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 "vmm_msr.h" 78 #include "vmm_ipi.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 int nmi_pending; /* (i) NMI pending */ 101 int extint_pending; /* (i) INTR pending */ 102 struct vm_exception exception; /* (x) exception collateral */ 103 int exception_pending; /* (i) exception pending */ 104 struct savefpu *guestfpu; /* (a,i) guest fpu state */ 105 uint64_t guest_xcr0; /* (i) guest %xcr0 register */ 106 void *stats; /* (a,i) statistics */ 107 uint64_t guest_msrs[VMM_MSR_NUM]; /* (i) emulated MSRs */ 108 struct vm_exit exitinfo; /* (x) exit reason and collateral */ 109 }; 110 111 #define vcpu_lock_initialized(v) mtx_initialized(&((v)->mtx)) 112 #define vcpu_lock_init(v) mtx_init(&((v)->mtx), "vcpu lock", 0, MTX_SPIN) 113 #define vcpu_lock(v) mtx_lock_spin(&((v)->mtx)) 114 #define vcpu_unlock(v) mtx_unlock_spin(&((v)->mtx)) 115 #define vcpu_assert_locked(v) mtx_assert(&((v)->mtx), MA_OWNED) 116 117 struct mem_seg { 118 vm_paddr_t gpa; 119 size_t len; 120 boolean_t wired; 121 vm_object_t object; 122 }; 123 #define VM_MAX_MEMORY_SEGMENTS 2 124 125 /* 126 * Initialization: 127 * (o) initialized the first time the VM is created 128 * (i) initialized when VM is created and when it is reinitialized 129 * (x) initialized before use 130 */ 131 struct vm { 132 void *cookie; /* (i) cpu-specific data */ 133 void *iommu; /* (x) iommu-specific data */ 134 struct vhpet *vhpet; /* (i) virtual HPET */ 135 struct vioapic *vioapic; /* (i) virtual ioapic */ 136 struct vatpic *vatpic; /* (i) virtual atpic */ 137 struct vatpit *vatpit; /* (i) virtual atpit */ 138 volatile cpuset_t active_cpus; /* (i) active vcpus */ 139 int suspend; /* (i) stop VM execution */ 140 volatile cpuset_t suspended_cpus; /* (i) suspended vcpus */ 141 volatile cpuset_t halted_cpus; /* (x) cpus in a hard halt */ 142 cpuset_t rendezvous_req_cpus; /* (x) rendezvous requested */ 143 cpuset_t rendezvous_done_cpus; /* (x) rendezvous finished */ 144 void *rendezvous_arg; /* (x) rendezvous func/arg */ 145 vm_rendezvous_func_t rendezvous_func; 146 struct mtx rendezvous_mtx; /* (o) rendezvous lock */ 147 int num_mem_segs; /* (o) guest memory segments */ 148 struct mem_seg mem_segs[VM_MAX_MEMORY_SEGMENTS]; 149 struct vmspace *vmspace; /* (o) guest's address space */ 150 char name[VM_MAX_NAMELEN]; /* (o) virtual machine name */ 151 struct vcpu vcpu[VM_MAXCPU]; /* (i) guest vcpus */ 152 }; 153 154 static int vmm_initialized; 155 156 static struct vmm_ops *ops; 157 #define VMM_INIT(num) (ops != NULL ? (*ops->init)(num) : 0) 158 #define VMM_CLEANUP() (ops != NULL ? (*ops->cleanup)() : 0) 159 #define VMM_RESUME() (ops != NULL ? (*ops->resume)() : 0) 160 161 #define VMINIT(vm, pmap) (ops != NULL ? (*ops->vminit)(vm, pmap): NULL) 162 #define VMRUN(vmi, vcpu, rip, pmap, rptr, sptr) \ 163 (ops != NULL ? (*ops->vmrun)(vmi, vcpu, rip, pmap, rptr, sptr) : ENXIO) 164 #define VMCLEANUP(vmi) (ops != NULL ? (*ops->vmcleanup)(vmi) : NULL) 165 #define VMSPACE_ALLOC(min, max) \ 166 (ops != NULL ? (*ops->vmspace_alloc)(min, max) : NULL) 167 #define VMSPACE_FREE(vmspace) \ 168 (ops != NULL ? (*ops->vmspace_free)(vmspace) : ENXIO) 169 #define VMGETREG(vmi, vcpu, num, retval) \ 170 (ops != NULL ? (*ops->vmgetreg)(vmi, vcpu, num, retval) : ENXIO) 171 #define VMSETREG(vmi, vcpu, num, val) \ 172 (ops != NULL ? (*ops->vmsetreg)(vmi, vcpu, num, val) : ENXIO) 173 #define VMGETDESC(vmi, vcpu, num, desc) \ 174 (ops != NULL ? (*ops->vmgetdesc)(vmi, vcpu, num, desc) : ENXIO) 175 #define VMSETDESC(vmi, vcpu, num, desc) \ 176 (ops != NULL ? (*ops->vmsetdesc)(vmi, vcpu, num, desc) : ENXIO) 177 #define VMGETCAP(vmi, vcpu, num, retval) \ 178 (ops != NULL ? (*ops->vmgetcap)(vmi, vcpu, num, retval) : ENXIO) 179 #define VMSETCAP(vmi, vcpu, num, val) \ 180 (ops != NULL ? (*ops->vmsetcap)(vmi, vcpu, num, val) : ENXIO) 181 #define VLAPIC_INIT(vmi, vcpu) \ 182 (ops != NULL ? (*ops->vlapic_init)(vmi, vcpu) : NULL) 183 #define VLAPIC_CLEANUP(vmi, vlapic) \ 184 (ops != NULL ? (*ops->vlapic_cleanup)(vmi, vlapic) : NULL) 185 186 #define fpu_start_emulating() load_cr0(rcr0() | CR0_TS) 187 #define fpu_stop_emulating() clts() 188 189 static MALLOC_DEFINE(M_VM, "vm", "vm"); 190 CTASSERT(VMM_MSR_NUM <= 64); /* msr_mask can keep track of up to 64 msrs */ 191 192 /* statistics */ 193 static VMM_STAT(VCPU_TOTAL_RUNTIME, "vcpu total runtime"); 194 195 SYSCTL_NODE(_hw, OID_AUTO, vmm, CTLFLAG_RW, NULL, NULL); 196 197 /* 198 * Halt the guest if all vcpus are executing a HLT instruction with 199 * interrupts disabled. 200 */ 201 static int halt_detection_enabled = 1; 202 TUNABLE_INT("hw.vmm.halt_detection", &halt_detection_enabled); 203 SYSCTL_INT(_hw_vmm, OID_AUTO, halt_detection, CTLFLAG_RDTUN, 204 &halt_detection_enabled, 0, 205 "Halt VM if all vcpus execute HLT with interrupts disabled"); 206 207 static int vmm_ipinum; 208 SYSCTL_INT(_hw_vmm, OID_AUTO, ipinum, CTLFLAG_RD, &vmm_ipinum, 0, 209 "IPI vector used for vcpu notifications"); 210 211 static void 212 vcpu_cleanup(struct vm *vm, int i, bool destroy) 213 { 214 struct vcpu *vcpu = &vm->vcpu[i]; 215 216 VLAPIC_CLEANUP(vm->cookie, vcpu->vlapic); 217 if (destroy) { 218 vmm_stat_free(vcpu->stats); 219 fpu_save_area_free(vcpu->guestfpu); 220 } 221 } 222 223 static void 224 vcpu_init(struct vm *vm, int vcpu_id, bool create) 225 { 226 struct vcpu *vcpu; 227 228 KASSERT(vcpu_id >= 0 && vcpu_id < VM_MAXCPU, 229 ("vcpu_init: invalid vcpu %d", vcpu_id)); 230 231 vcpu = &vm->vcpu[vcpu_id]; 232 233 if (create) { 234 KASSERT(!vcpu_lock_initialized(vcpu), ("vcpu %d already " 235 "initialized", vcpu_id)); 236 vcpu_lock_init(vcpu); 237 vcpu->state = VCPU_IDLE; 238 vcpu->hostcpu = NOCPU; 239 vcpu->guestfpu = fpu_save_area_alloc(); 240 vcpu->stats = vmm_stat_alloc(); 241 } 242 243 vcpu->vlapic = VLAPIC_INIT(vm->cookie, vcpu_id); 244 vm_set_x2apic_state(vm, vcpu_id, X2APIC_DISABLED); 245 vcpu->nmi_pending = 0; 246 vcpu->extint_pending = 0; 247 vcpu->exception_pending = 0; 248 vcpu->guest_xcr0 = XFEATURE_ENABLED_X87; 249 fpu_save_area_reset(vcpu->guestfpu); 250 vmm_stat_init(vcpu->stats); 251 guest_msrs_init(vm, vcpu_id); 252 } 253 254 struct vm_exit * 255 vm_exitinfo(struct vm *vm, int cpuid) 256 { 257 struct vcpu *vcpu; 258 259 if (cpuid < 0 || cpuid >= VM_MAXCPU) 260 panic("vm_exitinfo: invalid cpuid %d", cpuid); 261 262 vcpu = &vm->vcpu[cpuid]; 263 264 return (&vcpu->exitinfo); 265 } 266 267 static void 268 vmm_resume(void) 269 { 270 VMM_RESUME(); 271 } 272 273 static int 274 vmm_init(void) 275 { 276 int error; 277 278 vmm_host_state_init(); 279 280 vmm_ipinum = vmm_ipi_alloc(); 281 if (vmm_ipinum == 0) 282 vmm_ipinum = IPI_AST; 283 284 error = vmm_mem_init(); 285 if (error) 286 return (error); 287 288 if (vmm_is_intel()) 289 ops = &vmm_ops_intel; 290 else if (vmm_is_amd()) 291 ops = &vmm_ops_amd; 292 else 293 return (ENXIO); 294 295 vmm_msr_init(); 296 vmm_resume_p = vmm_resume; 297 298 return (VMM_INIT(vmm_ipinum)); 299 } 300 301 static int 302 vmm_handler(module_t mod, int what, void *arg) 303 { 304 int error; 305 306 switch (what) { 307 case MOD_LOAD: 308 vmmdev_init(); 309 if (ppt_avail_devices() > 0) 310 iommu_init(); 311 error = vmm_init(); 312 if (error == 0) 313 vmm_initialized = 1; 314 break; 315 case MOD_UNLOAD: 316 error = vmmdev_cleanup(); 317 if (error == 0) { 318 vmm_resume_p = NULL; 319 iommu_cleanup(); 320 if (vmm_ipinum != IPI_AST) 321 vmm_ipi_free(vmm_ipinum); 322 error = VMM_CLEANUP(); 323 /* 324 * Something bad happened - prevent new 325 * VMs from being created 326 */ 327 if (error) 328 vmm_initialized = 0; 329 } 330 break; 331 default: 332 error = 0; 333 break; 334 } 335 return (error); 336 } 337 338 static moduledata_t vmm_kmod = { 339 "vmm", 340 vmm_handler, 341 NULL 342 }; 343 344 /* 345 * vmm initialization has the following dependencies: 346 * 347 * - iommu initialization must happen after the pci passthru driver has had 348 * a chance to attach to any passthru devices (after SI_SUB_CONFIGURE). 349 * 350 * - VT-x initialization requires smp_rendezvous() and therefore must happen 351 * after SMP is fully functional (after SI_SUB_SMP). 352 */ 353 DECLARE_MODULE(vmm, vmm_kmod, SI_SUB_SMP + 1, SI_ORDER_ANY); 354 MODULE_VERSION(vmm, 1); 355 356 static void 357 vm_init(struct vm *vm, bool create) 358 { 359 int i; 360 361 vm->cookie = VMINIT(vm, vmspace_pmap(vm->vmspace)); 362 vm->iommu = NULL; 363 vm->vioapic = vioapic_init(vm); 364 vm->vhpet = vhpet_init(vm); 365 vm->vatpic = vatpic_init(vm); 366 vm->vatpit = vatpit_init(vm); 367 368 CPU_ZERO(&vm->active_cpus); 369 370 vm->suspend = 0; 371 CPU_ZERO(&vm->suspended_cpus); 372 373 for (i = 0; i < VM_MAXCPU; i++) 374 vcpu_init(vm, i, create); 375 } 376 377 int 378 vm_create(const char *name, struct vm **retvm) 379 { 380 struct vm *vm; 381 struct vmspace *vmspace; 382 383 /* 384 * If vmm.ko could not be successfully initialized then don't attempt 385 * to create the virtual machine. 386 */ 387 if (!vmm_initialized) 388 return (ENXIO); 389 390 if (name == NULL || strlen(name) >= VM_MAX_NAMELEN) 391 return (EINVAL); 392 393 vmspace = VMSPACE_ALLOC(VM_MIN_ADDRESS, VM_MAXUSER_ADDRESS); 394 if (vmspace == NULL) 395 return (ENOMEM); 396 397 vm = malloc(sizeof(struct vm), M_VM, M_WAITOK | M_ZERO); 398 strcpy(vm->name, name); 399 vm->num_mem_segs = 0; 400 vm->vmspace = vmspace; 401 mtx_init(&vm->rendezvous_mtx, "vm rendezvous lock", 0, MTX_DEF); 402 403 vm_init(vm, true); 404 405 *retvm = vm; 406 return (0); 407 } 408 409 static void 410 vm_free_mem_seg(struct vm *vm, struct mem_seg *seg) 411 { 412 413 if (seg->object != NULL) 414 vmm_mem_free(vm->vmspace, seg->gpa, seg->len); 415 416 bzero(seg, sizeof(*seg)); 417 } 418 419 static void 420 vm_cleanup(struct vm *vm, bool destroy) 421 { 422 int i; 423 424 ppt_unassign_all(vm); 425 426 if (vm->iommu != NULL) 427 iommu_destroy_domain(vm->iommu); 428 429 vatpit_cleanup(vm->vatpit); 430 vhpet_cleanup(vm->vhpet); 431 vatpic_cleanup(vm->vatpic); 432 vioapic_cleanup(vm->vioapic); 433 434 for (i = 0; i < VM_MAXCPU; i++) 435 vcpu_cleanup(vm, i, destroy); 436 437 VMCLEANUP(vm->cookie); 438 439 if (destroy) { 440 for (i = 0; i < vm->num_mem_segs; i++) 441 vm_free_mem_seg(vm, &vm->mem_segs[i]); 442 443 vm->num_mem_segs = 0; 444 445 VMSPACE_FREE(vm->vmspace); 446 vm->vmspace = NULL; 447 } 448 } 449 450 void 451 vm_destroy(struct vm *vm) 452 { 453 vm_cleanup(vm, true); 454 free(vm, M_VM); 455 } 456 457 int 458 vm_reinit(struct vm *vm) 459 { 460 int error; 461 462 /* 463 * A virtual machine can be reset only if all vcpus are suspended. 464 */ 465 if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) { 466 vm_cleanup(vm, false); 467 vm_init(vm, false); 468 error = 0; 469 } else { 470 error = EBUSY; 471 } 472 473 return (error); 474 } 475 476 const char * 477 vm_name(struct vm *vm) 478 { 479 return (vm->name); 480 } 481 482 int 483 vm_map_mmio(struct vm *vm, vm_paddr_t gpa, size_t len, vm_paddr_t hpa) 484 { 485 vm_object_t obj; 486 487 if ((obj = vmm_mmio_alloc(vm->vmspace, gpa, len, hpa)) == NULL) 488 return (ENOMEM); 489 else 490 return (0); 491 } 492 493 int 494 vm_unmap_mmio(struct vm *vm, vm_paddr_t gpa, size_t len) 495 { 496 497 vmm_mmio_free(vm->vmspace, gpa, len); 498 return (0); 499 } 500 501 boolean_t 502 vm_mem_allocated(struct vm *vm, vm_paddr_t gpa) 503 { 504 int i; 505 vm_paddr_t gpabase, gpalimit; 506 507 for (i = 0; i < vm->num_mem_segs; i++) { 508 gpabase = vm->mem_segs[i].gpa; 509 gpalimit = gpabase + vm->mem_segs[i].len; 510 if (gpa >= gpabase && gpa < gpalimit) 511 return (TRUE); /* 'gpa' is regular memory */ 512 } 513 514 if (ppt_is_mmio(vm, gpa)) 515 return (TRUE); /* 'gpa' is pci passthru mmio */ 516 517 return (FALSE); 518 } 519 520 int 521 vm_malloc(struct vm *vm, vm_paddr_t gpa, size_t len) 522 { 523 int available, allocated; 524 struct mem_seg *seg; 525 vm_object_t object; 526 vm_paddr_t g; 527 528 if ((gpa & PAGE_MASK) || (len & PAGE_MASK) || len == 0) 529 return (EINVAL); 530 531 available = allocated = 0; 532 g = gpa; 533 while (g < gpa + len) { 534 if (vm_mem_allocated(vm, g)) 535 allocated++; 536 else 537 available++; 538 539 g += PAGE_SIZE; 540 } 541 542 /* 543 * If there are some allocated and some available pages in the address 544 * range then it is an error. 545 */ 546 if (allocated && available) 547 return (EINVAL); 548 549 /* 550 * If the entire address range being requested has already been 551 * allocated then there isn't anything more to do. 552 */ 553 if (allocated && available == 0) 554 return (0); 555 556 if (vm->num_mem_segs >= VM_MAX_MEMORY_SEGMENTS) 557 return (E2BIG); 558 559 seg = &vm->mem_segs[vm->num_mem_segs]; 560 561 if ((object = vmm_mem_alloc(vm->vmspace, gpa, len)) == NULL) 562 return (ENOMEM); 563 564 seg->gpa = gpa; 565 seg->len = len; 566 seg->object = object; 567 seg->wired = FALSE; 568 569 vm->num_mem_segs++; 570 571 return (0); 572 } 573 574 static void 575 vm_gpa_unwire(struct vm *vm) 576 { 577 int i, rv; 578 struct mem_seg *seg; 579 580 for (i = 0; i < vm->num_mem_segs; i++) { 581 seg = &vm->mem_segs[i]; 582 if (!seg->wired) 583 continue; 584 585 rv = vm_map_unwire(&vm->vmspace->vm_map, 586 seg->gpa, seg->gpa + seg->len, 587 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES); 588 KASSERT(rv == KERN_SUCCESS, ("vm(%s) memory segment " 589 "%#lx/%ld could not be unwired: %d", 590 vm_name(vm), seg->gpa, seg->len, rv)); 591 592 seg->wired = FALSE; 593 } 594 } 595 596 static int 597 vm_gpa_wire(struct vm *vm) 598 { 599 int i, rv; 600 struct mem_seg *seg; 601 602 for (i = 0; i < vm->num_mem_segs; i++) { 603 seg = &vm->mem_segs[i]; 604 if (seg->wired) 605 continue; 606 607 /* XXX rlimits? */ 608 rv = vm_map_wire(&vm->vmspace->vm_map, 609 seg->gpa, seg->gpa + seg->len, 610 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES); 611 if (rv != KERN_SUCCESS) 612 break; 613 614 seg->wired = TRUE; 615 } 616 617 if (i < vm->num_mem_segs) { 618 /* 619 * Undo the wiring before returning an error. 620 */ 621 vm_gpa_unwire(vm); 622 return (EAGAIN); 623 } 624 625 return (0); 626 } 627 628 static void 629 vm_iommu_modify(struct vm *vm, boolean_t map) 630 { 631 int i, sz; 632 vm_paddr_t gpa, hpa; 633 struct mem_seg *seg; 634 void *vp, *cookie, *host_domain; 635 636 sz = PAGE_SIZE; 637 host_domain = iommu_host_domain(); 638 639 for (i = 0; i < vm->num_mem_segs; i++) { 640 seg = &vm->mem_segs[i]; 641 KASSERT(seg->wired, ("vm(%s) memory segment %#lx/%ld not wired", 642 vm_name(vm), seg->gpa, seg->len)); 643 644 gpa = seg->gpa; 645 while (gpa < seg->gpa + seg->len) { 646 vp = vm_gpa_hold(vm, gpa, PAGE_SIZE, VM_PROT_WRITE, 647 &cookie); 648 KASSERT(vp != NULL, ("vm(%s) could not map gpa %#lx", 649 vm_name(vm), gpa)); 650 651 vm_gpa_release(cookie); 652 653 hpa = DMAP_TO_PHYS((uintptr_t)vp); 654 if (map) { 655 iommu_create_mapping(vm->iommu, gpa, hpa, sz); 656 iommu_remove_mapping(host_domain, hpa, sz); 657 } else { 658 iommu_remove_mapping(vm->iommu, gpa, sz); 659 iommu_create_mapping(host_domain, hpa, hpa, sz); 660 } 661 662 gpa += PAGE_SIZE; 663 } 664 } 665 666 /* 667 * Invalidate the cached translations associated with the domain 668 * from which pages were removed. 669 */ 670 if (map) 671 iommu_invalidate_tlb(host_domain); 672 else 673 iommu_invalidate_tlb(vm->iommu); 674 } 675 676 #define vm_iommu_unmap(vm) vm_iommu_modify((vm), FALSE) 677 #define vm_iommu_map(vm) vm_iommu_modify((vm), TRUE) 678 679 int 680 vm_unassign_pptdev(struct vm *vm, int bus, int slot, int func) 681 { 682 int error; 683 684 error = ppt_unassign_device(vm, bus, slot, func); 685 if (error) 686 return (error); 687 688 if (ppt_assigned_devices(vm) == 0) { 689 vm_iommu_unmap(vm); 690 vm_gpa_unwire(vm); 691 } 692 return (0); 693 } 694 695 int 696 vm_assign_pptdev(struct vm *vm, int bus, int slot, int func) 697 { 698 int error; 699 vm_paddr_t maxaddr; 700 701 /* 702 * Virtual machines with pci passthru devices get special treatment: 703 * - the guest physical memory is wired 704 * - the iommu is programmed to do the 'gpa' to 'hpa' translation 705 * 706 * We need to do this before the first pci passthru device is attached. 707 */ 708 if (ppt_assigned_devices(vm) == 0) { 709 KASSERT(vm->iommu == NULL, 710 ("vm_assign_pptdev: iommu must be NULL")); 711 maxaddr = vmm_mem_maxaddr(); 712 vm->iommu = iommu_create_domain(maxaddr); 713 714 error = vm_gpa_wire(vm); 715 if (error) 716 return (error); 717 718 vm_iommu_map(vm); 719 } 720 721 error = ppt_assign_device(vm, bus, slot, func); 722 return (error); 723 } 724 725 void * 726 vm_gpa_hold(struct vm *vm, vm_paddr_t gpa, size_t len, int reqprot, 727 void **cookie) 728 { 729 int count, pageoff; 730 vm_page_t m; 731 732 pageoff = gpa & PAGE_MASK; 733 if (len > PAGE_SIZE - pageoff) 734 panic("vm_gpa_hold: invalid gpa/len: 0x%016lx/%lu", gpa, len); 735 736 count = vm_fault_quick_hold_pages(&vm->vmspace->vm_map, 737 trunc_page(gpa), PAGE_SIZE, reqprot, &m, 1); 738 739 if (count == 1) { 740 *cookie = m; 741 return ((void *)(PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m)) + pageoff)); 742 } else { 743 *cookie = NULL; 744 return (NULL); 745 } 746 } 747 748 void 749 vm_gpa_release(void *cookie) 750 { 751 vm_page_t m = cookie; 752 753 vm_page_lock(m); 754 vm_page_unhold(m); 755 vm_page_unlock(m); 756 } 757 758 int 759 vm_gpabase2memseg(struct vm *vm, vm_paddr_t gpabase, 760 struct vm_memory_segment *seg) 761 { 762 int i; 763 764 for (i = 0; i < vm->num_mem_segs; i++) { 765 if (gpabase == vm->mem_segs[i].gpa) { 766 seg->gpa = vm->mem_segs[i].gpa; 767 seg->len = vm->mem_segs[i].len; 768 seg->wired = vm->mem_segs[i].wired; 769 return (0); 770 } 771 } 772 return (-1); 773 } 774 775 int 776 vm_get_memobj(struct vm *vm, vm_paddr_t gpa, size_t len, 777 vm_offset_t *offset, struct vm_object **object) 778 { 779 int i; 780 size_t seg_len; 781 vm_paddr_t seg_gpa; 782 vm_object_t seg_obj; 783 784 for (i = 0; i < vm->num_mem_segs; i++) { 785 if ((seg_obj = vm->mem_segs[i].object) == NULL) 786 continue; 787 788 seg_gpa = vm->mem_segs[i].gpa; 789 seg_len = vm->mem_segs[i].len; 790 791 if (gpa >= seg_gpa && gpa < seg_gpa + seg_len) { 792 *offset = gpa - seg_gpa; 793 *object = seg_obj; 794 vm_object_reference(seg_obj); 795 return (0); 796 } 797 } 798 799 return (EINVAL); 800 } 801 802 int 803 vm_get_register(struct vm *vm, int vcpu, int reg, uint64_t *retval) 804 { 805 806 if (vcpu < 0 || vcpu >= VM_MAXCPU) 807 return (EINVAL); 808 809 if (reg >= VM_REG_LAST) 810 return (EINVAL); 811 812 return (VMGETREG(vm->cookie, vcpu, reg, retval)); 813 } 814 815 int 816 vm_set_register(struct vm *vm, int vcpu, int reg, uint64_t val) 817 { 818 819 if (vcpu < 0 || vcpu >= VM_MAXCPU) 820 return (EINVAL); 821 822 if (reg >= VM_REG_LAST) 823 return (EINVAL); 824 825 return (VMSETREG(vm->cookie, vcpu, reg, val)); 826 } 827 828 static boolean_t 829 is_descriptor_table(int reg) 830 { 831 832 switch (reg) { 833 case VM_REG_GUEST_IDTR: 834 case VM_REG_GUEST_GDTR: 835 return (TRUE); 836 default: 837 return (FALSE); 838 } 839 } 840 841 static boolean_t 842 is_segment_register(int reg) 843 { 844 845 switch (reg) { 846 case VM_REG_GUEST_ES: 847 case VM_REG_GUEST_CS: 848 case VM_REG_GUEST_SS: 849 case VM_REG_GUEST_DS: 850 case VM_REG_GUEST_FS: 851 case VM_REG_GUEST_GS: 852 case VM_REG_GUEST_TR: 853 case VM_REG_GUEST_LDTR: 854 return (TRUE); 855 default: 856 return (FALSE); 857 } 858 } 859 860 int 861 vm_get_seg_desc(struct vm *vm, int vcpu, int reg, 862 struct seg_desc *desc) 863 { 864 865 if (vcpu < 0 || vcpu >= VM_MAXCPU) 866 return (EINVAL); 867 868 if (!is_segment_register(reg) && !is_descriptor_table(reg)) 869 return (EINVAL); 870 871 return (VMGETDESC(vm->cookie, vcpu, reg, desc)); 872 } 873 874 int 875 vm_set_seg_desc(struct vm *vm, int vcpu, int reg, 876 struct seg_desc *desc) 877 { 878 if (vcpu < 0 || vcpu >= VM_MAXCPU) 879 return (EINVAL); 880 881 if (!is_segment_register(reg) && !is_descriptor_table(reg)) 882 return (EINVAL); 883 884 return (VMSETDESC(vm->cookie, vcpu, reg, desc)); 885 } 886 887 static void 888 restore_guest_fpustate(struct vcpu *vcpu) 889 { 890 891 /* flush host state to the pcb */ 892 fpuexit(curthread); 893 894 /* restore guest FPU state */ 895 fpu_stop_emulating(); 896 fpurestore(vcpu->guestfpu); 897 898 /* restore guest XCR0 if XSAVE is enabled in the host */ 899 if (rcr4() & CR4_XSAVE) 900 load_xcr(0, vcpu->guest_xcr0); 901 902 /* 903 * The FPU is now "dirty" with the guest's state so turn on emulation 904 * to trap any access to the FPU by the host. 905 */ 906 fpu_start_emulating(); 907 } 908 909 static void 910 save_guest_fpustate(struct vcpu *vcpu) 911 { 912 913 if ((rcr0() & CR0_TS) == 0) 914 panic("fpu emulation not enabled in host!"); 915 916 /* save guest XCR0 and restore host XCR0 */ 917 if (rcr4() & CR4_XSAVE) { 918 vcpu->guest_xcr0 = rxcr(0); 919 load_xcr(0, vmm_get_host_xcr0()); 920 } 921 922 /* save guest FPU state */ 923 fpu_stop_emulating(); 924 fpusave(vcpu->guestfpu); 925 fpu_start_emulating(); 926 } 927 928 static VMM_STAT(VCPU_IDLE_TICKS, "number of ticks vcpu was idle"); 929 930 static int 931 vcpu_set_state_locked(struct vcpu *vcpu, enum vcpu_state newstate, 932 bool from_idle) 933 { 934 int error; 935 936 vcpu_assert_locked(vcpu); 937 938 /* 939 * State transitions from the vmmdev_ioctl() must always begin from 940 * the VCPU_IDLE state. This guarantees that there is only a single 941 * ioctl() operating on a vcpu at any point. 942 */ 943 if (from_idle) { 944 while (vcpu->state != VCPU_IDLE) 945 msleep_spin(&vcpu->state, &vcpu->mtx, "vmstat", hz); 946 } else { 947 KASSERT(vcpu->state != VCPU_IDLE, ("invalid transition from " 948 "vcpu idle state")); 949 } 950 951 if (vcpu->state == VCPU_RUNNING) { 952 KASSERT(vcpu->hostcpu == curcpu, ("curcpu %d and hostcpu %d " 953 "mismatch for running vcpu", curcpu, vcpu->hostcpu)); 954 } else { 955 KASSERT(vcpu->hostcpu == NOCPU, ("Invalid hostcpu %d for a " 956 "vcpu that is not running", vcpu->hostcpu)); 957 } 958 959 /* 960 * The following state transitions are allowed: 961 * IDLE -> FROZEN -> IDLE 962 * FROZEN -> RUNNING -> FROZEN 963 * FROZEN -> SLEEPING -> FROZEN 964 */ 965 switch (vcpu->state) { 966 case VCPU_IDLE: 967 case VCPU_RUNNING: 968 case VCPU_SLEEPING: 969 error = (newstate != VCPU_FROZEN); 970 break; 971 case VCPU_FROZEN: 972 error = (newstate == VCPU_FROZEN); 973 break; 974 default: 975 error = 1; 976 break; 977 } 978 979 if (error) 980 return (EBUSY); 981 982 vcpu->state = newstate; 983 if (newstate == VCPU_RUNNING) 984 vcpu->hostcpu = curcpu; 985 else 986 vcpu->hostcpu = NOCPU; 987 988 if (newstate == VCPU_IDLE) 989 wakeup(&vcpu->state); 990 991 return (0); 992 } 993 994 static void 995 vcpu_require_state(struct vm *vm, int vcpuid, enum vcpu_state newstate) 996 { 997 int error; 998 999 if ((error = vcpu_set_state(vm, vcpuid, newstate, false)) != 0) 1000 panic("Error %d setting state to %d\n", error, newstate); 1001 } 1002 1003 static void 1004 vcpu_require_state_locked(struct vcpu *vcpu, enum vcpu_state newstate) 1005 { 1006 int error; 1007 1008 if ((error = vcpu_set_state_locked(vcpu, newstate, false)) != 0) 1009 panic("Error %d setting state to %d", error, newstate); 1010 } 1011 1012 static void 1013 vm_set_rendezvous_func(struct vm *vm, vm_rendezvous_func_t func) 1014 { 1015 1016 KASSERT(mtx_owned(&vm->rendezvous_mtx), ("rendezvous_mtx not locked")); 1017 1018 /* 1019 * Update 'rendezvous_func' and execute a write memory barrier to 1020 * ensure that it is visible across all host cpus. This is not needed 1021 * for correctness but it does ensure that all the vcpus will notice 1022 * that the rendezvous is requested immediately. 1023 */ 1024 vm->rendezvous_func = func; 1025 wmb(); 1026 } 1027 1028 #define RENDEZVOUS_CTR0(vm, vcpuid, fmt) \ 1029 do { \ 1030 if (vcpuid >= 0) \ 1031 VCPU_CTR0(vm, vcpuid, fmt); \ 1032 else \ 1033 VM_CTR0(vm, fmt); \ 1034 } while (0) 1035 1036 static void 1037 vm_handle_rendezvous(struct vm *vm, int vcpuid) 1038 { 1039 1040 KASSERT(vcpuid == -1 || (vcpuid >= 0 && vcpuid < VM_MAXCPU), 1041 ("vm_handle_rendezvous: invalid vcpuid %d", vcpuid)); 1042 1043 mtx_lock(&vm->rendezvous_mtx); 1044 while (vm->rendezvous_func != NULL) { 1045 /* 'rendezvous_req_cpus' must be a subset of 'active_cpus' */ 1046 CPU_AND(&vm->rendezvous_req_cpus, &vm->active_cpus); 1047 1048 if (vcpuid != -1 && 1049 CPU_ISSET(vcpuid, &vm->rendezvous_req_cpus) && 1050 !CPU_ISSET(vcpuid, &vm->rendezvous_done_cpus)) { 1051 VCPU_CTR0(vm, vcpuid, "Calling rendezvous func"); 1052 (*vm->rendezvous_func)(vm, vcpuid, vm->rendezvous_arg); 1053 CPU_SET(vcpuid, &vm->rendezvous_done_cpus); 1054 } 1055 if (CPU_CMP(&vm->rendezvous_req_cpus, 1056 &vm->rendezvous_done_cpus) == 0) { 1057 VCPU_CTR0(vm, vcpuid, "Rendezvous completed"); 1058 vm_set_rendezvous_func(vm, NULL); 1059 wakeup(&vm->rendezvous_func); 1060 break; 1061 } 1062 RENDEZVOUS_CTR0(vm, vcpuid, "Wait for rendezvous completion"); 1063 mtx_sleep(&vm->rendezvous_func, &vm->rendezvous_mtx, 0, 1064 "vmrndv", 0); 1065 } 1066 mtx_unlock(&vm->rendezvous_mtx); 1067 } 1068 1069 /* 1070 * Emulate a guest 'hlt' by sleeping until the vcpu is ready to run. 1071 */ 1072 static int 1073 vm_handle_hlt(struct vm *vm, int vcpuid, bool intr_disabled, bool *retu) 1074 { 1075 struct vcpu *vcpu; 1076 const char *wmesg; 1077 int t, vcpu_halted, vm_halted; 1078 1079 KASSERT(!CPU_ISSET(vcpuid, &vm->halted_cpus), ("vcpu already halted")); 1080 1081 vcpu = &vm->vcpu[vcpuid]; 1082 vcpu_halted = 0; 1083 vm_halted = 0; 1084 1085 vcpu_lock(vcpu); 1086 while (1) { 1087 /* 1088 * Do a final check for pending NMI or interrupts before 1089 * really putting this thread to sleep. Also check for 1090 * software events that would cause this vcpu to wakeup. 1091 * 1092 * These interrupts/events could have happened after the 1093 * vcpu returned from VMRUN() and before it acquired the 1094 * vcpu lock above. 1095 */ 1096 if (vm->rendezvous_func != NULL || vm->suspend) 1097 break; 1098 if (vm_nmi_pending(vm, vcpuid)) 1099 break; 1100 if (!intr_disabled) { 1101 if (vm_extint_pending(vm, vcpuid) || 1102 vlapic_pending_intr(vcpu->vlapic, NULL)) { 1103 break; 1104 } 1105 } 1106 1107 /* 1108 * Some Linux guests implement "halt" by having all vcpus 1109 * execute HLT with interrupts disabled. 'halted_cpus' keeps 1110 * track of the vcpus that have entered this state. When all 1111 * vcpus enter the halted state the virtual machine is halted. 1112 */ 1113 if (intr_disabled) { 1114 wmesg = "vmhalt"; 1115 VCPU_CTR0(vm, vcpuid, "Halted"); 1116 if (!vcpu_halted && halt_detection_enabled) { 1117 vcpu_halted = 1; 1118 CPU_SET_ATOMIC(vcpuid, &vm->halted_cpus); 1119 } 1120 if (CPU_CMP(&vm->halted_cpus, &vm->active_cpus) == 0) { 1121 vm_halted = 1; 1122 break; 1123 } 1124 } else { 1125 wmesg = "vmidle"; 1126 } 1127 1128 t = ticks; 1129 vcpu_require_state_locked(vcpu, VCPU_SLEEPING); 1130 msleep_spin(vcpu, &vcpu->mtx, wmesg, 0); 1131 vcpu_require_state_locked(vcpu, VCPU_FROZEN); 1132 vmm_stat_incr(vm, vcpuid, VCPU_IDLE_TICKS, ticks - t); 1133 } 1134 1135 if (vcpu_halted) 1136 CPU_CLR_ATOMIC(vcpuid, &vm->halted_cpus); 1137 1138 vcpu_unlock(vcpu); 1139 1140 if (vm_halted) 1141 vm_suspend(vm, VM_SUSPEND_HALT); 1142 1143 return (0); 1144 } 1145 1146 static int 1147 vm_handle_paging(struct vm *vm, int vcpuid, bool *retu) 1148 { 1149 int rv, ftype; 1150 struct vm_map *map; 1151 struct vcpu *vcpu; 1152 struct vm_exit *vme; 1153 1154 vcpu = &vm->vcpu[vcpuid]; 1155 vme = &vcpu->exitinfo; 1156 1157 ftype = vme->u.paging.fault_type; 1158 KASSERT(ftype == VM_PROT_READ || 1159 ftype == VM_PROT_WRITE || ftype == VM_PROT_EXECUTE, 1160 ("vm_handle_paging: invalid fault_type %d", ftype)); 1161 1162 if (ftype == VM_PROT_READ || ftype == VM_PROT_WRITE) { 1163 rv = pmap_emulate_accessed_dirty(vmspace_pmap(vm->vmspace), 1164 vme->u.paging.gpa, ftype); 1165 if (rv == 0) 1166 goto done; 1167 } 1168 1169 map = &vm->vmspace->vm_map; 1170 rv = vm_fault(map, vme->u.paging.gpa, ftype, VM_FAULT_NORMAL); 1171 1172 VCPU_CTR3(vm, vcpuid, "vm_handle_paging rv = %d, gpa = %#lx, " 1173 "ftype = %d", rv, vme->u.paging.gpa, ftype); 1174 1175 if (rv != KERN_SUCCESS) 1176 return (EFAULT); 1177 done: 1178 /* restart execution at the faulting instruction */ 1179 vme->inst_length = 0; 1180 1181 return (0); 1182 } 1183 1184 static int 1185 vm_handle_inst_emul(struct vm *vm, int vcpuid, bool *retu) 1186 { 1187 struct vie *vie; 1188 struct vcpu *vcpu; 1189 struct vm_exit *vme; 1190 uint64_t gla, gpa; 1191 struct vm_guest_paging *paging; 1192 mem_region_read_t mread; 1193 mem_region_write_t mwrite; 1194 int error; 1195 1196 vcpu = &vm->vcpu[vcpuid]; 1197 vme = &vcpu->exitinfo; 1198 1199 gla = vme->u.inst_emul.gla; 1200 gpa = vme->u.inst_emul.gpa; 1201 vie = &vme->u.inst_emul.vie; 1202 paging = &vme->u.inst_emul.paging; 1203 1204 vie_init(vie); 1205 1206 /* Fetch, decode and emulate the faulting instruction */ 1207 error = vmm_fetch_instruction(vm, vcpuid, paging, vme->rip, 1208 vme->inst_length, vie); 1209 if (error == 1) 1210 return (0); /* Resume guest to handle page fault */ 1211 else if (error == -1) 1212 return (EFAULT); 1213 else if (error != 0) 1214 panic("%s: vmm_fetch_instruction error %d", __func__, error); 1215 1216 if (vmm_decode_instruction(vm, vcpuid, gla, paging->cpu_mode, vie) != 0) 1217 return (EFAULT); 1218 1219 /* return to userland unless this is an in-kernel emulated device */ 1220 if (gpa >= DEFAULT_APIC_BASE && gpa < DEFAULT_APIC_BASE + PAGE_SIZE) { 1221 mread = lapic_mmio_read; 1222 mwrite = lapic_mmio_write; 1223 } else if (gpa >= VIOAPIC_BASE && gpa < VIOAPIC_BASE + VIOAPIC_SIZE) { 1224 mread = vioapic_mmio_read; 1225 mwrite = vioapic_mmio_write; 1226 } else if (gpa >= VHPET_BASE && gpa < VHPET_BASE + VHPET_SIZE) { 1227 mread = vhpet_mmio_read; 1228 mwrite = vhpet_mmio_write; 1229 } else { 1230 *retu = true; 1231 return (0); 1232 } 1233 1234 error = vmm_emulate_instruction(vm, vcpuid, gpa, vie, mread, mwrite, 1235 retu); 1236 1237 return (error); 1238 } 1239 1240 static int 1241 vm_handle_suspend(struct vm *vm, int vcpuid, bool *retu) 1242 { 1243 int i, done; 1244 struct vcpu *vcpu; 1245 1246 done = 0; 1247 vcpu = &vm->vcpu[vcpuid]; 1248 1249 CPU_SET_ATOMIC(vcpuid, &vm->suspended_cpus); 1250 1251 /* 1252 * Wait until all 'active_cpus' have suspended themselves. 1253 * 1254 * Since a VM may be suspended at any time including when one or 1255 * more vcpus are doing a rendezvous we need to call the rendezvous 1256 * handler while we are waiting to prevent a deadlock. 1257 */ 1258 vcpu_lock(vcpu); 1259 while (1) { 1260 if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) { 1261 VCPU_CTR0(vm, vcpuid, "All vcpus suspended"); 1262 break; 1263 } 1264 1265 if (vm->rendezvous_func == NULL) { 1266 VCPU_CTR0(vm, vcpuid, "Sleeping during suspend"); 1267 vcpu_require_state_locked(vcpu, VCPU_SLEEPING); 1268 msleep_spin(vcpu, &vcpu->mtx, "vmsusp", hz); 1269 vcpu_require_state_locked(vcpu, VCPU_FROZEN); 1270 } else { 1271 VCPU_CTR0(vm, vcpuid, "Rendezvous during suspend"); 1272 vcpu_unlock(vcpu); 1273 vm_handle_rendezvous(vm, vcpuid); 1274 vcpu_lock(vcpu); 1275 } 1276 } 1277 vcpu_unlock(vcpu); 1278 1279 /* 1280 * Wakeup the other sleeping vcpus and return to userspace. 1281 */ 1282 for (i = 0; i < VM_MAXCPU; i++) { 1283 if (CPU_ISSET(i, &vm->suspended_cpus)) { 1284 vcpu_notify_event(vm, i, false); 1285 } 1286 } 1287 1288 *retu = true; 1289 return (0); 1290 } 1291 1292 int 1293 vm_suspend(struct vm *vm, enum vm_suspend_how how) 1294 { 1295 int i; 1296 1297 if (how <= VM_SUSPEND_NONE || how >= VM_SUSPEND_LAST) 1298 return (EINVAL); 1299 1300 if (atomic_cmpset_int(&vm->suspend, 0, how) == 0) { 1301 VM_CTR2(vm, "virtual machine already suspended %d/%d", 1302 vm->suspend, how); 1303 return (EALREADY); 1304 } 1305 1306 VM_CTR1(vm, "virtual machine successfully suspended %d", how); 1307 1308 /* 1309 * Notify all active vcpus that they are now suspended. 1310 */ 1311 for (i = 0; i < VM_MAXCPU; i++) { 1312 if (CPU_ISSET(i, &vm->active_cpus)) 1313 vcpu_notify_event(vm, i, false); 1314 } 1315 1316 return (0); 1317 } 1318 1319 void 1320 vm_exit_suspended(struct vm *vm, int vcpuid, uint64_t rip) 1321 { 1322 struct vm_exit *vmexit; 1323 1324 KASSERT(vm->suspend > VM_SUSPEND_NONE && vm->suspend < VM_SUSPEND_LAST, 1325 ("vm_exit_suspended: invalid suspend type %d", vm->suspend)); 1326 1327 vmexit = vm_exitinfo(vm, vcpuid); 1328 vmexit->rip = rip; 1329 vmexit->inst_length = 0; 1330 vmexit->exitcode = VM_EXITCODE_SUSPENDED; 1331 vmexit->u.suspended.how = vm->suspend; 1332 } 1333 1334 void 1335 vm_exit_rendezvous(struct vm *vm, int vcpuid, uint64_t rip) 1336 { 1337 struct vm_exit *vmexit; 1338 1339 KASSERT(vm->rendezvous_func != NULL, ("rendezvous not in progress")); 1340 1341 vmexit = vm_exitinfo(vm, vcpuid); 1342 vmexit->rip = rip; 1343 vmexit->inst_length = 0; 1344 vmexit->exitcode = VM_EXITCODE_RENDEZVOUS; 1345 vmm_stat_incr(vm, vcpuid, VMEXIT_RENDEZVOUS, 1); 1346 } 1347 1348 void 1349 vm_exit_astpending(struct vm *vm, int vcpuid, uint64_t rip) 1350 { 1351 struct vm_exit *vmexit; 1352 1353 vmexit = vm_exitinfo(vm, vcpuid); 1354 vmexit->rip = rip; 1355 vmexit->inst_length = 0; 1356 vmexit->exitcode = VM_EXITCODE_BOGUS; 1357 vmm_stat_incr(vm, vcpuid, VMEXIT_ASTPENDING, 1); 1358 } 1359 1360 int 1361 vm_run(struct vm *vm, struct vm_run *vmrun) 1362 { 1363 int error, vcpuid; 1364 struct vcpu *vcpu; 1365 struct pcb *pcb; 1366 uint64_t tscval, rip; 1367 struct vm_exit *vme; 1368 bool retu, intr_disabled; 1369 pmap_t pmap; 1370 void *rptr, *sptr; 1371 1372 vcpuid = vmrun->cpuid; 1373 1374 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1375 return (EINVAL); 1376 1377 if (!CPU_ISSET(vcpuid, &vm->active_cpus)) 1378 return (EINVAL); 1379 1380 if (CPU_ISSET(vcpuid, &vm->suspended_cpus)) 1381 return (EINVAL); 1382 1383 rptr = &vm->rendezvous_func; 1384 sptr = &vm->suspend; 1385 pmap = vmspace_pmap(vm->vmspace); 1386 vcpu = &vm->vcpu[vcpuid]; 1387 vme = &vcpu->exitinfo; 1388 rip = vmrun->rip; 1389 restart: 1390 critical_enter(); 1391 1392 KASSERT(!CPU_ISSET(curcpu, &pmap->pm_active), 1393 ("vm_run: absurd pm_active")); 1394 1395 tscval = rdtsc(); 1396 1397 pcb = PCPU_GET(curpcb); 1398 set_pcb_flags(pcb, PCB_FULL_IRET); 1399 1400 restore_guest_msrs(vm, vcpuid); 1401 restore_guest_fpustate(vcpu); 1402 1403 vcpu_require_state(vm, vcpuid, VCPU_RUNNING); 1404 error = VMRUN(vm->cookie, vcpuid, rip, pmap, rptr, sptr); 1405 vcpu_require_state(vm, vcpuid, VCPU_FROZEN); 1406 1407 save_guest_fpustate(vcpu); 1408 restore_host_msrs(vm, vcpuid); 1409 1410 vmm_stat_incr(vm, vcpuid, VCPU_TOTAL_RUNTIME, rdtsc() - tscval); 1411 1412 critical_exit(); 1413 1414 if (error == 0) { 1415 retu = false; 1416 switch (vme->exitcode) { 1417 case VM_EXITCODE_SUSPENDED: 1418 error = vm_handle_suspend(vm, vcpuid, &retu); 1419 break; 1420 case VM_EXITCODE_IOAPIC_EOI: 1421 vioapic_process_eoi(vm, vcpuid, 1422 vme->u.ioapic_eoi.vector); 1423 break; 1424 case VM_EXITCODE_RENDEZVOUS: 1425 vm_handle_rendezvous(vm, vcpuid); 1426 error = 0; 1427 break; 1428 case VM_EXITCODE_HLT: 1429 intr_disabled = ((vme->u.hlt.rflags & PSL_I) == 0); 1430 error = vm_handle_hlt(vm, vcpuid, intr_disabled, &retu); 1431 break; 1432 case VM_EXITCODE_PAGING: 1433 error = vm_handle_paging(vm, vcpuid, &retu); 1434 break; 1435 case VM_EXITCODE_INST_EMUL: 1436 error = vm_handle_inst_emul(vm, vcpuid, &retu); 1437 break; 1438 case VM_EXITCODE_INOUT: 1439 case VM_EXITCODE_INOUT_STR: 1440 error = vm_handle_inout(vm, vcpuid, vme, &retu); 1441 break; 1442 default: 1443 retu = true; /* handled in userland */ 1444 break; 1445 } 1446 } 1447 1448 if (error == 0 && retu == false) { 1449 rip = vme->rip + vme->inst_length; 1450 goto restart; 1451 } 1452 1453 /* copy the exit information */ 1454 bcopy(vme, &vmrun->vm_exit, sizeof(struct vm_exit)); 1455 return (error); 1456 } 1457 1458 int 1459 vm_inject_exception(struct vm *vm, int vcpuid, struct vm_exception *exception) 1460 { 1461 struct vcpu *vcpu; 1462 1463 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1464 return (EINVAL); 1465 1466 if (exception->vector < 0 || exception->vector >= 32) 1467 return (EINVAL); 1468 1469 vcpu = &vm->vcpu[vcpuid]; 1470 1471 if (vcpu->exception_pending) { 1472 VCPU_CTR2(vm, vcpuid, "Unable to inject exception %d due to " 1473 "pending exception %d", exception->vector, 1474 vcpu->exception.vector); 1475 return (EBUSY); 1476 } 1477 1478 vcpu->exception_pending = 1; 1479 vcpu->exception = *exception; 1480 VCPU_CTR1(vm, vcpuid, "Exception %d pending", exception->vector); 1481 return (0); 1482 } 1483 1484 int 1485 vm_exception_pending(struct vm *vm, int vcpuid, struct vm_exception *exception) 1486 { 1487 struct vcpu *vcpu; 1488 int pending; 1489 1490 KASSERT(vcpuid >= 0 && vcpuid < VM_MAXCPU, ("invalid vcpu %d", vcpuid)); 1491 1492 vcpu = &vm->vcpu[vcpuid]; 1493 pending = vcpu->exception_pending; 1494 if (pending) { 1495 vcpu->exception_pending = 0; 1496 *exception = vcpu->exception; 1497 VCPU_CTR1(vm, vcpuid, "Exception %d delivered", 1498 exception->vector); 1499 } 1500 return (pending); 1501 } 1502 1503 static void 1504 vm_inject_fault(struct vm *vm, int vcpuid, struct vm_exception *exception) 1505 { 1506 struct vm_exit *vmexit; 1507 int error; 1508 1509 error = vm_inject_exception(vm, vcpuid, exception); 1510 KASSERT(error == 0, ("vm_inject_exception error %d", error)); 1511 1512 /* 1513 * A fault-like exception allows the instruction to be restarted 1514 * after the exception handler returns. 1515 * 1516 * By setting the inst_length to 0 we ensure that the instruction 1517 * pointer remains at the faulting instruction. 1518 */ 1519 vmexit = vm_exitinfo(vm, vcpuid); 1520 vmexit->inst_length = 0; 1521 } 1522 1523 void 1524 vm_inject_pf(struct vm *vm, int vcpuid, int error_code, uint64_t cr2) 1525 { 1526 struct vm_exception pf = { 1527 .vector = IDT_PF, 1528 .error_code_valid = 1, 1529 .error_code = error_code 1530 }; 1531 int error; 1532 1533 VCPU_CTR2(vm, vcpuid, "Injecting page fault: error_code %#x, cr2 %#lx", 1534 error_code, cr2); 1535 1536 error = vm_set_register(vm, vcpuid, VM_REG_GUEST_CR2, cr2); 1537 KASSERT(error == 0, ("vm_set_register(cr2) error %d", error)); 1538 1539 vm_inject_fault(vm, vcpuid, &pf); 1540 } 1541 1542 void 1543 vm_inject_gp(struct vm *vm, int vcpuid) 1544 { 1545 struct vm_exception gpf = { 1546 .vector = IDT_GP, 1547 .error_code_valid = 1, 1548 .error_code = 0 1549 }; 1550 1551 vm_inject_fault(vm, vcpuid, &gpf); 1552 } 1553 1554 void 1555 vm_inject_ud(struct vm *vm, int vcpuid) 1556 { 1557 struct vm_exception udf = { 1558 .vector = IDT_UD, 1559 .error_code_valid = 0 1560 }; 1561 1562 vm_inject_fault(vm, vcpuid, &udf); 1563 } 1564 1565 static VMM_STAT(VCPU_NMI_COUNT, "number of NMIs delivered to vcpu"); 1566 1567 int 1568 vm_inject_nmi(struct vm *vm, int vcpuid) 1569 { 1570 struct vcpu *vcpu; 1571 1572 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1573 return (EINVAL); 1574 1575 vcpu = &vm->vcpu[vcpuid]; 1576 1577 vcpu->nmi_pending = 1; 1578 vcpu_notify_event(vm, vcpuid, false); 1579 return (0); 1580 } 1581 1582 int 1583 vm_nmi_pending(struct vm *vm, int vcpuid) 1584 { 1585 struct vcpu *vcpu; 1586 1587 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1588 panic("vm_nmi_pending: invalid vcpuid %d", vcpuid); 1589 1590 vcpu = &vm->vcpu[vcpuid]; 1591 1592 return (vcpu->nmi_pending); 1593 } 1594 1595 void 1596 vm_nmi_clear(struct vm *vm, int vcpuid) 1597 { 1598 struct vcpu *vcpu; 1599 1600 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1601 panic("vm_nmi_pending: invalid vcpuid %d", vcpuid); 1602 1603 vcpu = &vm->vcpu[vcpuid]; 1604 1605 if (vcpu->nmi_pending == 0) 1606 panic("vm_nmi_clear: inconsistent nmi_pending state"); 1607 1608 vcpu->nmi_pending = 0; 1609 vmm_stat_incr(vm, vcpuid, VCPU_NMI_COUNT, 1); 1610 } 1611 1612 static VMM_STAT(VCPU_EXTINT_COUNT, "number of ExtINTs delivered to vcpu"); 1613 1614 int 1615 vm_inject_extint(struct vm *vm, int vcpuid) 1616 { 1617 struct vcpu *vcpu; 1618 1619 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1620 return (EINVAL); 1621 1622 vcpu = &vm->vcpu[vcpuid]; 1623 1624 vcpu->extint_pending = 1; 1625 vcpu_notify_event(vm, vcpuid, false); 1626 return (0); 1627 } 1628 1629 int 1630 vm_extint_pending(struct vm *vm, int vcpuid) 1631 { 1632 struct vcpu *vcpu; 1633 1634 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1635 panic("vm_extint_pending: invalid vcpuid %d", vcpuid); 1636 1637 vcpu = &vm->vcpu[vcpuid]; 1638 1639 return (vcpu->extint_pending); 1640 } 1641 1642 void 1643 vm_extint_clear(struct vm *vm, int vcpuid) 1644 { 1645 struct vcpu *vcpu; 1646 1647 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1648 panic("vm_extint_pending: invalid vcpuid %d", vcpuid); 1649 1650 vcpu = &vm->vcpu[vcpuid]; 1651 1652 if (vcpu->extint_pending == 0) 1653 panic("vm_extint_clear: inconsistent extint_pending state"); 1654 1655 vcpu->extint_pending = 0; 1656 vmm_stat_incr(vm, vcpuid, VCPU_EXTINT_COUNT, 1); 1657 } 1658 1659 int 1660 vm_get_capability(struct vm *vm, int vcpu, int type, int *retval) 1661 { 1662 if (vcpu < 0 || vcpu >= VM_MAXCPU) 1663 return (EINVAL); 1664 1665 if (type < 0 || type >= VM_CAP_MAX) 1666 return (EINVAL); 1667 1668 return (VMGETCAP(vm->cookie, vcpu, type, retval)); 1669 } 1670 1671 int 1672 vm_set_capability(struct vm *vm, int vcpu, int type, int val) 1673 { 1674 if (vcpu < 0 || vcpu >= VM_MAXCPU) 1675 return (EINVAL); 1676 1677 if (type < 0 || type >= VM_CAP_MAX) 1678 return (EINVAL); 1679 1680 return (VMSETCAP(vm->cookie, vcpu, type, val)); 1681 } 1682 1683 uint64_t * 1684 vm_guest_msrs(struct vm *vm, int cpu) 1685 { 1686 return (vm->vcpu[cpu].guest_msrs); 1687 } 1688 1689 struct vlapic * 1690 vm_lapic(struct vm *vm, int cpu) 1691 { 1692 return (vm->vcpu[cpu].vlapic); 1693 } 1694 1695 struct vioapic * 1696 vm_ioapic(struct vm *vm) 1697 { 1698 1699 return (vm->vioapic); 1700 } 1701 1702 struct vhpet * 1703 vm_hpet(struct vm *vm) 1704 { 1705 1706 return (vm->vhpet); 1707 } 1708 1709 boolean_t 1710 vmm_is_pptdev(int bus, int slot, int func) 1711 { 1712 int found, i, n; 1713 int b, s, f; 1714 char *val, *cp, *cp2; 1715 1716 /* 1717 * XXX 1718 * The length of an environment variable is limited to 128 bytes which 1719 * puts an upper limit on the number of passthru devices that may be 1720 * specified using a single environment variable. 1721 * 1722 * Work around this by scanning multiple environment variable 1723 * names instead of a single one - yuck! 1724 */ 1725 const char *names[] = { "pptdevs", "pptdevs2", "pptdevs3", NULL }; 1726 1727 /* set pptdevs="1/2/3 4/5/6 7/8/9 10/11/12" */ 1728 found = 0; 1729 for (i = 0; names[i] != NULL && !found; i++) { 1730 cp = val = getenv(names[i]); 1731 while (cp != NULL && *cp != '\0') { 1732 if ((cp2 = strchr(cp, ' ')) != NULL) 1733 *cp2 = '\0'; 1734 1735 n = sscanf(cp, "%d/%d/%d", &b, &s, &f); 1736 if (n == 3 && bus == b && slot == s && func == f) { 1737 found = 1; 1738 break; 1739 } 1740 1741 if (cp2 != NULL) 1742 *cp2++ = ' '; 1743 1744 cp = cp2; 1745 } 1746 freeenv(val); 1747 } 1748 return (found); 1749 } 1750 1751 void * 1752 vm_iommu_domain(struct vm *vm) 1753 { 1754 1755 return (vm->iommu); 1756 } 1757 1758 int 1759 vcpu_set_state(struct vm *vm, int vcpuid, enum vcpu_state newstate, 1760 bool from_idle) 1761 { 1762 int error; 1763 struct vcpu *vcpu; 1764 1765 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1766 panic("vm_set_run_state: invalid vcpuid %d", vcpuid); 1767 1768 vcpu = &vm->vcpu[vcpuid]; 1769 1770 vcpu_lock(vcpu); 1771 error = vcpu_set_state_locked(vcpu, newstate, from_idle); 1772 vcpu_unlock(vcpu); 1773 1774 return (error); 1775 } 1776 1777 enum vcpu_state 1778 vcpu_get_state(struct vm *vm, int vcpuid, int *hostcpu) 1779 { 1780 struct vcpu *vcpu; 1781 enum vcpu_state state; 1782 1783 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1784 panic("vm_get_run_state: invalid vcpuid %d", vcpuid); 1785 1786 vcpu = &vm->vcpu[vcpuid]; 1787 1788 vcpu_lock(vcpu); 1789 state = vcpu->state; 1790 if (hostcpu != NULL) 1791 *hostcpu = vcpu->hostcpu; 1792 vcpu_unlock(vcpu); 1793 1794 return (state); 1795 } 1796 1797 int 1798 vm_activate_cpu(struct vm *vm, int vcpuid) 1799 { 1800 1801 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1802 return (EINVAL); 1803 1804 if (CPU_ISSET(vcpuid, &vm->active_cpus)) 1805 return (EBUSY); 1806 1807 VCPU_CTR0(vm, vcpuid, "activated"); 1808 CPU_SET_ATOMIC(vcpuid, &vm->active_cpus); 1809 return (0); 1810 } 1811 1812 cpuset_t 1813 vm_active_cpus(struct vm *vm) 1814 { 1815 1816 return (vm->active_cpus); 1817 } 1818 1819 cpuset_t 1820 vm_suspended_cpus(struct vm *vm) 1821 { 1822 1823 return (vm->suspended_cpus); 1824 } 1825 1826 void * 1827 vcpu_stats(struct vm *vm, int vcpuid) 1828 { 1829 1830 return (vm->vcpu[vcpuid].stats); 1831 } 1832 1833 int 1834 vm_get_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state *state) 1835 { 1836 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1837 return (EINVAL); 1838 1839 *state = vm->vcpu[vcpuid].x2apic_state; 1840 1841 return (0); 1842 } 1843 1844 int 1845 vm_set_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state state) 1846 { 1847 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1848 return (EINVAL); 1849 1850 if (state >= X2APIC_STATE_LAST) 1851 return (EINVAL); 1852 1853 vm->vcpu[vcpuid].x2apic_state = state; 1854 1855 vlapic_set_x2apic_state(vm, vcpuid, state); 1856 1857 return (0); 1858 } 1859 1860 /* 1861 * This function is called to ensure that a vcpu "sees" a pending event 1862 * as soon as possible: 1863 * - If the vcpu thread is sleeping then it is woken up. 1864 * - If the vcpu is running on a different host_cpu then an IPI will be directed 1865 * to the host_cpu to cause the vcpu to trap into the hypervisor. 1866 */ 1867 void 1868 vcpu_notify_event(struct vm *vm, int vcpuid, bool lapic_intr) 1869 { 1870 int hostcpu; 1871 struct vcpu *vcpu; 1872 1873 vcpu = &vm->vcpu[vcpuid]; 1874 1875 vcpu_lock(vcpu); 1876 hostcpu = vcpu->hostcpu; 1877 if (vcpu->state == VCPU_RUNNING) { 1878 KASSERT(hostcpu != NOCPU, ("vcpu running on invalid hostcpu")); 1879 if (hostcpu != curcpu) { 1880 if (lapic_intr) { 1881 vlapic_post_intr(vcpu->vlapic, hostcpu, 1882 vmm_ipinum); 1883 } else { 1884 ipi_cpu(hostcpu, vmm_ipinum); 1885 } 1886 } else { 1887 /* 1888 * If the 'vcpu' is running on 'curcpu' then it must 1889 * be sending a notification to itself (e.g. SELF_IPI). 1890 * The pending event will be picked up when the vcpu 1891 * transitions back to guest context. 1892 */ 1893 } 1894 } else { 1895 KASSERT(hostcpu == NOCPU, ("vcpu state %d not consistent " 1896 "with hostcpu %d", vcpu->state, hostcpu)); 1897 if (vcpu->state == VCPU_SLEEPING) 1898 wakeup_one(vcpu); 1899 } 1900 vcpu_unlock(vcpu); 1901 } 1902 1903 struct vmspace * 1904 vm_get_vmspace(struct vm *vm) 1905 { 1906 1907 return (vm->vmspace); 1908 } 1909 1910 int 1911 vm_apicid2vcpuid(struct vm *vm, int apicid) 1912 { 1913 /* 1914 * XXX apic id is assumed to be numerically identical to vcpu id 1915 */ 1916 return (apicid); 1917 } 1918 1919 void 1920 vm_smp_rendezvous(struct vm *vm, int vcpuid, cpuset_t dest, 1921 vm_rendezvous_func_t func, void *arg) 1922 { 1923 int i; 1924 1925 /* 1926 * Enforce that this function is called without any locks 1927 */ 1928 WITNESS_WARN(WARN_PANIC, NULL, "vm_smp_rendezvous"); 1929 KASSERT(vcpuid == -1 || (vcpuid >= 0 && vcpuid < VM_MAXCPU), 1930 ("vm_smp_rendezvous: invalid vcpuid %d", vcpuid)); 1931 1932 restart: 1933 mtx_lock(&vm->rendezvous_mtx); 1934 if (vm->rendezvous_func != NULL) { 1935 /* 1936 * If a rendezvous is already in progress then we need to 1937 * call the rendezvous handler in case this 'vcpuid' is one 1938 * of the targets of the rendezvous. 1939 */ 1940 RENDEZVOUS_CTR0(vm, vcpuid, "Rendezvous already in progress"); 1941 mtx_unlock(&vm->rendezvous_mtx); 1942 vm_handle_rendezvous(vm, vcpuid); 1943 goto restart; 1944 } 1945 KASSERT(vm->rendezvous_func == NULL, ("vm_smp_rendezvous: previous " 1946 "rendezvous is still in progress")); 1947 1948 RENDEZVOUS_CTR0(vm, vcpuid, "Initiating rendezvous"); 1949 vm->rendezvous_req_cpus = dest; 1950 CPU_ZERO(&vm->rendezvous_done_cpus); 1951 vm->rendezvous_arg = arg; 1952 vm_set_rendezvous_func(vm, func); 1953 mtx_unlock(&vm->rendezvous_mtx); 1954 1955 /* 1956 * Wake up any sleeping vcpus and trigger a VM-exit in any running 1957 * vcpus so they handle the rendezvous as soon as possible. 1958 */ 1959 for (i = 0; i < VM_MAXCPU; i++) { 1960 if (CPU_ISSET(i, &dest)) 1961 vcpu_notify_event(vm, i, false); 1962 } 1963 1964 vm_handle_rendezvous(vm, vcpuid); 1965 } 1966 1967 struct vatpic * 1968 vm_atpic(struct vm *vm) 1969 { 1970 return (vm->vatpic); 1971 } 1972 1973 struct vatpit * 1974 vm_atpit(struct vm *vm) 1975 { 1976 return (vm->vatpit); 1977 } 1978 1979 enum vm_reg_name 1980 vm_segment_name(int seg) 1981 { 1982 static enum vm_reg_name seg_names[] = { 1983 VM_REG_GUEST_ES, 1984 VM_REG_GUEST_CS, 1985 VM_REG_GUEST_SS, 1986 VM_REG_GUEST_DS, 1987 VM_REG_GUEST_FS, 1988 VM_REG_GUEST_GS 1989 }; 1990 1991 KASSERT(seg >= 0 && seg < nitems(seg_names), 1992 ("%s: invalid segment encoding %d", __func__, seg)); 1993 return (seg_names[seg]); 1994 } 1995 1996 1997 /* 1998 * Return the amount of in-use and wired memory for the VM. Since 1999 * these are global stats, only return the values with for vCPU 0 2000 */ 2001 VMM_STAT_DECLARE(VMM_MEM_RESIDENT); 2002 VMM_STAT_DECLARE(VMM_MEM_WIRED); 2003 2004 static void 2005 vm_get_rescnt(struct vm *vm, int vcpu, struct vmm_stat_type *stat) 2006 { 2007 2008 if (vcpu == 0) { 2009 vmm_stat_set(vm, vcpu, VMM_MEM_RESIDENT, 2010 PAGE_SIZE * vmspace_resident_count(vm->vmspace)); 2011 } 2012 } 2013 2014 static void 2015 vm_get_wiredcnt(struct vm *vm, int vcpu, struct vmm_stat_type *stat) 2016 { 2017 2018 if (vcpu == 0) { 2019 vmm_stat_set(vm, vcpu, VMM_MEM_WIRED, 2020 PAGE_SIZE * pmap_wired_count(vmspace_pmap(vm->vmspace))); 2021 } 2022 } 2023 2024 VMM_STAT_FUNC(VMM_MEM_RESIDENT, "Resident memory", vm_get_rescnt); 2025 VMM_STAT_FUNC(VMM_MEM_WIRED, "Wired memory", vm_get_wiredcnt); 2026