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 2009 Sun Microsystems, Inc. All rights reserved. 24 * Use is subject to license terms. 25 * 26 * Copyright 2013 Joyent, Inc. All rights reserved. 27 */ 28 29 30 #include <sys/types.h> 31 #include <sys/machparam.h> 32 #include <sys/x86_archext.h> 33 #include <sys/systm.h> 34 #include <sys/mach_mmu.h> 35 #include <sys/multiboot.h> 36 #include <sys/multiboot2.h> 37 #include <sys/multiboot2_impl.h> 38 #include <sys/sysmacros.h> 39 #include <sys/framebuffer.h> 40 #include <sys/sha1.h> 41 #include <util/string.h> 42 #include <util/strtolctype.h> 43 #include <sys/efi.h> 44 45 /* 46 * Compile time debug knob. We do not have any early mechanism to control it 47 * as the boot is the earliest mechanism we have, and we do not want to have 48 * it being switched on by default. 49 */ 50 int dboot_debug = 0; 51 52 #if defined(__xpv) 53 54 #include <sys/hypervisor.h> 55 uintptr_t xen_virt_start; 56 pfn_t *mfn_to_pfn_mapping; 57 58 #else /* !__xpv */ 59 60 extern multiboot_header_t mb_header; 61 extern uint32_t mb2_load_addr; 62 extern int have_cpuid(void); 63 64 #endif /* !__xpv */ 65 66 #include <sys/inttypes.h> 67 #include <sys/bootinfo.h> 68 #include <sys/mach_mmu.h> 69 #include <sys/boot_console.h> 70 71 #include "dboot_asm.h" 72 #include "dboot_printf.h" 73 #include "dboot_xboot.h" 74 #include "dboot_elfload.h" 75 76 #define SHA1_ASCII_LENGTH (SHA1_DIGEST_LENGTH * 2) 77 78 /* 79 * This file contains code that runs to transition us from either a multiboot 80 * compliant loader (32 bit non-paging) or a XPV domain loader to 81 * regular kernel execution. Its task is to setup the kernel memory image 82 * and page tables. 83 * 84 * The code executes as: 85 * - 32 bits under GRUB (for 32 or 64 bit Solaris) 86 * - a 32 bit program for the 32-bit PV hypervisor 87 * - a 64 bit program for the 64-bit PV hypervisor (at least for now) 88 * 89 * Under the PV hypervisor, we must create mappings for any memory beyond the 90 * initial start of day allocation (such as the kernel itself). 91 * 92 * When on the metal, the mapping between maddr_t and paddr_t is 1:1. 93 * Since we are running in real mode, so all such memory is accessible. 94 */ 95 96 /* 97 * Standard bits used in PTE (page level) and PTP (internal levels) 98 */ 99 x86pte_t ptp_bits = PT_VALID | PT_REF | PT_WRITABLE | PT_USER; 100 x86pte_t pte_bits = PT_VALID | PT_REF | PT_WRITABLE | PT_MOD | PT_NOCONSIST; 101 102 /* 103 * This is the target addresses (physical) where the kernel text and data 104 * nucleus pages will be unpacked. On the hypervisor this is actually a 105 * virtual address. 106 */ 107 paddr_t ktext_phys; 108 uint32_t ksize = 2 * FOUR_MEG; /* kernel nucleus is 8Meg */ 109 110 static uint64_t target_kernel_text; /* value to use for KERNEL_TEXT */ 111 112 /* 113 * The stack is setup in assembler before entering startup_kernel() 114 */ 115 char stack_space[STACK_SIZE]; 116 117 /* 118 * Used to track physical memory allocation 119 */ 120 static paddr_t next_avail_addr = 0; 121 122 #if defined(__xpv) 123 /* 124 * Additional information needed for hypervisor memory allocation. 125 * Only memory up to scratch_end is mapped by page tables. 126 * mfn_base is the start of the hypervisor virtual image. It's ONE_GIG, so 127 * to derive a pfn from a pointer, you subtract mfn_base. 128 */ 129 130 static paddr_t scratch_end = 0; /* we can't write all of mem here */ 131 static paddr_t mfn_base; /* addr corresponding to mfn_list[0] */ 132 start_info_t *xen_info; 133 134 #else /* __xpv */ 135 136 /* 137 * If on the metal, then we have a multiboot loader. 138 */ 139 uint32_t mb_magic; /* magic from boot loader */ 140 uint32_t mb_addr; /* multiboot info package from loader */ 141 int multiboot_version; 142 multiboot_info_t *mb_info; 143 multiboot2_info_header_t *mb2_info; 144 multiboot_tag_mmap_t *mb2_mmap_tagp; 145 int num_entries; /* mmap entry count */ 146 boolean_t num_entries_set; /* is mmap entry count set */ 147 uintptr_t load_addr; 148 static boot_framebuffer_t framebuffer __aligned(16); 149 static boot_framebuffer_t *fb; 150 151 /* can not be automatic variables because of alignment */ 152 static efi_guid_t smbios3 = SMBIOS3_TABLE_GUID; 153 static efi_guid_t smbios = SMBIOS_TABLE_GUID; 154 static efi_guid_t acpi2 = EFI_ACPI_TABLE_GUID; 155 static efi_guid_t acpi1 = ACPI_10_TABLE_GUID; 156 #endif /* __xpv */ 157 158 /* 159 * This contains information passed to the kernel 160 */ 161 struct xboot_info boot_info __aligned(16); 162 struct xboot_info *bi; 163 164 /* 165 * Page table and memory stuff. 166 */ 167 static paddr_t max_mem; /* maximum memory address */ 168 169 /* 170 * Information about processor MMU 171 */ 172 int amd64_support = 0; 173 int largepage_support = 0; 174 int pae_support = 0; 175 int pge_support = 0; 176 int NX_support = 0; 177 int PAT_support = 0; 178 179 /* 180 * Low 32 bits of kernel entry address passed back to assembler. 181 * When running a 64 bit kernel, the high 32 bits are 0xffffffff. 182 */ 183 uint32_t entry_addr_low; 184 185 /* 186 * Memlists for the kernel. We shouldn't need a lot of these. 187 */ 188 #define MAX_MEMLIST (50) 189 struct boot_memlist memlists[MAX_MEMLIST]; 190 uint_t memlists_used = 0; 191 struct boot_memlist pcimemlists[MAX_MEMLIST]; 192 uint_t pcimemlists_used = 0; 193 struct boot_memlist rsvdmemlists[MAX_MEMLIST]; 194 uint_t rsvdmemlists_used = 0; 195 196 /* 197 * This should match what's in the bootloader. It's arbitrary, but GRUB 198 * in particular has limitations on how much space it can use before it 199 * stops working properly. This should be enough. 200 */ 201 struct boot_modules modules[MAX_BOOT_MODULES]; 202 uint_t modules_used = 0; 203 204 #ifdef __xpv 205 /* 206 * Xen strips the size field out of the mb_memory_map_t, see struct e820entry 207 * definition in Xen source. 208 */ 209 typedef struct { 210 uint32_t base_addr_low; 211 uint32_t base_addr_high; 212 uint32_t length_low; 213 uint32_t length_high; 214 uint32_t type; 215 } mmap_t; 216 217 /* 218 * There is 512KB of scratch area after the boot stack page. 219 * We'll use that for everything except the kernel nucleus pages which are too 220 * big to fit there and are allocated last anyway. 221 */ 222 #define MAXMAPS 100 223 static mmap_t map_buffer[MAXMAPS]; 224 #else 225 typedef mb_memory_map_t mmap_t; 226 #endif 227 228 /* 229 * Debugging macros 230 */ 231 uint_t prom_debug = 0; 232 uint_t map_debug = 0; 233 234 static char noname[2] = "-"; 235 236 /* 237 * Either hypervisor-specific or grub-specific code builds the initial 238 * memlists. This code does the sort/merge/link for final use. 239 */ 240 static void 241 sort_physinstall(void) 242 { 243 int i; 244 #if !defined(__xpv) 245 int j; 246 struct boot_memlist tmp; 247 248 /* 249 * Now sort the memlists, in case they weren't in order. 250 * Yeah, this is a bubble sort; small, simple and easy to get right. 251 */ 252 DBG_MSG("Sorting phys-installed list\n"); 253 for (j = memlists_used - 1; j > 0; --j) { 254 for (i = 0; i < j; ++i) { 255 if (memlists[i].addr < memlists[i + 1].addr) 256 continue; 257 tmp = memlists[i]; 258 memlists[i] = memlists[i + 1]; 259 memlists[i + 1] = tmp; 260 } 261 } 262 263 /* 264 * Merge any memlists that don't have holes between them. 265 */ 266 for (i = 0; i <= memlists_used - 1; ++i) { 267 if (memlists[i].addr + memlists[i].size != memlists[i + 1].addr) 268 continue; 269 270 if (prom_debug) 271 dboot_printf( 272 "merging mem segs %" PRIx64 "...%" PRIx64 273 " w/ %" PRIx64 "...%" PRIx64 "\n", 274 memlists[i].addr, 275 memlists[i].addr + memlists[i].size, 276 memlists[i + 1].addr, 277 memlists[i + 1].addr + memlists[i + 1].size); 278 279 memlists[i].size += memlists[i + 1].size; 280 for (j = i + 1; j < memlists_used - 1; ++j) 281 memlists[j] = memlists[j + 1]; 282 --memlists_used; 283 DBG(memlists_used); 284 --i; /* after merging we need to reexamine, so do this */ 285 } 286 #endif /* __xpv */ 287 288 if (prom_debug) { 289 dboot_printf("\nFinal memlists:\n"); 290 for (i = 0; i < memlists_used; ++i) { 291 dboot_printf("\t%d: addr=%" PRIx64 " size=%" 292 PRIx64 "\n", i, memlists[i].addr, memlists[i].size); 293 } 294 } 295 296 /* 297 * link together the memlists with native size pointers 298 */ 299 memlists[0].next = 0; 300 memlists[0].prev = 0; 301 for (i = 1; i < memlists_used; ++i) { 302 memlists[i].prev = (native_ptr_t)(uintptr_t)(memlists + i - 1); 303 memlists[i].next = 0; 304 memlists[i - 1].next = (native_ptr_t)(uintptr_t)(memlists + i); 305 } 306 bi->bi_phys_install = (native_ptr_t)(uintptr_t)memlists; 307 DBG(bi->bi_phys_install); 308 } 309 310 /* 311 * build bios reserved memlists 312 */ 313 static void 314 build_rsvdmemlists(void) 315 { 316 int i; 317 318 rsvdmemlists[0].next = 0; 319 rsvdmemlists[0].prev = 0; 320 for (i = 1; i < rsvdmemlists_used; ++i) { 321 rsvdmemlists[i].prev = 322 (native_ptr_t)(uintptr_t)(rsvdmemlists + i - 1); 323 rsvdmemlists[i].next = 0; 324 rsvdmemlists[i - 1].next = 325 (native_ptr_t)(uintptr_t)(rsvdmemlists + i); 326 } 327 bi->bi_rsvdmem = (native_ptr_t)(uintptr_t)rsvdmemlists; 328 DBG(bi->bi_rsvdmem); 329 } 330 331 #if defined(__xpv) 332 333 /* 334 * halt on the hypervisor after a delay to drain console output 335 */ 336 void 337 dboot_halt(void) 338 { 339 uint_t i = 10000; 340 341 while (--i) 342 (void) HYPERVISOR_yield(); 343 (void) HYPERVISOR_shutdown(SHUTDOWN_poweroff); 344 } 345 346 /* 347 * From a machine address, find the corresponding pseudo-physical address. 348 * Pseudo-physical address are contiguous and run from mfn_base in each VM. 349 * Machine addresses are the real underlying hardware addresses. 350 * These are needed for page table entries. Note that this routine is 351 * poorly protected. A bad value of "ma" will cause a page fault. 352 */ 353 paddr_t 354 ma_to_pa(maddr_t ma) 355 { 356 ulong_t pgoff = ma & MMU_PAGEOFFSET; 357 ulong_t pfn = mfn_to_pfn_mapping[mmu_btop(ma)]; 358 paddr_t pa; 359 360 if (pfn >= xen_info->nr_pages) 361 return (-(paddr_t)1); 362 pa = mfn_base + mmu_ptob((paddr_t)pfn) + pgoff; 363 #ifdef DEBUG 364 if (ma != pa_to_ma(pa)) 365 dboot_printf("ma_to_pa(%" PRIx64 ") got %" PRIx64 ", " 366 "pa_to_ma() says %" PRIx64 "\n", ma, pa, pa_to_ma(pa)); 367 #endif 368 return (pa); 369 } 370 371 /* 372 * From a pseudo-physical address, find the corresponding machine address. 373 */ 374 maddr_t 375 pa_to_ma(paddr_t pa) 376 { 377 pfn_t pfn; 378 ulong_t mfn; 379 380 pfn = mmu_btop(pa - mfn_base); 381 if (pa < mfn_base || pfn >= xen_info->nr_pages) 382 dboot_panic("pa_to_ma(): illegal address 0x%lx", (ulong_t)pa); 383 mfn = ((ulong_t *)xen_info->mfn_list)[pfn]; 384 #ifdef DEBUG 385 if (mfn_to_pfn_mapping[mfn] != pfn) 386 dboot_printf("pa_to_ma(pfn=%lx) got %lx ma_to_pa() says %lx\n", 387 pfn, mfn, mfn_to_pfn_mapping[mfn]); 388 #endif 389 return (mfn_to_ma(mfn) | (pa & MMU_PAGEOFFSET)); 390 } 391 392 #endif /* __xpv */ 393 394 x86pte_t 395 get_pteval(paddr_t table, uint_t index) 396 { 397 if (pae_support) 398 return (((x86pte_t *)(uintptr_t)table)[index]); 399 return (((x86pte32_t *)(uintptr_t)table)[index]); 400 } 401 402 /*ARGSUSED*/ 403 void 404 set_pteval(paddr_t table, uint_t index, uint_t level, x86pte_t pteval) 405 { 406 #ifdef __xpv 407 mmu_update_t t; 408 maddr_t mtable = pa_to_ma(table); 409 int retcnt; 410 411 t.ptr = (mtable + index * pte_size) | MMU_NORMAL_PT_UPDATE; 412 t.val = pteval; 413 if (HYPERVISOR_mmu_update(&t, 1, &retcnt, DOMID_SELF) || retcnt != 1) 414 dboot_panic("HYPERVISOR_mmu_update() failed"); 415 #else /* __xpv */ 416 uintptr_t tab_addr = (uintptr_t)table; 417 418 if (pae_support) 419 ((x86pte_t *)tab_addr)[index] = pteval; 420 else 421 ((x86pte32_t *)tab_addr)[index] = (x86pte32_t)pteval; 422 if (level == top_level && level == 2) 423 reload_cr3(); 424 #endif /* __xpv */ 425 } 426 427 paddr_t 428 make_ptable(x86pte_t *pteval, uint_t level) 429 { 430 paddr_t new_table = (paddr_t)(uintptr_t)mem_alloc(MMU_PAGESIZE); 431 432 if (level == top_level && level == 2) 433 *pteval = pa_to_ma((uintptr_t)new_table) | PT_VALID; 434 else 435 *pteval = pa_to_ma((uintptr_t)new_table) | ptp_bits; 436 437 #ifdef __xpv 438 /* Remove write permission to the new page table. */ 439 if (HYPERVISOR_update_va_mapping(new_table, 440 *pteval & ~(x86pte_t)PT_WRITABLE, UVMF_INVLPG | UVMF_LOCAL)) 441 dboot_panic("HYP_update_va_mapping error"); 442 #endif 443 444 if (map_debug) 445 dboot_printf("new page table lvl=%d paddr=0x%lx ptp=0x%" 446 PRIx64 "\n", level, (ulong_t)new_table, *pteval); 447 return (new_table); 448 } 449 450 x86pte_t * 451 map_pte(paddr_t table, uint_t index) 452 { 453 return ((x86pte_t *)(uintptr_t)(table + index * pte_size)); 454 } 455 456 /* 457 * dump out the contents of page tables... 458 */ 459 static void 460 dump_tables(void) 461 { 462 uint_t save_index[4]; /* for recursion */ 463 char *save_table[4]; /* for recursion */ 464 uint_t l; 465 uint64_t va; 466 uint64_t pgsize; 467 int index; 468 int i; 469 x86pte_t pteval; 470 char *table; 471 static char *tablist = "\t\t\t"; 472 char *tabs = tablist + 3 - top_level; 473 uint_t pa, pa1; 474 #if !defined(__xpv) 475 #define maddr_t paddr_t 476 #endif /* !__xpv */ 477 478 dboot_printf("Finished pagetables:\n"); 479 table = (char *)(uintptr_t)top_page_table; 480 l = top_level; 481 va = 0; 482 for (index = 0; index < ptes_per_table; ++index) { 483 pgsize = 1ull << shift_amt[l]; 484 if (pae_support) 485 pteval = ((x86pte_t *)table)[index]; 486 else 487 pteval = ((x86pte32_t *)table)[index]; 488 if (pteval == 0) 489 goto next_entry; 490 491 dboot_printf("%s %p[0x%x] = %" PRIx64 ", va=%" PRIx64, 492 tabs + l, (void *)table, index, (uint64_t)pteval, va); 493 pa = ma_to_pa(pteval & MMU_PAGEMASK); 494 dboot_printf(" physaddr=%x\n", pa); 495 496 /* 497 * Don't try to walk hypervisor private pagetables 498 */ 499 if ((l > 1 || (l == 1 && (pteval & PT_PAGESIZE) == 0))) { 500 save_table[l] = table; 501 save_index[l] = index; 502 --l; 503 index = -1; 504 table = (char *)(uintptr_t) 505 ma_to_pa(pteval & MMU_PAGEMASK); 506 goto recursion; 507 } 508 509 /* 510 * shorten dump for consecutive mappings 511 */ 512 for (i = 1; index + i < ptes_per_table; ++i) { 513 if (pae_support) 514 pteval = ((x86pte_t *)table)[index + i]; 515 else 516 pteval = ((x86pte32_t *)table)[index + i]; 517 if (pteval == 0) 518 break; 519 pa1 = ma_to_pa(pteval & MMU_PAGEMASK); 520 if (pa1 != pa + i * pgsize) 521 break; 522 } 523 if (i > 2) { 524 dboot_printf("%s...\n", tabs + l); 525 va += pgsize * (i - 2); 526 index += i - 2; 527 } 528 next_entry: 529 va += pgsize; 530 if (l == 3 && index == 256) /* VA hole */ 531 va = 0xffff800000000000ull; 532 recursion: 533 ; 534 } 535 if (l < top_level) { 536 ++l; 537 index = save_index[l]; 538 table = save_table[l]; 539 goto recursion; 540 } 541 } 542 543 /* 544 * Add a mapping for the machine page at the given virtual address. 545 */ 546 static void 547 map_ma_at_va(maddr_t ma, native_ptr_t va, uint_t level) 548 { 549 x86pte_t *ptep; 550 x86pte_t pteval; 551 552 pteval = ma | pte_bits; 553 if (level > 0) 554 pteval |= PT_PAGESIZE; 555 if (va >= target_kernel_text && pge_support) 556 pteval |= PT_GLOBAL; 557 558 if (map_debug && ma != va) 559 dboot_printf("mapping ma=0x%" PRIx64 " va=0x%" PRIx64 560 " pte=0x%" PRIx64 " l=%d\n", 561 (uint64_t)ma, (uint64_t)va, pteval, level); 562 563 #if defined(__xpv) 564 /* 565 * see if we can avoid find_pte() on the hypervisor 566 */ 567 if (HYPERVISOR_update_va_mapping(va, pteval, 568 UVMF_INVLPG | UVMF_LOCAL) == 0) 569 return; 570 #endif 571 572 /* 573 * Find the pte that will map this address. This creates any 574 * missing intermediate level page tables 575 */ 576 ptep = find_pte(va, NULL, level, 0); 577 578 /* 579 * When paravirtualized, we must use hypervisor calls to modify the 580 * PTE, since paging is active. On real hardware we just write to 581 * the pagetables which aren't in use yet. 582 */ 583 #if defined(__xpv) 584 ptep = ptep; /* shut lint up */ 585 if (HYPERVISOR_update_va_mapping(va, pteval, UVMF_INVLPG | UVMF_LOCAL)) 586 dboot_panic("mmu_update failed-map_pa_at_va va=0x%" PRIx64 587 " l=%d ma=0x%" PRIx64 ", pte=0x%" PRIx64 "", 588 (uint64_t)va, level, (uint64_t)ma, pteval); 589 #else 590 if (va < 1024 * 1024) 591 pteval |= PT_NOCACHE; /* for video RAM */ 592 if (pae_support) 593 *ptep = pteval; 594 else 595 *((x86pte32_t *)ptep) = (x86pte32_t)pteval; 596 #endif 597 } 598 599 /* 600 * Add a mapping for the physical page at the given virtual address. 601 */ 602 static void 603 map_pa_at_va(paddr_t pa, native_ptr_t va, uint_t level) 604 { 605 map_ma_at_va(pa_to_ma(pa), va, level); 606 } 607 608 /* 609 * This is called to remove start..end from the 610 * possible range of PCI addresses. 611 */ 612 const uint64_t pci_lo_limit = 0x00100000ul; 613 const uint64_t pci_hi_limit = 0xfff00000ul; 614 static void 615 exclude_from_pci(uint64_t start, uint64_t end) 616 { 617 int i; 618 int j; 619 struct boot_memlist *ml; 620 621 for (i = 0; i < pcimemlists_used; ++i) { 622 ml = &pcimemlists[i]; 623 624 /* delete the entire range? */ 625 if (start <= ml->addr && ml->addr + ml->size <= end) { 626 --pcimemlists_used; 627 for (j = i; j < pcimemlists_used; ++j) 628 pcimemlists[j] = pcimemlists[j + 1]; 629 --i; /* to revisit the new one at this index */ 630 } 631 632 /* split a range? */ 633 else if (ml->addr < start && end < ml->addr + ml->size) { 634 635 ++pcimemlists_used; 636 if (pcimemlists_used > MAX_MEMLIST) 637 dboot_panic("too many pcimemlists"); 638 639 for (j = pcimemlists_used - 1; j > i; --j) 640 pcimemlists[j] = pcimemlists[j - 1]; 641 ml->size = start - ml->addr; 642 643 ++ml; 644 ml->size = (ml->addr + ml->size) - end; 645 ml->addr = end; 646 ++i; /* skip on to next one */ 647 } 648 649 /* cut memory off the start? */ 650 else if (ml->addr < end && end < ml->addr + ml->size) { 651 ml->size -= end - ml->addr; 652 ml->addr = end; 653 } 654 655 /* cut memory off the end? */ 656 else if (ml->addr <= start && start < ml->addr + ml->size) { 657 ml->size = start - ml->addr; 658 } 659 } 660 } 661 662 /* 663 * During memory allocation, find the highest address not used yet. 664 */ 665 static void 666 check_higher(paddr_t a) 667 { 668 if (a < next_avail_addr) 669 return; 670 next_avail_addr = RNDUP(a + 1, MMU_PAGESIZE); 671 DBG(next_avail_addr); 672 } 673 674 static int 675 dboot_loader_mmap_entries(void) 676 { 677 #if !defined(__xpv) 678 if (num_entries_set == B_TRUE) 679 return (num_entries); 680 681 switch (multiboot_version) { 682 case 1: 683 DBG(mb_info->flags); 684 if (mb_info->flags & 0x40) { 685 mb_memory_map_t *mmap; 686 687 DBG(mb_info->mmap_addr); 688 DBG(mb_info->mmap_length); 689 check_higher(mb_info->mmap_addr + mb_info->mmap_length); 690 691 for (mmap = (mb_memory_map_t *)mb_info->mmap_addr; 692 (uint32_t)mmap < mb_info->mmap_addr + 693 mb_info->mmap_length; 694 mmap = (mb_memory_map_t *)((uint32_t)mmap + 695 mmap->size + sizeof (mmap->size))) 696 ++num_entries; 697 698 num_entries_set = B_TRUE; 699 } 700 break; 701 case 2: 702 num_entries_set = B_TRUE; 703 num_entries = dboot_multiboot2_mmap_nentries(mb2_info, 704 mb2_mmap_tagp); 705 break; 706 default: 707 dboot_panic("Unknown multiboot version: %d\n", 708 multiboot_version); 709 break; 710 } 711 return (num_entries); 712 #else 713 return (MAXMAPS); 714 #endif 715 } 716 717 static uint32_t 718 dboot_loader_mmap_get_type(int index) 719 { 720 #if !defined(__xpv) 721 mb_memory_map_t *mp, *mpend; 722 int i; 723 724 switch (multiboot_version) { 725 case 1: 726 mp = (mb_memory_map_t *)mb_info->mmap_addr; 727 mpend = (mb_memory_map_t *) 728 (mb_info->mmap_addr + mb_info->mmap_length); 729 730 for (i = 0; mp < mpend && i != index; i++) 731 mp = (mb_memory_map_t *)((uint32_t)mp + mp->size + 732 sizeof (mp->size)); 733 if (mp >= mpend) { 734 dboot_panic("dboot_loader_mmap_get_type(): index " 735 "out of bounds: %d\n", index); 736 } 737 return (mp->type); 738 739 case 2: 740 return (dboot_multiboot2_mmap_get_type(mb2_info, 741 mb2_mmap_tagp, index)); 742 743 default: 744 dboot_panic("Unknown multiboot version: %d\n", 745 multiboot_version); 746 break; 747 } 748 return (0); 749 #else 750 return (map_buffer[index].type); 751 #endif 752 } 753 754 static uint64_t 755 dboot_loader_mmap_get_base(int index) 756 { 757 #if !defined(__xpv) 758 mb_memory_map_t *mp, *mpend; 759 int i; 760 761 switch (multiboot_version) { 762 case 1: 763 mp = (mb_memory_map_t *)mb_info->mmap_addr; 764 mpend = (mb_memory_map_t *) 765 (mb_info->mmap_addr + mb_info->mmap_length); 766 767 for (i = 0; mp < mpend && i != index; i++) 768 mp = (mb_memory_map_t *)((uint32_t)mp + mp->size + 769 sizeof (mp->size)); 770 if (mp >= mpend) { 771 dboot_panic("dboot_loader_mmap_get_base(): index " 772 "out of bounds: %d\n", index); 773 } 774 return (((uint64_t)mp->base_addr_high << 32) + 775 (uint64_t)mp->base_addr_low); 776 777 case 2: 778 return (dboot_multiboot2_mmap_get_base(mb2_info, 779 mb2_mmap_tagp, index)); 780 781 default: 782 dboot_panic("Unknown multiboot version: %d\n", 783 multiboot_version); 784 break; 785 } 786 return (0); 787 #else 788 return (((uint64_t)map_buffer[index].base_addr_high << 32) + 789 (uint64_t)map_buffer[index].base_addr_low); 790 #endif 791 } 792 793 static uint64_t 794 dboot_loader_mmap_get_length(int index) 795 { 796 #if !defined(__xpv) 797 mb_memory_map_t *mp, *mpend; 798 int i; 799 800 switch (multiboot_version) { 801 case 1: 802 mp = (mb_memory_map_t *)mb_info->mmap_addr; 803 mpend = (mb_memory_map_t *) 804 (mb_info->mmap_addr + mb_info->mmap_length); 805 806 for (i = 0; mp < mpend && i != index; i++) 807 mp = (mb_memory_map_t *)((uint32_t)mp + mp->size + 808 sizeof (mp->size)); 809 if (mp >= mpend) { 810 dboot_panic("dboot_loader_mmap_get_length(): index " 811 "out of bounds: %d\n", index); 812 } 813 return (((uint64_t)mp->length_high << 32) + 814 (uint64_t)mp->length_low); 815 816 case 2: 817 return (dboot_multiboot2_mmap_get_length(mb2_info, 818 mb2_mmap_tagp, index)); 819 820 default: 821 dboot_panic("Unknown multiboot version: %d\n", 822 multiboot_version); 823 break; 824 } 825 return (0); 826 #else 827 return (((uint64_t)map_buffer[index].length_high << 32) + 828 (uint64_t)map_buffer[index].length_low); 829 #endif 830 } 831 832 static void 833 build_pcimemlists(void) 834 { 835 uint64_t page_offset = MMU_PAGEOFFSET; /* needs to be 64 bits */ 836 uint64_t start; 837 uint64_t end; 838 int i, num; 839 840 /* 841 * initialize 842 */ 843 pcimemlists[0].addr = pci_lo_limit; 844 pcimemlists[0].size = pci_hi_limit - pci_lo_limit; 845 pcimemlists_used = 1; 846 847 num = dboot_loader_mmap_entries(); 848 /* 849 * Fill in PCI memlists. 850 */ 851 for (i = 0; i < num; ++i) { 852 start = dboot_loader_mmap_get_base(i); 853 end = start + dboot_loader_mmap_get_length(i); 854 855 if (prom_debug) 856 dboot_printf("\ttype: %d %" PRIx64 "..%" 857 PRIx64 "\n", dboot_loader_mmap_get_type(i), 858 start, end); 859 860 /* 861 * page align start and end 862 */ 863 start = (start + page_offset) & ~page_offset; 864 end &= ~page_offset; 865 if (end <= start) 866 continue; 867 868 exclude_from_pci(start, end); 869 } 870 871 /* 872 * Finish off the pcimemlist 873 */ 874 if (prom_debug) { 875 for (i = 0; i < pcimemlists_used; ++i) { 876 dboot_printf("pcimemlist entry 0x%" PRIx64 "..0x%" 877 PRIx64 "\n", pcimemlists[i].addr, 878 pcimemlists[i].addr + pcimemlists[i].size); 879 } 880 } 881 pcimemlists[0].next = 0; 882 pcimemlists[0].prev = 0; 883 for (i = 1; i < pcimemlists_used; ++i) { 884 pcimemlists[i].prev = 885 (native_ptr_t)(uintptr_t)(pcimemlists + i - 1); 886 pcimemlists[i].next = 0; 887 pcimemlists[i - 1].next = 888 (native_ptr_t)(uintptr_t)(pcimemlists + i); 889 } 890 bi->bi_pcimem = (native_ptr_t)(uintptr_t)pcimemlists; 891 DBG(bi->bi_pcimem); 892 } 893 894 #if defined(__xpv) 895 /* 896 * Initialize memory allocator stuff from hypervisor-supplied start info. 897 */ 898 static void 899 init_mem_alloc(void) 900 { 901 int local; /* variables needed to find start region */ 902 paddr_t scratch_start; 903 xen_memory_map_t map; 904 905 DBG_MSG("Entered init_mem_alloc()\n"); 906 907 /* 908 * Free memory follows the stack. There's at least 512KB of scratch 909 * space, rounded up to at least 2Mb alignment. That should be enough 910 * for the page tables we'll need to build. The nucleus memory is 911 * allocated last and will be outside the addressible range. We'll 912 * switch to new page tables before we unpack the kernel 913 */ 914 scratch_start = RNDUP((paddr_t)(uintptr_t)&local, MMU_PAGESIZE); 915 DBG(scratch_start); 916 scratch_end = RNDUP((paddr_t)scratch_start + 512 * 1024, TWO_MEG); 917 DBG(scratch_end); 918 919 /* 920 * For paranoia, leave some space between hypervisor data and ours. 921 * Use 500 instead of 512. 922 */ 923 next_avail_addr = scratch_end - 500 * 1024; 924 DBG(next_avail_addr); 925 926 /* 927 * The domain builder gives us at most 1 module 928 */ 929 DBG(xen_info->mod_len); 930 if (xen_info->mod_len > 0) { 931 DBG(xen_info->mod_start); 932 modules[0].bm_addr = 933 (native_ptr_t)(uintptr_t)xen_info->mod_start; 934 modules[0].bm_size = xen_info->mod_len; 935 bi->bi_module_cnt = 1; 936 bi->bi_modules = (native_ptr_t)(uintptr_t)modules; 937 } else { 938 bi->bi_module_cnt = 0; 939 bi->bi_modules = (native_ptr_t)(uintptr_t)NULL; 940 } 941 DBG(bi->bi_module_cnt); 942 DBG(bi->bi_modules); 943 944 DBG(xen_info->mfn_list); 945 DBG(xen_info->nr_pages); 946 max_mem = (paddr_t)xen_info->nr_pages << MMU_PAGESHIFT; 947 DBG(max_mem); 948 949 /* 950 * Using pseudo-physical addresses, so only 1 memlist element 951 */ 952 memlists[0].addr = 0; 953 DBG(memlists[0].addr); 954 memlists[0].size = max_mem; 955 DBG(memlists[0].size); 956 memlists_used = 1; 957 DBG(memlists_used); 958 959 /* 960 * finish building physinstall list 961 */ 962 sort_physinstall(); 963 964 /* 965 * build bios reserved memlists 966 */ 967 build_rsvdmemlists(); 968 969 if (DOMAIN_IS_INITDOMAIN(xen_info)) { 970 /* 971 * build PCI Memory list 972 */ 973 map.nr_entries = MAXMAPS; 974 /*LINTED: constant in conditional context*/ 975 set_xen_guest_handle(map.buffer, map_buffer); 976 if (HYPERVISOR_memory_op(XENMEM_machine_memory_map, &map) != 0) 977 dboot_panic("getting XENMEM_machine_memory_map failed"); 978 build_pcimemlists(); 979 } 980 } 981 982 #else /* !__xpv */ 983 984 static void 985 dboot_multiboot1_xboot_consinfo(void) 986 { 987 fb->framebuffer = 0; 988 } 989 990 static void 991 dboot_multiboot2_xboot_consinfo(void) 992 { 993 multiboot_tag_framebuffer_t *fbtag; 994 fbtag = dboot_multiboot2_find_tag(mb2_info, 995 MULTIBOOT_TAG_TYPE_FRAMEBUFFER); 996 fb->framebuffer = (uint64_t)(uintptr_t)fbtag; 997 } 998 999 static int 1000 dboot_multiboot_modcount(void) 1001 { 1002 switch (multiboot_version) { 1003 case 1: 1004 return (mb_info->mods_count); 1005 1006 case 2: 1007 return (dboot_multiboot2_modcount(mb2_info)); 1008 1009 default: 1010 dboot_panic("Unknown multiboot version: %d\n", 1011 multiboot_version); 1012 break; 1013 } 1014 return (0); 1015 } 1016 1017 static uint32_t 1018 dboot_multiboot_modstart(int index) 1019 { 1020 switch (multiboot_version) { 1021 case 1: 1022 return (((mb_module_t *)mb_info->mods_addr)[index].mod_start); 1023 1024 case 2: 1025 return (dboot_multiboot2_modstart(mb2_info, index)); 1026 1027 default: 1028 dboot_panic("Unknown multiboot version: %d\n", 1029 multiboot_version); 1030 break; 1031 } 1032 return (0); 1033 } 1034 1035 static uint32_t 1036 dboot_multiboot_modend(int index) 1037 { 1038 switch (multiboot_version) { 1039 case 1: 1040 return (((mb_module_t *)mb_info->mods_addr)[index].mod_end); 1041 1042 case 2: 1043 return (dboot_multiboot2_modend(mb2_info, index)); 1044 1045 default: 1046 dboot_panic("Unknown multiboot version: %d\n", 1047 multiboot_version); 1048 break; 1049 } 1050 return (0); 1051 } 1052 1053 static char * 1054 dboot_multiboot_modcmdline(int index) 1055 { 1056 switch (multiboot_version) { 1057 case 1: 1058 return ((char *)((mb_module_t *) 1059 mb_info->mods_addr)[index].mod_name); 1060 1061 case 2: 1062 return (dboot_multiboot2_modcmdline(mb2_info, index)); 1063 1064 default: 1065 dboot_panic("Unknown multiboot version: %d\n", 1066 multiboot_version); 1067 break; 1068 } 1069 return (0); 1070 } 1071 1072 /* 1073 * Find the modules used by console setup. 1074 * Since we need the console to print early boot messages, the console is set up 1075 * before anything else and therefore we need to pick up the needed modules. 1076 * 1077 * Note, we just will search for and if found, will pass the modules 1078 * to console setup, the proper module list processing will happen later. 1079 * Currently used modules are boot environment and console font. 1080 */ 1081 static void 1082 dboot_find_console_modules(void) 1083 { 1084 int i, modcount; 1085 uint32_t mod_start, mod_end; 1086 char *cmdline; 1087 1088 modcount = dboot_multiboot_modcount(); 1089 bi->bi_module_cnt = 0; 1090 for (i = 0; i < modcount; ++i) { 1091 cmdline = dboot_multiboot_modcmdline(i); 1092 if (cmdline == NULL) 1093 continue; 1094 1095 if (strstr(cmdline, "type=console-font") != NULL) 1096 modules[bi->bi_module_cnt].bm_type = BMT_FONT; 1097 else if (strstr(cmdline, "type=environment") != NULL) 1098 modules[bi->bi_module_cnt].bm_type = BMT_ENV; 1099 else 1100 continue; 1101 1102 mod_start = dboot_multiboot_modstart(i); 1103 mod_end = dboot_multiboot_modend(i); 1104 modules[bi->bi_module_cnt].bm_addr = 1105 (native_ptr_t)(uintptr_t)mod_start; 1106 modules[bi->bi_module_cnt].bm_size = mod_end - mod_start; 1107 modules[bi->bi_module_cnt].bm_name = 1108 (native_ptr_t)(uintptr_t)NULL; 1109 modules[bi->bi_module_cnt].bm_hash = 1110 (native_ptr_t)(uintptr_t)NULL; 1111 bi->bi_module_cnt++; 1112 } 1113 if (bi->bi_module_cnt != 0) 1114 bi->bi_modules = (native_ptr_t)(uintptr_t)modules; 1115 } 1116 1117 static boolean_t 1118 dboot_multiboot_basicmeminfo(uint32_t *lower, uint32_t *upper) 1119 { 1120 boolean_t rv = B_FALSE; 1121 1122 switch (multiboot_version) { 1123 case 1: 1124 if (mb_info->flags & 0x01) { 1125 *lower = mb_info->mem_lower; 1126 *upper = mb_info->mem_upper; 1127 rv = B_TRUE; 1128 } 1129 break; 1130 1131 case 2: 1132 return (dboot_multiboot2_basicmeminfo(mb2_info, lower, upper)); 1133 1134 default: 1135 dboot_panic("Unknown multiboot version: %d\n", 1136 multiboot_version); 1137 break; 1138 } 1139 return (rv); 1140 } 1141 1142 static uint8_t 1143 dboot_a2h(char v) 1144 { 1145 if (v >= 'a') 1146 return (v - 'a' + 0xa); 1147 else if (v >= 'A') 1148 return (v - 'A' + 0xa); 1149 else if (v >= '0') 1150 return (v - '0'); 1151 else 1152 dboot_panic("bad ASCII hex character %c\n", v); 1153 1154 return (0); 1155 } 1156 1157 static void 1158 digest_a2h(const char *ascii, uint8_t *digest) 1159 { 1160 unsigned int i; 1161 1162 for (i = 0; i < SHA1_DIGEST_LENGTH; i++) { 1163 digest[i] = dboot_a2h(ascii[i * 2]) << 4; 1164 digest[i] |= dboot_a2h(ascii[i * 2 + 1]); 1165 } 1166 } 1167 1168 /* 1169 * Generate a SHA-1 hash of the first len bytes of image, and compare it with 1170 * the ASCII-format hash found in the 40-byte buffer at ascii. If they 1171 * match, return 0, otherwise -1. This works only for images smaller than 1172 * 4 GB, which should not be a problem. 1173 */ 1174 static int 1175 check_image_hash(uint_t midx) 1176 { 1177 const char *ascii; 1178 const void *image; 1179 size_t len; 1180 SHA1_CTX ctx; 1181 uint8_t digest[SHA1_DIGEST_LENGTH]; 1182 uint8_t baseline[SHA1_DIGEST_LENGTH]; 1183 unsigned int i; 1184 1185 ascii = (const char *)(uintptr_t)modules[midx].bm_hash; 1186 image = (const void *)(uintptr_t)modules[midx].bm_addr; 1187 len = (size_t)modules[midx].bm_size; 1188 1189 digest_a2h(ascii, baseline); 1190 1191 SHA1Init(&ctx); 1192 SHA1Update(&ctx, image, len); 1193 SHA1Final(digest, &ctx); 1194 1195 for (i = 0; i < SHA1_DIGEST_LENGTH; i++) { 1196 if (digest[i] != baseline[i]) 1197 return (-1); 1198 } 1199 1200 return (0); 1201 } 1202 1203 static const char * 1204 type_to_str(boot_module_type_t type) 1205 { 1206 switch (type) { 1207 case BMT_ROOTFS: 1208 return ("rootfs"); 1209 case BMT_FILE: 1210 return ("file"); 1211 case BMT_HASH: 1212 return ("hash"); 1213 case BMT_ENV: 1214 return ("environment"); 1215 case BMT_FONT: 1216 return ("console-font"); 1217 default: 1218 return ("unknown"); 1219 } 1220 } 1221 1222 static void 1223 check_images(void) 1224 { 1225 uint_t i; 1226 char displayhash[SHA1_ASCII_LENGTH + 1]; 1227 1228 for (i = 0; i < modules_used; i++) { 1229 if (prom_debug) { 1230 dboot_printf("module #%d: name %s type %s " 1231 "addr %lx size %lx\n", 1232 i, (char *)(uintptr_t)modules[i].bm_name, 1233 type_to_str(modules[i].bm_type), 1234 (ulong_t)modules[i].bm_addr, 1235 (ulong_t)modules[i].bm_size); 1236 } 1237 1238 if (modules[i].bm_type == BMT_HASH || 1239 modules[i].bm_hash == (native_ptr_t)(uintptr_t)NULL) { 1240 DBG_MSG("module has no hash; skipping check\n"); 1241 continue; 1242 } 1243 (void) memcpy(displayhash, 1244 (void *)(uintptr_t)modules[i].bm_hash, 1245 SHA1_ASCII_LENGTH); 1246 displayhash[SHA1_ASCII_LENGTH] = '\0'; 1247 if (prom_debug) { 1248 dboot_printf("checking expected hash [%s]: ", 1249 displayhash); 1250 } 1251 1252 if (check_image_hash(i) != 0) 1253 dboot_panic("hash mismatch!\n"); 1254 else 1255 DBG_MSG("OK\n"); 1256 } 1257 } 1258 1259 /* 1260 * Determine the module's starting address, size, name, and type, and fill the 1261 * boot_modules structure. This structure is used by the bop code, except for 1262 * hashes which are checked prior to transferring control to the kernel. 1263 */ 1264 static void 1265 process_module(int midx) 1266 { 1267 uint32_t mod_start = dboot_multiboot_modstart(midx); 1268 uint32_t mod_end = dboot_multiboot_modend(midx); 1269 char *cmdline = dboot_multiboot_modcmdline(midx); 1270 char *p, *q; 1271 1272 check_higher(mod_end); 1273 if (prom_debug) { 1274 dboot_printf("\tmodule #%d: '%s' at 0x%lx, end 0x%lx\n", 1275 midx, cmdline, (ulong_t)mod_start, (ulong_t)mod_end); 1276 } 1277 1278 if (mod_start > mod_end) { 1279 dboot_panic("module #%d: module start address 0x%lx greater " 1280 "than end address 0x%lx", midx, 1281 (ulong_t)mod_start, (ulong_t)mod_end); 1282 } 1283 1284 /* 1285 * A brief note on lengths and sizes: GRUB, for reasons unknown, passes 1286 * the address of the last valid byte in a module plus 1 as mod_end. 1287 * This is of course a bug; the multiboot specification simply states 1288 * that mod_start and mod_end "contain the start and end addresses of 1289 * the boot module itself" which is pretty obviously not what GRUB is 1290 * doing. However, fixing it requires that not only this code be 1291 * changed but also that other code consuming this value and values 1292 * derived from it be fixed, and that the kernel and GRUB must either 1293 * both have the bug or neither. While there are a lot of combinations 1294 * that will work, there are also some that won't, so for simplicity 1295 * we'll just cope with the bug. That means we won't actually hash the 1296 * byte at mod_end, and we will expect that mod_end for the hash file 1297 * itself is one greater than some multiple of 41 (40 bytes of ASCII 1298 * hash plus a newline for each module). We set bm_size to the true 1299 * correct number of bytes in each module, achieving exactly this. 1300 */ 1301 1302 modules[midx].bm_addr = (native_ptr_t)(uintptr_t)mod_start; 1303 modules[midx].bm_size = mod_end - mod_start; 1304 modules[midx].bm_name = (native_ptr_t)(uintptr_t)cmdline; 1305 modules[midx].bm_hash = (native_ptr_t)(uintptr_t)NULL; 1306 modules[midx].bm_type = BMT_FILE; 1307 1308 if (cmdline == NULL) { 1309 modules[midx].bm_name = (native_ptr_t)(uintptr_t)noname; 1310 return; 1311 } 1312 1313 p = cmdline; 1314 modules[midx].bm_name = 1315 (native_ptr_t)(uintptr_t)strsep(&p, " \t\f\n\r"); 1316 1317 while (p != NULL) { 1318 q = strsep(&p, " \t\f\n\r"); 1319 if (strncmp(q, "name=", 5) == 0) { 1320 if (q[5] != '\0' && !isspace(q[5])) { 1321 modules[midx].bm_name = 1322 (native_ptr_t)(uintptr_t)(q + 5); 1323 } 1324 continue; 1325 } 1326 1327 if (strncmp(q, "type=", 5) == 0) { 1328 if (q[5] == '\0' || isspace(q[5])) 1329 continue; 1330 q += 5; 1331 if (strcmp(q, "rootfs") == 0) { 1332 modules[midx].bm_type = BMT_ROOTFS; 1333 } else if (strcmp(q, "hash") == 0) { 1334 modules[midx].bm_type = BMT_HASH; 1335 } else if (strcmp(q, "environment") == 0) { 1336 modules[midx].bm_type = BMT_ENV; 1337 } else if (strcmp(q, "console-font") == 0) { 1338 modules[midx].bm_type = BMT_FONT; 1339 } else if (strcmp(q, "file") != 0) { 1340 dboot_printf("\tmodule #%d: unknown module " 1341 "type '%s'; defaulting to 'file'\n", 1342 midx, q); 1343 } 1344 continue; 1345 } 1346 1347 if (strncmp(q, "hash=", 5) == 0) { 1348 if (q[5] != '\0' && !isspace(q[5])) { 1349 modules[midx].bm_hash = 1350 (native_ptr_t)(uintptr_t)(q + 5); 1351 } 1352 continue; 1353 } 1354 1355 dboot_printf("ignoring unknown option '%s'\n", q); 1356 } 1357 } 1358 1359 /* 1360 * Backward compatibility: if there are exactly one or two modules, both 1361 * of type 'file' and neither with an embedded hash value, we have been 1362 * given the legacy style modules. In this case we need to treat the first 1363 * module as a rootfs and the second as a hash referencing that module. 1364 * Otherwise, even if the configuration is invalid, we assume that the 1365 * operator knows what he's doing or at least isn't being bitten by this 1366 * interface change. 1367 */ 1368 static void 1369 fixup_modules(void) 1370 { 1371 if (modules_used == 0 || modules_used > 2) 1372 return; 1373 1374 if (modules[0].bm_type != BMT_FILE || 1375 modules_used > 1 && modules[1].bm_type != BMT_FILE) { 1376 return; 1377 } 1378 1379 if (modules[0].bm_hash != (native_ptr_t)(uintptr_t)NULL || 1380 modules_used > 1 && 1381 modules[1].bm_hash != (native_ptr_t)(uintptr_t)NULL) { 1382 return; 1383 } 1384 1385 modules[0].bm_type = BMT_ROOTFS; 1386 if (modules_used > 1) { 1387 modules[1].bm_type = BMT_HASH; 1388 modules[1].bm_name = modules[0].bm_name; 1389 } 1390 } 1391 1392 /* 1393 * For modules that do not have assigned hashes but have a separate hash module, 1394 * find the assigned hash module and set the primary module's bm_hash to point 1395 * to the hash data from that module. We will then ignore modules of type 1396 * BMT_HASH from this point forward. 1397 */ 1398 static void 1399 assign_module_hashes(void) 1400 { 1401 uint_t i, j; 1402 1403 for (i = 0; i < modules_used; i++) { 1404 if (modules[i].bm_type == BMT_HASH || 1405 modules[i].bm_hash != (native_ptr_t)(uintptr_t)NULL) { 1406 continue; 1407 } 1408 1409 for (j = 0; j < modules_used; j++) { 1410 if (modules[j].bm_type != BMT_HASH || 1411 strcmp((char *)(uintptr_t)modules[j].bm_name, 1412 (char *)(uintptr_t)modules[i].bm_name) != 0) { 1413 continue; 1414 } 1415 1416 if (modules[j].bm_size < SHA1_ASCII_LENGTH) { 1417 dboot_printf("Short hash module of length " 1418 "0x%lx bytes; ignoring\n", 1419 (ulong_t)modules[j].bm_size); 1420 } else { 1421 modules[i].bm_hash = modules[j].bm_addr; 1422 } 1423 break; 1424 } 1425 } 1426 } 1427 1428 /* 1429 * Walk through the module information finding the last used address. 1430 * The first available address will become the top level page table. 1431 */ 1432 static void 1433 dboot_process_modules(void) 1434 { 1435 int i, modcount; 1436 extern char _end[]; 1437 1438 DBG_MSG("\nFinding Modules\n"); 1439 modcount = dboot_multiboot_modcount(); 1440 if (modcount > MAX_BOOT_MODULES) { 1441 dboot_panic("Too many modules (%d) -- the maximum is %d.", 1442 modcount, MAX_BOOT_MODULES); 1443 } 1444 /* 1445 * search the modules to find the last used address 1446 * we'll build the module list while we're walking through here 1447 */ 1448 check_higher((paddr_t)(uintptr_t)&_end); 1449 for (i = 0; i < modcount; ++i) { 1450 process_module(i); 1451 modules_used++; 1452 } 1453 bi->bi_modules = (native_ptr_t)(uintptr_t)modules; 1454 DBG(bi->bi_modules); 1455 bi->bi_module_cnt = modcount; 1456 DBG(bi->bi_module_cnt); 1457 1458 fixup_modules(); 1459 assign_module_hashes(); 1460 check_images(); 1461 } 1462 1463 /* 1464 * We then build the phys_install memlist from the multiboot information. 1465 */ 1466 static void 1467 dboot_process_mmap(void) 1468 { 1469 uint64_t start; 1470 uint64_t end; 1471 uint64_t page_offset = MMU_PAGEOFFSET; /* needs to be 64 bits */ 1472 uint32_t lower, upper; 1473 int i, mmap_entries; 1474 1475 /* 1476 * Walk through the memory map from multiboot and build our memlist 1477 * structures. Note these will have native format pointers. 1478 */ 1479 DBG_MSG("\nFinding Memory Map\n"); 1480 num_entries = 0; 1481 num_entries_set = B_FALSE; 1482 max_mem = 0; 1483 if ((mmap_entries = dboot_loader_mmap_entries()) > 0) { 1484 for (i = 0; i < mmap_entries; i++) { 1485 uint32_t type = dboot_loader_mmap_get_type(i); 1486 start = dboot_loader_mmap_get_base(i); 1487 end = start + dboot_loader_mmap_get_length(i); 1488 1489 if (prom_debug) 1490 dboot_printf("\ttype: %d %" PRIx64 "..%" 1491 PRIx64 "\n", type, start, end); 1492 1493 /* 1494 * page align start and end 1495 */ 1496 start = (start + page_offset) & ~page_offset; 1497 end &= ~page_offset; 1498 if (end <= start) 1499 continue; 1500 1501 /* 1502 * only type 1 is usable RAM 1503 */ 1504 switch (type) { 1505 case 1: 1506 if (end > max_mem) 1507 max_mem = end; 1508 memlists[memlists_used].addr = start; 1509 memlists[memlists_used].size = end - start; 1510 ++memlists_used; 1511 if (memlists_used > MAX_MEMLIST) 1512 dboot_panic("too many memlists"); 1513 break; 1514 case 2: 1515 rsvdmemlists[rsvdmemlists_used].addr = start; 1516 rsvdmemlists[rsvdmemlists_used].size = 1517 end - start; 1518 ++rsvdmemlists_used; 1519 if (rsvdmemlists_used > MAX_MEMLIST) 1520 dboot_panic("too many rsvdmemlists"); 1521 break; 1522 default: 1523 continue; 1524 } 1525 } 1526 build_pcimemlists(); 1527 } else if (dboot_multiboot_basicmeminfo(&lower, &upper)) { 1528 DBG(lower); 1529 memlists[memlists_used].addr = 0; 1530 memlists[memlists_used].size = lower * 1024; 1531 ++memlists_used; 1532 DBG(upper); 1533 memlists[memlists_used].addr = 1024 * 1024; 1534 memlists[memlists_used].size = upper * 1024; 1535 ++memlists_used; 1536 1537 /* 1538 * Old platform - assume I/O space at the end of memory. 1539 */ 1540 pcimemlists[0].addr = (upper * 1024) + (1024 * 1024); 1541 pcimemlists[0].size = pci_hi_limit - pcimemlists[0].addr; 1542 pcimemlists[0].next = 0; 1543 pcimemlists[0].prev = 0; 1544 bi->bi_pcimem = (native_ptr_t)(uintptr_t)pcimemlists; 1545 DBG(bi->bi_pcimem); 1546 } else { 1547 dboot_panic("No memory info from boot loader!!!"); 1548 } 1549 1550 /* 1551 * finish processing the physinstall list 1552 */ 1553 sort_physinstall(); 1554 1555 /* 1556 * build bios reserved mem lists 1557 */ 1558 build_rsvdmemlists(); 1559 } 1560 1561 /* 1562 * The highest address is used as the starting point for dboot's simple 1563 * memory allocator. 1564 * 1565 * Finding the highest address in case of Multiboot 1 protocol is 1566 * quite painful in the sense that some information provided by 1567 * the multiboot info structure points to BIOS data, and some to RAM. 1568 * 1569 * The module list was processed and checked already by dboot_process_modules(), 1570 * so we will check the command line string and the memory map. 1571 * 1572 * This list of to be checked items is based on our current knowledge of 1573 * allocations made by grub1 and will need to be reviewed if there 1574 * are updates about the information provided by Multiboot 1. 1575 * 1576 * In the case of the Multiboot 2, our life is much simpler, as the MB2 1577 * information tag list is one contiguous chunk of memory. 1578 */ 1579 static paddr_t 1580 dboot_multiboot1_highest_addr(void) 1581 { 1582 paddr_t addr = (paddr_t)(uintptr_t)NULL; 1583 char *cmdl = (char *)mb_info->cmdline; 1584 1585 if (mb_info->flags & MB_INFO_CMDLINE) 1586 addr = ((paddr_t)((uintptr_t)cmdl + strlen(cmdl) + 1)); 1587 1588 if (mb_info->flags & MB_INFO_MEM_MAP) 1589 addr = MAX(addr, 1590 ((paddr_t)(mb_info->mmap_addr + mb_info->mmap_length))); 1591 return (addr); 1592 } 1593 1594 static void 1595 dboot_multiboot_highest_addr(void) 1596 { 1597 paddr_t addr; 1598 1599 switch (multiboot_version) { 1600 case 1: 1601 addr = dboot_multiboot1_highest_addr(); 1602 if (addr != (paddr_t)(uintptr_t)NULL) 1603 check_higher(addr); 1604 break; 1605 case 2: 1606 addr = dboot_multiboot2_highest_addr(mb2_info); 1607 if (addr != (paddr_t)(uintptr_t)NULL) 1608 check_higher(addr); 1609 break; 1610 default: 1611 dboot_panic("Unknown multiboot version: %d\n", 1612 multiboot_version); 1613 break; 1614 } 1615 } 1616 1617 /* 1618 * Walk the boot loader provided information and find the highest free address. 1619 */ 1620 static void 1621 init_mem_alloc(void) 1622 { 1623 DBG_MSG("Entered init_mem_alloc()\n"); 1624 dboot_process_modules(); 1625 dboot_process_mmap(); 1626 dboot_multiboot_highest_addr(); 1627 } 1628 1629 static int 1630 dboot_same_guids(efi_guid_t *g1, efi_guid_t *g2) 1631 { 1632 int i; 1633 1634 if (g1->time_low != g2->time_low) 1635 return (0); 1636 if (g1->time_mid != g2->time_mid) 1637 return (0); 1638 if (g1->time_hi_and_version != g2->time_hi_and_version) 1639 return (0); 1640 if (g1->clock_seq_hi_and_reserved != g2->clock_seq_hi_and_reserved) 1641 return (0); 1642 if (g1->clock_seq_low != g2->clock_seq_low) 1643 return (0); 1644 1645 for (i = 0; i < 6; i++) { 1646 if (g1->node_addr[i] != g2->node_addr[i]) 1647 return (0); 1648 } 1649 return (1); 1650 } 1651 1652 static void 1653 process_efi32(EFI_SYSTEM_TABLE32 *efi) 1654 { 1655 uint32_t entries; 1656 EFI_CONFIGURATION_TABLE32 *config; 1657 int i; 1658 1659 entries = efi->NumberOfTableEntries; 1660 config = (EFI_CONFIGURATION_TABLE32 *)(uintptr_t) 1661 efi->ConfigurationTable; 1662 1663 for (i = 0; i < entries; i++) { 1664 if (dboot_same_guids(&config[i].VendorGuid, &smbios3)) { 1665 bi->bi_smbios = (native_ptr_t)(uintptr_t) 1666 config[i].VendorTable; 1667 } 1668 if (bi->bi_smbios == 0 && 1669 dboot_same_guids(&config[i].VendorGuid, &smbios)) { 1670 bi->bi_smbios = (native_ptr_t)(uintptr_t) 1671 config[i].VendorTable; 1672 } 1673 if (dboot_same_guids(&config[i].VendorGuid, &acpi2)) { 1674 bi->bi_acpi_rsdp = (native_ptr_t)(uintptr_t) 1675 config[i].VendorTable; 1676 } 1677 if (bi->bi_acpi_rsdp == 0 && 1678 dboot_same_guids(&config[i].VendorGuid, &acpi1)) { 1679 bi->bi_acpi_rsdp = (native_ptr_t)(uintptr_t) 1680 config[i].VendorTable; 1681 } 1682 } 1683 } 1684 1685 static void 1686 process_efi64(EFI_SYSTEM_TABLE64 *efi) 1687 { 1688 uint64_t entries; 1689 EFI_CONFIGURATION_TABLE64 *config; 1690 int i; 1691 1692 entries = efi->NumberOfTableEntries; 1693 config = (EFI_CONFIGURATION_TABLE64 *)(uintptr_t) 1694 efi->ConfigurationTable; 1695 1696 for (i = 0; i < entries; i++) { 1697 if (dboot_same_guids(&config[i].VendorGuid, &smbios3)) { 1698 bi->bi_smbios = (native_ptr_t)(uintptr_t) 1699 config[i].VendorTable; 1700 } 1701 if (bi->bi_smbios == 0 && 1702 dboot_same_guids(&config[i].VendorGuid, &smbios)) { 1703 bi->bi_smbios = (native_ptr_t)(uintptr_t) 1704 config[i].VendorTable; 1705 } 1706 /* Prefer acpi v2+ over v1. */ 1707 if (dboot_same_guids(&config[i].VendorGuid, &acpi2)) { 1708 bi->bi_acpi_rsdp = (native_ptr_t)(uintptr_t) 1709 config[i].VendorTable; 1710 } 1711 if (bi->bi_acpi_rsdp == 0 && 1712 dboot_same_guids(&config[i].VendorGuid, &acpi1)) { 1713 bi->bi_acpi_rsdp = (native_ptr_t)(uintptr_t) 1714 config[i].VendorTable; 1715 } 1716 } 1717 } 1718 1719 static void 1720 dboot_multiboot_get_fwtables(void) 1721 { 1722 multiboot_tag_new_acpi_t *nacpitagp; 1723 multiboot_tag_old_acpi_t *oacpitagp; 1724 multiboot_tag_efi64_t *efi64tagp = NULL; 1725 multiboot_tag_efi32_t *efi32tagp = NULL; 1726 1727 /* no fw tables from multiboot 1 */ 1728 if (multiboot_version != 2) 1729 return; 1730 1731 efi64tagp = (multiboot_tag_efi64_t *) 1732 dboot_multiboot2_find_tag(mb2_info, MULTIBOOT_TAG_TYPE_EFI64); 1733 if (efi64tagp != NULL) { 1734 bi->bi_uefi_arch = XBI_UEFI_ARCH_64; 1735 bi->bi_uefi_systab = (native_ptr_t)(uintptr_t) 1736 efi64tagp->mb_pointer; 1737 process_efi64((EFI_SYSTEM_TABLE64 *)(uintptr_t) 1738 efi64tagp->mb_pointer); 1739 } else { 1740 efi32tagp = (multiboot_tag_efi32_t *) 1741 dboot_multiboot2_find_tag(mb2_info, 1742 MULTIBOOT_TAG_TYPE_EFI32); 1743 if (efi32tagp != NULL) { 1744 bi->bi_uefi_arch = XBI_UEFI_ARCH_32; 1745 bi->bi_uefi_systab = (native_ptr_t)(uintptr_t) 1746 efi32tagp->mb_pointer; 1747 process_efi32((EFI_SYSTEM_TABLE32 *)(uintptr_t) 1748 efi32tagp->mb_pointer); 1749 } 1750 } 1751 1752 /* 1753 * The ACPI RSDP can be found by scanning the BIOS memory areas or 1754 * from the EFI system table. The boot loader may pass in the address 1755 * it found the ACPI tables at. 1756 */ 1757 nacpitagp = (multiboot_tag_new_acpi_t *) 1758 dboot_multiboot2_find_tag(mb2_info, 1759 MULTIBOOT_TAG_TYPE_ACPI_NEW); 1760 oacpitagp = (multiboot_tag_old_acpi_t *) 1761 dboot_multiboot2_find_tag(mb2_info, 1762 MULTIBOOT_TAG_TYPE_ACPI_OLD); 1763 1764 if (nacpitagp != NULL) { 1765 bi->bi_acpi_rsdp = (native_ptr_t)(uintptr_t) 1766 &nacpitagp->mb_rsdp[0]; 1767 } else if (oacpitagp != NULL) { 1768 bi->bi_acpi_rsdp = (native_ptr_t)(uintptr_t) 1769 &oacpitagp->mb_rsdp[0]; 1770 } 1771 } 1772 1773 /* print out EFI version string with newline */ 1774 static void 1775 dboot_print_efi_version(uint32_t ver) 1776 { 1777 int rev; 1778 1779 dboot_printf("%d.", EFI_REV_MAJOR(ver)); 1780 1781 rev = EFI_REV_MINOR(ver); 1782 if ((rev % 10) != 0) { 1783 dboot_printf("%d.%d\n", rev / 10, rev % 10); 1784 } else { 1785 dboot_printf("%d\n", rev / 10); 1786 } 1787 } 1788 1789 static void 1790 print_efi32(EFI_SYSTEM_TABLE32 *efi) 1791 { 1792 uint16_t *data; 1793 EFI_CONFIGURATION_TABLE32 *conf; 1794 int i; 1795 1796 dboot_printf("EFI32 signature: %llx\n", 1797 (unsigned long long)efi->Hdr.Signature); 1798 dboot_printf("EFI system version: "); 1799 dboot_print_efi_version(efi->Hdr.Revision); 1800 dboot_printf("EFI system vendor: "); 1801 data = (uint16_t *)(uintptr_t)efi->FirmwareVendor; 1802 for (i = 0; data[i] != 0; i++) 1803 dboot_printf("%c", (char)data[i]); 1804 dboot_printf("\nEFI firmware revision: "); 1805 dboot_print_efi_version(efi->FirmwareRevision); 1806 dboot_printf("EFI system table number of entries: %d\n", 1807 efi->NumberOfTableEntries); 1808 conf = (EFI_CONFIGURATION_TABLE32 *)(uintptr_t) 1809 efi->ConfigurationTable; 1810 for (i = 0; i < (int)efi->NumberOfTableEntries; i++) { 1811 dboot_printf("%d: 0x%x 0x%x 0x%x 0x%x 0x%x", i, 1812 conf[i].VendorGuid.time_low, 1813 conf[i].VendorGuid.time_mid, 1814 conf[i].VendorGuid.time_hi_and_version, 1815 conf[i].VendorGuid.clock_seq_hi_and_reserved, 1816 conf[i].VendorGuid.clock_seq_low); 1817 dboot_printf(" 0x%x 0x%x 0x%x 0x%x 0x%x 0x%x\n", 1818 conf[i].VendorGuid.node_addr[0], 1819 conf[i].VendorGuid.node_addr[1], 1820 conf[i].VendorGuid.node_addr[2], 1821 conf[i].VendorGuid.node_addr[3], 1822 conf[i].VendorGuid.node_addr[4], 1823 conf[i].VendorGuid.node_addr[5]); 1824 } 1825 } 1826 1827 static void 1828 print_efi64(EFI_SYSTEM_TABLE64 *efi) 1829 { 1830 uint16_t *data; 1831 EFI_CONFIGURATION_TABLE64 *conf; 1832 int i; 1833 1834 dboot_printf("EFI64 signature: %llx\n", 1835 (unsigned long long)efi->Hdr.Signature); 1836 dboot_printf("EFI system version: "); 1837 dboot_print_efi_version(efi->Hdr.Revision); 1838 dboot_printf("EFI system vendor: "); 1839 data = (uint16_t *)(uintptr_t)efi->FirmwareVendor; 1840 for (i = 0; data[i] != 0; i++) 1841 dboot_printf("%c", (char)data[i]); 1842 dboot_printf("\nEFI firmware revision: "); 1843 dboot_print_efi_version(efi->FirmwareRevision); 1844 dboot_printf("EFI system table number of entries: %lld\n", 1845 efi->NumberOfTableEntries); 1846 conf = (EFI_CONFIGURATION_TABLE64 *)(uintptr_t) 1847 efi->ConfigurationTable; 1848 for (i = 0; i < (int)efi->NumberOfTableEntries; i++) { 1849 dboot_printf("%d: 0x%x 0x%x 0x%x 0x%x 0x%x", i, 1850 conf[i].VendorGuid.time_low, 1851 conf[i].VendorGuid.time_mid, 1852 conf[i].VendorGuid.time_hi_and_version, 1853 conf[i].VendorGuid.clock_seq_hi_and_reserved, 1854 conf[i].VendorGuid.clock_seq_low); 1855 dboot_printf(" 0x%x 0x%x 0x%x 0x%x 0x%x 0x%x\n", 1856 conf[i].VendorGuid.node_addr[0], 1857 conf[i].VendorGuid.node_addr[1], 1858 conf[i].VendorGuid.node_addr[2], 1859 conf[i].VendorGuid.node_addr[3], 1860 conf[i].VendorGuid.node_addr[4], 1861 conf[i].VendorGuid.node_addr[5]); 1862 } 1863 } 1864 #endif /* !__xpv */ 1865 1866 /* 1867 * Simple memory allocator, allocates aligned physical memory. 1868 * Note that startup_kernel() only allocates memory, never frees. 1869 * Memory usage just grows in an upward direction. 1870 */ 1871 static void * 1872 do_mem_alloc(uint32_t size, uint32_t align) 1873 { 1874 uint_t i; 1875 uint64_t best; 1876 uint64_t start; 1877 uint64_t end; 1878 1879 /* 1880 * make sure size is a multiple of pagesize 1881 */ 1882 size = RNDUP(size, MMU_PAGESIZE); 1883 next_avail_addr = RNDUP(next_avail_addr, align); 1884 1885 /* 1886 * XXPV fixme joe 1887 * 1888 * a really large bootarchive that causes you to run out of memory 1889 * may cause this to blow up 1890 */ 1891 /* LINTED E_UNEXPECTED_UINT_PROMOTION */ 1892 best = (uint64_t)-size; 1893 for (i = 0; i < memlists_used; ++i) { 1894 start = memlists[i].addr; 1895 #if defined(__xpv) 1896 start += mfn_base; 1897 #endif 1898 end = start + memlists[i].size; 1899 1900 /* 1901 * did we find the desired address? 1902 */ 1903 if (start <= next_avail_addr && next_avail_addr + size <= end) { 1904 best = next_avail_addr; 1905 goto done; 1906 } 1907 1908 /* 1909 * if not is this address the best so far? 1910 */ 1911 if (start > next_avail_addr && start < best && 1912 RNDUP(start, align) + size <= end) 1913 best = RNDUP(start, align); 1914 } 1915 1916 /* 1917 * We didn't find exactly the address we wanted, due to going off the 1918 * end of a memory region. Return the best found memory address. 1919 */ 1920 done: 1921 next_avail_addr = best + size; 1922 #if defined(__xpv) 1923 if (next_avail_addr > scratch_end) 1924 dboot_panic("Out of mem next_avail: 0x%lx, scratch_end: " 1925 "0x%lx", (ulong_t)next_avail_addr, 1926 (ulong_t)scratch_end); 1927 #endif 1928 (void) memset((void *)(uintptr_t)best, 0, size); 1929 return ((void *)(uintptr_t)best); 1930 } 1931 1932 void * 1933 mem_alloc(uint32_t size) 1934 { 1935 return (do_mem_alloc(size, MMU_PAGESIZE)); 1936 } 1937 1938 1939 /* 1940 * Build page tables to map all of memory used so far as well as the kernel. 1941 */ 1942 static void 1943 build_page_tables(void) 1944 { 1945 uint32_t psize; 1946 uint32_t level; 1947 uint32_t off; 1948 uint64_t start; 1949 #if !defined(__xpv) 1950 uint32_t i; 1951 uint64_t end; 1952 #endif /* __xpv */ 1953 1954 /* 1955 * If we're on metal, we need to create the top level pagetable. 1956 */ 1957 #if defined(__xpv) 1958 top_page_table = (paddr_t)(uintptr_t)xen_info->pt_base; 1959 #else /* __xpv */ 1960 top_page_table = (paddr_t)(uintptr_t)mem_alloc(MMU_PAGESIZE); 1961 #endif /* __xpv */ 1962 DBG((uintptr_t)top_page_table); 1963 1964 /* 1965 * Determine if we'll use large mappings for kernel, then map it. 1966 */ 1967 if (largepage_support) { 1968 psize = lpagesize; 1969 level = 1; 1970 } else { 1971 psize = MMU_PAGESIZE; 1972 level = 0; 1973 } 1974 1975 DBG_MSG("Mapping kernel\n"); 1976 DBG(ktext_phys); 1977 DBG(target_kernel_text); 1978 DBG(ksize); 1979 DBG(psize); 1980 for (off = 0; off < ksize; off += psize) 1981 map_pa_at_va(ktext_phys + off, target_kernel_text + off, level); 1982 1983 /* 1984 * The kernel will need a 1 page window to work with page tables 1985 */ 1986 bi->bi_pt_window = (native_ptr_t)(uintptr_t)mem_alloc(MMU_PAGESIZE); 1987 DBG(bi->bi_pt_window); 1988 bi->bi_pte_to_pt_window = 1989 (native_ptr_t)(uintptr_t)find_pte(bi->bi_pt_window, NULL, 0, 0); 1990 DBG(bi->bi_pte_to_pt_window); 1991 1992 #if defined(__xpv) 1993 if (!DOMAIN_IS_INITDOMAIN(xen_info)) { 1994 /* If this is a domU we're done. */ 1995 DBG_MSG("\nPage tables constructed\n"); 1996 return; 1997 } 1998 #endif /* __xpv */ 1999 2000 /* 2001 * We need 1:1 mappings for the lower 1M of memory to access 2002 * BIOS tables used by a couple of drivers during boot. 2003 * 2004 * The following code works because our simple memory allocator 2005 * only grows usage in an upwards direction. 2006 * 2007 * Note that by this point in boot some mappings for low memory 2008 * may already exist because we've already accessed device in low 2009 * memory. (Specifically the video frame buffer and keyboard 2010 * status ports.) If we're booting on raw hardware then GRUB 2011 * created these mappings for us. If we're booting under a 2012 * hypervisor then we went ahead and remapped these devices into 2013 * memory allocated within dboot itself. 2014 */ 2015 if (map_debug) 2016 dboot_printf("1:1 map pa=0..1Meg\n"); 2017 for (start = 0; start < 1024 * 1024; start += MMU_PAGESIZE) { 2018 #if defined(__xpv) 2019 map_ma_at_va(start, start, 0); 2020 #else /* __xpv */ 2021 map_pa_at_va(start, start, 0); 2022 #endif /* __xpv */ 2023 } 2024 2025 #if !defined(__xpv) 2026 2027 for (i = 0; i < memlists_used; ++i) { 2028 start = memlists[i].addr; 2029 end = start + memlists[i].size; 2030 2031 if (map_debug) 2032 dboot_printf("1:1 map pa=%" PRIx64 "..%" PRIx64 "\n", 2033 start, end); 2034 while (start < end && start < next_avail_addr) { 2035 map_pa_at_va(start, start, 0); 2036 start += MMU_PAGESIZE; 2037 } 2038 if (start >= next_avail_addr) 2039 break; 2040 } 2041 2042 /* 2043 * Map framebuffer memory as PT_NOCACHE as this is memory from a 2044 * device and therefore must not be cached. 2045 */ 2046 if (bi->bi_framebuffer != NULL && fb->framebuffer != 0) { 2047 multiboot_tag_framebuffer_t *fb_tagp; 2048 fb_tagp = (multiboot_tag_framebuffer_t *)(uintptr_t) 2049 fb->framebuffer; 2050 2051 start = fb_tagp->framebuffer_common.framebuffer_addr; 2052 end = start + fb_tagp->framebuffer_common.framebuffer_height * 2053 fb_tagp->framebuffer_common.framebuffer_pitch; 2054 2055 if (map_debug) 2056 dboot_printf("FB 1:1 map pa=%" PRIx64 "..%" PRIx64 "\n", 2057 start, end); 2058 pte_bits |= PT_NOCACHE; 2059 if (PAT_support != 0) 2060 pte_bits |= PT_PAT_4K; 2061 2062 while (start < end) { 2063 map_pa_at_va(start, start, 0); 2064 start += MMU_PAGESIZE; 2065 } 2066 pte_bits &= ~PT_NOCACHE; 2067 if (PAT_support != 0) 2068 pte_bits &= ~PT_PAT_4K; 2069 } 2070 #endif /* !__xpv */ 2071 2072 DBG_MSG("\nPage tables constructed\n"); 2073 } 2074 2075 #define NO_MULTIBOOT \ 2076 "multiboot is no longer used to boot the Solaris Operating System.\n\ 2077 The grub entry should be changed to:\n\ 2078 kernel$ /platform/i86pc/kernel/$ISADIR/unix\n\ 2079 module$ /platform/i86pc/$ISADIR/boot_archive\n\ 2080 See http://illumos.org/msg/SUNOS-8000-AK for details.\n" 2081 2082 static void 2083 dboot_init_xboot_consinfo(void) 2084 { 2085 bi = &boot_info; 2086 2087 #if !defined(__xpv) 2088 fb = &framebuffer; 2089 bi->bi_framebuffer = (native_ptr_t)(uintptr_t)fb; 2090 2091 switch (multiboot_version) { 2092 case 1: 2093 dboot_multiboot1_xboot_consinfo(); 2094 break; 2095 case 2: 2096 dboot_multiboot2_xboot_consinfo(); 2097 break; 2098 default: 2099 dboot_panic("Unknown multiboot version: %d\n", 2100 multiboot_version); 2101 break; 2102 } 2103 dboot_find_console_modules(); 2104 #endif 2105 } 2106 2107 /* 2108 * Set up basic data from the boot loader. 2109 * The load_addr is part of AOUT kludge setup in dboot_grub.s, to support 2110 * 32-bit dboot code setup used to set up and start 64-bit kernel. 2111 * AOUT kludge does allow 32-bit boot loader, such as grub1, to load and 2112 * start 64-bit illumos kernel. 2113 */ 2114 static void 2115 dboot_loader_init(void) 2116 { 2117 #if !defined(__xpv) 2118 mb_info = NULL; 2119 mb2_info = NULL; 2120 2121 switch (mb_magic) { 2122 case MB_BOOTLOADER_MAGIC: 2123 multiboot_version = 1; 2124 mb_info = (multiboot_info_t *)(uintptr_t)mb_addr; 2125 #if defined(_BOOT_TARGET_amd64) 2126 load_addr = mb_header.load_addr; 2127 #endif 2128 break; 2129 2130 case MULTIBOOT2_BOOTLOADER_MAGIC: 2131 multiboot_version = 2; 2132 mb2_info = (multiboot2_info_header_t *)(uintptr_t)mb_addr; 2133 mb2_mmap_tagp = dboot_multiboot2_get_mmap_tagp(mb2_info); 2134 #if defined(_BOOT_TARGET_amd64) 2135 load_addr = mb2_load_addr; 2136 #endif 2137 break; 2138 2139 default: 2140 dboot_panic("Unknown bootloader magic: 0x%x\n", mb_magic); 2141 break; 2142 } 2143 #endif /* !defined(__xpv) */ 2144 } 2145 2146 /* Extract the kernel command line from [multi]boot information. */ 2147 static char * 2148 dboot_loader_cmdline(void) 2149 { 2150 char *line = NULL; 2151 2152 #if defined(__xpv) 2153 line = (char *)xen_info->cmd_line; 2154 #else /* __xpv */ 2155 2156 switch (multiboot_version) { 2157 case 1: 2158 if (mb_info->flags & MB_INFO_CMDLINE) 2159 line = (char *)mb_info->cmdline; 2160 break; 2161 2162 case 2: 2163 line = dboot_multiboot2_cmdline(mb2_info); 2164 break; 2165 2166 default: 2167 dboot_panic("Unknown multiboot version: %d\n", 2168 multiboot_version); 2169 break; 2170 } 2171 2172 #endif /* __xpv */ 2173 2174 /* 2175 * Make sure we have valid pointer so the string operations 2176 * will not crash us. 2177 */ 2178 if (line == NULL) 2179 line = ""; 2180 2181 return (line); 2182 } 2183 2184 static char * 2185 dboot_loader_name(void) 2186 { 2187 #if defined(__xpv) 2188 return (NULL); 2189 #else /* __xpv */ 2190 multiboot_tag_string_t *tag; 2191 2192 switch (multiboot_version) { 2193 case 1: 2194 return ((char *)mb_info->boot_loader_name); 2195 2196 case 2: 2197 tag = dboot_multiboot2_find_tag(mb2_info, 2198 MULTIBOOT_TAG_TYPE_BOOT_LOADER_NAME); 2199 return (tag->mb_string); 2200 default: 2201 dboot_panic("Unknown multiboot version: %d\n", 2202 multiboot_version); 2203 break; 2204 } 2205 2206 return (NULL); 2207 #endif /* __xpv */ 2208 } 2209 2210 /* 2211 * startup_kernel has a pretty simple job. It builds pagetables which reflect 2212 * 1:1 mappings for all memory in use. It then also adds mappings for 2213 * the kernel nucleus at virtual address of target_kernel_text using large page 2214 * mappings. The page table pages are also accessible at 1:1 mapped 2215 * virtual addresses. 2216 */ 2217 /*ARGSUSED*/ 2218 void 2219 startup_kernel(void) 2220 { 2221 char *cmdline; 2222 char *bootloader; 2223 #if defined(__xpv) 2224 physdev_set_iopl_t set_iopl; 2225 #endif /* __xpv */ 2226 2227 if (dboot_debug == 1) 2228 bcons_init(NULL); /* Set very early console to ttya. */ 2229 dboot_loader_init(); 2230 /* 2231 * At this point we are executing in a 32 bit real mode. 2232 */ 2233 2234 bootloader = dboot_loader_name(); 2235 cmdline = dboot_loader_cmdline(); 2236 2237 #if defined(__xpv) 2238 /* 2239 * For dom0, before we initialize the console subsystem we'll 2240 * need to enable io operations, so set I/O priveldge level to 1. 2241 */ 2242 if (DOMAIN_IS_INITDOMAIN(xen_info)) { 2243 set_iopl.iopl = 1; 2244 (void) HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl); 2245 } 2246 #endif /* __xpv */ 2247 2248 dboot_init_xboot_consinfo(); 2249 bi->bi_cmdline = (native_ptr_t)(uintptr_t)cmdline; 2250 bcons_init(bi); /* Now we can set the real console. */ 2251 2252 prom_debug = (find_boot_prop("prom_debug") != NULL); 2253 map_debug = (find_boot_prop("map_debug") != NULL); 2254 2255 #if !defined(__xpv) 2256 dboot_multiboot_get_fwtables(); 2257 #endif 2258 DBG_MSG("\n\nillumos prekernel set: "); 2259 DBG_MSG(cmdline); 2260 DBG_MSG("\n"); 2261 2262 if (bootloader != NULL && prom_debug) { 2263 dboot_printf("Kernel loaded by: %s\n", bootloader); 2264 #if !defined(__xpv) 2265 dboot_printf("Using multiboot %d boot protocol.\n", 2266 multiboot_version); 2267 #endif 2268 } 2269 2270 if (strstr(cmdline, "multiboot") != NULL) { 2271 dboot_panic(NO_MULTIBOOT); 2272 } 2273 2274 DBG((uintptr_t)bi); 2275 #if !defined(__xpv) 2276 DBG((uintptr_t)mb_info); 2277 DBG((uintptr_t)mb2_info); 2278 if (mb2_info != NULL) 2279 DBG(mb2_info->mbi_total_size); 2280 DBG(bi->bi_acpi_rsdp); 2281 DBG(bi->bi_smbios); 2282 DBG(bi->bi_uefi_arch); 2283 DBG(bi->bi_uefi_systab); 2284 2285 if (bi->bi_uefi_systab && prom_debug) { 2286 if (bi->bi_uefi_arch == XBI_UEFI_ARCH_64) { 2287 print_efi64((EFI_SYSTEM_TABLE64 *)(uintptr_t) 2288 bi->bi_uefi_systab); 2289 } else { 2290 print_efi32((EFI_SYSTEM_TABLE32 *)(uintptr_t) 2291 bi->bi_uefi_systab); 2292 } 2293 } 2294 #endif 2295 2296 /* 2297 * Need correct target_kernel_text value 2298 */ 2299 #if defined(_BOOT_TARGET_amd64) 2300 target_kernel_text = KERNEL_TEXT_amd64; 2301 #elif defined(__xpv) 2302 target_kernel_text = KERNEL_TEXT_i386_xpv; 2303 #else 2304 target_kernel_text = KERNEL_TEXT_i386; 2305 #endif 2306 DBG(target_kernel_text); 2307 2308 #if defined(__xpv) 2309 2310 /* 2311 * XXPV Derive this stuff from CPUID / what the hypervisor has enabled 2312 */ 2313 2314 #if defined(_BOOT_TARGET_amd64) 2315 /* 2316 * 64-bit hypervisor. 2317 */ 2318 amd64_support = 1; 2319 pae_support = 1; 2320 2321 #else /* _BOOT_TARGET_amd64 */ 2322 2323 /* 2324 * See if we are running on a PAE Hypervisor 2325 */ 2326 { 2327 xen_capabilities_info_t caps; 2328 2329 if (HYPERVISOR_xen_version(XENVER_capabilities, &caps) != 0) 2330 dboot_panic("HYPERVISOR_xen_version(caps) failed"); 2331 caps[sizeof (caps) - 1] = 0; 2332 if (prom_debug) 2333 dboot_printf("xen capabilities %s\n", caps); 2334 if (strstr(caps, "x86_32p") != NULL) 2335 pae_support = 1; 2336 } 2337 2338 #endif /* _BOOT_TARGET_amd64 */ 2339 { 2340 xen_platform_parameters_t p; 2341 2342 if (HYPERVISOR_xen_version(XENVER_platform_parameters, &p) != 0) 2343 dboot_panic("HYPERVISOR_xen_version(parms) failed"); 2344 DBG(p.virt_start); 2345 mfn_to_pfn_mapping = (pfn_t *)(xen_virt_start = p.virt_start); 2346 } 2347 2348 /* 2349 * The hypervisor loads stuff starting at 1Gig 2350 */ 2351 mfn_base = ONE_GIG; 2352 DBG(mfn_base); 2353 2354 /* 2355 * enable writable page table mode for the hypervisor 2356 */ 2357 if (HYPERVISOR_vm_assist(VMASST_CMD_enable, 2358 VMASST_TYPE_writable_pagetables) < 0) 2359 dboot_panic("HYPERVISOR_vm_assist(writable_pagetables) failed"); 2360 2361 /* 2362 * check for NX support 2363 */ 2364 if (pae_support) { 2365 uint32_t eax = 0x80000000; 2366 uint32_t edx = get_cpuid_edx(&eax); 2367 2368 if (eax >= 0x80000001) { 2369 eax = 0x80000001; 2370 edx = get_cpuid_edx(&eax); 2371 if (edx & CPUID_AMD_EDX_NX) 2372 NX_support = 1; 2373 } 2374 } 2375 2376 /* 2377 * check for PAT support 2378 */ 2379 { 2380 uint32_t eax = 1; 2381 uint32_t edx = get_cpuid_edx(&eax); 2382 2383 if (edx & CPUID_INTC_EDX_PAT) 2384 PAT_support = 1; 2385 } 2386 #if !defined(_BOOT_TARGET_amd64) 2387 2388 /* 2389 * The 32-bit hypervisor uses segmentation to protect itself from 2390 * guests. This means when a guest attempts to install a flat 4GB 2391 * code or data descriptor the 32-bit hypervisor will protect itself 2392 * by silently shrinking the segment such that if the guest attempts 2393 * any access where the hypervisor lives a #gp fault is generated. 2394 * The problem is that some applications expect a full 4GB flat 2395 * segment for their current thread pointer and will use negative 2396 * offset segment wrap around to access data. TLS support in linux 2397 * brand is one example of this. 2398 * 2399 * The 32-bit hypervisor can catch the #gp fault in these cases 2400 * and emulate the access without passing the #gp fault to the guest 2401 * but only if VMASST_TYPE_4gb_segments is explicitly turned on. 2402 * Seems like this should have been the default. 2403 * Either way, we want the hypervisor -- and not Solaris -- to deal 2404 * to deal with emulating these accesses. 2405 */ 2406 if (HYPERVISOR_vm_assist(VMASST_CMD_enable, 2407 VMASST_TYPE_4gb_segments) < 0) 2408 dboot_panic("HYPERVISOR_vm_assist(4gb_segments) failed"); 2409 #endif /* !_BOOT_TARGET_amd64 */ 2410 2411 #else /* __xpv */ 2412 2413 /* 2414 * use cpuid to enable MMU features 2415 */ 2416 if (have_cpuid()) { 2417 uint32_t eax, edx; 2418 2419 eax = 1; 2420 edx = get_cpuid_edx(&eax); 2421 if (edx & CPUID_INTC_EDX_PSE) 2422 largepage_support = 1; 2423 if (edx & CPUID_INTC_EDX_PGE) 2424 pge_support = 1; 2425 if (edx & CPUID_INTC_EDX_PAE) 2426 pae_support = 1; 2427 if (edx & CPUID_INTC_EDX_PAT) 2428 PAT_support = 1; 2429 2430 eax = 0x80000000; 2431 edx = get_cpuid_edx(&eax); 2432 if (eax >= 0x80000001) { 2433 eax = 0x80000001; 2434 edx = get_cpuid_edx(&eax); 2435 if (edx & CPUID_AMD_EDX_LM) 2436 amd64_support = 1; 2437 if (edx & CPUID_AMD_EDX_NX) 2438 NX_support = 1; 2439 } 2440 } else { 2441 dboot_printf("cpuid not supported\n"); 2442 } 2443 #endif /* __xpv */ 2444 2445 2446 #if defined(_BOOT_TARGET_amd64) 2447 if (amd64_support == 0) 2448 dboot_panic("long mode not supported, rebooting"); 2449 else if (pae_support == 0) 2450 dboot_panic("long mode, but no PAE; rebooting"); 2451 #else 2452 /* 2453 * Allow the command line to over-ride use of PAE for 32 bit. 2454 */ 2455 if (strstr(cmdline, "disablePAE=true") != NULL) { 2456 pae_support = 0; 2457 NX_support = 0; 2458 amd64_support = 0; 2459 } 2460 #endif 2461 2462 /* 2463 * initialize the simple memory allocator 2464 */ 2465 init_mem_alloc(); 2466 2467 #if !defined(__xpv) && !defined(_BOOT_TARGET_amd64) 2468 /* 2469 * disable PAE on 32 bit h/w w/o NX and < 4Gig of memory 2470 */ 2471 if (max_mem < FOUR_GIG && NX_support == 0) 2472 pae_support = 0; 2473 #endif 2474 2475 /* 2476 * configure mmu information 2477 */ 2478 if (pae_support) { 2479 shift_amt = shift_amt_pae; 2480 ptes_per_table = 512; 2481 pte_size = 8; 2482 lpagesize = TWO_MEG; 2483 #if defined(_BOOT_TARGET_amd64) 2484 top_level = 3; 2485 #else 2486 top_level = 2; 2487 #endif 2488 } else { 2489 pae_support = 0; 2490 NX_support = 0; 2491 shift_amt = shift_amt_nopae; 2492 ptes_per_table = 1024; 2493 pte_size = 4; 2494 lpagesize = FOUR_MEG; 2495 top_level = 1; 2496 } 2497 2498 DBG(PAT_support); 2499 DBG(pge_support); 2500 DBG(NX_support); 2501 DBG(largepage_support); 2502 DBG(amd64_support); 2503 DBG(top_level); 2504 DBG(pte_size); 2505 DBG(ptes_per_table); 2506 DBG(lpagesize); 2507 2508 #if defined(__xpv) 2509 ktext_phys = ONE_GIG; /* from UNIX Mapfile */ 2510 #else 2511 ktext_phys = FOUR_MEG; /* from UNIX Mapfile */ 2512 #endif 2513 2514 #if !defined(__xpv) && defined(_BOOT_TARGET_amd64) 2515 /* 2516 * For grub, copy kernel bits from the ELF64 file to final place. 2517 */ 2518 DBG_MSG("\nAllocating nucleus pages.\n"); 2519 ktext_phys = (uintptr_t)do_mem_alloc(ksize, FOUR_MEG); 2520 2521 if (ktext_phys == 0) 2522 dboot_panic("failed to allocate aligned kernel memory"); 2523 DBG(load_addr); 2524 if (dboot_elfload64(load_addr) != 0) 2525 dboot_panic("failed to parse kernel ELF image, rebooting"); 2526 #endif 2527 2528 DBG(ktext_phys); 2529 2530 /* 2531 * Allocate page tables. 2532 */ 2533 build_page_tables(); 2534 2535 /* 2536 * return to assembly code to switch to running kernel 2537 */ 2538 entry_addr_low = (uint32_t)target_kernel_text; 2539 DBG(entry_addr_low); 2540 bi->bi_use_largepage = largepage_support; 2541 bi->bi_use_pae = pae_support; 2542 bi->bi_use_pge = pge_support; 2543 bi->bi_use_nx = NX_support; 2544 2545 #if defined(__xpv) 2546 2547 bi->bi_next_paddr = next_avail_addr - mfn_base; 2548 DBG(bi->bi_next_paddr); 2549 bi->bi_next_vaddr = (native_ptr_t)(uintptr_t)next_avail_addr; 2550 DBG(bi->bi_next_vaddr); 2551 2552 /* 2553 * unmap unused pages in start area to make them available for DMA 2554 */ 2555 while (next_avail_addr < scratch_end) { 2556 (void) HYPERVISOR_update_va_mapping(next_avail_addr, 2557 0, UVMF_INVLPG | UVMF_LOCAL); 2558 next_avail_addr += MMU_PAGESIZE; 2559 } 2560 2561 bi->bi_xen_start_info = (native_ptr_t)(uintptr_t)xen_info; 2562 DBG((uintptr_t)HYPERVISOR_shared_info); 2563 bi->bi_shared_info = (native_ptr_t)HYPERVISOR_shared_info; 2564 bi->bi_top_page_table = (uintptr_t)top_page_table - mfn_base; 2565 2566 #else /* __xpv */ 2567 2568 bi->bi_next_paddr = next_avail_addr; 2569 DBG(bi->bi_next_paddr); 2570 bi->bi_next_vaddr = (native_ptr_t)(uintptr_t)next_avail_addr; 2571 DBG(bi->bi_next_vaddr); 2572 bi->bi_mb_version = multiboot_version; 2573 2574 switch (multiboot_version) { 2575 case 1: 2576 bi->bi_mb_info = (native_ptr_t)(uintptr_t)mb_info; 2577 break; 2578 case 2: 2579 bi->bi_mb_info = (native_ptr_t)(uintptr_t)mb2_info; 2580 break; 2581 default: 2582 dboot_panic("Unknown multiboot version: %d\n", 2583 multiboot_version); 2584 break; 2585 } 2586 bi->bi_top_page_table = (uintptr_t)top_page_table; 2587 2588 #endif /* __xpv */ 2589 2590 bi->bi_kseg_size = FOUR_MEG; 2591 DBG(bi->bi_kseg_size); 2592 2593 #ifndef __xpv 2594 if (map_debug) 2595 dump_tables(); 2596 #endif 2597 2598 DBG_MSG("\n\n*** DBOOT DONE -- back to asm to jump to kernel\n\n"); 2599 2600 #ifndef __xpv 2601 /* Update boot info with FB data */ 2602 fb->cursor.origin.x = fb_info.cursor.origin.x; 2603 fb->cursor.origin.y = fb_info.cursor.origin.y; 2604 fb->cursor.pos.x = fb_info.cursor.pos.x; 2605 fb->cursor.pos.y = fb_info.cursor.pos.y; 2606 fb->cursor.visible = fb_info.cursor.visible; 2607 #endif 2608 } 2609