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 */ 24 /* 25 * Copyright (c) 2010, Intel Corporation. 26 * All rights reserved. 27 */ 28 29 #include <sys/types.h> 30 #include <sys/t_lock.h> 31 #include <sys/param.h> 32 #include <sys/sysmacros.h> 33 #include <sys/signal.h> 34 #include <sys/systm.h> 35 #include <sys/user.h> 36 #include <sys/mman.h> 37 #include <sys/vm.h> 38 #include <sys/conf.h> 39 #include <sys/avintr.h> 40 #include <sys/autoconf.h> 41 #include <sys/disp.h> 42 #include <sys/class.h> 43 #include <sys/bitmap.h> 44 45 #include <sys/privregs.h> 46 47 #include <sys/proc.h> 48 #include <sys/buf.h> 49 #include <sys/kmem.h> 50 #include <sys/mem.h> 51 #include <sys/kstat.h> 52 53 #include <sys/reboot.h> 54 55 #include <sys/cred.h> 56 #include <sys/vnode.h> 57 #include <sys/file.h> 58 59 #include <sys/procfs.h> 60 61 #include <sys/vfs.h> 62 #include <sys/cmn_err.h> 63 #include <sys/utsname.h> 64 #include <sys/debug.h> 65 #include <sys/kdi.h> 66 67 #include <sys/dumphdr.h> 68 #include <sys/bootconf.h> 69 #include <sys/memlist_plat.h> 70 #include <sys/varargs.h> 71 #include <sys/promif.h> 72 #include <sys/modctl.h> 73 74 #include <sys/sunddi.h> 75 #include <sys/sunndi.h> 76 #include <sys/ndi_impldefs.h> 77 #include <sys/ddidmareq.h> 78 #include <sys/psw.h> 79 #include <sys/regset.h> 80 #include <sys/clock.h> 81 #include <sys/pte.h> 82 #include <sys/tss.h> 83 #include <sys/stack.h> 84 #include <sys/trap.h> 85 #include <sys/fp.h> 86 #include <vm/kboot_mmu.h> 87 #include <vm/anon.h> 88 #include <vm/as.h> 89 #include <vm/page.h> 90 #include <vm/seg.h> 91 #include <vm/seg_dev.h> 92 #include <vm/seg_kmem.h> 93 #include <vm/seg_kpm.h> 94 #include <vm/seg_map.h> 95 #include <vm/seg_vn.h> 96 #include <vm/seg_kp.h> 97 #include <sys/memnode.h> 98 #include <vm/vm_dep.h> 99 #include <sys/thread.h> 100 #include <sys/sysconf.h> 101 #include <sys/vm_machparam.h> 102 #include <sys/archsystm.h> 103 #include <sys/machsystm.h> 104 #include <vm/hat.h> 105 #include <vm/hat_i86.h> 106 #include <sys/pmem.h> 107 #include <sys/smp_impldefs.h> 108 #include <sys/x86_archext.h> 109 #include <sys/cpuvar.h> 110 #include <sys/segments.h> 111 #include <sys/clconf.h> 112 #include <sys/kobj.h> 113 #include <sys/kobj_lex.h> 114 #include <sys/cpc_impl.h> 115 #include <sys/cpu_module.h> 116 #include <sys/smbios.h> 117 #include <sys/debug_info.h> 118 #include <sys/bootinfo.h> 119 #include <sys/ddi_timer.h> 120 #include <sys/systeminfo.h> 121 #include <sys/multiboot.h> 122 123 #ifdef __xpv 124 125 #include <sys/hypervisor.h> 126 #include <sys/xen_mmu.h> 127 #include <sys/evtchn_impl.h> 128 #include <sys/gnttab.h> 129 #include <sys/xpv_panic.h> 130 #include <xen/sys/xenbus_comms.h> 131 #include <xen/public/physdev.h> 132 133 extern void xen_late_startup(void); 134 135 struct xen_evt_data cpu0_evt_data; 136 137 #else /* __xpv */ 138 #include <sys/memlist_impl.h> 139 140 extern void mem_config_init(void); 141 #endif /* __xpv */ 142 143 extern void progressbar_init(void); 144 extern void brand_init(void); 145 extern void pcf_init(void); 146 extern void pg_init(void); 147 148 extern int size_pse_array(pgcnt_t, int); 149 150 #if defined(_SOFT_HOSTID) 151 152 #include <sys/rtc.h> 153 154 static int32_t set_soft_hostid(void); 155 static char hostid_file[] = "/etc/hostid"; 156 157 #endif 158 159 void *gfx_devinfo_list; 160 161 #if defined(__amd64) && !defined(__xpv) 162 extern void immu_startup(void); 163 #endif 164 165 /* 166 * XXX make declaration below "static" when drivers no longer use this 167 * interface. 168 */ 169 extern caddr_t p0_va; /* Virtual address for accessing physical page 0 */ 170 171 /* 172 * segkp 173 */ 174 extern int segkp_fromheap; 175 176 static void kvm_init(void); 177 static void startup_init(void); 178 static void startup_memlist(void); 179 static void startup_kmem(void); 180 static void startup_modules(void); 181 static void startup_vm(void); 182 static void startup_end(void); 183 static void layout_kernel_va(void); 184 185 /* 186 * Declare these as initialized data so we can patch them. 187 */ 188 #ifdef __i386 189 190 /* 191 * Due to virtual address space limitations running in 32 bit mode, restrict 192 * the amount of physical memory configured to a max of PHYSMEM pages (16g). 193 * 194 * If the physical max memory size of 64g were allowed to be configured, the 195 * size of user virtual address space will be less than 1g. A limited user 196 * address space greatly reduces the range of applications that can run. 197 * 198 * If more physical memory than PHYSMEM is required, users should preferably 199 * run in 64 bit mode which has far looser virtual address space limitations. 200 * 201 * If 64 bit mode is not available (as in IA32) and/or more physical memory 202 * than PHYSMEM is required in 32 bit mode, physmem can be set to the desired 203 * value or to 0 (to configure all available memory) via eeprom(1M). kernelbase 204 * should also be carefully tuned to balance out the need of the user 205 * application while minimizing the risk of kernel heap exhaustion due to 206 * kernelbase being set too high. 207 */ 208 #define PHYSMEM 0x400000 209 210 #else /* __amd64 */ 211 212 /* 213 * For now we can handle memory with physical addresses up to about 214 * 64 Terabytes. This keeps the kernel above the VA hole, leaving roughly 215 * half the VA space for seg_kpm. When systems get bigger than 64TB this 216 * code will need revisiting. There is an implicit assumption that there 217 * are no *huge* holes in the physical address space too. 218 */ 219 #define TERABYTE (1ul << 40) 220 #define PHYSMEM_MAX64 mmu_btop(64 * TERABYTE) 221 #define PHYSMEM PHYSMEM_MAX64 222 #define AMD64_VA_HOLE_END 0xFFFF800000000000ul 223 224 #endif /* __amd64 */ 225 226 pgcnt_t physmem = PHYSMEM; 227 pgcnt_t obp_pages; /* Memory used by PROM for its text and data */ 228 229 char *kobj_file_buf; 230 int kobj_file_bufsize; /* set in /etc/system */ 231 232 /* Global variables for MP support. Used in mp_startup */ 233 caddr_t rm_platter_va = 0; 234 uint32_t rm_platter_pa; 235 236 int auto_lpg_disable = 1; 237 238 /* 239 * Some CPUs have holes in the middle of the 64-bit virtual address range. 240 */ 241 uintptr_t hole_start, hole_end; 242 243 /* 244 * kpm mapping window 245 */ 246 caddr_t kpm_vbase; 247 size_t kpm_size; 248 static int kpm_desired; 249 #ifdef __amd64 250 static uintptr_t segkpm_base = (uintptr_t)SEGKPM_BASE; 251 #endif 252 253 /* 254 * Configuration parameters set at boot time. 255 */ 256 257 caddr_t econtig; /* end of first block of contiguous kernel */ 258 259 struct bootops *bootops = 0; /* passed in from boot */ 260 struct bootops **bootopsp; 261 struct boot_syscalls *sysp; /* passed in from boot */ 262 263 char bootblock_fstype[16]; 264 265 char kern_bootargs[OBP_MAXPATHLEN]; 266 char kern_bootfile[OBP_MAXPATHLEN]; 267 268 /* 269 * ZFS zio segment. This allows us to exclude large portions of ZFS data that 270 * gets cached in kmem caches on the heap. If this is set to zero, we allocate 271 * zio buffers from their own segment, otherwise they are allocated from the 272 * heap. The optimization of allocating zio buffers from their own segment is 273 * only valid on 64-bit kernels. 274 */ 275 #if defined(__amd64) 276 int segzio_fromheap = 0; 277 #else 278 int segzio_fromheap = 1; 279 #endif 280 281 /* 282 * new memory fragmentations are possible in startup() due to BOP_ALLOCs. this 283 * depends on number of BOP_ALLOC calls made and requested size, memory size 284 * combination and whether boot.bin memory needs to be freed. 285 */ 286 #define POSS_NEW_FRAGMENTS 12 287 288 /* 289 * VM data structures 290 */ 291 long page_hashsz; /* Size of page hash table (power of two) */ 292 unsigned int page_hashsz_shift; /* log2(page_hashsz) */ 293 struct page *pp_base; /* Base of initial system page struct array */ 294 struct page **page_hash; /* Page hash table */ 295 pad_mutex_t *pse_mutex; /* Locks protecting pp->p_selock */ 296 size_t pse_table_size; /* Number of mutexes in pse_mutex[] */ 297 int pse_shift; /* log2(pse_table_size) */ 298 struct seg ktextseg; /* Segment used for kernel executable image */ 299 struct seg kvalloc; /* Segment used for "valloc" mapping */ 300 struct seg kpseg; /* Segment used for pageable kernel virt mem */ 301 struct seg kmapseg; /* Segment used for generic kernel mappings */ 302 struct seg kdebugseg; /* Segment used for the kernel debugger */ 303 304 struct seg *segkmap = &kmapseg; /* Kernel generic mapping segment */ 305 static struct seg *segmap = &kmapseg; /* easier to use name for in here */ 306 307 struct seg *segkp = &kpseg; /* Pageable kernel virtual memory segment */ 308 309 #if defined(__amd64) 310 struct seg kvseg_core; /* Segment used for the core heap */ 311 struct seg kpmseg; /* Segment used for physical mapping */ 312 struct seg *segkpm = &kpmseg; /* 64bit kernel physical mapping segment */ 313 #else 314 struct seg *segkpm = NULL; /* Unused on IA32 */ 315 #endif 316 317 caddr_t segkp_base; /* Base address of segkp */ 318 caddr_t segzio_base; /* Base address of segzio */ 319 #if defined(__amd64) 320 pgcnt_t segkpsize = btop(SEGKPDEFSIZE); /* size of segkp segment in pages */ 321 #else 322 pgcnt_t segkpsize = 0; 323 #endif 324 pgcnt_t segziosize = 0; /* size of zio segment in pages */ 325 326 /* 327 * A static DR page_t VA map is reserved that can map the page structures 328 * for a domain's entire RA space. The pages that back this space are 329 * dynamically allocated and need not be physically contiguous. The DR 330 * map size is derived from KPM size. 331 * This mechanism isn't used by x86 yet, so just stubs here. 332 */ 333 int ppvm_enable = 0; /* Static virtual map for page structs */ 334 page_t *ppvm_base = NULL; /* Base of page struct map */ 335 pgcnt_t ppvm_size = 0; /* Size of page struct map */ 336 337 /* 338 * VA range available to the debugger 339 */ 340 const caddr_t kdi_segdebugbase = (const caddr_t)SEGDEBUGBASE; 341 const size_t kdi_segdebugsize = SEGDEBUGSIZE; 342 343 struct memseg *memseg_base; 344 struct vnode unused_pages_vp; 345 346 #define FOURGB 0x100000000LL 347 348 struct memlist *memlist; 349 350 caddr_t s_text; /* start of kernel text segment */ 351 caddr_t e_text; /* end of kernel text segment */ 352 caddr_t s_data; /* start of kernel data segment */ 353 caddr_t e_data; /* end of kernel data segment */ 354 caddr_t modtext; /* start of loadable module text reserved */ 355 caddr_t e_modtext; /* end of loadable module text reserved */ 356 caddr_t moddata; /* start of loadable module data reserved */ 357 caddr_t e_moddata; /* end of loadable module data reserved */ 358 359 struct memlist *phys_install; /* Total installed physical memory */ 360 struct memlist *phys_avail; /* Total available physical memory */ 361 struct memlist *bios_rsvd; /* Bios reserved memory */ 362 363 /* 364 * kphysm_init returns the number of pages that were processed 365 */ 366 static pgcnt_t kphysm_init(page_t *, pgcnt_t); 367 368 #define IO_PROP_SIZE 64 /* device property size */ 369 370 /* 371 * a couple useful roundup macros 372 */ 373 #define ROUND_UP_PAGE(x) \ 374 ((uintptr_t)P2ROUNDUP((uintptr_t)(x), (uintptr_t)MMU_PAGESIZE)) 375 #define ROUND_UP_LPAGE(x) \ 376 ((uintptr_t)P2ROUNDUP((uintptr_t)(x), mmu.level_size[1])) 377 #define ROUND_UP_4MEG(x) \ 378 ((uintptr_t)P2ROUNDUP((uintptr_t)(x), (uintptr_t)FOUR_MEG)) 379 #define ROUND_UP_TOPLEVEL(x) \ 380 ((uintptr_t)P2ROUNDUP((uintptr_t)(x), mmu.level_size[mmu.max_level])) 381 382 /* 383 * 32-bit Kernel's Virtual memory layout. 384 * +-----------------------+ 385 * | | 386 * 0xFFC00000 -|-----------------------|- ARGSBASE 387 * | debugger | 388 * 0xFF800000 -|-----------------------|- SEGDEBUGBASE 389 * | Kernel Data | 390 * 0xFEC00000 -|-----------------------| 391 * | Kernel Text | 392 * 0xFE800000 -|-----------------------|- KERNEL_TEXT (0xFB400000 on Xen) 393 * |--- GDT ---|- GDT page (GDT_VA) 394 * |--- debug info ---|- debug info (DEBUG_INFO_VA) 395 * | | 396 * | page_t structures | 397 * | memsegs, memlists, | 398 * | page hash, etc. | 399 * --- -|-----------------------|- ekernelheap, valloc_base (floating) 400 * | | (segkp is just an arena in the heap) 401 * | | 402 * | kvseg | 403 * | | 404 * | | 405 * --- -|-----------------------|- kernelheap (floating) 406 * | Segkmap | 407 * 0xC3002000 -|-----------------------|- segmap_start (floating) 408 * | Red Zone | 409 * 0xC3000000 -|-----------------------|- kernelbase / userlimit (floating) 410 * | | || 411 * | Shared objects | \/ 412 * | | 413 * : : 414 * | user data | 415 * |-----------------------| 416 * | user text | 417 * 0x08048000 -|-----------------------| 418 * | user stack | 419 * : : 420 * | invalid | 421 * 0x00000000 +-----------------------+ 422 * 423 * 424 * 64-bit Kernel's Virtual memory layout. (assuming 64 bit app) 425 * +-----------------------+ 426 * | | 427 * 0xFFFFFFFF.FFC00000 |-----------------------|- ARGSBASE 428 * | debugger (?) | 429 * 0xFFFFFFFF.FF800000 |-----------------------|- SEGDEBUGBASE 430 * | unused | 431 * +-----------------------+ 432 * | Kernel Data | 433 * 0xFFFFFFFF.FBC00000 |-----------------------| 434 * | Kernel Text | 435 * 0xFFFFFFFF.FB800000 |-----------------------|- KERNEL_TEXT 436 * |--- GDT ---|- GDT page (GDT_VA) 437 * |--- debug info ---|- debug info (DEBUG_INFO_VA) 438 * | | 439 * | Core heap | (used for loadable modules) 440 * 0xFFFFFFFF.C0000000 |-----------------------|- core_base / ekernelheap 441 * | Kernel | 442 * | heap | 443 * 0xFFFFFXXX.XXX00000 |-----------------------|- kernelheap (floating) 444 * | segmap | 445 * 0xFFFFFXXX.XXX00000 |-----------------------|- segmap_start (floating) 446 * | device mappings | 447 * 0xFFFFFXXX.XXX00000 |-----------------------|- toxic_addr (floating) 448 * | segzio | 449 * 0xFFFFFXXX.XXX00000 |-----------------------|- segzio_base (floating) 450 * | segkp | 451 * --- |-----------------------|- segkp_base (floating) 452 * | page_t structures | valloc_base + valloc_sz 453 * | memsegs, memlists, | 454 * | page hash, etc. | 455 * 0xFFFFFF00.00000000 |-----------------------|- valloc_base (lower if > 1TB) 456 * | segkpm | 457 * 0xFFFFFE00.00000000 |-----------------------| 458 * | Red Zone | 459 * 0xFFFFFD80.00000000 |-----------------------|- KERNELBASE (lower if > 1TB) 460 * | User stack |- User space memory 461 * | | 462 * | shared objects, etc | (grows downwards) 463 * : : 464 * | | 465 * 0xFFFF8000.00000000 |-----------------------| 466 * | | 467 * | VA Hole / unused | 468 * | | 469 * 0x00008000.00000000 |-----------------------| 470 * | | 471 * | | 472 * : : 473 * | user heap | (grows upwards) 474 * | | 475 * | user data | 476 * |-----------------------| 477 * | user text | 478 * 0x00000000.04000000 |-----------------------| 479 * | invalid | 480 * 0x00000000.00000000 +-----------------------+ 481 * 482 * A 32 bit app on the 64 bit kernel sees the same layout as on the 32 bit 483 * kernel, except that userlimit is raised to 0xfe000000 484 * 485 * Floating values: 486 * 487 * valloc_base: start of the kernel's memory management/tracking data 488 * structures. This region contains page_t structures for 489 * physical memory, memsegs, memlists, and the page hash. 490 * 491 * core_base: start of the kernel's "core" heap area on 64-bit systems. 492 * This area is intended to be used for global data as well as for module 493 * text/data that does not fit into the nucleus pages. The core heap is 494 * restricted to a 2GB range, allowing every address within it to be 495 * accessed using rip-relative addressing 496 * 497 * ekernelheap: end of kernelheap and start of segmap. 498 * 499 * kernelheap: start of kernel heap. On 32-bit systems, this starts right 500 * above a red zone that separates the user's address space from the 501 * kernel's. On 64-bit systems, it sits above segkp and segkpm. 502 * 503 * segmap_start: start of segmap. The length of segmap can be modified 504 * through eeprom. The default length is 16MB on 32-bit systems and 64MB 505 * on 64-bit systems. 506 * 507 * kernelbase: On a 32-bit kernel the default value of 0xd4000000 will be 508 * decreased by 2X the size required for page_t. This allows the kernel 509 * heap to grow in size with physical memory. With sizeof(page_t) == 80 510 * bytes, the following shows the values of kernelbase and kernel heap 511 * sizes for different memory configurations (assuming default segmap and 512 * segkp sizes). 513 * 514 * mem size for kernelbase kernel heap 515 * size page_t's size 516 * ---- --------- ---------- ----------- 517 * 1gb 0x01400000 0xd1800000 684MB 518 * 2gb 0x02800000 0xcf000000 704MB 519 * 4gb 0x05000000 0xca000000 744MB 520 * 6gb 0x07800000 0xc5000000 784MB 521 * 8gb 0x0a000000 0xc0000000 824MB 522 * 16gb 0x14000000 0xac000000 984MB 523 * 32gb 0x28000000 0x84000000 1304MB 524 * 64gb 0x50000000 0x34000000 1944MB (*) 525 * 526 * kernelbase is less than the abi minimum of 0xc0000000 for memory 527 * configurations above 8gb. 528 * 529 * (*) support for memory configurations above 32gb will require manual tuning 530 * of kernelbase to balance out the need of user applications. 531 */ 532 533 /* real-time-clock initialization parameters */ 534 extern time_t process_rtc_config_file(void); 535 536 uintptr_t kernelbase; 537 uintptr_t postbootkernelbase; /* not set till boot loader is gone */ 538 uintptr_t eprom_kernelbase; 539 size_t segmapsize; 540 uintptr_t segmap_start; 541 int segmapfreelists; 542 pgcnt_t npages; 543 pgcnt_t orig_npages; 544 size_t core_size; /* size of "core" heap */ 545 uintptr_t core_base; /* base address of "core" heap */ 546 547 /* 548 * List of bootstrap pages. We mark these as allocated in startup. 549 * release_bootstrap() will free them when we're completely done with 550 * the bootstrap. 551 */ 552 static page_t *bootpages; 553 554 /* 555 * boot time pages that have a vnode from the ramdisk will keep that forever. 556 */ 557 static page_t *rd_pages; 558 559 /* 560 * Lower 64K 561 */ 562 static page_t *lower_pages = NULL; 563 static int lower_pages_count = 0; 564 565 struct system_hardware system_hardware; 566 567 /* 568 * Enable some debugging messages concerning memory usage... 569 */ 570 static void 571 print_memlist(char *title, struct memlist *mp) 572 { 573 prom_printf("MEMLIST: %s:\n", title); 574 while (mp != NULL) { 575 prom_printf("\tAddress 0x%" PRIx64 ", size 0x%" PRIx64 "\n", 576 mp->ml_address, mp->ml_size); 577 mp = mp->ml_next; 578 } 579 } 580 581 /* 582 * XX64 need a comment here.. are these just default values, surely 583 * we read the "cpuid" type information to figure this out. 584 */ 585 int l2cache_sz = 0x80000; 586 int l2cache_linesz = 0x40; 587 int l2cache_assoc = 1; 588 589 static size_t textrepl_min_gb = 10; 590 591 /* 592 * on 64 bit we use a predifined VA range for mapping devices in the kernel 593 * on 32 bit the mappings are intermixed in the heap, so we use a bit map 594 */ 595 #ifdef __amd64 596 597 vmem_t *device_arena; 598 uintptr_t toxic_addr = (uintptr_t)NULL; 599 size_t toxic_size = 1024 * 1024 * 1024; /* Sparc uses 1 gig too */ 600 601 #else /* __i386 */ 602 603 ulong_t *toxic_bit_map; /* one bit for each 4k of VA in heap_arena */ 604 size_t toxic_bit_map_len = 0; /* in bits */ 605 606 #endif /* __i386 */ 607 608 /* 609 * Simple boot time debug facilities 610 */ 611 static char *prm_dbg_str[] = { 612 "%s:%d: '%s' is 0x%x\n", 613 "%s:%d: '%s' is 0x%llx\n" 614 }; 615 616 int prom_debug; 617 618 #define PRM_DEBUG(q) if (prom_debug) \ 619 prom_printf(prm_dbg_str[sizeof (q) >> 3], "startup.c", __LINE__, #q, q); 620 #define PRM_POINT(q) if (prom_debug) \ 621 prom_printf("%s:%d: %s\n", "startup.c", __LINE__, q); 622 623 /* 624 * This structure is used to keep track of the intial allocations 625 * done in startup_memlist(). The value of NUM_ALLOCATIONS needs to 626 * be >= the number of ADD_TO_ALLOCATIONS() executed in the code. 627 */ 628 #define NUM_ALLOCATIONS 8 629 int num_allocations = 0; 630 struct { 631 void **al_ptr; 632 size_t al_size; 633 } allocations[NUM_ALLOCATIONS]; 634 size_t valloc_sz = 0; 635 uintptr_t valloc_base; 636 637 #define ADD_TO_ALLOCATIONS(ptr, size) { \ 638 size = ROUND_UP_PAGE(size); \ 639 if (num_allocations == NUM_ALLOCATIONS) \ 640 panic("too many ADD_TO_ALLOCATIONS()"); \ 641 allocations[num_allocations].al_ptr = (void**)&ptr; \ 642 allocations[num_allocations].al_size = size; \ 643 valloc_sz += size; \ 644 ++num_allocations; \ 645 } 646 647 /* 648 * Allocate all the initial memory needed by the page allocator. 649 */ 650 static void 651 perform_allocations(void) 652 { 653 caddr_t mem; 654 int i; 655 int valloc_align; 656 657 PRM_DEBUG(valloc_base); 658 PRM_DEBUG(valloc_sz); 659 valloc_align = mmu.level_size[mmu.max_page_level > 0]; 660 mem = BOP_ALLOC(bootops, (caddr_t)valloc_base, valloc_sz, valloc_align); 661 if (mem != (caddr_t)valloc_base) 662 panic("BOP_ALLOC() failed"); 663 bzero(mem, valloc_sz); 664 for (i = 0; i < num_allocations; ++i) { 665 *allocations[i].al_ptr = (void *)mem; 666 mem += allocations[i].al_size; 667 } 668 } 669 670 /* 671 * Our world looks like this at startup time. 672 * 673 * In a 32-bit OS, boot loads the kernel text at 0xfe800000 and kernel data 674 * at 0xfec00000. On a 64-bit OS, kernel text and data are loaded at 675 * 0xffffffff.fe800000 and 0xffffffff.fec00000 respectively. Those 676 * addresses are fixed in the binary at link time. 677 * 678 * On the text page: 679 * unix/genunix/krtld/module text loads. 680 * 681 * On the data page: 682 * unix/genunix/krtld/module data loads. 683 * 684 * Machine-dependent startup code 685 */ 686 void 687 startup(void) 688 { 689 #if !defined(__xpv) 690 extern void startup_pci_bios(void); 691 #endif 692 extern cpuset_t cpu_ready_set; 693 694 /* 695 * Make sure that nobody tries to use sekpm until we have 696 * initialized it properly. 697 */ 698 #if defined(__amd64) 699 kpm_desired = 1; 700 #endif 701 kpm_enable = 0; 702 CPUSET_ONLY(cpu_ready_set, 0); /* cpu 0 is boot cpu */ 703 704 #if defined(__xpv) /* XXPV fix me! */ 705 { 706 extern int segvn_use_regions; 707 segvn_use_regions = 0; 708 } 709 #endif 710 progressbar_init(); 711 startup_init(); 712 #if defined(__xpv) 713 startup_xen_version(); 714 #endif 715 startup_memlist(); 716 startup_kmem(); 717 startup_vm(); 718 #if !defined(__xpv) 719 /* 720 * Note we need to do this even on fast reboot in order to access 721 * the irq routing table (used for pci labels). 722 */ 723 startup_pci_bios(); 724 #endif 725 #if defined(__xpv) 726 startup_xen_mca(); 727 #endif 728 startup_modules(); 729 730 startup_end(); 731 } 732 733 static void 734 startup_init() 735 { 736 PRM_POINT("startup_init() starting..."); 737 738 /* 739 * Complete the extraction of cpuid data 740 */ 741 cpuid_pass2(CPU); 742 743 (void) check_boot_version(BOP_GETVERSION(bootops)); 744 745 /* 746 * Check for prom_debug in boot environment 747 */ 748 if (BOP_GETPROPLEN(bootops, "prom_debug") >= 0) { 749 ++prom_debug; 750 PRM_POINT("prom_debug found in boot enviroment"); 751 } 752 753 /* 754 * Collect node, cpu and memory configuration information. 755 */ 756 get_system_configuration(); 757 758 /* 759 * Halt if this is an unsupported processor. 760 */ 761 if (x86_type == X86_TYPE_486 || x86_type == X86_TYPE_CYRIX_486) { 762 printf("\n486 processor (\"%s\") detected.\n", 763 CPU->cpu_brandstr); 764 halt("This processor is not supported by this release " 765 "of Solaris."); 766 } 767 768 PRM_POINT("startup_init() done"); 769 } 770 771 /* 772 * Callback for copy_memlist_filter() to filter nucleus, kadb/kmdb, (ie. 773 * everything mapped above KERNEL_TEXT) pages from phys_avail. Note it 774 * also filters out physical page zero. There is some reliance on the 775 * boot loader allocating only a few contiguous physical memory chunks. 776 */ 777 static void 778 avail_filter(uint64_t *addr, uint64_t *size) 779 { 780 uintptr_t va; 781 uintptr_t next_va; 782 pfn_t pfn; 783 uint64_t pfn_addr; 784 uint64_t pfn_eaddr; 785 uint_t prot; 786 size_t len; 787 uint_t change; 788 789 if (prom_debug) 790 prom_printf("\tFilter: in: a=%" PRIx64 ", s=%" PRIx64 "\n", 791 *addr, *size); 792 793 /* 794 * page zero is required for BIOS.. never make it available 795 */ 796 if (*addr == 0) { 797 *addr += MMU_PAGESIZE; 798 *size -= MMU_PAGESIZE; 799 } 800 801 /* 802 * First we trim from the front of the range. Since kbm_probe() 803 * walks ranges in virtual order, but addr/size are physical, we need 804 * to the list until no changes are seen. This deals with the case 805 * where page "p" is mapped at v, page "p + PAGESIZE" is mapped at w 806 * but w < v. 807 */ 808 do { 809 change = 0; 810 for (va = KERNEL_TEXT; 811 *size > 0 && kbm_probe(&va, &len, &pfn, &prot) != 0; 812 va = next_va) { 813 814 next_va = va + len; 815 pfn_addr = pfn_to_pa(pfn); 816 pfn_eaddr = pfn_addr + len; 817 818 if (pfn_addr <= *addr && pfn_eaddr > *addr) { 819 change = 1; 820 while (*size > 0 && len > 0) { 821 *addr += MMU_PAGESIZE; 822 *size -= MMU_PAGESIZE; 823 len -= MMU_PAGESIZE; 824 } 825 } 826 } 827 if (change && prom_debug) 828 prom_printf("\t\ttrim: a=%" PRIx64 ", s=%" PRIx64 "\n", 829 *addr, *size); 830 } while (change); 831 832 /* 833 * Trim pages from the end of the range. 834 */ 835 for (va = KERNEL_TEXT; 836 *size > 0 && kbm_probe(&va, &len, &pfn, &prot) != 0; 837 va = next_va) { 838 839 next_va = va + len; 840 pfn_addr = pfn_to_pa(pfn); 841 842 if (pfn_addr >= *addr && pfn_addr < *addr + *size) 843 *size = pfn_addr - *addr; 844 } 845 846 if (prom_debug) 847 prom_printf("\tFilter out: a=%" PRIx64 ", s=%" PRIx64 "\n", 848 *addr, *size); 849 } 850 851 static void 852 kpm_init() 853 { 854 struct segkpm_crargs b; 855 856 /* 857 * These variables were all designed for sfmmu in which segkpm is 858 * mapped using a single pagesize - either 8KB or 4MB. On x86, we 859 * might use 2+ page sizes on a single machine, so none of these 860 * variables have a single correct value. They are set up as if we 861 * always use a 4KB pagesize, which should do no harm. In the long 862 * run, we should get rid of KPM's assumption that only a single 863 * pagesize is used. 864 */ 865 kpm_pgshft = MMU_PAGESHIFT; 866 kpm_pgsz = MMU_PAGESIZE; 867 kpm_pgoff = MMU_PAGEOFFSET; 868 kpmp2pshft = 0; 869 kpmpnpgs = 1; 870 ASSERT(((uintptr_t)kpm_vbase & (kpm_pgsz - 1)) == 0); 871 872 PRM_POINT("about to create segkpm"); 873 rw_enter(&kas.a_lock, RW_WRITER); 874 875 if (seg_attach(&kas, kpm_vbase, kpm_size, segkpm) < 0) 876 panic("cannot attach segkpm"); 877 878 b.prot = PROT_READ | PROT_WRITE; 879 b.nvcolors = 1; 880 881 if (segkpm_create(segkpm, (caddr_t)&b) != 0) 882 panic("segkpm_create segkpm"); 883 884 rw_exit(&kas.a_lock); 885 } 886 887 /* 888 * The debug info page provides enough information to allow external 889 * inspectors (e.g. when running under a hypervisor) to bootstrap 890 * themselves into allowing full-blown kernel debugging. 891 */ 892 static void 893 init_debug_info(void) 894 { 895 caddr_t mem; 896 debug_info_t *di; 897 898 #ifndef __lint 899 ASSERT(sizeof (debug_info_t) < MMU_PAGESIZE); 900 #endif 901 902 mem = BOP_ALLOC(bootops, (caddr_t)DEBUG_INFO_VA, MMU_PAGESIZE, 903 MMU_PAGESIZE); 904 905 if (mem != (caddr_t)DEBUG_INFO_VA) 906 panic("BOP_ALLOC() failed"); 907 bzero(mem, MMU_PAGESIZE); 908 909 di = (debug_info_t *)mem; 910 911 di->di_magic = DEBUG_INFO_MAGIC; 912 di->di_version = DEBUG_INFO_VERSION; 913 di->di_modules = (uintptr_t)&modules; 914 di->di_s_text = (uintptr_t)s_text; 915 di->di_e_text = (uintptr_t)e_text; 916 di->di_s_data = (uintptr_t)s_data; 917 di->di_e_data = (uintptr_t)e_data; 918 di->di_hat_htable_off = offsetof(hat_t, hat_htable); 919 di->di_ht_pfn_off = offsetof(htable_t, ht_pfn); 920 } 921 922 /* 923 * Build the memlists and other kernel essential memory system data structures. 924 * This is everything at valloc_base. 925 */ 926 static void 927 startup_memlist(void) 928 { 929 size_t memlist_sz; 930 size_t memseg_sz; 931 size_t pagehash_sz; 932 size_t pp_sz; 933 uintptr_t va; 934 size_t len; 935 uint_t prot; 936 pfn_t pfn; 937 int memblocks; 938 pfn_t rsvd_high_pfn; 939 pgcnt_t rsvd_pgcnt; 940 size_t rsvdmemlist_sz; 941 int rsvdmemblocks; 942 caddr_t pagecolor_mem; 943 size_t pagecolor_memsz; 944 caddr_t page_ctrs_mem; 945 size_t page_ctrs_size; 946 size_t pse_table_alloc_size; 947 struct memlist *current; 948 extern void startup_build_mem_nodes(struct memlist *); 949 950 /* XX64 fix these - they should be in include files */ 951 extern size_t page_coloring_init(uint_t, int, int); 952 extern void page_coloring_setup(caddr_t); 953 954 PRM_POINT("startup_memlist() starting..."); 955 956 /* 957 * Use leftover large page nucleus text/data space for loadable modules. 958 * Use at most MODTEXT/MODDATA. 959 */ 960 len = kbm_nucleus_size; 961 ASSERT(len > MMU_PAGESIZE); 962 963 moddata = (caddr_t)ROUND_UP_PAGE(e_data); 964 e_moddata = (caddr_t)P2ROUNDUP((uintptr_t)e_data, (uintptr_t)len); 965 if (e_moddata - moddata > MODDATA) 966 e_moddata = moddata + MODDATA; 967 968 modtext = (caddr_t)ROUND_UP_PAGE(e_text); 969 e_modtext = (caddr_t)P2ROUNDUP((uintptr_t)e_text, (uintptr_t)len); 970 if (e_modtext - modtext > MODTEXT) 971 e_modtext = modtext + MODTEXT; 972 973 econtig = e_moddata; 974 975 PRM_DEBUG(modtext); 976 PRM_DEBUG(e_modtext); 977 PRM_DEBUG(moddata); 978 PRM_DEBUG(e_moddata); 979 PRM_DEBUG(econtig); 980 981 /* 982 * Examine the boot loader physical memory map to find out: 983 * - total memory in system - physinstalled 984 * - the max physical address - physmax 985 * - the number of discontiguous segments of memory. 986 */ 987 if (prom_debug) 988 print_memlist("boot physinstalled", 989 bootops->boot_mem->physinstalled); 990 installed_top_size_ex(bootops->boot_mem->physinstalled, &physmax, 991 &physinstalled, &memblocks); 992 PRM_DEBUG(physmax); 993 PRM_DEBUG(physinstalled); 994 PRM_DEBUG(memblocks); 995 996 /* 997 * Compute maximum physical address for memory DR operations. 998 * Memory DR operations are unsupported on xpv or 32bit OSes. 999 */ 1000 #ifdef __amd64 1001 if (plat_dr_support_memory()) { 1002 if (plat_dr_physmax == 0) { 1003 uint_t pabits = UINT_MAX; 1004 1005 cpuid_get_addrsize(CPU, &pabits, NULL); 1006 plat_dr_physmax = btop(1ULL << pabits); 1007 } 1008 if (plat_dr_physmax > PHYSMEM_MAX64) 1009 plat_dr_physmax = PHYSMEM_MAX64; 1010 } else 1011 #endif 1012 plat_dr_physmax = 0; 1013 1014 /* 1015 * Examine the bios reserved memory to find out: 1016 * - the number of discontiguous segments of memory. 1017 */ 1018 if (prom_debug) 1019 print_memlist("boot reserved mem", 1020 bootops->boot_mem->rsvdmem); 1021 installed_top_size_ex(bootops->boot_mem->rsvdmem, &rsvd_high_pfn, 1022 &rsvd_pgcnt, &rsvdmemblocks); 1023 PRM_DEBUG(rsvd_high_pfn); 1024 PRM_DEBUG(rsvd_pgcnt); 1025 PRM_DEBUG(rsvdmemblocks); 1026 1027 /* 1028 * Initialize hat's mmu parameters. 1029 * Check for enforce-prot-exec in boot environment. It's used to 1030 * enable/disable support for the page table entry NX bit. 1031 * The default is to enforce PROT_EXEC on processors that support NX. 1032 * Boot seems to round up the "len", but 8 seems to be big enough. 1033 */ 1034 mmu_init(); 1035 1036 #ifdef __i386 1037 /* 1038 * physmax is lowered if there is more memory than can be 1039 * physically addressed in 32 bit (PAE/non-PAE) modes. 1040 */ 1041 if (mmu.pae_hat) { 1042 if (PFN_ABOVE64G(physmax)) { 1043 physinstalled -= (physmax - (PFN_64G - 1)); 1044 physmax = PFN_64G - 1; 1045 } 1046 } else { 1047 if (PFN_ABOVE4G(physmax)) { 1048 physinstalled -= (physmax - (PFN_4G - 1)); 1049 physmax = PFN_4G - 1; 1050 } 1051 } 1052 #endif 1053 1054 startup_build_mem_nodes(bootops->boot_mem->physinstalled); 1055 1056 if (BOP_GETPROPLEN(bootops, "enforce-prot-exec") >= 0) { 1057 int len = BOP_GETPROPLEN(bootops, "enforce-prot-exec"); 1058 char value[8]; 1059 1060 if (len < 8) 1061 (void) BOP_GETPROP(bootops, "enforce-prot-exec", value); 1062 else 1063 (void) strcpy(value, ""); 1064 if (strcmp(value, "off") == 0) 1065 mmu.pt_nx = 0; 1066 } 1067 PRM_DEBUG(mmu.pt_nx); 1068 1069 /* 1070 * We will need page_t's for every page in the system, except for 1071 * memory mapped at or above above the start of the kernel text segment. 1072 * 1073 * pages above e_modtext are attributed to kernel debugger (obp_pages) 1074 */ 1075 npages = physinstalled - 1; /* avail_filter() skips page 0, so "- 1" */ 1076 obp_pages = 0; 1077 va = KERNEL_TEXT; 1078 while (kbm_probe(&va, &len, &pfn, &prot) != 0) { 1079 npages -= len >> MMU_PAGESHIFT; 1080 if (va >= (uintptr_t)e_moddata) 1081 obp_pages += len >> MMU_PAGESHIFT; 1082 va += len; 1083 } 1084 PRM_DEBUG(npages); 1085 PRM_DEBUG(obp_pages); 1086 1087 /* 1088 * If physmem is patched to be non-zero, use it instead of the computed 1089 * value unless it is larger than the actual amount of memory on hand. 1090 */ 1091 if (physmem == 0 || physmem > npages) { 1092 physmem = npages; 1093 } else if (physmem < npages) { 1094 orig_npages = npages; 1095 npages = physmem; 1096 } 1097 PRM_DEBUG(physmem); 1098 1099 /* 1100 * We now compute the sizes of all the initial allocations for 1101 * structures the kernel needs in order do kmem_alloc(). These 1102 * include: 1103 * memsegs 1104 * memlists 1105 * page hash table 1106 * page_t's 1107 * page coloring data structs 1108 */ 1109 memseg_sz = sizeof (struct memseg) * (memblocks + POSS_NEW_FRAGMENTS); 1110 ADD_TO_ALLOCATIONS(memseg_base, memseg_sz); 1111 PRM_DEBUG(memseg_sz); 1112 1113 /* 1114 * Reserve space for memlists. There's no real good way to know exactly 1115 * how much room we'll need, but this should be a good upper bound. 1116 */ 1117 memlist_sz = ROUND_UP_PAGE(2 * sizeof (struct memlist) * 1118 (memblocks + POSS_NEW_FRAGMENTS)); 1119 ADD_TO_ALLOCATIONS(memlist, memlist_sz); 1120 PRM_DEBUG(memlist_sz); 1121 1122 /* 1123 * Reserve space for bios reserved memlists. 1124 */ 1125 rsvdmemlist_sz = ROUND_UP_PAGE(2 * sizeof (struct memlist) * 1126 (rsvdmemblocks + POSS_NEW_FRAGMENTS)); 1127 ADD_TO_ALLOCATIONS(bios_rsvd, rsvdmemlist_sz); 1128 PRM_DEBUG(rsvdmemlist_sz); 1129 1130 /* LINTED */ 1131 ASSERT(P2SAMEHIGHBIT((1 << PP_SHIFT), sizeof (struct page))); 1132 /* 1133 * The page structure hash table size is a power of 2 1134 * such that the average hash chain length is PAGE_HASHAVELEN. 1135 */ 1136 page_hashsz = npages / PAGE_HASHAVELEN; 1137 page_hashsz_shift = highbit(page_hashsz); 1138 page_hashsz = 1 << page_hashsz_shift; 1139 pagehash_sz = sizeof (struct page *) * page_hashsz; 1140 ADD_TO_ALLOCATIONS(page_hash, pagehash_sz); 1141 PRM_DEBUG(pagehash_sz); 1142 1143 /* 1144 * Set aside room for the page structures themselves. 1145 */ 1146 PRM_DEBUG(npages); 1147 pp_sz = sizeof (struct page) * npages; 1148 ADD_TO_ALLOCATIONS(pp_base, pp_sz); 1149 PRM_DEBUG(pp_sz); 1150 1151 /* 1152 * determine l2 cache info and memory size for page coloring 1153 */ 1154 (void) getl2cacheinfo(CPU, 1155 &l2cache_sz, &l2cache_linesz, &l2cache_assoc); 1156 pagecolor_memsz = 1157 page_coloring_init(l2cache_sz, l2cache_linesz, l2cache_assoc); 1158 ADD_TO_ALLOCATIONS(pagecolor_mem, pagecolor_memsz); 1159 PRM_DEBUG(pagecolor_memsz); 1160 1161 page_ctrs_size = page_ctrs_sz(); 1162 ADD_TO_ALLOCATIONS(page_ctrs_mem, page_ctrs_size); 1163 PRM_DEBUG(page_ctrs_size); 1164 1165 /* 1166 * Allocate the array that protects pp->p_selock. 1167 */ 1168 pse_shift = size_pse_array(physmem, max_ncpus); 1169 pse_table_size = 1 << pse_shift; 1170 pse_table_alloc_size = pse_table_size * sizeof (pad_mutex_t); 1171 ADD_TO_ALLOCATIONS(pse_mutex, pse_table_alloc_size); 1172 1173 #if defined(__amd64) 1174 valloc_sz = ROUND_UP_LPAGE(valloc_sz); 1175 valloc_base = VALLOC_BASE; 1176 1177 /* 1178 * The default values of VALLOC_BASE and SEGKPM_BASE should work 1179 * for values of physmax up to 1 Terabyte. They need adjusting when 1180 * memory is at addresses above 1 TB. When adjusted, segkpm_base must 1181 * be aligned on KERNEL_REDZONE_SIZE boundary (span of top level pte). 1182 */ 1183 if (physmax + 1 > mmu_btop(TERABYTE) || 1184 plat_dr_physmax > mmu_btop(TERABYTE)) { 1185 uint64_t kpm_resv_amount = mmu_ptob(physmax + 1); 1186 1187 if (kpm_resv_amount < mmu_ptob(plat_dr_physmax)) { 1188 kpm_resv_amount = mmu_ptob(plat_dr_physmax); 1189 } 1190 1191 segkpm_base = -(P2ROUNDUP((2 * kpm_resv_amount), 1192 KERNEL_REDZONE_SIZE)); /* down from top VA */ 1193 1194 /* make sure we leave some space for user apps above hole */ 1195 segkpm_base = MAX(segkpm_base, AMD64_VA_HOLE_END + TERABYTE); 1196 if (segkpm_base > SEGKPM_BASE) 1197 segkpm_base = SEGKPM_BASE; 1198 PRM_DEBUG(segkpm_base); 1199 1200 valloc_base = segkpm_base + P2ROUNDUP(kpm_resv_amount, ONE_GIG); 1201 if (valloc_base < segkpm_base) 1202 panic("not enough kernel VA to support memory size"); 1203 PRM_DEBUG(valloc_base); 1204 } 1205 #else /* __i386 */ 1206 valloc_base = (uintptr_t)(MISC_VA_BASE - valloc_sz); 1207 valloc_base = P2ALIGN(valloc_base, mmu.level_size[1]); 1208 PRM_DEBUG(valloc_base); 1209 #endif /* __i386 */ 1210 1211 /* 1212 * do all the initial allocations 1213 */ 1214 perform_allocations(); 1215 1216 /* 1217 * Build phys_install and phys_avail in kernel memspace. 1218 * - phys_install should be all memory in the system. 1219 * - phys_avail is phys_install minus any memory mapped before this 1220 * point above KERNEL_TEXT. 1221 */ 1222 current = phys_install = memlist; 1223 copy_memlist_filter(bootops->boot_mem->physinstalled, ¤t, NULL); 1224 if ((caddr_t)current > (caddr_t)memlist + memlist_sz) 1225 panic("physinstalled was too big!"); 1226 if (prom_debug) 1227 print_memlist("phys_install", phys_install); 1228 1229 phys_avail = current; 1230 PRM_POINT("Building phys_avail:\n"); 1231 copy_memlist_filter(bootops->boot_mem->physinstalled, ¤t, 1232 avail_filter); 1233 if ((caddr_t)current > (caddr_t)memlist + memlist_sz) 1234 panic("physavail was too big!"); 1235 if (prom_debug) 1236 print_memlist("phys_avail", phys_avail); 1237 #ifndef __xpv 1238 /* 1239 * Free unused memlist items, which may be used by memory DR driver 1240 * at runtime. 1241 */ 1242 if ((caddr_t)current < (caddr_t)memlist + memlist_sz) { 1243 memlist_free_block((caddr_t)current, 1244 (caddr_t)memlist + memlist_sz - (caddr_t)current); 1245 } 1246 #endif 1247 1248 /* 1249 * Build bios reserved memspace 1250 */ 1251 current = bios_rsvd; 1252 copy_memlist_filter(bootops->boot_mem->rsvdmem, ¤t, NULL); 1253 if ((caddr_t)current > (caddr_t)bios_rsvd + rsvdmemlist_sz) 1254 panic("bios_rsvd was too big!"); 1255 if (prom_debug) 1256 print_memlist("bios_rsvd", bios_rsvd); 1257 #ifndef __xpv 1258 /* 1259 * Free unused memlist items, which may be used by memory DR driver 1260 * at runtime. 1261 */ 1262 if ((caddr_t)current < (caddr_t)bios_rsvd + rsvdmemlist_sz) { 1263 memlist_free_block((caddr_t)current, 1264 (caddr_t)bios_rsvd + rsvdmemlist_sz - (caddr_t)current); 1265 } 1266 #endif 1267 1268 /* 1269 * setup page coloring 1270 */ 1271 page_coloring_setup(pagecolor_mem); 1272 page_lock_init(); /* currently a no-op */ 1273 1274 /* 1275 * free page list counters 1276 */ 1277 (void) page_ctrs_alloc(page_ctrs_mem); 1278 1279 /* 1280 * Size the pcf array based on the number of cpus in the box at 1281 * boot time. 1282 */ 1283 1284 pcf_init(); 1285 1286 /* 1287 * Initialize the page structures from the memory lists. 1288 */ 1289 availrmem_initial = availrmem = freemem = 0; 1290 PRM_POINT("Calling kphysm_init()..."); 1291 npages = kphysm_init(pp_base, npages); 1292 PRM_POINT("kphysm_init() done"); 1293 PRM_DEBUG(npages); 1294 1295 init_debug_info(); 1296 1297 /* 1298 * Now that page_t's have been initialized, remove all the 1299 * initial allocation pages from the kernel free page lists. 1300 */ 1301 boot_mapin((caddr_t)valloc_base, valloc_sz); 1302 boot_mapin((caddr_t)MISC_VA_BASE, MISC_VA_SIZE); 1303 PRM_POINT("startup_memlist() done"); 1304 1305 PRM_DEBUG(valloc_sz); 1306 1307 #if defined(__amd64) 1308 if ((availrmem >> (30 - MMU_PAGESHIFT)) >= 1309 textrepl_min_gb && l2cache_sz <= 2 << 20) { 1310 extern size_t textrepl_size_thresh; 1311 textrepl_size_thresh = (16 << 20) - 1; 1312 } 1313 #endif 1314 } 1315 1316 /* 1317 * Layout the kernel's part of address space and initialize kmem allocator. 1318 */ 1319 static void 1320 startup_kmem(void) 1321 { 1322 extern void page_set_colorequiv_arr(void); 1323 1324 PRM_POINT("startup_kmem() starting..."); 1325 1326 #if defined(__amd64) 1327 if (eprom_kernelbase && eprom_kernelbase != KERNELBASE) 1328 cmn_err(CE_NOTE, "!kernelbase cannot be changed on 64-bit " 1329 "systems."); 1330 kernelbase = segkpm_base - KERNEL_REDZONE_SIZE; 1331 core_base = (uintptr_t)COREHEAP_BASE; 1332 core_size = (size_t)MISC_VA_BASE - COREHEAP_BASE; 1333 #else /* __i386 */ 1334 /* 1335 * We configure kernelbase based on: 1336 * 1337 * 1. user specified kernelbase via eeprom command. Value cannot exceed 1338 * KERNELBASE_MAX. we large page align eprom_kernelbase 1339 * 1340 * 2. Default to KERNELBASE and adjust to 2X less the size for page_t. 1341 * On large memory systems we must lower kernelbase to allow 1342 * enough room for page_t's for all of memory. 1343 * 1344 * The value set here, might be changed a little later. 1345 */ 1346 if (eprom_kernelbase) { 1347 kernelbase = eprom_kernelbase & mmu.level_mask[1]; 1348 if (kernelbase > KERNELBASE_MAX) 1349 kernelbase = KERNELBASE_MAX; 1350 } else { 1351 kernelbase = (uintptr_t)KERNELBASE; 1352 kernelbase -= ROUND_UP_4MEG(2 * valloc_sz); 1353 } 1354 ASSERT((kernelbase & mmu.level_offset[1]) == 0); 1355 core_base = valloc_base; 1356 core_size = 0; 1357 #endif /* __i386 */ 1358 1359 PRM_DEBUG(core_base); 1360 PRM_DEBUG(core_size); 1361 PRM_DEBUG(kernelbase); 1362 1363 #if defined(__i386) 1364 segkp_fromheap = 1; 1365 #endif /* __i386 */ 1366 1367 ekernelheap = (char *)core_base; 1368 PRM_DEBUG(ekernelheap); 1369 1370 /* 1371 * Now that we know the real value of kernelbase, 1372 * update variables that were initialized with a value of 1373 * KERNELBASE (in common/conf/param.c). 1374 * 1375 * XXX The problem with this sort of hackery is that the 1376 * compiler just may feel like putting the const declarations 1377 * (in param.c) into the .text section. Perhaps they should 1378 * just be declared as variables there? 1379 */ 1380 1381 *(uintptr_t *)&_kernelbase = kernelbase; 1382 *(uintptr_t *)&_userlimit = kernelbase; 1383 #if defined(__amd64) 1384 *(uintptr_t *)&_userlimit -= KERNELBASE - USERLIMIT; 1385 #else 1386 *(uintptr_t *)&_userlimit32 = _userlimit; 1387 #endif 1388 PRM_DEBUG(_kernelbase); 1389 PRM_DEBUG(_userlimit); 1390 PRM_DEBUG(_userlimit32); 1391 1392 layout_kernel_va(); 1393 1394 #if defined(__i386) 1395 /* 1396 * If segmap is too large we can push the bottom of the kernel heap 1397 * higher than the base. Or worse, it could exceed the top of the 1398 * VA space entirely, causing it to wrap around. 1399 */ 1400 if (kernelheap >= ekernelheap || (uintptr_t)kernelheap < kernelbase) 1401 panic("too little address space available for kernelheap," 1402 " use eeprom for lower kernelbase or smaller segmapsize"); 1403 #endif /* __i386 */ 1404 1405 /* 1406 * Initialize the kernel heap. Note 3rd argument must be > 1st. 1407 */ 1408 kernelheap_init(kernelheap, ekernelheap, 1409 kernelheap + MMU_PAGESIZE, 1410 (void *)core_base, (void *)(core_base + core_size)); 1411 1412 #if defined(__xpv) 1413 /* 1414 * Link pending events struct into cpu struct 1415 */ 1416 CPU->cpu_m.mcpu_evt_pend = &cpu0_evt_data; 1417 #endif 1418 /* 1419 * Initialize kernel memory allocator. 1420 */ 1421 kmem_init(); 1422 1423 /* 1424 * Factor in colorequiv to check additional 'equivalent' bins 1425 */ 1426 page_set_colorequiv_arr(); 1427 1428 /* 1429 * print this out early so that we know what's going on 1430 */ 1431 print_x86_featureset(x86_featureset); 1432 1433 /* 1434 * Initialize bp_mapin(). 1435 */ 1436 bp_init(MMU_PAGESIZE, HAT_STORECACHING_OK); 1437 1438 /* 1439 * orig_npages is non-zero if physmem has been configured for less 1440 * than the available memory. 1441 */ 1442 if (orig_npages) { 1443 cmn_err(CE_WARN, "!%slimiting physmem to 0x%lx of 0x%lx pages", 1444 (npages == PHYSMEM ? "Due to virtual address space " : ""), 1445 npages, orig_npages); 1446 } 1447 #if defined(__i386) 1448 if (eprom_kernelbase && (eprom_kernelbase != kernelbase)) 1449 cmn_err(CE_WARN, "kernelbase value, User specified 0x%lx, " 1450 "System using 0x%lx", 1451 (uintptr_t)eprom_kernelbase, (uintptr_t)kernelbase); 1452 #endif 1453 1454 #ifdef KERNELBASE_ABI_MIN 1455 if (kernelbase < (uintptr_t)KERNELBASE_ABI_MIN) { 1456 cmn_err(CE_NOTE, "!kernelbase set to 0x%lx, system is not " 1457 "i386 ABI compliant.", (uintptr_t)kernelbase); 1458 } 1459 #endif 1460 1461 #ifndef __xpv 1462 if (plat_dr_support_memory()) { 1463 mem_config_init(); 1464 } 1465 #else /* __xpv */ 1466 /* 1467 * Some of the xen start information has to be relocated up 1468 * into the kernel's permanent address space. 1469 */ 1470 PRM_POINT("calling xen_relocate_start_info()"); 1471 xen_relocate_start_info(); 1472 PRM_POINT("xen_relocate_start_info() done"); 1473 1474 /* 1475 * (Update the vcpu pointer in our cpu structure to point into 1476 * the relocated shared info.) 1477 */ 1478 CPU->cpu_m.mcpu_vcpu_info = 1479 &HYPERVISOR_shared_info->vcpu_info[CPU->cpu_id]; 1480 #endif /* __xpv */ 1481 1482 PRM_POINT("startup_kmem() done"); 1483 } 1484 1485 #ifndef __xpv 1486 /* 1487 * If we have detected that we are running in an HVM environment, we need 1488 * to prepend the PV driver directory to the module search path. 1489 */ 1490 #define HVM_MOD_DIR "/platform/i86hvm/kernel" 1491 static void 1492 update_default_path() 1493 { 1494 char *current, *newpath; 1495 int newlen; 1496 1497 /* 1498 * We are about to resync with krtld. krtld will reset its 1499 * internal module search path iff Solaris has set default_path. 1500 * We want to be sure we're prepending this new directory to the 1501 * right search path. 1502 */ 1503 current = (default_path == NULL) ? kobj_module_path : default_path; 1504 1505 newlen = strlen(HVM_MOD_DIR) + strlen(current) + 2; 1506 newpath = kmem_alloc(newlen, KM_SLEEP); 1507 (void) strcpy(newpath, HVM_MOD_DIR); 1508 (void) strcat(newpath, " "); 1509 (void) strcat(newpath, current); 1510 1511 default_path = newpath; 1512 } 1513 #endif 1514 1515 static void 1516 startup_modules(void) 1517 { 1518 int cnt; 1519 extern void prom_setup(void); 1520 int32_t v, h; 1521 char d[11]; 1522 char *cp; 1523 cmi_hdl_t hdl; 1524 1525 PRM_POINT("startup_modules() starting..."); 1526 1527 #ifndef __xpv 1528 /* 1529 * Initialize ten-micro second timer so that drivers will 1530 * not get short changed in their init phase. This was 1531 * not getting called until clkinit which, on fast cpu's 1532 * caused the drv_usecwait to be way too short. 1533 */ 1534 microfind(); 1535 1536 if (get_hwenv() == HW_XEN_HVM) 1537 update_default_path(); 1538 #endif 1539 1540 /* 1541 * Read the GMT lag from /etc/rtc_config. 1542 */ 1543 sgmtl(process_rtc_config_file()); 1544 1545 /* 1546 * Calculate default settings of system parameters based upon 1547 * maxusers, yet allow to be overridden via the /etc/system file. 1548 */ 1549 param_calc(0); 1550 1551 mod_setup(); 1552 1553 /* 1554 * Initialize system parameters. 1555 */ 1556 param_init(); 1557 1558 /* 1559 * Initialize the default brands 1560 */ 1561 brand_init(); 1562 1563 /* 1564 * maxmem is the amount of physical memory we're playing with. 1565 */ 1566 maxmem = physmem; 1567 1568 /* 1569 * Initialize segment management stuff. 1570 */ 1571 seg_init(); 1572 1573 if (modload("fs", "specfs") == -1) 1574 halt("Can't load specfs"); 1575 1576 if (modload("fs", "devfs") == -1) 1577 halt("Can't load devfs"); 1578 1579 if (modload("fs", "dev") == -1) 1580 halt("Can't load dev"); 1581 1582 if (modload("fs", "procfs") == -1) 1583 halt("Can't load procfs"); 1584 1585 (void) modloadonly("sys", "lbl_edition"); 1586 1587 dispinit(); 1588 1589 /* 1590 * This is needed here to initialize hw_serial[] for cluster booting. 1591 */ 1592 if ((h = set_soft_hostid()) == HW_INVALID_HOSTID) { 1593 cmn_err(CE_WARN, "Unable to set hostid"); 1594 } else { 1595 for (v = h, cnt = 0; cnt < 10; cnt++) { 1596 d[cnt] = (char)(v % 10); 1597 v /= 10; 1598 if (v == 0) 1599 break; 1600 } 1601 for (cp = hw_serial; cnt >= 0; cnt--) 1602 *cp++ = d[cnt] + '0'; 1603 *cp = 0; 1604 } 1605 1606 /* Read cluster configuration data. */ 1607 clconf_init(); 1608 1609 #if defined(__xpv) 1610 (void) ec_init(); 1611 gnttab_init(); 1612 (void) xs_early_init(); 1613 #endif /* __xpv */ 1614 1615 /* 1616 * Create a kernel device tree. First, create rootnex and 1617 * then invoke bus specific code to probe devices. 1618 */ 1619 setup_ddi(); 1620 1621 #ifdef __xpv 1622 if (DOMAIN_IS_INITDOMAIN(xen_info)) 1623 #endif 1624 { 1625 /* 1626 * Load the System Management BIOS into the global ksmbios 1627 * handle, if an SMBIOS is present on this system. 1628 */ 1629 ksmbios = smbios_open(NULL, SMB_VERSION, ksmbios_flags, NULL); 1630 } 1631 1632 1633 /* 1634 * Set up the CPU module subsystem for the boot cpu in the native 1635 * case, and all physical cpu resource in the xpv dom0 case. 1636 * Modifies the device tree, so this must be done after 1637 * setup_ddi(). 1638 */ 1639 #ifdef __xpv 1640 /* 1641 * If paravirtualized and on dom0 then we initialize all physical 1642 * cpu handles now; if paravirtualized on a domU then do not 1643 * initialize. 1644 */ 1645 if (DOMAIN_IS_INITDOMAIN(xen_info)) { 1646 xen_mc_lcpu_cookie_t cpi; 1647 1648 for (cpi = xen_physcpu_next(NULL); cpi != NULL; 1649 cpi = xen_physcpu_next(cpi)) { 1650 if ((hdl = cmi_init(CMI_HDL_SOLARIS_xVM_MCA, 1651 xen_physcpu_chipid(cpi), xen_physcpu_coreid(cpi), 1652 xen_physcpu_strandid(cpi))) != NULL && 1653 is_x86_feature(x86_featureset, X86FSET_MCA)) 1654 cmi_mca_init(hdl); 1655 } 1656 } 1657 #else 1658 /* 1659 * Initialize a handle for the boot cpu - others will initialize 1660 * as they startup. Do not do this if we know we are in an HVM domU. 1661 */ 1662 if ((get_hwenv() != HW_XEN_HVM) && 1663 (hdl = cmi_init(CMI_HDL_NATIVE, cmi_ntv_hwchipid(CPU), 1664 cmi_ntv_hwcoreid(CPU), cmi_ntv_hwstrandid(CPU))) != NULL && 1665 is_x86_feature(x86_featureset, X86FSET_MCA)) { 1666 cmi_mca_init(hdl); 1667 CPU->cpu_m.mcpu_cmi_hdl = hdl; 1668 } 1669 #endif /* __xpv */ 1670 1671 /* 1672 * Fake a prom tree such that /dev/openprom continues to work 1673 */ 1674 PRM_POINT("startup_modules: calling prom_setup..."); 1675 prom_setup(); 1676 PRM_POINT("startup_modules: done"); 1677 1678 /* 1679 * Load all platform specific modules 1680 */ 1681 PRM_POINT("startup_modules: calling psm_modload..."); 1682 psm_modload(); 1683 1684 PRM_POINT("startup_modules() done"); 1685 } 1686 1687 /* 1688 * claim a "setaside" boot page for use in the kernel 1689 */ 1690 page_t * 1691 boot_claim_page(pfn_t pfn) 1692 { 1693 page_t *pp; 1694 1695 pp = page_numtopp_nolock(pfn); 1696 ASSERT(pp != NULL); 1697 1698 if (PP_ISBOOTPAGES(pp)) { 1699 if (pp->p_next != NULL) 1700 pp->p_next->p_prev = pp->p_prev; 1701 if (pp->p_prev == NULL) 1702 bootpages = pp->p_next; 1703 else 1704 pp->p_prev->p_next = pp->p_next; 1705 } else { 1706 /* 1707 * htable_attach() expects a base pagesize page 1708 */ 1709 if (pp->p_szc != 0) 1710 page_boot_demote(pp); 1711 pp = page_numtopp(pfn, SE_EXCL); 1712 } 1713 return (pp); 1714 } 1715 1716 /* 1717 * Walk through the pagetables looking for pages mapped in by boot. If the 1718 * setaside flag is set the pages are expected to be returned to the 1719 * kernel later in boot, so we add them to the bootpages list. 1720 */ 1721 static void 1722 protect_boot_range(uintptr_t low, uintptr_t high, int setaside) 1723 { 1724 uintptr_t va = low; 1725 size_t len; 1726 uint_t prot; 1727 pfn_t pfn; 1728 page_t *pp; 1729 pgcnt_t boot_protect_cnt = 0; 1730 1731 while (kbm_probe(&va, &len, &pfn, &prot) != 0 && va < high) { 1732 if (va + len >= high) 1733 panic("0x%lx byte mapping at 0x%p exceeds boot's " 1734 "legal range.", len, (void *)va); 1735 1736 while (len > 0) { 1737 pp = page_numtopp_alloc(pfn); 1738 if (pp != NULL) { 1739 if (setaside == 0) 1740 panic("Unexpected mapping by boot. " 1741 "addr=%p pfn=%lx\n", 1742 (void *)va, pfn); 1743 1744 pp->p_next = bootpages; 1745 pp->p_prev = NULL; 1746 PP_SETBOOTPAGES(pp); 1747 if (bootpages != NULL) { 1748 bootpages->p_prev = pp; 1749 } 1750 bootpages = pp; 1751 ++boot_protect_cnt; 1752 } 1753 1754 ++pfn; 1755 len -= MMU_PAGESIZE; 1756 va += MMU_PAGESIZE; 1757 } 1758 } 1759 PRM_DEBUG(boot_protect_cnt); 1760 } 1761 1762 /* 1763 * 1764 */ 1765 static void 1766 layout_kernel_va(void) 1767 { 1768 PRM_POINT("layout_kernel_va() starting..."); 1769 /* 1770 * Establish the final size of the kernel's heap, size of segmap, 1771 * segkp, etc. 1772 */ 1773 1774 #if defined(__amd64) 1775 1776 kpm_vbase = (caddr_t)segkpm_base; 1777 if (physmax + 1 < plat_dr_physmax) { 1778 kpm_size = ROUND_UP_LPAGE(mmu_ptob(plat_dr_physmax)); 1779 } else { 1780 kpm_size = ROUND_UP_LPAGE(mmu_ptob(physmax + 1)); 1781 } 1782 if ((uintptr_t)kpm_vbase + kpm_size > (uintptr_t)valloc_base) 1783 panic("not enough room for kpm!"); 1784 PRM_DEBUG(kpm_size); 1785 PRM_DEBUG(kpm_vbase); 1786 1787 /* 1788 * By default we create a seg_kp in 64 bit kernels, it's a little 1789 * faster to access than embedding it in the heap. 1790 */ 1791 segkp_base = (caddr_t)valloc_base + valloc_sz; 1792 if (!segkp_fromheap) { 1793 size_t sz = mmu_ptob(segkpsize); 1794 1795 /* 1796 * determine size of segkp 1797 */ 1798 if (sz < SEGKPMINSIZE || sz > SEGKPMAXSIZE) { 1799 sz = SEGKPDEFSIZE; 1800 cmn_err(CE_WARN, "!Illegal value for segkpsize. " 1801 "segkpsize has been reset to %ld pages", 1802 mmu_btop(sz)); 1803 } 1804 sz = MIN(sz, MAX(SEGKPMINSIZE, mmu_ptob(physmem))); 1805 1806 segkpsize = mmu_btop(ROUND_UP_LPAGE(sz)); 1807 } 1808 PRM_DEBUG(segkp_base); 1809 PRM_DEBUG(segkpsize); 1810 1811 /* 1812 * segzio is used for ZFS cached data. It uses a distinct VA 1813 * segment (from kernel heap) so that we can easily tell not to 1814 * include it in kernel crash dumps on 64 bit kernels. The trick is 1815 * to give it lots of VA, but not constrain the kernel heap. 1816 * We scale the size of segzio linearly with physmem up to 1817 * SEGZIOMAXSIZE. Above that amount it scales at 50% of physmem. 1818 */ 1819 segzio_base = segkp_base + mmu_ptob(segkpsize); 1820 if (segzio_fromheap) { 1821 segziosize = 0; 1822 } else { 1823 size_t physmem_size = mmu_ptob(physmem); 1824 size_t size = (segziosize == 0) ? 1825 physmem_size : mmu_ptob(segziosize); 1826 1827 if (size < SEGZIOMINSIZE) 1828 size = SEGZIOMINSIZE; 1829 if (size > SEGZIOMAXSIZE) { 1830 size = SEGZIOMAXSIZE; 1831 if (physmem_size > size) 1832 size += (physmem_size - size) / 2; 1833 } 1834 segziosize = mmu_btop(ROUND_UP_LPAGE(size)); 1835 } 1836 PRM_DEBUG(segziosize); 1837 PRM_DEBUG(segzio_base); 1838 1839 /* 1840 * Put the range of VA for device mappings next, kmdb knows to not 1841 * grep in this range of addresses. 1842 */ 1843 toxic_addr = 1844 ROUND_UP_LPAGE((uintptr_t)segzio_base + mmu_ptob(segziosize)); 1845 PRM_DEBUG(toxic_addr); 1846 segmap_start = ROUND_UP_LPAGE(toxic_addr + toxic_size); 1847 #else /* __i386 */ 1848 segmap_start = ROUND_UP_LPAGE(kernelbase); 1849 #endif /* __i386 */ 1850 PRM_DEBUG(segmap_start); 1851 1852 /* 1853 * Users can change segmapsize through eeprom. If the variable 1854 * is tuned through eeprom, there is no upper bound on the 1855 * size of segmap. 1856 */ 1857 segmapsize = MAX(ROUND_UP_LPAGE(segmapsize), SEGMAPDEFAULT); 1858 1859 #if defined(__i386) 1860 /* 1861 * 32-bit systems don't have segkpm or segkp, so segmap appears at 1862 * the bottom of the kernel's address range. Set aside space for a 1863 * small red zone just below the start of segmap. 1864 */ 1865 segmap_start += KERNEL_REDZONE_SIZE; 1866 segmapsize -= KERNEL_REDZONE_SIZE; 1867 #endif 1868 1869 PRM_DEBUG(segmap_start); 1870 PRM_DEBUG(segmapsize); 1871 kernelheap = (caddr_t)ROUND_UP_LPAGE(segmap_start + segmapsize); 1872 PRM_DEBUG(kernelheap); 1873 PRM_POINT("layout_kernel_va() done..."); 1874 } 1875 1876 /* 1877 * Finish initializing the VM system, now that we are no longer 1878 * relying on the boot time memory allocators. 1879 */ 1880 static void 1881 startup_vm(void) 1882 { 1883 struct segmap_crargs a; 1884 1885 extern int use_brk_lpg, use_stk_lpg; 1886 1887 PRM_POINT("startup_vm() starting..."); 1888 1889 /* 1890 * Initialize the hat layer. 1891 */ 1892 hat_init(); 1893 1894 /* 1895 * Do final allocations of HAT data structures that need to 1896 * be allocated before quiescing the boot loader. 1897 */ 1898 PRM_POINT("Calling hat_kern_alloc()..."); 1899 hat_kern_alloc((caddr_t)segmap_start, segmapsize, ekernelheap); 1900 PRM_POINT("hat_kern_alloc() done"); 1901 1902 #ifndef __xpv 1903 /* 1904 * Setup Page Attribute Table 1905 */ 1906 pat_sync(); 1907 #endif 1908 1909 /* 1910 * The next two loops are done in distinct steps in order 1911 * to be sure that any page that is doubly mapped (both above 1912 * KERNEL_TEXT and below kernelbase) is dealt with correctly. 1913 * Note this may never happen, but it might someday. 1914 */ 1915 bootpages = NULL; 1916 PRM_POINT("Protecting boot pages"); 1917 1918 /* 1919 * Protect any pages mapped above KERNEL_TEXT that somehow have 1920 * page_t's. This can only happen if something weird allocated 1921 * in this range (like kadb/kmdb). 1922 */ 1923 protect_boot_range(KERNEL_TEXT, (uintptr_t)-1, 0); 1924 1925 /* 1926 * Before we can take over memory allocation/mapping from the boot 1927 * loader we must remove from our free page lists any boot allocated 1928 * pages that stay mapped until release_bootstrap(). 1929 */ 1930 protect_boot_range(0, kernelbase, 1); 1931 1932 1933 /* 1934 * Switch to running on regular HAT (not boot_mmu) 1935 */ 1936 PRM_POINT("Calling hat_kern_setup()..."); 1937 hat_kern_setup(); 1938 1939 /* 1940 * It is no longer safe to call BOP_ALLOC(), so make sure we don't. 1941 */ 1942 bop_no_more_mem(); 1943 1944 PRM_POINT("hat_kern_setup() done"); 1945 1946 hat_cpu_online(CPU); 1947 1948 /* 1949 * Initialize VM system 1950 */ 1951 PRM_POINT("Calling kvm_init()..."); 1952 kvm_init(); 1953 PRM_POINT("kvm_init() done"); 1954 1955 /* 1956 * Tell kmdb that the VM system is now working 1957 */ 1958 if (boothowto & RB_DEBUG) 1959 kdi_dvec_vmready(); 1960 1961 #if defined(__xpv) 1962 /* 1963 * Populate the I/O pool on domain 0 1964 */ 1965 if (DOMAIN_IS_INITDOMAIN(xen_info)) { 1966 extern long populate_io_pool(void); 1967 long init_io_pool_cnt; 1968 1969 PRM_POINT("Populating reserve I/O page pool"); 1970 init_io_pool_cnt = populate_io_pool(); 1971 PRM_DEBUG(init_io_pool_cnt); 1972 } 1973 #endif 1974 /* 1975 * Mangle the brand string etc. 1976 */ 1977 cpuid_pass3(CPU); 1978 1979 #if defined(__amd64) 1980 1981 /* 1982 * Create the device arena for toxic (to dtrace/kmdb) mappings. 1983 */ 1984 device_arena = vmem_create("device", (void *)toxic_addr, 1985 toxic_size, MMU_PAGESIZE, NULL, NULL, NULL, 0, VM_SLEEP); 1986 1987 #else /* __i386 */ 1988 1989 /* 1990 * allocate the bit map that tracks toxic pages 1991 */ 1992 toxic_bit_map_len = btop((ulong_t)(valloc_base - kernelbase)); 1993 PRM_DEBUG(toxic_bit_map_len); 1994 toxic_bit_map = 1995 kmem_zalloc(BT_SIZEOFMAP(toxic_bit_map_len), KM_NOSLEEP); 1996 ASSERT(toxic_bit_map != NULL); 1997 PRM_DEBUG(toxic_bit_map); 1998 1999 #endif /* __i386 */ 2000 2001 2002 /* 2003 * Now that we've got more VA, as well as the ability to allocate from 2004 * it, tell the debugger. 2005 */ 2006 if (boothowto & RB_DEBUG) 2007 kdi_dvec_memavail(); 2008 2009 /* 2010 * The following code installs a special page fault handler (#pf) 2011 * to work around a pentium bug. 2012 */ 2013 #if !defined(__amd64) && !defined(__xpv) 2014 if (x86_type == X86_TYPE_P5) { 2015 desctbr_t idtr; 2016 gate_desc_t *newidt; 2017 2018 if ((newidt = kmem_zalloc(MMU_PAGESIZE, KM_NOSLEEP)) == NULL) 2019 panic("failed to install pentium_pftrap"); 2020 2021 bcopy(idt0, newidt, NIDT * sizeof (*idt0)); 2022 set_gatesegd(&newidt[T_PGFLT], &pentium_pftrap, 2023 KCS_SEL, SDT_SYSIGT, TRP_KPL, 0); 2024 2025 (void) as_setprot(&kas, (caddr_t)newidt, MMU_PAGESIZE, 2026 PROT_READ | PROT_EXEC); 2027 2028 CPU->cpu_idt = newidt; 2029 idtr.dtr_base = (uintptr_t)CPU->cpu_idt; 2030 idtr.dtr_limit = (NIDT * sizeof (*idt0)) - 1; 2031 wr_idtr(&idtr); 2032 } 2033 #endif /* !__amd64 */ 2034 2035 #if !defined(__xpv) 2036 /* 2037 * Map page pfn=0 for drivers, such as kd, that need to pick up 2038 * parameters left there by controllers/BIOS. 2039 */ 2040 PRM_POINT("setup up p0_va"); 2041 p0_va = i86devmap(0, 1, PROT_READ); 2042 PRM_DEBUG(p0_va); 2043 #endif 2044 2045 cmn_err(CE_CONT, "?mem = %luK (0x%lx)\n", 2046 physinstalled << (MMU_PAGESHIFT - 10), ptob(physinstalled)); 2047 2048 /* 2049 * disable automatic large pages for small memory systems or 2050 * when the disable flag is set. 2051 * 2052 * Do not yet consider page sizes larger than 2m/4m. 2053 */ 2054 if (!auto_lpg_disable && mmu.max_page_level > 0) { 2055 max_uheap_lpsize = LEVEL_SIZE(1); 2056 max_ustack_lpsize = LEVEL_SIZE(1); 2057 max_privmap_lpsize = LEVEL_SIZE(1); 2058 max_uidata_lpsize = LEVEL_SIZE(1); 2059 max_utext_lpsize = LEVEL_SIZE(1); 2060 max_shm_lpsize = LEVEL_SIZE(1); 2061 } 2062 if (physmem < privm_lpg_min_physmem || mmu.max_page_level == 0 || 2063 auto_lpg_disable) { 2064 use_brk_lpg = 0; 2065 use_stk_lpg = 0; 2066 } 2067 mcntl0_lpsize = LEVEL_SIZE(mmu.umax_page_level); 2068 2069 PRM_POINT("Calling hat_init_finish()..."); 2070 hat_init_finish(); 2071 PRM_POINT("hat_init_finish() done"); 2072 2073 /* 2074 * Initialize the segkp segment type. 2075 */ 2076 rw_enter(&kas.a_lock, RW_WRITER); 2077 PRM_POINT("Attaching segkp"); 2078 if (segkp_fromheap) { 2079 segkp->s_as = &kas; 2080 } else if (seg_attach(&kas, (caddr_t)segkp_base, mmu_ptob(segkpsize), 2081 segkp) < 0) { 2082 panic("startup: cannot attach segkp"); 2083 /*NOTREACHED*/ 2084 } 2085 PRM_POINT("Doing segkp_create()"); 2086 if (segkp_create(segkp) != 0) { 2087 panic("startup: segkp_create failed"); 2088 /*NOTREACHED*/ 2089 } 2090 PRM_DEBUG(segkp); 2091 rw_exit(&kas.a_lock); 2092 2093 /* 2094 * kpm segment 2095 */ 2096 segmap_kpm = 0; 2097 if (kpm_desired) { 2098 kpm_init(); 2099 kpm_enable = 1; 2100 } 2101 2102 /* 2103 * Now create segmap segment. 2104 */ 2105 rw_enter(&kas.a_lock, RW_WRITER); 2106 if (seg_attach(&kas, (caddr_t)segmap_start, segmapsize, segmap) < 0) { 2107 panic("cannot attach segmap"); 2108 /*NOTREACHED*/ 2109 } 2110 PRM_DEBUG(segmap); 2111 2112 a.prot = PROT_READ | PROT_WRITE; 2113 a.shmsize = 0; 2114 a.nfreelist = segmapfreelists; 2115 2116 if (segmap_create(segmap, (caddr_t)&a) != 0) 2117 panic("segmap_create segmap"); 2118 rw_exit(&kas.a_lock); 2119 2120 setup_vaddr_for_ppcopy(CPU); 2121 2122 segdev_init(); 2123 #if defined(__xpv) 2124 if (DOMAIN_IS_INITDOMAIN(xen_info)) 2125 #endif 2126 pmem_init(); 2127 2128 PRM_POINT("startup_vm() done"); 2129 } 2130 2131 /* 2132 * Load a tod module for the non-standard tod part found on this system. 2133 */ 2134 static void 2135 load_tod_module(char *todmod) 2136 { 2137 if (modload("tod", todmod) == -1) 2138 halt("Can't load TOD module"); 2139 } 2140 2141 static void 2142 startup_end(void) 2143 { 2144 int i; 2145 extern void setx86isalist(void); 2146 extern void cpu_event_init(void); 2147 2148 PRM_POINT("startup_end() starting..."); 2149 2150 /* 2151 * Perform tasks that get done after most of the VM 2152 * initialization has been done but before the clock 2153 * and other devices get started. 2154 */ 2155 kern_setup1(); 2156 2157 /* 2158 * Perform CPC initialization for this CPU. 2159 */ 2160 kcpc_hw_init(CPU); 2161 2162 /* 2163 * Initialize cpu event framework. 2164 */ 2165 cpu_event_init(); 2166 2167 #if defined(OPTERON_WORKAROUND_6323525) 2168 if (opteron_workaround_6323525) 2169 patch_workaround_6323525(); 2170 #endif 2171 /* 2172 * If needed, load TOD module now so that ddi_get_time(9F) etc. work 2173 * (For now, "needed" is defined as set tod_module_name in /etc/system) 2174 */ 2175 if (tod_module_name != NULL) { 2176 PRM_POINT("load_tod_module()"); 2177 load_tod_module(tod_module_name); 2178 } 2179 2180 #if defined(__xpv) 2181 /* 2182 * Forceload interposing TOD module for the hypervisor. 2183 */ 2184 PRM_POINT("load_tod_module()"); 2185 load_tod_module("xpvtod"); 2186 #endif 2187 2188 /* 2189 * Configure the system. 2190 */ 2191 PRM_POINT("Calling configure()..."); 2192 configure(); /* set up devices */ 2193 PRM_POINT("configure() done"); 2194 2195 /* 2196 * Set the isa_list string to the defined instruction sets we 2197 * support. 2198 */ 2199 setx86isalist(); 2200 cpu_intr_alloc(CPU, NINTR_THREADS); 2201 psm_install(); 2202 2203 /* 2204 * We're done with bootops. We don't unmap the bootstrap yet because 2205 * we're still using bootsvcs. 2206 */ 2207 PRM_POINT("NULLing out bootops"); 2208 *bootopsp = (struct bootops *)NULL; 2209 bootops = (struct bootops *)NULL; 2210 2211 #if defined(__xpv) 2212 ec_init_debug_irq(); 2213 xs_domu_init(); 2214 #endif 2215 2216 #if defined(__amd64) && !defined(__xpv) 2217 /* 2218 * Intel IOMMU has been setup/initialized in ddi_impl.c 2219 * Start it up now. 2220 */ 2221 immu_startup(); 2222 #endif 2223 2224 PRM_POINT("Enabling interrupts"); 2225 (*picinitf)(); 2226 sti(); 2227 #if defined(__xpv) 2228 ASSERT(CPU->cpu_m.mcpu_vcpu_info->evtchn_upcall_mask == 0); 2229 xen_late_startup(); 2230 #endif 2231 2232 (void) add_avsoftintr((void *)&softlevel1_hdl, 1, softlevel1, 2233 "softlevel1", NULL, NULL); /* XXX to be moved later */ 2234 2235 /* 2236 * Register these software interrupts for ddi timer. 2237 * Software interrupts up to the level 10 are supported. 2238 */ 2239 for (i = DDI_IPL_1; i <= DDI_IPL_10; i++) { 2240 char name[sizeof ("timer_softintr") + 2]; 2241 (void) sprintf(name, "timer_softintr%02d", i); 2242 (void) add_avsoftintr((void *)&softlevel_hdl[i-1], i, 2243 (avfunc)timer_softintr, name, (caddr_t)(uintptr_t)i, NULL); 2244 } 2245 2246 #if !defined(__xpv) 2247 if (modload("drv", "amd_iommu") < 0) { 2248 PRM_POINT("No AMD IOMMU present\n"); 2249 } else if (ddi_hold_installed_driver(ddi_name_to_major( 2250 "amd_iommu")) == NULL) { 2251 prom_printf("ERROR: failed to attach AMD IOMMU\n"); 2252 } 2253 #endif 2254 post_startup_cpu_fixups(); 2255 2256 PRM_POINT("startup_end() done"); 2257 } 2258 2259 /* 2260 * Don't remove the following 2 variables. They are necessary 2261 * for reading the hostid from the legacy file (/kernel/misc/sysinit). 2262 */ 2263 char *_hs1107 = hw_serial; 2264 ulong_t _bdhs34; 2265 2266 void 2267 post_startup(void) 2268 { 2269 extern void cpupm_init(cpu_t *); 2270 extern void cpu_event_init_cpu(cpu_t *); 2271 2272 /* 2273 * Set the system wide, processor-specific flags to be passed 2274 * to userland via the aux vector for performance hints and 2275 * instruction set extensions. 2276 */ 2277 bind_hwcap(); 2278 2279 #ifdef __xpv 2280 if (DOMAIN_IS_INITDOMAIN(xen_info)) 2281 #endif 2282 { 2283 #if defined(__xpv) 2284 xpv_panic_init(); 2285 #else 2286 /* 2287 * Startup the memory scrubber. 2288 * XXPV This should be running somewhere .. 2289 */ 2290 if (get_hwenv() != HW_XEN_HVM) 2291 memscrub_init(); 2292 #endif 2293 } 2294 2295 /* 2296 * Complete CPU module initialization 2297 */ 2298 cmi_post_startup(); 2299 2300 /* 2301 * Perform forceloading tasks for /etc/system. 2302 */ 2303 (void) mod_sysctl(SYS_FORCELOAD, NULL); 2304 2305 /* 2306 * ON4.0: Force /proc module in until clock interrupt handle fixed 2307 * ON4.0: This must be fixed or restated in /etc/systems. 2308 */ 2309 (void) modload("fs", "procfs"); 2310 2311 (void) i_ddi_attach_hw_nodes("pit_beep"); 2312 2313 #if defined(__i386) 2314 /* 2315 * Check for required functional Floating Point hardware, 2316 * unless FP hardware explicitly disabled. 2317 */ 2318 if (fpu_exists && (fpu_pentium_fdivbug || fp_kind == FP_NO)) 2319 halt("No working FP hardware found"); 2320 #endif 2321 2322 maxmem = freemem; 2323 2324 cpu_event_init_cpu(CPU); 2325 cpupm_init(CPU); 2326 (void) mach_cpu_create_device_node(CPU, NULL); 2327 2328 pg_init(); 2329 } 2330 2331 static int 2332 pp_in_range(page_t *pp, uint64_t low_addr, uint64_t high_addr) 2333 { 2334 return ((pp->p_pagenum >= btop(low_addr)) && 2335 (pp->p_pagenum < btopr(high_addr))); 2336 } 2337 2338 void 2339 release_bootstrap(void) 2340 { 2341 int root_is_ramdisk; 2342 page_t *pp; 2343 extern void kobj_boot_unmountroot(void); 2344 extern dev_t rootdev; 2345 #if !defined(__xpv) 2346 pfn_t pfn; 2347 #endif 2348 2349 /* unmount boot ramdisk and release kmem usage */ 2350 kobj_boot_unmountroot(); 2351 2352 /* 2353 * We're finished using the boot loader so free its pages. 2354 */ 2355 PRM_POINT("Unmapping lower boot pages"); 2356 2357 clear_boot_mappings(0, _userlimit); 2358 2359 postbootkernelbase = kernelbase; 2360 2361 /* 2362 * If root isn't on ramdisk, destroy the hardcoded 2363 * ramdisk node now and release the memory. Else, 2364 * ramdisk memory is kept in rd_pages. 2365 */ 2366 root_is_ramdisk = (getmajor(rootdev) == ddi_name_to_major("ramdisk")); 2367 if (!root_is_ramdisk) { 2368 dev_info_t *dip = ddi_find_devinfo("ramdisk", -1, 0); 2369 ASSERT(dip && ddi_get_parent(dip) == ddi_root_node()); 2370 ndi_rele_devi(dip); /* held from ddi_find_devinfo */ 2371 (void) ddi_remove_child(dip, 0); 2372 } 2373 2374 PRM_POINT("Releasing boot pages"); 2375 while (bootpages) { 2376 extern uint64_t ramdisk_start, ramdisk_end; 2377 pp = bootpages; 2378 bootpages = pp->p_next; 2379 2380 2381 /* Keep pages for the lower 64K */ 2382 if (pp_in_range(pp, 0, 0x40000)) { 2383 pp->p_next = lower_pages; 2384 lower_pages = pp; 2385 lower_pages_count++; 2386 continue; 2387 } 2388 2389 2390 if (root_is_ramdisk && pp_in_range(pp, ramdisk_start, 2391 ramdisk_end)) { 2392 pp->p_next = rd_pages; 2393 rd_pages = pp; 2394 continue; 2395 } 2396 pp->p_next = (struct page *)0; 2397 pp->p_prev = (struct page *)0; 2398 PP_CLRBOOTPAGES(pp); 2399 page_free(pp, 1); 2400 } 2401 PRM_POINT("Boot pages released"); 2402 2403 #if !defined(__xpv) 2404 /* XXPV -- note this following bunch of code needs to be revisited in Xen 3.0 */ 2405 /* 2406 * Find 1 page below 1 MB so that other processors can boot up or 2407 * so that any processor can resume. 2408 * Make sure it has a kernel VA as well as a 1:1 mapping. 2409 * We should have just free'd one up. 2410 */ 2411 2412 /* 2413 * 0x10 pages is 64K. Leave the bottom 64K alone 2414 * for BIOS. 2415 */ 2416 for (pfn = 0x10; pfn < btop(1*1024*1024); pfn++) { 2417 if (page_numtopp_alloc(pfn) == NULL) 2418 continue; 2419 rm_platter_va = i86devmap(pfn, 1, 2420 PROT_READ | PROT_WRITE | PROT_EXEC); 2421 rm_platter_pa = ptob(pfn); 2422 break; 2423 } 2424 if (pfn == btop(1*1024*1024) && use_mp) 2425 panic("No page below 1M available for starting " 2426 "other processors or for resuming from system-suspend"); 2427 #endif /* !__xpv */ 2428 } 2429 2430 /* 2431 * Initialize the platform-specific parts of a page_t. 2432 */ 2433 void 2434 add_physmem_cb(page_t *pp, pfn_t pnum) 2435 { 2436 pp->p_pagenum = pnum; 2437 pp->p_mapping = NULL; 2438 pp->p_embed = 0; 2439 pp->p_share = 0; 2440 pp->p_mlentry = 0; 2441 } 2442 2443 /* 2444 * kphysm_init() initializes physical memory. 2445 */ 2446 static pgcnt_t 2447 kphysm_init( 2448 page_t *pp, 2449 pgcnt_t npages) 2450 { 2451 struct memlist *pmem; 2452 struct memseg *cur_memseg; 2453 pfn_t base_pfn; 2454 pfn_t end_pfn; 2455 pgcnt_t num; 2456 pgcnt_t pages_done = 0; 2457 uint64_t addr; 2458 uint64_t size; 2459 extern pfn_t ddiphysmin; 2460 extern int mnode_xwa; 2461 int ms = 0, me = 0; 2462 2463 ASSERT(page_hash != NULL && page_hashsz != 0); 2464 2465 cur_memseg = memseg_base; 2466 for (pmem = phys_avail; pmem && npages; pmem = pmem->ml_next) { 2467 /* 2468 * In a 32 bit kernel can't use higher memory if we're 2469 * not booting in PAE mode. This check takes care of that. 2470 */ 2471 addr = pmem->ml_address; 2472 size = pmem->ml_size; 2473 if (btop(addr) > physmax) 2474 continue; 2475 2476 /* 2477 * align addr and size - they may not be at page boundaries 2478 */ 2479 if ((addr & MMU_PAGEOFFSET) != 0) { 2480 addr += MMU_PAGEOFFSET; 2481 addr &= ~(uint64_t)MMU_PAGEOFFSET; 2482 size -= addr - pmem->ml_address; 2483 } 2484 2485 /* only process pages below or equal to physmax */ 2486 if ((btop(addr + size) - 1) > physmax) 2487 size = ptob(physmax - btop(addr) + 1); 2488 2489 num = btop(size); 2490 if (num == 0) 2491 continue; 2492 2493 if (num > npages) 2494 num = npages; 2495 2496 npages -= num; 2497 pages_done += num; 2498 base_pfn = btop(addr); 2499 2500 if (prom_debug) 2501 prom_printf("MEMSEG addr=0x%" PRIx64 2502 " pgs=0x%lx pfn 0x%lx-0x%lx\n", 2503 addr, num, base_pfn, base_pfn + num); 2504 2505 /* 2506 * Ignore pages below ddiphysmin to simplify ddi memory 2507 * allocation with non-zero addr_lo requests. 2508 */ 2509 if (base_pfn < ddiphysmin) { 2510 if (base_pfn + num <= ddiphysmin) 2511 continue; 2512 pp += (ddiphysmin - base_pfn); 2513 num -= (ddiphysmin - base_pfn); 2514 base_pfn = ddiphysmin; 2515 } 2516 2517 /* 2518 * mnode_xwa is greater than 1 when large pages regions can 2519 * cross memory node boundaries. To prevent the formation 2520 * of these large pages, configure the memsegs based on the 2521 * memory node ranges which had been made non-contiguous. 2522 */ 2523 if (mnode_xwa > 1) { 2524 2525 end_pfn = base_pfn + num - 1; 2526 ms = PFN_2_MEM_NODE(base_pfn); 2527 me = PFN_2_MEM_NODE(end_pfn); 2528 2529 if (ms != me) { 2530 /* 2531 * current range spans more than 1 memory node. 2532 * Set num to only the pfn range in the start 2533 * memory node. 2534 */ 2535 num = mem_node_config[ms].physmax - base_pfn 2536 + 1; 2537 ASSERT(end_pfn > mem_node_config[ms].physmax); 2538 } 2539 } 2540 2541 for (;;) { 2542 /* 2543 * Build the memsegs entry 2544 */ 2545 cur_memseg->pages = pp; 2546 cur_memseg->epages = pp + num; 2547 cur_memseg->pages_base = base_pfn; 2548 cur_memseg->pages_end = base_pfn + num; 2549 2550 /* 2551 * Insert into memseg list in decreasing pfn range 2552 * order. Low memory is typically more fragmented such 2553 * that this ordering keeps the larger ranges at the 2554 * front of the list for code that searches memseg. 2555 * This ASSERTS that the memsegs coming in from boot 2556 * are in increasing physical address order and not 2557 * contiguous. 2558 */ 2559 if (memsegs != NULL) { 2560 ASSERT(cur_memseg->pages_base >= 2561 memsegs->pages_end); 2562 cur_memseg->next = memsegs; 2563 } 2564 memsegs = cur_memseg; 2565 2566 /* 2567 * add_physmem() initializes the PSM part of the page 2568 * struct by calling the PSM back with add_physmem_cb(). 2569 * In addition it coalesces pages into larger pages as 2570 * it initializes them. 2571 */ 2572 add_physmem(pp, num, base_pfn); 2573 cur_memseg++; 2574 availrmem_initial += num; 2575 availrmem += num; 2576 2577 pp += num; 2578 if (ms >= me) 2579 break; 2580 2581 /* process next memory node range */ 2582 ms++; 2583 base_pfn = mem_node_config[ms].physbase; 2584 num = MIN(mem_node_config[ms].physmax, 2585 end_pfn) - base_pfn + 1; 2586 } 2587 } 2588 2589 PRM_DEBUG(availrmem_initial); 2590 PRM_DEBUG(availrmem); 2591 PRM_DEBUG(freemem); 2592 build_pfn_hash(); 2593 return (pages_done); 2594 } 2595 2596 /* 2597 * Kernel VM initialization. 2598 */ 2599 static void 2600 kvm_init(void) 2601 { 2602 ASSERT((((uintptr_t)s_text) & MMU_PAGEOFFSET) == 0); 2603 2604 /* 2605 * Put the kernel segments in kernel address space. 2606 */ 2607 rw_enter(&kas.a_lock, RW_WRITER); 2608 as_avlinit(&kas); 2609 2610 (void) seg_attach(&kas, s_text, e_moddata - s_text, &ktextseg); 2611 (void) segkmem_create(&ktextseg); 2612 2613 (void) seg_attach(&kas, (caddr_t)valloc_base, valloc_sz, &kvalloc); 2614 (void) segkmem_create(&kvalloc); 2615 2616 (void) seg_attach(&kas, kernelheap, 2617 ekernelheap - kernelheap, &kvseg); 2618 (void) segkmem_create(&kvseg); 2619 2620 if (core_size > 0) { 2621 PRM_POINT("attaching kvseg_core"); 2622 (void) seg_attach(&kas, (caddr_t)core_base, core_size, 2623 &kvseg_core); 2624 (void) segkmem_create(&kvseg_core); 2625 } 2626 2627 if (segziosize > 0) { 2628 PRM_POINT("attaching segzio"); 2629 (void) seg_attach(&kas, segzio_base, mmu_ptob(segziosize), 2630 &kzioseg); 2631 (void) segkmem_zio_create(&kzioseg); 2632 2633 /* create zio area covering new segment */ 2634 segkmem_zio_init(segzio_base, mmu_ptob(segziosize)); 2635 } 2636 2637 (void) seg_attach(&kas, kdi_segdebugbase, kdi_segdebugsize, &kdebugseg); 2638 (void) segkmem_create(&kdebugseg); 2639 2640 rw_exit(&kas.a_lock); 2641 2642 /* 2643 * Ensure that the red zone at kernelbase is never accessible. 2644 */ 2645 PRM_POINT("protecting redzone"); 2646 (void) as_setprot(&kas, (caddr_t)kernelbase, KERNEL_REDZONE_SIZE, 0); 2647 2648 /* 2649 * Make the text writable so that it can be hot patched by DTrace. 2650 */ 2651 (void) as_setprot(&kas, s_text, e_modtext - s_text, 2652 PROT_READ | PROT_WRITE | PROT_EXEC); 2653 2654 /* 2655 * Make data writable until end. 2656 */ 2657 (void) as_setprot(&kas, s_data, e_moddata - s_data, 2658 PROT_READ | PROT_WRITE | PROT_EXEC); 2659 } 2660 2661 #ifndef __xpv 2662 /* 2663 * Solaris adds an entry for Write Combining caching to the PAT 2664 */ 2665 static uint64_t pat_attr_reg = PAT_DEFAULT_ATTRIBUTE; 2666 2667 void 2668 pat_sync(void) 2669 { 2670 ulong_t cr0, cr0_orig, cr4; 2671 2672 if (!is_x86_feature(x86_featureset, X86FSET_PAT)) 2673 return; 2674 cr0_orig = cr0 = getcr0(); 2675 cr4 = getcr4(); 2676 2677 /* disable caching and flush all caches and TLBs */ 2678 cr0 |= CR0_CD; 2679 cr0 &= ~CR0_NW; 2680 setcr0(cr0); 2681 invalidate_cache(); 2682 if (cr4 & CR4_PGE) { 2683 setcr4(cr4 & ~(ulong_t)CR4_PGE); 2684 setcr4(cr4); 2685 } else { 2686 reload_cr3(); 2687 } 2688 2689 /* add our entry to the PAT */ 2690 wrmsr(REG_PAT, pat_attr_reg); 2691 2692 /* flush TLBs and cache again, then reenable cr0 caching */ 2693 if (cr4 & CR4_PGE) { 2694 setcr4(cr4 & ~(ulong_t)CR4_PGE); 2695 setcr4(cr4); 2696 } else { 2697 reload_cr3(); 2698 } 2699 invalidate_cache(); 2700 setcr0(cr0_orig); 2701 } 2702 2703 #endif /* !__xpv */ 2704 2705 #if defined(_SOFT_HOSTID) 2706 /* 2707 * On platforms that do not have a hardware serial number, attempt 2708 * to set one based on the contents of /etc/hostid. If this file does 2709 * not exist, assume that we are to generate a new hostid and set 2710 * it in the kernel, for subsequent saving by a userland process 2711 * once the system is up and the root filesystem is mounted r/w. 2712 * 2713 * In order to gracefully support upgrade on OpenSolaris, if 2714 * /etc/hostid does not exist, we will attempt to get a serial number 2715 * using the legacy method (/kernel/misc/sysinit). 2716 * 2717 * In an attempt to make the hostid less prone to abuse 2718 * (for license circumvention, etc), we store it in /etc/hostid 2719 * in rot47 format. 2720 */ 2721 extern volatile unsigned long tenmicrodata; 2722 static int atoi(char *); 2723 2724 static int32_t 2725 set_soft_hostid(void) 2726 { 2727 struct _buf *file; 2728 char tokbuf[MAXNAMELEN]; 2729 token_t token; 2730 int done = 0; 2731 u_longlong_t tmp; 2732 int i; 2733 int32_t hostid = (int32_t)HW_INVALID_HOSTID; 2734 unsigned char *c; 2735 hrtime_t tsc; 2736 2737 /* 2738 * If /etc/hostid file not found, we'd like to get a pseudo 2739 * random number to use at the hostid. A nice way to do this 2740 * is to read the real time clock. To remain xen-compatible, 2741 * we can't poke the real hardware, so we use tsc_read() to 2742 * read the real time clock. However, there is an ominous 2743 * warning in tsc_read that says it can return zero, so we 2744 * deal with that possibility by falling back to using the 2745 * (hopefully random enough) value in tenmicrodata. 2746 */ 2747 2748 if ((file = kobj_open_file(hostid_file)) == (struct _buf *)-1) { 2749 /* 2750 * hostid file not found - try to load sysinit module 2751 * and see if it has a nonzero hostid value...use that 2752 * instead of generating a new hostid here if so. 2753 */ 2754 if ((i = modload("misc", "sysinit")) != -1) { 2755 if (strlen(hw_serial) > 0) 2756 hostid = (int32_t)atoi(hw_serial); 2757 (void) modunload(i); 2758 } 2759 if (hostid == HW_INVALID_HOSTID) { 2760 tsc = tsc_read(); 2761 if (tsc == 0) /* tsc_read can return zero sometimes */ 2762 hostid = (int32_t)tenmicrodata & 0x0CFFFFF; 2763 else 2764 hostid = (int32_t)tsc & 0x0CFFFFF; 2765 } 2766 } else { 2767 /* hostid file found */ 2768 while (!done) { 2769 token = kobj_lex(file, tokbuf, sizeof (tokbuf)); 2770 2771 switch (token) { 2772 case POUND: 2773 /* 2774 * skip comments 2775 */ 2776 kobj_find_eol(file); 2777 break; 2778 case STRING: 2779 /* 2780 * un-rot47 - obviously this 2781 * nonsense is ascii-specific 2782 */ 2783 for (c = (unsigned char *)tokbuf; 2784 *c != '\0'; c++) { 2785 *c += 47; 2786 if (*c > '~') 2787 *c -= 94; 2788 else if (*c < '!') 2789 *c += 94; 2790 } 2791 /* 2792 * now we should have a real number 2793 */ 2794 2795 if (kobj_getvalue(tokbuf, &tmp) != 0) 2796 kobj_file_err(CE_WARN, file, 2797 "Bad value %s for hostid", 2798 tokbuf); 2799 else 2800 hostid = (int32_t)tmp; 2801 2802 break; 2803 case EOF: 2804 done = 1; 2805 /* FALLTHROUGH */ 2806 case NEWLINE: 2807 kobj_newline(file); 2808 break; 2809 default: 2810 break; 2811 2812 } 2813 } 2814 if (hostid == HW_INVALID_HOSTID) /* didn't find a hostid */ 2815 kobj_file_err(CE_WARN, file, 2816 "hostid missing or corrupt"); 2817 2818 kobj_close_file(file); 2819 } 2820 /* 2821 * hostid is now the value read from /etc/hostid, or the 2822 * new hostid we generated in this routine or HW_INVALID_HOSTID if not 2823 * set. 2824 */ 2825 return (hostid); 2826 } 2827 2828 static int 2829 atoi(char *p) 2830 { 2831 int i = 0; 2832 2833 while (*p != '\0') 2834 i = 10 * i + (*p++ - '0'); 2835 2836 return (i); 2837 } 2838 2839 #endif /* _SOFT_HOSTID */ 2840 2841 void 2842 get_system_configuration(void) 2843 { 2844 char prop[32]; 2845 u_longlong_t nodes_ll, cpus_pernode_ll, lvalue; 2846 2847 if (BOP_GETPROPLEN(bootops, "nodes") > sizeof (prop) || 2848 BOP_GETPROP(bootops, "nodes", prop) < 0 || 2849 kobj_getvalue(prop, &nodes_ll) == -1 || 2850 nodes_ll > MAXNODES || 2851 BOP_GETPROPLEN(bootops, "cpus_pernode") > sizeof (prop) || 2852 BOP_GETPROP(bootops, "cpus_pernode", prop) < 0 || 2853 kobj_getvalue(prop, &cpus_pernode_ll) == -1) { 2854 system_hardware.hd_nodes = 1; 2855 system_hardware.hd_cpus_per_node = 0; 2856 } else { 2857 system_hardware.hd_nodes = (int)nodes_ll; 2858 system_hardware.hd_cpus_per_node = (int)cpus_pernode_ll; 2859 } 2860 2861 if (BOP_GETPROPLEN(bootops, "kernelbase") > sizeof (prop) || 2862 BOP_GETPROP(bootops, "kernelbase", prop) < 0 || 2863 kobj_getvalue(prop, &lvalue) == -1) 2864 eprom_kernelbase = NULL; 2865 else 2866 eprom_kernelbase = (uintptr_t)lvalue; 2867 2868 if (BOP_GETPROPLEN(bootops, "segmapsize") > sizeof (prop) || 2869 BOP_GETPROP(bootops, "segmapsize", prop) < 0 || 2870 kobj_getvalue(prop, &lvalue) == -1) 2871 segmapsize = SEGMAPDEFAULT; 2872 else 2873 segmapsize = (uintptr_t)lvalue; 2874 2875 if (BOP_GETPROPLEN(bootops, "segmapfreelists") > sizeof (prop) || 2876 BOP_GETPROP(bootops, "segmapfreelists", prop) < 0 || 2877 kobj_getvalue(prop, &lvalue) == -1) 2878 segmapfreelists = 0; /* use segmap driver default */ 2879 else 2880 segmapfreelists = (int)lvalue; 2881 2882 /* physmem used to be here, but moved much earlier to fakebop.c */ 2883 } 2884 2885 /* 2886 * Add to a memory list. 2887 * start = start of new memory segment 2888 * len = length of new memory segment in bytes 2889 * new = pointer to a new struct memlist 2890 * memlistp = memory list to which to add segment. 2891 */ 2892 void 2893 memlist_add( 2894 uint64_t start, 2895 uint64_t len, 2896 struct memlist *new, 2897 struct memlist **memlistp) 2898 { 2899 struct memlist *cur; 2900 uint64_t end = start + len; 2901 2902 new->ml_address = start; 2903 new->ml_size = len; 2904 2905 cur = *memlistp; 2906 2907 while (cur) { 2908 if (cur->ml_address >= end) { 2909 new->ml_next = cur; 2910 *memlistp = new; 2911 new->ml_prev = cur->ml_prev; 2912 cur->ml_prev = new; 2913 return; 2914 } 2915 ASSERT(cur->ml_address + cur->ml_size <= start); 2916 if (cur->ml_next == NULL) { 2917 cur->ml_next = new; 2918 new->ml_prev = cur; 2919 new->ml_next = NULL; 2920 return; 2921 } 2922 memlistp = &cur->ml_next; 2923 cur = cur->ml_next; 2924 } 2925 } 2926 2927 void 2928 kobj_vmem_init(vmem_t **text_arena, vmem_t **data_arena) 2929 { 2930 size_t tsize = e_modtext - modtext; 2931 size_t dsize = e_moddata - moddata; 2932 2933 *text_arena = vmem_create("module_text", tsize ? modtext : NULL, tsize, 2934 1, segkmem_alloc, segkmem_free, heaptext_arena, 0, VM_SLEEP); 2935 *data_arena = vmem_create("module_data", dsize ? moddata : NULL, dsize, 2936 1, segkmem_alloc, segkmem_free, heap32_arena, 0, VM_SLEEP); 2937 } 2938 2939 caddr_t 2940 kobj_text_alloc(vmem_t *arena, size_t size) 2941 { 2942 return (vmem_alloc(arena, size, VM_SLEEP | VM_BESTFIT)); 2943 } 2944 2945 /*ARGSUSED*/ 2946 caddr_t 2947 kobj_texthole_alloc(caddr_t addr, size_t size) 2948 { 2949 panic("unexpected call to kobj_texthole_alloc()"); 2950 /*NOTREACHED*/ 2951 return (0); 2952 } 2953 2954 /*ARGSUSED*/ 2955 void 2956 kobj_texthole_free(caddr_t addr, size_t size) 2957 { 2958 panic("unexpected call to kobj_texthole_free()"); 2959 } 2960 2961 /* 2962 * This is called just after configure() in startup(). 2963 * 2964 * The ISALIST concept is a bit hopeless on Intel, because 2965 * there's no guarantee of an ever-more-capable processor 2966 * given that various parts of the instruction set may appear 2967 * and disappear between different implementations. 2968 * 2969 * While it would be possible to correct it and even enhance 2970 * it somewhat, the explicit hardware capability bitmask allows 2971 * more flexibility. 2972 * 2973 * So, we just leave this alone. 2974 */ 2975 void 2976 setx86isalist(void) 2977 { 2978 char *tp; 2979 size_t len; 2980 extern char *isa_list; 2981 2982 #define TBUFSIZE 1024 2983 2984 tp = kmem_alloc(TBUFSIZE, KM_SLEEP); 2985 *tp = '\0'; 2986 2987 #if defined(__amd64) 2988 (void) strcpy(tp, "amd64 "); 2989 #endif 2990 2991 switch (x86_vendor) { 2992 case X86_VENDOR_Intel: 2993 case X86_VENDOR_AMD: 2994 case X86_VENDOR_TM: 2995 if (is_x86_feature(x86_featureset, X86FSET_CMOV)) { 2996 /* 2997 * Pentium Pro or later 2998 */ 2999 (void) strcat(tp, "pentium_pro"); 3000 (void) strcat(tp, 3001 is_x86_feature(x86_featureset, X86FSET_MMX) ? 3002 "+mmx pentium_pro " : " "); 3003 } 3004 /*FALLTHROUGH*/ 3005 case X86_VENDOR_Cyrix: 3006 /* 3007 * The Cyrix 6x86 does not have any Pentium features 3008 * accessible while not at privilege level 0. 3009 */ 3010 if (is_x86_feature(x86_featureset, X86FSET_CPUID)) { 3011 (void) strcat(tp, "pentium"); 3012 (void) strcat(tp, 3013 is_x86_feature(x86_featureset, X86FSET_MMX) ? 3014 "+mmx pentium " : " "); 3015 } 3016 break; 3017 default: 3018 break; 3019 } 3020 (void) strcat(tp, "i486 i386 i86"); 3021 len = strlen(tp) + 1; /* account for NULL at end of string */ 3022 isa_list = strcpy(kmem_alloc(len, KM_SLEEP), tp); 3023 kmem_free(tp, TBUFSIZE); 3024 3025 #undef TBUFSIZE 3026 } 3027 3028 3029 #ifdef __amd64 3030 3031 void * 3032 device_arena_alloc(size_t size, int vm_flag) 3033 { 3034 return (vmem_alloc(device_arena, size, vm_flag)); 3035 } 3036 3037 void 3038 device_arena_free(void *vaddr, size_t size) 3039 { 3040 vmem_free(device_arena, vaddr, size); 3041 } 3042 3043 #else /* __i386 */ 3044 3045 void * 3046 device_arena_alloc(size_t size, int vm_flag) 3047 { 3048 caddr_t vaddr; 3049 uintptr_t v; 3050 size_t start; 3051 size_t end; 3052 3053 vaddr = vmem_alloc(heap_arena, size, vm_flag); 3054 if (vaddr == NULL) 3055 return (NULL); 3056 3057 v = (uintptr_t)vaddr; 3058 ASSERT(v >= kernelbase); 3059 ASSERT(v + size <= valloc_base); 3060 3061 start = btop(v - kernelbase); 3062 end = btop(v + size - 1 - kernelbase); 3063 ASSERT(start < toxic_bit_map_len); 3064 ASSERT(end < toxic_bit_map_len); 3065 3066 while (start <= end) { 3067 BT_ATOMIC_SET(toxic_bit_map, start); 3068 ++start; 3069 } 3070 return (vaddr); 3071 } 3072 3073 void 3074 device_arena_free(void *vaddr, size_t size) 3075 { 3076 uintptr_t v = (uintptr_t)vaddr; 3077 size_t start; 3078 size_t end; 3079 3080 ASSERT(v >= kernelbase); 3081 ASSERT(v + size <= valloc_base); 3082 3083 start = btop(v - kernelbase); 3084 end = btop(v + size - 1 - kernelbase); 3085 ASSERT(start < toxic_bit_map_len); 3086 ASSERT(end < toxic_bit_map_len); 3087 3088 while (start <= end) { 3089 ASSERT(BT_TEST(toxic_bit_map, start) != 0); 3090 BT_ATOMIC_CLEAR(toxic_bit_map, start); 3091 ++start; 3092 } 3093 vmem_free(heap_arena, vaddr, size); 3094 } 3095 3096 /* 3097 * returns 1st address in range that is in device arena, or NULL 3098 * if len is not NULL it returns the length of the toxic range 3099 */ 3100 void * 3101 device_arena_contains(void *vaddr, size_t size, size_t *len) 3102 { 3103 uintptr_t v = (uintptr_t)vaddr; 3104 uintptr_t eaddr = v + size; 3105 size_t start; 3106 size_t end; 3107 3108 /* 3109 * if called very early by kmdb, just return NULL 3110 */ 3111 if (toxic_bit_map == NULL) 3112 return (NULL); 3113 3114 /* 3115 * First check if we're completely outside the bitmap range. 3116 */ 3117 if (v >= valloc_base || eaddr < kernelbase) 3118 return (NULL); 3119 3120 /* 3121 * Trim ends of search to look at only what the bitmap covers. 3122 */ 3123 if (v < kernelbase) 3124 v = kernelbase; 3125 start = btop(v - kernelbase); 3126 end = btop(eaddr - kernelbase); 3127 if (end >= toxic_bit_map_len) 3128 end = toxic_bit_map_len; 3129 3130 if (bt_range(toxic_bit_map, &start, &end, end) == 0) 3131 return (NULL); 3132 3133 v = kernelbase + ptob(start); 3134 if (len != NULL) 3135 *len = ptob(end - start); 3136 return ((void *)v); 3137 } 3138 3139 #endif /* __i386 */ 3140