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 /* Negative values represent VM control structure fields. */ 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 /* Negative values represent VM control structure fields. */ 886 if (reg >= VM_REG_LAST) 887 return (EINVAL); 888 889 error = vmmops_setreg(vcpu->cookie, reg, val); 890 if (error || reg != VM_REG_GUEST_RIP) 891 return (error); 892 893 /* Set 'nextrip' to match the value of %rip */ 894 VMM_CTR1(vcpu, "Setting nextrip to %#lx", val); 895 vcpu->nextrip = val; 896 return (0); 897 } 898 899 static bool 900 is_descriptor_table(int reg) 901 { 902 903 switch (reg) { 904 case VM_REG_GUEST_IDTR: 905 case VM_REG_GUEST_GDTR: 906 return (true); 907 default: 908 return (false); 909 } 910 } 911 912 static bool 913 is_segment_register(int reg) 914 { 915 916 switch (reg) { 917 case VM_REG_GUEST_ES: 918 case VM_REG_GUEST_CS: 919 case VM_REG_GUEST_SS: 920 case VM_REG_GUEST_DS: 921 case VM_REG_GUEST_FS: 922 case VM_REG_GUEST_GS: 923 case VM_REG_GUEST_TR: 924 case VM_REG_GUEST_LDTR: 925 return (true); 926 default: 927 return (false); 928 } 929 } 930 931 int 932 vm_get_seg_desc(struct vcpu *vcpu, int reg, struct seg_desc *desc) 933 { 934 935 if (!is_segment_register(reg) && !is_descriptor_table(reg)) 936 return (EINVAL); 937 938 return (vmmops_getdesc(vcpu->cookie, reg, desc)); 939 } 940 941 int 942 vm_set_seg_desc(struct vcpu *vcpu, int reg, struct seg_desc *desc) 943 { 944 945 if (!is_segment_register(reg) && !is_descriptor_table(reg)) 946 return (EINVAL); 947 948 return (vmmops_setdesc(vcpu->cookie, reg, desc)); 949 } 950 951 static void 952 restore_guest_fpustate(struct vcpu *vcpu) 953 { 954 955 /* flush host state to the pcb */ 956 fpuexit(curthread); 957 958 /* restore guest FPU state */ 959 fpu_enable(); 960 fpurestore(vcpu->guestfpu); 961 962 /* restore guest XCR0 if XSAVE is enabled in the host */ 963 if (rcr4() & CR4_XSAVE) 964 load_xcr(0, vcpu->guest_xcr0); 965 966 /* 967 * The FPU is now "dirty" with the guest's state so disable 968 * the FPU to trap any access by the host. 969 */ 970 fpu_disable(); 971 } 972 973 static void 974 save_guest_fpustate(struct vcpu *vcpu) 975 { 976 977 if ((rcr0() & CR0_TS) == 0) 978 panic("fpu emulation not enabled in host!"); 979 980 /* save guest XCR0 and restore host XCR0 */ 981 if (rcr4() & CR4_XSAVE) { 982 vcpu->guest_xcr0 = rxcr(0); 983 load_xcr(0, vmm_get_host_xcr0()); 984 } 985 986 /* save guest FPU state */ 987 fpu_enable(); 988 fpusave(vcpu->guestfpu); 989 fpu_disable(); 990 } 991 992 static VMM_STAT(VCPU_IDLE_TICKS, "number of ticks vcpu was idle"); 993 994 /* 995 * Invoke the rendezvous function on the specified vcpu if applicable. Return 996 * true if the rendezvous is finished, false otherwise. 997 */ 998 static bool 999 vm_rendezvous(struct vcpu *vcpu) 1000 { 1001 struct vm *vm = vcpu->vm; 1002 int vcpuid; 1003 1004 mtx_assert(&vcpu->vm->rendezvous_mtx, MA_OWNED); 1005 KASSERT(vcpu->vm->rendezvous_func != NULL, 1006 ("vm_rendezvous: no rendezvous pending")); 1007 1008 /* 'rendezvous_req_cpus' must be a subset of 'active_cpus' */ 1009 CPU_AND(&vm->rendezvous_req_cpus, &vm->rendezvous_req_cpus, 1010 &vm->active_cpus); 1011 1012 vcpuid = vcpu->vcpuid; 1013 if (CPU_ISSET(vcpuid, &vm->rendezvous_req_cpus) && 1014 !CPU_ISSET(vcpuid, &vm->rendezvous_done_cpus)) { 1015 VMM_CTR0(vcpu, "Calling rendezvous func"); 1016 (*vm->rendezvous_func)(vcpu, vm->rendezvous_arg); 1017 CPU_SET(vcpuid, &vm->rendezvous_done_cpus); 1018 } 1019 if (CPU_CMP(&vm->rendezvous_req_cpus, 1020 &vm->rendezvous_done_cpus) == 0) { 1021 VMM_CTR0(vcpu, "Rendezvous completed"); 1022 CPU_ZERO(&vm->rendezvous_req_cpus); 1023 vm->rendezvous_func = NULL; 1024 wakeup(&vm->rendezvous_func); 1025 return (true); 1026 } 1027 return (false); 1028 } 1029 1030 static void 1031 vcpu_wait_idle(struct vcpu *vcpu) 1032 { 1033 KASSERT(vcpu->state != VCPU_IDLE, ("vcpu already idle")); 1034 1035 vcpu->reqidle = 1; 1036 vcpu_notify_event_locked(vcpu, false); 1037 VMM_CTR1(vcpu, "vcpu state change from %s to " 1038 "idle requested", vcpu_state2str(vcpu->state)); 1039 msleep_spin(&vcpu->state, &vcpu->mtx, "vmstat", hz); 1040 } 1041 1042 static int 1043 vcpu_set_state_locked(struct vcpu *vcpu, enum vcpu_state newstate, 1044 bool from_idle) 1045 { 1046 int error; 1047 1048 vcpu_assert_locked(vcpu); 1049 1050 /* 1051 * State transitions from the vmmdev_ioctl() must always begin from 1052 * the VCPU_IDLE state. This guarantees that there is only a single 1053 * ioctl() operating on a vcpu at any point. 1054 */ 1055 if (from_idle) { 1056 while (vcpu->state != VCPU_IDLE) 1057 vcpu_wait_idle(vcpu); 1058 } else { 1059 KASSERT(vcpu->state != VCPU_IDLE, ("invalid transition from " 1060 "vcpu idle state")); 1061 } 1062 1063 if (vcpu->state == VCPU_RUNNING) { 1064 KASSERT(vcpu->hostcpu == curcpu, ("curcpu %d and hostcpu %d " 1065 "mismatch for running vcpu", curcpu, vcpu->hostcpu)); 1066 } else { 1067 KASSERT(vcpu->hostcpu == NOCPU, ("Invalid hostcpu %d for a " 1068 "vcpu that is not running", vcpu->hostcpu)); 1069 } 1070 1071 /* 1072 * The following state transitions are allowed: 1073 * IDLE -> FROZEN -> IDLE 1074 * FROZEN -> RUNNING -> FROZEN 1075 * FROZEN -> SLEEPING -> FROZEN 1076 */ 1077 switch (vcpu->state) { 1078 case VCPU_IDLE: 1079 case VCPU_RUNNING: 1080 case VCPU_SLEEPING: 1081 error = (newstate != VCPU_FROZEN); 1082 break; 1083 case VCPU_FROZEN: 1084 error = (newstate == VCPU_FROZEN); 1085 break; 1086 default: 1087 error = 1; 1088 break; 1089 } 1090 1091 if (error) 1092 return (EBUSY); 1093 1094 VMM_CTR2(vcpu, "vcpu state changed from %s to %s", 1095 vcpu_state2str(vcpu->state), vcpu_state2str(newstate)); 1096 1097 vcpu->state = newstate; 1098 if (newstate == VCPU_RUNNING) 1099 vcpu->hostcpu = curcpu; 1100 else 1101 vcpu->hostcpu = NOCPU; 1102 1103 if (newstate == VCPU_IDLE) 1104 wakeup(&vcpu->state); 1105 1106 return (0); 1107 } 1108 1109 /* 1110 * Try to lock all of the vCPUs in the VM while taking care to avoid deadlocks 1111 * with vm_smp_rendezvous(). 1112 * 1113 * The complexity here suggests that the rendezvous mechanism needs a rethink. 1114 */ 1115 int 1116 vcpu_set_state_all(struct vm *vm, enum vcpu_state newstate) 1117 { 1118 cpuset_t locked; 1119 struct vcpu *vcpu; 1120 int error, i; 1121 uint16_t maxcpus; 1122 1123 KASSERT(newstate != VCPU_IDLE, 1124 ("vcpu_set_state_all: invalid target state %d", newstate)); 1125 1126 error = 0; 1127 CPU_ZERO(&locked); 1128 maxcpus = vm->maxcpus; 1129 1130 mtx_lock(&vm->rendezvous_mtx); 1131 restart: 1132 if (vm->rendezvous_func != NULL) { 1133 /* 1134 * If we have a pending rendezvous, then the initiator may be 1135 * blocked waiting for other vCPUs to execute the callback. The 1136 * current thread may be a vCPU thread so we must not block 1137 * waiting for the initiator, otherwise we get a deadlock. 1138 * Thus, execute the callback on behalf of any idle vCPUs. 1139 */ 1140 for (i = 0; i < maxcpus; i++) { 1141 vcpu = vm_vcpu(vm, i); 1142 if (vcpu == NULL) 1143 continue; 1144 vcpu_lock(vcpu); 1145 if (vcpu->state == VCPU_IDLE) { 1146 (void)vcpu_set_state_locked(vcpu, VCPU_FROZEN, 1147 true); 1148 CPU_SET(i, &locked); 1149 } 1150 if (CPU_ISSET(i, &locked)) { 1151 /* 1152 * We can safely execute the callback on this 1153 * vCPU's behalf. 1154 */ 1155 vcpu_unlock(vcpu); 1156 (void)vm_rendezvous(vcpu); 1157 vcpu_lock(vcpu); 1158 } 1159 vcpu_unlock(vcpu); 1160 } 1161 } 1162 1163 /* 1164 * Now wait for remaining vCPUs to become idle. This may include the 1165 * initiator of a rendezvous that is currently blocked on the rendezvous 1166 * mutex. 1167 */ 1168 CPU_FOREACH_ISCLR(i, &locked) { 1169 if (i >= maxcpus) 1170 break; 1171 vcpu = vm_vcpu(vm, i); 1172 if (vcpu == NULL) 1173 continue; 1174 vcpu_lock(vcpu); 1175 while (vcpu->state != VCPU_IDLE) { 1176 mtx_unlock(&vm->rendezvous_mtx); 1177 vcpu_wait_idle(vcpu); 1178 vcpu_unlock(vcpu); 1179 mtx_lock(&vm->rendezvous_mtx); 1180 if (vm->rendezvous_func != NULL) 1181 goto restart; 1182 vcpu_lock(vcpu); 1183 } 1184 error = vcpu_set_state_locked(vcpu, newstate, true); 1185 vcpu_unlock(vcpu); 1186 if (error != 0) { 1187 /* Roll back state changes. */ 1188 CPU_FOREACH_ISSET(i, &locked) 1189 (void)vcpu_set_state(vcpu, VCPU_IDLE, false); 1190 break; 1191 } 1192 CPU_SET(i, &locked); 1193 } 1194 mtx_unlock(&vm->rendezvous_mtx); 1195 return (error); 1196 } 1197 1198 static void 1199 vcpu_require_state(struct vcpu *vcpu, enum vcpu_state newstate) 1200 { 1201 int error; 1202 1203 if ((error = vcpu_set_state(vcpu, newstate, false)) != 0) 1204 panic("Error %d setting state to %d\n", error, newstate); 1205 } 1206 1207 static void 1208 vcpu_require_state_locked(struct vcpu *vcpu, enum vcpu_state newstate) 1209 { 1210 int error; 1211 1212 if ((error = vcpu_set_state_locked(vcpu, newstate, false)) != 0) 1213 panic("Error %d setting state to %d", error, newstate); 1214 } 1215 1216 static int 1217 vm_handle_rendezvous(struct vcpu *vcpu) 1218 { 1219 struct vm *vm; 1220 struct thread *td; 1221 1222 td = curthread; 1223 vm = vcpu->vm; 1224 1225 mtx_lock(&vm->rendezvous_mtx); 1226 while (vm->rendezvous_func != NULL) { 1227 if (vm_rendezvous(vcpu)) 1228 break; 1229 1230 VMM_CTR0(vcpu, "Wait for rendezvous completion"); 1231 mtx_sleep(&vm->rendezvous_func, &vm->rendezvous_mtx, 0, 1232 "vmrndv", hz); 1233 if (td_ast_pending(td, TDA_SUSPEND)) { 1234 int error; 1235 1236 mtx_unlock(&vm->rendezvous_mtx); 1237 error = thread_check_susp(td, true); 1238 if (error != 0) 1239 return (error); 1240 mtx_lock(&vm->rendezvous_mtx); 1241 } 1242 } 1243 mtx_unlock(&vm->rendezvous_mtx); 1244 return (0); 1245 } 1246 1247 /* 1248 * Emulate a guest 'hlt' by sleeping until the vcpu is ready to run. 1249 */ 1250 static int 1251 vm_handle_hlt(struct vcpu *vcpu, bool intr_disabled, bool *retu) 1252 { 1253 struct vm *vm = vcpu->vm; 1254 const char *wmesg; 1255 struct thread *td; 1256 int error, t, vcpuid, vcpu_halted, vm_halted; 1257 1258 vcpuid = vcpu->vcpuid; 1259 vcpu_halted = 0; 1260 vm_halted = 0; 1261 error = 0; 1262 td = curthread; 1263 1264 KASSERT(!CPU_ISSET(vcpuid, &vm->halted_cpus), ("vcpu already halted")); 1265 1266 vcpu_lock(vcpu); 1267 while (1) { 1268 /* 1269 * Do a final check for pending NMI or interrupts before 1270 * really putting this thread to sleep. Also check for 1271 * software events that would cause this vcpu to wakeup. 1272 * 1273 * These interrupts/events could have happened after the 1274 * vcpu returned from vmmops_run() and before it acquired the 1275 * vcpu lock above. 1276 */ 1277 if (vm->rendezvous_func != NULL || vm->suspend || vcpu->reqidle) 1278 break; 1279 if (vm_nmi_pending(vcpu)) 1280 break; 1281 if (!intr_disabled) { 1282 if (vm_extint_pending(vcpu) || 1283 vlapic_pending_intr(vcpu->vlapic, NULL)) { 1284 break; 1285 } 1286 } 1287 1288 /* Don't go to sleep if the vcpu thread needs to yield */ 1289 if (vcpu_should_yield(vcpu)) 1290 break; 1291 1292 if (vcpu_debugged(vcpu)) 1293 break; 1294 1295 /* 1296 * Some Linux guests implement "halt" by having all vcpus 1297 * execute HLT with interrupts disabled. 'halted_cpus' keeps 1298 * track of the vcpus that have entered this state. When all 1299 * vcpus enter the halted state the virtual machine is halted. 1300 */ 1301 if (intr_disabled) { 1302 wmesg = "vmhalt"; 1303 VMM_CTR0(vcpu, "Halted"); 1304 if (!vcpu_halted && halt_detection_enabled) { 1305 vcpu_halted = 1; 1306 CPU_SET_ATOMIC(vcpuid, &vm->halted_cpus); 1307 } 1308 if (CPU_CMP(&vm->halted_cpus, &vm->active_cpus) == 0) { 1309 vm_halted = 1; 1310 break; 1311 } 1312 } else { 1313 wmesg = "vmidle"; 1314 } 1315 1316 t = ticks; 1317 vcpu_require_state_locked(vcpu, VCPU_SLEEPING); 1318 /* 1319 * XXX msleep_spin() cannot be interrupted by signals so 1320 * wake up periodically to check pending signals. 1321 */ 1322 msleep_spin(vcpu, &vcpu->mtx, wmesg, hz); 1323 vcpu_require_state_locked(vcpu, VCPU_FROZEN); 1324 vmm_stat_incr(vcpu, VCPU_IDLE_TICKS, ticks - t); 1325 if (td_ast_pending(td, TDA_SUSPEND)) { 1326 vcpu_unlock(vcpu); 1327 error = thread_check_susp(td, false); 1328 if (error != 0) { 1329 if (vcpu_halted) { 1330 CPU_CLR_ATOMIC(vcpuid, 1331 &vm->halted_cpus); 1332 } 1333 return (error); 1334 } 1335 vcpu_lock(vcpu); 1336 } 1337 } 1338 1339 if (vcpu_halted) 1340 CPU_CLR_ATOMIC(vcpuid, &vm->halted_cpus); 1341 1342 vcpu_unlock(vcpu); 1343 1344 if (vm_halted) 1345 vm_suspend(vm, VM_SUSPEND_HALT); 1346 1347 return (0); 1348 } 1349 1350 static int 1351 vm_handle_paging(struct vcpu *vcpu, bool *retu) 1352 { 1353 struct vm *vm = vcpu->vm; 1354 int rv, ftype; 1355 struct vm_map *map; 1356 struct vm_exit *vme; 1357 1358 vme = &vcpu->exitinfo; 1359 1360 KASSERT(vme->inst_length == 0, ("%s: invalid inst_length %d", 1361 __func__, vme->inst_length)); 1362 1363 ftype = vme->u.paging.fault_type; 1364 KASSERT(ftype == VM_PROT_READ || 1365 ftype == VM_PROT_WRITE || ftype == VM_PROT_EXECUTE, 1366 ("vm_handle_paging: invalid fault_type %d", ftype)); 1367 1368 if (ftype == VM_PROT_READ || ftype == VM_PROT_WRITE) { 1369 rv = pmap_emulate_accessed_dirty(vmspace_pmap(vm_vmspace(vm)), 1370 vme->u.paging.gpa, ftype); 1371 if (rv == 0) { 1372 VMM_CTR2(vcpu, "%s bit emulation for gpa %#lx", 1373 ftype == VM_PROT_READ ? "accessed" : "dirty", 1374 vme->u.paging.gpa); 1375 goto done; 1376 } 1377 } 1378 1379 map = &vm_vmspace(vm)->vm_map; 1380 rv = vm_fault(map, vme->u.paging.gpa, ftype, VM_FAULT_NORMAL, NULL); 1381 1382 VMM_CTR3(vcpu, "vm_handle_paging rv = %d, gpa = %#lx, " 1383 "ftype = %d", rv, vme->u.paging.gpa, ftype); 1384 1385 if (rv != KERN_SUCCESS) 1386 return (EFAULT); 1387 done: 1388 return (0); 1389 } 1390 1391 static int 1392 vm_handle_inst_emul(struct vcpu *vcpu, bool *retu) 1393 { 1394 struct vie *vie; 1395 struct vm_exit *vme; 1396 uint64_t gla, gpa, cs_base; 1397 struct vm_guest_paging *paging; 1398 mem_region_read_t mread; 1399 mem_region_write_t mwrite; 1400 enum vm_cpu_mode cpu_mode; 1401 int cs_d, error, fault; 1402 1403 vme = &vcpu->exitinfo; 1404 1405 KASSERT(vme->inst_length == 0, ("%s: invalid inst_length %d", 1406 __func__, vme->inst_length)); 1407 1408 gla = vme->u.inst_emul.gla; 1409 gpa = vme->u.inst_emul.gpa; 1410 cs_base = vme->u.inst_emul.cs_base; 1411 cs_d = vme->u.inst_emul.cs_d; 1412 vie = &vme->u.inst_emul.vie; 1413 paging = &vme->u.inst_emul.paging; 1414 cpu_mode = paging->cpu_mode; 1415 1416 VMM_CTR1(vcpu, "inst_emul fault accessing gpa %#lx", gpa); 1417 1418 /* Fetch, decode and emulate the faulting instruction */ 1419 if (vie->num_valid == 0) { 1420 error = vmm_fetch_instruction(vcpu, paging, vme->rip + cs_base, 1421 VIE_INST_SIZE, vie, &fault); 1422 } else { 1423 /* 1424 * The instruction bytes have already been copied into 'vie' 1425 */ 1426 error = fault = 0; 1427 } 1428 if (error || fault) 1429 return (error); 1430 1431 if (vmm_decode_instruction(vcpu, gla, cpu_mode, cs_d, vie) != 0) { 1432 VMM_CTR1(vcpu, "Error decoding instruction at %#lx", 1433 vme->rip + cs_base); 1434 *retu = true; /* dump instruction bytes in userspace */ 1435 return (0); 1436 } 1437 1438 /* 1439 * Update 'nextrip' based on the length of the emulated instruction. 1440 */ 1441 vme->inst_length = vie->num_processed; 1442 vcpu->nextrip += vie->num_processed; 1443 VMM_CTR1(vcpu, "nextrip updated to %#lx after instruction decoding", 1444 vcpu->nextrip); 1445 1446 /* return to userland unless this is an in-kernel emulated device */ 1447 if (gpa >= DEFAULT_APIC_BASE && gpa < DEFAULT_APIC_BASE + PAGE_SIZE) { 1448 mread = lapic_mmio_read; 1449 mwrite = lapic_mmio_write; 1450 } else if (gpa >= VIOAPIC_BASE && gpa < VIOAPIC_BASE + VIOAPIC_SIZE) { 1451 mread = vioapic_mmio_read; 1452 mwrite = vioapic_mmio_write; 1453 } else if (gpa >= VHPET_BASE && gpa < VHPET_BASE + VHPET_SIZE) { 1454 mread = vhpet_mmio_read; 1455 mwrite = vhpet_mmio_write; 1456 } else { 1457 *retu = true; 1458 return (0); 1459 } 1460 1461 error = vmm_emulate_instruction(vcpu, gpa, vie, paging, mread, mwrite, 1462 retu); 1463 1464 return (error); 1465 } 1466 1467 static int 1468 vm_handle_suspend(struct vcpu *vcpu, bool *retu) 1469 { 1470 struct vm *vm = vcpu->vm; 1471 int error, i; 1472 struct thread *td; 1473 1474 error = 0; 1475 td = curthread; 1476 1477 CPU_SET_ATOMIC(vcpu->vcpuid, &vm->suspended_cpus); 1478 1479 /* 1480 * Wait until all 'active_cpus' have suspended themselves. 1481 * 1482 * Since a VM may be suspended at any time including when one or 1483 * more vcpus are doing a rendezvous we need to call the rendezvous 1484 * handler while we are waiting to prevent a deadlock. 1485 */ 1486 vcpu_lock(vcpu); 1487 while (error == 0) { 1488 if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) { 1489 VMM_CTR0(vcpu, "All vcpus suspended"); 1490 break; 1491 } 1492 1493 if (vm->rendezvous_func == NULL) { 1494 VMM_CTR0(vcpu, "Sleeping during suspend"); 1495 vcpu_require_state_locked(vcpu, VCPU_SLEEPING); 1496 msleep_spin(vcpu, &vcpu->mtx, "vmsusp", hz); 1497 vcpu_require_state_locked(vcpu, VCPU_FROZEN); 1498 if (td_ast_pending(td, TDA_SUSPEND)) { 1499 vcpu_unlock(vcpu); 1500 error = thread_check_susp(td, false); 1501 vcpu_lock(vcpu); 1502 } 1503 } else { 1504 VMM_CTR0(vcpu, "Rendezvous during suspend"); 1505 vcpu_unlock(vcpu); 1506 error = vm_handle_rendezvous(vcpu); 1507 vcpu_lock(vcpu); 1508 } 1509 } 1510 vcpu_unlock(vcpu); 1511 1512 /* 1513 * Wakeup the other sleeping vcpus and return to userspace. 1514 */ 1515 for (i = 0; i < vm->maxcpus; i++) { 1516 if (CPU_ISSET(i, &vm->suspended_cpus)) { 1517 vcpu_notify_event(vm_vcpu(vm, i), false); 1518 } 1519 } 1520 1521 *retu = true; 1522 return (error); 1523 } 1524 1525 static int 1526 vm_handle_reqidle(struct vcpu *vcpu, bool *retu) 1527 { 1528 vcpu_lock(vcpu); 1529 KASSERT(vcpu->reqidle, ("invalid vcpu reqidle %d", vcpu->reqidle)); 1530 vcpu->reqidle = 0; 1531 vcpu_unlock(vcpu); 1532 *retu = true; 1533 return (0); 1534 } 1535 1536 static int 1537 vm_handle_db(struct vcpu *vcpu, struct vm_exit *vme, bool *retu) 1538 { 1539 int error, fault; 1540 uint64_t rsp; 1541 uint64_t rflags; 1542 struct vm_copyinfo copyinfo[2]; 1543 1544 *retu = true; 1545 if (!vme->u.dbg.pushf_intercept || vme->u.dbg.tf_shadow_val != 0) { 1546 return (0); 1547 } 1548 1549 vm_get_register(vcpu, VM_REG_GUEST_RSP, &rsp); 1550 error = vm_copy_setup(vcpu, &vme->u.dbg.paging, rsp, sizeof(uint64_t), 1551 VM_PROT_RW, copyinfo, nitems(copyinfo), &fault); 1552 if (error != 0 || fault != 0) { 1553 *retu = false; 1554 return (EINVAL); 1555 } 1556 1557 /* Read pushed rflags value from top of stack. */ 1558 vm_copyin(copyinfo, &rflags, sizeof(uint64_t)); 1559 1560 /* Clear TF bit. */ 1561 rflags &= ~(PSL_T); 1562 1563 /* Write updated value back to memory. */ 1564 vm_copyout(&rflags, copyinfo, sizeof(uint64_t)); 1565 vm_copy_teardown(copyinfo, nitems(copyinfo)); 1566 1567 return (0); 1568 } 1569 1570 int 1571 vm_suspend(struct vm *vm, enum vm_suspend_how how) 1572 { 1573 int i; 1574 1575 if (how <= VM_SUSPEND_NONE || how >= VM_SUSPEND_LAST) 1576 return (EINVAL); 1577 1578 if (atomic_cmpset_int(&vm->suspend, 0, how) == 0) { 1579 VM_CTR2(vm, "virtual machine already suspended %d/%d", 1580 vm->suspend, how); 1581 return (EALREADY); 1582 } 1583 1584 VM_CTR1(vm, "virtual machine successfully suspended %d", how); 1585 1586 /* 1587 * Notify all active vcpus that they are now suspended. 1588 */ 1589 for (i = 0; i < vm->maxcpus; i++) { 1590 if (CPU_ISSET(i, &vm->active_cpus)) 1591 vcpu_notify_event(vm_vcpu(vm, i), false); 1592 } 1593 1594 return (0); 1595 } 1596 1597 void 1598 vm_exit_suspended(struct vcpu *vcpu, uint64_t rip) 1599 { 1600 struct vm *vm = vcpu->vm; 1601 struct vm_exit *vmexit; 1602 1603 KASSERT(vm->suspend > VM_SUSPEND_NONE && vm->suspend < VM_SUSPEND_LAST, 1604 ("vm_exit_suspended: invalid suspend type %d", vm->suspend)); 1605 1606 vmexit = vm_exitinfo(vcpu); 1607 vmexit->rip = rip; 1608 vmexit->inst_length = 0; 1609 vmexit->exitcode = VM_EXITCODE_SUSPENDED; 1610 vmexit->u.suspended.how = vm->suspend; 1611 } 1612 1613 void 1614 vm_exit_debug(struct vcpu *vcpu, uint64_t rip) 1615 { 1616 struct vm_exit *vmexit; 1617 1618 vmexit = vm_exitinfo(vcpu); 1619 vmexit->rip = rip; 1620 vmexit->inst_length = 0; 1621 vmexit->exitcode = VM_EXITCODE_DEBUG; 1622 } 1623 1624 void 1625 vm_exit_rendezvous(struct vcpu *vcpu, uint64_t rip) 1626 { 1627 struct vm_exit *vmexit; 1628 1629 vmexit = vm_exitinfo(vcpu); 1630 vmexit->rip = rip; 1631 vmexit->inst_length = 0; 1632 vmexit->exitcode = VM_EXITCODE_RENDEZVOUS; 1633 vmm_stat_incr(vcpu, VMEXIT_RENDEZVOUS, 1); 1634 } 1635 1636 void 1637 vm_exit_reqidle(struct vcpu *vcpu, uint64_t rip) 1638 { 1639 struct vm_exit *vmexit; 1640 1641 vmexit = vm_exitinfo(vcpu); 1642 vmexit->rip = rip; 1643 vmexit->inst_length = 0; 1644 vmexit->exitcode = VM_EXITCODE_REQIDLE; 1645 vmm_stat_incr(vcpu, VMEXIT_REQIDLE, 1); 1646 } 1647 1648 void 1649 vm_exit_astpending(struct vcpu *vcpu, uint64_t rip) 1650 { 1651 struct vm_exit *vmexit; 1652 1653 vmexit = vm_exitinfo(vcpu); 1654 vmexit->rip = rip; 1655 vmexit->inst_length = 0; 1656 vmexit->exitcode = VM_EXITCODE_BOGUS; 1657 vmm_stat_incr(vcpu, VMEXIT_ASTPENDING, 1); 1658 } 1659 1660 int 1661 vm_run(struct vcpu *vcpu) 1662 { 1663 struct vm *vm = vcpu->vm; 1664 struct vm_eventinfo evinfo; 1665 int error, vcpuid; 1666 struct pcb *pcb; 1667 uint64_t tscval; 1668 struct vm_exit *vme; 1669 bool retu, intr_disabled; 1670 pmap_t pmap; 1671 1672 vcpuid = vcpu->vcpuid; 1673 1674 if (!CPU_ISSET(vcpuid, &vm->active_cpus)) 1675 return (EINVAL); 1676 1677 if (CPU_ISSET(vcpuid, &vm->suspended_cpus)) 1678 return (EINVAL); 1679 1680 pmap = vmspace_pmap(vm_vmspace(vm)); 1681 vme = &vcpu->exitinfo; 1682 evinfo.rptr = &vm->rendezvous_req_cpus; 1683 evinfo.sptr = &vm->suspend; 1684 evinfo.iptr = &vcpu->reqidle; 1685 restart: 1686 critical_enter(); 1687 1688 KASSERT(!CPU_ISSET(curcpu, &pmap->pm_active), 1689 ("vm_run: absurd pm_active")); 1690 1691 tscval = rdtsc(); 1692 1693 pcb = PCPU_GET(curpcb); 1694 set_pcb_flags(pcb, PCB_FULL_IRET); 1695 1696 restore_guest_fpustate(vcpu); 1697 1698 vcpu_require_state(vcpu, VCPU_RUNNING); 1699 error = vmmops_run(vcpu->cookie, vcpu->nextrip, pmap, &evinfo); 1700 vcpu_require_state(vcpu, VCPU_FROZEN); 1701 1702 save_guest_fpustate(vcpu); 1703 1704 vmm_stat_incr(vcpu, VCPU_TOTAL_RUNTIME, rdtsc() - tscval); 1705 1706 critical_exit(); 1707 1708 if (error == 0) { 1709 retu = false; 1710 vcpu->nextrip = vme->rip + vme->inst_length; 1711 switch (vme->exitcode) { 1712 case VM_EXITCODE_REQIDLE: 1713 error = vm_handle_reqidle(vcpu, &retu); 1714 break; 1715 case VM_EXITCODE_SUSPENDED: 1716 error = vm_handle_suspend(vcpu, &retu); 1717 break; 1718 case VM_EXITCODE_IOAPIC_EOI: 1719 vioapic_process_eoi(vm, vme->u.ioapic_eoi.vector); 1720 break; 1721 case VM_EXITCODE_RENDEZVOUS: 1722 error = vm_handle_rendezvous(vcpu); 1723 break; 1724 case VM_EXITCODE_HLT: 1725 intr_disabled = ((vme->u.hlt.rflags & PSL_I) == 0); 1726 error = vm_handle_hlt(vcpu, intr_disabled, &retu); 1727 break; 1728 case VM_EXITCODE_PAGING: 1729 error = vm_handle_paging(vcpu, &retu); 1730 break; 1731 case VM_EXITCODE_INST_EMUL: 1732 error = vm_handle_inst_emul(vcpu, &retu); 1733 break; 1734 case VM_EXITCODE_INOUT: 1735 case VM_EXITCODE_INOUT_STR: 1736 error = vm_handle_inout(vcpu, vme, &retu); 1737 break; 1738 case VM_EXITCODE_DB: 1739 error = vm_handle_db(vcpu, vme, &retu); 1740 break; 1741 case VM_EXITCODE_MONITOR: 1742 case VM_EXITCODE_MWAIT: 1743 case VM_EXITCODE_VMINSN: 1744 vm_inject_ud(vcpu); 1745 break; 1746 default: 1747 retu = true; /* handled in userland */ 1748 break; 1749 } 1750 } 1751 1752 /* 1753 * VM_EXITCODE_INST_EMUL could access the apic which could transform the 1754 * exit code into VM_EXITCODE_IPI. 1755 */ 1756 if (error == 0 && vme->exitcode == VM_EXITCODE_IPI) 1757 error = vm_handle_ipi(vcpu, vme, &retu); 1758 1759 if (error == 0 && retu == false) 1760 goto restart; 1761 1762 vmm_stat_incr(vcpu, VMEXIT_USERSPACE, 1); 1763 VMM_CTR2(vcpu, "retu %d/%d", error, vme->exitcode); 1764 1765 return (error); 1766 } 1767 1768 int 1769 vm_restart_instruction(struct vcpu *vcpu) 1770 { 1771 enum vcpu_state state; 1772 uint64_t rip; 1773 int error __diagused; 1774 1775 state = vcpu_get_state(vcpu, NULL); 1776 if (state == VCPU_RUNNING) { 1777 /* 1778 * When a vcpu is "running" the next instruction is determined 1779 * by adding 'rip' and 'inst_length' in the vcpu's 'exitinfo'. 1780 * Thus setting 'inst_length' to zero will cause the current 1781 * instruction to be restarted. 1782 */ 1783 vcpu->exitinfo.inst_length = 0; 1784 VMM_CTR1(vcpu, "restarting instruction at %#lx by " 1785 "setting inst_length to zero", vcpu->exitinfo.rip); 1786 } else if (state == VCPU_FROZEN) { 1787 /* 1788 * When a vcpu is "frozen" it is outside the critical section 1789 * around vmmops_run() and 'nextrip' points to the next 1790 * instruction. Thus instruction restart is achieved by setting 1791 * 'nextrip' to the vcpu's %rip. 1792 */ 1793 error = vm_get_register(vcpu, VM_REG_GUEST_RIP, &rip); 1794 KASSERT(!error, ("%s: error %d getting rip", __func__, error)); 1795 VMM_CTR2(vcpu, "restarting instruction by updating " 1796 "nextrip from %#lx to %#lx", vcpu->nextrip, rip); 1797 vcpu->nextrip = rip; 1798 } else { 1799 panic("%s: invalid state %d", __func__, state); 1800 } 1801 return (0); 1802 } 1803 1804 int 1805 vm_exit_intinfo(struct vcpu *vcpu, uint64_t info) 1806 { 1807 int type, vector; 1808 1809 if (info & VM_INTINFO_VALID) { 1810 type = info & VM_INTINFO_TYPE; 1811 vector = info & 0xff; 1812 if (type == VM_INTINFO_NMI && vector != IDT_NMI) 1813 return (EINVAL); 1814 if (type == VM_INTINFO_HWEXCEPTION && vector >= 32) 1815 return (EINVAL); 1816 if (info & VM_INTINFO_RSVD) 1817 return (EINVAL); 1818 } else { 1819 info = 0; 1820 } 1821 VMM_CTR2(vcpu, "%s: info1(%#lx)", __func__, info); 1822 vcpu->exitintinfo = info; 1823 return (0); 1824 } 1825 1826 enum exc_class { 1827 EXC_BENIGN, 1828 EXC_CONTRIBUTORY, 1829 EXC_PAGEFAULT 1830 }; 1831 1832 #define IDT_VE 20 /* Virtualization Exception (Intel specific) */ 1833 1834 static enum exc_class 1835 exception_class(uint64_t info) 1836 { 1837 int type, vector; 1838 1839 KASSERT(info & VM_INTINFO_VALID, ("intinfo must be valid: %#lx", info)); 1840 type = info & VM_INTINFO_TYPE; 1841 vector = info & 0xff; 1842 1843 /* Table 6-4, "Interrupt and Exception Classes", Intel SDM, Vol 3 */ 1844 switch (type) { 1845 case VM_INTINFO_HWINTR: 1846 case VM_INTINFO_SWINTR: 1847 case VM_INTINFO_NMI: 1848 return (EXC_BENIGN); 1849 default: 1850 /* 1851 * Hardware exception. 1852 * 1853 * SVM and VT-x use identical type values to represent NMI, 1854 * hardware interrupt and software interrupt. 1855 * 1856 * SVM uses type '3' for all exceptions. VT-x uses type '3' 1857 * for exceptions except #BP and #OF. #BP and #OF use a type 1858 * value of '5' or '6'. Therefore we don't check for explicit 1859 * values of 'type' to classify 'intinfo' into a hardware 1860 * exception. 1861 */ 1862 break; 1863 } 1864 1865 switch (vector) { 1866 case IDT_PF: 1867 case IDT_VE: 1868 return (EXC_PAGEFAULT); 1869 case IDT_DE: 1870 case IDT_TS: 1871 case IDT_NP: 1872 case IDT_SS: 1873 case IDT_GP: 1874 return (EXC_CONTRIBUTORY); 1875 default: 1876 return (EXC_BENIGN); 1877 } 1878 } 1879 1880 static int 1881 nested_fault(struct vcpu *vcpu, uint64_t info1, uint64_t info2, 1882 uint64_t *retinfo) 1883 { 1884 enum exc_class exc1, exc2; 1885 int type1, vector1; 1886 1887 KASSERT(info1 & VM_INTINFO_VALID, ("info1 %#lx is not valid", info1)); 1888 KASSERT(info2 & VM_INTINFO_VALID, ("info2 %#lx is not valid", info2)); 1889 1890 /* 1891 * If an exception occurs while attempting to call the double-fault 1892 * handler the processor enters shutdown mode (aka triple fault). 1893 */ 1894 type1 = info1 & VM_INTINFO_TYPE; 1895 vector1 = info1 & 0xff; 1896 if (type1 == VM_INTINFO_HWEXCEPTION && vector1 == IDT_DF) { 1897 VMM_CTR2(vcpu, "triple fault: info1(%#lx), info2(%#lx)", 1898 info1, info2); 1899 vm_suspend(vcpu->vm, VM_SUSPEND_TRIPLEFAULT); 1900 *retinfo = 0; 1901 return (0); 1902 } 1903 1904 /* 1905 * Table 6-5 "Conditions for Generating a Double Fault", Intel SDM, Vol3 1906 */ 1907 exc1 = exception_class(info1); 1908 exc2 = exception_class(info2); 1909 if ((exc1 == EXC_CONTRIBUTORY && exc2 == EXC_CONTRIBUTORY) || 1910 (exc1 == EXC_PAGEFAULT && exc2 != EXC_BENIGN)) { 1911 /* Convert nested fault into a double fault. */ 1912 *retinfo = IDT_DF; 1913 *retinfo |= VM_INTINFO_VALID | VM_INTINFO_HWEXCEPTION; 1914 *retinfo |= VM_INTINFO_DEL_ERRCODE; 1915 } else { 1916 /* Handle exceptions serially */ 1917 *retinfo = info2; 1918 } 1919 return (1); 1920 } 1921 1922 static uint64_t 1923 vcpu_exception_intinfo(struct vcpu *vcpu) 1924 { 1925 uint64_t info = 0; 1926 1927 if (vcpu->exception_pending) { 1928 info = vcpu->exc_vector & 0xff; 1929 info |= VM_INTINFO_VALID | VM_INTINFO_HWEXCEPTION; 1930 if (vcpu->exc_errcode_valid) { 1931 info |= VM_INTINFO_DEL_ERRCODE; 1932 info |= (uint64_t)vcpu->exc_errcode << 32; 1933 } 1934 } 1935 return (info); 1936 } 1937 1938 int 1939 vm_entry_intinfo(struct vcpu *vcpu, uint64_t *retinfo) 1940 { 1941 uint64_t info1, info2; 1942 int valid; 1943 1944 info1 = vcpu->exitintinfo; 1945 vcpu->exitintinfo = 0; 1946 1947 info2 = 0; 1948 if (vcpu->exception_pending) { 1949 info2 = vcpu_exception_intinfo(vcpu); 1950 vcpu->exception_pending = 0; 1951 VMM_CTR2(vcpu, "Exception %d delivered: %#lx", 1952 vcpu->exc_vector, info2); 1953 } 1954 1955 if ((info1 & VM_INTINFO_VALID) && (info2 & VM_INTINFO_VALID)) { 1956 valid = nested_fault(vcpu, info1, info2, retinfo); 1957 } else if (info1 & VM_INTINFO_VALID) { 1958 *retinfo = info1; 1959 valid = 1; 1960 } else if (info2 & VM_INTINFO_VALID) { 1961 *retinfo = info2; 1962 valid = 1; 1963 } else { 1964 valid = 0; 1965 } 1966 1967 if (valid) { 1968 VMM_CTR4(vcpu, "%s: info1(%#lx), info2(%#lx), " 1969 "retinfo(%#lx)", __func__, info1, info2, *retinfo); 1970 } 1971 1972 return (valid); 1973 } 1974 1975 int 1976 vm_get_intinfo(struct vcpu *vcpu, uint64_t *info1, uint64_t *info2) 1977 { 1978 *info1 = vcpu->exitintinfo; 1979 *info2 = vcpu_exception_intinfo(vcpu); 1980 return (0); 1981 } 1982 1983 int 1984 vm_inject_exception(struct vcpu *vcpu, int vector, int errcode_valid, 1985 uint32_t errcode, int restart_instruction) 1986 { 1987 uint64_t regval; 1988 int error __diagused; 1989 1990 if (vector < 0 || vector >= 32) 1991 return (EINVAL); 1992 1993 /* 1994 * A double fault exception should never be injected directly into 1995 * the guest. It is a derived exception that results from specific 1996 * combinations of nested faults. 1997 */ 1998 if (vector == IDT_DF) 1999 return (EINVAL); 2000 2001 if (vcpu->exception_pending) { 2002 VMM_CTR2(vcpu, "Unable to inject exception %d due to " 2003 "pending exception %d", vector, vcpu->exc_vector); 2004 return (EBUSY); 2005 } 2006 2007 if (errcode_valid) { 2008 /* 2009 * Exceptions don't deliver an error code in real mode. 2010 */ 2011 error = vm_get_register(vcpu, VM_REG_GUEST_CR0, ®val); 2012 KASSERT(!error, ("%s: error %d getting CR0", __func__, error)); 2013 if (!(regval & CR0_PE)) 2014 errcode_valid = 0; 2015 } 2016 2017 /* 2018 * From section 26.6.1 "Interruptibility State" in Intel SDM: 2019 * 2020 * Event blocking by "STI" or "MOV SS" is cleared after guest executes 2021 * one instruction or incurs an exception. 2022 */ 2023 error = vm_set_register(vcpu, VM_REG_GUEST_INTR_SHADOW, 0); 2024 KASSERT(error == 0, ("%s: error %d clearing interrupt shadow", 2025 __func__, error)); 2026 2027 if (restart_instruction) 2028 vm_restart_instruction(vcpu); 2029 2030 vcpu->exception_pending = 1; 2031 vcpu->exc_vector = vector; 2032 vcpu->exc_errcode = errcode; 2033 vcpu->exc_errcode_valid = errcode_valid; 2034 VMM_CTR1(vcpu, "Exception %d pending", vector); 2035 return (0); 2036 } 2037 2038 void 2039 vm_inject_fault(struct vcpu *vcpu, int vector, int errcode_valid, int errcode) 2040 { 2041 int error __diagused, restart_instruction; 2042 2043 restart_instruction = 1; 2044 2045 error = vm_inject_exception(vcpu, vector, errcode_valid, 2046 errcode, restart_instruction); 2047 KASSERT(error == 0, ("vm_inject_exception error %d", error)); 2048 } 2049 2050 void 2051 vm_inject_pf(struct vcpu *vcpu, int error_code, uint64_t cr2) 2052 { 2053 int error __diagused; 2054 2055 VMM_CTR2(vcpu, "Injecting page fault: error_code %#x, cr2 %#lx", 2056 error_code, cr2); 2057 2058 error = vm_set_register(vcpu, VM_REG_GUEST_CR2, cr2); 2059 KASSERT(error == 0, ("vm_set_register(cr2) error %d", error)); 2060 2061 vm_inject_fault(vcpu, IDT_PF, 1, error_code); 2062 } 2063 2064 static VMM_STAT(VCPU_NMI_COUNT, "number of NMIs delivered to vcpu"); 2065 2066 int 2067 vm_inject_nmi(struct vcpu *vcpu) 2068 { 2069 2070 vcpu->nmi_pending = 1; 2071 vcpu_notify_event(vcpu, false); 2072 return (0); 2073 } 2074 2075 int 2076 vm_nmi_pending(struct vcpu *vcpu) 2077 { 2078 return (vcpu->nmi_pending); 2079 } 2080 2081 void 2082 vm_nmi_clear(struct vcpu *vcpu) 2083 { 2084 if (vcpu->nmi_pending == 0) 2085 panic("vm_nmi_clear: inconsistent nmi_pending state"); 2086 2087 vcpu->nmi_pending = 0; 2088 vmm_stat_incr(vcpu, VCPU_NMI_COUNT, 1); 2089 } 2090 2091 static VMM_STAT(VCPU_EXTINT_COUNT, "number of ExtINTs delivered to vcpu"); 2092 2093 int 2094 vm_inject_extint(struct vcpu *vcpu) 2095 { 2096 2097 vcpu->extint_pending = 1; 2098 vcpu_notify_event(vcpu, false); 2099 return (0); 2100 } 2101 2102 int 2103 vm_extint_pending(struct vcpu *vcpu) 2104 { 2105 return (vcpu->extint_pending); 2106 } 2107 2108 void 2109 vm_extint_clear(struct vcpu *vcpu) 2110 { 2111 if (vcpu->extint_pending == 0) 2112 panic("vm_extint_clear: inconsistent extint_pending state"); 2113 2114 vcpu->extint_pending = 0; 2115 vmm_stat_incr(vcpu, VCPU_EXTINT_COUNT, 1); 2116 } 2117 2118 int 2119 vm_get_capability(struct vcpu *vcpu, int type, int *retval) 2120 { 2121 if (type < 0 || type >= VM_CAP_MAX) 2122 return (EINVAL); 2123 2124 return (vmmops_getcap(vcpu->cookie, type, retval)); 2125 } 2126 2127 int 2128 vm_set_capability(struct vcpu *vcpu, int type, int val) 2129 { 2130 if (type < 0 || type >= VM_CAP_MAX) 2131 return (EINVAL); 2132 2133 return (vmmops_setcap(vcpu->cookie, type, val)); 2134 } 2135 2136 struct vm * 2137 vcpu_vm(struct vcpu *vcpu) 2138 { 2139 return (vcpu->vm); 2140 } 2141 2142 int 2143 vcpu_vcpuid(struct vcpu *vcpu) 2144 { 2145 return (vcpu->vcpuid); 2146 } 2147 2148 struct vcpu * 2149 vm_vcpu(struct vm *vm, int vcpuid) 2150 { 2151 return (vm->vcpu[vcpuid]); 2152 } 2153 2154 struct vlapic * 2155 vm_lapic(struct vcpu *vcpu) 2156 { 2157 return (vcpu->vlapic); 2158 } 2159 2160 struct vioapic * 2161 vm_ioapic(struct vm *vm) 2162 { 2163 2164 return (vm->vioapic); 2165 } 2166 2167 struct vhpet * 2168 vm_hpet(struct vm *vm) 2169 { 2170 2171 return (vm->vhpet); 2172 } 2173 2174 bool 2175 vmm_is_pptdev(int bus, int slot, int func) 2176 { 2177 int b, f, i, n, s; 2178 char *val, *cp, *cp2; 2179 bool found; 2180 2181 /* 2182 * XXX 2183 * The length of an environment variable is limited to 128 bytes which 2184 * puts an upper limit on the number of passthru devices that may be 2185 * specified using a single environment variable. 2186 * 2187 * Work around this by scanning multiple environment variable 2188 * names instead of a single one - yuck! 2189 */ 2190 const char *names[] = { "pptdevs", "pptdevs2", "pptdevs3", NULL }; 2191 2192 /* set pptdevs="1/2/3 4/5/6 7/8/9 10/11/12" */ 2193 found = false; 2194 for (i = 0; names[i] != NULL && !found; i++) { 2195 cp = val = kern_getenv(names[i]); 2196 while (cp != NULL && *cp != '\0') { 2197 if ((cp2 = strchr(cp, ' ')) != NULL) 2198 *cp2 = '\0'; 2199 2200 n = sscanf(cp, "%d/%d/%d", &b, &s, &f); 2201 if (n == 3 && bus == b && slot == s && func == f) { 2202 found = true; 2203 break; 2204 } 2205 2206 if (cp2 != NULL) 2207 *cp2++ = ' '; 2208 2209 cp = cp2; 2210 } 2211 freeenv(val); 2212 } 2213 return (found); 2214 } 2215 2216 void * 2217 vm_iommu_domain(struct vm *vm) 2218 { 2219 2220 return (vm->iommu); 2221 } 2222 2223 int 2224 vcpu_set_state(struct vcpu *vcpu, enum vcpu_state newstate, bool from_idle) 2225 { 2226 int error; 2227 2228 vcpu_lock(vcpu); 2229 error = vcpu_set_state_locked(vcpu, newstate, from_idle); 2230 vcpu_unlock(vcpu); 2231 2232 return (error); 2233 } 2234 2235 enum vcpu_state 2236 vcpu_get_state(struct vcpu *vcpu, int *hostcpu) 2237 { 2238 enum vcpu_state state; 2239 2240 vcpu_lock(vcpu); 2241 state = vcpu->state; 2242 if (hostcpu != NULL) 2243 *hostcpu = vcpu->hostcpu; 2244 vcpu_unlock(vcpu); 2245 2246 return (state); 2247 } 2248 2249 int 2250 vm_activate_cpu(struct vcpu *vcpu) 2251 { 2252 struct vm *vm = vcpu->vm; 2253 2254 if (CPU_ISSET(vcpu->vcpuid, &vm->active_cpus)) 2255 return (EBUSY); 2256 2257 VMM_CTR0(vcpu, "activated"); 2258 CPU_SET_ATOMIC(vcpu->vcpuid, &vm->active_cpus); 2259 return (0); 2260 } 2261 2262 int 2263 vm_suspend_cpu(struct vm *vm, struct vcpu *vcpu) 2264 { 2265 if (vcpu == NULL) { 2266 vm->debug_cpus = vm->active_cpus; 2267 for (int i = 0; i < vm->maxcpus; i++) { 2268 if (CPU_ISSET(i, &vm->active_cpus)) 2269 vcpu_notify_event(vm_vcpu(vm, i), false); 2270 } 2271 } else { 2272 if (!CPU_ISSET(vcpu->vcpuid, &vm->active_cpus)) 2273 return (EINVAL); 2274 2275 CPU_SET_ATOMIC(vcpu->vcpuid, &vm->debug_cpus); 2276 vcpu_notify_event(vcpu, false); 2277 } 2278 return (0); 2279 } 2280 2281 int 2282 vm_resume_cpu(struct vm *vm, struct vcpu *vcpu) 2283 { 2284 2285 if (vcpu == NULL) { 2286 CPU_ZERO(&vm->debug_cpus); 2287 } else { 2288 if (!CPU_ISSET(vcpu->vcpuid, &vm->debug_cpus)) 2289 return (EINVAL); 2290 2291 CPU_CLR_ATOMIC(vcpu->vcpuid, &vm->debug_cpus); 2292 } 2293 return (0); 2294 } 2295 2296 int 2297 vcpu_debugged(struct vcpu *vcpu) 2298 { 2299 2300 return (CPU_ISSET(vcpu->vcpuid, &vcpu->vm->debug_cpus)); 2301 } 2302 2303 cpuset_t 2304 vm_active_cpus(struct vm *vm) 2305 { 2306 2307 return (vm->active_cpus); 2308 } 2309 2310 cpuset_t 2311 vm_debug_cpus(struct vm *vm) 2312 { 2313 2314 return (vm->debug_cpus); 2315 } 2316 2317 cpuset_t 2318 vm_suspended_cpus(struct vm *vm) 2319 { 2320 2321 return (vm->suspended_cpus); 2322 } 2323 2324 /* 2325 * Returns the subset of vCPUs in tostart that are awaiting startup. 2326 * These vCPUs are also marked as no longer awaiting startup. 2327 */ 2328 cpuset_t 2329 vm_start_cpus(struct vm *vm, const cpuset_t *tostart) 2330 { 2331 cpuset_t set; 2332 2333 mtx_lock(&vm->rendezvous_mtx); 2334 CPU_AND(&set, &vm->startup_cpus, tostart); 2335 CPU_ANDNOT(&vm->startup_cpus, &vm->startup_cpus, &set); 2336 mtx_unlock(&vm->rendezvous_mtx); 2337 return (set); 2338 } 2339 2340 void 2341 vm_await_start(struct vm *vm, const cpuset_t *waiting) 2342 { 2343 mtx_lock(&vm->rendezvous_mtx); 2344 CPU_OR(&vm->startup_cpus, &vm->startup_cpus, waiting); 2345 mtx_unlock(&vm->rendezvous_mtx); 2346 } 2347 2348 void * 2349 vcpu_stats(struct vcpu *vcpu) 2350 { 2351 2352 return (vcpu->stats); 2353 } 2354 2355 int 2356 vm_get_x2apic_state(struct vcpu *vcpu, enum x2apic_state *state) 2357 { 2358 *state = vcpu->x2apic_state; 2359 2360 return (0); 2361 } 2362 2363 int 2364 vm_set_x2apic_state(struct vcpu *vcpu, enum x2apic_state state) 2365 { 2366 if (state >= X2APIC_STATE_LAST) 2367 return (EINVAL); 2368 2369 vcpu->x2apic_state = state; 2370 2371 vlapic_set_x2apic_state(vcpu, state); 2372 2373 return (0); 2374 } 2375 2376 /* 2377 * This function is called to ensure that a vcpu "sees" a pending event 2378 * as soon as possible: 2379 * - If the vcpu thread is sleeping then it is woken up. 2380 * - If the vcpu is running on a different host_cpu then an IPI will be directed 2381 * to the host_cpu to cause the vcpu to trap into the hypervisor. 2382 */ 2383 static void 2384 vcpu_notify_event_locked(struct vcpu *vcpu, bool lapic_intr) 2385 { 2386 int hostcpu; 2387 2388 hostcpu = vcpu->hostcpu; 2389 if (vcpu->state == VCPU_RUNNING) { 2390 KASSERT(hostcpu != NOCPU, ("vcpu running on invalid hostcpu")); 2391 if (hostcpu != curcpu) { 2392 if (lapic_intr) { 2393 vlapic_post_intr(vcpu->vlapic, hostcpu, 2394 vmm_ipinum); 2395 } else { 2396 ipi_cpu(hostcpu, vmm_ipinum); 2397 } 2398 } else { 2399 /* 2400 * If the 'vcpu' is running on 'curcpu' then it must 2401 * be sending a notification to itself (e.g. SELF_IPI). 2402 * The pending event will be picked up when the vcpu 2403 * transitions back to guest context. 2404 */ 2405 } 2406 } else { 2407 KASSERT(hostcpu == NOCPU, ("vcpu state %d not consistent " 2408 "with hostcpu %d", vcpu->state, hostcpu)); 2409 if (vcpu->state == VCPU_SLEEPING) 2410 wakeup_one(vcpu); 2411 } 2412 } 2413 2414 void 2415 vcpu_notify_event(struct vcpu *vcpu, bool lapic_intr) 2416 { 2417 vcpu_lock(vcpu); 2418 vcpu_notify_event_locked(vcpu, lapic_intr); 2419 vcpu_unlock(vcpu); 2420 } 2421 2422 struct vm_mem * 2423 vm_mem(struct vm *vm) 2424 { 2425 return (&vm->mem); 2426 } 2427 2428 int 2429 vm_apicid2vcpuid(struct vm *vm, int apicid) 2430 { 2431 /* 2432 * XXX apic id is assumed to be numerically identical to vcpu id 2433 */ 2434 return (apicid); 2435 } 2436 2437 int 2438 vm_smp_rendezvous(struct vcpu *vcpu, cpuset_t dest, 2439 vm_rendezvous_func_t func, void *arg) 2440 { 2441 struct vm *vm = vcpu->vm; 2442 int error, i; 2443 2444 /* 2445 * Enforce that this function is called without any locks 2446 */ 2447 WITNESS_WARN(WARN_PANIC, NULL, "vm_smp_rendezvous"); 2448 2449 restart: 2450 mtx_lock(&vm->rendezvous_mtx); 2451 if (vm->rendezvous_func != NULL) { 2452 /* 2453 * If a rendezvous is already in progress then we need to 2454 * call the rendezvous handler in case this 'vcpu' is one 2455 * of the targets of the rendezvous. 2456 */ 2457 VMM_CTR0(vcpu, "Rendezvous already in progress"); 2458 mtx_unlock(&vm->rendezvous_mtx); 2459 error = vm_handle_rendezvous(vcpu); 2460 if (error != 0) 2461 return (error); 2462 goto restart; 2463 } 2464 KASSERT(vm->rendezvous_func == NULL, ("vm_smp_rendezvous: previous " 2465 "rendezvous is still in progress")); 2466 2467 VMM_CTR0(vcpu, "Initiating rendezvous"); 2468 vm->rendezvous_req_cpus = dest; 2469 CPU_ZERO(&vm->rendezvous_done_cpus); 2470 vm->rendezvous_arg = arg; 2471 vm->rendezvous_func = func; 2472 mtx_unlock(&vm->rendezvous_mtx); 2473 2474 /* 2475 * Wake up any sleeping vcpus and trigger a VM-exit in any running 2476 * vcpus so they handle the rendezvous as soon as possible. 2477 */ 2478 for (i = 0; i < vm->maxcpus; i++) { 2479 if (CPU_ISSET(i, &dest)) 2480 vcpu_notify_event(vm_vcpu(vm, i), false); 2481 } 2482 2483 return (vm_handle_rendezvous(vcpu)); 2484 } 2485 2486 struct vatpic * 2487 vm_atpic(struct vm *vm) 2488 { 2489 return (vm->vatpic); 2490 } 2491 2492 struct vatpit * 2493 vm_atpit(struct vm *vm) 2494 { 2495 return (vm->vatpit); 2496 } 2497 2498 struct vpmtmr * 2499 vm_pmtmr(struct vm *vm) 2500 { 2501 2502 return (vm->vpmtmr); 2503 } 2504 2505 struct vrtc * 2506 vm_rtc(struct vm *vm) 2507 { 2508 2509 return (vm->vrtc); 2510 } 2511 2512 enum vm_reg_name 2513 vm_segment_name(int seg) 2514 { 2515 static enum vm_reg_name seg_names[] = { 2516 VM_REG_GUEST_ES, 2517 VM_REG_GUEST_CS, 2518 VM_REG_GUEST_SS, 2519 VM_REG_GUEST_DS, 2520 VM_REG_GUEST_FS, 2521 VM_REG_GUEST_GS 2522 }; 2523 2524 KASSERT(seg >= 0 && seg < nitems(seg_names), 2525 ("%s: invalid segment encoding %d", __func__, seg)); 2526 return (seg_names[seg]); 2527 } 2528 2529 void 2530 vm_copy_teardown(struct vm_copyinfo *copyinfo, int num_copyinfo) 2531 { 2532 int idx; 2533 2534 for (idx = 0; idx < num_copyinfo; idx++) { 2535 if (copyinfo[idx].cookie != NULL) 2536 vm_gpa_release(copyinfo[idx].cookie); 2537 } 2538 bzero(copyinfo, num_copyinfo * sizeof(struct vm_copyinfo)); 2539 } 2540 2541 int 2542 vm_copy_setup(struct vcpu *vcpu, struct vm_guest_paging *paging, 2543 uint64_t gla, size_t len, int prot, struct vm_copyinfo *copyinfo, 2544 int num_copyinfo, int *fault) 2545 { 2546 int error, idx, nused; 2547 size_t n, off, remaining; 2548 void *hva, *cookie; 2549 uint64_t gpa; 2550 2551 bzero(copyinfo, sizeof(struct vm_copyinfo) * num_copyinfo); 2552 2553 nused = 0; 2554 remaining = len; 2555 while (remaining > 0) { 2556 if (nused >= num_copyinfo) 2557 return (EFAULT); 2558 error = vm_gla2gpa(vcpu, paging, gla, prot, &gpa, fault); 2559 if (error || *fault) 2560 return (error); 2561 off = gpa & PAGE_MASK; 2562 n = min(remaining, PAGE_SIZE - off); 2563 copyinfo[nused].gpa = gpa; 2564 copyinfo[nused].len = n; 2565 remaining -= n; 2566 gla += n; 2567 nused++; 2568 } 2569 2570 for (idx = 0; idx < nused; idx++) { 2571 hva = vm_gpa_hold(vcpu, copyinfo[idx].gpa, 2572 copyinfo[idx].len, prot, &cookie); 2573 if (hva == NULL) 2574 break; 2575 copyinfo[idx].hva = hva; 2576 copyinfo[idx].cookie = cookie; 2577 } 2578 2579 if (idx != nused) { 2580 vm_copy_teardown(copyinfo, num_copyinfo); 2581 return (EFAULT); 2582 } else { 2583 *fault = 0; 2584 return (0); 2585 } 2586 } 2587 2588 void 2589 vm_copyin(struct vm_copyinfo *copyinfo, void *kaddr, size_t len) 2590 { 2591 char *dst; 2592 int idx; 2593 2594 dst = kaddr; 2595 idx = 0; 2596 while (len > 0) { 2597 bcopy(copyinfo[idx].hva, dst, copyinfo[idx].len); 2598 len -= copyinfo[idx].len; 2599 dst += copyinfo[idx].len; 2600 idx++; 2601 } 2602 } 2603 2604 void 2605 vm_copyout(const void *kaddr, struct vm_copyinfo *copyinfo, size_t len) 2606 { 2607 const char *src; 2608 int idx; 2609 2610 src = kaddr; 2611 idx = 0; 2612 while (len > 0) { 2613 bcopy(src, copyinfo[idx].hva, copyinfo[idx].len); 2614 len -= copyinfo[idx].len; 2615 src += copyinfo[idx].len; 2616 idx++; 2617 } 2618 } 2619 2620 /* 2621 * Return the amount of in-use and wired memory for the VM. Since 2622 * these are global stats, only return the values with for vCPU 0 2623 */ 2624 VMM_STAT_DECLARE(VMM_MEM_RESIDENT); 2625 VMM_STAT_DECLARE(VMM_MEM_WIRED); 2626 2627 static void 2628 vm_get_rescnt(struct vcpu *vcpu, struct vmm_stat_type *stat) 2629 { 2630 2631 if (vcpu->vcpuid == 0) { 2632 vmm_stat_set(vcpu, VMM_MEM_RESIDENT, PAGE_SIZE * 2633 vmspace_resident_count(vm_vmspace(vcpu->vm))); 2634 } 2635 } 2636 2637 static void 2638 vm_get_wiredcnt(struct vcpu *vcpu, struct vmm_stat_type *stat) 2639 { 2640 2641 if (vcpu->vcpuid == 0) { 2642 vmm_stat_set(vcpu, VMM_MEM_WIRED, PAGE_SIZE * 2643 pmap_wired_count(vmspace_pmap(vm_vmspace(vcpu->vm)))); 2644 } 2645 } 2646 2647 VMM_STAT_FUNC(VMM_MEM_RESIDENT, "Resident memory", vm_get_rescnt); 2648 VMM_STAT_FUNC(VMM_MEM_WIRED, "Wired memory", vm_get_wiredcnt); 2649 2650 #ifdef BHYVE_SNAPSHOT 2651 static int 2652 vm_snapshot_vcpus(struct vm *vm, struct vm_snapshot_meta *meta) 2653 { 2654 uint64_t tsc, now; 2655 int ret; 2656 struct vcpu *vcpu; 2657 uint16_t i, maxcpus; 2658 2659 now = rdtsc(); 2660 maxcpus = vm_get_maxcpus(vm); 2661 for (i = 0; i < maxcpus; i++) { 2662 vcpu = vm->vcpu[i]; 2663 if (vcpu == NULL) 2664 continue; 2665 2666 SNAPSHOT_VAR_OR_LEAVE(vcpu->x2apic_state, meta, ret, done); 2667 SNAPSHOT_VAR_OR_LEAVE(vcpu->exitintinfo, meta, ret, done); 2668 SNAPSHOT_VAR_OR_LEAVE(vcpu->exc_vector, meta, ret, done); 2669 SNAPSHOT_VAR_OR_LEAVE(vcpu->exc_errcode_valid, meta, ret, done); 2670 SNAPSHOT_VAR_OR_LEAVE(vcpu->exc_errcode, meta, ret, done); 2671 SNAPSHOT_VAR_OR_LEAVE(vcpu->guest_xcr0, meta, ret, done); 2672 SNAPSHOT_VAR_OR_LEAVE(vcpu->exitinfo, meta, ret, done); 2673 SNAPSHOT_VAR_OR_LEAVE(vcpu->nextrip, meta, ret, done); 2674 2675 /* 2676 * Save the absolute TSC value by adding now to tsc_offset. 2677 * 2678 * It will be turned turned back into an actual offset when the 2679 * TSC restore function is called 2680 */ 2681 tsc = now + vcpu->tsc_offset; 2682 SNAPSHOT_VAR_OR_LEAVE(tsc, meta, ret, done); 2683 if (meta->op == VM_SNAPSHOT_RESTORE) 2684 vcpu->tsc_offset = tsc; 2685 } 2686 2687 done: 2688 return (ret); 2689 } 2690 2691 static int 2692 vm_snapshot_vm(struct vm *vm, struct vm_snapshot_meta *meta) 2693 { 2694 int ret; 2695 2696 ret = vm_snapshot_vcpus(vm, meta); 2697 if (ret != 0) 2698 goto done; 2699 2700 SNAPSHOT_VAR_OR_LEAVE(vm->startup_cpus, meta, ret, done); 2701 done: 2702 return (ret); 2703 } 2704 2705 static int 2706 vm_snapshot_vcpu(struct vm *vm, struct vm_snapshot_meta *meta) 2707 { 2708 int error; 2709 struct vcpu *vcpu; 2710 uint16_t i, maxcpus; 2711 2712 error = 0; 2713 2714 maxcpus = vm_get_maxcpus(vm); 2715 for (i = 0; i < maxcpus; i++) { 2716 vcpu = vm->vcpu[i]; 2717 if (vcpu == NULL) 2718 continue; 2719 2720 error = vmmops_vcpu_snapshot(vcpu->cookie, meta); 2721 if (error != 0) { 2722 printf("%s: failed to snapshot vmcs/vmcb data for " 2723 "vCPU: %d; error: %d\n", __func__, i, error); 2724 goto done; 2725 } 2726 } 2727 2728 done: 2729 return (error); 2730 } 2731 2732 /* 2733 * Save kernel-side structures to user-space for snapshotting. 2734 */ 2735 int 2736 vm_snapshot_req(struct vm *vm, struct vm_snapshot_meta *meta) 2737 { 2738 int ret = 0; 2739 2740 switch (meta->dev_req) { 2741 case STRUCT_VMCX: 2742 ret = vm_snapshot_vcpu(vm, meta); 2743 break; 2744 case STRUCT_VM: 2745 ret = vm_snapshot_vm(vm, meta); 2746 break; 2747 case STRUCT_VIOAPIC: 2748 ret = vioapic_snapshot(vm_ioapic(vm), meta); 2749 break; 2750 case STRUCT_VLAPIC: 2751 ret = vlapic_snapshot(vm, meta); 2752 break; 2753 case STRUCT_VHPET: 2754 ret = vhpet_snapshot(vm_hpet(vm), meta); 2755 break; 2756 case STRUCT_VATPIC: 2757 ret = vatpic_snapshot(vm_atpic(vm), meta); 2758 break; 2759 case STRUCT_VATPIT: 2760 ret = vatpit_snapshot(vm_atpit(vm), meta); 2761 break; 2762 case STRUCT_VPMTMR: 2763 ret = vpmtmr_snapshot(vm_pmtmr(vm), meta); 2764 break; 2765 case STRUCT_VRTC: 2766 ret = vrtc_snapshot(vm_rtc(vm), meta); 2767 break; 2768 default: 2769 printf("%s: failed to find the requested type %#x\n", 2770 __func__, meta->dev_req); 2771 ret = (EINVAL); 2772 } 2773 return (ret); 2774 } 2775 2776 void 2777 vm_set_tsc_offset(struct vcpu *vcpu, uint64_t offset) 2778 { 2779 vcpu->tsc_offset = offset; 2780 } 2781 2782 int 2783 vm_restore_time(struct vm *vm) 2784 { 2785 int error; 2786 uint64_t now; 2787 struct vcpu *vcpu; 2788 uint16_t i, maxcpus; 2789 2790 now = rdtsc(); 2791 2792 error = vhpet_restore_time(vm_hpet(vm)); 2793 if (error) 2794 return (error); 2795 2796 maxcpus = vm_get_maxcpus(vm); 2797 for (i = 0; i < maxcpus; i++) { 2798 vcpu = vm->vcpu[i]; 2799 if (vcpu == NULL) 2800 continue; 2801 2802 error = vmmops_restore_tsc(vcpu->cookie, 2803 vcpu->tsc_offset - now); 2804 if (error) 2805 return (error); 2806 } 2807 2808 return (0); 2809 } 2810 #endif 2811