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