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 if (destroy) 655 vm_xlock_memsegs(vm); 656 657 ppt_unassign_all(vm); 658 659 if (vm->iommu != NULL) 660 iommu_destroy_domain(vm->iommu); 661 662 if (destroy) 663 vrtc_cleanup(vm->vrtc); 664 else 665 vrtc_reset(vm->vrtc); 666 vpmtmr_cleanup(vm->vpmtmr); 667 vatpit_cleanup(vm->vatpit); 668 vhpet_cleanup(vm->vhpet); 669 vatpic_cleanup(vm->vatpic); 670 vioapic_cleanup(vm->vioapic); 671 672 for (i = 0; i < vm->maxcpus; i++) { 673 if (vm->vcpu[i] != NULL) 674 vcpu_cleanup(vm->vcpu[i], destroy); 675 } 676 677 vmmops_cleanup(vm->cookie); 678 679 /* 680 * System memory is removed from the guest address space only when 681 * the VM is destroyed. This is because the mapping remains the same 682 * across VM reset. 683 * 684 * Device memory can be relocated by the guest (e.g. using PCI BARs) 685 * so those mappings are removed on a VM reset. 686 */ 687 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 688 mm = &vm->mem_maps[i]; 689 if (destroy || !sysmem_mapping(vm, mm)) 690 vm_free_memmap(vm, i); 691 } 692 693 if (destroy) { 694 for (i = 0; i < VM_MAX_MEMSEGS; i++) 695 vm_free_memseg(vm, i); 696 vm_unlock_memsegs(vm); 697 698 vmmops_vmspace_free(vm->vmspace); 699 vm->vmspace = NULL; 700 701 free(vm->vcpu, M_VM); 702 sx_destroy(&vm->vcpus_init_lock); 703 sx_destroy(&vm->mem_segs_lock); 704 mtx_destroy(&vm->rendezvous_mtx); 705 } 706 } 707 708 void 709 vm_destroy(struct vm *vm) 710 { 711 vm_cleanup(vm, true); 712 free(vm, M_VM); 713 } 714 715 int 716 vm_reinit(struct vm *vm) 717 { 718 int error; 719 720 /* 721 * A virtual machine can be reset only if all vcpus are suspended. 722 */ 723 if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) { 724 vm_cleanup(vm, false); 725 vm_init(vm, false); 726 error = 0; 727 } else { 728 error = EBUSY; 729 } 730 731 return (error); 732 } 733 734 const char * 735 vm_name(struct vm *vm) 736 { 737 return (vm->name); 738 } 739 740 void 741 vm_slock_memsegs(struct vm *vm) 742 { 743 sx_slock(&vm->mem_segs_lock); 744 } 745 746 void 747 vm_xlock_memsegs(struct vm *vm) 748 { 749 sx_xlock(&vm->mem_segs_lock); 750 } 751 752 void 753 vm_unlock_memsegs(struct vm *vm) 754 { 755 sx_unlock(&vm->mem_segs_lock); 756 } 757 758 int 759 vm_map_mmio(struct vm *vm, vm_paddr_t gpa, size_t len, vm_paddr_t hpa) 760 { 761 vm_object_t obj; 762 763 if ((obj = vmm_mmio_alloc(vm->vmspace, gpa, len, hpa)) == NULL) 764 return (ENOMEM); 765 else 766 return (0); 767 } 768 769 int 770 vm_unmap_mmio(struct vm *vm, vm_paddr_t gpa, size_t len) 771 { 772 773 vmm_mmio_free(vm->vmspace, gpa, len); 774 return (0); 775 } 776 777 /* 778 * Return 'true' if 'gpa' is allocated in the guest address space. 779 * 780 * This function is called in the context of a running vcpu which acts as 781 * an implicit lock on 'vm->mem_maps[]'. 782 */ 783 bool 784 vm_mem_allocated(struct vcpu *vcpu, vm_paddr_t gpa) 785 { 786 struct vm *vm = vcpu->vm; 787 struct mem_map *mm; 788 int i; 789 790 #ifdef INVARIANTS 791 int hostcpu, state; 792 state = vcpu_get_state(vcpu, &hostcpu); 793 KASSERT(state == VCPU_RUNNING && hostcpu == curcpu, 794 ("%s: invalid vcpu state %d/%d", __func__, state, hostcpu)); 795 #endif 796 797 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 798 mm = &vm->mem_maps[i]; 799 if (mm->len != 0 && gpa >= mm->gpa && gpa < mm->gpa + mm->len) 800 return (true); /* 'gpa' is sysmem or devmem */ 801 } 802 803 if (ppt_is_mmio(vm, gpa)) 804 return (true); /* 'gpa' is pci passthru mmio */ 805 806 return (false); 807 } 808 809 int 810 vm_alloc_memseg(struct vm *vm, int ident, size_t len, bool sysmem) 811 { 812 struct mem_seg *seg; 813 vm_object_t obj; 814 815 sx_assert(&vm->mem_segs_lock, SX_XLOCKED); 816 817 if (ident < 0 || ident >= VM_MAX_MEMSEGS) 818 return (EINVAL); 819 820 if (len == 0 || (len & PAGE_MASK)) 821 return (EINVAL); 822 823 seg = &vm->mem_segs[ident]; 824 if (seg->object != NULL) { 825 if (seg->len == len && seg->sysmem == sysmem) 826 return (EEXIST); 827 else 828 return (EINVAL); 829 } 830 831 obj = vm_object_allocate(OBJT_SWAP, len >> PAGE_SHIFT); 832 if (obj == NULL) 833 return (ENOMEM); 834 835 seg->len = len; 836 seg->object = obj; 837 seg->sysmem = sysmem; 838 return (0); 839 } 840 841 int 842 vm_get_memseg(struct vm *vm, int ident, size_t *len, bool *sysmem, 843 vm_object_t *objptr) 844 { 845 struct mem_seg *seg; 846 847 sx_assert(&vm->mem_segs_lock, SX_LOCKED); 848 849 if (ident < 0 || ident >= VM_MAX_MEMSEGS) 850 return (EINVAL); 851 852 seg = &vm->mem_segs[ident]; 853 if (len) 854 *len = seg->len; 855 if (sysmem) 856 *sysmem = seg->sysmem; 857 if (objptr) 858 *objptr = seg->object; 859 return (0); 860 } 861 862 void 863 vm_free_memseg(struct vm *vm, int ident) 864 { 865 struct mem_seg *seg; 866 867 KASSERT(ident >= 0 && ident < VM_MAX_MEMSEGS, 868 ("%s: invalid memseg ident %d", __func__, ident)); 869 870 seg = &vm->mem_segs[ident]; 871 if (seg->object != NULL) { 872 vm_object_deallocate(seg->object); 873 bzero(seg, sizeof(struct mem_seg)); 874 } 875 } 876 877 int 878 vm_mmap_memseg(struct vm *vm, vm_paddr_t gpa, int segid, vm_ooffset_t first, 879 size_t len, int prot, int flags) 880 { 881 struct mem_seg *seg; 882 struct mem_map *m, *map; 883 vm_ooffset_t last; 884 int i, error; 885 886 if (prot == 0 || (prot & ~(VM_PROT_ALL)) != 0) 887 return (EINVAL); 888 889 if (flags & ~VM_MEMMAP_F_WIRED) 890 return (EINVAL); 891 892 if (segid < 0 || segid >= VM_MAX_MEMSEGS) 893 return (EINVAL); 894 895 seg = &vm->mem_segs[segid]; 896 if (seg->object == NULL) 897 return (EINVAL); 898 899 last = first + len; 900 if (first < 0 || first >= last || last > seg->len) 901 return (EINVAL); 902 903 if ((gpa | first | last) & PAGE_MASK) 904 return (EINVAL); 905 906 map = NULL; 907 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 908 m = &vm->mem_maps[i]; 909 if (m->len == 0) { 910 map = m; 911 break; 912 } 913 } 914 915 if (map == NULL) 916 return (ENOSPC); 917 918 error = vm_map_find(&vm->vmspace->vm_map, seg->object, first, &gpa, 919 len, 0, VMFS_NO_SPACE, prot, prot, 0); 920 if (error != KERN_SUCCESS) 921 return (EFAULT); 922 923 vm_object_reference(seg->object); 924 925 if (flags & VM_MEMMAP_F_WIRED) { 926 error = vm_map_wire(&vm->vmspace->vm_map, gpa, gpa + len, 927 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES); 928 if (error != KERN_SUCCESS) { 929 vm_map_remove(&vm->vmspace->vm_map, gpa, gpa + len); 930 return (error == KERN_RESOURCE_SHORTAGE ? ENOMEM : 931 EFAULT); 932 } 933 } 934 935 map->gpa = gpa; 936 map->len = len; 937 map->segoff = first; 938 map->segid = segid; 939 map->prot = prot; 940 map->flags = flags; 941 return (0); 942 } 943 944 int 945 vm_munmap_memseg(struct vm *vm, vm_paddr_t gpa, size_t len) 946 { 947 struct mem_map *m; 948 int i; 949 950 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 951 m = &vm->mem_maps[i]; 952 if (m->gpa == gpa && m->len == len && 953 (m->flags & VM_MEMMAP_F_IOMMU) == 0) { 954 vm_free_memmap(vm, i); 955 return (0); 956 } 957 } 958 959 return (EINVAL); 960 } 961 962 int 963 vm_mmap_getnext(struct vm *vm, vm_paddr_t *gpa, int *segid, 964 vm_ooffset_t *segoff, size_t *len, int *prot, int *flags) 965 { 966 struct mem_map *mm, *mmnext; 967 int i; 968 969 mmnext = NULL; 970 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 971 mm = &vm->mem_maps[i]; 972 if (mm->len == 0 || mm->gpa < *gpa) 973 continue; 974 if (mmnext == NULL || mm->gpa < mmnext->gpa) 975 mmnext = mm; 976 } 977 978 if (mmnext != NULL) { 979 *gpa = mmnext->gpa; 980 if (segid) 981 *segid = mmnext->segid; 982 if (segoff) 983 *segoff = mmnext->segoff; 984 if (len) 985 *len = mmnext->len; 986 if (prot) 987 *prot = mmnext->prot; 988 if (flags) 989 *flags = mmnext->flags; 990 return (0); 991 } else { 992 return (ENOENT); 993 } 994 } 995 996 static void 997 vm_free_memmap(struct vm *vm, int ident) 998 { 999 struct mem_map *mm; 1000 int error __diagused; 1001 1002 mm = &vm->mem_maps[ident]; 1003 if (mm->len) { 1004 error = vm_map_remove(&vm->vmspace->vm_map, mm->gpa, 1005 mm->gpa + mm->len); 1006 KASSERT(error == KERN_SUCCESS, ("%s: vm_map_remove error %d", 1007 __func__, error)); 1008 bzero(mm, sizeof(struct mem_map)); 1009 } 1010 } 1011 1012 static __inline bool 1013 sysmem_mapping(struct vm *vm, struct mem_map *mm) 1014 { 1015 1016 if (mm->len != 0 && vm->mem_segs[mm->segid].sysmem) 1017 return (true); 1018 else 1019 return (false); 1020 } 1021 1022 vm_paddr_t 1023 vmm_sysmem_maxaddr(struct vm *vm) 1024 { 1025 struct mem_map *mm; 1026 vm_paddr_t maxaddr; 1027 int i; 1028 1029 maxaddr = 0; 1030 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 1031 mm = &vm->mem_maps[i]; 1032 if (sysmem_mapping(vm, mm)) { 1033 if (maxaddr < mm->gpa + mm->len) 1034 maxaddr = mm->gpa + mm->len; 1035 } 1036 } 1037 return (maxaddr); 1038 } 1039 1040 static void 1041 vm_iommu_modify(struct vm *vm, bool map) 1042 { 1043 int i, sz; 1044 vm_paddr_t gpa, hpa; 1045 struct mem_map *mm; 1046 void *vp, *cookie, *host_domain; 1047 1048 sz = PAGE_SIZE; 1049 host_domain = iommu_host_domain(); 1050 1051 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 1052 mm = &vm->mem_maps[i]; 1053 if (!sysmem_mapping(vm, mm)) 1054 continue; 1055 1056 if (map) { 1057 KASSERT((mm->flags & VM_MEMMAP_F_IOMMU) == 0, 1058 ("iommu map found invalid memmap %#lx/%#lx/%#x", 1059 mm->gpa, mm->len, mm->flags)); 1060 if ((mm->flags & VM_MEMMAP_F_WIRED) == 0) 1061 continue; 1062 mm->flags |= VM_MEMMAP_F_IOMMU; 1063 } else { 1064 if ((mm->flags & VM_MEMMAP_F_IOMMU) == 0) 1065 continue; 1066 mm->flags &= ~VM_MEMMAP_F_IOMMU; 1067 KASSERT((mm->flags & VM_MEMMAP_F_WIRED) != 0, 1068 ("iommu unmap found invalid memmap %#lx/%#lx/%#x", 1069 mm->gpa, mm->len, mm->flags)); 1070 } 1071 1072 gpa = mm->gpa; 1073 while (gpa < mm->gpa + mm->len) { 1074 vp = vm_gpa_hold_global(vm, gpa, PAGE_SIZE, 1075 VM_PROT_WRITE, &cookie); 1076 KASSERT(vp != NULL, ("vm(%s) could not map gpa %#lx", 1077 vm_name(vm), gpa)); 1078 1079 vm_gpa_release(cookie); 1080 1081 hpa = DMAP_TO_PHYS((uintptr_t)vp); 1082 if (map) { 1083 iommu_create_mapping(vm->iommu, gpa, hpa, sz); 1084 } else { 1085 iommu_remove_mapping(vm->iommu, gpa, sz); 1086 } 1087 1088 gpa += PAGE_SIZE; 1089 } 1090 } 1091 1092 /* 1093 * Invalidate the cached translations associated with the domain 1094 * from which pages were removed. 1095 */ 1096 if (map) 1097 iommu_invalidate_tlb(host_domain); 1098 else 1099 iommu_invalidate_tlb(vm->iommu); 1100 } 1101 1102 #define vm_iommu_unmap(vm) vm_iommu_modify((vm), false) 1103 #define vm_iommu_map(vm) vm_iommu_modify((vm), true) 1104 1105 int 1106 vm_unassign_pptdev(struct vm *vm, int bus, int slot, int func) 1107 { 1108 int error; 1109 1110 error = ppt_unassign_device(vm, bus, slot, func); 1111 if (error) 1112 return (error); 1113 1114 if (ppt_assigned_devices(vm) == 0) 1115 vm_iommu_unmap(vm); 1116 1117 return (0); 1118 } 1119 1120 int 1121 vm_assign_pptdev(struct vm *vm, int bus, int slot, int func) 1122 { 1123 int error; 1124 vm_paddr_t maxaddr; 1125 1126 /* Set up the IOMMU to do the 'gpa' to 'hpa' translation */ 1127 if (ppt_assigned_devices(vm) == 0) { 1128 KASSERT(vm->iommu == NULL, 1129 ("vm_assign_pptdev: iommu must be NULL")); 1130 maxaddr = vmm_sysmem_maxaddr(vm); 1131 vm->iommu = iommu_create_domain(maxaddr); 1132 if (vm->iommu == NULL) 1133 return (ENXIO); 1134 vm_iommu_map(vm); 1135 } 1136 1137 error = ppt_assign_device(vm, bus, slot, func); 1138 return (error); 1139 } 1140 1141 static void * 1142 _vm_gpa_hold(struct vm *vm, vm_paddr_t gpa, size_t len, int reqprot, 1143 void **cookie) 1144 { 1145 int i, count, pageoff; 1146 struct mem_map *mm; 1147 vm_page_t m; 1148 1149 pageoff = gpa & PAGE_MASK; 1150 if (len > PAGE_SIZE - pageoff) 1151 panic("vm_gpa_hold: invalid gpa/len: 0x%016lx/%lu", gpa, len); 1152 1153 count = 0; 1154 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 1155 mm = &vm->mem_maps[i]; 1156 if (gpa >= mm->gpa && gpa < mm->gpa + mm->len) { 1157 count = vm_fault_quick_hold_pages(&vm->vmspace->vm_map, 1158 trunc_page(gpa), PAGE_SIZE, reqprot, &m, 1); 1159 break; 1160 } 1161 } 1162 1163 if (count == 1) { 1164 *cookie = m; 1165 return ((void *)(PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m)) + pageoff)); 1166 } else { 1167 *cookie = NULL; 1168 return (NULL); 1169 } 1170 } 1171 1172 void * 1173 vm_gpa_hold(struct vcpu *vcpu, vm_paddr_t gpa, size_t len, int reqprot, 1174 void **cookie) 1175 { 1176 #ifdef INVARIANTS 1177 /* 1178 * The current vcpu should be frozen to ensure 'vm_memmap[]' 1179 * stability. 1180 */ 1181 int state = vcpu_get_state(vcpu, NULL); 1182 KASSERT(state == VCPU_FROZEN, ("%s: invalid vcpu state %d", 1183 __func__, state)); 1184 #endif 1185 return (_vm_gpa_hold(vcpu->vm, gpa, len, reqprot, cookie)); 1186 } 1187 1188 void * 1189 vm_gpa_hold_global(struct vm *vm, vm_paddr_t gpa, size_t len, int reqprot, 1190 void **cookie) 1191 { 1192 sx_assert(&vm->mem_segs_lock, SX_LOCKED); 1193 return (_vm_gpa_hold(vm, gpa, len, reqprot, cookie)); 1194 } 1195 1196 void 1197 vm_gpa_release(void *cookie) 1198 { 1199 vm_page_t m = cookie; 1200 1201 vm_page_unwire(m, PQ_ACTIVE); 1202 } 1203 1204 int 1205 vm_get_register(struct vcpu *vcpu, int reg, uint64_t *retval) 1206 { 1207 1208 if (reg >= VM_REG_LAST) 1209 return (EINVAL); 1210 1211 return (vmmops_getreg(vcpu->cookie, reg, retval)); 1212 } 1213 1214 int 1215 vm_set_register(struct vcpu *vcpu, int reg, uint64_t val) 1216 { 1217 int error; 1218 1219 if (reg >= VM_REG_LAST) 1220 return (EINVAL); 1221 1222 error = vmmops_setreg(vcpu->cookie, reg, val); 1223 if (error || reg != VM_REG_GUEST_RIP) 1224 return (error); 1225 1226 /* Set 'nextrip' to match the value of %rip */ 1227 VMM_CTR1(vcpu, "Setting nextrip to %#lx", val); 1228 vcpu->nextrip = val; 1229 return (0); 1230 } 1231 1232 static bool 1233 is_descriptor_table(int reg) 1234 { 1235 1236 switch (reg) { 1237 case VM_REG_GUEST_IDTR: 1238 case VM_REG_GUEST_GDTR: 1239 return (true); 1240 default: 1241 return (false); 1242 } 1243 } 1244 1245 static bool 1246 is_segment_register(int reg) 1247 { 1248 1249 switch (reg) { 1250 case VM_REG_GUEST_ES: 1251 case VM_REG_GUEST_CS: 1252 case VM_REG_GUEST_SS: 1253 case VM_REG_GUEST_DS: 1254 case VM_REG_GUEST_FS: 1255 case VM_REG_GUEST_GS: 1256 case VM_REG_GUEST_TR: 1257 case VM_REG_GUEST_LDTR: 1258 return (true); 1259 default: 1260 return (false); 1261 } 1262 } 1263 1264 int 1265 vm_get_seg_desc(struct vcpu *vcpu, int reg, struct seg_desc *desc) 1266 { 1267 1268 if (!is_segment_register(reg) && !is_descriptor_table(reg)) 1269 return (EINVAL); 1270 1271 return (vmmops_getdesc(vcpu->cookie, reg, desc)); 1272 } 1273 1274 int 1275 vm_set_seg_desc(struct vcpu *vcpu, int reg, struct seg_desc *desc) 1276 { 1277 1278 if (!is_segment_register(reg) && !is_descriptor_table(reg)) 1279 return (EINVAL); 1280 1281 return (vmmops_setdesc(vcpu->cookie, reg, desc)); 1282 } 1283 1284 static void 1285 restore_guest_fpustate(struct vcpu *vcpu) 1286 { 1287 1288 /* flush host state to the pcb */ 1289 fpuexit(curthread); 1290 1291 /* restore guest FPU state */ 1292 fpu_stop_emulating(); 1293 fpurestore(vcpu->guestfpu); 1294 1295 /* restore guest XCR0 if XSAVE is enabled in the host */ 1296 if (rcr4() & CR4_XSAVE) 1297 load_xcr(0, vcpu->guest_xcr0); 1298 1299 /* 1300 * The FPU is now "dirty" with the guest's state so turn on emulation 1301 * to trap any access to the FPU by the host. 1302 */ 1303 fpu_start_emulating(); 1304 } 1305 1306 static void 1307 save_guest_fpustate(struct vcpu *vcpu) 1308 { 1309 1310 if ((rcr0() & CR0_TS) == 0) 1311 panic("fpu emulation not enabled in host!"); 1312 1313 /* save guest XCR0 and restore host XCR0 */ 1314 if (rcr4() & CR4_XSAVE) { 1315 vcpu->guest_xcr0 = rxcr(0); 1316 load_xcr(0, vmm_get_host_xcr0()); 1317 } 1318 1319 /* save guest FPU state */ 1320 fpu_stop_emulating(); 1321 fpusave(vcpu->guestfpu); 1322 fpu_start_emulating(); 1323 } 1324 1325 static VMM_STAT(VCPU_IDLE_TICKS, "number of ticks vcpu was idle"); 1326 1327 static int 1328 vcpu_set_state_locked(struct vcpu *vcpu, enum vcpu_state newstate, 1329 bool from_idle) 1330 { 1331 int error; 1332 1333 vcpu_assert_locked(vcpu); 1334 1335 /* 1336 * State transitions from the vmmdev_ioctl() must always begin from 1337 * the VCPU_IDLE state. This guarantees that there is only a single 1338 * ioctl() operating on a vcpu at any point. 1339 */ 1340 if (from_idle) { 1341 while (vcpu->state != VCPU_IDLE) { 1342 vcpu->reqidle = 1; 1343 vcpu_notify_event_locked(vcpu, false); 1344 VMM_CTR1(vcpu, "vcpu state change from %s to " 1345 "idle requested", vcpu_state2str(vcpu->state)); 1346 msleep_spin(&vcpu->state, &vcpu->mtx, "vmstat", hz); 1347 } 1348 } else { 1349 KASSERT(vcpu->state != VCPU_IDLE, ("invalid transition from " 1350 "vcpu idle state")); 1351 } 1352 1353 if (vcpu->state == VCPU_RUNNING) { 1354 KASSERT(vcpu->hostcpu == curcpu, ("curcpu %d and hostcpu %d " 1355 "mismatch for running vcpu", curcpu, vcpu->hostcpu)); 1356 } else { 1357 KASSERT(vcpu->hostcpu == NOCPU, ("Invalid hostcpu %d for a " 1358 "vcpu that is not running", vcpu->hostcpu)); 1359 } 1360 1361 /* 1362 * The following state transitions are allowed: 1363 * IDLE -> FROZEN -> IDLE 1364 * FROZEN -> RUNNING -> FROZEN 1365 * FROZEN -> SLEEPING -> FROZEN 1366 */ 1367 switch (vcpu->state) { 1368 case VCPU_IDLE: 1369 case VCPU_RUNNING: 1370 case VCPU_SLEEPING: 1371 error = (newstate != VCPU_FROZEN); 1372 break; 1373 case VCPU_FROZEN: 1374 error = (newstate == VCPU_FROZEN); 1375 break; 1376 default: 1377 error = 1; 1378 break; 1379 } 1380 1381 if (error) 1382 return (EBUSY); 1383 1384 VMM_CTR2(vcpu, "vcpu state changed from %s to %s", 1385 vcpu_state2str(vcpu->state), vcpu_state2str(newstate)); 1386 1387 vcpu->state = newstate; 1388 if (newstate == VCPU_RUNNING) 1389 vcpu->hostcpu = curcpu; 1390 else 1391 vcpu->hostcpu = NOCPU; 1392 1393 if (newstate == VCPU_IDLE) 1394 wakeup(&vcpu->state); 1395 1396 return (0); 1397 } 1398 1399 static void 1400 vcpu_require_state(struct vcpu *vcpu, enum vcpu_state newstate) 1401 { 1402 int error; 1403 1404 if ((error = vcpu_set_state(vcpu, newstate, false)) != 0) 1405 panic("Error %d setting state to %d\n", error, newstate); 1406 } 1407 1408 static void 1409 vcpu_require_state_locked(struct vcpu *vcpu, enum vcpu_state newstate) 1410 { 1411 int error; 1412 1413 if ((error = vcpu_set_state_locked(vcpu, newstate, false)) != 0) 1414 panic("Error %d setting state to %d", error, newstate); 1415 } 1416 1417 static int 1418 vm_handle_rendezvous(struct vcpu *vcpu) 1419 { 1420 struct vm *vm = vcpu->vm; 1421 struct thread *td; 1422 int error, vcpuid; 1423 1424 error = 0; 1425 vcpuid = vcpu->vcpuid; 1426 td = curthread; 1427 mtx_lock(&vm->rendezvous_mtx); 1428 while (vm->rendezvous_func != NULL) { 1429 /* 'rendezvous_req_cpus' must be a subset of 'active_cpus' */ 1430 CPU_AND(&vm->rendezvous_req_cpus, &vm->rendezvous_req_cpus, &vm->active_cpus); 1431 1432 if (CPU_ISSET(vcpuid, &vm->rendezvous_req_cpus) && 1433 !CPU_ISSET(vcpuid, &vm->rendezvous_done_cpus)) { 1434 VMM_CTR0(vcpu, "Calling rendezvous func"); 1435 (*vm->rendezvous_func)(vcpu, vm->rendezvous_arg); 1436 CPU_SET(vcpuid, &vm->rendezvous_done_cpus); 1437 } 1438 if (CPU_CMP(&vm->rendezvous_req_cpus, 1439 &vm->rendezvous_done_cpus) == 0) { 1440 VMM_CTR0(vcpu, "Rendezvous completed"); 1441 CPU_ZERO(&vm->rendezvous_req_cpus); 1442 vm->rendezvous_func = NULL; 1443 wakeup(&vm->rendezvous_func); 1444 break; 1445 } 1446 VMM_CTR0(vcpu, "Wait for rendezvous completion"); 1447 mtx_sleep(&vm->rendezvous_func, &vm->rendezvous_mtx, 0, 1448 "vmrndv", hz); 1449 if (td_ast_pending(td, TDA_SUSPEND)) { 1450 mtx_unlock(&vm->rendezvous_mtx); 1451 error = thread_check_susp(td, true); 1452 if (error != 0) 1453 return (error); 1454 mtx_lock(&vm->rendezvous_mtx); 1455 } 1456 } 1457 mtx_unlock(&vm->rendezvous_mtx); 1458 return (0); 1459 } 1460 1461 /* 1462 * Emulate a guest 'hlt' by sleeping until the vcpu is ready to run. 1463 */ 1464 static int 1465 vm_handle_hlt(struct vcpu *vcpu, bool intr_disabled, bool *retu) 1466 { 1467 struct vm *vm = vcpu->vm; 1468 const char *wmesg; 1469 struct thread *td; 1470 int error, t, vcpuid, vcpu_halted, vm_halted; 1471 1472 vcpuid = vcpu->vcpuid; 1473 vcpu_halted = 0; 1474 vm_halted = 0; 1475 error = 0; 1476 td = curthread; 1477 1478 KASSERT(!CPU_ISSET(vcpuid, &vm->halted_cpus), ("vcpu already halted")); 1479 1480 vcpu_lock(vcpu); 1481 while (1) { 1482 /* 1483 * Do a final check for pending NMI or interrupts before 1484 * really putting this thread to sleep. Also check for 1485 * software events that would cause this vcpu to wakeup. 1486 * 1487 * These interrupts/events could have happened after the 1488 * vcpu returned from vmmops_run() and before it acquired the 1489 * vcpu lock above. 1490 */ 1491 if (vm->rendezvous_func != NULL || vm->suspend || vcpu->reqidle) 1492 break; 1493 if (vm_nmi_pending(vcpu)) 1494 break; 1495 if (!intr_disabled) { 1496 if (vm_extint_pending(vcpu) || 1497 vlapic_pending_intr(vcpu->vlapic, NULL)) { 1498 break; 1499 } 1500 } 1501 1502 /* Don't go to sleep if the vcpu thread needs to yield */ 1503 if (vcpu_should_yield(vcpu)) 1504 break; 1505 1506 if (vcpu_debugged(vcpu)) 1507 break; 1508 1509 /* 1510 * Some Linux guests implement "halt" by having all vcpus 1511 * execute HLT with interrupts disabled. 'halted_cpus' keeps 1512 * track of the vcpus that have entered this state. When all 1513 * vcpus enter the halted state the virtual machine is halted. 1514 */ 1515 if (intr_disabled) { 1516 wmesg = "vmhalt"; 1517 VMM_CTR0(vcpu, "Halted"); 1518 if (!vcpu_halted && halt_detection_enabled) { 1519 vcpu_halted = 1; 1520 CPU_SET_ATOMIC(vcpuid, &vm->halted_cpus); 1521 } 1522 if (CPU_CMP(&vm->halted_cpus, &vm->active_cpus) == 0) { 1523 vm_halted = 1; 1524 break; 1525 } 1526 } else { 1527 wmesg = "vmidle"; 1528 } 1529 1530 t = ticks; 1531 vcpu_require_state_locked(vcpu, VCPU_SLEEPING); 1532 /* 1533 * XXX msleep_spin() cannot be interrupted by signals so 1534 * wake up periodically to check pending signals. 1535 */ 1536 msleep_spin(vcpu, &vcpu->mtx, wmesg, hz); 1537 vcpu_require_state_locked(vcpu, VCPU_FROZEN); 1538 vmm_stat_incr(vcpu, VCPU_IDLE_TICKS, ticks - t); 1539 if (td_ast_pending(td, TDA_SUSPEND)) { 1540 vcpu_unlock(vcpu); 1541 error = thread_check_susp(td, false); 1542 if (error != 0) { 1543 if (vcpu_halted) { 1544 CPU_CLR_ATOMIC(vcpuid, 1545 &vm->halted_cpus); 1546 } 1547 return (error); 1548 } 1549 vcpu_lock(vcpu); 1550 } 1551 } 1552 1553 if (vcpu_halted) 1554 CPU_CLR_ATOMIC(vcpuid, &vm->halted_cpus); 1555 1556 vcpu_unlock(vcpu); 1557 1558 if (vm_halted) 1559 vm_suspend(vm, VM_SUSPEND_HALT); 1560 1561 return (0); 1562 } 1563 1564 static int 1565 vm_handle_paging(struct vcpu *vcpu, bool *retu) 1566 { 1567 struct vm *vm = vcpu->vm; 1568 int rv, ftype; 1569 struct vm_map *map; 1570 struct vm_exit *vme; 1571 1572 vme = &vcpu->exitinfo; 1573 1574 KASSERT(vme->inst_length == 0, ("%s: invalid inst_length %d", 1575 __func__, vme->inst_length)); 1576 1577 ftype = vme->u.paging.fault_type; 1578 KASSERT(ftype == VM_PROT_READ || 1579 ftype == VM_PROT_WRITE || ftype == VM_PROT_EXECUTE, 1580 ("vm_handle_paging: invalid fault_type %d", ftype)); 1581 1582 if (ftype == VM_PROT_READ || ftype == VM_PROT_WRITE) { 1583 rv = pmap_emulate_accessed_dirty(vmspace_pmap(vm->vmspace), 1584 vme->u.paging.gpa, ftype); 1585 if (rv == 0) { 1586 VMM_CTR2(vcpu, "%s bit emulation for gpa %#lx", 1587 ftype == VM_PROT_READ ? "accessed" : "dirty", 1588 vme->u.paging.gpa); 1589 goto done; 1590 } 1591 } 1592 1593 map = &vm->vmspace->vm_map; 1594 rv = vm_fault(map, vme->u.paging.gpa, ftype, VM_FAULT_NORMAL, NULL); 1595 1596 VMM_CTR3(vcpu, "vm_handle_paging rv = %d, gpa = %#lx, " 1597 "ftype = %d", rv, vme->u.paging.gpa, ftype); 1598 1599 if (rv != KERN_SUCCESS) 1600 return (EFAULT); 1601 done: 1602 return (0); 1603 } 1604 1605 static int 1606 vm_handle_inst_emul(struct vcpu *vcpu, bool *retu) 1607 { 1608 struct vie *vie; 1609 struct vm_exit *vme; 1610 uint64_t gla, gpa, cs_base; 1611 struct vm_guest_paging *paging; 1612 mem_region_read_t mread; 1613 mem_region_write_t mwrite; 1614 enum vm_cpu_mode cpu_mode; 1615 int cs_d, error, fault; 1616 1617 vme = &vcpu->exitinfo; 1618 1619 KASSERT(vme->inst_length == 0, ("%s: invalid inst_length %d", 1620 __func__, vme->inst_length)); 1621 1622 gla = vme->u.inst_emul.gla; 1623 gpa = vme->u.inst_emul.gpa; 1624 cs_base = vme->u.inst_emul.cs_base; 1625 cs_d = vme->u.inst_emul.cs_d; 1626 vie = &vme->u.inst_emul.vie; 1627 paging = &vme->u.inst_emul.paging; 1628 cpu_mode = paging->cpu_mode; 1629 1630 VMM_CTR1(vcpu, "inst_emul fault accessing gpa %#lx", gpa); 1631 1632 /* Fetch, decode and emulate the faulting instruction */ 1633 if (vie->num_valid == 0) { 1634 error = vmm_fetch_instruction(vcpu, paging, vme->rip + cs_base, 1635 VIE_INST_SIZE, vie, &fault); 1636 } else { 1637 /* 1638 * The instruction bytes have already been copied into 'vie' 1639 */ 1640 error = fault = 0; 1641 } 1642 if (error || fault) 1643 return (error); 1644 1645 if (vmm_decode_instruction(vcpu, gla, cpu_mode, cs_d, vie) != 0) { 1646 VMM_CTR1(vcpu, "Error decoding instruction at %#lx", 1647 vme->rip + cs_base); 1648 *retu = true; /* dump instruction bytes in userspace */ 1649 return (0); 1650 } 1651 1652 /* 1653 * Update 'nextrip' based on the length of the emulated instruction. 1654 */ 1655 vme->inst_length = vie->num_processed; 1656 vcpu->nextrip += vie->num_processed; 1657 VMM_CTR1(vcpu, "nextrip updated to %#lx after instruction decoding", 1658 vcpu->nextrip); 1659 1660 /* return to userland unless this is an in-kernel emulated device */ 1661 if (gpa >= DEFAULT_APIC_BASE && gpa < DEFAULT_APIC_BASE + PAGE_SIZE) { 1662 mread = lapic_mmio_read; 1663 mwrite = lapic_mmio_write; 1664 } else if (gpa >= VIOAPIC_BASE && gpa < VIOAPIC_BASE + VIOAPIC_SIZE) { 1665 mread = vioapic_mmio_read; 1666 mwrite = vioapic_mmio_write; 1667 } else if (gpa >= VHPET_BASE && gpa < VHPET_BASE + VHPET_SIZE) { 1668 mread = vhpet_mmio_read; 1669 mwrite = vhpet_mmio_write; 1670 } else { 1671 *retu = true; 1672 return (0); 1673 } 1674 1675 error = vmm_emulate_instruction(vcpu, gpa, vie, paging, mread, mwrite, 1676 retu); 1677 1678 return (error); 1679 } 1680 1681 static int 1682 vm_handle_suspend(struct vcpu *vcpu, bool *retu) 1683 { 1684 struct vm *vm = vcpu->vm; 1685 int error, i; 1686 struct thread *td; 1687 1688 error = 0; 1689 td = curthread; 1690 1691 CPU_SET_ATOMIC(vcpu->vcpuid, &vm->suspended_cpus); 1692 1693 /* 1694 * Wait until all 'active_cpus' have suspended themselves. 1695 * 1696 * Since a VM may be suspended at any time including when one or 1697 * more vcpus are doing a rendezvous we need to call the rendezvous 1698 * handler while we are waiting to prevent a deadlock. 1699 */ 1700 vcpu_lock(vcpu); 1701 while (error == 0) { 1702 if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) { 1703 VMM_CTR0(vcpu, "All vcpus suspended"); 1704 break; 1705 } 1706 1707 if (vm->rendezvous_func == NULL) { 1708 VMM_CTR0(vcpu, "Sleeping during suspend"); 1709 vcpu_require_state_locked(vcpu, VCPU_SLEEPING); 1710 msleep_spin(vcpu, &vcpu->mtx, "vmsusp", hz); 1711 vcpu_require_state_locked(vcpu, VCPU_FROZEN); 1712 if (td_ast_pending(td, TDA_SUSPEND)) { 1713 vcpu_unlock(vcpu); 1714 error = thread_check_susp(td, false); 1715 vcpu_lock(vcpu); 1716 } 1717 } else { 1718 VMM_CTR0(vcpu, "Rendezvous during suspend"); 1719 vcpu_unlock(vcpu); 1720 error = vm_handle_rendezvous(vcpu); 1721 vcpu_lock(vcpu); 1722 } 1723 } 1724 vcpu_unlock(vcpu); 1725 1726 /* 1727 * Wakeup the other sleeping vcpus and return to userspace. 1728 */ 1729 for (i = 0; i < vm->maxcpus; i++) { 1730 if (CPU_ISSET(i, &vm->suspended_cpus)) { 1731 vcpu_notify_event(vm_vcpu(vm, i), false); 1732 } 1733 } 1734 1735 *retu = true; 1736 return (error); 1737 } 1738 1739 static int 1740 vm_handle_reqidle(struct vcpu *vcpu, bool *retu) 1741 { 1742 vcpu_lock(vcpu); 1743 KASSERT(vcpu->reqidle, ("invalid vcpu reqidle %d", vcpu->reqidle)); 1744 vcpu->reqidle = 0; 1745 vcpu_unlock(vcpu); 1746 *retu = true; 1747 return (0); 1748 } 1749 1750 int 1751 vm_suspend(struct vm *vm, enum vm_suspend_how how) 1752 { 1753 int i; 1754 1755 if (how <= VM_SUSPEND_NONE || how >= VM_SUSPEND_LAST) 1756 return (EINVAL); 1757 1758 if (atomic_cmpset_int(&vm->suspend, 0, how) == 0) { 1759 VM_CTR2(vm, "virtual machine already suspended %d/%d", 1760 vm->suspend, how); 1761 return (EALREADY); 1762 } 1763 1764 VM_CTR1(vm, "virtual machine successfully suspended %d", how); 1765 1766 /* 1767 * Notify all active vcpus that they are now suspended. 1768 */ 1769 for (i = 0; i < vm->maxcpus; i++) { 1770 if (CPU_ISSET(i, &vm->active_cpus)) 1771 vcpu_notify_event(vm_vcpu(vm, i), false); 1772 } 1773 1774 return (0); 1775 } 1776 1777 void 1778 vm_exit_suspended(struct vcpu *vcpu, uint64_t rip) 1779 { 1780 struct vm *vm = vcpu->vm; 1781 struct vm_exit *vmexit; 1782 1783 KASSERT(vm->suspend > VM_SUSPEND_NONE && vm->suspend < VM_SUSPEND_LAST, 1784 ("vm_exit_suspended: invalid suspend type %d", vm->suspend)); 1785 1786 vmexit = vm_exitinfo(vcpu); 1787 vmexit->rip = rip; 1788 vmexit->inst_length = 0; 1789 vmexit->exitcode = VM_EXITCODE_SUSPENDED; 1790 vmexit->u.suspended.how = vm->suspend; 1791 } 1792 1793 void 1794 vm_exit_debug(struct vcpu *vcpu, uint64_t rip) 1795 { 1796 struct vm_exit *vmexit; 1797 1798 vmexit = vm_exitinfo(vcpu); 1799 vmexit->rip = rip; 1800 vmexit->inst_length = 0; 1801 vmexit->exitcode = VM_EXITCODE_DEBUG; 1802 } 1803 1804 void 1805 vm_exit_rendezvous(struct vcpu *vcpu, uint64_t rip) 1806 { 1807 struct vm_exit *vmexit; 1808 1809 vmexit = vm_exitinfo(vcpu); 1810 vmexit->rip = rip; 1811 vmexit->inst_length = 0; 1812 vmexit->exitcode = VM_EXITCODE_RENDEZVOUS; 1813 vmm_stat_incr(vcpu, VMEXIT_RENDEZVOUS, 1); 1814 } 1815 1816 void 1817 vm_exit_reqidle(struct vcpu *vcpu, uint64_t rip) 1818 { 1819 struct vm_exit *vmexit; 1820 1821 vmexit = vm_exitinfo(vcpu); 1822 vmexit->rip = rip; 1823 vmexit->inst_length = 0; 1824 vmexit->exitcode = VM_EXITCODE_REQIDLE; 1825 vmm_stat_incr(vcpu, VMEXIT_REQIDLE, 1); 1826 } 1827 1828 void 1829 vm_exit_astpending(struct vcpu *vcpu, uint64_t rip) 1830 { 1831 struct vm_exit *vmexit; 1832 1833 vmexit = vm_exitinfo(vcpu); 1834 vmexit->rip = rip; 1835 vmexit->inst_length = 0; 1836 vmexit->exitcode = VM_EXITCODE_BOGUS; 1837 vmm_stat_incr(vcpu, VMEXIT_ASTPENDING, 1); 1838 } 1839 1840 int 1841 vm_run(struct vcpu *vcpu, struct vm_exit *vme_user) 1842 { 1843 struct vm *vm = vcpu->vm; 1844 struct vm_eventinfo evinfo; 1845 int error, vcpuid; 1846 struct pcb *pcb; 1847 uint64_t tscval; 1848 struct vm_exit *vme; 1849 bool retu, intr_disabled; 1850 pmap_t pmap; 1851 1852 vcpuid = vcpu->vcpuid; 1853 1854 if (!CPU_ISSET(vcpuid, &vm->active_cpus)) 1855 return (EINVAL); 1856 1857 if (CPU_ISSET(vcpuid, &vm->suspended_cpus)) 1858 return (EINVAL); 1859 1860 pmap = vmspace_pmap(vm->vmspace); 1861 vme = &vcpu->exitinfo; 1862 evinfo.rptr = &vm->rendezvous_req_cpus; 1863 evinfo.sptr = &vm->suspend; 1864 evinfo.iptr = &vcpu->reqidle; 1865 restart: 1866 critical_enter(); 1867 1868 KASSERT(!CPU_ISSET(curcpu, &pmap->pm_active), 1869 ("vm_run: absurd pm_active")); 1870 1871 tscval = rdtsc(); 1872 1873 pcb = PCPU_GET(curpcb); 1874 set_pcb_flags(pcb, PCB_FULL_IRET); 1875 1876 restore_guest_fpustate(vcpu); 1877 1878 vcpu_require_state(vcpu, VCPU_RUNNING); 1879 error = vmmops_run(vcpu->cookie, vcpu->nextrip, pmap, &evinfo); 1880 vcpu_require_state(vcpu, VCPU_FROZEN); 1881 1882 save_guest_fpustate(vcpu); 1883 1884 vmm_stat_incr(vcpu, VCPU_TOTAL_RUNTIME, rdtsc() - tscval); 1885 1886 critical_exit(); 1887 1888 if (error == 0) { 1889 retu = false; 1890 vcpu->nextrip = vme->rip + vme->inst_length; 1891 switch (vme->exitcode) { 1892 case VM_EXITCODE_REQIDLE: 1893 error = vm_handle_reqidle(vcpu, &retu); 1894 break; 1895 case VM_EXITCODE_SUSPENDED: 1896 error = vm_handle_suspend(vcpu, &retu); 1897 break; 1898 case VM_EXITCODE_IOAPIC_EOI: 1899 vioapic_process_eoi(vm, vme->u.ioapic_eoi.vector); 1900 break; 1901 case VM_EXITCODE_RENDEZVOUS: 1902 error = vm_handle_rendezvous(vcpu); 1903 break; 1904 case VM_EXITCODE_HLT: 1905 intr_disabled = ((vme->u.hlt.rflags & PSL_I) == 0); 1906 error = vm_handle_hlt(vcpu, intr_disabled, &retu); 1907 break; 1908 case VM_EXITCODE_PAGING: 1909 error = vm_handle_paging(vcpu, &retu); 1910 break; 1911 case VM_EXITCODE_INST_EMUL: 1912 error = vm_handle_inst_emul(vcpu, &retu); 1913 break; 1914 case VM_EXITCODE_INOUT: 1915 case VM_EXITCODE_INOUT_STR: 1916 error = vm_handle_inout(vcpu, vme, &retu); 1917 break; 1918 case VM_EXITCODE_MONITOR: 1919 case VM_EXITCODE_MWAIT: 1920 case VM_EXITCODE_VMINSN: 1921 vm_inject_ud(vcpu); 1922 break; 1923 default: 1924 retu = true; /* handled in userland */ 1925 break; 1926 } 1927 } 1928 1929 /* 1930 * VM_EXITCODE_INST_EMUL could access the apic which could transform the 1931 * exit code into VM_EXITCODE_IPI. 1932 */ 1933 if (error == 0 && vme->exitcode == VM_EXITCODE_IPI) { 1934 retu = false; 1935 error = vm_handle_ipi(vcpu, vme, &retu); 1936 } 1937 1938 if (error == 0 && retu == false) 1939 goto restart; 1940 1941 vmm_stat_incr(vcpu, VMEXIT_USERSPACE, 1); 1942 VMM_CTR2(vcpu, "retu %d/%d", error, vme->exitcode); 1943 1944 /* copy the exit information */ 1945 *vme_user = *vme; 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 } 2865 2866 done: 2867 return (ret); 2868 } 2869 2870 static int 2871 vm_snapshot_vm(struct vm *vm, struct vm_snapshot_meta *meta) 2872 { 2873 int ret; 2874 2875 ret = vm_snapshot_vcpus(vm, meta); 2876 if (ret != 0) 2877 goto done; 2878 2879 SNAPSHOT_VAR_OR_LEAVE(vm->startup_cpus, meta, ret, done); 2880 done: 2881 return (ret); 2882 } 2883 2884 static int 2885 vm_snapshot_vcpu(struct vm *vm, struct vm_snapshot_meta *meta) 2886 { 2887 int error; 2888 struct vcpu *vcpu; 2889 uint16_t i, maxcpus; 2890 2891 error = 0; 2892 2893 maxcpus = vm_get_maxcpus(vm); 2894 for (i = 0; i < maxcpus; i++) { 2895 vcpu = vm->vcpu[i]; 2896 if (vcpu == NULL) 2897 continue; 2898 2899 error = vmmops_vcpu_snapshot(vcpu->cookie, meta); 2900 if (error != 0) { 2901 printf("%s: failed to snapshot vmcs/vmcb data for " 2902 "vCPU: %d; error: %d\n", __func__, i, error); 2903 goto done; 2904 } 2905 } 2906 2907 done: 2908 return (error); 2909 } 2910 2911 /* 2912 * Save kernel-side structures to user-space for snapshotting. 2913 */ 2914 int 2915 vm_snapshot_req(struct vm *vm, struct vm_snapshot_meta *meta) 2916 { 2917 int ret = 0; 2918 2919 switch (meta->dev_req) { 2920 case STRUCT_VMX: 2921 ret = vmmops_snapshot(vm->cookie, meta); 2922 break; 2923 case STRUCT_VMCX: 2924 ret = vm_snapshot_vcpu(vm, meta); 2925 break; 2926 case STRUCT_VM: 2927 ret = vm_snapshot_vm(vm, meta); 2928 break; 2929 case STRUCT_VIOAPIC: 2930 ret = vioapic_snapshot(vm_ioapic(vm), meta); 2931 break; 2932 case STRUCT_VLAPIC: 2933 ret = vlapic_snapshot(vm, meta); 2934 break; 2935 case STRUCT_VHPET: 2936 ret = vhpet_snapshot(vm_hpet(vm), meta); 2937 break; 2938 case STRUCT_VATPIC: 2939 ret = vatpic_snapshot(vm_atpic(vm), meta); 2940 break; 2941 case STRUCT_VATPIT: 2942 ret = vatpit_snapshot(vm_atpit(vm), meta); 2943 break; 2944 case STRUCT_VPMTMR: 2945 ret = vpmtmr_snapshot(vm_pmtmr(vm), meta); 2946 break; 2947 case STRUCT_VRTC: 2948 ret = vrtc_snapshot(vm_rtc(vm), meta); 2949 break; 2950 default: 2951 printf("%s: failed to find the requested type %#x\n", 2952 __func__, meta->dev_req); 2953 ret = (EINVAL); 2954 } 2955 return (ret); 2956 } 2957 2958 void 2959 vm_set_tsc_offset(struct vcpu *vcpu, uint64_t offset) 2960 { 2961 vcpu->tsc_offset = offset; 2962 } 2963 2964 int 2965 vm_restore_time(struct vm *vm) 2966 { 2967 int error; 2968 uint64_t now; 2969 struct vcpu *vcpu; 2970 uint16_t i, maxcpus; 2971 2972 now = rdtsc(); 2973 2974 error = vhpet_restore_time(vm_hpet(vm)); 2975 if (error) 2976 return (error); 2977 2978 maxcpus = vm_get_maxcpus(vm); 2979 for (i = 0; i < maxcpus; i++) { 2980 vcpu = vm->vcpu[i]; 2981 if (vcpu == NULL) 2982 continue; 2983 2984 error = vmmops_restore_tsc(vcpu->cookie, 2985 vcpu->tsc_offset - now); 2986 if (error) 2987 return (error); 2988 } 2989 2990 return (0); 2991 } 2992 #endif 2993