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 SDT_PROVIDER_DEFINE(vmm); 260 261 static MALLOC_DEFINE(M_VM, "vm", "vm"); 262 263 /* statistics */ 264 static VMM_STAT(VCPU_TOTAL_RUNTIME, "vcpu total runtime"); 265 266 SYSCTL_NODE(_hw, OID_AUTO, vmm, CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, 267 NULL); 268 269 /* 270 * Halt the guest if all vcpus are executing a HLT instruction with 271 * interrupts disabled. 272 */ 273 static int halt_detection_enabled = 1; 274 SYSCTL_INT(_hw_vmm, OID_AUTO, halt_detection, CTLFLAG_RDTUN, 275 &halt_detection_enabled, 0, 276 "Halt VM if all vcpus execute HLT with interrupts disabled"); 277 278 static int vmm_ipinum; 279 SYSCTL_INT(_hw_vmm, OID_AUTO, ipinum, CTLFLAG_RD, &vmm_ipinum, 0, 280 "IPI vector used for vcpu notifications"); 281 282 static int trace_guest_exceptions; 283 SYSCTL_INT(_hw_vmm, OID_AUTO, trace_guest_exceptions, CTLFLAG_RDTUN, 284 &trace_guest_exceptions, 0, 285 "Trap into hypervisor on all guest exceptions and reflect them back"); 286 287 static int trap_wbinvd; 288 SYSCTL_INT(_hw_vmm, OID_AUTO, trap_wbinvd, CTLFLAG_RDTUN, &trap_wbinvd, 0, 289 "WBINVD triggers a VM-exit"); 290 291 u_int vm_maxcpu; 292 SYSCTL_UINT(_hw_vmm, OID_AUTO, maxcpu, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, 293 &vm_maxcpu, 0, "Maximum number of vCPUs"); 294 295 static void vm_free_memmap(struct vm *vm, int ident); 296 static bool sysmem_mapping(struct vm *vm, struct mem_map *mm); 297 static void vcpu_notify_event_locked(struct vcpu *vcpu, bool lapic_intr); 298 299 /* 300 * Upper limit on vm_maxcpu. Limited by use of uint16_t types for CPU 301 * counts as well as range of vpid values for VT-x and by the capacity 302 * of cpuset_t masks. The call to new_unrhdr() in vpid_init() in 303 * vmx.c requires 'vm_maxcpu + 1 <= 0xffff', hence the '- 1' below. 304 */ 305 #define VM_MAXCPU MIN(0xffff - 1, CPU_SETSIZE) 306 307 #ifdef KTR 308 static const char * 309 vcpu_state2str(enum vcpu_state state) 310 { 311 312 switch (state) { 313 case VCPU_IDLE: 314 return ("idle"); 315 case VCPU_FROZEN: 316 return ("frozen"); 317 case VCPU_RUNNING: 318 return ("running"); 319 case VCPU_SLEEPING: 320 return ("sleeping"); 321 default: 322 return ("unknown"); 323 } 324 } 325 #endif 326 327 static void 328 vcpu_cleanup(struct vcpu *vcpu, bool destroy) 329 { 330 vmmops_vlapic_cleanup(vcpu->vlapic); 331 vmmops_vcpu_cleanup(vcpu->cookie); 332 vcpu->cookie = NULL; 333 if (destroy) { 334 vmm_stat_free(vcpu->stats); 335 fpu_save_area_free(vcpu->guestfpu); 336 vcpu_lock_destroy(vcpu); 337 free(vcpu, M_VM); 338 } 339 } 340 341 static struct vcpu * 342 vcpu_alloc(struct vm *vm, int vcpu_id) 343 { 344 struct vcpu *vcpu; 345 346 KASSERT(vcpu_id >= 0 && vcpu_id < vm->maxcpus, 347 ("vcpu_init: invalid vcpu %d", vcpu_id)); 348 349 vcpu = malloc(sizeof(*vcpu), M_VM, M_WAITOK | M_ZERO); 350 vcpu_lock_init(vcpu); 351 vcpu->state = VCPU_IDLE; 352 vcpu->hostcpu = NOCPU; 353 vcpu->vcpuid = vcpu_id; 354 vcpu->vm = vm; 355 vcpu->guestfpu = fpu_save_area_alloc(); 356 vcpu->stats = vmm_stat_alloc(); 357 vcpu->tsc_offset = 0; 358 return (vcpu); 359 } 360 361 static void 362 vcpu_init(struct vcpu *vcpu) 363 { 364 vcpu->cookie = vmmops_vcpu_init(vcpu->vm->cookie, vcpu, vcpu->vcpuid); 365 vcpu->vlapic = vmmops_vlapic_init(vcpu->cookie); 366 vm_set_x2apic_state(vcpu, X2APIC_DISABLED); 367 vcpu->reqidle = 0; 368 vcpu->exitintinfo = 0; 369 vcpu->nmi_pending = 0; 370 vcpu->extint_pending = 0; 371 vcpu->exception_pending = 0; 372 vcpu->guest_xcr0 = XFEATURE_ENABLED_X87; 373 fpu_save_area_reset(vcpu->guestfpu); 374 vmm_stat_init(vcpu->stats); 375 } 376 377 int 378 vcpu_trace_exceptions(struct vcpu *vcpu) 379 { 380 381 return (trace_guest_exceptions); 382 } 383 384 int 385 vcpu_trap_wbinvd(struct vcpu *vcpu) 386 { 387 return (trap_wbinvd); 388 } 389 390 struct vm_exit * 391 vm_exitinfo(struct vcpu *vcpu) 392 { 393 return (&vcpu->exitinfo); 394 } 395 396 cpuset_t * 397 vm_exitinfo_cpuset(struct vcpu *vcpu) 398 { 399 return (&vcpu->exitinfo_cpuset); 400 } 401 402 static int 403 vmm_init(void) 404 { 405 int error; 406 407 if (!vmm_is_hw_supported()) 408 return (ENXIO); 409 410 vm_maxcpu = mp_ncpus; 411 TUNABLE_INT_FETCH("hw.vmm.maxcpu", &vm_maxcpu); 412 413 if (vm_maxcpu > VM_MAXCPU) { 414 printf("vmm: vm_maxcpu clamped to %u\n", VM_MAXCPU); 415 vm_maxcpu = VM_MAXCPU; 416 } 417 if (vm_maxcpu == 0) 418 vm_maxcpu = 1; 419 420 vmm_host_state_init(); 421 422 vmm_ipinum = lapic_ipi_alloc(pti ? &IDTVEC(justreturn1_pti) : 423 &IDTVEC(justreturn)); 424 if (vmm_ipinum < 0) 425 vmm_ipinum = IPI_AST; 426 427 error = vmm_mem_init(); 428 if (error) 429 return (error); 430 431 vmm_resume_p = vmmops_modresume; 432 433 return (vmmops_modinit(vmm_ipinum)); 434 } 435 436 static int 437 vmm_handler(module_t mod, int what, void *arg) 438 { 439 int error; 440 441 switch (what) { 442 case MOD_LOAD: 443 if (vmm_is_hw_supported()) { 444 vmmdev_init(); 445 error = vmm_init(); 446 if (error == 0) 447 vmm_initialized = 1; 448 } else { 449 error = ENXIO; 450 } 451 break; 452 case MOD_UNLOAD: 453 if (vmm_is_hw_supported()) { 454 error = vmmdev_cleanup(); 455 if (error == 0) { 456 vmm_resume_p = NULL; 457 iommu_cleanup(); 458 if (vmm_ipinum != IPI_AST) 459 lapic_ipi_free(vmm_ipinum); 460 error = vmmops_modcleanup(); 461 /* 462 * Something bad happened - prevent new 463 * VMs from being created 464 */ 465 if (error) 466 vmm_initialized = 0; 467 } 468 } else { 469 error = 0; 470 } 471 break; 472 default: 473 error = 0; 474 break; 475 } 476 return (error); 477 } 478 479 static moduledata_t vmm_kmod = { 480 "vmm", 481 vmm_handler, 482 NULL 483 }; 484 485 /* 486 * vmm initialization has the following dependencies: 487 * 488 * - VT-x initialization requires smp_rendezvous() and therefore must happen 489 * after SMP is fully functional (after SI_SUB_SMP). 490 */ 491 DECLARE_MODULE(vmm, vmm_kmod, SI_SUB_SMP + 1, SI_ORDER_ANY); 492 MODULE_VERSION(vmm, 1); 493 494 static void 495 vm_init(struct vm *vm, bool create) 496 { 497 vm->cookie = vmmops_init(vm, vmspace_pmap(vm->vmspace)); 498 vm->iommu = NULL; 499 vm->vioapic = vioapic_init(vm); 500 vm->vhpet = vhpet_init(vm); 501 vm->vatpic = vatpic_init(vm); 502 vm->vatpit = vatpit_init(vm); 503 vm->vpmtmr = vpmtmr_init(vm); 504 if (create) 505 vm->vrtc = vrtc_init(vm); 506 507 CPU_ZERO(&vm->active_cpus); 508 CPU_ZERO(&vm->debug_cpus); 509 CPU_ZERO(&vm->startup_cpus); 510 511 vm->suspend = 0; 512 CPU_ZERO(&vm->suspended_cpus); 513 514 if (!create) { 515 for (int i = 0; i < vm->maxcpus; i++) { 516 if (vm->vcpu[i] != NULL) 517 vcpu_init(vm->vcpu[i]); 518 } 519 } 520 } 521 522 void 523 vm_disable_vcpu_creation(struct vm *vm) 524 { 525 sx_xlock(&vm->vcpus_init_lock); 526 vm->dying = true; 527 sx_xunlock(&vm->vcpus_init_lock); 528 } 529 530 struct vcpu * 531 vm_alloc_vcpu(struct vm *vm, int vcpuid) 532 { 533 struct vcpu *vcpu; 534 535 if (vcpuid < 0 || vcpuid >= vm_get_maxcpus(vm)) 536 return (NULL); 537 538 vcpu = atomic_load_ptr(&vm->vcpu[vcpuid]); 539 if (__predict_true(vcpu != NULL)) 540 return (vcpu); 541 542 sx_xlock(&vm->vcpus_init_lock); 543 vcpu = vm->vcpu[vcpuid]; 544 if (vcpu == NULL && !vm->dying) { 545 vcpu = vcpu_alloc(vm, vcpuid); 546 vcpu_init(vcpu); 547 548 /* 549 * Ensure vCPU is fully created before updating pointer 550 * to permit unlocked reads above. 551 */ 552 atomic_store_rel_ptr((uintptr_t *)&vm->vcpu[vcpuid], 553 (uintptr_t)vcpu); 554 } 555 sx_xunlock(&vm->vcpus_init_lock); 556 return (vcpu); 557 } 558 559 void 560 vm_slock_vcpus(struct vm *vm) 561 { 562 sx_slock(&vm->vcpus_init_lock); 563 } 564 565 void 566 vm_unlock_vcpus(struct vm *vm) 567 { 568 sx_unlock(&vm->vcpus_init_lock); 569 } 570 571 /* 572 * The default CPU topology is a single thread per package. 573 */ 574 u_int cores_per_package = 1; 575 u_int threads_per_core = 1; 576 577 int 578 vm_create(const char *name, struct vm **retvm) 579 { 580 struct vm *vm; 581 struct vmspace *vmspace; 582 583 /* 584 * If vmm.ko could not be successfully initialized then don't attempt 585 * to create the virtual machine. 586 */ 587 if (!vmm_initialized) 588 return (ENXIO); 589 590 if (name == NULL || strnlen(name, VM_MAX_NAMELEN + 1) == 591 VM_MAX_NAMELEN + 1) 592 return (EINVAL); 593 594 vmspace = vmmops_vmspace_alloc(0, VM_MAXUSER_ADDRESS_LA48); 595 if (vmspace == NULL) 596 return (ENOMEM); 597 598 vm = malloc(sizeof(struct vm), M_VM, M_WAITOK | M_ZERO); 599 strcpy(vm->name, name); 600 vm->vmspace = vmspace; 601 mtx_init(&vm->rendezvous_mtx, "vm rendezvous lock", 0, MTX_DEF); 602 sx_init(&vm->mem_segs_lock, "vm mem_segs"); 603 sx_init(&vm->vcpus_init_lock, "vm vcpus"); 604 vm->vcpu = malloc(sizeof(*vm->vcpu) * vm_maxcpu, M_VM, M_WAITOK | 605 M_ZERO); 606 607 vm->sockets = 1; 608 vm->cores = cores_per_package; /* XXX backwards compatibility */ 609 vm->threads = threads_per_core; /* XXX backwards compatibility */ 610 vm->maxcpus = vm_maxcpu; 611 612 vm_init(vm, true); 613 614 *retvm = vm; 615 return (0); 616 } 617 618 void 619 vm_get_topology(struct vm *vm, uint16_t *sockets, uint16_t *cores, 620 uint16_t *threads, uint16_t *maxcpus) 621 { 622 *sockets = vm->sockets; 623 *cores = vm->cores; 624 *threads = vm->threads; 625 *maxcpus = vm->maxcpus; 626 } 627 628 uint16_t 629 vm_get_maxcpus(struct vm *vm) 630 { 631 return (vm->maxcpus); 632 } 633 634 int 635 vm_set_topology(struct vm *vm, uint16_t sockets, uint16_t cores, 636 uint16_t threads, uint16_t maxcpus __unused) 637 { 638 /* Ignore maxcpus. */ 639 if ((sockets * cores * threads) > vm->maxcpus) 640 return (EINVAL); 641 vm->sockets = sockets; 642 vm->cores = cores; 643 vm->threads = threads; 644 return(0); 645 } 646 647 static void 648 vm_cleanup(struct vm *vm, bool destroy) 649 { 650 struct mem_map *mm; 651 int i; 652 653 if (destroy) 654 vm_xlock_memsegs(vm); 655 656 ppt_unassign_all(vm); 657 658 if (vm->iommu != NULL) 659 iommu_destroy_domain(vm->iommu); 660 661 if (destroy) 662 vrtc_cleanup(vm->vrtc); 663 else 664 vrtc_reset(vm->vrtc); 665 vpmtmr_cleanup(vm->vpmtmr); 666 vatpit_cleanup(vm->vatpit); 667 vhpet_cleanup(vm->vhpet); 668 vatpic_cleanup(vm->vatpic); 669 vioapic_cleanup(vm->vioapic); 670 671 for (i = 0; i < vm->maxcpus; i++) { 672 if (vm->vcpu[i] != NULL) 673 vcpu_cleanup(vm->vcpu[i], destroy); 674 } 675 676 vmmops_cleanup(vm->cookie); 677 678 /* 679 * System memory is removed from the guest address space only when 680 * the VM is destroyed. This is because the mapping remains the same 681 * across VM reset. 682 * 683 * Device memory can be relocated by the guest (e.g. using PCI BARs) 684 * so those mappings are removed on a VM reset. 685 */ 686 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 687 mm = &vm->mem_maps[i]; 688 if (destroy || !sysmem_mapping(vm, mm)) 689 vm_free_memmap(vm, i); 690 } 691 692 if (destroy) { 693 for (i = 0; i < VM_MAX_MEMSEGS; i++) 694 vm_free_memseg(vm, i); 695 vm_unlock_memsegs(vm); 696 697 vmmops_vmspace_free(vm->vmspace); 698 vm->vmspace = NULL; 699 700 free(vm->vcpu, M_VM); 701 sx_destroy(&vm->vcpus_init_lock); 702 sx_destroy(&vm->mem_segs_lock); 703 mtx_destroy(&vm->rendezvous_mtx); 704 } 705 } 706 707 void 708 vm_destroy(struct vm *vm) 709 { 710 vm_cleanup(vm, true); 711 free(vm, M_VM); 712 } 713 714 int 715 vm_reinit(struct vm *vm) 716 { 717 int error; 718 719 /* 720 * A virtual machine can be reset only if all vcpus are suspended. 721 */ 722 if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) { 723 vm_cleanup(vm, false); 724 vm_init(vm, false); 725 error = 0; 726 } else { 727 error = EBUSY; 728 } 729 730 return (error); 731 } 732 733 const char * 734 vm_name(struct vm *vm) 735 { 736 return (vm->name); 737 } 738 739 void 740 vm_slock_memsegs(struct vm *vm) 741 { 742 sx_slock(&vm->mem_segs_lock); 743 } 744 745 void 746 vm_xlock_memsegs(struct vm *vm) 747 { 748 sx_xlock(&vm->mem_segs_lock); 749 } 750 751 void 752 vm_unlock_memsegs(struct vm *vm) 753 { 754 sx_unlock(&vm->mem_segs_lock); 755 } 756 757 int 758 vm_map_mmio(struct vm *vm, vm_paddr_t gpa, size_t len, vm_paddr_t hpa) 759 { 760 vm_object_t obj; 761 762 if ((obj = vmm_mmio_alloc(vm->vmspace, gpa, len, hpa)) == NULL) 763 return (ENOMEM); 764 else 765 return (0); 766 } 767 768 int 769 vm_unmap_mmio(struct vm *vm, vm_paddr_t gpa, size_t len) 770 { 771 772 vmm_mmio_free(vm->vmspace, gpa, len); 773 return (0); 774 } 775 776 /* 777 * Return 'true' if 'gpa' is allocated in the guest address space. 778 * 779 * This function is called in the context of a running vcpu which acts as 780 * an implicit lock on 'vm->mem_maps[]'. 781 */ 782 bool 783 vm_mem_allocated(struct vcpu *vcpu, vm_paddr_t gpa) 784 { 785 struct vm *vm = vcpu->vm; 786 struct mem_map *mm; 787 int i; 788 789 #ifdef INVARIANTS 790 int hostcpu, state; 791 state = vcpu_get_state(vcpu, &hostcpu); 792 KASSERT(state == VCPU_RUNNING && hostcpu == curcpu, 793 ("%s: invalid vcpu state %d/%d", __func__, state, hostcpu)); 794 #endif 795 796 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 797 mm = &vm->mem_maps[i]; 798 if (mm->len != 0 && gpa >= mm->gpa && gpa < mm->gpa + mm->len) 799 return (true); /* 'gpa' is sysmem or devmem */ 800 } 801 802 if (ppt_is_mmio(vm, gpa)) 803 return (true); /* 'gpa' is pci passthru mmio */ 804 805 return (false); 806 } 807 808 int 809 vm_alloc_memseg(struct vm *vm, int ident, size_t len, bool sysmem) 810 { 811 struct mem_seg *seg; 812 vm_object_t obj; 813 814 sx_assert(&vm->mem_segs_lock, SX_XLOCKED); 815 816 if (ident < 0 || ident >= VM_MAX_MEMSEGS) 817 return (EINVAL); 818 819 if (len == 0 || (len & PAGE_MASK)) 820 return (EINVAL); 821 822 seg = &vm->mem_segs[ident]; 823 if (seg->object != NULL) { 824 if (seg->len == len && seg->sysmem == sysmem) 825 return (EEXIST); 826 else 827 return (EINVAL); 828 } 829 830 obj = vm_object_allocate(OBJT_SWAP, len >> PAGE_SHIFT); 831 if (obj == NULL) 832 return (ENOMEM); 833 834 seg->len = len; 835 seg->object = obj; 836 seg->sysmem = sysmem; 837 return (0); 838 } 839 840 int 841 vm_get_memseg(struct vm *vm, int ident, size_t *len, bool *sysmem, 842 vm_object_t *objptr) 843 { 844 struct mem_seg *seg; 845 846 sx_assert(&vm->mem_segs_lock, SX_LOCKED); 847 848 if (ident < 0 || ident >= VM_MAX_MEMSEGS) 849 return (EINVAL); 850 851 seg = &vm->mem_segs[ident]; 852 if (len) 853 *len = seg->len; 854 if (sysmem) 855 *sysmem = seg->sysmem; 856 if (objptr) 857 *objptr = seg->object; 858 return (0); 859 } 860 861 void 862 vm_free_memseg(struct vm *vm, int ident) 863 { 864 struct mem_seg *seg; 865 866 KASSERT(ident >= 0 && ident < VM_MAX_MEMSEGS, 867 ("%s: invalid memseg ident %d", __func__, ident)); 868 869 seg = &vm->mem_segs[ident]; 870 if (seg->object != NULL) { 871 vm_object_deallocate(seg->object); 872 bzero(seg, sizeof(struct mem_seg)); 873 } 874 } 875 876 int 877 vm_mmap_memseg(struct vm *vm, vm_paddr_t gpa, int segid, vm_ooffset_t first, 878 size_t len, int prot, int flags) 879 { 880 struct mem_seg *seg; 881 struct mem_map *m, *map; 882 vm_ooffset_t last; 883 int i, error; 884 885 if (prot == 0 || (prot & ~(VM_PROT_ALL)) != 0) 886 return (EINVAL); 887 888 if (flags & ~VM_MEMMAP_F_WIRED) 889 return (EINVAL); 890 891 if (segid < 0 || segid >= VM_MAX_MEMSEGS) 892 return (EINVAL); 893 894 seg = &vm->mem_segs[segid]; 895 if (seg->object == NULL) 896 return (EINVAL); 897 898 last = first + len; 899 if (first < 0 || first >= last || last > seg->len) 900 return (EINVAL); 901 902 if ((gpa | first | last) & PAGE_MASK) 903 return (EINVAL); 904 905 map = NULL; 906 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 907 m = &vm->mem_maps[i]; 908 if (m->len == 0) { 909 map = m; 910 break; 911 } 912 } 913 914 if (map == NULL) 915 return (ENOSPC); 916 917 error = vm_map_find(&vm->vmspace->vm_map, seg->object, first, &gpa, 918 len, 0, VMFS_NO_SPACE, prot, prot, 0); 919 if (error != KERN_SUCCESS) 920 return (EFAULT); 921 922 vm_object_reference(seg->object); 923 924 if (flags & VM_MEMMAP_F_WIRED) { 925 error = vm_map_wire(&vm->vmspace->vm_map, gpa, gpa + len, 926 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES); 927 if (error != KERN_SUCCESS) { 928 vm_map_remove(&vm->vmspace->vm_map, gpa, gpa + len); 929 return (error == KERN_RESOURCE_SHORTAGE ? ENOMEM : 930 EFAULT); 931 } 932 } 933 934 map->gpa = gpa; 935 map->len = len; 936 map->segoff = first; 937 map->segid = segid; 938 map->prot = prot; 939 map->flags = flags; 940 return (0); 941 } 942 943 int 944 vm_munmap_memseg(struct vm *vm, vm_paddr_t gpa, size_t len) 945 { 946 struct mem_map *m; 947 int i; 948 949 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 950 m = &vm->mem_maps[i]; 951 if (m->gpa == gpa && m->len == len && 952 (m->flags & VM_MEMMAP_F_IOMMU) == 0) { 953 vm_free_memmap(vm, i); 954 return (0); 955 } 956 } 957 958 return (EINVAL); 959 } 960 961 int 962 vm_mmap_getnext(struct vm *vm, vm_paddr_t *gpa, int *segid, 963 vm_ooffset_t *segoff, size_t *len, int *prot, int *flags) 964 { 965 struct mem_map *mm, *mmnext; 966 int i; 967 968 mmnext = NULL; 969 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 970 mm = &vm->mem_maps[i]; 971 if (mm->len == 0 || mm->gpa < *gpa) 972 continue; 973 if (mmnext == NULL || mm->gpa < mmnext->gpa) 974 mmnext = mm; 975 } 976 977 if (mmnext != NULL) { 978 *gpa = mmnext->gpa; 979 if (segid) 980 *segid = mmnext->segid; 981 if (segoff) 982 *segoff = mmnext->segoff; 983 if (len) 984 *len = mmnext->len; 985 if (prot) 986 *prot = mmnext->prot; 987 if (flags) 988 *flags = mmnext->flags; 989 return (0); 990 } else { 991 return (ENOENT); 992 } 993 } 994 995 static void 996 vm_free_memmap(struct vm *vm, int ident) 997 { 998 struct mem_map *mm; 999 int error __diagused; 1000 1001 mm = &vm->mem_maps[ident]; 1002 if (mm->len) { 1003 error = vm_map_remove(&vm->vmspace->vm_map, mm->gpa, 1004 mm->gpa + mm->len); 1005 KASSERT(error == KERN_SUCCESS, ("%s: vm_map_remove error %d", 1006 __func__, error)); 1007 bzero(mm, sizeof(struct mem_map)); 1008 } 1009 } 1010 1011 static __inline bool 1012 sysmem_mapping(struct vm *vm, struct mem_map *mm) 1013 { 1014 1015 if (mm->len != 0 && vm->mem_segs[mm->segid].sysmem) 1016 return (true); 1017 else 1018 return (false); 1019 } 1020 1021 vm_paddr_t 1022 vmm_sysmem_maxaddr(struct vm *vm) 1023 { 1024 struct mem_map *mm; 1025 vm_paddr_t maxaddr; 1026 int i; 1027 1028 maxaddr = 0; 1029 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 1030 mm = &vm->mem_maps[i]; 1031 if (sysmem_mapping(vm, mm)) { 1032 if (maxaddr < mm->gpa + mm->len) 1033 maxaddr = mm->gpa + mm->len; 1034 } 1035 } 1036 return (maxaddr); 1037 } 1038 1039 static void 1040 vm_iommu_modify(struct vm *vm, bool map) 1041 { 1042 int i, sz; 1043 vm_paddr_t gpa, hpa; 1044 struct mem_map *mm; 1045 void *vp, *cookie, *host_domain; 1046 1047 sz = PAGE_SIZE; 1048 host_domain = iommu_host_domain(); 1049 1050 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 1051 mm = &vm->mem_maps[i]; 1052 if (!sysmem_mapping(vm, mm)) 1053 continue; 1054 1055 if (map) { 1056 KASSERT((mm->flags & VM_MEMMAP_F_IOMMU) == 0, 1057 ("iommu map found invalid memmap %#lx/%#lx/%#x", 1058 mm->gpa, mm->len, mm->flags)); 1059 if ((mm->flags & VM_MEMMAP_F_WIRED) == 0) 1060 continue; 1061 mm->flags |= VM_MEMMAP_F_IOMMU; 1062 } else { 1063 if ((mm->flags & VM_MEMMAP_F_IOMMU) == 0) 1064 continue; 1065 mm->flags &= ~VM_MEMMAP_F_IOMMU; 1066 KASSERT((mm->flags & VM_MEMMAP_F_WIRED) != 0, 1067 ("iommu unmap found invalid memmap %#lx/%#lx/%#x", 1068 mm->gpa, mm->len, mm->flags)); 1069 } 1070 1071 gpa = mm->gpa; 1072 while (gpa < mm->gpa + mm->len) { 1073 vp = vm_gpa_hold_global(vm, gpa, PAGE_SIZE, 1074 VM_PROT_WRITE, &cookie); 1075 KASSERT(vp != NULL, ("vm(%s) could not map gpa %#lx", 1076 vm_name(vm), gpa)); 1077 1078 vm_gpa_release(cookie); 1079 1080 hpa = DMAP_TO_PHYS((uintptr_t)vp); 1081 if (map) { 1082 iommu_create_mapping(vm->iommu, gpa, hpa, sz); 1083 } else { 1084 iommu_remove_mapping(vm->iommu, gpa, sz); 1085 } 1086 1087 gpa += PAGE_SIZE; 1088 } 1089 } 1090 1091 /* 1092 * Invalidate the cached translations associated with the domain 1093 * from which pages were removed. 1094 */ 1095 if (map) 1096 iommu_invalidate_tlb(host_domain); 1097 else 1098 iommu_invalidate_tlb(vm->iommu); 1099 } 1100 1101 #define vm_iommu_unmap(vm) vm_iommu_modify((vm), false) 1102 #define vm_iommu_map(vm) vm_iommu_modify((vm), true) 1103 1104 int 1105 vm_unassign_pptdev(struct vm *vm, int bus, int slot, int func) 1106 { 1107 int error; 1108 1109 error = ppt_unassign_device(vm, bus, slot, func); 1110 if (error) 1111 return (error); 1112 1113 if (ppt_assigned_devices(vm) == 0) 1114 vm_iommu_unmap(vm); 1115 1116 return (0); 1117 } 1118 1119 int 1120 vm_assign_pptdev(struct vm *vm, int bus, int slot, int func) 1121 { 1122 int error; 1123 vm_paddr_t maxaddr; 1124 1125 /* Set up the IOMMU to do the 'gpa' to 'hpa' translation */ 1126 if (ppt_assigned_devices(vm) == 0) { 1127 KASSERT(vm->iommu == NULL, 1128 ("vm_assign_pptdev: iommu must be NULL")); 1129 maxaddr = vmm_sysmem_maxaddr(vm); 1130 vm->iommu = iommu_create_domain(maxaddr); 1131 if (vm->iommu == NULL) 1132 return (ENXIO); 1133 vm_iommu_map(vm); 1134 } 1135 1136 error = ppt_assign_device(vm, bus, slot, func); 1137 return (error); 1138 } 1139 1140 static void * 1141 _vm_gpa_hold(struct vm *vm, vm_paddr_t gpa, size_t len, int reqprot, 1142 void **cookie) 1143 { 1144 int i, count, pageoff; 1145 struct mem_map *mm; 1146 vm_page_t m; 1147 1148 pageoff = gpa & PAGE_MASK; 1149 if (len > PAGE_SIZE - pageoff) 1150 panic("vm_gpa_hold: invalid gpa/len: 0x%016lx/%lu", gpa, len); 1151 1152 count = 0; 1153 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 1154 mm = &vm->mem_maps[i]; 1155 if (gpa >= mm->gpa && gpa < mm->gpa + mm->len) { 1156 count = vm_fault_quick_hold_pages(&vm->vmspace->vm_map, 1157 trunc_page(gpa), PAGE_SIZE, reqprot, &m, 1); 1158 break; 1159 } 1160 } 1161 1162 if (count == 1) { 1163 *cookie = m; 1164 return ((void *)(PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m)) + pageoff)); 1165 } else { 1166 *cookie = NULL; 1167 return (NULL); 1168 } 1169 } 1170 1171 void * 1172 vm_gpa_hold(struct vcpu *vcpu, vm_paddr_t gpa, size_t len, int reqprot, 1173 void **cookie) 1174 { 1175 #ifdef INVARIANTS 1176 /* 1177 * The current vcpu should be frozen to ensure 'vm_memmap[]' 1178 * stability. 1179 */ 1180 int state = vcpu_get_state(vcpu, NULL); 1181 KASSERT(state == VCPU_FROZEN, ("%s: invalid vcpu state %d", 1182 __func__, state)); 1183 #endif 1184 return (_vm_gpa_hold(vcpu->vm, gpa, len, reqprot, cookie)); 1185 } 1186 1187 void * 1188 vm_gpa_hold_global(struct vm *vm, vm_paddr_t gpa, size_t len, int reqprot, 1189 void **cookie) 1190 { 1191 sx_assert(&vm->mem_segs_lock, SX_LOCKED); 1192 return (_vm_gpa_hold(vm, gpa, len, reqprot, cookie)); 1193 } 1194 1195 void 1196 vm_gpa_release(void *cookie) 1197 { 1198 vm_page_t m = cookie; 1199 1200 vm_page_unwire(m, PQ_ACTIVE); 1201 } 1202 1203 int 1204 vm_get_register(struct vcpu *vcpu, int reg, uint64_t *retval) 1205 { 1206 1207 if (reg >= VM_REG_LAST) 1208 return (EINVAL); 1209 1210 return (vmmops_getreg(vcpu->cookie, reg, retval)); 1211 } 1212 1213 int 1214 vm_set_register(struct vcpu *vcpu, int reg, uint64_t val) 1215 { 1216 int error; 1217 1218 if (reg >= VM_REG_LAST) 1219 return (EINVAL); 1220 1221 error = vmmops_setreg(vcpu->cookie, reg, val); 1222 if (error || reg != VM_REG_GUEST_RIP) 1223 return (error); 1224 1225 /* Set 'nextrip' to match the value of %rip */ 1226 VMM_CTR1(vcpu, "Setting nextrip to %#lx", val); 1227 vcpu->nextrip = val; 1228 return (0); 1229 } 1230 1231 static bool 1232 is_descriptor_table(int reg) 1233 { 1234 1235 switch (reg) { 1236 case VM_REG_GUEST_IDTR: 1237 case VM_REG_GUEST_GDTR: 1238 return (true); 1239 default: 1240 return (false); 1241 } 1242 } 1243 1244 static bool 1245 is_segment_register(int reg) 1246 { 1247 1248 switch (reg) { 1249 case VM_REG_GUEST_ES: 1250 case VM_REG_GUEST_CS: 1251 case VM_REG_GUEST_SS: 1252 case VM_REG_GUEST_DS: 1253 case VM_REG_GUEST_FS: 1254 case VM_REG_GUEST_GS: 1255 case VM_REG_GUEST_TR: 1256 case VM_REG_GUEST_LDTR: 1257 return (true); 1258 default: 1259 return (false); 1260 } 1261 } 1262 1263 int 1264 vm_get_seg_desc(struct vcpu *vcpu, int reg, struct seg_desc *desc) 1265 { 1266 1267 if (!is_segment_register(reg) && !is_descriptor_table(reg)) 1268 return (EINVAL); 1269 1270 return (vmmops_getdesc(vcpu->cookie, reg, desc)); 1271 } 1272 1273 int 1274 vm_set_seg_desc(struct vcpu *vcpu, int reg, struct seg_desc *desc) 1275 { 1276 1277 if (!is_segment_register(reg) && !is_descriptor_table(reg)) 1278 return (EINVAL); 1279 1280 return (vmmops_setdesc(vcpu->cookie, reg, desc)); 1281 } 1282 1283 static void 1284 restore_guest_fpustate(struct vcpu *vcpu) 1285 { 1286 1287 /* flush host state to the pcb */ 1288 fpuexit(curthread); 1289 1290 /* restore guest FPU state */ 1291 fpu_enable(); 1292 fpurestore(vcpu->guestfpu); 1293 1294 /* restore guest XCR0 if XSAVE is enabled in the host */ 1295 if (rcr4() & CR4_XSAVE) 1296 load_xcr(0, vcpu->guest_xcr0); 1297 1298 /* 1299 * The FPU is now "dirty" with the guest's state so disable 1300 * the FPU to trap any access by the host. 1301 */ 1302 fpu_disable(); 1303 } 1304 1305 static void 1306 save_guest_fpustate(struct vcpu *vcpu) 1307 { 1308 1309 if ((rcr0() & CR0_TS) == 0) 1310 panic("fpu emulation not enabled in host!"); 1311 1312 /* save guest XCR0 and restore host XCR0 */ 1313 if (rcr4() & CR4_XSAVE) { 1314 vcpu->guest_xcr0 = rxcr(0); 1315 load_xcr(0, vmm_get_host_xcr0()); 1316 } 1317 1318 /* save guest FPU state */ 1319 fpu_enable(); 1320 fpusave(vcpu->guestfpu); 1321 fpu_disable(); 1322 } 1323 1324 static VMM_STAT(VCPU_IDLE_TICKS, "number of ticks vcpu was idle"); 1325 1326 static int 1327 vcpu_set_state_locked(struct vcpu *vcpu, enum vcpu_state newstate, 1328 bool from_idle) 1329 { 1330 int error; 1331 1332 vcpu_assert_locked(vcpu); 1333 1334 /* 1335 * State transitions from the vmmdev_ioctl() must always begin from 1336 * the VCPU_IDLE state. This guarantees that there is only a single 1337 * ioctl() operating on a vcpu at any point. 1338 */ 1339 if (from_idle) { 1340 while (vcpu->state != VCPU_IDLE) { 1341 vcpu->reqidle = 1; 1342 vcpu_notify_event_locked(vcpu, false); 1343 VMM_CTR1(vcpu, "vcpu state change from %s to " 1344 "idle requested", vcpu_state2str(vcpu->state)); 1345 msleep_spin(&vcpu->state, &vcpu->mtx, "vmstat", hz); 1346 } 1347 } else { 1348 KASSERT(vcpu->state != VCPU_IDLE, ("invalid transition from " 1349 "vcpu idle state")); 1350 } 1351 1352 if (vcpu->state == VCPU_RUNNING) { 1353 KASSERT(vcpu->hostcpu == curcpu, ("curcpu %d and hostcpu %d " 1354 "mismatch for running vcpu", curcpu, vcpu->hostcpu)); 1355 } else { 1356 KASSERT(vcpu->hostcpu == NOCPU, ("Invalid hostcpu %d for a " 1357 "vcpu that is not running", vcpu->hostcpu)); 1358 } 1359 1360 /* 1361 * The following state transitions are allowed: 1362 * IDLE -> FROZEN -> IDLE 1363 * FROZEN -> RUNNING -> FROZEN 1364 * FROZEN -> SLEEPING -> FROZEN 1365 */ 1366 switch (vcpu->state) { 1367 case VCPU_IDLE: 1368 case VCPU_RUNNING: 1369 case VCPU_SLEEPING: 1370 error = (newstate != VCPU_FROZEN); 1371 break; 1372 case VCPU_FROZEN: 1373 error = (newstate == VCPU_FROZEN); 1374 break; 1375 default: 1376 error = 1; 1377 break; 1378 } 1379 1380 if (error) 1381 return (EBUSY); 1382 1383 VMM_CTR2(vcpu, "vcpu state changed from %s to %s", 1384 vcpu_state2str(vcpu->state), vcpu_state2str(newstate)); 1385 1386 vcpu->state = newstate; 1387 if (newstate == VCPU_RUNNING) 1388 vcpu->hostcpu = curcpu; 1389 else 1390 vcpu->hostcpu = NOCPU; 1391 1392 if (newstate == VCPU_IDLE) 1393 wakeup(&vcpu->state); 1394 1395 return (0); 1396 } 1397 1398 static void 1399 vcpu_require_state(struct vcpu *vcpu, enum vcpu_state newstate) 1400 { 1401 int error; 1402 1403 if ((error = vcpu_set_state(vcpu, newstate, false)) != 0) 1404 panic("Error %d setting state to %d\n", error, newstate); 1405 } 1406 1407 static void 1408 vcpu_require_state_locked(struct vcpu *vcpu, enum vcpu_state newstate) 1409 { 1410 int error; 1411 1412 if ((error = vcpu_set_state_locked(vcpu, newstate, false)) != 0) 1413 panic("Error %d setting state to %d", error, newstate); 1414 } 1415 1416 static int 1417 vm_handle_rendezvous(struct vcpu *vcpu) 1418 { 1419 struct vm *vm = vcpu->vm; 1420 struct thread *td; 1421 int error, vcpuid; 1422 1423 error = 0; 1424 vcpuid = vcpu->vcpuid; 1425 td = curthread; 1426 mtx_lock(&vm->rendezvous_mtx); 1427 while (vm->rendezvous_func != NULL) { 1428 /* 'rendezvous_req_cpus' must be a subset of 'active_cpus' */ 1429 CPU_AND(&vm->rendezvous_req_cpus, &vm->rendezvous_req_cpus, &vm->active_cpus); 1430 1431 if (CPU_ISSET(vcpuid, &vm->rendezvous_req_cpus) && 1432 !CPU_ISSET(vcpuid, &vm->rendezvous_done_cpus)) { 1433 VMM_CTR0(vcpu, "Calling rendezvous func"); 1434 (*vm->rendezvous_func)(vcpu, vm->rendezvous_arg); 1435 CPU_SET(vcpuid, &vm->rendezvous_done_cpus); 1436 } 1437 if (CPU_CMP(&vm->rendezvous_req_cpus, 1438 &vm->rendezvous_done_cpus) == 0) { 1439 VMM_CTR0(vcpu, "Rendezvous completed"); 1440 CPU_ZERO(&vm->rendezvous_req_cpus); 1441 vm->rendezvous_func = NULL; 1442 wakeup(&vm->rendezvous_func); 1443 break; 1444 } 1445 VMM_CTR0(vcpu, "Wait for rendezvous completion"); 1446 mtx_sleep(&vm->rendezvous_func, &vm->rendezvous_mtx, 0, 1447 "vmrndv", hz); 1448 if (td_ast_pending(td, TDA_SUSPEND)) { 1449 mtx_unlock(&vm->rendezvous_mtx); 1450 error = thread_check_susp(td, true); 1451 if (error != 0) 1452 return (error); 1453 mtx_lock(&vm->rendezvous_mtx); 1454 } 1455 } 1456 mtx_unlock(&vm->rendezvous_mtx); 1457 return (0); 1458 } 1459 1460 /* 1461 * Emulate a guest 'hlt' by sleeping until the vcpu is ready to run. 1462 */ 1463 static int 1464 vm_handle_hlt(struct vcpu *vcpu, bool intr_disabled, bool *retu) 1465 { 1466 struct vm *vm = vcpu->vm; 1467 const char *wmesg; 1468 struct thread *td; 1469 int error, t, vcpuid, vcpu_halted, vm_halted; 1470 1471 vcpuid = vcpu->vcpuid; 1472 vcpu_halted = 0; 1473 vm_halted = 0; 1474 error = 0; 1475 td = curthread; 1476 1477 KASSERT(!CPU_ISSET(vcpuid, &vm->halted_cpus), ("vcpu already halted")); 1478 1479 vcpu_lock(vcpu); 1480 while (1) { 1481 /* 1482 * Do a final check for pending NMI or interrupts before 1483 * really putting this thread to sleep. Also check for 1484 * software events that would cause this vcpu to wakeup. 1485 * 1486 * These interrupts/events could have happened after the 1487 * vcpu returned from vmmops_run() and before it acquired the 1488 * vcpu lock above. 1489 */ 1490 if (vm->rendezvous_func != NULL || vm->suspend || vcpu->reqidle) 1491 break; 1492 if (vm_nmi_pending(vcpu)) 1493 break; 1494 if (!intr_disabled) { 1495 if (vm_extint_pending(vcpu) || 1496 vlapic_pending_intr(vcpu->vlapic, NULL)) { 1497 break; 1498 } 1499 } 1500 1501 /* Don't go to sleep if the vcpu thread needs to yield */ 1502 if (vcpu_should_yield(vcpu)) 1503 break; 1504 1505 if (vcpu_debugged(vcpu)) 1506 break; 1507 1508 /* 1509 * Some Linux guests implement "halt" by having all vcpus 1510 * execute HLT with interrupts disabled. 'halted_cpus' keeps 1511 * track of the vcpus that have entered this state. When all 1512 * vcpus enter the halted state the virtual machine is halted. 1513 */ 1514 if (intr_disabled) { 1515 wmesg = "vmhalt"; 1516 VMM_CTR0(vcpu, "Halted"); 1517 if (!vcpu_halted && halt_detection_enabled) { 1518 vcpu_halted = 1; 1519 CPU_SET_ATOMIC(vcpuid, &vm->halted_cpus); 1520 } 1521 if (CPU_CMP(&vm->halted_cpus, &vm->active_cpus) == 0) { 1522 vm_halted = 1; 1523 break; 1524 } 1525 } else { 1526 wmesg = "vmidle"; 1527 } 1528 1529 t = ticks; 1530 vcpu_require_state_locked(vcpu, VCPU_SLEEPING); 1531 /* 1532 * XXX msleep_spin() cannot be interrupted by signals so 1533 * wake up periodically to check pending signals. 1534 */ 1535 msleep_spin(vcpu, &vcpu->mtx, wmesg, hz); 1536 vcpu_require_state_locked(vcpu, VCPU_FROZEN); 1537 vmm_stat_incr(vcpu, VCPU_IDLE_TICKS, ticks - t); 1538 if (td_ast_pending(td, TDA_SUSPEND)) { 1539 vcpu_unlock(vcpu); 1540 error = thread_check_susp(td, false); 1541 if (error != 0) { 1542 if (vcpu_halted) { 1543 CPU_CLR_ATOMIC(vcpuid, 1544 &vm->halted_cpus); 1545 } 1546 return (error); 1547 } 1548 vcpu_lock(vcpu); 1549 } 1550 } 1551 1552 if (vcpu_halted) 1553 CPU_CLR_ATOMIC(vcpuid, &vm->halted_cpus); 1554 1555 vcpu_unlock(vcpu); 1556 1557 if (vm_halted) 1558 vm_suspend(vm, VM_SUSPEND_HALT); 1559 1560 return (0); 1561 } 1562 1563 static int 1564 vm_handle_paging(struct vcpu *vcpu, bool *retu) 1565 { 1566 struct vm *vm = vcpu->vm; 1567 int rv, ftype; 1568 struct vm_map *map; 1569 struct vm_exit *vme; 1570 1571 vme = &vcpu->exitinfo; 1572 1573 KASSERT(vme->inst_length == 0, ("%s: invalid inst_length %d", 1574 __func__, vme->inst_length)); 1575 1576 ftype = vme->u.paging.fault_type; 1577 KASSERT(ftype == VM_PROT_READ || 1578 ftype == VM_PROT_WRITE || ftype == VM_PROT_EXECUTE, 1579 ("vm_handle_paging: invalid fault_type %d", ftype)); 1580 1581 if (ftype == VM_PROT_READ || ftype == VM_PROT_WRITE) { 1582 rv = pmap_emulate_accessed_dirty(vmspace_pmap(vm->vmspace), 1583 vme->u.paging.gpa, ftype); 1584 if (rv == 0) { 1585 VMM_CTR2(vcpu, "%s bit emulation for gpa %#lx", 1586 ftype == VM_PROT_READ ? "accessed" : "dirty", 1587 vme->u.paging.gpa); 1588 goto done; 1589 } 1590 } 1591 1592 map = &vm->vmspace->vm_map; 1593 rv = vm_fault(map, vme->u.paging.gpa, ftype, VM_FAULT_NORMAL, NULL); 1594 1595 VMM_CTR3(vcpu, "vm_handle_paging rv = %d, gpa = %#lx, " 1596 "ftype = %d", rv, vme->u.paging.gpa, ftype); 1597 1598 if (rv != KERN_SUCCESS) 1599 return (EFAULT); 1600 done: 1601 return (0); 1602 } 1603 1604 static int 1605 vm_handle_inst_emul(struct vcpu *vcpu, bool *retu) 1606 { 1607 struct vie *vie; 1608 struct vm_exit *vme; 1609 uint64_t gla, gpa, cs_base; 1610 struct vm_guest_paging *paging; 1611 mem_region_read_t mread; 1612 mem_region_write_t mwrite; 1613 enum vm_cpu_mode cpu_mode; 1614 int cs_d, error, fault; 1615 1616 vme = &vcpu->exitinfo; 1617 1618 KASSERT(vme->inst_length == 0, ("%s: invalid inst_length %d", 1619 __func__, vme->inst_length)); 1620 1621 gla = vme->u.inst_emul.gla; 1622 gpa = vme->u.inst_emul.gpa; 1623 cs_base = vme->u.inst_emul.cs_base; 1624 cs_d = vme->u.inst_emul.cs_d; 1625 vie = &vme->u.inst_emul.vie; 1626 paging = &vme->u.inst_emul.paging; 1627 cpu_mode = paging->cpu_mode; 1628 1629 VMM_CTR1(vcpu, "inst_emul fault accessing gpa %#lx", gpa); 1630 1631 /* Fetch, decode and emulate the faulting instruction */ 1632 if (vie->num_valid == 0) { 1633 error = vmm_fetch_instruction(vcpu, paging, vme->rip + cs_base, 1634 VIE_INST_SIZE, vie, &fault); 1635 } else { 1636 /* 1637 * The instruction bytes have already been copied into 'vie' 1638 */ 1639 error = fault = 0; 1640 } 1641 if (error || fault) 1642 return (error); 1643 1644 if (vmm_decode_instruction(vcpu, gla, cpu_mode, cs_d, vie) != 0) { 1645 VMM_CTR1(vcpu, "Error decoding instruction at %#lx", 1646 vme->rip + cs_base); 1647 *retu = true; /* dump instruction bytes in userspace */ 1648 return (0); 1649 } 1650 1651 /* 1652 * Update 'nextrip' based on the length of the emulated instruction. 1653 */ 1654 vme->inst_length = vie->num_processed; 1655 vcpu->nextrip += vie->num_processed; 1656 VMM_CTR1(vcpu, "nextrip updated to %#lx after instruction decoding", 1657 vcpu->nextrip); 1658 1659 /* return to userland unless this is an in-kernel emulated device */ 1660 if (gpa >= DEFAULT_APIC_BASE && gpa < DEFAULT_APIC_BASE + PAGE_SIZE) { 1661 mread = lapic_mmio_read; 1662 mwrite = lapic_mmio_write; 1663 } else if (gpa >= VIOAPIC_BASE && gpa < VIOAPIC_BASE + VIOAPIC_SIZE) { 1664 mread = vioapic_mmio_read; 1665 mwrite = vioapic_mmio_write; 1666 } else if (gpa >= VHPET_BASE && gpa < VHPET_BASE + VHPET_SIZE) { 1667 mread = vhpet_mmio_read; 1668 mwrite = vhpet_mmio_write; 1669 } else { 1670 *retu = true; 1671 return (0); 1672 } 1673 1674 error = vmm_emulate_instruction(vcpu, gpa, vie, paging, mread, mwrite, 1675 retu); 1676 1677 return (error); 1678 } 1679 1680 static int 1681 vm_handle_suspend(struct vcpu *vcpu, bool *retu) 1682 { 1683 struct vm *vm = vcpu->vm; 1684 int error, i; 1685 struct thread *td; 1686 1687 error = 0; 1688 td = curthread; 1689 1690 CPU_SET_ATOMIC(vcpu->vcpuid, &vm->suspended_cpus); 1691 1692 /* 1693 * Wait until all 'active_cpus' have suspended themselves. 1694 * 1695 * Since a VM may be suspended at any time including when one or 1696 * more vcpus are doing a rendezvous we need to call the rendezvous 1697 * handler while we are waiting to prevent a deadlock. 1698 */ 1699 vcpu_lock(vcpu); 1700 while (error == 0) { 1701 if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) { 1702 VMM_CTR0(vcpu, "All vcpus suspended"); 1703 break; 1704 } 1705 1706 if (vm->rendezvous_func == NULL) { 1707 VMM_CTR0(vcpu, "Sleeping during suspend"); 1708 vcpu_require_state_locked(vcpu, VCPU_SLEEPING); 1709 msleep_spin(vcpu, &vcpu->mtx, "vmsusp", hz); 1710 vcpu_require_state_locked(vcpu, VCPU_FROZEN); 1711 if (td_ast_pending(td, TDA_SUSPEND)) { 1712 vcpu_unlock(vcpu); 1713 error = thread_check_susp(td, false); 1714 vcpu_lock(vcpu); 1715 } 1716 } else { 1717 VMM_CTR0(vcpu, "Rendezvous during suspend"); 1718 vcpu_unlock(vcpu); 1719 error = vm_handle_rendezvous(vcpu); 1720 vcpu_lock(vcpu); 1721 } 1722 } 1723 vcpu_unlock(vcpu); 1724 1725 /* 1726 * Wakeup the other sleeping vcpus and return to userspace. 1727 */ 1728 for (i = 0; i < vm->maxcpus; i++) { 1729 if (CPU_ISSET(i, &vm->suspended_cpus)) { 1730 vcpu_notify_event(vm_vcpu(vm, i), false); 1731 } 1732 } 1733 1734 *retu = true; 1735 return (error); 1736 } 1737 1738 static int 1739 vm_handle_reqidle(struct vcpu *vcpu, bool *retu) 1740 { 1741 vcpu_lock(vcpu); 1742 KASSERT(vcpu->reqidle, ("invalid vcpu reqidle %d", vcpu->reqidle)); 1743 vcpu->reqidle = 0; 1744 vcpu_unlock(vcpu); 1745 *retu = true; 1746 return (0); 1747 } 1748 1749 static int 1750 vm_handle_db(struct vcpu *vcpu, struct vm_exit *vme, bool *retu) 1751 { 1752 int error, fault; 1753 uint64_t rsp; 1754 uint64_t rflags; 1755 struct vm_copyinfo copyinfo; 1756 1757 *retu = true; 1758 if (!vme->u.dbg.pushf_intercept || vme->u.dbg.tf_shadow_val != 0) { 1759 return (0); 1760 } 1761 1762 vm_get_register(vcpu, VM_REG_GUEST_RSP, &rsp); 1763 error = vm_copy_setup(vcpu, &vme->u.dbg.paging, rsp, sizeof(uint64_t), 1764 VM_PROT_RW, ©info, 1, &fault); 1765 if (error != 0 || fault != 0) { 1766 *retu = false; 1767 return (EINVAL); 1768 } 1769 1770 /* Read pushed rflags value from top of stack. */ 1771 vm_copyin(©info, &rflags, sizeof(uint64_t)); 1772 1773 /* Clear TF bit. */ 1774 rflags &= ~(PSL_T); 1775 1776 /* Write updated value back to memory. */ 1777 vm_copyout(&rflags, ©info, sizeof(uint64_t)); 1778 vm_copy_teardown(©info, 1); 1779 1780 return (0); 1781 } 1782 1783 int 1784 vm_suspend(struct vm *vm, enum vm_suspend_how how) 1785 { 1786 int i; 1787 1788 if (how <= VM_SUSPEND_NONE || how >= VM_SUSPEND_LAST) 1789 return (EINVAL); 1790 1791 if (atomic_cmpset_int(&vm->suspend, 0, how) == 0) { 1792 VM_CTR2(vm, "virtual machine already suspended %d/%d", 1793 vm->suspend, how); 1794 return (EALREADY); 1795 } 1796 1797 VM_CTR1(vm, "virtual machine successfully suspended %d", how); 1798 1799 /* 1800 * Notify all active vcpus that they are now suspended. 1801 */ 1802 for (i = 0; i < vm->maxcpus; i++) { 1803 if (CPU_ISSET(i, &vm->active_cpus)) 1804 vcpu_notify_event(vm_vcpu(vm, i), false); 1805 } 1806 1807 return (0); 1808 } 1809 1810 void 1811 vm_exit_suspended(struct vcpu *vcpu, uint64_t rip) 1812 { 1813 struct vm *vm = vcpu->vm; 1814 struct vm_exit *vmexit; 1815 1816 KASSERT(vm->suspend > VM_SUSPEND_NONE && vm->suspend < VM_SUSPEND_LAST, 1817 ("vm_exit_suspended: invalid suspend type %d", vm->suspend)); 1818 1819 vmexit = vm_exitinfo(vcpu); 1820 vmexit->rip = rip; 1821 vmexit->inst_length = 0; 1822 vmexit->exitcode = VM_EXITCODE_SUSPENDED; 1823 vmexit->u.suspended.how = vm->suspend; 1824 } 1825 1826 void 1827 vm_exit_debug(struct vcpu *vcpu, uint64_t rip) 1828 { 1829 struct vm_exit *vmexit; 1830 1831 vmexit = vm_exitinfo(vcpu); 1832 vmexit->rip = rip; 1833 vmexit->inst_length = 0; 1834 vmexit->exitcode = VM_EXITCODE_DEBUG; 1835 } 1836 1837 void 1838 vm_exit_rendezvous(struct vcpu *vcpu, uint64_t rip) 1839 { 1840 struct vm_exit *vmexit; 1841 1842 vmexit = vm_exitinfo(vcpu); 1843 vmexit->rip = rip; 1844 vmexit->inst_length = 0; 1845 vmexit->exitcode = VM_EXITCODE_RENDEZVOUS; 1846 vmm_stat_incr(vcpu, VMEXIT_RENDEZVOUS, 1); 1847 } 1848 1849 void 1850 vm_exit_reqidle(struct vcpu *vcpu, uint64_t rip) 1851 { 1852 struct vm_exit *vmexit; 1853 1854 vmexit = vm_exitinfo(vcpu); 1855 vmexit->rip = rip; 1856 vmexit->inst_length = 0; 1857 vmexit->exitcode = VM_EXITCODE_REQIDLE; 1858 vmm_stat_incr(vcpu, VMEXIT_REQIDLE, 1); 1859 } 1860 1861 void 1862 vm_exit_astpending(struct vcpu *vcpu, uint64_t rip) 1863 { 1864 struct vm_exit *vmexit; 1865 1866 vmexit = vm_exitinfo(vcpu); 1867 vmexit->rip = rip; 1868 vmexit->inst_length = 0; 1869 vmexit->exitcode = VM_EXITCODE_BOGUS; 1870 vmm_stat_incr(vcpu, VMEXIT_ASTPENDING, 1); 1871 } 1872 1873 int 1874 vm_run(struct vcpu *vcpu) 1875 { 1876 struct vm *vm = vcpu->vm; 1877 struct vm_eventinfo evinfo; 1878 int error, vcpuid; 1879 struct pcb *pcb; 1880 uint64_t tscval; 1881 struct vm_exit *vme; 1882 bool retu, intr_disabled; 1883 pmap_t pmap; 1884 1885 vcpuid = vcpu->vcpuid; 1886 1887 if (!CPU_ISSET(vcpuid, &vm->active_cpus)) 1888 return (EINVAL); 1889 1890 if (CPU_ISSET(vcpuid, &vm->suspended_cpus)) 1891 return (EINVAL); 1892 1893 pmap = vmspace_pmap(vm->vmspace); 1894 vme = &vcpu->exitinfo; 1895 evinfo.rptr = &vm->rendezvous_req_cpus; 1896 evinfo.sptr = &vm->suspend; 1897 evinfo.iptr = &vcpu->reqidle; 1898 restart: 1899 critical_enter(); 1900 1901 KASSERT(!CPU_ISSET(curcpu, &pmap->pm_active), 1902 ("vm_run: absurd pm_active")); 1903 1904 tscval = rdtsc(); 1905 1906 pcb = PCPU_GET(curpcb); 1907 set_pcb_flags(pcb, PCB_FULL_IRET); 1908 1909 restore_guest_fpustate(vcpu); 1910 1911 vcpu_require_state(vcpu, VCPU_RUNNING); 1912 error = vmmops_run(vcpu->cookie, vcpu->nextrip, pmap, &evinfo); 1913 vcpu_require_state(vcpu, VCPU_FROZEN); 1914 1915 save_guest_fpustate(vcpu); 1916 1917 vmm_stat_incr(vcpu, VCPU_TOTAL_RUNTIME, rdtsc() - tscval); 1918 1919 critical_exit(); 1920 1921 if (error == 0) { 1922 retu = false; 1923 vcpu->nextrip = vme->rip + vme->inst_length; 1924 switch (vme->exitcode) { 1925 case VM_EXITCODE_REQIDLE: 1926 error = vm_handle_reqidle(vcpu, &retu); 1927 break; 1928 case VM_EXITCODE_SUSPENDED: 1929 error = vm_handle_suspend(vcpu, &retu); 1930 break; 1931 case VM_EXITCODE_IOAPIC_EOI: 1932 vioapic_process_eoi(vm, vme->u.ioapic_eoi.vector); 1933 break; 1934 case VM_EXITCODE_RENDEZVOUS: 1935 error = vm_handle_rendezvous(vcpu); 1936 break; 1937 case VM_EXITCODE_HLT: 1938 intr_disabled = ((vme->u.hlt.rflags & PSL_I) == 0); 1939 error = vm_handle_hlt(vcpu, intr_disabled, &retu); 1940 break; 1941 case VM_EXITCODE_PAGING: 1942 error = vm_handle_paging(vcpu, &retu); 1943 break; 1944 case VM_EXITCODE_INST_EMUL: 1945 error = vm_handle_inst_emul(vcpu, &retu); 1946 break; 1947 case VM_EXITCODE_INOUT: 1948 case VM_EXITCODE_INOUT_STR: 1949 error = vm_handle_inout(vcpu, vme, &retu); 1950 break; 1951 case VM_EXITCODE_DB: 1952 error = vm_handle_db(vcpu, vme, &retu); 1953 break; 1954 case VM_EXITCODE_MONITOR: 1955 case VM_EXITCODE_MWAIT: 1956 case VM_EXITCODE_VMINSN: 1957 vm_inject_ud(vcpu); 1958 break; 1959 default: 1960 retu = true; /* handled in userland */ 1961 break; 1962 } 1963 } 1964 1965 /* 1966 * VM_EXITCODE_INST_EMUL could access the apic which could transform the 1967 * exit code into VM_EXITCODE_IPI. 1968 */ 1969 if (error == 0 && vme->exitcode == VM_EXITCODE_IPI) 1970 error = vm_handle_ipi(vcpu, vme, &retu); 1971 1972 if (error == 0 && retu == false) 1973 goto restart; 1974 1975 vmm_stat_incr(vcpu, VMEXIT_USERSPACE, 1); 1976 VMM_CTR2(vcpu, "retu %d/%d", error, vme->exitcode); 1977 1978 return (error); 1979 } 1980 1981 int 1982 vm_restart_instruction(struct vcpu *vcpu) 1983 { 1984 enum vcpu_state state; 1985 uint64_t rip; 1986 int error __diagused; 1987 1988 state = vcpu_get_state(vcpu, NULL); 1989 if (state == VCPU_RUNNING) { 1990 /* 1991 * When a vcpu is "running" the next instruction is determined 1992 * by adding 'rip' and 'inst_length' in the vcpu's 'exitinfo'. 1993 * Thus setting 'inst_length' to zero will cause the current 1994 * instruction to be restarted. 1995 */ 1996 vcpu->exitinfo.inst_length = 0; 1997 VMM_CTR1(vcpu, "restarting instruction at %#lx by " 1998 "setting inst_length to zero", vcpu->exitinfo.rip); 1999 } else if (state == VCPU_FROZEN) { 2000 /* 2001 * When a vcpu is "frozen" it is outside the critical section 2002 * around vmmops_run() and 'nextrip' points to the next 2003 * instruction. Thus instruction restart is achieved by setting 2004 * 'nextrip' to the vcpu's %rip. 2005 */ 2006 error = vm_get_register(vcpu, VM_REG_GUEST_RIP, &rip); 2007 KASSERT(!error, ("%s: error %d getting rip", __func__, error)); 2008 VMM_CTR2(vcpu, "restarting instruction by updating " 2009 "nextrip from %#lx to %#lx", vcpu->nextrip, rip); 2010 vcpu->nextrip = rip; 2011 } else { 2012 panic("%s: invalid state %d", __func__, state); 2013 } 2014 return (0); 2015 } 2016 2017 int 2018 vm_exit_intinfo(struct vcpu *vcpu, uint64_t info) 2019 { 2020 int type, vector; 2021 2022 if (info & VM_INTINFO_VALID) { 2023 type = info & VM_INTINFO_TYPE; 2024 vector = info & 0xff; 2025 if (type == VM_INTINFO_NMI && vector != IDT_NMI) 2026 return (EINVAL); 2027 if (type == VM_INTINFO_HWEXCEPTION && vector >= 32) 2028 return (EINVAL); 2029 if (info & VM_INTINFO_RSVD) 2030 return (EINVAL); 2031 } else { 2032 info = 0; 2033 } 2034 VMM_CTR2(vcpu, "%s: info1(%#lx)", __func__, info); 2035 vcpu->exitintinfo = info; 2036 return (0); 2037 } 2038 2039 enum exc_class { 2040 EXC_BENIGN, 2041 EXC_CONTRIBUTORY, 2042 EXC_PAGEFAULT 2043 }; 2044 2045 #define IDT_VE 20 /* Virtualization Exception (Intel specific) */ 2046 2047 static enum exc_class 2048 exception_class(uint64_t info) 2049 { 2050 int type, vector; 2051 2052 KASSERT(info & VM_INTINFO_VALID, ("intinfo must be valid: %#lx", info)); 2053 type = info & VM_INTINFO_TYPE; 2054 vector = info & 0xff; 2055 2056 /* Table 6-4, "Interrupt and Exception Classes", Intel SDM, Vol 3 */ 2057 switch (type) { 2058 case VM_INTINFO_HWINTR: 2059 case VM_INTINFO_SWINTR: 2060 case VM_INTINFO_NMI: 2061 return (EXC_BENIGN); 2062 default: 2063 /* 2064 * Hardware exception. 2065 * 2066 * SVM and VT-x use identical type values to represent NMI, 2067 * hardware interrupt and software interrupt. 2068 * 2069 * SVM uses type '3' for all exceptions. VT-x uses type '3' 2070 * for exceptions except #BP and #OF. #BP and #OF use a type 2071 * value of '5' or '6'. Therefore we don't check for explicit 2072 * values of 'type' to classify 'intinfo' into a hardware 2073 * exception. 2074 */ 2075 break; 2076 } 2077 2078 switch (vector) { 2079 case IDT_PF: 2080 case IDT_VE: 2081 return (EXC_PAGEFAULT); 2082 case IDT_DE: 2083 case IDT_TS: 2084 case IDT_NP: 2085 case IDT_SS: 2086 case IDT_GP: 2087 return (EXC_CONTRIBUTORY); 2088 default: 2089 return (EXC_BENIGN); 2090 } 2091 } 2092 2093 static int 2094 nested_fault(struct vcpu *vcpu, uint64_t info1, uint64_t info2, 2095 uint64_t *retinfo) 2096 { 2097 enum exc_class exc1, exc2; 2098 int type1, vector1; 2099 2100 KASSERT(info1 & VM_INTINFO_VALID, ("info1 %#lx is not valid", info1)); 2101 KASSERT(info2 & VM_INTINFO_VALID, ("info2 %#lx is not valid", info2)); 2102 2103 /* 2104 * If an exception occurs while attempting to call the double-fault 2105 * handler the processor enters shutdown mode (aka triple fault). 2106 */ 2107 type1 = info1 & VM_INTINFO_TYPE; 2108 vector1 = info1 & 0xff; 2109 if (type1 == VM_INTINFO_HWEXCEPTION && vector1 == IDT_DF) { 2110 VMM_CTR2(vcpu, "triple fault: info1(%#lx), info2(%#lx)", 2111 info1, info2); 2112 vm_suspend(vcpu->vm, VM_SUSPEND_TRIPLEFAULT); 2113 *retinfo = 0; 2114 return (0); 2115 } 2116 2117 /* 2118 * Table 6-5 "Conditions for Generating a Double Fault", Intel SDM, Vol3 2119 */ 2120 exc1 = exception_class(info1); 2121 exc2 = exception_class(info2); 2122 if ((exc1 == EXC_CONTRIBUTORY && exc2 == EXC_CONTRIBUTORY) || 2123 (exc1 == EXC_PAGEFAULT && exc2 != EXC_BENIGN)) { 2124 /* Convert nested fault into a double fault. */ 2125 *retinfo = IDT_DF; 2126 *retinfo |= VM_INTINFO_VALID | VM_INTINFO_HWEXCEPTION; 2127 *retinfo |= VM_INTINFO_DEL_ERRCODE; 2128 } else { 2129 /* Handle exceptions serially */ 2130 *retinfo = info2; 2131 } 2132 return (1); 2133 } 2134 2135 static uint64_t 2136 vcpu_exception_intinfo(struct vcpu *vcpu) 2137 { 2138 uint64_t info = 0; 2139 2140 if (vcpu->exception_pending) { 2141 info = vcpu->exc_vector & 0xff; 2142 info |= VM_INTINFO_VALID | VM_INTINFO_HWEXCEPTION; 2143 if (vcpu->exc_errcode_valid) { 2144 info |= VM_INTINFO_DEL_ERRCODE; 2145 info |= (uint64_t)vcpu->exc_errcode << 32; 2146 } 2147 } 2148 return (info); 2149 } 2150 2151 int 2152 vm_entry_intinfo(struct vcpu *vcpu, uint64_t *retinfo) 2153 { 2154 uint64_t info1, info2; 2155 int valid; 2156 2157 info1 = vcpu->exitintinfo; 2158 vcpu->exitintinfo = 0; 2159 2160 info2 = 0; 2161 if (vcpu->exception_pending) { 2162 info2 = vcpu_exception_intinfo(vcpu); 2163 vcpu->exception_pending = 0; 2164 VMM_CTR2(vcpu, "Exception %d delivered: %#lx", 2165 vcpu->exc_vector, info2); 2166 } 2167 2168 if ((info1 & VM_INTINFO_VALID) && (info2 & VM_INTINFO_VALID)) { 2169 valid = nested_fault(vcpu, info1, info2, retinfo); 2170 } else if (info1 & VM_INTINFO_VALID) { 2171 *retinfo = info1; 2172 valid = 1; 2173 } else if (info2 & VM_INTINFO_VALID) { 2174 *retinfo = info2; 2175 valid = 1; 2176 } else { 2177 valid = 0; 2178 } 2179 2180 if (valid) { 2181 VMM_CTR4(vcpu, "%s: info1(%#lx), info2(%#lx), " 2182 "retinfo(%#lx)", __func__, info1, info2, *retinfo); 2183 } 2184 2185 return (valid); 2186 } 2187 2188 int 2189 vm_get_intinfo(struct vcpu *vcpu, uint64_t *info1, uint64_t *info2) 2190 { 2191 *info1 = vcpu->exitintinfo; 2192 *info2 = vcpu_exception_intinfo(vcpu); 2193 return (0); 2194 } 2195 2196 int 2197 vm_inject_exception(struct vcpu *vcpu, int vector, int errcode_valid, 2198 uint32_t errcode, int restart_instruction) 2199 { 2200 uint64_t regval; 2201 int error __diagused; 2202 2203 if (vector < 0 || vector >= 32) 2204 return (EINVAL); 2205 2206 /* 2207 * A double fault exception should never be injected directly into 2208 * the guest. It is a derived exception that results from specific 2209 * combinations of nested faults. 2210 */ 2211 if (vector == IDT_DF) 2212 return (EINVAL); 2213 2214 if (vcpu->exception_pending) { 2215 VMM_CTR2(vcpu, "Unable to inject exception %d due to " 2216 "pending exception %d", vector, vcpu->exc_vector); 2217 return (EBUSY); 2218 } 2219 2220 if (errcode_valid) { 2221 /* 2222 * Exceptions don't deliver an error code in real mode. 2223 */ 2224 error = vm_get_register(vcpu, VM_REG_GUEST_CR0, ®val); 2225 KASSERT(!error, ("%s: error %d getting CR0", __func__, error)); 2226 if (!(regval & CR0_PE)) 2227 errcode_valid = 0; 2228 } 2229 2230 /* 2231 * From section 26.6.1 "Interruptibility State" in Intel SDM: 2232 * 2233 * Event blocking by "STI" or "MOV SS" is cleared after guest executes 2234 * one instruction or incurs an exception. 2235 */ 2236 error = vm_set_register(vcpu, VM_REG_GUEST_INTR_SHADOW, 0); 2237 KASSERT(error == 0, ("%s: error %d clearing interrupt shadow", 2238 __func__, error)); 2239 2240 if (restart_instruction) 2241 vm_restart_instruction(vcpu); 2242 2243 vcpu->exception_pending = 1; 2244 vcpu->exc_vector = vector; 2245 vcpu->exc_errcode = errcode; 2246 vcpu->exc_errcode_valid = errcode_valid; 2247 VMM_CTR1(vcpu, "Exception %d pending", vector); 2248 return (0); 2249 } 2250 2251 void 2252 vm_inject_fault(struct vcpu *vcpu, int vector, int errcode_valid, int errcode) 2253 { 2254 int error __diagused, restart_instruction; 2255 2256 restart_instruction = 1; 2257 2258 error = vm_inject_exception(vcpu, vector, errcode_valid, 2259 errcode, restart_instruction); 2260 KASSERT(error == 0, ("vm_inject_exception error %d", error)); 2261 } 2262 2263 void 2264 vm_inject_pf(struct vcpu *vcpu, int error_code, uint64_t cr2) 2265 { 2266 int error __diagused; 2267 2268 VMM_CTR2(vcpu, "Injecting page fault: error_code %#x, cr2 %#lx", 2269 error_code, cr2); 2270 2271 error = vm_set_register(vcpu, VM_REG_GUEST_CR2, cr2); 2272 KASSERT(error == 0, ("vm_set_register(cr2) error %d", error)); 2273 2274 vm_inject_fault(vcpu, IDT_PF, 1, error_code); 2275 } 2276 2277 static VMM_STAT(VCPU_NMI_COUNT, "number of NMIs delivered to vcpu"); 2278 2279 int 2280 vm_inject_nmi(struct vcpu *vcpu) 2281 { 2282 2283 vcpu->nmi_pending = 1; 2284 vcpu_notify_event(vcpu, false); 2285 return (0); 2286 } 2287 2288 int 2289 vm_nmi_pending(struct vcpu *vcpu) 2290 { 2291 return (vcpu->nmi_pending); 2292 } 2293 2294 void 2295 vm_nmi_clear(struct vcpu *vcpu) 2296 { 2297 if (vcpu->nmi_pending == 0) 2298 panic("vm_nmi_clear: inconsistent nmi_pending state"); 2299 2300 vcpu->nmi_pending = 0; 2301 vmm_stat_incr(vcpu, VCPU_NMI_COUNT, 1); 2302 } 2303 2304 static VMM_STAT(VCPU_EXTINT_COUNT, "number of ExtINTs delivered to vcpu"); 2305 2306 int 2307 vm_inject_extint(struct vcpu *vcpu) 2308 { 2309 2310 vcpu->extint_pending = 1; 2311 vcpu_notify_event(vcpu, false); 2312 return (0); 2313 } 2314 2315 int 2316 vm_extint_pending(struct vcpu *vcpu) 2317 { 2318 return (vcpu->extint_pending); 2319 } 2320 2321 void 2322 vm_extint_clear(struct vcpu *vcpu) 2323 { 2324 if (vcpu->extint_pending == 0) 2325 panic("vm_extint_clear: inconsistent extint_pending state"); 2326 2327 vcpu->extint_pending = 0; 2328 vmm_stat_incr(vcpu, VCPU_EXTINT_COUNT, 1); 2329 } 2330 2331 int 2332 vm_get_capability(struct vcpu *vcpu, int type, int *retval) 2333 { 2334 if (type < 0 || type >= VM_CAP_MAX) 2335 return (EINVAL); 2336 2337 return (vmmops_getcap(vcpu->cookie, type, retval)); 2338 } 2339 2340 int 2341 vm_set_capability(struct vcpu *vcpu, int type, int val) 2342 { 2343 if (type < 0 || type >= VM_CAP_MAX) 2344 return (EINVAL); 2345 2346 return (vmmops_setcap(vcpu->cookie, type, val)); 2347 } 2348 2349 struct vm * 2350 vcpu_vm(struct vcpu *vcpu) 2351 { 2352 return (vcpu->vm); 2353 } 2354 2355 int 2356 vcpu_vcpuid(struct vcpu *vcpu) 2357 { 2358 return (vcpu->vcpuid); 2359 } 2360 2361 struct vcpu * 2362 vm_vcpu(struct vm *vm, int vcpuid) 2363 { 2364 return (vm->vcpu[vcpuid]); 2365 } 2366 2367 struct vlapic * 2368 vm_lapic(struct vcpu *vcpu) 2369 { 2370 return (vcpu->vlapic); 2371 } 2372 2373 struct vioapic * 2374 vm_ioapic(struct vm *vm) 2375 { 2376 2377 return (vm->vioapic); 2378 } 2379 2380 struct vhpet * 2381 vm_hpet(struct vm *vm) 2382 { 2383 2384 return (vm->vhpet); 2385 } 2386 2387 bool 2388 vmm_is_pptdev(int bus, int slot, int func) 2389 { 2390 int b, f, i, n, s; 2391 char *val, *cp, *cp2; 2392 bool found; 2393 2394 /* 2395 * XXX 2396 * The length of an environment variable is limited to 128 bytes which 2397 * puts an upper limit on the number of passthru devices that may be 2398 * specified using a single environment variable. 2399 * 2400 * Work around this by scanning multiple environment variable 2401 * names instead of a single one - yuck! 2402 */ 2403 const char *names[] = { "pptdevs", "pptdevs2", "pptdevs3", NULL }; 2404 2405 /* set pptdevs="1/2/3 4/5/6 7/8/9 10/11/12" */ 2406 found = false; 2407 for (i = 0; names[i] != NULL && !found; i++) { 2408 cp = val = kern_getenv(names[i]); 2409 while (cp != NULL && *cp != '\0') { 2410 if ((cp2 = strchr(cp, ' ')) != NULL) 2411 *cp2 = '\0'; 2412 2413 n = sscanf(cp, "%d/%d/%d", &b, &s, &f); 2414 if (n == 3 && bus == b && slot == s && func == f) { 2415 found = true; 2416 break; 2417 } 2418 2419 if (cp2 != NULL) 2420 *cp2++ = ' '; 2421 2422 cp = cp2; 2423 } 2424 freeenv(val); 2425 } 2426 return (found); 2427 } 2428 2429 void * 2430 vm_iommu_domain(struct vm *vm) 2431 { 2432 2433 return (vm->iommu); 2434 } 2435 2436 int 2437 vcpu_set_state(struct vcpu *vcpu, enum vcpu_state newstate, bool from_idle) 2438 { 2439 int error; 2440 2441 vcpu_lock(vcpu); 2442 error = vcpu_set_state_locked(vcpu, newstate, from_idle); 2443 vcpu_unlock(vcpu); 2444 2445 return (error); 2446 } 2447 2448 enum vcpu_state 2449 vcpu_get_state(struct vcpu *vcpu, int *hostcpu) 2450 { 2451 enum vcpu_state state; 2452 2453 vcpu_lock(vcpu); 2454 state = vcpu->state; 2455 if (hostcpu != NULL) 2456 *hostcpu = vcpu->hostcpu; 2457 vcpu_unlock(vcpu); 2458 2459 return (state); 2460 } 2461 2462 int 2463 vm_activate_cpu(struct vcpu *vcpu) 2464 { 2465 struct vm *vm = vcpu->vm; 2466 2467 if (CPU_ISSET(vcpu->vcpuid, &vm->active_cpus)) 2468 return (EBUSY); 2469 2470 VMM_CTR0(vcpu, "activated"); 2471 CPU_SET_ATOMIC(vcpu->vcpuid, &vm->active_cpus); 2472 return (0); 2473 } 2474 2475 int 2476 vm_suspend_cpu(struct vm *vm, struct vcpu *vcpu) 2477 { 2478 if (vcpu == NULL) { 2479 vm->debug_cpus = vm->active_cpus; 2480 for (int i = 0; i < vm->maxcpus; i++) { 2481 if (CPU_ISSET(i, &vm->active_cpus)) 2482 vcpu_notify_event(vm_vcpu(vm, i), false); 2483 } 2484 } else { 2485 if (!CPU_ISSET(vcpu->vcpuid, &vm->active_cpus)) 2486 return (EINVAL); 2487 2488 CPU_SET_ATOMIC(vcpu->vcpuid, &vm->debug_cpus); 2489 vcpu_notify_event(vcpu, false); 2490 } 2491 return (0); 2492 } 2493 2494 int 2495 vm_resume_cpu(struct vm *vm, struct vcpu *vcpu) 2496 { 2497 2498 if (vcpu == NULL) { 2499 CPU_ZERO(&vm->debug_cpus); 2500 } else { 2501 if (!CPU_ISSET(vcpu->vcpuid, &vm->debug_cpus)) 2502 return (EINVAL); 2503 2504 CPU_CLR_ATOMIC(vcpu->vcpuid, &vm->debug_cpus); 2505 } 2506 return (0); 2507 } 2508 2509 int 2510 vcpu_debugged(struct vcpu *vcpu) 2511 { 2512 2513 return (CPU_ISSET(vcpu->vcpuid, &vcpu->vm->debug_cpus)); 2514 } 2515 2516 cpuset_t 2517 vm_active_cpus(struct vm *vm) 2518 { 2519 2520 return (vm->active_cpus); 2521 } 2522 2523 cpuset_t 2524 vm_debug_cpus(struct vm *vm) 2525 { 2526 2527 return (vm->debug_cpus); 2528 } 2529 2530 cpuset_t 2531 vm_suspended_cpus(struct vm *vm) 2532 { 2533 2534 return (vm->suspended_cpus); 2535 } 2536 2537 /* 2538 * Returns the subset of vCPUs in tostart that are awaiting startup. 2539 * These vCPUs are also marked as no longer awaiting startup. 2540 */ 2541 cpuset_t 2542 vm_start_cpus(struct vm *vm, const cpuset_t *tostart) 2543 { 2544 cpuset_t set; 2545 2546 mtx_lock(&vm->rendezvous_mtx); 2547 CPU_AND(&set, &vm->startup_cpus, tostart); 2548 CPU_ANDNOT(&vm->startup_cpus, &vm->startup_cpus, &set); 2549 mtx_unlock(&vm->rendezvous_mtx); 2550 return (set); 2551 } 2552 2553 void 2554 vm_await_start(struct vm *vm, const cpuset_t *waiting) 2555 { 2556 mtx_lock(&vm->rendezvous_mtx); 2557 CPU_OR(&vm->startup_cpus, &vm->startup_cpus, waiting); 2558 mtx_unlock(&vm->rendezvous_mtx); 2559 } 2560 2561 void * 2562 vcpu_stats(struct vcpu *vcpu) 2563 { 2564 2565 return (vcpu->stats); 2566 } 2567 2568 int 2569 vm_get_x2apic_state(struct vcpu *vcpu, enum x2apic_state *state) 2570 { 2571 *state = vcpu->x2apic_state; 2572 2573 return (0); 2574 } 2575 2576 int 2577 vm_set_x2apic_state(struct vcpu *vcpu, enum x2apic_state state) 2578 { 2579 if (state >= X2APIC_STATE_LAST) 2580 return (EINVAL); 2581 2582 vcpu->x2apic_state = state; 2583 2584 vlapic_set_x2apic_state(vcpu, state); 2585 2586 return (0); 2587 } 2588 2589 /* 2590 * This function is called to ensure that a vcpu "sees" a pending event 2591 * as soon as possible: 2592 * - If the vcpu thread is sleeping then it is woken up. 2593 * - If the vcpu is running on a different host_cpu then an IPI will be directed 2594 * to the host_cpu to cause the vcpu to trap into the hypervisor. 2595 */ 2596 static void 2597 vcpu_notify_event_locked(struct vcpu *vcpu, bool lapic_intr) 2598 { 2599 int hostcpu; 2600 2601 hostcpu = vcpu->hostcpu; 2602 if (vcpu->state == VCPU_RUNNING) { 2603 KASSERT(hostcpu != NOCPU, ("vcpu running on invalid hostcpu")); 2604 if (hostcpu != curcpu) { 2605 if (lapic_intr) { 2606 vlapic_post_intr(vcpu->vlapic, hostcpu, 2607 vmm_ipinum); 2608 } else { 2609 ipi_cpu(hostcpu, vmm_ipinum); 2610 } 2611 } else { 2612 /* 2613 * If the 'vcpu' is running on 'curcpu' then it must 2614 * be sending a notification to itself (e.g. SELF_IPI). 2615 * The pending event will be picked up when the vcpu 2616 * transitions back to guest context. 2617 */ 2618 } 2619 } else { 2620 KASSERT(hostcpu == NOCPU, ("vcpu state %d not consistent " 2621 "with hostcpu %d", vcpu->state, hostcpu)); 2622 if (vcpu->state == VCPU_SLEEPING) 2623 wakeup_one(vcpu); 2624 } 2625 } 2626 2627 void 2628 vcpu_notify_event(struct vcpu *vcpu, bool lapic_intr) 2629 { 2630 vcpu_lock(vcpu); 2631 vcpu_notify_event_locked(vcpu, lapic_intr); 2632 vcpu_unlock(vcpu); 2633 } 2634 2635 struct vmspace * 2636 vm_get_vmspace(struct vm *vm) 2637 { 2638 2639 return (vm->vmspace); 2640 } 2641 2642 int 2643 vm_apicid2vcpuid(struct vm *vm, int apicid) 2644 { 2645 /* 2646 * XXX apic id is assumed to be numerically identical to vcpu id 2647 */ 2648 return (apicid); 2649 } 2650 2651 int 2652 vm_smp_rendezvous(struct vcpu *vcpu, cpuset_t dest, 2653 vm_rendezvous_func_t func, void *arg) 2654 { 2655 struct vm *vm = vcpu->vm; 2656 int error, i; 2657 2658 /* 2659 * Enforce that this function is called without any locks 2660 */ 2661 WITNESS_WARN(WARN_PANIC, NULL, "vm_smp_rendezvous"); 2662 2663 restart: 2664 mtx_lock(&vm->rendezvous_mtx); 2665 if (vm->rendezvous_func != NULL) { 2666 /* 2667 * If a rendezvous is already in progress then we need to 2668 * call the rendezvous handler in case this 'vcpu' is one 2669 * of the targets of the rendezvous. 2670 */ 2671 VMM_CTR0(vcpu, "Rendezvous already in progress"); 2672 mtx_unlock(&vm->rendezvous_mtx); 2673 error = vm_handle_rendezvous(vcpu); 2674 if (error != 0) 2675 return (error); 2676 goto restart; 2677 } 2678 KASSERT(vm->rendezvous_func == NULL, ("vm_smp_rendezvous: previous " 2679 "rendezvous is still in progress")); 2680 2681 VMM_CTR0(vcpu, "Initiating rendezvous"); 2682 vm->rendezvous_req_cpus = dest; 2683 CPU_ZERO(&vm->rendezvous_done_cpus); 2684 vm->rendezvous_arg = arg; 2685 vm->rendezvous_func = func; 2686 mtx_unlock(&vm->rendezvous_mtx); 2687 2688 /* 2689 * Wake up any sleeping vcpus and trigger a VM-exit in any running 2690 * vcpus so they handle the rendezvous as soon as possible. 2691 */ 2692 for (i = 0; i < vm->maxcpus; i++) { 2693 if (CPU_ISSET(i, &dest)) 2694 vcpu_notify_event(vm_vcpu(vm, i), false); 2695 } 2696 2697 return (vm_handle_rendezvous(vcpu)); 2698 } 2699 2700 struct vatpic * 2701 vm_atpic(struct vm *vm) 2702 { 2703 return (vm->vatpic); 2704 } 2705 2706 struct vatpit * 2707 vm_atpit(struct vm *vm) 2708 { 2709 return (vm->vatpit); 2710 } 2711 2712 struct vpmtmr * 2713 vm_pmtmr(struct vm *vm) 2714 { 2715 2716 return (vm->vpmtmr); 2717 } 2718 2719 struct vrtc * 2720 vm_rtc(struct vm *vm) 2721 { 2722 2723 return (vm->vrtc); 2724 } 2725 2726 enum vm_reg_name 2727 vm_segment_name(int seg) 2728 { 2729 static enum vm_reg_name seg_names[] = { 2730 VM_REG_GUEST_ES, 2731 VM_REG_GUEST_CS, 2732 VM_REG_GUEST_SS, 2733 VM_REG_GUEST_DS, 2734 VM_REG_GUEST_FS, 2735 VM_REG_GUEST_GS 2736 }; 2737 2738 KASSERT(seg >= 0 && seg < nitems(seg_names), 2739 ("%s: invalid segment encoding %d", __func__, seg)); 2740 return (seg_names[seg]); 2741 } 2742 2743 void 2744 vm_copy_teardown(struct vm_copyinfo *copyinfo, int num_copyinfo) 2745 { 2746 int idx; 2747 2748 for (idx = 0; idx < num_copyinfo; idx++) { 2749 if (copyinfo[idx].cookie != NULL) 2750 vm_gpa_release(copyinfo[idx].cookie); 2751 } 2752 bzero(copyinfo, num_copyinfo * sizeof(struct vm_copyinfo)); 2753 } 2754 2755 int 2756 vm_copy_setup(struct vcpu *vcpu, struct vm_guest_paging *paging, 2757 uint64_t gla, size_t len, int prot, struct vm_copyinfo *copyinfo, 2758 int num_copyinfo, int *fault) 2759 { 2760 int error, idx, nused; 2761 size_t n, off, remaining; 2762 void *hva, *cookie; 2763 uint64_t gpa; 2764 2765 bzero(copyinfo, sizeof(struct vm_copyinfo) * num_copyinfo); 2766 2767 nused = 0; 2768 remaining = len; 2769 while (remaining > 0) { 2770 KASSERT(nused < num_copyinfo, ("insufficient vm_copyinfo")); 2771 error = vm_gla2gpa(vcpu, paging, gla, prot, &gpa, fault); 2772 if (error || *fault) 2773 return (error); 2774 off = gpa & PAGE_MASK; 2775 n = min(remaining, PAGE_SIZE - off); 2776 copyinfo[nused].gpa = gpa; 2777 copyinfo[nused].len = n; 2778 remaining -= n; 2779 gla += n; 2780 nused++; 2781 } 2782 2783 for (idx = 0; idx < nused; idx++) { 2784 hva = vm_gpa_hold(vcpu, copyinfo[idx].gpa, 2785 copyinfo[idx].len, prot, &cookie); 2786 if (hva == NULL) 2787 break; 2788 copyinfo[idx].hva = hva; 2789 copyinfo[idx].cookie = cookie; 2790 } 2791 2792 if (idx != nused) { 2793 vm_copy_teardown(copyinfo, num_copyinfo); 2794 return (EFAULT); 2795 } else { 2796 *fault = 0; 2797 return (0); 2798 } 2799 } 2800 2801 void 2802 vm_copyin(struct vm_copyinfo *copyinfo, void *kaddr, size_t len) 2803 { 2804 char *dst; 2805 int idx; 2806 2807 dst = kaddr; 2808 idx = 0; 2809 while (len > 0) { 2810 bcopy(copyinfo[idx].hva, dst, copyinfo[idx].len); 2811 len -= copyinfo[idx].len; 2812 dst += copyinfo[idx].len; 2813 idx++; 2814 } 2815 } 2816 2817 void 2818 vm_copyout(const void *kaddr, struct vm_copyinfo *copyinfo, size_t len) 2819 { 2820 const char *src; 2821 int idx; 2822 2823 src = kaddr; 2824 idx = 0; 2825 while (len > 0) { 2826 bcopy(src, copyinfo[idx].hva, copyinfo[idx].len); 2827 len -= copyinfo[idx].len; 2828 src += copyinfo[idx].len; 2829 idx++; 2830 } 2831 } 2832 2833 /* 2834 * Return the amount of in-use and wired memory for the VM. Since 2835 * these are global stats, only return the values with for vCPU 0 2836 */ 2837 VMM_STAT_DECLARE(VMM_MEM_RESIDENT); 2838 VMM_STAT_DECLARE(VMM_MEM_WIRED); 2839 2840 static void 2841 vm_get_rescnt(struct vcpu *vcpu, struct vmm_stat_type *stat) 2842 { 2843 2844 if (vcpu->vcpuid == 0) { 2845 vmm_stat_set(vcpu, VMM_MEM_RESIDENT, PAGE_SIZE * 2846 vmspace_resident_count(vcpu->vm->vmspace)); 2847 } 2848 } 2849 2850 static void 2851 vm_get_wiredcnt(struct vcpu *vcpu, struct vmm_stat_type *stat) 2852 { 2853 2854 if (vcpu->vcpuid == 0) { 2855 vmm_stat_set(vcpu, VMM_MEM_WIRED, PAGE_SIZE * 2856 pmap_wired_count(vmspace_pmap(vcpu->vm->vmspace))); 2857 } 2858 } 2859 2860 VMM_STAT_FUNC(VMM_MEM_RESIDENT, "Resident memory", vm_get_rescnt); 2861 VMM_STAT_FUNC(VMM_MEM_WIRED, "Wired memory", vm_get_wiredcnt); 2862 2863 #ifdef BHYVE_SNAPSHOT 2864 static int 2865 vm_snapshot_vcpus(struct vm *vm, struct vm_snapshot_meta *meta) 2866 { 2867 uint64_t tsc, now; 2868 int ret; 2869 struct vcpu *vcpu; 2870 uint16_t i, maxcpus; 2871 2872 now = rdtsc(); 2873 maxcpus = vm_get_maxcpus(vm); 2874 for (i = 0; i < maxcpus; i++) { 2875 vcpu = vm->vcpu[i]; 2876 if (vcpu == NULL) 2877 continue; 2878 2879 SNAPSHOT_VAR_OR_LEAVE(vcpu->x2apic_state, meta, ret, done); 2880 SNAPSHOT_VAR_OR_LEAVE(vcpu->exitintinfo, meta, ret, done); 2881 SNAPSHOT_VAR_OR_LEAVE(vcpu->exc_vector, meta, ret, done); 2882 SNAPSHOT_VAR_OR_LEAVE(vcpu->exc_errcode_valid, meta, ret, done); 2883 SNAPSHOT_VAR_OR_LEAVE(vcpu->exc_errcode, meta, ret, done); 2884 SNAPSHOT_VAR_OR_LEAVE(vcpu->guest_xcr0, meta, ret, done); 2885 SNAPSHOT_VAR_OR_LEAVE(vcpu->exitinfo, meta, ret, done); 2886 SNAPSHOT_VAR_OR_LEAVE(vcpu->nextrip, meta, ret, done); 2887 2888 /* 2889 * Save the absolute TSC value by adding now to tsc_offset. 2890 * 2891 * It will be turned turned back into an actual offset when the 2892 * TSC restore function is called 2893 */ 2894 tsc = now + vcpu->tsc_offset; 2895 SNAPSHOT_VAR_OR_LEAVE(tsc, meta, ret, done); 2896 if (meta->op == VM_SNAPSHOT_RESTORE) 2897 vcpu->tsc_offset = tsc; 2898 } 2899 2900 done: 2901 return (ret); 2902 } 2903 2904 static int 2905 vm_snapshot_vm(struct vm *vm, struct vm_snapshot_meta *meta) 2906 { 2907 int ret; 2908 2909 ret = vm_snapshot_vcpus(vm, meta); 2910 if (ret != 0) 2911 goto done; 2912 2913 SNAPSHOT_VAR_OR_LEAVE(vm->startup_cpus, meta, ret, done); 2914 done: 2915 return (ret); 2916 } 2917 2918 static int 2919 vm_snapshot_vcpu(struct vm *vm, struct vm_snapshot_meta *meta) 2920 { 2921 int error; 2922 struct vcpu *vcpu; 2923 uint16_t i, maxcpus; 2924 2925 error = 0; 2926 2927 maxcpus = vm_get_maxcpus(vm); 2928 for (i = 0; i < maxcpus; i++) { 2929 vcpu = vm->vcpu[i]; 2930 if (vcpu == NULL) 2931 continue; 2932 2933 error = vmmops_vcpu_snapshot(vcpu->cookie, meta); 2934 if (error != 0) { 2935 printf("%s: failed to snapshot vmcs/vmcb data for " 2936 "vCPU: %d; error: %d\n", __func__, i, error); 2937 goto done; 2938 } 2939 } 2940 2941 done: 2942 return (error); 2943 } 2944 2945 /* 2946 * Save kernel-side structures to user-space for snapshotting. 2947 */ 2948 int 2949 vm_snapshot_req(struct vm *vm, struct vm_snapshot_meta *meta) 2950 { 2951 int ret = 0; 2952 2953 switch (meta->dev_req) { 2954 case STRUCT_VMCX: 2955 ret = vm_snapshot_vcpu(vm, meta); 2956 break; 2957 case STRUCT_VM: 2958 ret = vm_snapshot_vm(vm, meta); 2959 break; 2960 case STRUCT_VIOAPIC: 2961 ret = vioapic_snapshot(vm_ioapic(vm), meta); 2962 break; 2963 case STRUCT_VLAPIC: 2964 ret = vlapic_snapshot(vm, meta); 2965 break; 2966 case STRUCT_VHPET: 2967 ret = vhpet_snapshot(vm_hpet(vm), meta); 2968 break; 2969 case STRUCT_VATPIC: 2970 ret = vatpic_snapshot(vm_atpic(vm), meta); 2971 break; 2972 case STRUCT_VATPIT: 2973 ret = vatpit_snapshot(vm_atpit(vm), meta); 2974 break; 2975 case STRUCT_VPMTMR: 2976 ret = vpmtmr_snapshot(vm_pmtmr(vm), meta); 2977 break; 2978 case STRUCT_VRTC: 2979 ret = vrtc_snapshot(vm_rtc(vm), meta); 2980 break; 2981 default: 2982 printf("%s: failed to find the requested type %#x\n", 2983 __func__, meta->dev_req); 2984 ret = (EINVAL); 2985 } 2986 return (ret); 2987 } 2988 2989 void 2990 vm_set_tsc_offset(struct vcpu *vcpu, uint64_t offset) 2991 { 2992 vcpu->tsc_offset = offset; 2993 } 2994 2995 int 2996 vm_restore_time(struct vm *vm) 2997 { 2998 int error; 2999 uint64_t now; 3000 struct vcpu *vcpu; 3001 uint16_t i, maxcpus; 3002 3003 now = rdtsc(); 3004 3005 error = vhpet_restore_time(vm_hpet(vm)); 3006 if (error) 3007 return (error); 3008 3009 maxcpus = vm_get_maxcpus(vm); 3010 for (i = 0; i < maxcpus; i++) { 3011 vcpu = vm->vcpu[i]; 3012 if (vcpu == NULL) 3013 continue; 3014 3015 error = vmmops_restore_tsc(vcpu->cookie, 3016 vcpu->tsc_offset - now); 3017 if (error) 3018 return (error); 3019 } 3020 3021 return (0); 3022 } 3023 #endif 3024