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 (c) 1993, 2010, Oracle and/or its affiliates. All rights reserved. 23 * Copyright 2012 DEY Storage Systems, Inc. All rights reserved. 24 * Copyright 2013 Nexenta Systems, Inc. All rights reserved. 25 * Copyright 2013 Joyent, Inc. All rights reserved. 26 */ 27 /* 28 * Copyright (c) 2010, Intel Corporation. 29 * All rights reserved. 30 */ 31 32 #include <sys/types.h> 33 #include <sys/t_lock.h> 34 #include <sys/param.h> 35 #include <sys/sysmacros.h> 36 #include <sys/signal.h> 37 #include <sys/systm.h> 38 #include <sys/user.h> 39 #include <sys/mman.h> 40 #include <sys/vm.h> 41 #include <sys/conf.h> 42 #include <sys/avintr.h> 43 #include <sys/autoconf.h> 44 #include <sys/disp.h> 45 #include <sys/class.h> 46 #include <sys/bitmap.h> 47 48 #include <sys/privregs.h> 49 50 #include <sys/proc.h> 51 #include <sys/buf.h> 52 #include <sys/kmem.h> 53 #include <sys/mem.h> 54 #include <sys/kstat.h> 55 56 #include <sys/reboot.h> 57 58 #include <sys/cred.h> 59 #include <sys/vnode.h> 60 #include <sys/file.h> 61 62 #include <sys/procfs.h> 63 64 #include <sys/vfs.h> 65 #include <sys/cmn_err.h> 66 #include <sys/utsname.h> 67 #include <sys/debug.h> 68 #include <sys/kdi.h> 69 70 #include <sys/dumphdr.h> 71 #include <sys/bootconf.h> 72 #include <sys/memlist_plat.h> 73 #include <sys/varargs.h> 74 #include <sys/promif.h> 75 #include <sys/modctl.h> 76 77 #include <sys/sunddi.h> 78 #include <sys/sunndi.h> 79 #include <sys/ndi_impldefs.h> 80 #include <sys/ddidmareq.h> 81 #include <sys/psw.h> 82 #include <sys/regset.h> 83 #include <sys/clock.h> 84 #include <sys/pte.h> 85 #include <sys/tss.h> 86 #include <sys/stack.h> 87 #include <sys/trap.h> 88 #include <sys/fp.h> 89 #include <vm/kboot_mmu.h> 90 #include <vm/anon.h> 91 #include <vm/as.h> 92 #include <vm/page.h> 93 #include <vm/seg.h> 94 #include <vm/seg_dev.h> 95 #include <vm/seg_kmem.h> 96 #include <vm/seg_kpm.h> 97 #include <vm/seg_map.h> 98 #include <vm/seg_vn.h> 99 #include <vm/seg_kp.h> 100 #include <sys/memnode.h> 101 #include <vm/vm_dep.h> 102 #include <sys/thread.h> 103 #include <sys/sysconf.h> 104 #include <sys/vm_machparam.h> 105 #include <sys/archsystm.h> 106 #include <sys/machsystm.h> 107 #include <vm/hat.h> 108 #include <vm/hat_i86.h> 109 #include <sys/pmem.h> 110 #include <sys/smp_impldefs.h> 111 #include <sys/x86_archext.h> 112 #include <sys/cpuvar.h> 113 #include <sys/segments.h> 114 #include <sys/clconf.h> 115 #include <sys/kobj.h> 116 #include <sys/kobj_lex.h> 117 #include <sys/cpc_impl.h> 118 #include <sys/cpu_module.h> 119 #include <sys/smbios.h> 120 #include <sys/debug_info.h> 121 #include <sys/bootinfo.h> 122 #include <sys/ddi_periodic.h> 123 #include <sys/systeminfo.h> 124 #include <sys/multiboot.h> 125 #include <sys/ramdisk.h> 126 127 #ifdef __xpv 128 129 #include <sys/hypervisor.h> 130 #include <sys/xen_mmu.h> 131 #include <sys/evtchn_impl.h> 132 #include <sys/gnttab.h> 133 #include <sys/xpv_panic.h> 134 #include <xen/sys/xenbus_comms.h> 135 #include <xen/public/physdev.h> 136 137 extern void xen_late_startup(void); 138 139 struct xen_evt_data cpu0_evt_data; 140 141 #else /* __xpv */ 142 #include <sys/memlist_impl.h> 143 144 extern void mem_config_init(void); 145 #endif /* __xpv */ 146 147 extern void progressbar_init(void); 148 extern void brand_init(void); 149 extern void pcf_init(void); 150 extern void pg_init(void); 151 152 extern int size_pse_array(pgcnt_t, int); 153 154 #if defined(_SOFT_HOSTID) 155 156 #include <sys/rtc.h> 157 158 static int32_t set_soft_hostid(void); 159 static char hostid_file[] = "/etc/hostid"; 160 161 #endif 162 163 void *gfx_devinfo_list; 164 165 #if defined(__amd64) && !defined(__xpv) 166 extern void immu_startup(void); 167 #endif 168 169 /* 170 * XXX make declaration below "static" when drivers no longer use this 171 * interface. 172 */ 173 extern caddr_t p0_va; /* Virtual address for accessing physical page 0 */ 174 175 /* 176 * segkp 177 */ 178 extern int segkp_fromheap; 179 180 static void kvm_init(void); 181 static void startup_init(void); 182 static void startup_memlist(void); 183 static void startup_kmem(void); 184 static void startup_modules(void); 185 static void startup_vm(void); 186 static void startup_end(void); 187 static void layout_kernel_va(void); 188 189 /* 190 * Declare these as initialized data so we can patch them. 191 */ 192 #ifdef __i386 193 194 /* 195 * Due to virtual address space limitations running in 32 bit mode, restrict 196 * the amount of physical memory configured to a max of PHYSMEM pages (16g). 197 * 198 * If the physical max memory size of 64g were allowed to be configured, the 199 * size of user virtual address space will be less than 1g. A limited user 200 * address space greatly reduces the range of applications that can run. 201 * 202 * If more physical memory than PHYSMEM is required, users should preferably 203 * run in 64 bit mode which has far looser virtual address space limitations. 204 * 205 * If 64 bit mode is not available (as in IA32) and/or more physical memory 206 * than PHYSMEM is required in 32 bit mode, physmem can be set to the desired 207 * value or to 0 (to configure all available memory) via eeprom(1M). kernelbase 208 * should also be carefully tuned to balance out the need of the user 209 * application while minimizing the risk of kernel heap exhaustion due to 210 * kernelbase being set too high. 211 */ 212 #define PHYSMEM 0x400000 213 214 #else /* __amd64 */ 215 216 /* 217 * For now we can handle memory with physical addresses up to about 218 * 64 Terabytes. This keeps the kernel above the VA hole, leaving roughly 219 * half the VA space for seg_kpm. When systems get bigger than 64TB this 220 * code will need revisiting. There is an implicit assumption that there 221 * are no *huge* holes in the physical address space too. 222 */ 223 #define TERABYTE (1ul << 40) 224 #define PHYSMEM_MAX64 mmu_btop(64 * TERABYTE) 225 #define PHYSMEM PHYSMEM_MAX64 226 #define AMD64_VA_HOLE_END 0xFFFF800000000000ul 227 228 #endif /* __amd64 */ 229 230 pgcnt_t physmem = PHYSMEM; 231 pgcnt_t obp_pages; /* Memory used by PROM for its text and data */ 232 233 char *kobj_file_buf; 234 int kobj_file_bufsize; /* set in /etc/system */ 235 236 /* Global variables for MP support. Used in mp_startup */ 237 caddr_t rm_platter_va = 0; 238 uint32_t rm_platter_pa; 239 240 int auto_lpg_disable = 1; 241 242 /* 243 * Some CPUs have holes in the middle of the 64-bit virtual address range. 244 */ 245 uintptr_t hole_start, hole_end; 246 247 /* 248 * kpm mapping window 249 */ 250 caddr_t kpm_vbase; 251 size_t kpm_size; 252 static int kpm_desired; 253 #ifdef __amd64 254 static uintptr_t segkpm_base = (uintptr_t)SEGKPM_BASE; 255 #endif 256 257 /* 258 * Configuration parameters set at boot time. 259 */ 260 261 caddr_t econtig; /* end of first block of contiguous kernel */ 262 263 struct bootops *bootops = 0; /* passed in from boot */ 264 struct bootops **bootopsp; 265 struct boot_syscalls *sysp; /* passed in from boot */ 266 267 char bootblock_fstype[16]; 268 269 char kern_bootargs[OBP_MAXPATHLEN]; 270 char kern_bootfile[OBP_MAXPATHLEN]; 271 272 /* 273 * ZFS zio segment. This allows us to exclude large portions of ZFS data that 274 * gets cached in kmem caches on the heap. If this is set to zero, we allocate 275 * zio buffers from their own segment, otherwise they are allocated from the 276 * heap. The optimization of allocating zio buffers from their own segment is 277 * only valid on 64-bit kernels. 278 */ 279 #if defined(__amd64) 280 int segzio_fromheap = 0; 281 #else 282 int segzio_fromheap = 1; 283 #endif 284 285 /* 286 * new memory fragmentations are possible in startup() due to BOP_ALLOCs. this 287 * depends on number of BOP_ALLOC calls made and requested size, memory size 288 * combination and whether boot.bin memory needs to be freed. 289 */ 290 #define POSS_NEW_FRAGMENTS 12 291 292 /* 293 * VM data structures 294 */ 295 long page_hashsz; /* Size of page hash table (power of two) */ 296 unsigned int page_hashsz_shift; /* log2(page_hashsz) */ 297 struct page *pp_base; /* Base of initial system page struct array */ 298 struct page **page_hash; /* Page hash table */ 299 pad_mutex_t *pse_mutex; /* Locks protecting pp->p_selock */ 300 size_t pse_table_size; /* Number of mutexes in pse_mutex[] */ 301 int pse_shift; /* log2(pse_table_size) */ 302 struct seg ktextseg; /* Segment used for kernel executable image */ 303 struct seg kvalloc; /* Segment used for "valloc" mapping */ 304 struct seg kpseg; /* Segment used for pageable kernel virt mem */ 305 struct seg kmapseg; /* Segment used for generic kernel mappings */ 306 struct seg kdebugseg; /* Segment used for the kernel debugger */ 307 308 struct seg *segkmap = &kmapseg; /* Kernel generic mapping segment */ 309 static struct seg *segmap = &kmapseg; /* easier to use name for in here */ 310 311 struct seg *segkp = &kpseg; /* Pageable kernel virtual memory segment */ 312 313 #if defined(__amd64) 314 struct seg kvseg_core; /* Segment used for the core heap */ 315 struct seg kpmseg; /* Segment used for physical mapping */ 316 struct seg *segkpm = &kpmseg; /* 64bit kernel physical mapping segment */ 317 #else 318 struct seg *segkpm = NULL; /* Unused on IA32 */ 319 #endif 320 321 caddr_t segkp_base; /* Base address of segkp */ 322 caddr_t segzio_base; /* Base address of segzio */ 323 #if defined(__amd64) 324 pgcnt_t segkpsize = btop(SEGKPDEFSIZE); /* size of segkp segment in pages */ 325 #else 326 pgcnt_t segkpsize = 0; 327 #endif 328 pgcnt_t segziosize = 0; /* size of zio segment in pages */ 329 330 /* 331 * A static DR page_t VA map is reserved that can map the page structures 332 * for a domain's entire RA space. The pages that back this space are 333 * dynamically allocated and need not be physically contiguous. The DR 334 * map size is derived from KPM size. 335 * This mechanism isn't used by x86 yet, so just stubs here. 336 */ 337 int ppvm_enable = 0; /* Static virtual map for page structs */ 338 page_t *ppvm_base = NULL; /* Base of page struct map */ 339 pgcnt_t ppvm_size = 0; /* Size of page struct map */ 340 341 /* 342 * VA range available to the debugger 343 */ 344 const caddr_t kdi_segdebugbase = (const caddr_t)SEGDEBUGBASE; 345 const size_t kdi_segdebugsize = SEGDEBUGSIZE; 346 347 struct memseg *memseg_base; 348 struct vnode unused_pages_vp; 349 350 #define FOURGB 0x100000000LL 351 352 struct memlist *memlist; 353 354 caddr_t s_text; /* start of kernel text segment */ 355 caddr_t e_text; /* end of kernel text segment */ 356 caddr_t s_data; /* start of kernel data segment */ 357 caddr_t e_data; /* end of kernel data segment */ 358 caddr_t modtext; /* start of loadable module text reserved */ 359 caddr_t e_modtext; /* end of loadable module text reserved */ 360 caddr_t moddata; /* start of loadable module data reserved */ 361 caddr_t e_moddata; /* end of loadable module data reserved */ 362 363 struct memlist *phys_install; /* Total installed physical memory */ 364 struct memlist *phys_avail; /* Total available physical memory */ 365 struct memlist *bios_rsvd; /* Bios reserved memory */ 366 367 /* 368 * kphysm_init returns the number of pages that were processed 369 */ 370 static pgcnt_t kphysm_init(page_t *, pgcnt_t); 371 372 #define IO_PROP_SIZE 64 /* device property size */ 373 374 /* 375 * a couple useful roundup macros 376 */ 377 #define ROUND_UP_PAGE(x) \ 378 ((uintptr_t)P2ROUNDUP((uintptr_t)(x), (uintptr_t)MMU_PAGESIZE)) 379 #define ROUND_UP_LPAGE(x) \ 380 ((uintptr_t)P2ROUNDUP((uintptr_t)(x), mmu.level_size[1])) 381 #define ROUND_UP_4MEG(x) \ 382 ((uintptr_t)P2ROUNDUP((uintptr_t)(x), (uintptr_t)FOUR_MEG)) 383 #define ROUND_UP_TOPLEVEL(x) \ 384 ((uintptr_t)P2ROUNDUP((uintptr_t)(x), mmu.level_size[mmu.max_level])) 385 386 /* 387 * 32-bit Kernel's Virtual memory layout. 388 * +-----------------------+ 389 * | | 390 * 0xFFC00000 -|-----------------------|- ARGSBASE 391 * | debugger | 392 * 0xFF800000 -|-----------------------|- SEGDEBUGBASE 393 * | Kernel Data | 394 * 0xFEC00000 -|-----------------------| 395 * | Kernel Text | 396 * 0xFE800000 -|-----------------------|- KERNEL_TEXT (0xFB400000 on Xen) 397 * |--- GDT ---|- GDT page (GDT_VA) 398 * |--- debug info ---|- debug info (DEBUG_INFO_VA) 399 * | | 400 * | page_t structures | 401 * | memsegs, memlists, | 402 * | page hash, etc. | 403 * --- -|-----------------------|- ekernelheap, valloc_base (floating) 404 * | | (segkp is just an arena in the heap) 405 * | | 406 * | kvseg | 407 * | | 408 * | | 409 * --- -|-----------------------|- kernelheap (floating) 410 * | Segkmap | 411 * 0xC3002000 -|-----------------------|- segmap_start (floating) 412 * | Red Zone | 413 * 0xC3000000 -|-----------------------|- kernelbase / userlimit (floating) 414 * | | || 415 * | Shared objects | \/ 416 * | | 417 * : : 418 * | user data | 419 * |-----------------------| 420 * | user text | 421 * 0x08048000 -|-----------------------| 422 * | user stack | 423 * : : 424 * | invalid | 425 * 0x00000000 +-----------------------+ 426 * 427 * 428 * 64-bit Kernel's Virtual memory layout. (assuming 64 bit app) 429 * +-----------------------+ 430 * | | 431 * 0xFFFFFFFF.FFC00000 |-----------------------|- ARGSBASE 432 * | debugger (?) | 433 * 0xFFFFFFFF.FF800000 |-----------------------|- SEGDEBUGBASE 434 * | unused | 435 * +-----------------------+ 436 * | Kernel Data | 437 * 0xFFFFFFFF.FBC00000 |-----------------------| 438 * | Kernel Text | 439 * 0xFFFFFFFF.FB800000 |-----------------------|- KERNEL_TEXT 440 * |--- GDT ---|- GDT page (GDT_VA) 441 * |--- debug info ---|- debug info (DEBUG_INFO_VA) 442 * | | 443 * | Core heap | (used for loadable modules) 444 * 0xFFFFFFFF.C0000000 |-----------------------|- core_base / ekernelheap 445 * | Kernel | 446 * | heap | 447 * 0xFFFFFXXX.XXX00000 |-----------------------|- kernelheap (floating) 448 * | segmap | 449 * 0xFFFFFXXX.XXX00000 |-----------------------|- segmap_start (floating) 450 * | device mappings | 451 * 0xFFFFFXXX.XXX00000 |-----------------------|- toxic_addr (floating) 452 * | segzio | 453 * 0xFFFFFXXX.XXX00000 |-----------------------|- segzio_base (floating) 454 * | segkp | 455 * --- |-----------------------|- segkp_base (floating) 456 * | page_t structures | valloc_base + valloc_sz 457 * | memsegs, memlists, | 458 * | page hash, etc. | 459 * 0xFFFFFF00.00000000 |-----------------------|- valloc_base (lower if > 1TB) 460 * | segkpm | 461 * 0xFFFFFE00.00000000 |-----------------------| 462 * | Red Zone | 463 * 0xFFFFFD80.00000000 |-----------------------|- KERNELBASE (lower if > 1TB) 464 * | User stack |- User space memory 465 * | | 466 * | shared objects, etc | (grows downwards) 467 * : : 468 * | | 469 * 0xFFFF8000.00000000 |-----------------------| 470 * | | 471 * | VA Hole / unused | 472 * | | 473 * 0x00008000.00000000 |-----------------------| 474 * | | 475 * | | 476 * : : 477 * | user heap | (grows upwards) 478 * | | 479 * | user data | 480 * |-----------------------| 481 * | user text | 482 * 0x00000000.04000000 |-----------------------| 483 * | invalid | 484 * 0x00000000.00000000 +-----------------------+ 485 * 486 * A 32 bit app on the 64 bit kernel sees the same layout as on the 32 bit 487 * kernel, except that userlimit is raised to 0xfe000000 488 * 489 * Floating values: 490 * 491 * valloc_base: start of the kernel's memory management/tracking data 492 * structures. This region contains page_t structures for 493 * physical memory, memsegs, memlists, and the page hash. 494 * 495 * core_base: start of the kernel's "core" heap area on 64-bit systems. 496 * This area is intended to be used for global data as well as for module 497 * text/data that does not fit into the nucleus pages. The core heap is 498 * restricted to a 2GB range, allowing every address within it to be 499 * accessed using rip-relative addressing 500 * 501 * ekernelheap: end of kernelheap and start of segmap. 502 * 503 * kernelheap: start of kernel heap. On 32-bit systems, this starts right 504 * above a red zone that separates the user's address space from the 505 * kernel's. On 64-bit systems, it sits above segkp and segkpm. 506 * 507 * segmap_start: start of segmap. The length of segmap can be modified 508 * through eeprom. The default length is 16MB on 32-bit systems and 64MB 509 * on 64-bit systems. 510 * 511 * kernelbase: On a 32-bit kernel the default value of 0xd4000000 will be 512 * decreased by 2X the size required for page_t. This allows the kernel 513 * heap to grow in size with physical memory. With sizeof(page_t) == 80 514 * bytes, the following shows the values of kernelbase and kernel heap 515 * sizes for different memory configurations (assuming default segmap and 516 * segkp sizes). 517 * 518 * mem size for kernelbase kernel heap 519 * size page_t's size 520 * ---- --------- ---------- ----------- 521 * 1gb 0x01400000 0xd1800000 684MB 522 * 2gb 0x02800000 0xcf000000 704MB 523 * 4gb 0x05000000 0xca000000 744MB 524 * 6gb 0x07800000 0xc5000000 784MB 525 * 8gb 0x0a000000 0xc0000000 824MB 526 * 16gb 0x14000000 0xac000000 984MB 527 * 32gb 0x28000000 0x84000000 1304MB 528 * 64gb 0x50000000 0x34000000 1944MB (*) 529 * 530 * kernelbase is less than the abi minimum of 0xc0000000 for memory 531 * configurations above 8gb. 532 * 533 * (*) support for memory configurations above 32gb will require manual tuning 534 * of kernelbase to balance out the need of user applications. 535 */ 536 537 /* real-time-clock initialization parameters */ 538 extern time_t process_rtc_config_file(void); 539 540 uintptr_t kernelbase; 541 uintptr_t postbootkernelbase; /* not set till boot loader is gone */ 542 uintptr_t eprom_kernelbase; 543 size_t segmapsize; 544 uintptr_t segmap_start; 545 int segmapfreelists; 546 pgcnt_t npages; 547 pgcnt_t orig_npages; 548 size_t core_size; /* size of "core" heap */ 549 uintptr_t core_base; /* base address of "core" heap */ 550 551 /* 552 * List of bootstrap pages. We mark these as allocated in startup. 553 * release_bootstrap() will free them when we're completely done with 554 * the bootstrap. 555 */ 556 static page_t *bootpages; 557 558 /* 559 * boot time pages that have a vnode from the ramdisk will keep that forever. 560 */ 561 static page_t *rd_pages; 562 563 /* 564 * Lower 64K 565 */ 566 static page_t *lower_pages = NULL; 567 static int lower_pages_count = 0; 568 569 struct system_hardware system_hardware; 570 571 /* 572 * Enable some debugging messages concerning memory usage... 573 */ 574 static void 575 print_memlist(char *title, struct memlist *mp) 576 { 577 prom_printf("MEMLIST: %s:\n", title); 578 while (mp != NULL) { 579 prom_printf("\tAddress 0x%" PRIx64 ", size 0x%" PRIx64 "\n", 580 mp->ml_address, mp->ml_size); 581 mp = mp->ml_next; 582 } 583 } 584 585 /* 586 * XX64 need a comment here.. are these just default values, surely 587 * we read the "cpuid" type information to figure this out. 588 */ 589 int l2cache_sz = 0x80000; 590 int l2cache_linesz = 0x40; 591 int l2cache_assoc = 1; 592 593 static size_t textrepl_min_gb = 10; 594 595 /* 596 * on 64 bit we use a predifined VA range for mapping devices in the kernel 597 * on 32 bit the mappings are intermixed in the heap, so we use a bit map 598 */ 599 #ifdef __amd64 600 601 vmem_t *device_arena; 602 uintptr_t toxic_addr = (uintptr_t)NULL; 603 size_t toxic_size = 1024 * 1024 * 1024; /* Sparc uses 1 gig too */ 604 605 #else /* __i386 */ 606 607 ulong_t *toxic_bit_map; /* one bit for each 4k of VA in heap_arena */ 608 size_t toxic_bit_map_len = 0; /* in bits */ 609 610 #endif /* __i386 */ 611 612 /* 613 * Simple boot time debug facilities 614 */ 615 static char *prm_dbg_str[] = { 616 "%s:%d: '%s' is 0x%x\n", 617 "%s:%d: '%s' is 0x%llx\n" 618 }; 619 620 int prom_debug; 621 622 #define PRM_DEBUG(q) if (prom_debug) \ 623 prom_printf(prm_dbg_str[sizeof (q) >> 3], "startup.c", __LINE__, #q, q); 624 #define PRM_POINT(q) if (prom_debug) \ 625 prom_printf("%s:%d: %s\n", "startup.c", __LINE__, q); 626 627 /* 628 * This structure is used to keep track of the intial allocations 629 * done in startup_memlist(). The value of NUM_ALLOCATIONS needs to 630 * be >= the number of ADD_TO_ALLOCATIONS() executed in the code. 631 */ 632 #define NUM_ALLOCATIONS 8 633 int num_allocations = 0; 634 struct { 635 void **al_ptr; 636 size_t al_size; 637 } allocations[NUM_ALLOCATIONS]; 638 size_t valloc_sz = 0; 639 uintptr_t valloc_base; 640 641 #define ADD_TO_ALLOCATIONS(ptr, size) { \ 642 size = ROUND_UP_PAGE(size); \ 643 if (num_allocations == NUM_ALLOCATIONS) \ 644 panic("too many ADD_TO_ALLOCATIONS()"); \ 645 allocations[num_allocations].al_ptr = (void**)&ptr; \ 646 allocations[num_allocations].al_size = size; \ 647 valloc_sz += size; \ 648 ++num_allocations; \ 649 } 650 651 /* 652 * Allocate all the initial memory needed by the page allocator. 653 */ 654 static void 655 perform_allocations(void) 656 { 657 caddr_t mem; 658 int i; 659 int valloc_align; 660 661 PRM_DEBUG(valloc_base); 662 PRM_DEBUG(valloc_sz); 663 valloc_align = mmu.level_size[mmu.max_page_level > 0]; 664 mem = BOP_ALLOC(bootops, (caddr_t)valloc_base, valloc_sz, valloc_align); 665 if (mem != (caddr_t)valloc_base) 666 panic("BOP_ALLOC() failed"); 667 bzero(mem, valloc_sz); 668 for (i = 0; i < num_allocations; ++i) { 669 *allocations[i].al_ptr = (void *)mem; 670 mem += allocations[i].al_size; 671 } 672 } 673 674 /* 675 * Our world looks like this at startup time. 676 * 677 * In a 32-bit OS, boot loads the kernel text at 0xfe800000 and kernel data 678 * at 0xfec00000. On a 64-bit OS, kernel text and data are loaded at 679 * 0xffffffff.fe800000 and 0xffffffff.fec00000 respectively. Those 680 * addresses are fixed in the binary at link time. 681 * 682 * On the text page: 683 * unix/genunix/krtld/module text loads. 684 * 685 * On the data page: 686 * unix/genunix/krtld/module data loads. 687 * 688 * Machine-dependent startup code 689 */ 690 void 691 startup(void) 692 { 693 #if !defined(__xpv) 694 extern void startup_pci_bios(void); 695 #endif 696 extern cpuset_t cpu_ready_set; 697 698 /* 699 * Make sure that nobody tries to use sekpm until we have 700 * initialized it properly. 701 */ 702 #if defined(__amd64) 703 kpm_desired = 1; 704 #endif 705 kpm_enable = 0; 706 CPUSET_ONLY(cpu_ready_set, 0); /* cpu 0 is boot cpu */ 707 708 #if defined(__xpv) /* XXPV fix me! */ 709 { 710 extern int segvn_use_regions; 711 segvn_use_regions = 0; 712 } 713 #endif 714 progressbar_init(); 715 startup_init(); 716 #if defined(__xpv) 717 startup_xen_version(); 718 #endif 719 startup_memlist(); 720 startup_kmem(); 721 startup_vm(); 722 #if !defined(__xpv) 723 /* 724 * Note we need to do this even on fast reboot in order to access 725 * the irq routing table (used for pci labels). 726 */ 727 startup_pci_bios(); 728 #endif 729 #if defined(__xpv) 730 startup_xen_mca(); 731 #endif 732 startup_modules(); 733 734 startup_end(); 735 } 736 737 static void 738 startup_init() 739 { 740 PRM_POINT("startup_init() starting..."); 741 742 /* 743 * Complete the extraction of cpuid data 744 */ 745 cpuid_pass2(CPU); 746 747 (void) check_boot_version(BOP_GETVERSION(bootops)); 748 749 /* 750 * Check for prom_debug in boot environment 751 */ 752 if (BOP_GETPROPLEN(bootops, "prom_debug") >= 0) { 753 ++prom_debug; 754 PRM_POINT("prom_debug found in boot enviroment"); 755 } 756 757 /* 758 * Collect node, cpu and memory configuration information. 759 */ 760 get_system_configuration(); 761 762 /* 763 * Halt if this is an unsupported processor. 764 */ 765 if (x86_type == X86_TYPE_486 || x86_type == X86_TYPE_CYRIX_486) { 766 printf("\n486 processor (\"%s\") detected.\n", 767 CPU->cpu_brandstr); 768 halt("This processor is not supported by this release " 769 "of Solaris."); 770 } 771 772 PRM_POINT("startup_init() done"); 773 } 774 775 /* 776 * Callback for copy_memlist_filter() to filter nucleus, kadb/kmdb, (ie. 777 * everything mapped above KERNEL_TEXT) pages from phys_avail. Note it 778 * also filters out physical page zero. There is some reliance on the 779 * boot loader allocating only a few contiguous physical memory chunks. 780 */ 781 static void 782 avail_filter(uint64_t *addr, uint64_t *size) 783 { 784 uintptr_t va; 785 uintptr_t next_va; 786 pfn_t pfn; 787 uint64_t pfn_addr; 788 uint64_t pfn_eaddr; 789 uint_t prot; 790 size_t len; 791 uint_t change; 792 793 if (prom_debug) 794 prom_printf("\tFilter: in: a=%" PRIx64 ", s=%" PRIx64 "\n", 795 *addr, *size); 796 797 /* 798 * page zero is required for BIOS.. never make it available 799 */ 800 if (*addr == 0) { 801 *addr += MMU_PAGESIZE; 802 *size -= MMU_PAGESIZE; 803 } 804 805 /* 806 * First we trim from the front of the range. Since kbm_probe() 807 * walks ranges in virtual order, but addr/size are physical, we need 808 * to the list until no changes are seen. This deals with the case 809 * where page "p" is mapped at v, page "p + PAGESIZE" is mapped at w 810 * but w < v. 811 */ 812 do { 813 change = 0; 814 for (va = KERNEL_TEXT; 815 *size > 0 && kbm_probe(&va, &len, &pfn, &prot) != 0; 816 va = next_va) { 817 818 next_va = va + len; 819 pfn_addr = pfn_to_pa(pfn); 820 pfn_eaddr = pfn_addr + len; 821 822 if (pfn_addr <= *addr && pfn_eaddr > *addr) { 823 change = 1; 824 while (*size > 0 && len > 0) { 825 *addr += MMU_PAGESIZE; 826 *size -= MMU_PAGESIZE; 827 len -= MMU_PAGESIZE; 828 } 829 } 830 } 831 if (change && prom_debug) 832 prom_printf("\t\ttrim: a=%" PRIx64 ", s=%" PRIx64 "\n", 833 *addr, *size); 834 } while (change); 835 836 /* 837 * Trim pages from the end of the range. 838 */ 839 for (va = KERNEL_TEXT; 840 *size > 0 && kbm_probe(&va, &len, &pfn, &prot) != 0; 841 va = next_va) { 842 843 next_va = va + len; 844 pfn_addr = pfn_to_pa(pfn); 845 846 if (pfn_addr >= *addr && pfn_addr < *addr + *size) 847 *size = pfn_addr - *addr; 848 } 849 850 if (prom_debug) 851 prom_printf("\tFilter out: a=%" PRIx64 ", s=%" PRIx64 "\n", 852 *addr, *size); 853 } 854 855 static void 856 kpm_init() 857 { 858 struct segkpm_crargs b; 859 860 /* 861 * These variables were all designed for sfmmu in which segkpm is 862 * mapped using a single pagesize - either 8KB or 4MB. On x86, we 863 * might use 2+ page sizes on a single machine, so none of these 864 * variables have a single correct value. They are set up as if we 865 * always use a 4KB pagesize, which should do no harm. In the long 866 * run, we should get rid of KPM's assumption that only a single 867 * pagesize is used. 868 */ 869 kpm_pgshft = MMU_PAGESHIFT; 870 kpm_pgsz = MMU_PAGESIZE; 871 kpm_pgoff = MMU_PAGEOFFSET; 872 kpmp2pshft = 0; 873 kpmpnpgs = 1; 874 ASSERT(((uintptr_t)kpm_vbase & (kpm_pgsz - 1)) == 0); 875 876 PRM_POINT("about to create segkpm"); 877 rw_enter(&kas.a_lock, RW_WRITER); 878 879 if (seg_attach(&kas, kpm_vbase, kpm_size, segkpm) < 0) 880 panic("cannot attach segkpm"); 881 882 b.prot = PROT_READ | PROT_WRITE; 883 b.nvcolors = 1; 884 885 if (segkpm_create(segkpm, (caddr_t)&b) != 0) 886 panic("segkpm_create segkpm"); 887 888 rw_exit(&kas.a_lock); 889 } 890 891 /* 892 * The debug info page provides enough information to allow external 893 * inspectors (e.g. when running under a hypervisor) to bootstrap 894 * themselves into allowing full-blown kernel debugging. 895 */ 896 static void 897 init_debug_info(void) 898 { 899 caddr_t mem; 900 debug_info_t *di; 901 902 #ifndef __lint 903 ASSERT(sizeof (debug_info_t) < MMU_PAGESIZE); 904 #endif 905 906 mem = BOP_ALLOC(bootops, (caddr_t)DEBUG_INFO_VA, MMU_PAGESIZE, 907 MMU_PAGESIZE); 908 909 if (mem != (caddr_t)DEBUG_INFO_VA) 910 panic("BOP_ALLOC() failed"); 911 bzero(mem, MMU_PAGESIZE); 912 913 di = (debug_info_t *)mem; 914 915 di->di_magic = DEBUG_INFO_MAGIC; 916 di->di_version = DEBUG_INFO_VERSION; 917 di->di_modules = (uintptr_t)&modules; 918 di->di_s_text = (uintptr_t)s_text; 919 di->di_e_text = (uintptr_t)e_text; 920 di->di_s_data = (uintptr_t)s_data; 921 di->di_e_data = (uintptr_t)e_data; 922 di->di_hat_htable_off = offsetof(hat_t, hat_htable); 923 di->di_ht_pfn_off = offsetof(htable_t, ht_pfn); 924 } 925 926 /* 927 * Build the memlists and other kernel essential memory system data structures. 928 * This is everything at valloc_base. 929 */ 930 static void 931 startup_memlist(void) 932 { 933 size_t memlist_sz; 934 size_t memseg_sz; 935 size_t pagehash_sz; 936 size_t pp_sz; 937 uintptr_t va; 938 size_t len; 939 uint_t prot; 940 pfn_t pfn; 941 int memblocks; 942 pfn_t rsvd_high_pfn; 943 pgcnt_t rsvd_pgcnt; 944 size_t rsvdmemlist_sz; 945 int rsvdmemblocks; 946 caddr_t pagecolor_mem; 947 size_t pagecolor_memsz; 948 caddr_t page_ctrs_mem; 949 size_t page_ctrs_size; 950 size_t pse_table_alloc_size; 951 struct memlist *current; 952 extern void startup_build_mem_nodes(struct memlist *); 953 954 /* XX64 fix these - they should be in include files */ 955 extern size_t page_coloring_init(uint_t, int, int); 956 extern void page_coloring_setup(caddr_t); 957 958 PRM_POINT("startup_memlist() starting..."); 959 960 /* 961 * Use leftover large page nucleus text/data space for loadable modules. 962 * Use at most MODTEXT/MODDATA. 963 */ 964 len = kbm_nucleus_size; 965 ASSERT(len > MMU_PAGESIZE); 966 967 moddata = (caddr_t)ROUND_UP_PAGE(e_data); 968 e_moddata = (caddr_t)P2ROUNDUP((uintptr_t)e_data, (uintptr_t)len); 969 if (e_moddata - moddata > MODDATA) 970 e_moddata = moddata + MODDATA; 971 972 modtext = (caddr_t)ROUND_UP_PAGE(e_text); 973 e_modtext = (caddr_t)P2ROUNDUP((uintptr_t)e_text, (uintptr_t)len); 974 if (e_modtext - modtext > MODTEXT) 975 e_modtext = modtext + MODTEXT; 976 977 econtig = e_moddata; 978 979 PRM_DEBUG(modtext); 980 PRM_DEBUG(e_modtext); 981 PRM_DEBUG(moddata); 982 PRM_DEBUG(e_moddata); 983 PRM_DEBUG(econtig); 984 985 /* 986 * Examine the boot loader physical memory map to find out: 987 * - total memory in system - physinstalled 988 * - the max physical address - physmax 989 * - the number of discontiguous segments of memory. 990 */ 991 if (prom_debug) 992 print_memlist("boot physinstalled", 993 bootops->boot_mem->physinstalled); 994 installed_top_size_ex(bootops->boot_mem->physinstalled, &physmax, 995 &physinstalled, &memblocks); 996 PRM_DEBUG(physmax); 997 PRM_DEBUG(physinstalled); 998 PRM_DEBUG(memblocks); 999 1000 /* 1001 * Compute maximum physical address for memory DR operations. 1002 * Memory DR operations are unsupported on xpv or 32bit OSes. 1003 */ 1004 #ifdef __amd64 1005 if (plat_dr_support_memory()) { 1006 if (plat_dr_physmax == 0) { 1007 uint_t pabits = UINT_MAX; 1008 1009 cpuid_get_addrsize(CPU, &pabits, NULL); 1010 plat_dr_physmax = btop(1ULL << pabits); 1011 } 1012 if (plat_dr_physmax > PHYSMEM_MAX64) 1013 plat_dr_physmax = PHYSMEM_MAX64; 1014 } else 1015 #endif 1016 plat_dr_physmax = 0; 1017 1018 /* 1019 * Examine the bios reserved memory to find out: 1020 * - the number of discontiguous segments of memory. 1021 */ 1022 if (prom_debug) 1023 print_memlist("boot reserved mem", 1024 bootops->boot_mem->rsvdmem); 1025 installed_top_size_ex(bootops->boot_mem->rsvdmem, &rsvd_high_pfn, 1026 &rsvd_pgcnt, &rsvdmemblocks); 1027 PRM_DEBUG(rsvd_high_pfn); 1028 PRM_DEBUG(rsvd_pgcnt); 1029 PRM_DEBUG(rsvdmemblocks); 1030 1031 /* 1032 * Initialize hat's mmu parameters. 1033 * Check for enforce-prot-exec in boot environment. It's used to 1034 * enable/disable support for the page table entry NX bit. 1035 * The default is to enforce PROT_EXEC on processors that support NX. 1036 * Boot seems to round up the "len", but 8 seems to be big enough. 1037 */ 1038 mmu_init(); 1039 1040 #ifdef __i386 1041 /* 1042 * physmax is lowered if there is more memory than can be 1043 * physically addressed in 32 bit (PAE/non-PAE) modes. 1044 */ 1045 if (mmu.pae_hat) { 1046 if (PFN_ABOVE64G(physmax)) { 1047 physinstalled -= (physmax - (PFN_64G - 1)); 1048 physmax = PFN_64G - 1; 1049 } 1050 } else { 1051 if (PFN_ABOVE4G(physmax)) { 1052 physinstalled -= (physmax - (PFN_4G - 1)); 1053 physmax = PFN_4G - 1; 1054 } 1055 } 1056 #endif 1057 1058 startup_build_mem_nodes(bootops->boot_mem->physinstalled); 1059 1060 if (BOP_GETPROPLEN(bootops, "enforce-prot-exec") >= 0) { 1061 int len = BOP_GETPROPLEN(bootops, "enforce-prot-exec"); 1062 char value[8]; 1063 1064 if (len < 8) 1065 (void) BOP_GETPROP(bootops, "enforce-prot-exec", value); 1066 else 1067 (void) strcpy(value, ""); 1068 if (strcmp(value, "off") == 0) 1069 mmu.pt_nx = 0; 1070 } 1071 PRM_DEBUG(mmu.pt_nx); 1072 1073 /* 1074 * We will need page_t's for every page in the system, except for 1075 * memory mapped at or above above the start of the kernel text segment. 1076 * 1077 * pages above e_modtext are attributed to kernel debugger (obp_pages) 1078 */ 1079 npages = physinstalled - 1; /* avail_filter() skips page 0, so "- 1" */ 1080 obp_pages = 0; 1081 va = KERNEL_TEXT; 1082 while (kbm_probe(&va, &len, &pfn, &prot) != 0) { 1083 npages -= len >> MMU_PAGESHIFT; 1084 if (va >= (uintptr_t)e_moddata) 1085 obp_pages += len >> MMU_PAGESHIFT; 1086 va += len; 1087 } 1088 PRM_DEBUG(npages); 1089 PRM_DEBUG(obp_pages); 1090 1091 /* 1092 * If physmem is patched to be non-zero, use it instead of the computed 1093 * value unless it is larger than the actual amount of memory on hand. 1094 */ 1095 if (physmem == 0 || physmem > npages) { 1096 physmem = npages; 1097 } else if (physmem < npages) { 1098 orig_npages = npages; 1099 npages = physmem; 1100 } 1101 PRM_DEBUG(physmem); 1102 1103 /* 1104 * We now compute the sizes of all the initial allocations for 1105 * structures the kernel needs in order do kmem_alloc(). These 1106 * include: 1107 * memsegs 1108 * memlists 1109 * page hash table 1110 * page_t's 1111 * page coloring data structs 1112 */ 1113 memseg_sz = sizeof (struct memseg) * (memblocks + POSS_NEW_FRAGMENTS); 1114 ADD_TO_ALLOCATIONS(memseg_base, memseg_sz); 1115 PRM_DEBUG(memseg_sz); 1116 1117 /* 1118 * Reserve space for memlists. There's no real good way to know exactly 1119 * how much room we'll need, but this should be a good upper bound. 1120 */ 1121 memlist_sz = ROUND_UP_PAGE(2 * sizeof (struct memlist) * 1122 (memblocks + POSS_NEW_FRAGMENTS)); 1123 ADD_TO_ALLOCATIONS(memlist, memlist_sz); 1124 PRM_DEBUG(memlist_sz); 1125 1126 /* 1127 * Reserve space for bios reserved memlists. 1128 */ 1129 rsvdmemlist_sz = ROUND_UP_PAGE(2 * sizeof (struct memlist) * 1130 (rsvdmemblocks + POSS_NEW_FRAGMENTS)); 1131 ADD_TO_ALLOCATIONS(bios_rsvd, rsvdmemlist_sz); 1132 PRM_DEBUG(rsvdmemlist_sz); 1133 1134 /* LINTED */ 1135 ASSERT(P2SAMEHIGHBIT((1 << PP_SHIFT), sizeof (struct page))); 1136 /* 1137 * The page structure hash table size is a power of 2 1138 * such that the average hash chain length is PAGE_HASHAVELEN. 1139 */ 1140 page_hashsz = npages / PAGE_HASHAVELEN; 1141 page_hashsz_shift = highbit(page_hashsz); 1142 page_hashsz = 1 << page_hashsz_shift; 1143 pagehash_sz = sizeof (struct page *) * page_hashsz; 1144 ADD_TO_ALLOCATIONS(page_hash, pagehash_sz); 1145 PRM_DEBUG(pagehash_sz); 1146 1147 /* 1148 * Set aside room for the page structures themselves. 1149 */ 1150 PRM_DEBUG(npages); 1151 pp_sz = sizeof (struct page) * npages; 1152 ADD_TO_ALLOCATIONS(pp_base, pp_sz); 1153 PRM_DEBUG(pp_sz); 1154 1155 /* 1156 * determine l2 cache info and memory size for page coloring 1157 */ 1158 (void) getl2cacheinfo(CPU, 1159 &l2cache_sz, &l2cache_linesz, &l2cache_assoc); 1160 pagecolor_memsz = 1161 page_coloring_init(l2cache_sz, l2cache_linesz, l2cache_assoc); 1162 ADD_TO_ALLOCATIONS(pagecolor_mem, pagecolor_memsz); 1163 PRM_DEBUG(pagecolor_memsz); 1164 1165 page_ctrs_size = page_ctrs_sz(); 1166 ADD_TO_ALLOCATIONS(page_ctrs_mem, page_ctrs_size); 1167 PRM_DEBUG(page_ctrs_size); 1168 1169 /* 1170 * Allocate the array that protects pp->p_selock. 1171 */ 1172 pse_shift = size_pse_array(physmem, max_ncpus); 1173 pse_table_size = 1 << pse_shift; 1174 pse_table_alloc_size = pse_table_size * sizeof (pad_mutex_t); 1175 ADD_TO_ALLOCATIONS(pse_mutex, pse_table_alloc_size); 1176 1177 #if defined(__amd64) 1178 valloc_sz = ROUND_UP_LPAGE(valloc_sz); 1179 valloc_base = VALLOC_BASE; 1180 1181 /* 1182 * The default values of VALLOC_BASE and SEGKPM_BASE should work 1183 * for values of physmax up to 1 Terabyte. They need adjusting when 1184 * memory is at addresses above 1 TB. When adjusted, segkpm_base must 1185 * be aligned on KERNEL_REDZONE_SIZE boundary (span of top level pte). 1186 */ 1187 if (physmax + 1 > mmu_btop(TERABYTE) || 1188 plat_dr_physmax > mmu_btop(TERABYTE)) { 1189 uint64_t kpm_resv_amount = mmu_ptob(physmax + 1); 1190 1191 if (kpm_resv_amount < mmu_ptob(plat_dr_physmax)) { 1192 kpm_resv_amount = mmu_ptob(plat_dr_physmax); 1193 } 1194 1195 segkpm_base = -(P2ROUNDUP((2 * kpm_resv_amount), 1196 KERNEL_REDZONE_SIZE)); /* down from top VA */ 1197 1198 /* make sure we leave some space for user apps above hole */ 1199 segkpm_base = MAX(segkpm_base, AMD64_VA_HOLE_END + TERABYTE); 1200 if (segkpm_base > SEGKPM_BASE) 1201 segkpm_base = SEGKPM_BASE; 1202 PRM_DEBUG(segkpm_base); 1203 1204 valloc_base = segkpm_base + P2ROUNDUP(kpm_resv_amount, ONE_GIG); 1205 if (valloc_base < segkpm_base) 1206 panic("not enough kernel VA to support memory size"); 1207 PRM_DEBUG(valloc_base); 1208 } 1209 #else /* __i386 */ 1210 valloc_base = (uintptr_t)(MISC_VA_BASE - valloc_sz); 1211 valloc_base = P2ALIGN(valloc_base, mmu.level_size[1]); 1212 PRM_DEBUG(valloc_base); 1213 #endif /* __i386 */ 1214 1215 /* 1216 * do all the initial allocations 1217 */ 1218 perform_allocations(); 1219 1220 /* 1221 * Build phys_install and phys_avail in kernel memspace. 1222 * - phys_install should be all memory in the system. 1223 * - phys_avail is phys_install minus any memory mapped before this 1224 * point above KERNEL_TEXT. 1225 */ 1226 current = phys_install = memlist; 1227 copy_memlist_filter(bootops->boot_mem->physinstalled, ¤t, NULL); 1228 if ((caddr_t)current > (caddr_t)memlist + memlist_sz) 1229 panic("physinstalled was too big!"); 1230 if (prom_debug) 1231 print_memlist("phys_install", phys_install); 1232 1233 phys_avail = current; 1234 PRM_POINT("Building phys_avail:\n"); 1235 copy_memlist_filter(bootops->boot_mem->physinstalled, ¤t, 1236 avail_filter); 1237 if ((caddr_t)current > (caddr_t)memlist + memlist_sz) 1238 panic("physavail was too big!"); 1239 if (prom_debug) 1240 print_memlist("phys_avail", phys_avail); 1241 #ifndef __xpv 1242 /* 1243 * Free unused memlist items, which may be used by memory DR driver 1244 * at runtime. 1245 */ 1246 if ((caddr_t)current < (caddr_t)memlist + memlist_sz) { 1247 memlist_free_block((caddr_t)current, 1248 (caddr_t)memlist + memlist_sz - (caddr_t)current); 1249 } 1250 #endif 1251 1252 /* 1253 * Build bios reserved memspace 1254 */ 1255 current = bios_rsvd; 1256 copy_memlist_filter(bootops->boot_mem->rsvdmem, ¤t, NULL); 1257 if ((caddr_t)current > (caddr_t)bios_rsvd + rsvdmemlist_sz) 1258 panic("bios_rsvd was too big!"); 1259 if (prom_debug) 1260 print_memlist("bios_rsvd", bios_rsvd); 1261 #ifndef __xpv 1262 /* 1263 * Free unused memlist items, which may be used by memory DR driver 1264 * at runtime. 1265 */ 1266 if ((caddr_t)current < (caddr_t)bios_rsvd + rsvdmemlist_sz) { 1267 memlist_free_block((caddr_t)current, 1268 (caddr_t)bios_rsvd + rsvdmemlist_sz - (caddr_t)current); 1269 } 1270 #endif 1271 1272 /* 1273 * setup page coloring 1274 */ 1275 page_coloring_setup(pagecolor_mem); 1276 page_lock_init(); /* currently a no-op */ 1277 1278 /* 1279 * free page list counters 1280 */ 1281 (void) page_ctrs_alloc(page_ctrs_mem); 1282 1283 /* 1284 * Size the pcf array based on the number of cpus in the box at 1285 * boot time. 1286 */ 1287 1288 pcf_init(); 1289 1290 /* 1291 * Initialize the page structures from the memory lists. 1292 */ 1293 availrmem_initial = availrmem = freemem = 0; 1294 PRM_POINT("Calling kphysm_init()..."); 1295 npages = kphysm_init(pp_base, npages); 1296 PRM_POINT("kphysm_init() done"); 1297 PRM_DEBUG(npages); 1298 1299 init_debug_info(); 1300 1301 /* 1302 * Now that page_t's have been initialized, remove all the 1303 * initial allocation pages from the kernel free page lists. 1304 */ 1305 boot_mapin((caddr_t)valloc_base, valloc_sz); 1306 boot_mapin((caddr_t)MISC_VA_BASE, MISC_VA_SIZE); 1307 PRM_POINT("startup_memlist() done"); 1308 1309 PRM_DEBUG(valloc_sz); 1310 1311 #if defined(__amd64) 1312 if ((availrmem >> (30 - MMU_PAGESHIFT)) >= 1313 textrepl_min_gb && l2cache_sz <= 2 << 20) { 1314 extern size_t textrepl_size_thresh; 1315 textrepl_size_thresh = (16 << 20) - 1; 1316 } 1317 #endif 1318 } 1319 1320 /* 1321 * Layout the kernel's part of address space and initialize kmem allocator. 1322 */ 1323 static void 1324 startup_kmem(void) 1325 { 1326 extern void page_set_colorequiv_arr(void); 1327 1328 PRM_POINT("startup_kmem() starting..."); 1329 1330 #if defined(__amd64) 1331 if (eprom_kernelbase && eprom_kernelbase != KERNELBASE) 1332 cmn_err(CE_NOTE, "!kernelbase cannot be changed on 64-bit " 1333 "systems."); 1334 kernelbase = segkpm_base - KERNEL_REDZONE_SIZE; 1335 core_base = (uintptr_t)COREHEAP_BASE; 1336 core_size = (size_t)MISC_VA_BASE - COREHEAP_BASE; 1337 #else /* __i386 */ 1338 /* 1339 * We configure kernelbase based on: 1340 * 1341 * 1. user specified kernelbase via eeprom command. Value cannot exceed 1342 * KERNELBASE_MAX. we large page align eprom_kernelbase 1343 * 1344 * 2. Default to KERNELBASE and adjust to 2X less the size for page_t. 1345 * On large memory systems we must lower kernelbase to allow 1346 * enough room for page_t's for all of memory. 1347 * 1348 * The value set here, might be changed a little later. 1349 */ 1350 if (eprom_kernelbase) { 1351 kernelbase = eprom_kernelbase & mmu.level_mask[1]; 1352 if (kernelbase > KERNELBASE_MAX) 1353 kernelbase = KERNELBASE_MAX; 1354 } else { 1355 kernelbase = (uintptr_t)KERNELBASE; 1356 kernelbase -= ROUND_UP_4MEG(2 * valloc_sz); 1357 } 1358 ASSERT((kernelbase & mmu.level_offset[1]) == 0); 1359 core_base = valloc_base; 1360 core_size = 0; 1361 #endif /* __i386 */ 1362 1363 PRM_DEBUG(core_base); 1364 PRM_DEBUG(core_size); 1365 PRM_DEBUG(kernelbase); 1366 1367 #if defined(__i386) 1368 segkp_fromheap = 1; 1369 #endif /* __i386 */ 1370 1371 ekernelheap = (char *)core_base; 1372 PRM_DEBUG(ekernelheap); 1373 1374 /* 1375 * Now that we know the real value of kernelbase, 1376 * update variables that were initialized with a value of 1377 * KERNELBASE (in common/conf/param.c). 1378 * 1379 * XXX The problem with this sort of hackery is that the 1380 * compiler just may feel like putting the const declarations 1381 * (in param.c) into the .text section. Perhaps they should 1382 * just be declared as variables there? 1383 */ 1384 1385 *(uintptr_t *)&_kernelbase = kernelbase; 1386 *(uintptr_t *)&_userlimit = kernelbase; 1387 #if defined(__amd64) 1388 *(uintptr_t *)&_userlimit -= KERNELBASE - USERLIMIT; 1389 #else 1390 *(uintptr_t *)&_userlimit32 = _userlimit; 1391 #endif 1392 PRM_DEBUG(_kernelbase); 1393 PRM_DEBUG(_userlimit); 1394 PRM_DEBUG(_userlimit32); 1395 1396 layout_kernel_va(); 1397 1398 #if defined(__i386) 1399 /* 1400 * If segmap is too large we can push the bottom of the kernel heap 1401 * higher than the base. Or worse, it could exceed the top of the 1402 * VA space entirely, causing it to wrap around. 1403 */ 1404 if (kernelheap >= ekernelheap || (uintptr_t)kernelheap < kernelbase) 1405 panic("too little address space available for kernelheap," 1406 " use eeprom for lower kernelbase or smaller segmapsize"); 1407 #endif /* __i386 */ 1408 1409 /* 1410 * Initialize the kernel heap. Note 3rd argument must be > 1st. 1411 */ 1412 kernelheap_init(kernelheap, ekernelheap, 1413 kernelheap + MMU_PAGESIZE, 1414 (void *)core_base, (void *)(core_base + core_size)); 1415 1416 #if defined(__xpv) 1417 /* 1418 * Link pending events struct into cpu struct 1419 */ 1420 CPU->cpu_m.mcpu_evt_pend = &cpu0_evt_data; 1421 #endif 1422 /* 1423 * Initialize kernel memory allocator. 1424 */ 1425 kmem_init(); 1426 1427 /* 1428 * Factor in colorequiv to check additional 'equivalent' bins 1429 */ 1430 page_set_colorequiv_arr(); 1431 1432 /* 1433 * print this out early so that we know what's going on 1434 */ 1435 print_x86_featureset(x86_featureset); 1436 1437 /* 1438 * Initialize bp_mapin(). 1439 */ 1440 bp_init(MMU_PAGESIZE, HAT_STORECACHING_OK); 1441 1442 /* 1443 * orig_npages is non-zero if physmem has been configured for less 1444 * than the available memory. 1445 */ 1446 if (orig_npages) { 1447 cmn_err(CE_WARN, "!%slimiting physmem to 0x%lx of 0x%lx pages", 1448 (npages == PHYSMEM ? "Due to virtual address space " : ""), 1449 npages, orig_npages); 1450 } 1451 #if defined(__i386) 1452 if (eprom_kernelbase && (eprom_kernelbase != kernelbase)) 1453 cmn_err(CE_WARN, "kernelbase value, User specified 0x%lx, " 1454 "System using 0x%lx", 1455 (uintptr_t)eprom_kernelbase, (uintptr_t)kernelbase); 1456 #endif 1457 1458 #ifdef KERNELBASE_ABI_MIN 1459 if (kernelbase < (uintptr_t)KERNELBASE_ABI_MIN) { 1460 cmn_err(CE_NOTE, "!kernelbase set to 0x%lx, system is not " 1461 "i386 ABI compliant.", (uintptr_t)kernelbase); 1462 } 1463 #endif 1464 1465 #ifndef __xpv 1466 if (plat_dr_support_memory()) { 1467 mem_config_init(); 1468 } 1469 #else /* __xpv */ 1470 /* 1471 * Some of the xen start information has to be relocated up 1472 * into the kernel's permanent address space. 1473 */ 1474 PRM_POINT("calling xen_relocate_start_info()"); 1475 xen_relocate_start_info(); 1476 PRM_POINT("xen_relocate_start_info() done"); 1477 1478 /* 1479 * (Update the vcpu pointer in our cpu structure to point into 1480 * the relocated shared info.) 1481 */ 1482 CPU->cpu_m.mcpu_vcpu_info = 1483 &HYPERVISOR_shared_info->vcpu_info[CPU->cpu_id]; 1484 #endif /* __xpv */ 1485 1486 PRM_POINT("startup_kmem() done"); 1487 } 1488 1489 #ifndef __xpv 1490 /* 1491 * If we have detected that we are running in an HVM environment, we need 1492 * to prepend the PV driver directory to the module search path. 1493 */ 1494 #define HVM_MOD_DIR "/platform/i86hvm/kernel" 1495 static void 1496 update_default_path() 1497 { 1498 char *current, *newpath; 1499 int newlen; 1500 1501 /* 1502 * We are about to resync with krtld. krtld will reset its 1503 * internal module search path iff Solaris has set default_path. 1504 * We want to be sure we're prepending this new directory to the 1505 * right search path. 1506 */ 1507 current = (default_path == NULL) ? kobj_module_path : default_path; 1508 1509 newlen = strlen(HVM_MOD_DIR) + strlen(current) + 2; 1510 newpath = kmem_alloc(newlen, KM_SLEEP); 1511 (void) strcpy(newpath, HVM_MOD_DIR); 1512 (void) strcat(newpath, " "); 1513 (void) strcat(newpath, current); 1514 1515 default_path = newpath; 1516 } 1517 #endif 1518 1519 static void 1520 startup_modules(void) 1521 { 1522 int cnt; 1523 extern void prom_setup(void); 1524 int32_t v, h; 1525 char d[11]; 1526 char *cp; 1527 cmi_hdl_t hdl; 1528 1529 PRM_POINT("startup_modules() starting..."); 1530 1531 #ifndef __xpv 1532 /* 1533 * Initialize ten-micro second timer so that drivers will 1534 * not get short changed in their init phase. This was 1535 * not getting called until clkinit which, on fast cpu's 1536 * caused the drv_usecwait to be way too short. 1537 */ 1538 microfind(); 1539 1540 if ((get_hwenv() & HW_XEN_HVM) != 0) 1541 update_default_path(); 1542 #endif 1543 1544 /* 1545 * Read the GMT lag from /etc/rtc_config. 1546 */ 1547 sgmtl(process_rtc_config_file()); 1548 1549 /* 1550 * Calculate default settings of system parameters based upon 1551 * maxusers, yet allow to be overridden via the /etc/system file. 1552 */ 1553 param_calc(0); 1554 1555 mod_setup(); 1556 1557 /* 1558 * Initialize system parameters. 1559 */ 1560 param_init(); 1561 1562 /* 1563 * Initialize the default brands 1564 */ 1565 brand_init(); 1566 1567 /* 1568 * maxmem is the amount of physical memory we're playing with. 1569 */ 1570 maxmem = physmem; 1571 1572 /* 1573 * Initialize segment management stuff. 1574 */ 1575 seg_init(); 1576 1577 if (modload("fs", "specfs") == -1) 1578 halt("Can't load specfs"); 1579 1580 if (modload("fs", "devfs") == -1) 1581 halt("Can't load devfs"); 1582 1583 if (modload("fs", "dev") == -1) 1584 halt("Can't load dev"); 1585 1586 if (modload("fs", "procfs") == -1) 1587 halt("Can't load procfs"); 1588 1589 (void) modloadonly("sys", "lbl_edition"); 1590 1591 dispinit(); 1592 1593 /* Read cluster configuration data. */ 1594 clconf_init(); 1595 1596 #if defined(__xpv) 1597 (void) ec_init(); 1598 gnttab_init(); 1599 (void) xs_early_init(); 1600 #endif /* __xpv */ 1601 1602 /* 1603 * Create a kernel device tree. First, create rootnex and 1604 * then invoke bus specific code to probe devices. 1605 */ 1606 setup_ddi(); 1607 1608 #ifdef __xpv 1609 if (DOMAIN_IS_INITDOMAIN(xen_info)) 1610 #endif 1611 { 1612 /* 1613 * Load the System Management BIOS into the global ksmbios 1614 * handle, if an SMBIOS is present on this system. 1615 */ 1616 ksmbios = smbios_open(NULL, SMB_VERSION, ksmbios_flags, NULL); 1617 } 1618 1619 1620 /* 1621 * Originally clconf_init() apparently needed the hostid. But 1622 * this no longer appears to be true - it uses its own nodeid. 1623 * By placing the hostid logic here, we are able to make use of 1624 * the SMBIOS UUID. 1625 */ 1626 if ((h = set_soft_hostid()) == HW_INVALID_HOSTID) { 1627 cmn_err(CE_WARN, "Unable to set hostid"); 1628 } else { 1629 for (v = h, cnt = 0; cnt < 10; cnt++) { 1630 d[cnt] = (char)(v % 10); 1631 v /= 10; 1632 if (v == 0) 1633 break; 1634 } 1635 for (cp = hw_serial; cnt >= 0; cnt--) 1636 *cp++ = d[cnt] + '0'; 1637 *cp = 0; 1638 } 1639 1640 /* 1641 * Set up the CPU module subsystem for the boot cpu in the native 1642 * case, and all physical cpu resource in the xpv dom0 case. 1643 * Modifies the device tree, so this must be done after 1644 * setup_ddi(). 1645 */ 1646 #ifdef __xpv 1647 /* 1648 * If paravirtualized and on dom0 then we initialize all physical 1649 * cpu handles now; if paravirtualized on a domU then do not 1650 * initialize. 1651 */ 1652 if (DOMAIN_IS_INITDOMAIN(xen_info)) { 1653 xen_mc_lcpu_cookie_t cpi; 1654 1655 for (cpi = xen_physcpu_next(NULL); cpi != NULL; 1656 cpi = xen_physcpu_next(cpi)) { 1657 if ((hdl = cmi_init(CMI_HDL_SOLARIS_xVM_MCA, 1658 xen_physcpu_chipid(cpi), xen_physcpu_coreid(cpi), 1659 xen_physcpu_strandid(cpi))) != NULL && 1660 is_x86_feature(x86_featureset, X86FSET_MCA)) 1661 cmi_mca_init(hdl); 1662 } 1663 } 1664 #else 1665 /* 1666 * Initialize a handle for the boot cpu - others will initialize 1667 * as they startup. Do not do this if we know we are in an HVM domU. 1668 */ 1669 if ((get_hwenv() & HW_XEN_HVM) == 0 && 1670 (hdl = cmi_init(CMI_HDL_NATIVE, cmi_ntv_hwchipid(CPU), 1671 cmi_ntv_hwcoreid(CPU), cmi_ntv_hwstrandid(CPU))) != NULL && 1672 is_x86_feature(x86_featureset, X86FSET_MCA)) { 1673 cmi_mca_init(hdl); 1674 CPU->cpu_m.mcpu_cmi_hdl = hdl; 1675 } 1676 #endif /* __xpv */ 1677 1678 /* 1679 * Fake a prom tree such that /dev/openprom continues to work 1680 */ 1681 PRM_POINT("startup_modules: calling prom_setup..."); 1682 prom_setup(); 1683 PRM_POINT("startup_modules: done"); 1684 1685 /* 1686 * Load all platform specific modules 1687 */ 1688 PRM_POINT("startup_modules: calling psm_modload..."); 1689 psm_modload(); 1690 1691 PRM_POINT("startup_modules() done"); 1692 } 1693 1694 /* 1695 * claim a "setaside" boot page for use in the kernel 1696 */ 1697 page_t * 1698 boot_claim_page(pfn_t pfn) 1699 { 1700 page_t *pp; 1701 1702 pp = page_numtopp_nolock(pfn); 1703 ASSERT(pp != NULL); 1704 1705 if (PP_ISBOOTPAGES(pp)) { 1706 if (pp->p_next != NULL) 1707 pp->p_next->p_prev = pp->p_prev; 1708 if (pp->p_prev == NULL) 1709 bootpages = pp->p_next; 1710 else 1711 pp->p_prev->p_next = pp->p_next; 1712 } else { 1713 /* 1714 * htable_attach() expects a base pagesize page 1715 */ 1716 if (pp->p_szc != 0) 1717 page_boot_demote(pp); 1718 pp = page_numtopp(pfn, SE_EXCL); 1719 } 1720 return (pp); 1721 } 1722 1723 /* 1724 * Walk through the pagetables looking for pages mapped in by boot. If the 1725 * setaside flag is set the pages are expected to be returned to the 1726 * kernel later in boot, so we add them to the bootpages list. 1727 */ 1728 static void 1729 protect_boot_range(uintptr_t low, uintptr_t high, int setaside) 1730 { 1731 uintptr_t va = low; 1732 size_t len; 1733 uint_t prot; 1734 pfn_t pfn; 1735 page_t *pp; 1736 pgcnt_t boot_protect_cnt = 0; 1737 1738 while (kbm_probe(&va, &len, &pfn, &prot) != 0 && va < high) { 1739 if (va + len >= high) 1740 panic("0x%lx byte mapping at 0x%p exceeds boot's " 1741 "legal range.", len, (void *)va); 1742 1743 while (len > 0) { 1744 pp = page_numtopp_alloc(pfn); 1745 if (pp != NULL) { 1746 if (setaside == 0) 1747 panic("Unexpected mapping by boot. " 1748 "addr=%p pfn=%lx\n", 1749 (void *)va, pfn); 1750 1751 pp->p_next = bootpages; 1752 pp->p_prev = NULL; 1753 PP_SETBOOTPAGES(pp); 1754 if (bootpages != NULL) { 1755 bootpages->p_prev = pp; 1756 } 1757 bootpages = pp; 1758 ++boot_protect_cnt; 1759 } 1760 1761 ++pfn; 1762 len -= MMU_PAGESIZE; 1763 va += MMU_PAGESIZE; 1764 } 1765 } 1766 PRM_DEBUG(boot_protect_cnt); 1767 } 1768 1769 /* 1770 * 1771 */ 1772 static void 1773 layout_kernel_va(void) 1774 { 1775 PRM_POINT("layout_kernel_va() starting..."); 1776 /* 1777 * Establish the final size of the kernel's heap, size of segmap, 1778 * segkp, etc. 1779 */ 1780 1781 #if defined(__amd64) 1782 1783 kpm_vbase = (caddr_t)segkpm_base; 1784 if (physmax + 1 < plat_dr_physmax) { 1785 kpm_size = ROUND_UP_LPAGE(mmu_ptob(plat_dr_physmax)); 1786 } else { 1787 kpm_size = ROUND_UP_LPAGE(mmu_ptob(physmax + 1)); 1788 } 1789 if ((uintptr_t)kpm_vbase + kpm_size > (uintptr_t)valloc_base) 1790 panic("not enough room for kpm!"); 1791 PRM_DEBUG(kpm_size); 1792 PRM_DEBUG(kpm_vbase); 1793 1794 /* 1795 * By default we create a seg_kp in 64 bit kernels, it's a little 1796 * faster to access than embedding it in the heap. 1797 */ 1798 segkp_base = (caddr_t)valloc_base + valloc_sz; 1799 if (!segkp_fromheap) { 1800 size_t sz = mmu_ptob(segkpsize); 1801 1802 /* 1803 * determine size of segkp 1804 */ 1805 if (sz < SEGKPMINSIZE || sz > SEGKPMAXSIZE) { 1806 sz = SEGKPDEFSIZE; 1807 cmn_err(CE_WARN, "!Illegal value for segkpsize. " 1808 "segkpsize has been reset to %ld pages", 1809 mmu_btop(sz)); 1810 } 1811 sz = MIN(sz, MAX(SEGKPMINSIZE, mmu_ptob(physmem))); 1812 1813 segkpsize = mmu_btop(ROUND_UP_LPAGE(sz)); 1814 } 1815 PRM_DEBUG(segkp_base); 1816 PRM_DEBUG(segkpsize); 1817 1818 /* 1819 * segzio is used for ZFS cached data. It uses a distinct VA 1820 * segment (from kernel heap) so that we can easily tell not to 1821 * include it in kernel crash dumps on 64 bit kernels. The trick is 1822 * to give it lots of VA, but not constrain the kernel heap. 1823 * We scale the size of segzio linearly with physmem up to 1824 * SEGZIOMAXSIZE. Above that amount it scales at 50% of physmem. 1825 */ 1826 segzio_base = segkp_base + SEGKPMAXSIZE; 1827 if (segzio_fromheap) { 1828 segziosize = 0; 1829 } else { 1830 size_t physmem_size = mmu_ptob(physmem); 1831 size_t size = (segziosize == 0) ? 1832 physmem_size : mmu_ptob(segziosize); 1833 1834 if (size < SEGZIOMINSIZE) 1835 size = SEGZIOMINSIZE; 1836 if (size > SEGZIOMAXSIZE) { 1837 size = SEGZIOMAXSIZE; 1838 if (physmem_size > size) 1839 size += (physmem_size - size) / 2; 1840 } 1841 segziosize = mmu_btop(ROUND_UP_LPAGE(size)); 1842 } 1843 PRM_DEBUG(segziosize); 1844 PRM_DEBUG(segzio_base); 1845 1846 /* 1847 * Put the range of VA for device mappings next, kmdb knows to not 1848 * grep in this range of addresses. 1849 */ 1850 toxic_addr = 1851 ROUND_UP_LPAGE((uintptr_t)segzio_base + mmu_ptob(segziosize)); 1852 PRM_DEBUG(toxic_addr); 1853 segmap_start = ROUND_UP_LPAGE(toxic_addr + toxic_size); 1854 #else /* __i386 */ 1855 segmap_start = ROUND_UP_LPAGE(kernelbase); 1856 #endif /* __i386 */ 1857 PRM_DEBUG(segmap_start); 1858 1859 /* 1860 * Users can change segmapsize through eeprom. If the variable 1861 * is tuned through eeprom, there is no upper bound on the 1862 * size of segmap. 1863 */ 1864 segmapsize = MAX(ROUND_UP_LPAGE(segmapsize), SEGMAPDEFAULT); 1865 1866 #if defined(__i386) 1867 /* 1868 * 32-bit systems don't have segkpm or segkp, so segmap appears at 1869 * the bottom of the kernel's address range. Set aside space for a 1870 * small red zone just below the start of segmap. 1871 */ 1872 segmap_start += KERNEL_REDZONE_SIZE; 1873 segmapsize -= KERNEL_REDZONE_SIZE; 1874 #endif 1875 1876 PRM_DEBUG(segmap_start); 1877 PRM_DEBUG(segmapsize); 1878 kernelheap = (caddr_t)ROUND_UP_LPAGE(segmap_start + segmapsize); 1879 PRM_DEBUG(kernelheap); 1880 PRM_POINT("layout_kernel_va() done..."); 1881 } 1882 1883 /* 1884 * Finish initializing the VM system, now that we are no longer 1885 * relying on the boot time memory allocators. 1886 */ 1887 static void 1888 startup_vm(void) 1889 { 1890 struct segmap_crargs a; 1891 1892 extern int use_brk_lpg, use_stk_lpg; 1893 1894 PRM_POINT("startup_vm() starting..."); 1895 1896 /* 1897 * Initialize the hat layer. 1898 */ 1899 hat_init(); 1900 1901 /* 1902 * Do final allocations of HAT data structures that need to 1903 * be allocated before quiescing the boot loader. 1904 */ 1905 PRM_POINT("Calling hat_kern_alloc()..."); 1906 hat_kern_alloc((caddr_t)segmap_start, segmapsize, ekernelheap); 1907 PRM_POINT("hat_kern_alloc() done"); 1908 1909 #ifndef __xpv 1910 /* 1911 * Setup Page Attribute Table 1912 */ 1913 pat_sync(); 1914 #endif 1915 1916 /* 1917 * The next two loops are done in distinct steps in order 1918 * to be sure that any page that is doubly mapped (both above 1919 * KERNEL_TEXT and below kernelbase) is dealt with correctly. 1920 * Note this may never happen, but it might someday. 1921 */ 1922 bootpages = NULL; 1923 PRM_POINT("Protecting boot pages"); 1924 1925 /* 1926 * Protect any pages mapped above KERNEL_TEXT that somehow have 1927 * page_t's. This can only happen if something weird allocated 1928 * in this range (like kadb/kmdb). 1929 */ 1930 protect_boot_range(KERNEL_TEXT, (uintptr_t)-1, 0); 1931 1932 /* 1933 * Before we can take over memory allocation/mapping from the boot 1934 * loader we must remove from our free page lists any boot allocated 1935 * pages that stay mapped until release_bootstrap(). 1936 */ 1937 protect_boot_range(0, kernelbase, 1); 1938 1939 1940 /* 1941 * Switch to running on regular HAT (not boot_mmu) 1942 */ 1943 PRM_POINT("Calling hat_kern_setup()..."); 1944 hat_kern_setup(); 1945 1946 /* 1947 * It is no longer safe to call BOP_ALLOC(), so make sure we don't. 1948 */ 1949 bop_no_more_mem(); 1950 1951 PRM_POINT("hat_kern_setup() done"); 1952 1953 hat_cpu_online(CPU); 1954 1955 /* 1956 * Initialize VM system 1957 */ 1958 PRM_POINT("Calling kvm_init()..."); 1959 kvm_init(); 1960 PRM_POINT("kvm_init() done"); 1961 1962 /* 1963 * Tell kmdb that the VM system is now working 1964 */ 1965 if (boothowto & RB_DEBUG) 1966 kdi_dvec_vmready(); 1967 1968 #if defined(__xpv) 1969 /* 1970 * Populate the I/O pool on domain 0 1971 */ 1972 if (DOMAIN_IS_INITDOMAIN(xen_info)) { 1973 extern long populate_io_pool(void); 1974 long init_io_pool_cnt; 1975 1976 PRM_POINT("Populating reserve I/O page pool"); 1977 init_io_pool_cnt = populate_io_pool(); 1978 PRM_DEBUG(init_io_pool_cnt); 1979 } 1980 #endif 1981 /* 1982 * Mangle the brand string etc. 1983 */ 1984 cpuid_pass3(CPU); 1985 1986 #if defined(__amd64) 1987 1988 /* 1989 * Create the device arena for toxic (to dtrace/kmdb) mappings. 1990 */ 1991 device_arena = vmem_create("device", (void *)toxic_addr, 1992 toxic_size, MMU_PAGESIZE, NULL, NULL, NULL, 0, VM_SLEEP); 1993 1994 #else /* __i386 */ 1995 1996 /* 1997 * allocate the bit map that tracks toxic pages 1998 */ 1999 toxic_bit_map_len = btop((ulong_t)(valloc_base - kernelbase)); 2000 PRM_DEBUG(toxic_bit_map_len); 2001 toxic_bit_map = 2002 kmem_zalloc(BT_SIZEOFMAP(toxic_bit_map_len), KM_NOSLEEP); 2003 ASSERT(toxic_bit_map != NULL); 2004 PRM_DEBUG(toxic_bit_map); 2005 2006 #endif /* __i386 */ 2007 2008 2009 /* 2010 * Now that we've got more VA, as well as the ability to allocate from 2011 * it, tell the debugger. 2012 */ 2013 if (boothowto & RB_DEBUG) 2014 kdi_dvec_memavail(); 2015 2016 /* 2017 * The following code installs a special page fault handler (#pf) 2018 * to work around a pentium bug. 2019 */ 2020 #if !defined(__amd64) && !defined(__xpv) 2021 if (x86_type == X86_TYPE_P5) { 2022 desctbr_t idtr; 2023 gate_desc_t *newidt; 2024 2025 if ((newidt = kmem_zalloc(MMU_PAGESIZE, KM_NOSLEEP)) == NULL) 2026 panic("failed to install pentium_pftrap"); 2027 2028 bcopy(idt0, newidt, NIDT * sizeof (*idt0)); 2029 set_gatesegd(&newidt[T_PGFLT], &pentium_pftrap, 2030 KCS_SEL, SDT_SYSIGT, TRP_KPL, 0); 2031 2032 (void) as_setprot(&kas, (caddr_t)newidt, MMU_PAGESIZE, 2033 PROT_READ | PROT_EXEC); 2034 2035 CPU->cpu_idt = newidt; 2036 idtr.dtr_base = (uintptr_t)CPU->cpu_idt; 2037 idtr.dtr_limit = (NIDT * sizeof (*idt0)) - 1; 2038 wr_idtr(&idtr); 2039 } 2040 #endif /* !__amd64 */ 2041 2042 #if !defined(__xpv) 2043 /* 2044 * Map page pfn=0 for drivers, such as kd, that need to pick up 2045 * parameters left there by controllers/BIOS. 2046 */ 2047 PRM_POINT("setup up p0_va"); 2048 p0_va = i86devmap(0, 1, PROT_READ); 2049 PRM_DEBUG(p0_va); 2050 #endif 2051 2052 cmn_err(CE_CONT, "?mem = %luK (0x%lx)\n", 2053 physinstalled << (MMU_PAGESHIFT - 10), ptob(physinstalled)); 2054 2055 /* 2056 * disable automatic large pages for small memory systems or 2057 * when the disable flag is set. 2058 * 2059 * Do not yet consider page sizes larger than 2m/4m. 2060 */ 2061 if (!auto_lpg_disable && mmu.max_page_level > 0) { 2062 max_uheap_lpsize = LEVEL_SIZE(1); 2063 max_ustack_lpsize = LEVEL_SIZE(1); 2064 max_privmap_lpsize = LEVEL_SIZE(1); 2065 max_uidata_lpsize = LEVEL_SIZE(1); 2066 max_utext_lpsize = LEVEL_SIZE(1); 2067 max_shm_lpsize = LEVEL_SIZE(1); 2068 } 2069 if (physmem < privm_lpg_min_physmem || mmu.max_page_level == 0 || 2070 auto_lpg_disable) { 2071 use_brk_lpg = 0; 2072 use_stk_lpg = 0; 2073 } 2074 mcntl0_lpsize = LEVEL_SIZE(mmu.umax_page_level); 2075 2076 PRM_POINT("Calling hat_init_finish()..."); 2077 hat_init_finish(); 2078 PRM_POINT("hat_init_finish() done"); 2079 2080 /* 2081 * Initialize the segkp segment type. 2082 */ 2083 rw_enter(&kas.a_lock, RW_WRITER); 2084 PRM_POINT("Attaching segkp"); 2085 if (segkp_fromheap) { 2086 segkp->s_as = &kas; 2087 } else if (seg_attach(&kas, (caddr_t)segkp_base, mmu_ptob(segkpsize), 2088 segkp) < 0) { 2089 panic("startup: cannot attach segkp"); 2090 /*NOTREACHED*/ 2091 } 2092 PRM_POINT("Doing segkp_create()"); 2093 if (segkp_create(segkp) != 0) { 2094 panic("startup: segkp_create failed"); 2095 /*NOTREACHED*/ 2096 } 2097 PRM_DEBUG(segkp); 2098 rw_exit(&kas.a_lock); 2099 2100 /* 2101 * kpm segment 2102 */ 2103 segmap_kpm = 0; 2104 if (kpm_desired) { 2105 kpm_init(); 2106 kpm_enable = 1; 2107 } 2108 2109 /* 2110 * Now create segmap segment. 2111 */ 2112 rw_enter(&kas.a_lock, RW_WRITER); 2113 if (seg_attach(&kas, (caddr_t)segmap_start, segmapsize, segmap) < 0) { 2114 panic("cannot attach segmap"); 2115 /*NOTREACHED*/ 2116 } 2117 PRM_DEBUG(segmap); 2118 2119 a.prot = PROT_READ | PROT_WRITE; 2120 a.shmsize = 0; 2121 a.nfreelist = segmapfreelists; 2122 2123 if (segmap_create(segmap, (caddr_t)&a) != 0) 2124 panic("segmap_create segmap"); 2125 rw_exit(&kas.a_lock); 2126 2127 setup_vaddr_for_ppcopy(CPU); 2128 2129 segdev_init(); 2130 #if defined(__xpv) 2131 if (DOMAIN_IS_INITDOMAIN(xen_info)) 2132 #endif 2133 pmem_init(); 2134 2135 PRM_POINT("startup_vm() done"); 2136 } 2137 2138 /* 2139 * Load a tod module for the non-standard tod part found on this system. 2140 */ 2141 static void 2142 load_tod_module(char *todmod) 2143 { 2144 if (modload("tod", todmod) == -1) 2145 halt("Can't load TOD module"); 2146 } 2147 2148 static void 2149 startup_end(void) 2150 { 2151 int i; 2152 extern void setx86isalist(void); 2153 extern void cpu_event_init(void); 2154 2155 PRM_POINT("startup_end() starting..."); 2156 2157 /* 2158 * Perform tasks that get done after most of the VM 2159 * initialization has been done but before the clock 2160 * and other devices get started. 2161 */ 2162 kern_setup1(); 2163 2164 /* 2165 * Perform CPC initialization for this CPU. 2166 */ 2167 kcpc_hw_init(CPU); 2168 2169 /* 2170 * Initialize cpu event framework. 2171 */ 2172 cpu_event_init(); 2173 2174 #if defined(OPTERON_WORKAROUND_6323525) 2175 if (opteron_workaround_6323525) 2176 patch_workaround_6323525(); 2177 #endif 2178 /* 2179 * If needed, load TOD module now so that ddi_get_time(9F) etc. work 2180 * (For now, "needed" is defined as set tod_module_name in /etc/system) 2181 */ 2182 if (tod_module_name != NULL) { 2183 PRM_POINT("load_tod_module()"); 2184 load_tod_module(tod_module_name); 2185 } 2186 2187 #if defined(__xpv) 2188 /* 2189 * Forceload interposing TOD module for the hypervisor. 2190 */ 2191 PRM_POINT("load_tod_module()"); 2192 load_tod_module("xpvtod"); 2193 #endif 2194 2195 /* 2196 * Configure the system. 2197 */ 2198 PRM_POINT("Calling configure()..."); 2199 configure(); /* set up devices */ 2200 PRM_POINT("configure() done"); 2201 2202 /* 2203 * We can now setup for XSAVE because fpu_probe is done in configure(). 2204 */ 2205 if (fp_save_mech == FP_XSAVE) { 2206 xsave_setup_msr(CPU); 2207 } 2208 2209 /* 2210 * Set the isa_list string to the defined instruction sets we 2211 * support. 2212 */ 2213 setx86isalist(); 2214 cpu_intr_alloc(CPU, NINTR_THREADS); 2215 psm_install(); 2216 2217 /* 2218 * We're done with bootops. We don't unmap the bootstrap yet because 2219 * we're still using bootsvcs. 2220 */ 2221 PRM_POINT("NULLing out bootops"); 2222 *bootopsp = (struct bootops *)NULL; 2223 bootops = (struct bootops *)NULL; 2224 2225 #if defined(__xpv) 2226 ec_init_debug_irq(); 2227 xs_domu_init(); 2228 #endif 2229 2230 #if defined(__amd64) && !defined(__xpv) 2231 /* 2232 * Intel IOMMU has been setup/initialized in ddi_impl.c 2233 * Start it up now. 2234 */ 2235 immu_startup(); 2236 #endif 2237 2238 PRM_POINT("Enabling interrupts"); 2239 (*picinitf)(); 2240 sti(); 2241 #if defined(__xpv) 2242 ASSERT(CPU->cpu_m.mcpu_vcpu_info->evtchn_upcall_mask == 0); 2243 xen_late_startup(); 2244 #endif 2245 2246 (void) add_avsoftintr((void *)&softlevel1_hdl, 1, softlevel1, 2247 "softlevel1", NULL, NULL); /* XXX to be moved later */ 2248 2249 /* 2250 * Register software interrupt handlers for ddi_periodic_add(9F). 2251 * Software interrupts up to the level 10 are supported. 2252 */ 2253 for (i = DDI_IPL_1; i <= DDI_IPL_10; i++) { 2254 (void) add_avsoftintr((void *)&softlevel_hdl[i-1], i, 2255 (avfunc)ddi_periodic_softintr, "ddi_periodic", 2256 (caddr_t)(uintptr_t)i, NULL); 2257 } 2258 2259 #if !defined(__xpv) 2260 if (modload("drv", "amd_iommu") < 0) { 2261 PRM_POINT("No AMD IOMMU present\n"); 2262 } else if (ddi_hold_installed_driver(ddi_name_to_major( 2263 "amd_iommu")) == NULL) { 2264 prom_printf("ERROR: failed to attach AMD IOMMU\n"); 2265 } 2266 #endif 2267 post_startup_cpu_fixups(); 2268 2269 PRM_POINT("startup_end() done"); 2270 } 2271 2272 /* 2273 * Don't remove the following 2 variables. They are necessary 2274 * for reading the hostid from the legacy file (/kernel/misc/sysinit). 2275 */ 2276 char *_hs1107 = hw_serial; 2277 ulong_t _bdhs34; 2278 2279 void 2280 post_startup(void) 2281 { 2282 extern void cpupm_init(cpu_t *); 2283 extern void cpu_event_init_cpu(cpu_t *); 2284 2285 /* 2286 * Set the system wide, processor-specific flags to be passed 2287 * to userland via the aux vector for performance hints and 2288 * instruction set extensions. 2289 */ 2290 bind_hwcap(); 2291 2292 #ifdef __xpv 2293 if (DOMAIN_IS_INITDOMAIN(xen_info)) 2294 #endif 2295 { 2296 #if defined(__xpv) 2297 xpv_panic_init(); 2298 #else 2299 /* 2300 * Startup the memory scrubber. 2301 * XXPV This should be running somewhere .. 2302 */ 2303 if ((get_hwenv() & HW_VIRTUAL) == 0) 2304 memscrub_init(); 2305 #endif 2306 } 2307 2308 /* 2309 * Complete CPU module initialization 2310 */ 2311 cmi_post_startup(); 2312 2313 /* 2314 * Perform forceloading tasks for /etc/system. 2315 */ 2316 (void) mod_sysctl(SYS_FORCELOAD, NULL); 2317 2318 /* 2319 * ON4.0: Force /proc module in until clock interrupt handle fixed 2320 * ON4.0: This must be fixed or restated in /etc/systems. 2321 */ 2322 (void) modload("fs", "procfs"); 2323 2324 (void) i_ddi_attach_hw_nodes("pit_beep"); 2325 2326 #if defined(__i386) 2327 /* 2328 * Check for required functional Floating Point hardware, 2329 * unless FP hardware explicitly disabled. 2330 */ 2331 if (fpu_exists && (fpu_pentium_fdivbug || fp_kind == FP_NO)) 2332 halt("No working FP hardware found"); 2333 #endif 2334 2335 maxmem = freemem; 2336 2337 cpu_event_init_cpu(CPU); 2338 cpupm_init(CPU); 2339 (void) mach_cpu_create_device_node(CPU, NULL); 2340 2341 pg_init(); 2342 } 2343 2344 static int 2345 pp_in_range(page_t *pp, uint64_t low_addr, uint64_t high_addr) 2346 { 2347 return ((pp->p_pagenum >= btop(low_addr)) && 2348 (pp->p_pagenum < btopr(high_addr))); 2349 } 2350 2351 static int 2352 pp_in_module(page_t *pp, const rd_existing_t *modranges) 2353 { 2354 uint_t i; 2355 2356 for (i = 0; modranges[i].phys != 0; i++) { 2357 if (pp_in_range(pp, modranges[i].phys, 2358 modranges[i].phys + modranges[i].size)) 2359 return (1); 2360 } 2361 2362 return (0); 2363 } 2364 2365 void 2366 release_bootstrap(void) 2367 { 2368 int root_is_ramdisk; 2369 page_t *pp; 2370 extern void kobj_boot_unmountroot(void); 2371 extern dev_t rootdev; 2372 uint_t i; 2373 char propname[32]; 2374 rd_existing_t *modranges; 2375 #if !defined(__xpv) 2376 pfn_t pfn; 2377 #endif 2378 2379 /* 2380 * Save the bootfs module ranges so that we can reserve them below 2381 * for the real bootfs. 2382 */ 2383 modranges = kmem_alloc(sizeof (rd_existing_t) * MAX_BOOT_MODULES, 2384 KM_SLEEP); 2385 for (i = 0; ; i++) { 2386 uint64_t start, size; 2387 2388 modranges[i].phys = 0; 2389 2390 (void) snprintf(propname, sizeof (propname), 2391 "module-addr-%u", i); 2392 if (do_bsys_getproplen(NULL, propname) <= 0) 2393 break; 2394 (void) do_bsys_getprop(NULL, propname, &start); 2395 2396 (void) snprintf(propname, sizeof (propname), 2397 "module-size-%u", i); 2398 if (do_bsys_getproplen(NULL, propname) <= 0) 2399 break; 2400 (void) do_bsys_getprop(NULL, propname, &size); 2401 2402 modranges[i].phys = start; 2403 modranges[i].size = size; 2404 } 2405 2406 /* unmount boot ramdisk and release kmem usage */ 2407 kobj_boot_unmountroot(); 2408 2409 /* 2410 * We're finished using the boot loader so free its pages. 2411 */ 2412 PRM_POINT("Unmapping lower boot pages"); 2413 2414 clear_boot_mappings(0, _userlimit); 2415 2416 postbootkernelbase = kernelbase; 2417 2418 /* 2419 * If root isn't on ramdisk, destroy the hardcoded 2420 * ramdisk node now and release the memory. Else, 2421 * ramdisk memory is kept in rd_pages. 2422 */ 2423 root_is_ramdisk = (getmajor(rootdev) == ddi_name_to_major("ramdisk")); 2424 if (!root_is_ramdisk) { 2425 dev_info_t *dip = ddi_find_devinfo("ramdisk", -1, 0); 2426 ASSERT(dip && ddi_get_parent(dip) == ddi_root_node()); 2427 ndi_rele_devi(dip); /* held from ddi_find_devinfo */ 2428 (void) ddi_remove_child(dip, 0); 2429 } 2430 2431 PRM_POINT("Releasing boot pages"); 2432 while (bootpages) { 2433 extern uint64_t ramdisk_start, ramdisk_end; 2434 pp = bootpages; 2435 bootpages = pp->p_next; 2436 2437 2438 /* Keep pages for the lower 64K */ 2439 if (pp_in_range(pp, 0, 0x40000)) { 2440 pp->p_next = lower_pages; 2441 lower_pages = pp; 2442 lower_pages_count++; 2443 continue; 2444 } 2445 2446 if (root_is_ramdisk && pp_in_range(pp, ramdisk_start, 2447 ramdisk_end) || pp_in_module(pp, modranges)) { 2448 pp->p_next = rd_pages; 2449 rd_pages = pp; 2450 continue; 2451 } 2452 pp->p_next = (struct page *)0; 2453 pp->p_prev = (struct page *)0; 2454 PP_CLRBOOTPAGES(pp); 2455 page_free(pp, 1); 2456 } 2457 PRM_POINT("Boot pages released"); 2458 2459 kmem_free(modranges, sizeof (rd_existing_t) * 99); 2460 2461 #if !defined(__xpv) 2462 /* XXPV -- note this following bunch of code needs to be revisited in Xen 3.0 */ 2463 /* 2464 * Find 1 page below 1 MB so that other processors can boot up or 2465 * so that any processor can resume. 2466 * Make sure it has a kernel VA as well as a 1:1 mapping. 2467 * We should have just free'd one up. 2468 */ 2469 2470 /* 2471 * 0x10 pages is 64K. Leave the bottom 64K alone 2472 * for BIOS. 2473 */ 2474 for (pfn = 0x10; pfn < btop(1*1024*1024); pfn++) { 2475 if (page_numtopp_alloc(pfn) == NULL) 2476 continue; 2477 rm_platter_va = i86devmap(pfn, 1, 2478 PROT_READ | PROT_WRITE | PROT_EXEC); 2479 rm_platter_pa = ptob(pfn); 2480 break; 2481 } 2482 if (pfn == btop(1*1024*1024) && use_mp) 2483 panic("No page below 1M available for starting " 2484 "other processors or for resuming from system-suspend"); 2485 #endif /* !__xpv */ 2486 } 2487 2488 /* 2489 * Initialize the platform-specific parts of a page_t. 2490 */ 2491 void 2492 add_physmem_cb(page_t *pp, pfn_t pnum) 2493 { 2494 pp->p_pagenum = pnum; 2495 pp->p_mapping = NULL; 2496 pp->p_embed = 0; 2497 pp->p_share = 0; 2498 pp->p_mlentry = 0; 2499 } 2500 2501 /* 2502 * kphysm_init() initializes physical memory. 2503 */ 2504 static pgcnt_t 2505 kphysm_init( 2506 page_t *pp, 2507 pgcnt_t npages) 2508 { 2509 struct memlist *pmem; 2510 struct memseg *cur_memseg; 2511 pfn_t base_pfn; 2512 pfn_t end_pfn; 2513 pgcnt_t num; 2514 pgcnt_t pages_done = 0; 2515 uint64_t addr; 2516 uint64_t size; 2517 extern pfn_t ddiphysmin; 2518 extern int mnode_xwa; 2519 int ms = 0, me = 0; 2520 2521 ASSERT(page_hash != NULL && page_hashsz != 0); 2522 2523 cur_memseg = memseg_base; 2524 for (pmem = phys_avail; pmem && npages; pmem = pmem->ml_next) { 2525 /* 2526 * In a 32 bit kernel can't use higher memory if we're 2527 * not booting in PAE mode. This check takes care of that. 2528 */ 2529 addr = pmem->ml_address; 2530 size = pmem->ml_size; 2531 if (btop(addr) > physmax) 2532 continue; 2533 2534 /* 2535 * align addr and size - they may not be at page boundaries 2536 */ 2537 if ((addr & MMU_PAGEOFFSET) != 0) { 2538 addr += MMU_PAGEOFFSET; 2539 addr &= ~(uint64_t)MMU_PAGEOFFSET; 2540 size -= addr - pmem->ml_address; 2541 } 2542 2543 /* only process pages below or equal to physmax */ 2544 if ((btop(addr + size) - 1) > physmax) 2545 size = ptob(physmax - btop(addr) + 1); 2546 2547 num = btop(size); 2548 if (num == 0) 2549 continue; 2550 2551 if (num > npages) 2552 num = npages; 2553 2554 npages -= num; 2555 pages_done += num; 2556 base_pfn = btop(addr); 2557 2558 if (prom_debug) 2559 prom_printf("MEMSEG addr=0x%" PRIx64 2560 " pgs=0x%lx pfn 0x%lx-0x%lx\n", 2561 addr, num, base_pfn, base_pfn + num); 2562 2563 /* 2564 * Ignore pages below ddiphysmin to simplify ddi memory 2565 * allocation with non-zero addr_lo requests. 2566 */ 2567 if (base_pfn < ddiphysmin) { 2568 if (base_pfn + num <= ddiphysmin) 2569 continue; 2570 pp += (ddiphysmin - base_pfn); 2571 num -= (ddiphysmin - base_pfn); 2572 base_pfn = ddiphysmin; 2573 } 2574 2575 /* 2576 * mnode_xwa is greater than 1 when large pages regions can 2577 * cross memory node boundaries. To prevent the formation 2578 * of these large pages, configure the memsegs based on the 2579 * memory node ranges which had been made non-contiguous. 2580 */ 2581 if (mnode_xwa > 1) { 2582 2583 end_pfn = base_pfn + num - 1; 2584 ms = PFN_2_MEM_NODE(base_pfn); 2585 me = PFN_2_MEM_NODE(end_pfn); 2586 2587 if (ms != me) { 2588 /* 2589 * current range spans more than 1 memory node. 2590 * Set num to only the pfn range in the start 2591 * memory node. 2592 */ 2593 num = mem_node_config[ms].physmax - base_pfn 2594 + 1; 2595 ASSERT(end_pfn > mem_node_config[ms].physmax); 2596 } 2597 } 2598 2599 for (;;) { 2600 /* 2601 * Build the memsegs entry 2602 */ 2603 cur_memseg->pages = pp; 2604 cur_memseg->epages = pp + num; 2605 cur_memseg->pages_base = base_pfn; 2606 cur_memseg->pages_end = base_pfn + num; 2607 2608 /* 2609 * Insert into memseg list in decreasing pfn range 2610 * order. Low memory is typically more fragmented such 2611 * that this ordering keeps the larger ranges at the 2612 * front of the list for code that searches memseg. 2613 * This ASSERTS that the memsegs coming in from boot 2614 * are in increasing physical address order and not 2615 * contiguous. 2616 */ 2617 if (memsegs != NULL) { 2618 ASSERT(cur_memseg->pages_base >= 2619 memsegs->pages_end); 2620 cur_memseg->next = memsegs; 2621 } 2622 memsegs = cur_memseg; 2623 2624 /* 2625 * add_physmem() initializes the PSM part of the page 2626 * struct by calling the PSM back with add_physmem_cb(). 2627 * In addition it coalesces pages into larger pages as 2628 * it initializes them. 2629 */ 2630 add_physmem(pp, num, base_pfn); 2631 cur_memseg++; 2632 availrmem_initial += num; 2633 availrmem += num; 2634 2635 pp += num; 2636 if (ms >= me) 2637 break; 2638 2639 /* process next memory node range */ 2640 ms++; 2641 base_pfn = mem_node_config[ms].physbase; 2642 num = MIN(mem_node_config[ms].physmax, 2643 end_pfn) - base_pfn + 1; 2644 } 2645 } 2646 2647 PRM_DEBUG(availrmem_initial); 2648 PRM_DEBUG(availrmem); 2649 PRM_DEBUG(freemem); 2650 build_pfn_hash(); 2651 return (pages_done); 2652 } 2653 2654 /* 2655 * Kernel VM initialization. 2656 */ 2657 static void 2658 kvm_init(void) 2659 { 2660 ASSERT((((uintptr_t)s_text) & MMU_PAGEOFFSET) == 0); 2661 2662 /* 2663 * Put the kernel segments in kernel address space. 2664 */ 2665 rw_enter(&kas.a_lock, RW_WRITER); 2666 as_avlinit(&kas); 2667 2668 (void) seg_attach(&kas, s_text, e_moddata - s_text, &ktextseg); 2669 (void) segkmem_create(&ktextseg); 2670 2671 (void) seg_attach(&kas, (caddr_t)valloc_base, valloc_sz, &kvalloc); 2672 (void) segkmem_create(&kvalloc); 2673 2674 (void) seg_attach(&kas, kernelheap, 2675 ekernelheap - kernelheap, &kvseg); 2676 (void) segkmem_create(&kvseg); 2677 2678 if (core_size > 0) { 2679 PRM_POINT("attaching kvseg_core"); 2680 (void) seg_attach(&kas, (caddr_t)core_base, core_size, 2681 &kvseg_core); 2682 (void) segkmem_create(&kvseg_core); 2683 } 2684 2685 if (segziosize > 0) { 2686 PRM_POINT("attaching segzio"); 2687 (void) seg_attach(&kas, segzio_base, mmu_ptob(segziosize), 2688 &kzioseg); 2689 (void) segkmem_zio_create(&kzioseg); 2690 2691 /* create zio area covering new segment */ 2692 segkmem_zio_init(segzio_base, mmu_ptob(segziosize)); 2693 } 2694 2695 (void) seg_attach(&kas, kdi_segdebugbase, kdi_segdebugsize, &kdebugseg); 2696 (void) segkmem_create(&kdebugseg); 2697 2698 rw_exit(&kas.a_lock); 2699 2700 /* 2701 * Ensure that the red zone at kernelbase is never accessible. 2702 */ 2703 PRM_POINT("protecting redzone"); 2704 (void) as_setprot(&kas, (caddr_t)kernelbase, KERNEL_REDZONE_SIZE, 0); 2705 2706 /* 2707 * Make the text writable so that it can be hot patched by DTrace. 2708 */ 2709 (void) as_setprot(&kas, s_text, e_modtext - s_text, 2710 PROT_READ | PROT_WRITE | PROT_EXEC); 2711 2712 /* 2713 * Make data writable until end. 2714 */ 2715 (void) as_setprot(&kas, s_data, e_moddata - s_data, 2716 PROT_READ | PROT_WRITE | PROT_EXEC); 2717 } 2718 2719 #ifndef __xpv 2720 /* 2721 * Solaris adds an entry for Write Combining caching to the PAT 2722 */ 2723 static uint64_t pat_attr_reg = PAT_DEFAULT_ATTRIBUTE; 2724 2725 void 2726 pat_sync(void) 2727 { 2728 ulong_t cr0, cr0_orig, cr4; 2729 2730 if (!is_x86_feature(x86_featureset, X86FSET_PAT)) 2731 return; 2732 cr0_orig = cr0 = getcr0(); 2733 cr4 = getcr4(); 2734 2735 /* disable caching and flush all caches and TLBs */ 2736 cr0 |= CR0_CD; 2737 cr0 &= ~CR0_NW; 2738 setcr0(cr0); 2739 invalidate_cache(); 2740 if (cr4 & CR4_PGE) { 2741 setcr4(cr4 & ~(ulong_t)CR4_PGE); 2742 setcr4(cr4); 2743 } else { 2744 reload_cr3(); 2745 } 2746 2747 /* add our entry to the PAT */ 2748 wrmsr(REG_PAT, pat_attr_reg); 2749 2750 /* flush TLBs and cache again, then reenable cr0 caching */ 2751 if (cr4 & CR4_PGE) { 2752 setcr4(cr4 & ~(ulong_t)CR4_PGE); 2753 setcr4(cr4); 2754 } else { 2755 reload_cr3(); 2756 } 2757 invalidate_cache(); 2758 setcr0(cr0_orig); 2759 } 2760 2761 #endif /* !__xpv */ 2762 2763 #if defined(_SOFT_HOSTID) 2764 /* 2765 * On platforms that do not have a hardware serial number, attempt 2766 * to set one based on the contents of /etc/hostid. If this file does 2767 * not exist, assume that we are to generate a new hostid and set 2768 * it in the kernel, for subsequent saving by a userland process 2769 * once the system is up and the root filesystem is mounted r/w. 2770 * 2771 * In order to gracefully support upgrade on OpenSolaris, if 2772 * /etc/hostid does not exist, we will attempt to get a serial number 2773 * using the legacy method (/kernel/misc/sysinit). 2774 * 2775 * If that isn't present, we attempt to use an SMBIOS UUID, which is 2776 * a hardware serial number. Note that we don't automatically trust 2777 * all SMBIOS UUIDs (some older platforms are defective and ship duplicate 2778 * UUIDs in violation of the standard), we check against a blacklist. 2779 * 2780 * In an attempt to make the hostid less prone to abuse 2781 * (for license circumvention, etc), we store it in /etc/hostid 2782 * in rot47 format. 2783 */ 2784 extern volatile unsigned long tenmicrodata; 2785 static int atoi(char *); 2786 2787 /* 2788 * Set this to non-zero in /etc/system if you think your SMBIOS returns a 2789 * UUID that is not unique. (Also report it so that the smbios_uuid_blacklist 2790 * array can be updated.) 2791 */ 2792 int smbios_broken_uuid = 0; 2793 2794 /* 2795 * List of known bad UUIDs. This is just the lower 32-bit values, since 2796 * that's what we use for the host id. If your hostid falls here, you need 2797 * to contact your hardware OEM for a fix for your BIOS. 2798 */ 2799 static unsigned char 2800 smbios_uuid_blacklist[][16] = { 2801 2802 { /* Reported bad UUID (Google search) */ 2803 0x00, 0x02, 0x00, 0x03, 0x00, 0x04, 0x00, 0x05, 2804 0x00, 0x06, 0x00, 0x07, 0x00, 0x08, 0x00, 0x09, 2805 }, 2806 { /* Known bad DELL UUID */ 2807 0x4C, 0x4C, 0x45, 0x44, 0x00, 0x00, 0x20, 0x10, 2808 0x80, 0x20, 0x80, 0xC0, 0x4F, 0x20, 0x20, 0x20, 2809 }, 2810 { /* Uninitialized flash */ 2811 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 2812 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff 2813 }, 2814 { /* All zeros */ 2815 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 2816 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00 2817 }, 2818 }; 2819 2820 static int32_t 2821 uuid_to_hostid(const uint8_t *uuid) 2822 { 2823 /* 2824 * Although the UUIDs are 128-bits, they may not distribute entropy 2825 * evenly. We would like to use SHA or MD5, but those are located 2826 * in loadable modules and not available this early in boot. As we 2827 * don't need the values to be cryptographically strong, we just 2828 * generate 32-bit vaue by xor'ing the various sequences together, 2829 * which ensures that the entire UUID contributes to the hostid. 2830 */ 2831 uint32_t id = 0; 2832 2833 /* first check against the blacklist */ 2834 for (int i = 0; i < (sizeof (smbios_uuid_blacklist) / 16); i++) { 2835 if (bcmp(smbios_uuid_blacklist[0], uuid, 16) == 0) { 2836 cmn_err(CE_CONT, "?Broken SMBIOS UUID. " 2837 "Contact BIOS manufacturer for repair.\n"); 2838 return ((int32_t)HW_INVALID_HOSTID); 2839 } 2840 } 2841 2842 for (int i = 0; i < 16; i++) 2843 id ^= ((uuid[i]) << (8 * (i % sizeof (id)))); 2844 2845 /* Make sure return value is positive */ 2846 return (id & 0x7fffffff); 2847 } 2848 2849 static int32_t 2850 set_soft_hostid(void) 2851 { 2852 struct _buf *file; 2853 char tokbuf[MAXNAMELEN]; 2854 token_t token; 2855 int done = 0; 2856 u_longlong_t tmp; 2857 int i; 2858 int32_t hostid = (int32_t)HW_INVALID_HOSTID; 2859 unsigned char *c; 2860 hrtime_t tsc; 2861 smbios_system_t smsys; 2862 2863 /* 2864 * If /etc/hostid file not found, we'd like to get a pseudo 2865 * random number to use at the hostid. A nice way to do this 2866 * is to read the real time clock. To remain xen-compatible, 2867 * we can't poke the real hardware, so we use tsc_read() to 2868 * read the real time clock. However, there is an ominous 2869 * warning in tsc_read that says it can return zero, so we 2870 * deal with that possibility by falling back to using the 2871 * (hopefully random enough) value in tenmicrodata. 2872 */ 2873 2874 if ((file = kobj_open_file(hostid_file)) == (struct _buf *)-1) { 2875 /* 2876 * hostid file not found - try to load sysinit module 2877 * and see if it has a nonzero hostid value...use that 2878 * instead of generating a new hostid here if so. 2879 */ 2880 if ((i = modload("misc", "sysinit")) != -1) { 2881 if (strlen(hw_serial) > 0) 2882 hostid = (int32_t)atoi(hw_serial); 2883 (void) modunload(i); 2884 } 2885 2886 /* 2887 * We try to use the SMBIOS UUID. But not if it is blacklisted 2888 * in /etc/system. 2889 */ 2890 if ((hostid == HW_INVALID_HOSTID) && 2891 (smbios_broken_uuid == 0) && 2892 (ksmbios != NULL) && 2893 (smbios_info_system(ksmbios, &smsys) != SMB_ERR) && 2894 (smsys.smbs_uuidlen >= 16)) { 2895 hostid = uuid_to_hostid(smsys.smbs_uuid); 2896 } 2897 2898 /* 2899 * Generate a "random" hostid using the clock. These 2900 * hostids will change on each boot if the value is not 2901 * saved to a persistent /etc/hostid file. 2902 */ 2903 if (hostid == HW_INVALID_HOSTID) { 2904 tsc = tsc_read(); 2905 if (tsc == 0) /* tsc_read can return zero sometimes */ 2906 hostid = (int32_t)tenmicrodata & 0x0CFFFFF; 2907 else 2908 hostid = (int32_t)tsc & 0x0CFFFFF; 2909 } 2910 } else { 2911 /* hostid file found */ 2912 while (!done) { 2913 token = kobj_lex(file, tokbuf, sizeof (tokbuf)); 2914 2915 switch (token) { 2916 case POUND: 2917 /* 2918 * skip comments 2919 */ 2920 kobj_find_eol(file); 2921 break; 2922 case STRING: 2923 /* 2924 * un-rot47 - obviously this 2925 * nonsense is ascii-specific 2926 */ 2927 for (c = (unsigned char *)tokbuf; 2928 *c != '\0'; c++) { 2929 *c += 47; 2930 if (*c > '~') 2931 *c -= 94; 2932 else if (*c < '!') 2933 *c += 94; 2934 } 2935 /* 2936 * now we should have a real number 2937 */ 2938 2939 if (kobj_getvalue(tokbuf, &tmp) != 0) 2940 kobj_file_err(CE_WARN, file, 2941 "Bad value %s for hostid", 2942 tokbuf); 2943 else 2944 hostid = (int32_t)tmp; 2945 2946 break; 2947 case EOF: 2948 done = 1; 2949 /* FALLTHROUGH */ 2950 case NEWLINE: 2951 kobj_newline(file); 2952 break; 2953 default: 2954 break; 2955 2956 } 2957 } 2958 if (hostid == HW_INVALID_HOSTID) /* didn't find a hostid */ 2959 kobj_file_err(CE_WARN, file, 2960 "hostid missing or corrupt"); 2961 2962 kobj_close_file(file); 2963 } 2964 /* 2965 * hostid is now the value read from /etc/hostid, or the 2966 * new hostid we generated in this routine or HW_INVALID_HOSTID if not 2967 * set. 2968 */ 2969 return (hostid); 2970 } 2971 2972 static int 2973 atoi(char *p) 2974 { 2975 int i = 0; 2976 2977 while (*p != '\0') 2978 i = 10 * i + (*p++ - '0'); 2979 2980 return (i); 2981 } 2982 2983 #endif /* _SOFT_HOSTID */ 2984 2985 void 2986 get_system_configuration(void) 2987 { 2988 char prop[32]; 2989 u_longlong_t nodes_ll, cpus_pernode_ll, lvalue; 2990 2991 if (BOP_GETPROPLEN(bootops, "nodes") > sizeof (prop) || 2992 BOP_GETPROP(bootops, "nodes", prop) < 0 || 2993 kobj_getvalue(prop, &nodes_ll) == -1 || 2994 nodes_ll > MAXNODES || 2995 BOP_GETPROPLEN(bootops, "cpus_pernode") > sizeof (prop) || 2996 BOP_GETPROP(bootops, "cpus_pernode", prop) < 0 || 2997 kobj_getvalue(prop, &cpus_pernode_ll) == -1) { 2998 system_hardware.hd_nodes = 1; 2999 system_hardware.hd_cpus_per_node = 0; 3000 } else { 3001 system_hardware.hd_nodes = (int)nodes_ll; 3002 system_hardware.hd_cpus_per_node = (int)cpus_pernode_ll; 3003 } 3004 3005 if (BOP_GETPROPLEN(bootops, "kernelbase") > sizeof (prop) || 3006 BOP_GETPROP(bootops, "kernelbase", prop) < 0 || 3007 kobj_getvalue(prop, &lvalue) == -1) 3008 eprom_kernelbase = NULL; 3009 else 3010 eprom_kernelbase = (uintptr_t)lvalue; 3011 3012 if (BOP_GETPROPLEN(bootops, "segmapsize") > sizeof (prop) || 3013 BOP_GETPROP(bootops, "segmapsize", prop) < 0 || 3014 kobj_getvalue(prop, &lvalue) == -1) 3015 segmapsize = SEGMAPDEFAULT; 3016 else 3017 segmapsize = (uintptr_t)lvalue; 3018 3019 if (BOP_GETPROPLEN(bootops, "segmapfreelists") > sizeof (prop) || 3020 BOP_GETPROP(bootops, "segmapfreelists", prop) < 0 || 3021 kobj_getvalue(prop, &lvalue) == -1) 3022 segmapfreelists = 0; /* use segmap driver default */ 3023 else 3024 segmapfreelists = (int)lvalue; 3025 3026 /* physmem used to be here, but moved much earlier to fakebop.c */ 3027 } 3028 3029 /* 3030 * Add to a memory list. 3031 * start = start of new memory segment 3032 * len = length of new memory segment in bytes 3033 * new = pointer to a new struct memlist 3034 * memlistp = memory list to which to add segment. 3035 */ 3036 void 3037 memlist_add( 3038 uint64_t start, 3039 uint64_t len, 3040 struct memlist *new, 3041 struct memlist **memlistp) 3042 { 3043 struct memlist *cur; 3044 uint64_t end = start + len; 3045 3046 new->ml_address = start; 3047 new->ml_size = len; 3048 3049 cur = *memlistp; 3050 3051 while (cur) { 3052 if (cur->ml_address >= end) { 3053 new->ml_next = cur; 3054 *memlistp = new; 3055 new->ml_prev = cur->ml_prev; 3056 cur->ml_prev = new; 3057 return; 3058 } 3059 ASSERT(cur->ml_address + cur->ml_size <= start); 3060 if (cur->ml_next == NULL) { 3061 cur->ml_next = new; 3062 new->ml_prev = cur; 3063 new->ml_next = NULL; 3064 return; 3065 } 3066 memlistp = &cur->ml_next; 3067 cur = cur->ml_next; 3068 } 3069 } 3070 3071 void 3072 kobj_vmem_init(vmem_t **text_arena, vmem_t **data_arena) 3073 { 3074 size_t tsize = e_modtext - modtext; 3075 size_t dsize = e_moddata - moddata; 3076 3077 *text_arena = vmem_create("module_text", tsize ? modtext : NULL, tsize, 3078 1, segkmem_alloc, segkmem_free, heaptext_arena, 0, VM_SLEEP); 3079 *data_arena = vmem_create("module_data", dsize ? moddata : NULL, dsize, 3080 1, segkmem_alloc, segkmem_free, heap32_arena, 0, VM_SLEEP); 3081 } 3082 3083 caddr_t 3084 kobj_text_alloc(vmem_t *arena, size_t size) 3085 { 3086 return (vmem_alloc(arena, size, VM_SLEEP | VM_BESTFIT)); 3087 } 3088 3089 /*ARGSUSED*/ 3090 caddr_t 3091 kobj_texthole_alloc(caddr_t addr, size_t size) 3092 { 3093 panic("unexpected call to kobj_texthole_alloc()"); 3094 /*NOTREACHED*/ 3095 return (0); 3096 } 3097 3098 /*ARGSUSED*/ 3099 void 3100 kobj_texthole_free(caddr_t addr, size_t size) 3101 { 3102 panic("unexpected call to kobj_texthole_free()"); 3103 } 3104 3105 /* 3106 * This is called just after configure() in startup(). 3107 * 3108 * The ISALIST concept is a bit hopeless on Intel, because 3109 * there's no guarantee of an ever-more-capable processor 3110 * given that various parts of the instruction set may appear 3111 * and disappear between different implementations. 3112 * 3113 * While it would be possible to correct it and even enhance 3114 * it somewhat, the explicit hardware capability bitmask allows 3115 * more flexibility. 3116 * 3117 * So, we just leave this alone. 3118 */ 3119 void 3120 setx86isalist(void) 3121 { 3122 char *tp; 3123 size_t len; 3124 extern char *isa_list; 3125 3126 #define TBUFSIZE 1024 3127 3128 tp = kmem_alloc(TBUFSIZE, KM_SLEEP); 3129 *tp = '\0'; 3130 3131 #if defined(__amd64) 3132 (void) strcpy(tp, "amd64 "); 3133 #endif 3134 3135 switch (x86_vendor) { 3136 case X86_VENDOR_Intel: 3137 case X86_VENDOR_AMD: 3138 case X86_VENDOR_TM: 3139 if (is_x86_feature(x86_featureset, X86FSET_CMOV)) { 3140 /* 3141 * Pentium Pro or later 3142 */ 3143 (void) strcat(tp, "pentium_pro"); 3144 (void) strcat(tp, 3145 is_x86_feature(x86_featureset, X86FSET_MMX) ? 3146 "+mmx pentium_pro " : " "); 3147 } 3148 /*FALLTHROUGH*/ 3149 case X86_VENDOR_Cyrix: 3150 /* 3151 * The Cyrix 6x86 does not have any Pentium features 3152 * accessible while not at privilege level 0. 3153 */ 3154 if (is_x86_feature(x86_featureset, X86FSET_CPUID)) { 3155 (void) strcat(tp, "pentium"); 3156 (void) strcat(tp, 3157 is_x86_feature(x86_featureset, X86FSET_MMX) ? 3158 "+mmx pentium " : " "); 3159 } 3160 break; 3161 default: 3162 break; 3163 } 3164 (void) strcat(tp, "i486 i386 i86"); 3165 len = strlen(tp) + 1; /* account for NULL at end of string */ 3166 isa_list = strcpy(kmem_alloc(len, KM_SLEEP), tp); 3167 kmem_free(tp, TBUFSIZE); 3168 3169 #undef TBUFSIZE 3170 } 3171 3172 3173 #ifdef __amd64 3174 3175 void * 3176 device_arena_alloc(size_t size, int vm_flag) 3177 { 3178 return (vmem_alloc(device_arena, size, vm_flag)); 3179 } 3180 3181 void 3182 device_arena_free(void *vaddr, size_t size) 3183 { 3184 vmem_free(device_arena, vaddr, size); 3185 } 3186 3187 #else /* __i386 */ 3188 3189 void * 3190 device_arena_alloc(size_t size, int vm_flag) 3191 { 3192 caddr_t vaddr; 3193 uintptr_t v; 3194 size_t start; 3195 size_t end; 3196 3197 vaddr = vmem_alloc(heap_arena, size, vm_flag); 3198 if (vaddr == NULL) 3199 return (NULL); 3200 3201 v = (uintptr_t)vaddr; 3202 ASSERT(v >= kernelbase); 3203 ASSERT(v + size <= valloc_base); 3204 3205 start = btop(v - kernelbase); 3206 end = btop(v + size - 1 - kernelbase); 3207 ASSERT(start < toxic_bit_map_len); 3208 ASSERT(end < toxic_bit_map_len); 3209 3210 while (start <= end) { 3211 BT_ATOMIC_SET(toxic_bit_map, start); 3212 ++start; 3213 } 3214 return (vaddr); 3215 } 3216 3217 void 3218 device_arena_free(void *vaddr, size_t size) 3219 { 3220 uintptr_t v = (uintptr_t)vaddr; 3221 size_t start; 3222 size_t end; 3223 3224 ASSERT(v >= kernelbase); 3225 ASSERT(v + size <= valloc_base); 3226 3227 start = btop(v - kernelbase); 3228 end = btop(v + size - 1 - kernelbase); 3229 ASSERT(start < toxic_bit_map_len); 3230 ASSERT(end < toxic_bit_map_len); 3231 3232 while (start <= end) { 3233 ASSERT(BT_TEST(toxic_bit_map, start) != 0); 3234 BT_ATOMIC_CLEAR(toxic_bit_map, start); 3235 ++start; 3236 } 3237 vmem_free(heap_arena, vaddr, size); 3238 } 3239 3240 /* 3241 * returns 1st address in range that is in device arena, or NULL 3242 * if len is not NULL it returns the length of the toxic range 3243 */ 3244 void * 3245 device_arena_contains(void *vaddr, size_t size, size_t *len) 3246 { 3247 uintptr_t v = (uintptr_t)vaddr; 3248 uintptr_t eaddr = v + size; 3249 size_t start; 3250 size_t end; 3251 3252 /* 3253 * if called very early by kmdb, just return NULL 3254 */ 3255 if (toxic_bit_map == NULL) 3256 return (NULL); 3257 3258 /* 3259 * First check if we're completely outside the bitmap range. 3260 */ 3261 if (v >= valloc_base || eaddr < kernelbase) 3262 return (NULL); 3263 3264 /* 3265 * Trim ends of search to look at only what the bitmap covers. 3266 */ 3267 if (v < kernelbase) 3268 v = kernelbase; 3269 start = btop(v - kernelbase); 3270 end = btop(eaddr - kernelbase); 3271 if (end >= toxic_bit_map_len) 3272 end = toxic_bit_map_len; 3273 3274 if (bt_range(toxic_bit_map, &start, &end, end) == 0) 3275 return (NULL); 3276 3277 v = kernelbase + ptob(start); 3278 if (len != NULL) 3279 *len = ptob(end - start); 3280 return ((void *)v); 3281 } 3282 3283 #endif /* __i386 */ 3284