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/jail.h>
  37 #include <sys/queue.h>
  38 #include <sys/lock.h>
  39 #include <sys/mutex.h>
  40 #include <sys/malloc.h>
  41 #include <sys/conf.h>
  42 #include <sys/sysctl.h>
  43 #include <sys/libkern.h>
  44 #include <sys/ioccom.h>
  45 #include <sys/mman.h>
  46 #include <sys/uio.h>
  47 #include <sys/proc.h>
  48
  49 #include <vm/vm.h>
  50 #include <vm/pmap.h>
  51 #include <vm/vm_map.h>
  52 #include <vm/vm_object.h>
  53
  54 #include <machine/vmparam.h>
  55 #include <machine/vmm.h>
  56 #include <machine/vmm_instruction_emul.h>
  57 #include <machine/vmm_dev.h>
  58
  59 #include "vmm_lapic.h"
  60 #include "vmm_stat.h"
  61 #include "vmm_mem.h"
  62 #include "io/ppt.h"
  63 #include "io/vatpic.h"
  64 #include "io/vioapic.h"
  65 #include "io/vhpet.h"
  66 #include "io/vrtc.h"
  67
  68 struct devmem_softc {
  69         int     segid;
  70         char    *name;
  71         struct cdev *cdev;
  72         struct vmmdev_softc *sc;
  73         SLIST_ENTRY(devmem_softc) link;
  74 };
  75
  76 struct vmmdev_softc {
  77         struct vm       *vm;            /* vm instance cookie */
  78         struct cdev     *cdev;
  79         SLIST_ENTRY(vmmdev_softc) link;
  80         SLIST_HEAD(, devmem_softc) devmem;
  81         int             flags;
  82 };
  83 #define VSC_LINKED              0x01
  84
  85 static SLIST_HEAD(, vmmdev_softc) head;
  86
  87 static unsigned pr_allow_flag;
  88 static struct mtx vmmdev_mtx;
  89
  90 static MALLOC_DEFINE(M_VMMDEV, "vmmdev", "vmmdev");
  91
  92 SYSCTL_DECL(_hw_vmm);
  93
  94 static int vmm_priv_check(struct ucred *ucred);
  95 static int devmem_create_cdev(const char *vmname, int id, char *devmem);
  96 static void devmem_destroy(void *arg);
  97
  98 static int
  99 vmm_priv_check(struct ucred *ucred)
 100 {
 101
 102         if (jailed(ucred) &&
 103             !(ucred->cr_prison->pr_allow & pr_allow_flag))
 104                 return (EPERM);
 105
 106         return (0);
 107 }
 108
 109 static int
 110 vcpu_lock_one(struct vmmdev_softc *sc, int vcpu)
 111 {
 112         int error;
 113
 114         if (vcpu < 0 || vcpu >= vm_get_maxcpus(sc->vm))
 115                 return (EINVAL);
 116
 117         error = vcpu_set_state(sc->vm, vcpu, VCPU_FROZEN, true);
 118         return (error);
 119 }
 120
 121 static void
 122 vcpu_unlock_one(struct vmmdev_softc *sc, int vcpu)
 123 {
 124         enum vcpu_state state;
 125
 126         state = vcpu_get_state(sc->vm, vcpu, NULL);
 127         if (state != VCPU_FROZEN) {
 128                 panic("vcpu %s(%d) has invalid state %d", vm_name(sc->vm),
 129                     vcpu, state);
 130         }
 131
 132         vcpu_set_state(sc->vm, vcpu, VCPU_IDLE, false);
 133 }
 134
 135 static int
 136 vcpu_lock_all(struct vmmdev_softc *sc)
 137 {
 138         int error, vcpu;
 139         uint16_t maxcpus;
 140
 141         maxcpus = vm_get_maxcpus(sc->vm);
 142         for (vcpu = 0; vcpu < maxcpus; vcpu++) {
 143                 error = vcpu_lock_one(sc, vcpu);
 144                 if (error)
 145                         break;
 146         }
 147
 148         if (error) {
 149                 while (--vcpu >= 0)
 150                         vcpu_unlock_one(sc, vcpu);
 151         }
 152
 153         return (error);
 154 }
 155
 156 static void
 157 vcpu_unlock_all(struct vmmdev_softc *sc)
 158 {
 159         int vcpu;
 160         uint16_t maxcpus;
 161
 162         maxcpus = vm_get_maxcpus(sc->vm);
 163         for (vcpu = 0; vcpu < maxcpus; vcpu++)
 164                 vcpu_unlock_one(sc, vcpu);
 165 }
 166
 167 static struct vmmdev_softc *
 168 vmmdev_lookup(const char *name)
 169 {
 170         struct vmmdev_softc *sc;
 171
 172 #ifdef notyet   /* XXX kernel is not compiled with invariants */
 173         mtx_assert(&vmmdev_mtx, MA_OWNED);
 174 #endif
 175
 176         SLIST_FOREACH(sc, &head, link) {
 177                 if (strcmp(name, vm_name(sc->vm)) == 0)
 178                         break;
 179         }
 180
 181         return (sc);
 182 }
 183
 184 static struct vmmdev_softc *
 185 vmmdev_lookup2(struct cdev *cdev)
 186 {
 187
 188         return (cdev->si_drv1);
 189 }
 190
 191 static int
 192 vmmdev_rw(struct cdev *cdev, struct uio *uio, int flags)
 193 {
 194         int error, off, c, prot;
 195         vm_paddr_t gpa, maxaddr;
 196         void *hpa, *cookie;
 197         struct vmmdev_softc *sc;
 198         uint16_t lastcpu;
 199
 200         error = vmm_priv_check(curthread->td_ucred);
 201         if (error)
 202                 return (error);
 203
 204         sc = vmmdev_lookup2(cdev);
 205         if (sc == NULL)
 206                 return (ENXIO);
 207
 208         /*
 209          * Get a read lock on the guest memory map by freezing any vcpu.
 210          */
 211         lastcpu = vm_get_maxcpus(sc->vm) - 1;
 212         error = vcpu_lock_one(sc, lastcpu);
 213         if (error)
 214                 return (error);
 215
 216         prot = (uio->uio_rw == UIO_WRITE ? VM_PROT_WRITE : VM_PROT_READ);
 217         maxaddr = vmm_sysmem_maxaddr(sc->vm);
 218         while (uio->uio_resid > 0 && error == 0) {
 219                 gpa = uio->uio_offset;
 220                 off = gpa & PAGE_MASK;
 221                 c = min(uio->uio_resid, PAGE_SIZE - off);
 222
 223                 /*
 224                  * The VM has a hole in its physical memory map. If we want to
 225                  * use 'dd' to inspect memory beyond the hole we need to
 226                  * provide bogus data for memory that lies in the hole.
 227                  *
 228                  * Since this device does not support lseek(2), dd(1) will
 229                  * read(2) blocks of data to simulate the lseek(2).
 230                  */
 231                 hpa = vm_gpa_hold(sc->vm, lastcpu, gpa, c,
 232                     prot, &cookie);
 233                 if (hpa == NULL) {
 234                         if (uio->uio_rw == UIO_READ && gpa < maxaddr)
 235                                 error = uiomove(__DECONST(void *, zero_region),
 236                                     c, uio);
 237                         else
 238                                 error = EFAULT;
 239                 } else {
 240                         error = uiomove(hpa, c, uio);
 241                         vm_gpa_release(cookie);
 242                 }
 243         }
 244         vcpu_unlock_one(sc, lastcpu);
 245         return (error);
 246 }
 247
 248 CTASSERT(sizeof(((struct vm_memseg *)0)->name) >= SPECNAMELEN + 1);
 249
 250 static int
 251 get_memseg(struct vmmdev_softc *sc, struct vm_memseg *mseg)
 252 {
 253         struct devmem_softc *dsc;
 254         int error;
 255         bool sysmem;
 256
 257         error = vm_get_memseg(sc->vm, mseg->segid, &mseg->len, &sysmem, NULL);
 258         if (error || mseg->len == 0)
 259                 return (error);
 260
 261         if (!sysmem) {
 262                 SLIST_FOREACH(dsc, &sc->devmem, link) {
 263                         if (dsc->segid == mseg->segid)
 264                                 break;
 265                 }
 266                 KASSERT(dsc != NULL, ("%s: devmem segment %d not found",
 267                     __func__, mseg->segid));
 268                 error = copystr(dsc->name, mseg->name, SPECNAMELEN + 1, NULL);
 269         } else {
 270                 bzero(mseg->name, sizeof(mseg->name));
 271         }
 272
 273         return (error);
 274 }
 275
 276 static int
 277 alloc_memseg(struct vmmdev_softc *sc, struct vm_memseg *mseg)
 278 {
 279         char *name;
 280         int error;
 281         bool sysmem;
 282
 283         error = 0;
 284         name = NULL;
 285         sysmem = true;
 286
 287         if (VM_MEMSEG_NAME(mseg)) {
 288                 sysmem = false;
 289                 name = malloc(SPECNAMELEN + 1, M_VMMDEV, M_WAITOK);
 290                 error = copystr(mseg->name, name, SPECNAMELEN + 1, 0);
 291                 if (error)
 292                         goto done;
 293         }
 294
 295         error = vm_alloc_memseg(sc->vm, mseg->segid, mseg->len, sysmem);
 296         if (error)
 297                 goto done;
 298
 299         if (VM_MEMSEG_NAME(mseg)) {
 300                 error = devmem_create_cdev(vm_name(sc->vm), mseg->segid, name);
 301                 if (error)
 302                         vm_free_memseg(sc->vm, mseg->segid);
 303                 else
 304                         name = NULL;    /* freed when 'cdev' is destroyed */
 305         }
 306 done:
 307         free(name, M_VMMDEV);
 308         return (error);
 309 }
 310
 311 static int
 312 vm_get_register_set(struct vm *vm, int vcpu, unsigned int count, int *regnum,
 313     uint64_t *regval)
 314 {
 315         int error, i;
 316
 317         error = 0;
 318         for (i = 0; i < count; i++) {
 319                 error = vm_get_register(vm, vcpu, regnum[i], &regval[i]);
 320                 if (error)
 321                         break;
 322         }
 323         return (error);
 324 }
 325
 326 static int
 327 vm_set_register_set(struct vm *vm, int vcpu, unsigned int count, int *regnum,
 328     uint64_t *regval)
 329 {
 330         int error, i;
 331
 332         error = 0;
 333         for (i = 0; i < count; i++) {
 334                 error = vm_set_register(vm, vcpu, regnum[i], regval[i]);
 335                 if (error)
 336                         break;
 337         }
 338         return (error);
 339 }
 340
 341 static int
 342 vmmdev_ioctl(struct cdev *cdev, u_long cmd, caddr_t data, int fflag,
 343              struct thread *td)
 344 {
 345         int error, vcpu, state_changed, size;
 346         cpuset_t *cpuset;
 347         struct vmmdev_softc *sc;
 348         struct vm_register *vmreg;
 349         struct vm_seg_desc *vmsegdesc;
 350         struct vm_register_set *vmregset;
 351         struct vm_run *vmrun;
 352         struct vm_exception *vmexc;
 353         struct vm_lapic_irq *vmirq;
 354         struct vm_lapic_msi *vmmsi;
 355         struct vm_ioapic_irq *ioapic_irq;
 356         struct vm_isa_irq *isa_irq;
 357         struct vm_isa_irq_trigger *isa_irq_trigger;
 358         struct vm_capability *vmcap;
 359         struct vm_pptdev *pptdev;
 360         struct vm_pptdev_mmio *pptmmio;
 361         struct vm_pptdev_msi *pptmsi;
 362         struct vm_pptdev_msix *pptmsix;
 363         struct vm_nmi *vmnmi;
 364         struct vm_stats *vmstats;
 365         struct vm_stat_desc *statdesc;
 366         struct vm_x2apic *x2apic;
 367         struct vm_gpa_pte *gpapte;
 368         struct vm_suspend *vmsuspend;
 369         struct vm_gla2gpa *gg;
 370         struct vm_activate_cpu *vac;
 371         struct vm_cpuset *vm_cpuset;
 372         struct vm_intinfo *vmii;
 373         struct vm_rtc_time *rtctime;
 374         struct vm_rtc_data *rtcdata;
 375         struct vm_memmap *mm;
 376         struct vm_cpu_topology *topology;
 377         uint64_t *regvals;
 378         int *regnums;
 379
 380         error = vmm_priv_check(curthread->td_ucred);
 381         if (error)
 382                 return (error);
 383
 384         sc = vmmdev_lookup2(cdev);
 385         if (sc == NULL)
 386                 return (ENXIO);
 387
 388         vcpu = -1;
 389         state_changed = 0;
 390
 391         /*
 392          * Some VMM ioctls can operate only on vcpus that are not running.
 393          */
 394         switch (cmd) {
 395         case VM_RUN:
 396         case VM_GET_REGISTER:
 397         case VM_SET_REGISTER:
 398         case VM_GET_SEGMENT_DESCRIPTOR:
 399         case VM_SET_SEGMENT_DESCRIPTOR:
 400         case VM_GET_REGISTER_SET:
 401         case VM_SET_REGISTER_SET:
 402         case VM_INJECT_EXCEPTION:
 403         case VM_GET_CAPABILITY:
 404         case VM_SET_CAPABILITY:
 405         case VM_PPTDEV_MSI:
 406         case VM_PPTDEV_MSIX:
 407         case VM_SET_X2APIC_STATE:
 408         case VM_GLA2GPA:
 409         case VM_GLA2GPA_NOFAULT:
 410         case VM_ACTIVATE_CPU:
 411         case VM_SET_INTINFO:
 412         case VM_GET_INTINFO:
 413         case VM_RESTART_INSTRUCTION:
 414                 /*
 415                  * XXX fragile, handle with care
 416                  * Assumes that the first field of the ioctl data is the vcpu.
 417                  */
 418                 vcpu = *(int *)data;
 419                 error = vcpu_lock_one(sc, vcpu);
 420                 if (error)
 421                         goto done;
 422                 state_changed = 1;
 423                 break;
 424
 425         case VM_MAP_PPTDEV_MMIO:
 426         case VM_BIND_PPTDEV:
 427         case VM_UNBIND_PPTDEV:
 428         case VM_ALLOC_MEMSEG:
 429         case VM_MMAP_MEMSEG:
 430         case VM_REINIT:
 431                 /*
 432                  * ioctls that operate on the entire virtual machine must
 433                  * prevent all vcpus from running.
 434                  */
 435                 error = vcpu_lock_all(sc);
 436                 if (error)
 437                         goto done;
 438                 state_changed = 2;
 439                 break;
 440
 441         case VM_GET_MEMSEG:
 442         case VM_MMAP_GETNEXT:
 443                 /*
 444                  * Lock a vcpu to make sure that the memory map cannot be
 445                  * modified while it is being inspected.
 446                  */
 447                 vcpu = vm_get_maxcpus(sc->vm) - 1;
 448                 error = vcpu_lock_one(sc, vcpu);
 449                 if (error)
 450                         goto done;
 451                 state_changed = 1;
 452                 break;
 453
 454         default:
 455                 break;
 456         }
 457
 458         switch(cmd) {
 459         case VM_RUN:
 460                 vmrun = (struct vm_run *)data;
 461                 error = vm_run(sc->vm, vmrun);
 462                 break;
 463         case VM_SUSPEND:
 464                 vmsuspend = (struct vm_suspend *)data;
 465                 error = vm_suspend(sc->vm, vmsuspend->how);
 466                 break;
 467         case VM_REINIT:
 468                 error = vm_reinit(sc->vm);
 469                 break;
 470         case VM_STAT_DESC: {
 471                 statdesc = (struct vm_stat_desc *)data;
 472                 error = vmm_stat_desc_copy(statdesc->index,
 473                                         statdesc->desc, sizeof(statdesc->desc));
 474                 break;
 475         }
 476         case VM_STATS: {
 477                 CTASSERT(MAX_VM_STATS >= MAX_VMM_STAT_ELEMS);
 478                 vmstats = (struct vm_stats *)data;
 479                 getmicrotime(&vmstats->tv);
 480                 error = vmm_stat_copy(sc->vm, vmstats->cpuid,
 481                                       &vmstats->num_entries, vmstats->statbuf);
 482                 break;
 483         }
 484         case VM_PPTDEV_MSI:
 485                 pptmsi = (struct vm_pptdev_msi *)data;
 486                 error = ppt_setup_msi(sc->vm, pptmsi->vcpu,
 487                                       pptmsi->bus, pptmsi->slot, pptmsi->func,
 488                                       pptmsi->addr, pptmsi->msg,
 489                                       pptmsi->numvec);
 490                 break;
 491         case VM_PPTDEV_MSIX:
 492                 pptmsix = (struct vm_pptdev_msix *)data;
 493                 error = ppt_setup_msix(sc->vm, pptmsix->vcpu,
 494                                        pptmsix->bus, pptmsix->slot,
 495                                        pptmsix->func, pptmsix->idx,
 496                                        pptmsix->addr, pptmsix->msg,
 497                                        pptmsix->vector_control);
 498                 break;
 499         case VM_MAP_PPTDEV_MMIO:
 500                 pptmmio = (struct vm_pptdev_mmio *)data;
 501                 error = ppt_map_mmio(sc->vm, pptmmio->bus, pptmmio->slot,
 502                                      pptmmio->func, pptmmio->gpa, pptmmio->len,
 503                                      pptmmio->hpa);
 504                 break;
 505         case VM_BIND_PPTDEV:
 506                 pptdev = (struct vm_pptdev *)data;
 507                 error = vm_assign_pptdev(sc->vm, pptdev->bus, pptdev->slot,
 508                                          pptdev->func);
 509                 break;
 510         case VM_UNBIND_PPTDEV:
 511                 pptdev = (struct vm_pptdev *)data;
 512                 error = vm_unassign_pptdev(sc->vm, pptdev->bus, pptdev->slot,
 513                                            pptdev->func);
 514                 break;
 515         case VM_INJECT_EXCEPTION:
 516                 vmexc = (struct vm_exception *)data;
 517                 error = vm_inject_exception(sc->vm, vmexc->cpuid,
 518                     vmexc->vector, vmexc->error_code_valid, vmexc->error_code,
 519                     vmexc->restart_instruction);
 520                 break;
 521         case VM_INJECT_NMI:
 522                 vmnmi = (struct vm_nmi *)data;
 523                 error = vm_inject_nmi(sc->vm, vmnmi->cpuid);
 524                 break;
 525         case VM_LAPIC_IRQ:
 526                 vmirq = (struct vm_lapic_irq *)data;
 527                 error = lapic_intr_edge(sc->vm, vmirq->cpuid, vmirq->vector);
 528                 break;
 529         case VM_LAPIC_LOCAL_IRQ:
 530                 vmirq = (struct vm_lapic_irq *)data;
 531                 error = lapic_set_local_intr(sc->vm, vmirq->cpuid,
 532                     vmirq->vector);
 533                 break;
 534         case VM_LAPIC_MSI:
 535                 vmmsi = (struct vm_lapic_msi *)data;
 536                 error = lapic_intr_msi(sc->vm, vmmsi->addr, vmmsi->msg);
 537                 break;
 538         case VM_IOAPIC_ASSERT_IRQ:
 539                 ioapic_irq = (struct vm_ioapic_irq *)data;
 540                 error = vioapic_assert_irq(sc->vm, ioapic_irq->irq);
 541                 break;
 542         case VM_IOAPIC_DEASSERT_IRQ:
 543                 ioapic_irq = (struct vm_ioapic_irq *)data;
 544                 error = vioapic_deassert_irq(sc->vm, ioapic_irq->irq);
 545                 break;
 546         case VM_IOAPIC_PULSE_IRQ:
 547                 ioapic_irq = (struct vm_ioapic_irq *)data;
 548                 error = vioapic_pulse_irq(sc->vm, ioapic_irq->irq);
 549                 break;
 550         case VM_IOAPIC_PINCOUNT:
 551                 *(int *)data = vioapic_pincount(sc->vm);
 552                 break;
 553         case VM_ISA_ASSERT_IRQ:
 554                 isa_irq = (struct vm_isa_irq *)data;
 555                 error = vatpic_assert_irq(sc->vm, isa_irq->atpic_irq);
 556                 if (error == 0 && isa_irq->ioapic_irq != -1)
 557                         error = vioapic_assert_irq(sc->vm,
 558                             isa_irq->ioapic_irq);
 559                 break;
 560         case VM_ISA_DEASSERT_IRQ:
 561                 isa_irq = (struct vm_isa_irq *)data;
 562                 error = vatpic_deassert_irq(sc->vm, isa_irq->atpic_irq);
 563                 if (error == 0 && isa_irq->ioapic_irq != -1)
 564                         error = vioapic_deassert_irq(sc->vm,
 565                             isa_irq->ioapic_irq);
 566                 break;
 567         case VM_ISA_PULSE_IRQ:
 568                 isa_irq = (struct vm_isa_irq *)data;
 569                 error = vatpic_pulse_irq(sc->vm, isa_irq->atpic_irq);
 570                 if (error == 0 && isa_irq->ioapic_irq != -1)
 571                         error = vioapic_pulse_irq(sc->vm, isa_irq->ioapic_irq);
 572                 break;
 573         case VM_ISA_SET_IRQ_TRIGGER:
 574                 isa_irq_trigger = (struct vm_isa_irq_trigger *)data;
 575                 error = vatpic_set_irq_trigger(sc->vm,
 576                     isa_irq_trigger->atpic_irq, isa_irq_trigger->trigger);
 577                 break;
 578         case VM_MMAP_GETNEXT:
 579                 mm = (struct vm_memmap *)data;
 580                 error = vm_mmap_getnext(sc->vm, &mm->gpa, &mm->segid,
 581                     &mm->segoff, &mm->len, &mm->prot, &mm->flags);
 582                 break;
 583         case VM_MMAP_MEMSEG:
 584                 mm = (struct vm_memmap *)data;
 585                 error = vm_mmap_memseg(sc->vm, mm->gpa, mm->segid, mm->segoff,
 586                     mm->len, mm->prot, mm->flags);
 587                 break;
 588         case VM_ALLOC_MEMSEG:
 589                 error = alloc_memseg(sc, (struct vm_memseg *)data);
 590                 break;
 591         case VM_GET_MEMSEG:
 592                 error = get_memseg(sc, (struct vm_memseg *)data);
 593                 break;
 594         case VM_GET_REGISTER:
 595                 vmreg = (struct vm_register *)data;
 596                 error = vm_get_register(sc->vm, vmreg->cpuid, vmreg->regnum,
 597                                         &vmreg->regval);
 598                 break;
 599         case VM_SET_REGISTER:
 600                 vmreg = (struct vm_register *)data;
 601                 error = vm_set_register(sc->vm, vmreg->cpuid, vmreg->regnum,
 602                                         vmreg->regval);
 603                 break;
 604         case VM_SET_SEGMENT_DESCRIPTOR:
 605                 vmsegdesc = (struct vm_seg_desc *)data;
 606                 error = vm_set_seg_desc(sc->vm, vmsegdesc->cpuid,
 607                                         vmsegdesc->regnum,
 608                                         &vmsegdesc->desc);
 609                 break;
 610         case VM_GET_SEGMENT_DESCRIPTOR:
 611                 vmsegdesc = (struct vm_seg_desc *)data;
 612                 error = vm_get_seg_desc(sc->vm, vmsegdesc->cpuid,
 613                                         vmsegdesc->regnum,
 614                                         &vmsegdesc->desc);
 615                 break;
 616         case VM_GET_REGISTER_SET:
 617                 vmregset = (struct vm_register_set *)data;
 618                 if (vmregset->count > VM_REG_LAST) {
 619                         error = EINVAL;
 620                         break;
 621                 }
 622                 regvals = malloc(sizeof(regvals[0]) * vmregset->count, M_VMMDEV,
 623                     M_WAITOK);
 624                 regnums = malloc(sizeof(regnums[0]) * vmregset->count, M_VMMDEV,
 625                     M_WAITOK);
 626                 error = copyin(vmregset->regnums, regnums, sizeof(regnums[0]) *
 627                     vmregset->count);
 628                 if (error == 0)
 629                         error = vm_get_register_set(sc->vm, vmregset->cpuid,
 630                             vmregset->count, regnums, regvals);
 631                 if (error == 0)
 632                         error = copyout(regvals, vmregset->regvals,
 633                             sizeof(regvals[0]) * vmregset->count);
 634                 free(regvals, M_VMMDEV);
 635                 free(regnums, M_VMMDEV);
 636                 break;
 637         case VM_SET_REGISTER_SET:
 638                 vmregset = (struct vm_register_set *)data;
 639                 if (vmregset->count > VM_REG_LAST) {
 640                         error = EINVAL;
 641                         break;
 642                 }
 643                 regvals = malloc(sizeof(regvals[0]) * vmregset->count, M_VMMDEV,
 644                     M_WAITOK);
 645                 regnums = malloc(sizeof(regnums[0]) * vmregset->count, M_VMMDEV,
 646                     M_WAITOK);
 647                 error = copyin(vmregset->regnums, regnums, sizeof(regnums[0]) *
 648                     vmregset->count);
 649                 if (error == 0)
 650                         error = copyin(vmregset->regvals, regvals,
 651                             sizeof(regvals[0]) * vmregset->count);
 652                 if (error == 0)
 653                         error = vm_set_register_set(sc->vm, vmregset->cpuid,
 654                             vmregset->count, regnums, regvals);
 655                 free(regvals, M_VMMDEV);
 656                 free(regnums, M_VMMDEV);
 657                 break;
 658         case VM_GET_CAPABILITY:
 659                 vmcap = (struct vm_capability *)data;
 660                 error = vm_get_capability(sc->vm, vmcap->cpuid,
 661                                           vmcap->captype,
 662                                           &vmcap->capval);
 663                 break;
 664         case VM_SET_CAPABILITY:
 665                 vmcap = (struct vm_capability *)data;
 666                 error = vm_set_capability(sc->vm, vmcap->cpuid,
 667                                           vmcap->captype,
 668                                           vmcap->capval);
 669                 break;
 670         case VM_SET_X2APIC_STATE:
 671                 x2apic = (struct vm_x2apic *)data;
 672                 error = vm_set_x2apic_state(sc->vm,
 673                                             x2apic->cpuid, x2apic->state);
 674                 break;
 675         case VM_GET_X2APIC_STATE:
 676                 x2apic = (struct vm_x2apic *)data;
 677                 error = vm_get_x2apic_state(sc->vm,
 678                                             x2apic->cpuid, &x2apic->state);
 679                 break;
 680         case VM_GET_GPA_PMAP:
 681                 gpapte = (struct vm_gpa_pte *)data;
 682                 pmap_get_mapping(vmspace_pmap(vm_get_vmspace(sc->vm)),
 683                                  gpapte->gpa, gpapte->pte, &gpapte->ptenum);
 684                 error = 0;
 685                 break;
 686         case VM_GET_HPET_CAPABILITIES:
 687                 error = vhpet_getcap((struct vm_hpet_cap *)data);
 688                 break;
 689         case VM_GLA2GPA: {
 690                 CTASSERT(PROT_READ == VM_PROT_READ);
 691                 CTASSERT(PROT_WRITE == VM_PROT_WRITE);
 692                 CTASSERT(PROT_EXEC == VM_PROT_EXECUTE);
 693                 gg = (struct vm_gla2gpa *)data;
 694                 error = vm_gla2gpa(sc->vm, gg->vcpuid, &gg->paging, gg->gla,
 695                     gg->prot, &gg->gpa, &gg->fault);
 696                 KASSERT(error == 0 || error == EFAULT,
 697                     ("%s: vm_gla2gpa unknown error %d", __func__, error));
 698                 break;
 699         }
 700         case VM_GLA2GPA_NOFAULT:
 701                 gg = (struct vm_gla2gpa *)data;
 702                 error = vm_gla2gpa_nofault(sc->vm, gg->vcpuid, &gg->paging,
 703                     gg->gla, gg->prot, &gg->gpa, &gg->fault);
 704                 KASSERT(error == 0 || error == EFAULT,
 705                     ("%s: vm_gla2gpa unknown error %d", __func__, error));
 706                 break;
 707         case VM_ACTIVATE_CPU:
 708                 vac = (struct vm_activate_cpu *)data;
 709                 error = vm_activate_cpu(sc->vm, vac->vcpuid);
 710                 break;
 711         case VM_GET_CPUS:
 712                 error = 0;
 713                 vm_cpuset = (struct vm_cpuset *)data;
 714                 size = vm_cpuset->cpusetsize;
 715                 if (size < sizeof(cpuset_t) || size > CPU_MAXSIZE / NBBY) {
 716                         error = ERANGE;
 717                         break;
 718                 }
 719                 cpuset = malloc(size, M_TEMP, M_WAITOK | M_ZERO);
 720                 if (vm_cpuset->which == VM_ACTIVE_CPUS)
 721                         *cpuset = vm_active_cpus(sc->vm);
 722                 else if (vm_cpuset->which == VM_SUSPENDED_CPUS)
 723                         *cpuset = vm_suspended_cpus(sc->vm);
 724                 else if (vm_cpuset->which == VM_DEBUG_CPUS)
 725                         *cpuset = vm_debug_cpus(sc->vm);
 726                 else
 727                         error = EINVAL;
 728                 if (error == 0)
 729                         error = copyout(cpuset, vm_cpuset->cpus, size);
 730                 free(cpuset, M_TEMP);
 731                 break;
 732         case VM_SUSPEND_CPU:
 733                 vac = (struct vm_activate_cpu *)data;
 734                 error = vm_suspend_cpu(sc->vm, vac->vcpuid);
 735                 break;
 736         case VM_RESUME_CPU:
 737                 vac = (struct vm_activate_cpu *)data;
 738                 error = vm_resume_cpu(sc->vm, vac->vcpuid);
 739                 break;
 740         case VM_SET_INTINFO:
 741                 vmii = (struct vm_intinfo *)data;
 742                 error = vm_exit_intinfo(sc->vm, vmii->vcpuid, vmii->info1);
 743                 break;
 744         case VM_GET_INTINFO:
 745                 vmii = (struct vm_intinfo *)data;
 746                 error = vm_get_intinfo(sc->vm, vmii->vcpuid, &vmii->info1,
 747                     &vmii->info2);
 748                 break;
 749         case VM_RTC_WRITE:
 750                 rtcdata = (struct vm_rtc_data *)data;
 751                 error = vrtc_nvram_write(sc->vm, rtcdata->offset,
 752                     rtcdata->value);
 753                 break;
 754         case VM_RTC_READ:
 755                 rtcdata = (struct vm_rtc_data *)data;
 756                 error = vrtc_nvram_read(sc->vm, rtcdata->offset,
 757                     &rtcdata->value);
 758                 break;
 759         case VM_RTC_SETTIME:
 760                 rtctime = (struct vm_rtc_time *)data;
 761                 error = vrtc_set_time(sc->vm, rtctime->secs);
 762                 break;
 763         case VM_RTC_GETTIME:
 764                 error = 0;
 765                 rtctime = (struct vm_rtc_time *)data;
 766                 rtctime->secs = vrtc_get_time(sc->vm);
 767                 break;
 768         case VM_RESTART_INSTRUCTION:
 769                 error = vm_restart_instruction(sc->vm, vcpu);
 770                 break;
 771         case VM_SET_TOPOLOGY:
 772                 topology = (struct vm_cpu_topology *)data;
 773                 error = vm_set_topology(sc->vm, topology->sockets,
 774                     topology->cores, topology->threads, topology->maxcpus);
 775                 break;
 776         case VM_GET_TOPOLOGY:
 777                 topology = (struct vm_cpu_topology *)data;
 778                 vm_get_topology(sc->vm, &topology->sockets, &topology->cores,
 779                     &topology->threads, &topology->maxcpus);
 780                 error = 0;
 781                 break;
 782         default:
 783                 error = ENOTTY;
 784                 break;
 785         }
 786
 787         if (state_changed == 1)
 788                 vcpu_unlock_one(sc, vcpu);
 789         else if (state_changed == 2)
 790                 vcpu_unlock_all(sc);
 791
 792 done:
 793         /* Make sure that no handler returns a bogus value like ERESTART */
 794         KASSERT(error >= 0, ("vmmdev_ioctl: invalid error return %d", error));
 795         return (error);
 796 }
 797
 798 static int
 799 vmmdev_mmap_single(struct cdev *cdev, vm_ooffset_t *offset, vm_size_t mapsize,
 800     struct vm_object **objp, int nprot)
 801 {
 802         struct vmmdev_softc *sc;
 803         vm_paddr_t gpa;
 804         size_t len;
 805         vm_ooffset_t segoff, first, last;
 806         int error, found, segid;
 807         uint16_t lastcpu;
 808         bool sysmem;
 809
 810         error = vmm_priv_check(curthread->td_ucred);
 811         if (error)
 812                 return (error);
 813
 814         first = *offset;
 815         last = first + mapsize;
 816         if ((nprot & PROT_EXEC) || first < 0 || first >= last)
 817                 return (EINVAL);
 818
 819         sc = vmmdev_lookup2(cdev);
 820         if (sc == NULL) {
 821                 /* virtual machine is in the process of being created */
 822                 return (EINVAL);
 823         }
 824
 825         /*
 826          * Get a read lock on the guest memory map by freezing any vcpu.
 827          */
 828         lastcpu = vm_get_maxcpus(sc->vm) - 1;
 829         error = vcpu_lock_one(sc, lastcpu);
 830         if (error)
 831                 return (error);
 832
 833         gpa = 0;
 834         found = 0;
 835         while (!found) {
 836                 error = vm_mmap_getnext(sc->vm, &gpa, &segid, &segoff, &len,
 837                     NULL, NULL);
 838                 if (error)
 839                         break;
 840
 841                 if (first >= gpa && last <= gpa + len)
 842                         found = 1;
 843                 else
 844                         gpa += len;
 845         }
 846
 847         if (found) {
 848                 error = vm_get_memseg(sc->vm, segid, &len, &sysmem, objp);
 849                 KASSERT(error == 0 && *objp != NULL,
 850                     ("%s: invalid memory segment %d", __func__, segid));
 851                 if (sysmem) {
 852                         vm_object_reference(*objp);
 853                         *offset = segoff + (first - gpa);
 854                 } else {
 855                         error = EINVAL;
 856                 }
 857         }
 858         vcpu_unlock_one(sc, lastcpu);
 859         return (error);
 860 }
 861
 862 static void
 863 vmmdev_destroy(void *arg)
 864 {
 865         struct vmmdev_softc *sc = arg;
 866         struct devmem_softc *dsc;
 867         int error;
 868
 869         error = vcpu_lock_all(sc);
 870         KASSERT(error == 0, ("%s: error %d freezing vcpus", __func__, error));
 871
 872         while ((dsc = SLIST_FIRST(&sc->devmem)) != NULL) {
 873                 KASSERT(dsc->cdev == NULL, ("%s: devmem not free", __func__));
 874                 SLIST_REMOVE_HEAD(&sc->devmem, link);
 875                 free(dsc->name, M_VMMDEV);
 876                 free(dsc, M_VMMDEV);
 877         }
 878
 879         if (sc->cdev != NULL)
 880                 destroy_dev(sc->cdev);
 881
 882         if (sc->vm != NULL)
 883                 vm_destroy(sc->vm);
 884
 885         if ((sc->flags & VSC_LINKED) != 0) {
 886                 mtx_lock(&vmmdev_mtx);
 887                 SLIST_REMOVE(&head, sc, vmmdev_softc, link);
 888                 mtx_unlock(&vmmdev_mtx);
 889         }
 890
 891         free(sc, M_VMMDEV);
 892 }
 893
 894 static int
 895 sysctl_vmm_destroy(SYSCTL_HANDLER_ARGS)
 896 {
 897         int error;
 898         char buf[VM_MAX_NAMELEN];
 899         struct devmem_softc *dsc;
 900         struct vmmdev_softc *sc;
 901         struct cdev *cdev;
 902
 903         error = vmm_priv_check(req->td->td_ucred);
 904         if (error)
 905                 return (error);
 906
 907         strlcpy(buf, "beavis", sizeof(buf));
 908         error = sysctl_handle_string(oidp, buf, sizeof(buf), req);
 909         if (error != 0 || req->newptr == NULL)
 910                 return (error);
 911
 912         mtx_lock(&vmmdev_mtx);
 913         sc = vmmdev_lookup(buf);
 914         if (sc == NULL || sc->cdev == NULL) {
 915                 mtx_unlock(&vmmdev_mtx);
 916                 return (EINVAL);
 917         }
 918
 919         /*
 920          * The 'cdev' will be destroyed asynchronously when 'si_threadcount'
 921          * goes down to 0 so we should not do it again in the callback.
 922          *
 923          * Setting 'sc->cdev' to NULL is also used to indicate that the VM
 924          * is scheduled for destruction.
 925          */
 926         cdev = sc->cdev;
 927         sc->cdev = NULL;
 928         mtx_unlock(&vmmdev_mtx);
 929
 930         /*
 931          * Schedule all cdevs to be destroyed:
 932          *
 933          * - any new operations on the 'cdev' will return an error (ENXIO).
 934          *
 935          * - when the 'si_threadcount' dwindles down to zero the 'cdev' will
 936          *   be destroyed and the callback will be invoked in a taskqueue
 937          *   context.
 938          *
 939          * - the 'devmem' cdevs are destroyed before the virtual machine 'cdev'
 940          */
 941         SLIST_FOREACH(dsc, &sc->devmem, link) {
 942                 KASSERT(dsc->cdev != NULL, ("devmem cdev already destroyed"));
 943                 destroy_dev_sched_cb(dsc->cdev, devmem_destroy, dsc);
 944         }
 945         destroy_dev_sched_cb(cdev, vmmdev_destroy, sc);
 946         return (0);
 947 }
 948 SYSCTL_PROC(_hw_vmm, OID_AUTO, destroy,
 949             CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_PRISON,
 950             NULL, 0, sysctl_vmm_destroy, "A", NULL);
 951
 952 static struct cdevsw vmmdevsw = {
 953         .d_name         = "vmmdev",
 954         .d_version      = D_VERSION,
 955         .d_ioctl        = vmmdev_ioctl,
 956         .d_mmap_single  = vmmdev_mmap_single,
 957         .d_read         = vmmdev_rw,
 958         .d_write        = vmmdev_rw,
 959 };
 960
 961 static int
 962 sysctl_vmm_create(SYSCTL_HANDLER_ARGS)
 963 {
 964         int error;
 965         struct vm *vm;
 966         struct cdev *cdev;
 967         struct vmmdev_softc *sc, *sc2;
 968         char buf[VM_MAX_NAMELEN];
 969
 970         error = vmm_priv_check(req->td->td_ucred);
 971         if (error)
 972                 return (error);
 973
 974         strlcpy(buf, "beavis", sizeof(buf));
 975         error = sysctl_handle_string(oidp, buf, sizeof(buf), req);
 976         if (error != 0 || req->newptr == NULL)
 977                 return (error);
 978
 979         mtx_lock(&vmmdev_mtx);
 980         sc = vmmdev_lookup(buf);
 981         mtx_unlock(&vmmdev_mtx);
 982         if (sc != NULL)
 983                 return (EEXIST);
 984
 985         error = vm_create(buf, &vm);
 986         if (error != 0)
 987                 return (error);
 988
 989         sc = malloc(sizeof(struct vmmdev_softc), M_VMMDEV, M_WAITOK | M_ZERO);
 990         sc->vm = vm;
 991         SLIST_INIT(&sc->devmem);
 992
 993         /*
 994          * Lookup the name again just in case somebody sneaked in when we
 995          * dropped the lock.
 996          */
 997         mtx_lock(&vmmdev_mtx);
 998         sc2 = vmmdev_lookup(buf);
 999         if (sc2 == NULL) {
1000                 SLIST_INSERT_HEAD(&head, sc, link);
1001                 sc->flags |= VSC_LINKED;
1002         }
1003         mtx_unlock(&vmmdev_mtx);
1004
1005         if (sc2 != NULL) {
1006                 vmmdev_destroy(sc);
1007                 return (EEXIST);
1008         }
1009
1010         error = make_dev_p(MAKEDEV_CHECKNAME, &cdev, &vmmdevsw, NULL,
1011                            UID_ROOT, GID_WHEEL, 0600, "vmm/%s", buf);
1012         if (error != 0) {
1013                 vmmdev_destroy(sc);
1014                 return (error);
1015         }
1016
1017         mtx_lock(&vmmdev_mtx);
1018         sc->cdev = cdev;
1019         sc->cdev->si_drv1 = sc;
1020         mtx_unlock(&vmmdev_mtx);
1021
1022         return (0);
1023 }
1024 SYSCTL_PROC(_hw_vmm, OID_AUTO, create,
1025             CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_PRISON,
1026             NULL, 0, sysctl_vmm_create, "A", NULL);
1027
1028 void
1029 vmmdev_init(void)
1030 {
1031         mtx_init(&vmmdev_mtx, "vmm device mutex", NULL, MTX_DEF);
1032         pr_allow_flag = prison_add_allow(NULL, "vmm", NULL,
1033             "Allow use of vmm in a jail.");
1034 }
1035
1036 int
1037 vmmdev_cleanup(void)
1038 {
1039         int error;
1040
1041         if (SLIST_EMPTY(&head))
1042                 error = 0;
1043         else
1044                 error = EBUSY;
1045
1046         return (error);
1047 }
1048
1049 static int
1050 devmem_mmap_single(struct cdev *cdev, vm_ooffset_t *offset, vm_size_t len,
1051     struct vm_object **objp, int nprot)
1052 {
1053         struct devmem_softc *dsc;
1054         vm_ooffset_t first, last;
1055         size_t seglen;
1056         int error;
1057         uint16_t lastcpu;
1058         bool sysmem;
1059
1060         dsc = cdev->si_drv1;
1061         if (dsc == NULL) {
1062                 /* 'cdev' has been created but is not ready for use */
1063                 return (ENXIO);
1064         }
1065
1066         first = *offset;
1067         last = *offset + len;
1068         if ((nprot & PROT_EXEC) || first < 0 || first >= last)
1069                 return (EINVAL);
1070
1071         lastcpu = vm_get_maxcpus(dsc->sc->vm) - 1;
1072         error = vcpu_lock_one(dsc->sc, lastcpu);
1073         if (error)
1074                 return (error);
1075
1076         error = vm_get_memseg(dsc->sc->vm, dsc->segid, &seglen, &sysmem, objp);
1077         KASSERT(error == 0 && !sysmem && *objp != NULL,
1078             ("%s: invalid devmem segment %d", __func__, dsc->segid));
1079
1080         vcpu_unlock_one(dsc->sc, lastcpu);
1081
1082         if (seglen >= last) {
1083                 vm_object_reference(*objp);
1084                 return (0);
1085         } else {
1086                 return (EINVAL);
1087         }
1088 }
1089
1090 static struct cdevsw devmemsw = {
1091         .d_name         = "devmem",
1092         .d_version      = D_VERSION,
1093         .d_mmap_single  = devmem_mmap_single,
1094 };
1095
1096 static int
1097 devmem_create_cdev(const char *vmname, int segid, char *devname)
1098 {
1099         struct devmem_softc *dsc;
1100         struct vmmdev_softc *sc;
1101         struct cdev *cdev;
1102         int error;
1103
1104         error = make_dev_p(MAKEDEV_CHECKNAME, &cdev, &devmemsw, NULL,
1105             UID_ROOT, GID_WHEEL, 0600, "vmm.io/%s.%s", vmname, devname);
1106         if (error)
1107                 return (error);
1108
1109         dsc = malloc(sizeof(struct devmem_softc), M_VMMDEV, M_WAITOK | M_ZERO);
1110
1111         mtx_lock(&vmmdev_mtx);
1112         sc = vmmdev_lookup(vmname);
1113         KASSERT(sc != NULL, ("%s: vm %s softc not found", __func__, vmname));
1114         if (sc->cdev == NULL) {
1115                 /* virtual machine is being created or destroyed */
1116                 mtx_unlock(&vmmdev_mtx);
1117                 free(dsc, M_VMMDEV);
1118                 destroy_dev_sched_cb(cdev, NULL, 0);
1119                 return (ENODEV);
1120         }
1121
1122         dsc->segid = segid;
1123         dsc->name = devname;
1124         dsc->cdev = cdev;
1125         dsc->sc = sc;
1126         SLIST_INSERT_HEAD(&sc->devmem, dsc, link);
1127         mtx_unlock(&vmmdev_mtx);
1128
1129         /* The 'cdev' is ready for use after 'si_drv1' is initialized */
1130         cdev->si_drv1 = dsc;
1131         return (0);
1132 }
1133
1134 static void
1135 devmem_destroy(void *arg)
1136 {
1137         struct devmem_softc *dsc = arg;
1138
1139         KASSERT(dsc->cdev, ("%s: devmem cdev already destroyed", __func__));
1140         dsc->cdev = NULL;
1141         dsc->sc = NULL;
1142 }