1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 3 * 4 * Copyright (c) 2011 NetApp, Inc. 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 16 * THIS SOFTWARE IS PROVIDED BY NETAPP, INC ``AS IS'' AND 17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 19 * ARE DISCLAIMED. IN NO EVENT SHALL NETAPP, INC OR CONTRIBUTORS BE LIABLE 20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 26 * SUCH DAMAGE. 27 * 28 * $FreeBSD$ 29 */ 30 31 #include <sys/cdefs.h> 32 __FBSDID("$FreeBSD$"); 33 34 #include <sys/param.h> 35 #include <sys/systm.h> 36 #include <sys/kernel.h> 37 #include <sys/module.h> 38 #include <sys/sysctl.h> 39 #include <sys/malloc.h> 40 #include <sys/pcpu.h> 41 #include <sys/lock.h> 42 #include <sys/mutex.h> 43 #include <sys/proc.h> 44 #include <sys/rwlock.h> 45 #include <sys/sched.h> 46 #include <sys/smp.h> 47 #include <sys/systm.h> 48 49 #include <vm/vm.h> 50 #include <vm/vm_object.h> 51 #include <vm/vm_page.h> 52 #include <vm/pmap.h> 53 #include <vm/vm_map.h> 54 #include <vm/vm_extern.h> 55 #include <vm/vm_param.h> 56 57 #include <machine/cpu.h> 58 #include <machine/pcb.h> 59 #include <machine/smp.h> 60 #include <machine/md_var.h> 61 #include <x86/psl.h> 62 #include <x86/apicreg.h> 63 64 #include <machine/vmm.h> 65 #include <machine/vmm_dev.h> 66 #include <machine/vmm_instruction_emul.h> 67 68 #include "vmm_ioport.h" 69 #include "vmm_ktr.h" 70 #include "vmm_host.h" 71 #include "vmm_mem.h" 72 #include "vmm_util.h" 73 #include "vatpic.h" 74 #include "vatpit.h" 75 #include "vhpet.h" 76 #include "vioapic.h" 77 #include "vlapic.h" 78 #include "vpmtmr.h" 79 #include "vrtc.h" 80 #include "vmm_stat.h" 81 #include "vmm_lapic.h" 82 83 #include "io/ppt.h" 84 #include "io/iommu.h" 85 86 struct vlapic; 87 88 /* 89 * Initialization: 90 * (a) allocated when vcpu is created 91 * (i) initialized when vcpu is created and when it is reinitialized 92 * (o) initialized the first time the vcpu is created 93 * (x) initialized before use 94 */ 95 struct vcpu { 96 struct mtx mtx; /* (o) protects 'state' and 'hostcpu' */ 97 enum vcpu_state state; /* (o) vcpu state */ 98 int hostcpu; /* (o) vcpu's host cpu */ 99 int reqidle; /* (i) request vcpu to idle */ 100 struct vlapic *vlapic; /* (i) APIC device model */ 101 enum x2apic_state x2apic_state; /* (i) APIC mode */ 102 uint64_t exitintinfo; /* (i) events pending at VM exit */ 103 int nmi_pending; /* (i) NMI pending */ 104 int extint_pending; /* (i) INTR pending */ 105 int exception_pending; /* (i) exception pending */ 106 int exc_vector; /* (x) exception collateral */ 107 int exc_errcode_valid; 108 uint32_t exc_errcode; 109 struct savefpu *guestfpu; /* (a,i) guest fpu state */ 110 uint64_t guest_xcr0; /* (i) guest %xcr0 register */ 111 void *stats; /* (a,i) statistics */ 112 struct vm_exit exitinfo; /* (x) exit reason and collateral */ 113 uint64_t nextrip; /* (x) next instruction to execute */ 114 }; 115 116 #define vcpu_lock_initialized(v) mtx_initialized(&((v)->mtx)) 117 #define vcpu_lock_init(v) mtx_init(&((v)->mtx), "vcpu lock", 0, MTX_SPIN) 118 #define vcpu_lock(v) mtx_lock_spin(&((v)->mtx)) 119 #define vcpu_unlock(v) mtx_unlock_spin(&((v)->mtx)) 120 #define vcpu_assert_locked(v) mtx_assert(&((v)->mtx), MA_OWNED) 121 122 struct mem_seg { 123 size_t len; 124 bool sysmem; 125 struct vm_object *object; 126 }; 127 #define VM_MAX_MEMSEGS 3 128 129 struct mem_map { 130 vm_paddr_t gpa; 131 size_t len; 132 vm_ooffset_t segoff; 133 int segid; 134 int prot; 135 int flags; 136 }; 137 #define VM_MAX_MEMMAPS 4 138 139 /* 140 * Initialization: 141 * (o) initialized the first time the VM is created 142 * (i) initialized when VM is created and when it is reinitialized 143 * (x) initialized before use 144 */ 145 struct vm { 146 void *cookie; /* (i) cpu-specific data */ 147 void *iommu; /* (x) iommu-specific data */ 148 struct vhpet *vhpet; /* (i) virtual HPET */ 149 struct vioapic *vioapic; /* (i) virtual ioapic */ 150 struct vatpic *vatpic; /* (i) virtual atpic */ 151 struct vatpit *vatpit; /* (i) virtual atpit */ 152 struct vpmtmr *vpmtmr; /* (i) virtual ACPI PM timer */ 153 struct vrtc *vrtc; /* (o) virtual RTC */ 154 volatile cpuset_t active_cpus; /* (i) active vcpus */ 155 int suspend; /* (i) stop VM execution */ 156 volatile cpuset_t suspended_cpus; /* (i) suspended vcpus */ 157 volatile cpuset_t halted_cpus; /* (x) cpus in a hard halt */ 158 cpuset_t rendezvous_req_cpus; /* (x) rendezvous requested */ 159 cpuset_t rendezvous_done_cpus; /* (x) rendezvous finished */ 160 void *rendezvous_arg; /* (x) rendezvous func/arg */ 161 vm_rendezvous_func_t rendezvous_func; 162 struct mtx rendezvous_mtx; /* (o) rendezvous lock */ 163 struct mem_map mem_maps[VM_MAX_MEMMAPS]; /* (i) guest address space */ 164 struct mem_seg mem_segs[VM_MAX_MEMSEGS]; /* (o) guest memory regions */ 165 struct vmspace *vmspace; /* (o) guest's address space */ 166 char name[VM_MAX_NAMELEN]; /* (o) virtual machine name */ 167 struct vcpu vcpu[VM_MAXCPU]; /* (i) guest vcpus */ 168 }; 169 170 static int vmm_initialized; 171 172 static struct vmm_ops *ops; 173 #define VMM_INIT(num) (ops != NULL ? (*ops->init)(num) : 0) 174 #define VMM_CLEANUP() (ops != NULL ? (*ops->cleanup)() : 0) 175 #define VMM_RESUME() (ops != NULL ? (*ops->resume)() : 0) 176 177 #define VMINIT(vm, pmap) (ops != NULL ? (*ops->vminit)(vm, pmap): NULL) 178 #define VMRUN(vmi, vcpu, rip, pmap, evinfo) \ 179 (ops != NULL ? (*ops->vmrun)(vmi, vcpu, rip, pmap, evinfo) : ENXIO) 180 #define VMCLEANUP(vmi) (ops != NULL ? (*ops->vmcleanup)(vmi) : NULL) 181 #define VMSPACE_ALLOC(min, max) \ 182 (ops != NULL ? (*ops->vmspace_alloc)(min, max) : NULL) 183 #define VMSPACE_FREE(vmspace) \ 184 (ops != NULL ? (*ops->vmspace_free)(vmspace) : ENXIO) 185 #define VMGETREG(vmi, vcpu, num, retval) \ 186 (ops != NULL ? (*ops->vmgetreg)(vmi, vcpu, num, retval) : ENXIO) 187 #define VMSETREG(vmi, vcpu, num, val) \ 188 (ops != NULL ? (*ops->vmsetreg)(vmi, vcpu, num, val) : ENXIO) 189 #define VMGETDESC(vmi, vcpu, num, desc) \ 190 (ops != NULL ? (*ops->vmgetdesc)(vmi, vcpu, num, desc) : ENXIO) 191 #define VMSETDESC(vmi, vcpu, num, desc) \ 192 (ops != NULL ? (*ops->vmsetdesc)(vmi, vcpu, num, desc) : ENXIO) 193 #define VMGETCAP(vmi, vcpu, num, retval) \ 194 (ops != NULL ? (*ops->vmgetcap)(vmi, vcpu, num, retval) : ENXIO) 195 #define VMSETCAP(vmi, vcpu, num, val) \ 196 (ops != NULL ? (*ops->vmsetcap)(vmi, vcpu, num, val) : ENXIO) 197 #define VLAPIC_INIT(vmi, vcpu) \ 198 (ops != NULL ? (*ops->vlapic_init)(vmi, vcpu) : NULL) 199 #define VLAPIC_CLEANUP(vmi, vlapic) \ 200 (ops != NULL ? (*ops->vlapic_cleanup)(vmi, vlapic) : NULL) 201 202 #define fpu_start_emulating() load_cr0(rcr0() | CR0_TS) 203 #define fpu_stop_emulating() clts() 204 205 static MALLOC_DEFINE(M_VM, "vm", "vm"); 206 207 /* statistics */ 208 static VMM_STAT(VCPU_TOTAL_RUNTIME, "vcpu total runtime"); 209 210 SYSCTL_NODE(_hw, OID_AUTO, vmm, CTLFLAG_RW, NULL, NULL); 211 212 /* 213 * Halt the guest if all vcpus are executing a HLT instruction with 214 * interrupts disabled. 215 */ 216 static int halt_detection_enabled = 1; 217 SYSCTL_INT(_hw_vmm, OID_AUTO, halt_detection, CTLFLAG_RDTUN, 218 &halt_detection_enabled, 0, 219 "Halt VM if all vcpus execute HLT with interrupts disabled"); 220 221 static int vmm_ipinum; 222 SYSCTL_INT(_hw_vmm, OID_AUTO, ipinum, CTLFLAG_RD, &vmm_ipinum, 0, 223 "IPI vector used for vcpu notifications"); 224 225 static int trace_guest_exceptions; 226 SYSCTL_INT(_hw_vmm, OID_AUTO, trace_guest_exceptions, CTLFLAG_RDTUN, 227 &trace_guest_exceptions, 0, 228 "Trap into hypervisor on all guest exceptions and reflect them back"); 229 230 static void vm_free_memmap(struct vm *vm, int ident); 231 static bool sysmem_mapping(struct vm *vm, struct mem_map *mm); 232 static void vcpu_notify_event_locked(struct vcpu *vcpu, bool lapic_intr); 233 234 #ifdef KTR 235 static const char * 236 vcpu_state2str(enum vcpu_state state) 237 { 238 239 switch (state) { 240 case VCPU_IDLE: 241 return ("idle"); 242 case VCPU_FROZEN: 243 return ("frozen"); 244 case VCPU_RUNNING: 245 return ("running"); 246 case VCPU_SLEEPING: 247 return ("sleeping"); 248 default: 249 return ("unknown"); 250 } 251 } 252 #endif 253 254 static void 255 vcpu_cleanup(struct vm *vm, int i, bool destroy) 256 { 257 struct vcpu *vcpu = &vm->vcpu[i]; 258 259 VLAPIC_CLEANUP(vm->cookie, vcpu->vlapic); 260 if (destroy) { 261 vmm_stat_free(vcpu->stats); 262 fpu_save_area_free(vcpu->guestfpu); 263 } 264 } 265 266 static void 267 vcpu_init(struct vm *vm, int vcpu_id, bool create) 268 { 269 struct vcpu *vcpu; 270 271 KASSERT(vcpu_id >= 0 && vcpu_id < VM_MAXCPU, 272 ("vcpu_init: invalid vcpu %d", vcpu_id)); 273 274 vcpu = &vm->vcpu[vcpu_id]; 275 276 if (create) { 277 KASSERT(!vcpu_lock_initialized(vcpu), ("vcpu %d already " 278 "initialized", vcpu_id)); 279 vcpu_lock_init(vcpu); 280 vcpu->state = VCPU_IDLE; 281 vcpu->hostcpu = NOCPU; 282 vcpu->guestfpu = fpu_save_area_alloc(); 283 vcpu->stats = vmm_stat_alloc(); 284 } 285 286 vcpu->vlapic = VLAPIC_INIT(vm->cookie, vcpu_id); 287 vm_set_x2apic_state(vm, vcpu_id, X2APIC_DISABLED); 288 vcpu->reqidle = 0; 289 vcpu->exitintinfo = 0; 290 vcpu->nmi_pending = 0; 291 vcpu->extint_pending = 0; 292 vcpu->exception_pending = 0; 293 vcpu->guest_xcr0 = XFEATURE_ENABLED_X87; 294 fpu_save_area_reset(vcpu->guestfpu); 295 vmm_stat_init(vcpu->stats); 296 } 297 298 int 299 vcpu_trace_exceptions(struct vm *vm, int vcpuid) 300 { 301 302 return (trace_guest_exceptions); 303 } 304 305 struct vm_exit * 306 vm_exitinfo(struct vm *vm, int cpuid) 307 { 308 struct vcpu *vcpu; 309 310 if (cpuid < 0 || cpuid >= VM_MAXCPU) 311 panic("vm_exitinfo: invalid cpuid %d", cpuid); 312 313 vcpu = &vm->vcpu[cpuid]; 314 315 return (&vcpu->exitinfo); 316 } 317 318 static void 319 vmm_resume(void) 320 { 321 VMM_RESUME(); 322 } 323 324 static int 325 vmm_init(void) 326 { 327 int error; 328 329 vmm_host_state_init(); 330 331 vmm_ipinum = lapic_ipi_alloc(pti ? &IDTVEC(justreturn1_pti) : 332 &IDTVEC(justreturn)); 333 if (vmm_ipinum < 0) 334 vmm_ipinum = IPI_AST; 335 336 error = vmm_mem_init(); 337 if (error) 338 return (error); 339 340 if (vmm_is_intel()) 341 ops = &vmm_ops_intel; 342 else if (vmm_is_amd()) 343 ops = &vmm_ops_amd; 344 else 345 return (ENXIO); 346 347 vmm_resume_p = vmm_resume; 348 349 return (VMM_INIT(vmm_ipinum)); 350 } 351 352 static int 353 vmm_handler(module_t mod, int what, void *arg) 354 { 355 int error; 356 357 switch (what) { 358 case MOD_LOAD: 359 vmmdev_init(); 360 error = vmm_init(); 361 if (error == 0) 362 vmm_initialized = 1; 363 break; 364 case MOD_UNLOAD: 365 error = vmmdev_cleanup(); 366 if (error == 0) { 367 vmm_resume_p = NULL; 368 iommu_cleanup(); 369 if (vmm_ipinum != IPI_AST) 370 lapic_ipi_free(vmm_ipinum); 371 error = VMM_CLEANUP(); 372 /* 373 * Something bad happened - prevent new 374 * VMs from being created 375 */ 376 if (error) 377 vmm_initialized = 0; 378 } 379 break; 380 default: 381 error = 0; 382 break; 383 } 384 return (error); 385 } 386 387 static moduledata_t vmm_kmod = { 388 "vmm", 389 vmm_handler, 390 NULL 391 }; 392 393 /* 394 * vmm initialization has the following dependencies: 395 * 396 * - VT-x initialization requires smp_rendezvous() and therefore must happen 397 * after SMP is fully functional (after SI_SUB_SMP). 398 */ 399 DECLARE_MODULE(vmm, vmm_kmod, SI_SUB_SMP + 1, SI_ORDER_ANY); 400 MODULE_VERSION(vmm, 1); 401 402 static void 403 vm_init(struct vm *vm, bool create) 404 { 405 int i; 406 407 vm->cookie = VMINIT(vm, vmspace_pmap(vm->vmspace)); 408 vm->iommu = NULL; 409 vm->vioapic = vioapic_init(vm); 410 vm->vhpet = vhpet_init(vm); 411 vm->vatpic = vatpic_init(vm); 412 vm->vatpit = vatpit_init(vm); 413 vm->vpmtmr = vpmtmr_init(vm); 414 if (create) 415 vm->vrtc = vrtc_init(vm); 416 417 CPU_ZERO(&vm->active_cpus); 418 419 vm->suspend = 0; 420 CPU_ZERO(&vm->suspended_cpus); 421 422 for (i = 0; i < VM_MAXCPU; i++) 423 vcpu_init(vm, i, create); 424 } 425 426 int 427 vm_create(const char *name, struct vm **retvm) 428 { 429 struct vm *vm; 430 struct vmspace *vmspace; 431 432 /* 433 * If vmm.ko could not be successfully initialized then don't attempt 434 * to create the virtual machine. 435 */ 436 if (!vmm_initialized) 437 return (ENXIO); 438 439 if (name == NULL || strlen(name) >= VM_MAX_NAMELEN) 440 return (EINVAL); 441 442 vmspace = VMSPACE_ALLOC(0, VM_MAXUSER_ADDRESS); 443 if (vmspace == NULL) 444 return (ENOMEM); 445 446 vm = malloc(sizeof(struct vm), M_VM, M_WAITOK | M_ZERO); 447 strcpy(vm->name, name); 448 vm->vmspace = vmspace; 449 mtx_init(&vm->rendezvous_mtx, "vm rendezvous lock", 0, MTX_DEF); 450 451 vm_init(vm, true); 452 453 *retvm = vm; 454 return (0); 455 } 456 457 static void 458 vm_cleanup(struct vm *vm, bool destroy) 459 { 460 struct mem_map *mm; 461 int i; 462 463 ppt_unassign_all(vm); 464 465 if (vm->iommu != NULL) 466 iommu_destroy_domain(vm->iommu); 467 468 if (destroy) 469 vrtc_cleanup(vm->vrtc); 470 else 471 vrtc_reset(vm->vrtc); 472 vpmtmr_cleanup(vm->vpmtmr); 473 vatpit_cleanup(vm->vatpit); 474 vhpet_cleanup(vm->vhpet); 475 vatpic_cleanup(vm->vatpic); 476 vioapic_cleanup(vm->vioapic); 477 478 for (i = 0; i < VM_MAXCPU; i++) 479 vcpu_cleanup(vm, i, destroy); 480 481 VMCLEANUP(vm->cookie); 482 483 /* 484 * System memory is removed from the guest address space only when 485 * the VM is destroyed. This is because the mapping remains the same 486 * across VM reset. 487 * 488 * Device memory can be relocated by the guest (e.g. using PCI BARs) 489 * so those mappings are removed on a VM reset. 490 */ 491 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 492 mm = &vm->mem_maps[i]; 493 if (destroy || !sysmem_mapping(vm, mm)) 494 vm_free_memmap(vm, i); 495 } 496 497 if (destroy) { 498 for (i = 0; i < VM_MAX_MEMSEGS; i++) 499 vm_free_memseg(vm, i); 500 501 VMSPACE_FREE(vm->vmspace); 502 vm->vmspace = NULL; 503 } 504 } 505 506 void 507 vm_destroy(struct vm *vm) 508 { 509 vm_cleanup(vm, true); 510 free(vm, M_VM); 511 } 512 513 int 514 vm_reinit(struct vm *vm) 515 { 516 int error; 517 518 /* 519 * A virtual machine can be reset only if all vcpus are suspended. 520 */ 521 if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) { 522 vm_cleanup(vm, false); 523 vm_init(vm, false); 524 error = 0; 525 } else { 526 error = EBUSY; 527 } 528 529 return (error); 530 } 531 532 const char * 533 vm_name(struct vm *vm) 534 { 535 return (vm->name); 536 } 537 538 int 539 vm_map_mmio(struct vm *vm, vm_paddr_t gpa, size_t len, vm_paddr_t hpa) 540 { 541 vm_object_t obj; 542 543 if ((obj = vmm_mmio_alloc(vm->vmspace, gpa, len, hpa)) == NULL) 544 return (ENOMEM); 545 else 546 return (0); 547 } 548 549 int 550 vm_unmap_mmio(struct vm *vm, vm_paddr_t gpa, size_t len) 551 { 552 553 vmm_mmio_free(vm->vmspace, gpa, len); 554 return (0); 555 } 556 557 /* 558 * Return 'true' if 'gpa' is allocated in the guest address space. 559 * 560 * This function is called in the context of a running vcpu which acts as 561 * an implicit lock on 'vm->mem_maps[]'. 562 */ 563 bool 564 vm_mem_allocated(struct vm *vm, int vcpuid, vm_paddr_t gpa) 565 { 566 struct mem_map *mm; 567 int i; 568 569 #ifdef INVARIANTS 570 int hostcpu, state; 571 state = vcpu_get_state(vm, vcpuid, &hostcpu); 572 KASSERT(state == VCPU_RUNNING && hostcpu == curcpu, 573 ("%s: invalid vcpu state %d/%d", __func__, state, hostcpu)); 574 #endif 575 576 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 577 mm = &vm->mem_maps[i]; 578 if (mm->len != 0 && gpa >= mm->gpa && gpa < mm->gpa + mm->len) 579 return (true); /* 'gpa' is sysmem or devmem */ 580 } 581 582 if (ppt_is_mmio(vm, gpa)) 583 return (true); /* 'gpa' is pci passthru mmio */ 584 585 return (false); 586 } 587 588 int 589 vm_alloc_memseg(struct vm *vm, int ident, size_t len, bool sysmem) 590 { 591 struct mem_seg *seg; 592 vm_object_t obj; 593 594 if (ident < 0 || ident >= VM_MAX_MEMSEGS) 595 return (EINVAL); 596 597 if (len == 0 || (len & PAGE_MASK)) 598 return (EINVAL); 599 600 seg = &vm->mem_segs[ident]; 601 if (seg->object != NULL) { 602 if (seg->len == len && seg->sysmem == sysmem) 603 return (EEXIST); 604 else 605 return (EINVAL); 606 } 607 608 obj = vm_object_allocate(OBJT_DEFAULT, len >> PAGE_SHIFT); 609 if (obj == NULL) 610 return (ENOMEM); 611 612 seg->len = len; 613 seg->object = obj; 614 seg->sysmem = sysmem; 615 return (0); 616 } 617 618 int 619 vm_get_memseg(struct vm *vm, int ident, size_t *len, bool *sysmem, 620 vm_object_t *objptr) 621 { 622 struct mem_seg *seg; 623 624 if (ident < 0 || ident >= VM_MAX_MEMSEGS) 625 return (EINVAL); 626 627 seg = &vm->mem_segs[ident]; 628 if (len) 629 *len = seg->len; 630 if (sysmem) 631 *sysmem = seg->sysmem; 632 if (objptr) 633 *objptr = seg->object; 634 return (0); 635 } 636 637 void 638 vm_free_memseg(struct vm *vm, int ident) 639 { 640 struct mem_seg *seg; 641 642 KASSERT(ident >= 0 && ident < VM_MAX_MEMSEGS, 643 ("%s: invalid memseg ident %d", __func__, ident)); 644 645 seg = &vm->mem_segs[ident]; 646 if (seg->object != NULL) { 647 vm_object_deallocate(seg->object); 648 bzero(seg, sizeof(struct mem_seg)); 649 } 650 } 651 652 int 653 vm_mmap_memseg(struct vm *vm, vm_paddr_t gpa, int segid, vm_ooffset_t first, 654 size_t len, int prot, int flags) 655 { 656 struct mem_seg *seg; 657 struct mem_map *m, *map; 658 vm_ooffset_t last; 659 int i, error; 660 661 if (prot == 0 || (prot & ~(VM_PROT_ALL)) != 0) 662 return (EINVAL); 663 664 if (flags & ~VM_MEMMAP_F_WIRED) 665 return (EINVAL); 666 667 if (segid < 0 || segid >= VM_MAX_MEMSEGS) 668 return (EINVAL); 669 670 seg = &vm->mem_segs[segid]; 671 if (seg->object == NULL) 672 return (EINVAL); 673 674 last = first + len; 675 if (first < 0 || first >= last || last > seg->len) 676 return (EINVAL); 677 678 if ((gpa | first | last) & PAGE_MASK) 679 return (EINVAL); 680 681 map = NULL; 682 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 683 m = &vm->mem_maps[i]; 684 if (m->len == 0) { 685 map = m; 686 break; 687 } 688 } 689 690 if (map == NULL) 691 return (ENOSPC); 692 693 error = vm_map_find(&vm->vmspace->vm_map, seg->object, first, &gpa, 694 len, 0, VMFS_NO_SPACE, prot, prot, 0); 695 if (error != KERN_SUCCESS) 696 return (EFAULT); 697 698 vm_object_reference(seg->object); 699 700 if (flags & VM_MEMMAP_F_WIRED) { 701 error = vm_map_wire(&vm->vmspace->vm_map, gpa, gpa + len, 702 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES); 703 if (error != KERN_SUCCESS) { 704 vm_map_remove(&vm->vmspace->vm_map, gpa, gpa + len); 705 return (EFAULT); 706 } 707 } 708 709 map->gpa = gpa; 710 map->len = len; 711 map->segoff = first; 712 map->segid = segid; 713 map->prot = prot; 714 map->flags = flags; 715 return (0); 716 } 717 718 int 719 vm_mmap_getnext(struct vm *vm, vm_paddr_t *gpa, int *segid, 720 vm_ooffset_t *segoff, size_t *len, int *prot, int *flags) 721 { 722 struct mem_map *mm, *mmnext; 723 int i; 724 725 mmnext = NULL; 726 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 727 mm = &vm->mem_maps[i]; 728 if (mm->len == 0 || mm->gpa < *gpa) 729 continue; 730 if (mmnext == NULL || mm->gpa < mmnext->gpa) 731 mmnext = mm; 732 } 733 734 if (mmnext != NULL) { 735 *gpa = mmnext->gpa; 736 if (segid) 737 *segid = mmnext->segid; 738 if (segoff) 739 *segoff = mmnext->segoff; 740 if (len) 741 *len = mmnext->len; 742 if (prot) 743 *prot = mmnext->prot; 744 if (flags) 745 *flags = mmnext->flags; 746 return (0); 747 } else { 748 return (ENOENT); 749 } 750 } 751 752 static void 753 vm_free_memmap(struct vm *vm, int ident) 754 { 755 struct mem_map *mm; 756 int error; 757 758 mm = &vm->mem_maps[ident]; 759 if (mm->len) { 760 error = vm_map_remove(&vm->vmspace->vm_map, mm->gpa, 761 mm->gpa + mm->len); 762 KASSERT(error == KERN_SUCCESS, ("%s: vm_map_remove error %d", 763 __func__, error)); 764 bzero(mm, sizeof(struct mem_map)); 765 } 766 } 767 768 static __inline bool 769 sysmem_mapping(struct vm *vm, struct mem_map *mm) 770 { 771 772 if (mm->len != 0 && vm->mem_segs[mm->segid].sysmem) 773 return (true); 774 else 775 return (false); 776 } 777 778 static vm_paddr_t 779 sysmem_maxaddr(struct vm *vm) 780 { 781 struct mem_map *mm; 782 vm_paddr_t maxaddr; 783 int i; 784 785 maxaddr = 0; 786 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 787 mm = &vm->mem_maps[i]; 788 if (sysmem_mapping(vm, mm)) { 789 if (maxaddr < mm->gpa + mm->len) 790 maxaddr = mm->gpa + mm->len; 791 } 792 } 793 return (maxaddr); 794 } 795 796 static void 797 vm_iommu_modify(struct vm *vm, boolean_t map) 798 { 799 int i, sz; 800 vm_paddr_t gpa, hpa; 801 struct mem_map *mm; 802 void *vp, *cookie, *host_domain; 803 804 sz = PAGE_SIZE; 805 host_domain = iommu_host_domain(); 806 807 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 808 mm = &vm->mem_maps[i]; 809 if (!sysmem_mapping(vm, mm)) 810 continue; 811 812 if (map) { 813 KASSERT((mm->flags & VM_MEMMAP_F_IOMMU) == 0, 814 ("iommu map found invalid memmap %#lx/%#lx/%#x", 815 mm->gpa, mm->len, mm->flags)); 816 if ((mm->flags & VM_MEMMAP_F_WIRED) == 0) 817 continue; 818 mm->flags |= VM_MEMMAP_F_IOMMU; 819 } else { 820 if ((mm->flags & VM_MEMMAP_F_IOMMU) == 0) 821 continue; 822 mm->flags &= ~VM_MEMMAP_F_IOMMU; 823 KASSERT((mm->flags & VM_MEMMAP_F_WIRED) != 0, 824 ("iommu unmap found invalid memmap %#lx/%#lx/%#x", 825 mm->gpa, mm->len, mm->flags)); 826 } 827 828 gpa = mm->gpa; 829 while (gpa < mm->gpa + mm->len) { 830 vp = vm_gpa_hold(vm, -1, gpa, PAGE_SIZE, VM_PROT_WRITE, 831 &cookie); 832 KASSERT(vp != NULL, ("vm(%s) could not map gpa %#lx", 833 vm_name(vm), gpa)); 834 835 vm_gpa_release(cookie); 836 837 hpa = DMAP_TO_PHYS((uintptr_t)vp); 838 if (map) { 839 iommu_create_mapping(vm->iommu, gpa, hpa, sz); 840 iommu_remove_mapping(host_domain, hpa, sz); 841 } else { 842 iommu_remove_mapping(vm->iommu, gpa, sz); 843 iommu_create_mapping(host_domain, hpa, hpa, sz); 844 } 845 846 gpa += PAGE_SIZE; 847 } 848 } 849 850 /* 851 * Invalidate the cached translations associated with the domain 852 * from which pages were removed. 853 */ 854 if (map) 855 iommu_invalidate_tlb(host_domain); 856 else 857 iommu_invalidate_tlb(vm->iommu); 858 } 859 860 #define vm_iommu_unmap(vm) vm_iommu_modify((vm), FALSE) 861 #define vm_iommu_map(vm) vm_iommu_modify((vm), TRUE) 862 863 int 864 vm_unassign_pptdev(struct vm *vm, int bus, int slot, int func) 865 { 866 int error; 867 868 error = ppt_unassign_device(vm, bus, slot, func); 869 if (error) 870 return (error); 871 872 if (ppt_assigned_devices(vm) == 0) 873 vm_iommu_unmap(vm); 874 875 return (0); 876 } 877 878 int 879 vm_assign_pptdev(struct vm *vm, int bus, int slot, int func) 880 { 881 int error; 882 vm_paddr_t maxaddr; 883 884 /* Set up the IOMMU to do the 'gpa' to 'hpa' translation */ 885 if (ppt_assigned_devices(vm) == 0) { 886 KASSERT(vm->iommu == NULL, 887 ("vm_assign_pptdev: iommu must be NULL")); 888 maxaddr = sysmem_maxaddr(vm); 889 vm->iommu = iommu_create_domain(maxaddr); 890 if (vm->iommu == NULL) 891 return (ENXIO); 892 vm_iommu_map(vm); 893 } 894 895 error = ppt_assign_device(vm, bus, slot, func); 896 return (error); 897 } 898 899 void * 900 vm_gpa_hold(struct vm *vm, int vcpuid, vm_paddr_t gpa, size_t len, int reqprot, 901 void **cookie) 902 { 903 int i, count, pageoff; 904 struct mem_map *mm; 905 vm_page_t m; 906 #ifdef INVARIANTS 907 /* 908 * All vcpus are frozen by ioctls that modify the memory map 909 * (e.g. VM_MMAP_MEMSEG). Therefore 'vm->memmap[]' stability is 910 * guaranteed if at least one vcpu is in the VCPU_FROZEN state. 911 */ 912 int state; 913 KASSERT(vcpuid >= -1 && vcpuid < VM_MAXCPU, ("%s: invalid vcpuid %d", 914 __func__, vcpuid)); 915 for (i = 0; i < VM_MAXCPU; i++) { 916 if (vcpuid != -1 && vcpuid != i) 917 continue; 918 state = vcpu_get_state(vm, i, NULL); 919 KASSERT(state == VCPU_FROZEN, ("%s: invalid vcpu state %d", 920 __func__, state)); 921 } 922 #endif 923 pageoff = gpa & PAGE_MASK; 924 if (len > PAGE_SIZE - pageoff) 925 panic("vm_gpa_hold: invalid gpa/len: 0x%016lx/%lu", gpa, len); 926 927 count = 0; 928 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 929 mm = &vm->mem_maps[i]; 930 if (sysmem_mapping(vm, mm) && gpa >= mm->gpa && 931 gpa < mm->gpa + mm->len) { 932 count = vm_fault_quick_hold_pages(&vm->vmspace->vm_map, 933 trunc_page(gpa), PAGE_SIZE, reqprot, &m, 1); 934 break; 935 } 936 } 937 938 if (count == 1) { 939 *cookie = m; 940 return ((void *)(PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m)) + pageoff)); 941 } else { 942 *cookie = NULL; 943 return (NULL); 944 } 945 } 946 947 void 948 vm_gpa_release(void *cookie) 949 { 950 vm_page_t m = cookie; 951 952 vm_page_lock(m); 953 vm_page_unhold(m); 954 vm_page_unlock(m); 955 } 956 957 int 958 vm_get_register(struct vm *vm, int vcpu, int reg, uint64_t *retval) 959 { 960 961 if (vcpu < 0 || vcpu >= VM_MAXCPU) 962 return (EINVAL); 963 964 if (reg >= VM_REG_LAST) 965 return (EINVAL); 966 967 return (VMGETREG(vm->cookie, vcpu, reg, retval)); 968 } 969 970 int 971 vm_set_register(struct vm *vm, int vcpuid, int reg, uint64_t val) 972 { 973 struct vcpu *vcpu; 974 int error; 975 976 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 977 return (EINVAL); 978 979 if (reg >= VM_REG_LAST) 980 return (EINVAL); 981 982 error = VMSETREG(vm->cookie, vcpuid, reg, val); 983 if (error || reg != VM_REG_GUEST_RIP) 984 return (error); 985 986 /* Set 'nextrip' to match the value of %rip */ 987 VCPU_CTR1(vm, vcpuid, "Setting nextrip to %#lx", val); 988 vcpu = &vm->vcpu[vcpuid]; 989 vcpu->nextrip = val; 990 return (0); 991 } 992 993 static boolean_t 994 is_descriptor_table(int reg) 995 { 996 997 switch (reg) { 998 case VM_REG_GUEST_IDTR: 999 case VM_REG_GUEST_GDTR: 1000 return (TRUE); 1001 default: 1002 return (FALSE); 1003 } 1004 } 1005 1006 static boolean_t 1007 is_segment_register(int reg) 1008 { 1009 1010 switch (reg) { 1011 case VM_REG_GUEST_ES: 1012 case VM_REG_GUEST_CS: 1013 case VM_REG_GUEST_SS: 1014 case VM_REG_GUEST_DS: 1015 case VM_REG_GUEST_FS: 1016 case VM_REG_GUEST_GS: 1017 case VM_REG_GUEST_TR: 1018 case VM_REG_GUEST_LDTR: 1019 return (TRUE); 1020 default: 1021 return (FALSE); 1022 } 1023 } 1024 1025 int 1026 vm_get_seg_desc(struct vm *vm, int vcpu, int reg, 1027 struct seg_desc *desc) 1028 { 1029 1030 if (vcpu < 0 || vcpu >= VM_MAXCPU) 1031 return (EINVAL); 1032 1033 if (!is_segment_register(reg) && !is_descriptor_table(reg)) 1034 return (EINVAL); 1035 1036 return (VMGETDESC(vm->cookie, vcpu, reg, desc)); 1037 } 1038 1039 int 1040 vm_set_seg_desc(struct vm *vm, int vcpu, int reg, 1041 struct seg_desc *desc) 1042 { 1043 if (vcpu < 0 || vcpu >= VM_MAXCPU) 1044 return (EINVAL); 1045 1046 if (!is_segment_register(reg) && !is_descriptor_table(reg)) 1047 return (EINVAL); 1048 1049 return (VMSETDESC(vm->cookie, vcpu, reg, desc)); 1050 } 1051 1052 static void 1053 restore_guest_fpustate(struct vcpu *vcpu) 1054 { 1055 1056 /* flush host state to the pcb */ 1057 fpuexit(curthread); 1058 1059 /* restore guest FPU state */ 1060 fpu_stop_emulating(); 1061 fpurestore(vcpu->guestfpu); 1062 1063 /* restore guest XCR0 if XSAVE is enabled in the host */ 1064 if (rcr4() & CR4_XSAVE) 1065 load_xcr(0, vcpu->guest_xcr0); 1066 1067 /* 1068 * The FPU is now "dirty" with the guest's state so turn on emulation 1069 * to trap any access to the FPU by the host. 1070 */ 1071 fpu_start_emulating(); 1072 } 1073 1074 static void 1075 save_guest_fpustate(struct vcpu *vcpu) 1076 { 1077 1078 if ((rcr0() & CR0_TS) == 0) 1079 panic("fpu emulation not enabled in host!"); 1080 1081 /* save guest XCR0 and restore host XCR0 */ 1082 if (rcr4() & CR4_XSAVE) { 1083 vcpu->guest_xcr0 = rxcr(0); 1084 load_xcr(0, vmm_get_host_xcr0()); 1085 } 1086 1087 /* save guest FPU state */ 1088 fpu_stop_emulating(); 1089 fpusave(vcpu->guestfpu); 1090 fpu_start_emulating(); 1091 } 1092 1093 static VMM_STAT(VCPU_IDLE_TICKS, "number of ticks vcpu was idle"); 1094 1095 static int 1096 vcpu_set_state_locked(struct vm *vm, int vcpuid, enum vcpu_state newstate, 1097 bool from_idle) 1098 { 1099 struct vcpu *vcpu; 1100 int error; 1101 1102 vcpu = &vm->vcpu[vcpuid]; 1103 vcpu_assert_locked(vcpu); 1104 1105 /* 1106 * State transitions from the vmmdev_ioctl() must always begin from 1107 * the VCPU_IDLE state. This guarantees that there is only a single 1108 * ioctl() operating on a vcpu at any point. 1109 */ 1110 if (from_idle) { 1111 while (vcpu->state != VCPU_IDLE) { 1112 vcpu->reqidle = 1; 1113 vcpu_notify_event_locked(vcpu, false); 1114 VCPU_CTR1(vm, vcpuid, "vcpu state change from %s to " 1115 "idle requested", vcpu_state2str(vcpu->state)); 1116 msleep_spin(&vcpu->state, &vcpu->mtx, "vmstat", hz); 1117 } 1118 } else { 1119 KASSERT(vcpu->state != VCPU_IDLE, ("invalid transition from " 1120 "vcpu idle state")); 1121 } 1122 1123 if (vcpu->state == VCPU_RUNNING) { 1124 KASSERT(vcpu->hostcpu == curcpu, ("curcpu %d and hostcpu %d " 1125 "mismatch for running vcpu", curcpu, vcpu->hostcpu)); 1126 } else { 1127 KASSERT(vcpu->hostcpu == NOCPU, ("Invalid hostcpu %d for a " 1128 "vcpu that is not running", vcpu->hostcpu)); 1129 } 1130 1131 /* 1132 * The following state transitions are allowed: 1133 * IDLE -> FROZEN -> IDLE 1134 * FROZEN -> RUNNING -> FROZEN 1135 * FROZEN -> SLEEPING -> FROZEN 1136 */ 1137 switch (vcpu->state) { 1138 case VCPU_IDLE: 1139 case VCPU_RUNNING: 1140 case VCPU_SLEEPING: 1141 error = (newstate != VCPU_FROZEN); 1142 break; 1143 case VCPU_FROZEN: 1144 error = (newstate == VCPU_FROZEN); 1145 break; 1146 default: 1147 error = 1; 1148 break; 1149 } 1150 1151 if (error) 1152 return (EBUSY); 1153 1154 VCPU_CTR2(vm, vcpuid, "vcpu state changed from %s to %s", 1155 vcpu_state2str(vcpu->state), vcpu_state2str(newstate)); 1156 1157 vcpu->state = newstate; 1158 if (newstate == VCPU_RUNNING) 1159 vcpu->hostcpu = curcpu; 1160 else 1161 vcpu->hostcpu = NOCPU; 1162 1163 if (newstate == VCPU_IDLE) 1164 wakeup(&vcpu->state); 1165 1166 return (0); 1167 } 1168 1169 static void 1170 vcpu_require_state(struct vm *vm, int vcpuid, enum vcpu_state newstate) 1171 { 1172 int error; 1173 1174 if ((error = vcpu_set_state(vm, vcpuid, newstate, false)) != 0) 1175 panic("Error %d setting state to %d\n", error, newstate); 1176 } 1177 1178 static void 1179 vcpu_require_state_locked(struct vm *vm, int vcpuid, enum vcpu_state newstate) 1180 { 1181 int error; 1182 1183 if ((error = vcpu_set_state_locked(vm, vcpuid, newstate, false)) != 0) 1184 panic("Error %d setting state to %d", error, newstate); 1185 } 1186 1187 static void 1188 vm_set_rendezvous_func(struct vm *vm, vm_rendezvous_func_t func) 1189 { 1190 1191 KASSERT(mtx_owned(&vm->rendezvous_mtx), ("rendezvous_mtx not locked")); 1192 1193 /* 1194 * Update 'rendezvous_func' and execute a write memory barrier to 1195 * ensure that it is visible across all host cpus. This is not needed 1196 * for correctness but it does ensure that all the vcpus will notice 1197 * that the rendezvous is requested immediately. 1198 */ 1199 vm->rendezvous_func = func; 1200 wmb(); 1201 } 1202 1203 #define RENDEZVOUS_CTR0(vm, vcpuid, fmt) \ 1204 do { \ 1205 if (vcpuid >= 0) \ 1206 VCPU_CTR0(vm, vcpuid, fmt); \ 1207 else \ 1208 VM_CTR0(vm, fmt); \ 1209 } while (0) 1210 1211 static void 1212 vm_handle_rendezvous(struct vm *vm, int vcpuid) 1213 { 1214 1215 KASSERT(vcpuid == -1 || (vcpuid >= 0 && vcpuid < VM_MAXCPU), 1216 ("vm_handle_rendezvous: invalid vcpuid %d", vcpuid)); 1217 1218 mtx_lock(&vm->rendezvous_mtx); 1219 while (vm->rendezvous_func != NULL) { 1220 /* 'rendezvous_req_cpus' must be a subset of 'active_cpus' */ 1221 CPU_AND(&vm->rendezvous_req_cpus, &vm->active_cpus); 1222 1223 if (vcpuid != -1 && 1224 CPU_ISSET(vcpuid, &vm->rendezvous_req_cpus) && 1225 !CPU_ISSET(vcpuid, &vm->rendezvous_done_cpus)) { 1226 VCPU_CTR0(vm, vcpuid, "Calling rendezvous func"); 1227 (*vm->rendezvous_func)(vm, vcpuid, vm->rendezvous_arg); 1228 CPU_SET(vcpuid, &vm->rendezvous_done_cpus); 1229 } 1230 if (CPU_CMP(&vm->rendezvous_req_cpus, 1231 &vm->rendezvous_done_cpus) == 0) { 1232 VCPU_CTR0(vm, vcpuid, "Rendezvous completed"); 1233 vm_set_rendezvous_func(vm, NULL); 1234 wakeup(&vm->rendezvous_func); 1235 break; 1236 } 1237 RENDEZVOUS_CTR0(vm, vcpuid, "Wait for rendezvous completion"); 1238 mtx_sleep(&vm->rendezvous_func, &vm->rendezvous_mtx, 0, 1239 "vmrndv", 0); 1240 } 1241 mtx_unlock(&vm->rendezvous_mtx); 1242 } 1243 1244 /* 1245 * Emulate a guest 'hlt' by sleeping until the vcpu is ready to run. 1246 */ 1247 static int 1248 vm_handle_hlt(struct vm *vm, int vcpuid, bool intr_disabled, bool *retu) 1249 { 1250 struct vcpu *vcpu; 1251 const char *wmesg; 1252 int t, vcpu_halted, vm_halted; 1253 1254 KASSERT(!CPU_ISSET(vcpuid, &vm->halted_cpus), ("vcpu already halted")); 1255 1256 vcpu = &vm->vcpu[vcpuid]; 1257 vcpu_halted = 0; 1258 vm_halted = 0; 1259 1260 vcpu_lock(vcpu); 1261 while (1) { 1262 /* 1263 * Do a final check for pending NMI or interrupts before 1264 * really putting this thread to sleep. Also check for 1265 * software events that would cause this vcpu to wakeup. 1266 * 1267 * These interrupts/events could have happened after the 1268 * vcpu returned from VMRUN() and before it acquired the 1269 * vcpu lock above. 1270 */ 1271 if (vm->rendezvous_func != NULL || vm->suspend || vcpu->reqidle) 1272 break; 1273 if (vm_nmi_pending(vm, vcpuid)) 1274 break; 1275 if (!intr_disabled) { 1276 if (vm_extint_pending(vm, vcpuid) || 1277 vlapic_pending_intr(vcpu->vlapic, NULL)) { 1278 break; 1279 } 1280 } 1281 1282 /* Don't go to sleep if the vcpu thread needs to yield */ 1283 if (vcpu_should_yield(vm, vcpuid)) 1284 break; 1285 1286 /* 1287 * Some Linux guests implement "halt" by having all vcpus 1288 * execute HLT with interrupts disabled. 'halted_cpus' keeps 1289 * track of the vcpus that have entered this state. When all 1290 * vcpus enter the halted state the virtual machine is halted. 1291 */ 1292 if (intr_disabled) { 1293 wmesg = "vmhalt"; 1294 VCPU_CTR0(vm, vcpuid, "Halted"); 1295 if (!vcpu_halted && halt_detection_enabled) { 1296 vcpu_halted = 1; 1297 CPU_SET_ATOMIC(vcpuid, &vm->halted_cpus); 1298 } 1299 if (CPU_CMP(&vm->halted_cpus, &vm->active_cpus) == 0) { 1300 vm_halted = 1; 1301 break; 1302 } 1303 } else { 1304 wmesg = "vmidle"; 1305 } 1306 1307 t = ticks; 1308 vcpu_require_state_locked(vm, vcpuid, VCPU_SLEEPING); 1309 /* 1310 * XXX msleep_spin() cannot be interrupted by signals so 1311 * wake up periodically to check pending signals. 1312 */ 1313 msleep_spin(vcpu, &vcpu->mtx, wmesg, hz); 1314 vcpu_require_state_locked(vm, vcpuid, VCPU_FROZEN); 1315 vmm_stat_incr(vm, vcpuid, VCPU_IDLE_TICKS, ticks - t); 1316 } 1317 1318 if (vcpu_halted) 1319 CPU_CLR_ATOMIC(vcpuid, &vm->halted_cpus); 1320 1321 vcpu_unlock(vcpu); 1322 1323 if (vm_halted) 1324 vm_suspend(vm, VM_SUSPEND_HALT); 1325 1326 return (0); 1327 } 1328 1329 static int 1330 vm_handle_paging(struct vm *vm, int vcpuid, bool *retu) 1331 { 1332 int rv, ftype; 1333 struct vm_map *map; 1334 struct vcpu *vcpu; 1335 struct vm_exit *vme; 1336 1337 vcpu = &vm->vcpu[vcpuid]; 1338 vme = &vcpu->exitinfo; 1339 1340 KASSERT(vme->inst_length == 0, ("%s: invalid inst_length %d", 1341 __func__, vme->inst_length)); 1342 1343 ftype = vme->u.paging.fault_type; 1344 KASSERT(ftype == VM_PROT_READ || 1345 ftype == VM_PROT_WRITE || ftype == VM_PROT_EXECUTE, 1346 ("vm_handle_paging: invalid fault_type %d", ftype)); 1347 1348 if (ftype == VM_PROT_READ || ftype == VM_PROT_WRITE) { 1349 rv = pmap_emulate_accessed_dirty(vmspace_pmap(vm->vmspace), 1350 vme->u.paging.gpa, ftype); 1351 if (rv == 0) { 1352 VCPU_CTR2(vm, vcpuid, "%s bit emulation for gpa %#lx", 1353 ftype == VM_PROT_READ ? "accessed" : "dirty", 1354 vme->u.paging.gpa); 1355 goto done; 1356 } 1357 } 1358 1359 map = &vm->vmspace->vm_map; 1360 rv = vm_fault(map, vme->u.paging.gpa, ftype, VM_FAULT_NORMAL); 1361 1362 VCPU_CTR3(vm, vcpuid, "vm_handle_paging rv = %d, gpa = %#lx, " 1363 "ftype = %d", rv, vme->u.paging.gpa, ftype); 1364 1365 if (rv != KERN_SUCCESS) 1366 return (EFAULT); 1367 done: 1368 return (0); 1369 } 1370 1371 static int 1372 vm_handle_inst_emul(struct vm *vm, int vcpuid, bool *retu) 1373 { 1374 struct vie *vie; 1375 struct vcpu *vcpu; 1376 struct vm_exit *vme; 1377 uint64_t gla, gpa, cs_base; 1378 struct vm_guest_paging *paging; 1379 mem_region_read_t mread; 1380 mem_region_write_t mwrite; 1381 enum vm_cpu_mode cpu_mode; 1382 int cs_d, error, fault; 1383 1384 vcpu = &vm->vcpu[vcpuid]; 1385 vme = &vcpu->exitinfo; 1386 1387 KASSERT(vme->inst_length == 0, ("%s: invalid inst_length %d", 1388 __func__, vme->inst_length)); 1389 1390 gla = vme->u.inst_emul.gla; 1391 gpa = vme->u.inst_emul.gpa; 1392 cs_base = vme->u.inst_emul.cs_base; 1393 cs_d = vme->u.inst_emul.cs_d; 1394 vie = &vme->u.inst_emul.vie; 1395 paging = &vme->u.inst_emul.paging; 1396 cpu_mode = paging->cpu_mode; 1397 1398 VCPU_CTR1(vm, vcpuid, "inst_emul fault accessing gpa %#lx", gpa); 1399 1400 /* Fetch, decode and emulate the faulting instruction */ 1401 if (vie->num_valid == 0) { 1402 error = vmm_fetch_instruction(vm, vcpuid, paging, vme->rip + 1403 cs_base, VIE_INST_SIZE, vie, &fault); 1404 } else { 1405 /* 1406 * The instruction bytes have already been copied into 'vie' 1407 */ 1408 error = fault = 0; 1409 } 1410 if (error || fault) 1411 return (error); 1412 1413 if (vmm_decode_instruction(vm, vcpuid, gla, cpu_mode, cs_d, vie) != 0) { 1414 VCPU_CTR1(vm, vcpuid, "Error decoding instruction at %#lx", 1415 vme->rip + cs_base); 1416 *retu = true; /* dump instruction bytes in userspace */ 1417 return (0); 1418 } 1419 1420 /* 1421 * Update 'nextrip' based on the length of the emulated instruction. 1422 */ 1423 vme->inst_length = vie->num_processed; 1424 vcpu->nextrip += vie->num_processed; 1425 VCPU_CTR1(vm, vcpuid, "nextrip updated to %#lx after instruction " 1426 "decoding", vcpu->nextrip); 1427 1428 /* return to userland unless this is an in-kernel emulated device */ 1429 if (gpa >= DEFAULT_APIC_BASE && gpa < DEFAULT_APIC_BASE + PAGE_SIZE) { 1430 mread = lapic_mmio_read; 1431 mwrite = lapic_mmio_write; 1432 } else if (gpa >= VIOAPIC_BASE && gpa < VIOAPIC_BASE + VIOAPIC_SIZE) { 1433 mread = vioapic_mmio_read; 1434 mwrite = vioapic_mmio_write; 1435 } else if (gpa >= VHPET_BASE && gpa < VHPET_BASE + VHPET_SIZE) { 1436 mread = vhpet_mmio_read; 1437 mwrite = vhpet_mmio_write; 1438 } else { 1439 *retu = true; 1440 return (0); 1441 } 1442 1443 error = vmm_emulate_instruction(vm, vcpuid, gpa, vie, paging, 1444 mread, mwrite, retu); 1445 1446 return (error); 1447 } 1448 1449 static int 1450 vm_handle_suspend(struct vm *vm, int vcpuid, bool *retu) 1451 { 1452 int i, done; 1453 struct vcpu *vcpu; 1454 1455 done = 0; 1456 vcpu = &vm->vcpu[vcpuid]; 1457 1458 CPU_SET_ATOMIC(vcpuid, &vm->suspended_cpus); 1459 1460 /* 1461 * Wait until all 'active_cpus' have suspended themselves. 1462 * 1463 * Since a VM may be suspended at any time including when one or 1464 * more vcpus are doing a rendezvous we need to call the rendezvous 1465 * handler while we are waiting to prevent a deadlock. 1466 */ 1467 vcpu_lock(vcpu); 1468 while (1) { 1469 if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) { 1470 VCPU_CTR0(vm, vcpuid, "All vcpus suspended"); 1471 break; 1472 } 1473 1474 if (vm->rendezvous_func == NULL) { 1475 VCPU_CTR0(vm, vcpuid, "Sleeping during suspend"); 1476 vcpu_require_state_locked(vm, vcpuid, VCPU_SLEEPING); 1477 msleep_spin(vcpu, &vcpu->mtx, "vmsusp", hz); 1478 vcpu_require_state_locked(vm, vcpuid, VCPU_FROZEN); 1479 } else { 1480 VCPU_CTR0(vm, vcpuid, "Rendezvous during suspend"); 1481 vcpu_unlock(vcpu); 1482 vm_handle_rendezvous(vm, vcpuid); 1483 vcpu_lock(vcpu); 1484 } 1485 } 1486 vcpu_unlock(vcpu); 1487 1488 /* 1489 * Wakeup the other sleeping vcpus and return to userspace. 1490 */ 1491 for (i = 0; i < VM_MAXCPU; i++) { 1492 if (CPU_ISSET(i, &vm->suspended_cpus)) { 1493 vcpu_notify_event(vm, i, false); 1494 } 1495 } 1496 1497 *retu = true; 1498 return (0); 1499 } 1500 1501 static int 1502 vm_handle_reqidle(struct vm *vm, int vcpuid, bool *retu) 1503 { 1504 struct vcpu *vcpu = &vm->vcpu[vcpuid]; 1505 1506 vcpu_lock(vcpu); 1507 KASSERT(vcpu->reqidle, ("invalid vcpu reqidle %d", vcpu->reqidle)); 1508 vcpu->reqidle = 0; 1509 vcpu_unlock(vcpu); 1510 *retu = true; 1511 return (0); 1512 } 1513 1514 int 1515 vm_suspend(struct vm *vm, enum vm_suspend_how how) 1516 { 1517 int i; 1518 1519 if (how <= VM_SUSPEND_NONE || how >= VM_SUSPEND_LAST) 1520 return (EINVAL); 1521 1522 if (atomic_cmpset_int(&vm->suspend, 0, how) == 0) { 1523 VM_CTR2(vm, "virtual machine already suspended %d/%d", 1524 vm->suspend, how); 1525 return (EALREADY); 1526 } 1527 1528 VM_CTR1(vm, "virtual machine successfully suspended %d", how); 1529 1530 /* 1531 * Notify all active vcpus that they are now suspended. 1532 */ 1533 for (i = 0; i < VM_MAXCPU; i++) { 1534 if (CPU_ISSET(i, &vm->active_cpus)) 1535 vcpu_notify_event(vm, i, false); 1536 } 1537 1538 return (0); 1539 } 1540 1541 void 1542 vm_exit_suspended(struct vm *vm, int vcpuid, uint64_t rip) 1543 { 1544 struct vm_exit *vmexit; 1545 1546 KASSERT(vm->suspend > VM_SUSPEND_NONE && vm->suspend < VM_SUSPEND_LAST, 1547 ("vm_exit_suspended: invalid suspend type %d", vm->suspend)); 1548 1549 vmexit = vm_exitinfo(vm, vcpuid); 1550 vmexit->rip = rip; 1551 vmexit->inst_length = 0; 1552 vmexit->exitcode = VM_EXITCODE_SUSPENDED; 1553 vmexit->u.suspended.how = vm->suspend; 1554 } 1555 1556 void 1557 vm_exit_rendezvous(struct vm *vm, int vcpuid, uint64_t rip) 1558 { 1559 struct vm_exit *vmexit; 1560 1561 KASSERT(vm->rendezvous_func != NULL, ("rendezvous not in progress")); 1562 1563 vmexit = vm_exitinfo(vm, vcpuid); 1564 vmexit->rip = rip; 1565 vmexit->inst_length = 0; 1566 vmexit->exitcode = VM_EXITCODE_RENDEZVOUS; 1567 vmm_stat_incr(vm, vcpuid, VMEXIT_RENDEZVOUS, 1); 1568 } 1569 1570 void 1571 vm_exit_reqidle(struct vm *vm, int vcpuid, uint64_t rip) 1572 { 1573 struct vm_exit *vmexit; 1574 1575 vmexit = vm_exitinfo(vm, vcpuid); 1576 vmexit->rip = rip; 1577 vmexit->inst_length = 0; 1578 vmexit->exitcode = VM_EXITCODE_REQIDLE; 1579 vmm_stat_incr(vm, vcpuid, VMEXIT_REQIDLE, 1); 1580 } 1581 1582 void 1583 vm_exit_astpending(struct vm *vm, int vcpuid, uint64_t rip) 1584 { 1585 struct vm_exit *vmexit; 1586 1587 vmexit = vm_exitinfo(vm, vcpuid); 1588 vmexit->rip = rip; 1589 vmexit->inst_length = 0; 1590 vmexit->exitcode = VM_EXITCODE_BOGUS; 1591 vmm_stat_incr(vm, vcpuid, VMEXIT_ASTPENDING, 1); 1592 } 1593 1594 int 1595 vm_run(struct vm *vm, struct vm_run *vmrun) 1596 { 1597 struct vm_eventinfo evinfo; 1598 int error, vcpuid; 1599 struct vcpu *vcpu; 1600 struct pcb *pcb; 1601 uint64_t tscval; 1602 struct vm_exit *vme; 1603 bool retu, intr_disabled; 1604 pmap_t pmap; 1605 1606 vcpuid = vmrun->cpuid; 1607 1608 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1609 return (EINVAL); 1610 1611 if (!CPU_ISSET(vcpuid, &vm->active_cpus)) 1612 return (EINVAL); 1613 1614 if (CPU_ISSET(vcpuid, &vm->suspended_cpus)) 1615 return (EINVAL); 1616 1617 pmap = vmspace_pmap(vm->vmspace); 1618 vcpu = &vm->vcpu[vcpuid]; 1619 vme = &vcpu->exitinfo; 1620 evinfo.rptr = &vm->rendezvous_func; 1621 evinfo.sptr = &vm->suspend; 1622 evinfo.iptr = &vcpu->reqidle; 1623 restart: 1624 critical_enter(); 1625 1626 KASSERT(!CPU_ISSET(curcpu, &pmap->pm_active), 1627 ("vm_run: absurd pm_active")); 1628 1629 tscval = rdtsc(); 1630 1631 pcb = PCPU_GET(curpcb); 1632 set_pcb_flags(pcb, PCB_FULL_IRET); 1633 1634 restore_guest_fpustate(vcpu); 1635 1636 vcpu_require_state(vm, vcpuid, VCPU_RUNNING); 1637 error = VMRUN(vm->cookie, vcpuid, vcpu->nextrip, pmap, &evinfo); 1638 vcpu_require_state(vm, vcpuid, VCPU_FROZEN); 1639 1640 save_guest_fpustate(vcpu); 1641 1642 vmm_stat_incr(vm, vcpuid, VCPU_TOTAL_RUNTIME, rdtsc() - tscval); 1643 1644 critical_exit(); 1645 1646 if (error == 0) { 1647 retu = false; 1648 vcpu->nextrip = vme->rip + vme->inst_length; 1649 switch (vme->exitcode) { 1650 case VM_EXITCODE_REQIDLE: 1651 error = vm_handle_reqidle(vm, vcpuid, &retu); 1652 break; 1653 case VM_EXITCODE_SUSPENDED: 1654 error = vm_handle_suspend(vm, vcpuid, &retu); 1655 break; 1656 case VM_EXITCODE_IOAPIC_EOI: 1657 vioapic_process_eoi(vm, vcpuid, 1658 vme->u.ioapic_eoi.vector); 1659 break; 1660 case VM_EXITCODE_RENDEZVOUS: 1661 vm_handle_rendezvous(vm, vcpuid); 1662 error = 0; 1663 break; 1664 case VM_EXITCODE_HLT: 1665 intr_disabled = ((vme->u.hlt.rflags & PSL_I) == 0); 1666 error = vm_handle_hlt(vm, vcpuid, intr_disabled, &retu); 1667 break; 1668 case VM_EXITCODE_PAGING: 1669 error = vm_handle_paging(vm, vcpuid, &retu); 1670 break; 1671 case VM_EXITCODE_INST_EMUL: 1672 error = vm_handle_inst_emul(vm, vcpuid, &retu); 1673 break; 1674 case VM_EXITCODE_INOUT: 1675 case VM_EXITCODE_INOUT_STR: 1676 error = vm_handle_inout(vm, vcpuid, vme, &retu); 1677 break; 1678 case VM_EXITCODE_MONITOR: 1679 case VM_EXITCODE_MWAIT: 1680 vm_inject_ud(vm, vcpuid); 1681 break; 1682 default: 1683 retu = true; /* handled in userland */ 1684 break; 1685 } 1686 } 1687 1688 if (error == 0 && retu == false) 1689 goto restart; 1690 1691 VCPU_CTR2(vm, vcpuid, "retu %d/%d", error, vme->exitcode); 1692 1693 /* copy the exit information */ 1694 bcopy(vme, &vmrun->vm_exit, sizeof(struct vm_exit)); 1695 return (error); 1696 } 1697 1698 int 1699 vm_restart_instruction(void *arg, int vcpuid) 1700 { 1701 struct vm *vm; 1702 struct vcpu *vcpu; 1703 enum vcpu_state state; 1704 uint64_t rip; 1705 int error; 1706 1707 vm = arg; 1708 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1709 return (EINVAL); 1710 1711 vcpu = &vm->vcpu[vcpuid]; 1712 state = vcpu_get_state(vm, vcpuid, NULL); 1713 if (state == VCPU_RUNNING) { 1714 /* 1715 * When a vcpu is "running" the next instruction is determined 1716 * by adding 'rip' and 'inst_length' in the vcpu's 'exitinfo'. 1717 * Thus setting 'inst_length' to zero will cause the current 1718 * instruction to be restarted. 1719 */ 1720 vcpu->exitinfo.inst_length = 0; 1721 VCPU_CTR1(vm, vcpuid, "restarting instruction at %#lx by " 1722 "setting inst_length to zero", vcpu->exitinfo.rip); 1723 } else if (state == VCPU_FROZEN) { 1724 /* 1725 * When a vcpu is "frozen" it is outside the critical section 1726 * around VMRUN() and 'nextrip' points to the next instruction. 1727 * Thus instruction restart is achieved by setting 'nextrip' 1728 * to the vcpu's %rip. 1729 */ 1730 error = vm_get_register(vm, vcpuid, VM_REG_GUEST_RIP, &rip); 1731 KASSERT(!error, ("%s: error %d getting rip", __func__, error)); 1732 VCPU_CTR2(vm, vcpuid, "restarting instruction by updating " 1733 "nextrip from %#lx to %#lx", vcpu->nextrip, rip); 1734 vcpu->nextrip = rip; 1735 } else { 1736 panic("%s: invalid state %d", __func__, state); 1737 } 1738 return (0); 1739 } 1740 1741 int 1742 vm_exit_intinfo(struct vm *vm, int vcpuid, uint64_t info) 1743 { 1744 struct vcpu *vcpu; 1745 int type, vector; 1746 1747 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1748 return (EINVAL); 1749 1750 vcpu = &vm->vcpu[vcpuid]; 1751 1752 if (info & VM_INTINFO_VALID) { 1753 type = info & VM_INTINFO_TYPE; 1754 vector = info & 0xff; 1755 if (type == VM_INTINFO_NMI && vector != IDT_NMI) 1756 return (EINVAL); 1757 if (type == VM_INTINFO_HWEXCEPTION && vector >= 32) 1758 return (EINVAL); 1759 if (info & VM_INTINFO_RSVD) 1760 return (EINVAL); 1761 } else { 1762 info = 0; 1763 } 1764 VCPU_CTR2(vm, vcpuid, "%s: info1(%#lx)", __func__, info); 1765 vcpu->exitintinfo = info; 1766 return (0); 1767 } 1768 1769 enum exc_class { 1770 EXC_BENIGN, 1771 EXC_CONTRIBUTORY, 1772 EXC_PAGEFAULT 1773 }; 1774 1775 #define IDT_VE 20 /* Virtualization Exception (Intel specific) */ 1776 1777 static enum exc_class 1778 exception_class(uint64_t info) 1779 { 1780 int type, vector; 1781 1782 KASSERT(info & VM_INTINFO_VALID, ("intinfo must be valid: %#lx", info)); 1783 type = info & VM_INTINFO_TYPE; 1784 vector = info & 0xff; 1785 1786 /* Table 6-4, "Interrupt and Exception Classes", Intel SDM, Vol 3 */ 1787 switch (type) { 1788 case VM_INTINFO_HWINTR: 1789 case VM_INTINFO_SWINTR: 1790 case VM_INTINFO_NMI: 1791 return (EXC_BENIGN); 1792 default: 1793 /* 1794 * Hardware exception. 1795 * 1796 * SVM and VT-x use identical type values to represent NMI, 1797 * hardware interrupt and software interrupt. 1798 * 1799 * SVM uses type '3' for all exceptions. VT-x uses type '3' 1800 * for exceptions except #BP and #OF. #BP and #OF use a type 1801 * value of '5' or '6'. Therefore we don't check for explicit 1802 * values of 'type' to classify 'intinfo' into a hardware 1803 * exception. 1804 */ 1805 break; 1806 } 1807 1808 switch (vector) { 1809 case IDT_PF: 1810 case IDT_VE: 1811 return (EXC_PAGEFAULT); 1812 case IDT_DE: 1813 case IDT_TS: 1814 case IDT_NP: 1815 case IDT_SS: 1816 case IDT_GP: 1817 return (EXC_CONTRIBUTORY); 1818 default: 1819 return (EXC_BENIGN); 1820 } 1821 } 1822 1823 static int 1824 nested_fault(struct vm *vm, int vcpuid, uint64_t info1, uint64_t info2, 1825 uint64_t *retinfo) 1826 { 1827 enum exc_class exc1, exc2; 1828 int type1, vector1; 1829 1830 KASSERT(info1 & VM_INTINFO_VALID, ("info1 %#lx is not valid", info1)); 1831 KASSERT(info2 & VM_INTINFO_VALID, ("info2 %#lx is not valid", info2)); 1832 1833 /* 1834 * If an exception occurs while attempting to call the double-fault 1835 * handler the processor enters shutdown mode (aka triple fault). 1836 */ 1837 type1 = info1 & VM_INTINFO_TYPE; 1838 vector1 = info1 & 0xff; 1839 if (type1 == VM_INTINFO_HWEXCEPTION && vector1 == IDT_DF) { 1840 VCPU_CTR2(vm, vcpuid, "triple fault: info1(%#lx), info2(%#lx)", 1841 info1, info2); 1842 vm_suspend(vm, VM_SUSPEND_TRIPLEFAULT); 1843 *retinfo = 0; 1844 return (0); 1845 } 1846 1847 /* 1848 * Table 6-5 "Conditions for Generating a Double Fault", Intel SDM, Vol3 1849 */ 1850 exc1 = exception_class(info1); 1851 exc2 = exception_class(info2); 1852 if ((exc1 == EXC_CONTRIBUTORY && exc2 == EXC_CONTRIBUTORY) || 1853 (exc1 == EXC_PAGEFAULT && exc2 != EXC_BENIGN)) { 1854 /* Convert nested fault into a double fault. */ 1855 *retinfo = IDT_DF; 1856 *retinfo |= VM_INTINFO_VALID | VM_INTINFO_HWEXCEPTION; 1857 *retinfo |= VM_INTINFO_DEL_ERRCODE; 1858 } else { 1859 /* Handle exceptions serially */ 1860 *retinfo = info2; 1861 } 1862 return (1); 1863 } 1864 1865 static uint64_t 1866 vcpu_exception_intinfo(struct vcpu *vcpu) 1867 { 1868 uint64_t info = 0; 1869 1870 if (vcpu->exception_pending) { 1871 info = vcpu->exc_vector & 0xff; 1872 info |= VM_INTINFO_VALID | VM_INTINFO_HWEXCEPTION; 1873 if (vcpu->exc_errcode_valid) { 1874 info |= VM_INTINFO_DEL_ERRCODE; 1875 info |= (uint64_t)vcpu->exc_errcode << 32; 1876 } 1877 } 1878 return (info); 1879 } 1880 1881 int 1882 vm_entry_intinfo(struct vm *vm, int vcpuid, uint64_t *retinfo) 1883 { 1884 struct vcpu *vcpu; 1885 uint64_t info1, info2; 1886 int valid; 1887 1888 KASSERT(vcpuid >= 0 && vcpuid < VM_MAXCPU, ("invalid vcpu %d", vcpuid)); 1889 1890 vcpu = &vm->vcpu[vcpuid]; 1891 1892 info1 = vcpu->exitintinfo; 1893 vcpu->exitintinfo = 0; 1894 1895 info2 = 0; 1896 if (vcpu->exception_pending) { 1897 info2 = vcpu_exception_intinfo(vcpu); 1898 vcpu->exception_pending = 0; 1899 VCPU_CTR2(vm, vcpuid, "Exception %d delivered: %#lx", 1900 vcpu->exc_vector, info2); 1901 } 1902 1903 if ((info1 & VM_INTINFO_VALID) && (info2 & VM_INTINFO_VALID)) { 1904 valid = nested_fault(vm, vcpuid, info1, info2, retinfo); 1905 } else if (info1 & VM_INTINFO_VALID) { 1906 *retinfo = info1; 1907 valid = 1; 1908 } else if (info2 & VM_INTINFO_VALID) { 1909 *retinfo = info2; 1910 valid = 1; 1911 } else { 1912 valid = 0; 1913 } 1914 1915 if (valid) { 1916 VCPU_CTR4(vm, vcpuid, "%s: info1(%#lx), info2(%#lx), " 1917 "retinfo(%#lx)", __func__, info1, info2, *retinfo); 1918 } 1919 1920 return (valid); 1921 } 1922 1923 int 1924 vm_get_intinfo(struct vm *vm, int vcpuid, uint64_t *info1, uint64_t *info2) 1925 { 1926 struct vcpu *vcpu; 1927 1928 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1929 return (EINVAL); 1930 1931 vcpu = &vm->vcpu[vcpuid]; 1932 *info1 = vcpu->exitintinfo; 1933 *info2 = vcpu_exception_intinfo(vcpu); 1934 return (0); 1935 } 1936 1937 int 1938 vm_inject_exception(struct vm *vm, int vcpuid, int vector, int errcode_valid, 1939 uint32_t errcode, int restart_instruction) 1940 { 1941 struct vcpu *vcpu; 1942 uint64_t regval; 1943 int error; 1944 1945 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1946 return (EINVAL); 1947 1948 if (vector < 0 || vector >= 32) 1949 return (EINVAL); 1950 1951 /* 1952 * A double fault exception should never be injected directly into 1953 * the guest. It is a derived exception that results from specific 1954 * combinations of nested faults. 1955 */ 1956 if (vector == IDT_DF) 1957 return (EINVAL); 1958 1959 vcpu = &vm->vcpu[vcpuid]; 1960 1961 if (vcpu->exception_pending) { 1962 VCPU_CTR2(vm, vcpuid, "Unable to inject exception %d due to " 1963 "pending exception %d", vector, vcpu->exc_vector); 1964 return (EBUSY); 1965 } 1966 1967 if (errcode_valid) { 1968 /* 1969 * Exceptions don't deliver an error code in real mode. 1970 */ 1971 error = vm_get_register(vm, vcpuid, VM_REG_GUEST_CR0, ®val); 1972 KASSERT(!error, ("%s: error %d getting CR0", __func__, error)); 1973 if (!(regval & CR0_PE)) 1974 errcode_valid = 0; 1975 } 1976 1977 /* 1978 * From section 26.6.1 "Interruptibility State" in Intel SDM: 1979 * 1980 * Event blocking by "STI" or "MOV SS" is cleared after guest executes 1981 * one instruction or incurs an exception. 1982 */ 1983 error = vm_set_register(vm, vcpuid, VM_REG_GUEST_INTR_SHADOW, 0); 1984 KASSERT(error == 0, ("%s: error %d clearing interrupt shadow", 1985 __func__, error)); 1986 1987 if (restart_instruction) 1988 vm_restart_instruction(vm, vcpuid); 1989 1990 vcpu->exception_pending = 1; 1991 vcpu->exc_vector = vector; 1992 vcpu->exc_errcode = errcode; 1993 vcpu->exc_errcode_valid = errcode_valid; 1994 VCPU_CTR1(vm, vcpuid, "Exception %d pending", vector); 1995 return (0); 1996 } 1997 1998 void 1999 vm_inject_fault(void *vmarg, int vcpuid, int vector, int errcode_valid, 2000 int errcode) 2001 { 2002 struct vm *vm; 2003 int error, restart_instruction; 2004 2005 vm = vmarg; 2006 restart_instruction = 1; 2007 2008 error = vm_inject_exception(vm, vcpuid, vector, errcode_valid, 2009 errcode, restart_instruction); 2010 KASSERT(error == 0, ("vm_inject_exception error %d", error)); 2011 } 2012 2013 void 2014 vm_inject_pf(void *vmarg, int vcpuid, int error_code, uint64_t cr2) 2015 { 2016 struct vm *vm; 2017 int error; 2018 2019 vm = vmarg; 2020 VCPU_CTR2(vm, vcpuid, "Injecting page fault: error_code %#x, cr2 %#lx", 2021 error_code, cr2); 2022 2023 error = vm_set_register(vm, vcpuid, VM_REG_GUEST_CR2, cr2); 2024 KASSERT(error == 0, ("vm_set_register(cr2) error %d", error)); 2025 2026 vm_inject_fault(vm, vcpuid, IDT_PF, 1, error_code); 2027 } 2028 2029 static VMM_STAT(VCPU_NMI_COUNT, "number of NMIs delivered to vcpu"); 2030 2031 int 2032 vm_inject_nmi(struct vm *vm, int vcpuid) 2033 { 2034 struct vcpu *vcpu; 2035 2036 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 2037 return (EINVAL); 2038 2039 vcpu = &vm->vcpu[vcpuid]; 2040 2041 vcpu->nmi_pending = 1; 2042 vcpu_notify_event(vm, vcpuid, false); 2043 return (0); 2044 } 2045 2046 int 2047 vm_nmi_pending(struct vm *vm, int vcpuid) 2048 { 2049 struct vcpu *vcpu; 2050 2051 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 2052 panic("vm_nmi_pending: invalid vcpuid %d", vcpuid); 2053 2054 vcpu = &vm->vcpu[vcpuid]; 2055 2056 return (vcpu->nmi_pending); 2057 } 2058 2059 void 2060 vm_nmi_clear(struct vm *vm, int vcpuid) 2061 { 2062 struct vcpu *vcpu; 2063 2064 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 2065 panic("vm_nmi_pending: invalid vcpuid %d", vcpuid); 2066 2067 vcpu = &vm->vcpu[vcpuid]; 2068 2069 if (vcpu->nmi_pending == 0) 2070 panic("vm_nmi_clear: inconsistent nmi_pending state"); 2071 2072 vcpu->nmi_pending = 0; 2073 vmm_stat_incr(vm, vcpuid, VCPU_NMI_COUNT, 1); 2074 } 2075 2076 static VMM_STAT(VCPU_EXTINT_COUNT, "number of ExtINTs delivered to vcpu"); 2077 2078 int 2079 vm_inject_extint(struct vm *vm, int vcpuid) 2080 { 2081 struct vcpu *vcpu; 2082 2083 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 2084 return (EINVAL); 2085 2086 vcpu = &vm->vcpu[vcpuid]; 2087 2088 vcpu->extint_pending = 1; 2089 vcpu_notify_event(vm, vcpuid, false); 2090 return (0); 2091 } 2092 2093 int 2094 vm_extint_pending(struct vm *vm, int vcpuid) 2095 { 2096 struct vcpu *vcpu; 2097 2098 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 2099 panic("vm_extint_pending: invalid vcpuid %d", vcpuid); 2100 2101 vcpu = &vm->vcpu[vcpuid]; 2102 2103 return (vcpu->extint_pending); 2104 } 2105 2106 void 2107 vm_extint_clear(struct vm *vm, int vcpuid) 2108 { 2109 struct vcpu *vcpu; 2110 2111 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 2112 panic("vm_extint_pending: invalid vcpuid %d", vcpuid); 2113 2114 vcpu = &vm->vcpu[vcpuid]; 2115 2116 if (vcpu->extint_pending == 0) 2117 panic("vm_extint_clear: inconsistent extint_pending state"); 2118 2119 vcpu->extint_pending = 0; 2120 vmm_stat_incr(vm, vcpuid, VCPU_EXTINT_COUNT, 1); 2121 } 2122 2123 int 2124 vm_get_capability(struct vm *vm, int vcpu, int type, int *retval) 2125 { 2126 if (vcpu < 0 || vcpu >= VM_MAXCPU) 2127 return (EINVAL); 2128 2129 if (type < 0 || type >= VM_CAP_MAX) 2130 return (EINVAL); 2131 2132 return (VMGETCAP(vm->cookie, vcpu, type, retval)); 2133 } 2134 2135 int 2136 vm_set_capability(struct vm *vm, int vcpu, int type, int val) 2137 { 2138 if (vcpu < 0 || vcpu >= VM_MAXCPU) 2139 return (EINVAL); 2140 2141 if (type < 0 || type >= VM_CAP_MAX) 2142 return (EINVAL); 2143 2144 return (VMSETCAP(vm->cookie, vcpu, type, val)); 2145 } 2146 2147 struct vlapic * 2148 vm_lapic(struct vm *vm, int cpu) 2149 { 2150 return (vm->vcpu[cpu].vlapic); 2151 } 2152 2153 struct vioapic * 2154 vm_ioapic(struct vm *vm) 2155 { 2156 2157 return (vm->vioapic); 2158 } 2159 2160 struct vhpet * 2161 vm_hpet(struct vm *vm) 2162 { 2163 2164 return (vm->vhpet); 2165 } 2166 2167 boolean_t 2168 vmm_is_pptdev(int bus, int slot, int func) 2169 { 2170 int found, i, n; 2171 int b, s, f; 2172 char *val, *cp, *cp2; 2173 2174 /* 2175 * XXX 2176 * The length of an environment variable is limited to 128 bytes which 2177 * puts an upper limit on the number of passthru devices that may be 2178 * specified using a single environment variable. 2179 * 2180 * Work around this by scanning multiple environment variable 2181 * names instead of a single one - yuck! 2182 */ 2183 const char *names[] = { "pptdevs", "pptdevs2", "pptdevs3", NULL }; 2184 2185 /* set pptdevs="1/2/3 4/5/6 7/8/9 10/11/12" */ 2186 found = 0; 2187 for (i = 0; names[i] != NULL && !found; i++) { 2188 cp = val = kern_getenv(names[i]); 2189 while (cp != NULL && *cp != '\0') { 2190 if ((cp2 = strchr(cp, ' ')) != NULL) 2191 *cp2 = '\0'; 2192 2193 n = sscanf(cp, "%d/%d/%d", &b, &s, &f); 2194 if (n == 3 && bus == b && slot == s && func == f) { 2195 found = 1; 2196 break; 2197 } 2198 2199 if (cp2 != NULL) 2200 *cp2++ = ' '; 2201 2202 cp = cp2; 2203 } 2204 freeenv(val); 2205 } 2206 return (found); 2207 } 2208 2209 void * 2210 vm_iommu_domain(struct vm *vm) 2211 { 2212 2213 return (vm->iommu); 2214 } 2215 2216 int 2217 vcpu_set_state(struct vm *vm, int vcpuid, enum vcpu_state newstate, 2218 bool from_idle) 2219 { 2220 int error; 2221 struct vcpu *vcpu; 2222 2223 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 2224 panic("vm_set_run_state: invalid vcpuid %d", vcpuid); 2225 2226 vcpu = &vm->vcpu[vcpuid]; 2227 2228 vcpu_lock(vcpu); 2229 error = vcpu_set_state_locked(vm, vcpuid, newstate, from_idle); 2230 vcpu_unlock(vcpu); 2231 2232 return (error); 2233 } 2234 2235 enum vcpu_state 2236 vcpu_get_state(struct vm *vm, int vcpuid, int *hostcpu) 2237 { 2238 struct vcpu *vcpu; 2239 enum vcpu_state state; 2240 2241 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 2242 panic("vm_get_run_state: invalid vcpuid %d", vcpuid); 2243 2244 vcpu = &vm->vcpu[vcpuid]; 2245 2246 vcpu_lock(vcpu); 2247 state = vcpu->state; 2248 if (hostcpu != NULL) 2249 *hostcpu = vcpu->hostcpu; 2250 vcpu_unlock(vcpu); 2251 2252 return (state); 2253 } 2254 2255 int 2256 vm_activate_cpu(struct vm *vm, int vcpuid) 2257 { 2258 2259 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 2260 return (EINVAL); 2261 2262 if (CPU_ISSET(vcpuid, &vm->active_cpus)) 2263 return (EBUSY); 2264 2265 VCPU_CTR0(vm, vcpuid, "activated"); 2266 CPU_SET_ATOMIC(vcpuid, &vm->active_cpus); 2267 return (0); 2268 } 2269 2270 cpuset_t 2271 vm_active_cpus(struct vm *vm) 2272 { 2273 2274 return (vm->active_cpus); 2275 } 2276 2277 cpuset_t 2278 vm_suspended_cpus(struct vm *vm) 2279 { 2280 2281 return (vm->suspended_cpus); 2282 } 2283 2284 void * 2285 vcpu_stats(struct vm *vm, int vcpuid) 2286 { 2287 2288 return (vm->vcpu[vcpuid].stats); 2289 } 2290 2291 int 2292 vm_get_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state *state) 2293 { 2294 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 2295 return (EINVAL); 2296 2297 *state = vm->vcpu[vcpuid].x2apic_state; 2298 2299 return (0); 2300 } 2301 2302 int 2303 vm_set_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state state) 2304 { 2305 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 2306 return (EINVAL); 2307 2308 if (state >= X2APIC_STATE_LAST) 2309 return (EINVAL); 2310 2311 vm->vcpu[vcpuid].x2apic_state = state; 2312 2313 vlapic_set_x2apic_state(vm, vcpuid, state); 2314 2315 return (0); 2316 } 2317 2318 /* 2319 * This function is called to ensure that a vcpu "sees" a pending event 2320 * as soon as possible: 2321 * - If the vcpu thread is sleeping then it is woken up. 2322 * - If the vcpu is running on a different host_cpu then an IPI will be directed 2323 * to the host_cpu to cause the vcpu to trap into the hypervisor. 2324 */ 2325 static void 2326 vcpu_notify_event_locked(struct vcpu *vcpu, bool lapic_intr) 2327 { 2328 int hostcpu; 2329 2330 hostcpu = vcpu->hostcpu; 2331 if (vcpu->state == VCPU_RUNNING) { 2332 KASSERT(hostcpu != NOCPU, ("vcpu running on invalid hostcpu")); 2333 if (hostcpu != curcpu) { 2334 if (lapic_intr) { 2335 vlapic_post_intr(vcpu->vlapic, hostcpu, 2336 vmm_ipinum); 2337 } else { 2338 ipi_cpu(hostcpu, vmm_ipinum); 2339 } 2340 } else { 2341 /* 2342 * If the 'vcpu' is running on 'curcpu' then it must 2343 * be sending a notification to itself (e.g. SELF_IPI). 2344 * The pending event will be picked up when the vcpu 2345 * transitions back to guest context. 2346 */ 2347 } 2348 } else { 2349 KASSERT(hostcpu == NOCPU, ("vcpu state %d not consistent " 2350 "with hostcpu %d", vcpu->state, hostcpu)); 2351 if (vcpu->state == VCPU_SLEEPING) 2352 wakeup_one(vcpu); 2353 } 2354 } 2355 2356 void 2357 vcpu_notify_event(struct vm *vm, int vcpuid, bool lapic_intr) 2358 { 2359 struct vcpu *vcpu = &vm->vcpu[vcpuid]; 2360 2361 vcpu_lock(vcpu); 2362 vcpu_notify_event_locked(vcpu, lapic_intr); 2363 vcpu_unlock(vcpu); 2364 } 2365 2366 struct vmspace * 2367 vm_get_vmspace(struct vm *vm) 2368 { 2369 2370 return (vm->vmspace); 2371 } 2372 2373 int 2374 vm_apicid2vcpuid(struct vm *vm, int apicid) 2375 { 2376 /* 2377 * XXX apic id is assumed to be numerically identical to vcpu id 2378 */ 2379 return (apicid); 2380 } 2381 2382 void 2383 vm_smp_rendezvous(struct vm *vm, int vcpuid, cpuset_t dest, 2384 vm_rendezvous_func_t func, void *arg) 2385 { 2386 int i; 2387 2388 /* 2389 * Enforce that this function is called without any locks 2390 */ 2391 WITNESS_WARN(WARN_PANIC, NULL, "vm_smp_rendezvous"); 2392 KASSERT(vcpuid == -1 || (vcpuid >= 0 && vcpuid < VM_MAXCPU), 2393 ("vm_smp_rendezvous: invalid vcpuid %d", vcpuid)); 2394 2395 restart: 2396 mtx_lock(&vm->rendezvous_mtx); 2397 if (vm->rendezvous_func != NULL) { 2398 /* 2399 * If a rendezvous is already in progress then we need to 2400 * call the rendezvous handler in case this 'vcpuid' is one 2401 * of the targets of the rendezvous. 2402 */ 2403 RENDEZVOUS_CTR0(vm, vcpuid, "Rendezvous already in progress"); 2404 mtx_unlock(&vm->rendezvous_mtx); 2405 vm_handle_rendezvous(vm, vcpuid); 2406 goto restart; 2407 } 2408 KASSERT(vm->rendezvous_func == NULL, ("vm_smp_rendezvous: previous " 2409 "rendezvous is still in progress")); 2410 2411 RENDEZVOUS_CTR0(vm, vcpuid, "Initiating rendezvous"); 2412 vm->rendezvous_req_cpus = dest; 2413 CPU_ZERO(&vm->rendezvous_done_cpus); 2414 vm->rendezvous_arg = arg; 2415 vm_set_rendezvous_func(vm, func); 2416 mtx_unlock(&vm->rendezvous_mtx); 2417 2418 /* 2419 * Wake up any sleeping vcpus and trigger a VM-exit in any running 2420 * vcpus so they handle the rendezvous as soon as possible. 2421 */ 2422 for (i = 0; i < VM_MAXCPU; i++) { 2423 if (CPU_ISSET(i, &dest)) 2424 vcpu_notify_event(vm, i, false); 2425 } 2426 2427 vm_handle_rendezvous(vm, vcpuid); 2428 } 2429 2430 struct vatpic * 2431 vm_atpic(struct vm *vm) 2432 { 2433 return (vm->vatpic); 2434 } 2435 2436 struct vatpit * 2437 vm_atpit(struct vm *vm) 2438 { 2439 return (vm->vatpit); 2440 } 2441 2442 struct vpmtmr * 2443 vm_pmtmr(struct vm *vm) 2444 { 2445 2446 return (vm->vpmtmr); 2447 } 2448 2449 struct vrtc * 2450 vm_rtc(struct vm *vm) 2451 { 2452 2453 return (vm->vrtc); 2454 } 2455 2456 enum vm_reg_name 2457 vm_segment_name(int seg) 2458 { 2459 static enum vm_reg_name seg_names[] = { 2460 VM_REG_GUEST_ES, 2461 VM_REG_GUEST_CS, 2462 VM_REG_GUEST_SS, 2463 VM_REG_GUEST_DS, 2464 VM_REG_GUEST_FS, 2465 VM_REG_GUEST_GS 2466 }; 2467 2468 KASSERT(seg >= 0 && seg < nitems(seg_names), 2469 ("%s: invalid segment encoding %d", __func__, seg)); 2470 return (seg_names[seg]); 2471 } 2472 2473 void 2474 vm_copy_teardown(struct vm *vm, int vcpuid, struct vm_copyinfo *copyinfo, 2475 int num_copyinfo) 2476 { 2477 int idx; 2478 2479 for (idx = 0; idx < num_copyinfo; idx++) { 2480 if (copyinfo[idx].cookie != NULL) 2481 vm_gpa_release(copyinfo[idx].cookie); 2482 } 2483 bzero(copyinfo, num_copyinfo * sizeof(struct vm_copyinfo)); 2484 } 2485 2486 int 2487 vm_copy_setup(struct vm *vm, int vcpuid, struct vm_guest_paging *paging, 2488 uint64_t gla, size_t len, int prot, struct vm_copyinfo *copyinfo, 2489 int num_copyinfo, int *fault) 2490 { 2491 int error, idx, nused; 2492 size_t n, off, remaining; 2493 void *hva, *cookie; 2494 uint64_t gpa; 2495 2496 bzero(copyinfo, sizeof(struct vm_copyinfo) * num_copyinfo); 2497 2498 nused = 0; 2499 remaining = len; 2500 while (remaining > 0) { 2501 KASSERT(nused < num_copyinfo, ("insufficient vm_copyinfo")); 2502 error = vm_gla2gpa(vm, vcpuid, paging, gla, prot, &gpa, fault); 2503 if (error || *fault) 2504 return (error); 2505 off = gpa & PAGE_MASK; 2506 n = min(remaining, PAGE_SIZE - off); 2507 copyinfo[nused].gpa = gpa; 2508 copyinfo[nused].len = n; 2509 remaining -= n; 2510 gla += n; 2511 nused++; 2512 } 2513 2514 for (idx = 0; idx < nused; idx++) { 2515 hva = vm_gpa_hold(vm, vcpuid, copyinfo[idx].gpa, 2516 copyinfo[idx].len, prot, &cookie); 2517 if (hva == NULL) 2518 break; 2519 copyinfo[idx].hva = hva; 2520 copyinfo[idx].cookie = cookie; 2521 } 2522 2523 if (idx != nused) { 2524 vm_copy_teardown(vm, vcpuid, copyinfo, num_copyinfo); 2525 return (EFAULT); 2526 } else { 2527 *fault = 0; 2528 return (0); 2529 } 2530 } 2531 2532 void 2533 vm_copyin(struct vm *vm, int vcpuid, struct vm_copyinfo *copyinfo, void *kaddr, 2534 size_t len) 2535 { 2536 char *dst; 2537 int idx; 2538 2539 dst = kaddr; 2540 idx = 0; 2541 while (len > 0) { 2542 bcopy(copyinfo[idx].hva, dst, copyinfo[idx].len); 2543 len -= copyinfo[idx].len; 2544 dst += copyinfo[idx].len; 2545 idx++; 2546 } 2547 } 2548 2549 void 2550 vm_copyout(struct vm *vm, int vcpuid, const void *kaddr, 2551 struct vm_copyinfo *copyinfo, size_t len) 2552 { 2553 const char *src; 2554 int idx; 2555 2556 src = kaddr; 2557 idx = 0; 2558 while (len > 0) { 2559 bcopy(src, copyinfo[idx].hva, copyinfo[idx].len); 2560 len -= copyinfo[idx].len; 2561 src += copyinfo[idx].len; 2562 idx++; 2563 } 2564 } 2565 2566 /* 2567 * Return the amount of in-use and wired memory for the VM. Since 2568 * these are global stats, only return the values with for vCPU 0 2569 */ 2570 VMM_STAT_DECLARE(VMM_MEM_RESIDENT); 2571 VMM_STAT_DECLARE(VMM_MEM_WIRED); 2572 2573 static void 2574 vm_get_rescnt(struct vm *vm, int vcpu, struct vmm_stat_type *stat) 2575 { 2576 2577 if (vcpu == 0) { 2578 vmm_stat_set(vm, vcpu, VMM_MEM_RESIDENT, 2579 PAGE_SIZE * vmspace_resident_count(vm->vmspace)); 2580 } 2581 } 2582 2583 static void 2584 vm_get_wiredcnt(struct vm *vm, int vcpu, struct vmm_stat_type *stat) 2585 { 2586 2587 if (vcpu == 0) { 2588 vmm_stat_set(vm, vcpu, VMM_MEM_WIRED, 2589 PAGE_SIZE * pmap_wired_count(vmspace_pmap(vm->vmspace))); 2590 } 2591 } 2592 2593 VMM_STAT_FUNC(VMM_MEM_RESIDENT, "Resident memory", vm_get_rescnt); 2594 VMM_STAT_FUNC(VMM_MEM_WIRED, "Wired memory", vm_get_wiredcnt); 2595