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