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 2006 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 24 */ 25 26 #pragma ident "%Z%%M% %I% %E% SMI" 27 28 #include <sys/types.h> 29 #include <sys/t_lock.h> 30 #include <sys/param.h> 31 #include <sys/sysmacros.h> 32 #include <sys/signal.h> 33 #include <sys/systm.h> 34 #include <sys/user.h> 35 #include <sys/mman.h> 36 #include <sys/vm.h> 37 #include <sys/conf.h> 38 #include <sys/avintr.h> 39 #include <sys/autoconf.h> 40 #include <sys/disp.h> 41 #include <sys/class.h> 42 #include <sys/bitmap.h> 43 44 #include <sys/privregs.h> 45 46 #include <sys/proc.h> 47 #include <sys/buf.h> 48 #include <sys/kmem.h> 49 #include <sys/kstat.h> 50 51 #include <sys/reboot.h> 52 #include <sys/uadmin.h> 53 54 #include <sys/cred.h> 55 #include <sys/vnode.h> 56 #include <sys/file.h> 57 58 #include <sys/procfs.h> 59 #include <sys/acct.h> 60 61 #include <sys/vfs.h> 62 #include <sys/dnlc.h> 63 #include <sys/var.h> 64 #include <sys/cmn_err.h> 65 #include <sys/utsname.h> 66 #include <sys/debug.h> 67 #include <sys/kdi.h> 68 69 #include <sys/dumphdr.h> 70 #include <sys/bootconf.h> 71 #include <sys/varargs.h> 72 #include <sys/promif.h> 73 #include <sys/prom_emul.h> /* for create_prom_prop */ 74 #include <sys/modctl.h> /* for "procfs" hack */ 75 76 #include <sys/consdev.h> 77 #include <sys/frame.h> 78 79 #include <sys/sunddi.h> 80 #include <sys/sunndi.h> 81 #include <sys/ndi_impldefs.h> 82 #include <sys/ddidmareq.h> 83 #include <sys/psw.h> 84 #include <sys/regset.h> 85 #include <sys/clock.h> 86 #include <sys/pte.h> 87 #include <sys/mmu.h> 88 #include <sys/tss.h> 89 #include <sys/stack.h> 90 #include <sys/trap.h> 91 #include <sys/pic.h> 92 #include <sys/fp.h> 93 #include <vm/anon.h> 94 #include <vm/as.h> 95 #include <vm/page.h> 96 #include <vm/seg.h> 97 #include <vm/seg_dev.h> 98 #include <vm/seg_kmem.h> 99 #include <vm/seg_kpm.h> 100 #include <vm/seg_map.h> 101 #include <vm/seg_vn.h> 102 #include <vm/seg_kp.h> 103 #include <sys/memnode.h> 104 #include <vm/vm_dep.h> 105 #include <sys/swap.h> 106 #include <sys/thread.h> 107 #include <sys/sysconf.h> 108 #include <sys/vm_machparam.h> 109 #include <sys/archsystm.h> 110 #include <sys/machsystm.h> 111 #include <vm/hat.h> 112 #include <vm/hat_i86.h> 113 #include <sys/pmem.h> 114 #include <sys/instance.h> 115 #include <sys/smp_impldefs.h> 116 #include <sys/x86_archext.h> 117 #include <sys/segments.h> 118 #include <sys/clconf.h> 119 #include <sys/kobj.h> 120 #include <sys/kobj_lex.h> 121 #include <sys/prom_emul.h> 122 #include <sys/cpc_impl.h> 123 #include <sys/chip.h> 124 #include <sys/x86_archext.h> 125 #include <sys/cpu_module.h> 126 #include <sys/smbios.h> 127 128 extern void progressbar_init(void); 129 extern void progressbar_start(void); 130 131 /* 132 * XXX make declaration below "static" when drivers no longer use this 133 * interface. 134 */ 135 extern caddr_t p0_va; /* Virtual address for accessing physical page 0 */ 136 137 /* 138 * segkp 139 */ 140 extern int segkp_fromheap; 141 142 static void kvm_init(void); 143 static void startup_init(void); 144 static void startup_memlist(void); 145 static void startup_modules(void); 146 static void startup_bop_gone(void); 147 static void startup_vm(void); 148 static void startup_end(void); 149 150 /* 151 * Declare these as initialized data so we can patch them. 152 */ 153 #ifdef __i386 154 /* 155 * Due to virtual address space limitations running in 32 bit mode, restrict 156 * the amount of physical memory configured to a max of PHYSMEM32 pages (16g). 157 * 158 * If the physical max memory size of 64g were allowed to be configured, the 159 * size of user virtual address space will be less than 1g. A limited user 160 * address space greatly reduces the range of applications that can run. 161 * 162 * If more physical memory than PHYSMEM32 is required, users should preferably 163 * run in 64 bit mode which has no virtual address space limitation issues. 164 * 165 * If 64 bit mode is not available (as in IA32) and/or more physical memory 166 * than PHYSMEM32 is required in 32 bit mode, physmem can be set to the desired 167 * value or to 0 (to configure all available memory) via eeprom(1M). kernelbase 168 * should also be carefully tuned to balance out the need of the user 169 * application while minimizing the risk of kernel heap exhaustion due to 170 * kernelbase being set too high. 171 */ 172 #define PHYSMEM32 0x400000 173 174 pgcnt_t physmem = PHYSMEM32; 175 #else 176 pgcnt_t physmem = 0; /* memory size in pages, patch if you want less */ 177 #endif 178 pgcnt_t obp_pages; /* Memory used by PROM for its text and data */ 179 180 char *kobj_file_buf; 181 int kobj_file_bufsize; /* set in /etc/system */ 182 183 /* Global variables for MP support. Used in mp_startup */ 184 caddr_t rm_platter_va; 185 uint32_t rm_platter_pa; 186 187 int auto_lpg_disable = 1; 188 189 /* 190 * Some CPUs have holes in the middle of the 64-bit virtual address range. 191 */ 192 uintptr_t hole_start, hole_end; 193 194 /* 195 * kpm mapping window 196 */ 197 caddr_t kpm_vbase; 198 size_t kpm_size; 199 static int kpm_desired = 0; /* Do we want to try to use segkpm? */ 200 201 /* 202 * VA range that must be preserved for boot until we release all of its 203 * mappings. 204 */ 205 #if defined(__amd64) 206 static void *kmem_setaside; 207 #endif 208 209 /* 210 * Configuration parameters set at boot time. 211 */ 212 213 caddr_t econtig; /* end of first block of contiguous kernel */ 214 215 struct bootops *bootops = 0; /* passed in from boot */ 216 struct bootops **bootopsp; 217 struct boot_syscalls *sysp; /* passed in from boot */ 218 219 char bootblock_fstype[16]; 220 221 char kern_bootargs[OBP_MAXPATHLEN]; 222 223 /* 224 * new memory fragmentations are possible in startup() due to BOP_ALLOCs. this 225 * depends on number of BOP_ALLOC calls made and requested size, memory size 226 * combination and whether boot.bin memory needs to be freed. 227 */ 228 #define POSS_NEW_FRAGMENTS 12 229 230 /* 231 * VM data structures 232 */ 233 long page_hashsz; /* Size of page hash table (power of two) */ 234 struct page *pp_base; /* Base of initial system page struct array */ 235 struct page **page_hash; /* Page hash table */ 236 struct seg ktextseg; /* Segment used for kernel executable image */ 237 struct seg kvalloc; /* Segment used for "valloc" mapping */ 238 struct seg kpseg; /* Segment used for pageable kernel virt mem */ 239 struct seg kmapseg; /* Segment used for generic kernel mappings */ 240 struct seg kdebugseg; /* Segment used for the kernel debugger */ 241 242 struct seg *segkmap = &kmapseg; /* Kernel generic mapping segment */ 243 struct seg *segkp = &kpseg; /* Pageable kernel virtual memory segment */ 244 245 #if defined(__amd64) 246 struct seg kvseg_core; /* Segment used for the core heap */ 247 struct seg kpmseg; /* Segment used for physical mapping */ 248 struct seg *segkpm = &kpmseg; /* 64bit kernel physical mapping segment */ 249 #else 250 struct seg *segkpm = NULL; /* Unused on IA32 */ 251 #endif 252 253 caddr_t segkp_base; /* Base address of segkp */ 254 #if defined(__amd64) 255 pgcnt_t segkpsize = btop(SEGKPDEFSIZE); /* size of segkp segment in pages */ 256 #else 257 pgcnt_t segkpsize = 0; 258 #endif 259 260 struct memseg *memseg_base; 261 struct vnode unused_pages_vp; 262 263 #define FOURGB 0x100000000LL 264 265 struct memlist *memlist; 266 267 caddr_t s_text; /* start of kernel text segment */ 268 caddr_t e_text; /* end of kernel text segment */ 269 caddr_t s_data; /* start of kernel data segment */ 270 caddr_t e_data; /* end of kernel data segment */ 271 caddr_t modtext; /* start of loadable module text reserved */ 272 caddr_t e_modtext; /* end of loadable module text reserved */ 273 caddr_t moddata; /* start of loadable module data reserved */ 274 caddr_t e_moddata; /* end of loadable module data reserved */ 275 276 struct memlist *phys_install; /* Total installed physical memory */ 277 struct memlist *phys_avail; /* Total available physical memory */ 278 279 static void memlist_add(uint64_t, uint64_t, struct memlist *, 280 struct memlist **); 281 282 /* 283 * kphysm_init returns the number of pages that were processed 284 */ 285 static pgcnt_t kphysm_init(page_t *, struct memseg *, pgcnt_t, pgcnt_t); 286 287 #define IO_PROP_SIZE 64 /* device property size */ 288 289 /* 290 * a couple useful roundup macros 291 */ 292 #define ROUND_UP_PAGE(x) \ 293 ((uintptr_t)P2ROUNDUP((uintptr_t)(x), (uintptr_t)MMU_PAGESIZE)) 294 #define ROUND_UP_LPAGE(x) \ 295 ((uintptr_t)P2ROUNDUP((uintptr_t)(x), mmu.level_size[1])) 296 #define ROUND_UP_4MEG(x) \ 297 ((uintptr_t)P2ROUNDUP((uintptr_t)(x), (uintptr_t)FOURMB_PAGESIZE)) 298 #define ROUND_UP_TOPLEVEL(x) \ 299 ((uintptr_t)P2ROUNDUP((uintptr_t)(x), mmu.level_size[mmu.max_level])) 300 301 /* 302 * 32-bit Kernel's Virtual memory layout. 303 * +-----------------------+ 304 * | psm 1-1 map | 305 * | exec args area | 306 * 0xFFC00000 -|-----------------------|- ARGSBASE 307 * | debugger | 308 * 0xFF800000 -|-----------------------|- SEGDEBUGBASE 309 * | Kernel Data | 310 * 0xFEC00000 -|-----------------------| 311 * | Kernel Text | 312 * 0xFE800000 -|-----------------------|- KERNEL_TEXT 313 * | LUFS sinkhole | 314 * 0xFE000000 -|-----------------------|- lufs_addr 315 * --- -|-----------------------|- valloc_base + valloc_sz 316 * | early pp structures | 317 * | memsegs, memlists, | 318 * | page hash, etc. | 319 * --- -|-----------------------|- valloc_base (floating) 320 * | ptable_va | 321 * 0xFDFFE000 -|-----------------------|- ekernelheap, ptable_va 322 * | | (segkp is an arena under the heap) 323 * | | 324 * | kvseg | 325 * | | 326 * | | 327 * --- -|-----------------------|- kernelheap (floating) 328 * | Segkmap | 329 * 0xC3002000 -|-----------------------|- segkmap_start (floating) 330 * | Red Zone | 331 * 0xC3000000 -|-----------------------|- kernelbase / userlimit (floating) 332 * | | || 333 * | Shared objects | \/ 334 * | | 335 * : : 336 * | user data | 337 * |-----------------------| 338 * | user text | 339 * 0x08048000 -|-----------------------| 340 * | user stack | 341 * : : 342 * | invalid | 343 * 0x00000000 +-----------------------+ 344 * 345 * 346 * 64-bit Kernel's Virtual memory layout. (assuming 64 bit app) 347 * +-----------------------+ 348 * | psm 1-1 map | 349 * | exec args area | 350 * 0xFFFFFFFF.FFC00000 |-----------------------|- ARGSBASE 351 * | debugger (?) | 352 * 0xFFFFFFFF.FF800000 |-----------------------|- SEGDEBUGBASE 353 * | unused | 354 * +-----------------------+ 355 * | Kernel Data | 356 * 0xFFFFFFFF.FBC00000 |-----------------------| 357 * | Kernel Text | 358 * 0xFFFFFFFF.FB800000 |-----------------------|- KERNEL_TEXT 359 * | LUFS sinkhole | 360 * 0xFFFFFFFF.FB000000 -|-----------------------|- lufs_addr 361 * --- |-----------------------|- valloc_base + valloc_sz 362 * | early pp structures | 363 * | memsegs, memlists, | 364 * | page hash, etc. | 365 * --- |-----------------------|- valloc_base 366 * | ptable_va | 367 * --- |-----------------------|- ptable_va 368 * | Core heap | (used for loadable modules) 369 * 0xFFFFFFFF.C0000000 |-----------------------|- core_base / ekernelheap 370 * | Kernel | 371 * | heap | 372 * 0xFFFFFXXX.XXX00000 |-----------------------|- kernelheap (floating) 373 * | segkmap | 374 * 0xFFFFFXXX.XXX00000 |-----------------------|- segkmap_start (floating) 375 * | device mappings | 376 * 0xFFFFFXXX.XXX00000 |-----------------------|- toxic_addr (floating) 377 * | segkp | 378 * --- |-----------------------|- segkp_base 379 * | segkpm | 380 * 0xFFFFFE00.00000000 |-----------------------| 381 * | Red Zone | 382 * 0xFFFFFD80.00000000 |-----------------------|- KERNELBASE 383 * | User stack |- User space memory 384 * | | 385 * | shared objects, etc | (grows downwards) 386 * : : 387 * | | 388 * 0xFFFF8000.00000000 |-----------------------| 389 * | | 390 * | VA Hole / unused | 391 * | | 392 * 0x00008000.00000000 |-----------------------| 393 * | | 394 * | | 395 * : : 396 * | user heap | (grows upwards) 397 * | | 398 * | user data | 399 * |-----------------------| 400 * | user text | 401 * 0x00000000.04000000 |-----------------------| 402 * | invalid | 403 * 0x00000000.00000000 +-----------------------+ 404 * 405 * A 32 bit app on the 64 bit kernel sees the same layout as on the 32 bit 406 * kernel, except that userlimit is raised to 0xfe000000 407 * 408 * Floating values: 409 * 410 * valloc_base: start of the kernel's memory management/tracking data 411 * structures. This region contains page_t structures for the lowest 4GB 412 * of physical memory, memsegs, memlists, and the page hash. 413 * 414 * core_base: start of the kernel's "core" heap area on 64-bit systems. 415 * This area is intended to be used for global data as well as for module 416 * text/data that does not fit into the nucleus pages. The core heap is 417 * restricted to a 2GB range, allowing every address within it to be 418 * accessed using rip-relative addressing 419 * 420 * ekernelheap: end of kernelheap and start of segmap. 421 * 422 * kernelheap: start of kernel heap. On 32-bit systems, this starts right 423 * above a red zone that separates the user's address space from the 424 * kernel's. On 64-bit systems, it sits above segkp and segkpm. 425 * 426 * segkmap_start: start of segmap. The length of segmap can be modified 427 * by changing segmapsize in /etc/system (preferred) or eeprom (deprecated). 428 * The default length is 16MB on 32-bit systems and 64MB on 64-bit systems. 429 * 430 * kernelbase: On a 32-bit kernel the default value of 0xd4000000 will be 431 * decreased by 2X the size required for page_t. This allows the kernel 432 * heap to grow in size with physical memory. With sizeof(page_t) == 80 433 * bytes, the following shows the values of kernelbase and kernel heap 434 * sizes for different memory configurations (assuming default segmap and 435 * segkp sizes). 436 * 437 * mem size for kernelbase kernel heap 438 * size page_t's size 439 * ---- --------- ---------- ----------- 440 * 1gb 0x01400000 0xd1800000 684MB 441 * 2gb 0x02800000 0xcf000000 704MB 442 * 4gb 0x05000000 0xca000000 744MB 443 * 6gb 0x07800000 0xc5000000 784MB 444 * 8gb 0x0a000000 0xc0000000 824MB 445 * 16gb 0x14000000 0xac000000 984MB 446 * 32gb 0x28000000 0x84000000 1304MB 447 * 64gb 0x50000000 0x34000000 1944MB (*) 448 * 449 * kernelbase is less than the abi minimum of 0xc0000000 for memory 450 * configurations above 8gb. 451 * 452 * (*) support for memory configurations above 32gb will require manual tuning 453 * of kernelbase to balance out the need of user applications. 454 */ 455 456 /* real-time-clock initialization parameters */ 457 long gmt_lag; /* offset in seconds of gmt to local time */ 458 extern long process_rtc_config_file(void); 459 460 char *final_kernelheap; 461 char *boot_kernelheap; 462 uintptr_t kernelbase; 463 uintptr_t eprom_kernelbase; 464 size_t segmapsize; 465 static uintptr_t segmap_reserved; 466 uintptr_t segkmap_start; 467 int segmapfreelists; 468 pgcnt_t boot_npages; 469 pgcnt_t npages; 470 size_t core_size; /* size of "core" heap */ 471 uintptr_t core_base; /* base address of "core" heap */ 472 473 /* 474 * List of bootstrap pages. We mark these as allocated in startup. 475 * release_bootstrap() will free them when we're completely done with 476 * the bootstrap. 477 */ 478 static page_t *bootpages, *rd_pages; 479 480 struct system_hardware system_hardware; 481 482 /* 483 * Enable some debugging messages concerning memory usage... 484 * 485 * XX64 There should only be one print routine once memlist usage between 486 * vmx and the kernel is cleaned up and there is a single memlist structure 487 * shared between kernel and boot. 488 */ 489 static void 490 print_boot_memlist(char *title, struct memlist *mp) 491 { 492 prom_printf("MEMLIST: %s:\n", title); 493 while (mp != NULL) { 494 prom_printf("\tAddress 0x%" PRIx64 ", size 0x%" PRIx64 "\n", 495 mp->address, mp->size); 496 mp = mp->next; 497 } 498 } 499 500 static void 501 print_kernel_memlist(char *title, struct memlist *mp) 502 { 503 prom_printf("MEMLIST: %s:\n", title); 504 while (mp != NULL) { 505 prom_printf("\tAddress 0x%" PRIx64 ", size 0x%" PRIx64 "\n", 506 mp->address, mp->size); 507 mp = mp->next; 508 } 509 } 510 511 /* 512 * XX64 need a comment here.. are these just default values, surely 513 * we read the "cpuid" type information to figure this out. 514 */ 515 int l2cache_sz = 0x80000; 516 int l2cache_linesz = 0x40; 517 int l2cache_assoc = 1; 518 519 /* 520 * on 64 bit we use a predifined VA range for mapping devices in the kernel 521 * on 32 bit the mappings are intermixed in the heap, so we use a bit map 522 */ 523 #ifdef __amd64 524 525 vmem_t *device_arena; 526 uintptr_t toxic_addr = (uintptr_t)NULL; 527 size_t toxic_size = 1 * 1024 * 1024 * 1024; /* Sparc uses 1 gig too */ 528 529 #else /* __i386 */ 530 531 ulong_t *toxic_bit_map; /* one bit for each 4k of VA in heap_arena */ 532 size_t toxic_bit_map_len = 0; /* in bits */ 533 534 #endif /* __i386 */ 535 536 /* 537 * Simple boot time debug facilities 538 */ 539 static char *prm_dbg_str[] = { 540 "%s:%d: '%s' is 0x%x\n", 541 "%s:%d: '%s' is 0x%llx\n" 542 }; 543 544 int prom_debug; 545 546 #define PRM_DEBUG(q) if (prom_debug) \ 547 prom_printf(prm_dbg_str[sizeof (q) >> 3], "startup.c", __LINE__, #q, q); 548 #define PRM_POINT(q) if (prom_debug) \ 549 prom_printf("%s:%d: %s\n", "startup.c", __LINE__, q); 550 551 /* 552 * This structure is used to keep track of the intial allocations 553 * done in startup_memlist(). The value of NUM_ALLOCATIONS needs to 554 * be >= the number of ADD_TO_ALLOCATIONS() executed in the code. 555 */ 556 #define NUM_ALLOCATIONS 7 557 int num_allocations = 0; 558 struct { 559 void **al_ptr; 560 size_t al_size; 561 } allocations[NUM_ALLOCATIONS]; 562 size_t valloc_sz = 0; 563 uintptr_t valloc_base; 564 extern uintptr_t ptable_va; 565 extern size_t ptable_sz; 566 567 #define ADD_TO_ALLOCATIONS(ptr, size) { \ 568 size = ROUND_UP_PAGE(size); \ 569 if (num_allocations == NUM_ALLOCATIONS) \ 570 panic("too many ADD_TO_ALLOCATIONS()"); \ 571 allocations[num_allocations].al_ptr = (void**)&ptr; \ 572 allocations[num_allocations].al_size = size; \ 573 valloc_sz += size; \ 574 ++num_allocations; \ 575 } 576 577 static void 578 perform_allocations(void) 579 { 580 caddr_t mem; 581 int i; 582 583 mem = BOP_ALLOC(bootops, (caddr_t)valloc_base, valloc_sz, BO_NO_ALIGN); 584 if (mem != (caddr_t)valloc_base) 585 panic("BOP_ALLOC() failed"); 586 bzero(mem, valloc_sz); 587 for (i = 0; i < num_allocations; ++i) { 588 *allocations[i].al_ptr = (void *)mem; 589 mem += allocations[i].al_size; 590 } 591 } 592 593 /* 594 * Our world looks like this at startup time. 595 * 596 * In a 32-bit OS, boot loads the kernel text at 0xfe800000 and kernel data 597 * at 0xfec00000. On a 64-bit OS, kernel text and data are loaded at 598 * 0xffffffff.fe800000 and 0xffffffff.fec00000 respectively. Those 599 * addresses are fixed in the binary at link time. 600 * 601 * On the text page: 602 * unix/genunix/krtld/module text loads. 603 * 604 * On the data page: 605 * unix/genunix/krtld/module data loads and space for page_t's. 606 */ 607 /* 608 * Machine-dependent startup code 609 */ 610 void 611 startup(void) 612 { 613 extern void startup_bios_disk(void); 614 extern void startup_pci_bios(void); 615 /* 616 * Make sure that nobody tries to use sekpm until we have 617 * initialized it properly. 618 */ 619 #if defined(__amd64) 620 kpm_desired = kpm_enable; 621 #endif 622 kpm_enable = 0; 623 624 progressbar_init(); 625 startup_init(); 626 startup_memlist(); 627 startup_pci_bios(); 628 startup_modules(); 629 startup_bios_disk(); 630 startup_bop_gone(); 631 startup_vm(); 632 startup_end(); 633 progressbar_start(); 634 } 635 636 static void 637 startup_init() 638 { 639 PRM_POINT("startup_init() starting..."); 640 641 /* 642 * Complete the extraction of cpuid data 643 */ 644 cpuid_pass2(CPU); 645 646 (void) check_boot_version(BOP_GETVERSION(bootops)); 647 648 /* 649 * Check for prom_debug in boot environment 650 */ 651 if (BOP_GETPROPLEN(bootops, "prom_debug") >= 0) { 652 ++prom_debug; 653 PRM_POINT("prom_debug found in boot enviroment"); 654 } 655 656 /* 657 * Collect node, cpu and memory configuration information. 658 */ 659 get_system_configuration(); 660 661 /* 662 * Halt if this is an unsupported processor. 663 */ 664 if (x86_type == X86_TYPE_486 || x86_type == X86_TYPE_CYRIX_486) { 665 printf("\n486 processor (\"%s\") detected.\n", 666 CPU->cpu_brandstr); 667 halt("This processor is not supported by this release " 668 "of Solaris."); 669 } 670 671 PRM_POINT("startup_init() done"); 672 } 673 674 /* 675 * Callback for copy_memlist_filter() to filter nucleus, kadb/kmdb, (ie. 676 * everything mapped above KERNEL_TEXT) pages from phys_avail. Note it 677 * also filters out physical page zero. There is some reliance on the 678 * boot loader allocating only a few contiguous physical memory chunks. 679 */ 680 static void 681 avail_filter(uint64_t *addr, uint64_t *size) 682 { 683 uintptr_t va; 684 uintptr_t next_va; 685 pfn_t pfn; 686 uint64_t pfn_addr; 687 uint64_t pfn_eaddr; 688 uint_t prot; 689 size_t len; 690 uint_t change; 691 692 if (prom_debug) 693 prom_printf("\tFilter: in: a=%" PRIx64 ", s=%" PRIx64 "\n", 694 *addr, *size); 695 696 /* 697 * page zero is required for BIOS.. never make it available 698 */ 699 if (*addr == 0) { 700 *addr += MMU_PAGESIZE; 701 *size -= MMU_PAGESIZE; 702 } 703 704 /* 705 * First we trim from the front of the range. Since hat_boot_probe() 706 * walks ranges in virtual order, but addr/size are physical, we need 707 * to the list until no changes are seen. This deals with the case 708 * where page "p" is mapped at v, page "p + PAGESIZE" is mapped at w 709 * but w < v. 710 */ 711 do { 712 change = 0; 713 for (va = KERNEL_TEXT; 714 *size > 0 && hat_boot_probe(&va, &len, &pfn, &prot) != 0; 715 va = next_va) { 716 717 next_va = va + len; 718 pfn_addr = ptob((uint64_t)pfn); 719 pfn_eaddr = pfn_addr + len; 720 721 if (pfn_addr <= *addr && pfn_eaddr > *addr) { 722 change = 1; 723 while (*size > 0 && len > 0) { 724 *addr += MMU_PAGESIZE; 725 *size -= MMU_PAGESIZE; 726 len -= MMU_PAGESIZE; 727 } 728 } 729 } 730 if (change && prom_debug) 731 prom_printf("\t\ttrim: a=%" PRIx64 ", s=%" PRIx64 "\n", 732 *addr, *size); 733 } while (change); 734 735 /* 736 * Trim pages from the end of the range. 737 */ 738 for (va = KERNEL_TEXT; 739 *size > 0 && hat_boot_probe(&va, &len, &pfn, &prot) != 0; 740 va = next_va) { 741 742 next_va = va + len; 743 pfn_addr = ptob((uint64_t)pfn); 744 745 if (pfn_addr >= *addr && pfn_addr < *addr + *size) 746 *size = pfn_addr - *addr; 747 } 748 749 if (prom_debug) 750 prom_printf("\tFilter out: a=%" PRIx64 ", s=%" PRIx64 "\n", 751 *addr, *size); 752 } 753 754 static void 755 kpm_init() 756 { 757 struct segkpm_crargs b; 758 uintptr_t start, end; 759 struct memlist *pmem; 760 761 /* 762 * These variables were all designed for sfmmu in which segkpm is 763 * mapped using a single pagesize - either 8KB or 4MB. On x86, we 764 * might use 2+ page sizes on a single machine, so none of these 765 * variables have a single correct value. They are set up as if we 766 * always use a 4KB pagesize, which should do no harm. In the long 767 * run, we should get rid of KPM's assumption that only a single 768 * pagesize is used. 769 */ 770 kpm_pgshft = MMU_PAGESHIFT; 771 kpm_pgsz = MMU_PAGESIZE; 772 kpm_pgoff = MMU_PAGEOFFSET; 773 kpmp2pshft = 0; 774 kpmpnpgs = 1; 775 ASSERT(((uintptr_t)kpm_vbase & (kpm_pgsz - 1)) == 0); 776 777 PRM_POINT("about to create segkpm"); 778 rw_enter(&kas.a_lock, RW_WRITER); 779 780 if (seg_attach(&kas, kpm_vbase, kpm_size, segkpm) < 0) 781 panic("cannot attach segkpm"); 782 783 b.prot = PROT_READ | PROT_WRITE; 784 b.nvcolors = 1; 785 786 if (segkpm_create(segkpm, (caddr_t)&b) != 0) 787 panic("segkpm_create segkpm"); 788 789 rw_exit(&kas.a_lock); 790 791 /* 792 * Map each of the memsegs into the kpm segment, coalesing adjacent 793 * memsegs to allow mapping with the largest possible pages. 794 */ 795 pmem = phys_install; 796 start = pmem->address; 797 end = start + pmem->size; 798 for (;;) { 799 if (pmem == NULL || pmem->address > end) { 800 hat_devload(kas.a_hat, kpm_vbase + start, 801 end - start, mmu_btop(start), 802 PROT_READ | PROT_WRITE, 803 HAT_LOAD | HAT_LOAD_LOCK | HAT_LOAD_NOCONSIST); 804 if (pmem == NULL) 805 break; 806 start = pmem->address; 807 } 808 end = pmem->address + pmem->size; 809 pmem = pmem->next; 810 } 811 } 812 813 /* 814 * The purpose of startup memlist is to get the system to the 815 * point where it can use kmem_alloc()'s that operate correctly 816 * relying on BOP_ALLOC(). This includes allocating page_ts, 817 * page hash table, vmem initialized, etc. 818 * 819 * Boot's versions of physinstalled and physavail are insufficient for 820 * the kernel's purposes. Specifically we don't know which pages that 821 * are not in physavail can be reclaimed after boot is gone. 822 * 823 * This code solves the problem by dividing the address space 824 * into 3 regions as it takes over the MMU from the booter. 825 * 826 * 1) Any (non-nucleus) pages that are mapped at addresses above KERNEL_TEXT 827 * can not be used by the kernel. 828 * 829 * 2) Any free page that happens to be mapped below kernelbase 830 * is protected until the boot loader is released, but will then be reclaimed. 831 * 832 * 3) Boot shouldn't use any address in the remaining area between kernelbase 833 * and KERNEL_TEXT. 834 * 835 * In the case of multiple mappings to the same page, region 1 has precedence 836 * over region 2. 837 */ 838 static void 839 startup_memlist(void) 840 { 841 size_t memlist_sz; 842 size_t memseg_sz; 843 size_t pagehash_sz; 844 size_t pp_sz; 845 uintptr_t va; 846 size_t len; 847 uint_t prot; 848 pfn_t pfn; 849 int memblocks; 850 caddr_t pagecolor_mem; 851 size_t pagecolor_memsz; 852 caddr_t page_ctrs_mem; 853 size_t page_ctrs_size; 854 struct memlist *current; 855 pgcnt_t orig_npages = 0; 856 extern void startup_build_mem_nodes(struct memlist *); 857 858 /* XX64 fix these - they should be in include files */ 859 extern ulong_t cr4_value; 860 extern size_t page_coloring_init(uint_t, int, int); 861 extern void page_coloring_setup(caddr_t); 862 863 PRM_POINT("startup_memlist() starting..."); 864 865 /* 866 * Take the most current snapshot we can by calling mem-update. 867 * For this to work properly, we first have to ask boot for its 868 * end address. 869 */ 870 if (BOP_GETPROPLEN(bootops, "memory-update") == 0) 871 (void) BOP_GETPROP(bootops, "memory-update", NULL); 872 873 /* 874 * find if the kernel is mapped on a large page 875 */ 876 va = KERNEL_TEXT; 877 if (hat_boot_probe(&va, &len, &pfn, &prot) == 0) 878 panic("Couldn't find kernel text boot mapping"); 879 880 /* 881 * Use leftover large page nucleus text/data space for loadable modules. 882 * Use at most MODTEXT/MODDATA. 883 */ 884 if (len > MMU_PAGESIZE) { 885 886 moddata = (caddr_t)ROUND_UP_PAGE(e_data); 887 e_moddata = (caddr_t)ROUND_UP_4MEG(e_data); 888 if (e_moddata - moddata > MODDATA) 889 e_moddata = moddata + MODDATA; 890 891 modtext = (caddr_t)ROUND_UP_PAGE(e_text); 892 e_modtext = (caddr_t)ROUND_UP_4MEG(e_text); 893 if (e_modtext - modtext > MODTEXT) 894 e_modtext = modtext + MODTEXT; 895 896 897 } else { 898 899 PRM_POINT("Kernel NOT loaded on Large Page!"); 900 e_moddata = moddata = (caddr_t)ROUND_UP_PAGE(e_data); 901 e_modtext = modtext = (caddr_t)ROUND_UP_PAGE(e_text); 902 903 } 904 econtig = e_moddata; 905 906 PRM_DEBUG(modtext); 907 PRM_DEBUG(e_modtext); 908 PRM_DEBUG(moddata); 909 PRM_DEBUG(e_moddata); 910 PRM_DEBUG(econtig); 911 912 /* 913 * For MP machines cr4_value must be set or the non-boot 914 * CPUs will not be able to start. 915 */ 916 if (x86_feature & X86_LARGEPAGE) 917 cr4_value = getcr4(); 918 PRM_DEBUG(cr4_value); 919 920 /* 921 * Examine the boot loaders physical memory map to find out: 922 * - total memory in system - physinstalled 923 * - the max physical address - physmax 924 * - the number of segments the intsalled memory comes in 925 */ 926 if (prom_debug) 927 print_boot_memlist("boot physinstalled", 928 bootops->boot_mem->physinstalled); 929 installed_top_size(bootops->boot_mem->physinstalled, &physmax, 930 &physinstalled, &memblocks); 931 PRM_DEBUG(physmax); 932 PRM_DEBUG(physinstalled); 933 PRM_DEBUG(memblocks); 934 935 if (prom_debug) 936 print_boot_memlist("boot physavail", 937 bootops->boot_mem->physavail); 938 939 /* 940 * Initialize hat's mmu parameters. 941 * Check for enforce-prot-exec in boot environment. It's used to 942 * enable/disable support for the page table entry NX bit. 943 * The default is to enforce PROT_EXEC on processors that support NX. 944 * Boot seems to round up the "len", but 8 seems to be big enough. 945 */ 946 mmu_init(); 947 948 #ifdef __i386 949 /* 950 * physmax is lowered if there is more memory than can be 951 * physically addressed in 32 bit (PAE/non-PAE) modes. 952 */ 953 if (mmu.pae_hat) { 954 if (PFN_ABOVE64G(physmax)) { 955 physinstalled -= (physmax - (PFN_64G - 1)); 956 physmax = PFN_64G - 1; 957 } 958 } else { 959 if (PFN_ABOVE4G(physmax)) { 960 physinstalled -= (physmax - (PFN_4G - 1)); 961 physmax = PFN_4G - 1; 962 } 963 } 964 #endif 965 966 startup_build_mem_nodes(bootops->boot_mem->physinstalled); 967 968 if (BOP_GETPROPLEN(bootops, "enforce-prot-exec") >= 0) { 969 int len = BOP_GETPROPLEN(bootops, "enforce-prot-exec"); 970 char value[8]; 971 972 if (len < 8) 973 (void) BOP_GETPROP(bootops, "enforce-prot-exec", value); 974 else 975 (void) strcpy(value, ""); 976 if (strcmp(value, "off") == 0) 977 mmu.pt_nx = 0; 978 } 979 PRM_DEBUG(mmu.pt_nx); 980 981 /* 982 * We will need page_t's for every page in the system, except for 983 * memory mapped at or above above the start of the kernel text segment. 984 * 985 * pages above e_modtext are attributed to kernel debugger (obp_pages) 986 */ 987 npages = physinstalled - 1; /* avail_filter() skips page 0, so "- 1" */ 988 obp_pages = 0; 989 va = KERNEL_TEXT; 990 while (hat_boot_probe(&va, &len, &pfn, &prot) != 0) { 991 npages -= len >> MMU_PAGESHIFT; 992 if (va >= (uintptr_t)e_moddata) 993 obp_pages += len >> MMU_PAGESHIFT; 994 va += len; 995 } 996 PRM_DEBUG(npages); 997 PRM_DEBUG(obp_pages); 998 999 /* 1000 * If physmem is patched to be non-zero, use it instead of 1001 * the computed value unless it is larger than the real 1002 * amount of memory on hand. 1003 */ 1004 if (physmem == 0 || physmem > npages) { 1005 physmem = npages; 1006 } else if (physmem < npages) { 1007 orig_npages = npages; 1008 npages = physmem; 1009 } 1010 PRM_DEBUG(physmem); 1011 1012 /* 1013 * We now compute the sizes of all the initial allocations for 1014 * structures the kernel needs in order do kmem_alloc(). These 1015 * include: 1016 * memsegs 1017 * memlists 1018 * page hash table 1019 * page_t's 1020 * page coloring data structs 1021 */ 1022 memseg_sz = sizeof (struct memseg) * (memblocks + POSS_NEW_FRAGMENTS); 1023 ADD_TO_ALLOCATIONS(memseg_base, memseg_sz); 1024 PRM_DEBUG(memseg_sz); 1025 1026 /* 1027 * Reserve space for phys_avail/phys_install memlists. 1028 * There's no real good way to know exactly how much room we'll need, 1029 * but this should be a good upper bound. 1030 */ 1031 memlist_sz = ROUND_UP_PAGE(2 * sizeof (struct memlist) * 1032 (memblocks + POSS_NEW_FRAGMENTS)); 1033 ADD_TO_ALLOCATIONS(memlist, memlist_sz); 1034 PRM_DEBUG(memlist_sz); 1035 1036 /* 1037 * The page structure hash table size is a power of 2 1038 * such that the average hash chain length is PAGE_HASHAVELEN. 1039 */ 1040 page_hashsz = npages / PAGE_HASHAVELEN; 1041 page_hashsz = 1 << highbit(page_hashsz); 1042 pagehash_sz = sizeof (struct page *) * page_hashsz; 1043 ADD_TO_ALLOCATIONS(page_hash, pagehash_sz); 1044 PRM_DEBUG(pagehash_sz); 1045 1046 /* 1047 * Set aside room for the page structures themselves. Note: on 1048 * 64-bit systems we don't allocate page_t's for every page here. 1049 * We just allocate enough to map the lowest 4GB of physical 1050 * memory, minus those pages that are used for the "nucleus" kernel 1051 * text and data. The remaining pages are allocated once we can 1052 * map around boot. 1053 * 1054 * boot_npages is used to allocate an area big enough for our 1055 * initial page_t's. kphym_init may use less than that. 1056 */ 1057 boot_npages = npages; 1058 #if defined(__amd64) 1059 if (npages > mmu_btop(FOURGB - (econtig - s_text))) 1060 boot_npages = mmu_btop(FOURGB - (econtig - s_text)); 1061 #endif 1062 PRM_DEBUG(boot_npages); 1063 pp_sz = sizeof (struct page) * boot_npages; 1064 ADD_TO_ALLOCATIONS(pp_base, pp_sz); 1065 PRM_DEBUG(pp_sz); 1066 1067 /* 1068 * determine l2 cache info and memory size for page coloring 1069 */ 1070 (void) getl2cacheinfo(CPU, 1071 &l2cache_sz, &l2cache_linesz, &l2cache_assoc); 1072 pagecolor_memsz = 1073 page_coloring_init(l2cache_sz, l2cache_linesz, l2cache_assoc); 1074 ADD_TO_ALLOCATIONS(pagecolor_mem, pagecolor_memsz); 1075 PRM_DEBUG(pagecolor_memsz); 1076 1077 page_ctrs_size = page_ctrs_sz(); 1078 ADD_TO_ALLOCATIONS(page_ctrs_mem, page_ctrs_size); 1079 PRM_DEBUG(page_ctrs_size); 1080 1081 /* 1082 * valloc_base will be below kernel text 1083 * The extra pages are for the HAT and kmdb to map page tables. 1084 */ 1085 valloc_sz = ROUND_UP_LPAGE(valloc_sz); 1086 valloc_base = KERNEL_TEXT - valloc_sz; 1087 PRM_DEBUG(valloc_base); 1088 ptable_va = valloc_base - ptable_sz; 1089 1090 #if defined(__amd64) 1091 if (eprom_kernelbase && eprom_kernelbase != KERNELBASE) 1092 cmn_err(CE_NOTE, "!kernelbase cannot be changed on 64-bit " 1093 "systems."); 1094 kernelbase = (uintptr_t)KERNELBASE; 1095 core_base = (uintptr_t)COREHEAP_BASE; 1096 core_size = ptable_va - core_base; 1097 #else /* __i386 */ 1098 /* 1099 * We configure kernelbase based on: 1100 * 1101 * 1. user specified kernelbase via eeprom command. Value cannot exceed 1102 * KERNELBASE_MAX. we large page align eprom_kernelbase 1103 * 1104 * 2. Default to KERNELBASE and adjust to 2X less the size for page_t. 1105 * On large memory systems we must lower kernelbase to allow 1106 * enough room for page_t's for all of memory. 1107 * 1108 * The value set here, might be changed a little later. 1109 */ 1110 if (eprom_kernelbase) { 1111 kernelbase = eprom_kernelbase & mmu.level_mask[1]; 1112 if (kernelbase > KERNELBASE_MAX) 1113 kernelbase = KERNELBASE_MAX; 1114 } else { 1115 kernelbase = (uintptr_t)KERNELBASE; 1116 kernelbase -= ROUND_UP_4MEG(2 * valloc_sz); 1117 } 1118 ASSERT((kernelbase & mmu.level_offset[1]) == 0); 1119 core_base = ptable_va; 1120 core_size = 0; 1121 #endif 1122 1123 PRM_DEBUG(kernelbase); 1124 PRM_DEBUG(core_base); 1125 PRM_DEBUG(core_size); 1126 1127 /* 1128 * At this point, we can only use a portion of the kernelheap that 1129 * will be available after we boot. Both 32-bit and 64-bit systems 1130 * have this limitation, although the reasons are completely 1131 * different. 1132 * 1133 * On 64-bit systems, the booter only supports allocations in the 1134 * upper 4GB of memory, so we have to work with a reduced kernel 1135 * heap until we take over all allocations. The booter also sits 1136 * in the lower portion of that 4GB range, so we have to raise the 1137 * bottom of the heap even further. 1138 * 1139 * On 32-bit systems we have to leave room to place segmap below 1140 * the heap. We don't yet know how large segmap will be, so we 1141 * have to be very conservative. 1142 */ 1143 #if defined(__amd64) 1144 /* 1145 * XX64: For now, we let boot have the lower 2GB of the top 4GB 1146 * address range. In the long run, that should be fixed. It's 1147 * insane for a booter to need 2 2GB address ranges. 1148 */ 1149 boot_kernelheap = (caddr_t)(BOOT_DOUBLEMAP_BASE + BOOT_DOUBLEMAP_SIZE); 1150 segmap_reserved = 0; 1151 1152 #else /* __i386 */ 1153 segkp_fromheap = 1; 1154 segmap_reserved = ROUND_UP_LPAGE(MAX(segmapsize, SEGMAPMAX)); 1155 boot_kernelheap = (caddr_t)(ROUND_UP_LPAGE(kernelbase) + 1156 segmap_reserved); 1157 #endif 1158 PRM_DEBUG(boot_kernelheap); 1159 kernelheap = boot_kernelheap; 1160 ekernelheap = (char *)core_base; 1161 1162 /* 1163 * If segmap is too large we can push the bottom of the kernel heap 1164 * higher than the base. Or worse, it could exceed the top of the 1165 * VA space entirely, causing it to wrap around. 1166 */ 1167 if (kernelheap >= ekernelheap || (uintptr_t)kernelheap < kernelbase) 1168 panic("too little memory available for kernelheap," 1169 " use a different kernelbase"); 1170 1171 /* 1172 * Now that we know the real value of kernelbase, 1173 * update variables that were initialized with a value of 1174 * KERNELBASE (in common/conf/param.c). 1175 * 1176 * XXX The problem with this sort of hackery is that the 1177 * compiler just may feel like putting the const declarations 1178 * (in param.c) into the .text section. Perhaps they should 1179 * just be declared as variables there? 1180 */ 1181 1182 #if defined(__amd64) 1183 ASSERT(_kernelbase == KERNELBASE); 1184 ASSERT(_userlimit == USERLIMIT); 1185 /* 1186 * As one final sanity check, verify that the "red zone" between 1187 * kernel and userspace is exactly the size we expected. 1188 */ 1189 ASSERT(_kernelbase == (_userlimit + (2 * 1024 * 1024))); 1190 #else 1191 *(uintptr_t *)&_kernelbase = kernelbase; 1192 *(uintptr_t *)&_userlimit = kernelbase; 1193 *(uintptr_t *)&_userlimit32 = _userlimit; 1194 #endif 1195 PRM_DEBUG(_kernelbase); 1196 PRM_DEBUG(_userlimit); 1197 PRM_DEBUG(_userlimit32); 1198 1199 /* 1200 * do all the initial allocations 1201 */ 1202 perform_allocations(); 1203 1204 /* 1205 * Initialize the kernel heap. Note 3rd argument must be > 1st. 1206 */ 1207 kernelheap_init(kernelheap, ekernelheap, kernelheap + MMU_PAGESIZE, 1208 (void *)core_base, (void *)ptable_va); 1209 1210 /* 1211 * Build phys_install and phys_avail in kernel memspace. 1212 * - phys_install should be all memory in the system. 1213 * - phys_avail is phys_install minus any memory mapped before this 1214 * point above KERNEL_TEXT. 1215 */ 1216 current = phys_install = memlist; 1217 copy_memlist_filter(bootops->boot_mem->physinstalled, ¤t, NULL); 1218 if ((caddr_t)current > (caddr_t)memlist + memlist_sz) 1219 panic("physinstalled was too big!"); 1220 if (prom_debug) 1221 print_kernel_memlist("phys_install", phys_install); 1222 1223 phys_avail = current; 1224 PRM_POINT("Building phys_avail:\n"); 1225 copy_memlist_filter(bootops->boot_mem->physinstalled, ¤t, 1226 avail_filter); 1227 if ((caddr_t)current > (caddr_t)memlist + memlist_sz) 1228 panic("physavail was too big!"); 1229 if (prom_debug) 1230 print_kernel_memlist("phys_avail", phys_avail); 1231 1232 /* 1233 * setup page coloring 1234 */ 1235 page_coloring_setup(pagecolor_mem); 1236 page_lock_init(); /* currently a no-op */ 1237 1238 /* 1239 * free page list counters 1240 */ 1241 (void) page_ctrs_alloc(page_ctrs_mem); 1242 1243 /* 1244 * Initialize the page structures from the memory lists. 1245 */ 1246 availrmem_initial = availrmem = freemem = 0; 1247 PRM_POINT("Calling kphysm_init()..."); 1248 boot_npages = kphysm_init(pp_base, memseg_base, 0, boot_npages); 1249 PRM_POINT("kphysm_init() done"); 1250 PRM_DEBUG(boot_npages); 1251 1252 /* 1253 * Now that page_t's have been initialized, remove all the 1254 * initial allocation pages from the kernel free page lists. 1255 */ 1256 boot_mapin((caddr_t)valloc_base, valloc_sz); 1257 1258 /* 1259 * Initialize kernel memory allocator. 1260 */ 1261 kmem_init(); 1262 1263 /* 1264 * print this out early so that we know what's going on 1265 */ 1266 cmn_err(CE_CONT, "?features: %b\n", x86_feature, FMT_X86_FEATURE); 1267 1268 /* 1269 * Initialize bp_mapin(). 1270 */ 1271 bp_init(MMU_PAGESIZE, HAT_STORECACHING_OK); 1272 1273 /* 1274 * orig_npages is non-zero if physmem has been configured for less 1275 * than the available memory. 1276 */ 1277 if (orig_npages) { 1278 #ifdef __i386 1279 /* 1280 * use npages for physmem in case it has been temporarily 1281 * modified via /etc/system in kmem_init/mod_read_system_file. 1282 */ 1283 if (npages == PHYSMEM32) { 1284 cmn_err(CE_WARN, "!Due to 32 bit virtual" 1285 " address space limitations, limiting" 1286 " physmem to 0x%lx of 0x%lx available pages", 1287 npages, orig_npages); 1288 } else { 1289 cmn_err(CE_WARN, "!limiting physmem to 0x%lx of" 1290 " 0x%lx available pages", npages, orig_npages); 1291 } 1292 #else 1293 cmn_err(CE_WARN, "!limiting physmem to 0x%lx of" 1294 " 0x%lx available pages", npages, orig_npages); 1295 #endif 1296 } 1297 #if defined(__i386) 1298 if (eprom_kernelbase && (eprom_kernelbase != kernelbase)) 1299 cmn_err(CE_WARN, "kernelbase value, User specified 0x%lx, " 1300 "System using 0x%lx", 1301 (uintptr_t)eprom_kernelbase, (uintptr_t)kernelbase); 1302 #endif 1303 1304 #ifdef KERNELBASE_ABI_MIN 1305 if (kernelbase < (uintptr_t)KERNELBASE_ABI_MIN) { 1306 cmn_err(CE_NOTE, "!kernelbase set to 0x%lx, system is not " 1307 "i386 ABI compliant.", (uintptr_t)kernelbase); 1308 } 1309 #endif 1310 1311 PRM_POINT("startup_memlist() done"); 1312 } 1313 1314 static void 1315 startup_modules(void) 1316 { 1317 unsigned int i; 1318 extern void prom_setup(void); 1319 1320 PRM_POINT("startup_modules() starting..."); 1321 /* 1322 * Initialize ten-micro second timer so that drivers will 1323 * not get short changed in their init phase. This was 1324 * not getting called until clkinit which, on fast cpu's 1325 * caused the drv_usecwait to be way too short. 1326 */ 1327 microfind(); 1328 1329 /* 1330 * Read the GMT lag from /etc/rtc_config. 1331 */ 1332 gmt_lag = process_rtc_config_file(); 1333 1334 /* 1335 * Calculate default settings of system parameters based upon 1336 * maxusers, yet allow to be overridden via the /etc/system file. 1337 */ 1338 param_calc(0); 1339 1340 mod_setup(); 1341 1342 /* 1343 * Initialize system parameters. 1344 */ 1345 param_init(); 1346 1347 /* 1348 * maxmem is the amount of physical memory we're playing with. 1349 */ 1350 maxmem = physmem; 1351 1352 /* 1353 * Initialize the hat layer. 1354 */ 1355 hat_init(); 1356 1357 /* 1358 * Initialize segment management stuff. 1359 */ 1360 seg_init(); 1361 1362 if (modload("fs", "specfs") == -1) 1363 halt("Can't load specfs"); 1364 1365 if (modload("fs", "devfs") == -1) 1366 halt("Can't load devfs"); 1367 1368 dispinit(); 1369 1370 /* 1371 * This is needed here to initialize hw_serial[] for cluster booting. 1372 */ 1373 if ((i = modload("misc", "sysinit")) != (unsigned int)-1) 1374 (void) modunload(i); 1375 else 1376 cmn_err(CE_CONT, "sysinit load failed"); 1377 1378 /* Read cluster configuration data. */ 1379 clconf_init(); 1380 1381 /* 1382 * Create a kernel device tree. First, create rootnex and 1383 * then invoke bus specific code to probe devices. 1384 */ 1385 setup_ddi(); 1386 1387 /* 1388 * Set up the CPU module subsystem. Modifies the device tree, so it 1389 * must be done after setup_ddi(). 1390 */ 1391 cmi_init(); 1392 1393 /* 1394 * Initialize the MCA handlers 1395 */ 1396 if (x86_feature & X86_MCA) 1397 cmi_mca_init(); 1398 1399 /* 1400 * Fake a prom tree such that /dev/openprom continues to work 1401 */ 1402 prom_setup(); 1403 1404 /* 1405 * Load all platform specific modules 1406 */ 1407 psm_modload(); 1408 1409 PRM_POINT("startup_modules() done"); 1410 } 1411 1412 static void 1413 startup_bop_gone(void) 1414 { 1415 PRM_POINT("startup_bop_gone() starting..."); 1416 1417 /* 1418 * Do final allocations of HAT data structures that need to 1419 * be allocated before quiescing the boot loader. 1420 */ 1421 PRM_POINT("Calling hat_kern_alloc()..."); 1422 hat_kern_alloc(); 1423 PRM_POINT("hat_kern_alloc() done"); 1424 1425 /* 1426 * Setup MTRR (Memory type range registers) 1427 */ 1428 setup_mtrr(); 1429 PRM_POINT("startup_bop_gone() done"); 1430 } 1431 1432 /* 1433 * Walk through the pagetables looking for pages mapped in by boot. If the 1434 * setaside flag is set the pages are expected to be returned to the 1435 * kernel later in boot, so we add them to the bootpages list. 1436 */ 1437 static void 1438 protect_boot_range(uintptr_t low, uintptr_t high, int setaside) 1439 { 1440 uintptr_t va = low; 1441 size_t len; 1442 uint_t prot; 1443 pfn_t pfn; 1444 page_t *pp; 1445 pgcnt_t boot_protect_cnt = 0; 1446 1447 while (hat_boot_probe(&va, &len, &pfn, &prot) != 0 && va < high) { 1448 if (va + len >= high) 1449 panic("0x%lx byte mapping at 0x%p exceeds boot's " 1450 "legal range.", len, (void *)va); 1451 1452 while (len > 0) { 1453 pp = page_numtopp_alloc(pfn); 1454 if (pp != NULL) { 1455 if (setaside == 0) 1456 panic("Unexpected mapping by boot. " 1457 "addr=%p pfn=%lx\n", 1458 (void *)va, pfn); 1459 1460 pp->p_next = bootpages; 1461 bootpages = pp; 1462 ++boot_protect_cnt; 1463 } 1464 1465 ++pfn; 1466 len -= MMU_PAGESIZE; 1467 va += MMU_PAGESIZE; 1468 } 1469 } 1470 PRM_DEBUG(boot_protect_cnt); 1471 } 1472 1473 static void 1474 startup_vm(void) 1475 { 1476 struct segmap_crargs a; 1477 extern void hat_kern_setup(void); 1478 pgcnt_t pages_left; 1479 1480 extern int exec_lpg_disable, use_brk_lpg, use_stk_lpg, use_zmap_lpg; 1481 extern pgcnt_t auto_lpg_min_physmem; 1482 1483 PRM_POINT("startup_vm() starting..."); 1484 1485 /* 1486 * The next two loops are done in distinct steps in order 1487 * to be sure that any page that is doubly mapped (both above 1488 * KERNEL_TEXT and below kernelbase) is dealt with correctly. 1489 * Note this may never happen, but it might someday. 1490 */ 1491 1492 bootpages = NULL; 1493 PRM_POINT("Protecting boot pages"); 1494 /* 1495 * Protect any pages mapped above KERNEL_TEXT that somehow have 1496 * page_t's. This can only happen if something weird allocated 1497 * in this range (like kadb/kmdb). 1498 */ 1499 protect_boot_range(KERNEL_TEXT, (uintptr_t)-1, 0); 1500 1501 /* 1502 * Before we can take over memory allocation/mapping from the boot 1503 * loader we must remove from our free page lists any boot pages that 1504 * will stay mapped until release_bootstrap(). 1505 */ 1506 protect_boot_range(0, kernelbase, 1); 1507 #if defined(__amd64) 1508 protect_boot_range(BOOT_DOUBLEMAP_BASE, 1509 BOOT_DOUBLEMAP_BASE + BOOT_DOUBLEMAP_SIZE, 0); 1510 #endif 1511 1512 /* 1513 * Copy in boot's page tables, set up extra page tables for the kernel, 1514 * and switch to the kernel's context. 1515 */ 1516 PRM_POINT("Calling hat_kern_setup()..."); 1517 hat_kern_setup(); 1518 1519 /* 1520 * It is no longer safe to call BOP_ALLOC(), so make sure we don't. 1521 */ 1522 bootops->bsys_alloc = NULL; 1523 PRM_POINT("hat_kern_setup() done"); 1524 1525 hat_cpu_online(CPU); 1526 1527 /* 1528 * Before we call kvm_init(), we need to establish the final size 1529 * of the kernel's heap. So, we need to figure out how much space 1530 * to set aside for segkp, segkpm, and segmap. 1531 */ 1532 final_kernelheap = (caddr_t)ROUND_UP_LPAGE(kernelbase); 1533 #if defined(__amd64) 1534 if (kpm_desired) { 1535 /* 1536 * Segkpm appears at the bottom of the kernel's address 1537 * range. To detect accidental overruns of the user 1538 * address space, we leave a "red zone" of unmapped memory 1539 * between kernelbase and the beginning of segkpm. 1540 */ 1541 kpm_vbase = final_kernelheap + KERNEL_REDZONE_SIZE; 1542 kpm_size = mmu_ptob(physmax); 1543 PRM_DEBUG(kpm_vbase); 1544 PRM_DEBUG(kpm_size); 1545 final_kernelheap = 1546 (caddr_t)ROUND_UP_TOPLEVEL(kpm_vbase + kpm_size); 1547 } 1548 1549 if (!segkp_fromheap) { 1550 size_t sz = mmu_ptob(segkpsize); 1551 1552 /* 1553 * determine size of segkp and adjust the bottom of the 1554 * kernel's heap. 1555 */ 1556 if (sz < SEGKPMINSIZE || sz > SEGKPMAXSIZE) { 1557 sz = SEGKPDEFSIZE; 1558 cmn_err(CE_WARN, "!Illegal value for segkpsize. " 1559 "segkpsize has been reset to %ld pages", 1560 mmu_btop(sz)); 1561 } 1562 sz = MIN(sz, MAX(SEGKPMINSIZE, mmu_ptob(physmem))); 1563 1564 segkpsize = mmu_btop(ROUND_UP_LPAGE(sz)); 1565 segkp_base = final_kernelheap; 1566 PRM_DEBUG(segkpsize); 1567 PRM_DEBUG(segkp_base); 1568 final_kernelheap = segkp_base + mmu_ptob(segkpsize); 1569 PRM_DEBUG(final_kernelheap); 1570 } 1571 1572 /* 1573 * put the range of VA for device mappings next 1574 */ 1575 toxic_addr = (uintptr_t)final_kernelheap; 1576 PRM_DEBUG(toxic_addr); 1577 final_kernelheap = (char *)toxic_addr + toxic_size; 1578 #endif 1579 PRM_DEBUG(final_kernelheap); 1580 ASSERT(final_kernelheap < boot_kernelheap); 1581 1582 /* 1583 * Users can change segmapsize through eeprom or /etc/system. 1584 * If the variable is tuned through eeprom, there is no upper 1585 * bound on the size of segmap. If it is tuned through 1586 * /etc/system on 32-bit systems, it must be no larger than we 1587 * planned for in startup_memlist(). 1588 */ 1589 segmapsize = MAX(ROUND_UP_LPAGE(segmapsize), SEGMAPDEFAULT); 1590 segkmap_start = ROUND_UP_LPAGE((uintptr_t)final_kernelheap); 1591 1592 #if defined(__i386) 1593 if (segmapsize > segmap_reserved) { 1594 cmn_err(CE_NOTE, "!segmapsize may not be set > 0x%lx in " 1595 "/etc/system. Use eeprom.", (long)SEGMAPMAX); 1596 segmapsize = segmap_reserved; 1597 } 1598 /* 1599 * 32-bit systems don't have segkpm or segkp, so segmap appears at 1600 * the bottom of the kernel's address range. Set aside space for a 1601 * red zone just below the start of segmap. 1602 */ 1603 segkmap_start += KERNEL_REDZONE_SIZE; 1604 segmapsize -= KERNEL_REDZONE_SIZE; 1605 #endif 1606 final_kernelheap = (char *)(segkmap_start + segmapsize); 1607 1608 PRM_DEBUG(segkmap_start); 1609 PRM_DEBUG(segmapsize); 1610 PRM_DEBUG(final_kernelheap); 1611 1612 /* 1613 * Initialize VM system 1614 */ 1615 PRM_POINT("Calling kvm_init()..."); 1616 kvm_init(); 1617 PRM_POINT("kvm_init() done"); 1618 1619 /* 1620 * Tell kmdb that the VM system is now working 1621 */ 1622 if (boothowto & RB_DEBUG) 1623 kdi_dvec_vmready(); 1624 1625 /* 1626 * Mangle the brand string etc. 1627 */ 1628 cpuid_pass3(CPU); 1629 1630 PRM_DEBUG(final_kernelheap); 1631 1632 /* 1633 * Now that we can use memory outside the top 4GB (on 64-bit 1634 * systems) and we know the size of segmap, we can set the final 1635 * size of the kernel's heap. Note: on 64-bit systems we still 1636 * can't touch anything in the bottom half of the top 4GB range 1637 * because boot still has pages mapped there. 1638 */ 1639 if (final_kernelheap < boot_kernelheap) { 1640 kernelheap_extend(final_kernelheap, boot_kernelheap); 1641 #if defined(__amd64) 1642 kmem_setaside = vmem_xalloc(heap_arena, BOOT_DOUBLEMAP_SIZE, 1643 MMU_PAGESIZE, 0, 0, (void *)(BOOT_DOUBLEMAP_BASE), 1644 (void *)(BOOT_DOUBLEMAP_BASE + BOOT_DOUBLEMAP_SIZE), 1645 VM_NOSLEEP | VM_BESTFIT | VM_PANIC); 1646 PRM_DEBUG(kmem_setaside); 1647 if (kmem_setaside == NULL) 1648 panic("Could not protect boot's memory"); 1649 #endif 1650 } 1651 /* 1652 * Now that the kernel heap may have grown significantly, we need 1653 * to make all the remaining page_t's available to back that memory. 1654 * 1655 * XX64 this should probably wait till after release boot-strap too. 1656 */ 1657 pages_left = npages - boot_npages; 1658 if (pages_left > 0) { 1659 PRM_DEBUG(pages_left); 1660 (void) kphysm_init(NULL, memseg_base, boot_npages, pages_left); 1661 } 1662 1663 #if defined(__amd64) 1664 1665 /* 1666 * Create the device arena for toxic (to dtrace/kmdb) mappings. 1667 */ 1668 device_arena = vmem_create("device", (void *)toxic_addr, 1669 toxic_size, MMU_PAGESIZE, NULL, NULL, NULL, 0, VM_SLEEP); 1670 1671 #else /* __i386 */ 1672 1673 /* 1674 * allocate the bit map that tracks toxic pages 1675 */ 1676 toxic_bit_map_len = btop((ulong_t)(ptable_va - kernelbase)); 1677 PRM_DEBUG(toxic_bit_map_len); 1678 toxic_bit_map = 1679 kmem_zalloc(BT_SIZEOFMAP(toxic_bit_map_len), KM_NOSLEEP); 1680 ASSERT(toxic_bit_map != NULL); 1681 PRM_DEBUG(toxic_bit_map); 1682 1683 #endif /* __i386 */ 1684 1685 1686 /* 1687 * Now that we've got more VA, as well as the ability to allocate from 1688 * it, tell the debugger. 1689 */ 1690 if (boothowto & RB_DEBUG) 1691 kdi_dvec_memavail(); 1692 1693 /* 1694 * The following code installs a special page fault handler (#pf) 1695 * to work around a pentium bug. 1696 */ 1697 #if !defined(__amd64) 1698 if (x86_type == X86_TYPE_P5) { 1699 gate_desc_t *newidt; 1700 desctbr_t newidt_r; 1701 1702 if ((newidt = kmem_zalloc(MMU_PAGESIZE, KM_NOSLEEP)) == NULL) 1703 panic("failed to install pentium_pftrap"); 1704 1705 bcopy(idt0, newidt, sizeof (idt0)); 1706 set_gatesegd(&newidt[T_PGFLT], &pentium_pftrap, 1707 KCS_SEL, 0, SDT_SYSIGT, SEL_KPL); 1708 1709 (void) as_setprot(&kas, (caddr_t)newidt, MMU_PAGESIZE, 1710 PROT_READ|PROT_EXEC); 1711 1712 newidt_r.dtr_limit = sizeof (idt0) - 1; 1713 newidt_r.dtr_base = (uintptr_t)newidt; 1714 CPU->cpu_idt = newidt; 1715 wr_idtr(&newidt_r); 1716 } 1717 #endif /* !__amd64 */ 1718 1719 /* 1720 * Map page pfn=0 for drivers, such as kd, that need to pick up 1721 * parameters left there by controllers/BIOS. 1722 */ 1723 PRM_POINT("setup up p0_va"); 1724 p0_va = i86devmap(0, 1, PROT_READ); 1725 PRM_DEBUG(p0_va); 1726 1727 cmn_err(CE_CONT, "?mem = %luK (0x%lx)\n", 1728 physinstalled << (MMU_PAGESHIFT - 10), ptob(physinstalled)); 1729 1730 /* 1731 * disable automatic large pages for small memory systems or 1732 * when the disable flag is set. 1733 */ 1734 if (physmem < auto_lpg_min_physmem || auto_lpg_disable) { 1735 exec_lpg_disable = 1; 1736 use_brk_lpg = 0; 1737 use_stk_lpg = 0; 1738 use_zmap_lpg = 0; 1739 } 1740 1741 PRM_POINT("Calling hat_init_finish()..."); 1742 hat_init_finish(); 1743 PRM_POINT("hat_init_finish() done"); 1744 1745 /* 1746 * Initialize the segkp segment type. 1747 */ 1748 rw_enter(&kas.a_lock, RW_WRITER); 1749 if (!segkp_fromheap) { 1750 if (seg_attach(&kas, (caddr_t)segkp_base, mmu_ptob(segkpsize), 1751 segkp) < 0) { 1752 panic("startup: cannot attach segkp"); 1753 /*NOTREACHED*/ 1754 } 1755 } else { 1756 /* 1757 * For 32 bit x86 systems, we will have segkp under the heap. 1758 * There will not be a segkp segment. We do, however, need 1759 * to fill in the seg structure. 1760 */ 1761 segkp->s_as = &kas; 1762 } 1763 if (segkp_create(segkp) != 0) { 1764 panic("startup: segkp_create failed"); 1765 /*NOTREACHED*/ 1766 } 1767 PRM_DEBUG(segkp); 1768 rw_exit(&kas.a_lock); 1769 1770 /* 1771 * kpm segment 1772 */ 1773 segmap_kpm = 0; 1774 if (kpm_desired) { 1775 kpm_init(); 1776 kpm_enable = 1; 1777 } 1778 1779 /* 1780 * Now create segmap segment. 1781 */ 1782 rw_enter(&kas.a_lock, RW_WRITER); 1783 if (seg_attach(&kas, (caddr_t)segkmap_start, segmapsize, segkmap) < 0) { 1784 panic("cannot attach segkmap"); 1785 /*NOTREACHED*/ 1786 } 1787 PRM_DEBUG(segkmap); 1788 1789 /* 1790 * The 64 bit HAT permanently maps only segmap's page tables. 1791 * The 32 bit HAT maps the heap's page tables too. 1792 */ 1793 #if defined(__amd64) 1794 hat_kmap_init(segkmap_start, segmapsize); 1795 #else /* __i386 */ 1796 ASSERT(segkmap_start + segmapsize == (uintptr_t)final_kernelheap); 1797 hat_kmap_init(segkmap_start, (uintptr_t)ekernelheap - segkmap_start); 1798 #endif /* __i386 */ 1799 1800 a.prot = PROT_READ | PROT_WRITE; 1801 a.shmsize = 0; 1802 a.nfreelist = segmapfreelists; 1803 1804 if (segmap_create(segkmap, (caddr_t)&a) != 0) 1805 panic("segmap_create segkmap"); 1806 rw_exit(&kas.a_lock); 1807 1808 setup_vaddr_for_ppcopy(CPU); 1809 1810 segdev_init(); 1811 pmem_init(); 1812 PRM_POINT("startup_vm() done"); 1813 } 1814 1815 static void 1816 startup_end(void) 1817 { 1818 extern void setx86isalist(void); 1819 1820 PRM_POINT("startup_end() starting..."); 1821 1822 /* 1823 * Perform tasks that get done after most of the VM 1824 * initialization has been done but before the clock 1825 * and other devices get started. 1826 */ 1827 kern_setup1(); 1828 1829 /* 1830 * Perform CPC initialization for this CPU. 1831 */ 1832 kcpc_hw_init(CPU); 1833 1834 #if defined(__amd64) 1835 /* 1836 * Validate support for syscall/sysret 1837 * XX64 -- include SSE, SSE2, etc. here too? 1838 */ 1839 if ((x86_feature & X86_ASYSC) == 0) { 1840 cmn_err(CE_WARN, 1841 "cpu%d does not support syscall/sysret", CPU->cpu_id); 1842 } 1843 #endif 1844 1845 #if defined(OPTERON_WORKAROUND_6323525) 1846 if (opteron_workaround_6323525) 1847 patch_workaround_6323525(); 1848 #endif 1849 /* 1850 * Configure the system. 1851 */ 1852 PRM_POINT("Calling configure()..."); 1853 configure(); /* set up devices */ 1854 PRM_POINT("configure() done"); 1855 1856 /* 1857 * Set the isa_list string to the defined instruction sets we 1858 * support. 1859 */ 1860 setx86isalist(); 1861 cpu_intr_alloc(CPU, NINTR_THREADS); 1862 psm_install(); 1863 1864 /* 1865 * We're done with bootops. We don't unmap the bootstrap yet because 1866 * we're still using bootsvcs. 1867 */ 1868 PRM_POINT("zeroing out bootops"); 1869 *bootopsp = (struct bootops *)0; 1870 bootops = (struct bootops *)NULL; 1871 1872 PRM_POINT("Enabling interrupts"); 1873 (*picinitf)(); 1874 sti(); 1875 1876 (void) add_avsoftintr((void *)&softlevel1_hdl, 1, softlevel1, 1877 "softlevel1", NULL, NULL); /* XXX to be moved later */ 1878 1879 PRM_POINT("startup_end() done"); 1880 } 1881 1882 extern char hw_serial[]; 1883 char *_hs1107 = hw_serial; 1884 ulong_t _bdhs34; 1885 1886 void 1887 post_startup(void) 1888 { 1889 /* 1890 * Set the system wide, processor-specific flags to be passed 1891 * to userland via the aux vector for performance hints and 1892 * instruction set extensions. 1893 */ 1894 bind_hwcap(); 1895 1896 /* 1897 * Load the System Management BIOS into the global ksmbios handle, 1898 * if an SMBIOS is present on this system. 1899 */ 1900 ksmbios = smbios_open(NULL, SMB_VERSION, ksmbios_flags, NULL); 1901 1902 /* 1903 * Startup memory scrubber. 1904 */ 1905 memscrub_init(); 1906 1907 /* 1908 * Complete CPU module initialization 1909 */ 1910 cmi_post_init(); 1911 1912 /* 1913 * Perform forceloading tasks for /etc/system. 1914 */ 1915 (void) mod_sysctl(SYS_FORCELOAD, NULL); 1916 1917 /* 1918 * ON4.0: Force /proc module in until clock interrupt handle fixed 1919 * ON4.0: This must be fixed or restated in /etc/systems. 1920 */ 1921 (void) modload("fs", "procfs"); 1922 1923 #if defined(__i386) 1924 /* 1925 * Check for required functional Floating Point hardware, 1926 * unless FP hardware explicitly disabled. 1927 */ 1928 if (fpu_exists && (fpu_pentium_fdivbug || fp_kind == FP_NO)) 1929 halt("No working FP hardware found"); 1930 #endif 1931 1932 maxmem = freemem; 1933 1934 add_cpunode2devtree(CPU->cpu_id, CPU->cpu_m.mcpu_cpi); 1935 1936 /* 1937 * Perform the formal initialization of the boot chip, 1938 * and associate the boot cpu with it. 1939 * This must be done after the cpu node for CPU has been 1940 * added to the device tree, when the necessary probing to 1941 * know the chip type and chip "id" is performed. 1942 */ 1943 chip_cpu_init(CPU); 1944 chip_cpu_assign(CPU); 1945 } 1946 1947 static int 1948 pp_in_ramdisk(page_t *pp) 1949 { 1950 extern uint64_t ramdisk_start, ramdisk_end; 1951 1952 return ((pp->p_pagenum >= btop(ramdisk_start)) && 1953 (pp->p_pagenum < btopr(ramdisk_end))); 1954 } 1955 1956 void 1957 release_bootstrap(void) 1958 { 1959 int root_is_ramdisk; 1960 pfn_t pfn; 1961 page_t *pp; 1962 extern void kobj_boot_unmountroot(void); 1963 extern dev_t rootdev; 1964 1965 /* unmount boot ramdisk and release kmem usage */ 1966 kobj_boot_unmountroot(); 1967 1968 /* 1969 * We're finished using the boot loader so free its pages. 1970 */ 1971 PRM_POINT("Unmapping lower boot pages"); 1972 clear_boot_mappings(0, kernelbase); 1973 #if defined(__amd64) 1974 PRM_POINT("Unmapping upper boot pages"); 1975 clear_boot_mappings(BOOT_DOUBLEMAP_BASE, 1976 BOOT_DOUBLEMAP_BASE + BOOT_DOUBLEMAP_SIZE); 1977 #endif 1978 1979 /* 1980 * If root isn't on ramdisk, destroy the hardcoded 1981 * ramdisk node now and release the memory. Else, 1982 * ramdisk memory is kept in rd_pages. 1983 */ 1984 root_is_ramdisk = (getmajor(rootdev) == ddi_name_to_major("ramdisk")); 1985 if (!root_is_ramdisk) { 1986 dev_info_t *dip = ddi_find_devinfo("ramdisk", -1, 0); 1987 ASSERT(dip && ddi_get_parent(dip) == ddi_root_node()); 1988 ndi_rele_devi(dip); /* held from ddi_find_devinfo */ 1989 (void) ddi_remove_child(dip, 0); 1990 } 1991 1992 PRM_POINT("Releasing boot pages"); 1993 while (bootpages) { 1994 pp = bootpages; 1995 bootpages = pp->p_next; 1996 if (root_is_ramdisk && pp_in_ramdisk(pp)) { 1997 pp->p_next = rd_pages; 1998 rd_pages = pp; 1999 continue; 2000 } 2001 pp->p_next = (struct page *)0; 2002 page_free(pp, 1); 2003 } 2004 2005 /* 2006 * Find 1 page below 1 MB so that other processors can boot up. 2007 * Make sure it has a kernel VA as well as a 1:1 mapping. 2008 * We should have just free'd one up. 2009 */ 2010 if (use_mp) { 2011 for (pfn = 1; pfn < btop(1*1024*1024); pfn++) { 2012 if (page_numtopp_alloc(pfn) == NULL) 2013 continue; 2014 rm_platter_va = i86devmap(pfn, 1, 2015 PROT_READ | PROT_WRITE | PROT_EXEC); 2016 rm_platter_pa = ptob(pfn); 2017 hat_devload(kas.a_hat, 2018 (caddr_t)(uintptr_t)rm_platter_pa, MMU_PAGESIZE, 2019 pfn, PROT_READ | PROT_WRITE | PROT_EXEC, 2020 HAT_LOAD_NOCONSIST); 2021 break; 2022 } 2023 if (pfn == btop(1*1024*1024)) 2024 panic("No page available for starting " 2025 "other processors"); 2026 } 2027 2028 #if defined(__amd64) 2029 PRM_POINT("Returning boot's VA space to kernel heap"); 2030 if (kmem_setaside != NULL) 2031 vmem_free(heap_arena, kmem_setaside, BOOT_DOUBLEMAP_SIZE); 2032 #endif 2033 } 2034 2035 /* 2036 * Initialize the platform-specific parts of a page_t. 2037 */ 2038 void 2039 add_physmem_cb(page_t *pp, pfn_t pnum) 2040 { 2041 pp->p_pagenum = pnum; 2042 pp->p_mapping = NULL; 2043 pp->p_embed = 0; 2044 pp->p_share = 0; 2045 pp->p_mlentry = 0; 2046 } 2047 2048 /* 2049 * kphysm_init() initializes physical memory. 2050 */ 2051 static pgcnt_t 2052 kphysm_init( 2053 page_t *inpp, 2054 struct memseg *memsegp, 2055 pgcnt_t start, 2056 pgcnt_t npages) 2057 { 2058 struct memlist *pmem; 2059 struct memseg *cur_memseg; 2060 struct memseg **memsegpp; 2061 pfn_t base_pfn; 2062 pgcnt_t num; 2063 pgcnt_t total_skipped = 0; 2064 pgcnt_t skipping = 0; 2065 pgcnt_t pages_done = 0; 2066 pgcnt_t largepgcnt; 2067 uint64_t addr; 2068 uint64_t size; 2069 page_t *pp = inpp; 2070 int dobreak = 0; 2071 extern pfn_t ddiphysmin; 2072 2073 ASSERT(page_hash != NULL && page_hashsz != 0); 2074 2075 for (cur_memseg = memsegp; cur_memseg->pages != NULL; cur_memseg++); 2076 ASSERT(cur_memseg == memsegp || start > 0); 2077 2078 for (pmem = phys_avail; pmem && npages; pmem = pmem->next) { 2079 /* 2080 * In a 32 bit kernel can't use higher memory if we're 2081 * not booting in PAE mode. This check takes care of that. 2082 */ 2083 addr = pmem->address; 2084 size = pmem->size; 2085 if (btop(addr) > physmax) 2086 continue; 2087 2088 /* 2089 * align addr and size - they may not be at page boundaries 2090 */ 2091 if ((addr & MMU_PAGEOFFSET) != 0) { 2092 addr += MMU_PAGEOFFSET; 2093 addr &= ~(uint64_t)MMU_PAGEOFFSET; 2094 size -= addr - pmem->address; 2095 } 2096 2097 /* only process pages below or equal to physmax */ 2098 if ((btop(addr + size) - 1) > physmax) 2099 size = ptob(physmax - btop(addr) + 1); 2100 2101 num = btop(size); 2102 if (num == 0) 2103 continue; 2104 2105 if (total_skipped < start) { 2106 if (start - total_skipped > num) { 2107 total_skipped += num; 2108 continue; 2109 } 2110 skipping = start - total_skipped; 2111 num -= skipping; 2112 addr += (MMU_PAGESIZE * skipping); 2113 total_skipped = start; 2114 } 2115 if (num == 0) 2116 continue; 2117 2118 if (num > npages) 2119 num = npages; 2120 2121 npages -= num; 2122 pages_done += num; 2123 base_pfn = btop(addr); 2124 2125 /* 2126 * If the caller didn't provide space for the page 2127 * structures, carve them out of the memseg they will 2128 * represent. 2129 */ 2130 if (pp == NULL) { 2131 pgcnt_t pp_pgs; 2132 2133 if (num <= 1) 2134 continue; 2135 2136 /* 2137 * Compute how many of the pages we need to use for 2138 * page_ts 2139 */ 2140 pp_pgs = (num * sizeof (page_t)) / MMU_PAGESIZE + 1; 2141 while (mmu_ptob(pp_pgs - 1) / sizeof (page_t) >= 2142 num - pp_pgs + 1) 2143 --pp_pgs; 2144 PRM_DEBUG(pp_pgs); 2145 2146 pp = vmem_alloc(heap_arena, mmu_ptob(pp_pgs), 2147 VM_NOSLEEP); 2148 if (pp == NULL) { 2149 cmn_err(CE_WARN, "Unable to add %ld pages to " 2150 "the system.", num); 2151 continue; 2152 } 2153 2154 hat_devload(kas.a_hat, (void *)pp, mmu_ptob(pp_pgs), 2155 base_pfn, PROT_READ | PROT_WRITE | HAT_UNORDERED_OK, 2156 HAT_LOAD | HAT_LOAD_LOCK | HAT_LOAD_NOCONSIST); 2157 bzero(pp, mmu_ptob(pp_pgs)); 2158 num -= pp_pgs; 2159 base_pfn += pp_pgs; 2160 } 2161 2162 if (prom_debug) 2163 prom_printf("MEMSEG addr=0x%" PRIx64 2164 " pgs=0x%lx pfn 0x%lx-0x%lx\n", 2165 addr, num, base_pfn, base_pfn + num); 2166 2167 /* 2168 * drop pages below ddiphysmin to simplify ddi memory 2169 * allocation with non-zero addr_lo requests. 2170 */ 2171 if (base_pfn < ddiphysmin) { 2172 if (base_pfn + num <= ddiphysmin) { 2173 /* drop entire range below ddiphysmin */ 2174 continue; 2175 } 2176 /* adjust range to ddiphysmin */ 2177 pp += (ddiphysmin - base_pfn); 2178 num -= (ddiphysmin - base_pfn); 2179 base_pfn = ddiphysmin; 2180 } 2181 /* 2182 * Build the memsegs entry 2183 */ 2184 cur_memseg->pages = pp; 2185 cur_memseg->epages = pp + num; 2186 cur_memseg->pages_base = base_pfn; 2187 cur_memseg->pages_end = base_pfn + num; 2188 2189 /* 2190 * insert in memseg list in decreasing pfn range order. 2191 * Low memory is typically more fragmented such that this 2192 * ordering keeps the larger ranges at the front of the list 2193 * for code that searches memseg. 2194 */ 2195 memsegpp = &memsegs; 2196 for (;;) { 2197 if (*memsegpp == NULL) { 2198 /* empty memsegs */ 2199 memsegs = cur_memseg; 2200 break; 2201 } 2202 /* check for continuity with start of memsegpp */ 2203 if (cur_memseg->pages_end == (*memsegpp)->pages_base) { 2204 if (cur_memseg->epages == (*memsegpp)->pages) { 2205 /* 2206 * contiguous pfn and page_t's. Merge 2207 * cur_memseg into *memsegpp. Drop 2208 * cur_memseg 2209 */ 2210 (*memsegpp)->pages_base = 2211 cur_memseg->pages_base; 2212 (*memsegpp)->pages = 2213 cur_memseg->pages; 2214 /* 2215 * check if contiguous with the end of 2216 * the next memseg. 2217 */ 2218 if ((*memsegpp)->next && 2219 ((*memsegpp)->pages_base == 2220 (*memsegpp)->next->pages_end)) { 2221 cur_memseg = *memsegpp; 2222 memsegpp = &((*memsegpp)->next); 2223 dobreak = 1; 2224 } else { 2225 break; 2226 } 2227 } else { 2228 /* 2229 * contiguous pfn but not page_t's. 2230 * drop last pfn/page_t in cur_memseg 2231 * to prevent creation of large pages 2232 * with noncontiguous page_t's if not 2233 * aligned to largest page boundary. 2234 */ 2235 largepgcnt = page_get_pagecnt( 2236 page_num_pagesizes() - 1); 2237 2238 if (cur_memseg->pages_end & 2239 (largepgcnt - 1)) { 2240 num--; 2241 cur_memseg->epages--; 2242 cur_memseg->pages_end--; 2243 } 2244 } 2245 } 2246 2247 /* check for continuity with end of memsegpp */ 2248 if (cur_memseg->pages_base == (*memsegpp)->pages_end) { 2249 if (cur_memseg->pages == (*memsegpp)->epages) { 2250 /* 2251 * contiguous pfn and page_t's. Merge 2252 * cur_memseg into *memsegpp. Drop 2253 * cur_memseg. 2254 */ 2255 if (dobreak) { 2256 /* merge previously done */ 2257 cur_memseg->pages = 2258 (*memsegpp)->pages; 2259 cur_memseg->pages_base = 2260 (*memsegpp)->pages_base; 2261 cur_memseg->next = 2262 (*memsegpp)->next; 2263 } else { 2264 (*memsegpp)->pages_end = 2265 cur_memseg->pages_end; 2266 (*memsegpp)->epages = 2267 cur_memseg->epages; 2268 } 2269 break; 2270 } 2271 /* 2272 * contiguous pfn but not page_t's. 2273 * drop first pfn/page_t in cur_memseg 2274 * to prevent creation of large pages 2275 * with noncontiguous page_t's if not 2276 * aligned to largest page boundary. 2277 */ 2278 largepgcnt = page_get_pagecnt( 2279 page_num_pagesizes() - 1); 2280 if (base_pfn & (largepgcnt - 1)) { 2281 num--; 2282 base_pfn++; 2283 cur_memseg->pages++; 2284 cur_memseg->pages_base++; 2285 pp = cur_memseg->pages; 2286 } 2287 if (dobreak) 2288 break; 2289 } 2290 2291 if (cur_memseg->pages_base >= 2292 (*memsegpp)->pages_end) { 2293 cur_memseg->next = *memsegpp; 2294 *memsegpp = cur_memseg; 2295 break; 2296 } 2297 if ((*memsegpp)->next == NULL) { 2298 cur_memseg->next = NULL; 2299 (*memsegpp)->next = cur_memseg; 2300 break; 2301 } 2302 memsegpp = &((*memsegpp)->next); 2303 ASSERT(*memsegpp != NULL); 2304 } 2305 2306 /* 2307 * add_physmem() initializes the PSM part of the page 2308 * struct by calling the PSM back with add_physmem_cb(). 2309 * In addition it coalesces pages into larger pages as 2310 * it initializes them. 2311 */ 2312 add_physmem(pp, num, base_pfn); 2313 cur_memseg++; 2314 availrmem_initial += num; 2315 availrmem += num; 2316 2317 /* 2318 * If the caller provided the page frames to us, then 2319 * advance in that list. Otherwise, prepare to allocate 2320 * our own page frames for the next memseg. 2321 */ 2322 pp = (inpp == NULL) ? NULL : pp + num; 2323 } 2324 2325 PRM_DEBUG(availrmem_initial); 2326 PRM_DEBUG(availrmem); 2327 PRM_DEBUG(freemem); 2328 build_pfn_hash(); 2329 return (pages_done); 2330 } 2331 2332 /* 2333 * Kernel VM initialization. 2334 */ 2335 static void 2336 kvm_init(void) 2337 { 2338 #ifdef DEBUG 2339 extern void _start(); 2340 2341 ASSERT((caddr_t)_start == s_text); 2342 #endif 2343 ASSERT((((uintptr_t)s_text) & MMU_PAGEOFFSET) == 0); 2344 2345 /* 2346 * Put the kernel segments in kernel address space. 2347 */ 2348 rw_enter(&kas.a_lock, RW_WRITER); 2349 as_avlinit(&kas); 2350 2351 (void) seg_attach(&kas, s_text, e_moddata - s_text, &ktextseg); 2352 (void) segkmem_create(&ktextseg); 2353 2354 (void) seg_attach(&kas, (caddr_t)valloc_base, valloc_sz, &kvalloc); 2355 (void) segkmem_create(&kvalloc); 2356 2357 /* 2358 * We're about to map out /boot. This is the beginning of the 2359 * system resource management transition. We can no longer 2360 * call into /boot for I/O or memory allocations. 2361 * 2362 * XX64 - Is this still correct with kernelheap_extend() being called 2363 * later than this???? 2364 */ 2365 (void) seg_attach(&kas, final_kernelheap, 2366 ekernelheap - final_kernelheap, &kvseg); 2367 (void) segkmem_create(&kvseg); 2368 2369 #if defined(__amd64) 2370 (void) seg_attach(&kas, (caddr_t)core_base, core_size, &kvseg_core); 2371 (void) segkmem_create(&kvseg_core); 2372 #endif 2373 2374 (void) seg_attach(&kas, (caddr_t)SEGDEBUGBASE, (size_t)SEGDEBUGSIZE, 2375 &kdebugseg); 2376 (void) segkmem_create(&kdebugseg); 2377 2378 rw_exit(&kas.a_lock); 2379 2380 /* 2381 * Ensure that the red zone at kernelbase is never accessible. 2382 */ 2383 (void) as_setprot(&kas, (caddr_t)kernelbase, KERNEL_REDZONE_SIZE, 0); 2384 2385 /* 2386 * Make the text writable so that it can be hot patched by DTrace. 2387 */ 2388 (void) as_setprot(&kas, s_text, e_modtext - s_text, 2389 PROT_READ | PROT_WRITE | PROT_EXEC); 2390 2391 /* 2392 * Make data writable until end. 2393 */ 2394 (void) as_setprot(&kas, s_data, e_moddata - s_data, 2395 PROT_READ | PROT_WRITE | PROT_EXEC); 2396 } 2397 2398 /* 2399 * These are MTTR registers supported by P6 2400 */ 2401 static struct mtrrvar mtrrphys_arr[MAX_MTRRVAR]; 2402 static uint64_t mtrr64k, mtrr16k1, mtrr16k2; 2403 static uint64_t mtrr4k1, mtrr4k2, mtrr4k3; 2404 static uint64_t mtrr4k4, mtrr4k5, mtrr4k6; 2405 static uint64_t mtrr4k7, mtrr4k8, mtrrcap; 2406 uint64_t mtrrdef, pat_attr_reg; 2407 2408 /* 2409 * Disable reprogramming of MTRRs by default. 2410 */ 2411 int enable_relaxed_mtrr = 0; 2412 2413 void 2414 setup_mtrr() 2415 { 2416 int i, ecx; 2417 int vcnt; 2418 struct mtrrvar *mtrrphys; 2419 2420 if (!(x86_feature & X86_MTRR)) 2421 return; 2422 2423 mtrrcap = rdmsr(REG_MTRRCAP); 2424 mtrrdef = rdmsr(REG_MTRRDEF); 2425 if (mtrrcap & MTRRCAP_FIX) { 2426 mtrr64k = rdmsr(REG_MTRR64K); 2427 mtrr16k1 = rdmsr(REG_MTRR16K1); 2428 mtrr16k2 = rdmsr(REG_MTRR16K2); 2429 mtrr4k1 = rdmsr(REG_MTRR4K1); 2430 mtrr4k2 = rdmsr(REG_MTRR4K2); 2431 mtrr4k3 = rdmsr(REG_MTRR4K3); 2432 mtrr4k4 = rdmsr(REG_MTRR4K4); 2433 mtrr4k5 = rdmsr(REG_MTRR4K5); 2434 mtrr4k6 = rdmsr(REG_MTRR4K6); 2435 mtrr4k7 = rdmsr(REG_MTRR4K7); 2436 mtrr4k8 = rdmsr(REG_MTRR4K8); 2437 } 2438 if ((vcnt = (mtrrcap & MTRRCAP_VCNTMASK)) > MAX_MTRRVAR) 2439 vcnt = MAX_MTRRVAR; 2440 2441 for (i = 0, ecx = REG_MTRRPHYSBASE0, mtrrphys = mtrrphys_arr; 2442 i < vcnt - 1; i++, ecx += 2, mtrrphys++) { 2443 mtrrphys->mtrrphys_base = rdmsr(ecx); 2444 mtrrphys->mtrrphys_mask = rdmsr(ecx + 1); 2445 if ((x86_feature & X86_PAT) && enable_relaxed_mtrr) { 2446 mtrrphys->mtrrphys_mask &= ~MTRRPHYSMASK_V; 2447 } 2448 } 2449 if (x86_feature & X86_PAT) { 2450 if (enable_relaxed_mtrr) 2451 mtrrdef = MTRR_TYPE_WB|MTRRDEF_FE|MTRRDEF_E; 2452 pat_attr_reg = PAT_DEFAULT_ATTRIBUTE; 2453 } 2454 2455 mtrr_sync(); 2456 } 2457 2458 /* 2459 * Sync current cpu mtrr with the incore copy of mtrr. 2460 * This function has to be invoked with interrupts disabled 2461 * Currently we do not capture other cpu's. This is invoked on cpu0 2462 * just after reading /etc/system. 2463 * On other cpu's its invoked from mp_startup(). 2464 */ 2465 void 2466 mtrr_sync() 2467 { 2468 uint_t crvalue, cr0_orig; 2469 int vcnt, i, ecx; 2470 struct mtrrvar *mtrrphys; 2471 2472 cr0_orig = crvalue = getcr0(); 2473 crvalue |= CR0_CD; 2474 crvalue &= ~CR0_NW; 2475 setcr0(crvalue); 2476 invalidate_cache(); 2477 setcr3(getcr3()); 2478 2479 if (x86_feature & X86_PAT) 2480 wrmsr(REG_MTRRPAT, pat_attr_reg); 2481 2482 wrmsr(REG_MTRRDEF, rdmsr(REG_MTRRDEF) & 2483 ~((uint64_t)(uintptr_t)MTRRDEF_E)); 2484 2485 if (mtrrcap & MTRRCAP_FIX) { 2486 wrmsr(REG_MTRR64K, mtrr64k); 2487 wrmsr(REG_MTRR16K1, mtrr16k1); 2488 wrmsr(REG_MTRR16K2, mtrr16k2); 2489 wrmsr(REG_MTRR4K1, mtrr4k1); 2490 wrmsr(REG_MTRR4K2, mtrr4k2); 2491 wrmsr(REG_MTRR4K3, mtrr4k3); 2492 wrmsr(REG_MTRR4K4, mtrr4k4); 2493 wrmsr(REG_MTRR4K5, mtrr4k5); 2494 wrmsr(REG_MTRR4K6, mtrr4k6); 2495 wrmsr(REG_MTRR4K7, mtrr4k7); 2496 wrmsr(REG_MTRR4K8, mtrr4k8); 2497 } 2498 if ((vcnt = (mtrrcap & MTRRCAP_VCNTMASK)) > MAX_MTRRVAR) 2499 vcnt = MAX_MTRRVAR; 2500 for (i = 0, ecx = REG_MTRRPHYSBASE0, mtrrphys = mtrrphys_arr; 2501 i < vcnt - 1; i++, ecx += 2, mtrrphys++) { 2502 wrmsr(ecx, mtrrphys->mtrrphys_base); 2503 wrmsr(ecx + 1, mtrrphys->mtrrphys_mask); 2504 } 2505 wrmsr(REG_MTRRDEF, mtrrdef); 2506 setcr3(getcr3()); 2507 invalidate_cache(); 2508 setcr0(cr0_orig); 2509 } 2510 2511 /* 2512 * resync mtrr so that BIOS is happy. Called from mdboot 2513 */ 2514 void 2515 mtrr_resync() 2516 { 2517 if ((x86_feature & X86_PAT) && enable_relaxed_mtrr) { 2518 /* 2519 * We could have changed the default mtrr definition. 2520 * Put it back to uncached which is what it is at power on 2521 */ 2522 mtrrdef = MTRR_TYPE_UC|MTRRDEF_FE|MTRRDEF_E; 2523 mtrr_sync(); 2524 } 2525 } 2526 2527 void 2528 get_system_configuration() 2529 { 2530 char prop[32]; 2531 u_longlong_t nodes_ll, cpus_pernode_ll, lvalue; 2532 2533 if (((BOP_GETPROPLEN(bootops, "nodes") > sizeof (prop)) || 2534 (BOP_GETPROP(bootops, "nodes", prop) < 0) || 2535 (kobj_getvalue(prop, &nodes_ll) == -1) || 2536 (nodes_ll > MAXNODES)) || 2537 ((BOP_GETPROPLEN(bootops, "cpus_pernode") > sizeof (prop)) || 2538 (BOP_GETPROP(bootops, "cpus_pernode", prop) < 0) || 2539 (kobj_getvalue(prop, &cpus_pernode_ll) == -1))) { 2540 2541 system_hardware.hd_nodes = 1; 2542 system_hardware.hd_cpus_per_node = 0; 2543 } else { 2544 system_hardware.hd_nodes = (int)nodes_ll; 2545 system_hardware.hd_cpus_per_node = (int)cpus_pernode_ll; 2546 } 2547 if ((BOP_GETPROPLEN(bootops, "kernelbase") > sizeof (prop)) || 2548 (BOP_GETPROP(bootops, "kernelbase", prop) < 0) || 2549 (kobj_getvalue(prop, &lvalue) == -1)) 2550 eprom_kernelbase = NULL; 2551 else 2552 eprom_kernelbase = (uintptr_t)lvalue; 2553 2554 if ((BOP_GETPROPLEN(bootops, "segmapsize") > sizeof (prop)) || 2555 (BOP_GETPROP(bootops, "segmapsize", prop) < 0) || 2556 (kobj_getvalue(prop, &lvalue) == -1)) { 2557 segmapsize = SEGMAPDEFAULT; 2558 } else { 2559 segmapsize = (uintptr_t)lvalue; 2560 } 2561 2562 if ((BOP_GETPROPLEN(bootops, "segmapfreelists") > sizeof (prop)) || 2563 (BOP_GETPROP(bootops, "segmapfreelists", prop) < 0) || 2564 (kobj_getvalue(prop, &lvalue) == -1)) { 2565 segmapfreelists = 0; /* use segmap driver default */ 2566 } else { 2567 segmapfreelists = (int)lvalue; 2568 } 2569 2570 if ((BOP_GETPROPLEN(bootops, "physmem") <= sizeof (prop)) && 2571 (BOP_GETPROP(bootops, "physmem", prop) >= 0) && 2572 (kobj_getvalue(prop, &lvalue) != -1)) { 2573 physmem = (uintptr_t)lvalue; 2574 } 2575 } 2576 2577 /* 2578 * Add to a memory list. 2579 * start = start of new memory segment 2580 * len = length of new memory segment in bytes 2581 * new = pointer to a new struct memlist 2582 * memlistp = memory list to which to add segment. 2583 */ 2584 static void 2585 memlist_add( 2586 uint64_t start, 2587 uint64_t len, 2588 struct memlist *new, 2589 struct memlist **memlistp) 2590 { 2591 struct memlist *cur; 2592 uint64_t end = start + len; 2593 2594 new->address = start; 2595 new->size = len; 2596 2597 cur = *memlistp; 2598 2599 while (cur) { 2600 if (cur->address >= end) { 2601 new->next = cur; 2602 *memlistp = new; 2603 new->prev = cur->prev; 2604 cur->prev = new; 2605 return; 2606 } 2607 ASSERT(cur->address + cur->size <= start); 2608 if (cur->next == NULL) { 2609 cur->next = new; 2610 new->prev = cur; 2611 new->next = NULL; 2612 return; 2613 } 2614 memlistp = &cur->next; 2615 cur = cur->next; 2616 } 2617 } 2618 2619 void 2620 kobj_vmem_init(vmem_t **text_arena, vmem_t **data_arena) 2621 { 2622 size_t tsize = e_modtext - modtext; 2623 size_t dsize = e_moddata - moddata; 2624 2625 *text_arena = vmem_create("module_text", tsize ? modtext : NULL, tsize, 2626 1, segkmem_alloc, segkmem_free, heaptext_arena, 0, VM_SLEEP); 2627 *data_arena = vmem_create("module_data", dsize ? moddata : NULL, dsize, 2628 1, segkmem_alloc, segkmem_free, heap32_arena, 0, VM_SLEEP); 2629 } 2630 2631 caddr_t 2632 kobj_text_alloc(vmem_t *arena, size_t size) 2633 { 2634 return (vmem_alloc(arena, size, VM_SLEEP | VM_BESTFIT)); 2635 } 2636 2637 /*ARGSUSED*/ 2638 caddr_t 2639 kobj_texthole_alloc(caddr_t addr, size_t size) 2640 { 2641 panic("unexpected call to kobj_texthole_alloc()"); 2642 /*NOTREACHED*/ 2643 return (0); 2644 } 2645 2646 /*ARGSUSED*/ 2647 void 2648 kobj_texthole_free(caddr_t addr, size_t size) 2649 { 2650 panic("unexpected call to kobj_texthole_free()"); 2651 } 2652 2653 /* 2654 * This is called just after configure() in startup(). 2655 * 2656 * The ISALIST concept is a bit hopeless on Intel, because 2657 * there's no guarantee of an ever-more-capable processor 2658 * given that various parts of the instruction set may appear 2659 * and disappear between different implementations. 2660 * 2661 * While it would be possible to correct it and even enhance 2662 * it somewhat, the explicit hardware capability bitmask allows 2663 * more flexibility. 2664 * 2665 * So, we just leave this alone. 2666 */ 2667 void 2668 setx86isalist(void) 2669 { 2670 char *tp; 2671 size_t len; 2672 extern char *isa_list; 2673 2674 #define TBUFSIZE 1024 2675 2676 tp = kmem_alloc(TBUFSIZE, KM_SLEEP); 2677 *tp = '\0'; 2678 2679 #if defined(__amd64) 2680 (void) strcpy(tp, "amd64 "); 2681 #endif 2682 2683 switch (x86_vendor) { 2684 case X86_VENDOR_Intel: 2685 case X86_VENDOR_AMD: 2686 case X86_VENDOR_TM: 2687 if (x86_feature & X86_CMOV) { 2688 /* 2689 * Pentium Pro or later 2690 */ 2691 (void) strcat(tp, "pentium_pro"); 2692 (void) strcat(tp, x86_feature & X86_MMX ? 2693 "+mmx pentium_pro " : " "); 2694 } 2695 /*FALLTHROUGH*/ 2696 case X86_VENDOR_Cyrix: 2697 /* 2698 * The Cyrix 6x86 does not have any Pentium features 2699 * accessible while not at privilege level 0. 2700 */ 2701 if (x86_feature & X86_CPUID) { 2702 (void) strcat(tp, "pentium"); 2703 (void) strcat(tp, x86_feature & X86_MMX ? 2704 "+mmx pentium " : " "); 2705 } 2706 break; 2707 default: 2708 break; 2709 } 2710 (void) strcat(tp, "i486 i386 i86"); 2711 len = strlen(tp) + 1; /* account for NULL at end of string */ 2712 isa_list = strcpy(kmem_alloc(len, KM_SLEEP), tp); 2713 kmem_free(tp, TBUFSIZE); 2714 2715 #undef TBUFSIZE 2716 } 2717 2718 2719 #ifdef __amd64 2720 2721 void * 2722 device_arena_alloc(size_t size, int vm_flag) 2723 { 2724 return (vmem_alloc(device_arena, size, vm_flag)); 2725 } 2726 2727 void 2728 device_arena_free(void *vaddr, size_t size) 2729 { 2730 vmem_free(device_arena, vaddr, size); 2731 } 2732 2733 #else 2734 2735 void * 2736 device_arena_alloc(size_t size, int vm_flag) 2737 { 2738 caddr_t vaddr; 2739 uintptr_t v; 2740 size_t start; 2741 size_t end; 2742 2743 vaddr = vmem_alloc(heap_arena, size, vm_flag); 2744 if (vaddr == NULL) 2745 return (NULL); 2746 2747 v = (uintptr_t)vaddr; 2748 ASSERT(v >= kernelbase); 2749 ASSERT(v + size <= ptable_va); 2750 2751 start = btop(v - kernelbase); 2752 end = btop(v + size - 1 - kernelbase); 2753 ASSERT(start < toxic_bit_map_len); 2754 ASSERT(end < toxic_bit_map_len); 2755 2756 while (start <= end) { 2757 BT_ATOMIC_SET(toxic_bit_map, start); 2758 ++start; 2759 } 2760 return (vaddr); 2761 } 2762 2763 void 2764 device_arena_free(void *vaddr, size_t size) 2765 { 2766 uintptr_t v = (uintptr_t)vaddr; 2767 size_t start; 2768 size_t end; 2769 2770 ASSERT(v >= kernelbase); 2771 ASSERT(v + size <= ptable_va); 2772 2773 start = btop(v - kernelbase); 2774 end = btop(v + size - 1 - kernelbase); 2775 ASSERT(start < toxic_bit_map_len); 2776 ASSERT(end < toxic_bit_map_len); 2777 2778 while (start <= end) { 2779 ASSERT(BT_TEST(toxic_bit_map, start) != 0); 2780 BT_ATOMIC_CLEAR(toxic_bit_map, start); 2781 ++start; 2782 } 2783 vmem_free(heap_arena, vaddr, size); 2784 } 2785 2786 /* 2787 * returns 1st address in range that is in device arena, or NULL 2788 * if len is not NULL it returns the length of the toxic range 2789 */ 2790 void * 2791 device_arena_contains(void *vaddr, size_t size, size_t *len) 2792 { 2793 uintptr_t v = (uintptr_t)vaddr; 2794 uintptr_t eaddr = v + size; 2795 size_t start; 2796 size_t end; 2797 2798 /* 2799 * if called very early by kmdb, just return NULL 2800 */ 2801 if (toxic_bit_map == NULL) 2802 return (NULL); 2803 2804 /* 2805 * First check if we're completely outside the bitmap range. 2806 */ 2807 if (v >= ptable_va || eaddr < kernelbase) 2808 return (NULL); 2809 2810 /* 2811 * Trim ends of search to look at only what the bitmap covers. 2812 */ 2813 if (v < kernelbase) 2814 v = kernelbase; 2815 start = btop(v - kernelbase); 2816 end = btop(eaddr - kernelbase); 2817 if (end >= toxic_bit_map_len) 2818 end = toxic_bit_map_len; 2819 2820 if (bt_range(toxic_bit_map, &start, &end, end) == 0) 2821 return (NULL); 2822 2823 v = kernelbase + ptob(start); 2824 if (len != NULL) 2825 *len = ptob(end - start); 2826 return ((void *)v); 2827 } 2828 2829 #endif 2830