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/vm.h> 56 #include <machine/pcb.h> 57 #include <machine/smp.h> 58 #include <x86/apicreg.h> 59 #include <machine/vmparam.h> 60 61 #include <machine/vmm.h> 62 #include <machine/vmm_dev.h> 63 64 #include "vmm_ktr.h" 65 #include "vmm_host.h" 66 #include "vmm_mem.h" 67 #include "vmm_util.h" 68 #include "vioapic.h" 69 #include "vlapic.h" 70 #include "vmm_msr.h" 71 #include "vmm_ipi.h" 72 #include "vmm_stat.h" 73 #include "vmm_lapic.h" 74 75 #include "io/ppt.h" 76 #include "io/iommu.h" 77 78 struct vlapic; 79 80 struct vcpu { 81 int flags; 82 enum vcpu_state state; 83 struct mtx mtx; 84 int hostcpu; /* host cpuid this vcpu last ran on */ 85 uint64_t guest_msrs[VMM_MSR_NUM]; 86 struct vlapic *vlapic; 87 int vcpuid; 88 struct savefpu *guestfpu; /* guest fpu state */ 89 void *stats; 90 struct vm_exit exitinfo; 91 enum x2apic_state x2apic_state; 92 int nmi_pending; 93 }; 94 95 #define vcpu_lock_init(v) mtx_init(&((v)->mtx), "vcpu lock", 0, MTX_SPIN) 96 #define vcpu_lock(v) mtx_lock_spin(&((v)->mtx)) 97 #define vcpu_unlock(v) mtx_unlock_spin(&((v)->mtx)) 98 #define vcpu_assert_locked(v) mtx_assert(&((v)->mtx), MA_OWNED) 99 100 struct mem_seg { 101 vm_paddr_t gpa; 102 size_t len; 103 boolean_t wired; 104 vm_object_t object; 105 }; 106 #define VM_MAX_MEMORY_SEGMENTS 2 107 108 struct vm { 109 void *cookie; /* processor-specific data */ 110 void *iommu; /* iommu-specific data */ 111 struct vioapic *vioapic; /* virtual ioapic */ 112 struct vmspace *vmspace; /* guest's address space */ 113 struct vcpu vcpu[VM_MAXCPU]; 114 int num_mem_segs; 115 struct mem_seg mem_segs[VM_MAX_MEMORY_SEGMENTS]; 116 char name[VM_MAX_NAMELEN]; 117 118 /* 119 * Set of active vcpus. 120 * An active vcpu is one that has been started implicitly (BSP) or 121 * explicitly (AP) by sending it a startup ipi. 122 */ 123 cpuset_t active_cpus; 124 }; 125 126 static int vmm_initialized; 127 128 static struct vmm_ops *ops; 129 #define VMM_INIT() (ops != NULL ? (*ops->init)() : 0) 130 #define VMM_CLEANUP() (ops != NULL ? (*ops->cleanup)() : 0) 131 132 #define VMINIT(vm, pmap) (ops != NULL ? (*ops->vminit)(vm, pmap): NULL) 133 #define VMRUN(vmi, vcpu, rip, pmap) \ 134 (ops != NULL ? (*ops->vmrun)(vmi, vcpu, rip, pmap) : ENXIO) 135 #define VMCLEANUP(vmi) (ops != NULL ? (*ops->vmcleanup)(vmi) : NULL) 136 #define VMSPACE_ALLOC(min, max) \ 137 (ops != NULL ? (*ops->vmspace_alloc)(min, max) : NULL) 138 #define VMSPACE_FREE(vmspace) \ 139 (ops != NULL ? (*ops->vmspace_free)(vmspace) : ENXIO) 140 #define VMGETREG(vmi, vcpu, num, retval) \ 141 (ops != NULL ? (*ops->vmgetreg)(vmi, vcpu, num, retval) : ENXIO) 142 #define VMSETREG(vmi, vcpu, num, val) \ 143 (ops != NULL ? (*ops->vmsetreg)(vmi, vcpu, num, val) : ENXIO) 144 #define VMGETDESC(vmi, vcpu, num, desc) \ 145 (ops != NULL ? (*ops->vmgetdesc)(vmi, vcpu, num, desc) : ENXIO) 146 #define VMSETDESC(vmi, vcpu, num, desc) \ 147 (ops != NULL ? (*ops->vmsetdesc)(vmi, vcpu, num, desc) : ENXIO) 148 #define VMINJECT(vmi, vcpu, type, vec, ec, ecv) \ 149 (ops != NULL ? (*ops->vminject)(vmi, vcpu, type, vec, ec, ecv) : ENXIO) 150 #define VMGETCAP(vmi, vcpu, num, retval) \ 151 (ops != NULL ? (*ops->vmgetcap)(vmi, vcpu, num, retval) : ENXIO) 152 #define VMSETCAP(vmi, vcpu, num, val) \ 153 (ops != NULL ? (*ops->vmsetcap)(vmi, vcpu, num, val) : ENXIO) 154 155 #define fpu_start_emulating() load_cr0(rcr0() | CR0_TS) 156 #define fpu_stop_emulating() clts() 157 158 static MALLOC_DEFINE(M_VM, "vm", "vm"); 159 CTASSERT(VMM_MSR_NUM <= 64); /* msr_mask can keep track of up to 64 msrs */ 160 161 /* statistics */ 162 static VMM_STAT(VCPU_TOTAL_RUNTIME, "vcpu total runtime"); 163 164 static void 165 vcpu_cleanup(struct vcpu *vcpu) 166 { 167 vlapic_cleanup(vcpu->vlapic); 168 vmm_stat_free(vcpu->stats); 169 fpu_save_area_free(vcpu->guestfpu); 170 } 171 172 static void 173 vcpu_init(struct vm *vm, uint32_t vcpu_id) 174 { 175 struct vcpu *vcpu; 176 177 vcpu = &vm->vcpu[vcpu_id]; 178 179 vcpu_lock_init(vcpu); 180 vcpu->hostcpu = NOCPU; 181 vcpu->vcpuid = vcpu_id; 182 vcpu->vlapic = vlapic_init(vm, vcpu_id); 183 vm_set_x2apic_state(vm, vcpu_id, X2APIC_ENABLED); 184 vcpu->guestfpu = fpu_save_area_alloc(); 185 fpu_save_area_reset(vcpu->guestfpu); 186 vcpu->stats = vmm_stat_alloc(); 187 } 188 189 struct vm_exit * 190 vm_exitinfo(struct vm *vm, int cpuid) 191 { 192 struct vcpu *vcpu; 193 194 if (cpuid < 0 || cpuid >= VM_MAXCPU) 195 panic("vm_exitinfo: invalid cpuid %d", cpuid); 196 197 vcpu = &vm->vcpu[cpuid]; 198 199 return (&vcpu->exitinfo); 200 } 201 202 static int 203 vmm_init(void) 204 { 205 int error; 206 207 vmm_host_state_init(); 208 vmm_ipi_init(); 209 210 error = vmm_mem_init(); 211 if (error) 212 return (error); 213 214 if (vmm_is_intel()) 215 ops = &vmm_ops_intel; 216 else if (vmm_is_amd()) 217 ops = &vmm_ops_amd; 218 else 219 return (ENXIO); 220 221 vmm_msr_init(); 222 223 return (VMM_INIT()); 224 } 225 226 static int 227 vmm_handler(module_t mod, int what, void *arg) 228 { 229 int error; 230 231 switch (what) { 232 case MOD_LOAD: 233 vmmdev_init(); 234 iommu_init(); 235 error = vmm_init(); 236 if (error == 0) 237 vmm_initialized = 1; 238 break; 239 case MOD_UNLOAD: 240 error = vmmdev_cleanup(); 241 if (error == 0) { 242 iommu_cleanup(); 243 vmm_ipi_cleanup(); 244 error = VMM_CLEANUP(); 245 /* 246 * Something bad happened - prevent new 247 * VMs from being created 248 */ 249 if (error) 250 vmm_initialized = 0; 251 } 252 break; 253 default: 254 error = 0; 255 break; 256 } 257 return (error); 258 } 259 260 static moduledata_t vmm_kmod = { 261 "vmm", 262 vmm_handler, 263 NULL 264 }; 265 266 /* 267 * vmm initialization has the following dependencies: 268 * 269 * - iommu initialization must happen after the pci passthru driver has had 270 * a chance to attach to any passthru devices (after SI_SUB_CONFIGURE). 271 * 272 * - VT-x initialization requires smp_rendezvous() and therefore must happen 273 * after SMP is fully functional (after SI_SUB_SMP). 274 */ 275 DECLARE_MODULE(vmm, vmm_kmod, SI_SUB_SMP + 1, SI_ORDER_ANY); 276 MODULE_VERSION(vmm, 1); 277 278 SYSCTL_NODE(_hw, OID_AUTO, vmm, CTLFLAG_RW, NULL, NULL); 279 280 int 281 vm_create(const char *name, struct vm **retvm) 282 { 283 int i; 284 struct vm *vm; 285 struct vmspace *vmspace; 286 287 const int BSP = 0; 288 289 /* 290 * If vmm.ko could not be successfully initialized then don't attempt 291 * to create the virtual machine. 292 */ 293 if (!vmm_initialized) 294 return (ENXIO); 295 296 if (name == NULL || strlen(name) >= VM_MAX_NAMELEN) 297 return (EINVAL); 298 299 vmspace = VMSPACE_ALLOC(VM_MIN_ADDRESS, VM_MAXUSER_ADDRESS); 300 if (vmspace == NULL) 301 return (ENOMEM); 302 303 vm = malloc(sizeof(struct vm), M_VM, M_WAITOK | M_ZERO); 304 strcpy(vm->name, name); 305 vm->cookie = VMINIT(vm, vmspace_pmap(vmspace)); 306 vm->vioapic = vioapic_init(vm); 307 308 for (i = 0; i < VM_MAXCPU; i++) { 309 vcpu_init(vm, i); 310 guest_msrs_init(vm, i); 311 } 312 313 vm_activate_cpu(vm, BSP); 314 vm->vmspace = vmspace; 315 316 *retvm = vm; 317 return (0); 318 } 319 320 static void 321 vm_free_mem_seg(struct vm *vm, struct mem_seg *seg) 322 { 323 324 if (seg->object != NULL) 325 vmm_mem_free(vm->vmspace, seg->gpa, seg->len); 326 327 bzero(seg, sizeof(*seg)); 328 } 329 330 void 331 vm_destroy(struct vm *vm) 332 { 333 int i; 334 335 ppt_unassign_all(vm); 336 337 if (vm->iommu != NULL) 338 iommu_destroy_domain(vm->iommu); 339 340 for (i = 0; i < vm->num_mem_segs; i++) 341 vm_free_mem_seg(vm, &vm->mem_segs[i]); 342 343 vm->num_mem_segs = 0; 344 345 for (i = 0; i < VM_MAXCPU; i++) 346 vcpu_cleanup(&vm->vcpu[i]); 347 348 vioapic_cleanup(vm->vioapic); 349 350 VMSPACE_FREE(vm->vmspace); 351 352 VMCLEANUP(vm->cookie); 353 354 free(vm, M_VM); 355 } 356 357 const char * 358 vm_name(struct vm *vm) 359 { 360 return (vm->name); 361 } 362 363 int 364 vm_map_mmio(struct vm *vm, vm_paddr_t gpa, size_t len, vm_paddr_t hpa) 365 { 366 vm_object_t obj; 367 368 if ((obj = vmm_mmio_alloc(vm->vmspace, gpa, len, hpa)) == NULL) 369 return (ENOMEM); 370 else 371 return (0); 372 } 373 374 int 375 vm_unmap_mmio(struct vm *vm, vm_paddr_t gpa, size_t len) 376 { 377 378 vmm_mmio_free(vm->vmspace, gpa, len); 379 return (0); 380 } 381 382 boolean_t 383 vm_mem_allocated(struct vm *vm, vm_paddr_t gpa) 384 { 385 int i; 386 vm_paddr_t gpabase, gpalimit; 387 388 for (i = 0; i < vm->num_mem_segs; i++) { 389 gpabase = vm->mem_segs[i].gpa; 390 gpalimit = gpabase + vm->mem_segs[i].len; 391 if (gpa >= gpabase && gpa < gpalimit) 392 return (TRUE); /* 'gpa' is regular memory */ 393 } 394 395 if (ppt_is_mmio(vm, gpa)) 396 return (TRUE); /* 'gpa' is pci passthru mmio */ 397 398 return (FALSE); 399 } 400 401 int 402 vm_malloc(struct vm *vm, vm_paddr_t gpa, size_t len) 403 { 404 int available, allocated; 405 struct mem_seg *seg; 406 vm_object_t object; 407 vm_paddr_t g; 408 409 if ((gpa & PAGE_MASK) || (len & PAGE_MASK) || len == 0) 410 return (EINVAL); 411 412 available = allocated = 0; 413 g = gpa; 414 while (g < gpa + len) { 415 if (vm_mem_allocated(vm, g)) 416 allocated++; 417 else 418 available++; 419 420 g += PAGE_SIZE; 421 } 422 423 /* 424 * If there are some allocated and some available pages in the address 425 * range then it is an error. 426 */ 427 if (allocated && available) 428 return (EINVAL); 429 430 /* 431 * If the entire address range being requested has already been 432 * allocated then there isn't anything more to do. 433 */ 434 if (allocated && available == 0) 435 return (0); 436 437 if (vm->num_mem_segs >= VM_MAX_MEMORY_SEGMENTS) 438 return (E2BIG); 439 440 seg = &vm->mem_segs[vm->num_mem_segs]; 441 442 if ((object = vmm_mem_alloc(vm->vmspace, gpa, len)) == NULL) 443 return (ENOMEM); 444 445 seg->gpa = gpa; 446 seg->len = len; 447 seg->object = object; 448 seg->wired = FALSE; 449 450 vm->num_mem_segs++; 451 452 return (0); 453 } 454 455 static void 456 vm_gpa_unwire(struct vm *vm) 457 { 458 int i, rv; 459 struct mem_seg *seg; 460 461 for (i = 0; i < vm->num_mem_segs; i++) { 462 seg = &vm->mem_segs[i]; 463 if (!seg->wired) 464 continue; 465 466 rv = vm_map_unwire(&vm->vmspace->vm_map, 467 seg->gpa, seg->gpa + seg->len, 468 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES); 469 KASSERT(rv == KERN_SUCCESS, ("vm(%s) memory segment " 470 "%#lx/%ld could not be unwired: %d", 471 vm_name(vm), seg->gpa, seg->len, rv)); 472 473 seg->wired = FALSE; 474 } 475 } 476 477 static int 478 vm_gpa_wire(struct vm *vm) 479 { 480 int i, rv; 481 struct mem_seg *seg; 482 483 for (i = 0; i < vm->num_mem_segs; i++) { 484 seg = &vm->mem_segs[i]; 485 if (seg->wired) 486 continue; 487 488 /* XXX rlimits? */ 489 rv = vm_map_wire(&vm->vmspace->vm_map, 490 seg->gpa, seg->gpa + seg->len, 491 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES); 492 if (rv != KERN_SUCCESS) 493 break; 494 495 seg->wired = TRUE; 496 } 497 498 if (i < vm->num_mem_segs) { 499 /* 500 * Undo the wiring before returning an error. 501 */ 502 vm_gpa_unwire(vm); 503 return (EAGAIN); 504 } 505 506 return (0); 507 } 508 509 static void 510 vm_iommu_modify(struct vm *vm, boolean_t map) 511 { 512 int i, sz; 513 vm_paddr_t gpa, hpa; 514 struct mem_seg *seg; 515 void *vp, *cookie, *host_domain; 516 517 sz = PAGE_SIZE; 518 host_domain = iommu_host_domain(); 519 520 for (i = 0; i < vm->num_mem_segs; i++) { 521 seg = &vm->mem_segs[i]; 522 KASSERT(seg->wired, ("vm(%s) memory segment %#lx/%ld not wired", 523 vm_name(vm), seg->gpa, seg->len)); 524 525 gpa = seg->gpa; 526 while (gpa < seg->gpa + seg->len) { 527 vp = vm_gpa_hold(vm, gpa, PAGE_SIZE, VM_PROT_WRITE, 528 &cookie); 529 KASSERT(vp != NULL, ("vm(%s) could not map gpa %#lx", 530 vm_name(vm), gpa)); 531 532 vm_gpa_release(cookie); 533 534 hpa = DMAP_TO_PHYS((uintptr_t)vp); 535 if (map) { 536 iommu_create_mapping(vm->iommu, gpa, hpa, sz); 537 iommu_remove_mapping(host_domain, hpa, sz); 538 } else { 539 iommu_remove_mapping(vm->iommu, gpa, sz); 540 iommu_create_mapping(host_domain, hpa, hpa, sz); 541 } 542 543 gpa += PAGE_SIZE; 544 } 545 } 546 547 /* 548 * Invalidate the cached translations associated with the domain 549 * from which pages were removed. 550 */ 551 if (map) 552 iommu_invalidate_tlb(host_domain); 553 else 554 iommu_invalidate_tlb(vm->iommu); 555 } 556 557 #define vm_iommu_unmap(vm) vm_iommu_modify((vm), FALSE) 558 #define vm_iommu_map(vm) vm_iommu_modify((vm), TRUE) 559 560 int 561 vm_unassign_pptdev(struct vm *vm, int bus, int slot, int func) 562 { 563 int error; 564 565 error = ppt_unassign_device(vm, bus, slot, func); 566 if (error) 567 return (error); 568 569 if (ppt_num_devices(vm) == 0) { 570 vm_iommu_unmap(vm); 571 vm_gpa_unwire(vm); 572 } 573 return (0); 574 } 575 576 int 577 vm_assign_pptdev(struct vm *vm, int bus, int slot, int func) 578 { 579 int error; 580 vm_paddr_t maxaddr; 581 582 /* 583 * Virtual machines with pci passthru devices get special treatment: 584 * - the guest physical memory is wired 585 * - the iommu is programmed to do the 'gpa' to 'hpa' translation 586 * 587 * We need to do this before the first pci passthru device is attached. 588 */ 589 if (ppt_num_devices(vm) == 0) { 590 KASSERT(vm->iommu == NULL, 591 ("vm_assign_pptdev: iommu must be NULL")); 592 maxaddr = vmm_mem_maxaddr(); 593 vm->iommu = iommu_create_domain(maxaddr); 594 595 error = vm_gpa_wire(vm); 596 if (error) 597 return (error); 598 599 vm_iommu_map(vm); 600 } 601 602 error = ppt_assign_device(vm, bus, slot, func); 603 return (error); 604 } 605 606 void * 607 vm_gpa_hold(struct vm *vm, vm_paddr_t gpa, size_t len, int reqprot, 608 void **cookie) 609 { 610 int count, pageoff; 611 vm_page_t m; 612 613 pageoff = gpa & PAGE_MASK; 614 if (len > PAGE_SIZE - pageoff) 615 panic("vm_gpa_hold: invalid gpa/len: 0x%016lx/%lu", gpa, len); 616 617 count = vm_fault_quick_hold_pages(&vm->vmspace->vm_map, 618 trunc_page(gpa), PAGE_SIZE, reqprot, &m, 1); 619 620 if (count == 1) { 621 *cookie = m; 622 return ((void *)(PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m)) + pageoff)); 623 } else { 624 *cookie = NULL; 625 return (NULL); 626 } 627 } 628 629 void 630 vm_gpa_release(void *cookie) 631 { 632 vm_page_t m = cookie; 633 634 vm_page_lock(m); 635 vm_page_unhold(m); 636 vm_page_unlock(m); 637 } 638 639 int 640 vm_gpabase2memseg(struct vm *vm, vm_paddr_t gpabase, 641 struct vm_memory_segment *seg) 642 { 643 int i; 644 645 for (i = 0; i < vm->num_mem_segs; i++) { 646 if (gpabase == vm->mem_segs[i].gpa) { 647 seg->gpa = vm->mem_segs[i].gpa; 648 seg->len = vm->mem_segs[i].len; 649 seg->wired = vm->mem_segs[i].wired; 650 return (0); 651 } 652 } 653 return (-1); 654 } 655 656 int 657 vm_get_memobj(struct vm *vm, vm_paddr_t gpa, size_t len, 658 vm_offset_t *offset, struct vm_object **object) 659 { 660 int i; 661 size_t seg_len; 662 vm_paddr_t seg_gpa; 663 vm_object_t seg_obj; 664 665 for (i = 0; i < vm->num_mem_segs; i++) { 666 if ((seg_obj = vm->mem_segs[i].object) == NULL) 667 continue; 668 669 seg_gpa = vm->mem_segs[i].gpa; 670 seg_len = vm->mem_segs[i].len; 671 672 if (gpa >= seg_gpa && gpa < seg_gpa + seg_len) { 673 *offset = gpa - seg_gpa; 674 *object = seg_obj; 675 vm_object_reference(seg_obj); 676 return (0); 677 } 678 } 679 680 return (EINVAL); 681 } 682 683 int 684 vm_get_register(struct vm *vm, int vcpu, int reg, uint64_t *retval) 685 { 686 687 if (vcpu < 0 || vcpu >= VM_MAXCPU) 688 return (EINVAL); 689 690 if (reg >= VM_REG_LAST) 691 return (EINVAL); 692 693 return (VMGETREG(vm->cookie, vcpu, reg, retval)); 694 } 695 696 int 697 vm_set_register(struct vm *vm, int vcpu, int reg, uint64_t val) 698 { 699 700 if (vcpu < 0 || vcpu >= VM_MAXCPU) 701 return (EINVAL); 702 703 if (reg >= VM_REG_LAST) 704 return (EINVAL); 705 706 return (VMSETREG(vm->cookie, vcpu, reg, val)); 707 } 708 709 static boolean_t 710 is_descriptor_table(int reg) 711 { 712 713 switch (reg) { 714 case VM_REG_GUEST_IDTR: 715 case VM_REG_GUEST_GDTR: 716 return (TRUE); 717 default: 718 return (FALSE); 719 } 720 } 721 722 static boolean_t 723 is_segment_register(int reg) 724 { 725 726 switch (reg) { 727 case VM_REG_GUEST_ES: 728 case VM_REG_GUEST_CS: 729 case VM_REG_GUEST_SS: 730 case VM_REG_GUEST_DS: 731 case VM_REG_GUEST_FS: 732 case VM_REG_GUEST_GS: 733 case VM_REG_GUEST_TR: 734 case VM_REG_GUEST_LDTR: 735 return (TRUE); 736 default: 737 return (FALSE); 738 } 739 } 740 741 int 742 vm_get_seg_desc(struct vm *vm, int vcpu, int reg, 743 struct seg_desc *desc) 744 { 745 746 if (vcpu < 0 || vcpu >= VM_MAXCPU) 747 return (EINVAL); 748 749 if (!is_segment_register(reg) && !is_descriptor_table(reg)) 750 return (EINVAL); 751 752 return (VMGETDESC(vm->cookie, vcpu, reg, desc)); 753 } 754 755 int 756 vm_set_seg_desc(struct vm *vm, int vcpu, int reg, 757 struct seg_desc *desc) 758 { 759 if (vcpu < 0 || vcpu >= VM_MAXCPU) 760 return (EINVAL); 761 762 if (!is_segment_register(reg) && !is_descriptor_table(reg)) 763 return (EINVAL); 764 765 return (VMSETDESC(vm->cookie, vcpu, reg, desc)); 766 } 767 768 static void 769 restore_guest_fpustate(struct vcpu *vcpu) 770 { 771 772 /* flush host state to the pcb */ 773 fpuexit(curthread); 774 775 /* restore guest FPU state */ 776 fpu_stop_emulating(); 777 fpurestore(vcpu->guestfpu); 778 779 /* 780 * The FPU is now "dirty" with the guest's state so turn on emulation 781 * to trap any access to the FPU by the host. 782 */ 783 fpu_start_emulating(); 784 } 785 786 static void 787 save_guest_fpustate(struct vcpu *vcpu) 788 { 789 790 if ((rcr0() & CR0_TS) == 0) 791 panic("fpu emulation not enabled in host!"); 792 793 /* save guest FPU state */ 794 fpu_stop_emulating(); 795 fpusave(vcpu->guestfpu); 796 fpu_start_emulating(); 797 } 798 799 static VMM_STAT(VCPU_IDLE_TICKS, "number of ticks vcpu was idle"); 800 801 static int 802 vcpu_set_state_locked(struct vcpu *vcpu, enum vcpu_state newstate) 803 { 804 int error; 805 806 vcpu_assert_locked(vcpu); 807 808 /* 809 * The following state transitions are allowed: 810 * IDLE -> FROZEN -> IDLE 811 * FROZEN -> RUNNING -> FROZEN 812 * FROZEN -> SLEEPING -> FROZEN 813 */ 814 switch (vcpu->state) { 815 case VCPU_IDLE: 816 case VCPU_RUNNING: 817 case VCPU_SLEEPING: 818 error = (newstate != VCPU_FROZEN); 819 break; 820 case VCPU_FROZEN: 821 error = (newstate == VCPU_FROZEN); 822 break; 823 default: 824 error = 1; 825 break; 826 } 827 828 if (error == 0) 829 vcpu->state = newstate; 830 else 831 error = EBUSY; 832 833 return (error); 834 } 835 836 static void 837 vcpu_require_state(struct vm *vm, int vcpuid, enum vcpu_state newstate) 838 { 839 int error; 840 841 if ((error = vcpu_set_state(vm, vcpuid, newstate)) != 0) 842 panic("Error %d setting state to %d\n", error, newstate); 843 } 844 845 static void 846 vcpu_require_state_locked(struct vcpu *vcpu, enum vcpu_state newstate) 847 { 848 int error; 849 850 if ((error = vcpu_set_state_locked(vcpu, newstate)) != 0) 851 panic("Error %d setting state to %d", error, newstate); 852 } 853 854 /* 855 * Emulate a guest 'hlt' by sleeping until the vcpu is ready to run. 856 */ 857 static int 858 vm_handle_hlt(struct vm *vm, int vcpuid, boolean_t *retu) 859 { 860 struct vcpu *vcpu; 861 int sleepticks, t; 862 863 vcpu = &vm->vcpu[vcpuid]; 864 865 vcpu_lock(vcpu); 866 867 /* 868 * Figure out the number of host ticks until the next apic 869 * timer interrupt in the guest. 870 */ 871 sleepticks = lapic_timer_tick(vm, vcpuid); 872 873 /* 874 * If the guest local apic timer is disabled then sleep for 875 * a long time but not forever. 876 */ 877 if (sleepticks < 0) 878 sleepticks = hz; 879 880 /* 881 * Do a final check for pending NMI or interrupts before 882 * really putting this thread to sleep. 883 * 884 * These interrupts could have happened any time after we 885 * returned from VMRUN() and before we grabbed the vcpu lock. 886 */ 887 if (!vm_nmi_pending(vm, vcpuid) && lapic_pending_intr(vm, vcpuid) < 0) { 888 if (sleepticks <= 0) 889 panic("invalid sleepticks %d", sleepticks); 890 t = ticks; 891 vcpu_require_state_locked(vcpu, VCPU_SLEEPING); 892 msleep_spin(vcpu, &vcpu->mtx, "vmidle", sleepticks); 893 vcpu_require_state_locked(vcpu, VCPU_FROZEN); 894 vmm_stat_incr(vm, vcpuid, VCPU_IDLE_TICKS, ticks - t); 895 } 896 vcpu_unlock(vcpu); 897 898 return (0); 899 } 900 901 static int 902 vm_handle_paging(struct vm *vm, int vcpuid, boolean_t *retu) 903 { 904 int rv, ftype; 905 struct vm_map *map; 906 struct vcpu *vcpu; 907 struct vm_exit *vme; 908 909 vcpu = &vm->vcpu[vcpuid]; 910 vme = &vcpu->exitinfo; 911 912 ftype = vme->u.paging.fault_type; 913 KASSERT(ftype == VM_PROT_READ || 914 ftype == VM_PROT_WRITE || ftype == VM_PROT_EXECUTE, 915 ("vm_handle_paging: invalid fault_type %d", ftype)); 916 917 if (ftype == VM_PROT_READ || ftype == VM_PROT_WRITE) { 918 rv = pmap_emulate_accessed_dirty(vmspace_pmap(vm->vmspace), 919 vme->u.paging.gpa, ftype); 920 if (rv == 0) 921 goto done; 922 } 923 924 map = &vm->vmspace->vm_map; 925 rv = vm_fault(map, vme->u.paging.gpa, ftype, VM_FAULT_NORMAL); 926 927 VCPU_CTR3(vm, vcpuid, "vm_handle_paging rv = %d, gpa = %#lx, " 928 "ftype = %d", rv, vme->u.paging.gpa, ftype); 929 930 if (rv != KERN_SUCCESS) 931 return (EFAULT); 932 done: 933 /* restart execution at the faulting instruction */ 934 vme->inst_length = 0; 935 936 return (0); 937 } 938 939 static int 940 vm_handle_inst_emul(struct vm *vm, int vcpuid, boolean_t *retu) 941 { 942 struct vie *vie; 943 struct vcpu *vcpu; 944 struct vm_exit *vme; 945 int error, inst_length; 946 uint64_t rip, gla, gpa, cr3; 947 mem_region_read_t mread; 948 mem_region_write_t mwrite; 949 950 vcpu = &vm->vcpu[vcpuid]; 951 vme = &vcpu->exitinfo; 952 953 rip = vme->rip; 954 inst_length = vme->inst_length; 955 956 gla = vme->u.inst_emul.gla; 957 gpa = vme->u.inst_emul.gpa; 958 cr3 = vme->u.inst_emul.cr3; 959 vie = &vme->u.inst_emul.vie; 960 961 vie_init(vie); 962 963 /* Fetch, decode and emulate the faulting instruction */ 964 if (vmm_fetch_instruction(vm, vcpuid, rip, inst_length, cr3, vie) != 0) 965 return (EFAULT); 966 967 if (vmm_decode_instruction(vm, vcpuid, gla, vie) != 0) 968 return (EFAULT); 969 970 /* return to userland unless this is a local apic access */ 971 if (gpa >= DEFAULT_APIC_BASE && gpa < DEFAULT_APIC_BASE + PAGE_SIZE) { 972 mread = lapic_mmio_read; 973 mwrite = lapic_mmio_write; 974 } else if (gpa >= VIOAPIC_BASE && gpa < VIOAPIC_BASE + VIOAPIC_SIZE) { 975 mread = vioapic_mmio_read; 976 mwrite = vioapic_mmio_write; 977 } else { 978 *retu = TRUE; 979 return (0); 980 } 981 982 error = vmm_emulate_instruction(vm, vcpuid, gpa, vie, mread, mwrite, 0); 983 984 /* return to userland to spin up the AP */ 985 if (error == 0 && vme->exitcode == VM_EXITCODE_SPINUP_AP) 986 *retu = TRUE; 987 988 return (error); 989 } 990 991 int 992 vm_run(struct vm *vm, struct vm_run *vmrun) 993 { 994 int error, vcpuid; 995 struct vcpu *vcpu; 996 struct pcb *pcb; 997 uint64_t tscval, rip; 998 struct vm_exit *vme; 999 boolean_t retu; 1000 pmap_t pmap; 1001 1002 vcpuid = vmrun->cpuid; 1003 1004 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1005 return (EINVAL); 1006 1007 pmap = vmspace_pmap(vm->vmspace); 1008 vcpu = &vm->vcpu[vcpuid]; 1009 vme = &vcpu->exitinfo; 1010 rip = vmrun->rip; 1011 restart: 1012 critical_enter(); 1013 1014 KASSERT(!CPU_ISSET(curcpu, &pmap->pm_active), 1015 ("vm_run: absurd pm_active")); 1016 1017 tscval = rdtsc(); 1018 1019 pcb = PCPU_GET(curpcb); 1020 set_pcb_flags(pcb, PCB_FULL_IRET); 1021 1022 restore_guest_msrs(vm, vcpuid); 1023 restore_guest_fpustate(vcpu); 1024 1025 vcpu_require_state(vm, vcpuid, VCPU_RUNNING); 1026 vcpu->hostcpu = curcpu; 1027 error = VMRUN(vm->cookie, vcpuid, rip, pmap); 1028 vcpu->hostcpu = NOCPU; 1029 vcpu_require_state(vm, vcpuid, VCPU_FROZEN); 1030 1031 save_guest_fpustate(vcpu); 1032 restore_host_msrs(vm, vcpuid); 1033 1034 vmm_stat_incr(vm, vcpuid, VCPU_TOTAL_RUNTIME, rdtsc() - tscval); 1035 1036 critical_exit(); 1037 1038 if (error == 0) { 1039 retu = FALSE; 1040 switch (vme->exitcode) { 1041 case VM_EXITCODE_HLT: 1042 error = vm_handle_hlt(vm, vcpuid, &retu); 1043 break; 1044 case VM_EXITCODE_PAGING: 1045 error = vm_handle_paging(vm, vcpuid, &retu); 1046 break; 1047 case VM_EXITCODE_INST_EMUL: 1048 error = vm_handle_inst_emul(vm, vcpuid, &retu); 1049 break; 1050 default: 1051 retu = TRUE; /* handled in userland */ 1052 break; 1053 } 1054 } 1055 1056 if (error == 0 && retu == FALSE) { 1057 rip = vme->rip + vme->inst_length; 1058 goto restart; 1059 } 1060 1061 /* copy the exit information */ 1062 bcopy(vme, &vmrun->vm_exit, sizeof(struct vm_exit)); 1063 return (error); 1064 } 1065 1066 int 1067 vm_inject_event(struct vm *vm, int vcpuid, int type, 1068 int vector, uint32_t code, int code_valid) 1069 { 1070 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1071 return (EINVAL); 1072 1073 if ((type > VM_EVENT_NONE && type < VM_EVENT_MAX) == 0) 1074 return (EINVAL); 1075 1076 if (vector < 0 || vector > 255) 1077 return (EINVAL); 1078 1079 return (VMINJECT(vm->cookie, vcpuid, type, vector, code, code_valid)); 1080 } 1081 1082 static VMM_STAT(VCPU_NMI_COUNT, "number of NMIs delivered to vcpu"); 1083 1084 int 1085 vm_inject_nmi(struct vm *vm, int vcpuid) 1086 { 1087 struct vcpu *vcpu; 1088 1089 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1090 return (EINVAL); 1091 1092 vcpu = &vm->vcpu[vcpuid]; 1093 1094 vcpu->nmi_pending = 1; 1095 vm_interrupt_hostcpu(vm, vcpuid); 1096 return (0); 1097 } 1098 1099 int 1100 vm_nmi_pending(struct vm *vm, int vcpuid) 1101 { 1102 struct vcpu *vcpu; 1103 1104 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1105 panic("vm_nmi_pending: invalid vcpuid %d", vcpuid); 1106 1107 vcpu = &vm->vcpu[vcpuid]; 1108 1109 return (vcpu->nmi_pending); 1110 } 1111 1112 void 1113 vm_nmi_clear(struct vm *vm, int vcpuid) 1114 { 1115 struct vcpu *vcpu; 1116 1117 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1118 panic("vm_nmi_pending: invalid vcpuid %d", vcpuid); 1119 1120 vcpu = &vm->vcpu[vcpuid]; 1121 1122 if (vcpu->nmi_pending == 0) 1123 panic("vm_nmi_clear: inconsistent nmi_pending state"); 1124 1125 vcpu->nmi_pending = 0; 1126 vmm_stat_incr(vm, vcpuid, VCPU_NMI_COUNT, 1); 1127 } 1128 1129 int 1130 vm_get_capability(struct vm *vm, int vcpu, int type, int *retval) 1131 { 1132 if (vcpu < 0 || vcpu >= VM_MAXCPU) 1133 return (EINVAL); 1134 1135 if (type < 0 || type >= VM_CAP_MAX) 1136 return (EINVAL); 1137 1138 return (VMGETCAP(vm->cookie, vcpu, type, retval)); 1139 } 1140 1141 int 1142 vm_set_capability(struct vm *vm, int vcpu, int type, int val) 1143 { 1144 if (vcpu < 0 || vcpu >= VM_MAXCPU) 1145 return (EINVAL); 1146 1147 if (type < 0 || type >= VM_CAP_MAX) 1148 return (EINVAL); 1149 1150 return (VMSETCAP(vm->cookie, vcpu, type, val)); 1151 } 1152 1153 uint64_t * 1154 vm_guest_msrs(struct vm *vm, int cpu) 1155 { 1156 return (vm->vcpu[cpu].guest_msrs); 1157 } 1158 1159 struct vlapic * 1160 vm_lapic(struct vm *vm, int cpu) 1161 { 1162 return (vm->vcpu[cpu].vlapic); 1163 } 1164 1165 struct vioapic * 1166 vm_ioapic(struct vm *vm) 1167 { 1168 1169 return (vm->vioapic); 1170 } 1171 1172 boolean_t 1173 vmm_is_pptdev(int bus, int slot, int func) 1174 { 1175 int found, i, n; 1176 int b, s, f; 1177 char *val, *cp, *cp2; 1178 1179 /* 1180 * XXX 1181 * The length of an environment variable is limited to 128 bytes which 1182 * puts an upper limit on the number of passthru devices that may be 1183 * specified using a single environment variable. 1184 * 1185 * Work around this by scanning multiple environment variable 1186 * names instead of a single one - yuck! 1187 */ 1188 const char *names[] = { "pptdevs", "pptdevs2", "pptdevs3", NULL }; 1189 1190 /* set pptdevs="1/2/3 4/5/6 7/8/9 10/11/12" */ 1191 found = 0; 1192 for (i = 0; names[i] != NULL && !found; i++) { 1193 cp = val = getenv(names[i]); 1194 while (cp != NULL && *cp != '\0') { 1195 if ((cp2 = strchr(cp, ' ')) != NULL) 1196 *cp2 = '\0'; 1197 1198 n = sscanf(cp, "%d/%d/%d", &b, &s, &f); 1199 if (n == 3 && bus == b && slot == s && func == f) { 1200 found = 1; 1201 break; 1202 } 1203 1204 if (cp2 != NULL) 1205 *cp2++ = ' '; 1206 1207 cp = cp2; 1208 } 1209 freeenv(val); 1210 } 1211 return (found); 1212 } 1213 1214 void * 1215 vm_iommu_domain(struct vm *vm) 1216 { 1217 1218 return (vm->iommu); 1219 } 1220 1221 int 1222 vcpu_set_state(struct vm *vm, int vcpuid, enum vcpu_state newstate) 1223 { 1224 int error; 1225 struct vcpu *vcpu; 1226 1227 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1228 panic("vm_set_run_state: invalid vcpuid %d", vcpuid); 1229 1230 vcpu = &vm->vcpu[vcpuid]; 1231 1232 vcpu_lock(vcpu); 1233 error = vcpu_set_state_locked(vcpu, newstate); 1234 vcpu_unlock(vcpu); 1235 1236 return (error); 1237 } 1238 1239 enum vcpu_state 1240 vcpu_get_state(struct vm *vm, int vcpuid, int *hostcpu) 1241 { 1242 struct vcpu *vcpu; 1243 enum vcpu_state state; 1244 1245 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1246 panic("vm_get_run_state: invalid vcpuid %d", vcpuid); 1247 1248 vcpu = &vm->vcpu[vcpuid]; 1249 1250 vcpu_lock(vcpu); 1251 state = vcpu->state; 1252 if (hostcpu != NULL) 1253 *hostcpu = vcpu->hostcpu; 1254 vcpu_unlock(vcpu); 1255 1256 return (state); 1257 } 1258 1259 void 1260 vm_activate_cpu(struct vm *vm, int vcpuid) 1261 { 1262 1263 if (vcpuid >= 0 && vcpuid < VM_MAXCPU) 1264 CPU_SET(vcpuid, &vm->active_cpus); 1265 } 1266 1267 cpuset_t 1268 vm_active_cpus(struct vm *vm) 1269 { 1270 1271 return (vm->active_cpus); 1272 } 1273 1274 void * 1275 vcpu_stats(struct vm *vm, int vcpuid) 1276 { 1277 1278 return (vm->vcpu[vcpuid].stats); 1279 } 1280 1281 int 1282 vm_get_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state *state) 1283 { 1284 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1285 return (EINVAL); 1286 1287 *state = vm->vcpu[vcpuid].x2apic_state; 1288 1289 return (0); 1290 } 1291 1292 int 1293 vm_set_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state state) 1294 { 1295 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1296 return (EINVAL); 1297 1298 if (state >= X2APIC_STATE_LAST) 1299 return (EINVAL); 1300 1301 vm->vcpu[vcpuid].x2apic_state = state; 1302 1303 vlapic_set_x2apic_state(vm, vcpuid, state); 1304 1305 return (0); 1306 } 1307 1308 void 1309 vm_interrupt_hostcpu(struct vm *vm, int vcpuid) 1310 { 1311 int hostcpu; 1312 struct vcpu *vcpu; 1313 1314 vcpu = &vm->vcpu[vcpuid]; 1315 1316 vcpu_lock(vcpu); 1317 hostcpu = vcpu->hostcpu; 1318 if (hostcpu == NOCPU) { 1319 if (vcpu->state == VCPU_SLEEPING) 1320 wakeup_one(vcpu); 1321 } else { 1322 if (vcpu->state != VCPU_RUNNING) 1323 panic("invalid vcpu state %d", vcpu->state); 1324 if (hostcpu != curcpu) 1325 ipi_cpu(hostcpu, vmm_ipinum); 1326 } 1327 vcpu_unlock(vcpu); 1328 } 1329 1330 struct vmspace * 1331 vm_get_vmspace(struct vm *vm) 1332 { 1333 1334 return (vm->vmspace); 1335 } 1336 1337 int 1338 vm_apicid2vcpuid(struct vm *vm, int apicid) 1339 { 1340 /* 1341 * XXX apic id is assumed to be numerically identical to vcpu id 1342 */ 1343 return (apicid); 1344 } 1345