1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 3 * 4 * Copyright (c) 2011 NetApp, Inc. 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 16 * THIS SOFTWARE IS PROVIDED BY NETAPP, INC ``AS IS'' AND 17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 19 * ARE DISCLAIMED. IN NO EVENT SHALL NETAPP, INC OR CONTRIBUTORS BE LIABLE 20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 26 * SUCH DAMAGE. 27 * 28 * $FreeBSD$ 29 */ 30 31 #include <sys/cdefs.h> 32 __FBSDID("$FreeBSD$"); 33 34 #include <sys/param.h> 35 #include <sys/systm.h> 36 #include <sys/kernel.h> 37 #include <sys/module.h> 38 #include <sys/sysctl.h> 39 #include <sys/malloc.h> 40 #include <sys/pcpu.h> 41 #include <sys/lock.h> 42 #include <sys/mutex.h> 43 #include <sys/proc.h> 44 #include <sys/rwlock.h> 45 #include <sys/sched.h> 46 #include <sys/smp.h> 47 #include <sys/systm.h> 48 49 #include <vm/vm.h> 50 #include <vm/vm_object.h> 51 #include <vm/vm_page.h> 52 #include <vm/pmap.h> 53 #include <vm/vm_map.h> 54 #include <vm/vm_extern.h> 55 #include <vm/vm_param.h> 56 57 #include <machine/cpu.h> 58 #include <machine/pcb.h> 59 #include <machine/smp.h> 60 #include <machine/md_var.h> 61 #include <x86/psl.h> 62 #include <x86/apicreg.h> 63 64 #include <machine/vmm.h> 65 #include <machine/vmm_dev.h> 66 #include <machine/vmm_instruction_emul.h> 67 68 #include "vmm_ioport.h" 69 #include "vmm_ktr.h" 70 #include "vmm_host.h" 71 #include "vmm_mem.h" 72 #include "vmm_util.h" 73 #include "vatpic.h" 74 #include "vatpit.h" 75 #include "vhpet.h" 76 #include "vioapic.h" 77 #include "vlapic.h" 78 #include "vpmtmr.h" 79 #include "vrtc.h" 80 #include "vmm_stat.h" 81 #include "vmm_lapic.h" 82 83 #include "io/ppt.h" 84 #include "io/iommu.h" 85 86 struct vlapic; 87 88 /* 89 * Initialization: 90 * (a) allocated when vcpu is created 91 * (i) initialized when vcpu is created and when it is reinitialized 92 * (o) initialized the first time the vcpu is created 93 * (x) initialized before use 94 */ 95 struct vcpu { 96 struct mtx mtx; /* (o) protects 'state' and 'hostcpu' */ 97 enum vcpu_state state; /* (o) vcpu state */ 98 int hostcpu; /* (o) vcpu's host cpu */ 99 int reqidle; /* (i) request vcpu to idle */ 100 struct vlapic *vlapic; /* (i) APIC device model */ 101 enum x2apic_state x2apic_state; /* (i) APIC mode */ 102 uint64_t exitintinfo; /* (i) events pending at VM exit */ 103 int nmi_pending; /* (i) NMI pending */ 104 int extint_pending; /* (i) INTR pending */ 105 int exception_pending; /* (i) exception pending */ 106 int exc_vector; /* (x) exception collateral */ 107 int exc_errcode_valid; 108 uint32_t exc_errcode; 109 struct savefpu *guestfpu; /* (a,i) guest fpu state */ 110 uint64_t guest_xcr0; /* (i) guest %xcr0 register */ 111 void *stats; /* (a,i) statistics */ 112 struct vm_exit exitinfo; /* (x) exit reason and collateral */ 113 uint64_t nextrip; /* (x) next instruction to execute */ 114 }; 115 116 #define vcpu_lock_initialized(v) mtx_initialized(&((v)->mtx)) 117 #define vcpu_lock_init(v) mtx_init(&((v)->mtx), "vcpu lock", 0, MTX_SPIN) 118 #define vcpu_lock(v) mtx_lock_spin(&((v)->mtx)) 119 #define vcpu_unlock(v) mtx_unlock_spin(&((v)->mtx)) 120 #define vcpu_assert_locked(v) mtx_assert(&((v)->mtx), MA_OWNED) 121 122 struct mem_seg { 123 size_t len; 124 bool sysmem; 125 struct vm_object *object; 126 }; 127 #define VM_MAX_MEMSEGS 3 128 129 struct mem_map { 130 vm_paddr_t gpa; 131 size_t len; 132 vm_ooffset_t segoff; 133 int segid; 134 int prot; 135 int flags; 136 }; 137 #define VM_MAX_MEMMAPS 4 138 139 /* 140 * Initialization: 141 * (o) initialized the first time the VM is created 142 * (i) initialized when VM is created and when it is reinitialized 143 * (x) initialized before use 144 */ 145 struct vm { 146 void *cookie; /* (i) cpu-specific data */ 147 void *iommu; /* (x) iommu-specific data */ 148 struct vhpet *vhpet; /* (i) virtual HPET */ 149 struct vioapic *vioapic; /* (i) virtual ioapic */ 150 struct vatpic *vatpic; /* (i) virtual atpic */ 151 struct vatpit *vatpit; /* (i) virtual atpit */ 152 struct vpmtmr *vpmtmr; /* (i) virtual ACPI PM timer */ 153 struct vrtc *vrtc; /* (o) virtual RTC */ 154 volatile cpuset_t active_cpus; /* (i) active vcpus */ 155 volatile cpuset_t debug_cpus; /* (i) vcpus stopped for debug */ 156 int suspend; /* (i) stop VM execution */ 157 volatile cpuset_t suspended_cpus; /* (i) suspended vcpus */ 158 volatile cpuset_t halted_cpus; /* (x) cpus in a hard halt */ 159 cpuset_t rendezvous_req_cpus; /* (x) rendezvous requested */ 160 cpuset_t rendezvous_done_cpus; /* (x) rendezvous finished */ 161 void *rendezvous_arg; /* (x) rendezvous func/arg */ 162 vm_rendezvous_func_t rendezvous_func; 163 struct mtx rendezvous_mtx; /* (o) rendezvous lock */ 164 struct mem_map mem_maps[VM_MAX_MEMMAPS]; /* (i) guest address space */ 165 struct mem_seg mem_segs[VM_MAX_MEMSEGS]; /* (o) guest memory regions */ 166 struct vmspace *vmspace; /* (o) guest's address space */ 167 char name[VM_MAX_NAMELEN]; /* (o) virtual machine name */ 168 struct vcpu vcpu[VM_MAXCPU]; /* (i) guest vcpus */ 169 /* The following describe the vm cpu topology */ 170 uint16_t sockets; /* (o) num of sockets */ 171 uint16_t cores; /* (o) num of cores/socket */ 172 uint16_t threads; /* (o) num of threads/core */ 173 uint16_t maxcpus; /* (o) max pluggable cpus */ 174 }; 175 176 static int vmm_initialized; 177 178 static struct vmm_ops *ops; 179 #define VMM_INIT(num) (ops != NULL ? (*ops->init)(num) : 0) 180 #define VMM_CLEANUP() (ops != NULL ? (*ops->cleanup)() : 0) 181 #define VMM_RESUME() (ops != NULL ? (*ops->resume)() : 0) 182 183 #define VMINIT(vm, pmap) (ops != NULL ? (*ops->vminit)(vm, pmap): NULL) 184 #define VMRUN(vmi, vcpu, rip, pmap, evinfo) \ 185 (ops != NULL ? (*ops->vmrun)(vmi, vcpu, rip, pmap, evinfo) : ENXIO) 186 #define VMCLEANUP(vmi) (ops != NULL ? (*ops->vmcleanup)(vmi) : NULL) 187 #define VMSPACE_ALLOC(min, max) \ 188 (ops != NULL ? (*ops->vmspace_alloc)(min, max) : NULL) 189 #define VMSPACE_FREE(vmspace) \ 190 (ops != NULL ? (*ops->vmspace_free)(vmspace) : ENXIO) 191 #define VMGETREG(vmi, vcpu, num, retval) \ 192 (ops != NULL ? (*ops->vmgetreg)(vmi, vcpu, num, retval) : ENXIO) 193 #define VMSETREG(vmi, vcpu, num, val) \ 194 (ops != NULL ? (*ops->vmsetreg)(vmi, vcpu, num, val) : ENXIO) 195 #define VMGETDESC(vmi, vcpu, num, desc) \ 196 (ops != NULL ? (*ops->vmgetdesc)(vmi, vcpu, num, desc) : ENXIO) 197 #define VMSETDESC(vmi, vcpu, num, desc) \ 198 (ops != NULL ? (*ops->vmsetdesc)(vmi, vcpu, num, desc) : ENXIO) 199 #define VMGETCAP(vmi, vcpu, num, retval) \ 200 (ops != NULL ? (*ops->vmgetcap)(vmi, vcpu, num, retval) : ENXIO) 201 #define VMSETCAP(vmi, vcpu, num, val) \ 202 (ops != NULL ? (*ops->vmsetcap)(vmi, vcpu, num, val) : ENXIO) 203 #define VLAPIC_INIT(vmi, vcpu) \ 204 (ops != NULL ? (*ops->vlapic_init)(vmi, vcpu) : NULL) 205 #define VLAPIC_CLEANUP(vmi, vlapic) \ 206 (ops != NULL ? (*ops->vlapic_cleanup)(vmi, vlapic) : NULL) 207 208 #define fpu_start_emulating() load_cr0(rcr0() | CR0_TS) 209 #define fpu_stop_emulating() clts() 210 211 static MALLOC_DEFINE(M_VM, "vm", "vm"); 212 213 /* statistics */ 214 static VMM_STAT(VCPU_TOTAL_RUNTIME, "vcpu total runtime"); 215 216 SYSCTL_NODE(_hw, OID_AUTO, vmm, CTLFLAG_RW, NULL, NULL); 217 218 /* 219 * Halt the guest if all vcpus are executing a HLT instruction with 220 * interrupts disabled. 221 */ 222 static int halt_detection_enabled = 1; 223 SYSCTL_INT(_hw_vmm, OID_AUTO, halt_detection, CTLFLAG_RDTUN, 224 &halt_detection_enabled, 0, 225 "Halt VM if all vcpus execute HLT with interrupts disabled"); 226 227 static int vmm_ipinum; 228 SYSCTL_INT(_hw_vmm, OID_AUTO, ipinum, CTLFLAG_RD, &vmm_ipinum, 0, 229 "IPI vector used for vcpu notifications"); 230 231 static int trace_guest_exceptions; 232 SYSCTL_INT(_hw_vmm, OID_AUTO, trace_guest_exceptions, CTLFLAG_RDTUN, 233 &trace_guest_exceptions, 0, 234 "Trap into hypervisor on all guest exceptions and reflect them back"); 235 236 static void vm_free_memmap(struct vm *vm, int ident); 237 static bool sysmem_mapping(struct vm *vm, struct mem_map *mm); 238 static void vcpu_notify_event_locked(struct vcpu *vcpu, bool lapic_intr); 239 240 #ifdef KTR 241 static const char * 242 vcpu_state2str(enum vcpu_state state) 243 { 244 245 switch (state) { 246 case VCPU_IDLE: 247 return ("idle"); 248 case VCPU_FROZEN: 249 return ("frozen"); 250 case VCPU_RUNNING: 251 return ("running"); 252 case VCPU_SLEEPING: 253 return ("sleeping"); 254 default: 255 return ("unknown"); 256 } 257 } 258 #endif 259 260 static void 261 vcpu_cleanup(struct vm *vm, int i, bool destroy) 262 { 263 struct vcpu *vcpu = &vm->vcpu[i]; 264 265 VLAPIC_CLEANUP(vm->cookie, vcpu->vlapic); 266 if (destroy) { 267 vmm_stat_free(vcpu->stats); 268 fpu_save_area_free(vcpu->guestfpu); 269 } 270 } 271 272 static void 273 vcpu_init(struct vm *vm, int vcpu_id, bool create) 274 { 275 struct vcpu *vcpu; 276 277 KASSERT(vcpu_id >= 0 && vcpu_id < VM_MAXCPU, 278 ("vcpu_init: invalid vcpu %d", vcpu_id)); 279 280 vcpu = &vm->vcpu[vcpu_id]; 281 282 if (create) { 283 KASSERT(!vcpu_lock_initialized(vcpu), ("vcpu %d already " 284 "initialized", vcpu_id)); 285 vcpu_lock_init(vcpu); 286 vcpu->state = VCPU_IDLE; 287 vcpu->hostcpu = NOCPU; 288 vcpu->guestfpu = fpu_save_area_alloc(); 289 vcpu->stats = vmm_stat_alloc(); 290 } 291 292 vcpu->vlapic = VLAPIC_INIT(vm->cookie, vcpu_id); 293 vm_set_x2apic_state(vm, vcpu_id, X2APIC_DISABLED); 294 vcpu->reqidle = 0; 295 vcpu->exitintinfo = 0; 296 vcpu->nmi_pending = 0; 297 vcpu->extint_pending = 0; 298 vcpu->exception_pending = 0; 299 vcpu->guest_xcr0 = XFEATURE_ENABLED_X87; 300 fpu_save_area_reset(vcpu->guestfpu); 301 vmm_stat_init(vcpu->stats); 302 } 303 304 int 305 vcpu_trace_exceptions(struct vm *vm, int vcpuid) 306 { 307 308 return (trace_guest_exceptions); 309 } 310 311 struct vm_exit * 312 vm_exitinfo(struct vm *vm, int cpuid) 313 { 314 struct vcpu *vcpu; 315 316 if (cpuid < 0 || cpuid >= VM_MAXCPU) 317 panic("vm_exitinfo: invalid cpuid %d", cpuid); 318 319 vcpu = &vm->vcpu[cpuid]; 320 321 return (&vcpu->exitinfo); 322 } 323 324 static void 325 vmm_resume(void) 326 { 327 VMM_RESUME(); 328 } 329 330 static int 331 vmm_init(void) 332 { 333 int error; 334 335 vmm_host_state_init(); 336 337 vmm_ipinum = lapic_ipi_alloc(pti ? &IDTVEC(justreturn1_pti) : 338 &IDTVEC(justreturn)); 339 if (vmm_ipinum < 0) 340 vmm_ipinum = IPI_AST; 341 342 error = vmm_mem_init(); 343 if (error) 344 return (error); 345 346 if (vmm_is_intel()) 347 ops = &vmm_ops_intel; 348 else if (vmm_is_amd()) 349 ops = &vmm_ops_amd; 350 else 351 return (ENXIO); 352 353 vmm_resume_p = vmm_resume; 354 355 return (VMM_INIT(vmm_ipinum)); 356 } 357 358 static int 359 vmm_handler(module_t mod, int what, void *arg) 360 { 361 int error; 362 363 switch (what) { 364 case MOD_LOAD: 365 vmmdev_init(); 366 error = vmm_init(); 367 if (error == 0) 368 vmm_initialized = 1; 369 break; 370 case MOD_UNLOAD: 371 error = vmmdev_cleanup(); 372 if (error == 0) { 373 vmm_resume_p = NULL; 374 iommu_cleanup(); 375 if (vmm_ipinum != IPI_AST) 376 lapic_ipi_free(vmm_ipinum); 377 error = VMM_CLEANUP(); 378 /* 379 * Something bad happened - prevent new 380 * VMs from being created 381 */ 382 if (error) 383 vmm_initialized = 0; 384 } 385 break; 386 default: 387 error = 0; 388 break; 389 } 390 return (error); 391 } 392 393 static moduledata_t vmm_kmod = { 394 "vmm", 395 vmm_handler, 396 NULL 397 }; 398 399 /* 400 * vmm initialization has the following dependencies: 401 * 402 * - VT-x initialization requires smp_rendezvous() and therefore must happen 403 * after SMP is fully functional (after SI_SUB_SMP). 404 */ 405 DECLARE_MODULE(vmm, vmm_kmod, SI_SUB_SMP + 1, SI_ORDER_ANY); 406 MODULE_VERSION(vmm, 1); 407 408 static void 409 vm_init(struct vm *vm, bool create) 410 { 411 int i; 412 413 vm->cookie = VMINIT(vm, vmspace_pmap(vm->vmspace)); 414 vm->iommu = NULL; 415 vm->vioapic = vioapic_init(vm); 416 vm->vhpet = vhpet_init(vm); 417 vm->vatpic = vatpic_init(vm); 418 vm->vatpit = vatpit_init(vm); 419 vm->vpmtmr = vpmtmr_init(vm); 420 if (create) 421 vm->vrtc = vrtc_init(vm); 422 423 CPU_ZERO(&vm->active_cpus); 424 CPU_ZERO(&vm->debug_cpus); 425 426 vm->suspend = 0; 427 CPU_ZERO(&vm->suspended_cpus); 428 429 for (i = 0; i < VM_MAXCPU; i++) 430 vcpu_init(vm, i, create); 431 } 432 433 /* 434 * The default CPU topology is a single thread per package. 435 */ 436 u_int cores_per_package = 1; 437 u_int threads_per_core = 1; 438 439 int 440 vm_create(const char *name, struct vm **retvm) 441 { 442 struct vm *vm; 443 struct vmspace *vmspace; 444 445 /* 446 * If vmm.ko could not be successfully initialized then don't attempt 447 * to create the virtual machine. 448 */ 449 if (!vmm_initialized) 450 return (ENXIO); 451 452 if (name == NULL || strlen(name) >= VM_MAX_NAMELEN) 453 return (EINVAL); 454 455 vmspace = VMSPACE_ALLOC(0, VM_MAXUSER_ADDRESS); 456 if (vmspace == NULL) 457 return (ENOMEM); 458 459 vm = malloc(sizeof(struct vm), M_VM, M_WAITOK | M_ZERO); 460 strcpy(vm->name, name); 461 vm->vmspace = vmspace; 462 mtx_init(&vm->rendezvous_mtx, "vm rendezvous lock", 0, MTX_DEF); 463 464 vm->sockets = 1; 465 vm->cores = cores_per_package; /* XXX backwards compatibility */ 466 vm->threads = threads_per_core; /* XXX backwards compatibility */ 467 vm->maxcpus = 0; /* XXX not implemented */ 468 469 vm_init(vm, true); 470 471 *retvm = vm; 472 return (0); 473 } 474 475 void 476 vm_get_topology(struct vm *vm, uint16_t *sockets, uint16_t *cores, 477 uint16_t *threads, uint16_t *maxcpus) 478 { 479 *sockets = vm->sockets; 480 *cores = vm->cores; 481 *threads = vm->threads; 482 *maxcpus = vm->maxcpus; 483 } 484 485 int 486 vm_set_topology(struct vm *vm, uint16_t sockets, uint16_t cores, 487 uint16_t threads, uint16_t maxcpus) 488 { 489 if (maxcpus != 0) 490 return (EINVAL); /* XXX remove when supported */ 491 if ((sockets * cores * threads) > VM_MAXCPU) 492 return (EINVAL); 493 /* XXX need to check sockets * cores * threads == vCPU, how? */ 494 vm->sockets = sockets; 495 vm->cores = cores; 496 vm->threads = threads; 497 vm->maxcpus = maxcpus; 498 return(0); 499 } 500 501 static void 502 vm_cleanup(struct vm *vm, bool destroy) 503 { 504 struct mem_map *mm; 505 int i; 506 507 ppt_unassign_all(vm); 508 509 if (vm->iommu != NULL) 510 iommu_destroy_domain(vm->iommu); 511 512 if (destroy) 513 vrtc_cleanup(vm->vrtc); 514 else 515 vrtc_reset(vm->vrtc); 516 vpmtmr_cleanup(vm->vpmtmr); 517 vatpit_cleanup(vm->vatpit); 518 vhpet_cleanup(vm->vhpet); 519 vatpic_cleanup(vm->vatpic); 520 vioapic_cleanup(vm->vioapic); 521 522 for (i = 0; i < VM_MAXCPU; i++) 523 vcpu_cleanup(vm, i, destroy); 524 525 VMCLEANUP(vm->cookie); 526 527 /* 528 * System memory is removed from the guest address space only when 529 * the VM is destroyed. This is because the mapping remains the same 530 * across VM reset. 531 * 532 * Device memory can be relocated by the guest (e.g. using PCI BARs) 533 * so those mappings are removed on a VM reset. 534 */ 535 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 536 mm = &vm->mem_maps[i]; 537 if (destroy || !sysmem_mapping(vm, mm)) 538 vm_free_memmap(vm, i); 539 } 540 541 if (destroy) { 542 for (i = 0; i < VM_MAX_MEMSEGS; i++) 543 vm_free_memseg(vm, i); 544 545 VMSPACE_FREE(vm->vmspace); 546 vm->vmspace = NULL; 547 } 548 } 549 550 void 551 vm_destroy(struct vm *vm) 552 { 553 vm_cleanup(vm, true); 554 free(vm, M_VM); 555 } 556 557 int 558 vm_reinit(struct vm *vm) 559 { 560 int error; 561 562 /* 563 * A virtual machine can be reset only if all vcpus are suspended. 564 */ 565 if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) { 566 vm_cleanup(vm, false); 567 vm_init(vm, false); 568 error = 0; 569 } else { 570 error = EBUSY; 571 } 572 573 return (error); 574 } 575 576 const char * 577 vm_name(struct vm *vm) 578 { 579 return (vm->name); 580 } 581 582 int 583 vm_map_mmio(struct vm *vm, vm_paddr_t gpa, size_t len, vm_paddr_t hpa) 584 { 585 vm_object_t obj; 586 587 if ((obj = vmm_mmio_alloc(vm->vmspace, gpa, len, hpa)) == NULL) 588 return (ENOMEM); 589 else 590 return (0); 591 } 592 593 int 594 vm_unmap_mmio(struct vm *vm, vm_paddr_t gpa, size_t len) 595 { 596 597 vmm_mmio_free(vm->vmspace, gpa, len); 598 return (0); 599 } 600 601 /* 602 * Return 'true' if 'gpa' is allocated in the guest address space. 603 * 604 * This function is called in the context of a running vcpu which acts as 605 * an implicit lock on 'vm->mem_maps[]'. 606 */ 607 bool 608 vm_mem_allocated(struct vm *vm, int vcpuid, vm_paddr_t gpa) 609 { 610 struct mem_map *mm; 611 int i; 612 613 #ifdef INVARIANTS 614 int hostcpu, state; 615 state = vcpu_get_state(vm, vcpuid, &hostcpu); 616 KASSERT(state == VCPU_RUNNING && hostcpu == curcpu, 617 ("%s: invalid vcpu state %d/%d", __func__, state, hostcpu)); 618 #endif 619 620 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 621 mm = &vm->mem_maps[i]; 622 if (mm->len != 0 && gpa >= mm->gpa && gpa < mm->gpa + mm->len) 623 return (true); /* 'gpa' is sysmem or devmem */ 624 } 625 626 if (ppt_is_mmio(vm, gpa)) 627 return (true); /* 'gpa' is pci passthru mmio */ 628 629 return (false); 630 } 631 632 int 633 vm_alloc_memseg(struct vm *vm, int ident, size_t len, bool sysmem) 634 { 635 struct mem_seg *seg; 636 vm_object_t obj; 637 638 if (ident < 0 || ident >= VM_MAX_MEMSEGS) 639 return (EINVAL); 640 641 if (len == 0 || (len & PAGE_MASK)) 642 return (EINVAL); 643 644 seg = &vm->mem_segs[ident]; 645 if (seg->object != NULL) { 646 if (seg->len == len && seg->sysmem == sysmem) 647 return (EEXIST); 648 else 649 return (EINVAL); 650 } 651 652 obj = vm_object_allocate(OBJT_DEFAULT, len >> PAGE_SHIFT); 653 if (obj == NULL) 654 return (ENOMEM); 655 656 seg->len = len; 657 seg->object = obj; 658 seg->sysmem = sysmem; 659 return (0); 660 } 661 662 int 663 vm_get_memseg(struct vm *vm, int ident, size_t *len, bool *sysmem, 664 vm_object_t *objptr) 665 { 666 struct mem_seg *seg; 667 668 if (ident < 0 || ident >= VM_MAX_MEMSEGS) 669 return (EINVAL); 670 671 seg = &vm->mem_segs[ident]; 672 if (len) 673 *len = seg->len; 674 if (sysmem) 675 *sysmem = seg->sysmem; 676 if (objptr) 677 *objptr = seg->object; 678 return (0); 679 } 680 681 void 682 vm_free_memseg(struct vm *vm, int ident) 683 { 684 struct mem_seg *seg; 685 686 KASSERT(ident >= 0 && ident < VM_MAX_MEMSEGS, 687 ("%s: invalid memseg ident %d", __func__, ident)); 688 689 seg = &vm->mem_segs[ident]; 690 if (seg->object != NULL) { 691 vm_object_deallocate(seg->object); 692 bzero(seg, sizeof(struct mem_seg)); 693 } 694 } 695 696 int 697 vm_mmap_memseg(struct vm *vm, vm_paddr_t gpa, int segid, vm_ooffset_t first, 698 size_t len, int prot, int flags) 699 { 700 struct mem_seg *seg; 701 struct mem_map *m, *map; 702 vm_ooffset_t last; 703 int i, error; 704 705 if (prot == 0 || (prot & ~(VM_PROT_ALL)) != 0) 706 return (EINVAL); 707 708 if (flags & ~VM_MEMMAP_F_WIRED) 709 return (EINVAL); 710 711 if (segid < 0 || segid >= VM_MAX_MEMSEGS) 712 return (EINVAL); 713 714 seg = &vm->mem_segs[segid]; 715 if (seg->object == NULL) 716 return (EINVAL); 717 718 last = first + len; 719 if (first < 0 || first >= last || last > seg->len) 720 return (EINVAL); 721 722 if ((gpa | first | last) & PAGE_MASK) 723 return (EINVAL); 724 725 map = NULL; 726 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 727 m = &vm->mem_maps[i]; 728 if (m->len == 0) { 729 map = m; 730 break; 731 } 732 } 733 734 if (map == NULL) 735 return (ENOSPC); 736 737 error = vm_map_find(&vm->vmspace->vm_map, seg->object, first, &gpa, 738 len, 0, VMFS_NO_SPACE, prot, prot, 0); 739 if (error != KERN_SUCCESS) 740 return (EFAULT); 741 742 vm_object_reference(seg->object); 743 744 if (flags & VM_MEMMAP_F_WIRED) { 745 error = vm_map_wire(&vm->vmspace->vm_map, gpa, gpa + len, 746 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES); 747 if (error != KERN_SUCCESS) { 748 vm_map_remove(&vm->vmspace->vm_map, gpa, gpa + len); 749 return (EFAULT); 750 } 751 } 752 753 map->gpa = gpa; 754 map->len = len; 755 map->segoff = first; 756 map->segid = segid; 757 map->prot = prot; 758 map->flags = flags; 759 return (0); 760 } 761 762 int 763 vm_mmap_getnext(struct vm *vm, vm_paddr_t *gpa, int *segid, 764 vm_ooffset_t *segoff, size_t *len, int *prot, int *flags) 765 { 766 struct mem_map *mm, *mmnext; 767 int i; 768 769 mmnext = NULL; 770 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 771 mm = &vm->mem_maps[i]; 772 if (mm->len == 0 || mm->gpa < *gpa) 773 continue; 774 if (mmnext == NULL || mm->gpa < mmnext->gpa) 775 mmnext = mm; 776 } 777 778 if (mmnext != NULL) { 779 *gpa = mmnext->gpa; 780 if (segid) 781 *segid = mmnext->segid; 782 if (segoff) 783 *segoff = mmnext->segoff; 784 if (len) 785 *len = mmnext->len; 786 if (prot) 787 *prot = mmnext->prot; 788 if (flags) 789 *flags = mmnext->flags; 790 return (0); 791 } else { 792 return (ENOENT); 793 } 794 } 795 796 static void 797 vm_free_memmap(struct vm *vm, int ident) 798 { 799 struct mem_map *mm; 800 int error; 801 802 mm = &vm->mem_maps[ident]; 803 if (mm->len) { 804 error = vm_map_remove(&vm->vmspace->vm_map, mm->gpa, 805 mm->gpa + mm->len); 806 KASSERT(error == KERN_SUCCESS, ("%s: vm_map_remove error %d", 807 __func__, error)); 808 bzero(mm, sizeof(struct mem_map)); 809 } 810 } 811 812 static __inline bool 813 sysmem_mapping(struct vm *vm, struct mem_map *mm) 814 { 815 816 if (mm->len != 0 && vm->mem_segs[mm->segid].sysmem) 817 return (true); 818 else 819 return (false); 820 } 821 822 static vm_paddr_t 823 sysmem_maxaddr(struct vm *vm) 824 { 825 struct mem_map *mm; 826 vm_paddr_t maxaddr; 827 int i; 828 829 maxaddr = 0; 830 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 831 mm = &vm->mem_maps[i]; 832 if (sysmem_mapping(vm, mm)) { 833 if (maxaddr < mm->gpa + mm->len) 834 maxaddr = mm->gpa + mm->len; 835 } 836 } 837 return (maxaddr); 838 } 839 840 static void 841 vm_iommu_modify(struct vm *vm, boolean_t map) 842 { 843 int i, sz; 844 vm_paddr_t gpa, hpa; 845 struct mem_map *mm; 846 void *vp, *cookie, *host_domain; 847 848 sz = PAGE_SIZE; 849 host_domain = iommu_host_domain(); 850 851 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 852 mm = &vm->mem_maps[i]; 853 if (!sysmem_mapping(vm, mm)) 854 continue; 855 856 if (map) { 857 KASSERT((mm->flags & VM_MEMMAP_F_IOMMU) == 0, 858 ("iommu map found invalid memmap %#lx/%#lx/%#x", 859 mm->gpa, mm->len, mm->flags)); 860 if ((mm->flags & VM_MEMMAP_F_WIRED) == 0) 861 continue; 862 mm->flags |= VM_MEMMAP_F_IOMMU; 863 } else { 864 if ((mm->flags & VM_MEMMAP_F_IOMMU) == 0) 865 continue; 866 mm->flags &= ~VM_MEMMAP_F_IOMMU; 867 KASSERT((mm->flags & VM_MEMMAP_F_WIRED) != 0, 868 ("iommu unmap found invalid memmap %#lx/%#lx/%#x", 869 mm->gpa, mm->len, mm->flags)); 870 } 871 872 gpa = mm->gpa; 873 while (gpa < mm->gpa + mm->len) { 874 vp = vm_gpa_hold(vm, -1, gpa, PAGE_SIZE, VM_PROT_WRITE, 875 &cookie); 876 KASSERT(vp != NULL, ("vm(%s) could not map gpa %#lx", 877 vm_name(vm), gpa)); 878 879 vm_gpa_release(cookie); 880 881 hpa = DMAP_TO_PHYS((uintptr_t)vp); 882 if (map) { 883 iommu_create_mapping(vm->iommu, gpa, hpa, sz); 884 iommu_remove_mapping(host_domain, hpa, sz); 885 } else { 886 iommu_remove_mapping(vm->iommu, gpa, sz); 887 iommu_create_mapping(host_domain, hpa, hpa, sz); 888 } 889 890 gpa += PAGE_SIZE; 891 } 892 } 893 894 /* 895 * Invalidate the cached translations associated with the domain 896 * from which pages were removed. 897 */ 898 if (map) 899 iommu_invalidate_tlb(host_domain); 900 else 901 iommu_invalidate_tlb(vm->iommu); 902 } 903 904 #define vm_iommu_unmap(vm) vm_iommu_modify((vm), FALSE) 905 #define vm_iommu_map(vm) vm_iommu_modify((vm), TRUE) 906 907 int 908 vm_unassign_pptdev(struct vm *vm, int bus, int slot, int func) 909 { 910 int error; 911 912 error = ppt_unassign_device(vm, bus, slot, func); 913 if (error) 914 return (error); 915 916 if (ppt_assigned_devices(vm) == 0) 917 vm_iommu_unmap(vm); 918 919 return (0); 920 } 921 922 int 923 vm_assign_pptdev(struct vm *vm, int bus, int slot, int func) 924 { 925 int error; 926 vm_paddr_t maxaddr; 927 928 /* Set up the IOMMU to do the 'gpa' to 'hpa' translation */ 929 if (ppt_assigned_devices(vm) == 0) { 930 KASSERT(vm->iommu == NULL, 931 ("vm_assign_pptdev: iommu must be NULL")); 932 maxaddr = sysmem_maxaddr(vm); 933 vm->iommu = iommu_create_domain(maxaddr); 934 if (vm->iommu == NULL) 935 return (ENXIO); 936 vm_iommu_map(vm); 937 } 938 939 error = ppt_assign_device(vm, bus, slot, func); 940 return (error); 941 } 942 943 void * 944 vm_gpa_hold(struct vm *vm, int vcpuid, vm_paddr_t gpa, size_t len, int reqprot, 945 void **cookie) 946 { 947 int i, count, pageoff; 948 struct mem_map *mm; 949 vm_page_t m; 950 #ifdef INVARIANTS 951 /* 952 * All vcpus are frozen by ioctls that modify the memory map 953 * (e.g. VM_MMAP_MEMSEG). Therefore 'vm->memmap[]' stability is 954 * guaranteed if at least one vcpu is in the VCPU_FROZEN state. 955 */ 956 int state; 957 KASSERT(vcpuid >= -1 && vcpuid < VM_MAXCPU, ("%s: invalid vcpuid %d", 958 __func__, vcpuid)); 959 for (i = 0; i < VM_MAXCPU; i++) { 960 if (vcpuid != -1 && vcpuid != i) 961 continue; 962 state = vcpu_get_state(vm, i, NULL); 963 KASSERT(state == VCPU_FROZEN, ("%s: invalid vcpu state %d", 964 __func__, state)); 965 } 966 #endif 967 pageoff = gpa & PAGE_MASK; 968 if (len > PAGE_SIZE - pageoff) 969 panic("vm_gpa_hold: invalid gpa/len: 0x%016lx/%lu", gpa, len); 970 971 count = 0; 972 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 973 mm = &vm->mem_maps[i]; 974 if (sysmem_mapping(vm, mm) && gpa >= mm->gpa && 975 gpa < mm->gpa + mm->len) { 976 count = vm_fault_quick_hold_pages(&vm->vmspace->vm_map, 977 trunc_page(gpa), PAGE_SIZE, reqprot, &m, 1); 978 break; 979 } 980 } 981 982 if (count == 1) { 983 *cookie = m; 984 return ((void *)(PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m)) + pageoff)); 985 } else { 986 *cookie = NULL; 987 return (NULL); 988 } 989 } 990 991 void 992 vm_gpa_release(void *cookie) 993 { 994 vm_page_t m = cookie; 995 996 vm_page_lock(m); 997 vm_page_unhold(m); 998 vm_page_unlock(m); 999 } 1000 1001 int 1002 vm_get_register(struct vm *vm, int vcpu, int reg, uint64_t *retval) 1003 { 1004 1005 if (vcpu < 0 || vcpu >= VM_MAXCPU) 1006 return (EINVAL); 1007 1008 if (reg >= VM_REG_LAST) 1009 return (EINVAL); 1010 1011 return (VMGETREG(vm->cookie, vcpu, reg, retval)); 1012 } 1013 1014 int 1015 vm_set_register(struct vm *vm, int vcpuid, int reg, uint64_t val) 1016 { 1017 struct vcpu *vcpu; 1018 int error; 1019 1020 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1021 return (EINVAL); 1022 1023 if (reg >= VM_REG_LAST) 1024 return (EINVAL); 1025 1026 error = VMSETREG(vm->cookie, vcpuid, reg, val); 1027 if (error || reg != VM_REG_GUEST_RIP) 1028 return (error); 1029 1030 /* Set 'nextrip' to match the value of %rip */ 1031 VCPU_CTR1(vm, vcpuid, "Setting nextrip to %#lx", val); 1032 vcpu = &vm->vcpu[vcpuid]; 1033 vcpu->nextrip = val; 1034 return (0); 1035 } 1036 1037 static boolean_t 1038 is_descriptor_table(int reg) 1039 { 1040 1041 switch (reg) { 1042 case VM_REG_GUEST_IDTR: 1043 case VM_REG_GUEST_GDTR: 1044 return (TRUE); 1045 default: 1046 return (FALSE); 1047 } 1048 } 1049 1050 static boolean_t 1051 is_segment_register(int reg) 1052 { 1053 1054 switch (reg) { 1055 case VM_REG_GUEST_ES: 1056 case VM_REG_GUEST_CS: 1057 case VM_REG_GUEST_SS: 1058 case VM_REG_GUEST_DS: 1059 case VM_REG_GUEST_FS: 1060 case VM_REG_GUEST_GS: 1061 case VM_REG_GUEST_TR: 1062 case VM_REG_GUEST_LDTR: 1063 return (TRUE); 1064 default: 1065 return (FALSE); 1066 } 1067 } 1068 1069 int 1070 vm_get_seg_desc(struct vm *vm, int vcpu, int reg, 1071 struct seg_desc *desc) 1072 { 1073 1074 if (vcpu < 0 || vcpu >= VM_MAXCPU) 1075 return (EINVAL); 1076 1077 if (!is_segment_register(reg) && !is_descriptor_table(reg)) 1078 return (EINVAL); 1079 1080 return (VMGETDESC(vm->cookie, vcpu, reg, desc)); 1081 } 1082 1083 int 1084 vm_set_seg_desc(struct vm *vm, int vcpu, int reg, 1085 struct seg_desc *desc) 1086 { 1087 if (vcpu < 0 || vcpu >= VM_MAXCPU) 1088 return (EINVAL); 1089 1090 if (!is_segment_register(reg) && !is_descriptor_table(reg)) 1091 return (EINVAL); 1092 1093 return (VMSETDESC(vm->cookie, vcpu, reg, desc)); 1094 } 1095 1096 static void 1097 restore_guest_fpustate(struct vcpu *vcpu) 1098 { 1099 1100 /* flush host state to the pcb */ 1101 fpuexit(curthread); 1102 1103 /* restore guest FPU state */ 1104 fpu_stop_emulating(); 1105 fpurestore(vcpu->guestfpu); 1106 1107 /* restore guest XCR0 if XSAVE is enabled in the host */ 1108 if (rcr4() & CR4_XSAVE) 1109 load_xcr(0, vcpu->guest_xcr0); 1110 1111 /* 1112 * The FPU is now "dirty" with the guest's state so turn on emulation 1113 * to trap any access to the FPU by the host. 1114 */ 1115 fpu_start_emulating(); 1116 } 1117 1118 static void 1119 save_guest_fpustate(struct vcpu *vcpu) 1120 { 1121 1122 if ((rcr0() & CR0_TS) == 0) 1123 panic("fpu emulation not enabled in host!"); 1124 1125 /* save guest XCR0 and restore host XCR0 */ 1126 if (rcr4() & CR4_XSAVE) { 1127 vcpu->guest_xcr0 = rxcr(0); 1128 load_xcr(0, vmm_get_host_xcr0()); 1129 } 1130 1131 /* save guest FPU state */ 1132 fpu_stop_emulating(); 1133 fpusave(vcpu->guestfpu); 1134 fpu_start_emulating(); 1135 } 1136 1137 static VMM_STAT(VCPU_IDLE_TICKS, "number of ticks vcpu was idle"); 1138 1139 static int 1140 vcpu_set_state_locked(struct vm *vm, int vcpuid, enum vcpu_state newstate, 1141 bool from_idle) 1142 { 1143 struct vcpu *vcpu; 1144 int error; 1145 1146 vcpu = &vm->vcpu[vcpuid]; 1147 vcpu_assert_locked(vcpu); 1148 1149 /* 1150 * State transitions from the vmmdev_ioctl() must always begin from 1151 * the VCPU_IDLE state. This guarantees that there is only a single 1152 * ioctl() operating on a vcpu at any point. 1153 */ 1154 if (from_idle) { 1155 while (vcpu->state != VCPU_IDLE) { 1156 vcpu->reqidle = 1; 1157 vcpu_notify_event_locked(vcpu, false); 1158 VCPU_CTR1(vm, vcpuid, "vcpu state change from %s to " 1159 "idle requested", vcpu_state2str(vcpu->state)); 1160 msleep_spin(&vcpu->state, &vcpu->mtx, "vmstat", hz); 1161 } 1162 } else { 1163 KASSERT(vcpu->state != VCPU_IDLE, ("invalid transition from " 1164 "vcpu idle state")); 1165 } 1166 1167 if (vcpu->state == VCPU_RUNNING) { 1168 KASSERT(vcpu->hostcpu == curcpu, ("curcpu %d and hostcpu %d " 1169 "mismatch for running vcpu", curcpu, vcpu->hostcpu)); 1170 } else { 1171 KASSERT(vcpu->hostcpu == NOCPU, ("Invalid hostcpu %d for a " 1172 "vcpu that is not running", vcpu->hostcpu)); 1173 } 1174 1175 /* 1176 * The following state transitions are allowed: 1177 * IDLE -> FROZEN -> IDLE 1178 * FROZEN -> RUNNING -> FROZEN 1179 * FROZEN -> SLEEPING -> FROZEN 1180 */ 1181 switch (vcpu->state) { 1182 case VCPU_IDLE: 1183 case VCPU_RUNNING: 1184 case VCPU_SLEEPING: 1185 error = (newstate != VCPU_FROZEN); 1186 break; 1187 case VCPU_FROZEN: 1188 error = (newstate == VCPU_FROZEN); 1189 break; 1190 default: 1191 error = 1; 1192 break; 1193 } 1194 1195 if (error) 1196 return (EBUSY); 1197 1198 VCPU_CTR2(vm, vcpuid, "vcpu state changed from %s to %s", 1199 vcpu_state2str(vcpu->state), vcpu_state2str(newstate)); 1200 1201 vcpu->state = newstate; 1202 if (newstate == VCPU_RUNNING) 1203 vcpu->hostcpu = curcpu; 1204 else 1205 vcpu->hostcpu = NOCPU; 1206 1207 if (newstate == VCPU_IDLE) 1208 wakeup(&vcpu->state); 1209 1210 return (0); 1211 } 1212 1213 static void 1214 vcpu_require_state(struct vm *vm, int vcpuid, enum vcpu_state newstate) 1215 { 1216 int error; 1217 1218 if ((error = vcpu_set_state(vm, vcpuid, newstate, false)) != 0) 1219 panic("Error %d setting state to %d\n", error, newstate); 1220 } 1221 1222 static void 1223 vcpu_require_state_locked(struct vm *vm, int vcpuid, enum vcpu_state newstate) 1224 { 1225 int error; 1226 1227 if ((error = vcpu_set_state_locked(vm, vcpuid, newstate, false)) != 0) 1228 panic("Error %d setting state to %d", error, newstate); 1229 } 1230 1231 static void 1232 vm_set_rendezvous_func(struct vm *vm, vm_rendezvous_func_t func) 1233 { 1234 1235 KASSERT(mtx_owned(&vm->rendezvous_mtx), ("rendezvous_mtx not locked")); 1236 1237 /* 1238 * Update 'rendezvous_func' and execute a write memory barrier to 1239 * ensure that it is visible across all host cpus. This is not needed 1240 * for correctness but it does ensure that all the vcpus will notice 1241 * that the rendezvous is requested immediately. 1242 */ 1243 vm->rendezvous_func = func; 1244 wmb(); 1245 } 1246 1247 #define RENDEZVOUS_CTR0(vm, vcpuid, fmt) \ 1248 do { \ 1249 if (vcpuid >= 0) \ 1250 VCPU_CTR0(vm, vcpuid, fmt); \ 1251 else \ 1252 VM_CTR0(vm, fmt); \ 1253 } while (0) 1254 1255 static void 1256 vm_handle_rendezvous(struct vm *vm, int vcpuid) 1257 { 1258 1259 KASSERT(vcpuid == -1 || (vcpuid >= 0 && vcpuid < VM_MAXCPU), 1260 ("vm_handle_rendezvous: invalid vcpuid %d", vcpuid)); 1261 1262 mtx_lock(&vm->rendezvous_mtx); 1263 while (vm->rendezvous_func != NULL) { 1264 /* 'rendezvous_req_cpus' must be a subset of 'active_cpus' */ 1265 CPU_AND(&vm->rendezvous_req_cpus, &vm->active_cpus); 1266 1267 if (vcpuid != -1 && 1268 CPU_ISSET(vcpuid, &vm->rendezvous_req_cpus) && 1269 !CPU_ISSET(vcpuid, &vm->rendezvous_done_cpus)) { 1270 VCPU_CTR0(vm, vcpuid, "Calling rendezvous func"); 1271 (*vm->rendezvous_func)(vm, vcpuid, vm->rendezvous_arg); 1272 CPU_SET(vcpuid, &vm->rendezvous_done_cpus); 1273 } 1274 if (CPU_CMP(&vm->rendezvous_req_cpus, 1275 &vm->rendezvous_done_cpus) == 0) { 1276 VCPU_CTR0(vm, vcpuid, "Rendezvous completed"); 1277 vm_set_rendezvous_func(vm, NULL); 1278 wakeup(&vm->rendezvous_func); 1279 break; 1280 } 1281 RENDEZVOUS_CTR0(vm, vcpuid, "Wait for rendezvous completion"); 1282 mtx_sleep(&vm->rendezvous_func, &vm->rendezvous_mtx, 0, 1283 "vmrndv", 0); 1284 } 1285 mtx_unlock(&vm->rendezvous_mtx); 1286 } 1287 1288 /* 1289 * Emulate a guest 'hlt' by sleeping until the vcpu is ready to run. 1290 */ 1291 static int 1292 vm_handle_hlt(struct vm *vm, int vcpuid, bool intr_disabled, bool *retu) 1293 { 1294 struct vcpu *vcpu; 1295 const char *wmesg; 1296 int t, vcpu_halted, vm_halted; 1297 1298 KASSERT(!CPU_ISSET(vcpuid, &vm->halted_cpus), ("vcpu already halted")); 1299 1300 vcpu = &vm->vcpu[vcpuid]; 1301 vcpu_halted = 0; 1302 vm_halted = 0; 1303 1304 vcpu_lock(vcpu); 1305 while (1) { 1306 /* 1307 * Do a final check for pending NMI or interrupts before 1308 * really putting this thread to sleep. Also check for 1309 * software events that would cause this vcpu to wakeup. 1310 * 1311 * These interrupts/events could have happened after the 1312 * vcpu returned from VMRUN() and before it acquired the 1313 * vcpu lock above. 1314 */ 1315 if (vm->rendezvous_func != NULL || vm->suspend || vcpu->reqidle) 1316 break; 1317 if (vm_nmi_pending(vm, vcpuid)) 1318 break; 1319 if (!intr_disabled) { 1320 if (vm_extint_pending(vm, vcpuid) || 1321 vlapic_pending_intr(vcpu->vlapic, NULL)) { 1322 break; 1323 } 1324 } 1325 1326 /* Don't go to sleep if the vcpu thread needs to yield */ 1327 if (vcpu_should_yield(vm, vcpuid)) 1328 break; 1329 1330 if (vcpu_debugged(vm, vcpuid)) 1331 break; 1332 1333 /* 1334 * Some Linux guests implement "halt" by having all vcpus 1335 * execute HLT with interrupts disabled. 'halted_cpus' keeps 1336 * track of the vcpus that have entered this state. When all 1337 * vcpus enter the halted state the virtual machine is halted. 1338 */ 1339 if (intr_disabled) { 1340 wmesg = "vmhalt"; 1341 VCPU_CTR0(vm, vcpuid, "Halted"); 1342 if (!vcpu_halted && halt_detection_enabled) { 1343 vcpu_halted = 1; 1344 CPU_SET_ATOMIC(vcpuid, &vm->halted_cpus); 1345 } 1346 if (CPU_CMP(&vm->halted_cpus, &vm->active_cpus) == 0) { 1347 vm_halted = 1; 1348 break; 1349 } 1350 } else { 1351 wmesg = "vmidle"; 1352 } 1353 1354 t = ticks; 1355 vcpu_require_state_locked(vm, vcpuid, VCPU_SLEEPING); 1356 /* 1357 * XXX msleep_spin() cannot be interrupted by signals so 1358 * wake up periodically to check pending signals. 1359 */ 1360 msleep_spin(vcpu, &vcpu->mtx, wmesg, hz); 1361 vcpu_require_state_locked(vm, vcpuid, VCPU_FROZEN); 1362 vmm_stat_incr(vm, vcpuid, VCPU_IDLE_TICKS, ticks - t); 1363 } 1364 1365 if (vcpu_halted) 1366 CPU_CLR_ATOMIC(vcpuid, &vm->halted_cpus); 1367 1368 vcpu_unlock(vcpu); 1369 1370 if (vm_halted) 1371 vm_suspend(vm, VM_SUSPEND_HALT); 1372 1373 return (0); 1374 } 1375 1376 static int 1377 vm_handle_paging(struct vm *vm, int vcpuid, bool *retu) 1378 { 1379 int rv, ftype; 1380 struct vm_map *map; 1381 struct vcpu *vcpu; 1382 struct vm_exit *vme; 1383 1384 vcpu = &vm->vcpu[vcpuid]; 1385 vme = &vcpu->exitinfo; 1386 1387 KASSERT(vme->inst_length == 0, ("%s: invalid inst_length %d", 1388 __func__, vme->inst_length)); 1389 1390 ftype = vme->u.paging.fault_type; 1391 KASSERT(ftype == VM_PROT_READ || 1392 ftype == VM_PROT_WRITE || ftype == VM_PROT_EXECUTE, 1393 ("vm_handle_paging: invalid fault_type %d", ftype)); 1394 1395 if (ftype == VM_PROT_READ || ftype == VM_PROT_WRITE) { 1396 rv = pmap_emulate_accessed_dirty(vmspace_pmap(vm->vmspace), 1397 vme->u.paging.gpa, ftype); 1398 if (rv == 0) { 1399 VCPU_CTR2(vm, vcpuid, "%s bit emulation for gpa %#lx", 1400 ftype == VM_PROT_READ ? "accessed" : "dirty", 1401 vme->u.paging.gpa); 1402 goto done; 1403 } 1404 } 1405 1406 map = &vm->vmspace->vm_map; 1407 rv = vm_fault(map, vme->u.paging.gpa, ftype, VM_FAULT_NORMAL); 1408 1409 VCPU_CTR3(vm, vcpuid, "vm_handle_paging rv = %d, gpa = %#lx, " 1410 "ftype = %d", rv, vme->u.paging.gpa, ftype); 1411 1412 if (rv != KERN_SUCCESS) 1413 return (EFAULT); 1414 done: 1415 return (0); 1416 } 1417 1418 static int 1419 vm_handle_inst_emul(struct vm *vm, int vcpuid, bool *retu) 1420 { 1421 struct vie *vie; 1422 struct vcpu *vcpu; 1423 struct vm_exit *vme; 1424 uint64_t gla, gpa, cs_base; 1425 struct vm_guest_paging *paging; 1426 mem_region_read_t mread; 1427 mem_region_write_t mwrite; 1428 enum vm_cpu_mode cpu_mode; 1429 int cs_d, error, fault; 1430 1431 vcpu = &vm->vcpu[vcpuid]; 1432 vme = &vcpu->exitinfo; 1433 1434 KASSERT(vme->inst_length == 0, ("%s: invalid inst_length %d", 1435 __func__, vme->inst_length)); 1436 1437 gla = vme->u.inst_emul.gla; 1438 gpa = vme->u.inst_emul.gpa; 1439 cs_base = vme->u.inst_emul.cs_base; 1440 cs_d = vme->u.inst_emul.cs_d; 1441 vie = &vme->u.inst_emul.vie; 1442 paging = &vme->u.inst_emul.paging; 1443 cpu_mode = paging->cpu_mode; 1444 1445 VCPU_CTR1(vm, vcpuid, "inst_emul fault accessing gpa %#lx", gpa); 1446 1447 /* Fetch, decode and emulate the faulting instruction */ 1448 if (vie->num_valid == 0) { 1449 error = vmm_fetch_instruction(vm, vcpuid, paging, vme->rip + 1450 cs_base, VIE_INST_SIZE, vie, &fault); 1451 } else { 1452 /* 1453 * The instruction bytes have already been copied into 'vie' 1454 */ 1455 error = fault = 0; 1456 } 1457 if (error || fault) 1458 return (error); 1459 1460 if (vmm_decode_instruction(vm, vcpuid, gla, cpu_mode, cs_d, vie) != 0) { 1461 VCPU_CTR1(vm, vcpuid, "Error decoding instruction at %#lx", 1462 vme->rip + cs_base); 1463 *retu = true; /* dump instruction bytes in userspace */ 1464 return (0); 1465 } 1466 1467 /* 1468 * Update 'nextrip' based on the length of the emulated instruction. 1469 */ 1470 vme->inst_length = vie->num_processed; 1471 vcpu->nextrip += vie->num_processed; 1472 VCPU_CTR1(vm, vcpuid, "nextrip updated to %#lx after instruction " 1473 "decoding", vcpu->nextrip); 1474 1475 /* return to userland unless this is an in-kernel emulated device */ 1476 if (gpa >= DEFAULT_APIC_BASE && gpa < DEFAULT_APIC_BASE + PAGE_SIZE) { 1477 mread = lapic_mmio_read; 1478 mwrite = lapic_mmio_write; 1479 } else if (gpa >= VIOAPIC_BASE && gpa < VIOAPIC_BASE + VIOAPIC_SIZE) { 1480 mread = vioapic_mmio_read; 1481 mwrite = vioapic_mmio_write; 1482 } else if (gpa >= VHPET_BASE && gpa < VHPET_BASE + VHPET_SIZE) { 1483 mread = vhpet_mmio_read; 1484 mwrite = vhpet_mmio_write; 1485 } else { 1486 *retu = true; 1487 return (0); 1488 } 1489 1490 error = vmm_emulate_instruction(vm, vcpuid, gpa, vie, paging, 1491 mread, mwrite, retu); 1492 1493 return (error); 1494 } 1495 1496 static int 1497 vm_handle_suspend(struct vm *vm, int vcpuid, bool *retu) 1498 { 1499 int i, done; 1500 struct vcpu *vcpu; 1501 1502 done = 0; 1503 vcpu = &vm->vcpu[vcpuid]; 1504 1505 CPU_SET_ATOMIC(vcpuid, &vm->suspended_cpus); 1506 1507 /* 1508 * Wait until all 'active_cpus' have suspended themselves. 1509 * 1510 * Since a VM may be suspended at any time including when one or 1511 * more vcpus are doing a rendezvous we need to call the rendezvous 1512 * handler while we are waiting to prevent a deadlock. 1513 */ 1514 vcpu_lock(vcpu); 1515 while (1) { 1516 if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) { 1517 VCPU_CTR0(vm, vcpuid, "All vcpus suspended"); 1518 break; 1519 } 1520 1521 if (vm->rendezvous_func == NULL) { 1522 VCPU_CTR0(vm, vcpuid, "Sleeping during suspend"); 1523 vcpu_require_state_locked(vm, vcpuid, VCPU_SLEEPING); 1524 msleep_spin(vcpu, &vcpu->mtx, "vmsusp", hz); 1525 vcpu_require_state_locked(vm, vcpuid, VCPU_FROZEN); 1526 } else { 1527 VCPU_CTR0(vm, vcpuid, "Rendezvous during suspend"); 1528 vcpu_unlock(vcpu); 1529 vm_handle_rendezvous(vm, vcpuid); 1530 vcpu_lock(vcpu); 1531 } 1532 } 1533 vcpu_unlock(vcpu); 1534 1535 /* 1536 * Wakeup the other sleeping vcpus and return to userspace. 1537 */ 1538 for (i = 0; i < VM_MAXCPU; i++) { 1539 if (CPU_ISSET(i, &vm->suspended_cpus)) { 1540 vcpu_notify_event(vm, i, false); 1541 } 1542 } 1543 1544 *retu = true; 1545 return (0); 1546 } 1547 1548 static int 1549 vm_handle_reqidle(struct vm *vm, int vcpuid, bool *retu) 1550 { 1551 struct vcpu *vcpu = &vm->vcpu[vcpuid]; 1552 1553 vcpu_lock(vcpu); 1554 KASSERT(vcpu->reqidle, ("invalid vcpu reqidle %d", vcpu->reqidle)); 1555 vcpu->reqidle = 0; 1556 vcpu_unlock(vcpu); 1557 *retu = true; 1558 return (0); 1559 } 1560 1561 int 1562 vm_suspend(struct vm *vm, enum vm_suspend_how how) 1563 { 1564 int i; 1565 1566 if (how <= VM_SUSPEND_NONE || how >= VM_SUSPEND_LAST) 1567 return (EINVAL); 1568 1569 if (atomic_cmpset_int(&vm->suspend, 0, how) == 0) { 1570 VM_CTR2(vm, "virtual machine already suspended %d/%d", 1571 vm->suspend, how); 1572 return (EALREADY); 1573 } 1574 1575 VM_CTR1(vm, "virtual machine successfully suspended %d", how); 1576 1577 /* 1578 * Notify all active vcpus that they are now suspended. 1579 */ 1580 for (i = 0; i < VM_MAXCPU; i++) { 1581 if (CPU_ISSET(i, &vm->active_cpus)) 1582 vcpu_notify_event(vm, i, false); 1583 } 1584 1585 return (0); 1586 } 1587 1588 void 1589 vm_exit_suspended(struct vm *vm, int vcpuid, uint64_t rip) 1590 { 1591 struct vm_exit *vmexit; 1592 1593 KASSERT(vm->suspend > VM_SUSPEND_NONE && vm->suspend < VM_SUSPEND_LAST, 1594 ("vm_exit_suspended: invalid suspend type %d", vm->suspend)); 1595 1596 vmexit = vm_exitinfo(vm, vcpuid); 1597 vmexit->rip = rip; 1598 vmexit->inst_length = 0; 1599 vmexit->exitcode = VM_EXITCODE_SUSPENDED; 1600 vmexit->u.suspended.how = vm->suspend; 1601 } 1602 1603 void 1604 vm_exit_debug(struct vm *vm, int vcpuid, uint64_t rip) 1605 { 1606 struct vm_exit *vmexit; 1607 1608 vmexit = vm_exitinfo(vm, vcpuid); 1609 vmexit->rip = rip; 1610 vmexit->inst_length = 0; 1611 vmexit->exitcode = VM_EXITCODE_DEBUG; 1612 } 1613 1614 void 1615 vm_exit_rendezvous(struct vm *vm, int vcpuid, uint64_t rip) 1616 { 1617 struct vm_exit *vmexit; 1618 1619 KASSERT(vm->rendezvous_func != NULL, ("rendezvous not in progress")); 1620 1621 vmexit = vm_exitinfo(vm, vcpuid); 1622 vmexit->rip = rip; 1623 vmexit->inst_length = 0; 1624 vmexit->exitcode = VM_EXITCODE_RENDEZVOUS; 1625 vmm_stat_incr(vm, vcpuid, VMEXIT_RENDEZVOUS, 1); 1626 } 1627 1628 void 1629 vm_exit_reqidle(struct vm *vm, int vcpuid, uint64_t rip) 1630 { 1631 struct vm_exit *vmexit; 1632 1633 vmexit = vm_exitinfo(vm, vcpuid); 1634 vmexit->rip = rip; 1635 vmexit->inst_length = 0; 1636 vmexit->exitcode = VM_EXITCODE_REQIDLE; 1637 vmm_stat_incr(vm, vcpuid, VMEXIT_REQIDLE, 1); 1638 } 1639 1640 void 1641 vm_exit_astpending(struct vm *vm, int vcpuid, uint64_t rip) 1642 { 1643 struct vm_exit *vmexit; 1644 1645 vmexit = vm_exitinfo(vm, vcpuid); 1646 vmexit->rip = rip; 1647 vmexit->inst_length = 0; 1648 vmexit->exitcode = VM_EXITCODE_BOGUS; 1649 vmm_stat_incr(vm, vcpuid, VMEXIT_ASTPENDING, 1); 1650 } 1651 1652 int 1653 vm_run(struct vm *vm, struct vm_run *vmrun) 1654 { 1655 struct vm_eventinfo evinfo; 1656 int error, vcpuid; 1657 struct vcpu *vcpu; 1658 struct pcb *pcb; 1659 uint64_t tscval; 1660 struct vm_exit *vme; 1661 bool retu, intr_disabled; 1662 pmap_t pmap; 1663 1664 vcpuid = vmrun->cpuid; 1665 1666 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1667 return (EINVAL); 1668 1669 if (!CPU_ISSET(vcpuid, &vm->active_cpus)) 1670 return (EINVAL); 1671 1672 if (CPU_ISSET(vcpuid, &vm->suspended_cpus)) 1673 return (EINVAL); 1674 1675 pmap = vmspace_pmap(vm->vmspace); 1676 vcpu = &vm->vcpu[vcpuid]; 1677 vme = &vcpu->exitinfo; 1678 evinfo.rptr = &vm->rendezvous_func; 1679 evinfo.sptr = &vm->suspend; 1680 evinfo.iptr = &vcpu->reqidle; 1681 restart: 1682 critical_enter(); 1683 1684 KASSERT(!CPU_ISSET(curcpu, &pmap->pm_active), 1685 ("vm_run: absurd pm_active")); 1686 1687 tscval = rdtsc(); 1688 1689 pcb = PCPU_GET(curpcb); 1690 set_pcb_flags(pcb, PCB_FULL_IRET); 1691 1692 restore_guest_fpustate(vcpu); 1693 1694 vcpu_require_state(vm, vcpuid, VCPU_RUNNING); 1695 error = VMRUN(vm->cookie, vcpuid, vcpu->nextrip, pmap, &evinfo); 1696 vcpu_require_state(vm, vcpuid, VCPU_FROZEN); 1697 1698 save_guest_fpustate(vcpu); 1699 1700 vmm_stat_incr(vm, vcpuid, VCPU_TOTAL_RUNTIME, rdtsc() - tscval); 1701 1702 critical_exit(); 1703 1704 if (error == 0) { 1705 retu = false; 1706 vcpu->nextrip = vme->rip + vme->inst_length; 1707 switch (vme->exitcode) { 1708 case VM_EXITCODE_REQIDLE: 1709 error = vm_handle_reqidle(vm, vcpuid, &retu); 1710 break; 1711 case VM_EXITCODE_SUSPENDED: 1712 error = vm_handle_suspend(vm, vcpuid, &retu); 1713 break; 1714 case VM_EXITCODE_IOAPIC_EOI: 1715 vioapic_process_eoi(vm, vcpuid, 1716 vme->u.ioapic_eoi.vector); 1717 break; 1718 case VM_EXITCODE_RENDEZVOUS: 1719 vm_handle_rendezvous(vm, vcpuid); 1720 error = 0; 1721 break; 1722 case VM_EXITCODE_HLT: 1723 intr_disabled = ((vme->u.hlt.rflags & PSL_I) == 0); 1724 error = vm_handle_hlt(vm, vcpuid, intr_disabled, &retu); 1725 break; 1726 case VM_EXITCODE_PAGING: 1727 error = vm_handle_paging(vm, vcpuid, &retu); 1728 break; 1729 case VM_EXITCODE_INST_EMUL: 1730 error = vm_handle_inst_emul(vm, vcpuid, &retu); 1731 break; 1732 case VM_EXITCODE_INOUT: 1733 case VM_EXITCODE_INOUT_STR: 1734 error = vm_handle_inout(vm, vcpuid, vme, &retu); 1735 break; 1736 case VM_EXITCODE_MONITOR: 1737 case VM_EXITCODE_MWAIT: 1738 vm_inject_ud(vm, vcpuid); 1739 break; 1740 default: 1741 retu = true; /* handled in userland */ 1742 break; 1743 } 1744 } 1745 1746 if (error == 0 && retu == false) 1747 goto restart; 1748 1749 VCPU_CTR2(vm, vcpuid, "retu %d/%d", error, vme->exitcode); 1750 1751 /* copy the exit information */ 1752 bcopy(vme, &vmrun->vm_exit, sizeof(struct vm_exit)); 1753 return (error); 1754 } 1755 1756 int 1757 vm_restart_instruction(void *arg, int vcpuid) 1758 { 1759 struct vm *vm; 1760 struct vcpu *vcpu; 1761 enum vcpu_state state; 1762 uint64_t rip; 1763 int error; 1764 1765 vm = arg; 1766 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1767 return (EINVAL); 1768 1769 vcpu = &vm->vcpu[vcpuid]; 1770 state = vcpu_get_state(vm, vcpuid, NULL); 1771 if (state == VCPU_RUNNING) { 1772 /* 1773 * When a vcpu is "running" the next instruction is determined 1774 * by adding 'rip' and 'inst_length' in the vcpu's 'exitinfo'. 1775 * Thus setting 'inst_length' to zero will cause the current 1776 * instruction to be restarted. 1777 */ 1778 vcpu->exitinfo.inst_length = 0; 1779 VCPU_CTR1(vm, vcpuid, "restarting instruction at %#lx by " 1780 "setting inst_length to zero", vcpu->exitinfo.rip); 1781 } else if (state == VCPU_FROZEN) { 1782 /* 1783 * When a vcpu is "frozen" it is outside the critical section 1784 * around VMRUN() and 'nextrip' points to the next instruction. 1785 * Thus instruction restart is achieved by setting 'nextrip' 1786 * to the vcpu's %rip. 1787 */ 1788 error = vm_get_register(vm, vcpuid, VM_REG_GUEST_RIP, &rip); 1789 KASSERT(!error, ("%s: error %d getting rip", __func__, error)); 1790 VCPU_CTR2(vm, vcpuid, "restarting instruction by updating " 1791 "nextrip from %#lx to %#lx", vcpu->nextrip, rip); 1792 vcpu->nextrip = rip; 1793 } else { 1794 panic("%s: invalid state %d", __func__, state); 1795 } 1796 return (0); 1797 } 1798 1799 int 1800 vm_exit_intinfo(struct vm *vm, int vcpuid, uint64_t info) 1801 { 1802 struct vcpu *vcpu; 1803 int type, vector; 1804 1805 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1806 return (EINVAL); 1807 1808 vcpu = &vm->vcpu[vcpuid]; 1809 1810 if (info & VM_INTINFO_VALID) { 1811 type = info & VM_INTINFO_TYPE; 1812 vector = info & 0xff; 1813 if (type == VM_INTINFO_NMI && vector != IDT_NMI) 1814 return (EINVAL); 1815 if (type == VM_INTINFO_HWEXCEPTION && vector >= 32) 1816 return (EINVAL); 1817 if (info & VM_INTINFO_RSVD) 1818 return (EINVAL); 1819 } else { 1820 info = 0; 1821 } 1822 VCPU_CTR2(vm, vcpuid, "%s: info1(%#lx)", __func__, info); 1823 vcpu->exitintinfo = info; 1824 return (0); 1825 } 1826 1827 enum exc_class { 1828 EXC_BENIGN, 1829 EXC_CONTRIBUTORY, 1830 EXC_PAGEFAULT 1831 }; 1832 1833 #define IDT_VE 20 /* Virtualization Exception (Intel specific) */ 1834 1835 static enum exc_class 1836 exception_class(uint64_t info) 1837 { 1838 int type, vector; 1839 1840 KASSERT(info & VM_INTINFO_VALID, ("intinfo must be valid: %#lx", info)); 1841 type = info & VM_INTINFO_TYPE; 1842 vector = info & 0xff; 1843 1844 /* Table 6-4, "Interrupt and Exception Classes", Intel SDM, Vol 3 */ 1845 switch (type) { 1846 case VM_INTINFO_HWINTR: 1847 case VM_INTINFO_SWINTR: 1848 case VM_INTINFO_NMI: 1849 return (EXC_BENIGN); 1850 default: 1851 /* 1852 * Hardware exception. 1853 * 1854 * SVM and VT-x use identical type values to represent NMI, 1855 * hardware interrupt and software interrupt. 1856 * 1857 * SVM uses type '3' for all exceptions. VT-x uses type '3' 1858 * for exceptions except #BP and #OF. #BP and #OF use a type 1859 * value of '5' or '6'. Therefore we don't check for explicit 1860 * values of 'type' to classify 'intinfo' into a hardware 1861 * exception. 1862 */ 1863 break; 1864 } 1865 1866 switch (vector) { 1867 case IDT_PF: 1868 case IDT_VE: 1869 return (EXC_PAGEFAULT); 1870 case IDT_DE: 1871 case IDT_TS: 1872 case IDT_NP: 1873 case IDT_SS: 1874 case IDT_GP: 1875 return (EXC_CONTRIBUTORY); 1876 default: 1877 return (EXC_BENIGN); 1878 } 1879 } 1880 1881 static int 1882 nested_fault(struct vm *vm, int vcpuid, uint64_t info1, uint64_t info2, 1883 uint64_t *retinfo) 1884 { 1885 enum exc_class exc1, exc2; 1886 int type1, vector1; 1887 1888 KASSERT(info1 & VM_INTINFO_VALID, ("info1 %#lx is not valid", info1)); 1889 KASSERT(info2 & VM_INTINFO_VALID, ("info2 %#lx is not valid", info2)); 1890 1891 /* 1892 * If an exception occurs while attempting to call the double-fault 1893 * handler the processor enters shutdown mode (aka triple fault). 1894 */ 1895 type1 = info1 & VM_INTINFO_TYPE; 1896 vector1 = info1 & 0xff; 1897 if (type1 == VM_INTINFO_HWEXCEPTION && vector1 == IDT_DF) { 1898 VCPU_CTR2(vm, vcpuid, "triple fault: info1(%#lx), info2(%#lx)", 1899 info1, info2); 1900 vm_suspend(vm, VM_SUSPEND_TRIPLEFAULT); 1901 *retinfo = 0; 1902 return (0); 1903 } 1904 1905 /* 1906 * Table 6-5 "Conditions for Generating a Double Fault", Intel SDM, Vol3 1907 */ 1908 exc1 = exception_class(info1); 1909 exc2 = exception_class(info2); 1910 if ((exc1 == EXC_CONTRIBUTORY && exc2 == EXC_CONTRIBUTORY) || 1911 (exc1 == EXC_PAGEFAULT && exc2 != EXC_BENIGN)) { 1912 /* Convert nested fault into a double fault. */ 1913 *retinfo = IDT_DF; 1914 *retinfo |= VM_INTINFO_VALID | VM_INTINFO_HWEXCEPTION; 1915 *retinfo |= VM_INTINFO_DEL_ERRCODE; 1916 } else { 1917 /* Handle exceptions serially */ 1918 *retinfo = info2; 1919 } 1920 return (1); 1921 } 1922 1923 static uint64_t 1924 vcpu_exception_intinfo(struct vcpu *vcpu) 1925 { 1926 uint64_t info = 0; 1927 1928 if (vcpu->exception_pending) { 1929 info = vcpu->exc_vector & 0xff; 1930 info |= VM_INTINFO_VALID | VM_INTINFO_HWEXCEPTION; 1931 if (vcpu->exc_errcode_valid) { 1932 info |= VM_INTINFO_DEL_ERRCODE; 1933 info |= (uint64_t)vcpu->exc_errcode << 32; 1934 } 1935 } 1936 return (info); 1937 } 1938 1939 int 1940 vm_entry_intinfo(struct vm *vm, int vcpuid, uint64_t *retinfo) 1941 { 1942 struct vcpu *vcpu; 1943 uint64_t info1, info2; 1944 int valid; 1945 1946 KASSERT(vcpuid >= 0 && vcpuid < VM_MAXCPU, ("invalid vcpu %d", vcpuid)); 1947 1948 vcpu = &vm->vcpu[vcpuid]; 1949 1950 info1 = vcpu->exitintinfo; 1951 vcpu->exitintinfo = 0; 1952 1953 info2 = 0; 1954 if (vcpu->exception_pending) { 1955 info2 = vcpu_exception_intinfo(vcpu); 1956 vcpu->exception_pending = 0; 1957 VCPU_CTR2(vm, vcpuid, "Exception %d delivered: %#lx", 1958 vcpu->exc_vector, info2); 1959 } 1960 1961 if ((info1 & VM_INTINFO_VALID) && (info2 & VM_INTINFO_VALID)) { 1962 valid = nested_fault(vm, vcpuid, info1, info2, retinfo); 1963 } else if (info1 & VM_INTINFO_VALID) { 1964 *retinfo = info1; 1965 valid = 1; 1966 } else if (info2 & VM_INTINFO_VALID) { 1967 *retinfo = info2; 1968 valid = 1; 1969 } else { 1970 valid = 0; 1971 } 1972 1973 if (valid) { 1974 VCPU_CTR4(vm, vcpuid, "%s: info1(%#lx), info2(%#lx), " 1975 "retinfo(%#lx)", __func__, info1, info2, *retinfo); 1976 } 1977 1978 return (valid); 1979 } 1980 1981 int 1982 vm_get_intinfo(struct vm *vm, int vcpuid, uint64_t *info1, uint64_t *info2) 1983 { 1984 struct vcpu *vcpu; 1985 1986 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1987 return (EINVAL); 1988 1989 vcpu = &vm->vcpu[vcpuid]; 1990 *info1 = vcpu->exitintinfo; 1991 *info2 = vcpu_exception_intinfo(vcpu); 1992 return (0); 1993 } 1994 1995 int 1996 vm_inject_exception(struct vm *vm, int vcpuid, int vector, int errcode_valid, 1997 uint32_t errcode, int restart_instruction) 1998 { 1999 struct vcpu *vcpu; 2000 uint64_t regval; 2001 int error; 2002 2003 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 2004 return (EINVAL); 2005 2006 if (vector < 0 || vector >= 32) 2007 return (EINVAL); 2008 2009 /* 2010 * A double fault exception should never be injected directly into 2011 * the guest. It is a derived exception that results from specific 2012 * combinations of nested faults. 2013 */ 2014 if (vector == IDT_DF) 2015 return (EINVAL); 2016 2017 vcpu = &vm->vcpu[vcpuid]; 2018 2019 if (vcpu->exception_pending) { 2020 VCPU_CTR2(vm, vcpuid, "Unable to inject exception %d due to " 2021 "pending exception %d", vector, vcpu->exc_vector); 2022 return (EBUSY); 2023 } 2024 2025 if (errcode_valid) { 2026 /* 2027 * Exceptions don't deliver an error code in real mode. 2028 */ 2029 error = vm_get_register(vm, vcpuid, VM_REG_GUEST_CR0, ®val); 2030 KASSERT(!error, ("%s: error %d getting CR0", __func__, error)); 2031 if (!(regval & CR0_PE)) 2032 errcode_valid = 0; 2033 } 2034 2035 /* 2036 * From section 26.6.1 "Interruptibility State" in Intel SDM: 2037 * 2038 * Event blocking by "STI" or "MOV SS" is cleared after guest executes 2039 * one instruction or incurs an exception. 2040 */ 2041 error = vm_set_register(vm, vcpuid, VM_REG_GUEST_INTR_SHADOW, 0); 2042 KASSERT(error == 0, ("%s: error %d clearing interrupt shadow", 2043 __func__, error)); 2044 2045 if (restart_instruction) 2046 vm_restart_instruction(vm, vcpuid); 2047 2048 vcpu->exception_pending = 1; 2049 vcpu->exc_vector = vector; 2050 vcpu->exc_errcode = errcode; 2051 vcpu->exc_errcode_valid = errcode_valid; 2052 VCPU_CTR1(vm, vcpuid, "Exception %d pending", vector); 2053 return (0); 2054 } 2055 2056 void 2057 vm_inject_fault(void *vmarg, int vcpuid, int vector, int errcode_valid, 2058 int errcode) 2059 { 2060 struct vm *vm; 2061 int error, restart_instruction; 2062 2063 vm = vmarg; 2064 restart_instruction = 1; 2065 2066 error = vm_inject_exception(vm, vcpuid, vector, errcode_valid, 2067 errcode, restart_instruction); 2068 KASSERT(error == 0, ("vm_inject_exception error %d", error)); 2069 } 2070 2071 void 2072 vm_inject_pf(void *vmarg, int vcpuid, int error_code, uint64_t cr2) 2073 { 2074 struct vm *vm; 2075 int error; 2076 2077 vm = vmarg; 2078 VCPU_CTR2(vm, vcpuid, "Injecting page fault: error_code %#x, cr2 %#lx", 2079 error_code, cr2); 2080 2081 error = vm_set_register(vm, vcpuid, VM_REG_GUEST_CR2, cr2); 2082 KASSERT(error == 0, ("vm_set_register(cr2) error %d", error)); 2083 2084 vm_inject_fault(vm, vcpuid, IDT_PF, 1, error_code); 2085 } 2086 2087 static VMM_STAT(VCPU_NMI_COUNT, "number of NMIs delivered to vcpu"); 2088 2089 int 2090 vm_inject_nmi(struct vm *vm, int vcpuid) 2091 { 2092 struct vcpu *vcpu; 2093 2094 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 2095 return (EINVAL); 2096 2097 vcpu = &vm->vcpu[vcpuid]; 2098 2099 vcpu->nmi_pending = 1; 2100 vcpu_notify_event(vm, vcpuid, false); 2101 return (0); 2102 } 2103 2104 int 2105 vm_nmi_pending(struct vm *vm, int vcpuid) 2106 { 2107 struct vcpu *vcpu; 2108 2109 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 2110 panic("vm_nmi_pending: invalid vcpuid %d", vcpuid); 2111 2112 vcpu = &vm->vcpu[vcpuid]; 2113 2114 return (vcpu->nmi_pending); 2115 } 2116 2117 void 2118 vm_nmi_clear(struct vm *vm, int vcpuid) 2119 { 2120 struct vcpu *vcpu; 2121 2122 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 2123 panic("vm_nmi_pending: invalid vcpuid %d", vcpuid); 2124 2125 vcpu = &vm->vcpu[vcpuid]; 2126 2127 if (vcpu->nmi_pending == 0) 2128 panic("vm_nmi_clear: inconsistent nmi_pending state"); 2129 2130 vcpu->nmi_pending = 0; 2131 vmm_stat_incr(vm, vcpuid, VCPU_NMI_COUNT, 1); 2132 } 2133 2134 static VMM_STAT(VCPU_EXTINT_COUNT, "number of ExtINTs delivered to vcpu"); 2135 2136 int 2137 vm_inject_extint(struct vm *vm, int vcpuid) 2138 { 2139 struct vcpu *vcpu; 2140 2141 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 2142 return (EINVAL); 2143 2144 vcpu = &vm->vcpu[vcpuid]; 2145 2146 vcpu->extint_pending = 1; 2147 vcpu_notify_event(vm, vcpuid, false); 2148 return (0); 2149 } 2150 2151 int 2152 vm_extint_pending(struct vm *vm, int vcpuid) 2153 { 2154 struct vcpu *vcpu; 2155 2156 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 2157 panic("vm_extint_pending: invalid vcpuid %d", vcpuid); 2158 2159 vcpu = &vm->vcpu[vcpuid]; 2160 2161 return (vcpu->extint_pending); 2162 } 2163 2164 void 2165 vm_extint_clear(struct vm *vm, int vcpuid) 2166 { 2167 struct vcpu *vcpu; 2168 2169 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 2170 panic("vm_extint_pending: invalid vcpuid %d", vcpuid); 2171 2172 vcpu = &vm->vcpu[vcpuid]; 2173 2174 if (vcpu->extint_pending == 0) 2175 panic("vm_extint_clear: inconsistent extint_pending state"); 2176 2177 vcpu->extint_pending = 0; 2178 vmm_stat_incr(vm, vcpuid, VCPU_EXTINT_COUNT, 1); 2179 } 2180 2181 int 2182 vm_get_capability(struct vm *vm, int vcpu, int type, int *retval) 2183 { 2184 if (vcpu < 0 || vcpu >= VM_MAXCPU) 2185 return (EINVAL); 2186 2187 if (type < 0 || type >= VM_CAP_MAX) 2188 return (EINVAL); 2189 2190 return (VMGETCAP(vm->cookie, vcpu, type, retval)); 2191 } 2192 2193 int 2194 vm_set_capability(struct vm *vm, int vcpu, int type, int val) 2195 { 2196 if (vcpu < 0 || vcpu >= VM_MAXCPU) 2197 return (EINVAL); 2198 2199 if (type < 0 || type >= VM_CAP_MAX) 2200 return (EINVAL); 2201 2202 return (VMSETCAP(vm->cookie, vcpu, type, val)); 2203 } 2204 2205 struct vlapic * 2206 vm_lapic(struct vm *vm, int cpu) 2207 { 2208 return (vm->vcpu[cpu].vlapic); 2209 } 2210 2211 struct vioapic * 2212 vm_ioapic(struct vm *vm) 2213 { 2214 2215 return (vm->vioapic); 2216 } 2217 2218 struct vhpet * 2219 vm_hpet(struct vm *vm) 2220 { 2221 2222 return (vm->vhpet); 2223 } 2224 2225 boolean_t 2226 vmm_is_pptdev(int bus, int slot, int func) 2227 { 2228 int found, i, n; 2229 int b, s, f; 2230 char *val, *cp, *cp2; 2231 2232 /* 2233 * XXX 2234 * The length of an environment variable is limited to 128 bytes which 2235 * puts an upper limit on the number of passthru devices that may be 2236 * specified using a single environment variable. 2237 * 2238 * Work around this by scanning multiple environment variable 2239 * names instead of a single one - yuck! 2240 */ 2241 const char *names[] = { "pptdevs", "pptdevs2", "pptdevs3", NULL }; 2242 2243 /* set pptdevs="1/2/3 4/5/6 7/8/9 10/11/12" */ 2244 found = 0; 2245 for (i = 0; names[i] != NULL && !found; i++) { 2246 cp = val = kern_getenv(names[i]); 2247 while (cp != NULL && *cp != '\0') { 2248 if ((cp2 = strchr(cp, ' ')) != NULL) 2249 *cp2 = '\0'; 2250 2251 n = sscanf(cp, "%d/%d/%d", &b, &s, &f); 2252 if (n == 3 && bus == b && slot == s && func == f) { 2253 found = 1; 2254 break; 2255 } 2256 2257 if (cp2 != NULL) 2258 *cp2++ = ' '; 2259 2260 cp = cp2; 2261 } 2262 freeenv(val); 2263 } 2264 return (found); 2265 } 2266 2267 void * 2268 vm_iommu_domain(struct vm *vm) 2269 { 2270 2271 return (vm->iommu); 2272 } 2273 2274 int 2275 vcpu_set_state(struct vm *vm, int vcpuid, enum vcpu_state newstate, 2276 bool from_idle) 2277 { 2278 int error; 2279 struct vcpu *vcpu; 2280 2281 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 2282 panic("vm_set_run_state: invalid vcpuid %d", vcpuid); 2283 2284 vcpu = &vm->vcpu[vcpuid]; 2285 2286 vcpu_lock(vcpu); 2287 error = vcpu_set_state_locked(vm, vcpuid, newstate, from_idle); 2288 vcpu_unlock(vcpu); 2289 2290 return (error); 2291 } 2292 2293 enum vcpu_state 2294 vcpu_get_state(struct vm *vm, int vcpuid, int *hostcpu) 2295 { 2296 struct vcpu *vcpu; 2297 enum vcpu_state state; 2298 2299 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 2300 panic("vm_get_run_state: invalid vcpuid %d", vcpuid); 2301 2302 vcpu = &vm->vcpu[vcpuid]; 2303 2304 vcpu_lock(vcpu); 2305 state = vcpu->state; 2306 if (hostcpu != NULL) 2307 *hostcpu = vcpu->hostcpu; 2308 vcpu_unlock(vcpu); 2309 2310 return (state); 2311 } 2312 2313 int 2314 vm_activate_cpu(struct vm *vm, int vcpuid) 2315 { 2316 2317 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 2318 return (EINVAL); 2319 2320 if (CPU_ISSET(vcpuid, &vm->active_cpus)) 2321 return (EBUSY); 2322 2323 VCPU_CTR0(vm, vcpuid, "activated"); 2324 CPU_SET_ATOMIC(vcpuid, &vm->active_cpus); 2325 return (0); 2326 } 2327 2328 int 2329 vm_suspend_cpu(struct vm *vm, int vcpuid) 2330 { 2331 int i; 2332 2333 if (vcpuid < -1 || vcpuid >= VM_MAXCPU) 2334 return (EINVAL); 2335 2336 if (vcpuid == -1) { 2337 vm->debug_cpus = vm->active_cpus; 2338 for (i = 0; i < VM_MAXCPU; i++) { 2339 if (CPU_ISSET(i, &vm->active_cpus)) 2340 vcpu_notify_event(vm, i, false); 2341 } 2342 } else { 2343 if (!CPU_ISSET(vcpuid, &vm->active_cpus)) 2344 return (EINVAL); 2345 2346 CPU_SET_ATOMIC(vcpuid, &vm->debug_cpus); 2347 vcpu_notify_event(vm, vcpuid, false); 2348 } 2349 return (0); 2350 } 2351 2352 int 2353 vm_resume_cpu(struct vm *vm, int vcpuid) 2354 { 2355 2356 if (vcpuid < -1 || vcpuid >= VM_MAXCPU) 2357 return (EINVAL); 2358 2359 if (vcpuid == -1) { 2360 CPU_ZERO(&vm->debug_cpus); 2361 } else { 2362 if (!CPU_ISSET(vcpuid, &vm->debug_cpus)) 2363 return (EINVAL); 2364 2365 CPU_CLR_ATOMIC(vcpuid, &vm->debug_cpus); 2366 } 2367 return (0); 2368 } 2369 2370 int 2371 vcpu_debugged(struct vm *vm, int vcpuid) 2372 { 2373 2374 return (CPU_ISSET(vcpuid, &vm->debug_cpus)); 2375 } 2376 2377 cpuset_t 2378 vm_active_cpus(struct vm *vm) 2379 { 2380 2381 return (vm->active_cpus); 2382 } 2383 2384 cpuset_t 2385 vm_debug_cpus(struct vm *vm) 2386 { 2387 2388 return (vm->debug_cpus); 2389 } 2390 2391 cpuset_t 2392 vm_suspended_cpus(struct vm *vm) 2393 { 2394 2395 return (vm->suspended_cpus); 2396 } 2397 2398 void * 2399 vcpu_stats(struct vm *vm, int vcpuid) 2400 { 2401 2402 return (vm->vcpu[vcpuid].stats); 2403 } 2404 2405 int 2406 vm_get_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state *state) 2407 { 2408 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 2409 return (EINVAL); 2410 2411 *state = vm->vcpu[vcpuid].x2apic_state; 2412 2413 return (0); 2414 } 2415 2416 int 2417 vm_set_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state state) 2418 { 2419 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 2420 return (EINVAL); 2421 2422 if (state >= X2APIC_STATE_LAST) 2423 return (EINVAL); 2424 2425 vm->vcpu[vcpuid].x2apic_state = state; 2426 2427 vlapic_set_x2apic_state(vm, vcpuid, state); 2428 2429 return (0); 2430 } 2431 2432 /* 2433 * This function is called to ensure that a vcpu "sees" a pending event 2434 * as soon as possible: 2435 * - If the vcpu thread is sleeping then it is woken up. 2436 * - If the vcpu is running on a different host_cpu then an IPI will be directed 2437 * to the host_cpu to cause the vcpu to trap into the hypervisor. 2438 */ 2439 static void 2440 vcpu_notify_event_locked(struct vcpu *vcpu, bool lapic_intr) 2441 { 2442 int hostcpu; 2443 2444 hostcpu = vcpu->hostcpu; 2445 if (vcpu->state == VCPU_RUNNING) { 2446 KASSERT(hostcpu != NOCPU, ("vcpu running on invalid hostcpu")); 2447 if (hostcpu != curcpu) { 2448 if (lapic_intr) { 2449 vlapic_post_intr(vcpu->vlapic, hostcpu, 2450 vmm_ipinum); 2451 } else { 2452 ipi_cpu(hostcpu, vmm_ipinum); 2453 } 2454 } else { 2455 /* 2456 * If the 'vcpu' is running on 'curcpu' then it must 2457 * be sending a notification to itself (e.g. SELF_IPI). 2458 * The pending event will be picked up when the vcpu 2459 * transitions back to guest context. 2460 */ 2461 } 2462 } else { 2463 KASSERT(hostcpu == NOCPU, ("vcpu state %d not consistent " 2464 "with hostcpu %d", vcpu->state, hostcpu)); 2465 if (vcpu->state == VCPU_SLEEPING) 2466 wakeup_one(vcpu); 2467 } 2468 } 2469 2470 void 2471 vcpu_notify_event(struct vm *vm, int vcpuid, bool lapic_intr) 2472 { 2473 struct vcpu *vcpu = &vm->vcpu[vcpuid]; 2474 2475 vcpu_lock(vcpu); 2476 vcpu_notify_event_locked(vcpu, lapic_intr); 2477 vcpu_unlock(vcpu); 2478 } 2479 2480 struct vmspace * 2481 vm_get_vmspace(struct vm *vm) 2482 { 2483 2484 return (vm->vmspace); 2485 } 2486 2487 int 2488 vm_apicid2vcpuid(struct vm *vm, int apicid) 2489 { 2490 /* 2491 * XXX apic id is assumed to be numerically identical to vcpu id 2492 */ 2493 return (apicid); 2494 } 2495 2496 void 2497 vm_smp_rendezvous(struct vm *vm, int vcpuid, cpuset_t dest, 2498 vm_rendezvous_func_t func, void *arg) 2499 { 2500 int i; 2501 2502 /* 2503 * Enforce that this function is called without any locks 2504 */ 2505 WITNESS_WARN(WARN_PANIC, NULL, "vm_smp_rendezvous"); 2506 KASSERT(vcpuid == -1 || (vcpuid >= 0 && vcpuid < VM_MAXCPU), 2507 ("vm_smp_rendezvous: invalid vcpuid %d", vcpuid)); 2508 2509 restart: 2510 mtx_lock(&vm->rendezvous_mtx); 2511 if (vm->rendezvous_func != NULL) { 2512 /* 2513 * If a rendezvous is already in progress then we need to 2514 * call the rendezvous handler in case this 'vcpuid' is one 2515 * of the targets of the rendezvous. 2516 */ 2517 RENDEZVOUS_CTR0(vm, vcpuid, "Rendezvous already in progress"); 2518 mtx_unlock(&vm->rendezvous_mtx); 2519 vm_handle_rendezvous(vm, vcpuid); 2520 goto restart; 2521 } 2522 KASSERT(vm->rendezvous_func == NULL, ("vm_smp_rendezvous: previous " 2523 "rendezvous is still in progress")); 2524 2525 RENDEZVOUS_CTR0(vm, vcpuid, "Initiating rendezvous"); 2526 vm->rendezvous_req_cpus = dest; 2527 CPU_ZERO(&vm->rendezvous_done_cpus); 2528 vm->rendezvous_arg = arg; 2529 vm_set_rendezvous_func(vm, func); 2530 mtx_unlock(&vm->rendezvous_mtx); 2531 2532 /* 2533 * Wake up any sleeping vcpus and trigger a VM-exit in any running 2534 * vcpus so they handle the rendezvous as soon as possible. 2535 */ 2536 for (i = 0; i < VM_MAXCPU; i++) { 2537 if (CPU_ISSET(i, &dest)) 2538 vcpu_notify_event(vm, i, false); 2539 } 2540 2541 vm_handle_rendezvous(vm, vcpuid); 2542 } 2543 2544 struct vatpic * 2545 vm_atpic(struct vm *vm) 2546 { 2547 return (vm->vatpic); 2548 } 2549 2550 struct vatpit * 2551 vm_atpit(struct vm *vm) 2552 { 2553 return (vm->vatpit); 2554 } 2555 2556 struct vpmtmr * 2557 vm_pmtmr(struct vm *vm) 2558 { 2559 2560 return (vm->vpmtmr); 2561 } 2562 2563 struct vrtc * 2564 vm_rtc(struct vm *vm) 2565 { 2566 2567 return (vm->vrtc); 2568 } 2569 2570 enum vm_reg_name 2571 vm_segment_name(int seg) 2572 { 2573 static enum vm_reg_name seg_names[] = { 2574 VM_REG_GUEST_ES, 2575 VM_REG_GUEST_CS, 2576 VM_REG_GUEST_SS, 2577 VM_REG_GUEST_DS, 2578 VM_REG_GUEST_FS, 2579 VM_REG_GUEST_GS 2580 }; 2581 2582 KASSERT(seg >= 0 && seg < nitems(seg_names), 2583 ("%s: invalid segment encoding %d", __func__, seg)); 2584 return (seg_names[seg]); 2585 } 2586 2587 void 2588 vm_copy_teardown(struct vm *vm, int vcpuid, struct vm_copyinfo *copyinfo, 2589 int num_copyinfo) 2590 { 2591 int idx; 2592 2593 for (idx = 0; idx < num_copyinfo; idx++) { 2594 if (copyinfo[idx].cookie != NULL) 2595 vm_gpa_release(copyinfo[idx].cookie); 2596 } 2597 bzero(copyinfo, num_copyinfo * sizeof(struct vm_copyinfo)); 2598 } 2599 2600 int 2601 vm_copy_setup(struct vm *vm, int vcpuid, struct vm_guest_paging *paging, 2602 uint64_t gla, size_t len, int prot, struct vm_copyinfo *copyinfo, 2603 int num_copyinfo, int *fault) 2604 { 2605 int error, idx, nused; 2606 size_t n, off, remaining; 2607 void *hva, *cookie; 2608 uint64_t gpa; 2609 2610 bzero(copyinfo, sizeof(struct vm_copyinfo) * num_copyinfo); 2611 2612 nused = 0; 2613 remaining = len; 2614 while (remaining > 0) { 2615 KASSERT(nused < num_copyinfo, ("insufficient vm_copyinfo")); 2616 error = vm_gla2gpa(vm, vcpuid, paging, gla, prot, &gpa, fault); 2617 if (error || *fault) 2618 return (error); 2619 off = gpa & PAGE_MASK; 2620 n = min(remaining, PAGE_SIZE - off); 2621 copyinfo[nused].gpa = gpa; 2622 copyinfo[nused].len = n; 2623 remaining -= n; 2624 gla += n; 2625 nused++; 2626 } 2627 2628 for (idx = 0; idx < nused; idx++) { 2629 hva = vm_gpa_hold(vm, vcpuid, copyinfo[idx].gpa, 2630 copyinfo[idx].len, prot, &cookie); 2631 if (hva == NULL) 2632 break; 2633 copyinfo[idx].hva = hva; 2634 copyinfo[idx].cookie = cookie; 2635 } 2636 2637 if (idx != nused) { 2638 vm_copy_teardown(vm, vcpuid, copyinfo, num_copyinfo); 2639 return (EFAULT); 2640 } else { 2641 *fault = 0; 2642 return (0); 2643 } 2644 } 2645 2646 void 2647 vm_copyin(struct vm *vm, int vcpuid, struct vm_copyinfo *copyinfo, void *kaddr, 2648 size_t len) 2649 { 2650 char *dst; 2651 int idx; 2652 2653 dst = kaddr; 2654 idx = 0; 2655 while (len > 0) { 2656 bcopy(copyinfo[idx].hva, dst, copyinfo[idx].len); 2657 len -= copyinfo[idx].len; 2658 dst += copyinfo[idx].len; 2659 idx++; 2660 } 2661 } 2662 2663 void 2664 vm_copyout(struct vm *vm, int vcpuid, const void *kaddr, 2665 struct vm_copyinfo *copyinfo, size_t len) 2666 { 2667 const char *src; 2668 int idx; 2669 2670 src = kaddr; 2671 idx = 0; 2672 while (len > 0) { 2673 bcopy(src, copyinfo[idx].hva, copyinfo[idx].len); 2674 len -= copyinfo[idx].len; 2675 src += copyinfo[idx].len; 2676 idx++; 2677 } 2678 } 2679 2680 /* 2681 * Return the amount of in-use and wired memory for the VM. Since 2682 * these are global stats, only return the values with for vCPU 0 2683 */ 2684 VMM_STAT_DECLARE(VMM_MEM_RESIDENT); 2685 VMM_STAT_DECLARE(VMM_MEM_WIRED); 2686 2687 static void 2688 vm_get_rescnt(struct vm *vm, int vcpu, struct vmm_stat_type *stat) 2689 { 2690 2691 if (vcpu == 0) { 2692 vmm_stat_set(vm, vcpu, VMM_MEM_RESIDENT, 2693 PAGE_SIZE * vmspace_resident_count(vm->vmspace)); 2694 } 2695 } 2696 2697 static void 2698 vm_get_wiredcnt(struct vm *vm, int vcpu, struct vmm_stat_type *stat) 2699 { 2700 2701 if (vcpu == 0) { 2702 vmm_stat_set(vm, vcpu, VMM_MEM_WIRED, 2703 PAGE_SIZE * pmap_wired_count(vmspace_pmap(vm->vmspace))); 2704 } 2705 } 2706 2707 VMM_STAT_FUNC(VMM_MEM_RESIDENT, "Resident memory", vm_get_rescnt); 2708 VMM_STAT_FUNC(VMM_MEM_WIRED, "Wired memory", vm_get_wiredcnt); 2709