fakebop.c revision 6336:4eaf084434c9
1/* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 22/* 23 * Copyright 2008 Sun Microsystems, Inc. All rights reserved. 24 * Use is subject to license terms. 25 */ 26 27#pragma ident "%Z%%M% %I% %E% SMI" 28 29/* 30 * This file contains the functionality that mimics the boot operations 31 * on SPARC systems or the old boot.bin/multiboot programs on x86 systems. 32 * The x86 kernel now does everything on its own. 33 */ 34 35#include <sys/types.h> 36#include <sys/bootconf.h> 37#include <sys/bootsvcs.h> 38#include <sys/bootinfo.h> 39#include <sys/multiboot.h> 40#include <sys/bootvfs.h> 41#include <sys/bootprops.h> 42#include <sys/varargs.h> 43#include <sys/param.h> 44#include <sys/machparam.h> 45#include <sys/archsystm.h> 46#include <sys/boot_console.h> 47#include <sys/cmn_err.h> 48#include <sys/systm.h> 49#include <sys/promif.h> 50#include <sys/archsystm.h> 51#include <sys/x86_archext.h> 52#include <sys/kobj.h> 53#include <sys/privregs.h> 54#include <sys/sysmacros.h> 55#include <sys/ctype.h> 56#ifdef __xpv 57#include <sys/hypervisor.h> 58#include <net/if.h> 59#endif 60#include <vm/kboot_mmu.h> 61#include <vm/hat_pte.h> 62#include "acpi_fw.h" 63 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/* 125 * Allocate aligned physical memory at boot time. This allocator allocates 126 * from the highest possible addresses. This avoids exhausting memory that 127 * would be useful for DMA buffers. 128 */ 129paddr_t 130do_bop_phys_alloc(uint64_t size, uint64_t align) 131{ 132 paddr_t pa = 0; 133 paddr_t start; 134 paddr_t end; 135 struct memlist *ml = (struct memlist *)xbootp->bi_phys_install; 136 137 /* 138 * Be careful if high memory usage is limited in startup.c 139 * Since there are holes in the low part of the physical address 140 * space we can treat physmem as a pfn (not just a pgcnt) and 141 * get a conservative upper limit. 142 */ 143 if (physmem != 0 && high_phys > pfn_to_pa(physmem)) 144 high_phys = pfn_to_pa(physmem); 145 146 /* 147 * find the lowest or highest available memory in physinstalled 148 * On 32 bit avoid physmem above 4Gig if PAE isn't enabled 149 */ 150#if defined(__i386) 151 if (xbootp->bi_use_pae == 0 && high_phys > FOUR_GIG) 152 high_phys = FOUR_GIG; 153#endif 154 155 /* 156 * find the highest available memory in physinstalled 157 */ 158 size = P2ROUNDUP(size, align); 159 for (; ml; ml = ml->next) { 160 start = P2ROUNDUP(ml->address, align); 161 end = P2ALIGN(ml->address + ml->size, align); 162 if (start < next_phys) 163 start = P2ROUNDUP(next_phys, align); 164 if (end > high_phys) 165 end = P2ALIGN(high_phys, align); 166 167 if (end <= start) 168 continue; 169 if (end - start < size) 170 continue; 171 172 /* 173 * Early allocations need to use low memory, since 174 * physmem might be further limited by bootenv.rc 175 */ 176 if (early_allocation) { 177 if (pa == 0 || start < pa) 178 pa = start; 179 } else { 180 if (end - size > pa) 181 pa = end - size; 182 } 183 } 184 if (pa != 0) { 185 if (early_allocation) 186 next_phys = pa + size; 187 else 188 high_phys = pa; 189 return (pa); 190 } 191 bop_panic("do_bop_phys_alloc(0x%" PRIx64 ", 0x%" PRIx64 192 ") Out of memory\n", size, align); 193 /*NOTREACHED*/ 194} 195 196static uintptr_t 197alloc_vaddr(size_t size, paddr_t align) 198{ 199 uintptr_t rv; 200 201 next_virt = P2ROUNDUP(next_virt, (uintptr_t)align); 202 rv = (uintptr_t)next_virt; 203 next_virt += size; 204 return (rv); 205} 206 207/* 208 * Allocate virtual memory. The size is always rounded up to a multiple 209 * of base pagesize. 210 */ 211 212/*ARGSUSED*/ 213static caddr_t 214do_bsys_alloc(bootops_t *bop, caddr_t virthint, size_t size, int align) 215{ 216 paddr_t a = align; /* same type as pa for masking */ 217 uint_t pgsize; 218 paddr_t pa; 219 uintptr_t va; 220 ssize_t s; /* the aligned size */ 221 uint_t level; 222 uint_t is_kernel = (virthint != 0); 223 224 if (a < MMU_PAGESIZE) 225 a = MMU_PAGESIZE; 226 else if (!ISP2(a)) 227 prom_panic("do_bsys_alloc() incorrect alignment"); 228 size = P2ROUNDUP(size, MMU_PAGESIZE); 229 230 /* 231 * Use the next aligned virtual address if we weren't given one. 232 */ 233 if (virthint == NULL) { 234 virthint = (caddr_t)alloc_vaddr(size, a); 235 total_bop_alloc_scratch += size; 236 } else { 237 total_bop_alloc_kernel += size; 238 } 239 240 /* 241 * allocate the physical memory 242 */ 243 pa = do_bop_phys_alloc(size, a); 244 245 /* 246 * Add the mappings to the page tables, try large pages first. 247 */ 248 va = (uintptr_t)virthint; 249 s = size; 250 level = 1; 251 pgsize = xbootp->bi_use_pae ? TWO_MEG : FOUR_MEG; 252 if (xbootp->bi_use_largepage && a == pgsize) { 253 while (IS_P2ALIGNED(pa, pgsize) && IS_P2ALIGNED(va, pgsize) && 254 s >= pgsize) { 255 kbm_map(va, pa, level, is_kernel); 256 va += pgsize; 257 pa += pgsize; 258 s -= pgsize; 259 } 260 } 261 262 /* 263 * Map remaining pages use small mappings 264 */ 265 level = 0; 266 pgsize = MMU_PAGESIZE; 267 while (s > 0) { 268 kbm_map(va, pa, level, is_kernel); 269 va += pgsize; 270 pa += pgsize; 271 s -= pgsize; 272 } 273 return (virthint); 274} 275 276/* 277 * Free virtual memory - we'll just ignore these. 278 */ 279/*ARGSUSED*/ 280static void 281do_bsys_free(bootops_t *bop, caddr_t virt, size_t size) 282{ 283 bop_printf(NULL, "do_bsys_free(virt=0x%p, size=0x%lx) ignored\n", 284 (void *)virt, size); 285} 286 287/* 288 * Old interface 289 */ 290/*ARGSUSED*/ 291static caddr_t 292do_bsys_ealloc( 293 bootops_t *bop, 294 caddr_t virthint, 295 size_t size, 296 int align, 297 int flags) 298{ 299 prom_panic("unsupported call to BOP_EALLOC()\n"); 300 return (0); 301} 302 303 304static void 305bsetprop(char *name, int nlen, void *value, int vlen) 306{ 307 uint_t size; 308 uint_t need_size; 309 bootprop_t *b; 310 311 /* 312 * align the size to 16 byte boundary 313 */ 314 size = sizeof (bootprop_t) + nlen + 1 + vlen; 315 size = (size + 0xf) & ~0xf; 316 if (size > curr_space) { 317 need_size = (size + (MMU_PAGEOFFSET)) & MMU_PAGEMASK; 318 curr_page = do_bsys_alloc(NULL, 0, need_size, MMU_PAGESIZE); 319 curr_space = need_size; 320 } 321 322 /* 323 * use a bootprop_t at curr_page and link into list 324 */ 325 b = (bootprop_t *)curr_page; 326 curr_page += sizeof (bootprop_t); 327 curr_space -= sizeof (bootprop_t); 328 b->bp_next = bprops; 329 bprops = b; 330 331 /* 332 * follow by name and ending zero byte 333 */ 334 b->bp_name = curr_page; 335 bcopy(name, curr_page, nlen); 336 curr_page += nlen; 337 *curr_page++ = 0; 338 curr_space -= nlen + 1; 339 340 /* 341 * copy in value, but no ending zero byte 342 */ 343 b->bp_value = curr_page; 344 b->bp_vlen = vlen; 345 if (vlen > 0) { 346 bcopy(value, curr_page, vlen); 347 curr_page += vlen; 348 curr_space -= vlen; 349 } 350 351 /* 352 * align new values of curr_page, curr_space 353 */ 354 while (curr_space & 0xf) { 355 ++curr_page; 356 --curr_space; 357 } 358} 359 360static void 361bsetprops(char *name, char *value) 362{ 363 bsetprop(name, strlen(name), value, strlen(value) + 1); 364} 365 366static void 367bsetprop64(char *name, uint64_t value) 368{ 369 bsetprop(name, strlen(name), (void *)&value, sizeof (value)); 370} 371 372static void 373bsetpropsi(char *name, int value) 374{ 375 char prop_val[32]; 376 377 (void) snprintf(prop_val, sizeof (prop_val), "%d", value); 378 bsetprops(name, prop_val); 379} 380 381/* 382 * to find the size of the buffer to allocate 383 */ 384/*ARGSUSED*/ 385int 386do_bsys_getproplen(bootops_t *bop, const char *name) 387{ 388 bootprop_t *b; 389 390 for (b = bprops; b; b = b->bp_next) { 391 if (strcmp(name, b->bp_name) != 0) 392 continue; 393 return (b->bp_vlen); 394 } 395 return (-1); 396} 397 398/* 399 * get the value associated with this name 400 */ 401/*ARGSUSED*/ 402int 403do_bsys_getprop(bootops_t *bop, const char *name, void *value) 404{ 405 bootprop_t *b; 406 407 for (b = bprops; b; b = b->bp_next) { 408 if (strcmp(name, b->bp_name) != 0) 409 continue; 410 bcopy(b->bp_value, value, b->bp_vlen); 411 return (0); 412 } 413 return (-1); 414} 415 416/* 417 * get the name of the next property in succession from the standalone 418 */ 419/*ARGSUSED*/ 420static char * 421do_bsys_nextprop(bootops_t *bop, char *name) 422{ 423 bootprop_t *b; 424 425 /* 426 * A null name is a special signal for the 1st boot property 427 */ 428 if (name == NULL || strlen(name) == 0) { 429 if (bprops == NULL) 430 return (NULL); 431 return (bprops->bp_name); 432 } 433 434 for (b = bprops; b; b = b->bp_next) { 435 if (name != b->bp_name) 436 continue; 437 b = b->bp_next; 438 if (b == NULL) 439 return (NULL); 440 return (b->bp_name); 441 } 442 return (NULL); 443} 444 445/* 446 * Parse numeric value from a string. Understands decimal, hex, octal, - and ~ 447 */ 448static int 449parse_value(char *p, uint64_t *retval) 450{ 451 int adjust = 0; 452 uint64_t tmp = 0; 453 int digit; 454 int radix = 10; 455 456 *retval = 0; 457 if (*p == '-' || *p == '~') 458 adjust = *p++; 459 460 if (*p == '0') { 461 ++p; 462 if (*p == 0) 463 return (0); 464 if (*p == 'x' || *p == 'X') { 465 radix = 16; 466 ++p; 467 } else { 468 radix = 8; 469 ++p; 470 } 471 } 472 while (*p) { 473 if ('0' <= *p && *p <= '9') 474 digit = *p - '0'; 475 else if ('a' <= *p && *p <= 'f') 476 digit = 10 + *p - 'a'; 477 else if ('A' <= *p && *p <= 'F') 478 digit = 10 + *p - 'A'; 479 else 480 return (-1); 481 if (digit >= radix) 482 return (-1); 483 tmp = tmp * radix + digit; 484 ++p; 485 } 486 if (adjust == '-') 487 tmp = -tmp; 488 else if (adjust == '~') 489 tmp = ~tmp; 490 *retval = tmp; 491 return (0); 492} 493 494/* 495 * 2nd part of building the table of boot properties. This includes: 496 * - values from /boot/solaris/bootenv.rc (ie. eeprom(1m) values) 497 * 498 * lines look like one of: 499 * ^$ 500 * ^# comment till end of line 501 * setprop name 'value' 502 * setprop name value 503 * setprop name "value" 504 * 505 * we do single character I/O since this is really just looking at memory 506 */ 507void 508boot_prop_finish(void) 509{ 510 int fd; 511 char *line; 512 int c; 513 int bytes_read; 514 char *name; 515 int n_len; 516 char *value; 517 int v_len; 518 char *inputdev; /* these override the command line if serial ports */ 519 char *outputdev; 520 char *consoledev; 521 uint64_t lvalue; 522 int use_xencons = 0; 523 524#ifdef __xpv 525 if (!DOMAIN_IS_INITDOMAIN(xen_info)) 526 use_xencons = 1; 527#endif /* __xpv */ 528 529 DBG_MSG("Opening /boot/solaris/bootenv.rc\n"); 530 fd = BRD_OPEN(bfs_ops, "/boot/solaris/bootenv.rc", 0); 531 DBG(fd); 532 533 line = do_bsys_alloc(NULL, NULL, MMU_PAGESIZE, MMU_PAGESIZE); 534 while (fd >= 0) { 535 536 /* 537 * get a line 538 */ 539 for (c = 0; ; ++c) { 540 bytes_read = BRD_READ(bfs_ops, fd, line + c, 1); 541 if (bytes_read == 0) { 542 if (c == 0) 543 goto done; 544 break; 545 } 546 if (line[c] == '\n') 547 break; 548 } 549 line[c] = 0; 550 551 /* 552 * ignore comment lines 553 */ 554 c = 0; 555 while (ISSPACE(line[c])) 556 ++c; 557 if (line[c] == '#' || line[c] == 0) 558 continue; 559 560 /* 561 * must have "setprop " or "setprop\t" 562 */ 563 if (strncmp(line + c, "setprop ", 8) != 0 && 564 strncmp(line + c, "setprop\t", 8) != 0) 565 continue; 566 c += 8; 567 while (ISSPACE(line[c])) 568 ++c; 569 if (line[c] == 0) 570 continue; 571 572 /* 573 * gather up the property name 574 */ 575 name = line + c; 576 n_len = 0; 577 while (line[c] && !ISSPACE(line[c])) 578 ++n_len, ++c; 579 580 /* 581 * gather up the value, if any 582 */ 583 value = ""; 584 v_len = 0; 585 while (ISSPACE(line[c])) 586 ++c; 587 if (line[c] != 0) { 588 value = line + c; 589 while (line[c] && !ISSPACE(line[c])) 590 ++v_len, ++c; 591 } 592 593 if (v_len >= 2 && value[0] == value[v_len - 1] && 594 (value[0] == '\'' || value[0] == '"')) { 595 ++value; 596 v_len -= 2; 597 } 598 name[n_len] = 0; 599 if (v_len > 0) 600 value[v_len] = 0; 601 else 602 continue; 603 604 /* 605 * ignore "boot-file" property, it's now meaningless 606 */ 607 if (strcmp(name, "boot-file") == 0) 608 continue; 609 if (strcmp(name, "boot-args") == 0 && 610 strlen(boot_args) > 0) 611 continue; 612 613 /* 614 * If a property was explicitly set on the command line 615 * it will override a setting in bootenv.rc 616 */ 617 if (do_bsys_getproplen(NULL, name) > 0) 618 continue; 619 620 bsetprop(name, n_len, value, v_len + 1); 621 } 622done: 623 if (fd >= 0) 624 BRD_CLOSE(bfs_ops, fd); 625 626 /* 627 * Check if we have to limit the boot time allocator 628 */ 629 if (do_bsys_getproplen(NULL, "physmem") != -1 && 630 do_bsys_getprop(NULL, "physmem", line) >= 0 && 631 parse_value(line, &lvalue) != -1) { 632 if (0 < lvalue && (lvalue < physmem || physmem == 0)) { 633 physmem = (pgcnt_t)lvalue; 634 DBG(physmem); 635 } 636 } 637 early_allocation = 0; 638 639 /* 640 * check to see if we have to override the default value of the console 641 */ 642 if (!use_xencons) { 643 inputdev = line; 644 v_len = do_bsys_getproplen(NULL, "input-device"); 645 if (v_len > 0) 646 (void) do_bsys_getprop(NULL, "input-device", inputdev); 647 else 648 v_len = 0; 649 inputdev[v_len] = 0; 650 651 outputdev = inputdev + v_len + 1; 652 v_len = do_bsys_getproplen(NULL, "output-device"); 653 if (v_len > 0) 654 (void) do_bsys_getprop(NULL, "output-device", 655 outputdev); 656 else 657 v_len = 0; 658 outputdev[v_len] = 0; 659 660 consoledev = outputdev + v_len + 1; 661 v_len = do_bsys_getproplen(NULL, "console"); 662 if (v_len > 0) 663 (void) do_bsys_getprop(NULL, "console", consoledev); 664 else 665 v_len = 0; 666 consoledev[v_len] = 0; 667 bcons_init2(inputdev, outputdev, consoledev); 668 } else { 669 /* 670 * Ensure console property exists 671 * If not create it as "hypervisor" 672 */ 673 v_len = do_bsys_getproplen(NULL, "console"); 674 if (v_len < 0) 675 bsetprops("console", "hypervisor"); 676 inputdev = outputdev = consoledev = "hypervisor"; 677 bcons_init2(inputdev, outputdev, consoledev); 678 } 679 680 if (strstr((char *)xbootp->bi_cmdline, "prom_debug") || kbm_debug) { 681 value = line; 682 bop_printf(NULL, "\nBoot properties:\n"); 683 name = ""; 684 while ((name = do_bsys_nextprop(NULL, name)) != NULL) { 685 bop_printf(NULL, "\t0x%p %s = ", (void *)name, name); 686 (void) do_bsys_getprop(NULL, name, value); 687 v_len = do_bsys_getproplen(NULL, name); 688 bop_printf(NULL, "len=%d ", v_len); 689 value[v_len] = 0; 690 bop_printf(NULL, "%s\n", value); 691 } 692 } 693} 694 695/* 696 * print formatted output 697 */ 698/*PRINTFLIKE2*/ 699/*ARGSUSED*/ 700void 701bop_printf(bootops_t *bop, const char *fmt, ...) 702{ 703 va_list ap; 704 705 if (have_console == 0) 706 return; 707 708 va_start(ap, fmt); 709 (void) vsnprintf(buffer, BUFFERSIZE, fmt, ap); 710 va_end(ap); 711 PUT_STRING(buffer); 712} 713 714/* 715 * Another panic() variant; this one can be used even earlier during boot than 716 * prom_panic(). 717 */ 718/*PRINTFLIKE1*/ 719void 720bop_panic(const char *fmt, ...) 721{ 722 va_list ap; 723 724 va_start(ap, fmt); 725 bop_printf(NULL, fmt, ap); 726 va_end(ap); 727 728 bop_printf(NULL, "\nPress any key to reboot.\n"); 729 (void) bcons_getchar(); 730 bop_printf(NULL, "Resetting...\n"); 731 pc_reset(); 732} 733 734/* 735 * Do a real mode interrupt BIOS call 736 */ 737typedef struct bios_regs { 738 unsigned short ax, bx, cx, dx, si, di, bp, es, ds; 739} bios_regs_t; 740typedef int (*bios_func_t)(int, bios_regs_t *); 741 742/*ARGSUSED*/ 743static void 744do_bsys_doint(bootops_t *bop, int intnum, struct bop_regs *rp) 745{ 746#if defined(__xpv) 747 prom_panic("unsupported call to BOP_DOINT()\n"); 748#else /* __xpv */ 749 static int firsttime = 1; 750 bios_func_t bios_func = (bios_func_t)(void *)(uintptr_t)0x5000; 751 bios_regs_t br; 752 753 /* 754 * The first time we do this, we have to copy the pre-packaged 755 * low memory bios call code image into place. 756 */ 757 if (firsttime) { 758 extern char bios_image[]; 759 extern uint32_t bios_size; 760 761 bcopy(bios_image, (void *)bios_func, bios_size); 762 firsttime = 0; 763 } 764 765 br.ax = rp->eax.word.ax; 766 br.bx = rp->ebx.word.bx; 767 br.cx = rp->ecx.word.cx; 768 br.dx = rp->edx.word.dx; 769 br.bp = rp->ebp.word.bp; 770 br.si = rp->esi.word.si; 771 br.di = rp->edi.word.di; 772 br.ds = rp->ds; 773 br.es = rp->es; 774 775 DBG_MSG("Doing BIOS call..."); 776 rp->eflags = bios_func(intnum, &br); 777 DBG_MSG("done\n"); 778 779 rp->eax.word.ax = br.ax; 780 rp->ebx.word.bx = br.bx; 781 rp->ecx.word.cx = br.cx; 782 rp->edx.word.dx = br.dx; 783 rp->ebp.word.bp = br.bp; 784 rp->esi.word.si = br.si; 785 rp->edi.word.di = br.di; 786 rp->ds = br.ds; 787 rp->es = br.es; 788#endif /* __xpv */ 789} 790 791static struct boot_syscalls bop_sysp = { 792 bcons_getchar, 793 bcons_putchar, 794 bcons_ischar, 795}; 796 797static char *whoami; 798 799#define BUFLEN 64 800 801#if defined(__xpv) 802 803static char namebuf[32]; 804 805static void 806xen_parse_props(char *s, char *prop_map[], int n_prop) 807{ 808 char **prop_name = prop_map; 809 char *cp = s, *scp; 810 811 do { 812 scp = cp; 813 while ((*cp != NULL) && (*cp != ':')) 814 cp++; 815 816 if ((scp != cp) && (*prop_name != NULL)) { 817 *cp = NULL; 818 bsetprops(*prop_name, scp); 819 } 820 821 cp++; 822 prop_name++; 823 n_prop--; 824 } while (n_prop > 0); 825} 826 827#define VBDPATHLEN 64 828 829/* 830 * parse the 'xpv-root' property to create properties used by 831 * ufs_mountroot. 832 */ 833static void 834xen_vbdroot_props(char *s) 835{ 836 char vbdpath[VBDPATHLEN] = "/xpvd/xdf@"; 837 const char lnamefix[] = "/dev/dsk/c0d"; 838 char *pnp; 839 char *prop_p; 840 char mi; 841 short minor; 842 long addr = 0; 843 844 pnp = vbdpath + strlen(vbdpath); 845 prop_p = s + strlen(lnamefix); 846 while ((*prop_p != '\0') && (*prop_p != 's') && (*prop_p != 'p')) 847 addr = addr * 10 + *prop_p++ - '0'; 848 (void) snprintf(pnp, VBDPATHLEN, "%lx", addr); 849 pnp = vbdpath + strlen(vbdpath); 850 if (*prop_p == 's') 851 mi = 'a'; 852 else if (*prop_p == 'p') 853 mi = 'q'; 854 else 855 ASSERT(0); /* shouldn't be here */ 856 prop_p++; 857 ASSERT(*prop_p != '\0'); 858 if (ISDIGIT(*prop_p)) { 859 minor = *prop_p - '0'; 860 prop_p++; 861 if (ISDIGIT(*prop_p)) { 862 minor = minor * 10 + *prop_p - '0'; 863 } 864 } else { 865 /* malformed root path, use 0 as default */ 866 minor = 0; 867 } 868 ASSERT(minor < 16); /* at most 16 partitions */ 869 mi += minor; 870 *pnp++ = ':'; 871 *pnp++ = mi; 872 *pnp++ = '\0'; 873 bsetprops("fstype", "ufs"); 874 bsetprops("bootpath", vbdpath); 875 876 DBG_MSG("VBD bootpath set to "); 877 DBG_MSG(vbdpath); 878 DBG_MSG("\n"); 879} 880 881/* 882 * parse the xpv-nfsroot property to create properties used by 883 * nfs_mountroot. 884 */ 885static void 886xen_nfsroot_props(char *s) 887{ 888 char *prop_map[] = { 889 BP_SERVER_IP, /* server IP address */ 890 BP_SERVER_NAME, /* server hostname */ 891 BP_SERVER_PATH, /* root path */ 892 }; 893 int n_prop = sizeof (prop_map) / sizeof (prop_map[0]); 894 895 bsetprop("fstype", 6, "nfsdyn", 7); 896 897 xen_parse_props(s, prop_map, n_prop); 898 899 /* 900 * If a server name wasn't specified, use a default. 901 */ 902 if (do_bsys_getproplen(NULL, BP_SERVER_NAME) == -1) 903 bsetprops(BP_SERVER_NAME, "unknown"); 904} 905 906/* 907 * Extract our IP address, etc. from the "xpv-ip" property. 908 */ 909static void 910xen_ip_props(char *s) 911{ 912 char *prop_map[] = { 913 BP_HOST_IP, /* IP address */ 914 NULL, /* NFS server IP address (ignored in */ 915 /* favour of xpv-nfsroot) */ 916 BP_ROUTER_IP, /* IP gateway */ 917 BP_SUBNET_MASK, /* IP subnet mask */ 918 "xpv-hostname", /* hostname (ignored) */ 919 BP_NETWORK_INTERFACE, /* interface name */ 920 "xpv-hcp", /* host configuration protocol */ 921 }; 922 int n_prop = sizeof (prop_map) / sizeof (prop_map[0]); 923 char ifname[IFNAMSIZ]; 924 925 xen_parse_props(s, prop_map, n_prop); 926 927 /* 928 * A Linux dom0 administrator expects all interfaces to be 929 * called "ethX", which is not the case here. 930 * 931 * If the interface name specified is "eth0", presume that 932 * this is really intended to be "xnf0" (the first domU -> 933 * dom0 interface for this domain). 934 */ 935 if ((do_bsys_getprop(NULL, BP_NETWORK_INTERFACE, ifname) == 0) && 936 (strcmp("eth0", ifname) == 0)) { 937 bsetprops(BP_NETWORK_INTERFACE, "xnf0"); 938 bop_printf(NULL, 939 "network interface name 'eth0' replaced with 'xnf0'\n"); 940 } 941} 942 943#else /* __xpv */ 944 945static void 946setup_rarp_props(struct sol_netinfo *sip) 947{ 948 char buf[BUFLEN]; /* to hold ip/mac addrs */ 949 uint8_t *val; 950 951 val = (uint8_t *)&sip->sn_ciaddr; 952 (void) snprintf(buf, BUFLEN, "%d.%d.%d.%d", 953 val[0], val[1], val[2], val[3]); 954 bsetprops(BP_HOST_IP, buf); 955 956 val = (uint8_t *)&sip->sn_siaddr; 957 (void) snprintf(buf, BUFLEN, "%d.%d.%d.%d", 958 val[0], val[1], val[2], val[3]); 959 bsetprops(BP_SERVER_IP, buf); 960 961 if (sip->sn_giaddr != 0) { 962 val = (uint8_t *)&sip->sn_giaddr; 963 (void) snprintf(buf, BUFLEN, "%d.%d.%d.%d", 964 val[0], val[1], val[2], val[3]); 965 bsetprops(BP_ROUTER_IP, buf); 966 } 967 968 if (sip->sn_netmask != 0) { 969 val = (uint8_t *)&sip->sn_netmask; 970 (void) snprintf(buf, BUFLEN, "%d.%d.%d.%d", 971 val[0], val[1], val[2], val[3]); 972 bsetprops(BP_SUBNET_MASK, buf); 973 } 974 975 if (sip->sn_mactype != 4 || sip->sn_maclen != 6) { 976 bop_printf(NULL, "unsupported mac type %d, mac len %d\n", 977 sip->sn_mactype, sip->sn_maclen); 978 } else { 979 val = sip->sn_macaddr; 980 (void) snprintf(buf, BUFLEN, "%x:%x:%x:%x:%x:%x", 981 val[0], val[1], val[2], val[3], val[4], val[5]); 982 bsetprops(BP_BOOT_MAC, buf); 983 } 984} 985 986#endif /* __xpv */ 987 988/* 989 * 1st pass at building the table of boot properties. This includes: 990 * - values set on the command line: -B a=x,b=y,c=z .... 991 * - known values we just compute (ie. from xbootp) 992 * - values from /boot/solaris/bootenv.rc (ie. eeprom(1m) values) 993 * 994 * the grub command line looked like: 995 * kernel boot-file [-B prop=value[,prop=value]...] [boot-args] 996 * 997 * whoami is the same as boot-file 998 */ 999static void 1000build_boot_properties(void) 1001{ 1002 char *name; 1003 int name_len; 1004 char *value; 1005 int value_len; 1006 struct boot_modules *bm; 1007 char *propbuf; 1008 int quoted = 0; 1009 int boot_arg_len; 1010#ifndef __xpv 1011 static int stdout_val = 0; 1012 uchar_t boot_device; 1013 char str[3]; 1014 multiboot_info_t *mbi; 1015 int netboot; 1016 struct sol_netinfo *sip; 1017#endif 1018 1019 /* 1020 * These have to be done first, so that kobj_mount_root() works 1021 */ 1022 DBG_MSG("Building boot properties\n"); 1023 propbuf = do_bsys_alloc(NULL, NULL, MMU_PAGESIZE, 0); 1024 DBG((uintptr_t)propbuf); 1025 if (xbootp->bi_module_cnt > 0) { 1026 bm = xbootp->bi_modules; 1027 bsetprop64("ramdisk_start", (uint64_t)(uintptr_t)bm->bm_addr); 1028 bsetprop64("ramdisk_end", (uint64_t)(uintptr_t)bm->bm_addr + 1029 bm->bm_size); 1030 } 1031 1032 DBG_MSG("Parsing command line for boot properties\n"); 1033 value = xbootp->bi_cmdline; 1034 1035 /* 1036 * allocate memory to collect boot_args into 1037 */ 1038 boot_arg_len = strlen(xbootp->bi_cmdline) + 1; 1039 boot_args = do_bsys_alloc(NULL, NULL, boot_arg_len, MMU_PAGESIZE); 1040 boot_args[0] = 0; 1041 boot_arg_len = 0; 1042 1043#ifdef __xpv 1044 /* 1045 * Xen puts a lot of device information in front of the kernel name 1046 * let's grab them and make them boot properties. The first 1047 * string w/o an "=" in it will be the boot-file property. 1048 */ 1049 (void) strcpy(namebuf, "xpv-"); 1050 for (;;) { 1051 /* 1052 * get to next property 1053 */ 1054 while (ISSPACE(*value)) 1055 ++value; 1056 name = value; 1057 /* 1058 * look for an "=" 1059 */ 1060 while (*value && !ISSPACE(*value) && *value != '=') { 1061 value++; 1062 } 1063 if (*value != '=') { /* no "=" in the property */ 1064 value = name; 1065 break; 1066 } 1067 name_len = value - name; 1068 value_len = 0; 1069 /* 1070 * skip over the "=" 1071 */ 1072 value++; 1073 while (value[value_len] && !ISSPACE(value[value_len])) { 1074 ++value_len; 1075 } 1076 /* 1077 * build property name with "xpv-" prefix 1078 */ 1079 if (name_len + 4 > 32) { /* skip if name too long */ 1080 value += value_len; 1081 continue; 1082 } 1083 bcopy(name, &namebuf[4], name_len); 1084 name_len += 4; 1085 namebuf[name_len] = 0; 1086 bcopy(value, propbuf, value_len); 1087 propbuf[value_len] = 0; 1088 bsetprops(namebuf, propbuf); 1089 1090 /* 1091 * xpv-root is set to the logical disk name of the xen 1092 * VBD when booting from a disk-based filesystem. 1093 */ 1094 if (strcmp(namebuf, "xpv-root") == 0) 1095 xen_vbdroot_props(propbuf); 1096 /* 1097 * While we're here, if we have a "xpv-nfsroot" property 1098 * then we need to set "fstype" to "nfsdyn" so we mount 1099 * our root from the nfs server. Also parse the xpv-nfsroot 1100 * property to create the properties that nfs_mountroot will 1101 * need to find the root and mount it. 1102 */ 1103 if (strcmp(namebuf, "xpv-nfsroot") == 0) 1104 xen_nfsroot_props(propbuf); 1105 1106 if (strcmp(namebuf, "xpv-ip") == 0) 1107 xen_ip_props(propbuf); 1108 value += value_len; 1109 } 1110#endif 1111 1112 while (ISSPACE(*value)) 1113 ++value; 1114 /* 1115 * value now points at the boot-file 1116 */ 1117 value_len = 0; 1118 while (value[value_len] && !ISSPACE(value[value_len])) 1119 ++value_len; 1120 if (value_len > 0) { 1121 whoami = propbuf; 1122 bcopy(value, whoami, value_len); 1123 whoami[value_len] = 0; 1124 bsetprops("boot-file", whoami); 1125 /* 1126 * strip leading path stuff from whoami, so running from 1127 * PXE/miniroot makes sense. 1128 */ 1129 if (strstr(whoami, "/platform/") != NULL) 1130 whoami = strstr(whoami, "/platform/"); 1131 bsetprops("whoami", whoami); 1132 } 1133 1134 /* 1135 * Values forcibly set boot properties on the command line via -B. 1136 * Allow use of quotes in values. Other stuff goes on kernel 1137 * command line. 1138 */ 1139 name = value + value_len; 1140 while (*name != 0) { 1141 /* 1142 * anything not " -B" is copied to the command line 1143 */ 1144 if (!ISSPACE(name[0]) || name[1] != '-' || name[2] != 'B') { 1145 boot_args[boot_arg_len++] = *name; 1146 boot_args[boot_arg_len] = 0; 1147 ++name; 1148 continue; 1149 } 1150 1151 /* 1152 * skip the " -B" and following white space 1153 */ 1154 name += 3; 1155 while (ISSPACE(*name)) 1156 ++name; 1157 while (*name && !ISSPACE(*name)) { 1158 value = strstr(name, "="); 1159 if (value == NULL) 1160 break; 1161 name_len = value - name; 1162 ++value; 1163 value_len = 0; 1164 quoted = 0; 1165 for (; ; ++value_len) { 1166 if (!value[value_len]) 1167 break; 1168 1169 /* 1170 * is this value quoted? 1171 */ 1172 if (value_len == 0 && 1173 (value[0] == '\'' || value[0] == '"')) { 1174 quoted = value[0]; 1175 ++value_len; 1176 } 1177 1178 /* 1179 * In the quote accept any character, 1180 * but look for ending quote. 1181 */ 1182 if (quoted) { 1183 if (value[value_len] == quoted) 1184 quoted = 0; 1185 continue; 1186 } 1187 1188 /* 1189 * a comma or white space ends the value 1190 */ 1191 if (value[value_len] == ',' || 1192 ISSPACE(value[value_len])) 1193 break; 1194 } 1195 1196 if (value_len == 0) { 1197 bsetprop(name, name_len, "true", 5); 1198 } else { 1199 char *v = value; 1200 int l = value_len; 1201 if (v[0] == v[l - 1] && 1202 (v[0] == '\'' || v[0] == '"')) { 1203 ++v; 1204 l -= 2; 1205 } 1206 bcopy(v, propbuf, l); 1207 propbuf[l] = '\0'; 1208 bsetprop(name, name_len, propbuf, 1209 l + 1); 1210 } 1211 name = value + value_len; 1212 while (*name == ',') 1213 ++name; 1214 } 1215 } 1216 1217 /* 1218 * set boot-args property 1219 * 1275 name is bootargs, so set 1220 * that too 1221 */ 1222 bsetprops("boot-args", boot_args); 1223 bsetprops("bootargs", boot_args); 1224 1225#ifndef __xpv 1226 /* 1227 * set the BIOS boot device from GRUB 1228 */ 1229 netboot = 0; 1230 mbi = xbootp->bi_mb_info; 1231 if (mbi != NULL && mbi->flags & 0x2) { 1232 boot_device = mbi->boot_device >> 24; 1233 if (boot_device == 0x20) 1234 netboot++; 1235 str[0] = (boot_device >> 4) + '0'; 1236 str[1] = (boot_device & 0xf) + '0'; 1237 str[2] = 0; 1238 bsetprops("bios-boot-device", str); 1239 } else { 1240 netboot = 1; 1241 } 1242 1243 /* 1244 * In the netboot case, drives_info is overloaded with the dhcp ack. 1245 * This is not multiboot compliant and requires special pxegrub! 1246 */ 1247 if (netboot && mbi->drives_length != 0) { 1248 sip = (struct sol_netinfo *)(uintptr_t)mbi->drives_addr; 1249 if (sip->sn_infotype == SN_TYPE_BOOTP) 1250 bsetprop("bootp-response", sizeof ("bootp-response"), 1251 (void *)(uintptr_t)mbi->drives_addr, 1252 mbi->drives_length); 1253 else if (sip->sn_infotype == SN_TYPE_RARP) 1254 setup_rarp_props(sip); 1255 } 1256 bsetprop("stdout", strlen("stdout"), 1257 &stdout_val, sizeof (stdout_val)); 1258#endif /* __xpv */ 1259 1260 /* 1261 * more conjured up values for made up things.... 1262 */ 1263#if defined(__xpv) 1264 bsetprops("mfg-name", "i86xpv"); 1265 bsetprops("impl-arch-name", "i86xpv"); 1266#else 1267 bsetprops("mfg-name", "i86pc"); 1268 bsetprops("impl-arch-name", "i86pc"); 1269#endif 1270 1271 /* 1272 * Build firmware-provided system properties 1273 */ 1274 build_firmware_properties(); 1275 1276 /* 1277 * XXPV 1278 * 1279 * Find out what these are: 1280 * - cpuid_feature_ecx_include 1281 * - cpuid_feature_ecx_exclude 1282 * - cpuid_feature_edx_include 1283 * - cpuid_feature_edx_exclude 1284 * 1285 * Find out what these are in multiboot: 1286 * - netdev-path 1287 * - fstype 1288 */ 1289} 1290 1291#ifdef __xpv 1292/* 1293 * Under the Hypervisor, memory usable for DMA may be scarce. One 1294 * very likely large pool of DMA friendly memory is occupied by 1295 * the boot_archive, as it was loaded by grub into low MFNs. 1296 * 1297 * Here we free up that memory by copying the boot archive to what are 1298 * likely higher MFN pages and then swapping the mfn/pfn mappings. 1299 */ 1300#define PFN_2GIG 0x80000 1301static void 1302relocate_boot_archive(void) 1303{ 1304 mfn_t max_mfn = HYPERVISOR_memory_op(XENMEM_maximum_ram_page, NULL); 1305 struct boot_modules *bm = xbootp->bi_modules; 1306 uintptr_t va; 1307 pfn_t va_pfn; 1308 mfn_t va_mfn; 1309 caddr_t copy; 1310 pfn_t copy_pfn; 1311 mfn_t copy_mfn; 1312 size_t len; 1313 int slop; 1314 int total = 0; 1315 int relocated = 0; 1316 int mmu_update_return; 1317 mmu_update_t t[2]; 1318 x86pte_t pte; 1319 1320 /* 1321 * If all MFN's are below 2Gig, don't bother doing this. 1322 */ 1323 if (max_mfn < PFN_2GIG) 1324 return; 1325 if (xbootp->bi_module_cnt < 1) { 1326 DBG_MSG("no boot_archive!"); 1327 return; 1328 } 1329 1330 DBG_MSG("moving boot_archive to high MFN memory\n"); 1331 va = (uintptr_t)bm->bm_addr; 1332 len = bm->bm_size; 1333 slop = va & MMU_PAGEOFFSET; 1334 if (slop) { 1335 va += MMU_PAGESIZE - slop; 1336 len -= MMU_PAGESIZE - slop; 1337 } 1338 len = P2ALIGN(len, MMU_PAGESIZE); 1339 1340 /* 1341 * Go through all boot_archive pages, swapping any low MFN pages 1342 * with memory at next_phys. 1343 */ 1344 while (len != 0) { 1345 ++total; 1346 va_pfn = mmu_btop(va - ONE_GIG); 1347 va_mfn = mfn_list[va_pfn]; 1348 if (mfn_list[va_pfn] < PFN_2GIG) { 1349 copy = kbm_remap_window(next_phys, 1); 1350 bcopy((void *)va, copy, MMU_PAGESIZE); 1351 copy_pfn = mmu_btop(next_phys); 1352 copy_mfn = mfn_list[copy_pfn]; 1353 1354 pte = mfn_to_ma(copy_mfn) | PT_NOCONSIST | PT_VALID; 1355 if (HYPERVISOR_update_va_mapping(va, pte, 1356 UVMF_INVLPG | UVMF_LOCAL)) 1357 bop_panic("relocate_boot_archive(): " 1358 "HYPERVISOR_update_va_mapping() failed"); 1359 1360 mfn_list[va_pfn] = copy_mfn; 1361 mfn_list[copy_pfn] = va_mfn; 1362 1363 t[0].ptr = mfn_to_ma(copy_mfn) | MMU_MACHPHYS_UPDATE; 1364 t[0].val = va_pfn; 1365 t[1].ptr = mfn_to_ma(va_mfn) | MMU_MACHPHYS_UPDATE; 1366 t[1].val = copy_pfn; 1367 if (HYPERVISOR_mmu_update(t, 2, &mmu_update_return, 1368 DOMID_SELF) != 0 || mmu_update_return != 2) 1369 bop_panic("relocate_boot_archive(): " 1370 "HYPERVISOR_mmu_update() failed"); 1371 1372 next_phys += MMU_PAGESIZE; 1373 ++relocated; 1374 } 1375 len -= MMU_PAGESIZE; 1376 va += MMU_PAGESIZE; 1377 } 1378 DBG_MSG("Relocated pages:\n"); 1379 DBG(relocated); 1380 DBG_MSG("Out of total pages:\n"); 1381 DBG(total); 1382} 1383#endif /* __xpv */ 1384 1385#if !defined(__xpv) 1386/* 1387 * Install a temporary IDT that lets us catch errors in the boot time code. 1388 * We shouldn't get any faults at all while this is installed, so we'll 1389 * just generate a traceback and exit. 1390 */ 1391#ifdef __amd64 1392static const int bcode_sel = B64CODE_SEL; 1393#else 1394static const int bcode_sel = B32CODE_SEL; 1395#endif 1396 1397/* 1398 * simple description of a stack frame (args are 32 bit only currently) 1399 */ 1400typedef struct bop_frame { 1401 struct bop_frame *old_frame; 1402 pc_t retaddr; 1403 long arg[1]; 1404} bop_frame_t; 1405 1406void 1407bop_traceback(bop_frame_t *frame) 1408{ 1409 pc_t pc; 1410 int cnt; 1411 int a; 1412 char *ksym; 1413 ulong_t off; 1414 1415 bop_printf(NULL, "Stack traceback:\n"); 1416 for (cnt = 0; cnt < 30; ++cnt) { /* up to 30 frames */ 1417 pc = frame->retaddr; 1418 if (pc == 0) 1419 break; 1420 ksym = kobj_getsymname(pc, &off); 1421 if (ksym) 1422 bop_printf(NULL, " %s+%lx", ksym, off); 1423 else 1424 bop_printf(NULL, " 0x%lx", pc); 1425 1426 frame = frame->old_frame; 1427 if (frame == 0) { 1428 bop_printf(NULL, "\n"); 1429 break; 1430 } 1431 for (a = 0; a < 6; ++a) { /* try for 6 args */ 1432#if defined(__i386) 1433 if ((void *)&frame->arg[a] == (void *)frame->old_frame) 1434 break; 1435 if (a == 0) 1436 bop_printf(NULL, "("); 1437 else 1438 bop_printf(NULL, ","); 1439 bop_printf(NULL, "0x%lx", frame->arg[a]); 1440#endif 1441 } 1442 bop_printf(NULL, ")\n"); 1443 } 1444} 1445 1446struct trapframe { 1447 ulong_t frame_ptr; /* %[er]bp pushed by our code */ 1448 ulong_t error_code; /* optional */ 1449 ulong_t inst_ptr; 1450 ulong_t code_seg; 1451 ulong_t flags_reg; 1452#ifdef __amd64 1453 ulong_t stk_ptr; 1454 ulong_t stk_seg; 1455#endif 1456}; 1457 1458void 1459bop_trap(struct trapframe *tf) 1460{ 1461 bop_frame_t fakeframe; 1462 static int depth = 0; 1463 1464 /* 1465 * Check for an infinite loop of traps. 1466 */ 1467 if (++depth > 2) 1468 bop_panic("Nested trap"); 1469 1470 /* 1471 * adjust the tf for optional error_code by detecting the code selector 1472 */ 1473 if (tf->code_seg != bcode_sel) 1474 tf = (struct trapframe *)((uintptr_t)tf - sizeof (ulong_t)); 1475 1476 bop_printf(NULL, "Unexpected trap\n"); 1477 bop_printf(NULL, "instruction pointer 0x%lx\n", tf->inst_ptr); 1478 bop_printf(NULL, "error code, optional 0x%lx\n", 1479 tf->error_code & 0xffffffff); 1480 bop_printf(NULL, "code segment 0x%lx\n", tf->code_seg & 0xffff); 1481 bop_printf(NULL, "flags register 0x%lx\n", tf->flags_reg); 1482#ifdef __amd64 1483 bop_printf(NULL, "return %%rsp 0x%lx\n", tf->stk_ptr); 1484 bop_printf(NULL, "return %%ss 0x%lx\n", tf->stk_seg & 0xffff); 1485#endif 1486 fakeframe.old_frame = (bop_frame_t *)tf->frame_ptr; 1487 fakeframe.retaddr = (pc_t)tf->inst_ptr; 1488 bop_printf(NULL, "Attempting stack backtrace:\n"); 1489 bop_traceback(&fakeframe); 1490 bop_panic("unexpected trap in early boot"); 1491} 1492 1493extern void bop_trap_handler(void); 1494 1495static gate_desc_t *bop_idt; 1496 1497static desctbr_t bop_idt_info; 1498 1499static void 1500bop_idt_init(void) 1501{ 1502 int t; 1503 1504 bop_idt = (gate_desc_t *) 1505 do_bsys_alloc(NULL, NULL, MMU_PAGESIZE, MMU_PAGESIZE); 1506 bzero(bop_idt, MMU_PAGESIZE); 1507 for (t = 0; t < NIDT; ++t) { 1508 set_gatesegd(&bop_idt[t], &bop_trap_handler, bcode_sel, 1509 SDT_SYSIGT, TRP_KPL); 1510 } 1511 bop_idt_info.dtr_limit = (NIDT * sizeof (gate_desc_t)) - 1; 1512 bop_idt_info.dtr_base = (uintptr_t)bop_idt; 1513 wr_idtr(&bop_idt_info); 1514} 1515#endif /* !defined(__xpv) */ 1516 1517/* 1518 * This is where we enter the kernel. It dummies up the boot_ops and 1519 * boot_syscalls vectors and jumps off to _kobj_boot() 1520 */ 1521void 1522_start(struct xboot_info *xbp) 1523{ 1524 bootops_t *bops = &bootop; 1525 extern void _kobj_boot(); 1526 1527 /* 1528 * 1st off - initialize the console for any error messages 1529 */ 1530 xbootp = xbp; 1531#ifdef __xpv 1532 HYPERVISOR_shared_info = (void *)xbootp->bi_shared_info; 1533 xen_info = xbootp->bi_xen_start_info; 1534#endif 1535 bcons_init((void *)xbootp->bi_cmdline); 1536 have_console = 1; 1537 1538 /* 1539 * enable debugging 1540 */ 1541 if (strstr((char *)xbootp->bi_cmdline, "kbm_debug")) 1542 kbm_debug = 1; 1543 1544 DBG_MSG("\n\n*** Entered Solaris in _start() cmdline is: "); 1545 DBG_MSG((char *)xbootp->bi_cmdline); 1546 DBG_MSG("\n\n\n"); 1547 1548 /* 1549 * physavail is no longer used by startup 1550 */ 1551 bm.physinstalled = xbp->bi_phys_install; 1552 bm.pcimem = xbp->bi_pcimem; 1553 bm.physavail = NULL; 1554 1555 /* 1556 * initialize the boot time allocator 1557 */ 1558 next_phys = xbootp->bi_next_paddr; 1559 DBG(next_phys); 1560 next_virt = (uintptr_t)xbootp->bi_next_vaddr; 1561 DBG(next_virt); 1562 DBG_MSG("Initializing boot time memory management..."); 1563#ifdef __xpv 1564 { 1565 xen_platform_parameters_t p; 1566 1567 /* This call shouldn't fail, dboot already did it once. */ 1568 (void) HYPERVISOR_xen_version(XENVER_platform_parameters, &p); 1569 mfn_to_pfn_mapping = (pfn_t *)(xen_virt_start = p.virt_start); 1570 DBG(xen_virt_start); 1571 } 1572#endif 1573 kbm_init(xbootp); 1574 DBG_MSG("done\n"); 1575 1576 /* 1577 * Fill in the bootops vector 1578 */ 1579 bops->bsys_version = BO_VERSION; 1580 bops->boot_mem = &bm; 1581 bops->bsys_alloc = do_bsys_alloc; 1582 bops->bsys_free = do_bsys_free; 1583 bops->bsys_getproplen = do_bsys_getproplen; 1584 bops->bsys_getprop = do_bsys_getprop; 1585 bops->bsys_nextprop = do_bsys_nextprop; 1586 bops->bsys_printf = bop_printf; 1587 bops->bsys_doint = do_bsys_doint; 1588 1589 /* 1590 * BOP_EALLOC() is no longer needed 1591 */ 1592 bops->bsys_ealloc = do_bsys_ealloc; 1593 1594#ifdef __xpv 1595 /* 1596 * On domain 0 we need to free up some physical memory that is 1597 * usable for DMA. Since GRUB loaded the boot_archive, it is 1598 * sitting in low MFN memory. We'll relocated the boot archive 1599 * pages to high PFN memory. 1600 */ 1601 if (DOMAIN_IS_INITDOMAIN(xen_info)) 1602 relocate_boot_archive(); 1603#endif 1604 1605#ifndef __xpv 1606 /* 1607 * Install an IDT to catch early pagefaults (shouldn't have any). 1608 * Also needed for kmdb. 1609 */ 1610 bop_idt_init(); 1611#endif 1612 1613 /* 1614 * Start building the boot properties from the command line 1615 */ 1616 DBG_MSG("Initializing boot properties:\n"); 1617 build_boot_properties(); 1618 1619 if (strstr((char *)xbootp->bi_cmdline, "prom_debug") || kbm_debug) { 1620 char *name; 1621 char *value; 1622 char *cp; 1623 int len; 1624 1625 value = do_bsys_alloc(NULL, NULL, MMU_PAGESIZE, MMU_PAGESIZE); 1626 bop_printf(NULL, "\nBoot properties:\n"); 1627 name = ""; 1628 while ((name = do_bsys_nextprop(NULL, name)) != NULL) { 1629 bop_printf(NULL, "\t0x%p %s = ", (void *)name, name); 1630 (void) do_bsys_getprop(NULL, name, value); 1631 len = do_bsys_getproplen(NULL, name); 1632 bop_printf(NULL, "len=%d ", len); 1633 value[len] = 0; 1634 for (cp = value; *cp; ++cp) { 1635 if (' ' <= *cp && *cp <= '~') 1636 bop_printf(NULL, "%c", *cp); 1637 else 1638 bop_printf(NULL, "-0x%x-", *cp); 1639 } 1640 bop_printf(NULL, "\n"); 1641 } 1642 } 1643 1644 /* 1645 * jump into krtld... 1646 */ 1647 _kobj_boot(&bop_sysp, NULL, bops, NULL); 1648} 1649 1650 1651/*ARGSUSED*/ 1652static caddr_t 1653no_more_alloc(bootops_t *bop, caddr_t virthint, size_t size, int align) 1654{ 1655 panic("Attempt to bsys_alloc() too late\n"); 1656 return (NULL); 1657} 1658 1659/*ARGSUSED*/ 1660static void 1661no_more_free(bootops_t *bop, caddr_t virt, size_t size) 1662{ 1663 panic("Attempt to bsys_free() too late\n"); 1664} 1665 1666void 1667bop_no_more_mem(void) 1668{ 1669 DBG(total_bop_alloc_scratch); 1670 DBG(total_bop_alloc_kernel); 1671 bootops->bsys_alloc = no_more_alloc; 1672 bootops->bsys_free = no_more_free; 1673} 1674 1675 1676#ifndef __xpv 1677/* 1678 * Set ACPI firmware properties 1679 */ 1680 1681static caddr_t 1682vmap_phys(size_t length, paddr_t pa) 1683{ 1684 paddr_t start, end; 1685 caddr_t va; 1686 size_t len, page; 1687 1688 start = P2ALIGN(pa, MMU_PAGESIZE); 1689 end = P2ROUNDUP(pa + length, MMU_PAGESIZE); 1690 len = end - start; 1691 va = (caddr_t)alloc_vaddr(len, MMU_PAGESIZE); 1692 for (page = 0; page < len; page += MMU_PAGESIZE) 1693 kbm_map((uintptr_t)va + page, start + page, 0, 0); 1694 return (va + (pa & MMU_PAGEOFFSET)); 1695} 1696 1697static uint8_t 1698checksum_table(uint8_t *tp, size_t len) 1699{ 1700 uint8_t sum = 0; 1701 1702 while (len-- > 0) 1703 sum += *tp++; 1704 1705 return (sum); 1706} 1707 1708static int 1709valid_rsdp(struct rsdp *rp) 1710{ 1711 1712 /* validate the V1.x checksum */ 1713 if (checksum_table((uint8_t *)&rp->v1, sizeof (struct rsdp_v1)) != 0) 1714 return (0); 1715 1716 /* If pre-ACPI 2.0, this is a valid RSDP */ 1717 if (rp->v1.revision < 2) 1718 return (1); 1719 1720 /* validate the V2.x checksum */ 1721 if (checksum_table((uint8_t *)rp, sizeof (struct rsdp)) != 0) 1722 return (0); 1723 1724 return (1); 1725} 1726 1727/* 1728 * Scan memory range for an RSDP; 1729 * see ACPI 3.0 Spec, 5.2.5.1 1730 */ 1731static struct rsdp * 1732scan_rsdp(paddr_t start, paddr_t end) 1733{ 1734 size_t len = end - start + 1; 1735 caddr_t ptr; 1736 1737 ptr = vmap_phys(len, start); 1738 while (len > 0) { 1739 if (strncmp(ptr, ACPI_RSDP_SIG, ACPI_RSDP_SIG_LEN) == 0) 1740 if (valid_rsdp((struct rsdp *)ptr)) 1741 return ((struct rsdp *)ptr); 1742 ptr += 16; 1743 len -= 16; 1744 } 1745 1746 return (NULL); 1747} 1748 1749/* 1750 * Refer to ACPI 3.0 Spec, section 5.2.5.1 to understand this function 1751 */ 1752static struct rsdp * 1753find_rsdp() { 1754 struct rsdp *rsdp; 1755 uint16_t *ebda_seg; 1756 paddr_t ebda_addr; 1757 1758 /* 1759 * Get the EBDA segment and scan the first 1K 1760 */ 1761 ebda_seg = (uint16_t *)vmap_phys(sizeof (uint16_t), ACPI_EBDA_SEG_ADDR); 1762 ebda_addr = *ebda_seg << 4; 1763 rsdp = scan_rsdp(ebda_addr, ebda_addr + ACPI_EBDA_LEN - 1); 1764 if (rsdp == NULL) 1765 /* if EBDA doesn't contain RSDP, look in BIOS memory */ 1766 rsdp = scan_rsdp(0xe0000, 0xfffff); 1767 return (rsdp); 1768} 1769 1770static struct table_header * 1771map_fw_table(paddr_t table_addr) 1772{ 1773 struct table_header *tp; 1774 size_t len = MAX(sizeof (struct table_header), MMU_PAGESIZE); 1775 1776 /* 1777 * Map at least a page; if the table is larger than this, remap it 1778 */ 1779 tp = (struct table_header *)vmap_phys(len, table_addr); 1780 if (tp->len > len) 1781 tp = (struct table_header *)vmap_phys(tp->len, table_addr); 1782 return (tp); 1783} 1784 1785static struct table_header * 1786find_fw_table(char *signature) 1787{ 1788 static int revision = 0; 1789 static struct xsdt *xsdt; 1790 static int len; 1791 paddr_t xsdt_addr; 1792 struct rsdp *rsdp; 1793 struct table_header *tp; 1794 paddr_t table_addr; 1795 int n; 1796 1797 if (strlen(signature) != ACPI_TABLE_SIG_LEN) 1798 return (NULL); 1799 1800 /* 1801 * Reading the ACPI 3.0 Spec, section 5.2.5.3 will help 1802 * understand this code. If we haven't already found the RSDT/XSDT, 1803 * revision will be 0. Find the RSDP and check the revision 1804 * to find out whether to use the RSDT or XSDT. If revision is 1805 * 0 or 1, use the RSDT and set internal revision to 1; if it is 2, 1806 * use the XSDT. If the XSDT address is 0, though, fall back to 1807 * revision 1 and use the RSDT. 1808 */ 1809 if (revision == 0) { 1810 if ((rsdp = (struct rsdp *)find_rsdp()) != NULL) { 1811 revision = rsdp->v1.revision; 1812 switch (revision) { 1813 case 2: 1814 /* 1815 * Use the XSDT unless BIOS is buggy and 1816 * claims to be rev 2 but has a null XSDT 1817 * address 1818 */ 1819 xsdt_addr = rsdp->xsdt; 1820 if (xsdt_addr != 0) 1821 break; 1822 /* FALLTHROUGH */ 1823 case 0: 1824 /* treat RSDP rev 0 as revision 1 internally */ 1825 revision = 1; 1826 /* FALLTHROUGH */ 1827 case 1: 1828 /* use the RSDT for rev 0/1 */ 1829 xsdt_addr = rsdp->v1.rsdt; 1830 break; 1831 default: 1832 /* unknown revision */ 1833 revision = 0; 1834 break; 1835 } 1836 } 1837 if (revision == 0) 1838 return (NULL); 1839 1840 /* cache the XSDT info */ 1841 xsdt = (struct xsdt *)map_fw_table(xsdt_addr); 1842 len = (xsdt->hdr.len - sizeof (xsdt->hdr)) / 1843 ((revision == 1) ? sizeof (uint32_t) : sizeof (uint64_t)); 1844 } 1845 1846 /* 1847 * Scan the table headers looking for a signature match 1848 */ 1849 for (n = 0; n < len; n++) { 1850 table_addr = (revision == 1) ? xsdt->p.r[n] : xsdt->p.x[n]; 1851 if (table_addr == 0) 1852 continue; 1853 tp = map_fw_table(table_addr); 1854 if (strncmp(tp->sig, signature, ACPI_TABLE_SIG_LEN) == 0) { 1855 return (tp); 1856 } 1857 } 1858 return (NULL); 1859} 1860 1861static void 1862process_madt(struct madt *tp) 1863{ 1864 struct madt_processor *cpu, *end; 1865 uint32_t cpu_count = 0; 1866 1867 /* 1868 * User-set boot-ncpus overrides firmware count 1869 */ 1870 if (do_bsys_getproplen(NULL, "boot-ncpus") >= 0) 1871 return; 1872 1873 if (tp != NULL) { 1874 end = (struct madt_processor *)(tp->hdr.len + (uintptr_t)tp); 1875 cpu = tp->list; 1876 while (cpu < end) { 1877 if (cpu->type == MADT_PROCESSOR) 1878 if (cpu->flags & 1) 1879 cpu_count++; 1880 1881 cpu = (struct madt_processor *) 1882 (cpu->len + (uintptr_t)cpu); 1883 } 1884 bsetpropsi("boot-ncpus", cpu_count); 1885 } 1886 1887} 1888 1889static void 1890process_srat(struct srat *tp) 1891{ 1892 struct srat_item *item, *end; 1893 int i; 1894 int proc_num, mem_num; 1895#pragma pack(1) 1896 struct { 1897 uint32_t domain; 1898 uint32_t apic_id; 1899 uint32_t sapic_id; 1900 } processor; 1901 struct { 1902 uint32_t domain; 1903 uint64_t addr; 1904 uint64_t length; 1905 uint32_t flags; 1906 } memory; 1907#pragma pack() 1908 char prop_name[30]; 1909 1910 if (tp == NULL) 1911 return; 1912 1913 proc_num = mem_num = 0; 1914 end = (struct srat_item *)(tp->hdr.len + (uintptr_t)tp); 1915 item = tp->list; 1916 while (item < end) { 1917 switch (item->type) { 1918 case SRAT_PROCESSOR: 1919 if (!(item->i.p.flags & SRAT_ENABLED)) 1920 break; 1921 processor.domain = item->i.p.domain1; 1922 for (i = 0; i < 3; i++) 1923 processor.domain += 1924 item->i.p.domain2[i] << ((i + 1) * 8); 1925 processor.apic_id = item->i.p.apic_id; 1926 processor.sapic_id = item->i.p.local_sapic_eid; 1927 (void) snprintf(prop_name, 30, "acpi-srat-processor-%d", 1928 proc_num); 1929 bsetprop(prop_name, strlen(prop_name), &processor, 1930 sizeof (processor)); 1931 proc_num++; 1932 break; 1933 case SRAT_MEMORY: 1934 if (!(item->i.m.flags & SRAT_ENABLED)) 1935 break; 1936 memory.domain = item->i.m.domain; 1937 memory.addr = item->i.m.base_addr; 1938 memory.length = item->i.m.len; 1939 memory.flags = item->i.m.flags; 1940 (void) snprintf(prop_name, 30, "acpi-srat-memory-%d", 1941 mem_num); 1942 bsetprop(prop_name, strlen(prop_name), &memory, 1943 sizeof (memory)); 1944 mem_num++; 1945 break; 1946 } 1947 1948 item = (struct srat_item *) 1949 (item->len + (caddr_t)item); 1950 } 1951} 1952 1953static void 1954process_slit(struct slit *tp) 1955{ 1956 1957 /* 1958 * Check the number of localities; if it's too huge, we just 1959 * return and locality enumeration code will handle this later, 1960 * if possible. 1961 * 1962 * Note that the size of the table is the square of the 1963 * number of localities; if the number of localities exceeds 1964 * UINT16_MAX, the table size may overflow an int when being 1965 * passed to bsetprop() below. 1966 */ 1967 if (tp->number >= SLIT_LOCALITIES_MAX) 1968 return; 1969 1970 bsetprop(SLIT_NUM_PROPNAME, strlen(SLIT_NUM_PROPNAME), &tp->number, 1971 sizeof (tp->number)); 1972 bsetprop(SLIT_PROPNAME, strlen(SLIT_PROPNAME), &tp->entry, 1973 tp->number * tp->number); 1974} 1975#else /* __xpv */ 1976static void 1977enumerate_xen_cpus() 1978{ 1979 processorid_t id, max_id; 1980 1981 /* 1982 * User-set boot-ncpus overrides enumeration 1983 */ 1984 if (do_bsys_getproplen(NULL, "boot-ncpus") >= 0) 1985 return; 1986 1987 /* 1988 * Probe every possible virtual CPU id and remember the 1989 * highest id present; the count of CPUs is one greater 1990 * than this. This tacitly assumes at least cpu 0 is present. 1991 */ 1992 max_id = 0; 1993 for (id = 0; id < MAX_VIRT_CPUS; id++) 1994 if (HYPERVISOR_vcpu_op(VCPUOP_is_up, id, NULL) == 0) 1995 max_id = id; 1996 1997 bsetpropsi("boot-ncpus", max_id+1); 1998 1999} 2000#endif /* __xpv */ 2001 2002static void 2003build_firmware_properties(void) 2004{ 2005#ifndef __xpv 2006 struct table_header *tp; 2007 2008 if ((tp = find_fw_table("APIC")) != NULL) 2009 process_madt((struct madt *)tp); 2010 2011 if ((tp = find_fw_table("SRAT")) != NULL) 2012 process_srat((struct srat *)tp); 2013 2014 if (tp = find_fw_table("SLIT")) 2015 process_slit((struct slit *)tp); 2016#else /* __xpv */ 2017 enumerate_xen_cpus(); 2018#endif /* __xpv */ 2019} 2020 2021/* 2022 * fake up a boot property for USB serial console early boot output 2023 */ 2024void * 2025usbser_init(size_t size) 2026{ 2027 static char *p = NULL; 2028 2029 p = do_bsys_alloc(NULL, NULL, size, MMU_PAGESIZE); 2030 *p = 0; 2031 bsetprop("usb-serial-buf", strlen("usb-serial-buf") + 1, 2032 &p, sizeof (p)); 2033 return (p); 2034} 2035 2036/*ARGSUSED*/ 2037int 2038boot_compinfo(int fd, struct compinfo *cbp) 2039{ 2040 cbp->iscmp = 0; 2041 cbp->blksize = MAXBSIZE; 2042 return (0); 2043} 2044