fakebop.c revision 7656:2621e50fdf4a
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#include <sys/fastboot.h> 55#ifdef __xpv 56#include <sys/hypervisor.h> 57#include <net/if.h> 58#endif 59#include <vm/kboot_mmu.h> 60#include <vm/hat_pte.h> 61#include <sys/dmar_acpi.h> 62#include "acpi_fw.h" 63 64static int have_console = 0; /* set once primitive console is initialized */ 65static char *boot_args = ""; 66 67/* 68 * Debugging macros 69 */ 70static uint_t kbm_debug = 0; 71#define DBG_MSG(s) { if (kbm_debug) bop_printf(NULL, "%s", s); } 72#define DBG(x) { if (kbm_debug) \ 73 bop_printf(NULL, "%s is %" PRIx64 "\n", #x, (uint64_t)(x)); \ 74 } 75 76#define PUT_STRING(s) { \ 77 char *cp; \ 78 for (cp = (s); *cp; ++cp) \ 79 bcons_putchar(*cp); \ 80 } 81 82struct xboot_info *xbootp; /* boot info from "glue" code in low memory */ 83bootops_t bootop; /* simple bootops we'll pass on to kernel */ 84struct bsys_mem bm; 85 86static uintptr_t next_virt; /* next available virtual address */ 87static paddr_t next_phys; /* next available physical address from dboot */ 88static paddr_t high_phys = -(paddr_t)1; /* last used physical address */ 89 90/* 91 * buffer for vsnprintf for console I/O 92 */ 93#define BUFFERSIZE 256 94static char buffer[BUFFERSIZE]; 95/* 96 * stuff to store/report/manipulate boot property settings. 97 */ 98typedef struct bootprop { 99 struct bootprop *bp_next; 100 char *bp_name; 101 uint_t bp_vlen; 102 char *bp_value; 103} bootprop_t; 104 105static bootprop_t *bprops = NULL; 106static char *curr_page = NULL; /* ptr to avail bprop memory */ 107static int curr_space = 0; /* amount of memory at curr_page */ 108 109#ifdef __xpv 110start_info_t *xen_info; 111shared_info_t *HYPERVISOR_shared_info; 112#endif 113 114/* 115 * some allocator statistics 116 */ 117static ulong_t total_bop_alloc_scratch = 0; 118static ulong_t total_bop_alloc_kernel = 0; 119 120static void build_firmware_properties(void); 121 122static int early_allocation = 1; 123 124#ifdef __xpv 125int fastreboot_capable = 0; 126int force_fastreboot = 0; 127int post_fastreboot = 0; 128#else 129int fastreboot_capable = 1; 130int force_fastreboot = 0; 131int post_fastreboot = 0; 132#endif 133 134/* 135 * Information saved from current boot for fast reboot. 136 * If the information size exceeds what we have allocated, fast reboot 137 * will not be supported. 138 */ 139multiboot_info_t saved_mbi; 140mb_memory_map_t saved_mmap[FASTBOOT_SAVED_MMAP_COUNT]; 141struct sol_netinfo saved_drives[FASTBOOT_SAVED_DRIVES_COUNT]; 142char saved_cmdline[FASTBOOT_SAVED_CMDLINE_LEN]; 143int saved_cmdline_len = 0; 144 145/* 146 * Pointers to where System Resource Affinity Table (SRAT) and 147 * System Locality Information Table (SLIT) are mapped into virtual memory 148 */ 149struct srat *srat_ptr = NULL; 150struct slit *slit_ptr = NULL; 151 152 153/* 154 * Allocate aligned physical memory at boot time. This allocator allocates 155 * from the highest possible addresses. This avoids exhausting memory that 156 * would be useful for DMA buffers. 157 */ 158paddr_t 159do_bop_phys_alloc(uint64_t size, uint64_t align) 160{ 161 paddr_t pa = 0; 162 paddr_t start; 163 paddr_t end; 164 struct memlist *ml = (struct memlist *)xbootp->bi_phys_install; 165 166 /* 167 * Be careful if high memory usage is limited in startup.c 168 * Since there are holes in the low part of the physical address 169 * space we can treat physmem as a pfn (not just a pgcnt) and 170 * get a conservative upper limit. 171 */ 172 if (physmem != 0 && high_phys > pfn_to_pa(physmem)) 173 high_phys = pfn_to_pa(physmem); 174 175 /* 176 * find the lowest or highest available memory in physinstalled 177 * On 32 bit avoid physmem above 4Gig if PAE isn't enabled 178 */ 179#if defined(__i386) 180 if (xbootp->bi_use_pae == 0 && high_phys > FOUR_GIG) 181 high_phys = FOUR_GIG; 182#endif 183 184 /* 185 * find the highest available memory in physinstalled 186 */ 187 size = P2ROUNDUP(size, align); 188 for (; ml; ml = ml->next) { 189 start = P2ROUNDUP(ml->address, align); 190 end = P2ALIGN(ml->address + ml->size, align); 191 if (start < next_phys) 192 start = P2ROUNDUP(next_phys, align); 193 if (end > high_phys) 194 end = P2ALIGN(high_phys, align); 195 196 if (end <= start) 197 continue; 198 if (end - start < size) 199 continue; 200 201 /* 202 * Early allocations need to use low memory, since 203 * physmem might be further limited by bootenv.rc 204 */ 205 if (early_allocation) { 206 if (pa == 0 || start < pa) 207 pa = start; 208 } else { 209 if (end - size > pa) 210 pa = end - size; 211 } 212 } 213 if (pa != 0) { 214 if (early_allocation) 215 next_phys = pa + size; 216 else 217 high_phys = pa; 218 return (pa); 219 } 220 bop_panic("do_bop_phys_alloc(0x%" PRIx64 ", 0x%" PRIx64 221 ") Out of memory\n", size, align); 222 /*NOTREACHED*/ 223} 224 225static uintptr_t 226alloc_vaddr(size_t size, paddr_t align) 227{ 228 uintptr_t rv; 229 230 next_virt = P2ROUNDUP(next_virt, (uintptr_t)align); 231 rv = (uintptr_t)next_virt; 232 next_virt += size; 233 return (rv); 234} 235 236/* 237 * Allocate virtual memory. The size is always rounded up to a multiple 238 * of base pagesize. 239 */ 240 241/*ARGSUSED*/ 242static caddr_t 243do_bsys_alloc(bootops_t *bop, caddr_t virthint, size_t size, int align) 244{ 245 paddr_t a = align; /* same type as pa for masking */ 246 uint_t pgsize; 247 paddr_t pa; 248 uintptr_t va; 249 ssize_t s; /* the aligned size */ 250 uint_t level; 251 uint_t is_kernel = (virthint != 0); 252 253 if (a < MMU_PAGESIZE) 254 a = MMU_PAGESIZE; 255 else if (!ISP2(a)) 256 prom_panic("do_bsys_alloc() incorrect alignment"); 257 size = P2ROUNDUP(size, MMU_PAGESIZE); 258 259 /* 260 * Use the next aligned virtual address if we weren't given one. 261 */ 262 if (virthint == NULL) { 263 virthint = (caddr_t)alloc_vaddr(size, a); 264 total_bop_alloc_scratch += size; 265 } else { 266 total_bop_alloc_kernel += size; 267 } 268 269 /* 270 * allocate the physical memory 271 */ 272 pa = do_bop_phys_alloc(size, a); 273 274 /* 275 * Add the mappings to the page tables, try large pages first. 276 */ 277 va = (uintptr_t)virthint; 278 s = size; 279 level = 1; 280 pgsize = xbootp->bi_use_pae ? TWO_MEG : FOUR_MEG; 281 if (xbootp->bi_use_largepage && a == pgsize) { 282 while (IS_P2ALIGNED(pa, pgsize) && IS_P2ALIGNED(va, pgsize) && 283 s >= pgsize) { 284 kbm_map(va, pa, level, is_kernel); 285 va += pgsize; 286 pa += pgsize; 287 s -= pgsize; 288 } 289 } 290 291 /* 292 * Map remaining pages use small mappings 293 */ 294 level = 0; 295 pgsize = MMU_PAGESIZE; 296 while (s > 0) { 297 kbm_map(va, pa, level, is_kernel); 298 va += pgsize; 299 pa += pgsize; 300 s -= pgsize; 301 } 302 return (virthint); 303} 304 305/* 306 * Free virtual memory - we'll just ignore these. 307 */ 308/*ARGSUSED*/ 309static void 310do_bsys_free(bootops_t *bop, caddr_t virt, size_t size) 311{ 312 bop_printf(NULL, "do_bsys_free(virt=0x%p, size=0x%lx) ignored\n", 313 (void *)virt, size); 314} 315 316/* 317 * Old interface 318 */ 319/*ARGSUSED*/ 320static caddr_t 321do_bsys_ealloc( 322 bootops_t *bop, 323 caddr_t virthint, 324 size_t size, 325 int align, 326 int flags) 327{ 328 prom_panic("unsupported call to BOP_EALLOC()\n"); 329 return (0); 330} 331 332 333static void 334bsetprop(char *name, int nlen, void *value, int vlen) 335{ 336 uint_t size; 337 uint_t need_size; 338 bootprop_t *b; 339 340 /* 341 * align the size to 16 byte boundary 342 */ 343 size = sizeof (bootprop_t) + nlen + 1 + vlen; 344 size = (size + 0xf) & ~0xf; 345 if (size > curr_space) { 346 need_size = (size + (MMU_PAGEOFFSET)) & MMU_PAGEMASK; 347 curr_page = do_bsys_alloc(NULL, 0, need_size, MMU_PAGESIZE); 348 curr_space = need_size; 349 } 350 351 /* 352 * use a bootprop_t at curr_page and link into list 353 */ 354 b = (bootprop_t *)curr_page; 355 curr_page += sizeof (bootprop_t); 356 curr_space -= sizeof (bootprop_t); 357 b->bp_next = bprops; 358 bprops = b; 359 360 /* 361 * follow by name and ending zero byte 362 */ 363 b->bp_name = curr_page; 364 bcopy(name, curr_page, nlen); 365 curr_page += nlen; 366 *curr_page++ = 0; 367 curr_space -= nlen + 1; 368 369 /* 370 * copy in value, but no ending zero byte 371 */ 372 b->bp_value = curr_page; 373 b->bp_vlen = vlen; 374 if (vlen > 0) { 375 bcopy(value, curr_page, vlen); 376 curr_page += vlen; 377 curr_space -= vlen; 378 } 379 380 /* 381 * align new values of curr_page, curr_space 382 */ 383 while (curr_space & 0xf) { 384 ++curr_page; 385 --curr_space; 386 } 387} 388 389static void 390bsetprops(char *name, char *value) 391{ 392 bsetprop(name, strlen(name), value, strlen(value) + 1); 393} 394 395static void 396bsetprop64(char *name, uint64_t value) 397{ 398 bsetprop(name, strlen(name), (void *)&value, sizeof (value)); 399} 400 401static void 402bsetpropsi(char *name, int value) 403{ 404 char prop_val[32]; 405 406 (void) snprintf(prop_val, sizeof (prop_val), "%d", value); 407 bsetprops(name, prop_val); 408} 409 410/* 411 * to find the size of the buffer to allocate 412 */ 413/*ARGSUSED*/ 414int 415do_bsys_getproplen(bootops_t *bop, const char *name) 416{ 417 bootprop_t *b; 418 419 for (b = bprops; b; b = b->bp_next) { 420 if (strcmp(name, b->bp_name) != 0) 421 continue; 422 return (b->bp_vlen); 423 } 424 return (-1); 425} 426 427/* 428 * get the value associated with this name 429 */ 430/*ARGSUSED*/ 431int 432do_bsys_getprop(bootops_t *bop, const char *name, void *value) 433{ 434 bootprop_t *b; 435 436 for (b = bprops; b; b = b->bp_next) { 437 if (strcmp(name, b->bp_name) != 0) 438 continue; 439 bcopy(b->bp_value, value, b->bp_vlen); 440 return (0); 441 } 442 return (-1); 443} 444 445/* 446 * get the name of the next property in succession from the standalone 447 */ 448/*ARGSUSED*/ 449static char * 450do_bsys_nextprop(bootops_t *bop, char *name) 451{ 452 bootprop_t *b; 453 454 /* 455 * A null name is a special signal for the 1st boot property 456 */ 457 if (name == NULL || strlen(name) == 0) { 458 if (bprops == NULL) 459 return (NULL); 460 return (bprops->bp_name); 461 } 462 463 for (b = bprops; b; b = b->bp_next) { 464 if (name != b->bp_name) 465 continue; 466 b = b->bp_next; 467 if (b == NULL) 468 return (NULL); 469 return (b->bp_name); 470 } 471 return (NULL); 472} 473 474/* 475 * Parse numeric value from a string. Understands decimal, hex, octal, - and ~ 476 */ 477static int 478parse_value(char *p, uint64_t *retval) 479{ 480 int adjust = 0; 481 uint64_t tmp = 0; 482 int digit; 483 int radix = 10; 484 485 *retval = 0; 486 if (*p == '-' || *p == '~') 487 adjust = *p++; 488 489 if (*p == '0') { 490 ++p; 491 if (*p == 0) 492 return (0); 493 if (*p == 'x' || *p == 'X') { 494 radix = 16; 495 ++p; 496 } else { 497 radix = 8; 498 ++p; 499 } 500 } 501 while (*p) { 502 if ('0' <= *p && *p <= '9') 503 digit = *p - '0'; 504 else if ('a' <= *p && *p <= 'f') 505 digit = 10 + *p - 'a'; 506 else if ('A' <= *p && *p <= 'F') 507 digit = 10 + *p - 'A'; 508 else 509 return (-1); 510 if (digit >= radix) 511 return (-1); 512 tmp = tmp * radix + digit; 513 ++p; 514 } 515 if (adjust == '-') 516 tmp = -tmp; 517 else if (adjust == '~') 518 tmp = ~tmp; 519 *retval = tmp; 520 return (0); 521} 522 523/* 524 * 2nd part of building the table of boot properties. This includes: 525 * - values from /boot/solaris/bootenv.rc (ie. eeprom(1m) values) 526 * 527 * lines look like one of: 528 * ^$ 529 * ^# comment till end of line 530 * setprop name 'value' 531 * setprop name value 532 * setprop name "value" 533 * 534 * we do single character I/O since this is really just looking at memory 535 */ 536void 537boot_prop_finish(void) 538{ 539 int fd; 540 char *line; 541 int c; 542 int bytes_read; 543 char *name; 544 int n_len; 545 char *value; 546 int v_len; 547 char *inputdev; /* these override the command line if serial ports */ 548 char *outputdev; 549 char *consoledev; 550 uint64_t lvalue; 551 int use_xencons = 0; 552 553#ifdef __xpv 554 if (!DOMAIN_IS_INITDOMAIN(xen_info)) 555 use_xencons = 1; 556#endif /* __xpv */ 557 558 DBG_MSG("Opening /boot/solaris/bootenv.rc\n"); 559 fd = BRD_OPEN(bfs_ops, "/boot/solaris/bootenv.rc", 0); 560 DBG(fd); 561 562 line = do_bsys_alloc(NULL, NULL, MMU_PAGESIZE, MMU_PAGESIZE); 563 while (fd >= 0) { 564 565 /* 566 * get a line 567 */ 568 for (c = 0; ; ++c) { 569 bytes_read = BRD_READ(bfs_ops, fd, line + c, 1); 570 if (bytes_read == 0) { 571 if (c == 0) 572 goto done; 573 break; 574 } 575 if (line[c] == '\n') 576 break; 577 } 578 line[c] = 0; 579 580 /* 581 * ignore comment lines 582 */ 583 c = 0; 584 while (ISSPACE(line[c])) 585 ++c; 586 if (line[c] == '#' || line[c] == 0) 587 continue; 588 589 /* 590 * must have "setprop " or "setprop\t" 591 */ 592 if (strncmp(line + c, "setprop ", 8) != 0 && 593 strncmp(line + c, "setprop\t", 8) != 0) 594 continue; 595 c += 8; 596 while (ISSPACE(line[c])) 597 ++c; 598 if (line[c] == 0) 599 continue; 600 601 /* 602 * gather up the property name 603 */ 604 name = line + c; 605 n_len = 0; 606 while (line[c] && !ISSPACE(line[c])) 607 ++n_len, ++c; 608 609 /* 610 * gather up the value, if any 611 */ 612 value = ""; 613 v_len = 0; 614 while (ISSPACE(line[c])) 615 ++c; 616 if (line[c] != 0) { 617 value = line + c; 618 while (line[c] && !ISSPACE(line[c])) 619 ++v_len, ++c; 620 } 621 622 if (v_len >= 2 && value[0] == value[v_len - 1] && 623 (value[0] == '\'' || value[0] == '"')) { 624 ++value; 625 v_len -= 2; 626 } 627 name[n_len] = 0; 628 if (v_len > 0) 629 value[v_len] = 0; 630 else 631 continue; 632 633 /* 634 * ignore "boot-file" property, it's now meaningless 635 */ 636 if (strcmp(name, "boot-file") == 0) 637 continue; 638 if (strcmp(name, "boot-args") == 0 && 639 strlen(boot_args) > 0) 640 continue; 641 642 /* 643 * If a property was explicitly set on the command line 644 * it will override a setting in bootenv.rc 645 */ 646 if (do_bsys_getproplen(NULL, name) > 0) 647 continue; 648 649 bsetprop(name, n_len, value, v_len + 1); 650 } 651done: 652 if (fd >= 0) 653 BRD_CLOSE(bfs_ops, fd); 654 655 /* 656 * Check if we have to limit the boot time allocator 657 */ 658 if (do_bsys_getproplen(NULL, "physmem") != -1 && 659 do_bsys_getprop(NULL, "physmem", line) >= 0 && 660 parse_value(line, &lvalue) != -1) { 661 if (0 < lvalue && (lvalue < physmem || physmem == 0)) { 662 physmem = (pgcnt_t)lvalue; 663 DBG(physmem); 664 } 665 } 666 early_allocation = 0; 667 668 /* 669 * check to see if we have to override the default value of the console 670 */ 671 if (!use_xencons) { 672 inputdev = line; 673 v_len = do_bsys_getproplen(NULL, "input-device"); 674 if (v_len > 0) 675 (void) do_bsys_getprop(NULL, "input-device", inputdev); 676 else 677 v_len = 0; 678 inputdev[v_len] = 0; 679 680 outputdev = inputdev + v_len + 1; 681 v_len = do_bsys_getproplen(NULL, "output-device"); 682 if (v_len > 0) 683 (void) do_bsys_getprop(NULL, "output-device", 684 outputdev); 685 else 686 v_len = 0; 687 outputdev[v_len] = 0; 688 689 consoledev = outputdev + v_len + 1; 690 v_len = do_bsys_getproplen(NULL, "console"); 691 if (v_len > 0) 692 (void) do_bsys_getprop(NULL, "console", consoledev); 693 else 694 v_len = 0; 695 consoledev[v_len] = 0; 696 bcons_init2(inputdev, outputdev, consoledev); 697 } else { 698 /* 699 * Ensure console property exists 700 * If not create it as "hypervisor" 701 */ 702 v_len = do_bsys_getproplen(NULL, "console"); 703 if (v_len < 0) 704 bsetprops("console", "hypervisor"); 705 inputdev = outputdev = consoledev = "hypervisor"; 706 bcons_init2(inputdev, outputdev, consoledev); 707 } 708 709 if (strstr((char *)xbootp->bi_cmdline, "prom_debug") || kbm_debug) { 710 value = line; 711 bop_printf(NULL, "\nBoot properties:\n"); 712 name = ""; 713 while ((name = do_bsys_nextprop(NULL, name)) != NULL) { 714 bop_printf(NULL, "\t0x%p %s = ", (void *)name, name); 715 (void) do_bsys_getprop(NULL, name, value); 716 v_len = do_bsys_getproplen(NULL, name); 717 bop_printf(NULL, "len=%d ", v_len); 718 value[v_len] = 0; 719 bop_printf(NULL, "%s\n", value); 720 } 721 } 722} 723 724/* 725 * print formatted output 726 */ 727/*PRINTFLIKE2*/ 728/*ARGSUSED*/ 729void 730bop_printf(bootops_t *bop, const char *fmt, ...) 731{ 732 va_list ap; 733 734 if (have_console == 0) 735 return; 736 737 va_start(ap, fmt); 738 (void) vsnprintf(buffer, BUFFERSIZE, fmt, ap); 739 va_end(ap); 740 PUT_STRING(buffer); 741} 742 743/* 744 * Another panic() variant; this one can be used even earlier during boot than 745 * prom_panic(). 746 */ 747/*PRINTFLIKE1*/ 748void 749bop_panic(const char *fmt, ...) 750{ 751 va_list ap; 752 753 va_start(ap, fmt); 754 bop_printf(NULL, fmt, ap); 755 va_end(ap); 756 757 bop_printf(NULL, "\nPress any key to reboot.\n"); 758 (void) bcons_getchar(); 759 bop_printf(NULL, "Resetting...\n"); 760 pc_reset(); 761} 762 763/* 764 * Do a real mode interrupt BIOS call 765 */ 766typedef struct bios_regs { 767 unsigned short ax, bx, cx, dx, si, di, bp, es, ds; 768} bios_regs_t; 769typedef int (*bios_func_t)(int, bios_regs_t *); 770 771/*ARGSUSED*/ 772static void 773do_bsys_doint(bootops_t *bop, int intnum, struct bop_regs *rp) 774{ 775#if defined(__xpv) 776 prom_panic("unsupported call to BOP_DOINT()\n"); 777#else /* __xpv */ 778 static int firsttime = 1; 779 bios_func_t bios_func = (bios_func_t)(void *)(uintptr_t)0x5000; 780 bios_regs_t br; 781 782 /* 783 * The first time we do this, we have to copy the pre-packaged 784 * low memory bios call code image into place. 785 */ 786 if (firsttime) { 787 extern char bios_image[]; 788 extern uint32_t bios_size; 789 790 bcopy(bios_image, (void *)bios_func, bios_size); 791 firsttime = 0; 792 } 793 794 br.ax = rp->eax.word.ax; 795 br.bx = rp->ebx.word.bx; 796 br.cx = rp->ecx.word.cx; 797 br.dx = rp->edx.word.dx; 798 br.bp = rp->ebp.word.bp; 799 br.si = rp->esi.word.si; 800 br.di = rp->edi.word.di; 801 br.ds = rp->ds; 802 br.es = rp->es; 803 804 DBG_MSG("Doing BIOS call..."); 805 DBG(br.ax); 806 DBG(br.bx); 807 DBG(br.dx); 808 rp->eflags = bios_func(intnum, &br); 809 DBG_MSG("done\n"); 810 811 rp->eax.word.ax = br.ax; 812 rp->ebx.word.bx = br.bx; 813 rp->ecx.word.cx = br.cx; 814 rp->edx.word.dx = br.dx; 815 rp->ebp.word.bp = br.bp; 816 rp->esi.word.si = br.si; 817 rp->edi.word.di = br.di; 818 rp->ds = br.ds; 819 rp->es = br.es; 820#endif /* __xpv */ 821} 822 823static struct boot_syscalls bop_sysp = { 824 bcons_getchar, 825 bcons_putchar, 826 bcons_ischar, 827}; 828 829static char *whoami; 830 831#define BUFLEN 64 832 833#if defined(__xpv) 834 835static char namebuf[32]; 836 837static void 838xen_parse_props(char *s, char *prop_map[], int n_prop) 839{ 840 char **prop_name = prop_map; 841 char *cp = s, *scp; 842 843 do { 844 scp = cp; 845 while ((*cp != NULL) && (*cp != ':')) 846 cp++; 847 848 if ((scp != cp) && (*prop_name != NULL)) { 849 *cp = NULL; 850 bsetprops(*prop_name, scp); 851 } 852 853 cp++; 854 prop_name++; 855 n_prop--; 856 } while (n_prop > 0); 857} 858 859#define VBDPATHLEN 64 860 861/* 862 * parse the 'xpv-root' property to create properties used by 863 * ufs_mountroot. 864 */ 865static void 866xen_vbdroot_props(char *s) 867{ 868 char vbdpath[VBDPATHLEN] = "/xpvd/xdf@"; 869 const char lnamefix[] = "/dev/dsk/c0d"; 870 char *pnp; 871 char *prop_p; 872 char mi; 873 short minor; 874 long addr = 0; 875 876 pnp = vbdpath + strlen(vbdpath); 877 prop_p = s + strlen(lnamefix); 878 while ((*prop_p != '\0') && (*prop_p != 's') && (*prop_p != 'p')) 879 addr = addr * 10 + *prop_p++ - '0'; 880 (void) snprintf(pnp, VBDPATHLEN, "%lx", addr); 881 pnp = vbdpath + strlen(vbdpath); 882 if (*prop_p == 's') 883 mi = 'a'; 884 else if (*prop_p == 'p') 885 mi = 'q'; 886 else 887 ASSERT(0); /* shouldn't be here */ 888 prop_p++; 889 ASSERT(*prop_p != '\0'); 890 if (ISDIGIT(*prop_p)) { 891 minor = *prop_p - '0'; 892 prop_p++; 893 if (ISDIGIT(*prop_p)) { 894 minor = minor * 10 + *prop_p - '0'; 895 } 896 } else { 897 /* malformed root path, use 0 as default */ 898 minor = 0; 899 } 900 ASSERT(minor < 16); /* at most 16 partitions */ 901 mi += minor; 902 *pnp++ = ':'; 903 *pnp++ = mi; 904 *pnp++ = '\0'; 905 bsetprops("fstype", "ufs"); 906 bsetprops("bootpath", vbdpath); 907 908 DBG_MSG("VBD bootpath set to "); 909 DBG_MSG(vbdpath); 910 DBG_MSG("\n"); 911} 912 913/* 914 * parse the xpv-nfsroot property to create properties used by 915 * nfs_mountroot. 916 */ 917static void 918xen_nfsroot_props(char *s) 919{ 920 char *prop_map[] = { 921 BP_SERVER_IP, /* server IP address */ 922 BP_SERVER_NAME, /* server hostname */ 923 BP_SERVER_PATH, /* root path */ 924 }; 925 int n_prop = sizeof (prop_map) / sizeof (prop_map[0]); 926 927 bsetprop("fstype", 6, "nfs", 4); 928 929 xen_parse_props(s, prop_map, n_prop); 930 931 /* 932 * If a server name wasn't specified, use a default. 933 */ 934 if (do_bsys_getproplen(NULL, BP_SERVER_NAME) == -1) 935 bsetprops(BP_SERVER_NAME, "unknown"); 936} 937 938/* 939 * Extract our IP address, etc. from the "xpv-ip" property. 940 */ 941static void 942xen_ip_props(char *s) 943{ 944 char *prop_map[] = { 945 BP_HOST_IP, /* IP address */ 946 NULL, /* NFS server IP address (ignored in */ 947 /* favour of xpv-nfsroot) */ 948 BP_ROUTER_IP, /* IP gateway */ 949 BP_SUBNET_MASK, /* IP subnet mask */ 950 "xpv-hostname", /* hostname (ignored) */ 951 BP_NETWORK_INTERFACE, /* interface name */ 952 "xpv-hcp", /* host configuration protocol */ 953 }; 954 int n_prop = sizeof (prop_map) / sizeof (prop_map[0]); 955 char ifname[IFNAMSIZ]; 956 957 xen_parse_props(s, prop_map, n_prop); 958 959 /* 960 * A Linux dom0 administrator expects all interfaces to be 961 * called "ethX", which is not the case here. 962 * 963 * If the interface name specified is "eth0", presume that 964 * this is really intended to be "xnf0" (the first domU -> 965 * dom0 interface for this domain). 966 */ 967 if ((do_bsys_getprop(NULL, BP_NETWORK_INTERFACE, ifname) == 0) && 968 (strcmp("eth0", ifname) == 0)) { 969 bsetprops(BP_NETWORK_INTERFACE, "xnf0"); 970 bop_printf(NULL, 971 "network interface name 'eth0' replaced with 'xnf0'\n"); 972 } 973} 974 975#else /* __xpv */ 976 977static void 978setup_rarp_props(struct sol_netinfo *sip) 979{ 980 char buf[BUFLEN]; /* to hold ip/mac addrs */ 981 uint8_t *val; 982 983 val = (uint8_t *)&sip->sn_ciaddr; 984 (void) snprintf(buf, BUFLEN, "%d.%d.%d.%d", 985 val[0], val[1], val[2], val[3]); 986 bsetprops(BP_HOST_IP, buf); 987 988 val = (uint8_t *)&sip->sn_siaddr; 989 (void) snprintf(buf, BUFLEN, "%d.%d.%d.%d", 990 val[0], val[1], val[2], val[3]); 991 bsetprops(BP_SERVER_IP, buf); 992 993 if (sip->sn_giaddr != 0) { 994 val = (uint8_t *)&sip->sn_giaddr; 995 (void) snprintf(buf, BUFLEN, "%d.%d.%d.%d", 996 val[0], val[1], val[2], val[3]); 997 bsetprops(BP_ROUTER_IP, buf); 998 } 999 1000 if (sip->sn_netmask != 0) { 1001 val = (uint8_t *)&sip->sn_netmask; 1002 (void) snprintf(buf, BUFLEN, "%d.%d.%d.%d", 1003 val[0], val[1], val[2], val[3]); 1004 bsetprops(BP_SUBNET_MASK, buf); 1005 } 1006 1007 if (sip->sn_mactype != 4 || sip->sn_maclen != 6) { 1008 bop_printf(NULL, "unsupported mac type %d, mac len %d\n", 1009 sip->sn_mactype, sip->sn_maclen); 1010 } else { 1011 val = sip->sn_macaddr; 1012 (void) snprintf(buf, BUFLEN, "%x:%x:%x:%x:%x:%x", 1013 val[0], val[1], val[2], val[3], val[4], val[5]); 1014 bsetprops(BP_BOOT_MAC, buf); 1015 } 1016} 1017 1018#endif /* __xpv */ 1019 1020/* 1021 * 1st pass at building the table of boot properties. This includes: 1022 * - values set on the command line: -B a=x,b=y,c=z .... 1023 * - known values we just compute (ie. from xbootp) 1024 * - values from /boot/solaris/bootenv.rc (ie. eeprom(1m) values) 1025 * 1026 * the grub command line looked like: 1027 * kernel boot-file [-B prop=value[,prop=value]...] [boot-args] 1028 * 1029 * whoami is the same as boot-file 1030 */ 1031static void 1032build_boot_properties(void) 1033{ 1034 char *name; 1035 int name_len; 1036 char *value; 1037 int value_len; 1038 struct boot_modules *bm; 1039 char *propbuf; 1040 int quoted = 0; 1041 int boot_arg_len; 1042#ifndef __xpv 1043 static int stdout_val = 0; 1044 uchar_t boot_device; 1045 char str[3]; 1046 multiboot_info_t *mbi; 1047 int netboot; 1048 struct sol_netinfo *sip; 1049#endif 1050 1051 /* 1052 * These have to be done first, so that kobj_mount_root() works 1053 */ 1054 DBG_MSG("Building boot properties\n"); 1055 propbuf = do_bsys_alloc(NULL, NULL, MMU_PAGESIZE, 0); 1056 DBG((uintptr_t)propbuf); 1057 if (xbootp->bi_module_cnt > 0) { 1058 bm = xbootp->bi_modules; 1059 bsetprop64("ramdisk_start", (uint64_t)(uintptr_t)bm->bm_addr); 1060 bsetprop64("ramdisk_end", (uint64_t)(uintptr_t)bm->bm_addr + 1061 bm->bm_size); 1062 } 1063 1064 DBG_MSG("Parsing command line for boot properties\n"); 1065 value = xbootp->bi_cmdline; 1066 1067 /* 1068 * allocate memory to collect boot_args into 1069 */ 1070 boot_arg_len = strlen(xbootp->bi_cmdline) + 1; 1071 boot_args = do_bsys_alloc(NULL, NULL, boot_arg_len, MMU_PAGESIZE); 1072 boot_args[0] = 0; 1073 boot_arg_len = 0; 1074 1075#ifndef __xpv 1076 saved_cmdline_len = strlen(xbootp->bi_cmdline) + 1; 1077 if (saved_cmdline_len > FASTBOOT_SAVED_CMDLINE_LEN) { 1078 DBG(saved_cmdline_len); 1079 DBG_MSG("Command line too long: clearing fastreboot_capable\n"); 1080 fastreboot_capable = 0; 1081 } else { 1082 bcopy((void *)(xbootp->bi_cmdline), (void *)saved_cmdline, 1083 saved_cmdline_len); 1084 saved_cmdline[saved_cmdline_len - 1] = '\0'; 1085 } 1086#endif 1087 1088 1089#ifdef __xpv 1090 /* 1091 * Xen puts a lot of device information in front of the kernel name 1092 * let's grab them and make them boot properties. The first 1093 * string w/o an "=" in it will be the boot-file property. 1094 */ 1095 (void) strcpy(namebuf, "xpv-"); 1096 for (;;) { 1097 /* 1098 * get to next property 1099 */ 1100 while (ISSPACE(*value)) 1101 ++value; 1102 name = value; 1103 /* 1104 * look for an "=" 1105 */ 1106 while (*value && !ISSPACE(*value) && *value != '=') { 1107 value++; 1108 } 1109 if (*value != '=') { /* no "=" in the property */ 1110 value = name; 1111 break; 1112 } 1113 name_len = value - name; 1114 value_len = 0; 1115 /* 1116 * skip over the "=" 1117 */ 1118 value++; 1119 while (value[value_len] && !ISSPACE(value[value_len])) { 1120 ++value_len; 1121 } 1122 /* 1123 * build property name with "xpv-" prefix 1124 */ 1125 if (name_len + 4 > 32) { /* skip if name too long */ 1126 value += value_len; 1127 continue; 1128 } 1129 bcopy(name, &namebuf[4], name_len); 1130 name_len += 4; 1131 namebuf[name_len] = 0; 1132 bcopy(value, propbuf, value_len); 1133 propbuf[value_len] = 0; 1134 bsetprops(namebuf, propbuf); 1135 1136 /* 1137 * xpv-root is set to the logical disk name of the xen 1138 * VBD when booting from a disk-based filesystem. 1139 */ 1140 if (strcmp(namebuf, "xpv-root") == 0) 1141 xen_vbdroot_props(propbuf); 1142 /* 1143 * While we're here, if we have a "xpv-nfsroot" property 1144 * then we need to set "fstype" to "nfs" so we mount 1145 * our root from the nfs server. Also parse the xpv-nfsroot 1146 * property to create the properties that nfs_mountroot will 1147 * need to find the root and mount it. 1148 */ 1149 if (strcmp(namebuf, "xpv-nfsroot") == 0) 1150 xen_nfsroot_props(propbuf); 1151 1152 if (strcmp(namebuf, "xpv-ip") == 0) 1153 xen_ip_props(propbuf); 1154 value += value_len; 1155 } 1156#endif 1157 1158 while (ISSPACE(*value)) 1159 ++value; 1160 /* 1161 * value now points at the boot-file 1162 */ 1163 value_len = 0; 1164 while (value[value_len] && !ISSPACE(value[value_len])) 1165 ++value_len; 1166 if (value_len > 0) { 1167 whoami = propbuf; 1168 bcopy(value, whoami, value_len); 1169 whoami[value_len] = 0; 1170 bsetprops("boot-file", whoami); 1171 /* 1172 * strip leading path stuff from whoami, so running from 1173 * PXE/miniroot makes sense. 1174 */ 1175 if (strstr(whoami, "/platform/") != NULL) 1176 whoami = strstr(whoami, "/platform/"); 1177 bsetprops("whoami", whoami); 1178 } 1179 1180 /* 1181 * Values forcibly set boot properties on the command line via -B. 1182 * Allow use of quotes in values. Other stuff goes on kernel 1183 * command line. 1184 */ 1185 name = value + value_len; 1186 while (*name != 0) { 1187 /* 1188 * anything not " -B" is copied to the command line 1189 */ 1190 if (!ISSPACE(name[0]) || name[1] != '-' || name[2] != 'B') { 1191 boot_args[boot_arg_len++] = *name; 1192 boot_args[boot_arg_len] = 0; 1193 ++name; 1194 continue; 1195 } 1196 1197 /* 1198 * skip the " -B" and following white space 1199 */ 1200 name += 3; 1201 while (ISSPACE(*name)) 1202 ++name; 1203 while (*name && !ISSPACE(*name)) { 1204 value = strstr(name, "="); 1205 if (value == NULL) 1206 break; 1207 name_len = value - name; 1208 ++value; 1209 value_len = 0; 1210 quoted = 0; 1211 for (; ; ++value_len) { 1212 if (!value[value_len]) 1213 break; 1214 1215 /* 1216 * is this value quoted? 1217 */ 1218 if (value_len == 0 && 1219 (value[0] == '\'' || value[0] == '"')) { 1220 quoted = value[0]; 1221 ++value_len; 1222 } 1223 1224 /* 1225 * In the quote accept any character, 1226 * but look for ending quote. 1227 */ 1228 if (quoted) { 1229 if (value[value_len] == quoted) 1230 quoted = 0; 1231 continue; 1232 } 1233 1234 /* 1235 * a comma or white space ends the value 1236 */ 1237 if (value[value_len] == ',' || 1238 ISSPACE(value[value_len])) 1239 break; 1240 } 1241 1242 if (value_len == 0) { 1243 bsetprop(name, name_len, "true", 5); 1244 } else { 1245 char *v = value; 1246 int l = value_len; 1247 if (v[0] == v[l - 1] && 1248 (v[0] == '\'' || v[0] == '"')) { 1249 ++v; 1250 l -= 2; 1251 } 1252 bcopy(v, propbuf, l); 1253 propbuf[l] = '\0'; 1254 bsetprop(name, name_len, propbuf, 1255 l + 1); 1256 } 1257 name = value + value_len; 1258 while (*name == ',') 1259 ++name; 1260 } 1261 } 1262 1263 /* 1264 * set boot-args property 1265 * 1275 name is bootargs, so set 1266 * that too 1267 */ 1268 bsetprops("boot-args", boot_args); 1269 bsetprops("bootargs", boot_args); 1270 1271#ifndef __xpv 1272 /* 1273 * set the BIOS boot device from GRUB 1274 */ 1275 netboot = 0; 1276 mbi = xbootp->bi_mb_info; 1277 bcopy(mbi, &saved_mbi, sizeof (multiboot_info_t)); 1278 if (mbi->mmap_length > sizeof (saved_mmap)) { 1279 DBG_MSG("mbi->mmap_length too big: clearing " 1280 "fastreboot_capable\n"); 1281 fastreboot_capable = 0; 1282 } else { 1283 bcopy((void *)(uintptr_t)mbi->mmap_addr, (void *)saved_mmap, 1284 mbi->mmap_length); 1285 } 1286 1287 if (mbi->drives_length > sizeof (saved_drives)) { 1288 DBG(mbi->drives_length); 1289 DBG_MSG("mbi->drives_length too big: clearing " 1290 "fastreboot_capable\n"); 1291 fastreboot_capable = 0; 1292 } else { 1293 bcopy((void *)(uintptr_t)mbi->drives_addr, (void *)saved_drives, 1294 mbi->drives_length); 1295 } 1296 1297 if (mbi != NULL && mbi->flags & 0x2) { 1298 boot_device = mbi->boot_device >> 24; 1299 if (boot_device == 0x20) 1300 netboot++; 1301 str[0] = (boot_device >> 4) + '0'; 1302 str[1] = (boot_device & 0xf) + '0'; 1303 str[2] = 0; 1304 bsetprops("bios-boot-device", str); 1305 } else { 1306 netboot = 1; 1307 } 1308 1309 /* 1310 * In the netboot case, drives_info is overloaded with the dhcp ack. 1311 * This is not multiboot compliant and requires special pxegrub! 1312 */ 1313 if (netboot && mbi->drives_length != 0) { 1314 sip = (struct sol_netinfo *)(uintptr_t)mbi->drives_addr; 1315 if (sip->sn_infotype == SN_TYPE_BOOTP) 1316 bsetprop("bootp-response", sizeof ("bootp-response"), 1317 (void *)(uintptr_t)mbi->drives_addr, 1318 mbi->drives_length); 1319 else if (sip->sn_infotype == SN_TYPE_RARP) 1320 setup_rarp_props(sip); 1321 } 1322 bsetprop("stdout", strlen("stdout"), 1323 &stdout_val, sizeof (stdout_val)); 1324#endif /* __xpv */ 1325 1326 /* 1327 * more conjured up values for made up things.... 1328 */ 1329#if defined(__xpv) 1330 bsetprops("mfg-name", "i86xpv"); 1331 bsetprops("impl-arch-name", "i86xpv"); 1332#else 1333 bsetprops("mfg-name", "i86pc"); 1334 bsetprops("impl-arch-name", "i86pc"); 1335#endif 1336 1337 /* 1338 * Build firmware-provided system properties 1339 */ 1340 build_firmware_properties(); 1341 1342 /* 1343 * XXPV 1344 * 1345 * Find out what these are: 1346 * - cpuid_feature_ecx_include 1347 * - cpuid_feature_ecx_exclude 1348 * - cpuid_feature_edx_include 1349 * - cpuid_feature_edx_exclude 1350 * 1351 * Find out what these are in multiboot: 1352 * - netdev-path 1353 * - fstype 1354 */ 1355} 1356 1357#ifdef __xpv 1358/* 1359 * Under the Hypervisor, memory usable for DMA may be scarce. One 1360 * very likely large pool of DMA friendly memory is occupied by 1361 * the boot_archive, as it was loaded by grub into low MFNs. 1362 * 1363 * Here we free up that memory by copying the boot archive to what are 1364 * likely higher MFN pages and then swapping the mfn/pfn mappings. 1365 */ 1366#define PFN_2GIG 0x80000 1367static void 1368relocate_boot_archive(void) 1369{ 1370 mfn_t max_mfn = HYPERVISOR_memory_op(XENMEM_maximum_ram_page, NULL); 1371 struct boot_modules *bm = xbootp->bi_modules; 1372 uintptr_t va; 1373 pfn_t va_pfn; 1374 mfn_t va_mfn; 1375 caddr_t copy; 1376 pfn_t copy_pfn; 1377 mfn_t copy_mfn; 1378 size_t len; 1379 int slop; 1380 int total = 0; 1381 int relocated = 0; 1382 int mmu_update_return; 1383 mmu_update_t t[2]; 1384 x86pte_t pte; 1385 1386 /* 1387 * If all MFN's are below 2Gig, don't bother doing this. 1388 */ 1389 if (max_mfn < PFN_2GIG) 1390 return; 1391 if (xbootp->bi_module_cnt < 1) { 1392 DBG_MSG("no boot_archive!"); 1393 return; 1394 } 1395 1396 DBG_MSG("moving boot_archive to high MFN memory\n"); 1397 va = (uintptr_t)bm->bm_addr; 1398 len = bm->bm_size; 1399 slop = va & MMU_PAGEOFFSET; 1400 if (slop) { 1401 va += MMU_PAGESIZE - slop; 1402 len -= MMU_PAGESIZE - slop; 1403 } 1404 len = P2ALIGN(len, MMU_PAGESIZE); 1405 1406 /* 1407 * Go through all boot_archive pages, swapping any low MFN pages 1408 * with memory at next_phys. 1409 */ 1410 while (len != 0) { 1411 ++total; 1412 va_pfn = mmu_btop(va - ONE_GIG); 1413 va_mfn = mfn_list[va_pfn]; 1414 if (mfn_list[va_pfn] < PFN_2GIG) { 1415 copy = kbm_remap_window(next_phys, 1); 1416 bcopy((void *)va, copy, MMU_PAGESIZE); 1417 copy_pfn = mmu_btop(next_phys); 1418 copy_mfn = mfn_list[copy_pfn]; 1419 1420 pte = mfn_to_ma(copy_mfn) | PT_NOCONSIST | PT_VALID; 1421 if (HYPERVISOR_update_va_mapping(va, pte, 1422 UVMF_INVLPG | UVMF_LOCAL)) 1423 bop_panic("relocate_boot_archive(): " 1424 "HYPERVISOR_update_va_mapping() failed"); 1425 1426 mfn_list[va_pfn] = copy_mfn; 1427 mfn_list[copy_pfn] = va_mfn; 1428 1429 t[0].ptr = mfn_to_ma(copy_mfn) | MMU_MACHPHYS_UPDATE; 1430 t[0].val = va_pfn; 1431 t[1].ptr = mfn_to_ma(va_mfn) | MMU_MACHPHYS_UPDATE; 1432 t[1].val = copy_pfn; 1433 if (HYPERVISOR_mmu_update(t, 2, &mmu_update_return, 1434 DOMID_SELF) != 0 || mmu_update_return != 2) 1435 bop_panic("relocate_boot_archive(): " 1436 "HYPERVISOR_mmu_update() failed"); 1437 1438 next_phys += MMU_PAGESIZE; 1439 ++relocated; 1440 } 1441 len -= MMU_PAGESIZE; 1442 va += MMU_PAGESIZE; 1443 } 1444 DBG_MSG("Relocated pages:\n"); 1445 DBG(relocated); 1446 DBG_MSG("Out of total pages:\n"); 1447 DBG(total); 1448} 1449#endif /* __xpv */ 1450 1451#if !defined(__xpv) 1452/* 1453 * Install a temporary IDT that lets us catch errors in the boot time code. 1454 * We shouldn't get any faults at all while this is installed, so we'll 1455 * just generate a traceback and exit. 1456 */ 1457#ifdef __amd64 1458static const int bcode_sel = B64CODE_SEL; 1459#else 1460static const int bcode_sel = B32CODE_SEL; 1461#endif 1462 1463/* 1464 * simple description of a stack frame (args are 32 bit only currently) 1465 */ 1466typedef struct bop_frame { 1467 struct bop_frame *old_frame; 1468 pc_t retaddr; 1469 long arg[1]; 1470} bop_frame_t; 1471 1472void 1473bop_traceback(bop_frame_t *frame) 1474{ 1475 pc_t pc; 1476 int cnt; 1477 int a; 1478 char *ksym; 1479 ulong_t off; 1480 1481 bop_printf(NULL, "Stack traceback:\n"); 1482 for (cnt = 0; cnt < 30; ++cnt) { /* up to 30 frames */ 1483 pc = frame->retaddr; 1484 if (pc == 0) 1485 break; 1486 ksym = kobj_getsymname(pc, &off); 1487 if (ksym) 1488 bop_printf(NULL, " %s+%lx", ksym, off); 1489 else 1490 bop_printf(NULL, " 0x%lx", pc); 1491 1492 frame = frame->old_frame; 1493 if (frame == 0) { 1494 bop_printf(NULL, "\n"); 1495 break; 1496 } 1497 for (a = 0; a < 6; ++a) { /* try for 6 args */ 1498#if defined(__i386) 1499 if ((void *)&frame->arg[a] == (void *)frame->old_frame) 1500 break; 1501 if (a == 0) 1502 bop_printf(NULL, "("); 1503 else 1504 bop_printf(NULL, ","); 1505 bop_printf(NULL, "0x%lx", frame->arg[a]); 1506#endif 1507 } 1508 bop_printf(NULL, ")\n"); 1509 } 1510} 1511 1512struct trapframe { 1513 ulong_t frame_ptr; /* %[er]bp pushed by our code */ 1514 ulong_t error_code; /* optional */ 1515 ulong_t inst_ptr; 1516 ulong_t code_seg; 1517 ulong_t flags_reg; 1518#ifdef __amd64 1519 ulong_t stk_ptr; 1520 ulong_t stk_seg; 1521#endif 1522}; 1523 1524void 1525bop_trap(struct trapframe *tf) 1526{ 1527 bop_frame_t fakeframe; 1528 static int depth = 0; 1529 1530 /* 1531 * Check for an infinite loop of traps. 1532 */ 1533 if (++depth > 2) 1534 bop_panic("Nested trap"); 1535 1536 /* 1537 * adjust the tf for optional error_code by detecting the code selector 1538 */ 1539 if (tf->code_seg != bcode_sel) 1540 tf = (struct trapframe *)((uintptr_t)tf - sizeof (ulong_t)); 1541 1542 bop_printf(NULL, "Unexpected trap\n"); 1543 bop_printf(NULL, "instruction pointer 0x%lx\n", tf->inst_ptr); 1544 bop_printf(NULL, "error code, optional 0x%lx\n", 1545 tf->error_code & 0xffffffff); 1546 bop_printf(NULL, "code segment 0x%lx\n", tf->code_seg & 0xffff); 1547 bop_printf(NULL, "flags register 0x%lx\n", tf->flags_reg); 1548#ifdef __amd64 1549 bop_printf(NULL, "return %%rsp 0x%lx\n", tf->stk_ptr); 1550 bop_printf(NULL, "return %%ss 0x%lx\n", tf->stk_seg & 0xffff); 1551#endif 1552 fakeframe.old_frame = (bop_frame_t *)tf->frame_ptr; 1553 fakeframe.retaddr = (pc_t)tf->inst_ptr; 1554 bop_printf(NULL, "Attempting stack backtrace:\n"); 1555 bop_traceback(&fakeframe); 1556 bop_panic("unexpected trap in early boot"); 1557} 1558 1559extern void bop_trap_handler(void); 1560 1561static gate_desc_t *bop_idt; 1562 1563static desctbr_t bop_idt_info; 1564 1565static void 1566bop_idt_init(void) 1567{ 1568 int t; 1569 1570 bop_idt = (gate_desc_t *) 1571 do_bsys_alloc(NULL, NULL, MMU_PAGESIZE, MMU_PAGESIZE); 1572 bzero(bop_idt, MMU_PAGESIZE); 1573 for (t = 0; t < NIDT; ++t) { 1574 set_gatesegd(&bop_idt[t], &bop_trap_handler, bcode_sel, 1575 SDT_SYSIGT, TRP_KPL); 1576 } 1577 bop_idt_info.dtr_limit = (NIDT * sizeof (gate_desc_t)) - 1; 1578 bop_idt_info.dtr_base = (uintptr_t)bop_idt; 1579 wr_idtr(&bop_idt_info); 1580} 1581#endif /* !defined(__xpv) */ 1582 1583/* 1584 * This is where we enter the kernel. It dummies up the boot_ops and 1585 * boot_syscalls vectors and jumps off to _kobj_boot() 1586 */ 1587void 1588_start(struct xboot_info *xbp) 1589{ 1590 bootops_t *bops = &bootop; 1591 extern void _kobj_boot(); 1592 1593 /* 1594 * 1st off - initialize the console for any error messages 1595 */ 1596 xbootp = xbp; 1597#ifdef __xpv 1598 HYPERVISOR_shared_info = (void *)xbootp->bi_shared_info; 1599 xen_info = xbootp->bi_xen_start_info; 1600#endif 1601 bcons_init((void *)xbootp->bi_cmdline); 1602 have_console = 1; 1603 1604 if (*((uint32_t *)(FASTBOOT_SWTCH_PA + FASTBOOT_STACK_OFFSET)) == 1605 FASTBOOT_MAGIC) { 1606 post_fastreboot = 1; 1607 *((uint32_t *)(FASTBOOT_SWTCH_PA + FASTBOOT_STACK_OFFSET)) = 0; 1608 } 1609 1610 /* 1611 * enable debugging 1612 */ 1613 if (strstr((char *)xbootp->bi_cmdline, "kbm_debug")) 1614 kbm_debug = 1; 1615 1616 DBG_MSG("\n\n*** Entered Solaris in _start() cmdline is: "); 1617 DBG_MSG((char *)xbootp->bi_cmdline); 1618 DBG_MSG("\n\n\n"); 1619 1620 /* 1621 * physavail is no longer used by startup 1622 */ 1623 bm.physinstalled = xbp->bi_phys_install; 1624 bm.pcimem = xbp->bi_pcimem; 1625 bm.physavail = NULL; 1626 1627 /* 1628 * initialize the boot time allocator 1629 */ 1630 next_phys = xbootp->bi_next_paddr; 1631 DBG(next_phys); 1632 next_virt = (uintptr_t)xbootp->bi_next_vaddr; 1633 DBG(next_virt); 1634 DBG_MSG("Initializing boot time memory management..."); 1635#ifdef __xpv 1636 { 1637 xen_platform_parameters_t p; 1638 1639 /* This call shouldn't fail, dboot already did it once. */ 1640 (void) HYPERVISOR_xen_version(XENVER_platform_parameters, &p); 1641 mfn_to_pfn_mapping = (pfn_t *)(xen_virt_start = p.virt_start); 1642 DBG(xen_virt_start); 1643 } 1644#endif 1645 kbm_init(xbootp); 1646 DBG_MSG("done\n"); 1647 1648 /* 1649 * Fill in the bootops vector 1650 */ 1651 bops->bsys_version = BO_VERSION; 1652 bops->boot_mem = &bm; 1653 bops->bsys_alloc = do_bsys_alloc; 1654 bops->bsys_free = do_bsys_free; 1655 bops->bsys_getproplen = do_bsys_getproplen; 1656 bops->bsys_getprop = do_bsys_getprop; 1657 bops->bsys_nextprop = do_bsys_nextprop; 1658 bops->bsys_printf = bop_printf; 1659 bops->bsys_doint = do_bsys_doint; 1660 1661 /* 1662 * BOP_EALLOC() is no longer needed 1663 */ 1664 bops->bsys_ealloc = do_bsys_ealloc; 1665 1666#ifdef __xpv 1667 /* 1668 * On domain 0 we need to free up some physical memory that is 1669 * usable for DMA. Since GRUB loaded the boot_archive, it is 1670 * sitting in low MFN memory. We'll relocated the boot archive 1671 * pages to high PFN memory. 1672 */ 1673 if (DOMAIN_IS_INITDOMAIN(xen_info)) 1674 relocate_boot_archive(); 1675#endif 1676 1677#ifndef __xpv 1678 /* 1679 * Install an IDT to catch early pagefaults (shouldn't have any). 1680 * Also needed for kmdb. 1681 */ 1682 bop_idt_init(); 1683#endif 1684 1685 /* 1686 * Start building the boot properties from the command line 1687 */ 1688 DBG_MSG("Initializing boot properties:\n"); 1689 build_boot_properties(); 1690 1691 if (strstr((char *)xbootp->bi_cmdline, "prom_debug") || kbm_debug) { 1692 char *name; 1693 char *value; 1694 char *cp; 1695 int len; 1696 1697 value = do_bsys_alloc(NULL, NULL, MMU_PAGESIZE, MMU_PAGESIZE); 1698 bop_printf(NULL, "\nBoot properties:\n"); 1699 name = ""; 1700 while ((name = do_bsys_nextprop(NULL, name)) != NULL) { 1701 bop_printf(NULL, "\t0x%p %s = ", (void *)name, name); 1702 (void) do_bsys_getprop(NULL, name, value); 1703 len = do_bsys_getproplen(NULL, name); 1704 bop_printf(NULL, "len=%d ", len); 1705 value[len] = 0; 1706 for (cp = value; *cp; ++cp) { 1707 if (' ' <= *cp && *cp <= '~') 1708 bop_printf(NULL, "%c", *cp); 1709 else 1710 bop_printf(NULL, "-0x%x-", *cp); 1711 } 1712 bop_printf(NULL, "\n"); 1713 } 1714 } 1715 1716 /* 1717 * jump into krtld... 1718 */ 1719 _kobj_boot(&bop_sysp, NULL, bops, NULL); 1720} 1721 1722 1723/*ARGSUSED*/ 1724static caddr_t 1725no_more_alloc(bootops_t *bop, caddr_t virthint, size_t size, int align) 1726{ 1727 panic("Attempt to bsys_alloc() too late\n"); 1728 return (NULL); 1729} 1730 1731/*ARGSUSED*/ 1732static void 1733no_more_free(bootops_t *bop, caddr_t virt, size_t size) 1734{ 1735 panic("Attempt to bsys_free() too late\n"); 1736} 1737 1738void 1739bop_no_more_mem(void) 1740{ 1741 DBG(total_bop_alloc_scratch); 1742 DBG(total_bop_alloc_kernel); 1743 bootops->bsys_alloc = no_more_alloc; 1744 bootops->bsys_free = no_more_free; 1745} 1746 1747 1748#ifndef __xpv 1749/* 1750 * Set ACPI firmware properties 1751 */ 1752 1753static caddr_t 1754vmap_phys(size_t length, paddr_t pa) 1755{ 1756 paddr_t start, end; 1757 caddr_t va; 1758 size_t len, page; 1759 1760 start = P2ALIGN(pa, MMU_PAGESIZE); 1761 end = P2ROUNDUP(pa + length, MMU_PAGESIZE); 1762 len = end - start; 1763 va = (caddr_t)alloc_vaddr(len, MMU_PAGESIZE); 1764 for (page = 0; page < len; page += MMU_PAGESIZE) 1765 kbm_map((uintptr_t)va + page, start + page, 0, 0); 1766 return (va + (pa & MMU_PAGEOFFSET)); 1767} 1768 1769static uint8_t 1770checksum_table(uint8_t *tp, size_t len) 1771{ 1772 uint8_t sum = 0; 1773 1774 while (len-- > 0) 1775 sum += *tp++; 1776 1777 return (sum); 1778} 1779 1780static int 1781valid_rsdp(struct rsdp *rp) 1782{ 1783 1784 /* validate the V1.x checksum */ 1785 if (checksum_table((uint8_t *)&rp->v1, sizeof (struct rsdp_v1)) != 0) 1786 return (0); 1787 1788 /* If pre-ACPI 2.0, this is a valid RSDP */ 1789 if (rp->v1.revision < 2) 1790 return (1); 1791 1792 /* validate the V2.x checksum */ 1793 if (checksum_table((uint8_t *)rp, sizeof (struct rsdp)) != 0) 1794 return (0); 1795 1796 return (1); 1797} 1798 1799/* 1800 * Scan memory range for an RSDP; 1801 * see ACPI 3.0 Spec, 5.2.5.1 1802 */ 1803static struct rsdp * 1804scan_rsdp(paddr_t start, paddr_t end) 1805{ 1806 size_t len = end - start + 1; 1807 caddr_t ptr; 1808 1809 ptr = vmap_phys(len, start); 1810 while (len > 0) { 1811 if (strncmp(ptr, ACPI_RSDP_SIG, ACPI_RSDP_SIG_LEN) == 0) 1812 if (valid_rsdp((struct rsdp *)ptr)) 1813 return ((struct rsdp *)ptr); 1814 ptr += 16; 1815 len -= 16; 1816 } 1817 1818 return (NULL); 1819} 1820 1821/* 1822 * Refer to ACPI 3.0 Spec, section 5.2.5.1 to understand this function 1823 */ 1824static struct rsdp * 1825find_rsdp() { 1826 struct rsdp *rsdp; 1827 uint16_t *ebda_seg; 1828 paddr_t ebda_addr; 1829 1830 /* 1831 * Get the EBDA segment and scan the first 1K 1832 */ 1833 ebda_seg = (uint16_t *)vmap_phys(sizeof (uint16_t), ACPI_EBDA_SEG_ADDR); 1834 ebda_addr = *ebda_seg << 4; 1835 rsdp = scan_rsdp(ebda_addr, ebda_addr + ACPI_EBDA_LEN - 1); 1836 if (rsdp == NULL) 1837 /* if EBDA doesn't contain RSDP, look in BIOS memory */ 1838 rsdp = scan_rsdp(0xe0000, 0xfffff); 1839 return (rsdp); 1840} 1841 1842static struct table_header * 1843map_fw_table(paddr_t table_addr) 1844{ 1845 struct table_header *tp; 1846 size_t len = MAX(sizeof (struct table_header), MMU_PAGESIZE); 1847 1848 /* 1849 * Map at least a page; if the table is larger than this, remap it 1850 */ 1851 tp = (struct table_header *)vmap_phys(len, table_addr); 1852 if (tp->len > len) 1853 tp = (struct table_header *)vmap_phys(tp->len, table_addr); 1854 return (tp); 1855} 1856 1857static struct table_header * 1858find_fw_table(char *signature) 1859{ 1860 static int revision = 0; 1861 static struct xsdt *xsdt; 1862 static int len; 1863 paddr_t xsdt_addr; 1864 struct rsdp *rsdp; 1865 struct table_header *tp; 1866 paddr_t table_addr; 1867 int n; 1868 1869 if (strlen(signature) != ACPI_TABLE_SIG_LEN) 1870 return (NULL); 1871 1872 /* 1873 * Reading the ACPI 3.0 Spec, section 5.2.5.3 will help 1874 * understand this code. If we haven't already found the RSDT/XSDT, 1875 * revision will be 0. Find the RSDP and check the revision 1876 * to find out whether to use the RSDT or XSDT. If revision is 1877 * 0 or 1, use the RSDT and set internal revision to 1; if it is 2, 1878 * use the XSDT. If the XSDT address is 0, though, fall back to 1879 * revision 1 and use the RSDT. 1880 */ 1881 if (revision == 0) { 1882 if ((rsdp = (struct rsdp *)find_rsdp()) != NULL) { 1883 revision = rsdp->v1.revision; 1884 switch (revision) { 1885 case 2: 1886 /* 1887 * Use the XSDT unless BIOS is buggy and 1888 * claims to be rev 2 but has a null XSDT 1889 * address 1890 */ 1891 xsdt_addr = rsdp->xsdt; 1892 if (xsdt_addr != 0) 1893 break; 1894 /* FALLTHROUGH */ 1895 case 0: 1896 /* treat RSDP rev 0 as revision 1 internally */ 1897 revision = 1; 1898 /* FALLTHROUGH */ 1899 case 1: 1900 /* use the RSDT for rev 0/1 */ 1901 xsdt_addr = rsdp->v1.rsdt; 1902 break; 1903 default: 1904 /* unknown revision */ 1905 revision = 0; 1906 break; 1907 } 1908 } 1909 if (revision == 0) 1910 return (NULL); 1911 1912 /* cache the XSDT info */ 1913 xsdt = (struct xsdt *)map_fw_table(xsdt_addr); 1914 len = (xsdt->hdr.len - sizeof (xsdt->hdr)) / 1915 ((revision == 1) ? sizeof (uint32_t) : sizeof (uint64_t)); 1916 } 1917 1918 /* 1919 * Scan the table headers looking for a signature match 1920 */ 1921 for (n = 0; n < len; n++) { 1922 table_addr = (revision == 1) ? xsdt->p.r[n] : xsdt->p.x[n]; 1923 if (table_addr == 0) 1924 continue; 1925 tp = map_fw_table(table_addr); 1926 if (strncmp(tp->sig, signature, ACPI_TABLE_SIG_LEN) == 0) { 1927 return (tp); 1928 } 1929 } 1930 return (NULL); 1931} 1932 1933static void 1934process_madt(struct madt *tp) 1935{ 1936 struct madt_processor *cpu, *end; 1937 uint32_t cpu_count = 0; 1938 uint8_t cpu_apicid_array[UINT8_MAX + 1]; 1939 1940 if (tp != NULL) { 1941 /* 1942 * Determine number of CPUs and keep track of "final" APIC ID 1943 * for each CPU by walking through ACPI MADT processor list 1944 */ 1945 end = (struct madt_processor *)(tp->hdr.len + (uintptr_t)tp); 1946 cpu = tp->list; 1947 while (cpu < end) { 1948 if (cpu->type == MADT_PROCESSOR) { 1949 if (cpu->flags & 1) { 1950 if (cpu_count < UINT8_MAX) 1951 cpu_apicid_array[cpu_count] = 1952 cpu->apic_id; 1953 cpu_count++; 1954 } 1955 } 1956 1957 cpu = (struct madt_processor *) 1958 (cpu->len + (uintptr_t)cpu); 1959 } 1960 1961 /* 1962 * Make boot property for array of "final" APIC IDs for each 1963 * CPU 1964 */ 1965 bsetprop(BP_CPU_APICID_ARRAY, strlen(BP_CPU_APICID_ARRAY), 1966 cpu_apicid_array, cpu_count * sizeof (uint8_t)); 1967 } 1968 1969 /* 1970 * User-set boot-ncpus overrides firmware count 1971 */ 1972 if (do_bsys_getproplen(NULL, "boot-ncpus") >= 0) 1973 return; 1974 1975 /* 1976 * Set boot property for boot-ncpus to number of CPUs given in MADT 1977 * if user hasn't set the property already 1978 */ 1979 if (tp != NULL) 1980 bsetpropsi("boot-ncpus", cpu_count); 1981} 1982 1983static void 1984process_srat(struct srat *tp) 1985{ 1986 struct srat_item *item, *end; 1987 int i; 1988 int proc_num, mem_num; 1989#pragma pack(1) 1990 struct { 1991 uint32_t domain; 1992 uint32_t apic_id; 1993 uint32_t sapic_id; 1994 } processor; 1995 struct { 1996 uint32_t domain; 1997 uint32_t x2apic_id; 1998 } x2apic; 1999 struct { 2000 uint32_t domain; 2001 uint64_t addr; 2002 uint64_t length; 2003 uint32_t flags; 2004 } memory; 2005#pragma pack() 2006 char prop_name[30]; 2007 2008 if (tp == NULL) 2009 return; 2010 2011 proc_num = mem_num = 0; 2012 end = (struct srat_item *)(tp->hdr.len + (uintptr_t)tp); 2013 item = tp->list; 2014 while (item < end) { 2015 switch (item->type) { 2016 case SRAT_PROCESSOR: 2017 if (!(item->i.p.flags & SRAT_ENABLED)) 2018 break; 2019 processor.domain = item->i.p.domain1; 2020 for (i = 0; i < 3; i++) 2021 processor.domain += 2022 item->i.p.domain2[i] << ((i + 1) * 8); 2023 processor.apic_id = item->i.p.apic_id; 2024 processor.sapic_id = item->i.p.local_sapic_eid; 2025 (void) snprintf(prop_name, 30, "acpi-srat-processor-%d", 2026 proc_num); 2027 bsetprop(prop_name, strlen(prop_name), &processor, 2028 sizeof (processor)); 2029 proc_num++; 2030 break; 2031 case SRAT_MEMORY: 2032 if (!(item->i.m.flags & SRAT_ENABLED)) 2033 break; 2034 memory.domain = item->i.m.domain; 2035 memory.addr = item->i.m.base_addr; 2036 memory.length = item->i.m.len; 2037 memory.flags = item->i.m.flags; 2038 (void) snprintf(prop_name, 30, "acpi-srat-memory-%d", 2039 mem_num); 2040 bsetprop(prop_name, strlen(prop_name), &memory, 2041 sizeof (memory)); 2042 mem_num++; 2043 break; 2044 case SRAT_X2APIC: 2045 if (!(item->i.xp.flags & SRAT_ENABLED)) 2046 break; 2047 x2apic.domain = item->i.xp.domain; 2048 x2apic.x2apic_id = item->i.xp.x2apic_id; 2049 (void) snprintf(prop_name, 30, "acpi-srat-processor-%d", 2050 proc_num); 2051 bsetprop(prop_name, strlen(prop_name), &x2apic, 2052 sizeof (x2apic)); 2053 proc_num++; 2054 break; 2055 } 2056 2057 item = (struct srat_item *) 2058 (item->len + (caddr_t)item); 2059 } 2060} 2061 2062static void 2063process_slit(struct slit *tp) 2064{ 2065 2066 /* 2067 * Check the number of localities; if it's too huge, we just 2068 * return and locality enumeration code will handle this later, 2069 * if possible. 2070 * 2071 * Note that the size of the table is the square of the 2072 * number of localities; if the number of localities exceeds 2073 * UINT16_MAX, the table size may overflow an int when being 2074 * passed to bsetprop() below. 2075 */ 2076 if (tp->number >= SLIT_LOCALITIES_MAX) 2077 return; 2078 2079 bsetprop(SLIT_NUM_PROPNAME, strlen(SLIT_NUM_PROPNAME), &tp->number, 2080 sizeof (tp->number)); 2081 bsetprop(SLIT_PROPNAME, strlen(SLIT_PROPNAME), &tp->entry, 2082 tp->number * tp->number); 2083} 2084 2085static void 2086process_dmar(struct dmar *tp) 2087{ 2088 bsetprop(DMAR_TABLE_PROPNAME, strlen(DMAR_TABLE_PROPNAME), 2089 tp, tp->hdr.len); 2090} 2091 2092#else /* __xpv */ 2093static void 2094enumerate_xen_cpus() 2095{ 2096 processorid_t id, max_id; 2097 2098 /* 2099 * User-set boot-ncpus overrides enumeration 2100 */ 2101 if (do_bsys_getproplen(NULL, "boot-ncpus") >= 0) 2102 return; 2103 2104 /* 2105 * Probe every possible virtual CPU id and remember the 2106 * highest id present; the count of CPUs is one greater 2107 * than this. This tacitly assumes at least cpu 0 is present. 2108 */ 2109 max_id = 0; 2110 for (id = 0; id < MAX_VIRT_CPUS; id++) 2111 if (HYPERVISOR_vcpu_op(VCPUOP_is_up, id, NULL) == 0) 2112 max_id = id; 2113 2114 bsetpropsi("boot-ncpus", max_id+1); 2115 2116} 2117#endif /* __xpv */ 2118 2119static void 2120build_firmware_properties(void) 2121{ 2122#ifndef __xpv 2123 struct table_header *tp; 2124 2125 if ((tp = find_fw_table("APIC")) != NULL) 2126 process_madt((struct madt *)tp); 2127 2128 if ((srat_ptr = (struct srat *)find_fw_table("SRAT")) != NULL) 2129 process_srat(srat_ptr); 2130 2131 if (slit_ptr = (struct slit *)find_fw_table("SLIT")) 2132 process_slit(slit_ptr); 2133 2134 if (tp = find_fw_table("DMAR")) 2135 process_dmar((struct dmar *)tp); 2136#else /* __xpv */ 2137 enumerate_xen_cpus(); 2138#endif /* __xpv */ 2139} 2140 2141/* 2142 * fake up a boot property for USB serial console early boot output 2143 */ 2144void * 2145usbser_init(size_t size) 2146{ 2147 static char *p = NULL; 2148 2149 p = do_bsys_alloc(NULL, NULL, size, MMU_PAGESIZE); 2150 *p = 0; 2151 bsetprop("usb-serial-buf", strlen("usb-serial-buf") + 1, 2152 &p, sizeof (p)); 2153 return (p); 2154} 2155 2156/*ARGSUSED*/ 2157int 2158boot_compinfo(int fd, struct compinfo *cbp) 2159{ 2160 cbp->iscmp = 0; 2161 cbp->blksize = MAXBSIZE; 2162 return (0); 2163} 2164