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 * Copyright 2007 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 24 */ 25 26 #pragma ident "%Z%%M% %I% %E% SMI" 27 28 #include <sys/types.h> 29 #include <sys/t_lock.h> 30 #include <sys/param.h> 31 #include <sys/sysmacros.h> 32 #include <sys/signal.h> 33 #include <sys/systm.h> 34 #include <sys/user.h> 35 #include <sys/mman.h> 36 #include <sys/vm.h> 37 #include <sys/conf.h> 38 #include <sys/avintr.h> 39 #include <sys/autoconf.h> 40 #include <sys/disp.h> 41 #include <sys/class.h> 42 #include <sys/bitmap.h> 43 44 #include <sys/privregs.h> 45 46 #include <sys/proc.h> 47 #include <sys/buf.h> 48 #include <sys/kmem.h> 49 #include <sys/mem.h> 50 #include <sys/kstat.h> 51 52 #include <sys/reboot.h> 53 54 #include <sys/cred.h> 55 #include <sys/vnode.h> 56 #include <sys/file.h> 57 58 #include <sys/procfs.h> 59 60 #include <sys/vfs.h> 61 #include <sys/cmn_err.h> 62 #include <sys/utsname.h> 63 #include <sys/debug.h> 64 #include <sys/kdi.h> 65 66 #include <sys/dumphdr.h> 67 #include <sys/bootconf.h> 68 #include <sys/varargs.h> 69 #include <sys/promif.h> 70 #include <sys/modctl.h> 71 72 #include <sys/sunddi.h> 73 #include <sys/sunndi.h> 74 #include <sys/ndi_impldefs.h> 75 #include <sys/ddidmareq.h> 76 #include <sys/psw.h> 77 #include <sys/regset.h> 78 #include <sys/clock.h> 79 #include <sys/pte.h> 80 #include <sys/tss.h> 81 #include <sys/stack.h> 82 #include <sys/trap.h> 83 #include <sys/fp.h> 84 #include <vm/kboot_mmu.h> 85 #include <vm/anon.h> 86 #include <vm/as.h> 87 #include <vm/page.h> 88 #include <vm/seg.h> 89 #include <vm/seg_dev.h> 90 #include <vm/seg_kmem.h> 91 #include <vm/seg_kpm.h> 92 #include <vm/seg_map.h> 93 #include <vm/seg_vn.h> 94 #include <vm/seg_kp.h> 95 #include <sys/memnode.h> 96 #include <vm/vm_dep.h> 97 #include <sys/thread.h> 98 #include <sys/sysconf.h> 99 #include <sys/vm_machparam.h> 100 #include <sys/archsystm.h> 101 #include <sys/machsystm.h> 102 #include <vm/hat.h> 103 #include <vm/hat_i86.h> 104 #include <sys/pmem.h> 105 #include <sys/smp_impldefs.h> 106 #include <sys/x86_archext.h> 107 #include <sys/segments.h> 108 #include <sys/clconf.h> 109 #include <sys/kobj.h> 110 #include <sys/kobj_lex.h> 111 #include <sys/cpc_impl.h> 112 #include <sys/x86_archext.h> 113 #include <sys/cpu_module.h> 114 #include <sys/smbios.h> 115 #include <sys/debug_info.h> 116 #include <sys/bootinfo.h> 117 #include <sys/ddi_timer.h> 118 119 #ifdef __xpv 120 121 #include <sys/hypervisor.h> 122 #include <sys/xen_mmu.h> 123 #include <sys/evtchn_impl.h> 124 #include <sys/gnttab.h> 125 #include <sys/xpv_panic.h> 126 #include <xen/sys/xenbus_comms.h> 127 #include <xen/public/physdev.h> 128 129 extern void xen_late_startup(void); 130 131 struct xen_evt_data cpu0_evt_data; 132 133 #endif /* __xpv */ 134 135 extern void progressbar_init(void); 136 extern void progressbar_start(void); 137 extern void brand_init(void); 138 139 extern int size_pse_array(pgcnt_t, int); 140 141 /* 142 * XXX make declaration below "static" when drivers no longer use this 143 * interface. 144 */ 145 extern caddr_t p0_va; /* Virtual address for accessing physical page 0 */ 146 147 /* 148 * segkp 149 */ 150 extern int segkp_fromheap; 151 152 static void kvm_init(void); 153 static void startup_init(void); 154 static void startup_memlist(void); 155 static void startup_kmem(void); 156 static void startup_modules(void); 157 static void startup_vm(void); 158 static void startup_end(void); 159 static void layout_kernel_va(void); 160 161 /* 162 * Declare these as initialized data so we can patch them. 163 */ 164 #ifdef __i386 165 166 /* 167 * Due to virtual address space limitations running in 32 bit mode, restrict 168 * the amount of physical memory configured to a max of PHYSMEM pages (16g). 169 * 170 * If the physical max memory size of 64g were allowed to be configured, the 171 * size of user virtual address space will be less than 1g. A limited user 172 * address space greatly reduces the range of applications that can run. 173 * 174 * If more physical memory than PHYSMEM is required, users should preferably 175 * run in 64 bit mode which has far looser virtual address space limitations. 176 * 177 * If 64 bit mode is not available (as in IA32) and/or more physical memory 178 * than PHYSMEM is required in 32 bit mode, physmem can be set to the desired 179 * value or to 0 (to configure all available memory) via eeprom(1M). kernelbase 180 * should also be carefully tuned to balance out the need of the user 181 * application while minimizing the risk of kernel heap exhaustion due to 182 * kernelbase being set too high. 183 */ 184 #define PHYSMEM 0x400000 185 186 #else /* __amd64 */ 187 188 /* 189 * For now we can handle memory with physical addresses up to about 190 * 64 Terabytes. This keeps the kernel above the VA hole, leaving roughly 191 * half the VA space for seg_kpm. When systems get bigger than 64TB this 192 * code will need revisiting. There is an implicit assumption that there 193 * are no *huge* holes in the physical address space too. 194 */ 195 #define TERABYTE (1ul << 40) 196 #define PHYSMEM_MAX64 mmu_btop(64 * TERABYTE) 197 #define PHYSMEM PHYSMEM_MAX64 198 #define AMD64_VA_HOLE_END 0xFFFF800000000000ul 199 200 #endif /* __amd64 */ 201 202 pgcnt_t physmem = PHYSMEM; 203 pgcnt_t obp_pages; /* Memory used by PROM for its text and data */ 204 205 char *kobj_file_buf; 206 int kobj_file_bufsize; /* set in /etc/system */ 207 208 /* Global variables for MP support. Used in mp_startup */ 209 caddr_t rm_platter_va; 210 uint32_t rm_platter_pa; 211 212 int auto_lpg_disable = 1; 213 214 /* 215 * Some CPUs have holes in the middle of the 64-bit virtual address range. 216 */ 217 uintptr_t hole_start, hole_end; 218 219 /* 220 * kpm mapping window 221 */ 222 caddr_t kpm_vbase; 223 size_t kpm_size; 224 static int kpm_desired; 225 #ifdef __amd64 226 static uintptr_t segkpm_base = (uintptr_t)SEGKPM_BASE; 227 #endif 228 229 /* 230 * Configuration parameters set at boot time. 231 */ 232 233 caddr_t econtig; /* end of first block of contiguous kernel */ 234 235 struct bootops *bootops = 0; /* passed in from boot */ 236 struct bootops **bootopsp; 237 struct boot_syscalls *sysp; /* passed in from boot */ 238 239 char bootblock_fstype[16]; 240 241 char kern_bootargs[OBP_MAXPATHLEN]; 242 243 /* 244 * ZFS zio segment. This allows us to exclude large portions of ZFS data that 245 * gets cached in kmem caches on the heap. If this is set to zero, we allocate 246 * zio buffers from their own segment, otherwise they are allocated from the 247 * heap. The optimization of allocating zio buffers from their own segment is 248 * only valid on 64-bit kernels. 249 */ 250 #if defined(__amd64) 251 int segzio_fromheap = 0; 252 #else 253 int segzio_fromheap = 1; 254 #endif 255 256 /* 257 * new memory fragmentations are possible in startup() due to BOP_ALLOCs. this 258 * depends on number of BOP_ALLOC calls made and requested size, memory size 259 * combination and whether boot.bin memory needs to be freed. 260 */ 261 #define POSS_NEW_FRAGMENTS 12 262 263 /* 264 * VM data structures 265 */ 266 long page_hashsz; /* Size of page hash table (power of two) */ 267 struct page *pp_base; /* Base of initial system page struct array */ 268 struct page **page_hash; /* Page hash table */ 269 pad_mutex_t *pse_mutex; /* Locks protecting pp->p_selock */ 270 size_t pse_table_size; /* Number of mutexes in pse_mutex[] */ 271 int pse_shift; /* log2(pse_table_size) */ 272 struct seg ktextseg; /* Segment used for kernel executable image */ 273 struct seg kvalloc; /* Segment used for "valloc" mapping */ 274 struct seg kpseg; /* Segment used for pageable kernel virt mem */ 275 struct seg kmapseg; /* Segment used for generic kernel mappings */ 276 struct seg kdebugseg; /* Segment used for the kernel debugger */ 277 278 struct seg *segkmap = &kmapseg; /* Kernel generic mapping segment */ 279 static struct seg *segmap = &kmapseg; /* easier to use name for in here */ 280 281 struct seg *segkp = &kpseg; /* Pageable kernel virtual memory segment */ 282 283 #if defined(__amd64) 284 struct seg kvseg_core; /* Segment used for the core heap */ 285 struct seg kpmseg; /* Segment used for physical mapping */ 286 struct seg *segkpm = &kpmseg; /* 64bit kernel physical mapping segment */ 287 #else 288 struct seg *segkpm = NULL; /* Unused on IA32 */ 289 #endif 290 291 caddr_t segkp_base; /* Base address of segkp */ 292 caddr_t segzio_base; /* Base address of segzio */ 293 #if defined(__amd64) 294 pgcnt_t segkpsize = btop(SEGKPDEFSIZE); /* size of segkp segment in pages */ 295 #else 296 pgcnt_t segkpsize = 0; 297 #endif 298 pgcnt_t segziosize = 0; /* size of zio segment in pages */ 299 300 /* 301 * VA range available to the debugger 302 */ 303 const caddr_t kdi_segdebugbase = (const caddr_t)SEGDEBUGBASE; 304 const size_t kdi_segdebugsize = SEGDEBUGSIZE; 305 306 struct memseg *memseg_base; 307 struct vnode unused_pages_vp; 308 309 #define FOURGB 0x100000000LL 310 311 struct memlist *memlist; 312 313 caddr_t s_text; /* start of kernel text segment */ 314 caddr_t e_text; /* end of kernel text segment */ 315 caddr_t s_data; /* start of kernel data segment */ 316 caddr_t e_data; /* end of kernel data segment */ 317 caddr_t modtext; /* start of loadable module text reserved */ 318 caddr_t e_modtext; /* end of loadable module text reserved */ 319 caddr_t moddata; /* start of loadable module data reserved */ 320 caddr_t e_moddata; /* end of loadable module data reserved */ 321 322 struct memlist *phys_install; /* Total installed physical memory */ 323 struct memlist *phys_avail; /* Total available physical memory */ 324 325 /* 326 * kphysm_init returns the number of pages that were processed 327 */ 328 static pgcnt_t kphysm_init(page_t *, pgcnt_t); 329 330 #define IO_PROP_SIZE 64 /* device property size */ 331 332 /* 333 * a couple useful roundup macros 334 */ 335 #define ROUND_UP_PAGE(x) \ 336 ((uintptr_t)P2ROUNDUP((uintptr_t)(x), (uintptr_t)MMU_PAGESIZE)) 337 #define ROUND_UP_LPAGE(x) \ 338 ((uintptr_t)P2ROUNDUP((uintptr_t)(x), mmu.level_size[1])) 339 #define ROUND_UP_4MEG(x) \ 340 ((uintptr_t)P2ROUNDUP((uintptr_t)(x), (uintptr_t)FOUR_MEG)) 341 #define ROUND_UP_TOPLEVEL(x) \ 342 ((uintptr_t)P2ROUNDUP((uintptr_t)(x), mmu.level_size[mmu.max_level])) 343 344 /* 345 * 32-bit Kernel's Virtual memory layout. 346 * +-----------------------+ 347 * | | 348 * 0xFFC00000 -|-----------------------|- ARGSBASE 349 * | debugger | 350 * 0xFF800000 -|-----------------------|- SEGDEBUGBASE 351 * | Kernel Data | 352 * 0xFEC00000 -|-----------------------| 353 * | Kernel Text | 354 * 0xFE800000 -|-----------------------|- KERNEL_TEXT (0xFB400000 on Xen) 355 * |--- GDT ---|- GDT page (GDT_VA) 356 * |--- debug info ---|- debug info (DEBUG_INFO_VA) 357 * | | 358 * | page_t structures | 359 * | memsegs, memlists, | 360 * | page hash, etc. | 361 * --- -|-----------------------|- ekernelheap, valloc_base (floating) 362 * | | (segkp is just an arena in the heap) 363 * | | 364 * | kvseg | 365 * | | 366 * | | 367 * --- -|-----------------------|- kernelheap (floating) 368 * | Segkmap | 369 * 0xC3002000 -|-----------------------|- segmap_start (floating) 370 * | Red Zone | 371 * 0xC3000000 -|-----------------------|- kernelbase / userlimit (floating) 372 * | | || 373 * | Shared objects | \/ 374 * | | 375 * : : 376 * | user data | 377 * |-----------------------| 378 * | user text | 379 * 0x08048000 -|-----------------------| 380 * | user stack | 381 * : : 382 * | invalid | 383 * 0x00000000 +-----------------------+ 384 * 385 * 386 * 64-bit Kernel's Virtual memory layout. (assuming 64 bit app) 387 * +-----------------------+ 388 * | | 389 * 0xFFFFFFFF.FFC00000 |-----------------------|- ARGSBASE 390 * | debugger (?) | 391 * 0xFFFFFFFF.FF800000 |-----------------------|- SEGDEBUGBASE 392 * | unused | 393 * +-----------------------+ 394 * | Kernel Data | 395 * 0xFFFFFFFF.FBC00000 |-----------------------| 396 * | Kernel Text | 397 * 0xFFFFFFFF.FB800000 |-----------------------|- KERNEL_TEXT 398 * |--- GDT ---|- GDT page (GDT_VA) 399 * |--- debug info ---|- debug info (DEBUG_INFO_VA) 400 * | | 401 * | Core heap | (used for loadable modules) 402 * 0xFFFFFFFF.C0000000 |-----------------------|- core_base / ekernelheap 403 * | Kernel | 404 * | heap | 405 * 0xFFFFFXXX.XXX00000 |-----------------------|- kernelheap (floating) 406 * | segmap | 407 * 0xFFFFFXXX.XXX00000 |-----------------------|- segmap_start (floating) 408 * | device mappings | 409 * 0xFFFFFXXX.XXX00000 |-----------------------|- toxic_addr (floating) 410 * | segzio | 411 * 0xFFFFFXXX.XXX00000 |-----------------------|- segzio_base (floating) 412 * | segkp | 413 * --- |-----------------------|- segkp_base (floating) 414 * | page_t structures | valloc_base + valloc_sz 415 * | memsegs, memlists, | 416 * | page hash, etc. | 417 * 0xFFFFFF00.00000000 |-----------------------|- valloc_base (lower if > 1TB) 418 * | segkpm | 419 * 0xFFFFFE00.00000000 |-----------------------| 420 * | Red Zone | 421 * 0xFFFFFD80.00000000 |-----------------------|- KERNELBASE (lower if > 1TB) 422 * | User stack |- User space memory 423 * | | 424 * | shared objects, etc | (grows downwards) 425 * : : 426 * | | 427 * 0xFFFF8000.00000000 |-----------------------| 428 * | | 429 * | VA Hole / unused | 430 * | | 431 * 0x00008000.00000000 |-----------------------| 432 * | | 433 * | | 434 * : : 435 * | user heap | (grows upwards) 436 * | | 437 * | user data | 438 * |-----------------------| 439 * | user text | 440 * 0x00000000.04000000 |-----------------------| 441 * | invalid | 442 * 0x00000000.00000000 +-----------------------+ 443 * 444 * A 32 bit app on the 64 bit kernel sees the same layout as on the 32 bit 445 * kernel, except that userlimit is raised to 0xfe000000 446 * 447 * Floating values: 448 * 449 * valloc_base: start of the kernel's memory management/tracking data 450 * structures. This region contains page_t structures for 451 * physical memory, memsegs, memlists, and the page hash. 452 * 453 * core_base: start of the kernel's "core" heap area on 64-bit systems. 454 * This area is intended to be used for global data as well as for module 455 * text/data that does not fit into the nucleus pages. The core heap is 456 * restricted to a 2GB range, allowing every address within it to be 457 * accessed using rip-relative addressing 458 * 459 * ekernelheap: end of kernelheap and start of segmap. 460 * 461 * kernelheap: start of kernel heap. On 32-bit systems, this starts right 462 * above a red zone that separates the user's address space from the 463 * kernel's. On 64-bit systems, it sits above segkp and segkpm. 464 * 465 * segmap_start: start of segmap. The length of segmap can be modified 466 * by changing segmapsize in /etc/system (preferred) or eeprom (deprecated). 467 * The default length is 16MB on 32-bit systems and 64MB on 64-bit systems. 468 * 469 * kernelbase: On a 32-bit kernel the default value of 0xd4000000 will be 470 * decreased by 2X the size required for page_t. This allows the kernel 471 * heap to grow in size with physical memory. With sizeof(page_t) == 80 472 * bytes, the following shows the values of kernelbase and kernel heap 473 * sizes for different memory configurations (assuming default segmap and 474 * segkp sizes). 475 * 476 * mem size for kernelbase kernel heap 477 * size page_t's size 478 * ---- --------- ---------- ----------- 479 * 1gb 0x01400000 0xd1800000 684MB 480 * 2gb 0x02800000 0xcf000000 704MB 481 * 4gb 0x05000000 0xca000000 744MB 482 * 6gb 0x07800000 0xc5000000 784MB 483 * 8gb 0x0a000000 0xc0000000 824MB 484 * 16gb 0x14000000 0xac000000 984MB 485 * 32gb 0x28000000 0x84000000 1304MB 486 * 64gb 0x50000000 0x34000000 1944MB (*) 487 * 488 * kernelbase is less than the abi minimum of 0xc0000000 for memory 489 * configurations above 8gb. 490 * 491 * (*) support for memory configurations above 32gb will require manual tuning 492 * of kernelbase to balance out the need of user applications. 493 */ 494 495 /* real-time-clock initialization parameters */ 496 extern time_t process_rtc_config_file(void); 497 498 uintptr_t kernelbase; 499 uintptr_t postbootkernelbase; /* not set till boot loader is gone */ 500 uintptr_t eprom_kernelbase; 501 size_t segmapsize; 502 uintptr_t segmap_start; 503 int segmapfreelists; 504 pgcnt_t npages; 505 pgcnt_t orig_npages; 506 size_t core_size; /* size of "core" heap */ 507 uintptr_t core_base; /* base address of "core" heap */ 508 509 /* 510 * List of bootstrap pages. We mark these as allocated in startup. 511 * release_bootstrap() will free them when we're completely done with 512 * the bootstrap. 513 */ 514 static page_t *bootpages; 515 516 /* 517 * boot time pages that have a vnode from the ramdisk will keep that forever. 518 */ 519 static page_t *rd_pages; 520 521 struct system_hardware system_hardware; 522 523 /* 524 * Is this Solaris instance running in a fully virtualized xVM domain? 525 */ 526 int xpv_is_hvm = 0; 527 528 /* 529 * Enable some debugging messages concerning memory usage... 530 */ 531 static void 532 print_memlist(char *title, struct memlist *mp) 533 { 534 prom_printf("MEMLIST: %s:\n", title); 535 while (mp != NULL) { 536 prom_printf("\tAddress 0x%" PRIx64 ", size 0x%" PRIx64 "\n", 537 mp->address, mp->size); 538 mp = mp->next; 539 } 540 } 541 542 /* 543 * XX64 need a comment here.. are these just default values, surely 544 * we read the "cpuid" type information to figure this out. 545 */ 546 int l2cache_sz = 0x80000; 547 int l2cache_linesz = 0x40; 548 int l2cache_assoc = 1; 549 550 static size_t textrepl_min_gb = 10; 551 552 /* 553 * on 64 bit we use a predifined VA range for mapping devices in the kernel 554 * on 32 bit the mappings are intermixed in the heap, so we use a bit map 555 */ 556 #ifdef __amd64 557 558 vmem_t *device_arena; 559 uintptr_t toxic_addr = (uintptr_t)NULL; 560 size_t toxic_size = 1024 * 1024 * 1024; /* Sparc uses 1 gig too */ 561 562 #else /* __i386 */ 563 564 ulong_t *toxic_bit_map; /* one bit for each 4k of VA in heap_arena */ 565 size_t toxic_bit_map_len = 0; /* in bits */ 566 567 #endif /* __i386 */ 568 569 /* 570 * Simple boot time debug facilities 571 */ 572 static char *prm_dbg_str[] = { 573 "%s:%d: '%s' is 0x%x\n", 574 "%s:%d: '%s' is 0x%llx\n" 575 }; 576 577 int prom_debug; 578 579 #define PRM_DEBUG(q) if (prom_debug) \ 580 prom_printf(prm_dbg_str[sizeof (q) >> 3], "startup.c", __LINE__, #q, q); 581 #define PRM_POINT(q) if (prom_debug) \ 582 prom_printf("%s:%d: %s\n", "startup.c", __LINE__, q); 583 584 /* 585 * This structure is used to keep track of the intial allocations 586 * done in startup_memlist(). The value of NUM_ALLOCATIONS needs to 587 * be >= the number of ADD_TO_ALLOCATIONS() executed in the code. 588 */ 589 #define NUM_ALLOCATIONS 7 590 int num_allocations = 0; 591 struct { 592 void **al_ptr; 593 size_t al_size; 594 } allocations[NUM_ALLOCATIONS]; 595 size_t valloc_sz = 0; 596 uintptr_t valloc_base; 597 598 #define ADD_TO_ALLOCATIONS(ptr, size) { \ 599 size = ROUND_UP_PAGE(size); \ 600 if (num_allocations == NUM_ALLOCATIONS) \ 601 panic("too many ADD_TO_ALLOCATIONS()"); \ 602 allocations[num_allocations].al_ptr = (void**)&ptr; \ 603 allocations[num_allocations].al_size = size; \ 604 valloc_sz += size; \ 605 ++num_allocations; \ 606 } 607 608 /* 609 * Allocate all the initial memory needed by the page allocator. 610 */ 611 static void 612 perform_allocations(void) 613 { 614 caddr_t mem; 615 int i; 616 int valloc_align; 617 618 PRM_DEBUG(valloc_base); 619 PRM_DEBUG(valloc_sz); 620 valloc_align = mmu.level_size[mmu.max_page_level > 0]; 621 mem = BOP_ALLOC(bootops, (caddr_t)valloc_base, valloc_sz, valloc_align); 622 if (mem != (caddr_t)valloc_base) 623 panic("BOP_ALLOC() failed"); 624 bzero(mem, valloc_sz); 625 for (i = 0; i < num_allocations; ++i) { 626 *allocations[i].al_ptr = (void *)mem; 627 mem += allocations[i].al_size; 628 } 629 } 630 631 /* 632 * Our world looks like this at startup time. 633 * 634 * In a 32-bit OS, boot loads the kernel text at 0xfe800000 and kernel data 635 * at 0xfec00000. On a 64-bit OS, kernel text and data are loaded at 636 * 0xffffffff.fe800000 and 0xffffffff.fec00000 respectively. Those 637 * addresses are fixed in the binary at link time. 638 * 639 * On the text page: 640 * unix/genunix/krtld/module text loads. 641 * 642 * On the data page: 643 * unix/genunix/krtld/module data loads. 644 * 645 * Machine-dependent startup code 646 */ 647 void 648 startup(void) 649 { 650 #if !defined(__xpv) 651 extern void startup_bios_disk(void); 652 extern void startup_pci_bios(void); 653 #endif 654 /* 655 * Make sure that nobody tries to use sekpm until we have 656 * initialized it properly. 657 */ 658 #if defined(__amd64) 659 kpm_desired = 1; 660 #endif 661 kpm_enable = 0; 662 663 #if defined(__xpv) /* XXPV fix me! */ 664 { 665 extern int segvn_use_regions; 666 segvn_use_regions = 0; 667 } 668 #endif 669 progressbar_init(); 670 startup_init(); 671 #if defined(__xpv) 672 startup_xen_version(); 673 #endif 674 startup_memlist(); 675 startup_kmem(); 676 startup_vm(); 677 #if !defined(__xpv) 678 startup_pci_bios(); 679 #endif 680 startup_modules(); 681 #if !defined(__xpv) 682 startup_bios_disk(); 683 #endif 684 startup_end(); 685 progressbar_start(); 686 } 687 688 static void 689 startup_init() 690 { 691 PRM_POINT("startup_init() starting..."); 692 693 /* 694 * Complete the extraction of cpuid data 695 */ 696 cpuid_pass2(CPU); 697 698 (void) check_boot_version(BOP_GETVERSION(bootops)); 699 700 /* 701 * Check for prom_debug in boot environment 702 */ 703 if (BOP_GETPROPLEN(bootops, "prom_debug") >= 0) { 704 ++prom_debug; 705 PRM_POINT("prom_debug found in boot enviroment"); 706 } 707 708 /* 709 * Collect node, cpu and memory configuration information. 710 */ 711 get_system_configuration(); 712 713 /* 714 * Halt if this is an unsupported processor. 715 */ 716 if (x86_type == X86_TYPE_486 || x86_type == X86_TYPE_CYRIX_486) { 717 printf("\n486 processor (\"%s\") detected.\n", 718 CPU->cpu_brandstr); 719 halt("This processor is not supported by this release " 720 "of Solaris."); 721 } 722 723 PRM_POINT("startup_init() done"); 724 } 725 726 /* 727 * Callback for copy_memlist_filter() to filter nucleus, kadb/kmdb, (ie. 728 * everything mapped above KERNEL_TEXT) pages from phys_avail. Note it 729 * also filters out physical page zero. There is some reliance on the 730 * boot loader allocating only a few contiguous physical memory chunks. 731 */ 732 static void 733 avail_filter(uint64_t *addr, uint64_t *size) 734 { 735 uintptr_t va; 736 uintptr_t next_va; 737 pfn_t pfn; 738 uint64_t pfn_addr; 739 uint64_t pfn_eaddr; 740 uint_t prot; 741 size_t len; 742 uint_t change; 743 744 if (prom_debug) 745 prom_printf("\tFilter: in: a=%" PRIx64 ", s=%" PRIx64 "\n", 746 *addr, *size); 747 748 /* 749 * page zero is required for BIOS.. never make it available 750 */ 751 if (*addr == 0) { 752 *addr += MMU_PAGESIZE; 753 *size -= MMU_PAGESIZE; 754 } 755 756 /* 757 * First we trim from the front of the range. Since kbm_probe() 758 * walks ranges in virtual order, but addr/size are physical, we need 759 * to the list until no changes are seen. This deals with the case 760 * where page "p" is mapped at v, page "p + PAGESIZE" is mapped at w 761 * but w < v. 762 */ 763 do { 764 change = 0; 765 for (va = KERNEL_TEXT; 766 *size > 0 && kbm_probe(&va, &len, &pfn, &prot) != 0; 767 va = next_va) { 768 769 next_va = va + len; 770 pfn_addr = pfn_to_pa(pfn); 771 pfn_eaddr = pfn_addr + len; 772 773 if (pfn_addr <= *addr && pfn_eaddr > *addr) { 774 change = 1; 775 while (*size > 0 && len > 0) { 776 *addr += MMU_PAGESIZE; 777 *size -= MMU_PAGESIZE; 778 len -= MMU_PAGESIZE; 779 } 780 } 781 } 782 if (change && prom_debug) 783 prom_printf("\t\ttrim: a=%" PRIx64 ", s=%" PRIx64 "\n", 784 *addr, *size); 785 } while (change); 786 787 /* 788 * Trim pages from the end of the range. 789 */ 790 for (va = KERNEL_TEXT; 791 *size > 0 && kbm_probe(&va, &len, &pfn, &prot) != 0; 792 va = next_va) { 793 794 next_va = va + len; 795 pfn_addr = pfn_to_pa(pfn); 796 797 if (pfn_addr >= *addr && pfn_addr < *addr + *size) 798 *size = pfn_addr - *addr; 799 } 800 801 if (prom_debug) 802 prom_printf("\tFilter out: a=%" PRIx64 ", s=%" PRIx64 "\n", 803 *addr, *size); 804 } 805 806 static void 807 kpm_init() 808 { 809 struct segkpm_crargs b; 810 811 /* 812 * These variables were all designed for sfmmu in which segkpm is 813 * mapped using a single pagesize - either 8KB or 4MB. On x86, we 814 * might use 2+ page sizes on a single machine, so none of these 815 * variables have a single correct value. They are set up as if we 816 * always use a 4KB pagesize, which should do no harm. In the long 817 * run, we should get rid of KPM's assumption that only a single 818 * pagesize is used. 819 */ 820 kpm_pgshft = MMU_PAGESHIFT; 821 kpm_pgsz = MMU_PAGESIZE; 822 kpm_pgoff = MMU_PAGEOFFSET; 823 kpmp2pshft = 0; 824 kpmpnpgs = 1; 825 ASSERT(((uintptr_t)kpm_vbase & (kpm_pgsz - 1)) == 0); 826 827 PRM_POINT("about to create segkpm"); 828 rw_enter(&kas.a_lock, RW_WRITER); 829 830 if (seg_attach(&kas, kpm_vbase, kpm_size, segkpm) < 0) 831 panic("cannot attach segkpm"); 832 833 b.prot = PROT_READ | PROT_WRITE; 834 b.nvcolors = 1; 835 836 if (segkpm_create(segkpm, (caddr_t)&b) != 0) 837 panic("segkpm_create segkpm"); 838 839 rw_exit(&kas.a_lock); 840 } 841 842 /* 843 * The debug info page provides enough information to allow external 844 * inspectors (e.g. when running under a hypervisor) to bootstrap 845 * themselves into allowing full-blown kernel debugging. 846 */ 847 static void 848 init_debug_info(void) 849 { 850 caddr_t mem; 851 debug_info_t *di; 852 853 #ifndef __lint 854 ASSERT(sizeof (debug_info_t) < MMU_PAGESIZE); 855 #endif 856 857 mem = BOP_ALLOC(bootops, (caddr_t)DEBUG_INFO_VA, MMU_PAGESIZE, 858 MMU_PAGESIZE); 859 860 if (mem != (caddr_t)DEBUG_INFO_VA) 861 panic("BOP_ALLOC() failed"); 862 bzero(mem, MMU_PAGESIZE); 863 864 di = (debug_info_t *)mem; 865 866 di->di_magic = DEBUG_INFO_MAGIC; 867 di->di_version = DEBUG_INFO_VERSION; 868 di->di_modules = (uintptr_t)&modules; 869 di->di_s_text = (uintptr_t)s_text; 870 di->di_e_text = (uintptr_t)e_text; 871 di->di_s_data = (uintptr_t)s_data; 872 di->di_e_data = (uintptr_t)e_data; 873 di->di_hat_htable_off = offsetof(hat_t, hat_htable); 874 di->di_ht_pfn_off = offsetof(htable_t, ht_pfn); 875 } 876 877 /* 878 * Build the memlists and other kernel essential memory system data structures. 879 * This is everything at valloc_base. 880 */ 881 static void 882 startup_memlist(void) 883 { 884 size_t memlist_sz; 885 size_t memseg_sz; 886 size_t pagehash_sz; 887 size_t pp_sz; 888 uintptr_t va; 889 size_t len; 890 uint_t prot; 891 pfn_t pfn; 892 int memblocks; 893 caddr_t pagecolor_mem; 894 size_t pagecolor_memsz; 895 caddr_t page_ctrs_mem; 896 size_t page_ctrs_size; 897 size_t pse_table_alloc_size; 898 struct memlist *current; 899 extern void startup_build_mem_nodes(struct memlist *); 900 901 /* XX64 fix these - they should be in include files */ 902 extern size_t page_coloring_init(uint_t, int, int); 903 extern void page_coloring_setup(caddr_t); 904 905 PRM_POINT("startup_memlist() starting..."); 906 907 /* 908 * Use leftover large page nucleus text/data space for loadable modules. 909 * Use at most MODTEXT/MODDATA. 910 */ 911 len = kbm_nucleus_size; 912 ASSERT(len > MMU_PAGESIZE); 913 914 moddata = (caddr_t)ROUND_UP_PAGE(e_data); 915 e_moddata = (caddr_t)P2ROUNDUP((uintptr_t)e_data, (uintptr_t)len); 916 if (e_moddata - moddata > MODDATA) 917 e_moddata = moddata + MODDATA; 918 919 modtext = (caddr_t)ROUND_UP_PAGE(e_text); 920 e_modtext = (caddr_t)P2ROUNDUP((uintptr_t)e_text, (uintptr_t)len); 921 if (e_modtext - modtext > MODTEXT) 922 e_modtext = modtext + MODTEXT; 923 924 econtig = e_moddata; 925 926 PRM_DEBUG(modtext); 927 PRM_DEBUG(e_modtext); 928 PRM_DEBUG(moddata); 929 PRM_DEBUG(e_moddata); 930 PRM_DEBUG(econtig); 931 932 /* 933 * Examine the boot loader physical memory map to find out: 934 * - total memory in system - physinstalled 935 * - the max physical address - physmax 936 * - the number of discontiguous segments of memory. 937 */ 938 if (prom_debug) 939 print_memlist("boot physinstalled", 940 bootops->boot_mem->physinstalled); 941 installed_top_size(bootops->boot_mem->physinstalled, &physmax, 942 &physinstalled, &memblocks); 943 PRM_DEBUG(physmax); 944 PRM_DEBUG(physinstalled); 945 PRM_DEBUG(memblocks); 946 947 /* 948 * Initialize hat's mmu parameters. 949 * Check for enforce-prot-exec in boot environment. It's used to 950 * enable/disable support for the page table entry NX bit. 951 * The default is to enforce PROT_EXEC on processors that support NX. 952 * Boot seems to round up the "len", but 8 seems to be big enough. 953 */ 954 mmu_init(); 955 956 #ifdef __i386 957 /* 958 * physmax is lowered if there is more memory than can be 959 * physically addressed in 32 bit (PAE/non-PAE) modes. 960 */ 961 if (mmu.pae_hat) { 962 if (PFN_ABOVE64G(physmax)) { 963 physinstalled -= (physmax - (PFN_64G - 1)); 964 physmax = PFN_64G - 1; 965 } 966 } else { 967 if (PFN_ABOVE4G(physmax)) { 968 physinstalled -= (physmax - (PFN_4G - 1)); 969 physmax = PFN_4G - 1; 970 } 971 } 972 #endif 973 974 startup_build_mem_nodes(bootops->boot_mem->physinstalled); 975 976 if (BOP_GETPROPLEN(bootops, "enforce-prot-exec") >= 0) { 977 int len = BOP_GETPROPLEN(bootops, "enforce-prot-exec"); 978 char value[8]; 979 980 if (len < 8) 981 (void) BOP_GETPROP(bootops, "enforce-prot-exec", value); 982 else 983 (void) strcpy(value, ""); 984 if (strcmp(value, "off") == 0) 985 mmu.pt_nx = 0; 986 } 987 PRM_DEBUG(mmu.pt_nx); 988 989 /* 990 * We will need page_t's for every page in the system, except for 991 * memory mapped at or above above the start of the kernel text segment. 992 * 993 * pages above e_modtext are attributed to kernel debugger (obp_pages) 994 */ 995 npages = physinstalled - 1; /* avail_filter() skips page 0, so "- 1" */ 996 obp_pages = 0; 997 va = KERNEL_TEXT; 998 while (kbm_probe(&va, &len, &pfn, &prot) != 0) { 999 npages -= len >> MMU_PAGESHIFT; 1000 if (va >= (uintptr_t)e_moddata) 1001 obp_pages += len >> MMU_PAGESHIFT; 1002 va += len; 1003 } 1004 PRM_DEBUG(npages); 1005 PRM_DEBUG(obp_pages); 1006 1007 /* 1008 * If physmem is patched to be non-zero, use it instead of the computed 1009 * value unless it is larger than the actual amount of memory on hand. 1010 */ 1011 if (physmem == 0 || physmem > npages) { 1012 physmem = npages; 1013 } else if (physmem < npages) { 1014 orig_npages = npages; 1015 npages = physmem; 1016 } 1017 PRM_DEBUG(physmem); 1018 1019 /* 1020 * We now compute the sizes of all the initial allocations for 1021 * structures the kernel needs in order do kmem_alloc(). These 1022 * include: 1023 * memsegs 1024 * memlists 1025 * page hash table 1026 * page_t's 1027 * page coloring data structs 1028 */ 1029 memseg_sz = sizeof (struct memseg) * (memblocks + POSS_NEW_FRAGMENTS); 1030 ADD_TO_ALLOCATIONS(memseg_base, memseg_sz); 1031 PRM_DEBUG(memseg_sz); 1032 1033 /* 1034 * Reserve space for memlists. There's no real good way to know exactly 1035 * how much room we'll need, but this should be a good upper bound. 1036 */ 1037 memlist_sz = ROUND_UP_PAGE(2 * sizeof (struct memlist) * 1038 (memblocks + POSS_NEW_FRAGMENTS)); 1039 ADD_TO_ALLOCATIONS(memlist, memlist_sz); 1040 PRM_DEBUG(memlist_sz); 1041 1042 /* 1043 * The page structure hash table size is a power of 2 1044 * such that the average hash chain length is PAGE_HASHAVELEN. 1045 */ 1046 page_hashsz = npages / PAGE_HASHAVELEN; 1047 page_hashsz = 1 << highbit(page_hashsz); 1048 pagehash_sz = sizeof (struct page *) * page_hashsz; 1049 ADD_TO_ALLOCATIONS(page_hash, pagehash_sz); 1050 PRM_DEBUG(pagehash_sz); 1051 1052 /* 1053 * Set aside room for the page structures themselves. 1054 */ 1055 PRM_DEBUG(npages); 1056 pp_sz = sizeof (struct page) * npages; 1057 ADD_TO_ALLOCATIONS(pp_base, pp_sz); 1058 PRM_DEBUG(pp_sz); 1059 1060 /* 1061 * determine l2 cache info and memory size for page coloring 1062 */ 1063 (void) getl2cacheinfo(CPU, 1064 &l2cache_sz, &l2cache_linesz, &l2cache_assoc); 1065 pagecolor_memsz = 1066 page_coloring_init(l2cache_sz, l2cache_linesz, l2cache_assoc); 1067 ADD_TO_ALLOCATIONS(pagecolor_mem, pagecolor_memsz); 1068 PRM_DEBUG(pagecolor_memsz); 1069 1070 page_ctrs_size = page_ctrs_sz(); 1071 ADD_TO_ALLOCATIONS(page_ctrs_mem, page_ctrs_size); 1072 PRM_DEBUG(page_ctrs_size); 1073 1074 /* 1075 * Allocate the array that protects pp->p_selock. 1076 */ 1077 pse_shift = size_pse_array(physmem, max_ncpus); 1078 pse_table_size = 1 << pse_shift; 1079 pse_table_alloc_size = pse_table_size * sizeof (pad_mutex_t); 1080 ADD_TO_ALLOCATIONS(pse_mutex, pse_table_alloc_size); 1081 1082 #if defined(__amd64) 1083 valloc_sz = ROUND_UP_LPAGE(valloc_sz); 1084 valloc_base = VALLOC_BASE; 1085 1086 /* 1087 * The default values of VALLOC_BASE and SEGKPM_BASE should work 1088 * for values of physmax up to 1 Terabyte. They need adjusting when 1089 * memory is at addresses above 1 TB. 1090 */ 1091 if (physmax + 1 > mmu_btop(TERABYTE)) { 1092 uint64_t kpm_resv_amount = mmu_ptob(physmax + 1); 1093 1094 /* Round to largest possible pagesize for now */ 1095 kpm_resv_amount = P2ROUNDUP(kpm_resv_amount, ONE_GIG); 1096 1097 segkpm_base = -(2 * kpm_resv_amount); /* down from top VA */ 1098 1099 /* make sure we leave some space for user apps above hole */ 1100 segkpm_base = MAX(segkpm_base, AMD64_VA_HOLE_END + TERABYTE); 1101 if (segkpm_base > SEGKPM_BASE) 1102 segkpm_base = SEGKPM_BASE; 1103 PRM_DEBUG(segkpm_base); 1104 1105 valloc_base = segkpm_base + kpm_resv_amount; 1106 PRM_DEBUG(valloc_base); 1107 } 1108 #else /* __i386 */ 1109 valloc_base = (uintptr_t)(MISC_VA_BASE - valloc_sz); 1110 valloc_base = P2ALIGN(valloc_base, mmu.level_size[1]); 1111 PRM_DEBUG(valloc_base); 1112 #endif /* __i386 */ 1113 1114 /* 1115 * do all the initial allocations 1116 */ 1117 perform_allocations(); 1118 1119 /* 1120 * Build phys_install and phys_avail in kernel memspace. 1121 * - phys_install should be all memory in the system. 1122 * - phys_avail is phys_install minus any memory mapped before this 1123 * point above KERNEL_TEXT. 1124 */ 1125 current = phys_install = memlist; 1126 copy_memlist_filter(bootops->boot_mem->physinstalled, ¤t, NULL); 1127 if ((caddr_t)current > (caddr_t)memlist + memlist_sz) 1128 panic("physinstalled was too big!"); 1129 if (prom_debug) 1130 print_memlist("phys_install", phys_install); 1131 1132 phys_avail = current; 1133 PRM_POINT("Building phys_avail:\n"); 1134 copy_memlist_filter(bootops->boot_mem->physinstalled, ¤t, 1135 avail_filter); 1136 if ((caddr_t)current > (caddr_t)memlist + memlist_sz) 1137 panic("physavail was too big!"); 1138 if (prom_debug) 1139 print_memlist("phys_avail", phys_avail); 1140 1141 /* 1142 * setup page coloring 1143 */ 1144 page_coloring_setup(pagecolor_mem); 1145 page_lock_init(); /* currently a no-op */ 1146 1147 /* 1148 * free page list counters 1149 */ 1150 (void) page_ctrs_alloc(page_ctrs_mem); 1151 1152 /* 1153 * Initialize the page structures from the memory lists. 1154 */ 1155 availrmem_initial = availrmem = freemem = 0; 1156 PRM_POINT("Calling kphysm_init()..."); 1157 npages = kphysm_init(pp_base, npages); 1158 PRM_POINT("kphysm_init() done"); 1159 PRM_DEBUG(npages); 1160 1161 init_debug_info(); 1162 1163 /* 1164 * Now that page_t's have been initialized, remove all the 1165 * initial allocation pages from the kernel free page lists. 1166 */ 1167 boot_mapin((caddr_t)valloc_base, valloc_sz); 1168 boot_mapin((caddr_t)MISC_VA_BASE, MISC_VA_SIZE); 1169 PRM_POINT("startup_memlist() done"); 1170 1171 PRM_DEBUG(valloc_sz); 1172 1173 #if defined(__amd64) 1174 if ((availrmem >> (30 - MMU_PAGESHIFT)) >= 1175 textrepl_min_gb && l2cache_sz <= 2 << 20) { 1176 extern size_t textrepl_size_thresh; 1177 textrepl_size_thresh = (16 << 20) - 1; 1178 } 1179 #endif 1180 } 1181 1182 /* 1183 * Layout the kernel's part of address space and initialize kmem allocator. 1184 */ 1185 static void 1186 startup_kmem(void) 1187 { 1188 extern void page_set_colorequiv_arr(void); 1189 1190 PRM_POINT("startup_kmem() starting..."); 1191 1192 #if defined(__amd64) 1193 if (eprom_kernelbase && eprom_kernelbase != KERNELBASE) 1194 cmn_err(CE_NOTE, "!kernelbase cannot be changed on 64-bit " 1195 "systems."); 1196 kernelbase = segkpm_base - KERNEL_REDZONE_SIZE; 1197 core_base = (uintptr_t)COREHEAP_BASE; 1198 core_size = (size_t)MISC_VA_BASE - COREHEAP_BASE; 1199 #else /* __i386 */ 1200 /* 1201 * We configure kernelbase based on: 1202 * 1203 * 1. user specified kernelbase via eeprom command. Value cannot exceed 1204 * KERNELBASE_MAX. we large page align eprom_kernelbase 1205 * 1206 * 2. Default to KERNELBASE and adjust to 2X less the size for page_t. 1207 * On large memory systems we must lower kernelbase to allow 1208 * enough room for page_t's for all of memory. 1209 * 1210 * The value set here, might be changed a little later. 1211 */ 1212 if (eprom_kernelbase) { 1213 kernelbase = eprom_kernelbase & mmu.level_mask[1]; 1214 if (kernelbase > KERNELBASE_MAX) 1215 kernelbase = KERNELBASE_MAX; 1216 } else { 1217 kernelbase = (uintptr_t)KERNELBASE; 1218 kernelbase -= ROUND_UP_4MEG(2 * valloc_sz); 1219 } 1220 ASSERT((kernelbase & mmu.level_offset[1]) == 0); 1221 core_base = valloc_base; 1222 core_size = 0; 1223 #endif /* __i386 */ 1224 1225 PRM_DEBUG(core_base); 1226 PRM_DEBUG(core_size); 1227 PRM_DEBUG(kernelbase); 1228 1229 #if defined(__i386) 1230 segkp_fromheap = 1; 1231 #endif /* __i386 */ 1232 1233 ekernelheap = (char *)core_base; 1234 PRM_DEBUG(ekernelheap); 1235 1236 /* 1237 * Now that we know the real value of kernelbase, 1238 * update variables that were initialized with a value of 1239 * KERNELBASE (in common/conf/param.c). 1240 * 1241 * XXX The problem with this sort of hackery is that the 1242 * compiler just may feel like putting the const declarations 1243 * (in param.c) into the .text section. Perhaps they should 1244 * just be declared as variables there? 1245 */ 1246 1247 *(uintptr_t *)&_kernelbase = kernelbase; 1248 *(uintptr_t *)&_userlimit = kernelbase; 1249 #if defined(__amd64) 1250 *(uintptr_t *)&_userlimit -= KERNELBASE - USERLIMIT; 1251 #else 1252 *(uintptr_t *)&_userlimit32 = _userlimit; 1253 #endif 1254 PRM_DEBUG(_kernelbase); 1255 PRM_DEBUG(_userlimit); 1256 PRM_DEBUG(_userlimit32); 1257 1258 layout_kernel_va(); 1259 1260 #if defined(__i386) 1261 /* 1262 * If segmap is too large we can push the bottom of the kernel heap 1263 * higher than the base. Or worse, it could exceed the top of the 1264 * VA space entirely, causing it to wrap around. 1265 */ 1266 if (kernelheap >= ekernelheap || (uintptr_t)kernelheap < kernelbase) 1267 panic("too little address space available for kernelheap," 1268 " use eeprom for lower kernelbase or smaller segmapsize"); 1269 #endif /* __i386 */ 1270 1271 /* 1272 * Initialize the kernel heap. Note 3rd argument must be > 1st. 1273 */ 1274 kernelheap_init(kernelheap, ekernelheap, 1275 kernelheap + MMU_PAGESIZE, 1276 (void *)core_base, (void *)(core_base + core_size)); 1277 1278 #if defined(__xpv) 1279 /* 1280 * Link pending events struct into cpu struct 1281 */ 1282 CPU->cpu_m.mcpu_evt_pend = &cpu0_evt_data; 1283 #endif 1284 /* 1285 * Initialize kernel memory allocator. 1286 */ 1287 kmem_init(); 1288 1289 /* 1290 * Factor in colorequiv to check additional 'equivalent' bins 1291 */ 1292 page_set_colorequiv_arr(); 1293 1294 /* 1295 * print this out early so that we know what's going on 1296 */ 1297 cmn_err(CE_CONT, "?features: %b\n", x86_feature, FMT_X86_FEATURE); 1298 1299 /* 1300 * Initialize bp_mapin(). 1301 */ 1302 bp_init(MMU_PAGESIZE, HAT_STORECACHING_OK); 1303 1304 /* 1305 * orig_npages is non-zero if physmem has been configured for less 1306 * than the available memory. 1307 */ 1308 if (orig_npages) { 1309 cmn_err(CE_WARN, "!%slimiting physmem to 0x%lx of 0x%lx pages", 1310 (npages == PHYSMEM ? "Due to virtual address space " : ""), 1311 npages, orig_npages); 1312 } 1313 #if defined(__i386) 1314 if (eprom_kernelbase && (eprom_kernelbase != kernelbase)) 1315 cmn_err(CE_WARN, "kernelbase value, User specified 0x%lx, " 1316 "System using 0x%lx", 1317 (uintptr_t)eprom_kernelbase, (uintptr_t)kernelbase); 1318 #endif 1319 1320 #ifdef KERNELBASE_ABI_MIN 1321 if (kernelbase < (uintptr_t)KERNELBASE_ABI_MIN) { 1322 cmn_err(CE_NOTE, "!kernelbase set to 0x%lx, system is not " 1323 "i386 ABI compliant.", (uintptr_t)kernelbase); 1324 } 1325 #endif 1326 1327 #ifdef __xpv 1328 /* 1329 * Some of the xen start information has to be relocated up 1330 * into the kernel's permanent address space. 1331 */ 1332 PRM_POINT("calling xen_relocate_start_info()"); 1333 xen_relocate_start_info(); 1334 PRM_POINT("xen_relocate_start_info() done"); 1335 1336 /* 1337 * (Update the vcpu pointer in our cpu structure to point into 1338 * the relocated shared info.) 1339 */ 1340 CPU->cpu_m.mcpu_vcpu_info = 1341 &HYPERVISOR_shared_info->vcpu_info[CPU->cpu_id]; 1342 #endif 1343 1344 PRM_POINT("startup_kmem() done"); 1345 } 1346 1347 #ifndef __xpv 1348 /* 1349 * If we have detected that we are running in an HVM environment, we need 1350 * to prepend the PV driver directory to the module search path. 1351 */ 1352 #define HVM_MOD_DIR "/platform/i86hvm/kernel" 1353 static void 1354 update_default_path() 1355 { 1356 char *current, *newpath; 1357 int newlen; 1358 1359 /* 1360 * We are about to resync with krtld. krtld will reset its 1361 * internal module search path iff Solaris has set default_path. 1362 * We want to be sure we're prepending this new directory to the 1363 * right search path. 1364 */ 1365 current = (default_path == NULL) ? kobj_module_path : default_path; 1366 1367 newlen = strlen(HVM_MOD_DIR) + strlen(current) + 1; 1368 newpath = kmem_alloc(newlen, KM_SLEEP); 1369 (void) strcpy(newpath, HVM_MOD_DIR); 1370 (void) strcat(newpath, " "); 1371 (void) strcat(newpath, current); 1372 1373 default_path = newpath; 1374 } 1375 #endif 1376 1377 static void 1378 startup_modules(void) 1379 { 1380 unsigned int i; 1381 extern void prom_setup(void); 1382 1383 PRM_POINT("startup_modules() starting..."); 1384 1385 #ifndef __xpv 1386 /* 1387 * Initialize ten-micro second timer so that drivers will 1388 * not get short changed in their init phase. This was 1389 * not getting called until clkinit which, on fast cpu's 1390 * caused the drv_usecwait to be way too short. 1391 */ 1392 microfind(); 1393 1394 if (xpv_is_hvm) 1395 update_default_path(); 1396 #endif 1397 1398 /* 1399 * Read the GMT lag from /etc/rtc_config. 1400 */ 1401 sgmtl(process_rtc_config_file()); 1402 1403 /* 1404 * Calculate default settings of system parameters based upon 1405 * maxusers, yet allow to be overridden via the /etc/system file. 1406 */ 1407 param_calc(0); 1408 1409 mod_setup(); 1410 1411 /* 1412 * Initialize system parameters. 1413 */ 1414 param_init(); 1415 1416 /* 1417 * Initialize the default brands 1418 */ 1419 brand_init(); 1420 1421 /* 1422 * maxmem is the amount of physical memory we're playing with. 1423 */ 1424 maxmem = physmem; 1425 1426 /* 1427 * Initialize segment management stuff. 1428 */ 1429 seg_init(); 1430 1431 if (modload("fs", "specfs") == -1) 1432 halt("Can't load specfs"); 1433 1434 if (modload("fs", "devfs") == -1) 1435 halt("Can't load devfs"); 1436 1437 if (modload("fs", "dev") == -1) 1438 halt("Can't load dev"); 1439 1440 (void) modloadonly("sys", "lbl_edition"); 1441 1442 dispinit(); 1443 1444 /* 1445 * This is needed here to initialize hw_serial[] for cluster booting. 1446 */ 1447 if ((i = modload("misc", "sysinit")) != (unsigned int)-1) 1448 (void) modunload(i); 1449 else 1450 cmn_err(CE_CONT, "sysinit load failed"); 1451 1452 /* Read cluster configuration data. */ 1453 clconf_init(); 1454 1455 #if defined(__xpv) 1456 ec_init(); 1457 gnttab_init(); 1458 (void) xs_early_init(); 1459 #endif /* __xpv */ 1460 1461 /* 1462 * Create a kernel device tree. First, create rootnex and 1463 * then invoke bus specific code to probe devices. 1464 */ 1465 setup_ddi(); 1466 1467 #ifndef __xpv 1468 { 1469 /* 1470 * Set up the CPU module subsystem. Modifies the device tree, 1471 * so it must be done after setup_ddi(). 1472 */ 1473 1474 cmi_hdl_t hdl; 1475 1476 if ((hdl = cmi_init(CMI_HDL_NATIVE, cmi_ntv_hwchipid(CPU), 1477 cmi_ntv_hwcoreid(CPU), cmi_ntv_hwstrandid(CPU))) != NULL) { 1478 if (x86_feature & X86_MCA) 1479 cmi_mca_init(hdl); 1480 } 1481 } 1482 #endif /* __xpv */ 1483 1484 /* 1485 * Fake a prom tree such that /dev/openprom continues to work 1486 */ 1487 PRM_POINT("startup_modules: calling prom_setup..."); 1488 prom_setup(); 1489 PRM_POINT("startup_modules: done"); 1490 1491 /* 1492 * Load all platform specific modules 1493 */ 1494 PRM_POINT("startup_modules: calling psm_modload..."); 1495 psm_modload(); 1496 1497 PRM_POINT("startup_modules() done"); 1498 } 1499 1500 /* 1501 * claim a "setaside" boot page for use in the kernel 1502 */ 1503 page_t * 1504 boot_claim_page(pfn_t pfn) 1505 { 1506 page_t *pp; 1507 1508 pp = page_numtopp_nolock(pfn); 1509 ASSERT(pp != NULL); 1510 1511 if (PP_ISBOOTPAGES(pp)) { 1512 if (pp->p_next != NULL) 1513 pp->p_next->p_prev = pp->p_prev; 1514 if (pp->p_prev == NULL) 1515 bootpages = pp->p_next; 1516 else 1517 pp->p_prev->p_next = pp->p_next; 1518 } else { 1519 /* 1520 * htable_attach() expects a base pagesize page 1521 */ 1522 if (pp->p_szc != 0) 1523 page_boot_demote(pp); 1524 pp = page_numtopp(pfn, SE_EXCL); 1525 } 1526 return (pp); 1527 } 1528 1529 /* 1530 * Walk through the pagetables looking for pages mapped in by boot. If the 1531 * setaside flag is set the pages are expected to be returned to the 1532 * kernel later in boot, so we add them to the bootpages list. 1533 */ 1534 static void 1535 protect_boot_range(uintptr_t low, uintptr_t high, int setaside) 1536 { 1537 uintptr_t va = low; 1538 size_t len; 1539 uint_t prot; 1540 pfn_t pfn; 1541 page_t *pp; 1542 pgcnt_t boot_protect_cnt = 0; 1543 1544 while (kbm_probe(&va, &len, &pfn, &prot) != 0 && va < high) { 1545 if (va + len >= high) 1546 panic("0x%lx byte mapping at 0x%p exceeds boot's " 1547 "legal range.", len, (void *)va); 1548 1549 while (len > 0) { 1550 pp = page_numtopp_alloc(pfn); 1551 if (pp != NULL) { 1552 if (setaside == 0) 1553 panic("Unexpected mapping by boot. " 1554 "addr=%p pfn=%lx\n", 1555 (void *)va, pfn); 1556 1557 pp->p_next = bootpages; 1558 pp->p_prev = NULL; 1559 PP_SETBOOTPAGES(pp); 1560 if (bootpages != NULL) { 1561 bootpages->p_prev = pp; 1562 } 1563 bootpages = pp; 1564 ++boot_protect_cnt; 1565 } 1566 1567 ++pfn; 1568 len -= MMU_PAGESIZE; 1569 va += MMU_PAGESIZE; 1570 } 1571 } 1572 PRM_DEBUG(boot_protect_cnt); 1573 } 1574 1575 /* 1576 * 1577 */ 1578 static void 1579 layout_kernel_va(void) 1580 { 1581 PRM_POINT("layout_kernel_va() starting..."); 1582 /* 1583 * Establish the final size of the kernel's heap, size of segmap, 1584 * segkp, etc. 1585 */ 1586 1587 #if defined(__amd64) 1588 1589 kpm_vbase = (caddr_t)segkpm_base; 1590 kpm_size = ROUND_UP_LPAGE(mmu_ptob(physmax + 1)); 1591 if ((uintptr_t)kpm_vbase + kpm_size > (uintptr_t)valloc_base) 1592 panic("not enough room for kpm!"); 1593 PRM_DEBUG(kpm_size); 1594 PRM_DEBUG(kpm_vbase); 1595 1596 /* 1597 * By default we create a seg_kp in 64 bit kernels, it's a little 1598 * faster to access than embedding it in the heap. 1599 */ 1600 segkp_base = (caddr_t)valloc_base + valloc_sz; 1601 if (!segkp_fromheap) { 1602 size_t sz = mmu_ptob(segkpsize); 1603 1604 /* 1605 * determine size of segkp 1606 */ 1607 if (sz < SEGKPMINSIZE || sz > SEGKPMAXSIZE) { 1608 sz = SEGKPDEFSIZE; 1609 cmn_err(CE_WARN, "!Illegal value for segkpsize. " 1610 "segkpsize has been reset to %ld pages", 1611 mmu_btop(sz)); 1612 } 1613 sz = MIN(sz, MAX(SEGKPMINSIZE, mmu_ptob(physmem))); 1614 1615 segkpsize = mmu_btop(ROUND_UP_LPAGE(sz)); 1616 } 1617 PRM_DEBUG(segkp_base); 1618 PRM_DEBUG(segkpsize); 1619 1620 /* 1621 * segzio is used for ZFS cached data. It uses a distinct VA 1622 * segment (from kernel heap) so that we can easily tell not to 1623 * include it in kernel crash dumps on 64 bit kernels. The trick is 1624 * to give it lots of VA, but not constrain the kernel heap. 1625 * We scale the size of segzio linearly with physmem up to 1626 * SEGZIOMAXSIZE. Above that amount it scales at 50% of physmem. 1627 */ 1628 segzio_base = segkp_base + mmu_ptob(segkpsize); 1629 if (segzio_fromheap) { 1630 segziosize = 0; 1631 } else { 1632 size_t physmem_size = mmu_ptob(physmem); 1633 size_t size = (segziosize == 0) ? 1634 physmem_size : mmu_ptob(segziosize); 1635 1636 if (size < SEGZIOMINSIZE) 1637 size = SEGZIOMINSIZE; 1638 if (size > SEGZIOMAXSIZE) { 1639 size = SEGZIOMAXSIZE; 1640 if (physmem_size > size) 1641 size += (physmem_size - size) / 2; 1642 } 1643 segziosize = mmu_btop(ROUND_UP_LPAGE(size)); 1644 } 1645 PRM_DEBUG(segziosize); 1646 PRM_DEBUG(segzio_base); 1647 1648 /* 1649 * Put the range of VA for device mappings next, kmdb knows to not 1650 * grep in this range of addresses. 1651 */ 1652 toxic_addr = 1653 ROUND_UP_LPAGE((uintptr_t)segzio_base + mmu_ptob(segziosize)); 1654 PRM_DEBUG(toxic_addr); 1655 segmap_start = ROUND_UP_LPAGE(toxic_addr + toxic_size); 1656 #else /* __i386 */ 1657 segmap_start = ROUND_UP_LPAGE(kernelbase); 1658 #endif /* __i386 */ 1659 PRM_DEBUG(segmap_start); 1660 1661 /* 1662 * Users can change segmapsize through eeprom or /etc/system. 1663 * If the variable is tuned through eeprom, there is no upper 1664 * bound on the size of segmap. If it is tuned through 1665 * /etc/system on 32-bit systems, it must be no larger than we 1666 * planned for in startup_memlist(). 1667 */ 1668 segmapsize = MAX(ROUND_UP_LPAGE(segmapsize), SEGMAPDEFAULT); 1669 1670 #if defined(__i386) 1671 /* 1672 * 32-bit systems don't have segkpm or segkp, so segmap appears at 1673 * the bottom of the kernel's address range. Set aside space for a 1674 * small red zone just below the start of segmap. 1675 */ 1676 segmap_start += KERNEL_REDZONE_SIZE; 1677 segmapsize -= KERNEL_REDZONE_SIZE; 1678 #endif 1679 1680 PRM_DEBUG(segmap_start); 1681 PRM_DEBUG(segmapsize); 1682 kernelheap = (caddr_t)ROUND_UP_LPAGE(segmap_start + segmapsize); 1683 PRM_DEBUG(kernelheap); 1684 PRM_POINT("layout_kernel_va() done..."); 1685 } 1686 1687 /* 1688 * Finish initializing the VM system, now that we are no longer 1689 * relying on the boot time memory allocators. 1690 */ 1691 static void 1692 startup_vm(void) 1693 { 1694 struct segmap_crargs a; 1695 1696 extern int use_brk_lpg, use_stk_lpg; 1697 1698 PRM_POINT("startup_vm() starting..."); 1699 1700 /* 1701 * Initialize the hat layer. 1702 */ 1703 hat_init(); 1704 1705 /* 1706 * Do final allocations of HAT data structures that need to 1707 * be allocated before quiescing the boot loader. 1708 */ 1709 PRM_POINT("Calling hat_kern_alloc()..."); 1710 hat_kern_alloc((caddr_t)segmap_start, segmapsize, ekernelheap); 1711 PRM_POINT("hat_kern_alloc() done"); 1712 1713 #ifndef __xpv 1714 /* 1715 * Setup Page Attribute Table 1716 */ 1717 pat_sync(); 1718 #endif 1719 1720 /* 1721 * The next two loops are done in distinct steps in order 1722 * to be sure that any page that is doubly mapped (both above 1723 * KERNEL_TEXT and below kernelbase) is dealt with correctly. 1724 * Note this may never happen, but it might someday. 1725 */ 1726 bootpages = NULL; 1727 PRM_POINT("Protecting boot pages"); 1728 1729 /* 1730 * Protect any pages mapped above KERNEL_TEXT that somehow have 1731 * page_t's. This can only happen if something weird allocated 1732 * in this range (like kadb/kmdb). 1733 */ 1734 protect_boot_range(KERNEL_TEXT, (uintptr_t)-1, 0); 1735 1736 /* 1737 * Before we can take over memory allocation/mapping from the boot 1738 * loader we must remove from our free page lists any boot allocated 1739 * pages that stay mapped until release_bootstrap(). 1740 */ 1741 protect_boot_range(0, kernelbase, 1); 1742 1743 1744 /* 1745 * Switch to running on regular HAT (not boot_mmu) 1746 */ 1747 PRM_POINT("Calling hat_kern_setup()..."); 1748 hat_kern_setup(); 1749 1750 /* 1751 * It is no longer safe to call BOP_ALLOC(), so make sure we don't. 1752 */ 1753 bop_no_more_mem(); 1754 1755 PRM_POINT("hat_kern_setup() done"); 1756 1757 hat_cpu_online(CPU); 1758 1759 /* 1760 * Initialize VM system 1761 */ 1762 PRM_POINT("Calling kvm_init()..."); 1763 kvm_init(); 1764 PRM_POINT("kvm_init() done"); 1765 1766 /* 1767 * Tell kmdb that the VM system is now working 1768 */ 1769 if (boothowto & RB_DEBUG) 1770 kdi_dvec_vmready(); 1771 1772 #if defined(__xpv) 1773 /* 1774 * Populate the I/O pool on domain 0 1775 */ 1776 if (DOMAIN_IS_INITDOMAIN(xen_info)) { 1777 extern long populate_io_pool(void); 1778 long init_io_pool_cnt; 1779 1780 PRM_POINT("Populating reserve I/O page pool"); 1781 init_io_pool_cnt = populate_io_pool(); 1782 PRM_DEBUG(init_io_pool_cnt); 1783 } 1784 #endif 1785 /* 1786 * Mangle the brand string etc. 1787 */ 1788 cpuid_pass3(CPU); 1789 1790 #if defined(__amd64) 1791 1792 /* 1793 * Create the device arena for toxic (to dtrace/kmdb) mappings. 1794 */ 1795 device_arena = vmem_create("device", (void *)toxic_addr, 1796 toxic_size, MMU_PAGESIZE, NULL, NULL, NULL, 0, VM_SLEEP); 1797 1798 #else /* __i386 */ 1799 1800 /* 1801 * allocate the bit map that tracks toxic pages 1802 */ 1803 toxic_bit_map_len = btop((ulong_t)(valloc_base - kernelbase)); 1804 PRM_DEBUG(toxic_bit_map_len); 1805 toxic_bit_map = 1806 kmem_zalloc(BT_SIZEOFMAP(toxic_bit_map_len), KM_NOSLEEP); 1807 ASSERT(toxic_bit_map != NULL); 1808 PRM_DEBUG(toxic_bit_map); 1809 1810 #endif /* __i386 */ 1811 1812 1813 /* 1814 * Now that we've got more VA, as well as the ability to allocate from 1815 * it, tell the debugger. 1816 */ 1817 if (boothowto & RB_DEBUG) 1818 kdi_dvec_memavail(); 1819 1820 /* 1821 * The following code installs a special page fault handler (#pf) 1822 * to work around a pentium bug. 1823 */ 1824 #if !defined(__amd64) && !defined(__xpv) 1825 if (x86_type == X86_TYPE_P5) { 1826 desctbr_t idtr; 1827 gate_desc_t *newidt; 1828 1829 if ((newidt = kmem_zalloc(MMU_PAGESIZE, KM_NOSLEEP)) == NULL) 1830 panic("failed to install pentium_pftrap"); 1831 1832 bcopy(idt0, newidt, NIDT * sizeof (*idt0)); 1833 set_gatesegd(&newidt[T_PGFLT], &pentium_pftrap, 1834 KCS_SEL, SDT_SYSIGT, TRP_KPL); 1835 1836 (void) as_setprot(&kas, (caddr_t)newidt, MMU_PAGESIZE, 1837 PROT_READ | PROT_EXEC); 1838 1839 CPU->cpu_idt = newidt; 1840 idtr.dtr_base = (uintptr_t)CPU->cpu_idt; 1841 idtr.dtr_limit = (NIDT * sizeof (*idt0)) - 1; 1842 wr_idtr(&idtr); 1843 } 1844 #endif /* !__amd64 */ 1845 1846 #if !defined(__xpv) 1847 /* 1848 * Map page pfn=0 for drivers, such as kd, that need to pick up 1849 * parameters left there by controllers/BIOS. 1850 */ 1851 PRM_POINT("setup up p0_va"); 1852 p0_va = i86devmap(0, 1, PROT_READ); 1853 PRM_DEBUG(p0_va); 1854 #endif 1855 1856 cmn_err(CE_CONT, "?mem = %luK (0x%lx)\n", 1857 physinstalled << (MMU_PAGESHIFT - 10), ptob(physinstalled)); 1858 1859 /* 1860 * disable automatic large pages for small memory systems or 1861 * when the disable flag is set. 1862 * 1863 * Do not yet consider page sizes larger than 2m/4m. 1864 */ 1865 if (!auto_lpg_disable && mmu.max_page_level > 0) { 1866 max_uheap_lpsize = LEVEL_SIZE(1); 1867 max_ustack_lpsize = LEVEL_SIZE(1); 1868 max_privmap_lpsize = LEVEL_SIZE(1); 1869 max_uidata_lpsize = LEVEL_SIZE(1); 1870 max_utext_lpsize = LEVEL_SIZE(1); 1871 max_shm_lpsize = LEVEL_SIZE(1); 1872 } 1873 if (physmem < privm_lpg_min_physmem || mmu.max_page_level == 0 || 1874 auto_lpg_disable) { 1875 use_brk_lpg = 0; 1876 use_stk_lpg = 0; 1877 } 1878 mcntl0_lpsize = LEVEL_SIZE(mmu.umax_page_level); 1879 1880 PRM_POINT("Calling hat_init_finish()..."); 1881 hat_init_finish(); 1882 PRM_POINT("hat_init_finish() done"); 1883 1884 /* 1885 * Initialize the segkp segment type. 1886 */ 1887 rw_enter(&kas.a_lock, RW_WRITER); 1888 PRM_POINT("Attaching segkp"); 1889 if (segkp_fromheap) { 1890 segkp->s_as = &kas; 1891 } else if (seg_attach(&kas, (caddr_t)segkp_base, mmu_ptob(segkpsize), 1892 segkp) < 0) { 1893 panic("startup: cannot attach segkp"); 1894 /*NOTREACHED*/ 1895 } 1896 PRM_POINT("Doing segkp_create()"); 1897 if (segkp_create(segkp) != 0) { 1898 panic("startup: segkp_create failed"); 1899 /*NOTREACHED*/ 1900 } 1901 PRM_DEBUG(segkp); 1902 rw_exit(&kas.a_lock); 1903 1904 /* 1905 * kpm segment 1906 */ 1907 segmap_kpm = 0; 1908 if (kpm_desired) { 1909 kpm_init(); 1910 kpm_enable = 1; 1911 vpm_enable = 1; 1912 } 1913 1914 /* 1915 * Now create segmap segment. 1916 */ 1917 rw_enter(&kas.a_lock, RW_WRITER); 1918 if (seg_attach(&kas, (caddr_t)segmap_start, segmapsize, segmap) < 0) { 1919 panic("cannot attach segmap"); 1920 /*NOTREACHED*/ 1921 } 1922 PRM_DEBUG(segmap); 1923 1924 a.prot = PROT_READ | PROT_WRITE; 1925 a.shmsize = 0; 1926 a.nfreelist = segmapfreelists; 1927 1928 if (segmap_create(segmap, (caddr_t)&a) != 0) 1929 panic("segmap_create segmap"); 1930 rw_exit(&kas.a_lock); 1931 1932 setup_vaddr_for_ppcopy(CPU); 1933 1934 segdev_init(); 1935 #if defined(__xpv) 1936 if (DOMAIN_IS_INITDOMAIN(xen_info)) 1937 #endif 1938 pmem_init(); 1939 1940 PRM_POINT("startup_vm() done"); 1941 } 1942 1943 /* 1944 * Load a tod module for the non-standard tod part found on this system. 1945 */ 1946 static void 1947 load_tod_module(char *todmod) 1948 { 1949 if (modload("tod", todmod) == -1) 1950 halt("Can't load TOD module"); 1951 } 1952 1953 static void 1954 startup_end(void) 1955 { 1956 int i; 1957 extern void setx86isalist(void); 1958 1959 PRM_POINT("startup_end() starting..."); 1960 1961 /* 1962 * Perform tasks that get done after most of the VM 1963 * initialization has been done but before the clock 1964 * and other devices get started. 1965 */ 1966 kern_setup1(); 1967 1968 /* 1969 * Perform CPC initialization for this CPU. 1970 */ 1971 kcpc_hw_init(CPU); 1972 1973 #if defined(OPTERON_WORKAROUND_6323525) 1974 if (opteron_workaround_6323525) 1975 patch_workaround_6323525(); 1976 #endif 1977 /* 1978 * If needed, load TOD module now so that ddi_get_time(9F) etc. work 1979 * (For now, "needed" is defined as set tod_module_name in /etc/system) 1980 */ 1981 if (tod_module_name != NULL) { 1982 PRM_POINT("load_tod_module()"); 1983 load_tod_module(tod_module_name); 1984 } 1985 1986 #if defined(__xpv) 1987 /* 1988 * Forceload interposing TOD module for the hypervisor. 1989 */ 1990 PRM_POINT("load_tod_module()"); 1991 load_tod_module("xpvtod"); 1992 #endif 1993 1994 /* 1995 * Configure the system. 1996 */ 1997 PRM_POINT("Calling configure()..."); 1998 configure(); /* set up devices */ 1999 PRM_POINT("configure() done"); 2000 2001 /* 2002 * Set the isa_list string to the defined instruction sets we 2003 * support. 2004 */ 2005 setx86isalist(); 2006 cpu_intr_alloc(CPU, NINTR_THREADS); 2007 psm_install(); 2008 2009 /* 2010 * We're done with bootops. We don't unmap the bootstrap yet because 2011 * we're still using bootsvcs. 2012 */ 2013 PRM_POINT("NULLing out bootops"); 2014 *bootopsp = (struct bootops *)NULL; 2015 bootops = (struct bootops *)NULL; 2016 2017 #if defined(__xpv) 2018 ec_init_debug_irq(); 2019 xs_domu_init(); 2020 #endif 2021 PRM_POINT("Enabling interrupts"); 2022 (*picinitf)(); 2023 sti(); 2024 #if defined(__xpv) 2025 ASSERT(CPU->cpu_m.mcpu_vcpu_info->evtchn_upcall_mask == 0); 2026 xen_late_startup(); 2027 #endif 2028 2029 (void) add_avsoftintr((void *)&softlevel1_hdl, 1, softlevel1, 2030 "softlevel1", NULL, NULL); /* XXX to be moved later */ 2031 2032 /* 2033 * Register these software interrupts for ddi timer. 2034 * Software interrupts up to the level 10 are supported. 2035 */ 2036 for (i = DDI_IPL_1; i <= DDI_IPL_10; i++) { 2037 char name[sizeof ("timer_softintr") + 2]; 2038 (void) sprintf(name, "timer_softintr%02d", i); 2039 (void) add_avsoftintr((void *)&softlevel_hdl[i-1], i, 2040 (avfunc)timer_softintr, name, (caddr_t)(uintptr_t)i, NULL); 2041 } 2042 2043 PRM_POINT("startup_end() done"); 2044 } 2045 2046 extern char hw_serial[]; 2047 char *_hs1107 = hw_serial; 2048 ulong_t _bdhs34; 2049 2050 void 2051 post_startup(void) 2052 { 2053 /* 2054 * Set the system wide, processor-specific flags to be passed 2055 * to userland via the aux vector for performance hints and 2056 * instruction set extensions. 2057 */ 2058 bind_hwcap(); 2059 2060 #ifdef __xpv 2061 if (DOMAIN_IS_INITDOMAIN(xen_info)) 2062 #endif 2063 { 2064 /* 2065 * Load the System Management BIOS into the global ksmbios 2066 * handle, if an SMBIOS is present on this system. 2067 */ 2068 ksmbios = smbios_open(NULL, SMB_VERSION, ksmbios_flags, NULL); 2069 2070 #if defined(__xpv) 2071 xpv_panic_init(); 2072 #else 2073 /* 2074 * Startup the memory scrubber. 2075 * XXPV This should be running somewhere .. 2076 */ 2077 memscrub_init(); 2078 #endif 2079 } 2080 2081 /* 2082 * Complete CPU module initialization 2083 */ 2084 cmi_post_startup(); 2085 2086 /* 2087 * Perform forceloading tasks for /etc/system. 2088 */ 2089 (void) mod_sysctl(SYS_FORCELOAD, NULL); 2090 2091 /* 2092 * ON4.0: Force /proc module in until clock interrupt handle fixed 2093 * ON4.0: This must be fixed or restated in /etc/systems. 2094 */ 2095 (void) modload("fs", "procfs"); 2096 2097 (void) i_ddi_attach_hw_nodes("pit_beep"); 2098 2099 #if defined(__i386) 2100 /* 2101 * Check for required functional Floating Point hardware, 2102 * unless FP hardware explicitly disabled. 2103 */ 2104 if (fpu_exists && (fpu_pentium_fdivbug || fp_kind == FP_NO)) 2105 halt("No working FP hardware found"); 2106 #endif 2107 2108 maxmem = freemem; 2109 2110 add_cpunode2devtree(CPU->cpu_id, CPU->cpu_m.mcpu_cpi); 2111 } 2112 2113 static int 2114 pp_in_ramdisk(page_t *pp) 2115 { 2116 extern uint64_t ramdisk_start, ramdisk_end; 2117 2118 return ((pp->p_pagenum >= btop(ramdisk_start)) && 2119 (pp->p_pagenum < btopr(ramdisk_end))); 2120 } 2121 2122 void 2123 release_bootstrap(void) 2124 { 2125 int root_is_ramdisk; 2126 page_t *pp; 2127 extern void kobj_boot_unmountroot(void); 2128 extern dev_t rootdev; 2129 2130 /* unmount boot ramdisk and release kmem usage */ 2131 kobj_boot_unmountroot(); 2132 2133 /* 2134 * We're finished using the boot loader so free its pages. 2135 */ 2136 PRM_POINT("Unmapping lower boot pages"); 2137 clear_boot_mappings(0, _userlimit); 2138 postbootkernelbase = kernelbase; 2139 2140 /* 2141 * If root isn't on ramdisk, destroy the hardcoded 2142 * ramdisk node now and release the memory. Else, 2143 * ramdisk memory is kept in rd_pages. 2144 */ 2145 root_is_ramdisk = (getmajor(rootdev) == ddi_name_to_major("ramdisk")); 2146 if (!root_is_ramdisk) { 2147 dev_info_t *dip = ddi_find_devinfo("ramdisk", -1, 0); 2148 ASSERT(dip && ddi_get_parent(dip) == ddi_root_node()); 2149 ndi_rele_devi(dip); /* held from ddi_find_devinfo */ 2150 (void) ddi_remove_child(dip, 0); 2151 } 2152 2153 PRM_POINT("Releasing boot pages"); 2154 while (bootpages) { 2155 pp = bootpages; 2156 bootpages = pp->p_next; 2157 if (root_is_ramdisk && pp_in_ramdisk(pp)) { 2158 pp->p_next = rd_pages; 2159 rd_pages = pp; 2160 continue; 2161 } 2162 pp->p_next = (struct page *)0; 2163 pp->p_prev = (struct page *)0; 2164 PP_CLRBOOTPAGES(pp); 2165 page_free(pp, 1); 2166 } 2167 PRM_POINT("Boot pages released"); 2168 2169 #if !defined(__xpv) 2170 /* XXPV -- note this following bunch of code needs to be revisited in Xen 3.0 */ 2171 /* 2172 * Find 1 page below 1 MB so that other processors can boot up. 2173 * Make sure it has a kernel VA as well as a 1:1 mapping. 2174 * We should have just free'd one up. 2175 */ 2176 if (use_mp) { 2177 pfn_t pfn; 2178 2179 for (pfn = 1; pfn < btop(1*1024*1024); pfn++) { 2180 if (page_numtopp_alloc(pfn) == NULL) 2181 continue; 2182 rm_platter_va = i86devmap(pfn, 1, 2183 PROT_READ | PROT_WRITE | PROT_EXEC); 2184 rm_platter_pa = ptob(pfn); 2185 hat_devload(kas.a_hat, 2186 (caddr_t)(uintptr_t)rm_platter_pa, MMU_PAGESIZE, 2187 pfn, PROT_READ | PROT_WRITE | PROT_EXEC, 2188 HAT_LOAD_NOCONSIST); 2189 break; 2190 } 2191 if (pfn == btop(1*1024*1024)) 2192 panic("No page available for starting " 2193 "other processors"); 2194 } 2195 #endif /* !__xpv */ 2196 } 2197 2198 /* 2199 * Initialize the platform-specific parts of a page_t. 2200 */ 2201 void 2202 add_physmem_cb(page_t *pp, pfn_t pnum) 2203 { 2204 pp->p_pagenum = pnum; 2205 pp->p_mapping = NULL; 2206 pp->p_embed = 0; 2207 pp->p_share = 0; 2208 pp->p_mlentry = 0; 2209 } 2210 2211 /* 2212 * kphysm_init() initializes physical memory. 2213 */ 2214 static pgcnt_t 2215 kphysm_init( 2216 page_t *pp, 2217 pgcnt_t npages) 2218 { 2219 struct memlist *pmem; 2220 struct memseg *cur_memseg; 2221 pfn_t base_pfn; 2222 pgcnt_t num; 2223 pgcnt_t pages_done = 0; 2224 uint64_t addr; 2225 uint64_t size; 2226 extern pfn_t ddiphysmin; 2227 2228 ASSERT(page_hash != NULL && page_hashsz != 0); 2229 2230 cur_memseg = memseg_base; 2231 for (pmem = phys_avail; pmem && npages; pmem = pmem->next) { 2232 /* 2233 * In a 32 bit kernel can't use higher memory if we're 2234 * not booting in PAE mode. This check takes care of that. 2235 */ 2236 addr = pmem->address; 2237 size = pmem->size; 2238 if (btop(addr) > physmax) 2239 continue; 2240 2241 /* 2242 * align addr and size - they may not be at page boundaries 2243 */ 2244 if ((addr & MMU_PAGEOFFSET) != 0) { 2245 addr += MMU_PAGEOFFSET; 2246 addr &= ~(uint64_t)MMU_PAGEOFFSET; 2247 size -= addr - pmem->address; 2248 } 2249 2250 /* only process pages below or equal to physmax */ 2251 if ((btop(addr + size) - 1) > physmax) 2252 size = ptob(physmax - btop(addr) + 1); 2253 2254 num = btop(size); 2255 if (num == 0) 2256 continue; 2257 2258 if (num > npages) 2259 num = npages; 2260 2261 npages -= num; 2262 pages_done += num; 2263 base_pfn = btop(addr); 2264 2265 if (prom_debug) 2266 prom_printf("MEMSEG addr=0x%" PRIx64 2267 " pgs=0x%lx pfn 0x%lx-0x%lx\n", 2268 addr, num, base_pfn, base_pfn + num); 2269 2270 /* 2271 * Ignore pages below ddiphysmin to simplify ddi memory 2272 * allocation with non-zero addr_lo requests. 2273 */ 2274 if (base_pfn < ddiphysmin) { 2275 if (base_pfn + num <= ddiphysmin) 2276 continue; 2277 pp += (ddiphysmin - base_pfn); 2278 num -= (ddiphysmin - base_pfn); 2279 base_pfn = ddiphysmin; 2280 } 2281 2282 /* 2283 * Build the memsegs entry 2284 */ 2285 cur_memseg->pages = pp; 2286 cur_memseg->epages = pp + num; 2287 cur_memseg->pages_base = base_pfn; 2288 cur_memseg->pages_end = base_pfn + num; 2289 2290 /* 2291 * Insert into memseg list in decreasing pfn range order. 2292 * Low memory is typically more fragmented such that this 2293 * ordering keeps the larger ranges at the front of the list 2294 * for code that searches memseg. 2295 * This ASSERTS that the memsegs coming in from boot are in 2296 * increasing physical address order and not contiguous. 2297 */ 2298 if (memsegs != NULL) { 2299 ASSERT(cur_memseg->pages_base >= memsegs->pages_end); 2300 cur_memseg->next = memsegs; 2301 } 2302 memsegs = cur_memseg; 2303 2304 /* 2305 * add_physmem() initializes the PSM part of the page 2306 * struct by calling the PSM back with add_physmem_cb(). 2307 * In addition it coalesces pages into larger pages as 2308 * it initializes them. 2309 */ 2310 add_physmem(pp, num, base_pfn); 2311 cur_memseg++; 2312 availrmem_initial += num; 2313 availrmem += num; 2314 2315 pp += num; 2316 } 2317 2318 PRM_DEBUG(availrmem_initial); 2319 PRM_DEBUG(availrmem); 2320 PRM_DEBUG(freemem); 2321 build_pfn_hash(); 2322 return (pages_done); 2323 } 2324 2325 /* 2326 * Kernel VM initialization. 2327 */ 2328 static void 2329 kvm_init(void) 2330 { 2331 ASSERT((((uintptr_t)s_text) & MMU_PAGEOFFSET) == 0); 2332 2333 /* 2334 * Put the kernel segments in kernel address space. 2335 */ 2336 rw_enter(&kas.a_lock, RW_WRITER); 2337 as_avlinit(&kas); 2338 2339 (void) seg_attach(&kas, s_text, e_moddata - s_text, &ktextseg); 2340 (void) segkmem_create(&ktextseg); 2341 2342 (void) seg_attach(&kas, (caddr_t)valloc_base, valloc_sz, &kvalloc); 2343 (void) segkmem_create(&kvalloc); 2344 2345 (void) seg_attach(&kas, kernelheap, 2346 ekernelheap - kernelheap, &kvseg); 2347 (void) segkmem_create(&kvseg); 2348 2349 if (core_size > 0) { 2350 PRM_POINT("attaching kvseg_core"); 2351 (void) seg_attach(&kas, (caddr_t)core_base, core_size, 2352 &kvseg_core); 2353 (void) segkmem_create(&kvseg_core); 2354 } 2355 2356 if (segziosize > 0) { 2357 PRM_POINT("attaching segzio"); 2358 (void) seg_attach(&kas, segzio_base, mmu_ptob(segziosize), 2359 &kzioseg); 2360 (void) segkmem_zio_create(&kzioseg); 2361 2362 /* create zio area covering new segment */ 2363 segkmem_zio_init(segzio_base, mmu_ptob(segziosize)); 2364 } 2365 2366 (void) seg_attach(&kas, kdi_segdebugbase, kdi_segdebugsize, &kdebugseg); 2367 (void) segkmem_create(&kdebugseg); 2368 2369 rw_exit(&kas.a_lock); 2370 2371 /* 2372 * Ensure that the red zone at kernelbase is never accessible. 2373 */ 2374 PRM_POINT("protecting redzone"); 2375 (void) as_setprot(&kas, (caddr_t)kernelbase, KERNEL_REDZONE_SIZE, 0); 2376 2377 /* 2378 * Make the text writable so that it can be hot patched by DTrace. 2379 */ 2380 (void) as_setprot(&kas, s_text, e_modtext - s_text, 2381 PROT_READ | PROT_WRITE | PROT_EXEC); 2382 2383 /* 2384 * Make data writable until end. 2385 */ 2386 (void) as_setprot(&kas, s_data, e_moddata - s_data, 2387 PROT_READ | PROT_WRITE | PROT_EXEC); 2388 } 2389 2390 #ifndef __xpv 2391 /* 2392 * Solaris adds an entry for Write Combining caching to the PAT 2393 */ 2394 static uint64_t pat_attr_reg = PAT_DEFAULT_ATTRIBUTE; 2395 2396 void 2397 pat_sync(void) 2398 { 2399 ulong_t cr0, cr0_orig, cr4; 2400 2401 if (!(x86_feature & X86_PAT)) 2402 return; 2403 cr0_orig = cr0 = getcr0(); 2404 cr4 = getcr4(); 2405 2406 /* disable caching and flush all caches and TLBs */ 2407 cr0 |= CR0_CD; 2408 cr0 &= ~CR0_NW; 2409 setcr0(cr0); 2410 invalidate_cache(); 2411 if (cr4 & CR4_PGE) { 2412 setcr4(cr4 & ~(ulong_t)CR4_PGE); 2413 setcr4(cr4); 2414 } else { 2415 reload_cr3(); 2416 } 2417 2418 /* add our entry to the PAT */ 2419 wrmsr(REG_PAT, pat_attr_reg); 2420 2421 /* flush TLBs and cache again, then reenable cr0 caching */ 2422 if (cr4 & CR4_PGE) { 2423 setcr4(cr4 & ~(ulong_t)CR4_PGE); 2424 setcr4(cr4); 2425 } else { 2426 reload_cr3(); 2427 } 2428 invalidate_cache(); 2429 setcr0(cr0_orig); 2430 } 2431 2432 #endif /* !__xpv */ 2433 2434 void 2435 get_system_configuration(void) 2436 { 2437 char prop[32]; 2438 u_longlong_t nodes_ll, cpus_pernode_ll, lvalue; 2439 2440 if (BOP_GETPROPLEN(bootops, "nodes") > sizeof (prop) || 2441 BOP_GETPROP(bootops, "nodes", prop) < 0 || 2442 kobj_getvalue(prop, &nodes_ll) == -1 || 2443 nodes_ll > MAXNODES || 2444 BOP_GETPROPLEN(bootops, "cpus_pernode") > sizeof (prop) || 2445 BOP_GETPROP(bootops, "cpus_pernode", prop) < 0 || 2446 kobj_getvalue(prop, &cpus_pernode_ll) == -1) { 2447 system_hardware.hd_nodes = 1; 2448 system_hardware.hd_cpus_per_node = 0; 2449 } else { 2450 system_hardware.hd_nodes = (int)nodes_ll; 2451 system_hardware.hd_cpus_per_node = (int)cpus_pernode_ll; 2452 } 2453 2454 if (BOP_GETPROPLEN(bootops, "kernelbase") > sizeof (prop) || 2455 BOP_GETPROP(bootops, "kernelbase", prop) < 0 || 2456 kobj_getvalue(prop, &lvalue) == -1) 2457 eprom_kernelbase = NULL; 2458 else 2459 eprom_kernelbase = (uintptr_t)lvalue; 2460 2461 if (BOP_GETPROPLEN(bootops, "segmapsize") > sizeof (prop) || 2462 BOP_GETPROP(bootops, "segmapsize", prop) < 0 || 2463 kobj_getvalue(prop, &lvalue) == -1) 2464 segmapsize = SEGMAPDEFAULT; 2465 else 2466 segmapsize = (uintptr_t)lvalue; 2467 2468 if (BOP_GETPROPLEN(bootops, "segmapfreelists") > sizeof (prop) || 2469 BOP_GETPROP(bootops, "segmapfreelists", prop) < 0 || 2470 kobj_getvalue(prop, &lvalue) == -1) 2471 segmapfreelists = 0; /* use segmap driver default */ 2472 else 2473 segmapfreelists = (int)lvalue; 2474 2475 /* physmem used to be here, but moved much earlier to fakebop.c */ 2476 } 2477 2478 /* 2479 * Add to a memory list. 2480 * start = start of new memory segment 2481 * len = length of new memory segment in bytes 2482 * new = pointer to a new struct memlist 2483 * memlistp = memory list to which to add segment. 2484 */ 2485 void 2486 memlist_add( 2487 uint64_t start, 2488 uint64_t len, 2489 struct memlist *new, 2490 struct memlist **memlistp) 2491 { 2492 struct memlist *cur; 2493 uint64_t end = start + len; 2494 2495 new->address = start; 2496 new->size = len; 2497 2498 cur = *memlistp; 2499 2500 while (cur) { 2501 if (cur->address >= end) { 2502 new->next = cur; 2503 *memlistp = new; 2504 new->prev = cur->prev; 2505 cur->prev = new; 2506 return; 2507 } 2508 ASSERT(cur->address + cur->size <= start); 2509 if (cur->next == NULL) { 2510 cur->next = new; 2511 new->prev = cur; 2512 new->next = NULL; 2513 return; 2514 } 2515 memlistp = &cur->next; 2516 cur = cur->next; 2517 } 2518 } 2519 2520 void 2521 kobj_vmem_init(vmem_t **text_arena, vmem_t **data_arena) 2522 { 2523 size_t tsize = e_modtext - modtext; 2524 size_t dsize = e_moddata - moddata; 2525 2526 *text_arena = vmem_create("module_text", tsize ? modtext : NULL, tsize, 2527 1, segkmem_alloc, segkmem_free, heaptext_arena, 0, VM_SLEEP); 2528 *data_arena = vmem_create("module_data", dsize ? moddata : NULL, dsize, 2529 1, segkmem_alloc, segkmem_free, heap32_arena, 0, VM_SLEEP); 2530 } 2531 2532 caddr_t 2533 kobj_text_alloc(vmem_t *arena, size_t size) 2534 { 2535 return (vmem_alloc(arena, size, VM_SLEEP | VM_BESTFIT)); 2536 } 2537 2538 /*ARGSUSED*/ 2539 caddr_t 2540 kobj_texthole_alloc(caddr_t addr, size_t size) 2541 { 2542 panic("unexpected call to kobj_texthole_alloc()"); 2543 /*NOTREACHED*/ 2544 return (0); 2545 } 2546 2547 /*ARGSUSED*/ 2548 void 2549 kobj_texthole_free(caddr_t addr, size_t size) 2550 { 2551 panic("unexpected call to kobj_texthole_free()"); 2552 } 2553 2554 /* 2555 * This is called just after configure() in startup(). 2556 * 2557 * The ISALIST concept is a bit hopeless on Intel, because 2558 * there's no guarantee of an ever-more-capable processor 2559 * given that various parts of the instruction set may appear 2560 * and disappear between different implementations. 2561 * 2562 * While it would be possible to correct it and even enhance 2563 * it somewhat, the explicit hardware capability bitmask allows 2564 * more flexibility. 2565 * 2566 * So, we just leave this alone. 2567 */ 2568 void 2569 setx86isalist(void) 2570 { 2571 char *tp; 2572 size_t len; 2573 extern char *isa_list; 2574 2575 #define TBUFSIZE 1024 2576 2577 tp = kmem_alloc(TBUFSIZE, KM_SLEEP); 2578 *tp = '\0'; 2579 2580 #if defined(__amd64) 2581 (void) strcpy(tp, "amd64 "); 2582 #endif 2583 2584 switch (x86_vendor) { 2585 case X86_VENDOR_Intel: 2586 case X86_VENDOR_AMD: 2587 case X86_VENDOR_TM: 2588 if (x86_feature & X86_CMOV) { 2589 /* 2590 * Pentium Pro or later 2591 */ 2592 (void) strcat(tp, "pentium_pro"); 2593 (void) strcat(tp, x86_feature & X86_MMX ? 2594 "+mmx pentium_pro " : " "); 2595 } 2596 /*FALLTHROUGH*/ 2597 case X86_VENDOR_Cyrix: 2598 /* 2599 * The Cyrix 6x86 does not have any Pentium features 2600 * accessible while not at privilege level 0. 2601 */ 2602 if (x86_feature & X86_CPUID) { 2603 (void) strcat(tp, "pentium"); 2604 (void) strcat(tp, x86_feature & X86_MMX ? 2605 "+mmx pentium " : " "); 2606 } 2607 break; 2608 default: 2609 break; 2610 } 2611 (void) strcat(tp, "i486 i386 i86"); 2612 len = strlen(tp) + 1; /* account for NULL at end of string */ 2613 isa_list = strcpy(kmem_alloc(len, KM_SLEEP), tp); 2614 kmem_free(tp, TBUFSIZE); 2615 2616 #undef TBUFSIZE 2617 } 2618 2619 2620 #ifdef __amd64 2621 2622 void * 2623 device_arena_alloc(size_t size, int vm_flag) 2624 { 2625 return (vmem_alloc(device_arena, size, vm_flag)); 2626 } 2627 2628 void 2629 device_arena_free(void *vaddr, size_t size) 2630 { 2631 vmem_free(device_arena, vaddr, size); 2632 } 2633 2634 #else /* __i386 */ 2635 2636 void * 2637 device_arena_alloc(size_t size, int vm_flag) 2638 { 2639 caddr_t vaddr; 2640 uintptr_t v; 2641 size_t start; 2642 size_t end; 2643 2644 vaddr = vmem_alloc(heap_arena, size, vm_flag); 2645 if (vaddr == NULL) 2646 return (NULL); 2647 2648 v = (uintptr_t)vaddr; 2649 ASSERT(v >= kernelbase); 2650 ASSERT(v + size <= valloc_base); 2651 2652 start = btop(v - kernelbase); 2653 end = btop(v + size - 1 - kernelbase); 2654 ASSERT(start < toxic_bit_map_len); 2655 ASSERT(end < toxic_bit_map_len); 2656 2657 while (start <= end) { 2658 BT_ATOMIC_SET(toxic_bit_map, start); 2659 ++start; 2660 } 2661 return (vaddr); 2662 } 2663 2664 void 2665 device_arena_free(void *vaddr, size_t size) 2666 { 2667 uintptr_t v = (uintptr_t)vaddr; 2668 size_t start; 2669 size_t end; 2670 2671 ASSERT(v >= kernelbase); 2672 ASSERT(v + size <= valloc_base); 2673 2674 start = btop(v - kernelbase); 2675 end = btop(v + size - 1 - kernelbase); 2676 ASSERT(start < toxic_bit_map_len); 2677 ASSERT(end < toxic_bit_map_len); 2678 2679 while (start <= end) { 2680 ASSERT(BT_TEST(toxic_bit_map, start) != 0); 2681 BT_ATOMIC_CLEAR(toxic_bit_map, start); 2682 ++start; 2683 } 2684 vmem_free(heap_arena, vaddr, size); 2685 } 2686 2687 /* 2688 * returns 1st address in range that is in device arena, or NULL 2689 * if len is not NULL it returns the length of the toxic range 2690 */ 2691 void * 2692 device_arena_contains(void *vaddr, size_t size, size_t *len) 2693 { 2694 uintptr_t v = (uintptr_t)vaddr; 2695 uintptr_t eaddr = v + size; 2696 size_t start; 2697 size_t end; 2698 2699 /* 2700 * if called very early by kmdb, just return NULL 2701 */ 2702 if (toxic_bit_map == NULL) 2703 return (NULL); 2704 2705 /* 2706 * First check if we're completely outside the bitmap range. 2707 */ 2708 if (v >= valloc_base || eaddr < kernelbase) 2709 return (NULL); 2710 2711 /* 2712 * Trim ends of search to look at only what the bitmap covers. 2713 */ 2714 if (v < kernelbase) 2715 v = kernelbase; 2716 start = btop(v - kernelbase); 2717 end = btop(eaddr - kernelbase); 2718 if (end >= toxic_bit_map_len) 2719 end = toxic_bit_map_len; 2720 2721 if (bt_range(toxic_bit_map, &start, &end, end) == 0) 2722 return (NULL); 2723 2724 v = kernelbase + ptob(start); 2725 if (len != NULL) 2726 *len = ptob(end - start); 2727 return ((void *)v); 2728 } 2729 2730 #endif /* __i386 */ 2731