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 static int vmm_ipinum; 219 SYSCTL_INT(_hw_vmm, OID_AUTO, ipinum, CTLFLAG_RD, &vmm_ipinum, 0, 220 "IPI vector used for vcpu notifications"); 221 222 static int trace_guest_exceptions; 223 SYSCTL_INT(_hw_vmm, OID_AUTO, trace_guest_exceptions, CTLFLAG_RDTUN, 224 &trace_guest_exceptions, 0, 225 "Trap into hypervisor on all guest exceptions and reflect them back"); 226 227 static int vmm_force_iommu = 0; 228 TUNABLE_INT("hw.vmm.force_iommu", &vmm_force_iommu); 229 SYSCTL_INT(_hw_vmm, OID_AUTO, force_iommu, CTLFLAG_RDTUN, &vmm_force_iommu, 0, 230 "Force use of I/O MMU even if no passthrough devices were found."); 231 232 static void vm_free_memmap(struct vm *vm, int ident); 233 static bool sysmem_mapping(struct vm *vm, struct mem_map *mm); 234 static void vcpu_notify_event_locked(struct vcpu *vcpu, bool lapic_intr); 235 236 #ifdef KTR 237 static const char * 238 vcpu_state2str(enum vcpu_state state) 239 { 240 241 switch (state) { 242 case VCPU_IDLE: 243 return ("idle"); 244 case VCPU_FROZEN: 245 return ("frozen"); 246 case VCPU_RUNNING: 247 return ("running"); 248 case VCPU_SLEEPING: 249 return ("sleeping"); 250 default: 251 return ("unknown"); 252 } 253 } 254 #endif 255 256 static void 257 vcpu_cleanup(struct vm *vm, int i, bool destroy) 258 { 259 struct vcpu *vcpu = &vm->vcpu[i]; 260 261 VLAPIC_CLEANUP(vm->cookie, vcpu->vlapic); 262 if (destroy) { 263 vmm_stat_free(vcpu->stats); 264 fpu_save_area_free(vcpu->guestfpu); 265 } 266 } 267 268 static void 269 vcpu_init(struct vm *vm, int vcpu_id, bool create) 270 { 271 struct vcpu *vcpu; 272 273 KASSERT(vcpu_id >= 0 && vcpu_id < VM_MAXCPU, 274 ("vcpu_init: invalid vcpu %d", vcpu_id)); 275 276 vcpu = &vm->vcpu[vcpu_id]; 277 278 if (create) { 279 KASSERT(!vcpu_lock_initialized(vcpu), ("vcpu %d already " 280 "initialized", vcpu_id)); 281 vcpu_lock_init(vcpu); 282 vcpu->state = VCPU_IDLE; 283 vcpu->hostcpu = NOCPU; 284 vcpu->guestfpu = fpu_save_area_alloc(); 285 vcpu->stats = vmm_stat_alloc(); 286 } 287 288 vcpu->vlapic = VLAPIC_INIT(vm->cookie, vcpu_id); 289 vm_set_x2apic_state(vm, vcpu_id, X2APIC_DISABLED); 290 vcpu->reqidle = 0; 291 vcpu->exitintinfo = 0; 292 vcpu->nmi_pending = 0; 293 vcpu->extint_pending = 0; 294 vcpu->exception_pending = 0; 295 vcpu->guest_xcr0 = XFEATURE_ENABLED_X87; 296 fpu_save_area_reset(vcpu->guestfpu); 297 vmm_stat_init(vcpu->stats); 298 } 299 300 int 301 vcpu_trace_exceptions(struct vm *vm, int vcpuid) 302 { 303 304 return (trace_guest_exceptions); 305 } 306 307 struct vm_exit * 308 vm_exitinfo(struct vm *vm, int cpuid) 309 { 310 struct vcpu *vcpu; 311 312 if (cpuid < 0 || cpuid >= VM_MAXCPU) 313 panic("vm_exitinfo: invalid cpuid %d", cpuid); 314 315 vcpu = &vm->vcpu[cpuid]; 316 317 return (&vcpu->exitinfo); 318 } 319 320 static void 321 vmm_resume(void) 322 { 323 VMM_RESUME(); 324 } 325 326 static int 327 vmm_init(void) 328 { 329 int error; 330 331 vmm_host_state_init(); 332 333 vmm_ipinum = lapic_ipi_alloc(&IDTVEC(justreturn)); 334 if (vmm_ipinum < 0) 335 vmm_ipinum = IPI_AST; 336 337 error = vmm_mem_init(); 338 if (error) 339 return (error); 340 341 if (vmm_is_intel()) 342 ops = &vmm_ops_intel; 343 else if (vmm_is_amd()) 344 ops = &vmm_ops_amd; 345 else 346 return (ENXIO); 347 348 vmm_resume_p = vmm_resume; 349 350 return (VMM_INIT(vmm_ipinum)); 351 } 352 353 static int 354 vmm_handler(module_t mod, int what, void *arg) 355 { 356 int error; 357 358 switch (what) { 359 case MOD_LOAD: 360 vmmdev_init(); 361 if (vmm_force_iommu || ppt_avail_devices() > 0) 362 iommu_init(); 363 error = vmm_init(); 364 if (error == 0) 365 vmm_initialized = 1; 366 break; 367 case MOD_UNLOAD: 368 error = vmmdev_cleanup(); 369 if (error == 0) { 370 vmm_resume_p = NULL; 371 iommu_cleanup(); 372 if (vmm_ipinum != IPI_AST) 373 lapic_ipi_free(vmm_ipinum); 374 error = VMM_CLEANUP(); 375 /* 376 * Something bad happened - prevent new 377 * VMs from being created 378 */ 379 if (error) 380 vmm_initialized = 0; 381 } 382 break; 383 default: 384 error = 0; 385 break; 386 } 387 return (error); 388 } 389 390 static moduledata_t vmm_kmod = { 391 "vmm", 392 vmm_handler, 393 NULL 394 }; 395 396 /* 397 * vmm initialization has the following dependencies: 398 * 399 * - iommu initialization must happen after the pci passthru driver has had 400 * a chance to attach to any passthru devices (after SI_SUB_CONFIGURE). 401 * 402 * - VT-x initialization requires smp_rendezvous() and therefore must happen 403 * after SMP is fully functional (after SI_SUB_SMP). 404 */ 405 DECLARE_MODULE(vmm, vmm_kmod, SI_SUB_SMP + 1, SI_ORDER_ANY); 406 MODULE_VERSION(vmm, 1); 407 408 static void 409 vm_init(struct vm *vm, bool create) 410 { 411 int i; 412 413 vm->cookie = VMINIT(vm, vmspace_pmap(vm->vmspace)); 414 vm->iommu = NULL; 415 vm->vioapic = vioapic_init(vm); 416 vm->vhpet = vhpet_init(vm); 417 vm->vatpic = vatpic_init(vm); 418 vm->vatpit = vatpit_init(vm); 419 vm->vpmtmr = vpmtmr_init(vm); 420 if (create) 421 vm->vrtc = vrtc_init(vm); 422 423 CPU_ZERO(&vm->active_cpus); 424 425 vm->suspend = 0; 426 CPU_ZERO(&vm->suspended_cpus); 427 428 for (i = 0; i < VM_MAXCPU; i++) 429 vcpu_init(vm, i, create); 430 } 431 432 int 433 vm_create(const char *name, struct vm **retvm) 434 { 435 struct vm *vm; 436 struct vmspace *vmspace; 437 438 /* 439 * If vmm.ko could not be successfully initialized then don't attempt 440 * to create the virtual machine. 441 */ 442 if (!vmm_initialized) 443 return (ENXIO); 444 445 if (name == NULL || strlen(name) >= VM_MAX_NAMELEN) 446 return (EINVAL); 447 448 vmspace = VMSPACE_ALLOC(0, VM_MAXUSER_ADDRESS); 449 if (vmspace == NULL) 450 return (ENOMEM); 451 452 vm = malloc(sizeof(struct vm), M_VM, M_WAITOK | M_ZERO); 453 strcpy(vm->name, name); 454 vm->vmspace = vmspace; 455 mtx_init(&vm->rendezvous_mtx, "vm rendezvous lock", 0, MTX_DEF); 456 457 vm_init(vm, true); 458 459 *retvm = vm; 460 return (0); 461 } 462 463 static void 464 vm_cleanup(struct vm *vm, bool destroy) 465 { 466 struct mem_map *mm; 467 int i; 468 469 ppt_unassign_all(vm); 470 471 if (vm->iommu != NULL) 472 iommu_destroy_domain(vm->iommu); 473 474 if (destroy) 475 vrtc_cleanup(vm->vrtc); 476 else 477 vrtc_reset(vm->vrtc); 478 vpmtmr_cleanup(vm->vpmtmr); 479 vatpit_cleanup(vm->vatpit); 480 vhpet_cleanup(vm->vhpet); 481 vatpic_cleanup(vm->vatpic); 482 vioapic_cleanup(vm->vioapic); 483 484 for (i = 0; i < VM_MAXCPU; i++) 485 vcpu_cleanup(vm, i, destroy); 486 487 VMCLEANUP(vm->cookie); 488 489 /* 490 * System memory is removed from the guest address space only when 491 * the VM is destroyed. This is because the mapping remains the same 492 * across VM reset. 493 * 494 * Device memory can be relocated by the guest (e.g. using PCI BARs) 495 * so those mappings are removed on a VM reset. 496 */ 497 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 498 mm = &vm->mem_maps[i]; 499 if (destroy || !sysmem_mapping(vm, mm)) 500 vm_free_memmap(vm, i); 501 } 502 503 if (destroy) { 504 for (i = 0; i < VM_MAX_MEMSEGS; i++) 505 vm_free_memseg(vm, i); 506 507 VMSPACE_FREE(vm->vmspace); 508 vm->vmspace = NULL; 509 } 510 } 511 512 void 513 vm_destroy(struct vm *vm) 514 { 515 vm_cleanup(vm, true); 516 free(vm, M_VM); 517 } 518 519 int 520 vm_reinit(struct vm *vm) 521 { 522 int error; 523 524 /* 525 * A virtual machine can be reset only if all vcpus are suspended. 526 */ 527 if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) { 528 vm_cleanup(vm, false); 529 vm_init(vm, false); 530 error = 0; 531 } else { 532 error = EBUSY; 533 } 534 535 return (error); 536 } 537 538 const char * 539 vm_name(struct vm *vm) 540 { 541 return (vm->name); 542 } 543 544 int 545 vm_map_mmio(struct vm *vm, vm_paddr_t gpa, size_t len, vm_paddr_t hpa) 546 { 547 vm_object_t obj; 548 549 if ((obj = vmm_mmio_alloc(vm->vmspace, gpa, len, hpa)) == NULL) 550 return (ENOMEM); 551 else 552 return (0); 553 } 554 555 int 556 vm_unmap_mmio(struct vm *vm, vm_paddr_t gpa, size_t len) 557 { 558 559 vmm_mmio_free(vm->vmspace, gpa, len); 560 return (0); 561 } 562 563 /* 564 * Return 'true' if 'gpa' is allocated in the guest address space. 565 * 566 * This function is called in the context of a running vcpu which acts as 567 * an implicit lock on 'vm->mem_maps[]'. 568 */ 569 bool 570 vm_mem_allocated(struct vm *vm, int vcpuid, vm_paddr_t gpa) 571 { 572 struct mem_map *mm; 573 int i; 574 575 #ifdef INVARIANTS 576 int hostcpu, state; 577 state = vcpu_get_state(vm, vcpuid, &hostcpu); 578 KASSERT(state == VCPU_RUNNING && hostcpu == curcpu, 579 ("%s: invalid vcpu state %d/%d", __func__, state, hostcpu)); 580 #endif 581 582 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 583 mm = &vm->mem_maps[i]; 584 if (mm->len != 0 && gpa >= mm->gpa && gpa < mm->gpa + mm->len) 585 return (true); /* 'gpa' is sysmem or devmem */ 586 } 587 588 if (ppt_is_mmio(vm, gpa)) 589 return (true); /* 'gpa' is pci passthru mmio */ 590 591 return (false); 592 } 593 594 int 595 vm_alloc_memseg(struct vm *vm, int ident, size_t len, bool sysmem) 596 { 597 struct mem_seg *seg; 598 vm_object_t obj; 599 600 if (ident < 0 || ident >= VM_MAX_MEMSEGS) 601 return (EINVAL); 602 603 if (len == 0 || (len & PAGE_MASK)) 604 return (EINVAL); 605 606 seg = &vm->mem_segs[ident]; 607 if (seg->object != NULL) { 608 if (seg->len == len && seg->sysmem == sysmem) 609 return (EEXIST); 610 else 611 return (EINVAL); 612 } 613 614 obj = vm_object_allocate(OBJT_DEFAULT, len >> PAGE_SHIFT); 615 if (obj == NULL) 616 return (ENOMEM); 617 618 seg->len = len; 619 seg->object = obj; 620 seg->sysmem = sysmem; 621 return (0); 622 } 623 624 int 625 vm_get_memseg(struct vm *vm, int ident, size_t *len, bool *sysmem, 626 vm_object_t *objptr) 627 { 628 struct mem_seg *seg; 629 630 if (ident < 0 || ident >= VM_MAX_MEMSEGS) 631 return (EINVAL); 632 633 seg = &vm->mem_segs[ident]; 634 if (len) 635 *len = seg->len; 636 if (sysmem) 637 *sysmem = seg->sysmem; 638 if (objptr) 639 *objptr = seg->object; 640 return (0); 641 } 642 643 void 644 vm_free_memseg(struct vm *vm, int ident) 645 { 646 struct mem_seg *seg; 647 648 KASSERT(ident >= 0 && ident < VM_MAX_MEMSEGS, 649 ("%s: invalid memseg ident %d", __func__, ident)); 650 651 seg = &vm->mem_segs[ident]; 652 if (seg->object != NULL) { 653 vm_object_deallocate(seg->object); 654 bzero(seg, sizeof(struct mem_seg)); 655 } 656 } 657 658 int 659 vm_mmap_memseg(struct vm *vm, vm_paddr_t gpa, int segid, vm_ooffset_t first, 660 size_t len, int prot, int flags) 661 { 662 struct mem_seg *seg; 663 struct mem_map *m, *map; 664 vm_ooffset_t last; 665 int i, error; 666 667 if (prot == 0 || (prot & ~(VM_PROT_ALL)) != 0) 668 return (EINVAL); 669 670 if (flags & ~VM_MEMMAP_F_WIRED) 671 return (EINVAL); 672 673 if (segid < 0 || segid >= VM_MAX_MEMSEGS) 674 return (EINVAL); 675 676 seg = &vm->mem_segs[segid]; 677 if (seg->object == NULL) 678 return (EINVAL); 679 680 last = first + len; 681 if (first < 0 || first >= last || last > seg->len) 682 return (EINVAL); 683 684 if ((gpa | first | last) & PAGE_MASK) 685 return (EINVAL); 686 687 map = NULL; 688 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 689 m = &vm->mem_maps[i]; 690 if (m->len == 0) { 691 map = m; 692 break; 693 } 694 } 695 696 if (map == NULL) 697 return (ENOSPC); 698 699 error = vm_map_find(&vm->vmspace->vm_map, seg->object, first, &gpa, 700 len, 0, VMFS_NO_SPACE, prot, prot, 0); 701 if (error != KERN_SUCCESS) 702 return (EFAULT); 703 704 vm_object_reference(seg->object); 705 706 if (flags & VM_MEMMAP_F_WIRED) { 707 error = vm_map_wire(&vm->vmspace->vm_map, gpa, gpa + len, 708 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES); 709 if (error != KERN_SUCCESS) { 710 vm_map_remove(&vm->vmspace->vm_map, gpa, gpa + len); 711 return (EFAULT); 712 } 713 } 714 715 map->gpa = gpa; 716 map->len = len; 717 map->segoff = first; 718 map->segid = segid; 719 map->prot = prot; 720 map->flags = flags; 721 return (0); 722 } 723 724 int 725 vm_mmap_getnext(struct vm *vm, vm_paddr_t *gpa, int *segid, 726 vm_ooffset_t *segoff, size_t *len, int *prot, int *flags) 727 { 728 struct mem_map *mm, *mmnext; 729 int i; 730 731 mmnext = NULL; 732 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 733 mm = &vm->mem_maps[i]; 734 if (mm->len == 0 || mm->gpa < *gpa) 735 continue; 736 if (mmnext == NULL || mm->gpa < mmnext->gpa) 737 mmnext = mm; 738 } 739 740 if (mmnext != NULL) { 741 *gpa = mmnext->gpa; 742 if (segid) 743 *segid = mmnext->segid; 744 if (segoff) 745 *segoff = mmnext->segoff; 746 if (len) 747 *len = mmnext->len; 748 if (prot) 749 *prot = mmnext->prot; 750 if (flags) 751 *flags = mmnext->flags; 752 return (0); 753 } else { 754 return (ENOENT); 755 } 756 } 757 758 static void 759 vm_free_memmap(struct vm *vm, int ident) 760 { 761 struct mem_map *mm; 762 int error; 763 764 mm = &vm->mem_maps[ident]; 765 if (mm->len) { 766 error = vm_map_remove(&vm->vmspace->vm_map, mm->gpa, 767 mm->gpa + mm->len); 768 KASSERT(error == KERN_SUCCESS, ("%s: vm_map_remove error %d", 769 __func__, error)); 770 bzero(mm, sizeof(struct mem_map)); 771 } 772 } 773 774 static __inline bool 775 sysmem_mapping(struct vm *vm, struct mem_map *mm) 776 { 777 778 if (mm->len != 0 && vm->mem_segs[mm->segid].sysmem) 779 return (true); 780 else 781 return (false); 782 } 783 784 static vm_paddr_t 785 sysmem_maxaddr(struct vm *vm) 786 { 787 struct mem_map *mm; 788 vm_paddr_t maxaddr; 789 int i; 790 791 maxaddr = 0; 792 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 793 mm = &vm->mem_maps[i]; 794 if (sysmem_mapping(vm, mm)) { 795 if (maxaddr < mm->gpa + mm->len) 796 maxaddr = mm->gpa + mm->len; 797 } 798 } 799 return (maxaddr); 800 } 801 802 static void 803 vm_iommu_modify(struct vm *vm, boolean_t map) 804 { 805 int i, sz; 806 vm_paddr_t gpa, hpa; 807 struct mem_map *mm; 808 void *vp, *cookie, *host_domain; 809 810 sz = PAGE_SIZE; 811 host_domain = iommu_host_domain(); 812 813 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 814 mm = &vm->mem_maps[i]; 815 if (!sysmem_mapping(vm, mm)) 816 continue; 817 818 if (map) { 819 KASSERT((mm->flags & VM_MEMMAP_F_IOMMU) == 0, 820 ("iommu map found invalid memmap %#lx/%#lx/%#x", 821 mm->gpa, mm->len, mm->flags)); 822 if ((mm->flags & VM_MEMMAP_F_WIRED) == 0) 823 continue; 824 mm->flags |= VM_MEMMAP_F_IOMMU; 825 } else { 826 if ((mm->flags & VM_MEMMAP_F_IOMMU) == 0) 827 continue; 828 mm->flags &= ~VM_MEMMAP_F_IOMMU; 829 KASSERT((mm->flags & VM_MEMMAP_F_WIRED) != 0, 830 ("iommu unmap found invalid memmap %#lx/%#lx/%#x", 831 mm->gpa, mm->len, mm->flags)); 832 } 833 834 gpa = mm->gpa; 835 while (gpa < mm->gpa + mm->len) { 836 vp = vm_gpa_hold(vm, -1, gpa, PAGE_SIZE, VM_PROT_WRITE, 837 &cookie); 838 KASSERT(vp != NULL, ("vm(%s) could not map gpa %#lx", 839 vm_name(vm), gpa)); 840 841 vm_gpa_release(cookie); 842 843 hpa = DMAP_TO_PHYS((uintptr_t)vp); 844 if (map) { 845 iommu_create_mapping(vm->iommu, gpa, hpa, sz); 846 iommu_remove_mapping(host_domain, hpa, sz); 847 } else { 848 iommu_remove_mapping(vm->iommu, gpa, sz); 849 iommu_create_mapping(host_domain, hpa, hpa, sz); 850 } 851 852 gpa += PAGE_SIZE; 853 } 854 } 855 856 /* 857 * Invalidate the cached translations associated with the domain 858 * from which pages were removed. 859 */ 860 if (map) 861 iommu_invalidate_tlb(host_domain); 862 else 863 iommu_invalidate_tlb(vm->iommu); 864 } 865 866 #define vm_iommu_unmap(vm) vm_iommu_modify((vm), FALSE) 867 #define vm_iommu_map(vm) vm_iommu_modify((vm), TRUE) 868 869 int 870 vm_unassign_pptdev(struct vm *vm, int bus, int slot, int func) 871 { 872 int error; 873 874 error = ppt_unassign_device(vm, bus, slot, func); 875 if (error) 876 return (error); 877 878 if (ppt_assigned_devices(vm) == 0) 879 vm_iommu_unmap(vm); 880 881 return (0); 882 } 883 884 int 885 vm_assign_pptdev(struct vm *vm, int bus, int slot, int func) 886 { 887 int error; 888 vm_paddr_t maxaddr; 889 890 /* Set up the IOMMU to do the 'gpa' to 'hpa' translation */ 891 if (ppt_assigned_devices(vm) == 0) { 892 KASSERT(vm->iommu == NULL, 893 ("vm_assign_pptdev: iommu must be NULL")); 894 maxaddr = sysmem_maxaddr(vm); 895 vm->iommu = iommu_create_domain(maxaddr); 896 vm_iommu_map(vm); 897 } 898 899 error = ppt_assign_device(vm, bus, slot, func); 900 return (error); 901 } 902 903 void * 904 vm_gpa_hold(struct vm *vm, int vcpuid, vm_paddr_t gpa, size_t len, int reqprot, 905 void **cookie) 906 { 907 int i, count, pageoff; 908 struct mem_map *mm; 909 vm_page_t m; 910 #ifdef INVARIANTS 911 /* 912 * All vcpus are frozen by ioctls that modify the memory map 913 * (e.g. VM_MMAP_MEMSEG). Therefore 'vm->memmap[]' stability is 914 * guaranteed if at least one vcpu is in the VCPU_FROZEN state. 915 */ 916 int state; 917 KASSERT(vcpuid >= -1 || vcpuid < VM_MAXCPU, ("%s: invalid vcpuid %d", 918 __func__, vcpuid)); 919 for (i = 0; i < VM_MAXCPU; i++) { 920 if (vcpuid != -1 && vcpuid != i) 921 continue; 922 state = vcpu_get_state(vm, i, NULL); 923 KASSERT(state == VCPU_FROZEN, ("%s: invalid vcpu state %d", 924 __func__, state)); 925 } 926 #endif 927 pageoff = gpa & PAGE_MASK; 928 if (len > PAGE_SIZE - pageoff) 929 panic("vm_gpa_hold: invalid gpa/len: 0x%016lx/%lu", gpa, len); 930 931 count = 0; 932 for (i = 0; i < VM_MAX_MEMMAPS; i++) { 933 mm = &vm->mem_maps[i]; 934 if (sysmem_mapping(vm, mm) && gpa >= mm->gpa && 935 gpa < mm->gpa + mm->len) { 936 count = vm_fault_quick_hold_pages(&vm->vmspace->vm_map, 937 trunc_page(gpa), PAGE_SIZE, reqprot, &m, 1); 938 break; 939 } 940 } 941 942 if (count == 1) { 943 *cookie = m; 944 return ((void *)(PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m)) + pageoff)); 945 } else { 946 *cookie = NULL; 947 return (NULL); 948 } 949 } 950 951 void 952 vm_gpa_release(void *cookie) 953 { 954 vm_page_t m = cookie; 955 956 vm_page_lock(m); 957 vm_page_unhold(m); 958 vm_page_unlock(m); 959 } 960 961 int 962 vm_get_register(struct vm *vm, int vcpu, int reg, uint64_t *retval) 963 { 964 965 if (vcpu < 0 || vcpu >= VM_MAXCPU) 966 return (EINVAL); 967 968 if (reg >= VM_REG_LAST) 969 return (EINVAL); 970 971 return (VMGETREG(vm->cookie, vcpu, reg, retval)); 972 } 973 974 int 975 vm_set_register(struct vm *vm, int vcpuid, int reg, uint64_t val) 976 { 977 struct vcpu *vcpu; 978 int error; 979 980 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 981 return (EINVAL); 982 983 if (reg >= VM_REG_LAST) 984 return (EINVAL); 985 986 error = VMSETREG(vm->cookie, vcpuid, reg, val); 987 if (error || reg != VM_REG_GUEST_RIP) 988 return (error); 989 990 /* Set 'nextrip' to match the value of %rip */ 991 VCPU_CTR1(vm, vcpuid, "Setting nextrip to %#lx", val); 992 vcpu = &vm->vcpu[vcpuid]; 993 vcpu->nextrip = val; 994 return (0); 995 } 996 997 static boolean_t 998 is_descriptor_table(int reg) 999 { 1000 1001 switch (reg) { 1002 case VM_REG_GUEST_IDTR: 1003 case VM_REG_GUEST_GDTR: 1004 return (TRUE); 1005 default: 1006 return (FALSE); 1007 } 1008 } 1009 1010 static boolean_t 1011 is_segment_register(int reg) 1012 { 1013 1014 switch (reg) { 1015 case VM_REG_GUEST_ES: 1016 case VM_REG_GUEST_CS: 1017 case VM_REG_GUEST_SS: 1018 case VM_REG_GUEST_DS: 1019 case VM_REG_GUEST_FS: 1020 case VM_REG_GUEST_GS: 1021 case VM_REG_GUEST_TR: 1022 case VM_REG_GUEST_LDTR: 1023 return (TRUE); 1024 default: 1025 return (FALSE); 1026 } 1027 } 1028 1029 int 1030 vm_get_seg_desc(struct vm *vm, int vcpu, int reg, 1031 struct seg_desc *desc) 1032 { 1033 1034 if (vcpu < 0 || vcpu >= VM_MAXCPU) 1035 return (EINVAL); 1036 1037 if (!is_segment_register(reg) && !is_descriptor_table(reg)) 1038 return (EINVAL); 1039 1040 return (VMGETDESC(vm->cookie, vcpu, reg, desc)); 1041 } 1042 1043 int 1044 vm_set_seg_desc(struct vm *vm, int vcpu, int reg, 1045 struct seg_desc *desc) 1046 { 1047 if (vcpu < 0 || vcpu >= VM_MAXCPU) 1048 return (EINVAL); 1049 1050 if (!is_segment_register(reg) && !is_descriptor_table(reg)) 1051 return (EINVAL); 1052 1053 return (VMSETDESC(vm->cookie, vcpu, reg, desc)); 1054 } 1055 1056 static void 1057 restore_guest_fpustate(struct vcpu *vcpu) 1058 { 1059 1060 /* flush host state to the pcb */ 1061 fpuexit(curthread); 1062 1063 /* restore guest FPU state */ 1064 fpu_stop_emulating(); 1065 fpurestore(vcpu->guestfpu); 1066 1067 /* restore guest XCR0 if XSAVE is enabled in the host */ 1068 if (rcr4() & CR4_XSAVE) 1069 load_xcr(0, vcpu->guest_xcr0); 1070 1071 /* 1072 * The FPU is now "dirty" with the guest's state so turn on emulation 1073 * to trap any access to the FPU by the host. 1074 */ 1075 fpu_start_emulating(); 1076 } 1077 1078 static void 1079 save_guest_fpustate(struct vcpu *vcpu) 1080 { 1081 1082 if ((rcr0() & CR0_TS) == 0) 1083 panic("fpu emulation not enabled in host!"); 1084 1085 /* save guest XCR0 and restore host XCR0 */ 1086 if (rcr4() & CR4_XSAVE) { 1087 vcpu->guest_xcr0 = rxcr(0); 1088 load_xcr(0, vmm_get_host_xcr0()); 1089 } 1090 1091 /* save guest FPU state */ 1092 fpu_stop_emulating(); 1093 fpusave(vcpu->guestfpu); 1094 fpu_start_emulating(); 1095 } 1096 1097 static VMM_STAT(VCPU_IDLE_TICKS, "number of ticks vcpu was idle"); 1098 1099 static int 1100 vcpu_set_state_locked(struct vm *vm, int vcpuid, enum vcpu_state newstate, 1101 bool from_idle) 1102 { 1103 struct vcpu *vcpu; 1104 int error; 1105 1106 vcpu = &vm->vcpu[vcpuid]; 1107 vcpu_assert_locked(vcpu); 1108 1109 /* 1110 * State transitions from the vmmdev_ioctl() must always begin from 1111 * the VCPU_IDLE state. This guarantees that there is only a single 1112 * ioctl() operating on a vcpu at any point. 1113 */ 1114 if (from_idle) { 1115 while (vcpu->state != VCPU_IDLE) { 1116 vcpu->reqidle = 1; 1117 vcpu_notify_event_locked(vcpu, false); 1118 VCPU_CTR1(vm, vcpuid, "vcpu state change from %s to " 1119 "idle requested", vcpu_state2str(vcpu->state)); 1120 msleep_spin(&vcpu->state, &vcpu->mtx, "vmstat", hz); 1121 } 1122 } else { 1123 KASSERT(vcpu->state != VCPU_IDLE, ("invalid transition from " 1124 "vcpu idle state")); 1125 } 1126 1127 if (vcpu->state == VCPU_RUNNING) { 1128 KASSERT(vcpu->hostcpu == curcpu, ("curcpu %d and hostcpu %d " 1129 "mismatch for running vcpu", curcpu, vcpu->hostcpu)); 1130 } else { 1131 KASSERT(vcpu->hostcpu == NOCPU, ("Invalid hostcpu %d for a " 1132 "vcpu that is not running", vcpu->hostcpu)); 1133 } 1134 1135 /* 1136 * The following state transitions are allowed: 1137 * IDLE -> FROZEN -> IDLE 1138 * FROZEN -> RUNNING -> FROZEN 1139 * FROZEN -> SLEEPING -> FROZEN 1140 */ 1141 switch (vcpu->state) { 1142 case VCPU_IDLE: 1143 case VCPU_RUNNING: 1144 case VCPU_SLEEPING: 1145 error = (newstate != VCPU_FROZEN); 1146 break; 1147 case VCPU_FROZEN: 1148 error = (newstate == VCPU_FROZEN); 1149 break; 1150 default: 1151 error = 1; 1152 break; 1153 } 1154 1155 if (error) 1156 return (EBUSY); 1157 1158 VCPU_CTR2(vm, vcpuid, "vcpu state changed from %s to %s", 1159 vcpu_state2str(vcpu->state), vcpu_state2str(newstate)); 1160 1161 vcpu->state = newstate; 1162 if (newstate == VCPU_RUNNING) 1163 vcpu->hostcpu = curcpu; 1164 else 1165 vcpu->hostcpu = NOCPU; 1166 1167 if (newstate == VCPU_IDLE) 1168 wakeup(&vcpu->state); 1169 1170 return (0); 1171 } 1172 1173 static void 1174 vcpu_require_state(struct vm *vm, int vcpuid, enum vcpu_state newstate) 1175 { 1176 int error; 1177 1178 if ((error = vcpu_set_state(vm, vcpuid, newstate, false)) != 0) 1179 panic("Error %d setting state to %d\n", error, newstate); 1180 } 1181 1182 static void 1183 vcpu_require_state_locked(struct vm *vm, int vcpuid, enum vcpu_state newstate) 1184 { 1185 int error; 1186 1187 if ((error = vcpu_set_state_locked(vm, vcpuid, newstate, false)) != 0) 1188 panic("Error %d setting state to %d", error, newstate); 1189 } 1190 1191 static void 1192 vm_set_rendezvous_func(struct vm *vm, vm_rendezvous_func_t func) 1193 { 1194 1195 KASSERT(mtx_owned(&vm->rendezvous_mtx), ("rendezvous_mtx not locked")); 1196 1197 /* 1198 * Update 'rendezvous_func' and execute a write memory barrier to 1199 * ensure that it is visible across all host cpus. This is not needed 1200 * for correctness but it does ensure that all the vcpus will notice 1201 * that the rendezvous is requested immediately. 1202 */ 1203 vm->rendezvous_func = func; 1204 wmb(); 1205 } 1206 1207 #define RENDEZVOUS_CTR0(vm, vcpuid, fmt) \ 1208 do { \ 1209 if (vcpuid >= 0) \ 1210 VCPU_CTR0(vm, vcpuid, fmt); \ 1211 else \ 1212 VM_CTR0(vm, fmt); \ 1213 } while (0) 1214 1215 static void 1216 vm_handle_rendezvous(struct vm *vm, int vcpuid) 1217 { 1218 1219 KASSERT(vcpuid == -1 || (vcpuid >= 0 && vcpuid < VM_MAXCPU), 1220 ("vm_handle_rendezvous: invalid vcpuid %d", vcpuid)); 1221 1222 mtx_lock(&vm->rendezvous_mtx); 1223 while (vm->rendezvous_func != NULL) { 1224 /* 'rendezvous_req_cpus' must be a subset of 'active_cpus' */ 1225 CPU_AND(&vm->rendezvous_req_cpus, &vm->active_cpus); 1226 1227 if (vcpuid != -1 && 1228 CPU_ISSET(vcpuid, &vm->rendezvous_req_cpus) && 1229 !CPU_ISSET(vcpuid, &vm->rendezvous_done_cpus)) { 1230 VCPU_CTR0(vm, vcpuid, "Calling rendezvous func"); 1231 (*vm->rendezvous_func)(vm, vcpuid, vm->rendezvous_arg); 1232 CPU_SET(vcpuid, &vm->rendezvous_done_cpus); 1233 } 1234 if (CPU_CMP(&vm->rendezvous_req_cpus, 1235 &vm->rendezvous_done_cpus) == 0) { 1236 VCPU_CTR0(vm, vcpuid, "Rendezvous completed"); 1237 vm_set_rendezvous_func(vm, NULL); 1238 wakeup(&vm->rendezvous_func); 1239 break; 1240 } 1241 RENDEZVOUS_CTR0(vm, vcpuid, "Wait for rendezvous completion"); 1242 mtx_sleep(&vm->rendezvous_func, &vm->rendezvous_mtx, 0, 1243 "vmrndv", 0); 1244 } 1245 mtx_unlock(&vm->rendezvous_mtx); 1246 } 1247 1248 /* 1249 * Emulate a guest 'hlt' by sleeping until the vcpu is ready to run. 1250 */ 1251 static int 1252 vm_handle_hlt(struct vm *vm, int vcpuid, bool intr_disabled, bool *retu) 1253 { 1254 struct vcpu *vcpu; 1255 const char *wmesg; 1256 int t, vcpu_halted, vm_halted; 1257 1258 KASSERT(!CPU_ISSET(vcpuid, &vm->halted_cpus), ("vcpu already halted")); 1259 1260 vcpu = &vm->vcpu[vcpuid]; 1261 vcpu_halted = 0; 1262 vm_halted = 0; 1263 1264 vcpu_lock(vcpu); 1265 while (1) { 1266 /* 1267 * Do a final check for pending NMI or interrupts before 1268 * really putting this thread to sleep. Also check for 1269 * software events that would cause this vcpu to wakeup. 1270 * 1271 * These interrupts/events could have happened after the 1272 * vcpu returned from VMRUN() and before it acquired the 1273 * vcpu lock above. 1274 */ 1275 if (vm->rendezvous_func != NULL || vm->suspend || vcpu->reqidle) 1276 break; 1277 if (vm_nmi_pending(vm, vcpuid)) 1278 break; 1279 if (!intr_disabled) { 1280 if (vm_extint_pending(vm, vcpuid) || 1281 vlapic_pending_intr(vcpu->vlapic, NULL)) { 1282 break; 1283 } 1284 } 1285 1286 /* Don't go to sleep if the vcpu thread needs to yield */ 1287 if (vcpu_should_yield(vm, vcpuid)) 1288 break; 1289 1290 /* 1291 * Some Linux guests implement "halt" by having all vcpus 1292 * execute HLT with interrupts disabled. 'halted_cpus' keeps 1293 * track of the vcpus that have entered this state. When all 1294 * vcpus enter the halted state the virtual machine is halted. 1295 */ 1296 if (intr_disabled) { 1297 wmesg = "vmhalt"; 1298 VCPU_CTR0(vm, vcpuid, "Halted"); 1299 if (!vcpu_halted && halt_detection_enabled) { 1300 vcpu_halted = 1; 1301 CPU_SET_ATOMIC(vcpuid, &vm->halted_cpus); 1302 } 1303 if (CPU_CMP(&vm->halted_cpus, &vm->active_cpus) == 0) { 1304 vm_halted = 1; 1305 break; 1306 } 1307 } else { 1308 wmesg = "vmidle"; 1309 } 1310 1311 t = ticks; 1312 vcpu_require_state_locked(vm, vcpuid, VCPU_SLEEPING); 1313 /* 1314 * XXX msleep_spin() cannot be interrupted by signals so 1315 * wake up periodically to check pending signals. 1316 */ 1317 msleep_spin(vcpu, &vcpu->mtx, wmesg, hz); 1318 vcpu_require_state_locked(vm, vcpuid, VCPU_FROZEN); 1319 vmm_stat_incr(vm, vcpuid, VCPU_IDLE_TICKS, ticks - t); 1320 } 1321 1322 if (vcpu_halted) 1323 CPU_CLR_ATOMIC(vcpuid, &vm->halted_cpus); 1324 1325 vcpu_unlock(vcpu); 1326 1327 if (vm_halted) 1328 vm_suspend(vm, VM_SUSPEND_HALT); 1329 1330 return (0); 1331 } 1332 1333 static int 1334 vm_handle_paging(struct vm *vm, int vcpuid, bool *retu) 1335 { 1336 int rv, ftype; 1337 struct vm_map *map; 1338 struct vcpu *vcpu; 1339 struct vm_exit *vme; 1340 1341 vcpu = &vm->vcpu[vcpuid]; 1342 vme = &vcpu->exitinfo; 1343 1344 KASSERT(vme->inst_length == 0, ("%s: invalid inst_length %d", 1345 __func__, vme->inst_length)); 1346 1347 ftype = vme->u.paging.fault_type; 1348 KASSERT(ftype == VM_PROT_READ || 1349 ftype == VM_PROT_WRITE || ftype == VM_PROT_EXECUTE, 1350 ("vm_handle_paging: invalid fault_type %d", ftype)); 1351 1352 if (ftype == VM_PROT_READ || ftype == VM_PROT_WRITE) { 1353 rv = pmap_emulate_accessed_dirty(vmspace_pmap(vm->vmspace), 1354 vme->u.paging.gpa, ftype); 1355 if (rv == 0) { 1356 VCPU_CTR2(vm, vcpuid, "%s bit emulation for gpa %#lx", 1357 ftype == VM_PROT_READ ? "accessed" : "dirty", 1358 vme->u.paging.gpa); 1359 goto done; 1360 } 1361 } 1362 1363 map = &vm->vmspace->vm_map; 1364 rv = vm_fault(map, vme->u.paging.gpa, ftype, VM_FAULT_NORMAL); 1365 1366 VCPU_CTR3(vm, vcpuid, "vm_handle_paging rv = %d, gpa = %#lx, " 1367 "ftype = %d", rv, vme->u.paging.gpa, ftype); 1368 1369 if (rv != KERN_SUCCESS) 1370 return (EFAULT); 1371 done: 1372 return (0); 1373 } 1374 1375 static int 1376 vm_handle_inst_emul(struct vm *vm, int vcpuid, bool *retu) 1377 { 1378 struct vie *vie; 1379 struct vcpu *vcpu; 1380 struct vm_exit *vme; 1381 uint64_t gla, gpa, cs_base; 1382 struct vm_guest_paging *paging; 1383 mem_region_read_t mread; 1384 mem_region_write_t mwrite; 1385 enum vm_cpu_mode cpu_mode; 1386 int cs_d, error, fault; 1387 1388 vcpu = &vm->vcpu[vcpuid]; 1389 vme = &vcpu->exitinfo; 1390 1391 KASSERT(vme->inst_length == 0, ("%s: invalid inst_length %d", 1392 __func__, vme->inst_length)); 1393 1394 gla = vme->u.inst_emul.gla; 1395 gpa = vme->u.inst_emul.gpa; 1396 cs_base = vme->u.inst_emul.cs_base; 1397 cs_d = vme->u.inst_emul.cs_d; 1398 vie = &vme->u.inst_emul.vie; 1399 paging = &vme->u.inst_emul.paging; 1400 cpu_mode = paging->cpu_mode; 1401 1402 VCPU_CTR1(vm, vcpuid, "inst_emul fault accessing gpa %#lx", gpa); 1403 1404 /* Fetch, decode and emulate the faulting instruction */ 1405 if (vie->num_valid == 0) { 1406 error = vmm_fetch_instruction(vm, vcpuid, paging, vme->rip + 1407 cs_base, VIE_INST_SIZE, vie, &fault); 1408 } else { 1409 /* 1410 * The instruction bytes have already been copied into 'vie' 1411 */ 1412 error = fault = 0; 1413 } 1414 if (error || fault) 1415 return (error); 1416 1417 if (vmm_decode_instruction(vm, vcpuid, gla, cpu_mode, cs_d, vie) != 0) { 1418 VCPU_CTR1(vm, vcpuid, "Error decoding instruction at %#lx", 1419 vme->rip + cs_base); 1420 *retu = true; /* dump instruction bytes in userspace */ 1421 return (0); 1422 } 1423 1424 /* 1425 * Update 'nextrip' based on the length of the emulated instruction. 1426 */ 1427 vme->inst_length = vie->num_processed; 1428 vcpu->nextrip += vie->num_processed; 1429 VCPU_CTR1(vm, vcpuid, "nextrip updated to %#lx after instruction " 1430 "decoding", vcpu->nextrip); 1431 1432 /* return to userland unless this is an in-kernel emulated device */ 1433 if (gpa >= DEFAULT_APIC_BASE && gpa < DEFAULT_APIC_BASE + PAGE_SIZE) { 1434 mread = lapic_mmio_read; 1435 mwrite = lapic_mmio_write; 1436 } else if (gpa >= VIOAPIC_BASE && gpa < VIOAPIC_BASE + VIOAPIC_SIZE) { 1437 mread = vioapic_mmio_read; 1438 mwrite = vioapic_mmio_write; 1439 } else if (gpa >= VHPET_BASE && gpa < VHPET_BASE + VHPET_SIZE) { 1440 mread = vhpet_mmio_read; 1441 mwrite = vhpet_mmio_write; 1442 } else { 1443 *retu = true; 1444 return (0); 1445 } 1446 1447 error = vmm_emulate_instruction(vm, vcpuid, gpa, vie, paging, 1448 mread, mwrite, retu); 1449 1450 return (error); 1451 } 1452 1453 static int 1454 vm_handle_suspend(struct vm *vm, int vcpuid, bool *retu) 1455 { 1456 int i, done; 1457 struct vcpu *vcpu; 1458 1459 done = 0; 1460 vcpu = &vm->vcpu[vcpuid]; 1461 1462 CPU_SET_ATOMIC(vcpuid, &vm->suspended_cpus); 1463 1464 /* 1465 * Wait until all 'active_cpus' have suspended themselves. 1466 * 1467 * Since a VM may be suspended at any time including when one or 1468 * more vcpus are doing a rendezvous we need to call the rendezvous 1469 * handler while we are waiting to prevent a deadlock. 1470 */ 1471 vcpu_lock(vcpu); 1472 while (1) { 1473 if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) { 1474 VCPU_CTR0(vm, vcpuid, "All vcpus suspended"); 1475 break; 1476 } 1477 1478 if (vm->rendezvous_func == NULL) { 1479 VCPU_CTR0(vm, vcpuid, "Sleeping during suspend"); 1480 vcpu_require_state_locked(vm, vcpuid, VCPU_SLEEPING); 1481 msleep_spin(vcpu, &vcpu->mtx, "vmsusp", hz); 1482 vcpu_require_state_locked(vm, vcpuid, VCPU_FROZEN); 1483 } else { 1484 VCPU_CTR0(vm, vcpuid, "Rendezvous during suspend"); 1485 vcpu_unlock(vcpu); 1486 vm_handle_rendezvous(vm, vcpuid); 1487 vcpu_lock(vcpu); 1488 } 1489 } 1490 vcpu_unlock(vcpu); 1491 1492 /* 1493 * Wakeup the other sleeping vcpus and return to userspace. 1494 */ 1495 for (i = 0; i < VM_MAXCPU; i++) { 1496 if (CPU_ISSET(i, &vm->suspended_cpus)) { 1497 vcpu_notify_event(vm, i, false); 1498 } 1499 } 1500 1501 *retu = true; 1502 return (0); 1503 } 1504 1505 static int 1506 vm_handle_reqidle(struct vm *vm, int vcpuid, bool *retu) 1507 { 1508 struct vcpu *vcpu = &vm->vcpu[vcpuid]; 1509 1510 vcpu_lock(vcpu); 1511 KASSERT(vcpu->reqidle, ("invalid vcpu reqidle %d", vcpu->reqidle)); 1512 vcpu->reqidle = 0; 1513 vcpu_unlock(vcpu); 1514 *retu = true; 1515 return (0); 1516 } 1517 1518 int 1519 vm_suspend(struct vm *vm, enum vm_suspend_how how) 1520 { 1521 int i; 1522 1523 if (how <= VM_SUSPEND_NONE || how >= VM_SUSPEND_LAST) 1524 return (EINVAL); 1525 1526 if (atomic_cmpset_int(&vm->suspend, 0, how) == 0) { 1527 VM_CTR2(vm, "virtual machine already suspended %d/%d", 1528 vm->suspend, how); 1529 return (EALREADY); 1530 } 1531 1532 VM_CTR1(vm, "virtual machine successfully suspended %d", how); 1533 1534 /* 1535 * Notify all active vcpus that they are now suspended. 1536 */ 1537 for (i = 0; i < VM_MAXCPU; i++) { 1538 if (CPU_ISSET(i, &vm->active_cpus)) 1539 vcpu_notify_event(vm, i, false); 1540 } 1541 1542 return (0); 1543 } 1544 1545 void 1546 vm_exit_suspended(struct vm *vm, int vcpuid, uint64_t rip) 1547 { 1548 struct vm_exit *vmexit; 1549 1550 KASSERT(vm->suspend > VM_SUSPEND_NONE && vm->suspend < VM_SUSPEND_LAST, 1551 ("vm_exit_suspended: invalid suspend type %d", vm->suspend)); 1552 1553 vmexit = vm_exitinfo(vm, vcpuid); 1554 vmexit->rip = rip; 1555 vmexit->inst_length = 0; 1556 vmexit->exitcode = VM_EXITCODE_SUSPENDED; 1557 vmexit->u.suspended.how = vm->suspend; 1558 } 1559 1560 void 1561 vm_exit_rendezvous(struct vm *vm, int vcpuid, uint64_t rip) 1562 { 1563 struct vm_exit *vmexit; 1564 1565 KASSERT(vm->rendezvous_func != NULL, ("rendezvous not in progress")); 1566 1567 vmexit = vm_exitinfo(vm, vcpuid); 1568 vmexit->rip = rip; 1569 vmexit->inst_length = 0; 1570 vmexit->exitcode = VM_EXITCODE_RENDEZVOUS; 1571 vmm_stat_incr(vm, vcpuid, VMEXIT_RENDEZVOUS, 1); 1572 } 1573 1574 void 1575 vm_exit_reqidle(struct vm *vm, int vcpuid, uint64_t rip) 1576 { 1577 struct vm_exit *vmexit; 1578 1579 vmexit = vm_exitinfo(vm, vcpuid); 1580 vmexit->rip = rip; 1581 vmexit->inst_length = 0; 1582 vmexit->exitcode = VM_EXITCODE_REQIDLE; 1583 vmm_stat_incr(vm, vcpuid, VMEXIT_REQIDLE, 1); 1584 } 1585 1586 void 1587 vm_exit_astpending(struct vm *vm, int vcpuid, uint64_t rip) 1588 { 1589 struct vm_exit *vmexit; 1590 1591 vmexit = vm_exitinfo(vm, vcpuid); 1592 vmexit->rip = rip; 1593 vmexit->inst_length = 0; 1594 vmexit->exitcode = VM_EXITCODE_BOGUS; 1595 vmm_stat_incr(vm, vcpuid, VMEXIT_ASTPENDING, 1); 1596 } 1597 1598 int 1599 vm_run(struct vm *vm, struct vm_run *vmrun) 1600 { 1601 struct vm_eventinfo evinfo; 1602 int error, vcpuid; 1603 struct vcpu *vcpu; 1604 struct pcb *pcb; 1605 uint64_t tscval; 1606 struct vm_exit *vme; 1607 bool retu, intr_disabled; 1608 pmap_t pmap; 1609 1610 vcpuid = vmrun->cpuid; 1611 1612 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1613 return (EINVAL); 1614 1615 if (!CPU_ISSET(vcpuid, &vm->active_cpus)) 1616 return (EINVAL); 1617 1618 if (CPU_ISSET(vcpuid, &vm->suspended_cpus)) 1619 return (EINVAL); 1620 1621 pmap = vmspace_pmap(vm->vmspace); 1622 vcpu = &vm->vcpu[vcpuid]; 1623 vme = &vcpu->exitinfo; 1624 evinfo.rptr = &vm->rendezvous_func; 1625 evinfo.sptr = &vm->suspend; 1626 evinfo.iptr = &vcpu->reqidle; 1627 restart: 1628 critical_enter(); 1629 1630 KASSERT(!CPU_ISSET(curcpu, &pmap->pm_active), 1631 ("vm_run: absurd pm_active")); 1632 1633 tscval = rdtsc(); 1634 1635 pcb = PCPU_GET(curpcb); 1636 set_pcb_flags(pcb, PCB_FULL_IRET); 1637 1638 restore_guest_fpustate(vcpu); 1639 1640 vcpu_require_state(vm, vcpuid, VCPU_RUNNING); 1641 error = VMRUN(vm->cookie, vcpuid, vcpu->nextrip, pmap, &evinfo); 1642 vcpu_require_state(vm, vcpuid, VCPU_FROZEN); 1643 1644 save_guest_fpustate(vcpu); 1645 1646 vmm_stat_incr(vm, vcpuid, VCPU_TOTAL_RUNTIME, rdtsc() - tscval); 1647 1648 critical_exit(); 1649 1650 if (error == 0) { 1651 retu = false; 1652 vcpu->nextrip = vme->rip + vme->inst_length; 1653 switch (vme->exitcode) { 1654 case VM_EXITCODE_REQIDLE: 1655 error = vm_handle_reqidle(vm, vcpuid, &retu); 1656 break; 1657 case VM_EXITCODE_SUSPENDED: 1658 error = vm_handle_suspend(vm, vcpuid, &retu); 1659 break; 1660 case VM_EXITCODE_IOAPIC_EOI: 1661 vioapic_process_eoi(vm, vcpuid, 1662 vme->u.ioapic_eoi.vector); 1663 break; 1664 case VM_EXITCODE_RENDEZVOUS: 1665 vm_handle_rendezvous(vm, vcpuid); 1666 error = 0; 1667 break; 1668 case VM_EXITCODE_HLT: 1669 intr_disabled = ((vme->u.hlt.rflags & PSL_I) == 0); 1670 error = vm_handle_hlt(vm, vcpuid, intr_disabled, &retu); 1671 break; 1672 case VM_EXITCODE_PAGING: 1673 error = vm_handle_paging(vm, vcpuid, &retu); 1674 break; 1675 case VM_EXITCODE_INST_EMUL: 1676 error = vm_handle_inst_emul(vm, vcpuid, &retu); 1677 break; 1678 case VM_EXITCODE_INOUT: 1679 case VM_EXITCODE_INOUT_STR: 1680 error = vm_handle_inout(vm, vcpuid, vme, &retu); 1681 break; 1682 case VM_EXITCODE_MONITOR: 1683 case VM_EXITCODE_MWAIT: 1684 vm_inject_ud(vm, vcpuid); 1685 break; 1686 default: 1687 retu = true; /* handled in userland */ 1688 break; 1689 } 1690 } 1691 1692 if (error == 0 && retu == false) 1693 goto restart; 1694 1695 VCPU_CTR2(vm, vcpuid, "retu %d/%d", error, vme->exitcode); 1696 1697 /* copy the exit information */ 1698 bcopy(vme, &vmrun->vm_exit, sizeof(struct vm_exit)); 1699 return (error); 1700 } 1701 1702 int 1703 vm_restart_instruction(void *arg, int vcpuid) 1704 { 1705 struct vm *vm; 1706 struct vcpu *vcpu; 1707 enum vcpu_state state; 1708 uint64_t rip; 1709 int error; 1710 1711 vm = arg; 1712 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1713 return (EINVAL); 1714 1715 vcpu = &vm->vcpu[vcpuid]; 1716 state = vcpu_get_state(vm, vcpuid, NULL); 1717 if (state == VCPU_RUNNING) { 1718 /* 1719 * When a vcpu is "running" the next instruction is determined 1720 * by adding 'rip' and 'inst_length' in the vcpu's 'exitinfo'. 1721 * Thus setting 'inst_length' to zero will cause the current 1722 * instruction to be restarted. 1723 */ 1724 vcpu->exitinfo.inst_length = 0; 1725 VCPU_CTR1(vm, vcpuid, "restarting instruction at %#lx by " 1726 "setting inst_length to zero", vcpu->exitinfo.rip); 1727 } else if (state == VCPU_FROZEN) { 1728 /* 1729 * When a vcpu is "frozen" it is outside the critical section 1730 * around VMRUN() and 'nextrip' points to the next instruction. 1731 * Thus instruction restart is achieved by setting 'nextrip' 1732 * to the vcpu's %rip. 1733 */ 1734 error = vm_get_register(vm, vcpuid, VM_REG_GUEST_RIP, &rip); 1735 KASSERT(!error, ("%s: error %d getting rip", __func__, error)); 1736 VCPU_CTR2(vm, vcpuid, "restarting instruction by updating " 1737 "nextrip from %#lx to %#lx", vcpu->nextrip, rip); 1738 vcpu->nextrip = rip; 1739 } else { 1740 panic("%s: invalid state %d", __func__, state); 1741 } 1742 return (0); 1743 } 1744 1745 int 1746 vm_exit_intinfo(struct vm *vm, int vcpuid, uint64_t info) 1747 { 1748 struct vcpu *vcpu; 1749 int type, vector; 1750 1751 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1752 return (EINVAL); 1753 1754 vcpu = &vm->vcpu[vcpuid]; 1755 1756 if (info & VM_INTINFO_VALID) { 1757 type = info & VM_INTINFO_TYPE; 1758 vector = info & 0xff; 1759 if (type == VM_INTINFO_NMI && vector != IDT_NMI) 1760 return (EINVAL); 1761 if (type == VM_INTINFO_HWEXCEPTION && vector >= 32) 1762 return (EINVAL); 1763 if (info & VM_INTINFO_RSVD) 1764 return (EINVAL); 1765 } else { 1766 info = 0; 1767 } 1768 VCPU_CTR2(vm, vcpuid, "%s: info1(%#lx)", __func__, info); 1769 vcpu->exitintinfo = info; 1770 return (0); 1771 } 1772 1773 enum exc_class { 1774 EXC_BENIGN, 1775 EXC_CONTRIBUTORY, 1776 EXC_PAGEFAULT 1777 }; 1778 1779 #define IDT_VE 20 /* Virtualization Exception (Intel specific) */ 1780 1781 static enum exc_class 1782 exception_class(uint64_t info) 1783 { 1784 int type, vector; 1785 1786 KASSERT(info & VM_INTINFO_VALID, ("intinfo must be valid: %#lx", info)); 1787 type = info & VM_INTINFO_TYPE; 1788 vector = info & 0xff; 1789 1790 /* Table 6-4, "Interrupt and Exception Classes", Intel SDM, Vol 3 */ 1791 switch (type) { 1792 case VM_INTINFO_HWINTR: 1793 case VM_INTINFO_SWINTR: 1794 case VM_INTINFO_NMI: 1795 return (EXC_BENIGN); 1796 default: 1797 /* 1798 * Hardware exception. 1799 * 1800 * SVM and VT-x use identical type values to represent NMI, 1801 * hardware interrupt and software interrupt. 1802 * 1803 * SVM uses type '3' for all exceptions. VT-x uses type '3' 1804 * for exceptions except #BP and #OF. #BP and #OF use a type 1805 * value of '5' or '6'. Therefore we don't check for explicit 1806 * values of 'type' to classify 'intinfo' into a hardware 1807 * exception. 1808 */ 1809 break; 1810 } 1811 1812 switch (vector) { 1813 case IDT_PF: 1814 case IDT_VE: 1815 return (EXC_PAGEFAULT); 1816 case IDT_DE: 1817 case IDT_TS: 1818 case IDT_NP: 1819 case IDT_SS: 1820 case IDT_GP: 1821 return (EXC_CONTRIBUTORY); 1822 default: 1823 return (EXC_BENIGN); 1824 } 1825 } 1826 1827 static int 1828 nested_fault(struct vm *vm, int vcpuid, uint64_t info1, uint64_t info2, 1829 uint64_t *retinfo) 1830 { 1831 enum exc_class exc1, exc2; 1832 int type1, vector1; 1833 1834 KASSERT(info1 & VM_INTINFO_VALID, ("info1 %#lx is not valid", info1)); 1835 KASSERT(info2 & VM_INTINFO_VALID, ("info2 %#lx is not valid", info2)); 1836 1837 /* 1838 * If an exception occurs while attempting to call the double-fault 1839 * handler the processor enters shutdown mode (aka triple fault). 1840 */ 1841 type1 = info1 & VM_INTINFO_TYPE; 1842 vector1 = info1 & 0xff; 1843 if (type1 == VM_INTINFO_HWEXCEPTION && vector1 == IDT_DF) { 1844 VCPU_CTR2(vm, vcpuid, "triple fault: info1(%#lx), info2(%#lx)", 1845 info1, info2); 1846 vm_suspend(vm, VM_SUSPEND_TRIPLEFAULT); 1847 *retinfo = 0; 1848 return (0); 1849 } 1850 1851 /* 1852 * Table 6-5 "Conditions for Generating a Double Fault", Intel SDM, Vol3 1853 */ 1854 exc1 = exception_class(info1); 1855 exc2 = exception_class(info2); 1856 if ((exc1 == EXC_CONTRIBUTORY && exc2 == EXC_CONTRIBUTORY) || 1857 (exc1 == EXC_PAGEFAULT && exc2 != EXC_BENIGN)) { 1858 /* Convert nested fault into a double fault. */ 1859 *retinfo = IDT_DF; 1860 *retinfo |= VM_INTINFO_VALID | VM_INTINFO_HWEXCEPTION; 1861 *retinfo |= VM_INTINFO_DEL_ERRCODE; 1862 } else { 1863 /* Handle exceptions serially */ 1864 *retinfo = info2; 1865 } 1866 return (1); 1867 } 1868 1869 static uint64_t 1870 vcpu_exception_intinfo(struct vcpu *vcpu) 1871 { 1872 uint64_t info = 0; 1873 1874 if (vcpu->exception_pending) { 1875 info = vcpu->exc_vector & 0xff; 1876 info |= VM_INTINFO_VALID | VM_INTINFO_HWEXCEPTION; 1877 if (vcpu->exc_errcode_valid) { 1878 info |= VM_INTINFO_DEL_ERRCODE; 1879 info |= (uint64_t)vcpu->exc_errcode << 32; 1880 } 1881 } 1882 return (info); 1883 } 1884 1885 int 1886 vm_entry_intinfo(struct vm *vm, int vcpuid, uint64_t *retinfo) 1887 { 1888 struct vcpu *vcpu; 1889 uint64_t info1, info2; 1890 int valid; 1891 1892 KASSERT(vcpuid >= 0 && vcpuid < VM_MAXCPU, ("invalid vcpu %d", vcpuid)); 1893 1894 vcpu = &vm->vcpu[vcpuid]; 1895 1896 info1 = vcpu->exitintinfo; 1897 vcpu->exitintinfo = 0; 1898 1899 info2 = 0; 1900 if (vcpu->exception_pending) { 1901 info2 = vcpu_exception_intinfo(vcpu); 1902 vcpu->exception_pending = 0; 1903 VCPU_CTR2(vm, vcpuid, "Exception %d delivered: %#lx", 1904 vcpu->exc_vector, info2); 1905 } 1906 1907 if ((info1 & VM_INTINFO_VALID) && (info2 & VM_INTINFO_VALID)) { 1908 valid = nested_fault(vm, vcpuid, info1, info2, retinfo); 1909 } else if (info1 & VM_INTINFO_VALID) { 1910 *retinfo = info1; 1911 valid = 1; 1912 } else if (info2 & VM_INTINFO_VALID) { 1913 *retinfo = info2; 1914 valid = 1; 1915 } else { 1916 valid = 0; 1917 } 1918 1919 if (valid) { 1920 VCPU_CTR4(vm, vcpuid, "%s: info1(%#lx), info2(%#lx), " 1921 "retinfo(%#lx)", __func__, info1, info2, *retinfo); 1922 } 1923 1924 return (valid); 1925 } 1926 1927 int 1928 vm_get_intinfo(struct vm *vm, int vcpuid, uint64_t *info1, uint64_t *info2) 1929 { 1930 struct vcpu *vcpu; 1931 1932 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1933 return (EINVAL); 1934 1935 vcpu = &vm->vcpu[vcpuid]; 1936 *info1 = vcpu->exitintinfo; 1937 *info2 = vcpu_exception_intinfo(vcpu); 1938 return (0); 1939 } 1940 1941 int 1942 vm_inject_exception(struct vm *vm, int vcpuid, int vector, int errcode_valid, 1943 uint32_t errcode, int restart_instruction) 1944 { 1945 struct vcpu *vcpu; 1946 uint64_t regval; 1947 int error; 1948 1949 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1950 return (EINVAL); 1951 1952 if (vector < 0 || vector >= 32) 1953 return (EINVAL); 1954 1955 /* 1956 * A double fault exception should never be injected directly into 1957 * the guest. It is a derived exception that results from specific 1958 * combinations of nested faults. 1959 */ 1960 if (vector == IDT_DF) 1961 return (EINVAL); 1962 1963 vcpu = &vm->vcpu[vcpuid]; 1964 1965 if (vcpu->exception_pending) { 1966 VCPU_CTR2(vm, vcpuid, "Unable to inject exception %d due to " 1967 "pending exception %d", vector, vcpu->exc_vector); 1968 return (EBUSY); 1969 } 1970 1971 if (errcode_valid) { 1972 /* 1973 * Exceptions don't deliver an error code in real mode. 1974 */ 1975 error = vm_get_register(vm, vcpuid, VM_REG_GUEST_CR0, ®val); 1976 KASSERT(!error, ("%s: error %d getting CR0", __func__, error)); 1977 if (!(regval & CR0_PE)) 1978 errcode_valid = 0; 1979 } 1980 1981 /* 1982 * From section 26.6.1 "Interruptibility State" in Intel SDM: 1983 * 1984 * Event blocking by "STI" or "MOV SS" is cleared after guest executes 1985 * one instruction or incurs an exception. 1986 */ 1987 error = vm_set_register(vm, vcpuid, VM_REG_GUEST_INTR_SHADOW, 0); 1988 KASSERT(error == 0, ("%s: error %d clearing interrupt shadow", 1989 __func__, error)); 1990 1991 if (restart_instruction) 1992 vm_restart_instruction(vm, vcpuid); 1993 1994 vcpu->exception_pending = 1; 1995 vcpu->exc_vector = vector; 1996 vcpu->exc_errcode = errcode; 1997 vcpu->exc_errcode_valid = errcode_valid; 1998 VCPU_CTR1(vm, vcpuid, "Exception %d pending", vector); 1999 return (0); 2000 } 2001 2002 void 2003 vm_inject_fault(void *vmarg, int vcpuid, int vector, int errcode_valid, 2004 int errcode) 2005 { 2006 struct vm *vm; 2007 int error, restart_instruction; 2008 2009 vm = vmarg; 2010 restart_instruction = 1; 2011 2012 error = vm_inject_exception(vm, vcpuid, vector, errcode_valid, 2013 errcode, restart_instruction); 2014 KASSERT(error == 0, ("vm_inject_exception error %d", error)); 2015 } 2016 2017 void 2018 vm_inject_pf(void *vmarg, int vcpuid, int error_code, uint64_t cr2) 2019 { 2020 struct vm *vm; 2021 int error; 2022 2023 vm = vmarg; 2024 VCPU_CTR2(vm, vcpuid, "Injecting page fault: error_code %#x, cr2 %#lx", 2025 error_code, cr2); 2026 2027 error = vm_set_register(vm, vcpuid, VM_REG_GUEST_CR2, cr2); 2028 KASSERT(error == 0, ("vm_set_register(cr2) error %d", error)); 2029 2030 vm_inject_fault(vm, vcpuid, IDT_PF, 1, error_code); 2031 } 2032 2033 static VMM_STAT(VCPU_NMI_COUNT, "number of NMIs delivered to vcpu"); 2034 2035 int 2036 vm_inject_nmi(struct vm *vm, int vcpuid) 2037 { 2038 struct vcpu *vcpu; 2039 2040 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 2041 return (EINVAL); 2042 2043 vcpu = &vm->vcpu[vcpuid]; 2044 2045 vcpu->nmi_pending = 1; 2046 vcpu_notify_event(vm, vcpuid, false); 2047 return (0); 2048 } 2049 2050 int 2051 vm_nmi_pending(struct vm *vm, int vcpuid) 2052 { 2053 struct vcpu *vcpu; 2054 2055 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 2056 panic("vm_nmi_pending: invalid vcpuid %d", vcpuid); 2057 2058 vcpu = &vm->vcpu[vcpuid]; 2059 2060 return (vcpu->nmi_pending); 2061 } 2062 2063 void 2064 vm_nmi_clear(struct vm *vm, int vcpuid) 2065 { 2066 struct vcpu *vcpu; 2067 2068 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 2069 panic("vm_nmi_pending: invalid vcpuid %d", vcpuid); 2070 2071 vcpu = &vm->vcpu[vcpuid]; 2072 2073 if (vcpu->nmi_pending == 0) 2074 panic("vm_nmi_clear: inconsistent nmi_pending state"); 2075 2076 vcpu->nmi_pending = 0; 2077 vmm_stat_incr(vm, vcpuid, VCPU_NMI_COUNT, 1); 2078 } 2079 2080 static VMM_STAT(VCPU_EXTINT_COUNT, "number of ExtINTs delivered to vcpu"); 2081 2082 int 2083 vm_inject_extint(struct vm *vm, int vcpuid) 2084 { 2085 struct vcpu *vcpu; 2086 2087 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 2088 return (EINVAL); 2089 2090 vcpu = &vm->vcpu[vcpuid]; 2091 2092 vcpu->extint_pending = 1; 2093 vcpu_notify_event(vm, vcpuid, false); 2094 return (0); 2095 } 2096 2097 int 2098 vm_extint_pending(struct vm *vm, int vcpuid) 2099 { 2100 struct vcpu *vcpu; 2101 2102 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 2103 panic("vm_extint_pending: invalid vcpuid %d", vcpuid); 2104 2105 vcpu = &vm->vcpu[vcpuid]; 2106 2107 return (vcpu->extint_pending); 2108 } 2109 2110 void 2111 vm_extint_clear(struct vm *vm, int vcpuid) 2112 { 2113 struct vcpu *vcpu; 2114 2115 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 2116 panic("vm_extint_pending: invalid vcpuid %d", vcpuid); 2117 2118 vcpu = &vm->vcpu[vcpuid]; 2119 2120 if (vcpu->extint_pending == 0) 2121 panic("vm_extint_clear: inconsistent extint_pending state"); 2122 2123 vcpu->extint_pending = 0; 2124 vmm_stat_incr(vm, vcpuid, VCPU_EXTINT_COUNT, 1); 2125 } 2126 2127 int 2128 vm_get_capability(struct vm *vm, int vcpu, int type, int *retval) 2129 { 2130 if (vcpu < 0 || vcpu >= VM_MAXCPU) 2131 return (EINVAL); 2132 2133 if (type < 0 || type >= VM_CAP_MAX) 2134 return (EINVAL); 2135 2136 return (VMGETCAP(vm->cookie, vcpu, type, retval)); 2137 } 2138 2139 int 2140 vm_set_capability(struct vm *vm, int vcpu, int type, int val) 2141 { 2142 if (vcpu < 0 || vcpu >= VM_MAXCPU) 2143 return (EINVAL); 2144 2145 if (type < 0 || type >= VM_CAP_MAX) 2146 return (EINVAL); 2147 2148 return (VMSETCAP(vm->cookie, vcpu, type, val)); 2149 } 2150 2151 struct vlapic * 2152 vm_lapic(struct vm *vm, int cpu) 2153 { 2154 return (vm->vcpu[cpu].vlapic); 2155 } 2156 2157 struct vioapic * 2158 vm_ioapic(struct vm *vm) 2159 { 2160 2161 return (vm->vioapic); 2162 } 2163 2164 struct vhpet * 2165 vm_hpet(struct vm *vm) 2166 { 2167 2168 return (vm->vhpet); 2169 } 2170 2171 boolean_t 2172 vmm_is_pptdev(int bus, int slot, int func) 2173 { 2174 int found, i, n; 2175 int b, s, f; 2176 char *val, *cp, *cp2; 2177 2178 /* 2179 * XXX 2180 * The length of an environment variable is limited to 128 bytes which 2181 * puts an upper limit on the number of passthru devices that may be 2182 * specified using a single environment variable. 2183 * 2184 * Work around this by scanning multiple environment variable 2185 * names instead of a single one - yuck! 2186 */ 2187 const char *names[] = { "pptdevs", "pptdevs2", "pptdevs3", NULL }; 2188 2189 /* set pptdevs="1/2/3 4/5/6 7/8/9 10/11/12" */ 2190 found = 0; 2191 for (i = 0; names[i] != NULL && !found; i++) { 2192 cp = val = kern_getenv(names[i]); 2193 while (cp != NULL && *cp != '\0') { 2194 if ((cp2 = strchr(cp, ' ')) != NULL) 2195 *cp2 = '\0'; 2196 2197 n = sscanf(cp, "%d/%d/%d", &b, &s, &f); 2198 if (n == 3 && bus == b && slot == s && func == f) { 2199 found = 1; 2200 break; 2201 } 2202 2203 if (cp2 != NULL) 2204 *cp2++ = ' '; 2205 2206 cp = cp2; 2207 } 2208 freeenv(val); 2209 } 2210 return (found); 2211 } 2212 2213 void * 2214 vm_iommu_domain(struct vm *vm) 2215 { 2216 2217 return (vm->iommu); 2218 } 2219 2220 int 2221 vcpu_set_state(struct vm *vm, int vcpuid, enum vcpu_state newstate, 2222 bool from_idle) 2223 { 2224 int error; 2225 struct vcpu *vcpu; 2226 2227 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 2228 panic("vm_set_run_state: invalid vcpuid %d", vcpuid); 2229 2230 vcpu = &vm->vcpu[vcpuid]; 2231 2232 vcpu_lock(vcpu); 2233 error = vcpu_set_state_locked(vm, vcpuid, newstate, from_idle); 2234 vcpu_unlock(vcpu); 2235 2236 return (error); 2237 } 2238 2239 enum vcpu_state 2240 vcpu_get_state(struct vm *vm, int vcpuid, int *hostcpu) 2241 { 2242 struct vcpu *vcpu; 2243 enum vcpu_state state; 2244 2245 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 2246 panic("vm_get_run_state: invalid vcpuid %d", vcpuid); 2247 2248 vcpu = &vm->vcpu[vcpuid]; 2249 2250 vcpu_lock(vcpu); 2251 state = vcpu->state; 2252 if (hostcpu != NULL) 2253 *hostcpu = vcpu->hostcpu; 2254 vcpu_unlock(vcpu); 2255 2256 return (state); 2257 } 2258 2259 int 2260 vm_activate_cpu(struct vm *vm, int vcpuid) 2261 { 2262 2263 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 2264 return (EINVAL); 2265 2266 if (CPU_ISSET(vcpuid, &vm->active_cpus)) 2267 return (EBUSY); 2268 2269 VCPU_CTR0(vm, vcpuid, "activated"); 2270 CPU_SET_ATOMIC(vcpuid, &vm->active_cpus); 2271 return (0); 2272 } 2273 2274 cpuset_t 2275 vm_active_cpus(struct vm *vm) 2276 { 2277 2278 return (vm->active_cpus); 2279 } 2280 2281 cpuset_t 2282 vm_suspended_cpus(struct vm *vm) 2283 { 2284 2285 return (vm->suspended_cpus); 2286 } 2287 2288 void * 2289 vcpu_stats(struct vm *vm, int vcpuid) 2290 { 2291 2292 return (vm->vcpu[vcpuid].stats); 2293 } 2294 2295 int 2296 vm_get_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state *state) 2297 { 2298 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 2299 return (EINVAL); 2300 2301 *state = vm->vcpu[vcpuid].x2apic_state; 2302 2303 return (0); 2304 } 2305 2306 int 2307 vm_set_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state state) 2308 { 2309 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 2310 return (EINVAL); 2311 2312 if (state >= X2APIC_STATE_LAST) 2313 return (EINVAL); 2314 2315 vm->vcpu[vcpuid].x2apic_state = state; 2316 2317 vlapic_set_x2apic_state(vm, vcpuid, state); 2318 2319 return (0); 2320 } 2321 2322 /* 2323 * This function is called to ensure that a vcpu "sees" a pending event 2324 * as soon as possible: 2325 * - If the vcpu thread is sleeping then it is woken up. 2326 * - If the vcpu is running on a different host_cpu then an IPI will be directed 2327 * to the host_cpu to cause the vcpu to trap into the hypervisor. 2328 */ 2329 static void 2330 vcpu_notify_event_locked(struct vcpu *vcpu, bool lapic_intr) 2331 { 2332 int hostcpu; 2333 2334 hostcpu = vcpu->hostcpu; 2335 if (vcpu->state == VCPU_RUNNING) { 2336 KASSERT(hostcpu != NOCPU, ("vcpu running on invalid hostcpu")); 2337 if (hostcpu != curcpu) { 2338 if (lapic_intr) { 2339 vlapic_post_intr(vcpu->vlapic, hostcpu, 2340 vmm_ipinum); 2341 } else { 2342 ipi_cpu(hostcpu, vmm_ipinum); 2343 } 2344 } else { 2345 /* 2346 * If the 'vcpu' is running on 'curcpu' then it must 2347 * be sending a notification to itself (e.g. SELF_IPI). 2348 * The pending event will be picked up when the vcpu 2349 * transitions back to guest context. 2350 */ 2351 } 2352 } else { 2353 KASSERT(hostcpu == NOCPU, ("vcpu state %d not consistent " 2354 "with hostcpu %d", vcpu->state, hostcpu)); 2355 if (vcpu->state == VCPU_SLEEPING) 2356 wakeup_one(vcpu); 2357 } 2358 } 2359 2360 void 2361 vcpu_notify_event(struct vm *vm, int vcpuid, bool lapic_intr) 2362 { 2363 struct vcpu *vcpu = &vm->vcpu[vcpuid]; 2364 2365 vcpu_lock(vcpu); 2366 vcpu_notify_event_locked(vcpu, lapic_intr); 2367 vcpu_unlock(vcpu); 2368 } 2369 2370 struct vmspace * 2371 vm_get_vmspace(struct vm *vm) 2372 { 2373 2374 return (vm->vmspace); 2375 } 2376 2377 int 2378 vm_apicid2vcpuid(struct vm *vm, int apicid) 2379 { 2380 /* 2381 * XXX apic id is assumed to be numerically identical to vcpu id 2382 */ 2383 return (apicid); 2384 } 2385 2386 void 2387 vm_smp_rendezvous(struct vm *vm, int vcpuid, cpuset_t dest, 2388 vm_rendezvous_func_t func, void *arg) 2389 { 2390 int i; 2391 2392 /* 2393 * Enforce that this function is called without any locks 2394 */ 2395 WITNESS_WARN(WARN_PANIC, NULL, "vm_smp_rendezvous"); 2396 KASSERT(vcpuid == -1 || (vcpuid >= 0 && vcpuid < VM_MAXCPU), 2397 ("vm_smp_rendezvous: invalid vcpuid %d", vcpuid)); 2398 2399 restart: 2400 mtx_lock(&vm->rendezvous_mtx); 2401 if (vm->rendezvous_func != NULL) { 2402 /* 2403 * If a rendezvous is already in progress then we need to 2404 * call the rendezvous handler in case this 'vcpuid' is one 2405 * of the targets of the rendezvous. 2406 */ 2407 RENDEZVOUS_CTR0(vm, vcpuid, "Rendezvous already in progress"); 2408 mtx_unlock(&vm->rendezvous_mtx); 2409 vm_handle_rendezvous(vm, vcpuid); 2410 goto restart; 2411 } 2412 KASSERT(vm->rendezvous_func == NULL, ("vm_smp_rendezvous: previous " 2413 "rendezvous is still in progress")); 2414 2415 RENDEZVOUS_CTR0(vm, vcpuid, "Initiating rendezvous"); 2416 vm->rendezvous_req_cpus = dest; 2417 CPU_ZERO(&vm->rendezvous_done_cpus); 2418 vm->rendezvous_arg = arg; 2419 vm_set_rendezvous_func(vm, func); 2420 mtx_unlock(&vm->rendezvous_mtx); 2421 2422 /* 2423 * Wake up any sleeping vcpus and trigger a VM-exit in any running 2424 * vcpus so they handle the rendezvous as soon as possible. 2425 */ 2426 for (i = 0; i < VM_MAXCPU; i++) { 2427 if (CPU_ISSET(i, &dest)) 2428 vcpu_notify_event(vm, i, false); 2429 } 2430 2431 vm_handle_rendezvous(vm, vcpuid); 2432 } 2433 2434 struct vatpic * 2435 vm_atpic(struct vm *vm) 2436 { 2437 return (vm->vatpic); 2438 } 2439 2440 struct vatpit * 2441 vm_atpit(struct vm *vm) 2442 { 2443 return (vm->vatpit); 2444 } 2445 2446 struct vpmtmr * 2447 vm_pmtmr(struct vm *vm) 2448 { 2449 2450 return (vm->vpmtmr); 2451 } 2452 2453 struct vrtc * 2454 vm_rtc(struct vm *vm) 2455 { 2456 2457 return (vm->vrtc); 2458 } 2459 2460 enum vm_reg_name 2461 vm_segment_name(int seg) 2462 { 2463 static enum vm_reg_name seg_names[] = { 2464 VM_REG_GUEST_ES, 2465 VM_REG_GUEST_CS, 2466 VM_REG_GUEST_SS, 2467 VM_REG_GUEST_DS, 2468 VM_REG_GUEST_FS, 2469 VM_REG_GUEST_GS 2470 }; 2471 2472 KASSERT(seg >= 0 && seg < nitems(seg_names), 2473 ("%s: invalid segment encoding %d", __func__, seg)); 2474 return (seg_names[seg]); 2475 } 2476 2477 void 2478 vm_copy_teardown(struct vm *vm, int vcpuid, struct vm_copyinfo *copyinfo, 2479 int num_copyinfo) 2480 { 2481 int idx; 2482 2483 for (idx = 0; idx < num_copyinfo; idx++) { 2484 if (copyinfo[idx].cookie != NULL) 2485 vm_gpa_release(copyinfo[idx].cookie); 2486 } 2487 bzero(copyinfo, num_copyinfo * sizeof(struct vm_copyinfo)); 2488 } 2489 2490 int 2491 vm_copy_setup(struct vm *vm, int vcpuid, struct vm_guest_paging *paging, 2492 uint64_t gla, size_t len, int prot, struct vm_copyinfo *copyinfo, 2493 int num_copyinfo, int *fault) 2494 { 2495 int error, idx, nused; 2496 size_t n, off, remaining; 2497 void *hva, *cookie; 2498 uint64_t gpa; 2499 2500 bzero(copyinfo, sizeof(struct vm_copyinfo) * num_copyinfo); 2501 2502 nused = 0; 2503 remaining = len; 2504 while (remaining > 0) { 2505 KASSERT(nused < num_copyinfo, ("insufficient vm_copyinfo")); 2506 error = vm_gla2gpa(vm, vcpuid, paging, gla, prot, &gpa, fault); 2507 if (error || *fault) 2508 return (error); 2509 off = gpa & PAGE_MASK; 2510 n = min(remaining, PAGE_SIZE - off); 2511 copyinfo[nused].gpa = gpa; 2512 copyinfo[nused].len = n; 2513 remaining -= n; 2514 gla += n; 2515 nused++; 2516 } 2517 2518 for (idx = 0; idx < nused; idx++) { 2519 hva = vm_gpa_hold(vm, vcpuid, copyinfo[idx].gpa, 2520 copyinfo[idx].len, prot, &cookie); 2521 if (hva == NULL) 2522 break; 2523 copyinfo[idx].hva = hva; 2524 copyinfo[idx].cookie = cookie; 2525 } 2526 2527 if (idx != nused) { 2528 vm_copy_teardown(vm, vcpuid, copyinfo, num_copyinfo); 2529 return (EFAULT); 2530 } else { 2531 *fault = 0; 2532 return (0); 2533 } 2534 } 2535 2536 void 2537 vm_copyin(struct vm *vm, int vcpuid, struct vm_copyinfo *copyinfo, void *kaddr, 2538 size_t len) 2539 { 2540 char *dst; 2541 int idx; 2542 2543 dst = kaddr; 2544 idx = 0; 2545 while (len > 0) { 2546 bcopy(copyinfo[idx].hva, dst, copyinfo[idx].len); 2547 len -= copyinfo[idx].len; 2548 dst += copyinfo[idx].len; 2549 idx++; 2550 } 2551 } 2552 2553 void 2554 vm_copyout(struct vm *vm, int vcpuid, const void *kaddr, 2555 struct vm_copyinfo *copyinfo, size_t len) 2556 { 2557 const char *src; 2558 int idx; 2559 2560 src = kaddr; 2561 idx = 0; 2562 while (len > 0) { 2563 bcopy(src, copyinfo[idx].hva, copyinfo[idx].len); 2564 len -= copyinfo[idx].len; 2565 src += copyinfo[idx].len; 2566 idx++; 2567 } 2568 } 2569 2570 /* 2571 * Return the amount of in-use and wired memory for the VM. Since 2572 * these are global stats, only return the values with for vCPU 0 2573 */ 2574 VMM_STAT_DECLARE(VMM_MEM_RESIDENT); 2575 VMM_STAT_DECLARE(VMM_MEM_WIRED); 2576 2577 static void 2578 vm_get_rescnt(struct vm *vm, int vcpu, struct vmm_stat_type *stat) 2579 { 2580 2581 if (vcpu == 0) { 2582 vmm_stat_set(vm, vcpu, VMM_MEM_RESIDENT, 2583 PAGE_SIZE * vmspace_resident_count(vm->vmspace)); 2584 } 2585 } 2586 2587 static void 2588 vm_get_wiredcnt(struct vm *vm, int vcpu, struct vmm_stat_type *stat) 2589 { 2590 2591 if (vcpu == 0) { 2592 vmm_stat_set(vm, vcpu, VMM_MEM_WIRED, 2593 PAGE_SIZE * pmap_wired_count(vmspace_pmap(vm->vmspace))); 2594 } 2595 } 2596 2597 VMM_STAT_FUNC(VMM_MEM_RESIDENT, "Resident memory", vm_get_rescnt); 2598 VMM_STAT_FUNC(VMM_MEM_WIRED, "Wired memory", vm_get_wiredcnt); 2599