1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 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 * $FreeBSD$ 29 */ 30 31 #include <sys/cdefs.h> 32 __FBSDID("$FreeBSD$"); 33 34 #include "opt_bhyve_snapshot.h" 35 36 #include <sys/param.h> 37 #include <sys/systm.h> 38 #include <sys/kernel.h> 39 #include <sys/module.h> 40 #include <sys/sysctl.h> 41 #include <sys/malloc.h> 42 #include <sys/pcpu.h> 43 #include <sys/lock.h> 44 #include <sys/mutex.h> 45 #include <sys/proc.h> 46 #include <sys/rwlock.h> 47 #include <sys/sched.h> 48 #include <sys/smp.h> 49 #include <sys/sx.h> 50 #include <sys/vnode.h> 51 52 #include <vm/vm.h> 53 #include <vm/vm_param.h> 54 #include <vm/vm_extern.h> 55 #include <vm/vm_object.h> 56 #include <vm/vm_page.h> 57 #include <vm/pmap.h> 58 #include <vm/vm_map.h> 59 #include <vm/vm_pager.h> 60 #include <vm/vm_kern.h> 61 #include <vm/vnode_pager.h> 62 #include <vm/swap_pager.h> 63 #include <vm/uma.h> 64 65 #include <machine/cpu.h> 66 #include <machine/pcb.h> 67 #include <machine/smp.h> 68 #include <machine/md_var.h> 69 #include <x86/psl.h> 70 #include <x86/apicreg.h> 71 #include <x86/ifunc.h> 72 73 #include <machine/vmm.h> 74 #include <machine/vmm_dev.h> 75 #include <machine/vmm_instruction_emul.h> 76 #include <machine/vmm_snapshot.h> 77 78 #include "vmm_ioport.h" 79 #include "vmm_ktr.h" 80 #include "vmm_host.h" 81 #include "vmm_mem.h" 82 #include "vmm_util.h" 83 #include "vatpic.h" 84 #include "vatpit.h" 85 #include "vhpet.h" 86 #include "vioapic.h" 87 #include "vlapic.h" 88 #include "vpmtmr.h" 89 #include "vrtc.h" 90 #include "vmm_stat.h" 91 #include "vmm_lapic.h" 92 93 #include "io/ppt.h" 94 #include "io/iommu.h" 95 96 struct vlapic; 97 98 /* 99 * Initialization: 100 * (a) allocated when vcpu is created 101 * (i) initialized when vcpu is created and when it is reinitialized 102 * (o) initialized the first time the vcpu is created 103 * (x) initialized before use 104 */ 105 struct vcpu { 106 struct mtx mtx; /* (o) protects 'state' and 'hostcpu' */ 107 enum vcpu_state state; /* (o) vcpu state */ 108 int vcpuid; /* (o) */ 109 int hostcpu; /* (o) vcpu's host cpu */ 110 int reqidle; /* (i) request vcpu to idle */ 111 struct vm *vm; /* (o) */ 112 void *cookie; /* (i) cpu-specific data */ 113 struct vlapic *vlapic; /* (i) APIC device model */ 114 enum x2apic_state x2apic_state; /* (i) APIC mode */ 115 uint64_t exitintinfo; /* (i) events pending at VM exit */ 116 int nmi_pending; /* (i) NMI pending */ 117 int extint_pending; /* (i) INTR pending */ 118 int exception_pending; /* (i) exception pending */ 119 int exc_vector; /* (x) exception collateral */ 120 int exc_errcode_valid; 121 uint32_t exc_errcode; 122 struct savefpu *guestfpu; /* (a,i) guest fpu state */ 123 uint64_t guest_xcr0; /* (i) guest %xcr0 register */ 124 void *stats; /* (a,i) statistics */ 125 struct vm_exit exitinfo; /* (x) exit reason and collateral */ 126 uint64_t nextrip; /* (x) next instruction to execute */ 127 uint64_t tsc_offset; /* (o) TSC offsetting */ 128 }; 129 130 #define vcpu_lock_init(v) mtx_init(&((v)->mtx), "vcpu lock", 0, MTX_SPIN) 131 #define vcpu_lock_destroy(v) mtx_destroy(&((v)->mtx)) 132 #define vcpu_lock(v) mtx_lock_spin(&((v)->mtx)) 133 #define vcpu_unlock(v) mtx_unlock_spin(&((v)->mtx)) 134 #define vcpu_assert_locked(v) mtx_assert(&((v)->mtx), MA_OWNED) 135 136 struct mem_seg { 137 size_t len; 138 bool sysmem; 139 struct vm_object *object; 140 }; 141 #define VM_MAX_MEMSEGS 4 142 143 struct mem_map { 144 vm_paddr_t gpa; 145 size_t len; 146 vm_ooffset_t segoff; 147 int segid; 148 int prot; 149 int flags; 150 }; 151 #define VM_MAX_MEMMAPS 8 152 153 /* 154 * Initialization: 155 * (o) initialized the first time the VM is created 156 * (i) initialized when VM is created and when it is reinitialized 157 * (x) initialized before use 158 * 159 * Locking: 160 * [m] mem_segs_lock 161 * [r] rendezvous_mtx 162 * [v] reads require one frozen vcpu, writes require freezing all vcpus 163 */ 164 struct vm { 165 void *cookie; /* (i) cpu-specific data */ 166 void *iommu; /* (x) iommu-specific data */ 167 struct vhpet *vhpet; /* (i) virtual HPET */ 168 struct vioapic *vioapic; /* (i) virtual ioapic */ 169 struct vatpic *vatpic; /* (i) virtual atpic */ 170 struct vatpit *vatpit; /* (i) virtual atpit */ 171 struct vpmtmr *vpmtmr; /* (i) virtual ACPI PM timer */ 172 struct vrtc *vrtc; /* (o) virtual RTC */ 173 volatile cpuset_t active_cpus; /* (i) active vcpus */ 174 volatile cpuset_t debug_cpus; /* (i) vcpus stopped for debug */ 175 cpuset_t startup_cpus; /* (i) [r] waiting for startup */ 176 int suspend; /* (i) stop VM execution */ 177 bool dying; /* (o) is dying */ 178 volatile cpuset_t suspended_cpus; /* (i) suspended vcpus */ 179 volatile cpuset_t halted_cpus; /* (x) cpus in a hard halt */ 180 cpuset_t rendezvous_req_cpus; /* (x) [r] rendezvous requested */ 181 cpuset_t rendezvous_done_cpus; /* (x) [r] rendezvous finished */ 182 void *rendezvous_arg; /* (x) [r] rendezvous func/arg */ 183 vm_rendezvous_func_t rendezvous_func; 184 struct mtx rendezvous_mtx; /* (o) rendezvous lock */ 185 struct mem_map mem_maps[VM_MAX_MEMMAPS]; /* (i) [m+v] guest address space */ 186 struct mem_seg mem_segs[VM_MAX_MEMSEGS]; /* (o) [m+v] guest memory regions */ 187 struct vmspace *vmspace; /* (o) guest's address space */ 188 char name[VM_MAX_NAMELEN+1]; /* (o) virtual machine name */ 189 struct vcpu **vcpu; /* (o) guest vcpus */ 190 /* The following describe the vm cpu topology */ 191 uint16_t sockets; /* (o) num of sockets */ 192 uint16_t cores; /* (o) num of cores/socket */ 193 uint16_t threads; /* (o) num of threads/core */ 194 uint16_t maxcpus; /* (o) max pluggable cpus */ 195 struct sx mem_segs_lock; /* (o) */ 196 struct sx vcpus_init_lock; /* (o) */ 197 }; 198 199 #define VMM_CTR0(vcpu, format) \ 200 VCPU_CTR0((vcpu)->vm, (vcpu)->vcpuid, format) 201 202 #define VMM_CTR1(vcpu, format, p1) \ 203 VCPU_CTR1((vcpu)->vm, (vcpu)->vcpuid, format, p1) 204 205 #define VMM_CTR2(vcpu, format, p1, p2) \ 206 VCPU_CTR2((vcpu)->vm, (vcpu)->vcpuid, format, p1, p2) 207 208 #define VMM_CTR3(vcpu, format, p1, p2, p3) \ 209 VCPU_CTR3((vcpu)->vm, (vcpu)->vcpuid, format, p1, p2, p3) 210 211 #define VMM_CTR4(vcpu, format, p1, p2, p3, p4) \ 212 VCPU_CTR4((vcpu)->vm, (vcpu)->vcpuid, format, p1, p2, p3, p4) 213 214 static int vmm_initialized; 215 216 static void vmmops_panic(void); 217 218 static void 219 vmmops_panic(void) 220 { 221 panic("vmm_ops func called when !vmm_is_intel() && !vmm_is_svm()"); 222 } 223 224 #define DEFINE_VMMOPS_IFUNC(ret_type, opname, args) \ 225 DEFINE_IFUNC(static, ret_type, vmmops_##opname, args) \ 226 { \ 227 if (vmm_is_intel()) \ 228 return (vmm_ops_intel.opname); \ 229 else if (vmm_is_svm()) \ 230 return (vmm_ops_amd.opname); \ 231 else \ 232 return ((ret_type (*)args)vmmops_panic); \ 233 } 234 235 DEFINE_VMMOPS_IFUNC(int, modinit, (int ipinum)) 236 DEFINE_VMMOPS_IFUNC(int, modcleanup, (void)) 237 DEFINE_VMMOPS_IFUNC(void, modresume, (void)) 238 DEFINE_VMMOPS_IFUNC(void *, init, (struct vm *vm, struct pmap *pmap)) 239 DEFINE_VMMOPS_IFUNC(int, run, (void *vcpui, register_t rip, struct pmap *pmap, 240 struct vm_eventinfo *info)) 241 DEFINE_VMMOPS_IFUNC(void, cleanup, (void *vmi)) 242 DEFINE_VMMOPS_IFUNC(void *, vcpu_init, (void *vmi, struct vcpu *vcpu, 243 int vcpu_id)) 244 DEFINE_VMMOPS_IFUNC(void, vcpu_cleanup, (void *vcpui)) 245 DEFINE_VMMOPS_IFUNC(int, getreg, (void *vcpui, int num, uint64_t *retval)) 246 DEFINE_VMMOPS_IFUNC(int, setreg, (void *vcpui, int num, uint64_t val)) 247 DEFINE_VMMOPS_IFUNC(int, getdesc, (void *vcpui, int num, struct seg_desc *desc)) 248 DEFINE_VMMOPS_IFUNC(int, setdesc, (void *vcpui, int num, struct seg_desc *desc)) 249 DEFINE_VMMOPS_IFUNC(int, getcap, (void *vcpui, int num, int *retval)) 250 DEFINE_VMMOPS_IFUNC(int, setcap, (void *vcpui, int num, int val)) 251 DEFINE_VMMOPS_IFUNC(struct vmspace *, vmspace_alloc, (vm_offset_t min, 252 vm_offset_t max)) 253 DEFINE_VMMOPS_IFUNC(void, vmspace_free, (struct vmspace *vmspace)) 254 DEFINE_VMMOPS_IFUNC(struct vlapic *, vlapic_init, (void *vcpui)) 255 DEFINE_VMMOPS_IFUNC(void, vlapic_cleanup, (struct vlapic *vlapic)) 256 #ifdef BHYVE_SNAPSHOT 257 DEFINE_VMMOPS_IFUNC(int, snapshot, (void *vmi, struct vm_snapshot_meta *meta)) 258 DEFINE_VMMOPS_IFUNC(int, vcpu_snapshot, (void *vcpui, 259 struct vm_snapshot_meta *meta)) 260 DEFINE_VMMOPS_IFUNC(int, restore_tsc, (void *vcpui, uint64_t now)) 261 #endif 262 263 #define fpu_start_emulating() load_cr0(rcr0() | CR0_TS) 264 #define fpu_stop_emulating() clts() 265 266 SDT_PROVIDER_DEFINE(vmm); 267 268 static MALLOC_DEFINE(M_VM, "vm", "vm"); 269 270 /* statistics */ 271 static VMM_STAT(VCPU_TOTAL_RUNTIME, "vcpu total runtime"); 272 273 SYSCTL_NODE(_hw, OID_AUTO, vmm, CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, 274 NULL); 275 276 /* 277 * Halt the guest if all vcpus are executing a HLT instruction with 278 * interrupts disabled. 279 */ 280 static int halt_detection_enabled = 1; 281 SYSCTL_INT(_hw_vmm, OID_AUTO, halt_detection, CTLFLAG_RDTUN, 282 &halt_detection_enabled, 0, 283 "Halt VM if all vcpus execute HLT with interrupts disabled"); 284 285 static int vmm_ipinum; 286 SYSCTL_INT(_hw_vmm, OID_AUTO, ipinum, CTLFLAG_RD, &vmm_ipinum, 0, 287 "IPI vector used for vcpu notifications"); 288 289 static int trace_guest_exceptions; 290 SYSCTL_INT(_hw_vmm, OID_AUTO, trace_guest_exceptions, CTLFLAG_RDTUN, 291 &trace_guest_exceptions, 0, 292 "Trap into hypervisor on all guest exceptions and reflect them back"); 293 294 static int trap_wbinvd; 295 SYSCTL_INT(_hw_vmm, OID_AUTO, trap_wbinvd, CTLFLAG_RDTUN, &trap_wbinvd, 0, 296 "WBINVD triggers a VM-exit"); 297 298 u_int vm_maxcpu; 299 SYSCTL_UINT(_hw_vmm, OID_AUTO, maxcpu, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, 300 &vm_maxcpu, 0, "Maximum number of vCPUs"); 301 302 static void vm_free_memmap(struct vm *vm, int ident); 303 static bool sysmem_mapping(struct vm *vm, struct mem_map *mm); 304 static void vcpu_notify_event_locked(struct vcpu *vcpu, bool lapic_intr); 305 306 /* 307 * Upper limit on vm_maxcpu. Limited by use of uint16_t types for CPU 308 * counts as well as range of vpid values for VT-x and by the capacity 309 * of cpuset_t masks. The call to new_unrhdr() in vpid_init() in 310 * vmx.c requires 'vm_maxcpu + 1 <= 0xffff', hence the '- 1' below. 311 */ 312 #define VM_MAXCPU MIN(0xffff - 1, CPU_SETSIZE) 313 314 #ifdef KTR 315 static const char * 316 vcpu_state2str(enum vcpu_state state) 317 { 318 319 switch (state) { 320 case VCPU_IDLE: 321 return ("idle"); 322 case VCPU_FROZEN: 323 return ("frozen"); 324 case VCPU_RUNNING: 325 return ("running"); 326 case VCPU_SLEEPING: 327 return ("sleeping"); 328 default: 329 return ("unknown"); 330 } 331 } 332 #endif 333 334 static void 335 vcpu_cleanup(struct vcpu *vcpu, bool destroy) 336 { 337 vmmops_vlapic_cleanup(vcpu->vlapic); 338 vmmops_vcpu_cleanup(vcpu->cookie); 339 vcpu->cookie = NULL; 340 if (destroy) { 341 vmm_stat_free(vcpu->stats); 342 fpu_save_area_free(vcpu->guestfpu); 343 vcpu_lock_destroy(vcpu); 344 free(vcpu, M_VM); 345 } 346 } 347 348 static struct vcpu * 349 vcpu_alloc(struct vm *vm, int vcpu_id) 350 { 351 struct vcpu *vcpu; 352 353 KASSERT(vcpu_id >= 0 && vcpu_id < vm->maxcpus, 354 ("vcpu_init: invalid vcpu %d", vcpu_id)); 355 356 vcpu = malloc(sizeof(*vcpu), M_VM, M_WAITOK | M_ZERO); 357 vcpu_lock_init(vcpu); 358 vcpu->state = VCPU_IDLE; 359 vcpu->hostcpu = NOCPU; 360 vcpu->vcpuid = vcpu_id; 361 vcpu->vm = vm; 362 vcpu->guestfpu = fpu_save_area_alloc(); 363 vcpu->stats = vmm_stat_alloc(); 364 vcpu->tsc_offset = 0; 365 return (vcpu); 366 } 367 368 static void 369 vcpu_init(struct vcpu *vcpu) 370 { 371 vcpu->cookie = vmmops_vcpu_init(vcpu->vm->cookie, vcpu, vcpu->vcpuid); 372 vcpu->vlapic = vmmops_vlapic_init(vcpu->cookie); 373 vm_set_x2apic_state(vcpu, X2APIC_DISABLED); 374 vcpu->reqidle = 0; 375 vcpu->exitintinfo = 0; 376 vcpu->nmi_pending = 0; 377 vcpu->extint_pending = 0; 378 vcpu->exception_pending = 0; 379 vcpu->guest_xcr0 = XFEATURE_ENABLED_X87; 380 fpu_save_area_reset(vcpu->guestfpu); 381 vmm_stat_init(vcpu->stats); 382 } 383 384 int 385 vcpu_trace_exceptions(struct vcpu *vcpu) 386 { 387 388 return (trace_guest_exceptions); 389 } 390 391 int 392 vcpu_trap_wbinvd(struct vcpu *vcpu) 393 { 394 return (trap_wbinvd); 395 } 396 397 struct vm_exit * 398 vm_exitinfo(struct vcpu *vcpu) 399 { 400 return (&vcpu->exitinfo); 401 } 402 403 static int 404 vmm_init(void) 405 { 406 int error; 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 error = vmm_mem_init(); 429 if (error) 430 return (error); 431 432 vmm_resume_p = vmmops_modresume; 433 434 return (vmmops_modinit(vmm_ipinum)); 435 } 436 437 static int 438 vmm_handler(module_t mod, int what, void *arg) 439 { 440 int error; 441 442 switch (what) { 443 case MOD_LOAD: 444 if (vmm_is_hw_supported()) { 445 vmmdev_init(); 446 error = vmm_init(); 447 if (error == 0) 448 vmm_initialized = 1; 449 } else { 450 error = ENXIO; 451 } 452 break; 453 case MOD_UNLOAD: 454 if (vmm_is_hw_supported()) { 455 error = vmmdev_cleanup(); 456 if (error == 0) { 457 vmm_resume_p = NULL; 458 iommu_cleanup(); 459 if (vmm_ipinum != IPI_AST) 460 lapic_ipi_free(vmm_ipinum); 461 error = vmmops_modcleanup(); 462 /* 463 * Something bad happened - prevent new 464 * VMs from being created 465 */ 466 if (error) 467 vmm_initialized = 0; 468 } 469 } else { 470 error = 0; 471 } 472 break; 473 default: 474 error = 0; 475 break; 476 } 477 return (error); 478 } 479 480 static moduledata_t vmm_kmod = { 481 "vmm", 482 vmm_handler, 483 NULL 484 }; 485 486 /* 487 * vmm initialization has the following dependencies: 488 * 489 * - VT-x initialization requires smp_rendezvous() and therefore must happen 490 * after SMP is fully functional (after SI_SUB_SMP). 491 */ 492 DECLARE_MODULE(vmm, vmm_kmod, SI_SUB_SMP + 1, SI_ORDER_ANY); 493 MODULE_VERSION(vmm, 1); 494 495 static void 496 vm_init(struct vm *vm, bool create) 497 { 498 vm->cookie = vmmops_init(vm, vmspace_pmap(vm->vmspace)); 499 vm->iommu = NULL; 500 vm->vioapic = vioapic_init(vm); 501 vm->vhpet = vhpet_init(vm); 502 vm->vatpic = vatpic_init(vm); 503 vm->vatpit = vatpit_init(vm); 504 vm->vpmtmr = vpmtmr_init(vm); 505 if (create) 506 vm->vrtc = vrtc_init(vm); 507 508 CPU_ZERO(&vm->active_cpus); 509 CPU_ZERO(&vm->debug_cpus); 510 CPU_ZERO(&vm->startup_cpus); 511 512 vm->suspend = 0; 513 CPU_ZERO(&vm->suspended_cpus); 514 515 if (!create) { 516 for (int i = 0; i < vm->maxcpus; i++) { 517 if (vm->vcpu[i] != NULL) 518 vcpu_init(vm->vcpu[i]); 519 } 520 } 521 } 522 523 void 524 vm_disable_vcpu_creation(struct vm *vm) 525 { 526 sx_xlock(&vm->vcpus_init_lock); 527 vm->dying = true; 528 sx_xunlock(&vm->vcpus_init_lock); 529 } 530 531 struct vcpu * 532 vm_alloc_vcpu(struct vm *vm, int vcpuid) 533 { 534 struct vcpu *vcpu; 535 536 if (vcpuid < 0 || vcpuid >= vm_get_maxcpus(vm)) 537 return (NULL); 538 539 vcpu = atomic_load_ptr(&vm->vcpu[vcpuid]); 540 if (__predict_true(vcpu != NULL)) 541 return (vcpu); 542 543 sx_xlock(&vm->vcpus_init_lock); 544 vcpu = vm->vcpu[vcpuid]; 545 if (vcpu == NULL && !vm->dying) { 546 vcpu = vcpu_alloc(vm, vcpuid); 547 vcpu_init(vcpu); 548 549 /* 550 * Ensure vCPU is fully created before updating pointer 551 * to permit unlocked reads above. 552 */ 553 atomic_store_rel_ptr((uintptr_t *)&vm->vcpu[vcpuid], 554 (uintptr_t)vcpu); 555 } 556 sx_xunlock(&vm->vcpus_init_lock); 557 return (vcpu); 558 } 559 560 void 561 vm_slock_vcpus(struct vm *vm) 562 { 563 sx_slock(&vm->vcpus_init_lock); 564 } 565 566 void 567 vm_unlock_vcpus(struct vm *vm) 568 { 569 sx_unlock(&vm->vcpus_init_lock); 570 } 571 572 /* 573 * The default CPU topology is a single thread per package. 574 */ 575 u_int cores_per_package = 1; 576 u_int threads_per_core = 1; 577 578 int 579 vm_create(const char *name, struct vm **retvm) 580 { 581 struct vm *vm; 582 struct vmspace *vmspace; 583 584 /* 585 * If vmm.ko could not be successfully initialized then don't attempt 586 * to create the virtual machine. 587 */ 588 if (!vmm_initialized) 589 return (ENXIO); 590 591 if (name == NULL || strnlen(name, VM_MAX_NAMELEN + 1) == 592 VM_MAX_NAMELEN + 1) 593 return (EINVAL); 594 595 vmspace = vmmops_vmspace_alloc(0, VM_MAXUSER_ADDRESS_LA48); 596 if (vmspace == NULL) 597 return (ENOMEM); 598 599 vm = malloc(sizeof(struct vm), M_VM, M_WAITOK | M_ZERO); 600 strcpy(vm->name, name); 601 vm->vmspace = vmspace; 602 mtx_init(&vm->rendezvous_mtx, "vm rendezvous lock", 0, MTX_DEF); 603 sx_init(&vm->mem_segs_lock, "vm mem_segs"); 604 sx_init(&vm->vcpus_init_lock, "vm vcpus"); 605 vm->vcpu = malloc(sizeof(*vm->vcpu) * vm_maxcpu, M_VM, M_WAITOK | 606 M_ZERO); 607 608 vm->sockets = 1; 609 vm->cores = cores_per_package; /* XXX backwards compatibility */ 610 vm->threads = threads_per_core; /* XXX backwards compatibility */ 611 vm->maxcpus = vm_maxcpu; 612 613 vm_init(vm, true); 614 615 *retvm = vm; 616 return (0); 617 } 618 619 void 620 vm_get_topology(struct vm *vm, uint16_t *sockets, uint16_t *cores, 621 uint16_t *threads, uint16_t *maxcpus) 622 { 623 *sockets = vm->sockets; 624 *cores = vm->cores; 625 *threads = vm->threads; 626 *maxcpus = vm->maxcpus; 627 } 628 629 uint16_t 630 vm_get_maxcpus(struct vm *vm) 631 { 632 return (vm->maxcpus); 633 } 634 635 int 636 vm_set_topology(struct vm *vm, uint16_t sockets, uint16_t cores, 637 uint16_t threads, uint16_t maxcpus __unused) 638 { 639 /* Ignore maxcpus. */ 640 if ((sockets * cores * threads) > vm->maxcpus) 641 return (EINVAL); 642 vm->sockets = sockets; 643 vm->cores = cores; 644 vm->threads = threads; 645 return(0); 646 } 647 648 static void 649 vm_cleanup(struct vm *vm, bool destroy) 650 { 651 struct mem_map *mm; 652 int i; 653 654 ppt_unassign_all(vm); 655 656 if (vm->iommu != NULL) 657 iommu_destroy_domain(vm->iommu); 658 659 if (destroy) 660 vrtc_cleanup(vm->vrtc); 661 else 662 vrtc_reset(vm->vrtc); 663 vpmtmr_cleanup(vm->vpmtmr); 664 vatpit_cleanup(vm->vatpit); 665 vhpet_cleanup(vm->vhpet); 666 vatpic_cleanup(vm->vatpic); 667 vioapic_cleanup(vm->vioapic); 668 669 for (i = 0; i < vm->maxcpus; i++) { 670 if (vm->vcpu[i] != NULL) 671 vcpu_cleanup(vm->vcpu[i], destroy); 672 } 673 674 vmmops_cleanup(vm->cookie); 675 676 /* 677 * System memory is removed from the guest address space only when 678 * the VM is destroyed. This is because the mapping remains the same 679 * across VM reset. 680 * 681 * Device memory can be relocated by the guest (e.g. using PCI BARs) 682 * so those mappings are removed on a VM reset. 683 */ 684 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 685 mm = &vm->mem_maps[i]; 686 if (destroy || !sysmem_mapping(vm, mm)) 687 vm_free_memmap(vm, i); 688 } 689 690 if (destroy) { 691 for (i = 0; i < VM_MAX_MEMSEGS; i++) 692 vm_free_memseg(vm, i); 693 694 vmmops_vmspace_free(vm->vmspace); 695 vm->vmspace = NULL; 696 697 free(vm->vcpu, M_VM); 698 sx_destroy(&vm->vcpus_init_lock); 699 sx_destroy(&vm->mem_segs_lock); 700 mtx_destroy(&vm->rendezvous_mtx); 701 } 702 } 703 704 void 705 vm_destroy(struct vm *vm) 706 { 707 vm_cleanup(vm, true); 708 free(vm, M_VM); 709 } 710 711 int 712 vm_reinit(struct vm *vm) 713 { 714 int error; 715 716 /* 717 * A virtual machine can be reset only if all vcpus are suspended. 718 */ 719 if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) { 720 vm_cleanup(vm, false); 721 vm_init(vm, false); 722 error = 0; 723 } else { 724 error = EBUSY; 725 } 726 727 return (error); 728 } 729 730 const char * 731 vm_name(struct vm *vm) 732 { 733 return (vm->name); 734 } 735 736 void 737 vm_slock_memsegs(struct vm *vm) 738 { 739 sx_slock(&vm->mem_segs_lock); 740 } 741 742 void 743 vm_xlock_memsegs(struct vm *vm) 744 { 745 sx_xlock(&vm->mem_segs_lock); 746 } 747 748 void 749 vm_unlock_memsegs(struct vm *vm) 750 { 751 sx_unlock(&vm->mem_segs_lock); 752 } 753 754 int 755 vm_map_mmio(struct vm *vm, vm_paddr_t gpa, size_t len, vm_paddr_t hpa) 756 { 757 vm_object_t obj; 758 759 if ((obj = vmm_mmio_alloc(vm->vmspace, gpa, len, hpa)) == NULL) 760 return (ENOMEM); 761 else 762 return (0); 763 } 764 765 int 766 vm_unmap_mmio(struct vm *vm, vm_paddr_t gpa, size_t len) 767 { 768 769 vmm_mmio_free(vm->vmspace, gpa, len); 770 return (0); 771 } 772 773 /* 774 * Return 'true' if 'gpa' is allocated in the guest address space. 775 * 776 * This function is called in the context of a running vcpu which acts as 777 * an implicit lock on 'vm->mem_maps[]'. 778 */ 779 bool 780 vm_mem_allocated(struct vcpu *vcpu, vm_paddr_t gpa) 781 { 782 struct vm *vm = vcpu->vm; 783 struct mem_map *mm; 784 int i; 785 786 #ifdef INVARIANTS 787 int hostcpu, state; 788 state = vcpu_get_state(vcpu, &hostcpu); 789 KASSERT(state == VCPU_RUNNING && hostcpu == curcpu, 790 ("%s: invalid vcpu state %d/%d", __func__, state, hostcpu)); 791 #endif 792 793 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 794 mm = &vm->mem_maps[i]; 795 if (mm->len != 0 && gpa >= mm->gpa && gpa < mm->gpa + mm->len) 796 return (true); /* 'gpa' is sysmem or devmem */ 797 } 798 799 if (ppt_is_mmio(vm, gpa)) 800 return (true); /* 'gpa' is pci passthru mmio */ 801 802 return (false); 803 } 804 805 int 806 vm_alloc_memseg(struct vm *vm, int ident, size_t len, bool sysmem) 807 { 808 struct mem_seg *seg; 809 vm_object_t obj; 810 811 sx_assert(&vm->mem_segs_lock, SX_XLOCKED); 812 813 if (ident < 0 || ident >= VM_MAX_MEMSEGS) 814 return (EINVAL); 815 816 if (len == 0 || (len & PAGE_MASK)) 817 return (EINVAL); 818 819 seg = &vm->mem_segs[ident]; 820 if (seg->object != NULL) { 821 if (seg->len == len && seg->sysmem == sysmem) 822 return (EEXIST); 823 else 824 return (EINVAL); 825 } 826 827 obj = vm_object_allocate(OBJT_SWAP, len >> PAGE_SHIFT); 828 if (obj == NULL) 829 return (ENOMEM); 830 831 seg->len = len; 832 seg->object = obj; 833 seg->sysmem = sysmem; 834 return (0); 835 } 836 837 int 838 vm_get_memseg(struct vm *vm, int ident, size_t *len, bool *sysmem, 839 vm_object_t *objptr) 840 { 841 struct mem_seg *seg; 842 843 sx_assert(&vm->mem_segs_lock, SX_LOCKED); 844 845 if (ident < 0 || ident >= VM_MAX_MEMSEGS) 846 return (EINVAL); 847 848 seg = &vm->mem_segs[ident]; 849 if (len) 850 *len = seg->len; 851 if (sysmem) 852 *sysmem = seg->sysmem; 853 if (objptr) 854 *objptr = seg->object; 855 return (0); 856 } 857 858 void 859 vm_free_memseg(struct vm *vm, int ident) 860 { 861 struct mem_seg *seg; 862 863 KASSERT(ident >= 0 && ident < VM_MAX_MEMSEGS, 864 ("%s: invalid memseg ident %d", __func__, ident)); 865 866 seg = &vm->mem_segs[ident]; 867 if (seg->object != NULL) { 868 vm_object_deallocate(seg->object); 869 bzero(seg, sizeof(struct mem_seg)); 870 } 871 } 872 873 int 874 vm_mmap_memseg(struct vm *vm, vm_paddr_t gpa, int segid, vm_ooffset_t first, 875 size_t len, int prot, int flags) 876 { 877 struct mem_seg *seg; 878 struct mem_map *m, *map; 879 vm_ooffset_t last; 880 int i, error; 881 882 if (prot == 0 || (prot & ~(VM_PROT_ALL)) != 0) 883 return (EINVAL); 884 885 if (flags & ~VM_MEMMAP_F_WIRED) 886 return (EINVAL); 887 888 if (segid < 0 || segid >= VM_MAX_MEMSEGS) 889 return (EINVAL); 890 891 seg = &vm->mem_segs[segid]; 892 if (seg->object == NULL) 893 return (EINVAL); 894 895 last = first + len; 896 if (first < 0 || first >= last || last > seg->len) 897 return (EINVAL); 898 899 if ((gpa | first | last) & PAGE_MASK) 900 return (EINVAL); 901 902 map = NULL; 903 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 904 m = &vm->mem_maps[i]; 905 if (m->len == 0) { 906 map = m; 907 break; 908 } 909 } 910 911 if (map == NULL) 912 return (ENOSPC); 913 914 error = vm_map_find(&vm->vmspace->vm_map, seg->object, first, &gpa, 915 len, 0, VMFS_NO_SPACE, prot, prot, 0); 916 if (error != KERN_SUCCESS) 917 return (EFAULT); 918 919 vm_object_reference(seg->object); 920 921 if (flags & VM_MEMMAP_F_WIRED) { 922 error = vm_map_wire(&vm->vmspace->vm_map, gpa, gpa + len, 923 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES); 924 if (error != KERN_SUCCESS) { 925 vm_map_remove(&vm->vmspace->vm_map, gpa, gpa + len); 926 return (error == KERN_RESOURCE_SHORTAGE ? ENOMEM : 927 EFAULT); 928 } 929 } 930 931 map->gpa = gpa; 932 map->len = len; 933 map->segoff = first; 934 map->segid = segid; 935 map->prot = prot; 936 map->flags = flags; 937 return (0); 938 } 939 940 int 941 vm_munmap_memseg(struct vm *vm, vm_paddr_t gpa, size_t len) 942 { 943 struct mem_map *m; 944 int i; 945 946 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 947 m = &vm->mem_maps[i]; 948 if (m->gpa == gpa && m->len == len && 949 (m->flags & VM_MEMMAP_F_IOMMU) == 0) { 950 vm_free_memmap(vm, i); 951 return (0); 952 } 953 } 954 955 return (EINVAL); 956 } 957 958 int 959 vm_mmap_getnext(struct vm *vm, vm_paddr_t *gpa, int *segid, 960 vm_ooffset_t *segoff, size_t *len, int *prot, int *flags) 961 { 962 struct mem_map *mm, *mmnext; 963 int i; 964 965 mmnext = NULL; 966 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 967 mm = &vm->mem_maps[i]; 968 if (mm->len == 0 || mm->gpa < *gpa) 969 continue; 970 if (mmnext == NULL || mm->gpa < mmnext->gpa) 971 mmnext = mm; 972 } 973 974 if (mmnext != NULL) { 975 *gpa = mmnext->gpa; 976 if (segid) 977 *segid = mmnext->segid; 978 if (segoff) 979 *segoff = mmnext->segoff; 980 if (len) 981 *len = mmnext->len; 982 if (prot) 983 *prot = mmnext->prot; 984 if (flags) 985 *flags = mmnext->flags; 986 return (0); 987 } else { 988 return (ENOENT); 989 } 990 } 991 992 static void 993 vm_free_memmap(struct vm *vm, int ident) 994 { 995 struct mem_map *mm; 996 int error __diagused; 997 998 mm = &vm->mem_maps[ident]; 999 if (mm->len) { 1000 error = vm_map_remove(&vm->vmspace->vm_map, mm->gpa, 1001 mm->gpa + mm->len); 1002 KASSERT(error == KERN_SUCCESS, ("%s: vm_map_remove error %d", 1003 __func__, error)); 1004 bzero(mm, sizeof(struct mem_map)); 1005 } 1006 } 1007 1008 static __inline bool 1009 sysmem_mapping(struct vm *vm, struct mem_map *mm) 1010 { 1011 1012 if (mm->len != 0 && vm->mem_segs[mm->segid].sysmem) 1013 return (true); 1014 else 1015 return (false); 1016 } 1017 1018 vm_paddr_t 1019 vmm_sysmem_maxaddr(struct vm *vm) 1020 { 1021 struct mem_map *mm; 1022 vm_paddr_t maxaddr; 1023 int i; 1024 1025 maxaddr = 0; 1026 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 1027 mm = &vm->mem_maps[i]; 1028 if (sysmem_mapping(vm, mm)) { 1029 if (maxaddr < mm->gpa + mm->len) 1030 maxaddr = mm->gpa + mm->len; 1031 } 1032 } 1033 return (maxaddr); 1034 } 1035 1036 static void 1037 vm_iommu_modify(struct vm *vm, bool map) 1038 { 1039 int i, sz; 1040 vm_paddr_t gpa, hpa; 1041 struct mem_map *mm; 1042 void *vp, *cookie, *host_domain; 1043 1044 sz = PAGE_SIZE; 1045 host_domain = iommu_host_domain(); 1046 1047 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 1048 mm = &vm->mem_maps[i]; 1049 if (!sysmem_mapping(vm, mm)) 1050 continue; 1051 1052 if (map) { 1053 KASSERT((mm->flags & VM_MEMMAP_F_IOMMU) == 0, 1054 ("iommu map found invalid memmap %#lx/%#lx/%#x", 1055 mm->gpa, mm->len, mm->flags)); 1056 if ((mm->flags & VM_MEMMAP_F_WIRED) == 0) 1057 continue; 1058 mm->flags |= VM_MEMMAP_F_IOMMU; 1059 } else { 1060 if ((mm->flags & VM_MEMMAP_F_IOMMU) == 0) 1061 continue; 1062 mm->flags &= ~VM_MEMMAP_F_IOMMU; 1063 KASSERT((mm->flags & VM_MEMMAP_F_WIRED) != 0, 1064 ("iommu unmap found invalid memmap %#lx/%#lx/%#x", 1065 mm->gpa, mm->len, mm->flags)); 1066 } 1067 1068 gpa = mm->gpa; 1069 while (gpa < mm->gpa + mm->len) { 1070 vp = vm_gpa_hold_global(vm, gpa, PAGE_SIZE, 1071 VM_PROT_WRITE, &cookie); 1072 KASSERT(vp != NULL, ("vm(%s) could not map gpa %#lx", 1073 vm_name(vm), gpa)); 1074 1075 vm_gpa_release(cookie); 1076 1077 hpa = DMAP_TO_PHYS((uintptr_t)vp); 1078 if (map) { 1079 iommu_create_mapping(vm->iommu, gpa, hpa, sz); 1080 } else { 1081 iommu_remove_mapping(vm->iommu, gpa, sz); 1082 } 1083 1084 gpa += PAGE_SIZE; 1085 } 1086 } 1087 1088 /* 1089 * Invalidate the cached translations associated with the domain 1090 * from which pages were removed. 1091 */ 1092 if (map) 1093 iommu_invalidate_tlb(host_domain); 1094 else 1095 iommu_invalidate_tlb(vm->iommu); 1096 } 1097 1098 #define vm_iommu_unmap(vm) vm_iommu_modify((vm), false) 1099 #define vm_iommu_map(vm) vm_iommu_modify((vm), true) 1100 1101 int 1102 vm_unassign_pptdev(struct vm *vm, int bus, int slot, int func) 1103 { 1104 int error; 1105 1106 error = ppt_unassign_device(vm, bus, slot, func); 1107 if (error) 1108 return (error); 1109 1110 if (ppt_assigned_devices(vm) == 0) 1111 vm_iommu_unmap(vm); 1112 1113 return (0); 1114 } 1115 1116 int 1117 vm_assign_pptdev(struct vm *vm, int bus, int slot, int func) 1118 { 1119 int error; 1120 vm_paddr_t maxaddr; 1121 1122 /* Set up the IOMMU to do the 'gpa' to 'hpa' translation */ 1123 if (ppt_assigned_devices(vm) == 0) { 1124 KASSERT(vm->iommu == NULL, 1125 ("vm_assign_pptdev: iommu must be NULL")); 1126 maxaddr = vmm_sysmem_maxaddr(vm); 1127 vm->iommu = iommu_create_domain(maxaddr); 1128 if (vm->iommu == NULL) 1129 return (ENXIO); 1130 vm_iommu_map(vm); 1131 } 1132 1133 error = ppt_assign_device(vm, bus, slot, func); 1134 return (error); 1135 } 1136 1137 static void * 1138 _vm_gpa_hold(struct vm *vm, vm_paddr_t gpa, size_t len, int reqprot, 1139 void **cookie) 1140 { 1141 int i, count, pageoff; 1142 struct mem_map *mm; 1143 vm_page_t m; 1144 1145 pageoff = gpa & PAGE_MASK; 1146 if (len > PAGE_SIZE - pageoff) 1147 panic("vm_gpa_hold: invalid gpa/len: 0x%016lx/%lu", gpa, len); 1148 1149 count = 0; 1150 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 1151 mm = &vm->mem_maps[i]; 1152 if (gpa >= mm->gpa && gpa < mm->gpa + mm->len) { 1153 count = vm_fault_quick_hold_pages(&vm->vmspace->vm_map, 1154 trunc_page(gpa), PAGE_SIZE, reqprot, &m, 1); 1155 break; 1156 } 1157 } 1158 1159 if (count == 1) { 1160 *cookie = m; 1161 return ((void *)(PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m)) + pageoff)); 1162 } else { 1163 *cookie = NULL; 1164 return (NULL); 1165 } 1166 } 1167 1168 void * 1169 vm_gpa_hold(struct vcpu *vcpu, vm_paddr_t gpa, size_t len, int reqprot, 1170 void **cookie) 1171 { 1172 #ifdef INVARIANTS 1173 /* 1174 * The current vcpu should be frozen to ensure 'vm_memmap[]' 1175 * stability. 1176 */ 1177 int state = vcpu_get_state(vcpu, NULL); 1178 KASSERT(state == VCPU_FROZEN, ("%s: invalid vcpu state %d", 1179 __func__, state)); 1180 #endif 1181 return (_vm_gpa_hold(vcpu->vm, gpa, len, reqprot, cookie)); 1182 } 1183 1184 void * 1185 vm_gpa_hold_global(struct vm *vm, vm_paddr_t gpa, size_t len, int reqprot, 1186 void **cookie) 1187 { 1188 sx_assert(&vm->mem_segs_lock, SX_LOCKED); 1189 return (_vm_gpa_hold(vm, gpa, len, reqprot, cookie)); 1190 } 1191 1192 void 1193 vm_gpa_release(void *cookie) 1194 { 1195 vm_page_t m = cookie; 1196 1197 vm_page_unwire(m, PQ_ACTIVE); 1198 } 1199 1200 int 1201 vm_get_register(struct vcpu *vcpu, int reg, uint64_t *retval) 1202 { 1203 1204 if (reg >= VM_REG_LAST) 1205 return (EINVAL); 1206 1207 return (vmmops_getreg(vcpu->cookie, reg, retval)); 1208 } 1209 1210 int 1211 vm_set_register(struct vcpu *vcpu, int reg, uint64_t val) 1212 { 1213 int error; 1214 1215 if (reg >= VM_REG_LAST) 1216 return (EINVAL); 1217 1218 error = vmmops_setreg(vcpu->cookie, reg, val); 1219 if (error || reg != VM_REG_GUEST_RIP) 1220 return (error); 1221 1222 /* Set 'nextrip' to match the value of %rip */ 1223 VMM_CTR1(vcpu, "Setting nextrip to %#lx", val); 1224 vcpu->nextrip = val; 1225 return (0); 1226 } 1227 1228 static bool 1229 is_descriptor_table(int reg) 1230 { 1231 1232 switch (reg) { 1233 case VM_REG_GUEST_IDTR: 1234 case VM_REG_GUEST_GDTR: 1235 return (true); 1236 default: 1237 return (false); 1238 } 1239 } 1240 1241 static bool 1242 is_segment_register(int reg) 1243 { 1244 1245 switch (reg) { 1246 case VM_REG_GUEST_ES: 1247 case VM_REG_GUEST_CS: 1248 case VM_REG_GUEST_SS: 1249 case VM_REG_GUEST_DS: 1250 case VM_REG_GUEST_FS: 1251 case VM_REG_GUEST_GS: 1252 case VM_REG_GUEST_TR: 1253 case VM_REG_GUEST_LDTR: 1254 return (true); 1255 default: 1256 return (false); 1257 } 1258 } 1259 1260 int 1261 vm_get_seg_desc(struct vcpu *vcpu, int reg, struct seg_desc *desc) 1262 { 1263 1264 if (!is_segment_register(reg) && !is_descriptor_table(reg)) 1265 return (EINVAL); 1266 1267 return (vmmops_getdesc(vcpu->cookie, reg, desc)); 1268 } 1269 1270 int 1271 vm_set_seg_desc(struct vcpu *vcpu, int reg, struct seg_desc *desc) 1272 { 1273 1274 if (!is_segment_register(reg) && !is_descriptor_table(reg)) 1275 return (EINVAL); 1276 1277 return (vmmops_setdesc(vcpu->cookie, reg, desc)); 1278 } 1279 1280 static void 1281 restore_guest_fpustate(struct vcpu *vcpu) 1282 { 1283 1284 /* flush host state to the pcb */ 1285 fpuexit(curthread); 1286 1287 /* restore guest FPU state */ 1288 fpu_stop_emulating(); 1289 fpurestore(vcpu->guestfpu); 1290 1291 /* restore guest XCR0 if XSAVE is enabled in the host */ 1292 if (rcr4() & CR4_XSAVE) 1293 load_xcr(0, vcpu->guest_xcr0); 1294 1295 /* 1296 * The FPU is now "dirty" with the guest's state so turn on emulation 1297 * to trap any access to the FPU by the host. 1298 */ 1299 fpu_start_emulating(); 1300 } 1301 1302 static void 1303 save_guest_fpustate(struct vcpu *vcpu) 1304 { 1305 1306 if ((rcr0() & CR0_TS) == 0) 1307 panic("fpu emulation not enabled in host!"); 1308 1309 /* save guest XCR0 and restore host XCR0 */ 1310 if (rcr4() & CR4_XSAVE) { 1311 vcpu->guest_xcr0 = rxcr(0); 1312 load_xcr(0, vmm_get_host_xcr0()); 1313 } 1314 1315 /* save guest FPU state */ 1316 fpu_stop_emulating(); 1317 fpusave(vcpu->guestfpu); 1318 fpu_start_emulating(); 1319 } 1320 1321 static VMM_STAT(VCPU_IDLE_TICKS, "number of ticks vcpu was idle"); 1322 1323 static int 1324 vcpu_set_state_locked(struct vcpu *vcpu, enum vcpu_state newstate, 1325 bool from_idle) 1326 { 1327 int error; 1328 1329 vcpu_assert_locked(vcpu); 1330 1331 /* 1332 * State transitions from the vmmdev_ioctl() must always begin from 1333 * the VCPU_IDLE state. This guarantees that there is only a single 1334 * ioctl() operating on a vcpu at any point. 1335 */ 1336 if (from_idle) { 1337 while (vcpu->state != VCPU_IDLE) { 1338 vcpu->reqidle = 1; 1339 vcpu_notify_event_locked(vcpu, false); 1340 VMM_CTR1(vcpu, "vcpu state change from %s to " 1341 "idle requested", vcpu_state2str(vcpu->state)); 1342 msleep_spin(&vcpu->state, &vcpu->mtx, "vmstat", hz); 1343 } 1344 } else { 1345 KASSERT(vcpu->state != VCPU_IDLE, ("invalid transition from " 1346 "vcpu idle state")); 1347 } 1348 1349 if (vcpu->state == VCPU_RUNNING) { 1350 KASSERT(vcpu->hostcpu == curcpu, ("curcpu %d and hostcpu %d " 1351 "mismatch for running vcpu", curcpu, vcpu->hostcpu)); 1352 } else { 1353 KASSERT(vcpu->hostcpu == NOCPU, ("Invalid hostcpu %d for a " 1354 "vcpu that is not running", vcpu->hostcpu)); 1355 } 1356 1357 /* 1358 * The following state transitions are allowed: 1359 * IDLE -> FROZEN -> IDLE 1360 * FROZEN -> RUNNING -> FROZEN 1361 * FROZEN -> SLEEPING -> FROZEN 1362 */ 1363 switch (vcpu->state) { 1364 case VCPU_IDLE: 1365 case VCPU_RUNNING: 1366 case VCPU_SLEEPING: 1367 error = (newstate != VCPU_FROZEN); 1368 break; 1369 case VCPU_FROZEN: 1370 error = (newstate == VCPU_FROZEN); 1371 break; 1372 default: 1373 error = 1; 1374 break; 1375 } 1376 1377 if (error) 1378 return (EBUSY); 1379 1380 VMM_CTR2(vcpu, "vcpu state changed from %s to %s", 1381 vcpu_state2str(vcpu->state), vcpu_state2str(newstate)); 1382 1383 vcpu->state = newstate; 1384 if (newstate == VCPU_RUNNING) 1385 vcpu->hostcpu = curcpu; 1386 else 1387 vcpu->hostcpu = NOCPU; 1388 1389 if (newstate == VCPU_IDLE) 1390 wakeup(&vcpu->state); 1391 1392 return (0); 1393 } 1394 1395 static void 1396 vcpu_require_state(struct vcpu *vcpu, enum vcpu_state newstate) 1397 { 1398 int error; 1399 1400 if ((error = vcpu_set_state(vcpu, newstate, false)) != 0) 1401 panic("Error %d setting state to %d\n", error, newstate); 1402 } 1403 1404 static void 1405 vcpu_require_state_locked(struct vcpu *vcpu, enum vcpu_state newstate) 1406 { 1407 int error; 1408 1409 if ((error = vcpu_set_state_locked(vcpu, newstate, false)) != 0) 1410 panic("Error %d setting state to %d", error, newstate); 1411 } 1412 1413 static int 1414 vm_handle_rendezvous(struct vcpu *vcpu) 1415 { 1416 struct vm *vm = vcpu->vm; 1417 struct thread *td; 1418 int error, vcpuid; 1419 1420 error = 0; 1421 vcpuid = vcpu->vcpuid; 1422 td = curthread; 1423 mtx_lock(&vm->rendezvous_mtx); 1424 while (vm->rendezvous_func != NULL) { 1425 /* 'rendezvous_req_cpus' must be a subset of 'active_cpus' */ 1426 CPU_AND(&vm->rendezvous_req_cpus, &vm->rendezvous_req_cpus, &vm->active_cpus); 1427 1428 if (CPU_ISSET(vcpuid, &vm->rendezvous_req_cpus) && 1429 !CPU_ISSET(vcpuid, &vm->rendezvous_done_cpus)) { 1430 VMM_CTR0(vcpu, "Calling rendezvous func"); 1431 (*vm->rendezvous_func)(vcpu, vm->rendezvous_arg); 1432 CPU_SET(vcpuid, &vm->rendezvous_done_cpus); 1433 } 1434 if (CPU_CMP(&vm->rendezvous_req_cpus, 1435 &vm->rendezvous_done_cpus) == 0) { 1436 VMM_CTR0(vcpu, "Rendezvous completed"); 1437 vm->rendezvous_func = NULL; 1438 wakeup(&vm->rendezvous_func); 1439 break; 1440 } 1441 VMM_CTR0(vcpu, "Wait for rendezvous completion"); 1442 mtx_sleep(&vm->rendezvous_func, &vm->rendezvous_mtx, 0, 1443 "vmrndv", hz); 1444 if (td_ast_pending(td, TDA_SUSPEND)) { 1445 mtx_unlock(&vm->rendezvous_mtx); 1446 error = thread_check_susp(td, true); 1447 if (error != 0) 1448 return (error); 1449 mtx_lock(&vm->rendezvous_mtx); 1450 } 1451 } 1452 mtx_unlock(&vm->rendezvous_mtx); 1453 return (0); 1454 } 1455 1456 /* 1457 * Emulate a guest 'hlt' by sleeping until the vcpu is ready to run. 1458 */ 1459 static int 1460 vm_handle_hlt(struct vcpu *vcpu, bool intr_disabled, bool *retu) 1461 { 1462 struct vm *vm = vcpu->vm; 1463 const char *wmesg; 1464 struct thread *td; 1465 int error, t, vcpuid, vcpu_halted, vm_halted; 1466 1467 vcpuid = vcpu->vcpuid; 1468 vcpu_halted = 0; 1469 vm_halted = 0; 1470 error = 0; 1471 td = curthread; 1472 1473 KASSERT(!CPU_ISSET(vcpuid, &vm->halted_cpus), ("vcpu already halted")); 1474 1475 vcpu_lock(vcpu); 1476 while (1) { 1477 /* 1478 * Do a final check for pending NMI or interrupts before 1479 * really putting this thread to sleep. Also check for 1480 * software events that would cause this vcpu to wakeup. 1481 * 1482 * These interrupts/events could have happened after the 1483 * vcpu returned from vmmops_run() and before it acquired the 1484 * vcpu lock above. 1485 */ 1486 if (vm->rendezvous_func != NULL || vm->suspend || vcpu->reqidle) 1487 break; 1488 if (vm_nmi_pending(vcpu)) 1489 break; 1490 if (!intr_disabled) { 1491 if (vm_extint_pending(vcpu) || 1492 vlapic_pending_intr(vcpu->vlapic, NULL)) { 1493 break; 1494 } 1495 } 1496 1497 /* Don't go to sleep if the vcpu thread needs to yield */ 1498 if (vcpu_should_yield(vcpu)) 1499 break; 1500 1501 if (vcpu_debugged(vcpu)) 1502 break; 1503 1504 /* 1505 * Some Linux guests implement "halt" by having all vcpus 1506 * execute HLT with interrupts disabled. 'halted_cpus' keeps 1507 * track of the vcpus that have entered this state. When all 1508 * vcpus enter the halted state the virtual machine is halted. 1509 */ 1510 if (intr_disabled) { 1511 wmesg = "vmhalt"; 1512 VMM_CTR0(vcpu, "Halted"); 1513 if (!vcpu_halted && halt_detection_enabled) { 1514 vcpu_halted = 1; 1515 CPU_SET_ATOMIC(vcpuid, &vm->halted_cpus); 1516 } 1517 if (CPU_CMP(&vm->halted_cpus, &vm->active_cpus) == 0) { 1518 vm_halted = 1; 1519 break; 1520 } 1521 } else { 1522 wmesg = "vmidle"; 1523 } 1524 1525 t = ticks; 1526 vcpu_require_state_locked(vcpu, VCPU_SLEEPING); 1527 /* 1528 * XXX msleep_spin() cannot be interrupted by signals so 1529 * wake up periodically to check pending signals. 1530 */ 1531 msleep_spin(vcpu, &vcpu->mtx, wmesg, hz); 1532 vcpu_require_state_locked(vcpu, VCPU_FROZEN); 1533 vmm_stat_incr(vcpu, VCPU_IDLE_TICKS, ticks - t); 1534 if (td_ast_pending(td, TDA_SUSPEND)) { 1535 vcpu_unlock(vcpu); 1536 error = thread_check_susp(td, false); 1537 if (error != 0) { 1538 if (vcpu_halted) { 1539 CPU_CLR_ATOMIC(vcpuid, 1540 &vm->halted_cpus); 1541 } 1542 return (error); 1543 } 1544 vcpu_lock(vcpu); 1545 } 1546 } 1547 1548 if (vcpu_halted) 1549 CPU_CLR_ATOMIC(vcpuid, &vm->halted_cpus); 1550 1551 vcpu_unlock(vcpu); 1552 1553 if (vm_halted) 1554 vm_suspend(vm, VM_SUSPEND_HALT); 1555 1556 return (0); 1557 } 1558 1559 static int 1560 vm_handle_paging(struct vcpu *vcpu, bool *retu) 1561 { 1562 struct vm *vm = vcpu->vm; 1563 int rv, ftype; 1564 struct vm_map *map; 1565 struct vm_exit *vme; 1566 1567 vme = &vcpu->exitinfo; 1568 1569 KASSERT(vme->inst_length == 0, ("%s: invalid inst_length %d", 1570 __func__, vme->inst_length)); 1571 1572 ftype = vme->u.paging.fault_type; 1573 KASSERT(ftype == VM_PROT_READ || 1574 ftype == VM_PROT_WRITE || ftype == VM_PROT_EXECUTE, 1575 ("vm_handle_paging: invalid fault_type %d", ftype)); 1576 1577 if (ftype == VM_PROT_READ || ftype == VM_PROT_WRITE) { 1578 rv = pmap_emulate_accessed_dirty(vmspace_pmap(vm->vmspace), 1579 vme->u.paging.gpa, ftype); 1580 if (rv == 0) { 1581 VMM_CTR2(vcpu, "%s bit emulation for gpa %#lx", 1582 ftype == VM_PROT_READ ? "accessed" : "dirty", 1583 vme->u.paging.gpa); 1584 goto done; 1585 } 1586 } 1587 1588 map = &vm->vmspace->vm_map; 1589 rv = vm_fault(map, vme->u.paging.gpa, ftype, VM_FAULT_NORMAL, NULL); 1590 1591 VMM_CTR3(vcpu, "vm_handle_paging rv = %d, gpa = %#lx, " 1592 "ftype = %d", rv, vme->u.paging.gpa, ftype); 1593 1594 if (rv != KERN_SUCCESS) 1595 return (EFAULT); 1596 done: 1597 return (0); 1598 } 1599 1600 static int 1601 vm_handle_inst_emul(struct vcpu *vcpu, bool *retu) 1602 { 1603 struct vie *vie; 1604 struct vm_exit *vme; 1605 uint64_t gla, gpa, cs_base; 1606 struct vm_guest_paging *paging; 1607 mem_region_read_t mread; 1608 mem_region_write_t mwrite; 1609 enum vm_cpu_mode cpu_mode; 1610 int cs_d, error, fault; 1611 1612 vme = &vcpu->exitinfo; 1613 1614 KASSERT(vme->inst_length == 0, ("%s: invalid inst_length %d", 1615 __func__, vme->inst_length)); 1616 1617 gla = vme->u.inst_emul.gla; 1618 gpa = vme->u.inst_emul.gpa; 1619 cs_base = vme->u.inst_emul.cs_base; 1620 cs_d = vme->u.inst_emul.cs_d; 1621 vie = &vme->u.inst_emul.vie; 1622 paging = &vme->u.inst_emul.paging; 1623 cpu_mode = paging->cpu_mode; 1624 1625 VMM_CTR1(vcpu, "inst_emul fault accessing gpa %#lx", gpa); 1626 1627 /* Fetch, decode and emulate the faulting instruction */ 1628 if (vie->num_valid == 0) { 1629 error = vmm_fetch_instruction(vcpu, paging, vme->rip + cs_base, 1630 VIE_INST_SIZE, vie, &fault); 1631 } else { 1632 /* 1633 * The instruction bytes have already been copied into 'vie' 1634 */ 1635 error = fault = 0; 1636 } 1637 if (error || fault) 1638 return (error); 1639 1640 if (vmm_decode_instruction(vcpu, gla, cpu_mode, cs_d, vie) != 0) { 1641 VMM_CTR1(vcpu, "Error decoding instruction at %#lx", 1642 vme->rip + cs_base); 1643 *retu = true; /* dump instruction bytes in userspace */ 1644 return (0); 1645 } 1646 1647 /* 1648 * Update 'nextrip' based on the length of the emulated instruction. 1649 */ 1650 vme->inst_length = vie->num_processed; 1651 vcpu->nextrip += vie->num_processed; 1652 VMM_CTR1(vcpu, "nextrip updated to %#lx after instruction decoding", 1653 vcpu->nextrip); 1654 1655 /* return to userland unless this is an in-kernel emulated device */ 1656 if (gpa >= DEFAULT_APIC_BASE && gpa < DEFAULT_APIC_BASE + PAGE_SIZE) { 1657 mread = lapic_mmio_read; 1658 mwrite = lapic_mmio_write; 1659 } else if (gpa >= VIOAPIC_BASE && gpa < VIOAPIC_BASE + VIOAPIC_SIZE) { 1660 mread = vioapic_mmio_read; 1661 mwrite = vioapic_mmio_write; 1662 } else if (gpa >= VHPET_BASE && gpa < VHPET_BASE + VHPET_SIZE) { 1663 mread = vhpet_mmio_read; 1664 mwrite = vhpet_mmio_write; 1665 } else { 1666 *retu = true; 1667 return (0); 1668 } 1669 1670 error = vmm_emulate_instruction(vcpu, gpa, vie, paging, mread, mwrite, 1671 retu); 1672 1673 return (error); 1674 } 1675 1676 static int 1677 vm_handle_suspend(struct vcpu *vcpu, bool *retu) 1678 { 1679 struct vm *vm = vcpu->vm; 1680 int error, i; 1681 struct thread *td; 1682 1683 error = 0; 1684 td = curthread; 1685 1686 CPU_SET_ATOMIC(vcpu->vcpuid, &vm->suspended_cpus); 1687 1688 /* 1689 * Wait until all 'active_cpus' have suspended themselves. 1690 * 1691 * Since a VM may be suspended at any time including when one or 1692 * more vcpus are doing a rendezvous we need to call the rendezvous 1693 * handler while we are waiting to prevent a deadlock. 1694 */ 1695 vcpu_lock(vcpu); 1696 while (error == 0) { 1697 if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) { 1698 VMM_CTR0(vcpu, "All vcpus suspended"); 1699 break; 1700 } 1701 1702 if (vm->rendezvous_func == NULL) { 1703 VMM_CTR0(vcpu, "Sleeping during suspend"); 1704 vcpu_require_state_locked(vcpu, VCPU_SLEEPING); 1705 msleep_spin(vcpu, &vcpu->mtx, "vmsusp", hz); 1706 vcpu_require_state_locked(vcpu, VCPU_FROZEN); 1707 if (td_ast_pending(td, TDA_SUSPEND)) { 1708 vcpu_unlock(vcpu); 1709 error = thread_check_susp(td, false); 1710 vcpu_lock(vcpu); 1711 } 1712 } else { 1713 VMM_CTR0(vcpu, "Rendezvous during suspend"); 1714 vcpu_unlock(vcpu); 1715 error = vm_handle_rendezvous(vcpu); 1716 vcpu_lock(vcpu); 1717 } 1718 } 1719 vcpu_unlock(vcpu); 1720 1721 /* 1722 * Wakeup the other sleeping vcpus and return to userspace. 1723 */ 1724 for (i = 0; i < vm->maxcpus; i++) { 1725 if (CPU_ISSET(i, &vm->suspended_cpus)) { 1726 vcpu_notify_event(vm_vcpu(vm, i), false); 1727 } 1728 } 1729 1730 *retu = true; 1731 return (error); 1732 } 1733 1734 static int 1735 vm_handle_reqidle(struct vcpu *vcpu, bool *retu) 1736 { 1737 vcpu_lock(vcpu); 1738 KASSERT(vcpu->reqidle, ("invalid vcpu reqidle %d", vcpu->reqidle)); 1739 vcpu->reqidle = 0; 1740 vcpu_unlock(vcpu); 1741 *retu = true; 1742 return (0); 1743 } 1744 1745 int 1746 vm_suspend(struct vm *vm, enum vm_suspend_how how) 1747 { 1748 int i; 1749 1750 if (how <= VM_SUSPEND_NONE || how >= VM_SUSPEND_LAST) 1751 return (EINVAL); 1752 1753 if (atomic_cmpset_int(&vm->suspend, 0, how) == 0) { 1754 VM_CTR2(vm, "virtual machine already suspended %d/%d", 1755 vm->suspend, how); 1756 return (EALREADY); 1757 } 1758 1759 VM_CTR1(vm, "virtual machine successfully suspended %d", how); 1760 1761 /* 1762 * Notify all active vcpus that they are now suspended. 1763 */ 1764 for (i = 0; i < vm->maxcpus; i++) { 1765 if (CPU_ISSET(i, &vm->active_cpus)) 1766 vcpu_notify_event(vm_vcpu(vm, i), false); 1767 } 1768 1769 return (0); 1770 } 1771 1772 void 1773 vm_exit_suspended(struct vcpu *vcpu, uint64_t rip) 1774 { 1775 struct vm *vm = vcpu->vm; 1776 struct vm_exit *vmexit; 1777 1778 KASSERT(vm->suspend > VM_SUSPEND_NONE && vm->suspend < VM_SUSPEND_LAST, 1779 ("vm_exit_suspended: invalid suspend type %d", vm->suspend)); 1780 1781 vmexit = vm_exitinfo(vcpu); 1782 vmexit->rip = rip; 1783 vmexit->inst_length = 0; 1784 vmexit->exitcode = VM_EXITCODE_SUSPENDED; 1785 vmexit->u.suspended.how = vm->suspend; 1786 } 1787 1788 void 1789 vm_exit_debug(struct vcpu *vcpu, uint64_t rip) 1790 { 1791 struct vm_exit *vmexit; 1792 1793 vmexit = vm_exitinfo(vcpu); 1794 vmexit->rip = rip; 1795 vmexit->inst_length = 0; 1796 vmexit->exitcode = VM_EXITCODE_DEBUG; 1797 } 1798 1799 void 1800 vm_exit_rendezvous(struct vcpu *vcpu, uint64_t rip) 1801 { 1802 struct vm_exit *vmexit; 1803 1804 vmexit = vm_exitinfo(vcpu); 1805 vmexit->rip = rip; 1806 vmexit->inst_length = 0; 1807 vmexit->exitcode = VM_EXITCODE_RENDEZVOUS; 1808 vmm_stat_incr(vcpu, VMEXIT_RENDEZVOUS, 1); 1809 } 1810 1811 void 1812 vm_exit_reqidle(struct vcpu *vcpu, uint64_t rip) 1813 { 1814 struct vm_exit *vmexit; 1815 1816 vmexit = vm_exitinfo(vcpu); 1817 vmexit->rip = rip; 1818 vmexit->inst_length = 0; 1819 vmexit->exitcode = VM_EXITCODE_REQIDLE; 1820 vmm_stat_incr(vcpu, VMEXIT_REQIDLE, 1); 1821 } 1822 1823 void 1824 vm_exit_astpending(struct vcpu *vcpu, uint64_t rip) 1825 { 1826 struct vm_exit *vmexit; 1827 1828 vmexit = vm_exitinfo(vcpu); 1829 vmexit->rip = rip; 1830 vmexit->inst_length = 0; 1831 vmexit->exitcode = VM_EXITCODE_BOGUS; 1832 vmm_stat_incr(vcpu, VMEXIT_ASTPENDING, 1); 1833 } 1834 1835 int 1836 vm_run(struct vcpu *vcpu, struct vm_exit *vme_user) 1837 { 1838 struct vm *vm = vcpu->vm; 1839 struct vm_eventinfo evinfo; 1840 int error, vcpuid; 1841 struct pcb *pcb; 1842 uint64_t tscval; 1843 struct vm_exit *vme; 1844 bool retu, intr_disabled; 1845 pmap_t pmap; 1846 1847 vcpuid = vcpu->vcpuid; 1848 1849 if (!CPU_ISSET(vcpuid, &vm->active_cpus)) 1850 return (EINVAL); 1851 1852 if (CPU_ISSET(vcpuid, &vm->suspended_cpus)) 1853 return (EINVAL); 1854 1855 pmap = vmspace_pmap(vm->vmspace); 1856 vme = &vcpu->exitinfo; 1857 evinfo.rptr = &vm->rendezvous_func; 1858 evinfo.sptr = &vm->suspend; 1859 evinfo.iptr = &vcpu->reqidle; 1860 restart: 1861 critical_enter(); 1862 1863 KASSERT(!CPU_ISSET(curcpu, &pmap->pm_active), 1864 ("vm_run: absurd pm_active")); 1865 1866 tscval = rdtsc(); 1867 1868 pcb = PCPU_GET(curpcb); 1869 set_pcb_flags(pcb, PCB_FULL_IRET); 1870 1871 restore_guest_fpustate(vcpu); 1872 1873 vcpu_require_state(vcpu, VCPU_RUNNING); 1874 error = vmmops_run(vcpu->cookie, vcpu->nextrip, pmap, &evinfo); 1875 vcpu_require_state(vcpu, VCPU_FROZEN); 1876 1877 save_guest_fpustate(vcpu); 1878 1879 vmm_stat_incr(vcpu, VCPU_TOTAL_RUNTIME, rdtsc() - tscval); 1880 1881 critical_exit(); 1882 1883 if (error == 0) { 1884 retu = false; 1885 vcpu->nextrip = vme->rip + vme->inst_length; 1886 switch (vme->exitcode) { 1887 case VM_EXITCODE_REQIDLE: 1888 error = vm_handle_reqidle(vcpu, &retu); 1889 break; 1890 case VM_EXITCODE_SUSPENDED: 1891 error = vm_handle_suspend(vcpu, &retu); 1892 break; 1893 case VM_EXITCODE_IOAPIC_EOI: 1894 vioapic_process_eoi(vm, vme->u.ioapic_eoi.vector); 1895 break; 1896 case VM_EXITCODE_RENDEZVOUS: 1897 error = vm_handle_rendezvous(vcpu); 1898 break; 1899 case VM_EXITCODE_HLT: 1900 intr_disabled = ((vme->u.hlt.rflags & PSL_I) == 0); 1901 error = vm_handle_hlt(vcpu, intr_disabled, &retu); 1902 break; 1903 case VM_EXITCODE_PAGING: 1904 error = vm_handle_paging(vcpu, &retu); 1905 break; 1906 case VM_EXITCODE_INST_EMUL: 1907 error = vm_handle_inst_emul(vcpu, &retu); 1908 break; 1909 case VM_EXITCODE_INOUT: 1910 case VM_EXITCODE_INOUT_STR: 1911 error = vm_handle_inout(vcpu, vme, &retu); 1912 break; 1913 case VM_EXITCODE_MONITOR: 1914 case VM_EXITCODE_MWAIT: 1915 case VM_EXITCODE_VMINSN: 1916 vm_inject_ud(vcpu); 1917 break; 1918 default: 1919 retu = true; /* handled in userland */ 1920 break; 1921 } 1922 } 1923 1924 /* 1925 * VM_EXITCODE_INST_EMUL could access the apic which could transform the 1926 * exit code into VM_EXITCODE_IPI. 1927 */ 1928 if (error == 0 && vme->exitcode == VM_EXITCODE_IPI) { 1929 retu = false; 1930 error = vm_handle_ipi(vcpu, vme, &retu); 1931 } 1932 1933 if (error == 0 && retu == false) 1934 goto restart; 1935 1936 vmm_stat_incr(vcpu, VMEXIT_USERSPACE, 1); 1937 VMM_CTR2(vcpu, "retu %d/%d", error, vme->exitcode); 1938 1939 /* copy the exit information */ 1940 *vme_user = *vme; 1941 return (error); 1942 } 1943 1944 int 1945 vm_restart_instruction(struct vcpu *vcpu) 1946 { 1947 enum vcpu_state state; 1948 uint64_t rip; 1949 int error __diagused; 1950 1951 state = vcpu_get_state(vcpu, NULL); 1952 if (state == VCPU_RUNNING) { 1953 /* 1954 * When a vcpu is "running" the next instruction is determined 1955 * by adding 'rip' and 'inst_length' in the vcpu's 'exitinfo'. 1956 * Thus setting 'inst_length' to zero will cause the current 1957 * instruction to be restarted. 1958 */ 1959 vcpu->exitinfo.inst_length = 0; 1960 VMM_CTR1(vcpu, "restarting instruction at %#lx by " 1961 "setting inst_length to zero", vcpu->exitinfo.rip); 1962 } else if (state == VCPU_FROZEN) { 1963 /* 1964 * When a vcpu is "frozen" it is outside the critical section 1965 * around vmmops_run() and 'nextrip' points to the next 1966 * instruction. Thus instruction restart is achieved by setting 1967 * 'nextrip' to the vcpu's %rip. 1968 */ 1969 error = vm_get_register(vcpu, VM_REG_GUEST_RIP, &rip); 1970 KASSERT(!error, ("%s: error %d getting rip", __func__, error)); 1971 VMM_CTR2(vcpu, "restarting instruction by updating " 1972 "nextrip from %#lx to %#lx", vcpu->nextrip, rip); 1973 vcpu->nextrip = rip; 1974 } else { 1975 panic("%s: invalid state %d", __func__, state); 1976 } 1977 return (0); 1978 } 1979 1980 int 1981 vm_exit_intinfo(struct vcpu *vcpu, uint64_t info) 1982 { 1983 int type, vector; 1984 1985 if (info & VM_INTINFO_VALID) { 1986 type = info & VM_INTINFO_TYPE; 1987 vector = info & 0xff; 1988 if (type == VM_INTINFO_NMI && vector != IDT_NMI) 1989 return (EINVAL); 1990 if (type == VM_INTINFO_HWEXCEPTION && vector >= 32) 1991 return (EINVAL); 1992 if (info & VM_INTINFO_RSVD) 1993 return (EINVAL); 1994 } else { 1995 info = 0; 1996 } 1997 VMM_CTR2(vcpu, "%s: info1(%#lx)", __func__, info); 1998 vcpu->exitintinfo = info; 1999 return (0); 2000 } 2001 2002 enum exc_class { 2003 EXC_BENIGN, 2004 EXC_CONTRIBUTORY, 2005 EXC_PAGEFAULT 2006 }; 2007 2008 #define IDT_VE 20 /* Virtualization Exception (Intel specific) */ 2009 2010 static enum exc_class 2011 exception_class(uint64_t info) 2012 { 2013 int type, vector; 2014 2015 KASSERT(info & VM_INTINFO_VALID, ("intinfo must be valid: %#lx", info)); 2016 type = info & VM_INTINFO_TYPE; 2017 vector = info & 0xff; 2018 2019 /* Table 6-4, "Interrupt and Exception Classes", Intel SDM, Vol 3 */ 2020 switch (type) { 2021 case VM_INTINFO_HWINTR: 2022 case VM_INTINFO_SWINTR: 2023 case VM_INTINFO_NMI: 2024 return (EXC_BENIGN); 2025 default: 2026 /* 2027 * Hardware exception. 2028 * 2029 * SVM and VT-x use identical type values to represent NMI, 2030 * hardware interrupt and software interrupt. 2031 * 2032 * SVM uses type '3' for all exceptions. VT-x uses type '3' 2033 * for exceptions except #BP and #OF. #BP and #OF use a type 2034 * value of '5' or '6'. Therefore we don't check for explicit 2035 * values of 'type' to classify 'intinfo' into a hardware 2036 * exception. 2037 */ 2038 break; 2039 } 2040 2041 switch (vector) { 2042 case IDT_PF: 2043 case IDT_VE: 2044 return (EXC_PAGEFAULT); 2045 case IDT_DE: 2046 case IDT_TS: 2047 case IDT_NP: 2048 case IDT_SS: 2049 case IDT_GP: 2050 return (EXC_CONTRIBUTORY); 2051 default: 2052 return (EXC_BENIGN); 2053 } 2054 } 2055 2056 static int 2057 nested_fault(struct vcpu *vcpu, uint64_t info1, uint64_t info2, 2058 uint64_t *retinfo) 2059 { 2060 enum exc_class exc1, exc2; 2061 int type1, vector1; 2062 2063 KASSERT(info1 & VM_INTINFO_VALID, ("info1 %#lx is not valid", info1)); 2064 KASSERT(info2 & VM_INTINFO_VALID, ("info2 %#lx is not valid", info2)); 2065 2066 /* 2067 * If an exception occurs while attempting to call the double-fault 2068 * handler the processor enters shutdown mode (aka triple fault). 2069 */ 2070 type1 = info1 & VM_INTINFO_TYPE; 2071 vector1 = info1 & 0xff; 2072 if (type1 == VM_INTINFO_HWEXCEPTION && vector1 == IDT_DF) { 2073 VMM_CTR2(vcpu, "triple fault: info1(%#lx), info2(%#lx)", 2074 info1, info2); 2075 vm_suspend(vcpu->vm, VM_SUSPEND_TRIPLEFAULT); 2076 *retinfo = 0; 2077 return (0); 2078 } 2079 2080 /* 2081 * Table 6-5 "Conditions for Generating a Double Fault", Intel SDM, Vol3 2082 */ 2083 exc1 = exception_class(info1); 2084 exc2 = exception_class(info2); 2085 if ((exc1 == EXC_CONTRIBUTORY && exc2 == EXC_CONTRIBUTORY) || 2086 (exc1 == EXC_PAGEFAULT && exc2 != EXC_BENIGN)) { 2087 /* Convert nested fault into a double fault. */ 2088 *retinfo = IDT_DF; 2089 *retinfo |= VM_INTINFO_VALID | VM_INTINFO_HWEXCEPTION; 2090 *retinfo |= VM_INTINFO_DEL_ERRCODE; 2091 } else { 2092 /* Handle exceptions serially */ 2093 *retinfo = info2; 2094 } 2095 return (1); 2096 } 2097 2098 static uint64_t 2099 vcpu_exception_intinfo(struct vcpu *vcpu) 2100 { 2101 uint64_t info = 0; 2102 2103 if (vcpu->exception_pending) { 2104 info = vcpu->exc_vector & 0xff; 2105 info |= VM_INTINFO_VALID | VM_INTINFO_HWEXCEPTION; 2106 if (vcpu->exc_errcode_valid) { 2107 info |= VM_INTINFO_DEL_ERRCODE; 2108 info |= (uint64_t)vcpu->exc_errcode << 32; 2109 } 2110 } 2111 return (info); 2112 } 2113 2114 int 2115 vm_entry_intinfo(struct vcpu *vcpu, uint64_t *retinfo) 2116 { 2117 uint64_t info1, info2; 2118 int valid; 2119 2120 info1 = vcpu->exitintinfo; 2121 vcpu->exitintinfo = 0; 2122 2123 info2 = 0; 2124 if (vcpu->exception_pending) { 2125 info2 = vcpu_exception_intinfo(vcpu); 2126 vcpu->exception_pending = 0; 2127 VMM_CTR2(vcpu, "Exception %d delivered: %#lx", 2128 vcpu->exc_vector, info2); 2129 } 2130 2131 if ((info1 & VM_INTINFO_VALID) && (info2 & VM_INTINFO_VALID)) { 2132 valid = nested_fault(vcpu, info1, info2, retinfo); 2133 } else if (info1 & VM_INTINFO_VALID) { 2134 *retinfo = info1; 2135 valid = 1; 2136 } else if (info2 & VM_INTINFO_VALID) { 2137 *retinfo = info2; 2138 valid = 1; 2139 } else { 2140 valid = 0; 2141 } 2142 2143 if (valid) { 2144 VMM_CTR4(vcpu, "%s: info1(%#lx), info2(%#lx), " 2145 "retinfo(%#lx)", __func__, info1, info2, *retinfo); 2146 } 2147 2148 return (valid); 2149 } 2150 2151 int 2152 vm_get_intinfo(struct vcpu *vcpu, uint64_t *info1, uint64_t *info2) 2153 { 2154 *info1 = vcpu->exitintinfo; 2155 *info2 = vcpu_exception_intinfo(vcpu); 2156 return (0); 2157 } 2158 2159 int 2160 vm_inject_exception(struct vcpu *vcpu, int vector, int errcode_valid, 2161 uint32_t errcode, int restart_instruction) 2162 { 2163 uint64_t regval; 2164 int error __diagused; 2165 2166 if (vector < 0 || vector >= 32) 2167 return (EINVAL); 2168 2169 /* 2170 * A double fault exception should never be injected directly into 2171 * the guest. It is a derived exception that results from specific 2172 * combinations of nested faults. 2173 */ 2174 if (vector == IDT_DF) 2175 return (EINVAL); 2176 2177 if (vcpu->exception_pending) { 2178 VMM_CTR2(vcpu, "Unable to inject exception %d due to " 2179 "pending exception %d", vector, vcpu->exc_vector); 2180 return (EBUSY); 2181 } 2182 2183 if (errcode_valid) { 2184 /* 2185 * Exceptions don't deliver an error code in real mode. 2186 */ 2187 error = vm_get_register(vcpu, VM_REG_GUEST_CR0, ®val); 2188 KASSERT(!error, ("%s: error %d getting CR0", __func__, error)); 2189 if (!(regval & CR0_PE)) 2190 errcode_valid = 0; 2191 } 2192 2193 /* 2194 * From section 26.6.1 "Interruptibility State" in Intel SDM: 2195 * 2196 * Event blocking by "STI" or "MOV SS" is cleared after guest executes 2197 * one instruction or incurs an exception. 2198 */ 2199 error = vm_set_register(vcpu, VM_REG_GUEST_INTR_SHADOW, 0); 2200 KASSERT(error == 0, ("%s: error %d clearing interrupt shadow", 2201 __func__, error)); 2202 2203 if (restart_instruction) 2204 vm_restart_instruction(vcpu); 2205 2206 vcpu->exception_pending = 1; 2207 vcpu->exc_vector = vector; 2208 vcpu->exc_errcode = errcode; 2209 vcpu->exc_errcode_valid = errcode_valid; 2210 VMM_CTR1(vcpu, "Exception %d pending", vector); 2211 return (0); 2212 } 2213 2214 void 2215 vm_inject_fault(struct vcpu *vcpu, int vector, int errcode_valid, int errcode) 2216 { 2217 int error __diagused, restart_instruction; 2218 2219 restart_instruction = 1; 2220 2221 error = vm_inject_exception(vcpu, vector, errcode_valid, 2222 errcode, restart_instruction); 2223 KASSERT(error == 0, ("vm_inject_exception error %d", error)); 2224 } 2225 2226 void 2227 vm_inject_pf(struct vcpu *vcpu, int error_code, uint64_t cr2) 2228 { 2229 int error __diagused; 2230 2231 VMM_CTR2(vcpu, "Injecting page fault: error_code %#x, cr2 %#lx", 2232 error_code, cr2); 2233 2234 error = vm_set_register(vcpu, VM_REG_GUEST_CR2, cr2); 2235 KASSERT(error == 0, ("vm_set_register(cr2) error %d", error)); 2236 2237 vm_inject_fault(vcpu, IDT_PF, 1, error_code); 2238 } 2239 2240 static VMM_STAT(VCPU_NMI_COUNT, "number of NMIs delivered to vcpu"); 2241 2242 int 2243 vm_inject_nmi(struct vcpu *vcpu) 2244 { 2245 2246 vcpu->nmi_pending = 1; 2247 vcpu_notify_event(vcpu, false); 2248 return (0); 2249 } 2250 2251 int 2252 vm_nmi_pending(struct vcpu *vcpu) 2253 { 2254 return (vcpu->nmi_pending); 2255 } 2256 2257 void 2258 vm_nmi_clear(struct vcpu *vcpu) 2259 { 2260 if (vcpu->nmi_pending == 0) 2261 panic("vm_nmi_clear: inconsistent nmi_pending state"); 2262 2263 vcpu->nmi_pending = 0; 2264 vmm_stat_incr(vcpu, VCPU_NMI_COUNT, 1); 2265 } 2266 2267 static VMM_STAT(VCPU_EXTINT_COUNT, "number of ExtINTs delivered to vcpu"); 2268 2269 int 2270 vm_inject_extint(struct vcpu *vcpu) 2271 { 2272 2273 vcpu->extint_pending = 1; 2274 vcpu_notify_event(vcpu, false); 2275 return (0); 2276 } 2277 2278 int 2279 vm_extint_pending(struct vcpu *vcpu) 2280 { 2281 return (vcpu->extint_pending); 2282 } 2283 2284 void 2285 vm_extint_clear(struct vcpu *vcpu) 2286 { 2287 if (vcpu->extint_pending == 0) 2288 panic("vm_extint_clear: inconsistent extint_pending state"); 2289 2290 vcpu->extint_pending = 0; 2291 vmm_stat_incr(vcpu, VCPU_EXTINT_COUNT, 1); 2292 } 2293 2294 int 2295 vm_get_capability(struct vcpu *vcpu, int type, int *retval) 2296 { 2297 if (type < 0 || type >= VM_CAP_MAX) 2298 return (EINVAL); 2299 2300 return (vmmops_getcap(vcpu->cookie, type, retval)); 2301 } 2302 2303 int 2304 vm_set_capability(struct vcpu *vcpu, int type, int val) 2305 { 2306 if (type < 0 || type >= VM_CAP_MAX) 2307 return (EINVAL); 2308 2309 return (vmmops_setcap(vcpu->cookie, type, val)); 2310 } 2311 2312 struct vm * 2313 vcpu_vm(struct vcpu *vcpu) 2314 { 2315 return (vcpu->vm); 2316 } 2317 2318 int 2319 vcpu_vcpuid(struct vcpu *vcpu) 2320 { 2321 return (vcpu->vcpuid); 2322 } 2323 2324 struct vcpu * 2325 vm_vcpu(struct vm *vm, int vcpuid) 2326 { 2327 return (vm->vcpu[vcpuid]); 2328 } 2329 2330 struct vlapic * 2331 vm_lapic(struct vcpu *vcpu) 2332 { 2333 return (vcpu->vlapic); 2334 } 2335 2336 struct vioapic * 2337 vm_ioapic(struct vm *vm) 2338 { 2339 2340 return (vm->vioapic); 2341 } 2342 2343 struct vhpet * 2344 vm_hpet(struct vm *vm) 2345 { 2346 2347 return (vm->vhpet); 2348 } 2349 2350 bool 2351 vmm_is_pptdev(int bus, int slot, int func) 2352 { 2353 int b, f, i, n, s; 2354 char *val, *cp, *cp2; 2355 bool found; 2356 2357 /* 2358 * XXX 2359 * The length of an environment variable is limited to 128 bytes which 2360 * puts an upper limit on the number of passthru devices that may be 2361 * specified using a single environment variable. 2362 * 2363 * Work around this by scanning multiple environment variable 2364 * names instead of a single one - yuck! 2365 */ 2366 const char *names[] = { "pptdevs", "pptdevs2", "pptdevs3", NULL }; 2367 2368 /* set pptdevs="1/2/3 4/5/6 7/8/9 10/11/12" */ 2369 found = false; 2370 for (i = 0; names[i] != NULL && !found; i++) { 2371 cp = val = kern_getenv(names[i]); 2372 while (cp != NULL && *cp != '\0') { 2373 if ((cp2 = strchr(cp, ' ')) != NULL) 2374 *cp2 = '\0'; 2375 2376 n = sscanf(cp, "%d/%d/%d", &b, &s, &f); 2377 if (n == 3 && bus == b && slot == s && func == f) { 2378 found = true; 2379 break; 2380 } 2381 2382 if (cp2 != NULL) 2383 *cp2++ = ' '; 2384 2385 cp = cp2; 2386 } 2387 freeenv(val); 2388 } 2389 return (found); 2390 } 2391 2392 void * 2393 vm_iommu_domain(struct vm *vm) 2394 { 2395 2396 return (vm->iommu); 2397 } 2398 2399 int 2400 vcpu_set_state(struct vcpu *vcpu, enum vcpu_state newstate, bool from_idle) 2401 { 2402 int error; 2403 2404 vcpu_lock(vcpu); 2405 error = vcpu_set_state_locked(vcpu, newstate, from_idle); 2406 vcpu_unlock(vcpu); 2407 2408 return (error); 2409 } 2410 2411 enum vcpu_state 2412 vcpu_get_state(struct vcpu *vcpu, int *hostcpu) 2413 { 2414 enum vcpu_state state; 2415 2416 vcpu_lock(vcpu); 2417 state = vcpu->state; 2418 if (hostcpu != NULL) 2419 *hostcpu = vcpu->hostcpu; 2420 vcpu_unlock(vcpu); 2421 2422 return (state); 2423 } 2424 2425 int 2426 vm_activate_cpu(struct vcpu *vcpu) 2427 { 2428 struct vm *vm = vcpu->vm; 2429 2430 if (CPU_ISSET(vcpu->vcpuid, &vm->active_cpus)) 2431 return (EBUSY); 2432 2433 VMM_CTR0(vcpu, "activated"); 2434 CPU_SET_ATOMIC(vcpu->vcpuid, &vm->active_cpus); 2435 return (0); 2436 } 2437 2438 int 2439 vm_suspend_cpu(struct vm *vm, struct vcpu *vcpu) 2440 { 2441 if (vcpu == NULL) { 2442 vm->debug_cpus = vm->active_cpus; 2443 for (int i = 0; i < vm->maxcpus; i++) { 2444 if (CPU_ISSET(i, &vm->active_cpus)) 2445 vcpu_notify_event(vm_vcpu(vm, i), false); 2446 } 2447 } else { 2448 if (!CPU_ISSET(vcpu->vcpuid, &vm->active_cpus)) 2449 return (EINVAL); 2450 2451 CPU_SET_ATOMIC(vcpu->vcpuid, &vm->debug_cpus); 2452 vcpu_notify_event(vcpu, false); 2453 } 2454 return (0); 2455 } 2456 2457 int 2458 vm_resume_cpu(struct vm *vm, struct vcpu *vcpu) 2459 { 2460 2461 if (vcpu == NULL) { 2462 CPU_ZERO(&vm->debug_cpus); 2463 } else { 2464 if (!CPU_ISSET(vcpu->vcpuid, &vm->debug_cpus)) 2465 return (EINVAL); 2466 2467 CPU_CLR_ATOMIC(vcpu->vcpuid, &vm->debug_cpus); 2468 } 2469 return (0); 2470 } 2471 2472 int 2473 vcpu_debugged(struct vcpu *vcpu) 2474 { 2475 2476 return (CPU_ISSET(vcpu->vcpuid, &vcpu->vm->debug_cpus)); 2477 } 2478 2479 cpuset_t 2480 vm_active_cpus(struct vm *vm) 2481 { 2482 2483 return (vm->active_cpus); 2484 } 2485 2486 cpuset_t 2487 vm_debug_cpus(struct vm *vm) 2488 { 2489 2490 return (vm->debug_cpus); 2491 } 2492 2493 cpuset_t 2494 vm_suspended_cpus(struct vm *vm) 2495 { 2496 2497 return (vm->suspended_cpus); 2498 } 2499 2500 /* 2501 * Returns the subset of vCPUs in tostart that are awaiting startup. 2502 * These vCPUs are also marked as no longer awaiting startup. 2503 */ 2504 cpuset_t 2505 vm_start_cpus(struct vm *vm, const cpuset_t *tostart) 2506 { 2507 cpuset_t set; 2508 2509 mtx_lock(&vm->rendezvous_mtx); 2510 CPU_AND(&set, &vm->startup_cpus, tostart); 2511 CPU_ANDNOT(&vm->startup_cpus, &vm->startup_cpus, &set); 2512 mtx_unlock(&vm->rendezvous_mtx); 2513 return (set); 2514 } 2515 2516 void 2517 vm_await_start(struct vm *vm, const cpuset_t *waiting) 2518 { 2519 mtx_lock(&vm->rendezvous_mtx); 2520 CPU_OR(&vm->startup_cpus, &vm->startup_cpus, waiting); 2521 mtx_unlock(&vm->rendezvous_mtx); 2522 } 2523 2524 void * 2525 vcpu_stats(struct vcpu *vcpu) 2526 { 2527 2528 return (vcpu->stats); 2529 } 2530 2531 int 2532 vm_get_x2apic_state(struct vcpu *vcpu, enum x2apic_state *state) 2533 { 2534 *state = vcpu->x2apic_state; 2535 2536 return (0); 2537 } 2538 2539 int 2540 vm_set_x2apic_state(struct vcpu *vcpu, enum x2apic_state state) 2541 { 2542 if (state >= X2APIC_STATE_LAST) 2543 return (EINVAL); 2544 2545 vcpu->x2apic_state = state; 2546 2547 vlapic_set_x2apic_state(vcpu, state); 2548 2549 return (0); 2550 } 2551 2552 /* 2553 * This function is called to ensure that a vcpu "sees" a pending event 2554 * as soon as possible: 2555 * - If the vcpu thread is sleeping then it is woken up. 2556 * - If the vcpu is running on a different host_cpu then an IPI will be directed 2557 * to the host_cpu to cause the vcpu to trap into the hypervisor. 2558 */ 2559 static void 2560 vcpu_notify_event_locked(struct vcpu *vcpu, bool lapic_intr) 2561 { 2562 int hostcpu; 2563 2564 hostcpu = vcpu->hostcpu; 2565 if (vcpu->state == VCPU_RUNNING) { 2566 KASSERT(hostcpu != NOCPU, ("vcpu running on invalid hostcpu")); 2567 if (hostcpu != curcpu) { 2568 if (lapic_intr) { 2569 vlapic_post_intr(vcpu->vlapic, hostcpu, 2570 vmm_ipinum); 2571 } else { 2572 ipi_cpu(hostcpu, vmm_ipinum); 2573 } 2574 } else { 2575 /* 2576 * If the 'vcpu' is running on 'curcpu' then it must 2577 * be sending a notification to itself (e.g. SELF_IPI). 2578 * The pending event will be picked up when the vcpu 2579 * transitions back to guest context. 2580 */ 2581 } 2582 } else { 2583 KASSERT(hostcpu == NOCPU, ("vcpu state %d not consistent " 2584 "with hostcpu %d", vcpu->state, hostcpu)); 2585 if (vcpu->state == VCPU_SLEEPING) 2586 wakeup_one(vcpu); 2587 } 2588 } 2589 2590 void 2591 vcpu_notify_event(struct vcpu *vcpu, bool lapic_intr) 2592 { 2593 vcpu_lock(vcpu); 2594 vcpu_notify_event_locked(vcpu, lapic_intr); 2595 vcpu_unlock(vcpu); 2596 } 2597 2598 struct vmspace * 2599 vm_get_vmspace(struct vm *vm) 2600 { 2601 2602 return (vm->vmspace); 2603 } 2604 2605 int 2606 vm_apicid2vcpuid(struct vm *vm, int apicid) 2607 { 2608 /* 2609 * XXX apic id is assumed to be numerically identical to vcpu id 2610 */ 2611 return (apicid); 2612 } 2613 2614 int 2615 vm_smp_rendezvous(struct vcpu *vcpu, cpuset_t dest, 2616 vm_rendezvous_func_t func, void *arg) 2617 { 2618 struct vm *vm = vcpu->vm; 2619 int error, i; 2620 2621 /* 2622 * Enforce that this function is called without any locks 2623 */ 2624 WITNESS_WARN(WARN_PANIC, NULL, "vm_smp_rendezvous"); 2625 2626 restart: 2627 mtx_lock(&vm->rendezvous_mtx); 2628 if (vm->rendezvous_func != NULL) { 2629 /* 2630 * If a rendezvous is already in progress then we need to 2631 * call the rendezvous handler in case this 'vcpu' is one 2632 * of the targets of the rendezvous. 2633 */ 2634 VMM_CTR0(vcpu, "Rendezvous already in progress"); 2635 mtx_unlock(&vm->rendezvous_mtx); 2636 error = vm_handle_rendezvous(vcpu); 2637 if (error != 0) 2638 return (error); 2639 goto restart; 2640 } 2641 KASSERT(vm->rendezvous_func == NULL, ("vm_smp_rendezvous: previous " 2642 "rendezvous is still in progress")); 2643 2644 VMM_CTR0(vcpu, "Initiating rendezvous"); 2645 vm->rendezvous_req_cpus = dest; 2646 CPU_ZERO(&vm->rendezvous_done_cpus); 2647 vm->rendezvous_arg = arg; 2648 vm->rendezvous_func = func; 2649 mtx_unlock(&vm->rendezvous_mtx); 2650 2651 /* 2652 * Wake up any sleeping vcpus and trigger a VM-exit in any running 2653 * vcpus so they handle the rendezvous as soon as possible. 2654 */ 2655 for (i = 0; i < vm->maxcpus; i++) { 2656 if (CPU_ISSET(i, &dest)) 2657 vcpu_notify_event(vm_vcpu(vm, i), false); 2658 } 2659 2660 return (vm_handle_rendezvous(vcpu)); 2661 } 2662 2663 struct vatpic * 2664 vm_atpic(struct vm *vm) 2665 { 2666 return (vm->vatpic); 2667 } 2668 2669 struct vatpit * 2670 vm_atpit(struct vm *vm) 2671 { 2672 return (vm->vatpit); 2673 } 2674 2675 struct vpmtmr * 2676 vm_pmtmr(struct vm *vm) 2677 { 2678 2679 return (vm->vpmtmr); 2680 } 2681 2682 struct vrtc * 2683 vm_rtc(struct vm *vm) 2684 { 2685 2686 return (vm->vrtc); 2687 } 2688 2689 enum vm_reg_name 2690 vm_segment_name(int seg) 2691 { 2692 static enum vm_reg_name seg_names[] = { 2693 VM_REG_GUEST_ES, 2694 VM_REG_GUEST_CS, 2695 VM_REG_GUEST_SS, 2696 VM_REG_GUEST_DS, 2697 VM_REG_GUEST_FS, 2698 VM_REG_GUEST_GS 2699 }; 2700 2701 KASSERT(seg >= 0 && seg < nitems(seg_names), 2702 ("%s: invalid segment encoding %d", __func__, seg)); 2703 return (seg_names[seg]); 2704 } 2705 2706 void 2707 vm_copy_teardown(struct vm_copyinfo *copyinfo, int num_copyinfo) 2708 { 2709 int idx; 2710 2711 for (idx = 0; idx < num_copyinfo; idx++) { 2712 if (copyinfo[idx].cookie != NULL) 2713 vm_gpa_release(copyinfo[idx].cookie); 2714 } 2715 bzero(copyinfo, num_copyinfo * sizeof(struct vm_copyinfo)); 2716 } 2717 2718 int 2719 vm_copy_setup(struct vcpu *vcpu, struct vm_guest_paging *paging, 2720 uint64_t gla, size_t len, int prot, struct vm_copyinfo *copyinfo, 2721 int num_copyinfo, int *fault) 2722 { 2723 int error, idx, nused; 2724 size_t n, off, remaining; 2725 void *hva, *cookie; 2726 uint64_t gpa; 2727 2728 bzero(copyinfo, sizeof(struct vm_copyinfo) * num_copyinfo); 2729 2730 nused = 0; 2731 remaining = len; 2732 while (remaining > 0) { 2733 KASSERT(nused < num_copyinfo, ("insufficient vm_copyinfo")); 2734 error = vm_gla2gpa(vcpu, paging, gla, prot, &gpa, fault); 2735 if (error || *fault) 2736 return (error); 2737 off = gpa & PAGE_MASK; 2738 n = min(remaining, PAGE_SIZE - off); 2739 copyinfo[nused].gpa = gpa; 2740 copyinfo[nused].len = n; 2741 remaining -= n; 2742 gla += n; 2743 nused++; 2744 } 2745 2746 for (idx = 0; idx < nused; idx++) { 2747 hva = vm_gpa_hold(vcpu, copyinfo[idx].gpa, 2748 copyinfo[idx].len, prot, &cookie); 2749 if (hva == NULL) 2750 break; 2751 copyinfo[idx].hva = hva; 2752 copyinfo[idx].cookie = cookie; 2753 } 2754 2755 if (idx != nused) { 2756 vm_copy_teardown(copyinfo, num_copyinfo); 2757 return (EFAULT); 2758 } else { 2759 *fault = 0; 2760 return (0); 2761 } 2762 } 2763 2764 void 2765 vm_copyin(struct vm_copyinfo *copyinfo, void *kaddr, size_t len) 2766 { 2767 char *dst; 2768 int idx; 2769 2770 dst = kaddr; 2771 idx = 0; 2772 while (len > 0) { 2773 bcopy(copyinfo[idx].hva, dst, copyinfo[idx].len); 2774 len -= copyinfo[idx].len; 2775 dst += copyinfo[idx].len; 2776 idx++; 2777 } 2778 } 2779 2780 void 2781 vm_copyout(const void *kaddr, struct vm_copyinfo *copyinfo, size_t len) 2782 { 2783 const char *src; 2784 int idx; 2785 2786 src = kaddr; 2787 idx = 0; 2788 while (len > 0) { 2789 bcopy(src, copyinfo[idx].hva, copyinfo[idx].len); 2790 len -= copyinfo[idx].len; 2791 src += copyinfo[idx].len; 2792 idx++; 2793 } 2794 } 2795 2796 /* 2797 * Return the amount of in-use and wired memory for the VM. Since 2798 * these are global stats, only return the values with for vCPU 0 2799 */ 2800 VMM_STAT_DECLARE(VMM_MEM_RESIDENT); 2801 VMM_STAT_DECLARE(VMM_MEM_WIRED); 2802 2803 static void 2804 vm_get_rescnt(struct vcpu *vcpu, struct vmm_stat_type *stat) 2805 { 2806 2807 if (vcpu->vcpuid == 0) { 2808 vmm_stat_set(vcpu, VMM_MEM_RESIDENT, PAGE_SIZE * 2809 vmspace_resident_count(vcpu->vm->vmspace)); 2810 } 2811 } 2812 2813 static void 2814 vm_get_wiredcnt(struct vcpu *vcpu, struct vmm_stat_type *stat) 2815 { 2816 2817 if (vcpu->vcpuid == 0) { 2818 vmm_stat_set(vcpu, VMM_MEM_WIRED, PAGE_SIZE * 2819 pmap_wired_count(vmspace_pmap(vcpu->vm->vmspace))); 2820 } 2821 } 2822 2823 VMM_STAT_FUNC(VMM_MEM_RESIDENT, "Resident memory", vm_get_rescnt); 2824 VMM_STAT_FUNC(VMM_MEM_WIRED, "Wired memory", vm_get_wiredcnt); 2825 2826 #ifdef BHYVE_SNAPSHOT 2827 static int 2828 vm_snapshot_vcpus(struct vm *vm, struct vm_snapshot_meta *meta) 2829 { 2830 uint64_t tsc, now; 2831 int ret; 2832 struct vcpu *vcpu; 2833 uint16_t i, maxcpus; 2834 2835 now = rdtsc(); 2836 maxcpus = vm_get_maxcpus(vm); 2837 for (i = 0; i < maxcpus; i++) { 2838 vcpu = vm->vcpu[i]; 2839 if (vcpu == NULL) 2840 continue; 2841 2842 SNAPSHOT_VAR_OR_LEAVE(vcpu->x2apic_state, meta, ret, done); 2843 SNAPSHOT_VAR_OR_LEAVE(vcpu->exitintinfo, meta, ret, done); 2844 SNAPSHOT_VAR_OR_LEAVE(vcpu->exc_vector, meta, ret, done); 2845 SNAPSHOT_VAR_OR_LEAVE(vcpu->exc_errcode_valid, meta, ret, done); 2846 SNAPSHOT_VAR_OR_LEAVE(vcpu->exc_errcode, meta, ret, done); 2847 SNAPSHOT_VAR_OR_LEAVE(vcpu->guest_xcr0, meta, ret, done); 2848 SNAPSHOT_VAR_OR_LEAVE(vcpu->exitinfo, meta, ret, done); 2849 SNAPSHOT_VAR_OR_LEAVE(vcpu->nextrip, meta, ret, done); 2850 2851 /* 2852 * Save the absolute TSC value by adding now to tsc_offset. 2853 * 2854 * It will be turned turned back into an actual offset when the 2855 * TSC restore function is called 2856 */ 2857 tsc = now + vcpu->tsc_offset; 2858 SNAPSHOT_VAR_OR_LEAVE(tsc, meta, ret, done); 2859 } 2860 2861 done: 2862 return (ret); 2863 } 2864 2865 static int 2866 vm_snapshot_vm(struct vm *vm, struct vm_snapshot_meta *meta) 2867 { 2868 int ret; 2869 2870 ret = vm_snapshot_vcpus(vm, meta); 2871 if (ret != 0) 2872 goto done; 2873 2874 SNAPSHOT_VAR_OR_LEAVE(vm->startup_cpus, meta, ret, done); 2875 done: 2876 return (ret); 2877 } 2878 2879 static int 2880 vm_snapshot_vcpu(struct vm *vm, struct vm_snapshot_meta *meta) 2881 { 2882 int error; 2883 struct vcpu *vcpu; 2884 uint16_t i, maxcpus; 2885 2886 error = 0; 2887 2888 maxcpus = vm_get_maxcpus(vm); 2889 for (i = 0; i < maxcpus; i++) { 2890 vcpu = vm->vcpu[i]; 2891 if (vcpu == NULL) 2892 continue; 2893 2894 error = vmmops_vcpu_snapshot(vcpu->cookie, meta); 2895 if (error != 0) { 2896 printf("%s: failed to snapshot vmcs/vmcb data for " 2897 "vCPU: %d; error: %d\n", __func__, i, error); 2898 goto done; 2899 } 2900 } 2901 2902 done: 2903 return (error); 2904 } 2905 2906 /* 2907 * Save kernel-side structures to user-space for snapshotting. 2908 */ 2909 int 2910 vm_snapshot_req(struct vm *vm, struct vm_snapshot_meta *meta) 2911 { 2912 int ret = 0; 2913 2914 switch (meta->dev_req) { 2915 case STRUCT_VMX: 2916 ret = vmmops_snapshot(vm->cookie, meta); 2917 break; 2918 case STRUCT_VMCX: 2919 ret = vm_snapshot_vcpu(vm, meta); 2920 break; 2921 case STRUCT_VM: 2922 ret = vm_snapshot_vm(vm, meta); 2923 break; 2924 case STRUCT_VIOAPIC: 2925 ret = vioapic_snapshot(vm_ioapic(vm), meta); 2926 break; 2927 case STRUCT_VLAPIC: 2928 ret = vlapic_snapshot(vm, meta); 2929 break; 2930 case STRUCT_VHPET: 2931 ret = vhpet_snapshot(vm_hpet(vm), meta); 2932 break; 2933 case STRUCT_VATPIC: 2934 ret = vatpic_snapshot(vm_atpic(vm), meta); 2935 break; 2936 case STRUCT_VATPIT: 2937 ret = vatpit_snapshot(vm_atpit(vm), meta); 2938 break; 2939 case STRUCT_VPMTMR: 2940 ret = vpmtmr_snapshot(vm_pmtmr(vm), meta); 2941 break; 2942 case STRUCT_VRTC: 2943 ret = vrtc_snapshot(vm_rtc(vm), meta); 2944 break; 2945 default: 2946 printf("%s: failed to find the requested type %#x\n", 2947 __func__, meta->dev_req); 2948 ret = (EINVAL); 2949 } 2950 return (ret); 2951 } 2952 2953 void 2954 vm_set_tsc_offset(struct vcpu *vcpu, uint64_t offset) 2955 { 2956 vcpu->tsc_offset = offset; 2957 } 2958 2959 int 2960 vm_restore_time(struct vm *vm) 2961 { 2962 int error; 2963 uint64_t now; 2964 struct vcpu *vcpu; 2965 uint16_t i, maxcpus; 2966 2967 now = rdtsc(); 2968 2969 error = vhpet_restore_time(vm_hpet(vm)); 2970 if (error) 2971 return (error); 2972 2973 maxcpus = vm_get_maxcpus(vm); 2974 for (i = 0; i < maxcpus; i++) { 2975 vcpu = vm->vcpu[i]; 2976 if (vcpu == NULL) 2977 continue; 2978 2979 error = vmmops_restore_tsc(vcpu->cookie, 2980 vcpu->tsc_offset - now); 2981 if (error) 2982 return (error); 2983 } 2984 2985 return (0); 2986 } 2987 #endif 2988