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 /* 23 * Copyright 2007 Sun Microsystems, Inc. All rights reserved. 24 * Use is subject to license terms. 25 */ 26 27 #pragma ident "%Z%%M% %I% %E% SMI" 28 29 #include <sys/machsystm.h> 30 #include <sys/archsystm.h> 31 #include <sys/vm.h> 32 #include <sys/cpu.h> 33 #include <sys/atomic.h> 34 #include <sys/reboot.h> 35 #include <sys/kdi.h> 36 #include <sys/bootconf.h> 37 #include <sys/memlist_plat.h> 38 #include <sys/memlist_impl.h> 39 #include <sys/prom_plat.h> 40 #include <sys/prom_isa.h> 41 #include <sys/autoconf.h> 42 #include <sys/intreg.h> 43 #include <sys/ivintr.h> 44 #include <sys/fpu/fpusystm.h> 45 #include <sys/iommutsb.h> 46 #include <vm/vm_dep.h> 47 #include <vm/seg_dev.h> 48 #include <vm/seg_kmem.h> 49 #include <vm/seg_kpm.h> 50 #include <vm/seg_map.h> 51 #include <vm/seg_kp.h> 52 #include <sys/sysconf.h> 53 #include <vm/hat_sfmmu.h> 54 #include <sys/kobj.h> 55 #include <sys/sun4asi.h> 56 #include <sys/clconf.h> 57 #include <sys/platform_module.h> 58 #include <sys/panic.h> 59 #include <sys/cpu_sgnblk_defs.h> 60 #include <sys/clock.h> 61 #include <sys/cmn_err.h> 62 #include <sys/promif.h> 63 #include <sys/prom_debug.h> 64 #include <sys/traptrace.h> 65 #include <sys/memnode.h> 66 #include <sys/mem_cage.h> 67 #include <sys/mmu.h> 68 69 extern void setup_trap_table(void); 70 extern void cpu_intrq_setup(struct cpu *); 71 extern void cpu_intrq_register(struct cpu *); 72 extern void contig_mem_init(void); 73 extern void mach_dump_buffer_init(void); 74 extern void mach_descrip_init(void); 75 extern void mach_descrip_startup_fini(void); 76 extern void mach_memscrub(void); 77 extern void mach_fpras(void); 78 extern void mach_cpu_halt_idle(void); 79 extern void mach_hw_copy_limit(void); 80 extern void load_mach_drivers(void); 81 extern void load_tod_module(void); 82 #pragma weak load_tod_module 83 84 extern int ndata_alloc_mmfsa(struct memlist *ndata); 85 #pragma weak ndata_alloc_mmfsa 86 87 extern void cif_init(void); 88 #pragma weak cif_init 89 90 extern void parse_idprom(void); 91 extern void add_vx_handler(char *, int, void (*)(cell_t *)); 92 extern void mem_config_init(void); 93 extern void memseg_remap_init(void); 94 95 extern void mach_kpm_init(void); 96 97 /* 98 * External Data: 99 */ 100 extern int vac_size; /* cache size in bytes */ 101 extern uint_t vac_mask; /* VAC alignment consistency mask */ 102 extern uint_t vac_colors; 103 104 /* 105 * Global Data Definitions: 106 */ 107 108 /* 109 * XXX - Don't port this to new architectures 110 * A 3rd party volume manager driver (vxdm) depends on the symbol romp. 111 * 'romp' has no use with a prom with an IEEE 1275 client interface. 112 * The driver doesn't use the value, but it depends on the symbol. 113 */ 114 void *romp; /* veritas driver won't load without romp 4154976 */ 115 /* 116 * Declare these as initialized data so we can patch them. 117 */ 118 pgcnt_t physmem = 0; /* memory size in pages, patch if you want less */ 119 pgcnt_t segkpsize = 120 btop(SEGKPDEFSIZE); /* size of segkp segment in pages */ 121 uint_t segmap_percent = 12; /* Size of segmap segment */ 122 123 int use_cache = 1; /* cache not reliable (605 bugs) with MP */ 124 int vac_copyback = 1; 125 char *cache_mode = NULL; 126 int use_mix = 1; 127 int prom_debug = 0; 128 129 struct bootops *bootops = 0; /* passed in from boot in %o2 */ 130 caddr_t boot_tba; /* %tba at boot - used by kmdb */ 131 uint_t tba_taken_over = 0; 132 133 caddr_t s_text; /* start of kernel text segment */ 134 caddr_t e_text; /* end of kernel text segment */ 135 caddr_t s_data; /* start of kernel data segment */ 136 caddr_t e_data; /* end of kernel data segment */ 137 138 caddr_t modtext; /* beginning of module text */ 139 size_t modtext_sz; /* size of module text */ 140 caddr_t moddata; /* beginning of module data reserve */ 141 caddr_t e_moddata; /* end of module data reserve */ 142 143 /* 144 * End of first block of contiguous kernel in 32-bit virtual address space 145 */ 146 caddr_t econtig32; /* end of first blk of contiguous kernel */ 147 148 caddr_t ncbase; /* beginning of non-cached segment */ 149 caddr_t ncend; /* end of non-cached segment */ 150 caddr_t sdata; /* beginning of data segment */ 151 152 caddr_t extra_etva; /* beginning of unused nucleus text */ 153 pgcnt_t extra_etpg; /* number of pages of unused nucleus text */ 154 155 size_t ndata_remain_sz; /* bytes from end of data to 4MB boundary */ 156 caddr_t nalloc_base; /* beginning of nucleus allocation */ 157 caddr_t nalloc_end; /* end of nucleus allocatable memory */ 158 caddr_t valloc_base; /* beginning of kvalloc segment */ 159 160 caddr_t kmem64_base; /* base of kernel mem segment in 64-bit space */ 161 caddr_t kmem64_end; /* end of kernel mem segment in 64-bit space */ 162 163 uintptr_t shm_alignment; /* VAC address consistency modulus */ 164 struct memlist *phys_install; /* Total installed physical memory */ 165 struct memlist *phys_avail; /* Available (unreserved) physical memory */ 166 struct memlist *virt_avail; /* Available (unmapped?) virtual memory */ 167 struct memlist ndata; /* memlist of nucleus allocatable memory */ 168 int memexp_flag; /* memory expansion card flag */ 169 uint64_t ecache_flushaddr; /* physical address used for flushing E$ */ 170 pgcnt_t obp_pages; /* Physical pages used by OBP */ 171 172 /* 173 * VM data structures 174 */ 175 long page_hashsz; /* Size of page hash table (power of two) */ 176 struct page *pp_base; /* Base of system page struct array */ 177 size_t pp_sz; /* Size in bytes of page struct array */ 178 struct page **page_hash; /* Page hash table */ 179 struct seg ktextseg; /* Segment used for kernel executable image */ 180 struct seg kvalloc; /* Segment used for "valloc" mapping */ 181 struct seg kpseg; /* Segment used for pageable kernel virt mem */ 182 struct seg ktexthole; /* Segment used for nucleus text hole */ 183 struct seg kmapseg; /* Segment used for generic kernel mappings */ 184 struct seg kpmseg; /* Segment used for physical mapping */ 185 struct seg kdebugseg; /* Segment used for the kernel debugger */ 186 187 uintptr_t kpm_pp_base; /* Base of system kpm_page array */ 188 size_t kpm_pp_sz; /* Size of system kpm_page array */ 189 pgcnt_t kpm_npages; /* How many kpm pages are managed */ 190 191 struct seg *segkp = &kpseg; /* Pageable kernel virtual memory segment */ 192 struct seg *segkmap = &kmapseg; /* Kernel generic mapping segment */ 193 struct seg *segkpm = &kpmseg; /* 64bit kernel physical mapping segment */ 194 195 int segzio_fromheap = 0; /* zio allocations occur from heap */ 196 caddr_t segzio_base; /* Base address of segzio */ 197 pgcnt_t segziosize = 0; /* size of zio segment in pages */ 198 199 /* 200 * debugger pages (if allocated) 201 */ 202 struct vnode kdebugvp; 203 204 /* 205 * VA range available to the debugger 206 */ 207 const caddr_t kdi_segdebugbase = (const caddr_t)SEGDEBUGBASE; 208 const size_t kdi_segdebugsize = SEGDEBUGSIZE; 209 210 /* 211 * Segment for relocated kernel structures in 64-bit large RAM kernels 212 */ 213 struct seg kmem64; 214 215 struct memseg *memseg_base; 216 size_t memseg_sz; /* Used to translate a va to page */ 217 struct vnode unused_pages_vp; 218 219 /* 220 * VM data structures allocated early during boot. 221 */ 222 size_t pagehash_sz; 223 uint64_t memlist_sz; 224 225 char tbr_wr_addr_inited = 0; 226 227 228 /* 229 * Static Routines: 230 */ 231 static void memlist_add(uint64_t, uint64_t, struct memlist **, 232 struct memlist **); 233 static void kphysm_init(page_t *, struct memseg *, pgcnt_t, uintptr_t, 234 pgcnt_t); 235 static void kvm_init(void); 236 237 static void startup_init(void); 238 static void startup_memlist(void); 239 static void startup_modules(void); 240 static void startup_bop_gone(void); 241 static void startup_vm(void); 242 static void startup_end(void); 243 static void setup_cage_params(void); 244 static void startup_create_io_node(void); 245 246 static pgcnt_t npages; 247 static struct memlist *memlist; 248 void *memlist_end; 249 250 static pgcnt_t bop_alloc_pages; 251 static caddr_t hblk_base; 252 uint_t hblk_alloc_dynamic = 0; 253 uint_t hblk1_min = H1MIN; 254 uint_t hblk8_min; 255 256 257 /* 258 * Hooks for unsupported platforms and down-rev firmware 259 */ 260 int iam_positron(void); 261 #pragma weak iam_positron 262 static void do_prom_version_check(void); 263 static void kpm_init(void); 264 static void kpm_npages_setup(int); 265 static void kpm_memseg_init(void); 266 267 /* 268 * After receiving a thermal interrupt, this is the number of seconds 269 * to delay before shutting off the system, assuming 270 * shutdown fails. Use /etc/system to change the delay if this isn't 271 * large enough. 272 */ 273 int thermal_powerdown_delay = 1200; 274 275 /* 276 * Used to hold off page relocations into the cage until OBP has completed 277 * its boot-time handoff of its resources to the kernel. 278 */ 279 int page_relocate_ready = 0; 280 281 /* 282 * Enable some debugging messages concerning memory usage... 283 */ 284 #ifdef DEBUGGING_MEM 285 static int debugging_mem; 286 static void 287 printmemlist(char *title, struct memlist *list) 288 { 289 if (!debugging_mem) 290 return; 291 292 printf("%s\n", title); 293 294 while (list) { 295 prom_printf("\taddr = 0x%x %8x, size = 0x%x %8x\n", 296 (uint32_t)(list->address >> 32), (uint32_t)list->address, 297 (uint32_t)(list->size >> 32), (uint32_t)(list->size)); 298 list = list->next; 299 } 300 } 301 302 void 303 printmemseg(struct memseg *memseg) 304 { 305 if (!debugging_mem) 306 return; 307 308 printf("memseg\n"); 309 310 while (memseg) { 311 prom_printf("\tpage = 0x%p, epage = 0x%p, " 312 "pfn = 0x%x, epfn = 0x%x\n", 313 memseg->pages, memseg->epages, 314 memseg->pages_base, memseg->pages_end); 315 memseg = memseg->next; 316 } 317 } 318 319 #define debug_pause(str) halt((str)) 320 #define MPRINTF(str) if (debugging_mem) prom_printf((str)) 321 #define MPRINTF1(str, a) if (debugging_mem) prom_printf((str), (a)) 322 #define MPRINTF2(str, a, b) if (debugging_mem) prom_printf((str), (a), (b)) 323 #define MPRINTF3(str, a, b, c) \ 324 if (debugging_mem) prom_printf((str), (a), (b), (c)) 325 #else /* DEBUGGING_MEM */ 326 #define MPRINTF(str) 327 #define MPRINTF1(str, a) 328 #define MPRINTF2(str, a, b) 329 #define MPRINTF3(str, a, b, c) 330 #endif /* DEBUGGING_MEM */ 331 332 /* Simple message to indicate that the bootops pointer has been zeroed */ 333 #ifdef DEBUG 334 static int bootops_gone_on = 0; 335 #define BOOTOPS_GONE() \ 336 if (bootops_gone_on) \ 337 prom_printf("The bootops vec is zeroed now!\n"); 338 #else 339 #define BOOTOPS_GONE() 340 #endif /* DEBUG */ 341 342 /* 343 * Monitor pages may not be where this says they are. 344 * and the debugger may not be there either. 345 * 346 * Note that 'pages' here are *physical* pages, which are 8k on sun4u. 347 * 348 * Physical memory layout 349 * (not necessarily contiguous) 350 * (THIS IS SOMEWHAT WRONG) 351 * /-----------------------\ 352 * | monitor pages | 353 * availmem -|-----------------------| 354 * | | 355 * | page pool | 356 * | | 357 * |-----------------------| 358 * | configured tables | 359 * | buffers | 360 * firstaddr -|-----------------------| 361 * | hat data structures | 362 * |-----------------------| 363 * | kernel data, bss | 364 * |-----------------------| 365 * | interrupt stack | 366 * |-----------------------| 367 * | kernel text (RO) | 368 * |-----------------------| 369 * | trap table (4k) | 370 * |-----------------------| 371 * page 1 | panicbuf | 372 * |-----------------------| 373 * page 0 | reclaimed | 374 * |_______________________| 375 * 376 * 377 * 378 * Kernel's Virtual Memory Layout. 379 * /-----------------------\ 380 * 0xFFFFFFFF.FFFFFFFF -| |- 381 * | OBP's virtual page | 382 * | tables | 383 * 0xFFFFFFFC.00000000 -|-----------------------|- 384 * : : 385 * : : 386 * -|-----------------------|- 387 * | segzio | (base and size vary) 388 * 0xFFFFFE00.00000000 -|-----------------------|- 389 * | | Ultrasparc I/II support 390 * | segkpm segment | up to 2TB of physical 391 * | (64-bit kernel ONLY) | memory, VAC has 2 colors 392 * | | 393 * 0xFFFFFA00.00000000 -|-----------------------|- 2TB segkpm alignment 394 * : : 395 * : : 396 * 0xFFFFF810.00000000 -|-----------------------|- hole_end 397 * | | ^ 398 * | UltraSPARC I/II call | | 399 * | bug requires an extra | | 400 * | 4 GB of space between | | 401 * | hole and used RAM | | 402 * | | | 403 * 0xFFFFF800.00000000 -|-----------------------|- | 404 * | | | 405 * | Virtual Address Hole | UltraSPARC 406 * | on UltraSPARC I/II | I/II * ONLY * 407 * | | | 408 * 0x00000800.00000000 -|-----------------------|- | 409 * | | | 410 * | UltraSPARC I/II call | | 411 * | bug requires an extra | | 412 * | 4 GB of space between | | 413 * | hole and used RAM | | 414 * | | v 415 * 0x000007FF.00000000 -|-----------------------|- hole_start ----- 416 * : : ^ 417 * : : | 418 * 0x00000XXX.XXXXXXXX -|-----------------------|- kmem64_end | 419 * | | | 420 * | 64-bit kernel ONLY | | 421 * | | | 422 * | kmem64 segment | | 423 * | | | 424 * | (Relocated extra HME | Approximately 425 * | block allocations, | 1 TB of virtual 426 * | memnode freelists, | address space 427 * | HME hash buckets, | | 428 * | mml_table, kpmp_table,| | 429 * | page_t array and | | 430 * | hashblock pool to | | 431 * | avoid hard-coded | | 432 * | 32-bit vaddr | | 433 * | limitations) | | 434 * | | v 435 * 0x00000700.00000000 -|-----------------------|- SYSLIMIT (kmem64_base) 436 * | | 437 * | segkmem segment | (SYSLIMIT - SYSBASE = 4TB) 438 * | | 439 * 0x00000300.00000000 -|-----------------------|- SYSBASE 440 * : : 441 * : : 442 * -|-----------------------|- 443 * | | 444 * | segmap segment | SEGMAPSIZE (1/8th physmem, 445 * | | 256G MAX) 446 * 0x000002a7.50000000 -|-----------------------|- SEGMAPBASE 447 * : : 448 * : : 449 * -|-----------------------|- 450 * | | 451 * | segkp | SEGKPSIZE (2GB) 452 * | | 453 * | | 454 * 0x000002a1.00000000 -|-----------------------|- SEGKPBASE 455 * | | 456 * 0x000002a0.00000000 -|-----------------------|- MEMSCRUBBASE 457 * | | (SEGKPBASE - 0x400000) 458 * 0x0000029F.FFE00000 -|-----------------------|- ARGSBASE 459 * | | (MEMSCRUBBASE - NCARGS) 460 * 0x0000029F.FFD80000 -|-----------------------|- PPMAPBASE 461 * | | (ARGSBASE - PPMAPSIZE) 462 * 0x0000029F.FFD00000 -|-----------------------|- PPMAP_FAST_BASE 463 * | | 464 * 0x0000029F.FF980000 -|-----------------------|- PIOMAPBASE 465 * | | 466 * 0x0000029F.FF580000 -|-----------------------|- NARG_BASE 467 * : : 468 * : : 469 * 0x00000000.FFFFFFFF -|-----------------------|- OFW_END_ADDR 470 * | | 471 * | OBP | 472 * | | 473 * 0x00000000.F0000000 -|-----------------------|- OFW_START_ADDR 474 * | kmdb | 475 * 0x00000000.EDD00000 -|-----------------------|- SEGDEBUGBASE 476 * : : 477 * : : 478 * 0x00000000.7c000000 -|-----------------------|- SYSLIMIT32 479 * | | 480 * | segkmem32 segment | (SYSLIMIT32 - SYSBASE32 = 481 * | | ~64MB) 482 * 0x00000000.78002000 -|-----------------------| 483 * | panicbuf | 484 * 0x00000000.78000000 -|-----------------------|- SYSBASE32 485 * : : 486 * : : 487 * | | 488 * |-----------------------|- econtig32 489 * | vm structures | 490 * 0x00000000.01C00000 |-----------------------|- nalloc_end 491 * | TSBs | 492 * |-----------------------|- end/nalloc_base 493 * | kernel data & bss | 494 * 0x00000000.01800000 -|-----------------------| 495 * : nucleus text hole : 496 * 0x00000000.01400000 -|-----------------------| 497 * : : 498 * |-----------------------| 499 * | module text | 500 * |-----------------------|- e_text/modtext 501 * | kernel text | 502 * |-----------------------| 503 * | trap table (48k) | 504 * 0x00000000.01000000 -|-----------------------|- KERNELBASE 505 * | reserved for trapstat |} TSTAT_TOTAL_SIZE 506 * |-----------------------| 507 * | | 508 * | invalid | 509 * | | 510 * 0x00000000.00000000 _|_______________________| 511 * 512 * 513 * 514 * 32-bit User Virtual Memory Layout. 515 * /-----------------------\ 516 * | | 517 * | invalid | 518 * | | 519 * 0xFFC00000 -|-----------------------|- USERLIMIT 520 * | user stack | 521 * : : 522 * : : 523 * : : 524 * | user data | 525 * -|-----------------------|- 526 * | user text | 527 * 0x00002000 -|-----------------------|- 528 * | invalid | 529 * 0x00000000 _|_______________________| 530 * 531 * 532 * 533 * 64-bit User Virtual Memory Layout. 534 * /-----------------------\ 535 * | | 536 * | invalid | 537 * | | 538 * 0xFFFFFFFF.80000000 -|-----------------------|- USERLIMIT 539 * | user stack | 540 * : : 541 * : : 542 * : : 543 * | user data | 544 * -|-----------------------|- 545 * | user text | 546 * 0x00000000.00100000 -|-----------------------|- 547 * | invalid | 548 * 0x00000000.00000000 _|_______________________| 549 */ 550 551 extern caddr_t ecache_init_scrub_flush_area(caddr_t alloc_base); 552 extern uint64_t ecache_flush_address(void); 553 554 #pragma weak load_platform_modules 555 #pragma weak plat_startup_memlist 556 #pragma weak ecache_init_scrub_flush_area 557 #pragma weak ecache_flush_address 558 559 560 /* 561 * By default the DR Cage is enabled for maximum OS 562 * MPSS performance. Users needing to disable the cage mechanism 563 * can set this variable to zero via /etc/system. 564 * Disabling the cage on systems supporting Dynamic Reconfiguration (DR) 565 * will result in loss of DR functionality. 566 * Platforms wishing to disable kernel Cage by default 567 * should do so in their set_platform_defaults() routine. 568 */ 569 int kernel_cage_enable = 1; 570 571 static void 572 setup_cage_params(void) 573 { 574 void (*func)(void); 575 576 func = (void (*)(void))kobj_getsymvalue("set_platform_cage_params", 0); 577 if (func != NULL) { 578 (*func)(); 579 return; 580 } 581 582 if (kernel_cage_enable == 0) { 583 return; 584 } 585 kcage_range_lock(); 586 if (kcage_range_init(phys_avail, 1) == 0) { 587 kcage_init(total_pages / 256); 588 } 589 kcage_range_unlock(); 590 591 if (kcage_on) { 592 cmn_err(CE_NOTE, "!Kernel Cage is ENABLED"); 593 } else { 594 cmn_err(CE_NOTE, "!Kernel Cage is DISABLED"); 595 } 596 597 } 598 599 /* 600 * Machine-dependent startup code 601 */ 602 void 603 startup(void) 604 { 605 startup_init(); 606 if (&startup_platform) 607 startup_platform(); 608 startup_memlist(); 609 startup_modules(); 610 setup_cage_params(); 611 startup_bop_gone(); 612 startup_vm(); 613 startup_end(); 614 } 615 616 struct regs sync_reg_buf; 617 uint64_t sync_tt; 618 619 void 620 sync_handler(void) 621 { 622 struct trap_info ti; 623 int i; 624 625 /* 626 * Prevent trying to talk to the other CPUs since they are 627 * sitting in the prom and won't reply. 628 */ 629 for (i = 0; i < NCPU; i++) { 630 if ((i != CPU->cpu_id) && CPU_XCALL_READY(i)) { 631 cpu[i]->cpu_flags &= ~CPU_READY; 632 cpu[i]->cpu_flags |= CPU_QUIESCED; 633 CPUSET_DEL(cpu_ready_set, cpu[i]->cpu_id); 634 } 635 } 636 637 /* 638 * We've managed to get here without going through the 639 * normal panic code path. Try and save some useful 640 * information. 641 */ 642 if (!panicstr && (curthread->t_panic_trap == NULL)) { 643 ti.trap_type = sync_tt; 644 ti.trap_regs = &sync_reg_buf; 645 ti.trap_addr = NULL; 646 ti.trap_mmu_fsr = 0x0; 647 648 curthread->t_panic_trap = &ti; 649 } 650 651 /* 652 * If we're re-entering the panic path, update the signature 653 * block so that the SC knows we're in the second part of panic. 654 */ 655 if (panicstr) 656 CPU_SIGNATURE(OS_SIG, SIGST_EXIT, SIGSUBST_DUMP, -1); 657 658 nopanicdebug = 1; /* do not perform debug_enter() prior to dump */ 659 panic("sync initiated"); 660 } 661 662 663 static void 664 startup_init(void) 665 { 666 /* 667 * We want to save the registers while we're still in OBP 668 * so that we know they haven't been fiddled with since. 669 * (In principle, OBP can't change them just because it 670 * makes a callback, but we'd rather not depend on that 671 * behavior.) 672 */ 673 char sync_str[] = 674 "warning @ warning off : sync " 675 "%%tl-c %%tstate h# %p x! " 676 "%%g1 h# %p x! %%g2 h# %p x! %%g3 h# %p x! " 677 "%%g4 h# %p x! %%g5 h# %p x! %%g6 h# %p x! " 678 "%%g7 h# %p x! %%o0 h# %p x! %%o1 h# %p x! " 679 "%%o2 h# %p x! %%o3 h# %p x! %%o4 h# %p x! " 680 "%%o5 h# %p x! %%o6 h# %p x! %%o7 h# %p x! " 681 "%%tl-c %%tpc h# %p x! %%tl-c %%tnpc h# %p x! " 682 "%%y h# %p l! %%tl-c %%tt h# %p x! " 683 "sync ; warning !"; 684 685 /* 686 * 20 == num of %p substrings 687 * 16 == max num of chars %p will expand to. 688 */ 689 char bp[sizeof (sync_str) + 16 * 20]; 690 691 (void) check_boot_version(BOP_GETVERSION(bootops)); 692 693 /* 694 * Initialize ptl1 stack for the 1st CPU. 695 */ 696 ptl1_init_cpu(&cpu0); 697 698 /* 699 * Initialize the address map for cache consistent mappings 700 * to random pages; must be done after vac_size is set. 701 */ 702 ppmapinit(); 703 704 /* 705 * Initialize the PROM callback handler. 706 */ 707 init_vx_handler(); 708 709 /* 710 * have prom call sync_callback() to handle the sync and 711 * save some useful information which will be stored in the 712 * core file later. 713 */ 714 (void) sprintf((char *)bp, sync_str, 715 (void *)&sync_reg_buf.r_tstate, (void *)&sync_reg_buf.r_g1, 716 (void *)&sync_reg_buf.r_g2, (void *)&sync_reg_buf.r_g3, 717 (void *)&sync_reg_buf.r_g4, (void *)&sync_reg_buf.r_g5, 718 (void *)&sync_reg_buf.r_g6, (void *)&sync_reg_buf.r_g7, 719 (void *)&sync_reg_buf.r_o0, (void *)&sync_reg_buf.r_o1, 720 (void *)&sync_reg_buf.r_o2, (void *)&sync_reg_buf.r_o3, 721 (void *)&sync_reg_buf.r_o4, (void *)&sync_reg_buf.r_o5, 722 (void *)&sync_reg_buf.r_o6, (void *)&sync_reg_buf.r_o7, 723 (void *)&sync_reg_buf.r_pc, (void *)&sync_reg_buf.r_npc, 724 (void *)&sync_reg_buf.r_y, (void *)&sync_tt); 725 prom_interpret(bp, 0, 0, 0, 0, 0); 726 add_vx_handler("sync", 1, (void (*)(cell_t *))sync_handler); 727 } 728 729 static u_longlong_t *boot_physinstalled, *boot_physavail, *boot_virtavail; 730 static size_t boot_physinstalled_len, boot_physavail_len, boot_virtavail_len; 731 732 #define IVSIZE ((MAXIVNUM * sizeof (intr_vec_t *)) + \ 733 (MAX_RSVD_IV * sizeof (intr_vec_t)) + \ 734 (MAX_RSVD_IVX * sizeof (intr_vecx_t))) 735 736 /* 737 * As OBP takes up some RAM when the system boots, pages will already be "lost" 738 * to the system and reflected in npages by the time we see it. 739 * 740 * We only want to allocate kernel structures in the 64-bit virtual address 741 * space on systems with enough RAM to make the overhead of keeping track of 742 * an extra kernel memory segment worthwhile. 743 * 744 * Since OBP has already performed its memory allocations by this point, if we 745 * have more than MINMOVE_RAM_MB MB of RAM left free, go ahead and map 746 * memory in the 64-bit virtual address space; otherwise keep allocations 747 * contiguous with we've mapped so far in the 32-bit virtual address space. 748 */ 749 #define MINMOVE_RAM_MB ((size_t)1900) 750 #define MB_TO_BYTES(mb) ((mb) * 1048576ul) 751 752 pgcnt_t tune_npages = (pgcnt_t) 753 (MB_TO_BYTES(MINMOVE_RAM_MB)/ (size_t)MMU_PAGESIZE); 754 755 static void 756 startup_memlist(void) 757 { 758 size_t alloc_sz; 759 size_t ctrs_sz; 760 caddr_t alloc_base; 761 caddr_t ctrs_base, ctrs_end; 762 caddr_t memspace; 763 caddr_t va; 764 int memblocks = 0; 765 struct memlist *cur; 766 size_t syslimit = (size_t)SYSLIMIT; 767 size_t sysbase = (size_t)SYSBASE; 768 int alloc_alignsize = MMU_PAGESIZE; 769 extern void page_coloring_init(void); 770 extern void page_set_colorequiv_arr(void); 771 772 /* 773 * Initialize enough of the system to allow kmem_alloc to work by 774 * calling boot to allocate its memory until the time that 775 * kvm_init is completed. The page structs are allocated after 776 * rounding up end to the nearest page boundary; the memsegs are 777 * initialized and the space they use comes from the kernel heap. 778 * With appropriate initialization, they can be reallocated later 779 * to a size appropriate for the machine's configuration. 780 * 781 * At this point, memory is allocated for things that will never 782 * need to be freed, this used to be "valloced". This allows a 783 * savings as the pages don't need page structures to describe 784 * them because them will not be managed by the vm system. 785 */ 786 787 /* 788 * We're loaded by boot with the following configuration (as 789 * specified in the sun4u/conf/Mapfile): 790 * 791 * text: 4 MB chunk aligned on a 4MB boundary 792 * data & bss: 4 MB chunk aligned on a 4MB boundary 793 * 794 * These two chunks will eventually be mapped by 2 locked 4MB 795 * ttes and will represent the nucleus of the kernel. This gives 796 * us some free space that is already allocated, some or all of 797 * which is made available to kernel module text. 798 * 799 * The free space in the data-bss chunk is used for nucleus 800 * allocatable data structures and we reserve it using the 801 * nalloc_base and nalloc_end variables. This space is currently 802 * being used for hat data structures required for tlb miss 803 * handling operations. We align nalloc_base to a l2 cache 804 * linesize because this is the line size the hardware uses to 805 * maintain cache coherency. 806 * 256K is carved out for module data. 807 */ 808 809 nalloc_base = (caddr_t)roundup((uintptr_t)e_data, MMU_PAGESIZE); 810 moddata = nalloc_base; 811 e_moddata = nalloc_base + MODDATA; 812 nalloc_base = e_moddata; 813 814 nalloc_end = (caddr_t)roundup((uintptr_t)nalloc_base, MMU_PAGESIZE4M); 815 valloc_base = nalloc_base; 816 817 /* 818 * Calculate the start of the data segment. 819 */ 820 sdata = (caddr_t)((uintptr_t)e_data & MMU_PAGEMASK4M); 821 822 PRM_DEBUG(moddata); 823 PRM_DEBUG(nalloc_base); 824 PRM_DEBUG(nalloc_end); 825 PRM_DEBUG(sdata); 826 827 /* 828 * Remember any slop after e_text so we can give it to the modules. 829 */ 830 PRM_DEBUG(e_text); 831 modtext = (caddr_t)roundup((uintptr_t)e_text, MMU_PAGESIZE); 832 if (((uintptr_t)modtext & MMU_PAGEMASK4M) != (uintptr_t)s_text) 833 panic("nucleus text overflow"); 834 modtext_sz = (caddr_t)roundup((uintptr_t)modtext, MMU_PAGESIZE4M) - 835 modtext; 836 PRM_DEBUG(modtext); 837 PRM_DEBUG(modtext_sz); 838 839 copy_boot_memlists(&boot_physinstalled, &boot_physinstalled_len, 840 &boot_physavail, &boot_physavail_len, 841 &boot_virtavail, &boot_virtavail_len); 842 /* 843 * Remember what the physically available highest page is 844 * so that dumpsys works properly, and find out how much 845 * memory is installed. 846 */ 847 installed_top_size_memlist_array(boot_physinstalled, 848 boot_physinstalled_len, &physmax, &physinstalled); 849 PRM_DEBUG(physinstalled); 850 PRM_DEBUG(physmax); 851 852 /* Fill out memory nodes config structure */ 853 startup_build_mem_nodes(boot_physinstalled, boot_physinstalled_len); 854 855 /* 856 * Get the list of physically available memory to size 857 * the number of page structures needed. 858 */ 859 size_physavail(boot_physavail, boot_physavail_len, &npages, &memblocks); 860 /* 861 * This first snap shot of npages can represent the pages used 862 * by OBP's text and data approximately. This is used in the 863 * the calculation of the kernel size 864 */ 865 obp_pages = physinstalled - npages; 866 867 868 /* 869 * On small-memory systems (<MODTEXT_SM_SIZE MB, currently 256MB), the 870 * in-nucleus module text is capped to MODTEXT_SM_CAP bytes (currently 871 * 2MB) and any excess pages are put on physavail. The assumption is 872 * that small-memory systems will need more pages more than they'll 873 * need efficiently-mapped module texts. 874 */ 875 if ((physinstalled < mmu_btop(MODTEXT_SM_SIZE << 20)) && 876 modtext_sz > MODTEXT_SM_CAP) { 877 extra_etpg = mmu_btop(modtext_sz - MODTEXT_SM_CAP); 878 modtext_sz = MODTEXT_SM_CAP; 879 } else 880 extra_etpg = 0; 881 PRM_DEBUG(extra_etpg); 882 PRM_DEBUG(modtext_sz); 883 extra_etva = modtext + modtext_sz; 884 PRM_DEBUG(extra_etva); 885 886 /* 887 * Account for any pages after e_text and e_data. 888 */ 889 npages += extra_etpg; 890 npages += mmu_btopr(nalloc_end - nalloc_base); 891 PRM_DEBUG(npages); 892 893 /* 894 * npages is the maximum of available physical memory possible. 895 * (ie. it will never be more than this) 896 */ 897 898 /* 899 * initialize the nucleus memory allocator. 900 */ 901 ndata_alloc_init(&ndata, (uintptr_t)nalloc_base, (uintptr_t)nalloc_end); 902 903 /* 904 * Allocate mmu fault status area from the nucleus data area. 905 */ 906 if ((&ndata_alloc_mmfsa != NULL) && (ndata_alloc_mmfsa(&ndata) != 0)) 907 cmn_err(CE_PANIC, "no more nucleus memory after mfsa alloc"); 908 909 /* 910 * Allocate kernel TSBs from the nucleus data area. 911 */ 912 if (ndata_alloc_tsbs(&ndata, npages) != 0) 913 cmn_err(CE_PANIC, "no more nucleus memory after tsbs alloc"); 914 915 /* 916 * Allocate dmv dispatch table from the nucleus data area. 917 */ 918 if (ndata_alloc_dmv(&ndata) != 0) 919 cmn_err(CE_PANIC, "no more nucleus memory after dmv alloc"); 920 921 922 page_coloring_init(); 923 924 /* 925 * Allocate page_freelists bin headers for memnode 0 from the 926 * nucleus data area. 927 */ 928 if (ndata_alloc_page_freelists(&ndata, 0) != 0) 929 cmn_err(CE_PANIC, 930 "no more nucleus memory after page free lists alloc"); 931 932 if (kpm_enable) { 933 kpm_init(); 934 /* 935 * kpm page space -- Update kpm_npages and make the 936 * same assumption about fragmenting as it is done 937 * for memseg_sz. 938 */ 939 kpm_npages_setup(memblocks + 4); 940 } 941 942 /* 943 * Allocate hat related structs from the nucleus data area. 944 */ 945 if (ndata_alloc_hat(&ndata, npages, kpm_npages) != 0) 946 cmn_err(CE_PANIC, "no more nucleus memory after hat alloc"); 947 948 /* 949 * We want to do the BOP_ALLOCs before the real allocation of page 950 * structs in order to not have to allocate page structs for this 951 * memory. We need to calculate a virtual address because we want 952 * the page structs to come before other allocations in virtual address 953 * space. This is so some (if not all) of page structs can actually 954 * live in the nucleus. 955 */ 956 957 /* 958 * WARNING WARNING WARNING WARNING WARNING WARNING WARNING 959 * 960 * There are comments all over the SFMMU code warning of dire 961 * consequences if the TSBs are moved out of 32-bit space. This 962 * is largely because the asm code uses "sethi %hi(addr)"-type 963 * instructions which will not provide the expected result if the 964 * address is a 64-bit one. 965 * 966 * WARNING WARNING WARNING WARNING WARNING WARNING WARNING 967 */ 968 alloc_base = (caddr_t)roundup((uintptr_t)nalloc_end, MMU_PAGESIZE); 969 alloc_base = sfmmu_ktsb_alloc(alloc_base); 970 alloc_base = (caddr_t)roundup((uintptr_t)alloc_base, ecache_alignsize); 971 PRM_DEBUG(alloc_base); 972 973 /* 974 * Allocate IOMMU TSB array. We do this here so that the physical 975 * memory gets deducted from the PROM's physical memory list. 976 */ 977 alloc_base = iommu_tsb_init(alloc_base); 978 alloc_base = (caddr_t)roundup((uintptr_t)alloc_base, 979 ecache_alignsize); 980 PRM_DEBUG(alloc_base); 981 982 /* 983 * Platforms like Starcat and OPL need special structures assigned in 984 * 32-bit virtual address space because their probing routines execute 985 * FCode, and FCode can't handle 64-bit virtual addresses... 986 */ 987 if (&plat_startup_memlist) { 988 alloc_base = plat_startup_memlist(alloc_base); 989 alloc_base = (caddr_t)roundup((uintptr_t)alloc_base, 990 ecache_alignsize); 991 PRM_DEBUG(alloc_base); 992 } 993 994 /* 995 * If we have enough memory, use 4M pages for alignment because it 996 * greatly reduces the number of TLB misses we take albeit at the cost 997 * of possible RAM wastage (degenerate case of 4 MB - MMU_PAGESIZE per 998 * allocation.) Still, the speedup on large memory systems (e.g. > 64 999 * GB) is quite noticeable, so it is worth the effort to do if we can. 1000 * 1001 * Note, however, that this speedup will only occur if the boot PROM 1002 * uses the largest possible MMU page size possible to map memory 1003 * requests that are properly aligned and sized (for example, a request 1004 * for a multiple of 4MB of memory aligned to a 4MB boundary will 1005 * result in a mapping using a 4MB MMU page.) 1006 * 1007 * Even then, the large page mappings will only speed things up until 1008 * the startup process proceeds a bit further, as when 1009 * sfmmu_map_prom_mappings() copies page mappings from the PROM to the 1010 * kernel it remaps everything but the TSBs using 8K pages anyway... 1011 * 1012 * At some point in the future, sfmmu_map_prom_mappings() will be 1013 * rewritten to copy memory mappings to the kernel using the same MMU 1014 * page sizes the PROM used. When that occurs, if the PROM did use 1015 * large MMU pages to map memory, the alignment/sizing work we're 1016 * doing now should give us a nice extra performance boost, albeit at 1017 * the cost of greater RAM usage... 1018 */ 1019 alloc_alignsize = ((npages >= tune_npages) ? MMU_PAGESIZE4M : 1020 MMU_PAGESIZE); 1021 1022 PRM_DEBUG(tune_npages); 1023 PRM_DEBUG(alloc_alignsize); 1024 1025 /* 1026 * Save off where the contiguous allocations to date have ended 1027 * in econtig32. 1028 */ 1029 econtig32 = alloc_base; 1030 PRM_DEBUG(econtig32); 1031 1032 if (econtig32 > (caddr_t)KERNEL_LIMIT32) 1033 cmn_err(CE_PANIC, "econtig32 too big"); 1034 1035 /* 1036 * To avoid memory allocation collisions in the 32-bit virtual address 1037 * space, make allocations from this point forward in 64-bit virtual 1038 * address space starting at syslimit and working up. Also use the 1039 * alignment specified by alloc_alignsize, as we may be able to save 1040 * ourselves TLB misses by using larger page sizes if they're 1041 * available. 1042 * 1043 * All this is needed because on large memory systems, the default 1044 * Solaris allocations will collide with SYSBASE32, which is hard 1045 * coded to be at the virtual address 0x78000000. Therefore, on 64-bit 1046 * kernels, move the allocations to a location in the 64-bit virtual 1047 * address space space, allowing those structures to grow without 1048 * worry. 1049 * 1050 * On current CPUs we'll run out of physical memory address bits before 1051 * we need to worry about the allocations running into anything else in 1052 * VM or the virtual address holes on US-I and II, as there's currently 1053 * about 1 TB of addressable space before the US-I/II VA hole. 1054 */ 1055 kmem64_base = (caddr_t)syslimit; 1056 PRM_DEBUG(kmem64_base); 1057 1058 alloc_base = (caddr_t)roundup((uintptr_t)kmem64_base, alloc_alignsize); 1059 1060 /* 1061 * If KHME and/or UHME hash buckets won't fit in the nucleus, allocate 1062 * them here. 1063 */ 1064 if (khme_hash == NULL || uhme_hash == NULL) { 1065 /* 1066 * alloc_hme_buckets() will align alloc_base properly before 1067 * assigning the hash buckets, so we don't need to do it 1068 * before the call... 1069 */ 1070 alloc_base = alloc_hme_buckets(alloc_base, alloc_alignsize); 1071 1072 PRM_DEBUG(alloc_base); 1073 PRM_DEBUG(khme_hash); 1074 PRM_DEBUG(uhme_hash); 1075 } 1076 1077 /* 1078 * Allocate the remaining page freelists. NUMA systems can 1079 * have lots of page freelists, one per node, which quickly 1080 * outgrow the amount of nucleus memory available. 1081 */ 1082 if (max_mem_nodes > 1) { 1083 int mnode; 1084 caddr_t alloc_start = alloc_base; 1085 1086 for (mnode = 1; mnode < max_mem_nodes; mnode++) { 1087 alloc_base = alloc_page_freelists(mnode, alloc_base, 1088 ecache_alignsize); 1089 } 1090 1091 if (alloc_base > alloc_start) { 1092 alloc_base = (caddr_t)roundup((uintptr_t)alloc_base, 1093 alloc_alignsize); 1094 if ((caddr_t)BOP_ALLOC(bootops, alloc_start, 1095 alloc_base - alloc_start, 1096 alloc_alignsize) != alloc_start) 1097 cmn_err(CE_PANIC, 1098 "Unable to alloc page freelists\n"); 1099 } 1100 1101 PRM_DEBUG(alloc_base); 1102 } 1103 1104 if (!mml_table) { 1105 size_t mmltable_sz; 1106 1107 /* 1108 * We need to allocate the mml_table here because there 1109 * was not enough space within the nucleus. 1110 */ 1111 mmltable_sz = sizeof (kmutex_t) * mml_table_sz; 1112 alloc_sz = roundup(mmltable_sz, alloc_alignsize); 1113 alloc_base = (caddr_t)roundup((uintptr_t)alloc_base, 1114 alloc_alignsize); 1115 1116 if ((mml_table = (kmutex_t *)BOP_ALLOC(bootops, alloc_base, 1117 alloc_sz, alloc_alignsize)) != (kmutex_t *)alloc_base) 1118 panic("mml_table alloc failure"); 1119 1120 alloc_base += alloc_sz; 1121 PRM_DEBUG(mml_table); 1122 PRM_DEBUG(alloc_base); 1123 } 1124 1125 if (kpm_enable && !(kpmp_table || kpmp_stable)) { 1126 size_t kpmptable_sz; 1127 caddr_t table; 1128 1129 /* 1130 * We need to allocate either kpmp_table or kpmp_stable here 1131 * because there was not enough space within the nucleus. 1132 */ 1133 kpmptable_sz = (kpm_smallpages == 0) ? 1134 sizeof (kpm_hlk_t) * kpmp_table_sz : 1135 sizeof (kpm_shlk_t) * kpmp_stable_sz; 1136 1137 alloc_sz = roundup(kpmptable_sz, alloc_alignsize); 1138 alloc_base = (caddr_t)roundup((uintptr_t)alloc_base, 1139 alloc_alignsize); 1140 1141 table = BOP_ALLOC(bootops, alloc_base, alloc_sz, 1142 alloc_alignsize); 1143 1144 if (table != alloc_base) 1145 panic("kpmp_table or kpmp_stable alloc failure"); 1146 1147 if (kpm_smallpages == 0) { 1148 kpmp_table = (kpm_hlk_t *)table; 1149 PRM_DEBUG(kpmp_table); 1150 } else { 1151 kpmp_stable = (kpm_shlk_t *)table; 1152 PRM_DEBUG(kpmp_stable); 1153 } 1154 1155 alloc_base += alloc_sz; 1156 PRM_DEBUG(alloc_base); 1157 } 1158 1159 if (&ecache_init_scrub_flush_area) { 1160 /* 1161 * Pass alloc_base directly, as the routine itself is 1162 * responsible for any special alignment requirements... 1163 */ 1164 alloc_base = ecache_init_scrub_flush_area(alloc_base); 1165 PRM_DEBUG(alloc_base); 1166 } 1167 1168 /* 1169 * Take the most current snapshot we can by calling mem-update. 1170 */ 1171 copy_boot_memlists(&boot_physinstalled, &boot_physinstalled_len, 1172 &boot_physavail, &boot_physavail_len, 1173 &boot_virtavail, &boot_virtavail_len); 1174 1175 /* 1176 * Reset npages and memblocks based on boot_physavail list. 1177 */ 1178 size_physavail(boot_physavail, boot_physavail_len, &npages, &memblocks); 1179 PRM_DEBUG(npages); 1180 1181 /* 1182 * Account for extra memory after e_text. 1183 */ 1184 npages += extra_etpg; 1185 1186 /* 1187 * Calculate the largest free memory chunk in the nucleus data area. 1188 * We need to figure out if page structs can fit in there or not. 1189 * We also make sure enough page structs get created for any physical 1190 * memory we might be returning to the system. 1191 */ 1192 ndata_remain_sz = ndata_maxsize(&ndata); 1193 PRM_DEBUG(ndata_remain_sz); 1194 1195 pp_sz = sizeof (struct page) * npages; 1196 1197 /* 1198 * Here's a nice bit of code based on somewhat recursive logic: 1199 * 1200 * If the page array would fit within the nucleus, we want to 1201 * add npages to cover any extra memory we may be returning back 1202 * to the system. 1203 * 1204 * HOWEVER, the page array is sized by calculating the size of 1205 * (struct page * npages), as are the pagehash table, ctrs and 1206 * memseg_list, so the very act of performing the calculation below may 1207 * in fact make the array large enough that it no longer fits in the 1208 * nucleus, meaning there would now be a much larger area of the 1209 * nucleus free that should really be added to npages, which would 1210 * make the page array that much larger, and so on. 1211 * 1212 * This also ignores the memory possibly used in the nucleus for the 1213 * the page hash, ctrs and memseg list and the fact that whether they 1214 * fit there or not varies with the npages calculation below, but we 1215 * don't even factor them into the equation at this point; perhaps we 1216 * should or perhaps we should just take the approach that the few 1217 * extra pages we could add via this calculation REALLY aren't worth 1218 * the hassle... 1219 */ 1220 if (ndata_remain_sz > pp_sz) { 1221 size_t spare = ndata_spare(&ndata, pp_sz, ecache_alignsize); 1222 1223 npages += mmu_btop(spare); 1224 1225 pp_sz = npages * sizeof (struct page); 1226 1227 pp_base = ndata_alloc(&ndata, pp_sz, ecache_alignsize); 1228 } 1229 1230 /* 1231 * If physmem is patched to be non-zero, use it instead of 1232 * the monitor value unless physmem is larger than the total 1233 * amount of memory on hand. 1234 */ 1235 if (physmem == 0 || physmem > npages) 1236 physmem = npages; 1237 1238 /* 1239 * If pp_base is NULL that means the routines above have determined 1240 * the page array will not fit in the nucleus; we'll have to 1241 * BOP_ALLOC() ourselves some space for them. 1242 */ 1243 if (pp_base == NULL) { 1244 alloc_base = (caddr_t)roundup((uintptr_t)alloc_base, 1245 alloc_alignsize); 1246 1247 alloc_sz = roundup(pp_sz, alloc_alignsize); 1248 1249 if ((pp_base = (struct page *)BOP_ALLOC(bootops, 1250 alloc_base, alloc_sz, alloc_alignsize)) != 1251 (struct page *)alloc_base) 1252 panic("page alloc failure"); 1253 1254 alloc_base += alloc_sz; 1255 } 1256 1257 /* 1258 * The page structure hash table size is a power of 2 1259 * such that the average hash chain length is PAGE_HASHAVELEN. 1260 */ 1261 page_hashsz = npages / PAGE_HASHAVELEN; 1262 page_hashsz = 1 << highbit((ulong_t)page_hashsz); 1263 pagehash_sz = sizeof (struct page *) * page_hashsz; 1264 1265 /* 1266 * We want to TRY to fit the page structure hash table, 1267 * the page size free list counters, the memseg list and 1268 * and the kpm page space in the nucleus if possible. 1269 * 1270 * alloc_sz counts how much memory needs to be allocated by 1271 * BOP_ALLOC(). 1272 */ 1273 page_hash = ndata_alloc(&ndata, pagehash_sz, ecache_alignsize); 1274 1275 alloc_sz = (page_hash == NULL ? pagehash_sz : 0); 1276 1277 /* 1278 * Size up per page size free list counters. 1279 */ 1280 ctrs_sz = page_ctrs_sz(); 1281 ctrs_base = ndata_alloc(&ndata, ctrs_sz, ecache_alignsize); 1282 1283 if (ctrs_base == NULL) 1284 alloc_sz = roundup(alloc_sz, ecache_alignsize) + ctrs_sz; 1285 1286 /* 1287 * The memseg list is for the chunks of physical memory that 1288 * will be managed by the vm system. The number calculated is 1289 * a guess as boot may fragment it more when memory allocations 1290 * are made before kphysm_init(). Currently, there are two 1291 * allocations before then, so we assume each causes fragmen- 1292 * tation, and add a couple more for good measure. 1293 */ 1294 memseg_sz = sizeof (struct memseg) * (memblocks + 4); 1295 memseg_base = ndata_alloc(&ndata, memseg_sz, ecache_alignsize); 1296 1297 if (memseg_base == NULL) 1298 alloc_sz = roundup(alloc_sz, ecache_alignsize) + memseg_sz; 1299 1300 1301 if (kpm_enable) { 1302 /* 1303 * kpm page space -- Update kpm_npages and make the 1304 * same assumption about fragmenting as it is done 1305 * for memseg_sz above. 1306 */ 1307 kpm_npages_setup(memblocks + 4); 1308 kpm_pp_sz = (kpm_smallpages == 0) ? 1309 kpm_npages * sizeof (kpm_page_t): 1310 kpm_npages * sizeof (kpm_spage_t); 1311 1312 kpm_pp_base = (uintptr_t)ndata_alloc(&ndata, kpm_pp_sz, 1313 ecache_alignsize); 1314 1315 if (kpm_pp_base == NULL) 1316 alloc_sz = roundup(alloc_sz, ecache_alignsize) + 1317 kpm_pp_sz; 1318 } 1319 1320 if (alloc_sz > 0) { 1321 uintptr_t bop_base; 1322 1323 /* 1324 * We need extra memory allocated through BOP_ALLOC. 1325 */ 1326 alloc_base = (caddr_t)roundup((uintptr_t)alloc_base, 1327 alloc_alignsize); 1328 1329 alloc_sz = roundup(alloc_sz, alloc_alignsize); 1330 1331 if ((bop_base = (uintptr_t)BOP_ALLOC(bootops, alloc_base, 1332 alloc_sz, alloc_alignsize)) != (uintptr_t)alloc_base) 1333 panic("system page struct alloc failure"); 1334 1335 alloc_base += alloc_sz; 1336 1337 if (page_hash == NULL) { 1338 page_hash = (struct page **)bop_base; 1339 bop_base = roundup(bop_base + pagehash_sz, 1340 ecache_alignsize); 1341 } 1342 1343 if (ctrs_base == NULL) { 1344 ctrs_base = (caddr_t)bop_base; 1345 bop_base = roundup(bop_base + ctrs_sz, 1346 ecache_alignsize); 1347 } 1348 1349 if (memseg_base == NULL) { 1350 memseg_base = (struct memseg *)bop_base; 1351 bop_base = roundup(bop_base + memseg_sz, 1352 ecache_alignsize); 1353 } 1354 1355 if (kpm_enable && kpm_pp_base == NULL) { 1356 kpm_pp_base = (uintptr_t)bop_base; 1357 bop_base = roundup(bop_base + kpm_pp_sz, 1358 ecache_alignsize); 1359 } 1360 1361 ASSERT(bop_base <= (uintptr_t)alloc_base); 1362 } 1363 1364 /* 1365 * Initialize per page size free list counters. 1366 */ 1367 ctrs_end = page_ctrs_alloc(ctrs_base); 1368 ASSERT(ctrs_base + ctrs_sz >= ctrs_end); 1369 1370 PRM_DEBUG(page_hash); 1371 PRM_DEBUG(memseg_base); 1372 PRM_DEBUG(kpm_pp_base); 1373 PRM_DEBUG(kpm_pp_sz); 1374 PRM_DEBUG(pp_base); 1375 PRM_DEBUG(pp_sz); 1376 PRM_DEBUG(alloc_base); 1377 1378 #ifdef TRAPTRACE 1379 /* 1380 * Allocate trap trace buffer last so as not to affect 1381 * the 4M alignments of the allocations above on V9 SPARCs... 1382 */ 1383 alloc_base = trap_trace_alloc(alloc_base); 1384 PRM_DEBUG(alloc_base); 1385 #endif /* TRAPTRACE */ 1386 1387 if (kmem64_base) { 1388 /* 1389 * Set the end of the kmem64 segment for V9 SPARCs, if 1390 * appropriate... 1391 */ 1392 kmem64_end = (caddr_t)roundup((uintptr_t)alloc_base, 1393 alloc_alignsize); 1394 1395 PRM_DEBUG(kmem64_base); 1396 PRM_DEBUG(kmem64_end); 1397 } 1398 1399 /* 1400 * Allocate space for the interrupt vector table and also for the 1401 * reserved interrupt vector data structures. 1402 */ 1403 memspace = (caddr_t)BOP_ALLOC(bootops, (caddr_t)intr_vec_table, 1404 IVSIZE, MMU_PAGESIZE); 1405 if (memspace != (caddr_t)intr_vec_table) 1406 panic("interrupt vector table allocation failure"); 1407 1408 /* 1409 * The memory lists from boot are allocated from the heap arena 1410 * so that later they can be freed and/or reallocated. 1411 */ 1412 if (BOP_GETPROP(bootops, "extent", &memlist_sz) == -1) 1413 panic("could not retrieve property \"extent\""); 1414 1415 /* 1416 * Between now and when we finish copying in the memory lists, 1417 * allocations happen so the space gets fragmented and the 1418 * lists longer. Leave enough space for lists twice as long 1419 * as what boot says it has now; roundup to a pagesize. 1420 * Also add space for the final phys-avail copy in the fixup 1421 * routine. 1422 */ 1423 va = (caddr_t)(sysbase + PAGESIZE + PANICBUFSIZE + 1424 roundup(IVSIZE, MMU_PAGESIZE)); 1425 memlist_sz *= 4; 1426 memlist_sz = roundup(memlist_sz, MMU_PAGESIZE); 1427 memspace = (caddr_t)BOP_ALLOC(bootops, va, memlist_sz, BO_NO_ALIGN); 1428 if (memspace == NULL) 1429 halt("Boot allocation failed."); 1430 1431 memlist = (struct memlist *)memspace; 1432 memlist_end = (char *)memspace + memlist_sz; 1433 1434 PRM_DEBUG(memlist); 1435 PRM_DEBUG(memlist_end); 1436 PRM_DEBUG(sysbase); 1437 PRM_DEBUG(syslimit); 1438 1439 kernelheap_init((void *)sysbase, (void *)syslimit, 1440 (caddr_t)sysbase + PAGESIZE, NULL, NULL); 1441 1442 /* 1443 * Take the most current snapshot we can by calling mem-update. 1444 */ 1445 copy_boot_memlists(&boot_physinstalled, &boot_physinstalled_len, 1446 &boot_physavail, &boot_physavail_len, 1447 &boot_virtavail, &boot_virtavail_len); 1448 1449 /* 1450 * Remove the space used by BOP_ALLOC from the kernel heap 1451 * plus the area actually used by the OBP (if any) 1452 * ignoring virtual addresses in virt_avail, above syslimit. 1453 */ 1454 virt_avail = memlist; 1455 copy_memlist(boot_virtavail, boot_virtavail_len, &memlist); 1456 1457 for (cur = virt_avail; cur->next; cur = cur->next) { 1458 uint64_t range_base, range_size; 1459 1460 if ((range_base = cur->address + cur->size) < (uint64_t)sysbase) 1461 continue; 1462 if (range_base >= (uint64_t)syslimit) 1463 break; 1464 /* 1465 * Limit the range to end at syslimit. 1466 */ 1467 range_size = MIN(cur->next->address, 1468 (uint64_t)syslimit) - range_base; 1469 (void) vmem_xalloc(heap_arena, (size_t)range_size, PAGESIZE, 1470 0, 0, (void *)range_base, (void *)(range_base + range_size), 1471 VM_NOSLEEP | VM_BESTFIT | VM_PANIC); 1472 } 1473 1474 phys_avail = memlist; 1475 (void) copy_physavail(boot_physavail, boot_physavail_len, 1476 &memlist, 0, 0); 1477 1478 /* 1479 * Add any extra memory after e_text to the phys_avail list, as long 1480 * as there's at least a page to add. 1481 */ 1482 if (extra_etpg) 1483 memlist_add(va_to_pa(extra_etva), mmu_ptob(extra_etpg), 1484 &memlist, &phys_avail); 1485 1486 /* 1487 * Add any extra memory after e_data to the phys_avail list as long 1488 * as there's at least a page to add. Usually, there isn't any, 1489 * since extra HME blocks typically get allocated there first before 1490 * using RAM elsewhere. 1491 */ 1492 if ((nalloc_base = ndata_extra_base(&ndata, MMU_PAGESIZE)) == NULL) 1493 nalloc_base = nalloc_end; 1494 ndata_remain_sz = nalloc_end - nalloc_base; 1495 1496 if (ndata_remain_sz >= MMU_PAGESIZE) 1497 memlist_add(va_to_pa(nalloc_base), 1498 (uint64_t)ndata_remain_sz, &memlist, &phys_avail); 1499 1500 PRM_DEBUG(memlist); 1501 PRM_DEBUG(memlist_sz); 1502 PRM_DEBUG(memspace); 1503 1504 if ((caddr_t)memlist > (memspace + memlist_sz)) 1505 panic("memlist overflow"); 1506 1507 PRM_DEBUG(pp_base); 1508 PRM_DEBUG(memseg_base); 1509 PRM_DEBUG(npages); 1510 1511 /* 1512 * Initialize the page structures from the memory lists. 1513 */ 1514 kphysm_init(pp_base, memseg_base, npages, kpm_pp_base, kpm_npages); 1515 1516 availrmem_initial = availrmem = freemem; 1517 PRM_DEBUG(availrmem); 1518 1519 /* 1520 * Some of the locks depend on page_hashsz being set! 1521 * kmem_init() depends on this; so, keep it here. 1522 */ 1523 page_lock_init(); 1524 1525 /* 1526 * Initialize kernel memory allocator. 1527 */ 1528 kmem_init(); 1529 1530 /* 1531 * Factor in colorequiv to check additional 'equivalent' bins 1532 */ 1533 page_set_colorequiv_arr(); 1534 1535 /* 1536 * Initialize bp_mapin(). 1537 */ 1538 bp_init(shm_alignment, HAT_STRICTORDER); 1539 1540 /* 1541 * Reserve space for panicbuf, intr_vec_table and reserved interrupt 1542 * vector data structures from the 32-bit heap. 1543 */ 1544 (void) vmem_xalloc(heap32_arena, PANICBUFSIZE, PAGESIZE, 0, 0, 1545 panicbuf, panicbuf + PANICBUFSIZE, 1546 VM_NOSLEEP | VM_BESTFIT | VM_PANIC); 1547 1548 (void) vmem_xalloc(heap32_arena, IVSIZE, PAGESIZE, 0, 0, 1549 intr_vec_table, (caddr_t)intr_vec_table + IVSIZE, 1550 VM_NOSLEEP | VM_BESTFIT | VM_PANIC); 1551 1552 mem_config_init(); 1553 } 1554 1555 static void 1556 startup_modules(void) 1557 { 1558 int proplen, nhblk1, nhblk8; 1559 size_t nhblksz; 1560 pgcnt_t hblk_pages, pages_per_hblk; 1561 size_t hme8blk_sz, hme1blk_sz; 1562 1563 /* 1564 * Log any optional messages from the boot program 1565 */ 1566 proplen = (size_t)BOP_GETPROPLEN(bootops, "boot-message"); 1567 if (proplen > 0) { 1568 char *msg; 1569 size_t len = (size_t)proplen; 1570 1571 msg = kmem_zalloc(len, KM_SLEEP); 1572 (void) BOP_GETPROP(bootops, "boot-message", msg); 1573 cmn_err(CE_CONT, "?%s\n", msg); 1574 kmem_free(msg, len); 1575 } 1576 1577 /* 1578 * Let the platforms have a chance to change default 1579 * values before reading system file. 1580 */ 1581 if (&set_platform_defaults) 1582 set_platform_defaults(); 1583 1584 /* 1585 * Calculate default settings of system parameters based upon 1586 * maxusers, yet allow to be overridden via the /etc/system file. 1587 */ 1588 param_calc(0); 1589 1590 mod_setup(); 1591 1592 /* 1593 * If this is a positron, complain and halt. 1594 */ 1595 if (&iam_positron && iam_positron()) { 1596 cmn_err(CE_WARN, "This hardware platform is not supported" 1597 " by this release of Solaris.\n"); 1598 #ifdef DEBUG 1599 prom_enter_mon(); /* Type 'go' to resume */ 1600 cmn_err(CE_WARN, "Booting an unsupported platform.\n"); 1601 cmn_err(CE_WARN, "Booting with down-rev firmware.\n"); 1602 1603 #else /* DEBUG */ 1604 halt(0); 1605 #endif /* DEBUG */ 1606 } 1607 1608 /* 1609 * If we are running firmware that isn't 64-bit ready 1610 * then complain and halt. 1611 */ 1612 do_prom_version_check(); 1613 1614 /* 1615 * Initialize system parameters 1616 */ 1617 param_init(); 1618 1619 /* 1620 * maxmem is the amount of physical memory we're playing with. 1621 */ 1622 maxmem = physmem; 1623 1624 /* Set segkp limits. */ 1625 ncbase = kdi_segdebugbase; 1626 ncend = kdi_segdebugbase; 1627 1628 /* 1629 * Initialize the hat layer. 1630 */ 1631 hat_init(); 1632 1633 /* 1634 * Initialize segment management stuff. 1635 */ 1636 seg_init(); 1637 1638 /* 1639 * Create the va>tte handler, so the prom can understand 1640 * kernel translations. The handler is installed later, just 1641 * as we are about to take over the trap table from the prom. 1642 */ 1643 create_va_to_tte(); 1644 1645 /* 1646 * Load the forthdebugger (optional) 1647 */ 1648 forthdebug_init(); 1649 1650 /* 1651 * Create OBP node for console input callbacks 1652 * if it is needed. 1653 */ 1654 startup_create_io_node(); 1655 1656 if (modloadonly("fs", "specfs") == -1) 1657 halt("Can't load specfs"); 1658 1659 if (modloadonly("fs", "devfs") == -1) 1660 halt("Can't load devfs"); 1661 1662 if (modloadonly("misc", "swapgeneric") == -1) 1663 halt("Can't load swapgeneric"); 1664 1665 (void) modloadonly("sys", "lbl_edition"); 1666 1667 dispinit(); 1668 1669 /* 1670 * Infer meanings to the members of the idprom buffer. 1671 */ 1672 parse_idprom(); 1673 1674 /* Read cluster configuration data. */ 1675 clconf_init(); 1676 1677 setup_ddi(); 1678 1679 /* 1680 * Lets take this opportunity to load the root device. 1681 */ 1682 if (loadrootmodules() != 0) 1683 debug_enter("Can't load the root filesystem"); 1684 1685 /* 1686 * Load tod driver module for the tod part found on this system. 1687 * Recompute the cpu frequency/delays based on tod as tod part 1688 * tends to keep time more accurately. 1689 */ 1690 if (&load_tod_module) 1691 load_tod_module(); 1692 1693 /* 1694 * Allow platforms to load modules which might 1695 * be needed after bootops are gone. 1696 */ 1697 if (&load_platform_modules) 1698 load_platform_modules(); 1699 1700 setcpudelay(); 1701 1702 copy_boot_memlists(&boot_physinstalled, &boot_physinstalled_len, 1703 &boot_physavail, &boot_physavail_len, 1704 &boot_virtavail, &boot_virtavail_len); 1705 1706 bop_alloc_pages = size_virtalloc(boot_virtavail, boot_virtavail_len); 1707 1708 /* 1709 * Calculation and allocation of hmeblks needed to remap 1710 * the memory allocated by PROM till now: 1711 * 1712 * (1) calculate how much virtual memory has been bop_alloc'ed. 1713 * (2) roundup this memory to span of hme8blk, i.e. 64KB 1714 * (3) calculate number of hme8blk's needed to remap this memory 1715 * (4) calculate amount of memory that's consumed by these hme8blk's 1716 * (5) add memory calculated in steps (2) and (4) above. 1717 * (6) roundup this memory to span of hme8blk, i.e. 64KB 1718 * (7) calculate number of hme8blk's needed to remap this memory 1719 * (8) calculate amount of memory that's consumed by these hme8blk's 1720 * (9) allocate additional hme1blk's to hold large mappings. 1721 * H8TOH1 determines this. The current SWAG gives enough hblk1's 1722 * to remap everything with 4M mappings. 1723 * (10) account for partially used hblk8's due to non-64K aligned 1724 * PROM mapping entries. 1725 * (11) add memory calculated in steps (8), (9), and (10) above. 1726 * (12) kmem_zalloc the memory calculated in (11); since segkmem 1727 * is not ready yet, this gets bop_alloc'ed. 1728 * (13) there will be very few bop_alloc's after this point before 1729 * trap table takes over 1730 */ 1731 1732 /* sfmmu_init_nucleus_hblks expects properly aligned data structures. */ 1733 hme8blk_sz = roundup(HME8BLK_SZ, sizeof (int64_t)); 1734 hme1blk_sz = roundup(HME1BLK_SZ, sizeof (int64_t)); 1735 1736 pages_per_hblk = btop(HMEBLK_SPAN(TTE8K)); 1737 bop_alloc_pages = roundup(bop_alloc_pages, pages_per_hblk); 1738 nhblk8 = bop_alloc_pages / pages_per_hblk; 1739 nhblk1 = roundup(nhblk8, H8TOH1) / H8TOH1; 1740 hblk_pages = btopr(nhblk8 * hme8blk_sz + nhblk1 * hme1blk_sz); 1741 bop_alloc_pages += hblk_pages; 1742 bop_alloc_pages = roundup(bop_alloc_pages, pages_per_hblk); 1743 nhblk8 = bop_alloc_pages / pages_per_hblk; 1744 nhblk1 = roundup(nhblk8, H8TOH1) / H8TOH1; 1745 if (nhblk1 < hblk1_min) 1746 nhblk1 = hblk1_min; 1747 if (nhblk8 < hblk8_min) 1748 nhblk8 = hblk8_min; 1749 1750 /* 1751 * Since hblk8's can hold up to 64k of mappings aligned on a 64k 1752 * boundary, the number of hblk8's needed to map the entries in the 1753 * boot_virtavail list needs to be adjusted to take this into 1754 * consideration. Thus, we need to add additional hblk8's since it 1755 * is possible that an hblk8 will not have all 8 slots used due to 1756 * alignment constraints. Since there were boot_virtavail_len entries 1757 * in that list, we need to add that many hblk8's to the number 1758 * already calculated to make sure we don't underestimate. 1759 */ 1760 nhblk8 += boot_virtavail_len; 1761 nhblksz = nhblk8 * hme8blk_sz + nhblk1 * hme1blk_sz; 1762 1763 /* Allocate in pagesize chunks */ 1764 nhblksz = roundup(nhblksz, MMU_PAGESIZE); 1765 hblk_base = kmem_zalloc(nhblksz, KM_SLEEP); 1766 sfmmu_init_nucleus_hblks(hblk_base, nhblksz, nhblk8, nhblk1); 1767 } 1768 1769 static void 1770 startup_bop_gone(void) 1771 { 1772 extern int bop_io_quiesced; 1773 1774 /* 1775 * Destroy the MD initialized at startup 1776 * The startup initializes the MD framework 1777 * using prom and BOP alloc free it now. 1778 */ 1779 mach_descrip_startup_fini(); 1780 1781 /* 1782 * Call back into boot and release boots resources. 1783 */ 1784 BOP_QUIESCE_IO(bootops); 1785 bop_io_quiesced = 1; 1786 1787 copy_boot_memlists(&boot_physinstalled, &boot_physinstalled_len, 1788 &boot_physavail, &boot_physavail_len, 1789 &boot_virtavail, &boot_virtavail_len); 1790 /* 1791 * Copy physinstalled list into kernel space. 1792 */ 1793 phys_install = memlist; 1794 copy_memlist(boot_physinstalled, boot_physinstalled_len, &memlist); 1795 1796 /* 1797 * setup physically contiguous area twice as large as the ecache. 1798 * this is used while doing displacement flush of ecaches 1799 */ 1800 if (&ecache_flush_address) { 1801 ecache_flushaddr = ecache_flush_address(); 1802 if (ecache_flushaddr == (uint64_t)-1) { 1803 cmn_err(CE_PANIC, 1804 "startup: no memory to set ecache_flushaddr"); 1805 } 1806 } 1807 1808 /* 1809 * Virtual available next. 1810 */ 1811 ASSERT(virt_avail != NULL); 1812 memlist_free_list(virt_avail); 1813 virt_avail = memlist; 1814 copy_memlist(boot_virtavail, boot_virtavail_len, &memlist); 1815 1816 /* 1817 * Last chance to ask our booter questions .. 1818 */ 1819 } 1820 1821 1822 /* 1823 * startup_fixup_physavail - called from mach_sfmmu.c after the final 1824 * allocations have been performed. We can't call it in startup_bop_gone 1825 * since later operations can cause obp to allocate more memory. 1826 */ 1827 void 1828 startup_fixup_physavail(void) 1829 { 1830 struct memlist *cur; 1831 1832 /* 1833 * take the most current snapshot we can by calling mem-update 1834 */ 1835 copy_boot_memlists(&boot_physinstalled, &boot_physinstalled_len, 1836 &boot_physavail, &boot_physavail_len, 1837 &boot_virtavail, &boot_virtavail_len); 1838 1839 /* 1840 * Copy phys_avail list, again. 1841 * Both the kernel/boot and the prom have been allocating 1842 * from the original list we copied earlier. 1843 */ 1844 cur = memlist; 1845 (void) copy_physavail(boot_physavail, boot_physavail_len, 1846 &memlist, 0, 0); 1847 1848 /* 1849 * Add any extra memory after e_text we added to the phys_avail list 1850 * back to the old list. 1851 */ 1852 if (extra_etpg) 1853 memlist_add(va_to_pa(extra_etva), mmu_ptob(extra_etpg), 1854 &memlist, &cur); 1855 if (ndata_remain_sz >= MMU_PAGESIZE) 1856 memlist_add(va_to_pa(nalloc_base), 1857 (uint64_t)ndata_remain_sz, &memlist, &cur); 1858 1859 /* 1860 * There isn't any bounds checking on the memlist area 1861 * so ensure it hasn't overgrown. 1862 */ 1863 if ((caddr_t)memlist > (caddr_t)memlist_end) 1864 cmn_err(CE_PANIC, "startup: memlist size exceeded"); 1865 1866 /* 1867 * The kernel removes the pages that were allocated for it from 1868 * the freelist, but we now have to find any -extra- pages that 1869 * the prom has allocated for it's own book-keeping, and remove 1870 * them from the freelist too. sigh. 1871 */ 1872 fix_prom_pages(phys_avail, cur); 1873 1874 ASSERT(phys_avail != NULL); 1875 memlist_free_list(phys_avail); 1876 phys_avail = cur; 1877 1878 /* 1879 * We're done with boot. Just after this point in time, boot 1880 * gets unmapped, so we can no longer rely on its services. 1881 * Zero the bootops to indicate this fact. 1882 */ 1883 bootops = (struct bootops *)NULL; 1884 BOOTOPS_GONE(); 1885 } 1886 1887 static void 1888 startup_vm(void) 1889 { 1890 size_t i; 1891 struct segmap_crargs a; 1892 struct segkpm_crargs b; 1893 1894 uint64_t avmem; 1895 caddr_t va; 1896 pgcnt_t max_phys_segkp; 1897 int mnode; 1898 1899 extern int use_brk_lpg, use_stk_lpg; 1900 1901 /* 1902 * get prom's mappings, create hments for them and switch 1903 * to the kernel context. 1904 */ 1905 hat_kern_setup(); 1906 1907 /* 1908 * Take over trap table 1909 */ 1910 setup_trap_table(); 1911 1912 /* 1913 * Install the va>tte handler, so that the prom can handle 1914 * misses and understand the kernel table layout in case 1915 * we need call into the prom. 1916 */ 1917 install_va_to_tte(); 1918 1919 /* 1920 * Set a flag to indicate that the tba has been taken over. 1921 */ 1922 tba_taken_over = 1; 1923 1924 /* initialize MMU primary context register */ 1925 mmu_init_kcontext(); 1926 1927 /* 1928 * The boot cpu can now take interrupts, x-calls, x-traps 1929 */ 1930 CPUSET_ADD(cpu_ready_set, CPU->cpu_id); 1931 CPU->cpu_flags |= (CPU_READY | CPU_ENABLE | CPU_EXISTS); 1932 1933 /* 1934 * Set a flag to tell write_scb_int() that it can access V_TBR_WR_ADDR. 1935 */ 1936 tbr_wr_addr_inited = 1; 1937 1938 /* 1939 * Initialize VM system, and map kernel address space. 1940 */ 1941 kvm_init(); 1942 1943 /* 1944 * XXX4U: previously, we initialized and turned on 1945 * the caches at this point. But of course we have 1946 * nothing to do, as the prom has already done this 1947 * for us -- main memory must be E$able at all times. 1948 */ 1949 1950 /* 1951 * If the following is true, someone has patched 1952 * phsymem to be less than the number of pages that 1953 * the system actually has. Remove pages until system 1954 * memory is limited to the requested amount. Since we 1955 * have allocated page structures for all pages, we 1956 * correct the amount of memory we want to remove 1957 * by the size of the memory used to hold page structures 1958 * for the non-used pages. 1959 */ 1960 if (physmem < npages) { 1961 pgcnt_t diff, off; 1962 struct page *pp; 1963 struct seg kseg; 1964 1965 cmn_err(CE_WARN, "limiting physmem to %ld pages", physmem); 1966 1967 off = 0; 1968 diff = npages - physmem; 1969 diff -= mmu_btopr(diff * sizeof (struct page)); 1970 kseg.s_as = &kas; 1971 while (diff--) { 1972 pp = page_create_va(&unused_pages_vp, (offset_t)off, 1973 MMU_PAGESIZE, PG_WAIT | PG_EXCL, 1974 &kseg, (caddr_t)off); 1975 if (pp == NULL) 1976 cmn_err(CE_PANIC, "limited physmem too much!"); 1977 page_io_unlock(pp); 1978 page_downgrade(pp); 1979 availrmem--; 1980 off += MMU_PAGESIZE; 1981 } 1982 } 1983 1984 /* 1985 * When printing memory, show the total as physmem less 1986 * that stolen by a debugger. 1987 */ 1988 cmn_err(CE_CONT, "?mem = %ldK (0x%lx000)\n", 1989 (ulong_t)(physinstalled) << (PAGESHIFT - 10), 1990 (ulong_t)(physinstalled) << (PAGESHIFT - 12)); 1991 1992 avmem = (uint64_t)freemem << PAGESHIFT; 1993 cmn_err(CE_CONT, "?avail mem = %lld\n", (unsigned long long)avmem); 1994 1995 /* 1996 * For small memory systems disable automatic large pages. 1997 */ 1998 if (physmem < privm_lpg_min_physmem) { 1999 use_brk_lpg = 0; 2000 use_stk_lpg = 0; 2001 } 2002 2003 /* 2004 * Perform platform specific freelist processing 2005 */ 2006 if (&plat_freelist_process) { 2007 for (mnode = 0; mnode < max_mem_nodes; mnode++) 2008 if (mem_node_config[mnode].exists) 2009 plat_freelist_process(mnode); 2010 } 2011 2012 /* 2013 * Initialize the segkp segment type. We position it 2014 * after the configured tables and buffers (whose end 2015 * is given by econtig) and before V_WKBASE_ADDR. 2016 * Also in this area is segkmap (size SEGMAPSIZE). 2017 */ 2018 2019 /* XXX - cache alignment? */ 2020 va = (caddr_t)SEGKPBASE; 2021 ASSERT(((uintptr_t)va & PAGEOFFSET) == 0); 2022 2023 max_phys_segkp = (physmem * 2); 2024 2025 if (segkpsize < btop(SEGKPMINSIZE) || segkpsize > btop(SEGKPMAXSIZE)) { 2026 segkpsize = btop(SEGKPDEFSIZE); 2027 cmn_err(CE_WARN, "Illegal value for segkpsize. " 2028 "segkpsize has been reset to %ld pages", segkpsize); 2029 } 2030 2031 i = ptob(MIN(segkpsize, max_phys_segkp)); 2032 2033 rw_enter(&kas.a_lock, RW_WRITER); 2034 if (seg_attach(&kas, va, i, segkp) < 0) 2035 cmn_err(CE_PANIC, "startup: cannot attach segkp"); 2036 if (segkp_create(segkp) != 0) 2037 cmn_err(CE_PANIC, "startup: segkp_create failed"); 2038 rw_exit(&kas.a_lock); 2039 2040 /* 2041 * kpm segment 2042 */ 2043 segmap_kpm = kpm_enable && 2044 segmap_kpm && PAGESIZE == MAXBSIZE; 2045 2046 if (kpm_enable) { 2047 rw_enter(&kas.a_lock, RW_WRITER); 2048 2049 /* 2050 * The segkpm virtual range range is larger than the 2051 * actual physical memory size and also covers gaps in 2052 * the physical address range for the following reasons: 2053 * . keep conversion between segkpm and physical addresses 2054 * simple, cheap and unambiguous. 2055 * . avoid extension/shrink of the the segkpm in case of DR. 2056 * . avoid complexity for handling of virtual addressed 2057 * caches, segkpm and the regular mapping scheme must be 2058 * kept in sync wrt. the virtual color of mapped pages. 2059 * Any accesses to virtual segkpm ranges not backed by 2060 * physical memory will fall through the memseg pfn hash 2061 * and will be handled in segkpm_fault. 2062 * Additional kpm_size spaces needed for vac alias prevention. 2063 */ 2064 if (seg_attach(&kas, kpm_vbase, kpm_size * vac_colors, 2065 segkpm) < 0) 2066 cmn_err(CE_PANIC, "cannot attach segkpm"); 2067 2068 b.prot = PROT_READ | PROT_WRITE; 2069 b.nvcolors = shm_alignment >> MMU_PAGESHIFT; 2070 2071 if (segkpm_create(segkpm, (caddr_t)&b) != 0) 2072 panic("segkpm_create segkpm"); 2073 2074 rw_exit(&kas.a_lock); 2075 2076 mach_kpm_init(); 2077 } 2078 2079 if (!segzio_fromheap) { 2080 size_t size; 2081 size_t physmem_b = mmu_ptob(physmem); 2082 2083 /* size is in bytes, segziosize is in pages */ 2084 if (segziosize == 0) { 2085 size = physmem_b; 2086 } else { 2087 size = mmu_ptob(segziosize); 2088 } 2089 2090 if (size < SEGZIOMINSIZE) { 2091 size = SEGZIOMINSIZE; 2092 } else if (size > SEGZIOMAXSIZE) { 2093 size = SEGZIOMAXSIZE; 2094 /* 2095 * On 64-bit x86, we only have 2TB of KVA. This exists 2096 * for parity with x86. 2097 * 2098 * SEGZIOMAXSIZE is capped at 512gb so that segzio 2099 * doesn't consume all of KVA. However, if we have a 2100 * system that has more thant 512gb of physical memory, 2101 * we can actually consume about half of the difference 2102 * between 512gb and the rest of the available physical 2103 * memory. 2104 */ 2105 if (physmem_b > SEGZIOMAXSIZE) { 2106 size += (physmem_b - SEGZIOMAXSIZE) / 2; 2107 } 2108 } 2109 segziosize = mmu_btop(roundup(size, MMU_PAGESIZE)); 2110 /* put the base of the ZIO segment after the kpm segment */ 2111 segzio_base = kpm_vbase + (kpm_size * vac_colors); 2112 PRM_DEBUG(segziosize); 2113 PRM_DEBUG(segzio_base); 2114 2115 /* 2116 * On some platforms, kvm_init is called after the kpm 2117 * sizes have been determined. On SPARC, kvm_init is called 2118 * before, so we have to attach the kzioseg after kvm is 2119 * initialized, otherwise we'll try to allocate from the boot 2120 * area since the kernel heap hasn't yet been configured. 2121 */ 2122 rw_enter(&kas.a_lock, RW_WRITER); 2123 2124 (void) seg_attach(&kas, segzio_base, mmu_ptob(segziosize), 2125 &kzioseg); 2126 (void) segkmem_zio_create(&kzioseg); 2127 2128 /* create zio area covering new segment */ 2129 segkmem_zio_init(segzio_base, mmu_ptob(segziosize)); 2130 2131 rw_exit(&kas.a_lock); 2132 } 2133 2134 2135 /* 2136 * Now create generic mapping segment. This mapping 2137 * goes SEGMAPSIZE beyond SEGMAPBASE. But if the total 2138 * virtual address is greater than the amount of free 2139 * memory that is available, then we trim back the 2140 * segment size to that amount 2141 */ 2142 va = (caddr_t)SEGMAPBASE; 2143 2144 /* 2145 * 1201049: segkmap base address must be MAXBSIZE aligned 2146 */ 2147 ASSERT(((uintptr_t)va & MAXBOFFSET) == 0); 2148 2149 /* 2150 * Set size of segmap to percentage of freemem at boot, 2151 * but stay within the allowable range 2152 * Note we take percentage before converting from pages 2153 * to bytes to avoid an overflow on 32-bit kernels. 2154 */ 2155 i = mmu_ptob((freemem * segmap_percent) / 100); 2156 2157 if (i < MINMAPSIZE) 2158 i = MINMAPSIZE; 2159 2160 if (i > MIN(SEGMAPSIZE, mmu_ptob(freemem))) 2161 i = MIN(SEGMAPSIZE, mmu_ptob(freemem)); 2162 2163 i &= MAXBMASK; /* 1201049: segkmap size must be MAXBSIZE aligned */ 2164 2165 rw_enter(&kas.a_lock, RW_WRITER); 2166 if (seg_attach(&kas, va, i, segkmap) < 0) 2167 cmn_err(CE_PANIC, "cannot attach segkmap"); 2168 2169 a.prot = PROT_READ | PROT_WRITE; 2170 a.shmsize = shm_alignment; 2171 a.nfreelist = 0; /* use segmap driver defaults */ 2172 2173 if (segmap_create(segkmap, (caddr_t)&a) != 0) 2174 panic("segmap_create segkmap"); 2175 rw_exit(&kas.a_lock); 2176 2177 segdev_init(); 2178 } 2179 2180 static void 2181 startup_end(void) 2182 { 2183 if ((caddr_t)memlist > (caddr_t)memlist_end) 2184 panic("memlist overflow 2"); 2185 memlist_free_block((caddr_t)memlist, 2186 ((caddr_t)memlist_end - (caddr_t)memlist)); 2187 memlist = NULL; 2188 2189 /* enable page_relocation since OBP is now done */ 2190 page_relocate_ready = 1; 2191 2192 /* 2193 * Perform tasks that get done after most of the VM 2194 * initialization has been done but before the clock 2195 * and other devices get started. 2196 */ 2197 kern_setup1(); 2198 2199 /* 2200 * Intialize the VM arenas for allocating physically 2201 * contiguus memory chunk for interrupt queues snd 2202 * allocate/register boot cpu's queues, if any and 2203 * allocate dump buffer for sun4v systems to store 2204 * extra crash information during crash dump 2205 */ 2206 contig_mem_init(); 2207 mach_descrip_init(); 2208 cpu_intrq_setup(CPU); 2209 cpu_intrq_register(CPU); 2210 mach_htraptrace_setup(CPU->cpu_id); 2211 mach_htraptrace_configure(CPU->cpu_id); 2212 mach_dump_buffer_init(); 2213 2214 /* 2215 * Initialize interrupt related stuff 2216 */ 2217 cpu_intr_alloc(CPU, NINTR_THREADS); 2218 2219 (void) splzs(); /* allow hi clock ints but not zs */ 2220 2221 /* 2222 * Initialize errors. 2223 */ 2224 error_init(); 2225 2226 /* 2227 * Note that we may have already used kernel bcopy before this 2228 * point - but if you really care about this, adb the use_hw_* 2229 * variables to 0 before rebooting. 2230 */ 2231 mach_hw_copy_limit(); 2232 2233 /* 2234 * Install the "real" preemption guards before DDI services 2235 * are available. 2236 */ 2237 (void) prom_set_preprom(kern_preprom); 2238 (void) prom_set_postprom(kern_postprom); 2239 CPU->cpu_m.mutex_ready = 1; 2240 2241 /* 2242 * Initialize segnf (kernel support for non-faulting loads). 2243 */ 2244 segnf_init(); 2245 2246 /* 2247 * Configure the root devinfo node. 2248 */ 2249 configure(); /* set up devices */ 2250 mach_cpu_halt_idle(); 2251 } 2252 2253 2254 void 2255 post_startup(void) 2256 { 2257 #ifdef PTL1_PANIC_DEBUG 2258 extern void init_ptl1_thread(void); 2259 #endif /* PTL1_PANIC_DEBUG */ 2260 extern void abort_sequence_init(void); 2261 2262 /* 2263 * Set the system wide, processor-specific flags to be passed 2264 * to userland via the aux vector for performance hints and 2265 * instruction set extensions. 2266 */ 2267 bind_hwcap(); 2268 2269 /* 2270 * Startup memory scrubber (if any) 2271 */ 2272 mach_memscrub(); 2273 2274 /* 2275 * Allocate soft interrupt to handle abort sequence. 2276 */ 2277 abort_sequence_init(); 2278 2279 /* 2280 * Configure the rest of the system. 2281 * Perform forceloading tasks for /etc/system. 2282 */ 2283 (void) mod_sysctl(SYS_FORCELOAD, NULL); 2284 /* 2285 * ON4.0: Force /proc module in until clock interrupt handle fixed 2286 * ON4.0: This must be fixed or restated in /etc/systems. 2287 */ 2288 (void) modload("fs", "procfs"); 2289 2290 /* load machine class specific drivers */ 2291 load_mach_drivers(); 2292 2293 /* load platform specific drivers */ 2294 if (&load_platform_drivers) 2295 load_platform_drivers(); 2296 2297 /* load vis simulation module, if we are running w/fpu off */ 2298 if (!fpu_exists) { 2299 if (modload("misc", "vis") == -1) 2300 halt("Can't load vis"); 2301 } 2302 2303 mach_fpras(); 2304 2305 maxmem = freemem; 2306 2307 #ifdef PTL1_PANIC_DEBUG 2308 init_ptl1_thread(); 2309 #endif /* PTL1_PANIC_DEBUG */ 2310 2311 if (&cif_init) 2312 cif_init(); 2313 } 2314 2315 #ifdef PTL1_PANIC_DEBUG 2316 int ptl1_panic_test = 0; 2317 int ptl1_panic_xc_one_test = 0; 2318 int ptl1_panic_xc_all_test = 0; 2319 int ptl1_panic_xt_one_test = 0; 2320 int ptl1_panic_xt_all_test = 0; 2321 kthread_id_t ptl1_thread_p = NULL; 2322 kcondvar_t ptl1_cv; 2323 kmutex_t ptl1_mutex; 2324 int ptl1_recurse_count_threshold = 0x40; 2325 int ptl1_recurse_trap_threshold = 0x3d; 2326 extern void ptl1_recurse(int, int); 2327 extern void ptl1_panic_xt(int, int); 2328 2329 /* 2330 * Called once per second by timeout() to wake up 2331 * the ptl1_panic thread to see if it should cause 2332 * a trap to the ptl1_panic() code. 2333 */ 2334 /* ARGSUSED */ 2335 static void 2336 ptl1_wakeup(void *arg) 2337 { 2338 mutex_enter(&ptl1_mutex); 2339 cv_signal(&ptl1_cv); 2340 mutex_exit(&ptl1_mutex); 2341 } 2342 2343 /* 2344 * ptl1_panic cross call function: 2345 * Needed because xc_one() and xc_some() can pass 2346 * 64 bit args but ptl1_recurse() expects ints. 2347 */ 2348 static void 2349 ptl1_panic_xc(void) 2350 { 2351 ptl1_recurse(ptl1_recurse_count_threshold, 2352 ptl1_recurse_trap_threshold); 2353 } 2354 2355 /* 2356 * The ptl1 thread waits for a global flag to be set 2357 * and uses the recurse thresholds to set the stack depth 2358 * to cause a ptl1_panic() directly via a call to ptl1_recurse 2359 * or indirectly via the cross call and cross trap functions. 2360 * 2361 * This is useful testing stack overflows and normal 2362 * ptl1_panic() states with a know stack frame. 2363 * 2364 * ptl1_recurse() is an asm function in ptl1_panic.s that 2365 * sets the {In, Local, Out, and Global} registers to a 2366 * know state on the stack and just prior to causing a 2367 * test ptl1_panic trap. 2368 */ 2369 static void 2370 ptl1_thread(void) 2371 { 2372 mutex_enter(&ptl1_mutex); 2373 while (ptl1_thread_p) { 2374 cpuset_t other_cpus; 2375 int cpu_id; 2376 int my_cpu_id; 2377 int target_cpu_id; 2378 int target_found; 2379 2380 if (ptl1_panic_test) { 2381 ptl1_recurse(ptl1_recurse_count_threshold, 2382 ptl1_recurse_trap_threshold); 2383 } 2384 2385 /* 2386 * Find potential targets for x-call and x-trap, 2387 * if any exist while preempt is disabled we 2388 * start a ptl1_panic if requested via a 2389 * globals. 2390 */ 2391 kpreempt_disable(); 2392 my_cpu_id = CPU->cpu_id; 2393 other_cpus = cpu_ready_set; 2394 CPUSET_DEL(other_cpus, CPU->cpu_id); 2395 target_found = 0; 2396 if (!CPUSET_ISNULL(other_cpus)) { 2397 /* 2398 * Pick the first one 2399 */ 2400 for (cpu_id = 0; cpu_id < NCPU; cpu_id++) { 2401 if (cpu_id == my_cpu_id) 2402 continue; 2403 2404 if (CPU_XCALL_READY(cpu_id)) { 2405 target_cpu_id = cpu_id; 2406 target_found = 1; 2407 break; 2408 } 2409 } 2410 ASSERT(target_found); 2411 2412 if (ptl1_panic_xc_one_test) { 2413 xc_one(target_cpu_id, 2414 (xcfunc_t *)ptl1_panic_xc, 0, 0); 2415 } 2416 if (ptl1_panic_xc_all_test) { 2417 xc_some(other_cpus, 2418 (xcfunc_t *)ptl1_panic_xc, 0, 0); 2419 } 2420 if (ptl1_panic_xt_one_test) { 2421 xt_one(target_cpu_id, 2422 (xcfunc_t *)ptl1_panic_xt, 0, 0); 2423 } 2424 if (ptl1_panic_xt_all_test) { 2425 xt_some(other_cpus, 2426 (xcfunc_t *)ptl1_panic_xt, 0, 0); 2427 } 2428 } 2429 kpreempt_enable(); 2430 (void) timeout(ptl1_wakeup, NULL, hz); 2431 (void) cv_wait(&ptl1_cv, &ptl1_mutex); 2432 } 2433 mutex_exit(&ptl1_mutex); 2434 } 2435 2436 /* 2437 * Called during early startup to create the ptl1_thread 2438 */ 2439 void 2440 init_ptl1_thread(void) 2441 { 2442 ptl1_thread_p = thread_create(NULL, 0, ptl1_thread, NULL, 0, 2443 &p0, TS_RUN, 0); 2444 } 2445 #endif /* PTL1_PANIC_DEBUG */ 2446 2447 2448 /* 2449 * Add to a memory list. 2450 * start = start of new memory segment 2451 * len = length of new memory segment in bytes 2452 * memlistp = pointer to array of available memory segment structures 2453 * curmemlistp = memory list to which to add segment. 2454 */ 2455 static void 2456 memlist_add(uint64_t start, uint64_t len, struct memlist **memlistp, 2457 struct memlist **curmemlistp) 2458 { 2459 struct memlist *new; 2460 2461 new = *memlistp; 2462 new->address = start; 2463 new->size = len; 2464 *memlistp = new + 1; 2465 2466 memlist_insert(new, curmemlistp); 2467 } 2468 2469 /* 2470 * In the case of architectures that support dynamic addition of 2471 * memory at run-time there are two cases where memsegs need to 2472 * be initialized and added to the memseg list. 2473 * 1) memsegs that are constructed at startup. 2474 * 2) memsegs that are constructed at run-time on 2475 * hot-plug capable architectures. 2476 * This code was originally part of the function kphysm_init(). 2477 */ 2478 2479 static void 2480 memseg_list_add(struct memseg *memsegp) 2481 { 2482 struct memseg **prev_memsegp; 2483 pgcnt_t num; 2484 2485 /* insert in memseg list, decreasing number of pages order */ 2486 2487 num = MSEG_NPAGES(memsegp); 2488 2489 for (prev_memsegp = &memsegs; *prev_memsegp; 2490 prev_memsegp = &((*prev_memsegp)->next)) { 2491 if (num > MSEG_NPAGES(*prev_memsegp)) 2492 break; 2493 } 2494 2495 memsegp->next = *prev_memsegp; 2496 *prev_memsegp = memsegp; 2497 2498 if (kpm_enable) { 2499 memsegp->nextpa = (memsegp->next) ? 2500 va_to_pa(memsegp->next) : MSEG_NULLPTR_PA; 2501 2502 if (prev_memsegp != &memsegs) { 2503 struct memseg *msp; 2504 msp = (struct memseg *)((caddr_t)prev_memsegp - 2505 offsetof(struct memseg, next)); 2506 msp->nextpa = va_to_pa(memsegp); 2507 } else { 2508 memsegspa = va_to_pa(memsegs); 2509 } 2510 } 2511 } 2512 2513 /* 2514 * PSM add_physmem_cb(). US-II and newer processors have some 2515 * flavor of the prefetch capability implemented. We exploit 2516 * this capability for optimum performance. 2517 */ 2518 #define PREFETCH_BYTES 64 2519 2520 void 2521 add_physmem_cb(page_t *pp, pfn_t pnum) 2522 { 2523 extern void prefetch_page_w(void *); 2524 2525 pp->p_pagenum = pnum; 2526 2527 /* 2528 * Prefetch one more page_t into E$. To prevent future 2529 * mishaps with the sizeof(page_t) changing on us, we 2530 * catch this on debug kernels if we can't bring in the 2531 * entire hpage with 2 PREFETCH_BYTES reads. See 2532 * also, sun4u/cpu/cpu_module.c 2533 */ 2534 /*LINTED*/ 2535 ASSERT(sizeof (page_t) <= 2*PREFETCH_BYTES); 2536 prefetch_page_w((char *)pp); 2537 } 2538 2539 /* 2540 * kphysm_init() tackles the problem of initializing physical memory. 2541 * The old startup made some assumptions about the kernel living in 2542 * physically contiguous space which is no longer valid. 2543 */ 2544 static void 2545 kphysm_init(page_t *pp, struct memseg *memsegp, pgcnt_t npages, 2546 uintptr_t kpm_pp, pgcnt_t kpm_npages) 2547 { 2548 struct memlist *pmem; 2549 struct memseg *msp; 2550 pfn_t base; 2551 pgcnt_t num; 2552 pfn_t lastseg_pages_end = 0; 2553 pgcnt_t nelem_used = 0; 2554 2555 ASSERT(page_hash != NULL && page_hashsz != 0); 2556 2557 msp = memsegp; 2558 for (pmem = phys_avail; pmem && npages; pmem = pmem->next) { 2559 2560 /* 2561 * Build the memsegs entry 2562 */ 2563 num = btop(pmem->size); 2564 if (num > npages) 2565 num = npages; 2566 npages -= num; 2567 base = btop(pmem->address); 2568 2569 msp->pages = pp; 2570 msp->epages = pp + num; 2571 msp->pages_base = base; 2572 msp->pages_end = base + num; 2573 2574 if (kpm_enable) { 2575 pfn_t pbase_a; 2576 pfn_t pend_a; 2577 pfn_t prev_pend_a; 2578 pgcnt_t nelem; 2579 2580 msp->pagespa = va_to_pa(pp); 2581 msp->epagespa = va_to_pa(pp + num); 2582 pbase_a = kpmptop(ptokpmp(base)); 2583 pend_a = kpmptop(ptokpmp(base + num - 1)) + kpmpnpgs; 2584 nelem = ptokpmp(pend_a - pbase_a); 2585 msp->kpm_nkpmpgs = nelem; 2586 msp->kpm_pbase = pbase_a; 2587 if (lastseg_pages_end) { 2588 /* 2589 * Assume phys_avail is in ascending order 2590 * of physical addresses. 2591 */ 2592 ASSERT(base + num > lastseg_pages_end); 2593 prev_pend_a = kpmptop( 2594 ptokpmp(lastseg_pages_end - 1)) + kpmpnpgs; 2595 2596 if (prev_pend_a > pbase_a) { 2597 /* 2598 * Overlap, more than one memseg may 2599 * point to the same kpm_page range. 2600 */ 2601 if (kpm_smallpages == 0) { 2602 msp->kpm_pages = 2603 (kpm_page_t *)kpm_pp - 1; 2604 kpm_pp = (uintptr_t) 2605 ((kpm_page_t *)kpm_pp 2606 + nelem - 1); 2607 } else { 2608 msp->kpm_spages = 2609 (kpm_spage_t *)kpm_pp - 1; 2610 kpm_pp = (uintptr_t) 2611 ((kpm_spage_t *)kpm_pp 2612 + nelem - 1); 2613 } 2614 nelem_used += nelem - 1; 2615 2616 } else { 2617 if (kpm_smallpages == 0) { 2618 msp->kpm_pages = 2619 (kpm_page_t *)kpm_pp; 2620 kpm_pp = (uintptr_t) 2621 ((kpm_page_t *)kpm_pp 2622 + nelem); 2623 } else { 2624 msp->kpm_spages = 2625 (kpm_spage_t *)kpm_pp; 2626 kpm_pp = (uintptr_t) 2627 ((kpm_spage_t *) 2628 kpm_pp + nelem); 2629 } 2630 nelem_used += nelem; 2631 } 2632 2633 } else { 2634 if (kpm_smallpages == 0) { 2635 msp->kpm_pages = (kpm_page_t *)kpm_pp; 2636 kpm_pp = (uintptr_t) 2637 ((kpm_page_t *)kpm_pp + nelem); 2638 } else { 2639 msp->kpm_spages = (kpm_spage_t *)kpm_pp; 2640 kpm_pp = (uintptr_t) 2641 ((kpm_spage_t *)kpm_pp + nelem); 2642 } 2643 nelem_used = nelem; 2644 } 2645 2646 if (nelem_used > kpm_npages) 2647 panic("kphysm_init: kpm_pp overflow\n"); 2648 2649 msp->kpm_pagespa = va_to_pa(msp->kpm_pages); 2650 lastseg_pages_end = msp->pages_end; 2651 } 2652 2653 memseg_list_add(msp); 2654 2655 /* 2656 * add_physmem() initializes the PSM part of the page 2657 * struct by calling the PSM back with add_physmem_cb(). 2658 * In addition it coalesces pages into larger pages as 2659 * it initializes them. 2660 */ 2661 add_physmem(pp, num, base); 2662 pp += num; 2663 msp++; 2664 } 2665 2666 build_pfn_hash(); 2667 } 2668 2669 /* 2670 * Kernel VM initialization. 2671 * Assumptions about kernel address space ordering: 2672 * (1) gap (user space) 2673 * (2) kernel text 2674 * (3) kernel data/bss 2675 * (4) gap 2676 * (5) kernel data structures 2677 * (6) gap 2678 * (7) debugger (optional) 2679 * (8) monitor 2680 * (9) gap (possibly null) 2681 * (10) dvma 2682 * (11) devices 2683 */ 2684 static void 2685 kvm_init(void) 2686 { 2687 /* 2688 * Put the kernel segments in kernel address space. 2689 */ 2690 rw_enter(&kas.a_lock, RW_WRITER); 2691 as_avlinit(&kas); 2692 2693 (void) seg_attach(&kas, (caddr_t)KERNELBASE, 2694 (size_t)(e_moddata - KERNELBASE), &ktextseg); 2695 (void) segkmem_create(&ktextseg); 2696 2697 (void) seg_attach(&kas, (caddr_t)(KERNELBASE + MMU_PAGESIZE4M), 2698 (size_t)(MMU_PAGESIZE4M), &ktexthole); 2699 (void) segkmem_create(&ktexthole); 2700 2701 (void) seg_attach(&kas, (caddr_t)valloc_base, 2702 (size_t)(econtig32 - valloc_base), &kvalloc); 2703 (void) segkmem_create(&kvalloc); 2704 2705 if (kmem64_base) { 2706 (void) seg_attach(&kas, (caddr_t)kmem64_base, 2707 (size_t)(kmem64_end - kmem64_base), &kmem64); 2708 (void) segkmem_create(&kmem64); 2709 } 2710 2711 /* 2712 * We're about to map out /boot. This is the beginning of the 2713 * system resource management transition. We can no longer 2714 * call into /boot for I/O or memory allocations. 2715 */ 2716 (void) seg_attach(&kas, kernelheap, ekernelheap - kernelheap, &kvseg); 2717 (void) segkmem_create(&kvseg); 2718 hblk_alloc_dynamic = 1; 2719 2720 /* 2721 * we need to preallocate pages for DR operations before enabling large 2722 * page kernel heap because of memseg_remap_init() hat_unload() hack. 2723 */ 2724 memseg_remap_init(); 2725 2726 /* at this point we are ready to use large page heap */ 2727 segkmem_heap_lp_init(); 2728 2729 (void) seg_attach(&kas, (caddr_t)SYSBASE32, SYSLIMIT32 - SYSBASE32, 2730 &kvseg32); 2731 (void) segkmem_create(&kvseg32); 2732 2733 /* 2734 * Create a segment for the debugger. 2735 */ 2736 (void) seg_attach(&kas, kdi_segdebugbase, kdi_segdebugsize, &kdebugseg); 2737 (void) segkmem_create(&kdebugseg); 2738 2739 rw_exit(&kas.a_lock); 2740 } 2741 2742 char obp_tte_str[] = 2743 "h# %x constant MMU_PAGESHIFT " 2744 "h# %x constant TTE8K " 2745 "h# %x constant SFHME_SIZE " 2746 "h# %x constant SFHME_TTE " 2747 "h# %x constant HMEBLK_TAG " 2748 "h# %x constant HMEBLK_NEXT " 2749 "h# %x constant HMEBLK_MISC " 2750 "h# %x constant HMEBLK_HME1 " 2751 "h# %x constant NHMENTS " 2752 "h# %x constant HBLK_SZMASK " 2753 "h# %x constant HBLK_RANGE_SHIFT " 2754 "h# %x constant HMEBP_HBLK " 2755 "h# %x constant HMEBUCKET_SIZE " 2756 "h# %x constant HTAG_SFMMUPSZ " 2757 "h# %x constant HTAG_REHASHSZ " 2758 "h# %x constant mmu_hashcnt " 2759 "h# %p constant uhme_hash " 2760 "h# %p constant khme_hash " 2761 "h# %x constant UHMEHASH_SZ " 2762 "h# %x constant KHMEHASH_SZ " 2763 "h# %p constant KCONTEXT " 2764 "h# %p constant KHATID " 2765 "h# %x constant ASI_MEM " 2766 2767 ": PHYS-X@ ( phys -- data ) " 2768 " ASI_MEM spacex@ " 2769 "; " 2770 2771 ": PHYS-W@ ( phys -- data ) " 2772 " ASI_MEM spacew@ " 2773 "; " 2774 2775 ": PHYS-L@ ( phys -- data ) " 2776 " ASI_MEM spaceL@ " 2777 "; " 2778 2779 ": TTE_PAGE_SHIFT ( ttesz -- hmeshift ) " 2780 " 3 * MMU_PAGESHIFT + " 2781 "; " 2782 2783 ": TTE_IS_VALID ( ttep -- flag ) " 2784 " PHYS-X@ 0< " 2785 "; " 2786 2787 ": HME_HASH_SHIFT ( ttesz -- hmeshift ) " 2788 " dup TTE8K = if " 2789 " drop HBLK_RANGE_SHIFT " 2790 " else " 2791 " TTE_PAGE_SHIFT " 2792 " then " 2793 "; " 2794 2795 ": HME_HASH_BSPAGE ( addr hmeshift -- bspage ) " 2796 " tuck >> swap MMU_PAGESHIFT - << " 2797 "; " 2798 2799 ": HME_HASH_FUNCTION ( sfmmup addr hmeshift -- hmebp ) " 2800 " >> over xor swap ( hash sfmmup ) " 2801 " KHATID <> if ( hash ) " 2802 " UHMEHASH_SZ and ( bucket ) " 2803 " HMEBUCKET_SIZE * uhme_hash + ( hmebp ) " 2804 " else ( hash ) " 2805 " KHMEHASH_SZ and ( bucket ) " 2806 " HMEBUCKET_SIZE * khme_hash + ( hmebp ) " 2807 " then ( hmebp ) " 2808 "; " 2809 2810 ": HME_HASH_TABLE_SEARCH " 2811 " ( sfmmup hmebp hblktag -- sfmmup null | sfmmup hmeblkp ) " 2812 " >r hmebp_hblk + phys-x@ begin ( sfmmup hmeblkp ) ( r: hblktag ) " 2813 " dup if ( sfmmup hmeblkp ) ( r: hblktag ) " 2814 " dup hmeblk_tag + phys-x@ r@ = if ( sfmmup hmeblkp ) " 2815 " dup hmeblk_tag + 8 + phys-x@ 2 pick = if " 2816 " true ( sfmmup hmeblkp true ) ( r: hblktag ) " 2817 " else " 2818 " hmeblk_next + phys-x@ false " 2819 " ( sfmmup hmeblkp false ) ( r: hblktag ) " 2820 " then " 2821 " else " 2822 " hmeblk_next + phys-x@ false " 2823 " ( sfmmup hmeblkp false ) ( r: hblktag ) " 2824 " then " 2825 " else " 2826 " true " 2827 " then " 2828 " until r> drop " 2829 "; " 2830 2831 ": HME_HASH_TAG ( sfmmup rehash addr -- hblktag ) " 2832 " over HME_HASH_SHIFT HME_HASH_BSPAGE ( sfmmup rehash bspage ) " 2833 " HTAG_REHASHSZ << or nip ( hblktag ) " 2834 "; " 2835 2836 ": HBLK_TO_TTEP ( hmeblkp addr -- ttep ) " 2837 " over HMEBLK_MISC + PHYS-L@ HBLK_SZMASK and ( hmeblkp addr ttesz ) " 2838 " TTE8K = if ( hmeblkp addr ) " 2839 " MMU_PAGESHIFT >> NHMENTS 1- and ( hmeblkp hme-index ) " 2840 " else ( hmeblkp addr ) " 2841 " drop 0 ( hmeblkp 0 ) " 2842 " then ( hmeblkp hme-index ) " 2843 " SFHME_SIZE * + HMEBLK_HME1 + ( hmep ) " 2844 " SFHME_TTE + ( ttep ) " 2845 "; " 2846 2847 ": unix-tte ( addr cnum -- false | tte-data true ) " 2848 " KCONTEXT = if ( addr ) " 2849 " KHATID ( addr khatid ) " 2850 " else ( addr ) " 2851 " drop false exit ( false ) " 2852 " then " 2853 " ( addr khatid ) " 2854 " mmu_hashcnt 1+ 1 do ( addr sfmmup ) " 2855 " 2dup swap i HME_HASH_SHIFT " 2856 "( addr sfmmup sfmmup addr hmeshift ) " 2857 " HME_HASH_FUNCTION ( addr sfmmup hmebp ) " 2858 " over i 4 pick " 2859 "( addr sfmmup hmebp sfmmup rehash addr ) " 2860 " HME_HASH_TAG ( addr sfmmup hmebp hblktag ) " 2861 " HME_HASH_TABLE_SEARCH " 2862 "( addr sfmmup { null | hmeblkp } ) " 2863 " ?dup if ( addr sfmmup hmeblkp ) " 2864 " nip swap HBLK_TO_TTEP ( ttep ) " 2865 " dup TTE_IS_VALID if ( valid-ttep ) " 2866 " PHYS-X@ true ( tte-data true ) " 2867 " else ( invalid-tte ) " 2868 " drop false ( false ) " 2869 " then ( false | tte-data true ) " 2870 " unloop exit ( false | tte-data true ) " 2871 " then ( addr sfmmup ) " 2872 " loop ( addr sfmmup ) " 2873 " 2drop false ( false ) " 2874 "; " 2875 ; 2876 2877 void 2878 create_va_to_tte(void) 2879 { 2880 char *bp; 2881 extern int khmehash_num, uhmehash_num; 2882 extern struct hmehash_bucket *khme_hash, *uhme_hash; 2883 2884 #define OFFSET(type, field) ((uintptr_t)(&((type *)0)->field)) 2885 2886 bp = (char *)kobj_zalloc(MMU_PAGESIZE, KM_SLEEP); 2887 2888 /* 2889 * Teach obp how to parse our sw ttes. 2890 */ 2891 (void) sprintf(bp, obp_tte_str, 2892 MMU_PAGESHIFT, 2893 TTE8K, 2894 sizeof (struct sf_hment), 2895 OFFSET(struct sf_hment, hme_tte), 2896 OFFSET(struct hme_blk, hblk_tag), 2897 OFFSET(struct hme_blk, hblk_nextpa), 2898 OFFSET(struct hme_blk, hblk_misc), 2899 OFFSET(struct hme_blk, hblk_hme), 2900 NHMENTS, 2901 HBLK_SZMASK, 2902 HBLK_RANGE_SHIFT, 2903 OFFSET(struct hmehash_bucket, hmeh_nextpa), 2904 sizeof (struct hmehash_bucket), 2905 HTAG_SFMMUPSZ, 2906 HTAG_REHASHSZ, 2907 mmu_hashcnt, 2908 (caddr_t)va_to_pa((caddr_t)uhme_hash), 2909 (caddr_t)va_to_pa((caddr_t)khme_hash), 2910 UHMEHASH_SZ, 2911 KHMEHASH_SZ, 2912 KCONTEXT, 2913 KHATID, 2914 ASI_MEM); 2915 prom_interpret(bp, 0, 0, 0, 0, 0); 2916 2917 kobj_free(bp, MMU_PAGESIZE); 2918 } 2919 2920 void 2921 install_va_to_tte(void) 2922 { 2923 /* 2924 * advise prom that he can use unix-tte 2925 */ 2926 prom_interpret("' unix-tte is va>tte-data", 0, 0, 0, 0, 0); 2927 } 2928 2929 2930 /* 2931 * Because kmdb links prom_stdout_is_framebuffer into its own 2932 * module, we add "device-type=display" here for /os-io node, so that 2933 * prom_stdout_is_framebuffer still works corrrectly after /os-io node 2934 * is registered into OBP. 2935 */ 2936 static char *create_node = 2937 "\" /\" find-device " 2938 "new-device " 2939 "\" os-io\" device-name " 2940 "\" display\" device-type " 2941 ": cb-r/w ( adr,len method$ -- #read/#written ) " 2942 " 2>r swap 2 2r> ['] $callback catch if " 2943 " 2drop 3drop 0 " 2944 " then " 2945 "; " 2946 ": read ( adr,len -- #read ) " 2947 " \" read\" ['] cb-r/w catch if 2drop 2drop -2 exit then " 2948 " ( retN ... ret1 N ) " 2949 " ?dup if " 2950 " swap >r 1- 0 ?do drop loop r> " 2951 " else " 2952 " -2 " 2953 " then " 2954 "; " 2955 ": write ( adr,len -- #written ) " 2956 " \" write\" ['] cb-r/w catch if 2drop 2drop 0 exit then " 2957 " ( retN ... ret1 N ) " 2958 " ?dup if " 2959 " swap >r 1- 0 ?do drop loop r> " 2960 " else " 2961 " 0 " 2962 " then " 2963 "; " 2964 ": poll-tty ( -- ) ; " 2965 ": install-abort ( -- ) ['] poll-tty d# 10 alarm ; " 2966 ": remove-abort ( -- ) ['] poll-tty 0 alarm ; " 2967 ": cb-give/take ( $method -- ) " 2968 " 0 -rot ['] $callback catch ?dup if " 2969 " >r 2drop 2drop r> throw " 2970 " else " 2971 " 0 ?do drop loop " 2972 " then " 2973 "; " 2974 ": give ( -- ) \" exit-input\" cb-give/take ; " 2975 ": take ( -- ) \" enter-input\" cb-give/take ; " 2976 ": open ( -- ok? ) true ; " 2977 ": close ( -- ) ; " 2978 "finish-device " 2979 "device-end "; 2980 2981 /* 2982 * Create the OBP input/output node (FCode serial driver). 2983 * It is needed for both USB console keyboard and for 2984 * the kernel terminal emulator. It is too early to check for a 2985 * kernel console compatible framebuffer now, so we create this 2986 * so that we're ready if we need to enable kernel terminal emulation. 2987 * 2988 * When the USB software takes over the input device at the time 2989 * consconfig runs, OBP's stdin is redirected to this node. 2990 * Whenever the FORTH user interface is used after this switch, 2991 * the node will call back into the kernel for console input. 2992 * If a serial device such as ttya or a UART with a Type 5 keyboard 2993 * attached is used, OBP takes over the serial device when the system 2994 * goes to the debugger after the system is booted. This sharing 2995 * of the relatively simple serial device is difficult but possible. 2996 * Sharing the USB host controller is impossible due its complexity. 2997 * 2998 * Similarly to USB keyboard input redirection, after consconfig_dacf 2999 * configures a kernel console framebuffer as the standard output 3000 * device, OBP's stdout is switched to to vector through the 3001 * /os-io node into the kernel terminal emulator. 3002 */ 3003 static void 3004 startup_create_io_node(void) 3005 { 3006 prom_interpret(create_node, 0, 0, 0, 0, 0); 3007 } 3008 3009 3010 static void 3011 do_prom_version_check(void) 3012 { 3013 int i; 3014 pnode_t node; 3015 char buf[64]; 3016 static char drev[] = "Down-rev firmware detected%s\n" 3017 "\tPlease upgrade to the following minimum version:\n" 3018 "\t\t%s\n"; 3019 3020 i = prom_version_check(buf, sizeof (buf), &node); 3021 3022 if (i == PROM_VER64_OK) 3023 return; 3024 3025 if (i == PROM_VER64_UPGRADE) { 3026 cmn_err(CE_WARN, drev, "", buf); 3027 3028 #ifdef DEBUG 3029 prom_enter_mon(); /* Type 'go' to continue */ 3030 cmn_err(CE_WARN, "Booting with down-rev firmware\n"); 3031 return; 3032 #else 3033 halt(0); 3034 #endif 3035 } 3036 3037 /* 3038 * The other possibility is that this is a server running 3039 * good firmware, but down-rev firmware was detected on at 3040 * least one other cpu board. We just complain if we see 3041 * that. 3042 */ 3043 cmn_err(CE_WARN, drev, " on one or more CPU boards", buf); 3044 } 3045 3046 static void 3047 kpm_init() 3048 { 3049 kpm_pgshft = (kpm_smallpages == 0) ? MMU_PAGESHIFT4M : MMU_PAGESHIFT; 3050 kpm_pgsz = 1ull << kpm_pgshft; 3051 kpm_pgoff = kpm_pgsz - 1; 3052 kpmp2pshft = kpm_pgshft - PAGESHIFT; 3053 kpmpnpgs = 1 << kpmp2pshft; 3054 ASSERT(((uintptr_t)kpm_vbase & (kpm_pgsz - 1)) == 0); 3055 } 3056 3057 void 3058 kpm_npages_setup(int memblocks) 3059 { 3060 /* 3061 * npages can be scattered in a maximum of 'memblocks' 3062 */ 3063 kpm_npages = ptokpmpr(npages) + memblocks; 3064 } 3065 3066 /* 3067 * Must be defined in platform dependent code. 3068 */ 3069 extern caddr_t modtext; 3070 extern size_t modtext_sz; 3071 extern caddr_t moddata; 3072 3073 #define HEAPTEXT_ARENA(addr) \ 3074 ((uintptr_t)(addr) < KERNELBASE + 2 * MMU_PAGESIZE4M ? 0 : \ 3075 (((uintptr_t)(addr) - HEAPTEXT_BASE) / \ 3076 (HEAPTEXT_MAPPED + HEAPTEXT_UNMAPPED) + 1)) 3077 3078 #define HEAPTEXT_OVERSIZED(addr) \ 3079 ((uintptr_t)(addr) >= HEAPTEXT_BASE + HEAPTEXT_SIZE - HEAPTEXT_OVERSIZE) 3080 3081 vmem_t *texthole_source[HEAPTEXT_NARENAS]; 3082 vmem_t *texthole_arena[HEAPTEXT_NARENAS]; 3083 kmutex_t texthole_lock; 3084 3085 char kern_bootargs[OBP_MAXPATHLEN]; 3086 3087 void 3088 kobj_vmem_init(vmem_t **text_arena, vmem_t **data_arena) 3089 { 3090 uintptr_t addr, limit; 3091 3092 addr = HEAPTEXT_BASE; 3093 limit = addr + HEAPTEXT_SIZE - HEAPTEXT_OVERSIZE; 3094 3095 /* 3096 * Before we initialize the text_arena, we want to punch holes in the 3097 * underlying heaptext_arena. This guarantees that for any text 3098 * address we can find a text hole less than HEAPTEXT_MAPPED away. 3099 */ 3100 for (; addr + HEAPTEXT_UNMAPPED <= limit; 3101 addr += HEAPTEXT_MAPPED + HEAPTEXT_UNMAPPED) { 3102 (void) vmem_xalloc(heaptext_arena, HEAPTEXT_UNMAPPED, PAGESIZE, 3103 0, 0, (void *)addr, (void *)(addr + HEAPTEXT_UNMAPPED), 3104 VM_NOSLEEP | VM_BESTFIT | VM_PANIC); 3105 } 3106 3107 /* 3108 * Allocate one page at the oversize to break up the text region 3109 * from the oversized region. 3110 */ 3111 (void) vmem_xalloc(heaptext_arena, PAGESIZE, PAGESIZE, 0, 0, 3112 (void *)limit, (void *)(limit + PAGESIZE), 3113 VM_NOSLEEP | VM_BESTFIT | VM_PANIC); 3114 3115 *text_arena = vmem_create("module_text", modtext, modtext_sz, 3116 sizeof (uintptr_t), segkmem_alloc, segkmem_free, 3117 heaptext_arena, 0, VM_SLEEP); 3118 *data_arena = vmem_create("module_data", moddata, MODDATA, 1, 3119 segkmem_alloc, segkmem_free, heap32_arena, 0, VM_SLEEP); 3120 } 3121 3122 caddr_t 3123 kobj_text_alloc(vmem_t *arena, size_t size) 3124 { 3125 caddr_t rval, better; 3126 3127 /* 3128 * First, try a sleeping allocation. 3129 */ 3130 rval = vmem_alloc(arena, size, VM_SLEEP | VM_BESTFIT); 3131 3132 if (size >= HEAPTEXT_MAPPED || !HEAPTEXT_OVERSIZED(rval)) 3133 return (rval); 3134 3135 /* 3136 * We didn't get the area that we wanted. We're going to try to do an 3137 * allocation with explicit constraints. 3138 */ 3139 better = vmem_xalloc(arena, size, sizeof (uintptr_t), 0, 0, NULL, 3140 (void *)(HEAPTEXT_BASE + HEAPTEXT_SIZE - HEAPTEXT_OVERSIZE), 3141 VM_NOSLEEP | VM_BESTFIT); 3142 3143 if (better != NULL) { 3144 /* 3145 * That worked. Free our first attempt and return. 3146 */ 3147 vmem_free(arena, rval, size); 3148 return (better); 3149 } 3150 3151 /* 3152 * That didn't work; we'll have to return our first attempt. 3153 */ 3154 return (rval); 3155 } 3156 3157 caddr_t 3158 kobj_texthole_alloc(caddr_t addr, size_t size) 3159 { 3160 int arena = HEAPTEXT_ARENA(addr); 3161 char c[30]; 3162 uintptr_t base; 3163 3164 if (HEAPTEXT_OVERSIZED(addr)) { 3165 /* 3166 * If this is an oversized allocation, there is no text hole 3167 * available for it; return NULL. 3168 */ 3169 return (NULL); 3170 } 3171 3172 mutex_enter(&texthole_lock); 3173 3174 if (texthole_arena[arena] == NULL) { 3175 ASSERT(texthole_source[arena] == NULL); 3176 3177 if (arena == 0) { 3178 texthole_source[0] = vmem_create("module_text_holesrc", 3179 (void *)(KERNELBASE + MMU_PAGESIZE4M), 3180 MMU_PAGESIZE4M, PAGESIZE, NULL, NULL, NULL, 3181 0, VM_SLEEP); 3182 } else { 3183 base = HEAPTEXT_BASE + 3184 (arena - 1) * (HEAPTEXT_MAPPED + HEAPTEXT_UNMAPPED); 3185 3186 (void) snprintf(c, sizeof (c), 3187 "heaptext_holesrc_%d", arena); 3188 3189 texthole_source[arena] = vmem_create(c, (void *)base, 3190 HEAPTEXT_UNMAPPED, PAGESIZE, NULL, NULL, NULL, 3191 0, VM_SLEEP); 3192 } 3193 3194 (void) snprintf(c, sizeof (c), "heaptext_hole_%d", arena); 3195 3196 texthole_arena[arena] = vmem_create(c, NULL, 0, 3197 sizeof (uint32_t), segkmem_alloc_permanent, segkmem_free, 3198 texthole_source[arena], 0, VM_SLEEP); 3199 } 3200 3201 mutex_exit(&texthole_lock); 3202 3203 ASSERT(texthole_arena[arena] != NULL); 3204 ASSERT(arena >= 0 && arena < HEAPTEXT_NARENAS); 3205 return (vmem_alloc(texthole_arena[arena], size, 3206 VM_BESTFIT | VM_NOSLEEP)); 3207 } 3208 3209 void 3210 kobj_texthole_free(caddr_t addr, size_t size) 3211 { 3212 int arena = HEAPTEXT_ARENA(addr); 3213 3214 ASSERT(arena >= 0 && arena < HEAPTEXT_NARENAS); 3215 ASSERT(texthole_arena[arena] != NULL); 3216 vmem_free(texthole_arena[arena], addr, size); 3217 } 3218