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 return (vmm_mmio_alloc(vm_vmspace(vm), gpa, len, hpa)); 728 } 729 730 int 731 vm_unmap_mmio(struct vm *vm, vm_paddr_t gpa, size_t len) 732 { 733 734 vmm_mmio_free(vm_vmspace(vm), gpa, len); 735 return (0); 736 } 737 738 static int 739 vm_iommu_map(struct vm *vm) 740 { 741 pmap_t pmap; 742 vm_paddr_t gpa, hpa; 743 struct vm_mem_map *mm; 744 int error, i; 745 746 sx_assert(&vm->mem.mem_segs_lock, SX_LOCKED); 747 748 pmap = vmspace_pmap(vm_vmspace(vm)); 749 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 750 if (!vm_memseg_sysmem(vm, i)) 751 continue; 752 753 mm = &vm->mem.mem_maps[i]; 754 KASSERT((mm->flags & VM_MEMMAP_F_IOMMU) == 0, 755 ("iommu map found invalid memmap %#lx/%#lx/%#x", 756 mm->gpa, mm->len, mm->flags)); 757 if ((mm->flags & VM_MEMMAP_F_WIRED) == 0) 758 continue; 759 mm->flags |= VM_MEMMAP_F_IOMMU; 760 761 for (gpa = mm->gpa; gpa < mm->gpa + mm->len; gpa += PAGE_SIZE) { 762 hpa = pmap_extract(pmap, gpa); 763 764 /* 765 * All mappings in the vmm vmspace must be 766 * present since they are managed by vmm in this way. 767 * Because we are in pass-through mode, the 768 * mappings must also be wired. This implies 769 * that all pages must be mapped and wired, 770 * allowing to use pmap_extract() and avoiding the 771 * need to use vm_gpa_hold_global(). 772 * 773 * This could change if/when we start 774 * supporting page faults on IOMMU maps. 775 */ 776 KASSERT(vm_page_wired(PHYS_TO_VM_PAGE(hpa)), 777 ("vm_iommu_map: vm %p gpa %jx hpa %jx not wired", 778 vm, (uintmax_t)gpa, (uintmax_t)hpa)); 779 780 iommu_create_mapping(vm->iommu, gpa, hpa, PAGE_SIZE); 781 } 782 } 783 784 error = iommu_invalidate_tlb(iommu_host_domain()); 785 return (error); 786 } 787 788 static int 789 vm_iommu_unmap(struct vm *vm) 790 { 791 vm_paddr_t gpa; 792 struct vm_mem_map *mm; 793 int error, i; 794 795 sx_assert(&vm->mem.mem_segs_lock, SX_LOCKED); 796 797 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 798 if (!vm_memseg_sysmem(vm, i)) 799 continue; 800 801 mm = &vm->mem.mem_maps[i]; 802 if ((mm->flags & VM_MEMMAP_F_IOMMU) == 0) 803 continue; 804 mm->flags &= ~VM_MEMMAP_F_IOMMU; 805 KASSERT((mm->flags & VM_MEMMAP_F_WIRED) != 0, 806 ("iommu unmap found invalid memmap %#lx/%#lx/%#x", 807 mm->gpa, mm->len, mm->flags)); 808 809 for (gpa = mm->gpa; gpa < mm->gpa + mm->len; gpa += PAGE_SIZE) { 810 KASSERT(vm_page_wired(PHYS_TO_VM_PAGE(pmap_extract( 811 vmspace_pmap(vm_vmspace(vm)), gpa))), 812 ("vm_iommu_unmap: vm %p gpa %jx not wired", 813 vm, (uintmax_t)gpa)); 814 iommu_remove_mapping(vm->iommu, gpa, PAGE_SIZE); 815 } 816 } 817 818 /* 819 * Invalidate the cached translations associated with the domain 820 * from which pages were removed. 821 */ 822 error = iommu_invalidate_tlb(vm->iommu); 823 return (error); 824 } 825 826 int 827 vm_unassign_pptdev(struct vm *vm, int bus, int slot, int func) 828 { 829 int error; 830 831 error = ppt_unassign_device(vm, bus, slot, func); 832 if (error) 833 return (error); 834 835 if (ppt_assigned_devices(vm) == 0) 836 error = vm_iommu_unmap(vm); 837 838 return (error); 839 } 840 841 int 842 vm_assign_pptdev(struct vm *vm, int bus, int slot, int func) 843 { 844 int error; 845 vm_paddr_t maxaddr; 846 bool map = false; 847 848 /* Set up the IOMMU to do the 'gpa' to 'hpa' translation */ 849 if (ppt_assigned_devices(vm) == 0) { 850 KASSERT(vm->iommu == NULL, 851 ("vm_assign_pptdev: iommu must be NULL")); 852 maxaddr = vmm_sysmem_maxaddr(vm); 853 vm->iommu = iommu_create_domain(maxaddr); 854 if (vm->iommu == NULL) 855 return (ENXIO); 856 map = true; 857 } 858 859 error = ppt_assign_device(vm, bus, slot, func); 860 if (error == 0 && map) 861 error = vm_iommu_map(vm); 862 return (error); 863 } 864 865 int 866 vm_get_register(struct vcpu *vcpu, int reg, uint64_t *retval) 867 { 868 /* Negative values represent VM control structure fields. */ 869 if (reg >= VM_REG_LAST) 870 return (EINVAL); 871 872 return (vmmops_getreg(vcpu->cookie, reg, retval)); 873 } 874 875 int 876 vm_set_register(struct vcpu *vcpu, int reg, uint64_t val) 877 { 878 int error; 879 880 /* Negative values represent VM control structure fields. */ 881 if (reg >= VM_REG_LAST) 882 return (EINVAL); 883 884 error = vmmops_setreg(vcpu->cookie, reg, val); 885 if (error || reg != VM_REG_GUEST_RIP) 886 return (error); 887 888 /* Set 'nextrip' to match the value of %rip */ 889 VMM_CTR1(vcpu, "Setting nextrip to %#lx", val); 890 vcpu->nextrip = val; 891 return (0); 892 } 893 894 static bool 895 is_descriptor_table(int reg) 896 { 897 898 switch (reg) { 899 case VM_REG_GUEST_IDTR: 900 case VM_REG_GUEST_GDTR: 901 return (true); 902 default: 903 return (false); 904 } 905 } 906 907 static bool 908 is_segment_register(int reg) 909 { 910 911 switch (reg) { 912 case VM_REG_GUEST_ES: 913 case VM_REG_GUEST_CS: 914 case VM_REG_GUEST_SS: 915 case VM_REG_GUEST_DS: 916 case VM_REG_GUEST_FS: 917 case VM_REG_GUEST_GS: 918 case VM_REG_GUEST_TR: 919 case VM_REG_GUEST_LDTR: 920 return (true); 921 default: 922 return (false); 923 } 924 } 925 926 int 927 vm_get_seg_desc(struct vcpu *vcpu, int reg, struct seg_desc *desc) 928 { 929 930 if (!is_segment_register(reg) && !is_descriptor_table(reg)) 931 return (EINVAL); 932 933 return (vmmops_getdesc(vcpu->cookie, reg, desc)); 934 } 935 936 int 937 vm_set_seg_desc(struct vcpu *vcpu, int reg, struct seg_desc *desc) 938 { 939 940 if (!is_segment_register(reg) && !is_descriptor_table(reg)) 941 return (EINVAL); 942 943 return (vmmops_setdesc(vcpu->cookie, reg, desc)); 944 } 945 946 static void 947 restore_guest_fpustate(struct vcpu *vcpu) 948 { 949 950 /* flush host state to the pcb */ 951 fpuexit(curthread); 952 953 /* restore guest FPU state */ 954 fpu_enable(); 955 fpurestore(vcpu->guestfpu); 956 957 /* restore guest XCR0 if XSAVE is enabled in the host */ 958 if (rcr4() & CR4_XSAVE) 959 load_xcr(0, vcpu->guest_xcr0); 960 961 /* 962 * The FPU is now "dirty" with the guest's state so disable 963 * the FPU to trap any access by the host. 964 */ 965 fpu_disable(); 966 } 967 968 static void 969 save_guest_fpustate(struct vcpu *vcpu) 970 { 971 972 if ((rcr0() & CR0_TS) == 0) 973 panic("fpu emulation not enabled in host!"); 974 975 /* save guest XCR0 and restore host XCR0 */ 976 if (rcr4() & CR4_XSAVE) { 977 vcpu->guest_xcr0 = rxcr(0); 978 load_xcr(0, vmm_get_host_xcr0()); 979 } 980 981 /* save guest FPU state */ 982 fpu_enable(); 983 fpusave(vcpu->guestfpu); 984 fpu_disable(); 985 } 986 987 static VMM_STAT(VCPU_IDLE_TICKS, "number of ticks vcpu was idle"); 988 989 /* 990 * Invoke the rendezvous function on the specified vcpu if applicable. Return 991 * true if the rendezvous is finished, false otherwise. 992 */ 993 static bool 994 vm_rendezvous(struct vcpu *vcpu) 995 { 996 struct vm *vm = vcpu->vm; 997 int vcpuid; 998 999 mtx_assert(&vcpu->vm->rendezvous_mtx, MA_OWNED); 1000 KASSERT(vcpu->vm->rendezvous_func != NULL, 1001 ("vm_rendezvous: no rendezvous pending")); 1002 1003 /* 'rendezvous_req_cpus' must be a subset of 'active_cpus' */ 1004 CPU_AND(&vm->rendezvous_req_cpus, &vm->rendezvous_req_cpus, 1005 &vm->active_cpus); 1006 1007 vcpuid = vcpu->vcpuid; 1008 if (CPU_ISSET(vcpuid, &vm->rendezvous_req_cpus) && 1009 !CPU_ISSET(vcpuid, &vm->rendezvous_done_cpus)) { 1010 VMM_CTR0(vcpu, "Calling rendezvous func"); 1011 (*vm->rendezvous_func)(vcpu, vm->rendezvous_arg); 1012 CPU_SET(vcpuid, &vm->rendezvous_done_cpus); 1013 } 1014 if (CPU_CMP(&vm->rendezvous_req_cpus, 1015 &vm->rendezvous_done_cpus) == 0) { 1016 VMM_CTR0(vcpu, "Rendezvous completed"); 1017 CPU_ZERO(&vm->rendezvous_req_cpus); 1018 vm->rendezvous_func = NULL; 1019 wakeup(&vm->rendezvous_func); 1020 return (true); 1021 } 1022 return (false); 1023 } 1024 1025 static void 1026 vcpu_wait_idle(struct vcpu *vcpu) 1027 { 1028 KASSERT(vcpu->state != VCPU_IDLE, ("vcpu already idle")); 1029 1030 vcpu->reqidle = 1; 1031 vcpu_notify_event_locked(vcpu, false); 1032 VMM_CTR1(vcpu, "vcpu state change from %s to " 1033 "idle requested", vcpu_state2str(vcpu->state)); 1034 msleep_spin(&vcpu->state, &vcpu->mtx, "vmstat", hz); 1035 } 1036 1037 static int 1038 vcpu_set_state_locked(struct vcpu *vcpu, enum vcpu_state newstate, 1039 bool from_idle) 1040 { 1041 int error; 1042 1043 vcpu_assert_locked(vcpu); 1044 1045 /* 1046 * State transitions from the vmmdev_ioctl() must always begin from 1047 * the VCPU_IDLE state. This guarantees that there is only a single 1048 * ioctl() operating on a vcpu at any point. 1049 */ 1050 if (from_idle) { 1051 while (vcpu->state != VCPU_IDLE) 1052 vcpu_wait_idle(vcpu); 1053 } else { 1054 KASSERT(vcpu->state != VCPU_IDLE, ("invalid transition from " 1055 "vcpu idle state")); 1056 } 1057 1058 if (vcpu->state == VCPU_RUNNING) { 1059 KASSERT(vcpu->hostcpu == curcpu, ("curcpu %d and hostcpu %d " 1060 "mismatch for running vcpu", curcpu, vcpu->hostcpu)); 1061 } else { 1062 KASSERT(vcpu->hostcpu == NOCPU, ("Invalid hostcpu %d for a " 1063 "vcpu that is not running", vcpu->hostcpu)); 1064 } 1065 1066 /* 1067 * The following state transitions are allowed: 1068 * IDLE -> FROZEN -> IDLE 1069 * FROZEN -> RUNNING -> FROZEN 1070 * FROZEN -> SLEEPING -> FROZEN 1071 */ 1072 switch (vcpu->state) { 1073 case VCPU_IDLE: 1074 case VCPU_RUNNING: 1075 case VCPU_SLEEPING: 1076 error = (newstate != VCPU_FROZEN); 1077 break; 1078 case VCPU_FROZEN: 1079 error = (newstate == VCPU_FROZEN); 1080 break; 1081 default: 1082 error = 1; 1083 break; 1084 } 1085 1086 if (error) 1087 return (EBUSY); 1088 1089 VMM_CTR2(vcpu, "vcpu state changed from %s to %s", 1090 vcpu_state2str(vcpu->state), vcpu_state2str(newstate)); 1091 1092 vcpu->state = newstate; 1093 if (newstate == VCPU_RUNNING) 1094 vcpu->hostcpu = curcpu; 1095 else 1096 vcpu->hostcpu = NOCPU; 1097 1098 if (newstate == VCPU_IDLE) 1099 wakeup(&vcpu->state); 1100 1101 return (0); 1102 } 1103 1104 /* 1105 * Try to lock all of the vCPUs in the VM while taking care to avoid deadlocks 1106 * with vm_smp_rendezvous(). 1107 * 1108 * The complexity here suggests that the rendezvous mechanism needs a rethink. 1109 */ 1110 int 1111 vcpu_set_state_all(struct vm *vm, enum vcpu_state newstate) 1112 { 1113 cpuset_t locked; 1114 struct vcpu *vcpu; 1115 int error, i; 1116 uint16_t maxcpus; 1117 1118 KASSERT(newstate != VCPU_IDLE, 1119 ("vcpu_set_state_all: invalid target state %d", newstate)); 1120 1121 error = 0; 1122 CPU_ZERO(&locked); 1123 maxcpus = vm->maxcpus; 1124 1125 mtx_lock(&vm->rendezvous_mtx); 1126 restart: 1127 if (vm->rendezvous_func != NULL) { 1128 /* 1129 * If we have a pending rendezvous, then the initiator may be 1130 * blocked waiting for other vCPUs to execute the callback. The 1131 * current thread may be a vCPU thread so we must not block 1132 * waiting for the initiator, otherwise we get a deadlock. 1133 * Thus, execute the callback on behalf of any idle vCPUs. 1134 */ 1135 for (i = 0; i < maxcpus; i++) { 1136 vcpu = vm_vcpu(vm, i); 1137 if (vcpu == NULL) 1138 continue; 1139 vcpu_lock(vcpu); 1140 if (vcpu->state == VCPU_IDLE) { 1141 (void)vcpu_set_state_locked(vcpu, VCPU_FROZEN, 1142 true); 1143 CPU_SET(i, &locked); 1144 } 1145 if (CPU_ISSET(i, &locked)) { 1146 /* 1147 * We can safely execute the callback on this 1148 * vCPU's behalf. 1149 */ 1150 vcpu_unlock(vcpu); 1151 (void)vm_rendezvous(vcpu); 1152 vcpu_lock(vcpu); 1153 } 1154 vcpu_unlock(vcpu); 1155 } 1156 } 1157 1158 /* 1159 * Now wait for remaining vCPUs to become idle. This may include the 1160 * initiator of a rendezvous that is currently blocked on the rendezvous 1161 * mutex. 1162 */ 1163 CPU_FOREACH_ISCLR(i, &locked) { 1164 if (i >= maxcpus) 1165 break; 1166 vcpu = vm_vcpu(vm, i); 1167 if (vcpu == NULL) 1168 continue; 1169 vcpu_lock(vcpu); 1170 while (vcpu->state != VCPU_IDLE) { 1171 mtx_unlock(&vm->rendezvous_mtx); 1172 vcpu_wait_idle(vcpu); 1173 vcpu_unlock(vcpu); 1174 mtx_lock(&vm->rendezvous_mtx); 1175 if (vm->rendezvous_func != NULL) 1176 goto restart; 1177 vcpu_lock(vcpu); 1178 } 1179 error = vcpu_set_state_locked(vcpu, newstate, true); 1180 vcpu_unlock(vcpu); 1181 if (error != 0) { 1182 /* Roll back state changes. */ 1183 CPU_FOREACH_ISSET(i, &locked) 1184 (void)vcpu_set_state(vcpu, VCPU_IDLE, false); 1185 break; 1186 } 1187 CPU_SET(i, &locked); 1188 } 1189 mtx_unlock(&vm->rendezvous_mtx); 1190 return (error); 1191 } 1192 1193 static void 1194 vcpu_require_state(struct vcpu *vcpu, enum vcpu_state newstate) 1195 { 1196 int error; 1197 1198 if ((error = vcpu_set_state(vcpu, newstate, false)) != 0) 1199 panic("Error %d setting state to %d\n", error, newstate); 1200 } 1201 1202 static void 1203 vcpu_require_state_locked(struct vcpu *vcpu, enum vcpu_state newstate) 1204 { 1205 int error; 1206 1207 if ((error = vcpu_set_state_locked(vcpu, newstate, false)) != 0) 1208 panic("Error %d setting state to %d", error, newstate); 1209 } 1210 1211 static int 1212 vm_handle_rendezvous(struct vcpu *vcpu) 1213 { 1214 struct vm *vm; 1215 struct thread *td; 1216 1217 td = curthread; 1218 vm = vcpu->vm; 1219 1220 mtx_lock(&vm->rendezvous_mtx); 1221 while (vm->rendezvous_func != NULL) { 1222 if (vm_rendezvous(vcpu)) 1223 break; 1224 1225 VMM_CTR0(vcpu, "Wait for rendezvous completion"); 1226 mtx_sleep(&vm->rendezvous_func, &vm->rendezvous_mtx, 0, 1227 "vmrndv", hz); 1228 if (td_ast_pending(td, TDA_SUSPEND)) { 1229 int error; 1230 1231 mtx_unlock(&vm->rendezvous_mtx); 1232 error = thread_check_susp(td, true); 1233 if (error != 0) 1234 return (error); 1235 mtx_lock(&vm->rendezvous_mtx); 1236 } 1237 } 1238 mtx_unlock(&vm->rendezvous_mtx); 1239 return (0); 1240 } 1241 1242 /* 1243 * Emulate a guest 'hlt' by sleeping until the vcpu is ready to run. 1244 */ 1245 static int 1246 vm_handle_hlt(struct vcpu *vcpu, bool intr_disabled, bool *retu) 1247 { 1248 struct vm *vm = vcpu->vm; 1249 const char *wmesg; 1250 struct thread *td; 1251 int error, t, vcpuid, vcpu_halted, vm_halted; 1252 1253 vcpuid = vcpu->vcpuid; 1254 vcpu_halted = 0; 1255 vm_halted = 0; 1256 error = 0; 1257 td = curthread; 1258 1259 KASSERT(!CPU_ISSET(vcpuid, &vm->halted_cpus), ("vcpu already halted")); 1260 1261 vcpu_lock(vcpu); 1262 while (1) { 1263 /* 1264 * Do a final check for pending NMI or interrupts before 1265 * really putting this thread to sleep. Also check for 1266 * software events that would cause this vcpu to wakeup. 1267 * 1268 * These interrupts/events could have happened after the 1269 * vcpu returned from vmmops_run() and before it acquired the 1270 * vcpu lock above. 1271 */ 1272 if (vm->rendezvous_func != NULL || vm->suspend || vcpu->reqidle) 1273 break; 1274 if (vm_nmi_pending(vcpu)) 1275 break; 1276 if (!intr_disabled) { 1277 if (vm_extint_pending(vcpu) || 1278 vlapic_pending_intr(vcpu->vlapic, NULL)) { 1279 break; 1280 } 1281 } 1282 1283 /* Don't go to sleep if the vcpu thread needs to yield */ 1284 if (vcpu_should_yield(vcpu)) 1285 break; 1286 1287 if (vcpu_debugged(vcpu)) 1288 break; 1289 1290 /* 1291 * Some Linux guests implement "halt" by having all vcpus 1292 * execute HLT with interrupts disabled. 'halted_cpus' keeps 1293 * track of the vcpus that have entered this state. When all 1294 * vcpus enter the halted state the virtual machine is halted. 1295 */ 1296 if (intr_disabled) { 1297 wmesg = "vmhalt"; 1298 VMM_CTR0(vcpu, "Halted"); 1299 if (!vcpu_halted && halt_detection_enabled) { 1300 vcpu_halted = 1; 1301 CPU_SET_ATOMIC(vcpuid, &vm->halted_cpus); 1302 } 1303 if (CPU_CMP(&vm->halted_cpus, &vm->active_cpus) == 0) { 1304 vm_halted = 1; 1305 break; 1306 } 1307 } else { 1308 wmesg = "vmidle"; 1309 } 1310 1311 t = ticks; 1312 vcpu_require_state_locked(vcpu, VCPU_SLEEPING); 1313 /* 1314 * XXX msleep_spin() cannot be interrupted by signals so 1315 * wake up periodically to check pending signals. 1316 */ 1317 msleep_spin(vcpu, &vcpu->mtx, wmesg, hz); 1318 vcpu_require_state_locked(vcpu, VCPU_FROZEN); 1319 vmm_stat_incr(vcpu, VCPU_IDLE_TICKS, ticks - t); 1320 if (td_ast_pending(td, TDA_SUSPEND)) { 1321 vcpu_unlock(vcpu); 1322 error = thread_check_susp(td, false); 1323 if (error != 0) { 1324 if (vcpu_halted) { 1325 CPU_CLR_ATOMIC(vcpuid, 1326 &vm->halted_cpus); 1327 } 1328 return (error); 1329 } 1330 vcpu_lock(vcpu); 1331 } 1332 } 1333 1334 if (vcpu_halted) 1335 CPU_CLR_ATOMIC(vcpuid, &vm->halted_cpus); 1336 1337 vcpu_unlock(vcpu); 1338 1339 if (vm_halted) 1340 vm_suspend(vm, VM_SUSPEND_HALT); 1341 1342 return (0); 1343 } 1344 1345 static int 1346 vm_handle_paging(struct vcpu *vcpu, bool *retu) 1347 { 1348 struct vm *vm = vcpu->vm; 1349 int rv, ftype; 1350 struct vm_map *map; 1351 struct vm_exit *vme; 1352 1353 vme = &vcpu->exitinfo; 1354 1355 KASSERT(vme->inst_length == 0, ("%s: invalid inst_length %d", 1356 __func__, vme->inst_length)); 1357 1358 ftype = vme->u.paging.fault_type; 1359 KASSERT(ftype == VM_PROT_READ || 1360 ftype == VM_PROT_WRITE || ftype == VM_PROT_EXECUTE, 1361 ("vm_handle_paging: invalid fault_type %d", ftype)); 1362 1363 if (ftype == VM_PROT_READ || ftype == VM_PROT_WRITE) { 1364 rv = pmap_emulate_accessed_dirty(vmspace_pmap(vm_vmspace(vm)), 1365 vme->u.paging.gpa, ftype); 1366 if (rv == 0) { 1367 VMM_CTR2(vcpu, "%s bit emulation for gpa %#lx", 1368 ftype == VM_PROT_READ ? "accessed" : "dirty", 1369 vme->u.paging.gpa); 1370 goto done; 1371 } 1372 } 1373 1374 map = &vm_vmspace(vm)->vm_map; 1375 rv = vm_fault(map, vme->u.paging.gpa, ftype, VM_FAULT_NORMAL, NULL); 1376 1377 VMM_CTR3(vcpu, "vm_handle_paging rv = %d, gpa = %#lx, " 1378 "ftype = %d", rv, vme->u.paging.gpa, ftype); 1379 1380 if (rv != KERN_SUCCESS) 1381 return (EFAULT); 1382 done: 1383 return (0); 1384 } 1385 1386 static int 1387 vm_handle_inst_emul(struct vcpu *vcpu, bool *retu) 1388 { 1389 struct vie *vie; 1390 struct vm_exit *vme; 1391 uint64_t gla, gpa, cs_base; 1392 struct vm_guest_paging *paging; 1393 mem_region_read_t mread; 1394 mem_region_write_t mwrite; 1395 enum vm_cpu_mode cpu_mode; 1396 int cs_d, error, fault; 1397 1398 vme = &vcpu->exitinfo; 1399 1400 KASSERT(vme->inst_length == 0, ("%s: invalid inst_length %d", 1401 __func__, vme->inst_length)); 1402 1403 gla = vme->u.inst_emul.gla; 1404 gpa = vme->u.inst_emul.gpa; 1405 cs_base = vme->u.inst_emul.cs_base; 1406 cs_d = vme->u.inst_emul.cs_d; 1407 vie = &vme->u.inst_emul.vie; 1408 paging = &vme->u.inst_emul.paging; 1409 cpu_mode = paging->cpu_mode; 1410 1411 VMM_CTR1(vcpu, "inst_emul fault accessing gpa %#lx", gpa); 1412 1413 /* Fetch, decode and emulate the faulting instruction */ 1414 if (vie->num_valid == 0) { 1415 error = vmm_fetch_instruction(vcpu, paging, vme->rip + cs_base, 1416 VIE_INST_SIZE, vie, &fault); 1417 } else { 1418 /* 1419 * The instruction bytes have already been copied into 'vie' 1420 */ 1421 error = fault = 0; 1422 } 1423 if (error || fault) 1424 return (error); 1425 1426 if (vmm_decode_instruction(vcpu, gla, cpu_mode, cs_d, vie) != 0) { 1427 VMM_CTR1(vcpu, "Error decoding instruction at %#lx", 1428 vme->rip + cs_base); 1429 *retu = true; /* dump instruction bytes in userspace */ 1430 return (0); 1431 } 1432 1433 /* 1434 * Update 'nextrip' based on the length of the emulated instruction. 1435 */ 1436 vme->inst_length = vie->num_processed; 1437 vcpu->nextrip += vie->num_processed; 1438 VMM_CTR1(vcpu, "nextrip updated to %#lx after instruction decoding", 1439 vcpu->nextrip); 1440 1441 /* return to userland unless this is an in-kernel emulated device */ 1442 if (gpa >= DEFAULT_APIC_BASE && gpa < DEFAULT_APIC_BASE + PAGE_SIZE) { 1443 mread = lapic_mmio_read; 1444 mwrite = lapic_mmio_write; 1445 } else if (gpa >= VIOAPIC_BASE && gpa < VIOAPIC_BASE + VIOAPIC_SIZE) { 1446 mread = vioapic_mmio_read; 1447 mwrite = vioapic_mmio_write; 1448 } else if (gpa >= VHPET_BASE && gpa < VHPET_BASE + VHPET_SIZE) { 1449 mread = vhpet_mmio_read; 1450 mwrite = vhpet_mmio_write; 1451 } else { 1452 *retu = true; 1453 return (0); 1454 } 1455 1456 error = vmm_emulate_instruction(vcpu, gpa, vie, paging, mread, mwrite, 1457 retu); 1458 1459 return (error); 1460 } 1461 1462 static int 1463 vm_handle_suspend(struct vcpu *vcpu, bool *retu) 1464 { 1465 struct vm *vm = vcpu->vm; 1466 int error, i; 1467 struct thread *td; 1468 1469 error = 0; 1470 td = curthread; 1471 1472 CPU_SET_ATOMIC(vcpu->vcpuid, &vm->suspended_cpus); 1473 1474 /* 1475 * Wait until all 'active_cpus' have suspended themselves. 1476 * 1477 * Since a VM may be suspended at any time including when one or 1478 * more vcpus are doing a rendezvous we need to call the rendezvous 1479 * handler while we are waiting to prevent a deadlock. 1480 */ 1481 vcpu_lock(vcpu); 1482 while (error == 0) { 1483 if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) { 1484 VMM_CTR0(vcpu, "All vcpus suspended"); 1485 break; 1486 } 1487 1488 if (vm->rendezvous_func == NULL) { 1489 VMM_CTR0(vcpu, "Sleeping during suspend"); 1490 vcpu_require_state_locked(vcpu, VCPU_SLEEPING); 1491 msleep_spin(vcpu, &vcpu->mtx, "vmsusp", hz); 1492 vcpu_require_state_locked(vcpu, VCPU_FROZEN); 1493 if (td_ast_pending(td, TDA_SUSPEND)) { 1494 vcpu_unlock(vcpu); 1495 error = thread_check_susp(td, false); 1496 vcpu_lock(vcpu); 1497 } 1498 } else { 1499 VMM_CTR0(vcpu, "Rendezvous during suspend"); 1500 vcpu_unlock(vcpu); 1501 error = vm_handle_rendezvous(vcpu); 1502 vcpu_lock(vcpu); 1503 } 1504 } 1505 vcpu_unlock(vcpu); 1506 1507 /* 1508 * Wakeup the other sleeping vcpus and return to userspace. 1509 */ 1510 for (i = 0; i < vm->maxcpus; i++) { 1511 if (CPU_ISSET(i, &vm->suspended_cpus)) { 1512 vcpu_notify_event(vm_vcpu(vm, i), false); 1513 } 1514 } 1515 1516 *retu = true; 1517 return (error); 1518 } 1519 1520 static int 1521 vm_handle_reqidle(struct vcpu *vcpu, bool *retu) 1522 { 1523 vcpu_lock(vcpu); 1524 KASSERT(vcpu->reqidle, ("invalid vcpu reqidle %d", vcpu->reqidle)); 1525 vcpu->reqidle = 0; 1526 vcpu_unlock(vcpu); 1527 *retu = true; 1528 return (0); 1529 } 1530 1531 static int 1532 vm_handle_db(struct vcpu *vcpu, struct vm_exit *vme, bool *retu) 1533 { 1534 int error, fault; 1535 uint64_t rsp; 1536 uint64_t rflags; 1537 struct vm_copyinfo copyinfo[2]; 1538 1539 *retu = true; 1540 if (!vme->u.dbg.pushf_intercept || vme->u.dbg.tf_shadow_val != 0) { 1541 return (0); 1542 } 1543 1544 vm_get_register(vcpu, VM_REG_GUEST_RSP, &rsp); 1545 error = vm_copy_setup(vcpu, &vme->u.dbg.paging, rsp, sizeof(uint64_t), 1546 VM_PROT_RW, copyinfo, nitems(copyinfo), &fault); 1547 if (error != 0 || fault != 0) { 1548 *retu = false; 1549 return (EINVAL); 1550 } 1551 1552 /* Read pushed rflags value from top of stack. */ 1553 vm_copyin(copyinfo, &rflags, sizeof(uint64_t)); 1554 1555 /* Clear TF bit. */ 1556 rflags &= ~(PSL_T); 1557 1558 /* Write updated value back to memory. */ 1559 vm_copyout(&rflags, copyinfo, sizeof(uint64_t)); 1560 vm_copy_teardown(copyinfo, nitems(copyinfo)); 1561 1562 return (0); 1563 } 1564 1565 int 1566 vm_suspend(struct vm *vm, enum vm_suspend_how how) 1567 { 1568 int i; 1569 1570 if (how <= VM_SUSPEND_NONE || how >= VM_SUSPEND_LAST) 1571 return (EINVAL); 1572 1573 if (atomic_cmpset_int(&vm->suspend, 0, how) == 0) { 1574 VM_CTR2(vm, "virtual machine already suspended %d/%d", 1575 vm->suspend, how); 1576 return (EALREADY); 1577 } 1578 1579 VM_CTR1(vm, "virtual machine successfully suspended %d", how); 1580 1581 /* 1582 * Notify all active vcpus that they are now suspended. 1583 */ 1584 for (i = 0; i < vm->maxcpus; i++) { 1585 if (CPU_ISSET(i, &vm->active_cpus)) 1586 vcpu_notify_event(vm_vcpu(vm, i), false); 1587 } 1588 1589 return (0); 1590 } 1591 1592 void 1593 vm_exit_suspended(struct vcpu *vcpu, uint64_t rip) 1594 { 1595 struct vm *vm = vcpu->vm; 1596 struct vm_exit *vmexit; 1597 1598 KASSERT(vm->suspend > VM_SUSPEND_NONE && vm->suspend < VM_SUSPEND_LAST, 1599 ("vm_exit_suspended: invalid suspend type %d", vm->suspend)); 1600 1601 vmexit = vm_exitinfo(vcpu); 1602 vmexit->rip = rip; 1603 vmexit->inst_length = 0; 1604 vmexit->exitcode = VM_EXITCODE_SUSPENDED; 1605 vmexit->u.suspended.how = vm->suspend; 1606 } 1607 1608 void 1609 vm_exit_debug(struct vcpu *vcpu, uint64_t rip) 1610 { 1611 struct vm_exit *vmexit; 1612 1613 vmexit = vm_exitinfo(vcpu); 1614 vmexit->rip = rip; 1615 vmexit->inst_length = 0; 1616 vmexit->exitcode = VM_EXITCODE_DEBUG; 1617 } 1618 1619 void 1620 vm_exit_rendezvous(struct vcpu *vcpu, uint64_t rip) 1621 { 1622 struct vm_exit *vmexit; 1623 1624 vmexit = vm_exitinfo(vcpu); 1625 vmexit->rip = rip; 1626 vmexit->inst_length = 0; 1627 vmexit->exitcode = VM_EXITCODE_RENDEZVOUS; 1628 vmm_stat_incr(vcpu, VMEXIT_RENDEZVOUS, 1); 1629 } 1630 1631 void 1632 vm_exit_reqidle(struct vcpu *vcpu, uint64_t rip) 1633 { 1634 struct vm_exit *vmexit; 1635 1636 vmexit = vm_exitinfo(vcpu); 1637 vmexit->rip = rip; 1638 vmexit->inst_length = 0; 1639 vmexit->exitcode = VM_EXITCODE_REQIDLE; 1640 vmm_stat_incr(vcpu, VMEXIT_REQIDLE, 1); 1641 } 1642 1643 void 1644 vm_exit_astpending(struct vcpu *vcpu, uint64_t rip) 1645 { 1646 struct vm_exit *vmexit; 1647 1648 vmexit = vm_exitinfo(vcpu); 1649 vmexit->rip = rip; 1650 vmexit->inst_length = 0; 1651 vmexit->exitcode = VM_EXITCODE_BOGUS; 1652 vmm_stat_incr(vcpu, VMEXIT_ASTPENDING, 1); 1653 } 1654 1655 int 1656 vm_run(struct vcpu *vcpu) 1657 { 1658 struct vm *vm = vcpu->vm; 1659 struct vm_eventinfo evinfo; 1660 int error, vcpuid; 1661 struct pcb *pcb; 1662 uint64_t tscval; 1663 struct vm_exit *vme; 1664 bool retu, intr_disabled; 1665 pmap_t pmap; 1666 1667 vcpuid = vcpu->vcpuid; 1668 1669 if (!CPU_ISSET(vcpuid, &vm->active_cpus)) 1670 return (EINVAL); 1671 1672 if (CPU_ISSET(vcpuid, &vm->suspended_cpus)) 1673 return (EINVAL); 1674 1675 pmap = vmspace_pmap(vm_vmspace(vm)); 1676 vme = &vcpu->exitinfo; 1677 evinfo.rptr = &vm->rendezvous_req_cpus; 1678 evinfo.sptr = &vm->suspend; 1679 evinfo.iptr = &vcpu->reqidle; 1680 restart: 1681 critical_enter(); 1682 1683 KASSERT(!CPU_ISSET(curcpu, &pmap->pm_active), 1684 ("vm_run: absurd pm_active")); 1685 1686 tscval = rdtsc(); 1687 1688 pcb = PCPU_GET(curpcb); 1689 set_pcb_flags(pcb, PCB_FULL_IRET); 1690 1691 restore_guest_fpustate(vcpu); 1692 1693 vcpu_require_state(vcpu, VCPU_RUNNING); 1694 error = vmmops_run(vcpu->cookie, vcpu->nextrip, pmap, &evinfo); 1695 vcpu_require_state(vcpu, VCPU_FROZEN); 1696 1697 save_guest_fpustate(vcpu); 1698 1699 vmm_stat_incr(vcpu, VCPU_TOTAL_RUNTIME, rdtsc() - tscval); 1700 1701 critical_exit(); 1702 1703 if (error == 0) { 1704 retu = false; 1705 vcpu->nextrip = vme->rip + vme->inst_length; 1706 switch (vme->exitcode) { 1707 case VM_EXITCODE_REQIDLE: 1708 error = vm_handle_reqidle(vcpu, &retu); 1709 break; 1710 case VM_EXITCODE_SUSPENDED: 1711 error = vm_handle_suspend(vcpu, &retu); 1712 break; 1713 case VM_EXITCODE_IOAPIC_EOI: 1714 vioapic_process_eoi(vm, vme->u.ioapic_eoi.vector); 1715 break; 1716 case VM_EXITCODE_RENDEZVOUS: 1717 error = vm_handle_rendezvous(vcpu); 1718 break; 1719 case VM_EXITCODE_HLT: 1720 intr_disabled = ((vme->u.hlt.rflags & PSL_I) == 0); 1721 error = vm_handle_hlt(vcpu, intr_disabled, &retu); 1722 break; 1723 case VM_EXITCODE_PAGING: 1724 error = vm_handle_paging(vcpu, &retu); 1725 break; 1726 case VM_EXITCODE_INST_EMUL: 1727 error = vm_handle_inst_emul(vcpu, &retu); 1728 break; 1729 case VM_EXITCODE_INOUT: 1730 case VM_EXITCODE_INOUT_STR: 1731 error = vm_handle_inout(vcpu, vme, &retu); 1732 break; 1733 case VM_EXITCODE_DB: 1734 error = vm_handle_db(vcpu, vme, &retu); 1735 break; 1736 case VM_EXITCODE_MONITOR: 1737 case VM_EXITCODE_MWAIT: 1738 case VM_EXITCODE_VMINSN: 1739 vm_inject_ud(vcpu); 1740 break; 1741 default: 1742 retu = true; /* handled in userland */ 1743 break; 1744 } 1745 } 1746 1747 /* 1748 * VM_EXITCODE_INST_EMUL could access the apic which could transform the 1749 * exit code into VM_EXITCODE_IPI. 1750 */ 1751 if (error == 0 && vme->exitcode == VM_EXITCODE_IPI) 1752 error = vm_handle_ipi(vcpu, vme, &retu); 1753 1754 if (error == 0 && retu == false) 1755 goto restart; 1756 1757 vmm_stat_incr(vcpu, VMEXIT_USERSPACE, 1); 1758 VMM_CTR2(vcpu, "retu %d/%d", error, vme->exitcode); 1759 1760 return (error); 1761 } 1762 1763 int 1764 vm_restart_instruction(struct vcpu *vcpu) 1765 { 1766 enum vcpu_state state; 1767 uint64_t rip; 1768 int error __diagused; 1769 1770 state = vcpu_get_state(vcpu, NULL); 1771 if (state == VCPU_RUNNING) { 1772 /* 1773 * When a vcpu is "running" the next instruction is determined 1774 * by adding 'rip' and 'inst_length' in the vcpu's 'exitinfo'. 1775 * Thus setting 'inst_length' to zero will cause the current 1776 * instruction to be restarted. 1777 */ 1778 vcpu->exitinfo.inst_length = 0; 1779 VMM_CTR1(vcpu, "restarting instruction at %#lx by " 1780 "setting inst_length to zero", vcpu->exitinfo.rip); 1781 } else if (state == VCPU_FROZEN) { 1782 /* 1783 * When a vcpu is "frozen" it is outside the critical section 1784 * around vmmops_run() and 'nextrip' points to the next 1785 * instruction. Thus instruction restart is achieved by setting 1786 * 'nextrip' to the vcpu's %rip. 1787 */ 1788 error = vm_get_register(vcpu, VM_REG_GUEST_RIP, &rip); 1789 KASSERT(!error, ("%s: error %d getting rip", __func__, error)); 1790 VMM_CTR2(vcpu, "restarting instruction by updating " 1791 "nextrip from %#lx to %#lx", vcpu->nextrip, rip); 1792 vcpu->nextrip = rip; 1793 } else { 1794 panic("%s: invalid state %d", __func__, state); 1795 } 1796 return (0); 1797 } 1798 1799 int 1800 vm_exit_intinfo(struct vcpu *vcpu, uint64_t info) 1801 { 1802 int type, vector; 1803 1804 if (info & VM_INTINFO_VALID) { 1805 type = info & VM_INTINFO_TYPE; 1806 vector = info & 0xff; 1807 if (type == VM_INTINFO_NMI && vector != IDT_NMI) 1808 return (EINVAL); 1809 if (type == VM_INTINFO_HWEXCEPTION && vector >= 32) 1810 return (EINVAL); 1811 if (info & VM_INTINFO_RSVD) 1812 return (EINVAL); 1813 } else { 1814 info = 0; 1815 } 1816 VMM_CTR2(vcpu, "%s: info1(%#lx)", __func__, info); 1817 vcpu->exitintinfo = info; 1818 return (0); 1819 } 1820 1821 enum exc_class { 1822 EXC_BENIGN, 1823 EXC_CONTRIBUTORY, 1824 EXC_PAGEFAULT 1825 }; 1826 1827 #define IDT_VE 20 /* Virtualization Exception (Intel specific) */ 1828 1829 static enum exc_class 1830 exception_class(uint64_t info) 1831 { 1832 int type, vector; 1833 1834 KASSERT(info & VM_INTINFO_VALID, ("intinfo must be valid: %#lx", info)); 1835 type = info & VM_INTINFO_TYPE; 1836 vector = info & 0xff; 1837 1838 /* Table 6-4, "Interrupt and Exception Classes", Intel SDM, Vol 3 */ 1839 switch (type) { 1840 case VM_INTINFO_HWINTR: 1841 case VM_INTINFO_SWINTR: 1842 case VM_INTINFO_NMI: 1843 return (EXC_BENIGN); 1844 default: 1845 /* 1846 * Hardware exception. 1847 * 1848 * SVM and VT-x use identical type values to represent NMI, 1849 * hardware interrupt and software interrupt. 1850 * 1851 * SVM uses type '3' for all exceptions. VT-x uses type '3' 1852 * for exceptions except #BP and #OF. #BP and #OF use a type 1853 * value of '5' or '6'. Therefore we don't check for explicit 1854 * values of 'type' to classify 'intinfo' into a hardware 1855 * exception. 1856 */ 1857 break; 1858 } 1859 1860 switch (vector) { 1861 case IDT_PF: 1862 case IDT_VE: 1863 return (EXC_PAGEFAULT); 1864 case IDT_DE: 1865 case IDT_TS: 1866 case IDT_NP: 1867 case IDT_SS: 1868 case IDT_GP: 1869 return (EXC_CONTRIBUTORY); 1870 default: 1871 return (EXC_BENIGN); 1872 } 1873 } 1874 1875 static int 1876 nested_fault(struct vcpu *vcpu, uint64_t info1, uint64_t info2, 1877 uint64_t *retinfo) 1878 { 1879 enum exc_class exc1, exc2; 1880 int type1, vector1; 1881 1882 KASSERT(info1 & VM_INTINFO_VALID, ("info1 %#lx is not valid", info1)); 1883 KASSERT(info2 & VM_INTINFO_VALID, ("info2 %#lx is not valid", info2)); 1884 1885 /* 1886 * If an exception occurs while attempting to call the double-fault 1887 * handler the processor enters shutdown mode (aka triple fault). 1888 */ 1889 type1 = info1 & VM_INTINFO_TYPE; 1890 vector1 = info1 & 0xff; 1891 if (type1 == VM_INTINFO_HWEXCEPTION && vector1 == IDT_DF) { 1892 VMM_CTR2(vcpu, "triple fault: info1(%#lx), info2(%#lx)", 1893 info1, info2); 1894 vm_suspend(vcpu->vm, VM_SUSPEND_TRIPLEFAULT); 1895 *retinfo = 0; 1896 return (0); 1897 } 1898 1899 /* 1900 * Table 6-5 "Conditions for Generating a Double Fault", Intel SDM, Vol3 1901 */ 1902 exc1 = exception_class(info1); 1903 exc2 = exception_class(info2); 1904 if ((exc1 == EXC_CONTRIBUTORY && exc2 == EXC_CONTRIBUTORY) || 1905 (exc1 == EXC_PAGEFAULT && exc2 != EXC_BENIGN)) { 1906 /* Convert nested fault into a double fault. */ 1907 *retinfo = IDT_DF; 1908 *retinfo |= VM_INTINFO_VALID | VM_INTINFO_HWEXCEPTION; 1909 *retinfo |= VM_INTINFO_DEL_ERRCODE; 1910 } else { 1911 /* Handle exceptions serially */ 1912 *retinfo = info2; 1913 } 1914 return (1); 1915 } 1916 1917 static uint64_t 1918 vcpu_exception_intinfo(struct vcpu *vcpu) 1919 { 1920 uint64_t info = 0; 1921 1922 if (vcpu->exception_pending) { 1923 info = vcpu->exc_vector & 0xff; 1924 info |= VM_INTINFO_VALID | VM_INTINFO_HWEXCEPTION; 1925 if (vcpu->exc_errcode_valid) { 1926 info |= VM_INTINFO_DEL_ERRCODE; 1927 info |= (uint64_t)vcpu->exc_errcode << 32; 1928 } 1929 } 1930 return (info); 1931 } 1932 1933 int 1934 vm_entry_intinfo(struct vcpu *vcpu, uint64_t *retinfo) 1935 { 1936 uint64_t info1, info2; 1937 int valid; 1938 1939 info1 = vcpu->exitintinfo; 1940 vcpu->exitintinfo = 0; 1941 1942 info2 = 0; 1943 if (vcpu->exception_pending) { 1944 info2 = vcpu_exception_intinfo(vcpu); 1945 vcpu->exception_pending = 0; 1946 VMM_CTR2(vcpu, "Exception %d delivered: %#lx", 1947 vcpu->exc_vector, info2); 1948 } 1949 1950 if ((info1 & VM_INTINFO_VALID) && (info2 & VM_INTINFO_VALID)) { 1951 valid = nested_fault(vcpu, info1, info2, retinfo); 1952 } else if (info1 & VM_INTINFO_VALID) { 1953 *retinfo = info1; 1954 valid = 1; 1955 } else if (info2 & VM_INTINFO_VALID) { 1956 *retinfo = info2; 1957 valid = 1; 1958 } else { 1959 valid = 0; 1960 } 1961 1962 if (valid) { 1963 VMM_CTR4(vcpu, "%s: info1(%#lx), info2(%#lx), " 1964 "retinfo(%#lx)", __func__, info1, info2, *retinfo); 1965 } 1966 1967 return (valid); 1968 } 1969 1970 int 1971 vm_get_intinfo(struct vcpu *vcpu, uint64_t *info1, uint64_t *info2) 1972 { 1973 *info1 = vcpu->exitintinfo; 1974 *info2 = vcpu_exception_intinfo(vcpu); 1975 return (0); 1976 } 1977 1978 int 1979 vm_inject_exception(struct vcpu *vcpu, int vector, int errcode_valid, 1980 uint32_t errcode, int restart_instruction) 1981 { 1982 uint64_t regval; 1983 int error __diagused; 1984 1985 if (vector < 0 || vector >= 32) 1986 return (EINVAL); 1987 1988 /* 1989 * A double fault exception should never be injected directly into 1990 * the guest. It is a derived exception that results from specific 1991 * combinations of nested faults. 1992 */ 1993 if (vector == IDT_DF) 1994 return (EINVAL); 1995 1996 if (vcpu->exception_pending) { 1997 VMM_CTR2(vcpu, "Unable to inject exception %d due to " 1998 "pending exception %d", vector, vcpu->exc_vector); 1999 return (EBUSY); 2000 } 2001 2002 if (errcode_valid) { 2003 /* 2004 * Exceptions don't deliver an error code in real mode. 2005 */ 2006 error = vm_get_register(vcpu, VM_REG_GUEST_CR0, ®val); 2007 KASSERT(!error, ("%s: error %d getting CR0", __func__, error)); 2008 if (!(regval & CR0_PE)) 2009 errcode_valid = 0; 2010 } 2011 2012 /* 2013 * From section 26.6.1 "Interruptibility State" in Intel SDM: 2014 * 2015 * Event blocking by "STI" or "MOV SS" is cleared after guest executes 2016 * one instruction or incurs an exception. 2017 */ 2018 error = vm_set_register(vcpu, VM_REG_GUEST_INTR_SHADOW, 0); 2019 KASSERT(error == 0, ("%s: error %d clearing interrupt shadow", 2020 __func__, error)); 2021 2022 if (restart_instruction) 2023 vm_restart_instruction(vcpu); 2024 2025 vcpu->exception_pending = 1; 2026 vcpu->exc_vector = vector; 2027 vcpu->exc_errcode = errcode; 2028 vcpu->exc_errcode_valid = errcode_valid; 2029 VMM_CTR1(vcpu, "Exception %d pending", vector); 2030 return (0); 2031 } 2032 2033 void 2034 vm_inject_fault(struct vcpu *vcpu, int vector, int errcode_valid, int errcode) 2035 { 2036 int error __diagused, restart_instruction; 2037 2038 restart_instruction = 1; 2039 2040 error = vm_inject_exception(vcpu, vector, errcode_valid, 2041 errcode, restart_instruction); 2042 KASSERT(error == 0, ("vm_inject_exception error %d", error)); 2043 } 2044 2045 void 2046 vm_inject_pf(struct vcpu *vcpu, int error_code, uint64_t cr2) 2047 { 2048 int error __diagused; 2049 2050 VMM_CTR2(vcpu, "Injecting page fault: error_code %#x, cr2 %#lx", 2051 error_code, cr2); 2052 2053 error = vm_set_register(vcpu, VM_REG_GUEST_CR2, cr2); 2054 KASSERT(error == 0, ("vm_set_register(cr2) error %d", error)); 2055 2056 vm_inject_fault(vcpu, IDT_PF, 1, error_code); 2057 } 2058 2059 static VMM_STAT(VCPU_NMI_COUNT, "number of NMIs delivered to vcpu"); 2060 2061 int 2062 vm_inject_nmi(struct vcpu *vcpu) 2063 { 2064 2065 vcpu->nmi_pending = 1; 2066 vcpu_notify_event(vcpu, false); 2067 return (0); 2068 } 2069 2070 int 2071 vm_nmi_pending(struct vcpu *vcpu) 2072 { 2073 return (vcpu->nmi_pending); 2074 } 2075 2076 void 2077 vm_nmi_clear(struct vcpu *vcpu) 2078 { 2079 if (vcpu->nmi_pending == 0) 2080 panic("vm_nmi_clear: inconsistent nmi_pending state"); 2081 2082 vcpu->nmi_pending = 0; 2083 vmm_stat_incr(vcpu, VCPU_NMI_COUNT, 1); 2084 } 2085 2086 static VMM_STAT(VCPU_EXTINT_COUNT, "number of ExtINTs delivered to vcpu"); 2087 2088 int 2089 vm_inject_extint(struct vcpu *vcpu) 2090 { 2091 2092 vcpu->extint_pending = 1; 2093 vcpu_notify_event(vcpu, false); 2094 return (0); 2095 } 2096 2097 int 2098 vm_extint_pending(struct vcpu *vcpu) 2099 { 2100 return (vcpu->extint_pending); 2101 } 2102 2103 void 2104 vm_extint_clear(struct vcpu *vcpu) 2105 { 2106 if (vcpu->extint_pending == 0) 2107 panic("vm_extint_clear: inconsistent extint_pending state"); 2108 2109 vcpu->extint_pending = 0; 2110 vmm_stat_incr(vcpu, VCPU_EXTINT_COUNT, 1); 2111 } 2112 2113 int 2114 vm_get_capability(struct vcpu *vcpu, int type, int *retval) 2115 { 2116 if (type < 0 || type >= VM_CAP_MAX) 2117 return (EINVAL); 2118 2119 return (vmmops_getcap(vcpu->cookie, type, retval)); 2120 } 2121 2122 int 2123 vm_set_capability(struct vcpu *vcpu, int type, int val) 2124 { 2125 if (type < 0 || type >= VM_CAP_MAX) 2126 return (EINVAL); 2127 2128 return (vmmops_setcap(vcpu->cookie, type, val)); 2129 } 2130 2131 struct vm * 2132 vcpu_vm(struct vcpu *vcpu) 2133 { 2134 return (vcpu->vm); 2135 } 2136 2137 int 2138 vcpu_vcpuid(struct vcpu *vcpu) 2139 { 2140 return (vcpu->vcpuid); 2141 } 2142 2143 struct vcpu * 2144 vm_vcpu(struct vm *vm, int vcpuid) 2145 { 2146 return (vm->vcpu[vcpuid]); 2147 } 2148 2149 struct vlapic * 2150 vm_lapic(struct vcpu *vcpu) 2151 { 2152 return (vcpu->vlapic); 2153 } 2154 2155 struct vioapic * 2156 vm_ioapic(struct vm *vm) 2157 { 2158 2159 return (vm->vioapic); 2160 } 2161 2162 struct vhpet * 2163 vm_hpet(struct vm *vm) 2164 { 2165 2166 return (vm->vhpet); 2167 } 2168 2169 bool 2170 vmm_is_pptdev(int bus, int slot, int func) 2171 { 2172 int b, f, i, n, s; 2173 char *val, *cp, *cp2; 2174 bool found; 2175 2176 /* 2177 * XXX 2178 * The length of an environment variable is limited to 128 bytes which 2179 * puts an upper limit on the number of passthru devices that may be 2180 * specified using a single environment variable. 2181 * 2182 * Work around this by scanning multiple environment variable 2183 * names instead of a single one - yuck! 2184 */ 2185 const char *names[] = { "pptdevs", "pptdevs2", "pptdevs3", NULL }; 2186 2187 /* set pptdevs="1/2/3 4/5/6 7/8/9 10/11/12" */ 2188 found = false; 2189 for (i = 0; names[i] != NULL && !found; i++) { 2190 cp = val = kern_getenv(names[i]); 2191 while (cp != NULL && *cp != '\0') { 2192 if ((cp2 = strchr(cp, ' ')) != NULL) 2193 *cp2 = '\0'; 2194 2195 n = sscanf(cp, "%d/%d/%d", &b, &s, &f); 2196 if (n == 3 && bus == b && slot == s && func == f) { 2197 found = true; 2198 break; 2199 } 2200 2201 if (cp2 != NULL) 2202 *cp2++ = ' '; 2203 2204 cp = cp2; 2205 } 2206 freeenv(val); 2207 } 2208 return (found); 2209 } 2210 2211 void * 2212 vm_iommu_domain(struct vm *vm) 2213 { 2214 2215 return (vm->iommu); 2216 } 2217 2218 int 2219 vcpu_set_state(struct vcpu *vcpu, enum vcpu_state newstate, bool from_idle) 2220 { 2221 int error; 2222 2223 vcpu_lock(vcpu); 2224 error = vcpu_set_state_locked(vcpu, newstate, from_idle); 2225 vcpu_unlock(vcpu); 2226 2227 return (error); 2228 } 2229 2230 enum vcpu_state 2231 vcpu_get_state(struct vcpu *vcpu, int *hostcpu) 2232 { 2233 enum vcpu_state state; 2234 2235 vcpu_lock(vcpu); 2236 state = vcpu->state; 2237 if (hostcpu != NULL) 2238 *hostcpu = vcpu->hostcpu; 2239 vcpu_unlock(vcpu); 2240 2241 return (state); 2242 } 2243 2244 int 2245 vm_activate_cpu(struct vcpu *vcpu) 2246 { 2247 struct vm *vm = vcpu->vm; 2248 2249 if (CPU_ISSET(vcpu->vcpuid, &vm->active_cpus)) 2250 return (EBUSY); 2251 2252 VMM_CTR0(vcpu, "activated"); 2253 CPU_SET_ATOMIC(vcpu->vcpuid, &vm->active_cpus); 2254 return (0); 2255 } 2256 2257 int 2258 vm_suspend_cpu(struct vm *vm, struct vcpu *vcpu) 2259 { 2260 if (vcpu == NULL) { 2261 vm->debug_cpus = vm->active_cpus; 2262 for (int i = 0; i < vm->maxcpus; i++) { 2263 if (CPU_ISSET(i, &vm->active_cpus)) 2264 vcpu_notify_event(vm_vcpu(vm, i), false); 2265 } 2266 } else { 2267 if (!CPU_ISSET(vcpu->vcpuid, &vm->active_cpus)) 2268 return (EINVAL); 2269 2270 CPU_SET_ATOMIC(vcpu->vcpuid, &vm->debug_cpus); 2271 vcpu_notify_event(vcpu, false); 2272 } 2273 return (0); 2274 } 2275 2276 int 2277 vm_resume_cpu(struct vm *vm, struct vcpu *vcpu) 2278 { 2279 2280 if (vcpu == NULL) { 2281 CPU_ZERO(&vm->debug_cpus); 2282 } else { 2283 if (!CPU_ISSET(vcpu->vcpuid, &vm->debug_cpus)) 2284 return (EINVAL); 2285 2286 CPU_CLR_ATOMIC(vcpu->vcpuid, &vm->debug_cpus); 2287 } 2288 return (0); 2289 } 2290 2291 int 2292 vcpu_debugged(struct vcpu *vcpu) 2293 { 2294 2295 return (CPU_ISSET(vcpu->vcpuid, &vcpu->vm->debug_cpus)); 2296 } 2297 2298 cpuset_t 2299 vm_active_cpus(struct vm *vm) 2300 { 2301 2302 return (vm->active_cpus); 2303 } 2304 2305 cpuset_t 2306 vm_debug_cpus(struct vm *vm) 2307 { 2308 2309 return (vm->debug_cpus); 2310 } 2311 2312 cpuset_t 2313 vm_suspended_cpus(struct vm *vm) 2314 { 2315 2316 return (vm->suspended_cpus); 2317 } 2318 2319 /* 2320 * Returns the subset of vCPUs in tostart that are awaiting startup. 2321 * These vCPUs are also marked as no longer awaiting startup. 2322 */ 2323 cpuset_t 2324 vm_start_cpus(struct vm *vm, const cpuset_t *tostart) 2325 { 2326 cpuset_t set; 2327 2328 mtx_lock(&vm->rendezvous_mtx); 2329 CPU_AND(&set, &vm->startup_cpus, tostart); 2330 CPU_ANDNOT(&vm->startup_cpus, &vm->startup_cpus, &set); 2331 mtx_unlock(&vm->rendezvous_mtx); 2332 return (set); 2333 } 2334 2335 void 2336 vm_await_start(struct vm *vm, const cpuset_t *waiting) 2337 { 2338 mtx_lock(&vm->rendezvous_mtx); 2339 CPU_OR(&vm->startup_cpus, &vm->startup_cpus, waiting); 2340 mtx_unlock(&vm->rendezvous_mtx); 2341 } 2342 2343 void * 2344 vcpu_stats(struct vcpu *vcpu) 2345 { 2346 2347 return (vcpu->stats); 2348 } 2349 2350 int 2351 vm_get_x2apic_state(struct vcpu *vcpu, enum x2apic_state *state) 2352 { 2353 *state = vcpu->x2apic_state; 2354 2355 return (0); 2356 } 2357 2358 int 2359 vm_set_x2apic_state(struct vcpu *vcpu, enum x2apic_state state) 2360 { 2361 if (state >= X2APIC_STATE_LAST) 2362 return (EINVAL); 2363 2364 vcpu->x2apic_state = state; 2365 2366 vlapic_set_x2apic_state(vcpu, state); 2367 2368 return (0); 2369 } 2370 2371 /* 2372 * This function is called to ensure that a vcpu "sees" a pending event 2373 * as soon as possible: 2374 * - If the vcpu thread is sleeping then it is woken up. 2375 * - If the vcpu is running on a different host_cpu then an IPI will be directed 2376 * to the host_cpu to cause the vcpu to trap into the hypervisor. 2377 */ 2378 static void 2379 vcpu_notify_event_locked(struct vcpu *vcpu, bool lapic_intr) 2380 { 2381 int hostcpu; 2382 2383 hostcpu = vcpu->hostcpu; 2384 if (vcpu->state == VCPU_RUNNING) { 2385 KASSERT(hostcpu != NOCPU, ("vcpu running on invalid hostcpu")); 2386 if (hostcpu != curcpu) { 2387 if (lapic_intr) { 2388 vlapic_post_intr(vcpu->vlapic, hostcpu, 2389 vmm_ipinum); 2390 } else { 2391 ipi_cpu(hostcpu, vmm_ipinum); 2392 } 2393 } else { 2394 /* 2395 * If the 'vcpu' is running on 'curcpu' then it must 2396 * be sending a notification to itself (e.g. SELF_IPI). 2397 * The pending event will be picked up when the vcpu 2398 * transitions back to guest context. 2399 */ 2400 } 2401 } else { 2402 KASSERT(hostcpu == NOCPU, ("vcpu state %d not consistent " 2403 "with hostcpu %d", vcpu->state, hostcpu)); 2404 if (vcpu->state == VCPU_SLEEPING) 2405 wakeup_one(vcpu); 2406 } 2407 } 2408 2409 void 2410 vcpu_notify_event(struct vcpu *vcpu, bool lapic_intr) 2411 { 2412 vcpu_lock(vcpu); 2413 vcpu_notify_event_locked(vcpu, lapic_intr); 2414 vcpu_unlock(vcpu); 2415 } 2416 2417 struct vm_mem * 2418 vm_mem(struct vm *vm) 2419 { 2420 return (&vm->mem); 2421 } 2422 2423 int 2424 vm_apicid2vcpuid(struct vm *vm, int apicid) 2425 { 2426 /* 2427 * XXX apic id is assumed to be numerically identical to vcpu id 2428 */ 2429 return (apicid); 2430 } 2431 2432 int 2433 vm_smp_rendezvous(struct vcpu *vcpu, cpuset_t dest, 2434 vm_rendezvous_func_t func, void *arg) 2435 { 2436 struct vm *vm = vcpu->vm; 2437 int error, i; 2438 2439 /* 2440 * Enforce that this function is called without any locks 2441 */ 2442 WITNESS_WARN(WARN_PANIC, NULL, "vm_smp_rendezvous"); 2443 2444 restart: 2445 mtx_lock(&vm->rendezvous_mtx); 2446 if (vm->rendezvous_func != NULL) { 2447 /* 2448 * If a rendezvous is already in progress then we need to 2449 * call the rendezvous handler in case this 'vcpu' is one 2450 * of the targets of the rendezvous. 2451 */ 2452 VMM_CTR0(vcpu, "Rendezvous already in progress"); 2453 mtx_unlock(&vm->rendezvous_mtx); 2454 error = vm_handle_rendezvous(vcpu); 2455 if (error != 0) 2456 return (error); 2457 goto restart; 2458 } 2459 KASSERT(vm->rendezvous_func == NULL, ("vm_smp_rendezvous: previous " 2460 "rendezvous is still in progress")); 2461 2462 VMM_CTR0(vcpu, "Initiating rendezvous"); 2463 vm->rendezvous_req_cpus = dest; 2464 CPU_ZERO(&vm->rendezvous_done_cpus); 2465 vm->rendezvous_arg = arg; 2466 vm->rendezvous_func = func; 2467 mtx_unlock(&vm->rendezvous_mtx); 2468 2469 /* 2470 * Wake up any sleeping vcpus and trigger a VM-exit in any running 2471 * vcpus so they handle the rendezvous as soon as possible. 2472 */ 2473 for (i = 0; i < vm->maxcpus; i++) { 2474 if (CPU_ISSET(i, &dest)) 2475 vcpu_notify_event(vm_vcpu(vm, i), false); 2476 } 2477 2478 return (vm_handle_rendezvous(vcpu)); 2479 } 2480 2481 struct vatpic * 2482 vm_atpic(struct vm *vm) 2483 { 2484 return (vm->vatpic); 2485 } 2486 2487 struct vatpit * 2488 vm_atpit(struct vm *vm) 2489 { 2490 return (vm->vatpit); 2491 } 2492 2493 struct vpmtmr * 2494 vm_pmtmr(struct vm *vm) 2495 { 2496 2497 return (vm->vpmtmr); 2498 } 2499 2500 struct vrtc * 2501 vm_rtc(struct vm *vm) 2502 { 2503 2504 return (vm->vrtc); 2505 } 2506 2507 enum vm_reg_name 2508 vm_segment_name(int seg) 2509 { 2510 static enum vm_reg_name seg_names[] = { 2511 VM_REG_GUEST_ES, 2512 VM_REG_GUEST_CS, 2513 VM_REG_GUEST_SS, 2514 VM_REG_GUEST_DS, 2515 VM_REG_GUEST_FS, 2516 VM_REG_GUEST_GS 2517 }; 2518 2519 KASSERT(seg >= 0 && seg < nitems(seg_names), 2520 ("%s: invalid segment encoding %d", __func__, seg)); 2521 return (seg_names[seg]); 2522 } 2523 2524 void 2525 vm_copy_teardown(struct vm_copyinfo *copyinfo, int num_copyinfo) 2526 { 2527 int idx; 2528 2529 for (idx = 0; idx < num_copyinfo; idx++) { 2530 if (copyinfo[idx].cookie != NULL) 2531 vm_gpa_release(copyinfo[idx].cookie); 2532 } 2533 bzero(copyinfo, num_copyinfo * sizeof(struct vm_copyinfo)); 2534 } 2535 2536 int 2537 vm_copy_setup(struct vcpu *vcpu, struct vm_guest_paging *paging, 2538 uint64_t gla, size_t len, int prot, struct vm_copyinfo *copyinfo, 2539 int num_copyinfo, int *fault) 2540 { 2541 int error, idx, nused; 2542 size_t n, off, remaining; 2543 void *hva, *cookie; 2544 uint64_t gpa; 2545 2546 bzero(copyinfo, sizeof(struct vm_copyinfo) * num_copyinfo); 2547 2548 nused = 0; 2549 remaining = len; 2550 while (remaining > 0) { 2551 if (nused >= num_copyinfo) 2552 return (EFAULT); 2553 error = vm_gla2gpa(vcpu, paging, gla, prot, &gpa, fault); 2554 if (error || *fault) 2555 return (error); 2556 off = gpa & PAGE_MASK; 2557 n = min(remaining, PAGE_SIZE - off); 2558 copyinfo[nused].gpa = gpa; 2559 copyinfo[nused].len = n; 2560 remaining -= n; 2561 gla += n; 2562 nused++; 2563 } 2564 2565 for (idx = 0; idx < nused; idx++) { 2566 hva = vm_gpa_hold(vcpu, copyinfo[idx].gpa, 2567 copyinfo[idx].len, prot, &cookie); 2568 if (hva == NULL) 2569 break; 2570 copyinfo[idx].hva = hva; 2571 copyinfo[idx].cookie = cookie; 2572 } 2573 2574 if (idx != nused) { 2575 vm_copy_teardown(copyinfo, num_copyinfo); 2576 return (EFAULT); 2577 } else { 2578 *fault = 0; 2579 return (0); 2580 } 2581 } 2582 2583 void 2584 vm_copyin(struct vm_copyinfo *copyinfo, void *kaddr, size_t len) 2585 { 2586 char *dst; 2587 int idx; 2588 2589 dst = kaddr; 2590 idx = 0; 2591 while (len > 0) { 2592 bcopy(copyinfo[idx].hva, dst, copyinfo[idx].len); 2593 len -= copyinfo[idx].len; 2594 dst += copyinfo[idx].len; 2595 idx++; 2596 } 2597 } 2598 2599 void 2600 vm_copyout(const void *kaddr, struct vm_copyinfo *copyinfo, size_t len) 2601 { 2602 const char *src; 2603 int idx; 2604 2605 src = kaddr; 2606 idx = 0; 2607 while (len > 0) { 2608 bcopy(src, copyinfo[idx].hva, copyinfo[idx].len); 2609 len -= copyinfo[idx].len; 2610 src += copyinfo[idx].len; 2611 idx++; 2612 } 2613 } 2614 2615 /* 2616 * Return the amount of in-use and wired memory for the VM. Since 2617 * these are global stats, only return the values with for vCPU 0 2618 */ 2619 VMM_STAT_DECLARE(VMM_MEM_RESIDENT); 2620 VMM_STAT_DECLARE(VMM_MEM_WIRED); 2621 2622 static void 2623 vm_get_rescnt(struct vcpu *vcpu, struct vmm_stat_type *stat) 2624 { 2625 2626 if (vcpu->vcpuid == 0) { 2627 vmm_stat_set(vcpu, VMM_MEM_RESIDENT, PAGE_SIZE * 2628 vmspace_resident_count(vm_vmspace(vcpu->vm))); 2629 } 2630 } 2631 2632 static void 2633 vm_get_wiredcnt(struct vcpu *vcpu, struct vmm_stat_type *stat) 2634 { 2635 2636 if (vcpu->vcpuid == 0) { 2637 vmm_stat_set(vcpu, VMM_MEM_WIRED, PAGE_SIZE * 2638 pmap_wired_count(vmspace_pmap(vm_vmspace(vcpu->vm)))); 2639 } 2640 } 2641 2642 VMM_STAT_FUNC(VMM_MEM_RESIDENT, "Resident memory", vm_get_rescnt); 2643 VMM_STAT_FUNC(VMM_MEM_WIRED, "Wired memory", vm_get_wiredcnt); 2644 2645 #ifdef BHYVE_SNAPSHOT 2646 static int 2647 vm_snapshot_vcpus(struct vm *vm, struct vm_snapshot_meta *meta) 2648 { 2649 uint64_t tsc, now; 2650 int ret; 2651 struct vcpu *vcpu; 2652 uint16_t i, maxcpus; 2653 2654 now = rdtsc(); 2655 maxcpus = vm_get_maxcpus(vm); 2656 for (i = 0; i < maxcpus; i++) { 2657 vcpu = vm->vcpu[i]; 2658 if (vcpu == NULL) 2659 continue; 2660 2661 SNAPSHOT_VAR_OR_LEAVE(vcpu->x2apic_state, meta, ret, done); 2662 SNAPSHOT_VAR_OR_LEAVE(vcpu->exitintinfo, meta, ret, done); 2663 SNAPSHOT_VAR_OR_LEAVE(vcpu->exc_vector, meta, ret, done); 2664 SNAPSHOT_VAR_OR_LEAVE(vcpu->exc_errcode_valid, meta, ret, done); 2665 SNAPSHOT_VAR_OR_LEAVE(vcpu->exc_errcode, meta, ret, done); 2666 SNAPSHOT_VAR_OR_LEAVE(vcpu->guest_xcr0, meta, ret, done); 2667 SNAPSHOT_VAR_OR_LEAVE(vcpu->exitinfo, meta, ret, done); 2668 SNAPSHOT_VAR_OR_LEAVE(vcpu->nextrip, meta, ret, done); 2669 2670 /* 2671 * Save the absolute TSC value by adding now to tsc_offset. 2672 * 2673 * It will be turned turned back into an actual offset when the 2674 * TSC restore function is called 2675 */ 2676 tsc = now + vcpu->tsc_offset; 2677 SNAPSHOT_VAR_OR_LEAVE(tsc, meta, ret, done); 2678 if (meta->op == VM_SNAPSHOT_RESTORE) 2679 vcpu->tsc_offset = tsc; 2680 } 2681 2682 done: 2683 return (ret); 2684 } 2685 2686 static int 2687 vm_snapshot_vm(struct vm *vm, struct vm_snapshot_meta *meta) 2688 { 2689 int ret; 2690 2691 ret = vm_snapshot_vcpus(vm, meta); 2692 if (ret != 0) 2693 goto done; 2694 2695 SNAPSHOT_VAR_OR_LEAVE(vm->startup_cpus, meta, ret, done); 2696 done: 2697 return (ret); 2698 } 2699 2700 static int 2701 vm_snapshot_vcpu(struct vm *vm, struct vm_snapshot_meta *meta) 2702 { 2703 int error; 2704 struct vcpu *vcpu; 2705 uint16_t i, maxcpus; 2706 2707 error = 0; 2708 2709 maxcpus = vm_get_maxcpus(vm); 2710 for (i = 0; i < maxcpus; i++) { 2711 vcpu = vm->vcpu[i]; 2712 if (vcpu == NULL) 2713 continue; 2714 2715 error = vmmops_vcpu_snapshot(vcpu->cookie, meta); 2716 if (error != 0) { 2717 printf("%s: failed to snapshot vmcs/vmcb data for " 2718 "vCPU: %d; error: %d\n", __func__, i, error); 2719 goto done; 2720 } 2721 } 2722 2723 done: 2724 return (error); 2725 } 2726 2727 /* 2728 * Save kernel-side structures to user-space for snapshotting. 2729 */ 2730 int 2731 vm_snapshot_req(struct vm *vm, struct vm_snapshot_meta *meta) 2732 { 2733 int ret = 0; 2734 2735 switch (meta->dev_req) { 2736 case STRUCT_VMCX: 2737 ret = vm_snapshot_vcpu(vm, meta); 2738 break; 2739 case STRUCT_VM: 2740 ret = vm_snapshot_vm(vm, meta); 2741 break; 2742 case STRUCT_VIOAPIC: 2743 ret = vioapic_snapshot(vm_ioapic(vm), meta); 2744 break; 2745 case STRUCT_VLAPIC: 2746 ret = vlapic_snapshot(vm, meta); 2747 break; 2748 case STRUCT_VHPET: 2749 ret = vhpet_snapshot(vm_hpet(vm), meta); 2750 break; 2751 case STRUCT_VATPIC: 2752 ret = vatpic_snapshot(vm_atpic(vm), meta); 2753 break; 2754 case STRUCT_VATPIT: 2755 ret = vatpit_snapshot(vm_atpit(vm), meta); 2756 break; 2757 case STRUCT_VPMTMR: 2758 ret = vpmtmr_snapshot(vm_pmtmr(vm), meta); 2759 break; 2760 case STRUCT_VRTC: 2761 ret = vrtc_snapshot(vm_rtc(vm), meta); 2762 break; 2763 default: 2764 printf("%s: failed to find the requested type %#x\n", 2765 __func__, meta->dev_req); 2766 ret = (EINVAL); 2767 } 2768 return (ret); 2769 } 2770 2771 void 2772 vm_set_tsc_offset(struct vcpu *vcpu, uint64_t offset) 2773 { 2774 vcpu->tsc_offset = offset; 2775 } 2776 2777 int 2778 vm_restore_time(struct vm *vm) 2779 { 2780 int error; 2781 uint64_t now; 2782 struct vcpu *vcpu; 2783 uint16_t i, maxcpus; 2784 2785 now = rdtsc(); 2786 2787 error = vhpet_restore_time(vm_hpet(vm)); 2788 if (error) 2789 return (error); 2790 2791 maxcpus = vm_get_maxcpus(vm); 2792 for (i = 0; i < maxcpus; i++) { 2793 vcpu = vm->vcpu[i]; 2794 if (vcpu == NULL) 2795 continue; 2796 2797 error = vmmops_restore_tsc(vcpu->cookie, 2798 vcpu->tsc_offset - now); 2799 if (error) 2800 return (error); 2801 } 2802 2803 return (0); 2804 } 2805 #endif 2806