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 volatile cpuset_t halted_cpus; 147 }; 148 149 static int vmm_initialized; 150 151 static struct vmm_ops *ops; 152 #define VMM_INIT(num) (ops != NULL ? (*ops->init)(num) : 0) 153 #define VMM_CLEANUP() (ops != NULL ? (*ops->cleanup)() : 0) 154 #define VMM_RESUME() (ops != NULL ? (*ops->resume)() : 0) 155 156 #define VMINIT(vm, pmap) (ops != NULL ? (*ops->vminit)(vm, pmap): NULL) 157 #define VMRUN(vmi, vcpu, rip, pmap, rptr, sptr) \ 158 (ops != NULL ? (*ops->vmrun)(vmi, vcpu, rip, pmap, rptr, sptr) : ENXIO) 159 #define VMCLEANUP(vmi) (ops != NULL ? (*ops->vmcleanup)(vmi) : NULL) 160 #define VMSPACE_ALLOC(min, max) \ 161 (ops != NULL ? (*ops->vmspace_alloc)(min, max) : NULL) 162 #define VMSPACE_FREE(vmspace) \ 163 (ops != NULL ? (*ops->vmspace_free)(vmspace) : ENXIO) 164 #define VMGETREG(vmi, vcpu, num, retval) \ 165 (ops != NULL ? (*ops->vmgetreg)(vmi, vcpu, num, retval) : ENXIO) 166 #define VMSETREG(vmi, vcpu, num, val) \ 167 (ops != NULL ? (*ops->vmsetreg)(vmi, vcpu, num, val) : ENXIO) 168 #define VMGETDESC(vmi, vcpu, num, desc) \ 169 (ops != NULL ? (*ops->vmgetdesc)(vmi, vcpu, num, desc) : ENXIO) 170 #define VMSETDESC(vmi, vcpu, num, desc) \ 171 (ops != NULL ? (*ops->vmsetdesc)(vmi, vcpu, num, desc) : ENXIO) 172 #define VMGETCAP(vmi, vcpu, num, retval) \ 173 (ops != NULL ? (*ops->vmgetcap)(vmi, vcpu, num, retval) : ENXIO) 174 #define VMSETCAP(vmi, vcpu, num, val) \ 175 (ops != NULL ? (*ops->vmsetcap)(vmi, vcpu, num, val) : ENXIO) 176 #define VLAPIC_INIT(vmi, vcpu) \ 177 (ops != NULL ? (*ops->vlapic_init)(vmi, vcpu) : NULL) 178 #define VLAPIC_CLEANUP(vmi, vlapic) \ 179 (ops != NULL ? (*ops->vlapic_cleanup)(vmi, vlapic) : NULL) 180 181 #define fpu_start_emulating() load_cr0(rcr0() | CR0_TS) 182 #define fpu_stop_emulating() clts() 183 184 static MALLOC_DEFINE(M_VM, "vm", "vm"); 185 CTASSERT(VMM_MSR_NUM <= 64); /* msr_mask can keep track of up to 64 msrs */ 186 187 /* statistics */ 188 static VMM_STAT(VCPU_TOTAL_RUNTIME, "vcpu total runtime"); 189 190 SYSCTL_NODE(_hw, OID_AUTO, vmm, CTLFLAG_RW, NULL, NULL); 191 192 static int vmm_ipinum; 193 SYSCTL_INT(_hw_vmm, OID_AUTO, ipinum, CTLFLAG_RD, &vmm_ipinum, 0, 194 "IPI vector used for vcpu notifications"); 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 vcpu *vcpu; 1010 const char *wmesg; 1011 int t, vcpu_halted, vm_halted; 1012 1013 KASSERT(!CPU_ISSET(vcpuid, &vm->halted_cpus), ("vcpu already halted")); 1014 1015 vcpu = &vm->vcpu[vcpuid]; 1016 vcpu_halted = 0; 1017 vm_halted = 0; 1018 1019 vcpu_lock(vcpu); 1020 while (1) { 1021 /* 1022 * Do a final check for pending NMI or interrupts before 1023 * really putting this thread to sleep. Also check for 1024 * software events that would cause this vcpu to wakeup. 1025 * 1026 * These interrupts/events could have happened after the 1027 * vcpu returned from VMRUN() and before it acquired the 1028 * vcpu lock above. 1029 */ 1030 if (vm->rendezvous_func != NULL || vm->suspend) 1031 break; 1032 if (vm_nmi_pending(vm, vcpuid)) 1033 break; 1034 if (!intr_disabled) { 1035 if (vm_extint_pending(vm, vcpuid) || 1036 vlapic_pending_intr(vcpu->vlapic, NULL)) { 1037 break; 1038 } 1039 } 1040 1041 /* 1042 * Some Linux guests implement "halt" by having all vcpus 1043 * execute HLT with interrupts disabled. 'halted_cpus' keeps 1044 * track of the vcpus that have entered this state. When all 1045 * vcpus enter the halted state the virtual machine is halted. 1046 */ 1047 if (intr_disabled) { 1048 wmesg = "vmhalt"; 1049 VCPU_CTR0(vm, vcpuid, "Halted"); 1050 if (!vcpu_halted) { 1051 vcpu_halted = 1; 1052 CPU_SET_ATOMIC(vcpuid, &vm->halted_cpus); 1053 } 1054 if (CPU_CMP(&vm->halted_cpus, &vm->active_cpus) == 0) { 1055 vm_halted = 1; 1056 break; 1057 } 1058 } else { 1059 wmesg = "vmidle"; 1060 } 1061 1062 t = ticks; 1063 vcpu_require_state_locked(vcpu, VCPU_SLEEPING); 1064 msleep_spin(vcpu, &vcpu->mtx, wmesg, 0); 1065 vcpu_require_state_locked(vcpu, VCPU_FROZEN); 1066 vmm_stat_incr(vm, vcpuid, VCPU_IDLE_TICKS, ticks - t); 1067 } 1068 1069 if (vcpu_halted) 1070 CPU_CLR_ATOMIC(vcpuid, &vm->halted_cpus); 1071 1072 vcpu_unlock(vcpu); 1073 1074 if (vm_halted) 1075 vm_suspend(vm, VM_SUSPEND_HALT); 1076 1077 return (0); 1078 } 1079 1080 static int 1081 vm_handle_paging(struct vm *vm, int vcpuid, bool *retu) 1082 { 1083 int rv, ftype; 1084 struct vm_map *map; 1085 struct vcpu *vcpu; 1086 struct vm_exit *vme; 1087 1088 vcpu = &vm->vcpu[vcpuid]; 1089 vme = &vcpu->exitinfo; 1090 1091 ftype = vme->u.paging.fault_type; 1092 KASSERT(ftype == VM_PROT_READ || 1093 ftype == VM_PROT_WRITE || ftype == VM_PROT_EXECUTE, 1094 ("vm_handle_paging: invalid fault_type %d", ftype)); 1095 1096 if (ftype == VM_PROT_READ || ftype == VM_PROT_WRITE) { 1097 rv = pmap_emulate_accessed_dirty(vmspace_pmap(vm->vmspace), 1098 vme->u.paging.gpa, ftype); 1099 if (rv == 0) 1100 goto done; 1101 } 1102 1103 map = &vm->vmspace->vm_map; 1104 rv = vm_fault(map, vme->u.paging.gpa, ftype, VM_FAULT_NORMAL); 1105 1106 VCPU_CTR3(vm, vcpuid, "vm_handle_paging rv = %d, gpa = %#lx, " 1107 "ftype = %d", rv, vme->u.paging.gpa, ftype); 1108 1109 if (rv != KERN_SUCCESS) 1110 return (EFAULT); 1111 done: 1112 /* restart execution at the faulting instruction */ 1113 vme->inst_length = 0; 1114 1115 return (0); 1116 } 1117 1118 static int 1119 vm_handle_inst_emul(struct vm *vm, int vcpuid, bool *retu) 1120 { 1121 struct vie *vie; 1122 struct vcpu *vcpu; 1123 struct vm_exit *vme; 1124 int error, inst_length; 1125 uint64_t rip, gla, gpa, cr3; 1126 enum vie_cpu_mode cpu_mode; 1127 enum vie_paging_mode paging_mode; 1128 mem_region_read_t mread; 1129 mem_region_write_t mwrite; 1130 1131 vcpu = &vm->vcpu[vcpuid]; 1132 vme = &vcpu->exitinfo; 1133 1134 rip = vme->rip; 1135 inst_length = vme->inst_length; 1136 1137 gla = vme->u.inst_emul.gla; 1138 gpa = vme->u.inst_emul.gpa; 1139 cr3 = vme->u.inst_emul.cr3; 1140 cpu_mode = vme->u.inst_emul.cpu_mode; 1141 paging_mode = vme->u.inst_emul.paging_mode; 1142 vie = &vme->u.inst_emul.vie; 1143 1144 vie_init(vie); 1145 1146 /* Fetch, decode and emulate the faulting instruction */ 1147 if (vmm_fetch_instruction(vm, vcpuid, rip, inst_length, cr3, 1148 paging_mode, vie) != 0) 1149 return (EFAULT); 1150 1151 if (vmm_decode_instruction(vm, vcpuid, gla, cpu_mode, vie) != 0) 1152 return (EFAULT); 1153 1154 /* return to userland unless this is an in-kernel emulated device */ 1155 if (gpa >= DEFAULT_APIC_BASE && gpa < DEFAULT_APIC_BASE + PAGE_SIZE) { 1156 mread = lapic_mmio_read; 1157 mwrite = lapic_mmio_write; 1158 } else if (gpa >= VIOAPIC_BASE && gpa < VIOAPIC_BASE + VIOAPIC_SIZE) { 1159 mread = vioapic_mmio_read; 1160 mwrite = vioapic_mmio_write; 1161 } else if (gpa >= VHPET_BASE && gpa < VHPET_BASE + VHPET_SIZE) { 1162 mread = vhpet_mmio_read; 1163 mwrite = vhpet_mmio_write; 1164 } else { 1165 *retu = true; 1166 return (0); 1167 } 1168 1169 error = vmm_emulate_instruction(vm, vcpuid, gpa, vie, mread, mwrite, 1170 retu); 1171 1172 return (error); 1173 } 1174 1175 static int 1176 vm_handle_suspend(struct vm *vm, int vcpuid, bool *retu) 1177 { 1178 int i, done; 1179 struct vcpu *vcpu; 1180 1181 done = 0; 1182 vcpu = &vm->vcpu[vcpuid]; 1183 1184 CPU_SET_ATOMIC(vcpuid, &vm->suspended_cpus); 1185 1186 /* 1187 * Wait until all 'active_cpus' have suspended themselves. 1188 * 1189 * Since a VM may be suspended at any time including when one or 1190 * more vcpus are doing a rendezvous we need to call the rendezvous 1191 * handler while we are waiting to prevent a deadlock. 1192 */ 1193 vcpu_lock(vcpu); 1194 while (1) { 1195 if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) { 1196 VCPU_CTR0(vm, vcpuid, "All vcpus suspended"); 1197 break; 1198 } 1199 1200 if (vm->rendezvous_func == NULL) { 1201 VCPU_CTR0(vm, vcpuid, "Sleeping during suspend"); 1202 vcpu_require_state_locked(vcpu, VCPU_SLEEPING); 1203 msleep_spin(vcpu, &vcpu->mtx, "vmsusp", hz); 1204 vcpu_require_state_locked(vcpu, VCPU_FROZEN); 1205 } else { 1206 VCPU_CTR0(vm, vcpuid, "Rendezvous during suspend"); 1207 vcpu_unlock(vcpu); 1208 vm_handle_rendezvous(vm, vcpuid); 1209 vcpu_lock(vcpu); 1210 } 1211 } 1212 vcpu_unlock(vcpu); 1213 1214 /* 1215 * Wakeup the other sleeping vcpus and return to userspace. 1216 */ 1217 for (i = 0; i < VM_MAXCPU; i++) { 1218 if (CPU_ISSET(i, &vm->suspended_cpus)) { 1219 vcpu_notify_event(vm, i, false); 1220 } 1221 } 1222 1223 *retu = true; 1224 return (0); 1225 } 1226 1227 int 1228 vm_suspend(struct vm *vm, enum vm_suspend_how how) 1229 { 1230 int i; 1231 1232 if (how <= VM_SUSPEND_NONE || how >= VM_SUSPEND_LAST) 1233 return (EINVAL); 1234 1235 if (atomic_cmpset_int(&vm->suspend, 0, how) == 0) { 1236 VM_CTR2(vm, "virtual machine already suspended %d/%d", 1237 vm->suspend, how); 1238 return (EALREADY); 1239 } 1240 1241 VM_CTR1(vm, "virtual machine successfully suspended %d", how); 1242 1243 /* 1244 * Notify all active vcpus that they are now suspended. 1245 */ 1246 for (i = 0; i < VM_MAXCPU; i++) { 1247 if (CPU_ISSET(i, &vm->active_cpus)) 1248 vcpu_notify_event(vm, i, false); 1249 } 1250 1251 return (0); 1252 } 1253 1254 void 1255 vm_exit_suspended(struct vm *vm, int vcpuid, uint64_t rip) 1256 { 1257 struct vm_exit *vmexit; 1258 1259 KASSERT(vm->suspend > VM_SUSPEND_NONE && vm->suspend < VM_SUSPEND_LAST, 1260 ("vm_exit_suspended: invalid suspend type %d", vm->suspend)); 1261 1262 vmexit = vm_exitinfo(vm, vcpuid); 1263 vmexit->rip = rip; 1264 vmexit->inst_length = 0; 1265 vmexit->exitcode = VM_EXITCODE_SUSPENDED; 1266 vmexit->u.suspended.how = vm->suspend; 1267 } 1268 1269 int 1270 vm_run(struct vm *vm, struct vm_run *vmrun) 1271 { 1272 int error, vcpuid; 1273 struct vcpu *vcpu; 1274 struct pcb *pcb; 1275 uint64_t tscval, rip; 1276 struct vm_exit *vme; 1277 bool retu, intr_disabled; 1278 pmap_t pmap; 1279 void *rptr, *sptr; 1280 1281 vcpuid = vmrun->cpuid; 1282 1283 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1284 return (EINVAL); 1285 1286 rptr = &vm->rendezvous_func; 1287 sptr = &vm->suspend; 1288 pmap = vmspace_pmap(vm->vmspace); 1289 vcpu = &vm->vcpu[vcpuid]; 1290 vme = &vcpu->exitinfo; 1291 rip = vmrun->rip; 1292 restart: 1293 critical_enter(); 1294 1295 KASSERT(!CPU_ISSET(curcpu, &pmap->pm_active), 1296 ("vm_run: absurd pm_active")); 1297 1298 tscval = rdtsc(); 1299 1300 pcb = PCPU_GET(curpcb); 1301 set_pcb_flags(pcb, PCB_FULL_IRET); 1302 1303 restore_guest_msrs(vm, vcpuid); 1304 restore_guest_fpustate(vcpu); 1305 1306 vcpu_require_state(vm, vcpuid, VCPU_RUNNING); 1307 error = VMRUN(vm->cookie, vcpuid, rip, pmap, rptr, sptr); 1308 vcpu_require_state(vm, vcpuid, VCPU_FROZEN); 1309 1310 save_guest_fpustate(vcpu); 1311 restore_host_msrs(vm, vcpuid); 1312 1313 vmm_stat_incr(vm, vcpuid, VCPU_TOTAL_RUNTIME, rdtsc() - tscval); 1314 1315 critical_exit(); 1316 1317 if (error == 0) { 1318 retu = false; 1319 switch (vme->exitcode) { 1320 case VM_EXITCODE_SUSPENDED: 1321 error = vm_handle_suspend(vm, vcpuid, &retu); 1322 break; 1323 case VM_EXITCODE_IOAPIC_EOI: 1324 vioapic_process_eoi(vm, vcpuid, 1325 vme->u.ioapic_eoi.vector); 1326 break; 1327 case VM_EXITCODE_RENDEZVOUS: 1328 vm_handle_rendezvous(vm, vcpuid); 1329 error = 0; 1330 break; 1331 case VM_EXITCODE_HLT: 1332 intr_disabled = ((vme->u.hlt.rflags & PSL_I) == 0); 1333 error = vm_handle_hlt(vm, vcpuid, intr_disabled, &retu); 1334 break; 1335 case VM_EXITCODE_PAGING: 1336 error = vm_handle_paging(vm, vcpuid, &retu); 1337 break; 1338 case VM_EXITCODE_INST_EMUL: 1339 error = vm_handle_inst_emul(vm, vcpuid, &retu); 1340 break; 1341 default: 1342 retu = true; /* handled in userland */ 1343 break; 1344 } 1345 } 1346 1347 if (error == 0 && retu == false) { 1348 rip = vme->rip + vme->inst_length; 1349 goto restart; 1350 } 1351 1352 /* copy the exit information */ 1353 bcopy(vme, &vmrun->vm_exit, sizeof(struct vm_exit)); 1354 return (error); 1355 } 1356 1357 int 1358 vm_inject_exception(struct vm *vm, int vcpuid, struct vm_exception *exception) 1359 { 1360 struct vcpu *vcpu; 1361 1362 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1363 return (EINVAL); 1364 1365 if (exception->vector < 0 || exception->vector >= 32) 1366 return (EINVAL); 1367 1368 vcpu = &vm->vcpu[vcpuid]; 1369 1370 if (vcpu->exception_pending) { 1371 VCPU_CTR2(vm, vcpuid, "Unable to inject exception %d due to " 1372 "pending exception %d", exception->vector, 1373 vcpu->exception.vector); 1374 return (EBUSY); 1375 } 1376 1377 vcpu->exception_pending = 1; 1378 vcpu->exception = *exception; 1379 VCPU_CTR1(vm, vcpuid, "Exception %d pending", exception->vector); 1380 return (0); 1381 } 1382 1383 int 1384 vm_exception_pending(struct vm *vm, int vcpuid, struct vm_exception *exception) 1385 { 1386 struct vcpu *vcpu; 1387 int pending; 1388 1389 KASSERT(vcpuid >= 0 && vcpuid < VM_MAXCPU, ("invalid vcpu %d", vcpuid)); 1390 1391 vcpu = &vm->vcpu[vcpuid]; 1392 pending = vcpu->exception_pending; 1393 if (pending) { 1394 vcpu->exception_pending = 0; 1395 *exception = vcpu->exception; 1396 VCPU_CTR1(vm, vcpuid, "Exception %d delivered", 1397 exception->vector); 1398 } 1399 return (pending); 1400 } 1401 1402 static void 1403 vm_inject_fault(struct vm *vm, int vcpuid, struct vm_exception *exception) 1404 { 1405 struct vm_exit *vmexit; 1406 int error; 1407 1408 error = vm_inject_exception(vm, vcpuid, exception); 1409 KASSERT(error == 0, ("vm_inject_exception error %d", error)); 1410 1411 /* 1412 * A fault-like exception allows the instruction to be restarted 1413 * after the exception handler returns. 1414 * 1415 * By setting the inst_length to 0 we ensure that the instruction 1416 * pointer remains at the faulting instruction. 1417 */ 1418 vmexit = vm_exitinfo(vm, vcpuid); 1419 vmexit->inst_length = 0; 1420 } 1421 1422 void 1423 vm_inject_gp(struct vm *vm, int vcpuid) 1424 { 1425 struct vm_exception gpf = { 1426 .vector = IDT_GP, 1427 .error_code_valid = 1, 1428 .error_code = 0 1429 }; 1430 1431 vm_inject_fault(vm, vcpuid, &gpf); 1432 } 1433 1434 void 1435 vm_inject_ud(struct vm *vm, int vcpuid) 1436 { 1437 struct vm_exception udf = { 1438 .vector = IDT_UD, 1439 .error_code_valid = 0 1440 }; 1441 1442 vm_inject_fault(vm, vcpuid, &udf); 1443 } 1444 1445 static VMM_STAT(VCPU_NMI_COUNT, "number of NMIs delivered to vcpu"); 1446 1447 int 1448 vm_inject_nmi(struct vm *vm, int vcpuid) 1449 { 1450 struct vcpu *vcpu; 1451 1452 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1453 return (EINVAL); 1454 1455 vcpu = &vm->vcpu[vcpuid]; 1456 1457 vcpu->nmi_pending = 1; 1458 vcpu_notify_event(vm, vcpuid, false); 1459 return (0); 1460 } 1461 1462 int 1463 vm_nmi_pending(struct vm *vm, int vcpuid) 1464 { 1465 struct vcpu *vcpu; 1466 1467 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1468 panic("vm_nmi_pending: invalid vcpuid %d", vcpuid); 1469 1470 vcpu = &vm->vcpu[vcpuid]; 1471 1472 return (vcpu->nmi_pending); 1473 } 1474 1475 void 1476 vm_nmi_clear(struct vm *vm, int vcpuid) 1477 { 1478 struct vcpu *vcpu; 1479 1480 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1481 panic("vm_nmi_pending: invalid vcpuid %d", vcpuid); 1482 1483 vcpu = &vm->vcpu[vcpuid]; 1484 1485 if (vcpu->nmi_pending == 0) 1486 panic("vm_nmi_clear: inconsistent nmi_pending state"); 1487 1488 vcpu->nmi_pending = 0; 1489 vmm_stat_incr(vm, vcpuid, VCPU_NMI_COUNT, 1); 1490 } 1491 1492 static VMM_STAT(VCPU_EXTINT_COUNT, "number of ExtINTs delivered to vcpu"); 1493 1494 int 1495 vm_inject_extint(struct vm *vm, int vcpuid) 1496 { 1497 struct vcpu *vcpu; 1498 1499 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1500 return (EINVAL); 1501 1502 vcpu = &vm->vcpu[vcpuid]; 1503 1504 vcpu->extint_pending = 1; 1505 vcpu_notify_event(vm, vcpuid, false); 1506 return (0); 1507 } 1508 1509 int 1510 vm_extint_pending(struct vm *vm, int vcpuid) 1511 { 1512 struct vcpu *vcpu; 1513 1514 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1515 panic("vm_extint_pending: invalid vcpuid %d", vcpuid); 1516 1517 vcpu = &vm->vcpu[vcpuid]; 1518 1519 return (vcpu->extint_pending); 1520 } 1521 1522 void 1523 vm_extint_clear(struct vm *vm, int vcpuid) 1524 { 1525 struct vcpu *vcpu; 1526 1527 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1528 panic("vm_extint_pending: invalid vcpuid %d", vcpuid); 1529 1530 vcpu = &vm->vcpu[vcpuid]; 1531 1532 if (vcpu->extint_pending == 0) 1533 panic("vm_extint_clear: inconsistent extint_pending state"); 1534 1535 vcpu->extint_pending = 0; 1536 vmm_stat_incr(vm, vcpuid, VCPU_EXTINT_COUNT, 1); 1537 } 1538 1539 int 1540 vm_get_capability(struct vm *vm, int vcpu, int type, int *retval) 1541 { 1542 if (vcpu < 0 || vcpu >= VM_MAXCPU) 1543 return (EINVAL); 1544 1545 if (type < 0 || type >= VM_CAP_MAX) 1546 return (EINVAL); 1547 1548 return (VMGETCAP(vm->cookie, vcpu, type, retval)); 1549 } 1550 1551 int 1552 vm_set_capability(struct vm *vm, int vcpu, int type, int val) 1553 { 1554 if (vcpu < 0 || vcpu >= VM_MAXCPU) 1555 return (EINVAL); 1556 1557 if (type < 0 || type >= VM_CAP_MAX) 1558 return (EINVAL); 1559 1560 return (VMSETCAP(vm->cookie, vcpu, type, val)); 1561 } 1562 1563 uint64_t * 1564 vm_guest_msrs(struct vm *vm, int cpu) 1565 { 1566 return (vm->vcpu[cpu].guest_msrs); 1567 } 1568 1569 struct vlapic * 1570 vm_lapic(struct vm *vm, int cpu) 1571 { 1572 return (vm->vcpu[cpu].vlapic); 1573 } 1574 1575 struct vioapic * 1576 vm_ioapic(struct vm *vm) 1577 { 1578 1579 return (vm->vioapic); 1580 } 1581 1582 struct vhpet * 1583 vm_hpet(struct vm *vm) 1584 { 1585 1586 return (vm->vhpet); 1587 } 1588 1589 boolean_t 1590 vmm_is_pptdev(int bus, int slot, int func) 1591 { 1592 int found, i, n; 1593 int b, s, f; 1594 char *val, *cp, *cp2; 1595 1596 /* 1597 * XXX 1598 * The length of an environment variable is limited to 128 bytes which 1599 * puts an upper limit on the number of passthru devices that may be 1600 * specified using a single environment variable. 1601 * 1602 * Work around this by scanning multiple environment variable 1603 * names instead of a single one - yuck! 1604 */ 1605 const char *names[] = { "pptdevs", "pptdevs2", "pptdevs3", NULL }; 1606 1607 /* set pptdevs="1/2/3 4/5/6 7/8/9 10/11/12" */ 1608 found = 0; 1609 for (i = 0; names[i] != NULL && !found; i++) { 1610 cp = val = getenv(names[i]); 1611 while (cp != NULL && *cp != '\0') { 1612 if ((cp2 = strchr(cp, ' ')) != NULL) 1613 *cp2 = '\0'; 1614 1615 n = sscanf(cp, "%d/%d/%d", &b, &s, &f); 1616 if (n == 3 && bus == b && slot == s && func == f) { 1617 found = 1; 1618 break; 1619 } 1620 1621 if (cp2 != NULL) 1622 *cp2++ = ' '; 1623 1624 cp = cp2; 1625 } 1626 freeenv(val); 1627 } 1628 return (found); 1629 } 1630 1631 void * 1632 vm_iommu_domain(struct vm *vm) 1633 { 1634 1635 return (vm->iommu); 1636 } 1637 1638 int 1639 vcpu_set_state(struct vm *vm, int vcpuid, enum vcpu_state newstate, 1640 bool from_idle) 1641 { 1642 int error; 1643 struct vcpu *vcpu; 1644 1645 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1646 panic("vm_set_run_state: invalid vcpuid %d", vcpuid); 1647 1648 vcpu = &vm->vcpu[vcpuid]; 1649 1650 vcpu_lock(vcpu); 1651 error = vcpu_set_state_locked(vcpu, newstate, from_idle); 1652 vcpu_unlock(vcpu); 1653 1654 return (error); 1655 } 1656 1657 enum vcpu_state 1658 vcpu_get_state(struct vm *vm, int vcpuid, int *hostcpu) 1659 { 1660 struct vcpu *vcpu; 1661 enum vcpu_state state; 1662 1663 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1664 panic("vm_get_run_state: invalid vcpuid %d", vcpuid); 1665 1666 vcpu = &vm->vcpu[vcpuid]; 1667 1668 vcpu_lock(vcpu); 1669 state = vcpu->state; 1670 if (hostcpu != NULL) 1671 *hostcpu = vcpu->hostcpu; 1672 vcpu_unlock(vcpu); 1673 1674 return (state); 1675 } 1676 1677 void 1678 vm_activate_cpu(struct vm *vm, int vcpuid) 1679 { 1680 1681 KASSERT(vcpuid >= 0 && vcpuid < VM_MAXCPU, 1682 ("vm_activate_cpu: invalid vcpuid %d", vcpuid)); 1683 KASSERT(!CPU_ISSET(vcpuid, &vm->active_cpus), 1684 ("vm_activate_cpu: vcpuid %d is already active", vcpuid)); 1685 1686 VCPU_CTR0(vm, vcpuid, "activated"); 1687 CPU_SET_ATOMIC(vcpuid, &vm->active_cpus); 1688 } 1689 1690 cpuset_t 1691 vm_active_cpus(struct vm *vm) 1692 { 1693 1694 return (vm->active_cpus); 1695 } 1696 1697 void * 1698 vcpu_stats(struct vm *vm, int vcpuid) 1699 { 1700 1701 return (vm->vcpu[vcpuid].stats); 1702 } 1703 1704 int 1705 vm_get_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state *state) 1706 { 1707 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1708 return (EINVAL); 1709 1710 *state = vm->vcpu[vcpuid].x2apic_state; 1711 1712 return (0); 1713 } 1714 1715 int 1716 vm_set_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state state) 1717 { 1718 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1719 return (EINVAL); 1720 1721 if (state >= X2APIC_STATE_LAST) 1722 return (EINVAL); 1723 1724 vm->vcpu[vcpuid].x2apic_state = state; 1725 1726 vlapic_set_x2apic_state(vm, vcpuid, state); 1727 1728 return (0); 1729 } 1730 1731 /* 1732 * This function is called to ensure that a vcpu "sees" a pending event 1733 * as soon as possible: 1734 * - If the vcpu thread is sleeping then it is woken up. 1735 * - If the vcpu is running on a different host_cpu then an IPI will be directed 1736 * to the host_cpu to cause the vcpu to trap into the hypervisor. 1737 */ 1738 void 1739 vcpu_notify_event(struct vm *vm, int vcpuid, bool lapic_intr) 1740 { 1741 int hostcpu; 1742 struct vcpu *vcpu; 1743 1744 vcpu = &vm->vcpu[vcpuid]; 1745 1746 vcpu_lock(vcpu); 1747 hostcpu = vcpu->hostcpu; 1748 if (vcpu->state == VCPU_RUNNING) { 1749 KASSERT(hostcpu != NOCPU, ("vcpu running on invalid hostcpu")); 1750 if (hostcpu != curcpu) { 1751 if (lapic_intr) { 1752 vlapic_post_intr(vcpu->vlapic, hostcpu, 1753 vmm_ipinum); 1754 } else { 1755 ipi_cpu(hostcpu, vmm_ipinum); 1756 } 1757 } else { 1758 /* 1759 * If the 'vcpu' is running on 'curcpu' then it must 1760 * be sending a notification to itself (e.g. SELF_IPI). 1761 * The pending event will be picked up when the vcpu 1762 * transitions back to guest context. 1763 */ 1764 } 1765 } else { 1766 KASSERT(hostcpu == NOCPU, ("vcpu state %d not consistent " 1767 "with hostcpu %d", vcpu->state, hostcpu)); 1768 if (vcpu->state == VCPU_SLEEPING) 1769 wakeup_one(vcpu); 1770 } 1771 vcpu_unlock(vcpu); 1772 } 1773 1774 struct vmspace * 1775 vm_get_vmspace(struct vm *vm) 1776 { 1777 1778 return (vm->vmspace); 1779 } 1780 1781 int 1782 vm_apicid2vcpuid(struct vm *vm, int apicid) 1783 { 1784 /* 1785 * XXX apic id is assumed to be numerically identical to vcpu id 1786 */ 1787 return (apicid); 1788 } 1789 1790 void 1791 vm_smp_rendezvous(struct vm *vm, int vcpuid, cpuset_t dest, 1792 vm_rendezvous_func_t func, void *arg) 1793 { 1794 int i; 1795 1796 /* 1797 * Enforce that this function is called without any locks 1798 */ 1799 WITNESS_WARN(WARN_PANIC, NULL, "vm_smp_rendezvous"); 1800 KASSERT(vcpuid == -1 || (vcpuid >= 0 && vcpuid < VM_MAXCPU), 1801 ("vm_smp_rendezvous: invalid vcpuid %d", vcpuid)); 1802 1803 restart: 1804 mtx_lock(&vm->rendezvous_mtx); 1805 if (vm->rendezvous_func != NULL) { 1806 /* 1807 * If a rendezvous is already in progress then we need to 1808 * call the rendezvous handler in case this 'vcpuid' is one 1809 * of the targets of the rendezvous. 1810 */ 1811 RENDEZVOUS_CTR0(vm, vcpuid, "Rendezvous already in progress"); 1812 mtx_unlock(&vm->rendezvous_mtx); 1813 vm_handle_rendezvous(vm, vcpuid); 1814 goto restart; 1815 } 1816 KASSERT(vm->rendezvous_func == NULL, ("vm_smp_rendezvous: previous " 1817 "rendezvous is still in progress")); 1818 1819 RENDEZVOUS_CTR0(vm, vcpuid, "Initiating rendezvous"); 1820 vm->rendezvous_req_cpus = dest; 1821 CPU_ZERO(&vm->rendezvous_done_cpus); 1822 vm->rendezvous_arg = arg; 1823 vm_set_rendezvous_func(vm, func); 1824 mtx_unlock(&vm->rendezvous_mtx); 1825 1826 /* 1827 * Wake up any sleeping vcpus and trigger a VM-exit in any running 1828 * vcpus so they handle the rendezvous as soon as possible. 1829 */ 1830 for (i = 0; i < VM_MAXCPU; i++) { 1831 if (CPU_ISSET(i, &dest)) 1832 vcpu_notify_event(vm, i, false); 1833 } 1834 1835 vm_handle_rendezvous(vm, vcpuid); 1836 } 1837 1838 struct vatpic * 1839 vm_atpic(struct vm *vm) 1840 { 1841 return (vm->vatpic); 1842 } 1843 1844 struct vatpit * 1845 vm_atpit(struct vm *vm) 1846 { 1847 return (vm->vatpit); 1848 } 1849