1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 /* 22 * Copyright 2007 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 24 */ 25 26 #pragma ident "%Z%%M% %I% %E% SMI" 27 28 #include <sys/types.h> 29 #include <sys/t_lock.h> 30 #include <sys/param.h> 31 #include <sys/sysmacros.h> 32 #include <sys/signal.h> 33 #include <sys/systm.h> 34 #include <sys/user.h> 35 #include <sys/mman.h> 36 #include <sys/vm.h> 37 #include <sys/conf.h> 38 #include <sys/avintr.h> 39 #include <sys/autoconf.h> 40 #include <sys/disp.h> 41 #include <sys/class.h> 42 #include <sys/bitmap.h> 43 44 #include <sys/privregs.h> 45 46 #include <sys/proc.h> 47 #include <sys/buf.h> 48 #include <sys/kmem.h> 49 #include <sys/mem.h> 50 #include <sys/kstat.h> 51 52 #include <sys/reboot.h> 53 54 #include <sys/cred.h> 55 #include <sys/vnode.h> 56 #include <sys/file.h> 57 58 #include <sys/procfs.h> 59 60 #include <sys/vfs.h> 61 #include <sys/cmn_err.h> 62 #include <sys/utsname.h> 63 #include <sys/debug.h> 64 #include <sys/kdi.h> 65 66 #include <sys/dumphdr.h> 67 #include <sys/bootconf.h> 68 #include <sys/varargs.h> 69 #include <sys/promif.h> 70 #include <sys/modctl.h> /* for "procfs" hack */ 71 72 #include <sys/sunddi.h> 73 #include <sys/sunndi.h> 74 #include <sys/ndi_impldefs.h> 75 #include <sys/ddidmareq.h> 76 #include <sys/psw.h> 77 #include <sys/regset.h> 78 #include <sys/clock.h> 79 #include <sys/pte.h> 80 #include <sys/tss.h> 81 #include <sys/stack.h> 82 #include <sys/trap.h> 83 #include <sys/fp.h> 84 #include <vm/anon.h> 85 #include <vm/as.h> 86 #include <vm/page.h> 87 #include <vm/seg.h> 88 #include <vm/seg_dev.h> 89 #include <vm/seg_kmem.h> 90 #include <vm/seg_kpm.h> 91 #include <vm/seg_map.h> 92 #include <vm/seg_vn.h> 93 #include <vm/seg_kp.h> 94 #include <sys/memnode.h> 95 #include <vm/vm_dep.h> 96 #include <sys/thread.h> 97 #include <sys/sysconf.h> 98 #include <sys/vm_machparam.h> 99 #include <sys/archsystm.h> 100 #include <sys/machsystm.h> 101 #include <vm/hat.h> 102 #include <vm/hat_i86.h> 103 #include <sys/pmem.h> 104 #include <sys/smp_impldefs.h> 105 #include <sys/x86_archext.h> 106 #include <sys/segments.h> 107 #include <sys/clconf.h> 108 #include <sys/kobj.h> 109 #include <sys/kobj_lex.h> 110 #include <sys/cpc_impl.h> 111 #include <sys/pg.h> 112 #include <sys/x86_archext.h> 113 #include <sys/cpu_module.h> 114 #include <sys/smbios.h> 115 #include <sys/debug_info.h> 116 117 118 #include <sys/bootinfo.h> 119 #include <vm/kboot_mmu.h> 120 121 extern void progressbar_init(void); 122 extern void progressbar_start(void); 123 extern void brand_init(void); 124 125 /* 126 * XXX make declaration below "static" when drivers no longer use this 127 * interface. 128 */ 129 extern caddr_t p0_va; /* Virtual address for accessing physical page 0 */ 130 131 /* 132 * segkp 133 */ 134 extern int segkp_fromheap; 135 136 static void kvm_init(void); 137 static void startup_init(void); 138 static void startup_memlist(void); 139 static void startup_kmem(void); 140 static void startup_modules(void); 141 static void startup_vm(void); 142 static void startup_end(void); 143 144 /* 145 * Declare these as initialized data so we can patch them. 146 */ 147 #ifdef __i386 148 /* 149 * Due to virtual address space limitations running in 32 bit mode, restrict 150 * the amount of physical memory configured to a max of PHYSMEM32 pages (16g). 151 * 152 * If the physical max memory size of 64g were allowed to be configured, the 153 * size of user virtual address space will be less than 1g. A limited user 154 * address space greatly reduces the range of applications that can run. 155 * 156 * If more physical memory than PHYSMEM32 is required, users should preferably 157 * run in 64 bit mode which has no virtual address space limitation issues. 158 * 159 * If 64 bit mode is not available (as in IA32) and/or more physical memory 160 * than PHYSMEM32 is required in 32 bit mode, physmem can be set to the desired 161 * value or to 0 (to configure all available memory) via eeprom(1M). kernelbase 162 * should also be carefully tuned to balance out the need of the user 163 * application while minimizing the risk of kernel heap exhaustion due to 164 * kernelbase being set too high. 165 */ 166 #define PHYSMEM32 0x400000 167 168 pgcnt_t physmem = PHYSMEM32; 169 #else 170 pgcnt_t physmem = 0; /* memory size in pages, patch if you want less */ 171 #endif 172 pgcnt_t obp_pages; /* Memory used by PROM for its text and data */ 173 174 char *kobj_file_buf; 175 int kobj_file_bufsize; /* set in /etc/system */ 176 177 /* Global variables for MP support. Used in mp_startup */ 178 caddr_t rm_platter_va; 179 uint32_t rm_platter_pa; 180 181 int auto_lpg_disable = 1; 182 183 /* 184 * Some CPUs have holes in the middle of the 64-bit virtual address range. 185 */ 186 uintptr_t hole_start, hole_end; 187 188 /* 189 * kpm mapping window 190 */ 191 caddr_t kpm_vbase; 192 size_t kpm_size; 193 static int kpm_desired = 0; /* Do we want to try to use segkpm? */ 194 195 /* 196 * Configuration parameters set at boot time. 197 */ 198 199 caddr_t econtig; /* end of first block of contiguous kernel */ 200 201 struct bootops *bootops = 0; /* passed in from boot */ 202 struct bootops **bootopsp; 203 struct boot_syscalls *sysp; /* passed in from boot */ 204 205 char bootblock_fstype[16]; 206 207 char kern_bootargs[OBP_MAXPATHLEN]; 208 209 /* 210 * ZFS zio segment. This allows us to exclude large portions of ZFS data that 211 * gets cached in kmem caches on the heap. If this is set to zero, we allocate 212 * zio buffers from their own segment, otherwise they are allocated from the 213 * heap. The optimization of allocating zio buffers from their own segment is 214 * only valid on 64-bit kernels. 215 */ 216 #if defined(__amd64) 217 int segzio_fromheap = 0; 218 #else 219 int segzio_fromheap = 1; 220 #endif 221 222 /* 223 * new memory fragmentations are possible in startup() due to BOP_ALLOCs. this 224 * depends on number of BOP_ALLOC calls made and requested size, memory size 225 * combination and whether boot.bin memory needs to be freed. 226 */ 227 #define POSS_NEW_FRAGMENTS 12 228 229 /* 230 * VM data structures 231 */ 232 long page_hashsz; /* Size of page hash table (power of two) */ 233 struct page *pp_base; /* Base of initial system page struct array */ 234 struct page **page_hash; /* Page hash table */ 235 struct seg ktextseg; /* Segment used for kernel executable image */ 236 struct seg kvalloc; /* Segment used for "valloc" mapping */ 237 struct seg kpseg; /* Segment used for pageable kernel virt mem */ 238 struct seg kmapseg; /* Segment used for generic kernel mappings */ 239 struct seg kdebugseg; /* Segment used for the kernel debugger */ 240 241 struct seg *segkmap = &kmapseg; /* Kernel generic mapping segment */ 242 static struct seg *segmap = &kmapseg; /* easier to use name for in here */ 243 244 struct seg *segkp = &kpseg; /* Pageable kernel virtual memory segment */ 245 246 #if defined(__amd64) 247 struct seg kvseg_core; /* Segment used for the core heap */ 248 struct seg kpmseg; /* Segment used for physical mapping */ 249 struct seg *segkpm = &kpmseg; /* 64bit kernel physical mapping segment */ 250 #else 251 struct seg *segkpm = NULL; /* Unused on IA32 */ 252 #endif 253 254 caddr_t segkp_base; /* Base address of segkp */ 255 caddr_t segzio_base; /* Base address of segzio */ 256 #if defined(__amd64) 257 pgcnt_t segkpsize = btop(SEGKPDEFSIZE); /* size of segkp segment in pages */ 258 #else 259 pgcnt_t segkpsize = 0; 260 #endif 261 pgcnt_t segziosize = 0; /* size of zio segment in pages */ 262 263 /* 264 * VA range available to the debugger 265 */ 266 const caddr_t kdi_segdebugbase = (const caddr_t)SEGDEBUGBASE; 267 const size_t kdi_segdebugsize = SEGDEBUGSIZE; 268 269 struct memseg *memseg_base; 270 struct vnode unused_pages_vp; 271 272 #define FOURGB 0x100000000LL 273 274 struct memlist *memlist; 275 276 caddr_t s_text; /* start of kernel text segment */ 277 caddr_t e_text; /* end of kernel text segment */ 278 caddr_t s_data; /* start of kernel data segment */ 279 caddr_t e_data; /* end of kernel data segment */ 280 caddr_t modtext; /* start of loadable module text reserved */ 281 caddr_t e_modtext; /* end of loadable module text reserved */ 282 caddr_t moddata; /* start of loadable module data reserved */ 283 caddr_t e_moddata; /* end of loadable module data reserved */ 284 285 struct memlist *phys_install; /* Total installed physical memory */ 286 struct memlist *phys_avail; /* Total available physical memory */ 287 288 /* 289 * kphysm_init returns the number of pages that were processed 290 */ 291 static pgcnt_t kphysm_init(page_t *, pgcnt_t); 292 293 #define IO_PROP_SIZE 64 /* device property size */ 294 295 /* 296 * a couple useful roundup macros 297 */ 298 #define ROUND_UP_PAGE(x) \ 299 ((uintptr_t)P2ROUNDUP((uintptr_t)(x), (uintptr_t)MMU_PAGESIZE)) 300 #define ROUND_UP_LPAGE(x) \ 301 ((uintptr_t)P2ROUNDUP((uintptr_t)(x), mmu.level_size[1])) 302 #define ROUND_UP_4MEG(x) \ 303 ((uintptr_t)P2ROUNDUP((uintptr_t)(x), (uintptr_t)FOUR_MEG)) 304 #define ROUND_UP_TOPLEVEL(x) \ 305 ((uintptr_t)P2ROUNDUP((uintptr_t)(x), mmu.level_size[mmu.max_level])) 306 307 /* 308 * 32-bit Kernel's Virtual memory layout. 309 * +-----------------------+ 310 * | | 311 * 0xFFC00000 -|-----------------------|- ARGSBASE 312 * | debugger | 313 * 0xFF800000 -|-----------------------|- SEGDEBUGBASE 314 * | Kernel Data | 315 * 0xFEC00000 -|-----------------------| 316 * | Kernel Text | 317 * 0xFE800000 -|-----------------------|- KERNEL_TEXT 318 * |--- GDT ---|- GDT page (GDT_VA) 319 * |--- debug info ---|- debug info (DEBUG_INFO_VA) 320 * | | 321 * | page_t structures | 322 * | memsegs, memlists, | 323 * | page hash, etc. | 324 * --- -|-----------------------|- ekernelheap, valloc_base (floating) 325 * | | (segkp is just an arena in the heap) 326 * | | 327 * | kvseg | 328 * | | 329 * | | 330 * --- -|-----------------------|- kernelheap (floating) 331 * | Segkmap | 332 * 0xC3002000 -|-----------------------|- segmap_start (floating) 333 * | Red Zone | 334 * 0xC3000000 -|-----------------------|- kernelbase / userlimit (floating) 335 * | | || 336 * | Shared objects | \/ 337 * | | 338 * : : 339 * | user data | 340 * |-----------------------| 341 * | user text | 342 * 0x08048000 -|-----------------------| 343 * | user stack | 344 * : : 345 * | invalid | 346 * 0x00000000 +-----------------------+ 347 * 348 * 349 * 64-bit Kernel's Virtual memory layout. (assuming 64 bit app) 350 * +-----------------------+ 351 * | | 352 * 0xFFFFFFFF.FFC00000 |-----------------------|- ARGSBASE 353 * | debugger (?) | 354 * 0xFFFFFFFF.FF800000 |-----------------------|- SEGDEBUGBASE 355 * | unused | 356 * +-----------------------+ 357 * | Kernel Data | 358 * 0xFFFFFFFF.FBC00000 |-----------------------| 359 * | Kernel Text | 360 * 0xFFFFFFFF.FB800000 |-----------------------|- KERNEL_TEXT 361 * |--- GDT ---|- GDT page (GDT_VA) 362 * |--- debug info ---|- debug info (DEBUG_INFO_VA) 363 * | | 364 * | Core heap | (used for loadable modules) 365 * 0xFFFFFFFF.C0000000 |-----------------------|- core_base / ekernelheap 366 * | Kernel | 367 * | heap | 368 * 0xFFFFFXXX.XXX00000 |-----------------------|- kernelheap (floating) 369 * | segmap | 370 * 0xFFFFFXXX.XXX00000 |-----------------------|- segmap_start (floating) 371 * | device mappings | 372 * 0xFFFFFXXX.XXX00000 |-----------------------|- toxic_addr (floating) 373 * | segzio | 374 * 0xFFFFFXXX.XXX00000 |-----------------------|- segzio_base (floating) 375 * | segkp | 376 * --- |-----------------------|- segkp_base (floating) 377 * | page_t structures | valloc_base + valloc_sz 378 * | memsegs, memlists, | 379 * | page hash, etc. | 380 * 0xFFFFFF00.00000000 |-----------------------|- valloc_base 381 * | segkpm | 382 * 0xFFFFFE00.00000000 |-----------------------| 383 * | Red Zone | 384 * 0xFFFFFD80.00000000 |-----------------------|- KERNELBASE 385 * | User stack |- User space memory 386 * | | 387 * | shared objects, etc | (grows downwards) 388 * : : 389 * | | 390 * 0xFFFF8000.00000000 |-----------------------| 391 * | | 392 * | VA Hole / unused | 393 * | | 394 * 0x00008000.00000000 |-----------------------| 395 * | | 396 * | | 397 * : : 398 * | user heap | (grows upwards) 399 * | | 400 * | user data | 401 * |-----------------------| 402 * | user text | 403 * 0x00000000.04000000 |-----------------------| 404 * | invalid | 405 * 0x00000000.00000000 +-----------------------+ 406 * 407 * A 32 bit app on the 64 bit kernel sees the same layout as on the 32 bit 408 * kernel, except that userlimit is raised to 0xfe000000 409 * 410 * Floating values: 411 * 412 * valloc_base: start of the kernel's memory management/tracking data 413 * structures. This region contains page_t structures for 414 * physical memory, memsegs, memlists, and the page hash. 415 * 416 * core_base: start of the kernel's "core" heap area on 64-bit systems. 417 * This area is intended to be used for global data as well as for module 418 * text/data that does not fit into the nucleus pages. The core heap is 419 * restricted to a 2GB range, allowing every address within it to be 420 * accessed using rip-relative addressing 421 * 422 * ekernelheap: end of kernelheap and start of segmap. 423 * 424 * kernelheap: start of kernel heap. On 32-bit systems, this starts right 425 * above a red zone that separates the user's address space from the 426 * kernel's. On 64-bit systems, it sits above segkp and segkpm. 427 * 428 * segmap_start: start of segmap. The length of segmap can be modified 429 * by changing segmapsize in /etc/system (preferred) or eeprom (deprecated). 430 * The default length is 16MB on 32-bit systems and 64MB on 64-bit systems. 431 * 432 * kernelbase: On a 32-bit kernel the default value of 0xd4000000 will be 433 * decreased by 2X the size required for page_t. This allows the kernel 434 * heap to grow in size with physical memory. With sizeof(page_t) == 80 435 * bytes, the following shows the values of kernelbase and kernel heap 436 * sizes for different memory configurations (assuming default segmap and 437 * segkp sizes). 438 * 439 * mem size for kernelbase kernel heap 440 * size page_t's size 441 * ---- --------- ---------- ----------- 442 * 1gb 0x01400000 0xd1800000 684MB 443 * 2gb 0x02800000 0xcf000000 704MB 444 * 4gb 0x05000000 0xca000000 744MB 445 * 6gb 0x07800000 0xc5000000 784MB 446 * 8gb 0x0a000000 0xc0000000 824MB 447 * 16gb 0x14000000 0xac000000 984MB 448 * 32gb 0x28000000 0x84000000 1304MB 449 * 64gb 0x50000000 0x34000000 1944MB (*) 450 * 451 * kernelbase is less than the abi minimum of 0xc0000000 for memory 452 * configurations above 8gb. 453 * 454 * (*) support for memory configurations above 32gb will require manual tuning 455 * of kernelbase to balance out the need of user applications. 456 */ 457 458 /* real-time-clock initialization parameters */ 459 extern time_t process_rtc_config_file(void); 460 461 char *final_kernelheap; 462 char *boot_kernelheap; 463 uintptr_t kernelbase; 464 uintptr_t postbootkernelbase; /* not set till boot loader is gone */ 465 uintptr_t eprom_kernelbase; 466 size_t segmapsize; 467 static uintptr_t segmap_reserved; 468 uintptr_t segmap_start; 469 int segmapfreelists; 470 pgcnt_t npages; 471 pgcnt_t orig_npages; 472 size_t core_size; /* size of "core" heap */ 473 uintptr_t core_base; /* base address of "core" heap */ 474 475 /* 476 * List of bootstrap pages. We mark these as allocated in startup. 477 * release_bootstrap() will free them when we're completely done with 478 * the bootstrap. 479 */ 480 static page_t *bootpages; 481 482 /* 483 * boot time pages that have a vnode from the ramdisk will keep that forever. 484 */ 485 static page_t *rd_pages; 486 487 struct system_hardware system_hardware; 488 489 /* 490 * Enable some debugging messages concerning memory usage... 491 */ 492 static void 493 print_memlist(char *title, struct memlist *mp) 494 { 495 prom_printf("MEMLIST: %s:\n", title); 496 while (mp != NULL) { 497 prom_printf("\tAddress 0x%" PRIx64 ", size 0x%" PRIx64 "\n", 498 mp->address, mp->size); 499 mp = mp->next; 500 } 501 } 502 503 /* 504 * XX64 need a comment here.. are these just default values, surely 505 * we read the "cpuid" type information to figure this out. 506 */ 507 int l2cache_sz = 0x80000; 508 int l2cache_linesz = 0x40; 509 int l2cache_assoc = 1; 510 511 static size_t textrepl_min_gb = 10; 512 513 /* 514 * on 64 bit we use a predifined VA range for mapping devices in the kernel 515 * on 32 bit the mappings are intermixed in the heap, so we use a bit map 516 */ 517 #ifdef __amd64 518 519 vmem_t *device_arena; 520 uintptr_t toxic_addr = (uintptr_t)NULL; 521 size_t toxic_size = 1024 * 1024 * 1024; /* Sparc uses 1 gig too */ 522 523 #else /* __i386 */ 524 525 ulong_t *toxic_bit_map; /* one bit for each 4k of VA in heap_arena */ 526 size_t toxic_bit_map_len = 0; /* in bits */ 527 528 #endif /* __i386 */ 529 530 /* 531 * Simple boot time debug facilities 532 */ 533 static char *prm_dbg_str[] = { 534 "%s:%d: '%s' is 0x%x\n", 535 "%s:%d: '%s' is 0x%llx\n" 536 }; 537 538 int prom_debug; 539 540 #define PRM_DEBUG(q) if (prom_debug) \ 541 prom_printf(prm_dbg_str[sizeof (q) >> 3], "startup.c", __LINE__, #q, q); 542 #define PRM_POINT(q) if (prom_debug) \ 543 prom_printf("%s:%d: %s\n", "startup.c", __LINE__, q); 544 545 /* 546 * This structure is used to keep track of the intial allocations 547 * done in startup_memlist(). The value of NUM_ALLOCATIONS needs to 548 * be >= the number of ADD_TO_ALLOCATIONS() executed in the code. 549 */ 550 #define NUM_ALLOCATIONS 7 551 int num_allocations = 0; 552 struct { 553 void **al_ptr; 554 size_t al_size; 555 } allocations[NUM_ALLOCATIONS]; 556 size_t valloc_sz = 0; 557 uintptr_t valloc_base; 558 559 #define ADD_TO_ALLOCATIONS(ptr, size) { \ 560 size = ROUND_UP_PAGE(size); \ 561 if (num_allocations == NUM_ALLOCATIONS) \ 562 panic("too many ADD_TO_ALLOCATIONS()"); \ 563 allocations[num_allocations].al_ptr = (void**)&ptr; \ 564 allocations[num_allocations].al_size = size; \ 565 valloc_sz += size; \ 566 ++num_allocations; \ 567 } 568 569 /* 570 * Allocate all the initial memory needed by the page allocator. 571 */ 572 static void 573 perform_allocations(void) 574 { 575 caddr_t mem; 576 int i; 577 int valloc_align; 578 579 PRM_DEBUG(valloc_base); 580 PRM_DEBUG(valloc_sz); 581 valloc_align = mmu.level_size[mmu.max_page_level > 0]; 582 mem = BOP_ALLOC(bootops, (caddr_t)valloc_base, valloc_sz, valloc_align); 583 if (mem != (caddr_t)valloc_base) 584 panic("BOP_ALLOC() failed"); 585 bzero(mem, valloc_sz); 586 for (i = 0; i < num_allocations; ++i) { 587 *allocations[i].al_ptr = (void *)mem; 588 mem += allocations[i].al_size; 589 } 590 } 591 592 /* 593 * Our world looks like this at startup time. 594 * 595 * In a 32-bit OS, boot loads the kernel text at 0xfe800000 and kernel data 596 * at 0xfec00000. On a 64-bit OS, kernel text and data are loaded at 597 * 0xffffffff.fe800000 and 0xffffffff.fec00000 respectively. Those 598 * addresses are fixed in the binary at link time. 599 * 600 * On the text page: 601 * unix/genunix/krtld/module text loads. 602 * 603 * On the data page: 604 * unix/genunix/krtld/module data loads. 605 * 606 * Machine-dependent startup code 607 */ 608 void 609 startup(void) 610 { 611 extern void startup_bios_disk(void); 612 extern void startup_pci_bios(void); 613 /* 614 * Make sure that nobody tries to use sekpm until we have 615 * initialized it properly. 616 */ 617 #if defined(__amd64) 618 kpm_desired = kpm_enable; 619 #endif 620 kpm_enable = 0; 621 622 progressbar_init(); 623 startup_init(); 624 startup_memlist(); 625 startup_kmem(); 626 startup_pci_bios(); 627 startup_modules(); 628 startup_bios_disk(); 629 startup_vm(); 630 startup_end(); 631 progressbar_start(); 632 } 633 634 static void 635 startup_init() 636 { 637 PRM_POINT("startup_init() starting..."); 638 639 /* 640 * Complete the extraction of cpuid data 641 */ 642 cpuid_pass2(CPU); 643 644 (void) check_boot_version(BOP_GETVERSION(bootops)); 645 646 /* 647 * Check for prom_debug in boot environment 648 */ 649 if (BOP_GETPROPLEN(bootops, "prom_debug") >= 0) { 650 ++prom_debug; 651 PRM_POINT("prom_debug found in boot enviroment"); 652 } 653 654 /* 655 * Collect node, cpu and memory configuration information. 656 */ 657 get_system_configuration(); 658 659 /* 660 * Halt if this is an unsupported processor. 661 */ 662 if (x86_type == X86_TYPE_486 || x86_type == X86_TYPE_CYRIX_486) { 663 printf("\n486 processor (\"%s\") detected.\n", 664 CPU->cpu_brandstr); 665 halt("This processor is not supported by this release " 666 "of Solaris."); 667 } 668 669 PRM_POINT("startup_init() done"); 670 } 671 672 /* 673 * Callback for copy_memlist_filter() to filter nucleus, kadb/kmdb, (ie. 674 * everything mapped above KERNEL_TEXT) pages from phys_avail. Note it 675 * also filters out physical page zero. There is some reliance on the 676 * boot loader allocating only a few contiguous physical memory chunks. 677 */ 678 static void 679 avail_filter(uint64_t *addr, uint64_t *size) 680 { 681 uintptr_t va; 682 uintptr_t next_va; 683 pfn_t pfn; 684 uint64_t pfn_addr; 685 uint64_t pfn_eaddr; 686 uint_t prot; 687 size_t len; 688 uint_t change; 689 690 if (prom_debug) 691 prom_printf("\tFilter: in: a=%" PRIx64 ", s=%" PRIx64 "\n", 692 *addr, *size); 693 694 /* 695 * page zero is required for BIOS.. never make it available 696 */ 697 if (*addr == 0) { 698 *addr += MMU_PAGESIZE; 699 *size -= MMU_PAGESIZE; 700 } 701 702 /* 703 * First we trim from the front of the range. Since kbm_probe() 704 * walks ranges in virtual order, but addr/size are physical, we need 705 * to the list until no changes are seen. This deals with the case 706 * where page "p" is mapped at v, page "p + PAGESIZE" is mapped at w 707 * but w < v. 708 */ 709 do { 710 change = 0; 711 for (va = KERNEL_TEXT; 712 *size > 0 && kbm_probe(&va, &len, &pfn, &prot) != 0; 713 va = next_va) { 714 715 next_va = va + len; 716 pfn_addr = pfn_to_pa(pfn); 717 pfn_eaddr = pfn_addr + len; 718 719 if (pfn_addr <= *addr && pfn_eaddr > *addr) { 720 change = 1; 721 while (*size > 0 && len > 0) { 722 *addr += MMU_PAGESIZE; 723 *size -= MMU_PAGESIZE; 724 len -= MMU_PAGESIZE; 725 } 726 } 727 } 728 if (change && prom_debug) 729 prom_printf("\t\ttrim: a=%" PRIx64 ", s=%" PRIx64 "\n", 730 *addr, *size); 731 } while (change); 732 733 /* 734 * Trim pages from the end of the range. 735 */ 736 for (va = KERNEL_TEXT; 737 *size > 0 && kbm_probe(&va, &len, &pfn, &prot) != 0; 738 va = next_va) { 739 740 next_va = va + len; 741 pfn_addr = pfn_to_pa(pfn); 742 743 if (pfn_addr >= *addr && pfn_addr < *addr + *size) 744 *size = pfn_addr - *addr; 745 } 746 747 if (prom_debug) 748 prom_printf("\tFilter out: a=%" PRIx64 ", s=%" PRIx64 "\n", 749 *addr, *size); 750 } 751 752 static void 753 kpm_init() 754 { 755 struct segkpm_crargs b; 756 757 /* 758 * These variables were all designed for sfmmu in which segkpm is 759 * mapped using a single pagesize - either 8KB or 4MB. On x86, we 760 * might use 2+ page sizes on a single machine, so none of these 761 * variables have a single correct value. They are set up as if we 762 * always use a 4KB pagesize, which should do no harm. In the long 763 * run, we should get rid of KPM's assumption that only a single 764 * pagesize is used. 765 */ 766 kpm_pgshft = MMU_PAGESHIFT; 767 kpm_pgsz = MMU_PAGESIZE; 768 kpm_pgoff = MMU_PAGEOFFSET; 769 kpmp2pshft = 0; 770 kpmpnpgs = 1; 771 ASSERT(((uintptr_t)kpm_vbase & (kpm_pgsz - 1)) == 0); 772 773 PRM_POINT("about to create segkpm"); 774 rw_enter(&kas.a_lock, RW_WRITER); 775 776 if (seg_attach(&kas, kpm_vbase, kpm_size, segkpm) < 0) 777 panic("cannot attach segkpm"); 778 779 b.prot = PROT_READ | PROT_WRITE; 780 b.nvcolors = 1; 781 782 if (segkpm_create(segkpm, (caddr_t)&b) != 0) 783 panic("segkpm_create segkpm"); 784 785 rw_exit(&kas.a_lock); 786 } 787 788 /* 789 * The debug info page provides enough information to allow external 790 * inspectors (e.g. when running under a hypervisor) to bootstrap 791 * themselves into allowing full-blown kernel debugging. 792 */ 793 static void 794 init_debug_info(void) 795 { 796 caddr_t mem; 797 debug_info_t *di; 798 799 #ifndef __lint 800 ASSERT(sizeof (debug_info_t) < MMU_PAGESIZE); 801 #endif 802 803 mem = BOP_ALLOC(bootops, (caddr_t)DEBUG_INFO_VA, MMU_PAGESIZE, 804 MMU_PAGESIZE); 805 806 if (mem != (caddr_t)DEBUG_INFO_VA) 807 panic("BOP_ALLOC() failed"); 808 bzero(mem, MMU_PAGESIZE); 809 810 di = (debug_info_t *)mem; 811 812 di->di_magic = DEBUG_INFO_MAGIC; 813 di->di_version = DEBUG_INFO_VERSION; 814 } 815 816 /* 817 * Build the memlists and other kernel essential memory system data structures. 818 * This is everything at valloc_base. 819 */ 820 static void 821 startup_memlist(void) 822 { 823 size_t memlist_sz; 824 size_t memseg_sz; 825 size_t pagehash_sz; 826 size_t pp_sz; 827 uintptr_t va; 828 size_t len; 829 uint_t prot; 830 pfn_t pfn; 831 int memblocks; 832 caddr_t pagecolor_mem; 833 size_t pagecolor_memsz; 834 caddr_t page_ctrs_mem; 835 size_t page_ctrs_size; 836 struct memlist *current; 837 extern void startup_build_mem_nodes(struct memlist *); 838 839 /* XX64 fix these - they should be in include files */ 840 extern size_t page_coloring_init(uint_t, int, int); 841 extern void page_coloring_setup(caddr_t); 842 843 PRM_POINT("startup_memlist() starting..."); 844 845 /* 846 * Use leftover large page nucleus text/data space for loadable modules. 847 * Use at most MODTEXT/MODDATA. 848 */ 849 len = kbm_nucleus_size; 850 ASSERT(len > MMU_PAGESIZE); 851 852 moddata = (caddr_t)ROUND_UP_PAGE(e_data); 853 e_moddata = (caddr_t)P2ROUNDUP((uintptr_t)e_data, (uintptr_t)len); 854 if (e_moddata - moddata > MODDATA) 855 e_moddata = moddata + MODDATA; 856 857 modtext = (caddr_t)ROUND_UP_PAGE(e_text); 858 e_modtext = (caddr_t)P2ROUNDUP((uintptr_t)e_text, (uintptr_t)len); 859 if (e_modtext - modtext > MODTEXT) 860 e_modtext = modtext + MODTEXT; 861 862 econtig = e_moddata; 863 864 PRM_DEBUG(modtext); 865 PRM_DEBUG(e_modtext); 866 PRM_DEBUG(moddata); 867 PRM_DEBUG(e_moddata); 868 PRM_DEBUG(econtig); 869 870 /* 871 * Examine the boot loader physical memory map to find out: 872 * - total memory in system - physinstalled 873 * - the max physical address - physmax 874 * - the number of discontiguous segments of memory. 875 */ 876 if (prom_debug) 877 print_memlist("boot physinstalled", 878 bootops->boot_mem->physinstalled); 879 installed_top_size(bootops->boot_mem->physinstalled, &physmax, 880 &physinstalled, &memblocks); 881 PRM_DEBUG(physmax); 882 PRM_DEBUG(physinstalled); 883 PRM_DEBUG(memblocks); 884 885 /* 886 * Initialize hat's mmu parameters. 887 * Check for enforce-prot-exec in boot environment. It's used to 888 * enable/disable support for the page table entry NX bit. 889 * The default is to enforce PROT_EXEC on processors that support NX. 890 * Boot seems to round up the "len", but 8 seems to be big enough. 891 */ 892 mmu_init(); 893 894 #ifdef __i386 895 /* 896 * physmax is lowered if there is more memory than can be 897 * physically addressed in 32 bit (PAE/non-PAE) modes. 898 */ 899 if (mmu.pae_hat) { 900 if (PFN_ABOVE64G(physmax)) { 901 physinstalled -= (physmax - (PFN_64G - 1)); 902 physmax = PFN_64G - 1; 903 } 904 } else { 905 if (PFN_ABOVE4G(physmax)) { 906 physinstalled -= (physmax - (PFN_4G - 1)); 907 physmax = PFN_4G - 1; 908 } 909 } 910 #endif 911 912 startup_build_mem_nodes(bootops->boot_mem->physinstalled); 913 914 if (BOP_GETPROPLEN(bootops, "enforce-prot-exec") >= 0) { 915 int len = BOP_GETPROPLEN(bootops, "enforce-prot-exec"); 916 char value[8]; 917 918 if (len < 8) 919 (void) BOP_GETPROP(bootops, "enforce-prot-exec", value); 920 else 921 (void) strcpy(value, ""); 922 if (strcmp(value, "off") == 0) 923 mmu.pt_nx = 0; 924 } 925 PRM_DEBUG(mmu.pt_nx); 926 927 /* 928 * We will need page_t's for every page in the system, except for 929 * memory mapped at or above above the start of the kernel text segment. 930 * 931 * pages above e_modtext are attributed to kernel debugger (obp_pages) 932 */ 933 npages = physinstalled - 1; /* avail_filter() skips page 0, so "- 1" */ 934 obp_pages = 0; 935 va = KERNEL_TEXT; 936 while (kbm_probe(&va, &len, &pfn, &prot) != 0) { 937 npages -= len >> MMU_PAGESHIFT; 938 if (va >= (uintptr_t)e_moddata) 939 obp_pages += len >> MMU_PAGESHIFT; 940 va += len; 941 } 942 PRM_DEBUG(npages); 943 PRM_DEBUG(obp_pages); 944 945 /* 946 * If physmem is patched to be non-zero, use it instead of 947 * the computed value unless it is larger than the real 948 * amount of memory on hand. 949 */ 950 if (physmem == 0 || physmem > npages) { 951 physmem = npages; 952 } else if (physmem < npages) { 953 orig_npages = npages; 954 npages = physmem; 955 } 956 PRM_DEBUG(physmem); 957 958 /* 959 * We now compute the sizes of all the initial allocations for 960 * structures the kernel needs in order do kmem_alloc(). These 961 * include: 962 * memsegs 963 * memlists 964 * page hash table 965 * page_t's 966 * page coloring data structs 967 */ 968 memseg_sz = sizeof (struct memseg) * (memblocks + POSS_NEW_FRAGMENTS); 969 ADD_TO_ALLOCATIONS(memseg_base, memseg_sz); 970 PRM_DEBUG(memseg_sz); 971 972 /* 973 * Reserve space for memlists. There's no real good way to know exactly 974 * how much room we'll need, but this should be a good upper bound. 975 */ 976 memlist_sz = ROUND_UP_PAGE(2 * sizeof (struct memlist) * 977 (memblocks + POSS_NEW_FRAGMENTS)); 978 ADD_TO_ALLOCATIONS(memlist, memlist_sz); 979 PRM_DEBUG(memlist_sz); 980 981 /* 982 * The page structure hash table size is a power of 2 983 * such that the average hash chain length is PAGE_HASHAVELEN. 984 */ 985 page_hashsz = npages / PAGE_HASHAVELEN; 986 page_hashsz = 1 << highbit(page_hashsz); 987 pagehash_sz = sizeof (struct page *) * page_hashsz; 988 ADD_TO_ALLOCATIONS(page_hash, pagehash_sz); 989 PRM_DEBUG(pagehash_sz); 990 991 /* 992 * Set aside room for the page structures themselves. 993 */ 994 PRM_DEBUG(npages); 995 pp_sz = sizeof (struct page) * npages; 996 ADD_TO_ALLOCATIONS(pp_base, pp_sz); 997 PRM_DEBUG(pp_sz); 998 999 /* 1000 * determine l2 cache info and memory size for page coloring 1001 */ 1002 (void) getl2cacheinfo(CPU, 1003 &l2cache_sz, &l2cache_linesz, &l2cache_assoc); 1004 pagecolor_memsz = 1005 page_coloring_init(l2cache_sz, l2cache_linesz, l2cache_assoc); 1006 ADD_TO_ALLOCATIONS(pagecolor_mem, pagecolor_memsz); 1007 PRM_DEBUG(pagecolor_memsz); 1008 1009 page_ctrs_size = page_ctrs_sz(); 1010 ADD_TO_ALLOCATIONS(page_ctrs_mem, page_ctrs_size); 1011 PRM_DEBUG(page_ctrs_size); 1012 1013 #if defined(__amd64) 1014 valloc_sz = ROUND_UP_LPAGE(valloc_sz); 1015 valloc_base = VALLOC_BASE; 1016 #else /* __i386 */ 1017 valloc_base = (uintptr_t)(MISC_VA_BASE - valloc_sz); 1018 valloc_base = P2ALIGN(valloc_base, mmu.level_size[1]); 1019 #endif /* __i386 */ 1020 PRM_DEBUG(valloc_base); 1021 1022 /* 1023 * do all the initial allocations 1024 */ 1025 perform_allocations(); 1026 1027 /* 1028 * Build phys_install and phys_avail in kernel memspace. 1029 * - phys_install should be all memory in the system. 1030 * - phys_avail is phys_install minus any memory mapped before this 1031 * point above KERNEL_TEXT. 1032 */ 1033 current = phys_install = memlist; 1034 copy_memlist_filter(bootops->boot_mem->physinstalled, ¤t, NULL); 1035 if ((caddr_t)current > (caddr_t)memlist + memlist_sz) 1036 panic("physinstalled was too big!"); 1037 if (prom_debug) 1038 print_memlist("phys_install", phys_install); 1039 1040 phys_avail = current; 1041 PRM_POINT("Building phys_avail:\n"); 1042 copy_memlist_filter(bootops->boot_mem->physinstalled, ¤t, 1043 avail_filter); 1044 if ((caddr_t)current > (caddr_t)memlist + memlist_sz) 1045 panic("physavail was too big!"); 1046 if (prom_debug) 1047 print_memlist("phys_avail", phys_avail); 1048 1049 /* 1050 * setup page coloring 1051 */ 1052 page_coloring_setup(pagecolor_mem); 1053 page_lock_init(); /* currently a no-op */ 1054 1055 /* 1056 * free page list counters 1057 */ 1058 (void) page_ctrs_alloc(page_ctrs_mem); 1059 1060 /* 1061 * Initialize the page structures from the memory lists. 1062 */ 1063 availrmem_initial = availrmem = freemem = 0; 1064 PRM_POINT("Calling kphysm_init()..."); 1065 npages = kphysm_init(pp_base, npages); 1066 PRM_POINT("kphysm_init() done"); 1067 PRM_DEBUG(npages); 1068 1069 init_debug_info(); 1070 1071 /* 1072 * Now that page_t's have been initialized, remove all the 1073 * initial allocation pages from the kernel free page lists. 1074 */ 1075 boot_mapin((caddr_t)valloc_base, valloc_sz); 1076 boot_mapin((caddr_t)GDT_VA, MMU_PAGESIZE); 1077 boot_mapin((caddr_t)DEBUG_INFO_VA, MMU_PAGESIZE); 1078 PRM_POINT("startup_memlist() done"); 1079 1080 PRM_DEBUG(valloc_sz); 1081 1082 #if defined(__amd64) 1083 if ((availrmem >> (30 - MMU_PAGESHIFT)) >= 1084 textrepl_min_gb && l2cache_sz <= 2 << 20) { 1085 extern size_t textrepl_size_thresh; 1086 textrepl_size_thresh = (16 << 20) - 1; 1087 } 1088 #endif 1089 } 1090 1091 /* 1092 * Layout the kernel's part of address space and initialize kmem allocator. 1093 */ 1094 static void 1095 startup_kmem(void) 1096 { 1097 extern void page_set_colorequiv_arr(void); 1098 1099 PRM_POINT("startup_kmem() starting..."); 1100 1101 #if defined(__amd64) 1102 if (eprom_kernelbase && eprom_kernelbase != KERNELBASE) 1103 cmn_err(CE_NOTE, "!kernelbase cannot be changed on 64-bit " 1104 "systems."); 1105 kernelbase = (uintptr_t)KERNELBASE; 1106 core_base = (uintptr_t)COREHEAP_BASE; 1107 core_size = (size_t)MISC_VA_BASE - COREHEAP_BASE; 1108 #else /* __i386 */ 1109 /* 1110 * We configure kernelbase based on: 1111 * 1112 * 1. user specified kernelbase via eeprom command. Value cannot exceed 1113 * KERNELBASE_MAX. we large page align eprom_kernelbase 1114 * 1115 * 2. Default to KERNELBASE and adjust to 2X less the size for page_t. 1116 * On large memory systems we must lower kernelbase to allow 1117 * enough room for page_t's for all of memory. 1118 * 1119 * The value set here, might be changed a little later. 1120 */ 1121 if (eprom_kernelbase) { 1122 kernelbase = eprom_kernelbase & mmu.level_mask[1]; 1123 if (kernelbase > KERNELBASE_MAX) 1124 kernelbase = KERNELBASE_MAX; 1125 } else { 1126 kernelbase = (uintptr_t)KERNELBASE; 1127 kernelbase -= ROUND_UP_4MEG(2 * valloc_sz); 1128 } 1129 ASSERT((kernelbase & mmu.level_offset[1]) == 0); 1130 core_base = valloc_base; 1131 core_size = 0; 1132 #endif /* __i386 */ 1133 1134 PRM_DEBUG(core_base); 1135 PRM_DEBUG(core_size); 1136 PRM_DEBUG(kernelbase); 1137 1138 /* 1139 * At this point, we can only use a portion of the kernelheap that 1140 * will be available after we boot. 32-bit systems have this 1141 * limitation. 1142 * 1143 * On 32-bit systems we have to leave room to place segmap below 1144 * the heap. We don't yet know how large segmap will be, so we 1145 * have to be very conservative. 1146 * 1147 * On 64 bit systems there should be LOTS of room so just use 1148 * the next 4Gig below core_base. 1149 */ 1150 #if defined(__amd64) 1151 1152 boot_kernelheap = (caddr_t)core_base - FOURGB; 1153 segmap_reserved = 0; 1154 1155 #else /* __i386 */ 1156 1157 segkp_fromheap = 1; 1158 segmap_reserved = ROUND_UP_LPAGE(MAX(segmapsize, SEGMAPMAX)); 1159 boot_kernelheap = 1160 (caddr_t)ROUND_UP_LPAGE(kernelbase) + segmap_reserved; 1161 1162 #endif /* __i386 */ 1163 PRM_DEBUG(boot_kernelheap); 1164 ekernelheap = (char *)core_base; 1165 PRM_DEBUG(ekernelheap); 1166 kernelheap = boot_kernelheap; 1167 1168 /* 1169 * If segmap is too large we can push the bottom of the kernel heap 1170 * higher than the base. Or worse, it could exceed the top of the 1171 * VA space entirely, causing it to wrap around. 1172 */ 1173 if (kernelheap >= ekernelheap || (uintptr_t)kernelheap < kernelbase) 1174 panic("too little memory available for kernelheap," 1175 " use a different kernelbase"); 1176 1177 /* 1178 * Now that we know the real value of kernelbase, 1179 * update variables that were initialized with a value of 1180 * KERNELBASE (in common/conf/param.c). 1181 * 1182 * XXX The problem with this sort of hackery is that the 1183 * compiler just may feel like putting the const declarations 1184 * (in param.c) into the .text section. Perhaps they should 1185 * just be declared as variables there? 1186 */ 1187 1188 #if defined(__amd64) 1189 ASSERT(_kernelbase == KERNELBASE); 1190 ASSERT(_userlimit == USERLIMIT); 1191 #else 1192 *(uintptr_t *)&_kernelbase = kernelbase; 1193 *(uintptr_t *)&_userlimit = kernelbase; 1194 *(uintptr_t *)&_userlimit32 = _userlimit; 1195 #endif 1196 PRM_DEBUG(_kernelbase); 1197 PRM_DEBUG(_userlimit); 1198 PRM_DEBUG(_userlimit32); 1199 1200 /* 1201 * Initialize the kernel heap. Note 3rd argument must be > 1st. 1202 */ 1203 kernelheap_init(boot_kernelheap, ekernelheap, 1204 boot_kernelheap + MMU_PAGESIZE, 1205 (void *)core_base, (void *)(core_base + core_size)); 1206 1207 /* 1208 * Initialize kernel memory allocator. 1209 */ 1210 kmem_init(); 1211 1212 /* 1213 * Factor in colorequiv to check additional 'equivalent' bins 1214 */ 1215 page_set_colorequiv_arr(); 1216 1217 /* 1218 * print this out early so that we know what's going on 1219 */ 1220 cmn_err(CE_CONT, "?features: %b\n", x86_feature, FMT_X86_FEATURE); 1221 1222 /* 1223 * Initialize bp_mapin(). 1224 */ 1225 bp_init(MMU_PAGESIZE, HAT_STORECACHING_OK); 1226 1227 /* 1228 * orig_npages is non-zero if physmem has been configured for less 1229 * than the available memory. 1230 */ 1231 if (orig_npages) { 1232 #ifdef __i386 1233 /* 1234 * use npages for physmem in case it has been temporarily 1235 * modified via /etc/system in kmem_init/mod_read_system_file. 1236 */ 1237 if (npages == PHYSMEM32) { 1238 cmn_err(CE_WARN, "!Due to 32-bit virtual" 1239 " address space limitations, limiting" 1240 " physmem to 0x%lx of 0x%lx available pages", 1241 npages, orig_npages); 1242 } else { 1243 cmn_err(CE_WARN, "!limiting physmem to 0x%lx of" 1244 " 0x%lx available pages", npages, orig_npages); 1245 } 1246 #else 1247 cmn_err(CE_WARN, "!limiting physmem to 0x%lx of" 1248 " 0x%lx available pages", npages, orig_npages); 1249 #endif 1250 } 1251 #if defined(__i386) 1252 if (eprom_kernelbase && (eprom_kernelbase != kernelbase)) 1253 cmn_err(CE_WARN, "kernelbase value, User specified 0x%lx, " 1254 "System using 0x%lx", 1255 (uintptr_t)eprom_kernelbase, (uintptr_t)kernelbase); 1256 #endif 1257 1258 #ifdef KERNELBASE_ABI_MIN 1259 if (kernelbase < (uintptr_t)KERNELBASE_ABI_MIN) { 1260 cmn_err(CE_NOTE, "!kernelbase set to 0x%lx, system is not " 1261 "i386 ABI compliant.", (uintptr_t)kernelbase); 1262 } 1263 #endif 1264 1265 PRM_POINT("startup_kmem() done"); 1266 } 1267 1268 static void 1269 startup_modules(void) 1270 { 1271 unsigned int i; 1272 extern void prom_setup(void); 1273 1274 PRM_POINT("startup_modules() starting..."); 1275 /* 1276 * Initialize ten-micro second timer so that drivers will 1277 * not get short changed in their init phase. This was 1278 * not getting called until clkinit which, on fast cpu's 1279 * caused the drv_usecwait to be way too short. 1280 */ 1281 microfind(); 1282 1283 /* 1284 * Read the GMT lag from /etc/rtc_config. 1285 */ 1286 sgmtl(process_rtc_config_file()); 1287 1288 /* 1289 * Calculate default settings of system parameters based upon 1290 * maxusers, yet allow to be overridden via the /etc/system file. 1291 */ 1292 param_calc(0); 1293 1294 mod_setup(); 1295 1296 /* 1297 * Initialize system parameters. 1298 */ 1299 param_init(); 1300 1301 /* 1302 * Initialize the default brands 1303 */ 1304 brand_init(); 1305 1306 /* 1307 * maxmem is the amount of physical memory we're playing with. 1308 */ 1309 maxmem = physmem; 1310 1311 /* 1312 * Initialize the hat layer. 1313 */ 1314 hat_init(); 1315 1316 /* 1317 * Initialize segment management stuff. 1318 */ 1319 seg_init(); 1320 1321 if (modload("fs", "specfs") == -1) 1322 halt("Can't load specfs"); 1323 1324 if (modload("fs", "devfs") == -1) 1325 halt("Can't load devfs"); 1326 1327 if (modload("fs", "dev") == -1) 1328 halt("Can't load dev"); 1329 1330 (void) modloadonly("sys", "lbl_edition"); 1331 1332 dispinit(); 1333 1334 /* 1335 * This is needed here to initialize hw_serial[] for cluster booting. 1336 */ 1337 if ((i = modload("misc", "sysinit")) != (unsigned int)-1) 1338 (void) modunload(i); 1339 else 1340 cmn_err(CE_CONT, "sysinit load failed"); 1341 1342 /* Read cluster configuration data. */ 1343 clconf_init(); 1344 1345 /* 1346 * Create a kernel device tree. First, create rootnex and 1347 * then invoke bus specific code to probe devices. 1348 */ 1349 setup_ddi(); 1350 1351 /* 1352 * Set up the CPU module subsystem. Modifies the device tree, so it 1353 * must be done after setup_ddi(). 1354 */ 1355 cmi_init(); 1356 1357 /* 1358 * Initialize the MCA handlers 1359 */ 1360 if (x86_feature & X86_MCA) 1361 cmi_mca_init(); 1362 1363 /* 1364 * Fake a prom tree such that /dev/openprom continues to work 1365 */ 1366 PRM_POINT("startup_modules: calling prom_setup..."); 1367 prom_setup(); 1368 PRM_POINT("startup_modules: done"); 1369 1370 /* 1371 * Load all platform specific modules 1372 */ 1373 PRM_POINT("startup_modules: calling psm_modload..."); 1374 psm_modload(); 1375 1376 PRM_POINT("startup_modules() done"); 1377 } 1378 1379 /* 1380 * claim a "setaside" boot page for use in the kernel 1381 */ 1382 page_t * 1383 boot_claim_page(pfn_t pfn) 1384 { 1385 page_t *pp; 1386 1387 pp = page_numtopp_nolock(pfn); 1388 ASSERT(pp != NULL); 1389 1390 if (PP_ISBOOTPAGES(pp)) { 1391 if (pp->p_next != NULL) 1392 pp->p_next->p_prev = pp->p_prev; 1393 if (pp->p_prev == NULL) 1394 bootpages = pp->p_next; 1395 else 1396 pp->p_prev->p_next = pp->p_next; 1397 } else { 1398 /* 1399 * htable_attach() expects a base pagesize page 1400 */ 1401 if (pp->p_szc != 0) 1402 page_boot_demote(pp); 1403 pp = page_numtopp(pfn, SE_EXCL); 1404 } 1405 return (pp); 1406 } 1407 1408 /* 1409 * Walk through the pagetables looking for pages mapped in by boot. If the 1410 * setaside flag is set the pages are expected to be returned to the 1411 * kernel later in boot, so we add them to the bootpages list. 1412 */ 1413 static void 1414 protect_boot_range(uintptr_t low, uintptr_t high, int setaside) 1415 { 1416 uintptr_t va = low; 1417 size_t len; 1418 uint_t prot; 1419 pfn_t pfn; 1420 page_t *pp; 1421 pgcnt_t boot_protect_cnt = 0; 1422 1423 while (kbm_probe(&va, &len, &pfn, &prot) != 0 && va < high) { 1424 if (va + len >= high) 1425 panic("0x%lx byte mapping at 0x%p exceeds boot's " 1426 "legal range.", len, (void *)va); 1427 1428 while (len > 0) { 1429 pp = page_numtopp_alloc(pfn); 1430 if (pp != NULL) { 1431 if (setaside == 0) 1432 panic("Unexpected mapping by boot. " 1433 "addr=%p pfn=%lx\n", 1434 (void *)va, pfn); 1435 1436 pp->p_next = bootpages; 1437 pp->p_prev = NULL; 1438 PP_SETBOOTPAGES(pp); 1439 if (bootpages != NULL) { 1440 bootpages->p_prev = pp; 1441 } 1442 bootpages = pp; 1443 ++boot_protect_cnt; 1444 } 1445 1446 ++pfn; 1447 len -= MMU_PAGESIZE; 1448 va += MMU_PAGESIZE; 1449 } 1450 } 1451 PRM_DEBUG(boot_protect_cnt); 1452 } 1453 1454 /* 1455 * Finish initializing the VM system, now that we are no longer 1456 * relying on the boot time memory allocators. 1457 */ 1458 static void 1459 startup_vm(void) 1460 { 1461 struct segmap_crargs a; 1462 1463 extern int use_brk_lpg, use_stk_lpg; 1464 1465 PRM_POINT("startup_vm() starting..."); 1466 1467 /* 1468 * Establish the final size of the kernel's heap, size of segmap, 1469 * segkp, etc. 1470 */ 1471 1472 #if defined(__amd64) 1473 1474 /* 1475 * Check if there is enough virtual address space in KPM region to 1476 * map physmax. 1477 */ 1478 kpm_vbase = (caddr_t)(uintptr_t)SEGKPM_BASE; 1479 kpm_size = 0; 1480 if (kpm_desired) { 1481 kpm_size = ROUND_UP_LPAGE(mmu_ptob(physmax + 1)); 1482 if ((uintptr_t)kpm_vbase + kpm_size > (uintptr_t)VALLOC_BASE) { 1483 kpm_size = 0; 1484 kpm_desired = 0; 1485 } 1486 } 1487 1488 PRM_DEBUG(kpm_size); 1489 PRM_DEBUG(kpm_vbase); 1490 1491 /* 1492 * By default we create a seg_kp in 64 bit kernels, it's a little 1493 * faster to access than embedding it in the heap. 1494 */ 1495 segkp_base = (caddr_t)valloc_base + valloc_sz; 1496 if (!segkp_fromheap) { 1497 size_t sz = mmu_ptob(segkpsize); 1498 1499 /* 1500 * determine size of segkp 1501 */ 1502 if (sz < SEGKPMINSIZE || sz > SEGKPMAXSIZE) { 1503 sz = SEGKPDEFSIZE; 1504 cmn_err(CE_WARN, "!Illegal value for segkpsize. " 1505 "segkpsize has been reset to %ld pages", 1506 mmu_btop(sz)); 1507 } 1508 sz = MIN(sz, MAX(SEGKPMINSIZE, mmu_ptob(physmem))); 1509 1510 segkpsize = mmu_btop(ROUND_UP_LPAGE(sz)); 1511 } 1512 PRM_DEBUG(segkp_base); 1513 PRM_DEBUG(segkpsize); 1514 1515 segzio_base = segkp_base + mmu_ptob(segkpsize); 1516 if (segzio_fromheap) { 1517 segziosize = 0; 1518 } else { 1519 size_t size; 1520 size_t physmem_b = mmu_ptob(physmem); 1521 1522 /* size is in bytes, segziosize is in pages */ 1523 if (segziosize == 0) { 1524 size = physmem_b; 1525 } else { 1526 size = mmu_ptob(segziosize); 1527 } 1528 1529 if (size < SEGZIOMINSIZE) { 1530 size = SEGZIOMINSIZE; 1531 } else if (size > SEGZIOMAXSIZE) { 1532 size = SEGZIOMAXSIZE; 1533 /* 1534 * SEGZIOMAXSIZE is capped at 512gb so that segzio 1535 * doesn't consume all of KVA. However, if we have a 1536 * system that has more thant 512gb of physical memory, 1537 * we can actually consume about half of the difference 1538 * between 512gb and the rest of the available physical 1539 * memory. 1540 */ 1541 if (physmem_b > SEGZIOMAXSIZE) { 1542 size += (physmem_b - SEGZIOMAXSIZE) / 2; 1543 } 1544 } 1545 segziosize = mmu_btop(ROUND_UP_LPAGE(size)); 1546 } 1547 PRM_DEBUG(segziosize); 1548 PRM_DEBUG(segzio_base); 1549 1550 /* 1551 * Put the range of VA for device mappings next, kmdb knows to not 1552 * grep in this range of addresses. 1553 */ 1554 toxic_addr = 1555 ROUND_UP_LPAGE((uintptr_t)segzio_base + mmu_ptob(segziosize)); 1556 PRM_DEBUG(toxic_addr); 1557 segmap_start = ROUND_UP_LPAGE(toxic_addr + toxic_size); 1558 #else /* __i386 */ 1559 segmap_start = ROUND_UP_LPAGE(kernelbase); 1560 #endif /* __i386 */ 1561 PRM_DEBUG(segmap_start); 1562 ASSERT((caddr_t)segmap_start < boot_kernelheap); 1563 1564 /* 1565 * Users can change segmapsize through eeprom or /etc/system. 1566 * If the variable is tuned through eeprom, there is no upper 1567 * bound on the size of segmap. If it is tuned through 1568 * /etc/system on 32-bit systems, it must be no larger than we 1569 * planned for in startup_memlist(). 1570 */ 1571 segmapsize = MAX(ROUND_UP_LPAGE(segmapsize), SEGMAPDEFAULT); 1572 1573 #if defined(__i386) 1574 if (segmapsize > segmap_reserved) { 1575 cmn_err(CE_NOTE, "!segmapsize may not be set > 0x%lx in " 1576 "/etc/system. Use eeprom.", (long)SEGMAPMAX); 1577 segmapsize = segmap_reserved; 1578 } 1579 /* 1580 * 32-bit systems don't have segkpm or segkp, so segmap appears at 1581 * the bottom of the kernel's address range. Set aside space for a 1582 * small red zone just below the start of segmap. 1583 */ 1584 segmap_start += KERNEL_REDZONE_SIZE; 1585 segmapsize -= KERNEL_REDZONE_SIZE; 1586 #endif 1587 1588 PRM_DEBUG(segmap_start); 1589 PRM_DEBUG(segmapsize); 1590 final_kernelheap = (caddr_t)ROUND_UP_LPAGE(segmap_start + segmapsize); 1591 PRM_DEBUG(final_kernelheap); 1592 1593 /* 1594 * Do final allocations of HAT data structures that need to 1595 * be allocated before quiescing the boot loader. 1596 */ 1597 PRM_POINT("Calling hat_kern_alloc()..."); 1598 hat_kern_alloc((caddr_t)segmap_start, segmapsize, ekernelheap); 1599 PRM_POINT("hat_kern_alloc() done"); 1600 1601 /* 1602 * Setup MTRR (Memory type range registers) 1603 */ 1604 setup_mtrr(); 1605 1606 /* 1607 * The next two loops are done in distinct steps in order 1608 * to be sure that any page that is doubly mapped (both above 1609 * KERNEL_TEXT and below kernelbase) is dealt with correctly. 1610 * Note this may never happen, but it might someday. 1611 */ 1612 bootpages = NULL; 1613 PRM_POINT("Protecting boot pages"); 1614 1615 /* 1616 * Protect any pages mapped above KERNEL_TEXT that somehow have 1617 * page_t's. This can only happen if something weird allocated 1618 * in this range (like kadb/kmdb). 1619 */ 1620 protect_boot_range(KERNEL_TEXT, (uintptr_t)-1, 0); 1621 1622 /* 1623 * Before we can take over memory allocation/mapping from the boot 1624 * loader we must remove from our free page lists any boot allocated 1625 * pages that stay mapped until release_bootstrap(). 1626 */ 1627 protect_boot_range(0, kernelbase, 1); 1628 1629 /* 1630 * Switch to running on regular HAT (not boot_mmu) 1631 */ 1632 PRM_POINT("Calling hat_kern_setup()..."); 1633 hat_kern_setup(); 1634 1635 /* 1636 * It is no longer safe to call BOP_ALLOC(), so make sure we don't. 1637 */ 1638 bop_no_more_mem(); 1639 1640 PRM_POINT("hat_kern_setup() done"); 1641 1642 hat_cpu_online(CPU); 1643 1644 /* 1645 * Initialize VM system 1646 */ 1647 PRM_POINT("Calling kvm_init()..."); 1648 kvm_init(); 1649 PRM_POINT("kvm_init() done"); 1650 1651 /* 1652 * Tell kmdb that the VM system is now working 1653 */ 1654 if (boothowto & RB_DEBUG) 1655 kdi_dvec_vmready(); 1656 1657 /* 1658 * Mangle the brand string etc. 1659 */ 1660 cpuid_pass3(CPU); 1661 1662 /* 1663 * Now that we can use memory outside the top 4GB (on 64-bit 1664 * systems) and we know the size of segmap, we can set the final 1665 * size of the kernel's heap. 1666 */ 1667 if (final_kernelheap < boot_kernelheap) { 1668 PRM_POINT("kernelheap_extend()"); 1669 PRM_DEBUG(boot_kernelheap); 1670 PRM_DEBUG(final_kernelheap); 1671 kernelheap_extend(final_kernelheap, boot_kernelheap); 1672 } 1673 1674 #if defined(__amd64) 1675 1676 /* 1677 * Create the device arena for toxic (to dtrace/kmdb) mappings. 1678 */ 1679 device_arena = vmem_create("device", (void *)toxic_addr, 1680 toxic_size, MMU_PAGESIZE, NULL, NULL, NULL, 0, VM_SLEEP); 1681 1682 #else /* __i386 */ 1683 1684 /* 1685 * allocate the bit map that tracks toxic pages 1686 */ 1687 toxic_bit_map_len = btop((ulong_t)(valloc_base - kernelbase)); 1688 PRM_DEBUG(toxic_bit_map_len); 1689 toxic_bit_map = 1690 kmem_zalloc(BT_SIZEOFMAP(toxic_bit_map_len), KM_NOSLEEP); 1691 ASSERT(toxic_bit_map != NULL); 1692 PRM_DEBUG(toxic_bit_map); 1693 1694 #endif /* __i386 */ 1695 1696 1697 /* 1698 * Now that we've got more VA, as well as the ability to allocate from 1699 * it, tell the debugger. 1700 */ 1701 if (boothowto & RB_DEBUG) 1702 kdi_dvec_memavail(); 1703 1704 /* 1705 * The following code installs a special page fault handler (#pf) 1706 * to work around a pentium bug. 1707 */ 1708 #if !defined(__amd64) 1709 if (x86_type == X86_TYPE_P5) { 1710 desctbr_t idtr; 1711 gate_desc_t *newidt; 1712 struct machcpu *mcpu = &CPU->cpu_m; 1713 1714 if ((newidt = kmem_zalloc(MMU_PAGESIZE, KM_NOSLEEP)) == NULL) 1715 panic("failed to install pentium_pftrap"); 1716 1717 bcopy(idt0, newidt, sizeof (idt0)); 1718 set_gatesegd(&newidt[T_PGFLT], &pentium_pftrap, 1719 KCS_SEL, SDT_SYSIGT, SEL_KPL); 1720 1721 (void) as_setprot(&kas, (caddr_t)newidt, MMU_PAGESIZE, 1722 PROT_READ|PROT_EXEC); 1723 1724 mcpu->mcpu_idt = newidt; 1725 idtr.dtr_base = (uintptr_t)mcpu->mcpu_idt; 1726 idtr.dtr_limit = sizeof (idt0) - 1; 1727 wr_idtr(&idtr); 1728 } 1729 #endif /* !__amd64 */ 1730 1731 /* 1732 * Map page pfn=0 for drivers, such as kd, that need to pick up 1733 * parameters left there by controllers/BIOS. 1734 */ 1735 PRM_POINT("setup up p0_va"); 1736 p0_va = i86devmap(0, 1, PROT_READ); 1737 PRM_DEBUG(p0_va); 1738 1739 cmn_err(CE_CONT, "?mem = %luK (0x%lx)\n", 1740 physinstalled << (MMU_PAGESHIFT - 10), ptob(physinstalled)); 1741 1742 /* 1743 * disable automatic large pages for small memory systems or 1744 * when the disable flag is set. 1745 */ 1746 if (!auto_lpg_disable && mmu.max_page_level > 0) { 1747 max_uheap_lpsize = LEVEL_SIZE(1); 1748 max_ustack_lpsize = LEVEL_SIZE(1); 1749 max_privmap_lpsize = LEVEL_SIZE(1); 1750 max_uidata_lpsize = LEVEL_SIZE(1); 1751 max_utext_lpsize = LEVEL_SIZE(1); 1752 max_shm_lpsize = LEVEL_SIZE(1); 1753 } 1754 if (physmem < privm_lpg_min_physmem || mmu.max_page_level == 0 || 1755 auto_lpg_disable) { 1756 use_brk_lpg = 0; 1757 use_stk_lpg = 0; 1758 } 1759 if (mmu.max_page_level > 0) { 1760 mcntl0_lpsize = LEVEL_SIZE(1); 1761 } 1762 1763 PRM_POINT("Calling hat_init_finish()..."); 1764 hat_init_finish(); 1765 PRM_POINT("hat_init_finish() done"); 1766 1767 /* 1768 * Initialize the segkp segment type. 1769 */ 1770 rw_enter(&kas.a_lock, RW_WRITER); 1771 PRM_POINT("Attaching segkp"); 1772 if (segkp_fromheap) { 1773 segkp->s_as = &kas; 1774 } else if (seg_attach(&kas, (caddr_t)segkp_base, mmu_ptob(segkpsize), 1775 segkp) < 0) { 1776 panic("startup: cannot attach segkp"); 1777 /*NOTREACHED*/ 1778 } 1779 PRM_POINT("Doing segkp_create()"); 1780 if (segkp_create(segkp) != 0) { 1781 panic("startup: segkp_create failed"); 1782 /*NOTREACHED*/ 1783 } 1784 PRM_DEBUG(segkp); 1785 rw_exit(&kas.a_lock); 1786 1787 /* 1788 * kpm segment 1789 */ 1790 segmap_kpm = 0; 1791 if (kpm_desired) { 1792 kpm_init(); 1793 kpm_enable = 1; 1794 vpm_enable = 1; 1795 } 1796 1797 /* 1798 * Now create segmap segment. 1799 */ 1800 rw_enter(&kas.a_lock, RW_WRITER); 1801 if (seg_attach(&kas, (caddr_t)segmap_start, segmapsize, segmap) < 0) { 1802 panic("cannot attach segmap"); 1803 /*NOTREACHED*/ 1804 } 1805 PRM_DEBUG(segmap); 1806 1807 a.prot = PROT_READ | PROT_WRITE; 1808 a.shmsize = 0; 1809 a.nfreelist = segmapfreelists; 1810 1811 if (segmap_create(segmap, (caddr_t)&a) != 0) 1812 panic("segmap_create segmap"); 1813 rw_exit(&kas.a_lock); 1814 1815 setup_vaddr_for_ppcopy(CPU); 1816 1817 segdev_init(); 1818 pmem_init(); 1819 1820 PRM_POINT("startup_vm() done"); 1821 } 1822 1823 /* 1824 * Load a tod module for the non-standard tod part found on this system. 1825 */ 1826 static void 1827 load_tod_module(char *todmod) 1828 { 1829 if (modload("tod", todmod) == -1) 1830 halt("Can't load TOD module"); 1831 } 1832 1833 static void 1834 startup_end(void) 1835 { 1836 extern void setx86isalist(void); 1837 1838 PRM_POINT("startup_end() starting..."); 1839 1840 /* 1841 * Perform tasks that get done after most of the VM 1842 * initialization has been done but before the clock 1843 * and other devices get started. 1844 */ 1845 kern_setup1(); 1846 1847 /* 1848 * Perform CPC initialization for this CPU. 1849 */ 1850 kcpc_hw_init(CPU); 1851 1852 #if defined(OPTERON_WORKAROUND_6323525) 1853 if (opteron_workaround_6323525) 1854 patch_workaround_6323525(); 1855 #endif 1856 /* 1857 * If needed, load TOD module now so that ddi_get_time(9F) etc. work 1858 * (For now, "needed" is defined as set tod_module_name in /etc/system) 1859 */ 1860 if (tod_module_name != NULL) { 1861 PRM_POINT("load_tod_module()"); 1862 load_tod_module(tod_module_name); 1863 } 1864 1865 /* 1866 * Configure the system. 1867 */ 1868 PRM_POINT("Calling configure()..."); 1869 configure(); /* set up devices */ 1870 PRM_POINT("configure() done"); 1871 1872 /* 1873 * Set the isa_list string to the defined instruction sets we 1874 * support. 1875 */ 1876 setx86isalist(); 1877 cpu_intr_alloc(CPU, NINTR_THREADS); 1878 psm_install(); 1879 1880 /* 1881 * We're done with bootops. We don't unmap the bootstrap yet because 1882 * we're still using bootsvcs. 1883 */ 1884 PRM_POINT("NULLing out bootops"); 1885 *bootopsp = (struct bootops *)NULL; 1886 bootops = (struct bootops *)NULL; 1887 1888 PRM_POINT("Enabling interrupts"); 1889 (*picinitf)(); 1890 sti(); 1891 1892 (void) add_avsoftintr((void *)&softlevel1_hdl, 1, softlevel1, 1893 "softlevel1", NULL, NULL); /* XXX to be moved later */ 1894 1895 PRM_POINT("startup_end() done"); 1896 } 1897 1898 extern char hw_serial[]; 1899 char *_hs1107 = hw_serial; 1900 ulong_t _bdhs34; 1901 1902 void 1903 post_startup(void) 1904 { 1905 /* 1906 * Set the system wide, processor-specific flags to be passed 1907 * to userland via the aux vector for performance hints and 1908 * instruction set extensions. 1909 */ 1910 bind_hwcap(); 1911 1912 /* 1913 * Load the System Management BIOS into the global ksmbios 1914 * handle, if an SMBIOS is present on this system. 1915 */ 1916 ksmbios = smbios_open(NULL, SMB_VERSION, ksmbios_flags, NULL); 1917 1918 /* 1919 * Startup the memory scrubber. 1920 */ 1921 memscrub_init(); 1922 1923 /* 1924 * Complete CPU module initialization 1925 */ 1926 cmi_post_init(); 1927 1928 /* 1929 * Perform forceloading tasks for /etc/system. 1930 */ 1931 (void) mod_sysctl(SYS_FORCELOAD, NULL); 1932 1933 /* 1934 * ON4.0: Force /proc module in until clock interrupt handle fixed 1935 * ON4.0: This must be fixed or restated in /etc/systems. 1936 */ 1937 (void) modload("fs", "procfs"); 1938 1939 #if defined(__i386) 1940 /* 1941 * Check for required functional Floating Point hardware, 1942 * unless FP hardware explicitly disabled. 1943 */ 1944 if (fpu_exists && (fpu_pentium_fdivbug || fp_kind == FP_NO)) 1945 halt("No working FP hardware found"); 1946 #endif 1947 1948 maxmem = freemem; 1949 1950 add_cpunode2devtree(CPU->cpu_id, CPU->cpu_m.mcpu_cpi); 1951 } 1952 1953 static int 1954 pp_in_ramdisk(page_t *pp) 1955 { 1956 extern uint64_t ramdisk_start, ramdisk_end; 1957 1958 return ((pp->p_pagenum >= btop(ramdisk_start)) && 1959 (pp->p_pagenum < btopr(ramdisk_end))); 1960 } 1961 1962 void 1963 release_bootstrap(void) 1964 { 1965 int root_is_ramdisk; 1966 page_t *pp; 1967 extern void kobj_boot_unmountroot(void); 1968 extern dev_t rootdev; 1969 1970 /* unmount boot ramdisk and release kmem usage */ 1971 kobj_boot_unmountroot(); 1972 1973 /* 1974 * We're finished using the boot loader so free its pages. 1975 */ 1976 PRM_POINT("Unmapping lower boot pages"); 1977 clear_boot_mappings(0, _userlimit); 1978 postbootkernelbase = kernelbase; 1979 1980 /* 1981 * If root isn't on ramdisk, destroy the hardcoded 1982 * ramdisk node now and release the memory. Else, 1983 * ramdisk memory is kept in rd_pages. 1984 */ 1985 root_is_ramdisk = (getmajor(rootdev) == ddi_name_to_major("ramdisk")); 1986 if (!root_is_ramdisk) { 1987 dev_info_t *dip = ddi_find_devinfo("ramdisk", -1, 0); 1988 ASSERT(dip && ddi_get_parent(dip) == ddi_root_node()); 1989 ndi_rele_devi(dip); /* held from ddi_find_devinfo */ 1990 (void) ddi_remove_child(dip, 0); 1991 } 1992 1993 PRM_POINT("Releasing boot pages"); 1994 while (bootpages) { 1995 pp = bootpages; 1996 bootpages = pp->p_next; 1997 if (root_is_ramdisk && pp_in_ramdisk(pp)) { 1998 pp->p_next = rd_pages; 1999 rd_pages = pp; 2000 continue; 2001 } 2002 pp->p_next = (struct page *)0; 2003 pp->p_prev = (struct page *)0; 2004 PP_CLRBOOTPAGES(pp); 2005 page_free(pp, 1); 2006 } 2007 PRM_POINT("Boot pages released"); 2008 2009 /* 2010 * Find 1 page below 1 MB so that other processors can boot up. 2011 * Make sure it has a kernel VA as well as a 1:1 mapping. 2012 * We should have just free'd one up. 2013 */ 2014 if (use_mp) { 2015 pfn_t pfn; 2016 2017 for (pfn = 1; pfn < btop(1*1024*1024); pfn++) { 2018 if (page_numtopp_alloc(pfn) == NULL) 2019 continue; 2020 rm_platter_va = i86devmap(pfn, 1, 2021 PROT_READ | PROT_WRITE | PROT_EXEC); 2022 rm_platter_pa = ptob(pfn); 2023 hat_devload(kas.a_hat, 2024 (caddr_t)(uintptr_t)rm_platter_pa, MMU_PAGESIZE, 2025 pfn, PROT_READ | PROT_WRITE | PROT_EXEC, 2026 HAT_LOAD_NOCONSIST); 2027 break; 2028 } 2029 if (pfn == btop(1*1024*1024)) 2030 panic("No page available for starting " 2031 "other processors"); 2032 } 2033 2034 } 2035 2036 /* 2037 * Initialize the platform-specific parts of a page_t. 2038 */ 2039 void 2040 add_physmem_cb(page_t *pp, pfn_t pnum) 2041 { 2042 pp->p_pagenum = pnum; 2043 pp->p_mapping = NULL; 2044 pp->p_embed = 0; 2045 pp->p_share = 0; 2046 pp->p_mlentry = 0; 2047 } 2048 2049 /* 2050 * kphysm_init() initializes physical memory. 2051 */ 2052 static pgcnt_t 2053 kphysm_init( 2054 page_t *pp, 2055 pgcnt_t npages) 2056 { 2057 struct memlist *pmem; 2058 struct memseg *cur_memseg; 2059 pfn_t base_pfn; 2060 pgcnt_t num; 2061 pgcnt_t pages_done = 0; 2062 uint64_t addr; 2063 uint64_t size; 2064 extern pfn_t ddiphysmin; 2065 2066 ASSERT(page_hash != NULL && page_hashsz != 0); 2067 2068 cur_memseg = memseg_base; 2069 for (pmem = phys_avail; pmem && npages; pmem = pmem->next) { 2070 /* 2071 * In a 32 bit kernel can't use higher memory if we're 2072 * not booting in PAE mode. This check takes care of that. 2073 */ 2074 addr = pmem->address; 2075 size = pmem->size; 2076 if (btop(addr) > physmax) 2077 continue; 2078 2079 /* 2080 * align addr and size - they may not be at page boundaries 2081 */ 2082 if ((addr & MMU_PAGEOFFSET) != 0) { 2083 addr += MMU_PAGEOFFSET; 2084 addr &= ~(uint64_t)MMU_PAGEOFFSET; 2085 size -= addr - pmem->address; 2086 } 2087 2088 /* only process pages below or equal to physmax */ 2089 if ((btop(addr + size) - 1) > physmax) 2090 size = ptob(physmax - btop(addr) + 1); 2091 2092 num = btop(size); 2093 if (num == 0) 2094 continue; 2095 2096 if (num > npages) 2097 num = npages; 2098 2099 npages -= num; 2100 pages_done += num; 2101 base_pfn = btop(addr); 2102 2103 if (prom_debug) 2104 prom_printf("MEMSEG addr=0x%" PRIx64 2105 " pgs=0x%lx pfn 0x%lx-0x%lx\n", 2106 addr, num, base_pfn, base_pfn + num); 2107 2108 /* 2109 * Ignore pages below ddiphysmin to simplify ddi memory 2110 * allocation with non-zero addr_lo requests. 2111 */ 2112 if (base_pfn < ddiphysmin) { 2113 if (base_pfn + num <= ddiphysmin) 2114 continue; 2115 pp += (ddiphysmin - base_pfn); 2116 num -= (ddiphysmin - base_pfn); 2117 base_pfn = ddiphysmin; 2118 } 2119 2120 /* 2121 * Build the memsegs entry 2122 */ 2123 cur_memseg->pages = pp; 2124 cur_memseg->epages = pp + num; 2125 cur_memseg->pages_base = base_pfn; 2126 cur_memseg->pages_end = base_pfn + num; 2127 2128 /* 2129 * Insert into memseg list in decreasing pfn range order. 2130 * Low memory is typically more fragmented such that this 2131 * ordering keeps the larger ranges at the front of the list 2132 * for code that searches memseg. 2133 * This ASSERTS that the memsegs coming in from boot are in 2134 * increasing physical address order and not contiguous. 2135 */ 2136 if (memsegs != NULL) { 2137 ASSERT(cur_memseg->pages_base >= memsegs->pages_end); 2138 cur_memseg->next = memsegs; 2139 } 2140 memsegs = cur_memseg; 2141 2142 /* 2143 * add_physmem() initializes the PSM part of the page 2144 * struct by calling the PSM back with add_physmem_cb(). 2145 * In addition it coalesces pages into larger pages as 2146 * it initializes them. 2147 */ 2148 add_physmem(pp, num, base_pfn); 2149 cur_memseg++; 2150 availrmem_initial += num; 2151 availrmem += num; 2152 2153 pp += num; 2154 } 2155 2156 PRM_DEBUG(availrmem_initial); 2157 PRM_DEBUG(availrmem); 2158 PRM_DEBUG(freemem); 2159 build_pfn_hash(); 2160 return (pages_done); 2161 } 2162 2163 /* 2164 * Kernel VM initialization. 2165 */ 2166 static void 2167 kvm_init(void) 2168 { 2169 ASSERT((((uintptr_t)s_text) & MMU_PAGEOFFSET) == 0); 2170 2171 /* 2172 * Put the kernel segments in kernel address space. 2173 */ 2174 rw_enter(&kas.a_lock, RW_WRITER); 2175 as_avlinit(&kas); 2176 2177 (void) seg_attach(&kas, s_text, e_moddata - s_text, &ktextseg); 2178 (void) segkmem_create(&ktextseg); 2179 2180 (void) seg_attach(&kas, (caddr_t)valloc_base, valloc_sz, &kvalloc); 2181 (void) segkmem_create(&kvalloc); 2182 2183 /* 2184 * We're about to map out /boot. This is the beginning of the 2185 * system resource management transition. We can no longer 2186 * call into /boot for I/O or memory allocations. 2187 */ 2188 (void) seg_attach(&kas, final_kernelheap, 2189 ekernelheap - final_kernelheap, &kvseg); 2190 (void) segkmem_create(&kvseg); 2191 2192 if (core_size > 0) { 2193 PRM_POINT("attaching kvseg_core"); 2194 (void) seg_attach(&kas, (caddr_t)core_base, core_size, 2195 &kvseg_core); 2196 (void) segkmem_create(&kvseg_core); 2197 } 2198 2199 if (segziosize > 0) { 2200 PRM_POINT("attaching segzio"); 2201 (void) seg_attach(&kas, segzio_base, mmu_ptob(segziosize), 2202 &kzioseg); 2203 (void) segkmem_zio_create(&kzioseg); 2204 2205 /* create zio area covering new segment */ 2206 segkmem_zio_init(segzio_base, mmu_ptob(segziosize)); 2207 } 2208 2209 (void) seg_attach(&kas, kdi_segdebugbase, kdi_segdebugsize, &kdebugseg); 2210 (void) segkmem_create(&kdebugseg); 2211 2212 rw_exit(&kas.a_lock); 2213 2214 /* 2215 * Ensure that the red zone at kernelbase is never accessible. 2216 */ 2217 PRM_POINT("protecting redzone"); 2218 (void) as_setprot(&kas, (caddr_t)kernelbase, KERNEL_REDZONE_SIZE, 0); 2219 2220 /* 2221 * Make the text writable so that it can be hot patched by DTrace. 2222 */ 2223 (void) as_setprot(&kas, s_text, e_modtext - s_text, 2224 PROT_READ | PROT_WRITE | PROT_EXEC); 2225 2226 /* 2227 * Make data writable until end. 2228 */ 2229 (void) as_setprot(&kas, s_data, e_moddata - s_data, 2230 PROT_READ | PROT_WRITE | PROT_EXEC); 2231 } 2232 2233 /* 2234 * These are MTTR registers supported by P6 2235 */ 2236 static struct mtrrvar mtrrphys_arr[MAX_MTRRVAR]; 2237 static uint64_t mtrr64k, mtrr16k1, mtrr16k2; 2238 static uint64_t mtrr4k1, mtrr4k2, mtrr4k3; 2239 static uint64_t mtrr4k4, mtrr4k5, mtrr4k6; 2240 static uint64_t mtrr4k7, mtrr4k8, mtrrcap; 2241 uint64_t mtrrdef, pat_attr_reg; 2242 2243 /* 2244 * Disable reprogramming of MTRRs by default. 2245 */ 2246 int enable_relaxed_mtrr = 0; 2247 2248 void 2249 setup_mtrr(void) 2250 { 2251 int i, ecx; 2252 int vcnt; 2253 struct mtrrvar *mtrrphys; 2254 2255 if (!(x86_feature & X86_MTRR)) 2256 return; 2257 2258 mtrrcap = rdmsr(REG_MTRRCAP); 2259 mtrrdef = rdmsr(REG_MTRRDEF); 2260 if (mtrrcap & MTRRCAP_FIX) { 2261 mtrr64k = rdmsr(REG_MTRR64K); 2262 mtrr16k1 = rdmsr(REG_MTRR16K1); 2263 mtrr16k2 = rdmsr(REG_MTRR16K2); 2264 mtrr4k1 = rdmsr(REG_MTRR4K1); 2265 mtrr4k2 = rdmsr(REG_MTRR4K2); 2266 mtrr4k3 = rdmsr(REG_MTRR4K3); 2267 mtrr4k4 = rdmsr(REG_MTRR4K4); 2268 mtrr4k5 = rdmsr(REG_MTRR4K5); 2269 mtrr4k6 = rdmsr(REG_MTRR4K6); 2270 mtrr4k7 = rdmsr(REG_MTRR4K7); 2271 mtrr4k8 = rdmsr(REG_MTRR4K8); 2272 } 2273 if ((vcnt = (mtrrcap & MTRRCAP_VCNTMASK)) > MAX_MTRRVAR) 2274 vcnt = MAX_MTRRVAR; 2275 2276 for (i = 0, ecx = REG_MTRRPHYSBASE0, mtrrphys = mtrrphys_arr; 2277 i < vcnt - 1; i++, ecx += 2, mtrrphys++) { 2278 mtrrphys->mtrrphys_base = rdmsr(ecx); 2279 mtrrphys->mtrrphys_mask = rdmsr(ecx + 1); 2280 if ((x86_feature & X86_PAT) && enable_relaxed_mtrr) { 2281 mtrrphys->mtrrphys_mask &= ~MTRRPHYSMASK_V; 2282 } 2283 } 2284 if (x86_feature & X86_PAT) { 2285 if (enable_relaxed_mtrr) 2286 mtrrdef = MTRR_TYPE_WB|MTRRDEF_FE|MTRRDEF_E; 2287 pat_attr_reg = PAT_DEFAULT_ATTRIBUTE; 2288 } 2289 2290 mtrr_sync(); 2291 } 2292 2293 /* 2294 * Sync current cpu mtrr with the incore copy of mtrr. 2295 * This function has to be invoked with interrupts disabled 2296 * Currently we do not capture other cpu's. This is invoked on cpu0 2297 * just after reading /etc/system. 2298 * On other cpu's its invoked from mp_startup(). 2299 */ 2300 void 2301 mtrr_sync(void) 2302 { 2303 uint_t crvalue, cr0_orig; 2304 int vcnt, i, ecx; 2305 struct mtrrvar *mtrrphys; 2306 2307 cr0_orig = crvalue = getcr0(); 2308 crvalue |= CR0_CD; 2309 crvalue &= ~CR0_NW; 2310 setcr0(crvalue); 2311 invalidate_cache(); 2312 2313 reload_cr3(); 2314 if (x86_feature & X86_PAT) 2315 wrmsr(REG_MTRRPAT, pat_attr_reg); 2316 2317 wrmsr(REG_MTRRDEF, rdmsr(REG_MTRRDEF) & 2318 ~((uint64_t)(uintptr_t)MTRRDEF_E)); 2319 2320 if (mtrrcap & MTRRCAP_FIX) { 2321 wrmsr(REG_MTRR64K, mtrr64k); 2322 wrmsr(REG_MTRR16K1, mtrr16k1); 2323 wrmsr(REG_MTRR16K2, mtrr16k2); 2324 wrmsr(REG_MTRR4K1, mtrr4k1); 2325 wrmsr(REG_MTRR4K2, mtrr4k2); 2326 wrmsr(REG_MTRR4K3, mtrr4k3); 2327 wrmsr(REG_MTRR4K4, mtrr4k4); 2328 wrmsr(REG_MTRR4K5, mtrr4k5); 2329 wrmsr(REG_MTRR4K6, mtrr4k6); 2330 wrmsr(REG_MTRR4K7, mtrr4k7); 2331 wrmsr(REG_MTRR4K8, mtrr4k8); 2332 } 2333 if ((vcnt = (mtrrcap & MTRRCAP_VCNTMASK)) > MAX_MTRRVAR) 2334 vcnt = MAX_MTRRVAR; 2335 for (i = 0, ecx = REG_MTRRPHYSBASE0, mtrrphys = mtrrphys_arr; 2336 i < vcnt - 1; i++, ecx += 2, mtrrphys++) { 2337 wrmsr(ecx, mtrrphys->mtrrphys_base); 2338 wrmsr(ecx + 1, mtrrphys->mtrrphys_mask); 2339 } 2340 wrmsr(REG_MTRRDEF, mtrrdef); 2341 2342 reload_cr3(); 2343 invalidate_cache(); 2344 setcr0(cr0_orig); 2345 } 2346 2347 /* 2348 * resync mtrr so that BIOS is happy. Called from mdboot 2349 */ 2350 void 2351 mtrr_resync(void) 2352 { 2353 if ((x86_feature & X86_PAT) && enable_relaxed_mtrr) { 2354 /* 2355 * We could have changed the default mtrr definition. 2356 * Put it back to uncached which is what it is at power on 2357 */ 2358 mtrrdef = MTRR_TYPE_UC|MTRRDEF_FE|MTRRDEF_E; 2359 mtrr_sync(); 2360 } 2361 } 2362 2363 void 2364 get_system_configuration(void) 2365 { 2366 char prop[32]; 2367 u_longlong_t nodes_ll, cpus_pernode_ll, lvalue; 2368 2369 if (((BOP_GETPROPLEN(bootops, "nodes") > sizeof (prop)) || 2370 (BOP_GETPROP(bootops, "nodes", prop) < 0) || 2371 (kobj_getvalue(prop, &nodes_ll) == -1) || 2372 (nodes_ll > MAXNODES)) || 2373 ((BOP_GETPROPLEN(bootops, "cpus_pernode") > sizeof (prop)) || 2374 (BOP_GETPROP(bootops, "cpus_pernode", prop) < 0) || 2375 (kobj_getvalue(prop, &cpus_pernode_ll) == -1))) { 2376 2377 system_hardware.hd_nodes = 1; 2378 system_hardware.hd_cpus_per_node = 0; 2379 } else { 2380 system_hardware.hd_nodes = (int)nodes_ll; 2381 system_hardware.hd_cpus_per_node = (int)cpus_pernode_ll; 2382 } 2383 if ((BOP_GETPROPLEN(bootops, "kernelbase") > sizeof (prop)) || 2384 (BOP_GETPROP(bootops, "kernelbase", prop) < 0) || 2385 (kobj_getvalue(prop, &lvalue) == -1)) 2386 eprom_kernelbase = NULL; 2387 else 2388 eprom_kernelbase = (uintptr_t)lvalue; 2389 2390 if ((BOP_GETPROPLEN(bootops, "segmapsize") > sizeof (prop)) || 2391 (BOP_GETPROP(bootops, "segmapsize", prop) < 0) || 2392 (kobj_getvalue(prop, &lvalue) == -1)) { 2393 segmapsize = SEGMAPDEFAULT; 2394 } else { 2395 segmapsize = (uintptr_t)lvalue; 2396 } 2397 2398 if ((BOP_GETPROPLEN(bootops, "segmapfreelists") > sizeof (prop)) || 2399 (BOP_GETPROP(bootops, "segmapfreelists", prop) < 0) || 2400 (kobj_getvalue(prop, &lvalue) == -1)) { 2401 segmapfreelists = 0; /* use segmap driver default */ 2402 } else { 2403 segmapfreelists = (int)lvalue; 2404 } 2405 2406 /* physmem used to be here, but moved much earlier to fakebop.c */ 2407 } 2408 2409 /* 2410 * Add to a memory list. 2411 * start = start of new memory segment 2412 * len = length of new memory segment in bytes 2413 * new = pointer to a new struct memlist 2414 * memlistp = memory list to which to add segment. 2415 */ 2416 void 2417 memlist_add( 2418 uint64_t start, 2419 uint64_t len, 2420 struct memlist *new, 2421 struct memlist **memlistp) 2422 { 2423 struct memlist *cur; 2424 uint64_t end = start + len; 2425 2426 new->address = start; 2427 new->size = len; 2428 2429 cur = *memlistp; 2430 2431 while (cur) { 2432 if (cur->address >= end) { 2433 new->next = cur; 2434 *memlistp = new; 2435 new->prev = cur->prev; 2436 cur->prev = new; 2437 return; 2438 } 2439 ASSERT(cur->address + cur->size <= start); 2440 if (cur->next == NULL) { 2441 cur->next = new; 2442 new->prev = cur; 2443 new->next = NULL; 2444 return; 2445 } 2446 memlistp = &cur->next; 2447 cur = cur->next; 2448 } 2449 } 2450 2451 void 2452 kobj_vmem_init(vmem_t **text_arena, vmem_t **data_arena) 2453 { 2454 size_t tsize = e_modtext - modtext; 2455 size_t dsize = e_moddata - moddata; 2456 2457 *text_arena = vmem_create("module_text", tsize ? modtext : NULL, tsize, 2458 1, segkmem_alloc, segkmem_free, heaptext_arena, 0, VM_SLEEP); 2459 *data_arena = vmem_create("module_data", dsize ? moddata : NULL, dsize, 2460 1, segkmem_alloc, segkmem_free, heap32_arena, 0, VM_SLEEP); 2461 } 2462 2463 caddr_t 2464 kobj_text_alloc(vmem_t *arena, size_t size) 2465 { 2466 return (vmem_alloc(arena, size, VM_SLEEP | VM_BESTFIT)); 2467 } 2468 2469 /*ARGSUSED*/ 2470 caddr_t 2471 kobj_texthole_alloc(caddr_t addr, size_t size) 2472 { 2473 panic("unexpected call to kobj_texthole_alloc()"); 2474 /*NOTREACHED*/ 2475 return (0); 2476 } 2477 2478 /*ARGSUSED*/ 2479 void 2480 kobj_texthole_free(caddr_t addr, size_t size) 2481 { 2482 panic("unexpected call to kobj_texthole_free()"); 2483 } 2484 2485 /* 2486 * This is called just after configure() in startup(). 2487 * 2488 * The ISALIST concept is a bit hopeless on Intel, because 2489 * there's no guarantee of an ever-more-capable processor 2490 * given that various parts of the instruction set may appear 2491 * and disappear between different implementations. 2492 * 2493 * While it would be possible to correct it and even enhance 2494 * it somewhat, the explicit hardware capability bitmask allows 2495 * more flexibility. 2496 * 2497 * So, we just leave this alone. 2498 */ 2499 void 2500 setx86isalist(void) 2501 { 2502 char *tp; 2503 size_t len; 2504 extern char *isa_list; 2505 2506 #define TBUFSIZE 1024 2507 2508 tp = kmem_alloc(TBUFSIZE, KM_SLEEP); 2509 *tp = '\0'; 2510 2511 #if defined(__amd64) 2512 (void) strcpy(tp, "amd64 "); 2513 #endif 2514 2515 switch (x86_vendor) { 2516 case X86_VENDOR_Intel: 2517 case X86_VENDOR_AMD: 2518 case X86_VENDOR_TM: 2519 if (x86_feature & X86_CMOV) { 2520 /* 2521 * Pentium Pro or later 2522 */ 2523 (void) strcat(tp, "pentium_pro"); 2524 (void) strcat(tp, x86_feature & X86_MMX ? 2525 "+mmx pentium_pro " : " "); 2526 } 2527 /*FALLTHROUGH*/ 2528 case X86_VENDOR_Cyrix: 2529 /* 2530 * The Cyrix 6x86 does not have any Pentium features 2531 * accessible while not at privilege level 0. 2532 */ 2533 if (x86_feature & X86_CPUID) { 2534 (void) strcat(tp, "pentium"); 2535 (void) strcat(tp, x86_feature & X86_MMX ? 2536 "+mmx pentium " : " "); 2537 } 2538 break; 2539 default: 2540 break; 2541 } 2542 (void) strcat(tp, "i486 i386 i86"); 2543 len = strlen(tp) + 1; /* account for NULL at end of string */ 2544 isa_list = strcpy(kmem_alloc(len, KM_SLEEP), tp); 2545 kmem_free(tp, TBUFSIZE); 2546 2547 #undef TBUFSIZE 2548 } 2549 2550 2551 #ifdef __amd64 2552 2553 void * 2554 device_arena_alloc(size_t size, int vm_flag) 2555 { 2556 return (vmem_alloc(device_arena, size, vm_flag)); 2557 } 2558 2559 void 2560 device_arena_free(void *vaddr, size_t size) 2561 { 2562 vmem_free(device_arena, vaddr, size); 2563 } 2564 2565 #else /* __i386 */ 2566 2567 void * 2568 device_arena_alloc(size_t size, int vm_flag) 2569 { 2570 caddr_t vaddr; 2571 uintptr_t v; 2572 size_t start; 2573 size_t end; 2574 2575 vaddr = vmem_alloc(heap_arena, size, vm_flag); 2576 if (vaddr == NULL) 2577 return (NULL); 2578 2579 v = (uintptr_t)vaddr; 2580 ASSERT(v >= kernelbase); 2581 ASSERT(v + size <= valloc_base); 2582 2583 start = btop(v - kernelbase); 2584 end = btop(v + size - 1 - kernelbase); 2585 ASSERT(start < toxic_bit_map_len); 2586 ASSERT(end < toxic_bit_map_len); 2587 2588 while (start <= end) { 2589 BT_ATOMIC_SET(toxic_bit_map, start); 2590 ++start; 2591 } 2592 return (vaddr); 2593 } 2594 2595 void 2596 device_arena_free(void *vaddr, size_t size) 2597 { 2598 uintptr_t v = (uintptr_t)vaddr; 2599 size_t start; 2600 size_t end; 2601 2602 ASSERT(v >= kernelbase); 2603 ASSERT(v + size <= valloc_base); 2604 2605 start = btop(v - kernelbase); 2606 end = btop(v + size - 1 - kernelbase); 2607 ASSERT(start < toxic_bit_map_len); 2608 ASSERT(end < toxic_bit_map_len); 2609 2610 while (start <= end) { 2611 ASSERT(BT_TEST(toxic_bit_map, start) != 0); 2612 BT_ATOMIC_CLEAR(toxic_bit_map, start); 2613 ++start; 2614 } 2615 vmem_free(heap_arena, vaddr, size); 2616 } 2617 2618 /* 2619 * returns 1st address in range that is in device arena, or NULL 2620 * if len is not NULL it returns the length of the toxic range 2621 */ 2622 void * 2623 device_arena_contains(void *vaddr, size_t size, size_t *len) 2624 { 2625 uintptr_t v = (uintptr_t)vaddr; 2626 uintptr_t eaddr = v + size; 2627 size_t start; 2628 size_t end; 2629 2630 /* 2631 * if called very early by kmdb, just return NULL 2632 */ 2633 if (toxic_bit_map == NULL) 2634 return (NULL); 2635 2636 /* 2637 * First check if we're completely outside the bitmap range. 2638 */ 2639 if (v >= valloc_base || eaddr < kernelbase) 2640 return (NULL); 2641 2642 /* 2643 * Trim ends of search to look at only what the bitmap covers. 2644 */ 2645 if (v < kernelbase) 2646 v = kernelbase; 2647 start = btop(v - kernelbase); 2648 end = btop(eaddr - kernelbase); 2649 if (end >= toxic_bit_map_len) 2650 end = toxic_bit_map_len; 2651 2652 if (bt_range(toxic_bit_map, &start, &end, end) == 0) 2653 return (NULL); 2654 2655 v = kernelbase + ptob(start); 2656 if (len != NULL) 2657 *len = ptob(end - start); 2658 return ((void *)v); 2659 } 2660 2661 #endif /* __i386 */ 2662