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 2010 Sun Microsystems, Inc. All rights reserved. 24 * Use is subject to license terms. 25 */ 26 /* 27 * Copyright (c) 2010, Intel Corporation. 28 * All rights reserved. 29 */ 30 31 /* 32 * This file contains the functionality that mimics the boot operations 33 * on SPARC systems or the old boot.bin/multiboot programs on x86 systems. 34 * The x86 kernel now does everything on its own. 35 */ 36 37 #include <sys/types.h> 38 #include <sys/bootconf.h> 39 #include <sys/bootsvcs.h> 40 #include <sys/bootinfo.h> 41 #include <sys/multiboot.h> 42 #include <sys/bootvfs.h> 43 #include <sys/bootprops.h> 44 #include <sys/varargs.h> 45 #include <sys/param.h> 46 #include <sys/machparam.h> 47 #include <sys/machsystm.h> 48 #include <sys/archsystm.h> 49 #include <sys/boot_console.h> 50 #include <sys/cmn_err.h> 51 #include <sys/systm.h> 52 #include <sys/promif.h> 53 #include <sys/archsystm.h> 54 #include <sys/x86_archext.h> 55 #include <sys/kobj.h> 56 #include <sys/privregs.h> 57 #include <sys/sysmacros.h> 58 #include <sys/ctype.h> 59 #include <sys/fastboot.h> 60 #ifdef __xpv 61 #include <sys/hypervisor.h> 62 #include <net/if.h> 63 #endif 64 #include <vm/kboot_mmu.h> 65 #include <vm/hat_pte.h> 66 #include <sys/kobj.h> 67 #include <sys/kobj_lex.h> 68 #include <sys/pci_cfgspace_impl.h> 69 #include "acpi_fw.h" 70 71 static int have_console = 0; /* set once primitive console is initialized */ 72 static char *boot_args = ""; 73 74 /* 75 * Debugging macros 76 */ 77 static uint_t kbm_debug = 0; 78 #define DBG_MSG(s) { if (kbm_debug) bop_printf(NULL, "%s", s); } 79 #define DBG(x) { if (kbm_debug) \ 80 bop_printf(NULL, "%s is %" PRIx64 "\n", #x, (uint64_t)(x)); \ 81 } 82 83 #define PUT_STRING(s) { \ 84 char *cp; \ 85 for (cp = (s); *cp; ++cp) \ 86 bcons_putchar(*cp); \ 87 } 88 89 struct xboot_info *xbootp; /* boot info from "glue" code in low memory */ 90 bootops_t bootop; /* simple bootops we'll pass on to kernel */ 91 struct bsys_mem bm; 92 93 static uintptr_t next_virt; /* next available virtual address */ 94 static paddr_t next_phys; /* next available physical address from dboot */ 95 static paddr_t high_phys = -(paddr_t)1; /* last used physical address */ 96 97 /* 98 * buffer for vsnprintf for console I/O 99 */ 100 #define BUFFERSIZE 256 101 static char buffer[BUFFERSIZE]; 102 /* 103 * stuff to store/report/manipulate boot property settings. 104 */ 105 typedef struct bootprop { 106 struct bootprop *bp_next; 107 char *bp_name; 108 uint_t bp_vlen; 109 char *bp_value; 110 } bootprop_t; 111 112 static bootprop_t *bprops = NULL; 113 static char *curr_page = NULL; /* ptr to avail bprop memory */ 114 static int curr_space = 0; /* amount of memory at curr_page */ 115 116 #ifdef __xpv 117 start_info_t *xen_info; 118 shared_info_t *HYPERVISOR_shared_info; 119 #endif 120 121 /* 122 * some allocator statistics 123 */ 124 static ulong_t total_bop_alloc_scratch = 0; 125 static ulong_t total_bop_alloc_kernel = 0; 126 127 static void build_firmware_properties(void); 128 129 static int early_allocation = 1; 130 131 int force_fastreboot = 0; 132 volatile int fastreboot_onpanic = 0; 133 int post_fastreboot = 0; 134 #ifdef __xpv 135 volatile int fastreboot_capable = 0; 136 #else 137 volatile int fastreboot_capable = 1; 138 #endif 139 140 /* 141 * Information saved from current boot for fast reboot. 142 * If the information size exceeds what we have allocated, fast reboot 143 * will not be supported. 144 */ 145 multiboot_info_t saved_mbi; 146 mb_memory_map_t saved_mmap[FASTBOOT_SAVED_MMAP_COUNT]; 147 uint8_t saved_drives[FASTBOOT_SAVED_DRIVES_SIZE]; 148 char saved_cmdline[FASTBOOT_SAVED_CMDLINE_LEN]; 149 int saved_cmdline_len = 0; 150 size_t saved_file_size[FASTBOOT_MAX_FILES_MAP]; 151 152 /* 153 * Turn off fastreboot_onpanic to avoid panic loop. 154 */ 155 char fastreboot_onpanic_cmdline[FASTBOOT_SAVED_CMDLINE_LEN]; 156 static const char fastreboot_onpanic_args[] = " -B fastreboot_onpanic=0"; 157 158 /* 159 * Pointers to where System Resource Affinity Table (SRAT), System Locality 160 * Information Table (SLIT) and Maximum System Capability Table (MSCT) 161 * are mapped into virtual memory 162 */ 163 struct srat *srat_ptr = NULL; 164 struct slit *slit_ptr = NULL; 165 struct msct *msct_ptr = NULL; 166 167 /* 168 * Allocate aligned physical memory at boot time. This allocator allocates 169 * from the highest possible addresses. This avoids exhausting memory that 170 * would be useful for DMA buffers. 171 */ 172 paddr_t 173 do_bop_phys_alloc(uint64_t size, uint64_t align) 174 { 175 paddr_t pa = 0; 176 paddr_t start; 177 paddr_t end; 178 struct memlist *ml = (struct memlist *)xbootp->bi_phys_install; 179 180 /* 181 * Be careful if high memory usage is limited in startup.c 182 * Since there are holes in the low part of the physical address 183 * space we can treat physmem as a pfn (not just a pgcnt) and 184 * get a conservative upper limit. 185 */ 186 if (physmem != 0 && high_phys > pfn_to_pa(physmem)) 187 high_phys = pfn_to_pa(physmem); 188 189 /* 190 * find the lowest or highest available memory in physinstalled 191 * On 32 bit avoid physmem above 4Gig if PAE isn't enabled 192 */ 193 #if defined(__i386) 194 if (xbootp->bi_use_pae == 0 && high_phys > FOUR_GIG) 195 high_phys = FOUR_GIG; 196 #endif 197 198 /* 199 * find the highest available memory in physinstalled 200 */ 201 size = P2ROUNDUP(size, align); 202 for (; ml; ml = ml->ml_next) { 203 start = P2ROUNDUP(ml->ml_address, align); 204 end = P2ALIGN(ml->ml_address + ml->ml_size, align); 205 if (start < next_phys) 206 start = P2ROUNDUP(next_phys, align); 207 if (end > high_phys) 208 end = P2ALIGN(high_phys, align); 209 210 if (end <= start) 211 continue; 212 if (end - start < size) 213 continue; 214 215 /* 216 * Early allocations need to use low memory, since 217 * physmem might be further limited by bootenv.rc 218 */ 219 if (early_allocation) { 220 if (pa == 0 || start < pa) 221 pa = start; 222 } else { 223 if (end - size > pa) 224 pa = end - size; 225 } 226 } 227 if (pa != 0) { 228 if (early_allocation) 229 next_phys = pa + size; 230 else 231 high_phys = pa; 232 return (pa); 233 } 234 bop_panic("do_bop_phys_alloc(0x%" PRIx64 ", 0x%" PRIx64 235 ") Out of memory\n", size, align); 236 /*NOTREACHED*/ 237 } 238 239 uintptr_t 240 alloc_vaddr(size_t size, paddr_t align) 241 { 242 uintptr_t rv; 243 244 next_virt = P2ROUNDUP(next_virt, (uintptr_t)align); 245 rv = (uintptr_t)next_virt; 246 next_virt += size; 247 return (rv); 248 } 249 250 /* 251 * Allocate virtual memory. The size is always rounded up to a multiple 252 * of base pagesize. 253 */ 254 255 /*ARGSUSED*/ 256 static caddr_t 257 do_bsys_alloc(bootops_t *bop, caddr_t virthint, size_t size, int align) 258 { 259 paddr_t a = align; /* same type as pa for masking */ 260 uint_t pgsize; 261 paddr_t pa; 262 uintptr_t va; 263 ssize_t s; /* the aligned size */ 264 uint_t level; 265 uint_t is_kernel = (virthint != 0); 266 267 if (a < MMU_PAGESIZE) 268 a = MMU_PAGESIZE; 269 else if (!ISP2(a)) 270 prom_panic("do_bsys_alloc() incorrect alignment"); 271 size = P2ROUNDUP(size, MMU_PAGESIZE); 272 273 /* 274 * Use the next aligned virtual address if we weren't given one. 275 */ 276 if (virthint == NULL) { 277 virthint = (caddr_t)alloc_vaddr(size, a); 278 total_bop_alloc_scratch += size; 279 } else { 280 total_bop_alloc_kernel += size; 281 } 282 283 /* 284 * allocate the physical memory 285 */ 286 pa = do_bop_phys_alloc(size, a); 287 288 /* 289 * Add the mappings to the page tables, try large pages first. 290 */ 291 va = (uintptr_t)virthint; 292 s = size; 293 level = 1; 294 pgsize = xbootp->bi_use_pae ? TWO_MEG : FOUR_MEG; 295 if (xbootp->bi_use_largepage && a == pgsize) { 296 while (IS_P2ALIGNED(pa, pgsize) && IS_P2ALIGNED(va, pgsize) && 297 s >= pgsize) { 298 kbm_map(va, pa, level, is_kernel); 299 va += pgsize; 300 pa += pgsize; 301 s -= pgsize; 302 } 303 } 304 305 /* 306 * Map remaining pages use small mappings 307 */ 308 level = 0; 309 pgsize = MMU_PAGESIZE; 310 while (s > 0) { 311 kbm_map(va, pa, level, is_kernel); 312 va += pgsize; 313 pa += pgsize; 314 s -= pgsize; 315 } 316 return (virthint); 317 } 318 319 /* 320 * Free virtual memory - we'll just ignore these. 321 */ 322 /*ARGSUSED*/ 323 static void 324 do_bsys_free(bootops_t *bop, caddr_t virt, size_t size) 325 { 326 bop_printf(NULL, "do_bsys_free(virt=0x%p, size=0x%lx) ignored\n", 327 (void *)virt, size); 328 } 329 330 /* 331 * Old interface 332 */ 333 /*ARGSUSED*/ 334 static caddr_t 335 do_bsys_ealloc( 336 bootops_t *bop, 337 caddr_t virthint, 338 size_t size, 339 int align, 340 int flags) 341 { 342 prom_panic("unsupported call to BOP_EALLOC()\n"); 343 return (0); 344 } 345 346 347 static void 348 bsetprop(char *name, int nlen, void *value, int vlen) 349 { 350 uint_t size; 351 uint_t need_size; 352 bootprop_t *b; 353 354 /* 355 * align the size to 16 byte boundary 356 */ 357 size = sizeof (bootprop_t) + nlen + 1 + vlen; 358 size = (size + 0xf) & ~0xf; 359 if (size > curr_space) { 360 need_size = (size + (MMU_PAGEOFFSET)) & MMU_PAGEMASK; 361 curr_page = do_bsys_alloc(NULL, 0, need_size, MMU_PAGESIZE); 362 curr_space = need_size; 363 } 364 365 /* 366 * use a bootprop_t at curr_page and link into list 367 */ 368 b = (bootprop_t *)curr_page; 369 curr_page += sizeof (bootprop_t); 370 curr_space -= sizeof (bootprop_t); 371 b->bp_next = bprops; 372 bprops = b; 373 374 /* 375 * follow by name and ending zero byte 376 */ 377 b->bp_name = curr_page; 378 bcopy(name, curr_page, nlen); 379 curr_page += nlen; 380 *curr_page++ = 0; 381 curr_space -= nlen + 1; 382 383 /* 384 * copy in value, but no ending zero byte 385 */ 386 b->bp_value = curr_page; 387 b->bp_vlen = vlen; 388 if (vlen > 0) { 389 bcopy(value, curr_page, vlen); 390 curr_page += vlen; 391 curr_space -= vlen; 392 } 393 394 /* 395 * align new values of curr_page, curr_space 396 */ 397 while (curr_space & 0xf) { 398 ++curr_page; 399 --curr_space; 400 } 401 } 402 403 static void 404 bsetprops(char *name, char *value) 405 { 406 bsetprop(name, strlen(name), value, strlen(value) + 1); 407 } 408 409 static void 410 bsetprop64(char *name, uint64_t value) 411 { 412 bsetprop(name, strlen(name), (void *)&value, sizeof (value)); 413 } 414 415 static void 416 bsetpropsi(char *name, int value) 417 { 418 char prop_val[32]; 419 420 (void) snprintf(prop_val, sizeof (prop_val), "%d", value); 421 bsetprops(name, prop_val); 422 } 423 424 /* 425 * to find the size of the buffer to allocate 426 */ 427 /*ARGSUSED*/ 428 int 429 do_bsys_getproplen(bootops_t *bop, const char *name) 430 { 431 bootprop_t *b; 432 433 for (b = bprops; b; b = b->bp_next) { 434 if (strcmp(name, b->bp_name) != 0) 435 continue; 436 return (b->bp_vlen); 437 } 438 return (-1); 439 } 440 441 /* 442 * get the value associated with this name 443 */ 444 /*ARGSUSED*/ 445 int 446 do_bsys_getprop(bootops_t *bop, const char *name, void *value) 447 { 448 bootprop_t *b; 449 450 for (b = bprops; b; b = b->bp_next) { 451 if (strcmp(name, b->bp_name) != 0) 452 continue; 453 bcopy(b->bp_value, value, b->bp_vlen); 454 return (0); 455 } 456 return (-1); 457 } 458 459 /* 460 * get the name of the next property in succession from the standalone 461 */ 462 /*ARGSUSED*/ 463 static char * 464 do_bsys_nextprop(bootops_t *bop, char *name) 465 { 466 bootprop_t *b; 467 468 /* 469 * A null name is a special signal for the 1st boot property 470 */ 471 if (name == NULL || strlen(name) == 0) { 472 if (bprops == NULL) 473 return (NULL); 474 return (bprops->bp_name); 475 } 476 477 for (b = bprops; b; b = b->bp_next) { 478 if (name != b->bp_name) 479 continue; 480 b = b->bp_next; 481 if (b == NULL) 482 return (NULL); 483 return (b->bp_name); 484 } 485 return (NULL); 486 } 487 488 /* 489 * Parse numeric value from a string. Understands decimal, hex, octal, - and ~ 490 */ 491 static int 492 parse_value(char *p, uint64_t *retval) 493 { 494 int adjust = 0; 495 uint64_t tmp = 0; 496 int digit; 497 int radix = 10; 498 499 *retval = 0; 500 if (*p == '-' || *p == '~') 501 adjust = *p++; 502 503 if (*p == '0') { 504 ++p; 505 if (*p == 0) 506 return (0); 507 if (*p == 'x' || *p == 'X') { 508 radix = 16; 509 ++p; 510 } else { 511 radix = 8; 512 ++p; 513 } 514 } 515 while (*p) { 516 if ('0' <= *p && *p <= '9') 517 digit = *p - '0'; 518 else if ('a' <= *p && *p <= 'f') 519 digit = 10 + *p - 'a'; 520 else if ('A' <= *p && *p <= 'F') 521 digit = 10 + *p - 'A'; 522 else 523 return (-1); 524 if (digit >= radix) 525 return (-1); 526 tmp = tmp * radix + digit; 527 ++p; 528 } 529 if (adjust == '-') 530 tmp = -tmp; 531 else if (adjust == '~') 532 tmp = ~tmp; 533 *retval = tmp; 534 return (0); 535 } 536 537 /* 538 * 2nd part of building the table of boot properties. This includes: 539 * - values from /boot/solaris/bootenv.rc (ie. eeprom(1m) values) 540 * 541 * lines look like one of: 542 * ^$ 543 * ^# comment till end of line 544 * setprop name 'value' 545 * setprop name value 546 * setprop name "value" 547 * 548 * we do single character I/O since this is really just looking at memory 549 */ 550 void 551 boot_prop_finish(void) 552 { 553 int fd; 554 char *line; 555 int c; 556 int bytes_read; 557 char *name; 558 int n_len; 559 char *value; 560 int v_len; 561 char *inputdev; /* these override the command line if serial ports */ 562 char *outputdev; 563 char *consoledev; 564 uint64_t lvalue; 565 int use_xencons = 0; 566 567 #ifdef __xpv 568 if (!DOMAIN_IS_INITDOMAIN(xen_info)) 569 use_xencons = 1; 570 #endif /* __xpv */ 571 572 DBG_MSG("Opening /boot/solaris/bootenv.rc\n"); 573 fd = BRD_OPEN(bfs_ops, "/boot/solaris/bootenv.rc", 0); 574 DBG(fd); 575 576 line = do_bsys_alloc(NULL, NULL, MMU_PAGESIZE, MMU_PAGESIZE); 577 while (fd >= 0) { 578 579 /* 580 * get a line 581 */ 582 for (c = 0; ; ++c) { 583 bytes_read = BRD_READ(bfs_ops, fd, line + c, 1); 584 if (bytes_read == 0) { 585 if (c == 0) 586 goto done; 587 break; 588 } 589 if (line[c] == '\n') 590 break; 591 } 592 line[c] = 0; 593 594 /* 595 * ignore comment lines 596 */ 597 c = 0; 598 while (ISSPACE(line[c])) 599 ++c; 600 if (line[c] == '#' || line[c] == 0) 601 continue; 602 603 /* 604 * must have "setprop " or "setprop\t" 605 */ 606 if (strncmp(line + c, "setprop ", 8) != 0 && 607 strncmp(line + c, "setprop\t", 8) != 0) 608 continue; 609 c += 8; 610 while (ISSPACE(line[c])) 611 ++c; 612 if (line[c] == 0) 613 continue; 614 615 /* 616 * gather up the property name 617 */ 618 name = line + c; 619 n_len = 0; 620 while (line[c] && !ISSPACE(line[c])) 621 ++n_len, ++c; 622 623 /* 624 * gather up the value, if any 625 */ 626 value = ""; 627 v_len = 0; 628 while (ISSPACE(line[c])) 629 ++c; 630 if (line[c] != 0) { 631 value = line + c; 632 while (line[c] && !ISSPACE(line[c])) 633 ++v_len, ++c; 634 } 635 636 if (v_len >= 2 && value[0] == value[v_len - 1] && 637 (value[0] == '\'' || value[0] == '"')) { 638 ++value; 639 v_len -= 2; 640 } 641 name[n_len] = 0; 642 if (v_len > 0) 643 value[v_len] = 0; 644 else 645 continue; 646 647 /* 648 * ignore "boot-file" property, it's now meaningless 649 */ 650 if (strcmp(name, "boot-file") == 0) 651 continue; 652 if (strcmp(name, "boot-args") == 0 && 653 strlen(boot_args) > 0) 654 continue; 655 656 /* 657 * If a property was explicitly set on the command line 658 * it will override a setting in bootenv.rc 659 */ 660 if (do_bsys_getproplen(NULL, name) > 0) 661 continue; 662 663 bsetprop(name, n_len, value, v_len + 1); 664 } 665 done: 666 if (fd >= 0) 667 BRD_CLOSE(bfs_ops, fd); 668 669 /* 670 * Check if we have to limit the boot time allocator 671 */ 672 if (do_bsys_getproplen(NULL, "physmem") != -1 && 673 do_bsys_getprop(NULL, "physmem", line) >= 0 && 674 parse_value(line, &lvalue) != -1) { 675 if (0 < lvalue && (lvalue < physmem || physmem == 0)) { 676 physmem = (pgcnt_t)lvalue; 677 DBG(physmem); 678 } 679 } 680 early_allocation = 0; 681 682 /* 683 * check to see if we have to override the default value of the console 684 */ 685 if (!use_xencons) { 686 inputdev = line; 687 v_len = do_bsys_getproplen(NULL, "input-device"); 688 if (v_len > 0) 689 (void) do_bsys_getprop(NULL, "input-device", inputdev); 690 else 691 v_len = 0; 692 inputdev[v_len] = 0; 693 694 outputdev = inputdev + v_len + 1; 695 v_len = do_bsys_getproplen(NULL, "output-device"); 696 if (v_len > 0) 697 (void) do_bsys_getprop(NULL, "output-device", 698 outputdev); 699 else 700 v_len = 0; 701 outputdev[v_len] = 0; 702 703 consoledev = outputdev + v_len + 1; 704 v_len = do_bsys_getproplen(NULL, "console"); 705 if (v_len > 0) { 706 (void) do_bsys_getprop(NULL, "console", consoledev); 707 if (post_fastreboot && 708 strcmp(consoledev, "graphics") == 0) { 709 bsetprops("console", "text"); 710 v_len = strlen("text"); 711 bcopy("text", consoledev, v_len); 712 } 713 } else { 714 v_len = 0; 715 } 716 consoledev[v_len] = 0; 717 bcons_init2(inputdev, outputdev, consoledev); 718 } else { 719 /* 720 * Ensure console property exists 721 * If not create it as "hypervisor" 722 */ 723 v_len = do_bsys_getproplen(NULL, "console"); 724 if (v_len < 0) 725 bsetprops("console", "hypervisor"); 726 inputdev = outputdev = consoledev = "hypervisor"; 727 bcons_init2(inputdev, outputdev, consoledev); 728 } 729 730 if (strstr((char *)xbootp->bi_cmdline, "prom_debug") || kbm_debug) { 731 value = line; 732 bop_printf(NULL, "\nBoot properties:\n"); 733 name = ""; 734 while ((name = do_bsys_nextprop(NULL, name)) != NULL) { 735 bop_printf(NULL, "\t0x%p %s = ", (void *)name, name); 736 (void) do_bsys_getprop(NULL, name, value); 737 v_len = do_bsys_getproplen(NULL, name); 738 bop_printf(NULL, "len=%d ", v_len); 739 value[v_len] = 0; 740 bop_printf(NULL, "%s\n", value); 741 } 742 } 743 } 744 745 /* 746 * print formatted output 747 */ 748 /*PRINTFLIKE2*/ 749 /*ARGSUSED*/ 750 void 751 bop_printf(bootops_t *bop, const char *fmt, ...) 752 { 753 va_list ap; 754 755 if (have_console == 0) 756 return; 757 758 va_start(ap, fmt); 759 (void) vsnprintf(buffer, BUFFERSIZE, fmt, ap); 760 va_end(ap); 761 PUT_STRING(buffer); 762 } 763 764 /* 765 * Another panic() variant; this one can be used even earlier during boot than 766 * prom_panic(). 767 */ 768 /*PRINTFLIKE1*/ 769 void 770 bop_panic(const char *fmt, ...) 771 { 772 va_list ap; 773 774 va_start(ap, fmt); 775 bop_printf(NULL, fmt, ap); 776 va_end(ap); 777 778 bop_printf(NULL, "\nPress any key to reboot.\n"); 779 (void) bcons_getchar(); 780 bop_printf(NULL, "Resetting...\n"); 781 pc_reset(); 782 } 783 784 /* 785 * Do a real mode interrupt BIOS call 786 */ 787 typedef struct bios_regs { 788 unsigned short ax, bx, cx, dx, si, di, bp, es, ds; 789 } bios_regs_t; 790 typedef int (*bios_func_t)(int, bios_regs_t *); 791 792 /*ARGSUSED*/ 793 static void 794 do_bsys_doint(bootops_t *bop, int intnum, struct bop_regs *rp) 795 { 796 #if defined(__xpv) 797 prom_panic("unsupported call to BOP_DOINT()\n"); 798 #else /* __xpv */ 799 static int firsttime = 1; 800 bios_func_t bios_func = (bios_func_t)(void *)(uintptr_t)0x5000; 801 bios_regs_t br; 802 803 /* 804 * The first time we do this, we have to copy the pre-packaged 805 * low memory bios call code image into place. 806 */ 807 if (firsttime) { 808 extern char bios_image[]; 809 extern uint32_t bios_size; 810 811 bcopy(bios_image, (void *)bios_func, bios_size); 812 firsttime = 0; 813 } 814 815 br.ax = rp->eax.word.ax; 816 br.bx = rp->ebx.word.bx; 817 br.cx = rp->ecx.word.cx; 818 br.dx = rp->edx.word.dx; 819 br.bp = rp->ebp.word.bp; 820 br.si = rp->esi.word.si; 821 br.di = rp->edi.word.di; 822 br.ds = rp->ds; 823 br.es = rp->es; 824 825 DBG_MSG("Doing BIOS call..."); 826 DBG(br.ax); 827 DBG(br.bx); 828 DBG(br.dx); 829 rp->eflags = bios_func(intnum, &br); 830 DBG_MSG("done\n"); 831 832 rp->eax.word.ax = br.ax; 833 rp->ebx.word.bx = br.bx; 834 rp->ecx.word.cx = br.cx; 835 rp->edx.word.dx = br.dx; 836 rp->ebp.word.bp = br.bp; 837 rp->esi.word.si = br.si; 838 rp->edi.word.di = br.di; 839 rp->ds = br.ds; 840 rp->es = br.es; 841 #endif /* __xpv */ 842 } 843 844 static struct boot_syscalls bop_sysp = { 845 bcons_getchar, 846 bcons_putchar, 847 bcons_ischar, 848 }; 849 850 static char *whoami; 851 852 #define BUFLEN 64 853 854 #if defined(__xpv) 855 856 static char namebuf[32]; 857 858 static void 859 xen_parse_props(char *s, char *prop_map[], int n_prop) 860 { 861 char **prop_name = prop_map; 862 char *cp = s, *scp; 863 864 do { 865 scp = cp; 866 while ((*cp != NULL) && (*cp != ':')) 867 cp++; 868 869 if ((scp != cp) && (*prop_name != NULL)) { 870 *cp = NULL; 871 bsetprops(*prop_name, scp); 872 } 873 874 cp++; 875 prop_name++; 876 n_prop--; 877 } while (n_prop > 0); 878 } 879 880 #define VBDPATHLEN 64 881 882 /* 883 * parse the 'xpv-root' property to create properties used by 884 * ufs_mountroot. 885 */ 886 static void 887 xen_vbdroot_props(char *s) 888 { 889 char vbdpath[VBDPATHLEN] = "/xpvd/xdf@"; 890 const char lnamefix[] = "/dev/dsk/c0d"; 891 char *pnp; 892 char *prop_p; 893 char mi; 894 short minor; 895 long addr = 0; 896 897 pnp = vbdpath + strlen(vbdpath); 898 prop_p = s + strlen(lnamefix); 899 while ((*prop_p != '\0') && (*prop_p != 's') && (*prop_p != 'p')) 900 addr = addr * 10 + *prop_p++ - '0'; 901 (void) snprintf(pnp, VBDPATHLEN, "%lx", addr); 902 pnp = vbdpath + strlen(vbdpath); 903 if (*prop_p == 's') 904 mi = 'a'; 905 else if (*prop_p == 'p') 906 mi = 'q'; 907 else 908 ASSERT(0); /* shouldn't be here */ 909 prop_p++; 910 ASSERT(*prop_p != '\0'); 911 if (ISDIGIT(*prop_p)) { 912 minor = *prop_p - '0'; 913 prop_p++; 914 if (ISDIGIT(*prop_p)) { 915 minor = minor * 10 + *prop_p - '0'; 916 } 917 } else { 918 /* malformed root path, use 0 as default */ 919 minor = 0; 920 } 921 ASSERT(minor < 16); /* at most 16 partitions */ 922 mi += minor; 923 *pnp++ = ':'; 924 *pnp++ = mi; 925 *pnp++ = '\0'; 926 bsetprops("fstype", "ufs"); 927 bsetprops("bootpath", vbdpath); 928 929 DBG_MSG("VBD bootpath set to "); 930 DBG_MSG(vbdpath); 931 DBG_MSG("\n"); 932 } 933 934 /* 935 * parse the xpv-nfsroot property to create properties used by 936 * nfs_mountroot. 937 */ 938 static void 939 xen_nfsroot_props(char *s) 940 { 941 char *prop_map[] = { 942 BP_SERVER_IP, /* server IP address */ 943 BP_SERVER_NAME, /* server hostname */ 944 BP_SERVER_PATH, /* root path */ 945 }; 946 int n_prop = sizeof (prop_map) / sizeof (prop_map[0]); 947 948 bsetprop("fstype", 6, "nfs", 4); 949 950 xen_parse_props(s, prop_map, n_prop); 951 952 /* 953 * If a server name wasn't specified, use a default. 954 */ 955 if (do_bsys_getproplen(NULL, BP_SERVER_NAME) == -1) 956 bsetprops(BP_SERVER_NAME, "unknown"); 957 } 958 959 /* 960 * Extract our IP address, etc. from the "xpv-ip" property. 961 */ 962 static void 963 xen_ip_props(char *s) 964 { 965 char *prop_map[] = { 966 BP_HOST_IP, /* IP address */ 967 NULL, /* NFS server IP address (ignored in */ 968 /* favour of xpv-nfsroot) */ 969 BP_ROUTER_IP, /* IP gateway */ 970 BP_SUBNET_MASK, /* IP subnet mask */ 971 "xpv-hostname", /* hostname (ignored) */ 972 BP_NETWORK_INTERFACE, /* interface name */ 973 "xpv-hcp", /* host configuration protocol */ 974 }; 975 int n_prop = sizeof (prop_map) / sizeof (prop_map[0]); 976 char ifname[IFNAMSIZ]; 977 978 xen_parse_props(s, prop_map, n_prop); 979 980 /* 981 * A Linux dom0 administrator expects all interfaces to be 982 * called "ethX", which is not the case here. 983 * 984 * If the interface name specified is "eth0", presume that 985 * this is really intended to be "xnf0" (the first domU -> 986 * dom0 interface for this domain). 987 */ 988 if ((do_bsys_getprop(NULL, BP_NETWORK_INTERFACE, ifname) == 0) && 989 (strcmp("eth0", ifname) == 0)) { 990 bsetprops(BP_NETWORK_INTERFACE, "xnf0"); 991 bop_printf(NULL, 992 "network interface name 'eth0' replaced with 'xnf0'\n"); 993 } 994 } 995 996 #else /* __xpv */ 997 998 static void 999 setup_rarp_props(struct sol_netinfo *sip) 1000 { 1001 char buf[BUFLEN]; /* to hold ip/mac addrs */ 1002 uint8_t *val; 1003 1004 val = (uint8_t *)&sip->sn_ciaddr; 1005 (void) snprintf(buf, BUFLEN, "%d.%d.%d.%d", 1006 val[0], val[1], val[2], val[3]); 1007 bsetprops(BP_HOST_IP, buf); 1008 1009 val = (uint8_t *)&sip->sn_siaddr; 1010 (void) snprintf(buf, BUFLEN, "%d.%d.%d.%d", 1011 val[0], val[1], val[2], val[3]); 1012 bsetprops(BP_SERVER_IP, buf); 1013 1014 if (sip->sn_giaddr != 0) { 1015 val = (uint8_t *)&sip->sn_giaddr; 1016 (void) snprintf(buf, BUFLEN, "%d.%d.%d.%d", 1017 val[0], val[1], val[2], val[3]); 1018 bsetprops(BP_ROUTER_IP, buf); 1019 } 1020 1021 if (sip->sn_netmask != 0) { 1022 val = (uint8_t *)&sip->sn_netmask; 1023 (void) snprintf(buf, BUFLEN, "%d.%d.%d.%d", 1024 val[0], val[1], val[2], val[3]); 1025 bsetprops(BP_SUBNET_MASK, buf); 1026 } 1027 1028 if (sip->sn_mactype != 4 || sip->sn_maclen != 6) { 1029 bop_printf(NULL, "unsupported mac type %d, mac len %d\n", 1030 sip->sn_mactype, sip->sn_maclen); 1031 } else { 1032 val = sip->sn_macaddr; 1033 (void) snprintf(buf, BUFLEN, "%x:%x:%x:%x:%x:%x", 1034 val[0], val[1], val[2], val[3], val[4], val[5]); 1035 bsetprops(BP_BOOT_MAC, buf); 1036 } 1037 } 1038 1039 #endif /* __xpv */ 1040 1041 static void 1042 build_panic_cmdline(const char *cmd, int cmdlen) 1043 { 1044 int proplen; 1045 size_t arglen; 1046 1047 arglen = sizeof (fastreboot_onpanic_args); 1048 /* 1049 * If we allready have fastreboot-onpanic set to zero, 1050 * don't add them again. 1051 */ 1052 if ((proplen = do_bsys_getproplen(NULL, FASTREBOOT_ONPANIC)) > 0 && 1053 proplen <= sizeof (fastreboot_onpanic_cmdline)) { 1054 (void) do_bsys_getprop(NULL, FASTREBOOT_ONPANIC, 1055 fastreboot_onpanic_cmdline); 1056 if (FASTREBOOT_ONPANIC_NOTSET(fastreboot_onpanic_cmdline)) 1057 arglen = 1; 1058 } 1059 1060 /* 1061 * construct fastreboot_onpanic_cmdline 1062 */ 1063 if (cmdlen + arglen > sizeof (fastreboot_onpanic_cmdline)) { 1064 DBG_MSG("Command line too long: clearing " 1065 FASTREBOOT_ONPANIC "\n"); 1066 fastreboot_onpanic = 0; 1067 } else { 1068 bcopy(cmd, fastreboot_onpanic_cmdline, cmdlen); 1069 if (arglen != 1) 1070 bcopy(fastreboot_onpanic_args, 1071 fastreboot_onpanic_cmdline + cmdlen, arglen); 1072 else 1073 fastreboot_onpanic_cmdline[cmdlen] = 0; 1074 } 1075 } 1076 1077 1078 #ifndef __xpv 1079 /* 1080 * Construct boot command line for Fast Reboot 1081 */ 1082 static void 1083 build_fastboot_cmdline(void) 1084 { 1085 saved_cmdline_len = strlen(xbootp->bi_cmdline) + 1; 1086 if (saved_cmdline_len > FASTBOOT_SAVED_CMDLINE_LEN) { 1087 DBG(saved_cmdline_len); 1088 DBG_MSG("Command line too long: clearing fastreboot_capable\n"); 1089 fastreboot_capable = 0; 1090 } else { 1091 bcopy((void *)(xbootp->bi_cmdline), (void *)saved_cmdline, 1092 saved_cmdline_len); 1093 saved_cmdline[saved_cmdline_len - 1] = '\0'; 1094 build_panic_cmdline(saved_cmdline, saved_cmdline_len - 1); 1095 } 1096 } 1097 1098 /* 1099 * Save memory layout, disk drive information, unix and boot archive sizes for 1100 * Fast Reboot. 1101 */ 1102 static void 1103 save_boot_info(multiboot_info_t *mbi, struct xboot_info *xbi) 1104 { 1105 struct boot_modules *modp; 1106 int i; 1107 1108 bcopy(mbi, &saved_mbi, sizeof (multiboot_info_t)); 1109 if (mbi->mmap_length > sizeof (saved_mmap)) { 1110 DBG_MSG("mbi->mmap_length too big: clearing " 1111 "fastreboot_capable\n"); 1112 fastreboot_capable = 0; 1113 } else { 1114 bcopy((void *)(uintptr_t)mbi->mmap_addr, (void *)saved_mmap, 1115 mbi->mmap_length); 1116 } 1117 1118 if ((mbi->flags & MB_INFO_DRIVE_INFO) != 0) { 1119 if (mbi->drives_length > sizeof (saved_drives)) { 1120 DBG(mbi->drives_length); 1121 DBG_MSG("mbi->drives_length too big: clearing " 1122 "fastreboot_capable\n"); 1123 fastreboot_capable = 0; 1124 } else { 1125 bcopy((void *)(uintptr_t)mbi->drives_addr, 1126 (void *)saved_drives, mbi->drives_length); 1127 } 1128 } else { 1129 saved_mbi.drives_length = 0; 1130 saved_mbi.drives_addr = NULL; 1131 } 1132 1133 /* 1134 * Current file sizes. Used by fastboot.c to figure out how much 1135 * memory to reserve for panic reboot. 1136 * Use the module list from the dboot-constructed xboot_info 1137 * instead of the list referenced by the multiboot structure 1138 * because that structure may not be addressable now. 1139 */ 1140 saved_file_size[FASTBOOT_NAME_UNIX] = FOUR_MEG - PAGESIZE; 1141 for (i = 0, modp = (struct boot_modules *)(uintptr_t)xbi->bi_modules; 1142 i < xbi->bi_module_cnt; i++, modp++) { 1143 saved_file_size[FASTBOOT_NAME_BOOTARCHIVE] += modp->bm_size; 1144 } 1145 } 1146 #endif /* __xpv */ 1147 1148 1149 /* 1150 * 1st pass at building the table of boot properties. This includes: 1151 * - values set on the command line: -B a=x,b=y,c=z .... 1152 * - known values we just compute (ie. from xbootp) 1153 * - values from /boot/solaris/bootenv.rc (ie. eeprom(1m) values) 1154 * 1155 * the grub command line looked like: 1156 * kernel boot-file [-B prop=value[,prop=value]...] [boot-args] 1157 * 1158 * whoami is the same as boot-file 1159 */ 1160 static void 1161 build_boot_properties(void) 1162 { 1163 char *name; 1164 int name_len; 1165 char *value; 1166 int value_len; 1167 struct boot_modules *bm; 1168 char *propbuf; 1169 int quoted = 0; 1170 int boot_arg_len; 1171 #ifndef __xpv 1172 static int stdout_val = 0; 1173 uchar_t boot_device; 1174 char str[3]; 1175 multiboot_info_t *mbi; 1176 int netboot; 1177 struct sol_netinfo *sip; 1178 #endif 1179 1180 /* 1181 * These have to be done first, so that kobj_mount_root() works 1182 */ 1183 DBG_MSG("Building boot properties\n"); 1184 propbuf = do_bsys_alloc(NULL, NULL, MMU_PAGESIZE, 0); 1185 DBG((uintptr_t)propbuf); 1186 if (xbootp->bi_module_cnt > 0) { 1187 bm = xbootp->bi_modules; 1188 bsetprop64("ramdisk_start", (uint64_t)(uintptr_t)bm->bm_addr); 1189 bsetprop64("ramdisk_end", (uint64_t)(uintptr_t)bm->bm_addr + 1190 bm->bm_size); 1191 } 1192 1193 DBG_MSG("Parsing command line for boot properties\n"); 1194 value = xbootp->bi_cmdline; 1195 1196 /* 1197 * allocate memory to collect boot_args into 1198 */ 1199 boot_arg_len = strlen(xbootp->bi_cmdline) + 1; 1200 boot_args = do_bsys_alloc(NULL, NULL, boot_arg_len, MMU_PAGESIZE); 1201 boot_args[0] = 0; 1202 boot_arg_len = 0; 1203 1204 #ifdef __xpv 1205 /* 1206 * Xen puts a lot of device information in front of the kernel name 1207 * let's grab them and make them boot properties. The first 1208 * string w/o an "=" in it will be the boot-file property. 1209 */ 1210 (void) strcpy(namebuf, "xpv-"); 1211 for (;;) { 1212 /* 1213 * get to next property 1214 */ 1215 while (ISSPACE(*value)) 1216 ++value; 1217 name = value; 1218 /* 1219 * look for an "=" 1220 */ 1221 while (*value && !ISSPACE(*value) && *value != '=') { 1222 value++; 1223 } 1224 if (*value != '=') { /* no "=" in the property */ 1225 value = name; 1226 break; 1227 } 1228 name_len = value - name; 1229 value_len = 0; 1230 /* 1231 * skip over the "=" 1232 */ 1233 value++; 1234 while (value[value_len] && !ISSPACE(value[value_len])) { 1235 ++value_len; 1236 } 1237 /* 1238 * build property name with "xpv-" prefix 1239 */ 1240 if (name_len + 4 > 32) { /* skip if name too long */ 1241 value += value_len; 1242 continue; 1243 } 1244 bcopy(name, &namebuf[4], name_len); 1245 name_len += 4; 1246 namebuf[name_len] = 0; 1247 bcopy(value, propbuf, value_len); 1248 propbuf[value_len] = 0; 1249 bsetprops(namebuf, propbuf); 1250 1251 /* 1252 * xpv-root is set to the logical disk name of the xen 1253 * VBD when booting from a disk-based filesystem. 1254 */ 1255 if (strcmp(namebuf, "xpv-root") == 0) 1256 xen_vbdroot_props(propbuf); 1257 /* 1258 * While we're here, if we have a "xpv-nfsroot" property 1259 * then we need to set "fstype" to "nfs" so we mount 1260 * our root from the nfs server. Also parse the xpv-nfsroot 1261 * property to create the properties that nfs_mountroot will 1262 * need to find the root and mount it. 1263 */ 1264 if (strcmp(namebuf, "xpv-nfsroot") == 0) 1265 xen_nfsroot_props(propbuf); 1266 1267 if (strcmp(namebuf, "xpv-ip") == 0) 1268 xen_ip_props(propbuf); 1269 value += value_len; 1270 } 1271 #endif 1272 1273 while (ISSPACE(*value)) 1274 ++value; 1275 /* 1276 * value now points at the boot-file 1277 */ 1278 value_len = 0; 1279 while (value[value_len] && !ISSPACE(value[value_len])) 1280 ++value_len; 1281 if (value_len > 0) { 1282 whoami = propbuf; 1283 bcopy(value, whoami, value_len); 1284 whoami[value_len] = 0; 1285 bsetprops("boot-file", whoami); 1286 /* 1287 * strip leading path stuff from whoami, so running from 1288 * PXE/miniroot makes sense. 1289 */ 1290 if (strstr(whoami, "/platform/") != NULL) 1291 whoami = strstr(whoami, "/platform/"); 1292 bsetprops("whoami", whoami); 1293 } 1294 1295 /* 1296 * Values forcibly set boot properties on the command line via -B. 1297 * Allow use of quotes in values. Other stuff goes on kernel 1298 * command line. 1299 */ 1300 name = value + value_len; 1301 while (*name != 0) { 1302 /* 1303 * anything not " -B" is copied to the command line 1304 */ 1305 if (!ISSPACE(name[0]) || name[1] != '-' || name[2] != 'B') { 1306 boot_args[boot_arg_len++] = *name; 1307 boot_args[boot_arg_len] = 0; 1308 ++name; 1309 continue; 1310 } 1311 1312 /* 1313 * skip the " -B" and following white space 1314 */ 1315 name += 3; 1316 while (ISSPACE(*name)) 1317 ++name; 1318 while (*name && !ISSPACE(*name)) { 1319 value = strstr(name, "="); 1320 if (value == NULL) 1321 break; 1322 name_len = value - name; 1323 ++value; 1324 value_len = 0; 1325 quoted = 0; 1326 for (; ; ++value_len) { 1327 if (!value[value_len]) 1328 break; 1329 1330 /* 1331 * is this value quoted? 1332 */ 1333 if (value_len == 0 && 1334 (value[0] == '\'' || value[0] == '"')) { 1335 quoted = value[0]; 1336 ++value_len; 1337 } 1338 1339 /* 1340 * In the quote accept any character, 1341 * but look for ending quote. 1342 */ 1343 if (quoted) { 1344 if (value[value_len] == quoted) 1345 quoted = 0; 1346 continue; 1347 } 1348 1349 /* 1350 * a comma or white space ends the value 1351 */ 1352 if (value[value_len] == ',' || 1353 ISSPACE(value[value_len])) 1354 break; 1355 } 1356 1357 if (value_len == 0) { 1358 bsetprop(name, name_len, "true", 5); 1359 } else { 1360 char *v = value; 1361 int l = value_len; 1362 if (v[0] == v[l - 1] && 1363 (v[0] == '\'' || v[0] == '"')) { 1364 ++v; 1365 l -= 2; 1366 } 1367 bcopy(v, propbuf, l); 1368 propbuf[l] = '\0'; 1369 bsetprop(name, name_len, propbuf, 1370 l + 1); 1371 } 1372 name = value + value_len; 1373 while (*name == ',') 1374 ++name; 1375 } 1376 } 1377 1378 /* 1379 * set boot-args property 1380 * 1275 name is bootargs, so set 1381 * that too 1382 */ 1383 bsetprops("boot-args", boot_args); 1384 bsetprops("bootargs", boot_args); 1385 1386 #ifndef __xpv 1387 /* 1388 * set the BIOS boot device from GRUB 1389 */ 1390 netboot = 0; 1391 mbi = xbootp->bi_mb_info; 1392 1393 /* 1394 * Build boot command line for Fast Reboot 1395 */ 1396 build_fastboot_cmdline(); 1397 1398 /* 1399 * Save various boot information for Fast Reboot 1400 */ 1401 save_boot_info(mbi, xbootp); 1402 1403 if (mbi != NULL && mbi->flags & MB_INFO_BOOTDEV) { 1404 boot_device = mbi->boot_device >> 24; 1405 if (boot_device == 0x20) 1406 netboot++; 1407 str[0] = (boot_device >> 4) + '0'; 1408 str[1] = (boot_device & 0xf) + '0'; 1409 str[2] = 0; 1410 bsetprops("bios-boot-device", str); 1411 } else { 1412 netboot = 1; 1413 } 1414 1415 /* 1416 * In the netboot case, drives_info is overloaded with the dhcp ack. 1417 * This is not multiboot compliant and requires special pxegrub! 1418 */ 1419 if (netboot && mbi->drives_length != 0) { 1420 sip = (struct sol_netinfo *)(uintptr_t)mbi->drives_addr; 1421 if (sip->sn_infotype == SN_TYPE_BOOTP) 1422 bsetprop("bootp-response", sizeof ("bootp-response"), 1423 (void *)(uintptr_t)mbi->drives_addr, 1424 mbi->drives_length); 1425 else if (sip->sn_infotype == SN_TYPE_RARP) 1426 setup_rarp_props(sip); 1427 } 1428 bsetprop("stdout", strlen("stdout"), 1429 &stdout_val, sizeof (stdout_val)); 1430 #endif /* __xpv */ 1431 1432 /* 1433 * more conjured up values for made up things.... 1434 */ 1435 #if defined(__xpv) 1436 bsetprops("mfg-name", "i86xpv"); 1437 bsetprops("impl-arch-name", "i86xpv"); 1438 #else 1439 bsetprops("mfg-name", "i86pc"); 1440 bsetprops("impl-arch-name", "i86pc"); 1441 #endif 1442 1443 /* 1444 * Build firmware-provided system properties 1445 */ 1446 build_firmware_properties(); 1447 1448 /* 1449 * XXPV 1450 * 1451 * Find out what these are: 1452 * - cpuid_feature_ecx_include 1453 * - cpuid_feature_ecx_exclude 1454 * - cpuid_feature_edx_include 1455 * - cpuid_feature_edx_exclude 1456 * 1457 * Find out what these are in multiboot: 1458 * - netdev-path 1459 * - fstype 1460 */ 1461 } 1462 1463 #ifdef __xpv 1464 /* 1465 * Under the Hypervisor, memory usable for DMA may be scarce. One 1466 * very likely large pool of DMA friendly memory is occupied by 1467 * the boot_archive, as it was loaded by grub into low MFNs. 1468 * 1469 * Here we free up that memory by copying the boot archive to what are 1470 * likely higher MFN pages and then swapping the mfn/pfn mappings. 1471 */ 1472 #define PFN_2GIG 0x80000 1473 static void 1474 relocate_boot_archive(void) 1475 { 1476 mfn_t max_mfn = HYPERVISOR_memory_op(XENMEM_maximum_ram_page, NULL); 1477 struct boot_modules *bm = xbootp->bi_modules; 1478 uintptr_t va; 1479 pfn_t va_pfn; 1480 mfn_t va_mfn; 1481 caddr_t copy; 1482 pfn_t copy_pfn; 1483 mfn_t copy_mfn; 1484 size_t len; 1485 int slop; 1486 int total = 0; 1487 int relocated = 0; 1488 int mmu_update_return; 1489 mmu_update_t t[2]; 1490 x86pte_t pte; 1491 1492 /* 1493 * If all MFN's are below 2Gig, don't bother doing this. 1494 */ 1495 if (max_mfn < PFN_2GIG) 1496 return; 1497 if (xbootp->bi_module_cnt < 1) { 1498 DBG_MSG("no boot_archive!"); 1499 return; 1500 } 1501 1502 DBG_MSG("moving boot_archive to high MFN memory\n"); 1503 va = (uintptr_t)bm->bm_addr; 1504 len = bm->bm_size; 1505 slop = va & MMU_PAGEOFFSET; 1506 if (slop) { 1507 va += MMU_PAGESIZE - slop; 1508 len -= MMU_PAGESIZE - slop; 1509 } 1510 len = P2ALIGN(len, MMU_PAGESIZE); 1511 1512 /* 1513 * Go through all boot_archive pages, swapping any low MFN pages 1514 * with memory at next_phys. 1515 */ 1516 while (len != 0) { 1517 ++total; 1518 va_pfn = mmu_btop(va - ONE_GIG); 1519 va_mfn = mfn_list[va_pfn]; 1520 if (mfn_list[va_pfn] < PFN_2GIG) { 1521 copy = kbm_remap_window(next_phys, 1); 1522 bcopy((void *)va, copy, MMU_PAGESIZE); 1523 copy_pfn = mmu_btop(next_phys); 1524 copy_mfn = mfn_list[copy_pfn]; 1525 1526 pte = mfn_to_ma(copy_mfn) | PT_NOCONSIST | PT_VALID; 1527 if (HYPERVISOR_update_va_mapping(va, pte, 1528 UVMF_INVLPG | UVMF_LOCAL)) 1529 bop_panic("relocate_boot_archive(): " 1530 "HYPERVISOR_update_va_mapping() failed"); 1531 1532 mfn_list[va_pfn] = copy_mfn; 1533 mfn_list[copy_pfn] = va_mfn; 1534 1535 t[0].ptr = mfn_to_ma(copy_mfn) | MMU_MACHPHYS_UPDATE; 1536 t[0].val = va_pfn; 1537 t[1].ptr = mfn_to_ma(va_mfn) | MMU_MACHPHYS_UPDATE; 1538 t[1].val = copy_pfn; 1539 if (HYPERVISOR_mmu_update(t, 2, &mmu_update_return, 1540 DOMID_SELF) != 0 || mmu_update_return != 2) 1541 bop_panic("relocate_boot_archive(): " 1542 "HYPERVISOR_mmu_update() failed"); 1543 1544 next_phys += MMU_PAGESIZE; 1545 ++relocated; 1546 } 1547 len -= MMU_PAGESIZE; 1548 va += MMU_PAGESIZE; 1549 } 1550 DBG_MSG("Relocated pages:\n"); 1551 DBG(relocated); 1552 DBG_MSG("Out of total pages:\n"); 1553 DBG(total); 1554 } 1555 #endif /* __xpv */ 1556 1557 #if !defined(__xpv) 1558 /* 1559 * Install a temporary IDT that lets us catch errors in the boot time code. 1560 * We shouldn't get any faults at all while this is installed, so we'll 1561 * just generate a traceback and exit. 1562 */ 1563 #ifdef __amd64 1564 static const int bcode_sel = B64CODE_SEL; 1565 #else 1566 static const int bcode_sel = B32CODE_SEL; 1567 #endif 1568 1569 /* 1570 * simple description of a stack frame (args are 32 bit only currently) 1571 */ 1572 typedef struct bop_frame { 1573 struct bop_frame *old_frame; 1574 pc_t retaddr; 1575 long arg[1]; 1576 } bop_frame_t; 1577 1578 void 1579 bop_traceback(bop_frame_t *frame) 1580 { 1581 pc_t pc; 1582 int cnt; 1583 char *ksym; 1584 ulong_t off; 1585 #if defined(__i386) 1586 int a; 1587 #endif 1588 1589 bop_printf(NULL, "Stack traceback:\n"); 1590 for (cnt = 0; cnt < 30; ++cnt) { /* up to 30 frames */ 1591 pc = frame->retaddr; 1592 if (pc == 0) 1593 break; 1594 ksym = kobj_getsymname(pc, &off); 1595 if (ksym) 1596 bop_printf(NULL, " %s+%lx", ksym, off); 1597 else 1598 bop_printf(NULL, " 0x%lx", pc); 1599 1600 frame = frame->old_frame; 1601 if (frame == 0) { 1602 bop_printf(NULL, "\n"); 1603 break; 1604 } 1605 #if defined(__i386) 1606 for (a = 0; a < 6; ++a) { /* try for 6 args */ 1607 if ((void *)&frame->arg[a] == (void *)frame->old_frame) 1608 break; 1609 if (a == 0) 1610 bop_printf(NULL, "("); 1611 else 1612 bop_printf(NULL, ","); 1613 bop_printf(NULL, "0x%lx", frame->arg[a]); 1614 } 1615 bop_printf(NULL, ")"); 1616 #endif 1617 bop_printf(NULL, "\n"); 1618 } 1619 } 1620 1621 struct trapframe { 1622 ulong_t error_code; /* optional */ 1623 ulong_t inst_ptr; 1624 ulong_t code_seg; 1625 ulong_t flags_reg; 1626 #ifdef __amd64 1627 ulong_t stk_ptr; 1628 ulong_t stk_seg; 1629 #endif 1630 }; 1631 1632 void 1633 bop_trap(ulong_t *tfp) 1634 { 1635 struct trapframe *tf = (struct trapframe *)tfp; 1636 bop_frame_t fakeframe; 1637 static int depth = 0; 1638 1639 /* 1640 * Check for an infinite loop of traps. 1641 */ 1642 if (++depth > 2) 1643 bop_panic("Nested trap"); 1644 1645 bop_printf(NULL, "Unexpected trap\n"); 1646 1647 /* 1648 * adjust the tf for optional error_code by detecting the code selector 1649 */ 1650 if (tf->code_seg != bcode_sel) 1651 tf = (struct trapframe *)(tfp - 1); 1652 else 1653 bop_printf(NULL, "error code 0x%lx\n", 1654 tf->error_code & 0xffffffff); 1655 1656 bop_printf(NULL, "instruction pointer 0x%lx\n", tf->inst_ptr); 1657 bop_printf(NULL, "code segment 0x%lx\n", tf->code_seg & 0xffff); 1658 bop_printf(NULL, "flags register 0x%lx\n", tf->flags_reg); 1659 #ifdef __amd64 1660 bop_printf(NULL, "return %%rsp 0x%lx\n", tf->stk_ptr); 1661 bop_printf(NULL, "return %%ss 0x%lx\n", tf->stk_seg & 0xffff); 1662 #endif 1663 1664 /* grab %[er]bp pushed by our code from the stack */ 1665 fakeframe.old_frame = (bop_frame_t *)*(tfp - 3); 1666 fakeframe.retaddr = (pc_t)tf->inst_ptr; 1667 bop_printf(NULL, "Attempting stack backtrace:\n"); 1668 bop_traceback(&fakeframe); 1669 bop_panic("unexpected trap in early boot"); 1670 } 1671 1672 extern void bop_trap_handler(void); 1673 1674 static gate_desc_t *bop_idt; 1675 1676 static desctbr_t bop_idt_info; 1677 1678 static void 1679 bop_idt_init(void) 1680 { 1681 int t; 1682 1683 bop_idt = (gate_desc_t *) 1684 do_bsys_alloc(NULL, NULL, MMU_PAGESIZE, MMU_PAGESIZE); 1685 bzero(bop_idt, MMU_PAGESIZE); 1686 for (t = 0; t < NIDT; ++t) { 1687 /* 1688 * Note that since boot runs without a TSS, the 1689 * double fault handler cannot use an alternate stack 1690 * (64-bit) or a task gate (32-bit). 1691 */ 1692 set_gatesegd(&bop_idt[t], &bop_trap_handler, bcode_sel, 1693 SDT_SYSIGT, TRP_KPL, 0); 1694 } 1695 bop_idt_info.dtr_limit = (NIDT * sizeof (gate_desc_t)) - 1; 1696 bop_idt_info.dtr_base = (uintptr_t)bop_idt; 1697 wr_idtr(&bop_idt_info); 1698 } 1699 #endif /* !defined(__xpv) */ 1700 1701 /* 1702 * This is where we enter the kernel. It dummies up the boot_ops and 1703 * boot_syscalls vectors and jumps off to _kobj_boot() 1704 */ 1705 void 1706 _start(struct xboot_info *xbp) 1707 { 1708 bootops_t *bops = &bootop; 1709 extern void _kobj_boot(); 1710 1711 /* 1712 * 1st off - initialize the console for any error messages 1713 */ 1714 xbootp = xbp; 1715 #ifdef __xpv 1716 HYPERVISOR_shared_info = (void *)xbootp->bi_shared_info; 1717 xen_info = xbootp->bi_xen_start_info; 1718 #endif 1719 1720 #ifndef __xpv 1721 if (*((uint32_t *)(FASTBOOT_SWTCH_PA + FASTBOOT_STACK_OFFSET)) == 1722 FASTBOOT_MAGIC) { 1723 post_fastreboot = 1; 1724 *((uint32_t *)(FASTBOOT_SWTCH_PA + FASTBOOT_STACK_OFFSET)) = 0; 1725 } 1726 #endif 1727 1728 bcons_init((void *)xbootp->bi_cmdline); 1729 have_console = 1; 1730 1731 /* 1732 * enable debugging 1733 */ 1734 if (strstr((char *)xbootp->bi_cmdline, "kbm_debug")) 1735 kbm_debug = 1; 1736 1737 DBG_MSG("\n\n*** Entered Solaris in _start() cmdline is: "); 1738 DBG_MSG((char *)xbootp->bi_cmdline); 1739 DBG_MSG("\n\n\n"); 1740 1741 /* 1742 * physavail is no longer used by startup 1743 */ 1744 bm.physinstalled = xbp->bi_phys_install; 1745 bm.pcimem = xbp->bi_pcimem; 1746 bm.rsvdmem = xbp->bi_rsvdmem; 1747 bm.physavail = NULL; 1748 1749 /* 1750 * initialize the boot time allocator 1751 */ 1752 next_phys = xbootp->bi_next_paddr; 1753 DBG(next_phys); 1754 next_virt = (uintptr_t)xbootp->bi_next_vaddr; 1755 DBG(next_virt); 1756 DBG_MSG("Initializing boot time memory management..."); 1757 #ifdef __xpv 1758 { 1759 xen_platform_parameters_t p; 1760 1761 /* This call shouldn't fail, dboot already did it once. */ 1762 (void) HYPERVISOR_xen_version(XENVER_platform_parameters, &p); 1763 mfn_to_pfn_mapping = (pfn_t *)(xen_virt_start = p.virt_start); 1764 DBG(xen_virt_start); 1765 } 1766 #endif 1767 kbm_init(xbootp); 1768 DBG_MSG("done\n"); 1769 1770 /* 1771 * Fill in the bootops vector 1772 */ 1773 bops->bsys_version = BO_VERSION; 1774 bops->boot_mem = &bm; 1775 bops->bsys_alloc = do_bsys_alloc; 1776 bops->bsys_free = do_bsys_free; 1777 bops->bsys_getproplen = do_bsys_getproplen; 1778 bops->bsys_getprop = do_bsys_getprop; 1779 bops->bsys_nextprop = do_bsys_nextprop; 1780 bops->bsys_printf = bop_printf; 1781 bops->bsys_doint = do_bsys_doint; 1782 1783 /* 1784 * BOP_EALLOC() is no longer needed 1785 */ 1786 bops->bsys_ealloc = do_bsys_ealloc; 1787 1788 #ifdef __xpv 1789 /* 1790 * On domain 0 we need to free up some physical memory that is 1791 * usable for DMA. Since GRUB loaded the boot_archive, it is 1792 * sitting in low MFN memory. We'll relocated the boot archive 1793 * pages to high PFN memory. 1794 */ 1795 if (DOMAIN_IS_INITDOMAIN(xen_info)) 1796 relocate_boot_archive(); 1797 #endif 1798 1799 #ifndef __xpv 1800 /* 1801 * Install an IDT to catch early pagefaults (shouldn't have any). 1802 * Also needed for kmdb. 1803 */ 1804 bop_idt_init(); 1805 #endif 1806 1807 /* 1808 * Start building the boot properties from the command line 1809 */ 1810 DBG_MSG("Initializing boot properties:\n"); 1811 build_boot_properties(); 1812 1813 if (strstr((char *)xbootp->bi_cmdline, "prom_debug") || kbm_debug) { 1814 char *name; 1815 char *value; 1816 char *cp; 1817 int len; 1818 1819 value = do_bsys_alloc(NULL, NULL, MMU_PAGESIZE, MMU_PAGESIZE); 1820 bop_printf(NULL, "\nBoot properties:\n"); 1821 name = ""; 1822 while ((name = do_bsys_nextprop(NULL, name)) != NULL) { 1823 bop_printf(NULL, "\t0x%p %s = ", (void *)name, name); 1824 (void) do_bsys_getprop(NULL, name, value); 1825 len = do_bsys_getproplen(NULL, name); 1826 bop_printf(NULL, "len=%d ", len); 1827 value[len] = 0; 1828 for (cp = value; *cp; ++cp) { 1829 if (' ' <= *cp && *cp <= '~') 1830 bop_printf(NULL, "%c", *cp); 1831 else 1832 bop_printf(NULL, "-0x%x-", *cp); 1833 } 1834 bop_printf(NULL, "\n"); 1835 } 1836 } 1837 1838 /* 1839 * jump into krtld... 1840 */ 1841 _kobj_boot(&bop_sysp, NULL, bops, NULL); 1842 } 1843 1844 1845 /*ARGSUSED*/ 1846 static caddr_t 1847 no_more_alloc(bootops_t *bop, caddr_t virthint, size_t size, int align) 1848 { 1849 panic("Attempt to bsys_alloc() too late\n"); 1850 return (NULL); 1851 } 1852 1853 /*ARGSUSED*/ 1854 static void 1855 no_more_free(bootops_t *bop, caddr_t virt, size_t size) 1856 { 1857 panic("Attempt to bsys_free() too late\n"); 1858 } 1859 1860 void 1861 bop_no_more_mem(void) 1862 { 1863 DBG(total_bop_alloc_scratch); 1864 DBG(total_bop_alloc_kernel); 1865 bootops->bsys_alloc = no_more_alloc; 1866 bootops->bsys_free = no_more_free; 1867 } 1868 1869 1870 /* 1871 * Set ACPI firmware properties 1872 */ 1873 1874 static caddr_t 1875 vmap_phys(size_t length, paddr_t pa) 1876 { 1877 paddr_t start, end; 1878 caddr_t va; 1879 size_t len, page; 1880 1881 #ifdef __xpv 1882 pa = pfn_to_pa(xen_assign_pfn(mmu_btop(pa))) | (pa & MMU_PAGEOFFSET); 1883 #endif 1884 start = P2ALIGN(pa, MMU_PAGESIZE); 1885 end = P2ROUNDUP(pa + length, MMU_PAGESIZE); 1886 len = end - start; 1887 va = (caddr_t)alloc_vaddr(len, MMU_PAGESIZE); 1888 for (page = 0; page < len; page += MMU_PAGESIZE) 1889 kbm_map((uintptr_t)va + page, start + page, 0, 0); 1890 return (va + (pa & MMU_PAGEOFFSET)); 1891 } 1892 1893 static uint8_t 1894 checksum_table(uint8_t *tp, size_t len) 1895 { 1896 uint8_t sum = 0; 1897 1898 while (len-- > 0) 1899 sum += *tp++; 1900 1901 return (sum); 1902 } 1903 1904 static int 1905 valid_rsdp(struct rsdp *rp) 1906 { 1907 1908 /* validate the V1.x checksum */ 1909 if (checksum_table((uint8_t *)&rp->v1, sizeof (struct rsdp_v1)) != 0) 1910 return (0); 1911 1912 /* If pre-ACPI 2.0, this is a valid RSDP */ 1913 if (rp->v1.revision < 2) 1914 return (1); 1915 1916 /* validate the V2.x checksum */ 1917 if (checksum_table((uint8_t *)rp, sizeof (struct rsdp)) != 0) 1918 return (0); 1919 1920 return (1); 1921 } 1922 1923 /* 1924 * Scan memory range for an RSDP; 1925 * see ACPI 3.0 Spec, 5.2.5.1 1926 */ 1927 static struct rsdp * 1928 scan_rsdp(paddr_t start, paddr_t end) 1929 { 1930 size_t len = end - start + 1; 1931 caddr_t ptr; 1932 1933 ptr = vmap_phys(len, start); 1934 while (len > 0) { 1935 if (strncmp(ptr, ACPI_RSDP_SIG, ACPI_RSDP_SIG_LEN) == 0) 1936 if (valid_rsdp((struct rsdp *)ptr)) 1937 return ((struct rsdp *)ptr); 1938 ptr += 16; 1939 len -= 16; 1940 } 1941 1942 return (NULL); 1943 } 1944 1945 /* 1946 * Refer to ACPI 3.0 Spec, section 5.2.5.1 to understand this function 1947 */ 1948 static struct rsdp * 1949 find_rsdp() { 1950 struct rsdp *rsdp; 1951 uint16_t *ebda_seg; 1952 paddr_t ebda_addr; 1953 1954 /* 1955 * Get the EBDA segment and scan the first 1K 1956 */ 1957 ebda_seg = (uint16_t *)vmap_phys(sizeof (uint16_t), ACPI_EBDA_SEG_ADDR); 1958 ebda_addr = *ebda_seg << 4; 1959 rsdp = scan_rsdp(ebda_addr, ebda_addr + ACPI_EBDA_LEN - 1); 1960 if (rsdp == NULL) 1961 /* if EBDA doesn't contain RSDP, look in BIOS memory */ 1962 rsdp = scan_rsdp(0xe0000, 0xfffff); 1963 return (rsdp); 1964 } 1965 1966 static struct table_header * 1967 map_fw_table(paddr_t table_addr) 1968 { 1969 struct table_header *tp; 1970 size_t len = MAX(sizeof (struct table_header), MMU_PAGESIZE); 1971 1972 /* 1973 * Map at least a page; if the table is larger than this, remap it 1974 */ 1975 tp = (struct table_header *)vmap_phys(len, table_addr); 1976 if (tp->len > len) 1977 tp = (struct table_header *)vmap_phys(tp->len, table_addr); 1978 return (tp); 1979 } 1980 1981 static struct table_header * 1982 find_fw_table(char *signature) 1983 { 1984 static int revision = 0; 1985 static struct xsdt *xsdt; 1986 static int len; 1987 paddr_t xsdt_addr; 1988 struct rsdp *rsdp; 1989 struct table_header *tp; 1990 paddr_t table_addr; 1991 int n; 1992 1993 if (strlen(signature) != ACPI_TABLE_SIG_LEN) 1994 return (NULL); 1995 1996 /* 1997 * Reading the ACPI 3.0 Spec, section 5.2.5.3 will help 1998 * understand this code. If we haven't already found the RSDT/XSDT, 1999 * revision will be 0. Find the RSDP and check the revision 2000 * to find out whether to use the RSDT or XSDT. If revision is 2001 * 0 or 1, use the RSDT and set internal revision to 1; if it is 2, 2002 * use the XSDT. If the XSDT address is 0, though, fall back to 2003 * revision 1 and use the RSDT. 2004 */ 2005 if (revision == 0) { 2006 if ((rsdp = (struct rsdp *)find_rsdp()) != NULL) { 2007 revision = rsdp->v1.revision; 2008 switch (revision) { 2009 case 2: 2010 /* 2011 * Use the XSDT unless BIOS is buggy and 2012 * claims to be rev 2 but has a null XSDT 2013 * address 2014 */ 2015 xsdt_addr = rsdp->xsdt; 2016 if (xsdt_addr != 0) 2017 break; 2018 /* FALLTHROUGH */ 2019 case 0: 2020 /* treat RSDP rev 0 as revision 1 internally */ 2021 revision = 1; 2022 /* FALLTHROUGH */ 2023 case 1: 2024 /* use the RSDT for rev 0/1 */ 2025 xsdt_addr = rsdp->v1.rsdt; 2026 break; 2027 default: 2028 /* unknown revision */ 2029 revision = 0; 2030 break; 2031 } 2032 } 2033 if (revision == 0) 2034 return (NULL); 2035 2036 /* cache the XSDT info */ 2037 xsdt = (struct xsdt *)map_fw_table(xsdt_addr); 2038 len = (xsdt->hdr.len - sizeof (xsdt->hdr)) / 2039 ((revision == 1) ? sizeof (uint32_t) : sizeof (uint64_t)); 2040 } 2041 2042 /* 2043 * Scan the table headers looking for a signature match 2044 */ 2045 for (n = 0; n < len; n++) { 2046 table_addr = (revision == 1) ? xsdt->p.r[n] : xsdt->p.x[n]; 2047 if (table_addr == 0) 2048 continue; 2049 tp = map_fw_table(table_addr); 2050 if (strncmp(tp->sig, signature, ACPI_TABLE_SIG_LEN) == 0) { 2051 return (tp); 2052 } 2053 } 2054 return (NULL); 2055 } 2056 2057 static void 2058 process_mcfg(struct mcfg *tp) 2059 { 2060 struct cfg_base_addr_alloc *cfg_baap; 2061 char *cfg_baa_endp; 2062 int64_t ecfginfo[4]; 2063 2064 cfg_baap = tp->CfgBaseAddrAllocList; 2065 cfg_baa_endp = ((char *)tp) + tp->Length; 2066 while ((char *)cfg_baap < cfg_baa_endp) { 2067 if (cfg_baap->base_addr != 0 && cfg_baap->segment == 0) { 2068 ecfginfo[0] = cfg_baap->base_addr; 2069 ecfginfo[1] = cfg_baap->segment; 2070 ecfginfo[2] = cfg_baap->start_bno; 2071 ecfginfo[3] = cfg_baap->end_bno; 2072 bsetprop(MCFG_PROPNAME, strlen(MCFG_PROPNAME), 2073 ecfginfo, sizeof (ecfginfo)); 2074 break; 2075 } 2076 cfg_baap++; 2077 } 2078 } 2079 2080 #ifndef __xpv 2081 static void 2082 process_madt(struct madt *tp) 2083 { 2084 struct madt_processor *cpu, *end; 2085 uint32_t cpu_count = 0; 2086 uint32_t cpu_possible_count = 0; 2087 uint8_t cpu_apicid_array[UINT8_MAX + 1]; 2088 2089 if (tp != NULL) { 2090 /* 2091 * Determine number of CPUs and keep track of "final" APIC ID 2092 * for each CPU by walking through ACPI MADT processor list 2093 */ 2094 end = (struct madt_processor *)(tp->hdr.len + (uintptr_t)tp); 2095 cpu = tp->list; 2096 while (cpu < end) { 2097 if (cpu->type == MADT_PROCESSOR) { 2098 if (cpu->flags & 1) { 2099 if (cpu_count < UINT8_MAX) 2100 cpu_apicid_array[cpu_count] = 2101 cpu->apic_id; 2102 cpu_count++; 2103 } 2104 cpu_possible_count++; 2105 } 2106 2107 cpu = (struct madt_processor *) 2108 (cpu->len + (uintptr_t)cpu); 2109 } 2110 2111 /* 2112 * Make boot property for array of "final" APIC IDs for each 2113 * CPU 2114 */ 2115 bsetprop(BP_CPU_APICID_ARRAY, strlen(BP_CPU_APICID_ARRAY), 2116 cpu_apicid_array, cpu_count * sizeof (uint8_t)); 2117 } 2118 2119 /* 2120 * Check whehter property plat-max-ncpus is already set. 2121 */ 2122 if (do_bsys_getproplen(NULL, PLAT_MAX_NCPUS_NAME) < 0) { 2123 /* 2124 * Set plat-max-ncpus to number of maximum possible CPUs given 2125 * in MADT if it hasn't been set. 2126 * There's no formal way to detect max possible CPUs supported 2127 * by platform according to ACPI spec3.0b. So current CPU 2128 * hotplug implementation expects that all possible CPUs will 2129 * have an entry in MADT table and set plat-max-ncpus to number 2130 * of entries in MADT. 2131 * With introducing of ACPI4.0, Maximum System Capability Table 2132 * (MSCT) provides maximum number of CPUs supported by platform. 2133 * If MSCT is unavailable, fall back to old way. 2134 */ 2135 if (tp != NULL) 2136 bsetpropsi(PLAT_MAX_NCPUS_NAME, cpu_possible_count); 2137 } 2138 2139 /* 2140 * Set boot property boot-max-ncpus to number of CPUs existing at 2141 * boot time. boot-max-ncpus is mainly used for optimization. 2142 */ 2143 if (tp != NULL) 2144 bsetpropsi(BOOT_MAX_NCPUS_NAME, cpu_count); 2145 2146 /* 2147 * User-set boot-ncpus overrides firmware count 2148 */ 2149 if (do_bsys_getproplen(NULL, BOOT_NCPUS_NAME) >= 0) 2150 return; 2151 2152 /* 2153 * Set boot property boot-ncpus to number of active CPUs given in MADT 2154 * if it hasn't been set yet. 2155 */ 2156 if (tp != NULL) 2157 bsetpropsi(BOOT_NCPUS_NAME, cpu_count); 2158 } 2159 2160 static void 2161 process_srat(struct srat *tp) 2162 { 2163 struct srat_item *item, *end; 2164 int i; 2165 int proc_num, mem_num; 2166 #pragma pack(1) 2167 struct { 2168 uint32_t domain; 2169 uint32_t apic_id; 2170 uint32_t sapic_id; 2171 } processor; 2172 struct { 2173 uint32_t domain; 2174 uint32_t x2apic_id; 2175 } x2apic; 2176 struct { 2177 uint32_t domain; 2178 uint64_t addr; 2179 uint64_t length; 2180 uint32_t flags; 2181 } memory; 2182 #pragma pack() 2183 char prop_name[30]; 2184 uint64_t maxmem = 0; 2185 2186 if (tp == NULL) 2187 return; 2188 2189 proc_num = mem_num = 0; 2190 end = (struct srat_item *)(tp->hdr.len + (uintptr_t)tp); 2191 item = tp->list; 2192 while (item < end) { 2193 switch (item->type) { 2194 case SRAT_PROCESSOR: 2195 if (!(item->i.p.flags & SRAT_ENABLED)) 2196 break; 2197 processor.domain = item->i.p.domain1; 2198 for (i = 0; i < 3; i++) 2199 processor.domain += 2200 item->i.p.domain2[i] << ((i + 1) * 8); 2201 processor.apic_id = item->i.p.apic_id; 2202 processor.sapic_id = item->i.p.local_sapic_eid; 2203 (void) snprintf(prop_name, 30, "acpi-srat-processor-%d", 2204 proc_num); 2205 bsetprop(prop_name, strlen(prop_name), &processor, 2206 sizeof (processor)); 2207 proc_num++; 2208 break; 2209 case SRAT_MEMORY: 2210 if (!(item->i.m.flags & SRAT_ENABLED)) 2211 break; 2212 memory.domain = item->i.m.domain; 2213 memory.addr = item->i.m.base_addr; 2214 memory.length = item->i.m.len; 2215 memory.flags = item->i.m.flags; 2216 (void) snprintf(prop_name, 30, "acpi-srat-memory-%d", 2217 mem_num); 2218 bsetprop(prop_name, strlen(prop_name), &memory, 2219 sizeof (memory)); 2220 if ((item->i.m.flags & SRAT_HOT_PLUG) && 2221 (memory.addr + memory.length > maxmem)) { 2222 maxmem = memory.addr + memory.length; 2223 } 2224 mem_num++; 2225 break; 2226 case SRAT_X2APIC: 2227 if (!(item->i.xp.flags & SRAT_ENABLED)) 2228 break; 2229 x2apic.domain = item->i.xp.domain; 2230 x2apic.x2apic_id = item->i.xp.x2apic_id; 2231 (void) snprintf(prop_name, 30, "acpi-srat-processor-%d", 2232 proc_num); 2233 bsetprop(prop_name, strlen(prop_name), &x2apic, 2234 sizeof (x2apic)); 2235 proc_num++; 2236 break; 2237 } 2238 2239 item = (struct srat_item *) 2240 (item->len + (caddr_t)item); 2241 } 2242 2243 /* 2244 * The maximum physical address calculated from the SRAT table is more 2245 * accurate than that calculated from the MSCT table. 2246 */ 2247 if (maxmem != 0) { 2248 plat_dr_physmax = btop(maxmem); 2249 } 2250 } 2251 2252 static void 2253 process_slit(struct slit *tp) 2254 { 2255 2256 /* 2257 * Check the number of localities; if it's too huge, we just 2258 * return and locality enumeration code will handle this later, 2259 * if possible. 2260 * 2261 * Note that the size of the table is the square of the 2262 * number of localities; if the number of localities exceeds 2263 * UINT16_MAX, the table size may overflow an int when being 2264 * passed to bsetprop() below. 2265 */ 2266 if (tp->number >= SLIT_LOCALITIES_MAX) 2267 return; 2268 2269 bsetprop(SLIT_NUM_PROPNAME, strlen(SLIT_NUM_PROPNAME), &tp->number, 2270 sizeof (tp->number)); 2271 bsetprop(SLIT_PROPNAME, strlen(SLIT_PROPNAME), &tp->entry, 2272 tp->number * tp->number); 2273 } 2274 2275 static struct msct * 2276 process_msct(struct msct *tp) 2277 { 2278 int last_seen = 0; 2279 int proc_num = 0; 2280 struct msct_proximity_domain *item, *end; 2281 extern uint64_t plat_dr_options; 2282 2283 ASSERT(tp != NULL); 2284 2285 end = (void *)(tp->hdr.len + (uintptr_t)tp); 2286 for (item = (void *)((uintptr_t)tp + tp->proximity_domain_offset); 2287 item < end; 2288 item = (void *)(item->length + (uintptr_t)item)) { 2289 /* 2290 * Sanity check according to section 5.2.19.1 of ACPI 4.0. 2291 * Revision 1 2292 * Length 22 2293 */ 2294 if (item->revision != 1 || item->length != 22) { 2295 cmn_err(CE_CONT, 2296 "?boot: unknown proximity domain structure in MSCT " 2297 "with rev(%d), len(%d).\n", 2298 (int)item->revision, (int)item->length); 2299 return (NULL); 2300 } else if (item->domain_min > item->domain_max) { 2301 cmn_err(CE_CONT, 2302 "?boot: invalid proximity domain structure in MSCT " 2303 "with domain_min(%u), domain_max(%u).\n", 2304 item->domain_min, item->domain_max); 2305 return (NULL); 2306 } else if (item->domain_min != last_seen) { 2307 /* 2308 * Items must be organized in ascending order of the 2309 * proximity domain enumerations. 2310 */ 2311 cmn_err(CE_CONT, 2312 "?boot: invalid proximity domain structure in MSCT," 2313 " items are not orginized in ascending order.\n"); 2314 return (NULL); 2315 } 2316 2317 /* 2318 * If processor_max is 0 then there would be no CPUs in this 2319 * domain. 2320 */ 2321 if (item->processor_max != 0) { 2322 proc_num += (item->domain_max - item->domain_min + 1) * 2323 item->processor_max; 2324 } 2325 2326 last_seen = item->domain_max - item->domain_min + 1; 2327 /* 2328 * Break out if all proximity domains have been processed. 2329 * Some BIOSes may have unused items at the end of MSCT table. 2330 */ 2331 if (last_seen > tp->maximum_proximity_domains) { 2332 break; 2333 } 2334 } 2335 if (last_seen != tp->maximum_proximity_domains + 1) { 2336 cmn_err(CE_CONT, 2337 "?boot: invalid proximity domain structure in MSCT, " 2338 "proximity domain count doesn't match.\n"); 2339 return (NULL); 2340 } 2341 2342 /* 2343 * Set plat-max-ncpus property if it hasn't been set yet. 2344 */ 2345 if (do_bsys_getproplen(NULL, PLAT_MAX_NCPUS_NAME) < 0) { 2346 if (proc_num != 0) { 2347 bsetpropsi(PLAT_MAX_NCPUS_NAME, proc_num); 2348 } 2349 } 2350 2351 /* 2352 * Use Maximum Physical Address from the MSCT table as upper limit for 2353 * memory hot-adding by default. It may be overridden by value from 2354 * the SRAT table or the "plat-dr-physmax" boot option. 2355 */ 2356 plat_dr_physmax = btop(tp->maximum_physical_address + 1); 2357 2358 /* 2359 * Existence of MSCT implies CPU/memory hotplug-capability for the 2360 * platform. 2361 */ 2362 plat_dr_options |= PLAT_DR_FEATURE_CPU; 2363 plat_dr_options |= PLAT_DR_FEATURE_MEMORY; 2364 2365 return (tp); 2366 } 2367 2368 #else /* __xpv */ 2369 static void 2370 enumerate_xen_cpus() 2371 { 2372 processorid_t id, max_id; 2373 2374 /* 2375 * User-set boot-ncpus overrides enumeration 2376 */ 2377 if (do_bsys_getproplen(NULL, BOOT_NCPUS_NAME) >= 0) 2378 return; 2379 2380 /* 2381 * Probe every possible virtual CPU id and remember the 2382 * highest id present; the count of CPUs is one greater 2383 * than this. This tacitly assumes at least cpu 0 is present. 2384 */ 2385 max_id = 0; 2386 for (id = 0; id < MAX_VIRT_CPUS; id++) 2387 if (HYPERVISOR_vcpu_op(VCPUOP_is_up, id, NULL) == 0) 2388 max_id = id; 2389 2390 bsetpropsi(BOOT_NCPUS_NAME, max_id+1); 2391 2392 } 2393 #endif /* __xpv */ 2394 2395 static void 2396 build_firmware_properties(void) 2397 { 2398 struct table_header *tp = NULL; 2399 2400 #ifndef __xpv 2401 if ((msct_ptr = (struct msct *)find_fw_table("MSCT")) != NULL) 2402 msct_ptr = process_msct(msct_ptr); 2403 2404 if ((tp = find_fw_table("APIC")) != NULL) 2405 process_madt((struct madt *)tp); 2406 2407 if ((srat_ptr = (struct srat *)find_fw_table("SRAT")) != NULL) 2408 process_srat(srat_ptr); 2409 2410 if (slit_ptr = (struct slit *)find_fw_table("SLIT")) 2411 process_slit(slit_ptr); 2412 2413 tp = find_fw_table("MCFG"); 2414 #else /* __xpv */ 2415 enumerate_xen_cpus(); 2416 if (DOMAIN_IS_INITDOMAIN(xen_info)) 2417 tp = find_fw_table("MCFG"); 2418 #endif /* __xpv */ 2419 if (tp != NULL) 2420 process_mcfg((struct mcfg *)tp); 2421 } 2422 2423 /* 2424 * fake up a boot property for deferred early console output 2425 * this is used by both graphical boot and the (developer only) 2426 * USB serial console 2427 */ 2428 void * 2429 defcons_init(size_t size) 2430 { 2431 static char *p = NULL; 2432 2433 p = do_bsys_alloc(NULL, NULL, size, MMU_PAGESIZE); 2434 *p = 0; 2435 bsetprop("deferred-console-buf", strlen("deferred-console-buf") + 1, 2436 &p, sizeof (p)); 2437 return (p); 2438 } 2439 2440 /*ARGSUSED*/ 2441 int 2442 boot_compinfo(int fd, struct compinfo *cbp) 2443 { 2444 cbp->iscmp = 0; 2445 cbp->blksize = MAXBSIZE; 2446 return (0); 2447 } 2448 2449 #define BP_MAX_STRLEN 32 2450 2451 /* 2452 * Get value for given boot property 2453 */ 2454 int 2455 bootprop_getval(const char *prop_name, u_longlong_t *prop_value) 2456 { 2457 int boot_prop_len; 2458 char str[BP_MAX_STRLEN]; 2459 u_longlong_t value; 2460 2461 boot_prop_len = BOP_GETPROPLEN(bootops, prop_name); 2462 if (boot_prop_len < 0 || boot_prop_len > sizeof (str) || 2463 BOP_GETPROP(bootops, prop_name, str) < 0 || 2464 kobj_getvalue(str, &value) == -1) 2465 return (-1); 2466 2467 if (prop_value) 2468 *prop_value = value; 2469 2470 return (0); 2471 } 2472