1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 /* 22 * Copyright 2007 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 24 */ 25 26 #pragma ident "%Z%%M% %I% %E% SMI" 27 28 #include <sys/types.h> 29 #include <sys/t_lock.h> 30 #include <sys/param.h> 31 #include <sys/sysmacros.h> 32 #include <sys/signal.h> 33 #include <sys/systm.h> 34 #include <sys/user.h> 35 #include <sys/mman.h> 36 #include <sys/vm.h> 37 #include <sys/conf.h> 38 #include <sys/avintr.h> 39 #include <sys/autoconf.h> 40 #include <sys/disp.h> 41 #include <sys/class.h> 42 #include <sys/bitmap.h> 43 44 #include <sys/privregs.h> 45 46 #include <sys/proc.h> 47 #include <sys/buf.h> 48 #include <sys/kmem.h> 49 #include <sys/mem.h> 50 #include <sys/kstat.h> 51 52 #include <sys/reboot.h> 53 54 #include <sys/cred.h> 55 #include <sys/vnode.h> 56 #include <sys/file.h> 57 58 #include <sys/procfs.h> 59 60 #include <sys/vfs.h> 61 #include <sys/cmn_err.h> 62 #include <sys/utsname.h> 63 #include <sys/debug.h> 64 #include <sys/kdi.h> 65 66 #include <sys/dumphdr.h> 67 #include <sys/bootconf.h> 68 #include <sys/varargs.h> 69 #include <sys/promif.h> 70 #include <sys/modctl.h> 71 72 #include <sys/sunddi.h> 73 #include <sys/sunndi.h> 74 #include <sys/ndi_impldefs.h> 75 #include <sys/ddidmareq.h> 76 #include <sys/psw.h> 77 #include <sys/regset.h> 78 #include <sys/clock.h> 79 #include <sys/pte.h> 80 #include <sys/tss.h> 81 #include <sys/stack.h> 82 #include <sys/trap.h> 83 #include <sys/fp.h> 84 #include <vm/kboot_mmu.h> 85 #include <vm/anon.h> 86 #include <vm/as.h> 87 #include <vm/page.h> 88 #include <vm/seg.h> 89 #include <vm/seg_dev.h> 90 #include <vm/seg_kmem.h> 91 #include <vm/seg_kpm.h> 92 #include <vm/seg_map.h> 93 #include <vm/seg_vn.h> 94 #include <vm/seg_kp.h> 95 #include <sys/memnode.h> 96 #include <vm/vm_dep.h> 97 #include <sys/thread.h> 98 #include <sys/sysconf.h> 99 #include <sys/vm_machparam.h> 100 #include <sys/archsystm.h> 101 #include <sys/machsystm.h> 102 #include <vm/hat.h> 103 #include <vm/hat_i86.h> 104 #include <sys/pmem.h> 105 #include <sys/smp_impldefs.h> 106 #include <sys/x86_archext.h> 107 #include <sys/segments.h> 108 #include <sys/clconf.h> 109 #include <sys/kobj.h> 110 #include <sys/kobj_lex.h> 111 #include <sys/cpc_impl.h> 112 #include <sys/x86_archext.h> 113 #include <sys/cpu_module.h> 114 #include <sys/smbios.h> 115 #include <sys/debug_info.h> 116 #include <sys/bootinfo.h> 117 #include <sys/ddi_timer.h> 118 119 #ifdef __xpv 120 121 #include <sys/hypervisor.h> 122 #include <sys/xen_mmu.h> 123 #include <sys/evtchn_impl.h> 124 #include <sys/gnttab.h> 125 #include <sys/xpv_panic.h> 126 #include <xen/sys/xenbus_comms.h> 127 #include <xen/public/physdev.h> 128 129 extern void xen_late_startup(void); 130 131 struct xen_evt_data cpu0_evt_data; 132 133 #endif /* __xpv */ 134 135 extern void progressbar_init(void); 136 extern void progressbar_start(void); 137 extern void brand_init(void); 138 139 extern int size_pse_array(pgcnt_t, int); 140 141 /* 142 * XXX make declaration below "static" when drivers no longer use this 143 * interface. 144 */ 145 extern caddr_t p0_va; /* Virtual address for accessing physical page 0 */ 146 147 /* 148 * segkp 149 */ 150 extern int segkp_fromheap; 151 152 static void kvm_init(void); 153 static void startup_init(void); 154 static void startup_memlist(void); 155 static void startup_kmem(void); 156 static void startup_modules(void); 157 static void startup_vm(void); 158 static void startup_end(void); 159 static void layout_kernel_va(void); 160 161 /* 162 * Declare these as initialized data so we can patch them. 163 */ 164 #ifdef __i386 165 166 /* 167 * Due to virtual address space limitations running in 32 bit mode, restrict 168 * the amount of physical memory configured to a max of PHYSMEM pages (16g). 169 * 170 * If the physical max memory size of 64g were allowed to be configured, the 171 * size of user virtual address space will be less than 1g. A limited user 172 * address space greatly reduces the range of applications that can run. 173 * 174 * If more physical memory than PHYSMEM is required, users should preferably 175 * run in 64 bit mode which has far looser virtual address space limitations. 176 * 177 * If 64 bit mode is not available (as in IA32) and/or more physical memory 178 * than PHYSMEM is required in 32 bit mode, physmem can be set to the desired 179 * value or to 0 (to configure all available memory) via eeprom(1M). kernelbase 180 * should also be carefully tuned to balance out the need of the user 181 * application while minimizing the risk of kernel heap exhaustion due to 182 * kernelbase being set too high. 183 */ 184 #define PHYSMEM 0x400000 185 186 #else /* __amd64 */ 187 188 /* 189 * For now we can handle memory with physical addresses up to about 190 * 64 Terabytes. This keeps the kernel above the VA hole, leaving roughly 191 * half the VA space for seg_kpm. When systems get bigger than 64TB this 192 * code will need revisiting. There is an implicit assumption that there 193 * are no *huge* holes in the physical address space too. 194 */ 195 #define TERABYTE (1ul << 40) 196 #define PHYSMEM_MAX64 mmu_btop(64 * TERABYTE) 197 #define PHYSMEM PHYSMEM_MAX64 198 #define AMD64_VA_HOLE_END 0xFFFF800000000000ul 199 200 #endif /* __amd64 */ 201 202 pgcnt_t physmem = PHYSMEM; 203 pgcnt_t obp_pages; /* Memory used by PROM for its text and data */ 204 205 char *kobj_file_buf; 206 int kobj_file_bufsize; /* set in /etc/system */ 207 208 /* Global variables for MP support. Used in mp_startup */ 209 caddr_t rm_platter_va; 210 uint32_t rm_platter_pa; 211 212 int auto_lpg_disable = 1; 213 214 /* 215 * Some CPUs have holes in the middle of the 64-bit virtual address range. 216 */ 217 uintptr_t hole_start, hole_end; 218 219 /* 220 * kpm mapping window 221 */ 222 caddr_t kpm_vbase; 223 size_t kpm_size; 224 static int kpm_desired; 225 #ifdef __amd64 226 static uintptr_t segkpm_base = (uintptr_t)SEGKPM_BASE; 227 #endif 228 229 /* 230 * Configuration parameters set at boot time. 231 */ 232 233 caddr_t econtig; /* end of first block of contiguous kernel */ 234 235 struct bootops *bootops = 0; /* passed in from boot */ 236 struct bootops **bootopsp; 237 struct boot_syscalls *sysp; /* passed in from boot */ 238 239 char bootblock_fstype[16]; 240 241 char kern_bootargs[OBP_MAXPATHLEN]; 242 243 /* 244 * ZFS zio segment. This allows us to exclude large portions of ZFS data that 245 * gets cached in kmem caches on the heap. If this is set to zero, we allocate 246 * zio buffers from their own segment, otherwise they are allocated from the 247 * heap. The optimization of allocating zio buffers from their own segment is 248 * only valid on 64-bit kernels. 249 */ 250 #if defined(__amd64) 251 int segzio_fromheap = 0; 252 #else 253 int segzio_fromheap = 1; 254 #endif 255 256 /* 257 * new memory fragmentations are possible in startup() due to BOP_ALLOCs. this 258 * depends on number of BOP_ALLOC calls made and requested size, memory size 259 * combination and whether boot.bin memory needs to be freed. 260 */ 261 #define POSS_NEW_FRAGMENTS 12 262 263 /* 264 * VM data structures 265 */ 266 long page_hashsz; /* Size of page hash table (power of two) */ 267 struct page *pp_base; /* Base of initial system page struct array */ 268 struct page **page_hash; /* Page hash table */ 269 pad_mutex_t *pse_mutex; /* Locks protecting pp->p_selock */ 270 size_t pse_table_size; /* Number of mutexes in pse_mutex[] */ 271 int pse_shift; /* log2(pse_table_size) */ 272 struct seg ktextseg; /* Segment used for kernel executable image */ 273 struct seg kvalloc; /* Segment used for "valloc" mapping */ 274 struct seg kpseg; /* Segment used for pageable kernel virt mem */ 275 struct seg kmapseg; /* Segment used for generic kernel mappings */ 276 struct seg kdebugseg; /* Segment used for the kernel debugger */ 277 278 struct seg *segkmap = &kmapseg; /* Kernel generic mapping segment */ 279 static struct seg *segmap = &kmapseg; /* easier to use name for in here */ 280 281 struct seg *segkp = &kpseg; /* Pageable kernel virtual memory segment */ 282 283 #if defined(__amd64) 284 struct seg kvseg_core; /* Segment used for the core heap */ 285 struct seg kpmseg; /* Segment used for physical mapping */ 286 struct seg *segkpm = &kpmseg; /* 64bit kernel physical mapping segment */ 287 #else 288 struct seg *segkpm = NULL; /* Unused on IA32 */ 289 #endif 290 291 caddr_t segkp_base; /* Base address of segkp */ 292 caddr_t segzio_base; /* Base address of segzio */ 293 #if defined(__amd64) 294 pgcnt_t segkpsize = btop(SEGKPDEFSIZE); /* size of segkp segment in pages */ 295 #else 296 pgcnt_t segkpsize = 0; 297 #endif 298 pgcnt_t segziosize = 0; /* size of zio segment in pages */ 299 300 /* 301 * VA range available to the debugger 302 */ 303 const caddr_t kdi_segdebugbase = (const caddr_t)SEGDEBUGBASE; 304 const size_t kdi_segdebugsize = SEGDEBUGSIZE; 305 306 struct memseg *memseg_base; 307 struct vnode unused_pages_vp; 308 309 #define FOURGB 0x100000000LL 310 311 struct memlist *memlist; 312 313 caddr_t s_text; /* start of kernel text segment */ 314 caddr_t e_text; /* end of kernel text segment */ 315 caddr_t s_data; /* start of kernel data segment */ 316 caddr_t e_data; /* end of kernel data segment */ 317 caddr_t modtext; /* start of loadable module text reserved */ 318 caddr_t e_modtext; /* end of loadable module text reserved */ 319 caddr_t moddata; /* start of loadable module data reserved */ 320 caddr_t e_moddata; /* end of loadable module data reserved */ 321 322 struct memlist *phys_install; /* Total installed physical memory */ 323 struct memlist *phys_avail; /* Total available physical memory */ 324 325 /* 326 * kphysm_init returns the number of pages that were processed 327 */ 328 static pgcnt_t kphysm_init(page_t *, pgcnt_t); 329 330 #define IO_PROP_SIZE 64 /* device property size */ 331 332 /* 333 * a couple useful roundup macros 334 */ 335 #define ROUND_UP_PAGE(x) \ 336 ((uintptr_t)P2ROUNDUP((uintptr_t)(x), (uintptr_t)MMU_PAGESIZE)) 337 #define ROUND_UP_LPAGE(x) \ 338 ((uintptr_t)P2ROUNDUP((uintptr_t)(x), mmu.level_size[1])) 339 #define ROUND_UP_4MEG(x) \ 340 ((uintptr_t)P2ROUNDUP((uintptr_t)(x), (uintptr_t)FOUR_MEG)) 341 #define ROUND_UP_TOPLEVEL(x) \ 342 ((uintptr_t)P2ROUNDUP((uintptr_t)(x), mmu.level_size[mmu.max_level])) 343 344 /* 345 * 32-bit Kernel's Virtual memory layout. 346 * +-----------------------+ 347 * | | 348 * 0xFFC00000 -|-----------------------|- ARGSBASE 349 * | debugger | 350 * 0xFF800000 -|-----------------------|- SEGDEBUGBASE 351 * | Kernel Data | 352 * 0xFEC00000 -|-----------------------| 353 * | Kernel Text | 354 * 0xFE800000 -|-----------------------|- KERNEL_TEXT (0xFB400000 on Xen) 355 * |--- GDT ---|- GDT page (GDT_VA) 356 * |--- debug info ---|- debug info (DEBUG_INFO_VA) 357 * | | 358 * | page_t structures | 359 * | memsegs, memlists, | 360 * | page hash, etc. | 361 * --- -|-----------------------|- ekernelheap, valloc_base (floating) 362 * | | (segkp is just an arena in the heap) 363 * | | 364 * | kvseg | 365 * | | 366 * | | 367 * --- -|-----------------------|- kernelheap (floating) 368 * | Segkmap | 369 * 0xC3002000 -|-----------------------|- segmap_start (floating) 370 * | Red Zone | 371 * 0xC3000000 -|-----------------------|- kernelbase / userlimit (floating) 372 * | | || 373 * | Shared objects | \/ 374 * | | 375 * : : 376 * | user data | 377 * |-----------------------| 378 * | user text | 379 * 0x08048000 -|-----------------------| 380 * | user stack | 381 * : : 382 * | invalid | 383 * 0x00000000 +-----------------------+ 384 * 385 * 386 * 64-bit Kernel's Virtual memory layout. (assuming 64 bit app) 387 * +-----------------------+ 388 * | | 389 * 0xFFFFFFFF.FFC00000 |-----------------------|- ARGSBASE 390 * | debugger (?) | 391 * 0xFFFFFFFF.FF800000 |-----------------------|- SEGDEBUGBASE 392 * | unused | 393 * +-----------------------+ 394 * | Kernel Data | 395 * 0xFFFFFFFF.FBC00000 |-----------------------| 396 * | Kernel Text | 397 * 0xFFFFFFFF.FB800000 |-----------------------|- KERNEL_TEXT 398 * |--- GDT ---|- GDT page (GDT_VA) 399 * |--- debug info ---|- debug info (DEBUG_INFO_VA) 400 * | | 401 * | Core heap | (used for loadable modules) 402 * 0xFFFFFFFF.C0000000 |-----------------------|- core_base / ekernelheap 403 * | Kernel | 404 * | heap | 405 * 0xFFFFFXXX.XXX00000 |-----------------------|- kernelheap (floating) 406 * | segmap | 407 * 0xFFFFFXXX.XXX00000 |-----------------------|- segmap_start (floating) 408 * | device mappings | 409 * 0xFFFFFXXX.XXX00000 |-----------------------|- toxic_addr (floating) 410 * | segzio | 411 * 0xFFFFFXXX.XXX00000 |-----------------------|- segzio_base (floating) 412 * | segkp | 413 * --- |-----------------------|- segkp_base (floating) 414 * | page_t structures | valloc_base + valloc_sz 415 * | memsegs, memlists, | 416 * | page hash, etc. | 417 * 0xFFFFFF00.00000000 |-----------------------|- valloc_base (lower if > 1TB) 418 * | segkpm | 419 * 0xFFFFFE00.00000000 |-----------------------| 420 * | Red Zone | 421 * 0xFFFFFD80.00000000 |-----------------------|- KERNELBASE (lower if > 1TB) 422 * | User stack |- User space memory 423 * | | 424 * | shared objects, etc | (grows downwards) 425 * : : 426 * | | 427 * 0xFFFF8000.00000000 |-----------------------| 428 * | | 429 * | VA Hole / unused | 430 * | | 431 * 0x00008000.00000000 |-----------------------| 432 * | | 433 * | | 434 * : : 435 * | user heap | (grows upwards) 436 * | | 437 * | user data | 438 * |-----------------------| 439 * | user text | 440 * 0x00000000.04000000 |-----------------------| 441 * | invalid | 442 * 0x00000000.00000000 +-----------------------+ 443 * 444 * A 32 bit app on the 64 bit kernel sees the same layout as on the 32 bit 445 * kernel, except that userlimit is raised to 0xfe000000 446 * 447 * Floating values: 448 * 449 * valloc_base: start of the kernel's memory management/tracking data 450 * structures. This region contains page_t structures for 451 * physical memory, memsegs, memlists, and the page hash. 452 * 453 * core_base: start of the kernel's "core" heap area on 64-bit systems. 454 * This area is intended to be used for global data as well as for module 455 * text/data that does not fit into the nucleus pages. The core heap is 456 * restricted to a 2GB range, allowing every address within it to be 457 * accessed using rip-relative addressing 458 * 459 * ekernelheap: end of kernelheap and start of segmap. 460 * 461 * kernelheap: start of kernel heap. On 32-bit systems, this starts right 462 * above a red zone that separates the user's address space from the 463 * kernel's. On 64-bit systems, it sits above segkp and segkpm. 464 * 465 * segmap_start: start of segmap. The length of segmap can be modified 466 * by changing segmapsize in /etc/system (preferred) or eeprom (deprecated). 467 * The default length is 16MB on 32-bit systems and 64MB on 64-bit systems. 468 * 469 * kernelbase: On a 32-bit kernel the default value of 0xd4000000 will be 470 * decreased by 2X the size required for page_t. This allows the kernel 471 * heap to grow in size with physical memory. With sizeof(page_t) == 80 472 * bytes, the following shows the values of kernelbase and kernel heap 473 * sizes for different memory configurations (assuming default segmap and 474 * segkp sizes). 475 * 476 * mem size for kernelbase kernel heap 477 * size page_t's size 478 * ---- --------- ---------- ----------- 479 * 1gb 0x01400000 0xd1800000 684MB 480 * 2gb 0x02800000 0xcf000000 704MB 481 * 4gb 0x05000000 0xca000000 744MB 482 * 6gb 0x07800000 0xc5000000 784MB 483 * 8gb 0x0a000000 0xc0000000 824MB 484 * 16gb 0x14000000 0xac000000 984MB 485 * 32gb 0x28000000 0x84000000 1304MB 486 * 64gb 0x50000000 0x34000000 1944MB (*) 487 * 488 * kernelbase is less than the abi minimum of 0xc0000000 for memory 489 * configurations above 8gb. 490 * 491 * (*) support for memory configurations above 32gb will require manual tuning 492 * of kernelbase to balance out the need of user applications. 493 */ 494 495 /* real-time-clock initialization parameters */ 496 extern time_t process_rtc_config_file(void); 497 498 uintptr_t kernelbase; 499 uintptr_t postbootkernelbase; /* not set till boot loader is gone */ 500 uintptr_t eprom_kernelbase; 501 size_t segmapsize; 502 uintptr_t segmap_start; 503 int segmapfreelists; 504 pgcnt_t npages; 505 pgcnt_t orig_npages; 506 size_t core_size; /* size of "core" heap */ 507 uintptr_t core_base; /* base address of "core" heap */ 508 509 /* 510 * List of bootstrap pages. We mark these as allocated in startup. 511 * release_bootstrap() will free them when we're completely done with 512 * the bootstrap. 513 */ 514 static page_t *bootpages; 515 516 /* 517 * boot time pages that have a vnode from the ramdisk will keep that forever. 518 */ 519 static page_t *rd_pages; 520 521 struct system_hardware system_hardware; 522 523 /* 524 * Enable some debugging messages concerning memory usage... 525 */ 526 static void 527 print_memlist(char *title, struct memlist *mp) 528 { 529 prom_printf("MEMLIST: %s:\n", title); 530 while (mp != NULL) { 531 prom_printf("\tAddress 0x%" PRIx64 ", size 0x%" PRIx64 "\n", 532 mp->address, mp->size); 533 mp = mp->next; 534 } 535 } 536 537 /* 538 * XX64 need a comment here.. are these just default values, surely 539 * we read the "cpuid" type information to figure this out. 540 */ 541 int l2cache_sz = 0x80000; 542 int l2cache_linesz = 0x40; 543 int l2cache_assoc = 1; 544 545 static size_t textrepl_min_gb = 10; 546 547 /* 548 * on 64 bit we use a predifined VA range for mapping devices in the kernel 549 * on 32 bit the mappings are intermixed in the heap, so we use a bit map 550 */ 551 #ifdef __amd64 552 553 vmem_t *device_arena; 554 uintptr_t toxic_addr = (uintptr_t)NULL; 555 size_t toxic_size = 1024 * 1024 * 1024; /* Sparc uses 1 gig too */ 556 557 #else /* __i386 */ 558 559 ulong_t *toxic_bit_map; /* one bit for each 4k of VA in heap_arena */ 560 size_t toxic_bit_map_len = 0; /* in bits */ 561 562 #endif /* __i386 */ 563 564 /* 565 * Simple boot time debug facilities 566 */ 567 static char *prm_dbg_str[] = { 568 "%s:%d: '%s' is 0x%x\n", 569 "%s:%d: '%s' is 0x%llx\n" 570 }; 571 572 int prom_debug; 573 574 #define PRM_DEBUG(q) if (prom_debug) \ 575 prom_printf(prm_dbg_str[sizeof (q) >> 3], "startup.c", __LINE__, #q, q); 576 #define PRM_POINT(q) if (prom_debug) \ 577 prom_printf("%s:%d: %s\n", "startup.c", __LINE__, q); 578 579 /* 580 * This structure is used to keep track of the intial allocations 581 * done in startup_memlist(). The value of NUM_ALLOCATIONS needs to 582 * be >= the number of ADD_TO_ALLOCATIONS() executed in the code. 583 */ 584 #define NUM_ALLOCATIONS 7 585 int num_allocations = 0; 586 struct { 587 void **al_ptr; 588 size_t al_size; 589 } allocations[NUM_ALLOCATIONS]; 590 size_t valloc_sz = 0; 591 uintptr_t valloc_base; 592 593 #define ADD_TO_ALLOCATIONS(ptr, size) { \ 594 size = ROUND_UP_PAGE(size); \ 595 if (num_allocations == NUM_ALLOCATIONS) \ 596 panic("too many ADD_TO_ALLOCATIONS()"); \ 597 allocations[num_allocations].al_ptr = (void**)&ptr; \ 598 allocations[num_allocations].al_size = size; \ 599 valloc_sz += size; \ 600 ++num_allocations; \ 601 } 602 603 /* 604 * Allocate all the initial memory needed by the page allocator. 605 */ 606 static void 607 perform_allocations(void) 608 { 609 caddr_t mem; 610 int i; 611 int valloc_align; 612 613 PRM_DEBUG(valloc_base); 614 PRM_DEBUG(valloc_sz); 615 valloc_align = mmu.level_size[mmu.max_page_level > 0]; 616 mem = BOP_ALLOC(bootops, (caddr_t)valloc_base, valloc_sz, valloc_align); 617 if (mem != (caddr_t)valloc_base) 618 panic("BOP_ALLOC() failed"); 619 bzero(mem, valloc_sz); 620 for (i = 0; i < num_allocations; ++i) { 621 *allocations[i].al_ptr = (void *)mem; 622 mem += allocations[i].al_size; 623 } 624 } 625 626 /* 627 * Our world looks like this at startup time. 628 * 629 * In a 32-bit OS, boot loads the kernel text at 0xfe800000 and kernel data 630 * at 0xfec00000. On a 64-bit OS, kernel text and data are loaded at 631 * 0xffffffff.fe800000 and 0xffffffff.fec00000 respectively. Those 632 * addresses are fixed in the binary at link time. 633 * 634 * On the text page: 635 * unix/genunix/krtld/module text loads. 636 * 637 * On the data page: 638 * unix/genunix/krtld/module data loads. 639 * 640 * Machine-dependent startup code 641 */ 642 void 643 startup(void) 644 { 645 #if !defined(__xpv) 646 extern void startup_bios_disk(void); 647 extern void startup_pci_bios(void); 648 #endif 649 /* 650 * Make sure that nobody tries to use sekpm until we have 651 * initialized it properly. 652 */ 653 #if defined(__amd64) 654 kpm_desired = 1; 655 #endif 656 kpm_enable = 0; 657 658 #if defined(__xpv) /* XXPV fix me! */ 659 { 660 extern int segvn_use_regions; 661 segvn_use_regions = 0; 662 } 663 #endif 664 progressbar_init(); 665 startup_init(); 666 #if defined(__xpv) 667 startup_xen_version(); 668 #endif 669 startup_memlist(); 670 startup_kmem(); 671 startup_vm(); 672 #if !defined(__xpv) 673 startup_pci_bios(); 674 #endif 675 startup_modules(); 676 #if !defined(__xpv) 677 startup_bios_disk(); 678 #endif 679 startup_end(); 680 progressbar_start(); 681 } 682 683 static void 684 startup_init() 685 { 686 PRM_POINT("startup_init() starting..."); 687 688 /* 689 * Complete the extraction of cpuid data 690 */ 691 cpuid_pass2(CPU); 692 693 (void) check_boot_version(BOP_GETVERSION(bootops)); 694 695 /* 696 * Check for prom_debug in boot environment 697 */ 698 if (BOP_GETPROPLEN(bootops, "prom_debug") >= 0) { 699 ++prom_debug; 700 PRM_POINT("prom_debug found in boot enviroment"); 701 } 702 703 /* 704 * Collect node, cpu and memory configuration information. 705 */ 706 get_system_configuration(); 707 708 /* 709 * Halt if this is an unsupported processor. 710 */ 711 if (x86_type == X86_TYPE_486 || x86_type == X86_TYPE_CYRIX_486) { 712 printf("\n486 processor (\"%s\") detected.\n", 713 CPU->cpu_brandstr); 714 halt("This processor is not supported by this release " 715 "of Solaris."); 716 } 717 718 PRM_POINT("startup_init() done"); 719 } 720 721 /* 722 * Callback for copy_memlist_filter() to filter nucleus, kadb/kmdb, (ie. 723 * everything mapped above KERNEL_TEXT) pages from phys_avail. Note it 724 * also filters out physical page zero. There is some reliance on the 725 * boot loader allocating only a few contiguous physical memory chunks. 726 */ 727 static void 728 avail_filter(uint64_t *addr, uint64_t *size) 729 { 730 uintptr_t va; 731 uintptr_t next_va; 732 pfn_t pfn; 733 uint64_t pfn_addr; 734 uint64_t pfn_eaddr; 735 uint_t prot; 736 size_t len; 737 uint_t change; 738 739 if (prom_debug) 740 prom_printf("\tFilter: in: a=%" PRIx64 ", s=%" PRIx64 "\n", 741 *addr, *size); 742 743 /* 744 * page zero is required for BIOS.. never make it available 745 */ 746 if (*addr == 0) { 747 *addr += MMU_PAGESIZE; 748 *size -= MMU_PAGESIZE; 749 } 750 751 /* 752 * First we trim from the front of the range. Since kbm_probe() 753 * walks ranges in virtual order, but addr/size are physical, we need 754 * to the list until no changes are seen. This deals with the case 755 * where page "p" is mapped at v, page "p + PAGESIZE" is mapped at w 756 * but w < v. 757 */ 758 do { 759 change = 0; 760 for (va = KERNEL_TEXT; 761 *size > 0 && kbm_probe(&va, &len, &pfn, &prot) != 0; 762 va = next_va) { 763 764 next_va = va + len; 765 pfn_addr = pfn_to_pa(pfn); 766 pfn_eaddr = pfn_addr + len; 767 768 if (pfn_addr <= *addr && pfn_eaddr > *addr) { 769 change = 1; 770 while (*size > 0 && len > 0) { 771 *addr += MMU_PAGESIZE; 772 *size -= MMU_PAGESIZE; 773 len -= MMU_PAGESIZE; 774 } 775 } 776 } 777 if (change && prom_debug) 778 prom_printf("\t\ttrim: a=%" PRIx64 ", s=%" PRIx64 "\n", 779 *addr, *size); 780 } while (change); 781 782 /* 783 * Trim pages from the end of the range. 784 */ 785 for (va = KERNEL_TEXT; 786 *size > 0 && kbm_probe(&va, &len, &pfn, &prot) != 0; 787 va = next_va) { 788 789 next_va = va + len; 790 pfn_addr = pfn_to_pa(pfn); 791 792 if (pfn_addr >= *addr && pfn_addr < *addr + *size) 793 *size = pfn_addr - *addr; 794 } 795 796 if (prom_debug) 797 prom_printf("\tFilter out: a=%" PRIx64 ", s=%" PRIx64 "\n", 798 *addr, *size); 799 } 800 801 static void 802 kpm_init() 803 { 804 struct segkpm_crargs b; 805 806 /* 807 * These variables were all designed for sfmmu in which segkpm is 808 * mapped using a single pagesize - either 8KB or 4MB. On x86, we 809 * might use 2+ page sizes on a single machine, so none of these 810 * variables have a single correct value. They are set up as if we 811 * always use a 4KB pagesize, which should do no harm. In the long 812 * run, we should get rid of KPM's assumption that only a single 813 * pagesize is used. 814 */ 815 kpm_pgshft = MMU_PAGESHIFT; 816 kpm_pgsz = MMU_PAGESIZE; 817 kpm_pgoff = MMU_PAGEOFFSET; 818 kpmp2pshft = 0; 819 kpmpnpgs = 1; 820 ASSERT(((uintptr_t)kpm_vbase & (kpm_pgsz - 1)) == 0); 821 822 PRM_POINT("about to create segkpm"); 823 rw_enter(&kas.a_lock, RW_WRITER); 824 825 if (seg_attach(&kas, kpm_vbase, kpm_size, segkpm) < 0) 826 panic("cannot attach segkpm"); 827 828 b.prot = PROT_READ | PROT_WRITE; 829 b.nvcolors = 1; 830 831 if (segkpm_create(segkpm, (caddr_t)&b) != 0) 832 panic("segkpm_create segkpm"); 833 834 rw_exit(&kas.a_lock); 835 } 836 837 /* 838 * The debug info page provides enough information to allow external 839 * inspectors (e.g. when running under a hypervisor) to bootstrap 840 * themselves into allowing full-blown kernel debugging. 841 */ 842 static void 843 init_debug_info(void) 844 { 845 caddr_t mem; 846 debug_info_t *di; 847 848 #ifndef __lint 849 ASSERT(sizeof (debug_info_t) < MMU_PAGESIZE); 850 #endif 851 852 mem = BOP_ALLOC(bootops, (caddr_t)DEBUG_INFO_VA, MMU_PAGESIZE, 853 MMU_PAGESIZE); 854 855 if (mem != (caddr_t)DEBUG_INFO_VA) 856 panic("BOP_ALLOC() failed"); 857 bzero(mem, MMU_PAGESIZE); 858 859 di = (debug_info_t *)mem; 860 861 di->di_magic = DEBUG_INFO_MAGIC; 862 di->di_version = DEBUG_INFO_VERSION; 863 di->di_modules = (uintptr_t)&modules; 864 di->di_s_text = (uintptr_t)s_text; 865 di->di_e_text = (uintptr_t)e_text; 866 di->di_s_data = (uintptr_t)s_data; 867 di->di_e_data = (uintptr_t)e_data; 868 di->di_hat_htable_off = offsetof(hat_t, hat_htable); 869 di->di_ht_pfn_off = offsetof(htable_t, ht_pfn); 870 } 871 872 /* 873 * Build the memlists and other kernel essential memory system data structures. 874 * This is everything at valloc_base. 875 */ 876 static void 877 startup_memlist(void) 878 { 879 size_t memlist_sz; 880 size_t memseg_sz; 881 size_t pagehash_sz; 882 size_t pp_sz; 883 uintptr_t va; 884 size_t len; 885 uint_t prot; 886 pfn_t pfn; 887 int memblocks; 888 caddr_t pagecolor_mem; 889 size_t pagecolor_memsz; 890 caddr_t page_ctrs_mem; 891 size_t page_ctrs_size; 892 size_t pse_table_alloc_size; 893 struct memlist *current; 894 extern void startup_build_mem_nodes(struct memlist *); 895 896 /* XX64 fix these - they should be in include files */ 897 extern size_t page_coloring_init(uint_t, int, int); 898 extern void page_coloring_setup(caddr_t); 899 900 PRM_POINT("startup_memlist() starting..."); 901 902 /* 903 * Use leftover large page nucleus text/data space for loadable modules. 904 * Use at most MODTEXT/MODDATA. 905 */ 906 len = kbm_nucleus_size; 907 ASSERT(len > MMU_PAGESIZE); 908 909 moddata = (caddr_t)ROUND_UP_PAGE(e_data); 910 e_moddata = (caddr_t)P2ROUNDUP((uintptr_t)e_data, (uintptr_t)len); 911 if (e_moddata - moddata > MODDATA) 912 e_moddata = moddata + MODDATA; 913 914 modtext = (caddr_t)ROUND_UP_PAGE(e_text); 915 e_modtext = (caddr_t)P2ROUNDUP((uintptr_t)e_text, (uintptr_t)len); 916 if (e_modtext - modtext > MODTEXT) 917 e_modtext = modtext + MODTEXT; 918 919 econtig = e_moddata; 920 921 PRM_DEBUG(modtext); 922 PRM_DEBUG(e_modtext); 923 PRM_DEBUG(moddata); 924 PRM_DEBUG(e_moddata); 925 PRM_DEBUG(econtig); 926 927 /* 928 * Examine the boot loader physical memory map to find out: 929 * - total memory in system - physinstalled 930 * - the max physical address - physmax 931 * - the number of discontiguous segments of memory. 932 */ 933 if (prom_debug) 934 print_memlist("boot physinstalled", 935 bootops->boot_mem->physinstalled); 936 installed_top_size(bootops->boot_mem->physinstalled, &physmax, 937 &physinstalled, &memblocks); 938 PRM_DEBUG(physmax); 939 PRM_DEBUG(physinstalled); 940 PRM_DEBUG(memblocks); 941 942 /* 943 * Initialize hat's mmu parameters. 944 * Check for enforce-prot-exec in boot environment. It's used to 945 * enable/disable support for the page table entry NX bit. 946 * The default is to enforce PROT_EXEC on processors that support NX. 947 * Boot seems to round up the "len", but 8 seems to be big enough. 948 */ 949 mmu_init(); 950 951 #ifdef __i386 952 /* 953 * physmax is lowered if there is more memory than can be 954 * physically addressed in 32 bit (PAE/non-PAE) modes. 955 */ 956 if (mmu.pae_hat) { 957 if (PFN_ABOVE64G(physmax)) { 958 physinstalled -= (physmax - (PFN_64G - 1)); 959 physmax = PFN_64G - 1; 960 } 961 } else { 962 if (PFN_ABOVE4G(physmax)) { 963 physinstalled -= (physmax - (PFN_4G - 1)); 964 physmax = PFN_4G - 1; 965 } 966 } 967 #endif 968 969 startup_build_mem_nodes(bootops->boot_mem->physinstalled); 970 971 if (BOP_GETPROPLEN(bootops, "enforce-prot-exec") >= 0) { 972 int len = BOP_GETPROPLEN(bootops, "enforce-prot-exec"); 973 char value[8]; 974 975 if (len < 8) 976 (void) BOP_GETPROP(bootops, "enforce-prot-exec", value); 977 else 978 (void) strcpy(value, ""); 979 if (strcmp(value, "off") == 0) 980 mmu.pt_nx = 0; 981 } 982 PRM_DEBUG(mmu.pt_nx); 983 984 /* 985 * We will need page_t's for every page in the system, except for 986 * memory mapped at or above above the start of the kernel text segment. 987 * 988 * pages above e_modtext are attributed to kernel debugger (obp_pages) 989 */ 990 npages = physinstalled - 1; /* avail_filter() skips page 0, so "- 1" */ 991 obp_pages = 0; 992 va = KERNEL_TEXT; 993 while (kbm_probe(&va, &len, &pfn, &prot) != 0) { 994 npages -= len >> MMU_PAGESHIFT; 995 if (va >= (uintptr_t)e_moddata) 996 obp_pages += len >> MMU_PAGESHIFT; 997 va += len; 998 } 999 PRM_DEBUG(npages); 1000 PRM_DEBUG(obp_pages); 1001 1002 /* 1003 * If physmem is patched to be non-zero, use it instead of the computed 1004 * value unless it is larger than the actual amount of memory on hand. 1005 */ 1006 if (physmem == 0 || physmem > npages) { 1007 physmem = npages; 1008 } else if (physmem < npages) { 1009 orig_npages = npages; 1010 npages = physmem; 1011 } 1012 PRM_DEBUG(physmem); 1013 1014 /* 1015 * We now compute the sizes of all the initial allocations for 1016 * structures the kernel needs in order do kmem_alloc(). These 1017 * include: 1018 * memsegs 1019 * memlists 1020 * page hash table 1021 * page_t's 1022 * page coloring data structs 1023 */ 1024 memseg_sz = sizeof (struct memseg) * (memblocks + POSS_NEW_FRAGMENTS); 1025 ADD_TO_ALLOCATIONS(memseg_base, memseg_sz); 1026 PRM_DEBUG(memseg_sz); 1027 1028 /* 1029 * Reserve space for memlists. There's no real good way to know exactly 1030 * how much room we'll need, but this should be a good upper bound. 1031 */ 1032 memlist_sz = ROUND_UP_PAGE(2 * sizeof (struct memlist) * 1033 (memblocks + POSS_NEW_FRAGMENTS)); 1034 ADD_TO_ALLOCATIONS(memlist, memlist_sz); 1035 PRM_DEBUG(memlist_sz); 1036 1037 /* 1038 * The page structure hash table size is a power of 2 1039 * such that the average hash chain length is PAGE_HASHAVELEN. 1040 */ 1041 page_hashsz = npages / PAGE_HASHAVELEN; 1042 page_hashsz = 1 << highbit(page_hashsz); 1043 pagehash_sz = sizeof (struct page *) * page_hashsz; 1044 ADD_TO_ALLOCATIONS(page_hash, pagehash_sz); 1045 PRM_DEBUG(pagehash_sz); 1046 1047 /* 1048 * Set aside room for the page structures themselves. 1049 */ 1050 PRM_DEBUG(npages); 1051 pp_sz = sizeof (struct page) * npages; 1052 ADD_TO_ALLOCATIONS(pp_base, pp_sz); 1053 PRM_DEBUG(pp_sz); 1054 1055 /* 1056 * determine l2 cache info and memory size for page coloring 1057 */ 1058 (void) getl2cacheinfo(CPU, 1059 &l2cache_sz, &l2cache_linesz, &l2cache_assoc); 1060 pagecolor_memsz = 1061 page_coloring_init(l2cache_sz, l2cache_linesz, l2cache_assoc); 1062 ADD_TO_ALLOCATIONS(pagecolor_mem, pagecolor_memsz); 1063 PRM_DEBUG(pagecolor_memsz); 1064 1065 page_ctrs_size = page_ctrs_sz(); 1066 ADD_TO_ALLOCATIONS(page_ctrs_mem, page_ctrs_size); 1067 PRM_DEBUG(page_ctrs_size); 1068 1069 /* 1070 * Allocate the array that protects pp->p_selock. 1071 */ 1072 pse_shift = size_pse_array(physmem, max_ncpus); 1073 pse_table_size = 1 << pse_shift; 1074 pse_table_alloc_size = pse_table_size * sizeof (pad_mutex_t); 1075 ADD_TO_ALLOCATIONS(pse_mutex, pse_table_alloc_size); 1076 1077 #if defined(__amd64) 1078 valloc_sz = ROUND_UP_LPAGE(valloc_sz); 1079 valloc_base = VALLOC_BASE; 1080 1081 /* 1082 * The default values of VALLOC_BASE and SEGKPM_BASE should work 1083 * for values of physmax up to 1 Terabyte. They need adjusting when 1084 * memory is at addresses above 1 TB. 1085 */ 1086 if (physmax + 1 > mmu_btop(TERABYTE)) { 1087 uint64_t kpm_resv_amount = mmu_ptob(physmax + 1); 1088 1089 /* Round to largest possible pagesize for now */ 1090 kpm_resv_amount = P2ROUNDUP(kpm_resv_amount, ONE_GIG); 1091 1092 segkpm_base = -(2 * kpm_resv_amount); /* down from top VA */ 1093 1094 /* make sure we leave some space for user apps above hole */ 1095 segkpm_base = MAX(segkpm_base, AMD64_VA_HOLE_END + TERABYTE); 1096 if (segkpm_base > SEGKPM_BASE) 1097 segkpm_base = SEGKPM_BASE; 1098 PRM_DEBUG(segkpm_base); 1099 1100 valloc_base = segkpm_base + kpm_resv_amount; 1101 PRM_DEBUG(valloc_base); 1102 } 1103 #else /* __i386 */ 1104 valloc_base = (uintptr_t)(MISC_VA_BASE - valloc_sz); 1105 valloc_base = P2ALIGN(valloc_base, mmu.level_size[1]); 1106 PRM_DEBUG(valloc_base); 1107 #endif /* __i386 */ 1108 1109 /* 1110 * do all the initial allocations 1111 */ 1112 perform_allocations(); 1113 1114 /* 1115 * Build phys_install and phys_avail in kernel memspace. 1116 * - phys_install should be all memory in the system. 1117 * - phys_avail is phys_install minus any memory mapped before this 1118 * point above KERNEL_TEXT. 1119 */ 1120 current = phys_install = memlist; 1121 copy_memlist_filter(bootops->boot_mem->physinstalled, ¤t, NULL); 1122 if ((caddr_t)current > (caddr_t)memlist + memlist_sz) 1123 panic("physinstalled was too big!"); 1124 if (prom_debug) 1125 print_memlist("phys_install", phys_install); 1126 1127 phys_avail = current; 1128 PRM_POINT("Building phys_avail:\n"); 1129 copy_memlist_filter(bootops->boot_mem->physinstalled, ¤t, 1130 avail_filter); 1131 if ((caddr_t)current > (caddr_t)memlist + memlist_sz) 1132 panic("physavail was too big!"); 1133 if (prom_debug) 1134 print_memlist("phys_avail", phys_avail); 1135 1136 /* 1137 * setup page coloring 1138 */ 1139 page_coloring_setup(pagecolor_mem); 1140 page_lock_init(); /* currently a no-op */ 1141 1142 /* 1143 * free page list counters 1144 */ 1145 (void) page_ctrs_alloc(page_ctrs_mem); 1146 1147 /* 1148 * Initialize the page structures from the memory lists. 1149 */ 1150 availrmem_initial = availrmem = freemem = 0; 1151 PRM_POINT("Calling kphysm_init()..."); 1152 npages = kphysm_init(pp_base, npages); 1153 PRM_POINT("kphysm_init() done"); 1154 PRM_DEBUG(npages); 1155 1156 init_debug_info(); 1157 1158 /* 1159 * Now that page_t's have been initialized, remove all the 1160 * initial allocation pages from the kernel free page lists. 1161 */ 1162 boot_mapin((caddr_t)valloc_base, valloc_sz); 1163 boot_mapin((caddr_t)MISC_VA_BASE, MISC_VA_SIZE); 1164 PRM_POINT("startup_memlist() done"); 1165 1166 PRM_DEBUG(valloc_sz); 1167 1168 #if defined(__amd64) 1169 if ((availrmem >> (30 - MMU_PAGESHIFT)) >= 1170 textrepl_min_gb && l2cache_sz <= 2 << 20) { 1171 extern size_t textrepl_size_thresh; 1172 textrepl_size_thresh = (16 << 20) - 1; 1173 } 1174 #endif 1175 } 1176 1177 /* 1178 * Layout the kernel's part of address space and initialize kmem allocator. 1179 */ 1180 static void 1181 startup_kmem(void) 1182 { 1183 extern void page_set_colorequiv_arr(void); 1184 1185 PRM_POINT("startup_kmem() starting..."); 1186 1187 #if defined(__amd64) 1188 if (eprom_kernelbase && eprom_kernelbase != KERNELBASE) 1189 cmn_err(CE_NOTE, "!kernelbase cannot be changed on 64-bit " 1190 "systems."); 1191 kernelbase = segkpm_base - KERNEL_REDZONE_SIZE; 1192 core_base = (uintptr_t)COREHEAP_BASE; 1193 core_size = (size_t)MISC_VA_BASE - COREHEAP_BASE; 1194 #else /* __i386 */ 1195 /* 1196 * We configure kernelbase based on: 1197 * 1198 * 1. user specified kernelbase via eeprom command. Value cannot exceed 1199 * KERNELBASE_MAX. we large page align eprom_kernelbase 1200 * 1201 * 2. Default to KERNELBASE and adjust to 2X less the size for page_t. 1202 * On large memory systems we must lower kernelbase to allow 1203 * enough room for page_t's for all of memory. 1204 * 1205 * The value set here, might be changed a little later. 1206 */ 1207 if (eprom_kernelbase) { 1208 kernelbase = eprom_kernelbase & mmu.level_mask[1]; 1209 if (kernelbase > KERNELBASE_MAX) 1210 kernelbase = KERNELBASE_MAX; 1211 } else { 1212 kernelbase = (uintptr_t)KERNELBASE; 1213 kernelbase -= ROUND_UP_4MEG(2 * valloc_sz); 1214 } 1215 ASSERT((kernelbase & mmu.level_offset[1]) == 0); 1216 core_base = valloc_base; 1217 core_size = 0; 1218 #endif /* __i386 */ 1219 1220 PRM_DEBUG(core_base); 1221 PRM_DEBUG(core_size); 1222 PRM_DEBUG(kernelbase); 1223 1224 #if defined(__i386) 1225 segkp_fromheap = 1; 1226 #endif /* __i386 */ 1227 1228 ekernelheap = (char *)core_base; 1229 PRM_DEBUG(ekernelheap); 1230 1231 /* 1232 * Now that we know the real value of kernelbase, 1233 * update variables that were initialized with a value of 1234 * KERNELBASE (in common/conf/param.c). 1235 * 1236 * XXX The problem with this sort of hackery is that the 1237 * compiler just may feel like putting the const declarations 1238 * (in param.c) into the .text section. Perhaps they should 1239 * just be declared as variables there? 1240 */ 1241 1242 *(uintptr_t *)&_kernelbase = kernelbase; 1243 *(uintptr_t *)&_userlimit = kernelbase; 1244 #if defined(__amd64) 1245 *(uintptr_t *)&_userlimit -= KERNELBASE - USERLIMIT; 1246 #else 1247 *(uintptr_t *)&_userlimit32 = _userlimit; 1248 #endif 1249 PRM_DEBUG(_kernelbase); 1250 PRM_DEBUG(_userlimit); 1251 PRM_DEBUG(_userlimit32); 1252 1253 layout_kernel_va(); 1254 1255 #if defined(__i386) 1256 /* 1257 * If segmap is too large we can push the bottom of the kernel heap 1258 * higher than the base. Or worse, it could exceed the top of the 1259 * VA space entirely, causing it to wrap around. 1260 */ 1261 if (kernelheap >= ekernelheap || (uintptr_t)kernelheap < kernelbase) 1262 panic("too little address space available for kernelheap," 1263 " use eeprom for lower kernelbase or smaller segmapsize"); 1264 #endif /* __i386 */ 1265 1266 /* 1267 * Initialize the kernel heap. Note 3rd argument must be > 1st. 1268 */ 1269 kernelheap_init(kernelheap, ekernelheap, 1270 kernelheap + MMU_PAGESIZE, 1271 (void *)core_base, (void *)(core_base + core_size)); 1272 1273 #if defined(__xpv) 1274 /* 1275 * Link pending events struct into cpu struct 1276 */ 1277 CPU->cpu_m.mcpu_evt_pend = &cpu0_evt_data; 1278 #endif 1279 /* 1280 * Initialize kernel memory allocator. 1281 */ 1282 kmem_init(); 1283 1284 /* 1285 * Factor in colorequiv to check additional 'equivalent' bins 1286 */ 1287 page_set_colorequiv_arr(); 1288 1289 /* 1290 * print this out early so that we know what's going on 1291 */ 1292 cmn_err(CE_CONT, "?features: %b\n", x86_feature, FMT_X86_FEATURE); 1293 1294 /* 1295 * Initialize bp_mapin(). 1296 */ 1297 bp_init(MMU_PAGESIZE, HAT_STORECACHING_OK); 1298 1299 /* 1300 * orig_npages is non-zero if physmem has been configured for less 1301 * than the available memory. 1302 */ 1303 if (orig_npages) { 1304 cmn_err(CE_WARN, "!%slimiting physmem to 0x%lx of 0x%lx pages", 1305 (npages == PHYSMEM ? "Due to virtual address space " : ""), 1306 npages, orig_npages); 1307 } 1308 #if defined(__i386) 1309 if (eprom_kernelbase && (eprom_kernelbase != kernelbase)) 1310 cmn_err(CE_WARN, "kernelbase value, User specified 0x%lx, " 1311 "System using 0x%lx", 1312 (uintptr_t)eprom_kernelbase, (uintptr_t)kernelbase); 1313 #endif 1314 1315 #ifdef KERNELBASE_ABI_MIN 1316 if (kernelbase < (uintptr_t)KERNELBASE_ABI_MIN) { 1317 cmn_err(CE_NOTE, "!kernelbase set to 0x%lx, system is not " 1318 "i386 ABI compliant.", (uintptr_t)kernelbase); 1319 } 1320 #endif 1321 1322 #ifdef __xpv 1323 /* 1324 * Some of the xen start information has to be relocated up 1325 * into the kernel's permanent address space. 1326 */ 1327 PRM_POINT("calling xen_relocate_start_info()"); 1328 xen_relocate_start_info(); 1329 PRM_POINT("xen_relocate_start_info() done"); 1330 1331 /* 1332 * (Update the vcpu pointer in our cpu structure to point into 1333 * the relocated shared info.) 1334 */ 1335 CPU->cpu_m.mcpu_vcpu_info = 1336 &HYPERVISOR_shared_info->vcpu_info[CPU->cpu_id]; 1337 #endif 1338 1339 PRM_POINT("startup_kmem() done"); 1340 } 1341 1342 static void 1343 startup_modules(void) 1344 { 1345 unsigned int i; 1346 extern void prom_setup(void); 1347 1348 PRM_POINT("startup_modules() starting..."); 1349 1350 #ifndef __xpv 1351 /* 1352 * Initialize ten-micro second timer so that drivers will 1353 * not get short changed in their init phase. This was 1354 * not getting called until clkinit which, on fast cpu's 1355 * caused the drv_usecwait to be way too short. 1356 */ 1357 microfind(); 1358 #endif 1359 1360 /* 1361 * Read the GMT lag from /etc/rtc_config. 1362 */ 1363 sgmtl(process_rtc_config_file()); 1364 1365 /* 1366 * Calculate default settings of system parameters based upon 1367 * maxusers, yet allow to be overridden via the /etc/system file. 1368 */ 1369 param_calc(0); 1370 1371 mod_setup(); 1372 1373 /* 1374 * Initialize system parameters. 1375 */ 1376 param_init(); 1377 1378 /* 1379 * Initialize the default brands 1380 */ 1381 brand_init(); 1382 1383 /* 1384 * maxmem is the amount of physical memory we're playing with. 1385 */ 1386 maxmem = physmem; 1387 1388 /* 1389 * Initialize segment management stuff. 1390 */ 1391 seg_init(); 1392 1393 if (modload("fs", "specfs") == -1) 1394 halt("Can't load specfs"); 1395 1396 if (modload("fs", "devfs") == -1) 1397 halt("Can't load devfs"); 1398 1399 if (modload("fs", "dev") == -1) 1400 halt("Can't load dev"); 1401 1402 (void) modloadonly("sys", "lbl_edition"); 1403 1404 dispinit(); 1405 1406 /* 1407 * This is needed here to initialize hw_serial[] for cluster booting. 1408 */ 1409 if ((i = modload("misc", "sysinit")) != (unsigned int)-1) 1410 (void) modunload(i); 1411 else 1412 cmn_err(CE_CONT, "sysinit load failed"); 1413 1414 /* Read cluster configuration data. */ 1415 clconf_init(); 1416 1417 #if defined(__xpv) 1418 ec_init(); 1419 gnttab_init(); 1420 (void) xs_early_init(); 1421 #endif /* __xpv */ 1422 1423 /* 1424 * Create a kernel device tree. First, create rootnex and 1425 * then invoke bus specific code to probe devices. 1426 */ 1427 setup_ddi(); 1428 1429 #ifndef __xpv 1430 { 1431 /* 1432 * Set up the CPU module subsystem. Modifies the device tree, 1433 * so it must be done after setup_ddi(). 1434 */ 1435 1436 cmi_hdl_t hdl; 1437 1438 if ((hdl = cmi_init(CMI_HDL_NATIVE, cmi_ntv_hwchipid(CPU), 1439 cmi_ntv_hwcoreid(CPU), cmi_ntv_hwstrandid(CPU))) != NULL) { 1440 if (x86_feature & X86_MCA) 1441 cmi_mca_init(hdl); 1442 } 1443 } 1444 #endif /* __xpv */ 1445 1446 /* 1447 * Fake a prom tree such that /dev/openprom continues to work 1448 */ 1449 PRM_POINT("startup_modules: calling prom_setup..."); 1450 prom_setup(); 1451 PRM_POINT("startup_modules: done"); 1452 1453 /* 1454 * Load all platform specific modules 1455 */ 1456 PRM_POINT("startup_modules: calling psm_modload..."); 1457 psm_modload(); 1458 1459 PRM_POINT("startup_modules() done"); 1460 } 1461 1462 /* 1463 * claim a "setaside" boot page for use in the kernel 1464 */ 1465 page_t * 1466 boot_claim_page(pfn_t pfn) 1467 { 1468 page_t *pp; 1469 1470 pp = page_numtopp_nolock(pfn); 1471 ASSERT(pp != NULL); 1472 1473 if (PP_ISBOOTPAGES(pp)) { 1474 if (pp->p_next != NULL) 1475 pp->p_next->p_prev = pp->p_prev; 1476 if (pp->p_prev == NULL) 1477 bootpages = pp->p_next; 1478 else 1479 pp->p_prev->p_next = pp->p_next; 1480 } else { 1481 /* 1482 * htable_attach() expects a base pagesize page 1483 */ 1484 if (pp->p_szc != 0) 1485 page_boot_demote(pp); 1486 pp = page_numtopp(pfn, SE_EXCL); 1487 } 1488 return (pp); 1489 } 1490 1491 /* 1492 * Walk through the pagetables looking for pages mapped in by boot. If the 1493 * setaside flag is set the pages are expected to be returned to the 1494 * kernel later in boot, so we add them to the bootpages list. 1495 */ 1496 static void 1497 protect_boot_range(uintptr_t low, uintptr_t high, int setaside) 1498 { 1499 uintptr_t va = low; 1500 size_t len; 1501 uint_t prot; 1502 pfn_t pfn; 1503 page_t *pp; 1504 pgcnt_t boot_protect_cnt = 0; 1505 1506 while (kbm_probe(&va, &len, &pfn, &prot) != 0 && va < high) { 1507 if (va + len >= high) 1508 panic("0x%lx byte mapping at 0x%p exceeds boot's " 1509 "legal range.", len, (void *)va); 1510 1511 while (len > 0) { 1512 pp = page_numtopp_alloc(pfn); 1513 if (pp != NULL) { 1514 if (setaside == 0) 1515 panic("Unexpected mapping by boot. " 1516 "addr=%p pfn=%lx\n", 1517 (void *)va, pfn); 1518 1519 pp->p_next = bootpages; 1520 pp->p_prev = NULL; 1521 PP_SETBOOTPAGES(pp); 1522 if (bootpages != NULL) { 1523 bootpages->p_prev = pp; 1524 } 1525 bootpages = pp; 1526 ++boot_protect_cnt; 1527 } 1528 1529 ++pfn; 1530 len -= MMU_PAGESIZE; 1531 va += MMU_PAGESIZE; 1532 } 1533 } 1534 PRM_DEBUG(boot_protect_cnt); 1535 } 1536 1537 /* 1538 * 1539 */ 1540 static void 1541 layout_kernel_va(void) 1542 { 1543 PRM_POINT("layout_kernel_va() starting..."); 1544 /* 1545 * Establish the final size of the kernel's heap, size of segmap, 1546 * segkp, etc. 1547 */ 1548 1549 #if defined(__amd64) 1550 1551 kpm_vbase = (caddr_t)segkpm_base; 1552 kpm_size = ROUND_UP_LPAGE(mmu_ptob(physmax + 1)); 1553 if ((uintptr_t)kpm_vbase + kpm_size > (uintptr_t)valloc_base) 1554 panic("not enough room for kpm!"); 1555 PRM_DEBUG(kpm_size); 1556 PRM_DEBUG(kpm_vbase); 1557 1558 /* 1559 * By default we create a seg_kp in 64 bit kernels, it's a little 1560 * faster to access than embedding it in the heap. 1561 */ 1562 segkp_base = (caddr_t)valloc_base + valloc_sz; 1563 if (!segkp_fromheap) { 1564 size_t sz = mmu_ptob(segkpsize); 1565 1566 /* 1567 * determine size of segkp 1568 */ 1569 if (sz < SEGKPMINSIZE || sz > SEGKPMAXSIZE) { 1570 sz = SEGKPDEFSIZE; 1571 cmn_err(CE_WARN, "!Illegal value for segkpsize. " 1572 "segkpsize has been reset to %ld pages", 1573 mmu_btop(sz)); 1574 } 1575 sz = MIN(sz, MAX(SEGKPMINSIZE, mmu_ptob(physmem))); 1576 1577 segkpsize = mmu_btop(ROUND_UP_LPAGE(sz)); 1578 } 1579 PRM_DEBUG(segkp_base); 1580 PRM_DEBUG(segkpsize); 1581 1582 /* 1583 * segzio is used for ZFS cached data. It uses a distinct VA 1584 * segment (from kernel heap) so that we can easily tell not to 1585 * include it in kernel crash dumps on 64 bit kernels. The trick is 1586 * to give it lots of VA, but not constrain the kernel heap. 1587 * We scale the size of segzio linearly with physmem up to 1588 * SEGZIOMAXSIZE. Above that amount it scales at 50% of physmem. 1589 */ 1590 segzio_base = segkp_base + mmu_ptob(segkpsize); 1591 if (segzio_fromheap) { 1592 segziosize = 0; 1593 } else { 1594 size_t physmem_size = mmu_ptob(physmem); 1595 size_t size = (segziosize == 0) ? 1596 physmem_size : mmu_ptob(segziosize); 1597 1598 if (size < SEGZIOMINSIZE) 1599 size = SEGZIOMINSIZE; 1600 if (size > SEGZIOMAXSIZE) { 1601 size = SEGZIOMAXSIZE; 1602 if (physmem_size > size) 1603 size += (physmem_size - size) / 2; 1604 } 1605 segziosize = mmu_btop(ROUND_UP_LPAGE(size)); 1606 } 1607 PRM_DEBUG(segziosize); 1608 PRM_DEBUG(segzio_base); 1609 1610 /* 1611 * Put the range of VA for device mappings next, kmdb knows to not 1612 * grep in this range of addresses. 1613 */ 1614 toxic_addr = 1615 ROUND_UP_LPAGE((uintptr_t)segzio_base + mmu_ptob(segziosize)); 1616 PRM_DEBUG(toxic_addr); 1617 segmap_start = ROUND_UP_LPAGE(toxic_addr + toxic_size); 1618 #else /* __i386 */ 1619 segmap_start = ROUND_UP_LPAGE(kernelbase); 1620 #endif /* __i386 */ 1621 PRM_DEBUG(segmap_start); 1622 1623 /* 1624 * Users can change segmapsize through eeprom or /etc/system. 1625 * If the variable is tuned through eeprom, there is no upper 1626 * bound on the size of segmap. If it is tuned through 1627 * /etc/system on 32-bit systems, it must be no larger than we 1628 * planned for in startup_memlist(). 1629 */ 1630 segmapsize = MAX(ROUND_UP_LPAGE(segmapsize), SEGMAPDEFAULT); 1631 1632 #if defined(__i386) 1633 /* 1634 * 32-bit systems don't have segkpm or segkp, so segmap appears at 1635 * the bottom of the kernel's address range. Set aside space for a 1636 * small red zone just below the start of segmap. 1637 */ 1638 segmap_start += KERNEL_REDZONE_SIZE; 1639 segmapsize -= KERNEL_REDZONE_SIZE; 1640 #endif 1641 1642 PRM_DEBUG(segmap_start); 1643 PRM_DEBUG(segmapsize); 1644 kernelheap = (caddr_t)ROUND_UP_LPAGE(segmap_start + segmapsize); 1645 PRM_DEBUG(kernelheap); 1646 PRM_POINT("layout_kernel_va() done..."); 1647 } 1648 1649 /* 1650 * Finish initializing the VM system, now that we are no longer 1651 * relying on the boot time memory allocators. 1652 */ 1653 static void 1654 startup_vm(void) 1655 { 1656 struct segmap_crargs a; 1657 1658 extern int use_brk_lpg, use_stk_lpg; 1659 1660 PRM_POINT("startup_vm() starting..."); 1661 1662 /* 1663 * Initialize the hat layer. 1664 */ 1665 hat_init(); 1666 1667 /* 1668 * Do final allocations of HAT data structures that need to 1669 * be allocated before quiescing the boot loader. 1670 */ 1671 PRM_POINT("Calling hat_kern_alloc()..."); 1672 hat_kern_alloc((caddr_t)segmap_start, segmapsize, ekernelheap); 1673 PRM_POINT("hat_kern_alloc() done"); 1674 1675 #ifndef __xpv 1676 /* 1677 * Setup Page Attribute Table 1678 */ 1679 pat_sync(); 1680 #endif 1681 1682 /* 1683 * The next two loops are done in distinct steps in order 1684 * to be sure that any page that is doubly mapped (both above 1685 * KERNEL_TEXT and below kernelbase) is dealt with correctly. 1686 * Note this may never happen, but it might someday. 1687 */ 1688 bootpages = NULL; 1689 PRM_POINT("Protecting boot pages"); 1690 1691 /* 1692 * Protect any pages mapped above KERNEL_TEXT that somehow have 1693 * page_t's. This can only happen if something weird allocated 1694 * in this range (like kadb/kmdb). 1695 */ 1696 protect_boot_range(KERNEL_TEXT, (uintptr_t)-1, 0); 1697 1698 /* 1699 * Before we can take over memory allocation/mapping from the boot 1700 * loader we must remove from our free page lists any boot allocated 1701 * pages that stay mapped until release_bootstrap(). 1702 */ 1703 protect_boot_range(0, kernelbase, 1); 1704 1705 1706 /* 1707 * Switch to running on regular HAT (not boot_mmu) 1708 */ 1709 PRM_POINT("Calling hat_kern_setup()..."); 1710 hat_kern_setup(); 1711 1712 /* 1713 * It is no longer safe to call BOP_ALLOC(), so make sure we don't. 1714 */ 1715 bop_no_more_mem(); 1716 1717 PRM_POINT("hat_kern_setup() done"); 1718 1719 hat_cpu_online(CPU); 1720 1721 /* 1722 * Initialize VM system 1723 */ 1724 PRM_POINT("Calling kvm_init()..."); 1725 kvm_init(); 1726 PRM_POINT("kvm_init() done"); 1727 1728 /* 1729 * Tell kmdb that the VM system is now working 1730 */ 1731 if (boothowto & RB_DEBUG) 1732 kdi_dvec_vmready(); 1733 1734 #if defined(__xpv) 1735 /* 1736 * Populate the I/O pool on domain 0 1737 */ 1738 if (DOMAIN_IS_INITDOMAIN(xen_info)) { 1739 extern long populate_io_pool(void); 1740 long init_io_pool_cnt; 1741 1742 PRM_POINT("Populating reserve I/O page pool"); 1743 init_io_pool_cnt = populate_io_pool(); 1744 PRM_DEBUG(init_io_pool_cnt); 1745 } 1746 #endif 1747 /* 1748 * Mangle the brand string etc. 1749 */ 1750 cpuid_pass3(CPU); 1751 1752 #if defined(__amd64) 1753 1754 /* 1755 * Create the device arena for toxic (to dtrace/kmdb) mappings. 1756 */ 1757 device_arena = vmem_create("device", (void *)toxic_addr, 1758 toxic_size, MMU_PAGESIZE, NULL, NULL, NULL, 0, VM_SLEEP); 1759 1760 #else /* __i386 */ 1761 1762 /* 1763 * allocate the bit map that tracks toxic pages 1764 */ 1765 toxic_bit_map_len = btop((ulong_t)(valloc_base - kernelbase)); 1766 PRM_DEBUG(toxic_bit_map_len); 1767 toxic_bit_map = 1768 kmem_zalloc(BT_SIZEOFMAP(toxic_bit_map_len), KM_NOSLEEP); 1769 ASSERT(toxic_bit_map != NULL); 1770 PRM_DEBUG(toxic_bit_map); 1771 1772 #endif /* __i386 */ 1773 1774 1775 /* 1776 * Now that we've got more VA, as well as the ability to allocate from 1777 * it, tell the debugger. 1778 */ 1779 if (boothowto & RB_DEBUG) 1780 kdi_dvec_memavail(); 1781 1782 /* 1783 * The following code installs a special page fault handler (#pf) 1784 * to work around a pentium bug. 1785 */ 1786 #if !defined(__amd64) && !defined(__xpv) 1787 if (x86_type == X86_TYPE_P5) { 1788 desctbr_t idtr; 1789 gate_desc_t *newidt; 1790 1791 if ((newidt = kmem_zalloc(MMU_PAGESIZE, KM_NOSLEEP)) == NULL) 1792 panic("failed to install pentium_pftrap"); 1793 1794 bcopy(idt0, newidt, NIDT * sizeof (*idt0)); 1795 set_gatesegd(&newidt[T_PGFLT], &pentium_pftrap, 1796 KCS_SEL, SDT_SYSIGT, TRP_KPL); 1797 1798 (void) as_setprot(&kas, (caddr_t)newidt, MMU_PAGESIZE, 1799 PROT_READ | PROT_EXEC); 1800 1801 CPU->cpu_idt = newidt; 1802 idtr.dtr_base = (uintptr_t)CPU->cpu_idt; 1803 idtr.dtr_limit = (NIDT * sizeof (*idt0)) - 1; 1804 wr_idtr(&idtr); 1805 } 1806 #endif /* !__amd64 */ 1807 1808 #if !defined(__xpv) 1809 /* 1810 * Map page pfn=0 for drivers, such as kd, that need to pick up 1811 * parameters left there by controllers/BIOS. 1812 */ 1813 PRM_POINT("setup up p0_va"); 1814 p0_va = i86devmap(0, 1, PROT_READ); 1815 PRM_DEBUG(p0_va); 1816 #endif 1817 1818 cmn_err(CE_CONT, "?mem = %luK (0x%lx)\n", 1819 physinstalled << (MMU_PAGESHIFT - 10), ptob(physinstalled)); 1820 1821 /* 1822 * disable automatic large pages for small memory systems or 1823 * when the disable flag is set. 1824 * 1825 * Do not yet consider page sizes larger than 2m/4m. 1826 */ 1827 if (!auto_lpg_disable && mmu.max_page_level > 0) { 1828 max_uheap_lpsize = LEVEL_SIZE(1); 1829 max_ustack_lpsize = LEVEL_SIZE(1); 1830 max_privmap_lpsize = LEVEL_SIZE(1); 1831 max_uidata_lpsize = LEVEL_SIZE(1); 1832 max_utext_lpsize = LEVEL_SIZE(1); 1833 max_shm_lpsize = LEVEL_SIZE(1); 1834 } 1835 if (physmem < privm_lpg_min_physmem || mmu.max_page_level == 0 || 1836 auto_lpg_disable) { 1837 use_brk_lpg = 0; 1838 use_stk_lpg = 0; 1839 } 1840 mcntl0_lpsize = LEVEL_SIZE(mmu.umax_page_level); 1841 1842 PRM_POINT("Calling hat_init_finish()..."); 1843 hat_init_finish(); 1844 PRM_POINT("hat_init_finish() done"); 1845 1846 /* 1847 * Initialize the segkp segment type. 1848 */ 1849 rw_enter(&kas.a_lock, RW_WRITER); 1850 PRM_POINT("Attaching segkp"); 1851 if (segkp_fromheap) { 1852 segkp->s_as = &kas; 1853 } else if (seg_attach(&kas, (caddr_t)segkp_base, mmu_ptob(segkpsize), 1854 segkp) < 0) { 1855 panic("startup: cannot attach segkp"); 1856 /*NOTREACHED*/ 1857 } 1858 PRM_POINT("Doing segkp_create()"); 1859 if (segkp_create(segkp) != 0) { 1860 panic("startup: segkp_create failed"); 1861 /*NOTREACHED*/ 1862 } 1863 PRM_DEBUG(segkp); 1864 rw_exit(&kas.a_lock); 1865 1866 /* 1867 * kpm segment 1868 */ 1869 segmap_kpm = 0; 1870 if (kpm_desired) { 1871 kpm_init(); 1872 kpm_enable = 1; 1873 vpm_enable = 1; 1874 } 1875 1876 /* 1877 * Now create segmap segment. 1878 */ 1879 rw_enter(&kas.a_lock, RW_WRITER); 1880 if (seg_attach(&kas, (caddr_t)segmap_start, segmapsize, segmap) < 0) { 1881 panic("cannot attach segmap"); 1882 /*NOTREACHED*/ 1883 } 1884 PRM_DEBUG(segmap); 1885 1886 a.prot = PROT_READ | PROT_WRITE; 1887 a.shmsize = 0; 1888 a.nfreelist = segmapfreelists; 1889 1890 if (segmap_create(segmap, (caddr_t)&a) != 0) 1891 panic("segmap_create segmap"); 1892 rw_exit(&kas.a_lock); 1893 1894 setup_vaddr_for_ppcopy(CPU); 1895 1896 segdev_init(); 1897 #if defined(__xpv) 1898 if (DOMAIN_IS_INITDOMAIN(xen_info)) 1899 #endif 1900 pmem_init(); 1901 1902 PRM_POINT("startup_vm() done"); 1903 } 1904 1905 /* 1906 * Load a tod module for the non-standard tod part found on this system. 1907 */ 1908 static void 1909 load_tod_module(char *todmod) 1910 { 1911 if (modload("tod", todmod) == -1) 1912 halt("Can't load TOD module"); 1913 } 1914 1915 static void 1916 startup_end(void) 1917 { 1918 int i; 1919 extern void setx86isalist(void); 1920 1921 PRM_POINT("startup_end() starting..."); 1922 1923 /* 1924 * Perform tasks that get done after most of the VM 1925 * initialization has been done but before the clock 1926 * and other devices get started. 1927 */ 1928 kern_setup1(); 1929 1930 /* 1931 * Perform CPC initialization for this CPU. 1932 */ 1933 kcpc_hw_init(CPU); 1934 1935 #if defined(OPTERON_WORKAROUND_6323525) 1936 if (opteron_workaround_6323525) 1937 patch_workaround_6323525(); 1938 #endif 1939 /* 1940 * If needed, load TOD module now so that ddi_get_time(9F) etc. work 1941 * (For now, "needed" is defined as set tod_module_name in /etc/system) 1942 */ 1943 if (tod_module_name != NULL) { 1944 PRM_POINT("load_tod_module()"); 1945 load_tod_module(tod_module_name); 1946 } 1947 1948 #if defined(__xpv) 1949 /* 1950 * Forceload interposing TOD module for the hypervisor. 1951 */ 1952 PRM_POINT("load_tod_module()"); 1953 load_tod_module("xpvtod"); 1954 #endif 1955 1956 /* 1957 * Configure the system. 1958 */ 1959 PRM_POINT("Calling configure()..."); 1960 configure(); /* set up devices */ 1961 PRM_POINT("configure() done"); 1962 1963 /* 1964 * Set the isa_list string to the defined instruction sets we 1965 * support. 1966 */ 1967 setx86isalist(); 1968 cpu_intr_alloc(CPU, NINTR_THREADS); 1969 psm_install(); 1970 1971 /* 1972 * We're done with bootops. We don't unmap the bootstrap yet because 1973 * we're still using bootsvcs. 1974 */ 1975 PRM_POINT("NULLing out bootops"); 1976 *bootopsp = (struct bootops *)NULL; 1977 bootops = (struct bootops *)NULL; 1978 1979 #if defined(__xpv) 1980 ec_init_debug_irq(); 1981 xs_domu_init(); 1982 #endif 1983 PRM_POINT("Enabling interrupts"); 1984 (*picinitf)(); 1985 sti(); 1986 #if defined(__xpv) 1987 ASSERT(CPU->cpu_m.mcpu_vcpu_info->evtchn_upcall_mask == 0); 1988 xen_late_startup(); 1989 #endif 1990 1991 (void) add_avsoftintr((void *)&softlevel1_hdl, 1, softlevel1, 1992 "softlevel1", NULL, NULL); /* XXX to be moved later */ 1993 1994 /* 1995 * Register these software interrupts for ddi timer. 1996 * Software interrupts up to the level 10 are supported. 1997 */ 1998 for (i = DDI_IPL_1; i <= DDI_IPL_10; i++) { 1999 char name[sizeof ("timer_softintr") + 2]; 2000 (void) sprintf(name, "timer_softintr%02d", i); 2001 (void) add_avsoftintr((void *)&softlevel_hdl[i-1], i, 2002 (avfunc)timer_softintr, name, (caddr_t)(uintptr_t)i, NULL); 2003 } 2004 2005 PRM_POINT("startup_end() done"); 2006 } 2007 2008 extern char hw_serial[]; 2009 char *_hs1107 = hw_serial; 2010 ulong_t _bdhs34; 2011 2012 void 2013 post_startup(void) 2014 { 2015 /* 2016 * Set the system wide, processor-specific flags to be passed 2017 * to userland via the aux vector for performance hints and 2018 * instruction set extensions. 2019 */ 2020 bind_hwcap(); 2021 2022 #ifdef __xpv 2023 if (DOMAIN_IS_INITDOMAIN(xen_info)) 2024 #endif 2025 { 2026 /* 2027 * Load the System Management BIOS into the global ksmbios 2028 * handle, if an SMBIOS is present on this system. 2029 */ 2030 ksmbios = smbios_open(NULL, SMB_VERSION, ksmbios_flags, NULL); 2031 2032 #if defined(__xpv) 2033 xpv_panic_init(); 2034 #else 2035 /* 2036 * Startup the memory scrubber. 2037 * XXPV This should be running somewhere .. 2038 */ 2039 memscrub_init(); 2040 #endif 2041 } 2042 2043 /* 2044 * Complete CPU module initialization 2045 */ 2046 cmi_post_startup(); 2047 2048 /* 2049 * Perform forceloading tasks for /etc/system. 2050 */ 2051 (void) mod_sysctl(SYS_FORCELOAD, NULL); 2052 2053 /* 2054 * ON4.0: Force /proc module in until clock interrupt handle fixed 2055 * ON4.0: This must be fixed or restated in /etc/systems. 2056 */ 2057 (void) modload("fs", "procfs"); 2058 2059 (void) i_ddi_attach_hw_nodes("pit_beep"); 2060 2061 #if defined(__i386) 2062 /* 2063 * Check for required functional Floating Point hardware, 2064 * unless FP hardware explicitly disabled. 2065 */ 2066 if (fpu_exists && (fpu_pentium_fdivbug || fp_kind == FP_NO)) 2067 halt("No working FP hardware found"); 2068 #endif 2069 2070 maxmem = freemem; 2071 2072 add_cpunode2devtree(CPU->cpu_id, CPU->cpu_m.mcpu_cpi); 2073 } 2074 2075 static int 2076 pp_in_ramdisk(page_t *pp) 2077 { 2078 extern uint64_t ramdisk_start, ramdisk_end; 2079 2080 return ((pp->p_pagenum >= btop(ramdisk_start)) && 2081 (pp->p_pagenum < btopr(ramdisk_end))); 2082 } 2083 2084 void 2085 release_bootstrap(void) 2086 { 2087 int root_is_ramdisk; 2088 page_t *pp; 2089 extern void kobj_boot_unmountroot(void); 2090 extern dev_t rootdev; 2091 2092 /* unmount boot ramdisk and release kmem usage */ 2093 kobj_boot_unmountroot(); 2094 2095 /* 2096 * We're finished using the boot loader so free its pages. 2097 */ 2098 PRM_POINT("Unmapping lower boot pages"); 2099 clear_boot_mappings(0, _userlimit); 2100 postbootkernelbase = kernelbase; 2101 2102 /* 2103 * If root isn't on ramdisk, destroy the hardcoded 2104 * ramdisk node now and release the memory. Else, 2105 * ramdisk memory is kept in rd_pages. 2106 */ 2107 root_is_ramdisk = (getmajor(rootdev) == ddi_name_to_major("ramdisk")); 2108 if (!root_is_ramdisk) { 2109 dev_info_t *dip = ddi_find_devinfo("ramdisk", -1, 0); 2110 ASSERT(dip && ddi_get_parent(dip) == ddi_root_node()); 2111 ndi_rele_devi(dip); /* held from ddi_find_devinfo */ 2112 (void) ddi_remove_child(dip, 0); 2113 } 2114 2115 PRM_POINT("Releasing boot pages"); 2116 while (bootpages) { 2117 pp = bootpages; 2118 bootpages = pp->p_next; 2119 if (root_is_ramdisk && pp_in_ramdisk(pp)) { 2120 pp->p_next = rd_pages; 2121 rd_pages = pp; 2122 continue; 2123 } 2124 pp->p_next = (struct page *)0; 2125 pp->p_prev = (struct page *)0; 2126 PP_CLRBOOTPAGES(pp); 2127 page_free(pp, 1); 2128 } 2129 PRM_POINT("Boot pages released"); 2130 2131 #if !defined(__xpv) 2132 /* XXPV -- note this following bunch of code needs to be revisited in Xen 3.0 */ 2133 /* 2134 * Find 1 page below 1 MB so that other processors can boot up. 2135 * Make sure it has a kernel VA as well as a 1:1 mapping. 2136 * We should have just free'd one up. 2137 */ 2138 if (use_mp) { 2139 pfn_t pfn; 2140 2141 for (pfn = 1; pfn < btop(1*1024*1024); pfn++) { 2142 if (page_numtopp_alloc(pfn) == NULL) 2143 continue; 2144 rm_platter_va = i86devmap(pfn, 1, 2145 PROT_READ | PROT_WRITE | PROT_EXEC); 2146 rm_platter_pa = ptob(pfn); 2147 hat_devload(kas.a_hat, 2148 (caddr_t)(uintptr_t)rm_platter_pa, MMU_PAGESIZE, 2149 pfn, PROT_READ | PROT_WRITE | PROT_EXEC, 2150 HAT_LOAD_NOCONSIST); 2151 break; 2152 } 2153 if (pfn == btop(1*1024*1024)) 2154 panic("No page available for starting " 2155 "other processors"); 2156 } 2157 #endif /* !__xpv */ 2158 } 2159 2160 /* 2161 * Initialize the platform-specific parts of a page_t. 2162 */ 2163 void 2164 add_physmem_cb(page_t *pp, pfn_t pnum) 2165 { 2166 pp->p_pagenum = pnum; 2167 pp->p_mapping = NULL; 2168 pp->p_embed = 0; 2169 pp->p_share = 0; 2170 pp->p_mlentry = 0; 2171 } 2172 2173 /* 2174 * kphysm_init() initializes physical memory. 2175 */ 2176 static pgcnt_t 2177 kphysm_init( 2178 page_t *pp, 2179 pgcnt_t npages) 2180 { 2181 struct memlist *pmem; 2182 struct memseg *cur_memseg; 2183 pfn_t base_pfn; 2184 pgcnt_t num; 2185 pgcnt_t pages_done = 0; 2186 uint64_t addr; 2187 uint64_t size; 2188 extern pfn_t ddiphysmin; 2189 2190 ASSERT(page_hash != NULL && page_hashsz != 0); 2191 2192 cur_memseg = memseg_base; 2193 for (pmem = phys_avail; pmem && npages; pmem = pmem->next) { 2194 /* 2195 * In a 32 bit kernel can't use higher memory if we're 2196 * not booting in PAE mode. This check takes care of that. 2197 */ 2198 addr = pmem->address; 2199 size = pmem->size; 2200 if (btop(addr) > physmax) 2201 continue; 2202 2203 /* 2204 * align addr and size - they may not be at page boundaries 2205 */ 2206 if ((addr & MMU_PAGEOFFSET) != 0) { 2207 addr += MMU_PAGEOFFSET; 2208 addr &= ~(uint64_t)MMU_PAGEOFFSET; 2209 size -= addr - pmem->address; 2210 } 2211 2212 /* only process pages below or equal to physmax */ 2213 if ((btop(addr + size) - 1) > physmax) 2214 size = ptob(physmax - btop(addr) + 1); 2215 2216 num = btop(size); 2217 if (num == 0) 2218 continue; 2219 2220 if (num > npages) 2221 num = npages; 2222 2223 npages -= num; 2224 pages_done += num; 2225 base_pfn = btop(addr); 2226 2227 if (prom_debug) 2228 prom_printf("MEMSEG addr=0x%" PRIx64 2229 " pgs=0x%lx pfn 0x%lx-0x%lx\n", 2230 addr, num, base_pfn, base_pfn + num); 2231 2232 /* 2233 * Ignore pages below ddiphysmin to simplify ddi memory 2234 * allocation with non-zero addr_lo requests. 2235 */ 2236 if (base_pfn < ddiphysmin) { 2237 if (base_pfn + num <= ddiphysmin) 2238 continue; 2239 pp += (ddiphysmin - base_pfn); 2240 num -= (ddiphysmin - base_pfn); 2241 base_pfn = ddiphysmin; 2242 } 2243 2244 /* 2245 * Build the memsegs entry 2246 */ 2247 cur_memseg->pages = pp; 2248 cur_memseg->epages = pp + num; 2249 cur_memseg->pages_base = base_pfn; 2250 cur_memseg->pages_end = base_pfn + num; 2251 2252 /* 2253 * Insert into memseg list in decreasing pfn range order. 2254 * Low memory is typically more fragmented such that this 2255 * ordering keeps the larger ranges at the front of the list 2256 * for code that searches memseg. 2257 * This ASSERTS that the memsegs coming in from boot are in 2258 * increasing physical address order and not contiguous. 2259 */ 2260 if (memsegs != NULL) { 2261 ASSERT(cur_memseg->pages_base >= memsegs->pages_end); 2262 cur_memseg->next = memsegs; 2263 } 2264 memsegs = cur_memseg; 2265 2266 /* 2267 * add_physmem() initializes the PSM part of the page 2268 * struct by calling the PSM back with add_physmem_cb(). 2269 * In addition it coalesces pages into larger pages as 2270 * it initializes them. 2271 */ 2272 add_physmem(pp, num, base_pfn); 2273 cur_memseg++; 2274 availrmem_initial += num; 2275 availrmem += num; 2276 2277 pp += num; 2278 } 2279 2280 PRM_DEBUG(availrmem_initial); 2281 PRM_DEBUG(availrmem); 2282 PRM_DEBUG(freemem); 2283 build_pfn_hash(); 2284 return (pages_done); 2285 } 2286 2287 /* 2288 * Kernel VM initialization. 2289 */ 2290 static void 2291 kvm_init(void) 2292 { 2293 ASSERT((((uintptr_t)s_text) & MMU_PAGEOFFSET) == 0); 2294 2295 /* 2296 * Put the kernel segments in kernel address space. 2297 */ 2298 rw_enter(&kas.a_lock, RW_WRITER); 2299 as_avlinit(&kas); 2300 2301 (void) seg_attach(&kas, s_text, e_moddata - s_text, &ktextseg); 2302 (void) segkmem_create(&ktextseg); 2303 2304 (void) seg_attach(&kas, (caddr_t)valloc_base, valloc_sz, &kvalloc); 2305 (void) segkmem_create(&kvalloc); 2306 2307 (void) seg_attach(&kas, kernelheap, 2308 ekernelheap - kernelheap, &kvseg); 2309 (void) segkmem_create(&kvseg); 2310 2311 if (core_size > 0) { 2312 PRM_POINT("attaching kvseg_core"); 2313 (void) seg_attach(&kas, (caddr_t)core_base, core_size, 2314 &kvseg_core); 2315 (void) segkmem_create(&kvseg_core); 2316 } 2317 2318 if (segziosize > 0) { 2319 PRM_POINT("attaching segzio"); 2320 (void) seg_attach(&kas, segzio_base, mmu_ptob(segziosize), 2321 &kzioseg); 2322 (void) segkmem_zio_create(&kzioseg); 2323 2324 /* create zio area covering new segment */ 2325 segkmem_zio_init(segzio_base, mmu_ptob(segziosize)); 2326 } 2327 2328 (void) seg_attach(&kas, kdi_segdebugbase, kdi_segdebugsize, &kdebugseg); 2329 (void) segkmem_create(&kdebugseg); 2330 2331 rw_exit(&kas.a_lock); 2332 2333 /* 2334 * Ensure that the red zone at kernelbase is never accessible. 2335 */ 2336 PRM_POINT("protecting redzone"); 2337 (void) as_setprot(&kas, (caddr_t)kernelbase, KERNEL_REDZONE_SIZE, 0); 2338 2339 /* 2340 * Make the text writable so that it can be hot patched by DTrace. 2341 */ 2342 (void) as_setprot(&kas, s_text, e_modtext - s_text, 2343 PROT_READ | PROT_WRITE | PROT_EXEC); 2344 2345 /* 2346 * Make data writable until end. 2347 */ 2348 (void) as_setprot(&kas, s_data, e_moddata - s_data, 2349 PROT_READ | PROT_WRITE | PROT_EXEC); 2350 } 2351 2352 #ifndef __xpv 2353 /* 2354 * Solaris adds an entry for Write Combining caching to the PAT 2355 */ 2356 static uint64_t pat_attr_reg = PAT_DEFAULT_ATTRIBUTE; 2357 2358 void 2359 pat_sync(void) 2360 { 2361 ulong_t cr0, cr0_orig, cr4; 2362 2363 if (!(x86_feature & X86_PAT)) 2364 return; 2365 cr0_orig = cr0 = getcr0(); 2366 cr4 = getcr4(); 2367 2368 /* disable caching and flush all caches and TLBs */ 2369 cr0 |= CR0_CD; 2370 cr0 &= ~CR0_NW; 2371 setcr0(cr0); 2372 invalidate_cache(); 2373 if (cr4 & CR4_PGE) { 2374 setcr4(cr4 & ~(ulong_t)CR4_PGE); 2375 setcr4(cr4); 2376 } else { 2377 reload_cr3(); 2378 } 2379 2380 /* add our entry to the PAT */ 2381 wrmsr(REG_PAT, pat_attr_reg); 2382 2383 /* flush TLBs and cache again, then reenable cr0 caching */ 2384 if (cr4 & CR4_PGE) { 2385 setcr4(cr4 & ~(ulong_t)CR4_PGE); 2386 setcr4(cr4); 2387 } else { 2388 reload_cr3(); 2389 } 2390 invalidate_cache(); 2391 setcr0(cr0_orig); 2392 } 2393 2394 #endif /* !__xpv */ 2395 2396 void 2397 get_system_configuration(void) 2398 { 2399 char prop[32]; 2400 u_longlong_t nodes_ll, cpus_pernode_ll, lvalue; 2401 2402 if (BOP_GETPROPLEN(bootops, "nodes") > sizeof (prop) || 2403 BOP_GETPROP(bootops, "nodes", prop) < 0 || 2404 kobj_getvalue(prop, &nodes_ll) == -1 || 2405 nodes_ll > MAXNODES || 2406 BOP_GETPROPLEN(bootops, "cpus_pernode") > sizeof (prop) || 2407 BOP_GETPROP(bootops, "cpus_pernode", prop) < 0 || 2408 kobj_getvalue(prop, &cpus_pernode_ll) == -1) { 2409 system_hardware.hd_nodes = 1; 2410 system_hardware.hd_cpus_per_node = 0; 2411 } else { 2412 system_hardware.hd_nodes = (int)nodes_ll; 2413 system_hardware.hd_cpus_per_node = (int)cpus_pernode_ll; 2414 } 2415 2416 if (BOP_GETPROPLEN(bootops, "kernelbase") > sizeof (prop) || 2417 BOP_GETPROP(bootops, "kernelbase", prop) < 0 || 2418 kobj_getvalue(prop, &lvalue) == -1) 2419 eprom_kernelbase = NULL; 2420 else 2421 eprom_kernelbase = (uintptr_t)lvalue; 2422 2423 if (BOP_GETPROPLEN(bootops, "segmapsize") > sizeof (prop) || 2424 BOP_GETPROP(bootops, "segmapsize", prop) < 0 || 2425 kobj_getvalue(prop, &lvalue) == -1) 2426 segmapsize = SEGMAPDEFAULT; 2427 else 2428 segmapsize = (uintptr_t)lvalue; 2429 2430 if (BOP_GETPROPLEN(bootops, "segmapfreelists") > sizeof (prop) || 2431 BOP_GETPROP(bootops, "segmapfreelists", prop) < 0 || 2432 kobj_getvalue(prop, &lvalue) == -1) 2433 segmapfreelists = 0; /* use segmap driver default */ 2434 else 2435 segmapfreelists = (int)lvalue; 2436 2437 /* physmem used to be here, but moved much earlier to fakebop.c */ 2438 } 2439 2440 /* 2441 * Add to a memory list. 2442 * start = start of new memory segment 2443 * len = length of new memory segment in bytes 2444 * new = pointer to a new struct memlist 2445 * memlistp = memory list to which to add segment. 2446 */ 2447 void 2448 memlist_add( 2449 uint64_t start, 2450 uint64_t len, 2451 struct memlist *new, 2452 struct memlist **memlistp) 2453 { 2454 struct memlist *cur; 2455 uint64_t end = start + len; 2456 2457 new->address = start; 2458 new->size = len; 2459 2460 cur = *memlistp; 2461 2462 while (cur) { 2463 if (cur->address >= end) { 2464 new->next = cur; 2465 *memlistp = new; 2466 new->prev = cur->prev; 2467 cur->prev = new; 2468 return; 2469 } 2470 ASSERT(cur->address + cur->size <= start); 2471 if (cur->next == NULL) { 2472 cur->next = new; 2473 new->prev = cur; 2474 new->next = NULL; 2475 return; 2476 } 2477 memlistp = &cur->next; 2478 cur = cur->next; 2479 } 2480 } 2481 2482 void 2483 kobj_vmem_init(vmem_t **text_arena, vmem_t **data_arena) 2484 { 2485 size_t tsize = e_modtext - modtext; 2486 size_t dsize = e_moddata - moddata; 2487 2488 *text_arena = vmem_create("module_text", tsize ? modtext : NULL, tsize, 2489 1, segkmem_alloc, segkmem_free, heaptext_arena, 0, VM_SLEEP); 2490 *data_arena = vmem_create("module_data", dsize ? moddata : NULL, dsize, 2491 1, segkmem_alloc, segkmem_free, heap32_arena, 0, VM_SLEEP); 2492 } 2493 2494 caddr_t 2495 kobj_text_alloc(vmem_t *arena, size_t size) 2496 { 2497 return (vmem_alloc(arena, size, VM_SLEEP | VM_BESTFIT)); 2498 } 2499 2500 /*ARGSUSED*/ 2501 caddr_t 2502 kobj_texthole_alloc(caddr_t addr, size_t size) 2503 { 2504 panic("unexpected call to kobj_texthole_alloc()"); 2505 /*NOTREACHED*/ 2506 return (0); 2507 } 2508 2509 /*ARGSUSED*/ 2510 void 2511 kobj_texthole_free(caddr_t addr, size_t size) 2512 { 2513 panic("unexpected call to kobj_texthole_free()"); 2514 } 2515 2516 /* 2517 * This is called just after configure() in startup(). 2518 * 2519 * The ISALIST concept is a bit hopeless on Intel, because 2520 * there's no guarantee of an ever-more-capable processor 2521 * given that various parts of the instruction set may appear 2522 * and disappear between different implementations. 2523 * 2524 * While it would be possible to correct it and even enhance 2525 * it somewhat, the explicit hardware capability bitmask allows 2526 * more flexibility. 2527 * 2528 * So, we just leave this alone. 2529 */ 2530 void 2531 setx86isalist(void) 2532 { 2533 char *tp; 2534 size_t len; 2535 extern char *isa_list; 2536 2537 #define TBUFSIZE 1024 2538 2539 tp = kmem_alloc(TBUFSIZE, KM_SLEEP); 2540 *tp = '\0'; 2541 2542 #if defined(__amd64) 2543 (void) strcpy(tp, "amd64 "); 2544 #endif 2545 2546 switch (x86_vendor) { 2547 case X86_VENDOR_Intel: 2548 case X86_VENDOR_AMD: 2549 case X86_VENDOR_TM: 2550 if (x86_feature & X86_CMOV) { 2551 /* 2552 * Pentium Pro or later 2553 */ 2554 (void) strcat(tp, "pentium_pro"); 2555 (void) strcat(tp, x86_feature & X86_MMX ? 2556 "+mmx pentium_pro " : " "); 2557 } 2558 /*FALLTHROUGH*/ 2559 case X86_VENDOR_Cyrix: 2560 /* 2561 * The Cyrix 6x86 does not have any Pentium features 2562 * accessible while not at privilege level 0. 2563 */ 2564 if (x86_feature & X86_CPUID) { 2565 (void) strcat(tp, "pentium"); 2566 (void) strcat(tp, x86_feature & X86_MMX ? 2567 "+mmx pentium " : " "); 2568 } 2569 break; 2570 default: 2571 break; 2572 } 2573 (void) strcat(tp, "i486 i386 i86"); 2574 len = strlen(tp) + 1; /* account for NULL at end of string */ 2575 isa_list = strcpy(kmem_alloc(len, KM_SLEEP), tp); 2576 kmem_free(tp, TBUFSIZE); 2577 2578 #undef TBUFSIZE 2579 } 2580 2581 2582 #ifdef __amd64 2583 2584 void * 2585 device_arena_alloc(size_t size, int vm_flag) 2586 { 2587 return (vmem_alloc(device_arena, size, vm_flag)); 2588 } 2589 2590 void 2591 device_arena_free(void *vaddr, size_t size) 2592 { 2593 vmem_free(device_arena, vaddr, size); 2594 } 2595 2596 #else /* __i386 */ 2597 2598 void * 2599 device_arena_alloc(size_t size, int vm_flag) 2600 { 2601 caddr_t vaddr; 2602 uintptr_t v; 2603 size_t start; 2604 size_t end; 2605 2606 vaddr = vmem_alloc(heap_arena, size, vm_flag); 2607 if (vaddr == NULL) 2608 return (NULL); 2609 2610 v = (uintptr_t)vaddr; 2611 ASSERT(v >= kernelbase); 2612 ASSERT(v + size <= valloc_base); 2613 2614 start = btop(v - kernelbase); 2615 end = btop(v + size - 1 - kernelbase); 2616 ASSERT(start < toxic_bit_map_len); 2617 ASSERT(end < toxic_bit_map_len); 2618 2619 while (start <= end) { 2620 BT_ATOMIC_SET(toxic_bit_map, start); 2621 ++start; 2622 } 2623 return (vaddr); 2624 } 2625 2626 void 2627 device_arena_free(void *vaddr, size_t size) 2628 { 2629 uintptr_t v = (uintptr_t)vaddr; 2630 size_t start; 2631 size_t end; 2632 2633 ASSERT(v >= kernelbase); 2634 ASSERT(v + size <= valloc_base); 2635 2636 start = btop(v - kernelbase); 2637 end = btop(v + size - 1 - kernelbase); 2638 ASSERT(start < toxic_bit_map_len); 2639 ASSERT(end < toxic_bit_map_len); 2640 2641 while (start <= end) { 2642 ASSERT(BT_TEST(toxic_bit_map, start) != 0); 2643 BT_ATOMIC_CLEAR(toxic_bit_map, start); 2644 ++start; 2645 } 2646 vmem_free(heap_arena, vaddr, size); 2647 } 2648 2649 /* 2650 * returns 1st address in range that is in device arena, or NULL 2651 * if len is not NULL it returns the length of the toxic range 2652 */ 2653 void * 2654 device_arena_contains(void *vaddr, size_t size, size_t *len) 2655 { 2656 uintptr_t v = (uintptr_t)vaddr; 2657 uintptr_t eaddr = v + size; 2658 size_t start; 2659 size_t end; 2660 2661 /* 2662 * if called very early by kmdb, just return NULL 2663 */ 2664 if (toxic_bit_map == NULL) 2665 return (NULL); 2666 2667 /* 2668 * First check if we're completely outside the bitmap range. 2669 */ 2670 if (v >= valloc_base || eaddr < kernelbase) 2671 return (NULL); 2672 2673 /* 2674 * Trim ends of search to look at only what the bitmap covers. 2675 */ 2676 if (v < kernelbase) 2677 v = kernelbase; 2678 start = btop(v - kernelbase); 2679 end = btop(eaddr - kernelbase); 2680 if (end >= toxic_bit_map_len) 2681 end = toxic_bit_map_len; 2682 2683 if (bt_range(toxic_bit_map, &start, &end, end) == 0) 2684 return (NULL); 2685 2686 v = kernelbase + ptob(start); 2687 if (len != NULL) 2688 *len = ptob(end - start); 2689 return ((void *)v); 2690 } 2691 2692 #endif /* __i386 */ 2693