sys/platform/vkernel/platform/init.c

   1 /*
   2  * Copyright (c) 2006 The DragonFly Project.  All rights reserved.
   3  *
   4  * This code is derived from software contributed to The DragonFly Project
   5  * by Matthew Dillon <dillon@backplane.com>
   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  *
  11  * 1. Redistributions of source code must retain the above copyright
  12  *    notice, this list of conditions and the following disclaimer.
  13  * 2. Redistributions in binary form must reproduce the above copyright
  14  *    notice, this list of conditions and the following disclaimer in
  15  *    the documentation and/or other materials provided with the
  16  *    distribution.
  17  * 3. Neither the name of The DragonFly Project nor the names of its
  18  *    contributors may be used to endorse or promote products derived
  19  *    from this software without specific, prior written permission.
  20  *
  21  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  22  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  23  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
  24  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
  25  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
  26  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
  27  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
  28  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
  29  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
  30  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
  31  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  32  * SUCH DAMAGE.
  33  *
  34  * $DragonFly: src/sys/platform/vkernel/platform/init.c,v 1.48 2008/01/29 19:54:59 dillon Exp $
  35  */
  36
  37 #include <sys/types.h>
  38 #include <sys/systm.h>
  39 #include <sys/kernel.h>
  40 #include <sys/stat.h>
  41 #include <sys/mman.h>
  42 #include <sys/cons.h>
  43 #include <sys/random.h>
  44 #include <sys/vkernel.h>
  45 #include <sys/tls.h>
  46 #include <sys/reboot.h>
  47 #include <sys/proc.h>
  48 #include <sys/msgbuf.h>
  49 #include <sys/vmspace.h>
  50 #include <sys/socket.h>
  51 #include <sys/sockio.h>
  52 #include <sys/sysctl.h>
  53 #include <vm/vm_page.h>
  54
  55 #include <machine/cpu.h>
  56 #include <machine/globaldata.h>
  57 #include <machine/tls.h>
  58 #include <machine/md_var.h>
  59 #include <machine/vmparam.h>
  60
  61 #include <net/if.h>
  62 #include <net/if_arp.h>
  63 #include <net/ethernet.h>
  64 #include <net/bridge/if_bridgevar.h>
  65 #include <netinet/in.h>
  66 #include <arpa/inet.h>
  67
  68 #include <stdio.h>
  69 #include <stdlib.h>
  70 #include <stdarg.h>
  71 #include <unistd.h>
  72 #include <fcntl.h>
  73 #include <string.h>
  74 #include <err.h>
  75 #include <errno.h>
  76 #include <assert.h>
  77
  78 vm_paddr_t phys_avail[16];
  79 vm_paddr_t Maxmem;
  80 vm_paddr_t Maxmem_bytes;
  81 int MemImageFd = -1;
  82 struct vkdisk_info DiskInfo[VKDISK_MAX];
  83 int DiskNum;
  84 struct vknetif_info NetifInfo[VKNETIF_MAX];
  85 int NetifNum;
  86 char *pid_file;
  87 vm_offset_t KvaStart;
  88 vm_offset_t KvaEnd;
  89 vm_offset_t KvaSize;
  90 vm_offset_t virtual_start;
  91 vm_offset_t virtual_end;
  92 vm_offset_t kernel_vm_end;
  93 vm_offset_t crashdumpmap;
  94 vm_offset_t clean_sva;
  95 vm_offset_t clean_eva;
  96 struct msgbuf *msgbufp;
  97 caddr_t ptvmmap;
  98 vpte_t  *KernelPTD;
  99 vpte_t  *KernelPTA;     /* Warning: Offset for direct VA translation */
 100 u_int cpu_feature;      /* XXX */
 101 u_int tsc_present;      /* XXX */
 102 int optcpus;            /* number of cpus - see mp_start() */
 103 int lwp_cpu_lock;       /* if/how to lock virtual CPUs to real CPUs */
 104 int real_ncpus;         /* number of real CPUs */
 105 int next_cpu;           /* next real CPU to lock a virtual CPU to */
 106
 107 struct privatespace *CPU_prvspace;
 108
 109 static struct trapframe proc0_tf;
 110 static void *proc0paddr;
 111
 112 static void init_sys_memory(char *imageFile);
 113 static void init_kern_memory(void);
 114 static void init_globaldata(void);
 115 static void init_vkernel(void);
 116 static void init_disk(char *diskExp[], int diskFileNum, enum vkdisk_type type);
 117 static void init_netif(char *netifExp[], int netifFileNum);
 118 static void writepid( void );
 119 static void cleanpid( void );
 120 static void usage(const char *ctl, ...);
 121
 122 static int save_ac;
 123 static char **save_av;
 124
 125 /*
 126  * Kernel startup for virtual kernels - standard main()
 127  */
 128 int
 129 main(int ac, char **av)
 130 {
 131         char *memImageFile = NULL;
 132         char *netifFile[VKNETIF_MAX];
 133         char *diskFile[VKDISK_MAX];
 134         char *cdFile[VKDISK_MAX];
 135         char *suffix;
 136         char *endp;
 137         size_t real_ncpus_size;
 138         int netifFileNum = 0;
 139         int diskFileNum = 0;
 140         int cdFileNum = 0;
 141         int bootOnDisk = -1;    /* set below to vcd (0) or vkd (1) */
 142         int c;
 143         int i;
 144         int n;
 145         int real_vkernel_enable;
 146         size_t real_vkernel_enable_size;
 147         int supports_sse;
 148         int supports_sse_size;
 149
 150         save_ac = ac;
 151         save_av = av;
 152
 153         /*
 154          * Process options
 155          */
 156         kernel_mem_readonly = 1;
 157 #ifdef SMP
 158         optcpus = 2;
 159 #endif
 160         lwp_cpu_lock = LCL_NONE;
 161
 162         real_vkernel_enable_size = sizeof(real_vkernel_enable);
 163         sysctlbyname("vm.vkernel_enable", &real_vkernel_enable, &real_vkernel_enable_size, NULL, 0);
 164
 165         if (real_vkernel_enable == 0)
 166                 errx(1,"vm.vkernel_enable is %d, must be set to 1 to execute a vkernel!", real_vkernel_enable);
 167
 168         real_ncpus_size = sizeof(real_ncpus);
 169         sysctlbyname("hw.ncpu", &real_ncpus, &real_ncpus_size, NULL, 0);
 170
 171         while ((c = getopt(ac, av, "c:svl:m:n:r:e:i:p:I:U")) != -1) {
 172                 switch(c) {
 173                 case 'e':
 174                         /*
 175                          * name=value:name=value:name=value...
 176                          */
 177                         n = strlen(optarg);
 178                         kern_envp = malloc(n + 2);
 179                         for (i = 0; i < n; ++i) {
 180                                 if (optarg[i] == ':')
 181                                         kern_envp[i] = 0;
 182                                 else
 183                                         kern_envp[i] = optarg[i];
 184                         }
 185                         kern_envp[i++] = 0;
 186                         kern_envp[i++] = 0;
 187                         break;
 188                 case 's':
 189                         boothowto |= RB_SINGLE;
 190                         break;
 191                 case 'v':
 192                         bootverbose = 1;
 193                         break;
 194                 case 'i':
 195                         memImageFile = optarg;
 196                         break;
 197                 case 'I':
 198                         if (netifFileNum < VKNETIF_MAX)
 199                                 netifFile[netifFileNum++] = strdup(optarg);
 200                         break;
 201                 case 'r':
 202                         if (bootOnDisk < 0)
 203                                 bootOnDisk = 1;
 204                         if (diskFileNum + cdFileNum < VKDISK_MAX)
 205                                 diskFile[diskFileNum++] = strdup(optarg);
 206                         break;
 207                 case 'c':
 208                         if (bootOnDisk < 0)
 209                                 bootOnDisk = 0;
 210                         if (diskFileNum + cdFileNum < VKDISK_MAX)
 211                                 cdFile[cdFileNum++] = strdup(optarg);
 212                         break;
 213                 case 'm':
 214                         Maxmem_bytes = strtoull(optarg, &suffix, 0);
 215                         if (suffix) {
 216                                 switch(*suffix) {
 217                                 case 'g':
 218                                 case 'G':
 219                                         Maxmem_bytes <<= 30;
 220                                         break;
 221                                 case 'm':
 222                                 case 'M':
 223                                         Maxmem_bytes <<= 20;
 224                                         break;
 225                                 case 'k':
 226                                 case 'K':
 227                                         Maxmem_bytes <<= 10;
 228                                         break;
 229                                 default:
 230                                         Maxmem_bytes = 0;
 231                                         usage("Bad maxmem option");
 232                                         /* NOT REACHED */
 233                                         break;
 234                                 }
 235                         }
 236                         break;
 237                 case 'l':
 238                         next_cpu = -1;
 239                         if (strncmp("map", optarg, 3) == 0) {
 240                                 lwp_cpu_lock = LCL_PER_CPU;
 241                                 if (optarg[3] == ',') {
 242                                         next_cpu = strtol(optarg+4, &endp, 0);
 243                                         if (*endp != '\0')
 244                                                 usage("Bad target CPU number at '%s'", endp);
 245                                 } else {
 246                                         next_cpu = 0;
 247                                 }
 248                                 if (next_cpu < 0 || next_cpu > real_ncpus - 1)
 249                                         usage("Bad target CPU, valid range is 0-%d", real_ncpus - 1);
 250                         } else if (strncmp("any", optarg, 3) == 0) {
 251                                 lwp_cpu_lock = LCL_NONE;
 252                         } else {
 253                                 lwp_cpu_lock = LCL_SINGLE_CPU;
 254                                 next_cpu = strtol(optarg, &endp, 0);
 255                                 if (*endp != '\0')
 256                                         usage("Bad target CPU number at '%s'", endp);
 257                                 if (next_cpu < 0 || next_cpu > real_ncpus - 1)
 258                                         usage("Bad target CPU, valid range is 0-%d", real_ncpus - 1);
 259                         }
 260                         break;
 261                 case 'n':
 262                         /*
 263                          * This value is set up by mp_start(), don't just
 264                          * set ncpus here.
 265                          */
 266 #ifdef SMP
 267                         optcpus = strtol(optarg, NULL, 0);
 268                         if (optcpus < 1 || optcpus > MAXCPU)
 269                                 usage("Bad ncpus, valid range is 1-%d", MAXCPU);
 270 #else
 271                         if (strtol(optarg, NULL, 0) != 1) {
 272                                 usage("You built a UP vkernel, only 1 cpu!");
 273                         }
 274 #endif
 275
 276                         break;
 277                 case 'p':
 278                         pid_file = optarg;
 279                         break;
 280                 case 'U':
 281                         kernel_mem_readonly = 0;
 282                         break;
 283                 }
 284         }
 285
 286         writepid();
 287         cpu_disable_intr();
 288         init_sys_memory(memImageFile);
 289         init_kern_memory();
 290         init_globaldata();
 291         init_vkernel();
 292         setrealcpu();
 293         init_kqueue();
 294
 295         supports_sse_size = sizeof(supports_sse_size);
 296         sysctlbyname("hw.instruction_sse", &supports_sse, &supports_sse_size,
 297                      NULL, 0);
 298         init_fpu(supports_sse);
 299
 300         /*
 301          * We boot from the first installed disk.
 302          */
 303         if (bootOnDisk == 1) {
 304                 init_disk(diskFile, diskFileNum, VKD_DISK);
 305                 init_disk(cdFile, cdFileNum, VKD_CD);
 306         } else {
 307                 init_disk(cdFile, cdFileNum, VKD_CD);
 308                 init_disk(diskFile, diskFileNum, VKD_DISK);
 309         }
 310         init_netif(netifFile, netifFileNum);
 311         init_exceptions();
 312         mi_startup();
 313         /* NOT REACHED */
 314         exit(1);
 315 }
 316
 317 /*
 318  * Initialize system memory.  This is the virtual kernel's 'RAM'.
 319  */
 320 static
 321 void
 322 init_sys_memory(char *imageFile)
 323 {
 324         struct stat st;
 325         int i;
 326         int fd;
 327
 328         /*
 329          * Figure out the system memory image size.  If an image file was
 330          * specified and -m was not specified, use the image file's size.
 331          */
 332
 333         if (imageFile && stat(imageFile, &st) == 0 && Maxmem_bytes == 0)
 334                 Maxmem_bytes = (vm_paddr_t)st.st_size;
 335         if ((imageFile == NULL || stat(imageFile, &st) < 0) &&
 336             Maxmem_bytes == 0) {
 337                 err(1, "Cannot create new memory file %s unless "
 338                        "system memory size is specified with -m",
 339                        imageFile);
 340                 /* NOT REACHED */
 341         }
 342
 343         /*
 344          * Maxmem must be known at this time
 345          */
 346         if (Maxmem_bytes < 32 * 1024 * 1024 || (Maxmem_bytes & SEG_MASK)) {
 347                 err(1, "Bad maxmem specification: 32MB minimum, "
 348                        "multiples of %dMB only",
 349                        SEG_SIZE / 1024 / 1024);
 350                 /* NOT REACHED */
 351         }
 352
 353         /*
 354          * Generate an image file name if necessary, then open/create the
 355          * file exclusively locked.  Do not allow multiple virtual kernels
 356          * to use the same image file.
 357          */
 358         if (imageFile == NULL) {
 359                 for (i = 0; i < 1000000; ++i) {
 360                         asprintf(&imageFile, "/var/vkernel/memimg.%06d", i);
 361                         fd = open(imageFile,
 362                                   O_RDWR|O_CREAT|O_EXLOCK|O_NONBLOCK, 0644);
 363                         if (fd < 0 && errno == EWOULDBLOCK) {
 364                                 free(imageFile);
 365                                 continue;
 366                         }
 367                         break;
 368                 }
 369         } else {
 370                 fd = open(imageFile, O_RDWR|O_CREAT|O_EXLOCK|O_NONBLOCK, 0644);
 371         }
 372         printf("Using memory file: %s\n", imageFile);
 373         if (fd < 0 || fstat(fd, &st) < 0) {
 374                 err(1, "Unable to open/create %s", imageFile);
 375                 /* NOT REACHED */
 376         }
 377
 378         /*
 379          * Truncate or extend the file as necessary.
 380          */
 381         if (st.st_size > Maxmem_bytes) {
 382                 ftruncate(fd, Maxmem_bytes);
 383         } else if (st.st_size < Maxmem_bytes) {
 384                 char *zmem;
 385                 off_t off = st.st_size & ~SEG_MASK;
 386
 387                 kprintf("%s: Reserving blocks for memory image\n", imageFile);
 388                 zmem = malloc(SEG_SIZE);
 389                 bzero(zmem, SEG_SIZE);
 390                 lseek(fd, off, SEEK_SET);
 391                 while (off < Maxmem_bytes) {
 392                         if (write(fd, zmem, SEG_SIZE) != SEG_SIZE) {
 393                                 err(1, "Unable to reserve blocks for memory image");
 394                                 /* NOT REACHED */
 395                         }
 396                         off += SEG_SIZE;
 397                 }
 398                 if (fsync(fd) < 0)
 399                         err(1, "Unable to reserve blocks for memory image");
 400                 free(zmem);
 401         }
 402         MemImageFd = fd;
 403         Maxmem = Maxmem_bytes >> PAGE_SHIFT;
 404 }
 405
 406 /*
 407  * Initialize kernel memory.  This reserves kernel virtual memory by using
 408  * MAP_VPAGETABLE
 409  */
 410
 411 static
 412 void
 413 init_kern_memory(void)
 414 {
 415         void *base;
 416         void *try;
 417         char *zero;
 418         char dummy;
 419         char *topofstack = &dummy;
 420         vpte_t pte;
 421         int i;
 422
 423         /*
 424          * Memory map our kernel virtual memory space.  Note that the
 425          * kernel image itself is not made part of this memory for the
 426          * moment.
 427          *
 428          * The memory map must be segment-aligned so we can properly
 429          * offset KernelPTD.
 430          *
 431          * If the system kernel has a different MAXDSIZ, it might not
 432          * be possible to map kernel memory in its prefered location.
 433          * Try a number of different locations.
 434          */
 435         try = (void *)0x40000000;
 436         base = NULL;
 437         while ((char *)try + KERNEL_KVA_SIZE < topofstack) {
 438                 base = mmap(try, KERNEL_KVA_SIZE, PROT_READ|PROT_WRITE,
 439                             MAP_FILE|MAP_SHARED|MAP_VPAGETABLE,
 440                             MemImageFd, 0);
 441                 if (base == try)
 442                         break;
 443                 if (base != MAP_FAILED)
 444                         munmap(base, KERNEL_KVA_SIZE);
 445                 try = (char *)try + 0x10000000;
 446         }
 447         if (base != try) {
 448                 err(1, "Unable to mmap() kernel virtual memory!");
 449                 /* NOT REACHED */
 450         }
 451         madvise(base, KERNEL_KVA_SIZE, MADV_NOSYNC);
 452         KvaStart = (vm_offset_t)base;
 453         KvaSize = KERNEL_KVA_SIZE;
 454         KvaEnd = KvaStart + KvaSize;
 455         printf("KVM mapped at %p-%p\n", (void *)KvaStart, (void *)KvaEnd);
 456
 457         /*
 458          * Create a top-level page table self-mapping itself.
 459          *
 460          * Initialize the page directory at physical page index 0 to point
 461          * to an array of page table pages starting at physical page index 1
 462          */
 463         lseek(MemImageFd, 0L, 0);
 464         for (i = 0; i < KERNEL_KVA_SIZE / SEG_SIZE; ++i) {
 465                 pte = ((i + 1) * PAGE_SIZE) | VPTE_V | VPTE_R | VPTE_W;
 466                 write(MemImageFd, &pte, sizeof(pte));
 467         }
 468
 469         /*
 470          * Initialize the PTEs in the page table pages required to map the
 471          * page table itself.  This includes mapping the page directory page
 472          * at the base so we go one more loop then normal.
 473          */
 474         lseek(MemImageFd, PAGE_SIZE, 0);
 475         for (i = 0; i <= KERNEL_KVA_SIZE / SEG_SIZE * sizeof(vpte_t); ++i) {
 476                 pte = (i * PAGE_SIZE) | VPTE_V | VPTE_R | VPTE_W;
 477                 write(MemImageFd, &pte, sizeof(pte));
 478         }
 479
 480         /*
 481          * Initialize remaining PTEs to 0.  We may be reusing a memory image
 482          * file.  This is approximately a megabyte.
 483          */
 484         i = (KERNEL_KVA_SIZE / PAGE_SIZE - i) * sizeof(pte);
 485         zero = malloc(PAGE_SIZE);
 486         bzero(zero, PAGE_SIZE);
 487         while (i) {
 488                 write(MemImageFd, zero, (i > PAGE_SIZE) ? PAGE_SIZE : i);
 489                 i = i - ((i > PAGE_SIZE) ? PAGE_SIZE : i);
 490         }
 491         free(zero);
 492
 493         /*
 494          * Enable the page table and calculate pointers to our self-map
 495          * for easy kernel page table manipulation.
 496          *
 497          * KernelPTA must be offset so we can do direct VA translations
 498          */
 499         mcontrol(base, KERNEL_KVA_SIZE, MADV_SETMAP,
 500                  0 | VPTE_R | VPTE_W | VPTE_V);
 501         KernelPTD = (vpte_t *)base;                       /* pg directory */
 502         KernelPTA = (vpte_t *)((char *)base + PAGE_SIZE); /* pg table pages */
 503         KernelPTA -= KvaStart >> PAGE_SHIFT;
 504
 505         /*
 506          * phys_avail[] represents unallocated physical memory.  MI code
 507          * will use phys_avail[] to create the vm_page array.
 508          */
 509         phys_avail[0] = PAGE_SIZE +
 510                         KERNEL_KVA_SIZE / PAGE_SIZE * sizeof(vpte_t);
 511         phys_avail[0] = (phys_avail[0] + PAGE_MASK) & ~(vm_paddr_t)PAGE_MASK;
 512         phys_avail[1] = Maxmem_bytes;
 513
 514         /*
 515          * (virtual_start, virtual_end) represent unallocated kernel virtual
 516          * memory.  MI code will create kernel_map using these parameters.
 517          */
 518         virtual_start = KvaStart + PAGE_SIZE +
 519                         KERNEL_KVA_SIZE / PAGE_SIZE * sizeof(vpte_t);
 520         virtual_start = (virtual_start + PAGE_MASK) & ~(vm_offset_t)PAGE_MASK;
 521         virtual_end = KvaStart + KERNEL_KVA_SIZE;
 522
 523         /*
 524          * kernel_vm_end could be set to virtual_end but we want some
 525          * indication of how much of the kernel_map we've used, so
 526          * set it low and let pmap_growkernel increase it even though we
 527          * don't need to create any new page table pages.
 528          */
 529         kernel_vm_end = virtual_start;
 530
 531         /*
 532          * Allocate space for process 0's UAREA.
 533          */
 534         proc0paddr = (void *)virtual_start;
 535         for (i = 0; i < UPAGES; ++i) {
 536                 pmap_kenter_quick(virtual_start, phys_avail[0]);
 537                 virtual_start += PAGE_SIZE;
 538                 phys_avail[0] += PAGE_SIZE;
 539         }
 540
 541         /*
 542          * crashdumpmap
 543          */
 544         crashdumpmap = virtual_start;
 545         virtual_start += MAXDUMPPGS * PAGE_SIZE;
 546
 547         /*
 548          * msgbufp maps the system message buffer
 549          */
 550         assert((MSGBUF_SIZE & PAGE_MASK) == 0);
 551         msgbufp = (void *)virtual_start;
 552         for (i = 0; i < (MSGBUF_SIZE >> PAGE_SHIFT); ++i) {
 553                 pmap_kenter_quick(virtual_start, phys_avail[0]);
 554                 virtual_start += PAGE_SIZE;
 555                 phys_avail[0] += PAGE_SIZE;
 556         }
 557         msgbufinit(msgbufp, MSGBUF_SIZE);
 558
 559         /*
 560          * used by kern_memio for /dev/mem access
 561          */
 562         ptvmmap = (caddr_t)virtual_start;
 563         virtual_start += PAGE_SIZE;
 564
 565         /*
 566          * Bootstrap the kernel_pmap
 567          */
 568         pmap_bootstrap();
 569 }
 570
 571 /*
 572  * Map the per-cpu globaldata for cpu #0.  Allocate the space using
 573  * virtual_start and phys_avail[0]
 574  */
 575 static
 576 void
 577 init_globaldata(void)
 578 {
 579         int i;
 580         vm_paddr_t pa;
 581         vm_offset_t va;
 582
 583         /*
 584          * Reserve enough KVA to cover possible cpus.  This is a considerable
 585          * amount of KVA since the privatespace structure includes two
 586          * whole page table mappings.
 587          */
 588         virtual_start = (virtual_start + SEG_MASK) & ~(vm_offset_t)SEG_MASK;
 589         CPU_prvspace = (void *)virtual_start;
 590         virtual_start += sizeof(struct privatespace) * SMP_MAXCPU;
 591
 592         /*
 593          * Allocate enough physical memory to cover the mdglobaldata
 594          * portion of the space and the idle stack and map the pages
 595          * into KVA.  For cpu #0 only.
 596          */
 597         for (i = 0; i < sizeof(struct mdglobaldata); i += PAGE_SIZE) {
 598                 pa = phys_avail[0];
 599                 va = (vm_offset_t)&CPU_prvspace[0].mdglobaldata + i;
 600                 pmap_kenter_quick(va, pa);
 601                 phys_avail[0] += PAGE_SIZE;
 602         }
 603         for (i = 0; i < sizeof(CPU_prvspace[0].idlestack); i += PAGE_SIZE) {
 604                 pa = phys_avail[0];
 605                 va = (vm_offset_t)&CPU_prvspace[0].idlestack + i;
 606                 pmap_kenter_quick(va, pa);
 607                 phys_avail[0] += PAGE_SIZE;
 608         }
 609
 610         /*
 611          * Setup the %gs for cpu #0.  The mycpu macro works after this
 612          * point.
 613          */
 614         tls_set_fs(&CPU_prvspace[0], sizeof(struct privatespace));
 615 }
 616
 617 /*
 618  * Initialize very low level systems including thread0, proc0, etc.
 619  */
 620 static
 621 void
 622 init_vkernel(void)
 623 {
 624         struct mdglobaldata *gd;
 625
 626         gd = &CPU_prvspace[0].mdglobaldata;
 627         bzero(gd, sizeof(*gd));
 628
 629         gd->mi.gd_curthread = &thread0;
 630         thread0.td_gd = &gd->mi;
 631         ncpus = 1;
 632         ncpus2 = 1;     /* rounded down power of 2 */
 633         ncpus_fit = 1;  /* rounded up power of 2 */
 634         /* ncpus2_mask and ncpus_fit_mask are 0 */
 635         init_param1();
 636         gd->mi.gd_prvspace = &CPU_prvspace[0];
 637         mi_gdinit(&gd->mi, 0);
 638         cpu_gdinit(gd, 0);
 639         mi_proc0init(&gd->mi, proc0paddr);
 640         lwp0.lwp_md.md_regs = &proc0_tf;
 641
 642         /*init_locks();*/
 643         cninit();
 644         rand_initialize();
 645 #if 0   /* #ifdef DDB */
 646         kdb_init();
 647         if (boothowto & RB_KDB)
 648                 Debugger("Boot flags requested debugger");
 649 #endif
 650 #if 0
 651         initializecpu();        /* Initialize CPU registers */
 652 #endif
 653         init_param2((phys_avail[1] - phys_avail[0]) / PAGE_SIZE);
 654
 655 #if 0
 656         /*
 657          * Map the message buffer
 658          */
 659         for (off = 0; off < round_page(MSGBUF_SIZE); off += PAGE_SIZE)
 660                 pmap_kenter((vm_offset_t)msgbufp + off, avail_end + off);
 661         msgbufinit(msgbufp, MSGBUF_SIZE);
 662 #endif
 663 #if 0
 664         thread0.td_pcb_cr3 ... MMU
 665         lwp0.lwp_md.md_regs = &proc0_tf;
 666 #endif
 667 }
 668
 669 /*
 670  * Filesystem image paths for the virtual kernel are optional.
 671  * If specified they each should point to a disk image,
 672  * the first of which will become the root disk.
 673  *
 674  * The virtual kernel caches data from our 'disk' just like a normal kernel,
 675  * so we do not really want the real kernel to cache the data too.  Use
 676  * O_DIRECT to remove the duplication.
 677  */
 678 static
 679 void
 680 init_disk(char *diskExp[], int diskFileNum, enum vkdisk_type type)
 681 {
 682         int i;
 683
 684         if (diskFileNum == 0)
 685                 return;
 686
 687         for(i=0; i < diskFileNum; i++){
 688                 char *fname;
 689                 fname = diskExp[i];
 690
 691                 if (fname == NULL) {
 692                         warnx("Invalid argument to '-r'");
 693                         continue;
 694                 }
 695
 696                 if (DiskNum < VKDISK_MAX) {
 697                         struct stat st;
 698                         struct vkdisk_info* info = NULL;
 699                         int fd;
 700                         size_t l = 0;
 701
 702                         if (type == VKD_DISK)
 703                             fd = open(fname, O_RDWR|O_DIRECT|O_EXLOCK|O_NONBLOCK, 0644);
 704                         else
 705                             fd = open(fname, O_RDONLY|O_DIRECT, 0644);
 706                         if (fd < 0 || fstat(fd, &st) < 0) {
 707                                 if (errno == EAGAIN)
 708                                         fprintf(stderr, "You may already have a vkernel using this disk image!\n");
 709                                 err(1, "Unable to open/create %s", fname);
 710                                 /* NOT REACHED */
 711                         }
 712                         /* get rid of O_NONBLOCK, keep O_DIRECT */
 713                         if (type == VKD_DISK)
 714                                 fcntl(fd, F_SETFL, O_DIRECT);
 715
 716                         info = &DiskInfo[DiskNum];
 717                         l = strlen(fname);
 718
 719                         info->unit = i;
 720                         info->fd = fd;
 721                         info->type = type;
 722                         memcpy(info->fname, fname, l);
 723
 724                         if (DiskNum == 0) {
 725                                 if (type == VKD_CD)
 726                                     rootdevnames[0] = "cd9660:vcd0a";
 727                                 else if (type == VKD_DISK)
 728                                     rootdevnames[0] = "ufs:vkd0s0a";
 729                         }
 730
 731                         DiskNum++;
 732                 } else {
 733                         warnx("vkd%d (%s) > VKDISK_MAX", DiskNum, fname);
 734                         continue;
 735                 }
 736         }
 737 }
 738
 739 static
 740 int
 741 netif_set_tapflags(int tap_unit, int f, int s)
 742 {
 743         struct ifreq ifr;
 744         int flags;
 745
 746         bzero(&ifr, sizeof(ifr));
 747
 748         snprintf(ifr.ifr_name, sizeof(ifr.ifr_name), "tap%d", tap_unit);
 749         if (ioctl(s, SIOCGIFFLAGS, &ifr) < 0) {
 750                 warn("tap%d: ioctl(SIOCGIFFLAGS) failed", tap_unit);
 751                 return -1;
 752         }
 753
 754         /*
 755          * Adjust if_flags
 756          *
 757          * If the flags are already set/cleared, then we return
 758          * immediately to avoid extra syscalls
 759          */
 760         flags = (ifr.ifr_flags & 0xffff) | (ifr.ifr_flagshigh << 16);
 761         if (f < 0) {
 762                 /* Turn off flags */
 763                 f = -f;
 764                 if ((flags & f) == 0)
 765                         return 0;
 766                 flags &= ~f;
 767         } else {
 768                 /* Turn on flags */
 769                 if (flags & f)
 770                         return 0;
 771                 flags |= f;
 772         }
 773
 774         /*
 775          * Fix up ifreq.ifr_name, since it may be trashed
 776          * in previous ioctl(SIOCGIFFLAGS)
 777          */
 778         snprintf(ifr.ifr_name, sizeof(ifr.ifr_name), "tap%d", tap_unit);
 779
 780         ifr.ifr_flags = flags & 0xffff;
 781         ifr.ifr_flagshigh = flags >> 16;
 782         if (ioctl(s, SIOCSIFFLAGS, &ifr) < 0) {
 783                 warn("tap%d: ioctl(SIOCSIFFLAGS) failed", tap_unit);
 784                 return -1;
 785         }
 786         return 0;
 787 }
 788
 789 static
 790 int
 791 netif_set_tapaddr(int tap_unit, in_addr_t addr, in_addr_t mask, int s)
 792 {
 793         struct ifaliasreq ifra;
 794         struct sockaddr_in *in;
 795
 796         bzero(&ifra, sizeof(ifra));
 797         snprintf(ifra.ifra_name, sizeof(ifra.ifra_name), "tap%d", tap_unit);
 798
 799         /* Setup address */
 800         in = (struct sockaddr_in *)&ifra.ifra_addr;
 801         in->sin_family = AF_INET;
 802         in->sin_len = sizeof(*in);
 803         in->sin_addr.s_addr = addr;
 804
 805         if (mask != 0) {
 806                 /* Setup netmask */
 807                 in = (struct sockaddr_in *)&ifra.ifra_mask;
 808                 in->sin_len = sizeof(*in);
 809                 in->sin_addr.s_addr = mask;
 810         }
 811
 812         if (ioctl(s, SIOCAIFADDR, &ifra) < 0) {
 813                 warn("tap%d: ioctl(SIOCAIFADDR) failed", tap_unit);
 814                 return -1;
 815         }
 816         return 0;
 817 }
 818
 819 static
 820 int
 821 netif_add_tap2brg(int tap_unit, const char *ifbridge, int s)
 822 {
 823         struct ifbreq ifbr;
 824         struct ifdrv ifd;
 825
 826         bzero(&ifbr, sizeof(ifbr));
 827         snprintf(ifbr.ifbr_ifsname, sizeof(ifbr.ifbr_ifsname),
 828                  "tap%d", tap_unit);
 829
 830         bzero(&ifd, sizeof(ifd));
 831         strlcpy(ifd.ifd_name, ifbridge, sizeof(ifd.ifd_name));
 832         ifd.ifd_cmd = BRDGADD;
 833         ifd.ifd_len = sizeof(ifbr);
 834         ifd.ifd_data = &ifbr;
 835
 836         if (ioctl(s, SIOCSDRVSPEC, &ifd) < 0) {
 837                 /*
 838                  * 'errno == EEXIST' means that the tap(4) is already
 839                  * a member of the bridge(4)
 840                  */
 841                 if (errno != EEXIST) {
 842                         warn("ioctl(%s, SIOCSDRVSPEC) failed", ifbridge);
 843                         return -1;
 844                 }
 845         }
 846         return 0;
 847 }
 848
 849 #define TAPDEV_OFLAGS   (O_RDWR | O_NONBLOCK)
 850
 851 /* XXX major()/minor() can't be used in vkernel */
 852 #define TAPDEV_MAJOR(x) ((int)(((u_int)(x) >> 8) & 0xff))
 853 #define TAPDEV_MINOR(x) ((int)((x) & 0xffff00ff))
 854
 855 #ifndef TAP_CDEV_MAJOR
 856 #define TAP_CDEV_MAJOR  149
 857 #endif
 858
 859 /*
 860  * Locate the first unused tap(4) device file if auto mode is requested,
 861  * or open the user supplied device file, and bring up the corresponding
 862  * tap(4) interface.
 863  *
 864  * NOTE: Only tap(4) device file is supported currently
 865  */
 866 static
 867 int
 868 netif_open_tap(const char *netif, int *tap_unit, int s)
 869 {
 870         char tap_dev[MAXPATHLEN];
 871         int tap_fd, failed;
 872         struct stat st;
 873
 874         *tap_unit = -1;
 875
 876         if (strcmp(netif, "auto") == 0) {
 877                 int i;
 878
 879                 /*
 880                  * Find first unused tap(4) device file
 881                  */
 882                 for (i = 0; ; ++i) {
 883                         snprintf(tap_dev, sizeof(tap_dev), "/dev/tap%d", i);
 884                         tap_fd = open(tap_dev, TAPDEV_OFLAGS);
 885                         if (tap_fd >= 0 || errno == ENOENT)
 886                                 break;
 887                 }
 888                 if (tap_fd < 0) {
 889                         warnx("Unable to find a free tap(4)");
 890                         return -1;
 891                 }
 892         } else {
 893                 /*
 894                  * User supplied tap(4) device file
 895                  */
 896                 if (netif[0] == '/')    /* Absolute path */
 897                         strlcpy(tap_dev, netif, sizeof(tap_dev));
 898                 else
 899                         snprintf(tap_dev, sizeof(tap_dev), "/dev/%s", netif);
 900
 901                 tap_fd = open(tap_dev, TAPDEV_OFLAGS);
 902                 if (tap_fd < 0) {
 903                         warn("Unable to open %s", tap_dev);
 904                         return -1;
 905                 }
 906         }
 907
 908         /*
 909          * Check whether the device file is a tap(4)
 910          */
 911         failed = 1;
 912         if (fstat(tap_fd, &st) == 0 && S_ISCHR(st.st_mode) &&
 913             TAPDEV_MAJOR(st.st_rdev) == TAP_CDEV_MAJOR) {
 914                 *tap_unit = TAPDEV_MINOR(st.st_rdev);
 915
 916                 /*
 917                  * Bring up the corresponding tap(4) interface
 918                  */
 919                 if (netif_set_tapflags(*tap_unit, IFF_UP, s) == 0)
 920                         failed = 0;
 921         } else {
 922                 warnx("%s is not a tap(4) device", tap_dev);
 923         }
 924
 925         if (failed) {
 926                 close(tap_fd);
 927                 tap_fd = -1;
 928                 *tap_unit = -1;
 929         }
 930         return tap_fd;
 931 }
 932
 933 #undef TAPDEV_MAJOR
 934 #undef TAPDEV_MINOR
 935 #undef TAPDEV_OFLAGS
 936
 937 /*
 938  * Following syntax is supported,
 939  * 1) x.x.x.x             tap(4)'s address is x.x.x.x
 940  *
 941  * 2) x.x.x.x/z           tap(4)'s address is x.x.x.x
 942  *                        tap(4)'s netmask len is z
 943  *
 944  * 3) x.x.x.x:y.y.y.y     tap(4)'s address is x.x.x.x
 945  *                        pseudo netif's address is y.y.y.y
 946  *
 947  * 4) x.x.x.x:y.y.y.y/z   tap(4)'s address is x.x.x.x
 948  *                        pseudo netif's address is y.y.y.y
 949  *                        tap(4) and pseudo netif's netmask len are z
 950  *
 951  * 5) bridgeX             tap(4) will be added to bridgeX
 952  *
 953  * 6) bridgeX:y.y.y.y     tap(4) will be added to bridgeX
 954  *                        pseudo netif's address is y.y.y.y
 955  *
 956  * 7) bridgeX:y.y.y.y/z   tap(4) will be added to bridgeX
 957  *                        pseudo netif's address is y.y.y.y
 958  *                        pseudo netif's netmask len is z
 959  */
 960 static
 961 int
 962 netif_init_tap(int tap_unit, in_addr_t *addr, in_addr_t *mask, int s)
 963 {
 964         in_addr_t tap_addr, netmask, netif_addr;
 965         int next_netif_addr;
 966         char *tok, *masklen_str, *ifbridge;
 967
 968         *addr = 0;
 969         *mask = 0;
 970
 971         tok = strtok(NULL, ":/");
 972         if (tok == NULL) {
 973                 /*
 974                  * Nothing special, simply use tap(4) as backend
 975                  */
 976                 return 0;
 977         }
 978
 979         if (inet_pton(AF_INET, tok, &tap_addr) > 0) {
 980                 /*
 981                  * tap(4)'s address is supplied
 982                  */
 983                 ifbridge = NULL;
 984
 985                 /*
 986                  * If there is next token, then it may be pseudo
 987                  * netif's address or netmask len for tap(4)
 988                  */
 989                 next_netif_addr = 0;
 990         } else {
 991                 /*
 992                  * Not tap(4)'s address, assume it as a bridge(4)
 993                  * iface name
 994                  */
 995                 tap_addr = 0;
 996                 ifbridge = tok;
 997
 998                 /*
 999                  * If there is next token, then it must be pseudo
1000                  * netif's address
1001                  */
1002                 next_netif_addr = 1;
1003         }
1004
1005         netmask = netif_addr = 0;
1006
1007         tok = strtok(NULL, ":/");
1008         if (tok == NULL)
1009                 goto back;
1010
1011         if (inet_pton(AF_INET, tok, &netif_addr) <= 0) {
1012                 if (next_netif_addr) {
1013                         warnx("Invalid pseudo netif address: %s", tok);
1014                         return -1;
1015                 }
1016                 netif_addr = 0;
1017
1018                 /*
1019                  * Current token is not address, then it must be netmask len
1020                  */
1021                 masklen_str = tok;
1022         } else {
1023                 /*
1024                  * Current token is pseudo netif address, if there is next token
1025                  * it must be netmask len
1026                  */
1027                 masklen_str = strtok(NULL, "/");
1028         }
1029
1030         /* Calculate netmask */
1031         if (masklen_str != NULL) {
1032                 u_long masklen;
1033
1034                 masklen = strtoul(masklen_str, NULL, 10);
1035                 if (masklen < 32 && masklen > 0) {
1036                         netmask = htonl(~((1LL << (32 - masklen)) - 1)
1037                                         & 0xffffffff);
1038                 } else {
1039                         warnx("Invalid netmask len: %lu", masklen);
1040                         return -1;
1041                 }
1042         }
1043
1044         /* Make sure there is no more token left */
1045         if (strtok(NULL, ":/") != NULL) {
1046                 warnx("Invalid argument to '-I'");
1047                 return -1;
1048         }
1049
1050 back:
1051         if (ifbridge == NULL) {
1052                 /* Set tap(4) address/netmask */
1053                 if (netif_set_tapaddr(tap_unit, tap_addr, netmask, s) < 0)
1054                         return -1;
1055         } else {
1056                 /* Tie tap(4) to bridge(4) */
1057                 if (netif_add_tap2brg(tap_unit, ifbridge, s) < 0)
1058                         return -1;
1059         }
1060
1061         *addr = netif_addr;
1062         *mask = netmask;
1063         return 0;
1064 }
1065
1066 /*
1067  * NetifInfo[] will be filled for pseudo netif initialization.
1068  * NetifNum will be bumped to reflect the number of valid entries
1069  * in NetifInfo[].
1070  */
1071 static
1072 void
1073 init_netif(char *netifExp[], int netifExpNum)
1074 {
1075         int i, s;
1076
1077         if (netifExpNum == 0)
1078                 return;
1079
1080         s = socket(AF_INET, SOCK_DGRAM, 0);     /* for ioctl(SIOC) */
1081         if (s < 0)
1082                 return;
1083
1084         for (i = 0; i < netifExpNum; ++i) {
1085                 struct vknetif_info *info;
1086                 in_addr_t netif_addr, netif_mask;
1087                 int tap_fd, tap_unit;
1088                 char *netif;
1089
1090                 netif = strtok(netifExp[i], ":");
1091                 if (netif == NULL) {
1092                         warnx("Invalid argument to '-I'");
1093                         continue;
1094                 }
1095
1096                 /*
1097                  * Open tap(4) device file and bring up the
1098                  * corresponding interface
1099                  */
1100                 tap_fd = netif_open_tap(netif, &tap_unit, s);
1101                 if (tap_fd < 0)
1102                         continue;
1103
1104                 /*
1105                  * Initialize tap(4) and get address/netmask
1106                  * for pseudo netif
1107                  *
1108                  * NB: Rest part of netifExp[i] is passed
1109                  *     to netif_init_tap() implicitly.
1110                  */
1111                 if (netif_init_tap(tap_unit, &netif_addr, &netif_mask, s) < 0) {
1112                         /*
1113                          * NB: Closing tap(4) device file will bring
1114                          *     down the corresponding interface
1115                          */
1116                         close(tap_fd);
1117                         continue;
1118                 }
1119
1120                 info = &NetifInfo[NetifNum];
1121                 info->tap_fd = tap_fd;
1122                 info->tap_unit = tap_unit;
1123                 info->netif_addr = netif_addr;
1124                 info->netif_mask = netif_mask;
1125
1126                 NetifNum++;
1127                 if (NetifNum >= VKNETIF_MAX)    /* XXX will this happen? */
1128                         break;
1129         }
1130         close(s);
1131 }
1132
1133 static
1134 void
1135 writepid( void )
1136 {
1137         pid_t self;
1138         FILE *fp;
1139
1140         if (pid_file != NULL) {
1141                 self = getpid();
1142                 fp = fopen(pid_file, "w");
1143
1144                 if (fp != NULL) {
1145                         fprintf(fp, "%ld\n", (long)self);
1146                         fclose(fp);
1147                 }
1148                 else {
1149                         perror("Warning: couldn't open pidfile");
1150                 }
1151         }
1152 }
1153
1154 static
1155 void
1156 cleanpid( void )
1157 {
1158         if (pid_file != NULL) {
1159                 if ( unlink(pid_file) != 0 )
1160                         perror("Warning: couldn't remove pidfile");
1161         }
1162 }
1163
1164 static
1165 void
1166 usage(const char *ctl, ...)
1167 {
1168         va_list va;
1169
1170         va_start(va, ctl);
1171         vfprintf(stderr, ctl, va);
1172         va_end(va);
1173         fprintf(stderr, "\n");
1174         exit(1);
1175 }
1176
1177 void
1178 cpu_reset(void)
1179 {
1180         kprintf("cpu reset, rebooting vkernel\n");
1181         closefrom(3);
1182         cleanpid();
1183         execv(save_av[0], save_av);
1184 }
1185
1186 void
1187 cpu_halt(void)
1188 {
1189         kprintf("cpu halt, exiting vkernel\n");
1190         cleanpid();
1191         exit(0);
1192 }
1193
1194 void
1195 setrealcpu(void)
1196 {
1197         switch(lwp_cpu_lock) {
1198         case LCL_PER_CPU:
1199                 if (bootverbose)
1200                         kprintf("Locking CPU%d to real cpu %d\n",
1201                                 mycpuid, next_cpu);
1202                 usched_set(getpid(), USCHED_SET_CPU, &next_cpu, sizeof(next_cpu));
1203                 next_cpu++;
1204                 if (next_cpu >= real_ncpus)
1205                         next_cpu = 0;
1206                 break;
1207         case LCL_SINGLE_CPU:
1208                 if (bootverbose)
1209                         kprintf("Locking CPU%d to real cpu %d\n",
1210                                 mycpuid, next_cpu);
1211                 usched_set(getpid(), USCHED_SET_CPU, &next_cpu, sizeof(next_cpu));
1212                 break;
1213         default:
1214                 /* do not map virtual cpus to real cpus */
1215                 break;
1216         }
1217 }
1218