1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 /* 22 * Copyright (c) 1993, 2010, Oracle and/or its affiliates. All rights reserved. 23 * Copyright 2012 DEY Storage Systems, Inc. All rights reserved. 24 * Copyright 2013 Nexenta Systems, Inc. All rights reserved. 25 * Copyright 2013 Joyent, Inc. All rights reserved. 26 */ 27 /* 28 * Copyright (c) 2010, Intel Corporation. 29 * All rights reserved. 30 */ 31 32 #include <sys/types.h> 33 #include <sys/t_lock.h> 34 #include <sys/param.h> 35 #include <sys/sysmacros.h> 36 #include <sys/signal.h> 37 #include <sys/systm.h> 38 #include <sys/user.h> 39 #include <sys/mman.h> 40 #include <sys/vm.h> 41 #include <sys/conf.h> 42 #include <sys/avintr.h> 43 #include <sys/autoconf.h> 44 #include <sys/disp.h> 45 #include <sys/class.h> 46 #include <sys/bitmap.h> 47 48 #include <sys/privregs.h> 49 50 #include <sys/proc.h> 51 #include <sys/buf.h> 52 #include <sys/kmem.h> 53 #include <sys/mem.h> 54 #include <sys/kstat.h> 55 56 #include <sys/reboot.h> 57 58 #include <sys/cred.h> 59 #include <sys/vnode.h> 60 #include <sys/file.h> 61 62 #include <sys/procfs.h> 63 64 #include <sys/vfs.h> 65 #include <sys/cmn_err.h> 66 #include <sys/utsname.h> 67 #include <sys/debug.h> 68 #include <sys/kdi.h> 69 70 #include <sys/dumphdr.h> 71 #include <sys/bootconf.h> 72 #include <sys/memlist_plat.h> 73 #include <sys/varargs.h> 74 #include <sys/promif.h> 75 #include <sys/modctl.h> 76 77 #include <sys/sunddi.h> 78 #include <sys/sunndi.h> 79 #include <sys/ndi_impldefs.h> 80 #include <sys/ddidmareq.h> 81 #include <sys/psw.h> 82 #include <sys/regset.h> 83 #include <sys/clock.h> 84 #include <sys/pte.h> 85 #include <sys/tss.h> 86 #include <sys/stack.h> 87 #include <sys/trap.h> 88 #include <sys/fp.h> 89 #include <vm/kboot_mmu.h> 90 #include <vm/anon.h> 91 #include <vm/as.h> 92 #include <vm/page.h> 93 #include <vm/seg.h> 94 #include <vm/seg_dev.h> 95 #include <vm/seg_kmem.h> 96 #include <vm/seg_kpm.h> 97 #include <vm/seg_map.h> 98 #include <vm/seg_vn.h> 99 #include <vm/seg_kp.h> 100 #include <sys/memnode.h> 101 #include <vm/vm_dep.h> 102 #include <sys/thread.h> 103 #include <sys/sysconf.h> 104 #include <sys/vm_machparam.h> 105 #include <sys/archsystm.h> 106 #include <sys/machsystm.h> 107 #include <vm/hat.h> 108 #include <vm/hat_i86.h> 109 #include <sys/pmem.h> 110 #include <sys/smp_impldefs.h> 111 #include <sys/x86_archext.h> 112 #include <sys/cpuvar.h> 113 #include <sys/segments.h> 114 #include <sys/clconf.h> 115 #include <sys/kobj.h> 116 #include <sys/kobj_lex.h> 117 #include <sys/cpc_impl.h> 118 #include <sys/cpu_module.h> 119 #include <sys/smbios.h> 120 #include <sys/debug_info.h> 121 #include <sys/bootinfo.h> 122 #include <sys/ddi_periodic.h> 123 #include <sys/systeminfo.h> 124 #include <sys/multiboot.h> 125 #include <sys/ramdisk.h> 126 127 #ifdef __xpv 128 129 #include <sys/hypervisor.h> 130 #include <sys/xen_mmu.h> 131 #include <sys/evtchn_impl.h> 132 #include <sys/gnttab.h> 133 #include <sys/xpv_panic.h> 134 #include <xen/sys/xenbus_comms.h> 135 #include <xen/public/physdev.h> 136 137 extern void xen_late_startup(void); 138 139 struct xen_evt_data cpu0_evt_data; 140 141 #else /* __xpv */ 142 #include <sys/memlist_impl.h> 143 144 extern void mem_config_init(void); 145 #endif /* __xpv */ 146 147 extern void progressbar_init(void); 148 extern void brand_init(void); 149 extern void pcf_init(void); 150 extern void pg_init(void); 151 extern void ssp_init(void); 152 153 extern int size_pse_array(pgcnt_t, int); 154 155 #if defined(_SOFT_HOSTID) 156 157 #include <sys/rtc.h> 158 159 static int32_t set_soft_hostid(void); 160 static char hostid_file[] = "/etc/hostid"; 161 162 #endif 163 164 void *gfx_devinfo_list; 165 166 #if defined(__amd64) && !defined(__xpv) 167 extern void immu_startup(void); 168 #endif 169 170 /* 171 * XXX make declaration below "static" when drivers no longer use this 172 * interface. 173 */ 174 extern caddr_t p0_va; /* Virtual address for accessing physical page 0 */ 175 176 /* 177 * segkp 178 */ 179 extern int segkp_fromheap; 180 181 static void kvm_init(void); 182 static void startup_init(void); 183 static void startup_memlist(void); 184 static void startup_kmem(void); 185 static void startup_modules(void); 186 static void startup_vm(void); 187 static void startup_end(void); 188 static void layout_kernel_va(void); 189 190 /* 191 * Declare these as initialized data so we can patch them. 192 */ 193 #ifdef __i386 194 195 /* 196 * Due to virtual address space limitations running in 32 bit mode, restrict 197 * the amount of physical memory configured to a max of PHYSMEM pages (16g). 198 * 199 * If the physical max memory size of 64g were allowed to be configured, the 200 * size of user virtual address space will be less than 1g. A limited user 201 * address space greatly reduces the range of applications that can run. 202 * 203 * If more physical memory than PHYSMEM is required, users should preferably 204 * run in 64 bit mode which has far looser virtual address space limitations. 205 * 206 * If 64 bit mode is not available (as in IA32) and/or more physical memory 207 * than PHYSMEM is required in 32 bit mode, physmem can be set to the desired 208 * value or to 0 (to configure all available memory) via eeprom(1M). kernelbase 209 * should also be carefully tuned to balance out the need of the user 210 * application while minimizing the risk of kernel heap exhaustion due to 211 * kernelbase being set too high. 212 */ 213 #define PHYSMEM 0x400000 214 215 #else /* __amd64 */ 216 217 /* 218 * For now we can handle memory with physical addresses up to about 219 * 64 Terabytes. This keeps the kernel above the VA hole, leaving roughly 220 * half the VA space for seg_kpm. When systems get bigger than 64TB this 221 * code will need revisiting. There is an implicit assumption that there 222 * are no *huge* holes in the physical address space too. 223 */ 224 #define TERABYTE (1ul << 40) 225 #define PHYSMEM_MAX64 mmu_btop(64 * TERABYTE) 226 #define PHYSMEM PHYSMEM_MAX64 227 #define AMD64_VA_HOLE_END 0xFFFF800000000000ul 228 229 #endif /* __amd64 */ 230 231 pgcnt_t physmem = PHYSMEM; 232 pgcnt_t obp_pages; /* Memory used by PROM for its text and data */ 233 234 char *kobj_file_buf; 235 int kobj_file_bufsize; /* set in /etc/system */ 236 237 /* Global variables for MP support. Used in mp_startup */ 238 caddr_t rm_platter_va = 0; 239 uint32_t rm_platter_pa; 240 241 int auto_lpg_disable = 1; 242 243 /* 244 * Some CPUs have holes in the middle of the 64-bit virtual address range. 245 */ 246 uintptr_t hole_start, hole_end; 247 248 /* 249 * kpm mapping window 250 */ 251 caddr_t kpm_vbase; 252 size_t kpm_size; 253 static int kpm_desired; 254 #ifdef __amd64 255 static uintptr_t segkpm_base = (uintptr_t)SEGKPM_BASE; 256 #endif 257 258 /* 259 * Configuration parameters set at boot time. 260 */ 261 262 caddr_t econtig; /* end of first block of contiguous kernel */ 263 264 struct bootops *bootops = 0; /* passed in from boot */ 265 struct bootops **bootopsp; 266 struct boot_syscalls *sysp; /* passed in from boot */ 267 268 char bootblock_fstype[16]; 269 270 char kern_bootargs[OBP_MAXPATHLEN]; 271 char kern_bootfile[OBP_MAXPATHLEN]; 272 273 /* 274 * ZFS zio segment. This allows us to exclude large portions of ZFS data that 275 * gets cached in kmem caches on the heap. If this is set to zero, we allocate 276 * zio buffers from their own segment, otherwise they are allocated from the 277 * heap. The optimization of allocating zio buffers from their own segment is 278 * only valid on 64-bit kernels. 279 */ 280 #if defined(__amd64) 281 int segzio_fromheap = 0; 282 #else 283 int segzio_fromheap = 1; 284 #endif 285 286 /* 287 * new memory fragmentations are possible in startup() due to BOP_ALLOCs. this 288 * depends on number of BOP_ALLOC calls made and requested size, memory size 289 * combination and whether boot.bin memory needs to be freed. 290 */ 291 #define POSS_NEW_FRAGMENTS 12 292 293 /* 294 * VM data structures 295 */ 296 long page_hashsz; /* Size of page hash table (power of two) */ 297 unsigned int page_hashsz_shift; /* log2(page_hashsz) */ 298 struct page *pp_base; /* Base of initial system page struct array */ 299 struct page **page_hash; /* Page hash table */ 300 pad_mutex_t *pse_mutex; /* Locks protecting pp->p_selock */ 301 size_t pse_table_size; /* Number of mutexes in pse_mutex[] */ 302 int pse_shift; /* log2(pse_table_size) */ 303 struct seg ktextseg; /* Segment used for kernel executable image */ 304 struct seg kvalloc; /* Segment used for "valloc" mapping */ 305 struct seg kpseg; /* Segment used for pageable kernel virt mem */ 306 struct seg kmapseg; /* Segment used for generic kernel mappings */ 307 struct seg kdebugseg; /* Segment used for the kernel debugger */ 308 309 struct seg *segkmap = &kmapseg; /* Kernel generic mapping segment */ 310 static struct seg *segmap = &kmapseg; /* easier to use name for in here */ 311 312 struct seg *segkp = &kpseg; /* Pageable kernel virtual memory segment */ 313 314 #if defined(__amd64) 315 struct seg kvseg_core; /* Segment used for the core heap */ 316 struct seg kpmseg; /* Segment used for physical mapping */ 317 struct seg *segkpm = &kpmseg; /* 64bit kernel physical mapping segment */ 318 #else 319 struct seg *segkpm = NULL; /* Unused on IA32 */ 320 #endif 321 322 caddr_t segkp_base; /* Base address of segkp */ 323 caddr_t segzio_base; /* Base address of segzio */ 324 #if defined(__amd64) 325 pgcnt_t segkpsize = btop(SEGKPDEFSIZE); /* size of segkp segment in pages */ 326 #else 327 pgcnt_t segkpsize = 0; 328 #endif 329 pgcnt_t segziosize = 0; /* size of zio segment in pages */ 330 331 /* 332 * A static DR page_t VA map is reserved that can map the page structures 333 * for a domain's entire RA space. The pages that back this space are 334 * dynamically allocated and need not be physically contiguous. The DR 335 * map size is derived from KPM size. 336 * This mechanism isn't used by x86 yet, so just stubs here. 337 */ 338 int ppvm_enable = 0; /* Static virtual map for page structs */ 339 page_t *ppvm_base = NULL; /* Base of page struct map */ 340 pgcnt_t ppvm_size = 0; /* Size of page struct map */ 341 342 /* 343 * VA range available to the debugger 344 */ 345 const caddr_t kdi_segdebugbase = (const caddr_t)SEGDEBUGBASE; 346 const size_t kdi_segdebugsize = SEGDEBUGSIZE; 347 348 struct memseg *memseg_base; 349 struct vnode unused_pages_vp; 350 351 #define FOURGB 0x100000000LL 352 353 struct memlist *memlist; 354 355 caddr_t s_text; /* start of kernel text segment */ 356 caddr_t e_text; /* end of kernel text segment */ 357 caddr_t s_data; /* start of kernel data segment */ 358 caddr_t e_data; /* end of kernel data segment */ 359 caddr_t modtext; /* start of loadable module text reserved */ 360 caddr_t e_modtext; /* end of loadable module text reserved */ 361 caddr_t moddata; /* start of loadable module data reserved */ 362 caddr_t e_moddata; /* end of loadable module data reserved */ 363 364 struct memlist *phys_install; /* Total installed physical memory */ 365 struct memlist *phys_avail; /* Total available physical memory */ 366 struct memlist *bios_rsvd; /* Bios reserved memory */ 367 368 /* 369 * kphysm_init returns the number of pages that were processed 370 */ 371 static pgcnt_t kphysm_init(page_t *, pgcnt_t); 372 373 #define IO_PROP_SIZE 64 /* device property size */ 374 375 /* 376 * a couple useful roundup macros 377 */ 378 #define ROUND_UP_PAGE(x) \ 379 ((uintptr_t)P2ROUNDUP((uintptr_t)(x), (uintptr_t)MMU_PAGESIZE)) 380 #define ROUND_UP_LPAGE(x) \ 381 ((uintptr_t)P2ROUNDUP((uintptr_t)(x), mmu.level_size[1])) 382 #define ROUND_UP_4MEG(x) \ 383 ((uintptr_t)P2ROUNDUP((uintptr_t)(x), (uintptr_t)FOUR_MEG)) 384 #define ROUND_UP_TOPLEVEL(x) \ 385 ((uintptr_t)P2ROUNDUP((uintptr_t)(x), mmu.level_size[mmu.max_level])) 386 387 /* 388 * 32-bit Kernel's Virtual memory layout. 389 * +-----------------------+ 390 * | | 391 * 0xFFC00000 -|-----------------------|- ARGSBASE 392 * | debugger | 393 * 0xFF800000 -|-----------------------|- SEGDEBUGBASE 394 * | Kernel Data | 395 * 0xFEC00000 -|-----------------------| 396 * | Kernel Text | 397 * 0xFE800000 -|-----------------------|- KERNEL_TEXT (0xFB400000 on Xen) 398 * |--- GDT ---|- GDT page (GDT_VA) 399 * |--- debug info ---|- debug info (DEBUG_INFO_VA) 400 * | | 401 * | page_t structures | 402 * | memsegs, memlists, | 403 * | page hash, etc. | 404 * --- -|-----------------------|- ekernelheap, valloc_base (floating) 405 * | | (segkp is just an arena in the heap) 406 * | | 407 * | kvseg | 408 * | | 409 * | | 410 * --- -|-----------------------|- kernelheap (floating) 411 * | Segkmap | 412 * 0xC3002000 -|-----------------------|- segmap_start (floating) 413 * | Red Zone | 414 * 0xC3000000 -|-----------------------|- kernelbase / userlimit (floating) 415 * | | || 416 * | Shared objects | \/ 417 * | | 418 * : : 419 * | user data | 420 * |-----------------------| 421 * | user text | 422 * 0x08048000 -|-----------------------| 423 * | user stack | 424 * : : 425 * | invalid | 426 * 0x00000000 +-----------------------+ 427 * 428 * 429 * 64-bit Kernel's Virtual memory layout. (assuming 64 bit app) 430 * +-----------------------+ 431 * | | 432 * 0xFFFFFFFF.FFC00000 |-----------------------|- ARGSBASE 433 * | debugger (?) | 434 * 0xFFFFFFFF.FF800000 |-----------------------|- SEGDEBUGBASE 435 * | unused | 436 * +-----------------------+ 437 * | Kernel Data | 438 * 0xFFFFFFFF.FBC00000 |-----------------------| 439 * | Kernel Text | 440 * 0xFFFFFFFF.FB800000 |-----------------------|- KERNEL_TEXT 441 * |--- GDT ---|- GDT page (GDT_VA) 442 * |--- debug info ---|- debug info (DEBUG_INFO_VA) 443 * | | 444 * | Core heap | (used for loadable modules) 445 * 0xFFFFFFFF.C0000000 |-----------------------|- core_base / ekernelheap 446 * | Kernel | 447 * | heap | 448 * 0xFFFFFXXX.XXX00000 |-----------------------|- kernelheap (floating) 449 * | segmap | 450 * 0xFFFFFXXX.XXX00000 |-----------------------|- segmap_start (floating) 451 * | device mappings | 452 * 0xFFFFFXXX.XXX00000 |-----------------------|- toxic_addr (floating) 453 * | segzio | 454 * 0xFFFFFXXX.XXX00000 |-----------------------|- segzio_base (floating) 455 * | segkp | 456 * --- |-----------------------|- segkp_base (floating) 457 * | page_t structures | valloc_base + valloc_sz 458 * | memsegs, memlists, | 459 * | page hash, etc. | 460 * 0xFFFFFF00.00000000 |-----------------------|- valloc_base (lower if > 1TB) 461 * | segkpm | 462 * 0xFFFFFE00.00000000 |-----------------------| 463 * | Red Zone | 464 * 0xFFFFFD80.00000000 |-----------------------|- KERNELBASE (lower if > 1TB) 465 * | User stack |- User space memory 466 * | | 467 * | shared objects, etc | (grows downwards) 468 * : : 469 * | | 470 * 0xFFFF8000.00000000 |-----------------------| 471 * | | 472 * | VA Hole / unused | 473 * | | 474 * 0x00008000.00000000 |-----------------------| 475 * | | 476 * | | 477 * : : 478 * | user heap | (grows upwards) 479 * | | 480 * | user data | 481 * |-----------------------| 482 * | user text | 483 * 0x00000000.04000000 |-----------------------| 484 * | invalid | 485 * 0x00000000.00000000 +-----------------------+ 486 * 487 * A 32 bit app on the 64 bit kernel sees the same layout as on the 32 bit 488 * kernel, except that userlimit is raised to 0xfe000000 489 * 490 * Floating values: 491 * 492 * valloc_base: start of the kernel's memory management/tracking data 493 * structures. This region contains page_t structures for 494 * physical memory, memsegs, memlists, and the page hash. 495 * 496 * core_base: start of the kernel's "core" heap area on 64-bit systems. 497 * This area is intended to be used for global data as well as for module 498 * text/data that does not fit into the nucleus pages. The core heap is 499 * restricted to a 2GB range, allowing every address within it to be 500 * accessed using rip-relative addressing 501 * 502 * ekernelheap: end of kernelheap and start of segmap. 503 * 504 * kernelheap: start of kernel heap. On 32-bit systems, this starts right 505 * above a red zone that separates the user's address space from the 506 * kernel's. On 64-bit systems, it sits above segkp and segkpm. 507 * 508 * segmap_start: start of segmap. The length of segmap can be modified 509 * through eeprom. The default length is 16MB on 32-bit systems and 64MB 510 * on 64-bit systems. 511 * 512 * kernelbase: On a 32-bit kernel the default value of 0xd4000000 will be 513 * decreased by 2X the size required for page_t. This allows the kernel 514 * heap to grow in size with physical memory. With sizeof(page_t) == 80 515 * bytes, the following shows the values of kernelbase and kernel heap 516 * sizes for different memory configurations (assuming default segmap and 517 * segkp sizes). 518 * 519 * mem size for kernelbase kernel heap 520 * size page_t's size 521 * ---- --------- ---------- ----------- 522 * 1gb 0x01400000 0xd1800000 684MB 523 * 2gb 0x02800000 0xcf000000 704MB 524 * 4gb 0x05000000 0xca000000 744MB 525 * 6gb 0x07800000 0xc5000000 784MB 526 * 8gb 0x0a000000 0xc0000000 824MB 527 * 16gb 0x14000000 0xac000000 984MB 528 * 32gb 0x28000000 0x84000000 1304MB 529 * 64gb 0x50000000 0x34000000 1944MB (*) 530 * 531 * kernelbase is less than the abi minimum of 0xc0000000 for memory 532 * configurations above 8gb. 533 * 534 * (*) support for memory configurations above 32gb will require manual tuning 535 * of kernelbase to balance out the need of user applications. 536 */ 537 538 /* real-time-clock initialization parameters */ 539 extern time_t process_rtc_config_file(void); 540 541 uintptr_t kernelbase; 542 uintptr_t postbootkernelbase; /* not set till boot loader is gone */ 543 uintptr_t eprom_kernelbase; 544 size_t segmapsize; 545 uintptr_t segmap_start; 546 int segmapfreelists; 547 pgcnt_t npages; 548 pgcnt_t orig_npages; 549 size_t core_size; /* size of "core" heap */ 550 uintptr_t core_base; /* base address of "core" heap */ 551 552 /* 553 * List of bootstrap pages. We mark these as allocated in startup. 554 * release_bootstrap() will free them when we're completely done with 555 * the bootstrap. 556 */ 557 static page_t *bootpages; 558 559 /* 560 * boot time pages that have a vnode from the ramdisk will keep that forever. 561 */ 562 static page_t *rd_pages; 563 564 /* 565 * Lower 64K 566 */ 567 static page_t *lower_pages = NULL; 568 static int lower_pages_count = 0; 569 570 struct system_hardware system_hardware; 571 572 /* 573 * Enable some debugging messages concerning memory usage... 574 */ 575 static void 576 print_memlist(char *title, struct memlist *mp) 577 { 578 prom_printf("MEMLIST: %s:\n", title); 579 while (mp != NULL) { 580 prom_printf("\tAddress 0x%" PRIx64 ", size 0x%" PRIx64 "\n", 581 mp->ml_address, mp->ml_size); 582 mp = mp->ml_next; 583 } 584 } 585 586 /* 587 * XX64 need a comment here.. are these just default values, surely 588 * we read the "cpuid" type information to figure this out. 589 */ 590 int l2cache_sz = 0x80000; 591 int l2cache_linesz = 0x40; 592 int l2cache_assoc = 1; 593 594 static size_t textrepl_min_gb = 10; 595 596 /* 597 * on 64 bit we use a predifined VA range for mapping devices in the kernel 598 * on 32 bit the mappings are intermixed in the heap, so we use a bit map 599 */ 600 #ifdef __amd64 601 602 vmem_t *device_arena; 603 uintptr_t toxic_addr = (uintptr_t)NULL; 604 size_t toxic_size = 1024 * 1024 * 1024; /* Sparc uses 1 gig too */ 605 606 #else /* __i386 */ 607 608 ulong_t *toxic_bit_map; /* one bit for each 4k of VA in heap_arena */ 609 size_t toxic_bit_map_len = 0; /* in bits */ 610 611 #endif /* __i386 */ 612 613 /* 614 * Simple boot time debug facilities 615 */ 616 static char *prm_dbg_str[] = { 617 "%s:%d: '%s' is 0x%x\n", 618 "%s:%d: '%s' is 0x%llx\n" 619 }; 620 621 int prom_debug; 622 623 #define PRM_DEBUG(q) if (prom_debug) \ 624 prom_printf(prm_dbg_str[sizeof (q) >> 3], "startup.c", __LINE__, #q, q); 625 #define PRM_POINT(q) if (prom_debug) \ 626 prom_printf("%s:%d: %s\n", "startup.c", __LINE__, q); 627 628 /* 629 * This structure is used to keep track of the intial allocations 630 * done in startup_memlist(). The value of NUM_ALLOCATIONS needs to 631 * be >= the number of ADD_TO_ALLOCATIONS() executed in the code. 632 */ 633 #define NUM_ALLOCATIONS 8 634 int num_allocations = 0; 635 struct { 636 void **al_ptr; 637 size_t al_size; 638 } allocations[NUM_ALLOCATIONS]; 639 size_t valloc_sz = 0; 640 uintptr_t valloc_base; 641 642 #define ADD_TO_ALLOCATIONS(ptr, size) { \ 643 size = ROUND_UP_PAGE(size); \ 644 if (num_allocations == NUM_ALLOCATIONS) \ 645 panic("too many ADD_TO_ALLOCATIONS()"); \ 646 allocations[num_allocations].al_ptr = (void**)&ptr; \ 647 allocations[num_allocations].al_size = size; \ 648 valloc_sz += size; \ 649 ++num_allocations; \ 650 } 651 652 /* 653 * Allocate all the initial memory needed by the page allocator. 654 */ 655 static void 656 perform_allocations(void) 657 { 658 caddr_t mem; 659 int i; 660 int valloc_align; 661 662 PRM_DEBUG(valloc_base); 663 PRM_DEBUG(valloc_sz); 664 valloc_align = mmu.level_size[mmu.max_page_level > 0]; 665 mem = BOP_ALLOC(bootops, (caddr_t)valloc_base, valloc_sz, valloc_align); 666 if (mem != (caddr_t)valloc_base) 667 panic("BOP_ALLOC() failed"); 668 bzero(mem, valloc_sz); 669 for (i = 0; i < num_allocations; ++i) { 670 *allocations[i].al_ptr = (void *)mem; 671 mem += allocations[i].al_size; 672 } 673 } 674 675 /* 676 * Our world looks like this at startup time. 677 * 678 * In a 32-bit OS, boot loads the kernel text at 0xfe800000 and kernel data 679 * at 0xfec00000. On a 64-bit OS, kernel text and data are loaded at 680 * 0xffffffff.fe800000 and 0xffffffff.fec00000 respectively. Those 681 * addresses are fixed in the binary at link time. 682 * 683 * On the text page: 684 * unix/genunix/krtld/module text loads. 685 * 686 * On the data page: 687 * unix/genunix/krtld/module data loads. 688 * 689 * Machine-dependent startup code 690 */ 691 void 692 startup(void) 693 { 694 #if !defined(__xpv) 695 extern void startup_pci_bios(void); 696 #endif 697 extern cpuset_t cpu_ready_set; 698 699 /* 700 * Make sure that nobody tries to use sekpm until we have 701 * initialized it properly. 702 */ 703 #if defined(__amd64) 704 kpm_desired = 1; 705 #endif 706 kpm_enable = 0; 707 CPUSET_ONLY(cpu_ready_set, 0); /* cpu 0 is boot cpu */ 708 709 #if defined(__xpv) /* XXPV fix me! */ 710 { 711 extern int segvn_use_regions; 712 segvn_use_regions = 0; 713 } 714 #endif 715 ssp_init(); 716 progressbar_init(); 717 startup_init(); 718 #if defined(__xpv) 719 startup_xen_version(); 720 #endif 721 startup_memlist(); 722 startup_kmem(); 723 startup_vm(); 724 #if !defined(__xpv) 725 /* 726 * Note we need to do this even on fast reboot in order to access 727 * the irq routing table (used for pci labels). 728 */ 729 startup_pci_bios(); 730 #endif 731 #if defined(__xpv) 732 startup_xen_mca(); 733 #endif 734 startup_modules(); 735 736 startup_end(); 737 } 738 739 static void 740 startup_init() 741 { 742 PRM_POINT("startup_init() starting..."); 743 744 /* 745 * Complete the extraction of cpuid data 746 */ 747 cpuid_pass2(CPU); 748 749 (void) check_boot_version(BOP_GETVERSION(bootops)); 750 751 /* 752 * Check for prom_debug in boot environment 753 */ 754 if (BOP_GETPROPLEN(bootops, "prom_debug") >= 0) { 755 ++prom_debug; 756 PRM_POINT("prom_debug found in boot enviroment"); 757 } 758 759 /* 760 * Collect node, cpu and memory configuration information. 761 */ 762 get_system_configuration(); 763 764 /* 765 * Halt if this is an unsupported processor. 766 */ 767 if (x86_type == X86_TYPE_486 || x86_type == X86_TYPE_CYRIX_486) { 768 printf("\n486 processor (\"%s\") detected.\n", 769 CPU->cpu_brandstr); 770 halt("This processor is not supported by this release " 771 "of Solaris."); 772 } 773 774 PRM_POINT("startup_init() done"); 775 } 776 777 /* 778 * Callback for copy_memlist_filter() to filter nucleus, kadb/kmdb, (ie. 779 * everything mapped above KERNEL_TEXT) pages from phys_avail. Note it 780 * also filters out physical page zero. There is some reliance on the 781 * boot loader allocating only a few contiguous physical memory chunks. 782 */ 783 static void 784 avail_filter(uint64_t *addr, uint64_t *size) 785 { 786 uintptr_t va; 787 uintptr_t next_va; 788 pfn_t pfn; 789 uint64_t pfn_addr; 790 uint64_t pfn_eaddr; 791 uint_t prot; 792 size_t len; 793 uint_t change; 794 795 if (prom_debug) 796 prom_printf("\tFilter: in: a=%" PRIx64 ", s=%" PRIx64 "\n", 797 *addr, *size); 798 799 /* 800 * page zero is required for BIOS.. never make it available 801 */ 802 if (*addr == 0) { 803 *addr += MMU_PAGESIZE; 804 *size -= MMU_PAGESIZE; 805 } 806 807 /* 808 * First we trim from the front of the range. Since kbm_probe() 809 * walks ranges in virtual order, but addr/size are physical, we need 810 * to the list until no changes are seen. This deals with the case 811 * where page "p" is mapped at v, page "p + PAGESIZE" is mapped at w 812 * but w < v. 813 */ 814 do { 815 change = 0; 816 for (va = KERNEL_TEXT; 817 *size > 0 && kbm_probe(&va, &len, &pfn, &prot) != 0; 818 va = next_va) { 819 820 next_va = va + len; 821 pfn_addr = pfn_to_pa(pfn); 822 pfn_eaddr = pfn_addr + len; 823 824 if (pfn_addr <= *addr && pfn_eaddr > *addr) { 825 change = 1; 826 while (*size > 0 && len > 0) { 827 *addr += MMU_PAGESIZE; 828 *size -= MMU_PAGESIZE; 829 len -= MMU_PAGESIZE; 830 } 831 } 832 } 833 if (change && prom_debug) 834 prom_printf("\t\ttrim: a=%" PRIx64 ", s=%" PRIx64 "\n", 835 *addr, *size); 836 } while (change); 837 838 /* 839 * Trim pages from the end of the range. 840 */ 841 for (va = KERNEL_TEXT; 842 *size > 0 && kbm_probe(&va, &len, &pfn, &prot) != 0; 843 va = next_va) { 844 845 next_va = va + len; 846 pfn_addr = pfn_to_pa(pfn); 847 848 if (pfn_addr >= *addr && pfn_addr < *addr + *size) 849 *size = pfn_addr - *addr; 850 } 851 852 if (prom_debug) 853 prom_printf("\tFilter out: a=%" PRIx64 ", s=%" PRIx64 "\n", 854 *addr, *size); 855 } 856 857 static void 858 kpm_init() 859 { 860 struct segkpm_crargs b; 861 862 /* 863 * These variables were all designed for sfmmu in which segkpm is 864 * mapped using a single pagesize - either 8KB or 4MB. On x86, we 865 * might use 2+ page sizes on a single machine, so none of these 866 * variables have a single correct value. They are set up as if we 867 * always use a 4KB pagesize, which should do no harm. In the long 868 * run, we should get rid of KPM's assumption that only a single 869 * pagesize is used. 870 */ 871 kpm_pgshft = MMU_PAGESHIFT; 872 kpm_pgsz = MMU_PAGESIZE; 873 kpm_pgoff = MMU_PAGEOFFSET; 874 kpmp2pshft = 0; 875 kpmpnpgs = 1; 876 ASSERT(((uintptr_t)kpm_vbase & (kpm_pgsz - 1)) == 0); 877 878 PRM_POINT("about to create segkpm"); 879 rw_enter(&kas.a_lock, RW_WRITER); 880 881 if (seg_attach(&kas, kpm_vbase, kpm_size, segkpm) < 0) 882 panic("cannot attach segkpm"); 883 884 b.prot = PROT_READ | PROT_WRITE; 885 b.nvcolors = 1; 886 887 if (segkpm_create(segkpm, (caddr_t)&b) != 0) 888 panic("segkpm_create segkpm"); 889 890 rw_exit(&kas.a_lock); 891 } 892 893 /* 894 * The debug info page provides enough information to allow external 895 * inspectors (e.g. when running under a hypervisor) to bootstrap 896 * themselves into allowing full-blown kernel debugging. 897 */ 898 static void 899 init_debug_info(void) 900 { 901 caddr_t mem; 902 debug_info_t *di; 903 904 #ifndef __lint 905 ASSERT(sizeof (debug_info_t) < MMU_PAGESIZE); 906 #endif 907 908 mem = BOP_ALLOC(bootops, (caddr_t)DEBUG_INFO_VA, MMU_PAGESIZE, 909 MMU_PAGESIZE); 910 911 if (mem != (caddr_t)DEBUG_INFO_VA) 912 panic("BOP_ALLOC() failed"); 913 bzero(mem, MMU_PAGESIZE); 914 915 di = (debug_info_t *)mem; 916 917 di->di_magic = DEBUG_INFO_MAGIC; 918 di->di_version = DEBUG_INFO_VERSION; 919 di->di_modules = (uintptr_t)&modules; 920 di->di_s_text = (uintptr_t)s_text; 921 di->di_e_text = (uintptr_t)e_text; 922 di->di_s_data = (uintptr_t)s_data; 923 di->di_e_data = (uintptr_t)e_data; 924 di->di_hat_htable_off = offsetof(hat_t, hat_htable); 925 di->di_ht_pfn_off = offsetof(htable_t, ht_pfn); 926 } 927 928 /* 929 * Build the memlists and other kernel essential memory system data structures. 930 * This is everything at valloc_base. 931 */ 932 static void 933 startup_memlist(void) 934 { 935 size_t memlist_sz; 936 size_t memseg_sz; 937 size_t pagehash_sz; 938 size_t pp_sz; 939 uintptr_t va; 940 size_t len; 941 uint_t prot; 942 pfn_t pfn; 943 int memblocks; 944 pfn_t rsvd_high_pfn; 945 pgcnt_t rsvd_pgcnt; 946 size_t rsvdmemlist_sz; 947 int rsvdmemblocks; 948 caddr_t pagecolor_mem; 949 size_t pagecolor_memsz; 950 caddr_t page_ctrs_mem; 951 size_t page_ctrs_size; 952 size_t pse_table_alloc_size; 953 struct memlist *current; 954 extern void startup_build_mem_nodes(struct memlist *); 955 956 /* XX64 fix these - they should be in include files */ 957 extern size_t page_coloring_init(uint_t, int, int); 958 extern void page_coloring_setup(caddr_t); 959 960 PRM_POINT("startup_memlist() starting..."); 961 962 /* 963 * Use leftover large page nucleus text/data space for loadable modules. 964 * Use at most MODTEXT/MODDATA. 965 */ 966 len = kbm_nucleus_size; 967 ASSERT(len > MMU_PAGESIZE); 968 969 moddata = (caddr_t)ROUND_UP_PAGE(e_data); 970 e_moddata = (caddr_t)P2ROUNDUP((uintptr_t)e_data, (uintptr_t)len); 971 if (e_moddata - moddata > MODDATA) 972 e_moddata = moddata + MODDATA; 973 974 modtext = (caddr_t)ROUND_UP_PAGE(e_text); 975 e_modtext = (caddr_t)P2ROUNDUP((uintptr_t)e_text, (uintptr_t)len); 976 if (e_modtext - modtext > MODTEXT) 977 e_modtext = modtext + MODTEXT; 978 979 econtig = e_moddata; 980 981 PRM_DEBUG(modtext); 982 PRM_DEBUG(e_modtext); 983 PRM_DEBUG(moddata); 984 PRM_DEBUG(e_moddata); 985 PRM_DEBUG(econtig); 986 987 /* 988 * Examine the boot loader physical memory map to find out: 989 * - total memory in system - physinstalled 990 * - the max physical address - physmax 991 * - the number of discontiguous segments of memory. 992 */ 993 if (prom_debug) 994 print_memlist("boot physinstalled", 995 bootops->boot_mem->physinstalled); 996 installed_top_size_ex(bootops->boot_mem->physinstalled, &physmax, 997 &physinstalled, &memblocks); 998 PRM_DEBUG(physmax); 999 PRM_DEBUG(physinstalled); 1000 PRM_DEBUG(memblocks); 1001 1002 /* 1003 * Compute maximum physical address for memory DR operations. 1004 * Memory DR operations are unsupported on xpv or 32bit OSes. 1005 */ 1006 #ifdef __amd64 1007 if (plat_dr_support_memory()) { 1008 if (plat_dr_physmax == 0) { 1009 uint_t pabits = UINT_MAX; 1010 1011 cpuid_get_addrsize(CPU, &pabits, NULL); 1012 plat_dr_physmax = btop(1ULL << pabits); 1013 } 1014 if (plat_dr_physmax > PHYSMEM_MAX64) 1015 plat_dr_physmax = PHYSMEM_MAX64; 1016 } else 1017 #endif 1018 plat_dr_physmax = 0; 1019 1020 /* 1021 * Examine the bios reserved memory to find out: 1022 * - the number of discontiguous segments of memory. 1023 */ 1024 if (prom_debug) 1025 print_memlist("boot reserved mem", 1026 bootops->boot_mem->rsvdmem); 1027 installed_top_size_ex(bootops->boot_mem->rsvdmem, &rsvd_high_pfn, 1028 &rsvd_pgcnt, &rsvdmemblocks); 1029 PRM_DEBUG(rsvd_high_pfn); 1030 PRM_DEBUG(rsvd_pgcnt); 1031 PRM_DEBUG(rsvdmemblocks); 1032 1033 /* 1034 * Initialize hat's mmu parameters. 1035 * Check for enforce-prot-exec in boot environment. It's used to 1036 * enable/disable support for the page table entry NX bit. 1037 * The default is to enforce PROT_EXEC on processors that support NX. 1038 * Boot seems to round up the "len", but 8 seems to be big enough. 1039 */ 1040 mmu_init(); 1041 1042 #ifdef __i386 1043 /* 1044 * physmax is lowered if there is more memory than can be 1045 * physically addressed in 32 bit (PAE/non-PAE) modes. 1046 */ 1047 if (mmu.pae_hat) { 1048 if (PFN_ABOVE64G(physmax)) { 1049 physinstalled -= (physmax - (PFN_64G - 1)); 1050 physmax = PFN_64G - 1; 1051 } 1052 } else { 1053 if (PFN_ABOVE4G(physmax)) { 1054 physinstalled -= (physmax - (PFN_4G - 1)); 1055 physmax = PFN_4G - 1; 1056 } 1057 } 1058 #endif 1059 1060 startup_build_mem_nodes(bootops->boot_mem->physinstalled); 1061 1062 if (BOP_GETPROPLEN(bootops, "enforce-prot-exec") >= 0) { 1063 int len = BOP_GETPROPLEN(bootops, "enforce-prot-exec"); 1064 char value[8]; 1065 1066 if (len < 8) 1067 (void) BOP_GETPROP(bootops, "enforce-prot-exec", value); 1068 else 1069 (void) strcpy(value, ""); 1070 if (strcmp(value, "off") == 0) 1071 mmu.pt_nx = 0; 1072 } 1073 PRM_DEBUG(mmu.pt_nx); 1074 1075 /* 1076 * We will need page_t's for every page in the system, except for 1077 * memory mapped at or above above the start of the kernel text segment. 1078 * 1079 * pages above e_modtext are attributed to kernel debugger (obp_pages) 1080 */ 1081 npages = physinstalled - 1; /* avail_filter() skips page 0, so "- 1" */ 1082 obp_pages = 0; 1083 va = KERNEL_TEXT; 1084 while (kbm_probe(&va, &len, &pfn, &prot) != 0) { 1085 npages -= len >> MMU_PAGESHIFT; 1086 if (va >= (uintptr_t)e_moddata) 1087 obp_pages += len >> MMU_PAGESHIFT; 1088 va += len; 1089 } 1090 PRM_DEBUG(npages); 1091 PRM_DEBUG(obp_pages); 1092 1093 /* 1094 * If physmem is patched to be non-zero, use it instead of the computed 1095 * value unless it is larger than the actual amount of memory on hand. 1096 */ 1097 if (physmem == 0 || physmem > npages) { 1098 physmem = npages; 1099 } else if (physmem < npages) { 1100 orig_npages = npages; 1101 npages = physmem; 1102 } 1103 PRM_DEBUG(physmem); 1104 1105 /* 1106 * We now compute the sizes of all the initial allocations for 1107 * structures the kernel needs in order do kmem_alloc(). These 1108 * include: 1109 * memsegs 1110 * memlists 1111 * page hash table 1112 * page_t's 1113 * page coloring data structs 1114 */ 1115 memseg_sz = sizeof (struct memseg) * (memblocks + POSS_NEW_FRAGMENTS); 1116 ADD_TO_ALLOCATIONS(memseg_base, memseg_sz); 1117 PRM_DEBUG(memseg_sz); 1118 1119 /* 1120 * Reserve space for memlists. There's no real good way to know exactly 1121 * how much room we'll need, but this should be a good upper bound. 1122 */ 1123 memlist_sz = ROUND_UP_PAGE(2 * sizeof (struct memlist) * 1124 (memblocks + POSS_NEW_FRAGMENTS)); 1125 ADD_TO_ALLOCATIONS(memlist, memlist_sz); 1126 PRM_DEBUG(memlist_sz); 1127 1128 /* 1129 * Reserve space for bios reserved memlists. 1130 */ 1131 rsvdmemlist_sz = ROUND_UP_PAGE(2 * sizeof (struct memlist) * 1132 (rsvdmemblocks + POSS_NEW_FRAGMENTS)); 1133 ADD_TO_ALLOCATIONS(bios_rsvd, rsvdmemlist_sz); 1134 PRM_DEBUG(rsvdmemlist_sz); 1135 1136 /* LINTED */ 1137 ASSERT(P2SAMEHIGHBIT((1 << PP_SHIFT), sizeof (struct page))); 1138 /* 1139 * The page structure hash table size is a power of 2 1140 * such that the average hash chain length is PAGE_HASHAVELEN. 1141 */ 1142 page_hashsz = npages / PAGE_HASHAVELEN; 1143 page_hashsz_shift = highbit(page_hashsz); 1144 page_hashsz = 1 << page_hashsz_shift; 1145 pagehash_sz = sizeof (struct page *) * page_hashsz; 1146 ADD_TO_ALLOCATIONS(page_hash, pagehash_sz); 1147 PRM_DEBUG(pagehash_sz); 1148 1149 /* 1150 * Set aside room for the page structures themselves. 1151 */ 1152 PRM_DEBUG(npages); 1153 pp_sz = sizeof (struct page) * npages; 1154 ADD_TO_ALLOCATIONS(pp_base, pp_sz); 1155 PRM_DEBUG(pp_sz); 1156 1157 /* 1158 * determine l2 cache info and memory size for page coloring 1159 */ 1160 (void) getl2cacheinfo(CPU, 1161 &l2cache_sz, &l2cache_linesz, &l2cache_assoc); 1162 pagecolor_memsz = 1163 page_coloring_init(l2cache_sz, l2cache_linesz, l2cache_assoc); 1164 ADD_TO_ALLOCATIONS(pagecolor_mem, pagecolor_memsz); 1165 PRM_DEBUG(pagecolor_memsz); 1166 1167 page_ctrs_size = page_ctrs_sz(); 1168 ADD_TO_ALLOCATIONS(page_ctrs_mem, page_ctrs_size); 1169 PRM_DEBUG(page_ctrs_size); 1170 1171 /* 1172 * Allocate the array that protects pp->p_selock. 1173 */ 1174 pse_shift = size_pse_array(physmem, max_ncpus); 1175 pse_table_size = 1 << pse_shift; 1176 pse_table_alloc_size = pse_table_size * sizeof (pad_mutex_t); 1177 ADD_TO_ALLOCATIONS(pse_mutex, pse_table_alloc_size); 1178 1179 #if defined(__amd64) 1180 valloc_sz = ROUND_UP_LPAGE(valloc_sz); 1181 valloc_base = VALLOC_BASE; 1182 1183 /* 1184 * The default values of VALLOC_BASE and SEGKPM_BASE should work 1185 * for values of physmax up to 1 Terabyte. They need adjusting when 1186 * memory is at addresses above 1 TB. When adjusted, segkpm_base must 1187 * be aligned on KERNEL_REDZONE_SIZE boundary (span of top level pte). 1188 */ 1189 if (physmax + 1 > mmu_btop(TERABYTE) || 1190 plat_dr_physmax > mmu_btop(TERABYTE)) { 1191 uint64_t kpm_resv_amount = mmu_ptob(physmax + 1); 1192 1193 if (kpm_resv_amount < mmu_ptob(plat_dr_physmax)) { 1194 kpm_resv_amount = mmu_ptob(plat_dr_physmax); 1195 } 1196 1197 segkpm_base = -(P2ROUNDUP((2 * kpm_resv_amount), 1198 KERNEL_REDZONE_SIZE)); /* down from top VA */ 1199 1200 /* make sure we leave some space for user apps above hole */ 1201 segkpm_base = MAX(segkpm_base, AMD64_VA_HOLE_END + TERABYTE); 1202 if (segkpm_base > SEGKPM_BASE) 1203 segkpm_base = SEGKPM_BASE; 1204 PRM_DEBUG(segkpm_base); 1205 1206 valloc_base = segkpm_base + P2ROUNDUP(kpm_resv_amount, ONE_GIG); 1207 if (valloc_base < segkpm_base) 1208 panic("not enough kernel VA to support memory size"); 1209 PRM_DEBUG(valloc_base); 1210 } 1211 #else /* __i386 */ 1212 valloc_base = (uintptr_t)(MISC_VA_BASE - valloc_sz); 1213 valloc_base = P2ALIGN(valloc_base, mmu.level_size[1]); 1214 PRM_DEBUG(valloc_base); 1215 #endif /* __i386 */ 1216 1217 /* 1218 * do all the initial allocations 1219 */ 1220 perform_allocations(); 1221 1222 /* 1223 * Build phys_install and phys_avail in kernel memspace. 1224 * - phys_install should be all memory in the system. 1225 * - phys_avail is phys_install minus any memory mapped before this 1226 * point above KERNEL_TEXT. 1227 */ 1228 current = phys_install = memlist; 1229 copy_memlist_filter(bootops->boot_mem->physinstalled, ¤t, NULL); 1230 if ((caddr_t)current > (caddr_t)memlist + memlist_sz) 1231 panic("physinstalled was too big!"); 1232 if (prom_debug) 1233 print_memlist("phys_install", phys_install); 1234 1235 phys_avail = current; 1236 PRM_POINT("Building phys_avail:\n"); 1237 copy_memlist_filter(bootops->boot_mem->physinstalled, ¤t, 1238 avail_filter); 1239 if ((caddr_t)current > (caddr_t)memlist + memlist_sz) 1240 panic("physavail was too big!"); 1241 if (prom_debug) 1242 print_memlist("phys_avail", phys_avail); 1243 #ifndef __xpv 1244 /* 1245 * Free unused memlist items, which may be used by memory DR driver 1246 * at runtime. 1247 */ 1248 if ((caddr_t)current < (caddr_t)memlist + memlist_sz) { 1249 memlist_free_block((caddr_t)current, 1250 (caddr_t)memlist + memlist_sz - (caddr_t)current); 1251 } 1252 #endif 1253 1254 /* 1255 * Build bios reserved memspace 1256 */ 1257 current = bios_rsvd; 1258 copy_memlist_filter(bootops->boot_mem->rsvdmem, ¤t, NULL); 1259 if ((caddr_t)current > (caddr_t)bios_rsvd + rsvdmemlist_sz) 1260 panic("bios_rsvd was too big!"); 1261 if (prom_debug) 1262 print_memlist("bios_rsvd", bios_rsvd); 1263 #ifndef __xpv 1264 /* 1265 * Free unused memlist items, which may be used by memory DR driver 1266 * at runtime. 1267 */ 1268 if ((caddr_t)current < (caddr_t)bios_rsvd + rsvdmemlist_sz) { 1269 memlist_free_block((caddr_t)current, 1270 (caddr_t)bios_rsvd + rsvdmemlist_sz - (caddr_t)current); 1271 } 1272 #endif 1273 1274 /* 1275 * setup page coloring 1276 */ 1277 page_coloring_setup(pagecolor_mem); 1278 page_lock_init(); /* currently a no-op */ 1279 1280 /* 1281 * free page list counters 1282 */ 1283 (void) page_ctrs_alloc(page_ctrs_mem); 1284 1285 /* 1286 * Size the pcf array based on the number of cpus in the box at 1287 * boot time. 1288 */ 1289 1290 pcf_init(); 1291 1292 /* 1293 * Initialize the page structures from the memory lists. 1294 */ 1295 availrmem_initial = availrmem = freemem = 0; 1296 PRM_POINT("Calling kphysm_init()..."); 1297 npages = kphysm_init(pp_base, npages); 1298 PRM_POINT("kphysm_init() done"); 1299 PRM_DEBUG(npages); 1300 1301 init_debug_info(); 1302 1303 /* 1304 * Now that page_t's have been initialized, remove all the 1305 * initial allocation pages from the kernel free page lists. 1306 */ 1307 boot_mapin((caddr_t)valloc_base, valloc_sz); 1308 boot_mapin((caddr_t)MISC_VA_BASE, MISC_VA_SIZE); 1309 PRM_POINT("startup_memlist() done"); 1310 1311 PRM_DEBUG(valloc_sz); 1312 1313 #if defined(__amd64) 1314 if ((availrmem >> (30 - MMU_PAGESHIFT)) >= 1315 textrepl_min_gb && l2cache_sz <= 2 << 20) { 1316 extern size_t textrepl_size_thresh; 1317 textrepl_size_thresh = (16 << 20) - 1; 1318 } 1319 #endif 1320 } 1321 1322 /* 1323 * Layout the kernel's part of address space and initialize kmem allocator. 1324 */ 1325 static void 1326 startup_kmem(void) 1327 { 1328 extern void page_set_colorequiv_arr(void); 1329 1330 PRM_POINT("startup_kmem() starting..."); 1331 1332 #if defined(__amd64) 1333 if (eprom_kernelbase && eprom_kernelbase != KERNELBASE) 1334 cmn_err(CE_NOTE, "!kernelbase cannot be changed on 64-bit " 1335 "systems."); 1336 kernelbase = segkpm_base - KERNEL_REDZONE_SIZE; 1337 core_base = (uintptr_t)COREHEAP_BASE; 1338 core_size = (size_t)MISC_VA_BASE - COREHEAP_BASE; 1339 #else /* __i386 */ 1340 /* 1341 * We configure kernelbase based on: 1342 * 1343 * 1. user specified kernelbase via eeprom command. Value cannot exceed 1344 * KERNELBASE_MAX. we large page align eprom_kernelbase 1345 * 1346 * 2. Default to KERNELBASE and adjust to 2X less the size for page_t. 1347 * On large memory systems we must lower kernelbase to allow 1348 * enough room for page_t's for all of memory. 1349 * 1350 * The value set here, might be changed a little later. 1351 */ 1352 if (eprom_kernelbase) { 1353 kernelbase = eprom_kernelbase & mmu.level_mask[1]; 1354 if (kernelbase > KERNELBASE_MAX) 1355 kernelbase = KERNELBASE_MAX; 1356 } else { 1357 kernelbase = (uintptr_t)KERNELBASE; 1358 kernelbase -= ROUND_UP_4MEG(2 * valloc_sz); 1359 } 1360 ASSERT((kernelbase & mmu.level_offset[1]) == 0); 1361 core_base = valloc_base; 1362 core_size = 0; 1363 #endif /* __i386 */ 1364 1365 PRM_DEBUG(core_base); 1366 PRM_DEBUG(core_size); 1367 PRM_DEBUG(kernelbase); 1368 1369 #if defined(__i386) 1370 segkp_fromheap = 1; 1371 #endif /* __i386 */ 1372 1373 ekernelheap = (char *)core_base; 1374 PRM_DEBUG(ekernelheap); 1375 1376 /* 1377 * Now that we know the real value of kernelbase, 1378 * update variables that were initialized with a value of 1379 * KERNELBASE (in common/conf/param.c). 1380 * 1381 * XXX The problem with this sort of hackery is that the 1382 * compiler just may feel like putting the const declarations 1383 * (in param.c) into the .text section. Perhaps they should 1384 * just be declared as variables there? 1385 */ 1386 1387 *(uintptr_t *)&_kernelbase = kernelbase; 1388 *(uintptr_t *)&_userlimit = kernelbase; 1389 #if defined(__amd64) 1390 *(uintptr_t *)&_userlimit -= KERNELBASE - USERLIMIT; 1391 #else 1392 *(uintptr_t *)&_userlimit32 = _userlimit; 1393 #endif 1394 PRM_DEBUG(_kernelbase); 1395 PRM_DEBUG(_userlimit); 1396 PRM_DEBUG(_userlimit32); 1397 1398 layout_kernel_va(); 1399 1400 #if defined(__i386) 1401 /* 1402 * If segmap is too large we can push the bottom of the kernel heap 1403 * higher than the base. Or worse, it could exceed the top of the 1404 * VA space entirely, causing it to wrap around. 1405 */ 1406 if (kernelheap >= ekernelheap || (uintptr_t)kernelheap < kernelbase) 1407 panic("too little address space available for kernelheap," 1408 " use eeprom for lower kernelbase or smaller segmapsize"); 1409 #endif /* __i386 */ 1410 1411 /* 1412 * Initialize the kernel heap. Note 3rd argument must be > 1st. 1413 */ 1414 kernelheap_init(kernelheap, ekernelheap, 1415 kernelheap + MMU_PAGESIZE, 1416 (void *)core_base, (void *)(core_base + core_size)); 1417 1418 #if defined(__xpv) 1419 /* 1420 * Link pending events struct into cpu struct 1421 */ 1422 CPU->cpu_m.mcpu_evt_pend = &cpu0_evt_data; 1423 #endif 1424 /* 1425 * Initialize kernel memory allocator. 1426 */ 1427 kmem_init(); 1428 1429 /* 1430 * Factor in colorequiv to check additional 'equivalent' bins 1431 */ 1432 page_set_colorequiv_arr(); 1433 1434 /* 1435 * print this out early so that we know what's going on 1436 */ 1437 print_x86_featureset(x86_featureset); 1438 1439 /* 1440 * Initialize bp_mapin(). 1441 */ 1442 bp_init(MMU_PAGESIZE, HAT_STORECACHING_OK); 1443 1444 /* 1445 * orig_npages is non-zero if physmem has been configured for less 1446 * than the available memory. 1447 */ 1448 if (orig_npages) { 1449 cmn_err(CE_WARN, "!%slimiting physmem to 0x%lx of 0x%lx pages", 1450 (npages == PHYSMEM ? "Due to virtual address space " : ""), 1451 npages, orig_npages); 1452 } 1453 #if defined(__i386) 1454 if (eprom_kernelbase && (eprom_kernelbase != kernelbase)) 1455 cmn_err(CE_WARN, "kernelbase value, User specified 0x%lx, " 1456 "System using 0x%lx", 1457 (uintptr_t)eprom_kernelbase, (uintptr_t)kernelbase); 1458 #endif 1459 1460 #ifdef KERNELBASE_ABI_MIN 1461 if (kernelbase < (uintptr_t)KERNELBASE_ABI_MIN) { 1462 cmn_err(CE_NOTE, "!kernelbase set to 0x%lx, system is not " 1463 "i386 ABI compliant.", (uintptr_t)kernelbase); 1464 } 1465 #endif 1466 1467 #ifndef __xpv 1468 if (plat_dr_support_memory()) { 1469 mem_config_init(); 1470 } 1471 #else /* __xpv */ 1472 /* 1473 * Some of the xen start information has to be relocated up 1474 * into the kernel's permanent address space. 1475 */ 1476 PRM_POINT("calling xen_relocate_start_info()"); 1477 xen_relocate_start_info(); 1478 PRM_POINT("xen_relocate_start_info() done"); 1479 1480 /* 1481 * (Update the vcpu pointer in our cpu structure to point into 1482 * the relocated shared info.) 1483 */ 1484 CPU->cpu_m.mcpu_vcpu_info = 1485 &HYPERVISOR_shared_info->vcpu_info[CPU->cpu_id]; 1486 #endif /* __xpv */ 1487 1488 PRM_POINT("startup_kmem() done"); 1489 } 1490 1491 #ifndef __xpv 1492 /* 1493 * If we have detected that we are running in an HVM environment, we need 1494 * to prepend the PV driver directory to the module search path. 1495 */ 1496 #define HVM_MOD_DIR "/platform/i86hvm/kernel" 1497 static void 1498 update_default_path() 1499 { 1500 char *current, *newpath; 1501 int newlen; 1502 1503 /* 1504 * We are about to resync with krtld. krtld will reset its 1505 * internal module search path iff Solaris has set default_path. 1506 * We want to be sure we're prepending this new directory to the 1507 * right search path. 1508 */ 1509 current = (default_path == NULL) ? kobj_module_path : default_path; 1510 1511 newlen = strlen(HVM_MOD_DIR) + strlen(current) + 2; 1512 newpath = kmem_alloc(newlen, KM_SLEEP); 1513 (void) strcpy(newpath, HVM_MOD_DIR); 1514 (void) strcat(newpath, " "); 1515 (void) strcat(newpath, current); 1516 1517 default_path = newpath; 1518 } 1519 #endif 1520 1521 static void 1522 startup_modules(void) 1523 { 1524 int cnt; 1525 extern void prom_setup(void); 1526 int32_t v, h; 1527 char d[11]; 1528 char *cp; 1529 cmi_hdl_t hdl; 1530 1531 PRM_POINT("startup_modules() starting..."); 1532 1533 #ifndef __xpv 1534 /* 1535 * Initialize ten-micro second timer so that drivers will 1536 * not get short changed in their init phase. This was 1537 * not getting called until clkinit which, on fast cpu's 1538 * caused the drv_usecwait to be way too short. 1539 */ 1540 microfind(); 1541 1542 if ((get_hwenv() & HW_XEN_HVM) != 0) 1543 update_default_path(); 1544 #endif 1545 1546 /* 1547 * Read the GMT lag from /etc/rtc_config. 1548 */ 1549 sgmtl(process_rtc_config_file()); 1550 1551 /* 1552 * Calculate default settings of system parameters based upon 1553 * maxusers, yet allow to be overridden via the /etc/system file. 1554 */ 1555 param_calc(0); 1556 1557 mod_setup(); 1558 1559 /* 1560 * Initialize system parameters. 1561 */ 1562 param_init(); 1563 1564 /* 1565 * Initialize the default brands 1566 */ 1567 brand_init(); 1568 1569 /* 1570 * maxmem is the amount of physical memory we're playing with. 1571 */ 1572 maxmem = physmem; 1573 1574 /* 1575 * Initialize segment management stuff. 1576 */ 1577 seg_init(); 1578 1579 if (modload("fs", "specfs") == -1) 1580 halt("Can't load specfs"); 1581 1582 if (modload("fs", "devfs") == -1) 1583 halt("Can't load devfs"); 1584 1585 if (modload("fs", "dev") == -1) 1586 halt("Can't load dev"); 1587 1588 if (modload("fs", "procfs") == -1) 1589 halt("Can't load procfs"); 1590 1591 (void) modloadonly("sys", "lbl_edition"); 1592 1593 dispinit(); 1594 1595 /* Read cluster configuration data. */ 1596 clconf_init(); 1597 1598 #if defined(__xpv) 1599 (void) ec_init(); 1600 gnttab_init(); 1601 (void) xs_early_init(); 1602 #endif /* __xpv */ 1603 1604 /* 1605 * Create a kernel device tree. First, create rootnex and 1606 * then invoke bus specific code to probe devices. 1607 */ 1608 setup_ddi(); 1609 1610 #ifdef __xpv 1611 if (DOMAIN_IS_INITDOMAIN(xen_info)) 1612 #endif 1613 { 1614 id_t smid; 1615 smbios_system_t smsys; 1616 smbios_info_t sminfo; 1617 char *mfg; 1618 /* 1619 * Load the System Management BIOS into the global ksmbios 1620 * handle, if an SMBIOS is present on this system. 1621 * Also set "si-hw-provider" property, if not already set. 1622 */ 1623 ksmbios = smbios_open(NULL, SMB_VERSION, ksmbios_flags, NULL); 1624 if (ksmbios != NULL && 1625 ((smid = smbios_info_system(ksmbios, &smsys)) != SMB_ERR) && 1626 (smbios_info_common(ksmbios, smid, &sminfo)) != SMB_ERR) { 1627 mfg = (char *)sminfo.smbi_manufacturer; 1628 if (BOP_GETPROPLEN(bootops, "si-hw-provider") < 0) { 1629 extern char hw_provider[]; 1630 int i; 1631 for (i = 0; i < SYS_NMLN; i++) { 1632 if (isprint(mfg[i])) 1633 hw_provider[i] = mfg[i]; 1634 else { 1635 hw_provider[i] = '\0'; 1636 break; 1637 } 1638 } 1639 hw_provider[SYS_NMLN - 1] = '\0'; 1640 } 1641 } 1642 } 1643 1644 1645 /* 1646 * Originally clconf_init() apparently needed the hostid. But 1647 * this no longer appears to be true - it uses its own nodeid. 1648 * By placing the hostid logic here, we are able to make use of 1649 * the SMBIOS UUID. 1650 */ 1651 if ((h = set_soft_hostid()) == HW_INVALID_HOSTID) { 1652 cmn_err(CE_WARN, "Unable to set hostid"); 1653 } else { 1654 for (v = h, cnt = 0; cnt < 10; cnt++) { 1655 d[cnt] = (char)(v % 10); 1656 v /= 10; 1657 if (v == 0) 1658 break; 1659 } 1660 for (cp = hw_serial; cnt >= 0; cnt--) 1661 *cp++ = d[cnt] + '0'; 1662 *cp = 0; 1663 } 1664 1665 /* 1666 * Set up the CPU module subsystem for the boot cpu in the native 1667 * case, and all physical cpu resource in the xpv dom0 case. 1668 * Modifies the device tree, so this must be done after 1669 * setup_ddi(). 1670 */ 1671 #ifdef __xpv 1672 /* 1673 * If paravirtualized and on dom0 then we initialize all physical 1674 * cpu handles now; if paravirtualized on a domU then do not 1675 * initialize. 1676 */ 1677 if (DOMAIN_IS_INITDOMAIN(xen_info)) { 1678 xen_mc_lcpu_cookie_t cpi; 1679 1680 for (cpi = xen_physcpu_next(NULL); cpi != NULL; 1681 cpi = xen_physcpu_next(cpi)) { 1682 if ((hdl = cmi_init(CMI_HDL_SOLARIS_xVM_MCA, 1683 xen_physcpu_chipid(cpi), xen_physcpu_coreid(cpi), 1684 xen_physcpu_strandid(cpi))) != NULL && 1685 is_x86_feature(x86_featureset, X86FSET_MCA)) 1686 cmi_mca_init(hdl); 1687 } 1688 } 1689 #else 1690 /* 1691 * Initialize a handle for the boot cpu - others will initialize 1692 * as they startup. Do not do this if we know we are in an HVM domU. 1693 */ 1694 if ((get_hwenv() & HW_XEN_HVM) == 0 && 1695 (hdl = cmi_init(CMI_HDL_NATIVE, cmi_ntv_hwchipid(CPU), 1696 cmi_ntv_hwcoreid(CPU), cmi_ntv_hwstrandid(CPU))) != NULL && 1697 is_x86_feature(x86_featureset, X86FSET_MCA)) { 1698 cmi_mca_init(hdl); 1699 CPU->cpu_m.mcpu_cmi_hdl = hdl; 1700 } 1701 #endif /* __xpv */ 1702 1703 /* 1704 * Fake a prom tree such that /dev/openprom continues to work 1705 */ 1706 PRM_POINT("startup_modules: calling prom_setup..."); 1707 prom_setup(); 1708 PRM_POINT("startup_modules: done"); 1709 1710 /* 1711 * Load all platform specific modules 1712 */ 1713 PRM_POINT("startup_modules: calling psm_modload..."); 1714 psm_modload(); 1715 1716 PRM_POINT("startup_modules() done"); 1717 } 1718 1719 /* 1720 * claim a "setaside" boot page for use in the kernel 1721 */ 1722 page_t * 1723 boot_claim_page(pfn_t pfn) 1724 { 1725 page_t *pp; 1726 1727 pp = page_numtopp_nolock(pfn); 1728 ASSERT(pp != NULL); 1729 1730 if (PP_ISBOOTPAGES(pp)) { 1731 if (pp->p_next != NULL) 1732 pp->p_next->p_prev = pp->p_prev; 1733 if (pp->p_prev == NULL) 1734 bootpages = pp->p_next; 1735 else 1736 pp->p_prev->p_next = pp->p_next; 1737 } else { 1738 /* 1739 * htable_attach() expects a base pagesize page 1740 */ 1741 if (pp->p_szc != 0) 1742 page_boot_demote(pp); 1743 pp = page_numtopp(pfn, SE_EXCL); 1744 } 1745 return (pp); 1746 } 1747 1748 /* 1749 * Walk through the pagetables looking for pages mapped in by boot. If the 1750 * setaside flag is set the pages are expected to be returned to the 1751 * kernel later in boot, so we add them to the bootpages list. 1752 */ 1753 static void 1754 protect_boot_range(uintptr_t low, uintptr_t high, int setaside) 1755 { 1756 uintptr_t va = low; 1757 size_t len; 1758 uint_t prot; 1759 pfn_t pfn; 1760 page_t *pp; 1761 pgcnt_t boot_protect_cnt = 0; 1762 1763 while (kbm_probe(&va, &len, &pfn, &prot) != 0 && va < high) { 1764 if (va + len >= high) 1765 panic("0x%lx byte mapping at 0x%p exceeds boot's " 1766 "legal range.", len, (void *)va); 1767 1768 while (len > 0) { 1769 pp = page_numtopp_alloc(pfn); 1770 if (pp != NULL) { 1771 if (setaside == 0) 1772 panic("Unexpected mapping by boot. " 1773 "addr=%p pfn=%lx\n", 1774 (void *)va, pfn); 1775 1776 pp->p_next = bootpages; 1777 pp->p_prev = NULL; 1778 PP_SETBOOTPAGES(pp); 1779 if (bootpages != NULL) { 1780 bootpages->p_prev = pp; 1781 } 1782 bootpages = pp; 1783 ++boot_protect_cnt; 1784 } 1785 1786 ++pfn; 1787 len -= MMU_PAGESIZE; 1788 va += MMU_PAGESIZE; 1789 } 1790 } 1791 PRM_DEBUG(boot_protect_cnt); 1792 } 1793 1794 /* 1795 * 1796 */ 1797 static void 1798 layout_kernel_va(void) 1799 { 1800 PRM_POINT("layout_kernel_va() starting..."); 1801 /* 1802 * Establish the final size of the kernel's heap, size of segmap, 1803 * segkp, etc. 1804 */ 1805 1806 #if defined(__amd64) 1807 1808 kpm_vbase = (caddr_t)segkpm_base; 1809 if (physmax + 1 < plat_dr_physmax) { 1810 kpm_size = ROUND_UP_LPAGE(mmu_ptob(plat_dr_physmax)); 1811 } else { 1812 kpm_size = ROUND_UP_LPAGE(mmu_ptob(physmax + 1)); 1813 } 1814 if ((uintptr_t)kpm_vbase + kpm_size > (uintptr_t)valloc_base) 1815 panic("not enough room for kpm!"); 1816 PRM_DEBUG(kpm_size); 1817 PRM_DEBUG(kpm_vbase); 1818 1819 /* 1820 * By default we create a seg_kp in 64 bit kernels, it's a little 1821 * faster to access than embedding it in the heap. 1822 */ 1823 segkp_base = (caddr_t)valloc_base + valloc_sz; 1824 if (!segkp_fromheap) { 1825 size_t sz = mmu_ptob(segkpsize); 1826 1827 /* 1828 * determine size of segkp 1829 */ 1830 if (sz < SEGKPMINSIZE || sz > SEGKPMAXSIZE) { 1831 sz = SEGKPDEFSIZE; 1832 cmn_err(CE_WARN, "!Illegal value for segkpsize. " 1833 "segkpsize has been reset to %ld pages", 1834 mmu_btop(sz)); 1835 } 1836 sz = MIN(sz, MAX(SEGKPMINSIZE, mmu_ptob(physmem))); 1837 1838 segkpsize = mmu_btop(ROUND_UP_LPAGE(sz)); 1839 } 1840 PRM_DEBUG(segkp_base); 1841 PRM_DEBUG(segkpsize); 1842 1843 /* 1844 * segzio is used for ZFS cached data. It uses a distinct VA 1845 * segment (from kernel heap) so that we can easily tell not to 1846 * include it in kernel crash dumps on 64 bit kernels. The trick is 1847 * to give it lots of VA, but not constrain the kernel heap. 1848 * We scale the size of segzio linearly with physmem up to 1849 * SEGZIOMAXSIZE. Above that amount it scales at 50% of physmem. 1850 */ 1851 segzio_base = segkp_base + mmu_ptob(segkpsize); 1852 if (segzio_fromheap) { 1853 segziosize = 0; 1854 } else { 1855 size_t physmem_size = mmu_ptob(physmem); 1856 size_t size = (segziosize == 0) ? 1857 physmem_size : mmu_ptob(segziosize); 1858 1859 if (size < SEGZIOMINSIZE) 1860 size = SEGZIOMINSIZE; 1861 if (size > SEGZIOMAXSIZE) { 1862 size = SEGZIOMAXSIZE; 1863 if (physmem_size > size) 1864 size += (physmem_size - size) / 2; 1865 } 1866 segziosize = mmu_btop(ROUND_UP_LPAGE(size)); 1867 } 1868 PRM_DEBUG(segziosize); 1869 PRM_DEBUG(segzio_base); 1870 1871 /* 1872 * Put the range of VA for device mappings next, kmdb knows to not 1873 * grep in this range of addresses. 1874 */ 1875 toxic_addr = 1876 ROUND_UP_LPAGE((uintptr_t)segzio_base + mmu_ptob(segziosize)); 1877 PRM_DEBUG(toxic_addr); 1878 segmap_start = ROUND_UP_LPAGE(toxic_addr + toxic_size); 1879 #else /* __i386 */ 1880 segmap_start = ROUND_UP_LPAGE(kernelbase); 1881 #endif /* __i386 */ 1882 PRM_DEBUG(segmap_start); 1883 1884 /* 1885 * Users can change segmapsize through eeprom. If the variable 1886 * is tuned through eeprom, there is no upper bound on the 1887 * size of segmap. 1888 */ 1889 segmapsize = MAX(ROUND_UP_LPAGE(segmapsize), SEGMAPDEFAULT); 1890 1891 #if defined(__i386) 1892 /* 1893 * 32-bit systems don't have segkpm or segkp, so segmap appears at 1894 * the bottom of the kernel's address range. Set aside space for a 1895 * small red zone just below the start of segmap. 1896 */ 1897 segmap_start += KERNEL_REDZONE_SIZE; 1898 segmapsize -= KERNEL_REDZONE_SIZE; 1899 #endif 1900 1901 PRM_DEBUG(segmap_start); 1902 PRM_DEBUG(segmapsize); 1903 kernelheap = (caddr_t)ROUND_UP_LPAGE(segmap_start + segmapsize); 1904 PRM_DEBUG(kernelheap); 1905 PRM_POINT("layout_kernel_va() done..."); 1906 } 1907 1908 /* 1909 * Finish initializing the VM system, now that we are no longer 1910 * relying on the boot time memory allocators. 1911 */ 1912 static void 1913 startup_vm(void) 1914 { 1915 struct segmap_crargs a; 1916 1917 extern int use_brk_lpg, use_stk_lpg; 1918 1919 PRM_POINT("startup_vm() starting..."); 1920 1921 /* 1922 * Initialize the hat layer. 1923 */ 1924 hat_init(); 1925 1926 /* 1927 * Do final allocations of HAT data structures that need to 1928 * be allocated before quiescing the boot loader. 1929 */ 1930 PRM_POINT("Calling hat_kern_alloc()..."); 1931 hat_kern_alloc((caddr_t)segmap_start, segmapsize, ekernelheap); 1932 PRM_POINT("hat_kern_alloc() done"); 1933 1934 #ifndef __xpv 1935 /* 1936 * Setup Page Attribute Table 1937 */ 1938 pat_sync(); 1939 #endif 1940 1941 /* 1942 * The next two loops are done in distinct steps in order 1943 * to be sure that any page that is doubly mapped (both above 1944 * KERNEL_TEXT and below kernelbase) is dealt with correctly. 1945 * Note this may never happen, but it might someday. 1946 */ 1947 bootpages = NULL; 1948 PRM_POINT("Protecting boot pages"); 1949 1950 /* 1951 * Protect any pages mapped above KERNEL_TEXT that somehow have 1952 * page_t's. This can only happen if something weird allocated 1953 * in this range (like kadb/kmdb). 1954 */ 1955 protect_boot_range(KERNEL_TEXT, (uintptr_t)-1, 0); 1956 1957 /* 1958 * Before we can take over memory allocation/mapping from the boot 1959 * loader we must remove from our free page lists any boot allocated 1960 * pages that stay mapped until release_bootstrap(). 1961 */ 1962 protect_boot_range(0, kernelbase, 1); 1963 1964 1965 /* 1966 * Switch to running on regular HAT (not boot_mmu) 1967 */ 1968 PRM_POINT("Calling hat_kern_setup()..."); 1969 hat_kern_setup(); 1970 1971 /* 1972 * It is no longer safe to call BOP_ALLOC(), so make sure we don't. 1973 */ 1974 bop_no_more_mem(); 1975 1976 PRM_POINT("hat_kern_setup() done"); 1977 1978 hat_cpu_online(CPU); 1979 1980 /* 1981 * Initialize VM system 1982 */ 1983 PRM_POINT("Calling kvm_init()..."); 1984 kvm_init(); 1985 PRM_POINT("kvm_init() done"); 1986 1987 /* 1988 * Tell kmdb that the VM system is now working 1989 */ 1990 if (boothowto & RB_DEBUG) 1991 kdi_dvec_vmready(); 1992 1993 #if defined(__xpv) 1994 /* 1995 * Populate the I/O pool on domain 0 1996 */ 1997 if (DOMAIN_IS_INITDOMAIN(xen_info)) { 1998 extern long populate_io_pool(void); 1999 long init_io_pool_cnt; 2000 2001 PRM_POINT("Populating reserve I/O page pool"); 2002 init_io_pool_cnt = populate_io_pool(); 2003 PRM_DEBUG(init_io_pool_cnt); 2004 } 2005 #endif 2006 /* 2007 * Mangle the brand string etc. 2008 */ 2009 cpuid_pass3(CPU); 2010 2011 #if defined(__amd64) 2012 2013 /* 2014 * Create the device arena for toxic (to dtrace/kmdb) mappings. 2015 */ 2016 device_arena = vmem_create("device", (void *)toxic_addr, 2017 toxic_size, MMU_PAGESIZE, NULL, NULL, NULL, 0, VM_SLEEP); 2018 2019 #else /* __i386 */ 2020 2021 /* 2022 * allocate the bit map that tracks toxic pages 2023 */ 2024 toxic_bit_map_len = btop((ulong_t)(valloc_base - kernelbase)); 2025 PRM_DEBUG(toxic_bit_map_len); 2026 toxic_bit_map = 2027 kmem_zalloc(BT_SIZEOFMAP(toxic_bit_map_len), KM_NOSLEEP); 2028 ASSERT(toxic_bit_map != NULL); 2029 PRM_DEBUG(toxic_bit_map); 2030 2031 #endif /* __i386 */ 2032 2033 2034 /* 2035 * Now that we've got more VA, as well as the ability to allocate from 2036 * it, tell the debugger. 2037 */ 2038 if (boothowto & RB_DEBUG) 2039 kdi_dvec_memavail(); 2040 2041 /* 2042 * The following code installs a special page fault handler (#pf) 2043 * to work around a pentium bug. 2044 */ 2045 #if !defined(__amd64) && !defined(__xpv) 2046 if (x86_type == X86_TYPE_P5) { 2047 desctbr_t idtr; 2048 gate_desc_t *newidt; 2049 2050 if ((newidt = kmem_zalloc(MMU_PAGESIZE, KM_NOSLEEP)) == NULL) 2051 panic("failed to install pentium_pftrap"); 2052 2053 bcopy(idt0, newidt, NIDT * sizeof (*idt0)); 2054 set_gatesegd(&newidt[T_PGFLT], &pentium_pftrap, 2055 KCS_SEL, SDT_SYSIGT, TRP_KPL, 0); 2056 2057 (void) as_setprot(&kas, (caddr_t)newidt, MMU_PAGESIZE, 2058 PROT_READ | PROT_EXEC); 2059 2060 CPU->cpu_idt = newidt; 2061 idtr.dtr_base = (uintptr_t)CPU->cpu_idt; 2062 idtr.dtr_limit = (NIDT * sizeof (*idt0)) - 1; 2063 wr_idtr(&idtr); 2064 } 2065 #endif /* !__amd64 */ 2066 2067 #if !defined(__xpv) 2068 /* 2069 * Map page pfn=0 for drivers, such as kd, that need to pick up 2070 * parameters left there by controllers/BIOS. 2071 */ 2072 PRM_POINT("setup up p0_va"); 2073 p0_va = i86devmap(0, 1, PROT_READ); 2074 PRM_DEBUG(p0_va); 2075 #endif 2076 2077 cmn_err(CE_CONT, "?mem = %luK (0x%lx)\n", 2078 physinstalled << (MMU_PAGESHIFT - 10), ptob(physinstalled)); 2079 2080 /* 2081 * disable automatic large pages for small memory systems or 2082 * when the disable flag is set. 2083 * 2084 * Do not yet consider page sizes larger than 2m/4m. 2085 */ 2086 if (!auto_lpg_disable && mmu.max_page_level > 0) { 2087 max_uheap_lpsize = LEVEL_SIZE(1); 2088 max_ustack_lpsize = LEVEL_SIZE(1); 2089 max_privmap_lpsize = LEVEL_SIZE(1); 2090 max_uidata_lpsize = LEVEL_SIZE(1); 2091 max_utext_lpsize = LEVEL_SIZE(1); 2092 max_shm_lpsize = LEVEL_SIZE(1); 2093 } 2094 if (physmem < privm_lpg_min_physmem || mmu.max_page_level == 0 || 2095 auto_lpg_disable) { 2096 use_brk_lpg = 0; 2097 use_stk_lpg = 0; 2098 } 2099 mcntl0_lpsize = LEVEL_SIZE(mmu.umax_page_level); 2100 2101 PRM_POINT("Calling hat_init_finish()..."); 2102 hat_init_finish(); 2103 PRM_POINT("hat_init_finish() done"); 2104 2105 /* 2106 * Initialize the segkp segment type. 2107 */ 2108 rw_enter(&kas.a_lock, RW_WRITER); 2109 PRM_POINT("Attaching segkp"); 2110 if (segkp_fromheap) { 2111 segkp->s_as = &kas; 2112 } else if (seg_attach(&kas, (caddr_t)segkp_base, mmu_ptob(segkpsize), 2113 segkp) < 0) { 2114 panic("startup: cannot attach segkp"); 2115 /*NOTREACHED*/ 2116 } 2117 PRM_POINT("Doing segkp_create()"); 2118 if (segkp_create(segkp) != 0) { 2119 panic("startup: segkp_create failed"); 2120 /*NOTREACHED*/ 2121 } 2122 PRM_DEBUG(segkp); 2123 rw_exit(&kas.a_lock); 2124 2125 /* 2126 * kpm segment 2127 */ 2128 segmap_kpm = 0; 2129 if (kpm_desired) { 2130 kpm_init(); 2131 kpm_enable = 1; 2132 } 2133 2134 /* 2135 * Now create segmap segment. 2136 */ 2137 rw_enter(&kas.a_lock, RW_WRITER); 2138 if (seg_attach(&kas, (caddr_t)segmap_start, segmapsize, segmap) < 0) { 2139 panic("cannot attach segmap"); 2140 /*NOTREACHED*/ 2141 } 2142 PRM_DEBUG(segmap); 2143 2144 a.prot = PROT_READ | PROT_WRITE; 2145 a.shmsize = 0; 2146 a.nfreelist = segmapfreelists; 2147 2148 if (segmap_create(segmap, (caddr_t)&a) != 0) 2149 panic("segmap_create segmap"); 2150 rw_exit(&kas.a_lock); 2151 2152 setup_vaddr_for_ppcopy(CPU); 2153 2154 segdev_init(); 2155 #if defined(__xpv) 2156 if (DOMAIN_IS_INITDOMAIN(xen_info)) 2157 #endif 2158 pmem_init(); 2159 2160 PRM_POINT("startup_vm() done"); 2161 } 2162 2163 /* 2164 * Load a tod module for the non-standard tod part found on this system. 2165 */ 2166 static void 2167 load_tod_module(char *todmod) 2168 { 2169 if (modload("tod", todmod) == -1) 2170 halt("Can't load TOD module"); 2171 } 2172 2173 static void 2174 startup_end(void) 2175 { 2176 int i; 2177 extern void setx86isalist(void); 2178 extern void cpu_event_init(void); 2179 2180 PRM_POINT("startup_end() starting..."); 2181 2182 /* 2183 * Perform tasks that get done after most of the VM 2184 * initialization has been done but before the clock 2185 * and other devices get started. 2186 */ 2187 kern_setup1(); 2188 2189 /* 2190 * Perform CPC initialization for this CPU. 2191 */ 2192 kcpc_hw_init(CPU); 2193 2194 /* 2195 * Initialize cpu event framework. 2196 */ 2197 cpu_event_init(); 2198 2199 #if defined(OPTERON_WORKAROUND_6323525) 2200 if (opteron_workaround_6323525) 2201 patch_workaround_6323525(); 2202 #endif 2203 /* 2204 * If needed, load TOD module now so that ddi_get_time(9F) etc. work 2205 * (For now, "needed" is defined as set tod_module_name in /etc/system) 2206 */ 2207 if (tod_module_name != NULL) { 2208 PRM_POINT("load_tod_module()"); 2209 load_tod_module(tod_module_name); 2210 } 2211 2212 #if defined(__xpv) 2213 /* 2214 * Forceload interposing TOD module for the hypervisor. 2215 */ 2216 PRM_POINT("load_tod_module()"); 2217 load_tod_module("xpvtod"); 2218 #endif 2219 2220 /* 2221 * Configure the system. 2222 */ 2223 PRM_POINT("Calling configure()..."); 2224 configure(); /* set up devices */ 2225 PRM_POINT("configure() done"); 2226 2227 /* 2228 * We can now setup for XSAVE because fpu_probe is done in configure(). 2229 */ 2230 if (fp_save_mech == FP_XSAVE) { 2231 xsave_setup_msr(CPU); 2232 } 2233 2234 /* 2235 * Set the isa_list string to the defined instruction sets we 2236 * support. 2237 */ 2238 setx86isalist(); 2239 cpu_intr_alloc(CPU, NINTR_THREADS); 2240 psm_install(); 2241 2242 /* 2243 * We're done with bootops. We don't unmap the bootstrap yet because 2244 * we're still using bootsvcs. 2245 */ 2246 PRM_POINT("NULLing out bootops"); 2247 *bootopsp = (struct bootops *)NULL; 2248 bootops = (struct bootops *)NULL; 2249 2250 #if defined(__xpv) 2251 ec_init_debug_irq(); 2252 xs_domu_init(); 2253 #endif 2254 2255 #if defined(__amd64) && !defined(__xpv) 2256 /* 2257 * Intel IOMMU has been setup/initialized in ddi_impl.c 2258 * Start it up now. 2259 */ 2260 immu_startup(); 2261 #endif 2262 2263 PRM_POINT("Enabling interrupts"); 2264 (*picinitf)(); 2265 sti(); 2266 #if defined(__xpv) 2267 ASSERT(CPU->cpu_m.mcpu_vcpu_info->evtchn_upcall_mask == 0); 2268 xen_late_startup(); 2269 #endif 2270 2271 (void) add_avsoftintr((void *)&softlevel1_hdl, 1, softlevel1, 2272 "softlevel1", NULL, NULL); /* XXX to be moved later */ 2273 2274 /* 2275 * Register software interrupt handlers for ddi_periodic_add(9F). 2276 * Software interrupts up to the level 10 are supported. 2277 */ 2278 for (i = DDI_IPL_1; i <= DDI_IPL_10; i++) { 2279 (void) add_avsoftintr((void *)&softlevel_hdl[i-1], i, 2280 (avfunc)ddi_periodic_softintr, "ddi_periodic", 2281 (caddr_t)(uintptr_t)i, NULL); 2282 } 2283 2284 #if !defined(__xpv) 2285 if (modload("drv", "amd_iommu") < 0) { 2286 PRM_POINT("No AMD IOMMU present\n"); 2287 } else if (ddi_hold_installed_driver(ddi_name_to_major( 2288 "amd_iommu")) == NULL) { 2289 prom_printf("ERROR: failed to attach AMD IOMMU\n"); 2290 } 2291 #endif 2292 post_startup_cpu_fixups(); 2293 2294 PRM_POINT("startup_end() done"); 2295 } 2296 2297 /* 2298 * Don't remove the following 2 variables. They are necessary 2299 * for reading the hostid from the legacy file (/kernel/misc/sysinit). 2300 */ 2301 char *_hs1107 = hw_serial; 2302 ulong_t _bdhs34; 2303 2304 void 2305 post_startup(void) 2306 { 2307 extern void cpupm_init(cpu_t *); 2308 extern void cpu_event_init_cpu(cpu_t *); 2309 2310 /* 2311 * Set the system wide, processor-specific flags to be passed 2312 * to userland via the aux vector for performance hints and 2313 * instruction set extensions. 2314 */ 2315 bind_hwcap(); 2316 2317 #ifdef __xpv 2318 if (DOMAIN_IS_INITDOMAIN(xen_info)) 2319 #endif 2320 { 2321 #if defined(__xpv) 2322 xpv_panic_init(); 2323 #else 2324 /* 2325 * Startup the memory scrubber. 2326 * XXPV This should be running somewhere .. 2327 */ 2328 if ((get_hwenv() & HW_VIRTUAL) == 0) 2329 memscrub_init(); 2330 #endif 2331 } 2332 2333 /* 2334 * Complete CPU module initialization 2335 */ 2336 cmi_post_startup(); 2337 2338 /* 2339 * Perform forceloading tasks for /etc/system. 2340 */ 2341 (void) mod_sysctl(SYS_FORCELOAD, NULL); 2342 2343 /* 2344 * ON4.0: Force /proc module in until clock interrupt handle fixed 2345 * ON4.0: This must be fixed or restated in /etc/systems. 2346 */ 2347 (void) modload("fs", "procfs"); 2348 2349 (void) i_ddi_attach_hw_nodes("pit_beep"); 2350 2351 #if defined(__i386) 2352 /* 2353 * Check for required functional Floating Point hardware, 2354 * unless FP hardware explicitly disabled. 2355 */ 2356 if (fpu_exists && (fpu_pentium_fdivbug || fp_kind == FP_NO)) 2357 halt("No working FP hardware found"); 2358 #endif 2359 2360 maxmem = freemem; 2361 2362 cpu_event_init_cpu(CPU); 2363 cpupm_init(CPU); 2364 (void) mach_cpu_create_device_node(CPU, NULL); 2365 2366 pg_init(); 2367 } 2368 2369 static int 2370 pp_in_range(page_t *pp, uint64_t low_addr, uint64_t high_addr) 2371 { 2372 return ((pp->p_pagenum >= btop(low_addr)) && 2373 (pp->p_pagenum < btopr(high_addr))); 2374 } 2375 2376 static int 2377 pp_in_module(page_t *pp, const rd_existing_t *modranges) 2378 { 2379 uint_t i; 2380 2381 for (i = 0; modranges[i].phys != 0; i++) { 2382 if (pp_in_range(pp, modranges[i].phys, 2383 modranges[i].phys + modranges[i].size)) 2384 return (1); 2385 } 2386 2387 return (0); 2388 } 2389 2390 void 2391 release_bootstrap(void) 2392 { 2393 int root_is_ramdisk; 2394 page_t *pp; 2395 extern void kobj_boot_unmountroot(void); 2396 extern dev_t rootdev; 2397 uint_t i; 2398 char propname[32]; 2399 rd_existing_t *modranges; 2400 #if !defined(__xpv) 2401 pfn_t pfn; 2402 #endif 2403 2404 /* 2405 * Save the bootfs module ranges so that we can reserve them below 2406 * for the real bootfs. 2407 */ 2408 modranges = kmem_alloc(sizeof (rd_existing_t) * MAX_BOOT_MODULES, 2409 KM_SLEEP); 2410 for (i = 0; ; i++) { 2411 uint64_t start, size; 2412 2413 modranges[i].phys = 0; 2414 2415 (void) snprintf(propname, sizeof (propname), 2416 "module-addr-%u", i); 2417 if (do_bsys_getproplen(NULL, propname) <= 0) 2418 break; 2419 (void) do_bsys_getprop(NULL, propname, &start); 2420 2421 (void) snprintf(propname, sizeof (propname), 2422 "module-size-%u", i); 2423 if (do_bsys_getproplen(NULL, propname) <= 0) 2424 break; 2425 (void) do_bsys_getprop(NULL, propname, &size); 2426 2427 modranges[i].phys = start; 2428 modranges[i].size = size; 2429 } 2430 2431 /* unmount boot ramdisk and release kmem usage */ 2432 kobj_boot_unmountroot(); 2433 2434 /* 2435 * We're finished using the boot loader so free its pages. 2436 */ 2437 PRM_POINT("Unmapping lower boot pages"); 2438 2439 clear_boot_mappings(0, _userlimit); 2440 2441 postbootkernelbase = kernelbase; 2442 2443 /* 2444 * If root isn't on ramdisk, destroy the hardcoded 2445 * ramdisk node now and release the memory. Else, 2446 * ramdisk memory is kept in rd_pages. 2447 */ 2448 root_is_ramdisk = (getmajor(rootdev) == ddi_name_to_major("ramdisk")); 2449 if (!root_is_ramdisk) { 2450 dev_info_t *dip = ddi_find_devinfo("ramdisk", -1, 0); 2451 ASSERT(dip && ddi_get_parent(dip) == ddi_root_node()); 2452 ndi_rele_devi(dip); /* held from ddi_find_devinfo */ 2453 (void) ddi_remove_child(dip, 0); 2454 } 2455 2456 PRM_POINT("Releasing boot pages"); 2457 while (bootpages) { 2458 extern uint64_t ramdisk_start, ramdisk_end; 2459 pp = bootpages; 2460 bootpages = pp->p_next; 2461 2462 2463 /* Keep pages for the lower 64K */ 2464 if (pp_in_range(pp, 0, 0x40000)) { 2465 pp->p_next = lower_pages; 2466 lower_pages = pp; 2467 lower_pages_count++; 2468 continue; 2469 } 2470 2471 if (root_is_ramdisk && pp_in_range(pp, ramdisk_start, 2472 ramdisk_end) || pp_in_module(pp, modranges)) { 2473 pp->p_next = rd_pages; 2474 rd_pages = pp; 2475 continue; 2476 } 2477 pp->p_next = (struct page *)0; 2478 pp->p_prev = (struct page *)0; 2479 PP_CLRBOOTPAGES(pp); 2480 page_free(pp, 1); 2481 } 2482 PRM_POINT("Boot pages released"); 2483 2484 kmem_free(modranges, sizeof (rd_existing_t) * 99); 2485 2486 #if !defined(__xpv) 2487 /* XXPV -- note this following bunch of code needs to be revisited in Xen 3.0 */ 2488 /* 2489 * Find 1 page below 1 MB so that other processors can boot up or 2490 * so that any processor can resume. 2491 * Make sure it has a kernel VA as well as a 1:1 mapping. 2492 * We should have just free'd one up. 2493 */ 2494 2495 /* 2496 * 0x10 pages is 64K. Leave the bottom 64K alone 2497 * for BIOS. 2498 */ 2499 for (pfn = 0x10; pfn < btop(1*1024*1024); pfn++) { 2500 if (page_numtopp_alloc(pfn) == NULL) 2501 continue; 2502 rm_platter_va = i86devmap(pfn, 1, 2503 PROT_READ | PROT_WRITE | PROT_EXEC); 2504 rm_platter_pa = ptob(pfn); 2505 break; 2506 } 2507 if (pfn == btop(1*1024*1024) && use_mp) 2508 panic("No page below 1M available for starting " 2509 "other processors or for resuming from system-suspend"); 2510 #endif /* !__xpv */ 2511 } 2512 2513 /* 2514 * Initialize the platform-specific parts of a page_t. 2515 */ 2516 void 2517 add_physmem_cb(page_t *pp, pfn_t pnum) 2518 { 2519 pp->p_pagenum = pnum; 2520 pp->p_mapping = NULL; 2521 pp->p_embed = 0; 2522 pp->p_share = 0; 2523 pp->p_mlentry = 0; 2524 } 2525 2526 /* 2527 * kphysm_init() initializes physical memory. 2528 */ 2529 static pgcnt_t 2530 kphysm_init( 2531 page_t *pp, 2532 pgcnt_t npages) 2533 { 2534 struct memlist *pmem; 2535 struct memseg *cur_memseg; 2536 pfn_t base_pfn; 2537 pfn_t end_pfn; 2538 pgcnt_t num; 2539 pgcnt_t pages_done = 0; 2540 uint64_t addr; 2541 uint64_t size; 2542 extern pfn_t ddiphysmin; 2543 extern int mnode_xwa; 2544 int ms = 0, me = 0; 2545 2546 ASSERT(page_hash != NULL && page_hashsz != 0); 2547 2548 cur_memseg = memseg_base; 2549 for (pmem = phys_avail; pmem && npages; pmem = pmem->ml_next) { 2550 /* 2551 * In a 32 bit kernel can't use higher memory if we're 2552 * not booting in PAE mode. This check takes care of that. 2553 */ 2554 addr = pmem->ml_address; 2555 size = pmem->ml_size; 2556 if (btop(addr) > physmax) 2557 continue; 2558 2559 /* 2560 * align addr and size - they may not be at page boundaries 2561 */ 2562 if ((addr & MMU_PAGEOFFSET) != 0) { 2563 addr += MMU_PAGEOFFSET; 2564 addr &= ~(uint64_t)MMU_PAGEOFFSET; 2565 size -= addr - pmem->ml_address; 2566 } 2567 2568 /* only process pages below or equal to physmax */ 2569 if ((btop(addr + size) - 1) > physmax) 2570 size = ptob(physmax - btop(addr) + 1); 2571 2572 num = btop(size); 2573 if (num == 0) 2574 continue; 2575 2576 if (num > npages) 2577 num = npages; 2578 2579 npages -= num; 2580 pages_done += num; 2581 base_pfn = btop(addr); 2582 2583 if (prom_debug) 2584 prom_printf("MEMSEG addr=0x%" PRIx64 2585 " pgs=0x%lx pfn 0x%lx-0x%lx\n", 2586 addr, num, base_pfn, base_pfn + num); 2587 2588 /* 2589 * Ignore pages below ddiphysmin to simplify ddi memory 2590 * allocation with non-zero addr_lo requests. 2591 */ 2592 if (base_pfn < ddiphysmin) { 2593 if (base_pfn + num <= ddiphysmin) 2594 continue; 2595 pp += (ddiphysmin - base_pfn); 2596 num -= (ddiphysmin - base_pfn); 2597 base_pfn = ddiphysmin; 2598 } 2599 2600 /* 2601 * mnode_xwa is greater than 1 when large pages regions can 2602 * cross memory node boundaries. To prevent the formation 2603 * of these large pages, configure the memsegs based on the 2604 * memory node ranges which had been made non-contiguous. 2605 */ 2606 if (mnode_xwa > 1) { 2607 2608 end_pfn = base_pfn + num - 1; 2609 ms = PFN_2_MEM_NODE(base_pfn); 2610 me = PFN_2_MEM_NODE(end_pfn); 2611 2612 if (ms != me) { 2613 /* 2614 * current range spans more than 1 memory node. 2615 * Set num to only the pfn range in the start 2616 * memory node. 2617 */ 2618 num = mem_node_config[ms].physmax - base_pfn 2619 + 1; 2620 ASSERT(end_pfn > mem_node_config[ms].physmax); 2621 } 2622 } 2623 2624 for (;;) { 2625 /* 2626 * Build the memsegs entry 2627 */ 2628 cur_memseg->pages = pp; 2629 cur_memseg->epages = pp + num; 2630 cur_memseg->pages_base = base_pfn; 2631 cur_memseg->pages_end = base_pfn + num; 2632 2633 /* 2634 * Insert into memseg list in decreasing pfn range 2635 * order. Low memory is typically more fragmented such 2636 * that this ordering keeps the larger ranges at the 2637 * front of the list for code that searches memseg. 2638 * This ASSERTS that the memsegs coming in from boot 2639 * are in increasing physical address order and not 2640 * contiguous. 2641 */ 2642 if (memsegs != NULL) { 2643 ASSERT(cur_memseg->pages_base >= 2644 memsegs->pages_end); 2645 cur_memseg->next = memsegs; 2646 } 2647 memsegs = cur_memseg; 2648 2649 /* 2650 * add_physmem() initializes the PSM part of the page 2651 * struct by calling the PSM back with add_physmem_cb(). 2652 * In addition it coalesces pages into larger pages as 2653 * it initializes them. 2654 */ 2655 add_physmem(pp, num, base_pfn); 2656 cur_memseg++; 2657 availrmem_initial += num; 2658 availrmem += num; 2659 2660 pp += num; 2661 if (ms >= me) 2662 break; 2663 2664 /* process next memory node range */ 2665 ms++; 2666 base_pfn = mem_node_config[ms].physbase; 2667 num = MIN(mem_node_config[ms].physmax, 2668 end_pfn) - base_pfn + 1; 2669 } 2670 } 2671 2672 PRM_DEBUG(availrmem_initial); 2673 PRM_DEBUG(availrmem); 2674 PRM_DEBUG(freemem); 2675 build_pfn_hash(); 2676 return (pages_done); 2677 } 2678 2679 /* 2680 * Kernel VM initialization. 2681 */ 2682 static void 2683 kvm_init(void) 2684 { 2685 ASSERT((((uintptr_t)s_text) & MMU_PAGEOFFSET) == 0); 2686 2687 /* 2688 * Put the kernel segments in kernel address space. 2689 */ 2690 rw_enter(&kas.a_lock, RW_WRITER); 2691 as_avlinit(&kas); 2692 2693 (void) seg_attach(&kas, s_text, e_moddata - s_text, &ktextseg); 2694 (void) segkmem_create(&ktextseg); 2695 2696 (void) seg_attach(&kas, (caddr_t)valloc_base, valloc_sz, &kvalloc); 2697 (void) segkmem_create(&kvalloc); 2698 2699 (void) seg_attach(&kas, kernelheap, 2700 ekernelheap - kernelheap, &kvseg); 2701 (void) segkmem_create(&kvseg); 2702 2703 if (core_size > 0) { 2704 PRM_POINT("attaching kvseg_core"); 2705 (void) seg_attach(&kas, (caddr_t)core_base, core_size, 2706 &kvseg_core); 2707 (void) segkmem_create(&kvseg_core); 2708 } 2709 2710 if (segziosize > 0) { 2711 PRM_POINT("attaching segzio"); 2712 (void) seg_attach(&kas, segzio_base, mmu_ptob(segziosize), 2713 &kzioseg); 2714 (void) segkmem_zio_create(&kzioseg); 2715 2716 /* create zio area covering new segment */ 2717 segkmem_zio_init(segzio_base, mmu_ptob(segziosize)); 2718 } 2719 2720 (void) seg_attach(&kas, kdi_segdebugbase, kdi_segdebugsize, &kdebugseg); 2721 (void) segkmem_create(&kdebugseg); 2722 2723 rw_exit(&kas.a_lock); 2724 2725 /* 2726 * Ensure that the red zone at kernelbase is never accessible. 2727 */ 2728 PRM_POINT("protecting redzone"); 2729 (void) as_setprot(&kas, (caddr_t)kernelbase, KERNEL_REDZONE_SIZE, 0); 2730 2731 /* 2732 * Make the text writable so that it can be hot patched by DTrace. 2733 */ 2734 (void) as_setprot(&kas, s_text, e_modtext - s_text, 2735 PROT_READ | PROT_WRITE | PROT_EXEC); 2736 2737 /* 2738 * Make data writable until end. 2739 */ 2740 (void) as_setprot(&kas, s_data, e_moddata - s_data, 2741 PROT_READ | PROT_WRITE | PROT_EXEC); 2742 } 2743 2744 #ifndef __xpv 2745 /* 2746 * Solaris adds an entry for Write Combining caching to the PAT 2747 */ 2748 static uint64_t pat_attr_reg = PAT_DEFAULT_ATTRIBUTE; 2749 2750 void 2751 pat_sync(void) 2752 { 2753 ulong_t cr0, cr0_orig, cr4; 2754 2755 if (!is_x86_feature(x86_featureset, X86FSET_PAT)) 2756 return; 2757 cr0_orig = cr0 = getcr0(); 2758 cr4 = getcr4(); 2759 2760 /* disable caching and flush all caches and TLBs */ 2761 cr0 |= CR0_CD; 2762 cr0 &= ~CR0_NW; 2763 setcr0(cr0); 2764 invalidate_cache(); 2765 if (cr4 & CR4_PGE) { 2766 setcr4(cr4 & ~(ulong_t)CR4_PGE); 2767 setcr4(cr4); 2768 } else { 2769 reload_cr3(); 2770 } 2771 2772 /* add our entry to the PAT */ 2773 wrmsr(REG_PAT, pat_attr_reg); 2774 2775 /* flush TLBs and cache again, then reenable cr0 caching */ 2776 if (cr4 & CR4_PGE) { 2777 setcr4(cr4 & ~(ulong_t)CR4_PGE); 2778 setcr4(cr4); 2779 } else { 2780 reload_cr3(); 2781 } 2782 invalidate_cache(); 2783 setcr0(cr0_orig); 2784 } 2785 2786 #endif /* !__xpv */ 2787 2788 #if defined(_SOFT_HOSTID) 2789 /* 2790 * On platforms that do not have a hardware serial number, attempt 2791 * to set one based on the contents of /etc/hostid. If this file does 2792 * not exist, assume that we are to generate a new hostid and set 2793 * it in the kernel, for subsequent saving by a userland process 2794 * once the system is up and the root filesystem is mounted r/w. 2795 * 2796 * In order to gracefully support upgrade on OpenSolaris, if 2797 * /etc/hostid does not exist, we will attempt to get a serial number 2798 * using the legacy method (/kernel/misc/sysinit). 2799 * 2800 * If that isn't present, we attempt to use an SMBIOS UUID, which is 2801 * a hardware serial number. Note that we don't automatically trust 2802 * all SMBIOS UUIDs (some older platforms are defective and ship duplicate 2803 * UUIDs in violation of the standard), we check against a blacklist. 2804 * 2805 * In an attempt to make the hostid less prone to abuse 2806 * (for license circumvention, etc), we store it in /etc/hostid 2807 * in rot47 format. 2808 */ 2809 extern volatile unsigned long tenmicrodata; 2810 static int atoi(char *); 2811 2812 /* 2813 * Set this to non-zero in /etc/system if you think your SMBIOS returns a 2814 * UUID that is not unique. (Also report it so that the smbios_uuid_blacklist 2815 * array can be updated.) 2816 */ 2817 int smbios_broken_uuid = 0; 2818 2819 /* 2820 * List of known bad UUIDs. This is just the lower 32-bit values, since 2821 * that's what we use for the host id. If your hostid falls here, you need 2822 * to contact your hardware OEM for a fix for your BIOS. 2823 */ 2824 static unsigned char 2825 smbios_uuid_blacklist[][16] = { 2826 2827 { /* Reported bad UUID (Google search) */ 2828 0x00, 0x02, 0x00, 0x03, 0x00, 0x04, 0x00, 0x05, 2829 0x00, 0x06, 0x00, 0x07, 0x00, 0x08, 0x00, 0x09, 2830 }, 2831 { /* Known bad DELL UUID */ 2832 0x4C, 0x4C, 0x45, 0x44, 0x00, 0x00, 0x20, 0x10, 2833 0x80, 0x20, 0x80, 0xC0, 0x4F, 0x20, 0x20, 0x20, 2834 }, 2835 { /* Uninitialized flash */ 2836 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 2837 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff 2838 }, 2839 { /* All zeros */ 2840 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 2841 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00 2842 }, 2843 }; 2844 2845 static int32_t 2846 uuid_to_hostid(const uint8_t *uuid) 2847 { 2848 /* 2849 * Although the UUIDs are 128-bits, they may not distribute entropy 2850 * evenly. We would like to use SHA or MD5, but those are located 2851 * in loadable modules and not available this early in boot. As we 2852 * don't need the values to be cryptographically strong, we just 2853 * generate 32-bit vaue by xor'ing the various sequences together, 2854 * which ensures that the entire UUID contributes to the hostid. 2855 */ 2856 uint32_t id = 0; 2857 2858 /* first check against the blacklist */ 2859 for (int i = 0; i < (sizeof (smbios_uuid_blacklist) / 16); i++) { 2860 if (bcmp(smbios_uuid_blacklist[0], uuid, 16) == 0) { 2861 cmn_err(CE_CONT, "?Broken SMBIOS UUID. " 2862 "Contact BIOS manufacturer for repair.\n"); 2863 return ((int32_t)HW_INVALID_HOSTID); 2864 } 2865 } 2866 2867 for (int i = 0; i < 16; i++) 2868 id ^= ((uuid[i]) << (8 * (i % sizeof (id)))); 2869 2870 /* Make sure return value is positive */ 2871 return (id & 0x7fffffff); 2872 } 2873 2874 static int32_t 2875 set_soft_hostid(void) 2876 { 2877 struct _buf *file; 2878 char tokbuf[MAXNAMELEN]; 2879 token_t token; 2880 int done = 0; 2881 u_longlong_t tmp; 2882 int i; 2883 int32_t hostid = (int32_t)HW_INVALID_HOSTID; 2884 unsigned char *c; 2885 hrtime_t tsc; 2886 smbios_system_t smsys; 2887 2888 /* 2889 * If /etc/hostid file not found, we'd like to get a pseudo 2890 * random number to use at the hostid. A nice way to do this 2891 * is to read the real time clock. To remain xen-compatible, 2892 * we can't poke the real hardware, so we use tsc_read() to 2893 * read the real time clock. However, there is an ominous 2894 * warning in tsc_read that says it can return zero, so we 2895 * deal with that possibility by falling back to using the 2896 * (hopefully random enough) value in tenmicrodata. 2897 */ 2898 2899 if ((file = kobj_open_file(hostid_file)) == (struct _buf *)-1) { 2900 /* 2901 * hostid file not found - try to load sysinit module 2902 * and see if it has a nonzero hostid value...use that 2903 * instead of generating a new hostid here if so. 2904 */ 2905 if ((i = modload("misc", "sysinit")) != -1) { 2906 if (strlen(hw_serial) > 0) 2907 hostid = (int32_t)atoi(hw_serial); 2908 (void) modunload(i); 2909 } 2910 2911 /* 2912 * We try to use the SMBIOS UUID. But not if it is blacklisted 2913 * in /etc/system. 2914 */ 2915 if ((hostid == HW_INVALID_HOSTID) && 2916 (smbios_broken_uuid == 0) && 2917 (ksmbios != NULL) && 2918 (smbios_info_system(ksmbios, &smsys) != SMB_ERR) && 2919 (smsys.smbs_uuidlen >= 16)) { 2920 hostid = uuid_to_hostid(smsys.smbs_uuid); 2921 } 2922 2923 /* 2924 * Generate a "random" hostid using the clock. These 2925 * hostids will change on each boot if the value is not 2926 * saved to a persistent /etc/hostid file. 2927 */ 2928 if (hostid == HW_INVALID_HOSTID) { 2929 tsc = tsc_read(); 2930 if (tsc == 0) /* tsc_read can return zero sometimes */ 2931 hostid = (int32_t)tenmicrodata & 0x0CFFFFF; 2932 else 2933 hostid = (int32_t)tsc & 0x0CFFFFF; 2934 } 2935 } else { 2936 /* hostid file found */ 2937 while (!done) { 2938 token = kobj_lex(file, tokbuf, sizeof (tokbuf)); 2939 2940 switch (token) { 2941 case POUND: 2942 /* 2943 * skip comments 2944 */ 2945 kobj_find_eol(file); 2946 break; 2947 case STRING: 2948 /* 2949 * un-rot47 - obviously this 2950 * nonsense is ascii-specific 2951 */ 2952 for (c = (unsigned char *)tokbuf; 2953 *c != '\0'; c++) { 2954 *c += 47; 2955 if (*c > '~') 2956 *c -= 94; 2957 else if (*c < '!') 2958 *c += 94; 2959 } 2960 /* 2961 * now we should have a real number 2962 */ 2963 2964 if (kobj_getvalue(tokbuf, &tmp) != 0) 2965 kobj_file_err(CE_WARN, file, 2966 "Bad value %s for hostid", 2967 tokbuf); 2968 else 2969 hostid = (int32_t)tmp; 2970 2971 break; 2972 case EOF: 2973 done = 1; 2974 /* FALLTHROUGH */ 2975 case NEWLINE: 2976 kobj_newline(file); 2977 break; 2978 default: 2979 break; 2980 2981 } 2982 } 2983 if (hostid == HW_INVALID_HOSTID) /* didn't find a hostid */ 2984 kobj_file_err(CE_WARN, file, 2985 "hostid missing or corrupt"); 2986 2987 kobj_close_file(file); 2988 } 2989 /* 2990 * hostid is now the value read from /etc/hostid, or the 2991 * new hostid we generated in this routine or HW_INVALID_HOSTID if not 2992 * set. 2993 */ 2994 return (hostid); 2995 } 2996 2997 static int 2998 atoi(char *p) 2999 { 3000 int i = 0; 3001 3002 while (*p != '\0') 3003 i = 10 * i + (*p++ - '0'); 3004 3005 return (i); 3006 } 3007 3008 #endif /* _SOFT_HOSTID */ 3009 3010 void 3011 get_system_configuration(void) 3012 { 3013 char prop[32]; 3014 u_longlong_t nodes_ll, cpus_pernode_ll, lvalue; 3015 3016 if (BOP_GETPROPLEN(bootops, "nodes") > sizeof (prop) || 3017 BOP_GETPROP(bootops, "nodes", prop) < 0 || 3018 kobj_getvalue(prop, &nodes_ll) == -1 || 3019 nodes_ll > MAXNODES || 3020 BOP_GETPROPLEN(bootops, "cpus_pernode") > sizeof (prop) || 3021 BOP_GETPROP(bootops, "cpus_pernode", prop) < 0 || 3022 kobj_getvalue(prop, &cpus_pernode_ll) == -1) { 3023 system_hardware.hd_nodes = 1; 3024 system_hardware.hd_cpus_per_node = 0; 3025 } else { 3026 system_hardware.hd_nodes = (int)nodes_ll; 3027 system_hardware.hd_cpus_per_node = (int)cpus_pernode_ll; 3028 } 3029 3030 if (BOP_GETPROPLEN(bootops, "kernelbase") > sizeof (prop) || 3031 BOP_GETPROP(bootops, "kernelbase", prop) < 0 || 3032 kobj_getvalue(prop, &lvalue) == -1) 3033 eprom_kernelbase = NULL; 3034 else 3035 eprom_kernelbase = (uintptr_t)lvalue; 3036 3037 if (BOP_GETPROPLEN(bootops, "segmapsize") > sizeof (prop) || 3038 BOP_GETPROP(bootops, "segmapsize", prop) < 0 || 3039 kobj_getvalue(prop, &lvalue) == -1) 3040 segmapsize = SEGMAPDEFAULT; 3041 else 3042 segmapsize = (uintptr_t)lvalue; 3043 3044 if (BOP_GETPROPLEN(bootops, "segmapfreelists") > sizeof (prop) || 3045 BOP_GETPROP(bootops, "segmapfreelists", prop) < 0 || 3046 kobj_getvalue(prop, &lvalue) == -1) 3047 segmapfreelists = 0; /* use segmap driver default */ 3048 else 3049 segmapfreelists = (int)lvalue; 3050 3051 /* physmem used to be here, but moved much earlier to fakebop.c */ 3052 } 3053 3054 /* 3055 * Add to a memory list. 3056 * start = start of new memory segment 3057 * len = length of new memory segment in bytes 3058 * new = pointer to a new struct memlist 3059 * memlistp = memory list to which to add segment. 3060 */ 3061 void 3062 memlist_add( 3063 uint64_t start, 3064 uint64_t len, 3065 struct memlist *new, 3066 struct memlist **memlistp) 3067 { 3068 struct memlist *cur; 3069 uint64_t end = start + len; 3070 3071 new->ml_address = start; 3072 new->ml_size = len; 3073 3074 cur = *memlistp; 3075 3076 while (cur) { 3077 if (cur->ml_address >= end) { 3078 new->ml_next = cur; 3079 *memlistp = new; 3080 new->ml_prev = cur->ml_prev; 3081 cur->ml_prev = new; 3082 return; 3083 } 3084 ASSERT(cur->ml_address + cur->ml_size <= start); 3085 if (cur->ml_next == NULL) { 3086 cur->ml_next = new; 3087 new->ml_prev = cur; 3088 new->ml_next = NULL; 3089 return; 3090 } 3091 memlistp = &cur->ml_next; 3092 cur = cur->ml_next; 3093 } 3094 } 3095 3096 void 3097 kobj_vmem_init(vmem_t **text_arena, vmem_t **data_arena) 3098 { 3099 size_t tsize = e_modtext - modtext; 3100 size_t dsize = e_moddata - moddata; 3101 3102 *text_arena = vmem_create("module_text", tsize ? modtext : NULL, tsize, 3103 1, segkmem_alloc, segkmem_free, heaptext_arena, 0, VM_SLEEP); 3104 *data_arena = vmem_create("module_data", dsize ? moddata : NULL, dsize, 3105 1, segkmem_alloc, segkmem_free, heap32_arena, 0, VM_SLEEP); 3106 } 3107 3108 caddr_t 3109 kobj_text_alloc(vmem_t *arena, size_t size) 3110 { 3111 return (vmem_alloc(arena, size, VM_SLEEP | VM_BESTFIT)); 3112 } 3113 3114 /*ARGSUSED*/ 3115 caddr_t 3116 kobj_texthole_alloc(caddr_t addr, size_t size) 3117 { 3118 panic("unexpected call to kobj_texthole_alloc()"); 3119 /*NOTREACHED*/ 3120 return (0); 3121 } 3122 3123 /*ARGSUSED*/ 3124 void 3125 kobj_texthole_free(caddr_t addr, size_t size) 3126 { 3127 panic("unexpected call to kobj_texthole_free()"); 3128 } 3129 3130 /* 3131 * This is called just after configure() in startup(). 3132 * 3133 * The ISALIST concept is a bit hopeless on Intel, because 3134 * there's no guarantee of an ever-more-capable processor 3135 * given that various parts of the instruction set may appear 3136 * and disappear between different implementations. 3137 * 3138 * While it would be possible to correct it and even enhance 3139 * it somewhat, the explicit hardware capability bitmask allows 3140 * more flexibility. 3141 * 3142 * So, we just leave this alone. 3143 */ 3144 void 3145 setx86isalist(void) 3146 { 3147 char *tp; 3148 size_t len; 3149 extern char *isa_list; 3150 3151 #define TBUFSIZE 1024 3152 3153 tp = kmem_alloc(TBUFSIZE, KM_SLEEP); 3154 *tp = '\0'; 3155 3156 #if defined(__amd64) 3157 (void) strcpy(tp, "amd64 "); 3158 #endif 3159 3160 switch (x86_vendor) { 3161 case X86_VENDOR_Intel: 3162 case X86_VENDOR_AMD: 3163 case X86_VENDOR_TM: 3164 if (is_x86_feature(x86_featureset, X86FSET_CMOV)) { 3165 /* 3166 * Pentium Pro or later 3167 */ 3168 (void) strcat(tp, "pentium_pro"); 3169 (void) strcat(tp, 3170 is_x86_feature(x86_featureset, X86FSET_MMX) ? 3171 "+mmx pentium_pro " : " "); 3172 } 3173 /*FALLTHROUGH*/ 3174 case X86_VENDOR_Cyrix: 3175 /* 3176 * The Cyrix 6x86 does not have any Pentium features 3177 * accessible while not at privilege level 0. 3178 */ 3179 if (is_x86_feature(x86_featureset, X86FSET_CPUID)) { 3180 (void) strcat(tp, "pentium"); 3181 (void) strcat(tp, 3182 is_x86_feature(x86_featureset, X86FSET_MMX) ? 3183 "+mmx pentium " : " "); 3184 } 3185 break; 3186 default: 3187 break; 3188 } 3189 (void) strcat(tp, "i486 i386 i86"); 3190 len = strlen(tp) + 1; /* account for NULL at end of string */ 3191 isa_list = strcpy(kmem_alloc(len, KM_SLEEP), tp); 3192 kmem_free(tp, TBUFSIZE); 3193 3194 #undef TBUFSIZE 3195 } 3196 3197 3198 #ifdef __amd64 3199 3200 void * 3201 device_arena_alloc(size_t size, int vm_flag) 3202 { 3203 return (vmem_alloc(device_arena, size, vm_flag)); 3204 } 3205 3206 void 3207 device_arena_free(void *vaddr, size_t size) 3208 { 3209 vmem_free(device_arena, vaddr, size); 3210 } 3211 3212 #else /* __i386 */ 3213 3214 void * 3215 device_arena_alloc(size_t size, int vm_flag) 3216 { 3217 caddr_t vaddr; 3218 uintptr_t v; 3219 size_t start; 3220 size_t end; 3221 3222 vaddr = vmem_alloc(heap_arena, size, vm_flag); 3223 if (vaddr == NULL) 3224 return (NULL); 3225 3226 v = (uintptr_t)vaddr; 3227 ASSERT(v >= kernelbase); 3228 ASSERT(v + size <= valloc_base); 3229 3230 start = btop(v - kernelbase); 3231 end = btop(v + size - 1 - kernelbase); 3232 ASSERT(start < toxic_bit_map_len); 3233 ASSERT(end < toxic_bit_map_len); 3234 3235 while (start <= end) { 3236 BT_ATOMIC_SET(toxic_bit_map, start); 3237 ++start; 3238 } 3239 return (vaddr); 3240 } 3241 3242 void 3243 device_arena_free(void *vaddr, size_t size) 3244 { 3245 uintptr_t v = (uintptr_t)vaddr; 3246 size_t start; 3247 size_t end; 3248 3249 ASSERT(v >= kernelbase); 3250 ASSERT(v + size <= valloc_base); 3251 3252 start = btop(v - kernelbase); 3253 end = btop(v + size - 1 - kernelbase); 3254 ASSERT(start < toxic_bit_map_len); 3255 ASSERT(end < toxic_bit_map_len); 3256 3257 while (start <= end) { 3258 ASSERT(BT_TEST(toxic_bit_map, start) != 0); 3259 BT_ATOMIC_CLEAR(toxic_bit_map, start); 3260 ++start; 3261 } 3262 vmem_free(heap_arena, vaddr, size); 3263 } 3264 3265 /* 3266 * returns 1st address in range that is in device arena, or NULL 3267 * if len is not NULL it returns the length of the toxic range 3268 */ 3269 void * 3270 device_arena_contains(void *vaddr, size_t size, size_t *len) 3271 { 3272 uintptr_t v = (uintptr_t)vaddr; 3273 uintptr_t eaddr = v + size; 3274 size_t start; 3275 size_t end; 3276 3277 /* 3278 * if called very early by kmdb, just return NULL 3279 */ 3280 if (toxic_bit_map == NULL) 3281 return (NULL); 3282 3283 /* 3284 * First check if we're completely outside the bitmap range. 3285 */ 3286 if (v >= valloc_base || eaddr < kernelbase) 3287 return (NULL); 3288 3289 /* 3290 * Trim ends of search to look at only what the bitmap covers. 3291 */ 3292 if (v < kernelbase) 3293 v = kernelbase; 3294 start = btop(v - kernelbase); 3295 end = btop(eaddr - kernelbase); 3296 if (end >= toxic_bit_map_len) 3297 end = toxic_bit_map_len; 3298 3299 if (bt_range(toxic_bit_map, &start, &end, end) == 0) 3300 return (NULL); 3301 3302 v = kernelbase + ptob(start); 3303 if (len != NULL) 3304 *len = ptob(end - start); 3305 return ((void *)v); 3306 } 3307 3308 #endif /* __i386 */ 3309