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 */ 25 /* 26 * Copyright (c) 2010, Intel Corporation. 27 * All rights reserved. 28 */ 29 30 #include <sys/types.h> 31 #include <sys/t_lock.h> 32 #include <sys/param.h> 33 #include <sys/sysmacros.h> 34 #include <sys/signal.h> 35 #include <sys/systm.h> 36 #include <sys/user.h> 37 #include <sys/mman.h> 38 #include <sys/vm.h> 39 #include <sys/conf.h> 40 #include <sys/avintr.h> 41 #include <sys/autoconf.h> 42 #include <sys/disp.h> 43 #include <sys/class.h> 44 #include <sys/bitmap.h> 45 46 #include <sys/privregs.h> 47 48 #include <sys/proc.h> 49 #include <sys/buf.h> 50 #include <sys/kmem.h> 51 #include <sys/mem.h> 52 #include <sys/kstat.h> 53 54 #include <sys/reboot.h> 55 56 #include <sys/cred.h> 57 #include <sys/vnode.h> 58 #include <sys/file.h> 59 60 #include <sys/procfs.h> 61 62 #include <sys/vfs.h> 63 #include <sys/cmn_err.h> 64 #include <sys/utsname.h> 65 #include <sys/debug.h> 66 #include <sys/kdi.h> 67 68 #include <sys/dumphdr.h> 69 #include <sys/bootconf.h> 70 #include <sys/memlist_plat.h> 71 #include <sys/varargs.h> 72 #include <sys/promif.h> 73 #include <sys/modctl.h> 74 75 #include <sys/sunddi.h> 76 #include <sys/sunndi.h> 77 #include <sys/ndi_impldefs.h> 78 #include <sys/ddidmareq.h> 79 #include <sys/psw.h> 80 #include <sys/regset.h> 81 #include <sys/clock.h> 82 #include <sys/pte.h> 83 #include <sys/tss.h> 84 #include <sys/stack.h> 85 #include <sys/trap.h> 86 #include <sys/fp.h> 87 #include <vm/kboot_mmu.h> 88 #include <vm/anon.h> 89 #include <vm/as.h> 90 #include <vm/page.h> 91 #include <vm/seg.h> 92 #include <vm/seg_dev.h> 93 #include <vm/seg_kmem.h> 94 #include <vm/seg_kpm.h> 95 #include <vm/seg_map.h> 96 #include <vm/seg_vn.h> 97 #include <vm/seg_kp.h> 98 #include <sys/memnode.h> 99 #include <vm/vm_dep.h> 100 #include <sys/thread.h> 101 #include <sys/sysconf.h> 102 #include <sys/vm_machparam.h> 103 #include <sys/archsystm.h> 104 #include <sys/machsystm.h> 105 #include <vm/hat.h> 106 #include <vm/hat_i86.h> 107 #include <sys/pmem.h> 108 #include <sys/smp_impldefs.h> 109 #include <sys/x86_archext.h> 110 #include <sys/cpuvar.h> 111 #include <sys/segments.h> 112 #include <sys/clconf.h> 113 #include <sys/kobj.h> 114 #include <sys/kobj_lex.h> 115 #include <sys/cpc_impl.h> 116 #include <sys/cpu_module.h> 117 #include <sys/smbios.h> 118 #include <sys/debug_info.h> 119 #include <sys/bootinfo.h> 120 #include <sys/ddi_timer.h> 121 #include <sys/systeminfo.h> 122 #include <sys/multiboot.h> 123 124 #ifdef __xpv 125 126 #include <sys/hypervisor.h> 127 #include <sys/xen_mmu.h> 128 #include <sys/evtchn_impl.h> 129 #include <sys/gnttab.h> 130 #include <sys/xpv_panic.h> 131 #include <xen/sys/xenbus_comms.h> 132 #include <xen/public/physdev.h> 133 134 extern void xen_late_startup(void); 135 136 struct xen_evt_data cpu0_evt_data; 137 138 #else /* __xpv */ 139 #include <sys/memlist_impl.h> 140 141 extern void mem_config_init(void); 142 #endif /* __xpv */ 143 144 extern void progressbar_init(void); 145 extern void brand_init(void); 146 extern void pcf_init(void); 147 extern void pg_init(void); 148 149 extern int size_pse_array(pgcnt_t, int); 150 151 #if defined(_SOFT_HOSTID) 152 153 #include <sys/rtc.h> 154 155 static int32_t set_soft_hostid(void); 156 static char hostid_file[] = "/etc/hostid"; 157 158 #endif 159 160 void *gfx_devinfo_list; 161 162 #if defined(__amd64) && !defined(__xpv) 163 extern void immu_startup(void); 164 #endif 165 166 /* 167 * XXX make declaration below "static" when drivers no longer use this 168 * interface. 169 */ 170 extern caddr_t p0_va; /* Virtual address for accessing physical page 0 */ 171 172 /* 173 * segkp 174 */ 175 extern int segkp_fromheap; 176 177 static void kvm_init(void); 178 static void startup_init(void); 179 static void startup_memlist(void); 180 static void startup_kmem(void); 181 static void startup_modules(void); 182 static void startup_vm(void); 183 static void startup_end(void); 184 static void layout_kernel_va(void); 185 186 /* 187 * Declare these as initialized data so we can patch them. 188 */ 189 #ifdef __i386 190 191 /* 192 * Due to virtual address space limitations running in 32 bit mode, restrict 193 * the amount of physical memory configured to a max of PHYSMEM pages (16g). 194 * 195 * If the physical max memory size of 64g were allowed to be configured, the 196 * size of user virtual address space will be less than 1g. A limited user 197 * address space greatly reduces the range of applications that can run. 198 * 199 * If more physical memory than PHYSMEM is required, users should preferably 200 * run in 64 bit mode which has far looser virtual address space limitations. 201 * 202 * If 64 bit mode is not available (as in IA32) and/or more physical memory 203 * than PHYSMEM is required in 32 bit mode, physmem can be set to the desired 204 * value or to 0 (to configure all available memory) via eeprom(1M). kernelbase 205 * should also be carefully tuned to balance out the need of the user 206 * application while minimizing the risk of kernel heap exhaustion due to 207 * kernelbase being set too high. 208 */ 209 #define PHYSMEM 0x400000 210 211 #else /* __amd64 */ 212 213 /* 214 * For now we can handle memory with physical addresses up to about 215 * 64 Terabytes. This keeps the kernel above the VA hole, leaving roughly 216 * half the VA space for seg_kpm. When systems get bigger than 64TB this 217 * code will need revisiting. There is an implicit assumption that there 218 * are no *huge* holes in the physical address space too. 219 */ 220 #define TERABYTE (1ul << 40) 221 #define PHYSMEM_MAX64 mmu_btop(64 * TERABYTE) 222 #define PHYSMEM PHYSMEM_MAX64 223 #define AMD64_VA_HOLE_END 0xFFFF800000000000ul 224 225 #endif /* __amd64 */ 226 227 pgcnt_t physmem = PHYSMEM; 228 pgcnt_t obp_pages; /* Memory used by PROM for its text and data */ 229 230 char *kobj_file_buf; 231 int kobj_file_bufsize; /* set in /etc/system */ 232 233 /* Global variables for MP support. Used in mp_startup */ 234 caddr_t rm_platter_va = 0; 235 uint32_t rm_platter_pa; 236 237 int auto_lpg_disable = 1; 238 239 /* 240 * Some CPUs have holes in the middle of the 64-bit virtual address range. 241 */ 242 uintptr_t hole_start, hole_end; 243 244 /* 245 * kpm mapping window 246 */ 247 caddr_t kpm_vbase; 248 size_t kpm_size; 249 static int kpm_desired; 250 #ifdef __amd64 251 static uintptr_t segkpm_base = (uintptr_t)SEGKPM_BASE; 252 #endif 253 254 /* 255 * Configuration parameters set at boot time. 256 */ 257 258 caddr_t econtig; /* end of first block of contiguous kernel */ 259 260 struct bootops *bootops = 0; /* passed in from boot */ 261 struct bootops **bootopsp; 262 struct boot_syscalls *sysp; /* passed in from boot */ 263 264 char bootblock_fstype[16]; 265 266 char kern_bootargs[OBP_MAXPATHLEN]; 267 char kern_bootfile[OBP_MAXPATHLEN]; 268 269 /* 270 * ZFS zio segment. This allows us to exclude large portions of ZFS data that 271 * gets cached in kmem caches on the heap. If this is set to zero, we allocate 272 * zio buffers from their own segment, otherwise they are allocated from the 273 * heap. The optimization of allocating zio buffers from their own segment is 274 * only valid on 64-bit kernels. 275 */ 276 #if defined(__amd64) 277 int segzio_fromheap = 0; 278 #else 279 int segzio_fromheap = 1; 280 #endif 281 282 /* 283 * new memory fragmentations are possible in startup() due to BOP_ALLOCs. this 284 * depends on number of BOP_ALLOC calls made and requested size, memory size 285 * combination and whether boot.bin memory needs to be freed. 286 */ 287 #define POSS_NEW_FRAGMENTS 12 288 289 /* 290 * VM data structures 291 */ 292 long page_hashsz; /* Size of page hash table (power of two) */ 293 unsigned int page_hashsz_shift; /* log2(page_hashsz) */ 294 struct page *pp_base; /* Base of initial system page struct array */ 295 struct page **page_hash; /* Page hash table */ 296 pad_mutex_t *pse_mutex; /* Locks protecting pp->p_selock */ 297 size_t pse_table_size; /* Number of mutexes in pse_mutex[] */ 298 int pse_shift; /* log2(pse_table_size) */ 299 struct seg ktextseg; /* Segment used for kernel executable image */ 300 struct seg kvalloc; /* Segment used for "valloc" mapping */ 301 struct seg kpseg; /* Segment used for pageable kernel virt mem */ 302 struct seg kmapseg; /* Segment used for generic kernel mappings */ 303 struct seg kdebugseg; /* Segment used for the kernel debugger */ 304 305 struct seg *segkmap = &kmapseg; /* Kernel generic mapping segment */ 306 static struct seg *segmap = &kmapseg; /* easier to use name for in here */ 307 308 struct seg *segkp = &kpseg; /* Pageable kernel virtual memory segment */ 309 310 #if defined(__amd64) 311 struct seg kvseg_core; /* Segment used for the core heap */ 312 struct seg kpmseg; /* Segment used for physical mapping */ 313 struct seg *segkpm = &kpmseg; /* 64bit kernel physical mapping segment */ 314 #else 315 struct seg *segkpm = NULL; /* Unused on IA32 */ 316 #endif 317 318 caddr_t segkp_base; /* Base address of segkp */ 319 caddr_t segzio_base; /* Base address of segzio */ 320 #if defined(__amd64) 321 pgcnt_t segkpsize = btop(SEGKPDEFSIZE); /* size of segkp segment in pages */ 322 #else 323 pgcnt_t segkpsize = 0; 324 #endif 325 pgcnt_t segziosize = 0; /* size of zio segment in pages */ 326 327 /* 328 * A static DR page_t VA map is reserved that can map the page structures 329 * for a domain's entire RA space. The pages that back this space are 330 * dynamically allocated and need not be physically contiguous. The DR 331 * map size is derived from KPM size. 332 * This mechanism isn't used by x86 yet, so just stubs here. 333 */ 334 int ppvm_enable = 0; /* Static virtual map for page structs */ 335 page_t *ppvm_base = NULL; /* Base of page struct map */ 336 pgcnt_t ppvm_size = 0; /* Size of page struct map */ 337 338 /* 339 * VA range available to the debugger 340 */ 341 const caddr_t kdi_segdebugbase = (const caddr_t)SEGDEBUGBASE; 342 const size_t kdi_segdebugsize = SEGDEBUGSIZE; 343 344 struct memseg *memseg_base; 345 struct vnode unused_pages_vp; 346 347 #define FOURGB 0x100000000LL 348 349 struct memlist *memlist; 350 351 caddr_t s_text; /* start of kernel text segment */ 352 caddr_t e_text; /* end of kernel text segment */ 353 caddr_t s_data; /* start of kernel data segment */ 354 caddr_t e_data; /* end of kernel data segment */ 355 caddr_t modtext; /* start of loadable module text reserved */ 356 caddr_t e_modtext; /* end of loadable module text reserved */ 357 caddr_t moddata; /* start of loadable module data reserved */ 358 caddr_t e_moddata; /* end of loadable module data reserved */ 359 360 struct memlist *phys_install; /* Total installed physical memory */ 361 struct memlist *phys_avail; /* Total available physical memory */ 362 struct memlist *bios_rsvd; /* Bios reserved memory */ 363 364 /* 365 * kphysm_init returns the number of pages that were processed 366 */ 367 static pgcnt_t kphysm_init(page_t *, pgcnt_t); 368 369 #define IO_PROP_SIZE 64 /* device property size */ 370 371 /* 372 * a couple useful roundup macros 373 */ 374 #define ROUND_UP_PAGE(x) \ 375 ((uintptr_t)P2ROUNDUP((uintptr_t)(x), (uintptr_t)MMU_PAGESIZE)) 376 #define ROUND_UP_LPAGE(x) \ 377 ((uintptr_t)P2ROUNDUP((uintptr_t)(x), mmu.level_size[1])) 378 #define ROUND_UP_4MEG(x) \ 379 ((uintptr_t)P2ROUNDUP((uintptr_t)(x), (uintptr_t)FOUR_MEG)) 380 #define ROUND_UP_TOPLEVEL(x) \ 381 ((uintptr_t)P2ROUNDUP((uintptr_t)(x), mmu.level_size[mmu.max_level])) 382 383 /* 384 * 32-bit Kernel's Virtual memory layout. 385 * +-----------------------+ 386 * | | 387 * 0xFFC00000 -|-----------------------|- ARGSBASE 388 * | debugger | 389 * 0xFF800000 -|-----------------------|- SEGDEBUGBASE 390 * | Kernel Data | 391 * 0xFEC00000 -|-----------------------| 392 * | Kernel Text | 393 * 0xFE800000 -|-----------------------|- KERNEL_TEXT (0xFB400000 on Xen) 394 * |--- GDT ---|- GDT page (GDT_VA) 395 * |--- debug info ---|- debug info (DEBUG_INFO_VA) 396 * | | 397 * | page_t structures | 398 * | memsegs, memlists, | 399 * | page hash, etc. | 400 * --- -|-----------------------|- ekernelheap, valloc_base (floating) 401 * | | (segkp is just an arena in the heap) 402 * | | 403 * | kvseg | 404 * | | 405 * | | 406 * --- -|-----------------------|- kernelheap (floating) 407 * | Segkmap | 408 * 0xC3002000 -|-----------------------|- segmap_start (floating) 409 * | Red Zone | 410 * 0xC3000000 -|-----------------------|- kernelbase / userlimit (floating) 411 * | | || 412 * | Shared objects | \/ 413 * | | 414 * : : 415 * | user data | 416 * |-----------------------| 417 * | user text | 418 * 0x08048000 -|-----------------------| 419 * | user stack | 420 * : : 421 * | invalid | 422 * 0x00000000 +-----------------------+ 423 * 424 * 425 * 64-bit Kernel's Virtual memory layout. (assuming 64 bit app) 426 * +-----------------------+ 427 * | | 428 * 0xFFFFFFFF.FFC00000 |-----------------------|- ARGSBASE 429 * | debugger (?) | 430 * 0xFFFFFFFF.FF800000 |-----------------------|- SEGDEBUGBASE 431 * | unused | 432 * +-----------------------+ 433 * | Kernel Data | 434 * 0xFFFFFFFF.FBC00000 |-----------------------| 435 * | Kernel Text | 436 * 0xFFFFFFFF.FB800000 |-----------------------|- KERNEL_TEXT 437 * |--- GDT ---|- GDT page (GDT_VA) 438 * |--- debug info ---|- debug info (DEBUG_INFO_VA) 439 * | | 440 * | Core heap | (used for loadable modules) 441 * 0xFFFFFFFF.C0000000 |-----------------------|- core_base / ekernelheap 442 * | Kernel | 443 * | heap | 444 * 0xFFFFFXXX.XXX00000 |-----------------------|- kernelheap (floating) 445 * | segmap | 446 * 0xFFFFFXXX.XXX00000 |-----------------------|- segmap_start (floating) 447 * | device mappings | 448 * 0xFFFFFXXX.XXX00000 |-----------------------|- toxic_addr (floating) 449 * | segzio | 450 * 0xFFFFFXXX.XXX00000 |-----------------------|- segzio_base (floating) 451 * | segkp | 452 * --- |-----------------------|- segkp_base (floating) 453 * | page_t structures | valloc_base + valloc_sz 454 * | memsegs, memlists, | 455 * | page hash, etc. | 456 * 0xFFFFFF00.00000000 |-----------------------|- valloc_base (lower if > 1TB) 457 * | segkpm | 458 * 0xFFFFFE00.00000000 |-----------------------| 459 * | Red Zone | 460 * 0xFFFFFD80.00000000 |-----------------------|- KERNELBASE (lower if > 1TB) 461 * | User stack |- User space memory 462 * | | 463 * | shared objects, etc | (grows downwards) 464 * : : 465 * | | 466 * 0xFFFF8000.00000000 |-----------------------| 467 * | | 468 * | VA Hole / unused | 469 * | | 470 * 0x00008000.00000000 |-----------------------| 471 * | | 472 * | | 473 * : : 474 * | user heap | (grows upwards) 475 * | | 476 * | user data | 477 * |-----------------------| 478 * | user text | 479 * 0x00000000.04000000 |-----------------------| 480 * | invalid | 481 * 0x00000000.00000000 +-----------------------+ 482 * 483 * A 32 bit app on the 64 bit kernel sees the same layout as on the 32 bit 484 * kernel, except that userlimit is raised to 0xfe000000 485 * 486 * Floating values: 487 * 488 * valloc_base: start of the kernel's memory management/tracking data 489 * structures. This region contains page_t structures for 490 * physical memory, memsegs, memlists, and the page hash. 491 * 492 * core_base: start of the kernel's "core" heap area on 64-bit systems. 493 * This area is intended to be used for global data as well as for module 494 * text/data that does not fit into the nucleus pages. The core heap is 495 * restricted to a 2GB range, allowing every address within it to be 496 * accessed using rip-relative addressing 497 * 498 * ekernelheap: end of kernelheap and start of segmap. 499 * 500 * kernelheap: start of kernel heap. On 32-bit systems, this starts right 501 * above a red zone that separates the user's address space from the 502 * kernel's. On 64-bit systems, it sits above segkp and segkpm. 503 * 504 * segmap_start: start of segmap. The length of segmap can be modified 505 * through eeprom. The default length is 16MB on 32-bit systems and 64MB 506 * on 64-bit systems. 507 * 508 * kernelbase: On a 32-bit kernel the default value of 0xd4000000 will be 509 * decreased by 2X the size required for page_t. This allows the kernel 510 * heap to grow in size with physical memory. With sizeof(page_t) == 80 511 * bytes, the following shows the values of kernelbase and kernel heap 512 * sizes for different memory configurations (assuming default segmap and 513 * segkp sizes). 514 * 515 * mem size for kernelbase kernel heap 516 * size page_t's size 517 * ---- --------- ---------- ----------- 518 * 1gb 0x01400000 0xd1800000 684MB 519 * 2gb 0x02800000 0xcf000000 704MB 520 * 4gb 0x05000000 0xca000000 744MB 521 * 6gb 0x07800000 0xc5000000 784MB 522 * 8gb 0x0a000000 0xc0000000 824MB 523 * 16gb 0x14000000 0xac000000 984MB 524 * 32gb 0x28000000 0x84000000 1304MB 525 * 64gb 0x50000000 0x34000000 1944MB (*) 526 * 527 * kernelbase is less than the abi minimum of 0xc0000000 for memory 528 * configurations above 8gb. 529 * 530 * (*) support for memory configurations above 32gb will require manual tuning 531 * of kernelbase to balance out the need of user applications. 532 */ 533 534 /* real-time-clock initialization parameters */ 535 extern time_t process_rtc_config_file(void); 536 537 uintptr_t kernelbase; 538 uintptr_t postbootkernelbase; /* not set till boot loader is gone */ 539 uintptr_t eprom_kernelbase; 540 size_t segmapsize; 541 uintptr_t segmap_start; 542 int segmapfreelists; 543 pgcnt_t npages; 544 pgcnt_t orig_npages; 545 size_t core_size; /* size of "core" heap */ 546 uintptr_t core_base; /* base address of "core" heap */ 547 548 /* 549 * List of bootstrap pages. We mark these as allocated in startup. 550 * release_bootstrap() will free them when we're completely done with 551 * the bootstrap. 552 */ 553 static page_t *bootpages; 554 555 /* 556 * boot time pages that have a vnode from the ramdisk will keep that forever. 557 */ 558 static page_t *rd_pages; 559 560 /* 561 * Lower 64K 562 */ 563 static page_t *lower_pages = NULL; 564 static int lower_pages_count = 0; 565 566 struct system_hardware system_hardware; 567 568 /* 569 * Enable some debugging messages concerning memory usage... 570 */ 571 static void 572 print_memlist(char *title, struct memlist *mp) 573 { 574 prom_printf("MEMLIST: %s:\n", title); 575 while (mp != NULL) { 576 prom_printf("\tAddress 0x%" PRIx64 ", size 0x%" PRIx64 "\n", 577 mp->ml_address, mp->ml_size); 578 mp = mp->ml_next; 579 } 580 } 581 582 /* 583 * XX64 need a comment here.. are these just default values, surely 584 * we read the "cpuid" type information to figure this out. 585 */ 586 int l2cache_sz = 0x80000; 587 int l2cache_linesz = 0x40; 588 int l2cache_assoc = 1; 589 590 static size_t textrepl_min_gb = 10; 591 592 /* 593 * on 64 bit we use a predifined VA range for mapping devices in the kernel 594 * on 32 bit the mappings are intermixed in the heap, so we use a bit map 595 */ 596 #ifdef __amd64 597 598 vmem_t *device_arena; 599 uintptr_t toxic_addr = (uintptr_t)NULL; 600 size_t toxic_size = 1024 * 1024 * 1024; /* Sparc uses 1 gig too */ 601 602 #else /* __i386 */ 603 604 ulong_t *toxic_bit_map; /* one bit for each 4k of VA in heap_arena */ 605 size_t toxic_bit_map_len = 0; /* in bits */ 606 607 #endif /* __i386 */ 608 609 /* 610 * Simple boot time debug facilities 611 */ 612 static char *prm_dbg_str[] = { 613 "%s:%d: '%s' is 0x%x\n", 614 "%s:%d: '%s' is 0x%llx\n" 615 }; 616 617 int prom_debug; 618 619 #define PRM_DEBUG(q) if (prom_debug) \ 620 prom_printf(prm_dbg_str[sizeof (q) >> 3], "startup.c", __LINE__, #q, q); 621 #define PRM_POINT(q) if (prom_debug) \ 622 prom_printf("%s:%d: %s\n", "startup.c", __LINE__, q); 623 624 /* 625 * This structure is used to keep track of the intial allocations 626 * done in startup_memlist(). The value of NUM_ALLOCATIONS needs to 627 * be >= the number of ADD_TO_ALLOCATIONS() executed in the code. 628 */ 629 #define NUM_ALLOCATIONS 8 630 int num_allocations = 0; 631 struct { 632 void **al_ptr; 633 size_t al_size; 634 } allocations[NUM_ALLOCATIONS]; 635 size_t valloc_sz = 0; 636 uintptr_t valloc_base; 637 638 #define ADD_TO_ALLOCATIONS(ptr, size) { \ 639 size = ROUND_UP_PAGE(size); \ 640 if (num_allocations == NUM_ALLOCATIONS) \ 641 panic("too many ADD_TO_ALLOCATIONS()"); \ 642 allocations[num_allocations].al_ptr = (void**)&ptr; \ 643 allocations[num_allocations].al_size = size; \ 644 valloc_sz += size; \ 645 ++num_allocations; \ 646 } 647 648 /* 649 * Allocate all the initial memory needed by the page allocator. 650 */ 651 static void 652 perform_allocations(void) 653 { 654 caddr_t mem; 655 int i; 656 int valloc_align; 657 658 PRM_DEBUG(valloc_base); 659 PRM_DEBUG(valloc_sz); 660 valloc_align = mmu.level_size[mmu.max_page_level > 0]; 661 mem = BOP_ALLOC(bootops, (caddr_t)valloc_base, valloc_sz, valloc_align); 662 if (mem != (caddr_t)valloc_base) 663 panic("BOP_ALLOC() failed"); 664 bzero(mem, valloc_sz); 665 for (i = 0; i < num_allocations; ++i) { 666 *allocations[i].al_ptr = (void *)mem; 667 mem += allocations[i].al_size; 668 } 669 } 670 671 /* 672 * Our world looks like this at startup time. 673 * 674 * In a 32-bit OS, boot loads the kernel text at 0xfe800000 and kernel data 675 * at 0xfec00000. On a 64-bit OS, kernel text and data are loaded at 676 * 0xffffffff.fe800000 and 0xffffffff.fec00000 respectively. Those 677 * addresses are fixed in the binary at link time. 678 * 679 * On the text page: 680 * unix/genunix/krtld/module text loads. 681 * 682 * On the data page: 683 * unix/genunix/krtld/module data loads. 684 * 685 * Machine-dependent startup code 686 */ 687 void 688 startup(void) 689 { 690 #if !defined(__xpv) 691 extern void startup_pci_bios(void); 692 #endif 693 extern cpuset_t cpu_ready_set; 694 695 /* 696 * Make sure that nobody tries to use sekpm until we have 697 * initialized it properly. 698 */ 699 #if defined(__amd64) 700 kpm_desired = 1; 701 #endif 702 kpm_enable = 0; 703 CPUSET_ONLY(cpu_ready_set, 0); /* cpu 0 is boot cpu */ 704 705 #if defined(__xpv) /* XXPV fix me! */ 706 { 707 extern int segvn_use_regions; 708 segvn_use_regions = 0; 709 } 710 #endif 711 progressbar_init(); 712 startup_init(); 713 #if defined(__xpv) 714 startup_xen_version(); 715 #endif 716 startup_memlist(); 717 startup_kmem(); 718 startup_vm(); 719 #if !defined(__xpv) 720 /* 721 * Note we need to do this even on fast reboot in order to access 722 * the irq routing table (used for pci labels). 723 */ 724 startup_pci_bios(); 725 #endif 726 #if defined(__xpv) 727 startup_xen_mca(); 728 #endif 729 startup_modules(); 730 731 startup_end(); 732 } 733 734 static void 735 startup_init() 736 { 737 PRM_POINT("startup_init() starting..."); 738 739 /* 740 * Complete the extraction of cpuid data 741 */ 742 cpuid_pass2(CPU); 743 744 (void) check_boot_version(BOP_GETVERSION(bootops)); 745 746 /* 747 * Check for prom_debug in boot environment 748 */ 749 if (BOP_GETPROPLEN(bootops, "prom_debug") >= 0) { 750 ++prom_debug; 751 PRM_POINT("prom_debug found in boot enviroment"); 752 } 753 754 /* 755 * Collect node, cpu and memory configuration information. 756 */ 757 get_system_configuration(); 758 759 /* 760 * Halt if this is an unsupported processor. 761 */ 762 if (x86_type == X86_TYPE_486 || x86_type == X86_TYPE_CYRIX_486) { 763 printf("\n486 processor (\"%s\") detected.\n", 764 CPU->cpu_brandstr); 765 halt("This processor is not supported by this release " 766 "of Solaris."); 767 } 768 769 PRM_POINT("startup_init() done"); 770 } 771 772 /* 773 * Callback for copy_memlist_filter() to filter nucleus, kadb/kmdb, (ie. 774 * everything mapped above KERNEL_TEXT) pages from phys_avail. Note it 775 * also filters out physical page zero. There is some reliance on the 776 * boot loader allocating only a few contiguous physical memory chunks. 777 */ 778 static void 779 avail_filter(uint64_t *addr, uint64_t *size) 780 { 781 uintptr_t va; 782 uintptr_t next_va; 783 pfn_t pfn; 784 uint64_t pfn_addr; 785 uint64_t pfn_eaddr; 786 uint_t prot; 787 size_t len; 788 uint_t change; 789 790 if (prom_debug) 791 prom_printf("\tFilter: in: a=%" PRIx64 ", s=%" PRIx64 "\n", 792 *addr, *size); 793 794 /* 795 * page zero is required for BIOS.. never make it available 796 */ 797 if (*addr == 0) { 798 *addr += MMU_PAGESIZE; 799 *size -= MMU_PAGESIZE; 800 } 801 802 /* 803 * First we trim from the front of the range. Since kbm_probe() 804 * walks ranges in virtual order, but addr/size are physical, we need 805 * to the list until no changes are seen. This deals with the case 806 * where page "p" is mapped at v, page "p + PAGESIZE" is mapped at w 807 * but w < v. 808 */ 809 do { 810 change = 0; 811 for (va = KERNEL_TEXT; 812 *size > 0 && kbm_probe(&va, &len, &pfn, &prot) != 0; 813 va = next_va) { 814 815 next_va = va + len; 816 pfn_addr = pfn_to_pa(pfn); 817 pfn_eaddr = pfn_addr + len; 818 819 if (pfn_addr <= *addr && pfn_eaddr > *addr) { 820 change = 1; 821 while (*size > 0 && len > 0) { 822 *addr += MMU_PAGESIZE; 823 *size -= MMU_PAGESIZE; 824 len -= MMU_PAGESIZE; 825 } 826 } 827 } 828 if (change && prom_debug) 829 prom_printf("\t\ttrim: a=%" PRIx64 ", s=%" PRIx64 "\n", 830 *addr, *size); 831 } while (change); 832 833 /* 834 * Trim pages from the end of the range. 835 */ 836 for (va = KERNEL_TEXT; 837 *size > 0 && kbm_probe(&va, &len, &pfn, &prot) != 0; 838 va = next_va) { 839 840 next_va = va + len; 841 pfn_addr = pfn_to_pa(pfn); 842 843 if (pfn_addr >= *addr && pfn_addr < *addr + *size) 844 *size = pfn_addr - *addr; 845 } 846 847 if (prom_debug) 848 prom_printf("\tFilter out: a=%" PRIx64 ", s=%" PRIx64 "\n", 849 *addr, *size); 850 } 851 852 static void 853 kpm_init() 854 { 855 struct segkpm_crargs b; 856 857 /* 858 * These variables were all designed for sfmmu in which segkpm is 859 * mapped using a single pagesize - either 8KB or 4MB. On x86, we 860 * might use 2+ page sizes on a single machine, so none of these 861 * variables have a single correct value. They are set up as if we 862 * always use a 4KB pagesize, which should do no harm. In the long 863 * run, we should get rid of KPM's assumption that only a single 864 * pagesize is used. 865 */ 866 kpm_pgshft = MMU_PAGESHIFT; 867 kpm_pgsz = MMU_PAGESIZE; 868 kpm_pgoff = MMU_PAGEOFFSET; 869 kpmp2pshft = 0; 870 kpmpnpgs = 1; 871 ASSERT(((uintptr_t)kpm_vbase & (kpm_pgsz - 1)) == 0); 872 873 PRM_POINT("about to create segkpm"); 874 rw_enter(&kas.a_lock, RW_WRITER); 875 876 if (seg_attach(&kas, kpm_vbase, kpm_size, segkpm) < 0) 877 panic("cannot attach segkpm"); 878 879 b.prot = PROT_READ | PROT_WRITE; 880 b.nvcolors = 1; 881 882 if (segkpm_create(segkpm, (caddr_t)&b) != 0) 883 panic("segkpm_create segkpm"); 884 885 rw_exit(&kas.a_lock); 886 } 887 888 /* 889 * The debug info page provides enough information to allow external 890 * inspectors (e.g. when running under a hypervisor) to bootstrap 891 * themselves into allowing full-blown kernel debugging. 892 */ 893 static void 894 init_debug_info(void) 895 { 896 caddr_t mem; 897 debug_info_t *di; 898 899 #ifndef __lint 900 ASSERT(sizeof (debug_info_t) < MMU_PAGESIZE); 901 #endif 902 903 mem = BOP_ALLOC(bootops, (caddr_t)DEBUG_INFO_VA, MMU_PAGESIZE, 904 MMU_PAGESIZE); 905 906 if (mem != (caddr_t)DEBUG_INFO_VA) 907 panic("BOP_ALLOC() failed"); 908 bzero(mem, MMU_PAGESIZE); 909 910 di = (debug_info_t *)mem; 911 912 di->di_magic = DEBUG_INFO_MAGIC; 913 di->di_version = DEBUG_INFO_VERSION; 914 di->di_modules = (uintptr_t)&modules; 915 di->di_s_text = (uintptr_t)s_text; 916 di->di_e_text = (uintptr_t)e_text; 917 di->di_s_data = (uintptr_t)s_data; 918 di->di_e_data = (uintptr_t)e_data; 919 di->di_hat_htable_off = offsetof(hat_t, hat_htable); 920 di->di_ht_pfn_off = offsetof(htable_t, ht_pfn); 921 } 922 923 /* 924 * Build the memlists and other kernel essential memory system data structures. 925 * This is everything at valloc_base. 926 */ 927 static void 928 startup_memlist(void) 929 { 930 size_t memlist_sz; 931 size_t memseg_sz; 932 size_t pagehash_sz; 933 size_t pp_sz; 934 uintptr_t va; 935 size_t len; 936 uint_t prot; 937 pfn_t pfn; 938 int memblocks; 939 pfn_t rsvd_high_pfn; 940 pgcnt_t rsvd_pgcnt; 941 size_t rsvdmemlist_sz; 942 int rsvdmemblocks; 943 caddr_t pagecolor_mem; 944 size_t pagecolor_memsz; 945 caddr_t page_ctrs_mem; 946 size_t page_ctrs_size; 947 size_t pse_table_alloc_size; 948 struct memlist *current; 949 extern void startup_build_mem_nodes(struct memlist *); 950 951 /* XX64 fix these - they should be in include files */ 952 extern size_t page_coloring_init(uint_t, int, int); 953 extern void page_coloring_setup(caddr_t); 954 955 PRM_POINT("startup_memlist() starting..."); 956 957 /* 958 * Use leftover large page nucleus text/data space for loadable modules. 959 * Use at most MODTEXT/MODDATA. 960 */ 961 len = kbm_nucleus_size; 962 ASSERT(len > MMU_PAGESIZE); 963 964 moddata = (caddr_t)ROUND_UP_PAGE(e_data); 965 e_moddata = (caddr_t)P2ROUNDUP((uintptr_t)e_data, (uintptr_t)len); 966 if (e_moddata - moddata > MODDATA) 967 e_moddata = moddata + MODDATA; 968 969 modtext = (caddr_t)ROUND_UP_PAGE(e_text); 970 e_modtext = (caddr_t)P2ROUNDUP((uintptr_t)e_text, (uintptr_t)len); 971 if (e_modtext - modtext > MODTEXT) 972 e_modtext = modtext + MODTEXT; 973 974 econtig = e_moddata; 975 976 PRM_DEBUG(modtext); 977 PRM_DEBUG(e_modtext); 978 PRM_DEBUG(moddata); 979 PRM_DEBUG(e_moddata); 980 PRM_DEBUG(econtig); 981 982 /* 983 * Examine the boot loader physical memory map to find out: 984 * - total memory in system - physinstalled 985 * - the max physical address - physmax 986 * - the number of discontiguous segments of memory. 987 */ 988 if (prom_debug) 989 print_memlist("boot physinstalled", 990 bootops->boot_mem->physinstalled); 991 installed_top_size_ex(bootops->boot_mem->physinstalled, &physmax, 992 &physinstalled, &memblocks); 993 PRM_DEBUG(physmax); 994 PRM_DEBUG(physinstalled); 995 PRM_DEBUG(memblocks); 996 997 /* 998 * Compute maximum physical address for memory DR operations. 999 * Memory DR operations are unsupported on xpv or 32bit OSes. 1000 */ 1001 #ifdef __amd64 1002 if (plat_dr_support_memory()) { 1003 if (plat_dr_physmax == 0) { 1004 uint_t pabits = UINT_MAX; 1005 1006 cpuid_get_addrsize(CPU, &pabits, NULL); 1007 plat_dr_physmax = btop(1ULL << pabits); 1008 } 1009 if (plat_dr_physmax > PHYSMEM_MAX64) 1010 plat_dr_physmax = PHYSMEM_MAX64; 1011 } else 1012 #endif 1013 plat_dr_physmax = 0; 1014 1015 /* 1016 * Examine the bios reserved memory to find out: 1017 * - the number of discontiguous segments of memory. 1018 */ 1019 if (prom_debug) 1020 print_memlist("boot reserved mem", 1021 bootops->boot_mem->rsvdmem); 1022 installed_top_size_ex(bootops->boot_mem->rsvdmem, &rsvd_high_pfn, 1023 &rsvd_pgcnt, &rsvdmemblocks); 1024 PRM_DEBUG(rsvd_high_pfn); 1025 PRM_DEBUG(rsvd_pgcnt); 1026 PRM_DEBUG(rsvdmemblocks); 1027 1028 /* 1029 * Initialize hat's mmu parameters. 1030 * Check for enforce-prot-exec in boot environment. It's used to 1031 * enable/disable support for the page table entry NX bit. 1032 * The default is to enforce PROT_EXEC on processors that support NX. 1033 * Boot seems to round up the "len", but 8 seems to be big enough. 1034 */ 1035 mmu_init(); 1036 1037 #ifdef __i386 1038 /* 1039 * physmax is lowered if there is more memory than can be 1040 * physically addressed in 32 bit (PAE/non-PAE) modes. 1041 */ 1042 if (mmu.pae_hat) { 1043 if (PFN_ABOVE64G(physmax)) { 1044 physinstalled -= (physmax - (PFN_64G - 1)); 1045 physmax = PFN_64G - 1; 1046 } 1047 } else { 1048 if (PFN_ABOVE4G(physmax)) { 1049 physinstalled -= (physmax - (PFN_4G - 1)); 1050 physmax = PFN_4G - 1; 1051 } 1052 } 1053 #endif 1054 1055 startup_build_mem_nodes(bootops->boot_mem->physinstalled); 1056 1057 if (BOP_GETPROPLEN(bootops, "enforce-prot-exec") >= 0) { 1058 int len = BOP_GETPROPLEN(bootops, "enforce-prot-exec"); 1059 char value[8]; 1060 1061 if (len < 8) 1062 (void) BOP_GETPROP(bootops, "enforce-prot-exec", value); 1063 else 1064 (void) strcpy(value, ""); 1065 if (strcmp(value, "off") == 0) 1066 mmu.pt_nx = 0; 1067 } 1068 PRM_DEBUG(mmu.pt_nx); 1069 1070 /* 1071 * We will need page_t's for every page in the system, except for 1072 * memory mapped at or above above the start of the kernel text segment. 1073 * 1074 * pages above e_modtext are attributed to kernel debugger (obp_pages) 1075 */ 1076 npages = physinstalled - 1; /* avail_filter() skips page 0, so "- 1" */ 1077 obp_pages = 0; 1078 va = KERNEL_TEXT; 1079 while (kbm_probe(&va, &len, &pfn, &prot) != 0) { 1080 npages -= len >> MMU_PAGESHIFT; 1081 if (va >= (uintptr_t)e_moddata) 1082 obp_pages += len >> MMU_PAGESHIFT; 1083 va += len; 1084 } 1085 PRM_DEBUG(npages); 1086 PRM_DEBUG(obp_pages); 1087 1088 /* 1089 * If physmem is patched to be non-zero, use it instead of the computed 1090 * value unless it is larger than the actual amount of memory on hand. 1091 */ 1092 if (physmem == 0 || physmem > npages) { 1093 physmem = npages; 1094 } else if (physmem < npages) { 1095 orig_npages = npages; 1096 npages = physmem; 1097 } 1098 PRM_DEBUG(physmem); 1099 1100 /* 1101 * We now compute the sizes of all the initial allocations for 1102 * structures the kernel needs in order do kmem_alloc(). These 1103 * include: 1104 * memsegs 1105 * memlists 1106 * page hash table 1107 * page_t's 1108 * page coloring data structs 1109 */ 1110 memseg_sz = sizeof (struct memseg) * (memblocks + POSS_NEW_FRAGMENTS); 1111 ADD_TO_ALLOCATIONS(memseg_base, memseg_sz); 1112 PRM_DEBUG(memseg_sz); 1113 1114 /* 1115 * Reserve space for memlists. There's no real good way to know exactly 1116 * how much room we'll need, but this should be a good upper bound. 1117 */ 1118 memlist_sz = ROUND_UP_PAGE(2 * sizeof (struct memlist) * 1119 (memblocks + POSS_NEW_FRAGMENTS)); 1120 ADD_TO_ALLOCATIONS(memlist, memlist_sz); 1121 PRM_DEBUG(memlist_sz); 1122 1123 /* 1124 * Reserve space for bios reserved memlists. 1125 */ 1126 rsvdmemlist_sz = ROUND_UP_PAGE(2 * sizeof (struct memlist) * 1127 (rsvdmemblocks + POSS_NEW_FRAGMENTS)); 1128 ADD_TO_ALLOCATIONS(bios_rsvd, rsvdmemlist_sz); 1129 PRM_DEBUG(rsvdmemlist_sz); 1130 1131 /* LINTED */ 1132 ASSERT(P2SAMEHIGHBIT((1 << PP_SHIFT), sizeof (struct page))); 1133 /* 1134 * The page structure hash table size is a power of 2 1135 * such that the average hash chain length is PAGE_HASHAVELEN. 1136 */ 1137 page_hashsz = npages / PAGE_HASHAVELEN; 1138 page_hashsz_shift = highbit(page_hashsz); 1139 page_hashsz = 1 << page_hashsz_shift; 1140 pagehash_sz = sizeof (struct page *) * page_hashsz; 1141 ADD_TO_ALLOCATIONS(page_hash, pagehash_sz); 1142 PRM_DEBUG(pagehash_sz); 1143 1144 /* 1145 * Set aside room for the page structures themselves. 1146 */ 1147 PRM_DEBUG(npages); 1148 pp_sz = sizeof (struct page) * npages; 1149 ADD_TO_ALLOCATIONS(pp_base, pp_sz); 1150 PRM_DEBUG(pp_sz); 1151 1152 /* 1153 * determine l2 cache info and memory size for page coloring 1154 */ 1155 (void) getl2cacheinfo(CPU, 1156 &l2cache_sz, &l2cache_linesz, &l2cache_assoc); 1157 pagecolor_memsz = 1158 page_coloring_init(l2cache_sz, l2cache_linesz, l2cache_assoc); 1159 ADD_TO_ALLOCATIONS(pagecolor_mem, pagecolor_memsz); 1160 PRM_DEBUG(pagecolor_memsz); 1161 1162 page_ctrs_size = page_ctrs_sz(); 1163 ADD_TO_ALLOCATIONS(page_ctrs_mem, page_ctrs_size); 1164 PRM_DEBUG(page_ctrs_size); 1165 1166 /* 1167 * Allocate the array that protects pp->p_selock. 1168 */ 1169 pse_shift = size_pse_array(physmem, max_ncpus); 1170 pse_table_size = 1 << pse_shift; 1171 pse_table_alloc_size = pse_table_size * sizeof (pad_mutex_t); 1172 ADD_TO_ALLOCATIONS(pse_mutex, pse_table_alloc_size); 1173 1174 #if defined(__amd64) 1175 valloc_sz = ROUND_UP_LPAGE(valloc_sz); 1176 valloc_base = VALLOC_BASE; 1177 1178 /* 1179 * The default values of VALLOC_BASE and SEGKPM_BASE should work 1180 * for values of physmax up to 1 Terabyte. They need adjusting when 1181 * memory is at addresses above 1 TB. When adjusted, segkpm_base must 1182 * be aligned on KERNEL_REDZONE_SIZE boundary (span of top level pte). 1183 */ 1184 if (physmax + 1 > mmu_btop(TERABYTE) || 1185 plat_dr_physmax > mmu_btop(TERABYTE)) { 1186 uint64_t kpm_resv_amount = mmu_ptob(physmax + 1); 1187 1188 if (kpm_resv_amount < mmu_ptob(plat_dr_physmax)) { 1189 kpm_resv_amount = mmu_ptob(plat_dr_physmax); 1190 } 1191 1192 segkpm_base = -(P2ROUNDUP((2 * kpm_resv_amount), 1193 KERNEL_REDZONE_SIZE)); /* down from top VA */ 1194 1195 /* make sure we leave some space for user apps above hole */ 1196 segkpm_base = MAX(segkpm_base, AMD64_VA_HOLE_END + TERABYTE); 1197 if (segkpm_base > SEGKPM_BASE) 1198 segkpm_base = SEGKPM_BASE; 1199 PRM_DEBUG(segkpm_base); 1200 1201 valloc_base = segkpm_base + P2ROUNDUP(kpm_resv_amount, ONE_GIG); 1202 if (valloc_base < segkpm_base) 1203 panic("not enough kernel VA to support memory size"); 1204 PRM_DEBUG(valloc_base); 1205 } 1206 #else /* __i386 */ 1207 valloc_base = (uintptr_t)(MISC_VA_BASE - valloc_sz); 1208 valloc_base = P2ALIGN(valloc_base, mmu.level_size[1]); 1209 PRM_DEBUG(valloc_base); 1210 #endif /* __i386 */ 1211 1212 /* 1213 * do all the initial allocations 1214 */ 1215 perform_allocations(); 1216 1217 /* 1218 * Build phys_install and phys_avail in kernel memspace. 1219 * - phys_install should be all memory in the system. 1220 * - phys_avail is phys_install minus any memory mapped before this 1221 * point above KERNEL_TEXT. 1222 */ 1223 current = phys_install = memlist; 1224 copy_memlist_filter(bootops->boot_mem->physinstalled, ¤t, NULL); 1225 if ((caddr_t)current > (caddr_t)memlist + memlist_sz) 1226 panic("physinstalled was too big!"); 1227 if (prom_debug) 1228 print_memlist("phys_install", phys_install); 1229 1230 phys_avail = current; 1231 PRM_POINT("Building phys_avail:\n"); 1232 copy_memlist_filter(bootops->boot_mem->physinstalled, ¤t, 1233 avail_filter); 1234 if ((caddr_t)current > (caddr_t)memlist + memlist_sz) 1235 panic("physavail was too big!"); 1236 if (prom_debug) 1237 print_memlist("phys_avail", phys_avail); 1238 #ifndef __xpv 1239 /* 1240 * Free unused memlist items, which may be used by memory DR driver 1241 * at runtime. 1242 */ 1243 if ((caddr_t)current < (caddr_t)memlist + memlist_sz) { 1244 memlist_free_block((caddr_t)current, 1245 (caddr_t)memlist + memlist_sz - (caddr_t)current); 1246 } 1247 #endif 1248 1249 /* 1250 * Build bios reserved memspace 1251 */ 1252 current = bios_rsvd; 1253 copy_memlist_filter(bootops->boot_mem->rsvdmem, ¤t, NULL); 1254 if ((caddr_t)current > (caddr_t)bios_rsvd + rsvdmemlist_sz) 1255 panic("bios_rsvd was too big!"); 1256 if (prom_debug) 1257 print_memlist("bios_rsvd", bios_rsvd); 1258 #ifndef __xpv 1259 /* 1260 * Free unused memlist items, which may be used by memory DR driver 1261 * at runtime. 1262 */ 1263 if ((caddr_t)current < (caddr_t)bios_rsvd + rsvdmemlist_sz) { 1264 memlist_free_block((caddr_t)current, 1265 (caddr_t)bios_rsvd + rsvdmemlist_sz - (caddr_t)current); 1266 } 1267 #endif 1268 1269 /* 1270 * setup page coloring 1271 */ 1272 page_coloring_setup(pagecolor_mem); 1273 page_lock_init(); /* currently a no-op */ 1274 1275 /* 1276 * free page list counters 1277 */ 1278 (void) page_ctrs_alloc(page_ctrs_mem); 1279 1280 /* 1281 * Size the pcf array based on the number of cpus in the box at 1282 * boot time. 1283 */ 1284 1285 pcf_init(); 1286 1287 /* 1288 * Initialize the page structures from the memory lists. 1289 */ 1290 availrmem_initial = availrmem = freemem = 0; 1291 PRM_POINT("Calling kphysm_init()..."); 1292 npages = kphysm_init(pp_base, npages); 1293 PRM_POINT("kphysm_init() done"); 1294 PRM_DEBUG(npages); 1295 1296 init_debug_info(); 1297 1298 /* 1299 * Now that page_t's have been initialized, remove all the 1300 * initial allocation pages from the kernel free page lists. 1301 */ 1302 boot_mapin((caddr_t)valloc_base, valloc_sz); 1303 boot_mapin((caddr_t)MISC_VA_BASE, MISC_VA_SIZE); 1304 PRM_POINT("startup_memlist() done"); 1305 1306 PRM_DEBUG(valloc_sz); 1307 1308 #if defined(__amd64) 1309 if ((availrmem >> (30 - MMU_PAGESHIFT)) >= 1310 textrepl_min_gb && l2cache_sz <= 2 << 20) { 1311 extern size_t textrepl_size_thresh; 1312 textrepl_size_thresh = (16 << 20) - 1; 1313 } 1314 #endif 1315 } 1316 1317 /* 1318 * Layout the kernel's part of address space and initialize kmem allocator. 1319 */ 1320 static void 1321 startup_kmem(void) 1322 { 1323 extern void page_set_colorequiv_arr(void); 1324 1325 PRM_POINT("startup_kmem() starting..."); 1326 1327 #if defined(__amd64) 1328 if (eprom_kernelbase && eprom_kernelbase != KERNELBASE) 1329 cmn_err(CE_NOTE, "!kernelbase cannot be changed on 64-bit " 1330 "systems."); 1331 kernelbase = segkpm_base - KERNEL_REDZONE_SIZE; 1332 core_base = (uintptr_t)COREHEAP_BASE; 1333 core_size = (size_t)MISC_VA_BASE - COREHEAP_BASE; 1334 #else /* __i386 */ 1335 /* 1336 * We configure kernelbase based on: 1337 * 1338 * 1. user specified kernelbase via eeprom command. Value cannot exceed 1339 * KERNELBASE_MAX. we large page align eprom_kernelbase 1340 * 1341 * 2. Default to KERNELBASE and adjust to 2X less the size for page_t. 1342 * On large memory systems we must lower kernelbase to allow 1343 * enough room for page_t's for all of memory. 1344 * 1345 * The value set here, might be changed a little later. 1346 */ 1347 if (eprom_kernelbase) { 1348 kernelbase = eprom_kernelbase & mmu.level_mask[1]; 1349 if (kernelbase > KERNELBASE_MAX) 1350 kernelbase = KERNELBASE_MAX; 1351 } else { 1352 kernelbase = (uintptr_t)KERNELBASE; 1353 kernelbase -= ROUND_UP_4MEG(2 * valloc_sz); 1354 } 1355 ASSERT((kernelbase & mmu.level_offset[1]) == 0); 1356 core_base = valloc_base; 1357 core_size = 0; 1358 #endif /* __i386 */ 1359 1360 PRM_DEBUG(core_base); 1361 PRM_DEBUG(core_size); 1362 PRM_DEBUG(kernelbase); 1363 1364 #if defined(__i386) 1365 segkp_fromheap = 1; 1366 #endif /* __i386 */ 1367 1368 ekernelheap = (char *)core_base; 1369 PRM_DEBUG(ekernelheap); 1370 1371 /* 1372 * Now that we know the real value of kernelbase, 1373 * update variables that were initialized with a value of 1374 * KERNELBASE (in common/conf/param.c). 1375 * 1376 * XXX The problem with this sort of hackery is that the 1377 * compiler just may feel like putting the const declarations 1378 * (in param.c) into the .text section. Perhaps they should 1379 * just be declared as variables there? 1380 */ 1381 1382 *(uintptr_t *)&_kernelbase = kernelbase; 1383 *(uintptr_t *)&_userlimit = kernelbase; 1384 #if defined(__amd64) 1385 *(uintptr_t *)&_userlimit -= KERNELBASE - USERLIMIT; 1386 #else 1387 *(uintptr_t *)&_userlimit32 = _userlimit; 1388 #endif 1389 PRM_DEBUG(_kernelbase); 1390 PRM_DEBUG(_userlimit); 1391 PRM_DEBUG(_userlimit32); 1392 1393 layout_kernel_va(); 1394 1395 #if defined(__i386) 1396 /* 1397 * If segmap is too large we can push the bottom of the kernel heap 1398 * higher than the base. Or worse, it could exceed the top of the 1399 * VA space entirely, causing it to wrap around. 1400 */ 1401 if (kernelheap >= ekernelheap || (uintptr_t)kernelheap < kernelbase) 1402 panic("too little address space available for kernelheap," 1403 " use eeprom for lower kernelbase or smaller segmapsize"); 1404 #endif /* __i386 */ 1405 1406 /* 1407 * Initialize the kernel heap. Note 3rd argument must be > 1st. 1408 */ 1409 kernelheap_init(kernelheap, ekernelheap, 1410 kernelheap + MMU_PAGESIZE, 1411 (void *)core_base, (void *)(core_base + core_size)); 1412 1413 #if defined(__xpv) 1414 /* 1415 * Link pending events struct into cpu struct 1416 */ 1417 CPU->cpu_m.mcpu_evt_pend = &cpu0_evt_data; 1418 #endif 1419 /* 1420 * Initialize kernel memory allocator. 1421 */ 1422 kmem_init(); 1423 1424 /* 1425 * Factor in colorequiv to check additional 'equivalent' bins 1426 */ 1427 page_set_colorequiv_arr(); 1428 1429 /* 1430 * print this out early so that we know what's going on 1431 */ 1432 print_x86_featureset(x86_featureset); 1433 1434 /* 1435 * Initialize bp_mapin(). 1436 */ 1437 bp_init(MMU_PAGESIZE, HAT_STORECACHING_OK); 1438 1439 /* 1440 * orig_npages is non-zero if physmem has been configured for less 1441 * than the available memory. 1442 */ 1443 if (orig_npages) { 1444 cmn_err(CE_WARN, "!%slimiting physmem to 0x%lx of 0x%lx pages", 1445 (npages == PHYSMEM ? "Due to virtual address space " : ""), 1446 npages, orig_npages); 1447 } 1448 #if defined(__i386) 1449 if (eprom_kernelbase && (eprom_kernelbase != kernelbase)) 1450 cmn_err(CE_WARN, "kernelbase value, User specified 0x%lx, " 1451 "System using 0x%lx", 1452 (uintptr_t)eprom_kernelbase, (uintptr_t)kernelbase); 1453 #endif 1454 1455 #ifdef KERNELBASE_ABI_MIN 1456 if (kernelbase < (uintptr_t)KERNELBASE_ABI_MIN) { 1457 cmn_err(CE_NOTE, "!kernelbase set to 0x%lx, system is not " 1458 "i386 ABI compliant.", (uintptr_t)kernelbase); 1459 } 1460 #endif 1461 1462 #ifndef __xpv 1463 if (plat_dr_support_memory()) { 1464 mem_config_init(); 1465 } 1466 #else /* __xpv */ 1467 /* 1468 * Some of the xen start information has to be relocated up 1469 * into the kernel's permanent address space. 1470 */ 1471 PRM_POINT("calling xen_relocate_start_info()"); 1472 xen_relocate_start_info(); 1473 PRM_POINT("xen_relocate_start_info() done"); 1474 1475 /* 1476 * (Update the vcpu pointer in our cpu structure to point into 1477 * the relocated shared info.) 1478 */ 1479 CPU->cpu_m.mcpu_vcpu_info = 1480 &HYPERVISOR_shared_info->vcpu_info[CPU->cpu_id]; 1481 #endif /* __xpv */ 1482 1483 PRM_POINT("startup_kmem() done"); 1484 } 1485 1486 #ifndef __xpv 1487 /* 1488 * If we have detected that we are running in an HVM environment, we need 1489 * to prepend the PV driver directory to the module search path. 1490 */ 1491 #define HVM_MOD_DIR "/platform/i86hvm/kernel" 1492 static void 1493 update_default_path() 1494 { 1495 char *current, *newpath; 1496 int newlen; 1497 1498 /* 1499 * We are about to resync with krtld. krtld will reset its 1500 * internal module search path iff Solaris has set default_path. 1501 * We want to be sure we're prepending this new directory to the 1502 * right search path. 1503 */ 1504 current = (default_path == NULL) ? kobj_module_path : default_path; 1505 1506 newlen = strlen(HVM_MOD_DIR) + strlen(current) + 2; 1507 newpath = kmem_alloc(newlen, KM_SLEEP); 1508 (void) strcpy(newpath, HVM_MOD_DIR); 1509 (void) strcat(newpath, " "); 1510 (void) strcat(newpath, current); 1511 1512 default_path = newpath; 1513 } 1514 #endif 1515 1516 static void 1517 startup_modules(void) 1518 { 1519 int cnt; 1520 extern void prom_setup(void); 1521 int32_t v, h; 1522 char d[11]; 1523 char *cp; 1524 cmi_hdl_t hdl; 1525 1526 PRM_POINT("startup_modules() starting..."); 1527 1528 #ifndef __xpv 1529 /* 1530 * Initialize ten-micro second timer so that drivers will 1531 * not get short changed in their init phase. This was 1532 * not getting called until clkinit which, on fast cpu's 1533 * caused the drv_usecwait to be way too short. 1534 */ 1535 microfind(); 1536 1537 if (get_hwenv() == HW_XEN_HVM) 1538 update_default_path(); 1539 #endif 1540 1541 /* 1542 * Read the GMT lag from /etc/rtc_config. 1543 */ 1544 sgmtl(process_rtc_config_file()); 1545 1546 /* 1547 * Calculate default settings of system parameters based upon 1548 * maxusers, yet allow to be overridden via the /etc/system file. 1549 */ 1550 param_calc(0); 1551 1552 mod_setup(); 1553 1554 /* 1555 * Initialize system parameters. 1556 */ 1557 param_init(); 1558 1559 /* 1560 * Initialize the default brands 1561 */ 1562 brand_init(); 1563 1564 /* 1565 * maxmem is the amount of physical memory we're playing with. 1566 */ 1567 maxmem = physmem; 1568 1569 /* 1570 * Initialize segment management stuff. 1571 */ 1572 seg_init(); 1573 1574 if (modload("fs", "specfs") == -1) 1575 halt("Can't load specfs"); 1576 1577 if (modload("fs", "devfs") == -1) 1578 halt("Can't load devfs"); 1579 1580 if (modload("fs", "dev") == -1) 1581 halt("Can't load dev"); 1582 1583 if (modload("fs", "procfs") == -1) 1584 halt("Can't load procfs"); 1585 1586 (void) modloadonly("sys", "lbl_edition"); 1587 1588 dispinit(); 1589 1590 /* Read cluster configuration data. */ 1591 clconf_init(); 1592 1593 #if defined(__xpv) 1594 (void) ec_init(); 1595 gnttab_init(); 1596 (void) xs_early_init(); 1597 #endif /* __xpv */ 1598 1599 /* 1600 * Create a kernel device tree. First, create rootnex and 1601 * then invoke bus specific code to probe devices. 1602 */ 1603 setup_ddi(); 1604 1605 #ifdef __xpv 1606 if (DOMAIN_IS_INITDOMAIN(xen_info)) 1607 #endif 1608 { 1609 /* 1610 * Load the System Management BIOS into the global ksmbios 1611 * handle, if an SMBIOS is present on this system. 1612 */ 1613 ksmbios = smbios_open(NULL, SMB_VERSION, ksmbios_flags, NULL); 1614 } 1615 1616 1617 /* 1618 * Originally clconf_init() apparently needed the hostid. But 1619 * this no longer appears to be true - it uses its own nodeid. 1620 * By placing the hostid logic here, we are able to make use of 1621 * the SMBIOS UUID. 1622 */ 1623 if ((h = set_soft_hostid()) == HW_INVALID_HOSTID) { 1624 cmn_err(CE_WARN, "Unable to set hostid"); 1625 } else { 1626 for (v = h, cnt = 0; cnt < 10; cnt++) { 1627 d[cnt] = (char)(v % 10); 1628 v /= 10; 1629 if (v == 0) 1630 break; 1631 } 1632 for (cp = hw_serial; cnt >= 0; cnt--) 1633 *cp++ = d[cnt] + '0'; 1634 *cp = 0; 1635 } 1636 1637 /* 1638 * Set up the CPU module subsystem for the boot cpu in the native 1639 * case, and all physical cpu resource in the xpv dom0 case. 1640 * Modifies the device tree, so this must be done after 1641 * setup_ddi(). 1642 */ 1643 #ifdef __xpv 1644 /* 1645 * If paravirtualized and on dom0 then we initialize all physical 1646 * cpu handles now; if paravirtualized on a domU then do not 1647 * initialize. 1648 */ 1649 if (DOMAIN_IS_INITDOMAIN(xen_info)) { 1650 xen_mc_lcpu_cookie_t cpi; 1651 1652 for (cpi = xen_physcpu_next(NULL); cpi != NULL; 1653 cpi = xen_physcpu_next(cpi)) { 1654 if ((hdl = cmi_init(CMI_HDL_SOLARIS_xVM_MCA, 1655 xen_physcpu_chipid(cpi), xen_physcpu_coreid(cpi), 1656 xen_physcpu_strandid(cpi))) != NULL && 1657 is_x86_feature(x86_featureset, X86FSET_MCA)) 1658 cmi_mca_init(hdl); 1659 } 1660 } 1661 #else 1662 /* 1663 * Initialize a handle for the boot cpu - others will initialize 1664 * as they startup. Do not do this if we know we are in an HVM domU. 1665 */ 1666 if ((get_hwenv() != HW_XEN_HVM) && 1667 (hdl = cmi_init(CMI_HDL_NATIVE, cmi_ntv_hwchipid(CPU), 1668 cmi_ntv_hwcoreid(CPU), cmi_ntv_hwstrandid(CPU))) != NULL && 1669 is_x86_feature(x86_featureset, X86FSET_MCA)) { 1670 cmi_mca_init(hdl); 1671 CPU->cpu_m.mcpu_cmi_hdl = hdl; 1672 } 1673 #endif /* __xpv */ 1674 1675 /* 1676 * Fake a prom tree such that /dev/openprom continues to work 1677 */ 1678 PRM_POINT("startup_modules: calling prom_setup..."); 1679 prom_setup(); 1680 PRM_POINT("startup_modules: done"); 1681 1682 /* 1683 * Load all platform specific modules 1684 */ 1685 PRM_POINT("startup_modules: calling psm_modload..."); 1686 psm_modload(); 1687 1688 PRM_POINT("startup_modules() done"); 1689 } 1690 1691 /* 1692 * claim a "setaside" boot page for use in the kernel 1693 */ 1694 page_t * 1695 boot_claim_page(pfn_t pfn) 1696 { 1697 page_t *pp; 1698 1699 pp = page_numtopp_nolock(pfn); 1700 ASSERT(pp != NULL); 1701 1702 if (PP_ISBOOTPAGES(pp)) { 1703 if (pp->p_next != NULL) 1704 pp->p_next->p_prev = pp->p_prev; 1705 if (pp->p_prev == NULL) 1706 bootpages = pp->p_next; 1707 else 1708 pp->p_prev->p_next = pp->p_next; 1709 } else { 1710 /* 1711 * htable_attach() expects a base pagesize page 1712 */ 1713 if (pp->p_szc != 0) 1714 page_boot_demote(pp); 1715 pp = page_numtopp(pfn, SE_EXCL); 1716 } 1717 return (pp); 1718 } 1719 1720 /* 1721 * Walk through the pagetables looking for pages mapped in by boot. If the 1722 * setaside flag is set the pages are expected to be returned to the 1723 * kernel later in boot, so we add them to the bootpages list. 1724 */ 1725 static void 1726 protect_boot_range(uintptr_t low, uintptr_t high, int setaside) 1727 { 1728 uintptr_t va = low; 1729 size_t len; 1730 uint_t prot; 1731 pfn_t pfn; 1732 page_t *pp; 1733 pgcnt_t boot_protect_cnt = 0; 1734 1735 while (kbm_probe(&va, &len, &pfn, &prot) != 0 && va < high) { 1736 if (va + len >= high) 1737 panic("0x%lx byte mapping at 0x%p exceeds boot's " 1738 "legal range.", len, (void *)va); 1739 1740 while (len > 0) { 1741 pp = page_numtopp_alloc(pfn); 1742 if (pp != NULL) { 1743 if (setaside == 0) 1744 panic("Unexpected mapping by boot. " 1745 "addr=%p pfn=%lx\n", 1746 (void *)va, pfn); 1747 1748 pp->p_next = bootpages; 1749 pp->p_prev = NULL; 1750 PP_SETBOOTPAGES(pp); 1751 if (bootpages != NULL) { 1752 bootpages->p_prev = pp; 1753 } 1754 bootpages = pp; 1755 ++boot_protect_cnt; 1756 } 1757 1758 ++pfn; 1759 len -= MMU_PAGESIZE; 1760 va += MMU_PAGESIZE; 1761 } 1762 } 1763 PRM_DEBUG(boot_protect_cnt); 1764 } 1765 1766 /* 1767 * 1768 */ 1769 static void 1770 layout_kernel_va(void) 1771 { 1772 PRM_POINT("layout_kernel_va() starting..."); 1773 /* 1774 * Establish the final size of the kernel's heap, size of segmap, 1775 * segkp, etc. 1776 */ 1777 1778 #if defined(__amd64) 1779 1780 kpm_vbase = (caddr_t)segkpm_base; 1781 if (physmax + 1 < plat_dr_physmax) { 1782 kpm_size = ROUND_UP_LPAGE(mmu_ptob(plat_dr_physmax)); 1783 } else { 1784 kpm_size = ROUND_UP_LPAGE(mmu_ptob(physmax + 1)); 1785 } 1786 if ((uintptr_t)kpm_vbase + kpm_size > (uintptr_t)valloc_base) 1787 panic("not enough room for kpm!"); 1788 PRM_DEBUG(kpm_size); 1789 PRM_DEBUG(kpm_vbase); 1790 1791 /* 1792 * By default we create a seg_kp in 64 bit kernels, it's a little 1793 * faster to access than embedding it in the heap. 1794 */ 1795 segkp_base = (caddr_t)valloc_base + valloc_sz; 1796 if (!segkp_fromheap) { 1797 size_t sz = mmu_ptob(segkpsize); 1798 1799 /* 1800 * determine size of segkp 1801 */ 1802 if (sz < SEGKPMINSIZE || sz > SEGKPMAXSIZE) { 1803 sz = SEGKPDEFSIZE; 1804 cmn_err(CE_WARN, "!Illegal value for segkpsize. " 1805 "segkpsize has been reset to %ld pages", 1806 mmu_btop(sz)); 1807 } 1808 sz = MIN(sz, MAX(SEGKPMINSIZE, mmu_ptob(physmem))); 1809 1810 segkpsize = mmu_btop(ROUND_UP_LPAGE(sz)); 1811 } 1812 PRM_DEBUG(segkp_base); 1813 PRM_DEBUG(segkpsize); 1814 1815 /* 1816 * segzio is used for ZFS cached data. It uses a distinct VA 1817 * segment (from kernel heap) so that we can easily tell not to 1818 * include it in kernel crash dumps on 64 bit kernels. The trick is 1819 * to give it lots of VA, but not constrain the kernel heap. 1820 * We scale the size of segzio linearly with physmem up to 1821 * SEGZIOMAXSIZE. Above that amount it scales at 50% of physmem. 1822 */ 1823 segzio_base = segkp_base + mmu_ptob(segkpsize); 1824 if (segzio_fromheap) { 1825 segziosize = 0; 1826 } else { 1827 size_t physmem_size = mmu_ptob(physmem); 1828 size_t size = (segziosize == 0) ? 1829 physmem_size : mmu_ptob(segziosize); 1830 1831 if (size < SEGZIOMINSIZE) 1832 size = SEGZIOMINSIZE; 1833 if (size > SEGZIOMAXSIZE) { 1834 size = SEGZIOMAXSIZE; 1835 if (physmem_size > size) 1836 size += (physmem_size - size) / 2; 1837 } 1838 segziosize = mmu_btop(ROUND_UP_LPAGE(size)); 1839 } 1840 PRM_DEBUG(segziosize); 1841 PRM_DEBUG(segzio_base); 1842 1843 /* 1844 * Put the range of VA for device mappings next, kmdb knows to not 1845 * grep in this range of addresses. 1846 */ 1847 toxic_addr = 1848 ROUND_UP_LPAGE((uintptr_t)segzio_base + mmu_ptob(segziosize)); 1849 PRM_DEBUG(toxic_addr); 1850 segmap_start = ROUND_UP_LPAGE(toxic_addr + toxic_size); 1851 #else /* __i386 */ 1852 segmap_start = ROUND_UP_LPAGE(kernelbase); 1853 #endif /* __i386 */ 1854 PRM_DEBUG(segmap_start); 1855 1856 /* 1857 * Users can change segmapsize through eeprom. If the variable 1858 * is tuned through eeprom, there is no upper bound on the 1859 * size of segmap. 1860 */ 1861 segmapsize = MAX(ROUND_UP_LPAGE(segmapsize), SEGMAPDEFAULT); 1862 1863 #if defined(__i386) 1864 /* 1865 * 32-bit systems don't have segkpm or segkp, so segmap appears at 1866 * the bottom of the kernel's address range. Set aside space for a 1867 * small red zone just below the start of segmap. 1868 */ 1869 segmap_start += KERNEL_REDZONE_SIZE; 1870 segmapsize -= KERNEL_REDZONE_SIZE; 1871 #endif 1872 1873 PRM_DEBUG(segmap_start); 1874 PRM_DEBUG(segmapsize); 1875 kernelheap = (caddr_t)ROUND_UP_LPAGE(segmap_start + segmapsize); 1876 PRM_DEBUG(kernelheap); 1877 PRM_POINT("layout_kernel_va() done..."); 1878 } 1879 1880 /* 1881 * Finish initializing the VM system, now that we are no longer 1882 * relying on the boot time memory allocators. 1883 */ 1884 static void 1885 startup_vm(void) 1886 { 1887 struct segmap_crargs a; 1888 1889 extern int use_brk_lpg, use_stk_lpg; 1890 1891 PRM_POINT("startup_vm() starting..."); 1892 1893 /* 1894 * Initialize the hat layer. 1895 */ 1896 hat_init(); 1897 1898 /* 1899 * Do final allocations of HAT data structures that need to 1900 * be allocated before quiescing the boot loader. 1901 */ 1902 PRM_POINT("Calling hat_kern_alloc()..."); 1903 hat_kern_alloc((caddr_t)segmap_start, segmapsize, ekernelheap); 1904 PRM_POINT("hat_kern_alloc() done"); 1905 1906 #ifndef __xpv 1907 /* 1908 * Setup Page Attribute Table 1909 */ 1910 pat_sync(); 1911 #endif 1912 1913 /* 1914 * The next two loops are done in distinct steps in order 1915 * to be sure that any page that is doubly mapped (both above 1916 * KERNEL_TEXT and below kernelbase) is dealt with correctly. 1917 * Note this may never happen, but it might someday. 1918 */ 1919 bootpages = NULL; 1920 PRM_POINT("Protecting boot pages"); 1921 1922 /* 1923 * Protect any pages mapped above KERNEL_TEXT that somehow have 1924 * page_t's. This can only happen if something weird allocated 1925 * in this range (like kadb/kmdb). 1926 */ 1927 protect_boot_range(KERNEL_TEXT, (uintptr_t)-1, 0); 1928 1929 /* 1930 * Before we can take over memory allocation/mapping from the boot 1931 * loader we must remove from our free page lists any boot allocated 1932 * pages that stay mapped until release_bootstrap(). 1933 */ 1934 protect_boot_range(0, kernelbase, 1); 1935 1936 1937 /* 1938 * Switch to running on regular HAT (not boot_mmu) 1939 */ 1940 PRM_POINT("Calling hat_kern_setup()..."); 1941 hat_kern_setup(); 1942 1943 /* 1944 * It is no longer safe to call BOP_ALLOC(), so make sure we don't. 1945 */ 1946 bop_no_more_mem(); 1947 1948 PRM_POINT("hat_kern_setup() done"); 1949 1950 hat_cpu_online(CPU); 1951 1952 /* 1953 * Initialize VM system 1954 */ 1955 PRM_POINT("Calling kvm_init()..."); 1956 kvm_init(); 1957 PRM_POINT("kvm_init() done"); 1958 1959 /* 1960 * Tell kmdb that the VM system is now working 1961 */ 1962 if (boothowto & RB_DEBUG) 1963 kdi_dvec_vmready(); 1964 1965 #if defined(__xpv) 1966 /* 1967 * Populate the I/O pool on domain 0 1968 */ 1969 if (DOMAIN_IS_INITDOMAIN(xen_info)) { 1970 extern long populate_io_pool(void); 1971 long init_io_pool_cnt; 1972 1973 PRM_POINT("Populating reserve I/O page pool"); 1974 init_io_pool_cnt = populate_io_pool(); 1975 PRM_DEBUG(init_io_pool_cnt); 1976 } 1977 #endif 1978 /* 1979 * Mangle the brand string etc. 1980 */ 1981 cpuid_pass3(CPU); 1982 1983 #if defined(__amd64) 1984 1985 /* 1986 * Create the device arena for toxic (to dtrace/kmdb) mappings. 1987 */ 1988 device_arena = vmem_create("device", (void *)toxic_addr, 1989 toxic_size, MMU_PAGESIZE, NULL, NULL, NULL, 0, VM_SLEEP); 1990 1991 #else /* __i386 */ 1992 1993 /* 1994 * allocate the bit map that tracks toxic pages 1995 */ 1996 toxic_bit_map_len = btop((ulong_t)(valloc_base - kernelbase)); 1997 PRM_DEBUG(toxic_bit_map_len); 1998 toxic_bit_map = 1999 kmem_zalloc(BT_SIZEOFMAP(toxic_bit_map_len), KM_NOSLEEP); 2000 ASSERT(toxic_bit_map != NULL); 2001 PRM_DEBUG(toxic_bit_map); 2002 2003 #endif /* __i386 */ 2004 2005 2006 /* 2007 * Now that we've got more VA, as well as the ability to allocate from 2008 * it, tell the debugger. 2009 */ 2010 if (boothowto & RB_DEBUG) 2011 kdi_dvec_memavail(); 2012 2013 /* 2014 * The following code installs a special page fault handler (#pf) 2015 * to work around a pentium bug. 2016 */ 2017 #if !defined(__amd64) && !defined(__xpv) 2018 if (x86_type == X86_TYPE_P5) { 2019 desctbr_t idtr; 2020 gate_desc_t *newidt; 2021 2022 if ((newidt = kmem_zalloc(MMU_PAGESIZE, KM_NOSLEEP)) == NULL) 2023 panic("failed to install pentium_pftrap"); 2024 2025 bcopy(idt0, newidt, NIDT * sizeof (*idt0)); 2026 set_gatesegd(&newidt[T_PGFLT], &pentium_pftrap, 2027 KCS_SEL, SDT_SYSIGT, TRP_KPL, 0); 2028 2029 (void) as_setprot(&kas, (caddr_t)newidt, MMU_PAGESIZE, 2030 PROT_READ | PROT_EXEC); 2031 2032 CPU->cpu_idt = newidt; 2033 idtr.dtr_base = (uintptr_t)CPU->cpu_idt; 2034 idtr.dtr_limit = (NIDT * sizeof (*idt0)) - 1; 2035 wr_idtr(&idtr); 2036 } 2037 #endif /* !__amd64 */ 2038 2039 #if !defined(__xpv) 2040 /* 2041 * Map page pfn=0 for drivers, such as kd, that need to pick up 2042 * parameters left there by controllers/BIOS. 2043 */ 2044 PRM_POINT("setup up p0_va"); 2045 p0_va = i86devmap(0, 1, PROT_READ); 2046 PRM_DEBUG(p0_va); 2047 #endif 2048 2049 cmn_err(CE_CONT, "?mem = %luK (0x%lx)\n", 2050 physinstalled << (MMU_PAGESHIFT - 10), ptob(physinstalled)); 2051 2052 /* 2053 * disable automatic large pages for small memory systems or 2054 * when the disable flag is set. 2055 * 2056 * Do not yet consider page sizes larger than 2m/4m. 2057 */ 2058 if (!auto_lpg_disable && mmu.max_page_level > 0) { 2059 max_uheap_lpsize = LEVEL_SIZE(1); 2060 max_ustack_lpsize = LEVEL_SIZE(1); 2061 max_privmap_lpsize = LEVEL_SIZE(1); 2062 max_uidata_lpsize = LEVEL_SIZE(1); 2063 max_utext_lpsize = LEVEL_SIZE(1); 2064 max_shm_lpsize = LEVEL_SIZE(1); 2065 } 2066 if (physmem < privm_lpg_min_physmem || mmu.max_page_level == 0 || 2067 auto_lpg_disable) { 2068 use_brk_lpg = 0; 2069 use_stk_lpg = 0; 2070 } 2071 mcntl0_lpsize = LEVEL_SIZE(mmu.umax_page_level); 2072 2073 PRM_POINT("Calling hat_init_finish()..."); 2074 hat_init_finish(); 2075 PRM_POINT("hat_init_finish() done"); 2076 2077 /* 2078 * Initialize the segkp segment type. 2079 */ 2080 rw_enter(&kas.a_lock, RW_WRITER); 2081 PRM_POINT("Attaching segkp"); 2082 if (segkp_fromheap) { 2083 segkp->s_as = &kas; 2084 } else if (seg_attach(&kas, (caddr_t)segkp_base, mmu_ptob(segkpsize), 2085 segkp) < 0) { 2086 panic("startup: cannot attach segkp"); 2087 /*NOTREACHED*/ 2088 } 2089 PRM_POINT("Doing segkp_create()"); 2090 if (segkp_create(segkp) != 0) { 2091 panic("startup: segkp_create failed"); 2092 /*NOTREACHED*/ 2093 } 2094 PRM_DEBUG(segkp); 2095 rw_exit(&kas.a_lock); 2096 2097 /* 2098 * kpm segment 2099 */ 2100 segmap_kpm = 0; 2101 if (kpm_desired) { 2102 kpm_init(); 2103 kpm_enable = 1; 2104 } 2105 2106 /* 2107 * Now create segmap segment. 2108 */ 2109 rw_enter(&kas.a_lock, RW_WRITER); 2110 if (seg_attach(&kas, (caddr_t)segmap_start, segmapsize, segmap) < 0) { 2111 panic("cannot attach segmap"); 2112 /*NOTREACHED*/ 2113 } 2114 PRM_DEBUG(segmap); 2115 2116 a.prot = PROT_READ | PROT_WRITE; 2117 a.shmsize = 0; 2118 a.nfreelist = segmapfreelists; 2119 2120 if (segmap_create(segmap, (caddr_t)&a) != 0) 2121 panic("segmap_create segmap"); 2122 rw_exit(&kas.a_lock); 2123 2124 setup_vaddr_for_ppcopy(CPU); 2125 2126 segdev_init(); 2127 #if defined(__xpv) 2128 if (DOMAIN_IS_INITDOMAIN(xen_info)) 2129 #endif 2130 pmem_init(); 2131 2132 PRM_POINT("startup_vm() done"); 2133 } 2134 2135 /* 2136 * Load a tod module for the non-standard tod part found on this system. 2137 */ 2138 static void 2139 load_tod_module(char *todmod) 2140 { 2141 if (modload("tod", todmod) == -1) 2142 halt("Can't load TOD module"); 2143 } 2144 2145 static void 2146 startup_end(void) 2147 { 2148 int i; 2149 extern void setx86isalist(void); 2150 extern void cpu_event_init(void); 2151 2152 PRM_POINT("startup_end() starting..."); 2153 2154 /* 2155 * Perform tasks that get done after most of the VM 2156 * initialization has been done but before the clock 2157 * and other devices get started. 2158 */ 2159 kern_setup1(); 2160 2161 /* 2162 * Perform CPC initialization for this CPU. 2163 */ 2164 kcpc_hw_init(CPU); 2165 2166 /* 2167 * Initialize cpu event framework. 2168 */ 2169 cpu_event_init(); 2170 2171 #if defined(OPTERON_WORKAROUND_6323525) 2172 if (opteron_workaround_6323525) 2173 patch_workaround_6323525(); 2174 #endif 2175 /* 2176 * If needed, load TOD module now so that ddi_get_time(9F) etc. work 2177 * (For now, "needed" is defined as set tod_module_name in /etc/system) 2178 */ 2179 if (tod_module_name != NULL) { 2180 PRM_POINT("load_tod_module()"); 2181 load_tod_module(tod_module_name); 2182 } 2183 2184 #if defined(__xpv) 2185 /* 2186 * Forceload interposing TOD module for the hypervisor. 2187 */ 2188 PRM_POINT("load_tod_module()"); 2189 load_tod_module("xpvtod"); 2190 #endif 2191 2192 /* 2193 * Configure the system. 2194 */ 2195 PRM_POINT("Calling configure()..."); 2196 configure(); /* set up devices */ 2197 PRM_POINT("configure() done"); 2198 2199 /* 2200 * We can now setup for XSAVE because fpu_probe is done in configure(). 2201 */ 2202 if (fp_save_mech == FP_XSAVE) { 2203 xsave_setup_msr(CPU); 2204 } 2205 2206 /* 2207 * Set the isa_list string to the defined instruction sets we 2208 * support. 2209 */ 2210 setx86isalist(); 2211 cpu_intr_alloc(CPU, NINTR_THREADS); 2212 psm_install(); 2213 2214 /* 2215 * We're done with bootops. We don't unmap the bootstrap yet because 2216 * we're still using bootsvcs. 2217 */ 2218 PRM_POINT("NULLing out bootops"); 2219 *bootopsp = (struct bootops *)NULL; 2220 bootops = (struct bootops *)NULL; 2221 2222 #if defined(__xpv) 2223 ec_init_debug_irq(); 2224 xs_domu_init(); 2225 #endif 2226 2227 #if defined(__amd64) && !defined(__xpv) 2228 /* 2229 * Intel IOMMU has been setup/initialized in ddi_impl.c 2230 * Start it up now. 2231 */ 2232 immu_startup(); 2233 #endif 2234 2235 PRM_POINT("Enabling interrupts"); 2236 (*picinitf)(); 2237 sti(); 2238 #if defined(__xpv) 2239 ASSERT(CPU->cpu_m.mcpu_vcpu_info->evtchn_upcall_mask == 0); 2240 xen_late_startup(); 2241 #endif 2242 2243 (void) add_avsoftintr((void *)&softlevel1_hdl, 1, softlevel1, 2244 "softlevel1", NULL, NULL); /* XXX to be moved later */ 2245 2246 /* 2247 * Register these software interrupts for ddi timer. 2248 * Software interrupts up to the level 10 are supported. 2249 */ 2250 for (i = DDI_IPL_1; i <= DDI_IPL_10; i++) { 2251 char name[sizeof ("timer_softintr") + 2]; 2252 (void) sprintf(name, "timer_softintr%02d", i); 2253 (void) add_avsoftintr((void *)&softlevel_hdl[i-1], i, 2254 (avfunc)timer_softintr, name, (caddr_t)(uintptr_t)i, NULL); 2255 } 2256 2257 #if !defined(__xpv) 2258 if (modload("drv", "amd_iommu") < 0) { 2259 PRM_POINT("No AMD IOMMU present\n"); 2260 } else if (ddi_hold_installed_driver(ddi_name_to_major( 2261 "amd_iommu")) == NULL) { 2262 prom_printf("ERROR: failed to attach AMD IOMMU\n"); 2263 } 2264 #endif 2265 post_startup_cpu_fixups(); 2266 2267 PRM_POINT("startup_end() done"); 2268 } 2269 2270 /* 2271 * Don't remove the following 2 variables. They are necessary 2272 * for reading the hostid from the legacy file (/kernel/misc/sysinit). 2273 */ 2274 char *_hs1107 = hw_serial; 2275 ulong_t _bdhs34; 2276 2277 void 2278 post_startup(void) 2279 { 2280 extern void cpupm_init(cpu_t *); 2281 extern void cpu_event_init_cpu(cpu_t *); 2282 2283 /* 2284 * Set the system wide, processor-specific flags to be passed 2285 * to userland via the aux vector for performance hints and 2286 * instruction set extensions. 2287 */ 2288 bind_hwcap(); 2289 2290 #ifdef __xpv 2291 if (DOMAIN_IS_INITDOMAIN(xen_info)) 2292 #endif 2293 { 2294 #if defined(__xpv) 2295 xpv_panic_init(); 2296 #else 2297 /* 2298 * Startup the memory scrubber. 2299 * XXPV This should be running somewhere .. 2300 */ 2301 if (get_hwenv() != HW_XEN_HVM) 2302 memscrub_init(); 2303 #endif 2304 } 2305 2306 /* 2307 * Complete CPU module initialization 2308 */ 2309 cmi_post_startup(); 2310 2311 /* 2312 * Perform forceloading tasks for /etc/system. 2313 */ 2314 (void) mod_sysctl(SYS_FORCELOAD, NULL); 2315 2316 /* 2317 * ON4.0: Force /proc module in until clock interrupt handle fixed 2318 * ON4.0: This must be fixed or restated in /etc/systems. 2319 */ 2320 (void) modload("fs", "procfs"); 2321 2322 (void) i_ddi_attach_hw_nodes("pit_beep"); 2323 2324 #if defined(__i386) 2325 /* 2326 * Check for required functional Floating Point hardware, 2327 * unless FP hardware explicitly disabled. 2328 */ 2329 if (fpu_exists && (fpu_pentium_fdivbug || fp_kind == FP_NO)) 2330 halt("No working FP hardware found"); 2331 #endif 2332 2333 maxmem = freemem; 2334 2335 cpu_event_init_cpu(CPU); 2336 cpupm_init(CPU); 2337 (void) mach_cpu_create_device_node(CPU, NULL); 2338 2339 pg_init(); 2340 } 2341 2342 static int 2343 pp_in_range(page_t *pp, uint64_t low_addr, uint64_t high_addr) 2344 { 2345 return ((pp->p_pagenum >= btop(low_addr)) && 2346 (pp->p_pagenum < btopr(high_addr))); 2347 } 2348 2349 void 2350 release_bootstrap(void) 2351 { 2352 int root_is_ramdisk; 2353 page_t *pp; 2354 extern void kobj_boot_unmountroot(void); 2355 extern dev_t rootdev; 2356 #if !defined(__xpv) 2357 pfn_t pfn; 2358 #endif 2359 2360 /* unmount boot ramdisk and release kmem usage */ 2361 kobj_boot_unmountroot(); 2362 2363 /* 2364 * We're finished using the boot loader so free its pages. 2365 */ 2366 PRM_POINT("Unmapping lower boot pages"); 2367 2368 clear_boot_mappings(0, _userlimit); 2369 2370 postbootkernelbase = kernelbase; 2371 2372 /* 2373 * If root isn't on ramdisk, destroy the hardcoded 2374 * ramdisk node now and release the memory. Else, 2375 * ramdisk memory is kept in rd_pages. 2376 */ 2377 root_is_ramdisk = (getmajor(rootdev) == ddi_name_to_major("ramdisk")); 2378 if (!root_is_ramdisk) { 2379 dev_info_t *dip = ddi_find_devinfo("ramdisk", -1, 0); 2380 ASSERT(dip && ddi_get_parent(dip) == ddi_root_node()); 2381 ndi_rele_devi(dip); /* held from ddi_find_devinfo */ 2382 (void) ddi_remove_child(dip, 0); 2383 } 2384 2385 PRM_POINT("Releasing boot pages"); 2386 while (bootpages) { 2387 extern uint64_t ramdisk_start, ramdisk_end; 2388 pp = bootpages; 2389 bootpages = pp->p_next; 2390 2391 2392 /* Keep pages for the lower 64K */ 2393 if (pp_in_range(pp, 0, 0x40000)) { 2394 pp->p_next = lower_pages; 2395 lower_pages = pp; 2396 lower_pages_count++; 2397 continue; 2398 } 2399 2400 2401 if (root_is_ramdisk && pp_in_range(pp, ramdisk_start, 2402 ramdisk_end)) { 2403 pp->p_next = rd_pages; 2404 rd_pages = pp; 2405 continue; 2406 } 2407 pp->p_next = (struct page *)0; 2408 pp->p_prev = (struct page *)0; 2409 PP_CLRBOOTPAGES(pp); 2410 page_free(pp, 1); 2411 } 2412 PRM_POINT("Boot pages released"); 2413 2414 #if !defined(__xpv) 2415 /* XXPV -- note this following bunch of code needs to be revisited in Xen 3.0 */ 2416 /* 2417 * Find 1 page below 1 MB so that other processors can boot up or 2418 * so that any processor can resume. 2419 * Make sure it has a kernel VA as well as a 1:1 mapping. 2420 * We should have just free'd one up. 2421 */ 2422 2423 /* 2424 * 0x10 pages is 64K. Leave the bottom 64K alone 2425 * for BIOS. 2426 */ 2427 for (pfn = 0x10; pfn < btop(1*1024*1024); pfn++) { 2428 if (page_numtopp_alloc(pfn) == NULL) 2429 continue; 2430 rm_platter_va = i86devmap(pfn, 1, 2431 PROT_READ | PROT_WRITE | PROT_EXEC); 2432 rm_platter_pa = ptob(pfn); 2433 break; 2434 } 2435 if (pfn == btop(1*1024*1024) && use_mp) 2436 panic("No page below 1M available for starting " 2437 "other processors or for resuming from system-suspend"); 2438 #endif /* !__xpv */ 2439 } 2440 2441 /* 2442 * Initialize the platform-specific parts of a page_t. 2443 */ 2444 void 2445 add_physmem_cb(page_t *pp, pfn_t pnum) 2446 { 2447 pp->p_pagenum = pnum; 2448 pp->p_mapping = NULL; 2449 pp->p_embed = 0; 2450 pp->p_share = 0; 2451 pp->p_mlentry = 0; 2452 } 2453 2454 /* 2455 * kphysm_init() initializes physical memory. 2456 */ 2457 static pgcnt_t 2458 kphysm_init( 2459 page_t *pp, 2460 pgcnt_t npages) 2461 { 2462 struct memlist *pmem; 2463 struct memseg *cur_memseg; 2464 pfn_t base_pfn; 2465 pfn_t end_pfn; 2466 pgcnt_t num; 2467 pgcnt_t pages_done = 0; 2468 uint64_t addr; 2469 uint64_t size; 2470 extern pfn_t ddiphysmin; 2471 extern int mnode_xwa; 2472 int ms = 0, me = 0; 2473 2474 ASSERT(page_hash != NULL && page_hashsz != 0); 2475 2476 cur_memseg = memseg_base; 2477 for (pmem = phys_avail; pmem && npages; pmem = pmem->ml_next) { 2478 /* 2479 * In a 32 bit kernel can't use higher memory if we're 2480 * not booting in PAE mode. This check takes care of that. 2481 */ 2482 addr = pmem->ml_address; 2483 size = pmem->ml_size; 2484 if (btop(addr) > physmax) 2485 continue; 2486 2487 /* 2488 * align addr and size - they may not be at page boundaries 2489 */ 2490 if ((addr & MMU_PAGEOFFSET) != 0) { 2491 addr += MMU_PAGEOFFSET; 2492 addr &= ~(uint64_t)MMU_PAGEOFFSET; 2493 size -= addr - pmem->ml_address; 2494 } 2495 2496 /* only process pages below or equal to physmax */ 2497 if ((btop(addr + size) - 1) > physmax) 2498 size = ptob(physmax - btop(addr) + 1); 2499 2500 num = btop(size); 2501 if (num == 0) 2502 continue; 2503 2504 if (num > npages) 2505 num = npages; 2506 2507 npages -= num; 2508 pages_done += num; 2509 base_pfn = btop(addr); 2510 2511 if (prom_debug) 2512 prom_printf("MEMSEG addr=0x%" PRIx64 2513 " pgs=0x%lx pfn 0x%lx-0x%lx\n", 2514 addr, num, base_pfn, base_pfn + num); 2515 2516 /* 2517 * Ignore pages below ddiphysmin to simplify ddi memory 2518 * allocation with non-zero addr_lo requests. 2519 */ 2520 if (base_pfn < ddiphysmin) { 2521 if (base_pfn + num <= ddiphysmin) 2522 continue; 2523 pp += (ddiphysmin - base_pfn); 2524 num -= (ddiphysmin - base_pfn); 2525 base_pfn = ddiphysmin; 2526 } 2527 2528 /* 2529 * mnode_xwa is greater than 1 when large pages regions can 2530 * cross memory node boundaries. To prevent the formation 2531 * of these large pages, configure the memsegs based on the 2532 * memory node ranges which had been made non-contiguous. 2533 */ 2534 if (mnode_xwa > 1) { 2535 2536 end_pfn = base_pfn + num - 1; 2537 ms = PFN_2_MEM_NODE(base_pfn); 2538 me = PFN_2_MEM_NODE(end_pfn); 2539 2540 if (ms != me) { 2541 /* 2542 * current range spans more than 1 memory node. 2543 * Set num to only the pfn range in the start 2544 * memory node. 2545 */ 2546 num = mem_node_config[ms].physmax - base_pfn 2547 + 1; 2548 ASSERT(end_pfn > mem_node_config[ms].physmax); 2549 } 2550 } 2551 2552 for (;;) { 2553 /* 2554 * Build the memsegs entry 2555 */ 2556 cur_memseg->pages = pp; 2557 cur_memseg->epages = pp + num; 2558 cur_memseg->pages_base = base_pfn; 2559 cur_memseg->pages_end = base_pfn + num; 2560 2561 /* 2562 * Insert into memseg list in decreasing pfn range 2563 * order. Low memory is typically more fragmented such 2564 * that this ordering keeps the larger ranges at the 2565 * front of the list for code that searches memseg. 2566 * This ASSERTS that the memsegs coming in from boot 2567 * are in increasing physical address order and not 2568 * contiguous. 2569 */ 2570 if (memsegs != NULL) { 2571 ASSERT(cur_memseg->pages_base >= 2572 memsegs->pages_end); 2573 cur_memseg->next = memsegs; 2574 } 2575 memsegs = cur_memseg; 2576 2577 /* 2578 * add_physmem() initializes the PSM part of the page 2579 * struct by calling the PSM back with add_physmem_cb(). 2580 * In addition it coalesces pages into larger pages as 2581 * it initializes them. 2582 */ 2583 add_physmem(pp, num, base_pfn); 2584 cur_memseg++; 2585 availrmem_initial += num; 2586 availrmem += num; 2587 2588 pp += num; 2589 if (ms >= me) 2590 break; 2591 2592 /* process next memory node range */ 2593 ms++; 2594 base_pfn = mem_node_config[ms].physbase; 2595 num = MIN(mem_node_config[ms].physmax, 2596 end_pfn) - base_pfn + 1; 2597 } 2598 } 2599 2600 PRM_DEBUG(availrmem_initial); 2601 PRM_DEBUG(availrmem); 2602 PRM_DEBUG(freemem); 2603 build_pfn_hash(); 2604 return (pages_done); 2605 } 2606 2607 /* 2608 * Kernel VM initialization. 2609 */ 2610 static void 2611 kvm_init(void) 2612 { 2613 ASSERT((((uintptr_t)s_text) & MMU_PAGEOFFSET) == 0); 2614 2615 /* 2616 * Put the kernel segments in kernel address space. 2617 */ 2618 rw_enter(&kas.a_lock, RW_WRITER); 2619 as_avlinit(&kas); 2620 2621 (void) seg_attach(&kas, s_text, e_moddata - s_text, &ktextseg); 2622 (void) segkmem_create(&ktextseg); 2623 2624 (void) seg_attach(&kas, (caddr_t)valloc_base, valloc_sz, &kvalloc); 2625 (void) segkmem_create(&kvalloc); 2626 2627 (void) seg_attach(&kas, kernelheap, 2628 ekernelheap - kernelheap, &kvseg); 2629 (void) segkmem_create(&kvseg); 2630 2631 if (core_size > 0) { 2632 PRM_POINT("attaching kvseg_core"); 2633 (void) seg_attach(&kas, (caddr_t)core_base, core_size, 2634 &kvseg_core); 2635 (void) segkmem_create(&kvseg_core); 2636 } 2637 2638 if (segziosize > 0) { 2639 PRM_POINT("attaching segzio"); 2640 (void) seg_attach(&kas, segzio_base, mmu_ptob(segziosize), 2641 &kzioseg); 2642 (void) segkmem_zio_create(&kzioseg); 2643 2644 /* create zio area covering new segment */ 2645 segkmem_zio_init(segzio_base, mmu_ptob(segziosize)); 2646 } 2647 2648 (void) seg_attach(&kas, kdi_segdebugbase, kdi_segdebugsize, &kdebugseg); 2649 (void) segkmem_create(&kdebugseg); 2650 2651 rw_exit(&kas.a_lock); 2652 2653 /* 2654 * Ensure that the red zone at kernelbase is never accessible. 2655 */ 2656 PRM_POINT("protecting redzone"); 2657 (void) as_setprot(&kas, (caddr_t)kernelbase, KERNEL_REDZONE_SIZE, 0); 2658 2659 /* 2660 * Make the text writable so that it can be hot patched by DTrace. 2661 */ 2662 (void) as_setprot(&kas, s_text, e_modtext - s_text, 2663 PROT_READ | PROT_WRITE | PROT_EXEC); 2664 2665 /* 2666 * Make data writable until end. 2667 */ 2668 (void) as_setprot(&kas, s_data, e_moddata - s_data, 2669 PROT_READ | PROT_WRITE | PROT_EXEC); 2670 } 2671 2672 #ifndef __xpv 2673 /* 2674 * Solaris adds an entry for Write Combining caching to the PAT 2675 */ 2676 static uint64_t pat_attr_reg = PAT_DEFAULT_ATTRIBUTE; 2677 2678 void 2679 pat_sync(void) 2680 { 2681 ulong_t cr0, cr0_orig, cr4; 2682 2683 if (!is_x86_feature(x86_featureset, X86FSET_PAT)) 2684 return; 2685 cr0_orig = cr0 = getcr0(); 2686 cr4 = getcr4(); 2687 2688 /* disable caching and flush all caches and TLBs */ 2689 cr0 |= CR0_CD; 2690 cr0 &= ~CR0_NW; 2691 setcr0(cr0); 2692 invalidate_cache(); 2693 if (cr4 & CR4_PGE) { 2694 setcr4(cr4 & ~(ulong_t)CR4_PGE); 2695 setcr4(cr4); 2696 } else { 2697 reload_cr3(); 2698 } 2699 2700 /* add our entry to the PAT */ 2701 wrmsr(REG_PAT, pat_attr_reg); 2702 2703 /* flush TLBs and cache again, then reenable cr0 caching */ 2704 if (cr4 & CR4_PGE) { 2705 setcr4(cr4 & ~(ulong_t)CR4_PGE); 2706 setcr4(cr4); 2707 } else { 2708 reload_cr3(); 2709 } 2710 invalidate_cache(); 2711 setcr0(cr0_orig); 2712 } 2713 2714 #endif /* !__xpv */ 2715 2716 #if defined(_SOFT_HOSTID) 2717 /* 2718 * On platforms that do not have a hardware serial number, attempt 2719 * to set one based on the contents of /etc/hostid. If this file does 2720 * not exist, assume that we are to generate a new hostid and set 2721 * it in the kernel, for subsequent saving by a userland process 2722 * once the system is up and the root filesystem is mounted r/w. 2723 * 2724 * In order to gracefully support upgrade on OpenSolaris, if 2725 * /etc/hostid does not exist, we will attempt to get a serial number 2726 * using the legacy method (/kernel/misc/sysinit). 2727 * 2728 * If that isn't present, we attempt to use an SMBIOS UUID, which is 2729 * a hardware serial number. Note that we don't automatically trust 2730 * all SMBIOS UUIDs (some older platforms are defective and ship duplicate 2731 * UUIDs in violation of the standard), we check against a blacklist. 2732 * 2733 * In an attempt to make the hostid less prone to abuse 2734 * (for license circumvention, etc), we store it in /etc/hostid 2735 * in rot47 format. 2736 */ 2737 extern volatile unsigned long tenmicrodata; 2738 static int atoi(char *); 2739 2740 /* 2741 * Set this to non-zero in /etc/system if you think your SMBIOS returns a 2742 * UUID that is not unique. (Also report it so that the smbios_uuid_blacklist 2743 * array can be updated.) 2744 */ 2745 int smbios_broken_uuid = 0; 2746 2747 /* 2748 * List of known bad UUIDs. This is just the lower 32-bit values, since 2749 * that's what we use for the host id. If your hostid falls here, you need 2750 * to contact your hardware OEM for a fix for your BIOS. 2751 */ 2752 static unsigned char 2753 smbios_uuid_blacklist[][16] = { 2754 2755 { /* Reported bad UUID (Google search) */ 2756 0x00, 0x02, 0x00, 0x03, 0x00, 0x04, 0x00, 0x05, 2757 0x00, 0x06, 0x00, 0x07, 0x00, 0x08, 0x00, 0x09, 2758 }, 2759 { /* Known bad DELL UUID */ 2760 0x4C, 0x4C, 0x45, 0x44, 0x00, 0x00, 0x20, 0x10, 2761 0x80, 0x20, 0x80, 0xC0, 0x4F, 0x20, 0x20, 0x20, 2762 }, 2763 { /* Uninitialized flash */ 2764 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 2765 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff 2766 }, 2767 { /* All zeros */ 2768 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 2769 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00 2770 }, 2771 }; 2772 2773 static int32_t 2774 uuid_to_hostid(const uint8_t *uuid) 2775 { 2776 /* 2777 * Although the UUIDs are 128-bits, they may not distribute entropy 2778 * evenly. We would like to use SHA or MD5, but those are located 2779 * in loadable modules and not available this early in boot. As we 2780 * don't need the values to be cryptographically strong, we just 2781 * generate 32-bit vaue by xor'ing the various sequences together, 2782 * which ensures that the enire UUID contributes to the hostid. 2783 */ 2784 int32_t id = 0; 2785 2786 /* first check against the blacklist */ 2787 for (int i = 0; i < (sizeof (smbios_uuid_blacklist) / 16); i++) { 2788 if (bcmp(smbios_uuid_blacklist[0], uuid, 16) == 0) { 2789 cmn_err(CE_CONT, "?Broken SMBIOS UUID. " 2790 "Contact BIOS manufacturer for repair.\n"); 2791 return ((int32_t)HW_INVALID_HOSTID); 2792 } 2793 } 2794 2795 for (int i = 0; i < 16; i++) 2796 id ^= ((uuid[i]) << (8 * (i % sizeof (id)))); 2797 2798 return (id); 2799 } 2800 2801 static int32_t 2802 set_soft_hostid(void) 2803 { 2804 struct _buf *file; 2805 char tokbuf[MAXNAMELEN]; 2806 token_t token; 2807 int done = 0; 2808 u_longlong_t tmp; 2809 int i; 2810 int32_t hostid = (int32_t)HW_INVALID_HOSTID; 2811 unsigned char *c; 2812 hrtime_t tsc; 2813 smbios_system_t smsys; 2814 2815 /* 2816 * If /etc/hostid file not found, we'd like to get a pseudo 2817 * random number to use at the hostid. A nice way to do this 2818 * is to read the real time clock. To remain xen-compatible, 2819 * we can't poke the real hardware, so we use tsc_read() to 2820 * read the real time clock. However, there is an ominous 2821 * warning in tsc_read that says it can return zero, so we 2822 * deal with that possibility by falling back to using the 2823 * (hopefully random enough) value in tenmicrodata. 2824 */ 2825 2826 if ((file = kobj_open_file(hostid_file)) == (struct _buf *)-1) { 2827 /* 2828 * hostid file not found - try to load sysinit module 2829 * and see if it has a nonzero hostid value...use that 2830 * instead of generating a new hostid here if so. 2831 */ 2832 if ((i = modload("misc", "sysinit")) != -1) { 2833 if (strlen(hw_serial) > 0) 2834 hostid = (int32_t)atoi(hw_serial); 2835 (void) modunload(i); 2836 } 2837 2838 /* 2839 * We try to use the SMBIOS UUID. But not if it is blacklisted 2840 * in /etc/system. 2841 */ 2842 if ((hostid == HW_INVALID_HOSTID) && 2843 (smbios_broken_uuid == 0) && 2844 (ksmbios != NULL) && 2845 (smbios_info_system(ksmbios, &smsys) != SMB_ERR) && 2846 (smsys.smbs_uuidlen >= 16)) { 2847 hostid = uuid_to_hostid(smsys.smbs_uuid); 2848 } 2849 2850 /* 2851 * Generate a "random" hostid using the clock. These 2852 * hostids will change on each boot if the value is not 2853 * saved to a persistent /etc/hostid file. 2854 */ 2855 if (hostid == HW_INVALID_HOSTID) { 2856 tsc = tsc_read(); 2857 if (tsc == 0) /* tsc_read can return zero sometimes */ 2858 hostid = (int32_t)tenmicrodata & 0x0CFFFFF; 2859 else 2860 hostid = (int32_t)tsc & 0x0CFFFFF; 2861 } 2862 } else { 2863 /* hostid file found */ 2864 while (!done) { 2865 token = kobj_lex(file, tokbuf, sizeof (tokbuf)); 2866 2867 switch (token) { 2868 case POUND: 2869 /* 2870 * skip comments 2871 */ 2872 kobj_find_eol(file); 2873 break; 2874 case STRING: 2875 /* 2876 * un-rot47 - obviously this 2877 * nonsense is ascii-specific 2878 */ 2879 for (c = (unsigned char *)tokbuf; 2880 *c != '\0'; c++) { 2881 *c += 47; 2882 if (*c > '~') 2883 *c -= 94; 2884 else if (*c < '!') 2885 *c += 94; 2886 } 2887 /* 2888 * now we should have a real number 2889 */ 2890 2891 if (kobj_getvalue(tokbuf, &tmp) != 0) 2892 kobj_file_err(CE_WARN, file, 2893 "Bad value %s for hostid", 2894 tokbuf); 2895 else 2896 hostid = (int32_t)tmp; 2897 2898 break; 2899 case EOF: 2900 done = 1; 2901 /* FALLTHROUGH */ 2902 case NEWLINE: 2903 kobj_newline(file); 2904 break; 2905 default: 2906 break; 2907 2908 } 2909 } 2910 if (hostid == HW_INVALID_HOSTID) /* didn't find a hostid */ 2911 kobj_file_err(CE_WARN, file, 2912 "hostid missing or corrupt"); 2913 2914 kobj_close_file(file); 2915 } 2916 /* 2917 * hostid is now the value read from /etc/hostid, or the 2918 * new hostid we generated in this routine or HW_INVALID_HOSTID if not 2919 * set. 2920 */ 2921 return (hostid); 2922 } 2923 2924 static int 2925 atoi(char *p) 2926 { 2927 int i = 0; 2928 2929 while (*p != '\0') 2930 i = 10 * i + (*p++ - '0'); 2931 2932 return (i); 2933 } 2934 2935 #endif /* _SOFT_HOSTID */ 2936 2937 void 2938 get_system_configuration(void) 2939 { 2940 char prop[32]; 2941 u_longlong_t nodes_ll, cpus_pernode_ll, lvalue; 2942 2943 if (BOP_GETPROPLEN(bootops, "nodes") > sizeof (prop) || 2944 BOP_GETPROP(bootops, "nodes", prop) < 0 || 2945 kobj_getvalue(prop, &nodes_ll) == -1 || 2946 nodes_ll > MAXNODES || 2947 BOP_GETPROPLEN(bootops, "cpus_pernode") > sizeof (prop) || 2948 BOP_GETPROP(bootops, "cpus_pernode", prop) < 0 || 2949 kobj_getvalue(prop, &cpus_pernode_ll) == -1) { 2950 system_hardware.hd_nodes = 1; 2951 system_hardware.hd_cpus_per_node = 0; 2952 } else { 2953 system_hardware.hd_nodes = (int)nodes_ll; 2954 system_hardware.hd_cpus_per_node = (int)cpus_pernode_ll; 2955 } 2956 2957 if (BOP_GETPROPLEN(bootops, "kernelbase") > sizeof (prop) || 2958 BOP_GETPROP(bootops, "kernelbase", prop) < 0 || 2959 kobj_getvalue(prop, &lvalue) == -1) 2960 eprom_kernelbase = NULL; 2961 else 2962 eprom_kernelbase = (uintptr_t)lvalue; 2963 2964 if (BOP_GETPROPLEN(bootops, "segmapsize") > sizeof (prop) || 2965 BOP_GETPROP(bootops, "segmapsize", prop) < 0 || 2966 kobj_getvalue(prop, &lvalue) == -1) 2967 segmapsize = SEGMAPDEFAULT; 2968 else 2969 segmapsize = (uintptr_t)lvalue; 2970 2971 if (BOP_GETPROPLEN(bootops, "segmapfreelists") > sizeof (prop) || 2972 BOP_GETPROP(bootops, "segmapfreelists", prop) < 0 || 2973 kobj_getvalue(prop, &lvalue) == -1) 2974 segmapfreelists = 0; /* use segmap driver default */ 2975 else 2976 segmapfreelists = (int)lvalue; 2977 2978 /* physmem used to be here, but moved much earlier to fakebop.c */ 2979 } 2980 2981 /* 2982 * Add to a memory list. 2983 * start = start of new memory segment 2984 * len = length of new memory segment in bytes 2985 * new = pointer to a new struct memlist 2986 * memlistp = memory list to which to add segment. 2987 */ 2988 void 2989 memlist_add( 2990 uint64_t start, 2991 uint64_t len, 2992 struct memlist *new, 2993 struct memlist **memlistp) 2994 { 2995 struct memlist *cur; 2996 uint64_t end = start + len; 2997 2998 new->ml_address = start; 2999 new->ml_size = len; 3000 3001 cur = *memlistp; 3002 3003 while (cur) { 3004 if (cur->ml_address >= end) { 3005 new->ml_next = cur; 3006 *memlistp = new; 3007 new->ml_prev = cur->ml_prev; 3008 cur->ml_prev = new; 3009 return; 3010 } 3011 ASSERT(cur->ml_address + cur->ml_size <= start); 3012 if (cur->ml_next == NULL) { 3013 cur->ml_next = new; 3014 new->ml_prev = cur; 3015 new->ml_next = NULL; 3016 return; 3017 } 3018 memlistp = &cur->ml_next; 3019 cur = cur->ml_next; 3020 } 3021 } 3022 3023 void 3024 kobj_vmem_init(vmem_t **text_arena, vmem_t **data_arena) 3025 { 3026 size_t tsize = e_modtext - modtext; 3027 size_t dsize = e_moddata - moddata; 3028 3029 *text_arena = vmem_create("module_text", tsize ? modtext : NULL, tsize, 3030 1, segkmem_alloc, segkmem_free, heaptext_arena, 0, VM_SLEEP); 3031 *data_arena = vmem_create("module_data", dsize ? moddata : NULL, dsize, 3032 1, segkmem_alloc, segkmem_free, heap32_arena, 0, VM_SLEEP); 3033 } 3034 3035 caddr_t 3036 kobj_text_alloc(vmem_t *arena, size_t size) 3037 { 3038 return (vmem_alloc(arena, size, VM_SLEEP | VM_BESTFIT)); 3039 } 3040 3041 /*ARGSUSED*/ 3042 caddr_t 3043 kobj_texthole_alloc(caddr_t addr, size_t size) 3044 { 3045 panic("unexpected call to kobj_texthole_alloc()"); 3046 /*NOTREACHED*/ 3047 return (0); 3048 } 3049 3050 /*ARGSUSED*/ 3051 void 3052 kobj_texthole_free(caddr_t addr, size_t size) 3053 { 3054 panic("unexpected call to kobj_texthole_free()"); 3055 } 3056 3057 /* 3058 * This is called just after configure() in startup(). 3059 * 3060 * The ISALIST concept is a bit hopeless on Intel, because 3061 * there's no guarantee of an ever-more-capable processor 3062 * given that various parts of the instruction set may appear 3063 * and disappear between different implementations. 3064 * 3065 * While it would be possible to correct it and even enhance 3066 * it somewhat, the explicit hardware capability bitmask allows 3067 * more flexibility. 3068 * 3069 * So, we just leave this alone. 3070 */ 3071 void 3072 setx86isalist(void) 3073 { 3074 char *tp; 3075 size_t len; 3076 extern char *isa_list; 3077 3078 #define TBUFSIZE 1024 3079 3080 tp = kmem_alloc(TBUFSIZE, KM_SLEEP); 3081 *tp = '\0'; 3082 3083 #if defined(__amd64) 3084 (void) strcpy(tp, "amd64 "); 3085 #endif 3086 3087 switch (x86_vendor) { 3088 case X86_VENDOR_Intel: 3089 case X86_VENDOR_AMD: 3090 case X86_VENDOR_TM: 3091 if (is_x86_feature(x86_featureset, X86FSET_CMOV)) { 3092 /* 3093 * Pentium Pro or later 3094 */ 3095 (void) strcat(tp, "pentium_pro"); 3096 (void) strcat(tp, 3097 is_x86_feature(x86_featureset, X86FSET_MMX) ? 3098 "+mmx pentium_pro " : " "); 3099 } 3100 /*FALLTHROUGH*/ 3101 case X86_VENDOR_Cyrix: 3102 /* 3103 * The Cyrix 6x86 does not have any Pentium features 3104 * accessible while not at privilege level 0. 3105 */ 3106 if (is_x86_feature(x86_featureset, X86FSET_CPUID)) { 3107 (void) strcat(tp, "pentium"); 3108 (void) strcat(tp, 3109 is_x86_feature(x86_featureset, X86FSET_MMX) ? 3110 "+mmx pentium " : " "); 3111 } 3112 break; 3113 default: 3114 break; 3115 } 3116 (void) strcat(tp, "i486 i386 i86"); 3117 len = strlen(tp) + 1; /* account for NULL at end of string */ 3118 isa_list = strcpy(kmem_alloc(len, KM_SLEEP), tp); 3119 kmem_free(tp, TBUFSIZE); 3120 3121 #undef TBUFSIZE 3122 } 3123 3124 3125 #ifdef __amd64 3126 3127 void * 3128 device_arena_alloc(size_t size, int vm_flag) 3129 { 3130 return (vmem_alloc(device_arena, size, vm_flag)); 3131 } 3132 3133 void 3134 device_arena_free(void *vaddr, size_t size) 3135 { 3136 vmem_free(device_arena, vaddr, size); 3137 } 3138 3139 #else /* __i386 */ 3140 3141 void * 3142 device_arena_alloc(size_t size, int vm_flag) 3143 { 3144 caddr_t vaddr; 3145 uintptr_t v; 3146 size_t start; 3147 size_t end; 3148 3149 vaddr = vmem_alloc(heap_arena, size, vm_flag); 3150 if (vaddr == NULL) 3151 return (NULL); 3152 3153 v = (uintptr_t)vaddr; 3154 ASSERT(v >= kernelbase); 3155 ASSERT(v + size <= valloc_base); 3156 3157 start = btop(v - kernelbase); 3158 end = btop(v + size - 1 - kernelbase); 3159 ASSERT(start < toxic_bit_map_len); 3160 ASSERT(end < toxic_bit_map_len); 3161 3162 while (start <= end) { 3163 BT_ATOMIC_SET(toxic_bit_map, start); 3164 ++start; 3165 } 3166 return (vaddr); 3167 } 3168 3169 void 3170 device_arena_free(void *vaddr, size_t size) 3171 { 3172 uintptr_t v = (uintptr_t)vaddr; 3173 size_t start; 3174 size_t end; 3175 3176 ASSERT(v >= kernelbase); 3177 ASSERT(v + size <= valloc_base); 3178 3179 start = btop(v - kernelbase); 3180 end = btop(v + size - 1 - kernelbase); 3181 ASSERT(start < toxic_bit_map_len); 3182 ASSERT(end < toxic_bit_map_len); 3183 3184 while (start <= end) { 3185 ASSERT(BT_TEST(toxic_bit_map, start) != 0); 3186 BT_ATOMIC_CLEAR(toxic_bit_map, start); 3187 ++start; 3188 } 3189 vmem_free(heap_arena, vaddr, size); 3190 } 3191 3192 /* 3193 * returns 1st address in range that is in device arena, or NULL 3194 * if len is not NULL it returns the length of the toxic range 3195 */ 3196 void * 3197 device_arena_contains(void *vaddr, size_t size, size_t *len) 3198 { 3199 uintptr_t v = (uintptr_t)vaddr; 3200 uintptr_t eaddr = v + size; 3201 size_t start; 3202 size_t end; 3203 3204 /* 3205 * if called very early by kmdb, just return NULL 3206 */ 3207 if (toxic_bit_map == NULL) 3208 return (NULL); 3209 3210 /* 3211 * First check if we're completely outside the bitmap range. 3212 */ 3213 if (v >= valloc_base || eaddr < kernelbase) 3214 return (NULL); 3215 3216 /* 3217 * Trim ends of search to look at only what the bitmap covers. 3218 */ 3219 if (v < kernelbase) 3220 v = kernelbase; 3221 start = btop(v - kernelbase); 3222 end = btop(eaddr - kernelbase); 3223 if (end >= toxic_bit_map_len) 3224 end = toxic_bit_map_len; 3225 3226 if (bt_range(toxic_bit_map, &start, &end, end) == 0) 3227 return (NULL); 3228 3229 v = kernelbase + ptob(start); 3230 if (len != NULL) 3231 *len = ptob(end - start); 3232 return ((void *)v); 3233 } 3234 3235 #endif /* __i386 */ 3236