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 66 #include "vmm_ktr.h" 67 #include "vmm_host.h" 68 #include "vmm_mem.h" 69 #include "vmm_util.h" 70 #include "vatpic.h" 71 #include "vatpit.h" 72 #include "vhpet.h" 73 #include "vioapic.h" 74 #include "vlapic.h" 75 #include "vmm_msr.h" 76 #include "vmm_ipi.h" 77 #include "vmm_stat.h" 78 #include "vmm_lapic.h" 79 80 #include "io/ppt.h" 81 #include "io/iommu.h" 82 83 struct vlapic; 84 85 struct vcpu { 86 int flags; 87 enum vcpu_state state; 88 struct mtx mtx; 89 int hostcpu; /* host cpuid this vcpu last ran on */ 90 uint64_t guest_msrs[VMM_MSR_NUM]; 91 struct vlapic *vlapic; 92 int vcpuid; 93 struct savefpu *guestfpu; /* guest fpu state */ 94 uint64_t guest_xcr0; 95 void *stats; 96 struct vm_exit exitinfo; 97 enum x2apic_state x2apic_state; 98 int nmi_pending; 99 int extint_pending; 100 struct vm_exception exception; 101 int exception_pending; 102 }; 103 104 #define vcpu_lock_init(v) mtx_init(&((v)->mtx), "vcpu lock", 0, MTX_SPIN) 105 #define vcpu_lock(v) mtx_lock_spin(&((v)->mtx)) 106 #define vcpu_unlock(v) mtx_unlock_spin(&((v)->mtx)) 107 #define vcpu_assert_locked(v) mtx_assert(&((v)->mtx), MA_OWNED) 108 109 struct mem_seg { 110 vm_paddr_t gpa; 111 size_t len; 112 boolean_t wired; 113 vm_object_t object; 114 }; 115 #define VM_MAX_MEMORY_SEGMENTS 2 116 117 struct vm { 118 void *cookie; /* processor-specific data */ 119 void *iommu; /* iommu-specific data */ 120 struct vhpet *vhpet; /* virtual HPET */ 121 struct vioapic *vioapic; /* virtual ioapic */ 122 struct vatpic *vatpic; /* virtual atpic */ 123 struct vatpit *vatpit; /* virtual atpit */ 124 struct vmspace *vmspace; /* guest's address space */ 125 struct vcpu vcpu[VM_MAXCPU]; 126 int num_mem_segs; 127 struct mem_seg mem_segs[VM_MAX_MEMORY_SEGMENTS]; 128 char name[VM_MAX_NAMELEN]; 129 130 /* 131 * Set of active vcpus. 132 * An active vcpu is one that has been started implicitly (BSP) or 133 * explicitly (AP) by sending it a startup ipi. 134 */ 135 volatile cpuset_t active_cpus; 136 137 struct mtx rendezvous_mtx; 138 cpuset_t rendezvous_req_cpus; 139 cpuset_t rendezvous_done_cpus; 140 void *rendezvous_arg; 141 vm_rendezvous_func_t rendezvous_func; 142 143 int suspend; 144 volatile cpuset_t suspended_cpus; 145 }; 146 147 static int vmm_initialized; 148 149 static struct vmm_ops *ops; 150 #define VMM_INIT(num) (ops != NULL ? (*ops->init)(num) : 0) 151 #define VMM_CLEANUP() (ops != NULL ? (*ops->cleanup)() : 0) 152 #define VMM_RESUME() (ops != NULL ? (*ops->resume)() : 0) 153 154 #define VMINIT(vm, pmap) (ops != NULL ? (*ops->vminit)(vm, pmap): NULL) 155 #define VMRUN(vmi, vcpu, rip, pmap, rptr, sptr) \ 156 (ops != NULL ? (*ops->vmrun)(vmi, vcpu, rip, pmap, rptr, sptr) : ENXIO) 157 #define VMCLEANUP(vmi) (ops != NULL ? (*ops->vmcleanup)(vmi) : NULL) 158 #define VMSPACE_ALLOC(min, max) \ 159 (ops != NULL ? (*ops->vmspace_alloc)(min, max) : NULL) 160 #define VMSPACE_FREE(vmspace) \ 161 (ops != NULL ? (*ops->vmspace_free)(vmspace) : ENXIO) 162 #define VMGETREG(vmi, vcpu, num, retval) \ 163 (ops != NULL ? (*ops->vmgetreg)(vmi, vcpu, num, retval) : ENXIO) 164 #define VMSETREG(vmi, vcpu, num, val) \ 165 (ops != NULL ? (*ops->vmsetreg)(vmi, vcpu, num, val) : ENXIO) 166 #define VMGETDESC(vmi, vcpu, num, desc) \ 167 (ops != NULL ? (*ops->vmgetdesc)(vmi, vcpu, num, desc) : ENXIO) 168 #define VMSETDESC(vmi, vcpu, num, desc) \ 169 (ops != NULL ? (*ops->vmsetdesc)(vmi, vcpu, num, desc) : ENXIO) 170 #define VMGETCAP(vmi, vcpu, num, retval) \ 171 (ops != NULL ? (*ops->vmgetcap)(vmi, vcpu, num, retval) : ENXIO) 172 #define VMSETCAP(vmi, vcpu, num, val) \ 173 (ops != NULL ? (*ops->vmsetcap)(vmi, vcpu, num, val) : ENXIO) 174 #define VLAPIC_INIT(vmi, vcpu) \ 175 (ops != NULL ? (*ops->vlapic_init)(vmi, vcpu) : NULL) 176 #define VLAPIC_CLEANUP(vmi, vlapic) \ 177 (ops != NULL ? (*ops->vlapic_cleanup)(vmi, vlapic) : NULL) 178 179 #define fpu_start_emulating() load_cr0(rcr0() | CR0_TS) 180 #define fpu_stop_emulating() clts() 181 182 static MALLOC_DEFINE(M_VM, "vm", "vm"); 183 CTASSERT(VMM_MSR_NUM <= 64); /* msr_mask can keep track of up to 64 msrs */ 184 185 /* statistics */ 186 static VMM_STAT(VCPU_TOTAL_RUNTIME, "vcpu total runtime"); 187 188 SYSCTL_NODE(_hw, OID_AUTO, vmm, CTLFLAG_RW, NULL, NULL); 189 190 static int vmm_ipinum; 191 SYSCTL_INT(_hw_vmm, OID_AUTO, ipinum, CTLFLAG_RD, &vmm_ipinum, 0, 192 "IPI vector used for vcpu notifications"); 193 194 static void vm_deactivate_cpu(struct vm *vm, int vcpuid); 195 196 static void 197 vcpu_cleanup(struct vm *vm, int i) 198 { 199 struct vcpu *vcpu = &vm->vcpu[i]; 200 201 VLAPIC_CLEANUP(vm->cookie, vcpu->vlapic); 202 vmm_stat_free(vcpu->stats); 203 fpu_save_area_free(vcpu->guestfpu); 204 } 205 206 static void 207 vcpu_init(struct vm *vm, uint32_t vcpu_id) 208 { 209 struct vcpu *vcpu; 210 211 vcpu = &vm->vcpu[vcpu_id]; 212 213 vcpu_lock_init(vcpu); 214 vcpu->hostcpu = NOCPU; 215 vcpu->vcpuid = vcpu_id; 216 vcpu->vlapic = VLAPIC_INIT(vm->cookie, vcpu_id); 217 vm_set_x2apic_state(vm, vcpu_id, X2APIC_DISABLED); 218 vcpu->guest_xcr0 = XFEATURE_ENABLED_X87; 219 vcpu->guestfpu = fpu_save_area_alloc(); 220 fpu_save_area_reset(vcpu->guestfpu); 221 vcpu->stats = vmm_stat_alloc(); 222 } 223 224 struct vm_exit * 225 vm_exitinfo(struct vm *vm, int cpuid) 226 { 227 struct vcpu *vcpu; 228 229 if (cpuid < 0 || cpuid >= VM_MAXCPU) 230 panic("vm_exitinfo: invalid cpuid %d", cpuid); 231 232 vcpu = &vm->vcpu[cpuid]; 233 234 return (&vcpu->exitinfo); 235 } 236 237 static void 238 vmm_resume(void) 239 { 240 VMM_RESUME(); 241 } 242 243 static int 244 vmm_init(void) 245 { 246 int error; 247 248 vmm_host_state_init(); 249 250 vmm_ipinum = vmm_ipi_alloc(); 251 if (vmm_ipinum == 0) 252 vmm_ipinum = IPI_AST; 253 254 error = vmm_mem_init(); 255 if (error) 256 return (error); 257 258 if (vmm_is_intel()) 259 ops = &vmm_ops_intel; 260 else if (vmm_is_amd()) 261 ops = &vmm_ops_amd; 262 else 263 return (ENXIO); 264 265 vmm_msr_init(); 266 vmm_resume_p = vmm_resume; 267 268 return (VMM_INIT(vmm_ipinum)); 269 } 270 271 static int 272 vmm_handler(module_t mod, int what, void *arg) 273 { 274 int error; 275 276 switch (what) { 277 case MOD_LOAD: 278 vmmdev_init(); 279 if (ppt_avail_devices() > 0) 280 iommu_init(); 281 error = vmm_init(); 282 if (error == 0) 283 vmm_initialized = 1; 284 break; 285 case MOD_UNLOAD: 286 error = vmmdev_cleanup(); 287 if (error == 0) { 288 vmm_resume_p = NULL; 289 iommu_cleanup(); 290 if (vmm_ipinum != IPI_AST) 291 vmm_ipi_free(vmm_ipinum); 292 error = VMM_CLEANUP(); 293 /* 294 * Something bad happened - prevent new 295 * VMs from being created 296 */ 297 if (error) 298 vmm_initialized = 0; 299 } 300 break; 301 default: 302 error = 0; 303 break; 304 } 305 return (error); 306 } 307 308 static moduledata_t vmm_kmod = { 309 "vmm", 310 vmm_handler, 311 NULL 312 }; 313 314 /* 315 * vmm initialization has the following dependencies: 316 * 317 * - iommu initialization must happen after the pci passthru driver has had 318 * a chance to attach to any passthru devices (after SI_SUB_CONFIGURE). 319 * 320 * - VT-x initialization requires smp_rendezvous() and therefore must happen 321 * after SMP is fully functional (after SI_SUB_SMP). 322 */ 323 DECLARE_MODULE(vmm, vmm_kmod, SI_SUB_SMP + 1, SI_ORDER_ANY); 324 MODULE_VERSION(vmm, 1); 325 326 int 327 vm_create(const char *name, struct vm **retvm) 328 { 329 int i; 330 struct vm *vm; 331 struct vmspace *vmspace; 332 333 const int BSP = 0; 334 335 /* 336 * If vmm.ko could not be successfully initialized then don't attempt 337 * to create the virtual machine. 338 */ 339 if (!vmm_initialized) 340 return (ENXIO); 341 342 if (name == NULL || strlen(name) >= VM_MAX_NAMELEN) 343 return (EINVAL); 344 345 vmspace = VMSPACE_ALLOC(VM_MIN_ADDRESS, VM_MAXUSER_ADDRESS); 346 if (vmspace == NULL) 347 return (ENOMEM); 348 349 vm = malloc(sizeof(struct vm), M_VM, M_WAITOK | M_ZERO); 350 strcpy(vm->name, name); 351 vm->vmspace = vmspace; 352 mtx_init(&vm->rendezvous_mtx, "vm rendezvous lock", 0, MTX_DEF); 353 vm->cookie = VMINIT(vm, vmspace_pmap(vmspace)); 354 vm->vioapic = vioapic_init(vm); 355 vm->vhpet = vhpet_init(vm); 356 vm->vatpic = vatpic_init(vm); 357 vm->vatpit = vatpit_init(vm); 358 359 for (i = 0; i < VM_MAXCPU; i++) { 360 vcpu_init(vm, i); 361 guest_msrs_init(vm, i); 362 } 363 364 vm_activate_cpu(vm, BSP); 365 366 *retvm = vm; 367 return (0); 368 } 369 370 static void 371 vm_free_mem_seg(struct vm *vm, struct mem_seg *seg) 372 { 373 374 if (seg->object != NULL) 375 vmm_mem_free(vm->vmspace, seg->gpa, seg->len); 376 377 bzero(seg, sizeof(*seg)); 378 } 379 380 void 381 vm_destroy(struct vm *vm) 382 { 383 int i; 384 385 ppt_unassign_all(vm); 386 387 if (vm->iommu != NULL) 388 iommu_destroy_domain(vm->iommu); 389 390 vatpit_cleanup(vm->vatpit); 391 vhpet_cleanup(vm->vhpet); 392 vatpic_cleanup(vm->vatpic); 393 vioapic_cleanup(vm->vioapic); 394 395 for (i = 0; i < vm->num_mem_segs; i++) 396 vm_free_mem_seg(vm, &vm->mem_segs[i]); 397 398 vm->num_mem_segs = 0; 399 400 for (i = 0; i < VM_MAXCPU; i++) 401 vcpu_cleanup(vm, i); 402 403 VMSPACE_FREE(vm->vmspace); 404 405 VMCLEANUP(vm->cookie); 406 407 free(vm, M_VM); 408 } 409 410 const char * 411 vm_name(struct vm *vm) 412 { 413 return (vm->name); 414 } 415 416 int 417 vm_map_mmio(struct vm *vm, vm_paddr_t gpa, size_t len, vm_paddr_t hpa) 418 { 419 vm_object_t obj; 420 421 if ((obj = vmm_mmio_alloc(vm->vmspace, gpa, len, hpa)) == NULL) 422 return (ENOMEM); 423 else 424 return (0); 425 } 426 427 int 428 vm_unmap_mmio(struct vm *vm, vm_paddr_t gpa, size_t len) 429 { 430 431 vmm_mmio_free(vm->vmspace, gpa, len); 432 return (0); 433 } 434 435 boolean_t 436 vm_mem_allocated(struct vm *vm, vm_paddr_t gpa) 437 { 438 int i; 439 vm_paddr_t gpabase, gpalimit; 440 441 for (i = 0; i < vm->num_mem_segs; i++) { 442 gpabase = vm->mem_segs[i].gpa; 443 gpalimit = gpabase + vm->mem_segs[i].len; 444 if (gpa >= gpabase && gpa < gpalimit) 445 return (TRUE); /* 'gpa' is regular memory */ 446 } 447 448 if (ppt_is_mmio(vm, gpa)) 449 return (TRUE); /* 'gpa' is pci passthru mmio */ 450 451 return (FALSE); 452 } 453 454 int 455 vm_malloc(struct vm *vm, vm_paddr_t gpa, size_t len) 456 { 457 int available, allocated; 458 struct mem_seg *seg; 459 vm_object_t object; 460 vm_paddr_t g; 461 462 if ((gpa & PAGE_MASK) || (len & PAGE_MASK) || len == 0) 463 return (EINVAL); 464 465 available = allocated = 0; 466 g = gpa; 467 while (g < gpa + len) { 468 if (vm_mem_allocated(vm, g)) 469 allocated++; 470 else 471 available++; 472 473 g += PAGE_SIZE; 474 } 475 476 /* 477 * If there are some allocated and some available pages in the address 478 * range then it is an error. 479 */ 480 if (allocated && available) 481 return (EINVAL); 482 483 /* 484 * If the entire address range being requested has already been 485 * allocated then there isn't anything more to do. 486 */ 487 if (allocated && available == 0) 488 return (0); 489 490 if (vm->num_mem_segs >= VM_MAX_MEMORY_SEGMENTS) 491 return (E2BIG); 492 493 seg = &vm->mem_segs[vm->num_mem_segs]; 494 495 if ((object = vmm_mem_alloc(vm->vmspace, gpa, len)) == NULL) 496 return (ENOMEM); 497 498 seg->gpa = gpa; 499 seg->len = len; 500 seg->object = object; 501 seg->wired = FALSE; 502 503 vm->num_mem_segs++; 504 505 return (0); 506 } 507 508 static void 509 vm_gpa_unwire(struct vm *vm) 510 { 511 int i, rv; 512 struct mem_seg *seg; 513 514 for (i = 0; i < vm->num_mem_segs; i++) { 515 seg = &vm->mem_segs[i]; 516 if (!seg->wired) 517 continue; 518 519 rv = vm_map_unwire(&vm->vmspace->vm_map, 520 seg->gpa, seg->gpa + seg->len, 521 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES); 522 KASSERT(rv == KERN_SUCCESS, ("vm(%s) memory segment " 523 "%#lx/%ld could not be unwired: %d", 524 vm_name(vm), seg->gpa, seg->len, rv)); 525 526 seg->wired = FALSE; 527 } 528 } 529 530 static int 531 vm_gpa_wire(struct vm *vm) 532 { 533 int i, rv; 534 struct mem_seg *seg; 535 536 for (i = 0; i < vm->num_mem_segs; i++) { 537 seg = &vm->mem_segs[i]; 538 if (seg->wired) 539 continue; 540 541 /* XXX rlimits? */ 542 rv = vm_map_wire(&vm->vmspace->vm_map, 543 seg->gpa, seg->gpa + seg->len, 544 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES); 545 if (rv != KERN_SUCCESS) 546 break; 547 548 seg->wired = TRUE; 549 } 550 551 if (i < vm->num_mem_segs) { 552 /* 553 * Undo the wiring before returning an error. 554 */ 555 vm_gpa_unwire(vm); 556 return (EAGAIN); 557 } 558 559 return (0); 560 } 561 562 static void 563 vm_iommu_modify(struct vm *vm, boolean_t map) 564 { 565 int i, sz; 566 vm_paddr_t gpa, hpa; 567 struct mem_seg *seg; 568 void *vp, *cookie, *host_domain; 569 570 sz = PAGE_SIZE; 571 host_domain = iommu_host_domain(); 572 573 for (i = 0; i < vm->num_mem_segs; i++) { 574 seg = &vm->mem_segs[i]; 575 KASSERT(seg->wired, ("vm(%s) memory segment %#lx/%ld not wired", 576 vm_name(vm), seg->gpa, seg->len)); 577 578 gpa = seg->gpa; 579 while (gpa < seg->gpa + seg->len) { 580 vp = vm_gpa_hold(vm, gpa, PAGE_SIZE, VM_PROT_WRITE, 581 &cookie); 582 KASSERT(vp != NULL, ("vm(%s) could not map gpa %#lx", 583 vm_name(vm), gpa)); 584 585 vm_gpa_release(cookie); 586 587 hpa = DMAP_TO_PHYS((uintptr_t)vp); 588 if (map) { 589 iommu_create_mapping(vm->iommu, gpa, hpa, sz); 590 iommu_remove_mapping(host_domain, hpa, sz); 591 } else { 592 iommu_remove_mapping(vm->iommu, gpa, sz); 593 iommu_create_mapping(host_domain, hpa, hpa, sz); 594 } 595 596 gpa += PAGE_SIZE; 597 } 598 } 599 600 /* 601 * Invalidate the cached translations associated with the domain 602 * from which pages were removed. 603 */ 604 if (map) 605 iommu_invalidate_tlb(host_domain); 606 else 607 iommu_invalidate_tlb(vm->iommu); 608 } 609 610 #define vm_iommu_unmap(vm) vm_iommu_modify((vm), FALSE) 611 #define vm_iommu_map(vm) vm_iommu_modify((vm), TRUE) 612 613 int 614 vm_unassign_pptdev(struct vm *vm, int bus, int slot, int func) 615 { 616 int error; 617 618 error = ppt_unassign_device(vm, bus, slot, func); 619 if (error) 620 return (error); 621 622 if (ppt_assigned_devices(vm) == 0) { 623 vm_iommu_unmap(vm); 624 vm_gpa_unwire(vm); 625 } 626 return (0); 627 } 628 629 int 630 vm_assign_pptdev(struct vm *vm, int bus, int slot, int func) 631 { 632 int error; 633 vm_paddr_t maxaddr; 634 635 /* 636 * Virtual machines with pci passthru devices get special treatment: 637 * - the guest physical memory is wired 638 * - the iommu is programmed to do the 'gpa' to 'hpa' translation 639 * 640 * We need to do this before the first pci passthru device is attached. 641 */ 642 if (ppt_assigned_devices(vm) == 0) { 643 KASSERT(vm->iommu == NULL, 644 ("vm_assign_pptdev: iommu must be NULL")); 645 maxaddr = vmm_mem_maxaddr(); 646 vm->iommu = iommu_create_domain(maxaddr); 647 648 error = vm_gpa_wire(vm); 649 if (error) 650 return (error); 651 652 vm_iommu_map(vm); 653 } 654 655 error = ppt_assign_device(vm, bus, slot, func); 656 return (error); 657 } 658 659 void * 660 vm_gpa_hold(struct vm *vm, vm_paddr_t gpa, size_t len, int reqprot, 661 void **cookie) 662 { 663 int count, pageoff; 664 vm_page_t m; 665 666 pageoff = gpa & PAGE_MASK; 667 if (len > PAGE_SIZE - pageoff) 668 panic("vm_gpa_hold: invalid gpa/len: 0x%016lx/%lu", gpa, len); 669 670 count = vm_fault_quick_hold_pages(&vm->vmspace->vm_map, 671 trunc_page(gpa), PAGE_SIZE, reqprot, &m, 1); 672 673 if (count == 1) { 674 *cookie = m; 675 return ((void *)(PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m)) + pageoff)); 676 } else { 677 *cookie = NULL; 678 return (NULL); 679 } 680 } 681 682 void 683 vm_gpa_release(void *cookie) 684 { 685 vm_page_t m = cookie; 686 687 vm_page_lock(m); 688 vm_page_unhold(m); 689 vm_page_unlock(m); 690 } 691 692 int 693 vm_gpabase2memseg(struct vm *vm, vm_paddr_t gpabase, 694 struct vm_memory_segment *seg) 695 { 696 int i; 697 698 for (i = 0; i < vm->num_mem_segs; i++) { 699 if (gpabase == vm->mem_segs[i].gpa) { 700 seg->gpa = vm->mem_segs[i].gpa; 701 seg->len = vm->mem_segs[i].len; 702 seg->wired = vm->mem_segs[i].wired; 703 return (0); 704 } 705 } 706 return (-1); 707 } 708 709 int 710 vm_get_memobj(struct vm *vm, vm_paddr_t gpa, size_t len, 711 vm_offset_t *offset, struct vm_object **object) 712 { 713 int i; 714 size_t seg_len; 715 vm_paddr_t seg_gpa; 716 vm_object_t seg_obj; 717 718 for (i = 0; i < vm->num_mem_segs; i++) { 719 if ((seg_obj = vm->mem_segs[i].object) == NULL) 720 continue; 721 722 seg_gpa = vm->mem_segs[i].gpa; 723 seg_len = vm->mem_segs[i].len; 724 725 if (gpa >= seg_gpa && gpa < seg_gpa + seg_len) { 726 *offset = gpa - seg_gpa; 727 *object = seg_obj; 728 vm_object_reference(seg_obj); 729 return (0); 730 } 731 } 732 733 return (EINVAL); 734 } 735 736 int 737 vm_get_register(struct vm *vm, int vcpu, int reg, uint64_t *retval) 738 { 739 740 if (vcpu < 0 || vcpu >= VM_MAXCPU) 741 return (EINVAL); 742 743 if (reg >= VM_REG_LAST) 744 return (EINVAL); 745 746 return (VMGETREG(vm->cookie, vcpu, reg, retval)); 747 } 748 749 int 750 vm_set_register(struct vm *vm, int vcpu, int reg, uint64_t val) 751 { 752 753 if (vcpu < 0 || vcpu >= VM_MAXCPU) 754 return (EINVAL); 755 756 if (reg >= VM_REG_LAST) 757 return (EINVAL); 758 759 return (VMSETREG(vm->cookie, vcpu, reg, val)); 760 } 761 762 static boolean_t 763 is_descriptor_table(int reg) 764 { 765 766 switch (reg) { 767 case VM_REG_GUEST_IDTR: 768 case VM_REG_GUEST_GDTR: 769 return (TRUE); 770 default: 771 return (FALSE); 772 } 773 } 774 775 static boolean_t 776 is_segment_register(int reg) 777 { 778 779 switch (reg) { 780 case VM_REG_GUEST_ES: 781 case VM_REG_GUEST_CS: 782 case VM_REG_GUEST_SS: 783 case VM_REG_GUEST_DS: 784 case VM_REG_GUEST_FS: 785 case VM_REG_GUEST_GS: 786 case VM_REG_GUEST_TR: 787 case VM_REG_GUEST_LDTR: 788 return (TRUE); 789 default: 790 return (FALSE); 791 } 792 } 793 794 int 795 vm_get_seg_desc(struct vm *vm, int vcpu, int reg, 796 struct seg_desc *desc) 797 { 798 799 if (vcpu < 0 || vcpu >= VM_MAXCPU) 800 return (EINVAL); 801 802 if (!is_segment_register(reg) && !is_descriptor_table(reg)) 803 return (EINVAL); 804 805 return (VMGETDESC(vm->cookie, vcpu, reg, desc)); 806 } 807 808 int 809 vm_set_seg_desc(struct vm *vm, int vcpu, int reg, 810 struct seg_desc *desc) 811 { 812 if (vcpu < 0 || vcpu >= VM_MAXCPU) 813 return (EINVAL); 814 815 if (!is_segment_register(reg) && !is_descriptor_table(reg)) 816 return (EINVAL); 817 818 return (VMSETDESC(vm->cookie, vcpu, reg, desc)); 819 } 820 821 static void 822 restore_guest_fpustate(struct vcpu *vcpu) 823 { 824 825 /* flush host state to the pcb */ 826 fpuexit(curthread); 827 828 /* restore guest FPU state */ 829 fpu_stop_emulating(); 830 fpurestore(vcpu->guestfpu); 831 832 /* restore guest XCR0 if XSAVE is enabled in the host */ 833 if (rcr4() & CR4_XSAVE) 834 load_xcr(0, vcpu->guest_xcr0); 835 836 /* 837 * The FPU is now "dirty" with the guest's state so turn on emulation 838 * to trap any access to the FPU by the host. 839 */ 840 fpu_start_emulating(); 841 } 842 843 static void 844 save_guest_fpustate(struct vcpu *vcpu) 845 { 846 847 if ((rcr0() & CR0_TS) == 0) 848 panic("fpu emulation not enabled in host!"); 849 850 /* save guest XCR0 and restore host XCR0 */ 851 if (rcr4() & CR4_XSAVE) { 852 vcpu->guest_xcr0 = rxcr(0); 853 load_xcr(0, vmm_get_host_xcr0()); 854 } 855 856 /* save guest FPU state */ 857 fpu_stop_emulating(); 858 fpusave(vcpu->guestfpu); 859 fpu_start_emulating(); 860 } 861 862 static VMM_STAT(VCPU_IDLE_TICKS, "number of ticks vcpu was idle"); 863 864 static int 865 vcpu_set_state_locked(struct vcpu *vcpu, enum vcpu_state newstate, 866 bool from_idle) 867 { 868 int error; 869 870 vcpu_assert_locked(vcpu); 871 872 /* 873 * State transitions from the vmmdev_ioctl() must always begin from 874 * the VCPU_IDLE state. This guarantees that there is only a single 875 * ioctl() operating on a vcpu at any point. 876 */ 877 if (from_idle) { 878 while (vcpu->state != VCPU_IDLE) 879 msleep_spin(&vcpu->state, &vcpu->mtx, "vmstat", hz); 880 } else { 881 KASSERT(vcpu->state != VCPU_IDLE, ("invalid transition from " 882 "vcpu idle state")); 883 } 884 885 if (vcpu->state == VCPU_RUNNING) { 886 KASSERT(vcpu->hostcpu == curcpu, ("curcpu %d and hostcpu %d " 887 "mismatch for running vcpu", curcpu, vcpu->hostcpu)); 888 } else { 889 KASSERT(vcpu->hostcpu == NOCPU, ("Invalid hostcpu %d for a " 890 "vcpu that is not running", vcpu->hostcpu)); 891 } 892 893 /* 894 * The following state transitions are allowed: 895 * IDLE -> FROZEN -> IDLE 896 * FROZEN -> RUNNING -> FROZEN 897 * FROZEN -> SLEEPING -> FROZEN 898 */ 899 switch (vcpu->state) { 900 case VCPU_IDLE: 901 case VCPU_RUNNING: 902 case VCPU_SLEEPING: 903 error = (newstate != VCPU_FROZEN); 904 break; 905 case VCPU_FROZEN: 906 error = (newstate == VCPU_FROZEN); 907 break; 908 default: 909 error = 1; 910 break; 911 } 912 913 if (error) 914 return (EBUSY); 915 916 vcpu->state = newstate; 917 if (newstate == VCPU_RUNNING) 918 vcpu->hostcpu = curcpu; 919 else 920 vcpu->hostcpu = NOCPU; 921 922 if (newstate == VCPU_IDLE) 923 wakeup(&vcpu->state); 924 925 return (0); 926 } 927 928 static void 929 vcpu_require_state(struct vm *vm, int vcpuid, enum vcpu_state newstate) 930 { 931 int error; 932 933 if ((error = vcpu_set_state(vm, vcpuid, newstate, false)) != 0) 934 panic("Error %d setting state to %d\n", error, newstate); 935 } 936 937 static void 938 vcpu_require_state_locked(struct vcpu *vcpu, enum vcpu_state newstate) 939 { 940 int error; 941 942 if ((error = vcpu_set_state_locked(vcpu, newstate, false)) != 0) 943 panic("Error %d setting state to %d", error, newstate); 944 } 945 946 static void 947 vm_set_rendezvous_func(struct vm *vm, vm_rendezvous_func_t func) 948 { 949 950 KASSERT(mtx_owned(&vm->rendezvous_mtx), ("rendezvous_mtx not locked")); 951 952 /* 953 * Update 'rendezvous_func' and execute a write memory barrier to 954 * ensure that it is visible across all host cpus. This is not needed 955 * for correctness but it does ensure that all the vcpus will notice 956 * that the rendezvous is requested immediately. 957 */ 958 vm->rendezvous_func = func; 959 wmb(); 960 } 961 962 #define RENDEZVOUS_CTR0(vm, vcpuid, fmt) \ 963 do { \ 964 if (vcpuid >= 0) \ 965 VCPU_CTR0(vm, vcpuid, fmt); \ 966 else \ 967 VM_CTR0(vm, fmt); \ 968 } while (0) 969 970 static void 971 vm_handle_rendezvous(struct vm *vm, int vcpuid) 972 { 973 974 KASSERT(vcpuid == -1 || (vcpuid >= 0 && vcpuid < VM_MAXCPU), 975 ("vm_handle_rendezvous: invalid vcpuid %d", vcpuid)); 976 977 mtx_lock(&vm->rendezvous_mtx); 978 while (vm->rendezvous_func != NULL) { 979 /* 'rendezvous_req_cpus' must be a subset of 'active_cpus' */ 980 CPU_AND(&vm->rendezvous_req_cpus, &vm->active_cpus); 981 982 if (vcpuid != -1 && 983 CPU_ISSET(vcpuid, &vm->rendezvous_req_cpus) && 984 !CPU_ISSET(vcpuid, &vm->rendezvous_done_cpus)) { 985 VCPU_CTR0(vm, vcpuid, "Calling rendezvous func"); 986 (*vm->rendezvous_func)(vm, vcpuid, vm->rendezvous_arg); 987 CPU_SET(vcpuid, &vm->rendezvous_done_cpus); 988 } 989 if (CPU_CMP(&vm->rendezvous_req_cpus, 990 &vm->rendezvous_done_cpus) == 0) { 991 VCPU_CTR0(vm, vcpuid, "Rendezvous completed"); 992 vm_set_rendezvous_func(vm, NULL); 993 wakeup(&vm->rendezvous_func); 994 break; 995 } 996 RENDEZVOUS_CTR0(vm, vcpuid, "Wait for rendezvous completion"); 997 mtx_sleep(&vm->rendezvous_func, &vm->rendezvous_mtx, 0, 998 "vmrndv", 0); 999 } 1000 mtx_unlock(&vm->rendezvous_mtx); 1001 } 1002 1003 /* 1004 * Emulate a guest 'hlt' by sleeping until the vcpu is ready to run. 1005 */ 1006 static int 1007 vm_handle_hlt(struct vm *vm, int vcpuid, bool intr_disabled, bool *retu) 1008 { 1009 struct vm_exit *vmexit; 1010 struct vcpu *vcpu; 1011 int t, timo, spindown; 1012 1013 vcpu = &vm->vcpu[vcpuid]; 1014 spindown = 0; 1015 1016 vcpu_lock(vcpu); 1017 1018 /* 1019 * Do a final check for pending NMI or interrupts before 1020 * really putting this thread to sleep. 1021 * 1022 * These interrupts could have happened any time after we 1023 * returned from VMRUN() and before we grabbed the vcpu lock. 1024 */ 1025 if (vm->rendezvous_func == NULL && 1026 !vm_nmi_pending(vm, vcpuid) && 1027 (intr_disabled || !vlapic_pending_intr(vcpu->vlapic, NULL))) { 1028 t = ticks; 1029 vcpu_require_state_locked(vcpu, VCPU_SLEEPING); 1030 if (vlapic_enabled(vcpu->vlapic)) { 1031 /* 1032 * XXX msleep_spin() is not interruptible so use the 1033 * 'timo' to put an upper bound on the sleep time. 1034 */ 1035 timo = hz; 1036 msleep_spin(vcpu, &vcpu->mtx, "vmidle", timo); 1037 } else { 1038 /* 1039 * Spindown the vcpu if the apic is disabled and it 1040 * had entered the halted state. 1041 */ 1042 spindown = 1; 1043 } 1044 vcpu_require_state_locked(vcpu, VCPU_FROZEN); 1045 vmm_stat_incr(vm, vcpuid, VCPU_IDLE_TICKS, ticks - t); 1046 } 1047 vcpu_unlock(vcpu); 1048 1049 /* 1050 * Since 'vm_deactivate_cpu()' grabs a sleep mutex we must call it 1051 * outside the confines of the vcpu spinlock. 1052 */ 1053 if (spindown) { 1054 *retu = true; 1055 vmexit = vm_exitinfo(vm, vcpuid); 1056 vmexit->exitcode = VM_EXITCODE_SPINDOWN_CPU; 1057 vm_deactivate_cpu(vm, vcpuid); 1058 VCPU_CTR0(vm, vcpuid, "spinning down cpu"); 1059 } 1060 1061 return (0); 1062 } 1063 1064 static int 1065 vm_handle_paging(struct vm *vm, int vcpuid, bool *retu) 1066 { 1067 int rv, ftype; 1068 struct vm_map *map; 1069 struct vcpu *vcpu; 1070 struct vm_exit *vme; 1071 1072 vcpu = &vm->vcpu[vcpuid]; 1073 vme = &vcpu->exitinfo; 1074 1075 ftype = vme->u.paging.fault_type; 1076 KASSERT(ftype == VM_PROT_READ || 1077 ftype == VM_PROT_WRITE || ftype == VM_PROT_EXECUTE, 1078 ("vm_handle_paging: invalid fault_type %d", ftype)); 1079 1080 if (ftype == VM_PROT_READ || ftype == VM_PROT_WRITE) { 1081 rv = pmap_emulate_accessed_dirty(vmspace_pmap(vm->vmspace), 1082 vme->u.paging.gpa, ftype); 1083 if (rv == 0) 1084 goto done; 1085 } 1086 1087 map = &vm->vmspace->vm_map; 1088 rv = vm_fault(map, vme->u.paging.gpa, ftype, VM_FAULT_NORMAL); 1089 1090 VCPU_CTR3(vm, vcpuid, "vm_handle_paging rv = %d, gpa = %#lx, " 1091 "ftype = %d", rv, vme->u.paging.gpa, ftype); 1092 1093 if (rv != KERN_SUCCESS) 1094 return (EFAULT); 1095 done: 1096 /* restart execution at the faulting instruction */ 1097 vme->inst_length = 0; 1098 1099 return (0); 1100 } 1101 1102 static int 1103 vm_handle_inst_emul(struct vm *vm, int vcpuid, bool *retu) 1104 { 1105 struct vie *vie; 1106 struct vcpu *vcpu; 1107 struct vm_exit *vme; 1108 int error, inst_length; 1109 uint64_t rip, gla, gpa, cr3; 1110 enum vie_cpu_mode cpu_mode; 1111 enum vie_paging_mode paging_mode; 1112 mem_region_read_t mread; 1113 mem_region_write_t mwrite; 1114 1115 vcpu = &vm->vcpu[vcpuid]; 1116 vme = &vcpu->exitinfo; 1117 1118 rip = vme->rip; 1119 inst_length = vme->inst_length; 1120 1121 gla = vme->u.inst_emul.gla; 1122 gpa = vme->u.inst_emul.gpa; 1123 cr3 = vme->u.inst_emul.cr3; 1124 cpu_mode = vme->u.inst_emul.cpu_mode; 1125 paging_mode = vme->u.inst_emul.paging_mode; 1126 vie = &vme->u.inst_emul.vie; 1127 1128 vie_init(vie); 1129 1130 /* Fetch, decode and emulate the faulting instruction */ 1131 if (vmm_fetch_instruction(vm, vcpuid, rip, inst_length, cr3, 1132 paging_mode, vie) != 0) 1133 return (EFAULT); 1134 1135 if (vmm_decode_instruction(vm, vcpuid, gla, cpu_mode, vie) != 0) 1136 return (EFAULT); 1137 1138 /* return to userland unless this is an in-kernel emulated device */ 1139 if (gpa >= DEFAULT_APIC_BASE && gpa < DEFAULT_APIC_BASE + PAGE_SIZE) { 1140 mread = lapic_mmio_read; 1141 mwrite = lapic_mmio_write; 1142 } else if (gpa >= VIOAPIC_BASE && gpa < VIOAPIC_BASE + VIOAPIC_SIZE) { 1143 mread = vioapic_mmio_read; 1144 mwrite = vioapic_mmio_write; 1145 } else if (gpa >= VHPET_BASE && gpa < VHPET_BASE + VHPET_SIZE) { 1146 mread = vhpet_mmio_read; 1147 mwrite = vhpet_mmio_write; 1148 } else { 1149 *retu = true; 1150 return (0); 1151 } 1152 1153 error = vmm_emulate_instruction(vm, vcpuid, gpa, vie, mread, mwrite, 1154 retu); 1155 1156 return (error); 1157 } 1158 1159 static int 1160 vm_handle_suspend(struct vm *vm, int vcpuid, bool *retu) 1161 { 1162 int i, done; 1163 struct vcpu *vcpu; 1164 1165 done = 0; 1166 vcpu = &vm->vcpu[vcpuid]; 1167 1168 CPU_SET_ATOMIC(vcpuid, &vm->suspended_cpus); 1169 1170 /* 1171 * Wait until all 'active_cpus' have suspended themselves. 1172 * 1173 * Since a VM may be suspended at any time including when one or 1174 * more vcpus are doing a rendezvous we need to call the rendezvous 1175 * handler while we are waiting to prevent a deadlock. 1176 */ 1177 vcpu_lock(vcpu); 1178 while (1) { 1179 if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) { 1180 VCPU_CTR0(vm, vcpuid, "All vcpus suspended"); 1181 break; 1182 } 1183 1184 if (vm->rendezvous_func == NULL) { 1185 VCPU_CTR0(vm, vcpuid, "Sleeping during suspend"); 1186 vcpu_require_state_locked(vcpu, VCPU_SLEEPING); 1187 msleep_spin(vcpu, &vcpu->mtx, "vmsusp", hz); 1188 vcpu_require_state_locked(vcpu, VCPU_FROZEN); 1189 } else { 1190 VCPU_CTR0(vm, vcpuid, "Rendezvous during suspend"); 1191 vcpu_unlock(vcpu); 1192 vm_handle_rendezvous(vm, vcpuid); 1193 vcpu_lock(vcpu); 1194 } 1195 } 1196 vcpu_unlock(vcpu); 1197 1198 /* 1199 * Wakeup the other sleeping vcpus and return to userspace. 1200 */ 1201 for (i = 0; i < VM_MAXCPU; i++) { 1202 if (CPU_ISSET(i, &vm->suspended_cpus)) { 1203 vcpu_notify_event(vm, i, false); 1204 } 1205 } 1206 1207 *retu = true; 1208 return (0); 1209 } 1210 1211 int 1212 vm_suspend(struct vm *vm) 1213 { 1214 1215 if (atomic_cmpset_int(&vm->suspend, 0, 1)) { 1216 VM_CTR0(vm, "virtual machine suspended"); 1217 return (0); 1218 } else { 1219 VM_CTR0(vm, "virtual machine already suspended"); 1220 return (EALREADY); 1221 } 1222 } 1223 1224 int 1225 vm_run(struct vm *vm, struct vm_run *vmrun) 1226 { 1227 int error, vcpuid; 1228 struct vcpu *vcpu; 1229 struct pcb *pcb; 1230 uint64_t tscval, rip; 1231 struct vm_exit *vme; 1232 bool retu, intr_disabled; 1233 pmap_t pmap; 1234 void *rptr, *sptr; 1235 1236 vcpuid = vmrun->cpuid; 1237 1238 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1239 return (EINVAL); 1240 1241 rptr = &vm->rendezvous_func; 1242 sptr = &vm->suspend; 1243 pmap = vmspace_pmap(vm->vmspace); 1244 vcpu = &vm->vcpu[vcpuid]; 1245 vme = &vcpu->exitinfo; 1246 rip = vmrun->rip; 1247 restart: 1248 critical_enter(); 1249 1250 KASSERT(!CPU_ISSET(curcpu, &pmap->pm_active), 1251 ("vm_run: absurd pm_active")); 1252 1253 tscval = rdtsc(); 1254 1255 pcb = PCPU_GET(curpcb); 1256 set_pcb_flags(pcb, PCB_FULL_IRET); 1257 1258 restore_guest_msrs(vm, vcpuid); 1259 restore_guest_fpustate(vcpu); 1260 1261 vcpu_require_state(vm, vcpuid, VCPU_RUNNING); 1262 error = VMRUN(vm->cookie, vcpuid, rip, pmap, rptr, sptr); 1263 vcpu_require_state(vm, vcpuid, VCPU_FROZEN); 1264 1265 save_guest_fpustate(vcpu); 1266 restore_host_msrs(vm, vcpuid); 1267 1268 vmm_stat_incr(vm, vcpuid, VCPU_TOTAL_RUNTIME, rdtsc() - tscval); 1269 1270 critical_exit(); 1271 1272 if (error == 0) { 1273 retu = false; 1274 switch (vme->exitcode) { 1275 case VM_EXITCODE_SUSPENDED: 1276 error = vm_handle_suspend(vm, vcpuid, &retu); 1277 break; 1278 case VM_EXITCODE_IOAPIC_EOI: 1279 vioapic_process_eoi(vm, vcpuid, 1280 vme->u.ioapic_eoi.vector); 1281 break; 1282 case VM_EXITCODE_RENDEZVOUS: 1283 vm_handle_rendezvous(vm, vcpuid); 1284 error = 0; 1285 break; 1286 case VM_EXITCODE_HLT: 1287 intr_disabled = ((vme->u.hlt.rflags & PSL_I) == 0); 1288 error = vm_handle_hlt(vm, vcpuid, intr_disabled, &retu); 1289 break; 1290 case VM_EXITCODE_PAGING: 1291 error = vm_handle_paging(vm, vcpuid, &retu); 1292 break; 1293 case VM_EXITCODE_INST_EMUL: 1294 error = vm_handle_inst_emul(vm, vcpuid, &retu); 1295 break; 1296 default: 1297 retu = true; /* handled in userland */ 1298 break; 1299 } 1300 } 1301 1302 if (error == 0 && retu == false) { 1303 rip = vme->rip + vme->inst_length; 1304 goto restart; 1305 } 1306 1307 /* copy the exit information */ 1308 bcopy(vme, &vmrun->vm_exit, sizeof(struct vm_exit)); 1309 return (error); 1310 } 1311 1312 int 1313 vm_inject_exception(struct vm *vm, int vcpuid, struct vm_exception *exception) 1314 { 1315 struct vcpu *vcpu; 1316 1317 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1318 return (EINVAL); 1319 1320 if (exception->vector < 0 || exception->vector >= 32) 1321 return (EINVAL); 1322 1323 vcpu = &vm->vcpu[vcpuid]; 1324 1325 if (vcpu->exception_pending) { 1326 VCPU_CTR2(vm, vcpuid, "Unable to inject exception %d due to " 1327 "pending exception %d", exception->vector, 1328 vcpu->exception.vector); 1329 return (EBUSY); 1330 } 1331 1332 vcpu->exception_pending = 1; 1333 vcpu->exception = *exception; 1334 VCPU_CTR1(vm, vcpuid, "Exception %d pending", exception->vector); 1335 return (0); 1336 } 1337 1338 int 1339 vm_exception_pending(struct vm *vm, int vcpuid, struct vm_exception *exception) 1340 { 1341 struct vcpu *vcpu; 1342 int pending; 1343 1344 KASSERT(vcpuid >= 0 && vcpuid < VM_MAXCPU, ("invalid vcpu %d", vcpuid)); 1345 1346 vcpu = &vm->vcpu[vcpuid]; 1347 pending = vcpu->exception_pending; 1348 if (pending) { 1349 vcpu->exception_pending = 0; 1350 *exception = vcpu->exception; 1351 VCPU_CTR1(vm, vcpuid, "Exception %d delivered", 1352 exception->vector); 1353 } 1354 return (pending); 1355 } 1356 1357 static void 1358 vm_inject_fault(struct vm *vm, int vcpuid, struct vm_exception *exception) 1359 { 1360 struct vm_exit *vmexit; 1361 int error; 1362 1363 error = vm_inject_exception(vm, vcpuid, exception); 1364 KASSERT(error == 0, ("vm_inject_exception error %d", error)); 1365 1366 /* 1367 * A fault-like exception allows the instruction to be restarted 1368 * after the exception handler returns. 1369 * 1370 * By setting the inst_length to 0 we ensure that the instruction 1371 * pointer remains at the faulting instruction. 1372 */ 1373 vmexit = vm_exitinfo(vm, vcpuid); 1374 vmexit->inst_length = 0; 1375 } 1376 1377 void 1378 vm_inject_gp(struct vm *vm, int vcpuid) 1379 { 1380 struct vm_exception gpf = { 1381 .vector = IDT_GP, 1382 .error_code_valid = 1, 1383 .error_code = 0 1384 }; 1385 1386 vm_inject_fault(vm, vcpuid, &gpf); 1387 } 1388 1389 void 1390 vm_inject_ud(struct vm *vm, int vcpuid) 1391 { 1392 struct vm_exception udf = { 1393 .vector = IDT_UD, 1394 .error_code_valid = 0 1395 }; 1396 1397 vm_inject_fault(vm, vcpuid, &udf); 1398 } 1399 1400 static VMM_STAT(VCPU_NMI_COUNT, "number of NMIs delivered to vcpu"); 1401 1402 int 1403 vm_inject_nmi(struct vm *vm, int vcpuid) 1404 { 1405 struct vcpu *vcpu; 1406 1407 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1408 return (EINVAL); 1409 1410 vcpu = &vm->vcpu[vcpuid]; 1411 1412 vcpu->nmi_pending = 1; 1413 vcpu_notify_event(vm, vcpuid, false); 1414 return (0); 1415 } 1416 1417 int 1418 vm_nmi_pending(struct vm *vm, int vcpuid) 1419 { 1420 struct vcpu *vcpu; 1421 1422 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1423 panic("vm_nmi_pending: invalid vcpuid %d", vcpuid); 1424 1425 vcpu = &vm->vcpu[vcpuid]; 1426 1427 return (vcpu->nmi_pending); 1428 } 1429 1430 void 1431 vm_nmi_clear(struct vm *vm, int vcpuid) 1432 { 1433 struct vcpu *vcpu; 1434 1435 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1436 panic("vm_nmi_pending: invalid vcpuid %d", vcpuid); 1437 1438 vcpu = &vm->vcpu[vcpuid]; 1439 1440 if (vcpu->nmi_pending == 0) 1441 panic("vm_nmi_clear: inconsistent nmi_pending state"); 1442 1443 vcpu->nmi_pending = 0; 1444 vmm_stat_incr(vm, vcpuid, VCPU_NMI_COUNT, 1); 1445 } 1446 1447 static VMM_STAT(VCPU_EXTINT_COUNT, "number of ExtINTs delivered to vcpu"); 1448 1449 int 1450 vm_inject_extint(struct vm *vm, int vcpuid) 1451 { 1452 struct vcpu *vcpu; 1453 1454 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1455 return (EINVAL); 1456 1457 vcpu = &vm->vcpu[vcpuid]; 1458 1459 vcpu->extint_pending = 1; 1460 vcpu_notify_event(vm, vcpuid, false); 1461 return (0); 1462 } 1463 1464 int 1465 vm_extint_pending(struct vm *vm, int vcpuid) 1466 { 1467 struct vcpu *vcpu; 1468 1469 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1470 panic("vm_extint_pending: invalid vcpuid %d", vcpuid); 1471 1472 vcpu = &vm->vcpu[vcpuid]; 1473 1474 return (vcpu->extint_pending); 1475 } 1476 1477 void 1478 vm_extint_clear(struct vm *vm, int vcpuid) 1479 { 1480 struct vcpu *vcpu; 1481 1482 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1483 panic("vm_extint_pending: invalid vcpuid %d", vcpuid); 1484 1485 vcpu = &vm->vcpu[vcpuid]; 1486 1487 if (vcpu->extint_pending == 0) 1488 panic("vm_extint_clear: inconsistent extint_pending state"); 1489 1490 vcpu->extint_pending = 0; 1491 vmm_stat_incr(vm, vcpuid, VCPU_EXTINT_COUNT, 1); 1492 } 1493 1494 int 1495 vm_get_capability(struct vm *vm, int vcpu, int type, int *retval) 1496 { 1497 if (vcpu < 0 || vcpu >= VM_MAXCPU) 1498 return (EINVAL); 1499 1500 if (type < 0 || type >= VM_CAP_MAX) 1501 return (EINVAL); 1502 1503 return (VMGETCAP(vm->cookie, vcpu, type, retval)); 1504 } 1505 1506 int 1507 vm_set_capability(struct vm *vm, int vcpu, int type, int val) 1508 { 1509 if (vcpu < 0 || vcpu >= VM_MAXCPU) 1510 return (EINVAL); 1511 1512 if (type < 0 || type >= VM_CAP_MAX) 1513 return (EINVAL); 1514 1515 return (VMSETCAP(vm->cookie, vcpu, type, val)); 1516 } 1517 1518 uint64_t * 1519 vm_guest_msrs(struct vm *vm, int cpu) 1520 { 1521 return (vm->vcpu[cpu].guest_msrs); 1522 } 1523 1524 struct vlapic * 1525 vm_lapic(struct vm *vm, int cpu) 1526 { 1527 return (vm->vcpu[cpu].vlapic); 1528 } 1529 1530 struct vioapic * 1531 vm_ioapic(struct vm *vm) 1532 { 1533 1534 return (vm->vioapic); 1535 } 1536 1537 struct vhpet * 1538 vm_hpet(struct vm *vm) 1539 { 1540 1541 return (vm->vhpet); 1542 } 1543 1544 boolean_t 1545 vmm_is_pptdev(int bus, int slot, int func) 1546 { 1547 int found, i, n; 1548 int b, s, f; 1549 char *val, *cp, *cp2; 1550 1551 /* 1552 * XXX 1553 * The length of an environment variable is limited to 128 bytes which 1554 * puts an upper limit on the number of passthru devices that may be 1555 * specified using a single environment variable. 1556 * 1557 * Work around this by scanning multiple environment variable 1558 * names instead of a single one - yuck! 1559 */ 1560 const char *names[] = { "pptdevs", "pptdevs2", "pptdevs3", NULL }; 1561 1562 /* set pptdevs="1/2/3 4/5/6 7/8/9 10/11/12" */ 1563 found = 0; 1564 for (i = 0; names[i] != NULL && !found; i++) { 1565 cp = val = getenv(names[i]); 1566 while (cp != NULL && *cp != '\0') { 1567 if ((cp2 = strchr(cp, ' ')) != NULL) 1568 *cp2 = '\0'; 1569 1570 n = sscanf(cp, "%d/%d/%d", &b, &s, &f); 1571 if (n == 3 && bus == b && slot == s && func == f) { 1572 found = 1; 1573 break; 1574 } 1575 1576 if (cp2 != NULL) 1577 *cp2++ = ' '; 1578 1579 cp = cp2; 1580 } 1581 freeenv(val); 1582 } 1583 return (found); 1584 } 1585 1586 void * 1587 vm_iommu_domain(struct vm *vm) 1588 { 1589 1590 return (vm->iommu); 1591 } 1592 1593 int 1594 vcpu_set_state(struct vm *vm, int vcpuid, enum vcpu_state newstate, 1595 bool from_idle) 1596 { 1597 int error; 1598 struct vcpu *vcpu; 1599 1600 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1601 panic("vm_set_run_state: invalid vcpuid %d", vcpuid); 1602 1603 vcpu = &vm->vcpu[vcpuid]; 1604 1605 vcpu_lock(vcpu); 1606 error = vcpu_set_state_locked(vcpu, newstate, from_idle); 1607 vcpu_unlock(vcpu); 1608 1609 return (error); 1610 } 1611 1612 enum vcpu_state 1613 vcpu_get_state(struct vm *vm, int vcpuid, int *hostcpu) 1614 { 1615 struct vcpu *vcpu; 1616 enum vcpu_state state; 1617 1618 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1619 panic("vm_get_run_state: invalid vcpuid %d", vcpuid); 1620 1621 vcpu = &vm->vcpu[vcpuid]; 1622 1623 vcpu_lock(vcpu); 1624 state = vcpu->state; 1625 if (hostcpu != NULL) 1626 *hostcpu = vcpu->hostcpu; 1627 vcpu_unlock(vcpu); 1628 1629 return (state); 1630 } 1631 1632 void 1633 vm_activate_cpu(struct vm *vm, int vcpuid) 1634 { 1635 1636 KASSERT(vcpuid >= 0 && vcpuid < VM_MAXCPU, 1637 ("vm_activate_cpu: invalid vcpuid %d", vcpuid)); 1638 KASSERT(!CPU_ISSET(vcpuid, &vm->active_cpus), 1639 ("vm_activate_cpu: vcpuid %d is already active", vcpuid)); 1640 1641 VCPU_CTR0(vm, vcpuid, "activated"); 1642 CPU_SET_ATOMIC(vcpuid, &vm->active_cpus); 1643 } 1644 1645 static void 1646 vm_deactivate_cpu(struct vm *vm, int vcpuid) 1647 { 1648 1649 KASSERT(vcpuid >= 0 && vcpuid < VM_MAXCPU, 1650 ("vm_deactivate_cpu: invalid vcpuid %d", vcpuid)); 1651 KASSERT(CPU_ISSET(vcpuid, &vm->active_cpus), 1652 ("vm_deactivate_cpu: vcpuid %d is not active", vcpuid)); 1653 1654 VCPU_CTR0(vm, vcpuid, "deactivated"); 1655 CPU_CLR_ATOMIC(vcpuid, &vm->active_cpus); 1656 1657 /* 1658 * If a vcpu rendezvous is in progress then it could be blocked 1659 * on 'vcpuid' - unblock it before disappearing forever. 1660 */ 1661 mtx_lock(&vm->rendezvous_mtx); 1662 if (vm->rendezvous_func != NULL) { 1663 VCPU_CTR0(vm, vcpuid, "unblock rendezvous after deactivation"); 1664 wakeup(&vm->rendezvous_func); 1665 } 1666 mtx_unlock(&vm->rendezvous_mtx); 1667 } 1668 1669 cpuset_t 1670 vm_active_cpus(struct vm *vm) 1671 { 1672 1673 return (vm->active_cpus); 1674 } 1675 1676 void * 1677 vcpu_stats(struct vm *vm, int vcpuid) 1678 { 1679 1680 return (vm->vcpu[vcpuid].stats); 1681 } 1682 1683 int 1684 vm_get_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state *state) 1685 { 1686 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1687 return (EINVAL); 1688 1689 *state = vm->vcpu[vcpuid].x2apic_state; 1690 1691 return (0); 1692 } 1693 1694 int 1695 vm_set_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state state) 1696 { 1697 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1698 return (EINVAL); 1699 1700 if (state >= X2APIC_STATE_LAST) 1701 return (EINVAL); 1702 1703 vm->vcpu[vcpuid].x2apic_state = state; 1704 1705 vlapic_set_x2apic_state(vm, vcpuid, state); 1706 1707 return (0); 1708 } 1709 1710 /* 1711 * This function is called to ensure that a vcpu "sees" a pending event 1712 * as soon as possible: 1713 * - If the vcpu thread is sleeping then it is woken up. 1714 * - If the vcpu is running on a different host_cpu then an IPI will be directed 1715 * to the host_cpu to cause the vcpu to trap into the hypervisor. 1716 */ 1717 void 1718 vcpu_notify_event(struct vm *vm, int vcpuid, bool lapic_intr) 1719 { 1720 int hostcpu; 1721 struct vcpu *vcpu; 1722 1723 vcpu = &vm->vcpu[vcpuid]; 1724 1725 vcpu_lock(vcpu); 1726 hostcpu = vcpu->hostcpu; 1727 if (vcpu->state == VCPU_RUNNING) { 1728 KASSERT(hostcpu != NOCPU, ("vcpu running on invalid hostcpu")); 1729 if (hostcpu != curcpu) { 1730 if (lapic_intr) { 1731 vlapic_post_intr(vcpu->vlapic, hostcpu, 1732 vmm_ipinum); 1733 } else { 1734 ipi_cpu(hostcpu, vmm_ipinum); 1735 } 1736 } else { 1737 /* 1738 * If the 'vcpu' is running on 'curcpu' then it must 1739 * be sending a notification to itself (e.g. SELF_IPI). 1740 * The pending event will be picked up when the vcpu 1741 * transitions back to guest context. 1742 */ 1743 } 1744 } else { 1745 KASSERT(hostcpu == NOCPU, ("vcpu state %d not consistent " 1746 "with hostcpu %d", vcpu->state, hostcpu)); 1747 if (vcpu->state == VCPU_SLEEPING) 1748 wakeup_one(vcpu); 1749 } 1750 vcpu_unlock(vcpu); 1751 } 1752 1753 struct vmspace * 1754 vm_get_vmspace(struct vm *vm) 1755 { 1756 1757 return (vm->vmspace); 1758 } 1759 1760 int 1761 vm_apicid2vcpuid(struct vm *vm, int apicid) 1762 { 1763 /* 1764 * XXX apic id is assumed to be numerically identical to vcpu id 1765 */ 1766 return (apicid); 1767 } 1768 1769 void 1770 vm_smp_rendezvous(struct vm *vm, int vcpuid, cpuset_t dest, 1771 vm_rendezvous_func_t func, void *arg) 1772 { 1773 int i; 1774 1775 /* 1776 * Enforce that this function is called without any locks 1777 */ 1778 WITNESS_WARN(WARN_PANIC, NULL, "vm_smp_rendezvous"); 1779 KASSERT(vcpuid == -1 || (vcpuid >= 0 && vcpuid < VM_MAXCPU), 1780 ("vm_smp_rendezvous: invalid vcpuid %d", vcpuid)); 1781 1782 restart: 1783 mtx_lock(&vm->rendezvous_mtx); 1784 if (vm->rendezvous_func != NULL) { 1785 /* 1786 * If a rendezvous is already in progress then we need to 1787 * call the rendezvous handler in case this 'vcpuid' is one 1788 * of the targets of the rendezvous. 1789 */ 1790 RENDEZVOUS_CTR0(vm, vcpuid, "Rendezvous already in progress"); 1791 mtx_unlock(&vm->rendezvous_mtx); 1792 vm_handle_rendezvous(vm, vcpuid); 1793 goto restart; 1794 } 1795 KASSERT(vm->rendezvous_func == NULL, ("vm_smp_rendezvous: previous " 1796 "rendezvous is still in progress")); 1797 1798 RENDEZVOUS_CTR0(vm, vcpuid, "Initiating rendezvous"); 1799 vm->rendezvous_req_cpus = dest; 1800 CPU_ZERO(&vm->rendezvous_done_cpus); 1801 vm->rendezvous_arg = arg; 1802 vm_set_rendezvous_func(vm, func); 1803 mtx_unlock(&vm->rendezvous_mtx); 1804 1805 /* 1806 * Wake up any sleeping vcpus and trigger a VM-exit in any running 1807 * vcpus so they handle the rendezvous as soon as possible. 1808 */ 1809 for (i = 0; i < VM_MAXCPU; i++) { 1810 if (CPU_ISSET(i, &dest)) 1811 vcpu_notify_event(vm, i, false); 1812 } 1813 1814 vm_handle_rendezvous(vm, vcpuid); 1815 } 1816 1817 struct vatpic * 1818 vm_atpic(struct vm *vm) 1819 { 1820 return (vm->vatpic); 1821 } 1822 1823 struct vatpit * 1824 vm_atpit(struct vm *vm) 1825 { 1826 return (vm->vatpit); 1827 } 1828