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 2010 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 24 */ 25 /* 26 * Copyright (c) 2010, Intel Corporation. 27 * All rights reserved. 28 */ 29 30 #include <sys/types.h> 31 #include <sys/t_lock.h> 32 #include <sys/param.h> 33 #include <sys/sysmacros.h> 34 #include <sys/signal.h> 35 #include <sys/systm.h> 36 #include <sys/user.h> 37 #include <sys/mman.h> 38 #include <sys/vm.h> 39 #include <sys/conf.h> 40 #include <sys/avintr.h> 41 #include <sys/autoconf.h> 42 #include <sys/disp.h> 43 #include <sys/class.h> 44 #include <sys/bitmap.h> 45 46 #include <sys/privregs.h> 47 48 #include <sys/proc.h> 49 #include <sys/buf.h> 50 #include <sys/kmem.h> 51 #include <sys/mem.h> 52 #include <sys/kstat.h> 53 54 #include <sys/reboot.h> 55 56 #include <sys/cred.h> 57 #include <sys/vnode.h> 58 #include <sys/file.h> 59 60 #include <sys/procfs.h> 61 62 #include <sys/vfs.h> 63 #include <sys/cmn_err.h> 64 #include <sys/utsname.h> 65 #include <sys/debug.h> 66 #include <sys/kdi.h> 67 68 #include <sys/dumphdr.h> 69 #include <sys/bootconf.h> 70 #include <sys/memlist_plat.h> 71 #include <sys/varargs.h> 72 #include <sys/promif.h> 73 #include <sys/modctl.h> 74 75 #include <sys/sunddi.h> 76 #include <sys/sunndi.h> 77 #include <sys/ndi_impldefs.h> 78 #include <sys/ddidmareq.h> 79 #include <sys/psw.h> 80 #include <sys/regset.h> 81 #include <sys/clock.h> 82 #include <sys/pte.h> 83 #include <sys/tss.h> 84 #include <sys/stack.h> 85 #include <sys/trap.h> 86 #include <sys/fp.h> 87 #include <vm/kboot_mmu.h> 88 #include <vm/anon.h> 89 #include <vm/as.h> 90 #include <vm/page.h> 91 #include <vm/seg.h> 92 #include <vm/seg_dev.h> 93 #include <vm/seg_kmem.h> 94 #include <vm/seg_kpm.h> 95 #include <vm/seg_map.h> 96 #include <vm/seg_vn.h> 97 #include <vm/seg_kp.h> 98 #include <sys/memnode.h> 99 #include <vm/vm_dep.h> 100 #include <sys/thread.h> 101 #include <sys/sysconf.h> 102 #include <sys/vm_machparam.h> 103 #include <sys/archsystm.h> 104 #include <sys/machsystm.h> 105 #include <vm/hat.h> 106 #include <vm/hat_i86.h> 107 #include <sys/pmem.h> 108 #include <sys/smp_impldefs.h> 109 #include <sys/x86_archext.h> 110 #include <sys/cpuvar.h> 111 #include <sys/segments.h> 112 #include <sys/clconf.h> 113 #include <sys/kobj.h> 114 #include <sys/kobj_lex.h> 115 #include <sys/cpc_impl.h> 116 #include <sys/cpu_module.h> 117 #include <sys/smbios.h> 118 #include <sys/debug_info.h> 119 #include <sys/bootinfo.h> 120 #include <sys/ddi_timer.h> 121 #include <sys/systeminfo.h> 122 #include <sys/multiboot.h> 123 124 #ifdef __xpv 125 126 #include <sys/hypervisor.h> 127 #include <sys/xen_mmu.h> 128 #include <sys/evtchn_impl.h> 129 #include <sys/gnttab.h> 130 #include <sys/xpv_panic.h> 131 #include <xen/sys/xenbus_comms.h> 132 #include <xen/public/physdev.h> 133 134 extern void xen_late_startup(void); 135 136 struct xen_evt_data cpu0_evt_data; 137 138 #else /* __xpv */ 139 #include <sys/memlist_impl.h> 140 141 extern void mem_config_init(void); 142 #endif /* __xpv */ 143 144 extern void progressbar_init(void); 145 extern void brand_init(void); 146 extern void pcf_init(void); 147 extern void pg_init(void); 148 149 extern int size_pse_array(pgcnt_t, int); 150 151 #if defined(_SOFT_HOSTID) 152 153 #include <sys/rtc.h> 154 155 static int32_t set_soft_hostid(void); 156 static char hostid_file[] = "/etc/hostid"; 157 158 #endif 159 160 void *gfx_devinfo_list; 161 162 #if defined(__amd64) && !defined(__xpv) 163 extern void immu_startup(void); 164 #endif 165 166 /* 167 * XXX make declaration below "static" when drivers no longer use this 168 * interface. 169 */ 170 extern caddr_t p0_va; /* Virtual address for accessing physical page 0 */ 171 172 /* 173 * segkp 174 */ 175 extern int segkp_fromheap; 176 177 static void kvm_init(void); 178 static void startup_init(void); 179 static void startup_memlist(void); 180 static void startup_kmem(void); 181 static void startup_modules(void); 182 static void startup_vm(void); 183 static void startup_end(void); 184 static void layout_kernel_va(void); 185 186 /* 187 * Declare these as initialized data so we can patch them. 188 */ 189 #ifdef __i386 190 191 /* 192 * Due to virtual address space limitations running in 32 bit mode, restrict 193 * the amount of physical memory configured to a max of PHYSMEM pages (16g). 194 * 195 * If the physical max memory size of 64g were allowed to be configured, the 196 * size of user virtual address space will be less than 1g. A limited user 197 * address space greatly reduces the range of applications that can run. 198 * 199 * If more physical memory than PHYSMEM is required, users should preferably 200 * run in 64 bit mode which has far looser virtual address space limitations. 201 * 202 * If 64 bit mode is not available (as in IA32) and/or more physical memory 203 * than PHYSMEM is required in 32 bit mode, physmem can be set to the desired 204 * value or to 0 (to configure all available memory) via eeprom(1M). kernelbase 205 * should also be carefully tuned to balance out the need of the user 206 * application while minimizing the risk of kernel heap exhaustion due to 207 * kernelbase being set too high. 208 */ 209 #define PHYSMEM 0x400000 210 211 #else /* __amd64 */ 212 213 /* 214 * For now we can handle memory with physical addresses up to about 215 * 64 Terabytes. This keeps the kernel above the VA hole, leaving roughly 216 * half the VA space for seg_kpm. When systems get bigger than 64TB this 217 * code will need revisiting. There is an implicit assumption that there 218 * are no *huge* holes in the physical address space too. 219 */ 220 #define TERABYTE (1ul << 40) 221 #define PHYSMEM_MAX64 mmu_btop(64 * TERABYTE) 222 #define PHYSMEM PHYSMEM_MAX64 223 #define AMD64_VA_HOLE_END 0xFFFF800000000000ul 224 225 #endif /* __amd64 */ 226 227 pgcnt_t physmem = PHYSMEM; 228 pgcnt_t obp_pages; /* Memory used by PROM for its text and data */ 229 230 char *kobj_file_buf; 231 int kobj_file_bufsize; /* set in /etc/system */ 232 233 /* Global variables for MP support. Used in mp_startup */ 234 caddr_t rm_platter_va = 0; 235 uint32_t rm_platter_pa; 236 237 int auto_lpg_disable = 1; 238 239 /* 240 * Some CPUs have holes in the middle of the 64-bit virtual address range. 241 */ 242 uintptr_t hole_start, hole_end; 243 244 /* 245 * kpm mapping window 246 */ 247 caddr_t kpm_vbase; 248 size_t kpm_size; 249 static int kpm_desired; 250 #ifdef __amd64 251 static uintptr_t segkpm_base = (uintptr_t)SEGKPM_BASE; 252 #endif 253 254 /* 255 * Configuration parameters set at boot time. 256 */ 257 258 caddr_t econtig; /* end of first block of contiguous kernel */ 259 260 struct bootops *bootops = 0; /* passed in from boot */ 261 struct bootops **bootopsp; 262 struct boot_syscalls *sysp; /* passed in from boot */ 263 264 char bootblock_fstype[16]; 265 266 char kern_bootargs[OBP_MAXPATHLEN]; 267 char kern_bootfile[OBP_MAXPATHLEN]; 268 269 /* 270 * ZFS zio segment. This allows us to exclude large portions of ZFS data that 271 * gets cached in kmem caches on the heap. If this is set to zero, we allocate 272 * zio buffers from their own segment, otherwise they are allocated from the 273 * heap. The optimization of allocating zio buffers from their own segment is 274 * only valid on 64-bit kernels. 275 */ 276 #if defined(__amd64) 277 int segzio_fromheap = 0; 278 #else 279 int segzio_fromheap = 1; 280 #endif 281 282 /* 283 * new memory fragmentations are possible in startup() due to BOP_ALLOCs. this 284 * depends on number of BOP_ALLOC calls made and requested size, memory size 285 * combination and whether boot.bin memory needs to be freed. 286 */ 287 #define POSS_NEW_FRAGMENTS 12 288 289 /* 290 * VM data structures 291 */ 292 long page_hashsz; /* Size of page hash table (power of two) */ 293 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 extern int post_fastreboot; 692 #endif 693 extern cpuset_t cpu_ready_set; 694 695 /* 696 * Make sure that nobody tries to use sekpm until we have 697 * initialized it properly. 698 */ 699 #if defined(__amd64) 700 kpm_desired = 1; 701 #endif 702 kpm_enable = 0; 703 CPUSET_ONLY(cpu_ready_set, 0); /* cpu 0 is boot cpu */ 704 705 #if defined(__xpv) /* XXPV fix me! */ 706 { 707 extern int segvn_use_regions; 708 segvn_use_regions = 0; 709 } 710 #endif 711 progressbar_init(); 712 startup_init(); 713 #if defined(__xpv) 714 startup_xen_version(); 715 #endif 716 startup_memlist(); 717 startup_kmem(); 718 startup_vm(); 719 #if !defined(__xpv) 720 if (!post_fastreboot) 721 startup_pci_bios(); 722 #endif 723 #if defined(__xpv) 724 startup_xen_mca(); 725 #endif 726 startup_modules(); 727 728 startup_end(); 729 } 730 731 static void 732 startup_init() 733 { 734 PRM_POINT("startup_init() starting..."); 735 736 /* 737 * Complete the extraction of cpuid data 738 */ 739 cpuid_pass2(CPU); 740 741 (void) check_boot_version(BOP_GETVERSION(bootops)); 742 743 /* 744 * Check for prom_debug in boot environment 745 */ 746 if (BOP_GETPROPLEN(bootops, "prom_debug") >= 0) { 747 ++prom_debug; 748 PRM_POINT("prom_debug found in boot enviroment"); 749 } 750 751 /* 752 * Collect node, cpu and memory configuration information. 753 */ 754 get_system_configuration(); 755 756 /* 757 * Halt if this is an unsupported processor. 758 */ 759 if (x86_type == X86_TYPE_486 || x86_type == X86_TYPE_CYRIX_486) { 760 printf("\n486 processor (\"%s\") detected.\n", 761 CPU->cpu_brandstr); 762 halt("This processor is not supported by this release " 763 "of Solaris."); 764 } 765 766 PRM_POINT("startup_init() done"); 767 } 768 769 /* 770 * Callback for copy_memlist_filter() to filter nucleus, kadb/kmdb, (ie. 771 * everything mapped above KERNEL_TEXT) pages from phys_avail. Note it 772 * also filters out physical page zero. There is some reliance on the 773 * boot loader allocating only a few contiguous physical memory chunks. 774 */ 775 static void 776 avail_filter(uint64_t *addr, uint64_t *size) 777 { 778 uintptr_t va; 779 uintptr_t next_va; 780 pfn_t pfn; 781 uint64_t pfn_addr; 782 uint64_t pfn_eaddr; 783 uint_t prot; 784 size_t len; 785 uint_t change; 786 787 if (prom_debug) 788 prom_printf("\tFilter: in: a=%" PRIx64 ", s=%" PRIx64 "\n", 789 *addr, *size); 790 791 /* 792 * page zero is required for BIOS.. never make it available 793 */ 794 if (*addr == 0) { 795 *addr += MMU_PAGESIZE; 796 *size -= MMU_PAGESIZE; 797 } 798 799 /* 800 * First we trim from the front of the range. Since kbm_probe() 801 * walks ranges in virtual order, but addr/size are physical, we need 802 * to the list until no changes are seen. This deals with the case 803 * where page "p" is mapped at v, page "p + PAGESIZE" is mapped at w 804 * but w < v. 805 */ 806 do { 807 change = 0; 808 for (va = KERNEL_TEXT; 809 *size > 0 && kbm_probe(&va, &len, &pfn, &prot) != 0; 810 va = next_va) { 811 812 next_va = va + len; 813 pfn_addr = pfn_to_pa(pfn); 814 pfn_eaddr = pfn_addr + len; 815 816 if (pfn_addr <= *addr && pfn_eaddr > *addr) { 817 change = 1; 818 while (*size > 0 && len > 0) { 819 *addr += MMU_PAGESIZE; 820 *size -= MMU_PAGESIZE; 821 len -= MMU_PAGESIZE; 822 } 823 } 824 } 825 if (change && prom_debug) 826 prom_printf("\t\ttrim: a=%" PRIx64 ", s=%" PRIx64 "\n", 827 *addr, *size); 828 } while (change); 829 830 /* 831 * Trim pages from the end of the range. 832 */ 833 for (va = KERNEL_TEXT; 834 *size > 0 && kbm_probe(&va, &len, &pfn, &prot) != 0; 835 va = next_va) { 836 837 next_va = va + len; 838 pfn_addr = pfn_to_pa(pfn); 839 840 if (pfn_addr >= *addr && pfn_addr < *addr + *size) 841 *size = pfn_addr - *addr; 842 } 843 844 if (prom_debug) 845 prom_printf("\tFilter out: a=%" PRIx64 ", s=%" PRIx64 "\n", 846 *addr, *size); 847 } 848 849 static void 850 kpm_init() 851 { 852 struct segkpm_crargs b; 853 854 /* 855 * These variables were all designed for sfmmu in which segkpm is 856 * mapped using a single pagesize - either 8KB or 4MB. On x86, we 857 * might use 2+ page sizes on a single machine, so none of these 858 * variables have a single correct value. They are set up as if we 859 * always use a 4KB pagesize, which should do no harm. In the long 860 * run, we should get rid of KPM's assumption that only a single 861 * pagesize is used. 862 */ 863 kpm_pgshft = MMU_PAGESHIFT; 864 kpm_pgsz = MMU_PAGESIZE; 865 kpm_pgoff = MMU_PAGEOFFSET; 866 kpmp2pshft = 0; 867 kpmpnpgs = 1; 868 ASSERT(((uintptr_t)kpm_vbase & (kpm_pgsz - 1)) == 0); 869 870 PRM_POINT("about to create segkpm"); 871 rw_enter(&kas.a_lock, RW_WRITER); 872 873 if (seg_attach(&kas, kpm_vbase, kpm_size, segkpm) < 0) 874 panic("cannot attach segkpm"); 875 876 b.prot = PROT_READ | PROT_WRITE; 877 b.nvcolors = 1; 878 879 if (segkpm_create(segkpm, (caddr_t)&b) != 0) 880 panic("segkpm_create segkpm"); 881 882 rw_exit(&kas.a_lock); 883 } 884 885 /* 886 * The debug info page provides enough information to allow external 887 * inspectors (e.g. when running under a hypervisor) to bootstrap 888 * themselves into allowing full-blown kernel debugging. 889 */ 890 static void 891 init_debug_info(void) 892 { 893 caddr_t mem; 894 debug_info_t *di; 895 896 #ifndef __lint 897 ASSERT(sizeof (debug_info_t) < MMU_PAGESIZE); 898 #endif 899 900 mem = BOP_ALLOC(bootops, (caddr_t)DEBUG_INFO_VA, MMU_PAGESIZE, 901 MMU_PAGESIZE); 902 903 if (mem != (caddr_t)DEBUG_INFO_VA) 904 panic("BOP_ALLOC() failed"); 905 bzero(mem, MMU_PAGESIZE); 906 907 di = (debug_info_t *)mem; 908 909 di->di_magic = DEBUG_INFO_MAGIC; 910 di->di_version = DEBUG_INFO_VERSION; 911 di->di_modules = (uintptr_t)&modules; 912 di->di_s_text = (uintptr_t)s_text; 913 di->di_e_text = (uintptr_t)e_text; 914 di->di_s_data = (uintptr_t)s_data; 915 di->di_e_data = (uintptr_t)e_data; 916 di->di_hat_htable_off = offsetof(hat_t, hat_htable); 917 di->di_ht_pfn_off = offsetof(htable_t, ht_pfn); 918 } 919 920 /* 921 * Build the memlists and other kernel essential memory system data structures. 922 * This is everything at valloc_base. 923 */ 924 static void 925 startup_memlist(void) 926 { 927 size_t memlist_sz; 928 size_t memseg_sz; 929 size_t pagehash_sz; 930 size_t pp_sz; 931 uintptr_t va; 932 size_t len; 933 uint_t prot; 934 pfn_t pfn; 935 int memblocks; 936 pfn_t rsvd_high_pfn; 937 pgcnt_t rsvd_pgcnt; 938 size_t rsvdmemlist_sz; 939 int rsvdmemblocks; 940 caddr_t pagecolor_mem; 941 size_t pagecolor_memsz; 942 caddr_t page_ctrs_mem; 943 size_t page_ctrs_size; 944 size_t pse_table_alloc_size; 945 struct memlist *current; 946 extern void startup_build_mem_nodes(struct memlist *); 947 948 /* XX64 fix these - they should be in include files */ 949 extern size_t page_coloring_init(uint_t, int, int); 950 extern void page_coloring_setup(caddr_t); 951 952 PRM_POINT("startup_memlist() starting..."); 953 954 /* 955 * Use leftover large page nucleus text/data space for loadable modules. 956 * Use at most MODTEXT/MODDATA. 957 */ 958 len = kbm_nucleus_size; 959 ASSERT(len > MMU_PAGESIZE); 960 961 moddata = (caddr_t)ROUND_UP_PAGE(e_data); 962 e_moddata = (caddr_t)P2ROUNDUP((uintptr_t)e_data, (uintptr_t)len); 963 if (e_moddata - moddata > MODDATA) 964 e_moddata = moddata + MODDATA; 965 966 modtext = (caddr_t)ROUND_UP_PAGE(e_text); 967 e_modtext = (caddr_t)P2ROUNDUP((uintptr_t)e_text, (uintptr_t)len); 968 if (e_modtext - modtext > MODTEXT) 969 e_modtext = modtext + MODTEXT; 970 971 econtig = e_moddata; 972 973 PRM_DEBUG(modtext); 974 PRM_DEBUG(e_modtext); 975 PRM_DEBUG(moddata); 976 PRM_DEBUG(e_moddata); 977 PRM_DEBUG(econtig); 978 979 /* 980 * Examine the boot loader physical memory map to find out: 981 * - total memory in system - physinstalled 982 * - the max physical address - physmax 983 * - the number of discontiguous segments of memory. 984 */ 985 if (prom_debug) 986 print_memlist("boot physinstalled", 987 bootops->boot_mem->physinstalled); 988 installed_top_size_ex(bootops->boot_mem->physinstalled, &physmax, 989 &physinstalled, &memblocks); 990 PRM_DEBUG(physmax); 991 PRM_DEBUG(physinstalled); 992 PRM_DEBUG(memblocks); 993 994 /* 995 * Compute maximum physical address for memory DR operations. 996 * Memory DR operations are unsupported on xpv or 32bit OSes. 997 */ 998 #ifdef __amd64 999 if (plat_dr_support_memory()) { 1000 if (plat_dr_physmax == 0) { 1001 uint_t pabits = UINT_MAX; 1002 1003 cpuid_get_addrsize(CPU, &pabits, NULL); 1004 plat_dr_physmax = btop(1ULL << pabits); 1005 } 1006 if (plat_dr_physmax > PHYSMEM_MAX64) 1007 plat_dr_physmax = PHYSMEM_MAX64; 1008 } else 1009 #endif 1010 plat_dr_physmax = 0; 1011 1012 /* 1013 * Examine the bios reserved memory to find out: 1014 * - the number of discontiguous segments of memory. 1015 */ 1016 if (prom_debug) 1017 print_memlist("boot reserved mem", 1018 bootops->boot_mem->rsvdmem); 1019 installed_top_size_ex(bootops->boot_mem->rsvdmem, &rsvd_high_pfn, 1020 &rsvd_pgcnt, &rsvdmemblocks); 1021 PRM_DEBUG(rsvd_high_pfn); 1022 PRM_DEBUG(rsvd_pgcnt); 1023 PRM_DEBUG(rsvdmemblocks); 1024 1025 /* 1026 * Initialize hat's mmu parameters. 1027 * Check for enforce-prot-exec in boot environment. It's used to 1028 * enable/disable support for the page table entry NX bit. 1029 * The default is to enforce PROT_EXEC on processors that support NX. 1030 * Boot seems to round up the "len", but 8 seems to be big enough. 1031 */ 1032 mmu_init(); 1033 1034 #ifdef __i386 1035 /* 1036 * physmax is lowered if there is more memory than can be 1037 * physically addressed in 32 bit (PAE/non-PAE) modes. 1038 */ 1039 if (mmu.pae_hat) { 1040 if (PFN_ABOVE64G(physmax)) { 1041 physinstalled -= (physmax - (PFN_64G - 1)); 1042 physmax = PFN_64G - 1; 1043 } 1044 } else { 1045 if (PFN_ABOVE4G(physmax)) { 1046 physinstalled -= (physmax - (PFN_4G - 1)); 1047 physmax = PFN_4G - 1; 1048 } 1049 } 1050 #endif 1051 1052 startup_build_mem_nodes(bootops->boot_mem->physinstalled); 1053 1054 if (BOP_GETPROPLEN(bootops, "enforce-prot-exec") >= 0) { 1055 int len = BOP_GETPROPLEN(bootops, "enforce-prot-exec"); 1056 char value[8]; 1057 1058 if (len < 8) 1059 (void) BOP_GETPROP(bootops, "enforce-prot-exec", value); 1060 else 1061 (void) strcpy(value, ""); 1062 if (strcmp(value, "off") == 0) 1063 mmu.pt_nx = 0; 1064 } 1065 PRM_DEBUG(mmu.pt_nx); 1066 1067 /* 1068 * We will need page_t's for every page in the system, except for 1069 * memory mapped at or above above the start of the kernel text segment. 1070 * 1071 * pages above e_modtext are attributed to kernel debugger (obp_pages) 1072 */ 1073 npages = physinstalled - 1; /* avail_filter() skips page 0, so "- 1" */ 1074 obp_pages = 0; 1075 va = KERNEL_TEXT; 1076 while (kbm_probe(&va, &len, &pfn, &prot) != 0) { 1077 npages -= len >> MMU_PAGESHIFT; 1078 if (va >= (uintptr_t)e_moddata) 1079 obp_pages += len >> MMU_PAGESHIFT; 1080 va += len; 1081 } 1082 PRM_DEBUG(npages); 1083 PRM_DEBUG(obp_pages); 1084 1085 /* 1086 * If physmem is patched to be non-zero, use it instead of the computed 1087 * value unless it is larger than the actual amount of memory on hand. 1088 */ 1089 if (physmem == 0 || physmem > npages) { 1090 physmem = npages; 1091 } else if (physmem < npages) { 1092 orig_npages = npages; 1093 npages = physmem; 1094 } 1095 PRM_DEBUG(physmem); 1096 1097 /* 1098 * We now compute the sizes of all the initial allocations for 1099 * structures the kernel needs in order do kmem_alloc(). These 1100 * include: 1101 * memsegs 1102 * memlists 1103 * page hash table 1104 * page_t's 1105 * page coloring data structs 1106 */ 1107 memseg_sz = sizeof (struct memseg) * (memblocks + POSS_NEW_FRAGMENTS); 1108 ADD_TO_ALLOCATIONS(memseg_base, memseg_sz); 1109 PRM_DEBUG(memseg_sz); 1110 1111 /* 1112 * Reserve space for memlists. There's no real good way to know exactly 1113 * how much room we'll need, but this should be a good upper bound. 1114 */ 1115 memlist_sz = ROUND_UP_PAGE(2 * sizeof (struct memlist) * 1116 (memblocks + POSS_NEW_FRAGMENTS)); 1117 ADD_TO_ALLOCATIONS(memlist, memlist_sz); 1118 PRM_DEBUG(memlist_sz); 1119 1120 /* 1121 * Reserve space for bios reserved memlists. 1122 */ 1123 rsvdmemlist_sz = ROUND_UP_PAGE(2 * sizeof (struct memlist) * 1124 (rsvdmemblocks + POSS_NEW_FRAGMENTS)); 1125 ADD_TO_ALLOCATIONS(bios_rsvd, rsvdmemlist_sz); 1126 PRM_DEBUG(rsvdmemlist_sz); 1127 1128 /* 1129 * The page structure hash table size is a power of 2 1130 * such that the average hash chain length is PAGE_HASHAVELEN. 1131 */ 1132 page_hashsz = npages / PAGE_HASHAVELEN; 1133 page_hashsz = 1 << highbit(page_hashsz); 1134 pagehash_sz = sizeof (struct page *) * page_hashsz; 1135 ADD_TO_ALLOCATIONS(page_hash, pagehash_sz); 1136 PRM_DEBUG(pagehash_sz); 1137 1138 /* 1139 * Set aside room for the page structures themselves. 1140 */ 1141 PRM_DEBUG(npages); 1142 pp_sz = sizeof (struct page) * npages; 1143 ADD_TO_ALLOCATIONS(pp_base, pp_sz); 1144 PRM_DEBUG(pp_sz); 1145 1146 /* 1147 * determine l2 cache info and memory size for page coloring 1148 */ 1149 (void) getl2cacheinfo(CPU, 1150 &l2cache_sz, &l2cache_linesz, &l2cache_assoc); 1151 pagecolor_memsz = 1152 page_coloring_init(l2cache_sz, l2cache_linesz, l2cache_assoc); 1153 ADD_TO_ALLOCATIONS(pagecolor_mem, pagecolor_memsz); 1154 PRM_DEBUG(pagecolor_memsz); 1155 1156 page_ctrs_size = page_ctrs_sz(); 1157 ADD_TO_ALLOCATIONS(page_ctrs_mem, page_ctrs_size); 1158 PRM_DEBUG(page_ctrs_size); 1159 1160 /* 1161 * Allocate the array that protects pp->p_selock. 1162 */ 1163 pse_shift = size_pse_array(physmem, max_ncpus); 1164 pse_table_size = 1 << pse_shift; 1165 pse_table_alloc_size = pse_table_size * sizeof (pad_mutex_t); 1166 ADD_TO_ALLOCATIONS(pse_mutex, pse_table_alloc_size); 1167 1168 #if defined(__amd64) 1169 valloc_sz = ROUND_UP_LPAGE(valloc_sz); 1170 valloc_base = VALLOC_BASE; 1171 1172 /* 1173 * The default values of VALLOC_BASE and SEGKPM_BASE should work 1174 * for values of physmax up to 1 Terabyte. They need adjusting when 1175 * memory is at addresses above 1 TB. When adjusted, segkpm_base must 1176 * be aligned on KERNEL_REDZONE_SIZE boundary (span of top level pte). 1177 */ 1178 if (physmax + 1 > mmu_btop(TERABYTE) || 1179 plat_dr_physmax > mmu_btop(TERABYTE)) { 1180 uint64_t kpm_resv_amount = mmu_ptob(physmax + 1); 1181 1182 if (kpm_resv_amount < mmu_ptob(plat_dr_physmax)) { 1183 kpm_resv_amount = mmu_ptob(plat_dr_physmax); 1184 } 1185 1186 segkpm_base = -(P2ROUNDUP((2 * kpm_resv_amount), 1187 KERNEL_REDZONE_SIZE)); /* down from top VA */ 1188 1189 /* make sure we leave some space for user apps above hole */ 1190 segkpm_base = MAX(segkpm_base, AMD64_VA_HOLE_END + TERABYTE); 1191 if (segkpm_base > SEGKPM_BASE) 1192 segkpm_base = SEGKPM_BASE; 1193 PRM_DEBUG(segkpm_base); 1194 1195 valloc_base = segkpm_base + P2ROUNDUP(kpm_resv_amount, ONE_GIG); 1196 if (valloc_base < segkpm_base) 1197 panic("not enough kernel VA to support memory size"); 1198 PRM_DEBUG(valloc_base); 1199 } 1200 #else /* __i386 */ 1201 valloc_base = (uintptr_t)(MISC_VA_BASE - valloc_sz); 1202 valloc_base = P2ALIGN(valloc_base, mmu.level_size[1]); 1203 PRM_DEBUG(valloc_base); 1204 #endif /* __i386 */ 1205 1206 /* 1207 * do all the initial allocations 1208 */ 1209 perform_allocations(); 1210 1211 /* 1212 * Build phys_install and phys_avail in kernel memspace. 1213 * - phys_install should be all memory in the system. 1214 * - phys_avail is phys_install minus any memory mapped before this 1215 * point above KERNEL_TEXT. 1216 */ 1217 current = phys_install = memlist; 1218 copy_memlist_filter(bootops->boot_mem->physinstalled, ¤t, NULL); 1219 if ((caddr_t)current > (caddr_t)memlist + memlist_sz) 1220 panic("physinstalled was too big!"); 1221 if (prom_debug) 1222 print_memlist("phys_install", phys_install); 1223 1224 phys_avail = current; 1225 PRM_POINT("Building phys_avail:\n"); 1226 copy_memlist_filter(bootops->boot_mem->physinstalled, ¤t, 1227 avail_filter); 1228 if ((caddr_t)current > (caddr_t)memlist + memlist_sz) 1229 panic("physavail was too big!"); 1230 if (prom_debug) 1231 print_memlist("phys_avail", phys_avail); 1232 #ifndef __xpv 1233 /* 1234 * Free unused memlist items, which may be used by memory DR driver 1235 * at runtime. 1236 */ 1237 if ((caddr_t)current < (caddr_t)memlist + memlist_sz) { 1238 memlist_free_block((caddr_t)current, 1239 (caddr_t)memlist + memlist_sz - (caddr_t)current); 1240 } 1241 #endif 1242 1243 /* 1244 * Build bios reserved memspace 1245 */ 1246 current = bios_rsvd; 1247 copy_memlist_filter(bootops->boot_mem->rsvdmem, ¤t, NULL); 1248 if ((caddr_t)current > (caddr_t)bios_rsvd + rsvdmemlist_sz) 1249 panic("bios_rsvd was too big!"); 1250 if (prom_debug) 1251 print_memlist("bios_rsvd", bios_rsvd); 1252 #ifndef __xpv 1253 /* 1254 * Free unused memlist items, which may be used by memory DR driver 1255 * at runtime. 1256 */ 1257 if ((caddr_t)current < (caddr_t)bios_rsvd + rsvdmemlist_sz) { 1258 memlist_free_block((caddr_t)current, 1259 (caddr_t)bios_rsvd + rsvdmemlist_sz - (caddr_t)current); 1260 } 1261 #endif 1262 1263 /* 1264 * setup page coloring 1265 */ 1266 page_coloring_setup(pagecolor_mem); 1267 page_lock_init(); /* currently a no-op */ 1268 1269 /* 1270 * free page list counters 1271 */ 1272 (void) page_ctrs_alloc(page_ctrs_mem); 1273 1274 /* 1275 * Size the pcf array based on the number of cpus in the box at 1276 * boot time. 1277 */ 1278 1279 pcf_init(); 1280 1281 /* 1282 * Initialize the page structures from the memory lists. 1283 */ 1284 availrmem_initial = availrmem = freemem = 0; 1285 PRM_POINT("Calling kphysm_init()..."); 1286 npages = kphysm_init(pp_base, npages); 1287 PRM_POINT("kphysm_init() done"); 1288 PRM_DEBUG(npages); 1289 1290 init_debug_info(); 1291 1292 /* 1293 * Now that page_t's have been initialized, remove all the 1294 * initial allocation pages from the kernel free page lists. 1295 */ 1296 boot_mapin((caddr_t)valloc_base, valloc_sz); 1297 boot_mapin((caddr_t)MISC_VA_BASE, MISC_VA_SIZE); 1298 PRM_POINT("startup_memlist() done"); 1299 1300 PRM_DEBUG(valloc_sz); 1301 1302 #if defined(__amd64) 1303 if ((availrmem >> (30 - MMU_PAGESHIFT)) >= 1304 textrepl_min_gb && l2cache_sz <= 2 << 20) { 1305 extern size_t textrepl_size_thresh; 1306 textrepl_size_thresh = (16 << 20) - 1; 1307 } 1308 #endif 1309 } 1310 1311 /* 1312 * Layout the kernel's part of address space and initialize kmem allocator. 1313 */ 1314 static void 1315 startup_kmem(void) 1316 { 1317 extern void page_set_colorequiv_arr(void); 1318 const char *fmt = "?features: %b\n"; 1319 1320 PRM_POINT("startup_kmem() starting..."); 1321 1322 #if defined(__amd64) 1323 if (eprom_kernelbase && eprom_kernelbase != KERNELBASE) 1324 cmn_err(CE_NOTE, "!kernelbase cannot be changed on 64-bit " 1325 "systems."); 1326 kernelbase = segkpm_base - KERNEL_REDZONE_SIZE; 1327 core_base = (uintptr_t)COREHEAP_BASE; 1328 core_size = (size_t)MISC_VA_BASE - COREHEAP_BASE; 1329 #else /* __i386 */ 1330 /* 1331 * We configure kernelbase based on: 1332 * 1333 * 1. user specified kernelbase via eeprom command. Value cannot exceed 1334 * KERNELBASE_MAX. we large page align eprom_kernelbase 1335 * 1336 * 2. Default to KERNELBASE and adjust to 2X less the size for page_t. 1337 * On large memory systems we must lower kernelbase to allow 1338 * enough room for page_t's for all of memory. 1339 * 1340 * The value set here, might be changed a little later. 1341 */ 1342 if (eprom_kernelbase) { 1343 kernelbase = eprom_kernelbase & mmu.level_mask[1]; 1344 if (kernelbase > KERNELBASE_MAX) 1345 kernelbase = KERNELBASE_MAX; 1346 } else { 1347 kernelbase = (uintptr_t)KERNELBASE; 1348 kernelbase -= ROUND_UP_4MEG(2 * valloc_sz); 1349 } 1350 ASSERT((kernelbase & mmu.level_offset[1]) == 0); 1351 core_base = valloc_base; 1352 core_size = 0; 1353 #endif /* __i386 */ 1354 1355 PRM_DEBUG(core_base); 1356 PRM_DEBUG(core_size); 1357 PRM_DEBUG(kernelbase); 1358 1359 #if defined(__i386) 1360 segkp_fromheap = 1; 1361 #endif /* __i386 */ 1362 1363 ekernelheap = (char *)core_base; 1364 PRM_DEBUG(ekernelheap); 1365 1366 /* 1367 * Now that we know the real value of kernelbase, 1368 * update variables that were initialized with a value of 1369 * KERNELBASE (in common/conf/param.c). 1370 * 1371 * XXX The problem with this sort of hackery is that the 1372 * compiler just may feel like putting the const declarations 1373 * (in param.c) into the .text section. Perhaps they should 1374 * just be declared as variables there? 1375 */ 1376 1377 *(uintptr_t *)&_kernelbase = kernelbase; 1378 *(uintptr_t *)&_userlimit = kernelbase; 1379 #if defined(__amd64) 1380 *(uintptr_t *)&_userlimit -= KERNELBASE - USERLIMIT; 1381 #else 1382 *(uintptr_t *)&_userlimit32 = _userlimit; 1383 #endif 1384 PRM_DEBUG(_kernelbase); 1385 PRM_DEBUG(_userlimit); 1386 PRM_DEBUG(_userlimit32); 1387 1388 layout_kernel_va(); 1389 1390 #if defined(__i386) 1391 /* 1392 * If segmap is too large we can push the bottom of the kernel heap 1393 * higher than the base. Or worse, it could exceed the top of the 1394 * VA space entirely, causing it to wrap around. 1395 */ 1396 if (kernelheap >= ekernelheap || (uintptr_t)kernelheap < kernelbase) 1397 panic("too little address space available for kernelheap," 1398 " use eeprom for lower kernelbase or smaller segmapsize"); 1399 #endif /* __i386 */ 1400 1401 /* 1402 * Initialize the kernel heap. Note 3rd argument must be > 1st. 1403 */ 1404 kernelheap_init(kernelheap, ekernelheap, 1405 kernelheap + MMU_PAGESIZE, 1406 (void *)core_base, (void *)(core_base + core_size)); 1407 1408 #if defined(__xpv) 1409 /* 1410 * Link pending events struct into cpu struct 1411 */ 1412 CPU->cpu_m.mcpu_evt_pend = &cpu0_evt_data; 1413 #endif 1414 /* 1415 * Initialize kernel memory allocator. 1416 */ 1417 kmem_init(); 1418 1419 /* 1420 * Factor in colorequiv to check additional 'equivalent' bins 1421 */ 1422 page_set_colorequiv_arr(); 1423 1424 /* 1425 * print this out early so that we know what's going on 1426 */ 1427 cmn_err(CE_CONT, fmt, x86_feature, FMT_X86_FEATURE); 1428 1429 /* 1430 * Initialize bp_mapin(). 1431 */ 1432 bp_init(MMU_PAGESIZE, HAT_STORECACHING_OK); 1433 1434 /* 1435 * orig_npages is non-zero if physmem has been configured for less 1436 * than the available memory. 1437 */ 1438 if (orig_npages) { 1439 cmn_err(CE_WARN, "!%slimiting physmem to 0x%lx of 0x%lx pages", 1440 (npages == PHYSMEM ? "Due to virtual address space " : ""), 1441 npages, orig_npages); 1442 } 1443 #if defined(__i386) 1444 if (eprom_kernelbase && (eprom_kernelbase != kernelbase)) 1445 cmn_err(CE_WARN, "kernelbase value, User specified 0x%lx, " 1446 "System using 0x%lx", 1447 (uintptr_t)eprom_kernelbase, (uintptr_t)kernelbase); 1448 #endif 1449 1450 #ifdef KERNELBASE_ABI_MIN 1451 if (kernelbase < (uintptr_t)KERNELBASE_ABI_MIN) { 1452 cmn_err(CE_NOTE, "!kernelbase set to 0x%lx, system is not " 1453 "i386 ABI compliant.", (uintptr_t)kernelbase); 1454 } 1455 #endif 1456 1457 #ifndef __xpv 1458 if (plat_dr_support_memory()) { 1459 mem_config_init(); 1460 } 1461 #else /* __xpv */ 1462 /* 1463 * Some of the xen start information has to be relocated up 1464 * into the kernel's permanent address space. 1465 */ 1466 PRM_POINT("calling xen_relocate_start_info()"); 1467 xen_relocate_start_info(); 1468 PRM_POINT("xen_relocate_start_info() done"); 1469 1470 /* 1471 * (Update the vcpu pointer in our cpu structure to point into 1472 * the relocated shared info.) 1473 */ 1474 CPU->cpu_m.mcpu_vcpu_info = 1475 &HYPERVISOR_shared_info->vcpu_info[CPU->cpu_id]; 1476 #endif /* __xpv */ 1477 1478 PRM_POINT("startup_kmem() done"); 1479 } 1480 1481 #ifndef __xpv 1482 /* 1483 * If we have detected that we are running in an HVM environment, we need 1484 * to prepend the PV driver directory to the module search path. 1485 */ 1486 #define HVM_MOD_DIR "/platform/i86hvm/kernel" 1487 static void 1488 update_default_path() 1489 { 1490 char *current, *newpath; 1491 int newlen; 1492 1493 /* 1494 * We are about to resync with krtld. krtld will reset its 1495 * internal module search path iff Solaris has set default_path. 1496 * We want to be sure we're prepending this new directory to the 1497 * right search path. 1498 */ 1499 current = (default_path == NULL) ? kobj_module_path : default_path; 1500 1501 newlen = strlen(HVM_MOD_DIR) + strlen(current) + 2; 1502 newpath = kmem_alloc(newlen, KM_SLEEP); 1503 (void) strcpy(newpath, HVM_MOD_DIR); 1504 (void) strcat(newpath, " "); 1505 (void) strcat(newpath, current); 1506 1507 default_path = newpath; 1508 } 1509 #endif 1510 1511 static void 1512 startup_modules(void) 1513 { 1514 int cnt; 1515 extern void prom_setup(void); 1516 int32_t v, h; 1517 char d[11]; 1518 char *cp; 1519 cmi_hdl_t hdl; 1520 1521 PRM_POINT("startup_modules() starting..."); 1522 1523 #ifndef __xpv 1524 /* 1525 * Initialize ten-micro second timer so that drivers will 1526 * not get short changed in their init phase. This was 1527 * not getting called until clkinit which, on fast cpu's 1528 * caused the drv_usecwait to be way too short. 1529 */ 1530 microfind(); 1531 1532 if (get_hwenv() == HW_XEN_HVM) 1533 update_default_path(); 1534 #endif 1535 1536 /* 1537 * Read the GMT lag from /etc/rtc_config. 1538 */ 1539 sgmtl(process_rtc_config_file()); 1540 1541 /* 1542 * Calculate default settings of system parameters based upon 1543 * maxusers, yet allow to be overridden via the /etc/system file. 1544 */ 1545 param_calc(0); 1546 1547 mod_setup(); 1548 1549 /* 1550 * Initialize system parameters. 1551 */ 1552 param_init(); 1553 1554 /* 1555 * Initialize the default brands 1556 */ 1557 brand_init(); 1558 1559 /* 1560 * maxmem is the amount of physical memory we're playing with. 1561 */ 1562 maxmem = physmem; 1563 1564 /* 1565 * Initialize segment management stuff. 1566 */ 1567 seg_init(); 1568 1569 if (modload("fs", "specfs") == -1) 1570 halt("Can't load specfs"); 1571 1572 if (modload("fs", "devfs") == -1) 1573 halt("Can't load devfs"); 1574 1575 if (modload("fs", "dev") == -1) 1576 halt("Can't load dev"); 1577 1578 if (modload("fs", "procfs") == -1) 1579 halt("Can't load procfs"); 1580 1581 (void) modloadonly("sys", "lbl_edition"); 1582 1583 dispinit(); 1584 1585 /* 1586 * This is needed here to initialize hw_serial[] for cluster booting. 1587 */ 1588 if ((h = set_soft_hostid()) == HW_INVALID_HOSTID) { 1589 cmn_err(CE_WARN, "Unable to set hostid"); 1590 } else { 1591 for (v = h, cnt = 0; cnt < 10; cnt++) { 1592 d[cnt] = (char)(v % 10); 1593 v /= 10; 1594 if (v == 0) 1595 break; 1596 } 1597 for (cp = hw_serial; cnt >= 0; cnt--) 1598 *cp++ = d[cnt] + '0'; 1599 *cp = 0; 1600 } 1601 1602 /* Read cluster configuration data. */ 1603 clconf_init(); 1604 1605 #if defined(__xpv) 1606 (void) ec_init(); 1607 gnttab_init(); 1608 (void) xs_early_init(); 1609 #endif /* __xpv */ 1610 1611 /* 1612 * Create a kernel device tree. First, create rootnex and 1613 * then invoke bus specific code to probe devices. 1614 */ 1615 setup_ddi(); 1616 1617 #ifdef __xpv 1618 if (DOMAIN_IS_INITDOMAIN(xen_info)) 1619 #endif 1620 { 1621 /* 1622 * Load the System Management BIOS into the global ksmbios 1623 * handle, if an SMBIOS is present on this system. 1624 */ 1625 ksmbios = smbios_open(NULL, SMB_VERSION, ksmbios_flags, NULL); 1626 } 1627 1628 1629 /* 1630 * Set up the CPU module subsystem for the boot cpu in the native 1631 * case, and all physical cpu resource in the xpv dom0 case. 1632 * Modifies the device tree, so this must be done after 1633 * setup_ddi(). 1634 */ 1635 #ifdef __xpv 1636 /* 1637 * If paravirtualized and on dom0 then we initialize all physical 1638 * cpu handles now; if paravirtualized on a domU then do not 1639 * initialize. 1640 */ 1641 if (DOMAIN_IS_INITDOMAIN(xen_info)) { 1642 xen_mc_lcpu_cookie_t cpi; 1643 1644 for (cpi = xen_physcpu_next(NULL); cpi != NULL; 1645 cpi = xen_physcpu_next(cpi)) { 1646 if ((hdl = cmi_init(CMI_HDL_SOLARIS_xVM_MCA, 1647 xen_physcpu_chipid(cpi), xen_physcpu_coreid(cpi), 1648 xen_physcpu_strandid(cpi))) != NULL && 1649 (x86_feature & X86_MCA)) 1650 cmi_mca_init(hdl); 1651 } 1652 } 1653 #else 1654 /* 1655 * Initialize a handle for the boot cpu - others will initialize 1656 * as they startup. Do not do this if we know we are in an HVM domU. 1657 */ 1658 if ((get_hwenv() != HW_XEN_HVM) && 1659 (hdl = cmi_init(CMI_HDL_NATIVE, cmi_ntv_hwchipid(CPU), 1660 cmi_ntv_hwcoreid(CPU), cmi_ntv_hwstrandid(CPU))) != NULL && 1661 (x86_feature & X86_MCA)) { 1662 cmi_mca_init(hdl); 1663 CPU->cpu_m.mcpu_cmi_hdl = hdl; 1664 } 1665 #endif /* __xpv */ 1666 1667 /* 1668 * Fake a prom tree such that /dev/openprom continues to work 1669 */ 1670 PRM_POINT("startup_modules: calling prom_setup..."); 1671 prom_setup(); 1672 PRM_POINT("startup_modules: done"); 1673 1674 /* 1675 * Load all platform specific modules 1676 */ 1677 PRM_POINT("startup_modules: calling psm_modload..."); 1678 psm_modload(); 1679 1680 PRM_POINT("startup_modules() done"); 1681 } 1682 1683 /* 1684 * claim a "setaside" boot page for use in the kernel 1685 */ 1686 page_t * 1687 boot_claim_page(pfn_t pfn) 1688 { 1689 page_t *pp; 1690 1691 pp = page_numtopp_nolock(pfn); 1692 ASSERT(pp != NULL); 1693 1694 if (PP_ISBOOTPAGES(pp)) { 1695 if (pp->p_next != NULL) 1696 pp->p_next->p_prev = pp->p_prev; 1697 if (pp->p_prev == NULL) 1698 bootpages = pp->p_next; 1699 else 1700 pp->p_prev->p_next = pp->p_next; 1701 } else { 1702 /* 1703 * htable_attach() expects a base pagesize page 1704 */ 1705 if (pp->p_szc != 0) 1706 page_boot_demote(pp); 1707 pp = page_numtopp(pfn, SE_EXCL); 1708 } 1709 return (pp); 1710 } 1711 1712 /* 1713 * Walk through the pagetables looking for pages mapped in by boot. If the 1714 * setaside flag is set the pages are expected to be returned to the 1715 * kernel later in boot, so we add them to the bootpages list. 1716 */ 1717 static void 1718 protect_boot_range(uintptr_t low, uintptr_t high, int setaside) 1719 { 1720 uintptr_t va = low; 1721 size_t len; 1722 uint_t prot; 1723 pfn_t pfn; 1724 page_t *pp; 1725 pgcnt_t boot_protect_cnt = 0; 1726 1727 while (kbm_probe(&va, &len, &pfn, &prot) != 0 && va < high) { 1728 if (va + len >= high) 1729 panic("0x%lx byte mapping at 0x%p exceeds boot's " 1730 "legal range.", len, (void *)va); 1731 1732 while (len > 0) { 1733 pp = page_numtopp_alloc(pfn); 1734 if (pp != NULL) { 1735 if (setaside == 0) 1736 panic("Unexpected mapping by boot. " 1737 "addr=%p pfn=%lx\n", 1738 (void *)va, pfn); 1739 1740 pp->p_next = bootpages; 1741 pp->p_prev = NULL; 1742 PP_SETBOOTPAGES(pp); 1743 if (bootpages != NULL) { 1744 bootpages->p_prev = pp; 1745 } 1746 bootpages = pp; 1747 ++boot_protect_cnt; 1748 } 1749 1750 ++pfn; 1751 len -= MMU_PAGESIZE; 1752 va += MMU_PAGESIZE; 1753 } 1754 } 1755 PRM_DEBUG(boot_protect_cnt); 1756 } 1757 1758 /* 1759 * 1760 */ 1761 static void 1762 layout_kernel_va(void) 1763 { 1764 PRM_POINT("layout_kernel_va() starting..."); 1765 /* 1766 * Establish the final size of the kernel's heap, size of segmap, 1767 * segkp, etc. 1768 */ 1769 1770 #if defined(__amd64) 1771 1772 kpm_vbase = (caddr_t)segkpm_base; 1773 if (physmax + 1 < plat_dr_physmax) { 1774 kpm_size = ROUND_UP_LPAGE(mmu_ptob(plat_dr_physmax)); 1775 } else { 1776 kpm_size = ROUND_UP_LPAGE(mmu_ptob(physmax + 1)); 1777 } 1778 if ((uintptr_t)kpm_vbase + kpm_size > (uintptr_t)valloc_base) 1779 panic("not enough room for kpm!"); 1780 PRM_DEBUG(kpm_size); 1781 PRM_DEBUG(kpm_vbase); 1782 1783 /* 1784 * By default we create a seg_kp in 64 bit kernels, it's a little 1785 * faster to access than embedding it in the heap. 1786 */ 1787 segkp_base = (caddr_t)valloc_base + valloc_sz; 1788 if (!segkp_fromheap) { 1789 size_t sz = mmu_ptob(segkpsize); 1790 1791 /* 1792 * determine size of segkp 1793 */ 1794 if (sz < SEGKPMINSIZE || sz > SEGKPMAXSIZE) { 1795 sz = SEGKPDEFSIZE; 1796 cmn_err(CE_WARN, "!Illegal value for segkpsize. " 1797 "segkpsize has been reset to %ld pages", 1798 mmu_btop(sz)); 1799 } 1800 sz = MIN(sz, MAX(SEGKPMINSIZE, mmu_ptob(physmem))); 1801 1802 segkpsize = mmu_btop(ROUND_UP_LPAGE(sz)); 1803 } 1804 PRM_DEBUG(segkp_base); 1805 PRM_DEBUG(segkpsize); 1806 1807 /* 1808 * segzio is used for ZFS cached data. It uses a distinct VA 1809 * segment (from kernel heap) so that we can easily tell not to 1810 * include it in kernel crash dumps on 64 bit kernels. The trick is 1811 * to give it lots of VA, but not constrain the kernel heap. 1812 * We scale the size of segzio linearly with physmem up to 1813 * SEGZIOMAXSIZE. Above that amount it scales at 50% of physmem. 1814 */ 1815 segzio_base = segkp_base + mmu_ptob(segkpsize); 1816 if (segzio_fromheap) { 1817 segziosize = 0; 1818 } else { 1819 size_t physmem_size = mmu_ptob(physmem); 1820 size_t size = (segziosize == 0) ? 1821 physmem_size : mmu_ptob(segziosize); 1822 1823 if (size < SEGZIOMINSIZE) 1824 size = SEGZIOMINSIZE; 1825 if (size > SEGZIOMAXSIZE) { 1826 size = SEGZIOMAXSIZE; 1827 if (physmem_size > size) 1828 size += (physmem_size - size) / 2; 1829 } 1830 segziosize = mmu_btop(ROUND_UP_LPAGE(size)); 1831 } 1832 PRM_DEBUG(segziosize); 1833 PRM_DEBUG(segzio_base); 1834 1835 /* 1836 * Put the range of VA for device mappings next, kmdb knows to not 1837 * grep in this range of addresses. 1838 */ 1839 toxic_addr = 1840 ROUND_UP_LPAGE((uintptr_t)segzio_base + mmu_ptob(segziosize)); 1841 PRM_DEBUG(toxic_addr); 1842 segmap_start = ROUND_UP_LPAGE(toxic_addr + toxic_size); 1843 #else /* __i386 */ 1844 segmap_start = ROUND_UP_LPAGE(kernelbase); 1845 #endif /* __i386 */ 1846 PRM_DEBUG(segmap_start); 1847 1848 /* 1849 * Users can change segmapsize through eeprom. If the variable 1850 * is tuned through eeprom, there is no upper bound on the 1851 * size of segmap. 1852 */ 1853 segmapsize = MAX(ROUND_UP_LPAGE(segmapsize), SEGMAPDEFAULT); 1854 1855 #if defined(__i386) 1856 /* 1857 * 32-bit systems don't have segkpm or segkp, so segmap appears at 1858 * the bottom of the kernel's address range. Set aside space for a 1859 * small red zone just below the start of segmap. 1860 */ 1861 segmap_start += KERNEL_REDZONE_SIZE; 1862 segmapsize -= KERNEL_REDZONE_SIZE; 1863 #endif 1864 1865 PRM_DEBUG(segmap_start); 1866 PRM_DEBUG(segmapsize); 1867 kernelheap = (caddr_t)ROUND_UP_LPAGE(segmap_start + segmapsize); 1868 PRM_DEBUG(kernelheap); 1869 PRM_POINT("layout_kernel_va() done..."); 1870 } 1871 1872 /* 1873 * Finish initializing the VM system, now that we are no longer 1874 * relying on the boot time memory allocators. 1875 */ 1876 static void 1877 startup_vm(void) 1878 { 1879 struct segmap_crargs a; 1880 1881 extern int use_brk_lpg, use_stk_lpg; 1882 1883 PRM_POINT("startup_vm() starting..."); 1884 1885 /* 1886 * Initialize the hat layer. 1887 */ 1888 hat_init(); 1889 1890 /* 1891 * Do final allocations of HAT data structures that need to 1892 * be allocated before quiescing the boot loader. 1893 */ 1894 PRM_POINT("Calling hat_kern_alloc()..."); 1895 hat_kern_alloc((caddr_t)segmap_start, segmapsize, ekernelheap); 1896 PRM_POINT("hat_kern_alloc() done"); 1897 1898 #ifndef __xpv 1899 /* 1900 * Setup Page Attribute Table 1901 */ 1902 pat_sync(); 1903 #endif 1904 1905 /* 1906 * The next two loops are done in distinct steps in order 1907 * to be sure that any page that is doubly mapped (both above 1908 * KERNEL_TEXT and below kernelbase) is dealt with correctly. 1909 * Note this may never happen, but it might someday. 1910 */ 1911 bootpages = NULL; 1912 PRM_POINT("Protecting boot pages"); 1913 1914 /* 1915 * Protect any pages mapped above KERNEL_TEXT that somehow have 1916 * page_t's. This can only happen if something weird allocated 1917 * in this range (like kadb/kmdb). 1918 */ 1919 protect_boot_range(KERNEL_TEXT, (uintptr_t)-1, 0); 1920 1921 /* 1922 * Before we can take over memory allocation/mapping from the boot 1923 * loader we must remove from our free page lists any boot allocated 1924 * pages that stay mapped until release_bootstrap(). 1925 */ 1926 protect_boot_range(0, kernelbase, 1); 1927 1928 1929 /* 1930 * Switch to running on regular HAT (not boot_mmu) 1931 */ 1932 PRM_POINT("Calling hat_kern_setup()..."); 1933 hat_kern_setup(); 1934 1935 /* 1936 * It is no longer safe to call BOP_ALLOC(), so make sure we don't. 1937 */ 1938 bop_no_more_mem(); 1939 1940 PRM_POINT("hat_kern_setup() done"); 1941 1942 hat_cpu_online(CPU); 1943 1944 /* 1945 * Initialize VM system 1946 */ 1947 PRM_POINT("Calling kvm_init()..."); 1948 kvm_init(); 1949 PRM_POINT("kvm_init() done"); 1950 1951 /* 1952 * Tell kmdb that the VM system is now working 1953 */ 1954 if (boothowto & RB_DEBUG) 1955 kdi_dvec_vmready(); 1956 1957 #if defined(__xpv) 1958 /* 1959 * Populate the I/O pool on domain 0 1960 */ 1961 if (DOMAIN_IS_INITDOMAIN(xen_info)) { 1962 extern long populate_io_pool(void); 1963 long init_io_pool_cnt; 1964 1965 PRM_POINT("Populating reserve I/O page pool"); 1966 init_io_pool_cnt = populate_io_pool(); 1967 PRM_DEBUG(init_io_pool_cnt); 1968 } 1969 #endif 1970 /* 1971 * Mangle the brand string etc. 1972 */ 1973 cpuid_pass3(CPU); 1974 1975 #if defined(__amd64) 1976 1977 /* 1978 * Create the device arena for toxic (to dtrace/kmdb) mappings. 1979 */ 1980 device_arena = vmem_create("device", (void *)toxic_addr, 1981 toxic_size, MMU_PAGESIZE, NULL, NULL, NULL, 0, VM_SLEEP); 1982 1983 #else /* __i386 */ 1984 1985 /* 1986 * allocate the bit map that tracks toxic pages 1987 */ 1988 toxic_bit_map_len = btop((ulong_t)(valloc_base - kernelbase)); 1989 PRM_DEBUG(toxic_bit_map_len); 1990 toxic_bit_map = 1991 kmem_zalloc(BT_SIZEOFMAP(toxic_bit_map_len), KM_NOSLEEP); 1992 ASSERT(toxic_bit_map != NULL); 1993 PRM_DEBUG(toxic_bit_map); 1994 1995 #endif /* __i386 */ 1996 1997 1998 /* 1999 * Now that we've got more VA, as well as the ability to allocate from 2000 * it, tell the debugger. 2001 */ 2002 if (boothowto & RB_DEBUG) 2003 kdi_dvec_memavail(); 2004 2005 /* 2006 * The following code installs a special page fault handler (#pf) 2007 * to work around a pentium bug. 2008 */ 2009 #if !defined(__amd64) && !defined(__xpv) 2010 if (x86_type == X86_TYPE_P5) { 2011 desctbr_t idtr; 2012 gate_desc_t *newidt; 2013 2014 if ((newidt = kmem_zalloc(MMU_PAGESIZE, KM_NOSLEEP)) == NULL) 2015 panic("failed to install pentium_pftrap"); 2016 2017 bcopy(idt0, newidt, NIDT * sizeof (*idt0)); 2018 set_gatesegd(&newidt[T_PGFLT], &pentium_pftrap, 2019 KCS_SEL, SDT_SYSIGT, TRP_KPL, 0); 2020 2021 (void) as_setprot(&kas, (caddr_t)newidt, MMU_PAGESIZE, 2022 PROT_READ | PROT_EXEC); 2023 2024 CPU->cpu_idt = newidt; 2025 idtr.dtr_base = (uintptr_t)CPU->cpu_idt; 2026 idtr.dtr_limit = (NIDT * sizeof (*idt0)) - 1; 2027 wr_idtr(&idtr); 2028 } 2029 #endif /* !__amd64 */ 2030 2031 #if !defined(__xpv) 2032 /* 2033 * Map page pfn=0 for drivers, such as kd, that need to pick up 2034 * parameters left there by controllers/BIOS. 2035 */ 2036 PRM_POINT("setup up p0_va"); 2037 p0_va = i86devmap(0, 1, PROT_READ); 2038 PRM_DEBUG(p0_va); 2039 #endif 2040 2041 cmn_err(CE_CONT, "?mem = %luK (0x%lx)\n", 2042 physinstalled << (MMU_PAGESHIFT - 10), ptob(physinstalled)); 2043 2044 /* 2045 * disable automatic large pages for small memory systems or 2046 * when the disable flag is set. 2047 * 2048 * Do not yet consider page sizes larger than 2m/4m. 2049 */ 2050 if (!auto_lpg_disable && mmu.max_page_level > 0) { 2051 max_uheap_lpsize = LEVEL_SIZE(1); 2052 max_ustack_lpsize = LEVEL_SIZE(1); 2053 max_privmap_lpsize = LEVEL_SIZE(1); 2054 max_uidata_lpsize = LEVEL_SIZE(1); 2055 max_utext_lpsize = LEVEL_SIZE(1); 2056 max_shm_lpsize = LEVEL_SIZE(1); 2057 } 2058 if (physmem < privm_lpg_min_physmem || mmu.max_page_level == 0 || 2059 auto_lpg_disable) { 2060 use_brk_lpg = 0; 2061 use_stk_lpg = 0; 2062 } 2063 mcntl0_lpsize = LEVEL_SIZE(mmu.umax_page_level); 2064 2065 PRM_POINT("Calling hat_init_finish()..."); 2066 hat_init_finish(); 2067 PRM_POINT("hat_init_finish() done"); 2068 2069 /* 2070 * Initialize the segkp segment type. 2071 */ 2072 rw_enter(&kas.a_lock, RW_WRITER); 2073 PRM_POINT("Attaching segkp"); 2074 if (segkp_fromheap) { 2075 segkp->s_as = &kas; 2076 } else if (seg_attach(&kas, (caddr_t)segkp_base, mmu_ptob(segkpsize), 2077 segkp) < 0) { 2078 panic("startup: cannot attach segkp"); 2079 /*NOTREACHED*/ 2080 } 2081 PRM_POINT("Doing segkp_create()"); 2082 if (segkp_create(segkp) != 0) { 2083 panic("startup: segkp_create failed"); 2084 /*NOTREACHED*/ 2085 } 2086 PRM_DEBUG(segkp); 2087 rw_exit(&kas.a_lock); 2088 2089 /* 2090 * kpm segment 2091 */ 2092 segmap_kpm = 0; 2093 if (kpm_desired) { 2094 kpm_init(); 2095 kpm_enable = 1; 2096 } 2097 2098 /* 2099 * Now create segmap segment. 2100 */ 2101 rw_enter(&kas.a_lock, RW_WRITER); 2102 if (seg_attach(&kas, (caddr_t)segmap_start, segmapsize, segmap) < 0) { 2103 panic("cannot attach segmap"); 2104 /*NOTREACHED*/ 2105 } 2106 PRM_DEBUG(segmap); 2107 2108 a.prot = PROT_READ | PROT_WRITE; 2109 a.shmsize = 0; 2110 a.nfreelist = segmapfreelists; 2111 2112 if (segmap_create(segmap, (caddr_t)&a) != 0) 2113 panic("segmap_create segmap"); 2114 rw_exit(&kas.a_lock); 2115 2116 setup_vaddr_for_ppcopy(CPU); 2117 2118 segdev_init(); 2119 #if defined(__xpv) 2120 if (DOMAIN_IS_INITDOMAIN(xen_info)) 2121 #endif 2122 pmem_init(); 2123 2124 PRM_POINT("startup_vm() done"); 2125 } 2126 2127 /* 2128 * Load a tod module for the non-standard tod part found on this system. 2129 */ 2130 static void 2131 load_tod_module(char *todmod) 2132 { 2133 if (modload("tod", todmod) == -1) 2134 halt("Can't load TOD module"); 2135 } 2136 2137 static void 2138 startup_end(void) 2139 { 2140 int i; 2141 extern void setx86isalist(void); 2142 extern void cpu_event_init(void); 2143 2144 PRM_POINT("startup_end() starting..."); 2145 2146 /* 2147 * Perform tasks that get done after most of the VM 2148 * initialization has been done but before the clock 2149 * and other devices get started. 2150 */ 2151 kern_setup1(); 2152 2153 /* 2154 * Perform CPC initialization for this CPU. 2155 */ 2156 kcpc_hw_init(CPU); 2157 2158 /* 2159 * Initialize cpu event framework. 2160 */ 2161 cpu_event_init(); 2162 2163 #if defined(OPTERON_WORKAROUND_6323525) 2164 if (opteron_workaround_6323525) 2165 patch_workaround_6323525(); 2166 #endif 2167 /* 2168 * If needed, load TOD module now so that ddi_get_time(9F) etc. work 2169 * (For now, "needed" is defined as set tod_module_name in /etc/system) 2170 */ 2171 if (tod_module_name != NULL) { 2172 PRM_POINT("load_tod_module()"); 2173 load_tod_module(tod_module_name); 2174 } 2175 2176 #if defined(__xpv) 2177 /* 2178 * Forceload interposing TOD module for the hypervisor. 2179 */ 2180 PRM_POINT("load_tod_module()"); 2181 load_tod_module("xpvtod"); 2182 #endif 2183 2184 /* 2185 * Configure the system. 2186 */ 2187 PRM_POINT("Calling configure()..."); 2188 configure(); /* set up devices */ 2189 PRM_POINT("configure() done"); 2190 2191 /* 2192 * Set the isa_list string to the defined instruction sets we 2193 * support. 2194 */ 2195 setx86isalist(); 2196 cpu_intr_alloc(CPU, NINTR_THREADS); 2197 psm_install(); 2198 2199 /* 2200 * We're done with bootops. We don't unmap the bootstrap yet because 2201 * we're still using bootsvcs. 2202 */ 2203 PRM_POINT("NULLing out bootops"); 2204 *bootopsp = (struct bootops *)NULL; 2205 bootops = (struct bootops *)NULL; 2206 2207 #if defined(__xpv) 2208 ec_init_debug_irq(); 2209 xs_domu_init(); 2210 #endif 2211 2212 #if defined(__amd64) && !defined(__xpv) 2213 /* 2214 * Intel IOMMU has been setup/initialized in ddi_impl.c 2215 * Start it up now. 2216 */ 2217 immu_startup(); 2218 #endif 2219 2220 PRM_POINT("Enabling interrupts"); 2221 (*picinitf)(); 2222 sti(); 2223 #if defined(__xpv) 2224 ASSERT(CPU->cpu_m.mcpu_vcpu_info->evtchn_upcall_mask == 0); 2225 xen_late_startup(); 2226 #endif 2227 2228 (void) add_avsoftintr((void *)&softlevel1_hdl, 1, softlevel1, 2229 "softlevel1", NULL, NULL); /* XXX to be moved later */ 2230 2231 /* 2232 * Register these software interrupts for ddi timer. 2233 * Software interrupts up to the level 10 are supported. 2234 */ 2235 for (i = DDI_IPL_1; i <= DDI_IPL_10; i++) { 2236 char name[sizeof ("timer_softintr") + 2]; 2237 (void) sprintf(name, "timer_softintr%02d", i); 2238 (void) add_avsoftintr((void *)&softlevel_hdl[i-1], i, 2239 (avfunc)timer_softintr, name, (caddr_t)(uintptr_t)i, NULL); 2240 } 2241 2242 #if !defined(__xpv) 2243 if (modload("drv", "amd_iommu") < 0) { 2244 PRM_POINT("No AMD IOMMU present\n"); 2245 } else if (ddi_hold_installed_driver(ddi_name_to_major( 2246 "amd_iommu")) == NULL) { 2247 prom_printf("ERROR: failed to attach AMD IOMMU\n"); 2248 } 2249 #endif 2250 post_startup_cpu_fixups(); 2251 2252 PRM_POINT("startup_end() done"); 2253 } 2254 2255 /* 2256 * Don't remove the following 2 variables. They are necessary 2257 * for reading the hostid from the legacy file (/kernel/misc/sysinit). 2258 */ 2259 char *_hs1107 = hw_serial; 2260 ulong_t _bdhs34; 2261 2262 void 2263 post_startup(void) 2264 { 2265 extern void cpupm_init(cpu_t *); 2266 extern void cpu_event_init_cpu(cpu_t *); 2267 2268 /* 2269 * Set the system wide, processor-specific flags to be passed 2270 * to userland via the aux vector for performance hints and 2271 * instruction set extensions. 2272 */ 2273 bind_hwcap(); 2274 2275 #ifdef __xpv 2276 if (DOMAIN_IS_INITDOMAIN(xen_info)) 2277 #endif 2278 { 2279 #if defined(__xpv) 2280 xpv_panic_init(); 2281 #else 2282 /* 2283 * Startup the memory scrubber. 2284 * XXPV This should be running somewhere .. 2285 */ 2286 if (get_hwenv() != HW_XEN_HVM) 2287 memscrub_init(); 2288 #endif 2289 } 2290 2291 /* 2292 * Complete CPU module initialization 2293 */ 2294 cmi_post_startup(); 2295 2296 /* 2297 * Perform forceloading tasks for /etc/system. 2298 */ 2299 (void) mod_sysctl(SYS_FORCELOAD, NULL); 2300 2301 /* 2302 * ON4.0: Force /proc module in until clock interrupt handle fixed 2303 * ON4.0: This must be fixed or restated in /etc/systems. 2304 */ 2305 (void) modload("fs", "procfs"); 2306 2307 (void) i_ddi_attach_hw_nodes("pit_beep"); 2308 2309 #if defined(__i386) 2310 /* 2311 * Check for required functional Floating Point hardware, 2312 * unless FP hardware explicitly disabled. 2313 */ 2314 if (fpu_exists && (fpu_pentium_fdivbug || fp_kind == FP_NO)) 2315 halt("No working FP hardware found"); 2316 #endif 2317 2318 maxmem = freemem; 2319 2320 cpu_event_init_cpu(CPU); 2321 cpupm_init(CPU); 2322 (void) mach_cpu_create_device_node(CPU, NULL); 2323 2324 pg_init(); 2325 } 2326 2327 static int 2328 pp_in_range(page_t *pp, uint64_t low_addr, uint64_t high_addr) 2329 { 2330 return ((pp->p_pagenum >= btop(low_addr)) && 2331 (pp->p_pagenum < btopr(high_addr))); 2332 } 2333 2334 void 2335 release_bootstrap(void) 2336 { 2337 int root_is_ramdisk; 2338 page_t *pp; 2339 extern void kobj_boot_unmountroot(void); 2340 extern dev_t rootdev; 2341 #if !defined(__xpv) 2342 pfn_t pfn; 2343 #endif 2344 2345 /* unmount boot ramdisk and release kmem usage */ 2346 kobj_boot_unmountroot(); 2347 2348 /* 2349 * We're finished using the boot loader so free its pages. 2350 */ 2351 PRM_POINT("Unmapping lower boot pages"); 2352 2353 clear_boot_mappings(0, _userlimit); 2354 2355 postbootkernelbase = kernelbase; 2356 2357 /* 2358 * If root isn't on ramdisk, destroy the hardcoded 2359 * ramdisk node now and release the memory. Else, 2360 * ramdisk memory is kept in rd_pages. 2361 */ 2362 root_is_ramdisk = (getmajor(rootdev) == ddi_name_to_major("ramdisk")); 2363 if (!root_is_ramdisk) { 2364 dev_info_t *dip = ddi_find_devinfo("ramdisk", -1, 0); 2365 ASSERT(dip && ddi_get_parent(dip) == ddi_root_node()); 2366 ndi_rele_devi(dip); /* held from ddi_find_devinfo */ 2367 (void) ddi_remove_child(dip, 0); 2368 } 2369 2370 PRM_POINT("Releasing boot pages"); 2371 while (bootpages) { 2372 extern uint64_t ramdisk_start, ramdisk_end; 2373 pp = bootpages; 2374 bootpages = pp->p_next; 2375 2376 2377 /* Keep pages for the lower 64K */ 2378 if (pp_in_range(pp, 0, 0x40000)) { 2379 pp->p_next = lower_pages; 2380 lower_pages = pp; 2381 lower_pages_count++; 2382 continue; 2383 } 2384 2385 2386 if (root_is_ramdisk && pp_in_range(pp, ramdisk_start, 2387 ramdisk_end)) { 2388 pp->p_next = rd_pages; 2389 rd_pages = pp; 2390 continue; 2391 } 2392 pp->p_next = (struct page *)0; 2393 pp->p_prev = (struct page *)0; 2394 PP_CLRBOOTPAGES(pp); 2395 page_free(pp, 1); 2396 } 2397 PRM_POINT("Boot pages released"); 2398 2399 #if !defined(__xpv) 2400 /* XXPV -- note this following bunch of code needs to be revisited in Xen 3.0 */ 2401 /* 2402 * Find 1 page below 1 MB so that other processors can boot up or 2403 * so that any processor can resume. 2404 * Make sure it has a kernel VA as well as a 1:1 mapping. 2405 * We should have just free'd one up. 2406 */ 2407 2408 /* 2409 * 0x10 pages is 64K. Leave the bottom 64K alone 2410 * for BIOS. 2411 */ 2412 for (pfn = 0x10; pfn < btop(1*1024*1024); pfn++) { 2413 if (page_numtopp_alloc(pfn) == NULL) 2414 continue; 2415 rm_platter_va = i86devmap(pfn, 1, 2416 PROT_READ | PROT_WRITE | PROT_EXEC); 2417 rm_platter_pa = ptob(pfn); 2418 break; 2419 } 2420 if (pfn == btop(1*1024*1024) && use_mp) 2421 panic("No page below 1M available for starting " 2422 "other processors or for resuming from system-suspend"); 2423 #endif /* !__xpv */ 2424 } 2425 2426 /* 2427 * Initialize the platform-specific parts of a page_t. 2428 */ 2429 void 2430 add_physmem_cb(page_t *pp, pfn_t pnum) 2431 { 2432 pp->p_pagenum = pnum; 2433 pp->p_mapping = NULL; 2434 pp->p_embed = 0; 2435 pp->p_share = 0; 2436 pp->p_mlentry = 0; 2437 } 2438 2439 /* 2440 * kphysm_init() initializes physical memory. 2441 */ 2442 static pgcnt_t 2443 kphysm_init( 2444 page_t *pp, 2445 pgcnt_t npages) 2446 { 2447 struct memlist *pmem; 2448 struct memseg *cur_memseg; 2449 pfn_t base_pfn; 2450 pfn_t end_pfn; 2451 pgcnt_t num; 2452 pgcnt_t pages_done = 0; 2453 uint64_t addr; 2454 uint64_t size; 2455 extern pfn_t ddiphysmin; 2456 extern int mnode_xwa; 2457 int ms = 0, me = 0; 2458 2459 ASSERT(page_hash != NULL && page_hashsz != 0); 2460 2461 cur_memseg = memseg_base; 2462 for (pmem = phys_avail; pmem && npages; pmem = pmem->ml_next) { 2463 /* 2464 * In a 32 bit kernel can't use higher memory if we're 2465 * not booting in PAE mode. This check takes care of that. 2466 */ 2467 addr = pmem->ml_address; 2468 size = pmem->ml_size; 2469 if (btop(addr) > physmax) 2470 continue; 2471 2472 /* 2473 * align addr and size - they may not be at page boundaries 2474 */ 2475 if ((addr & MMU_PAGEOFFSET) != 0) { 2476 addr += MMU_PAGEOFFSET; 2477 addr &= ~(uint64_t)MMU_PAGEOFFSET; 2478 size -= addr - pmem->ml_address; 2479 } 2480 2481 /* only process pages below or equal to physmax */ 2482 if ((btop(addr + size) - 1) > physmax) 2483 size = ptob(physmax - btop(addr) + 1); 2484 2485 num = btop(size); 2486 if (num == 0) 2487 continue; 2488 2489 if (num > npages) 2490 num = npages; 2491 2492 npages -= num; 2493 pages_done += num; 2494 base_pfn = btop(addr); 2495 2496 if (prom_debug) 2497 prom_printf("MEMSEG addr=0x%" PRIx64 2498 " pgs=0x%lx pfn 0x%lx-0x%lx\n", 2499 addr, num, base_pfn, base_pfn + num); 2500 2501 /* 2502 * Ignore pages below ddiphysmin to simplify ddi memory 2503 * allocation with non-zero addr_lo requests. 2504 */ 2505 if (base_pfn < ddiphysmin) { 2506 if (base_pfn + num <= ddiphysmin) 2507 continue; 2508 pp += (ddiphysmin - base_pfn); 2509 num -= (ddiphysmin - base_pfn); 2510 base_pfn = ddiphysmin; 2511 } 2512 2513 /* 2514 * mnode_xwa is greater than 1 when large pages regions can 2515 * cross memory node boundaries. To prevent the formation 2516 * of these large pages, configure the memsegs based on the 2517 * memory node ranges which had been made non-contiguous. 2518 */ 2519 if (mnode_xwa > 1) { 2520 2521 end_pfn = base_pfn + num - 1; 2522 ms = PFN_2_MEM_NODE(base_pfn); 2523 me = PFN_2_MEM_NODE(end_pfn); 2524 2525 if (ms != me) { 2526 /* 2527 * current range spans more than 1 memory node. 2528 * Set num to only the pfn range in the start 2529 * memory node. 2530 */ 2531 num = mem_node_config[ms].physmax - base_pfn 2532 + 1; 2533 ASSERT(end_pfn > mem_node_config[ms].physmax); 2534 } 2535 } 2536 2537 for (;;) { 2538 /* 2539 * Build the memsegs entry 2540 */ 2541 cur_memseg->pages = pp; 2542 cur_memseg->epages = pp + num; 2543 cur_memseg->pages_base = base_pfn; 2544 cur_memseg->pages_end = base_pfn + num; 2545 2546 /* 2547 * Insert into memseg list in decreasing pfn range 2548 * order. Low memory is typically more fragmented such 2549 * that this ordering keeps the larger ranges at the 2550 * front of the list for code that searches memseg. 2551 * This ASSERTS that the memsegs coming in from boot 2552 * are in increasing physical address order and not 2553 * contiguous. 2554 */ 2555 if (memsegs != NULL) { 2556 ASSERT(cur_memseg->pages_base >= 2557 memsegs->pages_end); 2558 cur_memseg->next = memsegs; 2559 } 2560 memsegs = cur_memseg; 2561 2562 /* 2563 * add_physmem() initializes the PSM part of the page 2564 * struct by calling the PSM back with add_physmem_cb(). 2565 * In addition it coalesces pages into larger pages as 2566 * it initializes them. 2567 */ 2568 add_physmem(pp, num, base_pfn); 2569 cur_memseg++; 2570 availrmem_initial += num; 2571 availrmem += num; 2572 2573 pp += num; 2574 if (ms >= me) 2575 break; 2576 2577 /* process next memory node range */ 2578 ms++; 2579 base_pfn = mem_node_config[ms].physbase; 2580 num = MIN(mem_node_config[ms].physmax, 2581 end_pfn) - base_pfn + 1; 2582 } 2583 } 2584 2585 PRM_DEBUG(availrmem_initial); 2586 PRM_DEBUG(availrmem); 2587 PRM_DEBUG(freemem); 2588 build_pfn_hash(); 2589 return (pages_done); 2590 } 2591 2592 /* 2593 * Kernel VM initialization. 2594 */ 2595 static void 2596 kvm_init(void) 2597 { 2598 ASSERT((((uintptr_t)s_text) & MMU_PAGEOFFSET) == 0); 2599 2600 /* 2601 * Put the kernel segments in kernel address space. 2602 */ 2603 rw_enter(&kas.a_lock, RW_WRITER); 2604 as_avlinit(&kas); 2605 2606 (void) seg_attach(&kas, s_text, e_moddata - s_text, &ktextseg); 2607 (void) segkmem_create(&ktextseg); 2608 2609 (void) seg_attach(&kas, (caddr_t)valloc_base, valloc_sz, &kvalloc); 2610 (void) segkmem_create(&kvalloc); 2611 2612 (void) seg_attach(&kas, kernelheap, 2613 ekernelheap - kernelheap, &kvseg); 2614 (void) segkmem_create(&kvseg); 2615 2616 if (core_size > 0) { 2617 PRM_POINT("attaching kvseg_core"); 2618 (void) seg_attach(&kas, (caddr_t)core_base, core_size, 2619 &kvseg_core); 2620 (void) segkmem_create(&kvseg_core); 2621 } 2622 2623 if (segziosize > 0) { 2624 PRM_POINT("attaching segzio"); 2625 (void) seg_attach(&kas, segzio_base, mmu_ptob(segziosize), 2626 &kzioseg); 2627 (void) segkmem_zio_create(&kzioseg); 2628 2629 /* create zio area covering new segment */ 2630 segkmem_zio_init(segzio_base, mmu_ptob(segziosize)); 2631 } 2632 2633 (void) seg_attach(&kas, kdi_segdebugbase, kdi_segdebugsize, &kdebugseg); 2634 (void) segkmem_create(&kdebugseg); 2635 2636 rw_exit(&kas.a_lock); 2637 2638 /* 2639 * Ensure that the red zone at kernelbase is never accessible. 2640 */ 2641 PRM_POINT("protecting redzone"); 2642 (void) as_setprot(&kas, (caddr_t)kernelbase, KERNEL_REDZONE_SIZE, 0); 2643 2644 /* 2645 * Make the text writable so that it can be hot patched by DTrace. 2646 */ 2647 (void) as_setprot(&kas, s_text, e_modtext - s_text, 2648 PROT_READ | PROT_WRITE | PROT_EXEC); 2649 2650 /* 2651 * Make data writable until end. 2652 */ 2653 (void) as_setprot(&kas, s_data, e_moddata - s_data, 2654 PROT_READ | PROT_WRITE | PROT_EXEC); 2655 } 2656 2657 #ifndef __xpv 2658 /* 2659 * Solaris adds an entry for Write Combining caching to the PAT 2660 */ 2661 static uint64_t pat_attr_reg = PAT_DEFAULT_ATTRIBUTE; 2662 2663 void 2664 pat_sync(void) 2665 { 2666 ulong_t cr0, cr0_orig, cr4; 2667 2668 if (!(x86_feature & X86_PAT)) 2669 return; 2670 cr0_orig = cr0 = getcr0(); 2671 cr4 = getcr4(); 2672 2673 /* disable caching and flush all caches and TLBs */ 2674 cr0 |= CR0_CD; 2675 cr0 &= ~CR0_NW; 2676 setcr0(cr0); 2677 invalidate_cache(); 2678 if (cr4 & CR4_PGE) { 2679 setcr4(cr4 & ~(ulong_t)CR4_PGE); 2680 setcr4(cr4); 2681 } else { 2682 reload_cr3(); 2683 } 2684 2685 /* add our entry to the PAT */ 2686 wrmsr(REG_PAT, pat_attr_reg); 2687 2688 /* flush TLBs and cache again, then reenable cr0 caching */ 2689 if (cr4 & CR4_PGE) { 2690 setcr4(cr4 & ~(ulong_t)CR4_PGE); 2691 setcr4(cr4); 2692 } else { 2693 reload_cr3(); 2694 } 2695 invalidate_cache(); 2696 setcr0(cr0_orig); 2697 } 2698 2699 #endif /* !__xpv */ 2700 2701 #if defined(_SOFT_HOSTID) 2702 /* 2703 * On platforms that do not have a hardware serial number, attempt 2704 * to set one based on the contents of /etc/hostid. If this file does 2705 * not exist, assume that we are to generate a new hostid and set 2706 * it in the kernel, for subsequent saving by a userland process 2707 * once the system is up and the root filesystem is mounted r/w. 2708 * 2709 * In order to gracefully support upgrade on OpenSolaris, if 2710 * /etc/hostid does not exist, we will attempt to get a serial number 2711 * using the legacy method (/kernel/misc/sysinit). 2712 * 2713 * In an attempt to make the hostid less prone to abuse 2714 * (for license circumvention, etc), we store it in /etc/hostid 2715 * in rot47 format. 2716 */ 2717 extern volatile unsigned long tenmicrodata; 2718 static int atoi(char *); 2719 2720 static int32_t 2721 set_soft_hostid(void) 2722 { 2723 struct _buf *file; 2724 char tokbuf[MAXNAMELEN]; 2725 token_t token; 2726 int done = 0; 2727 u_longlong_t tmp; 2728 int i; 2729 int32_t hostid = (int32_t)HW_INVALID_HOSTID; 2730 unsigned char *c; 2731 hrtime_t tsc; 2732 2733 /* 2734 * If /etc/hostid file not found, we'd like to get a pseudo 2735 * random number to use at the hostid. A nice way to do this 2736 * is to read the real time clock. To remain xen-compatible, 2737 * we can't poke the real hardware, so we use tsc_read() to 2738 * read the real time clock. However, there is an ominous 2739 * warning in tsc_read that says it can return zero, so we 2740 * deal with that possibility by falling back to using the 2741 * (hopefully random enough) value in tenmicrodata. 2742 */ 2743 2744 if ((file = kobj_open_file(hostid_file)) == (struct _buf *)-1) { 2745 /* 2746 * hostid file not found - try to load sysinit module 2747 * and see if it has a nonzero hostid value...use that 2748 * instead of generating a new hostid here if so. 2749 */ 2750 if ((i = modload("misc", "sysinit")) != -1) { 2751 if (strlen(hw_serial) > 0) 2752 hostid = (int32_t)atoi(hw_serial); 2753 (void) modunload(i); 2754 } 2755 if (hostid == HW_INVALID_HOSTID) { 2756 tsc = tsc_read(); 2757 if (tsc == 0) /* tsc_read can return zero sometimes */ 2758 hostid = (int32_t)tenmicrodata & 0x0CFFFFF; 2759 else 2760 hostid = (int32_t)tsc & 0x0CFFFFF; 2761 } 2762 } else { 2763 /* hostid file found */ 2764 while (!done) { 2765 token = kobj_lex(file, tokbuf, sizeof (tokbuf)); 2766 2767 switch (token) { 2768 case POUND: 2769 /* 2770 * skip comments 2771 */ 2772 kobj_find_eol(file); 2773 break; 2774 case STRING: 2775 /* 2776 * un-rot47 - obviously this 2777 * nonsense is ascii-specific 2778 */ 2779 for (c = (unsigned char *)tokbuf; 2780 *c != '\0'; c++) { 2781 *c += 47; 2782 if (*c > '~') 2783 *c -= 94; 2784 else if (*c < '!') 2785 *c += 94; 2786 } 2787 /* 2788 * now we should have a real number 2789 */ 2790 2791 if (kobj_getvalue(tokbuf, &tmp) != 0) 2792 kobj_file_err(CE_WARN, file, 2793 "Bad value %s for hostid", 2794 tokbuf); 2795 else 2796 hostid = (int32_t)tmp; 2797 2798 break; 2799 case EOF: 2800 done = 1; 2801 /* FALLTHROUGH */ 2802 case NEWLINE: 2803 kobj_newline(file); 2804 break; 2805 default: 2806 break; 2807 2808 } 2809 } 2810 if (hostid == HW_INVALID_HOSTID) /* didn't find a hostid */ 2811 kobj_file_err(CE_WARN, file, 2812 "hostid missing or corrupt"); 2813 2814 kobj_close_file(file); 2815 } 2816 /* 2817 * hostid is now the value read from /etc/hostid, or the 2818 * new hostid we generated in this routine or HW_INVALID_HOSTID if not 2819 * set. 2820 */ 2821 return (hostid); 2822 } 2823 2824 static int 2825 atoi(char *p) 2826 { 2827 int i = 0; 2828 2829 while (*p != '\0') 2830 i = 10 * i + (*p++ - '0'); 2831 2832 return (i); 2833 } 2834 2835 #endif /* _SOFT_HOSTID */ 2836 2837 void 2838 get_system_configuration(void) 2839 { 2840 char prop[32]; 2841 u_longlong_t nodes_ll, cpus_pernode_ll, lvalue; 2842 2843 if (BOP_GETPROPLEN(bootops, "nodes") > sizeof (prop) || 2844 BOP_GETPROP(bootops, "nodes", prop) < 0 || 2845 kobj_getvalue(prop, &nodes_ll) == -1 || 2846 nodes_ll > MAXNODES || 2847 BOP_GETPROPLEN(bootops, "cpus_pernode") > sizeof (prop) || 2848 BOP_GETPROP(bootops, "cpus_pernode", prop) < 0 || 2849 kobj_getvalue(prop, &cpus_pernode_ll) == -1) { 2850 system_hardware.hd_nodes = 1; 2851 system_hardware.hd_cpus_per_node = 0; 2852 } else { 2853 system_hardware.hd_nodes = (int)nodes_ll; 2854 system_hardware.hd_cpus_per_node = (int)cpus_pernode_ll; 2855 } 2856 2857 if (BOP_GETPROPLEN(bootops, "kernelbase") > sizeof (prop) || 2858 BOP_GETPROP(bootops, "kernelbase", prop) < 0 || 2859 kobj_getvalue(prop, &lvalue) == -1) 2860 eprom_kernelbase = NULL; 2861 else 2862 eprom_kernelbase = (uintptr_t)lvalue; 2863 2864 if (BOP_GETPROPLEN(bootops, "segmapsize") > sizeof (prop) || 2865 BOP_GETPROP(bootops, "segmapsize", prop) < 0 || 2866 kobj_getvalue(prop, &lvalue) == -1) 2867 segmapsize = SEGMAPDEFAULT; 2868 else 2869 segmapsize = (uintptr_t)lvalue; 2870 2871 if (BOP_GETPROPLEN(bootops, "segmapfreelists") > sizeof (prop) || 2872 BOP_GETPROP(bootops, "segmapfreelists", prop) < 0 || 2873 kobj_getvalue(prop, &lvalue) == -1) 2874 segmapfreelists = 0; /* use segmap driver default */ 2875 else 2876 segmapfreelists = (int)lvalue; 2877 2878 /* physmem used to be here, but moved much earlier to fakebop.c */ 2879 } 2880 2881 /* 2882 * Add to a memory list. 2883 * start = start of new memory segment 2884 * len = length of new memory segment in bytes 2885 * new = pointer to a new struct memlist 2886 * memlistp = memory list to which to add segment. 2887 */ 2888 void 2889 memlist_add( 2890 uint64_t start, 2891 uint64_t len, 2892 struct memlist *new, 2893 struct memlist **memlistp) 2894 { 2895 struct memlist *cur; 2896 uint64_t end = start + len; 2897 2898 new->ml_address = start; 2899 new->ml_size = len; 2900 2901 cur = *memlistp; 2902 2903 while (cur) { 2904 if (cur->ml_address >= end) { 2905 new->ml_next = cur; 2906 *memlistp = new; 2907 new->ml_prev = cur->ml_prev; 2908 cur->ml_prev = new; 2909 return; 2910 } 2911 ASSERT(cur->ml_address + cur->ml_size <= start); 2912 if (cur->ml_next == NULL) { 2913 cur->ml_next = new; 2914 new->ml_prev = cur; 2915 new->ml_next = NULL; 2916 return; 2917 } 2918 memlistp = &cur->ml_next; 2919 cur = cur->ml_next; 2920 } 2921 } 2922 2923 void 2924 kobj_vmem_init(vmem_t **text_arena, vmem_t **data_arena) 2925 { 2926 size_t tsize = e_modtext - modtext; 2927 size_t dsize = e_moddata - moddata; 2928 2929 *text_arena = vmem_create("module_text", tsize ? modtext : NULL, tsize, 2930 1, segkmem_alloc, segkmem_free, heaptext_arena, 0, VM_SLEEP); 2931 *data_arena = vmem_create("module_data", dsize ? moddata : NULL, dsize, 2932 1, segkmem_alloc, segkmem_free, heap32_arena, 0, VM_SLEEP); 2933 } 2934 2935 caddr_t 2936 kobj_text_alloc(vmem_t *arena, size_t size) 2937 { 2938 return (vmem_alloc(arena, size, VM_SLEEP | VM_BESTFIT)); 2939 } 2940 2941 /*ARGSUSED*/ 2942 caddr_t 2943 kobj_texthole_alloc(caddr_t addr, size_t size) 2944 { 2945 panic("unexpected call to kobj_texthole_alloc()"); 2946 /*NOTREACHED*/ 2947 return (0); 2948 } 2949 2950 /*ARGSUSED*/ 2951 void 2952 kobj_texthole_free(caddr_t addr, size_t size) 2953 { 2954 panic("unexpected call to kobj_texthole_free()"); 2955 } 2956 2957 /* 2958 * This is called just after configure() in startup(). 2959 * 2960 * The ISALIST concept is a bit hopeless on Intel, because 2961 * there's no guarantee of an ever-more-capable processor 2962 * given that various parts of the instruction set may appear 2963 * and disappear between different implementations. 2964 * 2965 * While it would be possible to correct it and even enhance 2966 * it somewhat, the explicit hardware capability bitmask allows 2967 * more flexibility. 2968 * 2969 * So, we just leave this alone. 2970 */ 2971 void 2972 setx86isalist(void) 2973 { 2974 char *tp; 2975 size_t len; 2976 extern char *isa_list; 2977 2978 #define TBUFSIZE 1024 2979 2980 tp = kmem_alloc(TBUFSIZE, KM_SLEEP); 2981 *tp = '\0'; 2982 2983 #if defined(__amd64) 2984 (void) strcpy(tp, "amd64 "); 2985 #endif 2986 2987 switch (x86_vendor) { 2988 case X86_VENDOR_Intel: 2989 case X86_VENDOR_AMD: 2990 case X86_VENDOR_TM: 2991 if (x86_feature & X86_CMOV) { 2992 /* 2993 * Pentium Pro or later 2994 */ 2995 (void) strcat(tp, "pentium_pro"); 2996 (void) strcat(tp, x86_feature & X86_MMX ? 2997 "+mmx pentium_pro " : " "); 2998 } 2999 /*FALLTHROUGH*/ 3000 case X86_VENDOR_Cyrix: 3001 /* 3002 * The Cyrix 6x86 does not have any Pentium features 3003 * accessible while not at privilege level 0. 3004 */ 3005 if (x86_feature & X86_CPUID) { 3006 (void) strcat(tp, "pentium"); 3007 (void) strcat(tp, x86_feature & X86_MMX ? 3008 "+mmx pentium " : " "); 3009 } 3010 break; 3011 default: 3012 break; 3013 } 3014 (void) strcat(tp, "i486 i386 i86"); 3015 len = strlen(tp) + 1; /* account for NULL at end of string */ 3016 isa_list = strcpy(kmem_alloc(len, KM_SLEEP), tp); 3017 kmem_free(tp, TBUFSIZE); 3018 3019 #undef TBUFSIZE 3020 } 3021 3022 3023 #ifdef __amd64 3024 3025 void * 3026 device_arena_alloc(size_t size, int vm_flag) 3027 { 3028 return (vmem_alloc(device_arena, size, vm_flag)); 3029 } 3030 3031 void 3032 device_arena_free(void *vaddr, size_t size) 3033 { 3034 vmem_free(device_arena, vaddr, size); 3035 } 3036 3037 #else /* __i386 */ 3038 3039 void * 3040 device_arena_alloc(size_t size, int vm_flag) 3041 { 3042 caddr_t vaddr; 3043 uintptr_t v; 3044 size_t start; 3045 size_t end; 3046 3047 vaddr = vmem_alloc(heap_arena, size, vm_flag); 3048 if (vaddr == NULL) 3049 return (NULL); 3050 3051 v = (uintptr_t)vaddr; 3052 ASSERT(v >= kernelbase); 3053 ASSERT(v + size <= valloc_base); 3054 3055 start = btop(v - kernelbase); 3056 end = btop(v + size - 1 - kernelbase); 3057 ASSERT(start < toxic_bit_map_len); 3058 ASSERT(end < toxic_bit_map_len); 3059 3060 while (start <= end) { 3061 BT_ATOMIC_SET(toxic_bit_map, start); 3062 ++start; 3063 } 3064 return (vaddr); 3065 } 3066 3067 void 3068 device_arena_free(void *vaddr, size_t size) 3069 { 3070 uintptr_t v = (uintptr_t)vaddr; 3071 size_t start; 3072 size_t end; 3073 3074 ASSERT(v >= kernelbase); 3075 ASSERT(v + size <= valloc_base); 3076 3077 start = btop(v - kernelbase); 3078 end = btop(v + size - 1 - kernelbase); 3079 ASSERT(start < toxic_bit_map_len); 3080 ASSERT(end < toxic_bit_map_len); 3081 3082 while (start <= end) { 3083 ASSERT(BT_TEST(toxic_bit_map, start) != 0); 3084 BT_ATOMIC_CLEAR(toxic_bit_map, start); 3085 ++start; 3086 } 3087 vmem_free(heap_arena, vaddr, size); 3088 } 3089 3090 /* 3091 * returns 1st address in range that is in device arena, or NULL 3092 * if len is not NULL it returns the length of the toxic range 3093 */ 3094 void * 3095 device_arena_contains(void *vaddr, size_t size, size_t *len) 3096 { 3097 uintptr_t v = (uintptr_t)vaddr; 3098 uintptr_t eaddr = v + size; 3099 size_t start; 3100 size_t end; 3101 3102 /* 3103 * if called very early by kmdb, just return NULL 3104 */ 3105 if (toxic_bit_map == NULL) 3106 return (NULL); 3107 3108 /* 3109 * First check if we're completely outside the bitmap range. 3110 */ 3111 if (v >= valloc_base || eaddr < kernelbase) 3112 return (NULL); 3113 3114 /* 3115 * Trim ends of search to look at only what the bitmap covers. 3116 */ 3117 if (v < kernelbase) 3118 v = kernelbase; 3119 start = btop(v - kernelbase); 3120 end = btop(eaddr - kernelbase); 3121 if (end >= toxic_bit_map_len) 3122 end = toxic_bit_map_len; 3123 3124 if (bt_range(toxic_bit_map, &start, &end, end) == 0) 3125 return (NULL); 3126 3127 v = kernelbase + ptob(start); 3128 if (len != NULL) 3129 *len = ptob(end - start); 3130 return ((void *)v); 3131 } 3132 3133 #endif /* __i386 */ 3134