sys/amd64/vmm/vmm_dev.c

   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/kernel.h>
  34 #include <sys/queue.h>
  35 #include <sys/lock.h>
  36 #include <sys/mutex.h>
  37 #include <sys/malloc.h>
  38 #include <sys/conf.h>
  39 #include <sys/sysctl.h>
  40 #include <sys/libkern.h>
  41 #include <sys/ioccom.h>
  42 #include <sys/mman.h>
  43 #include <sys/uio.h>
  44
  45 #include <vm/vm.h>
  46 #include <vm/pmap.h>
  47 #include <vm/vm_map.h>
  48
  49 #include <machine/vmparam.h>
  50 #include <machine/vmm.h>
  51 #include <machine/vmm_instruction_emul.h>
  52 #include <machine/vmm_dev.h>
  53
  54 #include "vmm_lapic.h"
  55 #include "vmm_stat.h"
  56 #include "vmm_mem.h"
  57 #include "io/ppt.h"
  58 #include "io/vatpic.h"
  59 #include "io/vioapic.h"
  60 #include "io/vhpet.h"
  61
  62 struct vmmdev_softc {
  63         struct vm       *vm;            /* vm instance cookie */
  64         struct cdev     *cdev;
  65         SLIST_ENTRY(vmmdev_softc) link;
  66         int             flags;
  67 };
  68 #define VSC_LINKED              0x01
  69
  70 static SLIST_HEAD(, vmmdev_softc) head;
  71
  72 static struct mtx vmmdev_mtx;
  73
  74 static MALLOC_DEFINE(M_VMMDEV, "vmmdev", "vmmdev");
  75
  76 SYSCTL_DECL(_hw_vmm);
  77
  78 static struct vmmdev_softc *
  79 vmmdev_lookup(const char *name)
  80 {
  81         struct vmmdev_softc *sc;
  82
  83 #ifdef notyet   /* XXX kernel is not compiled with invariants */
  84         mtx_assert(&vmmdev_mtx, MA_OWNED);
  85 #endif
  86
  87         SLIST_FOREACH(sc, &head, link) {
  88                 if (strcmp(name, vm_name(sc->vm)) == 0)
  89                         break;
  90         }
  91
  92         return (sc);
  93 }
  94
  95 static struct vmmdev_softc *
  96 vmmdev_lookup2(struct cdev *cdev)
  97 {
  98
  99         return (cdev->si_drv1);
 100 }
 101
 102 static int
 103 vmmdev_rw(struct cdev *cdev, struct uio *uio, int flags)
 104 {
 105         int error, off, c, prot;
 106         vm_paddr_t gpa;
 107         void *hpa, *cookie;
 108         struct vmmdev_softc *sc;
 109
 110         static char zerobuf[PAGE_SIZE];
 111
 112         error = 0;
 113         sc = vmmdev_lookup2(cdev);
 114         if (sc == NULL)
 115                 error = ENXIO;
 116
 117         prot = (uio->uio_rw == UIO_WRITE ? VM_PROT_WRITE : VM_PROT_READ);
 118         while (uio->uio_resid > 0 && error == 0) {
 119                 gpa = uio->uio_offset;
 120                 off = gpa & PAGE_MASK;
 121                 c = min(uio->uio_resid, PAGE_SIZE - off);
 122
 123                 /*
 124                  * The VM has a hole in its physical memory map. If we want to
 125                  * use 'dd' to inspect memory beyond the hole we need to
 126                  * provide bogus data for memory that lies in the hole.
 127                  *
 128                  * Since this device does not support lseek(2), dd(1) will
 129                  * read(2) blocks of data to simulate the lseek(2).
 130                  */
 131                 hpa = vm_gpa_hold(sc->vm, gpa, c, prot, &cookie);
 132                 if (hpa == NULL) {
 133                         if (uio->uio_rw == UIO_READ)
 134                                 error = uiomove(zerobuf, c, uio);
 135                         else
 136                                 error = EFAULT;
 137                 } else {
 138                         error = uiomove(hpa, c, uio);
 139                         vm_gpa_release(cookie);
 140                 }
 141         }
 142         return (error);
 143 }
 144
 145 static int
 146 vmmdev_ioctl(struct cdev *cdev, u_long cmd, caddr_t data, int fflag,
 147              struct thread *td)
 148 {
 149         int error, vcpu, state_changed, size;
 150         cpuset_t *cpuset;
 151         struct vmmdev_softc *sc;
 152         struct vm_memory_segment *seg;
 153         struct vm_register *vmreg;
 154         struct vm_seg_desc *vmsegdesc;
 155         struct vm_run *vmrun;
 156         struct vm_exception *vmexc;
 157         struct vm_lapic_irq *vmirq;
 158         struct vm_lapic_msi *vmmsi;
 159         struct vm_ioapic_irq *ioapic_irq;
 160         struct vm_isa_irq *isa_irq;
 161         struct vm_isa_irq_trigger *isa_irq_trigger;
 162         struct vm_capability *vmcap;
 163         struct vm_pptdev *pptdev;
 164         struct vm_pptdev_mmio *pptmmio;
 165         struct vm_pptdev_msi *pptmsi;
 166         struct vm_pptdev_msix *pptmsix;
 167         struct vm_nmi *vmnmi;
 168         struct vm_stats *vmstats;
 169         struct vm_stat_desc *statdesc;
 170         struct vm_x2apic *x2apic;
 171         struct vm_gpa_pte *gpapte;
 172         struct vm_suspend *vmsuspend;
 173         struct vm_gla2gpa *gg;
 174         struct vm_activate_cpu *vac;
 175         struct vm_cpuset *vm_cpuset;
 176         struct vm_intinfo *vmii;
 177
 178         sc = vmmdev_lookup2(cdev);
 179         if (sc == NULL)
 180                 return (ENXIO);
 181
 182         error = 0;
 183         vcpu = -1;
 184         state_changed = 0;
 185
 186         /*
 187          * Some VMM ioctls can operate only on vcpus that are not running.
 188          */
 189         switch (cmd) {
 190         case VM_RUN:
 191         case VM_GET_REGISTER:
 192         case VM_SET_REGISTER:
 193         case VM_GET_SEGMENT_DESCRIPTOR:
 194         case VM_SET_SEGMENT_DESCRIPTOR:
 195         case VM_INJECT_EXCEPTION:
 196         case VM_GET_CAPABILITY:
 197         case VM_SET_CAPABILITY:
 198         case VM_PPTDEV_MSI:
 199         case VM_PPTDEV_MSIX:
 200         case VM_SET_X2APIC_STATE:
 201         case VM_GLA2GPA:
 202         case VM_ACTIVATE_CPU:
 203         case VM_SET_INTINFO:
 204         case VM_GET_INTINFO:
 205                 /*
 206                  * XXX fragile, handle with care
 207                  * Assumes that the first field of the ioctl data is the vcpu.
 208                  */
 209                 vcpu = *(int *)data;
 210                 if (vcpu < 0 || vcpu >= VM_MAXCPU) {
 211                         error = EINVAL;
 212                         goto done;
 213                 }
 214
 215                 error = vcpu_set_state(sc->vm, vcpu, VCPU_FROZEN, true);
 216                 if (error)
 217                         goto done;
 218
 219                 state_changed = 1;
 220                 break;
 221
 222         case VM_MAP_PPTDEV_MMIO:
 223         case VM_BIND_PPTDEV:
 224         case VM_UNBIND_PPTDEV:
 225         case VM_MAP_MEMORY:
 226         case VM_REINIT:
 227                 /*
 228                  * ioctls that operate on the entire virtual machine must
 229                  * prevent all vcpus from running.
 230                  */
 231                 error = 0;
 232                 for (vcpu = 0; vcpu < VM_MAXCPU; vcpu++) {
 233                         error = vcpu_set_state(sc->vm, vcpu, VCPU_FROZEN, true);
 234                         if (error)
 235                                 break;
 236                 }
 237
 238                 if (error) {
 239                         while (--vcpu >= 0)
 240                                 vcpu_set_state(sc->vm, vcpu, VCPU_IDLE, false);
 241                         goto done;
 242                 }
 243
 244                 state_changed = 2;
 245                 break;
 246
 247         default:
 248                 break;
 249         }
 250
 251         switch(cmd) {
 252         case VM_RUN:
 253                 vmrun = (struct vm_run *)data;
 254                 error = vm_run(sc->vm, vmrun);
 255                 break;
 256         case VM_SUSPEND:
 257                 vmsuspend = (struct vm_suspend *)data;
 258                 error = vm_suspend(sc->vm, vmsuspend->how);
 259                 break;
 260         case VM_REINIT:
 261                 error = vm_reinit(sc->vm);
 262                 break;
 263         case VM_STAT_DESC: {
 264                 statdesc = (struct vm_stat_desc *)data;
 265                 error = vmm_stat_desc_copy(statdesc->index,
 266                                         statdesc->desc, sizeof(statdesc->desc));
 267                 break;
 268         }
 269         case VM_STATS: {
 270                 CTASSERT(MAX_VM_STATS >= MAX_VMM_STAT_ELEMS);
 271                 vmstats = (struct vm_stats *)data;
 272                 getmicrotime(&vmstats->tv);
 273                 error = vmm_stat_copy(sc->vm, vmstats->cpuid,
 274                                       &vmstats->num_entries, vmstats->statbuf);
 275                 break;
 276         }
 277         case VM_PPTDEV_MSI:
 278                 pptmsi = (struct vm_pptdev_msi *)data;
 279                 error = ppt_setup_msi(sc->vm, pptmsi->vcpu,
 280                                       pptmsi->bus, pptmsi->slot, pptmsi->func,
 281                                       pptmsi->addr, pptmsi->msg,
 282                                       pptmsi->numvec);
 283                 break;
 284         case VM_PPTDEV_MSIX:
 285                 pptmsix = (struct vm_pptdev_msix *)data;
 286                 error = ppt_setup_msix(sc->vm, pptmsix->vcpu,
 287                                        pptmsix->bus, pptmsix->slot,
 288                                        pptmsix->func, pptmsix->idx,
 289                                        pptmsix->addr, pptmsix->msg,
 290                                        pptmsix->vector_control);
 291                 break;
 292         case VM_MAP_PPTDEV_MMIO:
 293                 pptmmio = (struct vm_pptdev_mmio *)data;
 294                 error = ppt_map_mmio(sc->vm, pptmmio->bus, pptmmio->slot,
 295                                      pptmmio->func, pptmmio->gpa, pptmmio->len,
 296                                      pptmmio->hpa);
 297                 break;
 298         case VM_BIND_PPTDEV:
 299                 pptdev = (struct vm_pptdev *)data;
 300                 error = vm_assign_pptdev(sc->vm, pptdev->bus, pptdev->slot,
 301                                          pptdev->func);
 302                 break;
 303         case VM_UNBIND_PPTDEV:
 304                 pptdev = (struct vm_pptdev *)data;
 305                 error = vm_unassign_pptdev(sc->vm, pptdev->bus, pptdev->slot,
 306                                            pptdev->func);
 307                 break;
 308         case VM_INJECT_EXCEPTION:
 309                 vmexc = (struct vm_exception *)data;
 310                 error = vm_inject_exception(sc->vm, vmexc->cpuid, vmexc);
 311                 break;
 312         case VM_INJECT_NMI:
 313                 vmnmi = (struct vm_nmi *)data;
 314                 error = vm_inject_nmi(sc->vm, vmnmi->cpuid);
 315                 break;
 316         case VM_LAPIC_IRQ:
 317                 vmirq = (struct vm_lapic_irq *)data;
 318                 error = lapic_intr_edge(sc->vm, vmirq->cpuid, vmirq->vector);
 319                 break;
 320         case VM_LAPIC_LOCAL_IRQ:
 321                 vmirq = (struct vm_lapic_irq *)data;
 322                 error = lapic_set_local_intr(sc->vm, vmirq->cpuid,
 323                     vmirq->vector);
 324                 break;
 325         case VM_LAPIC_MSI:
 326                 vmmsi = (struct vm_lapic_msi *)data;
 327                 error = lapic_intr_msi(sc->vm, vmmsi->addr, vmmsi->msg);
 328                 break;
 329         case VM_IOAPIC_ASSERT_IRQ:
 330                 ioapic_irq = (struct vm_ioapic_irq *)data;
 331                 error = vioapic_assert_irq(sc->vm, ioapic_irq->irq);
 332                 break;
 333         case VM_IOAPIC_DEASSERT_IRQ:
 334                 ioapic_irq = (struct vm_ioapic_irq *)data;
 335                 error = vioapic_deassert_irq(sc->vm, ioapic_irq->irq);
 336                 break;
 337         case VM_IOAPIC_PULSE_IRQ:
 338                 ioapic_irq = (struct vm_ioapic_irq *)data;
 339                 error = vioapic_pulse_irq(sc->vm, ioapic_irq->irq);
 340                 break;
 341         case VM_IOAPIC_PINCOUNT:
 342                 *(int *)data = vioapic_pincount(sc->vm);
 343                 break;
 344         case VM_ISA_ASSERT_IRQ:
 345                 isa_irq = (struct vm_isa_irq *)data;
 346                 error = vatpic_assert_irq(sc->vm, isa_irq->atpic_irq);
 347                 if (error == 0 && isa_irq->ioapic_irq != -1)
 348                         error = vioapic_assert_irq(sc->vm,
 349                             isa_irq->ioapic_irq);
 350                 break;
 351         case VM_ISA_DEASSERT_IRQ:
 352                 isa_irq = (struct vm_isa_irq *)data;
 353                 error = vatpic_deassert_irq(sc->vm, isa_irq->atpic_irq);
 354                 if (error == 0 && isa_irq->ioapic_irq != -1)
 355                         error = vioapic_deassert_irq(sc->vm,
 356                             isa_irq->ioapic_irq);
 357                 break;
 358         case VM_ISA_PULSE_IRQ:
 359                 isa_irq = (struct vm_isa_irq *)data;
 360                 error = vatpic_pulse_irq(sc->vm, isa_irq->atpic_irq);
 361                 if (error == 0 && isa_irq->ioapic_irq != -1)
 362                         error = vioapic_pulse_irq(sc->vm, isa_irq->ioapic_irq);
 363                 break;
 364         case VM_ISA_SET_IRQ_TRIGGER:
 365                 isa_irq_trigger = (struct vm_isa_irq_trigger *)data;
 366                 error = vatpic_set_irq_trigger(sc->vm,
 367                     isa_irq_trigger->atpic_irq, isa_irq_trigger->trigger);
 368                 break;
 369         case VM_MAP_MEMORY:
 370                 seg = (struct vm_memory_segment *)data;
 371                 error = vm_malloc(sc->vm, seg->gpa, seg->len);
 372                 break;
 373         case VM_GET_MEMORY_SEG:
 374                 seg = (struct vm_memory_segment *)data;
 375                 seg->len = 0;
 376                 (void)vm_gpabase2memseg(sc->vm, seg->gpa, seg);
 377                 error = 0;
 378                 break;
 379         case VM_GET_REGISTER:
 380                 vmreg = (struct vm_register *)data;
 381                 error = vm_get_register(sc->vm, vmreg->cpuid, vmreg->regnum,
 382                                         &vmreg->regval);
 383                 break;
 384         case VM_SET_REGISTER:
 385                 vmreg = (struct vm_register *)data;
 386                 error = vm_set_register(sc->vm, vmreg->cpuid, vmreg->regnum,
 387                                         vmreg->regval);
 388                 break;
 389         case VM_SET_SEGMENT_DESCRIPTOR:
 390                 vmsegdesc = (struct vm_seg_desc *)data;
 391                 error = vm_set_seg_desc(sc->vm, vmsegdesc->cpuid,
 392                                         vmsegdesc->regnum,
 393                                         &vmsegdesc->desc);
 394                 break;
 395         case VM_GET_SEGMENT_DESCRIPTOR:
 396                 vmsegdesc = (struct vm_seg_desc *)data;
 397                 error = vm_get_seg_desc(sc->vm, vmsegdesc->cpuid,
 398                                         vmsegdesc->regnum,
 399                                         &vmsegdesc->desc);
 400                 break;
 401         case VM_GET_CAPABILITY:
 402                 vmcap = (struct vm_capability *)data;
 403                 error = vm_get_capability(sc->vm, vmcap->cpuid,
 404                                           vmcap->captype,
 405                                           &vmcap->capval);
 406                 break;
 407         case VM_SET_CAPABILITY:
 408                 vmcap = (struct vm_capability *)data;
 409                 error = vm_set_capability(sc->vm, vmcap->cpuid,
 410                                           vmcap->captype,
 411                                           vmcap->capval);
 412                 break;
 413         case VM_SET_X2APIC_STATE:
 414                 x2apic = (struct vm_x2apic *)data;
 415                 error = vm_set_x2apic_state(sc->vm,
 416                                             x2apic->cpuid, x2apic->state);
 417                 break;
 418         case VM_GET_X2APIC_STATE:
 419                 x2apic = (struct vm_x2apic *)data;
 420                 error = vm_get_x2apic_state(sc->vm,
 421                                             x2apic->cpuid, &x2apic->state);
 422                 break;
 423         case VM_GET_GPA_PMAP:
 424                 gpapte = (struct vm_gpa_pte *)data;
 425                 pmap_get_mapping(vmspace_pmap(vm_get_vmspace(sc->vm)),
 426                                  gpapte->gpa, gpapte->pte, &gpapte->ptenum);
 427                 error = 0;
 428                 break;
 429         case VM_GET_HPET_CAPABILITIES:
 430                 error = vhpet_getcap((struct vm_hpet_cap *)data);
 431                 break;
 432         case VM_GLA2GPA: {
 433                 CTASSERT(PROT_READ == VM_PROT_READ);
 434                 CTASSERT(PROT_WRITE == VM_PROT_WRITE);
 435                 CTASSERT(PROT_EXEC == VM_PROT_EXECUTE);
 436                 gg = (struct vm_gla2gpa *)data;
 437                 error = vmm_gla2gpa(sc->vm, gg->vcpuid, &gg->paging, gg->gla,
 438                     gg->prot, &gg->gpa);
 439                 KASSERT(error == 0 || error == 1 || error == -1,
 440                     ("%s: vmm_gla2gpa unknown error %d", __func__, error));
 441                 if (error >= 0) {
 442                         /*
 443                          * error = 0: the translation was successful
 444                          * error = 1: a fault was injected into the guest
 445                          */
 446                         gg->fault = error;
 447                         error = 0;
 448                 } else {
 449                         error = EFAULT;
 450                 }
 451                 break;
 452         }
 453         case VM_ACTIVATE_CPU:
 454                 vac = (struct vm_activate_cpu *)data;
 455                 error = vm_activate_cpu(sc->vm, vac->vcpuid);
 456                 break;
 457         case VM_GET_CPUS:
 458                 error = 0;
 459                 vm_cpuset = (struct vm_cpuset *)data;
 460                 size = vm_cpuset->cpusetsize;
 461                 if (size < sizeof(cpuset_t) || size > CPU_MAXSIZE / NBBY) {
 462                         error = ERANGE;
 463                         break;
 464                 }
 465                 cpuset = malloc(size, M_TEMP, M_WAITOK | M_ZERO);
 466                 if (vm_cpuset->which == VM_ACTIVE_CPUS)
 467                         *cpuset = vm_active_cpus(sc->vm);
 468                 else if (vm_cpuset->which == VM_SUSPENDED_CPUS)
 469                         *cpuset = vm_suspended_cpus(sc->vm);
 470                 else
 471                         error = EINVAL;
 472                 if (error == 0)
 473                         error = copyout(cpuset, vm_cpuset->cpus, size);
 474                 free(cpuset, M_TEMP);
 475                 break;
 476         case VM_SET_INTINFO:
 477                 vmii = (struct vm_intinfo *)data;
 478                 error = vm_exit_intinfo(sc->vm, vmii->vcpuid, vmii->info1);
 479                 break;
 480         case VM_GET_INTINFO:
 481                 vmii = (struct vm_intinfo *)data;
 482                 error = vm_get_intinfo(sc->vm, vmii->vcpuid, &vmii->info1,
 483                     &vmii->info2);
 484                 break;
 485         default:
 486                 error = ENOTTY;
 487                 break;
 488         }
 489
 490         if (state_changed == 1) {
 491                 vcpu_set_state(sc->vm, vcpu, VCPU_IDLE, false);
 492         } else if (state_changed == 2) {
 493                 for (vcpu = 0; vcpu < VM_MAXCPU; vcpu++)
 494                         vcpu_set_state(sc->vm, vcpu, VCPU_IDLE, false);
 495         }
 496
 497 done:
 498         /* Make sure that no handler returns a bogus value like ERESTART */
 499         KASSERT(error >= 0, ("vmmdev_ioctl: invalid error return %d", error));
 500         return (error);
 501 }
 502
 503 static int
 504 vmmdev_mmap_single(struct cdev *cdev, vm_ooffset_t *offset,
 505                    vm_size_t size, struct vm_object **object, int nprot)
 506 {
 507         int error;
 508         struct vmmdev_softc *sc;
 509
 510         sc = vmmdev_lookup2(cdev);
 511         if (sc != NULL && (nprot & PROT_EXEC) == 0)
 512                 error = vm_get_memobj(sc->vm, *offset, size, offset, object);
 513         else
 514                 error = EINVAL;
 515
 516         return (error);
 517 }
 518
 519 static void
 520 vmmdev_destroy(void *arg)
 521 {
 522
 523         struct vmmdev_softc *sc = arg;
 524
 525         if (sc->cdev != NULL)
 526                 destroy_dev(sc->cdev);
 527
 528         if (sc->vm != NULL)
 529                 vm_destroy(sc->vm);
 530
 531         if ((sc->flags & VSC_LINKED) != 0) {
 532                 mtx_lock(&vmmdev_mtx);
 533                 SLIST_REMOVE(&head, sc, vmmdev_softc, link);
 534                 mtx_unlock(&vmmdev_mtx);
 535         }
 536
 537         free(sc, M_VMMDEV);
 538 }
 539
 540 static int
 541 sysctl_vmm_destroy(SYSCTL_HANDLER_ARGS)
 542 {
 543         int error;
 544         char buf[VM_MAX_NAMELEN];
 545         struct vmmdev_softc *sc;
 546         struct cdev *cdev;
 547
 548         strlcpy(buf, "beavis", sizeof(buf));
 549         error = sysctl_handle_string(oidp, buf, sizeof(buf), req);
 550         if (error != 0 || req->newptr == NULL)
 551                 return (error);
 552
 553         mtx_lock(&vmmdev_mtx);
 554         sc = vmmdev_lookup(buf);
 555         if (sc == NULL || sc->cdev == NULL) {
 556                 mtx_unlock(&vmmdev_mtx);
 557                 return (EINVAL);
 558         }
 559
 560         /*
 561          * The 'cdev' will be destroyed asynchronously when 'si_threadcount'
 562          * goes down to 0 so we should not do it again in the callback.
 563          */
 564         cdev = sc->cdev;
 565         sc->cdev = NULL;
 566         mtx_unlock(&vmmdev_mtx);
 567
 568         /*
 569          * Schedule the 'cdev' to be destroyed:
 570          *
 571          * - any new operations on this 'cdev' will return an error (ENXIO).
 572          *
 573          * - when the 'si_threadcount' dwindles down to zero the 'cdev' will
 574          *   be destroyed and the callback will be invoked in a taskqueue
 575          *   context.
 576          */
 577         destroy_dev_sched_cb(cdev, vmmdev_destroy, sc);
 578
 579         return (0);
 580 }
 581 SYSCTL_PROC(_hw_vmm, OID_AUTO, destroy, CTLTYPE_STRING | CTLFLAG_RW,
 582             NULL, 0, sysctl_vmm_destroy, "A", NULL);
 583
 584 static struct cdevsw vmmdevsw = {
 585         .d_name         = "vmmdev",
 586         .d_version      = D_VERSION,
 587         .d_ioctl        = vmmdev_ioctl,
 588         .d_mmap_single  = vmmdev_mmap_single,
 589         .d_read         = vmmdev_rw,
 590         .d_write        = vmmdev_rw,
 591 };
 592
 593 static int
 594 sysctl_vmm_create(SYSCTL_HANDLER_ARGS)
 595 {
 596         int error;
 597         struct vm *vm;
 598         struct cdev *cdev;
 599         struct vmmdev_softc *sc, *sc2;
 600         char buf[VM_MAX_NAMELEN];
 601
 602         strlcpy(buf, "beavis", sizeof(buf));
 603         error = sysctl_handle_string(oidp, buf, sizeof(buf), req);
 604         if (error != 0 || req->newptr == NULL)
 605                 return (error);
 606
 607         mtx_lock(&vmmdev_mtx);
 608         sc = vmmdev_lookup(buf);
 609         mtx_unlock(&vmmdev_mtx);
 610         if (sc != NULL)
 611                 return (EEXIST);
 612
 613         error = vm_create(buf, &vm);
 614         if (error != 0)
 615                 return (error);
 616
 617         sc = malloc(sizeof(struct vmmdev_softc), M_VMMDEV, M_WAITOK | M_ZERO);
 618         sc->vm = vm;
 619
 620         /*
 621          * Lookup the name again just in case somebody sneaked in when we
 622          * dropped the lock.
 623          */
 624         mtx_lock(&vmmdev_mtx);
 625         sc2 = vmmdev_lookup(buf);
 626         if (sc2 == NULL) {
 627                 SLIST_INSERT_HEAD(&head, sc, link);
 628                 sc->flags |= VSC_LINKED;
 629         }
 630         mtx_unlock(&vmmdev_mtx);
 631
 632         if (sc2 != NULL) {
 633                 vmmdev_destroy(sc);
 634                 return (EEXIST);
 635         }
 636
 637         error = make_dev_p(MAKEDEV_CHECKNAME, &cdev, &vmmdevsw, NULL,
 638                            UID_ROOT, GID_WHEEL, 0600, "vmm/%s", buf);
 639         if (error != 0) {
 640                 vmmdev_destroy(sc);
 641                 return (error);
 642         }
 643
 644         mtx_lock(&vmmdev_mtx);
 645         sc->cdev = cdev;
 646         sc->cdev->si_drv1 = sc;
 647         mtx_unlock(&vmmdev_mtx);
 648
 649         return (0);
 650 }
 651 SYSCTL_PROC(_hw_vmm, OID_AUTO, create, CTLTYPE_STRING | CTLFLAG_RW,
 652             NULL, 0, sysctl_vmm_create, "A", NULL);
 653
 654 void
 655 vmmdev_init(void)
 656 {
 657         mtx_init(&vmmdev_mtx, "vmm device mutex", NULL, MTX_DEF);
 658 }
 659
 660 int
 661 vmmdev_cleanup(void)
 662 {
 663         int error;
 664
 665         if (SLIST_EMPTY(&head))
 666                 error = 0;
 667         else
 668                 error = EBUSY;
 669
 670         return (error);
 671 }