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 2008 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 24 */ 25 26 #include <sys/types.h> 27 #include <sys/t_lock.h> 28 #include <sys/param.h> 29 #include <sys/sysmacros.h> 30 #include <sys/signal.h> 31 #include <sys/systm.h> 32 #include <sys/user.h> 33 #include <sys/mman.h> 34 #include <sys/vm.h> 35 #include <sys/conf.h> 36 #include <sys/avintr.h> 37 #include <sys/autoconf.h> 38 #include <sys/disp.h> 39 #include <sys/class.h> 40 #include <sys/bitmap.h> 41 42 #include <sys/privregs.h> 43 44 #include <sys/proc.h> 45 #include <sys/buf.h> 46 #include <sys/kmem.h> 47 #include <sys/mem.h> 48 #include <sys/kstat.h> 49 50 #include <sys/reboot.h> 51 52 #include <sys/cred.h> 53 #include <sys/vnode.h> 54 #include <sys/file.h> 55 56 #include <sys/procfs.h> 57 58 #include <sys/vfs.h> 59 #include <sys/cmn_err.h> 60 #include <sys/utsname.h> 61 #include <sys/debug.h> 62 #include <sys/kdi.h> 63 64 #include <sys/dumphdr.h> 65 #include <sys/bootconf.h> 66 #include <sys/varargs.h> 67 #include <sys/promif.h> 68 #include <sys/modctl.h> 69 70 #include <sys/sunddi.h> 71 #include <sys/sunndi.h> 72 #include <sys/ndi_impldefs.h> 73 #include <sys/ddidmareq.h> 74 #include <sys/psw.h> 75 #include <sys/regset.h> 76 #include <sys/clock.h> 77 #include <sys/pte.h> 78 #include <sys/tss.h> 79 #include <sys/stack.h> 80 #include <sys/trap.h> 81 #include <sys/fp.h> 82 #include <vm/kboot_mmu.h> 83 #include <vm/anon.h> 84 #include <vm/as.h> 85 #include <vm/page.h> 86 #include <vm/seg.h> 87 #include <vm/seg_dev.h> 88 #include <vm/seg_kmem.h> 89 #include <vm/seg_kpm.h> 90 #include <vm/seg_map.h> 91 #include <vm/seg_vn.h> 92 #include <vm/seg_kp.h> 93 #include <sys/memnode.h> 94 #include <vm/vm_dep.h> 95 #include <sys/thread.h> 96 #include <sys/sysconf.h> 97 #include <sys/vm_machparam.h> 98 #include <sys/archsystm.h> 99 #include <sys/machsystm.h> 100 #include <vm/hat.h> 101 #include <vm/hat_i86.h> 102 #include <sys/pmem.h> 103 #include <sys/smp_impldefs.h> 104 #include <sys/x86_archext.h> 105 #include <sys/cpuvar.h> 106 #include <sys/segments.h> 107 #include <sys/clconf.h> 108 #include <sys/kobj.h> 109 #include <sys/kobj_lex.h> 110 #include <sys/cpc_impl.h> 111 #include <sys/x86_archext.h> 112 #include <sys/cpu_module.h> 113 #include <sys/smbios.h> 114 #include <sys/debug_info.h> 115 #include <sys/bootinfo.h> 116 #include <sys/ddi_timer.h> 117 118 #ifdef __xpv 119 120 #include <sys/hypervisor.h> 121 #include <sys/xen_mmu.h> 122 #include <sys/evtchn_impl.h> 123 #include <sys/gnttab.h> 124 #include <sys/xpv_panic.h> 125 #include <xen/sys/xenbus_comms.h> 126 #include <xen/public/physdev.h> 127 128 extern void xen_late_startup(void); 129 130 struct xen_evt_data cpu0_evt_data; 131 132 #endif /* __xpv */ 133 134 extern void progressbar_init(void); 135 extern void progressbar_start(void); 136 extern void brand_init(void); 137 extern void pcf_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 extern cpuset_t cpu_ready_set; 655 656 /* 657 * Make sure that nobody tries to use sekpm until we have 658 * initialized it properly. 659 */ 660 #if defined(__amd64) 661 kpm_desired = 1; 662 #endif 663 kpm_enable = 0; 664 CPUSET_ONLY(cpu_ready_set, 0); /* cpu 0 is boot cpu */ 665 666 #if defined(__xpv) /* XXPV fix me! */ 667 { 668 extern int segvn_use_regions; 669 segvn_use_regions = 0; 670 } 671 #endif 672 progressbar_init(); 673 startup_init(); 674 #if defined(__xpv) 675 startup_xen_version(); 676 #endif 677 startup_memlist(); 678 startup_kmem(); 679 startup_vm(); 680 #if !defined(__xpv) 681 startup_pci_bios(); 682 #endif 683 startup_modules(); 684 #if !defined(__xpv) 685 startup_bios_disk(); 686 #endif 687 startup_end(); 688 progressbar_start(); 689 } 690 691 static void 692 startup_init() 693 { 694 PRM_POINT("startup_init() starting..."); 695 696 /* 697 * Complete the extraction of cpuid data 698 */ 699 cpuid_pass2(CPU); 700 701 (void) check_boot_version(BOP_GETVERSION(bootops)); 702 703 /* 704 * Check for prom_debug in boot environment 705 */ 706 if (BOP_GETPROPLEN(bootops, "prom_debug") >= 0) { 707 ++prom_debug; 708 PRM_POINT("prom_debug found in boot enviroment"); 709 } 710 711 /* 712 * Collect node, cpu and memory configuration information. 713 */ 714 get_system_configuration(); 715 716 /* 717 * Halt if this is an unsupported processor. 718 */ 719 if (x86_type == X86_TYPE_486 || x86_type == X86_TYPE_CYRIX_486) { 720 printf("\n486 processor (\"%s\") detected.\n", 721 CPU->cpu_brandstr); 722 halt("This processor is not supported by this release " 723 "of Solaris."); 724 } 725 726 PRM_POINT("startup_init() done"); 727 } 728 729 /* 730 * Callback for copy_memlist_filter() to filter nucleus, kadb/kmdb, (ie. 731 * everything mapped above KERNEL_TEXT) pages from phys_avail. Note it 732 * also filters out physical page zero. There is some reliance on the 733 * boot loader allocating only a few contiguous physical memory chunks. 734 */ 735 static void 736 avail_filter(uint64_t *addr, uint64_t *size) 737 { 738 uintptr_t va; 739 uintptr_t next_va; 740 pfn_t pfn; 741 uint64_t pfn_addr; 742 uint64_t pfn_eaddr; 743 uint_t prot; 744 size_t len; 745 uint_t change; 746 747 if (prom_debug) 748 prom_printf("\tFilter: in: a=%" PRIx64 ", s=%" PRIx64 "\n", 749 *addr, *size); 750 751 /* 752 * page zero is required for BIOS.. never make it available 753 */ 754 if (*addr == 0) { 755 *addr += MMU_PAGESIZE; 756 *size -= MMU_PAGESIZE; 757 } 758 759 /* 760 * First we trim from the front of the range. Since kbm_probe() 761 * walks ranges in virtual order, but addr/size are physical, we need 762 * to the list until no changes are seen. This deals with the case 763 * where page "p" is mapped at v, page "p + PAGESIZE" is mapped at w 764 * but w < v. 765 */ 766 do { 767 change = 0; 768 for (va = KERNEL_TEXT; 769 *size > 0 && kbm_probe(&va, &len, &pfn, &prot) != 0; 770 va = next_va) { 771 772 next_va = va + len; 773 pfn_addr = pfn_to_pa(pfn); 774 pfn_eaddr = pfn_addr + len; 775 776 if (pfn_addr <= *addr && pfn_eaddr > *addr) { 777 change = 1; 778 while (*size > 0 && len > 0) { 779 *addr += MMU_PAGESIZE; 780 *size -= MMU_PAGESIZE; 781 len -= MMU_PAGESIZE; 782 } 783 } 784 } 785 if (change && prom_debug) 786 prom_printf("\t\ttrim: a=%" PRIx64 ", s=%" PRIx64 "\n", 787 *addr, *size); 788 } while (change); 789 790 /* 791 * Trim pages from the end of the range. 792 */ 793 for (va = KERNEL_TEXT; 794 *size > 0 && kbm_probe(&va, &len, &pfn, &prot) != 0; 795 va = next_va) { 796 797 next_va = va + len; 798 pfn_addr = pfn_to_pa(pfn); 799 800 if (pfn_addr >= *addr && pfn_addr < *addr + *size) 801 *size = pfn_addr - *addr; 802 } 803 804 if (prom_debug) 805 prom_printf("\tFilter out: a=%" PRIx64 ", s=%" PRIx64 "\n", 806 *addr, *size); 807 } 808 809 static void 810 kpm_init() 811 { 812 struct segkpm_crargs b; 813 814 /* 815 * These variables were all designed for sfmmu in which segkpm is 816 * mapped using a single pagesize - either 8KB or 4MB. On x86, we 817 * might use 2+ page sizes on a single machine, so none of these 818 * variables have a single correct value. They are set up as if we 819 * always use a 4KB pagesize, which should do no harm. In the long 820 * run, we should get rid of KPM's assumption that only a single 821 * pagesize is used. 822 */ 823 kpm_pgshft = MMU_PAGESHIFT; 824 kpm_pgsz = MMU_PAGESIZE; 825 kpm_pgoff = MMU_PAGEOFFSET; 826 kpmp2pshft = 0; 827 kpmpnpgs = 1; 828 ASSERT(((uintptr_t)kpm_vbase & (kpm_pgsz - 1)) == 0); 829 830 PRM_POINT("about to create segkpm"); 831 rw_enter(&kas.a_lock, RW_WRITER); 832 833 if (seg_attach(&kas, kpm_vbase, kpm_size, segkpm) < 0) 834 panic("cannot attach segkpm"); 835 836 b.prot = PROT_READ | PROT_WRITE; 837 b.nvcolors = 1; 838 839 if (segkpm_create(segkpm, (caddr_t)&b) != 0) 840 panic("segkpm_create segkpm"); 841 842 rw_exit(&kas.a_lock); 843 } 844 845 /* 846 * The debug info page provides enough information to allow external 847 * inspectors (e.g. when running under a hypervisor) to bootstrap 848 * themselves into allowing full-blown kernel debugging. 849 */ 850 static void 851 init_debug_info(void) 852 { 853 caddr_t mem; 854 debug_info_t *di; 855 856 #ifndef __lint 857 ASSERT(sizeof (debug_info_t) < MMU_PAGESIZE); 858 #endif 859 860 mem = BOP_ALLOC(bootops, (caddr_t)DEBUG_INFO_VA, MMU_PAGESIZE, 861 MMU_PAGESIZE); 862 863 if (mem != (caddr_t)DEBUG_INFO_VA) 864 panic("BOP_ALLOC() failed"); 865 bzero(mem, MMU_PAGESIZE); 866 867 di = (debug_info_t *)mem; 868 869 di->di_magic = DEBUG_INFO_MAGIC; 870 di->di_version = DEBUG_INFO_VERSION; 871 di->di_modules = (uintptr_t)&modules; 872 di->di_s_text = (uintptr_t)s_text; 873 di->di_e_text = (uintptr_t)e_text; 874 di->di_s_data = (uintptr_t)s_data; 875 di->di_e_data = (uintptr_t)e_data; 876 di->di_hat_htable_off = offsetof(hat_t, hat_htable); 877 di->di_ht_pfn_off = offsetof(htable_t, ht_pfn); 878 } 879 880 /* 881 * Build the memlists and other kernel essential memory system data structures. 882 * This is everything at valloc_base. 883 */ 884 static void 885 startup_memlist(void) 886 { 887 size_t memlist_sz; 888 size_t memseg_sz; 889 size_t pagehash_sz; 890 size_t pp_sz; 891 uintptr_t va; 892 size_t len; 893 uint_t prot; 894 pfn_t pfn; 895 int memblocks; 896 caddr_t pagecolor_mem; 897 size_t pagecolor_memsz; 898 caddr_t page_ctrs_mem; 899 size_t page_ctrs_size; 900 size_t pse_table_alloc_size; 901 struct memlist *current; 902 extern void startup_build_mem_nodes(struct memlist *); 903 904 /* XX64 fix these - they should be in include files */ 905 extern size_t page_coloring_init(uint_t, int, int); 906 extern void page_coloring_setup(caddr_t); 907 908 PRM_POINT("startup_memlist() starting..."); 909 910 /* 911 * Use leftover large page nucleus text/data space for loadable modules. 912 * Use at most MODTEXT/MODDATA. 913 */ 914 len = kbm_nucleus_size; 915 ASSERT(len > MMU_PAGESIZE); 916 917 moddata = (caddr_t)ROUND_UP_PAGE(e_data); 918 e_moddata = (caddr_t)P2ROUNDUP((uintptr_t)e_data, (uintptr_t)len); 919 if (e_moddata - moddata > MODDATA) 920 e_moddata = moddata + MODDATA; 921 922 modtext = (caddr_t)ROUND_UP_PAGE(e_text); 923 e_modtext = (caddr_t)P2ROUNDUP((uintptr_t)e_text, (uintptr_t)len); 924 if (e_modtext - modtext > MODTEXT) 925 e_modtext = modtext + MODTEXT; 926 927 econtig = e_moddata; 928 929 PRM_DEBUG(modtext); 930 PRM_DEBUG(e_modtext); 931 PRM_DEBUG(moddata); 932 PRM_DEBUG(e_moddata); 933 PRM_DEBUG(econtig); 934 935 /* 936 * Examine the boot loader physical memory map to find out: 937 * - total memory in system - physinstalled 938 * - the max physical address - physmax 939 * - the number of discontiguous segments of memory. 940 */ 941 if (prom_debug) 942 print_memlist("boot physinstalled", 943 bootops->boot_mem->physinstalled); 944 installed_top_size(bootops->boot_mem->physinstalled, &physmax, 945 &physinstalled, &memblocks); 946 PRM_DEBUG(physmax); 947 PRM_DEBUG(physinstalled); 948 PRM_DEBUG(memblocks); 949 950 /* 951 * Initialize hat's mmu parameters. 952 * Check for enforce-prot-exec in boot environment. It's used to 953 * enable/disable support for the page table entry NX bit. 954 * The default is to enforce PROT_EXEC on processors that support NX. 955 * Boot seems to round up the "len", but 8 seems to be big enough. 956 */ 957 mmu_init(); 958 959 #ifdef __i386 960 /* 961 * physmax is lowered if there is more memory than can be 962 * physically addressed in 32 bit (PAE/non-PAE) modes. 963 */ 964 if (mmu.pae_hat) { 965 if (PFN_ABOVE64G(physmax)) { 966 physinstalled -= (physmax - (PFN_64G - 1)); 967 physmax = PFN_64G - 1; 968 } 969 } else { 970 if (PFN_ABOVE4G(physmax)) { 971 physinstalled -= (physmax - (PFN_4G - 1)); 972 physmax = PFN_4G - 1; 973 } 974 } 975 #endif 976 977 startup_build_mem_nodes(bootops->boot_mem->physinstalled); 978 979 if (BOP_GETPROPLEN(bootops, "enforce-prot-exec") >= 0) { 980 int len = BOP_GETPROPLEN(bootops, "enforce-prot-exec"); 981 char value[8]; 982 983 if (len < 8) 984 (void) BOP_GETPROP(bootops, "enforce-prot-exec", value); 985 else 986 (void) strcpy(value, ""); 987 if (strcmp(value, "off") == 0) 988 mmu.pt_nx = 0; 989 } 990 PRM_DEBUG(mmu.pt_nx); 991 992 /* 993 * We will need page_t's for every page in the system, except for 994 * memory mapped at or above above the start of the kernel text segment. 995 * 996 * pages above e_modtext are attributed to kernel debugger (obp_pages) 997 */ 998 npages = physinstalled - 1; /* avail_filter() skips page 0, so "- 1" */ 999 obp_pages = 0; 1000 va = KERNEL_TEXT; 1001 while (kbm_probe(&va, &len, &pfn, &prot) != 0) { 1002 npages -= len >> MMU_PAGESHIFT; 1003 if (va >= (uintptr_t)e_moddata) 1004 obp_pages += len >> MMU_PAGESHIFT; 1005 va += len; 1006 } 1007 PRM_DEBUG(npages); 1008 PRM_DEBUG(obp_pages); 1009 1010 /* 1011 * If physmem is patched to be non-zero, use it instead of the computed 1012 * value unless it is larger than the actual amount of memory on hand. 1013 */ 1014 if (physmem == 0 || physmem > npages) { 1015 physmem = npages; 1016 } else if (physmem < npages) { 1017 orig_npages = npages; 1018 npages = physmem; 1019 } 1020 PRM_DEBUG(physmem); 1021 1022 /* 1023 * We now compute the sizes of all the initial allocations for 1024 * structures the kernel needs in order do kmem_alloc(). These 1025 * include: 1026 * memsegs 1027 * memlists 1028 * page hash table 1029 * page_t's 1030 * page coloring data structs 1031 */ 1032 memseg_sz = sizeof (struct memseg) * (memblocks + POSS_NEW_FRAGMENTS); 1033 ADD_TO_ALLOCATIONS(memseg_base, memseg_sz); 1034 PRM_DEBUG(memseg_sz); 1035 1036 /* 1037 * Reserve space for memlists. There's no real good way to know exactly 1038 * how much room we'll need, but this should be a good upper bound. 1039 */ 1040 memlist_sz = ROUND_UP_PAGE(2 * sizeof (struct memlist) * 1041 (memblocks + POSS_NEW_FRAGMENTS)); 1042 ADD_TO_ALLOCATIONS(memlist, memlist_sz); 1043 PRM_DEBUG(memlist_sz); 1044 1045 /* 1046 * The page structure hash table size is a power of 2 1047 * such that the average hash chain length is PAGE_HASHAVELEN. 1048 */ 1049 page_hashsz = npages / PAGE_HASHAVELEN; 1050 page_hashsz = 1 << highbit(page_hashsz); 1051 pagehash_sz = sizeof (struct page *) * page_hashsz; 1052 ADD_TO_ALLOCATIONS(page_hash, pagehash_sz); 1053 PRM_DEBUG(pagehash_sz); 1054 1055 /* 1056 * Set aside room for the page structures themselves. 1057 */ 1058 PRM_DEBUG(npages); 1059 pp_sz = sizeof (struct page) * npages; 1060 ADD_TO_ALLOCATIONS(pp_base, pp_sz); 1061 PRM_DEBUG(pp_sz); 1062 1063 /* 1064 * determine l2 cache info and memory size for page coloring 1065 */ 1066 (void) getl2cacheinfo(CPU, 1067 &l2cache_sz, &l2cache_linesz, &l2cache_assoc); 1068 pagecolor_memsz = 1069 page_coloring_init(l2cache_sz, l2cache_linesz, l2cache_assoc); 1070 ADD_TO_ALLOCATIONS(pagecolor_mem, pagecolor_memsz); 1071 PRM_DEBUG(pagecolor_memsz); 1072 1073 page_ctrs_size = page_ctrs_sz(); 1074 ADD_TO_ALLOCATIONS(page_ctrs_mem, page_ctrs_size); 1075 PRM_DEBUG(page_ctrs_size); 1076 1077 /* 1078 * Allocate the array that protects pp->p_selock. 1079 */ 1080 pse_shift = size_pse_array(physmem, max_ncpus); 1081 pse_table_size = 1 << pse_shift; 1082 pse_table_alloc_size = pse_table_size * sizeof (pad_mutex_t); 1083 ADD_TO_ALLOCATIONS(pse_mutex, pse_table_alloc_size); 1084 1085 #if defined(__amd64) 1086 valloc_sz = ROUND_UP_LPAGE(valloc_sz); 1087 valloc_base = VALLOC_BASE; 1088 1089 /* 1090 * The default values of VALLOC_BASE and SEGKPM_BASE should work 1091 * for values of physmax up to 1 Terabyte. They need adjusting when 1092 * memory is at addresses above 1 TB. 1093 */ 1094 if (physmax + 1 > mmu_btop(TERABYTE)) { 1095 uint64_t kpm_resv_amount = mmu_ptob(physmax + 1); 1096 1097 /* Round to largest possible pagesize for now */ 1098 kpm_resv_amount = P2ROUNDUP(kpm_resv_amount, ONE_GIG); 1099 1100 segkpm_base = -(2 * kpm_resv_amount); /* down from top VA */ 1101 1102 /* make sure we leave some space for user apps above hole */ 1103 segkpm_base = MAX(segkpm_base, AMD64_VA_HOLE_END + TERABYTE); 1104 if (segkpm_base > SEGKPM_BASE) 1105 segkpm_base = SEGKPM_BASE; 1106 PRM_DEBUG(segkpm_base); 1107 1108 valloc_base = segkpm_base + kpm_resv_amount; 1109 PRM_DEBUG(valloc_base); 1110 } 1111 #else /* __i386 */ 1112 valloc_base = (uintptr_t)(MISC_VA_BASE - valloc_sz); 1113 valloc_base = P2ALIGN(valloc_base, mmu.level_size[1]); 1114 PRM_DEBUG(valloc_base); 1115 #endif /* __i386 */ 1116 1117 /* 1118 * do all the initial allocations 1119 */ 1120 perform_allocations(); 1121 1122 /* 1123 * Build phys_install and phys_avail in kernel memspace. 1124 * - phys_install should be all memory in the system. 1125 * - phys_avail is phys_install minus any memory mapped before this 1126 * point above KERNEL_TEXT. 1127 */ 1128 current = phys_install = memlist; 1129 copy_memlist_filter(bootops->boot_mem->physinstalled, ¤t, NULL); 1130 if ((caddr_t)current > (caddr_t)memlist + memlist_sz) 1131 panic("physinstalled was too big!"); 1132 if (prom_debug) 1133 print_memlist("phys_install", phys_install); 1134 1135 phys_avail = current; 1136 PRM_POINT("Building phys_avail:\n"); 1137 copy_memlist_filter(bootops->boot_mem->physinstalled, ¤t, 1138 avail_filter); 1139 if ((caddr_t)current > (caddr_t)memlist + memlist_sz) 1140 panic("physavail was too big!"); 1141 if (prom_debug) 1142 print_memlist("phys_avail", phys_avail); 1143 1144 /* 1145 * setup page coloring 1146 */ 1147 page_coloring_setup(pagecolor_mem); 1148 page_lock_init(); /* currently a no-op */ 1149 1150 /* 1151 * free page list counters 1152 */ 1153 (void) page_ctrs_alloc(page_ctrs_mem); 1154 1155 /* 1156 * Size the pcf array based on the number of cpus in the box at 1157 * boot time. 1158 */ 1159 1160 pcf_init(); 1161 1162 /* 1163 * Initialize the page structures from the memory lists. 1164 */ 1165 availrmem_initial = availrmem = freemem = 0; 1166 PRM_POINT("Calling kphysm_init()..."); 1167 npages = kphysm_init(pp_base, npages); 1168 PRM_POINT("kphysm_init() done"); 1169 PRM_DEBUG(npages); 1170 1171 init_debug_info(); 1172 1173 /* 1174 * Now that page_t's have been initialized, remove all the 1175 * initial allocation pages from the kernel free page lists. 1176 */ 1177 boot_mapin((caddr_t)valloc_base, valloc_sz); 1178 boot_mapin((caddr_t)MISC_VA_BASE, MISC_VA_SIZE); 1179 PRM_POINT("startup_memlist() done"); 1180 1181 PRM_DEBUG(valloc_sz); 1182 1183 #if defined(__amd64) 1184 if ((availrmem >> (30 - MMU_PAGESHIFT)) >= 1185 textrepl_min_gb && l2cache_sz <= 2 << 20) { 1186 extern size_t textrepl_size_thresh; 1187 textrepl_size_thresh = (16 << 20) - 1; 1188 } 1189 #endif 1190 } 1191 1192 /* 1193 * Layout the kernel's part of address space and initialize kmem allocator. 1194 */ 1195 static void 1196 startup_kmem(void) 1197 { 1198 extern void page_set_colorequiv_arr(void); 1199 1200 PRM_POINT("startup_kmem() starting..."); 1201 1202 #if defined(__amd64) 1203 if (eprom_kernelbase && eprom_kernelbase != KERNELBASE) 1204 cmn_err(CE_NOTE, "!kernelbase cannot be changed on 64-bit " 1205 "systems."); 1206 kernelbase = segkpm_base - KERNEL_REDZONE_SIZE; 1207 core_base = (uintptr_t)COREHEAP_BASE; 1208 core_size = (size_t)MISC_VA_BASE - COREHEAP_BASE; 1209 #else /* __i386 */ 1210 /* 1211 * We configure kernelbase based on: 1212 * 1213 * 1. user specified kernelbase via eeprom command. Value cannot exceed 1214 * KERNELBASE_MAX. we large page align eprom_kernelbase 1215 * 1216 * 2. Default to KERNELBASE and adjust to 2X less the size for page_t. 1217 * On large memory systems we must lower kernelbase to allow 1218 * enough room for page_t's for all of memory. 1219 * 1220 * The value set here, might be changed a little later. 1221 */ 1222 if (eprom_kernelbase) { 1223 kernelbase = eprom_kernelbase & mmu.level_mask[1]; 1224 if (kernelbase > KERNELBASE_MAX) 1225 kernelbase = KERNELBASE_MAX; 1226 } else { 1227 kernelbase = (uintptr_t)KERNELBASE; 1228 kernelbase -= ROUND_UP_4MEG(2 * valloc_sz); 1229 } 1230 ASSERT((kernelbase & mmu.level_offset[1]) == 0); 1231 core_base = valloc_base; 1232 core_size = 0; 1233 #endif /* __i386 */ 1234 1235 PRM_DEBUG(core_base); 1236 PRM_DEBUG(core_size); 1237 PRM_DEBUG(kernelbase); 1238 1239 #if defined(__i386) 1240 segkp_fromheap = 1; 1241 #endif /* __i386 */ 1242 1243 ekernelheap = (char *)core_base; 1244 PRM_DEBUG(ekernelheap); 1245 1246 /* 1247 * Now that we know the real value of kernelbase, 1248 * update variables that were initialized with a value of 1249 * KERNELBASE (in common/conf/param.c). 1250 * 1251 * XXX The problem with this sort of hackery is that the 1252 * compiler just may feel like putting the const declarations 1253 * (in param.c) into the .text section. Perhaps they should 1254 * just be declared as variables there? 1255 */ 1256 1257 *(uintptr_t *)&_kernelbase = kernelbase; 1258 *(uintptr_t *)&_userlimit = kernelbase; 1259 #if defined(__amd64) 1260 *(uintptr_t *)&_userlimit -= KERNELBASE - USERLIMIT; 1261 #else 1262 *(uintptr_t *)&_userlimit32 = _userlimit; 1263 #endif 1264 PRM_DEBUG(_kernelbase); 1265 PRM_DEBUG(_userlimit); 1266 PRM_DEBUG(_userlimit32); 1267 1268 layout_kernel_va(); 1269 1270 #if defined(__i386) 1271 /* 1272 * If segmap is too large we can push the bottom of the kernel heap 1273 * higher than the base. Or worse, it could exceed the top of the 1274 * VA space entirely, causing it to wrap around. 1275 */ 1276 if (kernelheap >= ekernelheap || (uintptr_t)kernelheap < kernelbase) 1277 panic("too little address space available for kernelheap," 1278 " use eeprom for lower kernelbase or smaller segmapsize"); 1279 #endif /* __i386 */ 1280 1281 /* 1282 * Initialize the kernel heap. Note 3rd argument must be > 1st. 1283 */ 1284 kernelheap_init(kernelheap, ekernelheap, 1285 kernelheap + MMU_PAGESIZE, 1286 (void *)core_base, (void *)(core_base + core_size)); 1287 1288 #if defined(__xpv) 1289 /* 1290 * Link pending events struct into cpu struct 1291 */ 1292 CPU->cpu_m.mcpu_evt_pend = &cpu0_evt_data; 1293 #endif 1294 /* 1295 * Initialize kernel memory allocator. 1296 */ 1297 kmem_init(); 1298 1299 /* 1300 * Factor in colorequiv to check additional 'equivalent' bins 1301 */ 1302 page_set_colorequiv_arr(); 1303 1304 /* 1305 * print this out early so that we know what's going on 1306 */ 1307 cmn_err(CE_CONT, "?features: %b\n", x86_feature, FMT_X86_FEATURE); 1308 1309 /* 1310 * Initialize bp_mapin(). 1311 */ 1312 bp_init(MMU_PAGESIZE, HAT_STORECACHING_OK); 1313 1314 /* 1315 * orig_npages is non-zero if physmem has been configured for less 1316 * than the available memory. 1317 */ 1318 if (orig_npages) { 1319 cmn_err(CE_WARN, "!%slimiting physmem to 0x%lx of 0x%lx pages", 1320 (npages == PHYSMEM ? "Due to virtual address space " : ""), 1321 npages, orig_npages); 1322 } 1323 #if defined(__i386) 1324 if (eprom_kernelbase && (eprom_kernelbase != kernelbase)) 1325 cmn_err(CE_WARN, "kernelbase value, User specified 0x%lx, " 1326 "System using 0x%lx", 1327 (uintptr_t)eprom_kernelbase, (uintptr_t)kernelbase); 1328 #endif 1329 1330 #ifdef KERNELBASE_ABI_MIN 1331 if (kernelbase < (uintptr_t)KERNELBASE_ABI_MIN) { 1332 cmn_err(CE_NOTE, "!kernelbase set to 0x%lx, system is not " 1333 "i386 ABI compliant.", (uintptr_t)kernelbase); 1334 } 1335 #endif 1336 1337 #ifdef __xpv 1338 /* 1339 * Some of the xen start information has to be relocated up 1340 * into the kernel's permanent address space. 1341 */ 1342 PRM_POINT("calling xen_relocate_start_info()"); 1343 xen_relocate_start_info(); 1344 PRM_POINT("xen_relocate_start_info() done"); 1345 1346 /* 1347 * (Update the vcpu pointer in our cpu structure to point into 1348 * the relocated shared info.) 1349 */ 1350 CPU->cpu_m.mcpu_vcpu_info = 1351 &HYPERVISOR_shared_info->vcpu_info[CPU->cpu_id]; 1352 #endif 1353 1354 PRM_POINT("startup_kmem() done"); 1355 } 1356 1357 #ifndef __xpv 1358 /* 1359 * If we have detected that we are running in an HVM environment, we need 1360 * to prepend the PV driver directory to the module search path. 1361 */ 1362 #define HVM_MOD_DIR "/platform/i86hvm/kernel" 1363 static void 1364 update_default_path() 1365 { 1366 char *current, *newpath; 1367 int newlen; 1368 1369 /* 1370 * We are about to resync with krtld. krtld will reset its 1371 * internal module search path iff Solaris has set default_path. 1372 * We want to be sure we're prepending this new directory to the 1373 * right search path. 1374 */ 1375 current = (default_path == NULL) ? kobj_module_path : default_path; 1376 1377 newlen = strlen(HVM_MOD_DIR) + strlen(current) + 1; 1378 newpath = kmem_alloc(newlen, KM_SLEEP); 1379 (void) strcpy(newpath, HVM_MOD_DIR); 1380 (void) strcat(newpath, " "); 1381 (void) strcat(newpath, current); 1382 1383 default_path = newpath; 1384 } 1385 #endif 1386 1387 static void 1388 startup_modules(void) 1389 { 1390 unsigned int i; 1391 extern void prom_setup(void); 1392 1393 PRM_POINT("startup_modules() starting..."); 1394 1395 #ifndef __xpv 1396 /* 1397 * Initialize ten-micro second timer so that drivers will 1398 * not get short changed in their init phase. This was 1399 * not getting called until clkinit which, on fast cpu's 1400 * caused the drv_usecwait to be way too short. 1401 */ 1402 microfind(); 1403 1404 if (xpv_is_hvm) 1405 update_default_path(); 1406 #endif 1407 1408 /* 1409 * Read the GMT lag from /etc/rtc_config. 1410 */ 1411 sgmtl(process_rtc_config_file()); 1412 1413 /* 1414 * Calculate default settings of system parameters based upon 1415 * maxusers, yet allow to be overridden via the /etc/system file. 1416 */ 1417 param_calc(0); 1418 1419 mod_setup(); 1420 1421 /* 1422 * Initialize system parameters. 1423 */ 1424 param_init(); 1425 1426 /* 1427 * Initialize the default brands 1428 */ 1429 brand_init(); 1430 1431 /* 1432 * maxmem is the amount of physical memory we're playing with. 1433 */ 1434 maxmem = physmem; 1435 1436 /* 1437 * Initialize segment management stuff. 1438 */ 1439 seg_init(); 1440 1441 if (modload("fs", "specfs") == -1) 1442 halt("Can't load specfs"); 1443 1444 if (modload("fs", "devfs") == -1) 1445 halt("Can't load devfs"); 1446 1447 if (modload("fs", "dev") == -1) 1448 halt("Can't load dev"); 1449 1450 (void) modloadonly("sys", "lbl_edition"); 1451 1452 dispinit(); 1453 1454 /* 1455 * This is needed here to initialize hw_serial[] for cluster booting. 1456 */ 1457 if ((i = modload("misc", "sysinit")) != (unsigned int)-1) 1458 (void) modunload(i); 1459 else 1460 cmn_err(CE_CONT, "sysinit load failed"); 1461 1462 /* Read cluster configuration data. */ 1463 clconf_init(); 1464 1465 #if defined(__xpv) 1466 ec_init(); 1467 gnttab_init(); 1468 (void) xs_early_init(); 1469 #endif /* __xpv */ 1470 1471 /* 1472 * Create a kernel device tree. First, create rootnex and 1473 * then invoke bus specific code to probe devices. 1474 */ 1475 setup_ddi(); 1476 1477 #ifndef __xpv 1478 { 1479 /* 1480 * Set up the CPU module subsystem. Modifies the device tree, 1481 * so it must be done after setup_ddi(). 1482 */ 1483 1484 cmi_hdl_t hdl; 1485 1486 if ((hdl = cmi_init(CMI_HDL_NATIVE, cmi_ntv_hwchipid(CPU), 1487 cmi_ntv_hwcoreid(CPU), cmi_ntv_hwstrandid(CPU), 1488 cmi_ntv_hwmstrand(CPU))) != NULL) { 1489 if (x86_feature & X86_MCA) 1490 cmi_mca_init(hdl); 1491 } 1492 } 1493 #endif /* __xpv */ 1494 1495 /* 1496 * Fake a prom tree such that /dev/openprom continues to work 1497 */ 1498 PRM_POINT("startup_modules: calling prom_setup..."); 1499 prom_setup(); 1500 PRM_POINT("startup_modules: done"); 1501 1502 /* 1503 * Load all platform specific modules 1504 */ 1505 PRM_POINT("startup_modules: calling psm_modload..."); 1506 psm_modload(); 1507 1508 PRM_POINT("startup_modules() done"); 1509 } 1510 1511 /* 1512 * claim a "setaside" boot page for use in the kernel 1513 */ 1514 page_t * 1515 boot_claim_page(pfn_t pfn) 1516 { 1517 page_t *pp; 1518 1519 pp = page_numtopp_nolock(pfn); 1520 ASSERT(pp != NULL); 1521 1522 if (PP_ISBOOTPAGES(pp)) { 1523 if (pp->p_next != NULL) 1524 pp->p_next->p_prev = pp->p_prev; 1525 if (pp->p_prev == NULL) 1526 bootpages = pp->p_next; 1527 else 1528 pp->p_prev->p_next = pp->p_next; 1529 } else { 1530 /* 1531 * htable_attach() expects a base pagesize page 1532 */ 1533 if (pp->p_szc != 0) 1534 page_boot_demote(pp); 1535 pp = page_numtopp(pfn, SE_EXCL); 1536 } 1537 return (pp); 1538 } 1539 1540 /* 1541 * Walk through the pagetables looking for pages mapped in by boot. If the 1542 * setaside flag is set the pages are expected to be returned to the 1543 * kernel later in boot, so we add them to the bootpages list. 1544 */ 1545 static void 1546 protect_boot_range(uintptr_t low, uintptr_t high, int setaside) 1547 { 1548 uintptr_t va = low; 1549 size_t len; 1550 uint_t prot; 1551 pfn_t pfn; 1552 page_t *pp; 1553 pgcnt_t boot_protect_cnt = 0; 1554 1555 while (kbm_probe(&va, &len, &pfn, &prot) != 0 && va < high) { 1556 if (va + len >= high) 1557 panic("0x%lx byte mapping at 0x%p exceeds boot's " 1558 "legal range.", len, (void *)va); 1559 1560 while (len > 0) { 1561 pp = page_numtopp_alloc(pfn); 1562 if (pp != NULL) { 1563 if (setaside == 0) 1564 panic("Unexpected mapping by boot. " 1565 "addr=%p pfn=%lx\n", 1566 (void *)va, pfn); 1567 1568 pp->p_next = bootpages; 1569 pp->p_prev = NULL; 1570 PP_SETBOOTPAGES(pp); 1571 if (bootpages != NULL) { 1572 bootpages->p_prev = pp; 1573 } 1574 bootpages = pp; 1575 ++boot_protect_cnt; 1576 } 1577 1578 ++pfn; 1579 len -= MMU_PAGESIZE; 1580 va += MMU_PAGESIZE; 1581 } 1582 } 1583 PRM_DEBUG(boot_protect_cnt); 1584 } 1585 1586 /* 1587 * 1588 */ 1589 static void 1590 layout_kernel_va(void) 1591 { 1592 PRM_POINT("layout_kernel_va() starting..."); 1593 /* 1594 * Establish the final size of the kernel's heap, size of segmap, 1595 * segkp, etc. 1596 */ 1597 1598 #if defined(__amd64) 1599 1600 kpm_vbase = (caddr_t)segkpm_base; 1601 kpm_size = ROUND_UP_LPAGE(mmu_ptob(physmax + 1)); 1602 if ((uintptr_t)kpm_vbase + kpm_size > (uintptr_t)valloc_base) 1603 panic("not enough room for kpm!"); 1604 PRM_DEBUG(kpm_size); 1605 PRM_DEBUG(kpm_vbase); 1606 1607 /* 1608 * By default we create a seg_kp in 64 bit kernels, it's a little 1609 * faster to access than embedding it in the heap. 1610 */ 1611 segkp_base = (caddr_t)valloc_base + valloc_sz; 1612 if (!segkp_fromheap) { 1613 size_t sz = mmu_ptob(segkpsize); 1614 1615 /* 1616 * determine size of segkp 1617 */ 1618 if (sz < SEGKPMINSIZE || sz > SEGKPMAXSIZE) { 1619 sz = SEGKPDEFSIZE; 1620 cmn_err(CE_WARN, "!Illegal value for segkpsize. " 1621 "segkpsize has been reset to %ld pages", 1622 mmu_btop(sz)); 1623 } 1624 sz = MIN(sz, MAX(SEGKPMINSIZE, mmu_ptob(physmem))); 1625 1626 segkpsize = mmu_btop(ROUND_UP_LPAGE(sz)); 1627 } 1628 PRM_DEBUG(segkp_base); 1629 PRM_DEBUG(segkpsize); 1630 1631 /* 1632 * segzio is used for ZFS cached data. It uses a distinct VA 1633 * segment (from kernel heap) so that we can easily tell not to 1634 * include it in kernel crash dumps on 64 bit kernels. The trick is 1635 * to give it lots of VA, but not constrain the kernel heap. 1636 * We scale the size of segzio linearly with physmem up to 1637 * SEGZIOMAXSIZE. Above that amount it scales at 50% of physmem. 1638 */ 1639 segzio_base = segkp_base + mmu_ptob(segkpsize); 1640 if (segzio_fromheap) { 1641 segziosize = 0; 1642 } else { 1643 size_t physmem_size = mmu_ptob(physmem); 1644 size_t size = (segziosize == 0) ? 1645 physmem_size : mmu_ptob(segziosize); 1646 1647 if (size < SEGZIOMINSIZE) 1648 size = SEGZIOMINSIZE; 1649 if (size > SEGZIOMAXSIZE) { 1650 size = SEGZIOMAXSIZE; 1651 if (physmem_size > size) 1652 size += (physmem_size - size) / 2; 1653 } 1654 segziosize = mmu_btop(ROUND_UP_LPAGE(size)); 1655 } 1656 PRM_DEBUG(segziosize); 1657 PRM_DEBUG(segzio_base); 1658 1659 /* 1660 * Put the range of VA for device mappings next, kmdb knows to not 1661 * grep in this range of addresses. 1662 */ 1663 toxic_addr = 1664 ROUND_UP_LPAGE((uintptr_t)segzio_base + mmu_ptob(segziosize)); 1665 PRM_DEBUG(toxic_addr); 1666 segmap_start = ROUND_UP_LPAGE(toxic_addr + toxic_size); 1667 #else /* __i386 */ 1668 segmap_start = ROUND_UP_LPAGE(kernelbase); 1669 #endif /* __i386 */ 1670 PRM_DEBUG(segmap_start); 1671 1672 /* 1673 * Users can change segmapsize through eeprom or /etc/system. 1674 * If the variable is tuned through eeprom, there is no upper 1675 * bound on the size of segmap. If it is tuned through 1676 * /etc/system on 32-bit systems, it must be no larger than we 1677 * planned for in startup_memlist(). 1678 */ 1679 segmapsize = MAX(ROUND_UP_LPAGE(segmapsize), SEGMAPDEFAULT); 1680 1681 #if defined(__i386) 1682 /* 1683 * 32-bit systems don't have segkpm or segkp, so segmap appears at 1684 * the bottom of the kernel's address range. Set aside space for a 1685 * small red zone just below the start of segmap. 1686 */ 1687 segmap_start += KERNEL_REDZONE_SIZE; 1688 segmapsize -= KERNEL_REDZONE_SIZE; 1689 #endif 1690 1691 PRM_DEBUG(segmap_start); 1692 PRM_DEBUG(segmapsize); 1693 kernelheap = (caddr_t)ROUND_UP_LPAGE(segmap_start + segmapsize); 1694 PRM_DEBUG(kernelheap); 1695 PRM_POINT("layout_kernel_va() done..."); 1696 } 1697 1698 /* 1699 * Finish initializing the VM system, now that we are no longer 1700 * relying on the boot time memory allocators. 1701 */ 1702 static void 1703 startup_vm(void) 1704 { 1705 struct segmap_crargs a; 1706 1707 extern int use_brk_lpg, use_stk_lpg; 1708 1709 PRM_POINT("startup_vm() starting..."); 1710 1711 /* 1712 * Initialize the hat layer. 1713 */ 1714 hat_init(); 1715 1716 /* 1717 * Do final allocations of HAT data structures that need to 1718 * be allocated before quiescing the boot loader. 1719 */ 1720 PRM_POINT("Calling hat_kern_alloc()..."); 1721 hat_kern_alloc((caddr_t)segmap_start, segmapsize, ekernelheap); 1722 PRM_POINT("hat_kern_alloc() done"); 1723 1724 #ifndef __xpv 1725 /* 1726 * Setup Page Attribute Table 1727 */ 1728 pat_sync(); 1729 #endif 1730 1731 /* 1732 * The next two loops are done in distinct steps in order 1733 * to be sure that any page that is doubly mapped (both above 1734 * KERNEL_TEXT and below kernelbase) is dealt with correctly. 1735 * Note this may never happen, but it might someday. 1736 */ 1737 bootpages = NULL; 1738 PRM_POINT("Protecting boot pages"); 1739 1740 /* 1741 * Protect any pages mapped above KERNEL_TEXT that somehow have 1742 * page_t's. This can only happen if something weird allocated 1743 * in this range (like kadb/kmdb). 1744 */ 1745 protect_boot_range(KERNEL_TEXT, (uintptr_t)-1, 0); 1746 1747 /* 1748 * Before we can take over memory allocation/mapping from the boot 1749 * loader we must remove from our free page lists any boot allocated 1750 * pages that stay mapped until release_bootstrap(). 1751 */ 1752 protect_boot_range(0, kernelbase, 1); 1753 1754 1755 /* 1756 * Switch to running on regular HAT (not boot_mmu) 1757 */ 1758 PRM_POINT("Calling hat_kern_setup()..."); 1759 hat_kern_setup(); 1760 1761 /* 1762 * It is no longer safe to call BOP_ALLOC(), so make sure we don't. 1763 */ 1764 bop_no_more_mem(); 1765 1766 PRM_POINT("hat_kern_setup() done"); 1767 1768 hat_cpu_online(CPU); 1769 1770 /* 1771 * Initialize VM system 1772 */ 1773 PRM_POINT("Calling kvm_init()..."); 1774 kvm_init(); 1775 PRM_POINT("kvm_init() done"); 1776 1777 /* 1778 * Tell kmdb that the VM system is now working 1779 */ 1780 if (boothowto & RB_DEBUG) 1781 kdi_dvec_vmready(); 1782 1783 #if defined(__xpv) 1784 /* 1785 * Populate the I/O pool on domain 0 1786 */ 1787 if (DOMAIN_IS_INITDOMAIN(xen_info)) { 1788 extern long populate_io_pool(void); 1789 long init_io_pool_cnt; 1790 1791 PRM_POINT("Populating reserve I/O page pool"); 1792 init_io_pool_cnt = populate_io_pool(); 1793 PRM_DEBUG(init_io_pool_cnt); 1794 } 1795 #endif 1796 /* 1797 * Mangle the brand string etc. 1798 */ 1799 cpuid_pass3(CPU); 1800 1801 #if defined(__amd64) 1802 1803 /* 1804 * Create the device arena for toxic (to dtrace/kmdb) mappings. 1805 */ 1806 device_arena = vmem_create("device", (void *)toxic_addr, 1807 toxic_size, MMU_PAGESIZE, NULL, NULL, NULL, 0, VM_SLEEP); 1808 1809 #else /* __i386 */ 1810 1811 /* 1812 * allocate the bit map that tracks toxic pages 1813 */ 1814 toxic_bit_map_len = btop((ulong_t)(valloc_base - kernelbase)); 1815 PRM_DEBUG(toxic_bit_map_len); 1816 toxic_bit_map = 1817 kmem_zalloc(BT_SIZEOFMAP(toxic_bit_map_len), KM_NOSLEEP); 1818 ASSERT(toxic_bit_map != NULL); 1819 PRM_DEBUG(toxic_bit_map); 1820 1821 #endif /* __i386 */ 1822 1823 1824 /* 1825 * Now that we've got more VA, as well as the ability to allocate from 1826 * it, tell the debugger. 1827 */ 1828 if (boothowto & RB_DEBUG) 1829 kdi_dvec_memavail(); 1830 1831 /* 1832 * The following code installs a special page fault handler (#pf) 1833 * to work around a pentium bug. 1834 */ 1835 #if !defined(__amd64) && !defined(__xpv) 1836 if (x86_type == X86_TYPE_P5) { 1837 desctbr_t idtr; 1838 gate_desc_t *newidt; 1839 1840 if ((newidt = kmem_zalloc(MMU_PAGESIZE, KM_NOSLEEP)) == NULL) 1841 panic("failed to install pentium_pftrap"); 1842 1843 bcopy(idt0, newidt, NIDT * sizeof (*idt0)); 1844 set_gatesegd(&newidt[T_PGFLT], &pentium_pftrap, 1845 KCS_SEL, SDT_SYSIGT, TRP_KPL); 1846 1847 (void) as_setprot(&kas, (caddr_t)newidt, MMU_PAGESIZE, 1848 PROT_READ | PROT_EXEC); 1849 1850 CPU->cpu_idt = newidt; 1851 idtr.dtr_base = (uintptr_t)CPU->cpu_idt; 1852 idtr.dtr_limit = (NIDT * sizeof (*idt0)) - 1; 1853 wr_idtr(&idtr); 1854 } 1855 #endif /* !__amd64 */ 1856 1857 #if !defined(__xpv) 1858 /* 1859 * Map page pfn=0 for drivers, such as kd, that need to pick up 1860 * parameters left there by controllers/BIOS. 1861 */ 1862 PRM_POINT("setup up p0_va"); 1863 p0_va = i86devmap(0, 1, PROT_READ); 1864 PRM_DEBUG(p0_va); 1865 #endif 1866 1867 cmn_err(CE_CONT, "?mem = %luK (0x%lx)\n", 1868 physinstalled << (MMU_PAGESHIFT - 10), ptob(physinstalled)); 1869 1870 /* 1871 * disable automatic large pages for small memory systems or 1872 * when the disable flag is set. 1873 * 1874 * Do not yet consider page sizes larger than 2m/4m. 1875 */ 1876 if (!auto_lpg_disable && mmu.max_page_level > 0) { 1877 max_uheap_lpsize = LEVEL_SIZE(1); 1878 max_ustack_lpsize = LEVEL_SIZE(1); 1879 max_privmap_lpsize = LEVEL_SIZE(1); 1880 max_uidata_lpsize = LEVEL_SIZE(1); 1881 max_utext_lpsize = LEVEL_SIZE(1); 1882 max_shm_lpsize = LEVEL_SIZE(1); 1883 } 1884 if (physmem < privm_lpg_min_physmem || mmu.max_page_level == 0 || 1885 auto_lpg_disable) { 1886 use_brk_lpg = 0; 1887 use_stk_lpg = 0; 1888 } 1889 mcntl0_lpsize = LEVEL_SIZE(mmu.umax_page_level); 1890 1891 PRM_POINT("Calling hat_init_finish()..."); 1892 hat_init_finish(); 1893 PRM_POINT("hat_init_finish() done"); 1894 1895 /* 1896 * Initialize the segkp segment type. 1897 */ 1898 rw_enter(&kas.a_lock, RW_WRITER); 1899 PRM_POINT("Attaching segkp"); 1900 if (segkp_fromheap) { 1901 segkp->s_as = &kas; 1902 } else if (seg_attach(&kas, (caddr_t)segkp_base, mmu_ptob(segkpsize), 1903 segkp) < 0) { 1904 panic("startup: cannot attach segkp"); 1905 /*NOTREACHED*/ 1906 } 1907 PRM_POINT("Doing segkp_create()"); 1908 if (segkp_create(segkp) != 0) { 1909 panic("startup: segkp_create failed"); 1910 /*NOTREACHED*/ 1911 } 1912 PRM_DEBUG(segkp); 1913 rw_exit(&kas.a_lock); 1914 1915 /* 1916 * kpm segment 1917 */ 1918 segmap_kpm = 0; 1919 if (kpm_desired) { 1920 kpm_init(); 1921 kpm_enable = 1; 1922 vpm_enable = 1; 1923 } 1924 1925 /* 1926 * Now create segmap segment. 1927 */ 1928 rw_enter(&kas.a_lock, RW_WRITER); 1929 if (seg_attach(&kas, (caddr_t)segmap_start, segmapsize, segmap) < 0) { 1930 panic("cannot attach segmap"); 1931 /*NOTREACHED*/ 1932 } 1933 PRM_DEBUG(segmap); 1934 1935 a.prot = PROT_READ | PROT_WRITE; 1936 a.shmsize = 0; 1937 a.nfreelist = segmapfreelists; 1938 1939 if (segmap_create(segmap, (caddr_t)&a) != 0) 1940 panic("segmap_create segmap"); 1941 rw_exit(&kas.a_lock); 1942 1943 setup_vaddr_for_ppcopy(CPU); 1944 1945 segdev_init(); 1946 #if defined(__xpv) 1947 if (DOMAIN_IS_INITDOMAIN(xen_info)) 1948 #endif 1949 pmem_init(); 1950 1951 PRM_POINT("startup_vm() done"); 1952 } 1953 1954 /* 1955 * Load a tod module for the non-standard tod part found on this system. 1956 */ 1957 static void 1958 load_tod_module(char *todmod) 1959 { 1960 if (modload("tod", todmod) == -1) 1961 halt("Can't load TOD module"); 1962 } 1963 1964 static void 1965 startup_end(void) 1966 { 1967 int i; 1968 extern void setx86isalist(void); 1969 1970 PRM_POINT("startup_end() starting..."); 1971 1972 /* 1973 * Perform tasks that get done after most of the VM 1974 * initialization has been done but before the clock 1975 * and other devices get started. 1976 */ 1977 kern_setup1(); 1978 1979 /* 1980 * Perform CPC initialization for this CPU. 1981 */ 1982 kcpc_hw_init(CPU); 1983 1984 #if defined(OPTERON_WORKAROUND_6323525) 1985 if (opteron_workaround_6323525) 1986 patch_workaround_6323525(); 1987 #endif 1988 /* 1989 * If needed, load TOD module now so that ddi_get_time(9F) etc. work 1990 * (For now, "needed" is defined as set tod_module_name in /etc/system) 1991 */ 1992 if (tod_module_name != NULL) { 1993 PRM_POINT("load_tod_module()"); 1994 load_tod_module(tod_module_name); 1995 } 1996 1997 #if defined(__xpv) 1998 /* 1999 * Forceload interposing TOD module for the hypervisor. 2000 */ 2001 PRM_POINT("load_tod_module()"); 2002 load_tod_module("xpvtod"); 2003 #endif 2004 2005 /* 2006 * Configure the system. 2007 */ 2008 PRM_POINT("Calling configure()..."); 2009 configure(); /* set up devices */ 2010 PRM_POINT("configure() done"); 2011 2012 /* 2013 * Set the isa_list string to the defined instruction sets we 2014 * support. 2015 */ 2016 setx86isalist(); 2017 cpu_intr_alloc(CPU, NINTR_THREADS); 2018 psm_install(); 2019 2020 /* 2021 * We're done with bootops. We don't unmap the bootstrap yet because 2022 * we're still using bootsvcs. 2023 */ 2024 PRM_POINT("NULLing out bootops"); 2025 *bootopsp = (struct bootops *)NULL; 2026 bootops = (struct bootops *)NULL; 2027 2028 #if defined(__xpv) 2029 ec_init_debug_irq(); 2030 xs_domu_init(); 2031 #endif 2032 PRM_POINT("Enabling interrupts"); 2033 (*picinitf)(); 2034 sti(); 2035 #if defined(__xpv) 2036 ASSERT(CPU->cpu_m.mcpu_vcpu_info->evtchn_upcall_mask == 0); 2037 xen_late_startup(); 2038 #endif 2039 2040 (void) add_avsoftintr((void *)&softlevel1_hdl, 1, softlevel1, 2041 "softlevel1", NULL, NULL); /* XXX to be moved later */ 2042 2043 /* 2044 * Register these software interrupts for ddi timer. 2045 * Software interrupts up to the level 10 are supported. 2046 */ 2047 for (i = DDI_IPL_1; i <= DDI_IPL_10; i++) { 2048 char name[sizeof ("timer_softintr") + 2]; 2049 (void) sprintf(name, "timer_softintr%02d", i); 2050 (void) add_avsoftintr((void *)&softlevel_hdl[i-1], i, 2051 (avfunc)timer_softintr, name, (caddr_t)(uintptr_t)i, NULL); 2052 } 2053 2054 PRM_POINT("startup_end() done"); 2055 } 2056 2057 extern char hw_serial[]; 2058 char *_hs1107 = hw_serial; 2059 ulong_t _bdhs34; 2060 2061 void 2062 post_startup(void) 2063 { 2064 /* 2065 * Set the system wide, processor-specific flags to be passed 2066 * to userland via the aux vector for performance hints and 2067 * instruction set extensions. 2068 */ 2069 bind_hwcap(); 2070 2071 #ifdef __xpv 2072 if (DOMAIN_IS_INITDOMAIN(xen_info)) 2073 #endif 2074 { 2075 /* 2076 * Load the System Management BIOS into the global ksmbios 2077 * handle, if an SMBIOS is present on this system. 2078 */ 2079 ksmbios = smbios_open(NULL, SMB_VERSION, ksmbios_flags, NULL); 2080 2081 #if defined(__xpv) 2082 xpv_panic_init(); 2083 #else 2084 /* 2085 * Startup the memory scrubber. 2086 * XXPV This should be running somewhere .. 2087 */ 2088 memscrub_init(); 2089 #endif 2090 } 2091 2092 /* 2093 * Complete CPU module initialization 2094 */ 2095 cmi_post_startup(); 2096 2097 /* 2098 * Perform forceloading tasks for /etc/system. 2099 */ 2100 (void) mod_sysctl(SYS_FORCELOAD, NULL); 2101 2102 /* 2103 * ON4.0: Force /proc module in until clock interrupt handle fixed 2104 * ON4.0: This must be fixed or restated in /etc/systems. 2105 */ 2106 (void) modload("fs", "procfs"); 2107 2108 (void) i_ddi_attach_hw_nodes("pit_beep"); 2109 2110 #if defined(__i386) 2111 /* 2112 * Check for required functional Floating Point hardware, 2113 * unless FP hardware explicitly disabled. 2114 */ 2115 if (fpu_exists && (fpu_pentium_fdivbug || fp_kind == FP_NO)) 2116 halt("No working FP hardware found"); 2117 #endif 2118 2119 maxmem = freemem; 2120 2121 add_cpunode2devtree(CPU->cpu_id, CPU->cpu_m.mcpu_cpi); 2122 } 2123 2124 static int 2125 pp_in_ramdisk(page_t *pp) 2126 { 2127 extern uint64_t ramdisk_start, ramdisk_end; 2128 2129 return ((pp->p_pagenum >= btop(ramdisk_start)) && 2130 (pp->p_pagenum < btopr(ramdisk_end))); 2131 } 2132 2133 void 2134 release_bootstrap(void) 2135 { 2136 int root_is_ramdisk; 2137 page_t *pp; 2138 extern void kobj_boot_unmountroot(void); 2139 extern dev_t rootdev; 2140 2141 /* unmount boot ramdisk and release kmem usage */ 2142 kobj_boot_unmountroot(); 2143 2144 /* 2145 * We're finished using the boot loader so free its pages. 2146 */ 2147 PRM_POINT("Unmapping lower boot pages"); 2148 clear_boot_mappings(0, _userlimit); 2149 postbootkernelbase = kernelbase; 2150 2151 /* 2152 * If root isn't on ramdisk, destroy the hardcoded 2153 * ramdisk node now and release the memory. Else, 2154 * ramdisk memory is kept in rd_pages. 2155 */ 2156 root_is_ramdisk = (getmajor(rootdev) == ddi_name_to_major("ramdisk")); 2157 if (!root_is_ramdisk) { 2158 dev_info_t *dip = ddi_find_devinfo("ramdisk", -1, 0); 2159 ASSERT(dip && ddi_get_parent(dip) == ddi_root_node()); 2160 ndi_rele_devi(dip); /* held from ddi_find_devinfo */ 2161 (void) ddi_remove_child(dip, 0); 2162 } 2163 2164 PRM_POINT("Releasing boot pages"); 2165 while (bootpages) { 2166 pp = bootpages; 2167 bootpages = pp->p_next; 2168 if (root_is_ramdisk && pp_in_ramdisk(pp)) { 2169 pp->p_next = rd_pages; 2170 rd_pages = pp; 2171 continue; 2172 } 2173 pp->p_next = (struct page *)0; 2174 pp->p_prev = (struct page *)0; 2175 PP_CLRBOOTPAGES(pp); 2176 page_free(pp, 1); 2177 } 2178 PRM_POINT("Boot pages released"); 2179 2180 #if !defined(__xpv) 2181 /* XXPV -- note this following bunch of code needs to be revisited in Xen 3.0 */ 2182 /* 2183 * Find 1 page below 1 MB so that other processors can boot up. 2184 * Make sure it has a kernel VA as well as a 1:1 mapping. 2185 * We should have just free'd one up. 2186 */ 2187 if (use_mp) { 2188 pfn_t pfn; 2189 2190 for (pfn = 1; pfn < btop(1*1024*1024); pfn++) { 2191 if (page_numtopp_alloc(pfn) == NULL) 2192 continue; 2193 rm_platter_va = i86devmap(pfn, 1, 2194 PROT_READ | PROT_WRITE | PROT_EXEC); 2195 rm_platter_pa = ptob(pfn); 2196 hat_devload(kas.a_hat, 2197 (caddr_t)(uintptr_t)rm_platter_pa, MMU_PAGESIZE, 2198 pfn, PROT_READ | PROT_WRITE | PROT_EXEC, 2199 HAT_LOAD_NOCONSIST); 2200 break; 2201 } 2202 if (pfn == btop(1*1024*1024)) 2203 panic("No page available for starting " 2204 "other processors"); 2205 } 2206 #endif /* !__xpv */ 2207 } 2208 2209 /* 2210 * Initialize the platform-specific parts of a page_t. 2211 */ 2212 void 2213 add_physmem_cb(page_t *pp, pfn_t pnum) 2214 { 2215 pp->p_pagenum = pnum; 2216 pp->p_mapping = NULL; 2217 pp->p_embed = 0; 2218 pp->p_share = 0; 2219 pp->p_mlentry = 0; 2220 } 2221 2222 /* 2223 * kphysm_init() initializes physical memory. 2224 */ 2225 static pgcnt_t 2226 kphysm_init( 2227 page_t *pp, 2228 pgcnt_t npages) 2229 { 2230 struct memlist *pmem; 2231 struct memseg *cur_memseg; 2232 pfn_t base_pfn; 2233 pgcnt_t num; 2234 pgcnt_t pages_done = 0; 2235 uint64_t addr; 2236 uint64_t size; 2237 extern pfn_t ddiphysmin; 2238 2239 ASSERT(page_hash != NULL && page_hashsz != 0); 2240 2241 cur_memseg = memseg_base; 2242 for (pmem = phys_avail; pmem && npages; pmem = pmem->next) { 2243 /* 2244 * In a 32 bit kernel can't use higher memory if we're 2245 * not booting in PAE mode. This check takes care of that. 2246 */ 2247 addr = pmem->address; 2248 size = pmem->size; 2249 if (btop(addr) > physmax) 2250 continue; 2251 2252 /* 2253 * align addr and size - they may not be at page boundaries 2254 */ 2255 if ((addr & MMU_PAGEOFFSET) != 0) { 2256 addr += MMU_PAGEOFFSET; 2257 addr &= ~(uint64_t)MMU_PAGEOFFSET; 2258 size -= addr - pmem->address; 2259 } 2260 2261 /* only process pages below or equal to physmax */ 2262 if ((btop(addr + size) - 1) > physmax) 2263 size = ptob(physmax - btop(addr) + 1); 2264 2265 num = btop(size); 2266 if (num == 0) 2267 continue; 2268 2269 if (num > npages) 2270 num = npages; 2271 2272 npages -= num; 2273 pages_done += num; 2274 base_pfn = btop(addr); 2275 2276 if (prom_debug) 2277 prom_printf("MEMSEG addr=0x%" PRIx64 2278 " pgs=0x%lx pfn 0x%lx-0x%lx\n", 2279 addr, num, base_pfn, base_pfn + num); 2280 2281 /* 2282 * Ignore pages below ddiphysmin to simplify ddi memory 2283 * allocation with non-zero addr_lo requests. 2284 */ 2285 if (base_pfn < ddiphysmin) { 2286 if (base_pfn + num <= ddiphysmin) 2287 continue; 2288 pp += (ddiphysmin - base_pfn); 2289 num -= (ddiphysmin - base_pfn); 2290 base_pfn = ddiphysmin; 2291 } 2292 2293 /* 2294 * Build the memsegs entry 2295 */ 2296 cur_memseg->pages = pp; 2297 cur_memseg->epages = pp + num; 2298 cur_memseg->pages_base = base_pfn; 2299 cur_memseg->pages_end = base_pfn + num; 2300 2301 /* 2302 * Insert into memseg list in decreasing pfn range order. 2303 * Low memory is typically more fragmented such that this 2304 * ordering keeps the larger ranges at the front of the list 2305 * for code that searches memseg. 2306 * This ASSERTS that the memsegs coming in from boot are in 2307 * increasing physical address order and not contiguous. 2308 */ 2309 if (memsegs != NULL) { 2310 ASSERT(cur_memseg->pages_base >= memsegs->pages_end); 2311 cur_memseg->next = memsegs; 2312 } 2313 memsegs = cur_memseg; 2314 2315 /* 2316 * add_physmem() initializes the PSM part of the page 2317 * struct by calling the PSM back with add_physmem_cb(). 2318 * In addition it coalesces pages into larger pages as 2319 * it initializes them. 2320 */ 2321 add_physmem(pp, num, base_pfn); 2322 cur_memseg++; 2323 availrmem_initial += num; 2324 availrmem += num; 2325 2326 pp += num; 2327 } 2328 2329 PRM_DEBUG(availrmem_initial); 2330 PRM_DEBUG(availrmem); 2331 PRM_DEBUG(freemem); 2332 build_pfn_hash(); 2333 return (pages_done); 2334 } 2335 2336 /* 2337 * Kernel VM initialization. 2338 */ 2339 static void 2340 kvm_init(void) 2341 { 2342 ASSERT((((uintptr_t)s_text) & MMU_PAGEOFFSET) == 0); 2343 2344 /* 2345 * Put the kernel segments in kernel address space. 2346 */ 2347 rw_enter(&kas.a_lock, RW_WRITER); 2348 as_avlinit(&kas); 2349 2350 (void) seg_attach(&kas, s_text, e_moddata - s_text, &ktextseg); 2351 (void) segkmem_create(&ktextseg); 2352 2353 (void) seg_attach(&kas, (caddr_t)valloc_base, valloc_sz, &kvalloc); 2354 (void) segkmem_create(&kvalloc); 2355 2356 (void) seg_attach(&kas, kernelheap, 2357 ekernelheap - kernelheap, &kvseg); 2358 (void) segkmem_create(&kvseg); 2359 2360 if (core_size > 0) { 2361 PRM_POINT("attaching kvseg_core"); 2362 (void) seg_attach(&kas, (caddr_t)core_base, core_size, 2363 &kvseg_core); 2364 (void) segkmem_create(&kvseg_core); 2365 } 2366 2367 if (segziosize > 0) { 2368 PRM_POINT("attaching segzio"); 2369 (void) seg_attach(&kas, segzio_base, mmu_ptob(segziosize), 2370 &kzioseg); 2371 (void) segkmem_zio_create(&kzioseg); 2372 2373 /* create zio area covering new segment */ 2374 segkmem_zio_init(segzio_base, mmu_ptob(segziosize)); 2375 } 2376 2377 (void) seg_attach(&kas, kdi_segdebugbase, kdi_segdebugsize, &kdebugseg); 2378 (void) segkmem_create(&kdebugseg); 2379 2380 rw_exit(&kas.a_lock); 2381 2382 /* 2383 * Ensure that the red zone at kernelbase is never accessible. 2384 */ 2385 PRM_POINT("protecting redzone"); 2386 (void) as_setprot(&kas, (caddr_t)kernelbase, KERNEL_REDZONE_SIZE, 0); 2387 2388 /* 2389 * Make the text writable so that it can be hot patched by DTrace. 2390 */ 2391 (void) as_setprot(&kas, s_text, e_modtext - s_text, 2392 PROT_READ | PROT_WRITE | PROT_EXEC); 2393 2394 /* 2395 * Make data writable until end. 2396 */ 2397 (void) as_setprot(&kas, s_data, e_moddata - s_data, 2398 PROT_READ | PROT_WRITE | PROT_EXEC); 2399 } 2400 2401 #ifndef __xpv 2402 /* 2403 * Solaris adds an entry for Write Combining caching to the PAT 2404 */ 2405 static uint64_t pat_attr_reg = PAT_DEFAULT_ATTRIBUTE; 2406 2407 void 2408 pat_sync(void) 2409 { 2410 ulong_t cr0, cr0_orig, cr4; 2411 2412 if (!(x86_feature & X86_PAT)) 2413 return; 2414 cr0_orig = cr0 = getcr0(); 2415 cr4 = getcr4(); 2416 2417 /* disable caching and flush all caches and TLBs */ 2418 cr0 |= CR0_CD; 2419 cr0 &= ~CR0_NW; 2420 setcr0(cr0); 2421 invalidate_cache(); 2422 if (cr4 & CR4_PGE) { 2423 setcr4(cr4 & ~(ulong_t)CR4_PGE); 2424 setcr4(cr4); 2425 } else { 2426 reload_cr3(); 2427 } 2428 2429 /* add our entry to the PAT */ 2430 wrmsr(REG_PAT, pat_attr_reg); 2431 2432 /* flush TLBs and cache again, then reenable cr0 caching */ 2433 if (cr4 & CR4_PGE) { 2434 setcr4(cr4 & ~(ulong_t)CR4_PGE); 2435 setcr4(cr4); 2436 } else { 2437 reload_cr3(); 2438 } 2439 invalidate_cache(); 2440 setcr0(cr0_orig); 2441 } 2442 2443 #endif /* !__xpv */ 2444 2445 void 2446 get_system_configuration(void) 2447 { 2448 char prop[32]; 2449 u_longlong_t nodes_ll, cpus_pernode_ll, lvalue; 2450 2451 if (BOP_GETPROPLEN(bootops, "nodes") > sizeof (prop) || 2452 BOP_GETPROP(bootops, "nodes", prop) < 0 || 2453 kobj_getvalue(prop, &nodes_ll) == -1 || 2454 nodes_ll > MAXNODES || 2455 BOP_GETPROPLEN(bootops, "cpus_pernode") > sizeof (prop) || 2456 BOP_GETPROP(bootops, "cpus_pernode", prop) < 0 || 2457 kobj_getvalue(prop, &cpus_pernode_ll) == -1) { 2458 system_hardware.hd_nodes = 1; 2459 system_hardware.hd_cpus_per_node = 0; 2460 } else { 2461 system_hardware.hd_nodes = (int)nodes_ll; 2462 system_hardware.hd_cpus_per_node = (int)cpus_pernode_ll; 2463 } 2464 2465 if (BOP_GETPROPLEN(bootops, "kernelbase") > sizeof (prop) || 2466 BOP_GETPROP(bootops, "kernelbase", prop) < 0 || 2467 kobj_getvalue(prop, &lvalue) == -1) 2468 eprom_kernelbase = NULL; 2469 else 2470 eprom_kernelbase = (uintptr_t)lvalue; 2471 2472 if (BOP_GETPROPLEN(bootops, "segmapsize") > sizeof (prop) || 2473 BOP_GETPROP(bootops, "segmapsize", prop) < 0 || 2474 kobj_getvalue(prop, &lvalue) == -1) 2475 segmapsize = SEGMAPDEFAULT; 2476 else 2477 segmapsize = (uintptr_t)lvalue; 2478 2479 if (BOP_GETPROPLEN(bootops, "segmapfreelists") > sizeof (prop) || 2480 BOP_GETPROP(bootops, "segmapfreelists", prop) < 0 || 2481 kobj_getvalue(prop, &lvalue) == -1) 2482 segmapfreelists = 0; /* use segmap driver default */ 2483 else 2484 segmapfreelists = (int)lvalue; 2485 2486 /* physmem used to be here, but moved much earlier to fakebop.c */ 2487 } 2488 2489 /* 2490 * Add to a memory list. 2491 * start = start of new memory segment 2492 * len = length of new memory segment in bytes 2493 * new = pointer to a new struct memlist 2494 * memlistp = memory list to which to add segment. 2495 */ 2496 void 2497 memlist_add( 2498 uint64_t start, 2499 uint64_t len, 2500 struct memlist *new, 2501 struct memlist **memlistp) 2502 { 2503 struct memlist *cur; 2504 uint64_t end = start + len; 2505 2506 new->address = start; 2507 new->size = len; 2508 2509 cur = *memlistp; 2510 2511 while (cur) { 2512 if (cur->address >= end) { 2513 new->next = cur; 2514 *memlistp = new; 2515 new->prev = cur->prev; 2516 cur->prev = new; 2517 return; 2518 } 2519 ASSERT(cur->address + cur->size <= start); 2520 if (cur->next == NULL) { 2521 cur->next = new; 2522 new->prev = cur; 2523 new->next = NULL; 2524 return; 2525 } 2526 memlistp = &cur->next; 2527 cur = cur->next; 2528 } 2529 } 2530 2531 void 2532 kobj_vmem_init(vmem_t **text_arena, vmem_t **data_arena) 2533 { 2534 size_t tsize = e_modtext - modtext; 2535 size_t dsize = e_moddata - moddata; 2536 2537 *text_arena = vmem_create("module_text", tsize ? modtext : NULL, tsize, 2538 1, segkmem_alloc, segkmem_free, heaptext_arena, 0, VM_SLEEP); 2539 *data_arena = vmem_create("module_data", dsize ? moddata : NULL, dsize, 2540 1, segkmem_alloc, segkmem_free, heap32_arena, 0, VM_SLEEP); 2541 } 2542 2543 caddr_t 2544 kobj_text_alloc(vmem_t *arena, size_t size) 2545 { 2546 return (vmem_alloc(arena, size, VM_SLEEP | VM_BESTFIT)); 2547 } 2548 2549 /*ARGSUSED*/ 2550 caddr_t 2551 kobj_texthole_alloc(caddr_t addr, size_t size) 2552 { 2553 panic("unexpected call to kobj_texthole_alloc()"); 2554 /*NOTREACHED*/ 2555 return (0); 2556 } 2557 2558 /*ARGSUSED*/ 2559 void 2560 kobj_texthole_free(caddr_t addr, size_t size) 2561 { 2562 panic("unexpected call to kobj_texthole_free()"); 2563 } 2564 2565 /* 2566 * This is called just after configure() in startup(). 2567 * 2568 * The ISALIST concept is a bit hopeless on Intel, because 2569 * there's no guarantee of an ever-more-capable processor 2570 * given that various parts of the instruction set may appear 2571 * and disappear between different implementations. 2572 * 2573 * While it would be possible to correct it and even enhance 2574 * it somewhat, the explicit hardware capability bitmask allows 2575 * more flexibility. 2576 * 2577 * So, we just leave this alone. 2578 */ 2579 void 2580 setx86isalist(void) 2581 { 2582 char *tp; 2583 size_t len; 2584 extern char *isa_list; 2585 2586 #define TBUFSIZE 1024 2587 2588 tp = kmem_alloc(TBUFSIZE, KM_SLEEP); 2589 *tp = '\0'; 2590 2591 #if defined(__amd64) 2592 (void) strcpy(tp, "amd64 "); 2593 #endif 2594 2595 switch (x86_vendor) { 2596 case X86_VENDOR_Intel: 2597 case X86_VENDOR_AMD: 2598 case X86_VENDOR_TM: 2599 if (x86_feature & X86_CMOV) { 2600 /* 2601 * Pentium Pro or later 2602 */ 2603 (void) strcat(tp, "pentium_pro"); 2604 (void) strcat(tp, x86_feature & X86_MMX ? 2605 "+mmx pentium_pro " : " "); 2606 } 2607 /*FALLTHROUGH*/ 2608 case X86_VENDOR_Cyrix: 2609 /* 2610 * The Cyrix 6x86 does not have any Pentium features 2611 * accessible while not at privilege level 0. 2612 */ 2613 if (x86_feature & X86_CPUID) { 2614 (void) strcat(tp, "pentium"); 2615 (void) strcat(tp, x86_feature & X86_MMX ? 2616 "+mmx pentium " : " "); 2617 } 2618 break; 2619 default: 2620 break; 2621 } 2622 (void) strcat(tp, "i486 i386 i86"); 2623 len = strlen(tp) + 1; /* account for NULL at end of string */ 2624 isa_list = strcpy(kmem_alloc(len, KM_SLEEP), tp); 2625 kmem_free(tp, TBUFSIZE); 2626 2627 #undef TBUFSIZE 2628 } 2629 2630 2631 #ifdef __amd64 2632 2633 void * 2634 device_arena_alloc(size_t size, int vm_flag) 2635 { 2636 return (vmem_alloc(device_arena, size, vm_flag)); 2637 } 2638 2639 void 2640 device_arena_free(void *vaddr, size_t size) 2641 { 2642 vmem_free(device_arena, vaddr, size); 2643 } 2644 2645 #else /* __i386 */ 2646 2647 void * 2648 device_arena_alloc(size_t size, int vm_flag) 2649 { 2650 caddr_t vaddr; 2651 uintptr_t v; 2652 size_t start; 2653 size_t end; 2654 2655 vaddr = vmem_alloc(heap_arena, size, vm_flag); 2656 if (vaddr == NULL) 2657 return (NULL); 2658 2659 v = (uintptr_t)vaddr; 2660 ASSERT(v >= kernelbase); 2661 ASSERT(v + size <= valloc_base); 2662 2663 start = btop(v - kernelbase); 2664 end = btop(v + size - 1 - kernelbase); 2665 ASSERT(start < toxic_bit_map_len); 2666 ASSERT(end < toxic_bit_map_len); 2667 2668 while (start <= end) { 2669 BT_ATOMIC_SET(toxic_bit_map, start); 2670 ++start; 2671 } 2672 return (vaddr); 2673 } 2674 2675 void 2676 device_arena_free(void *vaddr, size_t size) 2677 { 2678 uintptr_t v = (uintptr_t)vaddr; 2679 size_t start; 2680 size_t end; 2681 2682 ASSERT(v >= kernelbase); 2683 ASSERT(v + size <= valloc_base); 2684 2685 start = btop(v - kernelbase); 2686 end = btop(v + size - 1 - kernelbase); 2687 ASSERT(start < toxic_bit_map_len); 2688 ASSERT(end < toxic_bit_map_len); 2689 2690 while (start <= end) { 2691 ASSERT(BT_TEST(toxic_bit_map, start) != 0); 2692 BT_ATOMIC_CLEAR(toxic_bit_map, start); 2693 ++start; 2694 } 2695 vmem_free(heap_arena, vaddr, size); 2696 } 2697 2698 /* 2699 * returns 1st address in range that is in device arena, or NULL 2700 * if len is not NULL it returns the length of the toxic range 2701 */ 2702 void * 2703 device_arena_contains(void *vaddr, size_t size, size_t *len) 2704 { 2705 uintptr_t v = (uintptr_t)vaddr; 2706 uintptr_t eaddr = v + size; 2707 size_t start; 2708 size_t end; 2709 2710 /* 2711 * if called very early by kmdb, just return NULL 2712 */ 2713 if (toxic_bit_map == NULL) 2714 return (NULL); 2715 2716 /* 2717 * First check if we're completely outside the bitmap range. 2718 */ 2719 if (v >= valloc_base || eaddr < kernelbase) 2720 return (NULL); 2721 2722 /* 2723 * Trim ends of search to look at only what the bitmap covers. 2724 */ 2725 if (v < kernelbase) 2726 v = kernelbase; 2727 start = btop(v - kernelbase); 2728 end = btop(eaddr - kernelbase); 2729 if (end >= toxic_bit_map_len) 2730 end = toxic_bit_map_len; 2731 2732 if (bt_range(toxic_bit_map, &start, &end, end) == 0) 2733 return (NULL); 2734 2735 v = kernelbase + ptob(start); 2736 if (len != NULL) 2737 *len = ptob(end - start); 2738 return ((void *)v); 2739 } 2740 2741 #endif /* __i386 */ 2742