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 (c) 1993, 2010, Oracle and/or its affiliates. All rights reserved. 23 * Copyright 2012 DEY Storage Systems, Inc. All rights reserved. 24 * Copyright 2013 Nexenta Systems, Inc. All rights reserved. 25 * Copyright 2015 Joyent, Inc. 26 */ 27 /* 28 * Copyright (c) 2010, Intel Corporation. 29 * All rights reserved. 30 */ 31 32 #include <sys/types.h> 33 #include <sys/t_lock.h> 34 #include <sys/param.h> 35 #include <sys/sysmacros.h> 36 #include <sys/signal.h> 37 #include <sys/systm.h> 38 #include <sys/user.h> 39 #include <sys/mman.h> 40 #include <sys/vm.h> 41 #include <sys/conf.h> 42 #include <sys/avintr.h> 43 #include <sys/autoconf.h> 44 #include <sys/disp.h> 45 #include <sys/class.h> 46 #include <sys/bitmap.h> 47 48 #include <sys/privregs.h> 49 50 #include <sys/proc.h> 51 #include <sys/buf.h> 52 #include <sys/kmem.h> 53 #include <sys/mem.h> 54 #include <sys/kstat.h> 55 56 #include <sys/reboot.h> 57 58 #include <sys/cred.h> 59 #include <sys/vnode.h> 60 #include <sys/file.h> 61 62 #include <sys/procfs.h> 63 64 #include <sys/vfs.h> 65 #include <sys/cmn_err.h> 66 #include <sys/utsname.h> 67 #include <sys/debug.h> 68 #include <sys/kdi.h> 69 70 #include <sys/dumphdr.h> 71 #include <sys/bootconf.h> 72 #include <sys/memlist_plat.h> 73 #include <sys/varargs.h> 74 #include <sys/promif.h> 75 #include <sys/modctl.h> 76 77 #include <sys/sunddi.h> 78 #include <sys/sunndi.h> 79 #include <sys/ndi_impldefs.h> 80 #include <sys/ddidmareq.h> 81 #include <sys/psw.h> 82 #include <sys/regset.h> 83 #include <sys/clock.h> 84 #include <sys/pte.h> 85 #include <sys/tss.h> 86 #include <sys/stack.h> 87 #include <sys/trap.h> 88 #include <sys/fp.h> 89 #include <vm/kboot_mmu.h> 90 #include <vm/anon.h> 91 #include <vm/as.h> 92 #include <vm/page.h> 93 #include <vm/seg.h> 94 #include <vm/seg_dev.h> 95 #include <vm/seg_kmem.h> 96 #include <vm/seg_kpm.h> 97 #include <vm/seg_map.h> 98 #include <vm/seg_vn.h> 99 #include <vm/seg_kp.h> 100 #include <sys/memnode.h> 101 #include <vm/vm_dep.h> 102 #include <sys/thread.h> 103 #include <sys/sysconf.h> 104 #include <sys/vm_machparam.h> 105 #include <sys/archsystm.h> 106 #include <sys/machsystm.h> 107 #include <vm/hat.h> 108 #include <vm/hat_i86.h> 109 #include <sys/pmem.h> 110 #include <sys/smp_impldefs.h> 111 #include <sys/x86_archext.h> 112 #include <sys/cpuvar.h> 113 #include <sys/segments.h> 114 #include <sys/clconf.h> 115 #include <sys/kobj.h> 116 #include <sys/kobj_lex.h> 117 #include <sys/cpc_impl.h> 118 #include <sys/cpu_module.h> 119 #include <sys/smbios.h> 120 #include <sys/debug_info.h> 121 #include <sys/bootinfo.h> 122 #include <sys/ddi_periodic.h> 123 #include <sys/systeminfo.h> 124 #include <sys/multiboot.h> 125 #include <sys/ramdisk.h> 126 127 #ifdef __xpv 128 129 #include <sys/hypervisor.h> 130 #include <sys/xen_mmu.h> 131 #include <sys/evtchn_impl.h> 132 #include <sys/gnttab.h> 133 #include <sys/xpv_panic.h> 134 #include <xen/sys/xenbus_comms.h> 135 #include <xen/public/physdev.h> 136 137 extern void xen_late_startup(void); 138 139 struct xen_evt_data cpu0_evt_data; 140 141 #else /* __xpv */ 142 #include <sys/memlist_impl.h> 143 144 extern void mem_config_init(void); 145 #endif /* __xpv */ 146 147 extern void progressbar_init(void); 148 extern void brand_init(void); 149 extern void pcf_init(void); 150 extern void pg_init(void); 151 extern void ssp_init(void); 152 153 extern int size_pse_array(pgcnt_t, int); 154 155 #if defined(_SOFT_HOSTID) 156 157 #include <sys/rtc.h> 158 159 static int32_t set_soft_hostid(void); 160 static char hostid_file[] = "/etc/hostid"; 161 162 #endif 163 164 void *gfx_devinfo_list; 165 166 #if defined(__amd64) && !defined(__xpv) 167 extern void immu_startup(void); 168 #endif 169 170 /* 171 * XXX make declaration below "static" when drivers no longer use this 172 * interface. 173 */ 174 extern caddr_t p0_va; /* Virtual address for accessing physical page 0 */ 175 176 /* 177 * segkp 178 */ 179 extern int segkp_fromheap; 180 181 static void kvm_init(void); 182 static void startup_init(void); 183 static void startup_memlist(void); 184 static void startup_kmem(void); 185 static void startup_modules(void); 186 static void startup_vm(void); 187 static void startup_end(void); 188 static void layout_kernel_va(void); 189 190 /* 191 * Declare these as initialized data so we can patch them. 192 */ 193 #ifdef __i386 194 195 /* 196 * Due to virtual address space limitations running in 32 bit mode, restrict 197 * the amount of physical memory configured to a max of PHYSMEM pages (16g). 198 * 199 * If the physical max memory size of 64g were allowed to be configured, the 200 * size of user virtual address space will be less than 1g. A limited user 201 * address space greatly reduces the range of applications that can run. 202 * 203 * If more physical memory than PHYSMEM is required, users should preferably 204 * run in 64 bit mode which has far looser virtual address space limitations. 205 * 206 * If 64 bit mode is not available (as in IA32) and/or more physical memory 207 * than PHYSMEM is required in 32 bit mode, physmem can be set to the desired 208 * value or to 0 (to configure all available memory) via eeprom(1M). kernelbase 209 * should also be carefully tuned to balance out the need of the user 210 * application while minimizing the risk of kernel heap exhaustion due to 211 * kernelbase being set too high. 212 */ 213 #define PHYSMEM 0x400000 214 215 #else /* __amd64 */ 216 217 /* 218 * For now we can handle memory with physical addresses up to about 219 * 64 Terabytes. This keeps the kernel above the VA hole, leaving roughly 220 * half the VA space for seg_kpm. When systems get bigger than 64TB this 221 * code will need revisiting. There is an implicit assumption that there 222 * are no *huge* holes in the physical address space too. 223 */ 224 #define TERABYTE (1ul << 40) 225 #define PHYSMEM_MAX64 mmu_btop(64 * TERABYTE) 226 #define PHYSMEM PHYSMEM_MAX64 227 #define AMD64_VA_HOLE_END 0xFFFF800000000000ul 228 229 #endif /* __amd64 */ 230 231 pgcnt_t physmem = PHYSMEM; 232 pgcnt_t obp_pages; /* Memory used by PROM for its text and data */ 233 234 char *kobj_file_buf; 235 int kobj_file_bufsize; /* set in /etc/system */ 236 237 /* Global variables for MP support. Used in mp_startup */ 238 caddr_t rm_platter_va = 0; 239 uint32_t rm_platter_pa; 240 241 int auto_lpg_disable = 1; 242 243 /* 244 * Some CPUs have holes in the middle of the 64-bit virtual address range. 245 */ 246 uintptr_t hole_start, hole_end; 247 248 /* 249 * kpm mapping window 250 */ 251 caddr_t kpm_vbase; 252 size_t kpm_size; 253 static int kpm_desired; 254 #ifdef __amd64 255 static uintptr_t segkpm_base = (uintptr_t)SEGKPM_BASE; 256 #endif 257 258 /* 259 * Configuration parameters set at boot time. 260 */ 261 262 caddr_t econtig; /* end of first block of contiguous kernel */ 263 264 struct bootops *bootops = 0; /* passed in from boot */ 265 struct bootops **bootopsp; 266 struct boot_syscalls *sysp; /* passed in from boot */ 267 268 char bootblock_fstype[16]; 269 270 char kern_bootargs[OBP_MAXPATHLEN]; 271 char kern_bootfile[OBP_MAXPATHLEN]; 272 273 /* 274 * ZFS zio segment. This allows us to exclude large portions of ZFS data that 275 * gets cached in kmem caches on the heap. If this is set to zero, we allocate 276 * zio buffers from their own segment, otherwise they are allocated from the 277 * heap. The optimization of allocating zio buffers from their own segment is 278 * only valid on 64-bit kernels. 279 */ 280 #if defined(__amd64) 281 int segzio_fromheap = 0; 282 #else 283 int segzio_fromheap = 1; 284 #endif 285 286 /* 287 * Give folks an escape hatch for disabling SMAP via kmdb. Doesn't work 288 * post-boot. 289 */ 290 int disable_smap = 0; 291 292 /* 293 * new memory fragmentations are possible in startup() due to BOP_ALLOCs. this 294 * depends on number of BOP_ALLOC calls made and requested size, memory size 295 * combination and whether boot.bin memory needs to be freed. 296 */ 297 #define POSS_NEW_FRAGMENTS 12 298 299 /* 300 * VM data structures 301 */ 302 long page_hashsz; /* Size of page hash table (power of two) */ 303 unsigned int page_hashsz_shift; /* log2(page_hashsz) */ 304 struct page *pp_base; /* Base of initial system page struct array */ 305 struct page **page_hash; /* Page hash table */ 306 pad_mutex_t *pse_mutex; /* Locks protecting pp->p_selock */ 307 size_t pse_table_size; /* Number of mutexes in pse_mutex[] */ 308 int pse_shift; /* log2(pse_table_size) */ 309 struct seg ktextseg; /* Segment used for kernel executable image */ 310 struct seg kvalloc; /* Segment used for "valloc" mapping */ 311 struct seg kpseg; /* Segment used for pageable kernel virt mem */ 312 struct seg kmapseg; /* Segment used for generic kernel mappings */ 313 struct seg kdebugseg; /* Segment used for the kernel debugger */ 314 315 struct seg *segkmap = &kmapseg; /* Kernel generic mapping segment */ 316 static struct seg *segmap = &kmapseg; /* easier to use name for in here */ 317 318 struct seg *segkp = &kpseg; /* Pageable kernel virtual memory segment */ 319 320 #if defined(__amd64) 321 struct seg kvseg_core; /* Segment used for the core heap */ 322 struct seg kpmseg; /* Segment used for physical mapping */ 323 struct seg *segkpm = &kpmseg; /* 64bit kernel physical mapping segment */ 324 #else 325 struct seg *segkpm = NULL; /* Unused on IA32 */ 326 #endif 327 328 caddr_t segkp_base; /* Base address of segkp */ 329 caddr_t segzio_base; /* Base address of segzio */ 330 #if defined(__amd64) 331 pgcnt_t segkpsize = btop(SEGKPDEFSIZE); /* size of segkp segment in pages */ 332 #else 333 pgcnt_t segkpsize = 0; 334 #endif 335 pgcnt_t segziosize = 0; /* size of zio segment in pages */ 336 337 /* 338 * A static DR page_t VA map is reserved that can map the page structures 339 * for a domain's entire RA space. The pages that back this space are 340 * dynamically allocated and need not be physically contiguous. The DR 341 * map size is derived from KPM size. 342 * This mechanism isn't used by x86 yet, so just stubs here. 343 */ 344 int ppvm_enable = 0; /* Static virtual map for page structs */ 345 page_t *ppvm_base = NULL; /* Base of page struct map */ 346 pgcnt_t ppvm_size = 0; /* Size of page struct map */ 347 348 /* 349 * VA range available to the debugger 350 */ 351 const caddr_t kdi_segdebugbase = (const caddr_t)SEGDEBUGBASE; 352 const size_t kdi_segdebugsize = SEGDEBUGSIZE; 353 354 struct memseg *memseg_base; 355 struct vnode unused_pages_vp; 356 357 #define FOURGB 0x100000000LL 358 359 struct memlist *memlist; 360 361 caddr_t s_text; /* start of kernel text segment */ 362 caddr_t e_text; /* end of kernel text segment */ 363 caddr_t s_data; /* start of kernel data segment */ 364 caddr_t e_data; /* end of kernel data segment */ 365 caddr_t modtext; /* start of loadable module text reserved */ 366 caddr_t e_modtext; /* end of loadable module text reserved */ 367 caddr_t moddata; /* start of loadable module data reserved */ 368 caddr_t e_moddata; /* end of loadable module data reserved */ 369 370 struct memlist *phys_install; /* Total installed physical memory */ 371 struct memlist *phys_avail; /* Total available physical memory */ 372 struct memlist *bios_rsvd; /* Bios reserved memory */ 373 374 /* 375 * kphysm_init returns the number of pages that were processed 376 */ 377 static pgcnt_t kphysm_init(page_t *, pgcnt_t); 378 379 #define IO_PROP_SIZE 64 /* device property size */ 380 381 /* 382 * a couple useful roundup macros 383 */ 384 #define ROUND_UP_PAGE(x) \ 385 ((uintptr_t)P2ROUNDUP((uintptr_t)(x), (uintptr_t)MMU_PAGESIZE)) 386 #define ROUND_UP_LPAGE(x) \ 387 ((uintptr_t)P2ROUNDUP((uintptr_t)(x), mmu.level_size[1])) 388 #define ROUND_UP_4MEG(x) \ 389 ((uintptr_t)P2ROUNDUP((uintptr_t)(x), (uintptr_t)FOUR_MEG)) 390 #define ROUND_UP_TOPLEVEL(x) \ 391 ((uintptr_t)P2ROUNDUP((uintptr_t)(x), mmu.level_size[mmu.max_level])) 392 393 /* 394 * 32-bit Kernel's Virtual memory layout. 395 * +-----------------------+ 396 * | | 397 * 0xFFC00000 -|-----------------------|- ARGSBASE 398 * | debugger | 399 * 0xFF800000 -|-----------------------|- SEGDEBUGBASE 400 * | Kernel Data | 401 * 0xFEC00000 -|-----------------------| 402 * | Kernel Text | 403 * 0xFE800000 -|-----------------------|- KERNEL_TEXT (0xFB400000 on Xen) 404 * |--- GDT ---|- GDT page (GDT_VA) 405 * |--- debug info ---|- debug info (DEBUG_INFO_VA) 406 * | | 407 * | page_t structures | 408 * | memsegs, memlists, | 409 * | page hash, etc. | 410 * --- -|-----------------------|- ekernelheap, valloc_base (floating) 411 * | | (segkp is just an arena in the heap) 412 * | | 413 * | kvseg | 414 * | | 415 * | | 416 * --- -|-----------------------|- kernelheap (floating) 417 * | Segkmap | 418 * 0xC3002000 -|-----------------------|- segmap_start (floating) 419 * | Red Zone | 420 * 0xC3000000 -|-----------------------|- kernelbase / userlimit (floating) 421 * | | || 422 * | Shared objects | \/ 423 * | | 424 * : : 425 * | user data | 426 * |-----------------------| 427 * | user text | 428 * 0x08048000 -|-----------------------| 429 * | user stack | 430 * : : 431 * | invalid | 432 * 0x00000000 +-----------------------+ 433 * 434 * 435 * 64-bit Kernel's Virtual memory layout. (assuming 64 bit app) 436 * +-----------------------+ 437 * | | 438 * 0xFFFFFFFF.FFC00000 |-----------------------|- ARGSBASE 439 * | debugger (?) | 440 * 0xFFFFFFFF.FF800000 |-----------------------|- SEGDEBUGBASE 441 * | unused | 442 * +-----------------------+ 443 * | Kernel Data | 444 * 0xFFFFFFFF.FBC00000 |-----------------------| 445 * | Kernel Text | 446 * 0xFFFFFFFF.FB800000 |-----------------------|- KERNEL_TEXT 447 * |--- GDT ---|- GDT page (GDT_VA) 448 * |--- debug info ---|- debug info (DEBUG_INFO_VA) 449 * | | 450 * | Core heap | (used for loadable modules) 451 * 0xFFFFFFFF.C0000000 |-----------------------|- core_base / ekernelheap 452 * | Kernel | 453 * | heap | 454 * 0xFFFFFXXX.XXX00000 |-----------------------|- kernelheap (floating) 455 * | segmap | 456 * 0xFFFFFXXX.XXX00000 |-----------------------|- segmap_start (floating) 457 * | device mappings | 458 * 0xFFFFFXXX.XXX00000 |-----------------------|- toxic_addr (floating) 459 * | segzio | 460 * 0xFFFFFXXX.XXX00000 |-----------------------|- segzio_base (floating) 461 * | segkp | 462 * --- |-----------------------|- segkp_base (floating) 463 * | page_t structures | valloc_base + valloc_sz 464 * | memsegs, memlists, | 465 * | page hash, etc. | 466 * 0xFFFFFF00.00000000 |-----------------------|- valloc_base (lower if > 1TB) 467 * | segkpm | 468 * 0xFFFFFE00.00000000 |-----------------------| 469 * | Red Zone | 470 * 0xFFFFFD80.00000000 |-----------------------|- KERNELBASE (lower if > 1TB) 471 * | User stack |- User space memory 472 * | | 473 * | shared objects, etc | (grows downwards) 474 * : : 475 * | | 476 * 0xFFFF8000.00000000 |-----------------------| 477 * | | 478 * | VA Hole / unused | 479 * | | 480 * 0x00008000.00000000 |-----------------------| 481 * | | 482 * | | 483 * : : 484 * | user heap | (grows upwards) 485 * | | 486 * | user data | 487 * |-----------------------| 488 * | user text | 489 * 0x00000000.04000000 |-----------------------| 490 * | invalid | 491 * 0x00000000.00000000 +-----------------------+ 492 * 493 * A 32 bit app on the 64 bit kernel sees the same layout as on the 32 bit 494 * kernel, except that userlimit is raised to 0xfe000000 495 * 496 * Floating values: 497 * 498 * valloc_base: start of the kernel's memory management/tracking data 499 * structures. This region contains page_t structures for 500 * physical memory, memsegs, memlists, and the page hash. 501 * 502 * core_base: start of the kernel's "core" heap area on 64-bit systems. 503 * This area is intended to be used for global data as well as for module 504 * text/data that does not fit into the nucleus pages. The core heap is 505 * restricted to a 2GB range, allowing every address within it to be 506 * accessed using rip-relative addressing 507 * 508 * ekernelheap: end of kernelheap and start of segmap. 509 * 510 * kernelheap: start of kernel heap. On 32-bit systems, this starts right 511 * above a red zone that separates the user's address space from the 512 * kernel's. On 64-bit systems, it sits above segkp and segkpm. 513 * 514 * segmap_start: start of segmap. The length of segmap can be modified 515 * through eeprom. The default length is 16MB on 32-bit systems and 64MB 516 * on 64-bit systems. 517 * 518 * kernelbase: On a 32-bit kernel the default value of 0xd4000000 will be 519 * decreased by 2X the size required for page_t. This allows the kernel 520 * heap to grow in size with physical memory. With sizeof(page_t) == 80 521 * bytes, the following shows the values of kernelbase and kernel heap 522 * sizes for different memory configurations (assuming default segmap and 523 * segkp sizes). 524 * 525 * mem size for kernelbase kernel heap 526 * size page_t's size 527 * ---- --------- ---------- ----------- 528 * 1gb 0x01400000 0xd1800000 684MB 529 * 2gb 0x02800000 0xcf000000 704MB 530 * 4gb 0x05000000 0xca000000 744MB 531 * 6gb 0x07800000 0xc5000000 784MB 532 * 8gb 0x0a000000 0xc0000000 824MB 533 * 16gb 0x14000000 0xac000000 984MB 534 * 32gb 0x28000000 0x84000000 1304MB 535 * 64gb 0x50000000 0x34000000 1944MB (*) 536 * 537 * kernelbase is less than the abi minimum of 0xc0000000 for memory 538 * configurations above 8gb. 539 * 540 * (*) support for memory configurations above 32gb will require manual tuning 541 * of kernelbase to balance out the need of user applications. 542 */ 543 544 /* real-time-clock initialization parameters */ 545 extern time_t process_rtc_config_file(void); 546 547 uintptr_t kernelbase; 548 uintptr_t postbootkernelbase; /* not set till boot loader is gone */ 549 uintptr_t eprom_kernelbase; 550 size_t segmapsize; 551 uintptr_t segmap_start; 552 int segmapfreelists; 553 pgcnt_t npages; 554 pgcnt_t orig_npages; 555 size_t core_size; /* size of "core" heap */ 556 uintptr_t core_base; /* base address of "core" heap */ 557 558 /* 559 * List of bootstrap pages. We mark these as allocated in startup. 560 * release_bootstrap() will free them when we're completely done with 561 * the bootstrap. 562 */ 563 static page_t *bootpages; 564 565 /* 566 * boot time pages that have a vnode from the ramdisk will keep that forever. 567 */ 568 static page_t *rd_pages; 569 570 /* 571 * Lower 64K 572 */ 573 static page_t *lower_pages = NULL; 574 static int lower_pages_count = 0; 575 576 struct system_hardware system_hardware; 577 578 /* 579 * Enable some debugging messages concerning memory usage... 580 */ 581 static void 582 print_memlist(char *title, struct memlist *mp) 583 { 584 prom_printf("MEMLIST: %s:\n", title); 585 while (mp != NULL) { 586 prom_printf("\tAddress 0x%" PRIx64 ", size 0x%" PRIx64 "\n", 587 mp->ml_address, mp->ml_size); 588 mp = mp->ml_next; 589 } 590 } 591 592 /* 593 * XX64 need a comment here.. are these just default values, surely 594 * we read the "cpuid" type information to figure this out. 595 */ 596 int l2cache_sz = 0x80000; 597 int l2cache_linesz = 0x40; 598 int l2cache_assoc = 1; 599 600 static size_t textrepl_min_gb = 10; 601 602 /* 603 * on 64 bit we use a predifined VA range for mapping devices in the kernel 604 * on 32 bit the mappings are intermixed in the heap, so we use a bit map 605 */ 606 #ifdef __amd64 607 608 vmem_t *device_arena; 609 uintptr_t toxic_addr = (uintptr_t)NULL; 610 size_t toxic_size = 1024 * 1024 * 1024; /* Sparc uses 1 gig too */ 611 612 #else /* __i386 */ 613 614 ulong_t *toxic_bit_map; /* one bit for each 4k of VA in heap_arena */ 615 size_t toxic_bit_map_len = 0; /* in bits */ 616 617 #endif /* __i386 */ 618 619 /* 620 * Simple boot time debug facilities 621 */ 622 static char *prm_dbg_str[] = { 623 "%s:%d: '%s' is 0x%x\n", 624 "%s:%d: '%s' is 0x%llx\n" 625 }; 626 627 int prom_debug; 628 629 #define PRM_DEBUG(q) if (prom_debug) \ 630 prom_printf(prm_dbg_str[sizeof (q) >> 3], "startup.c", __LINE__, #q, q); 631 #define PRM_POINT(q) if (prom_debug) \ 632 prom_printf("%s:%d: %s\n", "startup.c", __LINE__, q); 633 634 /* 635 * This structure is used to keep track of the intial allocations 636 * done in startup_memlist(). The value of NUM_ALLOCATIONS needs to 637 * be >= the number of ADD_TO_ALLOCATIONS() executed in the code. 638 */ 639 #define NUM_ALLOCATIONS 8 640 int num_allocations = 0; 641 struct { 642 void **al_ptr; 643 size_t al_size; 644 } allocations[NUM_ALLOCATIONS]; 645 size_t valloc_sz = 0; 646 uintptr_t valloc_base; 647 648 #define ADD_TO_ALLOCATIONS(ptr, size) { \ 649 size = ROUND_UP_PAGE(size); \ 650 if (num_allocations == NUM_ALLOCATIONS) \ 651 panic("too many ADD_TO_ALLOCATIONS()"); \ 652 allocations[num_allocations].al_ptr = (void**)&ptr; \ 653 allocations[num_allocations].al_size = size; \ 654 valloc_sz += size; \ 655 ++num_allocations; \ 656 } 657 658 /* 659 * Allocate all the initial memory needed by the page allocator. 660 */ 661 static void 662 perform_allocations(void) 663 { 664 caddr_t mem; 665 int i; 666 int valloc_align; 667 668 PRM_DEBUG(valloc_base); 669 PRM_DEBUG(valloc_sz); 670 valloc_align = mmu.level_size[mmu.max_page_level > 0]; 671 mem = BOP_ALLOC(bootops, (caddr_t)valloc_base, valloc_sz, valloc_align); 672 if (mem != (caddr_t)valloc_base) 673 panic("BOP_ALLOC() failed"); 674 bzero(mem, valloc_sz); 675 for (i = 0; i < num_allocations; ++i) { 676 *allocations[i].al_ptr = (void *)mem; 677 mem += allocations[i].al_size; 678 } 679 } 680 681 /* 682 * Set up and enable SMAP now before we start other CPUs, but after the kernel's 683 * VM has been set up so we can use hot_patch_kernel_text(). 684 * 685 * We can only patch 1, 2, or 4 bytes, but not three bytes. So instead, we 686 * replace the four byte word at the patch point. See uts/intel/ia32/ml/copy.s 687 * for more information on what's going on here. 688 */ 689 static void 690 startup_smap(void) 691 { 692 int i; 693 uint32_t inst; 694 uint8_t *instp; 695 char sym[128]; 696 697 extern int _smap_enable_patch_count; 698 extern int _smap_disable_patch_count; 699 700 if (disable_smap != 0) 701 remove_x86_feature(x86_featureset, X86FSET_SMAP); 702 703 if (is_x86_feature(x86_featureset, X86FSET_SMAP) == B_FALSE) 704 return; 705 706 for (i = 0; i < _smap_enable_patch_count; i++) { 707 int sizep; 708 709 VERIFY3U(i, <, _smap_enable_patch_count); 710 VERIFY(snprintf(sym, sizeof (sym), "_smap_enable_patch_%d", i) < 711 sizeof (sym)); 712 instp = (uint8_t *)(void *)kobj_getelfsym(sym, NULL, &sizep); 713 VERIFY(instp != 0); 714 inst = (instp[3] << 24) | (SMAP_CLAC_INSTR & 0x00ffffff); 715 hot_patch_kernel_text((caddr_t)instp, inst, 4); 716 } 717 718 for (i = 0; i < _smap_disable_patch_count; i++) { 719 int sizep; 720 721 VERIFY(snprintf(sym, sizeof (sym), "_smap_disable_patch_%d", 722 i) < sizeof (sym)); 723 instp = (uint8_t *)(void *)kobj_getelfsym(sym, NULL, &sizep); 724 VERIFY(instp != 0); 725 inst = (instp[3] << 24) | (SMAP_STAC_INSTR & 0x00ffffff); 726 hot_patch_kernel_text((caddr_t)instp, inst, 4); 727 } 728 729 hot_patch_kernel_text((caddr_t)smap_enable, SMAP_CLAC_INSTR, 4); 730 hot_patch_kernel_text((caddr_t)smap_disable, SMAP_STAC_INSTR, 4); 731 setcr4(getcr4() | CR4_SMAP); 732 smap_enable(); 733 } 734 735 /* 736 * Our world looks like this at startup time. 737 * 738 * In a 32-bit OS, boot loads the kernel text at 0xfe800000 and kernel data 739 * at 0xfec00000. On a 64-bit OS, kernel text and data are loaded at 740 * 0xffffffff.fe800000 and 0xffffffff.fec00000 respectively. Those 741 * addresses are fixed in the binary at link time. 742 * 743 * On the text page: 744 * unix/genunix/krtld/module text loads. 745 * 746 * On the data page: 747 * unix/genunix/krtld/module data loads. 748 * 749 * Machine-dependent startup code 750 */ 751 void 752 startup(void) 753 { 754 #if !defined(__xpv) 755 extern void startup_pci_bios(void); 756 #endif 757 extern cpuset_t cpu_ready_set; 758 759 /* 760 * Make sure that nobody tries to use sekpm until we have 761 * initialized it properly. 762 */ 763 #if defined(__amd64) 764 kpm_desired = 1; 765 #endif 766 kpm_enable = 0; 767 CPUSET_ONLY(cpu_ready_set, 0); /* cpu 0 is boot cpu */ 768 769 #if defined(__xpv) /* XXPV fix me! */ 770 { 771 extern int segvn_use_regions; 772 segvn_use_regions = 0; 773 } 774 #endif 775 ssp_init(); 776 progressbar_init(); 777 startup_init(); 778 #if defined(__xpv) 779 startup_xen_version(); 780 #endif 781 startup_memlist(); 782 startup_kmem(); 783 startup_vm(); 784 #if !defined(__xpv) 785 /* 786 * Note we need to do this even on fast reboot in order to access 787 * the irq routing table (used for pci labels). 788 */ 789 startup_pci_bios(); 790 startup_smap(); 791 #endif 792 #if defined(__xpv) 793 startup_xen_mca(); 794 #endif 795 startup_modules(); 796 797 startup_end(); 798 } 799 800 static void 801 startup_init() 802 { 803 PRM_POINT("startup_init() starting..."); 804 805 /* 806 * Complete the extraction of cpuid data 807 */ 808 cpuid_pass2(CPU); 809 810 (void) check_boot_version(BOP_GETVERSION(bootops)); 811 812 /* 813 * Check for prom_debug in boot environment 814 */ 815 if (BOP_GETPROPLEN(bootops, "prom_debug") >= 0) { 816 ++prom_debug; 817 PRM_POINT("prom_debug found in boot enviroment"); 818 } 819 820 /* 821 * Collect node, cpu and memory configuration information. 822 */ 823 get_system_configuration(); 824 825 /* 826 * Halt if this is an unsupported processor. 827 */ 828 if (x86_type == X86_TYPE_486 || x86_type == X86_TYPE_CYRIX_486) { 829 printf("\n486 processor (\"%s\") detected.\n", 830 CPU->cpu_brandstr); 831 halt("This processor is not supported by this release " 832 "of Solaris."); 833 } 834 835 PRM_POINT("startup_init() done"); 836 } 837 838 /* 839 * Callback for copy_memlist_filter() to filter nucleus, kadb/kmdb, (ie. 840 * everything mapped above KERNEL_TEXT) pages from phys_avail. Note it 841 * also filters out physical page zero. There is some reliance on the 842 * boot loader allocating only a few contiguous physical memory chunks. 843 */ 844 static void 845 avail_filter(uint64_t *addr, uint64_t *size) 846 { 847 uintptr_t va; 848 uintptr_t next_va; 849 pfn_t pfn; 850 uint64_t pfn_addr; 851 uint64_t pfn_eaddr; 852 uint_t prot; 853 size_t len; 854 uint_t change; 855 856 if (prom_debug) 857 prom_printf("\tFilter: in: a=%" PRIx64 ", s=%" PRIx64 "\n", 858 *addr, *size); 859 860 /* 861 * page zero is required for BIOS.. never make it available 862 */ 863 if (*addr == 0) { 864 *addr += MMU_PAGESIZE; 865 *size -= MMU_PAGESIZE; 866 } 867 868 /* 869 * First we trim from the front of the range. Since kbm_probe() 870 * walks ranges in virtual order, but addr/size are physical, we need 871 * to the list until no changes are seen. This deals with the case 872 * where page "p" is mapped at v, page "p + PAGESIZE" is mapped at w 873 * but w < v. 874 */ 875 do { 876 change = 0; 877 for (va = KERNEL_TEXT; 878 *size > 0 && kbm_probe(&va, &len, &pfn, &prot) != 0; 879 va = next_va) { 880 881 next_va = va + len; 882 pfn_addr = pfn_to_pa(pfn); 883 pfn_eaddr = pfn_addr + len; 884 885 if (pfn_addr <= *addr && pfn_eaddr > *addr) { 886 change = 1; 887 while (*size > 0 && len > 0) { 888 *addr += MMU_PAGESIZE; 889 *size -= MMU_PAGESIZE; 890 len -= MMU_PAGESIZE; 891 } 892 } 893 } 894 if (change && prom_debug) 895 prom_printf("\t\ttrim: a=%" PRIx64 ", s=%" PRIx64 "\n", 896 *addr, *size); 897 } while (change); 898 899 /* 900 * Trim pages from the end of the range. 901 */ 902 for (va = KERNEL_TEXT; 903 *size > 0 && kbm_probe(&va, &len, &pfn, &prot) != 0; 904 va = next_va) { 905 906 next_va = va + len; 907 pfn_addr = pfn_to_pa(pfn); 908 909 if (pfn_addr >= *addr && pfn_addr < *addr + *size) 910 *size = pfn_addr - *addr; 911 } 912 913 if (prom_debug) 914 prom_printf("\tFilter out: a=%" PRIx64 ", s=%" PRIx64 "\n", 915 *addr, *size); 916 } 917 918 static void 919 kpm_init() 920 { 921 struct segkpm_crargs b; 922 923 /* 924 * These variables were all designed for sfmmu in which segkpm is 925 * mapped using a single pagesize - either 8KB or 4MB. On x86, we 926 * might use 2+ page sizes on a single machine, so none of these 927 * variables have a single correct value. They are set up as if we 928 * always use a 4KB pagesize, which should do no harm. In the long 929 * run, we should get rid of KPM's assumption that only a single 930 * pagesize is used. 931 */ 932 kpm_pgshft = MMU_PAGESHIFT; 933 kpm_pgsz = MMU_PAGESIZE; 934 kpm_pgoff = MMU_PAGEOFFSET; 935 kpmp2pshft = 0; 936 kpmpnpgs = 1; 937 ASSERT(((uintptr_t)kpm_vbase & (kpm_pgsz - 1)) == 0); 938 939 PRM_POINT("about to create segkpm"); 940 rw_enter(&kas.a_lock, RW_WRITER); 941 942 if (seg_attach(&kas, kpm_vbase, kpm_size, segkpm) < 0) 943 panic("cannot attach segkpm"); 944 945 b.prot = PROT_READ | PROT_WRITE; 946 b.nvcolors = 1; 947 948 if (segkpm_create(segkpm, (caddr_t)&b) != 0) 949 panic("segkpm_create segkpm"); 950 951 rw_exit(&kas.a_lock); 952 } 953 954 /* 955 * The debug info page provides enough information to allow external 956 * inspectors (e.g. when running under a hypervisor) to bootstrap 957 * themselves into allowing full-blown kernel debugging. 958 */ 959 static void 960 init_debug_info(void) 961 { 962 caddr_t mem; 963 debug_info_t *di; 964 965 #ifndef __lint 966 ASSERT(sizeof (debug_info_t) < MMU_PAGESIZE); 967 #endif 968 969 mem = BOP_ALLOC(bootops, (caddr_t)DEBUG_INFO_VA, MMU_PAGESIZE, 970 MMU_PAGESIZE); 971 972 if (mem != (caddr_t)DEBUG_INFO_VA) 973 panic("BOP_ALLOC() failed"); 974 bzero(mem, MMU_PAGESIZE); 975 976 di = (debug_info_t *)mem; 977 978 di->di_magic = DEBUG_INFO_MAGIC; 979 di->di_version = DEBUG_INFO_VERSION; 980 di->di_modules = (uintptr_t)&modules; 981 di->di_s_text = (uintptr_t)s_text; 982 di->di_e_text = (uintptr_t)e_text; 983 di->di_s_data = (uintptr_t)s_data; 984 di->di_e_data = (uintptr_t)e_data; 985 di->di_hat_htable_off = offsetof(hat_t, hat_htable); 986 di->di_ht_pfn_off = offsetof(htable_t, ht_pfn); 987 } 988 989 /* 990 * Build the memlists and other kernel essential memory system data structures. 991 * This is everything at valloc_base. 992 */ 993 static void 994 startup_memlist(void) 995 { 996 size_t memlist_sz; 997 size_t memseg_sz; 998 size_t pagehash_sz; 999 size_t pp_sz; 1000 uintptr_t va; 1001 size_t len; 1002 uint_t prot; 1003 pfn_t pfn; 1004 int memblocks; 1005 pfn_t rsvd_high_pfn; 1006 pgcnt_t rsvd_pgcnt; 1007 size_t rsvdmemlist_sz; 1008 int rsvdmemblocks; 1009 caddr_t pagecolor_mem; 1010 size_t pagecolor_memsz; 1011 caddr_t page_ctrs_mem; 1012 size_t page_ctrs_size; 1013 size_t pse_table_alloc_size; 1014 struct memlist *current; 1015 extern void startup_build_mem_nodes(struct memlist *); 1016 1017 /* XX64 fix these - they should be in include files */ 1018 extern size_t page_coloring_init(uint_t, int, int); 1019 extern void page_coloring_setup(caddr_t); 1020 1021 PRM_POINT("startup_memlist() starting..."); 1022 1023 /* 1024 * Use leftover large page nucleus text/data space for loadable modules. 1025 * Use at most MODTEXT/MODDATA. 1026 */ 1027 len = kbm_nucleus_size; 1028 ASSERT(len > MMU_PAGESIZE); 1029 1030 moddata = (caddr_t)ROUND_UP_PAGE(e_data); 1031 e_moddata = (caddr_t)P2ROUNDUP((uintptr_t)e_data, (uintptr_t)len); 1032 if (e_moddata - moddata > MODDATA) 1033 e_moddata = moddata + MODDATA; 1034 1035 modtext = (caddr_t)ROUND_UP_PAGE(e_text); 1036 e_modtext = (caddr_t)P2ROUNDUP((uintptr_t)e_text, (uintptr_t)len); 1037 if (e_modtext - modtext > MODTEXT) 1038 e_modtext = modtext + MODTEXT; 1039 1040 econtig = e_moddata; 1041 1042 PRM_DEBUG(modtext); 1043 PRM_DEBUG(e_modtext); 1044 PRM_DEBUG(moddata); 1045 PRM_DEBUG(e_moddata); 1046 PRM_DEBUG(econtig); 1047 1048 /* 1049 * Examine the boot loader physical memory map to find out: 1050 * - total memory in system - physinstalled 1051 * - the max physical address - physmax 1052 * - the number of discontiguous segments of memory. 1053 */ 1054 if (prom_debug) 1055 print_memlist("boot physinstalled", 1056 bootops->boot_mem->physinstalled); 1057 installed_top_size_ex(bootops->boot_mem->physinstalled, &physmax, 1058 &physinstalled, &memblocks); 1059 PRM_DEBUG(physmax); 1060 PRM_DEBUG(physinstalled); 1061 PRM_DEBUG(memblocks); 1062 1063 /* 1064 * Compute maximum physical address for memory DR operations. 1065 * Memory DR operations are unsupported on xpv or 32bit OSes. 1066 */ 1067 #ifdef __amd64 1068 if (plat_dr_support_memory()) { 1069 if (plat_dr_physmax == 0) { 1070 uint_t pabits = UINT_MAX; 1071 1072 cpuid_get_addrsize(CPU, &pabits, NULL); 1073 plat_dr_physmax = btop(1ULL << pabits); 1074 } 1075 if (plat_dr_physmax > PHYSMEM_MAX64) 1076 plat_dr_physmax = PHYSMEM_MAX64; 1077 } else 1078 #endif 1079 plat_dr_physmax = 0; 1080 1081 /* 1082 * Examine the bios reserved memory to find out: 1083 * - the number of discontiguous segments of memory. 1084 */ 1085 if (prom_debug) 1086 print_memlist("boot reserved mem", 1087 bootops->boot_mem->rsvdmem); 1088 installed_top_size_ex(bootops->boot_mem->rsvdmem, &rsvd_high_pfn, 1089 &rsvd_pgcnt, &rsvdmemblocks); 1090 PRM_DEBUG(rsvd_high_pfn); 1091 PRM_DEBUG(rsvd_pgcnt); 1092 PRM_DEBUG(rsvdmemblocks); 1093 1094 /* 1095 * Initialize hat's mmu parameters. 1096 * Check for enforce-prot-exec in boot environment. It's used to 1097 * enable/disable support for the page table entry NX bit. 1098 * The default is to enforce PROT_EXEC on processors that support NX. 1099 * Boot seems to round up the "len", but 8 seems to be big enough. 1100 */ 1101 mmu_init(); 1102 1103 #ifdef __i386 1104 /* 1105 * physmax is lowered if there is more memory than can be 1106 * physically addressed in 32 bit (PAE/non-PAE) modes. 1107 */ 1108 if (mmu.pae_hat) { 1109 if (PFN_ABOVE64G(physmax)) { 1110 physinstalled -= (physmax - (PFN_64G - 1)); 1111 physmax = PFN_64G - 1; 1112 } 1113 } else { 1114 if (PFN_ABOVE4G(physmax)) { 1115 physinstalled -= (physmax - (PFN_4G - 1)); 1116 physmax = PFN_4G - 1; 1117 } 1118 } 1119 #endif 1120 1121 startup_build_mem_nodes(bootops->boot_mem->physinstalled); 1122 1123 if (BOP_GETPROPLEN(bootops, "enforce-prot-exec") >= 0) { 1124 int len = BOP_GETPROPLEN(bootops, "enforce-prot-exec"); 1125 char value[8]; 1126 1127 if (len < 8) 1128 (void) BOP_GETPROP(bootops, "enforce-prot-exec", value); 1129 else 1130 (void) strcpy(value, ""); 1131 if (strcmp(value, "off") == 0) 1132 mmu.pt_nx = 0; 1133 } 1134 PRM_DEBUG(mmu.pt_nx); 1135 1136 /* 1137 * We will need page_t's for every page in the system, except for 1138 * memory mapped at or above above the start of the kernel text segment. 1139 * 1140 * pages above e_modtext are attributed to kernel debugger (obp_pages) 1141 */ 1142 npages = physinstalled - 1; /* avail_filter() skips page 0, so "- 1" */ 1143 obp_pages = 0; 1144 va = KERNEL_TEXT; 1145 while (kbm_probe(&va, &len, &pfn, &prot) != 0) { 1146 npages -= len >> MMU_PAGESHIFT; 1147 if (va >= (uintptr_t)e_moddata) 1148 obp_pages += len >> MMU_PAGESHIFT; 1149 va += len; 1150 } 1151 PRM_DEBUG(npages); 1152 PRM_DEBUG(obp_pages); 1153 1154 /* 1155 * If physmem is patched to be non-zero, use it instead of the computed 1156 * value unless it is larger than the actual amount of memory on hand. 1157 */ 1158 if (physmem == 0 || physmem > npages) { 1159 physmem = npages; 1160 } else if (physmem < npages) { 1161 orig_npages = npages; 1162 npages = physmem; 1163 } 1164 PRM_DEBUG(physmem); 1165 1166 /* 1167 * We now compute the sizes of all the initial allocations for 1168 * structures the kernel needs in order do kmem_alloc(). These 1169 * include: 1170 * memsegs 1171 * memlists 1172 * page hash table 1173 * page_t's 1174 * page coloring data structs 1175 */ 1176 memseg_sz = sizeof (struct memseg) * (memblocks + POSS_NEW_FRAGMENTS); 1177 ADD_TO_ALLOCATIONS(memseg_base, memseg_sz); 1178 PRM_DEBUG(memseg_sz); 1179 1180 /* 1181 * Reserve space for memlists. There's no real good way to know exactly 1182 * how much room we'll need, but this should be a good upper bound. 1183 */ 1184 memlist_sz = ROUND_UP_PAGE(2 * sizeof (struct memlist) * 1185 (memblocks + POSS_NEW_FRAGMENTS)); 1186 ADD_TO_ALLOCATIONS(memlist, memlist_sz); 1187 PRM_DEBUG(memlist_sz); 1188 1189 /* 1190 * Reserve space for bios reserved memlists. 1191 */ 1192 rsvdmemlist_sz = ROUND_UP_PAGE(2 * sizeof (struct memlist) * 1193 (rsvdmemblocks + POSS_NEW_FRAGMENTS)); 1194 ADD_TO_ALLOCATIONS(bios_rsvd, rsvdmemlist_sz); 1195 PRM_DEBUG(rsvdmemlist_sz); 1196 1197 /* LINTED */ 1198 ASSERT(P2SAMEHIGHBIT((1 << PP_SHIFT), sizeof (struct page))); 1199 /* 1200 * The page structure hash table size is a power of 2 1201 * such that the average hash chain length is PAGE_HASHAVELEN. 1202 */ 1203 page_hashsz = npages / PAGE_HASHAVELEN; 1204 page_hashsz_shift = highbit(page_hashsz); 1205 page_hashsz = 1 << page_hashsz_shift; 1206 pagehash_sz = sizeof (struct page *) * page_hashsz; 1207 ADD_TO_ALLOCATIONS(page_hash, pagehash_sz); 1208 PRM_DEBUG(pagehash_sz); 1209 1210 /* 1211 * Set aside room for the page structures themselves. 1212 */ 1213 PRM_DEBUG(npages); 1214 pp_sz = sizeof (struct page) * npages; 1215 ADD_TO_ALLOCATIONS(pp_base, pp_sz); 1216 PRM_DEBUG(pp_sz); 1217 1218 /* 1219 * determine l2 cache info and memory size for page coloring 1220 */ 1221 (void) getl2cacheinfo(CPU, 1222 &l2cache_sz, &l2cache_linesz, &l2cache_assoc); 1223 pagecolor_memsz = 1224 page_coloring_init(l2cache_sz, l2cache_linesz, l2cache_assoc); 1225 ADD_TO_ALLOCATIONS(pagecolor_mem, pagecolor_memsz); 1226 PRM_DEBUG(pagecolor_memsz); 1227 1228 page_ctrs_size = page_ctrs_sz(); 1229 ADD_TO_ALLOCATIONS(page_ctrs_mem, page_ctrs_size); 1230 PRM_DEBUG(page_ctrs_size); 1231 1232 /* 1233 * Allocate the array that protects pp->p_selock. 1234 */ 1235 pse_shift = size_pse_array(physmem, max_ncpus); 1236 pse_table_size = 1 << pse_shift; 1237 pse_table_alloc_size = pse_table_size * sizeof (pad_mutex_t); 1238 ADD_TO_ALLOCATIONS(pse_mutex, pse_table_alloc_size); 1239 1240 #if defined(__amd64) 1241 valloc_sz = ROUND_UP_LPAGE(valloc_sz); 1242 valloc_base = VALLOC_BASE; 1243 1244 /* 1245 * The default values of VALLOC_BASE and SEGKPM_BASE should work 1246 * for values of physmax up to 1 Terabyte. They need adjusting when 1247 * memory is at addresses above 1 TB. When adjusted, segkpm_base must 1248 * be aligned on KERNEL_REDZONE_SIZE boundary (span of top level pte). 1249 */ 1250 if (physmax + 1 > mmu_btop(TERABYTE) || 1251 plat_dr_physmax > mmu_btop(TERABYTE)) { 1252 uint64_t kpm_resv_amount = mmu_ptob(physmax + 1); 1253 1254 if (kpm_resv_amount < mmu_ptob(plat_dr_physmax)) { 1255 kpm_resv_amount = mmu_ptob(plat_dr_physmax); 1256 } 1257 1258 segkpm_base = -(P2ROUNDUP((2 * kpm_resv_amount), 1259 KERNEL_REDZONE_SIZE)); /* down from top VA */ 1260 1261 /* make sure we leave some space for user apps above hole */ 1262 segkpm_base = MAX(segkpm_base, AMD64_VA_HOLE_END + TERABYTE); 1263 if (segkpm_base > SEGKPM_BASE) 1264 segkpm_base = SEGKPM_BASE; 1265 PRM_DEBUG(segkpm_base); 1266 1267 valloc_base = segkpm_base + P2ROUNDUP(kpm_resv_amount, ONE_GIG); 1268 if (valloc_base < segkpm_base) 1269 panic("not enough kernel VA to support memory size"); 1270 PRM_DEBUG(valloc_base); 1271 } 1272 #else /* __i386 */ 1273 valloc_base = (uintptr_t)(MISC_VA_BASE - valloc_sz); 1274 valloc_base = P2ALIGN(valloc_base, mmu.level_size[1]); 1275 PRM_DEBUG(valloc_base); 1276 #endif /* __i386 */ 1277 1278 /* 1279 * do all the initial allocations 1280 */ 1281 perform_allocations(); 1282 1283 /* 1284 * Build phys_install and phys_avail in kernel memspace. 1285 * - phys_install should be all memory in the system. 1286 * - phys_avail is phys_install minus any memory mapped before this 1287 * point above KERNEL_TEXT. 1288 */ 1289 current = phys_install = memlist; 1290 copy_memlist_filter(bootops->boot_mem->physinstalled, ¤t, NULL); 1291 if ((caddr_t)current > (caddr_t)memlist + memlist_sz) 1292 panic("physinstalled was too big!"); 1293 if (prom_debug) 1294 print_memlist("phys_install", phys_install); 1295 1296 phys_avail = current; 1297 PRM_POINT("Building phys_avail:\n"); 1298 copy_memlist_filter(bootops->boot_mem->physinstalled, ¤t, 1299 avail_filter); 1300 if ((caddr_t)current > (caddr_t)memlist + memlist_sz) 1301 panic("physavail was too big!"); 1302 if (prom_debug) 1303 print_memlist("phys_avail", phys_avail); 1304 #ifndef __xpv 1305 /* 1306 * Free unused memlist items, which may be used by memory DR driver 1307 * at runtime. 1308 */ 1309 if ((caddr_t)current < (caddr_t)memlist + memlist_sz) { 1310 memlist_free_block((caddr_t)current, 1311 (caddr_t)memlist + memlist_sz - (caddr_t)current); 1312 } 1313 #endif 1314 1315 /* 1316 * Build bios reserved memspace 1317 */ 1318 current = bios_rsvd; 1319 copy_memlist_filter(bootops->boot_mem->rsvdmem, ¤t, NULL); 1320 if ((caddr_t)current > (caddr_t)bios_rsvd + rsvdmemlist_sz) 1321 panic("bios_rsvd was too big!"); 1322 if (prom_debug) 1323 print_memlist("bios_rsvd", bios_rsvd); 1324 #ifndef __xpv 1325 /* 1326 * Free unused memlist items, which may be used by memory DR driver 1327 * at runtime. 1328 */ 1329 if ((caddr_t)current < (caddr_t)bios_rsvd + rsvdmemlist_sz) { 1330 memlist_free_block((caddr_t)current, 1331 (caddr_t)bios_rsvd + rsvdmemlist_sz - (caddr_t)current); 1332 } 1333 #endif 1334 1335 /* 1336 * setup page coloring 1337 */ 1338 page_coloring_setup(pagecolor_mem); 1339 page_lock_init(); /* currently a no-op */ 1340 1341 /* 1342 * free page list counters 1343 */ 1344 (void) page_ctrs_alloc(page_ctrs_mem); 1345 1346 /* 1347 * Size the pcf array based on the number of cpus in the box at 1348 * boot time. 1349 */ 1350 1351 pcf_init(); 1352 1353 /* 1354 * Initialize the page structures from the memory lists. 1355 */ 1356 availrmem_initial = availrmem = freemem = 0; 1357 PRM_POINT("Calling kphysm_init()..."); 1358 npages = kphysm_init(pp_base, npages); 1359 PRM_POINT("kphysm_init() done"); 1360 PRM_DEBUG(npages); 1361 1362 init_debug_info(); 1363 1364 /* 1365 * Now that page_t's have been initialized, remove all the 1366 * initial allocation pages from the kernel free page lists. 1367 */ 1368 boot_mapin((caddr_t)valloc_base, valloc_sz); 1369 boot_mapin((caddr_t)MISC_VA_BASE, MISC_VA_SIZE); 1370 PRM_POINT("startup_memlist() done"); 1371 1372 PRM_DEBUG(valloc_sz); 1373 1374 #if defined(__amd64) 1375 if ((availrmem >> (30 - MMU_PAGESHIFT)) >= 1376 textrepl_min_gb && l2cache_sz <= 2 << 20) { 1377 extern size_t textrepl_size_thresh; 1378 textrepl_size_thresh = (16 << 20) - 1; 1379 } 1380 #endif 1381 } 1382 1383 /* 1384 * Layout the kernel's part of address space and initialize kmem allocator. 1385 */ 1386 static void 1387 startup_kmem(void) 1388 { 1389 extern void page_set_colorequiv_arr(void); 1390 1391 PRM_POINT("startup_kmem() starting..."); 1392 1393 #if defined(__amd64) 1394 if (eprom_kernelbase && eprom_kernelbase != KERNELBASE) 1395 cmn_err(CE_NOTE, "!kernelbase cannot be changed on 64-bit " 1396 "systems."); 1397 kernelbase = segkpm_base - KERNEL_REDZONE_SIZE; 1398 core_base = (uintptr_t)COREHEAP_BASE; 1399 core_size = (size_t)MISC_VA_BASE - COREHEAP_BASE; 1400 #else /* __i386 */ 1401 /* 1402 * We configure kernelbase based on: 1403 * 1404 * 1. user specified kernelbase via eeprom command. Value cannot exceed 1405 * KERNELBASE_MAX. we large page align eprom_kernelbase 1406 * 1407 * 2. Default to KERNELBASE and adjust to 2X less the size for page_t. 1408 * On large memory systems we must lower kernelbase to allow 1409 * enough room for page_t's for all of memory. 1410 * 1411 * The value set here, might be changed a little later. 1412 */ 1413 if (eprom_kernelbase) { 1414 kernelbase = eprom_kernelbase & mmu.level_mask[1]; 1415 if (kernelbase > KERNELBASE_MAX) 1416 kernelbase = KERNELBASE_MAX; 1417 } else { 1418 kernelbase = (uintptr_t)KERNELBASE; 1419 kernelbase -= ROUND_UP_4MEG(2 * valloc_sz); 1420 } 1421 ASSERT((kernelbase & mmu.level_offset[1]) == 0); 1422 core_base = valloc_base; 1423 core_size = 0; 1424 #endif /* __i386 */ 1425 1426 PRM_DEBUG(core_base); 1427 PRM_DEBUG(core_size); 1428 PRM_DEBUG(kernelbase); 1429 1430 #if defined(__i386) 1431 segkp_fromheap = 1; 1432 #endif /* __i386 */ 1433 1434 ekernelheap = (char *)core_base; 1435 PRM_DEBUG(ekernelheap); 1436 1437 /* 1438 * Now that we know the real value of kernelbase, 1439 * update variables that were initialized with a value of 1440 * KERNELBASE (in common/conf/param.c). 1441 * 1442 * XXX The problem with this sort of hackery is that the 1443 * compiler just may feel like putting the const declarations 1444 * (in param.c) into the .text section. Perhaps they should 1445 * just be declared as variables there? 1446 */ 1447 1448 *(uintptr_t *)&_kernelbase = kernelbase; 1449 *(uintptr_t *)&_userlimit = kernelbase; 1450 #if defined(__amd64) 1451 *(uintptr_t *)&_userlimit -= KERNELBASE - USERLIMIT; 1452 #else 1453 *(uintptr_t *)&_userlimit32 = _userlimit; 1454 #endif 1455 PRM_DEBUG(_kernelbase); 1456 PRM_DEBUG(_userlimit); 1457 PRM_DEBUG(_userlimit32); 1458 1459 layout_kernel_va(); 1460 1461 #if defined(__i386) 1462 /* 1463 * If segmap is too large we can push the bottom of the kernel heap 1464 * higher than the base. Or worse, it could exceed the top of the 1465 * VA space entirely, causing it to wrap around. 1466 */ 1467 if (kernelheap >= ekernelheap || (uintptr_t)kernelheap < kernelbase) 1468 panic("too little address space available for kernelheap," 1469 " use eeprom for lower kernelbase or smaller segmapsize"); 1470 #endif /* __i386 */ 1471 1472 /* 1473 * Initialize the kernel heap. Note 3rd argument must be > 1st. 1474 */ 1475 kernelheap_init(kernelheap, ekernelheap, 1476 kernelheap + MMU_PAGESIZE, 1477 (void *)core_base, (void *)(core_base + core_size)); 1478 1479 #if defined(__xpv) 1480 /* 1481 * Link pending events struct into cpu struct 1482 */ 1483 CPU->cpu_m.mcpu_evt_pend = &cpu0_evt_data; 1484 #endif 1485 /* 1486 * Initialize kernel memory allocator. 1487 */ 1488 kmem_init(); 1489 1490 /* 1491 * Factor in colorequiv to check additional 'equivalent' bins 1492 */ 1493 page_set_colorequiv_arr(); 1494 1495 /* 1496 * print this out early so that we know what's going on 1497 */ 1498 print_x86_featureset(x86_featureset); 1499 1500 /* 1501 * Initialize bp_mapin(). 1502 */ 1503 bp_init(MMU_PAGESIZE, HAT_STORECACHING_OK); 1504 1505 /* 1506 * orig_npages is non-zero if physmem has been configured for less 1507 * than the available memory. 1508 */ 1509 if (orig_npages) { 1510 cmn_err(CE_WARN, "!%slimiting physmem to 0x%lx of 0x%lx pages", 1511 (npages == PHYSMEM ? "Due to virtual address space " : ""), 1512 npages, orig_npages); 1513 } 1514 #if defined(__i386) 1515 if (eprom_kernelbase && (eprom_kernelbase != kernelbase)) 1516 cmn_err(CE_WARN, "kernelbase value, User specified 0x%lx, " 1517 "System using 0x%lx", 1518 (uintptr_t)eprom_kernelbase, (uintptr_t)kernelbase); 1519 #endif 1520 1521 #ifdef KERNELBASE_ABI_MIN 1522 if (kernelbase < (uintptr_t)KERNELBASE_ABI_MIN) { 1523 cmn_err(CE_NOTE, "!kernelbase set to 0x%lx, system is not " 1524 "i386 ABI compliant.", (uintptr_t)kernelbase); 1525 } 1526 #endif 1527 1528 #ifndef __xpv 1529 if (plat_dr_support_memory()) { 1530 mem_config_init(); 1531 } 1532 #else /* __xpv */ 1533 /* 1534 * Some of the xen start information has to be relocated up 1535 * into the kernel's permanent address space. 1536 */ 1537 PRM_POINT("calling xen_relocate_start_info()"); 1538 xen_relocate_start_info(); 1539 PRM_POINT("xen_relocate_start_info() done"); 1540 1541 /* 1542 * (Update the vcpu pointer in our cpu structure to point into 1543 * the relocated shared info.) 1544 */ 1545 CPU->cpu_m.mcpu_vcpu_info = 1546 &HYPERVISOR_shared_info->vcpu_info[CPU->cpu_id]; 1547 #endif /* __xpv */ 1548 1549 PRM_POINT("startup_kmem() done"); 1550 } 1551 1552 #ifndef __xpv 1553 /* 1554 * If we have detected that we are running in an HVM environment, we need 1555 * to prepend the PV driver directory to the module search path. 1556 */ 1557 #define HVM_MOD_DIR "/platform/i86hvm/kernel" 1558 static void 1559 update_default_path() 1560 { 1561 char *current, *newpath; 1562 int newlen; 1563 1564 /* 1565 * We are about to resync with krtld. krtld will reset its 1566 * internal module search path iff Solaris has set default_path. 1567 * We want to be sure we're prepending this new directory to the 1568 * right search path. 1569 */ 1570 current = (default_path == NULL) ? kobj_module_path : default_path; 1571 1572 newlen = strlen(HVM_MOD_DIR) + strlen(current) + 2; 1573 newpath = kmem_alloc(newlen, KM_SLEEP); 1574 (void) strcpy(newpath, HVM_MOD_DIR); 1575 (void) strcat(newpath, " "); 1576 (void) strcat(newpath, current); 1577 1578 default_path = newpath; 1579 } 1580 #endif 1581 1582 static void 1583 startup_modules(void) 1584 { 1585 int cnt; 1586 extern void prom_setup(void); 1587 int32_t v, h; 1588 char d[11]; 1589 char *cp; 1590 cmi_hdl_t hdl; 1591 1592 PRM_POINT("startup_modules() starting..."); 1593 1594 #ifndef __xpv 1595 /* 1596 * Initialize ten-micro second timer so that drivers will 1597 * not get short changed in their init phase. This was 1598 * not getting called until clkinit which, on fast cpu's 1599 * caused the drv_usecwait to be way too short. 1600 */ 1601 microfind(); 1602 1603 if ((get_hwenv() & HW_XEN_HVM) != 0) 1604 update_default_path(); 1605 #endif 1606 1607 /* 1608 * Read the GMT lag from /etc/rtc_config. 1609 */ 1610 sgmtl(process_rtc_config_file()); 1611 1612 /* 1613 * Calculate default settings of system parameters based upon 1614 * maxusers, yet allow to be overridden via the /etc/system file. 1615 */ 1616 param_calc(0); 1617 1618 mod_setup(); 1619 1620 /* 1621 * Initialize system parameters. 1622 */ 1623 param_init(); 1624 1625 /* 1626 * Initialize the default brands 1627 */ 1628 brand_init(); 1629 1630 /* 1631 * maxmem is the amount of physical memory we're playing with. 1632 */ 1633 maxmem = physmem; 1634 1635 /* 1636 * Initialize segment management stuff. 1637 */ 1638 seg_init(); 1639 1640 if (modload("fs", "specfs") == -1) 1641 halt("Can't load specfs"); 1642 1643 if (modload("fs", "devfs") == -1) 1644 halt("Can't load devfs"); 1645 1646 if (modload("fs", "dev") == -1) 1647 halt("Can't load dev"); 1648 1649 if (modload("fs", "procfs") == -1) 1650 halt("Can't load procfs"); 1651 1652 (void) modloadonly("sys", "lbl_edition"); 1653 1654 dispinit(); 1655 1656 /* Read cluster configuration data. */ 1657 clconf_init(); 1658 1659 #if defined(__xpv) 1660 (void) ec_init(); 1661 gnttab_init(); 1662 (void) xs_early_init(); 1663 #endif /* __xpv */ 1664 1665 /* 1666 * Create a kernel device tree. First, create rootnex and 1667 * then invoke bus specific code to probe devices. 1668 */ 1669 setup_ddi(); 1670 1671 #ifdef __xpv 1672 if (DOMAIN_IS_INITDOMAIN(xen_info)) 1673 #endif 1674 { 1675 id_t smid; 1676 smbios_system_t smsys; 1677 smbios_info_t sminfo; 1678 char *mfg; 1679 /* 1680 * Load the System Management BIOS into the global ksmbios 1681 * handle, if an SMBIOS is present on this system. 1682 * Also set "si-hw-provider" property, if not already set. 1683 */ 1684 ksmbios = smbios_open(NULL, SMB_VERSION, ksmbios_flags, NULL); 1685 if (ksmbios != NULL && 1686 ((smid = smbios_info_system(ksmbios, &smsys)) != SMB_ERR) && 1687 (smbios_info_common(ksmbios, smid, &sminfo)) != SMB_ERR) { 1688 mfg = (char *)sminfo.smbi_manufacturer; 1689 if (BOP_GETPROPLEN(bootops, "si-hw-provider") < 0) { 1690 extern char hw_provider[]; 1691 int i; 1692 for (i = 0; i < SYS_NMLN; i++) { 1693 if (isprint(mfg[i])) 1694 hw_provider[i] = mfg[i]; 1695 else { 1696 hw_provider[i] = '\0'; 1697 break; 1698 } 1699 } 1700 hw_provider[SYS_NMLN - 1] = '\0'; 1701 } 1702 } 1703 } 1704 1705 1706 /* 1707 * Originally clconf_init() apparently needed the hostid. But 1708 * this no longer appears to be true - it uses its own nodeid. 1709 * By placing the hostid logic here, we are able to make use of 1710 * the SMBIOS UUID. 1711 */ 1712 if ((h = set_soft_hostid()) == HW_INVALID_HOSTID) { 1713 cmn_err(CE_WARN, "Unable to set hostid"); 1714 } else { 1715 for (v = h, cnt = 0; cnt < 10; cnt++) { 1716 d[cnt] = (char)(v % 10); 1717 v /= 10; 1718 if (v == 0) 1719 break; 1720 } 1721 for (cp = hw_serial; cnt >= 0; cnt--) 1722 *cp++ = d[cnt] + '0'; 1723 *cp = 0; 1724 } 1725 1726 /* 1727 * Set up the CPU module subsystem for the boot cpu in the native 1728 * case, and all physical cpu resource in the xpv dom0 case. 1729 * Modifies the device tree, so this must be done after 1730 * setup_ddi(). 1731 */ 1732 #ifdef __xpv 1733 /* 1734 * If paravirtualized and on dom0 then we initialize all physical 1735 * cpu handles now; if paravirtualized on a domU then do not 1736 * initialize. 1737 */ 1738 if (DOMAIN_IS_INITDOMAIN(xen_info)) { 1739 xen_mc_lcpu_cookie_t cpi; 1740 1741 for (cpi = xen_physcpu_next(NULL); cpi != NULL; 1742 cpi = xen_physcpu_next(cpi)) { 1743 if ((hdl = cmi_init(CMI_HDL_SOLARIS_xVM_MCA, 1744 xen_physcpu_chipid(cpi), xen_physcpu_coreid(cpi), 1745 xen_physcpu_strandid(cpi))) != NULL && 1746 is_x86_feature(x86_featureset, X86FSET_MCA)) 1747 cmi_mca_init(hdl); 1748 } 1749 } 1750 #else 1751 /* 1752 * Initialize a handle for the boot cpu - others will initialize 1753 * as they startup. Do not do this if we know we are in an HVM domU. 1754 */ 1755 if ((get_hwenv() & HW_XEN_HVM) == 0 && 1756 (hdl = cmi_init(CMI_HDL_NATIVE, cmi_ntv_hwchipid(CPU), 1757 cmi_ntv_hwcoreid(CPU), cmi_ntv_hwstrandid(CPU))) != NULL && 1758 is_x86_feature(x86_featureset, X86FSET_MCA)) { 1759 cmi_mca_init(hdl); 1760 CPU->cpu_m.mcpu_cmi_hdl = hdl; 1761 } 1762 #endif /* __xpv */ 1763 1764 /* 1765 * Fake a prom tree such that /dev/openprom continues to work 1766 */ 1767 PRM_POINT("startup_modules: calling prom_setup..."); 1768 prom_setup(); 1769 PRM_POINT("startup_modules: done"); 1770 1771 /* 1772 * Load all platform specific modules 1773 */ 1774 PRM_POINT("startup_modules: calling psm_modload..."); 1775 psm_modload(); 1776 1777 PRM_POINT("startup_modules() done"); 1778 } 1779 1780 /* 1781 * claim a "setaside" boot page for use in the kernel 1782 */ 1783 page_t * 1784 boot_claim_page(pfn_t pfn) 1785 { 1786 page_t *pp; 1787 1788 pp = page_numtopp_nolock(pfn); 1789 ASSERT(pp != NULL); 1790 1791 if (PP_ISBOOTPAGES(pp)) { 1792 if (pp->p_next != NULL) 1793 pp->p_next->p_prev = pp->p_prev; 1794 if (pp->p_prev == NULL) 1795 bootpages = pp->p_next; 1796 else 1797 pp->p_prev->p_next = pp->p_next; 1798 } else { 1799 /* 1800 * htable_attach() expects a base pagesize page 1801 */ 1802 if (pp->p_szc != 0) 1803 page_boot_demote(pp); 1804 pp = page_numtopp(pfn, SE_EXCL); 1805 } 1806 return (pp); 1807 } 1808 1809 /* 1810 * Walk through the pagetables looking for pages mapped in by boot. If the 1811 * setaside flag is set the pages are expected to be returned to the 1812 * kernel later in boot, so we add them to the bootpages list. 1813 */ 1814 static void 1815 protect_boot_range(uintptr_t low, uintptr_t high, int setaside) 1816 { 1817 uintptr_t va = low; 1818 size_t len; 1819 uint_t prot; 1820 pfn_t pfn; 1821 page_t *pp; 1822 pgcnt_t boot_protect_cnt = 0; 1823 1824 while (kbm_probe(&va, &len, &pfn, &prot) != 0 && va < high) { 1825 if (va + len >= high) 1826 panic("0x%lx byte mapping at 0x%p exceeds boot's " 1827 "legal range.", len, (void *)va); 1828 1829 while (len > 0) { 1830 pp = page_numtopp_alloc(pfn); 1831 if (pp != NULL) { 1832 if (setaside == 0) 1833 panic("Unexpected mapping by boot. " 1834 "addr=%p pfn=%lx\n", 1835 (void *)va, pfn); 1836 1837 pp->p_next = bootpages; 1838 pp->p_prev = NULL; 1839 PP_SETBOOTPAGES(pp); 1840 if (bootpages != NULL) { 1841 bootpages->p_prev = pp; 1842 } 1843 bootpages = pp; 1844 ++boot_protect_cnt; 1845 } 1846 1847 ++pfn; 1848 len -= MMU_PAGESIZE; 1849 va += MMU_PAGESIZE; 1850 } 1851 } 1852 PRM_DEBUG(boot_protect_cnt); 1853 } 1854 1855 /* 1856 * 1857 */ 1858 static void 1859 layout_kernel_va(void) 1860 { 1861 PRM_POINT("layout_kernel_va() starting..."); 1862 /* 1863 * Establish the final size of the kernel's heap, size of segmap, 1864 * segkp, etc. 1865 */ 1866 1867 #if defined(__amd64) 1868 1869 kpm_vbase = (caddr_t)segkpm_base; 1870 if (physmax + 1 < plat_dr_physmax) { 1871 kpm_size = ROUND_UP_LPAGE(mmu_ptob(plat_dr_physmax)); 1872 } else { 1873 kpm_size = ROUND_UP_LPAGE(mmu_ptob(physmax + 1)); 1874 } 1875 if ((uintptr_t)kpm_vbase + kpm_size > (uintptr_t)valloc_base) 1876 panic("not enough room for kpm!"); 1877 PRM_DEBUG(kpm_size); 1878 PRM_DEBUG(kpm_vbase); 1879 1880 /* 1881 * By default we create a seg_kp in 64 bit kernels, it's a little 1882 * faster to access than embedding it in the heap. 1883 */ 1884 segkp_base = (caddr_t)valloc_base + valloc_sz; 1885 if (!segkp_fromheap) { 1886 size_t sz = mmu_ptob(segkpsize); 1887 1888 /* 1889 * determine size of segkp 1890 */ 1891 if (sz < SEGKPMINSIZE || sz > SEGKPMAXSIZE) { 1892 sz = SEGKPDEFSIZE; 1893 cmn_err(CE_WARN, "!Illegal value for segkpsize. " 1894 "segkpsize has been reset to %ld pages", 1895 mmu_btop(sz)); 1896 } 1897 sz = MIN(sz, MAX(SEGKPMINSIZE, mmu_ptob(physmem))); 1898 1899 segkpsize = mmu_btop(ROUND_UP_LPAGE(sz)); 1900 } 1901 PRM_DEBUG(segkp_base); 1902 PRM_DEBUG(segkpsize); 1903 1904 /* 1905 * segzio is used for ZFS cached data. It uses a distinct VA 1906 * segment (from kernel heap) so that we can easily tell not to 1907 * include it in kernel crash dumps on 64 bit kernels. The trick is 1908 * to give it lots of VA, but not constrain the kernel heap. 1909 * We scale the size of segzio linearly with physmem up to 1910 * SEGZIOMAXSIZE. Above that amount it scales at 50% of physmem. 1911 */ 1912 segzio_base = segkp_base + mmu_ptob(segkpsize); 1913 if (segzio_fromheap) { 1914 segziosize = 0; 1915 } else { 1916 size_t physmem_size = mmu_ptob(physmem); 1917 size_t size = (segziosize == 0) ? 1918 physmem_size : mmu_ptob(segziosize); 1919 1920 if (size < SEGZIOMINSIZE) 1921 size = SEGZIOMINSIZE; 1922 if (size > SEGZIOMAXSIZE) { 1923 size = SEGZIOMAXSIZE; 1924 if (physmem_size > size) 1925 size += (physmem_size - size) / 2; 1926 } 1927 segziosize = mmu_btop(ROUND_UP_LPAGE(size)); 1928 } 1929 PRM_DEBUG(segziosize); 1930 PRM_DEBUG(segzio_base); 1931 1932 /* 1933 * Put the range of VA for device mappings next, kmdb knows to not 1934 * grep in this range of addresses. 1935 */ 1936 toxic_addr = 1937 ROUND_UP_LPAGE((uintptr_t)segzio_base + mmu_ptob(segziosize)); 1938 PRM_DEBUG(toxic_addr); 1939 segmap_start = ROUND_UP_LPAGE(toxic_addr + toxic_size); 1940 #else /* __i386 */ 1941 segmap_start = ROUND_UP_LPAGE(kernelbase); 1942 #endif /* __i386 */ 1943 PRM_DEBUG(segmap_start); 1944 1945 /* 1946 * Users can change segmapsize through eeprom. If the variable 1947 * is tuned through eeprom, there is no upper bound on the 1948 * size of segmap. 1949 */ 1950 segmapsize = MAX(ROUND_UP_LPAGE(segmapsize), SEGMAPDEFAULT); 1951 1952 #if defined(__i386) 1953 /* 1954 * 32-bit systems don't have segkpm or segkp, so segmap appears at 1955 * the bottom of the kernel's address range. Set aside space for a 1956 * small red zone just below the start of segmap. 1957 */ 1958 segmap_start += KERNEL_REDZONE_SIZE; 1959 segmapsize -= KERNEL_REDZONE_SIZE; 1960 #endif 1961 1962 PRM_DEBUG(segmap_start); 1963 PRM_DEBUG(segmapsize); 1964 kernelheap = (caddr_t)ROUND_UP_LPAGE(segmap_start + segmapsize); 1965 PRM_DEBUG(kernelheap); 1966 PRM_POINT("layout_kernel_va() done..."); 1967 } 1968 1969 /* 1970 * Finish initializing the VM system, now that we are no longer 1971 * relying on the boot time memory allocators. 1972 */ 1973 static void 1974 startup_vm(void) 1975 { 1976 struct segmap_crargs a; 1977 1978 extern int use_brk_lpg, use_stk_lpg; 1979 1980 PRM_POINT("startup_vm() starting..."); 1981 1982 /* 1983 * Initialize the hat layer. 1984 */ 1985 hat_init(); 1986 1987 /* 1988 * Do final allocations of HAT data structures that need to 1989 * be allocated before quiescing the boot loader. 1990 */ 1991 PRM_POINT("Calling hat_kern_alloc()..."); 1992 hat_kern_alloc((caddr_t)segmap_start, segmapsize, ekernelheap); 1993 PRM_POINT("hat_kern_alloc() done"); 1994 1995 #ifndef __xpv 1996 /* 1997 * Setup Page Attribute Table 1998 */ 1999 pat_sync(); 2000 #endif 2001 2002 /* 2003 * The next two loops are done in distinct steps in order 2004 * to be sure that any page that is doubly mapped (both above 2005 * KERNEL_TEXT and below kernelbase) is dealt with correctly. 2006 * Note this may never happen, but it might someday. 2007 */ 2008 bootpages = NULL; 2009 PRM_POINT("Protecting boot pages"); 2010 2011 /* 2012 * Protect any pages mapped above KERNEL_TEXT that somehow have 2013 * page_t's. This can only happen if something weird allocated 2014 * in this range (like kadb/kmdb). 2015 */ 2016 protect_boot_range(KERNEL_TEXT, (uintptr_t)-1, 0); 2017 2018 /* 2019 * Before we can take over memory allocation/mapping from the boot 2020 * loader we must remove from our free page lists any boot allocated 2021 * pages that stay mapped until release_bootstrap(). 2022 */ 2023 protect_boot_range(0, kernelbase, 1); 2024 2025 2026 /* 2027 * Switch to running on regular HAT (not boot_mmu) 2028 */ 2029 PRM_POINT("Calling hat_kern_setup()..."); 2030 hat_kern_setup(); 2031 2032 /* 2033 * It is no longer safe to call BOP_ALLOC(), so make sure we don't. 2034 */ 2035 bop_no_more_mem(); 2036 2037 PRM_POINT("hat_kern_setup() done"); 2038 2039 hat_cpu_online(CPU); 2040 2041 /* 2042 * Initialize VM system 2043 */ 2044 PRM_POINT("Calling kvm_init()..."); 2045 kvm_init(); 2046 PRM_POINT("kvm_init() done"); 2047 2048 /* 2049 * Tell kmdb that the VM system is now working 2050 */ 2051 if (boothowto & RB_DEBUG) 2052 kdi_dvec_vmready(); 2053 2054 #if defined(__xpv) 2055 /* 2056 * Populate the I/O pool on domain 0 2057 */ 2058 if (DOMAIN_IS_INITDOMAIN(xen_info)) { 2059 extern long populate_io_pool(void); 2060 long init_io_pool_cnt; 2061 2062 PRM_POINT("Populating reserve I/O page pool"); 2063 init_io_pool_cnt = populate_io_pool(); 2064 PRM_DEBUG(init_io_pool_cnt); 2065 } 2066 #endif 2067 /* 2068 * Mangle the brand string etc. 2069 */ 2070 cpuid_pass3(CPU); 2071 2072 #if defined(__amd64) 2073 2074 /* 2075 * Create the device arena for toxic (to dtrace/kmdb) mappings. 2076 */ 2077 device_arena = vmem_create("device", (void *)toxic_addr, 2078 toxic_size, MMU_PAGESIZE, NULL, NULL, NULL, 0, VM_SLEEP); 2079 2080 #else /* __i386 */ 2081 2082 /* 2083 * allocate the bit map that tracks toxic pages 2084 */ 2085 toxic_bit_map_len = btop((ulong_t)(valloc_base - kernelbase)); 2086 PRM_DEBUG(toxic_bit_map_len); 2087 toxic_bit_map = 2088 kmem_zalloc(BT_SIZEOFMAP(toxic_bit_map_len), KM_NOSLEEP); 2089 ASSERT(toxic_bit_map != NULL); 2090 PRM_DEBUG(toxic_bit_map); 2091 2092 #endif /* __i386 */ 2093 2094 2095 /* 2096 * Now that we've got more VA, as well as the ability to allocate from 2097 * it, tell the debugger. 2098 */ 2099 if (boothowto & RB_DEBUG) 2100 kdi_dvec_memavail(); 2101 2102 /* 2103 * The following code installs a special page fault handler (#pf) 2104 * to work around a pentium bug. 2105 */ 2106 #if !defined(__amd64) && !defined(__xpv) 2107 if (x86_type == X86_TYPE_P5) { 2108 desctbr_t idtr; 2109 gate_desc_t *newidt; 2110 2111 if ((newidt = kmem_zalloc(MMU_PAGESIZE, KM_NOSLEEP)) == NULL) 2112 panic("failed to install pentium_pftrap"); 2113 2114 bcopy(idt0, newidt, NIDT * sizeof (*idt0)); 2115 set_gatesegd(&newidt[T_PGFLT], &pentium_pftrap, 2116 KCS_SEL, SDT_SYSIGT, TRP_KPL, 0); 2117 2118 (void) as_setprot(&kas, (caddr_t)newidt, MMU_PAGESIZE, 2119 PROT_READ | PROT_EXEC); 2120 2121 CPU->cpu_idt = newidt; 2122 idtr.dtr_base = (uintptr_t)CPU->cpu_idt; 2123 idtr.dtr_limit = (NIDT * sizeof (*idt0)) - 1; 2124 wr_idtr(&idtr); 2125 } 2126 #endif /* !__amd64 */ 2127 2128 #if !defined(__xpv) 2129 /* 2130 * Map page pfn=0 for drivers, such as kd, that need to pick up 2131 * parameters left there by controllers/BIOS. 2132 */ 2133 PRM_POINT("setup up p0_va"); 2134 p0_va = i86devmap(0, 1, PROT_READ); 2135 PRM_DEBUG(p0_va); 2136 #endif 2137 2138 cmn_err(CE_CONT, "?mem = %luK (0x%lx)\n", 2139 physinstalled << (MMU_PAGESHIFT - 10), ptob(physinstalled)); 2140 2141 /* 2142 * disable automatic large pages for small memory systems or 2143 * when the disable flag is set. 2144 * 2145 * Do not yet consider page sizes larger than 2m/4m. 2146 */ 2147 if (!auto_lpg_disable && mmu.max_page_level > 0) { 2148 max_uheap_lpsize = LEVEL_SIZE(1); 2149 max_ustack_lpsize = LEVEL_SIZE(1); 2150 max_privmap_lpsize = LEVEL_SIZE(1); 2151 max_uidata_lpsize = LEVEL_SIZE(1); 2152 max_utext_lpsize = LEVEL_SIZE(1); 2153 max_shm_lpsize = LEVEL_SIZE(1); 2154 } 2155 if (physmem < privm_lpg_min_physmem || mmu.max_page_level == 0 || 2156 auto_lpg_disable) { 2157 use_brk_lpg = 0; 2158 use_stk_lpg = 0; 2159 } 2160 mcntl0_lpsize = LEVEL_SIZE(mmu.umax_page_level); 2161 2162 PRM_POINT("Calling hat_init_finish()..."); 2163 hat_init_finish(); 2164 PRM_POINT("hat_init_finish() done"); 2165 2166 /* 2167 * Initialize the segkp segment type. 2168 */ 2169 rw_enter(&kas.a_lock, RW_WRITER); 2170 PRM_POINT("Attaching segkp"); 2171 if (segkp_fromheap) { 2172 segkp->s_as = &kas; 2173 } else if (seg_attach(&kas, (caddr_t)segkp_base, mmu_ptob(segkpsize), 2174 segkp) < 0) { 2175 panic("startup: cannot attach segkp"); 2176 /*NOTREACHED*/ 2177 } 2178 PRM_POINT("Doing segkp_create()"); 2179 if (segkp_create(segkp) != 0) { 2180 panic("startup: segkp_create failed"); 2181 /*NOTREACHED*/ 2182 } 2183 PRM_DEBUG(segkp); 2184 rw_exit(&kas.a_lock); 2185 2186 /* 2187 * kpm segment 2188 */ 2189 segmap_kpm = 0; 2190 if (kpm_desired) { 2191 kpm_init(); 2192 kpm_enable = 1; 2193 } 2194 2195 /* 2196 * Now create segmap segment. 2197 */ 2198 rw_enter(&kas.a_lock, RW_WRITER); 2199 if (seg_attach(&kas, (caddr_t)segmap_start, segmapsize, segmap) < 0) { 2200 panic("cannot attach segmap"); 2201 /*NOTREACHED*/ 2202 } 2203 PRM_DEBUG(segmap); 2204 2205 a.prot = PROT_READ | PROT_WRITE; 2206 a.shmsize = 0; 2207 a.nfreelist = segmapfreelists; 2208 2209 if (segmap_create(segmap, (caddr_t)&a) != 0) 2210 panic("segmap_create segmap"); 2211 rw_exit(&kas.a_lock); 2212 2213 setup_vaddr_for_ppcopy(CPU); 2214 2215 segdev_init(); 2216 #if defined(__xpv) 2217 if (DOMAIN_IS_INITDOMAIN(xen_info)) 2218 #endif 2219 pmem_init(); 2220 2221 PRM_POINT("startup_vm() done"); 2222 } 2223 2224 /* 2225 * Load a tod module for the non-standard tod part found on this system. 2226 */ 2227 static void 2228 load_tod_module(char *todmod) 2229 { 2230 if (modload("tod", todmod) == -1) 2231 halt("Can't load TOD module"); 2232 } 2233 2234 static void 2235 startup_end(void) 2236 { 2237 int i; 2238 extern void setx86isalist(void); 2239 extern void cpu_event_init(void); 2240 2241 PRM_POINT("startup_end() starting..."); 2242 2243 /* 2244 * Perform tasks that get done after most of the VM 2245 * initialization has been done but before the clock 2246 * and other devices get started. 2247 */ 2248 kern_setup1(); 2249 2250 /* 2251 * Perform CPC initialization for this CPU. 2252 */ 2253 kcpc_hw_init(CPU); 2254 2255 /* 2256 * Initialize cpu event framework. 2257 */ 2258 cpu_event_init(); 2259 2260 #if defined(OPTERON_WORKAROUND_6323525) 2261 if (opteron_workaround_6323525) 2262 patch_workaround_6323525(); 2263 #endif 2264 /* 2265 * If needed, load TOD module now so that ddi_get_time(9F) etc. work 2266 * (For now, "needed" is defined as set tod_module_name in /etc/system) 2267 */ 2268 if (tod_module_name != NULL) { 2269 PRM_POINT("load_tod_module()"); 2270 load_tod_module(tod_module_name); 2271 } 2272 2273 #if defined(__xpv) 2274 /* 2275 * Forceload interposing TOD module for the hypervisor. 2276 */ 2277 PRM_POINT("load_tod_module()"); 2278 load_tod_module("xpvtod"); 2279 #endif 2280 2281 /* 2282 * Configure the system. 2283 */ 2284 PRM_POINT("Calling configure()..."); 2285 configure(); /* set up devices */ 2286 PRM_POINT("configure() done"); 2287 2288 /* 2289 * We can now setup for XSAVE because fpu_probe is done in configure(). 2290 */ 2291 if (fp_save_mech == FP_XSAVE) { 2292 xsave_setup_msr(CPU); 2293 } 2294 2295 /* 2296 * Set the isa_list string to the defined instruction sets we 2297 * support. 2298 */ 2299 setx86isalist(); 2300 cpu_intr_alloc(CPU, NINTR_THREADS); 2301 psm_install(); 2302 2303 /* 2304 * We're done with bootops. We don't unmap the bootstrap yet because 2305 * we're still using bootsvcs. 2306 */ 2307 PRM_POINT("NULLing out bootops"); 2308 *bootopsp = (struct bootops *)NULL; 2309 bootops = (struct bootops *)NULL; 2310 2311 #if defined(__xpv) 2312 ec_init_debug_irq(); 2313 xs_domu_init(); 2314 #endif 2315 2316 #if defined(__amd64) && !defined(__xpv) 2317 /* 2318 * Intel IOMMU has been setup/initialized in ddi_impl.c 2319 * Start it up now. 2320 */ 2321 immu_startup(); 2322 #endif 2323 2324 PRM_POINT("Enabling interrupts"); 2325 (*picinitf)(); 2326 sti(); 2327 #if defined(__xpv) 2328 ASSERT(CPU->cpu_m.mcpu_vcpu_info->evtchn_upcall_mask == 0); 2329 xen_late_startup(); 2330 #endif 2331 2332 (void) add_avsoftintr((void *)&softlevel1_hdl, 1, softlevel1, 2333 "softlevel1", NULL, NULL); /* XXX to be moved later */ 2334 2335 /* 2336 * Register software interrupt handlers for ddi_periodic_add(9F). 2337 * Software interrupts up to the level 10 are supported. 2338 */ 2339 for (i = DDI_IPL_1; i <= DDI_IPL_10; i++) { 2340 (void) add_avsoftintr((void *)&softlevel_hdl[i-1], i, 2341 (avfunc)ddi_periodic_softintr, "ddi_periodic", 2342 (caddr_t)(uintptr_t)i, NULL); 2343 } 2344 2345 #if !defined(__xpv) 2346 if (modload("drv", "amd_iommu") < 0) { 2347 PRM_POINT("No AMD IOMMU present\n"); 2348 } else if (ddi_hold_installed_driver(ddi_name_to_major( 2349 "amd_iommu")) == NULL) { 2350 prom_printf("ERROR: failed to attach AMD IOMMU\n"); 2351 } 2352 #endif 2353 post_startup_cpu_fixups(); 2354 2355 PRM_POINT("startup_end() done"); 2356 } 2357 2358 /* 2359 * Don't remove the following 2 variables. They are necessary 2360 * for reading the hostid from the legacy file (/kernel/misc/sysinit). 2361 */ 2362 char *_hs1107 = hw_serial; 2363 ulong_t _bdhs34; 2364 2365 void 2366 post_startup(void) 2367 { 2368 extern void cpupm_init(cpu_t *); 2369 extern void cpu_event_init_cpu(cpu_t *); 2370 2371 /* 2372 * Set the system wide, processor-specific flags to be passed 2373 * to userland via the aux vector for performance hints and 2374 * instruction set extensions. 2375 */ 2376 bind_hwcap(); 2377 2378 #ifdef __xpv 2379 if (DOMAIN_IS_INITDOMAIN(xen_info)) 2380 #endif 2381 { 2382 #if defined(__xpv) 2383 xpv_panic_init(); 2384 #else 2385 /* 2386 * Startup the memory scrubber. 2387 * XXPV This should be running somewhere .. 2388 */ 2389 if ((get_hwenv() & HW_VIRTUAL) == 0) 2390 memscrub_init(); 2391 #endif 2392 } 2393 2394 /* 2395 * Complete CPU module initialization 2396 */ 2397 cmi_post_startup(); 2398 2399 /* 2400 * Perform forceloading tasks for /etc/system. 2401 */ 2402 (void) mod_sysctl(SYS_FORCELOAD, NULL); 2403 2404 /* 2405 * ON4.0: Force /proc module in until clock interrupt handle fixed 2406 * ON4.0: This must be fixed or restated in /etc/systems. 2407 */ 2408 (void) modload("fs", "procfs"); 2409 2410 (void) i_ddi_attach_hw_nodes("pit_beep"); 2411 2412 #if defined(__i386) 2413 /* 2414 * Check for required functional Floating Point hardware, 2415 * unless FP hardware explicitly disabled. 2416 */ 2417 if (fpu_exists && (fpu_pentium_fdivbug || fp_kind == FP_NO)) 2418 halt("No working FP hardware found"); 2419 #endif 2420 2421 maxmem = freemem; 2422 2423 cpu_event_init_cpu(CPU); 2424 cpupm_init(CPU); 2425 (void) mach_cpu_create_device_node(CPU, NULL); 2426 2427 pg_init(); 2428 } 2429 2430 static int 2431 pp_in_range(page_t *pp, uint64_t low_addr, uint64_t high_addr) 2432 { 2433 return ((pp->p_pagenum >= btop(low_addr)) && 2434 (pp->p_pagenum < btopr(high_addr))); 2435 } 2436 2437 static int 2438 pp_in_module(page_t *pp, const rd_existing_t *modranges) 2439 { 2440 uint_t i; 2441 2442 for (i = 0; modranges[i].phys != 0; i++) { 2443 if (pp_in_range(pp, modranges[i].phys, 2444 modranges[i].phys + modranges[i].size)) 2445 return (1); 2446 } 2447 2448 return (0); 2449 } 2450 2451 void 2452 release_bootstrap(void) 2453 { 2454 int root_is_ramdisk; 2455 page_t *pp; 2456 extern void kobj_boot_unmountroot(void); 2457 extern dev_t rootdev; 2458 uint_t i; 2459 char propname[32]; 2460 rd_existing_t *modranges; 2461 #if !defined(__xpv) 2462 pfn_t pfn; 2463 #endif 2464 2465 /* 2466 * Save the bootfs module ranges so that we can reserve them below 2467 * for the real bootfs. 2468 */ 2469 modranges = kmem_alloc(sizeof (rd_existing_t) * MAX_BOOT_MODULES, 2470 KM_SLEEP); 2471 for (i = 0; ; i++) { 2472 uint64_t start, size; 2473 2474 modranges[i].phys = 0; 2475 2476 (void) snprintf(propname, sizeof (propname), 2477 "module-addr-%u", i); 2478 if (do_bsys_getproplen(NULL, propname) <= 0) 2479 break; 2480 (void) do_bsys_getprop(NULL, propname, &start); 2481 2482 (void) snprintf(propname, sizeof (propname), 2483 "module-size-%u", i); 2484 if (do_bsys_getproplen(NULL, propname) <= 0) 2485 break; 2486 (void) do_bsys_getprop(NULL, propname, &size); 2487 2488 modranges[i].phys = start; 2489 modranges[i].size = size; 2490 } 2491 2492 /* unmount boot ramdisk and release kmem usage */ 2493 kobj_boot_unmountroot(); 2494 2495 /* 2496 * We're finished using the boot loader so free its pages. 2497 */ 2498 PRM_POINT("Unmapping lower boot pages"); 2499 2500 clear_boot_mappings(0, _userlimit); 2501 2502 postbootkernelbase = kernelbase; 2503 2504 /* 2505 * If root isn't on ramdisk, destroy the hardcoded 2506 * ramdisk node now and release the memory. Else, 2507 * ramdisk memory is kept in rd_pages. 2508 */ 2509 root_is_ramdisk = (getmajor(rootdev) == ddi_name_to_major("ramdisk")); 2510 if (!root_is_ramdisk) { 2511 dev_info_t *dip = ddi_find_devinfo("ramdisk", -1, 0); 2512 ASSERT(dip && ddi_get_parent(dip) == ddi_root_node()); 2513 ndi_rele_devi(dip); /* held from ddi_find_devinfo */ 2514 (void) ddi_remove_child(dip, 0); 2515 } 2516 2517 PRM_POINT("Releasing boot pages"); 2518 while (bootpages) { 2519 extern uint64_t ramdisk_start, ramdisk_end; 2520 pp = bootpages; 2521 bootpages = pp->p_next; 2522 2523 2524 /* Keep pages for the lower 64K */ 2525 if (pp_in_range(pp, 0, 0x40000)) { 2526 pp->p_next = lower_pages; 2527 lower_pages = pp; 2528 lower_pages_count++; 2529 continue; 2530 } 2531 2532 if (root_is_ramdisk && pp_in_range(pp, ramdisk_start, 2533 ramdisk_end) || pp_in_module(pp, modranges)) { 2534 pp->p_next = rd_pages; 2535 rd_pages = pp; 2536 continue; 2537 } 2538 pp->p_next = (struct page *)0; 2539 pp->p_prev = (struct page *)0; 2540 PP_CLRBOOTPAGES(pp); 2541 page_free(pp, 1); 2542 } 2543 PRM_POINT("Boot pages released"); 2544 2545 kmem_free(modranges, sizeof (rd_existing_t) * 99); 2546 2547 #if !defined(__xpv) 2548 /* XXPV -- note this following bunch of code needs to be revisited in Xen 3.0 */ 2549 /* 2550 * Find 1 page below 1 MB so that other processors can boot up or 2551 * so that any processor can resume. 2552 * Make sure it has a kernel VA as well as a 1:1 mapping. 2553 * We should have just free'd one up. 2554 */ 2555 2556 /* 2557 * 0x10 pages is 64K. Leave the bottom 64K alone 2558 * for BIOS. 2559 */ 2560 for (pfn = 0x10; pfn < btop(1*1024*1024); pfn++) { 2561 if (page_numtopp_alloc(pfn) == NULL) 2562 continue; 2563 rm_platter_va = i86devmap(pfn, 1, 2564 PROT_READ | PROT_WRITE | PROT_EXEC); 2565 rm_platter_pa = ptob(pfn); 2566 break; 2567 } 2568 if (pfn == btop(1*1024*1024) && use_mp) 2569 panic("No page below 1M available for starting " 2570 "other processors or for resuming from system-suspend"); 2571 #endif /* !__xpv */ 2572 } 2573 2574 /* 2575 * Initialize the platform-specific parts of a page_t. 2576 */ 2577 void 2578 add_physmem_cb(page_t *pp, pfn_t pnum) 2579 { 2580 pp->p_pagenum = pnum; 2581 pp->p_mapping = NULL; 2582 pp->p_embed = 0; 2583 pp->p_share = 0; 2584 pp->p_mlentry = 0; 2585 } 2586 2587 /* 2588 * kphysm_init() initializes physical memory. 2589 */ 2590 static pgcnt_t 2591 kphysm_init( 2592 page_t *pp, 2593 pgcnt_t npages) 2594 { 2595 struct memlist *pmem; 2596 struct memseg *cur_memseg; 2597 pfn_t base_pfn; 2598 pfn_t end_pfn; 2599 pgcnt_t num; 2600 pgcnt_t pages_done = 0; 2601 uint64_t addr; 2602 uint64_t size; 2603 extern pfn_t ddiphysmin; 2604 extern int mnode_xwa; 2605 int ms = 0, me = 0; 2606 2607 ASSERT(page_hash != NULL && page_hashsz != 0); 2608 2609 cur_memseg = memseg_base; 2610 for (pmem = phys_avail; pmem && npages; pmem = pmem->ml_next) { 2611 /* 2612 * In a 32 bit kernel can't use higher memory if we're 2613 * not booting in PAE mode. This check takes care of that. 2614 */ 2615 addr = pmem->ml_address; 2616 size = pmem->ml_size; 2617 if (btop(addr) > physmax) 2618 continue; 2619 2620 /* 2621 * align addr and size - they may not be at page boundaries 2622 */ 2623 if ((addr & MMU_PAGEOFFSET) != 0) { 2624 addr += MMU_PAGEOFFSET; 2625 addr &= ~(uint64_t)MMU_PAGEOFFSET; 2626 size -= addr - pmem->ml_address; 2627 } 2628 2629 /* only process pages below or equal to physmax */ 2630 if ((btop(addr + size) - 1) > physmax) 2631 size = ptob(physmax - btop(addr) + 1); 2632 2633 num = btop(size); 2634 if (num == 0) 2635 continue; 2636 2637 if (num > npages) 2638 num = npages; 2639 2640 npages -= num; 2641 pages_done += num; 2642 base_pfn = btop(addr); 2643 2644 if (prom_debug) 2645 prom_printf("MEMSEG addr=0x%" PRIx64 2646 " pgs=0x%lx pfn 0x%lx-0x%lx\n", 2647 addr, num, base_pfn, base_pfn + num); 2648 2649 /* 2650 * Ignore pages below ddiphysmin to simplify ddi memory 2651 * allocation with non-zero addr_lo requests. 2652 */ 2653 if (base_pfn < ddiphysmin) { 2654 if (base_pfn + num <= ddiphysmin) 2655 continue; 2656 pp += (ddiphysmin - base_pfn); 2657 num -= (ddiphysmin - base_pfn); 2658 base_pfn = ddiphysmin; 2659 } 2660 2661 /* 2662 * mnode_xwa is greater than 1 when large pages regions can 2663 * cross memory node boundaries. To prevent the formation 2664 * of these large pages, configure the memsegs based on the 2665 * memory node ranges which had been made non-contiguous. 2666 */ 2667 if (mnode_xwa > 1) { 2668 2669 end_pfn = base_pfn + num - 1; 2670 ms = PFN_2_MEM_NODE(base_pfn); 2671 me = PFN_2_MEM_NODE(end_pfn); 2672 2673 if (ms != me) { 2674 /* 2675 * current range spans more than 1 memory node. 2676 * Set num to only the pfn range in the start 2677 * memory node. 2678 */ 2679 num = mem_node_config[ms].physmax - base_pfn 2680 + 1; 2681 ASSERT(end_pfn > mem_node_config[ms].physmax); 2682 } 2683 } 2684 2685 for (;;) { 2686 /* 2687 * Build the memsegs entry 2688 */ 2689 cur_memseg->pages = pp; 2690 cur_memseg->epages = pp + num; 2691 cur_memseg->pages_base = base_pfn; 2692 cur_memseg->pages_end = base_pfn + num; 2693 2694 /* 2695 * Insert into memseg list in decreasing pfn range 2696 * order. Low memory is typically more fragmented such 2697 * that this ordering keeps the larger ranges at the 2698 * front of the list for code that searches memseg. 2699 * This ASSERTS that the memsegs coming in from boot 2700 * are in increasing physical address order and not 2701 * contiguous. 2702 */ 2703 if (memsegs != NULL) { 2704 ASSERT(cur_memseg->pages_base >= 2705 memsegs->pages_end); 2706 cur_memseg->next = memsegs; 2707 } 2708 memsegs = cur_memseg; 2709 2710 /* 2711 * add_physmem() initializes the PSM part of the page 2712 * struct by calling the PSM back with add_physmem_cb(). 2713 * In addition it coalesces pages into larger pages as 2714 * it initializes them. 2715 */ 2716 add_physmem(pp, num, base_pfn); 2717 cur_memseg++; 2718 availrmem_initial += num; 2719 availrmem += num; 2720 2721 pp += num; 2722 if (ms >= me) 2723 break; 2724 2725 /* process next memory node range */ 2726 ms++; 2727 base_pfn = mem_node_config[ms].physbase; 2728 num = MIN(mem_node_config[ms].physmax, 2729 end_pfn) - base_pfn + 1; 2730 } 2731 } 2732 2733 PRM_DEBUG(availrmem_initial); 2734 PRM_DEBUG(availrmem); 2735 PRM_DEBUG(freemem); 2736 build_pfn_hash(); 2737 return (pages_done); 2738 } 2739 2740 /* 2741 * Kernel VM initialization. 2742 */ 2743 static void 2744 kvm_init(void) 2745 { 2746 ASSERT((((uintptr_t)s_text) & MMU_PAGEOFFSET) == 0); 2747 2748 /* 2749 * Put the kernel segments in kernel address space. 2750 */ 2751 rw_enter(&kas.a_lock, RW_WRITER); 2752 as_avlinit(&kas); 2753 2754 (void) seg_attach(&kas, s_text, e_moddata - s_text, &ktextseg); 2755 (void) segkmem_create(&ktextseg); 2756 2757 (void) seg_attach(&kas, (caddr_t)valloc_base, valloc_sz, &kvalloc); 2758 (void) segkmem_create(&kvalloc); 2759 2760 (void) seg_attach(&kas, kernelheap, 2761 ekernelheap - kernelheap, &kvseg); 2762 (void) segkmem_create(&kvseg); 2763 2764 if (core_size > 0) { 2765 PRM_POINT("attaching kvseg_core"); 2766 (void) seg_attach(&kas, (caddr_t)core_base, core_size, 2767 &kvseg_core); 2768 (void) segkmem_create(&kvseg_core); 2769 } 2770 2771 if (segziosize > 0) { 2772 PRM_POINT("attaching segzio"); 2773 (void) seg_attach(&kas, segzio_base, mmu_ptob(segziosize), 2774 &kzioseg); 2775 (void) segkmem_zio_create(&kzioseg); 2776 2777 /* create zio area covering new segment */ 2778 segkmem_zio_init(segzio_base, mmu_ptob(segziosize)); 2779 } 2780 2781 (void) seg_attach(&kas, kdi_segdebugbase, kdi_segdebugsize, &kdebugseg); 2782 (void) segkmem_create(&kdebugseg); 2783 2784 rw_exit(&kas.a_lock); 2785 2786 /* 2787 * Ensure that the red zone at kernelbase is never accessible. 2788 */ 2789 PRM_POINT("protecting redzone"); 2790 (void) as_setprot(&kas, (caddr_t)kernelbase, KERNEL_REDZONE_SIZE, 0); 2791 2792 /* 2793 * Make the text writable so that it can be hot patched by DTrace. 2794 */ 2795 (void) as_setprot(&kas, s_text, e_modtext - s_text, 2796 PROT_READ | PROT_WRITE | PROT_EXEC); 2797 2798 /* 2799 * Make data writable until end. 2800 */ 2801 (void) as_setprot(&kas, s_data, e_moddata - s_data, 2802 PROT_READ | PROT_WRITE | PROT_EXEC); 2803 } 2804 2805 #ifndef __xpv 2806 /* 2807 * Solaris adds an entry for Write Combining caching to the PAT 2808 */ 2809 static uint64_t pat_attr_reg = PAT_DEFAULT_ATTRIBUTE; 2810 2811 void 2812 pat_sync(void) 2813 { 2814 ulong_t cr0, cr0_orig, cr4; 2815 2816 if (!is_x86_feature(x86_featureset, X86FSET_PAT)) 2817 return; 2818 cr0_orig = cr0 = getcr0(); 2819 cr4 = getcr4(); 2820 2821 /* disable caching and flush all caches and TLBs */ 2822 cr0 |= CR0_CD; 2823 cr0 &= ~CR0_NW; 2824 setcr0(cr0); 2825 invalidate_cache(); 2826 if (cr4 & CR4_PGE) { 2827 setcr4(cr4 & ~(ulong_t)CR4_PGE); 2828 setcr4(cr4); 2829 } else { 2830 reload_cr3(); 2831 } 2832 2833 /* add our entry to the PAT */ 2834 wrmsr(REG_PAT, pat_attr_reg); 2835 2836 /* flush TLBs and cache again, then reenable cr0 caching */ 2837 if (cr4 & CR4_PGE) { 2838 setcr4(cr4 & ~(ulong_t)CR4_PGE); 2839 setcr4(cr4); 2840 } else { 2841 reload_cr3(); 2842 } 2843 invalidate_cache(); 2844 setcr0(cr0_orig); 2845 } 2846 2847 #endif /* !__xpv */ 2848 2849 #if defined(_SOFT_HOSTID) 2850 /* 2851 * On platforms that do not have a hardware serial number, attempt 2852 * to set one based on the contents of /etc/hostid. If this file does 2853 * not exist, assume that we are to generate a new hostid and set 2854 * it in the kernel, for subsequent saving by a userland process 2855 * once the system is up and the root filesystem is mounted r/w. 2856 * 2857 * In order to gracefully support upgrade on OpenSolaris, if 2858 * /etc/hostid does not exist, we will attempt to get a serial number 2859 * using the legacy method (/kernel/misc/sysinit). 2860 * 2861 * If that isn't present, we attempt to use an SMBIOS UUID, which is 2862 * a hardware serial number. Note that we don't automatically trust 2863 * all SMBIOS UUIDs (some older platforms are defective and ship duplicate 2864 * UUIDs in violation of the standard), we check against a blacklist. 2865 * 2866 * In an attempt to make the hostid less prone to abuse 2867 * (for license circumvention, etc), we store it in /etc/hostid 2868 * in rot47 format. 2869 */ 2870 extern volatile unsigned long tenmicrodata; 2871 static int atoi(char *); 2872 2873 /* 2874 * Set this to non-zero in /etc/system if you think your SMBIOS returns a 2875 * UUID that is not unique. (Also report it so that the smbios_uuid_blacklist 2876 * array can be updated.) 2877 */ 2878 int smbios_broken_uuid = 0; 2879 2880 /* 2881 * List of known bad UUIDs. This is just the lower 32-bit values, since 2882 * that's what we use for the host id. If your hostid falls here, you need 2883 * to contact your hardware OEM for a fix for your BIOS. 2884 */ 2885 static unsigned char 2886 smbios_uuid_blacklist[][16] = { 2887 2888 { /* Reported bad UUID (Google search) */ 2889 0x00, 0x02, 0x00, 0x03, 0x00, 0x04, 0x00, 0x05, 2890 0x00, 0x06, 0x00, 0x07, 0x00, 0x08, 0x00, 0x09, 2891 }, 2892 { /* Known bad DELL UUID */ 2893 0x4C, 0x4C, 0x45, 0x44, 0x00, 0x00, 0x20, 0x10, 2894 0x80, 0x20, 0x80, 0xC0, 0x4F, 0x20, 0x20, 0x20, 2895 }, 2896 { /* Uninitialized flash */ 2897 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 2898 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff 2899 }, 2900 { /* All zeros */ 2901 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 2902 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00 2903 }, 2904 }; 2905 2906 static int32_t 2907 uuid_to_hostid(const uint8_t *uuid) 2908 { 2909 /* 2910 * Although the UUIDs are 128-bits, they may not distribute entropy 2911 * evenly. We would like to use SHA or MD5, but those are located 2912 * in loadable modules and not available this early in boot. As we 2913 * don't need the values to be cryptographically strong, we just 2914 * generate 32-bit vaue by xor'ing the various sequences together, 2915 * which ensures that the entire UUID contributes to the hostid. 2916 */ 2917 uint32_t id = 0; 2918 2919 /* first check against the blacklist */ 2920 for (int i = 0; i < (sizeof (smbios_uuid_blacklist) / 16); i++) { 2921 if (bcmp(smbios_uuid_blacklist[0], uuid, 16) == 0) { 2922 cmn_err(CE_CONT, "?Broken SMBIOS UUID. " 2923 "Contact BIOS manufacturer for repair.\n"); 2924 return ((int32_t)HW_INVALID_HOSTID); 2925 } 2926 } 2927 2928 for (int i = 0; i < 16; i++) 2929 id ^= ((uuid[i]) << (8 * (i % sizeof (id)))); 2930 2931 /* Make sure return value is positive */ 2932 return (id & 0x7fffffff); 2933 } 2934 2935 static int32_t 2936 set_soft_hostid(void) 2937 { 2938 struct _buf *file; 2939 char tokbuf[MAXNAMELEN]; 2940 token_t token; 2941 int done = 0; 2942 u_longlong_t tmp; 2943 int i; 2944 int32_t hostid = (int32_t)HW_INVALID_HOSTID; 2945 unsigned char *c; 2946 hrtime_t tsc; 2947 smbios_system_t smsys; 2948 2949 /* 2950 * If /etc/hostid file not found, we'd like to get a pseudo 2951 * random number to use at the hostid. A nice way to do this 2952 * is to read the real time clock. To remain xen-compatible, 2953 * we can't poke the real hardware, so we use tsc_read() to 2954 * read the real time clock. However, there is an ominous 2955 * warning in tsc_read that says it can return zero, so we 2956 * deal with that possibility by falling back to using the 2957 * (hopefully random enough) value in tenmicrodata. 2958 */ 2959 2960 if ((file = kobj_open_file(hostid_file)) == (struct _buf *)-1) { 2961 /* 2962 * hostid file not found - try to load sysinit module 2963 * and see if it has a nonzero hostid value...use that 2964 * instead of generating a new hostid here if so. 2965 */ 2966 if ((i = modload("misc", "sysinit")) != -1) { 2967 if (strlen(hw_serial) > 0) 2968 hostid = (int32_t)atoi(hw_serial); 2969 (void) modunload(i); 2970 } 2971 2972 /* 2973 * We try to use the SMBIOS UUID. But not if it is blacklisted 2974 * in /etc/system. 2975 */ 2976 if ((hostid == HW_INVALID_HOSTID) && 2977 (smbios_broken_uuid == 0) && 2978 (ksmbios != NULL) && 2979 (smbios_info_system(ksmbios, &smsys) != SMB_ERR) && 2980 (smsys.smbs_uuidlen >= 16)) { 2981 hostid = uuid_to_hostid(smsys.smbs_uuid); 2982 } 2983 2984 /* 2985 * Generate a "random" hostid using the clock. These 2986 * hostids will change on each boot if the value is not 2987 * saved to a persistent /etc/hostid file. 2988 */ 2989 if (hostid == HW_INVALID_HOSTID) { 2990 tsc = tsc_read(); 2991 if (tsc == 0) /* tsc_read can return zero sometimes */ 2992 hostid = (int32_t)tenmicrodata & 0x0CFFFFF; 2993 else 2994 hostid = (int32_t)tsc & 0x0CFFFFF; 2995 } 2996 } else { 2997 /* hostid file found */ 2998 while (!done) { 2999 token = kobj_lex(file, tokbuf, sizeof (tokbuf)); 3000 3001 switch (token) { 3002 case POUND: 3003 /* 3004 * skip comments 3005 */ 3006 kobj_find_eol(file); 3007 break; 3008 case STRING: 3009 /* 3010 * un-rot47 - obviously this 3011 * nonsense is ascii-specific 3012 */ 3013 for (c = (unsigned char *)tokbuf; 3014 *c != '\0'; c++) { 3015 *c += 47; 3016 if (*c > '~') 3017 *c -= 94; 3018 else if (*c < '!') 3019 *c += 94; 3020 } 3021 /* 3022 * now we should have a real number 3023 */ 3024 3025 if (kobj_getvalue(tokbuf, &tmp) != 0) 3026 kobj_file_err(CE_WARN, file, 3027 "Bad value %s for hostid", 3028 tokbuf); 3029 else 3030 hostid = (int32_t)tmp; 3031 3032 break; 3033 case EOF: 3034 done = 1; 3035 /* FALLTHROUGH */ 3036 case NEWLINE: 3037 kobj_newline(file); 3038 break; 3039 default: 3040 break; 3041 3042 } 3043 } 3044 if (hostid == HW_INVALID_HOSTID) /* didn't find a hostid */ 3045 kobj_file_err(CE_WARN, file, 3046 "hostid missing or corrupt"); 3047 3048 kobj_close_file(file); 3049 } 3050 /* 3051 * hostid is now the value read from /etc/hostid, or the 3052 * new hostid we generated in this routine or HW_INVALID_HOSTID if not 3053 * set. 3054 */ 3055 return (hostid); 3056 } 3057 3058 static int 3059 atoi(char *p) 3060 { 3061 int i = 0; 3062 3063 while (*p != '\0') 3064 i = 10 * i + (*p++ - '0'); 3065 3066 return (i); 3067 } 3068 3069 #endif /* _SOFT_HOSTID */ 3070 3071 void 3072 get_system_configuration(void) 3073 { 3074 char prop[32]; 3075 u_longlong_t nodes_ll, cpus_pernode_ll, lvalue; 3076 3077 if (BOP_GETPROPLEN(bootops, "nodes") > sizeof (prop) || 3078 BOP_GETPROP(bootops, "nodes", prop) < 0 || 3079 kobj_getvalue(prop, &nodes_ll) == -1 || 3080 nodes_ll > MAXNODES || 3081 BOP_GETPROPLEN(bootops, "cpus_pernode") > sizeof (prop) || 3082 BOP_GETPROP(bootops, "cpus_pernode", prop) < 0 || 3083 kobj_getvalue(prop, &cpus_pernode_ll) == -1) { 3084 system_hardware.hd_nodes = 1; 3085 system_hardware.hd_cpus_per_node = 0; 3086 } else { 3087 system_hardware.hd_nodes = (int)nodes_ll; 3088 system_hardware.hd_cpus_per_node = (int)cpus_pernode_ll; 3089 } 3090 3091 if (BOP_GETPROPLEN(bootops, "kernelbase") > sizeof (prop) || 3092 BOP_GETPROP(bootops, "kernelbase", prop) < 0 || 3093 kobj_getvalue(prop, &lvalue) == -1) 3094 eprom_kernelbase = NULL; 3095 else 3096 eprom_kernelbase = (uintptr_t)lvalue; 3097 3098 if (BOP_GETPROPLEN(bootops, "segmapsize") > sizeof (prop) || 3099 BOP_GETPROP(bootops, "segmapsize", prop) < 0 || 3100 kobj_getvalue(prop, &lvalue) == -1) 3101 segmapsize = SEGMAPDEFAULT; 3102 else 3103 segmapsize = (uintptr_t)lvalue; 3104 3105 if (BOP_GETPROPLEN(bootops, "segmapfreelists") > sizeof (prop) || 3106 BOP_GETPROP(bootops, "segmapfreelists", prop) < 0 || 3107 kobj_getvalue(prop, &lvalue) == -1) 3108 segmapfreelists = 0; /* use segmap driver default */ 3109 else 3110 segmapfreelists = (int)lvalue; 3111 3112 /* physmem used to be here, but moved much earlier to fakebop.c */ 3113 } 3114 3115 /* 3116 * Add to a memory list. 3117 * start = start of new memory segment 3118 * len = length of new memory segment in bytes 3119 * new = pointer to a new struct memlist 3120 * memlistp = memory list to which to add segment. 3121 */ 3122 void 3123 memlist_add( 3124 uint64_t start, 3125 uint64_t len, 3126 struct memlist *new, 3127 struct memlist **memlistp) 3128 { 3129 struct memlist *cur; 3130 uint64_t end = start + len; 3131 3132 new->ml_address = start; 3133 new->ml_size = len; 3134 3135 cur = *memlistp; 3136 3137 while (cur) { 3138 if (cur->ml_address >= end) { 3139 new->ml_next = cur; 3140 *memlistp = new; 3141 new->ml_prev = cur->ml_prev; 3142 cur->ml_prev = new; 3143 return; 3144 } 3145 ASSERT(cur->ml_address + cur->ml_size <= start); 3146 if (cur->ml_next == NULL) { 3147 cur->ml_next = new; 3148 new->ml_prev = cur; 3149 new->ml_next = NULL; 3150 return; 3151 } 3152 memlistp = &cur->ml_next; 3153 cur = cur->ml_next; 3154 } 3155 } 3156 3157 void 3158 kobj_vmem_init(vmem_t **text_arena, vmem_t **data_arena) 3159 { 3160 size_t tsize = e_modtext - modtext; 3161 size_t dsize = e_moddata - moddata; 3162 3163 *text_arena = vmem_create("module_text", tsize ? modtext : NULL, tsize, 3164 1, segkmem_alloc, segkmem_free, heaptext_arena, 0, VM_SLEEP); 3165 *data_arena = vmem_create("module_data", dsize ? moddata : NULL, dsize, 3166 1, segkmem_alloc, segkmem_free, heap32_arena, 0, VM_SLEEP); 3167 } 3168 3169 caddr_t 3170 kobj_text_alloc(vmem_t *arena, size_t size) 3171 { 3172 return (vmem_alloc(arena, size, VM_SLEEP | VM_BESTFIT)); 3173 } 3174 3175 /*ARGSUSED*/ 3176 caddr_t 3177 kobj_texthole_alloc(caddr_t addr, size_t size) 3178 { 3179 panic("unexpected call to kobj_texthole_alloc()"); 3180 /*NOTREACHED*/ 3181 return (0); 3182 } 3183 3184 /*ARGSUSED*/ 3185 void 3186 kobj_texthole_free(caddr_t addr, size_t size) 3187 { 3188 panic("unexpected call to kobj_texthole_free()"); 3189 } 3190 3191 /* 3192 * This is called just after configure() in startup(). 3193 * 3194 * The ISALIST concept is a bit hopeless on Intel, because 3195 * there's no guarantee of an ever-more-capable processor 3196 * given that various parts of the instruction set may appear 3197 * and disappear between different implementations. 3198 * 3199 * While it would be possible to correct it and even enhance 3200 * it somewhat, the explicit hardware capability bitmask allows 3201 * more flexibility. 3202 * 3203 * So, we just leave this alone. 3204 */ 3205 void 3206 setx86isalist(void) 3207 { 3208 char *tp; 3209 size_t len; 3210 extern char *isa_list; 3211 3212 #define TBUFSIZE 1024 3213 3214 tp = kmem_alloc(TBUFSIZE, KM_SLEEP); 3215 *tp = '\0'; 3216 3217 #if defined(__amd64) 3218 (void) strcpy(tp, "amd64 "); 3219 #endif 3220 3221 switch (x86_vendor) { 3222 case X86_VENDOR_Intel: 3223 case X86_VENDOR_AMD: 3224 case X86_VENDOR_TM: 3225 if (is_x86_feature(x86_featureset, X86FSET_CMOV)) { 3226 /* 3227 * Pentium Pro or later 3228 */ 3229 (void) strcat(tp, "pentium_pro"); 3230 (void) strcat(tp, 3231 is_x86_feature(x86_featureset, X86FSET_MMX) ? 3232 "+mmx pentium_pro " : " "); 3233 } 3234 /*FALLTHROUGH*/ 3235 case X86_VENDOR_Cyrix: 3236 /* 3237 * The Cyrix 6x86 does not have any Pentium features 3238 * accessible while not at privilege level 0. 3239 */ 3240 if (is_x86_feature(x86_featureset, X86FSET_CPUID)) { 3241 (void) strcat(tp, "pentium"); 3242 (void) strcat(tp, 3243 is_x86_feature(x86_featureset, X86FSET_MMX) ? 3244 "+mmx pentium " : " "); 3245 } 3246 break; 3247 default: 3248 break; 3249 } 3250 (void) strcat(tp, "i486 i386 i86"); 3251 len = strlen(tp) + 1; /* account for NULL at end of string */ 3252 isa_list = strcpy(kmem_alloc(len, KM_SLEEP), tp); 3253 kmem_free(tp, TBUFSIZE); 3254 3255 #undef TBUFSIZE 3256 } 3257 3258 3259 #ifdef __amd64 3260 3261 void * 3262 device_arena_alloc(size_t size, int vm_flag) 3263 { 3264 return (vmem_alloc(device_arena, size, vm_flag)); 3265 } 3266 3267 void 3268 device_arena_free(void *vaddr, size_t size) 3269 { 3270 vmem_free(device_arena, vaddr, size); 3271 } 3272 3273 #else /* __i386 */ 3274 3275 void * 3276 device_arena_alloc(size_t size, int vm_flag) 3277 { 3278 caddr_t vaddr; 3279 uintptr_t v; 3280 size_t start; 3281 size_t end; 3282 3283 vaddr = vmem_alloc(heap_arena, size, vm_flag); 3284 if (vaddr == NULL) 3285 return (NULL); 3286 3287 v = (uintptr_t)vaddr; 3288 ASSERT(v >= kernelbase); 3289 ASSERT(v + size <= valloc_base); 3290 3291 start = btop(v - kernelbase); 3292 end = btop(v + size - 1 - kernelbase); 3293 ASSERT(start < toxic_bit_map_len); 3294 ASSERT(end < toxic_bit_map_len); 3295 3296 while (start <= end) { 3297 BT_ATOMIC_SET(toxic_bit_map, start); 3298 ++start; 3299 } 3300 return (vaddr); 3301 } 3302 3303 void 3304 device_arena_free(void *vaddr, size_t size) 3305 { 3306 uintptr_t v = (uintptr_t)vaddr; 3307 size_t start; 3308 size_t end; 3309 3310 ASSERT(v >= kernelbase); 3311 ASSERT(v + size <= valloc_base); 3312 3313 start = btop(v - kernelbase); 3314 end = btop(v + size - 1 - kernelbase); 3315 ASSERT(start < toxic_bit_map_len); 3316 ASSERT(end < toxic_bit_map_len); 3317 3318 while (start <= end) { 3319 ASSERT(BT_TEST(toxic_bit_map, start) != 0); 3320 BT_ATOMIC_CLEAR(toxic_bit_map, start); 3321 ++start; 3322 } 3323 vmem_free(heap_arena, vaddr, size); 3324 } 3325 3326 /* 3327 * returns 1st address in range that is in device arena, or NULL 3328 * if len is not NULL it returns the length of the toxic range 3329 */ 3330 void * 3331 device_arena_contains(void *vaddr, size_t size, size_t *len) 3332 { 3333 uintptr_t v = (uintptr_t)vaddr; 3334 uintptr_t eaddr = v + size; 3335 size_t start; 3336 size_t end; 3337 3338 /* 3339 * if called very early by kmdb, just return NULL 3340 */ 3341 if (toxic_bit_map == NULL) 3342 return (NULL); 3343 3344 /* 3345 * First check if we're completely outside the bitmap range. 3346 */ 3347 if (v >= valloc_base || eaddr < kernelbase) 3348 return (NULL); 3349 3350 /* 3351 * Trim ends of search to look at only what the bitmap covers. 3352 */ 3353 if (v < kernelbase) 3354 v = kernelbase; 3355 start = btop(v - kernelbase); 3356 end = btop(eaddr - kernelbase); 3357 if (end >= toxic_bit_map_len) 3358 end = toxic_bit_map_len; 3359 3360 if (bt_range(toxic_bit_map, &start, &end, end) == 0) 3361 return (NULL); 3362 3363 v = kernelbase + ptob(start); 3364 if (len != NULL) 3365 *len = ptob(end - start); 3366 return ((void *)v); 3367 } 3368 3369 #endif /* __i386 */ 3370