1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 /* 22 * Copyright (c) 1993, 2010, Oracle and/or its affiliates. All rights reserved. 23 * Copyright 2012 DEY Storage Systems, Inc. All rights reserved. 24 * Copyright 2013 Nexenta Systems, Inc. All rights reserved. 25 * Copyright 2015 Joyent, Inc. 26 * Copyright (c) 2015 by Delphix. All rights reserved. 27 */ 28 /* 29 * Copyright (c) 2010, Intel Corporation. 30 * All rights reserved. 31 */ 32 33 #include <sys/types.h> 34 #include <sys/t_lock.h> 35 #include <sys/param.h> 36 #include <sys/sysmacros.h> 37 #include <sys/signal.h> 38 #include <sys/systm.h> 39 #include <sys/user.h> 40 #include <sys/mman.h> 41 #include <sys/vm.h> 42 #include <sys/conf.h> 43 #include <sys/avintr.h> 44 #include <sys/autoconf.h> 45 #include <sys/disp.h> 46 #include <sys/class.h> 47 #include <sys/bitmap.h> 48 49 #include <sys/privregs.h> 50 51 #include <sys/proc.h> 52 #include <sys/buf.h> 53 #include <sys/kmem.h> 54 #include <sys/mem.h> 55 #include <sys/kstat.h> 56 57 #include <sys/reboot.h> 58 59 #include <sys/cred.h> 60 #include <sys/vnode.h> 61 #include <sys/file.h> 62 63 #include <sys/procfs.h> 64 65 #include <sys/vfs.h> 66 #include <sys/cmn_err.h> 67 #include <sys/utsname.h> 68 #include <sys/debug.h> 69 #include <sys/kdi.h> 70 71 #include <sys/dumphdr.h> 72 #include <sys/bootconf.h> 73 #include <sys/memlist_plat.h> 74 #include <sys/varargs.h> 75 #include <sys/promif.h> 76 #include <sys/modctl.h> 77 78 #include <sys/sunddi.h> 79 #include <sys/sunndi.h> 80 #include <sys/ndi_impldefs.h> 81 #include <sys/ddidmareq.h> 82 #include <sys/psw.h> 83 #include <sys/regset.h> 84 #include <sys/clock.h> 85 #include <sys/pte.h> 86 #include <sys/tss.h> 87 #include <sys/stack.h> 88 #include <sys/trap.h> 89 #include <sys/fp.h> 90 #include <vm/kboot_mmu.h> 91 #include <vm/anon.h> 92 #include <vm/as.h> 93 #include <vm/page.h> 94 #include <vm/seg.h> 95 #include <vm/seg_dev.h> 96 #include <vm/seg_kmem.h> 97 #include <vm/seg_kpm.h> 98 #include <vm/seg_map.h> 99 #include <vm/seg_vn.h> 100 #include <vm/seg_kp.h> 101 #include <sys/memnode.h> 102 #include <vm/vm_dep.h> 103 #include <sys/thread.h> 104 #include <sys/sysconf.h> 105 #include <sys/vm_machparam.h> 106 #include <sys/archsystm.h> 107 #include <sys/machsystm.h> 108 #include <vm/hat.h> 109 #include <vm/hat_i86.h> 110 #include <sys/pmem.h> 111 #include <sys/smp_impldefs.h> 112 #include <sys/x86_archext.h> 113 #include <sys/cpuvar.h> 114 #include <sys/segments.h> 115 #include <sys/clconf.h> 116 #include <sys/kobj.h> 117 #include <sys/kobj_lex.h> 118 #include <sys/cpc_impl.h> 119 #include <sys/cpu_module.h> 120 #include <sys/smbios.h> 121 #include <sys/debug_info.h> 122 #include <sys/bootinfo.h> 123 #include <sys/ddi_periodic.h> 124 #include <sys/systeminfo.h> 125 #include <sys/multiboot.h> 126 #include <sys/ramdisk.h> 127 128 #ifdef __xpv 129 130 #include <sys/hypervisor.h> 131 #include <sys/xen_mmu.h> 132 #include <sys/evtchn_impl.h> 133 #include <sys/gnttab.h> 134 #include <sys/xpv_panic.h> 135 #include <xen/sys/xenbus_comms.h> 136 #include <xen/public/physdev.h> 137 138 extern void xen_late_startup(void); 139 140 struct xen_evt_data cpu0_evt_data; 141 142 #else /* __xpv */ 143 #include <sys/memlist_impl.h> 144 145 extern void mem_config_init(void); 146 #endif /* __xpv */ 147 148 extern void progressbar_init(void); 149 extern void brand_init(void); 150 extern void pcf_init(void); 151 extern void pg_init(void); 152 extern void ssp_init(void); 153 154 extern int size_pse_array(pgcnt_t, int); 155 156 #if defined(_SOFT_HOSTID) 157 158 #include <sys/rtc.h> 159 160 static int32_t set_soft_hostid(void); 161 static char hostid_file[] = "/etc/hostid"; 162 163 #endif 164 165 void *gfx_devinfo_list; 166 167 #if defined(__amd64) && !defined(__xpv) 168 extern void immu_startup(void); 169 #endif 170 171 /* 172 * XXX make declaration below "static" when drivers no longer use this 173 * interface. 174 */ 175 extern caddr_t p0_va; /* Virtual address for accessing physical page 0 */ 176 177 /* 178 * segkp 179 */ 180 extern int segkp_fromheap; 181 182 static void kvm_init(void); 183 static void startup_init(void); 184 static void startup_memlist(void); 185 static void startup_kmem(void); 186 static void startup_modules(void); 187 static void startup_vm(void); 188 static void startup_end(void); 189 static void layout_kernel_va(void); 190 191 /* 192 * Declare these as initialized data so we can patch them. 193 */ 194 #ifdef __i386 195 196 /* 197 * Due to virtual address space limitations running in 32 bit mode, restrict 198 * the amount of physical memory configured to a max of PHYSMEM pages (16g). 199 * 200 * If the physical max memory size of 64g were allowed to be configured, the 201 * size of user virtual address space will be less than 1g. A limited user 202 * address space greatly reduces the range of applications that can run. 203 * 204 * If more physical memory than PHYSMEM is required, users should preferably 205 * run in 64 bit mode which has far looser virtual address space limitations. 206 * 207 * If 64 bit mode is not available (as in IA32) and/or more physical memory 208 * than PHYSMEM is required in 32 bit mode, physmem can be set to the desired 209 * value or to 0 (to configure all available memory) via eeprom(1M). kernelbase 210 * should also be carefully tuned to balance out the need of the user 211 * application while minimizing the risk of kernel heap exhaustion due to 212 * kernelbase being set too high. 213 */ 214 #define PHYSMEM 0x400000 215 216 #else /* __amd64 */ 217 218 /* 219 * For now we can handle memory with physical addresses up to about 220 * 64 Terabytes. This keeps the kernel above the VA hole, leaving roughly 221 * half the VA space for seg_kpm. When systems get bigger than 64TB this 222 * code will need revisiting. There is an implicit assumption that there 223 * are no *huge* holes in the physical address space too. 224 */ 225 #define TERABYTE (1ul << 40) 226 #define PHYSMEM_MAX64 mmu_btop(64 * TERABYTE) 227 #define PHYSMEM PHYSMEM_MAX64 228 #define AMD64_VA_HOLE_END 0xFFFF800000000000ul 229 230 #endif /* __amd64 */ 231 232 pgcnt_t physmem = PHYSMEM; 233 pgcnt_t obp_pages; /* Memory used by PROM for its text and data */ 234 235 char *kobj_file_buf; 236 int kobj_file_bufsize; /* set in /etc/system */ 237 238 /* Global variables for MP support. Used in mp_startup */ 239 caddr_t rm_platter_va = 0; 240 uint32_t rm_platter_pa; 241 242 int auto_lpg_disable = 1; 243 244 /* 245 * Some CPUs have holes in the middle of the 64-bit virtual address range. 246 */ 247 uintptr_t hole_start, hole_end; 248 249 /* 250 * kpm mapping window 251 */ 252 caddr_t kpm_vbase; 253 size_t kpm_size; 254 static int kpm_desired; 255 #ifdef __amd64 256 static uintptr_t segkpm_base = (uintptr_t)SEGKPM_BASE; 257 #endif 258 259 /* 260 * Configuration parameters set at boot time. 261 */ 262 263 caddr_t econtig; /* end of first block of contiguous kernel */ 264 265 struct bootops *bootops = 0; /* passed in from boot */ 266 struct bootops **bootopsp; 267 struct boot_syscalls *sysp; /* passed in from boot */ 268 269 char bootblock_fstype[16]; 270 271 char kern_bootargs[OBP_MAXPATHLEN]; 272 char kern_bootfile[OBP_MAXPATHLEN]; 273 274 /* 275 * ZFS zio segment. This allows us to exclude large portions of ZFS data that 276 * gets cached in kmem caches on the heap. If this is set to zero, we allocate 277 * zio buffers from their own segment, otherwise they are allocated from the 278 * heap. The optimization of allocating zio buffers from their own segment is 279 * only valid on 64-bit kernels. 280 */ 281 #if defined(__amd64) 282 int segzio_fromheap = 0; 283 #else 284 int segzio_fromheap = 1; 285 #endif 286 287 /* 288 * Give folks an escape hatch for disabling SMAP via kmdb. Doesn't work 289 * post-boot. 290 */ 291 int disable_smap = 0; 292 293 /* 294 * new memory fragmentations are possible in startup() due to BOP_ALLOCs. this 295 * depends on number of BOP_ALLOC calls made and requested size, memory size 296 * combination and whether boot.bin memory needs to be freed. 297 */ 298 #define POSS_NEW_FRAGMENTS 12 299 300 /* 301 * VM data structures 302 */ 303 long page_hashsz; /* Size of page hash table (power of two) */ 304 unsigned int page_hashsz_shift; /* log2(page_hashsz) */ 305 struct page *pp_base; /* Base of initial system page struct array */ 306 struct page **page_hash; /* Page hash table */ 307 pad_mutex_t *pse_mutex; /* Locks protecting pp->p_selock */ 308 size_t pse_table_size; /* Number of mutexes in pse_mutex[] */ 309 int pse_shift; /* log2(pse_table_size) */ 310 struct seg ktextseg; /* Segment used for kernel executable image */ 311 struct seg kvalloc; /* Segment used for "valloc" mapping */ 312 struct seg kpseg; /* Segment used for pageable kernel virt mem */ 313 struct seg kmapseg; /* Segment used for generic kernel mappings */ 314 struct seg kdebugseg; /* Segment used for the kernel debugger */ 315 316 struct seg *segkmap = &kmapseg; /* Kernel generic mapping segment */ 317 static struct seg *segmap = &kmapseg; /* easier to use name for in here */ 318 319 struct seg *segkp = &kpseg; /* Pageable kernel virtual memory segment */ 320 321 #if defined(__amd64) 322 struct seg kvseg_core; /* Segment used for the core heap */ 323 struct seg kpmseg; /* Segment used for physical mapping */ 324 struct seg *segkpm = &kpmseg; /* 64bit kernel physical mapping segment */ 325 #else 326 struct seg *segkpm = NULL; /* Unused on IA32 */ 327 #endif 328 329 caddr_t segkp_base; /* Base address of segkp */ 330 caddr_t segzio_base; /* Base address of segzio */ 331 #if defined(__amd64) 332 pgcnt_t segkpsize = btop(SEGKPDEFSIZE); /* size of segkp segment in pages */ 333 #else 334 pgcnt_t segkpsize = 0; 335 #endif 336 pgcnt_t segziosize = 0; /* size of zio segment in pages */ 337 338 /* 339 * A static DR page_t VA map is reserved that can map the page structures 340 * for a domain's entire RA space. The pages that back this space are 341 * dynamically allocated and need not be physically contiguous. The DR 342 * map size is derived from KPM size. 343 * This mechanism isn't used by x86 yet, so just stubs here. 344 */ 345 int ppvm_enable = 0; /* Static virtual map for page structs */ 346 page_t *ppvm_base = NULL; /* Base of page struct map */ 347 pgcnt_t ppvm_size = 0; /* Size of page struct map */ 348 349 /* 350 * VA range available to the debugger 351 */ 352 const caddr_t kdi_segdebugbase = (const caddr_t)SEGDEBUGBASE; 353 const size_t kdi_segdebugsize = SEGDEBUGSIZE; 354 355 struct memseg *memseg_base; 356 struct vnode unused_pages_vp; 357 358 #define FOURGB 0x100000000LL 359 360 struct memlist *memlist; 361 362 caddr_t s_text; /* start of kernel text segment */ 363 caddr_t e_text; /* end of kernel text segment */ 364 caddr_t s_data; /* start of kernel data segment */ 365 caddr_t e_data; /* end of kernel data segment */ 366 caddr_t modtext; /* start of loadable module text reserved */ 367 caddr_t e_modtext; /* end of loadable module text reserved */ 368 caddr_t moddata; /* start of loadable module data reserved */ 369 caddr_t e_moddata; /* end of loadable module data reserved */ 370 371 struct memlist *phys_install; /* Total installed physical memory */ 372 struct memlist *phys_avail; /* Total available physical memory */ 373 struct memlist *bios_rsvd; /* Bios reserved memory */ 374 375 /* 376 * kphysm_init returns the number of pages that were processed 377 */ 378 static pgcnt_t kphysm_init(page_t *, pgcnt_t); 379 380 #define IO_PROP_SIZE 64 /* device property size */ 381 382 /* 383 * a couple useful roundup macros 384 */ 385 #define ROUND_UP_PAGE(x) \ 386 ((uintptr_t)P2ROUNDUP((uintptr_t)(x), (uintptr_t)MMU_PAGESIZE)) 387 #define ROUND_UP_LPAGE(x) \ 388 ((uintptr_t)P2ROUNDUP((uintptr_t)(x), mmu.level_size[1])) 389 #define ROUND_UP_4MEG(x) \ 390 ((uintptr_t)P2ROUNDUP((uintptr_t)(x), (uintptr_t)FOUR_MEG)) 391 #define ROUND_UP_TOPLEVEL(x) \ 392 ((uintptr_t)P2ROUNDUP((uintptr_t)(x), mmu.level_size[mmu.max_level])) 393 394 /* 395 * 32-bit Kernel's Virtual memory layout. 396 * +-----------------------+ 397 * | | 398 * 0xFFC00000 -|-----------------------|- ARGSBASE 399 * | debugger | 400 * 0xFF800000 -|-----------------------|- SEGDEBUGBASE 401 * | Kernel Data | 402 * 0xFEC00000 -|-----------------------| 403 * | Kernel Text | 404 * 0xFE800000 -|-----------------------|- KERNEL_TEXT (0xFB400000 on Xen) 405 * |--- GDT ---|- GDT page (GDT_VA) 406 * |--- debug info ---|- debug info (DEBUG_INFO_VA) 407 * | | 408 * | page_t structures | 409 * | memsegs, memlists, | 410 * | page hash, etc. | 411 * --- -|-----------------------|- ekernelheap, valloc_base (floating) 412 * | | (segkp is just an arena in the heap) 413 * | | 414 * | kvseg | 415 * | | 416 * | | 417 * --- -|-----------------------|- kernelheap (floating) 418 * | Segkmap | 419 * 0xC3002000 -|-----------------------|- segmap_start (floating) 420 * | Red Zone | 421 * 0xC3000000 -|-----------------------|- kernelbase / userlimit (floating) 422 * | | || 423 * | Shared objects | \/ 424 * | | 425 * : : 426 * | user data | 427 * |-----------------------| 428 * | user text | 429 * 0x08048000 -|-----------------------| 430 * | user stack | 431 * : : 432 * | invalid | 433 * 0x00000000 +-----------------------+ 434 * 435 * 436 * 64-bit Kernel's Virtual memory layout. (assuming 64 bit app) 437 * +-----------------------+ 438 * | | 439 * 0xFFFFFFFF.FFC00000 |-----------------------|- ARGSBASE 440 * | debugger (?) | 441 * 0xFFFFFFFF.FF800000 |-----------------------|- SEGDEBUGBASE 442 * | unused | 443 * +-----------------------+ 444 * | Kernel Data | 445 * 0xFFFFFFFF.FBC00000 |-----------------------| 446 * | Kernel Text | 447 * 0xFFFFFFFF.FB800000 |-----------------------|- KERNEL_TEXT 448 * |--- GDT ---|- GDT page (GDT_VA) 449 * |--- debug info ---|- debug info (DEBUG_INFO_VA) 450 * | | 451 * | Core heap | (used for loadable modules) 452 * 0xFFFFFFFF.C0000000 |-----------------------|- core_base / ekernelheap 453 * | Kernel | 454 * | heap | 455 * 0xFFFFFXXX.XXX00000 |-----------------------|- kernelheap (floating) 456 * | segmap | 457 * 0xFFFFFXXX.XXX00000 |-----------------------|- segmap_start (floating) 458 * | device mappings | 459 * 0xFFFFFXXX.XXX00000 |-----------------------|- toxic_addr (floating) 460 * | segzio | 461 * 0xFFFFFXXX.XXX00000 |-----------------------|- segzio_base (floating) 462 * | segkp | 463 * --- |-----------------------|- segkp_base (floating) 464 * | page_t structures | valloc_base + valloc_sz 465 * | memsegs, memlists, | 466 * | page hash, etc. | 467 * 0xFFFFFF00.00000000 |-----------------------|- valloc_base (lower if >256GB) 468 * | segkpm | 469 * 0xFFFFFE00.00000000 |-----------------------| 470 * | Red Zone | 471 * 0xFFFFFD80.00000000 |-----------------------|- KERNELBASE (lower if >256GB) 472 * | User stack |- User space memory 473 * | | 474 * | shared objects, etc | (grows downwards) 475 * : : 476 * | | 477 * 0xFFFF8000.00000000 |-----------------------| 478 * | | 479 * | VA Hole / unused | 480 * | | 481 * 0x00008000.00000000 |-----------------------| 482 * | | 483 * | | 484 * : : 485 * | user heap | (grows upwards) 486 * | | 487 * | user data | 488 * |-----------------------| 489 * | user text | 490 * 0x00000000.04000000 |-----------------------| 491 * | invalid | 492 * 0x00000000.00000000 +-----------------------+ 493 * 494 * A 32 bit app on the 64 bit kernel sees the same layout as on the 32 bit 495 * kernel, except that userlimit is raised to 0xfe000000 496 * 497 * Floating values: 498 * 499 * valloc_base: start of the kernel's memory management/tracking data 500 * structures. This region contains page_t structures for 501 * physical memory, memsegs, memlists, and the page hash. 502 * 503 * core_base: start of the kernel's "core" heap area on 64-bit systems. 504 * This area is intended to be used for global data as well as for module 505 * text/data that does not fit into the nucleus pages. The core heap is 506 * restricted to a 2GB range, allowing every address within it to be 507 * accessed using rip-relative addressing 508 * 509 * ekernelheap: end of kernelheap and start of segmap. 510 * 511 * kernelheap: start of kernel heap. On 32-bit systems, this starts right 512 * above a red zone that separates the user's address space from the 513 * kernel's. On 64-bit systems, it sits above segkp and segkpm. 514 * 515 * segmap_start: start of segmap. The length of segmap can be modified 516 * through eeprom. The default length is 16MB on 32-bit systems and 64MB 517 * on 64-bit systems. 518 * 519 * kernelbase: On a 32-bit kernel the default value of 0xd4000000 will be 520 * decreased by 2X the size required for page_t. This allows the kernel 521 * heap to grow in size with physical memory. With sizeof(page_t) == 80 522 * bytes, the following shows the values of kernelbase and kernel heap 523 * sizes for different memory configurations (assuming default segmap and 524 * segkp sizes). 525 * 526 * mem size for kernelbase kernel heap 527 * size page_t's size 528 * ---- --------- ---------- ----------- 529 * 1gb 0x01400000 0xd1800000 684MB 530 * 2gb 0x02800000 0xcf000000 704MB 531 * 4gb 0x05000000 0xca000000 744MB 532 * 6gb 0x07800000 0xc5000000 784MB 533 * 8gb 0x0a000000 0xc0000000 824MB 534 * 16gb 0x14000000 0xac000000 984MB 535 * 32gb 0x28000000 0x84000000 1304MB 536 * 64gb 0x50000000 0x34000000 1944MB (*) 537 * 538 * kernelbase is less than the abi minimum of 0xc0000000 for memory 539 * configurations above 8gb. 540 * 541 * (*) support for memory configurations above 32gb will require manual tuning 542 * of kernelbase to balance out the need of user applications. 543 */ 544 545 /* real-time-clock initialization parameters */ 546 extern time_t process_rtc_config_file(void); 547 548 uintptr_t kernelbase; 549 uintptr_t postbootkernelbase; /* not set till boot loader is gone */ 550 uintptr_t eprom_kernelbase; 551 size_t segmapsize; 552 uintptr_t segmap_start; 553 int segmapfreelists; 554 pgcnt_t npages; 555 pgcnt_t orig_npages; 556 size_t core_size; /* size of "core" heap */ 557 uintptr_t core_base; /* base address of "core" heap */ 558 559 /* 560 * List of bootstrap pages. We mark these as allocated in startup. 561 * release_bootstrap() will free them when we're completely done with 562 * the bootstrap. 563 */ 564 static page_t *bootpages; 565 566 /* 567 * boot time pages that have a vnode from the ramdisk will keep that forever. 568 */ 569 static page_t *rd_pages; 570 571 /* 572 * Lower 64K 573 */ 574 static page_t *lower_pages = NULL; 575 static int lower_pages_count = 0; 576 577 struct system_hardware system_hardware; 578 579 /* 580 * Enable some debugging messages concerning memory usage... 581 */ 582 static void 583 print_memlist(char *title, struct memlist *mp) 584 { 585 prom_printf("MEMLIST: %s:\n", title); 586 while (mp != NULL) { 587 prom_printf("\tAddress 0x%" PRIx64 ", size 0x%" PRIx64 "\n", 588 mp->ml_address, mp->ml_size); 589 mp = mp->ml_next; 590 } 591 } 592 593 /* 594 * XX64 need a comment here.. are these just default values, surely 595 * we read the "cpuid" type information to figure this out. 596 */ 597 int l2cache_sz = 0x80000; 598 int l2cache_linesz = 0x40; 599 int l2cache_assoc = 1; 600 601 static size_t textrepl_min_gb = 10; 602 603 /* 604 * on 64 bit we use a predifined VA range for mapping devices in the kernel 605 * on 32 bit the mappings are intermixed in the heap, so we use a bit map 606 */ 607 #ifdef __amd64 608 609 vmem_t *device_arena; 610 uintptr_t toxic_addr = (uintptr_t)NULL; 611 size_t toxic_size = 1024 * 1024 * 1024; /* Sparc uses 1 gig too */ 612 613 #else /* __i386 */ 614 615 ulong_t *toxic_bit_map; /* one bit for each 4k of VA in heap_arena */ 616 size_t toxic_bit_map_len = 0; /* in bits */ 617 618 #endif /* __i386 */ 619 620 /* 621 * Simple boot time debug facilities 622 */ 623 static char *prm_dbg_str[] = { 624 "%s:%d: '%s' is 0x%x\n", 625 "%s:%d: '%s' is 0x%llx\n" 626 }; 627 628 int prom_debug; 629 630 #define PRM_DEBUG(q) if (prom_debug) \ 631 prom_printf(prm_dbg_str[sizeof (q) >> 3], "startup.c", __LINE__, #q, q); 632 #define PRM_POINT(q) if (prom_debug) \ 633 prom_printf("%s:%d: %s\n", "startup.c", __LINE__, q); 634 635 /* 636 * This structure is used to keep track of the intial allocations 637 * done in startup_memlist(). The value of NUM_ALLOCATIONS needs to 638 * be >= the number of ADD_TO_ALLOCATIONS() executed in the code. 639 */ 640 #define NUM_ALLOCATIONS 8 641 int num_allocations = 0; 642 struct { 643 void **al_ptr; 644 size_t al_size; 645 } allocations[NUM_ALLOCATIONS]; 646 size_t valloc_sz = 0; 647 uintptr_t valloc_base; 648 649 #define ADD_TO_ALLOCATIONS(ptr, size) { \ 650 size = ROUND_UP_PAGE(size); \ 651 if (num_allocations == NUM_ALLOCATIONS) \ 652 panic("too many ADD_TO_ALLOCATIONS()"); \ 653 allocations[num_allocations].al_ptr = (void**)&ptr; \ 654 allocations[num_allocations].al_size = size; \ 655 valloc_sz += size; \ 656 ++num_allocations; \ 657 } 658 659 /* 660 * Allocate all the initial memory needed by the page allocator. 661 */ 662 static void 663 perform_allocations(void) 664 { 665 caddr_t mem; 666 int i; 667 int valloc_align; 668 669 PRM_DEBUG(valloc_base); 670 PRM_DEBUG(valloc_sz); 671 valloc_align = mmu.level_size[mmu.max_page_level > 0]; 672 mem = BOP_ALLOC(bootops, (caddr_t)valloc_base, valloc_sz, valloc_align); 673 if (mem != (caddr_t)valloc_base) 674 panic("BOP_ALLOC() failed"); 675 bzero(mem, valloc_sz); 676 for (i = 0; i < num_allocations; ++i) { 677 *allocations[i].al_ptr = (void *)mem; 678 mem += allocations[i].al_size; 679 } 680 } 681 682 /* 683 * Set up and enable SMAP now before we start other CPUs, but after the kernel's 684 * VM has been set up so we can use hot_patch_kernel_text(). 685 * 686 * We can only patch 1, 2, or 4 bytes, but not three bytes. So instead, we 687 * replace the four byte word at the patch point. See uts/intel/ia32/ml/copy.s 688 * for more information on what's going on here. 689 */ 690 static void 691 startup_smap(void) 692 { 693 int i; 694 uint32_t inst; 695 uint8_t *instp; 696 char sym[128]; 697 698 extern int _smap_enable_patch_count; 699 extern int _smap_disable_patch_count; 700 701 if (disable_smap != 0) 702 remove_x86_feature(x86_featureset, X86FSET_SMAP); 703 704 if (is_x86_feature(x86_featureset, X86FSET_SMAP) == B_FALSE) 705 return; 706 707 for (i = 0; i < _smap_enable_patch_count; i++) { 708 int sizep; 709 710 VERIFY3U(i, <, _smap_enable_patch_count); 711 VERIFY(snprintf(sym, sizeof (sym), "_smap_enable_patch_%d", i) < 712 sizeof (sym)); 713 instp = (uint8_t *)(void *)kobj_getelfsym(sym, NULL, &sizep); 714 VERIFY(instp != 0); 715 inst = (instp[3] << 24) | (SMAP_CLAC_INSTR & 0x00ffffff); 716 hot_patch_kernel_text((caddr_t)instp, inst, 4); 717 } 718 719 for (i = 0; i < _smap_disable_patch_count; i++) { 720 int sizep; 721 722 VERIFY(snprintf(sym, sizeof (sym), "_smap_disable_patch_%d", 723 i) < sizeof (sym)); 724 instp = (uint8_t *)(void *)kobj_getelfsym(sym, NULL, &sizep); 725 VERIFY(instp != 0); 726 inst = (instp[3] << 24) | (SMAP_STAC_INSTR & 0x00ffffff); 727 hot_patch_kernel_text((caddr_t)instp, inst, 4); 728 } 729 730 hot_patch_kernel_text((caddr_t)smap_enable, SMAP_CLAC_INSTR, 4); 731 hot_patch_kernel_text((caddr_t)smap_disable, SMAP_STAC_INSTR, 4); 732 setcr4(getcr4() | CR4_SMAP); 733 smap_enable(); 734 } 735 736 /* 737 * Our world looks like this at startup time. 738 * 739 * In a 32-bit OS, boot loads the kernel text at 0xfe800000 and kernel data 740 * at 0xfec00000. On a 64-bit OS, kernel text and data are loaded at 741 * 0xffffffff.fe800000 and 0xffffffff.fec00000 respectively. Those 742 * addresses are fixed in the binary at link time. 743 * 744 * On the text page: 745 * unix/genunix/krtld/module text loads. 746 * 747 * On the data page: 748 * unix/genunix/krtld/module data loads. 749 * 750 * Machine-dependent startup code 751 */ 752 void 753 startup(void) 754 { 755 #if !defined(__xpv) 756 extern void startup_pci_bios(void); 757 #endif 758 extern cpuset_t cpu_ready_set; 759 760 /* 761 * Make sure that nobody tries to use sekpm until we have 762 * initialized it properly. 763 */ 764 #if defined(__amd64) 765 kpm_desired = 1; 766 #endif 767 kpm_enable = 0; 768 CPUSET_ONLY(cpu_ready_set, 0); /* cpu 0 is boot cpu */ 769 770 #if defined(__xpv) /* XXPV fix me! */ 771 { 772 extern int segvn_use_regions; 773 segvn_use_regions = 0; 774 } 775 #endif 776 ssp_init(); 777 progressbar_init(); 778 startup_init(); 779 #if defined(__xpv) 780 startup_xen_version(); 781 #endif 782 startup_memlist(); 783 startup_kmem(); 784 startup_vm(); 785 #if !defined(__xpv) 786 /* 787 * Note we need to do this even on fast reboot in order to access 788 * the irq routing table (used for pci labels). 789 */ 790 startup_pci_bios(); 791 startup_smap(); 792 #endif 793 #if defined(__xpv) 794 startup_xen_mca(); 795 #endif 796 startup_modules(); 797 798 startup_end(); 799 } 800 801 static void 802 startup_init() 803 { 804 PRM_POINT("startup_init() starting..."); 805 806 /* 807 * Complete the extraction of cpuid data 808 */ 809 cpuid_pass2(CPU); 810 811 (void) check_boot_version(BOP_GETVERSION(bootops)); 812 813 /* 814 * Check for prom_debug in boot environment 815 */ 816 if (BOP_GETPROPLEN(bootops, "prom_debug") >= 0) { 817 ++prom_debug; 818 PRM_POINT("prom_debug found in boot enviroment"); 819 } 820 821 /* 822 * Collect node, cpu and memory configuration information. 823 */ 824 get_system_configuration(); 825 826 /* 827 * Halt if this is an unsupported processor. 828 */ 829 if (x86_type == X86_TYPE_486 || x86_type == X86_TYPE_CYRIX_486) { 830 printf("\n486 processor (\"%s\") detected.\n", 831 CPU->cpu_brandstr); 832 halt("This processor is not supported by this release " 833 "of Solaris."); 834 } 835 836 PRM_POINT("startup_init() done"); 837 } 838 839 /* 840 * Callback for copy_memlist_filter() to filter nucleus, kadb/kmdb, (ie. 841 * everything mapped above KERNEL_TEXT) pages from phys_avail. Note it 842 * also filters out physical page zero. There is some reliance on the 843 * boot loader allocating only a few contiguous physical memory chunks. 844 */ 845 static void 846 avail_filter(uint64_t *addr, uint64_t *size) 847 { 848 uintptr_t va; 849 uintptr_t next_va; 850 pfn_t pfn; 851 uint64_t pfn_addr; 852 uint64_t pfn_eaddr; 853 uint_t prot; 854 size_t len; 855 uint_t change; 856 857 if (prom_debug) 858 prom_printf("\tFilter: in: a=%" PRIx64 ", s=%" PRIx64 "\n", 859 *addr, *size); 860 861 /* 862 * page zero is required for BIOS.. never make it available 863 */ 864 if (*addr == 0) { 865 *addr += MMU_PAGESIZE; 866 *size -= MMU_PAGESIZE; 867 } 868 869 /* 870 * First we trim from the front of the range. Since kbm_probe() 871 * walks ranges in virtual order, but addr/size are physical, we need 872 * to the list until no changes are seen. This deals with the case 873 * where page "p" is mapped at v, page "p + PAGESIZE" is mapped at w 874 * but w < v. 875 */ 876 do { 877 change = 0; 878 for (va = KERNEL_TEXT; 879 *size > 0 && kbm_probe(&va, &len, &pfn, &prot) != 0; 880 va = next_va) { 881 882 next_va = va + len; 883 pfn_addr = pfn_to_pa(pfn); 884 pfn_eaddr = pfn_addr + len; 885 886 if (pfn_addr <= *addr && pfn_eaddr > *addr) { 887 change = 1; 888 while (*size > 0 && len > 0) { 889 *addr += MMU_PAGESIZE; 890 *size -= MMU_PAGESIZE; 891 len -= MMU_PAGESIZE; 892 } 893 } 894 } 895 if (change && prom_debug) 896 prom_printf("\t\ttrim: a=%" PRIx64 ", s=%" PRIx64 "\n", 897 *addr, *size); 898 } while (change); 899 900 /* 901 * Trim pages from the end of the range. 902 */ 903 for (va = KERNEL_TEXT; 904 *size > 0 && kbm_probe(&va, &len, &pfn, &prot) != 0; 905 va = next_va) { 906 907 next_va = va + len; 908 pfn_addr = pfn_to_pa(pfn); 909 910 if (pfn_addr >= *addr && pfn_addr < *addr + *size) 911 *size = pfn_addr - *addr; 912 } 913 914 if (prom_debug) 915 prom_printf("\tFilter out: a=%" PRIx64 ", s=%" PRIx64 "\n", 916 *addr, *size); 917 } 918 919 static void 920 kpm_init() 921 { 922 struct segkpm_crargs b; 923 924 /* 925 * These variables were all designed for sfmmu in which segkpm is 926 * mapped using a single pagesize - either 8KB or 4MB. On x86, we 927 * might use 2+ page sizes on a single machine, so none of these 928 * variables have a single correct value. They are set up as if we 929 * always use a 4KB pagesize, which should do no harm. In the long 930 * run, we should get rid of KPM's assumption that only a single 931 * pagesize is used. 932 */ 933 kpm_pgshft = MMU_PAGESHIFT; 934 kpm_pgsz = MMU_PAGESIZE; 935 kpm_pgoff = MMU_PAGEOFFSET; 936 kpmp2pshft = 0; 937 kpmpnpgs = 1; 938 ASSERT(((uintptr_t)kpm_vbase & (kpm_pgsz - 1)) == 0); 939 940 PRM_POINT("about to create segkpm"); 941 rw_enter(&kas.a_lock, RW_WRITER); 942 943 if (seg_attach(&kas, kpm_vbase, kpm_size, segkpm) < 0) 944 panic("cannot attach segkpm"); 945 946 b.prot = PROT_READ | PROT_WRITE; 947 b.nvcolors = 1; 948 949 if (segkpm_create(segkpm, (caddr_t)&b) != 0) 950 panic("segkpm_create segkpm"); 951 952 rw_exit(&kas.a_lock); 953 } 954 955 /* 956 * The debug info page provides enough information to allow external 957 * inspectors (e.g. when running under a hypervisor) to bootstrap 958 * themselves into allowing full-blown kernel debugging. 959 */ 960 static void 961 init_debug_info(void) 962 { 963 caddr_t mem; 964 debug_info_t *di; 965 966 #ifndef __lint 967 ASSERT(sizeof (debug_info_t) < MMU_PAGESIZE); 968 #endif 969 970 mem = BOP_ALLOC(bootops, (caddr_t)DEBUG_INFO_VA, MMU_PAGESIZE, 971 MMU_PAGESIZE); 972 973 if (mem != (caddr_t)DEBUG_INFO_VA) 974 panic("BOP_ALLOC() failed"); 975 bzero(mem, MMU_PAGESIZE); 976 977 di = (debug_info_t *)mem; 978 979 di->di_magic = DEBUG_INFO_MAGIC; 980 di->di_version = DEBUG_INFO_VERSION; 981 di->di_modules = (uintptr_t)&modules; 982 di->di_s_text = (uintptr_t)s_text; 983 di->di_e_text = (uintptr_t)e_text; 984 di->di_s_data = (uintptr_t)s_data; 985 di->di_e_data = (uintptr_t)e_data; 986 di->di_hat_htable_off = offsetof(hat_t, hat_htable); 987 di->di_ht_pfn_off = offsetof(htable_t, ht_pfn); 988 } 989 990 /* 991 * Build the memlists and other kernel essential memory system data structures. 992 * This is everything at valloc_base. 993 */ 994 static void 995 startup_memlist(void) 996 { 997 size_t memlist_sz; 998 size_t memseg_sz; 999 size_t pagehash_sz; 1000 size_t pp_sz; 1001 uintptr_t va; 1002 size_t len; 1003 uint_t prot; 1004 pfn_t pfn; 1005 int memblocks; 1006 pfn_t rsvd_high_pfn; 1007 pgcnt_t rsvd_pgcnt; 1008 size_t rsvdmemlist_sz; 1009 int rsvdmemblocks; 1010 caddr_t pagecolor_mem; 1011 size_t pagecolor_memsz; 1012 caddr_t page_ctrs_mem; 1013 size_t page_ctrs_size; 1014 size_t pse_table_alloc_size; 1015 struct memlist *current; 1016 extern void startup_build_mem_nodes(struct memlist *); 1017 1018 /* XX64 fix these - they should be in include files */ 1019 extern size_t page_coloring_init(uint_t, int, int); 1020 extern void page_coloring_setup(caddr_t); 1021 1022 PRM_POINT("startup_memlist() starting..."); 1023 1024 /* 1025 * Use leftover large page nucleus text/data space for loadable modules. 1026 * Use at most MODTEXT/MODDATA. 1027 */ 1028 len = kbm_nucleus_size; 1029 ASSERT(len > MMU_PAGESIZE); 1030 1031 moddata = (caddr_t)ROUND_UP_PAGE(e_data); 1032 e_moddata = (caddr_t)P2ROUNDUP((uintptr_t)e_data, (uintptr_t)len); 1033 if (e_moddata - moddata > MODDATA) 1034 e_moddata = moddata + MODDATA; 1035 1036 modtext = (caddr_t)ROUND_UP_PAGE(e_text); 1037 e_modtext = (caddr_t)P2ROUNDUP((uintptr_t)e_text, (uintptr_t)len); 1038 if (e_modtext - modtext > MODTEXT) 1039 e_modtext = modtext + MODTEXT; 1040 1041 econtig = e_moddata; 1042 1043 PRM_DEBUG(modtext); 1044 PRM_DEBUG(e_modtext); 1045 PRM_DEBUG(moddata); 1046 PRM_DEBUG(e_moddata); 1047 PRM_DEBUG(econtig); 1048 1049 /* 1050 * Examine the boot loader physical memory map to find out: 1051 * - total memory in system - physinstalled 1052 * - the max physical address - physmax 1053 * - the number of discontiguous segments of memory. 1054 */ 1055 if (prom_debug) 1056 print_memlist("boot physinstalled", 1057 bootops->boot_mem->physinstalled); 1058 installed_top_size_ex(bootops->boot_mem->physinstalled, &physmax, 1059 &physinstalled, &memblocks); 1060 PRM_DEBUG(physmax); 1061 PRM_DEBUG(physinstalled); 1062 PRM_DEBUG(memblocks); 1063 1064 /* 1065 * Compute maximum physical address for memory DR operations. 1066 * Memory DR operations are unsupported on xpv or 32bit OSes. 1067 */ 1068 #ifdef __amd64 1069 if (plat_dr_support_memory()) { 1070 if (plat_dr_physmax == 0) { 1071 uint_t pabits = UINT_MAX; 1072 1073 cpuid_get_addrsize(CPU, &pabits, NULL); 1074 plat_dr_physmax = btop(1ULL << pabits); 1075 } 1076 if (plat_dr_physmax > PHYSMEM_MAX64) 1077 plat_dr_physmax = PHYSMEM_MAX64; 1078 } else 1079 #endif 1080 plat_dr_physmax = 0; 1081 1082 /* 1083 * Examine the bios reserved memory to find out: 1084 * - the number of discontiguous segments of memory. 1085 */ 1086 if (prom_debug) 1087 print_memlist("boot reserved mem", 1088 bootops->boot_mem->rsvdmem); 1089 installed_top_size_ex(bootops->boot_mem->rsvdmem, &rsvd_high_pfn, 1090 &rsvd_pgcnt, &rsvdmemblocks); 1091 PRM_DEBUG(rsvd_high_pfn); 1092 PRM_DEBUG(rsvd_pgcnt); 1093 PRM_DEBUG(rsvdmemblocks); 1094 1095 /* 1096 * Initialize hat's mmu parameters. 1097 * Check for enforce-prot-exec in boot environment. It's used to 1098 * enable/disable support for the page table entry NX bit. 1099 * The default is to enforce PROT_EXEC on processors that support NX. 1100 * Boot seems to round up the "len", but 8 seems to be big enough. 1101 */ 1102 mmu_init(); 1103 1104 #ifdef __i386 1105 /* 1106 * physmax is lowered if there is more memory than can be 1107 * physically addressed in 32 bit (PAE/non-PAE) modes. 1108 */ 1109 if (mmu.pae_hat) { 1110 if (PFN_ABOVE64G(physmax)) { 1111 physinstalled -= (physmax - (PFN_64G - 1)); 1112 physmax = PFN_64G - 1; 1113 } 1114 } else { 1115 if (PFN_ABOVE4G(physmax)) { 1116 physinstalled -= (physmax - (PFN_4G - 1)); 1117 physmax = PFN_4G - 1; 1118 } 1119 } 1120 #endif 1121 1122 startup_build_mem_nodes(bootops->boot_mem->physinstalled); 1123 1124 if (BOP_GETPROPLEN(bootops, "enforce-prot-exec") >= 0) { 1125 int len = BOP_GETPROPLEN(bootops, "enforce-prot-exec"); 1126 char value[8]; 1127 1128 if (len < 8) 1129 (void) BOP_GETPROP(bootops, "enforce-prot-exec", value); 1130 else 1131 (void) strcpy(value, ""); 1132 if (strcmp(value, "off") == 0) 1133 mmu.pt_nx = 0; 1134 } 1135 PRM_DEBUG(mmu.pt_nx); 1136 1137 /* 1138 * We will need page_t's for every page in the system, except for 1139 * memory mapped at or above above the start of the kernel text segment. 1140 * 1141 * pages above e_modtext are attributed to kernel debugger (obp_pages) 1142 */ 1143 npages = physinstalled - 1; /* avail_filter() skips page 0, so "- 1" */ 1144 obp_pages = 0; 1145 va = KERNEL_TEXT; 1146 while (kbm_probe(&va, &len, &pfn, &prot) != 0) { 1147 npages -= len >> MMU_PAGESHIFT; 1148 if (va >= (uintptr_t)e_moddata) 1149 obp_pages += len >> MMU_PAGESHIFT; 1150 va += len; 1151 } 1152 PRM_DEBUG(npages); 1153 PRM_DEBUG(obp_pages); 1154 1155 /* 1156 * If physmem is patched to be non-zero, use it instead of the computed 1157 * value unless it is larger than the actual amount of memory on hand. 1158 */ 1159 if (physmem == 0 || physmem > npages) { 1160 physmem = npages; 1161 } else if (physmem < npages) { 1162 orig_npages = npages; 1163 npages = physmem; 1164 } 1165 PRM_DEBUG(physmem); 1166 1167 /* 1168 * We now compute the sizes of all the initial allocations for 1169 * structures the kernel needs in order do kmem_alloc(). These 1170 * include: 1171 * memsegs 1172 * memlists 1173 * page hash table 1174 * page_t's 1175 * page coloring data structs 1176 */ 1177 memseg_sz = sizeof (struct memseg) * (memblocks + POSS_NEW_FRAGMENTS); 1178 ADD_TO_ALLOCATIONS(memseg_base, memseg_sz); 1179 PRM_DEBUG(memseg_sz); 1180 1181 /* 1182 * Reserve space for memlists. There's no real good way to know exactly 1183 * how much room we'll need, but this should be a good upper bound. 1184 */ 1185 memlist_sz = ROUND_UP_PAGE(2 * sizeof (struct memlist) * 1186 (memblocks + POSS_NEW_FRAGMENTS)); 1187 ADD_TO_ALLOCATIONS(memlist, memlist_sz); 1188 PRM_DEBUG(memlist_sz); 1189 1190 /* 1191 * Reserve space for bios reserved memlists. 1192 */ 1193 rsvdmemlist_sz = ROUND_UP_PAGE(2 * sizeof (struct memlist) * 1194 (rsvdmemblocks + POSS_NEW_FRAGMENTS)); 1195 ADD_TO_ALLOCATIONS(bios_rsvd, rsvdmemlist_sz); 1196 PRM_DEBUG(rsvdmemlist_sz); 1197 1198 /* LINTED */ 1199 ASSERT(P2SAMEHIGHBIT((1 << PP_SHIFT), sizeof (struct page))); 1200 /* 1201 * The page structure hash table size is a power of 2 1202 * such that the average hash chain length is PAGE_HASHAVELEN. 1203 */ 1204 page_hashsz = npages / PAGE_HASHAVELEN; 1205 page_hashsz_shift = highbit(page_hashsz); 1206 page_hashsz = 1 << page_hashsz_shift; 1207 pagehash_sz = sizeof (struct page *) * page_hashsz; 1208 ADD_TO_ALLOCATIONS(page_hash, pagehash_sz); 1209 PRM_DEBUG(pagehash_sz); 1210 1211 /* 1212 * Set aside room for the page structures themselves. 1213 */ 1214 PRM_DEBUG(npages); 1215 pp_sz = sizeof (struct page) * npages; 1216 ADD_TO_ALLOCATIONS(pp_base, pp_sz); 1217 PRM_DEBUG(pp_sz); 1218 1219 /* 1220 * determine l2 cache info and memory size for page coloring 1221 */ 1222 (void) getl2cacheinfo(CPU, 1223 &l2cache_sz, &l2cache_linesz, &l2cache_assoc); 1224 pagecolor_memsz = 1225 page_coloring_init(l2cache_sz, l2cache_linesz, l2cache_assoc); 1226 ADD_TO_ALLOCATIONS(pagecolor_mem, pagecolor_memsz); 1227 PRM_DEBUG(pagecolor_memsz); 1228 1229 page_ctrs_size = page_ctrs_sz(); 1230 ADD_TO_ALLOCATIONS(page_ctrs_mem, page_ctrs_size); 1231 PRM_DEBUG(page_ctrs_size); 1232 1233 /* 1234 * Allocate the array that protects pp->p_selock. 1235 */ 1236 pse_shift = size_pse_array(physmem, max_ncpus); 1237 pse_table_size = 1 << pse_shift; 1238 pse_table_alloc_size = pse_table_size * sizeof (pad_mutex_t); 1239 ADD_TO_ALLOCATIONS(pse_mutex, pse_table_alloc_size); 1240 1241 #if defined(__amd64) 1242 valloc_sz = ROUND_UP_LPAGE(valloc_sz); 1243 valloc_base = VALLOC_BASE; 1244 1245 /* 1246 * The default values of VALLOC_BASE and SEGKPM_BASE should work 1247 * for values of physmax up to 256GB (1/4 TB). They need adjusting when 1248 * memory is at addresses above 256GB. When adjusted, segkpm_base must 1249 * be aligned on KERNEL_REDZONE_SIZE boundary (span of top level pte). 1250 * 1251 * In the general case (>256GB), we use (4 * physmem) for the 1252 * kernel's virtual addresses, which is divided approximately 1253 * as follows: 1254 * - 1 * physmem for segkpm 1255 * - 1.5 * physmem for segzio 1256 * - 1.5 * physmem for heap 1257 * Total: 4.0 * physmem 1258 * 1259 * Note that the segzio and heap sizes are more than physmem so that 1260 * VA fragmentation does not prevent either of them from being 1261 * able to use nearly all of physmem. The value of 1.5x is determined 1262 * experimentally and may need to change if the workload changes. 1263 */ 1264 if (physmax + 1 > mmu_btop(TERABYTE / 4) || 1265 plat_dr_physmax > mmu_btop(TERABYTE / 4)) { 1266 uint64_t kpm_resv_amount = mmu_ptob(physmax + 1); 1267 1268 if (kpm_resv_amount < mmu_ptob(plat_dr_physmax)) { 1269 kpm_resv_amount = mmu_ptob(plat_dr_physmax); 1270 } 1271 1272 /* 1273 * This is what actually controls the KVA : UVA split. 1274 * The kernel uses high VA, and this is lowering the 1275 * boundary, thus increasing the amount of VA for the kernel. 1276 * This gives the kernel 4 * (amount of physical memory) VA. 1277 * 1278 * The maximum VA is UINT64_MAX and we are using 1279 * 64-bit 2's complement math, so e.g. if you have 512GB 1280 * of memory, segkpm_base = -(4 * 512GB) == -2TB == 1281 * UINT64_MAX - 2TB (approximately). So the kernel's 1282 * VA is [UINT64_MAX-2TB to UINT64_MAX]. 1283 */ 1284 segkpm_base = -(P2ROUNDUP((4 * kpm_resv_amount), 1285 KERNEL_REDZONE_SIZE)); 1286 1287 /* make sure we leave some space for user apps above hole */ 1288 segkpm_base = MAX(segkpm_base, AMD64_VA_HOLE_END + TERABYTE); 1289 if (segkpm_base > SEGKPM_BASE) 1290 segkpm_base = SEGKPM_BASE; 1291 PRM_DEBUG(segkpm_base); 1292 1293 valloc_base = segkpm_base + P2ROUNDUP(kpm_resv_amount, ONE_GIG); 1294 if (valloc_base < segkpm_base) 1295 panic("not enough kernel VA to support memory size"); 1296 PRM_DEBUG(valloc_base); 1297 } 1298 #else /* __i386 */ 1299 valloc_base = (uintptr_t)(MISC_VA_BASE - valloc_sz); 1300 valloc_base = P2ALIGN(valloc_base, mmu.level_size[1]); 1301 PRM_DEBUG(valloc_base); 1302 #endif /* __i386 */ 1303 1304 /* 1305 * do all the initial allocations 1306 */ 1307 perform_allocations(); 1308 1309 /* 1310 * Build phys_install and phys_avail in kernel memspace. 1311 * - phys_install should be all memory in the system. 1312 * - phys_avail is phys_install minus any memory mapped before this 1313 * point above KERNEL_TEXT. 1314 */ 1315 current = phys_install = memlist; 1316 copy_memlist_filter(bootops->boot_mem->physinstalled, ¤t, NULL); 1317 if ((caddr_t)current > (caddr_t)memlist + memlist_sz) 1318 panic("physinstalled was too big!"); 1319 if (prom_debug) 1320 print_memlist("phys_install", phys_install); 1321 1322 phys_avail = current; 1323 PRM_POINT("Building phys_avail:\n"); 1324 copy_memlist_filter(bootops->boot_mem->physinstalled, ¤t, 1325 avail_filter); 1326 if ((caddr_t)current > (caddr_t)memlist + memlist_sz) 1327 panic("physavail was too big!"); 1328 if (prom_debug) 1329 print_memlist("phys_avail", phys_avail); 1330 #ifndef __xpv 1331 /* 1332 * Free unused memlist items, which may be used by memory DR driver 1333 * at runtime. 1334 */ 1335 if ((caddr_t)current < (caddr_t)memlist + memlist_sz) { 1336 memlist_free_block((caddr_t)current, 1337 (caddr_t)memlist + memlist_sz - (caddr_t)current); 1338 } 1339 #endif 1340 1341 /* 1342 * Build bios reserved memspace 1343 */ 1344 current = bios_rsvd; 1345 copy_memlist_filter(bootops->boot_mem->rsvdmem, ¤t, NULL); 1346 if ((caddr_t)current > (caddr_t)bios_rsvd + rsvdmemlist_sz) 1347 panic("bios_rsvd was too big!"); 1348 if (prom_debug) 1349 print_memlist("bios_rsvd", bios_rsvd); 1350 #ifndef __xpv 1351 /* 1352 * Free unused memlist items, which may be used by memory DR driver 1353 * at runtime. 1354 */ 1355 if ((caddr_t)current < (caddr_t)bios_rsvd + rsvdmemlist_sz) { 1356 memlist_free_block((caddr_t)current, 1357 (caddr_t)bios_rsvd + rsvdmemlist_sz - (caddr_t)current); 1358 } 1359 #endif 1360 1361 /* 1362 * setup page coloring 1363 */ 1364 page_coloring_setup(pagecolor_mem); 1365 page_lock_init(); /* currently a no-op */ 1366 1367 /* 1368 * free page list counters 1369 */ 1370 (void) page_ctrs_alloc(page_ctrs_mem); 1371 1372 /* 1373 * Size the pcf array based on the number of cpus in the box at 1374 * boot time. 1375 */ 1376 1377 pcf_init(); 1378 1379 /* 1380 * Initialize the page structures from the memory lists. 1381 */ 1382 availrmem_initial = availrmem = freemem = 0; 1383 PRM_POINT("Calling kphysm_init()..."); 1384 npages = kphysm_init(pp_base, npages); 1385 PRM_POINT("kphysm_init() done"); 1386 PRM_DEBUG(npages); 1387 1388 init_debug_info(); 1389 1390 /* 1391 * Now that page_t's have been initialized, remove all the 1392 * initial allocation pages from the kernel free page lists. 1393 */ 1394 boot_mapin((caddr_t)valloc_base, valloc_sz); 1395 boot_mapin((caddr_t)MISC_VA_BASE, MISC_VA_SIZE); 1396 PRM_POINT("startup_memlist() done"); 1397 1398 PRM_DEBUG(valloc_sz); 1399 1400 #if defined(__amd64) 1401 if ((availrmem >> (30 - MMU_PAGESHIFT)) >= 1402 textrepl_min_gb && l2cache_sz <= 2 << 20) { 1403 extern size_t textrepl_size_thresh; 1404 textrepl_size_thresh = (16 << 20) - 1; 1405 } 1406 #endif 1407 } 1408 1409 /* 1410 * Layout the kernel's part of address space and initialize kmem allocator. 1411 */ 1412 static void 1413 startup_kmem(void) 1414 { 1415 extern void page_set_colorequiv_arr(void); 1416 1417 PRM_POINT("startup_kmem() starting..."); 1418 1419 #if defined(__amd64) 1420 if (eprom_kernelbase && eprom_kernelbase != KERNELBASE) 1421 cmn_err(CE_NOTE, "!kernelbase cannot be changed on 64-bit " 1422 "systems."); 1423 kernelbase = segkpm_base - KERNEL_REDZONE_SIZE; 1424 core_base = (uintptr_t)COREHEAP_BASE; 1425 core_size = (size_t)MISC_VA_BASE - COREHEAP_BASE; 1426 #else /* __i386 */ 1427 /* 1428 * We configure kernelbase based on: 1429 * 1430 * 1. user specified kernelbase via eeprom command. Value cannot exceed 1431 * KERNELBASE_MAX. we large page align eprom_kernelbase 1432 * 1433 * 2. Default to KERNELBASE and adjust to 2X less the size for page_t. 1434 * On large memory systems we must lower kernelbase to allow 1435 * enough room for page_t's for all of memory. 1436 * 1437 * The value set here, might be changed a little later. 1438 */ 1439 if (eprom_kernelbase) { 1440 kernelbase = eprom_kernelbase & mmu.level_mask[1]; 1441 if (kernelbase > KERNELBASE_MAX) 1442 kernelbase = KERNELBASE_MAX; 1443 } else { 1444 kernelbase = (uintptr_t)KERNELBASE; 1445 kernelbase -= ROUND_UP_4MEG(2 * valloc_sz); 1446 } 1447 ASSERT((kernelbase & mmu.level_offset[1]) == 0); 1448 core_base = valloc_base; 1449 core_size = 0; 1450 #endif /* __i386 */ 1451 1452 PRM_DEBUG(core_base); 1453 PRM_DEBUG(core_size); 1454 PRM_DEBUG(kernelbase); 1455 1456 #if defined(__i386) 1457 segkp_fromheap = 1; 1458 #endif /* __i386 */ 1459 1460 ekernelheap = (char *)core_base; 1461 PRM_DEBUG(ekernelheap); 1462 1463 /* 1464 * Now that we know the real value of kernelbase, 1465 * update variables that were initialized with a value of 1466 * KERNELBASE (in common/conf/param.c). 1467 * 1468 * XXX The problem with this sort of hackery is that the 1469 * compiler just may feel like putting the const declarations 1470 * (in param.c) into the .text section. Perhaps they should 1471 * just be declared as variables there? 1472 */ 1473 1474 *(uintptr_t *)&_kernelbase = kernelbase; 1475 *(uintptr_t *)&_userlimit = kernelbase; 1476 #if defined(__amd64) 1477 *(uintptr_t *)&_userlimit -= KERNELBASE - USERLIMIT; 1478 #else 1479 *(uintptr_t *)&_userlimit32 = _userlimit; 1480 #endif 1481 PRM_DEBUG(_kernelbase); 1482 PRM_DEBUG(_userlimit); 1483 PRM_DEBUG(_userlimit32); 1484 1485 layout_kernel_va(); 1486 1487 #if defined(__i386) 1488 /* 1489 * If segmap is too large we can push the bottom of the kernel heap 1490 * higher than the base. Or worse, it could exceed the top of the 1491 * VA space entirely, causing it to wrap around. 1492 */ 1493 if (kernelheap >= ekernelheap || (uintptr_t)kernelheap < kernelbase) 1494 panic("too little address space available for kernelheap," 1495 " use eeprom for lower kernelbase or smaller segmapsize"); 1496 #endif /* __i386 */ 1497 1498 /* 1499 * Initialize the kernel heap. Note 3rd argument must be > 1st. 1500 */ 1501 kernelheap_init(kernelheap, ekernelheap, 1502 kernelheap + MMU_PAGESIZE, 1503 (void *)core_base, (void *)(core_base + core_size)); 1504 1505 #if defined(__xpv) 1506 /* 1507 * Link pending events struct into cpu struct 1508 */ 1509 CPU->cpu_m.mcpu_evt_pend = &cpu0_evt_data; 1510 #endif 1511 /* 1512 * Initialize kernel memory allocator. 1513 */ 1514 kmem_init(); 1515 1516 /* 1517 * Factor in colorequiv to check additional 'equivalent' bins 1518 */ 1519 page_set_colorequiv_arr(); 1520 1521 /* 1522 * print this out early so that we know what's going on 1523 */ 1524 print_x86_featureset(x86_featureset); 1525 1526 /* 1527 * Initialize bp_mapin(). 1528 */ 1529 bp_init(MMU_PAGESIZE, HAT_STORECACHING_OK); 1530 1531 /* 1532 * orig_npages is non-zero if physmem has been configured for less 1533 * than the available memory. 1534 */ 1535 if (orig_npages) { 1536 cmn_err(CE_WARN, "!%slimiting physmem to 0x%lx of 0x%lx pages", 1537 (npages == PHYSMEM ? "Due to virtual address space " : ""), 1538 npages, orig_npages); 1539 } 1540 #if defined(__i386) 1541 if (eprom_kernelbase && (eprom_kernelbase != kernelbase)) 1542 cmn_err(CE_WARN, "kernelbase value, User specified 0x%lx, " 1543 "System using 0x%lx", 1544 (uintptr_t)eprom_kernelbase, (uintptr_t)kernelbase); 1545 #endif 1546 1547 #ifdef KERNELBASE_ABI_MIN 1548 if (kernelbase < (uintptr_t)KERNELBASE_ABI_MIN) { 1549 cmn_err(CE_NOTE, "!kernelbase set to 0x%lx, system is not " 1550 "i386 ABI compliant.", (uintptr_t)kernelbase); 1551 } 1552 #endif 1553 1554 #ifndef __xpv 1555 if (plat_dr_support_memory()) { 1556 mem_config_init(); 1557 } 1558 #else /* __xpv */ 1559 /* 1560 * Some of the xen start information has to be relocated up 1561 * into the kernel's permanent address space. 1562 */ 1563 PRM_POINT("calling xen_relocate_start_info()"); 1564 xen_relocate_start_info(); 1565 PRM_POINT("xen_relocate_start_info() done"); 1566 1567 /* 1568 * (Update the vcpu pointer in our cpu structure to point into 1569 * the relocated shared info.) 1570 */ 1571 CPU->cpu_m.mcpu_vcpu_info = 1572 &HYPERVISOR_shared_info->vcpu_info[CPU->cpu_id]; 1573 #endif /* __xpv */ 1574 1575 PRM_POINT("startup_kmem() done"); 1576 } 1577 1578 #ifndef __xpv 1579 /* 1580 * If we have detected that we are running in an HVM environment, we need 1581 * to prepend the PV driver directory to the module search path. 1582 */ 1583 #define HVM_MOD_DIR "/platform/i86hvm/kernel" 1584 static void 1585 update_default_path() 1586 { 1587 char *current, *newpath; 1588 int newlen; 1589 1590 /* 1591 * We are about to resync with krtld. krtld will reset its 1592 * internal module search path iff Solaris has set default_path. 1593 * We want to be sure we're prepending this new directory to the 1594 * right search path. 1595 */ 1596 current = (default_path == NULL) ? kobj_module_path : default_path; 1597 1598 newlen = strlen(HVM_MOD_DIR) + strlen(current) + 2; 1599 newpath = kmem_alloc(newlen, KM_SLEEP); 1600 (void) strcpy(newpath, HVM_MOD_DIR); 1601 (void) strcat(newpath, " "); 1602 (void) strcat(newpath, current); 1603 1604 default_path = newpath; 1605 } 1606 #endif 1607 1608 static void 1609 startup_modules(void) 1610 { 1611 int cnt; 1612 extern void prom_setup(void); 1613 int32_t v, h; 1614 char d[11]; 1615 char *cp; 1616 cmi_hdl_t hdl; 1617 1618 PRM_POINT("startup_modules() starting..."); 1619 1620 #ifndef __xpv 1621 /* 1622 * Initialize ten-micro second timer so that drivers will 1623 * not get short changed in their init phase. This was 1624 * not getting called until clkinit which, on fast cpu's 1625 * caused the drv_usecwait to be way too short. 1626 */ 1627 microfind(); 1628 1629 if ((get_hwenv() & HW_XEN_HVM) != 0) 1630 update_default_path(); 1631 #endif 1632 1633 /* 1634 * Read the GMT lag from /etc/rtc_config. 1635 */ 1636 sgmtl(process_rtc_config_file()); 1637 1638 /* 1639 * Calculate default settings of system parameters based upon 1640 * maxusers, yet allow to be overridden via the /etc/system file. 1641 */ 1642 param_calc(0); 1643 1644 mod_setup(); 1645 1646 /* 1647 * Initialize system parameters. 1648 */ 1649 param_init(); 1650 1651 /* 1652 * Initialize the default brands 1653 */ 1654 brand_init(); 1655 1656 /* 1657 * maxmem is the amount of physical memory we're playing with. 1658 */ 1659 maxmem = physmem; 1660 1661 /* 1662 * Initialize segment management stuff. 1663 */ 1664 seg_init(); 1665 1666 if (modload("fs", "specfs") == -1) 1667 halt("Can't load specfs"); 1668 1669 if (modload("fs", "devfs") == -1) 1670 halt("Can't load devfs"); 1671 1672 if (modload("fs", "dev") == -1) 1673 halt("Can't load dev"); 1674 1675 if (modload("fs", "procfs") == -1) 1676 halt("Can't load procfs"); 1677 1678 (void) modloadonly("sys", "lbl_edition"); 1679 1680 dispinit(); 1681 1682 /* Read cluster configuration data. */ 1683 clconf_init(); 1684 1685 #if defined(__xpv) 1686 (void) ec_init(); 1687 gnttab_init(); 1688 (void) xs_early_init(); 1689 #endif /* __xpv */ 1690 1691 /* 1692 * Create a kernel device tree. First, create rootnex and 1693 * then invoke bus specific code to probe devices. 1694 */ 1695 setup_ddi(); 1696 1697 #ifdef __xpv 1698 if (DOMAIN_IS_INITDOMAIN(xen_info)) 1699 #endif 1700 { 1701 id_t smid; 1702 smbios_system_t smsys; 1703 smbios_info_t sminfo; 1704 char *mfg; 1705 /* 1706 * Load the System Management BIOS into the global ksmbios 1707 * handle, if an SMBIOS is present on this system. 1708 * Also set "si-hw-provider" property, if not already set. 1709 */ 1710 ksmbios = smbios_open(NULL, SMB_VERSION, ksmbios_flags, NULL); 1711 if (ksmbios != NULL && 1712 ((smid = smbios_info_system(ksmbios, &smsys)) != SMB_ERR) && 1713 (smbios_info_common(ksmbios, smid, &sminfo)) != SMB_ERR) { 1714 mfg = (char *)sminfo.smbi_manufacturer; 1715 if (BOP_GETPROPLEN(bootops, "si-hw-provider") < 0) { 1716 extern char hw_provider[]; 1717 int i; 1718 for (i = 0; i < SYS_NMLN; i++) { 1719 if (isprint(mfg[i])) 1720 hw_provider[i] = mfg[i]; 1721 else { 1722 hw_provider[i] = '\0'; 1723 break; 1724 } 1725 } 1726 hw_provider[SYS_NMLN - 1] = '\0'; 1727 } 1728 } 1729 } 1730 1731 1732 /* 1733 * Originally clconf_init() apparently needed the hostid. But 1734 * this no longer appears to be true - it uses its own nodeid. 1735 * By placing the hostid logic here, we are able to make use of 1736 * the SMBIOS UUID. 1737 */ 1738 if ((h = set_soft_hostid()) == HW_INVALID_HOSTID) { 1739 cmn_err(CE_WARN, "Unable to set hostid"); 1740 } else { 1741 for (v = h, cnt = 0; cnt < 10; cnt++) { 1742 d[cnt] = (char)(v % 10); 1743 v /= 10; 1744 if (v == 0) 1745 break; 1746 } 1747 for (cp = hw_serial; cnt >= 0; cnt--) 1748 *cp++ = d[cnt] + '0'; 1749 *cp = 0; 1750 } 1751 1752 /* 1753 * Set up the CPU module subsystem for the boot cpu in the native 1754 * case, and all physical cpu resource in the xpv dom0 case. 1755 * Modifies the device tree, so this must be done after 1756 * setup_ddi(). 1757 */ 1758 #ifdef __xpv 1759 /* 1760 * If paravirtualized and on dom0 then we initialize all physical 1761 * cpu handles now; if paravirtualized on a domU then do not 1762 * initialize. 1763 */ 1764 if (DOMAIN_IS_INITDOMAIN(xen_info)) { 1765 xen_mc_lcpu_cookie_t cpi; 1766 1767 for (cpi = xen_physcpu_next(NULL); cpi != NULL; 1768 cpi = xen_physcpu_next(cpi)) { 1769 if ((hdl = cmi_init(CMI_HDL_SOLARIS_xVM_MCA, 1770 xen_physcpu_chipid(cpi), xen_physcpu_coreid(cpi), 1771 xen_physcpu_strandid(cpi))) != NULL && 1772 is_x86_feature(x86_featureset, X86FSET_MCA)) 1773 cmi_mca_init(hdl); 1774 } 1775 } 1776 #else 1777 /* 1778 * Initialize a handle for the boot cpu - others will initialize 1779 * as they startup. Do not do this if we know we are in an HVM domU. 1780 */ 1781 if ((get_hwenv() & HW_XEN_HVM) == 0 && 1782 (hdl = cmi_init(CMI_HDL_NATIVE, cmi_ntv_hwchipid(CPU), 1783 cmi_ntv_hwcoreid(CPU), cmi_ntv_hwstrandid(CPU))) != NULL && 1784 is_x86_feature(x86_featureset, X86FSET_MCA)) { 1785 cmi_mca_init(hdl); 1786 CPU->cpu_m.mcpu_cmi_hdl = hdl; 1787 } 1788 #endif /* __xpv */ 1789 1790 /* 1791 * Fake a prom tree such that /dev/openprom continues to work 1792 */ 1793 PRM_POINT("startup_modules: calling prom_setup..."); 1794 prom_setup(); 1795 PRM_POINT("startup_modules: done"); 1796 1797 /* 1798 * Load all platform specific modules 1799 */ 1800 PRM_POINT("startup_modules: calling psm_modload..."); 1801 psm_modload(); 1802 1803 PRM_POINT("startup_modules() done"); 1804 } 1805 1806 /* 1807 * claim a "setaside" boot page for use in the kernel 1808 */ 1809 page_t * 1810 boot_claim_page(pfn_t pfn) 1811 { 1812 page_t *pp; 1813 1814 pp = page_numtopp_nolock(pfn); 1815 ASSERT(pp != NULL); 1816 1817 if (PP_ISBOOTPAGES(pp)) { 1818 if (pp->p_next != NULL) 1819 pp->p_next->p_prev = pp->p_prev; 1820 if (pp->p_prev == NULL) 1821 bootpages = pp->p_next; 1822 else 1823 pp->p_prev->p_next = pp->p_next; 1824 } else { 1825 /* 1826 * htable_attach() expects a base pagesize page 1827 */ 1828 if (pp->p_szc != 0) 1829 page_boot_demote(pp); 1830 pp = page_numtopp(pfn, SE_EXCL); 1831 } 1832 return (pp); 1833 } 1834 1835 /* 1836 * Walk through the pagetables looking for pages mapped in by boot. If the 1837 * setaside flag is set the pages are expected to be returned to the 1838 * kernel later in boot, so we add them to the bootpages list. 1839 */ 1840 static void 1841 protect_boot_range(uintptr_t low, uintptr_t high, int setaside) 1842 { 1843 uintptr_t va = low; 1844 size_t len; 1845 uint_t prot; 1846 pfn_t pfn; 1847 page_t *pp; 1848 pgcnt_t boot_protect_cnt = 0; 1849 1850 while (kbm_probe(&va, &len, &pfn, &prot) != 0 && va < high) { 1851 if (va + len >= high) 1852 panic("0x%lx byte mapping at 0x%p exceeds boot's " 1853 "legal range.", len, (void *)va); 1854 1855 while (len > 0) { 1856 pp = page_numtopp_alloc(pfn); 1857 if (pp != NULL) { 1858 if (setaside == 0) 1859 panic("Unexpected mapping by boot. " 1860 "addr=%p pfn=%lx\n", 1861 (void *)va, pfn); 1862 1863 pp->p_next = bootpages; 1864 pp->p_prev = NULL; 1865 PP_SETBOOTPAGES(pp); 1866 if (bootpages != NULL) { 1867 bootpages->p_prev = pp; 1868 } 1869 bootpages = pp; 1870 ++boot_protect_cnt; 1871 } 1872 1873 ++pfn; 1874 len -= MMU_PAGESIZE; 1875 va += MMU_PAGESIZE; 1876 } 1877 } 1878 PRM_DEBUG(boot_protect_cnt); 1879 } 1880 1881 /* 1882 * 1883 */ 1884 static void 1885 layout_kernel_va(void) 1886 { 1887 PRM_POINT("layout_kernel_va() starting..."); 1888 /* 1889 * Establish the final size of the kernel's heap, size of segmap, 1890 * segkp, etc. 1891 */ 1892 1893 #if defined(__amd64) 1894 1895 kpm_vbase = (caddr_t)segkpm_base; 1896 if (physmax + 1 < plat_dr_physmax) { 1897 kpm_size = ROUND_UP_LPAGE(mmu_ptob(plat_dr_physmax)); 1898 } else { 1899 kpm_size = ROUND_UP_LPAGE(mmu_ptob(physmax + 1)); 1900 } 1901 if ((uintptr_t)kpm_vbase + kpm_size > (uintptr_t)valloc_base) 1902 panic("not enough room for kpm!"); 1903 PRM_DEBUG(kpm_size); 1904 PRM_DEBUG(kpm_vbase); 1905 1906 /* 1907 * By default we create a seg_kp in 64 bit kernels, it's a little 1908 * faster to access than embedding it in the heap. 1909 */ 1910 segkp_base = (caddr_t)valloc_base + valloc_sz; 1911 if (!segkp_fromheap) { 1912 size_t sz = mmu_ptob(segkpsize); 1913 1914 /* 1915 * determine size of segkp 1916 */ 1917 if (sz < SEGKPMINSIZE || sz > SEGKPMAXSIZE) { 1918 sz = SEGKPDEFSIZE; 1919 cmn_err(CE_WARN, "!Illegal value for segkpsize. " 1920 "segkpsize has been reset to %ld pages", 1921 mmu_btop(sz)); 1922 } 1923 sz = MIN(sz, MAX(SEGKPMINSIZE, mmu_ptob(physmem))); 1924 1925 segkpsize = mmu_btop(ROUND_UP_LPAGE(sz)); 1926 } 1927 PRM_DEBUG(segkp_base); 1928 PRM_DEBUG(segkpsize); 1929 1930 /* 1931 * segzio is used for ZFS cached data. It uses a distinct VA 1932 * segment (from kernel heap) so that we can easily tell not to 1933 * include it in kernel crash dumps on 64 bit kernels. The trick is 1934 * to give it lots of VA, but not constrain the kernel heap. 1935 * We can use 1.5x physmem for segzio, leaving approximately 1936 * another 1.5x physmem for heap. See also the comment in 1937 * startup_memlist(). 1938 */ 1939 segzio_base = segkp_base + mmu_ptob(segkpsize); 1940 if (segzio_fromheap) { 1941 segziosize = 0; 1942 } else { 1943 size_t physmem_size = mmu_ptob(physmem); 1944 size_t size = (segziosize == 0) ? 1945 physmem_size * 3 / 2 : mmu_ptob(segziosize); 1946 1947 if (size < SEGZIOMINSIZE) 1948 size = SEGZIOMINSIZE; 1949 segziosize = mmu_btop(ROUND_UP_LPAGE(size)); 1950 } 1951 PRM_DEBUG(segziosize); 1952 PRM_DEBUG(segzio_base); 1953 1954 /* 1955 * Put the range of VA for device mappings next, kmdb knows to not 1956 * grep in this range of addresses. 1957 */ 1958 toxic_addr = 1959 ROUND_UP_LPAGE((uintptr_t)segzio_base + mmu_ptob(segziosize)); 1960 PRM_DEBUG(toxic_addr); 1961 segmap_start = ROUND_UP_LPAGE(toxic_addr + toxic_size); 1962 #else /* __i386 */ 1963 segmap_start = ROUND_UP_LPAGE(kernelbase); 1964 #endif /* __i386 */ 1965 PRM_DEBUG(segmap_start); 1966 1967 /* 1968 * Users can change segmapsize through eeprom. If the variable 1969 * is tuned through eeprom, there is no upper bound on the 1970 * size of segmap. 1971 */ 1972 segmapsize = MAX(ROUND_UP_LPAGE(segmapsize), SEGMAPDEFAULT); 1973 1974 #if defined(__i386) 1975 /* 1976 * 32-bit systems don't have segkpm or segkp, so segmap appears at 1977 * the bottom of the kernel's address range. Set aside space for a 1978 * small red zone just below the start of segmap. 1979 */ 1980 segmap_start += KERNEL_REDZONE_SIZE; 1981 segmapsize -= KERNEL_REDZONE_SIZE; 1982 #endif 1983 1984 PRM_DEBUG(segmap_start); 1985 PRM_DEBUG(segmapsize); 1986 kernelheap = (caddr_t)ROUND_UP_LPAGE(segmap_start + segmapsize); 1987 PRM_DEBUG(kernelheap); 1988 PRM_POINT("layout_kernel_va() done..."); 1989 } 1990 1991 /* 1992 * Finish initializing the VM system, now that we are no longer 1993 * relying on the boot time memory allocators. 1994 */ 1995 static void 1996 startup_vm(void) 1997 { 1998 struct segmap_crargs a; 1999 2000 extern int use_brk_lpg, use_stk_lpg; 2001 2002 PRM_POINT("startup_vm() starting..."); 2003 2004 /* 2005 * Initialize the hat layer. 2006 */ 2007 hat_init(); 2008 2009 /* 2010 * Do final allocations of HAT data structures that need to 2011 * be allocated before quiescing the boot loader. 2012 */ 2013 PRM_POINT("Calling hat_kern_alloc()..."); 2014 hat_kern_alloc((caddr_t)segmap_start, segmapsize, ekernelheap); 2015 PRM_POINT("hat_kern_alloc() done"); 2016 2017 #ifndef __xpv 2018 /* 2019 * Setup Page Attribute Table 2020 */ 2021 pat_sync(); 2022 #endif 2023 2024 /* 2025 * The next two loops are done in distinct steps in order 2026 * to be sure that any page that is doubly mapped (both above 2027 * KERNEL_TEXT and below kernelbase) is dealt with correctly. 2028 * Note this may never happen, but it might someday. 2029 */ 2030 bootpages = NULL; 2031 PRM_POINT("Protecting boot pages"); 2032 2033 /* 2034 * Protect any pages mapped above KERNEL_TEXT that somehow have 2035 * page_t's. This can only happen if something weird allocated 2036 * in this range (like kadb/kmdb). 2037 */ 2038 protect_boot_range(KERNEL_TEXT, (uintptr_t)-1, 0); 2039 2040 /* 2041 * Before we can take over memory allocation/mapping from the boot 2042 * loader we must remove from our free page lists any boot allocated 2043 * pages that stay mapped until release_bootstrap(). 2044 */ 2045 protect_boot_range(0, kernelbase, 1); 2046 2047 2048 /* 2049 * Switch to running on regular HAT (not boot_mmu) 2050 */ 2051 PRM_POINT("Calling hat_kern_setup()..."); 2052 hat_kern_setup(); 2053 2054 /* 2055 * It is no longer safe to call BOP_ALLOC(), so make sure we don't. 2056 */ 2057 bop_no_more_mem(); 2058 2059 PRM_POINT("hat_kern_setup() done"); 2060 2061 hat_cpu_online(CPU); 2062 2063 /* 2064 * Initialize VM system 2065 */ 2066 PRM_POINT("Calling kvm_init()..."); 2067 kvm_init(); 2068 PRM_POINT("kvm_init() done"); 2069 2070 /* 2071 * Tell kmdb that the VM system is now working 2072 */ 2073 if (boothowto & RB_DEBUG) 2074 kdi_dvec_vmready(); 2075 2076 #if defined(__xpv) 2077 /* 2078 * Populate the I/O pool on domain 0 2079 */ 2080 if (DOMAIN_IS_INITDOMAIN(xen_info)) { 2081 extern long populate_io_pool(void); 2082 long init_io_pool_cnt; 2083 2084 PRM_POINT("Populating reserve I/O page pool"); 2085 init_io_pool_cnt = populate_io_pool(); 2086 PRM_DEBUG(init_io_pool_cnt); 2087 } 2088 #endif 2089 /* 2090 * Mangle the brand string etc. 2091 */ 2092 cpuid_pass3(CPU); 2093 2094 #if defined(__amd64) 2095 2096 /* 2097 * Create the device arena for toxic (to dtrace/kmdb) mappings. 2098 */ 2099 device_arena = vmem_create("device", (void *)toxic_addr, 2100 toxic_size, MMU_PAGESIZE, NULL, NULL, NULL, 0, VM_SLEEP); 2101 2102 #else /* __i386 */ 2103 2104 /* 2105 * allocate the bit map that tracks toxic pages 2106 */ 2107 toxic_bit_map_len = btop((ulong_t)(valloc_base - kernelbase)); 2108 PRM_DEBUG(toxic_bit_map_len); 2109 toxic_bit_map = 2110 kmem_zalloc(BT_SIZEOFMAP(toxic_bit_map_len), KM_NOSLEEP); 2111 ASSERT(toxic_bit_map != NULL); 2112 PRM_DEBUG(toxic_bit_map); 2113 2114 #endif /* __i386 */ 2115 2116 2117 /* 2118 * Now that we've got more VA, as well as the ability to allocate from 2119 * it, tell the debugger. 2120 */ 2121 if (boothowto & RB_DEBUG) 2122 kdi_dvec_memavail(); 2123 2124 /* 2125 * The following code installs a special page fault handler (#pf) 2126 * to work around a pentium bug. 2127 */ 2128 #if !defined(__amd64) && !defined(__xpv) 2129 if (x86_type == X86_TYPE_P5) { 2130 desctbr_t idtr; 2131 gate_desc_t *newidt; 2132 2133 if ((newidt = kmem_zalloc(MMU_PAGESIZE, KM_NOSLEEP)) == NULL) 2134 panic("failed to install pentium_pftrap"); 2135 2136 bcopy(idt0, newidt, NIDT * sizeof (*idt0)); 2137 set_gatesegd(&newidt[T_PGFLT], &pentium_pftrap, 2138 KCS_SEL, SDT_SYSIGT, TRP_KPL, 0); 2139 2140 (void) as_setprot(&kas, (caddr_t)newidt, MMU_PAGESIZE, 2141 PROT_READ | PROT_EXEC); 2142 2143 CPU->cpu_idt = newidt; 2144 idtr.dtr_base = (uintptr_t)CPU->cpu_idt; 2145 idtr.dtr_limit = (NIDT * sizeof (*idt0)) - 1; 2146 wr_idtr(&idtr); 2147 } 2148 #endif /* !__amd64 */ 2149 2150 #if !defined(__xpv) 2151 /* 2152 * Map page pfn=0 for drivers, such as kd, that need to pick up 2153 * parameters left there by controllers/BIOS. 2154 */ 2155 PRM_POINT("setup up p0_va"); 2156 p0_va = i86devmap(0, 1, PROT_READ); 2157 PRM_DEBUG(p0_va); 2158 #endif 2159 2160 cmn_err(CE_CONT, "?mem = %luK (0x%lx)\n", 2161 physinstalled << (MMU_PAGESHIFT - 10), ptob(physinstalled)); 2162 2163 /* 2164 * disable automatic large pages for small memory systems or 2165 * when the disable flag is set. 2166 * 2167 * Do not yet consider page sizes larger than 2m/4m. 2168 */ 2169 if (!auto_lpg_disable && mmu.max_page_level > 0) { 2170 max_uheap_lpsize = LEVEL_SIZE(1); 2171 max_ustack_lpsize = LEVEL_SIZE(1); 2172 max_privmap_lpsize = LEVEL_SIZE(1); 2173 max_uidata_lpsize = LEVEL_SIZE(1); 2174 max_utext_lpsize = LEVEL_SIZE(1); 2175 max_shm_lpsize = LEVEL_SIZE(1); 2176 } 2177 if (physmem < privm_lpg_min_physmem || mmu.max_page_level == 0 || 2178 auto_lpg_disable) { 2179 use_brk_lpg = 0; 2180 use_stk_lpg = 0; 2181 } 2182 mcntl0_lpsize = LEVEL_SIZE(mmu.umax_page_level); 2183 2184 PRM_POINT("Calling hat_init_finish()..."); 2185 hat_init_finish(); 2186 PRM_POINT("hat_init_finish() done"); 2187 2188 /* 2189 * Initialize the segkp segment type. 2190 */ 2191 rw_enter(&kas.a_lock, RW_WRITER); 2192 PRM_POINT("Attaching segkp"); 2193 if (segkp_fromheap) { 2194 segkp->s_as = &kas; 2195 } else if (seg_attach(&kas, (caddr_t)segkp_base, mmu_ptob(segkpsize), 2196 segkp) < 0) { 2197 panic("startup: cannot attach segkp"); 2198 /*NOTREACHED*/ 2199 } 2200 PRM_POINT("Doing segkp_create()"); 2201 if (segkp_create(segkp) != 0) { 2202 panic("startup: segkp_create failed"); 2203 /*NOTREACHED*/ 2204 } 2205 PRM_DEBUG(segkp); 2206 rw_exit(&kas.a_lock); 2207 2208 /* 2209 * kpm segment 2210 */ 2211 segmap_kpm = 0; 2212 if (kpm_desired) { 2213 kpm_init(); 2214 kpm_enable = 1; 2215 } 2216 2217 /* 2218 * Now create segmap segment. 2219 */ 2220 rw_enter(&kas.a_lock, RW_WRITER); 2221 if (seg_attach(&kas, (caddr_t)segmap_start, segmapsize, segmap) < 0) { 2222 panic("cannot attach segmap"); 2223 /*NOTREACHED*/ 2224 } 2225 PRM_DEBUG(segmap); 2226 2227 a.prot = PROT_READ | PROT_WRITE; 2228 a.shmsize = 0; 2229 a.nfreelist = segmapfreelists; 2230 2231 if (segmap_create(segmap, (caddr_t)&a) != 0) 2232 panic("segmap_create segmap"); 2233 rw_exit(&kas.a_lock); 2234 2235 setup_vaddr_for_ppcopy(CPU); 2236 2237 segdev_init(); 2238 #if defined(__xpv) 2239 if (DOMAIN_IS_INITDOMAIN(xen_info)) 2240 #endif 2241 pmem_init(); 2242 2243 PRM_POINT("startup_vm() done"); 2244 } 2245 2246 /* 2247 * Load a tod module for the non-standard tod part found on this system. 2248 */ 2249 static void 2250 load_tod_module(char *todmod) 2251 { 2252 if (modload("tod", todmod) == -1) 2253 halt("Can't load TOD module"); 2254 } 2255 2256 static void 2257 startup_end(void) 2258 { 2259 int i; 2260 extern void setx86isalist(void); 2261 extern void cpu_event_init(void); 2262 2263 PRM_POINT("startup_end() starting..."); 2264 2265 /* 2266 * Perform tasks that get done after most of the VM 2267 * initialization has been done but before the clock 2268 * and other devices get started. 2269 */ 2270 kern_setup1(); 2271 2272 /* 2273 * Perform CPC initialization for this CPU. 2274 */ 2275 kcpc_hw_init(CPU); 2276 2277 /* 2278 * Initialize cpu event framework. 2279 */ 2280 cpu_event_init(); 2281 2282 #if defined(OPTERON_WORKAROUND_6323525) 2283 if (opteron_workaround_6323525) 2284 patch_workaround_6323525(); 2285 #endif 2286 /* 2287 * If needed, load TOD module now so that ddi_get_time(9F) etc. work 2288 * (For now, "needed" is defined as set tod_module_name in /etc/system) 2289 */ 2290 if (tod_module_name != NULL) { 2291 PRM_POINT("load_tod_module()"); 2292 load_tod_module(tod_module_name); 2293 } 2294 2295 #if defined(__xpv) 2296 /* 2297 * Forceload interposing TOD module for the hypervisor. 2298 */ 2299 PRM_POINT("load_tod_module()"); 2300 load_tod_module("xpvtod"); 2301 #endif 2302 2303 /* 2304 * Configure the system. 2305 */ 2306 PRM_POINT("Calling configure()..."); 2307 configure(); /* set up devices */ 2308 PRM_POINT("configure() done"); 2309 2310 /* 2311 * We can now setup for XSAVE because fpu_probe is done in configure(). 2312 */ 2313 if (fp_save_mech == FP_XSAVE) { 2314 xsave_setup_msr(CPU); 2315 } 2316 2317 /* 2318 * Set the isa_list string to the defined instruction sets we 2319 * support. 2320 */ 2321 setx86isalist(); 2322 cpu_intr_alloc(CPU, NINTR_THREADS); 2323 psm_install(); 2324 2325 /* 2326 * We're done with bootops. We don't unmap the bootstrap yet because 2327 * we're still using bootsvcs. 2328 */ 2329 PRM_POINT("NULLing out bootops"); 2330 *bootopsp = (struct bootops *)NULL; 2331 bootops = (struct bootops *)NULL; 2332 2333 #if defined(__xpv) 2334 ec_init_debug_irq(); 2335 xs_domu_init(); 2336 #endif 2337 2338 #if defined(__amd64) && !defined(__xpv) 2339 /* 2340 * Intel IOMMU has been setup/initialized in ddi_impl.c 2341 * Start it up now. 2342 */ 2343 immu_startup(); 2344 #endif 2345 2346 PRM_POINT("Enabling interrupts"); 2347 (*picinitf)(); 2348 sti(); 2349 #if defined(__xpv) 2350 ASSERT(CPU->cpu_m.mcpu_vcpu_info->evtchn_upcall_mask == 0); 2351 xen_late_startup(); 2352 #endif 2353 2354 (void) add_avsoftintr((void *)&softlevel1_hdl, 1, softlevel1, 2355 "softlevel1", NULL, NULL); /* XXX to be moved later */ 2356 2357 /* 2358 * Register software interrupt handlers for ddi_periodic_add(9F). 2359 * Software interrupts up to the level 10 are supported. 2360 */ 2361 for (i = DDI_IPL_1; i <= DDI_IPL_10; i++) { 2362 (void) add_avsoftintr((void *)&softlevel_hdl[i-1], i, 2363 (avfunc)ddi_periodic_softintr, "ddi_periodic", 2364 (caddr_t)(uintptr_t)i, NULL); 2365 } 2366 2367 #if !defined(__xpv) 2368 if (modload("drv", "amd_iommu") < 0) { 2369 PRM_POINT("No AMD IOMMU present\n"); 2370 } else if (ddi_hold_installed_driver(ddi_name_to_major( 2371 "amd_iommu")) == NULL) { 2372 prom_printf("ERROR: failed to attach AMD IOMMU\n"); 2373 } 2374 #endif 2375 post_startup_cpu_fixups(); 2376 2377 PRM_POINT("startup_end() done"); 2378 } 2379 2380 /* 2381 * Don't remove the following 2 variables. They are necessary 2382 * for reading the hostid from the legacy file (/kernel/misc/sysinit). 2383 */ 2384 char *_hs1107 = hw_serial; 2385 ulong_t _bdhs34; 2386 2387 void 2388 post_startup(void) 2389 { 2390 extern void cpupm_init(cpu_t *); 2391 extern void cpu_event_init_cpu(cpu_t *); 2392 2393 /* 2394 * Set the system wide, processor-specific flags to be passed 2395 * to userland via the aux vector for performance hints and 2396 * instruction set extensions. 2397 */ 2398 bind_hwcap(); 2399 2400 #ifdef __xpv 2401 if (DOMAIN_IS_INITDOMAIN(xen_info)) 2402 #endif 2403 { 2404 #if defined(__xpv) 2405 xpv_panic_init(); 2406 #else 2407 /* 2408 * Startup the memory scrubber. 2409 * XXPV This should be running somewhere .. 2410 */ 2411 if ((get_hwenv() & HW_VIRTUAL) == 0) 2412 memscrub_init(); 2413 #endif 2414 } 2415 2416 /* 2417 * Complete CPU module initialization 2418 */ 2419 cmi_post_startup(); 2420 2421 /* 2422 * Perform forceloading tasks for /etc/system. 2423 */ 2424 (void) mod_sysctl(SYS_FORCELOAD, NULL); 2425 2426 /* 2427 * ON4.0: Force /proc module in until clock interrupt handle fixed 2428 * ON4.0: This must be fixed or restated in /etc/systems. 2429 */ 2430 (void) modload("fs", "procfs"); 2431 2432 (void) i_ddi_attach_hw_nodes("pit_beep"); 2433 2434 #if defined(__i386) 2435 /* 2436 * Check for required functional Floating Point hardware, 2437 * unless FP hardware explicitly disabled. 2438 */ 2439 if (fpu_exists && (fpu_pentium_fdivbug || fp_kind == FP_NO)) 2440 halt("No working FP hardware found"); 2441 #endif 2442 2443 maxmem = freemem; 2444 2445 cpu_event_init_cpu(CPU); 2446 cpupm_init(CPU); 2447 (void) mach_cpu_create_device_node(CPU, NULL); 2448 2449 pg_init(); 2450 } 2451 2452 static int 2453 pp_in_range(page_t *pp, uint64_t low_addr, uint64_t high_addr) 2454 { 2455 return ((pp->p_pagenum >= btop(low_addr)) && 2456 (pp->p_pagenum < btopr(high_addr))); 2457 } 2458 2459 static int 2460 pp_in_module(page_t *pp, const rd_existing_t *modranges) 2461 { 2462 uint_t i; 2463 2464 for (i = 0; modranges[i].phys != 0; i++) { 2465 if (pp_in_range(pp, modranges[i].phys, 2466 modranges[i].phys + modranges[i].size)) 2467 return (1); 2468 } 2469 2470 return (0); 2471 } 2472 2473 void 2474 release_bootstrap(void) 2475 { 2476 int root_is_ramdisk; 2477 page_t *pp; 2478 extern void kobj_boot_unmountroot(void); 2479 extern dev_t rootdev; 2480 uint_t i; 2481 char propname[32]; 2482 rd_existing_t *modranges; 2483 #if !defined(__xpv) 2484 pfn_t pfn; 2485 #endif 2486 2487 /* 2488 * Save the bootfs module ranges so that we can reserve them below 2489 * for the real bootfs. 2490 */ 2491 modranges = kmem_alloc(sizeof (rd_existing_t) * MAX_BOOT_MODULES, 2492 KM_SLEEP); 2493 for (i = 0; ; i++) { 2494 uint64_t start, size; 2495 2496 modranges[i].phys = 0; 2497 2498 (void) snprintf(propname, sizeof (propname), 2499 "module-addr-%u", i); 2500 if (do_bsys_getproplen(NULL, propname) <= 0) 2501 break; 2502 (void) do_bsys_getprop(NULL, propname, &start); 2503 2504 (void) snprintf(propname, sizeof (propname), 2505 "module-size-%u", i); 2506 if (do_bsys_getproplen(NULL, propname) <= 0) 2507 break; 2508 (void) do_bsys_getprop(NULL, propname, &size); 2509 2510 modranges[i].phys = start; 2511 modranges[i].size = size; 2512 } 2513 2514 /* unmount boot ramdisk and release kmem usage */ 2515 kobj_boot_unmountroot(); 2516 2517 /* 2518 * We're finished using the boot loader so free its pages. 2519 */ 2520 PRM_POINT("Unmapping lower boot pages"); 2521 2522 clear_boot_mappings(0, _userlimit); 2523 2524 postbootkernelbase = kernelbase; 2525 2526 /* 2527 * If root isn't on ramdisk, destroy the hardcoded 2528 * ramdisk node now and release the memory. Else, 2529 * ramdisk memory is kept in rd_pages. 2530 */ 2531 root_is_ramdisk = (getmajor(rootdev) == ddi_name_to_major("ramdisk")); 2532 if (!root_is_ramdisk) { 2533 dev_info_t *dip = ddi_find_devinfo("ramdisk", -1, 0); 2534 ASSERT(dip && ddi_get_parent(dip) == ddi_root_node()); 2535 ndi_rele_devi(dip); /* held from ddi_find_devinfo */ 2536 (void) ddi_remove_child(dip, 0); 2537 } 2538 2539 PRM_POINT("Releasing boot pages"); 2540 while (bootpages) { 2541 extern uint64_t ramdisk_start, ramdisk_end; 2542 pp = bootpages; 2543 bootpages = pp->p_next; 2544 2545 2546 /* Keep pages for the lower 64K */ 2547 if (pp_in_range(pp, 0, 0x40000)) { 2548 pp->p_next = lower_pages; 2549 lower_pages = pp; 2550 lower_pages_count++; 2551 continue; 2552 } 2553 2554 if (root_is_ramdisk && pp_in_range(pp, ramdisk_start, 2555 ramdisk_end) || pp_in_module(pp, modranges)) { 2556 pp->p_next = rd_pages; 2557 rd_pages = pp; 2558 continue; 2559 } 2560 pp->p_next = (struct page *)0; 2561 pp->p_prev = (struct page *)0; 2562 PP_CLRBOOTPAGES(pp); 2563 page_free(pp, 1); 2564 } 2565 PRM_POINT("Boot pages released"); 2566 2567 kmem_free(modranges, sizeof (rd_existing_t) * 99); 2568 2569 #if !defined(__xpv) 2570 /* XXPV -- note this following bunch of code needs to be revisited in Xen 3.0 */ 2571 /* 2572 * Find 1 page below 1 MB so that other processors can boot up or 2573 * so that any processor can resume. 2574 * Make sure it has a kernel VA as well as a 1:1 mapping. 2575 * We should have just free'd one up. 2576 */ 2577 2578 /* 2579 * 0x10 pages is 64K. Leave the bottom 64K alone 2580 * for BIOS. 2581 */ 2582 for (pfn = 0x10; pfn < btop(1*1024*1024); pfn++) { 2583 if (page_numtopp_alloc(pfn) == NULL) 2584 continue; 2585 rm_platter_va = i86devmap(pfn, 1, 2586 PROT_READ | PROT_WRITE | PROT_EXEC); 2587 rm_platter_pa = ptob(pfn); 2588 break; 2589 } 2590 if (pfn == btop(1*1024*1024) && use_mp) 2591 panic("No page below 1M available for starting " 2592 "other processors or for resuming from system-suspend"); 2593 #endif /* !__xpv */ 2594 } 2595 2596 /* 2597 * Initialize the platform-specific parts of a page_t. 2598 */ 2599 void 2600 add_physmem_cb(page_t *pp, pfn_t pnum) 2601 { 2602 pp->p_pagenum = pnum; 2603 pp->p_mapping = NULL; 2604 pp->p_embed = 0; 2605 pp->p_share = 0; 2606 pp->p_mlentry = 0; 2607 } 2608 2609 /* 2610 * kphysm_init() initializes physical memory. 2611 */ 2612 static pgcnt_t 2613 kphysm_init( 2614 page_t *pp, 2615 pgcnt_t npages) 2616 { 2617 struct memlist *pmem; 2618 struct memseg *cur_memseg; 2619 pfn_t base_pfn; 2620 pfn_t end_pfn; 2621 pgcnt_t num; 2622 pgcnt_t pages_done = 0; 2623 uint64_t addr; 2624 uint64_t size; 2625 extern pfn_t ddiphysmin; 2626 extern int mnode_xwa; 2627 int ms = 0, me = 0; 2628 2629 ASSERT(page_hash != NULL && page_hashsz != 0); 2630 2631 cur_memseg = memseg_base; 2632 for (pmem = phys_avail; pmem && npages; pmem = pmem->ml_next) { 2633 /* 2634 * In a 32 bit kernel can't use higher memory if we're 2635 * not booting in PAE mode. This check takes care of that. 2636 */ 2637 addr = pmem->ml_address; 2638 size = pmem->ml_size; 2639 if (btop(addr) > physmax) 2640 continue; 2641 2642 /* 2643 * align addr and size - they may not be at page boundaries 2644 */ 2645 if ((addr & MMU_PAGEOFFSET) != 0) { 2646 addr += MMU_PAGEOFFSET; 2647 addr &= ~(uint64_t)MMU_PAGEOFFSET; 2648 size -= addr - pmem->ml_address; 2649 } 2650 2651 /* only process pages below or equal to physmax */ 2652 if ((btop(addr + size) - 1) > physmax) 2653 size = ptob(physmax - btop(addr) + 1); 2654 2655 num = btop(size); 2656 if (num == 0) 2657 continue; 2658 2659 if (num > npages) 2660 num = npages; 2661 2662 npages -= num; 2663 pages_done += num; 2664 base_pfn = btop(addr); 2665 2666 if (prom_debug) 2667 prom_printf("MEMSEG addr=0x%" PRIx64 2668 " pgs=0x%lx pfn 0x%lx-0x%lx\n", 2669 addr, num, base_pfn, base_pfn + num); 2670 2671 /* 2672 * Ignore pages below ddiphysmin to simplify ddi memory 2673 * allocation with non-zero addr_lo requests. 2674 */ 2675 if (base_pfn < ddiphysmin) { 2676 if (base_pfn + num <= ddiphysmin) 2677 continue; 2678 pp += (ddiphysmin - base_pfn); 2679 num -= (ddiphysmin - base_pfn); 2680 base_pfn = ddiphysmin; 2681 } 2682 2683 /* 2684 * mnode_xwa is greater than 1 when large pages regions can 2685 * cross memory node boundaries. To prevent the formation 2686 * of these large pages, configure the memsegs based on the 2687 * memory node ranges which had been made non-contiguous. 2688 */ 2689 if (mnode_xwa > 1) { 2690 2691 end_pfn = base_pfn + num - 1; 2692 ms = PFN_2_MEM_NODE(base_pfn); 2693 me = PFN_2_MEM_NODE(end_pfn); 2694 2695 if (ms != me) { 2696 /* 2697 * current range spans more than 1 memory node. 2698 * Set num to only the pfn range in the start 2699 * memory node. 2700 */ 2701 num = mem_node_config[ms].physmax - base_pfn 2702 + 1; 2703 ASSERT(end_pfn > mem_node_config[ms].physmax); 2704 } 2705 } 2706 2707 for (;;) { 2708 /* 2709 * Build the memsegs entry 2710 */ 2711 cur_memseg->pages = pp; 2712 cur_memseg->epages = pp + num; 2713 cur_memseg->pages_base = base_pfn; 2714 cur_memseg->pages_end = base_pfn + num; 2715 2716 /* 2717 * Insert into memseg list in decreasing pfn range 2718 * order. Low memory is typically more fragmented such 2719 * that this ordering keeps the larger ranges at the 2720 * front of the list for code that searches memseg. 2721 * This ASSERTS that the memsegs coming in from boot 2722 * are in increasing physical address order and not 2723 * contiguous. 2724 */ 2725 if (memsegs != NULL) { 2726 ASSERT(cur_memseg->pages_base >= 2727 memsegs->pages_end); 2728 cur_memseg->next = memsegs; 2729 } 2730 memsegs = cur_memseg; 2731 2732 /* 2733 * add_physmem() initializes the PSM part of the page 2734 * struct by calling the PSM back with add_physmem_cb(). 2735 * In addition it coalesces pages into larger pages as 2736 * it initializes them. 2737 */ 2738 add_physmem(pp, num, base_pfn); 2739 cur_memseg++; 2740 availrmem_initial += num; 2741 availrmem += num; 2742 2743 pp += num; 2744 if (ms >= me) 2745 break; 2746 2747 /* process next memory node range */ 2748 ms++; 2749 base_pfn = mem_node_config[ms].physbase; 2750 num = MIN(mem_node_config[ms].physmax, 2751 end_pfn) - base_pfn + 1; 2752 } 2753 } 2754 2755 PRM_DEBUG(availrmem_initial); 2756 PRM_DEBUG(availrmem); 2757 PRM_DEBUG(freemem); 2758 build_pfn_hash(); 2759 return (pages_done); 2760 } 2761 2762 /* 2763 * Kernel VM initialization. 2764 */ 2765 static void 2766 kvm_init(void) 2767 { 2768 ASSERT((((uintptr_t)s_text) & MMU_PAGEOFFSET) == 0); 2769 2770 /* 2771 * Put the kernel segments in kernel address space. 2772 */ 2773 rw_enter(&kas.a_lock, RW_WRITER); 2774 as_avlinit(&kas); 2775 2776 (void) seg_attach(&kas, s_text, e_moddata - s_text, &ktextseg); 2777 (void) segkmem_create(&ktextseg); 2778 2779 (void) seg_attach(&kas, (caddr_t)valloc_base, valloc_sz, &kvalloc); 2780 (void) segkmem_create(&kvalloc); 2781 2782 (void) seg_attach(&kas, kernelheap, 2783 ekernelheap - kernelheap, &kvseg); 2784 (void) segkmem_create(&kvseg); 2785 2786 if (core_size > 0) { 2787 PRM_POINT("attaching kvseg_core"); 2788 (void) seg_attach(&kas, (caddr_t)core_base, core_size, 2789 &kvseg_core); 2790 (void) segkmem_create(&kvseg_core); 2791 } 2792 2793 if (segziosize > 0) { 2794 PRM_POINT("attaching segzio"); 2795 (void) seg_attach(&kas, segzio_base, mmu_ptob(segziosize), 2796 &kzioseg); 2797 (void) segkmem_zio_create(&kzioseg); 2798 2799 /* create zio area covering new segment */ 2800 segkmem_zio_init(segzio_base, mmu_ptob(segziosize)); 2801 } 2802 2803 (void) seg_attach(&kas, kdi_segdebugbase, kdi_segdebugsize, &kdebugseg); 2804 (void) segkmem_create(&kdebugseg); 2805 2806 rw_exit(&kas.a_lock); 2807 2808 /* 2809 * Ensure that the red zone at kernelbase is never accessible. 2810 */ 2811 PRM_POINT("protecting redzone"); 2812 (void) as_setprot(&kas, (caddr_t)kernelbase, KERNEL_REDZONE_SIZE, 0); 2813 2814 /* 2815 * Make the text writable so that it can be hot patched by DTrace. 2816 */ 2817 (void) as_setprot(&kas, s_text, e_modtext - s_text, 2818 PROT_READ | PROT_WRITE | PROT_EXEC); 2819 2820 /* 2821 * Make data writable until end. 2822 */ 2823 (void) as_setprot(&kas, s_data, e_moddata - s_data, 2824 PROT_READ | PROT_WRITE | PROT_EXEC); 2825 } 2826 2827 #ifndef __xpv 2828 /* 2829 * Solaris adds an entry for Write Combining caching to the PAT 2830 */ 2831 static uint64_t pat_attr_reg = PAT_DEFAULT_ATTRIBUTE; 2832 2833 void 2834 pat_sync(void) 2835 { 2836 ulong_t cr0, cr0_orig, cr4; 2837 2838 if (!is_x86_feature(x86_featureset, X86FSET_PAT)) 2839 return; 2840 cr0_orig = cr0 = getcr0(); 2841 cr4 = getcr4(); 2842 2843 /* disable caching and flush all caches and TLBs */ 2844 cr0 |= CR0_CD; 2845 cr0 &= ~CR0_NW; 2846 setcr0(cr0); 2847 invalidate_cache(); 2848 if (cr4 & CR4_PGE) { 2849 setcr4(cr4 & ~(ulong_t)CR4_PGE); 2850 setcr4(cr4); 2851 } else { 2852 reload_cr3(); 2853 } 2854 2855 /* add our entry to the PAT */ 2856 wrmsr(REG_PAT, pat_attr_reg); 2857 2858 /* flush TLBs and cache again, then reenable cr0 caching */ 2859 if (cr4 & CR4_PGE) { 2860 setcr4(cr4 & ~(ulong_t)CR4_PGE); 2861 setcr4(cr4); 2862 } else { 2863 reload_cr3(); 2864 } 2865 invalidate_cache(); 2866 setcr0(cr0_orig); 2867 } 2868 2869 #endif /* !__xpv */ 2870 2871 #if defined(_SOFT_HOSTID) 2872 /* 2873 * On platforms that do not have a hardware serial number, attempt 2874 * to set one based on the contents of /etc/hostid. If this file does 2875 * not exist, assume that we are to generate a new hostid and set 2876 * it in the kernel, for subsequent saving by a userland process 2877 * once the system is up and the root filesystem is mounted r/w. 2878 * 2879 * In order to gracefully support upgrade on OpenSolaris, if 2880 * /etc/hostid does not exist, we will attempt to get a serial number 2881 * using the legacy method (/kernel/misc/sysinit). 2882 * 2883 * If that isn't present, we attempt to use an SMBIOS UUID, which is 2884 * a hardware serial number. Note that we don't automatically trust 2885 * all SMBIOS UUIDs (some older platforms are defective and ship duplicate 2886 * UUIDs in violation of the standard), we check against a blacklist. 2887 * 2888 * In an attempt to make the hostid less prone to abuse 2889 * (for license circumvention, etc), we store it in /etc/hostid 2890 * in rot47 format. 2891 */ 2892 extern volatile unsigned long tenmicrodata; 2893 static int atoi(char *); 2894 2895 /* 2896 * Set this to non-zero in /etc/system if you think your SMBIOS returns a 2897 * UUID that is not unique. (Also report it so that the smbios_uuid_blacklist 2898 * array can be updated.) 2899 */ 2900 int smbios_broken_uuid = 0; 2901 2902 /* 2903 * List of known bad UUIDs. This is just the lower 32-bit values, since 2904 * that's what we use for the host id. If your hostid falls here, you need 2905 * to contact your hardware OEM for a fix for your BIOS. 2906 */ 2907 static unsigned char 2908 smbios_uuid_blacklist[][16] = { 2909 2910 { /* Reported bad UUID (Google search) */ 2911 0x00, 0x02, 0x00, 0x03, 0x00, 0x04, 0x00, 0x05, 2912 0x00, 0x06, 0x00, 0x07, 0x00, 0x08, 0x00, 0x09, 2913 }, 2914 { /* Known bad DELL UUID */ 2915 0x4C, 0x4C, 0x45, 0x44, 0x00, 0x00, 0x20, 0x10, 2916 0x80, 0x20, 0x80, 0xC0, 0x4F, 0x20, 0x20, 0x20, 2917 }, 2918 { /* Uninitialized flash */ 2919 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 2920 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff 2921 }, 2922 { /* All zeros */ 2923 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 2924 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00 2925 }, 2926 }; 2927 2928 static int32_t 2929 uuid_to_hostid(const uint8_t *uuid) 2930 { 2931 /* 2932 * Although the UUIDs are 128-bits, they may not distribute entropy 2933 * evenly. We would like to use SHA or MD5, but those are located 2934 * in loadable modules and not available this early in boot. As we 2935 * don't need the values to be cryptographically strong, we just 2936 * generate 32-bit vaue by xor'ing the various sequences together, 2937 * which ensures that the entire UUID contributes to the hostid. 2938 */ 2939 uint32_t id = 0; 2940 2941 /* first check against the blacklist */ 2942 for (int i = 0; i < (sizeof (smbios_uuid_blacklist) / 16); i++) { 2943 if (bcmp(smbios_uuid_blacklist[0], uuid, 16) == 0) { 2944 cmn_err(CE_CONT, "?Broken SMBIOS UUID. " 2945 "Contact BIOS manufacturer for repair.\n"); 2946 return ((int32_t)HW_INVALID_HOSTID); 2947 } 2948 } 2949 2950 for (int i = 0; i < 16; i++) 2951 id ^= ((uuid[i]) << (8 * (i % sizeof (id)))); 2952 2953 /* Make sure return value is positive */ 2954 return (id & 0x7fffffff); 2955 } 2956 2957 static int32_t 2958 set_soft_hostid(void) 2959 { 2960 struct _buf *file; 2961 char tokbuf[MAXNAMELEN]; 2962 token_t token; 2963 int done = 0; 2964 u_longlong_t tmp; 2965 int i; 2966 int32_t hostid = (int32_t)HW_INVALID_HOSTID; 2967 unsigned char *c; 2968 hrtime_t tsc; 2969 smbios_system_t smsys; 2970 2971 /* 2972 * If /etc/hostid file not found, we'd like to get a pseudo 2973 * random number to use at the hostid. A nice way to do this 2974 * is to read the real time clock. To remain xen-compatible, 2975 * we can't poke the real hardware, so we use tsc_read() to 2976 * read the real time clock. However, there is an ominous 2977 * warning in tsc_read that says it can return zero, so we 2978 * deal with that possibility by falling back to using the 2979 * (hopefully random enough) value in tenmicrodata. 2980 */ 2981 2982 if ((file = kobj_open_file(hostid_file)) == (struct _buf *)-1) { 2983 /* 2984 * hostid file not found - try to load sysinit module 2985 * and see if it has a nonzero hostid value...use that 2986 * instead of generating a new hostid here if so. 2987 */ 2988 if ((i = modload("misc", "sysinit")) != -1) { 2989 if (strlen(hw_serial) > 0) 2990 hostid = (int32_t)atoi(hw_serial); 2991 (void) modunload(i); 2992 } 2993 2994 /* 2995 * We try to use the SMBIOS UUID. But not if it is blacklisted 2996 * in /etc/system. 2997 */ 2998 if ((hostid == HW_INVALID_HOSTID) && 2999 (smbios_broken_uuid == 0) && 3000 (ksmbios != NULL) && 3001 (smbios_info_system(ksmbios, &smsys) != SMB_ERR) && 3002 (smsys.smbs_uuidlen >= 16)) { 3003 hostid = uuid_to_hostid(smsys.smbs_uuid); 3004 } 3005 3006 /* 3007 * Generate a "random" hostid using the clock. These 3008 * hostids will change on each boot if the value is not 3009 * saved to a persistent /etc/hostid file. 3010 */ 3011 if (hostid == HW_INVALID_HOSTID) { 3012 tsc = tsc_read(); 3013 if (tsc == 0) /* tsc_read can return zero sometimes */ 3014 hostid = (int32_t)tenmicrodata & 0x0CFFFFF; 3015 else 3016 hostid = (int32_t)tsc & 0x0CFFFFF; 3017 } 3018 } else { 3019 /* hostid file found */ 3020 while (!done) { 3021 token = kobj_lex(file, tokbuf, sizeof (tokbuf)); 3022 3023 switch (token) { 3024 case POUND: 3025 /* 3026 * skip comments 3027 */ 3028 kobj_find_eol(file); 3029 break; 3030 case STRING: 3031 /* 3032 * un-rot47 - obviously this 3033 * nonsense is ascii-specific 3034 */ 3035 for (c = (unsigned char *)tokbuf; 3036 *c != '\0'; c++) { 3037 *c += 47; 3038 if (*c > '~') 3039 *c -= 94; 3040 else if (*c < '!') 3041 *c += 94; 3042 } 3043 /* 3044 * now we should have a real number 3045 */ 3046 3047 if (kobj_getvalue(tokbuf, &tmp) != 0) 3048 kobj_file_err(CE_WARN, file, 3049 "Bad value %s for hostid", 3050 tokbuf); 3051 else 3052 hostid = (int32_t)tmp; 3053 3054 break; 3055 case EOF: 3056 done = 1; 3057 /* FALLTHROUGH */ 3058 case NEWLINE: 3059 kobj_newline(file); 3060 break; 3061 default: 3062 break; 3063 3064 } 3065 } 3066 if (hostid == HW_INVALID_HOSTID) /* didn't find a hostid */ 3067 kobj_file_err(CE_WARN, file, 3068 "hostid missing or corrupt"); 3069 3070 kobj_close_file(file); 3071 } 3072 /* 3073 * hostid is now the value read from /etc/hostid, or the 3074 * new hostid we generated in this routine or HW_INVALID_HOSTID if not 3075 * set. 3076 */ 3077 return (hostid); 3078 } 3079 3080 static int 3081 atoi(char *p) 3082 { 3083 int i = 0; 3084 3085 while (*p != '\0') 3086 i = 10 * i + (*p++ - '0'); 3087 3088 return (i); 3089 } 3090 3091 #endif /* _SOFT_HOSTID */ 3092 3093 void 3094 get_system_configuration(void) 3095 { 3096 char prop[32]; 3097 u_longlong_t nodes_ll, cpus_pernode_ll, lvalue; 3098 3099 if (BOP_GETPROPLEN(bootops, "nodes") > sizeof (prop) || 3100 BOP_GETPROP(bootops, "nodes", prop) < 0 || 3101 kobj_getvalue(prop, &nodes_ll) == -1 || 3102 nodes_ll > MAXNODES || 3103 BOP_GETPROPLEN(bootops, "cpus_pernode") > sizeof (prop) || 3104 BOP_GETPROP(bootops, "cpus_pernode", prop) < 0 || 3105 kobj_getvalue(prop, &cpus_pernode_ll) == -1) { 3106 system_hardware.hd_nodes = 1; 3107 system_hardware.hd_cpus_per_node = 0; 3108 } else { 3109 system_hardware.hd_nodes = (int)nodes_ll; 3110 system_hardware.hd_cpus_per_node = (int)cpus_pernode_ll; 3111 } 3112 3113 if (BOP_GETPROPLEN(bootops, "kernelbase") > sizeof (prop) || 3114 BOP_GETPROP(bootops, "kernelbase", prop) < 0 || 3115 kobj_getvalue(prop, &lvalue) == -1) 3116 eprom_kernelbase = NULL; 3117 else 3118 eprom_kernelbase = (uintptr_t)lvalue; 3119 3120 if (BOP_GETPROPLEN(bootops, "segmapsize") > sizeof (prop) || 3121 BOP_GETPROP(bootops, "segmapsize", prop) < 0 || 3122 kobj_getvalue(prop, &lvalue) == -1) 3123 segmapsize = SEGMAPDEFAULT; 3124 else 3125 segmapsize = (uintptr_t)lvalue; 3126 3127 if (BOP_GETPROPLEN(bootops, "segmapfreelists") > sizeof (prop) || 3128 BOP_GETPROP(bootops, "segmapfreelists", prop) < 0 || 3129 kobj_getvalue(prop, &lvalue) == -1) 3130 segmapfreelists = 0; /* use segmap driver default */ 3131 else 3132 segmapfreelists = (int)lvalue; 3133 3134 /* physmem used to be here, but moved much earlier to fakebop.c */ 3135 } 3136 3137 /* 3138 * Add to a memory list. 3139 * start = start of new memory segment 3140 * len = length of new memory segment in bytes 3141 * new = pointer to a new struct memlist 3142 * memlistp = memory list to which to add segment. 3143 */ 3144 void 3145 memlist_add( 3146 uint64_t start, 3147 uint64_t len, 3148 struct memlist *new, 3149 struct memlist **memlistp) 3150 { 3151 struct memlist *cur; 3152 uint64_t end = start + len; 3153 3154 new->ml_address = start; 3155 new->ml_size = len; 3156 3157 cur = *memlistp; 3158 3159 while (cur) { 3160 if (cur->ml_address >= end) { 3161 new->ml_next = cur; 3162 *memlistp = new; 3163 new->ml_prev = cur->ml_prev; 3164 cur->ml_prev = new; 3165 return; 3166 } 3167 ASSERT(cur->ml_address + cur->ml_size <= start); 3168 if (cur->ml_next == NULL) { 3169 cur->ml_next = new; 3170 new->ml_prev = cur; 3171 new->ml_next = NULL; 3172 return; 3173 } 3174 memlistp = &cur->ml_next; 3175 cur = cur->ml_next; 3176 } 3177 } 3178 3179 void 3180 kobj_vmem_init(vmem_t **text_arena, vmem_t **data_arena) 3181 { 3182 size_t tsize = e_modtext - modtext; 3183 size_t dsize = e_moddata - moddata; 3184 3185 *text_arena = vmem_create("module_text", tsize ? modtext : NULL, tsize, 3186 1, segkmem_alloc, segkmem_free, heaptext_arena, 0, VM_SLEEP); 3187 *data_arena = vmem_create("module_data", dsize ? moddata : NULL, dsize, 3188 1, segkmem_alloc, segkmem_free, heap32_arena, 0, VM_SLEEP); 3189 } 3190 3191 caddr_t 3192 kobj_text_alloc(vmem_t *arena, size_t size) 3193 { 3194 return (vmem_alloc(arena, size, VM_SLEEP | VM_BESTFIT)); 3195 } 3196 3197 /*ARGSUSED*/ 3198 caddr_t 3199 kobj_texthole_alloc(caddr_t addr, size_t size) 3200 { 3201 panic("unexpected call to kobj_texthole_alloc()"); 3202 /*NOTREACHED*/ 3203 return (0); 3204 } 3205 3206 /*ARGSUSED*/ 3207 void 3208 kobj_texthole_free(caddr_t addr, size_t size) 3209 { 3210 panic("unexpected call to kobj_texthole_free()"); 3211 } 3212 3213 /* 3214 * This is called just after configure() in startup(). 3215 * 3216 * The ISALIST concept is a bit hopeless on Intel, because 3217 * there's no guarantee of an ever-more-capable processor 3218 * given that various parts of the instruction set may appear 3219 * and disappear between different implementations. 3220 * 3221 * While it would be possible to correct it and even enhance 3222 * it somewhat, the explicit hardware capability bitmask allows 3223 * more flexibility. 3224 * 3225 * So, we just leave this alone. 3226 */ 3227 void 3228 setx86isalist(void) 3229 { 3230 char *tp; 3231 size_t len; 3232 extern char *isa_list; 3233 3234 #define TBUFSIZE 1024 3235 3236 tp = kmem_alloc(TBUFSIZE, KM_SLEEP); 3237 *tp = '\0'; 3238 3239 #if defined(__amd64) 3240 (void) strcpy(tp, "amd64 "); 3241 #endif 3242 3243 switch (x86_vendor) { 3244 case X86_VENDOR_Intel: 3245 case X86_VENDOR_AMD: 3246 case X86_VENDOR_TM: 3247 if (is_x86_feature(x86_featureset, X86FSET_CMOV)) { 3248 /* 3249 * Pentium Pro or later 3250 */ 3251 (void) strcat(tp, "pentium_pro"); 3252 (void) strcat(tp, 3253 is_x86_feature(x86_featureset, X86FSET_MMX) ? 3254 "+mmx pentium_pro " : " "); 3255 } 3256 /*FALLTHROUGH*/ 3257 case X86_VENDOR_Cyrix: 3258 /* 3259 * The Cyrix 6x86 does not have any Pentium features 3260 * accessible while not at privilege level 0. 3261 */ 3262 if (is_x86_feature(x86_featureset, X86FSET_CPUID)) { 3263 (void) strcat(tp, "pentium"); 3264 (void) strcat(tp, 3265 is_x86_feature(x86_featureset, X86FSET_MMX) ? 3266 "+mmx pentium " : " "); 3267 } 3268 break; 3269 default: 3270 break; 3271 } 3272 (void) strcat(tp, "i486 i386 i86"); 3273 len = strlen(tp) + 1; /* account for NULL at end of string */ 3274 isa_list = strcpy(kmem_alloc(len, KM_SLEEP), tp); 3275 kmem_free(tp, TBUFSIZE); 3276 3277 #undef TBUFSIZE 3278 } 3279 3280 3281 #ifdef __amd64 3282 3283 void * 3284 device_arena_alloc(size_t size, int vm_flag) 3285 { 3286 return (vmem_alloc(device_arena, size, vm_flag)); 3287 } 3288 3289 void 3290 device_arena_free(void *vaddr, size_t size) 3291 { 3292 vmem_free(device_arena, vaddr, size); 3293 } 3294 3295 #else /* __i386 */ 3296 3297 void * 3298 device_arena_alloc(size_t size, int vm_flag) 3299 { 3300 caddr_t vaddr; 3301 uintptr_t v; 3302 size_t start; 3303 size_t end; 3304 3305 vaddr = vmem_alloc(heap_arena, size, vm_flag); 3306 if (vaddr == NULL) 3307 return (NULL); 3308 3309 v = (uintptr_t)vaddr; 3310 ASSERT(v >= kernelbase); 3311 ASSERT(v + size <= valloc_base); 3312 3313 start = btop(v - kernelbase); 3314 end = btop(v + size - 1 - kernelbase); 3315 ASSERT(start < toxic_bit_map_len); 3316 ASSERT(end < toxic_bit_map_len); 3317 3318 while (start <= end) { 3319 BT_ATOMIC_SET(toxic_bit_map, start); 3320 ++start; 3321 } 3322 return (vaddr); 3323 } 3324 3325 void 3326 device_arena_free(void *vaddr, size_t size) 3327 { 3328 uintptr_t v = (uintptr_t)vaddr; 3329 size_t start; 3330 size_t end; 3331 3332 ASSERT(v >= kernelbase); 3333 ASSERT(v + size <= valloc_base); 3334 3335 start = btop(v - kernelbase); 3336 end = btop(v + size - 1 - kernelbase); 3337 ASSERT(start < toxic_bit_map_len); 3338 ASSERT(end < toxic_bit_map_len); 3339 3340 while (start <= end) { 3341 ASSERT(BT_TEST(toxic_bit_map, start) != 0); 3342 BT_ATOMIC_CLEAR(toxic_bit_map, start); 3343 ++start; 3344 } 3345 vmem_free(heap_arena, vaddr, size); 3346 } 3347 3348 /* 3349 * returns 1st address in range that is in device arena, or NULL 3350 * if len is not NULL it returns the length of the toxic range 3351 */ 3352 void * 3353 device_arena_contains(void *vaddr, size_t size, size_t *len) 3354 { 3355 uintptr_t v = (uintptr_t)vaddr; 3356 uintptr_t eaddr = v + size; 3357 size_t start; 3358 size_t end; 3359 3360 /* 3361 * if called very early by kmdb, just return NULL 3362 */ 3363 if (toxic_bit_map == NULL) 3364 return (NULL); 3365 3366 /* 3367 * First check if we're completely outside the bitmap range. 3368 */ 3369 if (v >= valloc_base || eaddr < kernelbase) 3370 return (NULL); 3371 3372 /* 3373 * Trim ends of search to look at only what the bitmap covers. 3374 */ 3375 if (v < kernelbase) 3376 v = kernelbase; 3377 start = btop(v - kernelbase); 3378 end = btop(eaddr - kernelbase); 3379 if (end >= toxic_bit_map_len) 3380 end = toxic_bit_map_len; 3381 3382 if (bt_range(toxic_bit_map, &start, &end, end) == 0) 3383 return (NULL); 3384 3385 v = kernelbase + ptob(start); 3386 if (len != NULL) 3387 *len = ptob(end - start); 3388 return ((void *)v); 3389 } 3390 3391 #endif /* __i386 */ 3392