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)GDT_VA, MMU_PAGESIZE); 1164 boot_mapin((caddr_t)DEBUG_INFO_VA, MMU_PAGESIZE); 1165 PRM_POINT("startup_memlist() done"); 1166 1167 PRM_DEBUG(valloc_sz); 1168 1169 #if defined(__amd64) 1170 if ((availrmem >> (30 - MMU_PAGESHIFT)) >= 1171 textrepl_min_gb && l2cache_sz <= 2 << 20) { 1172 extern size_t textrepl_size_thresh; 1173 textrepl_size_thresh = (16 << 20) - 1; 1174 } 1175 #endif 1176 } 1177 1178 /* 1179 * Layout the kernel's part of address space and initialize kmem allocator. 1180 */ 1181 static void 1182 startup_kmem(void) 1183 { 1184 extern void page_set_colorequiv_arr(void); 1185 1186 PRM_POINT("startup_kmem() starting..."); 1187 1188 #if defined(__amd64) 1189 if (eprom_kernelbase && eprom_kernelbase != KERNELBASE) 1190 cmn_err(CE_NOTE, "!kernelbase cannot be changed on 64-bit " 1191 "systems."); 1192 kernelbase = segkpm_base - KERNEL_REDZONE_SIZE; 1193 core_base = (uintptr_t)COREHEAP_BASE; 1194 core_size = (size_t)MISC_VA_BASE - COREHEAP_BASE; 1195 #else /* __i386 */ 1196 /* 1197 * We configure kernelbase based on: 1198 * 1199 * 1. user specified kernelbase via eeprom command. Value cannot exceed 1200 * KERNELBASE_MAX. we large page align eprom_kernelbase 1201 * 1202 * 2. Default to KERNELBASE and adjust to 2X less the size for page_t. 1203 * On large memory systems we must lower kernelbase to allow 1204 * enough room for page_t's for all of memory. 1205 * 1206 * The value set here, might be changed a little later. 1207 */ 1208 if (eprom_kernelbase) { 1209 kernelbase = eprom_kernelbase & mmu.level_mask[1]; 1210 if (kernelbase > KERNELBASE_MAX) 1211 kernelbase = KERNELBASE_MAX; 1212 } else { 1213 kernelbase = (uintptr_t)KERNELBASE; 1214 kernelbase -= ROUND_UP_4MEG(2 * valloc_sz); 1215 } 1216 ASSERT((kernelbase & mmu.level_offset[1]) == 0); 1217 core_base = valloc_base; 1218 core_size = 0; 1219 #endif /* __i386 */ 1220 1221 PRM_DEBUG(core_base); 1222 PRM_DEBUG(core_size); 1223 PRM_DEBUG(kernelbase); 1224 1225 #if defined(__i386) 1226 segkp_fromheap = 1; 1227 #endif /* __i386 */ 1228 1229 ekernelheap = (char *)core_base; 1230 PRM_DEBUG(ekernelheap); 1231 1232 /* 1233 * Now that we know the real value of kernelbase, 1234 * update variables that were initialized with a value of 1235 * KERNELBASE (in common/conf/param.c). 1236 * 1237 * XXX The problem with this sort of hackery is that the 1238 * compiler just may feel like putting the const declarations 1239 * (in param.c) into the .text section. Perhaps they should 1240 * just be declared as variables there? 1241 */ 1242 1243 *(uintptr_t *)&_kernelbase = kernelbase; 1244 *(uintptr_t *)&_userlimit = kernelbase; 1245 #if defined(__amd64) 1246 *(uintptr_t *)&_userlimit -= KERNELBASE - USERLIMIT; 1247 #else 1248 *(uintptr_t *)&_userlimit32 = _userlimit; 1249 #endif 1250 PRM_DEBUG(_kernelbase); 1251 PRM_DEBUG(_userlimit); 1252 PRM_DEBUG(_userlimit32); 1253 1254 layout_kernel_va(); 1255 1256 #if defined(__i386) 1257 /* 1258 * If segmap is too large we can push the bottom of the kernel heap 1259 * higher than the base. Or worse, it could exceed the top of the 1260 * VA space entirely, causing it to wrap around. 1261 */ 1262 if (kernelheap >= ekernelheap || (uintptr_t)kernelheap < kernelbase) 1263 panic("too little address space available for kernelheap," 1264 " use eeprom for lower kernelbase or smaller segmapsize"); 1265 #endif /* __i386 */ 1266 1267 /* 1268 * Initialize the kernel heap. Note 3rd argument must be > 1st. 1269 */ 1270 kernelheap_init(kernelheap, ekernelheap, 1271 kernelheap + MMU_PAGESIZE, 1272 (void *)core_base, (void *)(core_base + core_size)); 1273 1274 #if defined(__xpv) 1275 /* 1276 * Link pending events struct into cpu struct 1277 */ 1278 CPU->cpu_m.mcpu_evt_pend = &cpu0_evt_data; 1279 #endif 1280 /* 1281 * Initialize kernel memory allocator. 1282 */ 1283 kmem_init(); 1284 1285 /* 1286 * Factor in colorequiv to check additional 'equivalent' bins 1287 */ 1288 page_set_colorequiv_arr(); 1289 1290 /* 1291 * print this out early so that we know what's going on 1292 */ 1293 cmn_err(CE_CONT, "?features: %b\n", x86_feature, FMT_X86_FEATURE); 1294 1295 /* 1296 * Initialize bp_mapin(). 1297 */ 1298 bp_init(MMU_PAGESIZE, HAT_STORECACHING_OK); 1299 1300 /* 1301 * orig_npages is non-zero if physmem has been configured for less 1302 * than the available memory. 1303 */ 1304 if (orig_npages) { 1305 cmn_err(CE_WARN, "!%slimiting physmem to 0x%lx of 0x%lx pages", 1306 (npages == PHYSMEM ? "Due to virtual address space " : ""), 1307 npages, orig_npages); 1308 } 1309 #if defined(__i386) 1310 if (eprom_kernelbase && (eprom_kernelbase != kernelbase)) 1311 cmn_err(CE_WARN, "kernelbase value, User specified 0x%lx, " 1312 "System using 0x%lx", 1313 (uintptr_t)eprom_kernelbase, (uintptr_t)kernelbase); 1314 #endif 1315 1316 #ifdef KERNELBASE_ABI_MIN 1317 if (kernelbase < (uintptr_t)KERNELBASE_ABI_MIN) { 1318 cmn_err(CE_NOTE, "!kernelbase set to 0x%lx, system is not " 1319 "i386 ABI compliant.", (uintptr_t)kernelbase); 1320 } 1321 #endif 1322 1323 #ifdef __xpv 1324 /* 1325 * Some of the xen start information has to be relocated up 1326 * into the kernel's permanent address space. 1327 */ 1328 PRM_POINT("calling xen_relocate_start_info()"); 1329 xen_relocate_start_info(); 1330 PRM_POINT("xen_relocate_start_info() done"); 1331 1332 /* 1333 * (Update the vcpu pointer in our cpu structure to point into 1334 * the relocated shared info.) 1335 */ 1336 CPU->cpu_m.mcpu_vcpu_info = 1337 &HYPERVISOR_shared_info->vcpu_info[CPU->cpu_id]; 1338 #endif 1339 1340 PRM_POINT("startup_kmem() done"); 1341 } 1342 1343 static void 1344 startup_modules(void) 1345 { 1346 unsigned int i; 1347 extern void prom_setup(void); 1348 1349 PRM_POINT("startup_modules() starting..."); 1350 1351 #ifndef __xpv 1352 /* 1353 * Initialize ten-micro second timer so that drivers will 1354 * not get short changed in their init phase. This was 1355 * not getting called until clkinit which, on fast cpu's 1356 * caused the drv_usecwait to be way too short. 1357 */ 1358 microfind(); 1359 #endif 1360 1361 /* 1362 * Read the GMT lag from /etc/rtc_config. 1363 */ 1364 sgmtl(process_rtc_config_file()); 1365 1366 /* 1367 * Calculate default settings of system parameters based upon 1368 * maxusers, yet allow to be overridden via the /etc/system file. 1369 */ 1370 param_calc(0); 1371 1372 mod_setup(); 1373 1374 /* 1375 * Initialize system parameters. 1376 */ 1377 param_init(); 1378 1379 /* 1380 * Initialize the default brands 1381 */ 1382 brand_init(); 1383 1384 /* 1385 * maxmem is the amount of physical memory we're playing with. 1386 */ 1387 maxmem = physmem; 1388 1389 /* 1390 * Initialize segment management stuff. 1391 */ 1392 seg_init(); 1393 1394 if (modload("fs", "specfs") == -1) 1395 halt("Can't load specfs"); 1396 1397 if (modload("fs", "devfs") == -1) 1398 halt("Can't load devfs"); 1399 1400 if (modload("fs", "dev") == -1) 1401 halt("Can't load dev"); 1402 1403 (void) modloadonly("sys", "lbl_edition"); 1404 1405 dispinit(); 1406 1407 /* 1408 * This is needed here to initialize hw_serial[] for cluster booting. 1409 */ 1410 if ((i = modload("misc", "sysinit")) != (unsigned int)-1) 1411 (void) modunload(i); 1412 else 1413 cmn_err(CE_CONT, "sysinit load failed"); 1414 1415 /* Read cluster configuration data. */ 1416 clconf_init(); 1417 1418 #if defined(__xpv) 1419 ec_init(); 1420 gnttab_init(); 1421 (void) xs_early_init(); 1422 #endif /* __xpv */ 1423 1424 /* 1425 * Create a kernel device tree. First, create rootnex and 1426 * then invoke bus specific code to probe devices. 1427 */ 1428 setup_ddi(); 1429 1430 #ifndef __xpv 1431 { 1432 /* 1433 * Set up the CPU module subsystem. Modifies the device tree, 1434 * so it must be done after setup_ddi(). 1435 */ 1436 1437 cmi_hdl_t hdl; 1438 1439 if ((hdl = cmi_init(CMI_HDL_NATIVE, cmi_ntv_hwchipid(CPU), 1440 cmi_ntv_hwcoreid(CPU), cmi_ntv_hwstrandid(CPU))) != NULL) { 1441 if (x86_feature & X86_MCA) 1442 cmi_mca_init(hdl); 1443 } 1444 } 1445 #endif /* __xpv */ 1446 1447 /* 1448 * Fake a prom tree such that /dev/openprom continues to work 1449 */ 1450 PRM_POINT("startup_modules: calling prom_setup..."); 1451 prom_setup(); 1452 PRM_POINT("startup_modules: done"); 1453 1454 /* 1455 * Load all platform specific modules 1456 */ 1457 PRM_POINT("startup_modules: calling psm_modload..."); 1458 psm_modload(); 1459 1460 PRM_POINT("startup_modules() done"); 1461 } 1462 1463 /* 1464 * claim a "setaside" boot page for use in the kernel 1465 */ 1466 page_t * 1467 boot_claim_page(pfn_t pfn) 1468 { 1469 page_t *pp; 1470 1471 pp = page_numtopp_nolock(pfn); 1472 ASSERT(pp != NULL); 1473 1474 if (PP_ISBOOTPAGES(pp)) { 1475 if (pp->p_next != NULL) 1476 pp->p_next->p_prev = pp->p_prev; 1477 if (pp->p_prev == NULL) 1478 bootpages = pp->p_next; 1479 else 1480 pp->p_prev->p_next = pp->p_next; 1481 } else { 1482 /* 1483 * htable_attach() expects a base pagesize page 1484 */ 1485 if (pp->p_szc != 0) 1486 page_boot_demote(pp); 1487 pp = page_numtopp(pfn, SE_EXCL); 1488 } 1489 return (pp); 1490 } 1491 1492 /* 1493 * Walk through the pagetables looking for pages mapped in by boot. If the 1494 * setaside flag is set the pages are expected to be returned to the 1495 * kernel later in boot, so we add them to the bootpages list. 1496 */ 1497 static void 1498 protect_boot_range(uintptr_t low, uintptr_t high, int setaside) 1499 { 1500 uintptr_t va = low; 1501 size_t len; 1502 uint_t prot; 1503 pfn_t pfn; 1504 page_t *pp; 1505 pgcnt_t boot_protect_cnt = 0; 1506 1507 while (kbm_probe(&va, &len, &pfn, &prot) != 0 && va < high) { 1508 if (va + len >= high) 1509 panic("0x%lx byte mapping at 0x%p exceeds boot's " 1510 "legal range.", len, (void *)va); 1511 1512 while (len > 0) { 1513 pp = page_numtopp_alloc(pfn); 1514 if (pp != NULL) { 1515 if (setaside == 0) 1516 panic("Unexpected mapping by boot. " 1517 "addr=%p pfn=%lx\n", 1518 (void *)va, pfn); 1519 1520 pp->p_next = bootpages; 1521 pp->p_prev = NULL; 1522 PP_SETBOOTPAGES(pp); 1523 if (bootpages != NULL) { 1524 bootpages->p_prev = pp; 1525 } 1526 bootpages = pp; 1527 ++boot_protect_cnt; 1528 } 1529 1530 ++pfn; 1531 len -= MMU_PAGESIZE; 1532 va += MMU_PAGESIZE; 1533 } 1534 } 1535 PRM_DEBUG(boot_protect_cnt); 1536 } 1537 1538 /* 1539 * 1540 */ 1541 static void 1542 layout_kernel_va(void) 1543 { 1544 PRM_POINT("layout_kernel_va() starting..."); 1545 /* 1546 * Establish the final size of the kernel's heap, size of segmap, 1547 * segkp, etc. 1548 */ 1549 1550 #if defined(__amd64) 1551 1552 kpm_vbase = (caddr_t)segkpm_base; 1553 kpm_size = ROUND_UP_LPAGE(mmu_ptob(physmax + 1)); 1554 if ((uintptr_t)kpm_vbase + kpm_size > (uintptr_t)valloc_base) 1555 panic("not enough room for kpm!"); 1556 PRM_DEBUG(kpm_size); 1557 PRM_DEBUG(kpm_vbase); 1558 1559 /* 1560 * By default we create a seg_kp in 64 bit kernels, it's a little 1561 * faster to access than embedding it in the heap. 1562 */ 1563 segkp_base = (caddr_t)valloc_base + valloc_sz; 1564 if (!segkp_fromheap) { 1565 size_t sz = mmu_ptob(segkpsize); 1566 1567 /* 1568 * determine size of segkp 1569 */ 1570 if (sz < SEGKPMINSIZE || sz > SEGKPMAXSIZE) { 1571 sz = SEGKPDEFSIZE; 1572 cmn_err(CE_WARN, "!Illegal value for segkpsize. " 1573 "segkpsize has been reset to %ld pages", 1574 mmu_btop(sz)); 1575 } 1576 sz = MIN(sz, MAX(SEGKPMINSIZE, mmu_ptob(physmem))); 1577 1578 segkpsize = mmu_btop(ROUND_UP_LPAGE(sz)); 1579 } 1580 PRM_DEBUG(segkp_base); 1581 PRM_DEBUG(segkpsize); 1582 1583 /* 1584 * segzio is used for ZFS cached data. It uses a distinct VA 1585 * segment (from kernel heap) so that we can easily tell not to 1586 * include it in kernel crash dumps on 64 bit kernels. The trick is 1587 * to give it lots of VA, but not constrain the kernel heap. 1588 * We scale the size of segzio linearly with physmem up to 1589 * SEGZIOMAXSIZE. Above that amount it scales at 50% of physmem. 1590 */ 1591 segzio_base = segkp_base + mmu_ptob(segkpsize); 1592 if (segzio_fromheap) { 1593 segziosize = 0; 1594 } else { 1595 size_t physmem_size = mmu_ptob(physmem); 1596 size_t size = (segziosize == 0) ? 1597 physmem_size : mmu_ptob(segziosize); 1598 1599 if (size < SEGZIOMINSIZE) 1600 size = SEGZIOMINSIZE; 1601 if (size > SEGZIOMAXSIZE) { 1602 size = SEGZIOMAXSIZE; 1603 if (physmem_size > size) 1604 size += (physmem_size - size) / 2; 1605 } 1606 segziosize = mmu_btop(ROUND_UP_LPAGE(size)); 1607 } 1608 PRM_DEBUG(segziosize); 1609 PRM_DEBUG(segzio_base); 1610 1611 /* 1612 * Put the range of VA for device mappings next, kmdb knows to not 1613 * grep in this range of addresses. 1614 */ 1615 toxic_addr = 1616 ROUND_UP_LPAGE((uintptr_t)segzio_base + mmu_ptob(segziosize)); 1617 PRM_DEBUG(toxic_addr); 1618 segmap_start = ROUND_UP_LPAGE(toxic_addr + toxic_size); 1619 #else /* __i386 */ 1620 segmap_start = ROUND_UP_LPAGE(kernelbase); 1621 #endif /* __i386 */ 1622 PRM_DEBUG(segmap_start); 1623 1624 /* 1625 * Users can change segmapsize through eeprom or /etc/system. 1626 * If the variable is tuned through eeprom, there is no upper 1627 * bound on the size of segmap. If it is tuned through 1628 * /etc/system on 32-bit systems, it must be no larger than we 1629 * planned for in startup_memlist(). 1630 */ 1631 segmapsize = MAX(ROUND_UP_LPAGE(segmapsize), SEGMAPDEFAULT); 1632 1633 #if defined(__i386) 1634 /* 1635 * 32-bit systems don't have segkpm or segkp, so segmap appears at 1636 * the bottom of the kernel's address range. Set aside space for a 1637 * small red zone just below the start of segmap. 1638 */ 1639 segmap_start += KERNEL_REDZONE_SIZE; 1640 segmapsize -= KERNEL_REDZONE_SIZE; 1641 #endif 1642 1643 PRM_DEBUG(segmap_start); 1644 PRM_DEBUG(segmapsize); 1645 kernelheap = (caddr_t)ROUND_UP_LPAGE(segmap_start + segmapsize); 1646 PRM_DEBUG(kernelheap); 1647 PRM_POINT("layout_kernel_va() done..."); 1648 } 1649 1650 /* 1651 * Finish initializing the VM system, now that we are no longer 1652 * relying on the boot time memory allocators. 1653 */ 1654 static void 1655 startup_vm(void) 1656 { 1657 struct segmap_crargs a; 1658 1659 extern int use_brk_lpg, use_stk_lpg; 1660 1661 PRM_POINT("startup_vm() starting..."); 1662 1663 /* 1664 * Initialize the hat layer. 1665 */ 1666 hat_init(); 1667 1668 /* 1669 * Do final allocations of HAT data structures that need to 1670 * be allocated before quiescing the boot loader. 1671 */ 1672 PRM_POINT("Calling hat_kern_alloc()..."); 1673 hat_kern_alloc((caddr_t)segmap_start, segmapsize, ekernelheap); 1674 PRM_POINT("hat_kern_alloc() done"); 1675 1676 #ifndef __xpv 1677 /* 1678 * Setup Page Attribute Table 1679 */ 1680 pat_sync(); 1681 #endif 1682 1683 /* 1684 * The next two loops are done in distinct steps in order 1685 * to be sure that any page that is doubly mapped (both above 1686 * KERNEL_TEXT and below kernelbase) is dealt with correctly. 1687 * Note this may never happen, but it might someday. 1688 */ 1689 bootpages = NULL; 1690 PRM_POINT("Protecting boot pages"); 1691 1692 /* 1693 * Protect any pages mapped above KERNEL_TEXT that somehow have 1694 * page_t's. This can only happen if something weird allocated 1695 * in this range (like kadb/kmdb). 1696 */ 1697 protect_boot_range(KERNEL_TEXT, (uintptr_t)-1, 0); 1698 1699 /* 1700 * Before we can take over memory allocation/mapping from the boot 1701 * loader we must remove from our free page lists any boot allocated 1702 * pages that stay mapped until release_bootstrap(). 1703 */ 1704 protect_boot_range(0, kernelbase, 1); 1705 1706 1707 /* 1708 * Switch to running on regular HAT (not boot_mmu) 1709 */ 1710 PRM_POINT("Calling hat_kern_setup()..."); 1711 hat_kern_setup(); 1712 1713 /* 1714 * It is no longer safe to call BOP_ALLOC(), so make sure we don't. 1715 */ 1716 bop_no_more_mem(); 1717 1718 PRM_POINT("hat_kern_setup() done"); 1719 1720 hat_cpu_online(CPU); 1721 1722 /* 1723 * Initialize VM system 1724 */ 1725 PRM_POINT("Calling kvm_init()..."); 1726 kvm_init(); 1727 PRM_POINT("kvm_init() done"); 1728 1729 /* 1730 * Tell kmdb that the VM system is now working 1731 */ 1732 if (boothowto & RB_DEBUG) 1733 kdi_dvec_vmready(); 1734 1735 #if defined(__xpv) 1736 /* 1737 * Populate the I/O pool on domain 0 1738 */ 1739 if (DOMAIN_IS_INITDOMAIN(xen_info)) { 1740 extern long populate_io_pool(void); 1741 long init_io_pool_cnt; 1742 1743 PRM_POINT("Populating reserve I/O page pool"); 1744 init_io_pool_cnt = populate_io_pool(); 1745 PRM_DEBUG(init_io_pool_cnt); 1746 } 1747 #endif 1748 /* 1749 * Mangle the brand string etc. 1750 */ 1751 cpuid_pass3(CPU); 1752 1753 #if defined(__amd64) 1754 1755 /* 1756 * Create the device arena for toxic (to dtrace/kmdb) mappings. 1757 */ 1758 device_arena = vmem_create("device", (void *)toxic_addr, 1759 toxic_size, MMU_PAGESIZE, NULL, NULL, NULL, 0, VM_SLEEP); 1760 1761 #else /* __i386 */ 1762 1763 /* 1764 * allocate the bit map that tracks toxic pages 1765 */ 1766 toxic_bit_map_len = btop((ulong_t)(valloc_base - kernelbase)); 1767 PRM_DEBUG(toxic_bit_map_len); 1768 toxic_bit_map = 1769 kmem_zalloc(BT_SIZEOFMAP(toxic_bit_map_len), KM_NOSLEEP); 1770 ASSERT(toxic_bit_map != NULL); 1771 PRM_DEBUG(toxic_bit_map); 1772 1773 #endif /* __i386 */ 1774 1775 1776 /* 1777 * Now that we've got more VA, as well as the ability to allocate from 1778 * it, tell the debugger. 1779 */ 1780 if (boothowto & RB_DEBUG) 1781 kdi_dvec_memavail(); 1782 1783 /* 1784 * The following code installs a special page fault handler (#pf) 1785 * to work around a pentium bug. 1786 */ 1787 #if !defined(__amd64) && !defined(__xpv) 1788 if (x86_type == X86_TYPE_P5) { 1789 desctbr_t idtr; 1790 gate_desc_t *newidt; 1791 struct machcpu *mcpu = &CPU->cpu_m; 1792 1793 if ((newidt = kmem_zalloc(MMU_PAGESIZE, KM_NOSLEEP)) == NULL) 1794 panic("failed to install pentium_pftrap"); 1795 1796 bcopy(idt0, newidt, sizeof (idt0)); 1797 set_gatesegd(&newidt[T_PGFLT], &pentium_pftrap, 1798 KCS_SEL, SDT_SYSIGT, TRP_KPL); 1799 1800 (void) as_setprot(&kas, (caddr_t)newidt, MMU_PAGESIZE, 1801 PROT_READ|PROT_EXEC); 1802 1803 mcpu->mcpu_idt = newidt; 1804 idtr.dtr_base = (uintptr_t)mcpu->mcpu_idt; 1805 idtr.dtr_limit = sizeof (idt0) - 1; 1806 wr_idtr(&idtr); 1807 } 1808 #endif /* !__amd64 */ 1809 1810 #if !defined(__xpv) 1811 /* 1812 * Map page pfn=0 for drivers, such as kd, that need to pick up 1813 * parameters left there by controllers/BIOS. 1814 */ 1815 PRM_POINT("setup up p0_va"); 1816 p0_va = i86devmap(0, 1, PROT_READ); 1817 PRM_DEBUG(p0_va); 1818 #endif 1819 1820 cmn_err(CE_CONT, "?mem = %luK (0x%lx)\n", 1821 physinstalled << (MMU_PAGESHIFT - 10), ptob(physinstalled)); 1822 1823 /* 1824 * disable automatic large pages for small memory systems or 1825 * when the disable flag is set. 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 if (mmu.max_page_level > 0) { 1841 mcntl0_lpsize = LEVEL_SIZE(1); 1842 } 1843 1844 PRM_POINT("Calling hat_init_finish()..."); 1845 hat_init_finish(); 1846 PRM_POINT("hat_init_finish() done"); 1847 1848 /* 1849 * Initialize the segkp segment type. 1850 */ 1851 rw_enter(&kas.a_lock, RW_WRITER); 1852 PRM_POINT("Attaching segkp"); 1853 if (segkp_fromheap) { 1854 segkp->s_as = &kas; 1855 } else if (seg_attach(&kas, (caddr_t)segkp_base, mmu_ptob(segkpsize), 1856 segkp) < 0) { 1857 panic("startup: cannot attach segkp"); 1858 /*NOTREACHED*/ 1859 } 1860 PRM_POINT("Doing segkp_create()"); 1861 if (segkp_create(segkp) != 0) { 1862 panic("startup: segkp_create failed"); 1863 /*NOTREACHED*/ 1864 } 1865 PRM_DEBUG(segkp); 1866 rw_exit(&kas.a_lock); 1867 1868 /* 1869 * kpm segment 1870 */ 1871 segmap_kpm = 0; 1872 if (kpm_desired) { 1873 kpm_init(); 1874 kpm_enable = 1; 1875 vpm_enable = 1; 1876 } 1877 1878 /* 1879 * Now create segmap segment. 1880 */ 1881 rw_enter(&kas.a_lock, RW_WRITER); 1882 if (seg_attach(&kas, (caddr_t)segmap_start, segmapsize, segmap) < 0) { 1883 panic("cannot attach segmap"); 1884 /*NOTREACHED*/ 1885 } 1886 PRM_DEBUG(segmap); 1887 1888 a.prot = PROT_READ | PROT_WRITE; 1889 a.shmsize = 0; 1890 a.nfreelist = segmapfreelists; 1891 1892 if (segmap_create(segmap, (caddr_t)&a) != 0) 1893 panic("segmap_create segmap"); 1894 rw_exit(&kas.a_lock); 1895 1896 setup_vaddr_for_ppcopy(CPU); 1897 1898 segdev_init(); 1899 #if defined(__xpv) 1900 if (DOMAIN_IS_INITDOMAIN(xen_info)) 1901 #endif 1902 pmem_init(); 1903 1904 PRM_POINT("startup_vm() done"); 1905 } 1906 1907 /* 1908 * Load a tod module for the non-standard tod part found on this system. 1909 */ 1910 static void 1911 load_tod_module(char *todmod) 1912 { 1913 if (modload("tod", todmod) == -1) 1914 halt("Can't load TOD module"); 1915 } 1916 1917 static void 1918 startup_end(void) 1919 { 1920 int i; 1921 extern void setx86isalist(void); 1922 1923 PRM_POINT("startup_end() starting..."); 1924 1925 /* 1926 * Perform tasks that get done after most of the VM 1927 * initialization has been done but before the clock 1928 * and other devices get started. 1929 */ 1930 kern_setup1(); 1931 1932 /* 1933 * Perform CPC initialization for this CPU. 1934 */ 1935 kcpc_hw_init(CPU); 1936 1937 #if defined(OPTERON_WORKAROUND_6323525) 1938 if (opteron_workaround_6323525) 1939 patch_workaround_6323525(); 1940 #endif 1941 /* 1942 * If needed, load TOD module now so that ddi_get_time(9F) etc. work 1943 * (For now, "needed" is defined as set tod_module_name in /etc/system) 1944 */ 1945 if (tod_module_name != NULL) { 1946 PRM_POINT("load_tod_module()"); 1947 load_tod_module(tod_module_name); 1948 } 1949 1950 #if defined(__xpv) 1951 /* 1952 * Forceload interposing TOD module for the hypervisor. 1953 */ 1954 PRM_POINT("load_tod_module()"); 1955 load_tod_module("xpvtod"); 1956 #endif 1957 1958 /* 1959 * Configure the system. 1960 */ 1961 PRM_POINT("Calling configure()..."); 1962 configure(); /* set up devices */ 1963 PRM_POINT("configure() done"); 1964 1965 /* 1966 * Set the isa_list string to the defined instruction sets we 1967 * support. 1968 */ 1969 setx86isalist(); 1970 cpu_intr_alloc(CPU, NINTR_THREADS); 1971 psm_install(); 1972 1973 /* 1974 * We're done with bootops. We don't unmap the bootstrap yet because 1975 * we're still using bootsvcs. 1976 */ 1977 PRM_POINT("NULLing out bootops"); 1978 *bootopsp = (struct bootops *)NULL; 1979 bootops = (struct bootops *)NULL; 1980 1981 #if defined(__xpv) 1982 ec_init_debug_irq(); 1983 xs_domu_init(); 1984 #endif 1985 PRM_POINT("Enabling interrupts"); 1986 (*picinitf)(); 1987 sti(); 1988 #if defined(__xpv) 1989 ASSERT(CPU->cpu_m.mcpu_vcpu_info->evtchn_upcall_mask == 0); 1990 xen_late_startup(); 1991 #endif 1992 1993 (void) add_avsoftintr((void *)&softlevel1_hdl, 1, softlevel1, 1994 "softlevel1", NULL, NULL); /* XXX to be moved later */ 1995 1996 /* 1997 * Register these software interrupts for ddi timer. 1998 * Software interrupts up to the level 10 are supported. 1999 */ 2000 for (i = DDI_IPL_1; i <= DDI_IPL_10; i++) { 2001 char name[sizeof ("timer_softintr") + 2]; 2002 (void) sprintf(name, "timer_softintr%02d", i); 2003 (void) add_avsoftintr((void *)&softlevel_hdl[i-1], i, 2004 (avfunc)timer_softintr, name, (caddr_t)(uintptr_t)i, NULL); 2005 } 2006 2007 PRM_POINT("startup_end() done"); 2008 } 2009 2010 extern char hw_serial[]; 2011 char *_hs1107 = hw_serial; 2012 ulong_t _bdhs34; 2013 2014 void 2015 post_startup(void) 2016 { 2017 /* 2018 * Set the system wide, processor-specific flags to be passed 2019 * to userland via the aux vector for performance hints and 2020 * instruction set extensions. 2021 */ 2022 bind_hwcap(); 2023 2024 #ifdef __xpv 2025 if (DOMAIN_IS_INITDOMAIN(xen_info)) 2026 #endif 2027 { 2028 /* 2029 * Load the System Management BIOS into the global ksmbios 2030 * handle, if an SMBIOS is present on this system. 2031 */ 2032 ksmbios = smbios_open(NULL, SMB_VERSION, ksmbios_flags, NULL); 2033 2034 #if defined(__xpv) 2035 xpv_panic_init(); 2036 #else 2037 /* 2038 * Startup the memory scrubber. 2039 * XXPV This should be running somewhere .. 2040 */ 2041 memscrub_init(); 2042 #endif 2043 } 2044 2045 /* 2046 * Complete CPU module initialization 2047 */ 2048 cmi_post_startup(); 2049 2050 /* 2051 * Perform forceloading tasks for /etc/system. 2052 */ 2053 (void) mod_sysctl(SYS_FORCELOAD, NULL); 2054 2055 /* 2056 * ON4.0: Force /proc module in until clock interrupt handle fixed 2057 * ON4.0: This must be fixed or restated in /etc/systems. 2058 */ 2059 (void) modload("fs", "procfs"); 2060 2061 (void) i_ddi_attach_hw_nodes("pit_beep"); 2062 2063 #if defined(__i386) 2064 /* 2065 * Check for required functional Floating Point hardware, 2066 * unless FP hardware explicitly disabled. 2067 */ 2068 if (fpu_exists && (fpu_pentium_fdivbug || fp_kind == FP_NO)) 2069 halt("No working FP hardware found"); 2070 #endif 2071 2072 maxmem = freemem; 2073 2074 add_cpunode2devtree(CPU->cpu_id, CPU->cpu_m.mcpu_cpi); 2075 } 2076 2077 static int 2078 pp_in_ramdisk(page_t *pp) 2079 { 2080 extern uint64_t ramdisk_start, ramdisk_end; 2081 2082 return ((pp->p_pagenum >= btop(ramdisk_start)) && 2083 (pp->p_pagenum < btopr(ramdisk_end))); 2084 } 2085 2086 void 2087 release_bootstrap(void) 2088 { 2089 int root_is_ramdisk; 2090 page_t *pp; 2091 extern void kobj_boot_unmountroot(void); 2092 extern dev_t rootdev; 2093 2094 /* unmount boot ramdisk and release kmem usage */ 2095 kobj_boot_unmountroot(); 2096 2097 /* 2098 * We're finished using the boot loader so free its pages. 2099 */ 2100 PRM_POINT("Unmapping lower boot pages"); 2101 clear_boot_mappings(0, _userlimit); 2102 postbootkernelbase = kernelbase; 2103 2104 /* 2105 * If root isn't on ramdisk, destroy the hardcoded 2106 * ramdisk node now and release the memory. Else, 2107 * ramdisk memory is kept in rd_pages. 2108 */ 2109 root_is_ramdisk = (getmajor(rootdev) == ddi_name_to_major("ramdisk")); 2110 if (!root_is_ramdisk) { 2111 dev_info_t *dip = ddi_find_devinfo("ramdisk", -1, 0); 2112 ASSERT(dip && ddi_get_parent(dip) == ddi_root_node()); 2113 ndi_rele_devi(dip); /* held from ddi_find_devinfo */ 2114 (void) ddi_remove_child(dip, 0); 2115 } 2116 2117 PRM_POINT("Releasing boot pages"); 2118 while (bootpages) { 2119 pp = bootpages; 2120 bootpages = pp->p_next; 2121 if (root_is_ramdisk && pp_in_ramdisk(pp)) { 2122 pp->p_next = rd_pages; 2123 rd_pages = pp; 2124 continue; 2125 } 2126 pp->p_next = (struct page *)0; 2127 pp->p_prev = (struct page *)0; 2128 PP_CLRBOOTPAGES(pp); 2129 page_free(pp, 1); 2130 } 2131 PRM_POINT("Boot pages released"); 2132 2133 #if !defined(__xpv) 2134 /* XXPV -- note this following bunch of code needs to be revisited in Xen 3.0 */ 2135 /* 2136 * Find 1 page below 1 MB so that other processors can boot up. 2137 * Make sure it has a kernel VA as well as a 1:1 mapping. 2138 * We should have just free'd one up. 2139 */ 2140 if (use_mp) { 2141 pfn_t pfn; 2142 2143 for (pfn = 1; pfn < btop(1*1024*1024); pfn++) { 2144 if (page_numtopp_alloc(pfn) == NULL) 2145 continue; 2146 rm_platter_va = i86devmap(pfn, 1, 2147 PROT_READ | PROT_WRITE | PROT_EXEC); 2148 rm_platter_pa = ptob(pfn); 2149 hat_devload(kas.a_hat, 2150 (caddr_t)(uintptr_t)rm_platter_pa, MMU_PAGESIZE, 2151 pfn, PROT_READ | PROT_WRITE | PROT_EXEC, 2152 HAT_LOAD_NOCONSIST); 2153 break; 2154 } 2155 if (pfn == btop(1*1024*1024)) 2156 panic("No page available for starting " 2157 "other processors"); 2158 } 2159 #endif /* !__xpv */ 2160 } 2161 2162 /* 2163 * Initialize the platform-specific parts of a page_t. 2164 */ 2165 void 2166 add_physmem_cb(page_t *pp, pfn_t pnum) 2167 { 2168 pp->p_pagenum = pnum; 2169 pp->p_mapping = NULL; 2170 pp->p_embed = 0; 2171 pp->p_share = 0; 2172 pp->p_mlentry = 0; 2173 } 2174 2175 /* 2176 * kphysm_init() initializes physical memory. 2177 */ 2178 static pgcnt_t 2179 kphysm_init( 2180 page_t *pp, 2181 pgcnt_t npages) 2182 { 2183 struct memlist *pmem; 2184 struct memseg *cur_memseg; 2185 pfn_t base_pfn; 2186 pgcnt_t num; 2187 pgcnt_t pages_done = 0; 2188 uint64_t addr; 2189 uint64_t size; 2190 extern pfn_t ddiphysmin; 2191 2192 ASSERT(page_hash != NULL && page_hashsz != 0); 2193 2194 cur_memseg = memseg_base; 2195 for (pmem = phys_avail; pmem && npages; pmem = pmem->next) { 2196 /* 2197 * In a 32 bit kernel can't use higher memory if we're 2198 * not booting in PAE mode. This check takes care of that. 2199 */ 2200 addr = pmem->address; 2201 size = pmem->size; 2202 if (btop(addr) > physmax) 2203 continue; 2204 2205 /* 2206 * align addr and size - they may not be at page boundaries 2207 */ 2208 if ((addr & MMU_PAGEOFFSET) != 0) { 2209 addr += MMU_PAGEOFFSET; 2210 addr &= ~(uint64_t)MMU_PAGEOFFSET; 2211 size -= addr - pmem->address; 2212 } 2213 2214 /* only process pages below or equal to physmax */ 2215 if ((btop(addr + size) - 1) > physmax) 2216 size = ptob(physmax - btop(addr) + 1); 2217 2218 num = btop(size); 2219 if (num == 0) 2220 continue; 2221 2222 if (num > npages) 2223 num = npages; 2224 2225 npages -= num; 2226 pages_done += num; 2227 base_pfn = btop(addr); 2228 2229 if (prom_debug) 2230 prom_printf("MEMSEG addr=0x%" PRIx64 2231 " pgs=0x%lx pfn 0x%lx-0x%lx\n", 2232 addr, num, base_pfn, base_pfn + num); 2233 2234 /* 2235 * Ignore pages below ddiphysmin to simplify ddi memory 2236 * allocation with non-zero addr_lo requests. 2237 */ 2238 if (base_pfn < ddiphysmin) { 2239 if (base_pfn + num <= ddiphysmin) 2240 continue; 2241 pp += (ddiphysmin - base_pfn); 2242 num -= (ddiphysmin - base_pfn); 2243 base_pfn = ddiphysmin; 2244 } 2245 2246 /* 2247 * Build the memsegs entry 2248 */ 2249 cur_memseg->pages = pp; 2250 cur_memseg->epages = pp + num; 2251 cur_memseg->pages_base = base_pfn; 2252 cur_memseg->pages_end = base_pfn + num; 2253 2254 /* 2255 * Insert into memseg list in decreasing pfn range order. 2256 * Low memory is typically more fragmented such that this 2257 * ordering keeps the larger ranges at the front of the list 2258 * for code that searches memseg. 2259 * This ASSERTS that the memsegs coming in from boot are in 2260 * increasing physical address order and not contiguous. 2261 */ 2262 if (memsegs != NULL) { 2263 ASSERT(cur_memseg->pages_base >= memsegs->pages_end); 2264 cur_memseg->next = memsegs; 2265 } 2266 memsegs = cur_memseg; 2267 2268 /* 2269 * add_physmem() initializes the PSM part of the page 2270 * struct by calling the PSM back with add_physmem_cb(). 2271 * In addition it coalesces pages into larger pages as 2272 * it initializes them. 2273 */ 2274 add_physmem(pp, num, base_pfn); 2275 cur_memseg++; 2276 availrmem_initial += num; 2277 availrmem += num; 2278 2279 pp += num; 2280 } 2281 2282 PRM_DEBUG(availrmem_initial); 2283 PRM_DEBUG(availrmem); 2284 PRM_DEBUG(freemem); 2285 build_pfn_hash(); 2286 return (pages_done); 2287 } 2288 2289 /* 2290 * Kernel VM initialization. 2291 */ 2292 static void 2293 kvm_init(void) 2294 { 2295 ASSERT((((uintptr_t)s_text) & MMU_PAGEOFFSET) == 0); 2296 2297 /* 2298 * Put the kernel segments in kernel address space. 2299 */ 2300 rw_enter(&kas.a_lock, RW_WRITER); 2301 as_avlinit(&kas); 2302 2303 (void) seg_attach(&kas, s_text, e_moddata - s_text, &ktextseg); 2304 (void) segkmem_create(&ktextseg); 2305 2306 (void) seg_attach(&kas, (caddr_t)valloc_base, valloc_sz, &kvalloc); 2307 (void) segkmem_create(&kvalloc); 2308 2309 (void) seg_attach(&kas, kernelheap, 2310 ekernelheap - kernelheap, &kvseg); 2311 (void) segkmem_create(&kvseg); 2312 2313 if (core_size > 0) { 2314 PRM_POINT("attaching kvseg_core"); 2315 (void) seg_attach(&kas, (caddr_t)core_base, core_size, 2316 &kvseg_core); 2317 (void) segkmem_create(&kvseg_core); 2318 } 2319 2320 if (segziosize > 0) { 2321 PRM_POINT("attaching segzio"); 2322 (void) seg_attach(&kas, segzio_base, mmu_ptob(segziosize), 2323 &kzioseg); 2324 (void) segkmem_zio_create(&kzioseg); 2325 2326 /* create zio area covering new segment */ 2327 segkmem_zio_init(segzio_base, mmu_ptob(segziosize)); 2328 } 2329 2330 (void) seg_attach(&kas, kdi_segdebugbase, kdi_segdebugsize, &kdebugseg); 2331 (void) segkmem_create(&kdebugseg); 2332 2333 rw_exit(&kas.a_lock); 2334 2335 /* 2336 * Ensure that the red zone at kernelbase is never accessible. 2337 */ 2338 PRM_POINT("protecting redzone"); 2339 (void) as_setprot(&kas, (caddr_t)kernelbase, KERNEL_REDZONE_SIZE, 0); 2340 2341 /* 2342 * Make the text writable so that it can be hot patched by DTrace. 2343 */ 2344 (void) as_setprot(&kas, s_text, e_modtext - s_text, 2345 PROT_READ | PROT_WRITE | PROT_EXEC); 2346 2347 /* 2348 * Make data writable until end. 2349 */ 2350 (void) as_setprot(&kas, s_data, e_moddata - s_data, 2351 PROT_READ | PROT_WRITE | PROT_EXEC); 2352 } 2353 2354 #ifndef __xpv 2355 /* 2356 * Solaris adds an entry for Write Combining caching to the PAT 2357 */ 2358 static uint64_t pat_attr_reg = PAT_DEFAULT_ATTRIBUTE; 2359 2360 void 2361 pat_sync(void) 2362 { 2363 ulong_t cr0, cr0_orig, cr4; 2364 2365 if (!(x86_feature & X86_PAT)) 2366 return; 2367 cr0_orig = cr0 = getcr0(); 2368 cr4 = getcr4(); 2369 2370 /* disable caching and flush all caches and TLBs */ 2371 cr0 |= CR0_CD; 2372 cr0 &= ~CR0_NW; 2373 setcr0(cr0); 2374 invalidate_cache(); 2375 if (cr4 & CR4_PGE) { 2376 setcr4(cr4 & ~(ulong_t)CR4_PGE); 2377 setcr4(cr4); 2378 } else { 2379 reload_cr3(); 2380 } 2381 2382 /* add our entry to the PAT */ 2383 wrmsr(REG_PAT, pat_attr_reg); 2384 2385 /* flush TLBs and cache again, then reenable cr0 caching */ 2386 if (cr4 & CR4_PGE) { 2387 setcr4(cr4 & ~(ulong_t)CR4_PGE); 2388 setcr4(cr4); 2389 } else { 2390 reload_cr3(); 2391 } 2392 invalidate_cache(); 2393 setcr0(cr0_orig); 2394 } 2395 2396 #endif /* !__xpv */ 2397 2398 void 2399 get_system_configuration(void) 2400 { 2401 char prop[32]; 2402 u_longlong_t nodes_ll, cpus_pernode_ll, lvalue; 2403 2404 if (BOP_GETPROPLEN(bootops, "nodes") > sizeof (prop) || 2405 BOP_GETPROP(bootops, "nodes", prop) < 0 || 2406 kobj_getvalue(prop, &nodes_ll) == -1 || 2407 nodes_ll > MAXNODES || 2408 BOP_GETPROPLEN(bootops, "cpus_pernode") > sizeof (prop) || 2409 BOP_GETPROP(bootops, "cpus_pernode", prop) < 0 || 2410 kobj_getvalue(prop, &cpus_pernode_ll) == -1) { 2411 system_hardware.hd_nodes = 1; 2412 system_hardware.hd_cpus_per_node = 0; 2413 } else { 2414 system_hardware.hd_nodes = (int)nodes_ll; 2415 system_hardware.hd_cpus_per_node = (int)cpus_pernode_ll; 2416 } 2417 2418 if (BOP_GETPROPLEN(bootops, "kernelbase") > sizeof (prop) || 2419 BOP_GETPROP(bootops, "kernelbase", prop) < 0 || 2420 kobj_getvalue(prop, &lvalue) == -1) 2421 eprom_kernelbase = NULL; 2422 else 2423 eprom_kernelbase = (uintptr_t)lvalue; 2424 2425 if (BOP_GETPROPLEN(bootops, "segmapsize") > sizeof (prop) || 2426 BOP_GETPROP(bootops, "segmapsize", prop) < 0 || 2427 kobj_getvalue(prop, &lvalue) == -1) 2428 segmapsize = SEGMAPDEFAULT; 2429 else 2430 segmapsize = (uintptr_t)lvalue; 2431 2432 if (BOP_GETPROPLEN(bootops, "segmapfreelists") > sizeof (prop) || 2433 BOP_GETPROP(bootops, "segmapfreelists", prop) < 0 || 2434 kobj_getvalue(prop, &lvalue) == -1) 2435 segmapfreelists = 0; /* use segmap driver default */ 2436 else 2437 segmapfreelists = (int)lvalue; 2438 2439 /* physmem used to be here, but moved much earlier to fakebop.c */ 2440 } 2441 2442 /* 2443 * Add to a memory list. 2444 * start = start of new memory segment 2445 * len = length of new memory segment in bytes 2446 * new = pointer to a new struct memlist 2447 * memlistp = memory list to which to add segment. 2448 */ 2449 void 2450 memlist_add( 2451 uint64_t start, 2452 uint64_t len, 2453 struct memlist *new, 2454 struct memlist **memlistp) 2455 { 2456 struct memlist *cur; 2457 uint64_t end = start + len; 2458 2459 new->address = start; 2460 new->size = len; 2461 2462 cur = *memlistp; 2463 2464 while (cur) { 2465 if (cur->address >= end) { 2466 new->next = cur; 2467 *memlistp = new; 2468 new->prev = cur->prev; 2469 cur->prev = new; 2470 return; 2471 } 2472 ASSERT(cur->address + cur->size <= start); 2473 if (cur->next == NULL) { 2474 cur->next = new; 2475 new->prev = cur; 2476 new->next = NULL; 2477 return; 2478 } 2479 memlistp = &cur->next; 2480 cur = cur->next; 2481 } 2482 } 2483 2484 void 2485 kobj_vmem_init(vmem_t **text_arena, vmem_t **data_arena) 2486 { 2487 size_t tsize = e_modtext - modtext; 2488 size_t dsize = e_moddata - moddata; 2489 2490 *text_arena = vmem_create("module_text", tsize ? modtext : NULL, tsize, 2491 1, segkmem_alloc, segkmem_free, heaptext_arena, 0, VM_SLEEP); 2492 *data_arena = vmem_create("module_data", dsize ? moddata : NULL, dsize, 2493 1, segkmem_alloc, segkmem_free, heap32_arena, 0, VM_SLEEP); 2494 } 2495 2496 caddr_t 2497 kobj_text_alloc(vmem_t *arena, size_t size) 2498 { 2499 return (vmem_alloc(arena, size, VM_SLEEP | VM_BESTFIT)); 2500 } 2501 2502 /*ARGSUSED*/ 2503 caddr_t 2504 kobj_texthole_alloc(caddr_t addr, size_t size) 2505 { 2506 panic("unexpected call to kobj_texthole_alloc()"); 2507 /*NOTREACHED*/ 2508 return (0); 2509 } 2510 2511 /*ARGSUSED*/ 2512 void 2513 kobj_texthole_free(caddr_t addr, size_t size) 2514 { 2515 panic("unexpected call to kobj_texthole_free()"); 2516 } 2517 2518 /* 2519 * This is called just after configure() in startup(). 2520 * 2521 * The ISALIST concept is a bit hopeless on Intel, because 2522 * there's no guarantee of an ever-more-capable processor 2523 * given that various parts of the instruction set may appear 2524 * and disappear between different implementations. 2525 * 2526 * While it would be possible to correct it and even enhance 2527 * it somewhat, the explicit hardware capability bitmask allows 2528 * more flexibility. 2529 * 2530 * So, we just leave this alone. 2531 */ 2532 void 2533 setx86isalist(void) 2534 { 2535 char *tp; 2536 size_t len; 2537 extern char *isa_list; 2538 2539 #define TBUFSIZE 1024 2540 2541 tp = kmem_alloc(TBUFSIZE, KM_SLEEP); 2542 *tp = '\0'; 2543 2544 #if defined(__amd64) 2545 (void) strcpy(tp, "amd64 "); 2546 #endif 2547 2548 switch (x86_vendor) { 2549 case X86_VENDOR_Intel: 2550 case X86_VENDOR_AMD: 2551 case X86_VENDOR_TM: 2552 if (x86_feature & X86_CMOV) { 2553 /* 2554 * Pentium Pro or later 2555 */ 2556 (void) strcat(tp, "pentium_pro"); 2557 (void) strcat(tp, x86_feature & X86_MMX ? 2558 "+mmx pentium_pro " : " "); 2559 } 2560 /*FALLTHROUGH*/ 2561 case X86_VENDOR_Cyrix: 2562 /* 2563 * The Cyrix 6x86 does not have any Pentium features 2564 * accessible while not at privilege level 0. 2565 */ 2566 if (x86_feature & X86_CPUID) { 2567 (void) strcat(tp, "pentium"); 2568 (void) strcat(tp, x86_feature & X86_MMX ? 2569 "+mmx pentium " : " "); 2570 } 2571 break; 2572 default: 2573 break; 2574 } 2575 (void) strcat(tp, "i486 i386 i86"); 2576 len = strlen(tp) + 1; /* account for NULL at end of string */ 2577 isa_list = strcpy(kmem_alloc(len, KM_SLEEP), tp); 2578 kmem_free(tp, TBUFSIZE); 2579 2580 #undef TBUFSIZE 2581 } 2582 2583 2584 #ifdef __amd64 2585 2586 void * 2587 device_arena_alloc(size_t size, int vm_flag) 2588 { 2589 return (vmem_alloc(device_arena, size, vm_flag)); 2590 } 2591 2592 void 2593 device_arena_free(void *vaddr, size_t size) 2594 { 2595 vmem_free(device_arena, vaddr, size); 2596 } 2597 2598 #else /* __i386 */ 2599 2600 void * 2601 device_arena_alloc(size_t size, int vm_flag) 2602 { 2603 caddr_t vaddr; 2604 uintptr_t v; 2605 size_t start; 2606 size_t end; 2607 2608 vaddr = vmem_alloc(heap_arena, size, vm_flag); 2609 if (vaddr == NULL) 2610 return (NULL); 2611 2612 v = (uintptr_t)vaddr; 2613 ASSERT(v >= kernelbase); 2614 ASSERT(v + size <= valloc_base); 2615 2616 start = btop(v - kernelbase); 2617 end = btop(v + size - 1 - kernelbase); 2618 ASSERT(start < toxic_bit_map_len); 2619 ASSERT(end < toxic_bit_map_len); 2620 2621 while (start <= end) { 2622 BT_ATOMIC_SET(toxic_bit_map, start); 2623 ++start; 2624 } 2625 return (vaddr); 2626 } 2627 2628 void 2629 device_arena_free(void *vaddr, size_t size) 2630 { 2631 uintptr_t v = (uintptr_t)vaddr; 2632 size_t start; 2633 size_t end; 2634 2635 ASSERT(v >= kernelbase); 2636 ASSERT(v + size <= valloc_base); 2637 2638 start = btop(v - kernelbase); 2639 end = btop(v + size - 1 - kernelbase); 2640 ASSERT(start < toxic_bit_map_len); 2641 ASSERT(end < toxic_bit_map_len); 2642 2643 while (start <= end) { 2644 ASSERT(BT_TEST(toxic_bit_map, start) != 0); 2645 BT_ATOMIC_CLEAR(toxic_bit_map, start); 2646 ++start; 2647 } 2648 vmem_free(heap_arena, vaddr, size); 2649 } 2650 2651 /* 2652 * returns 1st address in range that is in device arena, or NULL 2653 * if len is not NULL it returns the length of the toxic range 2654 */ 2655 void * 2656 device_arena_contains(void *vaddr, size_t size, size_t *len) 2657 { 2658 uintptr_t v = (uintptr_t)vaddr; 2659 uintptr_t eaddr = v + size; 2660 size_t start; 2661 size_t end; 2662 2663 /* 2664 * if called very early by kmdb, just return NULL 2665 */ 2666 if (toxic_bit_map == NULL) 2667 return (NULL); 2668 2669 /* 2670 * First check if we're completely outside the bitmap range. 2671 */ 2672 if (v >= valloc_base || eaddr < kernelbase) 2673 return (NULL); 2674 2675 /* 2676 * Trim ends of search to look at only what the bitmap covers. 2677 */ 2678 if (v < kernelbase) 2679 v = kernelbase; 2680 start = btop(v - kernelbase); 2681 end = btop(eaddr - kernelbase); 2682 if (end >= toxic_bit_map_len) 2683 end = toxic_bit_map_len; 2684 2685 if (bt_range(toxic_bit_map, &start, &end, end) == 0) 2686 return (NULL); 2687 2688 v = kernelbase + ptob(start); 2689 if (len != NULL) 2690 *len = ptob(end - start); 2691 return ((void *)v); 2692 } 2693 2694 #endif /* __i386 */ 2695