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/pmap.h>
  50 #include <machine/vmparam.h>
  51
  52 #include <machine/vmm.h>
  53 #include "vmm_lapic.h"
  54 #include "vmm_stat.h"
  55 #include "vmm_mem.h"
  56 #include "io/ppt.h"
  57 #include <machine/vmm_dev.h>
  58
  59 struct vmmdev_softc {
  60         struct vm       *vm;            /* vm instance cookie */
  61         struct cdev     *cdev;
  62         SLIST_ENTRY(vmmdev_softc) link;
  63         int             flags;
  64 };
  65 #define VSC_LINKED              0x01
  66
  67 static SLIST_HEAD(, vmmdev_softc) head;
  68
  69 static struct mtx vmmdev_mtx;
  70
  71 static MALLOC_DEFINE(M_VMMDEV, "vmmdev", "vmmdev");
  72
  73 SYSCTL_DECL(_hw_vmm);
  74
  75 static struct vmmdev_softc *
  76 vmmdev_lookup(const char *name)
  77 {
  78         struct vmmdev_softc *sc;
  79
  80 #ifdef notyet   /* XXX kernel is not compiled with invariants */
  81         mtx_assert(&vmmdev_mtx, MA_OWNED);
  82 #endif
  83
  84         SLIST_FOREACH(sc, &head, link) {
  85                 if (strcmp(name, vm_name(sc->vm)) == 0)
  86                         break;
  87         }
  88
  89         return (sc);
  90 }
  91
  92 static struct vmmdev_softc *
  93 vmmdev_lookup2(struct cdev *cdev)
  94 {
  95
  96         return (cdev->si_drv1);
  97 }
  98
  99 static int
 100 vmmdev_rw(struct cdev *cdev, struct uio *uio, int flags)
 101 {
 102         int error, off, c, prot;
 103         vm_paddr_t gpa;
 104         void *hpa, *cookie;
 105         struct vmmdev_softc *sc;
 106
 107         static char zerobuf[PAGE_SIZE];
 108
 109         error = 0;
 110         sc = vmmdev_lookup2(cdev);
 111         if (sc == NULL)
 112                 error = ENXIO;
 113
 114         prot = (uio->uio_rw == UIO_WRITE ? VM_PROT_WRITE : VM_PROT_READ);
 115         while (uio->uio_resid > 0 && error == 0) {
 116                 gpa = uio->uio_offset;
 117                 off = gpa & PAGE_MASK;
 118                 c = min(uio->uio_resid, PAGE_SIZE - off);
 119
 120                 /*
 121                  * The VM has a hole in its physical memory map. If we want to
 122                  * use 'dd' to inspect memory beyond the hole we need to
 123                  * provide bogus data for memory that lies in the hole.
 124                  *
 125                  * Since this device does not support lseek(2), dd(1) will
 126                  * read(2) blocks of data to simulate the lseek(2).
 127                  */
 128                 hpa = vm_gpa_hold(sc->vm, gpa, c, prot, &cookie);
 129                 if (hpa == NULL) {
 130                         if (uio->uio_rw == UIO_READ)
 131                                 error = uiomove(zerobuf, c, uio);
 132                         else
 133                                 error = EFAULT;
 134                 } else {
 135                         error = uiomove(hpa, c, uio);
 136                         vm_gpa_release(cookie);
 137                 }
 138         }
 139         return (error);
 140 }
 141
 142 static int
 143 vmmdev_ioctl(struct cdev *cdev, u_long cmd, caddr_t data, int fflag,
 144              struct thread *td)
 145 {
 146         int error, vcpu, state_changed;
 147         struct vmmdev_softc *sc;
 148         struct vm_memory_segment *seg;
 149         struct vm_register *vmreg;
 150         struct vm_seg_desc* vmsegdesc;
 151         struct vm_run *vmrun;
 152         struct vm_event *vmevent;
 153         struct vm_lapic_irq *vmirq;
 154         struct vm_capability *vmcap;
 155         struct vm_pptdev *pptdev;
 156         struct vm_pptdev_mmio *pptmmio;
 157         struct vm_pptdev_msi *pptmsi;
 158         struct vm_pptdev_msix *pptmsix;
 159         struct vm_nmi *vmnmi;
 160         struct vm_stats *vmstats;
 161         struct vm_stat_desc *statdesc;
 162         struct vm_x2apic *x2apic;
 163         struct vm_gpa_pte *gpapte;
 164
 165         sc = vmmdev_lookup2(cdev);
 166         if (sc == NULL)
 167                 return (ENXIO);
 168
 169         vcpu = -1;
 170         state_changed = 0;
 171
 172         /*
 173          * Some VMM ioctls can operate only on vcpus that are not running.
 174          */
 175         switch (cmd) {
 176         case VM_RUN:
 177         case VM_GET_REGISTER:
 178         case VM_SET_REGISTER:
 179         case VM_GET_SEGMENT_DESCRIPTOR:
 180         case VM_SET_SEGMENT_DESCRIPTOR:
 181         case VM_INJECT_EVENT:
 182         case VM_GET_CAPABILITY:
 183         case VM_SET_CAPABILITY:
 184         case VM_PPTDEV_MSI:
 185         case VM_PPTDEV_MSIX:
 186         case VM_SET_X2APIC_STATE:
 187                 /*
 188                  * XXX fragile, handle with care
 189                  * Assumes that the first field of the ioctl data is the vcpu.
 190                  */
 191                 vcpu = *(int *)data;
 192                 if (vcpu < 0 || vcpu >= VM_MAXCPU) {
 193                         error = EINVAL;
 194                         goto done;
 195                 }
 196
 197                 error = vcpu_set_state(sc->vm, vcpu, VCPU_FROZEN);
 198                 if (error)
 199                         goto done;
 200
 201                 state_changed = 1;
 202                 break;
 203
 204         case VM_MAP_PPTDEV_MMIO:
 205         case VM_BIND_PPTDEV:
 206         case VM_UNBIND_PPTDEV:
 207         case VM_MAP_MEMORY:
 208                 /*
 209                  * ioctls that operate on the entire virtual machine must
 210                  * prevent all vcpus from running.
 211                  */
 212                 error = 0;
 213                 for (vcpu = 0; vcpu < VM_MAXCPU; vcpu++) {
 214                         error = vcpu_set_state(sc->vm, vcpu, VCPU_FROZEN);
 215                         if (error)
 216                                 break;
 217                 }
 218
 219                 if (error) {
 220                         while (--vcpu >= 0)
 221                                 vcpu_set_state(sc->vm, vcpu, VCPU_IDLE);
 222                         goto done;
 223                 }
 224
 225                 state_changed = 2;
 226                 break;
 227
 228         default:
 229                 break;
 230         }
 231
 232         switch(cmd) {
 233         case VM_RUN:
 234                 vmrun = (struct vm_run *)data;
 235                 error = vm_run(sc->vm, vmrun);
 236                 break;
 237         case VM_STAT_DESC: {
 238                 statdesc = (struct vm_stat_desc *)data;
 239                 error = vmm_stat_desc_copy(statdesc->index,
 240                                         statdesc->desc, sizeof(statdesc->desc));
 241                 break;
 242         }
 243         case VM_STATS: {
 244                 CTASSERT(MAX_VM_STATS >= MAX_VMM_STAT_ELEMS);
 245                 vmstats = (struct vm_stats *)data;
 246                 getmicrotime(&vmstats->tv);
 247                 error = vmm_stat_copy(sc->vm, vmstats->cpuid,
 248                                       &vmstats->num_entries, vmstats->statbuf);
 249                 break;
 250         }
 251         case VM_PPTDEV_MSI:
 252                 pptmsi = (struct vm_pptdev_msi *)data;
 253                 error = ppt_setup_msi(sc->vm, pptmsi->vcpu,
 254                                       pptmsi->bus, pptmsi->slot, pptmsi->func,
 255                                       pptmsi->destcpu, pptmsi->vector,
 256                                       pptmsi->numvec);
 257                 break;
 258         case VM_PPTDEV_MSIX:
 259                 pptmsix = (struct vm_pptdev_msix *)data;
 260                 error = ppt_setup_msix(sc->vm, pptmsix->vcpu,
 261                                        pptmsix->bus, pptmsix->slot,
 262                                        pptmsix->func, pptmsix->idx,
 263                                        pptmsix->msg, pptmsix->vector_control,
 264                                        pptmsix->addr);
 265                 break;
 266         case VM_MAP_PPTDEV_MMIO:
 267                 pptmmio = (struct vm_pptdev_mmio *)data;
 268                 error = ppt_map_mmio(sc->vm, pptmmio->bus, pptmmio->slot,
 269                                      pptmmio->func, pptmmio->gpa, pptmmio->len,
 270                                      pptmmio->hpa);
 271                 break;
 272         case VM_BIND_PPTDEV:
 273                 pptdev = (struct vm_pptdev *)data;
 274                 error = vm_assign_pptdev(sc->vm, pptdev->bus, pptdev->slot,
 275                                          pptdev->func);
 276                 break;
 277         case VM_UNBIND_PPTDEV:
 278                 pptdev = (struct vm_pptdev *)data;
 279                 error = vm_unassign_pptdev(sc->vm, pptdev->bus, pptdev->slot,
 280                                            pptdev->func);
 281                 break;
 282         case VM_INJECT_EVENT:
 283                 vmevent = (struct vm_event *)data;
 284                 error = vm_inject_event(sc->vm, vmevent->cpuid, vmevent->type,
 285                                         vmevent->vector,
 286                                         vmevent->error_code,
 287                                         vmevent->error_code_valid);
 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_set_intr(sc->vm, vmirq->cpuid, vmirq->vector);
 296                 break;
 297         case VM_MAP_MEMORY:
 298                 seg = (struct vm_memory_segment *)data;
 299                 error = vm_malloc(sc->vm, seg->gpa, seg->len);
 300                 break;
 301         case VM_GET_MEMORY_SEG:
 302                 seg = (struct vm_memory_segment *)data;
 303                 seg->len = 0;
 304                 (void)vm_gpabase2memseg(sc->vm, seg->gpa, seg);
 305                 error = 0;
 306                 break;
 307         case VM_GET_REGISTER:
 308                 vmreg = (struct vm_register *)data;
 309                 error = vm_get_register(sc->vm, vmreg->cpuid, vmreg->regnum,
 310                                         &vmreg->regval);
 311                 break;
 312         case VM_SET_REGISTER:
 313                 vmreg = (struct vm_register *)data;
 314                 error = vm_set_register(sc->vm, vmreg->cpuid, vmreg->regnum,
 315                                         vmreg->regval);
 316                 break;
 317         case VM_SET_SEGMENT_DESCRIPTOR:
 318                 vmsegdesc = (struct vm_seg_desc *)data;
 319                 error = vm_set_seg_desc(sc->vm, vmsegdesc->cpuid,
 320                                         vmsegdesc->regnum,
 321                                         &vmsegdesc->desc);
 322                 break;
 323         case VM_GET_SEGMENT_DESCRIPTOR:
 324                 vmsegdesc = (struct vm_seg_desc *)data;
 325                 error = vm_get_seg_desc(sc->vm, vmsegdesc->cpuid,
 326                                         vmsegdesc->regnum,
 327                                         &vmsegdesc->desc);
 328                 break;
 329         case VM_GET_CAPABILITY:
 330                 vmcap = (struct vm_capability *)data;
 331                 error = vm_get_capability(sc->vm, vmcap->cpuid,
 332                                           vmcap->captype,
 333                                           &vmcap->capval);
 334                 break;
 335         case VM_SET_CAPABILITY:
 336                 vmcap = (struct vm_capability *)data;
 337                 error = vm_set_capability(sc->vm, vmcap->cpuid,
 338                                           vmcap->captype,
 339                                           vmcap->capval);
 340                 break;
 341         case VM_SET_X2APIC_STATE:
 342                 x2apic = (struct vm_x2apic *)data;
 343                 error = vm_set_x2apic_state(sc->vm,
 344                                             x2apic->cpuid, x2apic->state);
 345                 break;
 346         case VM_GET_X2APIC_STATE:
 347                 x2apic = (struct vm_x2apic *)data;
 348                 error = vm_get_x2apic_state(sc->vm,
 349                                             x2apic->cpuid, &x2apic->state);
 350                 break;
 351         case VM_GET_GPA_PMAP:
 352                 gpapte = (struct vm_gpa_pte *)data;
 353                 pmap_get_mapping(vmspace_pmap(vm_get_vmspace(sc->vm)),
 354                                  gpapte->gpa, gpapte->pte, &gpapte->ptenum);
 355                 error = 0;
 356                 break;
 357         default:
 358                 error = ENOTTY;
 359                 break;
 360         }
 361
 362         if (state_changed == 1) {
 363                 vcpu_set_state(sc->vm, vcpu, VCPU_IDLE);
 364         } else if (state_changed == 2) {
 365                 for (vcpu = 0; vcpu < VM_MAXCPU; vcpu++)
 366                         vcpu_set_state(sc->vm, vcpu, VCPU_IDLE);
 367         }
 368
 369 done:
 370         /* Make sure that no handler returns a bogus value like ERESTART */
 371         KASSERT(error >= 0, ("vmmdev_ioctl: invalid error return %d", error));
 372         return (error);
 373 }
 374
 375 static int
 376 vmmdev_mmap_single(struct cdev *cdev, vm_ooffset_t *offset,
 377                    vm_size_t size, struct vm_object **object, int nprot)
 378 {
 379         int error;
 380         struct vmmdev_softc *sc;
 381
 382         sc = vmmdev_lookup2(cdev);
 383         if (sc != NULL && (nprot & PROT_EXEC) == 0)
 384                 error = vm_get_memobj(sc->vm, *offset, size, offset, object);
 385         else
 386                 error = EINVAL;
 387
 388         return (error);
 389 }
 390
 391 static void
 392 vmmdev_destroy(void *arg)
 393 {
 394
 395         struct vmmdev_softc *sc = arg;
 396
 397         if (sc->cdev != NULL)
 398                 destroy_dev(sc->cdev);
 399
 400         if (sc->vm != NULL)
 401                 vm_destroy(sc->vm);
 402
 403         if ((sc->flags & VSC_LINKED) != 0) {
 404                 mtx_lock(&vmmdev_mtx);
 405                 SLIST_REMOVE(&head, sc, vmmdev_softc, link);
 406                 mtx_unlock(&vmmdev_mtx);
 407         }
 408
 409         free(sc, M_VMMDEV);
 410 }
 411
 412 static int
 413 sysctl_vmm_destroy(SYSCTL_HANDLER_ARGS)
 414 {
 415         int error;
 416         char buf[VM_MAX_NAMELEN];
 417         struct vmmdev_softc *sc;
 418         struct cdev *cdev;
 419
 420         strlcpy(buf, "beavis", sizeof(buf));
 421         error = sysctl_handle_string(oidp, buf, sizeof(buf), req);
 422         if (error != 0 || req->newptr == NULL)
 423                 return (error);
 424
 425         mtx_lock(&vmmdev_mtx);
 426         sc = vmmdev_lookup(buf);
 427         if (sc == NULL || sc->cdev == NULL) {
 428                 mtx_unlock(&vmmdev_mtx);
 429                 return (EINVAL);
 430         }
 431
 432         /*
 433          * The 'cdev' will be destroyed asynchronously when 'si_threadcount'
 434          * goes down to 0 so we should not do it again in the callback.
 435          */
 436         cdev = sc->cdev;
 437         sc->cdev = NULL;
 438         mtx_unlock(&vmmdev_mtx);
 439
 440         /*
 441          * Schedule the 'cdev' to be destroyed:
 442          *
 443          * - any new operations on this 'cdev' will return an error (ENXIO).
 444          *
 445          * - when the 'si_threadcount' dwindles down to zero the 'cdev' will
 446          *   be destroyed and the callback will be invoked in a taskqueue
 447          *   context.
 448          */
 449         destroy_dev_sched_cb(cdev, vmmdev_destroy, sc);
 450
 451         return (0);
 452 }
 453 SYSCTL_PROC(_hw_vmm, OID_AUTO, destroy, CTLTYPE_STRING | CTLFLAG_RW,
 454             NULL, 0, sysctl_vmm_destroy, "A", NULL);
 455
 456 static struct cdevsw vmmdevsw = {
 457         .d_name         = "vmmdev",
 458         .d_version      = D_VERSION,
 459         .d_ioctl        = vmmdev_ioctl,
 460         .d_mmap_single  = vmmdev_mmap_single,
 461         .d_read         = vmmdev_rw,
 462         .d_write        = vmmdev_rw,
 463 };
 464
 465 static int
 466 sysctl_vmm_create(SYSCTL_HANDLER_ARGS)
 467 {
 468         int error;
 469         struct vm *vm;
 470         struct cdev *cdev;
 471         struct vmmdev_softc *sc, *sc2;
 472         char buf[VM_MAX_NAMELEN];
 473
 474         strlcpy(buf, "beavis", sizeof(buf));
 475         error = sysctl_handle_string(oidp, buf, sizeof(buf), req);
 476         if (error != 0 || req->newptr == NULL)
 477                 return (error);
 478
 479         mtx_lock(&vmmdev_mtx);
 480         sc = vmmdev_lookup(buf);
 481         mtx_unlock(&vmmdev_mtx);
 482         if (sc != NULL)
 483                 return (EEXIST);
 484
 485         error = vm_create(buf, &vm);
 486         if (error != 0)
 487                 return (error);
 488
 489         sc = malloc(sizeof(struct vmmdev_softc), M_VMMDEV, M_WAITOK | M_ZERO);
 490         sc->vm = vm;
 491
 492         /*
 493          * Lookup the name again just in case somebody sneaked in when we
 494          * dropped the lock.
 495          */
 496         mtx_lock(&vmmdev_mtx);
 497         sc2 = vmmdev_lookup(buf);
 498         if (sc2 == NULL) {
 499                 SLIST_INSERT_HEAD(&head, sc, link);
 500                 sc->flags |= VSC_LINKED;
 501         }
 502         mtx_unlock(&vmmdev_mtx);
 503
 504         if (sc2 != NULL) {
 505                 vmmdev_destroy(sc);
 506                 return (EEXIST);
 507         }
 508
 509         error = make_dev_p(MAKEDEV_CHECKNAME, &cdev, &vmmdevsw, NULL,
 510                            UID_ROOT, GID_WHEEL, 0600, "vmm/%s", buf);
 511         if (error != 0) {
 512                 vmmdev_destroy(sc);
 513                 return (error);
 514         }
 515
 516         mtx_lock(&vmmdev_mtx);
 517         sc->cdev = cdev;
 518         sc->cdev->si_drv1 = sc;
 519         mtx_unlock(&vmmdev_mtx);
 520
 521         return (0);
 522 }
 523 SYSCTL_PROC(_hw_vmm, OID_AUTO, create, CTLTYPE_STRING | CTLFLAG_RW,
 524             NULL, 0, sysctl_vmm_create, "A", NULL);
 525
 526 void
 527 vmmdev_init(void)
 528 {
 529         mtx_init(&vmmdev_mtx, "vmm device mutex", NULL, MTX_DEF);
 530 }
 531
 532 int
 533 vmmdev_cleanup(void)
 534 {
 535         int error;
 536
 537         if (SLIST_EMPTY(&head))
 538                 error = 0;
 539         else
 540                 error = EBUSY;
 541
 542         return (error);
 543 }