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 <sys/cdefs.h> 30 __FBSDID("$FreeBSD$"); 31 32 #include "opt_bhyve_snapshot.h" 33 34 #include <sys/param.h> 35 #include <sys/systm.h> 36 #include <sys/kernel.h> 37 #include <sys/module.h> 38 #include <sys/sysctl.h> 39 #include <sys/malloc.h> 40 #include <sys/pcpu.h> 41 #include <sys/lock.h> 42 #include <sys/mutex.h> 43 #include <sys/proc.h> 44 #include <sys/rwlock.h> 45 #include <sys/sched.h> 46 #include <sys/smp.h> 47 #include <sys/sx.h> 48 #include <sys/vnode.h> 49 50 #include <vm/vm.h> 51 #include <vm/vm_param.h> 52 #include <vm/vm_extern.h> 53 #include <vm/vm_object.h> 54 #include <vm/vm_page.h> 55 #include <vm/pmap.h> 56 #include <vm/vm_map.h> 57 #include <vm/vm_pager.h> 58 #include <vm/vm_kern.h> 59 #include <vm/vnode_pager.h> 60 #include <vm/swap_pager.h> 61 #include <vm/uma.h> 62 63 #include <machine/cpu.h> 64 #include <machine/pcb.h> 65 #include <machine/smp.h> 66 #include <machine/md_var.h> 67 #include <x86/psl.h> 68 #include <x86/apicreg.h> 69 #include <x86/ifunc.h> 70 71 #include <machine/vmm.h> 72 #include <machine/vmm_dev.h> 73 #include <machine/vmm_instruction_emul.h> 74 #include <machine/vmm_snapshot.h> 75 76 #include "vmm_ioport.h" 77 #include "vmm_ktr.h" 78 #include "vmm_host.h" 79 #include "vmm_mem.h" 80 #include "vmm_util.h" 81 #include "vatpic.h" 82 #include "vatpit.h" 83 #include "vhpet.h" 84 #include "vioapic.h" 85 #include "vlapic.h" 86 #include "vpmtmr.h" 87 #include "vrtc.h" 88 #include "vmm_stat.h" 89 #include "vmm_lapic.h" 90 91 #include "io/ppt.h" 92 #include "io/iommu.h" 93 94 struct vlapic; 95 96 /* 97 * Initialization: 98 * (a) allocated when vcpu is created 99 * (i) initialized when vcpu is created and when it is reinitialized 100 * (o) initialized the first time the vcpu is created 101 * (x) initialized before use 102 */ 103 struct vcpu { 104 struct mtx mtx; /* (o) protects 'state' and 'hostcpu' */ 105 enum vcpu_state state; /* (o) vcpu state */ 106 int vcpuid; /* (o) */ 107 int hostcpu; /* (o) vcpu's host cpu */ 108 int reqidle; /* (i) request vcpu to idle */ 109 struct vm *vm; /* (o) */ 110 void *cookie; /* (i) cpu-specific data */ 111 struct vlapic *vlapic; /* (i) APIC device model */ 112 enum x2apic_state x2apic_state; /* (i) APIC mode */ 113 uint64_t exitintinfo; /* (i) events pending at VM exit */ 114 int nmi_pending; /* (i) NMI pending */ 115 int extint_pending; /* (i) INTR pending */ 116 int exception_pending; /* (i) exception pending */ 117 int exc_vector; /* (x) exception collateral */ 118 int exc_errcode_valid; 119 uint32_t exc_errcode; 120 struct savefpu *guestfpu; /* (a,i) guest fpu state */ 121 uint64_t guest_xcr0; /* (i) guest %xcr0 register */ 122 void *stats; /* (a,i) statistics */ 123 struct vm_exit exitinfo; /* (x) exit reason and collateral */ 124 cpuset_t exitinfo_cpuset; /* (x) storage for vmexit handlers */ 125 uint64_t nextrip; /* (x) next instruction to execute */ 126 uint64_t tsc_offset; /* (o) TSC offsetting */ 127 }; 128 129 #define vcpu_lock_init(v) mtx_init(&((v)->mtx), "vcpu lock", 0, MTX_SPIN) 130 #define vcpu_lock_destroy(v) mtx_destroy(&((v)->mtx)) 131 #define vcpu_lock(v) mtx_lock_spin(&((v)->mtx)) 132 #define vcpu_unlock(v) mtx_unlock_spin(&((v)->mtx)) 133 #define vcpu_assert_locked(v) mtx_assert(&((v)->mtx), MA_OWNED) 134 135 struct mem_seg { 136 size_t len; 137 bool sysmem; 138 struct vm_object *object; 139 }; 140 #define VM_MAX_MEMSEGS 4 141 142 struct mem_map { 143 vm_paddr_t gpa; 144 size_t len; 145 vm_ooffset_t segoff; 146 int segid; 147 int prot; 148 int flags; 149 }; 150 #define VM_MAX_MEMMAPS 8 151 152 /* 153 * Initialization: 154 * (o) initialized the first time the VM is created 155 * (i) initialized when VM is created and when it is reinitialized 156 * (x) initialized before use 157 * 158 * Locking: 159 * [m] mem_segs_lock 160 * [r] rendezvous_mtx 161 * [v] reads require one frozen vcpu, writes require freezing all vcpus 162 */ 163 struct vm { 164 void *cookie; /* (i) cpu-specific data */ 165 void *iommu; /* (x) iommu-specific data */ 166 struct vhpet *vhpet; /* (i) virtual HPET */ 167 struct vioapic *vioapic; /* (i) virtual ioapic */ 168 struct vatpic *vatpic; /* (i) virtual atpic */ 169 struct vatpit *vatpit; /* (i) virtual atpit */ 170 struct vpmtmr *vpmtmr; /* (i) virtual ACPI PM timer */ 171 struct vrtc *vrtc; /* (o) virtual RTC */ 172 volatile cpuset_t active_cpus; /* (i) active vcpus */ 173 volatile cpuset_t debug_cpus; /* (i) vcpus stopped for debug */ 174 cpuset_t startup_cpus; /* (i) [r] waiting for startup */ 175 int suspend; /* (i) stop VM execution */ 176 bool dying; /* (o) is dying */ 177 volatile cpuset_t suspended_cpus; /* (i) suspended vcpus */ 178 volatile cpuset_t halted_cpus; /* (x) cpus in a hard halt */ 179 cpuset_t rendezvous_req_cpus; /* (x) [r] rendezvous requested */ 180 cpuset_t rendezvous_done_cpus; /* (x) [r] rendezvous finished */ 181 void *rendezvous_arg; /* (x) [r] rendezvous func/arg */ 182 vm_rendezvous_func_t rendezvous_func; 183 struct mtx rendezvous_mtx; /* (o) rendezvous lock */ 184 struct mem_map mem_maps[VM_MAX_MEMMAPS]; /* (i) [m+v] guest address space */ 185 struct mem_seg mem_segs[VM_MAX_MEMSEGS]; /* (o) [m+v] guest memory regions */ 186 struct vmspace *vmspace; /* (o) guest's address space */ 187 char name[VM_MAX_NAMELEN+1]; /* (o) virtual machine name */ 188 struct vcpu **vcpu; /* (o) guest vcpus */ 189 /* The following describe the vm cpu topology */ 190 uint16_t sockets; /* (o) num of sockets */ 191 uint16_t cores; /* (o) num of cores/socket */ 192 uint16_t threads; /* (o) num of threads/core */ 193 uint16_t maxcpus; /* (o) max pluggable cpus */ 194 struct sx mem_segs_lock; /* (o) */ 195 struct sx vcpus_init_lock; /* (o) */ 196 }; 197 198 #define VMM_CTR0(vcpu, format) \ 199 VCPU_CTR0((vcpu)->vm, (vcpu)->vcpuid, format) 200 201 #define VMM_CTR1(vcpu, format, p1) \ 202 VCPU_CTR1((vcpu)->vm, (vcpu)->vcpuid, format, p1) 203 204 #define VMM_CTR2(vcpu, format, p1, p2) \ 205 VCPU_CTR2((vcpu)->vm, (vcpu)->vcpuid, format, p1, p2) 206 207 #define VMM_CTR3(vcpu, format, p1, p2, p3) \ 208 VCPU_CTR3((vcpu)->vm, (vcpu)->vcpuid, format, p1, p2, p3) 209 210 #define VMM_CTR4(vcpu, format, p1, p2, p3, p4) \ 211 VCPU_CTR4((vcpu)->vm, (vcpu)->vcpuid, format, p1, p2, p3, p4) 212 213 static int vmm_initialized; 214 215 static void vmmops_panic(void); 216 217 static void 218 vmmops_panic(void) 219 { 220 panic("vmm_ops func called when !vmm_is_intel() && !vmm_is_svm()"); 221 } 222 223 #define DEFINE_VMMOPS_IFUNC(ret_type, opname, args) \ 224 DEFINE_IFUNC(static, ret_type, vmmops_##opname, args) \ 225 { \ 226 if (vmm_is_intel()) \ 227 return (vmm_ops_intel.opname); \ 228 else if (vmm_is_svm()) \ 229 return (vmm_ops_amd.opname); \ 230 else \ 231 return ((ret_type (*)args)vmmops_panic); \ 232 } 233 234 DEFINE_VMMOPS_IFUNC(int, modinit, (int ipinum)) 235 DEFINE_VMMOPS_IFUNC(int, modcleanup, (void)) 236 DEFINE_VMMOPS_IFUNC(void, modresume, (void)) 237 DEFINE_VMMOPS_IFUNC(void *, init, (struct vm *vm, struct pmap *pmap)) 238 DEFINE_VMMOPS_IFUNC(int, run, (void *vcpui, register_t rip, struct pmap *pmap, 239 struct vm_eventinfo *info)) 240 DEFINE_VMMOPS_IFUNC(void, cleanup, (void *vmi)) 241 DEFINE_VMMOPS_IFUNC(void *, vcpu_init, (void *vmi, struct vcpu *vcpu, 242 int vcpu_id)) 243 DEFINE_VMMOPS_IFUNC(void, vcpu_cleanup, (void *vcpui)) 244 DEFINE_VMMOPS_IFUNC(int, getreg, (void *vcpui, int num, uint64_t *retval)) 245 DEFINE_VMMOPS_IFUNC(int, setreg, (void *vcpui, int num, uint64_t val)) 246 DEFINE_VMMOPS_IFUNC(int, getdesc, (void *vcpui, int num, struct seg_desc *desc)) 247 DEFINE_VMMOPS_IFUNC(int, setdesc, (void *vcpui, int num, struct seg_desc *desc)) 248 DEFINE_VMMOPS_IFUNC(int, getcap, (void *vcpui, int num, int *retval)) 249 DEFINE_VMMOPS_IFUNC(int, setcap, (void *vcpui, int num, int val)) 250 DEFINE_VMMOPS_IFUNC(struct vmspace *, vmspace_alloc, (vm_offset_t min, 251 vm_offset_t max)) 252 DEFINE_VMMOPS_IFUNC(void, vmspace_free, (struct vmspace *vmspace)) 253 DEFINE_VMMOPS_IFUNC(struct vlapic *, vlapic_init, (void *vcpui)) 254 DEFINE_VMMOPS_IFUNC(void, vlapic_cleanup, (struct vlapic *vlapic)) 255 #ifdef BHYVE_SNAPSHOT 256 DEFINE_VMMOPS_IFUNC(int, vcpu_snapshot, (void *vcpui, 257 struct vm_snapshot_meta *meta)) 258 DEFINE_VMMOPS_IFUNC(int, restore_tsc, (void *vcpui, uint64_t now)) 259 #endif 260 261 #define fpu_start_emulating() load_cr0(rcr0() | CR0_TS) 262 #define fpu_stop_emulating() clts() 263 264 SDT_PROVIDER_DEFINE(vmm); 265 266 static MALLOC_DEFINE(M_VM, "vm", "vm"); 267 268 /* statistics */ 269 static VMM_STAT(VCPU_TOTAL_RUNTIME, "vcpu total runtime"); 270 271 SYSCTL_NODE(_hw, OID_AUTO, vmm, CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, 272 NULL); 273 274 /* 275 * Halt the guest if all vcpus are executing a HLT instruction with 276 * interrupts disabled. 277 */ 278 static int halt_detection_enabled = 1; 279 SYSCTL_INT(_hw_vmm, OID_AUTO, halt_detection, CTLFLAG_RDTUN, 280 &halt_detection_enabled, 0, 281 "Halt VM if all vcpus execute HLT with interrupts disabled"); 282 283 static int vmm_ipinum; 284 SYSCTL_INT(_hw_vmm, OID_AUTO, ipinum, CTLFLAG_RD, &vmm_ipinum, 0, 285 "IPI vector used for vcpu notifications"); 286 287 static int trace_guest_exceptions; 288 SYSCTL_INT(_hw_vmm, OID_AUTO, trace_guest_exceptions, CTLFLAG_RDTUN, 289 &trace_guest_exceptions, 0, 290 "Trap into hypervisor on all guest exceptions and reflect them back"); 291 292 static int trap_wbinvd; 293 SYSCTL_INT(_hw_vmm, OID_AUTO, trap_wbinvd, CTLFLAG_RDTUN, &trap_wbinvd, 0, 294 "WBINVD triggers a VM-exit"); 295 296 u_int vm_maxcpu; 297 SYSCTL_UINT(_hw_vmm, OID_AUTO, maxcpu, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, 298 &vm_maxcpu, 0, "Maximum number of vCPUs"); 299 300 static void vm_free_memmap(struct vm *vm, int ident); 301 static bool sysmem_mapping(struct vm *vm, struct mem_map *mm); 302 static void vcpu_notify_event_locked(struct vcpu *vcpu, bool lapic_intr); 303 304 /* 305 * Upper limit on vm_maxcpu. Limited by use of uint16_t types for CPU 306 * counts as well as range of vpid values for VT-x and by the capacity 307 * of cpuset_t masks. The call to new_unrhdr() in vpid_init() in 308 * vmx.c requires 'vm_maxcpu + 1 <= 0xffff', hence the '- 1' below. 309 */ 310 #define VM_MAXCPU MIN(0xffff - 1, CPU_SETSIZE) 311 312 #ifdef KTR 313 static const char * 314 vcpu_state2str(enum vcpu_state state) 315 { 316 317 switch (state) { 318 case VCPU_IDLE: 319 return ("idle"); 320 case VCPU_FROZEN: 321 return ("frozen"); 322 case VCPU_RUNNING: 323 return ("running"); 324 case VCPU_SLEEPING: 325 return ("sleeping"); 326 default: 327 return ("unknown"); 328 } 329 } 330 #endif 331 332 static void 333 vcpu_cleanup(struct vcpu *vcpu, bool destroy) 334 { 335 vmmops_vlapic_cleanup(vcpu->vlapic); 336 vmmops_vcpu_cleanup(vcpu->cookie); 337 vcpu->cookie = NULL; 338 if (destroy) { 339 vmm_stat_free(vcpu->stats); 340 fpu_save_area_free(vcpu->guestfpu); 341 vcpu_lock_destroy(vcpu); 342 free(vcpu, M_VM); 343 } 344 } 345 346 static struct vcpu * 347 vcpu_alloc(struct vm *vm, int vcpu_id) 348 { 349 struct vcpu *vcpu; 350 351 KASSERT(vcpu_id >= 0 && vcpu_id < vm->maxcpus, 352 ("vcpu_init: invalid vcpu %d", vcpu_id)); 353 354 vcpu = malloc(sizeof(*vcpu), M_VM, M_WAITOK | M_ZERO); 355 vcpu_lock_init(vcpu); 356 vcpu->state = VCPU_IDLE; 357 vcpu->hostcpu = NOCPU; 358 vcpu->vcpuid = vcpu_id; 359 vcpu->vm = vm; 360 vcpu->guestfpu = fpu_save_area_alloc(); 361 vcpu->stats = vmm_stat_alloc(); 362 vcpu->tsc_offset = 0; 363 return (vcpu); 364 } 365 366 static void 367 vcpu_init(struct vcpu *vcpu) 368 { 369 vcpu->cookie = vmmops_vcpu_init(vcpu->vm->cookie, vcpu, vcpu->vcpuid); 370 vcpu->vlapic = vmmops_vlapic_init(vcpu->cookie); 371 vm_set_x2apic_state(vcpu, X2APIC_DISABLED); 372 vcpu->reqidle = 0; 373 vcpu->exitintinfo = 0; 374 vcpu->nmi_pending = 0; 375 vcpu->extint_pending = 0; 376 vcpu->exception_pending = 0; 377 vcpu->guest_xcr0 = XFEATURE_ENABLED_X87; 378 fpu_save_area_reset(vcpu->guestfpu); 379 vmm_stat_init(vcpu->stats); 380 } 381 382 int 383 vcpu_trace_exceptions(struct vcpu *vcpu) 384 { 385 386 return (trace_guest_exceptions); 387 } 388 389 int 390 vcpu_trap_wbinvd(struct vcpu *vcpu) 391 { 392 return (trap_wbinvd); 393 } 394 395 struct vm_exit * 396 vm_exitinfo(struct vcpu *vcpu) 397 { 398 return (&vcpu->exitinfo); 399 } 400 401 cpuset_t * 402 vm_exitinfo_cpuset(struct vcpu *vcpu) 403 { 404 return (&vcpu->exitinfo_cpuset); 405 } 406 407 static int 408 vmm_init(void) 409 { 410 int error; 411 412 if (!vmm_is_hw_supported()) 413 return (ENXIO); 414 415 vm_maxcpu = mp_ncpus; 416 TUNABLE_INT_FETCH("hw.vmm.maxcpu", &vm_maxcpu); 417 418 if (vm_maxcpu > VM_MAXCPU) { 419 printf("vmm: vm_maxcpu clamped to %u\n", VM_MAXCPU); 420 vm_maxcpu = VM_MAXCPU; 421 } 422 if (vm_maxcpu == 0) 423 vm_maxcpu = 1; 424 425 vmm_host_state_init(); 426 427 vmm_ipinum = lapic_ipi_alloc(pti ? &IDTVEC(justreturn1_pti) : 428 &IDTVEC(justreturn)); 429 if (vmm_ipinum < 0) 430 vmm_ipinum = IPI_AST; 431 432 error = vmm_mem_init(); 433 if (error) 434 return (error); 435 436 vmm_resume_p = vmmops_modresume; 437 438 return (vmmops_modinit(vmm_ipinum)); 439 } 440 441 static int 442 vmm_handler(module_t mod, int what, void *arg) 443 { 444 int error; 445 446 switch (what) { 447 case MOD_LOAD: 448 if (vmm_is_hw_supported()) { 449 vmmdev_init(); 450 error = vmm_init(); 451 if (error == 0) 452 vmm_initialized = 1; 453 } else { 454 error = ENXIO; 455 } 456 break; 457 case MOD_UNLOAD: 458 if (vmm_is_hw_supported()) { 459 error = vmmdev_cleanup(); 460 if (error == 0) { 461 vmm_resume_p = NULL; 462 iommu_cleanup(); 463 if (vmm_ipinum != IPI_AST) 464 lapic_ipi_free(vmm_ipinum); 465 error = vmmops_modcleanup(); 466 /* 467 * Something bad happened - prevent new 468 * VMs from being created 469 */ 470 if (error) 471 vmm_initialized = 0; 472 } 473 } else { 474 error = 0; 475 } 476 break; 477 default: 478 error = 0; 479 break; 480 } 481 return (error); 482 } 483 484 static moduledata_t vmm_kmod = { 485 "vmm", 486 vmm_handler, 487 NULL 488 }; 489 490 /* 491 * vmm initialization has the following dependencies: 492 * 493 * - VT-x initialization requires smp_rendezvous() and therefore must happen 494 * after SMP is fully functional (after SI_SUB_SMP). 495 */ 496 DECLARE_MODULE(vmm, vmm_kmod, SI_SUB_SMP + 1, SI_ORDER_ANY); 497 MODULE_VERSION(vmm, 1); 498 499 static void 500 vm_init(struct vm *vm, bool create) 501 { 502 vm->cookie = vmmops_init(vm, vmspace_pmap(vm->vmspace)); 503 vm->iommu = NULL; 504 vm->vioapic = vioapic_init(vm); 505 vm->vhpet = vhpet_init(vm); 506 vm->vatpic = vatpic_init(vm); 507 vm->vatpit = vatpit_init(vm); 508 vm->vpmtmr = vpmtmr_init(vm); 509 if (create) 510 vm->vrtc = vrtc_init(vm); 511 512 CPU_ZERO(&vm->active_cpus); 513 CPU_ZERO(&vm->debug_cpus); 514 CPU_ZERO(&vm->startup_cpus); 515 516 vm->suspend = 0; 517 CPU_ZERO(&vm->suspended_cpus); 518 519 if (!create) { 520 for (int i = 0; i < vm->maxcpus; i++) { 521 if (vm->vcpu[i] != NULL) 522 vcpu_init(vm->vcpu[i]); 523 } 524 } 525 } 526 527 void 528 vm_disable_vcpu_creation(struct vm *vm) 529 { 530 sx_xlock(&vm->vcpus_init_lock); 531 vm->dying = true; 532 sx_xunlock(&vm->vcpus_init_lock); 533 } 534 535 struct vcpu * 536 vm_alloc_vcpu(struct vm *vm, int vcpuid) 537 { 538 struct vcpu *vcpu; 539 540 if (vcpuid < 0 || vcpuid >= vm_get_maxcpus(vm)) 541 return (NULL); 542 543 vcpu = atomic_load_ptr(&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_slock_vcpus(struct vm *vm) 566 { 567 sx_slock(&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 struct vmspace *vmspace; 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 vmspace = vmmops_vmspace_alloc(0, VM_MAXUSER_ADDRESS_LA48); 600 if (vmspace == NULL) 601 return (ENOMEM); 602 603 vm = malloc(sizeof(struct vm), M_VM, M_WAITOK | M_ZERO); 604 strcpy(vm->name, name); 605 vm->vmspace = vmspace; 606 mtx_init(&vm->rendezvous_mtx, "vm rendezvous lock", 0, MTX_DEF); 607 sx_init(&vm->mem_segs_lock, "vm mem_segs"); 608 sx_init(&vm->vcpus_init_lock, "vm vcpus"); 609 vm->vcpu = malloc(sizeof(*vm->vcpu) * vm_maxcpu, M_VM, M_WAITOK | 610 M_ZERO); 611 612 vm->sockets = 1; 613 vm->cores = cores_per_package; /* XXX backwards compatibility */ 614 vm->threads = threads_per_core; /* XXX backwards compatibility */ 615 vm->maxcpus = vm_maxcpu; 616 617 vm_init(vm, true); 618 619 *retvm = vm; 620 return (0); 621 } 622 623 void 624 vm_get_topology(struct vm *vm, uint16_t *sockets, uint16_t *cores, 625 uint16_t *threads, uint16_t *maxcpus) 626 { 627 *sockets = vm->sockets; 628 *cores = vm->cores; 629 *threads = vm->threads; 630 *maxcpus = vm->maxcpus; 631 } 632 633 uint16_t 634 vm_get_maxcpus(struct vm *vm) 635 { 636 return (vm->maxcpus); 637 } 638 639 int 640 vm_set_topology(struct vm *vm, uint16_t sockets, uint16_t cores, 641 uint16_t threads, uint16_t maxcpus __unused) 642 { 643 /* Ignore maxcpus. */ 644 if ((sockets * cores * threads) > vm->maxcpus) 645 return (EINVAL); 646 vm->sockets = sockets; 647 vm->cores = cores; 648 vm->threads = threads; 649 return(0); 650 } 651 652 static void 653 vm_cleanup(struct vm *vm, bool destroy) 654 { 655 struct mem_map *mm; 656 int i; 657 658 if (destroy) 659 vm_xlock_memsegs(vm); 660 661 ppt_unassign_all(vm); 662 663 if (vm->iommu != NULL) 664 iommu_destroy_domain(vm->iommu); 665 666 if (destroy) 667 vrtc_cleanup(vm->vrtc); 668 else 669 vrtc_reset(vm->vrtc); 670 vpmtmr_cleanup(vm->vpmtmr); 671 vatpit_cleanup(vm->vatpit); 672 vhpet_cleanup(vm->vhpet); 673 vatpic_cleanup(vm->vatpic); 674 vioapic_cleanup(vm->vioapic); 675 676 for (i = 0; i < vm->maxcpus; i++) { 677 if (vm->vcpu[i] != NULL) 678 vcpu_cleanup(vm->vcpu[i], destroy); 679 } 680 681 vmmops_cleanup(vm->cookie); 682 683 /* 684 * System memory is removed from the guest address space only when 685 * the VM is destroyed. This is because the mapping remains the same 686 * across VM reset. 687 * 688 * Device memory can be relocated by the guest (e.g. using PCI BARs) 689 * so those mappings are removed on a VM reset. 690 */ 691 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 692 mm = &vm->mem_maps[i]; 693 if (destroy || !sysmem_mapping(vm, mm)) 694 vm_free_memmap(vm, i); 695 } 696 697 if (destroy) { 698 for (i = 0; i < VM_MAX_MEMSEGS; i++) 699 vm_free_memseg(vm, i); 700 vm_unlock_memsegs(vm); 701 702 vmmops_vmspace_free(vm->vmspace); 703 vm->vmspace = NULL; 704 705 free(vm->vcpu, M_VM); 706 sx_destroy(&vm->vcpus_init_lock); 707 sx_destroy(&vm->mem_segs_lock); 708 mtx_destroy(&vm->rendezvous_mtx); 709 } 710 } 711 712 void 713 vm_destroy(struct vm *vm) 714 { 715 vm_cleanup(vm, true); 716 free(vm, M_VM); 717 } 718 719 int 720 vm_reinit(struct vm *vm) 721 { 722 int error; 723 724 /* 725 * A virtual machine can be reset only if all vcpus are suspended. 726 */ 727 if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) { 728 vm_cleanup(vm, false); 729 vm_init(vm, false); 730 error = 0; 731 } else { 732 error = EBUSY; 733 } 734 735 return (error); 736 } 737 738 const char * 739 vm_name(struct vm *vm) 740 { 741 return (vm->name); 742 } 743 744 void 745 vm_slock_memsegs(struct vm *vm) 746 { 747 sx_slock(&vm->mem_segs_lock); 748 } 749 750 void 751 vm_xlock_memsegs(struct vm *vm) 752 { 753 sx_xlock(&vm->mem_segs_lock); 754 } 755 756 void 757 vm_unlock_memsegs(struct vm *vm) 758 { 759 sx_unlock(&vm->mem_segs_lock); 760 } 761 762 int 763 vm_map_mmio(struct vm *vm, vm_paddr_t gpa, size_t len, vm_paddr_t hpa) 764 { 765 vm_object_t obj; 766 767 if ((obj = vmm_mmio_alloc(vm->vmspace, gpa, len, hpa)) == NULL) 768 return (ENOMEM); 769 else 770 return (0); 771 } 772 773 int 774 vm_unmap_mmio(struct vm *vm, vm_paddr_t gpa, size_t len) 775 { 776 777 vmm_mmio_free(vm->vmspace, gpa, len); 778 return (0); 779 } 780 781 /* 782 * Return 'true' if 'gpa' is allocated in the guest address space. 783 * 784 * This function is called in the context of a running vcpu which acts as 785 * an implicit lock on 'vm->mem_maps[]'. 786 */ 787 bool 788 vm_mem_allocated(struct vcpu *vcpu, vm_paddr_t gpa) 789 { 790 struct vm *vm = vcpu->vm; 791 struct mem_map *mm; 792 int i; 793 794 #ifdef INVARIANTS 795 int hostcpu, state; 796 state = vcpu_get_state(vcpu, &hostcpu); 797 KASSERT(state == VCPU_RUNNING && hostcpu == curcpu, 798 ("%s: invalid vcpu state %d/%d", __func__, state, hostcpu)); 799 #endif 800 801 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 802 mm = &vm->mem_maps[i]; 803 if (mm->len != 0 && gpa >= mm->gpa && gpa < mm->gpa + mm->len) 804 return (true); /* 'gpa' is sysmem or devmem */ 805 } 806 807 if (ppt_is_mmio(vm, gpa)) 808 return (true); /* 'gpa' is pci passthru mmio */ 809 810 return (false); 811 } 812 813 int 814 vm_alloc_memseg(struct vm *vm, int ident, size_t len, bool sysmem) 815 { 816 struct mem_seg *seg; 817 vm_object_t obj; 818 819 sx_assert(&vm->mem_segs_lock, SX_XLOCKED); 820 821 if (ident < 0 || ident >= VM_MAX_MEMSEGS) 822 return (EINVAL); 823 824 if (len == 0 || (len & PAGE_MASK)) 825 return (EINVAL); 826 827 seg = &vm->mem_segs[ident]; 828 if (seg->object != NULL) { 829 if (seg->len == len && seg->sysmem == sysmem) 830 return (EEXIST); 831 else 832 return (EINVAL); 833 } 834 835 obj = vm_object_allocate(OBJT_SWAP, len >> PAGE_SHIFT); 836 if (obj == NULL) 837 return (ENOMEM); 838 839 seg->len = len; 840 seg->object = obj; 841 seg->sysmem = sysmem; 842 return (0); 843 } 844 845 int 846 vm_get_memseg(struct vm *vm, int ident, size_t *len, bool *sysmem, 847 vm_object_t *objptr) 848 { 849 struct mem_seg *seg; 850 851 sx_assert(&vm->mem_segs_lock, SX_LOCKED); 852 853 if (ident < 0 || ident >= VM_MAX_MEMSEGS) 854 return (EINVAL); 855 856 seg = &vm->mem_segs[ident]; 857 if (len) 858 *len = seg->len; 859 if (sysmem) 860 *sysmem = seg->sysmem; 861 if (objptr) 862 *objptr = seg->object; 863 return (0); 864 } 865 866 void 867 vm_free_memseg(struct vm *vm, int ident) 868 { 869 struct mem_seg *seg; 870 871 KASSERT(ident >= 0 && ident < VM_MAX_MEMSEGS, 872 ("%s: invalid memseg ident %d", __func__, ident)); 873 874 seg = &vm->mem_segs[ident]; 875 if (seg->object != NULL) { 876 vm_object_deallocate(seg->object); 877 bzero(seg, sizeof(struct mem_seg)); 878 } 879 } 880 881 int 882 vm_mmap_memseg(struct vm *vm, vm_paddr_t gpa, int segid, vm_ooffset_t first, 883 size_t len, int prot, int flags) 884 { 885 struct mem_seg *seg; 886 struct mem_map *m, *map; 887 vm_ooffset_t last; 888 int i, error; 889 890 if (prot == 0 || (prot & ~(VM_PROT_ALL)) != 0) 891 return (EINVAL); 892 893 if (flags & ~VM_MEMMAP_F_WIRED) 894 return (EINVAL); 895 896 if (segid < 0 || segid >= VM_MAX_MEMSEGS) 897 return (EINVAL); 898 899 seg = &vm->mem_segs[segid]; 900 if (seg->object == NULL) 901 return (EINVAL); 902 903 last = first + len; 904 if (first < 0 || first >= last || last > seg->len) 905 return (EINVAL); 906 907 if ((gpa | first | last) & PAGE_MASK) 908 return (EINVAL); 909 910 map = NULL; 911 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 912 m = &vm->mem_maps[i]; 913 if (m->len == 0) { 914 map = m; 915 break; 916 } 917 } 918 919 if (map == NULL) 920 return (ENOSPC); 921 922 error = vm_map_find(&vm->vmspace->vm_map, seg->object, first, &gpa, 923 len, 0, VMFS_NO_SPACE, prot, prot, 0); 924 if (error != KERN_SUCCESS) 925 return (EFAULT); 926 927 vm_object_reference(seg->object); 928 929 if (flags & VM_MEMMAP_F_WIRED) { 930 error = vm_map_wire(&vm->vmspace->vm_map, gpa, gpa + len, 931 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES); 932 if (error != KERN_SUCCESS) { 933 vm_map_remove(&vm->vmspace->vm_map, gpa, gpa + len); 934 return (error == KERN_RESOURCE_SHORTAGE ? ENOMEM : 935 EFAULT); 936 } 937 } 938 939 map->gpa = gpa; 940 map->len = len; 941 map->segoff = first; 942 map->segid = segid; 943 map->prot = prot; 944 map->flags = flags; 945 return (0); 946 } 947 948 int 949 vm_munmap_memseg(struct vm *vm, vm_paddr_t gpa, size_t len) 950 { 951 struct mem_map *m; 952 int i; 953 954 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 955 m = &vm->mem_maps[i]; 956 if (m->gpa == gpa && m->len == len && 957 (m->flags & VM_MEMMAP_F_IOMMU) == 0) { 958 vm_free_memmap(vm, i); 959 return (0); 960 } 961 } 962 963 return (EINVAL); 964 } 965 966 int 967 vm_mmap_getnext(struct vm *vm, vm_paddr_t *gpa, int *segid, 968 vm_ooffset_t *segoff, size_t *len, int *prot, int *flags) 969 { 970 struct mem_map *mm, *mmnext; 971 int i; 972 973 mmnext = NULL; 974 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 975 mm = &vm->mem_maps[i]; 976 if (mm->len == 0 || mm->gpa < *gpa) 977 continue; 978 if (mmnext == NULL || mm->gpa < mmnext->gpa) 979 mmnext = mm; 980 } 981 982 if (mmnext != NULL) { 983 *gpa = mmnext->gpa; 984 if (segid) 985 *segid = mmnext->segid; 986 if (segoff) 987 *segoff = mmnext->segoff; 988 if (len) 989 *len = mmnext->len; 990 if (prot) 991 *prot = mmnext->prot; 992 if (flags) 993 *flags = mmnext->flags; 994 return (0); 995 } else { 996 return (ENOENT); 997 } 998 } 999 1000 static void 1001 vm_free_memmap(struct vm *vm, int ident) 1002 { 1003 struct mem_map *mm; 1004 int error __diagused; 1005 1006 mm = &vm->mem_maps[ident]; 1007 if (mm->len) { 1008 error = vm_map_remove(&vm->vmspace->vm_map, mm->gpa, 1009 mm->gpa + mm->len); 1010 KASSERT(error == KERN_SUCCESS, ("%s: vm_map_remove error %d", 1011 __func__, error)); 1012 bzero(mm, sizeof(struct mem_map)); 1013 } 1014 } 1015 1016 static __inline bool 1017 sysmem_mapping(struct vm *vm, struct mem_map *mm) 1018 { 1019 1020 if (mm->len != 0 && vm->mem_segs[mm->segid].sysmem) 1021 return (true); 1022 else 1023 return (false); 1024 } 1025 1026 vm_paddr_t 1027 vmm_sysmem_maxaddr(struct vm *vm) 1028 { 1029 struct mem_map *mm; 1030 vm_paddr_t maxaddr; 1031 int i; 1032 1033 maxaddr = 0; 1034 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 1035 mm = &vm->mem_maps[i]; 1036 if (sysmem_mapping(vm, mm)) { 1037 if (maxaddr < mm->gpa + mm->len) 1038 maxaddr = mm->gpa + mm->len; 1039 } 1040 } 1041 return (maxaddr); 1042 } 1043 1044 static void 1045 vm_iommu_modify(struct vm *vm, bool map) 1046 { 1047 int i, sz; 1048 vm_paddr_t gpa, hpa; 1049 struct mem_map *mm; 1050 void *vp, *cookie, *host_domain; 1051 1052 sz = PAGE_SIZE; 1053 host_domain = iommu_host_domain(); 1054 1055 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 1056 mm = &vm->mem_maps[i]; 1057 if (!sysmem_mapping(vm, mm)) 1058 continue; 1059 1060 if (map) { 1061 KASSERT((mm->flags & VM_MEMMAP_F_IOMMU) == 0, 1062 ("iommu map found invalid memmap %#lx/%#lx/%#x", 1063 mm->gpa, mm->len, mm->flags)); 1064 if ((mm->flags & VM_MEMMAP_F_WIRED) == 0) 1065 continue; 1066 mm->flags |= VM_MEMMAP_F_IOMMU; 1067 } else { 1068 if ((mm->flags & VM_MEMMAP_F_IOMMU) == 0) 1069 continue; 1070 mm->flags &= ~VM_MEMMAP_F_IOMMU; 1071 KASSERT((mm->flags & VM_MEMMAP_F_WIRED) != 0, 1072 ("iommu unmap found invalid memmap %#lx/%#lx/%#x", 1073 mm->gpa, mm->len, mm->flags)); 1074 } 1075 1076 gpa = mm->gpa; 1077 while (gpa < mm->gpa + mm->len) { 1078 vp = vm_gpa_hold_global(vm, gpa, PAGE_SIZE, 1079 VM_PROT_WRITE, &cookie); 1080 KASSERT(vp != NULL, ("vm(%s) could not map gpa %#lx", 1081 vm_name(vm), gpa)); 1082 1083 vm_gpa_release(cookie); 1084 1085 hpa = DMAP_TO_PHYS((uintptr_t)vp); 1086 if (map) { 1087 iommu_create_mapping(vm->iommu, gpa, hpa, sz); 1088 } else { 1089 iommu_remove_mapping(vm->iommu, gpa, sz); 1090 } 1091 1092 gpa += PAGE_SIZE; 1093 } 1094 } 1095 1096 /* 1097 * Invalidate the cached translations associated with the domain 1098 * from which pages were removed. 1099 */ 1100 if (map) 1101 iommu_invalidate_tlb(host_domain); 1102 else 1103 iommu_invalidate_tlb(vm->iommu); 1104 } 1105 1106 #define vm_iommu_unmap(vm) vm_iommu_modify((vm), false) 1107 #define vm_iommu_map(vm) vm_iommu_modify((vm), true) 1108 1109 int 1110 vm_unassign_pptdev(struct vm *vm, int bus, int slot, int func) 1111 { 1112 int error; 1113 1114 error = ppt_unassign_device(vm, bus, slot, func); 1115 if (error) 1116 return (error); 1117 1118 if (ppt_assigned_devices(vm) == 0) 1119 vm_iommu_unmap(vm); 1120 1121 return (0); 1122 } 1123 1124 int 1125 vm_assign_pptdev(struct vm *vm, int bus, int slot, int func) 1126 { 1127 int error; 1128 vm_paddr_t maxaddr; 1129 1130 /* Set up the IOMMU to do the 'gpa' to 'hpa' translation */ 1131 if (ppt_assigned_devices(vm) == 0) { 1132 KASSERT(vm->iommu == NULL, 1133 ("vm_assign_pptdev: iommu must be NULL")); 1134 maxaddr = vmm_sysmem_maxaddr(vm); 1135 vm->iommu = iommu_create_domain(maxaddr); 1136 if (vm->iommu == NULL) 1137 return (ENXIO); 1138 vm_iommu_map(vm); 1139 } 1140 1141 error = ppt_assign_device(vm, bus, slot, func); 1142 return (error); 1143 } 1144 1145 static void * 1146 _vm_gpa_hold(struct vm *vm, vm_paddr_t gpa, size_t len, int reqprot, 1147 void **cookie) 1148 { 1149 int i, count, pageoff; 1150 struct mem_map *mm; 1151 vm_page_t m; 1152 1153 pageoff = gpa & PAGE_MASK; 1154 if (len > PAGE_SIZE - pageoff) 1155 panic("vm_gpa_hold: invalid gpa/len: 0x%016lx/%lu", gpa, len); 1156 1157 count = 0; 1158 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 1159 mm = &vm->mem_maps[i]; 1160 if (gpa >= mm->gpa && gpa < mm->gpa + mm->len) { 1161 count = vm_fault_quick_hold_pages(&vm->vmspace->vm_map, 1162 trunc_page(gpa), PAGE_SIZE, reqprot, &m, 1); 1163 break; 1164 } 1165 } 1166 1167 if (count == 1) { 1168 *cookie = m; 1169 return ((void *)(PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m)) + pageoff)); 1170 } else { 1171 *cookie = NULL; 1172 return (NULL); 1173 } 1174 } 1175 1176 void * 1177 vm_gpa_hold(struct vcpu *vcpu, vm_paddr_t gpa, size_t len, int reqprot, 1178 void **cookie) 1179 { 1180 #ifdef INVARIANTS 1181 /* 1182 * The current vcpu should be frozen to ensure 'vm_memmap[]' 1183 * stability. 1184 */ 1185 int state = vcpu_get_state(vcpu, NULL); 1186 KASSERT(state == VCPU_FROZEN, ("%s: invalid vcpu state %d", 1187 __func__, state)); 1188 #endif 1189 return (_vm_gpa_hold(vcpu->vm, gpa, len, reqprot, cookie)); 1190 } 1191 1192 void * 1193 vm_gpa_hold_global(struct vm *vm, vm_paddr_t gpa, size_t len, int reqprot, 1194 void **cookie) 1195 { 1196 sx_assert(&vm->mem_segs_lock, SX_LOCKED); 1197 return (_vm_gpa_hold(vm, gpa, len, reqprot, cookie)); 1198 } 1199 1200 void 1201 vm_gpa_release(void *cookie) 1202 { 1203 vm_page_t m = cookie; 1204 1205 vm_page_unwire(m, PQ_ACTIVE); 1206 } 1207 1208 int 1209 vm_get_register(struct vcpu *vcpu, int reg, uint64_t *retval) 1210 { 1211 1212 if (reg >= VM_REG_LAST) 1213 return (EINVAL); 1214 1215 return (vmmops_getreg(vcpu->cookie, reg, retval)); 1216 } 1217 1218 int 1219 vm_set_register(struct vcpu *vcpu, int reg, uint64_t val) 1220 { 1221 int error; 1222 1223 if (reg >= VM_REG_LAST) 1224 return (EINVAL); 1225 1226 error = vmmops_setreg(vcpu->cookie, reg, val); 1227 if (error || reg != VM_REG_GUEST_RIP) 1228 return (error); 1229 1230 /* Set 'nextrip' to match the value of %rip */ 1231 VMM_CTR1(vcpu, "Setting nextrip to %#lx", val); 1232 vcpu->nextrip = val; 1233 return (0); 1234 } 1235 1236 static bool 1237 is_descriptor_table(int reg) 1238 { 1239 1240 switch (reg) { 1241 case VM_REG_GUEST_IDTR: 1242 case VM_REG_GUEST_GDTR: 1243 return (true); 1244 default: 1245 return (false); 1246 } 1247 } 1248 1249 static bool 1250 is_segment_register(int reg) 1251 { 1252 1253 switch (reg) { 1254 case VM_REG_GUEST_ES: 1255 case VM_REG_GUEST_CS: 1256 case VM_REG_GUEST_SS: 1257 case VM_REG_GUEST_DS: 1258 case VM_REG_GUEST_FS: 1259 case VM_REG_GUEST_GS: 1260 case VM_REG_GUEST_TR: 1261 case VM_REG_GUEST_LDTR: 1262 return (true); 1263 default: 1264 return (false); 1265 } 1266 } 1267 1268 int 1269 vm_get_seg_desc(struct vcpu *vcpu, int reg, struct seg_desc *desc) 1270 { 1271 1272 if (!is_segment_register(reg) && !is_descriptor_table(reg)) 1273 return (EINVAL); 1274 1275 return (vmmops_getdesc(vcpu->cookie, reg, desc)); 1276 } 1277 1278 int 1279 vm_set_seg_desc(struct vcpu *vcpu, int reg, struct seg_desc *desc) 1280 { 1281 1282 if (!is_segment_register(reg) && !is_descriptor_table(reg)) 1283 return (EINVAL); 1284 1285 return (vmmops_setdesc(vcpu->cookie, reg, desc)); 1286 } 1287 1288 static void 1289 restore_guest_fpustate(struct vcpu *vcpu) 1290 { 1291 1292 /* flush host state to the pcb */ 1293 fpuexit(curthread); 1294 1295 /* restore guest FPU state */ 1296 fpu_stop_emulating(); 1297 fpurestore(vcpu->guestfpu); 1298 1299 /* restore guest XCR0 if XSAVE is enabled in the host */ 1300 if (rcr4() & CR4_XSAVE) 1301 load_xcr(0, vcpu->guest_xcr0); 1302 1303 /* 1304 * The FPU is now "dirty" with the guest's state so turn on emulation 1305 * to trap any access to the FPU by the host. 1306 */ 1307 fpu_start_emulating(); 1308 } 1309 1310 static void 1311 save_guest_fpustate(struct vcpu *vcpu) 1312 { 1313 1314 if ((rcr0() & CR0_TS) == 0) 1315 panic("fpu emulation not enabled in host!"); 1316 1317 /* save guest XCR0 and restore host XCR0 */ 1318 if (rcr4() & CR4_XSAVE) { 1319 vcpu->guest_xcr0 = rxcr(0); 1320 load_xcr(0, vmm_get_host_xcr0()); 1321 } 1322 1323 /* save guest FPU state */ 1324 fpu_stop_emulating(); 1325 fpusave(vcpu->guestfpu); 1326 fpu_start_emulating(); 1327 } 1328 1329 static VMM_STAT(VCPU_IDLE_TICKS, "number of ticks vcpu was idle"); 1330 1331 static int 1332 vcpu_set_state_locked(struct vcpu *vcpu, enum vcpu_state newstate, 1333 bool from_idle) 1334 { 1335 int error; 1336 1337 vcpu_assert_locked(vcpu); 1338 1339 /* 1340 * State transitions from the vmmdev_ioctl() must always begin from 1341 * the VCPU_IDLE state. This guarantees that there is only a single 1342 * ioctl() operating on a vcpu at any point. 1343 */ 1344 if (from_idle) { 1345 while (vcpu->state != VCPU_IDLE) { 1346 vcpu->reqidle = 1; 1347 vcpu_notify_event_locked(vcpu, false); 1348 VMM_CTR1(vcpu, "vcpu state change from %s to " 1349 "idle requested", vcpu_state2str(vcpu->state)); 1350 msleep_spin(&vcpu->state, &vcpu->mtx, "vmstat", hz); 1351 } 1352 } else { 1353 KASSERT(vcpu->state != VCPU_IDLE, ("invalid transition from " 1354 "vcpu idle state")); 1355 } 1356 1357 if (vcpu->state == VCPU_RUNNING) { 1358 KASSERT(vcpu->hostcpu == curcpu, ("curcpu %d and hostcpu %d " 1359 "mismatch for running vcpu", curcpu, vcpu->hostcpu)); 1360 } else { 1361 KASSERT(vcpu->hostcpu == NOCPU, ("Invalid hostcpu %d for a " 1362 "vcpu that is not running", vcpu->hostcpu)); 1363 } 1364 1365 /* 1366 * The following state transitions are allowed: 1367 * IDLE -> FROZEN -> IDLE 1368 * FROZEN -> RUNNING -> FROZEN 1369 * FROZEN -> SLEEPING -> FROZEN 1370 */ 1371 switch (vcpu->state) { 1372 case VCPU_IDLE: 1373 case VCPU_RUNNING: 1374 case VCPU_SLEEPING: 1375 error = (newstate != VCPU_FROZEN); 1376 break; 1377 case VCPU_FROZEN: 1378 error = (newstate == VCPU_FROZEN); 1379 break; 1380 default: 1381 error = 1; 1382 break; 1383 } 1384 1385 if (error) 1386 return (EBUSY); 1387 1388 VMM_CTR2(vcpu, "vcpu state changed from %s to %s", 1389 vcpu_state2str(vcpu->state), vcpu_state2str(newstate)); 1390 1391 vcpu->state = newstate; 1392 if (newstate == VCPU_RUNNING) 1393 vcpu->hostcpu = curcpu; 1394 else 1395 vcpu->hostcpu = NOCPU; 1396 1397 if (newstate == VCPU_IDLE) 1398 wakeup(&vcpu->state); 1399 1400 return (0); 1401 } 1402 1403 static void 1404 vcpu_require_state(struct vcpu *vcpu, enum vcpu_state newstate) 1405 { 1406 int error; 1407 1408 if ((error = vcpu_set_state(vcpu, newstate, false)) != 0) 1409 panic("Error %d setting state to %d\n", error, newstate); 1410 } 1411 1412 static void 1413 vcpu_require_state_locked(struct vcpu *vcpu, enum vcpu_state newstate) 1414 { 1415 int error; 1416 1417 if ((error = vcpu_set_state_locked(vcpu, newstate, false)) != 0) 1418 panic("Error %d setting state to %d", error, newstate); 1419 } 1420 1421 static int 1422 vm_handle_rendezvous(struct vcpu *vcpu) 1423 { 1424 struct vm *vm = vcpu->vm; 1425 struct thread *td; 1426 int error, vcpuid; 1427 1428 error = 0; 1429 vcpuid = vcpu->vcpuid; 1430 td = curthread; 1431 mtx_lock(&vm->rendezvous_mtx); 1432 while (vm->rendezvous_func != NULL) { 1433 /* 'rendezvous_req_cpus' must be a subset of 'active_cpus' */ 1434 CPU_AND(&vm->rendezvous_req_cpus, &vm->rendezvous_req_cpus, &vm->active_cpus); 1435 1436 if (CPU_ISSET(vcpuid, &vm->rendezvous_req_cpus) && 1437 !CPU_ISSET(vcpuid, &vm->rendezvous_done_cpus)) { 1438 VMM_CTR0(vcpu, "Calling rendezvous func"); 1439 (*vm->rendezvous_func)(vcpu, vm->rendezvous_arg); 1440 CPU_SET(vcpuid, &vm->rendezvous_done_cpus); 1441 } 1442 if (CPU_CMP(&vm->rendezvous_req_cpus, 1443 &vm->rendezvous_done_cpus) == 0) { 1444 VMM_CTR0(vcpu, "Rendezvous completed"); 1445 CPU_ZERO(&vm->rendezvous_req_cpus); 1446 vm->rendezvous_func = NULL; 1447 wakeup(&vm->rendezvous_func); 1448 break; 1449 } 1450 VMM_CTR0(vcpu, "Wait for rendezvous completion"); 1451 mtx_sleep(&vm->rendezvous_func, &vm->rendezvous_mtx, 0, 1452 "vmrndv", hz); 1453 if (td_ast_pending(td, TDA_SUSPEND)) { 1454 mtx_unlock(&vm->rendezvous_mtx); 1455 error = thread_check_susp(td, true); 1456 if (error != 0) 1457 return (error); 1458 mtx_lock(&vm->rendezvous_mtx); 1459 } 1460 } 1461 mtx_unlock(&vm->rendezvous_mtx); 1462 return (0); 1463 } 1464 1465 /* 1466 * Emulate a guest 'hlt' by sleeping until the vcpu is ready to run. 1467 */ 1468 static int 1469 vm_handle_hlt(struct vcpu *vcpu, bool intr_disabled, bool *retu) 1470 { 1471 struct vm *vm = vcpu->vm; 1472 const char *wmesg; 1473 struct thread *td; 1474 int error, t, vcpuid, vcpu_halted, vm_halted; 1475 1476 vcpuid = vcpu->vcpuid; 1477 vcpu_halted = 0; 1478 vm_halted = 0; 1479 error = 0; 1480 td = curthread; 1481 1482 KASSERT(!CPU_ISSET(vcpuid, &vm->halted_cpus), ("vcpu already halted")); 1483 1484 vcpu_lock(vcpu); 1485 while (1) { 1486 /* 1487 * Do a final check for pending NMI or interrupts before 1488 * really putting this thread to sleep. Also check for 1489 * software events that would cause this vcpu to wakeup. 1490 * 1491 * These interrupts/events could have happened after the 1492 * vcpu returned from vmmops_run() and before it acquired the 1493 * vcpu lock above. 1494 */ 1495 if (vm->rendezvous_func != NULL || vm->suspend || vcpu->reqidle) 1496 break; 1497 if (vm_nmi_pending(vcpu)) 1498 break; 1499 if (!intr_disabled) { 1500 if (vm_extint_pending(vcpu) || 1501 vlapic_pending_intr(vcpu->vlapic, NULL)) { 1502 break; 1503 } 1504 } 1505 1506 /* Don't go to sleep if the vcpu thread needs to yield */ 1507 if (vcpu_should_yield(vcpu)) 1508 break; 1509 1510 if (vcpu_debugged(vcpu)) 1511 break; 1512 1513 /* 1514 * Some Linux guests implement "halt" by having all vcpus 1515 * execute HLT with interrupts disabled. 'halted_cpus' keeps 1516 * track of the vcpus that have entered this state. When all 1517 * vcpus enter the halted state the virtual machine is halted. 1518 */ 1519 if (intr_disabled) { 1520 wmesg = "vmhalt"; 1521 VMM_CTR0(vcpu, "Halted"); 1522 if (!vcpu_halted && halt_detection_enabled) { 1523 vcpu_halted = 1; 1524 CPU_SET_ATOMIC(vcpuid, &vm->halted_cpus); 1525 } 1526 if (CPU_CMP(&vm->halted_cpus, &vm->active_cpus) == 0) { 1527 vm_halted = 1; 1528 break; 1529 } 1530 } else { 1531 wmesg = "vmidle"; 1532 } 1533 1534 t = ticks; 1535 vcpu_require_state_locked(vcpu, VCPU_SLEEPING); 1536 /* 1537 * XXX msleep_spin() cannot be interrupted by signals so 1538 * wake up periodically to check pending signals. 1539 */ 1540 msleep_spin(vcpu, &vcpu->mtx, wmesg, hz); 1541 vcpu_require_state_locked(vcpu, VCPU_FROZEN); 1542 vmm_stat_incr(vcpu, VCPU_IDLE_TICKS, ticks - t); 1543 if (td_ast_pending(td, TDA_SUSPEND)) { 1544 vcpu_unlock(vcpu); 1545 error = thread_check_susp(td, false); 1546 if (error != 0) { 1547 if (vcpu_halted) { 1548 CPU_CLR_ATOMIC(vcpuid, 1549 &vm->halted_cpus); 1550 } 1551 return (error); 1552 } 1553 vcpu_lock(vcpu); 1554 } 1555 } 1556 1557 if (vcpu_halted) 1558 CPU_CLR_ATOMIC(vcpuid, &vm->halted_cpus); 1559 1560 vcpu_unlock(vcpu); 1561 1562 if (vm_halted) 1563 vm_suspend(vm, VM_SUSPEND_HALT); 1564 1565 return (0); 1566 } 1567 1568 static int 1569 vm_handle_paging(struct vcpu *vcpu, bool *retu) 1570 { 1571 struct vm *vm = vcpu->vm; 1572 int rv, ftype; 1573 struct vm_map *map; 1574 struct vm_exit *vme; 1575 1576 vme = &vcpu->exitinfo; 1577 1578 KASSERT(vme->inst_length == 0, ("%s: invalid inst_length %d", 1579 __func__, vme->inst_length)); 1580 1581 ftype = vme->u.paging.fault_type; 1582 KASSERT(ftype == VM_PROT_READ || 1583 ftype == VM_PROT_WRITE || ftype == VM_PROT_EXECUTE, 1584 ("vm_handle_paging: invalid fault_type %d", ftype)); 1585 1586 if (ftype == VM_PROT_READ || ftype == VM_PROT_WRITE) { 1587 rv = pmap_emulate_accessed_dirty(vmspace_pmap(vm->vmspace), 1588 vme->u.paging.gpa, ftype); 1589 if (rv == 0) { 1590 VMM_CTR2(vcpu, "%s bit emulation for gpa %#lx", 1591 ftype == VM_PROT_READ ? "accessed" : "dirty", 1592 vme->u.paging.gpa); 1593 goto done; 1594 } 1595 } 1596 1597 map = &vm->vmspace->vm_map; 1598 rv = vm_fault(map, vme->u.paging.gpa, ftype, VM_FAULT_NORMAL, NULL); 1599 1600 VMM_CTR3(vcpu, "vm_handle_paging rv = %d, gpa = %#lx, " 1601 "ftype = %d", rv, vme->u.paging.gpa, ftype); 1602 1603 if (rv != KERN_SUCCESS) 1604 return (EFAULT); 1605 done: 1606 return (0); 1607 } 1608 1609 static int 1610 vm_handle_inst_emul(struct vcpu *vcpu, bool *retu) 1611 { 1612 struct vie *vie; 1613 struct vm_exit *vme; 1614 uint64_t gla, gpa, cs_base; 1615 struct vm_guest_paging *paging; 1616 mem_region_read_t mread; 1617 mem_region_write_t mwrite; 1618 enum vm_cpu_mode cpu_mode; 1619 int cs_d, error, fault; 1620 1621 vme = &vcpu->exitinfo; 1622 1623 KASSERT(vme->inst_length == 0, ("%s: invalid inst_length %d", 1624 __func__, vme->inst_length)); 1625 1626 gla = vme->u.inst_emul.gla; 1627 gpa = vme->u.inst_emul.gpa; 1628 cs_base = vme->u.inst_emul.cs_base; 1629 cs_d = vme->u.inst_emul.cs_d; 1630 vie = &vme->u.inst_emul.vie; 1631 paging = &vme->u.inst_emul.paging; 1632 cpu_mode = paging->cpu_mode; 1633 1634 VMM_CTR1(vcpu, "inst_emul fault accessing gpa %#lx", gpa); 1635 1636 /* Fetch, decode and emulate the faulting instruction */ 1637 if (vie->num_valid == 0) { 1638 error = vmm_fetch_instruction(vcpu, paging, vme->rip + cs_base, 1639 VIE_INST_SIZE, vie, &fault); 1640 } else { 1641 /* 1642 * The instruction bytes have already been copied into 'vie' 1643 */ 1644 error = fault = 0; 1645 } 1646 if (error || fault) 1647 return (error); 1648 1649 if (vmm_decode_instruction(vcpu, gla, cpu_mode, cs_d, vie) != 0) { 1650 VMM_CTR1(vcpu, "Error decoding instruction at %#lx", 1651 vme->rip + cs_base); 1652 *retu = true; /* dump instruction bytes in userspace */ 1653 return (0); 1654 } 1655 1656 /* 1657 * Update 'nextrip' based on the length of the emulated instruction. 1658 */ 1659 vme->inst_length = vie->num_processed; 1660 vcpu->nextrip += vie->num_processed; 1661 VMM_CTR1(vcpu, "nextrip updated to %#lx after instruction decoding", 1662 vcpu->nextrip); 1663 1664 /* return to userland unless this is an in-kernel emulated device */ 1665 if (gpa >= DEFAULT_APIC_BASE && gpa < DEFAULT_APIC_BASE + PAGE_SIZE) { 1666 mread = lapic_mmio_read; 1667 mwrite = lapic_mmio_write; 1668 } else if (gpa >= VIOAPIC_BASE && gpa < VIOAPIC_BASE + VIOAPIC_SIZE) { 1669 mread = vioapic_mmio_read; 1670 mwrite = vioapic_mmio_write; 1671 } else if (gpa >= VHPET_BASE && gpa < VHPET_BASE + VHPET_SIZE) { 1672 mread = vhpet_mmio_read; 1673 mwrite = vhpet_mmio_write; 1674 } else { 1675 *retu = true; 1676 return (0); 1677 } 1678 1679 error = vmm_emulate_instruction(vcpu, gpa, vie, paging, mread, mwrite, 1680 retu); 1681 1682 return (error); 1683 } 1684 1685 static int 1686 vm_handle_suspend(struct vcpu *vcpu, bool *retu) 1687 { 1688 struct vm *vm = vcpu->vm; 1689 int error, i; 1690 struct thread *td; 1691 1692 error = 0; 1693 td = curthread; 1694 1695 CPU_SET_ATOMIC(vcpu->vcpuid, &vm->suspended_cpus); 1696 1697 /* 1698 * Wait until all 'active_cpus' have suspended themselves. 1699 * 1700 * Since a VM may be suspended at any time including when one or 1701 * more vcpus are doing a rendezvous we need to call the rendezvous 1702 * handler while we are waiting to prevent a deadlock. 1703 */ 1704 vcpu_lock(vcpu); 1705 while (error == 0) { 1706 if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) { 1707 VMM_CTR0(vcpu, "All vcpus suspended"); 1708 break; 1709 } 1710 1711 if (vm->rendezvous_func == NULL) { 1712 VMM_CTR0(vcpu, "Sleeping during suspend"); 1713 vcpu_require_state_locked(vcpu, VCPU_SLEEPING); 1714 msleep_spin(vcpu, &vcpu->mtx, "vmsusp", hz); 1715 vcpu_require_state_locked(vcpu, VCPU_FROZEN); 1716 if (td_ast_pending(td, TDA_SUSPEND)) { 1717 vcpu_unlock(vcpu); 1718 error = thread_check_susp(td, false); 1719 vcpu_lock(vcpu); 1720 } 1721 } else { 1722 VMM_CTR0(vcpu, "Rendezvous during suspend"); 1723 vcpu_unlock(vcpu); 1724 error = vm_handle_rendezvous(vcpu); 1725 vcpu_lock(vcpu); 1726 } 1727 } 1728 vcpu_unlock(vcpu); 1729 1730 /* 1731 * Wakeup the other sleeping vcpus and return to userspace. 1732 */ 1733 for (i = 0; i < vm->maxcpus; i++) { 1734 if (CPU_ISSET(i, &vm->suspended_cpus)) { 1735 vcpu_notify_event(vm_vcpu(vm, i), false); 1736 } 1737 } 1738 1739 *retu = true; 1740 return (error); 1741 } 1742 1743 static int 1744 vm_handle_reqidle(struct vcpu *vcpu, bool *retu) 1745 { 1746 vcpu_lock(vcpu); 1747 KASSERT(vcpu->reqidle, ("invalid vcpu reqidle %d", vcpu->reqidle)); 1748 vcpu->reqidle = 0; 1749 vcpu_unlock(vcpu); 1750 *retu = true; 1751 return (0); 1752 } 1753 1754 int 1755 vm_suspend(struct vm *vm, enum vm_suspend_how how) 1756 { 1757 int i; 1758 1759 if (how <= VM_SUSPEND_NONE || how >= VM_SUSPEND_LAST) 1760 return (EINVAL); 1761 1762 if (atomic_cmpset_int(&vm->suspend, 0, how) == 0) { 1763 VM_CTR2(vm, "virtual machine already suspended %d/%d", 1764 vm->suspend, how); 1765 return (EALREADY); 1766 } 1767 1768 VM_CTR1(vm, "virtual machine successfully suspended %d", how); 1769 1770 /* 1771 * Notify all active vcpus that they are now suspended. 1772 */ 1773 for (i = 0; i < vm->maxcpus; i++) { 1774 if (CPU_ISSET(i, &vm->active_cpus)) 1775 vcpu_notify_event(vm_vcpu(vm, i), false); 1776 } 1777 1778 return (0); 1779 } 1780 1781 void 1782 vm_exit_suspended(struct vcpu *vcpu, uint64_t rip) 1783 { 1784 struct vm *vm = vcpu->vm; 1785 struct vm_exit *vmexit; 1786 1787 KASSERT(vm->suspend > VM_SUSPEND_NONE && vm->suspend < VM_SUSPEND_LAST, 1788 ("vm_exit_suspended: invalid suspend type %d", vm->suspend)); 1789 1790 vmexit = vm_exitinfo(vcpu); 1791 vmexit->rip = rip; 1792 vmexit->inst_length = 0; 1793 vmexit->exitcode = VM_EXITCODE_SUSPENDED; 1794 vmexit->u.suspended.how = vm->suspend; 1795 } 1796 1797 void 1798 vm_exit_debug(struct vcpu *vcpu, uint64_t rip) 1799 { 1800 struct vm_exit *vmexit; 1801 1802 vmexit = vm_exitinfo(vcpu); 1803 vmexit->rip = rip; 1804 vmexit->inst_length = 0; 1805 vmexit->exitcode = VM_EXITCODE_DEBUG; 1806 } 1807 1808 void 1809 vm_exit_rendezvous(struct vcpu *vcpu, uint64_t rip) 1810 { 1811 struct vm_exit *vmexit; 1812 1813 vmexit = vm_exitinfo(vcpu); 1814 vmexit->rip = rip; 1815 vmexit->inst_length = 0; 1816 vmexit->exitcode = VM_EXITCODE_RENDEZVOUS; 1817 vmm_stat_incr(vcpu, VMEXIT_RENDEZVOUS, 1); 1818 } 1819 1820 void 1821 vm_exit_reqidle(struct vcpu *vcpu, uint64_t rip) 1822 { 1823 struct vm_exit *vmexit; 1824 1825 vmexit = vm_exitinfo(vcpu); 1826 vmexit->rip = rip; 1827 vmexit->inst_length = 0; 1828 vmexit->exitcode = VM_EXITCODE_REQIDLE; 1829 vmm_stat_incr(vcpu, VMEXIT_REQIDLE, 1); 1830 } 1831 1832 void 1833 vm_exit_astpending(struct vcpu *vcpu, uint64_t rip) 1834 { 1835 struct vm_exit *vmexit; 1836 1837 vmexit = vm_exitinfo(vcpu); 1838 vmexit->rip = rip; 1839 vmexit->inst_length = 0; 1840 vmexit->exitcode = VM_EXITCODE_BOGUS; 1841 vmm_stat_incr(vcpu, VMEXIT_ASTPENDING, 1); 1842 } 1843 1844 int 1845 vm_run(struct vcpu *vcpu) 1846 { 1847 struct vm *vm = vcpu->vm; 1848 struct vm_eventinfo evinfo; 1849 int error, vcpuid; 1850 struct pcb *pcb; 1851 uint64_t tscval; 1852 struct vm_exit *vme; 1853 bool retu, intr_disabled; 1854 pmap_t pmap; 1855 1856 vcpuid = vcpu->vcpuid; 1857 1858 if (!CPU_ISSET(vcpuid, &vm->active_cpus)) 1859 return (EINVAL); 1860 1861 if (CPU_ISSET(vcpuid, &vm->suspended_cpus)) 1862 return (EINVAL); 1863 1864 pmap = vmspace_pmap(vm->vmspace); 1865 vme = &vcpu->exitinfo; 1866 evinfo.rptr = &vm->rendezvous_req_cpus; 1867 evinfo.sptr = &vm->suspend; 1868 evinfo.iptr = &vcpu->reqidle; 1869 restart: 1870 critical_enter(); 1871 1872 KASSERT(!CPU_ISSET(curcpu, &pmap->pm_active), 1873 ("vm_run: absurd pm_active")); 1874 1875 tscval = rdtsc(); 1876 1877 pcb = PCPU_GET(curpcb); 1878 set_pcb_flags(pcb, PCB_FULL_IRET); 1879 1880 restore_guest_fpustate(vcpu); 1881 1882 vcpu_require_state(vcpu, VCPU_RUNNING); 1883 error = vmmops_run(vcpu->cookie, vcpu->nextrip, pmap, &evinfo); 1884 vcpu_require_state(vcpu, VCPU_FROZEN); 1885 1886 save_guest_fpustate(vcpu); 1887 1888 vmm_stat_incr(vcpu, VCPU_TOTAL_RUNTIME, rdtsc() - tscval); 1889 1890 critical_exit(); 1891 1892 if (error == 0) { 1893 retu = false; 1894 vcpu->nextrip = vme->rip + vme->inst_length; 1895 switch (vme->exitcode) { 1896 case VM_EXITCODE_REQIDLE: 1897 error = vm_handle_reqidle(vcpu, &retu); 1898 break; 1899 case VM_EXITCODE_SUSPENDED: 1900 error = vm_handle_suspend(vcpu, &retu); 1901 break; 1902 case VM_EXITCODE_IOAPIC_EOI: 1903 vioapic_process_eoi(vm, vme->u.ioapic_eoi.vector); 1904 break; 1905 case VM_EXITCODE_RENDEZVOUS: 1906 error = vm_handle_rendezvous(vcpu); 1907 break; 1908 case VM_EXITCODE_HLT: 1909 intr_disabled = ((vme->u.hlt.rflags & PSL_I) == 0); 1910 error = vm_handle_hlt(vcpu, intr_disabled, &retu); 1911 break; 1912 case VM_EXITCODE_PAGING: 1913 error = vm_handle_paging(vcpu, &retu); 1914 break; 1915 case VM_EXITCODE_INST_EMUL: 1916 error = vm_handle_inst_emul(vcpu, &retu); 1917 break; 1918 case VM_EXITCODE_INOUT: 1919 case VM_EXITCODE_INOUT_STR: 1920 error = vm_handle_inout(vcpu, vme, &retu); 1921 break; 1922 case VM_EXITCODE_MONITOR: 1923 case VM_EXITCODE_MWAIT: 1924 case VM_EXITCODE_VMINSN: 1925 vm_inject_ud(vcpu); 1926 break; 1927 default: 1928 retu = true; /* handled in userland */ 1929 break; 1930 } 1931 } 1932 1933 /* 1934 * VM_EXITCODE_INST_EMUL could access the apic which could transform the 1935 * exit code into VM_EXITCODE_IPI. 1936 */ 1937 if (error == 0 && vme->exitcode == VM_EXITCODE_IPI) 1938 error = vm_handle_ipi(vcpu, vme, &retu); 1939 1940 if (error == 0 && retu == false) 1941 goto restart; 1942 1943 vmm_stat_incr(vcpu, VMEXIT_USERSPACE, 1); 1944 VMM_CTR2(vcpu, "retu %d/%d", error, vme->exitcode); 1945 1946 return (error); 1947 } 1948 1949 int 1950 vm_restart_instruction(struct vcpu *vcpu) 1951 { 1952 enum vcpu_state state; 1953 uint64_t rip; 1954 int error __diagused; 1955 1956 state = vcpu_get_state(vcpu, NULL); 1957 if (state == VCPU_RUNNING) { 1958 /* 1959 * When a vcpu is "running" the next instruction is determined 1960 * by adding 'rip' and 'inst_length' in the vcpu's 'exitinfo'. 1961 * Thus setting 'inst_length' to zero will cause the current 1962 * instruction to be restarted. 1963 */ 1964 vcpu->exitinfo.inst_length = 0; 1965 VMM_CTR1(vcpu, "restarting instruction at %#lx by " 1966 "setting inst_length to zero", vcpu->exitinfo.rip); 1967 } else if (state == VCPU_FROZEN) { 1968 /* 1969 * When a vcpu is "frozen" it is outside the critical section 1970 * around vmmops_run() and 'nextrip' points to the next 1971 * instruction. Thus instruction restart is achieved by setting 1972 * 'nextrip' to the vcpu's %rip. 1973 */ 1974 error = vm_get_register(vcpu, VM_REG_GUEST_RIP, &rip); 1975 KASSERT(!error, ("%s: error %d getting rip", __func__, error)); 1976 VMM_CTR2(vcpu, "restarting instruction by updating " 1977 "nextrip from %#lx to %#lx", vcpu->nextrip, rip); 1978 vcpu->nextrip = rip; 1979 } else { 1980 panic("%s: invalid state %d", __func__, state); 1981 } 1982 return (0); 1983 } 1984 1985 int 1986 vm_exit_intinfo(struct vcpu *vcpu, uint64_t info) 1987 { 1988 int type, vector; 1989 1990 if (info & VM_INTINFO_VALID) { 1991 type = info & VM_INTINFO_TYPE; 1992 vector = info & 0xff; 1993 if (type == VM_INTINFO_NMI && vector != IDT_NMI) 1994 return (EINVAL); 1995 if (type == VM_INTINFO_HWEXCEPTION && vector >= 32) 1996 return (EINVAL); 1997 if (info & VM_INTINFO_RSVD) 1998 return (EINVAL); 1999 } else { 2000 info = 0; 2001 } 2002 VMM_CTR2(vcpu, "%s: info1(%#lx)", __func__, info); 2003 vcpu->exitintinfo = info; 2004 return (0); 2005 } 2006 2007 enum exc_class { 2008 EXC_BENIGN, 2009 EXC_CONTRIBUTORY, 2010 EXC_PAGEFAULT 2011 }; 2012 2013 #define IDT_VE 20 /* Virtualization Exception (Intel specific) */ 2014 2015 static enum exc_class 2016 exception_class(uint64_t info) 2017 { 2018 int type, vector; 2019 2020 KASSERT(info & VM_INTINFO_VALID, ("intinfo must be valid: %#lx", info)); 2021 type = info & VM_INTINFO_TYPE; 2022 vector = info & 0xff; 2023 2024 /* Table 6-4, "Interrupt and Exception Classes", Intel SDM, Vol 3 */ 2025 switch (type) { 2026 case VM_INTINFO_HWINTR: 2027 case VM_INTINFO_SWINTR: 2028 case VM_INTINFO_NMI: 2029 return (EXC_BENIGN); 2030 default: 2031 /* 2032 * Hardware exception. 2033 * 2034 * SVM and VT-x use identical type values to represent NMI, 2035 * hardware interrupt and software interrupt. 2036 * 2037 * SVM uses type '3' for all exceptions. VT-x uses type '3' 2038 * for exceptions except #BP and #OF. #BP and #OF use a type 2039 * value of '5' or '6'. Therefore we don't check for explicit 2040 * values of 'type' to classify 'intinfo' into a hardware 2041 * exception. 2042 */ 2043 break; 2044 } 2045 2046 switch (vector) { 2047 case IDT_PF: 2048 case IDT_VE: 2049 return (EXC_PAGEFAULT); 2050 case IDT_DE: 2051 case IDT_TS: 2052 case IDT_NP: 2053 case IDT_SS: 2054 case IDT_GP: 2055 return (EXC_CONTRIBUTORY); 2056 default: 2057 return (EXC_BENIGN); 2058 } 2059 } 2060 2061 static int 2062 nested_fault(struct vcpu *vcpu, uint64_t info1, uint64_t info2, 2063 uint64_t *retinfo) 2064 { 2065 enum exc_class exc1, exc2; 2066 int type1, vector1; 2067 2068 KASSERT(info1 & VM_INTINFO_VALID, ("info1 %#lx is not valid", info1)); 2069 KASSERT(info2 & VM_INTINFO_VALID, ("info2 %#lx is not valid", info2)); 2070 2071 /* 2072 * If an exception occurs while attempting to call the double-fault 2073 * handler the processor enters shutdown mode (aka triple fault). 2074 */ 2075 type1 = info1 & VM_INTINFO_TYPE; 2076 vector1 = info1 & 0xff; 2077 if (type1 == VM_INTINFO_HWEXCEPTION && vector1 == IDT_DF) { 2078 VMM_CTR2(vcpu, "triple fault: info1(%#lx), info2(%#lx)", 2079 info1, info2); 2080 vm_suspend(vcpu->vm, VM_SUSPEND_TRIPLEFAULT); 2081 *retinfo = 0; 2082 return (0); 2083 } 2084 2085 /* 2086 * Table 6-5 "Conditions for Generating a Double Fault", Intel SDM, Vol3 2087 */ 2088 exc1 = exception_class(info1); 2089 exc2 = exception_class(info2); 2090 if ((exc1 == EXC_CONTRIBUTORY && exc2 == EXC_CONTRIBUTORY) || 2091 (exc1 == EXC_PAGEFAULT && exc2 != EXC_BENIGN)) { 2092 /* Convert nested fault into a double fault. */ 2093 *retinfo = IDT_DF; 2094 *retinfo |= VM_INTINFO_VALID | VM_INTINFO_HWEXCEPTION; 2095 *retinfo |= VM_INTINFO_DEL_ERRCODE; 2096 } else { 2097 /* Handle exceptions serially */ 2098 *retinfo = info2; 2099 } 2100 return (1); 2101 } 2102 2103 static uint64_t 2104 vcpu_exception_intinfo(struct vcpu *vcpu) 2105 { 2106 uint64_t info = 0; 2107 2108 if (vcpu->exception_pending) { 2109 info = vcpu->exc_vector & 0xff; 2110 info |= VM_INTINFO_VALID | VM_INTINFO_HWEXCEPTION; 2111 if (vcpu->exc_errcode_valid) { 2112 info |= VM_INTINFO_DEL_ERRCODE; 2113 info |= (uint64_t)vcpu->exc_errcode << 32; 2114 } 2115 } 2116 return (info); 2117 } 2118 2119 int 2120 vm_entry_intinfo(struct vcpu *vcpu, uint64_t *retinfo) 2121 { 2122 uint64_t info1, info2; 2123 int valid; 2124 2125 info1 = vcpu->exitintinfo; 2126 vcpu->exitintinfo = 0; 2127 2128 info2 = 0; 2129 if (vcpu->exception_pending) { 2130 info2 = vcpu_exception_intinfo(vcpu); 2131 vcpu->exception_pending = 0; 2132 VMM_CTR2(vcpu, "Exception %d delivered: %#lx", 2133 vcpu->exc_vector, info2); 2134 } 2135 2136 if ((info1 & VM_INTINFO_VALID) && (info2 & VM_INTINFO_VALID)) { 2137 valid = nested_fault(vcpu, info1, info2, retinfo); 2138 } else if (info1 & VM_INTINFO_VALID) { 2139 *retinfo = info1; 2140 valid = 1; 2141 } else if (info2 & VM_INTINFO_VALID) { 2142 *retinfo = info2; 2143 valid = 1; 2144 } else { 2145 valid = 0; 2146 } 2147 2148 if (valid) { 2149 VMM_CTR4(vcpu, "%s: info1(%#lx), info2(%#lx), " 2150 "retinfo(%#lx)", __func__, info1, info2, *retinfo); 2151 } 2152 2153 return (valid); 2154 } 2155 2156 int 2157 vm_get_intinfo(struct vcpu *vcpu, uint64_t *info1, uint64_t *info2) 2158 { 2159 *info1 = vcpu->exitintinfo; 2160 *info2 = vcpu_exception_intinfo(vcpu); 2161 return (0); 2162 } 2163 2164 int 2165 vm_inject_exception(struct vcpu *vcpu, int vector, int errcode_valid, 2166 uint32_t errcode, int restart_instruction) 2167 { 2168 uint64_t regval; 2169 int error __diagused; 2170 2171 if (vector < 0 || vector >= 32) 2172 return (EINVAL); 2173 2174 /* 2175 * A double fault exception should never be injected directly into 2176 * the guest. It is a derived exception that results from specific 2177 * combinations of nested faults. 2178 */ 2179 if (vector == IDT_DF) 2180 return (EINVAL); 2181 2182 if (vcpu->exception_pending) { 2183 VMM_CTR2(vcpu, "Unable to inject exception %d due to " 2184 "pending exception %d", vector, vcpu->exc_vector); 2185 return (EBUSY); 2186 } 2187 2188 if (errcode_valid) { 2189 /* 2190 * Exceptions don't deliver an error code in real mode. 2191 */ 2192 error = vm_get_register(vcpu, VM_REG_GUEST_CR0, ®val); 2193 KASSERT(!error, ("%s: error %d getting CR0", __func__, error)); 2194 if (!(regval & CR0_PE)) 2195 errcode_valid = 0; 2196 } 2197 2198 /* 2199 * From section 26.6.1 "Interruptibility State" in Intel SDM: 2200 * 2201 * Event blocking by "STI" or "MOV SS" is cleared after guest executes 2202 * one instruction or incurs an exception. 2203 */ 2204 error = vm_set_register(vcpu, VM_REG_GUEST_INTR_SHADOW, 0); 2205 KASSERT(error == 0, ("%s: error %d clearing interrupt shadow", 2206 __func__, error)); 2207 2208 if (restart_instruction) 2209 vm_restart_instruction(vcpu); 2210 2211 vcpu->exception_pending = 1; 2212 vcpu->exc_vector = vector; 2213 vcpu->exc_errcode = errcode; 2214 vcpu->exc_errcode_valid = errcode_valid; 2215 VMM_CTR1(vcpu, "Exception %d pending", vector); 2216 return (0); 2217 } 2218 2219 void 2220 vm_inject_fault(struct vcpu *vcpu, int vector, int errcode_valid, int errcode) 2221 { 2222 int error __diagused, restart_instruction; 2223 2224 restart_instruction = 1; 2225 2226 error = vm_inject_exception(vcpu, vector, errcode_valid, 2227 errcode, restart_instruction); 2228 KASSERT(error == 0, ("vm_inject_exception error %d", error)); 2229 } 2230 2231 void 2232 vm_inject_pf(struct vcpu *vcpu, int error_code, uint64_t cr2) 2233 { 2234 int error __diagused; 2235 2236 VMM_CTR2(vcpu, "Injecting page fault: error_code %#x, cr2 %#lx", 2237 error_code, cr2); 2238 2239 error = vm_set_register(vcpu, VM_REG_GUEST_CR2, cr2); 2240 KASSERT(error == 0, ("vm_set_register(cr2) error %d", error)); 2241 2242 vm_inject_fault(vcpu, IDT_PF, 1, error_code); 2243 } 2244 2245 static VMM_STAT(VCPU_NMI_COUNT, "number of NMIs delivered to vcpu"); 2246 2247 int 2248 vm_inject_nmi(struct vcpu *vcpu) 2249 { 2250 2251 vcpu->nmi_pending = 1; 2252 vcpu_notify_event(vcpu, false); 2253 return (0); 2254 } 2255 2256 int 2257 vm_nmi_pending(struct vcpu *vcpu) 2258 { 2259 return (vcpu->nmi_pending); 2260 } 2261 2262 void 2263 vm_nmi_clear(struct vcpu *vcpu) 2264 { 2265 if (vcpu->nmi_pending == 0) 2266 panic("vm_nmi_clear: inconsistent nmi_pending state"); 2267 2268 vcpu->nmi_pending = 0; 2269 vmm_stat_incr(vcpu, VCPU_NMI_COUNT, 1); 2270 } 2271 2272 static VMM_STAT(VCPU_EXTINT_COUNT, "number of ExtINTs delivered to vcpu"); 2273 2274 int 2275 vm_inject_extint(struct vcpu *vcpu) 2276 { 2277 2278 vcpu->extint_pending = 1; 2279 vcpu_notify_event(vcpu, false); 2280 return (0); 2281 } 2282 2283 int 2284 vm_extint_pending(struct vcpu *vcpu) 2285 { 2286 return (vcpu->extint_pending); 2287 } 2288 2289 void 2290 vm_extint_clear(struct vcpu *vcpu) 2291 { 2292 if (vcpu->extint_pending == 0) 2293 panic("vm_extint_clear: inconsistent extint_pending state"); 2294 2295 vcpu->extint_pending = 0; 2296 vmm_stat_incr(vcpu, VCPU_EXTINT_COUNT, 1); 2297 } 2298 2299 int 2300 vm_get_capability(struct vcpu *vcpu, int type, int *retval) 2301 { 2302 if (type < 0 || type >= VM_CAP_MAX) 2303 return (EINVAL); 2304 2305 return (vmmops_getcap(vcpu->cookie, type, retval)); 2306 } 2307 2308 int 2309 vm_set_capability(struct vcpu *vcpu, int type, int val) 2310 { 2311 if (type < 0 || type >= VM_CAP_MAX) 2312 return (EINVAL); 2313 2314 return (vmmops_setcap(vcpu->cookie, type, val)); 2315 } 2316 2317 struct vm * 2318 vcpu_vm(struct vcpu *vcpu) 2319 { 2320 return (vcpu->vm); 2321 } 2322 2323 int 2324 vcpu_vcpuid(struct vcpu *vcpu) 2325 { 2326 return (vcpu->vcpuid); 2327 } 2328 2329 struct vcpu * 2330 vm_vcpu(struct vm *vm, int vcpuid) 2331 { 2332 return (vm->vcpu[vcpuid]); 2333 } 2334 2335 struct vlapic * 2336 vm_lapic(struct vcpu *vcpu) 2337 { 2338 return (vcpu->vlapic); 2339 } 2340 2341 struct vioapic * 2342 vm_ioapic(struct vm *vm) 2343 { 2344 2345 return (vm->vioapic); 2346 } 2347 2348 struct vhpet * 2349 vm_hpet(struct vm *vm) 2350 { 2351 2352 return (vm->vhpet); 2353 } 2354 2355 bool 2356 vmm_is_pptdev(int bus, int slot, int func) 2357 { 2358 int b, f, i, n, s; 2359 char *val, *cp, *cp2; 2360 bool found; 2361 2362 /* 2363 * XXX 2364 * The length of an environment variable is limited to 128 bytes which 2365 * puts an upper limit on the number of passthru devices that may be 2366 * specified using a single environment variable. 2367 * 2368 * Work around this by scanning multiple environment variable 2369 * names instead of a single one - yuck! 2370 */ 2371 const char *names[] = { "pptdevs", "pptdevs2", "pptdevs3", NULL }; 2372 2373 /* set pptdevs="1/2/3 4/5/6 7/8/9 10/11/12" */ 2374 found = false; 2375 for (i = 0; names[i] != NULL && !found; i++) { 2376 cp = val = kern_getenv(names[i]); 2377 while (cp != NULL && *cp != '\0') { 2378 if ((cp2 = strchr(cp, ' ')) != NULL) 2379 *cp2 = '\0'; 2380 2381 n = sscanf(cp, "%d/%d/%d", &b, &s, &f); 2382 if (n == 3 && bus == b && slot == s && func == f) { 2383 found = true; 2384 break; 2385 } 2386 2387 if (cp2 != NULL) 2388 *cp2++ = ' '; 2389 2390 cp = cp2; 2391 } 2392 freeenv(val); 2393 } 2394 return (found); 2395 } 2396 2397 void * 2398 vm_iommu_domain(struct vm *vm) 2399 { 2400 2401 return (vm->iommu); 2402 } 2403 2404 int 2405 vcpu_set_state(struct vcpu *vcpu, enum vcpu_state newstate, bool from_idle) 2406 { 2407 int error; 2408 2409 vcpu_lock(vcpu); 2410 error = vcpu_set_state_locked(vcpu, newstate, from_idle); 2411 vcpu_unlock(vcpu); 2412 2413 return (error); 2414 } 2415 2416 enum vcpu_state 2417 vcpu_get_state(struct vcpu *vcpu, int *hostcpu) 2418 { 2419 enum vcpu_state state; 2420 2421 vcpu_lock(vcpu); 2422 state = vcpu->state; 2423 if (hostcpu != NULL) 2424 *hostcpu = vcpu->hostcpu; 2425 vcpu_unlock(vcpu); 2426 2427 return (state); 2428 } 2429 2430 int 2431 vm_activate_cpu(struct vcpu *vcpu) 2432 { 2433 struct vm *vm = vcpu->vm; 2434 2435 if (CPU_ISSET(vcpu->vcpuid, &vm->active_cpus)) 2436 return (EBUSY); 2437 2438 VMM_CTR0(vcpu, "activated"); 2439 CPU_SET_ATOMIC(vcpu->vcpuid, &vm->active_cpus); 2440 return (0); 2441 } 2442 2443 int 2444 vm_suspend_cpu(struct vm *vm, struct vcpu *vcpu) 2445 { 2446 if (vcpu == NULL) { 2447 vm->debug_cpus = vm->active_cpus; 2448 for (int i = 0; i < vm->maxcpus; i++) { 2449 if (CPU_ISSET(i, &vm->active_cpus)) 2450 vcpu_notify_event(vm_vcpu(vm, i), false); 2451 } 2452 } else { 2453 if (!CPU_ISSET(vcpu->vcpuid, &vm->active_cpus)) 2454 return (EINVAL); 2455 2456 CPU_SET_ATOMIC(vcpu->vcpuid, &vm->debug_cpus); 2457 vcpu_notify_event(vcpu, false); 2458 } 2459 return (0); 2460 } 2461 2462 int 2463 vm_resume_cpu(struct vm *vm, struct vcpu *vcpu) 2464 { 2465 2466 if (vcpu == NULL) { 2467 CPU_ZERO(&vm->debug_cpus); 2468 } else { 2469 if (!CPU_ISSET(vcpu->vcpuid, &vm->debug_cpus)) 2470 return (EINVAL); 2471 2472 CPU_CLR_ATOMIC(vcpu->vcpuid, &vm->debug_cpus); 2473 } 2474 return (0); 2475 } 2476 2477 int 2478 vcpu_debugged(struct vcpu *vcpu) 2479 { 2480 2481 return (CPU_ISSET(vcpu->vcpuid, &vcpu->vm->debug_cpus)); 2482 } 2483 2484 cpuset_t 2485 vm_active_cpus(struct vm *vm) 2486 { 2487 2488 return (vm->active_cpus); 2489 } 2490 2491 cpuset_t 2492 vm_debug_cpus(struct vm *vm) 2493 { 2494 2495 return (vm->debug_cpus); 2496 } 2497 2498 cpuset_t 2499 vm_suspended_cpus(struct vm *vm) 2500 { 2501 2502 return (vm->suspended_cpus); 2503 } 2504 2505 /* 2506 * Returns the subset of vCPUs in tostart that are awaiting startup. 2507 * These vCPUs are also marked as no longer awaiting startup. 2508 */ 2509 cpuset_t 2510 vm_start_cpus(struct vm *vm, const cpuset_t *tostart) 2511 { 2512 cpuset_t set; 2513 2514 mtx_lock(&vm->rendezvous_mtx); 2515 CPU_AND(&set, &vm->startup_cpus, tostart); 2516 CPU_ANDNOT(&vm->startup_cpus, &vm->startup_cpus, &set); 2517 mtx_unlock(&vm->rendezvous_mtx); 2518 return (set); 2519 } 2520 2521 void 2522 vm_await_start(struct vm *vm, const cpuset_t *waiting) 2523 { 2524 mtx_lock(&vm->rendezvous_mtx); 2525 CPU_OR(&vm->startup_cpus, &vm->startup_cpus, waiting); 2526 mtx_unlock(&vm->rendezvous_mtx); 2527 } 2528 2529 void * 2530 vcpu_stats(struct vcpu *vcpu) 2531 { 2532 2533 return (vcpu->stats); 2534 } 2535 2536 int 2537 vm_get_x2apic_state(struct vcpu *vcpu, enum x2apic_state *state) 2538 { 2539 *state = vcpu->x2apic_state; 2540 2541 return (0); 2542 } 2543 2544 int 2545 vm_set_x2apic_state(struct vcpu *vcpu, enum x2apic_state state) 2546 { 2547 if (state >= X2APIC_STATE_LAST) 2548 return (EINVAL); 2549 2550 vcpu->x2apic_state = state; 2551 2552 vlapic_set_x2apic_state(vcpu, state); 2553 2554 return (0); 2555 } 2556 2557 /* 2558 * This function is called to ensure that a vcpu "sees" a pending event 2559 * as soon as possible: 2560 * - If the vcpu thread is sleeping then it is woken up. 2561 * - If the vcpu is running on a different host_cpu then an IPI will be directed 2562 * to the host_cpu to cause the vcpu to trap into the hypervisor. 2563 */ 2564 static void 2565 vcpu_notify_event_locked(struct vcpu *vcpu, bool lapic_intr) 2566 { 2567 int hostcpu; 2568 2569 hostcpu = vcpu->hostcpu; 2570 if (vcpu->state == VCPU_RUNNING) { 2571 KASSERT(hostcpu != NOCPU, ("vcpu running on invalid hostcpu")); 2572 if (hostcpu != curcpu) { 2573 if (lapic_intr) { 2574 vlapic_post_intr(vcpu->vlapic, hostcpu, 2575 vmm_ipinum); 2576 } else { 2577 ipi_cpu(hostcpu, vmm_ipinum); 2578 } 2579 } else { 2580 /* 2581 * If the 'vcpu' is running on 'curcpu' then it must 2582 * be sending a notification to itself (e.g. SELF_IPI). 2583 * The pending event will be picked up when the vcpu 2584 * transitions back to guest context. 2585 */ 2586 } 2587 } else { 2588 KASSERT(hostcpu == NOCPU, ("vcpu state %d not consistent " 2589 "with hostcpu %d", vcpu->state, hostcpu)); 2590 if (vcpu->state == VCPU_SLEEPING) 2591 wakeup_one(vcpu); 2592 } 2593 } 2594 2595 void 2596 vcpu_notify_event(struct vcpu *vcpu, bool lapic_intr) 2597 { 2598 vcpu_lock(vcpu); 2599 vcpu_notify_event_locked(vcpu, lapic_intr); 2600 vcpu_unlock(vcpu); 2601 } 2602 2603 struct vmspace * 2604 vm_get_vmspace(struct vm *vm) 2605 { 2606 2607 return (vm->vmspace); 2608 } 2609 2610 int 2611 vm_apicid2vcpuid(struct vm *vm, int apicid) 2612 { 2613 /* 2614 * XXX apic id is assumed to be numerically identical to vcpu id 2615 */ 2616 return (apicid); 2617 } 2618 2619 int 2620 vm_smp_rendezvous(struct vcpu *vcpu, cpuset_t dest, 2621 vm_rendezvous_func_t func, void *arg) 2622 { 2623 struct vm *vm = vcpu->vm; 2624 int error, i; 2625 2626 /* 2627 * Enforce that this function is called without any locks 2628 */ 2629 WITNESS_WARN(WARN_PANIC, NULL, "vm_smp_rendezvous"); 2630 2631 restart: 2632 mtx_lock(&vm->rendezvous_mtx); 2633 if (vm->rendezvous_func != NULL) { 2634 /* 2635 * If a rendezvous is already in progress then we need to 2636 * call the rendezvous handler in case this 'vcpu' is one 2637 * of the targets of the rendezvous. 2638 */ 2639 VMM_CTR0(vcpu, "Rendezvous already in progress"); 2640 mtx_unlock(&vm->rendezvous_mtx); 2641 error = vm_handle_rendezvous(vcpu); 2642 if (error != 0) 2643 return (error); 2644 goto restart; 2645 } 2646 KASSERT(vm->rendezvous_func == NULL, ("vm_smp_rendezvous: previous " 2647 "rendezvous is still in progress")); 2648 2649 VMM_CTR0(vcpu, "Initiating rendezvous"); 2650 vm->rendezvous_req_cpus = dest; 2651 CPU_ZERO(&vm->rendezvous_done_cpus); 2652 vm->rendezvous_arg = arg; 2653 vm->rendezvous_func = func; 2654 mtx_unlock(&vm->rendezvous_mtx); 2655 2656 /* 2657 * Wake up any sleeping vcpus and trigger a VM-exit in any running 2658 * vcpus so they handle the rendezvous as soon as possible. 2659 */ 2660 for (i = 0; i < vm->maxcpus; i++) { 2661 if (CPU_ISSET(i, &dest)) 2662 vcpu_notify_event(vm_vcpu(vm, i), false); 2663 } 2664 2665 return (vm_handle_rendezvous(vcpu)); 2666 } 2667 2668 struct vatpic * 2669 vm_atpic(struct vm *vm) 2670 { 2671 return (vm->vatpic); 2672 } 2673 2674 struct vatpit * 2675 vm_atpit(struct vm *vm) 2676 { 2677 return (vm->vatpit); 2678 } 2679 2680 struct vpmtmr * 2681 vm_pmtmr(struct vm *vm) 2682 { 2683 2684 return (vm->vpmtmr); 2685 } 2686 2687 struct vrtc * 2688 vm_rtc(struct vm *vm) 2689 { 2690 2691 return (vm->vrtc); 2692 } 2693 2694 enum vm_reg_name 2695 vm_segment_name(int seg) 2696 { 2697 static enum vm_reg_name seg_names[] = { 2698 VM_REG_GUEST_ES, 2699 VM_REG_GUEST_CS, 2700 VM_REG_GUEST_SS, 2701 VM_REG_GUEST_DS, 2702 VM_REG_GUEST_FS, 2703 VM_REG_GUEST_GS 2704 }; 2705 2706 KASSERT(seg >= 0 && seg < nitems(seg_names), 2707 ("%s: invalid segment encoding %d", __func__, seg)); 2708 return (seg_names[seg]); 2709 } 2710 2711 void 2712 vm_copy_teardown(struct vm_copyinfo *copyinfo, int num_copyinfo) 2713 { 2714 int idx; 2715 2716 for (idx = 0; idx < num_copyinfo; idx++) { 2717 if (copyinfo[idx].cookie != NULL) 2718 vm_gpa_release(copyinfo[idx].cookie); 2719 } 2720 bzero(copyinfo, num_copyinfo * sizeof(struct vm_copyinfo)); 2721 } 2722 2723 int 2724 vm_copy_setup(struct vcpu *vcpu, struct vm_guest_paging *paging, 2725 uint64_t gla, size_t len, int prot, struct vm_copyinfo *copyinfo, 2726 int num_copyinfo, int *fault) 2727 { 2728 int error, idx, nused; 2729 size_t n, off, remaining; 2730 void *hva, *cookie; 2731 uint64_t gpa; 2732 2733 bzero(copyinfo, sizeof(struct vm_copyinfo) * num_copyinfo); 2734 2735 nused = 0; 2736 remaining = len; 2737 while (remaining > 0) { 2738 KASSERT(nused < num_copyinfo, ("insufficient vm_copyinfo")); 2739 error = vm_gla2gpa(vcpu, paging, gla, prot, &gpa, fault); 2740 if (error || *fault) 2741 return (error); 2742 off = gpa & PAGE_MASK; 2743 n = min(remaining, PAGE_SIZE - off); 2744 copyinfo[nused].gpa = gpa; 2745 copyinfo[nused].len = n; 2746 remaining -= n; 2747 gla += n; 2748 nused++; 2749 } 2750 2751 for (idx = 0; idx < nused; idx++) { 2752 hva = vm_gpa_hold(vcpu, copyinfo[idx].gpa, 2753 copyinfo[idx].len, prot, &cookie); 2754 if (hva == NULL) 2755 break; 2756 copyinfo[idx].hva = hva; 2757 copyinfo[idx].cookie = cookie; 2758 } 2759 2760 if (idx != nused) { 2761 vm_copy_teardown(copyinfo, num_copyinfo); 2762 return (EFAULT); 2763 } else { 2764 *fault = 0; 2765 return (0); 2766 } 2767 } 2768 2769 void 2770 vm_copyin(struct vm_copyinfo *copyinfo, void *kaddr, size_t len) 2771 { 2772 char *dst; 2773 int idx; 2774 2775 dst = kaddr; 2776 idx = 0; 2777 while (len > 0) { 2778 bcopy(copyinfo[idx].hva, dst, copyinfo[idx].len); 2779 len -= copyinfo[idx].len; 2780 dst += copyinfo[idx].len; 2781 idx++; 2782 } 2783 } 2784 2785 void 2786 vm_copyout(const void *kaddr, struct vm_copyinfo *copyinfo, size_t len) 2787 { 2788 const char *src; 2789 int idx; 2790 2791 src = kaddr; 2792 idx = 0; 2793 while (len > 0) { 2794 bcopy(src, copyinfo[idx].hva, copyinfo[idx].len); 2795 len -= copyinfo[idx].len; 2796 src += copyinfo[idx].len; 2797 idx++; 2798 } 2799 } 2800 2801 /* 2802 * Return the amount of in-use and wired memory for the VM. Since 2803 * these are global stats, only return the values with for vCPU 0 2804 */ 2805 VMM_STAT_DECLARE(VMM_MEM_RESIDENT); 2806 VMM_STAT_DECLARE(VMM_MEM_WIRED); 2807 2808 static void 2809 vm_get_rescnt(struct vcpu *vcpu, struct vmm_stat_type *stat) 2810 { 2811 2812 if (vcpu->vcpuid == 0) { 2813 vmm_stat_set(vcpu, VMM_MEM_RESIDENT, PAGE_SIZE * 2814 vmspace_resident_count(vcpu->vm->vmspace)); 2815 } 2816 } 2817 2818 static void 2819 vm_get_wiredcnt(struct vcpu *vcpu, struct vmm_stat_type *stat) 2820 { 2821 2822 if (vcpu->vcpuid == 0) { 2823 vmm_stat_set(vcpu, VMM_MEM_WIRED, PAGE_SIZE * 2824 pmap_wired_count(vmspace_pmap(vcpu->vm->vmspace))); 2825 } 2826 } 2827 2828 VMM_STAT_FUNC(VMM_MEM_RESIDENT, "Resident memory", vm_get_rescnt); 2829 VMM_STAT_FUNC(VMM_MEM_WIRED, "Wired memory", vm_get_wiredcnt); 2830 2831 #ifdef BHYVE_SNAPSHOT 2832 static int 2833 vm_snapshot_vcpus(struct vm *vm, struct vm_snapshot_meta *meta) 2834 { 2835 uint64_t tsc, now; 2836 int ret; 2837 struct vcpu *vcpu; 2838 uint16_t i, maxcpus; 2839 2840 now = rdtsc(); 2841 maxcpus = vm_get_maxcpus(vm); 2842 for (i = 0; i < maxcpus; i++) { 2843 vcpu = vm->vcpu[i]; 2844 if (vcpu == NULL) 2845 continue; 2846 2847 SNAPSHOT_VAR_OR_LEAVE(vcpu->x2apic_state, meta, ret, done); 2848 SNAPSHOT_VAR_OR_LEAVE(vcpu->exitintinfo, meta, ret, done); 2849 SNAPSHOT_VAR_OR_LEAVE(vcpu->exc_vector, meta, ret, done); 2850 SNAPSHOT_VAR_OR_LEAVE(vcpu->exc_errcode_valid, meta, ret, done); 2851 SNAPSHOT_VAR_OR_LEAVE(vcpu->exc_errcode, meta, ret, done); 2852 SNAPSHOT_VAR_OR_LEAVE(vcpu->guest_xcr0, meta, ret, done); 2853 SNAPSHOT_VAR_OR_LEAVE(vcpu->exitinfo, meta, ret, done); 2854 SNAPSHOT_VAR_OR_LEAVE(vcpu->nextrip, meta, ret, done); 2855 2856 /* 2857 * Save the absolute TSC value by adding now to tsc_offset. 2858 * 2859 * It will be turned turned back into an actual offset when the 2860 * TSC restore function is called 2861 */ 2862 tsc = now + vcpu->tsc_offset; 2863 SNAPSHOT_VAR_OR_LEAVE(tsc, meta, ret, done); 2864 if (meta->op == VM_SNAPSHOT_RESTORE) 2865 vcpu->tsc_offset = tsc; 2866 } 2867 2868 done: 2869 return (ret); 2870 } 2871 2872 static int 2873 vm_snapshot_vm(struct vm *vm, struct vm_snapshot_meta *meta) 2874 { 2875 int ret; 2876 2877 ret = vm_snapshot_vcpus(vm, meta); 2878 if (ret != 0) 2879 goto done; 2880 2881 SNAPSHOT_VAR_OR_LEAVE(vm->startup_cpus, meta, ret, done); 2882 done: 2883 return (ret); 2884 } 2885 2886 static int 2887 vm_snapshot_vcpu(struct vm *vm, struct vm_snapshot_meta *meta) 2888 { 2889 int error; 2890 struct vcpu *vcpu; 2891 uint16_t i, maxcpus; 2892 2893 error = 0; 2894 2895 maxcpus = vm_get_maxcpus(vm); 2896 for (i = 0; i < maxcpus; i++) { 2897 vcpu = vm->vcpu[i]; 2898 if (vcpu == NULL) 2899 continue; 2900 2901 error = vmmops_vcpu_snapshot(vcpu->cookie, meta); 2902 if (error != 0) { 2903 printf("%s: failed to snapshot vmcs/vmcb data for " 2904 "vCPU: %d; error: %d\n", __func__, i, error); 2905 goto done; 2906 } 2907 } 2908 2909 done: 2910 return (error); 2911 } 2912 2913 /* 2914 * Save kernel-side structures to user-space for snapshotting. 2915 */ 2916 int 2917 vm_snapshot_req(struct vm *vm, struct vm_snapshot_meta *meta) 2918 { 2919 int ret = 0; 2920 2921 switch (meta->dev_req) { 2922 case STRUCT_VMCX: 2923 ret = vm_snapshot_vcpu(vm, meta); 2924 break; 2925 case STRUCT_VM: 2926 ret = vm_snapshot_vm(vm, meta); 2927 break; 2928 case STRUCT_VIOAPIC: 2929 ret = vioapic_snapshot(vm_ioapic(vm), meta); 2930 break; 2931 case STRUCT_VLAPIC: 2932 ret = vlapic_snapshot(vm, meta); 2933 break; 2934 case STRUCT_VHPET: 2935 ret = vhpet_snapshot(vm_hpet(vm), meta); 2936 break; 2937 case STRUCT_VATPIC: 2938 ret = vatpic_snapshot(vm_atpic(vm), meta); 2939 break; 2940 case STRUCT_VATPIT: 2941 ret = vatpit_snapshot(vm_atpit(vm), meta); 2942 break; 2943 case STRUCT_VPMTMR: 2944 ret = vpmtmr_snapshot(vm_pmtmr(vm), meta); 2945 break; 2946 case STRUCT_VRTC: 2947 ret = vrtc_snapshot(vm_rtc(vm), meta); 2948 break; 2949 default: 2950 printf("%s: failed to find the requested type %#x\n", 2951 __func__, meta->dev_req); 2952 ret = (EINVAL); 2953 } 2954 return (ret); 2955 } 2956 2957 void 2958 vm_set_tsc_offset(struct vcpu *vcpu, uint64_t offset) 2959 { 2960 vcpu->tsc_offset = offset; 2961 } 2962 2963 int 2964 vm_restore_time(struct vm *vm) 2965 { 2966 int error; 2967 uint64_t now; 2968 struct vcpu *vcpu; 2969 uint16_t i, maxcpus; 2970 2971 now = rdtsc(); 2972 2973 error = vhpet_restore_time(vm_hpet(vm)); 2974 if (error) 2975 return (error); 2976 2977 maxcpus = vm_get_maxcpus(vm); 2978 for (i = 0; i < maxcpus; i++) { 2979 vcpu = vm->vcpu[i]; 2980 if (vcpu == NULL) 2981 continue; 2982 2983 error = vmmops_restore_tsc(vcpu->cookie, 2984 vcpu->tsc_offset - now); 2985 if (error) 2986 return (error); 2987 } 2988 2989 return (0); 2990 } 2991 #endif 2992