vmm.c revision 266339
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: stable/10/sys/amd64/vmm/vmm.c 266339 2014-05-17 19:11:08Z jhb $ 27 */ 28 29#include <sys/cdefs.h> 30__FBSDID("$FreeBSD: stable/10/sys/amd64/vmm/vmm.c 266339 2014-05-17 19:11:08Z jhb $"); 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 "vhpet.h" 71#include "vioapic.h" 72#include "vlapic.h" 73#include "vmm_msr.h" 74#include "vmm_ipi.h" 75#include "vmm_stat.h" 76#include "vmm_lapic.h" 77 78#include "io/ppt.h" 79#include "io/iommu.h" 80 81struct vlapic; 82 83struct vcpu { 84 int flags; 85 enum vcpu_state state; 86 struct mtx mtx; 87 int hostcpu; /* host cpuid this vcpu last ran on */ 88 uint64_t guest_msrs[VMM_MSR_NUM]; 89 struct vlapic *vlapic; 90 int vcpuid; 91 struct savefpu *guestfpu; /* guest fpu state */ 92 void *stats; 93 struct vm_exit exitinfo; 94 enum x2apic_state x2apic_state; 95 int nmi_pending; 96}; 97 98#define vcpu_lock_init(v) mtx_init(&((v)->mtx), "vcpu lock", 0, MTX_SPIN) 99#define vcpu_lock(v) mtx_lock_spin(&((v)->mtx)) 100#define vcpu_unlock(v) mtx_unlock_spin(&((v)->mtx)) 101#define vcpu_assert_locked(v) mtx_assert(&((v)->mtx), MA_OWNED) 102 103struct mem_seg { 104 vm_paddr_t gpa; 105 size_t len; 106 boolean_t wired; 107 vm_object_t object; 108}; 109#define VM_MAX_MEMORY_SEGMENTS 2 110 111struct vm { 112 void *cookie; /* processor-specific data */ 113 void *iommu; /* iommu-specific data */ 114 struct vhpet *vhpet; /* virtual HPET */ 115 struct vioapic *vioapic; /* virtual ioapic */ 116 struct vmspace *vmspace; /* guest's address space */ 117 struct vcpu vcpu[VM_MAXCPU]; 118 int num_mem_segs; 119 struct mem_seg mem_segs[VM_MAX_MEMORY_SEGMENTS]; 120 char name[VM_MAX_NAMELEN]; 121 122 /* 123 * Set of active vcpus. 124 * An active vcpu is one that has been started implicitly (BSP) or 125 * explicitly (AP) by sending it a startup ipi. 126 */ 127 volatile cpuset_t active_cpus; 128 129 struct mtx rendezvous_mtx; 130 cpuset_t rendezvous_req_cpus; 131 cpuset_t rendezvous_done_cpus; 132 void *rendezvous_arg; 133 vm_rendezvous_func_t rendezvous_func; 134}; 135 136static int vmm_initialized; 137 138static struct vmm_ops *ops; 139#define VMM_INIT(num) (ops != NULL ? (*ops->init)(num) : 0) 140#define VMM_CLEANUP() (ops != NULL ? (*ops->cleanup)() : 0) 141#define VMM_RESUME() (ops != NULL ? (*ops->resume)() : 0) 142 143#define VMINIT(vm, pmap) (ops != NULL ? (*ops->vminit)(vm, pmap): NULL) 144#define VMRUN(vmi, vcpu, rip, pmap, rptr) \ 145 (ops != NULL ? (*ops->vmrun)(vmi, vcpu, rip, pmap, rptr) : ENXIO) 146#define VMCLEANUP(vmi) (ops != NULL ? (*ops->vmcleanup)(vmi) : NULL) 147#define VMSPACE_ALLOC(min, max) \ 148 (ops != NULL ? (*ops->vmspace_alloc)(min, max) : NULL) 149#define VMSPACE_FREE(vmspace) \ 150 (ops != NULL ? (*ops->vmspace_free)(vmspace) : ENXIO) 151#define VMGETREG(vmi, vcpu, num, retval) \ 152 (ops != NULL ? (*ops->vmgetreg)(vmi, vcpu, num, retval) : ENXIO) 153#define VMSETREG(vmi, vcpu, num, val) \ 154 (ops != NULL ? (*ops->vmsetreg)(vmi, vcpu, num, val) : ENXIO) 155#define VMGETDESC(vmi, vcpu, num, desc) \ 156 (ops != NULL ? (*ops->vmgetdesc)(vmi, vcpu, num, desc) : ENXIO) 157#define VMSETDESC(vmi, vcpu, num, desc) \ 158 (ops != NULL ? (*ops->vmsetdesc)(vmi, vcpu, num, desc) : ENXIO) 159#define VMINJECT(vmi, vcpu, type, vec, ec, ecv) \ 160 (ops != NULL ? (*ops->vminject)(vmi, vcpu, type, vec, ec, ecv) : ENXIO) 161#define VMGETCAP(vmi, vcpu, num, retval) \ 162 (ops != NULL ? (*ops->vmgetcap)(vmi, vcpu, num, retval) : ENXIO) 163#define VMSETCAP(vmi, vcpu, num, val) \ 164 (ops != NULL ? (*ops->vmsetcap)(vmi, vcpu, num, val) : ENXIO) 165#define VLAPIC_INIT(vmi, vcpu) \ 166 (ops != NULL ? (*ops->vlapic_init)(vmi, vcpu) : NULL) 167#define VLAPIC_CLEANUP(vmi, vlapic) \ 168 (ops != NULL ? (*ops->vlapic_cleanup)(vmi, vlapic) : NULL) 169 170#define fpu_start_emulating() load_cr0(rcr0() | CR0_TS) 171#define fpu_stop_emulating() clts() 172 173static MALLOC_DEFINE(M_VM, "vm", "vm"); 174CTASSERT(VMM_MSR_NUM <= 64); /* msr_mask can keep track of up to 64 msrs */ 175 176/* statistics */ 177static VMM_STAT(VCPU_TOTAL_RUNTIME, "vcpu total runtime"); 178 179SYSCTL_NODE(_hw, OID_AUTO, vmm, CTLFLAG_RW, NULL, NULL); 180 181static int vmm_ipinum; 182SYSCTL_INT(_hw_vmm, OID_AUTO, ipinum, CTLFLAG_RD, &vmm_ipinum, 0, 183 "IPI vector used for vcpu notifications"); 184 185static void vm_deactivate_cpu(struct vm *vm, int vcpuid); 186 187static void 188vcpu_cleanup(struct vm *vm, int i) 189{ 190 struct vcpu *vcpu = &vm->vcpu[i]; 191 192 VLAPIC_CLEANUP(vm->cookie, vcpu->vlapic); 193 vmm_stat_free(vcpu->stats); 194 fpu_save_area_free(vcpu->guestfpu); 195} 196 197static void 198vcpu_init(struct vm *vm, uint32_t vcpu_id) 199{ 200 struct vcpu *vcpu; 201 202 vcpu = &vm->vcpu[vcpu_id]; 203 204 vcpu_lock_init(vcpu); 205 vcpu->hostcpu = NOCPU; 206 vcpu->vcpuid = vcpu_id; 207 vcpu->vlapic = VLAPIC_INIT(vm->cookie, vcpu_id); 208 vm_set_x2apic_state(vm, vcpu_id, X2APIC_ENABLED); 209 vcpu->guestfpu = fpu_save_area_alloc(); 210 fpu_save_area_reset(vcpu->guestfpu); 211 vcpu->stats = vmm_stat_alloc(); 212} 213 214struct vm_exit * 215vm_exitinfo(struct vm *vm, int cpuid) 216{ 217 struct vcpu *vcpu; 218 219 if (cpuid < 0 || cpuid >= VM_MAXCPU) 220 panic("vm_exitinfo: invalid cpuid %d", cpuid); 221 222 vcpu = &vm->vcpu[cpuid]; 223 224 return (&vcpu->exitinfo); 225} 226 227static void 228vmm_resume(void) 229{ 230 VMM_RESUME(); 231} 232 233static int 234vmm_init(void) 235{ 236 int error; 237 238 vmm_host_state_init(); 239 240 vmm_ipinum = vmm_ipi_alloc(); 241 if (vmm_ipinum == 0) 242 vmm_ipinum = IPI_AST; 243 244 error = vmm_mem_init(); 245 if (error) 246 return (error); 247 248 if (vmm_is_intel()) 249 ops = &vmm_ops_intel; 250 else if (vmm_is_amd()) 251 ops = &vmm_ops_amd; 252 else 253 return (ENXIO); 254 255 vmm_msr_init(); 256 vmm_resume_p = vmm_resume; 257 258 return (VMM_INIT(vmm_ipinum)); 259} 260 261static int 262vmm_handler(module_t mod, int what, void *arg) 263{ 264 int error; 265 266 switch (what) { 267 case MOD_LOAD: 268 vmmdev_init(); 269 iommu_init(); 270 error = vmm_init(); 271 if (error == 0) 272 vmm_initialized = 1; 273 break; 274 case MOD_UNLOAD: 275 error = vmmdev_cleanup(); 276 if (error == 0) { 277 vmm_resume_p = NULL; 278 iommu_cleanup(); 279 if (vmm_ipinum != IPI_AST) 280 vmm_ipi_free(vmm_ipinum); 281 error = VMM_CLEANUP(); 282 /* 283 * Something bad happened - prevent new 284 * VMs from being created 285 */ 286 if (error) 287 vmm_initialized = 0; 288 } 289 break; 290 default: 291 error = 0; 292 break; 293 } 294 return (error); 295} 296 297static moduledata_t vmm_kmod = { 298 "vmm", 299 vmm_handler, 300 NULL 301}; 302 303/* 304 * vmm initialization has the following dependencies: 305 * 306 * - iommu initialization must happen after the pci passthru driver has had 307 * a chance to attach to any passthru devices (after SI_SUB_CONFIGURE). 308 * 309 * - VT-x initialization requires smp_rendezvous() and therefore must happen 310 * after SMP is fully functional (after SI_SUB_SMP). 311 */ 312DECLARE_MODULE(vmm, vmm_kmod, SI_SUB_SMP + 1, SI_ORDER_ANY); 313MODULE_VERSION(vmm, 1); 314 315int 316vm_create(const char *name, struct vm **retvm) 317{ 318 int i; 319 struct vm *vm; 320 struct vmspace *vmspace; 321 322 const int BSP = 0; 323 324 /* 325 * If vmm.ko could not be successfully initialized then don't attempt 326 * to create the virtual machine. 327 */ 328 if (!vmm_initialized) 329 return (ENXIO); 330 331 if (name == NULL || strlen(name) >= VM_MAX_NAMELEN) 332 return (EINVAL); 333 334 vmspace = VMSPACE_ALLOC(VM_MIN_ADDRESS, VM_MAXUSER_ADDRESS); 335 if (vmspace == NULL) 336 return (ENOMEM); 337 338 vm = malloc(sizeof(struct vm), M_VM, M_WAITOK | M_ZERO); 339 strcpy(vm->name, name); 340 vm->vmspace = vmspace; 341 mtx_init(&vm->rendezvous_mtx, "vm rendezvous lock", 0, MTX_DEF); 342 vm->cookie = VMINIT(vm, vmspace_pmap(vmspace)); 343 vm->vioapic = vioapic_init(vm); 344 vm->vhpet = vhpet_init(vm); 345 346 for (i = 0; i < VM_MAXCPU; i++) { 347 vcpu_init(vm, i); 348 guest_msrs_init(vm, i); 349 } 350 351 vm_activate_cpu(vm, BSP); 352 353 *retvm = vm; 354 return (0); 355} 356 357static void 358vm_free_mem_seg(struct vm *vm, struct mem_seg *seg) 359{ 360 361 if (seg->object != NULL) 362 vmm_mem_free(vm->vmspace, seg->gpa, seg->len); 363 364 bzero(seg, sizeof(*seg)); 365} 366 367void 368vm_destroy(struct vm *vm) 369{ 370 int i; 371 372 ppt_unassign_all(vm); 373 374 if (vm->iommu != NULL) 375 iommu_destroy_domain(vm->iommu); 376 377 vhpet_cleanup(vm->vhpet); 378 vioapic_cleanup(vm->vioapic); 379 380 for (i = 0; i < vm->num_mem_segs; i++) 381 vm_free_mem_seg(vm, &vm->mem_segs[i]); 382 383 vm->num_mem_segs = 0; 384 385 for (i = 0; i < VM_MAXCPU; i++) 386 vcpu_cleanup(vm, i); 387 388 VMSPACE_FREE(vm->vmspace); 389 390 VMCLEANUP(vm->cookie); 391 392 free(vm, M_VM); 393} 394 395const char * 396vm_name(struct vm *vm) 397{ 398 return (vm->name); 399} 400 401int 402vm_map_mmio(struct vm *vm, vm_paddr_t gpa, size_t len, vm_paddr_t hpa) 403{ 404 vm_object_t obj; 405 406 if ((obj = vmm_mmio_alloc(vm->vmspace, gpa, len, hpa)) == NULL) 407 return (ENOMEM); 408 else 409 return (0); 410} 411 412int 413vm_unmap_mmio(struct vm *vm, vm_paddr_t gpa, size_t len) 414{ 415 416 vmm_mmio_free(vm->vmspace, gpa, len); 417 return (0); 418} 419 420boolean_t 421vm_mem_allocated(struct vm *vm, vm_paddr_t gpa) 422{ 423 int i; 424 vm_paddr_t gpabase, gpalimit; 425 426 for (i = 0; i < vm->num_mem_segs; i++) { 427 gpabase = vm->mem_segs[i].gpa; 428 gpalimit = gpabase + vm->mem_segs[i].len; 429 if (gpa >= gpabase && gpa < gpalimit) 430 return (TRUE); /* 'gpa' is regular memory */ 431 } 432 433 if (ppt_is_mmio(vm, gpa)) 434 return (TRUE); /* 'gpa' is pci passthru mmio */ 435 436 return (FALSE); 437} 438 439int 440vm_malloc(struct vm *vm, vm_paddr_t gpa, size_t len) 441{ 442 int available, allocated; 443 struct mem_seg *seg; 444 vm_object_t object; 445 vm_paddr_t g; 446 447 if ((gpa & PAGE_MASK) || (len & PAGE_MASK) || len == 0) 448 return (EINVAL); 449 450 available = allocated = 0; 451 g = gpa; 452 while (g < gpa + len) { 453 if (vm_mem_allocated(vm, g)) 454 allocated++; 455 else 456 available++; 457 458 g += PAGE_SIZE; 459 } 460 461 /* 462 * If there are some allocated and some available pages in the address 463 * range then it is an error. 464 */ 465 if (allocated && available) 466 return (EINVAL); 467 468 /* 469 * If the entire address range being requested has already been 470 * allocated then there isn't anything more to do. 471 */ 472 if (allocated && available == 0) 473 return (0); 474 475 if (vm->num_mem_segs >= VM_MAX_MEMORY_SEGMENTS) 476 return (E2BIG); 477 478 seg = &vm->mem_segs[vm->num_mem_segs]; 479 480 if ((object = vmm_mem_alloc(vm->vmspace, gpa, len)) == NULL) 481 return (ENOMEM); 482 483 seg->gpa = gpa; 484 seg->len = len; 485 seg->object = object; 486 seg->wired = FALSE; 487 488 vm->num_mem_segs++; 489 490 return (0); 491} 492 493static void 494vm_gpa_unwire(struct vm *vm) 495{ 496 int i, rv; 497 struct mem_seg *seg; 498 499 for (i = 0; i < vm->num_mem_segs; i++) { 500 seg = &vm->mem_segs[i]; 501 if (!seg->wired) 502 continue; 503 504 rv = vm_map_unwire(&vm->vmspace->vm_map, 505 seg->gpa, seg->gpa + seg->len, 506 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES); 507 KASSERT(rv == KERN_SUCCESS, ("vm(%s) memory segment " 508 "%#lx/%ld could not be unwired: %d", 509 vm_name(vm), seg->gpa, seg->len, rv)); 510 511 seg->wired = FALSE; 512 } 513} 514 515static int 516vm_gpa_wire(struct vm *vm) 517{ 518 int i, rv; 519 struct mem_seg *seg; 520 521 for (i = 0; i < vm->num_mem_segs; i++) { 522 seg = &vm->mem_segs[i]; 523 if (seg->wired) 524 continue; 525 526 /* XXX rlimits? */ 527 rv = vm_map_wire(&vm->vmspace->vm_map, 528 seg->gpa, seg->gpa + seg->len, 529 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES); 530 if (rv != KERN_SUCCESS) 531 break; 532 533 seg->wired = TRUE; 534 } 535 536 if (i < vm->num_mem_segs) { 537 /* 538 * Undo the wiring before returning an error. 539 */ 540 vm_gpa_unwire(vm); 541 return (EAGAIN); 542 } 543 544 return (0); 545} 546 547static void 548vm_iommu_modify(struct vm *vm, boolean_t map) 549{ 550 int i, sz; 551 vm_paddr_t gpa, hpa; 552 struct mem_seg *seg; 553 void *vp, *cookie, *host_domain; 554 555 sz = PAGE_SIZE; 556 host_domain = iommu_host_domain(); 557 558 for (i = 0; i < vm->num_mem_segs; i++) { 559 seg = &vm->mem_segs[i]; 560 KASSERT(seg->wired, ("vm(%s) memory segment %#lx/%ld not wired", 561 vm_name(vm), seg->gpa, seg->len)); 562 563 gpa = seg->gpa; 564 while (gpa < seg->gpa + seg->len) { 565 vp = vm_gpa_hold(vm, gpa, PAGE_SIZE, VM_PROT_WRITE, 566 &cookie); 567 KASSERT(vp != NULL, ("vm(%s) could not map gpa %#lx", 568 vm_name(vm), gpa)); 569 570 vm_gpa_release(cookie); 571 572 hpa = DMAP_TO_PHYS((uintptr_t)vp); 573 if (map) { 574 iommu_create_mapping(vm->iommu, gpa, hpa, sz); 575 iommu_remove_mapping(host_domain, hpa, sz); 576 } else { 577 iommu_remove_mapping(vm->iommu, gpa, sz); 578 iommu_create_mapping(host_domain, hpa, hpa, sz); 579 } 580 581 gpa += PAGE_SIZE; 582 } 583 } 584 585 /* 586 * Invalidate the cached translations associated with the domain 587 * from which pages were removed. 588 */ 589 if (map) 590 iommu_invalidate_tlb(host_domain); 591 else 592 iommu_invalidate_tlb(vm->iommu); 593} 594 595#define vm_iommu_unmap(vm) vm_iommu_modify((vm), FALSE) 596#define vm_iommu_map(vm) vm_iommu_modify((vm), TRUE) 597 598int 599vm_unassign_pptdev(struct vm *vm, int bus, int slot, int func) 600{ 601 int error; 602 603 error = ppt_unassign_device(vm, bus, slot, func); 604 if (error) 605 return (error); 606 607 if (ppt_num_devices(vm) == 0) { 608 vm_iommu_unmap(vm); 609 vm_gpa_unwire(vm); 610 } 611 return (0); 612} 613 614int 615vm_assign_pptdev(struct vm *vm, int bus, int slot, int func) 616{ 617 int error; 618 vm_paddr_t maxaddr; 619 620 /* 621 * Virtual machines with pci passthru devices get special treatment: 622 * - the guest physical memory is wired 623 * - the iommu is programmed to do the 'gpa' to 'hpa' translation 624 * 625 * We need to do this before the first pci passthru device is attached. 626 */ 627 if (ppt_num_devices(vm) == 0) { 628 KASSERT(vm->iommu == NULL, 629 ("vm_assign_pptdev: iommu must be NULL")); 630 maxaddr = vmm_mem_maxaddr(); 631 vm->iommu = iommu_create_domain(maxaddr); 632 633 error = vm_gpa_wire(vm); 634 if (error) 635 return (error); 636 637 vm_iommu_map(vm); 638 } 639 640 error = ppt_assign_device(vm, bus, slot, func); 641 return (error); 642} 643 644void * 645vm_gpa_hold(struct vm *vm, vm_paddr_t gpa, size_t len, int reqprot, 646 void **cookie) 647{ 648 int count, pageoff; 649 vm_page_t m; 650 651 pageoff = gpa & PAGE_MASK; 652 if (len > PAGE_SIZE - pageoff) 653 panic("vm_gpa_hold: invalid gpa/len: 0x%016lx/%lu", gpa, len); 654 655 count = vm_fault_quick_hold_pages(&vm->vmspace->vm_map, 656 trunc_page(gpa), PAGE_SIZE, reqprot, &m, 1); 657 658 if (count == 1) { 659 *cookie = m; 660 return ((void *)(PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m)) + pageoff)); 661 } else { 662 *cookie = NULL; 663 return (NULL); 664 } 665} 666 667void 668vm_gpa_release(void *cookie) 669{ 670 vm_page_t m = cookie; 671 672 vm_page_lock(m); 673 vm_page_unhold(m); 674 vm_page_unlock(m); 675} 676 677int 678vm_gpabase2memseg(struct vm *vm, vm_paddr_t gpabase, 679 struct vm_memory_segment *seg) 680{ 681 int i; 682 683 for (i = 0; i < vm->num_mem_segs; i++) { 684 if (gpabase == vm->mem_segs[i].gpa) { 685 seg->gpa = vm->mem_segs[i].gpa; 686 seg->len = vm->mem_segs[i].len; 687 seg->wired = vm->mem_segs[i].wired; 688 return (0); 689 } 690 } 691 return (-1); 692} 693 694int 695vm_get_memobj(struct vm *vm, vm_paddr_t gpa, size_t len, 696 vm_offset_t *offset, struct vm_object **object) 697{ 698 int i; 699 size_t seg_len; 700 vm_paddr_t seg_gpa; 701 vm_object_t seg_obj; 702 703 for (i = 0; i < vm->num_mem_segs; i++) { 704 if ((seg_obj = vm->mem_segs[i].object) == NULL) 705 continue; 706 707 seg_gpa = vm->mem_segs[i].gpa; 708 seg_len = vm->mem_segs[i].len; 709 710 if (gpa >= seg_gpa && gpa < seg_gpa + seg_len) { 711 *offset = gpa - seg_gpa; 712 *object = seg_obj; 713 vm_object_reference(seg_obj); 714 return (0); 715 } 716 } 717 718 return (EINVAL); 719} 720 721int 722vm_get_register(struct vm *vm, int vcpu, int reg, uint64_t *retval) 723{ 724 725 if (vcpu < 0 || vcpu >= VM_MAXCPU) 726 return (EINVAL); 727 728 if (reg >= VM_REG_LAST) 729 return (EINVAL); 730 731 return (VMGETREG(vm->cookie, vcpu, reg, retval)); 732} 733 734int 735vm_set_register(struct vm *vm, int vcpu, int reg, uint64_t val) 736{ 737 738 if (vcpu < 0 || vcpu >= VM_MAXCPU) 739 return (EINVAL); 740 741 if (reg >= VM_REG_LAST) 742 return (EINVAL); 743 744 return (VMSETREG(vm->cookie, vcpu, reg, val)); 745} 746 747static boolean_t 748is_descriptor_table(int reg) 749{ 750 751 switch (reg) { 752 case VM_REG_GUEST_IDTR: 753 case VM_REG_GUEST_GDTR: 754 return (TRUE); 755 default: 756 return (FALSE); 757 } 758} 759 760static boolean_t 761is_segment_register(int reg) 762{ 763 764 switch (reg) { 765 case VM_REG_GUEST_ES: 766 case VM_REG_GUEST_CS: 767 case VM_REG_GUEST_SS: 768 case VM_REG_GUEST_DS: 769 case VM_REG_GUEST_FS: 770 case VM_REG_GUEST_GS: 771 case VM_REG_GUEST_TR: 772 case VM_REG_GUEST_LDTR: 773 return (TRUE); 774 default: 775 return (FALSE); 776 } 777} 778 779int 780vm_get_seg_desc(struct vm *vm, int vcpu, int reg, 781 struct seg_desc *desc) 782{ 783 784 if (vcpu < 0 || vcpu >= VM_MAXCPU) 785 return (EINVAL); 786 787 if (!is_segment_register(reg) && !is_descriptor_table(reg)) 788 return (EINVAL); 789 790 return (VMGETDESC(vm->cookie, vcpu, reg, desc)); 791} 792 793int 794vm_set_seg_desc(struct vm *vm, int vcpu, int reg, 795 struct seg_desc *desc) 796{ 797 if (vcpu < 0 || vcpu >= VM_MAXCPU) 798 return (EINVAL); 799 800 if (!is_segment_register(reg) && !is_descriptor_table(reg)) 801 return (EINVAL); 802 803 return (VMSETDESC(vm->cookie, vcpu, reg, desc)); 804} 805 806static void 807restore_guest_fpustate(struct vcpu *vcpu) 808{ 809 810 /* flush host state to the pcb */ 811 fpuexit(curthread); 812 813 /* restore guest FPU state */ 814 fpu_stop_emulating(); 815 fpurestore(vcpu->guestfpu); 816 817 /* 818 * The FPU is now "dirty" with the guest's state so turn on emulation 819 * to trap any access to the FPU by the host. 820 */ 821 fpu_start_emulating(); 822} 823 824static void 825save_guest_fpustate(struct vcpu *vcpu) 826{ 827 828 if ((rcr0() & CR0_TS) == 0) 829 panic("fpu emulation not enabled in host!"); 830 831 /* save guest FPU state */ 832 fpu_stop_emulating(); 833 fpusave(vcpu->guestfpu); 834 fpu_start_emulating(); 835} 836 837static VMM_STAT(VCPU_IDLE_TICKS, "number of ticks vcpu was idle"); 838 839static int 840vcpu_set_state_locked(struct vcpu *vcpu, enum vcpu_state newstate) 841{ 842 int error; 843 844 vcpu_assert_locked(vcpu); 845 846 /* 847 * The following state transitions are allowed: 848 * IDLE -> FROZEN -> IDLE 849 * FROZEN -> RUNNING -> FROZEN 850 * FROZEN -> SLEEPING -> FROZEN 851 */ 852 switch (vcpu->state) { 853 case VCPU_IDLE: 854 case VCPU_RUNNING: 855 case VCPU_SLEEPING: 856 error = (newstate != VCPU_FROZEN); 857 break; 858 case VCPU_FROZEN: 859 error = (newstate == VCPU_FROZEN); 860 break; 861 default: 862 error = 1; 863 break; 864 } 865 866 if (error == 0) 867 vcpu->state = newstate; 868 else 869 error = EBUSY; 870 871 return (error); 872} 873 874static void 875vcpu_require_state(struct vm *vm, int vcpuid, enum vcpu_state newstate) 876{ 877 int error; 878 879 if ((error = vcpu_set_state(vm, vcpuid, newstate)) != 0) 880 panic("Error %d setting state to %d\n", error, newstate); 881} 882 883static void 884vcpu_require_state_locked(struct vcpu *vcpu, enum vcpu_state newstate) 885{ 886 int error; 887 888 if ((error = vcpu_set_state_locked(vcpu, newstate)) != 0) 889 panic("Error %d setting state to %d", error, newstate); 890} 891 892static void 893vm_set_rendezvous_func(struct vm *vm, vm_rendezvous_func_t func) 894{ 895 896 KASSERT(mtx_owned(&vm->rendezvous_mtx), ("rendezvous_mtx not locked")); 897 898 /* 899 * Update 'rendezvous_func' and execute a write memory barrier to 900 * ensure that it is visible across all host cpus. This is not needed 901 * for correctness but it does ensure that all the vcpus will notice 902 * that the rendezvous is requested immediately. 903 */ 904 vm->rendezvous_func = func; 905 wmb(); 906} 907 908#define RENDEZVOUS_CTR0(vm, vcpuid, fmt) \ 909 do { \ 910 if (vcpuid >= 0) \ 911 VCPU_CTR0(vm, vcpuid, fmt); \ 912 else \ 913 VM_CTR0(vm, fmt); \ 914 } while (0) 915 916static void 917vm_handle_rendezvous(struct vm *vm, int vcpuid) 918{ 919 920 KASSERT(vcpuid == -1 || (vcpuid >= 0 && vcpuid < VM_MAXCPU), 921 ("vm_handle_rendezvous: invalid vcpuid %d", vcpuid)); 922 923 mtx_lock(&vm->rendezvous_mtx); 924 while (vm->rendezvous_func != NULL) { 925 /* 'rendezvous_req_cpus' must be a subset of 'active_cpus' */ 926 CPU_AND(&vm->rendezvous_req_cpus, &vm->active_cpus); 927 928 if (vcpuid != -1 && 929 CPU_ISSET(vcpuid, &vm->rendezvous_req_cpus) && 930 !CPU_ISSET(vcpuid, &vm->rendezvous_done_cpus)) { 931 VCPU_CTR0(vm, vcpuid, "Calling rendezvous func"); 932 (*vm->rendezvous_func)(vm, vcpuid, vm->rendezvous_arg); 933 CPU_SET(vcpuid, &vm->rendezvous_done_cpus); 934 } 935 if (CPU_CMP(&vm->rendezvous_req_cpus, 936 &vm->rendezvous_done_cpus) == 0) { 937 VCPU_CTR0(vm, vcpuid, "Rendezvous completed"); 938 vm_set_rendezvous_func(vm, NULL); 939 wakeup(&vm->rendezvous_func); 940 break; 941 } 942 RENDEZVOUS_CTR0(vm, vcpuid, "Wait for rendezvous completion"); 943 mtx_sleep(&vm->rendezvous_func, &vm->rendezvous_mtx, 0, 944 "vmrndv", 0); 945 } 946 mtx_unlock(&vm->rendezvous_mtx); 947} 948 949/* 950 * Emulate a guest 'hlt' by sleeping until the vcpu is ready to run. 951 */ 952static int 953vm_handle_hlt(struct vm *vm, int vcpuid, bool intr_disabled, bool *retu) 954{ 955 struct vm_exit *vmexit; 956 struct vcpu *vcpu; 957 int t, timo, spindown; 958 959 vcpu = &vm->vcpu[vcpuid]; 960 spindown = 0; 961 962 vcpu_lock(vcpu); 963 964 /* 965 * Do a final check for pending NMI or interrupts before 966 * really putting this thread to sleep. 967 * 968 * These interrupts could have happened any time after we 969 * returned from VMRUN() and before we grabbed the vcpu lock. 970 */ 971 if (!vm_nmi_pending(vm, vcpuid) && 972 (intr_disabled || !vlapic_pending_intr(vcpu->vlapic, NULL))) { 973 t = ticks; 974 vcpu_require_state_locked(vcpu, VCPU_SLEEPING); 975 if (vlapic_enabled(vcpu->vlapic)) { 976 /* 977 * XXX msleep_spin() is not interruptible so use the 978 * 'timo' to put an upper bound on the sleep time. 979 */ 980 timo = hz; 981 msleep_spin(vcpu, &vcpu->mtx, "vmidle", timo); 982 } else { 983 /* 984 * Spindown the vcpu if the apic is disabled and it 985 * had entered the halted state. 986 */ 987 spindown = 1; 988 } 989 vcpu_require_state_locked(vcpu, VCPU_FROZEN); 990 vmm_stat_incr(vm, vcpuid, VCPU_IDLE_TICKS, ticks - t); 991 } 992 vcpu_unlock(vcpu); 993 994 /* 995 * Since 'vm_deactivate_cpu()' grabs a sleep mutex we must call it 996 * outside the confines of the vcpu spinlock. 997 */ 998 if (spindown) { 999 *retu = true; 1000 vmexit = vm_exitinfo(vm, vcpuid); 1001 vmexit->exitcode = VM_EXITCODE_SPINDOWN_CPU; 1002 vm_deactivate_cpu(vm, vcpuid); 1003 VCPU_CTR0(vm, vcpuid, "spinning down cpu"); 1004 } 1005 1006 return (0); 1007} 1008 1009static int 1010vm_handle_paging(struct vm *vm, int vcpuid, bool *retu) 1011{ 1012 int rv, ftype; 1013 struct vm_map *map; 1014 struct vcpu *vcpu; 1015 struct vm_exit *vme; 1016 1017 vcpu = &vm->vcpu[vcpuid]; 1018 vme = &vcpu->exitinfo; 1019 1020 ftype = vme->u.paging.fault_type; 1021 KASSERT(ftype == VM_PROT_READ || 1022 ftype == VM_PROT_WRITE || ftype == VM_PROT_EXECUTE, 1023 ("vm_handle_paging: invalid fault_type %d", ftype)); 1024 1025 if (ftype == VM_PROT_READ || ftype == VM_PROT_WRITE) { 1026 rv = pmap_emulate_accessed_dirty(vmspace_pmap(vm->vmspace), 1027 vme->u.paging.gpa, ftype); 1028 if (rv == 0) 1029 goto done; 1030 } 1031 1032 map = &vm->vmspace->vm_map; 1033 rv = vm_fault(map, vme->u.paging.gpa, ftype, VM_FAULT_NORMAL); 1034 1035 VCPU_CTR3(vm, vcpuid, "vm_handle_paging rv = %d, gpa = %#lx, " 1036 "ftype = %d", rv, vme->u.paging.gpa, ftype); 1037 1038 if (rv != KERN_SUCCESS) 1039 return (EFAULT); 1040done: 1041 /* restart execution at the faulting instruction */ 1042 vme->inst_length = 0; 1043 1044 return (0); 1045} 1046 1047static int 1048vm_handle_inst_emul(struct vm *vm, int vcpuid, bool *retu) 1049{ 1050 struct vie *vie; 1051 struct vcpu *vcpu; 1052 struct vm_exit *vme; 1053 int error, inst_length; 1054 uint64_t rip, gla, gpa, cr3; 1055 mem_region_read_t mread; 1056 mem_region_write_t mwrite; 1057 1058 vcpu = &vm->vcpu[vcpuid]; 1059 vme = &vcpu->exitinfo; 1060 1061 rip = vme->rip; 1062 inst_length = vme->inst_length; 1063 1064 gla = vme->u.inst_emul.gla; 1065 gpa = vme->u.inst_emul.gpa; 1066 cr3 = vme->u.inst_emul.cr3; 1067 vie = &vme->u.inst_emul.vie; 1068 1069 vie_init(vie); 1070 1071 /* Fetch, decode and emulate the faulting instruction */ 1072 if (vmm_fetch_instruction(vm, vcpuid, rip, inst_length, cr3, vie) != 0) 1073 return (EFAULT); 1074 1075 if (vmm_decode_instruction(vm, vcpuid, gla, vie) != 0) 1076 return (EFAULT); 1077 1078 /* return to userland unless this is an in-kernel emulated device */ 1079 if (gpa >= DEFAULT_APIC_BASE && gpa < DEFAULT_APIC_BASE + PAGE_SIZE) { 1080 mread = lapic_mmio_read; 1081 mwrite = lapic_mmio_write; 1082 } else if (gpa >= VIOAPIC_BASE && gpa < VIOAPIC_BASE + VIOAPIC_SIZE) { 1083 mread = vioapic_mmio_read; 1084 mwrite = vioapic_mmio_write; 1085 } else if (gpa >= VHPET_BASE && gpa < VHPET_BASE + VHPET_SIZE) { 1086 mread = vhpet_mmio_read; 1087 mwrite = vhpet_mmio_write; 1088 } else { 1089 *retu = true; 1090 return (0); 1091 } 1092 1093 error = vmm_emulate_instruction(vm, vcpuid, gpa, vie, mread, mwrite, 1094 retu); 1095 1096 return (error); 1097} 1098 1099int 1100vm_run(struct vm *vm, struct vm_run *vmrun) 1101{ 1102 int error, vcpuid; 1103 struct vcpu *vcpu; 1104 struct pcb *pcb; 1105 uint64_t tscval, rip; 1106 struct vm_exit *vme; 1107 bool retu, intr_disabled; 1108 pmap_t pmap; 1109 1110 vcpuid = vmrun->cpuid; 1111 1112 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1113 return (EINVAL); 1114 1115 pmap = vmspace_pmap(vm->vmspace); 1116 vcpu = &vm->vcpu[vcpuid]; 1117 vme = &vcpu->exitinfo; 1118 rip = vmrun->rip; 1119restart: 1120 critical_enter(); 1121 1122 KASSERT(!CPU_ISSET(curcpu, &pmap->pm_active), 1123 ("vm_run: absurd pm_active")); 1124 1125 tscval = rdtsc(); 1126 1127 pcb = PCPU_GET(curpcb); 1128 set_pcb_flags(pcb, PCB_FULL_IRET); 1129 1130 restore_guest_msrs(vm, vcpuid); 1131 restore_guest_fpustate(vcpu); 1132 1133 vcpu_require_state(vm, vcpuid, VCPU_RUNNING); 1134 vcpu->hostcpu = curcpu; 1135 error = VMRUN(vm->cookie, vcpuid, rip, pmap, &vm->rendezvous_func); 1136 vcpu->hostcpu = NOCPU; 1137 vcpu_require_state(vm, vcpuid, VCPU_FROZEN); 1138 1139 save_guest_fpustate(vcpu); 1140 restore_host_msrs(vm, vcpuid); 1141 1142 vmm_stat_incr(vm, vcpuid, VCPU_TOTAL_RUNTIME, rdtsc() - tscval); 1143 1144 critical_exit(); 1145 1146 if (error == 0) { 1147 retu = false; 1148 switch (vme->exitcode) { 1149 case VM_EXITCODE_IOAPIC_EOI: 1150 vioapic_process_eoi(vm, vcpuid, 1151 vme->u.ioapic_eoi.vector); 1152 break; 1153 case VM_EXITCODE_RENDEZVOUS: 1154 vm_handle_rendezvous(vm, vcpuid); 1155 error = 0; 1156 break; 1157 case VM_EXITCODE_HLT: 1158 intr_disabled = ((vme->u.hlt.rflags & PSL_I) == 0); 1159 error = vm_handle_hlt(vm, vcpuid, intr_disabled, &retu); 1160 break; 1161 case VM_EXITCODE_PAGING: 1162 error = vm_handle_paging(vm, vcpuid, &retu); 1163 break; 1164 case VM_EXITCODE_INST_EMUL: 1165 error = vm_handle_inst_emul(vm, vcpuid, &retu); 1166 break; 1167 default: 1168 retu = true; /* handled in userland */ 1169 break; 1170 } 1171 } 1172 1173 if (error == 0 && retu == false) { 1174 rip = vme->rip + vme->inst_length; 1175 goto restart; 1176 } 1177 1178 /* copy the exit information */ 1179 bcopy(vme, &vmrun->vm_exit, sizeof(struct vm_exit)); 1180 return (error); 1181} 1182 1183int 1184vm_inject_event(struct vm *vm, int vcpuid, int type, 1185 int vector, uint32_t code, int code_valid) 1186{ 1187 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1188 return (EINVAL); 1189 1190 if ((type > VM_EVENT_NONE && type < VM_EVENT_MAX) == 0) 1191 return (EINVAL); 1192 1193 if (vector < 0 || vector > 255) 1194 return (EINVAL); 1195 1196 return (VMINJECT(vm->cookie, vcpuid, type, vector, code, code_valid)); 1197} 1198 1199static VMM_STAT(VCPU_NMI_COUNT, "number of NMIs delivered to vcpu"); 1200 1201int 1202vm_inject_nmi(struct vm *vm, int vcpuid) 1203{ 1204 struct vcpu *vcpu; 1205 1206 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1207 return (EINVAL); 1208 1209 vcpu = &vm->vcpu[vcpuid]; 1210 1211 vcpu->nmi_pending = 1; 1212 vcpu_notify_event(vm, vcpuid, false); 1213 return (0); 1214} 1215 1216int 1217vm_nmi_pending(struct vm *vm, int vcpuid) 1218{ 1219 struct vcpu *vcpu; 1220 1221 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1222 panic("vm_nmi_pending: invalid vcpuid %d", vcpuid); 1223 1224 vcpu = &vm->vcpu[vcpuid]; 1225 1226 return (vcpu->nmi_pending); 1227} 1228 1229void 1230vm_nmi_clear(struct vm *vm, int vcpuid) 1231{ 1232 struct vcpu *vcpu; 1233 1234 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1235 panic("vm_nmi_pending: invalid vcpuid %d", vcpuid); 1236 1237 vcpu = &vm->vcpu[vcpuid]; 1238 1239 if (vcpu->nmi_pending == 0) 1240 panic("vm_nmi_clear: inconsistent nmi_pending state"); 1241 1242 vcpu->nmi_pending = 0; 1243 vmm_stat_incr(vm, vcpuid, VCPU_NMI_COUNT, 1); 1244} 1245 1246int 1247vm_get_capability(struct vm *vm, int vcpu, int type, int *retval) 1248{ 1249 if (vcpu < 0 || vcpu >= VM_MAXCPU) 1250 return (EINVAL); 1251 1252 if (type < 0 || type >= VM_CAP_MAX) 1253 return (EINVAL); 1254 1255 return (VMGETCAP(vm->cookie, vcpu, type, retval)); 1256} 1257 1258int 1259vm_set_capability(struct vm *vm, int vcpu, int type, int val) 1260{ 1261 if (vcpu < 0 || vcpu >= VM_MAXCPU) 1262 return (EINVAL); 1263 1264 if (type < 0 || type >= VM_CAP_MAX) 1265 return (EINVAL); 1266 1267 return (VMSETCAP(vm->cookie, vcpu, type, val)); 1268} 1269 1270uint64_t * 1271vm_guest_msrs(struct vm *vm, int cpu) 1272{ 1273 return (vm->vcpu[cpu].guest_msrs); 1274} 1275 1276struct vlapic * 1277vm_lapic(struct vm *vm, int cpu) 1278{ 1279 return (vm->vcpu[cpu].vlapic); 1280} 1281 1282struct vioapic * 1283vm_ioapic(struct vm *vm) 1284{ 1285 1286 return (vm->vioapic); 1287} 1288 1289struct vhpet * 1290vm_hpet(struct vm *vm) 1291{ 1292 1293 return (vm->vhpet); 1294} 1295 1296boolean_t 1297vmm_is_pptdev(int bus, int slot, int func) 1298{ 1299 int found, i, n; 1300 int b, s, f; 1301 char *val, *cp, *cp2; 1302 1303 /* 1304 * XXX 1305 * The length of an environment variable is limited to 128 bytes which 1306 * puts an upper limit on the number of passthru devices that may be 1307 * specified using a single environment variable. 1308 * 1309 * Work around this by scanning multiple environment variable 1310 * names instead of a single one - yuck! 1311 */ 1312 const char *names[] = { "pptdevs", "pptdevs2", "pptdevs3", NULL }; 1313 1314 /* set pptdevs="1/2/3 4/5/6 7/8/9 10/11/12" */ 1315 found = 0; 1316 for (i = 0; names[i] != NULL && !found; i++) { 1317 cp = val = getenv(names[i]); 1318 while (cp != NULL && *cp != '\0') { 1319 if ((cp2 = strchr(cp, ' ')) != NULL) 1320 *cp2 = '\0'; 1321 1322 n = sscanf(cp, "%d/%d/%d", &b, &s, &f); 1323 if (n == 3 && bus == b && slot == s && func == f) { 1324 found = 1; 1325 break; 1326 } 1327 1328 if (cp2 != NULL) 1329 *cp2++ = ' '; 1330 1331 cp = cp2; 1332 } 1333 freeenv(val); 1334 } 1335 return (found); 1336} 1337 1338void * 1339vm_iommu_domain(struct vm *vm) 1340{ 1341 1342 return (vm->iommu); 1343} 1344 1345int 1346vcpu_set_state(struct vm *vm, int vcpuid, enum vcpu_state newstate) 1347{ 1348 int error; 1349 struct vcpu *vcpu; 1350 1351 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1352 panic("vm_set_run_state: invalid vcpuid %d", vcpuid); 1353 1354 vcpu = &vm->vcpu[vcpuid]; 1355 1356 vcpu_lock(vcpu); 1357 error = vcpu_set_state_locked(vcpu, newstate); 1358 vcpu_unlock(vcpu); 1359 1360 return (error); 1361} 1362 1363enum vcpu_state 1364vcpu_get_state(struct vm *vm, int vcpuid, int *hostcpu) 1365{ 1366 struct vcpu *vcpu; 1367 enum vcpu_state state; 1368 1369 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1370 panic("vm_get_run_state: invalid vcpuid %d", vcpuid); 1371 1372 vcpu = &vm->vcpu[vcpuid]; 1373 1374 vcpu_lock(vcpu); 1375 state = vcpu->state; 1376 if (hostcpu != NULL) 1377 *hostcpu = vcpu->hostcpu; 1378 vcpu_unlock(vcpu); 1379 1380 return (state); 1381} 1382 1383void 1384vm_activate_cpu(struct vm *vm, int vcpuid) 1385{ 1386 1387 KASSERT(vcpuid >= 0 && vcpuid < VM_MAXCPU, 1388 ("vm_activate_cpu: invalid vcpuid %d", vcpuid)); 1389 KASSERT(!CPU_ISSET(vcpuid, &vm->active_cpus), 1390 ("vm_activate_cpu: vcpuid %d is already active", vcpuid)); 1391 1392 VCPU_CTR0(vm, vcpuid, "activated"); 1393 CPU_SET_ATOMIC(vcpuid, &vm->active_cpus); 1394} 1395 1396static void 1397vm_deactivate_cpu(struct vm *vm, int vcpuid) 1398{ 1399 1400 KASSERT(vcpuid >= 0 && vcpuid < VM_MAXCPU, 1401 ("vm_deactivate_cpu: invalid vcpuid %d", vcpuid)); 1402 KASSERT(CPU_ISSET(vcpuid, &vm->active_cpus), 1403 ("vm_deactivate_cpu: vcpuid %d is not active", vcpuid)); 1404 1405 VCPU_CTR0(vm, vcpuid, "deactivated"); 1406 CPU_CLR_ATOMIC(vcpuid, &vm->active_cpus); 1407 1408 /* 1409 * If a vcpu rendezvous is in progress then it could be blocked 1410 * on 'vcpuid' - unblock it before disappearing forever. 1411 */ 1412 mtx_lock(&vm->rendezvous_mtx); 1413 if (vm->rendezvous_func != NULL) { 1414 VCPU_CTR0(vm, vcpuid, "unblock rendezvous after deactivation"); 1415 wakeup(&vm->rendezvous_func); 1416 } 1417 mtx_unlock(&vm->rendezvous_mtx); 1418} 1419 1420cpuset_t 1421vm_active_cpus(struct vm *vm) 1422{ 1423 1424 return (vm->active_cpus); 1425} 1426 1427void * 1428vcpu_stats(struct vm *vm, int vcpuid) 1429{ 1430 1431 return (vm->vcpu[vcpuid].stats); 1432} 1433 1434int 1435vm_get_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state *state) 1436{ 1437 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1438 return (EINVAL); 1439 1440 *state = vm->vcpu[vcpuid].x2apic_state; 1441 1442 return (0); 1443} 1444 1445int 1446vm_set_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state state) 1447{ 1448 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1449 return (EINVAL); 1450 1451 if (state >= X2APIC_STATE_LAST) 1452 return (EINVAL); 1453 1454 vm->vcpu[vcpuid].x2apic_state = state; 1455 1456 vlapic_set_x2apic_state(vm, vcpuid, state); 1457 1458 return (0); 1459} 1460 1461/* 1462 * This function is called to ensure that a vcpu "sees" a pending event 1463 * as soon as possible: 1464 * - If the vcpu thread is sleeping then it is woken up. 1465 * - If the vcpu is running on a different host_cpu then an IPI will be directed 1466 * to the host_cpu to cause the vcpu to trap into the hypervisor. 1467 */ 1468void 1469vcpu_notify_event(struct vm *vm, int vcpuid, bool lapic_intr) 1470{ 1471 int hostcpu; 1472 struct vcpu *vcpu; 1473 1474 vcpu = &vm->vcpu[vcpuid]; 1475 1476 vcpu_lock(vcpu); 1477 hostcpu = vcpu->hostcpu; 1478 if (hostcpu == NOCPU) { 1479 if (vcpu->state == VCPU_SLEEPING) 1480 wakeup_one(vcpu); 1481 } else { 1482 if (vcpu->state != VCPU_RUNNING) 1483 panic("invalid vcpu state %d", vcpu->state); 1484 if (hostcpu != curcpu) { 1485 if (lapic_intr) 1486 vlapic_post_intr(vcpu->vlapic, hostcpu, 1487 vmm_ipinum); 1488 else 1489 ipi_cpu(hostcpu, vmm_ipinum); 1490 } 1491 } 1492 vcpu_unlock(vcpu); 1493} 1494 1495struct vmspace * 1496vm_get_vmspace(struct vm *vm) 1497{ 1498 1499 return (vm->vmspace); 1500} 1501 1502int 1503vm_apicid2vcpuid(struct vm *vm, int apicid) 1504{ 1505 /* 1506 * XXX apic id is assumed to be numerically identical to vcpu id 1507 */ 1508 return (apicid); 1509} 1510 1511void 1512vm_smp_rendezvous(struct vm *vm, int vcpuid, cpuset_t dest, 1513 vm_rendezvous_func_t func, void *arg) 1514{ 1515 int i; 1516 1517 /* 1518 * Enforce that this function is called without any locks 1519 */ 1520 WITNESS_WARN(WARN_PANIC, NULL, "vm_smp_rendezvous"); 1521 KASSERT(vcpuid == -1 || (vcpuid >= 0 && vcpuid < VM_MAXCPU), 1522 ("vm_smp_rendezvous: invalid vcpuid %d", vcpuid)); 1523 1524restart: 1525 mtx_lock(&vm->rendezvous_mtx); 1526 if (vm->rendezvous_func != NULL) { 1527 /* 1528 * If a rendezvous is already in progress then we need to 1529 * call the rendezvous handler in case this 'vcpuid' is one 1530 * of the targets of the rendezvous. 1531 */ 1532 RENDEZVOUS_CTR0(vm, vcpuid, "Rendezvous already in progress"); 1533 mtx_unlock(&vm->rendezvous_mtx); 1534 vm_handle_rendezvous(vm, vcpuid); 1535 goto restart; 1536 } 1537 KASSERT(vm->rendezvous_func == NULL, ("vm_smp_rendezvous: previous " 1538 "rendezvous is still in progress")); 1539 1540 RENDEZVOUS_CTR0(vm, vcpuid, "Initiating rendezvous"); 1541 vm->rendezvous_req_cpus = dest; 1542 CPU_ZERO(&vm->rendezvous_done_cpus); 1543 vm->rendezvous_arg = arg; 1544 vm_set_rendezvous_func(vm, func); 1545 mtx_unlock(&vm->rendezvous_mtx); 1546 1547 /* 1548 * Wake up any sleeping vcpus and trigger a VM-exit in any running 1549 * vcpus so they handle the rendezvous as soon as possible. 1550 */ 1551 for (i = 0; i < VM_MAXCPU; i++) { 1552 if (CPU_ISSET(i, &dest)) 1553 vcpu_notify_event(vm, i, false); 1554 } 1555 1556 vm_handle_rendezvous(vm, vcpuid); 1557} 1558