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/pcb.h> 57 #include <machine/smp.h> 58 #include <x86/psl.h> 59 #include <x86/apicreg.h> 60 61 #include <machine/vmm.h> 62 #include <machine/vmm_dev.h> 63 #include <machine/vmm_instruction_emul.h> 64 65 #include "vmm_ioport.h" 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 "vpmtmr.h" 76 #include "vrtc.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 /* 86 * Initialization: 87 * (a) allocated when vcpu is created 88 * (i) initialized when vcpu is created and when it is reinitialized 89 * (o) initialized the first time the vcpu is created 90 * (x) initialized before use 91 */ 92 struct vcpu { 93 struct mtx mtx; /* (o) protects 'state' and 'hostcpu' */ 94 enum vcpu_state state; /* (o) vcpu state */ 95 int hostcpu; /* (o) vcpu's host cpu */ 96 int reqidle; /* (i) request vcpu to idle */ 97 struct vlapic *vlapic; /* (i) APIC device model */ 98 enum x2apic_state x2apic_state; /* (i) APIC mode */ 99 uint64_t exitintinfo; /* (i) events pending at VM exit */ 100 int nmi_pending; /* (i) NMI pending */ 101 int extint_pending; /* (i) INTR pending */ 102 int exception_pending; /* (i) exception pending */ 103 int exc_vector; /* (x) exception collateral */ 104 int exc_errcode_valid; 105 uint32_t exc_errcode; 106 struct savefpu *guestfpu; /* (a,i) guest fpu state */ 107 uint64_t guest_xcr0; /* (i) guest %xcr0 register */ 108 void *stats; /* (a,i) statistics */ 109 struct vm_exit exitinfo; /* (x) exit reason and collateral */ 110 uint64_t nextrip; /* (x) next instruction to execute */ 111 }; 112 113 #define vcpu_lock_initialized(v) mtx_initialized(&((v)->mtx)) 114 #define vcpu_lock_init(v) mtx_init(&((v)->mtx), "vcpu lock", 0, MTX_SPIN) 115 #define vcpu_lock(v) mtx_lock_spin(&((v)->mtx)) 116 #define vcpu_unlock(v) mtx_unlock_spin(&((v)->mtx)) 117 #define vcpu_assert_locked(v) mtx_assert(&((v)->mtx), MA_OWNED) 118 119 struct mem_seg { 120 size_t len; 121 bool sysmem; 122 struct vm_object *object; 123 }; 124 #define VM_MAX_MEMSEGS 3 125 126 struct mem_map { 127 vm_paddr_t gpa; 128 size_t len; 129 vm_ooffset_t segoff; 130 int segid; 131 int prot; 132 int flags; 133 }; 134 #define VM_MAX_MEMMAPS 4 135 136 /* 137 * Initialization: 138 * (o) initialized the first time the VM is created 139 * (i) initialized when VM is created and when it is reinitialized 140 * (x) initialized before use 141 */ 142 struct vm { 143 void *cookie; /* (i) cpu-specific data */ 144 void *iommu; /* (x) iommu-specific data */ 145 struct vhpet *vhpet; /* (i) virtual HPET */ 146 struct vioapic *vioapic; /* (i) virtual ioapic */ 147 struct vatpic *vatpic; /* (i) virtual atpic */ 148 struct vatpit *vatpit; /* (i) virtual atpit */ 149 struct vpmtmr *vpmtmr; /* (i) virtual ACPI PM timer */ 150 struct vrtc *vrtc; /* (o) virtual RTC */ 151 volatile cpuset_t active_cpus; /* (i) active vcpus */ 152 int suspend; /* (i) stop VM execution */ 153 volatile cpuset_t suspended_cpus; /* (i) suspended vcpus */ 154 volatile cpuset_t halted_cpus; /* (x) cpus in a hard halt */ 155 cpuset_t rendezvous_req_cpus; /* (x) rendezvous requested */ 156 cpuset_t rendezvous_done_cpus; /* (x) rendezvous finished */ 157 void *rendezvous_arg; /* (x) rendezvous func/arg */ 158 vm_rendezvous_func_t rendezvous_func; 159 struct mtx rendezvous_mtx; /* (o) rendezvous lock */ 160 struct mem_map mem_maps[VM_MAX_MEMMAPS]; /* (i) guest address space */ 161 struct mem_seg mem_segs[VM_MAX_MEMSEGS]; /* (o) guest memory regions */ 162 struct vmspace *vmspace; /* (o) guest's address space */ 163 char name[VM_MAX_NAMELEN]; /* (o) virtual machine name */ 164 struct vcpu vcpu[VM_MAXCPU]; /* (i) guest vcpus */ 165 }; 166 167 static int vmm_initialized; 168 169 static struct vmm_ops *ops; 170 #define VMM_INIT(num) (ops != NULL ? (*ops->init)(num) : 0) 171 #define VMM_CLEANUP() (ops != NULL ? (*ops->cleanup)() : 0) 172 #define VMM_RESUME() (ops != NULL ? (*ops->resume)() : 0) 173 174 #define VMINIT(vm, pmap) (ops != NULL ? (*ops->vminit)(vm, pmap): NULL) 175 #define VMRUN(vmi, vcpu, rip, pmap, evinfo) \ 176 (ops != NULL ? (*ops->vmrun)(vmi, vcpu, rip, pmap, evinfo) : ENXIO) 177 #define VMCLEANUP(vmi) (ops != NULL ? (*ops->vmcleanup)(vmi) : NULL) 178 #define VMSPACE_ALLOC(min, max) \ 179 (ops != NULL ? (*ops->vmspace_alloc)(min, max) : NULL) 180 #define VMSPACE_FREE(vmspace) \ 181 (ops != NULL ? (*ops->vmspace_free)(vmspace) : ENXIO) 182 #define VMGETREG(vmi, vcpu, num, retval) \ 183 (ops != NULL ? (*ops->vmgetreg)(vmi, vcpu, num, retval) : ENXIO) 184 #define VMSETREG(vmi, vcpu, num, val) \ 185 (ops != NULL ? (*ops->vmsetreg)(vmi, vcpu, num, val) : ENXIO) 186 #define VMGETDESC(vmi, vcpu, num, desc) \ 187 (ops != NULL ? (*ops->vmgetdesc)(vmi, vcpu, num, desc) : ENXIO) 188 #define VMSETDESC(vmi, vcpu, num, desc) \ 189 (ops != NULL ? (*ops->vmsetdesc)(vmi, vcpu, num, desc) : ENXIO) 190 #define VMGETCAP(vmi, vcpu, num, retval) \ 191 (ops != NULL ? (*ops->vmgetcap)(vmi, vcpu, num, retval) : ENXIO) 192 #define VMSETCAP(vmi, vcpu, num, val) \ 193 (ops != NULL ? (*ops->vmsetcap)(vmi, vcpu, num, val) : ENXIO) 194 #define VLAPIC_INIT(vmi, vcpu) \ 195 (ops != NULL ? (*ops->vlapic_init)(vmi, vcpu) : NULL) 196 #define VLAPIC_CLEANUP(vmi, vlapic) \ 197 (ops != NULL ? (*ops->vlapic_cleanup)(vmi, vlapic) : NULL) 198 199 #define fpu_start_emulating() load_cr0(rcr0() | CR0_TS) 200 #define fpu_stop_emulating() clts() 201 202 static MALLOC_DEFINE(M_VM, "vm", "vm"); 203 204 /* statistics */ 205 static VMM_STAT(VCPU_TOTAL_RUNTIME, "vcpu total runtime"); 206 207 SYSCTL_NODE(_hw, OID_AUTO, vmm, CTLFLAG_RW, NULL, NULL); 208 209 /* 210 * Halt the guest if all vcpus are executing a HLT instruction with 211 * interrupts disabled. 212 */ 213 static int halt_detection_enabled = 1; 214 SYSCTL_INT(_hw_vmm, OID_AUTO, halt_detection, CTLFLAG_RDTUN, 215 &halt_detection_enabled, 0, 216 "Halt VM if all vcpus execute HLT with interrupts disabled"); 217 218 SYSCTL_INT(_hw_vmm, OID_AUTO, ipinum, CTLFLAG_RD, &vmm_ipinum, 0, 219 "IPI vector used for vcpu notifications"); 220 221 static int trace_guest_exceptions; 222 SYSCTL_INT(_hw_vmm, OID_AUTO, trace_guest_exceptions, CTLFLAG_RDTUN, 223 &trace_guest_exceptions, 0, 224 "Trap into hypervisor on all guest exceptions and reflect them back"); 225 226 static void vm_free_memmap(struct vm *vm, int ident); 227 static bool sysmem_mapping(struct vm *vm, struct mem_map *mm); 228 static void vcpu_notify_event_locked(struct vcpu *vcpu, bool lapic_intr); 229 230 #ifdef KTR 231 static const char * 232 vcpu_state2str(enum vcpu_state state) 233 { 234 235 switch (state) { 236 case VCPU_IDLE: 237 return ("idle"); 238 case VCPU_FROZEN: 239 return ("frozen"); 240 case VCPU_RUNNING: 241 return ("running"); 242 case VCPU_SLEEPING: 243 return ("sleeping"); 244 default: 245 return ("unknown"); 246 } 247 } 248 #endif 249 250 static void 251 vcpu_cleanup(struct vm *vm, int i, bool destroy) 252 { 253 struct vcpu *vcpu = &vm->vcpu[i]; 254 255 VLAPIC_CLEANUP(vm->cookie, vcpu->vlapic); 256 if (destroy) { 257 vmm_stat_free(vcpu->stats); 258 fpu_save_area_free(vcpu->guestfpu); 259 } 260 } 261 262 static void 263 vcpu_init(struct vm *vm, int vcpu_id, bool create) 264 { 265 struct vcpu *vcpu; 266 267 KASSERT(vcpu_id >= 0 && vcpu_id < VM_MAXCPU, 268 ("vcpu_init: invalid vcpu %d", vcpu_id)); 269 270 vcpu = &vm->vcpu[vcpu_id]; 271 272 if (create) { 273 KASSERT(!vcpu_lock_initialized(vcpu), ("vcpu %d already " 274 "initialized", vcpu_id)); 275 vcpu_lock_init(vcpu); 276 vcpu->state = VCPU_IDLE; 277 vcpu->hostcpu = NOCPU; 278 vcpu->guestfpu = fpu_save_area_alloc(); 279 vcpu->stats = vmm_stat_alloc(); 280 } 281 282 vcpu->vlapic = VLAPIC_INIT(vm->cookie, vcpu_id); 283 vm_set_x2apic_state(vm, vcpu_id, X2APIC_DISABLED); 284 vcpu->reqidle = 0; 285 vcpu->exitintinfo = 0; 286 vcpu->nmi_pending = 0; 287 vcpu->extint_pending = 0; 288 vcpu->exception_pending = 0; 289 vcpu->guest_xcr0 = XFEATURE_ENABLED_X87; 290 fpu_save_area_reset(vcpu->guestfpu); 291 vmm_stat_init(vcpu->stats); 292 } 293 294 int 295 vcpu_trace_exceptions(struct vm *vm, int vcpuid) 296 { 297 298 return (trace_guest_exceptions); 299 } 300 301 struct vm_exit * 302 vm_exitinfo(struct vm *vm, int cpuid) 303 { 304 struct vcpu *vcpu; 305 306 if (cpuid < 0 || cpuid >= VM_MAXCPU) 307 panic("vm_exitinfo: invalid cpuid %d", cpuid); 308 309 vcpu = &vm->vcpu[cpuid]; 310 311 return (&vcpu->exitinfo); 312 } 313 314 static void 315 vmm_resume(void) 316 { 317 VMM_RESUME(); 318 } 319 320 static int 321 vmm_init(void) 322 { 323 int error; 324 325 vmm_host_state_init(); 326 327 vmm_ipinum = lapic_ipi_alloc(&IDTVEC(justreturn)); 328 if (vmm_ipinum < 0) 329 vmm_ipinum = IPI_AST; 330 331 error = vmm_mem_init(); 332 if (error) 333 return (error); 334 335 if (vmm_is_intel()) 336 ops = &vmm_ops_intel; 337 else if (vmm_is_amd()) 338 ops = &vmm_ops_amd; 339 else 340 return (ENXIO); 341 342 vmm_resume_p = vmm_resume; 343 344 return (VMM_INIT(vmm_ipinum)); 345 } 346 347 static int 348 vmm_handler(module_t mod, int what, void *arg) 349 { 350 int error; 351 352 switch (what) { 353 case MOD_LOAD: 354 vmmdev_init(); 355 error = vmm_init(); 356 if (error == 0) 357 vmm_initialized = 1; 358 break; 359 case MOD_UNLOAD: 360 error = vmmdev_cleanup(); 361 if (error == 0) { 362 vmm_resume_p = NULL; 363 iommu_cleanup(); 364 if (vmm_ipinum != IPI_AST) 365 lapic_ipi_free(vmm_ipinum); 366 error = VMM_CLEANUP(); 367 /* 368 * Something bad happened - prevent new 369 * VMs from being created 370 */ 371 if (error) 372 vmm_initialized = 0; 373 } 374 break; 375 default: 376 error = 0; 377 break; 378 } 379 return (error); 380 } 381 382 static moduledata_t vmm_kmod = { 383 "vmm", 384 vmm_handler, 385 NULL 386 }; 387 388 /* 389 * vmm initialization has the following dependencies: 390 * 391 * - VT-x initialization requires smp_rendezvous() and therefore must happen 392 * after SMP is fully functional (after SI_SUB_SMP). 393 */ 394 DECLARE_MODULE(vmm, vmm_kmod, SI_SUB_SMP + 1, SI_ORDER_ANY); 395 MODULE_VERSION(vmm, 1); 396 397 static void 398 vm_init(struct vm *vm, bool create) 399 { 400 int i; 401 402 vm->cookie = VMINIT(vm, vmspace_pmap(vm->vmspace)); 403 vm->iommu = NULL; 404 vm->vioapic = vioapic_init(vm); 405 vm->vhpet = vhpet_init(vm); 406 vm->vatpic = vatpic_init(vm); 407 vm->vatpit = vatpit_init(vm); 408 vm->vpmtmr = vpmtmr_init(vm); 409 if (create) 410 vm->vrtc = vrtc_init(vm); 411 412 CPU_ZERO(&vm->active_cpus); 413 414 vm->suspend = 0; 415 CPU_ZERO(&vm->suspended_cpus); 416 417 for (i = 0; i < VM_MAXCPU; i++) 418 vcpu_init(vm, i, create); 419 } 420 421 int 422 vm_create(const char *name, struct vm **retvm) 423 { 424 struct vm *vm; 425 struct vmspace *vmspace; 426 427 /* 428 * If vmm.ko could not be successfully initialized then don't attempt 429 * to create the virtual machine. 430 */ 431 if (!vmm_initialized) 432 return (ENXIO); 433 434 if (name == NULL || strlen(name) >= VM_MAX_NAMELEN) 435 return (EINVAL); 436 437 vmspace = VMSPACE_ALLOC(0, VM_MAXUSER_ADDRESS); 438 if (vmspace == NULL) 439 return (ENOMEM); 440 441 vm = malloc(sizeof(struct vm), M_VM, M_WAITOK | M_ZERO); 442 strcpy(vm->name, name); 443 vm->vmspace = vmspace; 444 mtx_init(&vm->rendezvous_mtx, "vm rendezvous lock", 0, MTX_DEF); 445 446 vm_init(vm, true); 447 448 *retvm = vm; 449 return (0); 450 } 451 452 static void 453 vm_cleanup(struct vm *vm, bool destroy) 454 { 455 struct mem_map *mm; 456 int i; 457 458 ppt_unassign_all(vm); 459 460 if (vm->iommu != NULL) 461 iommu_destroy_domain(vm->iommu); 462 463 if (destroy) 464 vrtc_cleanup(vm->vrtc); 465 else 466 vrtc_reset(vm->vrtc); 467 vpmtmr_cleanup(vm->vpmtmr); 468 vatpit_cleanup(vm->vatpit); 469 vhpet_cleanup(vm->vhpet); 470 vatpic_cleanup(vm->vatpic); 471 vioapic_cleanup(vm->vioapic); 472 473 for (i = 0; i < VM_MAXCPU; i++) 474 vcpu_cleanup(vm, i, destroy); 475 476 VMCLEANUP(vm->cookie); 477 478 /* 479 * System memory is removed from the guest address space only when 480 * the VM is destroyed. This is because the mapping remains the same 481 * across VM reset. 482 * 483 * Device memory can be relocated by the guest (e.g. using PCI BARs) 484 * so those mappings are removed on a VM reset. 485 */ 486 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 487 mm = &vm->mem_maps[i]; 488 if (destroy || !sysmem_mapping(vm, mm)) 489 vm_free_memmap(vm, i); 490 } 491 492 if (destroy) { 493 for (i = 0; i < VM_MAX_MEMSEGS; i++) 494 vm_free_memseg(vm, i); 495 496 VMSPACE_FREE(vm->vmspace); 497 vm->vmspace = NULL; 498 } 499 } 500 501 void 502 vm_destroy(struct vm *vm) 503 { 504 vm_cleanup(vm, true); 505 free(vm, M_VM); 506 } 507 508 int 509 vm_reinit(struct vm *vm) 510 { 511 int error; 512 513 /* 514 * A virtual machine can be reset only if all vcpus are suspended. 515 */ 516 if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) { 517 vm_cleanup(vm, false); 518 vm_init(vm, false); 519 error = 0; 520 } else { 521 error = EBUSY; 522 } 523 524 return (error); 525 } 526 527 const char * 528 vm_name(struct vm *vm) 529 { 530 return (vm->name); 531 } 532 533 int 534 vm_map_mmio(struct vm *vm, vm_paddr_t gpa, size_t len, vm_paddr_t hpa) 535 { 536 vm_object_t obj; 537 538 if ((obj = vmm_mmio_alloc(vm->vmspace, gpa, len, hpa)) == NULL) 539 return (ENOMEM); 540 else 541 return (0); 542 } 543 544 int 545 vm_unmap_mmio(struct vm *vm, vm_paddr_t gpa, size_t len) 546 { 547 548 vmm_mmio_free(vm->vmspace, gpa, len); 549 return (0); 550 } 551 552 /* 553 * Return 'true' if 'gpa' is allocated in the guest address space. 554 * 555 * This function is called in the context of a running vcpu which acts as 556 * an implicit lock on 'vm->mem_maps[]'. 557 */ 558 bool 559 vm_mem_allocated(struct vm *vm, int vcpuid, vm_paddr_t gpa) 560 { 561 struct mem_map *mm; 562 int i; 563 564 #ifdef INVARIANTS 565 int hostcpu, state; 566 state = vcpu_get_state(vm, vcpuid, &hostcpu); 567 KASSERT(state == VCPU_RUNNING && hostcpu == curcpu, 568 ("%s: invalid vcpu state %d/%d", __func__, state, hostcpu)); 569 #endif 570 571 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 572 mm = &vm->mem_maps[i]; 573 if (mm->len != 0 && gpa >= mm->gpa && gpa < mm->gpa + mm->len) 574 return (true); /* 'gpa' is sysmem or devmem */ 575 } 576 577 if (ppt_is_mmio(vm, gpa)) 578 return (true); /* 'gpa' is pci passthru mmio */ 579 580 return (false); 581 } 582 583 int 584 vm_alloc_memseg(struct vm *vm, int ident, size_t len, bool sysmem) 585 { 586 struct mem_seg *seg; 587 vm_object_t obj; 588 589 if (ident < 0 || ident >= VM_MAX_MEMSEGS) 590 return (EINVAL); 591 592 if (len == 0 || (len & PAGE_MASK)) 593 return (EINVAL); 594 595 seg = &vm->mem_segs[ident]; 596 if (seg->object != NULL) { 597 if (seg->len == len && seg->sysmem == sysmem) 598 return (EEXIST); 599 else 600 return (EINVAL); 601 } 602 603 obj = vm_object_allocate(OBJT_DEFAULT, len >> PAGE_SHIFT); 604 if (obj == NULL) 605 return (ENOMEM); 606 607 seg->len = len; 608 seg->object = obj; 609 seg->sysmem = sysmem; 610 return (0); 611 } 612 613 int 614 vm_get_memseg(struct vm *vm, int ident, size_t *len, bool *sysmem, 615 vm_object_t *objptr) 616 { 617 struct mem_seg *seg; 618 619 if (ident < 0 || ident >= VM_MAX_MEMSEGS) 620 return (EINVAL); 621 622 seg = &vm->mem_segs[ident]; 623 if (len) 624 *len = seg->len; 625 if (sysmem) 626 *sysmem = seg->sysmem; 627 if (objptr) 628 *objptr = seg->object; 629 return (0); 630 } 631 632 void 633 vm_free_memseg(struct vm *vm, int ident) 634 { 635 struct mem_seg *seg; 636 637 KASSERT(ident >= 0 && ident < VM_MAX_MEMSEGS, 638 ("%s: invalid memseg ident %d", __func__, ident)); 639 640 seg = &vm->mem_segs[ident]; 641 if (seg->object != NULL) { 642 vm_object_deallocate(seg->object); 643 bzero(seg, sizeof(struct mem_seg)); 644 } 645 } 646 647 int 648 vm_mmap_memseg(struct vm *vm, vm_paddr_t gpa, int segid, vm_ooffset_t first, 649 size_t len, int prot, int flags) 650 { 651 struct mem_seg *seg; 652 struct mem_map *m, *map; 653 vm_ooffset_t last; 654 int i, error; 655 656 if (prot == 0 || (prot & ~(VM_PROT_ALL)) != 0) 657 return (EINVAL); 658 659 if (flags & ~VM_MEMMAP_F_WIRED) 660 return (EINVAL); 661 662 if (segid < 0 || segid >= VM_MAX_MEMSEGS) 663 return (EINVAL); 664 665 seg = &vm->mem_segs[segid]; 666 if (seg->object == NULL) 667 return (EINVAL); 668 669 last = first + len; 670 if (first < 0 || first >= last || last > seg->len) 671 return (EINVAL); 672 673 if ((gpa | first | last) & PAGE_MASK) 674 return (EINVAL); 675 676 map = NULL; 677 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 678 m = &vm->mem_maps[i]; 679 if (m->len == 0) { 680 map = m; 681 break; 682 } 683 } 684 685 if (map == NULL) 686 return (ENOSPC); 687 688 error = vm_map_find(&vm->vmspace->vm_map, seg->object, first, &gpa, 689 len, 0, VMFS_NO_SPACE, prot, prot, 0); 690 if (error != KERN_SUCCESS) 691 return (EFAULT); 692 693 vm_object_reference(seg->object); 694 695 if (flags & VM_MEMMAP_F_WIRED) { 696 error = vm_map_wire(&vm->vmspace->vm_map, gpa, gpa + len, 697 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES); 698 if (error != KERN_SUCCESS) { 699 vm_map_remove(&vm->vmspace->vm_map, gpa, gpa + len); 700 return (EFAULT); 701 } 702 } 703 704 map->gpa = gpa; 705 map->len = len; 706 map->segoff = first; 707 map->segid = segid; 708 map->prot = prot; 709 map->flags = flags; 710 return (0); 711 } 712 713 int 714 vm_mmap_getnext(struct vm *vm, vm_paddr_t *gpa, int *segid, 715 vm_ooffset_t *segoff, size_t *len, int *prot, int *flags) 716 { 717 struct mem_map *mm, *mmnext; 718 int i; 719 720 mmnext = NULL; 721 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 722 mm = &vm->mem_maps[i]; 723 if (mm->len == 0 || mm->gpa < *gpa) 724 continue; 725 if (mmnext == NULL || mm->gpa < mmnext->gpa) 726 mmnext = mm; 727 } 728 729 if (mmnext != NULL) { 730 *gpa = mmnext->gpa; 731 if (segid) 732 *segid = mmnext->segid; 733 if (segoff) 734 *segoff = mmnext->segoff; 735 if (len) 736 *len = mmnext->len; 737 if (prot) 738 *prot = mmnext->prot; 739 if (flags) 740 *flags = mmnext->flags; 741 return (0); 742 } else { 743 return (ENOENT); 744 } 745 } 746 747 static void 748 vm_free_memmap(struct vm *vm, int ident) 749 { 750 struct mem_map *mm; 751 int error; 752 753 mm = &vm->mem_maps[ident]; 754 if (mm->len) { 755 error = vm_map_remove(&vm->vmspace->vm_map, mm->gpa, 756 mm->gpa + mm->len); 757 KASSERT(error == KERN_SUCCESS, ("%s: vm_map_remove error %d", 758 __func__, error)); 759 bzero(mm, sizeof(struct mem_map)); 760 } 761 } 762 763 static __inline bool 764 sysmem_mapping(struct vm *vm, struct mem_map *mm) 765 { 766 767 if (mm->len != 0 && vm->mem_segs[mm->segid].sysmem) 768 return (true); 769 else 770 return (false); 771 } 772 773 static vm_paddr_t 774 sysmem_maxaddr(struct vm *vm) 775 { 776 struct mem_map *mm; 777 vm_paddr_t maxaddr; 778 int i; 779 780 maxaddr = 0; 781 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 782 mm = &vm->mem_maps[i]; 783 if (sysmem_mapping(vm, mm)) { 784 if (maxaddr < mm->gpa + mm->len) 785 maxaddr = mm->gpa + mm->len; 786 } 787 } 788 return (maxaddr); 789 } 790 791 static void 792 vm_iommu_modify(struct vm *vm, boolean_t map) 793 { 794 int i, sz; 795 vm_paddr_t gpa, hpa; 796 struct mem_map *mm; 797 void *vp, *cookie, *host_domain; 798 799 sz = PAGE_SIZE; 800 host_domain = iommu_host_domain(); 801 802 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 803 mm = &vm->mem_maps[i]; 804 if (!sysmem_mapping(vm, mm)) 805 continue; 806 807 if (map) { 808 KASSERT((mm->flags & VM_MEMMAP_F_IOMMU) == 0, 809 ("iommu map found invalid memmap %#lx/%#lx/%#x", 810 mm->gpa, mm->len, mm->flags)); 811 if ((mm->flags & VM_MEMMAP_F_WIRED) == 0) 812 continue; 813 mm->flags |= VM_MEMMAP_F_IOMMU; 814 } else { 815 if ((mm->flags & VM_MEMMAP_F_IOMMU) == 0) 816 continue; 817 mm->flags &= ~VM_MEMMAP_F_IOMMU; 818 KASSERT((mm->flags & VM_MEMMAP_F_WIRED) != 0, 819 ("iommu unmap found invalid memmap %#lx/%#lx/%#x", 820 mm->gpa, mm->len, mm->flags)); 821 } 822 823 gpa = mm->gpa; 824 while (gpa < mm->gpa + mm->len) { 825 vp = vm_gpa_hold(vm, -1, gpa, PAGE_SIZE, VM_PROT_WRITE, 826 &cookie); 827 KASSERT(vp != NULL, ("vm(%s) could not map gpa %#lx", 828 vm_name(vm), gpa)); 829 830 vm_gpa_release(cookie); 831 832 hpa = DMAP_TO_PHYS((uintptr_t)vp); 833 if (map) { 834 iommu_create_mapping(vm->iommu, gpa, hpa, sz); 835 iommu_remove_mapping(host_domain, hpa, sz); 836 } else { 837 iommu_remove_mapping(vm->iommu, gpa, sz); 838 iommu_create_mapping(host_domain, hpa, hpa, sz); 839 } 840 841 gpa += PAGE_SIZE; 842 } 843 } 844 845 /* 846 * Invalidate the cached translations associated with the domain 847 * from which pages were removed. 848 */ 849 if (map) 850 iommu_invalidate_tlb(host_domain); 851 else 852 iommu_invalidate_tlb(vm->iommu); 853 } 854 855 #define vm_iommu_unmap(vm) vm_iommu_modify((vm), FALSE) 856 #define vm_iommu_map(vm) vm_iommu_modify((vm), TRUE) 857 858 int 859 vm_unassign_pptdev(struct vm *vm, int bus, int slot, int func) 860 { 861 int error; 862 863 error = ppt_unassign_device(vm, bus, slot, func); 864 if (error) 865 return (error); 866 867 if (ppt_assigned_devices(vm) == 0) 868 vm_iommu_unmap(vm); 869 870 return (0); 871 } 872 873 int 874 vm_assign_pptdev(struct vm *vm, int bus, int slot, int func) 875 { 876 int error; 877 vm_paddr_t maxaddr; 878 879 /* Set up the IOMMU to do the 'gpa' to 'hpa' translation */ 880 if (ppt_assigned_devices(vm) == 0) { 881 KASSERT(vm->iommu == NULL, 882 ("vm_assign_pptdev: iommu must be NULL")); 883 maxaddr = sysmem_maxaddr(vm); 884 vm->iommu = iommu_create_domain(maxaddr); 885 if (vm->iommu == NULL) 886 return (ENXIO); 887 vm_iommu_map(vm); 888 } 889 890 error = ppt_assign_device(vm, bus, slot, func); 891 return (error); 892 } 893 894 void * 895 vm_gpa_hold(struct vm *vm, int vcpuid, vm_paddr_t gpa, size_t len, int reqprot, 896 void **cookie) 897 { 898 int i, count, pageoff; 899 struct mem_map *mm; 900 vm_page_t m; 901 #ifdef INVARIANTS 902 /* 903 * All vcpus are frozen by ioctls that modify the memory map 904 * (e.g. VM_MMAP_MEMSEG). Therefore 'vm->memmap[]' stability is 905 * guaranteed if at least one vcpu is in the VCPU_FROZEN state. 906 */ 907 int state; 908 KASSERT(vcpuid >= -1 && vcpuid < VM_MAXCPU, ("%s: invalid vcpuid %d", 909 __func__, vcpuid)); 910 for (i = 0; i < VM_MAXCPU; i++) { 911 if (vcpuid != -1 && vcpuid != i) 912 continue; 913 state = vcpu_get_state(vm, i, NULL); 914 KASSERT(state == VCPU_FROZEN, ("%s: invalid vcpu state %d", 915 __func__, state)); 916 } 917 #endif 918 pageoff = gpa & PAGE_MASK; 919 if (len > PAGE_SIZE - pageoff) 920 panic("vm_gpa_hold: invalid gpa/len: 0x%016lx/%lu", gpa, len); 921 922 count = 0; 923 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 924 mm = &vm->mem_maps[i]; 925 if (sysmem_mapping(vm, mm) && gpa >= mm->gpa && 926 gpa < mm->gpa + mm->len) { 927 count = vm_fault_quick_hold_pages(&vm->vmspace->vm_map, 928 trunc_page(gpa), PAGE_SIZE, reqprot, &m, 1); 929 break; 930 } 931 } 932 933 if (count == 1) { 934 *cookie = m; 935 return ((void *)(PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m)) + pageoff)); 936 } else { 937 *cookie = NULL; 938 return (NULL); 939 } 940 } 941 942 void 943 vm_gpa_release(void *cookie) 944 { 945 vm_page_t m = cookie; 946 947 vm_page_lock(m); 948 vm_page_unhold(m); 949 vm_page_unlock(m); 950 } 951 952 int 953 vm_get_register(struct vm *vm, int vcpu, int reg, uint64_t *retval) 954 { 955 956 if (vcpu < 0 || vcpu >= VM_MAXCPU) 957 return (EINVAL); 958 959 if (reg >= VM_REG_LAST) 960 return (EINVAL); 961 962 return (VMGETREG(vm->cookie, vcpu, reg, retval)); 963 } 964 965 int 966 vm_set_register(struct vm *vm, int vcpuid, int reg, uint64_t val) 967 { 968 struct vcpu *vcpu; 969 int error; 970 971 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 972 return (EINVAL); 973 974 if (reg >= VM_REG_LAST) 975 return (EINVAL); 976 977 error = VMSETREG(vm->cookie, vcpuid, reg, val); 978 if (error || reg != VM_REG_GUEST_RIP) 979 return (error); 980 981 /* Set 'nextrip' to match the value of %rip */ 982 VCPU_CTR1(vm, vcpuid, "Setting nextrip to %#lx", val); 983 vcpu = &vm->vcpu[vcpuid]; 984 vcpu->nextrip = val; 985 return (0); 986 } 987 988 static boolean_t 989 is_descriptor_table(int reg) 990 { 991 992 switch (reg) { 993 case VM_REG_GUEST_IDTR: 994 case VM_REG_GUEST_GDTR: 995 return (TRUE); 996 default: 997 return (FALSE); 998 } 999 } 1000 1001 static boolean_t 1002 is_segment_register(int reg) 1003 { 1004 1005 switch (reg) { 1006 case VM_REG_GUEST_ES: 1007 case VM_REG_GUEST_CS: 1008 case VM_REG_GUEST_SS: 1009 case VM_REG_GUEST_DS: 1010 case VM_REG_GUEST_FS: 1011 case VM_REG_GUEST_GS: 1012 case VM_REG_GUEST_TR: 1013 case VM_REG_GUEST_LDTR: 1014 return (TRUE); 1015 default: 1016 return (FALSE); 1017 } 1018 } 1019 1020 int 1021 vm_get_seg_desc(struct vm *vm, int vcpu, int reg, 1022 struct seg_desc *desc) 1023 { 1024 1025 if (vcpu < 0 || vcpu >= VM_MAXCPU) 1026 return (EINVAL); 1027 1028 if (!is_segment_register(reg) && !is_descriptor_table(reg)) 1029 return (EINVAL); 1030 1031 return (VMGETDESC(vm->cookie, vcpu, reg, desc)); 1032 } 1033 1034 int 1035 vm_set_seg_desc(struct vm *vm, int vcpu, int reg, 1036 struct seg_desc *desc) 1037 { 1038 if (vcpu < 0 || vcpu >= VM_MAXCPU) 1039 return (EINVAL); 1040 1041 if (!is_segment_register(reg) && !is_descriptor_table(reg)) 1042 return (EINVAL); 1043 1044 return (VMSETDESC(vm->cookie, vcpu, reg, desc)); 1045 } 1046 1047 static void 1048 restore_guest_fpustate(struct vcpu *vcpu) 1049 { 1050 1051 /* flush host state to the pcb */ 1052 fpuexit(curthread); 1053 1054 /* restore guest FPU state */ 1055 fpu_stop_emulating(); 1056 fpurestore(vcpu->guestfpu); 1057 1058 /* restore guest XCR0 if XSAVE is enabled in the host */ 1059 if (rcr4() & CR4_XSAVE) 1060 load_xcr(0, vcpu->guest_xcr0); 1061 1062 /* 1063 * The FPU is now "dirty" with the guest's state so turn on emulation 1064 * to trap any access to the FPU by the host. 1065 */ 1066 fpu_start_emulating(); 1067 } 1068 1069 static void 1070 save_guest_fpustate(struct vcpu *vcpu) 1071 { 1072 1073 if ((rcr0() & CR0_TS) == 0) 1074 panic("fpu emulation not enabled in host!"); 1075 1076 /* save guest XCR0 and restore host XCR0 */ 1077 if (rcr4() & CR4_XSAVE) { 1078 vcpu->guest_xcr0 = rxcr(0); 1079 load_xcr(0, vmm_get_host_xcr0()); 1080 } 1081 1082 /* save guest FPU state */ 1083 fpu_stop_emulating(); 1084 fpusave(vcpu->guestfpu); 1085 fpu_start_emulating(); 1086 } 1087 1088 static VMM_STAT(VCPU_IDLE_TICKS, "number of ticks vcpu was idle"); 1089 1090 static int 1091 vcpu_set_state_locked(struct vm *vm, int vcpuid, enum vcpu_state newstate, 1092 bool from_idle) 1093 { 1094 struct vcpu *vcpu; 1095 int error; 1096 1097 vcpu = &vm->vcpu[vcpuid]; 1098 vcpu_assert_locked(vcpu); 1099 1100 /* 1101 * State transitions from the vmmdev_ioctl() must always begin from 1102 * the VCPU_IDLE state. This guarantees that there is only a single 1103 * ioctl() operating on a vcpu at any point. 1104 */ 1105 if (from_idle) { 1106 while (vcpu->state != VCPU_IDLE) { 1107 vcpu->reqidle = 1; 1108 vcpu_notify_event_locked(vcpu, false); 1109 VCPU_CTR1(vm, vcpuid, "vcpu state change from %s to " 1110 "idle requested", vcpu_state2str(vcpu->state)); 1111 msleep_spin(&vcpu->state, &vcpu->mtx, "vmstat", hz); 1112 } 1113 } else { 1114 KASSERT(vcpu->state != VCPU_IDLE, ("invalid transition from " 1115 "vcpu idle state")); 1116 } 1117 1118 if (vcpu->state == VCPU_RUNNING) { 1119 KASSERT(vcpu->hostcpu == curcpu, ("curcpu %d and hostcpu %d " 1120 "mismatch for running vcpu", curcpu, vcpu->hostcpu)); 1121 } else { 1122 KASSERT(vcpu->hostcpu == NOCPU, ("Invalid hostcpu %d for a " 1123 "vcpu that is not running", vcpu->hostcpu)); 1124 } 1125 1126 /* 1127 * The following state transitions are allowed: 1128 * IDLE -> FROZEN -> IDLE 1129 * FROZEN -> RUNNING -> FROZEN 1130 * FROZEN -> SLEEPING -> FROZEN 1131 */ 1132 switch (vcpu->state) { 1133 case VCPU_IDLE: 1134 case VCPU_RUNNING: 1135 case VCPU_SLEEPING: 1136 error = (newstate != VCPU_FROZEN); 1137 break; 1138 case VCPU_FROZEN: 1139 error = (newstate == VCPU_FROZEN); 1140 break; 1141 default: 1142 error = 1; 1143 break; 1144 } 1145 1146 if (error) 1147 return (EBUSY); 1148 1149 VCPU_CTR2(vm, vcpuid, "vcpu state changed from %s to %s", 1150 vcpu_state2str(vcpu->state), vcpu_state2str(newstate)); 1151 1152 vcpu->state = newstate; 1153 if (newstate == VCPU_RUNNING) 1154 vcpu->hostcpu = curcpu; 1155 else 1156 vcpu->hostcpu = NOCPU; 1157 1158 if (newstate == VCPU_IDLE) 1159 wakeup(&vcpu->state); 1160 1161 return (0); 1162 } 1163 1164 static void 1165 vcpu_require_state(struct vm *vm, int vcpuid, enum vcpu_state newstate) 1166 { 1167 int error; 1168 1169 if ((error = vcpu_set_state(vm, vcpuid, newstate, false)) != 0) 1170 panic("Error %d setting state to %d\n", error, newstate); 1171 } 1172 1173 static void 1174 vcpu_require_state_locked(struct vm *vm, int vcpuid, enum vcpu_state newstate) 1175 { 1176 int error; 1177 1178 if ((error = vcpu_set_state_locked(vm, vcpuid, newstate, false)) != 0) 1179 panic("Error %d setting state to %d", error, newstate); 1180 } 1181 1182 static void 1183 vm_set_rendezvous_func(struct vm *vm, vm_rendezvous_func_t func) 1184 { 1185 1186 KASSERT(mtx_owned(&vm->rendezvous_mtx), ("rendezvous_mtx not locked")); 1187 1188 /* 1189 * Update 'rendezvous_func' and execute a write memory barrier to 1190 * ensure that it is visible across all host cpus. This is not needed 1191 * for correctness but it does ensure that all the vcpus will notice 1192 * that the rendezvous is requested immediately. 1193 */ 1194 vm->rendezvous_func = func; 1195 wmb(); 1196 } 1197 1198 #define RENDEZVOUS_CTR0(vm, vcpuid, fmt) \ 1199 do { \ 1200 if (vcpuid >= 0) \ 1201 VCPU_CTR0(vm, vcpuid, fmt); \ 1202 else \ 1203 VM_CTR0(vm, fmt); \ 1204 } while (0) 1205 1206 static void 1207 vm_handle_rendezvous(struct vm *vm, int vcpuid) 1208 { 1209 1210 KASSERT(vcpuid == -1 || (vcpuid >= 0 && vcpuid < VM_MAXCPU), 1211 ("vm_handle_rendezvous: invalid vcpuid %d", vcpuid)); 1212 1213 mtx_lock(&vm->rendezvous_mtx); 1214 while (vm->rendezvous_func != NULL) { 1215 /* 'rendezvous_req_cpus' must be a subset of 'active_cpus' */ 1216 CPU_AND(&vm->rendezvous_req_cpus, &vm->active_cpus); 1217 1218 if (vcpuid != -1 && 1219 CPU_ISSET(vcpuid, &vm->rendezvous_req_cpus) && 1220 !CPU_ISSET(vcpuid, &vm->rendezvous_done_cpus)) { 1221 VCPU_CTR0(vm, vcpuid, "Calling rendezvous func"); 1222 (*vm->rendezvous_func)(vm, vcpuid, vm->rendezvous_arg); 1223 CPU_SET(vcpuid, &vm->rendezvous_done_cpus); 1224 } 1225 if (CPU_CMP(&vm->rendezvous_req_cpus, 1226 &vm->rendezvous_done_cpus) == 0) { 1227 VCPU_CTR0(vm, vcpuid, "Rendezvous completed"); 1228 vm_set_rendezvous_func(vm, NULL); 1229 wakeup(&vm->rendezvous_func); 1230 break; 1231 } 1232 RENDEZVOUS_CTR0(vm, vcpuid, "Wait for rendezvous completion"); 1233 mtx_sleep(&vm->rendezvous_func, &vm->rendezvous_mtx, 0, 1234 "vmrndv", 0); 1235 } 1236 mtx_unlock(&vm->rendezvous_mtx); 1237 } 1238 1239 /* 1240 * Emulate a guest 'hlt' by sleeping until the vcpu is ready to run. 1241 */ 1242 static int 1243 vm_handle_hlt(struct vm *vm, int vcpuid, bool intr_disabled, bool *retu) 1244 { 1245 struct vcpu *vcpu; 1246 const char *wmesg; 1247 int t, vcpu_halted, vm_halted; 1248 1249 KASSERT(!CPU_ISSET(vcpuid, &vm->halted_cpus), ("vcpu already halted")); 1250 1251 vcpu = &vm->vcpu[vcpuid]; 1252 vcpu_halted = 0; 1253 vm_halted = 0; 1254 1255 vcpu_lock(vcpu); 1256 while (1) { 1257 /* 1258 * Do a final check for pending NMI or interrupts before 1259 * really putting this thread to sleep. Also check for 1260 * software events that would cause this vcpu to wakeup. 1261 * 1262 * These interrupts/events could have happened after the 1263 * vcpu returned from VMRUN() and before it acquired the 1264 * vcpu lock above. 1265 */ 1266 if (vm->rendezvous_func != NULL || vm->suspend || vcpu->reqidle) 1267 break; 1268 if (vm_nmi_pending(vm, vcpuid)) 1269 break; 1270 if (!intr_disabled) { 1271 if (vm_extint_pending(vm, vcpuid) || 1272 vlapic_pending_intr(vcpu->vlapic, NULL)) { 1273 break; 1274 } 1275 } 1276 1277 /* Don't go to sleep if the vcpu thread needs to yield */ 1278 if (vcpu_should_yield(vm, vcpuid)) 1279 break; 1280 1281 /* 1282 * Some Linux guests implement "halt" by having all vcpus 1283 * execute HLT with interrupts disabled. 'halted_cpus' keeps 1284 * track of the vcpus that have entered this state. When all 1285 * vcpus enter the halted state the virtual machine is halted. 1286 */ 1287 if (intr_disabled) { 1288 wmesg = "vmhalt"; 1289 VCPU_CTR0(vm, vcpuid, "Halted"); 1290 if (!vcpu_halted && halt_detection_enabled) { 1291 vcpu_halted = 1; 1292 CPU_SET_ATOMIC(vcpuid, &vm->halted_cpus); 1293 } 1294 if (CPU_CMP(&vm->halted_cpus, &vm->active_cpus) == 0) { 1295 vm_halted = 1; 1296 break; 1297 } 1298 } else { 1299 wmesg = "vmidle"; 1300 } 1301 1302 t = ticks; 1303 vcpu_require_state_locked(vm, vcpuid, VCPU_SLEEPING); 1304 /* 1305 * XXX msleep_spin() cannot be interrupted by signals so 1306 * wake up periodically to check pending signals. 1307 */ 1308 msleep_spin(vcpu, &vcpu->mtx, wmesg, hz); 1309 vcpu_require_state_locked(vm, vcpuid, VCPU_FROZEN); 1310 vmm_stat_incr(vm, vcpuid, VCPU_IDLE_TICKS, ticks - t); 1311 } 1312 1313 if (vcpu_halted) 1314 CPU_CLR_ATOMIC(vcpuid, &vm->halted_cpus); 1315 1316 vcpu_unlock(vcpu); 1317 1318 if (vm_halted) 1319 vm_suspend(vm, VM_SUSPEND_HALT); 1320 1321 return (0); 1322 } 1323 1324 static int 1325 vm_handle_paging(struct vm *vm, int vcpuid, bool *retu) 1326 { 1327 int rv, ftype; 1328 struct vm_map *map; 1329 struct vcpu *vcpu; 1330 struct vm_exit *vme; 1331 1332 vcpu = &vm->vcpu[vcpuid]; 1333 vme = &vcpu->exitinfo; 1334 1335 KASSERT(vme->inst_length == 0, ("%s: invalid inst_length %d", 1336 __func__, vme->inst_length)); 1337 1338 ftype = vme->u.paging.fault_type; 1339 KASSERT(ftype == VM_PROT_READ || 1340 ftype == VM_PROT_WRITE || ftype == VM_PROT_EXECUTE, 1341 ("vm_handle_paging: invalid fault_type %d", ftype)); 1342 1343 if (ftype == VM_PROT_READ || ftype == VM_PROT_WRITE) { 1344 rv = pmap_emulate_accessed_dirty(vmspace_pmap(vm->vmspace), 1345 vme->u.paging.gpa, ftype); 1346 if (rv == 0) { 1347 VCPU_CTR2(vm, vcpuid, "%s bit emulation for gpa %#lx", 1348 ftype == VM_PROT_READ ? "accessed" : "dirty", 1349 vme->u.paging.gpa); 1350 goto done; 1351 } 1352 } 1353 1354 map = &vm->vmspace->vm_map; 1355 rv = vm_fault(map, vme->u.paging.gpa, ftype, VM_FAULT_NORMAL); 1356 1357 VCPU_CTR3(vm, vcpuid, "vm_handle_paging rv = %d, gpa = %#lx, " 1358 "ftype = %d", rv, vme->u.paging.gpa, ftype); 1359 1360 if (rv != KERN_SUCCESS) 1361 return (EFAULT); 1362 done: 1363 return (0); 1364 } 1365 1366 static int 1367 vm_handle_inst_emul(struct vm *vm, int vcpuid, bool *retu) 1368 { 1369 struct vie *vie; 1370 struct vcpu *vcpu; 1371 struct vm_exit *vme; 1372 uint64_t gla, gpa, cs_base; 1373 struct vm_guest_paging *paging; 1374 mem_region_read_t mread; 1375 mem_region_write_t mwrite; 1376 enum vm_cpu_mode cpu_mode; 1377 int cs_d, error, fault; 1378 1379 vcpu = &vm->vcpu[vcpuid]; 1380 vme = &vcpu->exitinfo; 1381 1382 KASSERT(vme->inst_length == 0, ("%s: invalid inst_length %d", 1383 __func__, vme->inst_length)); 1384 1385 gla = vme->u.inst_emul.gla; 1386 gpa = vme->u.inst_emul.gpa; 1387 cs_base = vme->u.inst_emul.cs_base; 1388 cs_d = vme->u.inst_emul.cs_d; 1389 vie = &vme->u.inst_emul.vie; 1390 paging = &vme->u.inst_emul.paging; 1391 cpu_mode = paging->cpu_mode; 1392 1393 VCPU_CTR1(vm, vcpuid, "inst_emul fault accessing gpa %#lx", gpa); 1394 1395 /* Fetch, decode and emulate the faulting instruction */ 1396 if (vie->num_valid == 0) { 1397 error = vmm_fetch_instruction(vm, vcpuid, paging, vme->rip + 1398 cs_base, VIE_INST_SIZE, vie, &fault); 1399 } else { 1400 /* 1401 * The instruction bytes have already been copied into 'vie' 1402 */ 1403 error = fault = 0; 1404 } 1405 if (error || fault) 1406 return (error); 1407 1408 if (vmm_decode_instruction(vm, vcpuid, gla, cpu_mode, cs_d, vie) != 0) { 1409 VCPU_CTR1(vm, vcpuid, "Error decoding instruction at %#lx", 1410 vme->rip + cs_base); 1411 *retu = true; /* dump instruction bytes in userspace */ 1412 return (0); 1413 } 1414 1415 /* 1416 * Update 'nextrip' based on the length of the emulated instruction. 1417 */ 1418 vme->inst_length = vie->num_processed; 1419 vcpu->nextrip += vie->num_processed; 1420 VCPU_CTR1(vm, vcpuid, "nextrip updated to %#lx after instruction " 1421 "decoding", vcpu->nextrip); 1422 1423 /* return to userland unless this is an in-kernel emulated device */ 1424 if (gpa >= DEFAULT_APIC_BASE && gpa < DEFAULT_APIC_BASE + PAGE_SIZE) { 1425 mread = lapic_mmio_read; 1426 mwrite = lapic_mmio_write; 1427 } else if (gpa >= VIOAPIC_BASE && gpa < VIOAPIC_BASE + VIOAPIC_SIZE) { 1428 mread = vioapic_mmio_read; 1429 mwrite = vioapic_mmio_write; 1430 } else if (gpa >= VHPET_BASE && gpa < VHPET_BASE + VHPET_SIZE) { 1431 mread = vhpet_mmio_read; 1432 mwrite = vhpet_mmio_write; 1433 } else { 1434 *retu = true; 1435 return (0); 1436 } 1437 1438 error = vmm_emulate_instruction(vm, vcpuid, gpa, vie, paging, 1439 mread, mwrite, retu); 1440 1441 return (error); 1442 } 1443 1444 static int 1445 vm_handle_suspend(struct vm *vm, int vcpuid, bool *retu) 1446 { 1447 int i, done; 1448 struct vcpu *vcpu; 1449 1450 done = 0; 1451 vcpu = &vm->vcpu[vcpuid]; 1452 1453 CPU_SET_ATOMIC(vcpuid, &vm->suspended_cpus); 1454 1455 /* 1456 * Wait until all 'active_cpus' have suspended themselves. 1457 * 1458 * Since a VM may be suspended at any time including when one or 1459 * more vcpus are doing a rendezvous we need to call the rendezvous 1460 * handler while we are waiting to prevent a deadlock. 1461 */ 1462 vcpu_lock(vcpu); 1463 while (1) { 1464 if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) { 1465 VCPU_CTR0(vm, vcpuid, "All vcpus suspended"); 1466 break; 1467 } 1468 1469 if (vm->rendezvous_func == NULL) { 1470 VCPU_CTR0(vm, vcpuid, "Sleeping during suspend"); 1471 vcpu_require_state_locked(vm, vcpuid, VCPU_SLEEPING); 1472 msleep_spin(vcpu, &vcpu->mtx, "vmsusp", hz); 1473 vcpu_require_state_locked(vm, vcpuid, VCPU_FROZEN); 1474 } else { 1475 VCPU_CTR0(vm, vcpuid, "Rendezvous during suspend"); 1476 vcpu_unlock(vcpu); 1477 vm_handle_rendezvous(vm, vcpuid); 1478 vcpu_lock(vcpu); 1479 } 1480 } 1481 vcpu_unlock(vcpu); 1482 1483 /* 1484 * Wakeup the other sleeping vcpus and return to userspace. 1485 */ 1486 for (i = 0; i < VM_MAXCPU; i++) { 1487 if (CPU_ISSET(i, &vm->suspended_cpus)) { 1488 vcpu_notify_event(vm, i, false); 1489 } 1490 } 1491 1492 *retu = true; 1493 return (0); 1494 } 1495 1496 static int 1497 vm_handle_reqidle(struct vm *vm, int vcpuid, bool *retu) 1498 { 1499 struct vcpu *vcpu = &vm->vcpu[vcpuid]; 1500 1501 vcpu_lock(vcpu); 1502 KASSERT(vcpu->reqidle, ("invalid vcpu reqidle %d", vcpu->reqidle)); 1503 vcpu->reqidle = 0; 1504 vcpu_unlock(vcpu); 1505 *retu = true; 1506 return (0); 1507 } 1508 1509 int 1510 vm_suspend(struct vm *vm, enum vm_suspend_how how) 1511 { 1512 int i; 1513 1514 if (how <= VM_SUSPEND_NONE || how >= VM_SUSPEND_LAST) 1515 return (EINVAL); 1516 1517 if (atomic_cmpset_int(&vm->suspend, 0, how) == 0) { 1518 VM_CTR2(vm, "virtual machine already suspended %d/%d", 1519 vm->suspend, how); 1520 return (EALREADY); 1521 } 1522 1523 VM_CTR1(vm, "virtual machine successfully suspended %d", how); 1524 1525 /* 1526 * Notify all active vcpus that they are now suspended. 1527 */ 1528 for (i = 0; i < VM_MAXCPU; i++) { 1529 if (CPU_ISSET(i, &vm->active_cpus)) 1530 vcpu_notify_event(vm, i, false); 1531 } 1532 1533 return (0); 1534 } 1535 1536 void 1537 vm_exit_suspended(struct vm *vm, int vcpuid, uint64_t rip) 1538 { 1539 struct vm_exit *vmexit; 1540 1541 KASSERT(vm->suspend > VM_SUSPEND_NONE && vm->suspend < VM_SUSPEND_LAST, 1542 ("vm_exit_suspended: invalid suspend type %d", vm->suspend)); 1543 1544 vmexit = vm_exitinfo(vm, vcpuid); 1545 vmexit->rip = rip; 1546 vmexit->inst_length = 0; 1547 vmexit->exitcode = VM_EXITCODE_SUSPENDED; 1548 vmexit->u.suspended.how = vm->suspend; 1549 } 1550 1551 void 1552 vm_exit_rendezvous(struct vm *vm, int vcpuid, uint64_t rip) 1553 { 1554 struct vm_exit *vmexit; 1555 1556 KASSERT(vm->rendezvous_func != NULL, ("rendezvous not in progress")); 1557 1558 vmexit = vm_exitinfo(vm, vcpuid); 1559 vmexit->rip = rip; 1560 vmexit->inst_length = 0; 1561 vmexit->exitcode = VM_EXITCODE_RENDEZVOUS; 1562 vmm_stat_incr(vm, vcpuid, VMEXIT_RENDEZVOUS, 1); 1563 } 1564 1565 void 1566 vm_exit_reqidle(struct vm *vm, int vcpuid, uint64_t rip) 1567 { 1568 struct vm_exit *vmexit; 1569 1570 vmexit = vm_exitinfo(vm, vcpuid); 1571 vmexit->rip = rip; 1572 vmexit->inst_length = 0; 1573 vmexit->exitcode = VM_EXITCODE_REQIDLE; 1574 vmm_stat_incr(vm, vcpuid, VMEXIT_REQIDLE, 1); 1575 } 1576 1577 void 1578 vm_exit_astpending(struct vm *vm, int vcpuid, uint64_t rip) 1579 { 1580 struct vm_exit *vmexit; 1581 1582 vmexit = vm_exitinfo(vm, vcpuid); 1583 vmexit->rip = rip; 1584 vmexit->inst_length = 0; 1585 vmexit->exitcode = VM_EXITCODE_BOGUS; 1586 vmm_stat_incr(vm, vcpuid, VMEXIT_ASTPENDING, 1); 1587 } 1588 1589 int 1590 vm_run(struct vm *vm, struct vm_run *vmrun) 1591 { 1592 struct vm_eventinfo evinfo; 1593 int error, vcpuid; 1594 struct vcpu *vcpu; 1595 struct pcb *pcb; 1596 uint64_t tscval; 1597 struct vm_exit *vme; 1598 bool retu, intr_disabled; 1599 pmap_t pmap; 1600 1601 vcpuid = vmrun->cpuid; 1602 1603 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1604 return (EINVAL); 1605 1606 if (!CPU_ISSET(vcpuid, &vm->active_cpus)) 1607 return (EINVAL); 1608 1609 if (CPU_ISSET(vcpuid, &vm->suspended_cpus)) 1610 return (EINVAL); 1611 1612 pmap = vmspace_pmap(vm->vmspace); 1613 vcpu = &vm->vcpu[vcpuid]; 1614 vme = &vcpu->exitinfo; 1615 evinfo.rptr = &vm->rendezvous_func; 1616 evinfo.sptr = &vm->suspend; 1617 evinfo.iptr = &vcpu->reqidle; 1618 restart: 1619 critical_enter(); 1620 1621 KASSERT(!CPU_ISSET(curcpu, &pmap->pm_active), 1622 ("vm_run: absurd pm_active")); 1623 1624 tscval = rdtsc(); 1625 1626 pcb = PCPU_GET(curpcb); 1627 set_pcb_flags(pcb, PCB_FULL_IRET); 1628 1629 restore_guest_fpustate(vcpu); 1630 1631 vcpu_require_state(vm, vcpuid, VCPU_RUNNING); 1632 error = VMRUN(vm->cookie, vcpuid, vcpu->nextrip, pmap, &evinfo); 1633 vcpu_require_state(vm, vcpuid, VCPU_FROZEN); 1634 1635 save_guest_fpustate(vcpu); 1636 1637 vmm_stat_incr(vm, vcpuid, VCPU_TOTAL_RUNTIME, rdtsc() - tscval); 1638 1639 critical_exit(); 1640 1641 if (error == 0) { 1642 retu = false; 1643 vcpu->nextrip = vme->rip + vme->inst_length; 1644 switch (vme->exitcode) { 1645 case VM_EXITCODE_REQIDLE: 1646 error = vm_handle_reqidle(vm, vcpuid, &retu); 1647 break; 1648 case VM_EXITCODE_SUSPENDED: 1649 error = vm_handle_suspend(vm, vcpuid, &retu); 1650 break; 1651 case VM_EXITCODE_IOAPIC_EOI: 1652 vioapic_process_eoi(vm, vcpuid, 1653 vme->u.ioapic_eoi.vector); 1654 break; 1655 case VM_EXITCODE_RENDEZVOUS: 1656 vm_handle_rendezvous(vm, vcpuid); 1657 error = 0; 1658 break; 1659 case VM_EXITCODE_HLT: 1660 intr_disabled = ((vme->u.hlt.rflags & PSL_I) == 0); 1661 error = vm_handle_hlt(vm, vcpuid, intr_disabled, &retu); 1662 break; 1663 case VM_EXITCODE_PAGING: 1664 error = vm_handle_paging(vm, vcpuid, &retu); 1665 break; 1666 case VM_EXITCODE_INST_EMUL: 1667 error = vm_handle_inst_emul(vm, vcpuid, &retu); 1668 break; 1669 case VM_EXITCODE_INOUT: 1670 case VM_EXITCODE_INOUT_STR: 1671 error = vm_handle_inout(vm, vcpuid, vme, &retu); 1672 break; 1673 case VM_EXITCODE_MONITOR: 1674 case VM_EXITCODE_MWAIT: 1675 vm_inject_ud(vm, vcpuid); 1676 break; 1677 default: 1678 retu = true; /* handled in userland */ 1679 break; 1680 } 1681 } 1682 1683 if (error == 0 && retu == false) 1684 goto restart; 1685 1686 VCPU_CTR2(vm, vcpuid, "retu %d/%d", error, vme->exitcode); 1687 1688 /* copy the exit information */ 1689 bcopy(vme, &vmrun->vm_exit, sizeof(struct vm_exit)); 1690 return (error); 1691 } 1692 1693 int 1694 vm_restart_instruction(void *arg, int vcpuid) 1695 { 1696 struct vm *vm; 1697 struct vcpu *vcpu; 1698 enum vcpu_state state; 1699 uint64_t rip; 1700 int error; 1701 1702 vm = arg; 1703 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1704 return (EINVAL); 1705 1706 vcpu = &vm->vcpu[vcpuid]; 1707 state = vcpu_get_state(vm, vcpuid, NULL); 1708 if (state == VCPU_RUNNING) { 1709 /* 1710 * When a vcpu is "running" the next instruction is determined 1711 * by adding 'rip' and 'inst_length' in the vcpu's 'exitinfo'. 1712 * Thus setting 'inst_length' to zero will cause the current 1713 * instruction to be restarted. 1714 */ 1715 vcpu->exitinfo.inst_length = 0; 1716 VCPU_CTR1(vm, vcpuid, "restarting instruction at %#lx by " 1717 "setting inst_length to zero", vcpu->exitinfo.rip); 1718 } else if (state == VCPU_FROZEN) { 1719 /* 1720 * When a vcpu is "frozen" it is outside the critical section 1721 * around VMRUN() and 'nextrip' points to the next instruction. 1722 * Thus instruction restart is achieved by setting 'nextrip' 1723 * to the vcpu's %rip. 1724 */ 1725 error = vm_get_register(vm, vcpuid, VM_REG_GUEST_RIP, &rip); 1726 KASSERT(!error, ("%s: error %d getting rip", __func__, error)); 1727 VCPU_CTR2(vm, vcpuid, "restarting instruction by updating " 1728 "nextrip from %#lx to %#lx", vcpu->nextrip, rip); 1729 vcpu->nextrip = rip; 1730 } else { 1731 panic("%s: invalid state %d", __func__, state); 1732 } 1733 return (0); 1734 } 1735 1736 int 1737 vm_exit_intinfo(struct vm *vm, int vcpuid, uint64_t info) 1738 { 1739 struct vcpu *vcpu; 1740 int type, vector; 1741 1742 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1743 return (EINVAL); 1744 1745 vcpu = &vm->vcpu[vcpuid]; 1746 1747 if (info & VM_INTINFO_VALID) { 1748 type = info & VM_INTINFO_TYPE; 1749 vector = info & 0xff; 1750 if (type == VM_INTINFO_NMI && vector != IDT_NMI) 1751 return (EINVAL); 1752 if (type == VM_INTINFO_HWEXCEPTION && vector >= 32) 1753 return (EINVAL); 1754 if (info & VM_INTINFO_RSVD) 1755 return (EINVAL); 1756 } else { 1757 info = 0; 1758 } 1759 VCPU_CTR2(vm, vcpuid, "%s: info1(%#lx)", __func__, info); 1760 vcpu->exitintinfo = info; 1761 return (0); 1762 } 1763 1764 enum exc_class { 1765 EXC_BENIGN, 1766 EXC_CONTRIBUTORY, 1767 EXC_PAGEFAULT 1768 }; 1769 1770 #define IDT_VE 20 /* Virtualization Exception (Intel specific) */ 1771 1772 static enum exc_class 1773 exception_class(uint64_t info) 1774 { 1775 int type, vector; 1776 1777 KASSERT(info & VM_INTINFO_VALID, ("intinfo must be valid: %#lx", info)); 1778 type = info & VM_INTINFO_TYPE; 1779 vector = info & 0xff; 1780 1781 /* Table 6-4, "Interrupt and Exception Classes", Intel SDM, Vol 3 */ 1782 switch (type) { 1783 case VM_INTINFO_HWINTR: 1784 case VM_INTINFO_SWINTR: 1785 case VM_INTINFO_NMI: 1786 return (EXC_BENIGN); 1787 default: 1788 /* 1789 * Hardware exception. 1790 * 1791 * SVM and VT-x use identical type values to represent NMI, 1792 * hardware interrupt and software interrupt. 1793 * 1794 * SVM uses type '3' for all exceptions. VT-x uses type '3' 1795 * for exceptions except #BP and #OF. #BP and #OF use a type 1796 * value of '5' or '6'. Therefore we don't check for explicit 1797 * values of 'type' to classify 'intinfo' into a hardware 1798 * exception. 1799 */ 1800 break; 1801 } 1802 1803 switch (vector) { 1804 case IDT_PF: 1805 case IDT_VE: 1806 return (EXC_PAGEFAULT); 1807 case IDT_DE: 1808 case IDT_TS: 1809 case IDT_NP: 1810 case IDT_SS: 1811 case IDT_GP: 1812 return (EXC_CONTRIBUTORY); 1813 default: 1814 return (EXC_BENIGN); 1815 } 1816 } 1817 1818 static int 1819 nested_fault(struct vm *vm, int vcpuid, uint64_t info1, uint64_t info2, 1820 uint64_t *retinfo) 1821 { 1822 enum exc_class exc1, exc2; 1823 int type1, vector1; 1824 1825 KASSERT(info1 & VM_INTINFO_VALID, ("info1 %#lx is not valid", info1)); 1826 KASSERT(info2 & VM_INTINFO_VALID, ("info2 %#lx is not valid", info2)); 1827 1828 /* 1829 * If an exception occurs while attempting to call the double-fault 1830 * handler the processor enters shutdown mode (aka triple fault). 1831 */ 1832 type1 = info1 & VM_INTINFO_TYPE; 1833 vector1 = info1 & 0xff; 1834 if (type1 == VM_INTINFO_HWEXCEPTION && vector1 == IDT_DF) { 1835 VCPU_CTR2(vm, vcpuid, "triple fault: info1(%#lx), info2(%#lx)", 1836 info1, info2); 1837 vm_suspend(vm, VM_SUSPEND_TRIPLEFAULT); 1838 *retinfo = 0; 1839 return (0); 1840 } 1841 1842 /* 1843 * Table 6-5 "Conditions for Generating a Double Fault", Intel SDM, Vol3 1844 */ 1845 exc1 = exception_class(info1); 1846 exc2 = exception_class(info2); 1847 if ((exc1 == EXC_CONTRIBUTORY && exc2 == EXC_CONTRIBUTORY) || 1848 (exc1 == EXC_PAGEFAULT && exc2 != EXC_BENIGN)) { 1849 /* Convert nested fault into a double fault. */ 1850 *retinfo = IDT_DF; 1851 *retinfo |= VM_INTINFO_VALID | VM_INTINFO_HWEXCEPTION; 1852 *retinfo |= VM_INTINFO_DEL_ERRCODE; 1853 } else { 1854 /* Handle exceptions serially */ 1855 *retinfo = info2; 1856 } 1857 return (1); 1858 } 1859 1860 static uint64_t 1861 vcpu_exception_intinfo(struct vcpu *vcpu) 1862 { 1863 uint64_t info = 0; 1864 1865 if (vcpu->exception_pending) { 1866 info = vcpu->exc_vector & 0xff; 1867 info |= VM_INTINFO_VALID | VM_INTINFO_HWEXCEPTION; 1868 if (vcpu->exc_errcode_valid) { 1869 info |= VM_INTINFO_DEL_ERRCODE; 1870 info |= (uint64_t)vcpu->exc_errcode << 32; 1871 } 1872 } 1873 return (info); 1874 } 1875 1876 int 1877 vm_entry_intinfo(struct vm *vm, int vcpuid, uint64_t *retinfo) 1878 { 1879 struct vcpu *vcpu; 1880 uint64_t info1, info2; 1881 int valid; 1882 1883 KASSERT(vcpuid >= 0 && vcpuid < VM_MAXCPU, ("invalid vcpu %d", vcpuid)); 1884 1885 vcpu = &vm->vcpu[vcpuid]; 1886 1887 info1 = vcpu->exitintinfo; 1888 vcpu->exitintinfo = 0; 1889 1890 info2 = 0; 1891 if (vcpu->exception_pending) { 1892 info2 = vcpu_exception_intinfo(vcpu); 1893 vcpu->exception_pending = 0; 1894 VCPU_CTR2(vm, vcpuid, "Exception %d delivered: %#lx", 1895 vcpu->exc_vector, info2); 1896 } 1897 1898 if ((info1 & VM_INTINFO_VALID) && (info2 & VM_INTINFO_VALID)) { 1899 valid = nested_fault(vm, vcpuid, info1, info2, retinfo); 1900 } else if (info1 & VM_INTINFO_VALID) { 1901 *retinfo = info1; 1902 valid = 1; 1903 } else if (info2 & VM_INTINFO_VALID) { 1904 *retinfo = info2; 1905 valid = 1; 1906 } else { 1907 valid = 0; 1908 } 1909 1910 if (valid) { 1911 VCPU_CTR4(vm, vcpuid, "%s: info1(%#lx), info2(%#lx), " 1912 "retinfo(%#lx)", __func__, info1, info2, *retinfo); 1913 } 1914 1915 return (valid); 1916 } 1917 1918 int 1919 vm_get_intinfo(struct vm *vm, int vcpuid, uint64_t *info1, uint64_t *info2) 1920 { 1921 struct vcpu *vcpu; 1922 1923 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1924 return (EINVAL); 1925 1926 vcpu = &vm->vcpu[vcpuid]; 1927 *info1 = vcpu->exitintinfo; 1928 *info2 = vcpu_exception_intinfo(vcpu); 1929 return (0); 1930 } 1931 1932 int 1933 vm_inject_exception(struct vm *vm, int vcpuid, int vector, int errcode_valid, 1934 uint32_t errcode, int restart_instruction) 1935 { 1936 struct vcpu *vcpu; 1937 uint64_t regval; 1938 int error; 1939 1940 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1941 return (EINVAL); 1942 1943 if (vector < 0 || vector >= 32) 1944 return (EINVAL); 1945 1946 /* 1947 * A double fault exception should never be injected directly into 1948 * the guest. It is a derived exception that results from specific 1949 * combinations of nested faults. 1950 */ 1951 if (vector == IDT_DF) 1952 return (EINVAL); 1953 1954 vcpu = &vm->vcpu[vcpuid]; 1955 1956 if (vcpu->exception_pending) { 1957 VCPU_CTR2(vm, vcpuid, "Unable to inject exception %d due to " 1958 "pending exception %d", vector, vcpu->exc_vector); 1959 return (EBUSY); 1960 } 1961 1962 if (errcode_valid) { 1963 /* 1964 * Exceptions don't deliver an error code in real mode. 1965 */ 1966 error = vm_get_register(vm, vcpuid, VM_REG_GUEST_CR0, ®val); 1967 KASSERT(!error, ("%s: error %d getting CR0", __func__, error)); 1968 if (!(regval & CR0_PE)) 1969 errcode_valid = 0; 1970 } 1971 1972 /* 1973 * From section 26.6.1 "Interruptibility State" in Intel SDM: 1974 * 1975 * Event blocking by "STI" or "MOV SS" is cleared after guest executes 1976 * one instruction or incurs an exception. 1977 */ 1978 error = vm_set_register(vm, vcpuid, VM_REG_GUEST_INTR_SHADOW, 0); 1979 KASSERT(error == 0, ("%s: error %d clearing interrupt shadow", 1980 __func__, error)); 1981 1982 if (restart_instruction) 1983 vm_restart_instruction(vm, vcpuid); 1984 1985 vcpu->exception_pending = 1; 1986 vcpu->exc_vector = vector; 1987 vcpu->exc_errcode = errcode; 1988 vcpu->exc_errcode_valid = errcode_valid; 1989 VCPU_CTR1(vm, vcpuid, "Exception %d pending", vector); 1990 return (0); 1991 } 1992 1993 void 1994 vm_inject_fault(void *vmarg, int vcpuid, int vector, int errcode_valid, 1995 int errcode) 1996 { 1997 struct vm *vm; 1998 int error, restart_instruction; 1999 2000 vm = vmarg; 2001 restart_instruction = 1; 2002 2003 error = vm_inject_exception(vm, vcpuid, vector, errcode_valid, 2004 errcode, restart_instruction); 2005 KASSERT(error == 0, ("vm_inject_exception error %d", error)); 2006 } 2007 2008 void 2009 vm_inject_pf(void *vmarg, int vcpuid, int error_code, uint64_t cr2) 2010 { 2011 struct vm *vm; 2012 int error; 2013 2014 vm = vmarg; 2015 VCPU_CTR2(vm, vcpuid, "Injecting page fault: error_code %#x, cr2 %#lx", 2016 error_code, cr2); 2017 2018 error = vm_set_register(vm, vcpuid, VM_REG_GUEST_CR2, cr2); 2019 KASSERT(error == 0, ("vm_set_register(cr2) error %d", error)); 2020 2021 vm_inject_fault(vm, vcpuid, IDT_PF, 1, error_code); 2022 } 2023 2024 static VMM_STAT(VCPU_NMI_COUNT, "number of NMIs delivered to vcpu"); 2025 2026 int 2027 vm_inject_nmi(struct vm *vm, int vcpuid) 2028 { 2029 struct vcpu *vcpu; 2030 2031 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 2032 return (EINVAL); 2033 2034 vcpu = &vm->vcpu[vcpuid]; 2035 2036 vcpu->nmi_pending = 1; 2037 vcpu_notify_event(vm, vcpuid, false); 2038 return (0); 2039 } 2040 2041 int 2042 vm_nmi_pending(struct vm *vm, int vcpuid) 2043 { 2044 struct vcpu *vcpu; 2045 2046 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 2047 panic("vm_nmi_pending: invalid vcpuid %d", vcpuid); 2048 2049 vcpu = &vm->vcpu[vcpuid]; 2050 2051 return (vcpu->nmi_pending); 2052 } 2053 2054 void 2055 vm_nmi_clear(struct vm *vm, int vcpuid) 2056 { 2057 struct vcpu *vcpu; 2058 2059 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 2060 panic("vm_nmi_pending: invalid vcpuid %d", vcpuid); 2061 2062 vcpu = &vm->vcpu[vcpuid]; 2063 2064 if (vcpu->nmi_pending == 0) 2065 panic("vm_nmi_clear: inconsistent nmi_pending state"); 2066 2067 vcpu->nmi_pending = 0; 2068 vmm_stat_incr(vm, vcpuid, VCPU_NMI_COUNT, 1); 2069 } 2070 2071 static VMM_STAT(VCPU_EXTINT_COUNT, "number of ExtINTs delivered to vcpu"); 2072 2073 int 2074 vm_inject_extint(struct vm *vm, int vcpuid) 2075 { 2076 struct vcpu *vcpu; 2077 2078 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 2079 return (EINVAL); 2080 2081 vcpu = &vm->vcpu[vcpuid]; 2082 2083 vcpu->extint_pending = 1; 2084 vcpu_notify_event(vm, vcpuid, false); 2085 return (0); 2086 } 2087 2088 int 2089 vm_extint_pending(struct vm *vm, int vcpuid) 2090 { 2091 struct vcpu *vcpu; 2092 2093 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 2094 panic("vm_extint_pending: invalid vcpuid %d", vcpuid); 2095 2096 vcpu = &vm->vcpu[vcpuid]; 2097 2098 return (vcpu->extint_pending); 2099 } 2100 2101 void 2102 vm_extint_clear(struct vm *vm, int vcpuid) 2103 { 2104 struct vcpu *vcpu; 2105 2106 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 2107 panic("vm_extint_pending: invalid vcpuid %d", vcpuid); 2108 2109 vcpu = &vm->vcpu[vcpuid]; 2110 2111 if (vcpu->extint_pending == 0) 2112 panic("vm_extint_clear: inconsistent extint_pending state"); 2113 2114 vcpu->extint_pending = 0; 2115 vmm_stat_incr(vm, vcpuid, VCPU_EXTINT_COUNT, 1); 2116 } 2117 2118 int 2119 vm_get_capability(struct vm *vm, int vcpu, int type, int *retval) 2120 { 2121 if (vcpu < 0 || vcpu >= VM_MAXCPU) 2122 return (EINVAL); 2123 2124 if (type < 0 || type >= VM_CAP_MAX) 2125 return (EINVAL); 2126 2127 return (VMGETCAP(vm->cookie, vcpu, type, retval)); 2128 } 2129 2130 int 2131 vm_set_capability(struct vm *vm, int vcpu, int type, int val) 2132 { 2133 if (vcpu < 0 || vcpu >= VM_MAXCPU) 2134 return (EINVAL); 2135 2136 if (type < 0 || type >= VM_CAP_MAX) 2137 return (EINVAL); 2138 2139 return (VMSETCAP(vm->cookie, vcpu, type, val)); 2140 } 2141 2142 struct vlapic * 2143 vm_lapic(struct vm *vm, int cpu) 2144 { 2145 return (vm->vcpu[cpu].vlapic); 2146 } 2147 2148 struct vioapic * 2149 vm_ioapic(struct vm *vm) 2150 { 2151 2152 return (vm->vioapic); 2153 } 2154 2155 struct vhpet * 2156 vm_hpet(struct vm *vm) 2157 { 2158 2159 return (vm->vhpet); 2160 } 2161 2162 boolean_t 2163 vmm_is_pptdev(int bus, int slot, int func) 2164 { 2165 int found, i, n; 2166 int b, s, f; 2167 char *val, *cp, *cp2; 2168 2169 /* 2170 * XXX 2171 * The length of an environment variable is limited to 128 bytes which 2172 * puts an upper limit on the number of passthru devices that may be 2173 * specified using a single environment variable. 2174 * 2175 * Work around this by scanning multiple environment variable 2176 * names instead of a single one - yuck! 2177 */ 2178 const char *names[] = { "pptdevs", "pptdevs2", "pptdevs3", NULL }; 2179 2180 /* set pptdevs="1/2/3 4/5/6 7/8/9 10/11/12" */ 2181 found = 0; 2182 for (i = 0; names[i] != NULL && !found; i++) { 2183 cp = val = kern_getenv(names[i]); 2184 while (cp != NULL && *cp != '\0') { 2185 if ((cp2 = strchr(cp, ' ')) != NULL) 2186 *cp2 = '\0'; 2187 2188 n = sscanf(cp, "%d/%d/%d", &b, &s, &f); 2189 if (n == 3 && bus == b && slot == s && func == f) { 2190 found = 1; 2191 break; 2192 } 2193 2194 if (cp2 != NULL) 2195 *cp2++ = ' '; 2196 2197 cp = cp2; 2198 } 2199 freeenv(val); 2200 } 2201 return (found); 2202 } 2203 2204 void * 2205 vm_iommu_domain(struct vm *vm) 2206 { 2207 2208 return (vm->iommu); 2209 } 2210 2211 int 2212 vcpu_set_state(struct vm *vm, int vcpuid, enum vcpu_state newstate, 2213 bool from_idle) 2214 { 2215 int error; 2216 struct vcpu *vcpu; 2217 2218 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 2219 panic("vm_set_run_state: invalid vcpuid %d", vcpuid); 2220 2221 vcpu = &vm->vcpu[vcpuid]; 2222 2223 vcpu_lock(vcpu); 2224 error = vcpu_set_state_locked(vm, vcpuid, newstate, from_idle); 2225 vcpu_unlock(vcpu); 2226 2227 return (error); 2228 } 2229 2230 enum vcpu_state 2231 vcpu_get_state(struct vm *vm, int vcpuid, int *hostcpu) 2232 { 2233 struct vcpu *vcpu; 2234 enum vcpu_state state; 2235 2236 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 2237 panic("vm_get_run_state: invalid vcpuid %d", vcpuid); 2238 2239 vcpu = &vm->vcpu[vcpuid]; 2240 2241 vcpu_lock(vcpu); 2242 state = vcpu->state; 2243 if (hostcpu != NULL) 2244 *hostcpu = vcpu->hostcpu; 2245 vcpu_unlock(vcpu); 2246 2247 return (state); 2248 } 2249 2250 int 2251 vm_activate_cpu(struct vm *vm, int vcpuid) 2252 { 2253 2254 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 2255 return (EINVAL); 2256 2257 if (CPU_ISSET(vcpuid, &vm->active_cpus)) 2258 return (EBUSY); 2259 2260 VCPU_CTR0(vm, vcpuid, "activated"); 2261 CPU_SET_ATOMIC(vcpuid, &vm->active_cpus); 2262 return (0); 2263 } 2264 2265 cpuset_t 2266 vm_active_cpus(struct vm *vm) 2267 { 2268 2269 return (vm->active_cpus); 2270 } 2271 2272 cpuset_t 2273 vm_suspended_cpus(struct vm *vm) 2274 { 2275 2276 return (vm->suspended_cpus); 2277 } 2278 2279 void * 2280 vcpu_stats(struct vm *vm, int vcpuid) 2281 { 2282 2283 return (vm->vcpu[vcpuid].stats); 2284 } 2285 2286 int 2287 vm_get_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state *state) 2288 { 2289 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 2290 return (EINVAL); 2291 2292 *state = vm->vcpu[vcpuid].x2apic_state; 2293 2294 return (0); 2295 } 2296 2297 int 2298 vm_set_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state state) 2299 { 2300 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 2301 return (EINVAL); 2302 2303 if (state >= X2APIC_STATE_LAST) 2304 return (EINVAL); 2305 2306 vm->vcpu[vcpuid].x2apic_state = state; 2307 2308 vlapic_set_x2apic_state(vm, vcpuid, state); 2309 2310 return (0); 2311 } 2312 2313 /* 2314 * This function is called to ensure that a vcpu "sees" a pending event 2315 * as soon as possible: 2316 * - If the vcpu thread is sleeping then it is woken up. 2317 * - If the vcpu is running on a different host_cpu then an IPI will be directed 2318 * to the host_cpu to cause the vcpu to trap into the hypervisor. 2319 */ 2320 static void 2321 vcpu_notify_event_locked(struct vcpu *vcpu, bool lapic_intr) 2322 { 2323 int hostcpu; 2324 2325 hostcpu = vcpu->hostcpu; 2326 if (vcpu->state == VCPU_RUNNING) { 2327 KASSERT(hostcpu != NOCPU, ("vcpu running on invalid hostcpu")); 2328 if (hostcpu != curcpu) { 2329 if (lapic_intr) { 2330 vlapic_post_intr(vcpu->vlapic, hostcpu, 2331 vmm_ipinum); 2332 } else { 2333 ipi_cpu(hostcpu, vmm_ipinum); 2334 } 2335 } else { 2336 /* 2337 * If the 'vcpu' is running on 'curcpu' then it must 2338 * be sending a notification to itself (e.g. SELF_IPI). 2339 * The pending event will be picked up when the vcpu 2340 * transitions back to guest context. 2341 */ 2342 } 2343 } else { 2344 KASSERT(hostcpu == NOCPU, ("vcpu state %d not consistent " 2345 "with hostcpu %d", vcpu->state, hostcpu)); 2346 if (vcpu->state == VCPU_SLEEPING) 2347 wakeup_one(vcpu); 2348 } 2349 } 2350 2351 void 2352 vcpu_notify_event(struct vm *vm, int vcpuid, bool lapic_intr) 2353 { 2354 struct vcpu *vcpu = &vm->vcpu[vcpuid]; 2355 2356 vcpu_lock(vcpu); 2357 vcpu_notify_event_locked(vcpu, lapic_intr); 2358 vcpu_unlock(vcpu); 2359 } 2360 2361 struct vmspace * 2362 vm_get_vmspace(struct vm *vm) 2363 { 2364 2365 return (vm->vmspace); 2366 } 2367 2368 int 2369 vm_apicid2vcpuid(struct vm *vm, int apicid) 2370 { 2371 /* 2372 * XXX apic id is assumed to be numerically identical to vcpu id 2373 */ 2374 return (apicid); 2375 } 2376 2377 void 2378 vm_smp_rendezvous(struct vm *vm, int vcpuid, cpuset_t dest, 2379 vm_rendezvous_func_t func, void *arg) 2380 { 2381 int i; 2382 2383 /* 2384 * Enforce that this function is called without any locks 2385 */ 2386 WITNESS_WARN(WARN_PANIC, NULL, "vm_smp_rendezvous"); 2387 KASSERT(vcpuid == -1 || (vcpuid >= 0 && vcpuid < VM_MAXCPU), 2388 ("vm_smp_rendezvous: invalid vcpuid %d", vcpuid)); 2389 2390 restart: 2391 mtx_lock(&vm->rendezvous_mtx); 2392 if (vm->rendezvous_func != NULL) { 2393 /* 2394 * If a rendezvous is already in progress then we need to 2395 * call the rendezvous handler in case this 'vcpuid' is one 2396 * of the targets of the rendezvous. 2397 */ 2398 RENDEZVOUS_CTR0(vm, vcpuid, "Rendezvous already in progress"); 2399 mtx_unlock(&vm->rendezvous_mtx); 2400 vm_handle_rendezvous(vm, vcpuid); 2401 goto restart; 2402 } 2403 KASSERT(vm->rendezvous_func == NULL, ("vm_smp_rendezvous: previous " 2404 "rendezvous is still in progress")); 2405 2406 RENDEZVOUS_CTR0(vm, vcpuid, "Initiating rendezvous"); 2407 vm->rendezvous_req_cpus = dest; 2408 CPU_ZERO(&vm->rendezvous_done_cpus); 2409 vm->rendezvous_arg = arg; 2410 vm_set_rendezvous_func(vm, func); 2411 mtx_unlock(&vm->rendezvous_mtx); 2412 2413 /* 2414 * Wake up any sleeping vcpus and trigger a VM-exit in any running 2415 * vcpus so they handle the rendezvous as soon as possible. 2416 */ 2417 for (i = 0; i < VM_MAXCPU; i++) { 2418 if (CPU_ISSET(i, &dest)) 2419 vcpu_notify_event(vm, i, false); 2420 } 2421 2422 vm_handle_rendezvous(vm, vcpuid); 2423 } 2424 2425 struct vatpic * 2426 vm_atpic(struct vm *vm) 2427 { 2428 return (vm->vatpic); 2429 } 2430 2431 struct vatpit * 2432 vm_atpit(struct vm *vm) 2433 { 2434 return (vm->vatpit); 2435 } 2436 2437 struct vpmtmr * 2438 vm_pmtmr(struct vm *vm) 2439 { 2440 2441 return (vm->vpmtmr); 2442 } 2443 2444 struct vrtc * 2445 vm_rtc(struct vm *vm) 2446 { 2447 2448 return (vm->vrtc); 2449 } 2450 2451 enum vm_reg_name 2452 vm_segment_name(int seg) 2453 { 2454 static enum vm_reg_name seg_names[] = { 2455 VM_REG_GUEST_ES, 2456 VM_REG_GUEST_CS, 2457 VM_REG_GUEST_SS, 2458 VM_REG_GUEST_DS, 2459 VM_REG_GUEST_FS, 2460 VM_REG_GUEST_GS 2461 }; 2462 2463 KASSERT(seg >= 0 && seg < nitems(seg_names), 2464 ("%s: invalid segment encoding %d", __func__, seg)); 2465 return (seg_names[seg]); 2466 } 2467 2468 void 2469 vm_copy_teardown(struct vm *vm, int vcpuid, struct vm_copyinfo *copyinfo, 2470 int num_copyinfo) 2471 { 2472 int idx; 2473 2474 for (idx = 0; idx < num_copyinfo; idx++) { 2475 if (copyinfo[idx].cookie != NULL) 2476 vm_gpa_release(copyinfo[idx].cookie); 2477 } 2478 bzero(copyinfo, num_copyinfo * sizeof(struct vm_copyinfo)); 2479 } 2480 2481 int 2482 vm_copy_setup(struct vm *vm, int vcpuid, struct vm_guest_paging *paging, 2483 uint64_t gla, size_t len, int prot, struct vm_copyinfo *copyinfo, 2484 int num_copyinfo, int *fault) 2485 { 2486 int error, idx, nused; 2487 size_t n, off, remaining; 2488 void *hva, *cookie; 2489 uint64_t gpa; 2490 2491 bzero(copyinfo, sizeof(struct vm_copyinfo) * num_copyinfo); 2492 2493 nused = 0; 2494 remaining = len; 2495 while (remaining > 0) { 2496 KASSERT(nused < num_copyinfo, ("insufficient vm_copyinfo")); 2497 error = vm_gla2gpa(vm, vcpuid, paging, gla, prot, &gpa, fault); 2498 if (error || *fault) 2499 return (error); 2500 off = gpa & PAGE_MASK; 2501 n = min(remaining, PAGE_SIZE - off); 2502 copyinfo[nused].gpa = gpa; 2503 copyinfo[nused].len = n; 2504 remaining -= n; 2505 gla += n; 2506 nused++; 2507 } 2508 2509 for (idx = 0; idx < nused; idx++) { 2510 hva = vm_gpa_hold(vm, vcpuid, copyinfo[idx].gpa, 2511 copyinfo[idx].len, prot, &cookie); 2512 if (hva == NULL) 2513 break; 2514 copyinfo[idx].hva = hva; 2515 copyinfo[idx].cookie = cookie; 2516 } 2517 2518 if (idx != nused) { 2519 vm_copy_teardown(vm, vcpuid, copyinfo, num_copyinfo); 2520 return (EFAULT); 2521 } else { 2522 *fault = 0; 2523 return (0); 2524 } 2525 } 2526 2527 void 2528 vm_copyin(struct vm *vm, int vcpuid, struct vm_copyinfo *copyinfo, void *kaddr, 2529 size_t len) 2530 { 2531 char *dst; 2532 int idx; 2533 2534 dst = kaddr; 2535 idx = 0; 2536 while (len > 0) { 2537 bcopy(copyinfo[idx].hva, dst, copyinfo[idx].len); 2538 len -= copyinfo[idx].len; 2539 dst += copyinfo[idx].len; 2540 idx++; 2541 } 2542 } 2543 2544 void 2545 vm_copyout(struct vm *vm, int vcpuid, const void *kaddr, 2546 struct vm_copyinfo *copyinfo, size_t len) 2547 { 2548 const char *src; 2549 int idx; 2550 2551 src = kaddr; 2552 idx = 0; 2553 while (len > 0) { 2554 bcopy(src, copyinfo[idx].hva, copyinfo[idx].len); 2555 len -= copyinfo[idx].len; 2556 src += copyinfo[idx].len; 2557 idx++; 2558 } 2559 } 2560 2561 /* 2562 * Return the amount of in-use and wired memory for the VM. Since 2563 * these are global stats, only return the values with for vCPU 0 2564 */ 2565 VMM_STAT_DECLARE(VMM_MEM_RESIDENT); 2566 VMM_STAT_DECLARE(VMM_MEM_WIRED); 2567 2568 static void 2569 vm_get_rescnt(struct vm *vm, int vcpu, struct vmm_stat_type *stat) 2570 { 2571 2572 if (vcpu == 0) { 2573 vmm_stat_set(vm, vcpu, VMM_MEM_RESIDENT, 2574 PAGE_SIZE * vmspace_resident_count(vm->vmspace)); 2575 } 2576 } 2577 2578 static void 2579 vm_get_wiredcnt(struct vm *vm, int vcpu, struct vmm_stat_type *stat) 2580 { 2581 2582 if (vcpu == 0) { 2583 vmm_stat_set(vm, vcpu, VMM_MEM_WIRED, 2584 PAGE_SIZE * pmap_wired_count(vmspace_pmap(vm->vmspace))); 2585 } 2586 } 2587 2588 VMM_STAT_FUNC(VMM_MEM_RESIDENT, "Resident memory", vm_get_rescnt); 2589 VMM_STAT_FUNC(VMM_MEM_WIRED, "Wired memory", vm_get_wiredcnt); 2590