1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause 3 * 4 * Copyright (c) 2011 NetApp, Inc. 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 16 * THIS SOFTWARE IS PROVIDED BY NETAPP, INC ``AS IS'' AND 17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 19 * ARE DISCLAIMED. IN NO EVENT SHALL NETAPP, INC OR CONTRIBUTORS BE LIABLE 20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 26 * SUCH DAMAGE. 27 */ 28 29 #include <sys/cdefs.h> 30 #include "opt_bhyve_snapshot.h" 31 32 #include <sys/param.h> 33 #include <sys/systm.h> 34 #include <sys/kernel.h> 35 #include <sys/module.h> 36 #include <sys/sysctl.h> 37 #include <sys/malloc.h> 38 #include <sys/pcpu.h> 39 #include <sys/lock.h> 40 #include <sys/mutex.h> 41 #include <sys/proc.h> 42 #include <sys/rwlock.h> 43 #include <sys/sched.h> 44 #include <sys/smp.h> 45 #include <sys/sx.h> 46 #include <sys/vnode.h> 47 48 #include <vm/vm.h> 49 #include <vm/vm_param.h> 50 #include <vm/vm_extern.h> 51 #include <vm/vm_object.h> 52 #include <vm/vm_page.h> 53 #include <vm/pmap.h> 54 #include <vm/vm_map.h> 55 #include <vm/vm_pager.h> 56 #include <vm/vm_kern.h> 57 #include <vm/vnode_pager.h> 58 #include <vm/swap_pager.h> 59 #include <vm/uma.h> 60 61 #include <machine/cpu.h> 62 #include <machine/pcb.h> 63 #include <machine/smp.h> 64 #include <machine/md_var.h> 65 #include <x86/psl.h> 66 #include <x86/apicreg.h> 67 #include <x86/ifunc.h> 68 69 #include <machine/vmm.h> 70 #include <machine/vmm_dev.h> 71 #include <machine/vmm_instruction_emul.h> 72 #include <machine/vmm_snapshot.h> 73 74 #include "vmm_ioport.h" 75 #include "vmm_ktr.h" 76 #include "vmm_host.h" 77 #include "vmm_mem.h" 78 #include "vmm_util.h" 79 #include "vatpic.h" 80 #include "vatpit.h" 81 #include "vhpet.h" 82 #include "vioapic.h" 83 #include "vlapic.h" 84 #include "vpmtmr.h" 85 #include "vrtc.h" 86 #include "vmm_stat.h" 87 #include "vmm_lapic.h" 88 89 #include "io/ppt.h" 90 #include "io/iommu.h" 91 92 struct vlapic; 93 94 /* 95 * Initialization: 96 * (a) allocated when vcpu is created 97 * (i) initialized when vcpu is created and when it is reinitialized 98 * (o) initialized the first time the vcpu is created 99 * (x) initialized before use 100 */ 101 struct vcpu { 102 struct mtx mtx; /* (o) protects 'state' and 'hostcpu' */ 103 enum vcpu_state state; /* (o) vcpu state */ 104 int vcpuid; /* (o) */ 105 int hostcpu; /* (o) vcpu's host cpu */ 106 int reqidle; /* (i) request vcpu to idle */ 107 struct vm *vm; /* (o) */ 108 void *cookie; /* (i) cpu-specific data */ 109 struct vlapic *vlapic; /* (i) APIC device model */ 110 enum x2apic_state x2apic_state; /* (i) APIC mode */ 111 uint64_t exitintinfo; /* (i) events pending at VM exit */ 112 int nmi_pending; /* (i) NMI pending */ 113 int extint_pending; /* (i) INTR pending */ 114 int exception_pending; /* (i) exception pending */ 115 int exc_vector; /* (x) exception collateral */ 116 int exc_errcode_valid; 117 uint32_t exc_errcode; 118 struct savefpu *guestfpu; /* (a,i) guest fpu state */ 119 uint64_t guest_xcr0; /* (i) guest %xcr0 register */ 120 void *stats; /* (a,i) statistics */ 121 struct vm_exit exitinfo; /* (x) exit reason and collateral */ 122 cpuset_t exitinfo_cpuset; /* (x) storage for vmexit handlers */ 123 uint64_t nextrip; /* (x) next instruction to execute */ 124 uint64_t tsc_offset; /* (o) TSC offsetting */ 125 }; 126 127 #define vcpu_lock_init(v) mtx_init(&((v)->mtx), "vcpu lock", 0, MTX_SPIN) 128 #define vcpu_lock_destroy(v) mtx_destroy(&((v)->mtx)) 129 #define vcpu_lock(v) mtx_lock_spin(&((v)->mtx)) 130 #define vcpu_unlock(v) mtx_unlock_spin(&((v)->mtx)) 131 #define vcpu_assert_locked(v) mtx_assert(&((v)->mtx), MA_OWNED) 132 133 struct mem_seg { 134 size_t len; 135 bool sysmem; 136 struct vm_object *object; 137 }; 138 #define VM_MAX_MEMSEGS 4 139 140 struct mem_map { 141 vm_paddr_t gpa; 142 size_t len; 143 vm_ooffset_t segoff; 144 int segid; 145 int prot; 146 int flags; 147 }; 148 #define VM_MAX_MEMMAPS 8 149 150 /* 151 * Initialization: 152 * (o) initialized the first time the VM is created 153 * (i) initialized when VM is created and when it is reinitialized 154 * (x) initialized before use 155 * 156 * Locking: 157 * [m] mem_segs_lock 158 * [r] rendezvous_mtx 159 * [v] reads require one frozen vcpu, writes require freezing all vcpus 160 */ 161 struct vm { 162 void *cookie; /* (i) cpu-specific data */ 163 void *iommu; /* (x) iommu-specific data */ 164 struct vhpet *vhpet; /* (i) virtual HPET */ 165 struct vioapic *vioapic; /* (i) virtual ioapic */ 166 struct vatpic *vatpic; /* (i) virtual atpic */ 167 struct vatpit *vatpit; /* (i) virtual atpit */ 168 struct vpmtmr *vpmtmr; /* (i) virtual ACPI PM timer */ 169 struct vrtc *vrtc; /* (o) virtual RTC */ 170 volatile cpuset_t active_cpus; /* (i) active vcpus */ 171 volatile cpuset_t debug_cpus; /* (i) vcpus stopped for debug */ 172 cpuset_t startup_cpus; /* (i) [r] waiting for startup */ 173 int suspend; /* (i) stop VM execution */ 174 bool dying; /* (o) is dying */ 175 volatile cpuset_t suspended_cpus; /* (i) suspended vcpus */ 176 volatile cpuset_t halted_cpus; /* (x) cpus in a hard halt */ 177 cpuset_t rendezvous_req_cpus; /* (x) [r] rendezvous requested */ 178 cpuset_t rendezvous_done_cpus; /* (x) [r] rendezvous finished */ 179 void *rendezvous_arg; /* (x) [r] rendezvous func/arg */ 180 vm_rendezvous_func_t rendezvous_func; 181 struct mtx rendezvous_mtx; /* (o) rendezvous lock */ 182 struct mem_map mem_maps[VM_MAX_MEMMAPS]; /* (i) [m+v] guest address space */ 183 struct mem_seg mem_segs[VM_MAX_MEMSEGS]; /* (o) [m+v] guest memory regions */ 184 struct vmspace *vmspace; /* (o) guest's address space */ 185 char name[VM_MAX_NAMELEN+1]; /* (o) virtual machine name */ 186 struct vcpu **vcpu; /* (o) guest vcpus */ 187 /* The following describe the vm cpu topology */ 188 uint16_t sockets; /* (o) num of sockets */ 189 uint16_t cores; /* (o) num of cores/socket */ 190 uint16_t threads; /* (o) num of threads/core */ 191 uint16_t maxcpus; /* (o) max pluggable cpus */ 192 struct sx mem_segs_lock; /* (o) */ 193 struct sx vcpus_init_lock; /* (o) */ 194 }; 195 196 #define VMM_CTR0(vcpu, format) \ 197 VCPU_CTR0((vcpu)->vm, (vcpu)->vcpuid, format) 198 199 #define VMM_CTR1(vcpu, format, p1) \ 200 VCPU_CTR1((vcpu)->vm, (vcpu)->vcpuid, format, p1) 201 202 #define VMM_CTR2(vcpu, format, p1, p2) \ 203 VCPU_CTR2((vcpu)->vm, (vcpu)->vcpuid, format, p1, p2) 204 205 #define VMM_CTR3(vcpu, format, p1, p2, p3) \ 206 VCPU_CTR3((vcpu)->vm, (vcpu)->vcpuid, format, p1, p2, p3) 207 208 #define VMM_CTR4(vcpu, format, p1, p2, p3, p4) \ 209 VCPU_CTR4((vcpu)->vm, (vcpu)->vcpuid, format, p1, p2, p3, p4) 210 211 static int vmm_initialized; 212 213 static void vmmops_panic(void); 214 215 static void 216 vmmops_panic(void) 217 { 218 panic("vmm_ops func called when !vmm_is_intel() && !vmm_is_svm()"); 219 } 220 221 #define DEFINE_VMMOPS_IFUNC(ret_type, opname, args) \ 222 DEFINE_IFUNC(static, ret_type, vmmops_##opname, args) \ 223 { \ 224 if (vmm_is_intel()) \ 225 return (vmm_ops_intel.opname); \ 226 else if (vmm_is_svm()) \ 227 return (vmm_ops_amd.opname); \ 228 else \ 229 return ((ret_type (*)args)vmmops_panic); \ 230 } 231 232 DEFINE_VMMOPS_IFUNC(int, modinit, (int ipinum)) 233 DEFINE_VMMOPS_IFUNC(int, modcleanup, (void)) 234 DEFINE_VMMOPS_IFUNC(void, modresume, (void)) 235 DEFINE_VMMOPS_IFUNC(void *, init, (struct vm *vm, struct pmap *pmap)) 236 DEFINE_VMMOPS_IFUNC(int, run, (void *vcpui, register_t rip, struct pmap *pmap, 237 struct vm_eventinfo *info)) 238 DEFINE_VMMOPS_IFUNC(void, cleanup, (void *vmi)) 239 DEFINE_VMMOPS_IFUNC(void *, vcpu_init, (void *vmi, struct vcpu *vcpu, 240 int vcpu_id)) 241 DEFINE_VMMOPS_IFUNC(void, vcpu_cleanup, (void *vcpui)) 242 DEFINE_VMMOPS_IFUNC(int, getreg, (void *vcpui, int num, uint64_t *retval)) 243 DEFINE_VMMOPS_IFUNC(int, setreg, (void *vcpui, int num, uint64_t val)) 244 DEFINE_VMMOPS_IFUNC(int, getdesc, (void *vcpui, int num, struct seg_desc *desc)) 245 DEFINE_VMMOPS_IFUNC(int, setdesc, (void *vcpui, int num, struct seg_desc *desc)) 246 DEFINE_VMMOPS_IFUNC(int, getcap, (void *vcpui, int num, int *retval)) 247 DEFINE_VMMOPS_IFUNC(int, setcap, (void *vcpui, int num, int val)) 248 DEFINE_VMMOPS_IFUNC(struct vmspace *, vmspace_alloc, (vm_offset_t min, 249 vm_offset_t max)) 250 DEFINE_VMMOPS_IFUNC(void, vmspace_free, (struct vmspace *vmspace)) 251 DEFINE_VMMOPS_IFUNC(struct vlapic *, vlapic_init, (void *vcpui)) 252 DEFINE_VMMOPS_IFUNC(void, vlapic_cleanup, (struct vlapic *vlapic)) 253 #ifdef BHYVE_SNAPSHOT 254 DEFINE_VMMOPS_IFUNC(int, vcpu_snapshot, (void *vcpui, 255 struct vm_snapshot_meta *meta)) 256 DEFINE_VMMOPS_IFUNC(int, restore_tsc, (void *vcpui, uint64_t now)) 257 #endif 258 259 #define fpu_start_emulating() load_cr0(rcr0() | CR0_TS) 260 #define fpu_stop_emulating() clts() 261 262 SDT_PROVIDER_DEFINE(vmm); 263 264 static MALLOC_DEFINE(M_VM, "vm", "vm"); 265 266 /* statistics */ 267 static VMM_STAT(VCPU_TOTAL_RUNTIME, "vcpu total runtime"); 268 269 SYSCTL_NODE(_hw, OID_AUTO, vmm, CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, 270 NULL); 271 272 /* 273 * Halt the guest if all vcpus are executing a HLT instruction with 274 * interrupts disabled. 275 */ 276 static int halt_detection_enabled = 1; 277 SYSCTL_INT(_hw_vmm, OID_AUTO, halt_detection, CTLFLAG_RDTUN, 278 &halt_detection_enabled, 0, 279 "Halt VM if all vcpus execute HLT with interrupts disabled"); 280 281 static int vmm_ipinum; 282 SYSCTL_INT(_hw_vmm, OID_AUTO, ipinum, CTLFLAG_RD, &vmm_ipinum, 0, 283 "IPI vector used for vcpu notifications"); 284 285 static int trace_guest_exceptions; 286 SYSCTL_INT(_hw_vmm, OID_AUTO, trace_guest_exceptions, CTLFLAG_RDTUN, 287 &trace_guest_exceptions, 0, 288 "Trap into hypervisor on all guest exceptions and reflect them back"); 289 290 static int trap_wbinvd; 291 SYSCTL_INT(_hw_vmm, OID_AUTO, trap_wbinvd, CTLFLAG_RDTUN, &trap_wbinvd, 0, 292 "WBINVD triggers a VM-exit"); 293 294 u_int vm_maxcpu; 295 SYSCTL_UINT(_hw_vmm, OID_AUTO, maxcpu, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, 296 &vm_maxcpu, 0, "Maximum number of vCPUs"); 297 298 static void vm_free_memmap(struct vm *vm, int ident); 299 static bool sysmem_mapping(struct vm *vm, struct mem_map *mm); 300 static void vcpu_notify_event_locked(struct vcpu *vcpu, bool lapic_intr); 301 302 /* 303 * Upper limit on vm_maxcpu. Limited by use of uint16_t types for CPU 304 * counts as well as range of vpid values for VT-x and by the capacity 305 * of cpuset_t masks. The call to new_unrhdr() in vpid_init() in 306 * vmx.c requires 'vm_maxcpu + 1 <= 0xffff', hence the '- 1' below. 307 */ 308 #define VM_MAXCPU MIN(0xffff - 1, CPU_SETSIZE) 309 310 #ifdef KTR 311 static const char * 312 vcpu_state2str(enum vcpu_state state) 313 { 314 315 switch (state) { 316 case VCPU_IDLE: 317 return ("idle"); 318 case VCPU_FROZEN: 319 return ("frozen"); 320 case VCPU_RUNNING: 321 return ("running"); 322 case VCPU_SLEEPING: 323 return ("sleeping"); 324 default: 325 return ("unknown"); 326 } 327 } 328 #endif 329 330 static void 331 vcpu_cleanup(struct vcpu *vcpu, bool destroy) 332 { 333 vmmops_vlapic_cleanup(vcpu->vlapic); 334 vmmops_vcpu_cleanup(vcpu->cookie); 335 vcpu->cookie = NULL; 336 if (destroy) { 337 vmm_stat_free(vcpu->stats); 338 fpu_save_area_free(vcpu->guestfpu); 339 vcpu_lock_destroy(vcpu); 340 free(vcpu, M_VM); 341 } 342 } 343 344 static struct vcpu * 345 vcpu_alloc(struct vm *vm, int vcpu_id) 346 { 347 struct vcpu *vcpu; 348 349 KASSERT(vcpu_id >= 0 && vcpu_id < vm->maxcpus, 350 ("vcpu_init: invalid vcpu %d", vcpu_id)); 351 352 vcpu = malloc(sizeof(*vcpu), M_VM, M_WAITOK | M_ZERO); 353 vcpu_lock_init(vcpu); 354 vcpu->state = VCPU_IDLE; 355 vcpu->hostcpu = NOCPU; 356 vcpu->vcpuid = vcpu_id; 357 vcpu->vm = vm; 358 vcpu->guestfpu = fpu_save_area_alloc(); 359 vcpu->stats = vmm_stat_alloc(); 360 vcpu->tsc_offset = 0; 361 return (vcpu); 362 } 363 364 static void 365 vcpu_init(struct vcpu *vcpu) 366 { 367 vcpu->cookie = vmmops_vcpu_init(vcpu->vm->cookie, vcpu, vcpu->vcpuid); 368 vcpu->vlapic = vmmops_vlapic_init(vcpu->cookie); 369 vm_set_x2apic_state(vcpu, X2APIC_DISABLED); 370 vcpu->reqidle = 0; 371 vcpu->exitintinfo = 0; 372 vcpu->nmi_pending = 0; 373 vcpu->extint_pending = 0; 374 vcpu->exception_pending = 0; 375 vcpu->guest_xcr0 = XFEATURE_ENABLED_X87; 376 fpu_save_area_reset(vcpu->guestfpu); 377 vmm_stat_init(vcpu->stats); 378 } 379 380 int 381 vcpu_trace_exceptions(struct vcpu *vcpu) 382 { 383 384 return (trace_guest_exceptions); 385 } 386 387 int 388 vcpu_trap_wbinvd(struct vcpu *vcpu) 389 { 390 return (trap_wbinvd); 391 } 392 393 struct vm_exit * 394 vm_exitinfo(struct vcpu *vcpu) 395 { 396 return (&vcpu->exitinfo); 397 } 398 399 cpuset_t * 400 vm_exitinfo_cpuset(struct vcpu *vcpu) 401 { 402 return (&vcpu->exitinfo_cpuset); 403 } 404 405 static int 406 vmm_init(void) 407 { 408 int error; 409 410 if (!vmm_is_hw_supported()) 411 return (ENXIO); 412 413 vm_maxcpu = mp_ncpus; 414 TUNABLE_INT_FETCH("hw.vmm.maxcpu", &vm_maxcpu); 415 416 if (vm_maxcpu > VM_MAXCPU) { 417 printf("vmm: vm_maxcpu clamped to %u\n", VM_MAXCPU); 418 vm_maxcpu = VM_MAXCPU; 419 } 420 if (vm_maxcpu == 0) 421 vm_maxcpu = 1; 422 423 vmm_host_state_init(); 424 425 vmm_ipinum = lapic_ipi_alloc(pti ? &IDTVEC(justreturn1_pti) : 426 &IDTVEC(justreturn)); 427 if (vmm_ipinum < 0) 428 vmm_ipinum = IPI_AST; 429 430 error = vmm_mem_init(); 431 if (error) 432 return (error); 433 434 vmm_resume_p = vmmops_modresume; 435 436 return (vmmops_modinit(vmm_ipinum)); 437 } 438 439 static int 440 vmm_handler(module_t mod, int what, void *arg) 441 { 442 int error; 443 444 switch (what) { 445 case MOD_LOAD: 446 if (vmm_is_hw_supported()) { 447 vmmdev_init(); 448 error = vmm_init(); 449 if (error == 0) 450 vmm_initialized = 1; 451 } else { 452 error = ENXIO; 453 } 454 break; 455 case MOD_UNLOAD: 456 if (vmm_is_hw_supported()) { 457 error = vmmdev_cleanup(); 458 if (error == 0) { 459 vmm_resume_p = NULL; 460 iommu_cleanup(); 461 if (vmm_ipinum != IPI_AST) 462 lapic_ipi_free(vmm_ipinum); 463 error = vmmops_modcleanup(); 464 /* 465 * Something bad happened - prevent new 466 * VMs from being created 467 */ 468 if (error) 469 vmm_initialized = 0; 470 } 471 } else { 472 error = 0; 473 } 474 break; 475 default: 476 error = 0; 477 break; 478 } 479 return (error); 480 } 481 482 static moduledata_t vmm_kmod = { 483 "vmm", 484 vmm_handler, 485 NULL 486 }; 487 488 /* 489 * vmm initialization has the following dependencies: 490 * 491 * - VT-x initialization requires smp_rendezvous() and therefore must happen 492 * after SMP is fully functional (after SI_SUB_SMP). 493 */ 494 DECLARE_MODULE(vmm, vmm_kmod, SI_SUB_SMP + 1, SI_ORDER_ANY); 495 MODULE_VERSION(vmm, 1); 496 497 static void 498 vm_init(struct vm *vm, bool create) 499 { 500 vm->cookie = vmmops_init(vm, vmspace_pmap(vm->vmspace)); 501 vm->iommu = NULL; 502 vm->vioapic = vioapic_init(vm); 503 vm->vhpet = vhpet_init(vm); 504 vm->vatpic = vatpic_init(vm); 505 vm->vatpit = vatpit_init(vm); 506 vm->vpmtmr = vpmtmr_init(vm); 507 if (create) 508 vm->vrtc = vrtc_init(vm); 509 510 CPU_ZERO(&vm->active_cpus); 511 CPU_ZERO(&vm->debug_cpus); 512 CPU_ZERO(&vm->startup_cpus); 513 514 vm->suspend = 0; 515 CPU_ZERO(&vm->suspended_cpus); 516 517 if (!create) { 518 for (int i = 0; i < vm->maxcpus; i++) { 519 if (vm->vcpu[i] != NULL) 520 vcpu_init(vm->vcpu[i]); 521 } 522 } 523 } 524 525 void 526 vm_disable_vcpu_creation(struct vm *vm) 527 { 528 sx_xlock(&vm->vcpus_init_lock); 529 vm->dying = true; 530 sx_xunlock(&vm->vcpus_init_lock); 531 } 532 533 struct vcpu * 534 vm_alloc_vcpu(struct vm *vm, int vcpuid) 535 { 536 struct vcpu *vcpu; 537 538 if (vcpuid < 0 || vcpuid >= vm_get_maxcpus(vm)) 539 return (NULL); 540 541 vcpu = atomic_load_ptr(&vm->vcpu[vcpuid]); 542 if (__predict_true(vcpu != NULL)) 543 return (vcpu); 544 545 sx_xlock(&vm->vcpus_init_lock); 546 vcpu = vm->vcpu[vcpuid]; 547 if (vcpu == NULL && !vm->dying) { 548 vcpu = vcpu_alloc(vm, vcpuid); 549 vcpu_init(vcpu); 550 551 /* 552 * Ensure vCPU is fully created before updating pointer 553 * to permit unlocked reads above. 554 */ 555 atomic_store_rel_ptr((uintptr_t *)&vm->vcpu[vcpuid], 556 (uintptr_t)vcpu); 557 } 558 sx_xunlock(&vm->vcpus_init_lock); 559 return (vcpu); 560 } 561 562 void 563 vm_slock_vcpus(struct vm *vm) 564 { 565 sx_slock(&vm->vcpus_init_lock); 566 } 567 568 void 569 vm_unlock_vcpus(struct vm *vm) 570 { 571 sx_unlock(&vm->vcpus_init_lock); 572 } 573 574 /* 575 * The default CPU topology is a single thread per package. 576 */ 577 u_int cores_per_package = 1; 578 u_int threads_per_core = 1; 579 580 int 581 vm_create(const char *name, struct vm **retvm) 582 { 583 struct vm *vm; 584 struct vmspace *vmspace; 585 586 /* 587 * If vmm.ko could not be successfully initialized then don't attempt 588 * to create the virtual machine. 589 */ 590 if (!vmm_initialized) 591 return (ENXIO); 592 593 if (name == NULL || strnlen(name, VM_MAX_NAMELEN + 1) == 594 VM_MAX_NAMELEN + 1) 595 return (EINVAL); 596 597 vmspace = vmmops_vmspace_alloc(0, VM_MAXUSER_ADDRESS_LA48); 598 if (vmspace == NULL) 599 return (ENOMEM); 600 601 vm = malloc(sizeof(struct vm), M_VM, M_WAITOK | M_ZERO); 602 strcpy(vm->name, name); 603 vm->vmspace = vmspace; 604 mtx_init(&vm->rendezvous_mtx, "vm rendezvous lock", 0, MTX_DEF); 605 sx_init(&vm->mem_segs_lock, "vm mem_segs"); 606 sx_init(&vm->vcpus_init_lock, "vm vcpus"); 607 vm->vcpu = malloc(sizeof(*vm->vcpu) * vm_maxcpu, M_VM, M_WAITOK | 608 M_ZERO); 609 610 vm->sockets = 1; 611 vm->cores = cores_per_package; /* XXX backwards compatibility */ 612 vm->threads = threads_per_core; /* XXX backwards compatibility */ 613 vm->maxcpus = vm_maxcpu; 614 615 vm_init(vm, true); 616 617 *retvm = vm; 618 return (0); 619 } 620 621 void 622 vm_get_topology(struct vm *vm, uint16_t *sockets, uint16_t *cores, 623 uint16_t *threads, uint16_t *maxcpus) 624 { 625 *sockets = vm->sockets; 626 *cores = vm->cores; 627 *threads = vm->threads; 628 *maxcpus = vm->maxcpus; 629 } 630 631 uint16_t 632 vm_get_maxcpus(struct vm *vm) 633 { 634 return (vm->maxcpus); 635 } 636 637 int 638 vm_set_topology(struct vm *vm, uint16_t sockets, uint16_t cores, 639 uint16_t threads, uint16_t maxcpus __unused) 640 { 641 /* Ignore maxcpus. */ 642 if ((sockets * cores * threads) > vm->maxcpus) 643 return (EINVAL); 644 vm->sockets = sockets; 645 vm->cores = cores; 646 vm->threads = threads; 647 return(0); 648 } 649 650 static void 651 vm_cleanup(struct vm *vm, bool destroy) 652 { 653 struct mem_map *mm; 654 int i; 655 656 if (destroy) 657 vm_xlock_memsegs(vm); 658 659 ppt_unassign_all(vm); 660 661 if (vm->iommu != NULL) 662 iommu_destroy_domain(vm->iommu); 663 664 if (destroy) 665 vrtc_cleanup(vm->vrtc); 666 else 667 vrtc_reset(vm->vrtc); 668 vpmtmr_cleanup(vm->vpmtmr); 669 vatpit_cleanup(vm->vatpit); 670 vhpet_cleanup(vm->vhpet); 671 vatpic_cleanup(vm->vatpic); 672 vioapic_cleanup(vm->vioapic); 673 674 for (i = 0; i < vm->maxcpus; i++) { 675 if (vm->vcpu[i] != NULL) 676 vcpu_cleanup(vm->vcpu[i], destroy); 677 } 678 679 vmmops_cleanup(vm->cookie); 680 681 /* 682 * System memory is removed from the guest address space only when 683 * the VM is destroyed. This is because the mapping remains the same 684 * across VM reset. 685 * 686 * Device memory can be relocated by the guest (e.g. using PCI BARs) 687 * so those mappings are removed on a VM reset. 688 */ 689 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 690 mm = &vm->mem_maps[i]; 691 if (destroy || !sysmem_mapping(vm, mm)) 692 vm_free_memmap(vm, i); 693 } 694 695 if (destroy) { 696 for (i = 0; i < VM_MAX_MEMSEGS; i++) 697 vm_free_memseg(vm, i); 698 vm_unlock_memsegs(vm); 699 700 vmmops_vmspace_free(vm->vmspace); 701 vm->vmspace = NULL; 702 703 free(vm->vcpu, M_VM); 704 sx_destroy(&vm->vcpus_init_lock); 705 sx_destroy(&vm->mem_segs_lock); 706 mtx_destroy(&vm->rendezvous_mtx); 707 } 708 } 709 710 void 711 vm_destroy(struct vm *vm) 712 { 713 vm_cleanup(vm, true); 714 free(vm, M_VM); 715 } 716 717 int 718 vm_reinit(struct vm *vm) 719 { 720 int error; 721 722 /* 723 * A virtual machine can be reset only if all vcpus are suspended. 724 */ 725 if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) { 726 vm_cleanup(vm, false); 727 vm_init(vm, false); 728 error = 0; 729 } else { 730 error = EBUSY; 731 } 732 733 return (error); 734 } 735 736 const char * 737 vm_name(struct vm *vm) 738 { 739 return (vm->name); 740 } 741 742 void 743 vm_slock_memsegs(struct vm *vm) 744 { 745 sx_slock(&vm->mem_segs_lock); 746 } 747 748 void 749 vm_xlock_memsegs(struct vm *vm) 750 { 751 sx_xlock(&vm->mem_segs_lock); 752 } 753 754 void 755 vm_unlock_memsegs(struct vm *vm) 756 { 757 sx_unlock(&vm->mem_segs_lock); 758 } 759 760 int 761 vm_map_mmio(struct vm *vm, vm_paddr_t gpa, size_t len, vm_paddr_t hpa) 762 { 763 vm_object_t obj; 764 765 if ((obj = vmm_mmio_alloc(vm->vmspace, gpa, len, hpa)) == NULL) 766 return (ENOMEM); 767 else 768 return (0); 769 } 770 771 int 772 vm_unmap_mmio(struct vm *vm, vm_paddr_t gpa, size_t len) 773 { 774 775 vmm_mmio_free(vm->vmspace, gpa, len); 776 return (0); 777 } 778 779 /* 780 * Return 'true' if 'gpa' is allocated in the guest address space. 781 * 782 * This function is called in the context of a running vcpu which acts as 783 * an implicit lock on 'vm->mem_maps[]'. 784 */ 785 bool 786 vm_mem_allocated(struct vcpu *vcpu, vm_paddr_t gpa) 787 { 788 struct vm *vm = vcpu->vm; 789 struct mem_map *mm; 790 int i; 791 792 #ifdef INVARIANTS 793 int hostcpu, state; 794 state = vcpu_get_state(vcpu, &hostcpu); 795 KASSERT(state == VCPU_RUNNING && hostcpu == curcpu, 796 ("%s: invalid vcpu state %d/%d", __func__, state, hostcpu)); 797 #endif 798 799 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 800 mm = &vm->mem_maps[i]; 801 if (mm->len != 0 && gpa >= mm->gpa && gpa < mm->gpa + mm->len) 802 return (true); /* 'gpa' is sysmem or devmem */ 803 } 804 805 if (ppt_is_mmio(vm, gpa)) 806 return (true); /* 'gpa' is pci passthru mmio */ 807 808 return (false); 809 } 810 811 int 812 vm_alloc_memseg(struct vm *vm, int ident, size_t len, bool sysmem) 813 { 814 struct mem_seg *seg; 815 vm_object_t obj; 816 817 sx_assert(&vm->mem_segs_lock, SX_XLOCKED); 818 819 if (ident < 0 || ident >= VM_MAX_MEMSEGS) 820 return (EINVAL); 821 822 if (len == 0 || (len & PAGE_MASK)) 823 return (EINVAL); 824 825 seg = &vm->mem_segs[ident]; 826 if (seg->object != NULL) { 827 if (seg->len == len && seg->sysmem == sysmem) 828 return (EEXIST); 829 else 830 return (EINVAL); 831 } 832 833 obj = vm_object_allocate(OBJT_SWAP, len >> PAGE_SHIFT); 834 if (obj == NULL) 835 return (ENOMEM); 836 837 seg->len = len; 838 seg->object = obj; 839 seg->sysmem = sysmem; 840 return (0); 841 } 842 843 int 844 vm_get_memseg(struct vm *vm, int ident, size_t *len, bool *sysmem, 845 vm_object_t *objptr) 846 { 847 struct mem_seg *seg; 848 849 sx_assert(&vm->mem_segs_lock, SX_LOCKED); 850 851 if (ident < 0 || ident >= VM_MAX_MEMSEGS) 852 return (EINVAL); 853 854 seg = &vm->mem_segs[ident]; 855 if (len) 856 *len = seg->len; 857 if (sysmem) 858 *sysmem = seg->sysmem; 859 if (objptr) 860 *objptr = seg->object; 861 return (0); 862 } 863 864 void 865 vm_free_memseg(struct vm *vm, int ident) 866 { 867 struct mem_seg *seg; 868 869 KASSERT(ident >= 0 && ident < VM_MAX_MEMSEGS, 870 ("%s: invalid memseg ident %d", __func__, ident)); 871 872 seg = &vm->mem_segs[ident]; 873 if (seg->object != NULL) { 874 vm_object_deallocate(seg->object); 875 bzero(seg, sizeof(struct mem_seg)); 876 } 877 } 878 879 int 880 vm_mmap_memseg(struct vm *vm, vm_paddr_t gpa, int segid, vm_ooffset_t first, 881 size_t len, int prot, int flags) 882 { 883 struct mem_seg *seg; 884 struct mem_map *m, *map; 885 vm_ooffset_t last; 886 int i, error; 887 888 if (prot == 0 || (prot & ~(VM_PROT_ALL)) != 0) 889 return (EINVAL); 890 891 if (flags & ~VM_MEMMAP_F_WIRED) 892 return (EINVAL); 893 894 if (segid < 0 || segid >= VM_MAX_MEMSEGS) 895 return (EINVAL); 896 897 seg = &vm->mem_segs[segid]; 898 if (seg->object == NULL) 899 return (EINVAL); 900 901 last = first + len; 902 if (first < 0 || first >= last || last > seg->len) 903 return (EINVAL); 904 905 if ((gpa | first | last) & PAGE_MASK) 906 return (EINVAL); 907 908 map = NULL; 909 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 910 m = &vm->mem_maps[i]; 911 if (m->len == 0) { 912 map = m; 913 break; 914 } 915 } 916 917 if (map == NULL) 918 return (ENOSPC); 919 920 error = vm_map_find(&vm->vmspace->vm_map, seg->object, first, &gpa, 921 len, 0, VMFS_NO_SPACE, prot, prot, 0); 922 if (error != KERN_SUCCESS) 923 return (EFAULT); 924 925 vm_object_reference(seg->object); 926 927 if (flags & VM_MEMMAP_F_WIRED) { 928 error = vm_map_wire(&vm->vmspace->vm_map, gpa, gpa + len, 929 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES); 930 if (error != KERN_SUCCESS) { 931 vm_map_remove(&vm->vmspace->vm_map, gpa, gpa + len); 932 return (error == KERN_RESOURCE_SHORTAGE ? ENOMEM : 933 EFAULT); 934 } 935 } 936 937 map->gpa = gpa; 938 map->len = len; 939 map->segoff = first; 940 map->segid = segid; 941 map->prot = prot; 942 map->flags = flags; 943 return (0); 944 } 945 946 int 947 vm_munmap_memseg(struct vm *vm, vm_paddr_t gpa, size_t len) 948 { 949 struct mem_map *m; 950 int i; 951 952 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 953 m = &vm->mem_maps[i]; 954 if (m->gpa == gpa && m->len == len && 955 (m->flags & VM_MEMMAP_F_IOMMU) == 0) { 956 vm_free_memmap(vm, i); 957 return (0); 958 } 959 } 960 961 return (EINVAL); 962 } 963 964 int 965 vm_mmap_getnext(struct vm *vm, vm_paddr_t *gpa, int *segid, 966 vm_ooffset_t *segoff, size_t *len, int *prot, int *flags) 967 { 968 struct mem_map *mm, *mmnext; 969 int i; 970 971 mmnext = NULL; 972 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 973 mm = &vm->mem_maps[i]; 974 if (mm->len == 0 || mm->gpa < *gpa) 975 continue; 976 if (mmnext == NULL || mm->gpa < mmnext->gpa) 977 mmnext = mm; 978 } 979 980 if (mmnext != NULL) { 981 *gpa = mmnext->gpa; 982 if (segid) 983 *segid = mmnext->segid; 984 if (segoff) 985 *segoff = mmnext->segoff; 986 if (len) 987 *len = mmnext->len; 988 if (prot) 989 *prot = mmnext->prot; 990 if (flags) 991 *flags = mmnext->flags; 992 return (0); 993 } else { 994 return (ENOENT); 995 } 996 } 997 998 static void 999 vm_free_memmap(struct vm *vm, int ident) 1000 { 1001 struct mem_map *mm; 1002 int error __diagused; 1003 1004 mm = &vm->mem_maps[ident]; 1005 if (mm->len) { 1006 error = vm_map_remove(&vm->vmspace->vm_map, mm->gpa, 1007 mm->gpa + mm->len); 1008 KASSERT(error == KERN_SUCCESS, ("%s: vm_map_remove error %d", 1009 __func__, error)); 1010 bzero(mm, sizeof(struct mem_map)); 1011 } 1012 } 1013 1014 static __inline bool 1015 sysmem_mapping(struct vm *vm, struct mem_map *mm) 1016 { 1017 1018 if (mm->len != 0 && vm->mem_segs[mm->segid].sysmem) 1019 return (true); 1020 else 1021 return (false); 1022 } 1023 1024 vm_paddr_t 1025 vmm_sysmem_maxaddr(struct vm *vm) 1026 { 1027 struct mem_map *mm; 1028 vm_paddr_t maxaddr; 1029 int i; 1030 1031 maxaddr = 0; 1032 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 1033 mm = &vm->mem_maps[i]; 1034 if (sysmem_mapping(vm, mm)) { 1035 if (maxaddr < mm->gpa + mm->len) 1036 maxaddr = mm->gpa + mm->len; 1037 } 1038 } 1039 return (maxaddr); 1040 } 1041 1042 static void 1043 vm_iommu_modify(struct vm *vm, bool map) 1044 { 1045 int i, sz; 1046 vm_paddr_t gpa, hpa; 1047 struct mem_map *mm; 1048 void *vp, *cookie, *host_domain; 1049 1050 sz = PAGE_SIZE; 1051 host_domain = iommu_host_domain(); 1052 1053 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 1054 mm = &vm->mem_maps[i]; 1055 if (!sysmem_mapping(vm, mm)) 1056 continue; 1057 1058 if (map) { 1059 KASSERT((mm->flags & VM_MEMMAP_F_IOMMU) == 0, 1060 ("iommu map found invalid memmap %#lx/%#lx/%#x", 1061 mm->gpa, mm->len, mm->flags)); 1062 if ((mm->flags & VM_MEMMAP_F_WIRED) == 0) 1063 continue; 1064 mm->flags |= VM_MEMMAP_F_IOMMU; 1065 } else { 1066 if ((mm->flags & VM_MEMMAP_F_IOMMU) == 0) 1067 continue; 1068 mm->flags &= ~VM_MEMMAP_F_IOMMU; 1069 KASSERT((mm->flags & VM_MEMMAP_F_WIRED) != 0, 1070 ("iommu unmap found invalid memmap %#lx/%#lx/%#x", 1071 mm->gpa, mm->len, mm->flags)); 1072 } 1073 1074 gpa = mm->gpa; 1075 while (gpa < mm->gpa + mm->len) { 1076 vp = vm_gpa_hold_global(vm, gpa, PAGE_SIZE, 1077 VM_PROT_WRITE, &cookie); 1078 KASSERT(vp != NULL, ("vm(%s) could not map gpa %#lx", 1079 vm_name(vm), gpa)); 1080 1081 vm_gpa_release(cookie); 1082 1083 hpa = DMAP_TO_PHYS((uintptr_t)vp); 1084 if (map) { 1085 iommu_create_mapping(vm->iommu, gpa, hpa, sz); 1086 } else { 1087 iommu_remove_mapping(vm->iommu, gpa, sz); 1088 } 1089 1090 gpa += PAGE_SIZE; 1091 } 1092 } 1093 1094 /* 1095 * Invalidate the cached translations associated with the domain 1096 * from which pages were removed. 1097 */ 1098 if (map) 1099 iommu_invalidate_tlb(host_domain); 1100 else 1101 iommu_invalidate_tlb(vm->iommu); 1102 } 1103 1104 #define vm_iommu_unmap(vm) vm_iommu_modify((vm), false) 1105 #define vm_iommu_map(vm) vm_iommu_modify((vm), true) 1106 1107 int 1108 vm_unassign_pptdev(struct vm *vm, int bus, int slot, int func) 1109 { 1110 int error; 1111 1112 error = ppt_unassign_device(vm, bus, slot, func); 1113 if (error) 1114 return (error); 1115 1116 if (ppt_assigned_devices(vm) == 0) 1117 vm_iommu_unmap(vm); 1118 1119 return (0); 1120 } 1121 1122 int 1123 vm_assign_pptdev(struct vm *vm, int bus, int slot, int func) 1124 { 1125 int error; 1126 vm_paddr_t maxaddr; 1127 1128 /* Set up the IOMMU to do the 'gpa' to 'hpa' translation */ 1129 if (ppt_assigned_devices(vm) == 0) { 1130 KASSERT(vm->iommu == NULL, 1131 ("vm_assign_pptdev: iommu must be NULL")); 1132 maxaddr = vmm_sysmem_maxaddr(vm); 1133 vm->iommu = iommu_create_domain(maxaddr); 1134 if (vm->iommu == NULL) 1135 return (ENXIO); 1136 vm_iommu_map(vm); 1137 } 1138 1139 error = ppt_assign_device(vm, bus, slot, func); 1140 return (error); 1141 } 1142 1143 static void * 1144 _vm_gpa_hold(struct vm *vm, vm_paddr_t gpa, size_t len, int reqprot, 1145 void **cookie) 1146 { 1147 int i, count, pageoff; 1148 struct mem_map *mm; 1149 vm_page_t m; 1150 1151 pageoff = gpa & PAGE_MASK; 1152 if (len > PAGE_SIZE - pageoff) 1153 panic("vm_gpa_hold: invalid gpa/len: 0x%016lx/%lu", gpa, len); 1154 1155 count = 0; 1156 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 1157 mm = &vm->mem_maps[i]; 1158 if (gpa >= mm->gpa && gpa < mm->gpa + mm->len) { 1159 count = vm_fault_quick_hold_pages(&vm->vmspace->vm_map, 1160 trunc_page(gpa), PAGE_SIZE, reqprot, &m, 1); 1161 break; 1162 } 1163 } 1164 1165 if (count == 1) { 1166 *cookie = m; 1167 return ((void *)(PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m)) + pageoff)); 1168 } else { 1169 *cookie = NULL; 1170 return (NULL); 1171 } 1172 } 1173 1174 void * 1175 vm_gpa_hold(struct vcpu *vcpu, vm_paddr_t gpa, size_t len, int reqprot, 1176 void **cookie) 1177 { 1178 #ifdef INVARIANTS 1179 /* 1180 * The current vcpu should be frozen to ensure 'vm_memmap[]' 1181 * stability. 1182 */ 1183 int state = vcpu_get_state(vcpu, NULL); 1184 KASSERT(state == VCPU_FROZEN, ("%s: invalid vcpu state %d", 1185 __func__, state)); 1186 #endif 1187 return (_vm_gpa_hold(vcpu->vm, gpa, len, reqprot, cookie)); 1188 } 1189 1190 void * 1191 vm_gpa_hold_global(struct vm *vm, vm_paddr_t gpa, size_t len, int reqprot, 1192 void **cookie) 1193 { 1194 sx_assert(&vm->mem_segs_lock, SX_LOCKED); 1195 return (_vm_gpa_hold(vm, gpa, len, reqprot, cookie)); 1196 } 1197 1198 void 1199 vm_gpa_release(void *cookie) 1200 { 1201 vm_page_t m = cookie; 1202 1203 vm_page_unwire(m, PQ_ACTIVE); 1204 } 1205 1206 int 1207 vm_get_register(struct vcpu *vcpu, int reg, uint64_t *retval) 1208 { 1209 1210 if (reg >= VM_REG_LAST) 1211 return (EINVAL); 1212 1213 return (vmmops_getreg(vcpu->cookie, reg, retval)); 1214 } 1215 1216 int 1217 vm_set_register(struct vcpu *vcpu, int reg, uint64_t val) 1218 { 1219 int error; 1220 1221 if (reg >= VM_REG_LAST) 1222 return (EINVAL); 1223 1224 error = vmmops_setreg(vcpu->cookie, reg, val); 1225 if (error || reg != VM_REG_GUEST_RIP) 1226 return (error); 1227 1228 /* Set 'nextrip' to match the value of %rip */ 1229 VMM_CTR1(vcpu, "Setting nextrip to %#lx", val); 1230 vcpu->nextrip = val; 1231 return (0); 1232 } 1233 1234 static bool 1235 is_descriptor_table(int reg) 1236 { 1237 1238 switch (reg) { 1239 case VM_REG_GUEST_IDTR: 1240 case VM_REG_GUEST_GDTR: 1241 return (true); 1242 default: 1243 return (false); 1244 } 1245 } 1246 1247 static bool 1248 is_segment_register(int reg) 1249 { 1250 1251 switch (reg) { 1252 case VM_REG_GUEST_ES: 1253 case VM_REG_GUEST_CS: 1254 case VM_REG_GUEST_SS: 1255 case VM_REG_GUEST_DS: 1256 case VM_REG_GUEST_FS: 1257 case VM_REG_GUEST_GS: 1258 case VM_REG_GUEST_TR: 1259 case VM_REG_GUEST_LDTR: 1260 return (true); 1261 default: 1262 return (false); 1263 } 1264 } 1265 1266 int 1267 vm_get_seg_desc(struct vcpu *vcpu, int reg, struct seg_desc *desc) 1268 { 1269 1270 if (!is_segment_register(reg) && !is_descriptor_table(reg)) 1271 return (EINVAL); 1272 1273 return (vmmops_getdesc(vcpu->cookie, reg, desc)); 1274 } 1275 1276 int 1277 vm_set_seg_desc(struct vcpu *vcpu, int reg, struct seg_desc *desc) 1278 { 1279 1280 if (!is_segment_register(reg) && !is_descriptor_table(reg)) 1281 return (EINVAL); 1282 1283 return (vmmops_setdesc(vcpu->cookie, reg, desc)); 1284 } 1285 1286 static void 1287 restore_guest_fpustate(struct vcpu *vcpu) 1288 { 1289 1290 /* flush host state to the pcb */ 1291 fpuexit(curthread); 1292 1293 /* restore guest FPU state */ 1294 fpu_stop_emulating(); 1295 fpurestore(vcpu->guestfpu); 1296 1297 /* restore guest XCR0 if XSAVE is enabled in the host */ 1298 if (rcr4() & CR4_XSAVE) 1299 load_xcr(0, vcpu->guest_xcr0); 1300 1301 /* 1302 * The FPU is now "dirty" with the guest's state so turn on emulation 1303 * to trap any access to the FPU by the host. 1304 */ 1305 fpu_start_emulating(); 1306 } 1307 1308 static void 1309 save_guest_fpustate(struct vcpu *vcpu) 1310 { 1311 1312 if ((rcr0() & CR0_TS) == 0) 1313 panic("fpu emulation not enabled in host!"); 1314 1315 /* save guest XCR0 and restore host XCR0 */ 1316 if (rcr4() & CR4_XSAVE) { 1317 vcpu->guest_xcr0 = rxcr(0); 1318 load_xcr(0, vmm_get_host_xcr0()); 1319 } 1320 1321 /* save guest FPU state */ 1322 fpu_stop_emulating(); 1323 fpusave(vcpu->guestfpu); 1324 fpu_start_emulating(); 1325 } 1326 1327 static VMM_STAT(VCPU_IDLE_TICKS, "number of ticks vcpu was idle"); 1328 1329 static int 1330 vcpu_set_state_locked(struct vcpu *vcpu, enum vcpu_state newstate, 1331 bool from_idle) 1332 { 1333 int error; 1334 1335 vcpu_assert_locked(vcpu); 1336 1337 /* 1338 * State transitions from the vmmdev_ioctl() must always begin from 1339 * the VCPU_IDLE state. This guarantees that there is only a single 1340 * ioctl() operating on a vcpu at any point. 1341 */ 1342 if (from_idle) { 1343 while (vcpu->state != VCPU_IDLE) { 1344 vcpu->reqidle = 1; 1345 vcpu_notify_event_locked(vcpu, false); 1346 VMM_CTR1(vcpu, "vcpu state change from %s to " 1347 "idle requested", vcpu_state2str(vcpu->state)); 1348 msleep_spin(&vcpu->state, &vcpu->mtx, "vmstat", hz); 1349 } 1350 } else { 1351 KASSERT(vcpu->state != VCPU_IDLE, ("invalid transition from " 1352 "vcpu idle state")); 1353 } 1354 1355 if (vcpu->state == VCPU_RUNNING) { 1356 KASSERT(vcpu->hostcpu == curcpu, ("curcpu %d and hostcpu %d " 1357 "mismatch for running vcpu", curcpu, vcpu->hostcpu)); 1358 } else { 1359 KASSERT(vcpu->hostcpu == NOCPU, ("Invalid hostcpu %d for a " 1360 "vcpu that is not running", vcpu->hostcpu)); 1361 } 1362 1363 /* 1364 * The following state transitions are allowed: 1365 * IDLE -> FROZEN -> IDLE 1366 * FROZEN -> RUNNING -> FROZEN 1367 * FROZEN -> SLEEPING -> FROZEN 1368 */ 1369 switch (vcpu->state) { 1370 case VCPU_IDLE: 1371 case VCPU_RUNNING: 1372 case VCPU_SLEEPING: 1373 error = (newstate != VCPU_FROZEN); 1374 break; 1375 case VCPU_FROZEN: 1376 error = (newstate == VCPU_FROZEN); 1377 break; 1378 default: 1379 error = 1; 1380 break; 1381 } 1382 1383 if (error) 1384 return (EBUSY); 1385 1386 VMM_CTR2(vcpu, "vcpu state changed from %s to %s", 1387 vcpu_state2str(vcpu->state), vcpu_state2str(newstate)); 1388 1389 vcpu->state = newstate; 1390 if (newstate == VCPU_RUNNING) 1391 vcpu->hostcpu = curcpu; 1392 else 1393 vcpu->hostcpu = NOCPU; 1394 1395 if (newstate == VCPU_IDLE) 1396 wakeup(&vcpu->state); 1397 1398 return (0); 1399 } 1400 1401 static void 1402 vcpu_require_state(struct vcpu *vcpu, enum vcpu_state newstate) 1403 { 1404 int error; 1405 1406 if ((error = vcpu_set_state(vcpu, newstate, false)) != 0) 1407 panic("Error %d setting state to %d\n", error, newstate); 1408 } 1409 1410 static void 1411 vcpu_require_state_locked(struct vcpu *vcpu, enum vcpu_state newstate) 1412 { 1413 int error; 1414 1415 if ((error = vcpu_set_state_locked(vcpu, newstate, false)) != 0) 1416 panic("Error %d setting state to %d", error, newstate); 1417 } 1418 1419 static int 1420 vm_handle_rendezvous(struct vcpu *vcpu) 1421 { 1422 struct vm *vm = vcpu->vm; 1423 struct thread *td; 1424 int error, vcpuid; 1425 1426 error = 0; 1427 vcpuid = vcpu->vcpuid; 1428 td = curthread; 1429 mtx_lock(&vm->rendezvous_mtx); 1430 while (vm->rendezvous_func != NULL) { 1431 /* 'rendezvous_req_cpus' must be a subset of 'active_cpus' */ 1432 CPU_AND(&vm->rendezvous_req_cpus, &vm->rendezvous_req_cpus, &vm->active_cpus); 1433 1434 if (CPU_ISSET(vcpuid, &vm->rendezvous_req_cpus) && 1435 !CPU_ISSET(vcpuid, &vm->rendezvous_done_cpus)) { 1436 VMM_CTR0(vcpu, "Calling rendezvous func"); 1437 (*vm->rendezvous_func)(vcpu, vm->rendezvous_arg); 1438 CPU_SET(vcpuid, &vm->rendezvous_done_cpus); 1439 } 1440 if (CPU_CMP(&vm->rendezvous_req_cpus, 1441 &vm->rendezvous_done_cpus) == 0) { 1442 VMM_CTR0(vcpu, "Rendezvous completed"); 1443 CPU_ZERO(&vm->rendezvous_req_cpus); 1444 vm->rendezvous_func = NULL; 1445 wakeup(&vm->rendezvous_func); 1446 break; 1447 } 1448 VMM_CTR0(vcpu, "Wait for rendezvous completion"); 1449 mtx_sleep(&vm->rendezvous_func, &vm->rendezvous_mtx, 0, 1450 "vmrndv", hz); 1451 if (td_ast_pending(td, TDA_SUSPEND)) { 1452 mtx_unlock(&vm->rendezvous_mtx); 1453 error = thread_check_susp(td, true); 1454 if (error != 0) 1455 return (error); 1456 mtx_lock(&vm->rendezvous_mtx); 1457 } 1458 } 1459 mtx_unlock(&vm->rendezvous_mtx); 1460 return (0); 1461 } 1462 1463 /* 1464 * Emulate a guest 'hlt' by sleeping until the vcpu is ready to run. 1465 */ 1466 static int 1467 vm_handle_hlt(struct vcpu *vcpu, bool intr_disabled, bool *retu) 1468 { 1469 struct vm *vm = vcpu->vm; 1470 const char *wmesg; 1471 struct thread *td; 1472 int error, t, vcpuid, vcpu_halted, vm_halted; 1473 1474 vcpuid = vcpu->vcpuid; 1475 vcpu_halted = 0; 1476 vm_halted = 0; 1477 error = 0; 1478 td = curthread; 1479 1480 KASSERT(!CPU_ISSET(vcpuid, &vm->halted_cpus), ("vcpu already halted")); 1481 1482 vcpu_lock(vcpu); 1483 while (1) { 1484 /* 1485 * Do a final check for pending NMI or interrupts before 1486 * really putting this thread to sleep. Also check for 1487 * software events that would cause this vcpu to wakeup. 1488 * 1489 * These interrupts/events could have happened after the 1490 * vcpu returned from vmmops_run() and before it acquired the 1491 * vcpu lock above. 1492 */ 1493 if (vm->rendezvous_func != NULL || vm->suspend || vcpu->reqidle) 1494 break; 1495 if (vm_nmi_pending(vcpu)) 1496 break; 1497 if (!intr_disabled) { 1498 if (vm_extint_pending(vcpu) || 1499 vlapic_pending_intr(vcpu->vlapic, NULL)) { 1500 break; 1501 } 1502 } 1503 1504 /* Don't go to sleep if the vcpu thread needs to yield */ 1505 if (vcpu_should_yield(vcpu)) 1506 break; 1507 1508 if (vcpu_debugged(vcpu)) 1509 break; 1510 1511 /* 1512 * Some Linux guests implement "halt" by having all vcpus 1513 * execute HLT with interrupts disabled. 'halted_cpus' keeps 1514 * track of the vcpus that have entered this state. When all 1515 * vcpus enter the halted state the virtual machine is halted. 1516 */ 1517 if (intr_disabled) { 1518 wmesg = "vmhalt"; 1519 VMM_CTR0(vcpu, "Halted"); 1520 if (!vcpu_halted && halt_detection_enabled) { 1521 vcpu_halted = 1; 1522 CPU_SET_ATOMIC(vcpuid, &vm->halted_cpus); 1523 } 1524 if (CPU_CMP(&vm->halted_cpus, &vm->active_cpus) == 0) { 1525 vm_halted = 1; 1526 break; 1527 } 1528 } else { 1529 wmesg = "vmidle"; 1530 } 1531 1532 t = ticks; 1533 vcpu_require_state_locked(vcpu, VCPU_SLEEPING); 1534 /* 1535 * XXX msleep_spin() cannot be interrupted by signals so 1536 * wake up periodically to check pending signals. 1537 */ 1538 msleep_spin(vcpu, &vcpu->mtx, wmesg, hz); 1539 vcpu_require_state_locked(vcpu, VCPU_FROZEN); 1540 vmm_stat_incr(vcpu, VCPU_IDLE_TICKS, ticks - t); 1541 if (td_ast_pending(td, TDA_SUSPEND)) { 1542 vcpu_unlock(vcpu); 1543 error = thread_check_susp(td, false); 1544 if (error != 0) { 1545 if (vcpu_halted) { 1546 CPU_CLR_ATOMIC(vcpuid, 1547 &vm->halted_cpus); 1548 } 1549 return (error); 1550 } 1551 vcpu_lock(vcpu); 1552 } 1553 } 1554 1555 if (vcpu_halted) 1556 CPU_CLR_ATOMIC(vcpuid, &vm->halted_cpus); 1557 1558 vcpu_unlock(vcpu); 1559 1560 if (vm_halted) 1561 vm_suspend(vm, VM_SUSPEND_HALT); 1562 1563 return (0); 1564 } 1565 1566 static int 1567 vm_handle_paging(struct vcpu *vcpu, bool *retu) 1568 { 1569 struct vm *vm = vcpu->vm; 1570 int rv, ftype; 1571 struct vm_map *map; 1572 struct vm_exit *vme; 1573 1574 vme = &vcpu->exitinfo; 1575 1576 KASSERT(vme->inst_length == 0, ("%s: invalid inst_length %d", 1577 __func__, vme->inst_length)); 1578 1579 ftype = vme->u.paging.fault_type; 1580 KASSERT(ftype == VM_PROT_READ || 1581 ftype == VM_PROT_WRITE || ftype == VM_PROT_EXECUTE, 1582 ("vm_handle_paging: invalid fault_type %d", ftype)); 1583 1584 if (ftype == VM_PROT_READ || ftype == VM_PROT_WRITE) { 1585 rv = pmap_emulate_accessed_dirty(vmspace_pmap(vm->vmspace), 1586 vme->u.paging.gpa, ftype); 1587 if (rv == 0) { 1588 VMM_CTR2(vcpu, "%s bit emulation for gpa %#lx", 1589 ftype == VM_PROT_READ ? "accessed" : "dirty", 1590 vme->u.paging.gpa); 1591 goto done; 1592 } 1593 } 1594 1595 map = &vm->vmspace->vm_map; 1596 rv = vm_fault(map, vme->u.paging.gpa, ftype, VM_FAULT_NORMAL, NULL); 1597 1598 VMM_CTR3(vcpu, "vm_handle_paging rv = %d, gpa = %#lx, " 1599 "ftype = %d", rv, vme->u.paging.gpa, ftype); 1600 1601 if (rv != KERN_SUCCESS) 1602 return (EFAULT); 1603 done: 1604 return (0); 1605 } 1606 1607 static int 1608 vm_handle_inst_emul(struct vcpu *vcpu, bool *retu) 1609 { 1610 struct vie *vie; 1611 struct vm_exit *vme; 1612 uint64_t gla, gpa, cs_base; 1613 struct vm_guest_paging *paging; 1614 mem_region_read_t mread; 1615 mem_region_write_t mwrite; 1616 enum vm_cpu_mode cpu_mode; 1617 int cs_d, error, fault; 1618 1619 vme = &vcpu->exitinfo; 1620 1621 KASSERT(vme->inst_length == 0, ("%s: invalid inst_length %d", 1622 __func__, vme->inst_length)); 1623 1624 gla = vme->u.inst_emul.gla; 1625 gpa = vme->u.inst_emul.gpa; 1626 cs_base = vme->u.inst_emul.cs_base; 1627 cs_d = vme->u.inst_emul.cs_d; 1628 vie = &vme->u.inst_emul.vie; 1629 paging = &vme->u.inst_emul.paging; 1630 cpu_mode = paging->cpu_mode; 1631 1632 VMM_CTR1(vcpu, "inst_emul fault accessing gpa %#lx", gpa); 1633 1634 /* Fetch, decode and emulate the faulting instruction */ 1635 if (vie->num_valid == 0) { 1636 error = vmm_fetch_instruction(vcpu, paging, vme->rip + cs_base, 1637 VIE_INST_SIZE, vie, &fault); 1638 } else { 1639 /* 1640 * The instruction bytes have already been copied into 'vie' 1641 */ 1642 error = fault = 0; 1643 } 1644 if (error || fault) 1645 return (error); 1646 1647 if (vmm_decode_instruction(vcpu, gla, cpu_mode, cs_d, vie) != 0) { 1648 VMM_CTR1(vcpu, "Error decoding instruction at %#lx", 1649 vme->rip + cs_base); 1650 *retu = true; /* dump instruction bytes in userspace */ 1651 return (0); 1652 } 1653 1654 /* 1655 * Update 'nextrip' based on the length of the emulated instruction. 1656 */ 1657 vme->inst_length = vie->num_processed; 1658 vcpu->nextrip += vie->num_processed; 1659 VMM_CTR1(vcpu, "nextrip updated to %#lx after instruction decoding", 1660 vcpu->nextrip); 1661 1662 /* return to userland unless this is an in-kernel emulated device */ 1663 if (gpa >= DEFAULT_APIC_BASE && gpa < DEFAULT_APIC_BASE + PAGE_SIZE) { 1664 mread = lapic_mmio_read; 1665 mwrite = lapic_mmio_write; 1666 } else if (gpa >= VIOAPIC_BASE && gpa < VIOAPIC_BASE + VIOAPIC_SIZE) { 1667 mread = vioapic_mmio_read; 1668 mwrite = vioapic_mmio_write; 1669 } else if (gpa >= VHPET_BASE && gpa < VHPET_BASE + VHPET_SIZE) { 1670 mread = vhpet_mmio_read; 1671 mwrite = vhpet_mmio_write; 1672 } else { 1673 *retu = true; 1674 return (0); 1675 } 1676 1677 error = vmm_emulate_instruction(vcpu, gpa, vie, paging, mread, mwrite, 1678 retu); 1679 1680 return (error); 1681 } 1682 1683 static int 1684 vm_handle_suspend(struct vcpu *vcpu, bool *retu) 1685 { 1686 struct vm *vm = vcpu->vm; 1687 int error, i; 1688 struct thread *td; 1689 1690 error = 0; 1691 td = curthread; 1692 1693 CPU_SET_ATOMIC(vcpu->vcpuid, &vm->suspended_cpus); 1694 1695 /* 1696 * Wait until all 'active_cpus' have suspended themselves. 1697 * 1698 * Since a VM may be suspended at any time including when one or 1699 * more vcpus are doing a rendezvous we need to call the rendezvous 1700 * handler while we are waiting to prevent a deadlock. 1701 */ 1702 vcpu_lock(vcpu); 1703 while (error == 0) { 1704 if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) { 1705 VMM_CTR0(vcpu, "All vcpus suspended"); 1706 break; 1707 } 1708 1709 if (vm->rendezvous_func == NULL) { 1710 VMM_CTR0(vcpu, "Sleeping during suspend"); 1711 vcpu_require_state_locked(vcpu, VCPU_SLEEPING); 1712 msleep_spin(vcpu, &vcpu->mtx, "vmsusp", hz); 1713 vcpu_require_state_locked(vcpu, VCPU_FROZEN); 1714 if (td_ast_pending(td, TDA_SUSPEND)) { 1715 vcpu_unlock(vcpu); 1716 error = thread_check_susp(td, false); 1717 vcpu_lock(vcpu); 1718 } 1719 } else { 1720 VMM_CTR0(vcpu, "Rendezvous during suspend"); 1721 vcpu_unlock(vcpu); 1722 error = vm_handle_rendezvous(vcpu); 1723 vcpu_lock(vcpu); 1724 } 1725 } 1726 vcpu_unlock(vcpu); 1727 1728 /* 1729 * Wakeup the other sleeping vcpus and return to userspace. 1730 */ 1731 for (i = 0; i < vm->maxcpus; i++) { 1732 if (CPU_ISSET(i, &vm->suspended_cpus)) { 1733 vcpu_notify_event(vm_vcpu(vm, i), false); 1734 } 1735 } 1736 1737 *retu = true; 1738 return (error); 1739 } 1740 1741 static int 1742 vm_handle_reqidle(struct vcpu *vcpu, bool *retu) 1743 { 1744 vcpu_lock(vcpu); 1745 KASSERT(vcpu->reqidle, ("invalid vcpu reqidle %d", vcpu->reqidle)); 1746 vcpu->reqidle = 0; 1747 vcpu_unlock(vcpu); 1748 *retu = true; 1749 return (0); 1750 } 1751 1752 int 1753 vm_suspend(struct vm *vm, enum vm_suspend_how how) 1754 { 1755 int i; 1756 1757 if (how <= VM_SUSPEND_NONE || how >= VM_SUSPEND_LAST) 1758 return (EINVAL); 1759 1760 if (atomic_cmpset_int(&vm->suspend, 0, how) == 0) { 1761 VM_CTR2(vm, "virtual machine already suspended %d/%d", 1762 vm->suspend, how); 1763 return (EALREADY); 1764 } 1765 1766 VM_CTR1(vm, "virtual machine successfully suspended %d", how); 1767 1768 /* 1769 * Notify all active vcpus that they are now suspended. 1770 */ 1771 for (i = 0; i < vm->maxcpus; i++) { 1772 if (CPU_ISSET(i, &vm->active_cpus)) 1773 vcpu_notify_event(vm_vcpu(vm, i), false); 1774 } 1775 1776 return (0); 1777 } 1778 1779 void 1780 vm_exit_suspended(struct vcpu *vcpu, uint64_t rip) 1781 { 1782 struct vm *vm = vcpu->vm; 1783 struct vm_exit *vmexit; 1784 1785 KASSERT(vm->suspend > VM_SUSPEND_NONE && vm->suspend < VM_SUSPEND_LAST, 1786 ("vm_exit_suspended: invalid suspend type %d", vm->suspend)); 1787 1788 vmexit = vm_exitinfo(vcpu); 1789 vmexit->rip = rip; 1790 vmexit->inst_length = 0; 1791 vmexit->exitcode = VM_EXITCODE_SUSPENDED; 1792 vmexit->u.suspended.how = vm->suspend; 1793 } 1794 1795 void 1796 vm_exit_debug(struct vcpu *vcpu, uint64_t rip) 1797 { 1798 struct vm_exit *vmexit; 1799 1800 vmexit = vm_exitinfo(vcpu); 1801 vmexit->rip = rip; 1802 vmexit->inst_length = 0; 1803 vmexit->exitcode = VM_EXITCODE_DEBUG; 1804 } 1805 1806 void 1807 vm_exit_rendezvous(struct vcpu *vcpu, uint64_t rip) 1808 { 1809 struct vm_exit *vmexit; 1810 1811 vmexit = vm_exitinfo(vcpu); 1812 vmexit->rip = rip; 1813 vmexit->inst_length = 0; 1814 vmexit->exitcode = VM_EXITCODE_RENDEZVOUS; 1815 vmm_stat_incr(vcpu, VMEXIT_RENDEZVOUS, 1); 1816 } 1817 1818 void 1819 vm_exit_reqidle(struct vcpu *vcpu, uint64_t rip) 1820 { 1821 struct vm_exit *vmexit; 1822 1823 vmexit = vm_exitinfo(vcpu); 1824 vmexit->rip = rip; 1825 vmexit->inst_length = 0; 1826 vmexit->exitcode = VM_EXITCODE_REQIDLE; 1827 vmm_stat_incr(vcpu, VMEXIT_REQIDLE, 1); 1828 } 1829 1830 void 1831 vm_exit_astpending(struct vcpu *vcpu, uint64_t rip) 1832 { 1833 struct vm_exit *vmexit; 1834 1835 vmexit = vm_exitinfo(vcpu); 1836 vmexit->rip = rip; 1837 vmexit->inst_length = 0; 1838 vmexit->exitcode = VM_EXITCODE_BOGUS; 1839 vmm_stat_incr(vcpu, VMEXIT_ASTPENDING, 1); 1840 } 1841 1842 int 1843 vm_run(struct vcpu *vcpu) 1844 { 1845 struct vm *vm = vcpu->vm; 1846 struct vm_eventinfo evinfo; 1847 int error, vcpuid; 1848 struct pcb *pcb; 1849 uint64_t tscval; 1850 struct vm_exit *vme; 1851 bool retu, intr_disabled; 1852 pmap_t pmap; 1853 1854 vcpuid = vcpu->vcpuid; 1855 1856 if (!CPU_ISSET(vcpuid, &vm->active_cpus)) 1857 return (EINVAL); 1858 1859 if (CPU_ISSET(vcpuid, &vm->suspended_cpus)) 1860 return (EINVAL); 1861 1862 pmap = vmspace_pmap(vm->vmspace); 1863 vme = &vcpu->exitinfo; 1864 evinfo.rptr = &vm->rendezvous_req_cpus; 1865 evinfo.sptr = &vm->suspend; 1866 evinfo.iptr = &vcpu->reqidle; 1867 restart: 1868 critical_enter(); 1869 1870 KASSERT(!CPU_ISSET(curcpu, &pmap->pm_active), 1871 ("vm_run: absurd pm_active")); 1872 1873 tscval = rdtsc(); 1874 1875 pcb = PCPU_GET(curpcb); 1876 set_pcb_flags(pcb, PCB_FULL_IRET); 1877 1878 restore_guest_fpustate(vcpu); 1879 1880 vcpu_require_state(vcpu, VCPU_RUNNING); 1881 error = vmmops_run(vcpu->cookie, vcpu->nextrip, pmap, &evinfo); 1882 vcpu_require_state(vcpu, VCPU_FROZEN); 1883 1884 save_guest_fpustate(vcpu); 1885 1886 vmm_stat_incr(vcpu, VCPU_TOTAL_RUNTIME, rdtsc() - tscval); 1887 1888 critical_exit(); 1889 1890 if (error == 0) { 1891 retu = false; 1892 vcpu->nextrip = vme->rip + vme->inst_length; 1893 switch (vme->exitcode) { 1894 case VM_EXITCODE_REQIDLE: 1895 error = vm_handle_reqidle(vcpu, &retu); 1896 break; 1897 case VM_EXITCODE_SUSPENDED: 1898 error = vm_handle_suspend(vcpu, &retu); 1899 break; 1900 case VM_EXITCODE_IOAPIC_EOI: 1901 vioapic_process_eoi(vm, vme->u.ioapic_eoi.vector); 1902 break; 1903 case VM_EXITCODE_RENDEZVOUS: 1904 error = vm_handle_rendezvous(vcpu); 1905 break; 1906 case VM_EXITCODE_HLT: 1907 intr_disabled = ((vme->u.hlt.rflags & PSL_I) == 0); 1908 error = vm_handle_hlt(vcpu, intr_disabled, &retu); 1909 break; 1910 case VM_EXITCODE_PAGING: 1911 error = vm_handle_paging(vcpu, &retu); 1912 break; 1913 case VM_EXITCODE_INST_EMUL: 1914 error = vm_handle_inst_emul(vcpu, &retu); 1915 break; 1916 case VM_EXITCODE_INOUT: 1917 case VM_EXITCODE_INOUT_STR: 1918 error = vm_handle_inout(vcpu, vme, &retu); 1919 break; 1920 case VM_EXITCODE_MONITOR: 1921 case VM_EXITCODE_MWAIT: 1922 case VM_EXITCODE_VMINSN: 1923 vm_inject_ud(vcpu); 1924 break; 1925 default: 1926 retu = true; /* handled in userland */ 1927 break; 1928 } 1929 } 1930 1931 /* 1932 * VM_EXITCODE_INST_EMUL could access the apic which could transform the 1933 * exit code into VM_EXITCODE_IPI. 1934 */ 1935 if (error == 0 && vme->exitcode == VM_EXITCODE_IPI) 1936 error = vm_handle_ipi(vcpu, vme, &retu); 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 return (error); 1945 } 1946 1947 int 1948 vm_restart_instruction(struct vcpu *vcpu) 1949 { 1950 enum vcpu_state state; 1951 uint64_t rip; 1952 int error __diagused; 1953 1954 state = vcpu_get_state(vcpu, NULL); 1955 if (state == VCPU_RUNNING) { 1956 /* 1957 * When a vcpu is "running" the next instruction is determined 1958 * by adding 'rip' and 'inst_length' in the vcpu's 'exitinfo'. 1959 * Thus setting 'inst_length' to zero will cause the current 1960 * instruction to be restarted. 1961 */ 1962 vcpu->exitinfo.inst_length = 0; 1963 VMM_CTR1(vcpu, "restarting instruction at %#lx by " 1964 "setting inst_length to zero", vcpu->exitinfo.rip); 1965 } else if (state == VCPU_FROZEN) { 1966 /* 1967 * When a vcpu is "frozen" it is outside the critical section 1968 * around vmmops_run() and 'nextrip' points to the next 1969 * instruction. Thus instruction restart is achieved by setting 1970 * 'nextrip' to the vcpu's %rip. 1971 */ 1972 error = vm_get_register(vcpu, VM_REG_GUEST_RIP, &rip); 1973 KASSERT(!error, ("%s: error %d getting rip", __func__, error)); 1974 VMM_CTR2(vcpu, "restarting instruction by updating " 1975 "nextrip from %#lx to %#lx", vcpu->nextrip, rip); 1976 vcpu->nextrip = rip; 1977 } else { 1978 panic("%s: invalid state %d", __func__, state); 1979 } 1980 return (0); 1981 } 1982 1983 int 1984 vm_exit_intinfo(struct vcpu *vcpu, uint64_t info) 1985 { 1986 int type, vector; 1987 1988 if (info & VM_INTINFO_VALID) { 1989 type = info & VM_INTINFO_TYPE; 1990 vector = info & 0xff; 1991 if (type == VM_INTINFO_NMI && vector != IDT_NMI) 1992 return (EINVAL); 1993 if (type == VM_INTINFO_HWEXCEPTION && vector >= 32) 1994 return (EINVAL); 1995 if (info & VM_INTINFO_RSVD) 1996 return (EINVAL); 1997 } else { 1998 info = 0; 1999 } 2000 VMM_CTR2(vcpu, "%s: info1(%#lx)", __func__, info); 2001 vcpu->exitintinfo = info; 2002 return (0); 2003 } 2004 2005 enum exc_class { 2006 EXC_BENIGN, 2007 EXC_CONTRIBUTORY, 2008 EXC_PAGEFAULT 2009 }; 2010 2011 #define IDT_VE 20 /* Virtualization Exception (Intel specific) */ 2012 2013 static enum exc_class 2014 exception_class(uint64_t info) 2015 { 2016 int type, vector; 2017 2018 KASSERT(info & VM_INTINFO_VALID, ("intinfo must be valid: %#lx", info)); 2019 type = info & VM_INTINFO_TYPE; 2020 vector = info & 0xff; 2021 2022 /* Table 6-4, "Interrupt and Exception Classes", Intel SDM, Vol 3 */ 2023 switch (type) { 2024 case VM_INTINFO_HWINTR: 2025 case VM_INTINFO_SWINTR: 2026 case VM_INTINFO_NMI: 2027 return (EXC_BENIGN); 2028 default: 2029 /* 2030 * Hardware exception. 2031 * 2032 * SVM and VT-x use identical type values to represent NMI, 2033 * hardware interrupt and software interrupt. 2034 * 2035 * SVM uses type '3' for all exceptions. VT-x uses type '3' 2036 * for exceptions except #BP and #OF. #BP and #OF use a type 2037 * value of '5' or '6'. Therefore we don't check for explicit 2038 * values of 'type' to classify 'intinfo' into a hardware 2039 * exception. 2040 */ 2041 break; 2042 } 2043 2044 switch (vector) { 2045 case IDT_PF: 2046 case IDT_VE: 2047 return (EXC_PAGEFAULT); 2048 case IDT_DE: 2049 case IDT_TS: 2050 case IDT_NP: 2051 case IDT_SS: 2052 case IDT_GP: 2053 return (EXC_CONTRIBUTORY); 2054 default: 2055 return (EXC_BENIGN); 2056 } 2057 } 2058 2059 static int 2060 nested_fault(struct vcpu *vcpu, uint64_t info1, uint64_t info2, 2061 uint64_t *retinfo) 2062 { 2063 enum exc_class exc1, exc2; 2064 int type1, vector1; 2065 2066 KASSERT(info1 & VM_INTINFO_VALID, ("info1 %#lx is not valid", info1)); 2067 KASSERT(info2 & VM_INTINFO_VALID, ("info2 %#lx is not valid", info2)); 2068 2069 /* 2070 * If an exception occurs while attempting to call the double-fault 2071 * handler the processor enters shutdown mode (aka triple fault). 2072 */ 2073 type1 = info1 & VM_INTINFO_TYPE; 2074 vector1 = info1 & 0xff; 2075 if (type1 == VM_INTINFO_HWEXCEPTION && vector1 == IDT_DF) { 2076 VMM_CTR2(vcpu, "triple fault: info1(%#lx), info2(%#lx)", 2077 info1, info2); 2078 vm_suspend(vcpu->vm, VM_SUSPEND_TRIPLEFAULT); 2079 *retinfo = 0; 2080 return (0); 2081 } 2082 2083 /* 2084 * Table 6-5 "Conditions for Generating a Double Fault", Intel SDM, Vol3 2085 */ 2086 exc1 = exception_class(info1); 2087 exc2 = exception_class(info2); 2088 if ((exc1 == EXC_CONTRIBUTORY && exc2 == EXC_CONTRIBUTORY) || 2089 (exc1 == EXC_PAGEFAULT && exc2 != EXC_BENIGN)) { 2090 /* Convert nested fault into a double fault. */ 2091 *retinfo = IDT_DF; 2092 *retinfo |= VM_INTINFO_VALID | VM_INTINFO_HWEXCEPTION; 2093 *retinfo |= VM_INTINFO_DEL_ERRCODE; 2094 } else { 2095 /* Handle exceptions serially */ 2096 *retinfo = info2; 2097 } 2098 return (1); 2099 } 2100 2101 static uint64_t 2102 vcpu_exception_intinfo(struct vcpu *vcpu) 2103 { 2104 uint64_t info = 0; 2105 2106 if (vcpu->exception_pending) { 2107 info = vcpu->exc_vector & 0xff; 2108 info |= VM_INTINFO_VALID | VM_INTINFO_HWEXCEPTION; 2109 if (vcpu->exc_errcode_valid) { 2110 info |= VM_INTINFO_DEL_ERRCODE; 2111 info |= (uint64_t)vcpu->exc_errcode << 32; 2112 } 2113 } 2114 return (info); 2115 } 2116 2117 int 2118 vm_entry_intinfo(struct vcpu *vcpu, uint64_t *retinfo) 2119 { 2120 uint64_t info1, info2; 2121 int valid; 2122 2123 info1 = vcpu->exitintinfo; 2124 vcpu->exitintinfo = 0; 2125 2126 info2 = 0; 2127 if (vcpu->exception_pending) { 2128 info2 = vcpu_exception_intinfo(vcpu); 2129 vcpu->exception_pending = 0; 2130 VMM_CTR2(vcpu, "Exception %d delivered: %#lx", 2131 vcpu->exc_vector, info2); 2132 } 2133 2134 if ((info1 & VM_INTINFO_VALID) && (info2 & VM_INTINFO_VALID)) { 2135 valid = nested_fault(vcpu, info1, info2, retinfo); 2136 } else if (info1 & VM_INTINFO_VALID) { 2137 *retinfo = info1; 2138 valid = 1; 2139 } else if (info2 & VM_INTINFO_VALID) { 2140 *retinfo = info2; 2141 valid = 1; 2142 } else { 2143 valid = 0; 2144 } 2145 2146 if (valid) { 2147 VMM_CTR4(vcpu, "%s: info1(%#lx), info2(%#lx), " 2148 "retinfo(%#lx)", __func__, info1, info2, *retinfo); 2149 } 2150 2151 return (valid); 2152 } 2153 2154 int 2155 vm_get_intinfo(struct vcpu *vcpu, uint64_t *info1, uint64_t *info2) 2156 { 2157 *info1 = vcpu->exitintinfo; 2158 *info2 = vcpu_exception_intinfo(vcpu); 2159 return (0); 2160 } 2161 2162 int 2163 vm_inject_exception(struct vcpu *vcpu, int vector, int errcode_valid, 2164 uint32_t errcode, int restart_instruction) 2165 { 2166 uint64_t regval; 2167 int error __diagused; 2168 2169 if (vector < 0 || vector >= 32) 2170 return (EINVAL); 2171 2172 /* 2173 * A double fault exception should never be injected directly into 2174 * the guest. It is a derived exception that results from specific 2175 * combinations of nested faults. 2176 */ 2177 if (vector == IDT_DF) 2178 return (EINVAL); 2179 2180 if (vcpu->exception_pending) { 2181 VMM_CTR2(vcpu, "Unable to inject exception %d due to " 2182 "pending exception %d", vector, vcpu->exc_vector); 2183 return (EBUSY); 2184 } 2185 2186 if (errcode_valid) { 2187 /* 2188 * Exceptions don't deliver an error code in real mode. 2189 */ 2190 error = vm_get_register(vcpu, VM_REG_GUEST_CR0, ®val); 2191 KASSERT(!error, ("%s: error %d getting CR0", __func__, error)); 2192 if (!(regval & CR0_PE)) 2193 errcode_valid = 0; 2194 } 2195 2196 /* 2197 * From section 26.6.1 "Interruptibility State" in Intel SDM: 2198 * 2199 * Event blocking by "STI" or "MOV SS" is cleared after guest executes 2200 * one instruction or incurs an exception. 2201 */ 2202 error = vm_set_register(vcpu, VM_REG_GUEST_INTR_SHADOW, 0); 2203 KASSERT(error == 0, ("%s: error %d clearing interrupt shadow", 2204 __func__, error)); 2205 2206 if (restart_instruction) 2207 vm_restart_instruction(vcpu); 2208 2209 vcpu->exception_pending = 1; 2210 vcpu->exc_vector = vector; 2211 vcpu->exc_errcode = errcode; 2212 vcpu->exc_errcode_valid = errcode_valid; 2213 VMM_CTR1(vcpu, "Exception %d pending", vector); 2214 return (0); 2215 } 2216 2217 void 2218 vm_inject_fault(struct vcpu *vcpu, int vector, int errcode_valid, int errcode) 2219 { 2220 int error __diagused, restart_instruction; 2221 2222 restart_instruction = 1; 2223 2224 error = vm_inject_exception(vcpu, vector, errcode_valid, 2225 errcode, restart_instruction); 2226 KASSERT(error == 0, ("vm_inject_exception error %d", error)); 2227 } 2228 2229 void 2230 vm_inject_pf(struct vcpu *vcpu, int error_code, uint64_t cr2) 2231 { 2232 int error __diagused; 2233 2234 VMM_CTR2(vcpu, "Injecting page fault: error_code %#x, cr2 %#lx", 2235 error_code, cr2); 2236 2237 error = vm_set_register(vcpu, VM_REG_GUEST_CR2, cr2); 2238 KASSERT(error == 0, ("vm_set_register(cr2) error %d", error)); 2239 2240 vm_inject_fault(vcpu, IDT_PF, 1, error_code); 2241 } 2242 2243 static VMM_STAT(VCPU_NMI_COUNT, "number of NMIs delivered to vcpu"); 2244 2245 int 2246 vm_inject_nmi(struct vcpu *vcpu) 2247 { 2248 2249 vcpu->nmi_pending = 1; 2250 vcpu_notify_event(vcpu, false); 2251 return (0); 2252 } 2253 2254 int 2255 vm_nmi_pending(struct vcpu *vcpu) 2256 { 2257 return (vcpu->nmi_pending); 2258 } 2259 2260 void 2261 vm_nmi_clear(struct vcpu *vcpu) 2262 { 2263 if (vcpu->nmi_pending == 0) 2264 panic("vm_nmi_clear: inconsistent nmi_pending state"); 2265 2266 vcpu->nmi_pending = 0; 2267 vmm_stat_incr(vcpu, VCPU_NMI_COUNT, 1); 2268 } 2269 2270 static VMM_STAT(VCPU_EXTINT_COUNT, "number of ExtINTs delivered to vcpu"); 2271 2272 int 2273 vm_inject_extint(struct vcpu *vcpu) 2274 { 2275 2276 vcpu->extint_pending = 1; 2277 vcpu_notify_event(vcpu, false); 2278 return (0); 2279 } 2280 2281 int 2282 vm_extint_pending(struct vcpu *vcpu) 2283 { 2284 return (vcpu->extint_pending); 2285 } 2286 2287 void 2288 vm_extint_clear(struct vcpu *vcpu) 2289 { 2290 if (vcpu->extint_pending == 0) 2291 panic("vm_extint_clear: inconsistent extint_pending state"); 2292 2293 vcpu->extint_pending = 0; 2294 vmm_stat_incr(vcpu, VCPU_EXTINT_COUNT, 1); 2295 } 2296 2297 int 2298 vm_get_capability(struct vcpu *vcpu, int type, int *retval) 2299 { 2300 if (type < 0 || type >= VM_CAP_MAX) 2301 return (EINVAL); 2302 2303 return (vmmops_getcap(vcpu->cookie, type, retval)); 2304 } 2305 2306 int 2307 vm_set_capability(struct vcpu *vcpu, int type, int val) 2308 { 2309 if (type < 0 || type >= VM_CAP_MAX) 2310 return (EINVAL); 2311 2312 return (vmmops_setcap(vcpu->cookie, type, val)); 2313 } 2314 2315 struct vm * 2316 vcpu_vm(struct vcpu *vcpu) 2317 { 2318 return (vcpu->vm); 2319 } 2320 2321 int 2322 vcpu_vcpuid(struct vcpu *vcpu) 2323 { 2324 return (vcpu->vcpuid); 2325 } 2326 2327 struct vcpu * 2328 vm_vcpu(struct vm *vm, int vcpuid) 2329 { 2330 return (vm->vcpu[vcpuid]); 2331 } 2332 2333 struct vlapic * 2334 vm_lapic(struct vcpu *vcpu) 2335 { 2336 return (vcpu->vlapic); 2337 } 2338 2339 struct vioapic * 2340 vm_ioapic(struct vm *vm) 2341 { 2342 2343 return (vm->vioapic); 2344 } 2345 2346 struct vhpet * 2347 vm_hpet(struct vm *vm) 2348 { 2349 2350 return (vm->vhpet); 2351 } 2352 2353 bool 2354 vmm_is_pptdev(int bus, int slot, int func) 2355 { 2356 int b, f, i, n, s; 2357 char *val, *cp, *cp2; 2358 bool found; 2359 2360 /* 2361 * XXX 2362 * The length of an environment variable is limited to 128 bytes which 2363 * puts an upper limit on the number of passthru devices that may be 2364 * specified using a single environment variable. 2365 * 2366 * Work around this by scanning multiple environment variable 2367 * names instead of a single one - yuck! 2368 */ 2369 const char *names[] = { "pptdevs", "pptdevs2", "pptdevs3", NULL }; 2370 2371 /* set pptdevs="1/2/3 4/5/6 7/8/9 10/11/12" */ 2372 found = false; 2373 for (i = 0; names[i] != NULL && !found; i++) { 2374 cp = val = kern_getenv(names[i]); 2375 while (cp != NULL && *cp != '\0') { 2376 if ((cp2 = strchr(cp, ' ')) != NULL) 2377 *cp2 = '\0'; 2378 2379 n = sscanf(cp, "%d/%d/%d", &b, &s, &f); 2380 if (n == 3 && bus == b && slot == s && func == f) { 2381 found = true; 2382 break; 2383 } 2384 2385 if (cp2 != NULL) 2386 *cp2++ = ' '; 2387 2388 cp = cp2; 2389 } 2390 freeenv(val); 2391 } 2392 return (found); 2393 } 2394 2395 void * 2396 vm_iommu_domain(struct vm *vm) 2397 { 2398 2399 return (vm->iommu); 2400 } 2401 2402 int 2403 vcpu_set_state(struct vcpu *vcpu, enum vcpu_state newstate, bool from_idle) 2404 { 2405 int error; 2406 2407 vcpu_lock(vcpu); 2408 error = vcpu_set_state_locked(vcpu, newstate, from_idle); 2409 vcpu_unlock(vcpu); 2410 2411 return (error); 2412 } 2413 2414 enum vcpu_state 2415 vcpu_get_state(struct vcpu *vcpu, int *hostcpu) 2416 { 2417 enum vcpu_state state; 2418 2419 vcpu_lock(vcpu); 2420 state = vcpu->state; 2421 if (hostcpu != NULL) 2422 *hostcpu = vcpu->hostcpu; 2423 vcpu_unlock(vcpu); 2424 2425 return (state); 2426 } 2427 2428 int 2429 vm_activate_cpu(struct vcpu *vcpu) 2430 { 2431 struct vm *vm = vcpu->vm; 2432 2433 if (CPU_ISSET(vcpu->vcpuid, &vm->active_cpus)) 2434 return (EBUSY); 2435 2436 VMM_CTR0(vcpu, "activated"); 2437 CPU_SET_ATOMIC(vcpu->vcpuid, &vm->active_cpus); 2438 return (0); 2439 } 2440 2441 int 2442 vm_suspend_cpu(struct vm *vm, struct vcpu *vcpu) 2443 { 2444 if (vcpu == NULL) { 2445 vm->debug_cpus = vm->active_cpus; 2446 for (int i = 0; i < vm->maxcpus; i++) { 2447 if (CPU_ISSET(i, &vm->active_cpus)) 2448 vcpu_notify_event(vm_vcpu(vm, i), false); 2449 } 2450 } else { 2451 if (!CPU_ISSET(vcpu->vcpuid, &vm->active_cpus)) 2452 return (EINVAL); 2453 2454 CPU_SET_ATOMIC(vcpu->vcpuid, &vm->debug_cpus); 2455 vcpu_notify_event(vcpu, false); 2456 } 2457 return (0); 2458 } 2459 2460 int 2461 vm_resume_cpu(struct vm *vm, struct vcpu *vcpu) 2462 { 2463 2464 if (vcpu == NULL) { 2465 CPU_ZERO(&vm->debug_cpus); 2466 } else { 2467 if (!CPU_ISSET(vcpu->vcpuid, &vm->debug_cpus)) 2468 return (EINVAL); 2469 2470 CPU_CLR_ATOMIC(vcpu->vcpuid, &vm->debug_cpus); 2471 } 2472 return (0); 2473 } 2474 2475 int 2476 vcpu_debugged(struct vcpu *vcpu) 2477 { 2478 2479 return (CPU_ISSET(vcpu->vcpuid, &vcpu->vm->debug_cpus)); 2480 } 2481 2482 cpuset_t 2483 vm_active_cpus(struct vm *vm) 2484 { 2485 2486 return (vm->active_cpus); 2487 } 2488 2489 cpuset_t 2490 vm_debug_cpus(struct vm *vm) 2491 { 2492 2493 return (vm->debug_cpus); 2494 } 2495 2496 cpuset_t 2497 vm_suspended_cpus(struct vm *vm) 2498 { 2499 2500 return (vm->suspended_cpus); 2501 } 2502 2503 /* 2504 * Returns the subset of vCPUs in tostart that are awaiting startup. 2505 * These vCPUs are also marked as no longer awaiting startup. 2506 */ 2507 cpuset_t 2508 vm_start_cpus(struct vm *vm, const cpuset_t *tostart) 2509 { 2510 cpuset_t set; 2511 2512 mtx_lock(&vm->rendezvous_mtx); 2513 CPU_AND(&set, &vm->startup_cpus, tostart); 2514 CPU_ANDNOT(&vm->startup_cpus, &vm->startup_cpus, &set); 2515 mtx_unlock(&vm->rendezvous_mtx); 2516 return (set); 2517 } 2518 2519 void 2520 vm_await_start(struct vm *vm, const cpuset_t *waiting) 2521 { 2522 mtx_lock(&vm->rendezvous_mtx); 2523 CPU_OR(&vm->startup_cpus, &vm->startup_cpus, waiting); 2524 mtx_unlock(&vm->rendezvous_mtx); 2525 } 2526 2527 void * 2528 vcpu_stats(struct vcpu *vcpu) 2529 { 2530 2531 return (vcpu->stats); 2532 } 2533 2534 int 2535 vm_get_x2apic_state(struct vcpu *vcpu, enum x2apic_state *state) 2536 { 2537 *state = vcpu->x2apic_state; 2538 2539 return (0); 2540 } 2541 2542 int 2543 vm_set_x2apic_state(struct vcpu *vcpu, enum x2apic_state state) 2544 { 2545 if (state >= X2APIC_STATE_LAST) 2546 return (EINVAL); 2547 2548 vcpu->x2apic_state = state; 2549 2550 vlapic_set_x2apic_state(vcpu, state); 2551 2552 return (0); 2553 } 2554 2555 /* 2556 * This function is called to ensure that a vcpu "sees" a pending event 2557 * as soon as possible: 2558 * - If the vcpu thread is sleeping then it is woken up. 2559 * - If the vcpu is running on a different host_cpu then an IPI will be directed 2560 * to the host_cpu to cause the vcpu to trap into the hypervisor. 2561 */ 2562 static void 2563 vcpu_notify_event_locked(struct vcpu *vcpu, bool lapic_intr) 2564 { 2565 int hostcpu; 2566 2567 hostcpu = vcpu->hostcpu; 2568 if (vcpu->state == VCPU_RUNNING) { 2569 KASSERT(hostcpu != NOCPU, ("vcpu running on invalid hostcpu")); 2570 if (hostcpu != curcpu) { 2571 if (lapic_intr) { 2572 vlapic_post_intr(vcpu->vlapic, hostcpu, 2573 vmm_ipinum); 2574 } else { 2575 ipi_cpu(hostcpu, vmm_ipinum); 2576 } 2577 } else { 2578 /* 2579 * If the 'vcpu' is running on 'curcpu' then it must 2580 * be sending a notification to itself (e.g. SELF_IPI). 2581 * The pending event will be picked up when the vcpu 2582 * transitions back to guest context. 2583 */ 2584 } 2585 } else { 2586 KASSERT(hostcpu == NOCPU, ("vcpu state %d not consistent " 2587 "with hostcpu %d", vcpu->state, hostcpu)); 2588 if (vcpu->state == VCPU_SLEEPING) 2589 wakeup_one(vcpu); 2590 } 2591 } 2592 2593 void 2594 vcpu_notify_event(struct vcpu *vcpu, bool lapic_intr) 2595 { 2596 vcpu_lock(vcpu); 2597 vcpu_notify_event_locked(vcpu, lapic_intr); 2598 vcpu_unlock(vcpu); 2599 } 2600 2601 struct vmspace * 2602 vm_get_vmspace(struct vm *vm) 2603 { 2604 2605 return (vm->vmspace); 2606 } 2607 2608 int 2609 vm_apicid2vcpuid(struct vm *vm, int apicid) 2610 { 2611 /* 2612 * XXX apic id is assumed to be numerically identical to vcpu id 2613 */ 2614 return (apicid); 2615 } 2616 2617 int 2618 vm_smp_rendezvous(struct vcpu *vcpu, cpuset_t dest, 2619 vm_rendezvous_func_t func, void *arg) 2620 { 2621 struct vm *vm = vcpu->vm; 2622 int error, i; 2623 2624 /* 2625 * Enforce that this function is called without any locks 2626 */ 2627 WITNESS_WARN(WARN_PANIC, NULL, "vm_smp_rendezvous"); 2628 2629 restart: 2630 mtx_lock(&vm->rendezvous_mtx); 2631 if (vm->rendezvous_func != NULL) { 2632 /* 2633 * If a rendezvous is already in progress then we need to 2634 * call the rendezvous handler in case this 'vcpu' is one 2635 * of the targets of the rendezvous. 2636 */ 2637 VMM_CTR0(vcpu, "Rendezvous already in progress"); 2638 mtx_unlock(&vm->rendezvous_mtx); 2639 error = vm_handle_rendezvous(vcpu); 2640 if (error != 0) 2641 return (error); 2642 goto restart; 2643 } 2644 KASSERT(vm->rendezvous_func == NULL, ("vm_smp_rendezvous: previous " 2645 "rendezvous is still in progress")); 2646 2647 VMM_CTR0(vcpu, "Initiating rendezvous"); 2648 vm->rendezvous_req_cpus = dest; 2649 CPU_ZERO(&vm->rendezvous_done_cpus); 2650 vm->rendezvous_arg = arg; 2651 vm->rendezvous_func = func; 2652 mtx_unlock(&vm->rendezvous_mtx); 2653 2654 /* 2655 * Wake up any sleeping vcpus and trigger a VM-exit in any running 2656 * vcpus so they handle the rendezvous as soon as possible. 2657 */ 2658 for (i = 0; i < vm->maxcpus; i++) { 2659 if (CPU_ISSET(i, &dest)) 2660 vcpu_notify_event(vm_vcpu(vm, i), false); 2661 } 2662 2663 return (vm_handle_rendezvous(vcpu)); 2664 } 2665 2666 struct vatpic * 2667 vm_atpic(struct vm *vm) 2668 { 2669 return (vm->vatpic); 2670 } 2671 2672 struct vatpit * 2673 vm_atpit(struct vm *vm) 2674 { 2675 return (vm->vatpit); 2676 } 2677 2678 struct vpmtmr * 2679 vm_pmtmr(struct vm *vm) 2680 { 2681 2682 return (vm->vpmtmr); 2683 } 2684 2685 struct vrtc * 2686 vm_rtc(struct vm *vm) 2687 { 2688 2689 return (vm->vrtc); 2690 } 2691 2692 enum vm_reg_name 2693 vm_segment_name(int seg) 2694 { 2695 static enum vm_reg_name seg_names[] = { 2696 VM_REG_GUEST_ES, 2697 VM_REG_GUEST_CS, 2698 VM_REG_GUEST_SS, 2699 VM_REG_GUEST_DS, 2700 VM_REG_GUEST_FS, 2701 VM_REG_GUEST_GS 2702 }; 2703 2704 KASSERT(seg >= 0 && seg < nitems(seg_names), 2705 ("%s: invalid segment encoding %d", __func__, seg)); 2706 return (seg_names[seg]); 2707 } 2708 2709 void 2710 vm_copy_teardown(struct vm_copyinfo *copyinfo, int num_copyinfo) 2711 { 2712 int idx; 2713 2714 for (idx = 0; idx < num_copyinfo; idx++) { 2715 if (copyinfo[idx].cookie != NULL) 2716 vm_gpa_release(copyinfo[idx].cookie); 2717 } 2718 bzero(copyinfo, num_copyinfo * sizeof(struct vm_copyinfo)); 2719 } 2720 2721 int 2722 vm_copy_setup(struct vcpu *vcpu, struct vm_guest_paging *paging, 2723 uint64_t gla, size_t len, int prot, struct vm_copyinfo *copyinfo, 2724 int num_copyinfo, int *fault) 2725 { 2726 int error, idx, nused; 2727 size_t n, off, remaining; 2728 void *hva, *cookie; 2729 uint64_t gpa; 2730 2731 bzero(copyinfo, sizeof(struct vm_copyinfo) * num_copyinfo); 2732 2733 nused = 0; 2734 remaining = len; 2735 while (remaining > 0) { 2736 KASSERT(nused < num_copyinfo, ("insufficient vm_copyinfo")); 2737 error = vm_gla2gpa(vcpu, paging, gla, prot, &gpa, fault); 2738 if (error || *fault) 2739 return (error); 2740 off = gpa & PAGE_MASK; 2741 n = min(remaining, PAGE_SIZE - off); 2742 copyinfo[nused].gpa = gpa; 2743 copyinfo[nused].len = n; 2744 remaining -= n; 2745 gla += n; 2746 nused++; 2747 } 2748 2749 for (idx = 0; idx < nused; idx++) { 2750 hva = vm_gpa_hold(vcpu, copyinfo[idx].gpa, 2751 copyinfo[idx].len, prot, &cookie); 2752 if (hva == NULL) 2753 break; 2754 copyinfo[idx].hva = hva; 2755 copyinfo[idx].cookie = cookie; 2756 } 2757 2758 if (idx != nused) { 2759 vm_copy_teardown(copyinfo, num_copyinfo); 2760 return (EFAULT); 2761 } else { 2762 *fault = 0; 2763 return (0); 2764 } 2765 } 2766 2767 void 2768 vm_copyin(struct vm_copyinfo *copyinfo, void *kaddr, size_t len) 2769 { 2770 char *dst; 2771 int idx; 2772 2773 dst = kaddr; 2774 idx = 0; 2775 while (len > 0) { 2776 bcopy(copyinfo[idx].hva, dst, copyinfo[idx].len); 2777 len -= copyinfo[idx].len; 2778 dst += copyinfo[idx].len; 2779 idx++; 2780 } 2781 } 2782 2783 void 2784 vm_copyout(const void *kaddr, struct vm_copyinfo *copyinfo, size_t len) 2785 { 2786 const char *src; 2787 int idx; 2788 2789 src = kaddr; 2790 idx = 0; 2791 while (len > 0) { 2792 bcopy(src, copyinfo[idx].hva, copyinfo[idx].len); 2793 len -= copyinfo[idx].len; 2794 src += copyinfo[idx].len; 2795 idx++; 2796 } 2797 } 2798 2799 /* 2800 * Return the amount of in-use and wired memory for the VM. Since 2801 * these are global stats, only return the values with for vCPU 0 2802 */ 2803 VMM_STAT_DECLARE(VMM_MEM_RESIDENT); 2804 VMM_STAT_DECLARE(VMM_MEM_WIRED); 2805 2806 static void 2807 vm_get_rescnt(struct vcpu *vcpu, struct vmm_stat_type *stat) 2808 { 2809 2810 if (vcpu->vcpuid == 0) { 2811 vmm_stat_set(vcpu, VMM_MEM_RESIDENT, PAGE_SIZE * 2812 vmspace_resident_count(vcpu->vm->vmspace)); 2813 } 2814 } 2815 2816 static void 2817 vm_get_wiredcnt(struct vcpu *vcpu, struct vmm_stat_type *stat) 2818 { 2819 2820 if (vcpu->vcpuid == 0) { 2821 vmm_stat_set(vcpu, VMM_MEM_WIRED, PAGE_SIZE * 2822 pmap_wired_count(vmspace_pmap(vcpu->vm->vmspace))); 2823 } 2824 } 2825 2826 VMM_STAT_FUNC(VMM_MEM_RESIDENT, "Resident memory", vm_get_rescnt); 2827 VMM_STAT_FUNC(VMM_MEM_WIRED, "Wired memory", vm_get_wiredcnt); 2828 2829 #ifdef BHYVE_SNAPSHOT 2830 static int 2831 vm_snapshot_vcpus(struct vm *vm, struct vm_snapshot_meta *meta) 2832 { 2833 uint64_t tsc, now; 2834 int ret; 2835 struct vcpu *vcpu; 2836 uint16_t i, maxcpus; 2837 2838 now = rdtsc(); 2839 maxcpus = vm_get_maxcpus(vm); 2840 for (i = 0; i < maxcpus; i++) { 2841 vcpu = vm->vcpu[i]; 2842 if (vcpu == NULL) 2843 continue; 2844 2845 SNAPSHOT_VAR_OR_LEAVE(vcpu->x2apic_state, meta, ret, done); 2846 SNAPSHOT_VAR_OR_LEAVE(vcpu->exitintinfo, meta, ret, done); 2847 SNAPSHOT_VAR_OR_LEAVE(vcpu->exc_vector, meta, ret, done); 2848 SNAPSHOT_VAR_OR_LEAVE(vcpu->exc_errcode_valid, meta, ret, done); 2849 SNAPSHOT_VAR_OR_LEAVE(vcpu->exc_errcode, meta, ret, done); 2850 SNAPSHOT_VAR_OR_LEAVE(vcpu->guest_xcr0, meta, ret, done); 2851 SNAPSHOT_VAR_OR_LEAVE(vcpu->exitinfo, meta, ret, done); 2852 SNAPSHOT_VAR_OR_LEAVE(vcpu->nextrip, meta, ret, done); 2853 2854 /* 2855 * Save the absolute TSC value by adding now to tsc_offset. 2856 * 2857 * It will be turned turned back into an actual offset when the 2858 * TSC restore function is called 2859 */ 2860 tsc = now + vcpu->tsc_offset; 2861 SNAPSHOT_VAR_OR_LEAVE(tsc, meta, ret, done); 2862 if (meta->op == VM_SNAPSHOT_RESTORE) 2863 vcpu->tsc_offset = tsc; 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_VMCX: 2921 ret = vm_snapshot_vcpu(vm, meta); 2922 break; 2923 case STRUCT_VM: 2924 ret = vm_snapshot_vm(vm, meta); 2925 break; 2926 case STRUCT_VIOAPIC: 2927 ret = vioapic_snapshot(vm_ioapic(vm), meta); 2928 break; 2929 case STRUCT_VLAPIC: 2930 ret = vlapic_snapshot(vm, meta); 2931 break; 2932 case STRUCT_VHPET: 2933 ret = vhpet_snapshot(vm_hpet(vm), meta); 2934 break; 2935 case STRUCT_VATPIC: 2936 ret = vatpic_snapshot(vm_atpic(vm), meta); 2937 break; 2938 case STRUCT_VATPIT: 2939 ret = vatpit_snapshot(vm_atpit(vm), meta); 2940 break; 2941 case STRUCT_VPMTMR: 2942 ret = vpmtmr_snapshot(vm_pmtmr(vm), meta); 2943 break; 2944 case STRUCT_VRTC: 2945 ret = vrtc_snapshot(vm_rtc(vm), meta); 2946 break; 2947 default: 2948 printf("%s: failed to find the requested type %#x\n", 2949 __func__, meta->dev_req); 2950 ret = (EINVAL); 2951 } 2952 return (ret); 2953 } 2954 2955 void 2956 vm_set_tsc_offset(struct vcpu *vcpu, uint64_t offset) 2957 { 2958 vcpu->tsc_offset = offset; 2959 } 2960 2961 int 2962 vm_restore_time(struct vm *vm) 2963 { 2964 int error; 2965 uint64_t now; 2966 struct vcpu *vcpu; 2967 uint16_t i, maxcpus; 2968 2969 now = rdtsc(); 2970 2971 error = vhpet_restore_time(vm_hpet(vm)); 2972 if (error) 2973 return (error); 2974 2975 maxcpus = vm_get_maxcpus(vm); 2976 for (i = 0; i < maxcpus; i++) { 2977 vcpu = vm->vcpu[i]; 2978 if (vcpu == NULL) 2979 continue; 2980 2981 error = vmmops_restore_tsc(vcpu->cookie, 2982 vcpu->tsc_offset - now); 2983 if (error) 2984 return (error); 2985 } 2986 2987 return (0); 2988 } 2989 #endif 2990