sys/amd64/vmm/vmm_dev.c

   1 /*-
   2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
   3  *
   4  * Copyright (c) 2011 NetApp, Inc.
   5  * All rights reserved.
   6  *
   7  * Redistribution and use in source and binary forms, with or without
   8  * modification, are permitted provided that the following conditions
   9  * are met:
  10  * 1. Redistributions of source code must retain the above copyright
  11  *    notice, this list of conditions and the following disclaimer.
  12  * 2. Redistributions in binary form must reproduce the above copyright
  13  *    notice, this list of conditions and the following disclaimer in the
  14  *    documentation and/or other materials provided with the distribution.
  15  *
  16  * THIS SOFTWARE IS PROVIDED BY NETAPP, INC ``AS IS'' AND
  17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  19  * ARE DISCLAIMED.  IN NO EVENT SHALL NETAPP, INC OR CONTRIBUTORS BE LIABLE
  20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  26  * SUCH DAMAGE.
  27  *
  28  * $FreeBSD$
  29  */
  30
  31 #include <sys/cdefs.h>
  32 __FBSDID("$FreeBSD$");
  33
  34 #include <sys/param.h>
  35 #include <sys/kernel.h>
  36 #include <sys/queue.h>
  37 #include <sys/lock.h>
  38 #include <sys/mutex.h>
  39 #include <sys/malloc.h>
  40 #include <sys/conf.h>
  41 #include <sys/sysctl.h>
  42 #include <sys/libkern.h>
  43 #include <sys/ioccom.h>
  44 #include <sys/mman.h>
  45 #include <sys/uio.h>
  46
  47 #include <vm/vm.h>
  48 #include <vm/pmap.h>
  49 #include <vm/vm_map.h>
  50 #include <vm/vm_object.h>
  51
  52 #include <machine/vmparam.h>
  53 #include <machine/vmm.h>
  54 #include <machine/vmm_instruction_emul.h>
  55 #include <machine/vmm_dev.h>
  56
  57 #include "vmm_lapic.h"
  58 #include "vmm_stat.h"
  59 #include "vmm_mem.h"
  60 #include "io/ppt.h"
  61 #include "io/vatpic.h"
  62 #include "io/vioapic.h"
  63 #include "io/vhpet.h"
  64 #include "io/vrtc.h"
  65
  66 struct devmem_softc {
  67         int     segid;
  68         char    *name;
  69         struct cdev *cdev;
  70         struct vmmdev_softc *sc;
  71         SLIST_ENTRY(devmem_softc) link;
  72 };
  73
  74 struct vmmdev_softc {
  75         struct vm       *vm;            /* vm instance cookie */
  76         struct cdev     *cdev;
  77         SLIST_ENTRY(vmmdev_softc) link;
  78         SLIST_HEAD(, devmem_softc) devmem;
  79         int             flags;
  80 };
  81 #define VSC_LINKED              0x01
  82
  83 static SLIST_HEAD(, vmmdev_softc) head;
  84
  85 static struct mtx vmmdev_mtx;
  86
  87 static MALLOC_DEFINE(M_VMMDEV, "vmmdev", "vmmdev");
  88
  89 SYSCTL_DECL(_hw_vmm);
  90
  91 static int devmem_create_cdev(const char *vmname, int id, char *devmem);
  92 static void devmem_destroy(void *arg);
  93
  94 static int
  95 vcpu_lock_one(struct vmmdev_softc *sc, int vcpu)
  96 {
  97         int error;
  98
  99         if (vcpu < 0 || vcpu >= VM_MAXCPU)
 100                 return (EINVAL);
 101
 102         error = vcpu_set_state(sc->vm, vcpu, VCPU_FROZEN, true);
 103         return (error);
 104 }
 105
 106 static void
 107 vcpu_unlock_one(struct vmmdev_softc *sc, int vcpu)
 108 {
 109         enum vcpu_state state;
 110
 111         state = vcpu_get_state(sc->vm, vcpu, NULL);
 112         if (state != VCPU_FROZEN) {
 113                 panic("vcpu %s(%d) has invalid state %d", vm_name(sc->vm),
 114                     vcpu, state);
 115         }
 116
 117         vcpu_set_state(sc->vm, vcpu, VCPU_IDLE, false);
 118 }
 119
 120 static int
 121 vcpu_lock_all(struct vmmdev_softc *sc)
 122 {
 123         int error, vcpu;
 124
 125         for (vcpu = 0; vcpu < VM_MAXCPU; vcpu++) {
 126                 error = vcpu_lock_one(sc, vcpu);
 127                 if (error)
 128                         break;
 129         }
 130
 131         if (error) {
 132                 while (--vcpu >= 0)
 133                         vcpu_unlock_one(sc, vcpu);
 134         }
 135
 136         return (error);
 137 }
 138
 139 static void
 140 vcpu_unlock_all(struct vmmdev_softc *sc)
 141 {
 142         int vcpu;
 143
 144         for (vcpu = 0; vcpu < VM_MAXCPU; vcpu++)
 145                 vcpu_unlock_one(sc, vcpu);
 146 }
 147
 148 static struct vmmdev_softc *
 149 vmmdev_lookup(const char *name)
 150 {
 151         struct vmmdev_softc *sc;
 152
 153 #ifdef notyet   /* XXX kernel is not compiled with invariants */
 154         mtx_assert(&vmmdev_mtx, MA_OWNED);
 155 #endif
 156
 157         SLIST_FOREACH(sc, &head, link) {
 158                 if (strcmp(name, vm_name(sc->vm)) == 0)
 159                         break;
 160         }
 161
 162         return (sc);
 163 }
 164
 165 static struct vmmdev_softc *
 166 vmmdev_lookup2(struct cdev *cdev)
 167 {
 168
 169         return (cdev->si_drv1);
 170 }
 171
 172 static int
 173 vmmdev_rw(struct cdev *cdev, struct uio *uio, int flags)
 174 {
 175         int error, off, c, prot;
 176         vm_paddr_t gpa;
 177         void *hpa, *cookie;
 178         struct vmmdev_softc *sc;
 179
 180         sc = vmmdev_lookup2(cdev);
 181         if (sc == NULL)
 182                 return (ENXIO);
 183
 184         /*
 185          * Get a read lock on the guest memory map by freezing any vcpu.
 186          */
 187         error = vcpu_lock_one(sc, VM_MAXCPU - 1);
 188         if (error)
 189                 return (error);
 190
 191         prot = (uio->uio_rw == UIO_WRITE ? VM_PROT_WRITE : VM_PROT_READ);
 192         while (uio->uio_resid > 0 && error == 0) {
 193                 gpa = uio->uio_offset;
 194                 off = gpa & PAGE_MASK;
 195                 c = min(uio->uio_resid, PAGE_SIZE - off);
 196
 197                 /*
 198                  * The VM has a hole in its physical memory map. If we want to
 199                  * use 'dd' to inspect memory beyond the hole we need to
 200                  * provide bogus data for memory that lies in the hole.
 201                  *
 202                  * Since this device does not support lseek(2), dd(1) will
 203                  * read(2) blocks of data to simulate the lseek(2).
 204                  */
 205                 hpa = vm_gpa_hold(sc->vm, VM_MAXCPU - 1, gpa, c, prot, &cookie);
 206                 if (hpa == NULL) {
 207                         if (uio->uio_rw == UIO_READ)
 208                                 error = uiomove(__DECONST(void *, zero_region),
 209                                     c, uio);
 210                         else
 211                                 error = EFAULT;
 212                 } else {
 213                         error = uiomove(hpa, c, uio);
 214                         vm_gpa_release(cookie);
 215                 }
 216         }
 217         vcpu_unlock_one(sc, VM_MAXCPU - 1);
 218         return (error);
 219 }
 220
 221 CTASSERT(sizeof(((struct vm_memseg *)0)->name) >= SPECNAMELEN + 1);
 222
 223 static int
 224 get_memseg(struct vmmdev_softc *sc, struct vm_memseg *mseg)
 225 {
 226         struct devmem_softc *dsc;
 227         int error;
 228         bool sysmem;
 229
 230         error = vm_get_memseg(sc->vm, mseg->segid, &mseg->len, &sysmem, NULL);
 231         if (error || mseg->len == 0)
 232                 return (error);
 233
 234         if (!sysmem) {
 235                 SLIST_FOREACH(dsc, &sc->devmem, link) {
 236                         if (dsc->segid == mseg->segid)
 237                                 break;
 238                 }
 239                 KASSERT(dsc != NULL, ("%s: devmem segment %d not found",
 240                     __func__, mseg->segid));
 241                 error = copystr(dsc->name, mseg->name, SPECNAMELEN + 1, NULL);
 242         } else {
 243                 bzero(mseg->name, sizeof(mseg->name));
 244         }
 245
 246         return (error);
 247 }
 248
 249 static int
 250 alloc_memseg(struct vmmdev_softc *sc, struct vm_memseg *mseg)
 251 {
 252         char *name;
 253         int error;
 254         bool sysmem;
 255
 256         error = 0;
 257         name = NULL;
 258         sysmem = true;
 259
 260         if (VM_MEMSEG_NAME(mseg)) {
 261                 sysmem = false;
 262                 name = malloc(SPECNAMELEN + 1, M_VMMDEV, M_WAITOK);
 263                 error = copystr(mseg->name, name, SPECNAMELEN + 1, 0);
 264                 if (error)
 265                         goto done;
 266         }
 267
 268         error = vm_alloc_memseg(sc->vm, mseg->segid, mseg->len, sysmem);
 269         if (error)
 270                 goto done;
 271
 272         if (VM_MEMSEG_NAME(mseg)) {
 273                 error = devmem_create_cdev(vm_name(sc->vm), mseg->segid, name);
 274                 if (error)
 275                         vm_free_memseg(sc->vm, mseg->segid);
 276                 else
 277                         name = NULL;    /* freed when 'cdev' is destroyed */
 278         }
 279 done:
 280         free(name, M_VMMDEV);
 281         return (error);
 282 }
 283
 284 static int
 285 vm_get_register_set(struct vm *vm, int vcpu, unsigned int count, int *regnum,
 286     uint64_t *regval)
 287 {
 288         int error, i;
 289
 290         error = 0;
 291         for (i = 0; i < count; i++) {
 292                 error = vm_get_register(vm, vcpu, regnum[i], &regval[i]);
 293                 if (error)
 294                         break;
 295         }
 296         return (error);
 297 }
 298
 299 static int
 300 vm_set_register_set(struct vm *vm, int vcpu, unsigned int count, int *regnum,
 301     uint64_t *regval)
 302 {
 303         int error, i;
 304
 305         error = 0;
 306         for (i = 0; i < count; i++) {
 307                 error = vm_set_register(vm, vcpu, regnum[i], regval[i]);
 308                 if (error)
 309                         break;
 310         }
 311         return (error);
 312 }
 313
 314 static int
 315 vmmdev_ioctl(struct cdev *cdev, u_long cmd, caddr_t data, int fflag,
 316              struct thread *td)
 317 {
 318         int error, vcpu, state_changed, size;
 319         cpuset_t *cpuset;
 320         struct vmmdev_softc *sc;
 321         struct vm_register *vmreg;
 322         struct vm_seg_desc *vmsegdesc;
 323         struct vm_register_set *vmregset;
 324         struct vm_run *vmrun;
 325         struct vm_exception *vmexc;
 326         struct vm_lapic_irq *vmirq;
 327         struct vm_lapic_msi *vmmsi;
 328         struct vm_ioapic_irq *ioapic_irq;
 329         struct vm_isa_irq *isa_irq;
 330         struct vm_isa_irq_trigger *isa_irq_trigger;
 331         struct vm_capability *vmcap;
 332         struct vm_pptdev *pptdev;
 333         struct vm_pptdev_mmio *pptmmio;
 334         struct vm_pptdev_msi *pptmsi;
 335         struct vm_pptdev_msix *pptmsix;
 336         struct vm_nmi *vmnmi;
 337         struct vm_stats *vmstats;
 338         struct vm_stat_desc *statdesc;
 339         struct vm_x2apic *x2apic;
 340         struct vm_gpa_pte *gpapte;
 341         struct vm_suspend *vmsuspend;
 342         struct vm_gla2gpa *gg;
 343         struct vm_activate_cpu *vac;
 344         struct vm_cpuset *vm_cpuset;
 345         struct vm_intinfo *vmii;
 346         struct vm_rtc_time *rtctime;
 347         struct vm_rtc_data *rtcdata;
 348         struct vm_memmap *mm;
 349         uint64_t *regvals;
 350         int *regnums;
 351
 352         sc = vmmdev_lookup2(cdev);
 353         if (sc == NULL)
 354                 return (ENXIO);
 355
 356         error = 0;
 357         vcpu = -1;
 358         state_changed = 0;
 359
 360         /*
 361          * Some VMM ioctls can operate only on vcpus that are not running.
 362          */
 363         switch (cmd) {
 364         case VM_RUN:
 365         case VM_GET_REGISTER:
 366         case VM_SET_REGISTER:
 367         case VM_GET_SEGMENT_DESCRIPTOR:
 368         case VM_SET_SEGMENT_DESCRIPTOR:
 369         case VM_GET_REGISTER_SET:
 370         case VM_SET_REGISTER_SET:
 371         case VM_INJECT_EXCEPTION:
 372         case VM_GET_CAPABILITY:
 373         case VM_SET_CAPABILITY:
 374         case VM_PPTDEV_MSI:
 375         case VM_PPTDEV_MSIX:
 376         case VM_SET_X2APIC_STATE:
 377         case VM_GLA2GPA:
 378         case VM_ACTIVATE_CPU:
 379         case VM_SET_INTINFO:
 380         case VM_GET_INTINFO:
 381         case VM_RESTART_INSTRUCTION:
 382                 /*
 383                  * XXX fragile, handle with care
 384                  * Assumes that the first field of the ioctl data is the vcpu.
 385                  */
 386                 vcpu = *(int *)data;
 387                 error = vcpu_lock_one(sc, vcpu);
 388                 if (error)
 389                         goto done;
 390                 state_changed = 1;
 391                 break;
 392
 393         case VM_MAP_PPTDEV_MMIO:
 394         case VM_BIND_PPTDEV:
 395         case VM_UNBIND_PPTDEV:
 396         case VM_ALLOC_MEMSEG:
 397         case VM_MMAP_MEMSEG:
 398         case VM_REINIT:
 399                 /*
 400                  * ioctls that operate on the entire virtual machine must
 401                  * prevent all vcpus from running.
 402                  */
 403                 error = vcpu_lock_all(sc);
 404                 if (error)
 405                         goto done;
 406                 state_changed = 2;
 407                 break;
 408
 409         case VM_GET_MEMSEG:
 410         case VM_MMAP_GETNEXT:
 411                 /*
 412                  * Lock a vcpu to make sure that the memory map cannot be
 413                  * modified while it is being inspected.
 414                  */
 415                 vcpu = VM_MAXCPU - 1;
 416                 error = vcpu_lock_one(sc, vcpu);
 417                 if (error)
 418                         goto done;
 419                 state_changed = 1;
 420                 break;
 421
 422         default:
 423                 break;
 424         }
 425
 426         switch(cmd) {
 427         case VM_RUN:
 428                 vmrun = (struct vm_run *)data;
 429                 error = vm_run(sc->vm, vmrun);
 430                 break;
 431         case VM_SUSPEND:
 432                 vmsuspend = (struct vm_suspend *)data;
 433                 error = vm_suspend(sc->vm, vmsuspend->how);
 434                 break;
 435         case VM_REINIT:
 436                 error = vm_reinit(sc->vm);
 437                 break;
 438         case VM_STAT_DESC: {
 439                 statdesc = (struct vm_stat_desc *)data;
 440                 error = vmm_stat_desc_copy(statdesc->index,
 441                                         statdesc->desc, sizeof(statdesc->desc));
 442                 break;
 443         }
 444         case VM_STATS: {
 445                 CTASSERT(MAX_VM_STATS >= MAX_VMM_STAT_ELEMS);
 446                 vmstats = (struct vm_stats *)data;
 447                 getmicrotime(&vmstats->tv);
 448                 error = vmm_stat_copy(sc->vm, vmstats->cpuid,
 449                                       &vmstats->num_entries, vmstats->statbuf);
 450                 break;
 451         }
 452         case VM_PPTDEV_MSI:
 453                 pptmsi = (struct vm_pptdev_msi *)data;
 454                 error = ppt_setup_msi(sc->vm, pptmsi->vcpu,
 455                                       pptmsi->bus, pptmsi->slot, pptmsi->func,
 456                                       pptmsi->addr, pptmsi->msg,
 457                                       pptmsi->numvec);
 458                 break;
 459         case VM_PPTDEV_MSIX:
 460                 pptmsix = (struct vm_pptdev_msix *)data;
 461                 error = ppt_setup_msix(sc->vm, pptmsix->vcpu,
 462                                        pptmsix->bus, pptmsix->slot,
 463                                        pptmsix->func, pptmsix->idx,
 464                                        pptmsix->addr, pptmsix->msg,
 465                                        pptmsix->vector_control);
 466                 break;
 467         case VM_MAP_PPTDEV_MMIO:
 468                 pptmmio = (struct vm_pptdev_mmio *)data;
 469                 error = ppt_map_mmio(sc->vm, pptmmio->bus, pptmmio->slot,
 470                                      pptmmio->func, pptmmio->gpa, pptmmio->len,
 471                                      pptmmio->hpa);
 472                 break;
 473         case VM_BIND_PPTDEV:
 474                 pptdev = (struct vm_pptdev *)data;
 475                 error = vm_assign_pptdev(sc->vm, pptdev->bus, pptdev->slot,
 476                                          pptdev->func);
 477                 break;
 478         case VM_UNBIND_PPTDEV:
 479                 pptdev = (struct vm_pptdev *)data;
 480                 error = vm_unassign_pptdev(sc->vm, pptdev->bus, pptdev->slot,
 481                                            pptdev->func);
 482                 break;
 483         case VM_INJECT_EXCEPTION:
 484                 vmexc = (struct vm_exception *)data;
 485                 error = vm_inject_exception(sc->vm, vmexc->cpuid,
 486                     vmexc->vector, vmexc->error_code_valid, vmexc->error_code,
 487                     vmexc->restart_instruction);
 488                 break;
 489         case VM_INJECT_NMI:
 490                 vmnmi = (struct vm_nmi *)data;
 491                 error = vm_inject_nmi(sc->vm, vmnmi->cpuid);
 492                 break;
 493         case VM_LAPIC_IRQ:
 494                 vmirq = (struct vm_lapic_irq *)data;
 495                 error = lapic_intr_edge(sc->vm, vmirq->cpuid, vmirq->vector);
 496                 break;
 497         case VM_LAPIC_LOCAL_IRQ:
 498                 vmirq = (struct vm_lapic_irq *)data;
 499                 error = lapic_set_local_intr(sc->vm, vmirq->cpuid,
 500                     vmirq->vector);
 501                 break;
 502         case VM_LAPIC_MSI:
 503                 vmmsi = (struct vm_lapic_msi *)data;
 504                 error = lapic_intr_msi(sc->vm, vmmsi->addr, vmmsi->msg);
 505                 break;
 506         case VM_IOAPIC_ASSERT_IRQ:
 507                 ioapic_irq = (struct vm_ioapic_irq *)data;
 508                 error = vioapic_assert_irq(sc->vm, ioapic_irq->irq);
 509                 break;
 510         case VM_IOAPIC_DEASSERT_IRQ:
 511                 ioapic_irq = (struct vm_ioapic_irq *)data;
 512                 error = vioapic_deassert_irq(sc->vm, ioapic_irq->irq);
 513                 break;
 514         case VM_IOAPIC_PULSE_IRQ:
 515                 ioapic_irq = (struct vm_ioapic_irq *)data;
 516                 error = vioapic_pulse_irq(sc->vm, ioapic_irq->irq);
 517                 break;
 518         case VM_IOAPIC_PINCOUNT:
 519                 *(int *)data = vioapic_pincount(sc->vm);
 520                 break;
 521         case VM_ISA_ASSERT_IRQ:
 522                 isa_irq = (struct vm_isa_irq *)data;
 523                 error = vatpic_assert_irq(sc->vm, isa_irq->atpic_irq);
 524                 if (error == 0 && isa_irq->ioapic_irq != -1)
 525                         error = vioapic_assert_irq(sc->vm,
 526                             isa_irq->ioapic_irq);
 527                 break;
 528         case VM_ISA_DEASSERT_IRQ:
 529                 isa_irq = (struct vm_isa_irq *)data;
 530                 error = vatpic_deassert_irq(sc->vm, isa_irq->atpic_irq);
 531                 if (error == 0 && isa_irq->ioapic_irq != -1)
 532                         error = vioapic_deassert_irq(sc->vm,
 533                             isa_irq->ioapic_irq);
 534                 break;
 535         case VM_ISA_PULSE_IRQ:
 536                 isa_irq = (struct vm_isa_irq *)data;
 537                 error = vatpic_pulse_irq(sc->vm, isa_irq->atpic_irq);
 538                 if (error == 0 && isa_irq->ioapic_irq != -1)
 539                         error = vioapic_pulse_irq(sc->vm, isa_irq->ioapic_irq);
 540                 break;
 541         case VM_ISA_SET_IRQ_TRIGGER:
 542                 isa_irq_trigger = (struct vm_isa_irq_trigger *)data;
 543                 error = vatpic_set_irq_trigger(sc->vm,
 544                     isa_irq_trigger->atpic_irq, isa_irq_trigger->trigger);
 545                 break;
 546         case VM_MMAP_GETNEXT:
 547                 mm = (struct vm_memmap *)data;
 548                 error = vm_mmap_getnext(sc->vm, &mm->gpa, &mm->segid,
 549                     &mm->segoff, &mm->len, &mm->prot, &mm->flags);
 550                 break;
 551         case VM_MMAP_MEMSEG:
 552                 mm = (struct vm_memmap *)data;
 553                 error = vm_mmap_memseg(sc->vm, mm->gpa, mm->segid, mm->segoff,
 554                     mm->len, mm->prot, mm->flags);
 555                 break;
 556         case VM_ALLOC_MEMSEG:
 557                 error = alloc_memseg(sc, (struct vm_memseg *)data);
 558                 break;
 559         case VM_GET_MEMSEG:
 560                 error = get_memseg(sc, (struct vm_memseg *)data);
 561                 break;
 562         case VM_GET_REGISTER:
 563                 vmreg = (struct vm_register *)data;
 564                 error = vm_get_register(sc->vm, vmreg->cpuid, vmreg->regnum,
 565                                         &vmreg->regval);
 566                 break;
 567         case VM_SET_REGISTER:
 568                 vmreg = (struct vm_register *)data;
 569                 error = vm_set_register(sc->vm, vmreg->cpuid, vmreg->regnum,
 570                                         vmreg->regval);
 571                 break;
 572         case VM_SET_SEGMENT_DESCRIPTOR:
 573                 vmsegdesc = (struct vm_seg_desc *)data;
 574                 error = vm_set_seg_desc(sc->vm, vmsegdesc->cpuid,
 575                                         vmsegdesc->regnum,
 576                                         &vmsegdesc->desc);
 577                 break;
 578         case VM_GET_SEGMENT_DESCRIPTOR:
 579                 vmsegdesc = (struct vm_seg_desc *)data;
 580                 error = vm_get_seg_desc(sc->vm, vmsegdesc->cpuid,
 581                                         vmsegdesc->regnum,
 582                                         &vmsegdesc->desc);
 583                 break;
 584         case VM_GET_REGISTER_SET:
 585                 vmregset = (struct vm_register_set *)data;
 586                 if (vmregset->count > VM_REG_LAST) {
 587                         error = EINVAL;
 588                         break;
 589                 }
 590                 regvals = malloc(sizeof(regvals[0]) * vmregset->count, M_VMMDEV,
 591                     M_WAITOK);
 592                 regnums = malloc(sizeof(regnums[0]) * vmregset->count, M_VMMDEV,
 593                     M_WAITOK);
 594                 error = copyin(vmregset->regnums, regnums, sizeof(regnums[0]) *
 595                     vmregset->count);
 596                 if (error == 0)
 597                         error = vm_get_register_set(sc->vm, vmregset->cpuid,
 598                             vmregset->count, regnums, regvals);
 599                 if (error == 0)
 600                         error = copyout(regvals, vmregset->regvals,
 601                             sizeof(regvals[0]) * vmregset->count);
 602                 free(regvals, M_VMMDEV);
 603                 free(regnums, M_VMMDEV);
 604                 break;
 605         case VM_SET_REGISTER_SET:
 606                 vmregset = (struct vm_register_set *)data;
 607                 if (vmregset->count > VM_REG_LAST) {
 608                         error = EINVAL;
 609                         break;
 610                 }
 611                 regvals = malloc(sizeof(regvals[0]) * vmregset->count, M_VMMDEV,
 612                     M_WAITOK);
 613                 regnums = malloc(sizeof(regnums[0]) * vmregset->count, M_VMMDEV,
 614                     M_WAITOK);
 615                 error = copyin(vmregset->regnums, regnums, sizeof(regnums[0]) *
 616                     vmregset->count);
 617                 if (error == 0)
 618                         error = copyin(vmregset->regvals, regvals,
 619                             sizeof(regvals[0]) * vmregset->count);
 620                 if (error == 0)
 621                         error = vm_set_register_set(sc->vm, vmregset->cpuid,
 622                             vmregset->count, regnums, regvals);
 623                 free(regvals, M_VMMDEV);
 624                 free(regnums, M_VMMDEV);
 625                 break;
 626         case VM_GET_CAPABILITY:
 627                 vmcap = (struct vm_capability *)data;
 628                 error = vm_get_capability(sc->vm, vmcap->cpuid,
 629                                           vmcap->captype,
 630                                           &vmcap->capval);
 631                 break;
 632         case VM_SET_CAPABILITY:
 633                 vmcap = (struct vm_capability *)data;
 634                 error = vm_set_capability(sc->vm, vmcap->cpuid,
 635                                           vmcap->captype,
 636                                           vmcap->capval);
 637                 break;
 638         case VM_SET_X2APIC_STATE:
 639                 x2apic = (struct vm_x2apic *)data;
 640                 error = vm_set_x2apic_state(sc->vm,
 641                                             x2apic->cpuid, x2apic->state);
 642                 break;
 643         case VM_GET_X2APIC_STATE:
 644                 x2apic = (struct vm_x2apic *)data;
 645                 error = vm_get_x2apic_state(sc->vm,
 646                                             x2apic->cpuid, &x2apic->state);
 647                 break;
 648         case VM_GET_GPA_PMAP:
 649                 gpapte = (struct vm_gpa_pte *)data;
 650                 pmap_get_mapping(vmspace_pmap(vm_get_vmspace(sc->vm)),
 651                                  gpapte->gpa, gpapte->pte, &gpapte->ptenum);
 652                 error = 0;
 653                 break;
 654         case VM_GET_HPET_CAPABILITIES:
 655                 error = vhpet_getcap((struct vm_hpet_cap *)data);
 656                 break;
 657         case VM_GLA2GPA: {
 658                 CTASSERT(PROT_READ == VM_PROT_READ);
 659                 CTASSERT(PROT_WRITE == VM_PROT_WRITE);
 660                 CTASSERT(PROT_EXEC == VM_PROT_EXECUTE);
 661                 gg = (struct vm_gla2gpa *)data;
 662                 error = vm_gla2gpa(sc->vm, gg->vcpuid, &gg->paging, gg->gla,
 663                     gg->prot, &gg->gpa, &gg->fault);
 664                 KASSERT(error == 0 || error == EFAULT,
 665                     ("%s: vm_gla2gpa unknown error %d", __func__, error));
 666                 break;
 667         }
 668         case VM_ACTIVATE_CPU:
 669                 vac = (struct vm_activate_cpu *)data;
 670                 error = vm_activate_cpu(sc->vm, vac->vcpuid);
 671                 break;
 672         case VM_GET_CPUS:
 673                 error = 0;
 674                 vm_cpuset = (struct vm_cpuset *)data;
 675                 size = vm_cpuset->cpusetsize;
 676                 if (size < sizeof(cpuset_t) || size > CPU_MAXSIZE / NBBY) {
 677                         error = ERANGE;
 678                         break;
 679                 }
 680                 cpuset = malloc(size, M_TEMP, M_WAITOK | M_ZERO);
 681                 if (vm_cpuset->which == VM_ACTIVE_CPUS)
 682                         *cpuset = vm_active_cpus(sc->vm);
 683                 else if (vm_cpuset->which == VM_SUSPENDED_CPUS)
 684                         *cpuset = vm_suspended_cpus(sc->vm);
 685                 else
 686                         error = EINVAL;
 687                 if (error == 0)
 688                         error = copyout(cpuset, vm_cpuset->cpus, size);
 689                 free(cpuset, M_TEMP);
 690                 break;
 691         case VM_SET_INTINFO:
 692                 vmii = (struct vm_intinfo *)data;
 693                 error = vm_exit_intinfo(sc->vm, vmii->vcpuid, vmii->info1);
 694                 break;
 695         case VM_GET_INTINFO:
 696                 vmii = (struct vm_intinfo *)data;
 697                 error = vm_get_intinfo(sc->vm, vmii->vcpuid, &vmii->info1,
 698                     &vmii->info2);
 699                 break;
 700         case VM_RTC_WRITE:
 701                 rtcdata = (struct vm_rtc_data *)data;
 702                 error = vrtc_nvram_write(sc->vm, rtcdata->offset,
 703                     rtcdata->value);
 704                 break;
 705         case VM_RTC_READ:
 706                 rtcdata = (struct vm_rtc_data *)data;
 707                 error = vrtc_nvram_read(sc->vm, rtcdata->offset,
 708                     &rtcdata->value);
 709                 break;
 710         case VM_RTC_SETTIME:
 711                 rtctime = (struct vm_rtc_time *)data;
 712                 error = vrtc_set_time(sc->vm, rtctime->secs);
 713                 break;
 714         case VM_RTC_GETTIME:
 715                 error = 0;
 716                 rtctime = (struct vm_rtc_time *)data;
 717                 rtctime->secs = vrtc_get_time(sc->vm);
 718                 break;
 719         case VM_RESTART_INSTRUCTION:
 720                 error = vm_restart_instruction(sc->vm, vcpu);
 721                 break;
 722         default:
 723                 error = ENOTTY;
 724                 break;
 725         }
 726
 727         if (state_changed == 1)
 728                 vcpu_unlock_one(sc, vcpu);
 729         else if (state_changed == 2)
 730                 vcpu_unlock_all(sc);
 731
 732 done:
 733         /* Make sure that no handler returns a bogus value like ERESTART */
 734         KASSERT(error >= 0, ("vmmdev_ioctl: invalid error return %d", error));
 735         return (error);
 736 }
 737
 738 static int
 739 vmmdev_mmap_single(struct cdev *cdev, vm_ooffset_t *offset, vm_size_t mapsize,
 740     struct vm_object **objp, int nprot)
 741 {
 742         struct vmmdev_softc *sc;
 743         vm_paddr_t gpa;
 744         size_t len;
 745         vm_ooffset_t segoff, first, last;
 746         int error, found, segid;
 747         bool sysmem;
 748
 749         first = *offset;
 750         last = first + mapsize;
 751         if ((nprot & PROT_EXEC) || first < 0 || first >= last)
 752                 return (EINVAL);
 753
 754         sc = vmmdev_lookup2(cdev);
 755         if (sc == NULL) {
 756                 /* virtual machine is in the process of being created */
 757                 return (EINVAL);
 758         }
 759
 760         /*
 761          * Get a read lock on the guest memory map by freezing any vcpu.
 762          */
 763         error = vcpu_lock_one(sc, VM_MAXCPU - 1);
 764         if (error)
 765                 return (error);
 766
 767         gpa = 0;
 768         found = 0;
 769         while (!found) {
 770                 error = vm_mmap_getnext(sc->vm, &gpa, &segid, &segoff, &len,
 771                     NULL, NULL);
 772                 if (error)
 773                         break;
 774
 775                 if (first >= gpa && last <= gpa + len)
 776                         found = 1;
 777                 else
 778                         gpa += len;
 779         }
 780
 781         if (found) {
 782                 error = vm_get_memseg(sc->vm, segid, &len, &sysmem, objp);
 783                 KASSERT(error == 0 && *objp != NULL,
 784                     ("%s: invalid memory segment %d", __func__, segid));
 785                 if (sysmem) {
 786                         vm_object_reference(*objp);
 787                         *offset = segoff + (first - gpa);
 788                 } else {
 789                         error = EINVAL;
 790                 }
 791         }
 792         vcpu_unlock_one(sc, VM_MAXCPU - 1);
 793         return (error);
 794 }
 795
 796 static void
 797 vmmdev_destroy(void *arg)
 798 {
 799         struct vmmdev_softc *sc = arg;
 800         struct devmem_softc *dsc;
 801         int error;
 802
 803         error = vcpu_lock_all(sc);
 804         KASSERT(error == 0, ("%s: error %d freezing vcpus", __func__, error));
 805
 806         while ((dsc = SLIST_FIRST(&sc->devmem)) != NULL) {
 807                 KASSERT(dsc->cdev == NULL, ("%s: devmem not free", __func__));
 808                 SLIST_REMOVE_HEAD(&sc->devmem, link);
 809                 free(dsc->name, M_VMMDEV);
 810                 free(dsc, M_VMMDEV);
 811         }
 812
 813         if (sc->cdev != NULL)
 814                 destroy_dev(sc->cdev);
 815
 816         if (sc->vm != NULL)
 817                 vm_destroy(sc->vm);
 818
 819         if ((sc->flags & VSC_LINKED) != 0) {
 820                 mtx_lock(&vmmdev_mtx);
 821                 SLIST_REMOVE(&head, sc, vmmdev_softc, link);
 822                 mtx_unlock(&vmmdev_mtx);
 823         }
 824
 825         free(sc, M_VMMDEV);
 826 }
 827
 828 static int
 829 sysctl_vmm_destroy(SYSCTL_HANDLER_ARGS)
 830 {
 831         int error;
 832         char buf[VM_MAX_NAMELEN];
 833         struct devmem_softc *dsc;
 834         struct vmmdev_softc *sc;
 835         struct cdev *cdev;
 836
 837         strlcpy(buf, "beavis", sizeof(buf));
 838         error = sysctl_handle_string(oidp, buf, sizeof(buf), req);
 839         if (error != 0 || req->newptr == NULL)
 840                 return (error);
 841
 842         mtx_lock(&vmmdev_mtx);
 843         sc = vmmdev_lookup(buf);
 844         if (sc == NULL || sc->cdev == NULL) {
 845                 mtx_unlock(&vmmdev_mtx);
 846                 return (EINVAL);
 847         }
 848
 849         /*
 850          * The 'cdev' will be destroyed asynchronously when 'si_threadcount'
 851          * goes down to 0 so we should not do it again in the callback.
 852          *
 853          * Setting 'sc->cdev' to NULL is also used to indicate that the VM
 854          * is scheduled for destruction.
 855          */
 856         cdev = sc->cdev;
 857         sc->cdev = NULL;
 858         mtx_unlock(&vmmdev_mtx);
 859
 860         /*
 861          * Schedule all cdevs to be destroyed:
 862          *
 863          * - any new operations on the 'cdev' will return an error (ENXIO).
 864          *
 865          * - when the 'si_threadcount' dwindles down to zero the 'cdev' will
 866          *   be destroyed and the callback will be invoked in a taskqueue
 867          *   context.
 868          *
 869          * - the 'devmem' cdevs are destroyed before the virtual machine 'cdev'
 870          */
 871         SLIST_FOREACH(dsc, &sc->devmem, link) {
 872                 KASSERT(dsc->cdev != NULL, ("devmem cdev already destroyed"));
 873                 destroy_dev_sched_cb(dsc->cdev, devmem_destroy, dsc);
 874         }
 875         destroy_dev_sched_cb(cdev, vmmdev_destroy, sc);
 876         return (0);
 877 }
 878 SYSCTL_PROC(_hw_vmm, OID_AUTO, destroy, CTLTYPE_STRING | CTLFLAG_RW,
 879             NULL, 0, sysctl_vmm_destroy, "A", NULL);
 880
 881 static struct cdevsw vmmdevsw = {
 882         .d_name         = "vmmdev",
 883         .d_version      = D_VERSION,
 884         .d_ioctl        = vmmdev_ioctl,
 885         .d_mmap_single  = vmmdev_mmap_single,
 886         .d_read         = vmmdev_rw,
 887         .d_write        = vmmdev_rw,
 888 };
 889
 890 static int
 891 sysctl_vmm_create(SYSCTL_HANDLER_ARGS)
 892 {
 893         int error;
 894         struct vm *vm;
 895         struct cdev *cdev;
 896         struct vmmdev_softc *sc, *sc2;
 897         char buf[VM_MAX_NAMELEN];
 898
 899         strlcpy(buf, "beavis", sizeof(buf));
 900         error = sysctl_handle_string(oidp, buf, sizeof(buf), req);
 901         if (error != 0 || req->newptr == NULL)
 902                 return (error);
 903
 904         mtx_lock(&vmmdev_mtx);
 905         sc = vmmdev_lookup(buf);
 906         mtx_unlock(&vmmdev_mtx);
 907         if (sc != NULL)
 908                 return (EEXIST);
 909
 910         error = vm_create(buf, &vm);
 911         if (error != 0)
 912                 return (error);
 913
 914         sc = malloc(sizeof(struct vmmdev_softc), M_VMMDEV, M_WAITOK | M_ZERO);
 915         sc->vm = vm;
 916         SLIST_INIT(&sc->devmem);
 917
 918         /*
 919          * Lookup the name again just in case somebody sneaked in when we
 920          * dropped the lock.
 921          */
 922         mtx_lock(&vmmdev_mtx);
 923         sc2 = vmmdev_lookup(buf);
 924         if (sc2 == NULL) {
 925                 SLIST_INSERT_HEAD(&head, sc, link);
 926                 sc->flags |= VSC_LINKED;
 927         }
 928         mtx_unlock(&vmmdev_mtx);
 929
 930         if (sc2 != NULL) {
 931                 vmmdev_destroy(sc);
 932                 return (EEXIST);
 933         }
 934
 935         error = make_dev_p(MAKEDEV_CHECKNAME, &cdev, &vmmdevsw, NULL,
 936                            UID_ROOT, GID_WHEEL, 0600, "vmm/%s", buf);
 937         if (error != 0) {
 938                 vmmdev_destroy(sc);
 939                 return (error);
 940         }
 941
 942         mtx_lock(&vmmdev_mtx);
 943         sc->cdev = cdev;
 944         sc->cdev->si_drv1 = sc;
 945         mtx_unlock(&vmmdev_mtx);
 946
 947         return (0);
 948 }
 949 SYSCTL_PROC(_hw_vmm, OID_AUTO, create, CTLTYPE_STRING | CTLFLAG_RW,
 950             NULL, 0, sysctl_vmm_create, "A", NULL);
 951
 952 void
 953 vmmdev_init(void)
 954 {
 955         mtx_init(&vmmdev_mtx, "vmm device mutex", NULL, MTX_DEF);
 956 }
 957
 958 int
 959 vmmdev_cleanup(void)
 960 {
 961         int error;
 962
 963         if (SLIST_EMPTY(&head))
 964                 error = 0;
 965         else
 966                 error = EBUSY;
 967
 968         return (error);
 969 }
 970
 971 static int
 972 devmem_mmap_single(struct cdev *cdev, vm_ooffset_t *offset, vm_size_t len,
 973     struct vm_object **objp, int nprot)
 974 {
 975         struct devmem_softc *dsc;
 976         vm_ooffset_t first, last;
 977         size_t seglen;
 978         int error;
 979         bool sysmem;
 980
 981         dsc = cdev->si_drv1;
 982         if (dsc == NULL) {
 983                 /* 'cdev' has been created but is not ready for use */
 984                 return (ENXIO);
 985         }
 986
 987         first = *offset;
 988         last = *offset + len;
 989         if ((nprot & PROT_EXEC) || first < 0 || first >= last)
 990                 return (EINVAL);
 991
 992         error = vcpu_lock_one(dsc->sc, VM_MAXCPU - 1);
 993         if (error)
 994                 return (error);
 995
 996         error = vm_get_memseg(dsc->sc->vm, dsc->segid, &seglen, &sysmem, objp);
 997         KASSERT(error == 0 && !sysmem && *objp != NULL,
 998             ("%s: invalid devmem segment %d", __func__, dsc->segid));
 999
1000         vcpu_unlock_one(dsc->sc, VM_MAXCPU - 1);
1001
1002         if (seglen >= last) {
1003                 vm_object_reference(*objp);
1004                 return (0);
1005         } else {
1006                 return (EINVAL);
1007         }
1008 }
1009
1010 static struct cdevsw devmemsw = {
1011         .d_name         = "devmem",
1012         .d_version      = D_VERSION,
1013         .d_mmap_single  = devmem_mmap_single,
1014 };
1015
1016 static int
1017 devmem_create_cdev(const char *vmname, int segid, char *devname)
1018 {
1019         struct devmem_softc *dsc;
1020         struct vmmdev_softc *sc;
1021         struct cdev *cdev;
1022         int error;
1023
1024         error = make_dev_p(MAKEDEV_CHECKNAME, &cdev, &devmemsw, NULL,
1025             UID_ROOT, GID_WHEEL, 0600, "vmm.io/%s.%s", vmname, devname);
1026         if (error)
1027                 return (error);
1028
1029         dsc = malloc(sizeof(struct devmem_softc), M_VMMDEV, M_WAITOK | M_ZERO);
1030
1031         mtx_lock(&vmmdev_mtx);
1032         sc = vmmdev_lookup(vmname);
1033         KASSERT(sc != NULL, ("%s: vm %s softc not found", __func__, vmname));
1034         if (sc->cdev == NULL) {
1035                 /* virtual machine is being created or destroyed */
1036                 mtx_unlock(&vmmdev_mtx);
1037                 free(dsc, M_VMMDEV);
1038                 destroy_dev_sched_cb(cdev, NULL, 0);
1039                 return (ENODEV);
1040         }
1041
1042         dsc->segid = segid;
1043         dsc->name = devname;
1044         dsc->cdev = cdev;
1045         dsc->sc = sc;
1046         SLIST_INSERT_HEAD(&sc->devmem, dsc, link);
1047         mtx_unlock(&vmmdev_mtx);
1048
1049         /* The 'cdev' is ready for use after 'si_drv1' is initialized */
1050         cdev->si_drv1 = dsc;
1051         return (0);
1052 }
1053
1054 static void
1055 devmem_destroy(void *arg)
1056 {
1057         struct devmem_softc *dsc = arg;
1058
1059         KASSERT(dsc->cdev, ("%s: devmem cdev already destroyed", __func__));
1060         dsc->cdev = NULL;
1061         dsc->sc = NULL;
1062 }