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