1 /*- 2 * Copyright (c) 2011 NetApp, Inc. 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 14 * THIS SOFTWARE IS PROVIDED BY NETAPP, INC ``AS IS'' AND 15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 17 * ARE DISCLAIMED. IN NO EVENT SHALL NETAPP, INC OR CONTRIBUTORS BE LIABLE 18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 24 * SUCH DAMAGE. 25 * 26 * $FreeBSD$ 27 */ 28 29 #include <sys/cdefs.h> 30 __FBSDID("$FreeBSD$"); 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/systm.h> 46 47 #include <vm/vm.h> 48 #include <vm/vm_object.h> 49 #include <vm/vm_page.h> 50 #include <vm/pmap.h> 51 #include <vm/vm_map.h> 52 #include <vm/vm_extern.h> 53 #include <vm/vm_param.h> 54 55 #include <machine/cpu.h> 56 #include <machine/vm.h> 57 #include <machine/pcb.h> 58 #include <machine/smp.h> 59 #include <x86/psl.h> 60 #include <x86/apicreg.h> 61 #include <machine/vmparam.h> 62 63 #include <machine/vmm.h> 64 #include <machine/vmm_dev.h> 65 #include <machine/vmm_instruction_emul.h> 66 67 #include "vmm_ioport.h" 68 #include "vmm_ktr.h" 69 #include "vmm_host.h" 70 #include "vmm_mem.h" 71 #include "vmm_util.h" 72 #include "vatpic.h" 73 #include "vatpit.h" 74 #include "vhpet.h" 75 #include "vioapic.h" 76 #include "vlapic.h" 77 #include "vpmtmr.h" 78 #include "vrtc.h" 79 #include "vmm_stat.h" 80 #include "vmm_lapic.h" 81 82 #include "io/ppt.h" 83 #include "io/iommu.h" 84 85 struct vlapic; 86 87 /* 88 * Initialization: 89 * (a) allocated when vcpu is created 90 * (i) initialized when vcpu is created and when it is reinitialized 91 * (o) initialized the first time the vcpu is created 92 * (x) initialized before use 93 */ 94 struct vcpu { 95 struct mtx mtx; /* (o) protects 'state' and 'hostcpu' */ 96 enum vcpu_state state; /* (o) vcpu state */ 97 int hostcpu; /* (o) vcpu's host cpu */ 98 struct vlapic *vlapic; /* (i) APIC device model */ 99 enum x2apic_state x2apic_state; /* (i) APIC mode */ 100 uint64_t exitintinfo; /* (i) events pending at VM exit */ 101 int nmi_pending; /* (i) NMI pending */ 102 int extint_pending; /* (i) INTR pending */ 103 int exception_pending; /* (i) exception pending */ 104 int exc_vector; /* (x) exception collateral */ 105 int exc_errcode_valid; 106 uint32_t exc_errcode; 107 struct savefpu *guestfpu; /* (a,i) guest fpu state */ 108 uint64_t guest_xcr0; /* (i) guest %xcr0 register */ 109 void *stats; /* (a,i) statistics */ 110 struct vm_exit exitinfo; /* (x) exit reason and collateral */ 111 uint64_t nextrip; /* (x) next instruction to execute */ 112 }; 113 114 #define vcpu_lock_initialized(v) mtx_initialized(&((v)->mtx)) 115 #define vcpu_lock_init(v) mtx_init(&((v)->mtx), "vcpu lock", 0, MTX_SPIN) 116 #define vcpu_lock(v) mtx_lock_spin(&((v)->mtx)) 117 #define vcpu_unlock(v) mtx_unlock_spin(&((v)->mtx)) 118 #define vcpu_assert_locked(v) mtx_assert(&((v)->mtx), MA_OWNED) 119 120 struct mem_seg { 121 vm_paddr_t gpa; 122 size_t len; 123 boolean_t wired; 124 vm_object_t object; 125 }; 126 #define VM_MAX_MEMORY_SEGMENTS 2 127 128 /* 129 * Initialization: 130 * (o) initialized the first time the VM is created 131 * (i) initialized when VM is created and when it is reinitialized 132 * (x) initialized before use 133 */ 134 struct vm { 135 void *cookie; /* (i) cpu-specific data */ 136 void *iommu; /* (x) iommu-specific data */ 137 struct vhpet *vhpet; /* (i) virtual HPET */ 138 struct vioapic *vioapic; /* (i) virtual ioapic */ 139 struct vatpic *vatpic; /* (i) virtual atpic */ 140 struct vatpit *vatpit; /* (i) virtual atpit */ 141 struct vpmtmr *vpmtmr; /* (i) virtual ACPI PM timer */ 142 struct vrtc *vrtc; /* (o) virtual RTC */ 143 volatile cpuset_t active_cpus; /* (i) active vcpus */ 144 int suspend; /* (i) stop VM execution */ 145 volatile cpuset_t suspended_cpus; /* (i) suspended vcpus */ 146 volatile cpuset_t halted_cpus; /* (x) cpus in a hard halt */ 147 cpuset_t rendezvous_req_cpus; /* (x) rendezvous requested */ 148 cpuset_t rendezvous_done_cpus; /* (x) rendezvous finished */ 149 void *rendezvous_arg; /* (x) rendezvous func/arg */ 150 vm_rendezvous_func_t rendezvous_func; 151 struct mtx rendezvous_mtx; /* (o) rendezvous lock */ 152 int num_mem_segs; /* (o) guest memory segments */ 153 struct mem_seg mem_segs[VM_MAX_MEMORY_SEGMENTS]; 154 struct vmspace *vmspace; /* (o) guest's address space */ 155 char name[VM_MAX_NAMELEN]; /* (o) virtual machine name */ 156 struct vcpu vcpu[VM_MAXCPU]; /* (i) guest vcpus */ 157 }; 158 159 static int vmm_initialized; 160 161 static struct vmm_ops *ops; 162 #define VMM_INIT(num) (ops != NULL ? (*ops->init)(num) : 0) 163 #define VMM_CLEANUP() (ops != NULL ? (*ops->cleanup)() : 0) 164 #define VMM_RESUME() (ops != NULL ? (*ops->resume)() : 0) 165 166 #define VMINIT(vm, pmap) (ops != NULL ? (*ops->vminit)(vm, pmap): NULL) 167 #define VMRUN(vmi, vcpu, rip, pmap, rptr, sptr) \ 168 (ops != NULL ? (*ops->vmrun)(vmi, vcpu, rip, pmap, rptr, sptr) : ENXIO) 169 #define VMCLEANUP(vmi) (ops != NULL ? (*ops->vmcleanup)(vmi) : NULL) 170 #define VMSPACE_ALLOC(min, max) \ 171 (ops != NULL ? (*ops->vmspace_alloc)(min, max) : NULL) 172 #define VMSPACE_FREE(vmspace) \ 173 (ops != NULL ? (*ops->vmspace_free)(vmspace) : ENXIO) 174 #define VMGETREG(vmi, vcpu, num, retval) \ 175 (ops != NULL ? (*ops->vmgetreg)(vmi, vcpu, num, retval) : ENXIO) 176 #define VMSETREG(vmi, vcpu, num, val) \ 177 (ops != NULL ? (*ops->vmsetreg)(vmi, vcpu, num, val) : ENXIO) 178 #define VMGETDESC(vmi, vcpu, num, desc) \ 179 (ops != NULL ? (*ops->vmgetdesc)(vmi, vcpu, num, desc) : ENXIO) 180 #define VMSETDESC(vmi, vcpu, num, desc) \ 181 (ops != NULL ? (*ops->vmsetdesc)(vmi, vcpu, num, desc) : ENXIO) 182 #define VMGETCAP(vmi, vcpu, num, retval) \ 183 (ops != NULL ? (*ops->vmgetcap)(vmi, vcpu, num, retval) : ENXIO) 184 #define VMSETCAP(vmi, vcpu, num, val) \ 185 (ops != NULL ? (*ops->vmsetcap)(vmi, vcpu, num, val) : ENXIO) 186 #define VLAPIC_INIT(vmi, vcpu) \ 187 (ops != NULL ? (*ops->vlapic_init)(vmi, vcpu) : NULL) 188 #define VLAPIC_CLEANUP(vmi, vlapic) \ 189 (ops != NULL ? (*ops->vlapic_cleanup)(vmi, vlapic) : NULL) 190 191 #define fpu_start_emulating() load_cr0(rcr0() | CR0_TS) 192 #define fpu_stop_emulating() clts() 193 194 static MALLOC_DEFINE(M_VM, "vm", "vm"); 195 196 /* statistics */ 197 static VMM_STAT(VCPU_TOTAL_RUNTIME, "vcpu total runtime"); 198 199 SYSCTL_NODE(_hw, OID_AUTO, vmm, CTLFLAG_RW, NULL, NULL); 200 201 /* 202 * Halt the guest if all vcpus are executing a HLT instruction with 203 * interrupts disabled. 204 */ 205 static int halt_detection_enabled = 1; 206 SYSCTL_INT(_hw_vmm, OID_AUTO, halt_detection, CTLFLAG_RDTUN, 207 &halt_detection_enabled, 0, 208 "Halt VM if all vcpus execute HLT with interrupts disabled"); 209 210 static int vmm_ipinum; 211 SYSCTL_INT(_hw_vmm, OID_AUTO, ipinum, CTLFLAG_RD, &vmm_ipinum, 0, 212 "IPI vector used for vcpu notifications"); 213 214 static int trace_guest_exceptions; 215 SYSCTL_INT(_hw_vmm, OID_AUTO, trace_guest_exceptions, CTLFLAG_RDTUN, 216 &trace_guest_exceptions, 0, 217 "Trap into hypervisor on all guest exceptions and reflect them back"); 218 219 static int vmm_force_iommu = 0; 220 TUNABLE_INT("hw.vmm.force_iommu", &vmm_force_iommu); 221 SYSCTL_INT(_hw_vmm, OID_AUTO, force_iommu, CTLFLAG_RDTUN, &vmm_force_iommu, 0, 222 "Force use of I/O MMU even if no passthrough devices were found."); 223 224 static void 225 vcpu_cleanup(struct vm *vm, int i, bool destroy) 226 { 227 struct vcpu *vcpu = &vm->vcpu[i]; 228 229 VLAPIC_CLEANUP(vm->cookie, vcpu->vlapic); 230 if (destroy) { 231 vmm_stat_free(vcpu->stats); 232 fpu_save_area_free(vcpu->guestfpu); 233 } 234 } 235 236 static void 237 vcpu_init(struct vm *vm, int vcpu_id, bool create) 238 { 239 struct vcpu *vcpu; 240 241 KASSERT(vcpu_id >= 0 && vcpu_id < VM_MAXCPU, 242 ("vcpu_init: invalid vcpu %d", vcpu_id)); 243 244 vcpu = &vm->vcpu[vcpu_id]; 245 246 if (create) { 247 KASSERT(!vcpu_lock_initialized(vcpu), ("vcpu %d already " 248 "initialized", vcpu_id)); 249 vcpu_lock_init(vcpu); 250 vcpu->state = VCPU_IDLE; 251 vcpu->hostcpu = NOCPU; 252 vcpu->guestfpu = fpu_save_area_alloc(); 253 vcpu->stats = vmm_stat_alloc(); 254 } 255 256 vcpu->vlapic = VLAPIC_INIT(vm->cookie, vcpu_id); 257 vm_set_x2apic_state(vm, vcpu_id, X2APIC_DISABLED); 258 vcpu->exitintinfo = 0; 259 vcpu->nmi_pending = 0; 260 vcpu->extint_pending = 0; 261 vcpu->exception_pending = 0; 262 vcpu->guest_xcr0 = XFEATURE_ENABLED_X87; 263 fpu_save_area_reset(vcpu->guestfpu); 264 vmm_stat_init(vcpu->stats); 265 } 266 267 int 268 vcpu_trace_exceptions(struct vm *vm, int vcpuid) 269 { 270 271 return (trace_guest_exceptions); 272 } 273 274 struct vm_exit * 275 vm_exitinfo(struct vm *vm, int cpuid) 276 { 277 struct vcpu *vcpu; 278 279 if (cpuid < 0 || cpuid >= VM_MAXCPU) 280 panic("vm_exitinfo: invalid cpuid %d", cpuid); 281 282 vcpu = &vm->vcpu[cpuid]; 283 284 return (&vcpu->exitinfo); 285 } 286 287 static void 288 vmm_resume(void) 289 { 290 VMM_RESUME(); 291 } 292 293 static int 294 vmm_init(void) 295 { 296 int error; 297 298 vmm_host_state_init(); 299 300 vmm_ipinum = lapic_ipi_alloc(&IDTVEC(justreturn)); 301 if (vmm_ipinum < 0) 302 vmm_ipinum = IPI_AST; 303 304 error = vmm_mem_init(); 305 if (error) 306 return (error); 307 308 if (vmm_is_intel()) 309 ops = &vmm_ops_intel; 310 else if (vmm_is_amd()) 311 ops = &vmm_ops_amd; 312 else 313 return (ENXIO); 314 315 vmm_resume_p = vmm_resume; 316 317 return (VMM_INIT(vmm_ipinum)); 318 } 319 320 static int 321 vmm_handler(module_t mod, int what, void *arg) 322 { 323 int error; 324 325 switch (what) { 326 case MOD_LOAD: 327 vmmdev_init(); 328 if (vmm_force_iommu || ppt_avail_devices() > 0) 329 iommu_init(); 330 error = vmm_init(); 331 if (error == 0) 332 vmm_initialized = 1; 333 break; 334 case MOD_UNLOAD: 335 error = vmmdev_cleanup(); 336 if (error == 0) { 337 vmm_resume_p = NULL; 338 iommu_cleanup(); 339 if (vmm_ipinum != IPI_AST) 340 lapic_ipi_free(vmm_ipinum); 341 error = VMM_CLEANUP(); 342 /* 343 * Something bad happened - prevent new 344 * VMs from being created 345 */ 346 if (error) 347 vmm_initialized = 0; 348 } 349 break; 350 default: 351 error = 0; 352 break; 353 } 354 return (error); 355 } 356 357 static moduledata_t vmm_kmod = { 358 "vmm", 359 vmm_handler, 360 NULL 361 }; 362 363 /* 364 * vmm initialization has the following dependencies: 365 * 366 * - iommu initialization must happen after the pci passthru driver has had 367 * a chance to attach to any passthru devices (after SI_SUB_CONFIGURE). 368 * 369 * - VT-x initialization requires smp_rendezvous() and therefore must happen 370 * after SMP is fully functional (after SI_SUB_SMP). 371 */ 372 DECLARE_MODULE(vmm, vmm_kmod, SI_SUB_SMP + 1, SI_ORDER_ANY); 373 MODULE_VERSION(vmm, 1); 374 375 static void 376 vm_init(struct vm *vm, bool create) 377 { 378 int i; 379 380 vm->cookie = VMINIT(vm, vmspace_pmap(vm->vmspace)); 381 vm->iommu = NULL; 382 vm->vioapic = vioapic_init(vm); 383 vm->vhpet = vhpet_init(vm); 384 vm->vatpic = vatpic_init(vm); 385 vm->vatpit = vatpit_init(vm); 386 vm->vpmtmr = vpmtmr_init(vm); 387 if (create) 388 vm->vrtc = vrtc_init(vm); 389 390 CPU_ZERO(&vm->active_cpus); 391 392 vm->suspend = 0; 393 CPU_ZERO(&vm->suspended_cpus); 394 395 for (i = 0; i < VM_MAXCPU; i++) 396 vcpu_init(vm, i, create); 397 } 398 399 int 400 vm_create(const char *name, struct vm **retvm) 401 { 402 struct vm *vm; 403 struct vmspace *vmspace; 404 405 /* 406 * If vmm.ko could not be successfully initialized then don't attempt 407 * to create the virtual machine. 408 */ 409 if (!vmm_initialized) 410 return (ENXIO); 411 412 if (name == NULL || strlen(name) >= VM_MAX_NAMELEN) 413 return (EINVAL); 414 415 vmspace = VMSPACE_ALLOC(0, VM_MAXUSER_ADDRESS); 416 if (vmspace == NULL) 417 return (ENOMEM); 418 419 vm = malloc(sizeof(struct vm), M_VM, M_WAITOK | M_ZERO); 420 strcpy(vm->name, name); 421 vm->num_mem_segs = 0; 422 vm->vmspace = vmspace; 423 mtx_init(&vm->rendezvous_mtx, "vm rendezvous lock", 0, MTX_DEF); 424 425 vm_init(vm, true); 426 427 *retvm = vm; 428 return (0); 429 } 430 431 static void 432 vm_free_mem_seg(struct vm *vm, struct mem_seg *seg) 433 { 434 435 if (seg->object != NULL) 436 vmm_mem_free(vm->vmspace, seg->gpa, seg->len); 437 438 bzero(seg, sizeof(*seg)); 439 } 440 441 static void 442 vm_cleanup(struct vm *vm, bool destroy) 443 { 444 int i; 445 446 ppt_unassign_all(vm); 447 448 if (vm->iommu != NULL) 449 iommu_destroy_domain(vm->iommu); 450 451 if (destroy) 452 vrtc_cleanup(vm->vrtc); 453 else 454 vrtc_reset(vm->vrtc); 455 vpmtmr_cleanup(vm->vpmtmr); 456 vatpit_cleanup(vm->vatpit); 457 vhpet_cleanup(vm->vhpet); 458 vatpic_cleanup(vm->vatpic); 459 vioapic_cleanup(vm->vioapic); 460 461 for (i = 0; i < VM_MAXCPU; i++) 462 vcpu_cleanup(vm, i, destroy); 463 464 VMCLEANUP(vm->cookie); 465 466 if (destroy) { 467 for (i = 0; i < vm->num_mem_segs; i++) 468 vm_free_mem_seg(vm, &vm->mem_segs[i]); 469 470 vm->num_mem_segs = 0; 471 472 VMSPACE_FREE(vm->vmspace); 473 vm->vmspace = NULL; 474 } 475 } 476 477 void 478 vm_destroy(struct vm *vm) 479 { 480 vm_cleanup(vm, true); 481 free(vm, M_VM); 482 } 483 484 int 485 vm_reinit(struct vm *vm) 486 { 487 int error; 488 489 /* 490 * A virtual machine can be reset only if all vcpus are suspended. 491 */ 492 if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) { 493 vm_cleanup(vm, false); 494 vm_init(vm, false); 495 error = 0; 496 } else { 497 error = EBUSY; 498 } 499 500 return (error); 501 } 502 503 const char * 504 vm_name(struct vm *vm) 505 { 506 return (vm->name); 507 } 508 509 int 510 vm_map_mmio(struct vm *vm, vm_paddr_t gpa, size_t len, vm_paddr_t hpa) 511 { 512 vm_object_t obj; 513 514 if ((obj = vmm_mmio_alloc(vm->vmspace, gpa, len, hpa)) == NULL) 515 return (ENOMEM); 516 else 517 return (0); 518 } 519 520 int 521 vm_unmap_mmio(struct vm *vm, vm_paddr_t gpa, size_t len) 522 { 523 524 vmm_mmio_free(vm->vmspace, gpa, len); 525 return (0); 526 } 527 528 boolean_t 529 vm_mem_allocated(struct vm *vm, vm_paddr_t gpa) 530 { 531 int i; 532 vm_paddr_t gpabase, gpalimit; 533 534 for (i = 0; i < vm->num_mem_segs; i++) { 535 gpabase = vm->mem_segs[i].gpa; 536 gpalimit = gpabase + vm->mem_segs[i].len; 537 if (gpa >= gpabase && gpa < gpalimit) 538 return (TRUE); /* 'gpa' is regular memory */ 539 } 540 541 if (ppt_is_mmio(vm, gpa)) 542 return (TRUE); /* 'gpa' is pci passthru mmio */ 543 544 return (FALSE); 545 } 546 547 int 548 vm_malloc(struct vm *vm, vm_paddr_t gpa, size_t len) 549 { 550 int available, allocated; 551 struct mem_seg *seg; 552 vm_object_t object; 553 vm_paddr_t g; 554 555 if ((gpa & PAGE_MASK) || (len & PAGE_MASK) || len == 0) 556 return (EINVAL); 557 558 available = allocated = 0; 559 g = gpa; 560 while (g < gpa + len) { 561 if (vm_mem_allocated(vm, g)) 562 allocated++; 563 else 564 available++; 565 566 g += PAGE_SIZE; 567 } 568 569 /* 570 * If there are some allocated and some available pages in the address 571 * range then it is an error. 572 */ 573 if (allocated && available) 574 return (EINVAL); 575 576 /* 577 * If the entire address range being requested has already been 578 * allocated then there isn't anything more to do. 579 */ 580 if (allocated && available == 0) 581 return (0); 582 583 if (vm->num_mem_segs >= VM_MAX_MEMORY_SEGMENTS) 584 return (E2BIG); 585 586 seg = &vm->mem_segs[vm->num_mem_segs]; 587 588 if ((object = vmm_mem_alloc(vm->vmspace, gpa, len)) == NULL) 589 return (ENOMEM); 590 591 seg->gpa = gpa; 592 seg->len = len; 593 seg->object = object; 594 seg->wired = FALSE; 595 596 vm->num_mem_segs++; 597 598 return (0); 599 } 600 601 static vm_paddr_t 602 vm_maxmem(struct vm *vm) 603 { 604 int i; 605 vm_paddr_t gpa, maxmem; 606 607 maxmem = 0; 608 for (i = 0; i < vm->num_mem_segs; i++) { 609 gpa = vm->mem_segs[i].gpa + vm->mem_segs[i].len; 610 if (gpa > maxmem) 611 maxmem = gpa; 612 } 613 return (maxmem); 614 } 615 616 static void 617 vm_gpa_unwire(struct vm *vm) 618 { 619 int i, rv; 620 struct mem_seg *seg; 621 622 for (i = 0; i < vm->num_mem_segs; i++) { 623 seg = &vm->mem_segs[i]; 624 if (!seg->wired) 625 continue; 626 627 rv = vm_map_unwire(&vm->vmspace->vm_map, 628 seg->gpa, seg->gpa + seg->len, 629 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES); 630 KASSERT(rv == KERN_SUCCESS, ("vm(%s) memory segment " 631 "%#lx/%ld could not be unwired: %d", 632 vm_name(vm), seg->gpa, seg->len, rv)); 633 634 seg->wired = FALSE; 635 } 636 } 637 638 static int 639 vm_gpa_wire(struct vm *vm) 640 { 641 int i, rv; 642 struct mem_seg *seg; 643 644 for (i = 0; i < vm->num_mem_segs; i++) { 645 seg = &vm->mem_segs[i]; 646 if (seg->wired) 647 continue; 648 649 /* XXX rlimits? */ 650 rv = vm_map_wire(&vm->vmspace->vm_map, 651 seg->gpa, seg->gpa + seg->len, 652 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES); 653 if (rv != KERN_SUCCESS) 654 break; 655 656 seg->wired = TRUE; 657 } 658 659 if (i < vm->num_mem_segs) { 660 /* 661 * Undo the wiring before returning an error. 662 */ 663 vm_gpa_unwire(vm); 664 return (EAGAIN); 665 } 666 667 return (0); 668 } 669 670 static void 671 vm_iommu_modify(struct vm *vm, boolean_t map) 672 { 673 int i, sz; 674 vm_paddr_t gpa, hpa; 675 struct mem_seg *seg; 676 void *vp, *cookie, *host_domain; 677 678 sz = PAGE_SIZE; 679 host_domain = iommu_host_domain(); 680 681 for (i = 0; i < vm->num_mem_segs; i++) { 682 seg = &vm->mem_segs[i]; 683 KASSERT(seg->wired, ("vm(%s) memory segment %#lx/%ld not wired", 684 vm_name(vm), seg->gpa, seg->len)); 685 686 gpa = seg->gpa; 687 while (gpa < seg->gpa + seg->len) { 688 vp = vm_gpa_hold(vm, gpa, PAGE_SIZE, VM_PROT_WRITE, 689 &cookie); 690 KASSERT(vp != NULL, ("vm(%s) could not map gpa %#lx", 691 vm_name(vm), gpa)); 692 693 vm_gpa_release(cookie); 694 695 hpa = DMAP_TO_PHYS((uintptr_t)vp); 696 if (map) { 697 iommu_create_mapping(vm->iommu, gpa, hpa, sz); 698 iommu_remove_mapping(host_domain, hpa, sz); 699 } else { 700 iommu_remove_mapping(vm->iommu, gpa, sz); 701 iommu_create_mapping(host_domain, hpa, hpa, sz); 702 } 703 704 gpa += PAGE_SIZE; 705 } 706 } 707 708 /* 709 * Invalidate the cached translations associated with the domain 710 * from which pages were removed. 711 */ 712 if (map) 713 iommu_invalidate_tlb(host_domain); 714 else 715 iommu_invalidate_tlb(vm->iommu); 716 } 717 718 #define vm_iommu_unmap(vm) vm_iommu_modify((vm), FALSE) 719 #define vm_iommu_map(vm) vm_iommu_modify((vm), TRUE) 720 721 int 722 vm_unassign_pptdev(struct vm *vm, int bus, int slot, int func) 723 { 724 int error; 725 726 error = ppt_unassign_device(vm, bus, slot, func); 727 if (error) 728 return (error); 729 730 if (ppt_assigned_devices(vm) == 0) { 731 vm_iommu_unmap(vm); 732 vm_gpa_unwire(vm); 733 } 734 return (0); 735 } 736 737 int 738 vm_assign_pptdev(struct vm *vm, int bus, int slot, int func) 739 { 740 int error; 741 vm_paddr_t maxaddr; 742 743 /* 744 * Virtual machines with pci passthru devices get special treatment: 745 * - the guest physical memory is wired 746 * - the iommu is programmed to do the 'gpa' to 'hpa' translation 747 * 748 * We need to do this before the first pci passthru device is attached. 749 */ 750 if (ppt_assigned_devices(vm) == 0) { 751 KASSERT(vm->iommu == NULL, 752 ("vm_assign_pptdev: iommu must be NULL")); 753 maxaddr = vm_maxmem(vm); 754 vm->iommu = iommu_create_domain(maxaddr); 755 756 error = vm_gpa_wire(vm); 757 if (error) 758 return (error); 759 760 vm_iommu_map(vm); 761 } 762 763 error = ppt_assign_device(vm, bus, slot, func); 764 return (error); 765 } 766 767 void * 768 vm_gpa_hold(struct vm *vm, vm_paddr_t gpa, size_t len, int reqprot, 769 void **cookie) 770 { 771 int count, pageoff; 772 vm_page_t m; 773 774 pageoff = gpa & PAGE_MASK; 775 if (len > PAGE_SIZE - pageoff) 776 panic("vm_gpa_hold: invalid gpa/len: 0x%016lx/%lu", gpa, len); 777 778 count = vm_fault_quick_hold_pages(&vm->vmspace->vm_map, 779 trunc_page(gpa), PAGE_SIZE, reqprot, &m, 1); 780 781 if (count == 1) { 782 *cookie = m; 783 return ((void *)(PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m)) + pageoff)); 784 } else { 785 *cookie = NULL; 786 return (NULL); 787 } 788 } 789 790 void 791 vm_gpa_release(void *cookie) 792 { 793 vm_page_t m = cookie; 794 795 vm_page_lock(m); 796 vm_page_unhold(m); 797 vm_page_unlock(m); 798 } 799 800 int 801 vm_gpabase2memseg(struct vm *vm, vm_paddr_t gpabase, 802 struct vm_memory_segment *seg) 803 { 804 int i; 805 806 for (i = 0; i < vm->num_mem_segs; i++) { 807 if (gpabase == vm->mem_segs[i].gpa) { 808 seg->gpa = vm->mem_segs[i].gpa; 809 seg->len = vm->mem_segs[i].len; 810 seg->wired = vm->mem_segs[i].wired; 811 return (0); 812 } 813 } 814 return (-1); 815 } 816 817 int 818 vm_get_memobj(struct vm *vm, vm_paddr_t gpa, size_t len, 819 vm_offset_t *offset, struct vm_object **object) 820 { 821 int i; 822 size_t seg_len; 823 vm_paddr_t seg_gpa; 824 vm_object_t seg_obj; 825 826 for (i = 0; i < vm->num_mem_segs; i++) { 827 if ((seg_obj = vm->mem_segs[i].object) == NULL) 828 continue; 829 830 seg_gpa = vm->mem_segs[i].gpa; 831 seg_len = vm->mem_segs[i].len; 832 833 if (gpa >= seg_gpa && gpa < seg_gpa + seg_len) { 834 *offset = gpa - seg_gpa; 835 *object = seg_obj; 836 vm_object_reference(seg_obj); 837 return (0); 838 } 839 } 840 841 return (EINVAL); 842 } 843 844 int 845 vm_get_register(struct vm *vm, int vcpu, int reg, uint64_t *retval) 846 { 847 848 if (vcpu < 0 || vcpu >= VM_MAXCPU) 849 return (EINVAL); 850 851 if (reg >= VM_REG_LAST) 852 return (EINVAL); 853 854 return (VMGETREG(vm->cookie, vcpu, reg, retval)); 855 } 856 857 int 858 vm_set_register(struct vm *vm, int vcpuid, int reg, uint64_t val) 859 { 860 struct vcpu *vcpu; 861 int error; 862 863 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 864 return (EINVAL); 865 866 if (reg >= VM_REG_LAST) 867 return (EINVAL); 868 869 error = VMSETREG(vm->cookie, vcpuid, reg, val); 870 if (error || reg != VM_REG_GUEST_RIP) 871 return (error); 872 873 /* Set 'nextrip' to match the value of %rip */ 874 VCPU_CTR1(vm, vcpuid, "Setting nextrip to %#lx", val); 875 vcpu = &vm->vcpu[vcpuid]; 876 vcpu->nextrip = val; 877 return (0); 878 } 879 880 static boolean_t 881 is_descriptor_table(int reg) 882 { 883 884 switch (reg) { 885 case VM_REG_GUEST_IDTR: 886 case VM_REG_GUEST_GDTR: 887 return (TRUE); 888 default: 889 return (FALSE); 890 } 891 } 892 893 static boolean_t 894 is_segment_register(int reg) 895 { 896 897 switch (reg) { 898 case VM_REG_GUEST_ES: 899 case VM_REG_GUEST_CS: 900 case VM_REG_GUEST_SS: 901 case VM_REG_GUEST_DS: 902 case VM_REG_GUEST_FS: 903 case VM_REG_GUEST_GS: 904 case VM_REG_GUEST_TR: 905 case VM_REG_GUEST_LDTR: 906 return (TRUE); 907 default: 908 return (FALSE); 909 } 910 } 911 912 int 913 vm_get_seg_desc(struct vm *vm, int vcpu, int reg, 914 struct seg_desc *desc) 915 { 916 917 if (vcpu < 0 || vcpu >= VM_MAXCPU) 918 return (EINVAL); 919 920 if (!is_segment_register(reg) && !is_descriptor_table(reg)) 921 return (EINVAL); 922 923 return (VMGETDESC(vm->cookie, vcpu, reg, desc)); 924 } 925 926 int 927 vm_set_seg_desc(struct vm *vm, int vcpu, int reg, 928 struct seg_desc *desc) 929 { 930 if (vcpu < 0 || vcpu >= VM_MAXCPU) 931 return (EINVAL); 932 933 if (!is_segment_register(reg) && !is_descriptor_table(reg)) 934 return (EINVAL); 935 936 return (VMSETDESC(vm->cookie, vcpu, reg, desc)); 937 } 938 939 static void 940 restore_guest_fpustate(struct vcpu *vcpu) 941 { 942 943 /* flush host state to the pcb */ 944 fpuexit(curthread); 945 946 /* restore guest FPU state */ 947 fpu_stop_emulating(); 948 fpurestore(vcpu->guestfpu); 949 950 /* restore guest XCR0 if XSAVE is enabled in the host */ 951 if (rcr4() & CR4_XSAVE) 952 load_xcr(0, vcpu->guest_xcr0); 953 954 /* 955 * The FPU is now "dirty" with the guest's state so turn on emulation 956 * to trap any access to the FPU by the host. 957 */ 958 fpu_start_emulating(); 959 } 960 961 static void 962 save_guest_fpustate(struct vcpu *vcpu) 963 { 964 965 if ((rcr0() & CR0_TS) == 0) 966 panic("fpu emulation not enabled in host!"); 967 968 /* save guest XCR0 and restore host XCR0 */ 969 if (rcr4() & CR4_XSAVE) { 970 vcpu->guest_xcr0 = rxcr(0); 971 load_xcr(0, vmm_get_host_xcr0()); 972 } 973 974 /* save guest FPU state */ 975 fpu_stop_emulating(); 976 fpusave(vcpu->guestfpu); 977 fpu_start_emulating(); 978 } 979 980 static VMM_STAT(VCPU_IDLE_TICKS, "number of ticks vcpu was idle"); 981 982 static int 983 vcpu_set_state_locked(struct vcpu *vcpu, enum vcpu_state newstate, 984 bool from_idle) 985 { 986 int error; 987 988 vcpu_assert_locked(vcpu); 989 990 /* 991 * State transitions from the vmmdev_ioctl() must always begin from 992 * the VCPU_IDLE state. This guarantees that there is only a single 993 * ioctl() operating on a vcpu at any point. 994 */ 995 if (from_idle) { 996 while (vcpu->state != VCPU_IDLE) 997 msleep_spin(&vcpu->state, &vcpu->mtx, "vmstat", hz); 998 } else { 999 KASSERT(vcpu->state != VCPU_IDLE, ("invalid transition from " 1000 "vcpu idle state")); 1001 } 1002 1003 if (vcpu->state == VCPU_RUNNING) { 1004 KASSERT(vcpu->hostcpu == curcpu, ("curcpu %d and hostcpu %d " 1005 "mismatch for running vcpu", curcpu, vcpu->hostcpu)); 1006 } else { 1007 KASSERT(vcpu->hostcpu == NOCPU, ("Invalid hostcpu %d for a " 1008 "vcpu that is not running", vcpu->hostcpu)); 1009 } 1010 1011 /* 1012 * The following state transitions are allowed: 1013 * IDLE -> FROZEN -> IDLE 1014 * FROZEN -> RUNNING -> FROZEN 1015 * FROZEN -> SLEEPING -> FROZEN 1016 */ 1017 switch (vcpu->state) { 1018 case VCPU_IDLE: 1019 case VCPU_RUNNING: 1020 case VCPU_SLEEPING: 1021 error = (newstate != VCPU_FROZEN); 1022 break; 1023 case VCPU_FROZEN: 1024 error = (newstate == VCPU_FROZEN); 1025 break; 1026 default: 1027 error = 1; 1028 break; 1029 } 1030 1031 if (error) 1032 return (EBUSY); 1033 1034 vcpu->state = newstate; 1035 if (newstate == VCPU_RUNNING) 1036 vcpu->hostcpu = curcpu; 1037 else 1038 vcpu->hostcpu = NOCPU; 1039 1040 if (newstate == VCPU_IDLE) 1041 wakeup(&vcpu->state); 1042 1043 return (0); 1044 } 1045 1046 static void 1047 vcpu_require_state(struct vm *vm, int vcpuid, enum vcpu_state newstate) 1048 { 1049 int error; 1050 1051 if ((error = vcpu_set_state(vm, vcpuid, newstate, false)) != 0) 1052 panic("Error %d setting state to %d\n", error, newstate); 1053 } 1054 1055 static void 1056 vcpu_require_state_locked(struct vcpu *vcpu, enum vcpu_state newstate) 1057 { 1058 int error; 1059 1060 if ((error = vcpu_set_state_locked(vcpu, newstate, false)) != 0) 1061 panic("Error %d setting state to %d", error, newstate); 1062 } 1063 1064 static void 1065 vm_set_rendezvous_func(struct vm *vm, vm_rendezvous_func_t func) 1066 { 1067 1068 KASSERT(mtx_owned(&vm->rendezvous_mtx), ("rendezvous_mtx not locked")); 1069 1070 /* 1071 * Update 'rendezvous_func' and execute a write memory barrier to 1072 * ensure that it is visible across all host cpus. This is not needed 1073 * for correctness but it does ensure that all the vcpus will notice 1074 * that the rendezvous is requested immediately. 1075 */ 1076 vm->rendezvous_func = func; 1077 wmb(); 1078 } 1079 1080 #define RENDEZVOUS_CTR0(vm, vcpuid, fmt) \ 1081 do { \ 1082 if (vcpuid >= 0) \ 1083 VCPU_CTR0(vm, vcpuid, fmt); \ 1084 else \ 1085 VM_CTR0(vm, fmt); \ 1086 } while (0) 1087 1088 static void 1089 vm_handle_rendezvous(struct vm *vm, int vcpuid) 1090 { 1091 1092 KASSERT(vcpuid == -1 || (vcpuid >= 0 && vcpuid < VM_MAXCPU), 1093 ("vm_handle_rendezvous: invalid vcpuid %d", vcpuid)); 1094 1095 mtx_lock(&vm->rendezvous_mtx); 1096 while (vm->rendezvous_func != NULL) { 1097 /* 'rendezvous_req_cpus' must be a subset of 'active_cpus' */ 1098 CPU_AND(&vm->rendezvous_req_cpus, &vm->active_cpus); 1099 1100 if (vcpuid != -1 && 1101 CPU_ISSET(vcpuid, &vm->rendezvous_req_cpus) && 1102 !CPU_ISSET(vcpuid, &vm->rendezvous_done_cpus)) { 1103 VCPU_CTR0(vm, vcpuid, "Calling rendezvous func"); 1104 (*vm->rendezvous_func)(vm, vcpuid, vm->rendezvous_arg); 1105 CPU_SET(vcpuid, &vm->rendezvous_done_cpus); 1106 } 1107 if (CPU_CMP(&vm->rendezvous_req_cpus, 1108 &vm->rendezvous_done_cpus) == 0) { 1109 VCPU_CTR0(vm, vcpuid, "Rendezvous completed"); 1110 vm_set_rendezvous_func(vm, NULL); 1111 wakeup(&vm->rendezvous_func); 1112 break; 1113 } 1114 RENDEZVOUS_CTR0(vm, vcpuid, "Wait for rendezvous completion"); 1115 mtx_sleep(&vm->rendezvous_func, &vm->rendezvous_mtx, 0, 1116 "vmrndv", 0); 1117 } 1118 mtx_unlock(&vm->rendezvous_mtx); 1119 } 1120 1121 /* 1122 * Emulate a guest 'hlt' by sleeping until the vcpu is ready to run. 1123 */ 1124 static int 1125 vm_handle_hlt(struct vm *vm, int vcpuid, bool intr_disabled, bool *retu) 1126 { 1127 struct vcpu *vcpu; 1128 const char *wmesg; 1129 int t, vcpu_halted, vm_halted; 1130 1131 KASSERT(!CPU_ISSET(vcpuid, &vm->halted_cpus), ("vcpu already halted")); 1132 1133 vcpu = &vm->vcpu[vcpuid]; 1134 vcpu_halted = 0; 1135 vm_halted = 0; 1136 1137 vcpu_lock(vcpu); 1138 while (1) { 1139 /* 1140 * Do a final check for pending NMI or interrupts before 1141 * really putting this thread to sleep. Also check for 1142 * software events that would cause this vcpu to wakeup. 1143 * 1144 * These interrupts/events could have happened after the 1145 * vcpu returned from VMRUN() and before it acquired the 1146 * vcpu lock above. 1147 */ 1148 if (vm->rendezvous_func != NULL || vm->suspend) 1149 break; 1150 if (vm_nmi_pending(vm, vcpuid)) 1151 break; 1152 if (!intr_disabled) { 1153 if (vm_extint_pending(vm, vcpuid) || 1154 vlapic_pending_intr(vcpu->vlapic, NULL)) { 1155 break; 1156 } 1157 } 1158 1159 /* Don't go to sleep if the vcpu thread needs to yield */ 1160 if (vcpu_should_yield(vm, vcpuid)) 1161 break; 1162 1163 /* 1164 * Some Linux guests implement "halt" by having all vcpus 1165 * execute HLT with interrupts disabled. 'halted_cpus' keeps 1166 * track of the vcpus that have entered this state. When all 1167 * vcpus enter the halted state the virtual machine is halted. 1168 */ 1169 if (intr_disabled) { 1170 wmesg = "vmhalt"; 1171 VCPU_CTR0(vm, vcpuid, "Halted"); 1172 if (!vcpu_halted && halt_detection_enabled) { 1173 vcpu_halted = 1; 1174 CPU_SET_ATOMIC(vcpuid, &vm->halted_cpus); 1175 } 1176 if (CPU_CMP(&vm->halted_cpus, &vm->active_cpus) == 0) { 1177 vm_halted = 1; 1178 break; 1179 } 1180 } else { 1181 wmesg = "vmidle"; 1182 } 1183 1184 t = ticks; 1185 vcpu_require_state_locked(vcpu, VCPU_SLEEPING); 1186 /* 1187 * XXX msleep_spin() cannot be interrupted by signals so 1188 * wake up periodically to check pending signals. 1189 */ 1190 msleep_spin(vcpu, &vcpu->mtx, wmesg, hz); 1191 vcpu_require_state_locked(vcpu, VCPU_FROZEN); 1192 vmm_stat_incr(vm, vcpuid, VCPU_IDLE_TICKS, ticks - t); 1193 } 1194 1195 if (vcpu_halted) 1196 CPU_CLR_ATOMIC(vcpuid, &vm->halted_cpus); 1197 1198 vcpu_unlock(vcpu); 1199 1200 if (vm_halted) 1201 vm_suspend(vm, VM_SUSPEND_HALT); 1202 1203 return (0); 1204 } 1205 1206 static int 1207 vm_handle_paging(struct vm *vm, int vcpuid, bool *retu) 1208 { 1209 int rv, ftype; 1210 struct vm_map *map; 1211 struct vcpu *vcpu; 1212 struct vm_exit *vme; 1213 1214 vcpu = &vm->vcpu[vcpuid]; 1215 vme = &vcpu->exitinfo; 1216 1217 KASSERT(vme->inst_length == 0, ("%s: invalid inst_length %d", 1218 __func__, vme->inst_length)); 1219 1220 ftype = vme->u.paging.fault_type; 1221 KASSERT(ftype == VM_PROT_READ || 1222 ftype == VM_PROT_WRITE || ftype == VM_PROT_EXECUTE, 1223 ("vm_handle_paging: invalid fault_type %d", ftype)); 1224 1225 if (ftype == VM_PROT_READ || ftype == VM_PROT_WRITE) { 1226 rv = pmap_emulate_accessed_dirty(vmspace_pmap(vm->vmspace), 1227 vme->u.paging.gpa, ftype); 1228 if (rv == 0) { 1229 VCPU_CTR2(vm, vcpuid, "%s bit emulation for gpa %#lx", 1230 ftype == VM_PROT_READ ? "accessed" : "dirty", 1231 vme->u.paging.gpa); 1232 goto done; 1233 } 1234 } 1235 1236 map = &vm->vmspace->vm_map; 1237 rv = vm_fault(map, vme->u.paging.gpa, ftype, VM_FAULT_NORMAL); 1238 1239 VCPU_CTR3(vm, vcpuid, "vm_handle_paging rv = %d, gpa = %#lx, " 1240 "ftype = %d", rv, vme->u.paging.gpa, ftype); 1241 1242 if (rv != KERN_SUCCESS) 1243 return (EFAULT); 1244 done: 1245 return (0); 1246 } 1247 1248 static int 1249 vm_handle_inst_emul(struct vm *vm, int vcpuid, bool *retu) 1250 { 1251 struct vie *vie; 1252 struct vcpu *vcpu; 1253 struct vm_exit *vme; 1254 uint64_t gla, gpa, cs_base; 1255 struct vm_guest_paging *paging; 1256 mem_region_read_t mread; 1257 mem_region_write_t mwrite; 1258 enum vm_cpu_mode cpu_mode; 1259 int cs_d, error, fault; 1260 1261 vcpu = &vm->vcpu[vcpuid]; 1262 vme = &vcpu->exitinfo; 1263 1264 KASSERT(vme->inst_length == 0, ("%s: invalid inst_length %d", 1265 __func__, vme->inst_length)); 1266 1267 gla = vme->u.inst_emul.gla; 1268 gpa = vme->u.inst_emul.gpa; 1269 cs_base = vme->u.inst_emul.cs_base; 1270 cs_d = vme->u.inst_emul.cs_d; 1271 vie = &vme->u.inst_emul.vie; 1272 paging = &vme->u.inst_emul.paging; 1273 cpu_mode = paging->cpu_mode; 1274 1275 VCPU_CTR1(vm, vcpuid, "inst_emul fault accessing gpa %#lx", gpa); 1276 1277 /* Fetch, decode and emulate the faulting instruction */ 1278 if (vie->num_valid == 0) { 1279 error = vmm_fetch_instruction(vm, vcpuid, paging, vme->rip + 1280 cs_base, VIE_INST_SIZE, vie, &fault); 1281 } else { 1282 /* 1283 * The instruction bytes have already been copied into 'vie' 1284 */ 1285 error = fault = 0; 1286 } 1287 if (error || fault) 1288 return (error); 1289 1290 if (vmm_decode_instruction(vm, vcpuid, gla, cpu_mode, cs_d, vie) != 0) { 1291 VCPU_CTR1(vm, vcpuid, "Error decoding instruction at %#lx", 1292 vme->rip + cs_base); 1293 *retu = true; /* dump instruction bytes in userspace */ 1294 return (0); 1295 } 1296 1297 /* 1298 * Update 'nextrip' based on the length of the emulated instruction. 1299 */ 1300 vme->inst_length = vie->num_processed; 1301 vcpu->nextrip += vie->num_processed; 1302 VCPU_CTR1(vm, vcpuid, "nextrip updated to %#lx after instruction " 1303 "decoding", vcpu->nextrip); 1304 1305 /* return to userland unless this is an in-kernel emulated device */ 1306 if (gpa >= DEFAULT_APIC_BASE && gpa < DEFAULT_APIC_BASE + PAGE_SIZE) { 1307 mread = lapic_mmio_read; 1308 mwrite = lapic_mmio_write; 1309 } else if (gpa >= VIOAPIC_BASE && gpa < VIOAPIC_BASE + VIOAPIC_SIZE) { 1310 mread = vioapic_mmio_read; 1311 mwrite = vioapic_mmio_write; 1312 } else if (gpa >= VHPET_BASE && gpa < VHPET_BASE + VHPET_SIZE) { 1313 mread = vhpet_mmio_read; 1314 mwrite = vhpet_mmio_write; 1315 } else { 1316 *retu = true; 1317 return (0); 1318 } 1319 1320 error = vmm_emulate_instruction(vm, vcpuid, gpa, vie, paging, 1321 mread, mwrite, retu); 1322 1323 return (error); 1324 } 1325 1326 static int 1327 vm_handle_suspend(struct vm *vm, int vcpuid, bool *retu) 1328 { 1329 int i, done; 1330 struct vcpu *vcpu; 1331 1332 done = 0; 1333 vcpu = &vm->vcpu[vcpuid]; 1334 1335 CPU_SET_ATOMIC(vcpuid, &vm->suspended_cpus); 1336 1337 /* 1338 * Wait until all 'active_cpus' have suspended themselves. 1339 * 1340 * Since a VM may be suspended at any time including when one or 1341 * more vcpus are doing a rendezvous we need to call the rendezvous 1342 * handler while we are waiting to prevent a deadlock. 1343 */ 1344 vcpu_lock(vcpu); 1345 while (1) { 1346 if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) { 1347 VCPU_CTR0(vm, vcpuid, "All vcpus suspended"); 1348 break; 1349 } 1350 1351 if (vm->rendezvous_func == NULL) { 1352 VCPU_CTR0(vm, vcpuid, "Sleeping during suspend"); 1353 vcpu_require_state_locked(vcpu, VCPU_SLEEPING); 1354 msleep_spin(vcpu, &vcpu->mtx, "vmsusp", hz); 1355 vcpu_require_state_locked(vcpu, VCPU_FROZEN); 1356 } else { 1357 VCPU_CTR0(vm, vcpuid, "Rendezvous during suspend"); 1358 vcpu_unlock(vcpu); 1359 vm_handle_rendezvous(vm, vcpuid); 1360 vcpu_lock(vcpu); 1361 } 1362 } 1363 vcpu_unlock(vcpu); 1364 1365 /* 1366 * Wakeup the other sleeping vcpus and return to userspace. 1367 */ 1368 for (i = 0; i < VM_MAXCPU; i++) { 1369 if (CPU_ISSET(i, &vm->suspended_cpus)) { 1370 vcpu_notify_event(vm, i, false); 1371 } 1372 } 1373 1374 *retu = true; 1375 return (0); 1376 } 1377 1378 int 1379 vm_suspend(struct vm *vm, enum vm_suspend_how how) 1380 { 1381 int i; 1382 1383 if (how <= VM_SUSPEND_NONE || how >= VM_SUSPEND_LAST) 1384 return (EINVAL); 1385 1386 if (atomic_cmpset_int(&vm->suspend, 0, how) == 0) { 1387 VM_CTR2(vm, "virtual machine already suspended %d/%d", 1388 vm->suspend, how); 1389 return (EALREADY); 1390 } 1391 1392 VM_CTR1(vm, "virtual machine successfully suspended %d", how); 1393 1394 /* 1395 * Notify all active vcpus that they are now suspended. 1396 */ 1397 for (i = 0; i < VM_MAXCPU; i++) { 1398 if (CPU_ISSET(i, &vm->active_cpus)) 1399 vcpu_notify_event(vm, i, false); 1400 } 1401 1402 return (0); 1403 } 1404 1405 void 1406 vm_exit_suspended(struct vm *vm, int vcpuid, uint64_t rip) 1407 { 1408 struct vm_exit *vmexit; 1409 1410 KASSERT(vm->suspend > VM_SUSPEND_NONE && vm->suspend < VM_SUSPEND_LAST, 1411 ("vm_exit_suspended: invalid suspend type %d", vm->suspend)); 1412 1413 vmexit = vm_exitinfo(vm, vcpuid); 1414 vmexit->rip = rip; 1415 vmexit->inst_length = 0; 1416 vmexit->exitcode = VM_EXITCODE_SUSPENDED; 1417 vmexit->u.suspended.how = vm->suspend; 1418 } 1419 1420 void 1421 vm_exit_rendezvous(struct vm *vm, int vcpuid, uint64_t rip) 1422 { 1423 struct vm_exit *vmexit; 1424 1425 KASSERT(vm->rendezvous_func != NULL, ("rendezvous not in progress")); 1426 1427 vmexit = vm_exitinfo(vm, vcpuid); 1428 vmexit->rip = rip; 1429 vmexit->inst_length = 0; 1430 vmexit->exitcode = VM_EXITCODE_RENDEZVOUS; 1431 vmm_stat_incr(vm, vcpuid, VMEXIT_RENDEZVOUS, 1); 1432 } 1433 1434 void 1435 vm_exit_astpending(struct vm *vm, int vcpuid, uint64_t rip) 1436 { 1437 struct vm_exit *vmexit; 1438 1439 vmexit = vm_exitinfo(vm, vcpuid); 1440 vmexit->rip = rip; 1441 vmexit->inst_length = 0; 1442 vmexit->exitcode = VM_EXITCODE_BOGUS; 1443 vmm_stat_incr(vm, vcpuid, VMEXIT_ASTPENDING, 1); 1444 } 1445 1446 int 1447 vm_run(struct vm *vm, struct vm_run *vmrun) 1448 { 1449 int error, vcpuid; 1450 struct vcpu *vcpu; 1451 struct pcb *pcb; 1452 uint64_t tscval; 1453 struct vm_exit *vme; 1454 bool retu, intr_disabled; 1455 pmap_t pmap; 1456 void *rptr, *sptr; 1457 1458 vcpuid = vmrun->cpuid; 1459 1460 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1461 return (EINVAL); 1462 1463 if (!CPU_ISSET(vcpuid, &vm->active_cpus)) 1464 return (EINVAL); 1465 1466 if (CPU_ISSET(vcpuid, &vm->suspended_cpus)) 1467 return (EINVAL); 1468 1469 rptr = &vm->rendezvous_func; 1470 sptr = &vm->suspend; 1471 pmap = vmspace_pmap(vm->vmspace); 1472 vcpu = &vm->vcpu[vcpuid]; 1473 vme = &vcpu->exitinfo; 1474 restart: 1475 critical_enter(); 1476 1477 KASSERT(!CPU_ISSET(curcpu, &pmap->pm_active), 1478 ("vm_run: absurd pm_active")); 1479 1480 tscval = rdtsc(); 1481 1482 pcb = PCPU_GET(curpcb); 1483 set_pcb_flags(pcb, PCB_FULL_IRET); 1484 1485 restore_guest_fpustate(vcpu); 1486 1487 vcpu_require_state(vm, vcpuid, VCPU_RUNNING); 1488 error = VMRUN(vm->cookie, vcpuid, vcpu->nextrip, pmap, rptr, sptr); 1489 vcpu_require_state(vm, vcpuid, VCPU_FROZEN); 1490 1491 save_guest_fpustate(vcpu); 1492 1493 vmm_stat_incr(vm, vcpuid, VCPU_TOTAL_RUNTIME, rdtsc() - tscval); 1494 1495 critical_exit(); 1496 1497 if (error == 0) { 1498 retu = false; 1499 vcpu->nextrip = vme->rip + vme->inst_length; 1500 switch (vme->exitcode) { 1501 case VM_EXITCODE_SUSPENDED: 1502 error = vm_handle_suspend(vm, vcpuid, &retu); 1503 break; 1504 case VM_EXITCODE_IOAPIC_EOI: 1505 vioapic_process_eoi(vm, vcpuid, 1506 vme->u.ioapic_eoi.vector); 1507 break; 1508 case VM_EXITCODE_RENDEZVOUS: 1509 vm_handle_rendezvous(vm, vcpuid); 1510 error = 0; 1511 break; 1512 case VM_EXITCODE_HLT: 1513 intr_disabled = ((vme->u.hlt.rflags & PSL_I) == 0); 1514 error = vm_handle_hlt(vm, vcpuid, intr_disabled, &retu); 1515 break; 1516 case VM_EXITCODE_PAGING: 1517 error = vm_handle_paging(vm, vcpuid, &retu); 1518 break; 1519 case VM_EXITCODE_INST_EMUL: 1520 error = vm_handle_inst_emul(vm, vcpuid, &retu); 1521 break; 1522 case VM_EXITCODE_INOUT: 1523 case VM_EXITCODE_INOUT_STR: 1524 error = vm_handle_inout(vm, vcpuid, vme, &retu); 1525 break; 1526 case VM_EXITCODE_MONITOR: 1527 case VM_EXITCODE_MWAIT: 1528 vm_inject_ud(vm, vcpuid); 1529 break; 1530 default: 1531 retu = true; /* handled in userland */ 1532 break; 1533 } 1534 } 1535 1536 if (error == 0 && retu == false) 1537 goto restart; 1538 1539 /* copy the exit information */ 1540 bcopy(vme, &vmrun->vm_exit, sizeof(struct vm_exit)); 1541 return (error); 1542 } 1543 1544 int 1545 vm_restart_instruction(void *arg, int vcpuid) 1546 { 1547 struct vm *vm; 1548 struct vcpu *vcpu; 1549 enum vcpu_state state; 1550 uint64_t rip; 1551 int error; 1552 1553 vm = arg; 1554 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1555 return (EINVAL); 1556 1557 vcpu = &vm->vcpu[vcpuid]; 1558 state = vcpu_get_state(vm, vcpuid, NULL); 1559 if (state == VCPU_RUNNING) { 1560 /* 1561 * When a vcpu is "running" the next instruction is determined 1562 * by adding 'rip' and 'inst_length' in the vcpu's 'exitinfo'. 1563 * Thus setting 'inst_length' to zero will cause the current 1564 * instruction to be restarted. 1565 */ 1566 vcpu->exitinfo.inst_length = 0; 1567 VCPU_CTR1(vm, vcpuid, "restarting instruction at %#lx by " 1568 "setting inst_length to zero", vcpu->exitinfo.rip); 1569 } else if (state == VCPU_FROZEN) { 1570 /* 1571 * When a vcpu is "frozen" it is outside the critical section 1572 * around VMRUN() and 'nextrip' points to the next instruction. 1573 * Thus instruction restart is achieved by setting 'nextrip' 1574 * to the vcpu's %rip. 1575 */ 1576 error = vm_get_register(vm, vcpuid, VM_REG_GUEST_RIP, &rip); 1577 KASSERT(!error, ("%s: error %d getting rip", __func__, error)); 1578 VCPU_CTR2(vm, vcpuid, "restarting instruction by updating " 1579 "nextrip from %#lx to %#lx", vcpu->nextrip, rip); 1580 vcpu->nextrip = rip; 1581 } else { 1582 panic("%s: invalid state %d", __func__, state); 1583 } 1584 return (0); 1585 } 1586 1587 int 1588 vm_exit_intinfo(struct vm *vm, int vcpuid, uint64_t info) 1589 { 1590 struct vcpu *vcpu; 1591 int type, vector; 1592 1593 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1594 return (EINVAL); 1595 1596 vcpu = &vm->vcpu[vcpuid]; 1597 1598 if (info & VM_INTINFO_VALID) { 1599 type = info & VM_INTINFO_TYPE; 1600 vector = info & 0xff; 1601 if (type == VM_INTINFO_NMI && vector != IDT_NMI) 1602 return (EINVAL); 1603 if (type == VM_INTINFO_HWEXCEPTION && vector >= 32) 1604 return (EINVAL); 1605 if (info & VM_INTINFO_RSVD) 1606 return (EINVAL); 1607 } else { 1608 info = 0; 1609 } 1610 VCPU_CTR2(vm, vcpuid, "%s: info1(%#lx)", __func__, info); 1611 vcpu->exitintinfo = info; 1612 return (0); 1613 } 1614 1615 enum exc_class { 1616 EXC_BENIGN, 1617 EXC_CONTRIBUTORY, 1618 EXC_PAGEFAULT 1619 }; 1620 1621 #define IDT_VE 20 /* Virtualization Exception (Intel specific) */ 1622 1623 static enum exc_class 1624 exception_class(uint64_t info) 1625 { 1626 int type, vector; 1627 1628 KASSERT(info & VM_INTINFO_VALID, ("intinfo must be valid: %#lx", info)); 1629 type = info & VM_INTINFO_TYPE; 1630 vector = info & 0xff; 1631 1632 /* Table 6-4, "Interrupt and Exception Classes", Intel SDM, Vol 3 */ 1633 switch (type) { 1634 case VM_INTINFO_HWINTR: 1635 case VM_INTINFO_SWINTR: 1636 case VM_INTINFO_NMI: 1637 return (EXC_BENIGN); 1638 default: 1639 /* 1640 * Hardware exception. 1641 * 1642 * SVM and VT-x use identical type values to represent NMI, 1643 * hardware interrupt and software interrupt. 1644 * 1645 * SVM uses type '3' for all exceptions. VT-x uses type '3' 1646 * for exceptions except #BP and #OF. #BP and #OF use a type 1647 * value of '5' or '6'. Therefore we don't check for explicit 1648 * values of 'type' to classify 'intinfo' into a hardware 1649 * exception. 1650 */ 1651 break; 1652 } 1653 1654 switch (vector) { 1655 case IDT_PF: 1656 case IDT_VE: 1657 return (EXC_PAGEFAULT); 1658 case IDT_DE: 1659 case IDT_TS: 1660 case IDT_NP: 1661 case IDT_SS: 1662 case IDT_GP: 1663 return (EXC_CONTRIBUTORY); 1664 default: 1665 return (EXC_BENIGN); 1666 } 1667 } 1668 1669 static int 1670 nested_fault(struct vm *vm, int vcpuid, uint64_t info1, uint64_t info2, 1671 uint64_t *retinfo) 1672 { 1673 enum exc_class exc1, exc2; 1674 int type1, vector1; 1675 1676 KASSERT(info1 & VM_INTINFO_VALID, ("info1 %#lx is not valid", info1)); 1677 KASSERT(info2 & VM_INTINFO_VALID, ("info2 %#lx is not valid", info2)); 1678 1679 /* 1680 * If an exception occurs while attempting to call the double-fault 1681 * handler the processor enters shutdown mode (aka triple fault). 1682 */ 1683 type1 = info1 & VM_INTINFO_TYPE; 1684 vector1 = info1 & 0xff; 1685 if (type1 == VM_INTINFO_HWEXCEPTION && vector1 == IDT_DF) { 1686 VCPU_CTR2(vm, vcpuid, "triple fault: info1(%#lx), info2(%#lx)", 1687 info1, info2); 1688 vm_suspend(vm, VM_SUSPEND_TRIPLEFAULT); 1689 *retinfo = 0; 1690 return (0); 1691 } 1692 1693 /* 1694 * Table 6-5 "Conditions for Generating a Double Fault", Intel SDM, Vol3 1695 */ 1696 exc1 = exception_class(info1); 1697 exc2 = exception_class(info2); 1698 if ((exc1 == EXC_CONTRIBUTORY && exc2 == EXC_CONTRIBUTORY) || 1699 (exc1 == EXC_PAGEFAULT && exc2 != EXC_BENIGN)) { 1700 /* Convert nested fault into a double fault. */ 1701 *retinfo = IDT_DF; 1702 *retinfo |= VM_INTINFO_VALID | VM_INTINFO_HWEXCEPTION; 1703 *retinfo |= VM_INTINFO_DEL_ERRCODE; 1704 } else { 1705 /* Handle exceptions serially */ 1706 *retinfo = info2; 1707 } 1708 return (1); 1709 } 1710 1711 static uint64_t 1712 vcpu_exception_intinfo(struct vcpu *vcpu) 1713 { 1714 uint64_t info = 0; 1715 1716 if (vcpu->exception_pending) { 1717 info = vcpu->exc_vector & 0xff; 1718 info |= VM_INTINFO_VALID | VM_INTINFO_HWEXCEPTION; 1719 if (vcpu->exc_errcode_valid) { 1720 info |= VM_INTINFO_DEL_ERRCODE; 1721 info |= (uint64_t)vcpu->exc_errcode << 32; 1722 } 1723 } 1724 return (info); 1725 } 1726 1727 int 1728 vm_entry_intinfo(struct vm *vm, int vcpuid, uint64_t *retinfo) 1729 { 1730 struct vcpu *vcpu; 1731 uint64_t info1, info2; 1732 int valid; 1733 1734 KASSERT(vcpuid >= 0 && vcpuid < VM_MAXCPU, ("invalid vcpu %d", vcpuid)); 1735 1736 vcpu = &vm->vcpu[vcpuid]; 1737 1738 info1 = vcpu->exitintinfo; 1739 vcpu->exitintinfo = 0; 1740 1741 info2 = 0; 1742 if (vcpu->exception_pending) { 1743 info2 = vcpu_exception_intinfo(vcpu); 1744 vcpu->exception_pending = 0; 1745 VCPU_CTR2(vm, vcpuid, "Exception %d delivered: %#lx", 1746 vcpu->exc_vector, info2); 1747 } 1748 1749 if ((info1 & VM_INTINFO_VALID) && (info2 & VM_INTINFO_VALID)) { 1750 valid = nested_fault(vm, vcpuid, info1, info2, retinfo); 1751 } else if (info1 & VM_INTINFO_VALID) { 1752 *retinfo = info1; 1753 valid = 1; 1754 } else if (info2 & VM_INTINFO_VALID) { 1755 *retinfo = info2; 1756 valid = 1; 1757 } else { 1758 valid = 0; 1759 } 1760 1761 if (valid) { 1762 VCPU_CTR4(vm, vcpuid, "%s: info1(%#lx), info2(%#lx), " 1763 "retinfo(%#lx)", __func__, info1, info2, *retinfo); 1764 } 1765 1766 return (valid); 1767 } 1768 1769 int 1770 vm_get_intinfo(struct vm *vm, int vcpuid, uint64_t *info1, uint64_t *info2) 1771 { 1772 struct vcpu *vcpu; 1773 1774 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1775 return (EINVAL); 1776 1777 vcpu = &vm->vcpu[vcpuid]; 1778 *info1 = vcpu->exitintinfo; 1779 *info2 = vcpu_exception_intinfo(vcpu); 1780 return (0); 1781 } 1782 1783 int 1784 vm_inject_exception(struct vm *vm, int vcpuid, int vector, int errcode_valid, 1785 uint32_t errcode, int restart_instruction) 1786 { 1787 struct vcpu *vcpu; 1788 uint64_t regval; 1789 int error; 1790 1791 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1792 return (EINVAL); 1793 1794 if (vector < 0 || vector >= 32) 1795 return (EINVAL); 1796 1797 /* 1798 * A double fault exception should never be injected directly into 1799 * the guest. It is a derived exception that results from specific 1800 * combinations of nested faults. 1801 */ 1802 if (vector == IDT_DF) 1803 return (EINVAL); 1804 1805 vcpu = &vm->vcpu[vcpuid]; 1806 1807 if (vcpu->exception_pending) { 1808 VCPU_CTR2(vm, vcpuid, "Unable to inject exception %d due to " 1809 "pending exception %d", vector, vcpu->exc_vector); 1810 return (EBUSY); 1811 } 1812 1813 if (errcode_valid) { 1814 /* 1815 * Exceptions don't deliver an error code in real mode. 1816 */ 1817 error = vm_get_register(vm, vcpuid, VM_REG_GUEST_CR0, ®val); 1818 KASSERT(!error, ("%s: error %d getting CR0", __func__, error)); 1819 if (!(regval & CR0_PE)) 1820 errcode_valid = 0; 1821 } 1822 1823 /* 1824 * From section 26.6.1 "Interruptibility State" in Intel SDM: 1825 * 1826 * Event blocking by "STI" or "MOV SS" is cleared after guest executes 1827 * one instruction or incurs an exception. 1828 */ 1829 error = vm_set_register(vm, vcpuid, VM_REG_GUEST_INTR_SHADOW, 0); 1830 KASSERT(error == 0, ("%s: error %d clearing interrupt shadow", 1831 __func__, error)); 1832 1833 if (restart_instruction) 1834 vm_restart_instruction(vm, vcpuid); 1835 1836 vcpu->exception_pending = 1; 1837 vcpu->exc_vector = vector; 1838 vcpu->exc_errcode = errcode; 1839 vcpu->exc_errcode_valid = errcode_valid; 1840 VCPU_CTR1(vm, vcpuid, "Exception %d pending", vector); 1841 return (0); 1842 } 1843 1844 void 1845 vm_inject_fault(void *vmarg, int vcpuid, int vector, int errcode_valid, 1846 int errcode) 1847 { 1848 struct vm *vm; 1849 int error, restart_instruction; 1850 1851 vm = vmarg; 1852 restart_instruction = 1; 1853 1854 error = vm_inject_exception(vm, vcpuid, vector, errcode_valid, 1855 errcode, restart_instruction); 1856 KASSERT(error == 0, ("vm_inject_exception error %d", error)); 1857 } 1858 1859 void 1860 vm_inject_pf(void *vmarg, int vcpuid, int error_code, uint64_t cr2) 1861 { 1862 struct vm *vm; 1863 int error; 1864 1865 vm = vmarg; 1866 VCPU_CTR2(vm, vcpuid, "Injecting page fault: error_code %#x, cr2 %#lx", 1867 error_code, cr2); 1868 1869 error = vm_set_register(vm, vcpuid, VM_REG_GUEST_CR2, cr2); 1870 KASSERT(error == 0, ("vm_set_register(cr2) error %d", error)); 1871 1872 vm_inject_fault(vm, vcpuid, IDT_PF, 1, error_code); 1873 } 1874 1875 static VMM_STAT(VCPU_NMI_COUNT, "number of NMIs delivered to vcpu"); 1876 1877 int 1878 vm_inject_nmi(struct vm *vm, int vcpuid) 1879 { 1880 struct vcpu *vcpu; 1881 1882 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1883 return (EINVAL); 1884 1885 vcpu = &vm->vcpu[vcpuid]; 1886 1887 vcpu->nmi_pending = 1; 1888 vcpu_notify_event(vm, vcpuid, false); 1889 return (0); 1890 } 1891 1892 int 1893 vm_nmi_pending(struct vm *vm, int vcpuid) 1894 { 1895 struct vcpu *vcpu; 1896 1897 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1898 panic("vm_nmi_pending: invalid vcpuid %d", vcpuid); 1899 1900 vcpu = &vm->vcpu[vcpuid]; 1901 1902 return (vcpu->nmi_pending); 1903 } 1904 1905 void 1906 vm_nmi_clear(struct vm *vm, int vcpuid) 1907 { 1908 struct vcpu *vcpu; 1909 1910 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1911 panic("vm_nmi_pending: invalid vcpuid %d", vcpuid); 1912 1913 vcpu = &vm->vcpu[vcpuid]; 1914 1915 if (vcpu->nmi_pending == 0) 1916 panic("vm_nmi_clear: inconsistent nmi_pending state"); 1917 1918 vcpu->nmi_pending = 0; 1919 vmm_stat_incr(vm, vcpuid, VCPU_NMI_COUNT, 1); 1920 } 1921 1922 static VMM_STAT(VCPU_EXTINT_COUNT, "number of ExtINTs delivered to vcpu"); 1923 1924 int 1925 vm_inject_extint(struct vm *vm, int vcpuid) 1926 { 1927 struct vcpu *vcpu; 1928 1929 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1930 return (EINVAL); 1931 1932 vcpu = &vm->vcpu[vcpuid]; 1933 1934 vcpu->extint_pending = 1; 1935 vcpu_notify_event(vm, vcpuid, false); 1936 return (0); 1937 } 1938 1939 int 1940 vm_extint_pending(struct vm *vm, int vcpuid) 1941 { 1942 struct vcpu *vcpu; 1943 1944 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1945 panic("vm_extint_pending: invalid vcpuid %d", vcpuid); 1946 1947 vcpu = &vm->vcpu[vcpuid]; 1948 1949 return (vcpu->extint_pending); 1950 } 1951 1952 void 1953 vm_extint_clear(struct vm *vm, int vcpuid) 1954 { 1955 struct vcpu *vcpu; 1956 1957 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1958 panic("vm_extint_pending: invalid vcpuid %d", vcpuid); 1959 1960 vcpu = &vm->vcpu[vcpuid]; 1961 1962 if (vcpu->extint_pending == 0) 1963 panic("vm_extint_clear: inconsistent extint_pending state"); 1964 1965 vcpu->extint_pending = 0; 1966 vmm_stat_incr(vm, vcpuid, VCPU_EXTINT_COUNT, 1); 1967 } 1968 1969 int 1970 vm_get_capability(struct vm *vm, int vcpu, int type, int *retval) 1971 { 1972 if (vcpu < 0 || vcpu >= VM_MAXCPU) 1973 return (EINVAL); 1974 1975 if (type < 0 || type >= VM_CAP_MAX) 1976 return (EINVAL); 1977 1978 return (VMGETCAP(vm->cookie, vcpu, type, retval)); 1979 } 1980 1981 int 1982 vm_set_capability(struct vm *vm, int vcpu, int type, int val) 1983 { 1984 if (vcpu < 0 || vcpu >= VM_MAXCPU) 1985 return (EINVAL); 1986 1987 if (type < 0 || type >= VM_CAP_MAX) 1988 return (EINVAL); 1989 1990 return (VMSETCAP(vm->cookie, vcpu, type, val)); 1991 } 1992 1993 struct vlapic * 1994 vm_lapic(struct vm *vm, int cpu) 1995 { 1996 return (vm->vcpu[cpu].vlapic); 1997 } 1998 1999 struct vioapic * 2000 vm_ioapic(struct vm *vm) 2001 { 2002 2003 return (vm->vioapic); 2004 } 2005 2006 struct vhpet * 2007 vm_hpet(struct vm *vm) 2008 { 2009 2010 return (vm->vhpet); 2011 } 2012 2013 boolean_t 2014 vmm_is_pptdev(int bus, int slot, int func) 2015 { 2016 int found, i, n; 2017 int b, s, f; 2018 char *val, *cp, *cp2; 2019 2020 /* 2021 * XXX 2022 * The length of an environment variable is limited to 128 bytes which 2023 * puts an upper limit on the number of passthru devices that may be 2024 * specified using a single environment variable. 2025 * 2026 * Work around this by scanning multiple environment variable 2027 * names instead of a single one - yuck! 2028 */ 2029 const char *names[] = { "pptdevs", "pptdevs2", "pptdevs3", NULL }; 2030 2031 /* set pptdevs="1/2/3 4/5/6 7/8/9 10/11/12" */ 2032 found = 0; 2033 for (i = 0; names[i] != NULL && !found; i++) { 2034 cp = val = kern_getenv(names[i]); 2035 while (cp != NULL && *cp != '\0') { 2036 if ((cp2 = strchr(cp, ' ')) != NULL) 2037 *cp2 = '\0'; 2038 2039 n = sscanf(cp, "%d/%d/%d", &b, &s, &f); 2040 if (n == 3 && bus == b && slot == s && func == f) { 2041 found = 1; 2042 break; 2043 } 2044 2045 if (cp2 != NULL) 2046 *cp2++ = ' '; 2047 2048 cp = cp2; 2049 } 2050 freeenv(val); 2051 } 2052 return (found); 2053 } 2054 2055 void * 2056 vm_iommu_domain(struct vm *vm) 2057 { 2058 2059 return (vm->iommu); 2060 } 2061 2062 int 2063 vcpu_set_state(struct vm *vm, int vcpuid, enum vcpu_state newstate, 2064 bool from_idle) 2065 { 2066 int error; 2067 struct vcpu *vcpu; 2068 2069 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 2070 panic("vm_set_run_state: invalid vcpuid %d", vcpuid); 2071 2072 vcpu = &vm->vcpu[vcpuid]; 2073 2074 vcpu_lock(vcpu); 2075 error = vcpu_set_state_locked(vcpu, newstate, from_idle); 2076 vcpu_unlock(vcpu); 2077 2078 return (error); 2079 } 2080 2081 enum vcpu_state 2082 vcpu_get_state(struct vm *vm, int vcpuid, int *hostcpu) 2083 { 2084 struct vcpu *vcpu; 2085 enum vcpu_state state; 2086 2087 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 2088 panic("vm_get_run_state: invalid vcpuid %d", vcpuid); 2089 2090 vcpu = &vm->vcpu[vcpuid]; 2091 2092 vcpu_lock(vcpu); 2093 state = vcpu->state; 2094 if (hostcpu != NULL) 2095 *hostcpu = vcpu->hostcpu; 2096 vcpu_unlock(vcpu); 2097 2098 return (state); 2099 } 2100 2101 int 2102 vm_activate_cpu(struct vm *vm, int vcpuid) 2103 { 2104 2105 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 2106 return (EINVAL); 2107 2108 if (CPU_ISSET(vcpuid, &vm->active_cpus)) 2109 return (EBUSY); 2110 2111 VCPU_CTR0(vm, vcpuid, "activated"); 2112 CPU_SET_ATOMIC(vcpuid, &vm->active_cpus); 2113 return (0); 2114 } 2115 2116 cpuset_t 2117 vm_active_cpus(struct vm *vm) 2118 { 2119 2120 return (vm->active_cpus); 2121 } 2122 2123 cpuset_t 2124 vm_suspended_cpus(struct vm *vm) 2125 { 2126 2127 return (vm->suspended_cpus); 2128 } 2129 2130 void * 2131 vcpu_stats(struct vm *vm, int vcpuid) 2132 { 2133 2134 return (vm->vcpu[vcpuid].stats); 2135 } 2136 2137 int 2138 vm_get_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state *state) 2139 { 2140 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 2141 return (EINVAL); 2142 2143 *state = vm->vcpu[vcpuid].x2apic_state; 2144 2145 return (0); 2146 } 2147 2148 int 2149 vm_set_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state state) 2150 { 2151 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 2152 return (EINVAL); 2153 2154 if (state >= X2APIC_STATE_LAST) 2155 return (EINVAL); 2156 2157 vm->vcpu[vcpuid].x2apic_state = state; 2158 2159 vlapic_set_x2apic_state(vm, vcpuid, state); 2160 2161 return (0); 2162 } 2163 2164 /* 2165 * This function is called to ensure that a vcpu "sees" a pending event 2166 * as soon as possible: 2167 * - If the vcpu thread is sleeping then it is woken up. 2168 * - If the vcpu is running on a different host_cpu then an IPI will be directed 2169 * to the host_cpu to cause the vcpu to trap into the hypervisor. 2170 */ 2171 void 2172 vcpu_notify_event(struct vm *vm, int vcpuid, bool lapic_intr) 2173 { 2174 int hostcpu; 2175 struct vcpu *vcpu; 2176 2177 vcpu = &vm->vcpu[vcpuid]; 2178 2179 vcpu_lock(vcpu); 2180 hostcpu = vcpu->hostcpu; 2181 if (vcpu->state == VCPU_RUNNING) { 2182 KASSERT(hostcpu != NOCPU, ("vcpu running on invalid hostcpu")); 2183 if (hostcpu != curcpu) { 2184 if (lapic_intr) { 2185 vlapic_post_intr(vcpu->vlapic, hostcpu, 2186 vmm_ipinum); 2187 } else { 2188 ipi_cpu(hostcpu, vmm_ipinum); 2189 } 2190 } else { 2191 /* 2192 * If the 'vcpu' is running on 'curcpu' then it must 2193 * be sending a notification to itself (e.g. SELF_IPI). 2194 * The pending event will be picked up when the vcpu 2195 * transitions back to guest context. 2196 */ 2197 } 2198 } else { 2199 KASSERT(hostcpu == NOCPU, ("vcpu state %d not consistent " 2200 "with hostcpu %d", vcpu->state, hostcpu)); 2201 if (vcpu->state == VCPU_SLEEPING) 2202 wakeup_one(vcpu); 2203 } 2204 vcpu_unlock(vcpu); 2205 } 2206 2207 struct vmspace * 2208 vm_get_vmspace(struct vm *vm) 2209 { 2210 2211 return (vm->vmspace); 2212 } 2213 2214 int 2215 vm_apicid2vcpuid(struct vm *vm, int apicid) 2216 { 2217 /* 2218 * XXX apic id is assumed to be numerically identical to vcpu id 2219 */ 2220 return (apicid); 2221 } 2222 2223 void 2224 vm_smp_rendezvous(struct vm *vm, int vcpuid, cpuset_t dest, 2225 vm_rendezvous_func_t func, void *arg) 2226 { 2227 int i; 2228 2229 /* 2230 * Enforce that this function is called without any locks 2231 */ 2232 WITNESS_WARN(WARN_PANIC, NULL, "vm_smp_rendezvous"); 2233 KASSERT(vcpuid == -1 || (vcpuid >= 0 && vcpuid < VM_MAXCPU), 2234 ("vm_smp_rendezvous: invalid vcpuid %d", vcpuid)); 2235 2236 restart: 2237 mtx_lock(&vm->rendezvous_mtx); 2238 if (vm->rendezvous_func != NULL) { 2239 /* 2240 * If a rendezvous is already in progress then we need to 2241 * call the rendezvous handler in case this 'vcpuid' is one 2242 * of the targets of the rendezvous. 2243 */ 2244 RENDEZVOUS_CTR0(vm, vcpuid, "Rendezvous already in progress"); 2245 mtx_unlock(&vm->rendezvous_mtx); 2246 vm_handle_rendezvous(vm, vcpuid); 2247 goto restart; 2248 } 2249 KASSERT(vm->rendezvous_func == NULL, ("vm_smp_rendezvous: previous " 2250 "rendezvous is still in progress")); 2251 2252 RENDEZVOUS_CTR0(vm, vcpuid, "Initiating rendezvous"); 2253 vm->rendezvous_req_cpus = dest; 2254 CPU_ZERO(&vm->rendezvous_done_cpus); 2255 vm->rendezvous_arg = arg; 2256 vm_set_rendezvous_func(vm, func); 2257 mtx_unlock(&vm->rendezvous_mtx); 2258 2259 /* 2260 * Wake up any sleeping vcpus and trigger a VM-exit in any running 2261 * vcpus so they handle the rendezvous as soon as possible. 2262 */ 2263 for (i = 0; i < VM_MAXCPU; i++) { 2264 if (CPU_ISSET(i, &dest)) 2265 vcpu_notify_event(vm, i, false); 2266 } 2267 2268 vm_handle_rendezvous(vm, vcpuid); 2269 } 2270 2271 struct vatpic * 2272 vm_atpic(struct vm *vm) 2273 { 2274 return (vm->vatpic); 2275 } 2276 2277 struct vatpit * 2278 vm_atpit(struct vm *vm) 2279 { 2280 return (vm->vatpit); 2281 } 2282 2283 struct vpmtmr * 2284 vm_pmtmr(struct vm *vm) 2285 { 2286 2287 return (vm->vpmtmr); 2288 } 2289 2290 struct vrtc * 2291 vm_rtc(struct vm *vm) 2292 { 2293 2294 return (vm->vrtc); 2295 } 2296 2297 enum vm_reg_name 2298 vm_segment_name(int seg) 2299 { 2300 static enum vm_reg_name seg_names[] = { 2301 VM_REG_GUEST_ES, 2302 VM_REG_GUEST_CS, 2303 VM_REG_GUEST_SS, 2304 VM_REG_GUEST_DS, 2305 VM_REG_GUEST_FS, 2306 VM_REG_GUEST_GS 2307 }; 2308 2309 KASSERT(seg >= 0 && seg < nitems(seg_names), 2310 ("%s: invalid segment encoding %d", __func__, seg)); 2311 return (seg_names[seg]); 2312 } 2313 2314 void 2315 vm_copy_teardown(struct vm *vm, int vcpuid, struct vm_copyinfo *copyinfo, 2316 int num_copyinfo) 2317 { 2318 int idx; 2319 2320 for (idx = 0; idx < num_copyinfo; idx++) { 2321 if (copyinfo[idx].cookie != NULL) 2322 vm_gpa_release(copyinfo[idx].cookie); 2323 } 2324 bzero(copyinfo, num_copyinfo * sizeof(struct vm_copyinfo)); 2325 } 2326 2327 int 2328 vm_copy_setup(struct vm *vm, int vcpuid, struct vm_guest_paging *paging, 2329 uint64_t gla, size_t len, int prot, struct vm_copyinfo *copyinfo, 2330 int num_copyinfo, int *fault) 2331 { 2332 int error, idx, nused; 2333 size_t n, off, remaining; 2334 void *hva, *cookie; 2335 uint64_t gpa; 2336 2337 bzero(copyinfo, sizeof(struct vm_copyinfo) * num_copyinfo); 2338 2339 nused = 0; 2340 remaining = len; 2341 while (remaining > 0) { 2342 KASSERT(nused < num_copyinfo, ("insufficient vm_copyinfo")); 2343 error = vm_gla2gpa(vm, vcpuid, paging, gla, prot, &gpa, fault); 2344 if (error || *fault) 2345 return (error); 2346 off = gpa & PAGE_MASK; 2347 n = min(remaining, PAGE_SIZE - off); 2348 copyinfo[nused].gpa = gpa; 2349 copyinfo[nused].len = n; 2350 remaining -= n; 2351 gla += n; 2352 nused++; 2353 } 2354 2355 for (idx = 0; idx < nused; idx++) { 2356 hva = vm_gpa_hold(vm, copyinfo[idx].gpa, copyinfo[idx].len, 2357 prot, &cookie); 2358 if (hva == NULL) 2359 break; 2360 copyinfo[idx].hva = hva; 2361 copyinfo[idx].cookie = cookie; 2362 } 2363 2364 if (idx != nused) { 2365 vm_copy_teardown(vm, vcpuid, copyinfo, num_copyinfo); 2366 return (EFAULT); 2367 } else { 2368 *fault = 0; 2369 return (0); 2370 } 2371 } 2372 2373 void 2374 vm_copyin(struct vm *vm, int vcpuid, struct vm_copyinfo *copyinfo, void *kaddr, 2375 size_t len) 2376 { 2377 char *dst; 2378 int idx; 2379 2380 dst = kaddr; 2381 idx = 0; 2382 while (len > 0) { 2383 bcopy(copyinfo[idx].hva, dst, copyinfo[idx].len); 2384 len -= copyinfo[idx].len; 2385 dst += copyinfo[idx].len; 2386 idx++; 2387 } 2388 } 2389 2390 void 2391 vm_copyout(struct vm *vm, int vcpuid, const void *kaddr, 2392 struct vm_copyinfo *copyinfo, size_t len) 2393 { 2394 const char *src; 2395 int idx; 2396 2397 src = kaddr; 2398 idx = 0; 2399 while (len > 0) { 2400 bcopy(src, copyinfo[idx].hva, copyinfo[idx].len); 2401 len -= copyinfo[idx].len; 2402 src += copyinfo[idx].len; 2403 idx++; 2404 } 2405 } 2406 2407 /* 2408 * Return the amount of in-use and wired memory for the VM. Since 2409 * these are global stats, only return the values with for vCPU 0 2410 */ 2411 VMM_STAT_DECLARE(VMM_MEM_RESIDENT); 2412 VMM_STAT_DECLARE(VMM_MEM_WIRED); 2413 2414 static void 2415 vm_get_rescnt(struct vm *vm, int vcpu, struct vmm_stat_type *stat) 2416 { 2417 2418 if (vcpu == 0) { 2419 vmm_stat_set(vm, vcpu, VMM_MEM_RESIDENT, 2420 PAGE_SIZE * vmspace_resident_count(vm->vmspace)); 2421 } 2422 } 2423 2424 static void 2425 vm_get_wiredcnt(struct vm *vm, int vcpu, struct vmm_stat_type *stat) 2426 { 2427 2428 if (vcpu == 0) { 2429 vmm_stat_set(vm, vcpu, VMM_MEM_WIRED, 2430 PAGE_SIZE * pmap_wired_count(vmspace_pmap(vm->vmspace))); 2431 } 2432 } 2433 2434 VMM_STAT_FUNC(VMM_MEM_RESIDENT, "Resident memory", vm_get_rescnt); 2435 VMM_STAT_FUNC(VMM_MEM_WIRED, "Wired memory", vm_get_wiredcnt); 2436