sys/vm/vm_vmspace.c

   1 /*
   2  * (MPSAFE)
   3  *
   4  * Copyright (c) 2006 The DragonFly Project.  All rights reserved.
   5  *
   6  * This code is derived from software contributed to The DragonFly Project
   7  * by Matthew Dillon <dillon@backplane.com>
   8  *
   9  * Redistribution and use in source and binary forms, with or without
  10  * modification, are permitted provided that the following conditions
  11  * are met:
  12  *
  13  * 1. Redistributions of source code must retain the above copyright
  14  *    notice, this list of conditions and the following disclaimer.
  15  * 2. Redistributions in binary form must reproduce the above copyright
  16  *    notice, this list of conditions and the following disclaimer in
  17  *    the documentation and/or other materials provided with the
  18  *    distribution.
  19  * 3. Neither the name of The DragonFly Project nor the names of its
  20  *    contributors may be used to endorse or promote products derived
  21  *    from this software without specific, prior written permission.
  22  *
  23  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  24  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  25  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
  26  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
  27  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
  28  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
  29  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
  30  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
  31  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
  32  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
  33  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  34  * SUCH DAMAGE.
  35  */
  36
  37 #include <sys/param.h>
  38 #include <sys/kernel.h>
  39 #include <sys/systm.h>
  40 #include <sys/sysproto.h>
  41 #include <sys/kern_syscall.h>
  42 #include <sys/mman.h>
  43 #include <sys/thread.h>
  44 #include <sys/proc.h>
  45 #include <sys/malloc.h>
  46 #include <sys/sysctl.h>
  47 #include <sys/vkernel.h>
  48 #include <sys/vmspace.h>
  49
  50 #include <vm/vm_extern.h>
  51 #include <vm/pmap.h>
  52
  53 #include <machine/vmparam.h>
  54 #include <machine/vmm.h>
  55
  56 #include <sys/sysref2.h>
  57
  58 static struct vmspace_entry *vkernel_find_vmspace(struct vkernel_proc *vkp,
  59                                                   void *id);
  60 static void vmspace_entry_delete(struct vmspace_entry *ve,
  61                                  struct vkernel_proc *vkp);
  62
  63 static MALLOC_DEFINE(M_VKERNEL, "vkernel", "VKernel structures");
  64
  65 /*
  66  * vmspace_create (void *id, int type, void *data)
  67  *
  68  * Create a VMSPACE under the control of the caller with the specified id.
  69  * An id of NULL cannot be used.  The type and data fields must currently
  70  * be 0.
  71  *
  72  * The vmspace starts out completely empty.  Memory may be mapped into the
  73  * VMSPACE with vmspace_mmap() and MAP_VPAGETABLE section(s) controlled
  74  * with vmspace_mcontrol().
  75  *
  76  * No requirements.
  77  */
  78 int
  79 sys_vmspace_create(struct vmspace_create_args *uap)
  80 {
  81         struct vmspace_entry *ve;
  82         struct vkernel_proc *vkp;
  83         struct proc *p = curproc;
  84         int error;
  85
  86         if (vkernel_enable == 0)
  87                 return (EOPNOTSUPP);
  88
  89         /*
  90          * Create a virtual kernel side-structure for the process if one
  91          * does not exist.
  92          *
  93          * Implement a simple resolution for SMP races.
  94          */
  95         if ((vkp = p->p_vkernel) == NULL) {
  96                 vkp = kmalloc(sizeof(*vkp), M_VKERNEL, M_WAITOK|M_ZERO);
  97                 lwkt_gettoken(&p->p_token);
  98                 if (p->p_vkernel == NULL) {
  99                         vkp->refs = 1;
 100                         lwkt_token_init(&vkp->token, "vkernel");
 101                         RB_INIT(&vkp->root);
 102                         p->p_vkernel = vkp;
 103                 } else {
 104                         kfree(vkp, M_VKERNEL);
 105                         vkp = p->p_vkernel;
 106                 }
 107                 lwkt_reltoken(&p->p_token);
 108         }
 109
 110         if (curthread->td_vmm)
 111                 return 0;
 112
 113         /*
 114          * Create a new VMSPACE, disallow conflicting ids
 115          */
 116         ve = kmalloc(sizeof(struct vmspace_entry), M_VKERNEL, M_WAITOK|M_ZERO);
 117         ve->vmspace = vmspace_alloc(VM_MIN_USER_ADDRESS, VM_MAX_USER_ADDRESS);
 118         ve->id = uap->id;
 119         pmap_pinit2(vmspace_pmap(ve->vmspace));
 120
 121         lwkt_gettoken(&vkp->token);
 122         if (RB_INSERT(vmspace_rb_tree, &vkp->root, ve)) {
 123                 vmspace_rel(ve->vmspace);
 124                 ve->vmspace = NULL; /* safety */
 125                 kfree(ve, M_VKERNEL);
 126                 error = EEXIST;
 127         } else {
 128                 error = 0;
 129         }
 130         lwkt_reltoken(&vkp->token);
 131
 132         return (error);
 133 }
 134
 135 /*
 136  * Destroy a VMSPACE given its identifier.
 137  *
 138  * No requirements.
 139  */
 140 int
 141 sys_vmspace_destroy(struct vmspace_destroy_args *uap)
 142 {
 143         struct vkernel_proc *vkp;
 144         struct vmspace_entry *ve;
 145         int error;
 146
 147         if ((vkp = curproc->p_vkernel) == NULL) {
 148                 error = EINVAL;
 149                 goto done3;
 150         }
 151         lwkt_gettoken(&vkp->token);
 152         if ((ve = vkernel_find_vmspace(vkp, uap->id)) == NULL) {
 153                 error = ENOENT;
 154                 goto done2;
 155         }
 156         if (ve->refs) {
 157                 error = EBUSY;
 158                 goto done2;
 159         }
 160         vmspace_entry_delete(ve, vkp);
 161         error = 0;
 162 done2:
 163         lwkt_reltoken(&vkp->token);
 164 done3:
 165         return(error);
 166 }
 167
 168 /*
 169  * vmspace_ctl (void *id, int cmd, struct trapframe *tframe,
 170  *              struct vextframe *vframe);
 171  *
 172  * Transfer control to a VMSPACE.  Control is returned after the specified
 173  * number of microseconds or if a page fault, signal, trap, or system call
 174  * occurs.  The context is updated as appropriate.
 175  *
 176  * No requirements.
 177  */
 178 int
 179 sys_vmspace_ctl(struct vmspace_ctl_args *uap)
 180 {
 181         struct vkernel_proc *vkp;
 182         struct vkernel_lwp *vklp;
 183         struct vmspace_entry *ve = NULL;
 184         struct lwp *lp;
 185         struct proc *p;
 186         int framesz;
 187         int error;
 188
 189         lp = curthread->td_lwp;
 190         p = lp->lwp_proc;
 191
 192         if ((vkp = p->p_vkernel) == NULL)
 193                 return (EINVAL);
 194
 195         /*
 196          * ve only matters when VMM is not used.
 197          */
 198         if (curthread->td_vmm == NULL) {
 199                 lwkt_gettoken(&vkp->token);
 200                 if ((ve = vkernel_find_vmspace(vkp, uap->id)) == NULL) {
 201                         error = ENOENT;
 202                         goto done;
 203                 }
 204         }
 205
 206         switch(uap->cmd) {
 207         case VMSPACE_CTL_RUN:
 208                 /*
 209                  * Save the caller's register context, swap VM spaces, and
 210                  * install the passed register context.  Return with
 211                  * EJUSTRETURN so the syscall code doesn't adjust the context.
 212                  */
 213                 if (curthread->td_vmm == NULL)
 214                         atomic_add_int(&ve->refs, 1);
 215
 216                 framesz = sizeof(struct trapframe);
 217                 if ((vklp = lp->lwp_vkernel) == NULL) {
 218                         vklp = kmalloc(sizeof(*vklp), M_VKERNEL,
 219                                        M_WAITOK|M_ZERO);
 220                         lp->lwp_vkernel = vklp;
 221                 }
 222                 vklp->user_trapframe = uap->tframe;
 223                 vklp->user_vextframe = uap->vframe;
 224                 bcopy(uap->sysmsg_frame, &vklp->save_trapframe, framesz);
 225                 bcopy(&curthread->td_tls, &vklp->save_vextframe.vx_tls,
 226                       sizeof(vklp->save_vextframe.vx_tls));
 227                 error = copyin(uap->tframe, uap->sysmsg_frame, framesz);
 228                 if (error == 0) {
 229                         error = copyin(&uap->vframe->vx_tls,
 230                                        &curthread->td_tls,
 231                                        sizeof(struct savetls));
 232                 }
 233                 if (error == 0)
 234                         error = cpu_sanitize_frame(uap->sysmsg_frame);
 235                 if (error == 0)
 236                         error = cpu_sanitize_tls(&curthread->td_tls);
 237                 if (error) {
 238                         bcopy(&vklp->save_trapframe, uap->sysmsg_frame,
 239                               framesz);
 240                         bcopy(&vklp->save_vextframe.vx_tls, &curthread->td_tls,
 241                               sizeof(vklp->save_vextframe.vx_tls));
 242                         set_user_TLS();
 243                         if (curthread->td_vmm == NULL)
 244                                 atomic_subtract_int(&ve->refs, 1);
 245                 } else {
 246                         /* If it's a VMM thread just set the CR3. We also set the
 247                          * vklp->ve to a key to be able to distinguish when a
 248                          * vkernel user process runs and when not (when it's NULL)
 249                          */
 250                         if (curthread->td_vmm == NULL) {
 251                                 vklp->ve = ve;
 252                                 pmap_setlwpvm(lp, ve->vmspace);
 253                         } else {
 254                                 vklp->ve = uap->id;
 255                                 vmm_vm_set_guest_cr3((register_t)uap->id);
 256                         }
 257                         set_user_TLS();
 258                         set_vkernel_fp(uap->sysmsg_frame);
 259                         error = EJUSTRETURN;
 260                 }
 261                 break;
 262         default:
 263                 error = EOPNOTSUPP;
 264                 break;
 265         }
 266 done:
 267         if (curthread->td_vmm == NULL)
 268                 lwkt_reltoken(&vkp->token);
 269         return(error);
 270 }
 271
 272 /*
 273  * vmspace_mmap(id, addr, len, prot, flags, fd, offset)
 274  *
 275  * map memory within a VMSPACE.  This function is just like a normal mmap()
 276  * but operates on the vmspace's memory map.  Most callers use this to create
 277  * a MAP_VPAGETABLE mapping.
 278  *
 279  * No requirements.
 280  */
 281 int
 282 sys_vmspace_mmap(struct vmspace_mmap_args *uap)
 283 {
 284         struct vkernel_proc *vkp;
 285         struct vmspace_entry *ve;
 286         int error;
 287
 288         /*
 289          * We hold the vmspace token to serialize calls to vkernel_find_vmspace.
 290          */
 291         lwkt_gettoken(&vmspace_token);
 292         if ((vkp = curproc->p_vkernel) == NULL) {
 293                 error = EINVAL;
 294                 goto done3;
 295         }
 296
 297         /*
 298          * NOTE: kern_mmap() can block so we need to temporarily ref ve->refs.
 299          */
 300         lwkt_gettoken(&vkp->token);
 301         if ((ve = vkernel_find_vmspace(vkp, uap->id)) != NULL) {
 302                 atomic_add_int(&ve->refs, 1);
 303                 error = kern_mmap(ve->vmspace, uap->addr, uap->len,
 304                                   uap->prot, uap->flags,
 305                                   uap->fd, uap->offset, &uap->sysmsg_resultp);
 306                 atomic_subtract_int(&ve->refs, 1);
 307         } else {
 308                 error = ENOENT;
 309         }
 310         lwkt_reltoken(&vkp->token);
 311 done3:
 312         lwkt_reltoken(&vmspace_token);
 313         return (error);
 314 }
 315
 316 /*
 317  * vmspace_munmap(id, addr, len)
 318  *
 319  * unmap memory within a VMSPACE.
 320  *
 321  * No requirements.
 322  */
 323 int
 324 sys_vmspace_munmap(struct vmspace_munmap_args *uap)
 325 {
 326         struct vkernel_proc *vkp;
 327         struct vmspace_entry *ve;
 328         vm_offset_t addr;
 329         vm_offset_t tmpaddr;
 330         vm_size_t size, pageoff;
 331         vm_map_t map;
 332         int error;
 333
 334         if ((vkp = curproc->p_vkernel) == NULL) {
 335                 error = EINVAL;
 336                 goto done3;
 337         }
 338         lwkt_gettoken(&vkp->token);
 339         if ((ve = vkernel_find_vmspace(vkp, uap->id)) == NULL) {
 340                 error = ENOENT;
 341                 goto done2;
 342         }
 343
 344         /*
 345          * NOTE: kern_munmap() can block so we need to temporarily
 346          *       ref ve->refs.
 347          */
 348         atomic_add_int(&ve->refs, 1);
 349
 350         /*
 351          * Copied from sys_munmap()
 352          */
 353         addr = (vm_offset_t)uap->addr;
 354         size = uap->len;
 355
 356         pageoff = (addr & PAGE_MASK);
 357         addr -= pageoff;
 358         size += pageoff;
 359         size = (vm_size_t)round_page(size);
 360         if (size < uap->len) {          /* wrap */
 361                 error = EINVAL;
 362                 goto done1;
 363         }
 364         tmpaddr = addr + size;          /* workaround gcc4 opt */
 365         if (tmpaddr < addr) {           /* wrap */
 366                 error = EINVAL;
 367                 goto done1;
 368         }
 369         if (size == 0) {
 370                 error = 0;
 371                 goto done1;
 372         }
 373
 374         if (VM_MAX_USER_ADDRESS > 0 && tmpaddr > VM_MAX_USER_ADDRESS) {
 375                 error = EINVAL;
 376                 goto done1;
 377         }
 378         if (VM_MIN_USER_ADDRESS > 0 && addr < VM_MIN_USER_ADDRESS) {
 379                 error = EINVAL;
 380                 goto done1;
 381         }
 382         map = &ve->vmspace->vm_map;
 383         if (!vm_map_check_protection(map, addr, tmpaddr, VM_PROT_NONE, FALSE)) {
 384                 error = EINVAL;
 385                 goto done1;
 386         }
 387         vm_map_remove(map, addr, addr + size);
 388         error = 0;
 389 done1:
 390         atomic_subtract_int(&ve->refs, 1);
 391 done2:
 392         lwkt_reltoken(&vkp->token);
 393 done3:
 394         return (error);
 395 }
 396
 397 /*
 398  * vmspace_pread(id, buf, nbyte, flags, offset)
 399  *
 400  * Read data from a vmspace.  The number of bytes read is returned or
 401  * -1 if an unrecoverable error occured.  If the number of bytes read is
 402  * less then the request size, a page fault occured in the VMSPACE which
 403  * the caller must resolve in order to proceed.
 404  *
 405  * (not implemented yet)
 406  * No requirements.
 407  */
 408 int
 409 sys_vmspace_pread(struct vmspace_pread_args *uap)
 410 {
 411         struct vkernel_proc *vkp;
 412         struct vmspace_entry *ve;
 413         int error;
 414
 415         if ((vkp = curproc->p_vkernel) == NULL) {
 416                 error = EINVAL;
 417                 goto done3;
 418         }
 419         lwkt_gettoken(&vkp->token);
 420         if ((ve = vkernel_find_vmspace(vkp, uap->id)) == NULL) {
 421                 error = ENOENT;
 422                 goto done2;
 423         }
 424         error = EINVAL;
 425 done2:
 426         lwkt_reltoken(&vkp->token);
 427 done3:
 428         return (error);
 429 }
 430
 431 /*
 432  * vmspace_pwrite(id, buf, nbyte, flags, offset)
 433  *
 434  * Write data to a vmspace.  The number of bytes written is returned or
 435  * -1 if an unrecoverable error occured.  If the number of bytes written is
 436  * less then the request size, a page fault occured in the VMSPACE which
 437  * the caller must resolve in order to proceed.
 438  *
 439  * (not implemented yet)
 440  * No requirements.
 441  */
 442 int
 443 sys_vmspace_pwrite(struct vmspace_pwrite_args *uap)
 444 {
 445         struct vkernel_proc *vkp;
 446         struct vmspace_entry *ve;
 447         int error;
 448
 449         if ((vkp = curproc->p_vkernel) == NULL) {
 450                 error = EINVAL;
 451                 goto done3;
 452         }
 453         lwkt_gettoken(&vkp->token);
 454         if ((ve = vkernel_find_vmspace(vkp, uap->id)) == NULL) {
 455                 error = ENOENT;
 456                 goto done2;
 457         }
 458         error = EINVAL;
 459 done2:
 460         lwkt_reltoken(&vkp->token);
 461 done3:
 462         return (error);
 463 }
 464
 465 /*
 466  * vmspace_mcontrol(id, addr, len, behav, value)
 467  *
 468  * madvise/mcontrol support for a vmspace.
 469  *
 470  * No requirements.
 471  */
 472 int
 473 sys_vmspace_mcontrol(struct vmspace_mcontrol_args *uap)
 474 {
 475         struct vkernel_proc *vkp;
 476         struct vmspace_entry *ve;
 477         vm_offset_t start, end;
 478         vm_offset_t tmpaddr = (vm_offset_t)uap->addr + uap->len;
 479         int error;
 480
 481         if ((vkp = curproc->p_vkernel) == NULL) {
 482                 error = EINVAL;
 483                 goto done3;
 484         }
 485         lwkt_gettoken(&vkp->token);
 486         if ((ve = vkernel_find_vmspace(vkp, uap->id)) == NULL) {
 487                 error = ENOENT;
 488                 goto done2;
 489         }
 490
 491         /*
 492          * NOTE: kern_madvise() can block so we need to temporarily
 493          *       ref ve->refs.
 494          */
 495         atomic_add_int(&ve->refs, 1);
 496
 497         /*
 498          * This code is basically copied from sys_mcontrol()
 499          */
 500         if (uap->behav < 0 || uap->behav > MADV_CONTROL_END) {
 501                 error = EINVAL;
 502                 goto done1;
 503         }
 504
 505         if (tmpaddr < (vm_offset_t)uap->addr) {
 506                 error = EINVAL;
 507                 goto done1;
 508         }
 509         if (VM_MAX_USER_ADDRESS > 0 && tmpaddr > VM_MAX_USER_ADDRESS) {
 510                 error = EINVAL;
 511                 goto done1;
 512         }
 513         if (VM_MIN_USER_ADDRESS > 0 && uap->addr < VM_MIN_USER_ADDRESS) {
 514                 error = EINVAL;
 515                 goto done1;
 516         }
 517
 518         start = trunc_page((vm_offset_t) uap->addr);
 519         end = round_page(tmpaddr);
 520
 521         error = vm_map_madvise(&ve->vmspace->vm_map, start, end,
 522                                 uap->behav, uap->value);
 523 done1:
 524         atomic_subtract_int(&ve->refs, 1);
 525 done2:
 526         lwkt_reltoken(&vkp->token);
 527 done3:
 528         return (error);
 529 }
 530
 531 /*
 532  * Red black tree functions
 533  */
 534 static int rb_vmspace_compare(struct vmspace_entry *, struct vmspace_entry *);
 535 RB_GENERATE(vmspace_rb_tree, vmspace_entry, rb_entry, rb_vmspace_compare);
 536
 537 /*
 538  * a->start is address, and the only field has to be initialized.
 539  * The caller must hold vkp->token.
 540  *
 541  * The caller must hold vkp->token.
 542  */
 543 static int
 544 rb_vmspace_compare(struct vmspace_entry *a, struct vmspace_entry *b)
 545 {
 546         if ((char *)a->id < (char *)b->id)
 547                 return(-1);
 548         else if ((char *)a->id > (char *)b->id)
 549                 return(1);
 550         return(0);
 551 }
 552
 553 /*
 554  * The caller must hold vkp->token.
 555  */
 556 static
 557 int
 558 rb_vmspace_delete(struct vmspace_entry *ve, void *data)
 559 {
 560         struct vkernel_proc *vkp = data;
 561
 562         KKASSERT(ve->refs == 0);
 563         vmspace_entry_delete(ve, vkp);
 564         return(0);
 565 }
 566
 567 /*
 568  * Remove a vmspace_entry from the RB tree and destroy it.  We have to clean
 569  * up the pmap, the vm_map, then destroy the vmspace.
 570  *
 571  * This function must remove the ve immediately before it might potentially
 572  * block.
 573  *
 574  * The caller must hold vkp->token.
 575  */
 576 static
 577 void
 578 vmspace_entry_delete(struct vmspace_entry *ve, struct vkernel_proc *vkp)
 579 {
 580         RB_REMOVE(vmspace_rb_tree, &vkp->root, ve);
 581
 582         pmap_remove_pages(vmspace_pmap(ve->vmspace),
 583                           VM_MIN_USER_ADDRESS, VM_MAX_USER_ADDRESS);
 584         vm_map_remove(&ve->vmspace->vm_map,
 585                       VM_MIN_USER_ADDRESS, VM_MAX_USER_ADDRESS);
 586         vmspace_rel(ve->vmspace);
 587         ve->vmspace = NULL; /* safety */
 588         kfree(ve, M_VKERNEL);
 589 }
 590
 591 /*
 592  * Locate the ve for (id), return the ve or NULL.  If found this function
 593  * will bump ve->refs which prevents the ve from being immediately destroyed
 594  * (but it can still be removed).
 595  *
 596  * The caller must hold vkp->token.
 597  */
 598 static
 599 struct vmspace_entry *
 600 vkernel_find_vmspace(struct vkernel_proc *vkp, void *id)
 601 {
 602         struct vmspace_entry *ve;
 603         struct vmspace_entry key;
 604
 605         key.id = id;
 606         ve = RB_FIND(vmspace_rb_tree, &vkp->root, &key);
 607         return (ve);
 608 }
 609
 610 /*
 611  * Manage vkernel refs, used by the kernel when fork()ing or exit()ing
 612  * a vkernel process.
 613  *
 614  * No requirements.
 615  */
 616 void
 617 vkernel_inherit(struct proc *p1, struct proc *p2)
 618 {
 619         struct vkernel_proc *vkp;
 620
 621         vkp = p1->p_vkernel;
 622         KKASSERT(vkp->refs > 0);
 623         atomic_add_int(&vkp->refs, 1);
 624         p2->p_vkernel = vkp;
 625 }
 626
 627 /*
 628  * No requirements.
 629  */
 630 void
 631 vkernel_exit(struct proc *p)
 632 {
 633         struct vkernel_proc *vkp;
 634         struct lwp *lp;
 635
 636         vkp = p->p_vkernel;
 637
 638         /*
 639          * Restore the original VM context if we are killed while running
 640          * a different one.
 641          *
 642          * This isn't supposed to happen.  What is supposed to happen is
 643          * that the process should enter vkernel_trap() before the handling
 644          * the signal.
 645          */
 646         RB_FOREACH(lp, lwp_rb_tree, &p->p_lwp_tree) {
 647                 vkernel_lwp_exit(lp);
 648         }
 649
 650         /*
 651          * Dereference the common area
 652          */
 653         p->p_vkernel = NULL;
 654         KKASSERT(vkp->refs > 0);
 655
 656         if (atomic_fetchadd_int(&vkp->refs, -1) == 1) {
 657                 lwkt_gettoken(&vkp->token);
 658                 RB_SCAN(vmspace_rb_tree, &vkp->root, NULL,
 659                         rb_vmspace_delete, vkp);
 660                 lwkt_reltoken(&vkp->token);
 661                 kfree(vkp, M_VKERNEL);
 662         }
 663 }
 664
 665 /*
 666  * No requirements.
 667  */
 668 void
 669 vkernel_lwp_exit(struct lwp *lp)
 670 {
 671         struct vkernel_lwp *vklp;
 672         struct vmspace_entry *ve;
 673
 674         if ((vklp = lp->lwp_vkernel) != NULL) {
 675                 if ((ve = vklp->ve) != NULL) {
 676                         kprintf("Warning, pid %d killed with "
 677                                 "active VC!\n", lp->lwp_proc->p_pid);
 678                         pmap_setlwpvm(lp, lp->lwp_proc->p_vmspace);
 679                         vklp->ve = NULL;
 680                         KKASSERT(ve->refs > 0);
 681                         atomic_subtract_int(&ve->refs, 1);
 682                 }
 683                 lp->lwp_vkernel = NULL;
 684                 kfree(vklp, M_VKERNEL);
 685         }
 686 }
 687
 688 /*
 689  * A VM space under virtual kernel control trapped out or made a system call
 690  * or otherwise needs to return control to the virtual kernel context.
 691  *
 692  * No requirements.
 693  */
 694 void
 695 vkernel_trap(struct lwp *lp, struct trapframe *frame)
 696 {
 697         struct proc *p = lp->lwp_proc;
 698         struct vmspace_entry *ve;
 699         struct vkernel_lwp *vklp;
 700         int error;
 701
 702         /*
 703          * Which vmspace entry was running?
 704          */
 705         vklp = lp->lwp_vkernel;
 706         KKASSERT(vklp);
 707
 708         /* If it's a VMM thread just set the vkernel CR3 back */
 709         if (curthread->td_vmm == NULL) {
 710                 ve = vklp->ve;
 711                 KKASSERT(ve != NULL);
 712
 713                 /*
 714                  * Switch the LWP vmspace back to the virtual kernel's VM space.
 715                  */
 716                 vklp->ve = NULL;
 717                 pmap_setlwpvm(lp, p->p_vmspace);
 718                 KKASSERT(ve->refs > 0);
 719                 atomic_subtract_int(&ve->refs, 1);
 720                 /* ve is invalid once we kill our ref */
 721         } else {
 722                 vklp->ve = NULL;
 723                 vmm_vm_set_guest_cr3(p->p_vkernel->vkernel_cr3);
 724         }
 725         /*
 726          * Copy the emulated process frame to the virtual kernel process.
 727          * The emulated process cannot change TLS descriptors so don't
 728          * bother saving them, we already have a copy.
 729          *
 730          * Restore the virtual kernel's saved context so the virtual kernel
 731          * process can resume.
 732          */
 733         error = copyout(frame, vklp->user_trapframe, sizeof(*frame));
 734         bcopy(&vklp->save_trapframe, frame, sizeof(*frame));
 735         bcopy(&vklp->save_vextframe.vx_tls, &curthread->td_tls,
 736               sizeof(vklp->save_vextframe.vx_tls));
 737         set_user_TLS();
 738         cpu_vkernel_trap(frame, error);
 739 }