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 125*d7d93655Sblakej extern int size_pse_array(pgcnt_t, int); 126*d7d93655Sblakej 1277c478bd9Sstevel@tonic-gate /* 1287c478bd9Sstevel@tonic-gate * XXX make declaration below "static" when drivers no longer use this 1297c478bd9Sstevel@tonic-gate * interface. 1307c478bd9Sstevel@tonic-gate */ 1317c478bd9Sstevel@tonic-gate extern caddr_t p0_va; /* Virtual address for accessing physical page 0 */ 1327c478bd9Sstevel@tonic-gate 1337c478bd9Sstevel@tonic-gate /* 1347c478bd9Sstevel@tonic-gate * segkp 1357c478bd9Sstevel@tonic-gate */ 1367c478bd9Sstevel@tonic-gate extern int segkp_fromheap; 1377c478bd9Sstevel@tonic-gate 1387c478bd9Sstevel@tonic-gate static void kvm_init(void); 1397c478bd9Sstevel@tonic-gate static void startup_init(void); 1407c478bd9Sstevel@tonic-gate static void startup_memlist(void); 141ae115bc7Smrj static void startup_kmem(void); 1427c478bd9Sstevel@tonic-gate static void startup_modules(void); 1437c478bd9Sstevel@tonic-gate static void startup_vm(void); 1447c478bd9Sstevel@tonic-gate static void startup_end(void); 14535b1ab99Sjosephb static void layout_kernel_va(void); 1467c478bd9Sstevel@tonic-gate 1477c478bd9Sstevel@tonic-gate /* 1487c478bd9Sstevel@tonic-gate * Declare these as initialized data so we can patch them. 1497c478bd9Sstevel@tonic-gate */ 150f53ad214Skchow #ifdef __i386 15135b1ab99Sjosephb 152f53ad214Skchow /* 153f53ad214Skchow * Due to virtual address space limitations running in 32 bit mode, restrict 15435b1ab99Sjosephb * the amount of physical memory configured to a max of PHYSMEM pages (16g). 155f53ad214Skchow * 156f53ad214Skchow * If the physical max memory size of 64g were allowed to be configured, the 157f53ad214Skchow * size of user virtual address space will be less than 1g. A limited user 158f53ad214Skchow * address space greatly reduces the range of applications that can run. 159f53ad214Skchow * 16035b1ab99Sjosephb * If more physical memory than PHYSMEM is required, users should preferably 16135b1ab99Sjosephb * run in 64 bit mode which has far looser virtual address space limitations. 162f53ad214Skchow * 163f53ad214Skchow * If 64 bit mode is not available (as in IA32) and/or more physical memory 16435b1ab99Sjosephb * than PHYSMEM is required in 32 bit mode, physmem can be set to the desired 165f53ad214Skchow * value or to 0 (to configure all available memory) via eeprom(1M). kernelbase 166f53ad214Skchow * should also be carefully tuned to balance out the need of the user 167f53ad214Skchow * application while minimizing the risk of kernel heap exhaustion due to 168f53ad214Skchow * kernelbase being set too high. 169f53ad214Skchow */ 17035b1ab99Sjosephb #define PHYSMEM 0x400000 171f53ad214Skchow 17235b1ab99Sjosephb #else /* __amd64 */ 17335b1ab99Sjosephb 17435b1ab99Sjosephb /* 17535b1ab99Sjosephb * For now we can handle memory with physical addresses up to about 17635b1ab99Sjosephb * 64 Terabytes. This keeps the kernel above the VA hole, leaving roughly 17735b1ab99Sjosephb * half the VA space for seg_kpm. When systems get bigger than 64TB this 17835b1ab99Sjosephb * code will need revisiting. There is an implicit assumption that there 17935b1ab99Sjosephb * are no *huge* holes in the physical address space too. 18035b1ab99Sjosephb */ 18135b1ab99Sjosephb #define TERABYTE (1ul << 40) 18235b1ab99Sjosephb #define PHYSMEM_MAX64 mmu_btop(64 * TERABYTE) 18335b1ab99Sjosephb #define PHYSMEM PHYSMEM_MAX64 18435b1ab99Sjosephb #define AMD64_VA_HOLE_END 0xFFFF800000000000ul 18535b1ab99Sjosephb 18635b1ab99Sjosephb #endif /* __amd64 */ 18735b1ab99Sjosephb 18835b1ab99Sjosephb pgcnt_t physmem = PHYSMEM; 1897c478bd9Sstevel@tonic-gate pgcnt_t obp_pages; /* Memory used by PROM for its text and data */ 1907c478bd9Sstevel@tonic-gate 1917c478bd9Sstevel@tonic-gate char *kobj_file_buf; 1927c478bd9Sstevel@tonic-gate int kobj_file_bufsize; /* set in /etc/system */ 1937c478bd9Sstevel@tonic-gate 1947c478bd9Sstevel@tonic-gate /* Global variables for MP support. Used in mp_startup */ 1957c478bd9Sstevel@tonic-gate caddr_t rm_platter_va; 1967c478bd9Sstevel@tonic-gate uint32_t rm_platter_pa; 1977c478bd9Sstevel@tonic-gate 19883f9b804Skchow int auto_lpg_disable = 1; 19983f9b804Skchow 2007c478bd9Sstevel@tonic-gate /* 2017c478bd9Sstevel@tonic-gate * Some CPUs have holes in the middle of the 64-bit virtual address range. 2027c478bd9Sstevel@tonic-gate */ 2037c478bd9Sstevel@tonic-gate uintptr_t hole_start, hole_end; 2047c478bd9Sstevel@tonic-gate 2057c478bd9Sstevel@tonic-gate /* 2067c478bd9Sstevel@tonic-gate * kpm mapping window 2077c478bd9Sstevel@tonic-gate */ 2087c478bd9Sstevel@tonic-gate caddr_t kpm_vbase; 2097c478bd9Sstevel@tonic-gate size_t kpm_size; 21035b1ab99Sjosephb static int kpm_desired; 21135b1ab99Sjosephb #ifdef __amd64 21235b1ab99Sjosephb static uintptr_t segkpm_base = (uintptr_t)SEGKPM_BASE; 21335b1ab99Sjosephb #endif 2147c478bd9Sstevel@tonic-gate 2157c478bd9Sstevel@tonic-gate /* 2167c478bd9Sstevel@tonic-gate * Configuration parameters set at boot time. 2177c478bd9Sstevel@tonic-gate */ 2187c478bd9Sstevel@tonic-gate 2197c478bd9Sstevel@tonic-gate caddr_t econtig; /* end of first block of contiguous kernel */ 2207c478bd9Sstevel@tonic-gate 2217c478bd9Sstevel@tonic-gate struct bootops *bootops = 0; /* passed in from boot */ 2227c478bd9Sstevel@tonic-gate struct bootops **bootopsp; 2237c478bd9Sstevel@tonic-gate struct boot_syscalls *sysp; /* passed in from boot */ 2247c478bd9Sstevel@tonic-gate 2257c478bd9Sstevel@tonic-gate char bootblock_fstype[16]; 2267c478bd9Sstevel@tonic-gate 2277c478bd9Sstevel@tonic-gate char kern_bootargs[OBP_MAXPATHLEN]; 2287c478bd9Sstevel@tonic-gate 229ad23a2dbSjohansen /* 230ad23a2dbSjohansen * ZFS zio segment. This allows us to exclude large portions of ZFS data that 231ad23a2dbSjohansen * gets cached in kmem caches on the heap. If this is set to zero, we allocate 232ad23a2dbSjohansen * zio buffers from their own segment, otherwise they are allocated from the 233ad23a2dbSjohansen * heap. The optimization of allocating zio buffers from their own segment is 234ad23a2dbSjohansen * only valid on 64-bit kernels. 235ad23a2dbSjohansen */ 236ad23a2dbSjohansen #if defined(__amd64) 237ad23a2dbSjohansen int segzio_fromheap = 0; 238ad23a2dbSjohansen #else 239ad23a2dbSjohansen int segzio_fromheap = 1; 240ad23a2dbSjohansen #endif 241ad23a2dbSjohansen 2427c478bd9Sstevel@tonic-gate /* 2437c478bd9Sstevel@tonic-gate * new memory fragmentations are possible in startup() due to BOP_ALLOCs. this 2447c478bd9Sstevel@tonic-gate * depends on number of BOP_ALLOC calls made and requested size, memory size 2457c478bd9Sstevel@tonic-gate * combination and whether boot.bin memory needs to be freed. 2467c478bd9Sstevel@tonic-gate */ 2477c478bd9Sstevel@tonic-gate #define POSS_NEW_FRAGMENTS 12 2487c478bd9Sstevel@tonic-gate 2497c478bd9Sstevel@tonic-gate /* 2507c478bd9Sstevel@tonic-gate * VM data structures 2517c478bd9Sstevel@tonic-gate */ 2527c478bd9Sstevel@tonic-gate long page_hashsz; /* Size of page hash table (power of two) */ 2537c478bd9Sstevel@tonic-gate struct page *pp_base; /* Base of initial system page struct array */ 2547c478bd9Sstevel@tonic-gate struct page **page_hash; /* Page hash table */ 255*d7d93655Sblakej pad_mutex_t *pse_mutex; /* Locks protecting pp->p_selock */ 256*d7d93655Sblakej size_t pse_table_size; /* Number of mutexes in pse_mutex[] */ 257*d7d93655Sblakej int pse_shift; /* log2(pse_table_size) */ 2587c478bd9Sstevel@tonic-gate struct seg ktextseg; /* Segment used for kernel executable image */ 2597c478bd9Sstevel@tonic-gate struct seg kvalloc; /* Segment used for "valloc" mapping */ 2607c478bd9Sstevel@tonic-gate struct seg kpseg; /* Segment used for pageable kernel virt mem */ 2617c478bd9Sstevel@tonic-gate struct seg kmapseg; /* Segment used for generic kernel mappings */ 2627c478bd9Sstevel@tonic-gate struct seg kdebugseg; /* Segment used for the kernel debugger */ 2637c478bd9Sstevel@tonic-gate 2647c478bd9Sstevel@tonic-gate struct seg *segkmap = &kmapseg; /* Kernel generic mapping segment */ 265ae115bc7Smrj static struct seg *segmap = &kmapseg; /* easier to use name for in here */ 266ae115bc7Smrj 2677c478bd9Sstevel@tonic-gate struct seg *segkp = &kpseg; /* Pageable kernel virtual memory segment */ 2687c478bd9Sstevel@tonic-gate 2697c478bd9Sstevel@tonic-gate #if defined(__amd64) 2707c478bd9Sstevel@tonic-gate struct seg kvseg_core; /* Segment used for the core heap */ 2717c478bd9Sstevel@tonic-gate struct seg kpmseg; /* Segment used for physical mapping */ 2727c478bd9Sstevel@tonic-gate struct seg *segkpm = &kpmseg; /* 64bit kernel physical mapping segment */ 2737c478bd9Sstevel@tonic-gate #else 2747c478bd9Sstevel@tonic-gate struct seg *segkpm = NULL; /* Unused on IA32 */ 2757c478bd9Sstevel@tonic-gate #endif 2767c478bd9Sstevel@tonic-gate 2777c478bd9Sstevel@tonic-gate caddr_t segkp_base; /* Base address of segkp */ 278ad23a2dbSjohansen caddr_t segzio_base; /* Base address of segzio */ 2797c478bd9Sstevel@tonic-gate #if defined(__amd64) 2807c478bd9Sstevel@tonic-gate pgcnt_t segkpsize = btop(SEGKPDEFSIZE); /* size of segkp segment in pages */ 2817c478bd9Sstevel@tonic-gate #else 2827c478bd9Sstevel@tonic-gate pgcnt_t segkpsize = 0; 2837c478bd9Sstevel@tonic-gate #endif 284ad23a2dbSjohansen pgcnt_t segziosize = 0; /* size of zio segment in pages */ 2857c478bd9Sstevel@tonic-gate 286ae115bc7Smrj /* 287ae115bc7Smrj * VA range available to the debugger 288ae115bc7Smrj */ 289ae115bc7Smrj const caddr_t kdi_segdebugbase = (const caddr_t)SEGDEBUGBASE; 290ae115bc7Smrj const size_t kdi_segdebugsize = SEGDEBUGSIZE; 291ae115bc7Smrj 2927c478bd9Sstevel@tonic-gate struct memseg *memseg_base; 2937c478bd9Sstevel@tonic-gate struct vnode unused_pages_vp; 2947c478bd9Sstevel@tonic-gate 2957c478bd9Sstevel@tonic-gate #define FOURGB 0x100000000LL 2967c478bd9Sstevel@tonic-gate 2977c478bd9Sstevel@tonic-gate struct memlist *memlist; 2987c478bd9Sstevel@tonic-gate 2997c478bd9Sstevel@tonic-gate caddr_t s_text; /* start of kernel text segment */ 3007c478bd9Sstevel@tonic-gate caddr_t e_text; /* end of kernel text segment */ 3017c478bd9Sstevel@tonic-gate caddr_t s_data; /* start of kernel data segment */ 3027c478bd9Sstevel@tonic-gate caddr_t e_data; /* end of kernel data segment */ 3037c478bd9Sstevel@tonic-gate caddr_t modtext; /* start of loadable module text reserved */ 3047c478bd9Sstevel@tonic-gate caddr_t e_modtext; /* end of loadable module text reserved */ 3057c478bd9Sstevel@tonic-gate caddr_t moddata; /* start of loadable module data reserved */ 3067c478bd9Sstevel@tonic-gate caddr_t e_moddata; /* end of loadable module data reserved */ 3077c478bd9Sstevel@tonic-gate 3087c478bd9Sstevel@tonic-gate struct memlist *phys_install; /* Total installed physical memory */ 3097c478bd9Sstevel@tonic-gate struct memlist *phys_avail; /* Total available physical memory */ 3107c478bd9Sstevel@tonic-gate 3117c478bd9Sstevel@tonic-gate /* 3127c478bd9Sstevel@tonic-gate * kphysm_init returns the number of pages that were processed 3137c478bd9Sstevel@tonic-gate */ 314ae115bc7Smrj static pgcnt_t kphysm_init(page_t *, pgcnt_t); 3157c478bd9Sstevel@tonic-gate 3167c478bd9Sstevel@tonic-gate #define IO_PROP_SIZE 64 /* device property size */ 3177c478bd9Sstevel@tonic-gate 3187c478bd9Sstevel@tonic-gate /* 3197c478bd9Sstevel@tonic-gate * a couple useful roundup macros 3207c478bd9Sstevel@tonic-gate */ 3217c478bd9Sstevel@tonic-gate #define ROUND_UP_PAGE(x) \ 3227c478bd9Sstevel@tonic-gate ((uintptr_t)P2ROUNDUP((uintptr_t)(x), (uintptr_t)MMU_PAGESIZE)) 3237c478bd9Sstevel@tonic-gate #define ROUND_UP_LPAGE(x) \ 3247c478bd9Sstevel@tonic-gate ((uintptr_t)P2ROUNDUP((uintptr_t)(x), mmu.level_size[1])) 3257c478bd9Sstevel@tonic-gate #define ROUND_UP_4MEG(x) \ 326ae115bc7Smrj ((uintptr_t)P2ROUNDUP((uintptr_t)(x), (uintptr_t)FOUR_MEG)) 3277c478bd9Sstevel@tonic-gate #define ROUND_UP_TOPLEVEL(x) \ 3287c478bd9Sstevel@tonic-gate ((uintptr_t)P2ROUNDUP((uintptr_t)(x), mmu.level_size[mmu.max_level])) 3297c478bd9Sstevel@tonic-gate 3307c478bd9Sstevel@tonic-gate /* 3317c478bd9Sstevel@tonic-gate * 32-bit Kernel's Virtual memory layout. 3327c478bd9Sstevel@tonic-gate * +-----------------------+ 333ae115bc7Smrj * | | 3347c478bd9Sstevel@tonic-gate * 0xFFC00000 -|-----------------------|- ARGSBASE 3357c478bd9Sstevel@tonic-gate * | debugger | 3367c478bd9Sstevel@tonic-gate * 0xFF800000 -|-----------------------|- SEGDEBUGBASE 3377c478bd9Sstevel@tonic-gate * | Kernel Data | 3387c478bd9Sstevel@tonic-gate * 0xFEC00000 -|-----------------------| 3397c478bd9Sstevel@tonic-gate * | Kernel Text | 3407c478bd9Sstevel@tonic-gate * 0xFE800000 -|-----------------------|- KERNEL_TEXT 341ae115bc7Smrj * |--- GDT ---|- GDT page (GDT_VA) 342ae115bc7Smrj * |--- debug info ---|- debug info (DEBUG_INFO_VA) 343ae115bc7Smrj * | | 344ae115bc7Smrj * | page_t structures | 3457c478bd9Sstevel@tonic-gate * | memsegs, memlists, | 3467c478bd9Sstevel@tonic-gate * | page hash, etc. | 347ae115bc7Smrj * --- -|-----------------------|- ekernelheap, valloc_base (floating) 348ae115bc7Smrj * | | (segkp is just an arena in the heap) 3497c478bd9Sstevel@tonic-gate * | | 3507c478bd9Sstevel@tonic-gate * | kvseg | 3517c478bd9Sstevel@tonic-gate * | | 3527c478bd9Sstevel@tonic-gate * | | 3537c478bd9Sstevel@tonic-gate * --- -|-----------------------|- kernelheap (floating) 3547c478bd9Sstevel@tonic-gate * | Segkmap | 355ae115bc7Smrj * 0xC3002000 -|-----------------------|- segmap_start (floating) 3567c478bd9Sstevel@tonic-gate * | Red Zone | 3577c478bd9Sstevel@tonic-gate * 0xC3000000 -|-----------------------|- kernelbase / userlimit (floating) 3587c478bd9Sstevel@tonic-gate * | | || 3597c478bd9Sstevel@tonic-gate * | Shared objects | \/ 3607c478bd9Sstevel@tonic-gate * | | 3617c478bd9Sstevel@tonic-gate * : : 3627c478bd9Sstevel@tonic-gate * | user data | 3637c478bd9Sstevel@tonic-gate * |-----------------------| 3647c478bd9Sstevel@tonic-gate * | user text | 3657c478bd9Sstevel@tonic-gate * 0x08048000 -|-----------------------| 3667c478bd9Sstevel@tonic-gate * | user stack | 3677c478bd9Sstevel@tonic-gate * : : 3687c478bd9Sstevel@tonic-gate * | invalid | 3697c478bd9Sstevel@tonic-gate * 0x00000000 +-----------------------+ 3707c478bd9Sstevel@tonic-gate * 3717c478bd9Sstevel@tonic-gate * 3727c478bd9Sstevel@tonic-gate * 64-bit Kernel's Virtual memory layout. (assuming 64 bit app) 3737c478bd9Sstevel@tonic-gate * +-----------------------+ 374ae115bc7Smrj * | | 3757c478bd9Sstevel@tonic-gate * 0xFFFFFFFF.FFC00000 |-----------------------|- ARGSBASE 3767c478bd9Sstevel@tonic-gate * | debugger (?) | 3777c478bd9Sstevel@tonic-gate * 0xFFFFFFFF.FF800000 |-----------------------|- SEGDEBUGBASE 3787c478bd9Sstevel@tonic-gate * | unused | 3797c478bd9Sstevel@tonic-gate * +-----------------------+ 3807c478bd9Sstevel@tonic-gate * | Kernel Data | 3817c478bd9Sstevel@tonic-gate * 0xFFFFFFFF.FBC00000 |-----------------------| 3827c478bd9Sstevel@tonic-gate * | Kernel Text | 3837c478bd9Sstevel@tonic-gate * 0xFFFFFFFF.FB800000 |-----------------------|- KERNEL_TEXT 384ae115bc7Smrj * |--- GDT ---|- GDT page (GDT_VA) 385ae115bc7Smrj * |--- debug info ---|- debug info (DEBUG_INFO_VA) 386ae115bc7Smrj * | | 3877c478bd9Sstevel@tonic-gate * | Core heap | (used for loadable modules) 3887c478bd9Sstevel@tonic-gate * 0xFFFFFFFF.C0000000 |-----------------------|- core_base / ekernelheap 3897c478bd9Sstevel@tonic-gate * | Kernel | 3907c478bd9Sstevel@tonic-gate * | heap | 3917c478bd9Sstevel@tonic-gate * 0xFFFFFXXX.XXX00000 |-----------------------|- kernelheap (floating) 392ae115bc7Smrj * | segmap | 393ae115bc7Smrj * 0xFFFFFXXX.XXX00000 |-----------------------|- segmap_start (floating) 3947c478bd9Sstevel@tonic-gate * | device mappings | 3957c478bd9Sstevel@tonic-gate * 0xFFFFFXXX.XXX00000 |-----------------------|- toxic_addr (floating) 396ad23a2dbSjohansen * | segzio | 397ad23a2dbSjohansen * 0xFFFFFXXX.XXX00000 |-----------------------|- segzio_base (floating) 3987c478bd9Sstevel@tonic-gate * | segkp | 399ae115bc7Smrj * --- |-----------------------|- segkp_base (floating) 400ae115bc7Smrj * | page_t structures | valloc_base + valloc_sz 401ae115bc7Smrj * | memsegs, memlists, | 402ae115bc7Smrj * | page hash, etc. | 40335b1ab99Sjosephb * 0xFFFFFF00.00000000 |-----------------------|- valloc_base (lower if > 1TB) 4047c478bd9Sstevel@tonic-gate * | segkpm | 4057c478bd9Sstevel@tonic-gate * 0xFFFFFE00.00000000 |-----------------------| 4067c478bd9Sstevel@tonic-gate * | Red Zone | 40735b1ab99Sjosephb * 0xFFFFFD80.00000000 |-----------------------|- KERNELBASE (lower if > 1TB) 4087c478bd9Sstevel@tonic-gate * | User stack |- User space memory 4097c478bd9Sstevel@tonic-gate * | | 4107c478bd9Sstevel@tonic-gate * | shared objects, etc | (grows downwards) 4117c478bd9Sstevel@tonic-gate * : : 4127c478bd9Sstevel@tonic-gate * | | 4137c478bd9Sstevel@tonic-gate * 0xFFFF8000.00000000 |-----------------------| 4147c478bd9Sstevel@tonic-gate * | | 4157c478bd9Sstevel@tonic-gate * | VA Hole / unused | 4167c478bd9Sstevel@tonic-gate * | | 4177c478bd9Sstevel@tonic-gate * 0x00008000.00000000 |-----------------------| 4187c478bd9Sstevel@tonic-gate * | | 4197c478bd9Sstevel@tonic-gate * | | 4207c478bd9Sstevel@tonic-gate * : : 4217c478bd9Sstevel@tonic-gate * | user heap | (grows upwards) 4227c478bd9Sstevel@tonic-gate * | | 4237c478bd9Sstevel@tonic-gate * | user data | 4247c478bd9Sstevel@tonic-gate * |-----------------------| 4257c478bd9Sstevel@tonic-gate * | user text | 4267c478bd9Sstevel@tonic-gate * 0x00000000.04000000 |-----------------------| 4277c478bd9Sstevel@tonic-gate * | invalid | 4287c478bd9Sstevel@tonic-gate * 0x00000000.00000000 +-----------------------+ 4297c478bd9Sstevel@tonic-gate * 4307c478bd9Sstevel@tonic-gate * A 32 bit app on the 64 bit kernel sees the same layout as on the 32 bit 4317c478bd9Sstevel@tonic-gate * kernel, except that userlimit is raised to 0xfe000000 4327c478bd9Sstevel@tonic-gate * 4337c478bd9Sstevel@tonic-gate * Floating values: 4347c478bd9Sstevel@tonic-gate * 4357c478bd9Sstevel@tonic-gate * valloc_base: start of the kernel's memory management/tracking data 436ae115bc7Smrj * structures. This region contains page_t structures for 437ae115bc7Smrj * physical memory, memsegs, memlists, and the page hash. 4387c478bd9Sstevel@tonic-gate * 4397c478bd9Sstevel@tonic-gate * core_base: start of the kernel's "core" heap area on 64-bit systems. 4407c478bd9Sstevel@tonic-gate * This area is intended to be used for global data as well as for module 4417c478bd9Sstevel@tonic-gate * text/data that does not fit into the nucleus pages. The core heap is 4427c478bd9Sstevel@tonic-gate * restricted to a 2GB range, allowing every address within it to be 4437c478bd9Sstevel@tonic-gate * accessed using rip-relative addressing 4447c478bd9Sstevel@tonic-gate * 4457c478bd9Sstevel@tonic-gate * ekernelheap: end of kernelheap and start of segmap. 4467c478bd9Sstevel@tonic-gate * 4477c478bd9Sstevel@tonic-gate * kernelheap: start of kernel heap. On 32-bit systems, this starts right 4487c478bd9Sstevel@tonic-gate * above a red zone that separates the user's address space from the 4497c478bd9Sstevel@tonic-gate * kernel's. On 64-bit systems, it sits above segkp and segkpm. 4507c478bd9Sstevel@tonic-gate * 451ae115bc7Smrj * segmap_start: start of segmap. The length of segmap can be modified 4527c478bd9Sstevel@tonic-gate * by changing segmapsize in /etc/system (preferred) or eeprom (deprecated). 4537c478bd9Sstevel@tonic-gate * The default length is 16MB on 32-bit systems and 64MB on 64-bit systems. 4547c478bd9Sstevel@tonic-gate * 4557c478bd9Sstevel@tonic-gate * kernelbase: On a 32-bit kernel the default value of 0xd4000000 will be 4567c478bd9Sstevel@tonic-gate * decreased by 2X the size required for page_t. This allows the kernel 4577c478bd9Sstevel@tonic-gate * heap to grow in size with physical memory. With sizeof(page_t) == 80 4587c478bd9Sstevel@tonic-gate * bytes, the following shows the values of kernelbase and kernel heap 4597c478bd9Sstevel@tonic-gate * sizes for different memory configurations (assuming default segmap and 4607c478bd9Sstevel@tonic-gate * segkp sizes). 4617c478bd9Sstevel@tonic-gate * 4627c478bd9Sstevel@tonic-gate * mem size for kernelbase kernel heap 4637c478bd9Sstevel@tonic-gate * size page_t's size 4647c478bd9Sstevel@tonic-gate * ---- --------- ---------- ----------- 4657c478bd9Sstevel@tonic-gate * 1gb 0x01400000 0xd1800000 684MB 4667c478bd9Sstevel@tonic-gate * 2gb 0x02800000 0xcf000000 704MB 4677c478bd9Sstevel@tonic-gate * 4gb 0x05000000 0xca000000 744MB 4687c478bd9Sstevel@tonic-gate * 6gb 0x07800000 0xc5000000 784MB 4697c478bd9Sstevel@tonic-gate * 8gb 0x0a000000 0xc0000000 824MB 4707c478bd9Sstevel@tonic-gate * 16gb 0x14000000 0xac000000 984MB 4717c478bd9Sstevel@tonic-gate * 32gb 0x28000000 0x84000000 1304MB 4727c478bd9Sstevel@tonic-gate * 64gb 0x50000000 0x34000000 1944MB (*) 4737c478bd9Sstevel@tonic-gate * 4747c478bd9Sstevel@tonic-gate * kernelbase is less than the abi minimum of 0xc0000000 for memory 4757c478bd9Sstevel@tonic-gate * configurations above 8gb. 4767c478bd9Sstevel@tonic-gate * 4777c478bd9Sstevel@tonic-gate * (*) support for memory configurations above 32gb will require manual tuning 4787c478bd9Sstevel@tonic-gate * of kernelbase to balance out the need of user applications. 4797c478bd9Sstevel@tonic-gate */ 4807c478bd9Sstevel@tonic-gate 4817c478bd9Sstevel@tonic-gate /* real-time-clock initialization parameters */ 482ae115bc7Smrj extern time_t process_rtc_config_file(void); 4837c478bd9Sstevel@tonic-gate 4847c478bd9Sstevel@tonic-gate uintptr_t kernelbase; 485ae115bc7Smrj uintptr_t postbootkernelbase; /* not set till boot loader is gone */ 4867c478bd9Sstevel@tonic-gate uintptr_t eprom_kernelbase; 4877c478bd9Sstevel@tonic-gate size_t segmapsize; 488ae115bc7Smrj uintptr_t segmap_start; 4897c478bd9Sstevel@tonic-gate int segmapfreelists; 4907c478bd9Sstevel@tonic-gate pgcnt_t npages; 491ae115bc7Smrj pgcnt_t orig_npages; 4927c478bd9Sstevel@tonic-gate size_t core_size; /* size of "core" heap */ 4937c478bd9Sstevel@tonic-gate uintptr_t core_base; /* base address of "core" heap */ 4947c478bd9Sstevel@tonic-gate 4957c478bd9Sstevel@tonic-gate /* 4967c478bd9Sstevel@tonic-gate * List of bootstrap pages. We mark these as allocated in startup. 4977c478bd9Sstevel@tonic-gate * release_bootstrap() will free them when we're completely done with 4987c478bd9Sstevel@tonic-gate * the bootstrap. 4997c478bd9Sstevel@tonic-gate */ 500ae115bc7Smrj static page_t *bootpages; 501ae115bc7Smrj 502ae115bc7Smrj /* 503ae115bc7Smrj * boot time pages that have a vnode from the ramdisk will keep that forever. 504ae115bc7Smrj */ 505ae115bc7Smrj static page_t *rd_pages; 5067c478bd9Sstevel@tonic-gate 5077c478bd9Sstevel@tonic-gate struct system_hardware system_hardware; 5087c478bd9Sstevel@tonic-gate 5097c478bd9Sstevel@tonic-gate /* 5107c478bd9Sstevel@tonic-gate * Enable some debugging messages concerning memory usage... 5117c478bd9Sstevel@tonic-gate */ 5127c478bd9Sstevel@tonic-gate static void 513ae115bc7Smrj print_memlist(char *title, struct memlist *mp) 5147c478bd9Sstevel@tonic-gate { 5157c478bd9Sstevel@tonic-gate prom_printf("MEMLIST: %s:\n", title); 5167c478bd9Sstevel@tonic-gate while (mp != NULL) { 5177c478bd9Sstevel@tonic-gate prom_printf("\tAddress 0x%" PRIx64 ", size 0x%" PRIx64 "\n", 5187c478bd9Sstevel@tonic-gate mp->address, mp->size); 5197c478bd9Sstevel@tonic-gate mp = mp->next; 5207c478bd9Sstevel@tonic-gate } 5217c478bd9Sstevel@tonic-gate } 5227c478bd9Sstevel@tonic-gate 5237c478bd9Sstevel@tonic-gate /* 5247c478bd9Sstevel@tonic-gate * XX64 need a comment here.. are these just default values, surely 5257c478bd9Sstevel@tonic-gate * we read the "cpuid" type information to figure this out. 5267c478bd9Sstevel@tonic-gate */ 5277c478bd9Sstevel@tonic-gate int l2cache_sz = 0x80000; 5287c478bd9Sstevel@tonic-gate int l2cache_linesz = 0x40; 5297c478bd9Sstevel@tonic-gate int l2cache_assoc = 1; 5307c478bd9Sstevel@tonic-gate 5312cb27123Saguzovsk static size_t textrepl_min_gb = 10; 5322cb27123Saguzovsk 5337c478bd9Sstevel@tonic-gate /* 5347c478bd9Sstevel@tonic-gate * on 64 bit we use a predifined VA range for mapping devices in the kernel 5357c478bd9Sstevel@tonic-gate * on 32 bit the mappings are intermixed in the heap, so we use a bit map 5367c478bd9Sstevel@tonic-gate */ 5377c478bd9Sstevel@tonic-gate #ifdef __amd64 5387c478bd9Sstevel@tonic-gate 5397c478bd9Sstevel@tonic-gate vmem_t *device_arena; 5407c478bd9Sstevel@tonic-gate uintptr_t toxic_addr = (uintptr_t)NULL; 541ae115bc7Smrj size_t toxic_size = 1024 * 1024 * 1024; /* Sparc uses 1 gig too */ 5427c478bd9Sstevel@tonic-gate 5437c478bd9Sstevel@tonic-gate #else /* __i386 */ 5447c478bd9Sstevel@tonic-gate 5457c478bd9Sstevel@tonic-gate ulong_t *toxic_bit_map; /* one bit for each 4k of VA in heap_arena */ 5467c478bd9Sstevel@tonic-gate size_t toxic_bit_map_len = 0; /* in bits */ 5477c478bd9Sstevel@tonic-gate 5487c478bd9Sstevel@tonic-gate #endif /* __i386 */ 5497c478bd9Sstevel@tonic-gate 5507c478bd9Sstevel@tonic-gate /* 5517c478bd9Sstevel@tonic-gate * Simple boot time debug facilities 5527c478bd9Sstevel@tonic-gate */ 5537c478bd9Sstevel@tonic-gate static char *prm_dbg_str[] = { 5547c478bd9Sstevel@tonic-gate "%s:%d: '%s' is 0x%x\n", 5557c478bd9Sstevel@tonic-gate "%s:%d: '%s' is 0x%llx\n" 5567c478bd9Sstevel@tonic-gate }; 5577c478bd9Sstevel@tonic-gate 5587c478bd9Sstevel@tonic-gate int prom_debug; 5597c478bd9Sstevel@tonic-gate 5607c478bd9Sstevel@tonic-gate #define PRM_DEBUG(q) if (prom_debug) \ 5617c478bd9Sstevel@tonic-gate prom_printf(prm_dbg_str[sizeof (q) >> 3], "startup.c", __LINE__, #q, q); 5627c478bd9Sstevel@tonic-gate #define PRM_POINT(q) if (prom_debug) \ 5637c478bd9Sstevel@tonic-gate prom_printf("%s:%d: %s\n", "startup.c", __LINE__, q); 5647c478bd9Sstevel@tonic-gate 5657c478bd9Sstevel@tonic-gate /* 5667c478bd9Sstevel@tonic-gate * This structure is used to keep track of the intial allocations 5677c478bd9Sstevel@tonic-gate * done in startup_memlist(). The value of NUM_ALLOCATIONS needs to 5687c478bd9Sstevel@tonic-gate * be >= the number of ADD_TO_ALLOCATIONS() executed in the code. 5697c478bd9Sstevel@tonic-gate */ 5707c478bd9Sstevel@tonic-gate #define NUM_ALLOCATIONS 7 5717c478bd9Sstevel@tonic-gate int num_allocations = 0; 5727c478bd9Sstevel@tonic-gate struct { 5737c478bd9Sstevel@tonic-gate void **al_ptr; 5747c478bd9Sstevel@tonic-gate size_t al_size; 5757c478bd9Sstevel@tonic-gate } allocations[NUM_ALLOCATIONS]; 5767c478bd9Sstevel@tonic-gate size_t valloc_sz = 0; 5777c478bd9Sstevel@tonic-gate uintptr_t valloc_base; 5787c478bd9Sstevel@tonic-gate 5797c478bd9Sstevel@tonic-gate #define ADD_TO_ALLOCATIONS(ptr, size) { \ 5807c478bd9Sstevel@tonic-gate size = ROUND_UP_PAGE(size); \ 5817c478bd9Sstevel@tonic-gate if (num_allocations == NUM_ALLOCATIONS) \ 5827c478bd9Sstevel@tonic-gate panic("too many ADD_TO_ALLOCATIONS()"); \ 5837c478bd9Sstevel@tonic-gate allocations[num_allocations].al_ptr = (void**)&ptr; \ 5847c478bd9Sstevel@tonic-gate allocations[num_allocations].al_size = size; \ 5857c478bd9Sstevel@tonic-gate valloc_sz += size; \ 5867c478bd9Sstevel@tonic-gate ++num_allocations; \ 5877c478bd9Sstevel@tonic-gate } 5887c478bd9Sstevel@tonic-gate 589ae115bc7Smrj /* 590ae115bc7Smrj * Allocate all the initial memory needed by the page allocator. 591ae115bc7Smrj */ 5927c478bd9Sstevel@tonic-gate static void 5937c478bd9Sstevel@tonic-gate perform_allocations(void) 5947c478bd9Sstevel@tonic-gate { 5957c478bd9Sstevel@tonic-gate caddr_t mem; 5967c478bd9Sstevel@tonic-gate int i; 597ae115bc7Smrj int valloc_align; 5987c478bd9Sstevel@tonic-gate 599ae115bc7Smrj PRM_DEBUG(valloc_base); 600ae115bc7Smrj PRM_DEBUG(valloc_sz); 601ae115bc7Smrj valloc_align = mmu.level_size[mmu.max_page_level > 0]; 602ae115bc7Smrj mem = BOP_ALLOC(bootops, (caddr_t)valloc_base, valloc_sz, valloc_align); 6037c478bd9Sstevel@tonic-gate if (mem != (caddr_t)valloc_base) 6047c478bd9Sstevel@tonic-gate panic("BOP_ALLOC() failed"); 6057c478bd9Sstevel@tonic-gate bzero(mem, valloc_sz); 6067c478bd9Sstevel@tonic-gate for (i = 0; i < num_allocations; ++i) { 6077c478bd9Sstevel@tonic-gate *allocations[i].al_ptr = (void *)mem; 6087c478bd9Sstevel@tonic-gate mem += allocations[i].al_size; 6097c478bd9Sstevel@tonic-gate } 6107c478bd9Sstevel@tonic-gate } 6117c478bd9Sstevel@tonic-gate 6127c478bd9Sstevel@tonic-gate /* 6137c478bd9Sstevel@tonic-gate * Our world looks like this at startup time. 6147c478bd9Sstevel@tonic-gate * 6157c478bd9Sstevel@tonic-gate * In a 32-bit OS, boot loads the kernel text at 0xfe800000 and kernel data 6167c478bd9Sstevel@tonic-gate * at 0xfec00000. On a 64-bit OS, kernel text and data are loaded at 6177c478bd9Sstevel@tonic-gate * 0xffffffff.fe800000 and 0xffffffff.fec00000 respectively. Those 6187c478bd9Sstevel@tonic-gate * addresses are fixed in the binary at link time. 6197c478bd9Sstevel@tonic-gate * 6207c478bd9Sstevel@tonic-gate * On the text page: 6217c478bd9Sstevel@tonic-gate * unix/genunix/krtld/module text loads. 6227c478bd9Sstevel@tonic-gate * 6237c478bd9Sstevel@tonic-gate * On the data page: 624ae115bc7Smrj * unix/genunix/krtld/module data loads. 625ae115bc7Smrj * 6267c478bd9Sstevel@tonic-gate * Machine-dependent startup code 6277c478bd9Sstevel@tonic-gate */ 6287c478bd9Sstevel@tonic-gate void 6297c478bd9Sstevel@tonic-gate startup(void) 6307c478bd9Sstevel@tonic-gate { 63175bcd456Sjg extern void startup_bios_disk(void); 63275bcd456Sjg extern void startup_pci_bios(void); 6337c478bd9Sstevel@tonic-gate #if defined(__amd64) 63435b1ab99Sjosephb kpm_desired = 1; 6357c478bd9Sstevel@tonic-gate #endif 6367c478bd9Sstevel@tonic-gate kpm_enable = 0; 6377c478bd9Sstevel@tonic-gate 6387c478bd9Sstevel@tonic-gate progressbar_init(); 6397c478bd9Sstevel@tonic-gate startup_init(); 6407c478bd9Sstevel@tonic-gate startup_memlist(); 641ae115bc7Smrj startup_kmem(); 64235b1ab99Sjosephb startup_vm(); 64375bcd456Sjg startup_pci_bios(); 6447c478bd9Sstevel@tonic-gate startup_modules(); 6457c478bd9Sstevel@tonic-gate startup_bios_disk(); 6467c478bd9Sstevel@tonic-gate startup_end(); 6477c478bd9Sstevel@tonic-gate progressbar_start(); 6487c478bd9Sstevel@tonic-gate } 6497c478bd9Sstevel@tonic-gate 6507c478bd9Sstevel@tonic-gate static void 6517c478bd9Sstevel@tonic-gate startup_init() 6527c478bd9Sstevel@tonic-gate { 6537c478bd9Sstevel@tonic-gate PRM_POINT("startup_init() starting..."); 6547c478bd9Sstevel@tonic-gate 6557c478bd9Sstevel@tonic-gate /* 6567c478bd9Sstevel@tonic-gate * Complete the extraction of cpuid data 6577c478bd9Sstevel@tonic-gate */ 6587c478bd9Sstevel@tonic-gate cpuid_pass2(CPU); 6597c478bd9Sstevel@tonic-gate 6607c478bd9Sstevel@tonic-gate (void) check_boot_version(BOP_GETVERSION(bootops)); 6617c478bd9Sstevel@tonic-gate 6627c478bd9Sstevel@tonic-gate /* 6637c478bd9Sstevel@tonic-gate * Check for prom_debug in boot environment 6647c478bd9Sstevel@tonic-gate */ 6657c478bd9Sstevel@tonic-gate if (BOP_GETPROPLEN(bootops, "prom_debug") >= 0) { 6667c478bd9Sstevel@tonic-gate ++prom_debug; 6677c478bd9Sstevel@tonic-gate PRM_POINT("prom_debug found in boot enviroment"); 6687c478bd9Sstevel@tonic-gate } 6697c478bd9Sstevel@tonic-gate 6707c478bd9Sstevel@tonic-gate /* 6717c478bd9Sstevel@tonic-gate * Collect node, cpu and memory configuration information. 6727c478bd9Sstevel@tonic-gate */ 6737c478bd9Sstevel@tonic-gate get_system_configuration(); 6747c478bd9Sstevel@tonic-gate 6757c478bd9Sstevel@tonic-gate /* 6767c478bd9Sstevel@tonic-gate * Halt if this is an unsupported processor. 6777c478bd9Sstevel@tonic-gate */ 6787c478bd9Sstevel@tonic-gate if (x86_type == X86_TYPE_486 || x86_type == X86_TYPE_CYRIX_486) { 6797c478bd9Sstevel@tonic-gate printf("\n486 processor (\"%s\") detected.\n", 6807c478bd9Sstevel@tonic-gate CPU->cpu_brandstr); 6817c478bd9Sstevel@tonic-gate halt("This processor is not supported by this release " 6827c478bd9Sstevel@tonic-gate "of Solaris."); 6837c478bd9Sstevel@tonic-gate } 6847c478bd9Sstevel@tonic-gate 6857c478bd9Sstevel@tonic-gate PRM_POINT("startup_init() done"); 6867c478bd9Sstevel@tonic-gate } 6877c478bd9Sstevel@tonic-gate 6887c478bd9Sstevel@tonic-gate /* 6897c478bd9Sstevel@tonic-gate * Callback for copy_memlist_filter() to filter nucleus, kadb/kmdb, (ie. 6907c478bd9Sstevel@tonic-gate * everything mapped above KERNEL_TEXT) pages from phys_avail. Note it 6917c478bd9Sstevel@tonic-gate * also filters out physical page zero. There is some reliance on the 6927c478bd9Sstevel@tonic-gate * boot loader allocating only a few contiguous physical memory chunks. 6937c478bd9Sstevel@tonic-gate */ 6947c478bd9Sstevel@tonic-gate static void 6957c478bd9Sstevel@tonic-gate avail_filter(uint64_t *addr, uint64_t *size) 6967c478bd9Sstevel@tonic-gate { 6977c478bd9Sstevel@tonic-gate uintptr_t va; 6987c478bd9Sstevel@tonic-gate uintptr_t next_va; 6997c478bd9Sstevel@tonic-gate pfn_t pfn; 7007c478bd9Sstevel@tonic-gate uint64_t pfn_addr; 7017c478bd9Sstevel@tonic-gate uint64_t pfn_eaddr; 7027c478bd9Sstevel@tonic-gate uint_t prot; 7037c478bd9Sstevel@tonic-gate size_t len; 7047c478bd9Sstevel@tonic-gate uint_t change; 7057c478bd9Sstevel@tonic-gate 7067c478bd9Sstevel@tonic-gate if (prom_debug) 7077c478bd9Sstevel@tonic-gate prom_printf("\tFilter: in: a=%" PRIx64 ", s=%" PRIx64 "\n", 7087c478bd9Sstevel@tonic-gate *addr, *size); 7097c478bd9Sstevel@tonic-gate 7107c478bd9Sstevel@tonic-gate /* 7117c478bd9Sstevel@tonic-gate * page zero is required for BIOS.. never make it available 7127c478bd9Sstevel@tonic-gate */ 7137c478bd9Sstevel@tonic-gate if (*addr == 0) { 7147c478bd9Sstevel@tonic-gate *addr += MMU_PAGESIZE; 7157c478bd9Sstevel@tonic-gate *size -= MMU_PAGESIZE; 7167c478bd9Sstevel@tonic-gate } 7177c478bd9Sstevel@tonic-gate 7187c478bd9Sstevel@tonic-gate /* 719ae115bc7Smrj * First we trim from the front of the range. Since kbm_probe() 7207c478bd9Sstevel@tonic-gate * walks ranges in virtual order, but addr/size are physical, we need 7217c478bd9Sstevel@tonic-gate * to the list until no changes are seen. This deals with the case 7227c478bd9Sstevel@tonic-gate * where page "p" is mapped at v, page "p + PAGESIZE" is mapped at w 7237c478bd9Sstevel@tonic-gate * but w < v. 7247c478bd9Sstevel@tonic-gate */ 7257c478bd9Sstevel@tonic-gate do { 7267c478bd9Sstevel@tonic-gate change = 0; 7277c478bd9Sstevel@tonic-gate for (va = KERNEL_TEXT; 728ae115bc7Smrj *size > 0 && kbm_probe(&va, &len, &pfn, &prot) != 0; 7297c478bd9Sstevel@tonic-gate va = next_va) { 7307c478bd9Sstevel@tonic-gate 7317c478bd9Sstevel@tonic-gate next_va = va + len; 732ae115bc7Smrj pfn_addr = pfn_to_pa(pfn); 7337c478bd9Sstevel@tonic-gate pfn_eaddr = pfn_addr + len; 7347c478bd9Sstevel@tonic-gate 7357c478bd9Sstevel@tonic-gate if (pfn_addr <= *addr && pfn_eaddr > *addr) { 7367c478bd9Sstevel@tonic-gate change = 1; 7377c478bd9Sstevel@tonic-gate while (*size > 0 && len > 0) { 7387c478bd9Sstevel@tonic-gate *addr += MMU_PAGESIZE; 7397c478bd9Sstevel@tonic-gate *size -= MMU_PAGESIZE; 7407c478bd9Sstevel@tonic-gate len -= MMU_PAGESIZE; 7417c478bd9Sstevel@tonic-gate } 7427c478bd9Sstevel@tonic-gate } 7437c478bd9Sstevel@tonic-gate } 7447c478bd9Sstevel@tonic-gate if (change && prom_debug) 7457c478bd9Sstevel@tonic-gate prom_printf("\t\ttrim: a=%" PRIx64 ", s=%" PRIx64 "\n", 7467c478bd9Sstevel@tonic-gate *addr, *size); 7477c478bd9Sstevel@tonic-gate } while (change); 7487c478bd9Sstevel@tonic-gate 7497c478bd9Sstevel@tonic-gate /* 7507c478bd9Sstevel@tonic-gate * Trim pages from the end of the range. 7517c478bd9Sstevel@tonic-gate */ 7527c478bd9Sstevel@tonic-gate for (va = KERNEL_TEXT; 753ae115bc7Smrj *size > 0 && kbm_probe(&va, &len, &pfn, &prot) != 0; 7547c478bd9Sstevel@tonic-gate va = next_va) { 7557c478bd9Sstevel@tonic-gate 7567c478bd9Sstevel@tonic-gate next_va = va + len; 757ae115bc7Smrj pfn_addr = pfn_to_pa(pfn); 7587c478bd9Sstevel@tonic-gate 7597c478bd9Sstevel@tonic-gate if (pfn_addr >= *addr && pfn_addr < *addr + *size) 7607c478bd9Sstevel@tonic-gate *size = pfn_addr - *addr; 7617c478bd9Sstevel@tonic-gate } 7627c478bd9Sstevel@tonic-gate 7637c478bd9Sstevel@tonic-gate if (prom_debug) 7647c478bd9Sstevel@tonic-gate prom_printf("\tFilter out: a=%" PRIx64 ", s=%" PRIx64 "\n", 7657c478bd9Sstevel@tonic-gate *addr, *size); 7667c478bd9Sstevel@tonic-gate } 7677c478bd9Sstevel@tonic-gate 7687c478bd9Sstevel@tonic-gate static void 7697c478bd9Sstevel@tonic-gate kpm_init() 7707c478bd9Sstevel@tonic-gate { 7717c478bd9Sstevel@tonic-gate struct segkpm_crargs b; 7727c478bd9Sstevel@tonic-gate 7737c478bd9Sstevel@tonic-gate /* 7747c478bd9Sstevel@tonic-gate * These variables were all designed for sfmmu in which segkpm is 7757c478bd9Sstevel@tonic-gate * mapped using a single pagesize - either 8KB or 4MB. On x86, we 7767c478bd9Sstevel@tonic-gate * might use 2+ page sizes on a single machine, so none of these 7777c478bd9Sstevel@tonic-gate * variables have a single correct value. They are set up as if we 7787c478bd9Sstevel@tonic-gate * always use a 4KB pagesize, which should do no harm. In the long 7797c478bd9Sstevel@tonic-gate * run, we should get rid of KPM's assumption that only a single 7807c478bd9Sstevel@tonic-gate * pagesize is used. 7817c478bd9Sstevel@tonic-gate */ 7827c478bd9Sstevel@tonic-gate kpm_pgshft = MMU_PAGESHIFT; 7837c478bd9Sstevel@tonic-gate kpm_pgsz = MMU_PAGESIZE; 7847c478bd9Sstevel@tonic-gate kpm_pgoff = MMU_PAGEOFFSET; 7857c478bd9Sstevel@tonic-gate kpmp2pshft = 0; 7867c478bd9Sstevel@tonic-gate kpmpnpgs = 1; 7877c478bd9Sstevel@tonic-gate ASSERT(((uintptr_t)kpm_vbase & (kpm_pgsz - 1)) == 0); 7887c478bd9Sstevel@tonic-gate 7897c478bd9Sstevel@tonic-gate PRM_POINT("about to create segkpm"); 7907c478bd9Sstevel@tonic-gate rw_enter(&kas.a_lock, RW_WRITER); 7917c478bd9Sstevel@tonic-gate 7927c478bd9Sstevel@tonic-gate if (seg_attach(&kas, kpm_vbase, kpm_size, segkpm) < 0) 7937c478bd9Sstevel@tonic-gate panic("cannot attach segkpm"); 7947c478bd9Sstevel@tonic-gate 7957c478bd9Sstevel@tonic-gate b.prot = PROT_READ | PROT_WRITE; 7967c478bd9Sstevel@tonic-gate b.nvcolors = 1; 7977c478bd9Sstevel@tonic-gate 7987c478bd9Sstevel@tonic-gate if (segkpm_create(segkpm, (caddr_t)&b) != 0) 7997c478bd9Sstevel@tonic-gate panic("segkpm_create segkpm"); 8007c478bd9Sstevel@tonic-gate 8017c478bd9Sstevel@tonic-gate rw_exit(&kas.a_lock); 802ae115bc7Smrj } 8037c478bd9Sstevel@tonic-gate 804ae115bc7Smrj /* 805ae115bc7Smrj * The debug info page provides enough information to allow external 806ae115bc7Smrj * inspectors (e.g. when running under a hypervisor) to bootstrap 807ae115bc7Smrj * themselves into allowing full-blown kernel debugging. 808ae115bc7Smrj */ 809ae115bc7Smrj static void 810ae115bc7Smrj init_debug_info(void) 811ae115bc7Smrj { 812ae115bc7Smrj caddr_t mem; 813ae115bc7Smrj debug_info_t *di; 814ae115bc7Smrj 815ae115bc7Smrj #ifndef __lint 816ae115bc7Smrj ASSERT(sizeof (debug_info_t) < MMU_PAGESIZE); 817ae115bc7Smrj #endif 818ae115bc7Smrj 819ae115bc7Smrj mem = BOP_ALLOC(bootops, (caddr_t)DEBUG_INFO_VA, MMU_PAGESIZE, 820ae115bc7Smrj MMU_PAGESIZE); 821ae115bc7Smrj 822ae115bc7Smrj if (mem != (caddr_t)DEBUG_INFO_VA) 823ae115bc7Smrj panic("BOP_ALLOC() failed"); 824ae115bc7Smrj bzero(mem, MMU_PAGESIZE); 825ae115bc7Smrj 826ae115bc7Smrj di = (debug_info_t *)mem; 827ae115bc7Smrj 828ae115bc7Smrj di->di_magic = DEBUG_INFO_MAGIC; 829ae115bc7Smrj di->di_version = DEBUG_INFO_VERSION; 8307c478bd9Sstevel@tonic-gate } 8317c478bd9Sstevel@tonic-gate 8327c478bd9Sstevel@tonic-gate /* 833ae115bc7Smrj * Build the memlists and other kernel essential memory system data structures. 834ae115bc7Smrj * This is everything at valloc_base. 8357c478bd9Sstevel@tonic-gate */ 8367c478bd9Sstevel@tonic-gate static void 8377c478bd9Sstevel@tonic-gate startup_memlist(void) 8387c478bd9Sstevel@tonic-gate { 8397c478bd9Sstevel@tonic-gate size_t memlist_sz; 8407c478bd9Sstevel@tonic-gate size_t memseg_sz; 8417c478bd9Sstevel@tonic-gate size_t pagehash_sz; 8427c478bd9Sstevel@tonic-gate size_t pp_sz; 8437c478bd9Sstevel@tonic-gate uintptr_t va; 8447c478bd9Sstevel@tonic-gate size_t len; 8457c478bd9Sstevel@tonic-gate uint_t prot; 8467c478bd9Sstevel@tonic-gate pfn_t pfn; 8477c478bd9Sstevel@tonic-gate int memblocks; 8487c478bd9Sstevel@tonic-gate caddr_t pagecolor_mem; 8497c478bd9Sstevel@tonic-gate size_t pagecolor_memsz; 8507c478bd9Sstevel@tonic-gate caddr_t page_ctrs_mem; 8517c478bd9Sstevel@tonic-gate size_t page_ctrs_size; 852*d7d93655Sblakej size_t pse_table_alloc_size; 8537c478bd9Sstevel@tonic-gate struct memlist *current; 8547c478bd9Sstevel@tonic-gate extern void startup_build_mem_nodes(struct memlist *); 8557c478bd9Sstevel@tonic-gate 8567c478bd9Sstevel@tonic-gate /* XX64 fix these - they should be in include files */ 8577c478bd9Sstevel@tonic-gate extern size_t page_coloring_init(uint_t, int, int); 8587c478bd9Sstevel@tonic-gate extern void page_coloring_setup(caddr_t); 8597c478bd9Sstevel@tonic-gate 8607c478bd9Sstevel@tonic-gate PRM_POINT("startup_memlist() starting..."); 8617c478bd9Sstevel@tonic-gate 8627c478bd9Sstevel@tonic-gate /* 8637c478bd9Sstevel@tonic-gate * Use leftover large page nucleus text/data space for loadable modules. 8647c478bd9Sstevel@tonic-gate * Use at most MODTEXT/MODDATA. 8657c478bd9Sstevel@tonic-gate */ 866ae115bc7Smrj len = kbm_nucleus_size; 867ae115bc7Smrj ASSERT(len > MMU_PAGESIZE); 8687c478bd9Sstevel@tonic-gate 869ae115bc7Smrj moddata = (caddr_t)ROUND_UP_PAGE(e_data); 870ae115bc7Smrj e_moddata = (caddr_t)P2ROUNDUP((uintptr_t)e_data, (uintptr_t)len); 871ae115bc7Smrj if (e_moddata - moddata > MODDATA) 872ae115bc7Smrj e_moddata = moddata + MODDATA; 8737c478bd9Sstevel@tonic-gate 874ae115bc7Smrj modtext = (caddr_t)ROUND_UP_PAGE(e_text); 875ae115bc7Smrj e_modtext = (caddr_t)P2ROUNDUP((uintptr_t)e_text, (uintptr_t)len); 876ae115bc7Smrj if (e_modtext - modtext > MODTEXT) 877ae115bc7Smrj e_modtext = modtext + MODTEXT; 8787c478bd9Sstevel@tonic-gate 8797c478bd9Sstevel@tonic-gate econtig = e_moddata; 8807c478bd9Sstevel@tonic-gate 8817c478bd9Sstevel@tonic-gate PRM_DEBUG(modtext); 8827c478bd9Sstevel@tonic-gate PRM_DEBUG(e_modtext); 8837c478bd9Sstevel@tonic-gate PRM_DEBUG(moddata); 8847c478bd9Sstevel@tonic-gate PRM_DEBUG(e_moddata); 8857c478bd9Sstevel@tonic-gate PRM_DEBUG(econtig); 8867c478bd9Sstevel@tonic-gate 8877c478bd9Sstevel@tonic-gate /* 888ae115bc7Smrj * Examine the boot loader physical memory map to find out: 8897c478bd9Sstevel@tonic-gate * - total memory in system - physinstalled 8907c478bd9Sstevel@tonic-gate * - the max physical address - physmax 891ae115bc7Smrj * - the number of discontiguous segments of memory. 8927c478bd9Sstevel@tonic-gate */ 8937c478bd9Sstevel@tonic-gate if (prom_debug) 894ae115bc7Smrj print_memlist("boot physinstalled", 8957c478bd9Sstevel@tonic-gate bootops->boot_mem->physinstalled); 8967c478bd9Sstevel@tonic-gate installed_top_size(bootops->boot_mem->physinstalled, &physmax, 8977c478bd9Sstevel@tonic-gate &physinstalled, &memblocks); 8987c478bd9Sstevel@tonic-gate PRM_DEBUG(physmax); 8997c478bd9Sstevel@tonic-gate PRM_DEBUG(physinstalled); 9007c478bd9Sstevel@tonic-gate PRM_DEBUG(memblocks); 9017c478bd9Sstevel@tonic-gate 9027c478bd9Sstevel@tonic-gate /* 9037c478bd9Sstevel@tonic-gate * Initialize hat's mmu parameters. 9047c478bd9Sstevel@tonic-gate * Check for enforce-prot-exec in boot environment. It's used to 9057c478bd9Sstevel@tonic-gate * enable/disable support for the page table entry NX bit. 9067c478bd9Sstevel@tonic-gate * The default is to enforce PROT_EXEC on processors that support NX. 9077c478bd9Sstevel@tonic-gate * Boot seems to round up the "len", but 8 seems to be big enough. 9087c478bd9Sstevel@tonic-gate */ 9097c478bd9Sstevel@tonic-gate mmu_init(); 9107c478bd9Sstevel@tonic-gate 9117c478bd9Sstevel@tonic-gate #ifdef __i386 9127c478bd9Sstevel@tonic-gate /* 9137c478bd9Sstevel@tonic-gate * physmax is lowered if there is more memory than can be 9147c478bd9Sstevel@tonic-gate * physically addressed in 32 bit (PAE/non-PAE) modes. 9157c478bd9Sstevel@tonic-gate */ 9167c478bd9Sstevel@tonic-gate if (mmu.pae_hat) { 9177c478bd9Sstevel@tonic-gate if (PFN_ABOVE64G(physmax)) { 9187c478bd9Sstevel@tonic-gate physinstalled -= (physmax - (PFN_64G - 1)); 9197c478bd9Sstevel@tonic-gate physmax = PFN_64G - 1; 9207c478bd9Sstevel@tonic-gate } 9217c478bd9Sstevel@tonic-gate } else { 9227c478bd9Sstevel@tonic-gate if (PFN_ABOVE4G(physmax)) { 9237c478bd9Sstevel@tonic-gate physinstalled -= (physmax - (PFN_4G - 1)); 9247c478bd9Sstevel@tonic-gate physmax = PFN_4G - 1; 9257c478bd9Sstevel@tonic-gate } 9267c478bd9Sstevel@tonic-gate } 9277c478bd9Sstevel@tonic-gate #endif 9287c478bd9Sstevel@tonic-gate 9297c478bd9Sstevel@tonic-gate startup_build_mem_nodes(bootops->boot_mem->physinstalled); 9307c478bd9Sstevel@tonic-gate 9317c478bd9Sstevel@tonic-gate if (BOP_GETPROPLEN(bootops, "enforce-prot-exec") >= 0) { 9327c478bd9Sstevel@tonic-gate int len = BOP_GETPROPLEN(bootops, "enforce-prot-exec"); 9337c478bd9Sstevel@tonic-gate char value[8]; 9347c478bd9Sstevel@tonic-gate 9357c478bd9Sstevel@tonic-gate if (len < 8) 9367c478bd9Sstevel@tonic-gate (void) BOP_GETPROP(bootops, "enforce-prot-exec", value); 9377c478bd9Sstevel@tonic-gate else 9387c478bd9Sstevel@tonic-gate (void) strcpy(value, ""); 9397c478bd9Sstevel@tonic-gate if (strcmp(value, "off") == 0) 9407c478bd9Sstevel@tonic-gate mmu.pt_nx = 0; 9417c478bd9Sstevel@tonic-gate } 9427c478bd9Sstevel@tonic-gate PRM_DEBUG(mmu.pt_nx); 9437c478bd9Sstevel@tonic-gate 9447c478bd9Sstevel@tonic-gate /* 9457c478bd9Sstevel@tonic-gate * We will need page_t's for every page in the system, except for 9467c478bd9Sstevel@tonic-gate * memory mapped at or above above the start of the kernel text segment. 9477c478bd9Sstevel@tonic-gate * 9487c478bd9Sstevel@tonic-gate * pages above e_modtext are attributed to kernel debugger (obp_pages) 9497c478bd9Sstevel@tonic-gate */ 9507c478bd9Sstevel@tonic-gate npages = physinstalled - 1; /* avail_filter() skips page 0, so "- 1" */ 9517c478bd9Sstevel@tonic-gate obp_pages = 0; 9527c478bd9Sstevel@tonic-gate va = KERNEL_TEXT; 953ae115bc7Smrj while (kbm_probe(&va, &len, &pfn, &prot) != 0) { 9547c478bd9Sstevel@tonic-gate npages -= len >> MMU_PAGESHIFT; 9557c478bd9Sstevel@tonic-gate if (va >= (uintptr_t)e_moddata) 9567c478bd9Sstevel@tonic-gate obp_pages += len >> MMU_PAGESHIFT; 9577c478bd9Sstevel@tonic-gate va += len; 9587c478bd9Sstevel@tonic-gate } 9597c478bd9Sstevel@tonic-gate PRM_DEBUG(npages); 9607c478bd9Sstevel@tonic-gate PRM_DEBUG(obp_pages); 9617c478bd9Sstevel@tonic-gate 9627c478bd9Sstevel@tonic-gate /* 96335b1ab99Sjosephb * If physmem is patched to be non-zero, use it instead of the computed 96435b1ab99Sjosephb * value unless it is larger than the actual amount of memory on hand. 9657c478bd9Sstevel@tonic-gate */ 9664944b02eSkchow if (physmem == 0 || physmem > npages) { 9677c478bd9Sstevel@tonic-gate physmem = npages; 9684944b02eSkchow } else if (physmem < npages) { 969f53ad214Skchow orig_npages = npages; 9707c478bd9Sstevel@tonic-gate npages = physmem; 9714944b02eSkchow } 9727c478bd9Sstevel@tonic-gate PRM_DEBUG(physmem); 9737c478bd9Sstevel@tonic-gate 9747c478bd9Sstevel@tonic-gate /* 9757c478bd9Sstevel@tonic-gate * We now compute the sizes of all the initial allocations for 9767c478bd9Sstevel@tonic-gate * structures the kernel needs in order do kmem_alloc(). These 9777c478bd9Sstevel@tonic-gate * include: 9787c478bd9Sstevel@tonic-gate * memsegs 9797c478bd9Sstevel@tonic-gate * memlists 9807c478bd9Sstevel@tonic-gate * page hash table 9817c478bd9Sstevel@tonic-gate * page_t's 9827c478bd9Sstevel@tonic-gate * page coloring data structs 9837c478bd9Sstevel@tonic-gate */ 9847c478bd9Sstevel@tonic-gate memseg_sz = sizeof (struct memseg) * (memblocks + POSS_NEW_FRAGMENTS); 9857c478bd9Sstevel@tonic-gate ADD_TO_ALLOCATIONS(memseg_base, memseg_sz); 9867c478bd9Sstevel@tonic-gate PRM_DEBUG(memseg_sz); 9877c478bd9Sstevel@tonic-gate 9887c478bd9Sstevel@tonic-gate /* 989ae115bc7Smrj * Reserve space for memlists. There's no real good way to know exactly 990ae115bc7Smrj * how much room we'll need, but this should be a good upper bound. 9917c478bd9Sstevel@tonic-gate */ 9927c478bd9Sstevel@tonic-gate memlist_sz = ROUND_UP_PAGE(2 * sizeof (struct memlist) * 9937c478bd9Sstevel@tonic-gate (memblocks + POSS_NEW_FRAGMENTS)); 9947c478bd9Sstevel@tonic-gate ADD_TO_ALLOCATIONS(memlist, memlist_sz); 9957c478bd9Sstevel@tonic-gate PRM_DEBUG(memlist_sz); 9967c478bd9Sstevel@tonic-gate 9977c478bd9Sstevel@tonic-gate /* 9987c478bd9Sstevel@tonic-gate * The page structure hash table size is a power of 2 9997c478bd9Sstevel@tonic-gate * such that the average hash chain length is PAGE_HASHAVELEN. 10007c478bd9Sstevel@tonic-gate */ 10017c478bd9Sstevel@tonic-gate page_hashsz = npages / PAGE_HASHAVELEN; 10027c478bd9Sstevel@tonic-gate page_hashsz = 1 << highbit(page_hashsz); 10037c478bd9Sstevel@tonic-gate pagehash_sz = sizeof (struct page *) * page_hashsz; 10047c478bd9Sstevel@tonic-gate ADD_TO_ALLOCATIONS(page_hash, pagehash_sz); 10057c478bd9Sstevel@tonic-gate PRM_DEBUG(pagehash_sz); 10067c478bd9Sstevel@tonic-gate 10077c478bd9Sstevel@tonic-gate /* 1008ae115bc7Smrj * Set aside room for the page structures themselves. 10097c478bd9Sstevel@tonic-gate */ 1010ae115bc7Smrj PRM_DEBUG(npages); 1011ae115bc7Smrj pp_sz = sizeof (struct page) * npages; 10127c478bd9Sstevel@tonic-gate ADD_TO_ALLOCATIONS(pp_base, pp_sz); 10137c478bd9Sstevel@tonic-gate PRM_DEBUG(pp_sz); 10147c478bd9Sstevel@tonic-gate 10157c478bd9Sstevel@tonic-gate /* 10167c478bd9Sstevel@tonic-gate * determine l2 cache info and memory size for page coloring 10177c478bd9Sstevel@tonic-gate */ 10187c478bd9Sstevel@tonic-gate (void) getl2cacheinfo(CPU, 10197c478bd9Sstevel@tonic-gate &l2cache_sz, &l2cache_linesz, &l2cache_assoc); 10207c478bd9Sstevel@tonic-gate pagecolor_memsz = 10217c478bd9Sstevel@tonic-gate page_coloring_init(l2cache_sz, l2cache_linesz, l2cache_assoc); 10227c478bd9Sstevel@tonic-gate ADD_TO_ALLOCATIONS(pagecolor_mem, pagecolor_memsz); 10237c478bd9Sstevel@tonic-gate PRM_DEBUG(pagecolor_memsz); 10247c478bd9Sstevel@tonic-gate 10257c478bd9Sstevel@tonic-gate page_ctrs_size = page_ctrs_sz(); 10267c478bd9Sstevel@tonic-gate ADD_TO_ALLOCATIONS(page_ctrs_mem, page_ctrs_size); 10277c478bd9Sstevel@tonic-gate PRM_DEBUG(page_ctrs_size); 10287c478bd9Sstevel@tonic-gate 1029*d7d93655Sblakej /* 1030*d7d93655Sblakej * Allocate the array that protects pp->p_selock. 1031*d7d93655Sblakej */ 1032*d7d93655Sblakej pse_shift = size_pse_array(physmem, max_ncpus); 1033*d7d93655Sblakej pse_table_size = 1 << pse_shift; 1034*d7d93655Sblakej pse_table_alloc_size = pse_table_size * sizeof (pad_mutex_t); 1035*d7d93655Sblakej ADD_TO_ALLOCATIONS(pse_mutex, pse_table_alloc_size); 1036*d7d93655Sblakej 1037ae115bc7Smrj #if defined(__amd64) 10387c478bd9Sstevel@tonic-gate valloc_sz = ROUND_UP_LPAGE(valloc_sz); 1039ae115bc7Smrj valloc_base = VALLOC_BASE; 104035b1ab99Sjosephb 104135b1ab99Sjosephb /* 104235b1ab99Sjosephb * The default values of VALLOC_BASE and SEGKPM_BASE should work 104335b1ab99Sjosephb * for values of physmax up to 1 Terabyte. They need adjusting when 104435b1ab99Sjosephb * memory is at addresses above 1 TB. 104535b1ab99Sjosephb */ 104635b1ab99Sjosephb if (physmax + 1 > mmu_btop(TERABYTE)) { 104735b1ab99Sjosephb uint64_t kpm_resv_amount = mmu_ptob(physmax + 1); 104835b1ab99Sjosephb 104935b1ab99Sjosephb /* Round to largest possible pagesize for now */ 105035b1ab99Sjosephb kpm_resv_amount = P2ROUNDUP(kpm_resv_amount, ONE_GIG); 105135b1ab99Sjosephb 105235b1ab99Sjosephb segkpm_base = -(2 * kpm_resv_amount); /* down from top VA */ 105335b1ab99Sjosephb 105435b1ab99Sjosephb /* make sure we leave some space for user apps above hole */ 105535b1ab99Sjosephb segkpm_base = MAX(segkpm_base, AMD64_VA_HOLE_END + TERABYTE); 105635b1ab99Sjosephb if (segkpm_base > SEGKPM_BASE) 105735b1ab99Sjosephb segkpm_base = SEGKPM_BASE; 105835b1ab99Sjosephb PRM_DEBUG(segkpm_base); 105935b1ab99Sjosephb 106035b1ab99Sjosephb valloc_base = segkpm_base + kpm_resv_amount; 106135b1ab99Sjosephb PRM_DEBUG(valloc_base); 106235b1ab99Sjosephb } 1063ae115bc7Smrj #else /* __i386 */ 1064ae115bc7Smrj valloc_base = (uintptr_t)(MISC_VA_BASE - valloc_sz); 1065ae115bc7Smrj valloc_base = P2ALIGN(valloc_base, mmu.level_size[1]); 10667c478bd9Sstevel@tonic-gate PRM_DEBUG(valloc_base); 106735b1ab99Sjosephb #endif /* __i386 */ 1068ae115bc7Smrj 1069ae115bc7Smrj /* 1070ae115bc7Smrj * do all the initial allocations 1071ae115bc7Smrj */ 1072ae115bc7Smrj perform_allocations(); 1073ae115bc7Smrj 1074ae115bc7Smrj /* 1075ae115bc7Smrj * Build phys_install and phys_avail in kernel memspace. 1076ae115bc7Smrj * - phys_install should be all memory in the system. 1077ae115bc7Smrj * - phys_avail is phys_install minus any memory mapped before this 1078ae115bc7Smrj * point above KERNEL_TEXT. 1079ae115bc7Smrj */ 1080ae115bc7Smrj current = phys_install = memlist; 1081ae115bc7Smrj copy_memlist_filter(bootops->boot_mem->physinstalled, ¤t, NULL); 1082ae115bc7Smrj if ((caddr_t)current > (caddr_t)memlist + memlist_sz) 1083ae115bc7Smrj panic("physinstalled was too big!"); 1084ae115bc7Smrj if (prom_debug) 1085ae115bc7Smrj print_memlist("phys_install", phys_install); 1086ae115bc7Smrj 1087ae115bc7Smrj phys_avail = current; 1088ae115bc7Smrj PRM_POINT("Building phys_avail:\n"); 1089ae115bc7Smrj copy_memlist_filter(bootops->boot_mem->physinstalled, ¤t, 1090ae115bc7Smrj avail_filter); 1091ae115bc7Smrj if ((caddr_t)current > (caddr_t)memlist + memlist_sz) 1092ae115bc7Smrj panic("physavail was too big!"); 1093ae115bc7Smrj if (prom_debug) 1094ae115bc7Smrj print_memlist("phys_avail", phys_avail); 1095ae115bc7Smrj 1096ae115bc7Smrj /* 1097ae115bc7Smrj * setup page coloring 1098ae115bc7Smrj */ 1099ae115bc7Smrj page_coloring_setup(pagecolor_mem); 1100ae115bc7Smrj page_lock_init(); /* currently a no-op */ 1101ae115bc7Smrj 1102ae115bc7Smrj /* 1103ae115bc7Smrj * free page list counters 1104ae115bc7Smrj */ 1105ae115bc7Smrj (void) page_ctrs_alloc(page_ctrs_mem); 1106ae115bc7Smrj 1107ae115bc7Smrj /* 1108ae115bc7Smrj * Initialize the page structures from the memory lists. 1109ae115bc7Smrj */ 1110ae115bc7Smrj availrmem_initial = availrmem = freemem = 0; 1111ae115bc7Smrj PRM_POINT("Calling kphysm_init()..."); 1112ae115bc7Smrj npages = kphysm_init(pp_base, npages); 1113ae115bc7Smrj PRM_POINT("kphysm_init() done"); 1114ae115bc7Smrj PRM_DEBUG(npages); 1115ae115bc7Smrj 1116ae115bc7Smrj init_debug_info(); 1117ae115bc7Smrj 1118ae115bc7Smrj /* 1119ae115bc7Smrj * Now that page_t's have been initialized, remove all the 1120ae115bc7Smrj * initial allocation pages from the kernel free page lists. 1121ae115bc7Smrj */ 1122ae115bc7Smrj boot_mapin((caddr_t)valloc_base, valloc_sz); 1123ae115bc7Smrj boot_mapin((caddr_t)GDT_VA, MMU_PAGESIZE); 1124ae115bc7Smrj boot_mapin((caddr_t)DEBUG_INFO_VA, MMU_PAGESIZE); 1125ae115bc7Smrj PRM_POINT("startup_memlist() done"); 1126ae115bc7Smrj 1127ae115bc7Smrj PRM_DEBUG(valloc_sz); 11282cb27123Saguzovsk 1129567d55e1Saguzovsk #if defined(__amd64) 1130567d55e1Saguzovsk if ((availrmem >> (30 - MMU_PAGESHIFT)) >= 1131567d55e1Saguzovsk textrepl_min_gb && l2cache_sz <= 2 << 20) { 1132567d55e1Saguzovsk extern size_t textrepl_size_thresh; 11332cb27123Saguzovsk textrepl_size_thresh = (16 << 20) - 1; 11342cb27123Saguzovsk } 1135567d55e1Saguzovsk #endif 1136ae115bc7Smrj } 1137ae115bc7Smrj 1138ae115bc7Smrj /* 1139ae115bc7Smrj * Layout the kernel's part of address space and initialize kmem allocator. 1140ae115bc7Smrj */ 1141ae115bc7Smrj static void 1142ae115bc7Smrj startup_kmem(void) 1143ae115bc7Smrj { 1144932dc8e5Sdp extern void page_set_colorequiv_arr(void); 1145932dc8e5Sdp 1146ae115bc7Smrj PRM_POINT("startup_kmem() starting..."); 11477c478bd9Sstevel@tonic-gate 11487c478bd9Sstevel@tonic-gate #if defined(__amd64) 11497c478bd9Sstevel@tonic-gate if (eprom_kernelbase && eprom_kernelbase != KERNELBASE) 11507c478bd9Sstevel@tonic-gate cmn_err(CE_NOTE, "!kernelbase cannot be changed on 64-bit " 11517c478bd9Sstevel@tonic-gate "systems."); 115235b1ab99Sjosephb kernelbase = segkpm_base - KERNEL_REDZONE_SIZE; 11537c478bd9Sstevel@tonic-gate core_base = (uintptr_t)COREHEAP_BASE; 1154ae115bc7Smrj core_size = (size_t)MISC_VA_BASE - COREHEAP_BASE; 11557c478bd9Sstevel@tonic-gate #else /* __i386 */ 11567c478bd9Sstevel@tonic-gate /* 11577c478bd9Sstevel@tonic-gate * We configure kernelbase based on: 11587c478bd9Sstevel@tonic-gate * 11597c478bd9Sstevel@tonic-gate * 1. user specified kernelbase via eeprom command. Value cannot exceed 11607c478bd9Sstevel@tonic-gate * KERNELBASE_MAX. we large page align eprom_kernelbase 11617c478bd9Sstevel@tonic-gate * 11627c478bd9Sstevel@tonic-gate * 2. Default to KERNELBASE and adjust to 2X less the size for page_t. 11637c478bd9Sstevel@tonic-gate * On large memory systems we must lower kernelbase to allow 11647c478bd9Sstevel@tonic-gate * enough room for page_t's for all of memory. 11657c478bd9Sstevel@tonic-gate * 11667c478bd9Sstevel@tonic-gate * The value set here, might be changed a little later. 11677c478bd9Sstevel@tonic-gate */ 11687c478bd9Sstevel@tonic-gate if (eprom_kernelbase) { 11697c478bd9Sstevel@tonic-gate kernelbase = eprom_kernelbase & mmu.level_mask[1]; 11707c478bd9Sstevel@tonic-gate if (kernelbase > KERNELBASE_MAX) 11717c478bd9Sstevel@tonic-gate kernelbase = KERNELBASE_MAX; 11727c478bd9Sstevel@tonic-gate } else { 11737c478bd9Sstevel@tonic-gate kernelbase = (uintptr_t)KERNELBASE; 11747c478bd9Sstevel@tonic-gate kernelbase -= ROUND_UP_4MEG(2 * valloc_sz); 11757c478bd9Sstevel@tonic-gate } 11767c478bd9Sstevel@tonic-gate ASSERT((kernelbase & mmu.level_offset[1]) == 0); 1177ae115bc7Smrj core_base = valloc_base; 11787c478bd9Sstevel@tonic-gate core_size = 0; 1179ae115bc7Smrj #endif /* __i386 */ 11807c478bd9Sstevel@tonic-gate 11817c478bd9Sstevel@tonic-gate PRM_DEBUG(core_base); 11827c478bd9Sstevel@tonic-gate PRM_DEBUG(core_size); 1183ae115bc7Smrj PRM_DEBUG(kernelbase); 11847c478bd9Sstevel@tonic-gate 118535b1ab99Sjosephb #if defined(__i386) 11867c478bd9Sstevel@tonic-gate segkp_fromheap = 1; 1187ae115bc7Smrj #endif /* __i386 */ 118835b1ab99Sjosephb 11897c478bd9Sstevel@tonic-gate ekernelheap = (char *)core_base; 1190ae115bc7Smrj PRM_DEBUG(ekernelheap); 11917c478bd9Sstevel@tonic-gate 11927c478bd9Sstevel@tonic-gate /* 11937c478bd9Sstevel@tonic-gate * Now that we know the real value of kernelbase, 11947c478bd9Sstevel@tonic-gate * update variables that were initialized with a value of 11957c478bd9Sstevel@tonic-gate * KERNELBASE (in common/conf/param.c). 11967c478bd9Sstevel@tonic-gate * 11977c478bd9Sstevel@tonic-gate * XXX The problem with this sort of hackery is that the 11987c478bd9Sstevel@tonic-gate * compiler just may feel like putting the const declarations 11997c478bd9Sstevel@tonic-gate * (in param.c) into the .text section. Perhaps they should 12007c478bd9Sstevel@tonic-gate * just be declared as variables there? 12017c478bd9Sstevel@tonic-gate */ 12027c478bd9Sstevel@tonic-gate 12037c478bd9Sstevel@tonic-gate *(uintptr_t *)&_kernelbase = kernelbase; 12047c478bd9Sstevel@tonic-gate *(uintptr_t *)&_userlimit = kernelbase; 1205ccbaea4fSjosephb #if defined(__amd64) 1206ccbaea4fSjosephb *(uintptr_t *)&_userlimit -= KERNELBASE - USERLIMIT; 1207ccbaea4fSjosephb #else 12087c478bd9Sstevel@tonic-gate *(uintptr_t *)&_userlimit32 = _userlimit; 12097c478bd9Sstevel@tonic-gate #endif 12107c478bd9Sstevel@tonic-gate PRM_DEBUG(_kernelbase); 12117c478bd9Sstevel@tonic-gate PRM_DEBUG(_userlimit); 12127c478bd9Sstevel@tonic-gate PRM_DEBUG(_userlimit32); 12137c478bd9Sstevel@tonic-gate 121435b1ab99Sjosephb layout_kernel_va(); 121535b1ab99Sjosephb 121635b1ab99Sjosephb #if defined(__i386) 121735b1ab99Sjosephb /* 121835b1ab99Sjosephb * If segmap is too large we can push the bottom of the kernel heap 121935b1ab99Sjosephb * higher than the base. Or worse, it could exceed the top of the 122035b1ab99Sjosephb * VA space entirely, causing it to wrap around. 122135b1ab99Sjosephb */ 122235b1ab99Sjosephb if (kernelheap >= ekernelheap || (uintptr_t)kernelheap < kernelbase) 122335b1ab99Sjosephb panic("too little address space available for kernelheap," 122435b1ab99Sjosephb " use eeprom for lower kernelbase or smaller segmapsize"); 122535b1ab99Sjosephb #endif /* __i386 */ 122635b1ab99Sjosephb 12277c478bd9Sstevel@tonic-gate /* 12287c478bd9Sstevel@tonic-gate * Initialize the kernel heap. Note 3rd argument must be > 1st. 12297c478bd9Sstevel@tonic-gate */ 123035b1ab99Sjosephb kernelheap_init(kernelheap, ekernelheap, 123135b1ab99Sjosephb kernelheap + MMU_PAGESIZE, 1232ae115bc7Smrj (void *)core_base, (void *)(core_base + core_size)); 12337c478bd9Sstevel@tonic-gate 12347c478bd9Sstevel@tonic-gate /* 12357c478bd9Sstevel@tonic-gate * Initialize kernel memory allocator. 12367c478bd9Sstevel@tonic-gate */ 12377c478bd9Sstevel@tonic-gate kmem_init(); 12387c478bd9Sstevel@tonic-gate 1239932dc8e5Sdp /* 1240932dc8e5Sdp * Factor in colorequiv to check additional 'equivalent' bins 1241932dc8e5Sdp */ 1242932dc8e5Sdp page_set_colorequiv_arr(); 1243932dc8e5Sdp 12447c478bd9Sstevel@tonic-gate /* 12457c478bd9Sstevel@tonic-gate * print this out early so that we know what's going on 12467c478bd9Sstevel@tonic-gate */ 12477c478bd9Sstevel@tonic-gate cmn_err(CE_CONT, "?features: %b\n", x86_feature, FMT_X86_FEATURE); 12487c478bd9Sstevel@tonic-gate 12497c478bd9Sstevel@tonic-gate /* 12507c478bd9Sstevel@tonic-gate * Initialize bp_mapin(). 12517c478bd9Sstevel@tonic-gate */ 12527c478bd9Sstevel@tonic-gate bp_init(MMU_PAGESIZE, HAT_STORECACHING_OK); 12537c478bd9Sstevel@tonic-gate 1254f53ad214Skchow /* 1255f53ad214Skchow * orig_npages is non-zero if physmem has been configured for less 1256f53ad214Skchow * than the available memory. 1257f53ad214Skchow */ 1258f53ad214Skchow if (orig_npages) { 125935b1ab99Sjosephb cmn_err(CE_WARN, "!%slimiting physmem to 0x%lx of 0x%lx pages", 126035b1ab99Sjosephb (npages == PHYSMEM ? "Due to virtual address space " : ""), 126135b1ab99Sjosephb npages, orig_npages); 1262f53ad214Skchow } 12637c478bd9Sstevel@tonic-gate #if defined(__i386) 12647c478bd9Sstevel@tonic-gate if (eprom_kernelbase && (eprom_kernelbase != kernelbase)) 12657c478bd9Sstevel@tonic-gate cmn_err(CE_WARN, "kernelbase value, User specified 0x%lx, " 12667c478bd9Sstevel@tonic-gate "System using 0x%lx", 12677c478bd9Sstevel@tonic-gate (uintptr_t)eprom_kernelbase, (uintptr_t)kernelbase); 12687c478bd9Sstevel@tonic-gate #endif 12697c478bd9Sstevel@tonic-gate 12707c478bd9Sstevel@tonic-gate #ifdef KERNELBASE_ABI_MIN 12717c478bd9Sstevel@tonic-gate if (kernelbase < (uintptr_t)KERNELBASE_ABI_MIN) { 12727c478bd9Sstevel@tonic-gate cmn_err(CE_NOTE, "!kernelbase set to 0x%lx, system is not " 12737c478bd9Sstevel@tonic-gate "i386 ABI compliant.", (uintptr_t)kernelbase); 12747c478bd9Sstevel@tonic-gate } 12757c478bd9Sstevel@tonic-gate #endif 12767c478bd9Sstevel@tonic-gate 1277ae115bc7Smrj PRM_POINT("startup_kmem() done"); 12787c478bd9Sstevel@tonic-gate } 12797c478bd9Sstevel@tonic-gate 12807c478bd9Sstevel@tonic-gate static void 12817c478bd9Sstevel@tonic-gate startup_modules(void) 12827c478bd9Sstevel@tonic-gate { 12837c478bd9Sstevel@tonic-gate unsigned int i; 12847c478bd9Sstevel@tonic-gate extern void prom_setup(void); 12857c478bd9Sstevel@tonic-gate 12867c478bd9Sstevel@tonic-gate PRM_POINT("startup_modules() starting..."); 12877c478bd9Sstevel@tonic-gate /* 12887c478bd9Sstevel@tonic-gate * Initialize ten-micro second timer so that drivers will 12897c478bd9Sstevel@tonic-gate * not get short changed in their init phase. This was 12907c478bd9Sstevel@tonic-gate * not getting called until clkinit which, on fast cpu's 12917c478bd9Sstevel@tonic-gate * caused the drv_usecwait to be way too short. 12927c478bd9Sstevel@tonic-gate */ 12937c478bd9Sstevel@tonic-gate microfind(); 12947c478bd9Sstevel@tonic-gate 12957c478bd9Sstevel@tonic-gate /* 12967c478bd9Sstevel@tonic-gate * Read the GMT lag from /etc/rtc_config. 12977c478bd9Sstevel@tonic-gate */ 1298ae115bc7Smrj sgmtl(process_rtc_config_file()); 12997c478bd9Sstevel@tonic-gate 13007c478bd9Sstevel@tonic-gate /* 13017c478bd9Sstevel@tonic-gate * Calculate default settings of system parameters based upon 13027c478bd9Sstevel@tonic-gate * maxusers, yet allow to be overridden via the /etc/system file. 13037c478bd9Sstevel@tonic-gate */ 13047c478bd9Sstevel@tonic-gate param_calc(0); 13057c478bd9Sstevel@tonic-gate 13067c478bd9Sstevel@tonic-gate mod_setup(); 13077c478bd9Sstevel@tonic-gate 13087c478bd9Sstevel@tonic-gate /* 13097c478bd9Sstevel@tonic-gate * Initialize system parameters. 13107c478bd9Sstevel@tonic-gate */ 13117c478bd9Sstevel@tonic-gate param_init(); 13127c478bd9Sstevel@tonic-gate 13139acbbeafSnn /* 13149acbbeafSnn * Initialize the default brands 13159acbbeafSnn */ 13169acbbeafSnn brand_init(); 13179acbbeafSnn 13187c478bd9Sstevel@tonic-gate /* 13197c478bd9Sstevel@tonic-gate * maxmem is the amount of physical memory we're playing with. 13207c478bd9Sstevel@tonic-gate */ 13217c478bd9Sstevel@tonic-gate maxmem = physmem; 13227c478bd9Sstevel@tonic-gate 13237c478bd9Sstevel@tonic-gate /* 13247c478bd9Sstevel@tonic-gate * Initialize segment management stuff. 13257c478bd9Sstevel@tonic-gate */ 13267c478bd9Sstevel@tonic-gate seg_init(); 13277c478bd9Sstevel@tonic-gate 13287c478bd9Sstevel@tonic-gate if (modload("fs", "specfs") == -1) 13297c478bd9Sstevel@tonic-gate halt("Can't load specfs"); 13307c478bd9Sstevel@tonic-gate 13317c478bd9Sstevel@tonic-gate if (modload("fs", "devfs") == -1) 13327c478bd9Sstevel@tonic-gate halt("Can't load devfs"); 13337c478bd9Sstevel@tonic-gate 1334facf4a8dSllai if (modload("fs", "dev") == -1) 1335facf4a8dSllai halt("Can't load dev"); 1336facf4a8dSllai 133745916cd2Sjpk (void) modloadonly("sys", "lbl_edition"); 133845916cd2Sjpk 13397c478bd9Sstevel@tonic-gate dispinit(); 13407c478bd9Sstevel@tonic-gate 13417c478bd9Sstevel@tonic-gate /* 13427c478bd9Sstevel@tonic-gate * This is needed here to initialize hw_serial[] for cluster booting. 13437c478bd9Sstevel@tonic-gate */ 13447c478bd9Sstevel@tonic-gate if ((i = modload("misc", "sysinit")) != (unsigned int)-1) 13457c478bd9Sstevel@tonic-gate (void) modunload(i); 13467c478bd9Sstevel@tonic-gate else 13477c478bd9Sstevel@tonic-gate cmn_err(CE_CONT, "sysinit load failed"); 13487c478bd9Sstevel@tonic-gate 13497c478bd9Sstevel@tonic-gate /* Read cluster configuration data. */ 13507c478bd9Sstevel@tonic-gate clconf_init(); 13517c478bd9Sstevel@tonic-gate 13527c478bd9Sstevel@tonic-gate /* 13537c478bd9Sstevel@tonic-gate * Create a kernel device tree. First, create rootnex and 13547c478bd9Sstevel@tonic-gate * then invoke bus specific code to probe devices. 13557c478bd9Sstevel@tonic-gate */ 13567c478bd9Sstevel@tonic-gate setup_ddi(); 13577aec1d6eScindi 13587aec1d6eScindi /* 13597aec1d6eScindi * Set up the CPU module subsystem. Modifies the device tree, so it 13607aec1d6eScindi * must be done after setup_ddi(). 13617aec1d6eScindi */ 13627aec1d6eScindi cmi_init(); 13637aec1d6eScindi 13647aec1d6eScindi /* 13657aec1d6eScindi * Initialize the MCA handlers 13667aec1d6eScindi */ 13677aec1d6eScindi if (x86_feature & X86_MCA) 13687aec1d6eScindi cmi_mca_init(); 13697aec1d6eScindi 13707c478bd9Sstevel@tonic-gate /* 13717c478bd9Sstevel@tonic-gate * Fake a prom tree such that /dev/openprom continues to work 13727c478bd9Sstevel@tonic-gate */ 1373ae115bc7Smrj PRM_POINT("startup_modules: calling prom_setup..."); 13747c478bd9Sstevel@tonic-gate prom_setup(); 1375ae115bc7Smrj PRM_POINT("startup_modules: done"); 13767c478bd9Sstevel@tonic-gate 13777c478bd9Sstevel@tonic-gate /* 13787c478bd9Sstevel@tonic-gate * Load all platform specific modules 13797c478bd9Sstevel@tonic-gate */ 1380ae115bc7Smrj PRM_POINT("startup_modules: calling psm_modload..."); 13817c478bd9Sstevel@tonic-gate psm_modload(); 13827c478bd9Sstevel@tonic-gate 13837c478bd9Sstevel@tonic-gate PRM_POINT("startup_modules() done"); 13847c478bd9Sstevel@tonic-gate } 13857c478bd9Sstevel@tonic-gate 1386ae115bc7Smrj /* 1387ae115bc7Smrj * claim a "setaside" boot page for use in the kernel 1388ae115bc7Smrj */ 1389ae115bc7Smrj page_t * 1390ae115bc7Smrj boot_claim_page(pfn_t pfn) 13917c478bd9Sstevel@tonic-gate { 1392ae115bc7Smrj page_t *pp; 13937c478bd9Sstevel@tonic-gate 1394ae115bc7Smrj pp = page_numtopp_nolock(pfn); 1395ae115bc7Smrj ASSERT(pp != NULL); 13967c478bd9Sstevel@tonic-gate 1397ae115bc7Smrj if (PP_ISBOOTPAGES(pp)) { 1398ae115bc7Smrj if (pp->p_next != NULL) 1399ae115bc7Smrj pp->p_next->p_prev = pp->p_prev; 1400ae115bc7Smrj if (pp->p_prev == NULL) 1401ae115bc7Smrj bootpages = pp->p_next; 1402ae115bc7Smrj else 1403ae115bc7Smrj pp->p_prev->p_next = pp->p_next; 1404ae115bc7Smrj } else { 1405ae115bc7Smrj /* 1406ae115bc7Smrj * htable_attach() expects a base pagesize page 1407ae115bc7Smrj */ 1408ae115bc7Smrj if (pp->p_szc != 0) 1409ae115bc7Smrj page_boot_demote(pp); 1410ae115bc7Smrj pp = page_numtopp(pfn, SE_EXCL); 1411ae115bc7Smrj } 1412ae115bc7Smrj return (pp); 14137c478bd9Sstevel@tonic-gate } 14147c478bd9Sstevel@tonic-gate 14157c478bd9Sstevel@tonic-gate /* 14167c478bd9Sstevel@tonic-gate * Walk through the pagetables looking for pages mapped in by boot. If the 14177c478bd9Sstevel@tonic-gate * setaside flag is set the pages are expected to be returned to the 14187c478bd9Sstevel@tonic-gate * kernel later in boot, so we add them to the bootpages list. 14197c478bd9Sstevel@tonic-gate */ 14207c478bd9Sstevel@tonic-gate static void 14217c478bd9Sstevel@tonic-gate protect_boot_range(uintptr_t low, uintptr_t high, int setaside) 14227c478bd9Sstevel@tonic-gate { 14237c478bd9Sstevel@tonic-gate uintptr_t va = low; 14247c478bd9Sstevel@tonic-gate size_t len; 14257c478bd9Sstevel@tonic-gate uint_t prot; 14267c478bd9Sstevel@tonic-gate pfn_t pfn; 14277c478bd9Sstevel@tonic-gate page_t *pp; 14287c478bd9Sstevel@tonic-gate pgcnt_t boot_protect_cnt = 0; 14297c478bd9Sstevel@tonic-gate 1430ae115bc7Smrj while (kbm_probe(&va, &len, &pfn, &prot) != 0 && va < high) { 14317c478bd9Sstevel@tonic-gate if (va + len >= high) 14327c478bd9Sstevel@tonic-gate panic("0x%lx byte mapping at 0x%p exceeds boot's " 14337c478bd9Sstevel@tonic-gate "legal range.", len, (void *)va); 14347c478bd9Sstevel@tonic-gate 14357c478bd9Sstevel@tonic-gate while (len > 0) { 14367c478bd9Sstevel@tonic-gate pp = page_numtopp_alloc(pfn); 14377c478bd9Sstevel@tonic-gate if (pp != NULL) { 14387c478bd9Sstevel@tonic-gate if (setaside == 0) 14397c478bd9Sstevel@tonic-gate panic("Unexpected mapping by boot. " 14407c478bd9Sstevel@tonic-gate "addr=%p pfn=%lx\n", 14417c478bd9Sstevel@tonic-gate (void *)va, pfn); 14427c478bd9Sstevel@tonic-gate 14437c478bd9Sstevel@tonic-gate pp->p_next = bootpages; 1444ae115bc7Smrj pp->p_prev = NULL; 1445ae115bc7Smrj PP_SETBOOTPAGES(pp); 1446ae115bc7Smrj if (bootpages != NULL) { 1447ae115bc7Smrj bootpages->p_prev = pp; 1448ae115bc7Smrj } 14497c478bd9Sstevel@tonic-gate bootpages = pp; 14507c478bd9Sstevel@tonic-gate ++boot_protect_cnt; 14517c478bd9Sstevel@tonic-gate } 14527c478bd9Sstevel@tonic-gate 14537c478bd9Sstevel@tonic-gate ++pfn; 14547c478bd9Sstevel@tonic-gate len -= MMU_PAGESIZE; 14557c478bd9Sstevel@tonic-gate va += MMU_PAGESIZE; 14567c478bd9Sstevel@tonic-gate } 14577c478bd9Sstevel@tonic-gate } 14587c478bd9Sstevel@tonic-gate PRM_DEBUG(boot_protect_cnt); 14597c478bd9Sstevel@tonic-gate } 14607c478bd9Sstevel@tonic-gate 1461ae115bc7Smrj /* 146235b1ab99Sjosephb * 1463ae115bc7Smrj */ 14647c478bd9Sstevel@tonic-gate static void 146535b1ab99Sjosephb layout_kernel_va(void) 14667c478bd9Sstevel@tonic-gate { 146735b1ab99Sjosephb PRM_POINT("layout_kernel_va() starting..."); 14687c478bd9Sstevel@tonic-gate /* 1469ae115bc7Smrj * Establish the final size of the kernel's heap, size of segmap, 1470ae115bc7Smrj * segkp, etc. 14717c478bd9Sstevel@tonic-gate */ 14727c478bd9Sstevel@tonic-gate 14737c478bd9Sstevel@tonic-gate #if defined(__amd64) 14747c478bd9Sstevel@tonic-gate 147535b1ab99Sjosephb kpm_vbase = (caddr_t)segkpm_base; 147635b1ab99Sjosephb kpm_size = ROUND_UP_LPAGE(mmu_ptob(physmax + 1)); 147735b1ab99Sjosephb if ((uintptr_t)kpm_vbase + kpm_size > (uintptr_t)valloc_base) 147835b1ab99Sjosephb panic("not enough room for kpm!"); 1479ae115bc7Smrj PRM_DEBUG(kpm_size); 1480ae115bc7Smrj PRM_DEBUG(kpm_vbase); 14817c478bd9Sstevel@tonic-gate 14827c478bd9Sstevel@tonic-gate /* 1483ae115bc7Smrj * By default we create a seg_kp in 64 bit kernels, it's a little 1484ae115bc7Smrj * faster to access than embedding it in the heap. 14857c478bd9Sstevel@tonic-gate */ 1486ae115bc7Smrj segkp_base = (caddr_t)valloc_base + valloc_sz; 14877c478bd9Sstevel@tonic-gate if (!segkp_fromheap) { 14887c478bd9Sstevel@tonic-gate size_t sz = mmu_ptob(segkpsize); 14897c478bd9Sstevel@tonic-gate 14907c478bd9Sstevel@tonic-gate /* 1491ae115bc7Smrj * determine size of segkp 14927c478bd9Sstevel@tonic-gate */ 14937c478bd9Sstevel@tonic-gate if (sz < SEGKPMINSIZE || sz > SEGKPMAXSIZE) { 14947c478bd9Sstevel@tonic-gate sz = SEGKPDEFSIZE; 14957c478bd9Sstevel@tonic-gate cmn_err(CE_WARN, "!Illegal value for segkpsize. " 14967c478bd9Sstevel@tonic-gate "segkpsize has been reset to %ld pages", 14977c478bd9Sstevel@tonic-gate mmu_btop(sz)); 14987c478bd9Sstevel@tonic-gate } 14997c478bd9Sstevel@tonic-gate sz = MIN(sz, MAX(SEGKPMINSIZE, mmu_ptob(physmem))); 15007c478bd9Sstevel@tonic-gate 15017c478bd9Sstevel@tonic-gate segkpsize = mmu_btop(ROUND_UP_LPAGE(sz)); 15027c478bd9Sstevel@tonic-gate } 1503ae115bc7Smrj PRM_DEBUG(segkp_base); 1504ae115bc7Smrj PRM_DEBUG(segkpsize); 15057c478bd9Sstevel@tonic-gate 150635b1ab99Sjosephb /* 150735b1ab99Sjosephb * segzio is used for ZFS cached data. It uses a distinct VA 150835b1ab99Sjosephb * segment (from kernel heap) so that we can easily tell not to 150935b1ab99Sjosephb * include it in kernel crash dumps on 64 bit kernels. The trick is 151035b1ab99Sjosephb * to give it lots of VA, but not constrain the kernel heap. 151135b1ab99Sjosephb * We scale the size of segzio linearly with physmem up to 151235b1ab99Sjosephb * SEGZIOMAXSIZE. Above that amount it scales at 50% of physmem. 151335b1ab99Sjosephb */ 1514ae115bc7Smrj segzio_base = segkp_base + mmu_ptob(segkpsize); 1515ae115bc7Smrj if (segzio_fromheap) { 1516ae115bc7Smrj segziosize = 0; 1517ae115bc7Smrj } else { 151835b1ab99Sjosephb size_t physmem_size = mmu_ptob(physmem); 151935b1ab99Sjosephb size_t size = (segziosize == 0) ? 152035b1ab99Sjosephb physmem_size : mmu_ptob(segziosize); 1521ad23a2dbSjohansen 152235b1ab99Sjosephb if (size < SEGZIOMINSIZE) 1523ad23a2dbSjohansen size = SEGZIOMINSIZE; 152435b1ab99Sjosephb if (size > SEGZIOMAXSIZE) { 1525a778305aSjohansen size = SEGZIOMAXSIZE; 152635b1ab99Sjosephb if (physmem_size > size) 152735b1ab99Sjosephb size += (physmem_size - size) / 2; 1528ad23a2dbSjohansen } 1529ad23a2dbSjohansen segziosize = mmu_btop(ROUND_UP_LPAGE(size)); 1530ad23a2dbSjohansen } 1531ae115bc7Smrj PRM_DEBUG(segziosize); 1532ae115bc7Smrj PRM_DEBUG(segzio_base); 1533ad23a2dbSjohansen 15347c478bd9Sstevel@tonic-gate /* 1535ae115bc7Smrj * Put the range of VA for device mappings next, kmdb knows to not 1536ae115bc7Smrj * grep in this range of addresses. 15377c478bd9Sstevel@tonic-gate */ 1538ae115bc7Smrj toxic_addr = 1539ae115bc7Smrj ROUND_UP_LPAGE((uintptr_t)segzio_base + mmu_ptob(segziosize)); 15407c478bd9Sstevel@tonic-gate PRM_DEBUG(toxic_addr); 1541ae115bc7Smrj segmap_start = ROUND_UP_LPAGE(toxic_addr + toxic_size); 1542ae115bc7Smrj #else /* __i386 */ 1543ae115bc7Smrj segmap_start = ROUND_UP_LPAGE(kernelbase); 1544ae115bc7Smrj #endif /* __i386 */ 1545ae115bc7Smrj PRM_DEBUG(segmap_start); 15467c478bd9Sstevel@tonic-gate 15477c478bd9Sstevel@tonic-gate /* 15487c478bd9Sstevel@tonic-gate * Users can change segmapsize through eeprom or /etc/system. 15497c478bd9Sstevel@tonic-gate * If the variable is tuned through eeprom, there is no upper 15507c478bd9Sstevel@tonic-gate * bound on the size of segmap. If it is tuned through 15517c478bd9Sstevel@tonic-gate * /etc/system on 32-bit systems, it must be no larger than we 15527c478bd9Sstevel@tonic-gate * planned for in startup_memlist(). 15537c478bd9Sstevel@tonic-gate */ 15547c478bd9Sstevel@tonic-gate segmapsize = MAX(ROUND_UP_LPAGE(segmapsize), SEGMAPDEFAULT); 15557c478bd9Sstevel@tonic-gate 15567c478bd9Sstevel@tonic-gate #if defined(__i386) 15577c478bd9Sstevel@tonic-gate /* 15587c478bd9Sstevel@tonic-gate * 32-bit systems don't have segkpm or segkp, so segmap appears at 15597c478bd9Sstevel@tonic-gate * the bottom of the kernel's address range. Set aside space for a 1560ae115bc7Smrj * small red zone just below the start of segmap. 15617c478bd9Sstevel@tonic-gate */ 1562ae115bc7Smrj segmap_start += KERNEL_REDZONE_SIZE; 15637c478bd9Sstevel@tonic-gate segmapsize -= KERNEL_REDZONE_SIZE; 15647c478bd9Sstevel@tonic-gate #endif 15657c478bd9Sstevel@tonic-gate 1566ae115bc7Smrj PRM_DEBUG(segmap_start); 15677c478bd9Sstevel@tonic-gate PRM_DEBUG(segmapsize); 156835b1ab99Sjosephb kernelheap = (caddr_t)ROUND_UP_LPAGE(segmap_start + segmapsize); 156935b1ab99Sjosephb PRM_DEBUG(kernelheap); 157035b1ab99Sjosephb PRM_POINT("layout_kernel_va() done..."); 157135b1ab99Sjosephb } 157235b1ab99Sjosephb 157335b1ab99Sjosephb /* 157435b1ab99Sjosephb * Finish initializing the VM system, now that we are no longer 157535b1ab99Sjosephb * relying on the boot time memory allocators. 157635b1ab99Sjosephb */ 157735b1ab99Sjosephb static void 157835b1ab99Sjosephb startup_vm(void) 157935b1ab99Sjosephb { 158035b1ab99Sjosephb struct segmap_crargs a; 158135b1ab99Sjosephb 158235b1ab99Sjosephb extern int use_brk_lpg, use_stk_lpg; 158335b1ab99Sjosephb 158435b1ab99Sjosephb PRM_POINT("startup_vm() starting..."); 158535b1ab99Sjosephb 158635b1ab99Sjosephb /* 158735b1ab99Sjosephb * Initialize the hat layer. 158835b1ab99Sjosephb */ 158935b1ab99Sjosephb hat_init(); 15907c478bd9Sstevel@tonic-gate 1591ae115bc7Smrj /* 1592ae115bc7Smrj * Do final allocations of HAT data structures that need to 1593ae115bc7Smrj * be allocated before quiescing the boot loader. 1594ae115bc7Smrj */ 1595ae115bc7Smrj PRM_POINT("Calling hat_kern_alloc()..."); 1596ae115bc7Smrj hat_kern_alloc((caddr_t)segmap_start, segmapsize, ekernelheap); 1597ae115bc7Smrj PRM_POINT("hat_kern_alloc() done"); 1598ae115bc7Smrj 1599ae115bc7Smrj /* 1600ae115bc7Smrj * Setup MTRR (Memory type range registers) 1601ae115bc7Smrj */ 1602ae115bc7Smrj setup_mtrr(); 1603ae115bc7Smrj 1604ae115bc7Smrj /* 1605ae115bc7Smrj * The next two loops are done in distinct steps in order 1606ae115bc7Smrj * to be sure that any page that is doubly mapped (both above 1607ae115bc7Smrj * KERNEL_TEXT and below kernelbase) is dealt with correctly. 1608ae115bc7Smrj * Note this may never happen, but it might someday. 1609ae115bc7Smrj */ 1610ae115bc7Smrj bootpages = NULL; 1611ae115bc7Smrj PRM_POINT("Protecting boot pages"); 1612ae115bc7Smrj 1613ae115bc7Smrj /* 1614ae115bc7Smrj * Protect any pages mapped above KERNEL_TEXT that somehow have 1615ae115bc7Smrj * page_t's. This can only happen if something weird allocated 1616ae115bc7Smrj * in this range (like kadb/kmdb). 1617ae115bc7Smrj */ 1618ae115bc7Smrj protect_boot_range(KERNEL_TEXT, (uintptr_t)-1, 0); 1619ae115bc7Smrj 1620ae115bc7Smrj /* 1621ae115bc7Smrj * Before we can take over memory allocation/mapping from the boot 1622ae115bc7Smrj * loader we must remove from our free page lists any boot allocated 1623ae115bc7Smrj * pages that stay mapped until release_bootstrap(). 1624ae115bc7Smrj */ 1625ae115bc7Smrj protect_boot_range(0, kernelbase, 1); 1626ae115bc7Smrj 1627ae115bc7Smrj /* 1628ae115bc7Smrj * Switch to running on regular HAT (not boot_mmu) 1629ae115bc7Smrj */ 1630ae115bc7Smrj PRM_POINT("Calling hat_kern_setup()..."); 1631ae115bc7Smrj hat_kern_setup(); 1632ae115bc7Smrj 1633ae115bc7Smrj /* 1634ae115bc7Smrj * It is no longer safe to call BOP_ALLOC(), so make sure we don't. 1635ae115bc7Smrj */ 1636ae115bc7Smrj bop_no_more_mem(); 1637ae115bc7Smrj 1638ae115bc7Smrj PRM_POINT("hat_kern_setup() done"); 1639ae115bc7Smrj 1640ae115bc7Smrj hat_cpu_online(CPU); 1641ae115bc7Smrj 16427c478bd9Sstevel@tonic-gate /* 16437c478bd9Sstevel@tonic-gate * Initialize VM system 16447c478bd9Sstevel@tonic-gate */ 16457c478bd9Sstevel@tonic-gate PRM_POINT("Calling kvm_init()..."); 16467c478bd9Sstevel@tonic-gate kvm_init(); 16477c478bd9Sstevel@tonic-gate PRM_POINT("kvm_init() done"); 16487c478bd9Sstevel@tonic-gate 16497c478bd9Sstevel@tonic-gate /* 16507c478bd9Sstevel@tonic-gate * Tell kmdb that the VM system is now working 16517c478bd9Sstevel@tonic-gate */ 16527c478bd9Sstevel@tonic-gate if (boothowto & RB_DEBUG) 16537c478bd9Sstevel@tonic-gate kdi_dvec_vmready(); 16547c478bd9Sstevel@tonic-gate 16557c478bd9Sstevel@tonic-gate /* 16567c478bd9Sstevel@tonic-gate * Mangle the brand string etc. 16577c478bd9Sstevel@tonic-gate */ 16587c478bd9Sstevel@tonic-gate cpuid_pass3(CPU); 16597c478bd9Sstevel@tonic-gate 16607c478bd9Sstevel@tonic-gate #if defined(__amd64) 16617c478bd9Sstevel@tonic-gate 16627c478bd9Sstevel@tonic-gate /* 16637c478bd9Sstevel@tonic-gate * Create the device arena for toxic (to dtrace/kmdb) mappings. 16647c478bd9Sstevel@tonic-gate */ 16657c478bd9Sstevel@tonic-gate device_arena = vmem_create("device", (void *)toxic_addr, 16667c478bd9Sstevel@tonic-gate toxic_size, MMU_PAGESIZE, NULL, NULL, NULL, 0, VM_SLEEP); 16677c478bd9Sstevel@tonic-gate 16687c478bd9Sstevel@tonic-gate #else /* __i386 */ 16697c478bd9Sstevel@tonic-gate 16707c478bd9Sstevel@tonic-gate /* 16717c478bd9Sstevel@tonic-gate * allocate the bit map that tracks toxic pages 16727c478bd9Sstevel@tonic-gate */ 1673ae115bc7Smrj toxic_bit_map_len = btop((ulong_t)(valloc_base - kernelbase)); 16747c478bd9Sstevel@tonic-gate PRM_DEBUG(toxic_bit_map_len); 16757c478bd9Sstevel@tonic-gate toxic_bit_map = 16767c478bd9Sstevel@tonic-gate kmem_zalloc(BT_SIZEOFMAP(toxic_bit_map_len), KM_NOSLEEP); 16777c478bd9Sstevel@tonic-gate ASSERT(toxic_bit_map != NULL); 16787c478bd9Sstevel@tonic-gate PRM_DEBUG(toxic_bit_map); 16797c478bd9Sstevel@tonic-gate 16807c478bd9Sstevel@tonic-gate #endif /* __i386 */ 16817c478bd9Sstevel@tonic-gate 16827c478bd9Sstevel@tonic-gate 16837c478bd9Sstevel@tonic-gate /* 16847c478bd9Sstevel@tonic-gate * Now that we've got more VA, as well as the ability to allocate from 16857c478bd9Sstevel@tonic-gate * it, tell the debugger. 16867c478bd9Sstevel@tonic-gate */ 16877c478bd9Sstevel@tonic-gate if (boothowto & RB_DEBUG) 16887c478bd9Sstevel@tonic-gate kdi_dvec_memavail(); 16897c478bd9Sstevel@tonic-gate 16907c478bd9Sstevel@tonic-gate /* 16917c478bd9Sstevel@tonic-gate * The following code installs a special page fault handler (#pf) 16927c478bd9Sstevel@tonic-gate * to work around a pentium bug. 16937c478bd9Sstevel@tonic-gate */ 16947c478bd9Sstevel@tonic-gate #if !defined(__amd64) 16957c478bd9Sstevel@tonic-gate if (x86_type == X86_TYPE_P5) { 1696ae115bc7Smrj desctbr_t idtr; 16977c478bd9Sstevel@tonic-gate gate_desc_t *newidt; 1698ae115bc7Smrj struct machcpu *mcpu = &CPU->cpu_m; 16997c478bd9Sstevel@tonic-gate 17007c478bd9Sstevel@tonic-gate if ((newidt = kmem_zalloc(MMU_PAGESIZE, KM_NOSLEEP)) == NULL) 17017c478bd9Sstevel@tonic-gate panic("failed to install pentium_pftrap"); 17027c478bd9Sstevel@tonic-gate 17037c478bd9Sstevel@tonic-gate bcopy(idt0, newidt, sizeof (idt0)); 17047c478bd9Sstevel@tonic-gate set_gatesegd(&newidt[T_PGFLT], &pentium_pftrap, 1705ae115bc7Smrj KCS_SEL, SDT_SYSIGT, SEL_KPL); 17067c478bd9Sstevel@tonic-gate 17077c478bd9Sstevel@tonic-gate (void) as_setprot(&kas, (caddr_t)newidt, MMU_PAGESIZE, 17087c478bd9Sstevel@tonic-gate PROT_READ|PROT_EXEC); 17097c478bd9Sstevel@tonic-gate 1710ae115bc7Smrj mcpu->mcpu_idt = newidt; 1711ae115bc7Smrj idtr.dtr_base = (uintptr_t)mcpu->mcpu_idt; 1712ae115bc7Smrj idtr.dtr_limit = sizeof (idt0) - 1; 1713ae115bc7Smrj wr_idtr(&idtr); 17147c478bd9Sstevel@tonic-gate } 17157c478bd9Sstevel@tonic-gate #endif /* !__amd64 */ 17167c478bd9Sstevel@tonic-gate 17177c478bd9Sstevel@tonic-gate /* 17187c478bd9Sstevel@tonic-gate * Map page pfn=0 for drivers, such as kd, that need to pick up 17197c478bd9Sstevel@tonic-gate * parameters left there by controllers/BIOS. 17207c478bd9Sstevel@tonic-gate */ 17217c478bd9Sstevel@tonic-gate PRM_POINT("setup up p0_va"); 17227c478bd9Sstevel@tonic-gate p0_va = i86devmap(0, 1, PROT_READ); 17237c478bd9Sstevel@tonic-gate PRM_DEBUG(p0_va); 17247c478bd9Sstevel@tonic-gate 17257c478bd9Sstevel@tonic-gate cmn_err(CE_CONT, "?mem = %luK (0x%lx)\n", 17267c478bd9Sstevel@tonic-gate physinstalled << (MMU_PAGESHIFT - 10), ptob(physinstalled)); 17277c478bd9Sstevel@tonic-gate 172883f9b804Skchow /* 172983f9b804Skchow * disable automatic large pages for small memory systems or 173083f9b804Skchow * when the disable flag is set. 173183f9b804Skchow */ 1732ec25b48fSsusans if (!auto_lpg_disable && mmu.max_page_level > 0) { 1733ec25b48fSsusans max_uheap_lpsize = LEVEL_SIZE(1); 1734ec25b48fSsusans max_ustack_lpsize = LEVEL_SIZE(1); 1735ec25b48fSsusans max_privmap_lpsize = LEVEL_SIZE(1); 1736ec25b48fSsusans max_uidata_lpsize = LEVEL_SIZE(1); 1737ec25b48fSsusans max_utext_lpsize = LEVEL_SIZE(1); 1738ec25b48fSsusans max_shm_lpsize = LEVEL_SIZE(1); 1739ec25b48fSsusans } 1740ec25b48fSsusans if (physmem < privm_lpg_min_physmem || mmu.max_page_level == 0 || 1741ec25b48fSsusans auto_lpg_disable) { 1742beb1bda0Sdavemq use_brk_lpg = 0; 1743beb1bda0Sdavemq use_stk_lpg = 0; 1744ec25b48fSsusans } 1745ec25b48fSsusans if (mmu.max_page_level > 0) { 1746ec25b48fSsusans mcntl0_lpsize = LEVEL_SIZE(1); 1747beb1bda0Sdavemq } 1748beb1bda0Sdavemq 17497c478bd9Sstevel@tonic-gate PRM_POINT("Calling hat_init_finish()..."); 17507c478bd9Sstevel@tonic-gate hat_init_finish(); 17517c478bd9Sstevel@tonic-gate PRM_POINT("hat_init_finish() done"); 17527c478bd9Sstevel@tonic-gate 17537c478bd9Sstevel@tonic-gate /* 17547c478bd9Sstevel@tonic-gate * Initialize the segkp segment type. 17557c478bd9Sstevel@tonic-gate */ 17567c478bd9Sstevel@tonic-gate rw_enter(&kas.a_lock, RW_WRITER); 1757ae115bc7Smrj PRM_POINT("Attaching segkp"); 1758ae115bc7Smrj if (segkp_fromheap) { 17597c478bd9Sstevel@tonic-gate segkp->s_as = &kas; 1760ae115bc7Smrj } else if (seg_attach(&kas, (caddr_t)segkp_base, mmu_ptob(segkpsize), 1761ae115bc7Smrj segkp) < 0) { 1762ae115bc7Smrj panic("startup: cannot attach segkp"); 1763ae115bc7Smrj /*NOTREACHED*/ 17647c478bd9Sstevel@tonic-gate } 1765ae115bc7Smrj PRM_POINT("Doing segkp_create()"); 17667c478bd9Sstevel@tonic-gate if (segkp_create(segkp) != 0) { 17677c478bd9Sstevel@tonic-gate panic("startup: segkp_create failed"); 17687c478bd9Sstevel@tonic-gate /*NOTREACHED*/ 17697c478bd9Sstevel@tonic-gate } 17707c478bd9Sstevel@tonic-gate PRM_DEBUG(segkp); 17717c478bd9Sstevel@tonic-gate rw_exit(&kas.a_lock); 17727c478bd9Sstevel@tonic-gate 17737c478bd9Sstevel@tonic-gate /* 17747c478bd9Sstevel@tonic-gate * kpm segment 17757c478bd9Sstevel@tonic-gate */ 17767c478bd9Sstevel@tonic-gate segmap_kpm = 0; 17777c478bd9Sstevel@tonic-gate if (kpm_desired) { 17787c478bd9Sstevel@tonic-gate kpm_init(); 17797c478bd9Sstevel@tonic-gate kpm_enable = 1; 1780a5652762Spraks vpm_enable = 1; 17817c478bd9Sstevel@tonic-gate } 17827c478bd9Sstevel@tonic-gate 17837c478bd9Sstevel@tonic-gate /* 17847c478bd9Sstevel@tonic-gate * Now create segmap segment. 17857c478bd9Sstevel@tonic-gate */ 17867c478bd9Sstevel@tonic-gate rw_enter(&kas.a_lock, RW_WRITER); 1787ae115bc7Smrj if (seg_attach(&kas, (caddr_t)segmap_start, segmapsize, segmap) < 0) { 1788ae115bc7Smrj panic("cannot attach segmap"); 17897c478bd9Sstevel@tonic-gate /*NOTREACHED*/ 17907c478bd9Sstevel@tonic-gate } 1791ae115bc7Smrj PRM_DEBUG(segmap); 17927c478bd9Sstevel@tonic-gate 17937c478bd9Sstevel@tonic-gate a.prot = PROT_READ | PROT_WRITE; 17947c478bd9Sstevel@tonic-gate a.shmsize = 0; 17957c478bd9Sstevel@tonic-gate a.nfreelist = segmapfreelists; 17967c478bd9Sstevel@tonic-gate 1797ae115bc7Smrj if (segmap_create(segmap, (caddr_t)&a) != 0) 1798ae115bc7Smrj panic("segmap_create segmap"); 17997c478bd9Sstevel@tonic-gate rw_exit(&kas.a_lock); 18007c478bd9Sstevel@tonic-gate 18017c478bd9Sstevel@tonic-gate setup_vaddr_for_ppcopy(CPU); 18027c478bd9Sstevel@tonic-gate 18037c478bd9Sstevel@tonic-gate segdev_init(); 18047c478bd9Sstevel@tonic-gate pmem_init(); 1805ae115bc7Smrj 18067c478bd9Sstevel@tonic-gate PRM_POINT("startup_vm() done"); 18077c478bd9Sstevel@tonic-gate } 18087c478bd9Sstevel@tonic-gate 1809ae115bc7Smrj /* 1810ae115bc7Smrj * Load a tod module for the non-standard tod part found on this system. 1811ae115bc7Smrj */ 1812ae115bc7Smrj static void 1813ae115bc7Smrj load_tod_module(char *todmod) 1814ae115bc7Smrj { 1815ae115bc7Smrj if (modload("tod", todmod) == -1) 1816ae115bc7Smrj halt("Can't load TOD module"); 1817ae115bc7Smrj } 1818ae115bc7Smrj 18197c478bd9Sstevel@tonic-gate static void 18207c478bd9Sstevel@tonic-gate startup_end(void) 18217c478bd9Sstevel@tonic-gate { 18227c478bd9Sstevel@tonic-gate extern void setx86isalist(void); 18237c478bd9Sstevel@tonic-gate 18247c478bd9Sstevel@tonic-gate PRM_POINT("startup_end() starting..."); 18257c478bd9Sstevel@tonic-gate 18267c478bd9Sstevel@tonic-gate /* 18277c478bd9Sstevel@tonic-gate * Perform tasks that get done after most of the VM 18287c478bd9Sstevel@tonic-gate * initialization has been done but before the clock 18297c478bd9Sstevel@tonic-gate * and other devices get started. 18307c478bd9Sstevel@tonic-gate */ 18317c478bd9Sstevel@tonic-gate kern_setup1(); 18327c478bd9Sstevel@tonic-gate 18337c478bd9Sstevel@tonic-gate /* 18347c478bd9Sstevel@tonic-gate * Perform CPC initialization for this CPU. 18357c478bd9Sstevel@tonic-gate */ 18367c478bd9Sstevel@tonic-gate kcpc_hw_init(CPU); 18377c478bd9Sstevel@tonic-gate 1838ee88d2b9Skchow #if defined(OPTERON_WORKAROUND_6323525) 1839ee88d2b9Skchow if (opteron_workaround_6323525) 1840ee88d2b9Skchow patch_workaround_6323525(); 1841ee88d2b9Skchow #endif 1842ae115bc7Smrj /* 1843ae115bc7Smrj * If needed, load TOD module now so that ddi_get_time(9F) etc. work 1844ae115bc7Smrj * (For now, "needed" is defined as set tod_module_name in /etc/system) 1845ae115bc7Smrj */ 1846ae115bc7Smrj if (tod_module_name != NULL) { 1847ae115bc7Smrj PRM_POINT("load_tod_module()"); 1848ae115bc7Smrj load_tod_module(tod_module_name); 1849ae115bc7Smrj } 1850ae115bc7Smrj 18517c478bd9Sstevel@tonic-gate /* 18527c478bd9Sstevel@tonic-gate * Configure the system. 18537c478bd9Sstevel@tonic-gate */ 18547c478bd9Sstevel@tonic-gate PRM_POINT("Calling configure()..."); 18557c478bd9Sstevel@tonic-gate configure(); /* set up devices */ 18567c478bd9Sstevel@tonic-gate PRM_POINT("configure() done"); 18577c478bd9Sstevel@tonic-gate 18587c478bd9Sstevel@tonic-gate /* 18597c478bd9Sstevel@tonic-gate * Set the isa_list string to the defined instruction sets we 18607c478bd9Sstevel@tonic-gate * support. 18617c478bd9Sstevel@tonic-gate */ 18627c478bd9Sstevel@tonic-gate setx86isalist(); 1863100b72f4Sandrei cpu_intr_alloc(CPU, NINTR_THREADS); 18647c478bd9Sstevel@tonic-gate psm_install(); 18657c478bd9Sstevel@tonic-gate 18667c478bd9Sstevel@tonic-gate /* 18677c478bd9Sstevel@tonic-gate * We're done with bootops. We don't unmap the bootstrap yet because 18687c478bd9Sstevel@tonic-gate * we're still using bootsvcs. 18697c478bd9Sstevel@tonic-gate */ 1870ae115bc7Smrj PRM_POINT("NULLing out bootops"); 1871ae115bc7Smrj *bootopsp = (struct bootops *)NULL; 18727c478bd9Sstevel@tonic-gate bootops = (struct bootops *)NULL; 18737c478bd9Sstevel@tonic-gate 18747c478bd9Sstevel@tonic-gate PRM_POINT("Enabling interrupts"); 18757c478bd9Sstevel@tonic-gate (*picinitf)(); 18767c478bd9Sstevel@tonic-gate sti(); 18777c478bd9Sstevel@tonic-gate 18787c478bd9Sstevel@tonic-gate (void) add_avsoftintr((void *)&softlevel1_hdl, 1, softlevel1, 187935b1ab99Sjosephb "softlevel1", NULL, NULL); /* XXX to be moved later */ 18807c478bd9Sstevel@tonic-gate 18817c478bd9Sstevel@tonic-gate PRM_POINT("startup_end() done"); 18827c478bd9Sstevel@tonic-gate } 18837c478bd9Sstevel@tonic-gate 18847c478bd9Sstevel@tonic-gate extern char hw_serial[]; 18857c478bd9Sstevel@tonic-gate char *_hs1107 = hw_serial; 18867c478bd9Sstevel@tonic-gate ulong_t _bdhs34; 18877c478bd9Sstevel@tonic-gate 18887c478bd9Sstevel@tonic-gate void 18897c478bd9Sstevel@tonic-gate post_startup(void) 18907c478bd9Sstevel@tonic-gate { 18917c478bd9Sstevel@tonic-gate /* 18927c478bd9Sstevel@tonic-gate * Set the system wide, processor-specific flags to be passed 18937c478bd9Sstevel@tonic-gate * to userland via the aux vector for performance hints and 18947c478bd9Sstevel@tonic-gate * instruction set extensions. 18957c478bd9Sstevel@tonic-gate */ 18967c478bd9Sstevel@tonic-gate bind_hwcap(); 18977c478bd9Sstevel@tonic-gate 18987c478bd9Sstevel@tonic-gate /* 1899ae115bc7Smrj * Load the System Management BIOS into the global ksmbios 1900ae115bc7Smrj * handle, if an SMBIOS is present on this system. 19017c478bd9Sstevel@tonic-gate */ 190284ab085aSmws ksmbios = smbios_open(NULL, SMB_VERSION, ksmbios_flags, NULL); 19037c478bd9Sstevel@tonic-gate 19047c478bd9Sstevel@tonic-gate /* 1905ae115bc7Smrj * Startup the memory scrubber. 19067c478bd9Sstevel@tonic-gate */ 190784ab085aSmws memscrub_init(); 19087c478bd9Sstevel@tonic-gate 19097aec1d6eScindi /* 19107aec1d6eScindi * Complete CPU module initialization 19117aec1d6eScindi */ 19127aec1d6eScindi cmi_post_init(); 19137aec1d6eScindi 19147c478bd9Sstevel@tonic-gate /* 191584ab085aSmws * Perform forceloading tasks for /etc/system. 19167c478bd9Sstevel@tonic-gate */ 191784ab085aSmws (void) mod_sysctl(SYS_FORCELOAD, NULL); 19187c478bd9Sstevel@tonic-gate 19197c478bd9Sstevel@tonic-gate /* 19207c478bd9Sstevel@tonic-gate * ON4.0: Force /proc module in until clock interrupt handle fixed 19217c478bd9Sstevel@tonic-gate * ON4.0: This must be fixed or restated in /etc/systems. 19227c478bd9Sstevel@tonic-gate */ 19237c478bd9Sstevel@tonic-gate (void) modload("fs", "procfs"); 19247c478bd9Sstevel@tonic-gate 19257c478bd9Sstevel@tonic-gate #if defined(__i386) 19267c478bd9Sstevel@tonic-gate /* 19277c478bd9Sstevel@tonic-gate * Check for required functional Floating Point hardware, 19287c478bd9Sstevel@tonic-gate * unless FP hardware explicitly disabled. 19297c478bd9Sstevel@tonic-gate */ 19307c478bd9Sstevel@tonic-gate if (fpu_exists && (fpu_pentium_fdivbug || fp_kind == FP_NO)) 19317c478bd9Sstevel@tonic-gate halt("No working FP hardware found"); 19327c478bd9Sstevel@tonic-gate #endif 19337c478bd9Sstevel@tonic-gate 19347c478bd9Sstevel@tonic-gate maxmem = freemem; 19357c478bd9Sstevel@tonic-gate 19367c478bd9Sstevel@tonic-gate add_cpunode2devtree(CPU->cpu_id, CPU->cpu_m.mcpu_cpi); 19377c478bd9Sstevel@tonic-gate } 19387c478bd9Sstevel@tonic-gate 19397c478bd9Sstevel@tonic-gate static int 19407c478bd9Sstevel@tonic-gate pp_in_ramdisk(page_t *pp) 19417c478bd9Sstevel@tonic-gate { 19427c478bd9Sstevel@tonic-gate extern uint64_t ramdisk_start, ramdisk_end; 19437c478bd9Sstevel@tonic-gate 19447c478bd9Sstevel@tonic-gate return ((pp->p_pagenum >= btop(ramdisk_start)) && 19457c478bd9Sstevel@tonic-gate (pp->p_pagenum < btopr(ramdisk_end))); 19467c478bd9Sstevel@tonic-gate } 19477c478bd9Sstevel@tonic-gate 19487c478bd9Sstevel@tonic-gate void 19497c478bd9Sstevel@tonic-gate release_bootstrap(void) 19507c478bd9Sstevel@tonic-gate { 19517c478bd9Sstevel@tonic-gate int root_is_ramdisk; 19527c478bd9Sstevel@tonic-gate page_t *pp; 19537c478bd9Sstevel@tonic-gate extern void kobj_boot_unmountroot(void); 19547c478bd9Sstevel@tonic-gate extern dev_t rootdev; 19557c478bd9Sstevel@tonic-gate 19567c478bd9Sstevel@tonic-gate /* unmount boot ramdisk and release kmem usage */ 19577c478bd9Sstevel@tonic-gate kobj_boot_unmountroot(); 19587c478bd9Sstevel@tonic-gate 19597c478bd9Sstevel@tonic-gate /* 19607c478bd9Sstevel@tonic-gate * We're finished using the boot loader so free its pages. 19617c478bd9Sstevel@tonic-gate */ 19627c478bd9Sstevel@tonic-gate PRM_POINT("Unmapping lower boot pages"); 1963ae115bc7Smrj clear_boot_mappings(0, _userlimit); 1964ae115bc7Smrj postbootkernelbase = kernelbase; 19657c478bd9Sstevel@tonic-gate 19667c478bd9Sstevel@tonic-gate /* 19677c478bd9Sstevel@tonic-gate * If root isn't on ramdisk, destroy the hardcoded 19687c478bd9Sstevel@tonic-gate * ramdisk node now and release the memory. Else, 19697c478bd9Sstevel@tonic-gate * ramdisk memory is kept in rd_pages. 19707c478bd9Sstevel@tonic-gate */ 19717c478bd9Sstevel@tonic-gate root_is_ramdisk = (getmajor(rootdev) == ddi_name_to_major("ramdisk")); 19727c478bd9Sstevel@tonic-gate if (!root_is_ramdisk) { 19737c478bd9Sstevel@tonic-gate dev_info_t *dip = ddi_find_devinfo("ramdisk", -1, 0); 19747c478bd9Sstevel@tonic-gate ASSERT(dip && ddi_get_parent(dip) == ddi_root_node()); 19757c478bd9Sstevel@tonic-gate ndi_rele_devi(dip); /* held from ddi_find_devinfo */ 19767c478bd9Sstevel@tonic-gate (void) ddi_remove_child(dip, 0); 19777c478bd9Sstevel@tonic-gate } 19787c478bd9Sstevel@tonic-gate 19797c478bd9Sstevel@tonic-gate PRM_POINT("Releasing boot pages"); 19807c478bd9Sstevel@tonic-gate while (bootpages) { 19817c478bd9Sstevel@tonic-gate pp = bootpages; 19827c478bd9Sstevel@tonic-gate bootpages = pp->p_next; 19837c478bd9Sstevel@tonic-gate if (root_is_ramdisk && pp_in_ramdisk(pp)) { 19847c478bd9Sstevel@tonic-gate pp->p_next = rd_pages; 19857c478bd9Sstevel@tonic-gate rd_pages = pp; 19867c478bd9Sstevel@tonic-gate continue; 19877c478bd9Sstevel@tonic-gate } 19887c478bd9Sstevel@tonic-gate pp->p_next = (struct page *)0; 1989ae115bc7Smrj pp->p_prev = (struct page *)0; 1990ae115bc7Smrj PP_CLRBOOTPAGES(pp); 19917c478bd9Sstevel@tonic-gate page_free(pp, 1); 19927c478bd9Sstevel@tonic-gate } 1993ae115bc7Smrj PRM_POINT("Boot pages released"); 19947c478bd9Sstevel@tonic-gate 19957c478bd9Sstevel@tonic-gate /* 19967c478bd9Sstevel@tonic-gate * Find 1 page below 1 MB so that other processors can boot up. 19977c478bd9Sstevel@tonic-gate * Make sure it has a kernel VA as well as a 1:1 mapping. 19987c478bd9Sstevel@tonic-gate * We should have just free'd one up. 19997c478bd9Sstevel@tonic-gate */ 20007c478bd9Sstevel@tonic-gate if (use_mp) { 2001ae115bc7Smrj pfn_t pfn; 2002ae115bc7Smrj 20037c478bd9Sstevel@tonic-gate for (pfn = 1; pfn < btop(1*1024*1024); pfn++) { 20047c478bd9Sstevel@tonic-gate if (page_numtopp_alloc(pfn) == NULL) 20057c478bd9Sstevel@tonic-gate continue; 20067c478bd9Sstevel@tonic-gate rm_platter_va = i86devmap(pfn, 1, 20077c478bd9Sstevel@tonic-gate PROT_READ | PROT_WRITE | PROT_EXEC); 20087c478bd9Sstevel@tonic-gate rm_platter_pa = ptob(pfn); 20097c478bd9Sstevel@tonic-gate hat_devload(kas.a_hat, 20107c478bd9Sstevel@tonic-gate (caddr_t)(uintptr_t)rm_platter_pa, MMU_PAGESIZE, 20117c478bd9Sstevel@tonic-gate pfn, PROT_READ | PROT_WRITE | PROT_EXEC, 20127c478bd9Sstevel@tonic-gate HAT_LOAD_NOCONSIST); 20137c478bd9Sstevel@tonic-gate break; 20147c478bd9Sstevel@tonic-gate } 20157c478bd9Sstevel@tonic-gate if (pfn == btop(1*1024*1024)) 20167c478bd9Sstevel@tonic-gate panic("No page available for starting " 20177c478bd9Sstevel@tonic-gate "other processors"); 20187c478bd9Sstevel@tonic-gate } 20197c478bd9Sstevel@tonic-gate 20207c478bd9Sstevel@tonic-gate } 20217c478bd9Sstevel@tonic-gate 20227c478bd9Sstevel@tonic-gate /* 20237c478bd9Sstevel@tonic-gate * Initialize the platform-specific parts of a page_t. 20247c478bd9Sstevel@tonic-gate */ 20257c478bd9Sstevel@tonic-gate void 20267c478bd9Sstevel@tonic-gate add_physmem_cb(page_t *pp, pfn_t pnum) 20277c478bd9Sstevel@tonic-gate { 20287c478bd9Sstevel@tonic-gate pp->p_pagenum = pnum; 20297c478bd9Sstevel@tonic-gate pp->p_mapping = NULL; 20307c478bd9Sstevel@tonic-gate pp->p_embed = 0; 20317c478bd9Sstevel@tonic-gate pp->p_share = 0; 20327c478bd9Sstevel@tonic-gate pp->p_mlentry = 0; 20337c478bd9Sstevel@tonic-gate } 20347c478bd9Sstevel@tonic-gate 20357c478bd9Sstevel@tonic-gate /* 20367c478bd9Sstevel@tonic-gate * kphysm_init() initializes physical memory. 20377c478bd9Sstevel@tonic-gate */ 20387c478bd9Sstevel@tonic-gate static pgcnt_t 20397c478bd9Sstevel@tonic-gate kphysm_init( 2040ae115bc7Smrj page_t *pp, 20417c478bd9Sstevel@tonic-gate pgcnt_t npages) 20427c478bd9Sstevel@tonic-gate { 20437c478bd9Sstevel@tonic-gate struct memlist *pmem; 20447c478bd9Sstevel@tonic-gate struct memseg *cur_memseg; 20457c478bd9Sstevel@tonic-gate pfn_t base_pfn; 20467c478bd9Sstevel@tonic-gate pgcnt_t num; 20477c478bd9Sstevel@tonic-gate pgcnt_t pages_done = 0; 20487c478bd9Sstevel@tonic-gate uint64_t addr; 20497c478bd9Sstevel@tonic-gate uint64_t size; 20507c478bd9Sstevel@tonic-gate extern pfn_t ddiphysmin; 20517c478bd9Sstevel@tonic-gate 20527c478bd9Sstevel@tonic-gate ASSERT(page_hash != NULL && page_hashsz != 0); 20537c478bd9Sstevel@tonic-gate 2054ae115bc7Smrj cur_memseg = memseg_base; 20557c478bd9Sstevel@tonic-gate for (pmem = phys_avail; pmem && npages; pmem = pmem->next) { 20567c478bd9Sstevel@tonic-gate /* 20577c478bd9Sstevel@tonic-gate * In a 32 bit kernel can't use higher memory if we're 20587c478bd9Sstevel@tonic-gate * not booting in PAE mode. This check takes care of that. 20597c478bd9Sstevel@tonic-gate */ 20607c478bd9Sstevel@tonic-gate addr = pmem->address; 20617c478bd9Sstevel@tonic-gate size = pmem->size; 20627c478bd9Sstevel@tonic-gate if (btop(addr) > physmax) 20637c478bd9Sstevel@tonic-gate continue; 20647c478bd9Sstevel@tonic-gate 20657c478bd9Sstevel@tonic-gate /* 20667c478bd9Sstevel@tonic-gate * align addr and size - they may not be at page boundaries 20677c478bd9Sstevel@tonic-gate */ 20687c478bd9Sstevel@tonic-gate if ((addr & MMU_PAGEOFFSET) != 0) { 20697c478bd9Sstevel@tonic-gate addr += MMU_PAGEOFFSET; 20707c478bd9Sstevel@tonic-gate addr &= ~(uint64_t)MMU_PAGEOFFSET; 20717c478bd9Sstevel@tonic-gate size -= addr - pmem->address; 20727c478bd9Sstevel@tonic-gate } 20737c478bd9Sstevel@tonic-gate 20746bb54764Skchow /* only process pages below or equal to physmax */ 20756bb54764Skchow if ((btop(addr + size) - 1) > physmax) 20766bb54764Skchow size = ptob(physmax - btop(addr) + 1); 20777c478bd9Sstevel@tonic-gate 20787c478bd9Sstevel@tonic-gate num = btop(size); 20797c478bd9Sstevel@tonic-gate if (num == 0) 20807c478bd9Sstevel@tonic-gate continue; 20817c478bd9Sstevel@tonic-gate 20827c478bd9Sstevel@tonic-gate if (num > npages) 20837c478bd9Sstevel@tonic-gate num = npages; 20847c478bd9Sstevel@tonic-gate 20857c478bd9Sstevel@tonic-gate npages -= num; 20867c478bd9Sstevel@tonic-gate pages_done += num; 20877c478bd9Sstevel@tonic-gate base_pfn = btop(addr); 20887c478bd9Sstevel@tonic-gate 20897c478bd9Sstevel@tonic-gate if (prom_debug) 20907c478bd9Sstevel@tonic-gate prom_printf("MEMSEG addr=0x%" PRIx64 20917c478bd9Sstevel@tonic-gate " pgs=0x%lx pfn 0x%lx-0x%lx\n", 20927c478bd9Sstevel@tonic-gate addr, num, base_pfn, base_pfn + num); 20937c478bd9Sstevel@tonic-gate 20947c478bd9Sstevel@tonic-gate /* 2095ae115bc7Smrj * Ignore pages below ddiphysmin to simplify ddi memory 20967c478bd9Sstevel@tonic-gate * allocation with non-zero addr_lo requests. 20977c478bd9Sstevel@tonic-gate */ 20987c478bd9Sstevel@tonic-gate if (base_pfn < ddiphysmin) { 2099ae115bc7Smrj if (base_pfn + num <= ddiphysmin) 21007c478bd9Sstevel@tonic-gate continue; 21017c478bd9Sstevel@tonic-gate pp += (ddiphysmin - base_pfn); 21027c478bd9Sstevel@tonic-gate num -= (ddiphysmin - base_pfn); 21037c478bd9Sstevel@tonic-gate base_pfn = ddiphysmin; 21047c478bd9Sstevel@tonic-gate } 2105ae115bc7Smrj 21067c478bd9Sstevel@tonic-gate /* 21077c478bd9Sstevel@tonic-gate * Build the memsegs entry 21087c478bd9Sstevel@tonic-gate */ 21097c478bd9Sstevel@tonic-gate cur_memseg->pages = pp; 21107c478bd9Sstevel@tonic-gate cur_memseg->epages = pp + num; 21117c478bd9Sstevel@tonic-gate cur_memseg->pages_base = base_pfn; 21127c478bd9Sstevel@tonic-gate cur_memseg->pages_end = base_pfn + num; 21137c478bd9Sstevel@tonic-gate 21147c478bd9Sstevel@tonic-gate /* 2115ae115bc7Smrj * Insert into memseg list in decreasing pfn range order. 21167c478bd9Sstevel@tonic-gate * Low memory is typically more fragmented such that this 21177c478bd9Sstevel@tonic-gate * ordering keeps the larger ranges at the front of the list 21187c478bd9Sstevel@tonic-gate * for code that searches memseg. 2119ae115bc7Smrj * This ASSERTS that the memsegs coming in from boot are in 2120ae115bc7Smrj * increasing physical address order and not contiguous. 21217c478bd9Sstevel@tonic-gate */ 2122ae115bc7Smrj if (memsegs != NULL) { 2123ae115bc7Smrj ASSERT(cur_memseg->pages_base >= memsegs->pages_end); 2124ae115bc7Smrj cur_memseg->next = memsegs; 21257c478bd9Sstevel@tonic-gate } 2126ae115bc7Smrj memsegs = cur_memseg; 21277c478bd9Sstevel@tonic-gate 21287c478bd9Sstevel@tonic-gate /* 21297c478bd9Sstevel@tonic-gate * add_physmem() initializes the PSM part of the page 21307c478bd9Sstevel@tonic-gate * struct by calling the PSM back with add_physmem_cb(). 21317c478bd9Sstevel@tonic-gate * In addition it coalesces pages into larger pages as 21327c478bd9Sstevel@tonic-gate * it initializes them. 21337c478bd9Sstevel@tonic-gate */ 21347c478bd9Sstevel@tonic-gate add_physmem(pp, num, base_pfn); 21357c478bd9Sstevel@tonic-gate cur_memseg++; 21367c478bd9Sstevel@tonic-gate availrmem_initial += num; 21377c478bd9Sstevel@tonic-gate availrmem += num; 21387c478bd9Sstevel@tonic-gate 2139ae115bc7Smrj pp += num; 21407c478bd9Sstevel@tonic-gate } 21417c478bd9Sstevel@tonic-gate 21427c478bd9Sstevel@tonic-gate PRM_DEBUG(availrmem_initial); 21437c478bd9Sstevel@tonic-gate PRM_DEBUG(availrmem); 21447c478bd9Sstevel@tonic-gate PRM_DEBUG(freemem); 21457c478bd9Sstevel@tonic-gate build_pfn_hash(); 21467c478bd9Sstevel@tonic-gate return (pages_done); 21477c478bd9Sstevel@tonic-gate } 21487c478bd9Sstevel@tonic-gate 21497c478bd9Sstevel@tonic-gate /* 21507c478bd9Sstevel@tonic-gate * Kernel VM initialization. 21517c478bd9Sstevel@tonic-gate */ 21527c478bd9Sstevel@tonic-gate static void 21537c478bd9Sstevel@tonic-gate kvm_init(void) 21547c478bd9Sstevel@tonic-gate { 21557c478bd9Sstevel@tonic-gate ASSERT((((uintptr_t)s_text) & MMU_PAGEOFFSET) == 0); 21567c478bd9Sstevel@tonic-gate 21577c478bd9Sstevel@tonic-gate /* 21587c478bd9Sstevel@tonic-gate * Put the kernel segments in kernel address space. 21597c478bd9Sstevel@tonic-gate */ 21607c478bd9Sstevel@tonic-gate rw_enter(&kas.a_lock, RW_WRITER); 21617c478bd9Sstevel@tonic-gate as_avlinit(&kas); 21627c478bd9Sstevel@tonic-gate 21637c478bd9Sstevel@tonic-gate (void) seg_attach(&kas, s_text, e_moddata - s_text, &ktextseg); 21647c478bd9Sstevel@tonic-gate (void) segkmem_create(&ktextseg); 21657c478bd9Sstevel@tonic-gate 21667c478bd9Sstevel@tonic-gate (void) seg_attach(&kas, (caddr_t)valloc_base, valloc_sz, &kvalloc); 21677c478bd9Sstevel@tonic-gate (void) segkmem_create(&kvalloc); 21687c478bd9Sstevel@tonic-gate 216935b1ab99Sjosephb (void) seg_attach(&kas, kernelheap, 217035b1ab99Sjosephb ekernelheap - kernelheap, &kvseg); 21717c478bd9Sstevel@tonic-gate (void) segkmem_create(&kvseg); 21727c478bd9Sstevel@tonic-gate 2173ae115bc7Smrj if (core_size > 0) { 2174ae115bc7Smrj PRM_POINT("attaching kvseg_core"); 2175ae115bc7Smrj (void) seg_attach(&kas, (caddr_t)core_base, core_size, 2176ae115bc7Smrj &kvseg_core); 2177ae115bc7Smrj (void) segkmem_create(&kvseg_core); 2178ae115bc7Smrj } 2179ad23a2dbSjohansen 2180ae115bc7Smrj if (segziosize > 0) { 2181ae115bc7Smrj PRM_POINT("attaching segzio"); 2182ad23a2dbSjohansen (void) seg_attach(&kas, segzio_base, mmu_ptob(segziosize), 2183ad23a2dbSjohansen &kzioseg); 2184ad23a2dbSjohansen (void) segkmem_zio_create(&kzioseg); 2185ad23a2dbSjohansen 2186ad23a2dbSjohansen /* create zio area covering new segment */ 2187ad23a2dbSjohansen segkmem_zio_init(segzio_base, mmu_ptob(segziosize)); 2188ad23a2dbSjohansen } 21897c478bd9Sstevel@tonic-gate 2190ae115bc7Smrj (void) seg_attach(&kas, kdi_segdebugbase, kdi_segdebugsize, &kdebugseg); 21917c478bd9Sstevel@tonic-gate (void) segkmem_create(&kdebugseg); 21927c478bd9Sstevel@tonic-gate 21937c478bd9Sstevel@tonic-gate rw_exit(&kas.a_lock); 21947c478bd9Sstevel@tonic-gate 21957c478bd9Sstevel@tonic-gate /* 21967c478bd9Sstevel@tonic-gate * Ensure that the red zone at kernelbase is never accessible. 21977c478bd9Sstevel@tonic-gate */ 2198ae115bc7Smrj PRM_POINT("protecting redzone"); 21997c478bd9Sstevel@tonic-gate (void) as_setprot(&kas, (caddr_t)kernelbase, KERNEL_REDZONE_SIZE, 0); 22007c478bd9Sstevel@tonic-gate 22017c478bd9Sstevel@tonic-gate /* 22027c478bd9Sstevel@tonic-gate * Make the text writable so that it can be hot patched by DTrace. 22037c478bd9Sstevel@tonic-gate */ 22047c478bd9Sstevel@tonic-gate (void) as_setprot(&kas, s_text, e_modtext - s_text, 22057c478bd9Sstevel@tonic-gate PROT_READ | PROT_WRITE | PROT_EXEC); 22067c478bd9Sstevel@tonic-gate 22077c478bd9Sstevel@tonic-gate /* 22087c478bd9Sstevel@tonic-gate * Make data writable until end. 22097c478bd9Sstevel@tonic-gate */ 22107c478bd9Sstevel@tonic-gate (void) as_setprot(&kas, s_data, e_moddata - s_data, 22117c478bd9Sstevel@tonic-gate PROT_READ | PROT_WRITE | PROT_EXEC); 22127c478bd9Sstevel@tonic-gate } 22137c478bd9Sstevel@tonic-gate 22147c478bd9Sstevel@tonic-gate /* 22157c478bd9Sstevel@tonic-gate * These are MTTR registers supported by P6 22167c478bd9Sstevel@tonic-gate */ 22177c478bd9Sstevel@tonic-gate static struct mtrrvar mtrrphys_arr[MAX_MTRRVAR]; 22187c478bd9Sstevel@tonic-gate static uint64_t mtrr64k, mtrr16k1, mtrr16k2; 22197c478bd9Sstevel@tonic-gate static uint64_t mtrr4k1, mtrr4k2, mtrr4k3; 22207c478bd9Sstevel@tonic-gate static uint64_t mtrr4k4, mtrr4k5, mtrr4k6; 22217c478bd9Sstevel@tonic-gate static uint64_t mtrr4k7, mtrr4k8, mtrrcap; 22227c478bd9Sstevel@tonic-gate uint64_t mtrrdef, pat_attr_reg; 22237c478bd9Sstevel@tonic-gate 22247c478bd9Sstevel@tonic-gate /* 22257c478bd9Sstevel@tonic-gate * Disable reprogramming of MTRRs by default. 22267c478bd9Sstevel@tonic-gate */ 22277c478bd9Sstevel@tonic-gate int enable_relaxed_mtrr = 0; 22287c478bd9Sstevel@tonic-gate 22297c478bd9Sstevel@tonic-gate void 223045916cd2Sjpk setup_mtrr(void) 22317c478bd9Sstevel@tonic-gate { 22327c478bd9Sstevel@tonic-gate int i, ecx; 22337c478bd9Sstevel@tonic-gate int vcnt; 22347c478bd9Sstevel@tonic-gate struct mtrrvar *mtrrphys; 22357c478bd9Sstevel@tonic-gate 22367c478bd9Sstevel@tonic-gate if (!(x86_feature & X86_MTRR)) 22377c478bd9Sstevel@tonic-gate return; 22387c478bd9Sstevel@tonic-gate 22390ac7d7d8Skucharsk mtrrcap = rdmsr(REG_MTRRCAP); 22400ac7d7d8Skucharsk mtrrdef = rdmsr(REG_MTRRDEF); 22417c478bd9Sstevel@tonic-gate if (mtrrcap & MTRRCAP_FIX) { 22420ac7d7d8Skucharsk mtrr64k = rdmsr(REG_MTRR64K); 22430ac7d7d8Skucharsk mtrr16k1 = rdmsr(REG_MTRR16K1); 22440ac7d7d8Skucharsk mtrr16k2 = rdmsr(REG_MTRR16K2); 22450ac7d7d8Skucharsk mtrr4k1 = rdmsr(REG_MTRR4K1); 22460ac7d7d8Skucharsk mtrr4k2 = rdmsr(REG_MTRR4K2); 22470ac7d7d8Skucharsk mtrr4k3 = rdmsr(REG_MTRR4K3); 22480ac7d7d8Skucharsk mtrr4k4 = rdmsr(REG_MTRR4K4); 22490ac7d7d8Skucharsk mtrr4k5 = rdmsr(REG_MTRR4K5); 22500ac7d7d8Skucharsk mtrr4k6 = rdmsr(REG_MTRR4K6); 22510ac7d7d8Skucharsk mtrr4k7 = rdmsr(REG_MTRR4K7); 22520ac7d7d8Skucharsk mtrr4k8 = rdmsr(REG_MTRR4K8); 22537c478bd9Sstevel@tonic-gate } 22547c478bd9Sstevel@tonic-gate if ((vcnt = (mtrrcap & MTRRCAP_VCNTMASK)) > MAX_MTRRVAR) 22557c478bd9Sstevel@tonic-gate vcnt = MAX_MTRRVAR; 22567c478bd9Sstevel@tonic-gate 22577c478bd9Sstevel@tonic-gate for (i = 0, ecx = REG_MTRRPHYSBASE0, mtrrphys = mtrrphys_arr; 225835b1ab99Sjosephb i < vcnt - 1; i++, ecx += 2, mtrrphys++) { 22590ac7d7d8Skucharsk mtrrphys->mtrrphys_base = rdmsr(ecx); 22600ac7d7d8Skucharsk mtrrphys->mtrrphys_mask = rdmsr(ecx + 1); 226135b1ab99Sjosephb if ((x86_feature & X86_PAT) && enable_relaxed_mtrr) 22627c478bd9Sstevel@tonic-gate mtrrphys->mtrrphys_mask &= ~MTRRPHYSMASK_V; 22637c478bd9Sstevel@tonic-gate } 22647c478bd9Sstevel@tonic-gate if (x86_feature & X86_PAT) { 22657c478bd9Sstevel@tonic-gate if (enable_relaxed_mtrr) 22667c478bd9Sstevel@tonic-gate mtrrdef = MTRR_TYPE_WB|MTRRDEF_FE|MTRRDEF_E; 22677c478bd9Sstevel@tonic-gate pat_attr_reg = PAT_DEFAULT_ATTRIBUTE; 22687c478bd9Sstevel@tonic-gate } 22697c478bd9Sstevel@tonic-gate 22707c478bd9Sstevel@tonic-gate mtrr_sync(); 22717c478bd9Sstevel@tonic-gate } 22727c478bd9Sstevel@tonic-gate 22737c478bd9Sstevel@tonic-gate /* 22747c478bd9Sstevel@tonic-gate * Sync current cpu mtrr with the incore copy of mtrr. 22757c478bd9Sstevel@tonic-gate * This function has to be invoked with interrupts disabled 22767c478bd9Sstevel@tonic-gate * Currently we do not capture other cpu's. This is invoked on cpu0 22777c478bd9Sstevel@tonic-gate * just after reading /etc/system. 22787c478bd9Sstevel@tonic-gate * On other cpu's its invoked from mp_startup(). 22797c478bd9Sstevel@tonic-gate */ 22807c478bd9Sstevel@tonic-gate void 228145916cd2Sjpk mtrr_sync(void) 22827c478bd9Sstevel@tonic-gate { 22837c478bd9Sstevel@tonic-gate uint_t crvalue, cr0_orig; 22847c478bd9Sstevel@tonic-gate int vcnt, i, ecx; 22857c478bd9Sstevel@tonic-gate struct mtrrvar *mtrrphys; 22867c478bd9Sstevel@tonic-gate 22877c478bd9Sstevel@tonic-gate cr0_orig = crvalue = getcr0(); 22887c478bd9Sstevel@tonic-gate crvalue |= CR0_CD; 22897c478bd9Sstevel@tonic-gate crvalue &= ~CR0_NW; 22907c478bd9Sstevel@tonic-gate setcr0(crvalue); 22917c478bd9Sstevel@tonic-gate invalidate_cache(); 22927c478bd9Sstevel@tonic-gate 2293ae115bc7Smrj reload_cr3(); 22940ac7d7d8Skucharsk if (x86_feature & X86_PAT) 22950ac7d7d8Skucharsk wrmsr(REG_MTRRPAT, pat_attr_reg); 22960ac7d7d8Skucharsk 22970ac7d7d8Skucharsk wrmsr(REG_MTRRDEF, rdmsr(REG_MTRRDEF) & 22980ac7d7d8Skucharsk ~((uint64_t)(uintptr_t)MTRRDEF_E)); 22990ac7d7d8Skucharsk 23007c478bd9Sstevel@tonic-gate if (mtrrcap & MTRRCAP_FIX) { 23010ac7d7d8Skucharsk wrmsr(REG_MTRR64K, mtrr64k); 23020ac7d7d8Skucharsk wrmsr(REG_MTRR16K1, mtrr16k1); 23030ac7d7d8Skucharsk wrmsr(REG_MTRR16K2, mtrr16k2); 23040ac7d7d8Skucharsk wrmsr(REG_MTRR4K1, mtrr4k1); 23050ac7d7d8Skucharsk wrmsr(REG_MTRR4K2, mtrr4k2); 23060ac7d7d8Skucharsk wrmsr(REG_MTRR4K3, mtrr4k3); 23070ac7d7d8Skucharsk wrmsr(REG_MTRR4K4, mtrr4k4); 23080ac7d7d8Skucharsk wrmsr(REG_MTRR4K5, mtrr4k5); 23090ac7d7d8Skucharsk wrmsr(REG_MTRR4K6, mtrr4k6); 23100ac7d7d8Skucharsk wrmsr(REG_MTRR4K7, mtrr4k7); 23110ac7d7d8Skucharsk wrmsr(REG_MTRR4K8, mtrr4k8); 23127c478bd9Sstevel@tonic-gate } 23137c478bd9Sstevel@tonic-gate if ((vcnt = (mtrrcap & MTRRCAP_VCNTMASK)) > MAX_MTRRVAR) 23147c478bd9Sstevel@tonic-gate vcnt = MAX_MTRRVAR; 23157c478bd9Sstevel@tonic-gate for (i = 0, ecx = REG_MTRRPHYSBASE0, mtrrphys = mtrrphys_arr; 23160ac7d7d8Skucharsk i < vcnt - 1; i++, ecx += 2, mtrrphys++) { 23170ac7d7d8Skucharsk wrmsr(ecx, mtrrphys->mtrrphys_base); 23180ac7d7d8Skucharsk wrmsr(ecx + 1, mtrrphys->mtrrphys_mask); 23197c478bd9Sstevel@tonic-gate } 23200ac7d7d8Skucharsk wrmsr(REG_MTRRDEF, mtrrdef); 2321ae115bc7Smrj 2322ae115bc7Smrj reload_cr3(); 23237c478bd9Sstevel@tonic-gate invalidate_cache(); 23247c478bd9Sstevel@tonic-gate setcr0(cr0_orig); 23257c478bd9Sstevel@tonic-gate } 23267c478bd9Sstevel@tonic-gate 23277c478bd9Sstevel@tonic-gate /* 23287c478bd9Sstevel@tonic-gate * resync mtrr so that BIOS is happy. Called from mdboot 23297c478bd9Sstevel@tonic-gate */ 23307c478bd9Sstevel@tonic-gate void 233145916cd2Sjpk mtrr_resync(void) 23327c478bd9Sstevel@tonic-gate { 23337c478bd9Sstevel@tonic-gate if ((x86_feature & X86_PAT) && enable_relaxed_mtrr) { 23347c478bd9Sstevel@tonic-gate /* 23357c478bd9Sstevel@tonic-gate * We could have changed the default mtrr definition. 23367c478bd9Sstevel@tonic-gate * Put it back to uncached which is what it is at power on 23377c478bd9Sstevel@tonic-gate */ 23387c478bd9Sstevel@tonic-gate mtrrdef = MTRR_TYPE_UC|MTRRDEF_FE|MTRRDEF_E; 23397c478bd9Sstevel@tonic-gate mtrr_sync(); 23407c478bd9Sstevel@tonic-gate } 23417c478bd9Sstevel@tonic-gate } 23427c478bd9Sstevel@tonic-gate 23437c478bd9Sstevel@tonic-gate void 234445916cd2Sjpk get_system_configuration(void) 23457c478bd9Sstevel@tonic-gate { 23467c478bd9Sstevel@tonic-gate char prop[32]; 23477c478bd9Sstevel@tonic-gate u_longlong_t nodes_ll, cpus_pernode_ll, lvalue; 23487c478bd9Sstevel@tonic-gate 234935b1ab99Sjosephb if (BOP_GETPROPLEN(bootops, "nodes") > sizeof (prop) || 235035b1ab99Sjosephb BOP_GETPROP(bootops, "nodes", prop) < 0 || 235135b1ab99Sjosephb kobj_getvalue(prop, &nodes_ll) == -1 || 235235b1ab99Sjosephb nodes_ll > MAXNODES || 235335b1ab99Sjosephb BOP_GETPROPLEN(bootops, "cpus_pernode") > sizeof (prop) || 235435b1ab99Sjosephb BOP_GETPROP(bootops, "cpus_pernode", prop) < 0 || 235535b1ab99Sjosephb kobj_getvalue(prop, &cpus_pernode_ll) == -1) { 23567c478bd9Sstevel@tonic-gate system_hardware.hd_nodes = 1; 23577c478bd9Sstevel@tonic-gate system_hardware.hd_cpus_per_node = 0; 23587c478bd9Sstevel@tonic-gate } else { 23597c478bd9Sstevel@tonic-gate system_hardware.hd_nodes = (int)nodes_ll; 23607c478bd9Sstevel@tonic-gate system_hardware.hd_cpus_per_node = (int)cpus_pernode_ll; 23617c478bd9Sstevel@tonic-gate } 236235b1ab99Sjosephb 236335b1ab99Sjosephb if (BOP_GETPROPLEN(bootops, "kernelbase") > sizeof (prop) || 236435b1ab99Sjosephb BOP_GETPROP(bootops, "kernelbase", prop) < 0 || 236535b1ab99Sjosephb kobj_getvalue(prop, &lvalue) == -1) 236635b1ab99Sjosephb eprom_kernelbase = NULL; 23677c478bd9Sstevel@tonic-gate else 236835b1ab99Sjosephb eprom_kernelbase = (uintptr_t)lvalue; 23697c478bd9Sstevel@tonic-gate 237035b1ab99Sjosephb if (BOP_GETPROPLEN(bootops, "segmapsize") > sizeof (prop) || 237135b1ab99Sjosephb BOP_GETPROP(bootops, "segmapsize", prop) < 0 || 237235b1ab99Sjosephb kobj_getvalue(prop, &lvalue) == -1) 23737c478bd9Sstevel@tonic-gate segmapsize = SEGMAPDEFAULT; 237435b1ab99Sjosephb else 23757c478bd9Sstevel@tonic-gate segmapsize = (uintptr_t)lvalue; 23767c478bd9Sstevel@tonic-gate 237735b1ab99Sjosephb if (BOP_GETPROPLEN(bootops, "segmapfreelists") > sizeof (prop) || 237835b1ab99Sjosephb BOP_GETPROP(bootops, "segmapfreelists", prop) < 0 || 237935b1ab99Sjosephb kobj_getvalue(prop, &lvalue) == -1) 23807c478bd9Sstevel@tonic-gate segmapfreelists = 0; /* use segmap driver default */ 238135b1ab99Sjosephb else 23827c478bd9Sstevel@tonic-gate segmapfreelists = (int)lvalue; 23834944b02eSkchow 2384aac11643Sjosephb /* physmem used to be here, but moved much earlier to fakebop.c */ 23857c478bd9Sstevel@tonic-gate } 23867c478bd9Sstevel@tonic-gate 23877c478bd9Sstevel@tonic-gate /* 23887c478bd9Sstevel@tonic-gate * Add to a memory list. 23897c478bd9Sstevel@tonic-gate * start = start of new memory segment 23907c478bd9Sstevel@tonic-gate * len = length of new memory segment in bytes 23917c478bd9Sstevel@tonic-gate * new = pointer to a new struct memlist 23927c478bd9Sstevel@tonic-gate * memlistp = memory list to which to add segment. 23937c478bd9Sstevel@tonic-gate */ 2394ae115bc7Smrj void 23957c478bd9Sstevel@tonic-gate memlist_add( 23967c478bd9Sstevel@tonic-gate uint64_t start, 23977c478bd9Sstevel@tonic-gate uint64_t len, 23987c478bd9Sstevel@tonic-gate struct memlist *new, 23997c478bd9Sstevel@tonic-gate struct memlist **memlistp) 24007c478bd9Sstevel@tonic-gate { 24017c478bd9Sstevel@tonic-gate struct memlist *cur; 24027c478bd9Sstevel@tonic-gate uint64_t end = start + len; 24037c478bd9Sstevel@tonic-gate 24047c478bd9Sstevel@tonic-gate new->address = start; 24057c478bd9Sstevel@tonic-gate new->size = len; 24067c478bd9Sstevel@tonic-gate 24077c478bd9Sstevel@tonic-gate cur = *memlistp; 24087c478bd9Sstevel@tonic-gate 24097c478bd9Sstevel@tonic-gate while (cur) { 24107c478bd9Sstevel@tonic-gate if (cur->address >= end) { 24117c478bd9Sstevel@tonic-gate new->next = cur; 24127c478bd9Sstevel@tonic-gate *memlistp = new; 24137c478bd9Sstevel@tonic-gate new->prev = cur->prev; 24147c478bd9Sstevel@tonic-gate cur->prev = new; 24157c478bd9Sstevel@tonic-gate return; 24167c478bd9Sstevel@tonic-gate } 24177c478bd9Sstevel@tonic-gate ASSERT(cur->address + cur->size <= start); 24187c478bd9Sstevel@tonic-gate if (cur->next == NULL) { 24197c478bd9Sstevel@tonic-gate cur->next = new; 24207c478bd9Sstevel@tonic-gate new->prev = cur; 24217c478bd9Sstevel@tonic-gate new->next = NULL; 24227c478bd9Sstevel@tonic-gate return; 24237c478bd9Sstevel@tonic-gate } 24247c478bd9Sstevel@tonic-gate memlistp = &cur->next; 24257c478bd9Sstevel@tonic-gate cur = cur->next; 24267c478bd9Sstevel@tonic-gate } 24277c478bd9Sstevel@tonic-gate } 24287c478bd9Sstevel@tonic-gate 24297c478bd9Sstevel@tonic-gate void 24307c478bd9Sstevel@tonic-gate kobj_vmem_init(vmem_t **text_arena, vmem_t **data_arena) 24317c478bd9Sstevel@tonic-gate { 24327c478bd9Sstevel@tonic-gate size_t tsize = e_modtext - modtext; 24337c478bd9Sstevel@tonic-gate size_t dsize = e_moddata - moddata; 24347c478bd9Sstevel@tonic-gate 24357c478bd9Sstevel@tonic-gate *text_arena = vmem_create("module_text", tsize ? modtext : NULL, tsize, 24367c478bd9Sstevel@tonic-gate 1, segkmem_alloc, segkmem_free, heaptext_arena, 0, VM_SLEEP); 24377c478bd9Sstevel@tonic-gate *data_arena = vmem_create("module_data", dsize ? moddata : NULL, dsize, 24387c478bd9Sstevel@tonic-gate 1, segkmem_alloc, segkmem_free, heap32_arena, 0, VM_SLEEP); 24397c478bd9Sstevel@tonic-gate } 24407c478bd9Sstevel@tonic-gate 24417c478bd9Sstevel@tonic-gate caddr_t 24427c478bd9Sstevel@tonic-gate kobj_text_alloc(vmem_t *arena, size_t size) 24437c478bd9Sstevel@tonic-gate { 24447c478bd9Sstevel@tonic-gate return (vmem_alloc(arena, size, VM_SLEEP | VM_BESTFIT)); 24457c478bd9Sstevel@tonic-gate } 24467c478bd9Sstevel@tonic-gate 24477c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 24487c478bd9Sstevel@tonic-gate caddr_t 24497c478bd9Sstevel@tonic-gate kobj_texthole_alloc(caddr_t addr, size_t size) 24507c478bd9Sstevel@tonic-gate { 24517c478bd9Sstevel@tonic-gate panic("unexpected call to kobj_texthole_alloc()"); 24527c478bd9Sstevel@tonic-gate /*NOTREACHED*/ 24537c478bd9Sstevel@tonic-gate return (0); 24547c478bd9Sstevel@tonic-gate } 24557c478bd9Sstevel@tonic-gate 24567c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 24577c478bd9Sstevel@tonic-gate void 24587c478bd9Sstevel@tonic-gate kobj_texthole_free(caddr_t addr, size_t size) 24597c478bd9Sstevel@tonic-gate { 24607c478bd9Sstevel@tonic-gate panic("unexpected call to kobj_texthole_free()"); 24617c478bd9Sstevel@tonic-gate } 24627c478bd9Sstevel@tonic-gate 24637c478bd9Sstevel@tonic-gate /* 24647c478bd9Sstevel@tonic-gate * This is called just after configure() in startup(). 24657c478bd9Sstevel@tonic-gate * 24667c478bd9Sstevel@tonic-gate * The ISALIST concept is a bit hopeless on Intel, because 24677c478bd9Sstevel@tonic-gate * there's no guarantee of an ever-more-capable processor 24687c478bd9Sstevel@tonic-gate * given that various parts of the instruction set may appear 24697c478bd9Sstevel@tonic-gate * and disappear between different implementations. 24707c478bd9Sstevel@tonic-gate * 24717c478bd9Sstevel@tonic-gate * While it would be possible to correct it and even enhance 24727c478bd9Sstevel@tonic-gate * it somewhat, the explicit hardware capability bitmask allows 24737c478bd9Sstevel@tonic-gate * more flexibility. 24747c478bd9Sstevel@tonic-gate * 24757c478bd9Sstevel@tonic-gate * So, we just leave this alone. 24767c478bd9Sstevel@tonic-gate */ 24777c478bd9Sstevel@tonic-gate void 24787c478bd9Sstevel@tonic-gate setx86isalist(void) 24797c478bd9Sstevel@tonic-gate { 24807c478bd9Sstevel@tonic-gate char *tp; 24817c478bd9Sstevel@tonic-gate size_t len; 24827c478bd9Sstevel@tonic-gate extern char *isa_list; 24837c478bd9Sstevel@tonic-gate 24847c478bd9Sstevel@tonic-gate #define TBUFSIZE 1024 24857c478bd9Sstevel@tonic-gate 24867c478bd9Sstevel@tonic-gate tp = kmem_alloc(TBUFSIZE, KM_SLEEP); 24877c478bd9Sstevel@tonic-gate *tp = '\0'; 24887c478bd9Sstevel@tonic-gate 24897c478bd9Sstevel@tonic-gate #if defined(__amd64) 24907c478bd9Sstevel@tonic-gate (void) strcpy(tp, "amd64 "); 24917c478bd9Sstevel@tonic-gate #endif 24927c478bd9Sstevel@tonic-gate 24937c478bd9Sstevel@tonic-gate switch (x86_vendor) { 24947c478bd9Sstevel@tonic-gate case X86_VENDOR_Intel: 24957c478bd9Sstevel@tonic-gate case X86_VENDOR_AMD: 24967c478bd9Sstevel@tonic-gate case X86_VENDOR_TM: 24977c478bd9Sstevel@tonic-gate if (x86_feature & X86_CMOV) { 24987c478bd9Sstevel@tonic-gate /* 24997c478bd9Sstevel@tonic-gate * Pentium Pro or later 25007c478bd9Sstevel@tonic-gate */ 25017c478bd9Sstevel@tonic-gate (void) strcat(tp, "pentium_pro"); 25027c478bd9Sstevel@tonic-gate (void) strcat(tp, x86_feature & X86_MMX ? 25037c478bd9Sstevel@tonic-gate "+mmx pentium_pro " : " "); 25047c478bd9Sstevel@tonic-gate } 25057c478bd9Sstevel@tonic-gate /*FALLTHROUGH*/ 25067c478bd9Sstevel@tonic-gate case X86_VENDOR_Cyrix: 25077c478bd9Sstevel@tonic-gate /* 25087c478bd9Sstevel@tonic-gate * The Cyrix 6x86 does not have any Pentium features 25097c478bd9Sstevel@tonic-gate * accessible while not at privilege level 0. 25107c478bd9Sstevel@tonic-gate */ 25117c478bd9Sstevel@tonic-gate if (x86_feature & X86_CPUID) { 25127c478bd9Sstevel@tonic-gate (void) strcat(tp, "pentium"); 25137c478bd9Sstevel@tonic-gate (void) strcat(tp, x86_feature & X86_MMX ? 25147c478bd9Sstevel@tonic-gate "+mmx pentium " : " "); 25157c478bd9Sstevel@tonic-gate } 25167c478bd9Sstevel@tonic-gate break; 25177c478bd9Sstevel@tonic-gate default: 25187c478bd9Sstevel@tonic-gate break; 25197c478bd9Sstevel@tonic-gate } 25207c478bd9Sstevel@tonic-gate (void) strcat(tp, "i486 i386 i86"); 25217c478bd9Sstevel@tonic-gate len = strlen(tp) + 1; /* account for NULL at end of string */ 25227c478bd9Sstevel@tonic-gate isa_list = strcpy(kmem_alloc(len, KM_SLEEP), tp); 25237c478bd9Sstevel@tonic-gate kmem_free(tp, TBUFSIZE); 25247c478bd9Sstevel@tonic-gate 25257c478bd9Sstevel@tonic-gate #undef TBUFSIZE 25267c478bd9Sstevel@tonic-gate } 25277c478bd9Sstevel@tonic-gate 25287c478bd9Sstevel@tonic-gate 25297c478bd9Sstevel@tonic-gate #ifdef __amd64 25307c478bd9Sstevel@tonic-gate 25317c478bd9Sstevel@tonic-gate void * 25327c478bd9Sstevel@tonic-gate device_arena_alloc(size_t size, int vm_flag) 25337c478bd9Sstevel@tonic-gate { 25347c478bd9Sstevel@tonic-gate return (vmem_alloc(device_arena, size, vm_flag)); 25357c478bd9Sstevel@tonic-gate } 25367c478bd9Sstevel@tonic-gate 25377c478bd9Sstevel@tonic-gate void 25387c478bd9Sstevel@tonic-gate device_arena_free(void *vaddr, size_t size) 25397c478bd9Sstevel@tonic-gate { 25407c478bd9Sstevel@tonic-gate vmem_free(device_arena, vaddr, size); 25417c478bd9Sstevel@tonic-gate } 25427c478bd9Sstevel@tonic-gate 2543ae115bc7Smrj #else /* __i386 */ 25447c478bd9Sstevel@tonic-gate 25457c478bd9Sstevel@tonic-gate void * 25467c478bd9Sstevel@tonic-gate device_arena_alloc(size_t size, int vm_flag) 25477c478bd9Sstevel@tonic-gate { 25487c478bd9Sstevel@tonic-gate caddr_t vaddr; 25497c478bd9Sstevel@tonic-gate uintptr_t v; 25507c478bd9Sstevel@tonic-gate size_t start; 25517c478bd9Sstevel@tonic-gate size_t end; 25527c478bd9Sstevel@tonic-gate 25537c478bd9Sstevel@tonic-gate vaddr = vmem_alloc(heap_arena, size, vm_flag); 25547c478bd9Sstevel@tonic-gate if (vaddr == NULL) 25557c478bd9Sstevel@tonic-gate return (NULL); 25567c478bd9Sstevel@tonic-gate 25577c478bd9Sstevel@tonic-gate v = (uintptr_t)vaddr; 25587c478bd9Sstevel@tonic-gate ASSERT(v >= kernelbase); 2559ae115bc7Smrj ASSERT(v + size <= valloc_base); 25607c478bd9Sstevel@tonic-gate 25617c478bd9Sstevel@tonic-gate start = btop(v - kernelbase); 25627c478bd9Sstevel@tonic-gate end = btop(v + size - 1 - kernelbase); 25637c478bd9Sstevel@tonic-gate ASSERT(start < toxic_bit_map_len); 25647c478bd9Sstevel@tonic-gate ASSERT(end < toxic_bit_map_len); 25657c478bd9Sstevel@tonic-gate 25667c478bd9Sstevel@tonic-gate while (start <= end) { 25677c478bd9Sstevel@tonic-gate BT_ATOMIC_SET(toxic_bit_map, start); 25687c478bd9Sstevel@tonic-gate ++start; 25697c478bd9Sstevel@tonic-gate } 25707c478bd9Sstevel@tonic-gate return (vaddr); 25717c478bd9Sstevel@tonic-gate } 25727c478bd9Sstevel@tonic-gate 25737c478bd9Sstevel@tonic-gate void 25747c478bd9Sstevel@tonic-gate device_arena_free(void *vaddr, size_t size) 25757c478bd9Sstevel@tonic-gate { 25767c478bd9Sstevel@tonic-gate uintptr_t v = (uintptr_t)vaddr; 25777c478bd9Sstevel@tonic-gate size_t start; 25787c478bd9Sstevel@tonic-gate size_t end; 25797c478bd9Sstevel@tonic-gate 25807c478bd9Sstevel@tonic-gate ASSERT(v >= kernelbase); 2581ae115bc7Smrj ASSERT(v + size <= valloc_base); 25827c478bd9Sstevel@tonic-gate 25837c478bd9Sstevel@tonic-gate start = btop(v - kernelbase); 25847c478bd9Sstevel@tonic-gate end = btop(v + size - 1 - kernelbase); 25857c478bd9Sstevel@tonic-gate ASSERT(start < toxic_bit_map_len); 25867c478bd9Sstevel@tonic-gate ASSERT(end < toxic_bit_map_len); 25877c478bd9Sstevel@tonic-gate 25887c478bd9Sstevel@tonic-gate while (start <= end) { 25897c478bd9Sstevel@tonic-gate ASSERT(BT_TEST(toxic_bit_map, start) != 0); 25907c478bd9Sstevel@tonic-gate BT_ATOMIC_CLEAR(toxic_bit_map, start); 25917c478bd9Sstevel@tonic-gate ++start; 25927c478bd9Sstevel@tonic-gate } 25937c478bd9Sstevel@tonic-gate vmem_free(heap_arena, vaddr, size); 25947c478bd9Sstevel@tonic-gate } 25957c478bd9Sstevel@tonic-gate 25967c478bd9Sstevel@tonic-gate /* 25977c478bd9Sstevel@tonic-gate * returns 1st address in range that is in device arena, or NULL 25987c478bd9Sstevel@tonic-gate * if len is not NULL it returns the length of the toxic range 25997c478bd9Sstevel@tonic-gate */ 26007c478bd9Sstevel@tonic-gate void * 26017c478bd9Sstevel@tonic-gate device_arena_contains(void *vaddr, size_t size, size_t *len) 26027c478bd9Sstevel@tonic-gate { 26037c478bd9Sstevel@tonic-gate uintptr_t v = (uintptr_t)vaddr; 26047c478bd9Sstevel@tonic-gate uintptr_t eaddr = v + size; 26057c478bd9Sstevel@tonic-gate size_t start; 26067c478bd9Sstevel@tonic-gate size_t end; 26077c478bd9Sstevel@tonic-gate 26087c478bd9Sstevel@tonic-gate /* 26097c478bd9Sstevel@tonic-gate * if called very early by kmdb, just return NULL 26107c478bd9Sstevel@tonic-gate */ 26117c478bd9Sstevel@tonic-gate if (toxic_bit_map == NULL) 26127c478bd9Sstevel@tonic-gate return (NULL); 26137c478bd9Sstevel@tonic-gate 26147c478bd9Sstevel@tonic-gate /* 26157c478bd9Sstevel@tonic-gate * First check if we're completely outside the bitmap range. 26167c478bd9Sstevel@tonic-gate */ 2617ae115bc7Smrj if (v >= valloc_base || eaddr < kernelbase) 26187c478bd9Sstevel@tonic-gate return (NULL); 26197c478bd9Sstevel@tonic-gate 26207c478bd9Sstevel@tonic-gate /* 26217c478bd9Sstevel@tonic-gate * Trim ends of search to look at only what the bitmap covers. 26227c478bd9Sstevel@tonic-gate */ 26237c478bd9Sstevel@tonic-gate if (v < kernelbase) 26247c478bd9Sstevel@tonic-gate v = kernelbase; 26257c478bd9Sstevel@tonic-gate start = btop(v - kernelbase); 26267c478bd9Sstevel@tonic-gate end = btop(eaddr - kernelbase); 26277c478bd9Sstevel@tonic-gate if (end >= toxic_bit_map_len) 26287c478bd9Sstevel@tonic-gate end = toxic_bit_map_len; 26297c478bd9Sstevel@tonic-gate 26307c478bd9Sstevel@tonic-gate if (bt_range(toxic_bit_map, &start, &end, end) == 0) 26317c478bd9Sstevel@tonic-gate return (NULL); 26327c478bd9Sstevel@tonic-gate 26337c478bd9Sstevel@tonic-gate v = kernelbase + ptob(start); 26347c478bd9Sstevel@tonic-gate if (len != NULL) 26357c478bd9Sstevel@tonic-gate *len = ptob(end - start); 26367c478bd9Sstevel@tonic-gate return ((void *)v); 26377c478bd9Sstevel@tonic-gate } 26387c478bd9Sstevel@tonic-gate 2639ae115bc7Smrj #endif /* __i386 */ 2640