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 #include "io/vrtc.h"
  62
  63 struct vmmdev_softc {
  64         struct vm       *vm;            /* vm instance cookie */
  65         struct cdev     *cdev;
  66         SLIST_ENTRY(vmmdev_softc) link;
  67         int             flags;
  68 };
  69 #define VSC_LINKED              0x01
  70
  71 static SLIST_HEAD(, vmmdev_softc) head;
  72
  73 static struct mtx vmmdev_mtx;
  74
  75 static MALLOC_DEFINE(M_VMMDEV, "vmmdev", "vmmdev");
  76
  77 SYSCTL_DECL(_hw_vmm);
  78
  79 static struct vmmdev_softc *
  80 vmmdev_lookup(const char *name)
  81 {
  82         struct vmmdev_softc *sc;
  83
  84 #ifdef notyet   /* XXX kernel is not compiled with invariants */
  85         mtx_assert(&vmmdev_mtx, MA_OWNED);
  86 #endif
  87
  88         SLIST_FOREACH(sc, &head, link) {
  89                 if (strcmp(name, vm_name(sc->vm)) == 0)
  90                         break;
  91         }
  92
  93         return (sc);
  94 }
  95
  96 static struct vmmdev_softc *
  97 vmmdev_lookup2(struct cdev *cdev)
  98 {
  99
 100         return (cdev->si_drv1);
 101 }
 102
 103 static int
 104 vmmdev_rw(struct cdev *cdev, struct uio *uio, int flags)
 105 {
 106         int error, off, c, prot;
 107         vm_paddr_t gpa;
 108         void *hpa, *cookie;
 109         struct vmmdev_softc *sc;
 110
 111         static char zerobuf[PAGE_SIZE];
 112
 113         error = 0;
 114         sc = vmmdev_lookup2(cdev);
 115         if (sc == NULL)
 116                 error = ENXIO;
 117
 118         prot = (uio->uio_rw == UIO_WRITE ? VM_PROT_WRITE : VM_PROT_READ);
 119         while (uio->uio_resid > 0 && error == 0) {
 120                 gpa = uio->uio_offset;
 121                 off = gpa & PAGE_MASK;
 122                 c = min(uio->uio_resid, PAGE_SIZE - off);
 123
 124                 /*
 125                  * The VM has a hole in its physical memory map. If we want to
 126                  * use 'dd' to inspect memory beyond the hole we need to
 127                  * provide bogus data for memory that lies in the hole.
 128                  *
 129                  * Since this device does not support lseek(2), dd(1) will
 130                  * read(2) blocks of data to simulate the lseek(2).
 131                  */
 132                 hpa = vm_gpa_hold(sc->vm, gpa, c, prot, &cookie);
 133                 if (hpa == NULL) {
 134                         if (uio->uio_rw == UIO_READ)
 135                                 error = uiomove(zerobuf, c, uio);
 136                         else
 137                                 error = EFAULT;
 138                 } else {
 139                         error = uiomove(hpa, c, uio);
 140                         vm_gpa_release(cookie);
 141                 }
 142         }
 143         return (error);
 144 }
 145
 146 static int
 147 vmmdev_ioctl(struct cdev *cdev, u_long cmd, caddr_t data, int fflag,
 148              struct thread *td)
 149 {
 150         int error, vcpu, state_changed, size;
 151         cpuset_t *cpuset;
 152         struct vmmdev_softc *sc;
 153         struct vm_memory_segment *seg;
 154         struct vm_register *vmreg;
 155         struct vm_seg_desc *vmsegdesc;
 156         struct vm_run *vmrun;
 157         struct vm_exception *vmexc;
 158         struct vm_lapic_irq *vmirq;
 159         struct vm_lapic_msi *vmmsi;
 160         struct vm_ioapic_irq *ioapic_irq;
 161         struct vm_isa_irq *isa_irq;
 162         struct vm_isa_irq_trigger *isa_irq_trigger;
 163         struct vm_capability *vmcap;
 164         struct vm_pptdev *pptdev;
 165         struct vm_pptdev_mmio *pptmmio;
 166         struct vm_pptdev_msi *pptmsi;
 167         struct vm_pptdev_msix *pptmsix;
 168         struct vm_nmi *vmnmi;
 169         struct vm_stats *vmstats;
 170         struct vm_stat_desc *statdesc;
 171         struct vm_x2apic *x2apic;
 172         struct vm_gpa_pte *gpapte;
 173         struct vm_suspend *vmsuspend;
 174         struct vm_gla2gpa *gg;
 175         struct vm_activate_cpu *vac;
 176         struct vm_cpuset *vm_cpuset;
 177         struct vm_intinfo *vmii;
 178         struct vm_rtc_time *rtctime;
 179         struct vm_rtc_data *rtcdata;
 180
 181         sc = vmmdev_lookup2(cdev);
 182         if (sc == NULL)
 183                 return (ENXIO);
 184
 185         error = 0;
 186         vcpu = -1;
 187         state_changed = 0;
 188
 189         /*
 190          * Some VMM ioctls can operate only on vcpus that are not running.
 191          */
 192         switch (cmd) {
 193         case VM_RUN:
 194         case VM_GET_REGISTER:
 195         case VM_SET_REGISTER:
 196         case VM_GET_SEGMENT_DESCRIPTOR:
 197         case VM_SET_SEGMENT_DESCRIPTOR:
 198         case VM_INJECT_EXCEPTION:
 199         case VM_GET_CAPABILITY:
 200         case VM_SET_CAPABILITY:
 201         case VM_PPTDEV_MSI:
 202         case VM_PPTDEV_MSIX:
 203         case VM_SET_X2APIC_STATE:
 204         case VM_GLA2GPA:
 205         case VM_ACTIVATE_CPU:
 206         case VM_SET_INTINFO:
 207         case VM_GET_INTINFO:
 208         case VM_RESTART_INSTRUCTION:
 209                 /*
 210                  * XXX fragile, handle with care
 211                  * Assumes that the first field of the ioctl data is the vcpu.
 212                  */
 213                 vcpu = *(int *)data;
 214                 if (vcpu < 0 || vcpu >= VM_MAXCPU) {
 215                         error = EINVAL;
 216                         goto done;
 217                 }
 218
 219                 error = vcpu_set_state(sc->vm, vcpu, VCPU_FROZEN, true);
 220                 if (error)
 221                         goto done;
 222
 223                 state_changed = 1;
 224                 break;
 225
 226         case VM_MAP_PPTDEV_MMIO:
 227         case VM_BIND_PPTDEV:
 228         case VM_UNBIND_PPTDEV:
 229         case VM_MAP_MEMORY:
 230         case VM_REINIT:
 231                 /*
 232                  * ioctls that operate on the entire virtual machine must
 233                  * prevent all vcpus from running.
 234                  */
 235                 error = 0;
 236                 for (vcpu = 0; vcpu < VM_MAXCPU; vcpu++) {
 237                         error = vcpu_set_state(sc->vm, vcpu, VCPU_FROZEN, true);
 238                         if (error)
 239                                 break;
 240                 }
 241
 242                 if (error) {
 243                         while (--vcpu >= 0)
 244                                 vcpu_set_state(sc->vm, vcpu, VCPU_IDLE, false);
 245                         goto done;
 246                 }
 247
 248                 state_changed = 2;
 249                 break;
 250
 251         default:
 252                 break;
 253         }
 254
 255         switch(cmd) {
 256         case VM_RUN:
 257                 vmrun = (struct vm_run *)data;
 258                 error = vm_run(sc->vm, vmrun);
 259                 break;
 260         case VM_SUSPEND:
 261                 vmsuspend = (struct vm_suspend *)data;
 262                 error = vm_suspend(sc->vm, vmsuspend->how);
 263                 break;
 264         case VM_REINIT:
 265                 error = vm_reinit(sc->vm);
 266                 break;
 267         case VM_STAT_DESC: {
 268                 statdesc = (struct vm_stat_desc *)data;
 269                 error = vmm_stat_desc_copy(statdesc->index,
 270                                         statdesc->desc, sizeof(statdesc->desc));
 271                 break;
 272         }
 273         case VM_STATS: {
 274                 CTASSERT(MAX_VM_STATS >= MAX_VMM_STAT_ELEMS);
 275                 vmstats = (struct vm_stats *)data;
 276                 getmicrotime(&vmstats->tv);
 277                 error = vmm_stat_copy(sc->vm, vmstats->cpuid,
 278                                       &vmstats->num_entries, vmstats->statbuf);
 279                 break;
 280         }
 281         case VM_PPTDEV_MSI:
 282                 pptmsi = (struct vm_pptdev_msi *)data;
 283                 error = ppt_setup_msi(sc->vm, pptmsi->vcpu,
 284                                       pptmsi->bus, pptmsi->slot, pptmsi->func,
 285                                       pptmsi->addr, pptmsi->msg,
 286                                       pptmsi->numvec);
 287                 break;
 288         case VM_PPTDEV_MSIX:
 289                 pptmsix = (struct vm_pptdev_msix *)data;
 290                 error = ppt_setup_msix(sc->vm, pptmsix->vcpu,
 291                                        pptmsix->bus, pptmsix->slot,
 292                                        pptmsix->func, pptmsix->idx,
 293                                        pptmsix->addr, pptmsix->msg,
 294                                        pptmsix->vector_control);
 295                 break;
 296         case VM_MAP_PPTDEV_MMIO:
 297                 pptmmio = (struct vm_pptdev_mmio *)data;
 298                 error = ppt_map_mmio(sc->vm, pptmmio->bus, pptmmio->slot,
 299                                      pptmmio->func, pptmmio->gpa, pptmmio->len,
 300                                      pptmmio->hpa);
 301                 break;
 302         case VM_BIND_PPTDEV:
 303                 pptdev = (struct vm_pptdev *)data;
 304                 error = vm_assign_pptdev(sc->vm, pptdev->bus, pptdev->slot,
 305                                          pptdev->func);
 306                 break;
 307         case VM_UNBIND_PPTDEV:
 308                 pptdev = (struct vm_pptdev *)data;
 309                 error = vm_unassign_pptdev(sc->vm, pptdev->bus, pptdev->slot,
 310                                            pptdev->func);
 311                 break;
 312         case VM_INJECT_EXCEPTION:
 313                 vmexc = (struct vm_exception *)data;
 314                 error = vm_inject_exception(sc->vm, vmexc->cpuid,
 315                     vmexc->vector, vmexc->error_code_valid, vmexc->error_code,
 316                     vmexc->restart_instruction);
 317                 break;
 318         case VM_INJECT_NMI:
 319                 vmnmi = (struct vm_nmi *)data;
 320                 error = vm_inject_nmi(sc->vm, vmnmi->cpuid);
 321                 break;
 322         case VM_LAPIC_IRQ:
 323                 vmirq = (struct vm_lapic_irq *)data;
 324                 error = lapic_intr_edge(sc->vm, vmirq->cpuid, vmirq->vector);
 325                 break;
 326         case VM_LAPIC_LOCAL_IRQ:
 327                 vmirq = (struct vm_lapic_irq *)data;
 328                 error = lapic_set_local_intr(sc->vm, vmirq->cpuid,
 329                     vmirq->vector);
 330                 break;
 331         case VM_LAPIC_MSI:
 332                 vmmsi = (struct vm_lapic_msi *)data;
 333                 error = lapic_intr_msi(sc->vm, vmmsi->addr, vmmsi->msg);
 334                 break;
 335         case VM_IOAPIC_ASSERT_IRQ:
 336                 ioapic_irq = (struct vm_ioapic_irq *)data;
 337                 error = vioapic_assert_irq(sc->vm, ioapic_irq->irq);
 338                 break;
 339         case VM_IOAPIC_DEASSERT_IRQ:
 340                 ioapic_irq = (struct vm_ioapic_irq *)data;
 341                 error = vioapic_deassert_irq(sc->vm, ioapic_irq->irq);
 342                 break;
 343         case VM_IOAPIC_PULSE_IRQ:
 344                 ioapic_irq = (struct vm_ioapic_irq *)data;
 345                 error = vioapic_pulse_irq(sc->vm, ioapic_irq->irq);
 346                 break;
 347         case VM_IOAPIC_PINCOUNT:
 348                 *(int *)data = vioapic_pincount(sc->vm);
 349                 break;
 350         case VM_ISA_ASSERT_IRQ:
 351                 isa_irq = (struct vm_isa_irq *)data;
 352                 error = vatpic_assert_irq(sc->vm, isa_irq->atpic_irq);
 353                 if (error == 0 && isa_irq->ioapic_irq != -1)
 354                         error = vioapic_assert_irq(sc->vm,
 355                             isa_irq->ioapic_irq);
 356                 break;
 357         case VM_ISA_DEASSERT_IRQ:
 358                 isa_irq = (struct vm_isa_irq *)data;
 359                 error = vatpic_deassert_irq(sc->vm, isa_irq->atpic_irq);
 360                 if (error == 0 && isa_irq->ioapic_irq != -1)
 361                         error = vioapic_deassert_irq(sc->vm,
 362                             isa_irq->ioapic_irq);
 363                 break;
 364         case VM_ISA_PULSE_IRQ:
 365                 isa_irq = (struct vm_isa_irq *)data;
 366                 error = vatpic_pulse_irq(sc->vm, isa_irq->atpic_irq);
 367                 if (error == 0 && isa_irq->ioapic_irq != -1)
 368                         error = vioapic_pulse_irq(sc->vm, isa_irq->ioapic_irq);
 369                 break;
 370         case VM_ISA_SET_IRQ_TRIGGER:
 371                 isa_irq_trigger = (struct vm_isa_irq_trigger *)data;
 372                 error = vatpic_set_irq_trigger(sc->vm,
 373                     isa_irq_trigger->atpic_irq, isa_irq_trigger->trigger);
 374                 break;
 375         case VM_MAP_MEMORY:
 376                 seg = (struct vm_memory_segment *)data;
 377                 error = vm_malloc(sc->vm, seg->gpa, seg->len);
 378                 break;
 379         case VM_GET_MEMORY_SEG:
 380                 seg = (struct vm_memory_segment *)data;
 381                 seg->len = 0;
 382                 (void)vm_gpabase2memseg(sc->vm, seg->gpa, seg);
 383                 error = 0;
 384                 break;
 385         case VM_GET_REGISTER:
 386                 vmreg = (struct vm_register *)data;
 387                 error = vm_get_register(sc->vm, vmreg->cpuid, vmreg->regnum,
 388                                         &vmreg->regval);
 389                 break;
 390         case VM_SET_REGISTER:
 391                 vmreg = (struct vm_register *)data;
 392                 error = vm_set_register(sc->vm, vmreg->cpuid, vmreg->regnum,
 393                                         vmreg->regval);
 394                 break;
 395         case VM_SET_SEGMENT_DESCRIPTOR:
 396                 vmsegdesc = (struct vm_seg_desc *)data;
 397                 error = vm_set_seg_desc(sc->vm, vmsegdesc->cpuid,
 398                                         vmsegdesc->regnum,
 399                                         &vmsegdesc->desc);
 400                 break;
 401         case VM_GET_SEGMENT_DESCRIPTOR:
 402                 vmsegdesc = (struct vm_seg_desc *)data;
 403                 error = vm_get_seg_desc(sc->vm, vmsegdesc->cpuid,
 404                                         vmsegdesc->regnum,
 405                                         &vmsegdesc->desc);
 406                 break;
 407         case VM_GET_CAPABILITY:
 408                 vmcap = (struct vm_capability *)data;
 409                 error = vm_get_capability(sc->vm, vmcap->cpuid,
 410                                           vmcap->captype,
 411                                           &vmcap->capval);
 412                 break;
 413         case VM_SET_CAPABILITY:
 414                 vmcap = (struct vm_capability *)data;
 415                 error = vm_set_capability(sc->vm, vmcap->cpuid,
 416                                           vmcap->captype,
 417                                           vmcap->capval);
 418                 break;
 419         case VM_SET_X2APIC_STATE:
 420                 x2apic = (struct vm_x2apic *)data;
 421                 error = vm_set_x2apic_state(sc->vm,
 422                                             x2apic->cpuid, x2apic->state);
 423                 break;
 424         case VM_GET_X2APIC_STATE:
 425                 x2apic = (struct vm_x2apic *)data;
 426                 error = vm_get_x2apic_state(sc->vm,
 427                                             x2apic->cpuid, &x2apic->state);
 428                 break;
 429         case VM_GET_GPA_PMAP:
 430                 gpapte = (struct vm_gpa_pte *)data;
 431                 pmap_get_mapping(vmspace_pmap(vm_get_vmspace(sc->vm)),
 432                                  gpapte->gpa, gpapte->pte, &gpapte->ptenum);
 433                 error = 0;
 434                 break;
 435         case VM_GET_HPET_CAPABILITIES:
 436                 error = vhpet_getcap((struct vm_hpet_cap *)data);
 437                 break;
 438         case VM_GLA2GPA: {
 439                 CTASSERT(PROT_READ == VM_PROT_READ);
 440                 CTASSERT(PROT_WRITE == VM_PROT_WRITE);
 441                 CTASSERT(PROT_EXEC == VM_PROT_EXECUTE);
 442                 gg = (struct vm_gla2gpa *)data;
 443                 error = vm_gla2gpa(sc->vm, gg->vcpuid, &gg->paging, gg->gla,
 444                     gg->prot, &gg->gpa, &gg->fault);
 445                 KASSERT(error == 0 || error == EFAULT,
 446                     ("%s: vm_gla2gpa unknown error %d", __func__, error));
 447                 break;
 448         }
 449         case VM_ACTIVATE_CPU:
 450                 vac = (struct vm_activate_cpu *)data;
 451                 error = vm_activate_cpu(sc->vm, vac->vcpuid);
 452                 break;
 453         case VM_GET_CPUS:
 454                 error = 0;
 455                 vm_cpuset = (struct vm_cpuset *)data;
 456                 size = vm_cpuset->cpusetsize;
 457                 if (size < sizeof(cpuset_t) || size > CPU_MAXSIZE / NBBY) {
 458                         error = ERANGE;
 459                         break;
 460                 }
 461                 cpuset = malloc(size, M_TEMP, M_WAITOK | M_ZERO);
 462                 if (vm_cpuset->which == VM_ACTIVE_CPUS)
 463                         *cpuset = vm_active_cpus(sc->vm);
 464                 else if (vm_cpuset->which == VM_SUSPENDED_CPUS)
 465                         *cpuset = vm_suspended_cpus(sc->vm);
 466                 else
 467                         error = EINVAL;
 468                 if (error == 0)
 469                         error = copyout(cpuset, vm_cpuset->cpus, size);
 470                 free(cpuset, M_TEMP);
 471                 break;
 472         case VM_SET_INTINFO:
 473                 vmii = (struct vm_intinfo *)data;
 474                 error = vm_exit_intinfo(sc->vm, vmii->vcpuid, vmii->info1);
 475                 break;
 476         case VM_GET_INTINFO:
 477                 vmii = (struct vm_intinfo *)data;
 478                 error = vm_get_intinfo(sc->vm, vmii->vcpuid, &vmii->info1,
 479                     &vmii->info2);
 480                 break;
 481         case VM_RTC_WRITE:
 482                 rtcdata = (struct vm_rtc_data *)data;
 483                 error = vrtc_nvram_write(sc->vm, rtcdata->offset,
 484                     rtcdata->value);
 485                 break;
 486         case VM_RTC_READ:
 487                 rtcdata = (struct vm_rtc_data *)data;
 488                 error = vrtc_nvram_read(sc->vm, rtcdata->offset,
 489                     &rtcdata->value);
 490                 break;
 491         case VM_RTC_SETTIME:
 492                 rtctime = (struct vm_rtc_time *)data;
 493                 error = vrtc_set_time(sc->vm, rtctime->secs);
 494                 break;
 495         case VM_RTC_GETTIME:
 496                 error = 0;
 497                 rtctime = (struct vm_rtc_time *)data;
 498                 rtctime->secs = vrtc_get_time(sc->vm);
 499                 break;
 500         case VM_RESTART_INSTRUCTION:
 501                 error = vm_restart_instruction(sc->vm, vcpu);
 502                 break;
 503         default:
 504                 error = ENOTTY;
 505                 break;
 506         }
 507
 508         if (state_changed == 1) {
 509                 vcpu_set_state(sc->vm, vcpu, VCPU_IDLE, false);
 510         } else if (state_changed == 2) {
 511                 for (vcpu = 0; vcpu < VM_MAXCPU; vcpu++)
 512                         vcpu_set_state(sc->vm, vcpu, VCPU_IDLE, false);
 513         }
 514
 515 done:
 516         /* Make sure that no handler returns a bogus value like ERESTART */
 517         KASSERT(error >= 0, ("vmmdev_ioctl: invalid error return %d", error));
 518         return (error);
 519 }
 520
 521 static int
 522 vmmdev_mmap_single(struct cdev *cdev, vm_ooffset_t *offset,
 523                    vm_size_t size, struct vm_object **object, int nprot)
 524 {
 525         int error;
 526         struct vmmdev_softc *sc;
 527
 528         sc = vmmdev_lookup2(cdev);
 529         if (sc != NULL && (nprot & PROT_EXEC) == 0)
 530                 error = vm_get_memobj(sc->vm, *offset, size, offset, object);
 531         else
 532                 error = EINVAL;
 533
 534         return (error);
 535 }
 536
 537 static void
 538 vmmdev_destroy(void *arg)
 539 {
 540
 541         struct vmmdev_softc *sc = arg;
 542
 543         if (sc->cdev != NULL)
 544                 destroy_dev(sc->cdev);
 545
 546         if (sc->vm != NULL)
 547                 vm_destroy(sc->vm);
 548
 549         if ((sc->flags & VSC_LINKED) != 0) {
 550                 mtx_lock(&vmmdev_mtx);
 551                 SLIST_REMOVE(&head, sc, vmmdev_softc, link);
 552                 mtx_unlock(&vmmdev_mtx);
 553         }
 554
 555         free(sc, M_VMMDEV);
 556 }
 557
 558 static int
 559 sysctl_vmm_destroy(SYSCTL_HANDLER_ARGS)
 560 {
 561         int error;
 562         char buf[VM_MAX_NAMELEN];
 563         struct vmmdev_softc *sc;
 564         struct cdev *cdev;
 565
 566         strlcpy(buf, "beavis", sizeof(buf));
 567         error = sysctl_handle_string(oidp, buf, sizeof(buf), req);
 568         if (error != 0 || req->newptr == NULL)
 569                 return (error);
 570
 571         mtx_lock(&vmmdev_mtx);
 572         sc = vmmdev_lookup(buf);
 573         if (sc == NULL || sc->cdev == NULL) {
 574                 mtx_unlock(&vmmdev_mtx);
 575                 return (EINVAL);
 576         }
 577
 578         /*
 579          * The 'cdev' will be destroyed asynchronously when 'si_threadcount'
 580          * goes down to 0 so we should not do it again in the callback.
 581          */
 582         cdev = sc->cdev;
 583         sc->cdev = NULL;
 584         mtx_unlock(&vmmdev_mtx);
 585
 586         /*
 587          * Schedule the 'cdev' to be destroyed:
 588          *
 589          * - any new operations on this 'cdev' will return an error (ENXIO).
 590          *
 591          * - when the 'si_threadcount' dwindles down to zero the 'cdev' will
 592          *   be destroyed and the callback will be invoked in a taskqueue
 593          *   context.
 594          */
 595         destroy_dev_sched_cb(cdev, vmmdev_destroy, sc);
 596
 597         return (0);
 598 }
 599 SYSCTL_PROC(_hw_vmm, OID_AUTO, destroy, CTLTYPE_STRING | CTLFLAG_RW,
 600             NULL, 0, sysctl_vmm_destroy, "A", NULL);
 601
 602 static struct cdevsw vmmdevsw = {
 603         .d_name         = "vmmdev",
 604         .d_version      = D_VERSION,
 605         .d_ioctl        = vmmdev_ioctl,
 606         .d_mmap_single  = vmmdev_mmap_single,
 607         .d_read         = vmmdev_rw,
 608         .d_write        = vmmdev_rw,
 609 };
 610
 611 static int
 612 sysctl_vmm_create(SYSCTL_HANDLER_ARGS)
 613 {
 614         int error;
 615         struct vm *vm;
 616         struct cdev *cdev;
 617         struct vmmdev_softc *sc, *sc2;
 618         char buf[VM_MAX_NAMELEN];
 619
 620         strlcpy(buf, "beavis", sizeof(buf));
 621         error = sysctl_handle_string(oidp, buf, sizeof(buf), req);
 622         if (error != 0 || req->newptr == NULL)
 623                 return (error);
 624
 625         mtx_lock(&vmmdev_mtx);
 626         sc = vmmdev_lookup(buf);
 627         mtx_unlock(&vmmdev_mtx);
 628         if (sc != NULL)
 629                 return (EEXIST);
 630
 631         error = vm_create(buf, &vm);
 632         if (error != 0)
 633                 return (error);
 634
 635         sc = malloc(sizeof(struct vmmdev_softc), M_VMMDEV, M_WAITOK | M_ZERO);
 636         sc->vm = vm;
 637
 638         /*
 639          * Lookup the name again just in case somebody sneaked in when we
 640          * dropped the lock.
 641          */
 642         mtx_lock(&vmmdev_mtx);
 643         sc2 = vmmdev_lookup(buf);
 644         if (sc2 == NULL) {
 645                 SLIST_INSERT_HEAD(&head, sc, link);
 646                 sc->flags |= VSC_LINKED;
 647         }
 648         mtx_unlock(&vmmdev_mtx);
 649
 650         if (sc2 != NULL) {
 651                 vmmdev_destroy(sc);
 652                 return (EEXIST);
 653         }
 654
 655         error = make_dev_p(MAKEDEV_CHECKNAME, &cdev, &vmmdevsw, NULL,
 656                            UID_ROOT, GID_WHEEL, 0600, "vmm/%s", buf);
 657         if (error != 0) {
 658                 vmmdev_destroy(sc);
 659                 return (error);
 660         }
 661
 662         mtx_lock(&vmmdev_mtx);
 663         sc->cdev = cdev;
 664         sc->cdev->si_drv1 = sc;
 665         mtx_unlock(&vmmdev_mtx);
 666
 667         return (0);
 668 }
 669 SYSCTL_PROC(_hw_vmm, OID_AUTO, create, CTLTYPE_STRING | CTLFLAG_RW,
 670             NULL, 0, sysctl_vmm_create, "A", NULL);
 671
 672 void
 673 vmmdev_init(void)
 674 {
 675         mtx_init(&vmmdev_mtx, "vmm device mutex", NULL, MTX_DEF);
 676 }
 677
 678 int
 679 vmmdev_cleanup(void)
 680 {
 681         int error;
 682
 683         if (SLIST_EMPTY(&head))
 684                 error = 0;
 685         else
 686                 error = EBUSY;
 687
 688         return (error);
 689 }