1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 22 /* 23 * Copyright 2008 Sun Microsystems, Inc. All rights reserved. 24 * Use is subject to license terms. 25 */ 26 27 #pragma ident "%Z%%M% %I% %E% SMI" 28 29 /* 30 * This file contains the functionality that mimics the boot operations 31 * on SPARC systems or the old boot.bin/multiboot programs on x86 systems. 32 * The x86 kernel now does everything on its own. 33 */ 34 35 #include <sys/types.h> 36 #include <sys/bootconf.h> 37 #include <sys/bootsvcs.h> 38 #include <sys/bootinfo.h> 39 #include <sys/multiboot.h> 40 #include <sys/bootvfs.h> 41 #include <sys/bootprops.h> 42 #include <sys/varargs.h> 43 #include <sys/param.h> 44 #include <sys/machparam.h> 45 #include <sys/archsystm.h> 46 #include <sys/boot_console.h> 47 #include <sys/cmn_err.h> 48 #include <sys/systm.h> 49 #include <sys/promif.h> 50 #include <sys/archsystm.h> 51 #include <sys/x86_archext.h> 52 #include <sys/kobj.h> 53 #include <sys/privregs.h> 54 #include <sys/sysmacros.h> 55 #include <sys/ctype.h> 56 #ifdef __xpv 57 #include <sys/hypervisor.h> 58 #include <net/if.h> 59 #endif 60 #include <vm/kboot_mmu.h> 61 #include <vm/hat_pte.h> 62 #include "acpi_fw.h" 63 64 static int have_console = 0; /* set once primitive console is initialized */ 65 static char *boot_args = ""; 66 67 /* 68 * Debugging macros 69 */ 70 static uint_t kbm_debug = 0; 71 #define DBG_MSG(s) { if (kbm_debug) bop_printf(NULL, "%s", s); } 72 #define DBG(x) { if (kbm_debug) \ 73 bop_printf(NULL, "%s is %" PRIx64 "\n", #x, (uint64_t)(x)); \ 74 } 75 76 #define PUT_STRING(s) { \ 77 char *cp; \ 78 for (cp = (s); *cp; ++cp) \ 79 bcons_putchar(*cp); \ 80 } 81 82 struct xboot_info *xbootp; /* boot info from "glue" code in low memory */ 83 bootops_t bootop; /* simple bootops we'll pass on to kernel */ 84 struct bsys_mem bm; 85 86 static uintptr_t next_virt; /* next available virtual address */ 87 static paddr_t next_phys; /* next available physical address from dboot */ 88 static paddr_t high_phys = -(paddr_t)1; /* last used physical address */ 89 90 /* 91 * buffer for vsnprintf for console I/O 92 */ 93 #define BUFFERSIZE 256 94 static char buffer[BUFFERSIZE]; 95 /* 96 * stuff to store/report/manipulate boot property settings. 97 */ 98 typedef struct bootprop { 99 struct bootprop *bp_next; 100 char *bp_name; 101 uint_t bp_vlen; 102 char *bp_value; 103 } bootprop_t; 104 105 static bootprop_t *bprops = NULL; 106 static char *curr_page = NULL; /* ptr to avail bprop memory */ 107 static int curr_space = 0; /* amount of memory at curr_page */ 108 109 #ifdef __xpv 110 start_info_t *xen_info; 111 shared_info_t *HYPERVISOR_shared_info; 112 #endif 113 114 /* 115 * some allocator statistics 116 */ 117 static ulong_t total_bop_alloc_scratch = 0; 118 static ulong_t total_bop_alloc_kernel = 0; 119 120 static void build_firmware_properties(void); 121 122 static int early_allocation = 1; 123 124 /* 125 * Pointers to where System Resource Affinity Table (SRAT) and 126 * System Locality Information Table (SLIT) are mapped into virtual memory 127 */ 128 struct srat *srat_ptr = NULL; 129 struct slit *slit_ptr = NULL; 130 131 132 /* 133 * Allocate aligned physical memory at boot time. This allocator allocates 134 * from the highest possible addresses. This avoids exhausting memory that 135 * would be useful for DMA buffers. 136 */ 137 paddr_t 138 do_bop_phys_alloc(uint64_t size, uint64_t align) 139 { 140 paddr_t pa = 0; 141 paddr_t start; 142 paddr_t end; 143 struct memlist *ml = (struct memlist *)xbootp->bi_phys_install; 144 145 /* 146 * Be careful if high memory usage is limited in startup.c 147 * Since there are holes in the low part of the physical address 148 * space we can treat physmem as a pfn (not just a pgcnt) and 149 * get a conservative upper limit. 150 */ 151 if (physmem != 0 && high_phys > pfn_to_pa(physmem)) 152 high_phys = pfn_to_pa(physmem); 153 154 /* 155 * find the lowest or highest available memory in physinstalled 156 * On 32 bit avoid physmem above 4Gig if PAE isn't enabled 157 */ 158 #if defined(__i386) 159 if (xbootp->bi_use_pae == 0 && high_phys > FOUR_GIG) 160 high_phys = FOUR_GIG; 161 #endif 162 163 /* 164 * find the highest available memory in physinstalled 165 */ 166 size = P2ROUNDUP(size, align); 167 for (; ml; ml = ml->next) { 168 start = P2ROUNDUP(ml->address, align); 169 end = P2ALIGN(ml->address + ml->size, align); 170 if (start < next_phys) 171 start = P2ROUNDUP(next_phys, align); 172 if (end > high_phys) 173 end = P2ALIGN(high_phys, align); 174 175 if (end <= start) 176 continue; 177 if (end - start < size) 178 continue; 179 180 /* 181 * Early allocations need to use low memory, since 182 * physmem might be further limited by bootenv.rc 183 */ 184 if (early_allocation) { 185 if (pa == 0 || start < pa) 186 pa = start; 187 } else { 188 if (end - size > pa) 189 pa = end - size; 190 } 191 } 192 if (pa != 0) { 193 if (early_allocation) 194 next_phys = pa + size; 195 else 196 high_phys = pa; 197 return (pa); 198 } 199 bop_panic("do_bop_phys_alloc(0x%" PRIx64 ", 0x%" PRIx64 200 ") Out of memory\n", size, align); 201 /*NOTREACHED*/ 202 } 203 204 static uintptr_t 205 alloc_vaddr(size_t size, paddr_t align) 206 { 207 uintptr_t rv; 208 209 next_virt = P2ROUNDUP(next_virt, (uintptr_t)align); 210 rv = (uintptr_t)next_virt; 211 next_virt += size; 212 return (rv); 213 } 214 215 /* 216 * Allocate virtual memory. The size is always rounded up to a multiple 217 * of base pagesize. 218 */ 219 220 /*ARGSUSED*/ 221 static caddr_t 222 do_bsys_alloc(bootops_t *bop, caddr_t virthint, size_t size, int align) 223 { 224 paddr_t a = align; /* same type as pa for masking */ 225 uint_t pgsize; 226 paddr_t pa; 227 uintptr_t va; 228 ssize_t s; /* the aligned size */ 229 uint_t level; 230 uint_t is_kernel = (virthint != 0); 231 232 if (a < MMU_PAGESIZE) 233 a = MMU_PAGESIZE; 234 else if (!ISP2(a)) 235 prom_panic("do_bsys_alloc() incorrect alignment"); 236 size = P2ROUNDUP(size, MMU_PAGESIZE); 237 238 /* 239 * Use the next aligned virtual address if we weren't given one. 240 */ 241 if (virthint == NULL) { 242 virthint = (caddr_t)alloc_vaddr(size, a); 243 total_bop_alloc_scratch += size; 244 } else { 245 total_bop_alloc_kernel += size; 246 } 247 248 /* 249 * allocate the physical memory 250 */ 251 pa = do_bop_phys_alloc(size, a); 252 253 /* 254 * Add the mappings to the page tables, try large pages first. 255 */ 256 va = (uintptr_t)virthint; 257 s = size; 258 level = 1; 259 pgsize = xbootp->bi_use_pae ? TWO_MEG : FOUR_MEG; 260 if (xbootp->bi_use_largepage && a == pgsize) { 261 while (IS_P2ALIGNED(pa, pgsize) && IS_P2ALIGNED(va, pgsize) && 262 s >= pgsize) { 263 kbm_map(va, pa, level, is_kernel); 264 va += pgsize; 265 pa += pgsize; 266 s -= pgsize; 267 } 268 } 269 270 /* 271 * Map remaining pages use small mappings 272 */ 273 level = 0; 274 pgsize = MMU_PAGESIZE; 275 while (s > 0) { 276 kbm_map(va, pa, level, is_kernel); 277 va += pgsize; 278 pa += pgsize; 279 s -= pgsize; 280 } 281 return (virthint); 282 } 283 284 /* 285 * Free virtual memory - we'll just ignore these. 286 */ 287 /*ARGSUSED*/ 288 static void 289 do_bsys_free(bootops_t *bop, caddr_t virt, size_t size) 290 { 291 bop_printf(NULL, "do_bsys_free(virt=0x%p, size=0x%lx) ignored\n", 292 (void *)virt, size); 293 } 294 295 /* 296 * Old interface 297 */ 298 /*ARGSUSED*/ 299 static caddr_t 300 do_bsys_ealloc( 301 bootops_t *bop, 302 caddr_t virthint, 303 size_t size, 304 int align, 305 int flags) 306 { 307 prom_panic("unsupported call to BOP_EALLOC()\n"); 308 return (0); 309 } 310 311 312 static void 313 bsetprop(char *name, int nlen, void *value, int vlen) 314 { 315 uint_t size; 316 uint_t need_size; 317 bootprop_t *b; 318 319 /* 320 * align the size to 16 byte boundary 321 */ 322 size = sizeof (bootprop_t) + nlen + 1 + vlen; 323 size = (size + 0xf) & ~0xf; 324 if (size > curr_space) { 325 need_size = (size + (MMU_PAGEOFFSET)) & MMU_PAGEMASK; 326 curr_page = do_bsys_alloc(NULL, 0, need_size, MMU_PAGESIZE); 327 curr_space = need_size; 328 } 329 330 /* 331 * use a bootprop_t at curr_page and link into list 332 */ 333 b = (bootprop_t *)curr_page; 334 curr_page += sizeof (bootprop_t); 335 curr_space -= sizeof (bootprop_t); 336 b->bp_next = bprops; 337 bprops = b; 338 339 /* 340 * follow by name and ending zero byte 341 */ 342 b->bp_name = curr_page; 343 bcopy(name, curr_page, nlen); 344 curr_page += nlen; 345 *curr_page++ = 0; 346 curr_space -= nlen + 1; 347 348 /* 349 * copy in value, but no ending zero byte 350 */ 351 b->bp_value = curr_page; 352 b->bp_vlen = vlen; 353 if (vlen > 0) { 354 bcopy(value, curr_page, vlen); 355 curr_page += vlen; 356 curr_space -= vlen; 357 } 358 359 /* 360 * align new values of curr_page, curr_space 361 */ 362 while (curr_space & 0xf) { 363 ++curr_page; 364 --curr_space; 365 } 366 } 367 368 static void 369 bsetprops(char *name, char *value) 370 { 371 bsetprop(name, strlen(name), value, strlen(value) + 1); 372 } 373 374 static void 375 bsetprop64(char *name, uint64_t value) 376 { 377 bsetprop(name, strlen(name), (void *)&value, sizeof (value)); 378 } 379 380 static void 381 bsetpropsi(char *name, int value) 382 { 383 char prop_val[32]; 384 385 (void) snprintf(prop_val, sizeof (prop_val), "%d", value); 386 bsetprops(name, prop_val); 387 } 388 389 /* 390 * to find the size of the buffer to allocate 391 */ 392 /*ARGSUSED*/ 393 int 394 do_bsys_getproplen(bootops_t *bop, const char *name) 395 { 396 bootprop_t *b; 397 398 for (b = bprops; b; b = b->bp_next) { 399 if (strcmp(name, b->bp_name) != 0) 400 continue; 401 return (b->bp_vlen); 402 } 403 return (-1); 404 } 405 406 /* 407 * get the value associated with this name 408 */ 409 /*ARGSUSED*/ 410 int 411 do_bsys_getprop(bootops_t *bop, const char *name, void *value) 412 { 413 bootprop_t *b; 414 415 for (b = bprops; b; b = b->bp_next) { 416 if (strcmp(name, b->bp_name) != 0) 417 continue; 418 bcopy(b->bp_value, value, b->bp_vlen); 419 return (0); 420 } 421 return (-1); 422 } 423 424 /* 425 * get the name of the next property in succession from the standalone 426 */ 427 /*ARGSUSED*/ 428 static char * 429 do_bsys_nextprop(bootops_t *bop, char *name) 430 { 431 bootprop_t *b; 432 433 /* 434 * A null name is a special signal for the 1st boot property 435 */ 436 if (name == NULL || strlen(name) == 0) { 437 if (bprops == NULL) 438 return (NULL); 439 return (bprops->bp_name); 440 } 441 442 for (b = bprops; b; b = b->bp_next) { 443 if (name != b->bp_name) 444 continue; 445 b = b->bp_next; 446 if (b == NULL) 447 return (NULL); 448 return (b->bp_name); 449 } 450 return (NULL); 451 } 452 453 /* 454 * Parse numeric value from a string. Understands decimal, hex, octal, - and ~ 455 */ 456 static int 457 parse_value(char *p, uint64_t *retval) 458 { 459 int adjust = 0; 460 uint64_t tmp = 0; 461 int digit; 462 int radix = 10; 463 464 *retval = 0; 465 if (*p == '-' || *p == '~') 466 adjust = *p++; 467 468 if (*p == '0') { 469 ++p; 470 if (*p == 0) 471 return (0); 472 if (*p == 'x' || *p == 'X') { 473 radix = 16; 474 ++p; 475 } else { 476 radix = 8; 477 ++p; 478 } 479 } 480 while (*p) { 481 if ('0' <= *p && *p <= '9') 482 digit = *p - '0'; 483 else if ('a' <= *p && *p <= 'f') 484 digit = 10 + *p - 'a'; 485 else if ('A' <= *p && *p <= 'F') 486 digit = 10 + *p - 'A'; 487 else 488 return (-1); 489 if (digit >= radix) 490 return (-1); 491 tmp = tmp * radix + digit; 492 ++p; 493 } 494 if (adjust == '-') 495 tmp = -tmp; 496 else if (adjust == '~') 497 tmp = ~tmp; 498 *retval = tmp; 499 return (0); 500 } 501 502 /* 503 * 2nd part of building the table of boot properties. This includes: 504 * - values from /boot/solaris/bootenv.rc (ie. eeprom(1m) values) 505 * 506 * lines look like one of: 507 * ^$ 508 * ^# comment till end of line 509 * setprop name 'value' 510 * setprop name value 511 * setprop name "value" 512 * 513 * we do single character I/O since this is really just looking at memory 514 */ 515 void 516 boot_prop_finish(void) 517 { 518 int fd; 519 char *line; 520 int c; 521 int bytes_read; 522 char *name; 523 int n_len; 524 char *value; 525 int v_len; 526 char *inputdev; /* these override the command line if serial ports */ 527 char *outputdev; 528 char *consoledev; 529 uint64_t lvalue; 530 int use_xencons = 0; 531 532 #ifdef __xpv 533 if (!DOMAIN_IS_INITDOMAIN(xen_info)) 534 use_xencons = 1; 535 #endif /* __xpv */ 536 537 DBG_MSG("Opening /boot/solaris/bootenv.rc\n"); 538 fd = BRD_OPEN(bfs_ops, "/boot/solaris/bootenv.rc", 0); 539 DBG(fd); 540 541 line = do_bsys_alloc(NULL, NULL, MMU_PAGESIZE, MMU_PAGESIZE); 542 while (fd >= 0) { 543 544 /* 545 * get a line 546 */ 547 for (c = 0; ; ++c) { 548 bytes_read = BRD_READ(bfs_ops, fd, line + c, 1); 549 if (bytes_read == 0) { 550 if (c == 0) 551 goto done; 552 break; 553 } 554 if (line[c] == '\n') 555 break; 556 } 557 line[c] = 0; 558 559 /* 560 * ignore comment lines 561 */ 562 c = 0; 563 while (ISSPACE(line[c])) 564 ++c; 565 if (line[c] == '#' || line[c] == 0) 566 continue; 567 568 /* 569 * must have "setprop " or "setprop\t" 570 */ 571 if (strncmp(line + c, "setprop ", 8) != 0 && 572 strncmp(line + c, "setprop\t", 8) != 0) 573 continue; 574 c += 8; 575 while (ISSPACE(line[c])) 576 ++c; 577 if (line[c] == 0) 578 continue; 579 580 /* 581 * gather up the property name 582 */ 583 name = line + c; 584 n_len = 0; 585 while (line[c] && !ISSPACE(line[c])) 586 ++n_len, ++c; 587 588 /* 589 * gather up the value, if any 590 */ 591 value = ""; 592 v_len = 0; 593 while (ISSPACE(line[c])) 594 ++c; 595 if (line[c] != 0) { 596 value = line + c; 597 while (line[c] && !ISSPACE(line[c])) 598 ++v_len, ++c; 599 } 600 601 if (v_len >= 2 && value[0] == value[v_len - 1] && 602 (value[0] == '\'' || value[0] == '"')) { 603 ++value; 604 v_len -= 2; 605 } 606 name[n_len] = 0; 607 if (v_len > 0) 608 value[v_len] = 0; 609 else 610 continue; 611 612 /* 613 * ignore "boot-file" property, it's now meaningless 614 */ 615 if (strcmp(name, "boot-file") == 0) 616 continue; 617 if (strcmp(name, "boot-args") == 0 && 618 strlen(boot_args) > 0) 619 continue; 620 621 /* 622 * If a property was explicitly set on the command line 623 * it will override a setting in bootenv.rc 624 */ 625 if (do_bsys_getproplen(NULL, name) > 0) 626 continue; 627 628 bsetprop(name, n_len, value, v_len + 1); 629 } 630 done: 631 if (fd >= 0) 632 BRD_CLOSE(bfs_ops, fd); 633 634 /* 635 * Check if we have to limit the boot time allocator 636 */ 637 if (do_bsys_getproplen(NULL, "physmem") != -1 && 638 do_bsys_getprop(NULL, "physmem", line) >= 0 && 639 parse_value(line, &lvalue) != -1) { 640 if (0 < lvalue && (lvalue < physmem || physmem == 0)) { 641 physmem = (pgcnt_t)lvalue; 642 DBG(physmem); 643 } 644 } 645 early_allocation = 0; 646 647 /* 648 * check to see if we have to override the default value of the console 649 */ 650 if (!use_xencons) { 651 inputdev = line; 652 v_len = do_bsys_getproplen(NULL, "input-device"); 653 if (v_len > 0) 654 (void) do_bsys_getprop(NULL, "input-device", inputdev); 655 else 656 v_len = 0; 657 inputdev[v_len] = 0; 658 659 outputdev = inputdev + v_len + 1; 660 v_len = do_bsys_getproplen(NULL, "output-device"); 661 if (v_len > 0) 662 (void) do_bsys_getprop(NULL, "output-device", 663 outputdev); 664 else 665 v_len = 0; 666 outputdev[v_len] = 0; 667 668 consoledev = outputdev + v_len + 1; 669 v_len = do_bsys_getproplen(NULL, "console"); 670 if (v_len > 0) 671 (void) do_bsys_getprop(NULL, "console", consoledev); 672 else 673 v_len = 0; 674 consoledev[v_len] = 0; 675 bcons_init2(inputdev, outputdev, consoledev); 676 } else { 677 /* 678 * Ensure console property exists 679 * If not create it as "hypervisor" 680 */ 681 v_len = do_bsys_getproplen(NULL, "console"); 682 if (v_len < 0) 683 bsetprops("console", "hypervisor"); 684 inputdev = outputdev = consoledev = "hypervisor"; 685 bcons_init2(inputdev, outputdev, consoledev); 686 } 687 688 if (strstr((char *)xbootp->bi_cmdline, "prom_debug") || kbm_debug) { 689 value = line; 690 bop_printf(NULL, "\nBoot properties:\n"); 691 name = ""; 692 while ((name = do_bsys_nextprop(NULL, name)) != NULL) { 693 bop_printf(NULL, "\t0x%p %s = ", (void *)name, name); 694 (void) do_bsys_getprop(NULL, name, value); 695 v_len = do_bsys_getproplen(NULL, name); 696 bop_printf(NULL, "len=%d ", v_len); 697 value[v_len] = 0; 698 bop_printf(NULL, "%s\n", value); 699 } 700 } 701 } 702 703 /* 704 * print formatted output 705 */ 706 /*PRINTFLIKE2*/ 707 /*ARGSUSED*/ 708 void 709 bop_printf(bootops_t *bop, const char *fmt, ...) 710 { 711 va_list ap; 712 713 if (have_console == 0) 714 return; 715 716 va_start(ap, fmt); 717 (void) vsnprintf(buffer, BUFFERSIZE, fmt, ap); 718 va_end(ap); 719 PUT_STRING(buffer); 720 } 721 722 /* 723 * Another panic() variant; this one can be used even earlier during boot than 724 * prom_panic(). 725 */ 726 /*PRINTFLIKE1*/ 727 void 728 bop_panic(const char *fmt, ...) 729 { 730 va_list ap; 731 732 va_start(ap, fmt); 733 bop_printf(NULL, fmt, ap); 734 va_end(ap); 735 736 bop_printf(NULL, "\nPress any key to reboot.\n"); 737 (void) bcons_getchar(); 738 bop_printf(NULL, "Resetting...\n"); 739 pc_reset(); 740 } 741 742 /* 743 * Do a real mode interrupt BIOS call 744 */ 745 typedef struct bios_regs { 746 unsigned short ax, bx, cx, dx, si, di, bp, es, ds; 747 } bios_regs_t; 748 typedef int (*bios_func_t)(int, bios_regs_t *); 749 750 /*ARGSUSED*/ 751 static void 752 do_bsys_doint(bootops_t *bop, int intnum, struct bop_regs *rp) 753 { 754 #if defined(__xpv) 755 prom_panic("unsupported call to BOP_DOINT()\n"); 756 #else /* __xpv */ 757 static int firsttime = 1; 758 bios_func_t bios_func = (bios_func_t)(void *)(uintptr_t)0x5000; 759 bios_regs_t br; 760 761 /* 762 * The first time we do this, we have to copy the pre-packaged 763 * low memory bios call code image into place. 764 */ 765 if (firsttime) { 766 extern char bios_image[]; 767 extern uint32_t bios_size; 768 769 bcopy(bios_image, (void *)bios_func, bios_size); 770 firsttime = 0; 771 } 772 773 br.ax = rp->eax.word.ax; 774 br.bx = rp->ebx.word.bx; 775 br.cx = rp->ecx.word.cx; 776 br.dx = rp->edx.word.dx; 777 br.bp = rp->ebp.word.bp; 778 br.si = rp->esi.word.si; 779 br.di = rp->edi.word.di; 780 br.ds = rp->ds; 781 br.es = rp->es; 782 783 DBG_MSG("Doing BIOS call..."); 784 rp->eflags = bios_func(intnum, &br); 785 DBG_MSG("done\n"); 786 787 rp->eax.word.ax = br.ax; 788 rp->ebx.word.bx = br.bx; 789 rp->ecx.word.cx = br.cx; 790 rp->edx.word.dx = br.dx; 791 rp->ebp.word.bp = br.bp; 792 rp->esi.word.si = br.si; 793 rp->edi.word.di = br.di; 794 rp->ds = br.ds; 795 rp->es = br.es; 796 #endif /* __xpv */ 797 } 798 799 static struct boot_syscalls bop_sysp = { 800 bcons_getchar, 801 bcons_putchar, 802 bcons_ischar, 803 }; 804 805 static char *whoami; 806 807 #define BUFLEN 64 808 809 #if defined(__xpv) 810 811 static char namebuf[32]; 812 813 static void 814 xen_parse_props(char *s, char *prop_map[], int n_prop) 815 { 816 char **prop_name = prop_map; 817 char *cp = s, *scp; 818 819 do { 820 scp = cp; 821 while ((*cp != NULL) && (*cp != ':')) 822 cp++; 823 824 if ((scp != cp) && (*prop_name != NULL)) { 825 *cp = NULL; 826 bsetprops(*prop_name, scp); 827 } 828 829 cp++; 830 prop_name++; 831 n_prop--; 832 } while (n_prop > 0); 833 } 834 835 #define VBDPATHLEN 64 836 837 /* 838 * parse the 'xpv-root' property to create properties used by 839 * ufs_mountroot. 840 */ 841 static void 842 xen_vbdroot_props(char *s) 843 { 844 char vbdpath[VBDPATHLEN] = "/xpvd/xdf@"; 845 const char lnamefix[] = "/dev/dsk/c0d"; 846 char *pnp; 847 char *prop_p; 848 char mi; 849 short minor; 850 long addr = 0; 851 852 pnp = vbdpath + strlen(vbdpath); 853 prop_p = s + strlen(lnamefix); 854 while ((*prop_p != '\0') && (*prop_p != 's') && (*prop_p != 'p')) 855 addr = addr * 10 + *prop_p++ - '0'; 856 (void) snprintf(pnp, VBDPATHLEN, "%lx", addr); 857 pnp = vbdpath + strlen(vbdpath); 858 if (*prop_p == 's') 859 mi = 'a'; 860 else if (*prop_p == 'p') 861 mi = 'q'; 862 else 863 ASSERT(0); /* shouldn't be here */ 864 prop_p++; 865 ASSERT(*prop_p != '\0'); 866 if (ISDIGIT(*prop_p)) { 867 minor = *prop_p - '0'; 868 prop_p++; 869 if (ISDIGIT(*prop_p)) { 870 minor = minor * 10 + *prop_p - '0'; 871 } 872 } else { 873 /* malformed root path, use 0 as default */ 874 minor = 0; 875 } 876 ASSERT(minor < 16); /* at most 16 partitions */ 877 mi += minor; 878 *pnp++ = ':'; 879 *pnp++ = mi; 880 *pnp++ = '\0'; 881 bsetprops("fstype", "ufs"); 882 bsetprops("bootpath", vbdpath); 883 884 DBG_MSG("VBD bootpath set to "); 885 DBG_MSG(vbdpath); 886 DBG_MSG("\n"); 887 } 888 889 /* 890 * parse the xpv-nfsroot property to create properties used by 891 * nfs_mountroot. 892 */ 893 static void 894 xen_nfsroot_props(char *s) 895 { 896 char *prop_map[] = { 897 BP_SERVER_IP, /* server IP address */ 898 BP_SERVER_NAME, /* server hostname */ 899 BP_SERVER_PATH, /* root path */ 900 }; 901 int n_prop = sizeof (prop_map) / sizeof (prop_map[0]); 902 903 bsetprop("fstype", 6, "nfsdyn", 7); 904 905 xen_parse_props(s, prop_map, n_prop); 906 907 /* 908 * If a server name wasn't specified, use a default. 909 */ 910 if (do_bsys_getproplen(NULL, BP_SERVER_NAME) == -1) 911 bsetprops(BP_SERVER_NAME, "unknown"); 912 } 913 914 /* 915 * Extract our IP address, etc. from the "xpv-ip" property. 916 */ 917 static void 918 xen_ip_props(char *s) 919 { 920 char *prop_map[] = { 921 BP_HOST_IP, /* IP address */ 922 NULL, /* NFS server IP address (ignored in */ 923 /* favour of xpv-nfsroot) */ 924 BP_ROUTER_IP, /* IP gateway */ 925 BP_SUBNET_MASK, /* IP subnet mask */ 926 "xpv-hostname", /* hostname (ignored) */ 927 BP_NETWORK_INTERFACE, /* interface name */ 928 "xpv-hcp", /* host configuration protocol */ 929 }; 930 int n_prop = sizeof (prop_map) / sizeof (prop_map[0]); 931 char ifname[IFNAMSIZ]; 932 933 xen_parse_props(s, prop_map, n_prop); 934 935 /* 936 * A Linux dom0 administrator expects all interfaces to be 937 * called "ethX", which is not the case here. 938 * 939 * If the interface name specified is "eth0", presume that 940 * this is really intended to be "xnf0" (the first domU -> 941 * dom0 interface for this domain). 942 */ 943 if ((do_bsys_getprop(NULL, BP_NETWORK_INTERFACE, ifname) == 0) && 944 (strcmp("eth0", ifname) == 0)) { 945 bsetprops(BP_NETWORK_INTERFACE, "xnf0"); 946 bop_printf(NULL, 947 "network interface name 'eth0' replaced with 'xnf0'\n"); 948 } 949 } 950 951 #else /* __xpv */ 952 953 static void 954 setup_rarp_props(struct sol_netinfo *sip) 955 { 956 char buf[BUFLEN]; /* to hold ip/mac addrs */ 957 uint8_t *val; 958 959 val = (uint8_t *)&sip->sn_ciaddr; 960 (void) snprintf(buf, BUFLEN, "%d.%d.%d.%d", 961 val[0], val[1], val[2], val[3]); 962 bsetprops(BP_HOST_IP, buf); 963 964 val = (uint8_t *)&sip->sn_siaddr; 965 (void) snprintf(buf, BUFLEN, "%d.%d.%d.%d", 966 val[0], val[1], val[2], val[3]); 967 bsetprops(BP_SERVER_IP, buf); 968 969 if (sip->sn_giaddr != 0) { 970 val = (uint8_t *)&sip->sn_giaddr; 971 (void) snprintf(buf, BUFLEN, "%d.%d.%d.%d", 972 val[0], val[1], val[2], val[3]); 973 bsetprops(BP_ROUTER_IP, buf); 974 } 975 976 if (sip->sn_netmask != 0) { 977 val = (uint8_t *)&sip->sn_netmask; 978 (void) snprintf(buf, BUFLEN, "%d.%d.%d.%d", 979 val[0], val[1], val[2], val[3]); 980 bsetprops(BP_SUBNET_MASK, buf); 981 } 982 983 if (sip->sn_mactype != 4 || sip->sn_maclen != 6) { 984 bop_printf(NULL, "unsupported mac type %d, mac len %d\n", 985 sip->sn_mactype, sip->sn_maclen); 986 } else { 987 val = sip->sn_macaddr; 988 (void) snprintf(buf, BUFLEN, "%x:%x:%x:%x:%x:%x", 989 val[0], val[1], val[2], val[3], val[4], val[5]); 990 bsetprops(BP_BOOT_MAC, buf); 991 } 992 } 993 994 #endif /* __xpv */ 995 996 /* 997 * 1st pass at building the table of boot properties. This includes: 998 * - values set on the command line: -B a=x,b=y,c=z .... 999 * - known values we just compute (ie. from xbootp) 1000 * - values from /boot/solaris/bootenv.rc (ie. eeprom(1m) values) 1001 * 1002 * the grub command line looked like: 1003 * kernel boot-file [-B prop=value[,prop=value]...] [boot-args] 1004 * 1005 * whoami is the same as boot-file 1006 */ 1007 static void 1008 build_boot_properties(void) 1009 { 1010 char *name; 1011 int name_len; 1012 char *value; 1013 int value_len; 1014 struct boot_modules *bm; 1015 char *propbuf; 1016 int quoted = 0; 1017 int boot_arg_len; 1018 #ifndef __xpv 1019 static int stdout_val = 0; 1020 uchar_t boot_device; 1021 char str[3]; 1022 multiboot_info_t *mbi; 1023 int netboot; 1024 struct sol_netinfo *sip; 1025 #endif 1026 1027 /* 1028 * These have to be done first, so that kobj_mount_root() works 1029 */ 1030 DBG_MSG("Building boot properties\n"); 1031 propbuf = do_bsys_alloc(NULL, NULL, MMU_PAGESIZE, 0); 1032 DBG((uintptr_t)propbuf); 1033 if (xbootp->bi_module_cnt > 0) { 1034 bm = xbootp->bi_modules; 1035 bsetprop64("ramdisk_start", (uint64_t)(uintptr_t)bm->bm_addr); 1036 bsetprop64("ramdisk_end", (uint64_t)(uintptr_t)bm->bm_addr + 1037 bm->bm_size); 1038 } 1039 1040 DBG_MSG("Parsing command line for boot properties\n"); 1041 value = xbootp->bi_cmdline; 1042 1043 /* 1044 * allocate memory to collect boot_args into 1045 */ 1046 boot_arg_len = strlen(xbootp->bi_cmdline) + 1; 1047 boot_args = do_bsys_alloc(NULL, NULL, boot_arg_len, MMU_PAGESIZE); 1048 boot_args[0] = 0; 1049 boot_arg_len = 0; 1050 1051 #ifdef __xpv 1052 /* 1053 * Xen puts a lot of device information in front of the kernel name 1054 * let's grab them and make them boot properties. The first 1055 * string w/o an "=" in it will be the boot-file property. 1056 */ 1057 (void) strcpy(namebuf, "xpv-"); 1058 for (;;) { 1059 /* 1060 * get to next property 1061 */ 1062 while (ISSPACE(*value)) 1063 ++value; 1064 name = value; 1065 /* 1066 * look for an "=" 1067 */ 1068 while (*value && !ISSPACE(*value) && *value != '=') { 1069 value++; 1070 } 1071 if (*value != '=') { /* no "=" in the property */ 1072 value = name; 1073 break; 1074 } 1075 name_len = value - name; 1076 value_len = 0; 1077 /* 1078 * skip over the "=" 1079 */ 1080 value++; 1081 while (value[value_len] && !ISSPACE(value[value_len])) { 1082 ++value_len; 1083 } 1084 /* 1085 * build property name with "xpv-" prefix 1086 */ 1087 if (name_len + 4 > 32) { /* skip if name too long */ 1088 value += value_len; 1089 continue; 1090 } 1091 bcopy(name, &namebuf[4], name_len); 1092 name_len += 4; 1093 namebuf[name_len] = 0; 1094 bcopy(value, propbuf, value_len); 1095 propbuf[value_len] = 0; 1096 bsetprops(namebuf, propbuf); 1097 1098 /* 1099 * xpv-root is set to the logical disk name of the xen 1100 * VBD when booting from a disk-based filesystem. 1101 */ 1102 if (strcmp(namebuf, "xpv-root") == 0) 1103 xen_vbdroot_props(propbuf); 1104 /* 1105 * While we're here, if we have a "xpv-nfsroot" property 1106 * then we need to set "fstype" to "nfsdyn" so we mount 1107 * our root from the nfs server. Also parse the xpv-nfsroot 1108 * property to create the properties that nfs_mountroot will 1109 * need to find the root and mount it. 1110 */ 1111 if (strcmp(namebuf, "xpv-nfsroot") == 0) 1112 xen_nfsroot_props(propbuf); 1113 1114 if (strcmp(namebuf, "xpv-ip") == 0) 1115 xen_ip_props(propbuf); 1116 value += value_len; 1117 } 1118 #endif 1119 1120 while (ISSPACE(*value)) 1121 ++value; 1122 /* 1123 * value now points at the boot-file 1124 */ 1125 value_len = 0; 1126 while (value[value_len] && !ISSPACE(value[value_len])) 1127 ++value_len; 1128 if (value_len > 0) { 1129 whoami = propbuf; 1130 bcopy(value, whoami, value_len); 1131 whoami[value_len] = 0; 1132 bsetprops("boot-file", whoami); 1133 /* 1134 * strip leading path stuff from whoami, so running from 1135 * PXE/miniroot makes sense. 1136 */ 1137 if (strstr(whoami, "/platform/") != NULL) 1138 whoami = strstr(whoami, "/platform/"); 1139 bsetprops("whoami", whoami); 1140 } 1141 1142 /* 1143 * Values forcibly set boot properties on the command line via -B. 1144 * Allow use of quotes in values. Other stuff goes on kernel 1145 * command line. 1146 */ 1147 name = value + value_len; 1148 while (*name != 0) { 1149 /* 1150 * anything not " -B" is copied to the command line 1151 */ 1152 if (!ISSPACE(name[0]) || name[1] != '-' || name[2] != 'B') { 1153 boot_args[boot_arg_len++] = *name; 1154 boot_args[boot_arg_len] = 0; 1155 ++name; 1156 continue; 1157 } 1158 1159 /* 1160 * skip the " -B" and following white space 1161 */ 1162 name += 3; 1163 while (ISSPACE(*name)) 1164 ++name; 1165 while (*name && !ISSPACE(*name)) { 1166 value = strstr(name, "="); 1167 if (value == NULL) 1168 break; 1169 name_len = value - name; 1170 ++value; 1171 value_len = 0; 1172 quoted = 0; 1173 for (; ; ++value_len) { 1174 if (!value[value_len]) 1175 break; 1176 1177 /* 1178 * is this value quoted? 1179 */ 1180 if (value_len == 0 && 1181 (value[0] == '\'' || value[0] == '"')) { 1182 quoted = value[0]; 1183 ++value_len; 1184 } 1185 1186 /* 1187 * In the quote accept any character, 1188 * but look for ending quote. 1189 */ 1190 if (quoted) { 1191 if (value[value_len] == quoted) 1192 quoted = 0; 1193 continue; 1194 } 1195 1196 /* 1197 * a comma or white space ends the value 1198 */ 1199 if (value[value_len] == ',' || 1200 ISSPACE(value[value_len])) 1201 break; 1202 } 1203 1204 if (value_len == 0) { 1205 bsetprop(name, name_len, "true", 5); 1206 } else { 1207 char *v = value; 1208 int l = value_len; 1209 if (v[0] == v[l - 1] && 1210 (v[0] == '\'' || v[0] == '"')) { 1211 ++v; 1212 l -= 2; 1213 } 1214 bcopy(v, propbuf, l); 1215 propbuf[l] = '\0'; 1216 bsetprop(name, name_len, propbuf, 1217 l + 1); 1218 } 1219 name = value + value_len; 1220 while (*name == ',') 1221 ++name; 1222 } 1223 } 1224 1225 /* 1226 * set boot-args property 1227 * 1275 name is bootargs, so set 1228 * that too 1229 */ 1230 bsetprops("boot-args", boot_args); 1231 bsetprops("bootargs", boot_args); 1232 1233 #ifndef __xpv 1234 /* 1235 * set the BIOS boot device from GRUB 1236 */ 1237 netboot = 0; 1238 mbi = xbootp->bi_mb_info; 1239 if (mbi != NULL && mbi->flags & 0x2) { 1240 boot_device = mbi->boot_device >> 24; 1241 if (boot_device == 0x20) 1242 netboot++; 1243 str[0] = (boot_device >> 4) + '0'; 1244 str[1] = (boot_device & 0xf) + '0'; 1245 str[2] = 0; 1246 bsetprops("bios-boot-device", str); 1247 } else { 1248 netboot = 1; 1249 } 1250 1251 /* 1252 * In the netboot case, drives_info is overloaded with the dhcp ack. 1253 * This is not multiboot compliant and requires special pxegrub! 1254 */ 1255 if (netboot && mbi->drives_length != 0) { 1256 sip = (struct sol_netinfo *)(uintptr_t)mbi->drives_addr; 1257 if (sip->sn_infotype == SN_TYPE_BOOTP) 1258 bsetprop("bootp-response", sizeof ("bootp-response"), 1259 (void *)(uintptr_t)mbi->drives_addr, 1260 mbi->drives_length); 1261 else if (sip->sn_infotype == SN_TYPE_RARP) 1262 setup_rarp_props(sip); 1263 } 1264 bsetprop("stdout", strlen("stdout"), 1265 &stdout_val, sizeof (stdout_val)); 1266 #endif /* __xpv */ 1267 1268 /* 1269 * more conjured up values for made up things.... 1270 */ 1271 #if defined(__xpv) 1272 bsetprops("mfg-name", "i86xpv"); 1273 bsetprops("impl-arch-name", "i86xpv"); 1274 #else 1275 bsetprops("mfg-name", "i86pc"); 1276 bsetprops("impl-arch-name", "i86pc"); 1277 #endif 1278 1279 /* 1280 * Build firmware-provided system properties 1281 */ 1282 build_firmware_properties(); 1283 1284 /* 1285 * XXPV 1286 * 1287 * Find out what these are: 1288 * - cpuid_feature_ecx_include 1289 * - cpuid_feature_ecx_exclude 1290 * - cpuid_feature_edx_include 1291 * - cpuid_feature_edx_exclude 1292 * 1293 * Find out what these are in multiboot: 1294 * - netdev-path 1295 * - fstype 1296 */ 1297 } 1298 1299 #ifdef __xpv 1300 /* 1301 * Under the Hypervisor, memory usable for DMA may be scarce. One 1302 * very likely large pool of DMA friendly memory is occupied by 1303 * the boot_archive, as it was loaded by grub into low MFNs. 1304 * 1305 * Here we free up that memory by copying the boot archive to what are 1306 * likely higher MFN pages and then swapping the mfn/pfn mappings. 1307 */ 1308 #define PFN_2GIG 0x80000 1309 static void 1310 relocate_boot_archive(void) 1311 { 1312 mfn_t max_mfn = HYPERVISOR_memory_op(XENMEM_maximum_ram_page, NULL); 1313 struct boot_modules *bm = xbootp->bi_modules; 1314 uintptr_t va; 1315 pfn_t va_pfn; 1316 mfn_t va_mfn; 1317 caddr_t copy; 1318 pfn_t copy_pfn; 1319 mfn_t copy_mfn; 1320 size_t len; 1321 int slop; 1322 int total = 0; 1323 int relocated = 0; 1324 int mmu_update_return; 1325 mmu_update_t t[2]; 1326 x86pte_t pte; 1327 1328 /* 1329 * If all MFN's are below 2Gig, don't bother doing this. 1330 */ 1331 if (max_mfn < PFN_2GIG) 1332 return; 1333 if (xbootp->bi_module_cnt < 1) { 1334 DBG_MSG("no boot_archive!"); 1335 return; 1336 } 1337 1338 DBG_MSG("moving boot_archive to high MFN memory\n"); 1339 va = (uintptr_t)bm->bm_addr; 1340 len = bm->bm_size; 1341 slop = va & MMU_PAGEOFFSET; 1342 if (slop) { 1343 va += MMU_PAGESIZE - slop; 1344 len -= MMU_PAGESIZE - slop; 1345 } 1346 len = P2ALIGN(len, MMU_PAGESIZE); 1347 1348 /* 1349 * Go through all boot_archive pages, swapping any low MFN pages 1350 * with memory at next_phys. 1351 */ 1352 while (len != 0) { 1353 ++total; 1354 va_pfn = mmu_btop(va - ONE_GIG); 1355 va_mfn = mfn_list[va_pfn]; 1356 if (mfn_list[va_pfn] < PFN_2GIG) { 1357 copy = kbm_remap_window(next_phys, 1); 1358 bcopy((void *)va, copy, MMU_PAGESIZE); 1359 copy_pfn = mmu_btop(next_phys); 1360 copy_mfn = mfn_list[copy_pfn]; 1361 1362 pte = mfn_to_ma(copy_mfn) | PT_NOCONSIST | PT_VALID; 1363 if (HYPERVISOR_update_va_mapping(va, pte, 1364 UVMF_INVLPG | UVMF_LOCAL)) 1365 bop_panic("relocate_boot_archive(): " 1366 "HYPERVISOR_update_va_mapping() failed"); 1367 1368 mfn_list[va_pfn] = copy_mfn; 1369 mfn_list[copy_pfn] = va_mfn; 1370 1371 t[0].ptr = mfn_to_ma(copy_mfn) | MMU_MACHPHYS_UPDATE; 1372 t[0].val = va_pfn; 1373 t[1].ptr = mfn_to_ma(va_mfn) | MMU_MACHPHYS_UPDATE; 1374 t[1].val = copy_pfn; 1375 if (HYPERVISOR_mmu_update(t, 2, &mmu_update_return, 1376 DOMID_SELF) != 0 || mmu_update_return != 2) 1377 bop_panic("relocate_boot_archive(): " 1378 "HYPERVISOR_mmu_update() failed"); 1379 1380 next_phys += MMU_PAGESIZE; 1381 ++relocated; 1382 } 1383 len -= MMU_PAGESIZE; 1384 va += MMU_PAGESIZE; 1385 } 1386 DBG_MSG("Relocated pages:\n"); 1387 DBG(relocated); 1388 DBG_MSG("Out of total pages:\n"); 1389 DBG(total); 1390 } 1391 #endif /* __xpv */ 1392 1393 #if !defined(__xpv) 1394 /* 1395 * Install a temporary IDT that lets us catch errors in the boot time code. 1396 * We shouldn't get any faults at all while this is installed, so we'll 1397 * just generate a traceback and exit. 1398 */ 1399 #ifdef __amd64 1400 static const int bcode_sel = B64CODE_SEL; 1401 #else 1402 static const int bcode_sel = B32CODE_SEL; 1403 #endif 1404 1405 /* 1406 * simple description of a stack frame (args are 32 bit only currently) 1407 */ 1408 typedef struct bop_frame { 1409 struct bop_frame *old_frame; 1410 pc_t retaddr; 1411 long arg[1]; 1412 } bop_frame_t; 1413 1414 void 1415 bop_traceback(bop_frame_t *frame) 1416 { 1417 pc_t pc; 1418 int cnt; 1419 int a; 1420 char *ksym; 1421 ulong_t off; 1422 1423 bop_printf(NULL, "Stack traceback:\n"); 1424 for (cnt = 0; cnt < 30; ++cnt) { /* up to 30 frames */ 1425 pc = frame->retaddr; 1426 if (pc == 0) 1427 break; 1428 ksym = kobj_getsymname(pc, &off); 1429 if (ksym) 1430 bop_printf(NULL, " %s+%lx", ksym, off); 1431 else 1432 bop_printf(NULL, " 0x%lx", pc); 1433 1434 frame = frame->old_frame; 1435 if (frame == 0) { 1436 bop_printf(NULL, "\n"); 1437 break; 1438 } 1439 for (a = 0; a < 6; ++a) { /* try for 6 args */ 1440 #if defined(__i386) 1441 if ((void *)&frame->arg[a] == (void *)frame->old_frame) 1442 break; 1443 if (a == 0) 1444 bop_printf(NULL, "("); 1445 else 1446 bop_printf(NULL, ","); 1447 bop_printf(NULL, "0x%lx", frame->arg[a]); 1448 #endif 1449 } 1450 bop_printf(NULL, ")\n"); 1451 } 1452 } 1453 1454 struct trapframe { 1455 ulong_t frame_ptr; /* %[er]bp pushed by our code */ 1456 ulong_t error_code; /* optional */ 1457 ulong_t inst_ptr; 1458 ulong_t code_seg; 1459 ulong_t flags_reg; 1460 #ifdef __amd64 1461 ulong_t stk_ptr; 1462 ulong_t stk_seg; 1463 #endif 1464 }; 1465 1466 void 1467 bop_trap(struct trapframe *tf) 1468 { 1469 bop_frame_t fakeframe; 1470 static int depth = 0; 1471 1472 /* 1473 * Check for an infinite loop of traps. 1474 */ 1475 if (++depth > 2) 1476 bop_panic("Nested trap"); 1477 1478 /* 1479 * adjust the tf for optional error_code by detecting the code selector 1480 */ 1481 if (tf->code_seg != bcode_sel) 1482 tf = (struct trapframe *)((uintptr_t)tf - sizeof (ulong_t)); 1483 1484 bop_printf(NULL, "Unexpected trap\n"); 1485 bop_printf(NULL, "instruction pointer 0x%lx\n", tf->inst_ptr); 1486 bop_printf(NULL, "error code, optional 0x%lx\n", 1487 tf->error_code & 0xffffffff); 1488 bop_printf(NULL, "code segment 0x%lx\n", tf->code_seg & 0xffff); 1489 bop_printf(NULL, "flags register 0x%lx\n", tf->flags_reg); 1490 #ifdef __amd64 1491 bop_printf(NULL, "return %%rsp 0x%lx\n", tf->stk_ptr); 1492 bop_printf(NULL, "return %%ss 0x%lx\n", tf->stk_seg & 0xffff); 1493 #endif 1494 fakeframe.old_frame = (bop_frame_t *)tf->frame_ptr; 1495 fakeframe.retaddr = (pc_t)tf->inst_ptr; 1496 bop_printf(NULL, "Attempting stack backtrace:\n"); 1497 bop_traceback(&fakeframe); 1498 bop_panic("unexpected trap in early boot"); 1499 } 1500 1501 extern void bop_trap_handler(void); 1502 1503 static gate_desc_t *bop_idt; 1504 1505 static desctbr_t bop_idt_info; 1506 1507 static void 1508 bop_idt_init(void) 1509 { 1510 int t; 1511 1512 bop_idt = (gate_desc_t *) 1513 do_bsys_alloc(NULL, NULL, MMU_PAGESIZE, MMU_PAGESIZE); 1514 bzero(bop_idt, MMU_PAGESIZE); 1515 for (t = 0; t < NIDT; ++t) { 1516 set_gatesegd(&bop_idt[t], &bop_trap_handler, bcode_sel, 1517 SDT_SYSIGT, TRP_KPL); 1518 } 1519 bop_idt_info.dtr_limit = (NIDT * sizeof (gate_desc_t)) - 1; 1520 bop_idt_info.dtr_base = (uintptr_t)bop_idt; 1521 wr_idtr(&bop_idt_info); 1522 } 1523 #endif /* !defined(__xpv) */ 1524 1525 /* 1526 * This is where we enter the kernel. It dummies up the boot_ops and 1527 * boot_syscalls vectors and jumps off to _kobj_boot() 1528 */ 1529 void 1530 _start(struct xboot_info *xbp) 1531 { 1532 bootops_t *bops = &bootop; 1533 extern void _kobj_boot(); 1534 1535 /* 1536 * 1st off - initialize the console for any error messages 1537 */ 1538 xbootp = xbp; 1539 #ifdef __xpv 1540 HYPERVISOR_shared_info = (void *)xbootp->bi_shared_info; 1541 xen_info = xbootp->bi_xen_start_info; 1542 #endif 1543 bcons_init((void *)xbootp->bi_cmdline); 1544 have_console = 1; 1545 1546 /* 1547 * enable debugging 1548 */ 1549 if (strstr((char *)xbootp->bi_cmdline, "kbm_debug")) 1550 kbm_debug = 1; 1551 1552 DBG_MSG("\n\n*** Entered Solaris in _start() cmdline is: "); 1553 DBG_MSG((char *)xbootp->bi_cmdline); 1554 DBG_MSG("\n\n\n"); 1555 1556 /* 1557 * physavail is no longer used by startup 1558 */ 1559 bm.physinstalled = xbp->bi_phys_install; 1560 bm.pcimem = xbp->bi_pcimem; 1561 bm.physavail = NULL; 1562 1563 /* 1564 * initialize the boot time allocator 1565 */ 1566 next_phys = xbootp->bi_next_paddr; 1567 DBG(next_phys); 1568 next_virt = (uintptr_t)xbootp->bi_next_vaddr; 1569 DBG(next_virt); 1570 DBG_MSG("Initializing boot time memory management..."); 1571 #ifdef __xpv 1572 { 1573 xen_platform_parameters_t p; 1574 1575 /* This call shouldn't fail, dboot already did it once. */ 1576 (void) HYPERVISOR_xen_version(XENVER_platform_parameters, &p); 1577 mfn_to_pfn_mapping = (pfn_t *)(xen_virt_start = p.virt_start); 1578 DBG(xen_virt_start); 1579 } 1580 #endif 1581 kbm_init(xbootp); 1582 DBG_MSG("done\n"); 1583 1584 /* 1585 * Fill in the bootops vector 1586 */ 1587 bops->bsys_version = BO_VERSION; 1588 bops->boot_mem = &bm; 1589 bops->bsys_alloc = do_bsys_alloc; 1590 bops->bsys_free = do_bsys_free; 1591 bops->bsys_getproplen = do_bsys_getproplen; 1592 bops->bsys_getprop = do_bsys_getprop; 1593 bops->bsys_nextprop = do_bsys_nextprop; 1594 bops->bsys_printf = bop_printf; 1595 bops->bsys_doint = do_bsys_doint; 1596 1597 /* 1598 * BOP_EALLOC() is no longer needed 1599 */ 1600 bops->bsys_ealloc = do_bsys_ealloc; 1601 1602 #ifdef __xpv 1603 /* 1604 * On domain 0 we need to free up some physical memory that is 1605 * usable for DMA. Since GRUB loaded the boot_archive, it is 1606 * sitting in low MFN memory. We'll relocated the boot archive 1607 * pages to high PFN memory. 1608 */ 1609 if (DOMAIN_IS_INITDOMAIN(xen_info)) 1610 relocate_boot_archive(); 1611 #endif 1612 1613 #ifndef __xpv 1614 /* 1615 * Install an IDT to catch early pagefaults (shouldn't have any). 1616 * Also needed for kmdb. 1617 */ 1618 bop_idt_init(); 1619 #endif 1620 1621 /* 1622 * Start building the boot properties from the command line 1623 */ 1624 DBG_MSG("Initializing boot properties:\n"); 1625 build_boot_properties(); 1626 1627 if (strstr((char *)xbootp->bi_cmdline, "prom_debug") || kbm_debug) { 1628 char *name; 1629 char *value; 1630 char *cp; 1631 int len; 1632 1633 value = do_bsys_alloc(NULL, NULL, MMU_PAGESIZE, MMU_PAGESIZE); 1634 bop_printf(NULL, "\nBoot properties:\n"); 1635 name = ""; 1636 while ((name = do_bsys_nextprop(NULL, name)) != NULL) { 1637 bop_printf(NULL, "\t0x%p %s = ", (void *)name, name); 1638 (void) do_bsys_getprop(NULL, name, value); 1639 len = do_bsys_getproplen(NULL, name); 1640 bop_printf(NULL, "len=%d ", len); 1641 value[len] = 0; 1642 for (cp = value; *cp; ++cp) { 1643 if (' ' <= *cp && *cp <= '~') 1644 bop_printf(NULL, "%c", *cp); 1645 else 1646 bop_printf(NULL, "-0x%x-", *cp); 1647 } 1648 bop_printf(NULL, "\n"); 1649 } 1650 } 1651 1652 /* 1653 * jump into krtld... 1654 */ 1655 _kobj_boot(&bop_sysp, NULL, bops, NULL); 1656 } 1657 1658 1659 /*ARGSUSED*/ 1660 static caddr_t 1661 no_more_alloc(bootops_t *bop, caddr_t virthint, size_t size, int align) 1662 { 1663 panic("Attempt to bsys_alloc() too late\n"); 1664 return (NULL); 1665 } 1666 1667 /*ARGSUSED*/ 1668 static void 1669 no_more_free(bootops_t *bop, caddr_t virt, size_t size) 1670 { 1671 panic("Attempt to bsys_free() too late\n"); 1672 } 1673 1674 void 1675 bop_no_more_mem(void) 1676 { 1677 DBG(total_bop_alloc_scratch); 1678 DBG(total_bop_alloc_kernel); 1679 bootops->bsys_alloc = no_more_alloc; 1680 bootops->bsys_free = no_more_free; 1681 } 1682 1683 1684 #ifndef __xpv 1685 /* 1686 * Set ACPI firmware properties 1687 */ 1688 1689 static caddr_t 1690 vmap_phys(size_t length, paddr_t pa) 1691 { 1692 paddr_t start, end; 1693 caddr_t va; 1694 size_t len, page; 1695 1696 start = P2ALIGN(pa, MMU_PAGESIZE); 1697 end = P2ROUNDUP(pa + length, MMU_PAGESIZE); 1698 len = end - start; 1699 va = (caddr_t)alloc_vaddr(len, MMU_PAGESIZE); 1700 for (page = 0; page < len; page += MMU_PAGESIZE) 1701 kbm_map((uintptr_t)va + page, start + page, 0, 0); 1702 return (va + (pa & MMU_PAGEOFFSET)); 1703 } 1704 1705 static uint8_t 1706 checksum_table(uint8_t *tp, size_t len) 1707 { 1708 uint8_t sum = 0; 1709 1710 while (len-- > 0) 1711 sum += *tp++; 1712 1713 return (sum); 1714 } 1715 1716 static int 1717 valid_rsdp(struct rsdp *rp) 1718 { 1719 1720 /* validate the V1.x checksum */ 1721 if (checksum_table((uint8_t *)&rp->v1, sizeof (struct rsdp_v1)) != 0) 1722 return (0); 1723 1724 /* If pre-ACPI 2.0, this is a valid RSDP */ 1725 if (rp->v1.revision < 2) 1726 return (1); 1727 1728 /* validate the V2.x checksum */ 1729 if (checksum_table((uint8_t *)rp, sizeof (struct rsdp)) != 0) 1730 return (0); 1731 1732 return (1); 1733 } 1734 1735 /* 1736 * Scan memory range for an RSDP; 1737 * see ACPI 3.0 Spec, 5.2.5.1 1738 */ 1739 static struct rsdp * 1740 scan_rsdp(paddr_t start, paddr_t end) 1741 { 1742 size_t len = end - start + 1; 1743 caddr_t ptr; 1744 1745 ptr = vmap_phys(len, start); 1746 while (len > 0) { 1747 if (strncmp(ptr, ACPI_RSDP_SIG, ACPI_RSDP_SIG_LEN) == 0) 1748 if (valid_rsdp((struct rsdp *)ptr)) 1749 return ((struct rsdp *)ptr); 1750 ptr += 16; 1751 len -= 16; 1752 } 1753 1754 return (NULL); 1755 } 1756 1757 /* 1758 * Refer to ACPI 3.0 Spec, section 5.2.5.1 to understand this function 1759 */ 1760 static struct rsdp * 1761 find_rsdp() { 1762 struct rsdp *rsdp; 1763 uint16_t *ebda_seg; 1764 paddr_t ebda_addr; 1765 1766 /* 1767 * Get the EBDA segment and scan the first 1K 1768 */ 1769 ebda_seg = (uint16_t *)vmap_phys(sizeof (uint16_t), ACPI_EBDA_SEG_ADDR); 1770 ebda_addr = *ebda_seg << 4; 1771 rsdp = scan_rsdp(ebda_addr, ebda_addr + ACPI_EBDA_LEN - 1); 1772 if (rsdp == NULL) 1773 /* if EBDA doesn't contain RSDP, look in BIOS memory */ 1774 rsdp = scan_rsdp(0xe0000, 0xfffff); 1775 return (rsdp); 1776 } 1777 1778 static struct table_header * 1779 map_fw_table(paddr_t table_addr) 1780 { 1781 struct table_header *tp; 1782 size_t len = MAX(sizeof (struct table_header), MMU_PAGESIZE); 1783 1784 /* 1785 * Map at least a page; if the table is larger than this, remap it 1786 */ 1787 tp = (struct table_header *)vmap_phys(len, table_addr); 1788 if (tp->len > len) 1789 tp = (struct table_header *)vmap_phys(tp->len, table_addr); 1790 return (tp); 1791 } 1792 1793 static struct table_header * 1794 find_fw_table(char *signature) 1795 { 1796 static int revision = 0; 1797 static struct xsdt *xsdt; 1798 static int len; 1799 paddr_t xsdt_addr; 1800 struct rsdp *rsdp; 1801 struct table_header *tp; 1802 paddr_t table_addr; 1803 int n; 1804 1805 if (strlen(signature) != ACPI_TABLE_SIG_LEN) 1806 return (NULL); 1807 1808 /* 1809 * Reading the ACPI 3.0 Spec, section 5.2.5.3 will help 1810 * understand this code. If we haven't already found the RSDT/XSDT, 1811 * revision will be 0. Find the RSDP and check the revision 1812 * to find out whether to use the RSDT or XSDT. If revision is 1813 * 0 or 1, use the RSDT and set internal revision to 1; if it is 2, 1814 * use the XSDT. If the XSDT address is 0, though, fall back to 1815 * revision 1 and use the RSDT. 1816 */ 1817 if (revision == 0) { 1818 if ((rsdp = (struct rsdp *)find_rsdp()) != NULL) { 1819 revision = rsdp->v1.revision; 1820 switch (revision) { 1821 case 2: 1822 /* 1823 * Use the XSDT unless BIOS is buggy and 1824 * claims to be rev 2 but has a null XSDT 1825 * address 1826 */ 1827 xsdt_addr = rsdp->xsdt; 1828 if (xsdt_addr != 0) 1829 break; 1830 /* FALLTHROUGH */ 1831 case 0: 1832 /* treat RSDP rev 0 as revision 1 internally */ 1833 revision = 1; 1834 /* FALLTHROUGH */ 1835 case 1: 1836 /* use the RSDT for rev 0/1 */ 1837 xsdt_addr = rsdp->v1.rsdt; 1838 break; 1839 default: 1840 /* unknown revision */ 1841 revision = 0; 1842 break; 1843 } 1844 } 1845 if (revision == 0) 1846 return (NULL); 1847 1848 /* cache the XSDT info */ 1849 xsdt = (struct xsdt *)map_fw_table(xsdt_addr); 1850 len = (xsdt->hdr.len - sizeof (xsdt->hdr)) / 1851 ((revision == 1) ? sizeof (uint32_t) : sizeof (uint64_t)); 1852 } 1853 1854 /* 1855 * Scan the table headers looking for a signature match 1856 */ 1857 for (n = 0; n < len; n++) { 1858 table_addr = (revision == 1) ? xsdt->p.r[n] : xsdt->p.x[n]; 1859 if (table_addr == 0) 1860 continue; 1861 tp = map_fw_table(table_addr); 1862 if (strncmp(tp->sig, signature, ACPI_TABLE_SIG_LEN) == 0) { 1863 return (tp); 1864 } 1865 } 1866 return (NULL); 1867 } 1868 1869 static void 1870 process_madt(struct madt *tp) 1871 { 1872 struct madt_processor *cpu, *end; 1873 uint32_t cpu_count = 0; 1874 uint8_t cpu_apicid_array[UINT8_MAX + 1]; 1875 1876 if (tp != NULL) { 1877 /* 1878 * Determine number of CPUs and keep track of "final" APIC ID 1879 * for each CPU by walking through ACPI MADT processor list 1880 */ 1881 end = (struct madt_processor *)(tp->hdr.len + (uintptr_t)tp); 1882 cpu = tp->list; 1883 while (cpu < end) { 1884 if (cpu->type == MADT_PROCESSOR) { 1885 if (cpu->flags & 1) { 1886 if (cpu_count < UINT8_MAX) 1887 cpu_apicid_array[cpu_count] = 1888 cpu->apic_id; 1889 cpu_count++; 1890 } 1891 } 1892 1893 cpu = (struct madt_processor *) 1894 (cpu->len + (uintptr_t)cpu); 1895 } 1896 1897 /* 1898 * Make boot property for array of "final" APIC IDs for each 1899 * CPU 1900 */ 1901 bsetprop(BP_CPU_APICID_ARRAY, strlen(BP_CPU_APICID_ARRAY), 1902 cpu_apicid_array, cpu_count * sizeof (uint8_t)); 1903 } 1904 1905 /* 1906 * User-set boot-ncpus overrides firmware count 1907 */ 1908 if (do_bsys_getproplen(NULL, "boot-ncpus") >= 0) 1909 return; 1910 1911 /* 1912 * Set boot property for boot-ncpus to number of CPUs given in MADT 1913 * if user hasn't set the property already 1914 */ 1915 if (tp != NULL) 1916 bsetpropsi("boot-ncpus", cpu_count); 1917 } 1918 1919 static void 1920 process_srat(struct srat *tp) 1921 { 1922 struct srat_item *item, *end; 1923 int i; 1924 int proc_num, mem_num; 1925 #pragma pack(1) 1926 struct { 1927 uint32_t domain; 1928 uint32_t apic_id; 1929 uint32_t sapic_id; 1930 } processor; 1931 struct { 1932 uint32_t domain; 1933 uint32_t x2apic_id; 1934 } x2apic; 1935 struct { 1936 uint32_t domain; 1937 uint64_t addr; 1938 uint64_t length; 1939 uint32_t flags; 1940 } memory; 1941 #pragma pack() 1942 char prop_name[30]; 1943 1944 if (tp == NULL) 1945 return; 1946 1947 proc_num = mem_num = 0; 1948 end = (struct srat_item *)(tp->hdr.len + (uintptr_t)tp); 1949 item = tp->list; 1950 while (item < end) { 1951 switch (item->type) { 1952 case SRAT_PROCESSOR: 1953 if (!(item->i.p.flags & SRAT_ENABLED)) 1954 break; 1955 processor.domain = item->i.p.domain1; 1956 for (i = 0; i < 3; i++) 1957 processor.domain += 1958 item->i.p.domain2[i] << ((i + 1) * 8); 1959 processor.apic_id = item->i.p.apic_id; 1960 processor.sapic_id = item->i.p.local_sapic_eid; 1961 (void) snprintf(prop_name, 30, "acpi-srat-processor-%d", 1962 proc_num); 1963 bsetprop(prop_name, strlen(prop_name), &processor, 1964 sizeof (processor)); 1965 proc_num++; 1966 break; 1967 case SRAT_MEMORY: 1968 if (!(item->i.m.flags & SRAT_ENABLED)) 1969 break; 1970 memory.domain = item->i.m.domain; 1971 memory.addr = item->i.m.base_addr; 1972 memory.length = item->i.m.len; 1973 memory.flags = item->i.m.flags; 1974 (void) snprintf(prop_name, 30, "acpi-srat-memory-%d", 1975 mem_num); 1976 bsetprop(prop_name, strlen(prop_name), &memory, 1977 sizeof (memory)); 1978 mem_num++; 1979 break; 1980 case SRAT_X2APIC: 1981 if (!(item->i.xp.flags & SRAT_ENABLED)) 1982 break; 1983 x2apic.domain = item->i.xp.domain; 1984 x2apic.x2apic_id = item->i.xp.x2apic_id; 1985 (void) snprintf(prop_name, 30, "acpi-srat-processor-%d", 1986 proc_num); 1987 bsetprop(prop_name, strlen(prop_name), &x2apic, 1988 sizeof (x2apic)); 1989 proc_num++; 1990 break; 1991 } 1992 1993 item = (struct srat_item *) 1994 (item->len + (caddr_t)item); 1995 } 1996 } 1997 1998 static void 1999 process_slit(struct slit *tp) 2000 { 2001 2002 /* 2003 * Check the number of localities; if it's too huge, we just 2004 * return and locality enumeration code will handle this later, 2005 * if possible. 2006 * 2007 * Note that the size of the table is the square of the 2008 * number of localities; if the number of localities exceeds 2009 * UINT16_MAX, the table size may overflow an int when being 2010 * passed to bsetprop() below. 2011 */ 2012 if (tp->number >= SLIT_LOCALITIES_MAX) 2013 return; 2014 2015 bsetprop(SLIT_NUM_PROPNAME, strlen(SLIT_NUM_PROPNAME), &tp->number, 2016 sizeof (tp->number)); 2017 bsetprop(SLIT_PROPNAME, strlen(SLIT_PROPNAME), &tp->entry, 2018 tp->number * tp->number); 2019 } 2020 #else /* __xpv */ 2021 static void 2022 enumerate_xen_cpus() 2023 { 2024 processorid_t id, max_id; 2025 2026 /* 2027 * User-set boot-ncpus overrides enumeration 2028 */ 2029 if (do_bsys_getproplen(NULL, "boot-ncpus") >= 0) 2030 return; 2031 2032 /* 2033 * Probe every possible virtual CPU id and remember the 2034 * highest id present; the count of CPUs is one greater 2035 * than this. This tacitly assumes at least cpu 0 is present. 2036 */ 2037 max_id = 0; 2038 for (id = 0; id < MAX_VIRT_CPUS; id++) 2039 if (HYPERVISOR_vcpu_op(VCPUOP_is_up, id, NULL) == 0) 2040 max_id = id; 2041 2042 bsetpropsi("boot-ncpus", max_id+1); 2043 2044 } 2045 #endif /* __xpv */ 2046 2047 static void 2048 build_firmware_properties(void) 2049 { 2050 #ifndef __xpv 2051 struct table_header *tp; 2052 2053 if ((tp = find_fw_table("APIC")) != NULL) 2054 process_madt((struct madt *)tp); 2055 2056 if ((srat_ptr = (struct srat *)find_fw_table("SRAT")) != NULL) 2057 process_srat(srat_ptr); 2058 2059 if (slit_ptr = (struct slit *)find_fw_table("SLIT")) 2060 process_slit(slit_ptr); 2061 #else /* __xpv */ 2062 enumerate_xen_cpus(); 2063 #endif /* __xpv */ 2064 } 2065 2066 /* 2067 * fake up a boot property for USB serial console early boot output 2068 */ 2069 void * 2070 usbser_init(size_t size) 2071 { 2072 static char *p = NULL; 2073 2074 p = do_bsys_alloc(NULL, NULL, size, MMU_PAGESIZE); 2075 *p = 0; 2076 bsetprop("usb-serial-buf", strlen("usb-serial-buf") + 1, 2077 &p, sizeof (p)); 2078 return (p); 2079 } 2080 2081 /*ARGSUSED*/ 2082 int 2083 boot_compinfo(int fd, struct compinfo *cbp) 2084 { 2085 cbp->iscmp = 0; 2086 cbp->blksize = MAXBSIZE; 2087 return (0); 2088 } 2089