (vie->num_valid == 0) { 1419 error = vmm_fetch_instruction(vcpu, paging, vme->rip + cs_base, 1420 VIE_INST_SIZE, vie, &fault); 1421 } else { 1422 /* 1423 * The instruction bytes have already been copied into 'vie' 1424 */ 1425 error = fault = 0; 1426 } 1427 if (error || fault) 1428 return (error); 1429 1430 if (vmm_decode_instruction(vcpu, gla, cpu_mode, cs_d, vie) != 0) { 1431 VMM_CTR1(vcpu, "Error decoding instruction at %#lx", 1432 vme->rip + cs_base); 1433 *retu = true; /* dump instruction bytes in userspace */ 1434 return (0); 1435 } 1436 1437 /* 1438 * Update 'nextrip' based on the length of the emulated instruction. 1439 */ 1440 vme->inst_length = vie->num_processed; 1441 vcpu->nextrip += vie->num_processed; 1442 VMM_CTR1(vcpu, "nextrip updated to %#lx after instruction decoding", 1443 vcpu->nextrip); 1444 1445 /* return to userland unless this is an in-kernel emulated device */ 1446 if (gpa >= DEFAULT_APIC_BASE && gpa < DEFAULT_APIC_BASE + PAGE_SIZE) { 1447 mread = lapic_mmio_read; 1448 mwrite = lapic_mmio_write; 1449 } else if (gpa >= VIOAPIC_BASE && gpa < VIOAPIC_BASE + VIOAPIC_SIZE) { 1450 mread = vioapic_mmio_read; 1451 mwrite = vioapic_mmio_write; 1452 } else if (gpa >= VHPET_BASE && gpa < VHPET_BASE + VHPET_SIZE) { 1453 mread = vhpet_mmio_read; 1454 mwrite = vhpet_mmio_write; 1455 } else { 1456 *retu = true; 1457 return (0); 1458 } 1459 1460 error = vmm_emulate_instruction(vcpu, gpa, vie, paging, mread, mwrite, 1461 retu); 1462 1463 return (error); 1464 } 1465 1466 static int 1467 vm_handle_suspend(struct vcpu *vcpu, bool *retu) 1468 { 1469 struct vm *vm = vcpu->vm; 1470 int error, i; 1471 struct thread *td; 1472 1473 error = 0; 1474 td = curthread; 1475 1476 CPU_SET_ATOMIC(vcpu->vcpuid, &vm->suspended_cpus); 1477 1478 /* 1479 * Wait until all 'active_cpus' have suspended themselves. 1480 * 1481 * Since a VM may be suspended at any time including when one or 1482 * more vcpus are doing a rendezvous we need to call the rendezvous 1483 * handler while we are waiting to prevent a deadlock. 1484 */ 1485 vcpu_lock(vcpu); 1486 while (error == 0) { 1487 if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) { 1488 VMM_CTR0(vcpu, "All vcpus suspended"); 1489 break; 1490 } 1491 1492 if (vm->rendezvous_func == NULL) { 1493 VMM_CTR0(vcpu, "Sleeping during suspend"); 1494 vcpu_require_state_locked(vcpu, VCPU_SLEEPING); 1495 msleep_spin(vcpu, &vcpu->mtx, "vmsusp", hz); 1496 vcpu_require_state_locked(vcpu, VCPU_FROZEN); 1497 if (td_ast_pending(td, TDA_SUSPEND)) { 1498 vcpu_unlock(vcpu); 1499 error = thread_check_susp(td, false); 1500 vcpu_lock(vcpu); 1501 } 1502 } else { 1503 VMM_CTR0(vcpu, "Rendezvous during suspend"); 1504 vcpu_unlock(vcpu); 1505 error = vm_handle_rendezvous(vcpu); 1506 vcpu_lock(vcpu); 1507 } 1508 } 1509 vcpu_unlock(vcpu); 1510 1511 /* 1512 * Wakeup the other sleeping vcpus and return to userspace. 1513 */ 1514 for (i = 0; i < vm->maxcpus; i++) { 1515 if (CPU_ISSET(i, &vm->suspended_cpus)) { 1516 vcpu_notify_event(vm_vcpu(vm, i), false); 1517 } 1518 } 1519 1520 *retu = true; 1521 return (error); 1522 } 1523 1524 static int 1525 vm_handle_reqidle(struct vcpu *vcpu, bool *retu) 1526 { 1527 vcpu_lock(vcpu); 1528 KASSERT(vcpu->reqidle, ("invalid vcpu reqidle %d", vcpu->reqidle)); 1529 vcpu->reqidle = 0; 1530 vcpu_unlock(vcpu); 1531 *retu = true; 1532 return (0); 1533 } 1534 1535 static int 1536 vm_handle_db(struct vcpu *vcpu, struct vm_exit *vme, bool *retu) 1537 { 1538 int error, fault; 1539 uint64_t rsp; 1540 uint64_t rflags; 1541 struct vm_copyinfo copyinfo[2]; 1542 1543 *retu = true; 1544 if (!vme->u.dbg.pushf_intercept || vme->u.dbg.tf_shadow_val != 0) { 1545 return (0); 1546 } 1547 1548 vm_get_register(vcpu, VM_REG_GUEST_RSP, &rsp); 1549 error = vm_copy_setup(vcpu, &vme->u.dbg.paging, rsp, sizeof(uint64_t), 1550 VM_PROT_RW, copyinfo, nitems(copyinfo), &fault); 1551 if (error != 0 || fault != 0) { 1552 *retu = false; 1553 return (EINVAL); 1554 } 1555 1556 /* Read pushed rflags value from top of stack. */ 1557 vm_copyin(copyinfo, &rflags, sizeof(uint64_t)); 1558 1559 /* Clear TF bit. */ 1560 rflags &= ~(PSL_T); 1561 1562 /* Write updated value back to memory. */ 1563 vm_copyout(&rflags, copyinfo, sizeof(uint64_t)); 1564 vm_copy_teardown(copyinfo, nitems(copyinfo)); 1565 1566 return (0); 1567 } 1568 1569 int 1570 vm_suspend(struct vm *vm, enum vm_suspend_how how) 1571 { 1572 int i; 1573 1574 if (how <= VM_SUSPEND_NONE || how >= VM_SUSPEND_LAST) 1575 return (EINVAL); 1576 1577 if (atomic_cmpset_int(&vm->suspend, 0, how) == 0) { 1578 VM_CTR2(vm, "virtual machine already suspended %d/%d", 1579 vm->suspend, how); 1580 return (EALREADY); 1581 } 1582 1583 VM_CTR1(vm, "virtual machine successfully suspended %d", how); 1584 1585 /* 1586 * Notify all active vcpus that they are now suspended. 1587 */ 1588 for (i = 0; i < vm->maxcpus; i++) { 1589 if (CPU_ISSET(i, &vm->active_cpus)) 1590 vcpu_notify_event(vm_vcpu(vm, i), false); 1591 } 1592 1593 return (0); 1594 } 1595 1596 void 1597 vm_exit_suspended(struct vcpu *vcpu, uint64_t rip) 1598 { 1599 struct vm *vm = vcpu->vm; 1600 struct vm_exit *vmexit; 1601 1602 KASSERT(vm->suspend > VM_SUSPEND_NONE && vm->suspend < VM_SUSPEND_LAST, 1603 ("vm_exit_suspended: invalid suspend type %d", vm->suspend)); 1604 1605 vmexit = vm_exitinfo(vcpu); 1606 vmexit->rip = rip; 1607 vmexit->inst_length = 0; 1608 vmexit->exitcode = VM_EXITCODE_SUSPENDED; 1609 vmexit->u.suspended.how = vm->suspend; 1610 } 1611 1612 void 1613 vm_exit_debug(struct vcpu *vcpu, uint64_t rip) 1614 { 1615 struct vm_exit *vmexit; 1616 1617 vmexit = vm_exitinfo(vcpu); 1618 vmexit->rip = rip; 1619 vmexit->inst_length = 0; 1620 vmexit->exitcode = VM_EXITCODE_DEBUG; 1621 } 1622 1623 void 1624 vm_exit_rendezvous(struct vcpu *vcpu, uint64_t rip) 1625 { 1626 struct vm_exit *vmexit; 1627 1628 vmexit = vm_exitinfo(vcpu); 1629 vmexit->rip = rip; 1630 vmexit->inst_length = 0; 1631 vmexit->exitcode = VM_EXITCODE_RENDEZVOUS; 1632 vmm_stat_incr(vcpu, VMEXIT_RENDEZVOUS, 1); 1633 } 1634 1635 void 1636 vm_exit_reqidle(struct vcpu *vcpu, uint64_t rip) 1637 { 1638 struct vm_exit *vmexit; 1639 1640 vmexit = vm_exitinfo(vcpu); 1641 vmexit->rip = rip; 1642 vmexit->inst_length = 0; 1643 vmexit->exitcode = VM_EXITCODE_REQIDLE; 1644 vmm_stat_incr(vcpu, VMEXIT_REQIDLE, 1); 1645 } 1646 1647 void 1648 vm_exit_astpending(struct vcpu *vcpu, uint64_t rip) 1649 { 1650 struct vm_exit *vmexit; 1651 1652 vmexit = vm_exitinfo(vcpu); 1653 vmexit->rip = rip; 1654 vmexit->inst_length = 0; 1655 vmexit->exitcode = VM_EXITCODE_BOGUS; 1656 vmm_stat_incr(vcpu, VMEXIT_ASTPENDING, 1); 1657 } 1658 1659 int 1660 vm_run(struct vcpu *vcpu) 1661 { 1662 struct vm *vm = vcpu->vm; 1663 struct vm_eventinfo evinfo; 1664 int error, vcpuid; 1665 struct pcb *pcb; 1666 uint64_t tscval; 1667 struct vm_exit *vme; 1668 bool retu, intr_disabled; 1669 pmap_t pmap; 1670 1671 vcpuid = vcpu->vcpuid; 1672 1673 if (!CPU_ISSET(vcpuid, &vm->active_cpus)) 1674 return (EINVAL); 1675 1676 if (CPU_ISSET(vcpuid, &vm->suspended_cpus)) 1677 return (EINVAL); 1678 1679 pmap = vmspace_pmap(vm_vmspace(vm)); 1680 vme = &vcpu->exitinfo; 1681 evinfo.rptr = &vm->rendezvous_req_cpus; 1682 evinfo.sptr = &vm->suspend; 1683 evinfo.iptr = &vcpu->reqidle; 1684 restart: 1685 critical_enter(); 1686 1687 KASSERT(!CPU_ISSET(curcpu, &pmap->pm_active), 1688 ("vm_run: absurd pm_active")); 1689 1690 tscval = rdtsc(); 1691 1692 pcb = PCPU_GET(curpcb); 1693 set_pcb_flags(pcb, PCB_FULL_IRET); 1694 1695 restore_guest_fpustate(vcpu); 1696 1697 vcpu_require_state(vcpu, VCPU_RUNNING); 1698 error = vmmops_run(vcpu->cookie, vcpu->nextrip, pmap, &evinfo); 1699 vcpu_require_state(vcpu, VCPU_FROZEN); 1700 1701 save_guest_fpustate(vcpu); 1702 1703 vmm_stat_incr(vcpu, VCPU_TOTAL_RUNTIME, rdtsc() - tscval); 1704 1705 critical_exit(); 1706 1707 if (error == 0) { 1708 retu = false; 1709 vcpu->nextrip = vme->rip + vme->inst_length; 1710 switch (vme->exitcode) { 1711 case VM_EXITCODE_REQIDLE: 1712 error = vm_handle_reqidle(vcpu, &retu); 1713 break; 1714 case VM_EXITCODE_SUSPENDED: 1715 error = vm_handle_suspend(vcpu, &retu); 1716 break; 1717 case VM_EXITCODE_IOAPIC_EOI: 1718 vioapic_process_eoi(vm, vme->u.ioapic_eoi.vector); 1719 break; 1720 case VM_EXITCODE_RENDEZVOUS: 1721 error = vm_handle_rendezvous(vcpu); 1722 break; 1723 case VM_EXITCODE_HLT: 1724 intr_disabled = ((vme->u.hlt.rflags & PSL_I) == 0); 1725 error = vm_handle_hlt(vcpu, intr_disabled, &retu); 1726 break; 1727 case VM_EXITCODE_PAGING: 1728 error = vm_handle_paging(vcpu, &retu); 1729 break; 1730 case VM_EXITCODE_INST_EMUL: 1731 error = vm_handle_inst_emul(vcpu, &retu); 1732 break; 1733 case VM_EXITCODE_INOUT: 1734 case VM_EXITCODE_INOUT_STR: 1735 error = vm_handle_inout(vcpu, vme, &retu); 1736 break; 1737 case VM_EXITCODE_DB: 1738 error = vm_handle_db(vcpu, vme, &retu); 1739 break; 1740 case VM_EXITCODE_MONITOR: 1741 case VM_EXITCODE_MWAIT: 1742 case VM_EXITCODE_VMINSN: 1743 vm_inject_ud(vcpu); 1744 break; 1745 default: 1746 retu = true; /* handled in userland */ 1747 break; 1748 } 1749 } 1750 1751 /* 1752 * VM_EXITCODE_INST_EMUL could access the apic which could transform the 1753 * exit code into VM_EXITCODE_IPI. 1754 */ 1755 if (error == 0 && vme->exitcode == VM_EXITCODE_IPI) 1756 error = vm_handle_ipi(vcpu, vme, &retu); 1757 1758 if (error == 0 && retu == false) 1759 goto restart; 1760 1761 vmm_stat_incr(vcpu, VMEXIT_USERSPACE, 1); 1762 VMM_CTR2(vcpu, "retu %d/%d", error, vme->exitcode); 1763 1764 return (error); 1765 } 1766 1767 int 1768 vm_restart_instruction(struct vcpu *vcpu) 1769 { 1770 enum vcpu_state state; 1771 uint64_t rip; 1772 int error __diagused; 1773 1774 state = vcpu_get_state(vcpu, NULL); 1775 if (state == VCPU_RUNNING) { 1776 /* 1777 * When a vcpu is "running" the next instruction is determined 1778 * by adding 'rip' and 'inst_length' in the vcpu's 'exitinfo'. 1779 * Thus setting 'inst_length' to zero will cause the current 1780 * instruction to be restarted. 1781 */ 1782 vcpu->exitinfo.inst_length = 0; 1783 VMM_CTR1(vcpu, "restarting instruction at %#lx by " 1784 "setting inst_length to zero", vcpu->exitinfo.rip); 1785 } else if (state == VCPU_FROZEN) { 1786 /* 1787 * When a vcpu is "frozen" it is outside the critical section 1788 * around vmmops_run() and 'nextrip' points to the next 1789 * instruction. Thus instruction restart is achieved by setting 1790 * 'nextrip' to the vcpu's %rip. 1791 */ 1792 error = vm_get_register(vcpu, VM_REG_GUEST_RIP, &rip); 1793 KASSERT(!error, ("%s: error %d getting rip", __func__, error)); 1794 VMM_CTR2(vcpu, "restarting instruction by updating " 1795 "nextrip from %#lx to %#lx", vcpu->nextrip, rip); 1796 vcpu->nextrip = rip; 1797 } else { 1798 panic("%s: invalid state %d", __func__, state); 1799 } 1800 return (0); 1801 } 1802 1803 int 1804 vm_exit_intinfo(struct vcpu *vcpu, uint64_t info) 1805 { 1806 int type, vector; 1807 1808 if (info & VM_INTINFO_VALID) { 1809 type = info & VM_INTINFO_TYPE; 1810 vector = info & 0xff; 1811 if (type == VM_INTINFO_NMI && vector != IDT_NMI) 1812 return (EINVAL); 1813 if (type == VM_INTINFO_HWEXCEPTION && vector >= 32) 1814 return (EINVAL); 1815 if (info & VM_INTINFO_RSVD) 1816 return (EINVAL); 1817 } else { 1818 info = 0; 1819 } 1820 VMM_CTR2(vcpu, "%s: info1(%#lx)", __func__, info); 1821 vcpu->exitintinfo = info; 1822 return (0); 1823 } 1824 1825 enum exc_class { 1826 EXC_BENIGN, 1827 EXC_CONTRIBUTORY, 1828 EXC_PAGEFAULT 1829 }; 1830 1831 #define IDT_VE 20 /* Virtualization Exception (Intel specific) */ 1832 1833 static enum exc_class 1834 exception_class(uint64_t info) 1835 { 1836 int type, vector; 1837 1838 KASSERT(info & VM_INTINFO_VALID, ("intinfo must be valid: %#lx", info)); 1839 type = info & VM_INTINFO_TYPE; 1840 vector = info & 0xff; 1841 1842 /* Table 6-4, "Interrupt and Exception Classes", Intel SDM, Vol 3 */ 1843 switch (type) { 1844 case VM_INTINFO_HWINTR: 1845 case VM_INTINFO_SWINTR: 1846 case VM_INTINFO_NMI: 1847 return (EXC_BENIGN); 1848 default: 1849 /* 1850 * Hardware exception. 1851 * 1852 * SVM and VT-x use identical type values to represent NMI, 1853 * hardware interrupt and software interrupt. 1854 * 1855 * SVM uses type '3' for all exceptions. VT-x uses type '3' 1856 * for exceptions except #BP and #OF. #BP and #OF use a type 1857 * value of '5' or '6'. Therefore we don't check for explicit 1858 * values of 'type' to classify 'intinfo' into a hardware 1859 * exception. 1860 */ 1861 break; 1862 } 1863 1864 switch (vector) { 1865 case IDT_PF: 1866 case IDT_VE: 1867 return (EXC_PAGEFAULT); 1868 case IDT_DE: 1869 case IDT_TS: 1870 case IDT_NP: 1871 case IDT_SS: 1872 case IDT_GP: 1873 return (EXC_CONTRIBUTORY); 1874 default: 1875 return (EXC_BENIGN); 1876 } 1877 } 1878 1879 static int 1880 nested_fault(struct vcpu *vcpu, uint64_t info1, uint64_t info2, 1881 uint64_t *retinfo) 1882 { 1883 enum exc_class exc1, exc2; 1884 int type1, vector1; 1885 1886 KASSERT(info1 & VM_INTINFO_VALID, ("info1 %#lx is not valid", info1)); 1887 KASSERT(info2 & VM_INTINFO_VALID, ("info2 %#lx is not valid", info2)); 1888 1889 /* 1890 * If an exception occurs while attempting to call the double-fault 1891 * handler the processor enters shutdown mode (aka triple fault). 1892 */ 1893 type1 = info1 & VM_INTINFO_TYPE; 1894 vector1 = info1 & 0xff; 1895 if (type1 == VM_INTINFO_HWEXCEPTION && vector1 == IDT_DF) { 1896 VMM_CTR2(vcpu, "triple fault: info1(%#lx), info2(%#lx)", 1897 info1, info2); 1898 vm_suspend(vcpu->vm, VM_SUSPEND_TRIPLEFAULT); 1899 *retinfo = 0; 1900 return (0); 1901 } 1902 1903 /* 1904 * Table 6-5 "Conditions for Generating a Double Fault", Intel SDM, Vol3 1905 */ 1906 exc1 = exception_class(info1); 1907 exc2 = exception_class(info2); 1908 if ((exc1 == EXC_CONTRIBUTORY && exc2 == EXC_CONTRIBUTORY) || 1909 (exc1 == EXC_PAGEFAULT && exc2 != EXC_BENIGN)) { 1910 /* Convert nested fault into a double fault. */ 1911 *retinfo = IDT_DF; 1912 *retinfo |= VM_INTINFO_VALID | VM_INTINFO_HWEXCEPTION; 1913 *retinfo |= VM_INTINFO_DEL_ERRCODE; 1914 } else { 1915 /* Handle exceptions serially */ 1916 *retinfo = info2; 1917 } 1918 return (1); 1919 } 1920 1921 static uint64_t 1922 vcpu_exception_intinfo(struct vcpu *vcpu) 1923 { 1924 uint64_t info = 0; 1925 1926 if (vcpu->exception_pending) { 1927 info = vcpu->exc_vector & 0xff; 1928 info |= VM_INTINFO_VALID | VM_INTINFO_HWEXCEPTION; 1929 if (vcpu->exc_errcode_valid) { 1930 info |= VM_INTINFO_DEL_ERRCODE; 1931 info |= (uint64_t)vcpu->exc_errcode << 32; 1932 } 1933 } 1934 return (info); 1935 } 1936 1937 int 1938 vm_entry_intinfo(struct vcpu *vcpu, uint64_t *retinfo) 1939 { 1940 uint64_t info1, info2; 1941 int valid; 1942 1943 info1 = vcpu->exitintinfo; 1944 vcpu->exitintinfo = 0; 1945 1946 info2 = 0; 1947 if (vcpu->exception_pending) { 1948 info2 = vcpu_exception_intinfo(vcpu); 1949 vcpu->exception_pending = 0; 1950 VMM_CTR2(vcpu, "Exception %d delivered: %#lx", 1951 vcpu->exc_vector, info2); 1952 } 1953 1954 if ((info1 & VM_INTINFO_VALID) && (info2 & VM_INTINFO_VALID)) { 1955 valid = nested_fault(vcpu, info1, info2, retinfo); 1956 } else if (info1 & VM_INTINFO_VALID) { 1957 *retinfo = info1; 1958 valid = 1; 1959 } else if (info2 & VM_INTINFO_VALID) { 1960 *retinfo = info2; 1961 valid = 1; 1962 } else { 1963 valid = 0; 1964 } 1965 1966 if (valid) { 1967 VMM_CTR4(vcpu, "%s: info1(%#lx), info2(%#lx), " 1968 "retinfo(%#lx)", __func__, info1, info2, *retinfo); 1969 } 1970 1971 return (valid); 1972 } 1973 1974 int 1975 vm_get_intinfo(struct vcpu *vcpu, uint64_t *info1, uint64_t *info2) 1976 { 1977 *info1 = vcpu->exitintinfo; 1978 *info2 = vcpu_exception_intinfo(vcpu); 1979 return (0); 1980 } 1981 1982 int 1983 vm_inject_exception(struct vcpu *vcpu, int vector, int errcode_valid, 1984 uint32_t errcode, int restart_instruction) 1985 { 1986 uint64_t regval; 1987 int error __diagused; 1988 1989 if (vector < 0 || vector >= 32) 1990 return (EINVAL); 1991 1992 /* 1993 * A double fault exception should never be injected directly into 1994 * the guest. It is a derived exception that results from specific 1995 * combinations of nested faults. 1996 */ 1997 if (vector == IDT_DF) 1998 return (EINVAL); 1999 2000 if (vcpu->exception_pending) { 2001 VMM_CTR2(vcpu, "Unable to inject exception %d due to " 2002 "pending exception %d", vector, vcpu->exc_vector); 2003 return (EBUSY); 2004 } 2005 2006 if (errcode_valid) { 2007 /* 2008 * Exceptions don't deliver an error code in real mode. 2009 */ 2010 error = vm_get_register(vcpu, VM_REG_GUEST_CR0, ®val); 2011 KASSERT(!error, ("%s: error %d getting CR0", __func__, error)); 2012 if (!(regval & CR0_PE)) 2013 errcode_valid = 0; 2014 } 2015 2016 /* 2017 * From section 26.6.1 "Interruptibility State" in Intel SDM: 2018 * 2019 * Event blocking by "STI" or "MOV SS" is cleared after guest executes 2020 * one instruction or incurs an exception. 2021 */ 2022 error = vm_set_register(vcpu, VM_REG_GUEST_INTR_SHADOW, 0); 2023 KASSERT(error == 0, ("%s: error %d clearing interrupt shadow", 2024 __func__, error)); 2025 2026 if (restart_instruction) 2027 vm_restart_instruction(vcpu); 2028 2029 vcpu->exception_pending = 1; 2030 vcpu->exc_vector = vector; 2031 vcpu->exc_errcode = errcode; 2032 vcpu->exc_errcode_valid = errcode_valid; 2033 VMM_CTR1(vcpu, "Exception %d pending", vector); 2034 return (0); 2035 } 2036 2037 void 2038 vm_inject_fault(struct vcpu *vcpu, int vector, int errcode_valid, int errcode) 2039 { 2040 int error __diagused, restart_instruction; 2041 2042 restart_instruction = 1; 2043 2044 error = vm_inject_exception(vcpu, vector, errcode_valid, 2045 errcode, restart_instruction); 2046 KASSERT(error == 0, ("vm_inject_exception error %d", error)); 2047 } 2048 2049 void 2050 vm_inject_pf(struct vcpu *vcpu, int error_code, uint64_t cr2) 2051 { 2052 int error __diagused; 2053 2054 VMM_CTR2(vcpu, "Injecting page fault: error_code %#x, cr2 %#lx", 2055 error_code, cr2); 2056 2057 error = vm_set_register(vcpu, VM_REG_GUEST_CR2, cr2); 2058 KASSERT(error == 0, ("vm_set_register(cr2) error %d", error)); 2059 2060 vm_inject_fault(vcpu, IDT_PF, 1, error_code); 2061 } 2062 2063 static VMM_STAT(VCPU_NMI_COUNT, "number of NMIs delivered to vcpu"); 2064 2065 int 2066 vm_inject_nmi(struct vcpu *vcpu) 2067 { 2068 2069 vcpu->nmi_pending = 1; 2070 vcpu_notify_event(vcpu, false); 2071 return (0); 2072 } 2073 2074 int 2075 vm_nmi_pending(struct vcpu *vcpu) 2076 { 2077 return (vcpu->nmi_pending); 2078 } 2079 2080 void 2081 vm_nmi_clear(struct vcpu *vcpu) 2082 { 2083 if (vcpu->nmi_pending == 0) 2084 panic("vm_nmi_clear: inconsistent nmi_pending state"); 2085 2086 vcpu->nmi_pending = 0; 2087 vmm_stat_incr(vcpu, VCPU_NMI_COUNT, 1); 2088 } 2089 2090 static VMM_STAT(VCPU_EXTINT_COUNT, "number of ExtINTs delivered to vcpu"); 2091 2092 int 2093 vm_inject_extint(struct vcpu *vcpu) 2094 { 2095 2096 vcpu->extint_pending = 1; 2097 vcpu_notify_event(vcpu, false); 2098 return (0); 2099 } 2100 2101 int 2102 vm_extint_pending(struct vcpu *vcpu) 2103 { 2104 return (vcpu->extint_pending); 2105 } 2106 2107 void 2108 vm_extint_clear(struct vcpu *vcpu) 2109 { 2110 if (vcpu->extint_pending == 0) 2111 panic("vm_extint_clear: inconsistent extint_pending state"); 2112 2113 vcpu->extint_pending = 0; 2114 vmm_stat_incr(vcpu, VCPU_EXTINT_COUNT, 1); 2115 } 2116 2117 int 2118 vm_get_capability(struct vcpu *vcpu, int type, int *retval) 2119 { 2120 if (type < 0 || type >= VM_CAP_MAX) 2121 return (EINVAL); 2122 2123 return (vmmops_getcap(vcpu->cookie, type, retval)); 2124 } 2125 2126 int 2127 vm_set_capability(struct vcpu *vcpu, int type, int val) 2128 { 2129 if (type < 0 || type >= VM_CAP_MAX) 2130 return (EINVAL); 2131 2132 return (vmmops_setcap(vcpu->cookie, type, val)); 2133 } 2134 2135 struct vm * 2136 vcpu_vm(struct vcpu *vcpu) 2137 { 2138 return (vcpu->vm); 2139 } 2140 2141 int 2142 vcpu_vcpuid(struct vcpu *vcpu) 2143 { 2144 return (vcpu->vcpuid); 2145 } 2146 2147 struct vcpu * 2148 vm_vcpu(struct vm *vm, int vcpuid) 2149 { 2150 return (vm->vcpu[vcpuid]); 2151 } 2152 2153 struct vlapic * 2154 vm_lapic(struct vcpu *vcpu) 2155 { 2156 return (vcpu->vlapic); 2157 } 2158 2159 struct vioapic * 2160 vm_ioapic(struct vm *vm) 2161 { 2162 2163 return (vm->vioapic); 2164 } 2165 2166 struct vhpet * 2167 vm_hpet(struct vm *vm) 2168 { 2169 2170 return (vm->vhpet); 2171 } 2172 2173 bool 2174 vmm_is_pptdev(int bus, int slot, int func) 2175 { 2176 int b, f, i, n, s; 2177 char *val, *cp, *cp2; 2178 bool found; 2179 2180 /* 2181 * XXX 2182 * The length of an environment variable is limited to 128 bytes which 2183 * puts an upper limit on the number of passthru devices that may be 2184 * specified using a single environment variable. 2185 * 2186 * Work around this by scanning multiple environment variable 2187 * names instead of a single one - yuck! 2188 */ 2189 const char *names[] = { "pptdevs", "pptdevs2", "pptdevs3", NULL }; 2190 2191 /* set pptdevs="1/2/3 4/5/6 7/8/9 10/11/12" */ 2192 found = false; 2193 for (i = 0; names[i] != NULL && !found; i++) { 2194 cp = val = kern_getenv(names[i]); 2195 while (cp != NULL && *cp != '\0') { 2196 if ((cp2 = strchr(cp, ' ')) != NULL) 2197 *cp2 = '\0'; 2198 2199 n = sscanf(cp, "%d/%d/%d", &b, &s, &f); 2200 if (n == 3 && bus == b && slot == s && func == f) { 2201 found = true; 2202 break; 2203 } 2204 2205 if (cp2 != NULL) 2206 *cp2++ = ' '; 2207 2208 cp = cp2; 2209 } 2210 freeenv(val); 2211 } 2212 return (found); 2213 } 2214 2215 void * 2216 vm_iommu_domain(struct vm *vm) 2217 { 2218 2219 return (vm->iommu); 2220 } 2221 2222 int 2223 vcpu_set_state(struct vcpu *vcpu, enum vcpu_state newstate, bool from_idle) 2224 { 2225 int error; 2226 2227 vcpu_lock(vcpu); 2228 error = vcpu_set_state_locked(vcpu, newstate, from_idle); 2229 vcpu_unlock(vcpu); 2230 2231 return (error); 2232 } 2233 2234 enum vcpu_state 2235 vcpu_get_state(struct vcpu *vcpu, int *hostcpu) 2236 { 2237 enum vcpu_state state; 2238 2239 vcpu_lock(vcpu); 2240 state = vcpu->state; 2241 if (hostcpu != NULL) 2242 *hostcpu = vcpu->hostcpu; 2243 vcpu_unlock(vcpu); 2244 2245 return (state); 2246 } 2247 2248 int 2249 vm_activate_cpu(struct vcpu *vcpu) 2250 { 2251 struct vm *vm = vcpu->vm; 2252 2253 if (CPU_ISSET(vcpu->vcpuid, &vm->active_cpus)) 2254 return (EBUSY); 2255 2256 VMM_CTR0(vcpu, "activated"); 2257 CPU_SET_ATOMIC(vcpu->vcpuid, &vm->active_cpus); 2258 return (0); 2259 } 2260 2261 int 2262 vm_suspend_cpu(struct vm *vm, struct vcpu *vcpu) 2263 { 2264 if (vcpu == NULL) { 2265 vm->debug_cpus = vm->active_cpus; 2266 for (int i = 0; i < vm->maxcpus; i++) { 2267 if (CPU_ISSET(i, &vm->active_cpus)) 2268 vcpu_notify_event(vm_vcpu(vm, i), false); 2269 } 2270 } else { 2271 if (!CPU_ISSET(vcpu->vcpuid, &vm->active_cpus)) 2272 return (EINVAL); 2273 2274 CPU_SET_ATOMIC(vcpu->vcpuid, &vm->debug_cpus); 2275 vcpu_notify_event(vcpu, false); 2276 } 2277 return (0); 2278 } 2279 2280 int 2281 vm_resume_cpu(struct vm *vm, struct vcpu *vcpu) 2282 { 2283 2284 if (vcpu == NULL) { 2285 CPU_ZERO(&vm->debug_cpus); 2286 } else { 2287 if (!CPU_ISSET(vcpu->vcpuid, &vm->debug_cpus)) 2288 return (EINVAL); 2289 2290 CPU_CLR_ATOMIC(vcpu->vcpuid, &vm->debug_cpus); 2291 } 2292 return (0); 2293 } 2294 2295 int 2296 vcpu_debugged(struct vcpu *vcpu) 2297 { 2298 2299 return (CPU_ISSET(vcpu->vcpuid, &vcpu->vm->debug_cpus)); 2300 } 2301 2302 cpuset_t 2303 vm_active_cpus(struct vm *vm) 2304 { 2305 2306 return (vm->active_cpus); 2307 } 2308 2309 cpuset_t 2310 vm_debug_cpus(struct vm *vm) 2311 { 2312 2313 return (vm->debug_cpus); 2314 } 2315 2316 cpuset_t 2317 vm_suspended_cpus(struct vm *vm) 2318 { 2319 2320 return (vm->suspended_cpus); 2321 } 2322 2323 /* 2324 * Returns the subset of vCPUs in tostart that are awaiting startup. 2325 * These vCPUs are also marked as no longer awaiting startup. 2326 */ 2327 cpuset_t 2328 vm_start_cpus(struct vm *vm, const cpuset_t *tostart) 2329 { 2330 cpuset_t set; 2331 2332 mtx_lock(&vm->rendezvous_mtx); 2333 CPU_AND(&set, &vm->startup_cpus, tostart); 2334 CPU_ANDNOT(&vm->startup_cpus, &vm->startup_cpus, &set); 2335 mtx_unlock(&vm->rendezvous_mtx); 2336 return (set); 2337 } 2338 2339 void 2340 vm_await_start(struct vm *vm, const cpuset_t *waiting) 2341 { 2342 mtx_lock(&vm->rendezvous_mtx); 2343 CPU_OR(&vm->startup_cpus, &vm->startup_cpus, waiting); 2344 mtx_unlock(&vm->rendezvous_mtx); 2345 } 2346 2347 void * 2348 vcpu_stats(struct vcpu *vcpu) 2349 { 2350 2351 return (vcpu->stats); 2352 } 2353 2354 int 2355 vm_get_x2apic_state(struct vcpu *vcpu, enum x2apic_state *state) 2356 { 2357 *state = vcpu->x2apic_state; 2358 2359 return (0); 2360 } 2361 2362 int 2363 vm_set_x2apic_state(struct vcpu *vcpu, enum x2apic_state state) 2364 { 2365 if (state >= X2APIC_STATE_LAST) 2366 return (EINVAL); 2367 2368 vcpu->x2apic_state = state; 2369 2370 vlapic_set_x2apic_state(vcpu, state); 2371 2372 return (0); 2373 } 2374 2375 /* 2376 * This function is called to ensure that a vcpu "sees" a pending event 2377 * as soon as possible: 2378 * - If the vcpu thread is sleeping then it is woken up. 2379 * - If the vcpu is running on a different host_cpu then an IPI will be directed 2380 * to the host_cpu to cause the vcpu to trap into the hypervisor. 2381 */ 2382 static void 2383 vcpu_notify_event_locked(struct vcpu *vcpu, bool lapic_intr) 2384 { 2385 int hostcpu; 2386 2387 hostcpu = vcpu->hostcpu; 2388 if (vcpu->state == VCPU_RUNNING) { 2389 KASSERT(hostcpu != NOCPU, ("vcpu running on invalid hostcpu")); 2390 if (hostcpu != curcpu) { 2391 if (lapic_intr) { 2392 vlapic_post_intr(vcpu->vlapic, hostcpu, 2393 vmm_ipinum); 2394 } else { 2395 ipi_cpu(hostcpu, vmm_ipinum); 2396 } 2397 } else { 2398 /* 2399 * If the 'vcpu' is running on 'curcpu' then it must 2400 * be sending a notification to itself (e.g. SELF_IPI). 2401 * The pending event will be picked up when the vcpu 2402 * transitions back to guest context. 2403 */ 2404 } 2405 } else { 2406 KASSERT(hostcpu == NOCPU, ("vcpu state %d not consistent " 2407 "with hostcpu %d", vcpu->state, hostcpu)); 2408 if (vcpu->state == VCPU_SLEEPING) 2409 wakeup_one(vcpu); 2410 } 2411 } 2412 2413 void 2414 vcpu_notify_event(struct vcpu *vcpu, bool lapic_intr) 2415 { 2416 vcpu_lock(vcpu); 2417 vcpu_notify_event_locked(vcpu, lapic_intr); 2418 vcpu_unlock(vcpu); 2419 } 2420 2421 struct vm_mem * 2422 vm_mem(struct vm *vm) 2423 { 2424 return (&vm->mem); 2425 } 2426 2427 int 2428 vm_apicid2vcpuid(struct vm *vm, int apicid) 2429 { 2430 /* 2431 * XXX apic id is assumed to be numerically identical to vcpu id 2432 */ 2433 return (apicid); 2434 } 2435 2436 int 2437 vm_smp_rendezvous(struct vcpu *vcpu, cpuset_t dest, 2438 vm_rendezvous_func_t func, void *arg) 2439 { 2440 struct vm *vm = vcpu->vm; 2441 int error, i; 2442 2443 /* 2444 * Enforce that this function is called without any locks 2445 */ 2446 WITNESS_WARN(WARN_PANIC, NULL, "vm_smp_rendezvous"); 2447 2448 restart: 2449 mtx_lock(&vm->rendezvous_mtx); 2450 if (vm->rendezvous_func != NULL) { 2451 /* 2452 * If a rendezvous is already in progress then we need to 2453 * call the rendezvous handler in case this 'vcpu' is one 2454 * of the targets of the rendezvous. 2455 */ 2456 VMM_CTR0(vcpu, "Rendezvous already in progress"); 2457 mtx_unlock(&vm->rendezvous_mtx); 2458 error = vm_handle_rendezvous(vcpu); 2459 if (error != 0) 2460 return (error); 2461 goto restart; 2462 } 2463 KASSERT(vm->rendezvous_func == NULL, ("vm_smp_rendezvous: previous " 2464 "rendezvous is still in progress")); 2465 2466 VMM_CTR0(vcpu, "Initiating rendezvous"); 2467 vm->rendezvous_req_cpus = dest; 2468 CPU_ZERO(&vm->rendezvous_done_cpus); 2469 vm->rendezvous_arg = arg; 2470 vm->rendezvous_func = func; 2471 mtx_unlock(&vm->rendezvous_mtx); 2472 2473 /* 2474 * Wake up any sleeping vcpus and trigger a VM-exit in any running 2475 * vcpus so they handle the rendezvous as soon as possible. 2476 */ 2477 for (i = 0; i < vm->maxcpus; i++) { 2478 if (CPU_ISSET(i, &dest)) 2479 vcpu_notify_event(vm_vcpu(vm, i), false); 2480 } 2481 2482 return (vm_handle_rendezvous(vcpu)); 2483 } 2484 2485 struct vatpic * 2486 vm_atpic(struct vm *vm) 2487 { 2488 return (vm->vatpic); 2489 } 2490 2491 struct vatpit * 2492 vm_atpit(struct vm *vm) 2493 { 2494 return (vm->vatpit); 2495 } 2496 2497 struct vpmtmr * 2498 vm_pmtmr(struct vm *vm) 2499 { 2500 2501 return (vm->vpmtmr); 2502 } 2503 2504 struct vrtc * 2505 vm_rtc(struct vm *vm) 2506 { 2507 2508 return (vm->vrtc); 2509 } 2510 2511 enum vm_reg_name 2512 vm_segment_name(int seg) 2513 { 2514 static enum vm_reg_name seg_names[] = { 2515 VM_REG_GUEST_ES, 2516 VM_REG_GUEST_CS, 2517 VM_REG_GUEST_SS, 2518 VM_REG_GUEST_DS, 2519 VM_REG_GUEST_FS, 2520 VM_REG_GUEST_GS 2521 }; 2522 2523 KASSERT(seg >= 0 && seg < nitems(seg_names), 2524 ("%s: invalid segment encoding %d", __func__, seg)); 2525 return (seg_names[seg]); 2526 } 2527 2528 void 2529 vm_copy_teardown(struct vm_copyinfo *copyinfo, int num_copyinfo) 2530 { 2531 int idx; 2532 2533 for (idx = 0; idx < num_copyinfo; idx++) { 2534 if (copyinfo[idx].cookie != NULL) 2535 vm_gpa_release(copyinfo[idx].cookie); 2536 } 2537 bzero(copyinfo, num_copyinfo * sizeof(struct vm_copyinfo)); 2538 } 2539 2540 int 2541 vm_copy_setup(struct vcpu *vcpu, struct vm_guest_paging *paging, 2542 uint64_t gla, size_t len, int prot, struct vm_copyinfo *copyinfo, 2543 int num_copyinfo, int *fault) 2544 { 2545 int error, idx, nused; 2546 size_t n, off, remaining; 2547 void *hva, *cookie; 2548 uint64_t gpa; 2549 2550 bzero(copyinfo, sizeof(struct vm_copyinfo) * num_copyinfo); 2551 2552 nused = 0; 2553 remaining = len; 2554 while (remaining > 0) { 2555 if (nused >= num_copyinfo) 2556 return (EFAULT); 2557 error = vm_gla2gpa(vcpu, paging, gla, prot, &gpa, fault); 2558 if (error || *fault) 2559 return (error); 2560 off = gpa & PAGE_MASK; 2561 n = min(remaining, PAGE_SIZE - off); 2562 copyinfo[nused].gpa = gpa; 2563 copyinfo[nused].len = n; 2564 remaining -= n; 2565 gla += n; 2566 nused++; 2567 } 2568 2569 for (idx = 0; idx < nused; idx++) { 2570 hva = vm_gpa_hold(vcpu, copyinfo[idx].gpa, 2571 copyinfo[idx].len, prot, &cookie); 2572 if (hva == NULL) 2573 break; 2574 copyinfo[idx].hva = hva; 2575 copyinfo[idx].cookie = cookie; 2576 } 2577 2578 if (idx != nused) { 2579 vm_copy_teardown(copyinfo, num_copyinfo); 2580 return (EFAULT); 2581 } else { 2582 *fault = 0; 2583 return (0); 2584 } 2585 } 2586 2587 void 2588 vm_copyin(struct vm_copyinfo *copyinfo, void *kaddr, size_t len) 2589 { 2590 char *dst; 2591 int idx; 2592 2593 dst = kaddr; 2594 idx = 0; 2595 while (len > 0) { 2596 bcopy(copyinfo[idx].hva, dst, copyinfo[idx].len); 2597 len -= copyinfo[idx].len; 2598 dst += copyinfo[idx].len; 2599 idx++; 2600 } 2601 } 2602 2603 void 2604 vm_copyout(const void *kaddr, struct vm_copyinfo *copyinfo, size_t len) 2605 { 2606 const char *src; 2607 int idx; 2608 2609 src = kaddr; 2610 idx = 0; 2611 while (len > 0) { 2612 bcopy(src, copyinfo[idx].hva, copyinfo[idx].len); 2613 len -= copyinfo[idx].len; 2614 src += copyinfo[idx].len; 2615 idx++; 2616 } 2617 } 2618 2619 /* 2620 * Return the amount of in-use and wired memory for the VM. Since 2621 * these are global stats, only return the values with for vCPU 0 2622 */ 2623 VMM_STAT_DECLARE(VMM_MEM_RESIDENT); 2624 VMM_STAT_DECLARE(VMM_MEM_WIRED); 2625 2626 static void 2627 vm_get_rescnt(struct vcpu *vcpu, struct vmm_stat_type *stat) 2628 { 2629 2630 if (vcpu->vcpuid == 0) { 2631 vmm_stat_set(vcpu, VMM_MEM_RESIDENT, PAGE_SIZE * 2632 vmspace_resident_count(vm_vmspace(vcpu->vm))); 2633 } 2634 } 2635 2636 static void 2637 vm_get_wiredcnt(struct vcpu *vcpu, struct vmm_stat_type *stat) 2638 { 2639 2640 if (vcpu->vcpuid == 0) { 2641 vmm_stat_set(vcpu, VMM_MEM_WIRED, PAGE_SIZE * 2642 pmap_wired_count(vmspace_pmap(vm_vmspace(vcpu->vm)))); 2643 } 2644 } 2645 2646 VMM_STAT_FUNC(VMM_MEM_RESIDENT, "Resident memory", vm_get_rescnt); 2647 VMM_STAT_FUNC(VMM_MEM_WIRED, "Wired memory", vm_get_wiredcnt); 2648 2649 #ifdef BHYVE_SNAPSHOT 2650 static int 2651 vm_snapshot_vcpus(struct vm *vm, struct vm_snapshot_meta *meta) 2652 { 2653 uint64_t tsc, now; 2654 int ret; 2655 struct vcpu *vcpu; 2656 uint16_t i, maxcpus; 2657 2658 now = rdtsc(); 2659 maxcpus = vm_get_maxcpus(vm); 2660 for (i = 0; i < maxcpus; i++) { 2661 vcpu = vm->vcpu[i]; 2662 if (vcpu == NULL) 2663 continue; 2664 2665 SNAPSHOT_VAR_OR_LEAVE(vcpu->x2apic_state, meta, ret, done); 2666 SNAPSHOT_VAR_OR_LEAVE(vcpu->exitintinfo, meta, ret, done); 2667 SNAPSHOT_VAR_OR_LEAVE(vcpu->exc_vector, meta, ret, done); 2668 SNAPSHOT_VAR_OR_LEAVE(vcpu->exc_errcode_valid, meta, ret, done); 2669 SNAPSHOT_VAR_OR_LEAVE(vcpu->exc_errcode, meta, ret, done); 2670 SNAPSHOT_VAR_OR_LEAVE(vcpu->guest_xcr0, meta, ret, done); 2671 SNAPSHOT_VAR_OR_LEAVE(vcpu->exitinfo, meta, ret, done); 2672 SNAPSHOT_VAR_OR_LEAVE(vcpu->nextrip, meta, ret, done); 2673 2674 /* 2675 * Save the absolute TSC value by adding now to tsc_offset. 2676 * 2677 * It will be turned turned back into an actual offset when the 2678 * TSC restore function is called 2679 */ 2680 tsc = now + vcpu->tsc_offset; 2681 SNAPSHOT_VAR_OR_LEAVE(tsc, meta, ret, done); 2682 if (meta->op == VM_SNAPSHOT_RESTORE) 2683 vcpu->tsc_offset = tsc; 2684 } 2685 2686 done: 2687 return (ret); 2688 } 2689 2690 static int 2691 vm_snapshot_vm(struct vm *vm, struct vm_snapshot_meta *meta) 2692 { 2693 int ret; 2694 2695 ret = vm_snapshot_vcpus(vm, meta); 2696 if (ret != 0) 2697 goto done; 2698 2699 SNAPSHOT_VAR_OR_LEAVE(vm->startup_cpus, meta, ret, done); 2700 done: 2701 return (ret); 2702 } 2703 2704 static int 2705 vm_snapshot_vcpu(struct vm *vm, struct vm_snapshot_meta *meta) 2706 { 2707 int error; 2708 struct vcpu *vcpu; 2709 uint16_t i, maxcpus; 2710 2711 error = 0; 2712 2713 maxcpus = vm_get_maxcpus(vm); 2714 for (i = 0; i < maxcpus; i++) { 2715 vcpu = vm->vcpu[i]; 2716 if (vcpu == NULL) 2717 continue; 2718 2719 error = vmmops_vcpu_snapshot(vcpu->cookie, meta); 2720 if (error != 0) { 2721 printf("%s: failed to snapshot vmcs/vmcb data for " 2722 "vCPU: %d; error: %d\n", __func__, i, error); 2723 goto done; 2724 } 2725 } 2726 2727 done: 2728 return (error); 2729 } 2730 2731 /* 2732 * Save kernel-side structures to user-space for snapshotting. 2733 */ 2734 int 2735 vm_snapshot_req(struct vm *vm, struct vm_snapshot_meta *meta) 2736 { 2737 int ret = 0; 2738 2739 switch (meta->dev_req) { 2740 case STRUCT_VMCX: 2741 ret = vm_snapshot_vcpu(vm, meta); 2742 break; 2743 case STRUCT_VM: 2744 ret = vm_snapshot_vm(vm, meta); 2745 break; 2746 case STRUCT_VIOAPIC: 2747 ret = vioapic_snapshot(vm_ioapic(vm), meta); 2748 break; 2749 case STRUCT_VLAPIC: 2750 ret = vlapic_snapshot(vm, meta); 2751 break; 2752 case STRUCT_VHPET: 2753 ret = vhpet_snapshot(vm_hpet(vm), meta); 2754 break; 2755 case STRUCT_VATPIC: 2756 ret = vatpic_snapshot(vm_atpic(vm), meta); 2757 break; 2758 case STRUCT_VATPIT: 2759 ret = vatpit_snapshot(vm_atpit(vm), meta); 2760 break; 2761 case STRUCT_VPMTMR: 2762 ret = vpmtmr_snapshot(vm_pmtmr(vm), meta); 2763 break; 2764 case STRUCT_VRTC: 2765 ret = vrtc_snapshot(vm_rtc(vm), meta); 2766 break; 2767 default: 2768 printf("%s: failed to find the requested type %#x\n", 2769 __func__, meta->dev_req); 2770 ret = (EINVAL); 2771 } 2772 return (ret); 2773 } 2774 2775 void 2776 vm_set_tsc_offset(struct vcpu *vcpu, uint64_t offset) 2777 { 2778 vcpu->tsc_offset = offset; 2779 } 2780 2781 int 2782 vm_restore_time(struct vm *vm) 2783 { 2784 int error; 2785 uint64_t now; 2786 struct vcpu *vcpu; 2787 uint16_t i, maxcpus; 2788 2789 now = rdtsc(); 2790 2791 error = vhpet_restore_time(vm_hpet(vm)); 2792 if (error) 2793 return (error); 2794 2795 maxcpus = vm_get_maxcpus(vm); 2796 for (i = 0; i < maxcpus; i++) { 2797 vcpu = vm->vcpu[i]; 2798 if (vcpu == NULL) 2799 continue; 2800 2801 error = vmmops_restore_tsc(vcpu->cookie, 2802 vcpu->tsc_offset - now); 2803 if (error) 2804 return (error); 2805 } 2806 2807 return (0); 2808 } 2809 #endif 2810