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 size_t textrepl_size_thresh; 838 extern void startup_build_mem_nodes(struct memlist *); 839 840 /* XX64 fix these - they should be in include files */ 841 extern size_t page_coloring_init(uint_t, int, int); 842 extern void page_coloring_setup(caddr_t); 843 844 PRM_POINT("startup_memlist() starting..."); 845 846 /* 847 * Use leftover large page nucleus text/data space for loadable modules. 848 * Use at most MODTEXT/MODDATA. 849 */ 850 len = kbm_nucleus_size; 851 ASSERT(len > MMU_PAGESIZE); 852 853 moddata = (caddr_t)ROUND_UP_PAGE(e_data); 854 e_moddata = (caddr_t)P2ROUNDUP((uintptr_t)e_data, (uintptr_t)len); 855 if (e_moddata - moddata > MODDATA) 856 e_moddata = moddata + MODDATA; 857 858 modtext = (caddr_t)ROUND_UP_PAGE(e_text); 859 e_modtext = (caddr_t)P2ROUNDUP((uintptr_t)e_text, (uintptr_t)len); 860 if (e_modtext - modtext > MODTEXT) 861 e_modtext = modtext + MODTEXT; 862 863 econtig = e_moddata; 864 865 PRM_DEBUG(modtext); 866 PRM_DEBUG(e_modtext); 867 PRM_DEBUG(moddata); 868 PRM_DEBUG(e_moddata); 869 PRM_DEBUG(econtig); 870 871 /* 872 * Examine the boot loader physical memory map to find out: 873 * - total memory in system - physinstalled 874 * - the max physical address - physmax 875 * - the number of discontiguous segments of memory. 876 */ 877 if (prom_debug) 878 print_memlist("boot physinstalled", 879 bootops->boot_mem->physinstalled); 880 installed_top_size(bootops->boot_mem->physinstalled, &physmax, 881 &physinstalled, &memblocks); 882 PRM_DEBUG(physmax); 883 PRM_DEBUG(physinstalled); 884 PRM_DEBUG(memblocks); 885 886 /* 887 * Initialize hat's mmu parameters. 888 * Check for enforce-prot-exec in boot environment. It's used to 889 * enable/disable support for the page table entry NX bit. 890 * The default is to enforce PROT_EXEC on processors that support NX. 891 * Boot seems to round up the "len", but 8 seems to be big enough. 892 */ 893 mmu_init(); 894 895 #ifdef __i386 896 /* 897 * physmax is lowered if there is more memory than can be 898 * physically addressed in 32 bit (PAE/non-PAE) modes. 899 */ 900 if (mmu.pae_hat) { 901 if (PFN_ABOVE64G(physmax)) { 902 physinstalled -= (physmax - (PFN_64G - 1)); 903 physmax = PFN_64G - 1; 904 } 905 } else { 906 if (PFN_ABOVE4G(physmax)) { 907 physinstalled -= (physmax - (PFN_4G - 1)); 908 physmax = PFN_4G - 1; 909 } 910 } 911 #endif 912 913 startup_build_mem_nodes(bootops->boot_mem->physinstalled); 914 915 if (BOP_GETPROPLEN(bootops, "enforce-prot-exec") >= 0) { 916 int len = BOP_GETPROPLEN(bootops, "enforce-prot-exec"); 917 char value[8]; 918 919 if (len < 8) 920 (void) BOP_GETPROP(bootops, "enforce-prot-exec", value); 921 else 922 (void) strcpy(value, ""); 923 if (strcmp(value, "off") == 0) 924 mmu.pt_nx = 0; 925 } 926 PRM_DEBUG(mmu.pt_nx); 927 928 /* 929 * We will need page_t's for every page in the system, except for 930 * memory mapped at or above above the start of the kernel text segment. 931 * 932 * pages above e_modtext are attributed to kernel debugger (obp_pages) 933 */ 934 npages = physinstalled - 1; /* avail_filter() skips page 0, so "- 1" */ 935 obp_pages = 0; 936 va = KERNEL_TEXT; 937 while (kbm_probe(&va, &len, &pfn, &prot) != 0) { 938 npages -= len >> MMU_PAGESHIFT; 939 if (va >= (uintptr_t)e_moddata) 940 obp_pages += len >> MMU_PAGESHIFT; 941 va += len; 942 } 943 PRM_DEBUG(npages); 944 PRM_DEBUG(obp_pages); 945 946 /* 947 * If physmem is patched to be non-zero, use it instead of 948 * the computed value unless it is larger than the real 949 * amount of memory on hand. 950 */ 951 if (physmem == 0 || physmem > npages) { 952 physmem = npages; 953 } else if (physmem < npages) { 954 orig_npages = npages; 955 npages = physmem; 956 } 957 PRM_DEBUG(physmem); 958 959 /* 960 * We now compute the sizes of all the initial allocations for 961 * structures the kernel needs in order do kmem_alloc(). These 962 * include: 963 * memsegs 964 * memlists 965 * page hash table 966 * page_t's 967 * page coloring data structs 968 */ 969 memseg_sz = sizeof (struct memseg) * (memblocks + POSS_NEW_FRAGMENTS); 970 ADD_TO_ALLOCATIONS(memseg_base, memseg_sz); 971 PRM_DEBUG(memseg_sz); 972 973 /* 974 * Reserve space for memlists. There's no real good way to know exactly 975 * how much room we'll need, but this should be a good upper bound. 976 */ 977 memlist_sz = ROUND_UP_PAGE(2 * sizeof (struct memlist) * 978 (memblocks + POSS_NEW_FRAGMENTS)); 979 ADD_TO_ALLOCATIONS(memlist, memlist_sz); 980 PRM_DEBUG(memlist_sz); 981 982 /* 983 * The page structure hash table size is a power of 2 984 * such that the average hash chain length is PAGE_HASHAVELEN. 985 */ 986 page_hashsz = npages / PAGE_HASHAVELEN; 987 page_hashsz = 1 << highbit(page_hashsz); 988 pagehash_sz = sizeof (struct page *) * page_hashsz; 989 ADD_TO_ALLOCATIONS(page_hash, pagehash_sz); 990 PRM_DEBUG(pagehash_sz); 991 992 /* 993 * Set aside room for the page structures themselves. 994 */ 995 PRM_DEBUG(npages); 996 pp_sz = sizeof (struct page) * npages; 997 ADD_TO_ALLOCATIONS(pp_base, pp_sz); 998 PRM_DEBUG(pp_sz); 999 1000 /* 1001 * determine l2 cache info and memory size for page coloring 1002 */ 1003 (void) getl2cacheinfo(CPU, 1004 &l2cache_sz, &l2cache_linesz, &l2cache_assoc); 1005 pagecolor_memsz = 1006 page_coloring_init(l2cache_sz, l2cache_linesz, l2cache_assoc); 1007 ADD_TO_ALLOCATIONS(pagecolor_mem, pagecolor_memsz); 1008 PRM_DEBUG(pagecolor_memsz); 1009 1010 page_ctrs_size = page_ctrs_sz(); 1011 ADD_TO_ALLOCATIONS(page_ctrs_mem, page_ctrs_size); 1012 PRM_DEBUG(page_ctrs_size); 1013 1014 #if defined(__amd64) 1015 valloc_sz = ROUND_UP_LPAGE(valloc_sz); 1016 valloc_base = VALLOC_BASE; 1017 #else /* __i386 */ 1018 valloc_base = (uintptr_t)(MISC_VA_BASE - valloc_sz); 1019 valloc_base = P2ALIGN(valloc_base, mmu.level_size[1]); 1020 #endif /* __i386 */ 1021 PRM_DEBUG(valloc_base); 1022 1023 /* 1024 * do all the initial allocations 1025 */ 1026 perform_allocations(); 1027 1028 /* 1029 * Build phys_install and phys_avail in kernel memspace. 1030 * - phys_install should be all memory in the system. 1031 * - phys_avail is phys_install minus any memory mapped before this 1032 * point above KERNEL_TEXT. 1033 */ 1034 current = phys_install = memlist; 1035 copy_memlist_filter(bootops->boot_mem->physinstalled, ¤t, NULL); 1036 if ((caddr_t)current > (caddr_t)memlist + memlist_sz) 1037 panic("physinstalled was too big!"); 1038 if (prom_debug) 1039 print_memlist("phys_install", phys_install); 1040 1041 phys_avail = current; 1042 PRM_POINT("Building phys_avail:\n"); 1043 copy_memlist_filter(bootops->boot_mem->physinstalled, ¤t, 1044 avail_filter); 1045 if ((caddr_t)current > (caddr_t)memlist + memlist_sz) 1046 panic("physavail was too big!"); 1047 if (prom_debug) 1048 print_memlist("phys_avail", phys_avail); 1049 1050 /* 1051 * setup page coloring 1052 */ 1053 page_coloring_setup(pagecolor_mem); 1054 page_lock_init(); /* currently a no-op */ 1055 1056 /* 1057 * free page list counters 1058 */ 1059 (void) page_ctrs_alloc(page_ctrs_mem); 1060 1061 /* 1062 * Initialize the page structures from the memory lists. 1063 */ 1064 availrmem_initial = availrmem = freemem = 0; 1065 PRM_POINT("Calling kphysm_init()..."); 1066 npages = kphysm_init(pp_base, npages); 1067 PRM_POINT("kphysm_init() done"); 1068 PRM_DEBUG(npages); 1069 1070 init_debug_info(); 1071 1072 /* 1073 * Now that page_t's have been initialized, remove all the 1074 * initial allocation pages from the kernel free page lists. 1075 */ 1076 boot_mapin((caddr_t)valloc_base, valloc_sz); 1077 boot_mapin((caddr_t)GDT_VA, MMU_PAGESIZE); 1078 boot_mapin((caddr_t)DEBUG_INFO_VA, MMU_PAGESIZE); 1079 PRM_POINT("startup_memlist() done"); 1080 1081 PRM_DEBUG(valloc_sz); 1082 1083 if ((availrmem >> (30 - MMU_PAGESHIFT)) >= textrepl_min_gb && 1084 l2cache_sz <= 2 << 20) { 1085 textrepl_size_thresh = (16 << 20) - 1; 1086 } 1087 } 1088 1089 /* 1090 * Layout the kernel's part of address space and initialize kmem allocator. 1091 */ 1092 static void 1093 startup_kmem(void) 1094 { 1095 extern void page_set_colorequiv_arr(void); 1096 1097 PRM_POINT("startup_kmem() starting..."); 1098 1099 #if defined(__amd64) 1100 if (eprom_kernelbase && eprom_kernelbase != KERNELBASE) 1101 cmn_err(CE_NOTE, "!kernelbase cannot be changed on 64-bit " 1102 "systems."); 1103 kernelbase = (uintptr_t)KERNELBASE; 1104 core_base = (uintptr_t)COREHEAP_BASE; 1105 core_size = (size_t)MISC_VA_BASE - COREHEAP_BASE; 1106 #else /* __i386 */ 1107 /* 1108 * We configure kernelbase based on: 1109 * 1110 * 1. user specified kernelbase via eeprom command. Value cannot exceed 1111 * KERNELBASE_MAX. we large page align eprom_kernelbase 1112 * 1113 * 2. Default to KERNELBASE and adjust to 2X less the size for page_t. 1114 * On large memory systems we must lower kernelbase to allow 1115 * enough room for page_t's for all of memory. 1116 * 1117 * The value set here, might be changed a little later. 1118 */ 1119 if (eprom_kernelbase) { 1120 kernelbase = eprom_kernelbase & mmu.level_mask[1]; 1121 if (kernelbase > KERNELBASE_MAX) 1122 kernelbase = KERNELBASE_MAX; 1123 } else { 1124 kernelbase = (uintptr_t)KERNELBASE; 1125 kernelbase -= ROUND_UP_4MEG(2 * valloc_sz); 1126 } 1127 ASSERT((kernelbase & mmu.level_offset[1]) == 0); 1128 core_base = valloc_base; 1129 core_size = 0; 1130 #endif /* __i386 */ 1131 1132 PRM_DEBUG(core_base); 1133 PRM_DEBUG(core_size); 1134 PRM_DEBUG(kernelbase); 1135 1136 /* 1137 * At this point, we can only use a portion of the kernelheap that 1138 * will be available after we boot. 32-bit systems have this 1139 * limitation. 1140 * 1141 * On 32-bit systems we have to leave room to place segmap below 1142 * the heap. We don't yet know how large segmap will be, so we 1143 * have to be very conservative. 1144 * 1145 * On 64 bit systems there should be LOTS of room so just use 1146 * the next 4Gig below core_base. 1147 */ 1148 #if defined(__amd64) 1149 1150 boot_kernelheap = (caddr_t)core_base - FOURGB; 1151 segmap_reserved = 0; 1152 1153 #else /* __i386 */ 1154 1155 segkp_fromheap = 1; 1156 segmap_reserved = ROUND_UP_LPAGE(MAX(segmapsize, SEGMAPMAX)); 1157 boot_kernelheap = 1158 (caddr_t)ROUND_UP_LPAGE(kernelbase) + segmap_reserved; 1159 1160 #endif /* __i386 */ 1161 PRM_DEBUG(boot_kernelheap); 1162 ekernelheap = (char *)core_base; 1163 PRM_DEBUG(ekernelheap); 1164 kernelheap = boot_kernelheap; 1165 1166 /* 1167 * If segmap is too large we can push the bottom of the kernel heap 1168 * higher than the base. Or worse, it could exceed the top of the 1169 * VA space entirely, causing it to wrap around. 1170 */ 1171 if (kernelheap >= ekernelheap || (uintptr_t)kernelheap < kernelbase) 1172 panic("too little memory available for kernelheap," 1173 " use a different kernelbase"); 1174 1175 /* 1176 * Now that we know the real value of kernelbase, 1177 * update variables that were initialized with a value of 1178 * KERNELBASE (in common/conf/param.c). 1179 * 1180 * XXX The problem with this sort of hackery is that the 1181 * compiler just may feel like putting the const declarations 1182 * (in param.c) into the .text section. Perhaps they should 1183 * just be declared as variables there? 1184 */ 1185 1186 #if defined(__amd64) 1187 ASSERT(_kernelbase == KERNELBASE); 1188 ASSERT(_userlimit == USERLIMIT); 1189 #else 1190 *(uintptr_t *)&_kernelbase = kernelbase; 1191 *(uintptr_t *)&_userlimit = kernelbase; 1192 *(uintptr_t *)&_userlimit32 = _userlimit; 1193 #endif 1194 PRM_DEBUG(_kernelbase); 1195 PRM_DEBUG(_userlimit); 1196 PRM_DEBUG(_userlimit32); 1197 1198 /* 1199 * Initialize the kernel heap. Note 3rd argument must be > 1st. 1200 */ 1201 kernelheap_init(boot_kernelheap, ekernelheap, 1202 boot_kernelheap + MMU_PAGESIZE, 1203 (void *)core_base, (void *)(core_base + core_size)); 1204 1205 /* 1206 * Initialize kernel memory allocator. 1207 */ 1208 kmem_init(); 1209 1210 /* 1211 * Factor in colorequiv to check additional 'equivalent' bins 1212 */ 1213 page_set_colorequiv_arr(); 1214 1215 /* 1216 * print this out early so that we know what's going on 1217 */ 1218 cmn_err(CE_CONT, "?features: %b\n", x86_feature, FMT_X86_FEATURE); 1219 1220 /* 1221 * Initialize bp_mapin(). 1222 */ 1223 bp_init(MMU_PAGESIZE, HAT_STORECACHING_OK); 1224 1225 /* 1226 * orig_npages is non-zero if physmem has been configured for less 1227 * than the available memory. 1228 */ 1229 if (orig_npages) { 1230 #ifdef __i386 1231 /* 1232 * use npages for physmem in case it has been temporarily 1233 * modified via /etc/system in kmem_init/mod_read_system_file. 1234 */ 1235 if (npages == PHYSMEM32) { 1236 cmn_err(CE_WARN, "!Due to 32-bit virtual" 1237 " address space limitations, limiting" 1238 " physmem to 0x%lx of 0x%lx available pages", 1239 npages, orig_npages); 1240 } else { 1241 cmn_err(CE_WARN, "!limiting physmem to 0x%lx of" 1242 " 0x%lx available pages", npages, orig_npages); 1243 } 1244 #else 1245 cmn_err(CE_WARN, "!limiting physmem to 0x%lx of" 1246 " 0x%lx available pages", npages, orig_npages); 1247 #endif 1248 } 1249 #if defined(__i386) 1250 if (eprom_kernelbase && (eprom_kernelbase != kernelbase)) 1251 cmn_err(CE_WARN, "kernelbase value, User specified 0x%lx, " 1252 "System using 0x%lx", 1253 (uintptr_t)eprom_kernelbase, (uintptr_t)kernelbase); 1254 #endif 1255 1256 #ifdef KERNELBASE_ABI_MIN 1257 if (kernelbase < (uintptr_t)KERNELBASE_ABI_MIN) { 1258 cmn_err(CE_NOTE, "!kernelbase set to 0x%lx, system is not " 1259 "i386 ABI compliant.", (uintptr_t)kernelbase); 1260 } 1261 #endif 1262 1263 PRM_POINT("startup_kmem() done"); 1264 } 1265 1266 static void 1267 startup_modules(void) 1268 { 1269 unsigned int i; 1270 extern void prom_setup(void); 1271 1272 PRM_POINT("startup_modules() starting..."); 1273 /* 1274 * Initialize ten-micro second timer so that drivers will 1275 * not get short changed in their init phase. This was 1276 * not getting called until clkinit which, on fast cpu's 1277 * caused the drv_usecwait to be way too short. 1278 */ 1279 microfind(); 1280 1281 /* 1282 * Read the GMT lag from /etc/rtc_config. 1283 */ 1284 sgmtl(process_rtc_config_file()); 1285 1286 /* 1287 * Calculate default settings of system parameters based upon 1288 * maxusers, yet allow to be overridden via the /etc/system file. 1289 */ 1290 param_calc(0); 1291 1292 mod_setup(); 1293 1294 /* 1295 * Initialize system parameters. 1296 */ 1297 param_init(); 1298 1299 /* 1300 * Initialize the default brands 1301 */ 1302 brand_init(); 1303 1304 /* 1305 * maxmem is the amount of physical memory we're playing with. 1306 */ 1307 maxmem = physmem; 1308 1309 /* 1310 * Initialize the hat layer. 1311 */ 1312 hat_init(); 1313 1314 /* 1315 * Initialize segment management stuff. 1316 */ 1317 seg_init(); 1318 1319 if (modload("fs", "specfs") == -1) 1320 halt("Can't load specfs"); 1321 1322 if (modload("fs", "devfs") == -1) 1323 halt("Can't load devfs"); 1324 1325 if (modload("fs", "dev") == -1) 1326 halt("Can't load dev"); 1327 1328 (void) modloadonly("sys", "lbl_edition"); 1329 1330 dispinit(); 1331 1332 /* 1333 * This is needed here to initialize hw_serial[] for cluster booting. 1334 */ 1335 if ((i = modload("misc", "sysinit")) != (unsigned int)-1) 1336 (void) modunload(i); 1337 else 1338 cmn_err(CE_CONT, "sysinit load failed"); 1339 1340 /* Read cluster configuration data. */ 1341 clconf_init(); 1342 1343 /* 1344 * Create a kernel device tree. First, create rootnex and 1345 * then invoke bus specific code to probe devices. 1346 */ 1347 setup_ddi(); 1348 1349 /* 1350 * Set up the CPU module subsystem. Modifies the device tree, so it 1351 * must be done after setup_ddi(). 1352 */ 1353 cmi_init(); 1354 1355 /* 1356 * Initialize the MCA handlers 1357 */ 1358 if (x86_feature & X86_MCA) 1359 cmi_mca_init(); 1360 1361 /* 1362 * Fake a prom tree such that /dev/openprom continues to work 1363 */ 1364 PRM_POINT("startup_modules: calling prom_setup..."); 1365 prom_setup(); 1366 PRM_POINT("startup_modules: done"); 1367 1368 /* 1369 * Load all platform specific modules 1370 */ 1371 PRM_POINT("startup_modules: calling psm_modload..."); 1372 psm_modload(); 1373 1374 PRM_POINT("startup_modules() done"); 1375 } 1376 1377 /* 1378 * claim a "setaside" boot page for use in the kernel 1379 */ 1380 page_t * 1381 boot_claim_page(pfn_t pfn) 1382 { 1383 page_t *pp; 1384 1385 pp = page_numtopp_nolock(pfn); 1386 ASSERT(pp != NULL); 1387 1388 if (PP_ISBOOTPAGES(pp)) { 1389 if (pp->p_next != NULL) 1390 pp->p_next->p_prev = pp->p_prev; 1391 if (pp->p_prev == NULL) 1392 bootpages = pp->p_next; 1393 else 1394 pp->p_prev->p_next = pp->p_next; 1395 } else { 1396 /* 1397 * htable_attach() expects a base pagesize page 1398 */ 1399 if (pp->p_szc != 0) 1400 page_boot_demote(pp); 1401 pp = page_numtopp(pfn, SE_EXCL); 1402 } 1403 return (pp); 1404 } 1405 1406 /* 1407 * Walk through the pagetables looking for pages mapped in by boot. If the 1408 * setaside flag is set the pages are expected to be returned to the 1409 * kernel later in boot, so we add them to the bootpages list. 1410 */ 1411 static void 1412 protect_boot_range(uintptr_t low, uintptr_t high, int setaside) 1413 { 1414 uintptr_t va = low; 1415 size_t len; 1416 uint_t prot; 1417 pfn_t pfn; 1418 page_t *pp; 1419 pgcnt_t boot_protect_cnt = 0; 1420 1421 while (kbm_probe(&va, &len, &pfn, &prot) != 0 && va < high) { 1422 if (va + len >= high) 1423 panic("0x%lx byte mapping at 0x%p exceeds boot's " 1424 "legal range.", len, (void *)va); 1425 1426 while (len > 0) { 1427 pp = page_numtopp_alloc(pfn); 1428 if (pp != NULL) { 1429 if (setaside == 0) 1430 panic("Unexpected mapping by boot. " 1431 "addr=%p pfn=%lx\n", 1432 (void *)va, pfn); 1433 1434 pp->p_next = bootpages; 1435 pp->p_prev = NULL; 1436 PP_SETBOOTPAGES(pp); 1437 if (bootpages != NULL) { 1438 bootpages->p_prev = pp; 1439 } 1440 bootpages = pp; 1441 ++boot_protect_cnt; 1442 } 1443 1444 ++pfn; 1445 len -= MMU_PAGESIZE; 1446 va += MMU_PAGESIZE; 1447 } 1448 } 1449 PRM_DEBUG(boot_protect_cnt); 1450 } 1451 1452 /* 1453 * Finish initializing the VM system, now that we are no longer 1454 * relying on the boot time memory allocators. 1455 */ 1456 static void 1457 startup_vm(void) 1458 { 1459 struct segmap_crargs a; 1460 1461 extern int use_brk_lpg, use_stk_lpg; 1462 1463 PRM_POINT("startup_vm() starting..."); 1464 1465 /* 1466 * Establish the final size of the kernel's heap, size of segmap, 1467 * segkp, etc. 1468 */ 1469 1470 #if defined(__amd64) 1471 1472 /* 1473 * Check if there is enough virtual address space in KPM region to 1474 * map physmax. 1475 */ 1476 kpm_vbase = (caddr_t)(uintptr_t)SEGKPM_BASE; 1477 kpm_size = 0; 1478 if (kpm_desired) { 1479 kpm_size = ROUND_UP_LPAGE(mmu_ptob(physmax + 1)); 1480 if ((uintptr_t)kpm_vbase + kpm_size > (uintptr_t)VALLOC_BASE) { 1481 kpm_size = 0; 1482 kpm_desired = 0; 1483 } 1484 } 1485 1486 PRM_DEBUG(kpm_size); 1487 PRM_DEBUG(kpm_vbase); 1488 1489 /* 1490 * By default we create a seg_kp in 64 bit kernels, it's a little 1491 * faster to access than embedding it in the heap. 1492 */ 1493 segkp_base = (caddr_t)valloc_base + valloc_sz; 1494 if (!segkp_fromheap) { 1495 size_t sz = mmu_ptob(segkpsize); 1496 1497 /* 1498 * determine size of segkp 1499 */ 1500 if (sz < SEGKPMINSIZE || sz > SEGKPMAXSIZE) { 1501 sz = SEGKPDEFSIZE; 1502 cmn_err(CE_WARN, "!Illegal value for segkpsize. " 1503 "segkpsize has been reset to %ld pages", 1504 mmu_btop(sz)); 1505 } 1506 sz = MIN(sz, MAX(SEGKPMINSIZE, mmu_ptob(physmem))); 1507 1508 segkpsize = mmu_btop(ROUND_UP_LPAGE(sz)); 1509 } 1510 PRM_DEBUG(segkp_base); 1511 PRM_DEBUG(segkpsize); 1512 1513 segzio_base = segkp_base + mmu_ptob(segkpsize); 1514 if (segzio_fromheap) { 1515 segziosize = 0; 1516 } else { 1517 size_t size; 1518 size_t physmem_b = mmu_ptob(physmem); 1519 1520 /* size is in bytes, segziosize is in pages */ 1521 if (segziosize == 0) { 1522 size = physmem_b; 1523 } else { 1524 size = mmu_ptob(segziosize); 1525 } 1526 1527 if (size < SEGZIOMINSIZE) { 1528 size = SEGZIOMINSIZE; 1529 } else if (size > SEGZIOMAXSIZE) { 1530 size = SEGZIOMAXSIZE; 1531 /* 1532 * SEGZIOMAXSIZE is capped at 512gb so that segzio 1533 * doesn't consume all of KVA. However, if we have a 1534 * system that has more thant 512gb of physical memory, 1535 * we can actually consume about half of the difference 1536 * between 512gb and the rest of the available physical 1537 * memory. 1538 */ 1539 if (physmem_b > SEGZIOMAXSIZE) { 1540 size += (physmem_b - SEGZIOMAXSIZE) / 2; 1541 } 1542 } 1543 segziosize = mmu_btop(ROUND_UP_LPAGE(size)); 1544 } 1545 PRM_DEBUG(segziosize); 1546 PRM_DEBUG(segzio_base); 1547 1548 /* 1549 * Put the range of VA for device mappings next, kmdb knows to not 1550 * grep in this range of addresses. 1551 */ 1552 toxic_addr = 1553 ROUND_UP_LPAGE((uintptr_t)segzio_base + mmu_ptob(segziosize)); 1554 PRM_DEBUG(toxic_addr); 1555 segmap_start = ROUND_UP_LPAGE(toxic_addr + toxic_size); 1556 #else /* __i386 */ 1557 segmap_start = ROUND_UP_LPAGE(kernelbase); 1558 #endif /* __i386 */ 1559 PRM_DEBUG(segmap_start); 1560 ASSERT((caddr_t)segmap_start < boot_kernelheap); 1561 1562 /* 1563 * Users can change segmapsize through eeprom or /etc/system. 1564 * If the variable is tuned through eeprom, there is no upper 1565 * bound on the size of segmap. If it is tuned through 1566 * /etc/system on 32-bit systems, it must be no larger than we 1567 * planned for in startup_memlist(). 1568 */ 1569 segmapsize = MAX(ROUND_UP_LPAGE(segmapsize), SEGMAPDEFAULT); 1570 1571 #if defined(__i386) 1572 if (segmapsize > segmap_reserved) { 1573 cmn_err(CE_NOTE, "!segmapsize may not be set > 0x%lx in " 1574 "/etc/system. Use eeprom.", (long)SEGMAPMAX); 1575 segmapsize = segmap_reserved; 1576 } 1577 /* 1578 * 32-bit systems don't have segkpm or segkp, so segmap appears at 1579 * the bottom of the kernel's address range. Set aside space for a 1580 * small red zone just below the start of segmap. 1581 */ 1582 segmap_start += KERNEL_REDZONE_SIZE; 1583 segmapsize -= KERNEL_REDZONE_SIZE; 1584 #endif 1585 1586 PRM_DEBUG(segmap_start); 1587 PRM_DEBUG(segmapsize); 1588 final_kernelheap = (caddr_t)ROUND_UP_LPAGE(segmap_start + segmapsize); 1589 PRM_DEBUG(final_kernelheap); 1590 1591 /* 1592 * Do final allocations of HAT data structures that need to 1593 * be allocated before quiescing the boot loader. 1594 */ 1595 PRM_POINT("Calling hat_kern_alloc()..."); 1596 hat_kern_alloc((caddr_t)segmap_start, segmapsize, ekernelheap); 1597 PRM_POINT("hat_kern_alloc() done"); 1598 1599 /* 1600 * Setup MTRR (Memory type range registers) 1601 */ 1602 setup_mtrr(); 1603 1604 /* 1605 * The next two loops are done in distinct steps in order 1606 * to be sure that any page that is doubly mapped (both above 1607 * KERNEL_TEXT and below kernelbase) is dealt with correctly. 1608 * Note this may never happen, but it might someday. 1609 */ 1610 bootpages = NULL; 1611 PRM_POINT("Protecting boot pages"); 1612 1613 /* 1614 * Protect any pages mapped above KERNEL_TEXT that somehow have 1615 * page_t's. This can only happen if something weird allocated 1616 * in this range (like kadb/kmdb). 1617 */ 1618 protect_boot_range(KERNEL_TEXT, (uintptr_t)-1, 0); 1619 1620 /* 1621 * Before we can take over memory allocation/mapping from the boot 1622 * loader we must remove from our free page lists any boot allocated 1623 * pages that stay mapped until release_bootstrap(). 1624 */ 1625 protect_boot_range(0, kernelbase, 1); 1626 1627 /* 1628 * Switch to running on regular HAT (not boot_mmu) 1629 */ 1630 PRM_POINT("Calling hat_kern_setup()..."); 1631 hat_kern_setup(); 1632 1633 /* 1634 * It is no longer safe to call BOP_ALLOC(), so make sure we don't. 1635 */ 1636 bop_no_more_mem(); 1637 1638 PRM_POINT("hat_kern_setup() done"); 1639 1640 hat_cpu_online(CPU); 1641 1642 /* 1643 * Initialize VM system 1644 */ 1645 PRM_POINT("Calling kvm_init()..."); 1646 kvm_init(); 1647 PRM_POINT("kvm_init() done"); 1648 1649 /* 1650 * Tell kmdb that the VM system is now working 1651 */ 1652 if (boothowto & RB_DEBUG) 1653 kdi_dvec_vmready(); 1654 1655 /* 1656 * Mangle the brand string etc. 1657 */ 1658 cpuid_pass3(CPU); 1659 1660 /* 1661 * Now that we can use memory outside the top 4GB (on 64-bit 1662 * systems) and we know the size of segmap, we can set the final 1663 * size of the kernel's heap. 1664 */ 1665 if (final_kernelheap < boot_kernelheap) { 1666 PRM_POINT("kernelheap_extend()"); 1667 PRM_DEBUG(boot_kernelheap); 1668 PRM_DEBUG(final_kernelheap); 1669 kernelheap_extend(final_kernelheap, boot_kernelheap); 1670 } 1671 1672 #if defined(__amd64) 1673 1674 /* 1675 * Create the device arena for toxic (to dtrace/kmdb) mappings. 1676 */ 1677 device_arena = vmem_create("device", (void *)toxic_addr, 1678 toxic_size, MMU_PAGESIZE, NULL, NULL, NULL, 0, VM_SLEEP); 1679 1680 #else /* __i386 */ 1681 1682 /* 1683 * allocate the bit map that tracks toxic pages 1684 */ 1685 toxic_bit_map_len = btop((ulong_t)(valloc_base - kernelbase)); 1686 PRM_DEBUG(toxic_bit_map_len); 1687 toxic_bit_map = 1688 kmem_zalloc(BT_SIZEOFMAP(toxic_bit_map_len), KM_NOSLEEP); 1689 ASSERT(toxic_bit_map != NULL); 1690 PRM_DEBUG(toxic_bit_map); 1691 1692 #endif /* __i386 */ 1693 1694 1695 /* 1696 * Now that we've got more VA, as well as the ability to allocate from 1697 * it, tell the debugger. 1698 */ 1699 if (boothowto & RB_DEBUG) 1700 kdi_dvec_memavail(); 1701 1702 /* 1703 * The following code installs a special page fault handler (#pf) 1704 * to work around a pentium bug. 1705 */ 1706 #if !defined(__amd64) 1707 if (x86_type == X86_TYPE_P5) { 1708 desctbr_t idtr; 1709 gate_desc_t *newidt; 1710 struct machcpu *mcpu = &CPU->cpu_m; 1711 1712 if ((newidt = kmem_zalloc(MMU_PAGESIZE, KM_NOSLEEP)) == NULL) 1713 panic("failed to install pentium_pftrap"); 1714 1715 bcopy(idt0, newidt, sizeof (idt0)); 1716 set_gatesegd(&newidt[T_PGFLT], &pentium_pftrap, 1717 KCS_SEL, SDT_SYSIGT, SEL_KPL); 1718 1719 (void) as_setprot(&kas, (caddr_t)newidt, MMU_PAGESIZE, 1720 PROT_READ|PROT_EXEC); 1721 1722 mcpu->mcpu_idt = newidt; 1723 idtr.dtr_base = (uintptr_t)mcpu->mcpu_idt; 1724 idtr.dtr_limit = sizeof (idt0) - 1; 1725 wr_idtr(&idtr); 1726 } 1727 #endif /* !__amd64 */ 1728 1729 /* 1730 * Map page pfn=0 for drivers, such as kd, that need to pick up 1731 * parameters left there by controllers/BIOS. 1732 */ 1733 PRM_POINT("setup up p0_va"); 1734 p0_va = i86devmap(0, 1, PROT_READ); 1735 PRM_DEBUG(p0_va); 1736 1737 cmn_err(CE_CONT, "?mem = %luK (0x%lx)\n", 1738 physinstalled << (MMU_PAGESHIFT - 10), ptob(physinstalled)); 1739 1740 /* 1741 * disable automatic large pages for small memory systems or 1742 * when the disable flag is set. 1743 */ 1744 if (!auto_lpg_disable && mmu.max_page_level > 0) { 1745 max_uheap_lpsize = LEVEL_SIZE(1); 1746 max_ustack_lpsize = LEVEL_SIZE(1); 1747 max_privmap_lpsize = LEVEL_SIZE(1); 1748 max_uidata_lpsize = LEVEL_SIZE(1); 1749 max_utext_lpsize = LEVEL_SIZE(1); 1750 max_shm_lpsize = LEVEL_SIZE(1); 1751 } 1752 if (physmem < privm_lpg_min_physmem || mmu.max_page_level == 0 || 1753 auto_lpg_disable) { 1754 use_brk_lpg = 0; 1755 use_stk_lpg = 0; 1756 } 1757 if (mmu.max_page_level > 0) { 1758 mcntl0_lpsize = LEVEL_SIZE(1); 1759 } 1760 1761 PRM_POINT("Calling hat_init_finish()..."); 1762 hat_init_finish(); 1763 PRM_POINT("hat_init_finish() done"); 1764 1765 /* 1766 * Initialize the segkp segment type. 1767 */ 1768 rw_enter(&kas.a_lock, RW_WRITER); 1769 PRM_POINT("Attaching segkp"); 1770 if (segkp_fromheap) { 1771 segkp->s_as = &kas; 1772 } else if (seg_attach(&kas, (caddr_t)segkp_base, mmu_ptob(segkpsize), 1773 segkp) < 0) { 1774 panic("startup: cannot attach segkp"); 1775 /*NOTREACHED*/ 1776 } 1777 PRM_POINT("Doing segkp_create()"); 1778 if (segkp_create(segkp) != 0) { 1779 panic("startup: segkp_create failed"); 1780 /*NOTREACHED*/ 1781 } 1782 PRM_DEBUG(segkp); 1783 rw_exit(&kas.a_lock); 1784 1785 /* 1786 * kpm segment 1787 */ 1788 segmap_kpm = 0; 1789 if (kpm_desired) { 1790 kpm_init(); 1791 kpm_enable = 1; 1792 vpm_enable = 1; 1793 } 1794 1795 /* 1796 * Now create segmap segment. 1797 */ 1798 rw_enter(&kas.a_lock, RW_WRITER); 1799 if (seg_attach(&kas, (caddr_t)segmap_start, segmapsize, segmap) < 0) { 1800 panic("cannot attach segmap"); 1801 /*NOTREACHED*/ 1802 } 1803 PRM_DEBUG(segmap); 1804 1805 a.prot = PROT_READ | PROT_WRITE; 1806 a.shmsize = 0; 1807 a.nfreelist = segmapfreelists; 1808 1809 if (segmap_create(segmap, (caddr_t)&a) != 0) 1810 panic("segmap_create segmap"); 1811 rw_exit(&kas.a_lock); 1812 1813 setup_vaddr_for_ppcopy(CPU); 1814 1815 segdev_init(); 1816 pmem_init(); 1817 1818 PRM_POINT("startup_vm() done"); 1819 } 1820 1821 /* 1822 * Load a tod module for the non-standard tod part found on this system. 1823 */ 1824 static void 1825 load_tod_module(char *todmod) 1826 { 1827 if (modload("tod", todmod) == -1) 1828 halt("Can't load TOD module"); 1829 } 1830 1831 static void 1832 startup_end(void) 1833 { 1834 extern void setx86isalist(void); 1835 1836 PRM_POINT("startup_end() starting..."); 1837 1838 /* 1839 * Perform tasks that get done after most of the VM 1840 * initialization has been done but before the clock 1841 * and other devices get started. 1842 */ 1843 kern_setup1(); 1844 1845 /* 1846 * Perform CPC initialization for this CPU. 1847 */ 1848 kcpc_hw_init(CPU); 1849 1850 #if defined(OPTERON_WORKAROUND_6323525) 1851 if (opteron_workaround_6323525) 1852 patch_workaround_6323525(); 1853 #endif 1854 /* 1855 * If needed, load TOD module now so that ddi_get_time(9F) etc. work 1856 * (For now, "needed" is defined as set tod_module_name in /etc/system) 1857 */ 1858 if (tod_module_name != NULL) { 1859 PRM_POINT("load_tod_module()"); 1860 load_tod_module(tod_module_name); 1861 } 1862 1863 /* 1864 * Configure the system. 1865 */ 1866 PRM_POINT("Calling configure()..."); 1867 configure(); /* set up devices */ 1868 PRM_POINT("configure() done"); 1869 1870 /* 1871 * Set the isa_list string to the defined instruction sets we 1872 * support. 1873 */ 1874 setx86isalist(); 1875 cpu_intr_alloc(CPU, NINTR_THREADS); 1876 psm_install(); 1877 1878 /* 1879 * We're done with bootops. We don't unmap the bootstrap yet because 1880 * we're still using bootsvcs. 1881 */ 1882 PRM_POINT("NULLing out bootops"); 1883 *bootopsp = (struct bootops *)NULL; 1884 bootops = (struct bootops *)NULL; 1885 1886 PRM_POINT("Enabling interrupts"); 1887 (*picinitf)(); 1888 sti(); 1889 1890 (void) add_avsoftintr((void *)&softlevel1_hdl, 1, softlevel1, 1891 "softlevel1", NULL, NULL); /* XXX to be moved later */ 1892 1893 PRM_POINT("startup_end() done"); 1894 } 1895 1896 extern char hw_serial[]; 1897 char *_hs1107 = hw_serial; 1898 ulong_t _bdhs34; 1899 1900 void 1901 post_startup(void) 1902 { 1903 /* 1904 * Set the system wide, processor-specific flags to be passed 1905 * to userland via the aux vector for performance hints and 1906 * instruction set extensions. 1907 */ 1908 bind_hwcap(); 1909 1910 /* 1911 * Load the System Management BIOS into the global ksmbios 1912 * handle, if an SMBIOS is present on this system. 1913 */ 1914 ksmbios = smbios_open(NULL, SMB_VERSION, ksmbios_flags, NULL); 1915 1916 /* 1917 * Startup the memory scrubber. 1918 */ 1919 memscrub_init(); 1920 1921 /* 1922 * Complete CPU module initialization 1923 */ 1924 cmi_post_init(); 1925 1926 /* 1927 * Perform forceloading tasks for /etc/system. 1928 */ 1929 (void) mod_sysctl(SYS_FORCELOAD, NULL); 1930 1931 /* 1932 * ON4.0: Force /proc module in until clock interrupt handle fixed 1933 * ON4.0: This must be fixed or restated in /etc/systems. 1934 */ 1935 (void) modload("fs", "procfs"); 1936 1937 #if defined(__i386) 1938 /* 1939 * Check for required functional Floating Point hardware, 1940 * unless FP hardware explicitly disabled. 1941 */ 1942 if (fpu_exists && (fpu_pentium_fdivbug || fp_kind == FP_NO)) 1943 halt("No working FP hardware found"); 1944 #endif 1945 1946 maxmem = freemem; 1947 1948 add_cpunode2devtree(CPU->cpu_id, CPU->cpu_m.mcpu_cpi); 1949 } 1950 1951 static int 1952 pp_in_ramdisk(page_t *pp) 1953 { 1954 extern uint64_t ramdisk_start, ramdisk_end; 1955 1956 return ((pp->p_pagenum >= btop(ramdisk_start)) && 1957 (pp->p_pagenum < btopr(ramdisk_end))); 1958 } 1959 1960 void 1961 release_bootstrap(void) 1962 { 1963 int root_is_ramdisk; 1964 page_t *pp; 1965 extern void kobj_boot_unmountroot(void); 1966 extern dev_t rootdev; 1967 1968 /* unmount boot ramdisk and release kmem usage */ 1969 kobj_boot_unmountroot(); 1970 1971 /* 1972 * We're finished using the boot loader so free its pages. 1973 */ 1974 PRM_POINT("Unmapping lower boot pages"); 1975 clear_boot_mappings(0, _userlimit); 1976 postbootkernelbase = kernelbase; 1977 1978 /* 1979 * If root isn't on ramdisk, destroy the hardcoded 1980 * ramdisk node now and release the memory. Else, 1981 * ramdisk memory is kept in rd_pages. 1982 */ 1983 root_is_ramdisk = (getmajor(rootdev) == ddi_name_to_major("ramdisk")); 1984 if (!root_is_ramdisk) { 1985 dev_info_t *dip = ddi_find_devinfo("ramdisk", -1, 0); 1986 ASSERT(dip && ddi_get_parent(dip) == ddi_root_node()); 1987 ndi_rele_devi(dip); /* held from ddi_find_devinfo */ 1988 (void) ddi_remove_child(dip, 0); 1989 } 1990 1991 PRM_POINT("Releasing boot pages"); 1992 while (bootpages) { 1993 pp = bootpages; 1994 bootpages = pp->p_next; 1995 if (root_is_ramdisk && pp_in_ramdisk(pp)) { 1996 pp->p_next = rd_pages; 1997 rd_pages = pp; 1998 continue; 1999 } 2000 pp->p_next = (struct page *)0; 2001 pp->p_prev = (struct page *)0; 2002 PP_CLRBOOTPAGES(pp); 2003 page_free(pp, 1); 2004 } 2005 PRM_POINT("Boot pages released"); 2006 2007 /* 2008 * Find 1 page below 1 MB so that other processors can boot up. 2009 * Make sure it has a kernel VA as well as a 1:1 mapping. 2010 * We should have just free'd one up. 2011 */ 2012 if (use_mp) { 2013 pfn_t pfn; 2014 2015 for (pfn = 1; pfn < btop(1*1024*1024); pfn++) { 2016 if (page_numtopp_alloc(pfn) == NULL) 2017 continue; 2018 rm_platter_va = i86devmap(pfn, 1, 2019 PROT_READ | PROT_WRITE | PROT_EXEC); 2020 rm_platter_pa = ptob(pfn); 2021 hat_devload(kas.a_hat, 2022 (caddr_t)(uintptr_t)rm_platter_pa, MMU_PAGESIZE, 2023 pfn, PROT_READ | PROT_WRITE | PROT_EXEC, 2024 HAT_LOAD_NOCONSIST); 2025 break; 2026 } 2027 if (pfn == btop(1*1024*1024)) 2028 panic("No page available for starting " 2029 "other processors"); 2030 } 2031 2032 } 2033 2034 /* 2035 * Initialize the platform-specific parts of a page_t. 2036 */ 2037 void 2038 add_physmem_cb(page_t *pp, pfn_t pnum) 2039 { 2040 pp->p_pagenum = pnum; 2041 pp->p_mapping = NULL; 2042 pp->p_embed = 0; 2043 pp->p_share = 0; 2044 pp->p_mlentry = 0; 2045 } 2046 2047 /* 2048 * kphysm_init() initializes physical memory. 2049 */ 2050 static pgcnt_t 2051 kphysm_init( 2052 page_t *pp, 2053 pgcnt_t npages) 2054 { 2055 struct memlist *pmem; 2056 struct memseg *cur_memseg; 2057 pfn_t base_pfn; 2058 pgcnt_t num; 2059 pgcnt_t pages_done = 0; 2060 uint64_t addr; 2061 uint64_t size; 2062 extern pfn_t ddiphysmin; 2063 2064 ASSERT(page_hash != NULL && page_hashsz != 0); 2065 2066 cur_memseg = memseg_base; 2067 for (pmem = phys_avail; pmem && npages; pmem = pmem->next) { 2068 /* 2069 * In a 32 bit kernel can't use higher memory if we're 2070 * not booting in PAE mode. This check takes care of that. 2071 */ 2072 addr = pmem->address; 2073 size = pmem->size; 2074 if (btop(addr) > physmax) 2075 continue; 2076 2077 /* 2078 * align addr and size - they may not be at page boundaries 2079 */ 2080 if ((addr & MMU_PAGEOFFSET) != 0) { 2081 addr += MMU_PAGEOFFSET; 2082 addr &= ~(uint64_t)MMU_PAGEOFFSET; 2083 size -= addr - pmem->address; 2084 } 2085 2086 /* only process pages below or equal to physmax */ 2087 if ((btop(addr + size) - 1) > physmax) 2088 size = ptob(physmax - btop(addr) + 1); 2089 2090 num = btop(size); 2091 if (num == 0) 2092 continue; 2093 2094 if (num > npages) 2095 num = npages; 2096 2097 npages -= num; 2098 pages_done += num; 2099 base_pfn = btop(addr); 2100 2101 if (prom_debug) 2102 prom_printf("MEMSEG addr=0x%" PRIx64 2103 " pgs=0x%lx pfn 0x%lx-0x%lx\n", 2104 addr, num, base_pfn, base_pfn + num); 2105 2106 /* 2107 * Ignore pages below ddiphysmin to simplify ddi memory 2108 * allocation with non-zero addr_lo requests. 2109 */ 2110 if (base_pfn < ddiphysmin) { 2111 if (base_pfn + num <= ddiphysmin) 2112 continue; 2113 pp += (ddiphysmin - base_pfn); 2114 num -= (ddiphysmin - base_pfn); 2115 base_pfn = ddiphysmin; 2116 } 2117 2118 /* 2119 * Build the memsegs entry 2120 */ 2121 cur_memseg->pages = pp; 2122 cur_memseg->epages = pp + num; 2123 cur_memseg->pages_base = base_pfn; 2124 cur_memseg->pages_end = base_pfn + num; 2125 2126 /* 2127 * Insert into memseg list in decreasing pfn range order. 2128 * Low memory is typically more fragmented such that this 2129 * ordering keeps the larger ranges at the front of the list 2130 * for code that searches memseg. 2131 * This ASSERTS that the memsegs coming in from boot are in 2132 * increasing physical address order and not contiguous. 2133 */ 2134 if (memsegs != NULL) { 2135 ASSERT(cur_memseg->pages_base >= memsegs->pages_end); 2136 cur_memseg->next = memsegs; 2137 } 2138 memsegs = cur_memseg; 2139 2140 /* 2141 * add_physmem() initializes the PSM part of the page 2142 * struct by calling the PSM back with add_physmem_cb(). 2143 * In addition it coalesces pages into larger pages as 2144 * it initializes them. 2145 */ 2146 add_physmem(pp, num, base_pfn); 2147 cur_memseg++; 2148 availrmem_initial += num; 2149 availrmem += num; 2150 2151 pp += num; 2152 } 2153 2154 PRM_DEBUG(availrmem_initial); 2155 PRM_DEBUG(availrmem); 2156 PRM_DEBUG(freemem); 2157 build_pfn_hash(); 2158 return (pages_done); 2159 } 2160 2161 /* 2162 * Kernel VM initialization. 2163 */ 2164 static void 2165 kvm_init(void) 2166 { 2167 ASSERT((((uintptr_t)s_text) & MMU_PAGEOFFSET) == 0); 2168 2169 /* 2170 * Put the kernel segments in kernel address space. 2171 */ 2172 rw_enter(&kas.a_lock, RW_WRITER); 2173 as_avlinit(&kas); 2174 2175 (void) seg_attach(&kas, s_text, e_moddata - s_text, &ktextseg); 2176 (void) segkmem_create(&ktextseg); 2177 2178 (void) seg_attach(&kas, (caddr_t)valloc_base, valloc_sz, &kvalloc); 2179 (void) segkmem_create(&kvalloc); 2180 2181 /* 2182 * We're about to map out /boot. This is the beginning of the 2183 * system resource management transition. We can no longer 2184 * call into /boot for I/O or memory allocations. 2185 */ 2186 (void) seg_attach(&kas, final_kernelheap, 2187 ekernelheap - final_kernelheap, &kvseg); 2188 (void) segkmem_create(&kvseg); 2189 2190 if (core_size > 0) { 2191 PRM_POINT("attaching kvseg_core"); 2192 (void) seg_attach(&kas, (caddr_t)core_base, core_size, 2193 &kvseg_core); 2194 (void) segkmem_create(&kvseg_core); 2195 } 2196 2197 if (segziosize > 0) { 2198 PRM_POINT("attaching segzio"); 2199 (void) seg_attach(&kas, segzio_base, mmu_ptob(segziosize), 2200 &kzioseg); 2201 (void) segkmem_zio_create(&kzioseg); 2202 2203 /* create zio area covering new segment */ 2204 segkmem_zio_init(segzio_base, mmu_ptob(segziosize)); 2205 } 2206 2207 (void) seg_attach(&kas, kdi_segdebugbase, kdi_segdebugsize, &kdebugseg); 2208 (void) segkmem_create(&kdebugseg); 2209 2210 rw_exit(&kas.a_lock); 2211 2212 /* 2213 * Ensure that the red zone at kernelbase is never accessible. 2214 */ 2215 PRM_POINT("protecting redzone"); 2216 (void) as_setprot(&kas, (caddr_t)kernelbase, KERNEL_REDZONE_SIZE, 0); 2217 2218 /* 2219 * Make the text writable so that it can be hot patched by DTrace. 2220 */ 2221 (void) as_setprot(&kas, s_text, e_modtext - s_text, 2222 PROT_READ | PROT_WRITE | PROT_EXEC); 2223 2224 /* 2225 * Make data writable until end. 2226 */ 2227 (void) as_setprot(&kas, s_data, e_moddata - s_data, 2228 PROT_READ | PROT_WRITE | PROT_EXEC); 2229 } 2230 2231 /* 2232 * These are MTTR registers supported by P6 2233 */ 2234 static struct mtrrvar mtrrphys_arr[MAX_MTRRVAR]; 2235 static uint64_t mtrr64k, mtrr16k1, mtrr16k2; 2236 static uint64_t mtrr4k1, mtrr4k2, mtrr4k3; 2237 static uint64_t mtrr4k4, mtrr4k5, mtrr4k6; 2238 static uint64_t mtrr4k7, mtrr4k8, mtrrcap; 2239 uint64_t mtrrdef, pat_attr_reg; 2240 2241 /* 2242 * Disable reprogramming of MTRRs by default. 2243 */ 2244 int enable_relaxed_mtrr = 0; 2245 2246 void 2247 setup_mtrr(void) 2248 { 2249 int i, ecx; 2250 int vcnt; 2251 struct mtrrvar *mtrrphys; 2252 2253 if (!(x86_feature & X86_MTRR)) 2254 return; 2255 2256 mtrrcap = rdmsr(REG_MTRRCAP); 2257 mtrrdef = rdmsr(REG_MTRRDEF); 2258 if (mtrrcap & MTRRCAP_FIX) { 2259 mtrr64k = rdmsr(REG_MTRR64K); 2260 mtrr16k1 = rdmsr(REG_MTRR16K1); 2261 mtrr16k2 = rdmsr(REG_MTRR16K2); 2262 mtrr4k1 = rdmsr(REG_MTRR4K1); 2263 mtrr4k2 = rdmsr(REG_MTRR4K2); 2264 mtrr4k3 = rdmsr(REG_MTRR4K3); 2265 mtrr4k4 = rdmsr(REG_MTRR4K4); 2266 mtrr4k5 = rdmsr(REG_MTRR4K5); 2267 mtrr4k6 = rdmsr(REG_MTRR4K6); 2268 mtrr4k7 = rdmsr(REG_MTRR4K7); 2269 mtrr4k8 = rdmsr(REG_MTRR4K8); 2270 } 2271 if ((vcnt = (mtrrcap & MTRRCAP_VCNTMASK)) > MAX_MTRRVAR) 2272 vcnt = MAX_MTRRVAR; 2273 2274 for (i = 0, ecx = REG_MTRRPHYSBASE0, mtrrphys = mtrrphys_arr; 2275 i < vcnt - 1; i++, ecx += 2, mtrrphys++) { 2276 mtrrphys->mtrrphys_base = rdmsr(ecx); 2277 mtrrphys->mtrrphys_mask = rdmsr(ecx + 1); 2278 if ((x86_feature & X86_PAT) && enable_relaxed_mtrr) { 2279 mtrrphys->mtrrphys_mask &= ~MTRRPHYSMASK_V; 2280 } 2281 } 2282 if (x86_feature & X86_PAT) { 2283 if (enable_relaxed_mtrr) 2284 mtrrdef = MTRR_TYPE_WB|MTRRDEF_FE|MTRRDEF_E; 2285 pat_attr_reg = PAT_DEFAULT_ATTRIBUTE; 2286 } 2287 2288 mtrr_sync(); 2289 } 2290 2291 /* 2292 * Sync current cpu mtrr with the incore copy of mtrr. 2293 * This function has to be invoked with interrupts disabled 2294 * Currently we do not capture other cpu's. This is invoked on cpu0 2295 * just after reading /etc/system. 2296 * On other cpu's its invoked from mp_startup(). 2297 */ 2298 void 2299 mtrr_sync(void) 2300 { 2301 uint_t crvalue, cr0_orig; 2302 int vcnt, i, ecx; 2303 struct mtrrvar *mtrrphys; 2304 2305 cr0_orig = crvalue = getcr0(); 2306 crvalue |= CR0_CD; 2307 crvalue &= ~CR0_NW; 2308 setcr0(crvalue); 2309 invalidate_cache(); 2310 2311 reload_cr3(); 2312 if (x86_feature & X86_PAT) 2313 wrmsr(REG_MTRRPAT, pat_attr_reg); 2314 2315 wrmsr(REG_MTRRDEF, rdmsr(REG_MTRRDEF) & 2316 ~((uint64_t)(uintptr_t)MTRRDEF_E)); 2317 2318 if (mtrrcap & MTRRCAP_FIX) { 2319 wrmsr(REG_MTRR64K, mtrr64k); 2320 wrmsr(REG_MTRR16K1, mtrr16k1); 2321 wrmsr(REG_MTRR16K2, mtrr16k2); 2322 wrmsr(REG_MTRR4K1, mtrr4k1); 2323 wrmsr(REG_MTRR4K2, mtrr4k2); 2324 wrmsr(REG_MTRR4K3, mtrr4k3); 2325 wrmsr(REG_MTRR4K4, mtrr4k4); 2326 wrmsr(REG_MTRR4K5, mtrr4k5); 2327 wrmsr(REG_MTRR4K6, mtrr4k6); 2328 wrmsr(REG_MTRR4K7, mtrr4k7); 2329 wrmsr(REG_MTRR4K8, mtrr4k8); 2330 } 2331 if ((vcnt = (mtrrcap & MTRRCAP_VCNTMASK)) > MAX_MTRRVAR) 2332 vcnt = MAX_MTRRVAR; 2333 for (i = 0, ecx = REG_MTRRPHYSBASE0, mtrrphys = mtrrphys_arr; 2334 i < vcnt - 1; i++, ecx += 2, mtrrphys++) { 2335 wrmsr(ecx, mtrrphys->mtrrphys_base); 2336 wrmsr(ecx + 1, mtrrphys->mtrrphys_mask); 2337 } 2338 wrmsr(REG_MTRRDEF, mtrrdef); 2339 2340 reload_cr3(); 2341 invalidate_cache(); 2342 setcr0(cr0_orig); 2343 } 2344 2345 /* 2346 * resync mtrr so that BIOS is happy. Called from mdboot 2347 */ 2348 void 2349 mtrr_resync(void) 2350 { 2351 if ((x86_feature & X86_PAT) && enable_relaxed_mtrr) { 2352 /* 2353 * We could have changed the default mtrr definition. 2354 * Put it back to uncached which is what it is at power on 2355 */ 2356 mtrrdef = MTRR_TYPE_UC|MTRRDEF_FE|MTRRDEF_E; 2357 mtrr_sync(); 2358 } 2359 } 2360 2361 void 2362 get_system_configuration(void) 2363 { 2364 char prop[32]; 2365 u_longlong_t nodes_ll, cpus_pernode_ll, lvalue; 2366 2367 if (((BOP_GETPROPLEN(bootops, "nodes") > sizeof (prop)) || 2368 (BOP_GETPROP(bootops, "nodes", prop) < 0) || 2369 (kobj_getvalue(prop, &nodes_ll) == -1) || 2370 (nodes_ll > MAXNODES)) || 2371 ((BOP_GETPROPLEN(bootops, "cpus_pernode") > sizeof (prop)) || 2372 (BOP_GETPROP(bootops, "cpus_pernode", prop) < 0) || 2373 (kobj_getvalue(prop, &cpus_pernode_ll) == -1))) { 2374 2375 system_hardware.hd_nodes = 1; 2376 system_hardware.hd_cpus_per_node = 0; 2377 } else { 2378 system_hardware.hd_nodes = (int)nodes_ll; 2379 system_hardware.hd_cpus_per_node = (int)cpus_pernode_ll; 2380 } 2381 if ((BOP_GETPROPLEN(bootops, "kernelbase") > sizeof (prop)) || 2382 (BOP_GETPROP(bootops, "kernelbase", prop) < 0) || 2383 (kobj_getvalue(prop, &lvalue) == -1)) 2384 eprom_kernelbase = NULL; 2385 else 2386 eprom_kernelbase = (uintptr_t)lvalue; 2387 2388 if ((BOP_GETPROPLEN(bootops, "segmapsize") > sizeof (prop)) || 2389 (BOP_GETPROP(bootops, "segmapsize", prop) < 0) || 2390 (kobj_getvalue(prop, &lvalue) == -1)) { 2391 segmapsize = SEGMAPDEFAULT; 2392 } else { 2393 segmapsize = (uintptr_t)lvalue; 2394 } 2395 2396 if ((BOP_GETPROPLEN(bootops, "segmapfreelists") > sizeof (prop)) || 2397 (BOP_GETPROP(bootops, "segmapfreelists", prop) < 0) || 2398 (kobj_getvalue(prop, &lvalue) == -1)) { 2399 segmapfreelists = 0; /* use segmap driver default */ 2400 } else { 2401 segmapfreelists = (int)lvalue; 2402 } 2403 2404 /* physmem used to be here, but moved much earlier to fakebop.c */ 2405 } 2406 2407 /* 2408 * Add to a memory list. 2409 * start = start of new memory segment 2410 * len = length of new memory segment in bytes 2411 * new = pointer to a new struct memlist 2412 * memlistp = memory list to which to add segment. 2413 */ 2414 void 2415 memlist_add( 2416 uint64_t start, 2417 uint64_t len, 2418 struct memlist *new, 2419 struct memlist **memlistp) 2420 { 2421 struct memlist *cur; 2422 uint64_t end = start + len; 2423 2424 new->address = start; 2425 new->size = len; 2426 2427 cur = *memlistp; 2428 2429 while (cur) { 2430 if (cur->address >= end) { 2431 new->next = cur; 2432 *memlistp = new; 2433 new->prev = cur->prev; 2434 cur->prev = new; 2435 return; 2436 } 2437 ASSERT(cur->address + cur->size <= start); 2438 if (cur->next == NULL) { 2439 cur->next = new; 2440 new->prev = cur; 2441 new->next = NULL; 2442 return; 2443 } 2444 memlistp = &cur->next; 2445 cur = cur->next; 2446 } 2447 } 2448 2449 void 2450 kobj_vmem_init(vmem_t **text_arena, vmem_t **data_arena) 2451 { 2452 size_t tsize = e_modtext - modtext; 2453 size_t dsize = e_moddata - moddata; 2454 2455 *text_arena = vmem_create("module_text", tsize ? modtext : NULL, tsize, 2456 1, segkmem_alloc, segkmem_free, heaptext_arena, 0, VM_SLEEP); 2457 *data_arena = vmem_create("module_data", dsize ? moddata : NULL, dsize, 2458 1, segkmem_alloc, segkmem_free, heap32_arena, 0, VM_SLEEP); 2459 } 2460 2461 caddr_t 2462 kobj_text_alloc(vmem_t *arena, size_t size) 2463 { 2464 return (vmem_alloc(arena, size, VM_SLEEP | VM_BESTFIT)); 2465 } 2466 2467 /*ARGSUSED*/ 2468 caddr_t 2469 kobj_texthole_alloc(caddr_t addr, size_t size) 2470 { 2471 panic("unexpected call to kobj_texthole_alloc()"); 2472 /*NOTREACHED*/ 2473 return (0); 2474 } 2475 2476 /*ARGSUSED*/ 2477 void 2478 kobj_texthole_free(caddr_t addr, size_t size) 2479 { 2480 panic("unexpected call to kobj_texthole_free()"); 2481 } 2482 2483 /* 2484 * This is called just after configure() in startup(). 2485 * 2486 * The ISALIST concept is a bit hopeless on Intel, because 2487 * there's no guarantee of an ever-more-capable processor 2488 * given that various parts of the instruction set may appear 2489 * and disappear between different implementations. 2490 * 2491 * While it would be possible to correct it and even enhance 2492 * it somewhat, the explicit hardware capability bitmask allows 2493 * more flexibility. 2494 * 2495 * So, we just leave this alone. 2496 */ 2497 void 2498 setx86isalist(void) 2499 { 2500 char *tp; 2501 size_t len; 2502 extern char *isa_list; 2503 2504 #define TBUFSIZE 1024 2505 2506 tp = kmem_alloc(TBUFSIZE, KM_SLEEP); 2507 *tp = '\0'; 2508 2509 #if defined(__amd64) 2510 (void) strcpy(tp, "amd64 "); 2511 #endif 2512 2513 switch (x86_vendor) { 2514 case X86_VENDOR_Intel: 2515 case X86_VENDOR_AMD: 2516 case X86_VENDOR_TM: 2517 if (x86_feature & X86_CMOV) { 2518 /* 2519 * Pentium Pro or later 2520 */ 2521 (void) strcat(tp, "pentium_pro"); 2522 (void) strcat(tp, x86_feature & X86_MMX ? 2523 "+mmx pentium_pro " : " "); 2524 } 2525 /*FALLTHROUGH*/ 2526 case X86_VENDOR_Cyrix: 2527 /* 2528 * The Cyrix 6x86 does not have any Pentium features 2529 * accessible while not at privilege level 0. 2530 */ 2531 if (x86_feature & X86_CPUID) { 2532 (void) strcat(tp, "pentium"); 2533 (void) strcat(tp, x86_feature & X86_MMX ? 2534 "+mmx pentium " : " "); 2535 } 2536 break; 2537 default: 2538 break; 2539 } 2540 (void) strcat(tp, "i486 i386 i86"); 2541 len = strlen(tp) + 1; /* account for NULL at end of string */ 2542 isa_list = strcpy(kmem_alloc(len, KM_SLEEP), tp); 2543 kmem_free(tp, TBUFSIZE); 2544 2545 #undef TBUFSIZE 2546 } 2547 2548 2549 #ifdef __amd64 2550 2551 void * 2552 device_arena_alloc(size_t size, int vm_flag) 2553 { 2554 return (vmem_alloc(device_arena, size, vm_flag)); 2555 } 2556 2557 void 2558 device_arena_free(void *vaddr, size_t size) 2559 { 2560 vmem_free(device_arena, vaddr, size); 2561 } 2562 2563 #else /* __i386 */ 2564 2565 void * 2566 device_arena_alloc(size_t size, int vm_flag) 2567 { 2568 caddr_t vaddr; 2569 uintptr_t v; 2570 size_t start; 2571 size_t end; 2572 2573 vaddr = vmem_alloc(heap_arena, size, vm_flag); 2574 if (vaddr == NULL) 2575 return (NULL); 2576 2577 v = (uintptr_t)vaddr; 2578 ASSERT(v >= kernelbase); 2579 ASSERT(v + size <= valloc_base); 2580 2581 start = btop(v - kernelbase); 2582 end = btop(v + size - 1 - kernelbase); 2583 ASSERT(start < toxic_bit_map_len); 2584 ASSERT(end < toxic_bit_map_len); 2585 2586 while (start <= end) { 2587 BT_ATOMIC_SET(toxic_bit_map, start); 2588 ++start; 2589 } 2590 return (vaddr); 2591 } 2592 2593 void 2594 device_arena_free(void *vaddr, size_t size) 2595 { 2596 uintptr_t v = (uintptr_t)vaddr; 2597 size_t start; 2598 size_t end; 2599 2600 ASSERT(v >= kernelbase); 2601 ASSERT(v + size <= valloc_base); 2602 2603 start = btop(v - kernelbase); 2604 end = btop(v + size - 1 - kernelbase); 2605 ASSERT(start < toxic_bit_map_len); 2606 ASSERT(end < toxic_bit_map_len); 2607 2608 while (start <= end) { 2609 ASSERT(BT_TEST(toxic_bit_map, start) != 0); 2610 BT_ATOMIC_CLEAR(toxic_bit_map, start); 2611 ++start; 2612 } 2613 vmem_free(heap_arena, vaddr, size); 2614 } 2615 2616 /* 2617 * returns 1st address in range that is in device arena, or NULL 2618 * if len is not NULL it returns the length of the toxic range 2619 */ 2620 void * 2621 device_arena_contains(void *vaddr, size_t size, size_t *len) 2622 { 2623 uintptr_t v = (uintptr_t)vaddr; 2624 uintptr_t eaddr = v + size; 2625 size_t start; 2626 size_t end; 2627 2628 /* 2629 * if called very early by kmdb, just return NULL 2630 */ 2631 if (toxic_bit_map == NULL) 2632 return (NULL); 2633 2634 /* 2635 * First check if we're completely outside the bitmap range. 2636 */ 2637 if (v >= valloc_base || eaddr < kernelbase) 2638 return (NULL); 2639 2640 /* 2641 * Trim ends of search to look at only what the bitmap covers. 2642 */ 2643 if (v < kernelbase) 2644 v = kernelbase; 2645 start = btop(v - kernelbase); 2646 end = btop(eaddr - kernelbase); 2647 if (end >= toxic_bit_map_len) 2648 end = toxic_bit_map_len; 2649 2650 if (bt_range(toxic_bit_map, &start, &end, end) == 0) 2651 return (NULL); 2652 2653 v = kernelbase + ptob(start); 2654 if (len != NULL) 2655 *len = ptob(end - start); 2656 return ((void *)v); 2657 } 2658 2659 #endif /* __i386 */ 2660