vmmapi.c revision 295881
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: head/lib/libvmmapi/vmmapi.c 295881 2016-02-22 09:04:36Z skra $ 27 */ 28 29#include <sys/cdefs.h> 30__FBSDID("$FreeBSD: head/lib/libvmmapi/vmmapi.c 295881 2016-02-22 09:04:36Z skra $"); 31 32#include <sys/param.h> 33#include <sys/sysctl.h> 34#include <sys/ioctl.h> 35#include <sys/mman.h> 36#include <sys/_iovec.h> 37#include <sys/cpuset.h> 38 39#include <x86/segments.h> 40#include <machine/specialreg.h> 41 42#include <errno.h> 43#include <stdio.h> 44#include <stdlib.h> 45#include <assert.h> 46#include <string.h> 47#include <fcntl.h> 48#include <unistd.h> 49 50#include <libutil.h> 51 52#include <machine/vmm.h> 53#include <machine/vmm_dev.h> 54 55#include "vmmapi.h" 56 57#define MB (1024 * 1024UL) 58#define GB (1024 * 1024 * 1024UL) 59 60/* 61 * Size of the guard region before and after the virtual address space 62 * mapping the guest physical memory. This must be a multiple of the 63 * superpage size for performance reasons. 64 */ 65#define VM_MMAP_GUARD_SIZE (4 * MB) 66 67#define PROT_RW (PROT_READ | PROT_WRITE) 68#define PROT_ALL (PROT_READ | PROT_WRITE | PROT_EXEC) 69 70struct vmctx { 71 int fd; 72 uint32_t lowmem_limit; 73 int memflags; 74 size_t lowmem; 75 size_t highmem; 76 char *baseaddr; 77 char *name; 78}; 79 80#define CREATE(x) sysctlbyname("hw.vmm.create", NULL, NULL, (x), strlen((x))) 81#define DESTROY(x) sysctlbyname("hw.vmm.destroy", NULL, NULL, (x), strlen((x))) 82 83static int 84vm_device_open(const char *name) 85{ 86 int fd, len; 87 char *vmfile; 88 89 len = strlen("/dev/vmm/") + strlen(name) + 1; 90 vmfile = malloc(len); 91 assert(vmfile != NULL); 92 snprintf(vmfile, len, "/dev/vmm/%s", name); 93 94 /* Open the device file */ 95 fd = open(vmfile, O_RDWR, 0); 96 97 free(vmfile); 98 return (fd); 99} 100 101int 102vm_create(const char *name) 103{ 104 105 return (CREATE((char *)name)); 106} 107 108struct vmctx * 109vm_open(const char *name) 110{ 111 struct vmctx *vm; 112 113 vm = malloc(sizeof(struct vmctx) + strlen(name) + 1); 114 assert(vm != NULL); 115 116 vm->fd = -1; 117 vm->memflags = 0; 118 vm->lowmem_limit = 3 * GB; 119 vm->name = (char *)(vm + 1); 120 strcpy(vm->name, name); 121 122 if ((vm->fd = vm_device_open(vm->name)) < 0) 123 goto err; 124 125 return (vm); 126err: 127 vm_destroy(vm); 128 return (NULL); 129} 130 131void 132vm_destroy(struct vmctx *vm) 133{ 134 assert(vm != NULL); 135 136 if (vm->fd >= 0) 137 close(vm->fd); 138 DESTROY(vm->name); 139 140 free(vm); 141} 142 143int 144vm_parse_memsize(const char *optarg, size_t *ret_memsize) 145{ 146 char *endptr; 147 size_t optval; 148 int error; 149 150 optval = strtoul(optarg, &endptr, 0); 151 if (*optarg != '\0' && *endptr == '\0') { 152 /* 153 * For the sake of backward compatibility if the memory size 154 * specified on the command line is less than a megabyte then 155 * it is interpreted as being in units of MB. 156 */ 157 if (optval < MB) 158 optval *= MB; 159 *ret_memsize = optval; 160 error = 0; 161 } else 162 error = expand_number(optarg, ret_memsize); 163 164 return (error); 165} 166 167uint32_t 168vm_get_lowmem_limit(struct vmctx *ctx) 169{ 170 171 return (ctx->lowmem_limit); 172} 173 174void 175vm_set_lowmem_limit(struct vmctx *ctx, uint32_t limit) 176{ 177 178 ctx->lowmem_limit = limit; 179} 180 181void 182vm_set_memflags(struct vmctx *ctx, int flags) 183{ 184 185 ctx->memflags = flags; 186} 187 188int 189vm_get_memflags(struct vmctx *ctx) 190{ 191 192 return (ctx->memflags); 193} 194 195/* 196 * Map segment 'segid' starting at 'off' into guest address range [gpa,gpa+len). 197 */ 198int 199vm_mmap_memseg(struct vmctx *ctx, vm_paddr_t gpa, int segid, vm_ooffset_t off, 200 size_t len, int prot) 201{ 202 struct vm_memmap memmap; 203 int error, flags; 204 205 memmap.gpa = gpa; 206 memmap.segid = segid; 207 memmap.segoff = off; 208 memmap.len = len; 209 memmap.prot = prot; 210 memmap.flags = 0; 211 212 if (ctx->memflags & VM_MEM_F_WIRED) 213 memmap.flags |= VM_MEMMAP_F_WIRED; 214 215 /* 216 * If this mapping already exists then don't create it again. This 217 * is the common case for SYSMEM mappings created by bhyveload(8). 218 */ 219 error = vm_mmap_getnext(ctx, &gpa, &segid, &off, &len, &prot, &flags); 220 if (error == 0 && gpa == memmap.gpa) { 221 if (segid != memmap.segid || off != memmap.segoff || 222 prot != memmap.prot || flags != memmap.flags) { 223 errno = EEXIST; 224 return (-1); 225 } else { 226 return (0); 227 } 228 } 229 230 error = ioctl(ctx->fd, VM_MMAP_MEMSEG, &memmap); 231 return (error); 232} 233 234int 235vm_mmap_getnext(struct vmctx *ctx, vm_paddr_t *gpa, int *segid, 236 vm_ooffset_t *segoff, size_t *len, int *prot, int *flags) 237{ 238 struct vm_memmap memmap; 239 int error; 240 241 bzero(&memmap, sizeof(struct vm_memmap)); 242 memmap.gpa = *gpa; 243 error = ioctl(ctx->fd, VM_MMAP_GETNEXT, &memmap); 244 if (error == 0) { 245 *gpa = memmap.gpa; 246 *segid = memmap.segid; 247 *segoff = memmap.segoff; 248 *len = memmap.len; 249 *prot = memmap.prot; 250 *flags = memmap.flags; 251 } 252 return (error); 253} 254 255/* 256 * Return 0 if the segments are identical and non-zero otherwise. 257 * 258 * This is slightly complicated by the fact that only device memory segments 259 * are named. 260 */ 261static int 262cmpseg(size_t len, const char *str, size_t len2, const char *str2) 263{ 264 265 if (len == len2) { 266 if ((!str && !str2) || (str && str2 && !strcmp(str, str2))) 267 return (0); 268 } 269 return (-1); 270} 271 272static int 273vm_alloc_memseg(struct vmctx *ctx, int segid, size_t len, const char *name) 274{ 275 struct vm_memseg memseg; 276 size_t n; 277 int error; 278 279 /* 280 * If the memory segment has already been created then just return. 281 * This is the usual case for the SYSMEM segment created by userspace 282 * loaders like bhyveload(8). 283 */ 284 error = vm_get_memseg(ctx, segid, &memseg.len, memseg.name, 285 sizeof(memseg.name)); 286 if (error) 287 return (error); 288 289 if (memseg.len != 0) { 290 if (cmpseg(len, name, memseg.len, VM_MEMSEG_NAME(&memseg))) { 291 errno = EINVAL; 292 return (-1); 293 } else { 294 return (0); 295 } 296 } 297 298 bzero(&memseg, sizeof(struct vm_memseg)); 299 memseg.segid = segid; 300 memseg.len = len; 301 if (name != NULL) { 302 n = strlcpy(memseg.name, name, sizeof(memseg.name)); 303 if (n >= sizeof(memseg.name)) { 304 errno = ENAMETOOLONG; 305 return (-1); 306 } 307 } 308 309 error = ioctl(ctx->fd, VM_ALLOC_MEMSEG, &memseg); 310 return (error); 311} 312 313int 314vm_get_memseg(struct vmctx *ctx, int segid, size_t *lenp, char *namebuf, 315 size_t bufsize) 316{ 317 struct vm_memseg memseg; 318 size_t n; 319 int error; 320 321 memseg.segid = segid; 322 error = ioctl(ctx->fd, VM_GET_MEMSEG, &memseg); 323 if (error == 0) { 324 *lenp = memseg.len; 325 n = strlcpy(namebuf, memseg.name, bufsize); 326 if (n >= bufsize) { 327 errno = ENAMETOOLONG; 328 error = -1; 329 } 330 } 331 return (error); 332} 333 334static int 335setup_memory_segment(struct vmctx *ctx, vm_paddr_t gpa, size_t len, char *base) 336{ 337 char *ptr; 338 int error, flags; 339 340 /* Map 'len' bytes starting at 'gpa' in the guest address space */ 341 error = vm_mmap_memseg(ctx, gpa, VM_SYSMEM, gpa, len, PROT_ALL); 342 if (error) 343 return (error); 344 345 flags = MAP_SHARED | MAP_FIXED; 346 if ((ctx->memflags & VM_MEM_F_INCORE) == 0) 347 flags |= MAP_NOCORE; 348 349 /* mmap into the process address space on the host */ 350 ptr = mmap(base + gpa, len, PROT_RW, flags, ctx->fd, gpa); 351 if (ptr == MAP_FAILED) 352 return (-1); 353 354 return (0); 355} 356 357int 358vm_setup_memory(struct vmctx *ctx, size_t memsize, enum vm_mmap_style vms) 359{ 360 size_t objsize, len; 361 vm_paddr_t gpa; 362 char *baseaddr, *ptr; 363 int error, flags; 364 365 assert(vms == VM_MMAP_ALL); 366 367 /* 368 * If 'memsize' cannot fit entirely in the 'lowmem' segment then 369 * create another 'highmem' segment above 4GB for the remainder. 370 */ 371 if (memsize > ctx->lowmem_limit) { 372 ctx->lowmem = ctx->lowmem_limit; 373 ctx->highmem = memsize - ctx->lowmem_limit; 374 objsize = 4*GB + ctx->highmem; 375 } else { 376 ctx->lowmem = memsize; 377 ctx->highmem = 0; 378 objsize = ctx->lowmem; 379 } 380 381 error = vm_alloc_memseg(ctx, VM_SYSMEM, objsize, NULL); 382 if (error) 383 return (error); 384 385 /* 386 * Stake out a contiguous region covering the guest physical memory 387 * and the adjoining guard regions. 388 */ 389 len = VM_MMAP_GUARD_SIZE + objsize + VM_MMAP_GUARD_SIZE; 390 flags = MAP_PRIVATE | MAP_ANON | MAP_NOCORE | MAP_ALIGNED_SUPER; 391 ptr = mmap(NULL, len, PROT_NONE, flags, -1, 0); 392 if (ptr == MAP_FAILED) 393 return (-1); 394 395 baseaddr = ptr + VM_MMAP_GUARD_SIZE; 396 if (ctx->highmem > 0) { 397 gpa = 4*GB; 398 len = ctx->highmem; 399 error = setup_memory_segment(ctx, gpa, len, baseaddr); 400 if (error) 401 return (error); 402 } 403 404 if (ctx->lowmem > 0) { 405 gpa = 0; 406 len = ctx->lowmem; 407 error = setup_memory_segment(ctx, gpa, len, baseaddr); 408 if (error) 409 return (error); 410 } 411 412 ctx->baseaddr = baseaddr; 413 414 return (0); 415} 416 417/* 418 * Returns a non-NULL pointer if [gaddr, gaddr+len) is entirely contained in 419 * the lowmem or highmem regions. 420 * 421 * In particular return NULL if [gaddr, gaddr+len) falls in guest MMIO region. 422 * The instruction emulation code depends on this behavior. 423 */ 424void * 425vm_map_gpa(struct vmctx *ctx, vm_paddr_t gaddr, size_t len) 426{ 427 428 if (ctx->lowmem > 0) { 429 if (gaddr < ctx->lowmem && gaddr + len <= ctx->lowmem) 430 return (ctx->baseaddr + gaddr); 431 } 432 433 if (ctx->highmem > 0) { 434 if (gaddr >= 4*GB && gaddr + len <= 4*GB + ctx->highmem) 435 return (ctx->baseaddr + gaddr); 436 } 437 438 return (NULL); 439} 440 441size_t 442vm_get_lowmem_size(struct vmctx *ctx) 443{ 444 445 return (ctx->lowmem); 446} 447 448size_t 449vm_get_highmem_size(struct vmctx *ctx) 450{ 451 452 return (ctx->highmem); 453} 454 455void * 456vm_create_devmem(struct vmctx *ctx, int segid, const char *name, size_t len) 457{ 458 char pathname[MAXPATHLEN]; 459 size_t len2; 460 char *base, *ptr; 461 int fd, error, flags; 462 463 fd = -1; 464 ptr = MAP_FAILED; 465 if (name == NULL || strlen(name) == 0) { 466 errno = EINVAL; 467 goto done; 468 } 469 470 error = vm_alloc_memseg(ctx, segid, len, name); 471 if (error) 472 goto done; 473 474 strlcpy(pathname, "/dev/vmm.io/", sizeof(pathname)); 475 strlcat(pathname, ctx->name, sizeof(pathname)); 476 strlcat(pathname, ".", sizeof(pathname)); 477 strlcat(pathname, name, sizeof(pathname)); 478 479 fd = open(pathname, O_RDWR); 480 if (fd < 0) 481 goto done; 482 483 /* 484 * Stake out a contiguous region covering the device memory and the 485 * adjoining guard regions. 486 */ 487 len2 = VM_MMAP_GUARD_SIZE + len + VM_MMAP_GUARD_SIZE; 488 flags = MAP_PRIVATE | MAP_ANON | MAP_NOCORE | MAP_ALIGNED_SUPER; 489 base = mmap(NULL, len2, PROT_NONE, flags, -1, 0); 490 if (base == MAP_FAILED) 491 goto done; 492 493 flags = MAP_SHARED | MAP_FIXED; 494 if ((ctx->memflags & VM_MEM_F_INCORE) == 0) 495 flags |= MAP_NOCORE; 496 497 /* mmap the devmem region in the host address space */ 498 ptr = mmap(base + VM_MMAP_GUARD_SIZE, len, PROT_RW, flags, fd, 0); 499done: 500 if (fd >= 0) 501 close(fd); 502 return (ptr); 503} 504 505int 506vm_set_desc(struct vmctx *ctx, int vcpu, int reg, 507 uint64_t base, uint32_t limit, uint32_t access) 508{ 509 int error; 510 struct vm_seg_desc vmsegdesc; 511 512 bzero(&vmsegdesc, sizeof(vmsegdesc)); 513 vmsegdesc.cpuid = vcpu; 514 vmsegdesc.regnum = reg; 515 vmsegdesc.desc.base = base; 516 vmsegdesc.desc.limit = limit; 517 vmsegdesc.desc.access = access; 518 519 error = ioctl(ctx->fd, VM_SET_SEGMENT_DESCRIPTOR, &vmsegdesc); 520 return (error); 521} 522 523int 524vm_get_desc(struct vmctx *ctx, int vcpu, int reg, 525 uint64_t *base, uint32_t *limit, uint32_t *access) 526{ 527 int error; 528 struct vm_seg_desc vmsegdesc; 529 530 bzero(&vmsegdesc, sizeof(vmsegdesc)); 531 vmsegdesc.cpuid = vcpu; 532 vmsegdesc.regnum = reg; 533 534 error = ioctl(ctx->fd, VM_GET_SEGMENT_DESCRIPTOR, &vmsegdesc); 535 if (error == 0) { 536 *base = vmsegdesc.desc.base; 537 *limit = vmsegdesc.desc.limit; 538 *access = vmsegdesc.desc.access; 539 } 540 return (error); 541} 542 543int 544vm_get_seg_desc(struct vmctx *ctx, int vcpu, int reg, struct seg_desc *seg_desc) 545{ 546 int error; 547 548 error = vm_get_desc(ctx, vcpu, reg, &seg_desc->base, &seg_desc->limit, 549 &seg_desc->access); 550 return (error); 551} 552 553int 554vm_set_register(struct vmctx *ctx, int vcpu, int reg, uint64_t val) 555{ 556 int error; 557 struct vm_register vmreg; 558 559 bzero(&vmreg, sizeof(vmreg)); 560 vmreg.cpuid = vcpu; 561 vmreg.regnum = reg; 562 vmreg.regval = val; 563 564 error = ioctl(ctx->fd, VM_SET_REGISTER, &vmreg); 565 return (error); 566} 567 568int 569vm_get_register(struct vmctx *ctx, int vcpu, int reg, uint64_t *ret_val) 570{ 571 int error; 572 struct vm_register vmreg; 573 574 bzero(&vmreg, sizeof(vmreg)); 575 vmreg.cpuid = vcpu; 576 vmreg.regnum = reg; 577 578 error = ioctl(ctx->fd, VM_GET_REGISTER, &vmreg); 579 *ret_val = vmreg.regval; 580 return (error); 581} 582 583int 584vm_run(struct vmctx *ctx, int vcpu, struct vm_exit *vmexit) 585{ 586 int error; 587 struct vm_run vmrun; 588 589 bzero(&vmrun, sizeof(vmrun)); 590 vmrun.cpuid = vcpu; 591 592 error = ioctl(ctx->fd, VM_RUN, &vmrun); 593 bcopy(&vmrun.vm_exit, vmexit, sizeof(struct vm_exit)); 594 return (error); 595} 596 597int 598vm_suspend(struct vmctx *ctx, enum vm_suspend_how how) 599{ 600 struct vm_suspend vmsuspend; 601 602 bzero(&vmsuspend, sizeof(vmsuspend)); 603 vmsuspend.how = how; 604 return (ioctl(ctx->fd, VM_SUSPEND, &vmsuspend)); 605} 606 607int 608vm_reinit(struct vmctx *ctx) 609{ 610 611 return (ioctl(ctx->fd, VM_REINIT, 0)); 612} 613 614int 615vm_inject_exception(struct vmctx *ctx, int vcpu, int vector, int errcode_valid, 616 uint32_t errcode, int restart_instruction) 617{ 618 struct vm_exception exc; 619 620 exc.cpuid = vcpu; 621 exc.vector = vector; 622 exc.error_code = errcode; 623 exc.error_code_valid = errcode_valid; 624 exc.restart_instruction = restart_instruction; 625 626 return (ioctl(ctx->fd, VM_INJECT_EXCEPTION, &exc)); 627} 628 629int 630vm_apicid2vcpu(struct vmctx *ctx, int apicid) 631{ 632 /* 633 * The apic id associated with the 'vcpu' has the same numerical value 634 * as the 'vcpu' itself. 635 */ 636 return (apicid); 637} 638 639int 640vm_lapic_irq(struct vmctx *ctx, int vcpu, int vector) 641{ 642 struct vm_lapic_irq vmirq; 643 644 bzero(&vmirq, sizeof(vmirq)); 645 vmirq.cpuid = vcpu; 646 vmirq.vector = vector; 647 648 return (ioctl(ctx->fd, VM_LAPIC_IRQ, &vmirq)); 649} 650 651int 652vm_lapic_local_irq(struct vmctx *ctx, int vcpu, int vector) 653{ 654 struct vm_lapic_irq vmirq; 655 656 bzero(&vmirq, sizeof(vmirq)); 657 vmirq.cpuid = vcpu; 658 vmirq.vector = vector; 659 660 return (ioctl(ctx->fd, VM_LAPIC_LOCAL_IRQ, &vmirq)); 661} 662 663int 664vm_lapic_msi(struct vmctx *ctx, uint64_t addr, uint64_t msg) 665{ 666 struct vm_lapic_msi vmmsi; 667 668 bzero(&vmmsi, sizeof(vmmsi)); 669 vmmsi.addr = addr; 670 vmmsi.msg = msg; 671 672 return (ioctl(ctx->fd, VM_LAPIC_MSI, &vmmsi)); 673} 674 675int 676vm_ioapic_assert_irq(struct vmctx *ctx, int irq) 677{ 678 struct vm_ioapic_irq ioapic_irq; 679 680 bzero(&ioapic_irq, sizeof(struct vm_ioapic_irq)); 681 ioapic_irq.irq = irq; 682 683 return (ioctl(ctx->fd, VM_IOAPIC_ASSERT_IRQ, &ioapic_irq)); 684} 685 686int 687vm_ioapic_deassert_irq(struct vmctx *ctx, int irq) 688{ 689 struct vm_ioapic_irq ioapic_irq; 690 691 bzero(&ioapic_irq, sizeof(struct vm_ioapic_irq)); 692 ioapic_irq.irq = irq; 693 694 return (ioctl(ctx->fd, VM_IOAPIC_DEASSERT_IRQ, &ioapic_irq)); 695} 696 697int 698vm_ioapic_pulse_irq(struct vmctx *ctx, int irq) 699{ 700 struct vm_ioapic_irq ioapic_irq; 701 702 bzero(&ioapic_irq, sizeof(struct vm_ioapic_irq)); 703 ioapic_irq.irq = irq; 704 705 return (ioctl(ctx->fd, VM_IOAPIC_PULSE_IRQ, &ioapic_irq)); 706} 707 708int 709vm_ioapic_pincount(struct vmctx *ctx, int *pincount) 710{ 711 712 return (ioctl(ctx->fd, VM_IOAPIC_PINCOUNT, pincount)); 713} 714 715int 716vm_isa_assert_irq(struct vmctx *ctx, int atpic_irq, int ioapic_irq) 717{ 718 struct vm_isa_irq isa_irq; 719 720 bzero(&isa_irq, sizeof(struct vm_isa_irq)); 721 isa_irq.atpic_irq = atpic_irq; 722 isa_irq.ioapic_irq = ioapic_irq; 723 724 return (ioctl(ctx->fd, VM_ISA_ASSERT_IRQ, &isa_irq)); 725} 726 727int 728vm_isa_deassert_irq(struct vmctx *ctx, int atpic_irq, int ioapic_irq) 729{ 730 struct vm_isa_irq isa_irq; 731 732 bzero(&isa_irq, sizeof(struct vm_isa_irq)); 733 isa_irq.atpic_irq = atpic_irq; 734 isa_irq.ioapic_irq = ioapic_irq; 735 736 return (ioctl(ctx->fd, VM_ISA_DEASSERT_IRQ, &isa_irq)); 737} 738 739int 740vm_isa_pulse_irq(struct vmctx *ctx, int atpic_irq, int ioapic_irq) 741{ 742 struct vm_isa_irq isa_irq; 743 744 bzero(&isa_irq, sizeof(struct vm_isa_irq)); 745 isa_irq.atpic_irq = atpic_irq; 746 isa_irq.ioapic_irq = ioapic_irq; 747 748 return (ioctl(ctx->fd, VM_ISA_PULSE_IRQ, &isa_irq)); 749} 750 751int 752vm_isa_set_irq_trigger(struct vmctx *ctx, int atpic_irq, 753 enum vm_intr_trigger trigger) 754{ 755 struct vm_isa_irq_trigger isa_irq_trigger; 756 757 bzero(&isa_irq_trigger, sizeof(struct vm_isa_irq_trigger)); 758 isa_irq_trigger.atpic_irq = atpic_irq; 759 isa_irq_trigger.trigger = trigger; 760 761 return (ioctl(ctx->fd, VM_ISA_SET_IRQ_TRIGGER, &isa_irq_trigger)); 762} 763 764int 765vm_inject_nmi(struct vmctx *ctx, int vcpu) 766{ 767 struct vm_nmi vmnmi; 768 769 bzero(&vmnmi, sizeof(vmnmi)); 770 vmnmi.cpuid = vcpu; 771 772 return (ioctl(ctx->fd, VM_INJECT_NMI, &vmnmi)); 773} 774 775static struct { 776 const char *name; 777 int type; 778} capstrmap[] = { 779 { "hlt_exit", VM_CAP_HALT_EXIT }, 780 { "mtrap_exit", VM_CAP_MTRAP_EXIT }, 781 { "pause_exit", VM_CAP_PAUSE_EXIT }, 782 { "unrestricted_guest", VM_CAP_UNRESTRICTED_GUEST }, 783 { "enable_invpcid", VM_CAP_ENABLE_INVPCID }, 784 { 0 } 785}; 786 787int 788vm_capability_name2type(const char *capname) 789{ 790 int i; 791 792 for (i = 0; capstrmap[i].name != NULL && capname != NULL; i++) { 793 if (strcmp(capstrmap[i].name, capname) == 0) 794 return (capstrmap[i].type); 795 } 796 797 return (-1); 798} 799 800const char * 801vm_capability_type2name(int type) 802{ 803 int i; 804 805 for (i = 0; capstrmap[i].name != NULL; i++) { 806 if (capstrmap[i].type == type) 807 return (capstrmap[i].name); 808 } 809 810 return (NULL); 811} 812 813int 814vm_get_capability(struct vmctx *ctx, int vcpu, enum vm_cap_type cap, 815 int *retval) 816{ 817 int error; 818 struct vm_capability vmcap; 819 820 bzero(&vmcap, sizeof(vmcap)); 821 vmcap.cpuid = vcpu; 822 vmcap.captype = cap; 823 824 error = ioctl(ctx->fd, VM_GET_CAPABILITY, &vmcap); 825 *retval = vmcap.capval; 826 return (error); 827} 828 829int 830vm_set_capability(struct vmctx *ctx, int vcpu, enum vm_cap_type cap, int val) 831{ 832 struct vm_capability vmcap; 833 834 bzero(&vmcap, sizeof(vmcap)); 835 vmcap.cpuid = vcpu; 836 vmcap.captype = cap; 837 vmcap.capval = val; 838 839 return (ioctl(ctx->fd, VM_SET_CAPABILITY, &vmcap)); 840} 841 842int 843vm_assign_pptdev(struct vmctx *ctx, int bus, int slot, int func) 844{ 845 struct vm_pptdev pptdev; 846 847 bzero(&pptdev, sizeof(pptdev)); 848 pptdev.bus = bus; 849 pptdev.slot = slot; 850 pptdev.func = func; 851 852 return (ioctl(ctx->fd, VM_BIND_PPTDEV, &pptdev)); 853} 854 855int 856vm_unassign_pptdev(struct vmctx *ctx, int bus, int slot, int func) 857{ 858 struct vm_pptdev pptdev; 859 860 bzero(&pptdev, sizeof(pptdev)); 861 pptdev.bus = bus; 862 pptdev.slot = slot; 863 pptdev.func = func; 864 865 return (ioctl(ctx->fd, VM_UNBIND_PPTDEV, &pptdev)); 866} 867 868int 869vm_map_pptdev_mmio(struct vmctx *ctx, int bus, int slot, int func, 870 vm_paddr_t gpa, size_t len, vm_paddr_t hpa) 871{ 872 struct vm_pptdev_mmio pptmmio; 873 874 bzero(&pptmmio, sizeof(pptmmio)); 875 pptmmio.bus = bus; 876 pptmmio.slot = slot; 877 pptmmio.func = func; 878 pptmmio.gpa = gpa; 879 pptmmio.len = len; 880 pptmmio.hpa = hpa; 881 882 return (ioctl(ctx->fd, VM_MAP_PPTDEV_MMIO, &pptmmio)); 883} 884 885int 886vm_setup_pptdev_msi(struct vmctx *ctx, int vcpu, int bus, int slot, int func, 887 uint64_t addr, uint64_t msg, int numvec) 888{ 889 struct vm_pptdev_msi pptmsi; 890 891 bzero(&pptmsi, sizeof(pptmsi)); 892 pptmsi.vcpu = vcpu; 893 pptmsi.bus = bus; 894 pptmsi.slot = slot; 895 pptmsi.func = func; 896 pptmsi.msg = msg; 897 pptmsi.addr = addr; 898 pptmsi.numvec = numvec; 899 900 return (ioctl(ctx->fd, VM_PPTDEV_MSI, &pptmsi)); 901} 902 903int 904vm_setup_pptdev_msix(struct vmctx *ctx, int vcpu, int bus, int slot, int func, 905 int idx, uint64_t addr, uint64_t msg, uint32_t vector_control) 906{ 907 struct vm_pptdev_msix pptmsix; 908 909 bzero(&pptmsix, sizeof(pptmsix)); 910 pptmsix.vcpu = vcpu; 911 pptmsix.bus = bus; 912 pptmsix.slot = slot; 913 pptmsix.func = func; 914 pptmsix.idx = idx; 915 pptmsix.msg = msg; 916 pptmsix.addr = addr; 917 pptmsix.vector_control = vector_control; 918 919 return ioctl(ctx->fd, VM_PPTDEV_MSIX, &pptmsix); 920} 921 922uint64_t * 923vm_get_stats(struct vmctx *ctx, int vcpu, struct timeval *ret_tv, 924 int *ret_entries) 925{ 926 int error; 927 928 static struct vm_stats vmstats; 929 930 vmstats.cpuid = vcpu; 931 932 error = ioctl(ctx->fd, VM_STATS, &vmstats); 933 if (error == 0) { 934 if (ret_entries) 935 *ret_entries = vmstats.num_entries; 936 if (ret_tv) 937 *ret_tv = vmstats.tv; 938 return (vmstats.statbuf); 939 } else 940 return (NULL); 941} 942 943const char * 944vm_get_stat_desc(struct vmctx *ctx, int index) 945{ 946 static struct vm_stat_desc statdesc; 947 948 statdesc.index = index; 949 if (ioctl(ctx->fd, VM_STAT_DESC, &statdesc) == 0) 950 return (statdesc.desc); 951 else 952 return (NULL); 953} 954 955int 956vm_get_x2apic_state(struct vmctx *ctx, int vcpu, enum x2apic_state *state) 957{ 958 int error; 959 struct vm_x2apic x2apic; 960 961 bzero(&x2apic, sizeof(x2apic)); 962 x2apic.cpuid = vcpu; 963 964 error = ioctl(ctx->fd, VM_GET_X2APIC_STATE, &x2apic); 965 *state = x2apic.state; 966 return (error); 967} 968 969int 970vm_set_x2apic_state(struct vmctx *ctx, int vcpu, enum x2apic_state state) 971{ 972 int error; 973 struct vm_x2apic x2apic; 974 975 bzero(&x2apic, sizeof(x2apic)); 976 x2apic.cpuid = vcpu; 977 x2apic.state = state; 978 979 error = ioctl(ctx->fd, VM_SET_X2APIC_STATE, &x2apic); 980 981 return (error); 982} 983 984/* 985 * From Intel Vol 3a: 986 * Table 9-1. IA-32 Processor States Following Power-up, Reset or INIT 987 */ 988int 989vcpu_reset(struct vmctx *vmctx, int vcpu) 990{ 991 int error; 992 uint64_t rflags, rip, cr0, cr4, zero, desc_base, rdx; 993 uint32_t desc_access, desc_limit; 994 uint16_t sel; 995 996 zero = 0; 997 998 rflags = 0x2; 999 error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_RFLAGS, rflags); 1000 if (error) 1001 goto done; 1002 1003 rip = 0xfff0; 1004 if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_RIP, rip)) != 0) 1005 goto done; 1006 1007 cr0 = CR0_NE; 1008 if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_CR0, cr0)) != 0) 1009 goto done; 1010 1011 if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_CR3, zero)) != 0) 1012 goto done; 1013 1014 cr4 = 0; 1015 if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_CR4, cr4)) != 0) 1016 goto done; 1017 1018 /* 1019 * CS: present, r/w, accessed, 16-bit, byte granularity, usable 1020 */ 1021 desc_base = 0xffff0000; 1022 desc_limit = 0xffff; 1023 desc_access = 0x0093; 1024 error = vm_set_desc(vmctx, vcpu, VM_REG_GUEST_CS, 1025 desc_base, desc_limit, desc_access); 1026 if (error) 1027 goto done; 1028 1029 sel = 0xf000; 1030 if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_CS, sel)) != 0) 1031 goto done; 1032 1033 /* 1034 * SS,DS,ES,FS,GS: present, r/w, accessed, 16-bit, byte granularity 1035 */ 1036 desc_base = 0; 1037 desc_limit = 0xffff; 1038 desc_access = 0x0093; 1039 error = vm_set_desc(vmctx, vcpu, VM_REG_GUEST_SS, 1040 desc_base, desc_limit, desc_access); 1041 if (error) 1042 goto done; 1043 1044 error = vm_set_desc(vmctx, vcpu, VM_REG_GUEST_DS, 1045 desc_base, desc_limit, desc_access); 1046 if (error) 1047 goto done; 1048 1049 error = vm_set_desc(vmctx, vcpu, VM_REG_GUEST_ES, 1050 desc_base, desc_limit, desc_access); 1051 if (error) 1052 goto done; 1053 1054 error = vm_set_desc(vmctx, vcpu, VM_REG_GUEST_FS, 1055 desc_base, desc_limit, desc_access); 1056 if (error) 1057 goto done; 1058 1059 error = vm_set_desc(vmctx, vcpu, VM_REG_GUEST_GS, 1060 desc_base, desc_limit, desc_access); 1061 if (error) 1062 goto done; 1063 1064 sel = 0; 1065 if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_SS, sel)) != 0) 1066 goto done; 1067 if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_DS, sel)) != 0) 1068 goto done; 1069 if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_ES, sel)) != 0) 1070 goto done; 1071 if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_FS, sel)) != 0) 1072 goto done; 1073 if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_GS, sel)) != 0) 1074 goto done; 1075 1076 /* General purpose registers */ 1077 rdx = 0xf00; 1078 if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_RAX, zero)) != 0) 1079 goto done; 1080 if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_RBX, zero)) != 0) 1081 goto done; 1082 if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_RCX, zero)) != 0) 1083 goto done; 1084 if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_RDX, rdx)) != 0) 1085 goto done; 1086 if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_RSI, zero)) != 0) 1087 goto done; 1088 if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_RDI, zero)) != 0) 1089 goto done; 1090 if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_RBP, zero)) != 0) 1091 goto done; 1092 if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_RSP, zero)) != 0) 1093 goto done; 1094 1095 /* GDTR, IDTR */ 1096 desc_base = 0; 1097 desc_limit = 0xffff; 1098 desc_access = 0; 1099 error = vm_set_desc(vmctx, vcpu, VM_REG_GUEST_GDTR, 1100 desc_base, desc_limit, desc_access); 1101 if (error != 0) 1102 goto done; 1103 1104 error = vm_set_desc(vmctx, vcpu, VM_REG_GUEST_IDTR, 1105 desc_base, desc_limit, desc_access); 1106 if (error != 0) 1107 goto done; 1108 1109 /* TR */ 1110 desc_base = 0; 1111 desc_limit = 0xffff; 1112 desc_access = 0x0000008b; 1113 error = vm_set_desc(vmctx, vcpu, VM_REG_GUEST_TR, 0, 0, desc_access); 1114 if (error) 1115 goto done; 1116 1117 sel = 0; 1118 if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_TR, sel)) != 0) 1119 goto done; 1120 1121 /* LDTR */ 1122 desc_base = 0; 1123 desc_limit = 0xffff; 1124 desc_access = 0x00000082; 1125 error = vm_set_desc(vmctx, vcpu, VM_REG_GUEST_LDTR, desc_base, 1126 desc_limit, desc_access); 1127 if (error) 1128 goto done; 1129 1130 sel = 0; 1131 if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_LDTR, 0)) != 0) 1132 goto done; 1133 1134 /* XXX cr2, debug registers */ 1135 1136 error = 0; 1137done: 1138 return (error); 1139} 1140 1141int 1142vm_get_gpa_pmap(struct vmctx *ctx, uint64_t gpa, uint64_t *pte, int *num) 1143{ 1144 int error, i; 1145 struct vm_gpa_pte gpapte; 1146 1147 bzero(&gpapte, sizeof(gpapte)); 1148 gpapte.gpa = gpa; 1149 1150 error = ioctl(ctx->fd, VM_GET_GPA_PMAP, &gpapte); 1151 1152 if (error == 0) { 1153 *num = gpapte.ptenum; 1154 for (i = 0; i < gpapte.ptenum; i++) 1155 pte[i] = gpapte.pte[i]; 1156 } 1157 1158 return (error); 1159} 1160 1161int 1162vm_get_hpet_capabilities(struct vmctx *ctx, uint32_t *capabilities) 1163{ 1164 int error; 1165 struct vm_hpet_cap cap; 1166 1167 bzero(&cap, sizeof(struct vm_hpet_cap)); 1168 error = ioctl(ctx->fd, VM_GET_HPET_CAPABILITIES, &cap); 1169 if (capabilities != NULL) 1170 *capabilities = cap.capabilities; 1171 return (error); 1172} 1173 1174int 1175vm_gla2gpa(struct vmctx *ctx, int vcpu, struct vm_guest_paging *paging, 1176 uint64_t gla, int prot, uint64_t *gpa, int *fault) 1177{ 1178 struct vm_gla2gpa gg; 1179 int error; 1180 1181 bzero(&gg, sizeof(struct vm_gla2gpa)); 1182 gg.vcpuid = vcpu; 1183 gg.prot = prot; 1184 gg.gla = gla; 1185 gg.paging = *paging; 1186 1187 error = ioctl(ctx->fd, VM_GLA2GPA, &gg); 1188 if (error == 0) { 1189 *fault = gg.fault; 1190 *gpa = gg.gpa; 1191 } 1192 return (error); 1193} 1194 1195#ifndef min 1196#define min(a,b) (((a) < (b)) ? (a) : (b)) 1197#endif 1198 1199int 1200vm_copy_setup(struct vmctx *ctx, int vcpu, struct vm_guest_paging *paging, 1201 uint64_t gla, size_t len, int prot, struct iovec *iov, int iovcnt, 1202 int *fault) 1203{ 1204 void *va; 1205 uint64_t gpa; 1206 int error, i, n, off; 1207 1208 for (i = 0; i < iovcnt; i++) { 1209 iov[i].iov_base = 0; 1210 iov[i].iov_len = 0; 1211 } 1212 1213 while (len) { 1214 assert(iovcnt > 0); 1215 error = vm_gla2gpa(ctx, vcpu, paging, gla, prot, &gpa, fault); 1216 if (error || *fault) 1217 return (error); 1218 1219 off = gpa & PAGE_MASK; 1220 n = min(len, PAGE_SIZE - off); 1221 1222 va = vm_map_gpa(ctx, gpa, n); 1223 if (va == NULL) 1224 return (EFAULT); 1225 1226 iov->iov_base = va; 1227 iov->iov_len = n; 1228 iov++; 1229 iovcnt--; 1230 1231 gla += n; 1232 len -= n; 1233 } 1234 return (0); 1235} 1236 1237void 1238vm_copy_teardown(struct vmctx *ctx, int vcpu, struct iovec *iov, int iovcnt) 1239{ 1240 1241 return; 1242} 1243 1244void 1245vm_copyin(struct vmctx *ctx, int vcpu, struct iovec *iov, void *vp, size_t len) 1246{ 1247 const char *src; 1248 char *dst; 1249 size_t n; 1250 1251 dst = vp; 1252 while (len) { 1253 assert(iov->iov_len); 1254 n = min(len, iov->iov_len); 1255 src = iov->iov_base; 1256 bcopy(src, dst, n); 1257 1258 iov++; 1259 dst += n; 1260 len -= n; 1261 } 1262} 1263 1264void 1265vm_copyout(struct vmctx *ctx, int vcpu, const void *vp, struct iovec *iov, 1266 size_t len) 1267{ 1268 const char *src; 1269 char *dst; 1270 size_t n; 1271 1272 src = vp; 1273 while (len) { 1274 assert(iov->iov_len); 1275 n = min(len, iov->iov_len); 1276 dst = iov->iov_base; 1277 bcopy(src, dst, n); 1278 1279 iov++; 1280 src += n; 1281 len -= n; 1282 } 1283} 1284 1285static int 1286vm_get_cpus(struct vmctx *ctx, int which, cpuset_t *cpus) 1287{ 1288 struct vm_cpuset vm_cpuset; 1289 int error; 1290 1291 bzero(&vm_cpuset, sizeof(struct vm_cpuset)); 1292 vm_cpuset.which = which; 1293 vm_cpuset.cpusetsize = sizeof(cpuset_t); 1294 vm_cpuset.cpus = cpus; 1295 1296 error = ioctl(ctx->fd, VM_GET_CPUS, &vm_cpuset); 1297 return (error); 1298} 1299 1300int 1301vm_active_cpus(struct vmctx *ctx, cpuset_t *cpus) 1302{ 1303 1304 return (vm_get_cpus(ctx, VM_ACTIVE_CPUS, cpus)); 1305} 1306 1307int 1308vm_suspended_cpus(struct vmctx *ctx, cpuset_t *cpus) 1309{ 1310 1311 return (vm_get_cpus(ctx, VM_SUSPENDED_CPUS, cpus)); 1312} 1313 1314int 1315vm_activate_cpu(struct vmctx *ctx, int vcpu) 1316{ 1317 struct vm_activate_cpu ac; 1318 int error; 1319 1320 bzero(&ac, sizeof(struct vm_activate_cpu)); 1321 ac.vcpuid = vcpu; 1322 error = ioctl(ctx->fd, VM_ACTIVATE_CPU, &ac); 1323 return (error); 1324} 1325 1326int 1327vm_get_intinfo(struct vmctx *ctx, int vcpu, uint64_t *info1, uint64_t *info2) 1328{ 1329 struct vm_intinfo vmii; 1330 int error; 1331 1332 bzero(&vmii, sizeof(struct vm_intinfo)); 1333 vmii.vcpuid = vcpu; 1334 error = ioctl(ctx->fd, VM_GET_INTINFO, &vmii); 1335 if (error == 0) { 1336 *info1 = vmii.info1; 1337 *info2 = vmii.info2; 1338 } 1339 return (error); 1340} 1341 1342int 1343vm_set_intinfo(struct vmctx *ctx, int vcpu, uint64_t info1) 1344{ 1345 struct vm_intinfo vmii; 1346 int error; 1347 1348 bzero(&vmii, sizeof(struct vm_intinfo)); 1349 vmii.vcpuid = vcpu; 1350 vmii.info1 = info1; 1351 error = ioctl(ctx->fd, VM_SET_INTINFO, &vmii); 1352 return (error); 1353} 1354 1355int 1356vm_rtc_write(struct vmctx *ctx, int offset, uint8_t value) 1357{ 1358 struct vm_rtc_data rtcdata; 1359 int error; 1360 1361 bzero(&rtcdata, sizeof(struct vm_rtc_data)); 1362 rtcdata.offset = offset; 1363 rtcdata.value = value; 1364 error = ioctl(ctx->fd, VM_RTC_WRITE, &rtcdata); 1365 return (error); 1366} 1367 1368int 1369vm_rtc_read(struct vmctx *ctx, int offset, uint8_t *retval) 1370{ 1371 struct vm_rtc_data rtcdata; 1372 int error; 1373 1374 bzero(&rtcdata, sizeof(struct vm_rtc_data)); 1375 rtcdata.offset = offset; 1376 error = ioctl(ctx->fd, VM_RTC_READ, &rtcdata); 1377 if (error == 0) 1378 *retval = rtcdata.value; 1379 return (error); 1380} 1381 1382int 1383vm_rtc_settime(struct vmctx *ctx, time_t secs) 1384{ 1385 struct vm_rtc_time rtctime; 1386 int error; 1387 1388 bzero(&rtctime, sizeof(struct vm_rtc_time)); 1389 rtctime.secs = secs; 1390 error = ioctl(ctx->fd, VM_RTC_SETTIME, &rtctime); 1391 return (error); 1392} 1393 1394int 1395vm_rtc_gettime(struct vmctx *ctx, time_t *secs) 1396{ 1397 struct vm_rtc_time rtctime; 1398 int error; 1399 1400 bzero(&rtctime, sizeof(struct vm_rtc_time)); 1401 error = ioctl(ctx->fd, VM_RTC_GETTIME, &rtctime); 1402 if (error == 0) 1403 *secs = rtctime.secs; 1404 return (error); 1405} 1406 1407int 1408vm_restart_instruction(void *arg, int vcpu) 1409{ 1410 struct vmctx *ctx = arg; 1411 1412 return (ioctl(ctx->fd, VM_RESTART_INSTRUCTION, &vcpu)); 1413} 1414