17c478bd9Sstevel@tonic-gate /* 27c478bd9Sstevel@tonic-gate * CDDL HEADER START 37c478bd9Sstevel@tonic-gate * 47c478bd9Sstevel@tonic-gate * The contents of this file are subject to the terms of the 5100b72f4Sandrei * Common Development and Distribution License (the "License"). 6100b72f4Sandrei * You may not use this file except in compliance with the License. 77c478bd9Sstevel@tonic-gate * 87c478bd9Sstevel@tonic-gate * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 97c478bd9Sstevel@tonic-gate * or http://www.opensolaris.org/os/licensing. 107c478bd9Sstevel@tonic-gate * See the License for the specific language governing permissions 117c478bd9Sstevel@tonic-gate * and limitations under the License. 127c478bd9Sstevel@tonic-gate * 137c478bd9Sstevel@tonic-gate * When distributing Covered Code, include this CDDL HEADER in each 147c478bd9Sstevel@tonic-gate * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 157c478bd9Sstevel@tonic-gate * If applicable, add the following below this CDDL HEADER, with the 167c478bd9Sstevel@tonic-gate * fields enclosed by brackets "[]" replaced with your own identifying 177c478bd9Sstevel@tonic-gate * information: Portions Copyright [yyyy] [name of copyright owner] 187c478bd9Sstevel@tonic-gate * 197c478bd9Sstevel@tonic-gate * CDDL HEADER END 207c478bd9Sstevel@tonic-gate */ 217c478bd9Sstevel@tonic-gate /* 22fb2f18f8Sesaxe * Copyright 2007 Sun Microsystems, Inc. All rights reserved. 237c478bd9Sstevel@tonic-gate * Use is subject to license terms. 247c478bd9Sstevel@tonic-gate */ 257c478bd9Sstevel@tonic-gate 267c478bd9Sstevel@tonic-gate #pragma ident "%Z%%M% %I% %E% SMI" 277c478bd9Sstevel@tonic-gate 287c478bd9Sstevel@tonic-gate #include <sys/types.h> 297c478bd9Sstevel@tonic-gate #include <sys/t_lock.h> 307c478bd9Sstevel@tonic-gate #include <sys/param.h> 317c478bd9Sstevel@tonic-gate #include <sys/sysmacros.h> 327c478bd9Sstevel@tonic-gate #include <sys/signal.h> 337c478bd9Sstevel@tonic-gate #include <sys/systm.h> 347c478bd9Sstevel@tonic-gate #include <sys/user.h> 357c478bd9Sstevel@tonic-gate #include <sys/mman.h> 367c478bd9Sstevel@tonic-gate #include <sys/vm.h> 377c478bd9Sstevel@tonic-gate #include <sys/conf.h> 387c478bd9Sstevel@tonic-gate #include <sys/avintr.h> 397c478bd9Sstevel@tonic-gate #include <sys/autoconf.h> 407c478bd9Sstevel@tonic-gate #include <sys/disp.h> 417c478bd9Sstevel@tonic-gate #include <sys/class.h> 427c478bd9Sstevel@tonic-gate #include <sys/bitmap.h> 437c478bd9Sstevel@tonic-gate 447c478bd9Sstevel@tonic-gate #include <sys/privregs.h> 457c478bd9Sstevel@tonic-gate 467c478bd9Sstevel@tonic-gate #include <sys/proc.h> 477c478bd9Sstevel@tonic-gate #include <sys/buf.h> 487c478bd9Sstevel@tonic-gate #include <sys/kmem.h> 49ae115bc7Smrj #include <sys/mem.h> 507c478bd9Sstevel@tonic-gate #include <sys/kstat.h> 517c478bd9Sstevel@tonic-gate 527c478bd9Sstevel@tonic-gate #include <sys/reboot.h> 537c478bd9Sstevel@tonic-gate 547c478bd9Sstevel@tonic-gate #include <sys/cred.h> 557c478bd9Sstevel@tonic-gate #include <sys/vnode.h> 567c478bd9Sstevel@tonic-gate #include <sys/file.h> 577c478bd9Sstevel@tonic-gate 587c478bd9Sstevel@tonic-gate #include <sys/procfs.h> 597c478bd9Sstevel@tonic-gate 607c478bd9Sstevel@tonic-gate #include <sys/vfs.h> 617c478bd9Sstevel@tonic-gate #include <sys/cmn_err.h> 627c478bd9Sstevel@tonic-gate #include <sys/utsname.h> 637c478bd9Sstevel@tonic-gate #include <sys/debug.h> 647c478bd9Sstevel@tonic-gate #include <sys/kdi.h> 657c478bd9Sstevel@tonic-gate 667c478bd9Sstevel@tonic-gate #include <sys/dumphdr.h> 677c478bd9Sstevel@tonic-gate #include <sys/bootconf.h> 687c478bd9Sstevel@tonic-gate #include <sys/varargs.h> 697c478bd9Sstevel@tonic-gate #include <sys/promif.h> 707c478bd9Sstevel@tonic-gate #include <sys/modctl.h> /* for "procfs" hack */ 717c478bd9Sstevel@tonic-gate 727c478bd9Sstevel@tonic-gate #include <sys/sunddi.h> 737c478bd9Sstevel@tonic-gate #include <sys/sunndi.h> 747c478bd9Sstevel@tonic-gate #include <sys/ndi_impldefs.h> 757c478bd9Sstevel@tonic-gate #include <sys/ddidmareq.h> 767c478bd9Sstevel@tonic-gate #include <sys/psw.h> 777c478bd9Sstevel@tonic-gate #include <sys/regset.h> 787c478bd9Sstevel@tonic-gate #include <sys/clock.h> 797c478bd9Sstevel@tonic-gate #include <sys/pte.h> 807c478bd9Sstevel@tonic-gate #include <sys/tss.h> 817c478bd9Sstevel@tonic-gate #include <sys/stack.h> 827c478bd9Sstevel@tonic-gate #include <sys/trap.h> 837c478bd9Sstevel@tonic-gate #include <sys/fp.h> 847c478bd9Sstevel@tonic-gate #include <vm/anon.h> 857c478bd9Sstevel@tonic-gate #include <vm/as.h> 867c478bd9Sstevel@tonic-gate #include <vm/page.h> 877c478bd9Sstevel@tonic-gate #include <vm/seg.h> 887c478bd9Sstevel@tonic-gate #include <vm/seg_dev.h> 897c478bd9Sstevel@tonic-gate #include <vm/seg_kmem.h> 907c478bd9Sstevel@tonic-gate #include <vm/seg_kpm.h> 917c478bd9Sstevel@tonic-gate #include <vm/seg_map.h> 927c478bd9Sstevel@tonic-gate #include <vm/seg_vn.h> 937c478bd9Sstevel@tonic-gate #include <vm/seg_kp.h> 947c478bd9Sstevel@tonic-gate #include <sys/memnode.h> 957c478bd9Sstevel@tonic-gate #include <vm/vm_dep.h> 967c478bd9Sstevel@tonic-gate #include <sys/thread.h> 977c478bd9Sstevel@tonic-gate #include <sys/sysconf.h> 987c478bd9Sstevel@tonic-gate #include <sys/vm_machparam.h> 997c478bd9Sstevel@tonic-gate #include <sys/archsystm.h> 1007c478bd9Sstevel@tonic-gate #include <sys/machsystm.h> 1017c478bd9Sstevel@tonic-gate #include <vm/hat.h> 1027c478bd9Sstevel@tonic-gate #include <vm/hat_i86.h> 1037c478bd9Sstevel@tonic-gate #include <sys/pmem.h> 1047c478bd9Sstevel@tonic-gate #include <sys/smp_impldefs.h> 1057c478bd9Sstevel@tonic-gate #include <sys/x86_archext.h> 1067c478bd9Sstevel@tonic-gate #include <sys/segments.h> 1077c478bd9Sstevel@tonic-gate #include <sys/clconf.h> 1087c478bd9Sstevel@tonic-gate #include <sys/kobj.h> 1097c478bd9Sstevel@tonic-gate #include <sys/kobj_lex.h> 1107c478bd9Sstevel@tonic-gate #include <sys/cpc_impl.h> 111fb2f18f8Sesaxe #include <sys/pg.h> 1127c478bd9Sstevel@tonic-gate #include <sys/x86_archext.h> 1137aec1d6eScindi #include <sys/cpu_module.h> 11484ab085aSmws #include <sys/smbios.h> 115ae115bc7Smrj #include <sys/debug_info.h> 116ae115bc7Smrj 117ae115bc7Smrj 118ae115bc7Smrj #include <sys/bootinfo.h> 119ae115bc7Smrj #include <vm/kboot_mmu.h> 1207c478bd9Sstevel@tonic-gate 1217c478bd9Sstevel@tonic-gate extern void progressbar_init(void); 1227c478bd9Sstevel@tonic-gate extern void progressbar_start(void); 1239acbbeafSnn extern void brand_init(void); 1247c478bd9Sstevel@tonic-gate 1257c478bd9Sstevel@tonic-gate /* 1267c478bd9Sstevel@tonic-gate * XXX make declaration below "static" when drivers no longer use this 1277c478bd9Sstevel@tonic-gate * interface. 1287c478bd9Sstevel@tonic-gate */ 1297c478bd9Sstevel@tonic-gate extern caddr_t p0_va; /* Virtual address for accessing physical page 0 */ 1307c478bd9Sstevel@tonic-gate 1317c478bd9Sstevel@tonic-gate /* 1327c478bd9Sstevel@tonic-gate * segkp 1337c478bd9Sstevel@tonic-gate */ 1347c478bd9Sstevel@tonic-gate extern int segkp_fromheap; 1357c478bd9Sstevel@tonic-gate 1367c478bd9Sstevel@tonic-gate static void kvm_init(void); 1377c478bd9Sstevel@tonic-gate static void startup_init(void); 1387c478bd9Sstevel@tonic-gate static void startup_memlist(void); 139ae115bc7Smrj static void startup_kmem(void); 1407c478bd9Sstevel@tonic-gate static void startup_modules(void); 1417c478bd9Sstevel@tonic-gate static void startup_vm(void); 1427c478bd9Sstevel@tonic-gate static void startup_end(void); 143*35b1ab99Sjosephb static void layout_kernel_va(void); 1447c478bd9Sstevel@tonic-gate 1457c478bd9Sstevel@tonic-gate /* 1467c478bd9Sstevel@tonic-gate * Declare these as initialized data so we can patch them. 1477c478bd9Sstevel@tonic-gate */ 148f53ad214Skchow #ifdef __i386 149*35b1ab99Sjosephb 150f53ad214Skchow /* 151f53ad214Skchow * Due to virtual address space limitations running in 32 bit mode, restrict 152*35b1ab99Sjosephb * the amount of physical memory configured to a max of PHYSMEM pages (16g). 153f53ad214Skchow * 154f53ad214Skchow * If the physical max memory size of 64g were allowed to be configured, the 155f53ad214Skchow * size of user virtual address space will be less than 1g. A limited user 156f53ad214Skchow * address space greatly reduces the range of applications that can run. 157f53ad214Skchow * 158*35b1ab99Sjosephb * If more physical memory than PHYSMEM is required, users should preferably 159*35b1ab99Sjosephb * run in 64 bit mode which has far looser virtual address space limitations. 160f53ad214Skchow * 161f53ad214Skchow * If 64 bit mode is not available (as in IA32) and/or more physical memory 162*35b1ab99Sjosephb * than PHYSMEM is required in 32 bit mode, physmem can be set to the desired 163f53ad214Skchow * value or to 0 (to configure all available memory) via eeprom(1M). kernelbase 164f53ad214Skchow * should also be carefully tuned to balance out the need of the user 165f53ad214Skchow * application while minimizing the risk of kernel heap exhaustion due to 166f53ad214Skchow * kernelbase being set too high. 167f53ad214Skchow */ 168*35b1ab99Sjosephb #define PHYSMEM 0x400000 169f53ad214Skchow 170*35b1ab99Sjosephb #else /* __amd64 */ 171*35b1ab99Sjosephb 172*35b1ab99Sjosephb /* 173*35b1ab99Sjosephb * For now we can handle memory with physical addresses up to about 174*35b1ab99Sjosephb * 64 Terabytes. This keeps the kernel above the VA hole, leaving roughly 175*35b1ab99Sjosephb * half the VA space for seg_kpm. When systems get bigger than 64TB this 176*35b1ab99Sjosephb * code will need revisiting. There is an implicit assumption that there 177*35b1ab99Sjosephb * are no *huge* holes in the physical address space too. 178*35b1ab99Sjosephb */ 179*35b1ab99Sjosephb #define TERABYTE (1ul << 40) 180*35b1ab99Sjosephb #define PHYSMEM_MAX64 mmu_btop(64 * TERABYTE) 181*35b1ab99Sjosephb #define PHYSMEM PHYSMEM_MAX64 182*35b1ab99Sjosephb #define AMD64_VA_HOLE_END 0xFFFF800000000000ul 183*35b1ab99Sjosephb 184*35b1ab99Sjosephb #endif /* __amd64 */ 185*35b1ab99Sjosephb 186*35b1ab99Sjosephb pgcnt_t physmem = PHYSMEM; 1877c478bd9Sstevel@tonic-gate pgcnt_t obp_pages; /* Memory used by PROM for its text and data */ 1887c478bd9Sstevel@tonic-gate 1897c478bd9Sstevel@tonic-gate char *kobj_file_buf; 1907c478bd9Sstevel@tonic-gate int kobj_file_bufsize; /* set in /etc/system */ 1917c478bd9Sstevel@tonic-gate 1927c478bd9Sstevel@tonic-gate /* Global variables for MP support. Used in mp_startup */ 1937c478bd9Sstevel@tonic-gate caddr_t rm_platter_va; 1947c478bd9Sstevel@tonic-gate uint32_t rm_platter_pa; 1957c478bd9Sstevel@tonic-gate 19683f9b804Skchow int auto_lpg_disable = 1; 19783f9b804Skchow 1987c478bd9Sstevel@tonic-gate /* 1997c478bd9Sstevel@tonic-gate * Some CPUs have holes in the middle of the 64-bit virtual address range. 2007c478bd9Sstevel@tonic-gate */ 2017c478bd9Sstevel@tonic-gate uintptr_t hole_start, hole_end; 2027c478bd9Sstevel@tonic-gate 2037c478bd9Sstevel@tonic-gate /* 2047c478bd9Sstevel@tonic-gate * kpm mapping window 2057c478bd9Sstevel@tonic-gate */ 2067c478bd9Sstevel@tonic-gate caddr_t kpm_vbase; 2077c478bd9Sstevel@tonic-gate size_t kpm_size; 208*35b1ab99Sjosephb static int kpm_desired; 209*35b1ab99Sjosephb #ifdef __amd64 210*35b1ab99Sjosephb static uintptr_t segkpm_base = (uintptr_t)SEGKPM_BASE; 211*35b1ab99Sjosephb #endif 2127c478bd9Sstevel@tonic-gate 2137c478bd9Sstevel@tonic-gate /* 2147c478bd9Sstevel@tonic-gate * Configuration parameters set at boot time. 2157c478bd9Sstevel@tonic-gate */ 2167c478bd9Sstevel@tonic-gate 2177c478bd9Sstevel@tonic-gate caddr_t econtig; /* end of first block of contiguous kernel */ 2187c478bd9Sstevel@tonic-gate 2197c478bd9Sstevel@tonic-gate struct bootops *bootops = 0; /* passed in from boot */ 2207c478bd9Sstevel@tonic-gate struct bootops **bootopsp; 2217c478bd9Sstevel@tonic-gate struct boot_syscalls *sysp; /* passed in from boot */ 2227c478bd9Sstevel@tonic-gate 2237c478bd9Sstevel@tonic-gate char bootblock_fstype[16]; 2247c478bd9Sstevel@tonic-gate 2257c478bd9Sstevel@tonic-gate char kern_bootargs[OBP_MAXPATHLEN]; 2267c478bd9Sstevel@tonic-gate 227ad23a2dbSjohansen /* 228ad23a2dbSjohansen * ZFS zio segment. This allows us to exclude large portions of ZFS data that 229ad23a2dbSjohansen * gets cached in kmem caches on the heap. If this is set to zero, we allocate 230ad23a2dbSjohansen * zio buffers from their own segment, otherwise they are allocated from the 231ad23a2dbSjohansen * heap. The optimization of allocating zio buffers from their own segment is 232ad23a2dbSjohansen * only valid on 64-bit kernels. 233ad23a2dbSjohansen */ 234ad23a2dbSjohansen #if defined(__amd64) 235ad23a2dbSjohansen int segzio_fromheap = 0; 236ad23a2dbSjohansen #else 237ad23a2dbSjohansen int segzio_fromheap = 1; 238ad23a2dbSjohansen #endif 239ad23a2dbSjohansen 2407c478bd9Sstevel@tonic-gate /* 2417c478bd9Sstevel@tonic-gate * new memory fragmentations are possible in startup() due to BOP_ALLOCs. this 2427c478bd9Sstevel@tonic-gate * depends on number of BOP_ALLOC calls made and requested size, memory size 2437c478bd9Sstevel@tonic-gate * combination and whether boot.bin memory needs to be freed. 2447c478bd9Sstevel@tonic-gate */ 2457c478bd9Sstevel@tonic-gate #define POSS_NEW_FRAGMENTS 12 2467c478bd9Sstevel@tonic-gate 2477c478bd9Sstevel@tonic-gate /* 2487c478bd9Sstevel@tonic-gate * VM data structures 2497c478bd9Sstevel@tonic-gate */ 2507c478bd9Sstevel@tonic-gate long page_hashsz; /* Size of page hash table (power of two) */ 2517c478bd9Sstevel@tonic-gate struct page *pp_base; /* Base of initial system page struct array */ 2527c478bd9Sstevel@tonic-gate struct page **page_hash; /* Page hash table */ 2537c478bd9Sstevel@tonic-gate struct seg ktextseg; /* Segment used for kernel executable image */ 2547c478bd9Sstevel@tonic-gate struct seg kvalloc; /* Segment used for "valloc" mapping */ 2557c478bd9Sstevel@tonic-gate struct seg kpseg; /* Segment used for pageable kernel virt mem */ 2567c478bd9Sstevel@tonic-gate struct seg kmapseg; /* Segment used for generic kernel mappings */ 2577c478bd9Sstevel@tonic-gate struct seg kdebugseg; /* Segment used for the kernel debugger */ 2587c478bd9Sstevel@tonic-gate 2597c478bd9Sstevel@tonic-gate struct seg *segkmap = &kmapseg; /* Kernel generic mapping segment */ 260ae115bc7Smrj static struct seg *segmap = &kmapseg; /* easier to use name for in here */ 261ae115bc7Smrj 2627c478bd9Sstevel@tonic-gate struct seg *segkp = &kpseg; /* Pageable kernel virtual memory segment */ 2637c478bd9Sstevel@tonic-gate 2647c478bd9Sstevel@tonic-gate #if defined(__amd64) 2657c478bd9Sstevel@tonic-gate struct seg kvseg_core; /* Segment used for the core heap */ 2667c478bd9Sstevel@tonic-gate struct seg kpmseg; /* Segment used for physical mapping */ 2677c478bd9Sstevel@tonic-gate struct seg *segkpm = &kpmseg; /* 64bit kernel physical mapping segment */ 2687c478bd9Sstevel@tonic-gate #else 2697c478bd9Sstevel@tonic-gate struct seg *segkpm = NULL; /* Unused on IA32 */ 2707c478bd9Sstevel@tonic-gate #endif 2717c478bd9Sstevel@tonic-gate 2727c478bd9Sstevel@tonic-gate caddr_t segkp_base; /* Base address of segkp */ 273ad23a2dbSjohansen caddr_t segzio_base; /* Base address of segzio */ 2747c478bd9Sstevel@tonic-gate #if defined(__amd64) 2757c478bd9Sstevel@tonic-gate pgcnt_t segkpsize = btop(SEGKPDEFSIZE); /* size of segkp segment in pages */ 2767c478bd9Sstevel@tonic-gate #else 2777c478bd9Sstevel@tonic-gate pgcnt_t segkpsize = 0; 2787c478bd9Sstevel@tonic-gate #endif 279ad23a2dbSjohansen pgcnt_t segziosize = 0; /* size of zio segment in pages */ 2807c478bd9Sstevel@tonic-gate 281ae115bc7Smrj /* 282ae115bc7Smrj * VA range available to the debugger 283ae115bc7Smrj */ 284ae115bc7Smrj const caddr_t kdi_segdebugbase = (const caddr_t)SEGDEBUGBASE; 285ae115bc7Smrj const size_t kdi_segdebugsize = SEGDEBUGSIZE; 286ae115bc7Smrj 2877c478bd9Sstevel@tonic-gate struct memseg *memseg_base; 2887c478bd9Sstevel@tonic-gate struct vnode unused_pages_vp; 2897c478bd9Sstevel@tonic-gate 2907c478bd9Sstevel@tonic-gate #define FOURGB 0x100000000LL 2917c478bd9Sstevel@tonic-gate 2927c478bd9Sstevel@tonic-gate struct memlist *memlist; 2937c478bd9Sstevel@tonic-gate 2947c478bd9Sstevel@tonic-gate caddr_t s_text; /* start of kernel text segment */ 2957c478bd9Sstevel@tonic-gate caddr_t e_text; /* end of kernel text segment */ 2967c478bd9Sstevel@tonic-gate caddr_t s_data; /* start of kernel data segment */ 2977c478bd9Sstevel@tonic-gate caddr_t e_data; /* end of kernel data segment */ 2987c478bd9Sstevel@tonic-gate caddr_t modtext; /* start of loadable module text reserved */ 2997c478bd9Sstevel@tonic-gate caddr_t e_modtext; /* end of loadable module text reserved */ 3007c478bd9Sstevel@tonic-gate caddr_t moddata; /* start of loadable module data reserved */ 3017c478bd9Sstevel@tonic-gate caddr_t e_moddata; /* end of loadable module data reserved */ 3027c478bd9Sstevel@tonic-gate 3037c478bd9Sstevel@tonic-gate struct memlist *phys_install; /* Total installed physical memory */ 3047c478bd9Sstevel@tonic-gate struct memlist *phys_avail; /* Total available physical memory */ 3057c478bd9Sstevel@tonic-gate 3067c478bd9Sstevel@tonic-gate /* 3077c478bd9Sstevel@tonic-gate * kphysm_init returns the number of pages that were processed 3087c478bd9Sstevel@tonic-gate */ 309ae115bc7Smrj static pgcnt_t kphysm_init(page_t *, pgcnt_t); 3107c478bd9Sstevel@tonic-gate 3117c478bd9Sstevel@tonic-gate #define IO_PROP_SIZE 64 /* device property size */ 3127c478bd9Sstevel@tonic-gate 3137c478bd9Sstevel@tonic-gate /* 3147c478bd9Sstevel@tonic-gate * a couple useful roundup macros 3157c478bd9Sstevel@tonic-gate */ 3167c478bd9Sstevel@tonic-gate #define ROUND_UP_PAGE(x) \ 3177c478bd9Sstevel@tonic-gate ((uintptr_t)P2ROUNDUP((uintptr_t)(x), (uintptr_t)MMU_PAGESIZE)) 3187c478bd9Sstevel@tonic-gate #define ROUND_UP_LPAGE(x) \ 3197c478bd9Sstevel@tonic-gate ((uintptr_t)P2ROUNDUP((uintptr_t)(x), mmu.level_size[1])) 3207c478bd9Sstevel@tonic-gate #define ROUND_UP_4MEG(x) \ 321ae115bc7Smrj ((uintptr_t)P2ROUNDUP((uintptr_t)(x), (uintptr_t)FOUR_MEG)) 3227c478bd9Sstevel@tonic-gate #define ROUND_UP_TOPLEVEL(x) \ 3237c478bd9Sstevel@tonic-gate ((uintptr_t)P2ROUNDUP((uintptr_t)(x), mmu.level_size[mmu.max_level])) 3247c478bd9Sstevel@tonic-gate 3257c478bd9Sstevel@tonic-gate /* 3267c478bd9Sstevel@tonic-gate * 32-bit Kernel's Virtual memory layout. 3277c478bd9Sstevel@tonic-gate * +-----------------------+ 328ae115bc7Smrj * | | 3297c478bd9Sstevel@tonic-gate * 0xFFC00000 -|-----------------------|- ARGSBASE 3307c478bd9Sstevel@tonic-gate * | debugger | 3317c478bd9Sstevel@tonic-gate * 0xFF800000 -|-----------------------|- SEGDEBUGBASE 3327c478bd9Sstevel@tonic-gate * | Kernel Data | 3337c478bd9Sstevel@tonic-gate * 0xFEC00000 -|-----------------------| 3347c478bd9Sstevel@tonic-gate * | Kernel Text | 3357c478bd9Sstevel@tonic-gate * 0xFE800000 -|-----------------------|- KERNEL_TEXT 336ae115bc7Smrj * |--- GDT ---|- GDT page (GDT_VA) 337ae115bc7Smrj * |--- debug info ---|- debug info (DEBUG_INFO_VA) 338ae115bc7Smrj * | | 339ae115bc7Smrj * | page_t structures | 3407c478bd9Sstevel@tonic-gate * | memsegs, memlists, | 3417c478bd9Sstevel@tonic-gate * | page hash, etc. | 342ae115bc7Smrj * --- -|-----------------------|- ekernelheap, valloc_base (floating) 343ae115bc7Smrj * | | (segkp is just an arena in the heap) 3447c478bd9Sstevel@tonic-gate * | | 3457c478bd9Sstevel@tonic-gate * | kvseg | 3467c478bd9Sstevel@tonic-gate * | | 3477c478bd9Sstevel@tonic-gate * | | 3487c478bd9Sstevel@tonic-gate * --- -|-----------------------|- kernelheap (floating) 3497c478bd9Sstevel@tonic-gate * | Segkmap | 350ae115bc7Smrj * 0xC3002000 -|-----------------------|- segmap_start (floating) 3517c478bd9Sstevel@tonic-gate * | Red Zone | 3527c478bd9Sstevel@tonic-gate * 0xC3000000 -|-----------------------|- kernelbase / userlimit (floating) 3537c478bd9Sstevel@tonic-gate * | | || 3547c478bd9Sstevel@tonic-gate * | Shared objects | \/ 3557c478bd9Sstevel@tonic-gate * | | 3567c478bd9Sstevel@tonic-gate * : : 3577c478bd9Sstevel@tonic-gate * | user data | 3587c478bd9Sstevel@tonic-gate * |-----------------------| 3597c478bd9Sstevel@tonic-gate * | user text | 3607c478bd9Sstevel@tonic-gate * 0x08048000 -|-----------------------| 3617c478bd9Sstevel@tonic-gate * | user stack | 3627c478bd9Sstevel@tonic-gate * : : 3637c478bd9Sstevel@tonic-gate * | invalid | 3647c478bd9Sstevel@tonic-gate * 0x00000000 +-----------------------+ 3657c478bd9Sstevel@tonic-gate * 3667c478bd9Sstevel@tonic-gate * 3677c478bd9Sstevel@tonic-gate * 64-bit Kernel's Virtual memory layout. (assuming 64 bit app) 3687c478bd9Sstevel@tonic-gate * +-----------------------+ 369ae115bc7Smrj * | | 3707c478bd9Sstevel@tonic-gate * 0xFFFFFFFF.FFC00000 |-----------------------|- ARGSBASE 3717c478bd9Sstevel@tonic-gate * | debugger (?) | 3727c478bd9Sstevel@tonic-gate * 0xFFFFFFFF.FF800000 |-----------------------|- SEGDEBUGBASE 3737c478bd9Sstevel@tonic-gate * | unused | 3747c478bd9Sstevel@tonic-gate * +-----------------------+ 3757c478bd9Sstevel@tonic-gate * | Kernel Data | 3767c478bd9Sstevel@tonic-gate * 0xFFFFFFFF.FBC00000 |-----------------------| 3777c478bd9Sstevel@tonic-gate * | Kernel Text | 3787c478bd9Sstevel@tonic-gate * 0xFFFFFFFF.FB800000 |-----------------------|- KERNEL_TEXT 379ae115bc7Smrj * |--- GDT ---|- GDT page (GDT_VA) 380ae115bc7Smrj * |--- debug info ---|- debug info (DEBUG_INFO_VA) 381ae115bc7Smrj * | | 3827c478bd9Sstevel@tonic-gate * | Core heap | (used for loadable modules) 3837c478bd9Sstevel@tonic-gate * 0xFFFFFFFF.C0000000 |-----------------------|- core_base / ekernelheap 3847c478bd9Sstevel@tonic-gate * | Kernel | 3857c478bd9Sstevel@tonic-gate * | heap | 3867c478bd9Sstevel@tonic-gate * 0xFFFFFXXX.XXX00000 |-----------------------|- kernelheap (floating) 387ae115bc7Smrj * | segmap | 388ae115bc7Smrj * 0xFFFFFXXX.XXX00000 |-----------------------|- segmap_start (floating) 3897c478bd9Sstevel@tonic-gate * | device mappings | 3907c478bd9Sstevel@tonic-gate * 0xFFFFFXXX.XXX00000 |-----------------------|- toxic_addr (floating) 391ad23a2dbSjohansen * | segzio | 392ad23a2dbSjohansen * 0xFFFFFXXX.XXX00000 |-----------------------|- segzio_base (floating) 3937c478bd9Sstevel@tonic-gate * | segkp | 394ae115bc7Smrj * --- |-----------------------|- segkp_base (floating) 395ae115bc7Smrj * | page_t structures | valloc_base + valloc_sz 396ae115bc7Smrj * | memsegs, memlists, | 397ae115bc7Smrj * | page hash, etc. | 398*35b1ab99Sjosephb * 0xFFFFFF00.00000000 |-----------------------|- valloc_base (lower if > 1TB) 3997c478bd9Sstevel@tonic-gate * | segkpm | 4007c478bd9Sstevel@tonic-gate * 0xFFFFFE00.00000000 |-----------------------| 4017c478bd9Sstevel@tonic-gate * | Red Zone | 402*35b1ab99Sjosephb * 0xFFFFFD80.00000000 |-----------------------|- KERNELBASE (lower if > 1TB) 4037c478bd9Sstevel@tonic-gate * | User stack |- User space memory 4047c478bd9Sstevel@tonic-gate * | | 4057c478bd9Sstevel@tonic-gate * | shared objects, etc | (grows downwards) 4067c478bd9Sstevel@tonic-gate * : : 4077c478bd9Sstevel@tonic-gate * | | 4087c478bd9Sstevel@tonic-gate * 0xFFFF8000.00000000 |-----------------------| 4097c478bd9Sstevel@tonic-gate * | | 4107c478bd9Sstevel@tonic-gate * | VA Hole / unused | 4117c478bd9Sstevel@tonic-gate * | | 4127c478bd9Sstevel@tonic-gate * 0x00008000.00000000 |-----------------------| 4137c478bd9Sstevel@tonic-gate * | | 4147c478bd9Sstevel@tonic-gate * | | 4157c478bd9Sstevel@tonic-gate * : : 4167c478bd9Sstevel@tonic-gate * | user heap | (grows upwards) 4177c478bd9Sstevel@tonic-gate * | | 4187c478bd9Sstevel@tonic-gate * | user data | 4197c478bd9Sstevel@tonic-gate * |-----------------------| 4207c478bd9Sstevel@tonic-gate * | user text | 4217c478bd9Sstevel@tonic-gate * 0x00000000.04000000 |-----------------------| 4227c478bd9Sstevel@tonic-gate * | invalid | 4237c478bd9Sstevel@tonic-gate * 0x00000000.00000000 +-----------------------+ 4247c478bd9Sstevel@tonic-gate * 4257c478bd9Sstevel@tonic-gate * A 32 bit app on the 64 bit kernel sees the same layout as on the 32 bit 4267c478bd9Sstevel@tonic-gate * kernel, except that userlimit is raised to 0xfe000000 4277c478bd9Sstevel@tonic-gate * 4287c478bd9Sstevel@tonic-gate * Floating values: 4297c478bd9Sstevel@tonic-gate * 4307c478bd9Sstevel@tonic-gate * valloc_base: start of the kernel's memory management/tracking data 431ae115bc7Smrj * structures. This region contains page_t structures for 432ae115bc7Smrj * physical memory, memsegs, memlists, and the page hash. 4337c478bd9Sstevel@tonic-gate * 4347c478bd9Sstevel@tonic-gate * core_base: start of the kernel's "core" heap area on 64-bit systems. 4357c478bd9Sstevel@tonic-gate * This area is intended to be used for global data as well as for module 4367c478bd9Sstevel@tonic-gate * text/data that does not fit into the nucleus pages. The core heap is 4377c478bd9Sstevel@tonic-gate * restricted to a 2GB range, allowing every address within it to be 4387c478bd9Sstevel@tonic-gate * accessed using rip-relative addressing 4397c478bd9Sstevel@tonic-gate * 4407c478bd9Sstevel@tonic-gate * ekernelheap: end of kernelheap and start of segmap. 4417c478bd9Sstevel@tonic-gate * 4427c478bd9Sstevel@tonic-gate * kernelheap: start of kernel heap. On 32-bit systems, this starts right 4437c478bd9Sstevel@tonic-gate * above a red zone that separates the user's address space from the 4447c478bd9Sstevel@tonic-gate * kernel's. On 64-bit systems, it sits above segkp and segkpm. 4457c478bd9Sstevel@tonic-gate * 446ae115bc7Smrj * segmap_start: start of segmap. The length of segmap can be modified 4477c478bd9Sstevel@tonic-gate * by changing segmapsize in /etc/system (preferred) or eeprom (deprecated). 4487c478bd9Sstevel@tonic-gate * The default length is 16MB on 32-bit systems and 64MB on 64-bit systems. 4497c478bd9Sstevel@tonic-gate * 4507c478bd9Sstevel@tonic-gate * kernelbase: On a 32-bit kernel the default value of 0xd4000000 will be 4517c478bd9Sstevel@tonic-gate * decreased by 2X the size required for page_t. This allows the kernel 4527c478bd9Sstevel@tonic-gate * heap to grow in size with physical memory. With sizeof(page_t) == 80 4537c478bd9Sstevel@tonic-gate * bytes, the following shows the values of kernelbase and kernel heap 4547c478bd9Sstevel@tonic-gate * sizes for different memory configurations (assuming default segmap and 4557c478bd9Sstevel@tonic-gate * segkp sizes). 4567c478bd9Sstevel@tonic-gate * 4577c478bd9Sstevel@tonic-gate * mem size for kernelbase kernel heap 4587c478bd9Sstevel@tonic-gate * size page_t's size 4597c478bd9Sstevel@tonic-gate * ---- --------- ---------- ----------- 4607c478bd9Sstevel@tonic-gate * 1gb 0x01400000 0xd1800000 684MB 4617c478bd9Sstevel@tonic-gate * 2gb 0x02800000 0xcf000000 704MB 4627c478bd9Sstevel@tonic-gate * 4gb 0x05000000 0xca000000 744MB 4637c478bd9Sstevel@tonic-gate * 6gb 0x07800000 0xc5000000 784MB 4647c478bd9Sstevel@tonic-gate * 8gb 0x0a000000 0xc0000000 824MB 4657c478bd9Sstevel@tonic-gate * 16gb 0x14000000 0xac000000 984MB 4667c478bd9Sstevel@tonic-gate * 32gb 0x28000000 0x84000000 1304MB 4677c478bd9Sstevel@tonic-gate * 64gb 0x50000000 0x34000000 1944MB (*) 4687c478bd9Sstevel@tonic-gate * 4697c478bd9Sstevel@tonic-gate * kernelbase is less than the abi minimum of 0xc0000000 for memory 4707c478bd9Sstevel@tonic-gate * configurations above 8gb. 4717c478bd9Sstevel@tonic-gate * 4727c478bd9Sstevel@tonic-gate * (*) support for memory configurations above 32gb will require manual tuning 4737c478bd9Sstevel@tonic-gate * of kernelbase to balance out the need of user applications. 4747c478bd9Sstevel@tonic-gate */ 4757c478bd9Sstevel@tonic-gate 4767c478bd9Sstevel@tonic-gate /* real-time-clock initialization parameters */ 477ae115bc7Smrj extern time_t process_rtc_config_file(void); 4787c478bd9Sstevel@tonic-gate 4797c478bd9Sstevel@tonic-gate uintptr_t kernelbase; 480ae115bc7Smrj uintptr_t postbootkernelbase; /* not set till boot loader is gone */ 4817c478bd9Sstevel@tonic-gate uintptr_t eprom_kernelbase; 4827c478bd9Sstevel@tonic-gate size_t segmapsize; 483ae115bc7Smrj uintptr_t segmap_start; 4847c478bd9Sstevel@tonic-gate int segmapfreelists; 4857c478bd9Sstevel@tonic-gate pgcnt_t npages; 486ae115bc7Smrj pgcnt_t orig_npages; 4877c478bd9Sstevel@tonic-gate size_t core_size; /* size of "core" heap */ 4887c478bd9Sstevel@tonic-gate uintptr_t core_base; /* base address of "core" heap */ 4897c478bd9Sstevel@tonic-gate 4907c478bd9Sstevel@tonic-gate /* 4917c478bd9Sstevel@tonic-gate * List of bootstrap pages. We mark these as allocated in startup. 4927c478bd9Sstevel@tonic-gate * release_bootstrap() will free them when we're completely done with 4937c478bd9Sstevel@tonic-gate * the bootstrap. 4947c478bd9Sstevel@tonic-gate */ 495ae115bc7Smrj static page_t *bootpages; 496ae115bc7Smrj 497ae115bc7Smrj /* 498ae115bc7Smrj * boot time pages that have a vnode from the ramdisk will keep that forever. 499ae115bc7Smrj */ 500ae115bc7Smrj static page_t *rd_pages; 5017c478bd9Sstevel@tonic-gate 5027c478bd9Sstevel@tonic-gate struct system_hardware system_hardware; 5037c478bd9Sstevel@tonic-gate 5047c478bd9Sstevel@tonic-gate /* 5057c478bd9Sstevel@tonic-gate * Enable some debugging messages concerning memory usage... 5067c478bd9Sstevel@tonic-gate */ 5077c478bd9Sstevel@tonic-gate static void 508ae115bc7Smrj print_memlist(char *title, struct memlist *mp) 5097c478bd9Sstevel@tonic-gate { 5107c478bd9Sstevel@tonic-gate prom_printf("MEMLIST: %s:\n", title); 5117c478bd9Sstevel@tonic-gate while (mp != NULL) { 5127c478bd9Sstevel@tonic-gate prom_printf("\tAddress 0x%" PRIx64 ", size 0x%" PRIx64 "\n", 5137c478bd9Sstevel@tonic-gate mp->address, mp->size); 5147c478bd9Sstevel@tonic-gate mp = mp->next; 5157c478bd9Sstevel@tonic-gate } 5167c478bd9Sstevel@tonic-gate } 5177c478bd9Sstevel@tonic-gate 5187c478bd9Sstevel@tonic-gate /* 5197c478bd9Sstevel@tonic-gate * XX64 need a comment here.. are these just default values, surely 5207c478bd9Sstevel@tonic-gate * we read the "cpuid" type information to figure this out. 5217c478bd9Sstevel@tonic-gate */ 5227c478bd9Sstevel@tonic-gate int l2cache_sz = 0x80000; 5237c478bd9Sstevel@tonic-gate int l2cache_linesz = 0x40; 5247c478bd9Sstevel@tonic-gate int l2cache_assoc = 1; 5257c478bd9Sstevel@tonic-gate 5262cb27123Saguzovsk static size_t textrepl_min_gb = 10; 5272cb27123Saguzovsk 5287c478bd9Sstevel@tonic-gate /* 5297c478bd9Sstevel@tonic-gate * on 64 bit we use a predifined VA range for mapping devices in the kernel 5307c478bd9Sstevel@tonic-gate * on 32 bit the mappings are intermixed in the heap, so we use a bit map 5317c478bd9Sstevel@tonic-gate */ 5327c478bd9Sstevel@tonic-gate #ifdef __amd64 5337c478bd9Sstevel@tonic-gate 5347c478bd9Sstevel@tonic-gate vmem_t *device_arena; 5357c478bd9Sstevel@tonic-gate uintptr_t toxic_addr = (uintptr_t)NULL; 536ae115bc7Smrj size_t toxic_size = 1024 * 1024 * 1024; /* Sparc uses 1 gig too */ 5377c478bd9Sstevel@tonic-gate 5387c478bd9Sstevel@tonic-gate #else /* __i386 */ 5397c478bd9Sstevel@tonic-gate 5407c478bd9Sstevel@tonic-gate ulong_t *toxic_bit_map; /* one bit for each 4k of VA in heap_arena */ 5417c478bd9Sstevel@tonic-gate size_t toxic_bit_map_len = 0; /* in bits */ 5427c478bd9Sstevel@tonic-gate 5437c478bd9Sstevel@tonic-gate #endif /* __i386 */ 5447c478bd9Sstevel@tonic-gate 5457c478bd9Sstevel@tonic-gate /* 5467c478bd9Sstevel@tonic-gate * Simple boot time debug facilities 5477c478bd9Sstevel@tonic-gate */ 5487c478bd9Sstevel@tonic-gate static char *prm_dbg_str[] = { 5497c478bd9Sstevel@tonic-gate "%s:%d: '%s' is 0x%x\n", 5507c478bd9Sstevel@tonic-gate "%s:%d: '%s' is 0x%llx\n" 5517c478bd9Sstevel@tonic-gate }; 5527c478bd9Sstevel@tonic-gate 5537c478bd9Sstevel@tonic-gate int prom_debug; 5547c478bd9Sstevel@tonic-gate 5557c478bd9Sstevel@tonic-gate #define PRM_DEBUG(q) if (prom_debug) \ 5567c478bd9Sstevel@tonic-gate prom_printf(prm_dbg_str[sizeof (q) >> 3], "startup.c", __LINE__, #q, q); 5577c478bd9Sstevel@tonic-gate #define PRM_POINT(q) if (prom_debug) \ 5587c478bd9Sstevel@tonic-gate prom_printf("%s:%d: %s\n", "startup.c", __LINE__, q); 5597c478bd9Sstevel@tonic-gate 5607c478bd9Sstevel@tonic-gate /* 5617c478bd9Sstevel@tonic-gate * This structure is used to keep track of the intial allocations 5627c478bd9Sstevel@tonic-gate * done in startup_memlist(). The value of NUM_ALLOCATIONS needs to 5637c478bd9Sstevel@tonic-gate * be >= the number of ADD_TO_ALLOCATIONS() executed in the code. 5647c478bd9Sstevel@tonic-gate */ 5657c478bd9Sstevel@tonic-gate #define NUM_ALLOCATIONS 7 5667c478bd9Sstevel@tonic-gate int num_allocations = 0; 5677c478bd9Sstevel@tonic-gate struct { 5687c478bd9Sstevel@tonic-gate void **al_ptr; 5697c478bd9Sstevel@tonic-gate size_t al_size; 5707c478bd9Sstevel@tonic-gate } allocations[NUM_ALLOCATIONS]; 5717c478bd9Sstevel@tonic-gate size_t valloc_sz = 0; 5727c478bd9Sstevel@tonic-gate uintptr_t valloc_base; 5737c478bd9Sstevel@tonic-gate 5747c478bd9Sstevel@tonic-gate #define ADD_TO_ALLOCATIONS(ptr, size) { \ 5757c478bd9Sstevel@tonic-gate size = ROUND_UP_PAGE(size); \ 5767c478bd9Sstevel@tonic-gate if (num_allocations == NUM_ALLOCATIONS) \ 5777c478bd9Sstevel@tonic-gate panic("too many ADD_TO_ALLOCATIONS()"); \ 5787c478bd9Sstevel@tonic-gate allocations[num_allocations].al_ptr = (void**)&ptr; \ 5797c478bd9Sstevel@tonic-gate allocations[num_allocations].al_size = size; \ 5807c478bd9Sstevel@tonic-gate valloc_sz += size; \ 5817c478bd9Sstevel@tonic-gate ++num_allocations; \ 5827c478bd9Sstevel@tonic-gate } 5837c478bd9Sstevel@tonic-gate 584ae115bc7Smrj /* 585ae115bc7Smrj * Allocate all the initial memory needed by the page allocator. 586ae115bc7Smrj */ 5877c478bd9Sstevel@tonic-gate static void 5887c478bd9Sstevel@tonic-gate perform_allocations(void) 5897c478bd9Sstevel@tonic-gate { 5907c478bd9Sstevel@tonic-gate caddr_t mem; 5917c478bd9Sstevel@tonic-gate int i; 592ae115bc7Smrj int valloc_align; 5937c478bd9Sstevel@tonic-gate 594ae115bc7Smrj PRM_DEBUG(valloc_base); 595ae115bc7Smrj PRM_DEBUG(valloc_sz); 596ae115bc7Smrj valloc_align = mmu.level_size[mmu.max_page_level > 0]; 597ae115bc7Smrj mem = BOP_ALLOC(bootops, (caddr_t)valloc_base, valloc_sz, valloc_align); 5987c478bd9Sstevel@tonic-gate if (mem != (caddr_t)valloc_base) 5997c478bd9Sstevel@tonic-gate panic("BOP_ALLOC() failed"); 6007c478bd9Sstevel@tonic-gate bzero(mem, valloc_sz); 6017c478bd9Sstevel@tonic-gate for (i = 0; i < num_allocations; ++i) { 6027c478bd9Sstevel@tonic-gate *allocations[i].al_ptr = (void *)mem; 6037c478bd9Sstevel@tonic-gate mem += allocations[i].al_size; 6047c478bd9Sstevel@tonic-gate } 6057c478bd9Sstevel@tonic-gate } 6067c478bd9Sstevel@tonic-gate 6077c478bd9Sstevel@tonic-gate /* 6087c478bd9Sstevel@tonic-gate * Our world looks like this at startup time. 6097c478bd9Sstevel@tonic-gate * 6107c478bd9Sstevel@tonic-gate * In a 32-bit OS, boot loads the kernel text at 0xfe800000 and kernel data 6117c478bd9Sstevel@tonic-gate * at 0xfec00000. On a 64-bit OS, kernel text and data are loaded at 6127c478bd9Sstevel@tonic-gate * 0xffffffff.fe800000 and 0xffffffff.fec00000 respectively. Those 6137c478bd9Sstevel@tonic-gate * addresses are fixed in the binary at link time. 6147c478bd9Sstevel@tonic-gate * 6157c478bd9Sstevel@tonic-gate * On the text page: 6167c478bd9Sstevel@tonic-gate * unix/genunix/krtld/module text loads. 6177c478bd9Sstevel@tonic-gate * 6187c478bd9Sstevel@tonic-gate * On the data page: 619ae115bc7Smrj * unix/genunix/krtld/module data loads. 620ae115bc7Smrj * 6217c478bd9Sstevel@tonic-gate * Machine-dependent startup code 6227c478bd9Sstevel@tonic-gate */ 6237c478bd9Sstevel@tonic-gate void 6247c478bd9Sstevel@tonic-gate startup(void) 6257c478bd9Sstevel@tonic-gate { 62675bcd456Sjg extern void startup_bios_disk(void); 62775bcd456Sjg extern void startup_pci_bios(void); 6287c478bd9Sstevel@tonic-gate #if defined(__amd64) 629*35b1ab99Sjosephb kpm_desired = 1; 6307c478bd9Sstevel@tonic-gate #endif 6317c478bd9Sstevel@tonic-gate kpm_enable = 0; 6327c478bd9Sstevel@tonic-gate 6337c478bd9Sstevel@tonic-gate progressbar_init(); 6347c478bd9Sstevel@tonic-gate startup_init(); 6357c478bd9Sstevel@tonic-gate startup_memlist(); 636ae115bc7Smrj startup_kmem(); 637*35b1ab99Sjosephb startup_vm(); 63875bcd456Sjg startup_pci_bios(); 6397c478bd9Sstevel@tonic-gate startup_modules(); 6407c478bd9Sstevel@tonic-gate startup_bios_disk(); 6417c478bd9Sstevel@tonic-gate startup_end(); 6427c478bd9Sstevel@tonic-gate progressbar_start(); 6437c478bd9Sstevel@tonic-gate } 6447c478bd9Sstevel@tonic-gate 6457c478bd9Sstevel@tonic-gate static void 6467c478bd9Sstevel@tonic-gate startup_init() 6477c478bd9Sstevel@tonic-gate { 6487c478bd9Sstevel@tonic-gate PRM_POINT("startup_init() starting..."); 6497c478bd9Sstevel@tonic-gate 6507c478bd9Sstevel@tonic-gate /* 6517c478bd9Sstevel@tonic-gate * Complete the extraction of cpuid data 6527c478bd9Sstevel@tonic-gate */ 6537c478bd9Sstevel@tonic-gate cpuid_pass2(CPU); 6547c478bd9Sstevel@tonic-gate 6557c478bd9Sstevel@tonic-gate (void) check_boot_version(BOP_GETVERSION(bootops)); 6567c478bd9Sstevel@tonic-gate 6577c478bd9Sstevel@tonic-gate /* 6587c478bd9Sstevel@tonic-gate * Check for prom_debug in boot environment 6597c478bd9Sstevel@tonic-gate */ 6607c478bd9Sstevel@tonic-gate if (BOP_GETPROPLEN(bootops, "prom_debug") >= 0) { 6617c478bd9Sstevel@tonic-gate ++prom_debug; 6627c478bd9Sstevel@tonic-gate PRM_POINT("prom_debug found in boot enviroment"); 6637c478bd9Sstevel@tonic-gate } 6647c478bd9Sstevel@tonic-gate 6657c478bd9Sstevel@tonic-gate /* 6667c478bd9Sstevel@tonic-gate * Collect node, cpu and memory configuration information. 6677c478bd9Sstevel@tonic-gate */ 6687c478bd9Sstevel@tonic-gate get_system_configuration(); 6697c478bd9Sstevel@tonic-gate 6707c478bd9Sstevel@tonic-gate /* 6717c478bd9Sstevel@tonic-gate * Halt if this is an unsupported processor. 6727c478bd9Sstevel@tonic-gate */ 6737c478bd9Sstevel@tonic-gate if (x86_type == X86_TYPE_486 || x86_type == X86_TYPE_CYRIX_486) { 6747c478bd9Sstevel@tonic-gate printf("\n486 processor (\"%s\") detected.\n", 6757c478bd9Sstevel@tonic-gate CPU->cpu_brandstr); 6767c478bd9Sstevel@tonic-gate halt("This processor is not supported by this release " 6777c478bd9Sstevel@tonic-gate "of Solaris."); 6787c478bd9Sstevel@tonic-gate } 6797c478bd9Sstevel@tonic-gate 6807c478bd9Sstevel@tonic-gate PRM_POINT("startup_init() done"); 6817c478bd9Sstevel@tonic-gate } 6827c478bd9Sstevel@tonic-gate 6837c478bd9Sstevel@tonic-gate /* 6847c478bd9Sstevel@tonic-gate * Callback for copy_memlist_filter() to filter nucleus, kadb/kmdb, (ie. 6857c478bd9Sstevel@tonic-gate * everything mapped above KERNEL_TEXT) pages from phys_avail. Note it 6867c478bd9Sstevel@tonic-gate * also filters out physical page zero. There is some reliance on the 6877c478bd9Sstevel@tonic-gate * boot loader allocating only a few contiguous physical memory chunks. 6887c478bd9Sstevel@tonic-gate */ 6897c478bd9Sstevel@tonic-gate static void 6907c478bd9Sstevel@tonic-gate avail_filter(uint64_t *addr, uint64_t *size) 6917c478bd9Sstevel@tonic-gate { 6927c478bd9Sstevel@tonic-gate uintptr_t va; 6937c478bd9Sstevel@tonic-gate uintptr_t next_va; 6947c478bd9Sstevel@tonic-gate pfn_t pfn; 6957c478bd9Sstevel@tonic-gate uint64_t pfn_addr; 6967c478bd9Sstevel@tonic-gate uint64_t pfn_eaddr; 6977c478bd9Sstevel@tonic-gate uint_t prot; 6987c478bd9Sstevel@tonic-gate size_t len; 6997c478bd9Sstevel@tonic-gate uint_t change; 7007c478bd9Sstevel@tonic-gate 7017c478bd9Sstevel@tonic-gate if (prom_debug) 7027c478bd9Sstevel@tonic-gate prom_printf("\tFilter: in: a=%" PRIx64 ", s=%" PRIx64 "\n", 7037c478bd9Sstevel@tonic-gate *addr, *size); 7047c478bd9Sstevel@tonic-gate 7057c478bd9Sstevel@tonic-gate /* 7067c478bd9Sstevel@tonic-gate * page zero is required for BIOS.. never make it available 7077c478bd9Sstevel@tonic-gate */ 7087c478bd9Sstevel@tonic-gate if (*addr == 0) { 7097c478bd9Sstevel@tonic-gate *addr += MMU_PAGESIZE; 7107c478bd9Sstevel@tonic-gate *size -= MMU_PAGESIZE; 7117c478bd9Sstevel@tonic-gate } 7127c478bd9Sstevel@tonic-gate 7137c478bd9Sstevel@tonic-gate /* 714ae115bc7Smrj * First we trim from the front of the range. Since kbm_probe() 7157c478bd9Sstevel@tonic-gate * walks ranges in virtual order, but addr/size are physical, we need 7167c478bd9Sstevel@tonic-gate * to the list until no changes are seen. This deals with the case 7177c478bd9Sstevel@tonic-gate * where page "p" is mapped at v, page "p + PAGESIZE" is mapped at w 7187c478bd9Sstevel@tonic-gate * but w < v. 7197c478bd9Sstevel@tonic-gate */ 7207c478bd9Sstevel@tonic-gate do { 7217c478bd9Sstevel@tonic-gate change = 0; 7227c478bd9Sstevel@tonic-gate for (va = KERNEL_TEXT; 723ae115bc7Smrj *size > 0 && kbm_probe(&va, &len, &pfn, &prot) != 0; 7247c478bd9Sstevel@tonic-gate va = next_va) { 7257c478bd9Sstevel@tonic-gate 7267c478bd9Sstevel@tonic-gate next_va = va + len; 727ae115bc7Smrj pfn_addr = pfn_to_pa(pfn); 7287c478bd9Sstevel@tonic-gate pfn_eaddr = pfn_addr + len; 7297c478bd9Sstevel@tonic-gate 7307c478bd9Sstevel@tonic-gate if (pfn_addr <= *addr && pfn_eaddr > *addr) { 7317c478bd9Sstevel@tonic-gate change = 1; 7327c478bd9Sstevel@tonic-gate while (*size > 0 && len > 0) { 7337c478bd9Sstevel@tonic-gate *addr += MMU_PAGESIZE; 7347c478bd9Sstevel@tonic-gate *size -= MMU_PAGESIZE; 7357c478bd9Sstevel@tonic-gate len -= MMU_PAGESIZE; 7367c478bd9Sstevel@tonic-gate } 7377c478bd9Sstevel@tonic-gate } 7387c478bd9Sstevel@tonic-gate } 7397c478bd9Sstevel@tonic-gate if (change && prom_debug) 7407c478bd9Sstevel@tonic-gate prom_printf("\t\ttrim: a=%" PRIx64 ", s=%" PRIx64 "\n", 7417c478bd9Sstevel@tonic-gate *addr, *size); 7427c478bd9Sstevel@tonic-gate } while (change); 7437c478bd9Sstevel@tonic-gate 7447c478bd9Sstevel@tonic-gate /* 7457c478bd9Sstevel@tonic-gate * Trim pages from the end of the range. 7467c478bd9Sstevel@tonic-gate */ 7477c478bd9Sstevel@tonic-gate for (va = KERNEL_TEXT; 748ae115bc7Smrj *size > 0 && kbm_probe(&va, &len, &pfn, &prot) != 0; 7497c478bd9Sstevel@tonic-gate va = next_va) { 7507c478bd9Sstevel@tonic-gate 7517c478bd9Sstevel@tonic-gate next_va = va + len; 752ae115bc7Smrj pfn_addr = pfn_to_pa(pfn); 7537c478bd9Sstevel@tonic-gate 7547c478bd9Sstevel@tonic-gate if (pfn_addr >= *addr && pfn_addr < *addr + *size) 7557c478bd9Sstevel@tonic-gate *size = pfn_addr - *addr; 7567c478bd9Sstevel@tonic-gate } 7577c478bd9Sstevel@tonic-gate 7587c478bd9Sstevel@tonic-gate if (prom_debug) 7597c478bd9Sstevel@tonic-gate prom_printf("\tFilter out: a=%" PRIx64 ", s=%" PRIx64 "\n", 7607c478bd9Sstevel@tonic-gate *addr, *size); 7617c478bd9Sstevel@tonic-gate } 7627c478bd9Sstevel@tonic-gate 7637c478bd9Sstevel@tonic-gate static void 7647c478bd9Sstevel@tonic-gate kpm_init() 7657c478bd9Sstevel@tonic-gate { 7667c478bd9Sstevel@tonic-gate struct segkpm_crargs b; 7677c478bd9Sstevel@tonic-gate 7687c478bd9Sstevel@tonic-gate /* 7697c478bd9Sstevel@tonic-gate * These variables were all designed for sfmmu in which segkpm is 7707c478bd9Sstevel@tonic-gate * mapped using a single pagesize - either 8KB or 4MB. On x86, we 7717c478bd9Sstevel@tonic-gate * might use 2+ page sizes on a single machine, so none of these 7727c478bd9Sstevel@tonic-gate * variables have a single correct value. They are set up as if we 7737c478bd9Sstevel@tonic-gate * always use a 4KB pagesize, which should do no harm. In the long 7747c478bd9Sstevel@tonic-gate * run, we should get rid of KPM's assumption that only a single 7757c478bd9Sstevel@tonic-gate * pagesize is used. 7767c478bd9Sstevel@tonic-gate */ 7777c478bd9Sstevel@tonic-gate kpm_pgshft = MMU_PAGESHIFT; 7787c478bd9Sstevel@tonic-gate kpm_pgsz = MMU_PAGESIZE; 7797c478bd9Sstevel@tonic-gate kpm_pgoff = MMU_PAGEOFFSET; 7807c478bd9Sstevel@tonic-gate kpmp2pshft = 0; 7817c478bd9Sstevel@tonic-gate kpmpnpgs = 1; 7827c478bd9Sstevel@tonic-gate ASSERT(((uintptr_t)kpm_vbase & (kpm_pgsz - 1)) == 0); 7837c478bd9Sstevel@tonic-gate 7847c478bd9Sstevel@tonic-gate PRM_POINT("about to create segkpm"); 7857c478bd9Sstevel@tonic-gate rw_enter(&kas.a_lock, RW_WRITER); 7867c478bd9Sstevel@tonic-gate 7877c478bd9Sstevel@tonic-gate if (seg_attach(&kas, kpm_vbase, kpm_size, segkpm) < 0) 7887c478bd9Sstevel@tonic-gate panic("cannot attach segkpm"); 7897c478bd9Sstevel@tonic-gate 7907c478bd9Sstevel@tonic-gate b.prot = PROT_READ | PROT_WRITE; 7917c478bd9Sstevel@tonic-gate b.nvcolors = 1; 7927c478bd9Sstevel@tonic-gate 7937c478bd9Sstevel@tonic-gate if (segkpm_create(segkpm, (caddr_t)&b) != 0) 7947c478bd9Sstevel@tonic-gate panic("segkpm_create segkpm"); 7957c478bd9Sstevel@tonic-gate 7967c478bd9Sstevel@tonic-gate rw_exit(&kas.a_lock); 797ae115bc7Smrj } 7987c478bd9Sstevel@tonic-gate 799ae115bc7Smrj /* 800ae115bc7Smrj * The debug info page provides enough information to allow external 801ae115bc7Smrj * inspectors (e.g. when running under a hypervisor) to bootstrap 802ae115bc7Smrj * themselves into allowing full-blown kernel debugging. 803ae115bc7Smrj */ 804ae115bc7Smrj static void 805ae115bc7Smrj init_debug_info(void) 806ae115bc7Smrj { 807ae115bc7Smrj caddr_t mem; 808ae115bc7Smrj debug_info_t *di; 809ae115bc7Smrj 810ae115bc7Smrj #ifndef __lint 811ae115bc7Smrj ASSERT(sizeof (debug_info_t) < MMU_PAGESIZE); 812ae115bc7Smrj #endif 813ae115bc7Smrj 814ae115bc7Smrj mem = BOP_ALLOC(bootops, (caddr_t)DEBUG_INFO_VA, MMU_PAGESIZE, 815ae115bc7Smrj MMU_PAGESIZE); 816ae115bc7Smrj 817ae115bc7Smrj if (mem != (caddr_t)DEBUG_INFO_VA) 818ae115bc7Smrj panic("BOP_ALLOC() failed"); 819ae115bc7Smrj bzero(mem, MMU_PAGESIZE); 820ae115bc7Smrj 821ae115bc7Smrj di = (debug_info_t *)mem; 822ae115bc7Smrj 823ae115bc7Smrj di->di_magic = DEBUG_INFO_MAGIC; 824ae115bc7Smrj di->di_version = DEBUG_INFO_VERSION; 8257c478bd9Sstevel@tonic-gate } 8267c478bd9Sstevel@tonic-gate 8277c478bd9Sstevel@tonic-gate /* 828ae115bc7Smrj * Build the memlists and other kernel essential memory system data structures. 829ae115bc7Smrj * This is everything at valloc_base. 8307c478bd9Sstevel@tonic-gate */ 8317c478bd9Sstevel@tonic-gate static void 8327c478bd9Sstevel@tonic-gate startup_memlist(void) 8337c478bd9Sstevel@tonic-gate { 8347c478bd9Sstevel@tonic-gate size_t memlist_sz; 8357c478bd9Sstevel@tonic-gate size_t memseg_sz; 8367c478bd9Sstevel@tonic-gate size_t pagehash_sz; 8377c478bd9Sstevel@tonic-gate size_t pp_sz; 8387c478bd9Sstevel@tonic-gate uintptr_t va; 8397c478bd9Sstevel@tonic-gate size_t len; 8407c478bd9Sstevel@tonic-gate uint_t prot; 8417c478bd9Sstevel@tonic-gate pfn_t pfn; 8427c478bd9Sstevel@tonic-gate int memblocks; 8437c478bd9Sstevel@tonic-gate caddr_t pagecolor_mem; 8447c478bd9Sstevel@tonic-gate size_t pagecolor_memsz; 8457c478bd9Sstevel@tonic-gate caddr_t page_ctrs_mem; 8467c478bd9Sstevel@tonic-gate size_t page_ctrs_size; 8477c478bd9Sstevel@tonic-gate struct memlist *current; 8487c478bd9Sstevel@tonic-gate extern void startup_build_mem_nodes(struct memlist *); 8497c478bd9Sstevel@tonic-gate 8507c478bd9Sstevel@tonic-gate /* XX64 fix these - they should be in include files */ 8517c478bd9Sstevel@tonic-gate extern size_t page_coloring_init(uint_t, int, int); 8527c478bd9Sstevel@tonic-gate extern void page_coloring_setup(caddr_t); 8537c478bd9Sstevel@tonic-gate 8547c478bd9Sstevel@tonic-gate PRM_POINT("startup_memlist() starting..."); 8557c478bd9Sstevel@tonic-gate 8567c478bd9Sstevel@tonic-gate /* 8577c478bd9Sstevel@tonic-gate * Use leftover large page nucleus text/data space for loadable modules. 8587c478bd9Sstevel@tonic-gate * Use at most MODTEXT/MODDATA. 8597c478bd9Sstevel@tonic-gate */ 860ae115bc7Smrj len = kbm_nucleus_size; 861ae115bc7Smrj ASSERT(len > MMU_PAGESIZE); 8627c478bd9Sstevel@tonic-gate 863ae115bc7Smrj moddata = (caddr_t)ROUND_UP_PAGE(e_data); 864ae115bc7Smrj e_moddata = (caddr_t)P2ROUNDUP((uintptr_t)e_data, (uintptr_t)len); 865ae115bc7Smrj if (e_moddata - moddata > MODDATA) 866ae115bc7Smrj e_moddata = moddata + MODDATA; 8677c478bd9Sstevel@tonic-gate 868ae115bc7Smrj modtext = (caddr_t)ROUND_UP_PAGE(e_text); 869ae115bc7Smrj e_modtext = (caddr_t)P2ROUNDUP((uintptr_t)e_text, (uintptr_t)len); 870ae115bc7Smrj if (e_modtext - modtext > MODTEXT) 871ae115bc7Smrj e_modtext = modtext + MODTEXT; 8727c478bd9Sstevel@tonic-gate 8737c478bd9Sstevel@tonic-gate econtig = e_moddata; 8747c478bd9Sstevel@tonic-gate 8757c478bd9Sstevel@tonic-gate PRM_DEBUG(modtext); 8767c478bd9Sstevel@tonic-gate PRM_DEBUG(e_modtext); 8777c478bd9Sstevel@tonic-gate PRM_DEBUG(moddata); 8787c478bd9Sstevel@tonic-gate PRM_DEBUG(e_moddata); 8797c478bd9Sstevel@tonic-gate PRM_DEBUG(econtig); 8807c478bd9Sstevel@tonic-gate 8817c478bd9Sstevel@tonic-gate /* 882ae115bc7Smrj * Examine the boot loader physical memory map to find out: 8837c478bd9Sstevel@tonic-gate * - total memory in system - physinstalled 8847c478bd9Sstevel@tonic-gate * - the max physical address - physmax 885ae115bc7Smrj * - the number of discontiguous segments of memory. 8867c478bd9Sstevel@tonic-gate */ 8877c478bd9Sstevel@tonic-gate if (prom_debug) 888ae115bc7Smrj print_memlist("boot physinstalled", 8897c478bd9Sstevel@tonic-gate bootops->boot_mem->physinstalled); 8907c478bd9Sstevel@tonic-gate installed_top_size(bootops->boot_mem->physinstalled, &physmax, 8917c478bd9Sstevel@tonic-gate &physinstalled, &memblocks); 8927c478bd9Sstevel@tonic-gate PRM_DEBUG(physmax); 8937c478bd9Sstevel@tonic-gate PRM_DEBUG(physinstalled); 8947c478bd9Sstevel@tonic-gate PRM_DEBUG(memblocks); 8957c478bd9Sstevel@tonic-gate 8967c478bd9Sstevel@tonic-gate /* 8977c478bd9Sstevel@tonic-gate * Initialize hat's mmu parameters. 8987c478bd9Sstevel@tonic-gate * Check for enforce-prot-exec in boot environment. It's used to 8997c478bd9Sstevel@tonic-gate * enable/disable support for the page table entry NX bit. 9007c478bd9Sstevel@tonic-gate * The default is to enforce PROT_EXEC on processors that support NX. 9017c478bd9Sstevel@tonic-gate * Boot seems to round up the "len", but 8 seems to be big enough. 9027c478bd9Sstevel@tonic-gate */ 9037c478bd9Sstevel@tonic-gate mmu_init(); 9047c478bd9Sstevel@tonic-gate 9057c478bd9Sstevel@tonic-gate #ifdef __i386 9067c478bd9Sstevel@tonic-gate /* 9077c478bd9Sstevel@tonic-gate * physmax is lowered if there is more memory than can be 9087c478bd9Sstevel@tonic-gate * physically addressed in 32 bit (PAE/non-PAE) modes. 9097c478bd9Sstevel@tonic-gate */ 9107c478bd9Sstevel@tonic-gate if (mmu.pae_hat) { 9117c478bd9Sstevel@tonic-gate if (PFN_ABOVE64G(physmax)) { 9127c478bd9Sstevel@tonic-gate physinstalled -= (physmax - (PFN_64G - 1)); 9137c478bd9Sstevel@tonic-gate physmax = PFN_64G - 1; 9147c478bd9Sstevel@tonic-gate } 9157c478bd9Sstevel@tonic-gate } else { 9167c478bd9Sstevel@tonic-gate if (PFN_ABOVE4G(physmax)) { 9177c478bd9Sstevel@tonic-gate physinstalled -= (physmax - (PFN_4G - 1)); 9187c478bd9Sstevel@tonic-gate physmax = PFN_4G - 1; 9197c478bd9Sstevel@tonic-gate } 9207c478bd9Sstevel@tonic-gate } 9217c478bd9Sstevel@tonic-gate #endif 9227c478bd9Sstevel@tonic-gate 9237c478bd9Sstevel@tonic-gate startup_build_mem_nodes(bootops->boot_mem->physinstalled); 9247c478bd9Sstevel@tonic-gate 9257c478bd9Sstevel@tonic-gate if (BOP_GETPROPLEN(bootops, "enforce-prot-exec") >= 0) { 9267c478bd9Sstevel@tonic-gate int len = BOP_GETPROPLEN(bootops, "enforce-prot-exec"); 9277c478bd9Sstevel@tonic-gate char value[8]; 9287c478bd9Sstevel@tonic-gate 9297c478bd9Sstevel@tonic-gate if (len < 8) 9307c478bd9Sstevel@tonic-gate (void) BOP_GETPROP(bootops, "enforce-prot-exec", value); 9317c478bd9Sstevel@tonic-gate else 9327c478bd9Sstevel@tonic-gate (void) strcpy(value, ""); 9337c478bd9Sstevel@tonic-gate if (strcmp(value, "off") == 0) 9347c478bd9Sstevel@tonic-gate mmu.pt_nx = 0; 9357c478bd9Sstevel@tonic-gate } 9367c478bd9Sstevel@tonic-gate PRM_DEBUG(mmu.pt_nx); 9377c478bd9Sstevel@tonic-gate 9387c478bd9Sstevel@tonic-gate /* 9397c478bd9Sstevel@tonic-gate * We will need page_t's for every page in the system, except for 9407c478bd9Sstevel@tonic-gate * memory mapped at or above above the start of the kernel text segment. 9417c478bd9Sstevel@tonic-gate * 9427c478bd9Sstevel@tonic-gate * pages above e_modtext are attributed to kernel debugger (obp_pages) 9437c478bd9Sstevel@tonic-gate */ 9447c478bd9Sstevel@tonic-gate npages = physinstalled - 1; /* avail_filter() skips page 0, so "- 1" */ 9457c478bd9Sstevel@tonic-gate obp_pages = 0; 9467c478bd9Sstevel@tonic-gate va = KERNEL_TEXT; 947ae115bc7Smrj while (kbm_probe(&va, &len, &pfn, &prot) != 0) { 9487c478bd9Sstevel@tonic-gate npages -= len >> MMU_PAGESHIFT; 9497c478bd9Sstevel@tonic-gate if (va >= (uintptr_t)e_moddata) 9507c478bd9Sstevel@tonic-gate obp_pages += len >> MMU_PAGESHIFT; 9517c478bd9Sstevel@tonic-gate va += len; 9527c478bd9Sstevel@tonic-gate } 9537c478bd9Sstevel@tonic-gate PRM_DEBUG(npages); 9547c478bd9Sstevel@tonic-gate PRM_DEBUG(obp_pages); 9557c478bd9Sstevel@tonic-gate 9567c478bd9Sstevel@tonic-gate /* 957*35b1ab99Sjosephb * If physmem is patched to be non-zero, use it instead of the computed 958*35b1ab99Sjosephb * value unless it is larger than the actual amount of memory on hand. 9597c478bd9Sstevel@tonic-gate */ 9604944b02eSkchow if (physmem == 0 || physmem > npages) { 9617c478bd9Sstevel@tonic-gate physmem = npages; 9624944b02eSkchow } else if (physmem < npages) { 963f53ad214Skchow orig_npages = npages; 9647c478bd9Sstevel@tonic-gate npages = physmem; 9654944b02eSkchow } 9667c478bd9Sstevel@tonic-gate PRM_DEBUG(physmem); 9677c478bd9Sstevel@tonic-gate 9687c478bd9Sstevel@tonic-gate /* 9697c478bd9Sstevel@tonic-gate * We now compute the sizes of all the initial allocations for 9707c478bd9Sstevel@tonic-gate * structures the kernel needs in order do kmem_alloc(). These 9717c478bd9Sstevel@tonic-gate * include: 9727c478bd9Sstevel@tonic-gate * memsegs 9737c478bd9Sstevel@tonic-gate * memlists 9747c478bd9Sstevel@tonic-gate * page hash table 9757c478bd9Sstevel@tonic-gate * page_t's 9767c478bd9Sstevel@tonic-gate * page coloring data structs 9777c478bd9Sstevel@tonic-gate */ 9787c478bd9Sstevel@tonic-gate memseg_sz = sizeof (struct memseg) * (memblocks + POSS_NEW_FRAGMENTS); 9797c478bd9Sstevel@tonic-gate ADD_TO_ALLOCATIONS(memseg_base, memseg_sz); 9807c478bd9Sstevel@tonic-gate PRM_DEBUG(memseg_sz); 9817c478bd9Sstevel@tonic-gate 9827c478bd9Sstevel@tonic-gate /* 983ae115bc7Smrj * Reserve space for memlists. There's no real good way to know exactly 984ae115bc7Smrj * how much room we'll need, but this should be a good upper bound. 9857c478bd9Sstevel@tonic-gate */ 9867c478bd9Sstevel@tonic-gate memlist_sz = ROUND_UP_PAGE(2 * sizeof (struct memlist) * 9877c478bd9Sstevel@tonic-gate (memblocks + POSS_NEW_FRAGMENTS)); 9887c478bd9Sstevel@tonic-gate ADD_TO_ALLOCATIONS(memlist, memlist_sz); 9897c478bd9Sstevel@tonic-gate PRM_DEBUG(memlist_sz); 9907c478bd9Sstevel@tonic-gate 9917c478bd9Sstevel@tonic-gate /* 9927c478bd9Sstevel@tonic-gate * The page structure hash table size is a power of 2 9937c478bd9Sstevel@tonic-gate * such that the average hash chain length is PAGE_HASHAVELEN. 9947c478bd9Sstevel@tonic-gate */ 9957c478bd9Sstevel@tonic-gate page_hashsz = npages / PAGE_HASHAVELEN; 9967c478bd9Sstevel@tonic-gate page_hashsz = 1 << highbit(page_hashsz); 9977c478bd9Sstevel@tonic-gate pagehash_sz = sizeof (struct page *) * page_hashsz; 9987c478bd9Sstevel@tonic-gate ADD_TO_ALLOCATIONS(page_hash, pagehash_sz); 9997c478bd9Sstevel@tonic-gate PRM_DEBUG(pagehash_sz); 10007c478bd9Sstevel@tonic-gate 10017c478bd9Sstevel@tonic-gate /* 1002ae115bc7Smrj * Set aside room for the page structures themselves. 10037c478bd9Sstevel@tonic-gate */ 1004ae115bc7Smrj PRM_DEBUG(npages); 1005ae115bc7Smrj pp_sz = sizeof (struct page) * npages; 10067c478bd9Sstevel@tonic-gate ADD_TO_ALLOCATIONS(pp_base, pp_sz); 10077c478bd9Sstevel@tonic-gate PRM_DEBUG(pp_sz); 10087c478bd9Sstevel@tonic-gate 10097c478bd9Sstevel@tonic-gate /* 10107c478bd9Sstevel@tonic-gate * determine l2 cache info and memory size for page coloring 10117c478bd9Sstevel@tonic-gate */ 10127c478bd9Sstevel@tonic-gate (void) getl2cacheinfo(CPU, 10137c478bd9Sstevel@tonic-gate &l2cache_sz, &l2cache_linesz, &l2cache_assoc); 10147c478bd9Sstevel@tonic-gate pagecolor_memsz = 10157c478bd9Sstevel@tonic-gate page_coloring_init(l2cache_sz, l2cache_linesz, l2cache_assoc); 10167c478bd9Sstevel@tonic-gate ADD_TO_ALLOCATIONS(pagecolor_mem, pagecolor_memsz); 10177c478bd9Sstevel@tonic-gate PRM_DEBUG(pagecolor_memsz); 10187c478bd9Sstevel@tonic-gate 10197c478bd9Sstevel@tonic-gate page_ctrs_size = page_ctrs_sz(); 10207c478bd9Sstevel@tonic-gate ADD_TO_ALLOCATIONS(page_ctrs_mem, page_ctrs_size); 10217c478bd9Sstevel@tonic-gate PRM_DEBUG(page_ctrs_size); 10227c478bd9Sstevel@tonic-gate 1023ae115bc7Smrj #if defined(__amd64) 10247c478bd9Sstevel@tonic-gate valloc_sz = ROUND_UP_LPAGE(valloc_sz); 1025ae115bc7Smrj valloc_base = VALLOC_BASE; 1026*35b1ab99Sjosephb 1027*35b1ab99Sjosephb /* 1028*35b1ab99Sjosephb * The default values of VALLOC_BASE and SEGKPM_BASE should work 1029*35b1ab99Sjosephb * for values of physmax up to 1 Terabyte. They need adjusting when 1030*35b1ab99Sjosephb * memory is at addresses above 1 TB. 1031*35b1ab99Sjosephb */ 1032*35b1ab99Sjosephb if (physmax + 1 > mmu_btop(TERABYTE)) { 1033*35b1ab99Sjosephb uint64_t kpm_resv_amount = mmu_ptob(physmax + 1); 1034*35b1ab99Sjosephb 1035*35b1ab99Sjosephb /* Round to largest possible pagesize for now */ 1036*35b1ab99Sjosephb kpm_resv_amount = P2ROUNDUP(kpm_resv_amount, ONE_GIG); 1037*35b1ab99Sjosephb 1038*35b1ab99Sjosephb segkpm_base = -(2 * kpm_resv_amount); /* down from top VA */ 1039*35b1ab99Sjosephb 1040*35b1ab99Sjosephb /* make sure we leave some space for user apps above hole */ 1041*35b1ab99Sjosephb segkpm_base = MAX(segkpm_base, AMD64_VA_HOLE_END + TERABYTE); 1042*35b1ab99Sjosephb if (segkpm_base > SEGKPM_BASE) 1043*35b1ab99Sjosephb segkpm_base = SEGKPM_BASE; 1044*35b1ab99Sjosephb PRM_DEBUG(segkpm_base); 1045*35b1ab99Sjosephb 1046*35b1ab99Sjosephb valloc_base = segkpm_base + kpm_resv_amount; 1047*35b1ab99Sjosephb PRM_DEBUG(valloc_base); 1048*35b1ab99Sjosephb } 1049ae115bc7Smrj #else /* __i386 */ 1050ae115bc7Smrj valloc_base = (uintptr_t)(MISC_VA_BASE - valloc_sz); 1051ae115bc7Smrj valloc_base = P2ALIGN(valloc_base, mmu.level_size[1]); 10527c478bd9Sstevel@tonic-gate PRM_DEBUG(valloc_base); 1053*35b1ab99Sjosephb #endif /* __i386 */ 1054ae115bc7Smrj 1055ae115bc7Smrj /* 1056ae115bc7Smrj * do all the initial allocations 1057ae115bc7Smrj */ 1058ae115bc7Smrj perform_allocations(); 1059ae115bc7Smrj 1060ae115bc7Smrj /* 1061ae115bc7Smrj * Build phys_install and phys_avail in kernel memspace. 1062ae115bc7Smrj * - phys_install should be all memory in the system. 1063ae115bc7Smrj * - phys_avail is phys_install minus any memory mapped before this 1064ae115bc7Smrj * point above KERNEL_TEXT. 1065ae115bc7Smrj */ 1066ae115bc7Smrj current = phys_install = memlist; 1067ae115bc7Smrj copy_memlist_filter(bootops->boot_mem->physinstalled, ¤t, NULL); 1068ae115bc7Smrj if ((caddr_t)current > (caddr_t)memlist + memlist_sz) 1069ae115bc7Smrj panic("physinstalled was too big!"); 1070ae115bc7Smrj if (prom_debug) 1071ae115bc7Smrj print_memlist("phys_install", phys_install); 1072ae115bc7Smrj 1073ae115bc7Smrj phys_avail = current; 1074ae115bc7Smrj PRM_POINT("Building phys_avail:\n"); 1075ae115bc7Smrj copy_memlist_filter(bootops->boot_mem->physinstalled, ¤t, 1076ae115bc7Smrj avail_filter); 1077ae115bc7Smrj if ((caddr_t)current > (caddr_t)memlist + memlist_sz) 1078ae115bc7Smrj panic("physavail was too big!"); 1079ae115bc7Smrj if (prom_debug) 1080ae115bc7Smrj print_memlist("phys_avail", phys_avail); 1081ae115bc7Smrj 1082ae115bc7Smrj /* 1083ae115bc7Smrj * setup page coloring 1084ae115bc7Smrj */ 1085ae115bc7Smrj page_coloring_setup(pagecolor_mem); 1086ae115bc7Smrj page_lock_init(); /* currently a no-op */ 1087ae115bc7Smrj 1088ae115bc7Smrj /* 1089ae115bc7Smrj * free page list counters 1090ae115bc7Smrj */ 1091ae115bc7Smrj (void) page_ctrs_alloc(page_ctrs_mem); 1092ae115bc7Smrj 1093ae115bc7Smrj /* 1094ae115bc7Smrj * Initialize the page structures from the memory lists. 1095ae115bc7Smrj */ 1096ae115bc7Smrj availrmem_initial = availrmem = freemem = 0; 1097ae115bc7Smrj PRM_POINT("Calling kphysm_init()..."); 1098ae115bc7Smrj npages = kphysm_init(pp_base, npages); 1099ae115bc7Smrj PRM_POINT("kphysm_init() done"); 1100ae115bc7Smrj PRM_DEBUG(npages); 1101ae115bc7Smrj 1102ae115bc7Smrj init_debug_info(); 1103ae115bc7Smrj 1104ae115bc7Smrj /* 1105ae115bc7Smrj * Now that page_t's have been initialized, remove all the 1106ae115bc7Smrj * initial allocation pages from the kernel free page lists. 1107ae115bc7Smrj */ 1108ae115bc7Smrj boot_mapin((caddr_t)valloc_base, valloc_sz); 1109ae115bc7Smrj boot_mapin((caddr_t)GDT_VA, MMU_PAGESIZE); 1110ae115bc7Smrj boot_mapin((caddr_t)DEBUG_INFO_VA, MMU_PAGESIZE); 1111ae115bc7Smrj PRM_POINT("startup_memlist() done"); 1112ae115bc7Smrj 1113ae115bc7Smrj PRM_DEBUG(valloc_sz); 11142cb27123Saguzovsk 1115567d55e1Saguzovsk #if defined(__amd64) 1116567d55e1Saguzovsk if ((availrmem >> (30 - MMU_PAGESHIFT)) >= 1117567d55e1Saguzovsk textrepl_min_gb && l2cache_sz <= 2 << 20) { 1118567d55e1Saguzovsk extern size_t textrepl_size_thresh; 11192cb27123Saguzovsk textrepl_size_thresh = (16 << 20) - 1; 11202cb27123Saguzovsk } 1121567d55e1Saguzovsk #endif 1122ae115bc7Smrj } 1123ae115bc7Smrj 1124ae115bc7Smrj /* 1125ae115bc7Smrj * Layout the kernel's part of address space and initialize kmem allocator. 1126ae115bc7Smrj */ 1127ae115bc7Smrj static void 1128ae115bc7Smrj startup_kmem(void) 1129ae115bc7Smrj { 1130932dc8e5Sdp extern void page_set_colorequiv_arr(void); 1131932dc8e5Sdp 1132ae115bc7Smrj PRM_POINT("startup_kmem() starting..."); 11337c478bd9Sstevel@tonic-gate 11347c478bd9Sstevel@tonic-gate #if defined(__amd64) 11357c478bd9Sstevel@tonic-gate if (eprom_kernelbase && eprom_kernelbase != KERNELBASE) 11367c478bd9Sstevel@tonic-gate cmn_err(CE_NOTE, "!kernelbase cannot be changed on 64-bit " 11377c478bd9Sstevel@tonic-gate "systems."); 1138*35b1ab99Sjosephb kernelbase = segkpm_base - KERNEL_REDZONE_SIZE; 11397c478bd9Sstevel@tonic-gate core_base = (uintptr_t)COREHEAP_BASE; 1140ae115bc7Smrj core_size = (size_t)MISC_VA_BASE - COREHEAP_BASE; 11417c478bd9Sstevel@tonic-gate #else /* __i386 */ 11427c478bd9Sstevel@tonic-gate /* 11437c478bd9Sstevel@tonic-gate * We configure kernelbase based on: 11447c478bd9Sstevel@tonic-gate * 11457c478bd9Sstevel@tonic-gate * 1. user specified kernelbase via eeprom command. Value cannot exceed 11467c478bd9Sstevel@tonic-gate * KERNELBASE_MAX. we large page align eprom_kernelbase 11477c478bd9Sstevel@tonic-gate * 11487c478bd9Sstevel@tonic-gate * 2. Default to KERNELBASE and adjust to 2X less the size for page_t. 11497c478bd9Sstevel@tonic-gate * On large memory systems we must lower kernelbase to allow 11507c478bd9Sstevel@tonic-gate * enough room for page_t's for all of memory. 11517c478bd9Sstevel@tonic-gate * 11527c478bd9Sstevel@tonic-gate * The value set here, might be changed a little later. 11537c478bd9Sstevel@tonic-gate */ 11547c478bd9Sstevel@tonic-gate if (eprom_kernelbase) { 11557c478bd9Sstevel@tonic-gate kernelbase = eprom_kernelbase & mmu.level_mask[1]; 11567c478bd9Sstevel@tonic-gate if (kernelbase > KERNELBASE_MAX) 11577c478bd9Sstevel@tonic-gate kernelbase = KERNELBASE_MAX; 11587c478bd9Sstevel@tonic-gate } else { 11597c478bd9Sstevel@tonic-gate kernelbase = (uintptr_t)KERNELBASE; 11607c478bd9Sstevel@tonic-gate kernelbase -= ROUND_UP_4MEG(2 * valloc_sz); 11617c478bd9Sstevel@tonic-gate } 11627c478bd9Sstevel@tonic-gate ASSERT((kernelbase & mmu.level_offset[1]) == 0); 1163ae115bc7Smrj core_base = valloc_base; 11647c478bd9Sstevel@tonic-gate core_size = 0; 1165ae115bc7Smrj #endif /* __i386 */ 11667c478bd9Sstevel@tonic-gate 11677c478bd9Sstevel@tonic-gate PRM_DEBUG(core_base); 11687c478bd9Sstevel@tonic-gate PRM_DEBUG(core_size); 1169ae115bc7Smrj PRM_DEBUG(kernelbase); 11707c478bd9Sstevel@tonic-gate 1171*35b1ab99Sjosephb #if defined(__i386) 11727c478bd9Sstevel@tonic-gate segkp_fromheap = 1; 1173ae115bc7Smrj #endif /* __i386 */ 1174*35b1ab99Sjosephb 11757c478bd9Sstevel@tonic-gate ekernelheap = (char *)core_base; 1176ae115bc7Smrj PRM_DEBUG(ekernelheap); 11777c478bd9Sstevel@tonic-gate 11787c478bd9Sstevel@tonic-gate /* 11797c478bd9Sstevel@tonic-gate * Now that we know the real value of kernelbase, 11807c478bd9Sstevel@tonic-gate * update variables that were initialized with a value of 11817c478bd9Sstevel@tonic-gate * KERNELBASE (in common/conf/param.c). 11827c478bd9Sstevel@tonic-gate * 11837c478bd9Sstevel@tonic-gate * XXX The problem with this sort of hackery is that the 11847c478bd9Sstevel@tonic-gate * compiler just may feel like putting the const declarations 11857c478bd9Sstevel@tonic-gate * (in param.c) into the .text section. Perhaps they should 11867c478bd9Sstevel@tonic-gate * just be declared as variables there? 11877c478bd9Sstevel@tonic-gate */ 11887c478bd9Sstevel@tonic-gate 11897c478bd9Sstevel@tonic-gate *(uintptr_t *)&_kernelbase = kernelbase; 11907c478bd9Sstevel@tonic-gate *(uintptr_t *)&_userlimit = kernelbase; 1191*35b1ab99Sjosephb #if !defined(__amd64) 11927c478bd9Sstevel@tonic-gate *(uintptr_t *)&_userlimit32 = _userlimit; 11937c478bd9Sstevel@tonic-gate #endif 11947c478bd9Sstevel@tonic-gate PRM_DEBUG(_kernelbase); 11957c478bd9Sstevel@tonic-gate PRM_DEBUG(_userlimit); 11967c478bd9Sstevel@tonic-gate PRM_DEBUG(_userlimit32); 11977c478bd9Sstevel@tonic-gate 1198*35b1ab99Sjosephb layout_kernel_va(); 1199*35b1ab99Sjosephb 1200*35b1ab99Sjosephb #if defined(__i386) 1201*35b1ab99Sjosephb /* 1202*35b1ab99Sjosephb * If segmap is too large we can push the bottom of the kernel heap 1203*35b1ab99Sjosephb * higher than the base. Or worse, it could exceed the top of the 1204*35b1ab99Sjosephb * VA space entirely, causing it to wrap around. 1205*35b1ab99Sjosephb */ 1206*35b1ab99Sjosephb if (kernelheap >= ekernelheap || (uintptr_t)kernelheap < kernelbase) 1207*35b1ab99Sjosephb panic("too little address space available for kernelheap," 1208*35b1ab99Sjosephb " use eeprom for lower kernelbase or smaller segmapsize"); 1209*35b1ab99Sjosephb #endif /* __i386 */ 1210*35b1ab99Sjosephb 12117c478bd9Sstevel@tonic-gate /* 12127c478bd9Sstevel@tonic-gate * Initialize the kernel heap. Note 3rd argument must be > 1st. 12137c478bd9Sstevel@tonic-gate */ 1214*35b1ab99Sjosephb kernelheap_init(kernelheap, ekernelheap, 1215*35b1ab99Sjosephb kernelheap + MMU_PAGESIZE, 1216ae115bc7Smrj (void *)core_base, (void *)(core_base + core_size)); 12177c478bd9Sstevel@tonic-gate 12187c478bd9Sstevel@tonic-gate /* 12197c478bd9Sstevel@tonic-gate * Initialize kernel memory allocator. 12207c478bd9Sstevel@tonic-gate */ 12217c478bd9Sstevel@tonic-gate kmem_init(); 12227c478bd9Sstevel@tonic-gate 1223932dc8e5Sdp /* 1224932dc8e5Sdp * Factor in colorequiv to check additional 'equivalent' bins 1225932dc8e5Sdp */ 1226932dc8e5Sdp page_set_colorequiv_arr(); 1227932dc8e5Sdp 12287c478bd9Sstevel@tonic-gate /* 12297c478bd9Sstevel@tonic-gate * print this out early so that we know what's going on 12307c478bd9Sstevel@tonic-gate */ 12317c478bd9Sstevel@tonic-gate cmn_err(CE_CONT, "?features: %b\n", x86_feature, FMT_X86_FEATURE); 12327c478bd9Sstevel@tonic-gate 12337c478bd9Sstevel@tonic-gate /* 12347c478bd9Sstevel@tonic-gate * Initialize bp_mapin(). 12357c478bd9Sstevel@tonic-gate */ 12367c478bd9Sstevel@tonic-gate bp_init(MMU_PAGESIZE, HAT_STORECACHING_OK); 12377c478bd9Sstevel@tonic-gate 1238f53ad214Skchow /* 1239f53ad214Skchow * orig_npages is non-zero if physmem has been configured for less 1240f53ad214Skchow * than the available memory. 1241f53ad214Skchow */ 1242f53ad214Skchow if (orig_npages) { 1243*35b1ab99Sjosephb cmn_err(CE_WARN, "!%slimiting physmem to 0x%lx of 0x%lx pages", 1244*35b1ab99Sjosephb (npages == PHYSMEM ? "Due to virtual address space " : ""), 1245*35b1ab99Sjosephb npages, orig_npages); 1246f53ad214Skchow } 12477c478bd9Sstevel@tonic-gate #if defined(__i386) 12487c478bd9Sstevel@tonic-gate if (eprom_kernelbase && (eprom_kernelbase != kernelbase)) 12497c478bd9Sstevel@tonic-gate cmn_err(CE_WARN, "kernelbase value, User specified 0x%lx, " 12507c478bd9Sstevel@tonic-gate "System using 0x%lx", 12517c478bd9Sstevel@tonic-gate (uintptr_t)eprom_kernelbase, (uintptr_t)kernelbase); 12527c478bd9Sstevel@tonic-gate #endif 12537c478bd9Sstevel@tonic-gate 12547c478bd9Sstevel@tonic-gate #ifdef KERNELBASE_ABI_MIN 12557c478bd9Sstevel@tonic-gate if (kernelbase < (uintptr_t)KERNELBASE_ABI_MIN) { 12567c478bd9Sstevel@tonic-gate cmn_err(CE_NOTE, "!kernelbase set to 0x%lx, system is not " 12577c478bd9Sstevel@tonic-gate "i386 ABI compliant.", (uintptr_t)kernelbase); 12587c478bd9Sstevel@tonic-gate } 12597c478bd9Sstevel@tonic-gate #endif 12607c478bd9Sstevel@tonic-gate 1261ae115bc7Smrj PRM_POINT("startup_kmem() done"); 12627c478bd9Sstevel@tonic-gate } 12637c478bd9Sstevel@tonic-gate 12647c478bd9Sstevel@tonic-gate static void 12657c478bd9Sstevel@tonic-gate startup_modules(void) 12667c478bd9Sstevel@tonic-gate { 12677c478bd9Sstevel@tonic-gate unsigned int i; 12687c478bd9Sstevel@tonic-gate extern void prom_setup(void); 12697c478bd9Sstevel@tonic-gate 12707c478bd9Sstevel@tonic-gate PRM_POINT("startup_modules() starting..."); 12717c478bd9Sstevel@tonic-gate /* 12727c478bd9Sstevel@tonic-gate * Initialize ten-micro second timer so that drivers will 12737c478bd9Sstevel@tonic-gate * not get short changed in their init phase. This was 12747c478bd9Sstevel@tonic-gate * not getting called until clkinit which, on fast cpu's 12757c478bd9Sstevel@tonic-gate * caused the drv_usecwait to be way too short. 12767c478bd9Sstevel@tonic-gate */ 12777c478bd9Sstevel@tonic-gate microfind(); 12787c478bd9Sstevel@tonic-gate 12797c478bd9Sstevel@tonic-gate /* 12807c478bd9Sstevel@tonic-gate * Read the GMT lag from /etc/rtc_config. 12817c478bd9Sstevel@tonic-gate */ 1282ae115bc7Smrj sgmtl(process_rtc_config_file()); 12837c478bd9Sstevel@tonic-gate 12847c478bd9Sstevel@tonic-gate /* 12857c478bd9Sstevel@tonic-gate * Calculate default settings of system parameters based upon 12867c478bd9Sstevel@tonic-gate * maxusers, yet allow to be overridden via the /etc/system file. 12877c478bd9Sstevel@tonic-gate */ 12887c478bd9Sstevel@tonic-gate param_calc(0); 12897c478bd9Sstevel@tonic-gate 12907c478bd9Sstevel@tonic-gate mod_setup(); 12917c478bd9Sstevel@tonic-gate 12927c478bd9Sstevel@tonic-gate /* 12937c478bd9Sstevel@tonic-gate * Initialize system parameters. 12947c478bd9Sstevel@tonic-gate */ 12957c478bd9Sstevel@tonic-gate param_init(); 12967c478bd9Sstevel@tonic-gate 12979acbbeafSnn /* 12989acbbeafSnn * Initialize the default brands 12999acbbeafSnn */ 13009acbbeafSnn brand_init(); 13019acbbeafSnn 13027c478bd9Sstevel@tonic-gate /* 13037c478bd9Sstevel@tonic-gate * maxmem is the amount of physical memory we're playing with. 13047c478bd9Sstevel@tonic-gate */ 13057c478bd9Sstevel@tonic-gate maxmem = physmem; 13067c478bd9Sstevel@tonic-gate 13077c478bd9Sstevel@tonic-gate /* 13087c478bd9Sstevel@tonic-gate * Initialize segment management stuff. 13097c478bd9Sstevel@tonic-gate */ 13107c478bd9Sstevel@tonic-gate seg_init(); 13117c478bd9Sstevel@tonic-gate 13127c478bd9Sstevel@tonic-gate if (modload("fs", "specfs") == -1) 13137c478bd9Sstevel@tonic-gate halt("Can't load specfs"); 13147c478bd9Sstevel@tonic-gate 13157c478bd9Sstevel@tonic-gate if (modload("fs", "devfs") == -1) 13167c478bd9Sstevel@tonic-gate halt("Can't load devfs"); 13177c478bd9Sstevel@tonic-gate 1318facf4a8dSllai if (modload("fs", "dev") == -1) 1319facf4a8dSllai halt("Can't load dev"); 1320facf4a8dSllai 132145916cd2Sjpk (void) modloadonly("sys", "lbl_edition"); 132245916cd2Sjpk 13237c478bd9Sstevel@tonic-gate dispinit(); 13247c478bd9Sstevel@tonic-gate 13257c478bd9Sstevel@tonic-gate /* 13267c478bd9Sstevel@tonic-gate * This is needed here to initialize hw_serial[] for cluster booting. 13277c478bd9Sstevel@tonic-gate */ 13287c478bd9Sstevel@tonic-gate if ((i = modload("misc", "sysinit")) != (unsigned int)-1) 13297c478bd9Sstevel@tonic-gate (void) modunload(i); 13307c478bd9Sstevel@tonic-gate else 13317c478bd9Sstevel@tonic-gate cmn_err(CE_CONT, "sysinit load failed"); 13327c478bd9Sstevel@tonic-gate 13337c478bd9Sstevel@tonic-gate /* Read cluster configuration data. */ 13347c478bd9Sstevel@tonic-gate clconf_init(); 13357c478bd9Sstevel@tonic-gate 13367c478bd9Sstevel@tonic-gate /* 13377c478bd9Sstevel@tonic-gate * Create a kernel device tree. First, create rootnex and 13387c478bd9Sstevel@tonic-gate * then invoke bus specific code to probe devices. 13397c478bd9Sstevel@tonic-gate */ 13407c478bd9Sstevel@tonic-gate setup_ddi(); 13417aec1d6eScindi 13427aec1d6eScindi /* 13437aec1d6eScindi * Set up the CPU module subsystem. Modifies the device tree, so it 13447aec1d6eScindi * must be done after setup_ddi(). 13457aec1d6eScindi */ 13467aec1d6eScindi cmi_init(); 13477aec1d6eScindi 13487aec1d6eScindi /* 13497aec1d6eScindi * Initialize the MCA handlers 13507aec1d6eScindi */ 13517aec1d6eScindi if (x86_feature & X86_MCA) 13527aec1d6eScindi cmi_mca_init(); 13537aec1d6eScindi 13547c478bd9Sstevel@tonic-gate /* 13557c478bd9Sstevel@tonic-gate * Fake a prom tree such that /dev/openprom continues to work 13567c478bd9Sstevel@tonic-gate */ 1357ae115bc7Smrj PRM_POINT("startup_modules: calling prom_setup..."); 13587c478bd9Sstevel@tonic-gate prom_setup(); 1359ae115bc7Smrj PRM_POINT("startup_modules: done"); 13607c478bd9Sstevel@tonic-gate 13617c478bd9Sstevel@tonic-gate /* 13627c478bd9Sstevel@tonic-gate * Load all platform specific modules 13637c478bd9Sstevel@tonic-gate */ 1364ae115bc7Smrj PRM_POINT("startup_modules: calling psm_modload..."); 13657c478bd9Sstevel@tonic-gate psm_modload(); 13667c478bd9Sstevel@tonic-gate 13677c478bd9Sstevel@tonic-gate PRM_POINT("startup_modules() done"); 13687c478bd9Sstevel@tonic-gate } 13697c478bd9Sstevel@tonic-gate 1370ae115bc7Smrj /* 1371ae115bc7Smrj * claim a "setaside" boot page for use in the kernel 1372ae115bc7Smrj */ 1373ae115bc7Smrj page_t * 1374ae115bc7Smrj boot_claim_page(pfn_t pfn) 13757c478bd9Sstevel@tonic-gate { 1376ae115bc7Smrj page_t *pp; 13777c478bd9Sstevel@tonic-gate 1378ae115bc7Smrj pp = page_numtopp_nolock(pfn); 1379ae115bc7Smrj ASSERT(pp != NULL); 13807c478bd9Sstevel@tonic-gate 1381ae115bc7Smrj if (PP_ISBOOTPAGES(pp)) { 1382ae115bc7Smrj if (pp->p_next != NULL) 1383ae115bc7Smrj pp->p_next->p_prev = pp->p_prev; 1384ae115bc7Smrj if (pp->p_prev == NULL) 1385ae115bc7Smrj bootpages = pp->p_next; 1386ae115bc7Smrj else 1387ae115bc7Smrj pp->p_prev->p_next = pp->p_next; 1388ae115bc7Smrj } else { 1389ae115bc7Smrj /* 1390ae115bc7Smrj * htable_attach() expects a base pagesize page 1391ae115bc7Smrj */ 1392ae115bc7Smrj if (pp->p_szc != 0) 1393ae115bc7Smrj page_boot_demote(pp); 1394ae115bc7Smrj pp = page_numtopp(pfn, SE_EXCL); 1395ae115bc7Smrj } 1396ae115bc7Smrj return (pp); 13977c478bd9Sstevel@tonic-gate } 13987c478bd9Sstevel@tonic-gate 13997c478bd9Sstevel@tonic-gate /* 14007c478bd9Sstevel@tonic-gate * Walk through the pagetables looking for pages mapped in by boot. If the 14017c478bd9Sstevel@tonic-gate * setaside flag is set the pages are expected to be returned to the 14027c478bd9Sstevel@tonic-gate * kernel later in boot, so we add them to the bootpages list. 14037c478bd9Sstevel@tonic-gate */ 14047c478bd9Sstevel@tonic-gate static void 14057c478bd9Sstevel@tonic-gate protect_boot_range(uintptr_t low, uintptr_t high, int setaside) 14067c478bd9Sstevel@tonic-gate { 14077c478bd9Sstevel@tonic-gate uintptr_t va = low; 14087c478bd9Sstevel@tonic-gate size_t len; 14097c478bd9Sstevel@tonic-gate uint_t prot; 14107c478bd9Sstevel@tonic-gate pfn_t pfn; 14117c478bd9Sstevel@tonic-gate page_t *pp; 14127c478bd9Sstevel@tonic-gate pgcnt_t boot_protect_cnt = 0; 14137c478bd9Sstevel@tonic-gate 1414ae115bc7Smrj while (kbm_probe(&va, &len, &pfn, &prot) != 0 && va < high) { 14157c478bd9Sstevel@tonic-gate if (va + len >= high) 14167c478bd9Sstevel@tonic-gate panic("0x%lx byte mapping at 0x%p exceeds boot's " 14177c478bd9Sstevel@tonic-gate "legal range.", len, (void *)va); 14187c478bd9Sstevel@tonic-gate 14197c478bd9Sstevel@tonic-gate while (len > 0) { 14207c478bd9Sstevel@tonic-gate pp = page_numtopp_alloc(pfn); 14217c478bd9Sstevel@tonic-gate if (pp != NULL) { 14227c478bd9Sstevel@tonic-gate if (setaside == 0) 14237c478bd9Sstevel@tonic-gate panic("Unexpected mapping by boot. " 14247c478bd9Sstevel@tonic-gate "addr=%p pfn=%lx\n", 14257c478bd9Sstevel@tonic-gate (void *)va, pfn); 14267c478bd9Sstevel@tonic-gate 14277c478bd9Sstevel@tonic-gate pp->p_next = bootpages; 1428ae115bc7Smrj pp->p_prev = NULL; 1429ae115bc7Smrj PP_SETBOOTPAGES(pp); 1430ae115bc7Smrj if (bootpages != NULL) { 1431ae115bc7Smrj bootpages->p_prev = pp; 1432ae115bc7Smrj } 14337c478bd9Sstevel@tonic-gate bootpages = pp; 14347c478bd9Sstevel@tonic-gate ++boot_protect_cnt; 14357c478bd9Sstevel@tonic-gate } 14367c478bd9Sstevel@tonic-gate 14377c478bd9Sstevel@tonic-gate ++pfn; 14387c478bd9Sstevel@tonic-gate len -= MMU_PAGESIZE; 14397c478bd9Sstevel@tonic-gate va += MMU_PAGESIZE; 14407c478bd9Sstevel@tonic-gate } 14417c478bd9Sstevel@tonic-gate } 14427c478bd9Sstevel@tonic-gate PRM_DEBUG(boot_protect_cnt); 14437c478bd9Sstevel@tonic-gate } 14447c478bd9Sstevel@tonic-gate 1445ae115bc7Smrj /* 1446*35b1ab99Sjosephb * 1447ae115bc7Smrj */ 14487c478bd9Sstevel@tonic-gate static void 1449*35b1ab99Sjosephb layout_kernel_va(void) 14507c478bd9Sstevel@tonic-gate { 1451*35b1ab99Sjosephb PRM_POINT("layout_kernel_va() starting..."); 14527c478bd9Sstevel@tonic-gate /* 1453ae115bc7Smrj * Establish the final size of the kernel's heap, size of segmap, 1454ae115bc7Smrj * segkp, etc. 14557c478bd9Sstevel@tonic-gate */ 14567c478bd9Sstevel@tonic-gate 14577c478bd9Sstevel@tonic-gate #if defined(__amd64) 14587c478bd9Sstevel@tonic-gate 1459*35b1ab99Sjosephb kpm_vbase = (caddr_t)segkpm_base; 1460*35b1ab99Sjosephb kpm_size = ROUND_UP_LPAGE(mmu_ptob(physmax + 1)); 1461*35b1ab99Sjosephb if ((uintptr_t)kpm_vbase + kpm_size > (uintptr_t)valloc_base) 1462*35b1ab99Sjosephb panic("not enough room for kpm!"); 1463ae115bc7Smrj PRM_DEBUG(kpm_size); 1464ae115bc7Smrj PRM_DEBUG(kpm_vbase); 14657c478bd9Sstevel@tonic-gate 14667c478bd9Sstevel@tonic-gate /* 1467ae115bc7Smrj * By default we create a seg_kp in 64 bit kernels, it's a little 1468ae115bc7Smrj * faster to access than embedding it in the heap. 14697c478bd9Sstevel@tonic-gate */ 1470ae115bc7Smrj segkp_base = (caddr_t)valloc_base + valloc_sz; 14717c478bd9Sstevel@tonic-gate if (!segkp_fromheap) { 14727c478bd9Sstevel@tonic-gate size_t sz = mmu_ptob(segkpsize); 14737c478bd9Sstevel@tonic-gate 14747c478bd9Sstevel@tonic-gate /* 1475ae115bc7Smrj * determine size of segkp 14767c478bd9Sstevel@tonic-gate */ 14777c478bd9Sstevel@tonic-gate if (sz < SEGKPMINSIZE || sz > SEGKPMAXSIZE) { 14787c478bd9Sstevel@tonic-gate sz = SEGKPDEFSIZE; 14797c478bd9Sstevel@tonic-gate cmn_err(CE_WARN, "!Illegal value for segkpsize. " 14807c478bd9Sstevel@tonic-gate "segkpsize has been reset to %ld pages", 14817c478bd9Sstevel@tonic-gate mmu_btop(sz)); 14827c478bd9Sstevel@tonic-gate } 14837c478bd9Sstevel@tonic-gate sz = MIN(sz, MAX(SEGKPMINSIZE, mmu_ptob(physmem))); 14847c478bd9Sstevel@tonic-gate 14857c478bd9Sstevel@tonic-gate segkpsize = mmu_btop(ROUND_UP_LPAGE(sz)); 14867c478bd9Sstevel@tonic-gate } 1487ae115bc7Smrj PRM_DEBUG(segkp_base); 1488ae115bc7Smrj PRM_DEBUG(segkpsize); 14897c478bd9Sstevel@tonic-gate 1490*35b1ab99Sjosephb /* 1491*35b1ab99Sjosephb * segzio is used for ZFS cached data. It uses a distinct VA 1492*35b1ab99Sjosephb * segment (from kernel heap) so that we can easily tell not to 1493*35b1ab99Sjosephb * include it in kernel crash dumps on 64 bit kernels. The trick is 1494*35b1ab99Sjosephb * to give it lots of VA, but not constrain the kernel heap. 1495*35b1ab99Sjosephb * We scale the size of segzio linearly with physmem up to 1496*35b1ab99Sjosephb * SEGZIOMAXSIZE. Above that amount it scales at 50% of physmem. 1497*35b1ab99Sjosephb */ 1498ae115bc7Smrj segzio_base = segkp_base + mmu_ptob(segkpsize); 1499ae115bc7Smrj if (segzio_fromheap) { 1500ae115bc7Smrj segziosize = 0; 1501ae115bc7Smrj } else { 1502*35b1ab99Sjosephb size_t physmem_size = mmu_ptob(physmem); 1503*35b1ab99Sjosephb size_t size = (segziosize == 0) ? 1504*35b1ab99Sjosephb physmem_size : mmu_ptob(segziosize); 1505ad23a2dbSjohansen 1506*35b1ab99Sjosephb if (size < SEGZIOMINSIZE) 1507ad23a2dbSjohansen size = SEGZIOMINSIZE; 1508*35b1ab99Sjosephb if (size > SEGZIOMAXSIZE) { 1509a778305aSjohansen size = SEGZIOMAXSIZE; 1510*35b1ab99Sjosephb if (physmem_size > size) 1511*35b1ab99Sjosephb size += (physmem_size - size) / 2; 1512ad23a2dbSjohansen } 1513ad23a2dbSjohansen segziosize = mmu_btop(ROUND_UP_LPAGE(size)); 1514ad23a2dbSjohansen } 1515ae115bc7Smrj PRM_DEBUG(segziosize); 1516ae115bc7Smrj PRM_DEBUG(segzio_base); 1517ad23a2dbSjohansen 15187c478bd9Sstevel@tonic-gate /* 1519ae115bc7Smrj * Put the range of VA for device mappings next, kmdb knows to not 1520ae115bc7Smrj * grep in this range of addresses. 15217c478bd9Sstevel@tonic-gate */ 1522ae115bc7Smrj toxic_addr = 1523ae115bc7Smrj ROUND_UP_LPAGE((uintptr_t)segzio_base + mmu_ptob(segziosize)); 15247c478bd9Sstevel@tonic-gate PRM_DEBUG(toxic_addr); 1525ae115bc7Smrj segmap_start = ROUND_UP_LPAGE(toxic_addr + toxic_size); 1526ae115bc7Smrj #else /* __i386 */ 1527ae115bc7Smrj segmap_start = ROUND_UP_LPAGE(kernelbase); 1528ae115bc7Smrj #endif /* __i386 */ 1529ae115bc7Smrj PRM_DEBUG(segmap_start); 15307c478bd9Sstevel@tonic-gate 15317c478bd9Sstevel@tonic-gate /* 15327c478bd9Sstevel@tonic-gate * Users can change segmapsize through eeprom or /etc/system. 15337c478bd9Sstevel@tonic-gate * If the variable is tuned through eeprom, there is no upper 15347c478bd9Sstevel@tonic-gate * bound on the size of segmap. If it is tuned through 15357c478bd9Sstevel@tonic-gate * /etc/system on 32-bit systems, it must be no larger than we 15367c478bd9Sstevel@tonic-gate * planned for in startup_memlist(). 15377c478bd9Sstevel@tonic-gate */ 15387c478bd9Sstevel@tonic-gate segmapsize = MAX(ROUND_UP_LPAGE(segmapsize), SEGMAPDEFAULT); 15397c478bd9Sstevel@tonic-gate 15407c478bd9Sstevel@tonic-gate #if defined(__i386) 15417c478bd9Sstevel@tonic-gate /* 15427c478bd9Sstevel@tonic-gate * 32-bit systems don't have segkpm or segkp, so segmap appears at 15437c478bd9Sstevel@tonic-gate * the bottom of the kernel's address range. Set aside space for a 1544ae115bc7Smrj * small red zone just below the start of segmap. 15457c478bd9Sstevel@tonic-gate */ 1546ae115bc7Smrj segmap_start += KERNEL_REDZONE_SIZE; 15477c478bd9Sstevel@tonic-gate segmapsize -= KERNEL_REDZONE_SIZE; 15487c478bd9Sstevel@tonic-gate #endif 15497c478bd9Sstevel@tonic-gate 1550ae115bc7Smrj PRM_DEBUG(segmap_start); 15517c478bd9Sstevel@tonic-gate PRM_DEBUG(segmapsize); 1552*35b1ab99Sjosephb kernelheap = (caddr_t)ROUND_UP_LPAGE(segmap_start + segmapsize); 1553*35b1ab99Sjosephb PRM_DEBUG(kernelheap); 1554*35b1ab99Sjosephb PRM_POINT("layout_kernel_va() done..."); 1555*35b1ab99Sjosephb } 1556*35b1ab99Sjosephb 1557*35b1ab99Sjosephb /* 1558*35b1ab99Sjosephb * Finish initializing the VM system, now that we are no longer 1559*35b1ab99Sjosephb * relying on the boot time memory allocators. 1560*35b1ab99Sjosephb */ 1561*35b1ab99Sjosephb static void 1562*35b1ab99Sjosephb startup_vm(void) 1563*35b1ab99Sjosephb { 1564*35b1ab99Sjosephb struct segmap_crargs a; 1565*35b1ab99Sjosephb 1566*35b1ab99Sjosephb extern int use_brk_lpg, use_stk_lpg; 1567*35b1ab99Sjosephb 1568*35b1ab99Sjosephb PRM_POINT("startup_vm() starting..."); 1569*35b1ab99Sjosephb 1570*35b1ab99Sjosephb /* 1571*35b1ab99Sjosephb * Initialize the hat layer. 1572*35b1ab99Sjosephb */ 1573*35b1ab99Sjosephb hat_init(); 15747c478bd9Sstevel@tonic-gate 1575ae115bc7Smrj /* 1576ae115bc7Smrj * Do final allocations of HAT data structures that need to 1577ae115bc7Smrj * be allocated before quiescing the boot loader. 1578ae115bc7Smrj */ 1579ae115bc7Smrj PRM_POINT("Calling hat_kern_alloc()..."); 1580ae115bc7Smrj hat_kern_alloc((caddr_t)segmap_start, segmapsize, ekernelheap); 1581ae115bc7Smrj PRM_POINT("hat_kern_alloc() done"); 1582ae115bc7Smrj 1583ae115bc7Smrj /* 1584ae115bc7Smrj * Setup MTRR (Memory type range registers) 1585ae115bc7Smrj */ 1586ae115bc7Smrj setup_mtrr(); 1587ae115bc7Smrj 1588ae115bc7Smrj /* 1589ae115bc7Smrj * The next two loops are done in distinct steps in order 1590ae115bc7Smrj * to be sure that any page that is doubly mapped (both above 1591ae115bc7Smrj * KERNEL_TEXT and below kernelbase) is dealt with correctly. 1592ae115bc7Smrj * Note this may never happen, but it might someday. 1593ae115bc7Smrj */ 1594ae115bc7Smrj bootpages = NULL; 1595ae115bc7Smrj PRM_POINT("Protecting boot pages"); 1596ae115bc7Smrj 1597ae115bc7Smrj /* 1598ae115bc7Smrj * Protect any pages mapped above KERNEL_TEXT that somehow have 1599ae115bc7Smrj * page_t's. This can only happen if something weird allocated 1600ae115bc7Smrj * in this range (like kadb/kmdb). 1601ae115bc7Smrj */ 1602ae115bc7Smrj protect_boot_range(KERNEL_TEXT, (uintptr_t)-1, 0); 1603ae115bc7Smrj 1604ae115bc7Smrj /* 1605ae115bc7Smrj * Before we can take over memory allocation/mapping from the boot 1606ae115bc7Smrj * loader we must remove from our free page lists any boot allocated 1607ae115bc7Smrj * pages that stay mapped until release_bootstrap(). 1608ae115bc7Smrj */ 1609ae115bc7Smrj protect_boot_range(0, kernelbase, 1); 1610ae115bc7Smrj 1611ae115bc7Smrj /* 1612ae115bc7Smrj * Switch to running on regular HAT (not boot_mmu) 1613ae115bc7Smrj */ 1614ae115bc7Smrj PRM_POINT("Calling hat_kern_setup()..."); 1615ae115bc7Smrj hat_kern_setup(); 1616ae115bc7Smrj 1617ae115bc7Smrj /* 1618ae115bc7Smrj * It is no longer safe to call BOP_ALLOC(), so make sure we don't. 1619ae115bc7Smrj */ 1620ae115bc7Smrj bop_no_more_mem(); 1621ae115bc7Smrj 1622ae115bc7Smrj PRM_POINT("hat_kern_setup() done"); 1623ae115bc7Smrj 1624ae115bc7Smrj hat_cpu_online(CPU); 1625ae115bc7Smrj 16267c478bd9Sstevel@tonic-gate /* 16277c478bd9Sstevel@tonic-gate * Initialize VM system 16287c478bd9Sstevel@tonic-gate */ 16297c478bd9Sstevel@tonic-gate PRM_POINT("Calling kvm_init()..."); 16307c478bd9Sstevel@tonic-gate kvm_init(); 16317c478bd9Sstevel@tonic-gate PRM_POINT("kvm_init() done"); 16327c478bd9Sstevel@tonic-gate 16337c478bd9Sstevel@tonic-gate /* 16347c478bd9Sstevel@tonic-gate * Tell kmdb that the VM system is now working 16357c478bd9Sstevel@tonic-gate */ 16367c478bd9Sstevel@tonic-gate if (boothowto & RB_DEBUG) 16377c478bd9Sstevel@tonic-gate kdi_dvec_vmready(); 16387c478bd9Sstevel@tonic-gate 16397c478bd9Sstevel@tonic-gate /* 16407c478bd9Sstevel@tonic-gate * Mangle the brand string etc. 16417c478bd9Sstevel@tonic-gate */ 16427c478bd9Sstevel@tonic-gate cpuid_pass3(CPU); 16437c478bd9Sstevel@tonic-gate 16447c478bd9Sstevel@tonic-gate #if defined(__amd64) 16457c478bd9Sstevel@tonic-gate 16467c478bd9Sstevel@tonic-gate /* 16477c478bd9Sstevel@tonic-gate * Create the device arena for toxic (to dtrace/kmdb) mappings. 16487c478bd9Sstevel@tonic-gate */ 16497c478bd9Sstevel@tonic-gate device_arena = vmem_create("device", (void *)toxic_addr, 16507c478bd9Sstevel@tonic-gate toxic_size, MMU_PAGESIZE, NULL, NULL, NULL, 0, VM_SLEEP); 16517c478bd9Sstevel@tonic-gate 16527c478bd9Sstevel@tonic-gate #else /* __i386 */ 16537c478bd9Sstevel@tonic-gate 16547c478bd9Sstevel@tonic-gate /* 16557c478bd9Sstevel@tonic-gate * allocate the bit map that tracks toxic pages 16567c478bd9Sstevel@tonic-gate */ 1657ae115bc7Smrj toxic_bit_map_len = btop((ulong_t)(valloc_base - kernelbase)); 16587c478bd9Sstevel@tonic-gate PRM_DEBUG(toxic_bit_map_len); 16597c478bd9Sstevel@tonic-gate toxic_bit_map = 16607c478bd9Sstevel@tonic-gate kmem_zalloc(BT_SIZEOFMAP(toxic_bit_map_len), KM_NOSLEEP); 16617c478bd9Sstevel@tonic-gate ASSERT(toxic_bit_map != NULL); 16627c478bd9Sstevel@tonic-gate PRM_DEBUG(toxic_bit_map); 16637c478bd9Sstevel@tonic-gate 16647c478bd9Sstevel@tonic-gate #endif /* __i386 */ 16657c478bd9Sstevel@tonic-gate 16667c478bd9Sstevel@tonic-gate 16677c478bd9Sstevel@tonic-gate /* 16687c478bd9Sstevel@tonic-gate * Now that we've got more VA, as well as the ability to allocate from 16697c478bd9Sstevel@tonic-gate * it, tell the debugger. 16707c478bd9Sstevel@tonic-gate */ 16717c478bd9Sstevel@tonic-gate if (boothowto & RB_DEBUG) 16727c478bd9Sstevel@tonic-gate kdi_dvec_memavail(); 16737c478bd9Sstevel@tonic-gate 16747c478bd9Sstevel@tonic-gate /* 16757c478bd9Sstevel@tonic-gate * The following code installs a special page fault handler (#pf) 16767c478bd9Sstevel@tonic-gate * to work around a pentium bug. 16777c478bd9Sstevel@tonic-gate */ 16787c478bd9Sstevel@tonic-gate #if !defined(__amd64) 16797c478bd9Sstevel@tonic-gate if (x86_type == X86_TYPE_P5) { 1680ae115bc7Smrj desctbr_t idtr; 16817c478bd9Sstevel@tonic-gate gate_desc_t *newidt; 1682ae115bc7Smrj struct machcpu *mcpu = &CPU->cpu_m; 16837c478bd9Sstevel@tonic-gate 16847c478bd9Sstevel@tonic-gate if ((newidt = kmem_zalloc(MMU_PAGESIZE, KM_NOSLEEP)) == NULL) 16857c478bd9Sstevel@tonic-gate panic("failed to install pentium_pftrap"); 16867c478bd9Sstevel@tonic-gate 16877c478bd9Sstevel@tonic-gate bcopy(idt0, newidt, sizeof (idt0)); 16887c478bd9Sstevel@tonic-gate set_gatesegd(&newidt[T_PGFLT], &pentium_pftrap, 1689ae115bc7Smrj KCS_SEL, SDT_SYSIGT, SEL_KPL); 16907c478bd9Sstevel@tonic-gate 16917c478bd9Sstevel@tonic-gate (void) as_setprot(&kas, (caddr_t)newidt, MMU_PAGESIZE, 16927c478bd9Sstevel@tonic-gate PROT_READ|PROT_EXEC); 16937c478bd9Sstevel@tonic-gate 1694ae115bc7Smrj mcpu->mcpu_idt = newidt; 1695ae115bc7Smrj idtr.dtr_base = (uintptr_t)mcpu->mcpu_idt; 1696ae115bc7Smrj idtr.dtr_limit = sizeof (idt0) - 1; 1697ae115bc7Smrj wr_idtr(&idtr); 16987c478bd9Sstevel@tonic-gate } 16997c478bd9Sstevel@tonic-gate #endif /* !__amd64 */ 17007c478bd9Sstevel@tonic-gate 17017c478bd9Sstevel@tonic-gate /* 17027c478bd9Sstevel@tonic-gate * Map page pfn=0 for drivers, such as kd, that need to pick up 17037c478bd9Sstevel@tonic-gate * parameters left there by controllers/BIOS. 17047c478bd9Sstevel@tonic-gate */ 17057c478bd9Sstevel@tonic-gate PRM_POINT("setup up p0_va"); 17067c478bd9Sstevel@tonic-gate p0_va = i86devmap(0, 1, PROT_READ); 17077c478bd9Sstevel@tonic-gate PRM_DEBUG(p0_va); 17087c478bd9Sstevel@tonic-gate 17097c478bd9Sstevel@tonic-gate cmn_err(CE_CONT, "?mem = %luK (0x%lx)\n", 17107c478bd9Sstevel@tonic-gate physinstalled << (MMU_PAGESHIFT - 10), ptob(physinstalled)); 17117c478bd9Sstevel@tonic-gate 171283f9b804Skchow /* 171383f9b804Skchow * disable automatic large pages for small memory systems or 171483f9b804Skchow * when the disable flag is set. 171583f9b804Skchow */ 1716ec25b48fSsusans if (!auto_lpg_disable && mmu.max_page_level > 0) { 1717ec25b48fSsusans max_uheap_lpsize = LEVEL_SIZE(1); 1718ec25b48fSsusans max_ustack_lpsize = LEVEL_SIZE(1); 1719ec25b48fSsusans max_privmap_lpsize = LEVEL_SIZE(1); 1720ec25b48fSsusans max_uidata_lpsize = LEVEL_SIZE(1); 1721ec25b48fSsusans max_utext_lpsize = LEVEL_SIZE(1); 1722ec25b48fSsusans max_shm_lpsize = LEVEL_SIZE(1); 1723ec25b48fSsusans } 1724ec25b48fSsusans if (physmem < privm_lpg_min_physmem || mmu.max_page_level == 0 || 1725ec25b48fSsusans auto_lpg_disable) { 1726beb1bda0Sdavemq use_brk_lpg = 0; 1727beb1bda0Sdavemq use_stk_lpg = 0; 1728ec25b48fSsusans } 1729ec25b48fSsusans if (mmu.max_page_level > 0) { 1730ec25b48fSsusans mcntl0_lpsize = LEVEL_SIZE(1); 1731beb1bda0Sdavemq } 1732beb1bda0Sdavemq 17337c478bd9Sstevel@tonic-gate PRM_POINT("Calling hat_init_finish()..."); 17347c478bd9Sstevel@tonic-gate hat_init_finish(); 17357c478bd9Sstevel@tonic-gate PRM_POINT("hat_init_finish() done"); 17367c478bd9Sstevel@tonic-gate 17377c478bd9Sstevel@tonic-gate /* 17387c478bd9Sstevel@tonic-gate * Initialize the segkp segment type. 17397c478bd9Sstevel@tonic-gate */ 17407c478bd9Sstevel@tonic-gate rw_enter(&kas.a_lock, RW_WRITER); 1741ae115bc7Smrj PRM_POINT("Attaching segkp"); 1742ae115bc7Smrj if (segkp_fromheap) { 17437c478bd9Sstevel@tonic-gate segkp->s_as = &kas; 1744ae115bc7Smrj } else if (seg_attach(&kas, (caddr_t)segkp_base, mmu_ptob(segkpsize), 1745ae115bc7Smrj segkp) < 0) { 1746ae115bc7Smrj panic("startup: cannot attach segkp"); 1747ae115bc7Smrj /*NOTREACHED*/ 17487c478bd9Sstevel@tonic-gate } 1749ae115bc7Smrj PRM_POINT("Doing segkp_create()"); 17507c478bd9Sstevel@tonic-gate if (segkp_create(segkp) != 0) { 17517c478bd9Sstevel@tonic-gate panic("startup: segkp_create failed"); 17527c478bd9Sstevel@tonic-gate /*NOTREACHED*/ 17537c478bd9Sstevel@tonic-gate } 17547c478bd9Sstevel@tonic-gate PRM_DEBUG(segkp); 17557c478bd9Sstevel@tonic-gate rw_exit(&kas.a_lock); 17567c478bd9Sstevel@tonic-gate 17577c478bd9Sstevel@tonic-gate /* 17587c478bd9Sstevel@tonic-gate * kpm segment 17597c478bd9Sstevel@tonic-gate */ 17607c478bd9Sstevel@tonic-gate segmap_kpm = 0; 17617c478bd9Sstevel@tonic-gate if (kpm_desired) { 17627c478bd9Sstevel@tonic-gate kpm_init(); 17637c478bd9Sstevel@tonic-gate kpm_enable = 1; 1764a5652762Spraks vpm_enable = 1; 17657c478bd9Sstevel@tonic-gate } 17667c478bd9Sstevel@tonic-gate 17677c478bd9Sstevel@tonic-gate /* 17687c478bd9Sstevel@tonic-gate * Now create segmap segment. 17697c478bd9Sstevel@tonic-gate */ 17707c478bd9Sstevel@tonic-gate rw_enter(&kas.a_lock, RW_WRITER); 1771ae115bc7Smrj if (seg_attach(&kas, (caddr_t)segmap_start, segmapsize, segmap) < 0) { 1772ae115bc7Smrj panic("cannot attach segmap"); 17737c478bd9Sstevel@tonic-gate /*NOTREACHED*/ 17747c478bd9Sstevel@tonic-gate } 1775ae115bc7Smrj PRM_DEBUG(segmap); 17767c478bd9Sstevel@tonic-gate 17777c478bd9Sstevel@tonic-gate a.prot = PROT_READ | PROT_WRITE; 17787c478bd9Sstevel@tonic-gate a.shmsize = 0; 17797c478bd9Sstevel@tonic-gate a.nfreelist = segmapfreelists; 17807c478bd9Sstevel@tonic-gate 1781ae115bc7Smrj if (segmap_create(segmap, (caddr_t)&a) != 0) 1782ae115bc7Smrj panic("segmap_create segmap"); 17837c478bd9Sstevel@tonic-gate rw_exit(&kas.a_lock); 17847c478bd9Sstevel@tonic-gate 17857c478bd9Sstevel@tonic-gate setup_vaddr_for_ppcopy(CPU); 17867c478bd9Sstevel@tonic-gate 17877c478bd9Sstevel@tonic-gate segdev_init(); 17887c478bd9Sstevel@tonic-gate pmem_init(); 1789ae115bc7Smrj 17907c478bd9Sstevel@tonic-gate PRM_POINT("startup_vm() done"); 17917c478bd9Sstevel@tonic-gate } 17927c478bd9Sstevel@tonic-gate 1793ae115bc7Smrj /* 1794ae115bc7Smrj * Load a tod module for the non-standard tod part found on this system. 1795ae115bc7Smrj */ 1796ae115bc7Smrj static void 1797ae115bc7Smrj load_tod_module(char *todmod) 1798ae115bc7Smrj { 1799ae115bc7Smrj if (modload("tod", todmod) == -1) 1800ae115bc7Smrj halt("Can't load TOD module"); 1801ae115bc7Smrj } 1802ae115bc7Smrj 18037c478bd9Sstevel@tonic-gate static void 18047c478bd9Sstevel@tonic-gate startup_end(void) 18057c478bd9Sstevel@tonic-gate { 18067c478bd9Sstevel@tonic-gate extern void setx86isalist(void); 18077c478bd9Sstevel@tonic-gate 18087c478bd9Sstevel@tonic-gate PRM_POINT("startup_end() starting..."); 18097c478bd9Sstevel@tonic-gate 18107c478bd9Sstevel@tonic-gate /* 18117c478bd9Sstevel@tonic-gate * Perform tasks that get done after most of the VM 18127c478bd9Sstevel@tonic-gate * initialization has been done but before the clock 18137c478bd9Sstevel@tonic-gate * and other devices get started. 18147c478bd9Sstevel@tonic-gate */ 18157c478bd9Sstevel@tonic-gate kern_setup1(); 18167c478bd9Sstevel@tonic-gate 18177c478bd9Sstevel@tonic-gate /* 18187c478bd9Sstevel@tonic-gate * Perform CPC initialization for this CPU. 18197c478bd9Sstevel@tonic-gate */ 18207c478bd9Sstevel@tonic-gate kcpc_hw_init(CPU); 18217c478bd9Sstevel@tonic-gate 1822ee88d2b9Skchow #if defined(OPTERON_WORKAROUND_6323525) 1823ee88d2b9Skchow if (opteron_workaround_6323525) 1824ee88d2b9Skchow patch_workaround_6323525(); 1825ee88d2b9Skchow #endif 1826ae115bc7Smrj /* 1827ae115bc7Smrj * If needed, load TOD module now so that ddi_get_time(9F) etc. work 1828ae115bc7Smrj * (For now, "needed" is defined as set tod_module_name in /etc/system) 1829ae115bc7Smrj */ 1830ae115bc7Smrj if (tod_module_name != NULL) { 1831ae115bc7Smrj PRM_POINT("load_tod_module()"); 1832ae115bc7Smrj load_tod_module(tod_module_name); 1833ae115bc7Smrj } 1834ae115bc7Smrj 18357c478bd9Sstevel@tonic-gate /* 18367c478bd9Sstevel@tonic-gate * Configure the system. 18377c478bd9Sstevel@tonic-gate */ 18387c478bd9Sstevel@tonic-gate PRM_POINT("Calling configure()..."); 18397c478bd9Sstevel@tonic-gate configure(); /* set up devices */ 18407c478bd9Sstevel@tonic-gate PRM_POINT("configure() done"); 18417c478bd9Sstevel@tonic-gate 18427c478bd9Sstevel@tonic-gate /* 18437c478bd9Sstevel@tonic-gate * Set the isa_list string to the defined instruction sets we 18447c478bd9Sstevel@tonic-gate * support. 18457c478bd9Sstevel@tonic-gate */ 18467c478bd9Sstevel@tonic-gate setx86isalist(); 1847100b72f4Sandrei cpu_intr_alloc(CPU, NINTR_THREADS); 18487c478bd9Sstevel@tonic-gate psm_install(); 18497c478bd9Sstevel@tonic-gate 18507c478bd9Sstevel@tonic-gate /* 18517c478bd9Sstevel@tonic-gate * We're done with bootops. We don't unmap the bootstrap yet because 18527c478bd9Sstevel@tonic-gate * we're still using bootsvcs. 18537c478bd9Sstevel@tonic-gate */ 1854ae115bc7Smrj PRM_POINT("NULLing out bootops"); 1855ae115bc7Smrj *bootopsp = (struct bootops *)NULL; 18567c478bd9Sstevel@tonic-gate bootops = (struct bootops *)NULL; 18577c478bd9Sstevel@tonic-gate 18587c478bd9Sstevel@tonic-gate PRM_POINT("Enabling interrupts"); 18597c478bd9Sstevel@tonic-gate (*picinitf)(); 18607c478bd9Sstevel@tonic-gate sti(); 18617c478bd9Sstevel@tonic-gate 18627c478bd9Sstevel@tonic-gate (void) add_avsoftintr((void *)&softlevel1_hdl, 1, softlevel1, 1863*35b1ab99Sjosephb "softlevel1", NULL, NULL); /* XXX to be moved later */ 18647c478bd9Sstevel@tonic-gate 18657c478bd9Sstevel@tonic-gate PRM_POINT("startup_end() done"); 18667c478bd9Sstevel@tonic-gate } 18677c478bd9Sstevel@tonic-gate 18687c478bd9Sstevel@tonic-gate extern char hw_serial[]; 18697c478bd9Sstevel@tonic-gate char *_hs1107 = hw_serial; 18707c478bd9Sstevel@tonic-gate ulong_t _bdhs34; 18717c478bd9Sstevel@tonic-gate 18727c478bd9Sstevel@tonic-gate void 18737c478bd9Sstevel@tonic-gate post_startup(void) 18747c478bd9Sstevel@tonic-gate { 18757c478bd9Sstevel@tonic-gate /* 18767c478bd9Sstevel@tonic-gate * Set the system wide, processor-specific flags to be passed 18777c478bd9Sstevel@tonic-gate * to userland via the aux vector for performance hints and 18787c478bd9Sstevel@tonic-gate * instruction set extensions. 18797c478bd9Sstevel@tonic-gate */ 18807c478bd9Sstevel@tonic-gate bind_hwcap(); 18817c478bd9Sstevel@tonic-gate 18827c478bd9Sstevel@tonic-gate /* 1883ae115bc7Smrj * Load the System Management BIOS into the global ksmbios 1884ae115bc7Smrj * handle, if an SMBIOS is present on this system. 18857c478bd9Sstevel@tonic-gate */ 188684ab085aSmws ksmbios = smbios_open(NULL, SMB_VERSION, ksmbios_flags, NULL); 18877c478bd9Sstevel@tonic-gate 18887c478bd9Sstevel@tonic-gate /* 1889ae115bc7Smrj * Startup the memory scrubber. 18907c478bd9Sstevel@tonic-gate */ 189184ab085aSmws memscrub_init(); 18927c478bd9Sstevel@tonic-gate 18937aec1d6eScindi /* 18947aec1d6eScindi * Complete CPU module initialization 18957aec1d6eScindi */ 18967aec1d6eScindi cmi_post_init(); 18977aec1d6eScindi 18987c478bd9Sstevel@tonic-gate /* 189984ab085aSmws * Perform forceloading tasks for /etc/system. 19007c478bd9Sstevel@tonic-gate */ 190184ab085aSmws (void) mod_sysctl(SYS_FORCELOAD, NULL); 19027c478bd9Sstevel@tonic-gate 19037c478bd9Sstevel@tonic-gate /* 19047c478bd9Sstevel@tonic-gate * ON4.0: Force /proc module in until clock interrupt handle fixed 19057c478bd9Sstevel@tonic-gate * ON4.0: This must be fixed or restated in /etc/systems. 19067c478bd9Sstevel@tonic-gate */ 19077c478bd9Sstevel@tonic-gate (void) modload("fs", "procfs"); 19087c478bd9Sstevel@tonic-gate 19097c478bd9Sstevel@tonic-gate #if defined(__i386) 19107c478bd9Sstevel@tonic-gate /* 19117c478bd9Sstevel@tonic-gate * Check for required functional Floating Point hardware, 19127c478bd9Sstevel@tonic-gate * unless FP hardware explicitly disabled. 19137c478bd9Sstevel@tonic-gate */ 19147c478bd9Sstevel@tonic-gate if (fpu_exists && (fpu_pentium_fdivbug || fp_kind == FP_NO)) 19157c478bd9Sstevel@tonic-gate halt("No working FP hardware found"); 19167c478bd9Sstevel@tonic-gate #endif 19177c478bd9Sstevel@tonic-gate 19187c478bd9Sstevel@tonic-gate maxmem = freemem; 19197c478bd9Sstevel@tonic-gate 19207c478bd9Sstevel@tonic-gate add_cpunode2devtree(CPU->cpu_id, CPU->cpu_m.mcpu_cpi); 19217c478bd9Sstevel@tonic-gate } 19227c478bd9Sstevel@tonic-gate 19237c478bd9Sstevel@tonic-gate static int 19247c478bd9Sstevel@tonic-gate pp_in_ramdisk(page_t *pp) 19257c478bd9Sstevel@tonic-gate { 19267c478bd9Sstevel@tonic-gate extern uint64_t ramdisk_start, ramdisk_end; 19277c478bd9Sstevel@tonic-gate 19287c478bd9Sstevel@tonic-gate return ((pp->p_pagenum >= btop(ramdisk_start)) && 19297c478bd9Sstevel@tonic-gate (pp->p_pagenum < btopr(ramdisk_end))); 19307c478bd9Sstevel@tonic-gate } 19317c478bd9Sstevel@tonic-gate 19327c478bd9Sstevel@tonic-gate void 19337c478bd9Sstevel@tonic-gate release_bootstrap(void) 19347c478bd9Sstevel@tonic-gate { 19357c478bd9Sstevel@tonic-gate int root_is_ramdisk; 19367c478bd9Sstevel@tonic-gate page_t *pp; 19377c478bd9Sstevel@tonic-gate extern void kobj_boot_unmountroot(void); 19387c478bd9Sstevel@tonic-gate extern dev_t rootdev; 19397c478bd9Sstevel@tonic-gate 19407c478bd9Sstevel@tonic-gate /* unmount boot ramdisk and release kmem usage */ 19417c478bd9Sstevel@tonic-gate kobj_boot_unmountroot(); 19427c478bd9Sstevel@tonic-gate 19437c478bd9Sstevel@tonic-gate /* 19447c478bd9Sstevel@tonic-gate * We're finished using the boot loader so free its pages. 19457c478bd9Sstevel@tonic-gate */ 19467c478bd9Sstevel@tonic-gate PRM_POINT("Unmapping lower boot pages"); 1947ae115bc7Smrj clear_boot_mappings(0, _userlimit); 1948ae115bc7Smrj postbootkernelbase = kernelbase; 19497c478bd9Sstevel@tonic-gate 19507c478bd9Sstevel@tonic-gate /* 19517c478bd9Sstevel@tonic-gate * If root isn't on ramdisk, destroy the hardcoded 19527c478bd9Sstevel@tonic-gate * ramdisk node now and release the memory. Else, 19537c478bd9Sstevel@tonic-gate * ramdisk memory is kept in rd_pages. 19547c478bd9Sstevel@tonic-gate */ 19557c478bd9Sstevel@tonic-gate root_is_ramdisk = (getmajor(rootdev) == ddi_name_to_major("ramdisk")); 19567c478bd9Sstevel@tonic-gate if (!root_is_ramdisk) { 19577c478bd9Sstevel@tonic-gate dev_info_t *dip = ddi_find_devinfo("ramdisk", -1, 0); 19587c478bd9Sstevel@tonic-gate ASSERT(dip && ddi_get_parent(dip) == ddi_root_node()); 19597c478bd9Sstevel@tonic-gate ndi_rele_devi(dip); /* held from ddi_find_devinfo */ 19607c478bd9Sstevel@tonic-gate (void) ddi_remove_child(dip, 0); 19617c478bd9Sstevel@tonic-gate } 19627c478bd9Sstevel@tonic-gate 19637c478bd9Sstevel@tonic-gate PRM_POINT("Releasing boot pages"); 19647c478bd9Sstevel@tonic-gate while (bootpages) { 19657c478bd9Sstevel@tonic-gate pp = bootpages; 19667c478bd9Sstevel@tonic-gate bootpages = pp->p_next; 19677c478bd9Sstevel@tonic-gate if (root_is_ramdisk && pp_in_ramdisk(pp)) { 19687c478bd9Sstevel@tonic-gate pp->p_next = rd_pages; 19697c478bd9Sstevel@tonic-gate rd_pages = pp; 19707c478bd9Sstevel@tonic-gate continue; 19717c478bd9Sstevel@tonic-gate } 19727c478bd9Sstevel@tonic-gate pp->p_next = (struct page *)0; 1973ae115bc7Smrj pp->p_prev = (struct page *)0; 1974ae115bc7Smrj PP_CLRBOOTPAGES(pp); 19757c478bd9Sstevel@tonic-gate page_free(pp, 1); 19767c478bd9Sstevel@tonic-gate } 1977ae115bc7Smrj PRM_POINT("Boot pages released"); 19787c478bd9Sstevel@tonic-gate 19797c478bd9Sstevel@tonic-gate /* 19807c478bd9Sstevel@tonic-gate * Find 1 page below 1 MB so that other processors can boot up. 19817c478bd9Sstevel@tonic-gate * Make sure it has a kernel VA as well as a 1:1 mapping. 19827c478bd9Sstevel@tonic-gate * We should have just free'd one up. 19837c478bd9Sstevel@tonic-gate */ 19847c478bd9Sstevel@tonic-gate if (use_mp) { 1985ae115bc7Smrj pfn_t pfn; 1986ae115bc7Smrj 19877c478bd9Sstevel@tonic-gate for (pfn = 1; pfn < btop(1*1024*1024); pfn++) { 19887c478bd9Sstevel@tonic-gate if (page_numtopp_alloc(pfn) == NULL) 19897c478bd9Sstevel@tonic-gate continue; 19907c478bd9Sstevel@tonic-gate rm_platter_va = i86devmap(pfn, 1, 19917c478bd9Sstevel@tonic-gate PROT_READ | PROT_WRITE | PROT_EXEC); 19927c478bd9Sstevel@tonic-gate rm_platter_pa = ptob(pfn); 19937c478bd9Sstevel@tonic-gate hat_devload(kas.a_hat, 19947c478bd9Sstevel@tonic-gate (caddr_t)(uintptr_t)rm_platter_pa, MMU_PAGESIZE, 19957c478bd9Sstevel@tonic-gate pfn, PROT_READ | PROT_WRITE | PROT_EXEC, 19967c478bd9Sstevel@tonic-gate HAT_LOAD_NOCONSIST); 19977c478bd9Sstevel@tonic-gate break; 19987c478bd9Sstevel@tonic-gate } 19997c478bd9Sstevel@tonic-gate if (pfn == btop(1*1024*1024)) 20007c478bd9Sstevel@tonic-gate panic("No page available for starting " 20017c478bd9Sstevel@tonic-gate "other processors"); 20027c478bd9Sstevel@tonic-gate } 20037c478bd9Sstevel@tonic-gate 20047c478bd9Sstevel@tonic-gate } 20057c478bd9Sstevel@tonic-gate 20067c478bd9Sstevel@tonic-gate /* 20077c478bd9Sstevel@tonic-gate * Initialize the platform-specific parts of a page_t. 20087c478bd9Sstevel@tonic-gate */ 20097c478bd9Sstevel@tonic-gate void 20107c478bd9Sstevel@tonic-gate add_physmem_cb(page_t *pp, pfn_t pnum) 20117c478bd9Sstevel@tonic-gate { 20127c478bd9Sstevel@tonic-gate pp->p_pagenum = pnum; 20137c478bd9Sstevel@tonic-gate pp->p_mapping = NULL; 20147c478bd9Sstevel@tonic-gate pp->p_embed = 0; 20157c478bd9Sstevel@tonic-gate pp->p_share = 0; 20167c478bd9Sstevel@tonic-gate pp->p_mlentry = 0; 20177c478bd9Sstevel@tonic-gate } 20187c478bd9Sstevel@tonic-gate 20197c478bd9Sstevel@tonic-gate /* 20207c478bd9Sstevel@tonic-gate * kphysm_init() initializes physical memory. 20217c478bd9Sstevel@tonic-gate */ 20227c478bd9Sstevel@tonic-gate static pgcnt_t 20237c478bd9Sstevel@tonic-gate kphysm_init( 2024ae115bc7Smrj page_t *pp, 20257c478bd9Sstevel@tonic-gate pgcnt_t npages) 20267c478bd9Sstevel@tonic-gate { 20277c478bd9Sstevel@tonic-gate struct memlist *pmem; 20287c478bd9Sstevel@tonic-gate struct memseg *cur_memseg; 20297c478bd9Sstevel@tonic-gate pfn_t base_pfn; 20307c478bd9Sstevel@tonic-gate pgcnt_t num; 20317c478bd9Sstevel@tonic-gate pgcnt_t pages_done = 0; 20327c478bd9Sstevel@tonic-gate uint64_t addr; 20337c478bd9Sstevel@tonic-gate uint64_t size; 20347c478bd9Sstevel@tonic-gate extern pfn_t ddiphysmin; 20357c478bd9Sstevel@tonic-gate 20367c478bd9Sstevel@tonic-gate ASSERT(page_hash != NULL && page_hashsz != 0); 20377c478bd9Sstevel@tonic-gate 2038ae115bc7Smrj cur_memseg = memseg_base; 20397c478bd9Sstevel@tonic-gate for (pmem = phys_avail; pmem && npages; pmem = pmem->next) { 20407c478bd9Sstevel@tonic-gate /* 20417c478bd9Sstevel@tonic-gate * In a 32 bit kernel can't use higher memory if we're 20427c478bd9Sstevel@tonic-gate * not booting in PAE mode. This check takes care of that. 20437c478bd9Sstevel@tonic-gate */ 20447c478bd9Sstevel@tonic-gate addr = pmem->address; 20457c478bd9Sstevel@tonic-gate size = pmem->size; 20467c478bd9Sstevel@tonic-gate if (btop(addr) > physmax) 20477c478bd9Sstevel@tonic-gate continue; 20487c478bd9Sstevel@tonic-gate 20497c478bd9Sstevel@tonic-gate /* 20507c478bd9Sstevel@tonic-gate * align addr and size - they may not be at page boundaries 20517c478bd9Sstevel@tonic-gate */ 20527c478bd9Sstevel@tonic-gate if ((addr & MMU_PAGEOFFSET) != 0) { 20537c478bd9Sstevel@tonic-gate addr += MMU_PAGEOFFSET; 20547c478bd9Sstevel@tonic-gate addr &= ~(uint64_t)MMU_PAGEOFFSET; 20557c478bd9Sstevel@tonic-gate size -= addr - pmem->address; 20567c478bd9Sstevel@tonic-gate } 20577c478bd9Sstevel@tonic-gate 20586bb54764Skchow /* only process pages below or equal to physmax */ 20596bb54764Skchow if ((btop(addr + size) - 1) > physmax) 20606bb54764Skchow size = ptob(physmax - btop(addr) + 1); 20617c478bd9Sstevel@tonic-gate 20627c478bd9Sstevel@tonic-gate num = btop(size); 20637c478bd9Sstevel@tonic-gate if (num == 0) 20647c478bd9Sstevel@tonic-gate continue; 20657c478bd9Sstevel@tonic-gate 20667c478bd9Sstevel@tonic-gate if (num > npages) 20677c478bd9Sstevel@tonic-gate num = npages; 20687c478bd9Sstevel@tonic-gate 20697c478bd9Sstevel@tonic-gate npages -= num; 20707c478bd9Sstevel@tonic-gate pages_done += num; 20717c478bd9Sstevel@tonic-gate base_pfn = btop(addr); 20727c478bd9Sstevel@tonic-gate 20737c478bd9Sstevel@tonic-gate if (prom_debug) 20747c478bd9Sstevel@tonic-gate prom_printf("MEMSEG addr=0x%" PRIx64 20757c478bd9Sstevel@tonic-gate " pgs=0x%lx pfn 0x%lx-0x%lx\n", 20767c478bd9Sstevel@tonic-gate addr, num, base_pfn, base_pfn + num); 20777c478bd9Sstevel@tonic-gate 20787c478bd9Sstevel@tonic-gate /* 2079ae115bc7Smrj * Ignore pages below ddiphysmin to simplify ddi memory 20807c478bd9Sstevel@tonic-gate * allocation with non-zero addr_lo requests. 20817c478bd9Sstevel@tonic-gate */ 20827c478bd9Sstevel@tonic-gate if (base_pfn < ddiphysmin) { 2083ae115bc7Smrj if (base_pfn + num <= ddiphysmin) 20847c478bd9Sstevel@tonic-gate continue; 20857c478bd9Sstevel@tonic-gate pp += (ddiphysmin - base_pfn); 20867c478bd9Sstevel@tonic-gate num -= (ddiphysmin - base_pfn); 20877c478bd9Sstevel@tonic-gate base_pfn = ddiphysmin; 20887c478bd9Sstevel@tonic-gate } 2089ae115bc7Smrj 20907c478bd9Sstevel@tonic-gate /* 20917c478bd9Sstevel@tonic-gate * Build the memsegs entry 20927c478bd9Sstevel@tonic-gate */ 20937c478bd9Sstevel@tonic-gate cur_memseg->pages = pp; 20947c478bd9Sstevel@tonic-gate cur_memseg->epages = pp + num; 20957c478bd9Sstevel@tonic-gate cur_memseg->pages_base = base_pfn; 20967c478bd9Sstevel@tonic-gate cur_memseg->pages_end = base_pfn + num; 20977c478bd9Sstevel@tonic-gate 20987c478bd9Sstevel@tonic-gate /* 2099ae115bc7Smrj * Insert into memseg list in decreasing pfn range order. 21007c478bd9Sstevel@tonic-gate * Low memory is typically more fragmented such that this 21017c478bd9Sstevel@tonic-gate * ordering keeps the larger ranges at the front of the list 21027c478bd9Sstevel@tonic-gate * for code that searches memseg. 2103ae115bc7Smrj * This ASSERTS that the memsegs coming in from boot are in 2104ae115bc7Smrj * increasing physical address order and not contiguous. 21057c478bd9Sstevel@tonic-gate */ 2106ae115bc7Smrj if (memsegs != NULL) { 2107ae115bc7Smrj ASSERT(cur_memseg->pages_base >= memsegs->pages_end); 2108ae115bc7Smrj cur_memseg->next = memsegs; 21097c478bd9Sstevel@tonic-gate } 2110ae115bc7Smrj memsegs = cur_memseg; 21117c478bd9Sstevel@tonic-gate 21127c478bd9Sstevel@tonic-gate /* 21137c478bd9Sstevel@tonic-gate * add_physmem() initializes the PSM part of the page 21147c478bd9Sstevel@tonic-gate * struct by calling the PSM back with add_physmem_cb(). 21157c478bd9Sstevel@tonic-gate * In addition it coalesces pages into larger pages as 21167c478bd9Sstevel@tonic-gate * it initializes them. 21177c478bd9Sstevel@tonic-gate */ 21187c478bd9Sstevel@tonic-gate add_physmem(pp, num, base_pfn); 21197c478bd9Sstevel@tonic-gate cur_memseg++; 21207c478bd9Sstevel@tonic-gate availrmem_initial += num; 21217c478bd9Sstevel@tonic-gate availrmem += num; 21227c478bd9Sstevel@tonic-gate 2123ae115bc7Smrj pp += num; 21247c478bd9Sstevel@tonic-gate } 21257c478bd9Sstevel@tonic-gate 21267c478bd9Sstevel@tonic-gate PRM_DEBUG(availrmem_initial); 21277c478bd9Sstevel@tonic-gate PRM_DEBUG(availrmem); 21287c478bd9Sstevel@tonic-gate PRM_DEBUG(freemem); 21297c478bd9Sstevel@tonic-gate build_pfn_hash(); 21307c478bd9Sstevel@tonic-gate return (pages_done); 21317c478bd9Sstevel@tonic-gate } 21327c478bd9Sstevel@tonic-gate 21337c478bd9Sstevel@tonic-gate /* 21347c478bd9Sstevel@tonic-gate * Kernel VM initialization. 21357c478bd9Sstevel@tonic-gate */ 21367c478bd9Sstevel@tonic-gate static void 21377c478bd9Sstevel@tonic-gate kvm_init(void) 21387c478bd9Sstevel@tonic-gate { 21397c478bd9Sstevel@tonic-gate ASSERT((((uintptr_t)s_text) & MMU_PAGEOFFSET) == 0); 21407c478bd9Sstevel@tonic-gate 21417c478bd9Sstevel@tonic-gate /* 21427c478bd9Sstevel@tonic-gate * Put the kernel segments in kernel address space. 21437c478bd9Sstevel@tonic-gate */ 21447c478bd9Sstevel@tonic-gate rw_enter(&kas.a_lock, RW_WRITER); 21457c478bd9Sstevel@tonic-gate as_avlinit(&kas); 21467c478bd9Sstevel@tonic-gate 21477c478bd9Sstevel@tonic-gate (void) seg_attach(&kas, s_text, e_moddata - s_text, &ktextseg); 21487c478bd9Sstevel@tonic-gate (void) segkmem_create(&ktextseg); 21497c478bd9Sstevel@tonic-gate 21507c478bd9Sstevel@tonic-gate (void) seg_attach(&kas, (caddr_t)valloc_base, valloc_sz, &kvalloc); 21517c478bd9Sstevel@tonic-gate (void) segkmem_create(&kvalloc); 21527c478bd9Sstevel@tonic-gate 2153*35b1ab99Sjosephb (void) seg_attach(&kas, kernelheap, 2154*35b1ab99Sjosephb ekernelheap - kernelheap, &kvseg); 21557c478bd9Sstevel@tonic-gate (void) segkmem_create(&kvseg); 21567c478bd9Sstevel@tonic-gate 2157ae115bc7Smrj if (core_size > 0) { 2158ae115bc7Smrj PRM_POINT("attaching kvseg_core"); 2159ae115bc7Smrj (void) seg_attach(&kas, (caddr_t)core_base, core_size, 2160ae115bc7Smrj &kvseg_core); 2161ae115bc7Smrj (void) segkmem_create(&kvseg_core); 2162ae115bc7Smrj } 2163ad23a2dbSjohansen 2164ae115bc7Smrj if (segziosize > 0) { 2165ae115bc7Smrj PRM_POINT("attaching segzio"); 2166ad23a2dbSjohansen (void) seg_attach(&kas, segzio_base, mmu_ptob(segziosize), 2167ad23a2dbSjohansen &kzioseg); 2168ad23a2dbSjohansen (void) segkmem_zio_create(&kzioseg); 2169ad23a2dbSjohansen 2170ad23a2dbSjohansen /* create zio area covering new segment */ 2171ad23a2dbSjohansen segkmem_zio_init(segzio_base, mmu_ptob(segziosize)); 2172ad23a2dbSjohansen } 21737c478bd9Sstevel@tonic-gate 2174ae115bc7Smrj (void) seg_attach(&kas, kdi_segdebugbase, kdi_segdebugsize, &kdebugseg); 21757c478bd9Sstevel@tonic-gate (void) segkmem_create(&kdebugseg); 21767c478bd9Sstevel@tonic-gate 21777c478bd9Sstevel@tonic-gate rw_exit(&kas.a_lock); 21787c478bd9Sstevel@tonic-gate 21797c478bd9Sstevel@tonic-gate /* 21807c478bd9Sstevel@tonic-gate * Ensure that the red zone at kernelbase is never accessible. 21817c478bd9Sstevel@tonic-gate */ 2182ae115bc7Smrj PRM_POINT("protecting redzone"); 21837c478bd9Sstevel@tonic-gate (void) as_setprot(&kas, (caddr_t)kernelbase, KERNEL_REDZONE_SIZE, 0); 21847c478bd9Sstevel@tonic-gate 21857c478bd9Sstevel@tonic-gate /* 21867c478bd9Sstevel@tonic-gate * Make the text writable so that it can be hot patched by DTrace. 21877c478bd9Sstevel@tonic-gate */ 21887c478bd9Sstevel@tonic-gate (void) as_setprot(&kas, s_text, e_modtext - s_text, 21897c478bd9Sstevel@tonic-gate PROT_READ | PROT_WRITE | PROT_EXEC); 21907c478bd9Sstevel@tonic-gate 21917c478bd9Sstevel@tonic-gate /* 21927c478bd9Sstevel@tonic-gate * Make data writable until end. 21937c478bd9Sstevel@tonic-gate */ 21947c478bd9Sstevel@tonic-gate (void) as_setprot(&kas, s_data, e_moddata - s_data, 21957c478bd9Sstevel@tonic-gate PROT_READ | PROT_WRITE | PROT_EXEC); 21967c478bd9Sstevel@tonic-gate } 21977c478bd9Sstevel@tonic-gate 21987c478bd9Sstevel@tonic-gate /* 21997c478bd9Sstevel@tonic-gate * These are MTTR registers supported by P6 22007c478bd9Sstevel@tonic-gate */ 22017c478bd9Sstevel@tonic-gate static struct mtrrvar mtrrphys_arr[MAX_MTRRVAR]; 22027c478bd9Sstevel@tonic-gate static uint64_t mtrr64k, mtrr16k1, mtrr16k2; 22037c478bd9Sstevel@tonic-gate static uint64_t mtrr4k1, mtrr4k2, mtrr4k3; 22047c478bd9Sstevel@tonic-gate static uint64_t mtrr4k4, mtrr4k5, mtrr4k6; 22057c478bd9Sstevel@tonic-gate static uint64_t mtrr4k7, mtrr4k8, mtrrcap; 22067c478bd9Sstevel@tonic-gate uint64_t mtrrdef, pat_attr_reg; 22077c478bd9Sstevel@tonic-gate 22087c478bd9Sstevel@tonic-gate /* 22097c478bd9Sstevel@tonic-gate * Disable reprogramming of MTRRs by default. 22107c478bd9Sstevel@tonic-gate */ 22117c478bd9Sstevel@tonic-gate int enable_relaxed_mtrr = 0; 22127c478bd9Sstevel@tonic-gate 22137c478bd9Sstevel@tonic-gate void 221445916cd2Sjpk setup_mtrr(void) 22157c478bd9Sstevel@tonic-gate { 22167c478bd9Sstevel@tonic-gate int i, ecx; 22177c478bd9Sstevel@tonic-gate int vcnt; 22187c478bd9Sstevel@tonic-gate struct mtrrvar *mtrrphys; 22197c478bd9Sstevel@tonic-gate 22207c478bd9Sstevel@tonic-gate if (!(x86_feature & X86_MTRR)) 22217c478bd9Sstevel@tonic-gate return; 22227c478bd9Sstevel@tonic-gate 22230ac7d7d8Skucharsk mtrrcap = rdmsr(REG_MTRRCAP); 22240ac7d7d8Skucharsk mtrrdef = rdmsr(REG_MTRRDEF); 22257c478bd9Sstevel@tonic-gate if (mtrrcap & MTRRCAP_FIX) { 22260ac7d7d8Skucharsk mtrr64k = rdmsr(REG_MTRR64K); 22270ac7d7d8Skucharsk mtrr16k1 = rdmsr(REG_MTRR16K1); 22280ac7d7d8Skucharsk mtrr16k2 = rdmsr(REG_MTRR16K2); 22290ac7d7d8Skucharsk mtrr4k1 = rdmsr(REG_MTRR4K1); 22300ac7d7d8Skucharsk mtrr4k2 = rdmsr(REG_MTRR4K2); 22310ac7d7d8Skucharsk mtrr4k3 = rdmsr(REG_MTRR4K3); 22320ac7d7d8Skucharsk mtrr4k4 = rdmsr(REG_MTRR4K4); 22330ac7d7d8Skucharsk mtrr4k5 = rdmsr(REG_MTRR4K5); 22340ac7d7d8Skucharsk mtrr4k6 = rdmsr(REG_MTRR4K6); 22350ac7d7d8Skucharsk mtrr4k7 = rdmsr(REG_MTRR4K7); 22360ac7d7d8Skucharsk mtrr4k8 = rdmsr(REG_MTRR4K8); 22377c478bd9Sstevel@tonic-gate } 22387c478bd9Sstevel@tonic-gate if ((vcnt = (mtrrcap & MTRRCAP_VCNTMASK)) > MAX_MTRRVAR) 22397c478bd9Sstevel@tonic-gate vcnt = MAX_MTRRVAR; 22407c478bd9Sstevel@tonic-gate 22417c478bd9Sstevel@tonic-gate for (i = 0, ecx = REG_MTRRPHYSBASE0, mtrrphys = mtrrphys_arr; 2242*35b1ab99Sjosephb i < vcnt - 1; i++, ecx += 2, mtrrphys++) { 22430ac7d7d8Skucharsk mtrrphys->mtrrphys_base = rdmsr(ecx); 22440ac7d7d8Skucharsk mtrrphys->mtrrphys_mask = rdmsr(ecx + 1); 2245*35b1ab99Sjosephb if ((x86_feature & X86_PAT) && enable_relaxed_mtrr) 22467c478bd9Sstevel@tonic-gate mtrrphys->mtrrphys_mask &= ~MTRRPHYSMASK_V; 22477c478bd9Sstevel@tonic-gate } 22487c478bd9Sstevel@tonic-gate if (x86_feature & X86_PAT) { 22497c478bd9Sstevel@tonic-gate if (enable_relaxed_mtrr) 22507c478bd9Sstevel@tonic-gate mtrrdef = MTRR_TYPE_WB|MTRRDEF_FE|MTRRDEF_E; 22517c478bd9Sstevel@tonic-gate pat_attr_reg = PAT_DEFAULT_ATTRIBUTE; 22527c478bd9Sstevel@tonic-gate } 22537c478bd9Sstevel@tonic-gate 22547c478bd9Sstevel@tonic-gate mtrr_sync(); 22557c478bd9Sstevel@tonic-gate } 22567c478bd9Sstevel@tonic-gate 22577c478bd9Sstevel@tonic-gate /* 22587c478bd9Sstevel@tonic-gate * Sync current cpu mtrr with the incore copy of mtrr. 22597c478bd9Sstevel@tonic-gate * This function has to be invoked with interrupts disabled 22607c478bd9Sstevel@tonic-gate * Currently we do not capture other cpu's. This is invoked on cpu0 22617c478bd9Sstevel@tonic-gate * just after reading /etc/system. 22627c478bd9Sstevel@tonic-gate * On other cpu's its invoked from mp_startup(). 22637c478bd9Sstevel@tonic-gate */ 22647c478bd9Sstevel@tonic-gate void 226545916cd2Sjpk mtrr_sync(void) 22667c478bd9Sstevel@tonic-gate { 22677c478bd9Sstevel@tonic-gate uint_t crvalue, cr0_orig; 22687c478bd9Sstevel@tonic-gate int vcnt, i, ecx; 22697c478bd9Sstevel@tonic-gate struct mtrrvar *mtrrphys; 22707c478bd9Sstevel@tonic-gate 22717c478bd9Sstevel@tonic-gate cr0_orig = crvalue = getcr0(); 22727c478bd9Sstevel@tonic-gate crvalue |= CR0_CD; 22737c478bd9Sstevel@tonic-gate crvalue &= ~CR0_NW; 22747c478bd9Sstevel@tonic-gate setcr0(crvalue); 22757c478bd9Sstevel@tonic-gate invalidate_cache(); 22767c478bd9Sstevel@tonic-gate 2277ae115bc7Smrj reload_cr3(); 22780ac7d7d8Skucharsk if (x86_feature & X86_PAT) 22790ac7d7d8Skucharsk wrmsr(REG_MTRRPAT, pat_attr_reg); 22800ac7d7d8Skucharsk 22810ac7d7d8Skucharsk wrmsr(REG_MTRRDEF, rdmsr(REG_MTRRDEF) & 22820ac7d7d8Skucharsk ~((uint64_t)(uintptr_t)MTRRDEF_E)); 22830ac7d7d8Skucharsk 22847c478bd9Sstevel@tonic-gate if (mtrrcap & MTRRCAP_FIX) { 22850ac7d7d8Skucharsk wrmsr(REG_MTRR64K, mtrr64k); 22860ac7d7d8Skucharsk wrmsr(REG_MTRR16K1, mtrr16k1); 22870ac7d7d8Skucharsk wrmsr(REG_MTRR16K2, mtrr16k2); 22880ac7d7d8Skucharsk wrmsr(REG_MTRR4K1, mtrr4k1); 22890ac7d7d8Skucharsk wrmsr(REG_MTRR4K2, mtrr4k2); 22900ac7d7d8Skucharsk wrmsr(REG_MTRR4K3, mtrr4k3); 22910ac7d7d8Skucharsk wrmsr(REG_MTRR4K4, mtrr4k4); 22920ac7d7d8Skucharsk wrmsr(REG_MTRR4K5, mtrr4k5); 22930ac7d7d8Skucharsk wrmsr(REG_MTRR4K6, mtrr4k6); 22940ac7d7d8Skucharsk wrmsr(REG_MTRR4K7, mtrr4k7); 22950ac7d7d8Skucharsk wrmsr(REG_MTRR4K8, mtrr4k8); 22967c478bd9Sstevel@tonic-gate } 22977c478bd9Sstevel@tonic-gate if ((vcnt = (mtrrcap & MTRRCAP_VCNTMASK)) > MAX_MTRRVAR) 22987c478bd9Sstevel@tonic-gate vcnt = MAX_MTRRVAR; 22997c478bd9Sstevel@tonic-gate for (i = 0, ecx = REG_MTRRPHYSBASE0, mtrrphys = mtrrphys_arr; 23000ac7d7d8Skucharsk i < vcnt - 1; i++, ecx += 2, mtrrphys++) { 23010ac7d7d8Skucharsk wrmsr(ecx, mtrrphys->mtrrphys_base); 23020ac7d7d8Skucharsk wrmsr(ecx + 1, mtrrphys->mtrrphys_mask); 23037c478bd9Sstevel@tonic-gate } 23040ac7d7d8Skucharsk wrmsr(REG_MTRRDEF, mtrrdef); 2305ae115bc7Smrj 2306ae115bc7Smrj reload_cr3(); 23077c478bd9Sstevel@tonic-gate invalidate_cache(); 23087c478bd9Sstevel@tonic-gate setcr0(cr0_orig); 23097c478bd9Sstevel@tonic-gate } 23107c478bd9Sstevel@tonic-gate 23117c478bd9Sstevel@tonic-gate /* 23127c478bd9Sstevel@tonic-gate * resync mtrr so that BIOS is happy. Called from mdboot 23137c478bd9Sstevel@tonic-gate */ 23147c478bd9Sstevel@tonic-gate void 231545916cd2Sjpk mtrr_resync(void) 23167c478bd9Sstevel@tonic-gate { 23177c478bd9Sstevel@tonic-gate if ((x86_feature & X86_PAT) && enable_relaxed_mtrr) { 23187c478bd9Sstevel@tonic-gate /* 23197c478bd9Sstevel@tonic-gate * We could have changed the default mtrr definition. 23207c478bd9Sstevel@tonic-gate * Put it back to uncached which is what it is at power on 23217c478bd9Sstevel@tonic-gate */ 23227c478bd9Sstevel@tonic-gate mtrrdef = MTRR_TYPE_UC|MTRRDEF_FE|MTRRDEF_E; 23237c478bd9Sstevel@tonic-gate mtrr_sync(); 23247c478bd9Sstevel@tonic-gate } 23257c478bd9Sstevel@tonic-gate } 23267c478bd9Sstevel@tonic-gate 23277c478bd9Sstevel@tonic-gate void 232845916cd2Sjpk get_system_configuration(void) 23297c478bd9Sstevel@tonic-gate { 23307c478bd9Sstevel@tonic-gate char prop[32]; 23317c478bd9Sstevel@tonic-gate u_longlong_t nodes_ll, cpus_pernode_ll, lvalue; 23327c478bd9Sstevel@tonic-gate 2333*35b1ab99Sjosephb if (BOP_GETPROPLEN(bootops, "nodes") > sizeof (prop) || 2334*35b1ab99Sjosephb BOP_GETPROP(bootops, "nodes", prop) < 0 || 2335*35b1ab99Sjosephb kobj_getvalue(prop, &nodes_ll) == -1 || 2336*35b1ab99Sjosephb nodes_ll > MAXNODES || 2337*35b1ab99Sjosephb BOP_GETPROPLEN(bootops, "cpus_pernode") > sizeof (prop) || 2338*35b1ab99Sjosephb BOP_GETPROP(bootops, "cpus_pernode", prop) < 0 || 2339*35b1ab99Sjosephb kobj_getvalue(prop, &cpus_pernode_ll) == -1) { 23407c478bd9Sstevel@tonic-gate system_hardware.hd_nodes = 1; 23417c478bd9Sstevel@tonic-gate system_hardware.hd_cpus_per_node = 0; 23427c478bd9Sstevel@tonic-gate } else { 23437c478bd9Sstevel@tonic-gate system_hardware.hd_nodes = (int)nodes_ll; 23447c478bd9Sstevel@tonic-gate system_hardware.hd_cpus_per_node = (int)cpus_pernode_ll; 23457c478bd9Sstevel@tonic-gate } 2346*35b1ab99Sjosephb 2347*35b1ab99Sjosephb if (BOP_GETPROPLEN(bootops, "kernelbase") > sizeof (prop) || 2348*35b1ab99Sjosephb BOP_GETPROP(bootops, "kernelbase", prop) < 0 || 2349*35b1ab99Sjosephb kobj_getvalue(prop, &lvalue) == -1) 2350*35b1ab99Sjosephb eprom_kernelbase = NULL; 23517c478bd9Sstevel@tonic-gate else 2352*35b1ab99Sjosephb eprom_kernelbase = (uintptr_t)lvalue; 23537c478bd9Sstevel@tonic-gate 2354*35b1ab99Sjosephb if (BOP_GETPROPLEN(bootops, "segmapsize") > sizeof (prop) || 2355*35b1ab99Sjosephb BOP_GETPROP(bootops, "segmapsize", prop) < 0 || 2356*35b1ab99Sjosephb kobj_getvalue(prop, &lvalue) == -1) 23577c478bd9Sstevel@tonic-gate segmapsize = SEGMAPDEFAULT; 2358*35b1ab99Sjosephb else 23597c478bd9Sstevel@tonic-gate segmapsize = (uintptr_t)lvalue; 23607c478bd9Sstevel@tonic-gate 2361*35b1ab99Sjosephb if (BOP_GETPROPLEN(bootops, "segmapfreelists") > sizeof (prop) || 2362*35b1ab99Sjosephb BOP_GETPROP(bootops, "segmapfreelists", prop) < 0 || 2363*35b1ab99Sjosephb kobj_getvalue(prop, &lvalue) == -1) 23647c478bd9Sstevel@tonic-gate segmapfreelists = 0; /* use segmap driver default */ 2365*35b1ab99Sjosephb else 23667c478bd9Sstevel@tonic-gate segmapfreelists = (int)lvalue; 23674944b02eSkchow 2368aac11643Sjosephb /* physmem used to be here, but moved much earlier to fakebop.c */ 23697c478bd9Sstevel@tonic-gate } 23707c478bd9Sstevel@tonic-gate 23717c478bd9Sstevel@tonic-gate /* 23727c478bd9Sstevel@tonic-gate * Add to a memory list. 23737c478bd9Sstevel@tonic-gate * start = start of new memory segment 23747c478bd9Sstevel@tonic-gate * len = length of new memory segment in bytes 23757c478bd9Sstevel@tonic-gate * new = pointer to a new struct memlist 23767c478bd9Sstevel@tonic-gate * memlistp = memory list to which to add segment. 23777c478bd9Sstevel@tonic-gate */ 2378ae115bc7Smrj void 23797c478bd9Sstevel@tonic-gate memlist_add( 23807c478bd9Sstevel@tonic-gate uint64_t start, 23817c478bd9Sstevel@tonic-gate uint64_t len, 23827c478bd9Sstevel@tonic-gate struct memlist *new, 23837c478bd9Sstevel@tonic-gate struct memlist **memlistp) 23847c478bd9Sstevel@tonic-gate { 23857c478bd9Sstevel@tonic-gate struct memlist *cur; 23867c478bd9Sstevel@tonic-gate uint64_t end = start + len; 23877c478bd9Sstevel@tonic-gate 23887c478bd9Sstevel@tonic-gate new->address = start; 23897c478bd9Sstevel@tonic-gate new->size = len; 23907c478bd9Sstevel@tonic-gate 23917c478bd9Sstevel@tonic-gate cur = *memlistp; 23927c478bd9Sstevel@tonic-gate 23937c478bd9Sstevel@tonic-gate while (cur) { 23947c478bd9Sstevel@tonic-gate if (cur->address >= end) { 23957c478bd9Sstevel@tonic-gate new->next = cur; 23967c478bd9Sstevel@tonic-gate *memlistp = new; 23977c478bd9Sstevel@tonic-gate new->prev = cur->prev; 23987c478bd9Sstevel@tonic-gate cur->prev = new; 23997c478bd9Sstevel@tonic-gate return; 24007c478bd9Sstevel@tonic-gate } 24017c478bd9Sstevel@tonic-gate ASSERT(cur->address + cur->size <= start); 24027c478bd9Sstevel@tonic-gate if (cur->next == NULL) { 24037c478bd9Sstevel@tonic-gate cur->next = new; 24047c478bd9Sstevel@tonic-gate new->prev = cur; 24057c478bd9Sstevel@tonic-gate new->next = NULL; 24067c478bd9Sstevel@tonic-gate return; 24077c478bd9Sstevel@tonic-gate } 24087c478bd9Sstevel@tonic-gate memlistp = &cur->next; 24097c478bd9Sstevel@tonic-gate cur = cur->next; 24107c478bd9Sstevel@tonic-gate } 24117c478bd9Sstevel@tonic-gate } 24127c478bd9Sstevel@tonic-gate 24137c478bd9Sstevel@tonic-gate void 24147c478bd9Sstevel@tonic-gate kobj_vmem_init(vmem_t **text_arena, vmem_t **data_arena) 24157c478bd9Sstevel@tonic-gate { 24167c478bd9Sstevel@tonic-gate size_t tsize = e_modtext - modtext; 24177c478bd9Sstevel@tonic-gate size_t dsize = e_moddata - moddata; 24187c478bd9Sstevel@tonic-gate 24197c478bd9Sstevel@tonic-gate *text_arena = vmem_create("module_text", tsize ? modtext : NULL, tsize, 24207c478bd9Sstevel@tonic-gate 1, segkmem_alloc, segkmem_free, heaptext_arena, 0, VM_SLEEP); 24217c478bd9Sstevel@tonic-gate *data_arena = vmem_create("module_data", dsize ? moddata : NULL, dsize, 24227c478bd9Sstevel@tonic-gate 1, segkmem_alloc, segkmem_free, heap32_arena, 0, VM_SLEEP); 24237c478bd9Sstevel@tonic-gate } 24247c478bd9Sstevel@tonic-gate 24257c478bd9Sstevel@tonic-gate caddr_t 24267c478bd9Sstevel@tonic-gate kobj_text_alloc(vmem_t *arena, size_t size) 24277c478bd9Sstevel@tonic-gate { 24287c478bd9Sstevel@tonic-gate return (vmem_alloc(arena, size, VM_SLEEP | VM_BESTFIT)); 24297c478bd9Sstevel@tonic-gate } 24307c478bd9Sstevel@tonic-gate 24317c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 24327c478bd9Sstevel@tonic-gate caddr_t 24337c478bd9Sstevel@tonic-gate kobj_texthole_alloc(caddr_t addr, size_t size) 24347c478bd9Sstevel@tonic-gate { 24357c478bd9Sstevel@tonic-gate panic("unexpected call to kobj_texthole_alloc()"); 24367c478bd9Sstevel@tonic-gate /*NOTREACHED*/ 24377c478bd9Sstevel@tonic-gate return (0); 24387c478bd9Sstevel@tonic-gate } 24397c478bd9Sstevel@tonic-gate 24407c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 24417c478bd9Sstevel@tonic-gate void 24427c478bd9Sstevel@tonic-gate kobj_texthole_free(caddr_t addr, size_t size) 24437c478bd9Sstevel@tonic-gate { 24447c478bd9Sstevel@tonic-gate panic("unexpected call to kobj_texthole_free()"); 24457c478bd9Sstevel@tonic-gate } 24467c478bd9Sstevel@tonic-gate 24477c478bd9Sstevel@tonic-gate /* 24487c478bd9Sstevel@tonic-gate * This is called just after configure() in startup(). 24497c478bd9Sstevel@tonic-gate * 24507c478bd9Sstevel@tonic-gate * The ISALIST concept is a bit hopeless on Intel, because 24517c478bd9Sstevel@tonic-gate * there's no guarantee of an ever-more-capable processor 24527c478bd9Sstevel@tonic-gate * given that various parts of the instruction set may appear 24537c478bd9Sstevel@tonic-gate * and disappear between different implementations. 24547c478bd9Sstevel@tonic-gate * 24557c478bd9Sstevel@tonic-gate * While it would be possible to correct it and even enhance 24567c478bd9Sstevel@tonic-gate * it somewhat, the explicit hardware capability bitmask allows 24577c478bd9Sstevel@tonic-gate * more flexibility. 24587c478bd9Sstevel@tonic-gate * 24597c478bd9Sstevel@tonic-gate * So, we just leave this alone. 24607c478bd9Sstevel@tonic-gate */ 24617c478bd9Sstevel@tonic-gate void 24627c478bd9Sstevel@tonic-gate setx86isalist(void) 24637c478bd9Sstevel@tonic-gate { 24647c478bd9Sstevel@tonic-gate char *tp; 24657c478bd9Sstevel@tonic-gate size_t len; 24667c478bd9Sstevel@tonic-gate extern char *isa_list; 24677c478bd9Sstevel@tonic-gate 24687c478bd9Sstevel@tonic-gate #define TBUFSIZE 1024 24697c478bd9Sstevel@tonic-gate 24707c478bd9Sstevel@tonic-gate tp = kmem_alloc(TBUFSIZE, KM_SLEEP); 24717c478bd9Sstevel@tonic-gate *tp = '\0'; 24727c478bd9Sstevel@tonic-gate 24737c478bd9Sstevel@tonic-gate #if defined(__amd64) 24747c478bd9Sstevel@tonic-gate (void) strcpy(tp, "amd64 "); 24757c478bd9Sstevel@tonic-gate #endif 24767c478bd9Sstevel@tonic-gate 24777c478bd9Sstevel@tonic-gate switch (x86_vendor) { 24787c478bd9Sstevel@tonic-gate case X86_VENDOR_Intel: 24797c478bd9Sstevel@tonic-gate case X86_VENDOR_AMD: 24807c478bd9Sstevel@tonic-gate case X86_VENDOR_TM: 24817c478bd9Sstevel@tonic-gate if (x86_feature & X86_CMOV) { 24827c478bd9Sstevel@tonic-gate /* 24837c478bd9Sstevel@tonic-gate * Pentium Pro or later 24847c478bd9Sstevel@tonic-gate */ 24857c478bd9Sstevel@tonic-gate (void) strcat(tp, "pentium_pro"); 24867c478bd9Sstevel@tonic-gate (void) strcat(tp, x86_feature & X86_MMX ? 24877c478bd9Sstevel@tonic-gate "+mmx pentium_pro " : " "); 24887c478bd9Sstevel@tonic-gate } 24897c478bd9Sstevel@tonic-gate /*FALLTHROUGH*/ 24907c478bd9Sstevel@tonic-gate case X86_VENDOR_Cyrix: 24917c478bd9Sstevel@tonic-gate /* 24927c478bd9Sstevel@tonic-gate * The Cyrix 6x86 does not have any Pentium features 24937c478bd9Sstevel@tonic-gate * accessible while not at privilege level 0. 24947c478bd9Sstevel@tonic-gate */ 24957c478bd9Sstevel@tonic-gate if (x86_feature & X86_CPUID) { 24967c478bd9Sstevel@tonic-gate (void) strcat(tp, "pentium"); 24977c478bd9Sstevel@tonic-gate (void) strcat(tp, x86_feature & X86_MMX ? 24987c478bd9Sstevel@tonic-gate "+mmx pentium " : " "); 24997c478bd9Sstevel@tonic-gate } 25007c478bd9Sstevel@tonic-gate break; 25017c478bd9Sstevel@tonic-gate default: 25027c478bd9Sstevel@tonic-gate break; 25037c478bd9Sstevel@tonic-gate } 25047c478bd9Sstevel@tonic-gate (void) strcat(tp, "i486 i386 i86"); 25057c478bd9Sstevel@tonic-gate len = strlen(tp) + 1; /* account for NULL at end of string */ 25067c478bd9Sstevel@tonic-gate isa_list = strcpy(kmem_alloc(len, KM_SLEEP), tp); 25077c478bd9Sstevel@tonic-gate kmem_free(tp, TBUFSIZE); 25087c478bd9Sstevel@tonic-gate 25097c478bd9Sstevel@tonic-gate #undef TBUFSIZE 25107c478bd9Sstevel@tonic-gate } 25117c478bd9Sstevel@tonic-gate 25127c478bd9Sstevel@tonic-gate 25137c478bd9Sstevel@tonic-gate #ifdef __amd64 25147c478bd9Sstevel@tonic-gate 25157c478bd9Sstevel@tonic-gate void * 25167c478bd9Sstevel@tonic-gate device_arena_alloc(size_t size, int vm_flag) 25177c478bd9Sstevel@tonic-gate { 25187c478bd9Sstevel@tonic-gate return (vmem_alloc(device_arena, size, vm_flag)); 25197c478bd9Sstevel@tonic-gate } 25207c478bd9Sstevel@tonic-gate 25217c478bd9Sstevel@tonic-gate void 25227c478bd9Sstevel@tonic-gate device_arena_free(void *vaddr, size_t size) 25237c478bd9Sstevel@tonic-gate { 25247c478bd9Sstevel@tonic-gate vmem_free(device_arena, vaddr, size); 25257c478bd9Sstevel@tonic-gate } 25267c478bd9Sstevel@tonic-gate 2527ae115bc7Smrj #else /* __i386 */ 25287c478bd9Sstevel@tonic-gate 25297c478bd9Sstevel@tonic-gate void * 25307c478bd9Sstevel@tonic-gate device_arena_alloc(size_t size, int vm_flag) 25317c478bd9Sstevel@tonic-gate { 25327c478bd9Sstevel@tonic-gate caddr_t vaddr; 25337c478bd9Sstevel@tonic-gate uintptr_t v; 25347c478bd9Sstevel@tonic-gate size_t start; 25357c478bd9Sstevel@tonic-gate size_t end; 25367c478bd9Sstevel@tonic-gate 25377c478bd9Sstevel@tonic-gate vaddr = vmem_alloc(heap_arena, size, vm_flag); 25387c478bd9Sstevel@tonic-gate if (vaddr == NULL) 25397c478bd9Sstevel@tonic-gate return (NULL); 25407c478bd9Sstevel@tonic-gate 25417c478bd9Sstevel@tonic-gate v = (uintptr_t)vaddr; 25427c478bd9Sstevel@tonic-gate ASSERT(v >= kernelbase); 2543ae115bc7Smrj ASSERT(v + size <= valloc_base); 25447c478bd9Sstevel@tonic-gate 25457c478bd9Sstevel@tonic-gate start = btop(v - kernelbase); 25467c478bd9Sstevel@tonic-gate end = btop(v + size - 1 - kernelbase); 25477c478bd9Sstevel@tonic-gate ASSERT(start < toxic_bit_map_len); 25487c478bd9Sstevel@tonic-gate ASSERT(end < toxic_bit_map_len); 25497c478bd9Sstevel@tonic-gate 25507c478bd9Sstevel@tonic-gate while (start <= end) { 25517c478bd9Sstevel@tonic-gate BT_ATOMIC_SET(toxic_bit_map, start); 25527c478bd9Sstevel@tonic-gate ++start; 25537c478bd9Sstevel@tonic-gate } 25547c478bd9Sstevel@tonic-gate return (vaddr); 25557c478bd9Sstevel@tonic-gate } 25567c478bd9Sstevel@tonic-gate 25577c478bd9Sstevel@tonic-gate void 25587c478bd9Sstevel@tonic-gate device_arena_free(void *vaddr, size_t size) 25597c478bd9Sstevel@tonic-gate { 25607c478bd9Sstevel@tonic-gate uintptr_t v = (uintptr_t)vaddr; 25617c478bd9Sstevel@tonic-gate size_t start; 25627c478bd9Sstevel@tonic-gate size_t end; 25637c478bd9Sstevel@tonic-gate 25647c478bd9Sstevel@tonic-gate ASSERT(v >= kernelbase); 2565ae115bc7Smrj ASSERT(v + size <= valloc_base); 25667c478bd9Sstevel@tonic-gate 25677c478bd9Sstevel@tonic-gate start = btop(v - kernelbase); 25687c478bd9Sstevel@tonic-gate end = btop(v + size - 1 - kernelbase); 25697c478bd9Sstevel@tonic-gate ASSERT(start < toxic_bit_map_len); 25707c478bd9Sstevel@tonic-gate ASSERT(end < toxic_bit_map_len); 25717c478bd9Sstevel@tonic-gate 25727c478bd9Sstevel@tonic-gate while (start <= end) { 25737c478bd9Sstevel@tonic-gate ASSERT(BT_TEST(toxic_bit_map, start) != 0); 25747c478bd9Sstevel@tonic-gate BT_ATOMIC_CLEAR(toxic_bit_map, start); 25757c478bd9Sstevel@tonic-gate ++start; 25767c478bd9Sstevel@tonic-gate } 25777c478bd9Sstevel@tonic-gate vmem_free(heap_arena, vaddr, size); 25787c478bd9Sstevel@tonic-gate } 25797c478bd9Sstevel@tonic-gate 25807c478bd9Sstevel@tonic-gate /* 25817c478bd9Sstevel@tonic-gate * returns 1st address in range that is in device arena, or NULL 25827c478bd9Sstevel@tonic-gate * if len is not NULL it returns the length of the toxic range 25837c478bd9Sstevel@tonic-gate */ 25847c478bd9Sstevel@tonic-gate void * 25857c478bd9Sstevel@tonic-gate device_arena_contains(void *vaddr, size_t size, size_t *len) 25867c478bd9Sstevel@tonic-gate { 25877c478bd9Sstevel@tonic-gate uintptr_t v = (uintptr_t)vaddr; 25887c478bd9Sstevel@tonic-gate uintptr_t eaddr = v + size; 25897c478bd9Sstevel@tonic-gate size_t start; 25907c478bd9Sstevel@tonic-gate size_t end; 25917c478bd9Sstevel@tonic-gate 25927c478bd9Sstevel@tonic-gate /* 25937c478bd9Sstevel@tonic-gate * if called very early by kmdb, just return NULL 25947c478bd9Sstevel@tonic-gate */ 25957c478bd9Sstevel@tonic-gate if (toxic_bit_map == NULL) 25967c478bd9Sstevel@tonic-gate return (NULL); 25977c478bd9Sstevel@tonic-gate 25987c478bd9Sstevel@tonic-gate /* 25997c478bd9Sstevel@tonic-gate * First check if we're completely outside the bitmap range. 26007c478bd9Sstevel@tonic-gate */ 2601ae115bc7Smrj if (v >= valloc_base || eaddr < kernelbase) 26027c478bd9Sstevel@tonic-gate return (NULL); 26037c478bd9Sstevel@tonic-gate 26047c478bd9Sstevel@tonic-gate /* 26057c478bd9Sstevel@tonic-gate * Trim ends of search to look at only what the bitmap covers. 26067c478bd9Sstevel@tonic-gate */ 26077c478bd9Sstevel@tonic-gate if (v < kernelbase) 26087c478bd9Sstevel@tonic-gate v = kernelbase; 26097c478bd9Sstevel@tonic-gate start = btop(v - kernelbase); 26107c478bd9Sstevel@tonic-gate end = btop(eaddr - kernelbase); 26117c478bd9Sstevel@tonic-gate if (end >= toxic_bit_map_len) 26127c478bd9Sstevel@tonic-gate end = toxic_bit_map_len; 26137c478bd9Sstevel@tonic-gate 26147c478bd9Sstevel@tonic-gate if (bt_range(toxic_bit_map, &start, &end, end) == 0) 26157c478bd9Sstevel@tonic-gate return (NULL); 26167c478bd9Sstevel@tonic-gate 26177c478bd9Sstevel@tonic-gate v = kernelbase + ptob(start); 26187c478bd9Sstevel@tonic-gate if (len != NULL) 26197c478bd9Sstevel@tonic-gate *len = ptob(end - start); 26207c478bd9Sstevel@tonic-gate return ((void *)v); 26217c478bd9Sstevel@tonic-gate } 26227c478bd9Sstevel@tonic-gate 2623ae115bc7Smrj #endif /* __i386 */ 2624