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 /* 220db3240dSStephen Hanson * Copyright (c) 1993, 2010, Oracle and/or its affiliates. All rights reserved. 23*9e3e4df2SGarrett D'Amore * Copyright 2012 DEY Storage Systems, Inc. All rights reserved. 247c478bd9Sstevel@tonic-gate */ 25a3114836SGerry Liu /* 26a3114836SGerry Liu * Copyright (c) 2010, Intel Corporation. 27a3114836SGerry Liu * All rights reserved. 28a3114836SGerry Liu */ 297c478bd9Sstevel@tonic-gate 307c478bd9Sstevel@tonic-gate #include <sys/types.h> 317c478bd9Sstevel@tonic-gate #include <sys/t_lock.h> 327c478bd9Sstevel@tonic-gate #include <sys/param.h> 337c478bd9Sstevel@tonic-gate #include <sys/sysmacros.h> 347c478bd9Sstevel@tonic-gate #include <sys/signal.h> 357c478bd9Sstevel@tonic-gate #include <sys/systm.h> 367c478bd9Sstevel@tonic-gate #include <sys/user.h> 377c478bd9Sstevel@tonic-gate #include <sys/mman.h> 387c478bd9Sstevel@tonic-gate #include <sys/vm.h> 397c478bd9Sstevel@tonic-gate #include <sys/conf.h> 407c478bd9Sstevel@tonic-gate #include <sys/avintr.h> 417c478bd9Sstevel@tonic-gate #include <sys/autoconf.h> 427c478bd9Sstevel@tonic-gate #include <sys/disp.h> 437c478bd9Sstevel@tonic-gate #include <sys/class.h> 447c478bd9Sstevel@tonic-gate #include <sys/bitmap.h> 457c478bd9Sstevel@tonic-gate 467c478bd9Sstevel@tonic-gate #include <sys/privregs.h> 477c478bd9Sstevel@tonic-gate 487c478bd9Sstevel@tonic-gate #include <sys/proc.h> 497c478bd9Sstevel@tonic-gate #include <sys/buf.h> 507c478bd9Sstevel@tonic-gate #include <sys/kmem.h> 51ae115bc7Smrj #include <sys/mem.h> 527c478bd9Sstevel@tonic-gate #include <sys/kstat.h> 537c478bd9Sstevel@tonic-gate 547c478bd9Sstevel@tonic-gate #include <sys/reboot.h> 557c478bd9Sstevel@tonic-gate 567c478bd9Sstevel@tonic-gate #include <sys/cred.h> 577c478bd9Sstevel@tonic-gate #include <sys/vnode.h> 587c478bd9Sstevel@tonic-gate #include <sys/file.h> 597c478bd9Sstevel@tonic-gate 607c478bd9Sstevel@tonic-gate #include <sys/procfs.h> 617c478bd9Sstevel@tonic-gate 627c478bd9Sstevel@tonic-gate #include <sys/vfs.h> 637c478bd9Sstevel@tonic-gate #include <sys/cmn_err.h> 647c478bd9Sstevel@tonic-gate #include <sys/utsname.h> 657c478bd9Sstevel@tonic-gate #include <sys/debug.h> 667c478bd9Sstevel@tonic-gate #include <sys/kdi.h> 677c478bd9Sstevel@tonic-gate 687c478bd9Sstevel@tonic-gate #include <sys/dumphdr.h> 697c478bd9Sstevel@tonic-gate #include <sys/bootconf.h> 70a3114836SGerry Liu #include <sys/memlist_plat.h> 717c478bd9Sstevel@tonic-gate #include <sys/varargs.h> 727c478bd9Sstevel@tonic-gate #include <sys/promif.h> 73843e1988Sjohnlev #include <sys/modctl.h> 747c478bd9Sstevel@tonic-gate 757c478bd9Sstevel@tonic-gate #include <sys/sunddi.h> 767c478bd9Sstevel@tonic-gate #include <sys/sunndi.h> 777c478bd9Sstevel@tonic-gate #include <sys/ndi_impldefs.h> 787c478bd9Sstevel@tonic-gate #include <sys/ddidmareq.h> 797c478bd9Sstevel@tonic-gate #include <sys/psw.h> 807c478bd9Sstevel@tonic-gate #include <sys/regset.h> 817c478bd9Sstevel@tonic-gate #include <sys/clock.h> 827c478bd9Sstevel@tonic-gate #include <sys/pte.h> 837c478bd9Sstevel@tonic-gate #include <sys/tss.h> 847c478bd9Sstevel@tonic-gate #include <sys/stack.h> 857c478bd9Sstevel@tonic-gate #include <sys/trap.h> 867c478bd9Sstevel@tonic-gate #include <sys/fp.h> 871d03c31eSjohnlev #include <vm/kboot_mmu.h> 887c478bd9Sstevel@tonic-gate #include <vm/anon.h> 897c478bd9Sstevel@tonic-gate #include <vm/as.h> 907c478bd9Sstevel@tonic-gate #include <vm/page.h> 917c478bd9Sstevel@tonic-gate #include <vm/seg.h> 927c478bd9Sstevel@tonic-gate #include <vm/seg_dev.h> 937c478bd9Sstevel@tonic-gate #include <vm/seg_kmem.h> 947c478bd9Sstevel@tonic-gate #include <vm/seg_kpm.h> 957c478bd9Sstevel@tonic-gate #include <vm/seg_map.h> 967c478bd9Sstevel@tonic-gate #include <vm/seg_vn.h> 977c478bd9Sstevel@tonic-gate #include <vm/seg_kp.h> 987c478bd9Sstevel@tonic-gate #include <sys/memnode.h> 997c478bd9Sstevel@tonic-gate #include <vm/vm_dep.h> 1007c478bd9Sstevel@tonic-gate #include <sys/thread.h> 1017c478bd9Sstevel@tonic-gate #include <sys/sysconf.h> 1027c478bd9Sstevel@tonic-gate #include <sys/vm_machparam.h> 1037c478bd9Sstevel@tonic-gate #include <sys/archsystm.h> 1047c478bd9Sstevel@tonic-gate #include <sys/machsystm.h> 1057c478bd9Sstevel@tonic-gate #include <vm/hat.h> 1067c478bd9Sstevel@tonic-gate #include <vm/hat_i86.h> 1077c478bd9Sstevel@tonic-gate #include <sys/pmem.h> 1087c478bd9Sstevel@tonic-gate #include <sys/smp_impldefs.h> 1097c478bd9Sstevel@tonic-gate #include <sys/x86_archext.h> 110a563a037Sbholler #include <sys/cpuvar.h> 1117c478bd9Sstevel@tonic-gate #include <sys/segments.h> 1127c478bd9Sstevel@tonic-gate #include <sys/clconf.h> 1137c478bd9Sstevel@tonic-gate #include <sys/kobj.h> 1147c478bd9Sstevel@tonic-gate #include <sys/kobj_lex.h> 1157c478bd9Sstevel@tonic-gate #include <sys/cpc_impl.h> 1167aec1d6eScindi #include <sys/cpu_module.h> 11784ab085aSmws #include <sys/smbios.h> 118ae115bc7Smrj #include <sys/debug_info.h> 1191d03c31eSjohnlev #include <sys/bootinfo.h> 120dd4eeefdSeota #include <sys/ddi_timer.h> 1215679c89fSjv #include <sys/systeminfo.h> 12219397407SSherry Moore #include <sys/multiboot.h> 123ae115bc7Smrj 124a3114836SGerry Liu #ifdef __xpv 1251d03c31eSjohnlev 126843e1988Sjohnlev #include <sys/hypervisor.h> 127843e1988Sjohnlev #include <sys/xen_mmu.h> 128843e1988Sjohnlev #include <sys/evtchn_impl.h> 129843e1988Sjohnlev #include <sys/gnttab.h> 130843e1988Sjohnlev #include <sys/xpv_panic.h> 131843e1988Sjohnlev #include <xen/sys/xenbus_comms.h> 132843e1988Sjohnlev #include <xen/public/physdev.h> 1331d03c31eSjohnlev 134843e1988Sjohnlev extern void xen_late_startup(void); 135ae115bc7Smrj 1361d03c31eSjohnlev struct xen_evt_data cpu0_evt_data; 1371d03c31eSjohnlev 138a3114836SGerry Liu #else /* __xpv */ 139a3114836SGerry Liu #include <sys/memlist_impl.h> 140a3114836SGerry Liu 141a3114836SGerry Liu extern void mem_config_init(void); 1421d03c31eSjohnlev #endif /* __xpv */ 1437c478bd9Sstevel@tonic-gate 1447c478bd9Sstevel@tonic-gate extern void progressbar_init(void); 1459acbbeafSnn extern void brand_init(void); 14606fb6a36Sdv extern void pcf_init(void); 1470e751525SEric Saxe extern void pg_init(void); 1487c478bd9Sstevel@tonic-gate 149d7d93655Sblakej extern int size_pse_array(pgcnt_t, int); 150d7d93655Sblakej 151ed5289f9SKen Erickson #if defined(_SOFT_HOSTID) 152ed5289f9SKen Erickson 153ed5289f9SKen Erickson #include <sys/rtc.h> 154ed5289f9SKen Erickson 155ed5289f9SKen Erickson static int32_t set_soft_hostid(void); 156ed5289f9SKen Erickson static char hostid_file[] = "/etc/hostid"; 157ed5289f9SKen Erickson 158ed5289f9SKen Erickson #endif 159ed5289f9SKen Erickson 16094f1124eSVikram Hegde void *gfx_devinfo_list; 161ed5289f9SKen Erickson 1623a634bfcSVikram Hegde #if defined(__amd64) && !defined(__xpv) 1633a634bfcSVikram Hegde extern void immu_startup(void); 1643a634bfcSVikram Hegde #endif 1653a634bfcSVikram Hegde 1667c478bd9Sstevel@tonic-gate /* 1677c478bd9Sstevel@tonic-gate * XXX make declaration below "static" when drivers no longer use this 1687c478bd9Sstevel@tonic-gate * interface. 1697c478bd9Sstevel@tonic-gate */ 1707c478bd9Sstevel@tonic-gate extern caddr_t p0_va; /* Virtual address for accessing physical page 0 */ 1717c478bd9Sstevel@tonic-gate 1727c478bd9Sstevel@tonic-gate /* 1737c478bd9Sstevel@tonic-gate * segkp 1747c478bd9Sstevel@tonic-gate */ 1757c478bd9Sstevel@tonic-gate extern int segkp_fromheap; 1767c478bd9Sstevel@tonic-gate 1777c478bd9Sstevel@tonic-gate static void kvm_init(void); 1787c478bd9Sstevel@tonic-gate static void startup_init(void); 1797c478bd9Sstevel@tonic-gate static void startup_memlist(void); 180ae115bc7Smrj static void startup_kmem(void); 1817c478bd9Sstevel@tonic-gate static void startup_modules(void); 1827c478bd9Sstevel@tonic-gate static void startup_vm(void); 1837c478bd9Sstevel@tonic-gate static void startup_end(void); 18435b1ab99Sjosephb static void layout_kernel_va(void); 1857c478bd9Sstevel@tonic-gate 1867c478bd9Sstevel@tonic-gate /* 1877c478bd9Sstevel@tonic-gate * Declare these as initialized data so we can patch them. 1887c478bd9Sstevel@tonic-gate */ 189f53ad214Skchow #ifdef __i386 19035b1ab99Sjosephb 191f53ad214Skchow /* 192f53ad214Skchow * Due to virtual address space limitations running in 32 bit mode, restrict 19335b1ab99Sjosephb * the amount of physical memory configured to a max of PHYSMEM pages (16g). 194f53ad214Skchow * 195f53ad214Skchow * If the physical max memory size of 64g were allowed to be configured, the 196f53ad214Skchow * size of user virtual address space will be less than 1g. A limited user 197f53ad214Skchow * address space greatly reduces the range of applications that can run. 198f53ad214Skchow * 19935b1ab99Sjosephb * If more physical memory than PHYSMEM is required, users should preferably 20035b1ab99Sjosephb * run in 64 bit mode which has far looser virtual address space limitations. 201f53ad214Skchow * 202f53ad214Skchow * If 64 bit mode is not available (as in IA32) and/or more physical memory 20335b1ab99Sjosephb * than PHYSMEM is required in 32 bit mode, physmem can be set to the desired 204f53ad214Skchow * value or to 0 (to configure all available memory) via eeprom(1M). kernelbase 205f53ad214Skchow * should also be carefully tuned to balance out the need of the user 206f53ad214Skchow * application while minimizing the risk of kernel heap exhaustion due to 207f53ad214Skchow * kernelbase being set too high. 208f53ad214Skchow */ 20935b1ab99Sjosephb #define PHYSMEM 0x400000 210f53ad214Skchow 21135b1ab99Sjosephb #else /* __amd64 */ 21235b1ab99Sjosephb 21335b1ab99Sjosephb /* 21435b1ab99Sjosephb * For now we can handle memory with physical addresses up to about 21535b1ab99Sjosephb * 64 Terabytes. This keeps the kernel above the VA hole, leaving roughly 21635b1ab99Sjosephb * half the VA space for seg_kpm. When systems get bigger than 64TB this 21735b1ab99Sjosephb * code will need revisiting. There is an implicit assumption that there 21835b1ab99Sjosephb * are no *huge* holes in the physical address space too. 21935b1ab99Sjosephb */ 22035b1ab99Sjosephb #define TERABYTE (1ul << 40) 22135b1ab99Sjosephb #define PHYSMEM_MAX64 mmu_btop(64 * TERABYTE) 22235b1ab99Sjosephb #define PHYSMEM PHYSMEM_MAX64 22335b1ab99Sjosephb #define AMD64_VA_HOLE_END 0xFFFF800000000000ul 22435b1ab99Sjosephb 22535b1ab99Sjosephb #endif /* __amd64 */ 22635b1ab99Sjosephb 22735b1ab99Sjosephb pgcnt_t physmem = PHYSMEM; 2287c478bd9Sstevel@tonic-gate pgcnt_t obp_pages; /* Memory used by PROM for its text and data */ 2297c478bd9Sstevel@tonic-gate 2307c478bd9Sstevel@tonic-gate char *kobj_file_buf; 2317c478bd9Sstevel@tonic-gate int kobj_file_bufsize; /* set in /etc/system */ 2327c478bd9Sstevel@tonic-gate 2337c478bd9Sstevel@tonic-gate /* Global variables for MP support. Used in mp_startup */ 23419397407SSherry Moore caddr_t rm_platter_va = 0; 2357c478bd9Sstevel@tonic-gate uint32_t rm_platter_pa; 2367c478bd9Sstevel@tonic-gate 23783f9b804Skchow int auto_lpg_disable = 1; 23883f9b804Skchow 2397c478bd9Sstevel@tonic-gate /* 2407c478bd9Sstevel@tonic-gate * Some CPUs have holes in the middle of the 64-bit virtual address range. 2417c478bd9Sstevel@tonic-gate */ 2427c478bd9Sstevel@tonic-gate uintptr_t hole_start, hole_end; 2437c478bd9Sstevel@tonic-gate 2447c478bd9Sstevel@tonic-gate /* 2457c478bd9Sstevel@tonic-gate * kpm mapping window 2467c478bd9Sstevel@tonic-gate */ 2477c478bd9Sstevel@tonic-gate caddr_t kpm_vbase; 2487c478bd9Sstevel@tonic-gate size_t kpm_size; 24935b1ab99Sjosephb static int kpm_desired; 25035b1ab99Sjosephb #ifdef __amd64 25135b1ab99Sjosephb static uintptr_t segkpm_base = (uintptr_t)SEGKPM_BASE; 25235b1ab99Sjosephb #endif 2537c478bd9Sstevel@tonic-gate 2547c478bd9Sstevel@tonic-gate /* 2557c478bd9Sstevel@tonic-gate * Configuration parameters set at boot time. 2567c478bd9Sstevel@tonic-gate */ 2577c478bd9Sstevel@tonic-gate 2587c478bd9Sstevel@tonic-gate caddr_t econtig; /* end of first block of contiguous kernel */ 2597c478bd9Sstevel@tonic-gate 2607c478bd9Sstevel@tonic-gate struct bootops *bootops = 0; /* passed in from boot */ 2617c478bd9Sstevel@tonic-gate struct bootops **bootopsp; 2627c478bd9Sstevel@tonic-gate struct boot_syscalls *sysp; /* passed in from boot */ 2637c478bd9Sstevel@tonic-gate 2647c478bd9Sstevel@tonic-gate char bootblock_fstype[16]; 2657c478bd9Sstevel@tonic-gate 2667c478bd9Sstevel@tonic-gate char kern_bootargs[OBP_MAXPATHLEN]; 26719397407SSherry Moore char kern_bootfile[OBP_MAXPATHLEN]; 2687c478bd9Sstevel@tonic-gate 269ad23a2dbSjohansen /* 270ad23a2dbSjohansen * ZFS zio segment. This allows us to exclude large portions of ZFS data that 271ad23a2dbSjohansen * gets cached in kmem caches on the heap. If this is set to zero, we allocate 272ad23a2dbSjohansen * zio buffers from their own segment, otherwise they are allocated from the 273ad23a2dbSjohansen * heap. The optimization of allocating zio buffers from their own segment is 274ad23a2dbSjohansen * only valid on 64-bit kernels. 275ad23a2dbSjohansen */ 276ad23a2dbSjohansen #if defined(__amd64) 277ad23a2dbSjohansen int segzio_fromheap = 0; 278ad23a2dbSjohansen #else 279ad23a2dbSjohansen int segzio_fromheap = 1; 280ad23a2dbSjohansen #endif 281ad23a2dbSjohansen 2827c478bd9Sstevel@tonic-gate /* 2837c478bd9Sstevel@tonic-gate * new memory fragmentations are possible in startup() due to BOP_ALLOCs. this 2847c478bd9Sstevel@tonic-gate * depends on number of BOP_ALLOC calls made and requested size, memory size 2857c478bd9Sstevel@tonic-gate * combination and whether boot.bin memory needs to be freed. 2867c478bd9Sstevel@tonic-gate */ 2877c478bd9Sstevel@tonic-gate #define POSS_NEW_FRAGMENTS 12 2887c478bd9Sstevel@tonic-gate 2897c478bd9Sstevel@tonic-gate /* 2907c478bd9Sstevel@tonic-gate * VM data structures 2917c478bd9Sstevel@tonic-gate */ 2927c478bd9Sstevel@tonic-gate long page_hashsz; /* Size of page hash table (power of two) */ 293cb15d5d9SPeter Rival unsigned int page_hashsz_shift; /* log2(page_hashsz) */ 2947c478bd9Sstevel@tonic-gate struct page *pp_base; /* Base of initial system page struct array */ 2957c478bd9Sstevel@tonic-gate struct page **page_hash; /* Page hash table */ 296d7d93655Sblakej pad_mutex_t *pse_mutex; /* Locks protecting pp->p_selock */ 297d7d93655Sblakej size_t pse_table_size; /* Number of mutexes in pse_mutex[] */ 298d7d93655Sblakej int pse_shift; /* log2(pse_table_size) */ 2997c478bd9Sstevel@tonic-gate struct seg ktextseg; /* Segment used for kernel executable image */ 3007c478bd9Sstevel@tonic-gate struct seg kvalloc; /* Segment used for "valloc" mapping */ 3017c478bd9Sstevel@tonic-gate struct seg kpseg; /* Segment used for pageable kernel virt mem */ 3027c478bd9Sstevel@tonic-gate struct seg kmapseg; /* Segment used for generic kernel mappings */ 3037c478bd9Sstevel@tonic-gate struct seg kdebugseg; /* Segment used for the kernel debugger */ 3047c478bd9Sstevel@tonic-gate 3057c478bd9Sstevel@tonic-gate struct seg *segkmap = &kmapseg; /* Kernel generic mapping segment */ 306ae115bc7Smrj static struct seg *segmap = &kmapseg; /* easier to use name for in here */ 307ae115bc7Smrj 3087c478bd9Sstevel@tonic-gate struct seg *segkp = &kpseg; /* Pageable kernel virtual memory segment */ 3097c478bd9Sstevel@tonic-gate 3107c478bd9Sstevel@tonic-gate #if defined(__amd64) 3117c478bd9Sstevel@tonic-gate struct seg kvseg_core; /* Segment used for the core heap */ 3127c478bd9Sstevel@tonic-gate struct seg kpmseg; /* Segment used for physical mapping */ 3137c478bd9Sstevel@tonic-gate struct seg *segkpm = &kpmseg; /* 64bit kernel physical mapping segment */ 3147c478bd9Sstevel@tonic-gate #else 3157c478bd9Sstevel@tonic-gate struct seg *segkpm = NULL; /* Unused on IA32 */ 3167c478bd9Sstevel@tonic-gate #endif 3177c478bd9Sstevel@tonic-gate 3187c478bd9Sstevel@tonic-gate caddr_t segkp_base; /* Base address of segkp */ 319ad23a2dbSjohansen caddr_t segzio_base; /* Base address of segzio */ 3207c478bd9Sstevel@tonic-gate #if defined(__amd64) 3217c478bd9Sstevel@tonic-gate pgcnt_t segkpsize = btop(SEGKPDEFSIZE); /* size of segkp segment in pages */ 3227c478bd9Sstevel@tonic-gate #else 3237c478bd9Sstevel@tonic-gate pgcnt_t segkpsize = 0; 3247c478bd9Sstevel@tonic-gate #endif 325ad23a2dbSjohansen pgcnt_t segziosize = 0; /* size of zio segment in pages */ 3267c478bd9Sstevel@tonic-gate 327a3114836SGerry Liu /* 328a3114836SGerry Liu * A static DR page_t VA map is reserved that can map the page structures 329a3114836SGerry Liu * for a domain's entire RA space. The pages that back this space are 330a3114836SGerry Liu * dynamically allocated and need not be physically contiguous. The DR 331a3114836SGerry Liu * map size is derived from KPM size. 332a3114836SGerry Liu * This mechanism isn't used by x86 yet, so just stubs here. 333a3114836SGerry Liu */ 334a3114836SGerry Liu int ppvm_enable = 0; /* Static virtual map for page structs */ 335a3114836SGerry Liu page_t *ppvm_base = NULL; /* Base of page struct map */ 336a3114836SGerry Liu pgcnt_t ppvm_size = 0; /* Size of page struct map */ 337a3114836SGerry Liu 338ae115bc7Smrj /* 339ae115bc7Smrj * VA range available to the debugger 340ae115bc7Smrj */ 341ae115bc7Smrj const caddr_t kdi_segdebugbase = (const caddr_t)SEGDEBUGBASE; 342ae115bc7Smrj const size_t kdi_segdebugsize = SEGDEBUGSIZE; 343ae115bc7Smrj 3447c478bd9Sstevel@tonic-gate struct memseg *memseg_base; 3457c478bd9Sstevel@tonic-gate struct vnode unused_pages_vp; 3467c478bd9Sstevel@tonic-gate 3477c478bd9Sstevel@tonic-gate #define FOURGB 0x100000000LL 3487c478bd9Sstevel@tonic-gate 3497c478bd9Sstevel@tonic-gate struct memlist *memlist; 3507c478bd9Sstevel@tonic-gate 3517c478bd9Sstevel@tonic-gate caddr_t s_text; /* start of kernel text segment */ 3527c478bd9Sstevel@tonic-gate caddr_t e_text; /* end of kernel text segment */ 3537c478bd9Sstevel@tonic-gate caddr_t s_data; /* start of kernel data segment */ 3547c478bd9Sstevel@tonic-gate caddr_t e_data; /* end of kernel data segment */ 3557c478bd9Sstevel@tonic-gate caddr_t modtext; /* start of loadable module text reserved */ 3567c478bd9Sstevel@tonic-gate caddr_t e_modtext; /* end of loadable module text reserved */ 3577c478bd9Sstevel@tonic-gate caddr_t moddata; /* start of loadable module data reserved */ 3587c478bd9Sstevel@tonic-gate caddr_t e_moddata; /* end of loadable module data reserved */ 3597c478bd9Sstevel@tonic-gate 3607c478bd9Sstevel@tonic-gate struct memlist *phys_install; /* Total installed physical memory */ 3617c478bd9Sstevel@tonic-gate struct memlist *phys_avail; /* Total available physical memory */ 3621de082f7SVikram Hegde struct memlist *bios_rsvd; /* Bios reserved memory */ 3637c478bd9Sstevel@tonic-gate 3647c478bd9Sstevel@tonic-gate /* 3657c478bd9Sstevel@tonic-gate * kphysm_init returns the number of pages that were processed 3667c478bd9Sstevel@tonic-gate */ 367ae115bc7Smrj static pgcnt_t kphysm_init(page_t *, pgcnt_t); 3687c478bd9Sstevel@tonic-gate 3697c478bd9Sstevel@tonic-gate #define IO_PROP_SIZE 64 /* device property size */ 3707c478bd9Sstevel@tonic-gate 3717c478bd9Sstevel@tonic-gate /* 3727c478bd9Sstevel@tonic-gate * a couple useful roundup macros 3737c478bd9Sstevel@tonic-gate */ 3747c478bd9Sstevel@tonic-gate #define ROUND_UP_PAGE(x) \ 3757c478bd9Sstevel@tonic-gate ((uintptr_t)P2ROUNDUP((uintptr_t)(x), (uintptr_t)MMU_PAGESIZE)) 3767c478bd9Sstevel@tonic-gate #define ROUND_UP_LPAGE(x) \ 3777c478bd9Sstevel@tonic-gate ((uintptr_t)P2ROUNDUP((uintptr_t)(x), mmu.level_size[1])) 3787c478bd9Sstevel@tonic-gate #define ROUND_UP_4MEG(x) \ 379ae115bc7Smrj ((uintptr_t)P2ROUNDUP((uintptr_t)(x), (uintptr_t)FOUR_MEG)) 3807c478bd9Sstevel@tonic-gate #define ROUND_UP_TOPLEVEL(x) \ 3817c478bd9Sstevel@tonic-gate ((uintptr_t)P2ROUNDUP((uintptr_t)(x), mmu.level_size[mmu.max_level])) 3827c478bd9Sstevel@tonic-gate 3837c478bd9Sstevel@tonic-gate /* 3847c478bd9Sstevel@tonic-gate * 32-bit Kernel's Virtual memory layout. 3857c478bd9Sstevel@tonic-gate * +-----------------------+ 386ae115bc7Smrj * | | 3877c478bd9Sstevel@tonic-gate * 0xFFC00000 -|-----------------------|- ARGSBASE 3887c478bd9Sstevel@tonic-gate * | debugger | 3897c478bd9Sstevel@tonic-gate * 0xFF800000 -|-----------------------|- SEGDEBUGBASE 3907c478bd9Sstevel@tonic-gate * | Kernel Data | 3917c478bd9Sstevel@tonic-gate * 0xFEC00000 -|-----------------------| 3927c478bd9Sstevel@tonic-gate * | Kernel Text | 393843e1988Sjohnlev * 0xFE800000 -|-----------------------|- KERNEL_TEXT (0xFB400000 on Xen) 394ae115bc7Smrj * |--- GDT ---|- GDT page (GDT_VA) 395ae115bc7Smrj * |--- debug info ---|- debug info (DEBUG_INFO_VA) 396ae115bc7Smrj * | | 397ae115bc7Smrj * | page_t structures | 3987c478bd9Sstevel@tonic-gate * | memsegs, memlists, | 3997c478bd9Sstevel@tonic-gate * | page hash, etc. | 400ae115bc7Smrj * --- -|-----------------------|- ekernelheap, valloc_base (floating) 401ae115bc7Smrj * | | (segkp is just an arena in the heap) 4027c478bd9Sstevel@tonic-gate * | | 4037c478bd9Sstevel@tonic-gate * | kvseg | 4047c478bd9Sstevel@tonic-gate * | | 4057c478bd9Sstevel@tonic-gate * | | 4067c478bd9Sstevel@tonic-gate * --- -|-----------------------|- kernelheap (floating) 4077c478bd9Sstevel@tonic-gate * | Segkmap | 408ae115bc7Smrj * 0xC3002000 -|-----------------------|- segmap_start (floating) 4097c478bd9Sstevel@tonic-gate * | Red Zone | 4107c478bd9Sstevel@tonic-gate * 0xC3000000 -|-----------------------|- kernelbase / userlimit (floating) 4117c478bd9Sstevel@tonic-gate * | | || 4127c478bd9Sstevel@tonic-gate * | Shared objects | \/ 4137c478bd9Sstevel@tonic-gate * | | 4147c478bd9Sstevel@tonic-gate * : : 4157c478bd9Sstevel@tonic-gate * | user data | 4167c478bd9Sstevel@tonic-gate * |-----------------------| 4177c478bd9Sstevel@tonic-gate * | user text | 4187c478bd9Sstevel@tonic-gate * 0x08048000 -|-----------------------| 4197c478bd9Sstevel@tonic-gate * | user stack | 4207c478bd9Sstevel@tonic-gate * : : 4217c478bd9Sstevel@tonic-gate * | invalid | 4227c478bd9Sstevel@tonic-gate * 0x00000000 +-----------------------+ 4237c478bd9Sstevel@tonic-gate * 4247c478bd9Sstevel@tonic-gate * 4257c478bd9Sstevel@tonic-gate * 64-bit Kernel's Virtual memory layout. (assuming 64 bit app) 4267c478bd9Sstevel@tonic-gate * +-----------------------+ 427ae115bc7Smrj * | | 4287c478bd9Sstevel@tonic-gate * 0xFFFFFFFF.FFC00000 |-----------------------|- ARGSBASE 4297c478bd9Sstevel@tonic-gate * | debugger (?) | 4307c478bd9Sstevel@tonic-gate * 0xFFFFFFFF.FF800000 |-----------------------|- SEGDEBUGBASE 4317c478bd9Sstevel@tonic-gate * | unused | 4327c478bd9Sstevel@tonic-gate * +-----------------------+ 4337c478bd9Sstevel@tonic-gate * | Kernel Data | 4347c478bd9Sstevel@tonic-gate * 0xFFFFFFFF.FBC00000 |-----------------------| 4357c478bd9Sstevel@tonic-gate * | Kernel Text | 4367c478bd9Sstevel@tonic-gate * 0xFFFFFFFF.FB800000 |-----------------------|- KERNEL_TEXT 437ae115bc7Smrj * |--- GDT ---|- GDT page (GDT_VA) 438ae115bc7Smrj * |--- debug info ---|- debug info (DEBUG_INFO_VA) 439ae115bc7Smrj * | | 4407c478bd9Sstevel@tonic-gate * | Core heap | (used for loadable modules) 4417c478bd9Sstevel@tonic-gate * 0xFFFFFFFF.C0000000 |-----------------------|- core_base / ekernelheap 4427c478bd9Sstevel@tonic-gate * | Kernel | 4437c478bd9Sstevel@tonic-gate * | heap | 4447c478bd9Sstevel@tonic-gate * 0xFFFFFXXX.XXX00000 |-----------------------|- kernelheap (floating) 445ae115bc7Smrj * | segmap | 446ae115bc7Smrj * 0xFFFFFXXX.XXX00000 |-----------------------|- segmap_start (floating) 4477c478bd9Sstevel@tonic-gate * | device mappings | 4487c478bd9Sstevel@tonic-gate * 0xFFFFFXXX.XXX00000 |-----------------------|- toxic_addr (floating) 449ad23a2dbSjohansen * | segzio | 450ad23a2dbSjohansen * 0xFFFFFXXX.XXX00000 |-----------------------|- segzio_base (floating) 4517c478bd9Sstevel@tonic-gate * | segkp | 452ae115bc7Smrj * --- |-----------------------|- segkp_base (floating) 453ae115bc7Smrj * | page_t structures | valloc_base + valloc_sz 454ae115bc7Smrj * | memsegs, memlists, | 455ae115bc7Smrj * | page hash, etc. | 45635b1ab99Sjosephb * 0xFFFFFF00.00000000 |-----------------------|- valloc_base (lower if > 1TB) 4577c478bd9Sstevel@tonic-gate * | segkpm | 4587c478bd9Sstevel@tonic-gate * 0xFFFFFE00.00000000 |-----------------------| 4597c478bd9Sstevel@tonic-gate * | Red Zone | 46035b1ab99Sjosephb * 0xFFFFFD80.00000000 |-----------------------|- KERNELBASE (lower if > 1TB) 4617c478bd9Sstevel@tonic-gate * | User stack |- User space memory 4627c478bd9Sstevel@tonic-gate * | | 4637c478bd9Sstevel@tonic-gate * | shared objects, etc | (grows downwards) 4647c478bd9Sstevel@tonic-gate * : : 4657c478bd9Sstevel@tonic-gate * | | 4667c478bd9Sstevel@tonic-gate * 0xFFFF8000.00000000 |-----------------------| 4677c478bd9Sstevel@tonic-gate * | | 4687c478bd9Sstevel@tonic-gate * | VA Hole / unused | 4697c478bd9Sstevel@tonic-gate * | | 4707c478bd9Sstevel@tonic-gate * 0x00008000.00000000 |-----------------------| 4717c478bd9Sstevel@tonic-gate * | | 4727c478bd9Sstevel@tonic-gate * | | 4737c478bd9Sstevel@tonic-gate * : : 4747c478bd9Sstevel@tonic-gate * | user heap | (grows upwards) 4757c478bd9Sstevel@tonic-gate * | | 4767c478bd9Sstevel@tonic-gate * | user data | 4777c478bd9Sstevel@tonic-gate * |-----------------------| 4787c478bd9Sstevel@tonic-gate * | user text | 4797c478bd9Sstevel@tonic-gate * 0x00000000.04000000 |-----------------------| 4807c478bd9Sstevel@tonic-gate * | invalid | 4817c478bd9Sstevel@tonic-gate * 0x00000000.00000000 +-----------------------+ 4827c478bd9Sstevel@tonic-gate * 4837c478bd9Sstevel@tonic-gate * A 32 bit app on the 64 bit kernel sees the same layout as on the 32 bit 4847c478bd9Sstevel@tonic-gate * kernel, except that userlimit is raised to 0xfe000000 4857c478bd9Sstevel@tonic-gate * 4867c478bd9Sstevel@tonic-gate * Floating values: 4877c478bd9Sstevel@tonic-gate * 4887c478bd9Sstevel@tonic-gate * valloc_base: start of the kernel's memory management/tracking data 489ae115bc7Smrj * structures. This region contains page_t structures for 490ae115bc7Smrj * physical memory, memsegs, memlists, and the page hash. 4917c478bd9Sstevel@tonic-gate * 4927c478bd9Sstevel@tonic-gate * core_base: start of the kernel's "core" heap area on 64-bit systems. 4937c478bd9Sstevel@tonic-gate * This area is intended to be used for global data as well as for module 4947c478bd9Sstevel@tonic-gate * text/data that does not fit into the nucleus pages. The core heap is 4957c478bd9Sstevel@tonic-gate * restricted to a 2GB range, allowing every address within it to be 4967c478bd9Sstevel@tonic-gate * accessed using rip-relative addressing 4977c478bd9Sstevel@tonic-gate * 4987c478bd9Sstevel@tonic-gate * ekernelheap: end of kernelheap and start of segmap. 4997c478bd9Sstevel@tonic-gate * 5007c478bd9Sstevel@tonic-gate * kernelheap: start of kernel heap. On 32-bit systems, this starts right 5017c478bd9Sstevel@tonic-gate * above a red zone that separates the user's address space from the 5027c478bd9Sstevel@tonic-gate * kernel's. On 64-bit systems, it sits above segkp and segkpm. 5037c478bd9Sstevel@tonic-gate * 504ae115bc7Smrj * segmap_start: start of segmap. The length of segmap can be modified 50552eef812SAmrita Sadhukhan * through eeprom. The default length is 16MB on 32-bit systems and 64MB 50652eef812SAmrita Sadhukhan * on 64-bit systems. 5077c478bd9Sstevel@tonic-gate * 5087c478bd9Sstevel@tonic-gate * kernelbase: On a 32-bit kernel the default value of 0xd4000000 will be 5097c478bd9Sstevel@tonic-gate * decreased by 2X the size required for page_t. This allows the kernel 5107c478bd9Sstevel@tonic-gate * heap to grow in size with physical memory. With sizeof(page_t) == 80 5117c478bd9Sstevel@tonic-gate * bytes, the following shows the values of kernelbase and kernel heap 5127c478bd9Sstevel@tonic-gate * sizes for different memory configurations (assuming default segmap and 5137c478bd9Sstevel@tonic-gate * segkp sizes). 5147c478bd9Sstevel@tonic-gate * 5157c478bd9Sstevel@tonic-gate * mem size for kernelbase kernel heap 5167c478bd9Sstevel@tonic-gate * size page_t's size 5177c478bd9Sstevel@tonic-gate * ---- --------- ---------- ----------- 5187c478bd9Sstevel@tonic-gate * 1gb 0x01400000 0xd1800000 684MB 5197c478bd9Sstevel@tonic-gate * 2gb 0x02800000 0xcf000000 704MB 5207c478bd9Sstevel@tonic-gate * 4gb 0x05000000 0xca000000 744MB 5217c478bd9Sstevel@tonic-gate * 6gb 0x07800000 0xc5000000 784MB 5227c478bd9Sstevel@tonic-gate * 8gb 0x0a000000 0xc0000000 824MB 5237c478bd9Sstevel@tonic-gate * 16gb 0x14000000 0xac000000 984MB 5247c478bd9Sstevel@tonic-gate * 32gb 0x28000000 0x84000000 1304MB 5257c478bd9Sstevel@tonic-gate * 64gb 0x50000000 0x34000000 1944MB (*) 5267c478bd9Sstevel@tonic-gate * 5277c478bd9Sstevel@tonic-gate * kernelbase is less than the abi minimum of 0xc0000000 for memory 5287c478bd9Sstevel@tonic-gate * configurations above 8gb. 5297c478bd9Sstevel@tonic-gate * 5307c478bd9Sstevel@tonic-gate * (*) support for memory configurations above 32gb will require manual tuning 5317c478bd9Sstevel@tonic-gate * of kernelbase to balance out the need of user applications. 5327c478bd9Sstevel@tonic-gate */ 5337c478bd9Sstevel@tonic-gate 5347c478bd9Sstevel@tonic-gate /* real-time-clock initialization parameters */ 535ae115bc7Smrj extern time_t process_rtc_config_file(void); 5367c478bd9Sstevel@tonic-gate 5377c478bd9Sstevel@tonic-gate uintptr_t kernelbase; 538ae115bc7Smrj uintptr_t postbootkernelbase; /* not set till boot loader is gone */ 5397c478bd9Sstevel@tonic-gate uintptr_t eprom_kernelbase; 5407c478bd9Sstevel@tonic-gate size_t segmapsize; 541ae115bc7Smrj uintptr_t segmap_start; 5427c478bd9Sstevel@tonic-gate int segmapfreelists; 5437c478bd9Sstevel@tonic-gate pgcnt_t npages; 544ae115bc7Smrj pgcnt_t orig_npages; 5457c478bd9Sstevel@tonic-gate size_t core_size; /* size of "core" heap */ 5467c478bd9Sstevel@tonic-gate uintptr_t core_base; /* base address of "core" heap */ 5477c478bd9Sstevel@tonic-gate 5487c478bd9Sstevel@tonic-gate /* 5497c478bd9Sstevel@tonic-gate * List of bootstrap pages. We mark these as allocated in startup. 5507c478bd9Sstevel@tonic-gate * release_bootstrap() will free them when we're completely done with 5517c478bd9Sstevel@tonic-gate * the bootstrap. 5527c478bd9Sstevel@tonic-gate */ 553ae115bc7Smrj static page_t *bootpages; 554ae115bc7Smrj 555ae115bc7Smrj /* 556ae115bc7Smrj * boot time pages that have a vnode from the ramdisk will keep that forever. 557ae115bc7Smrj */ 558ae115bc7Smrj static page_t *rd_pages; 5597c478bd9Sstevel@tonic-gate 56019397407SSherry Moore /* 56119397407SSherry Moore * Lower 64K 56219397407SSherry Moore */ 56319397407SSherry Moore static page_t *lower_pages = NULL; 56419397407SSherry Moore static int lower_pages_count = 0; 56519397407SSherry Moore 5667c478bd9Sstevel@tonic-gate struct system_hardware system_hardware; 5677c478bd9Sstevel@tonic-gate 5687c478bd9Sstevel@tonic-gate /* 5697c478bd9Sstevel@tonic-gate * Enable some debugging messages concerning memory usage... 5707c478bd9Sstevel@tonic-gate */ 5717c478bd9Sstevel@tonic-gate static void 572ae115bc7Smrj print_memlist(char *title, struct memlist *mp) 5737c478bd9Sstevel@tonic-gate { 5747c478bd9Sstevel@tonic-gate prom_printf("MEMLIST: %s:\n", title); 5757c478bd9Sstevel@tonic-gate while (mp != NULL) { 5767c478bd9Sstevel@tonic-gate prom_printf("\tAddress 0x%" PRIx64 ", size 0x%" PRIx64 "\n", 57756f33205SJonathan Adams mp->ml_address, mp->ml_size); 57856f33205SJonathan Adams mp = mp->ml_next; 5797c478bd9Sstevel@tonic-gate } 5807c478bd9Sstevel@tonic-gate } 5817c478bd9Sstevel@tonic-gate 5827c478bd9Sstevel@tonic-gate /* 5837c478bd9Sstevel@tonic-gate * XX64 need a comment here.. are these just default values, surely 5847c478bd9Sstevel@tonic-gate * we read the "cpuid" type information to figure this out. 5857c478bd9Sstevel@tonic-gate */ 5867c478bd9Sstevel@tonic-gate int l2cache_sz = 0x80000; 5877c478bd9Sstevel@tonic-gate int l2cache_linesz = 0x40; 5887c478bd9Sstevel@tonic-gate int l2cache_assoc = 1; 5897c478bd9Sstevel@tonic-gate 5902cb27123Saguzovsk static size_t textrepl_min_gb = 10; 5912cb27123Saguzovsk 5927c478bd9Sstevel@tonic-gate /* 5937c478bd9Sstevel@tonic-gate * on 64 bit we use a predifined VA range for mapping devices in the kernel 5947c478bd9Sstevel@tonic-gate * on 32 bit the mappings are intermixed in the heap, so we use a bit map 5957c478bd9Sstevel@tonic-gate */ 5967c478bd9Sstevel@tonic-gate #ifdef __amd64 5977c478bd9Sstevel@tonic-gate 5987c478bd9Sstevel@tonic-gate vmem_t *device_arena; 5997c478bd9Sstevel@tonic-gate uintptr_t toxic_addr = (uintptr_t)NULL; 600ae115bc7Smrj size_t toxic_size = 1024 * 1024 * 1024; /* Sparc uses 1 gig too */ 6017c478bd9Sstevel@tonic-gate 6027c478bd9Sstevel@tonic-gate #else /* __i386 */ 6037c478bd9Sstevel@tonic-gate 6047c478bd9Sstevel@tonic-gate ulong_t *toxic_bit_map; /* one bit for each 4k of VA in heap_arena */ 6057c478bd9Sstevel@tonic-gate size_t toxic_bit_map_len = 0; /* in bits */ 6067c478bd9Sstevel@tonic-gate 6077c478bd9Sstevel@tonic-gate #endif /* __i386 */ 6087c478bd9Sstevel@tonic-gate 6097c478bd9Sstevel@tonic-gate /* 6107c478bd9Sstevel@tonic-gate * Simple boot time debug facilities 6117c478bd9Sstevel@tonic-gate */ 6127c478bd9Sstevel@tonic-gate static char *prm_dbg_str[] = { 6137c478bd9Sstevel@tonic-gate "%s:%d: '%s' is 0x%x\n", 6147c478bd9Sstevel@tonic-gate "%s:%d: '%s' is 0x%llx\n" 6157c478bd9Sstevel@tonic-gate }; 6167c478bd9Sstevel@tonic-gate 6177c478bd9Sstevel@tonic-gate int prom_debug; 6187c478bd9Sstevel@tonic-gate 6197c478bd9Sstevel@tonic-gate #define PRM_DEBUG(q) if (prom_debug) \ 6207c478bd9Sstevel@tonic-gate prom_printf(prm_dbg_str[sizeof (q) >> 3], "startup.c", __LINE__, #q, q); 6217c478bd9Sstevel@tonic-gate #define PRM_POINT(q) if (prom_debug) \ 6227c478bd9Sstevel@tonic-gate prom_printf("%s:%d: %s\n", "startup.c", __LINE__, q); 6237c478bd9Sstevel@tonic-gate 6247c478bd9Sstevel@tonic-gate /* 6257c478bd9Sstevel@tonic-gate * This structure is used to keep track of the intial allocations 6267c478bd9Sstevel@tonic-gate * done in startup_memlist(). The value of NUM_ALLOCATIONS needs to 6277c478bd9Sstevel@tonic-gate * be >= the number of ADD_TO_ALLOCATIONS() executed in the code. 6287c478bd9Sstevel@tonic-gate */ 6291de082f7SVikram Hegde #define NUM_ALLOCATIONS 8 6307c478bd9Sstevel@tonic-gate int num_allocations = 0; 6317c478bd9Sstevel@tonic-gate struct { 6327c478bd9Sstevel@tonic-gate void **al_ptr; 6337c478bd9Sstevel@tonic-gate size_t al_size; 6347c478bd9Sstevel@tonic-gate } allocations[NUM_ALLOCATIONS]; 6357c478bd9Sstevel@tonic-gate size_t valloc_sz = 0; 6367c478bd9Sstevel@tonic-gate uintptr_t valloc_base; 6377c478bd9Sstevel@tonic-gate 6387c478bd9Sstevel@tonic-gate #define ADD_TO_ALLOCATIONS(ptr, size) { \ 6397c478bd9Sstevel@tonic-gate size = ROUND_UP_PAGE(size); \ 6407c478bd9Sstevel@tonic-gate if (num_allocations == NUM_ALLOCATIONS) \ 6417c478bd9Sstevel@tonic-gate panic("too many ADD_TO_ALLOCATIONS()"); \ 6427c478bd9Sstevel@tonic-gate allocations[num_allocations].al_ptr = (void**)&ptr; \ 6437c478bd9Sstevel@tonic-gate allocations[num_allocations].al_size = size; \ 6447c478bd9Sstevel@tonic-gate valloc_sz += size; \ 6457c478bd9Sstevel@tonic-gate ++num_allocations; \ 6467c478bd9Sstevel@tonic-gate } 6477c478bd9Sstevel@tonic-gate 648ae115bc7Smrj /* 649ae115bc7Smrj * Allocate all the initial memory needed by the page allocator. 650ae115bc7Smrj */ 6517c478bd9Sstevel@tonic-gate static void 6527c478bd9Sstevel@tonic-gate perform_allocations(void) 6537c478bd9Sstevel@tonic-gate { 6547c478bd9Sstevel@tonic-gate caddr_t mem; 6557c478bd9Sstevel@tonic-gate int i; 656ae115bc7Smrj int valloc_align; 6577c478bd9Sstevel@tonic-gate 658ae115bc7Smrj PRM_DEBUG(valloc_base); 659ae115bc7Smrj PRM_DEBUG(valloc_sz); 660ae115bc7Smrj valloc_align = mmu.level_size[mmu.max_page_level > 0]; 661ae115bc7Smrj mem = BOP_ALLOC(bootops, (caddr_t)valloc_base, valloc_sz, valloc_align); 6627c478bd9Sstevel@tonic-gate if (mem != (caddr_t)valloc_base) 6637c478bd9Sstevel@tonic-gate panic("BOP_ALLOC() failed"); 6647c478bd9Sstevel@tonic-gate bzero(mem, valloc_sz); 6657c478bd9Sstevel@tonic-gate for (i = 0; i < num_allocations; ++i) { 6667c478bd9Sstevel@tonic-gate *allocations[i].al_ptr = (void *)mem; 6677c478bd9Sstevel@tonic-gate mem += allocations[i].al_size; 6687c478bd9Sstevel@tonic-gate } 6697c478bd9Sstevel@tonic-gate } 6707c478bd9Sstevel@tonic-gate 6717c478bd9Sstevel@tonic-gate /* 6727c478bd9Sstevel@tonic-gate * Our world looks like this at startup time. 6737c478bd9Sstevel@tonic-gate * 6747c478bd9Sstevel@tonic-gate * In a 32-bit OS, boot loads the kernel text at 0xfe800000 and kernel data 6757c478bd9Sstevel@tonic-gate * at 0xfec00000. On a 64-bit OS, kernel text and data are loaded at 6767c478bd9Sstevel@tonic-gate * 0xffffffff.fe800000 and 0xffffffff.fec00000 respectively. Those 6777c478bd9Sstevel@tonic-gate * addresses are fixed in the binary at link time. 6787c478bd9Sstevel@tonic-gate * 6797c478bd9Sstevel@tonic-gate * On the text page: 6807c478bd9Sstevel@tonic-gate * unix/genunix/krtld/module text loads. 6817c478bd9Sstevel@tonic-gate * 6827c478bd9Sstevel@tonic-gate * On the data page: 683ae115bc7Smrj * unix/genunix/krtld/module data loads. 684ae115bc7Smrj * 6857c478bd9Sstevel@tonic-gate * Machine-dependent startup code 6867c478bd9Sstevel@tonic-gate */ 6877c478bd9Sstevel@tonic-gate void 6887c478bd9Sstevel@tonic-gate startup(void) 6897c478bd9Sstevel@tonic-gate { 690843e1988Sjohnlev #if !defined(__xpv) 69175bcd456Sjg extern void startup_pci_bios(void); 692843e1988Sjohnlev #endif 693a563a037Sbholler extern cpuset_t cpu_ready_set; 694a563a037Sbholler 695843e1988Sjohnlev /* 696843e1988Sjohnlev * Make sure that nobody tries to use sekpm until we have 697843e1988Sjohnlev * initialized it properly. 698843e1988Sjohnlev */ 6997c478bd9Sstevel@tonic-gate #if defined(__amd64) 70035b1ab99Sjosephb kpm_desired = 1; 7017c478bd9Sstevel@tonic-gate #endif 7027c478bd9Sstevel@tonic-gate kpm_enable = 0; 703a563a037Sbholler CPUSET_ONLY(cpu_ready_set, 0); /* cpu 0 is boot cpu */ 7047c478bd9Sstevel@tonic-gate 705843e1988Sjohnlev #if defined(__xpv) /* XXPV fix me! */ 706843e1988Sjohnlev { 707843e1988Sjohnlev extern int segvn_use_regions; 708843e1988Sjohnlev segvn_use_regions = 0; 709843e1988Sjohnlev } 710843e1988Sjohnlev #endif 7117c478bd9Sstevel@tonic-gate progressbar_init(); 7127c478bd9Sstevel@tonic-gate startup_init(); 713ab4a9bebSjohnlev #if defined(__xpv) 714ab4a9bebSjohnlev startup_xen_version(); 715ab4a9bebSjohnlev #endif 7167c478bd9Sstevel@tonic-gate startup_memlist(); 717ae115bc7Smrj startup_kmem(); 71835b1ab99Sjosephb startup_vm(); 719843e1988Sjohnlev #if !defined(__xpv) 7200db3240dSStephen Hanson /* 7210db3240dSStephen Hanson * Note we need to do this even on fast reboot in order to access 7220db3240dSStephen Hanson * the irq routing table (used for pci labels). 7230db3240dSStephen Hanson */ 7240db3240dSStephen Hanson startup_pci_bios(); 725e4b86885SCheng Sean Ye #endif 726e4b86885SCheng Sean Ye #if defined(__xpv) 727e4b86885SCheng Sean Ye startup_xen_mca(); 728843e1988Sjohnlev #endif 7297c478bd9Sstevel@tonic-gate startup_modules(); 7308770118eSlipeng sang - Sun Microsystems - Beijing China 7317c478bd9Sstevel@tonic-gate startup_end(); 7327c478bd9Sstevel@tonic-gate } 7337c478bd9Sstevel@tonic-gate 7347c478bd9Sstevel@tonic-gate static void 7357c478bd9Sstevel@tonic-gate startup_init() 7367c478bd9Sstevel@tonic-gate { 7377c478bd9Sstevel@tonic-gate PRM_POINT("startup_init() starting..."); 7387c478bd9Sstevel@tonic-gate 7397c478bd9Sstevel@tonic-gate /* 7407c478bd9Sstevel@tonic-gate * Complete the extraction of cpuid data 7417c478bd9Sstevel@tonic-gate */ 7427c478bd9Sstevel@tonic-gate cpuid_pass2(CPU); 7437c478bd9Sstevel@tonic-gate 7447c478bd9Sstevel@tonic-gate (void) check_boot_version(BOP_GETVERSION(bootops)); 7457c478bd9Sstevel@tonic-gate 7467c478bd9Sstevel@tonic-gate /* 7477c478bd9Sstevel@tonic-gate * Check for prom_debug in boot environment 7487c478bd9Sstevel@tonic-gate */ 7497c478bd9Sstevel@tonic-gate if (BOP_GETPROPLEN(bootops, "prom_debug") >= 0) { 7507c478bd9Sstevel@tonic-gate ++prom_debug; 7517c478bd9Sstevel@tonic-gate PRM_POINT("prom_debug found in boot enviroment"); 7527c478bd9Sstevel@tonic-gate } 7537c478bd9Sstevel@tonic-gate 7547c478bd9Sstevel@tonic-gate /* 7557c478bd9Sstevel@tonic-gate * Collect node, cpu and memory configuration information. 7567c478bd9Sstevel@tonic-gate */ 7577c478bd9Sstevel@tonic-gate get_system_configuration(); 7587c478bd9Sstevel@tonic-gate 7597c478bd9Sstevel@tonic-gate /* 7607c478bd9Sstevel@tonic-gate * Halt if this is an unsupported processor. 7617c478bd9Sstevel@tonic-gate */ 7627c478bd9Sstevel@tonic-gate if (x86_type == X86_TYPE_486 || x86_type == X86_TYPE_CYRIX_486) { 7637c478bd9Sstevel@tonic-gate printf("\n486 processor (\"%s\") detected.\n", 7647c478bd9Sstevel@tonic-gate CPU->cpu_brandstr); 7657c478bd9Sstevel@tonic-gate halt("This processor is not supported by this release " 7667c478bd9Sstevel@tonic-gate "of Solaris."); 7677c478bd9Sstevel@tonic-gate } 7687c478bd9Sstevel@tonic-gate 7697c478bd9Sstevel@tonic-gate PRM_POINT("startup_init() done"); 7707c478bd9Sstevel@tonic-gate } 7717c478bd9Sstevel@tonic-gate 7727c478bd9Sstevel@tonic-gate /* 7737c478bd9Sstevel@tonic-gate * Callback for copy_memlist_filter() to filter nucleus, kadb/kmdb, (ie. 7747c478bd9Sstevel@tonic-gate * everything mapped above KERNEL_TEXT) pages from phys_avail. Note it 7757c478bd9Sstevel@tonic-gate * also filters out physical page zero. There is some reliance on the 7767c478bd9Sstevel@tonic-gate * boot loader allocating only a few contiguous physical memory chunks. 7777c478bd9Sstevel@tonic-gate */ 7787c478bd9Sstevel@tonic-gate static void 7797c478bd9Sstevel@tonic-gate avail_filter(uint64_t *addr, uint64_t *size) 7807c478bd9Sstevel@tonic-gate { 7817c478bd9Sstevel@tonic-gate uintptr_t va; 7827c478bd9Sstevel@tonic-gate uintptr_t next_va; 7837c478bd9Sstevel@tonic-gate pfn_t pfn; 7847c478bd9Sstevel@tonic-gate uint64_t pfn_addr; 7857c478bd9Sstevel@tonic-gate uint64_t pfn_eaddr; 7867c478bd9Sstevel@tonic-gate uint_t prot; 7877c478bd9Sstevel@tonic-gate size_t len; 7887c478bd9Sstevel@tonic-gate uint_t change; 7897c478bd9Sstevel@tonic-gate 7907c478bd9Sstevel@tonic-gate if (prom_debug) 7917c478bd9Sstevel@tonic-gate prom_printf("\tFilter: in: a=%" PRIx64 ", s=%" PRIx64 "\n", 7927c478bd9Sstevel@tonic-gate *addr, *size); 7937c478bd9Sstevel@tonic-gate 7947c478bd9Sstevel@tonic-gate /* 7957c478bd9Sstevel@tonic-gate * page zero is required for BIOS.. never make it available 7967c478bd9Sstevel@tonic-gate */ 7977c478bd9Sstevel@tonic-gate if (*addr == 0) { 7987c478bd9Sstevel@tonic-gate *addr += MMU_PAGESIZE; 7997c478bd9Sstevel@tonic-gate *size -= MMU_PAGESIZE; 8007c478bd9Sstevel@tonic-gate } 8017c478bd9Sstevel@tonic-gate 8027c478bd9Sstevel@tonic-gate /* 803ae115bc7Smrj * First we trim from the front of the range. Since kbm_probe() 8047c478bd9Sstevel@tonic-gate * walks ranges in virtual order, but addr/size are physical, we need 8057c478bd9Sstevel@tonic-gate * to the list until no changes are seen. This deals with the case 8067c478bd9Sstevel@tonic-gate * where page "p" is mapped at v, page "p + PAGESIZE" is mapped at w 8077c478bd9Sstevel@tonic-gate * but w < v. 8087c478bd9Sstevel@tonic-gate */ 8097c478bd9Sstevel@tonic-gate do { 8107c478bd9Sstevel@tonic-gate change = 0; 8117c478bd9Sstevel@tonic-gate for (va = KERNEL_TEXT; 812ae115bc7Smrj *size > 0 && kbm_probe(&va, &len, &pfn, &prot) != 0; 8137c478bd9Sstevel@tonic-gate va = next_va) { 8147c478bd9Sstevel@tonic-gate 8157c478bd9Sstevel@tonic-gate next_va = va + len; 816ae115bc7Smrj pfn_addr = pfn_to_pa(pfn); 8177c478bd9Sstevel@tonic-gate pfn_eaddr = pfn_addr + len; 8187c478bd9Sstevel@tonic-gate 8197c478bd9Sstevel@tonic-gate if (pfn_addr <= *addr && pfn_eaddr > *addr) { 8207c478bd9Sstevel@tonic-gate change = 1; 8217c478bd9Sstevel@tonic-gate while (*size > 0 && len > 0) { 8227c478bd9Sstevel@tonic-gate *addr += MMU_PAGESIZE; 8237c478bd9Sstevel@tonic-gate *size -= MMU_PAGESIZE; 8247c478bd9Sstevel@tonic-gate len -= MMU_PAGESIZE; 8257c478bd9Sstevel@tonic-gate } 8267c478bd9Sstevel@tonic-gate } 8277c478bd9Sstevel@tonic-gate } 8287c478bd9Sstevel@tonic-gate if (change && prom_debug) 8297c478bd9Sstevel@tonic-gate prom_printf("\t\ttrim: a=%" PRIx64 ", s=%" PRIx64 "\n", 8307c478bd9Sstevel@tonic-gate *addr, *size); 8317c478bd9Sstevel@tonic-gate } while (change); 8327c478bd9Sstevel@tonic-gate 8337c478bd9Sstevel@tonic-gate /* 8347c478bd9Sstevel@tonic-gate * Trim pages from the end of the range. 8357c478bd9Sstevel@tonic-gate */ 8367c478bd9Sstevel@tonic-gate for (va = KERNEL_TEXT; 837ae115bc7Smrj *size > 0 && kbm_probe(&va, &len, &pfn, &prot) != 0; 8387c478bd9Sstevel@tonic-gate va = next_va) { 8397c478bd9Sstevel@tonic-gate 8407c478bd9Sstevel@tonic-gate next_va = va + len; 841ae115bc7Smrj pfn_addr = pfn_to_pa(pfn); 8427c478bd9Sstevel@tonic-gate 8437c478bd9Sstevel@tonic-gate if (pfn_addr >= *addr && pfn_addr < *addr + *size) 8447c478bd9Sstevel@tonic-gate *size = pfn_addr - *addr; 8457c478bd9Sstevel@tonic-gate } 8467c478bd9Sstevel@tonic-gate 8477c478bd9Sstevel@tonic-gate if (prom_debug) 8487c478bd9Sstevel@tonic-gate prom_printf("\tFilter out: a=%" PRIx64 ", s=%" PRIx64 "\n", 8497c478bd9Sstevel@tonic-gate *addr, *size); 8507c478bd9Sstevel@tonic-gate } 8517c478bd9Sstevel@tonic-gate 8527c478bd9Sstevel@tonic-gate static void 8537c478bd9Sstevel@tonic-gate kpm_init() 8547c478bd9Sstevel@tonic-gate { 8557c478bd9Sstevel@tonic-gate struct segkpm_crargs b; 8567c478bd9Sstevel@tonic-gate 8577c478bd9Sstevel@tonic-gate /* 8587c478bd9Sstevel@tonic-gate * These variables were all designed for sfmmu in which segkpm is 8597c478bd9Sstevel@tonic-gate * mapped using a single pagesize - either 8KB or 4MB. On x86, we 8607c478bd9Sstevel@tonic-gate * might use 2+ page sizes on a single machine, so none of these 8617c478bd9Sstevel@tonic-gate * variables have a single correct value. They are set up as if we 8627c478bd9Sstevel@tonic-gate * always use a 4KB pagesize, which should do no harm. In the long 8637c478bd9Sstevel@tonic-gate * run, we should get rid of KPM's assumption that only a single 8647c478bd9Sstevel@tonic-gate * pagesize is used. 8657c478bd9Sstevel@tonic-gate */ 8667c478bd9Sstevel@tonic-gate kpm_pgshft = MMU_PAGESHIFT; 8677c478bd9Sstevel@tonic-gate kpm_pgsz = MMU_PAGESIZE; 8687c478bd9Sstevel@tonic-gate kpm_pgoff = MMU_PAGEOFFSET; 8697c478bd9Sstevel@tonic-gate kpmp2pshft = 0; 8707c478bd9Sstevel@tonic-gate kpmpnpgs = 1; 8717c478bd9Sstevel@tonic-gate ASSERT(((uintptr_t)kpm_vbase & (kpm_pgsz - 1)) == 0); 8727c478bd9Sstevel@tonic-gate 8737c478bd9Sstevel@tonic-gate PRM_POINT("about to create segkpm"); 8747c478bd9Sstevel@tonic-gate rw_enter(&kas.a_lock, RW_WRITER); 8757c478bd9Sstevel@tonic-gate 8767c478bd9Sstevel@tonic-gate if (seg_attach(&kas, kpm_vbase, kpm_size, segkpm) < 0) 8777c478bd9Sstevel@tonic-gate panic("cannot attach segkpm"); 8787c478bd9Sstevel@tonic-gate 8797c478bd9Sstevel@tonic-gate b.prot = PROT_READ | PROT_WRITE; 8807c478bd9Sstevel@tonic-gate b.nvcolors = 1; 8817c478bd9Sstevel@tonic-gate 8827c478bd9Sstevel@tonic-gate if (segkpm_create(segkpm, (caddr_t)&b) != 0) 8837c478bd9Sstevel@tonic-gate panic("segkpm_create segkpm"); 8847c478bd9Sstevel@tonic-gate 8857c478bd9Sstevel@tonic-gate rw_exit(&kas.a_lock); 886ae115bc7Smrj } 8877c478bd9Sstevel@tonic-gate 888ae115bc7Smrj /* 889ae115bc7Smrj * The debug info page provides enough information to allow external 890ae115bc7Smrj * inspectors (e.g. when running under a hypervisor) to bootstrap 891ae115bc7Smrj * themselves into allowing full-blown kernel debugging. 892ae115bc7Smrj */ 893ae115bc7Smrj static void 894ae115bc7Smrj init_debug_info(void) 895ae115bc7Smrj { 896ae115bc7Smrj caddr_t mem; 897ae115bc7Smrj debug_info_t *di; 898ae115bc7Smrj 899ae115bc7Smrj #ifndef __lint 900ae115bc7Smrj ASSERT(sizeof (debug_info_t) < MMU_PAGESIZE); 901ae115bc7Smrj #endif 902ae115bc7Smrj 903ae115bc7Smrj mem = BOP_ALLOC(bootops, (caddr_t)DEBUG_INFO_VA, MMU_PAGESIZE, 904ae115bc7Smrj MMU_PAGESIZE); 905ae115bc7Smrj 906ae115bc7Smrj if (mem != (caddr_t)DEBUG_INFO_VA) 907ae115bc7Smrj panic("BOP_ALLOC() failed"); 908ae115bc7Smrj bzero(mem, MMU_PAGESIZE); 909ae115bc7Smrj 910ae115bc7Smrj di = (debug_info_t *)mem; 911ae115bc7Smrj 912ae115bc7Smrj di->di_magic = DEBUG_INFO_MAGIC; 913ae115bc7Smrj di->di_version = DEBUG_INFO_VERSION; 914843e1988Sjohnlev di->di_modules = (uintptr_t)&modules; 915843e1988Sjohnlev di->di_s_text = (uintptr_t)s_text; 916843e1988Sjohnlev di->di_e_text = (uintptr_t)e_text; 917843e1988Sjohnlev di->di_s_data = (uintptr_t)s_data; 918843e1988Sjohnlev di->di_e_data = (uintptr_t)e_data; 919843e1988Sjohnlev di->di_hat_htable_off = offsetof(hat_t, hat_htable); 920843e1988Sjohnlev di->di_ht_pfn_off = offsetof(htable_t, ht_pfn); 9217c478bd9Sstevel@tonic-gate } 9227c478bd9Sstevel@tonic-gate 9237c478bd9Sstevel@tonic-gate /* 924ae115bc7Smrj * Build the memlists and other kernel essential memory system data structures. 925ae115bc7Smrj * This is everything at valloc_base. 9267c478bd9Sstevel@tonic-gate */ 9277c478bd9Sstevel@tonic-gate static void 9287c478bd9Sstevel@tonic-gate startup_memlist(void) 9297c478bd9Sstevel@tonic-gate { 9307c478bd9Sstevel@tonic-gate size_t memlist_sz; 9317c478bd9Sstevel@tonic-gate size_t memseg_sz; 9327c478bd9Sstevel@tonic-gate size_t pagehash_sz; 9337c478bd9Sstevel@tonic-gate size_t pp_sz; 9347c478bd9Sstevel@tonic-gate uintptr_t va; 9357c478bd9Sstevel@tonic-gate size_t len; 9367c478bd9Sstevel@tonic-gate uint_t prot; 9377c478bd9Sstevel@tonic-gate pfn_t pfn; 9387c478bd9Sstevel@tonic-gate int memblocks; 9391de082f7SVikram Hegde pfn_t rsvd_high_pfn; 9401de082f7SVikram Hegde pgcnt_t rsvd_pgcnt; 9411de082f7SVikram Hegde size_t rsvdmemlist_sz; 9421de082f7SVikram Hegde int rsvdmemblocks; 9437c478bd9Sstevel@tonic-gate caddr_t pagecolor_mem; 9447c478bd9Sstevel@tonic-gate size_t pagecolor_memsz; 9457c478bd9Sstevel@tonic-gate caddr_t page_ctrs_mem; 9467c478bd9Sstevel@tonic-gate size_t page_ctrs_size; 947d7d93655Sblakej size_t pse_table_alloc_size; 9487c478bd9Sstevel@tonic-gate struct memlist *current; 9497c478bd9Sstevel@tonic-gate extern void startup_build_mem_nodes(struct memlist *); 9507c478bd9Sstevel@tonic-gate 9517c478bd9Sstevel@tonic-gate /* XX64 fix these - they should be in include files */ 9527c478bd9Sstevel@tonic-gate extern size_t page_coloring_init(uint_t, int, int); 9537c478bd9Sstevel@tonic-gate extern void page_coloring_setup(caddr_t); 9547c478bd9Sstevel@tonic-gate 9557c478bd9Sstevel@tonic-gate PRM_POINT("startup_memlist() starting..."); 9567c478bd9Sstevel@tonic-gate 9577c478bd9Sstevel@tonic-gate /* 9587c478bd9Sstevel@tonic-gate * Use leftover large page nucleus text/data space for loadable modules. 9597c478bd9Sstevel@tonic-gate * Use at most MODTEXT/MODDATA. 9607c478bd9Sstevel@tonic-gate */ 961ae115bc7Smrj len = kbm_nucleus_size; 962ae115bc7Smrj ASSERT(len > MMU_PAGESIZE); 9637c478bd9Sstevel@tonic-gate 964ae115bc7Smrj moddata = (caddr_t)ROUND_UP_PAGE(e_data); 965ae115bc7Smrj e_moddata = (caddr_t)P2ROUNDUP((uintptr_t)e_data, (uintptr_t)len); 966ae115bc7Smrj if (e_moddata - moddata > MODDATA) 967ae115bc7Smrj e_moddata = moddata + MODDATA; 9687c478bd9Sstevel@tonic-gate 969ae115bc7Smrj modtext = (caddr_t)ROUND_UP_PAGE(e_text); 970ae115bc7Smrj e_modtext = (caddr_t)P2ROUNDUP((uintptr_t)e_text, (uintptr_t)len); 971ae115bc7Smrj if (e_modtext - modtext > MODTEXT) 972ae115bc7Smrj e_modtext = modtext + MODTEXT; 9737c478bd9Sstevel@tonic-gate 9747c478bd9Sstevel@tonic-gate econtig = e_moddata; 9757c478bd9Sstevel@tonic-gate 9767c478bd9Sstevel@tonic-gate PRM_DEBUG(modtext); 9777c478bd9Sstevel@tonic-gate PRM_DEBUG(e_modtext); 9787c478bd9Sstevel@tonic-gate PRM_DEBUG(moddata); 9797c478bd9Sstevel@tonic-gate PRM_DEBUG(e_moddata); 9807c478bd9Sstevel@tonic-gate PRM_DEBUG(econtig); 9817c478bd9Sstevel@tonic-gate 9827c478bd9Sstevel@tonic-gate /* 983ae115bc7Smrj * Examine the boot loader physical memory map to find out: 9847c478bd9Sstevel@tonic-gate * - total memory in system - physinstalled 9857c478bd9Sstevel@tonic-gate * - the max physical address - physmax 986ae115bc7Smrj * - the number of discontiguous segments of memory. 9877c478bd9Sstevel@tonic-gate */ 9887c478bd9Sstevel@tonic-gate if (prom_debug) 989ae115bc7Smrj print_memlist("boot physinstalled", 9907c478bd9Sstevel@tonic-gate bootops->boot_mem->physinstalled); 991a3114836SGerry Liu installed_top_size_ex(bootops->boot_mem->physinstalled, &physmax, 9927c478bd9Sstevel@tonic-gate &physinstalled, &memblocks); 9937c478bd9Sstevel@tonic-gate PRM_DEBUG(physmax); 9947c478bd9Sstevel@tonic-gate PRM_DEBUG(physinstalled); 9957c478bd9Sstevel@tonic-gate PRM_DEBUG(memblocks); 9967c478bd9Sstevel@tonic-gate 997a3114836SGerry Liu /* 998a3114836SGerry Liu * Compute maximum physical address for memory DR operations. 999a3114836SGerry Liu * Memory DR operations are unsupported on xpv or 32bit OSes. 1000a3114836SGerry Liu */ 1001a3114836SGerry Liu #ifdef __amd64 1002a3114836SGerry Liu if (plat_dr_support_memory()) { 1003a3114836SGerry Liu if (plat_dr_physmax == 0) { 1004a3114836SGerry Liu uint_t pabits = UINT_MAX; 1005a3114836SGerry Liu 1006a3114836SGerry Liu cpuid_get_addrsize(CPU, &pabits, NULL); 1007a3114836SGerry Liu plat_dr_physmax = btop(1ULL << pabits); 1008a3114836SGerry Liu } 1009a3114836SGerry Liu if (plat_dr_physmax > PHYSMEM_MAX64) 1010a3114836SGerry Liu plat_dr_physmax = PHYSMEM_MAX64; 1011a3114836SGerry Liu } else 1012a3114836SGerry Liu #endif 1013a3114836SGerry Liu plat_dr_physmax = 0; 1014a3114836SGerry Liu 10151de082f7SVikram Hegde /* 10161de082f7SVikram Hegde * Examine the bios reserved memory to find out: 10171de082f7SVikram Hegde * - the number of discontiguous segments of memory. 10181de082f7SVikram Hegde */ 10191de082f7SVikram Hegde if (prom_debug) 10201de082f7SVikram Hegde print_memlist("boot reserved mem", 10211de082f7SVikram Hegde bootops->boot_mem->rsvdmem); 1022a3114836SGerry Liu installed_top_size_ex(bootops->boot_mem->rsvdmem, &rsvd_high_pfn, 10231de082f7SVikram Hegde &rsvd_pgcnt, &rsvdmemblocks); 10241de082f7SVikram Hegde PRM_DEBUG(rsvd_high_pfn); 10251de082f7SVikram Hegde PRM_DEBUG(rsvd_pgcnt); 10261de082f7SVikram Hegde PRM_DEBUG(rsvdmemblocks); 10271de082f7SVikram Hegde 10287c478bd9Sstevel@tonic-gate /* 10297c478bd9Sstevel@tonic-gate * Initialize hat's mmu parameters. 10307c478bd9Sstevel@tonic-gate * Check for enforce-prot-exec in boot environment. It's used to 10317c478bd9Sstevel@tonic-gate * enable/disable support for the page table entry NX bit. 10327c478bd9Sstevel@tonic-gate * The default is to enforce PROT_EXEC on processors that support NX. 10337c478bd9Sstevel@tonic-gate * Boot seems to round up the "len", but 8 seems to be big enough. 10347c478bd9Sstevel@tonic-gate */ 10357c478bd9Sstevel@tonic-gate mmu_init(); 10367c478bd9Sstevel@tonic-gate 10377c478bd9Sstevel@tonic-gate #ifdef __i386 10387c478bd9Sstevel@tonic-gate /* 10397c478bd9Sstevel@tonic-gate * physmax is lowered if there is more memory than can be 10407c478bd9Sstevel@tonic-gate * physically addressed in 32 bit (PAE/non-PAE) modes. 10417c478bd9Sstevel@tonic-gate */ 10427c478bd9Sstevel@tonic-gate if (mmu.pae_hat) { 10437c478bd9Sstevel@tonic-gate if (PFN_ABOVE64G(physmax)) { 10447c478bd9Sstevel@tonic-gate physinstalled -= (physmax - (PFN_64G - 1)); 10457c478bd9Sstevel@tonic-gate physmax = PFN_64G - 1; 10467c478bd9Sstevel@tonic-gate } 10477c478bd9Sstevel@tonic-gate } else { 10487c478bd9Sstevel@tonic-gate if (PFN_ABOVE4G(physmax)) { 10497c478bd9Sstevel@tonic-gate physinstalled -= (physmax - (PFN_4G - 1)); 10507c478bd9Sstevel@tonic-gate physmax = PFN_4G - 1; 10517c478bd9Sstevel@tonic-gate } 10527c478bd9Sstevel@tonic-gate } 10537c478bd9Sstevel@tonic-gate #endif 10547c478bd9Sstevel@tonic-gate 10557c478bd9Sstevel@tonic-gate startup_build_mem_nodes(bootops->boot_mem->physinstalled); 10567c478bd9Sstevel@tonic-gate 10577c478bd9Sstevel@tonic-gate if (BOP_GETPROPLEN(bootops, "enforce-prot-exec") >= 0) { 10587c478bd9Sstevel@tonic-gate int len = BOP_GETPROPLEN(bootops, "enforce-prot-exec"); 10597c478bd9Sstevel@tonic-gate char value[8]; 10607c478bd9Sstevel@tonic-gate 10617c478bd9Sstevel@tonic-gate if (len < 8) 10627c478bd9Sstevel@tonic-gate (void) BOP_GETPROP(bootops, "enforce-prot-exec", value); 10637c478bd9Sstevel@tonic-gate else 10647c478bd9Sstevel@tonic-gate (void) strcpy(value, ""); 10657c478bd9Sstevel@tonic-gate if (strcmp(value, "off") == 0) 10667c478bd9Sstevel@tonic-gate mmu.pt_nx = 0; 10677c478bd9Sstevel@tonic-gate } 10687c478bd9Sstevel@tonic-gate PRM_DEBUG(mmu.pt_nx); 10697c478bd9Sstevel@tonic-gate 10707c478bd9Sstevel@tonic-gate /* 10717c478bd9Sstevel@tonic-gate * We will need page_t's for every page in the system, except for 10727c478bd9Sstevel@tonic-gate * memory mapped at or above above the start of the kernel text segment. 10737c478bd9Sstevel@tonic-gate * 10747c478bd9Sstevel@tonic-gate * pages above e_modtext are attributed to kernel debugger (obp_pages) 10757c478bd9Sstevel@tonic-gate */ 10767c478bd9Sstevel@tonic-gate npages = physinstalled - 1; /* avail_filter() skips page 0, so "- 1" */ 10777c478bd9Sstevel@tonic-gate obp_pages = 0; 10787c478bd9Sstevel@tonic-gate va = KERNEL_TEXT; 1079ae115bc7Smrj while (kbm_probe(&va, &len, &pfn, &prot) != 0) { 10807c478bd9Sstevel@tonic-gate npages -= len >> MMU_PAGESHIFT; 10817c478bd9Sstevel@tonic-gate if (va >= (uintptr_t)e_moddata) 10827c478bd9Sstevel@tonic-gate obp_pages += len >> MMU_PAGESHIFT; 10837c478bd9Sstevel@tonic-gate va += len; 10847c478bd9Sstevel@tonic-gate } 10857c478bd9Sstevel@tonic-gate PRM_DEBUG(npages); 10867c478bd9Sstevel@tonic-gate PRM_DEBUG(obp_pages); 10877c478bd9Sstevel@tonic-gate 10887c478bd9Sstevel@tonic-gate /* 108935b1ab99Sjosephb * If physmem is patched to be non-zero, use it instead of the computed 109035b1ab99Sjosephb * value unless it is larger than the actual amount of memory on hand. 10917c478bd9Sstevel@tonic-gate */ 10924944b02eSkchow if (physmem == 0 || physmem > npages) { 10937c478bd9Sstevel@tonic-gate physmem = npages; 10944944b02eSkchow } else if (physmem < npages) { 1095f53ad214Skchow orig_npages = npages; 10967c478bd9Sstevel@tonic-gate npages = physmem; 10974944b02eSkchow } 10987c478bd9Sstevel@tonic-gate PRM_DEBUG(physmem); 10997c478bd9Sstevel@tonic-gate 11007c478bd9Sstevel@tonic-gate /* 11017c478bd9Sstevel@tonic-gate * We now compute the sizes of all the initial allocations for 11027c478bd9Sstevel@tonic-gate * structures the kernel needs in order do kmem_alloc(). These 11037c478bd9Sstevel@tonic-gate * include: 11047c478bd9Sstevel@tonic-gate * memsegs 11057c478bd9Sstevel@tonic-gate * memlists 11067c478bd9Sstevel@tonic-gate * page hash table 11077c478bd9Sstevel@tonic-gate * page_t's 11087c478bd9Sstevel@tonic-gate * page coloring data structs 11097c478bd9Sstevel@tonic-gate */ 11107c478bd9Sstevel@tonic-gate memseg_sz = sizeof (struct memseg) * (memblocks + POSS_NEW_FRAGMENTS); 11117c478bd9Sstevel@tonic-gate ADD_TO_ALLOCATIONS(memseg_base, memseg_sz); 11127c478bd9Sstevel@tonic-gate PRM_DEBUG(memseg_sz); 11137c478bd9Sstevel@tonic-gate 11147c478bd9Sstevel@tonic-gate /* 1115ae115bc7Smrj * Reserve space for memlists. There's no real good way to know exactly 1116ae115bc7Smrj * how much room we'll need, but this should be a good upper bound. 11177c478bd9Sstevel@tonic-gate */ 11187c478bd9Sstevel@tonic-gate memlist_sz = ROUND_UP_PAGE(2 * sizeof (struct memlist) * 11197c478bd9Sstevel@tonic-gate (memblocks + POSS_NEW_FRAGMENTS)); 11207c478bd9Sstevel@tonic-gate ADD_TO_ALLOCATIONS(memlist, memlist_sz); 11217c478bd9Sstevel@tonic-gate PRM_DEBUG(memlist_sz); 11227c478bd9Sstevel@tonic-gate 11231de082f7SVikram Hegde /* 11241de082f7SVikram Hegde * Reserve space for bios reserved memlists. 11251de082f7SVikram Hegde */ 11261de082f7SVikram Hegde rsvdmemlist_sz = ROUND_UP_PAGE(2 * sizeof (struct memlist) * 11271de082f7SVikram Hegde (rsvdmemblocks + POSS_NEW_FRAGMENTS)); 11281de082f7SVikram Hegde ADD_TO_ALLOCATIONS(bios_rsvd, rsvdmemlist_sz); 11291de082f7SVikram Hegde PRM_DEBUG(rsvdmemlist_sz); 11301de082f7SVikram Hegde 1131cb15d5d9SPeter Rival /* LINTED */ 1132cb15d5d9SPeter Rival ASSERT(P2SAMEHIGHBIT((1 << PP_SHIFT), sizeof (struct page))); 11337c478bd9Sstevel@tonic-gate /* 11347c478bd9Sstevel@tonic-gate * The page structure hash table size is a power of 2 11357c478bd9Sstevel@tonic-gate * such that the average hash chain length is PAGE_HASHAVELEN. 11367c478bd9Sstevel@tonic-gate */ 11377c478bd9Sstevel@tonic-gate page_hashsz = npages / PAGE_HASHAVELEN; 1138cb15d5d9SPeter Rival page_hashsz_shift = highbit(page_hashsz); 1139cb15d5d9SPeter Rival page_hashsz = 1 << page_hashsz_shift; 11407c478bd9Sstevel@tonic-gate pagehash_sz = sizeof (struct page *) * page_hashsz; 11417c478bd9Sstevel@tonic-gate ADD_TO_ALLOCATIONS(page_hash, pagehash_sz); 11427c478bd9Sstevel@tonic-gate PRM_DEBUG(pagehash_sz); 11437c478bd9Sstevel@tonic-gate 11447c478bd9Sstevel@tonic-gate /* 1145ae115bc7Smrj * Set aside room for the page structures themselves. 11467c478bd9Sstevel@tonic-gate */ 1147ae115bc7Smrj PRM_DEBUG(npages); 1148ae115bc7Smrj pp_sz = sizeof (struct page) * npages; 11497c478bd9Sstevel@tonic-gate ADD_TO_ALLOCATIONS(pp_base, pp_sz); 11507c478bd9Sstevel@tonic-gate PRM_DEBUG(pp_sz); 11517c478bd9Sstevel@tonic-gate 11527c478bd9Sstevel@tonic-gate /* 11537c478bd9Sstevel@tonic-gate * determine l2 cache info and memory size for page coloring 11547c478bd9Sstevel@tonic-gate */ 11557c478bd9Sstevel@tonic-gate (void) getl2cacheinfo(CPU, 11567c478bd9Sstevel@tonic-gate &l2cache_sz, &l2cache_linesz, &l2cache_assoc); 11577c478bd9Sstevel@tonic-gate pagecolor_memsz = 11587c478bd9Sstevel@tonic-gate page_coloring_init(l2cache_sz, l2cache_linesz, l2cache_assoc); 11597c478bd9Sstevel@tonic-gate ADD_TO_ALLOCATIONS(pagecolor_mem, pagecolor_memsz); 11607c478bd9Sstevel@tonic-gate PRM_DEBUG(pagecolor_memsz); 11617c478bd9Sstevel@tonic-gate 11627c478bd9Sstevel@tonic-gate page_ctrs_size = page_ctrs_sz(); 11637c478bd9Sstevel@tonic-gate ADD_TO_ALLOCATIONS(page_ctrs_mem, page_ctrs_size); 11647c478bd9Sstevel@tonic-gate PRM_DEBUG(page_ctrs_size); 11657c478bd9Sstevel@tonic-gate 1166d7d93655Sblakej /* 1167d7d93655Sblakej * Allocate the array that protects pp->p_selock. 1168d7d93655Sblakej */ 1169d7d93655Sblakej pse_shift = size_pse_array(physmem, max_ncpus); 1170d7d93655Sblakej pse_table_size = 1 << pse_shift; 1171d7d93655Sblakej pse_table_alloc_size = pse_table_size * sizeof (pad_mutex_t); 1172d7d93655Sblakej ADD_TO_ALLOCATIONS(pse_mutex, pse_table_alloc_size); 1173d7d93655Sblakej 1174ae115bc7Smrj #if defined(__amd64) 11757c478bd9Sstevel@tonic-gate valloc_sz = ROUND_UP_LPAGE(valloc_sz); 1176ae115bc7Smrj valloc_base = VALLOC_BASE; 117735b1ab99Sjosephb 117835b1ab99Sjosephb /* 117935b1ab99Sjosephb * The default values of VALLOC_BASE and SEGKPM_BASE should work 118035b1ab99Sjosephb * for values of physmax up to 1 Terabyte. They need adjusting when 11816c8c89eaSKit Chow * memory is at addresses above 1 TB. When adjusted, segkpm_base must 11826c8c89eaSKit Chow * be aligned on KERNEL_REDZONE_SIZE boundary (span of top level pte). 118335b1ab99Sjosephb */ 1184a3114836SGerry Liu if (physmax + 1 > mmu_btop(TERABYTE) || 1185a3114836SGerry Liu plat_dr_physmax > mmu_btop(TERABYTE)) { 118635b1ab99Sjosephb uint64_t kpm_resv_amount = mmu_ptob(physmax + 1); 118735b1ab99Sjosephb 1188a3114836SGerry Liu if (kpm_resv_amount < mmu_ptob(plat_dr_physmax)) { 1189a3114836SGerry Liu kpm_resv_amount = mmu_ptob(plat_dr_physmax); 1190a3114836SGerry Liu } 1191a3114836SGerry Liu 11926c8c89eaSKit Chow segkpm_base = -(P2ROUNDUP((2 * kpm_resv_amount), 11936c8c89eaSKit Chow KERNEL_REDZONE_SIZE)); /* down from top VA */ 119435b1ab99Sjosephb 119535b1ab99Sjosephb /* make sure we leave some space for user apps above hole */ 119635b1ab99Sjosephb segkpm_base = MAX(segkpm_base, AMD64_VA_HOLE_END + TERABYTE); 119735b1ab99Sjosephb if (segkpm_base > SEGKPM_BASE) 119835b1ab99Sjosephb segkpm_base = SEGKPM_BASE; 119935b1ab99Sjosephb PRM_DEBUG(segkpm_base); 120035b1ab99Sjosephb 12016c8c89eaSKit Chow valloc_base = segkpm_base + P2ROUNDUP(kpm_resv_amount, ONE_GIG); 12026c8c89eaSKit Chow if (valloc_base < segkpm_base) 12036c8c89eaSKit Chow panic("not enough kernel VA to support memory size"); 120435b1ab99Sjosephb PRM_DEBUG(valloc_base); 120535b1ab99Sjosephb } 1206ae115bc7Smrj #else /* __i386 */ 1207ae115bc7Smrj valloc_base = (uintptr_t)(MISC_VA_BASE - valloc_sz); 1208ae115bc7Smrj valloc_base = P2ALIGN(valloc_base, mmu.level_size[1]); 12097c478bd9Sstevel@tonic-gate PRM_DEBUG(valloc_base); 121035b1ab99Sjosephb #endif /* __i386 */ 1211ae115bc7Smrj 1212ae115bc7Smrj /* 1213ae115bc7Smrj * do all the initial allocations 1214ae115bc7Smrj */ 1215ae115bc7Smrj perform_allocations(); 1216ae115bc7Smrj 1217ae115bc7Smrj /* 1218ae115bc7Smrj * Build phys_install and phys_avail in kernel memspace. 1219ae115bc7Smrj * - phys_install should be all memory in the system. 1220ae115bc7Smrj * - phys_avail is phys_install minus any memory mapped before this 1221ae115bc7Smrj * point above KERNEL_TEXT. 1222ae115bc7Smrj */ 1223ae115bc7Smrj current = phys_install = memlist; 1224ae115bc7Smrj copy_memlist_filter(bootops->boot_mem->physinstalled, ¤t, NULL); 1225ae115bc7Smrj if ((caddr_t)current > (caddr_t)memlist + memlist_sz) 1226ae115bc7Smrj panic("physinstalled was too big!"); 1227ae115bc7Smrj if (prom_debug) 1228ae115bc7Smrj print_memlist("phys_install", phys_install); 1229ae115bc7Smrj 1230ae115bc7Smrj phys_avail = current; 1231ae115bc7Smrj PRM_POINT("Building phys_avail:\n"); 1232ae115bc7Smrj copy_memlist_filter(bootops->boot_mem->physinstalled, ¤t, 1233ae115bc7Smrj avail_filter); 1234ae115bc7Smrj if ((caddr_t)current > (caddr_t)memlist + memlist_sz) 1235ae115bc7Smrj panic("physavail was too big!"); 1236ae115bc7Smrj if (prom_debug) 1237ae115bc7Smrj print_memlist("phys_avail", phys_avail); 1238a3114836SGerry Liu #ifndef __xpv 1239a3114836SGerry Liu /* 1240a3114836SGerry Liu * Free unused memlist items, which may be used by memory DR driver 1241a3114836SGerry Liu * at runtime. 1242a3114836SGerry Liu */ 1243a3114836SGerry Liu if ((caddr_t)current < (caddr_t)memlist + memlist_sz) { 1244a3114836SGerry Liu memlist_free_block((caddr_t)current, 1245a3114836SGerry Liu (caddr_t)memlist + memlist_sz - (caddr_t)current); 1246a3114836SGerry Liu } 1247a3114836SGerry Liu #endif 1248ae115bc7Smrj 12491de082f7SVikram Hegde /* 12501de082f7SVikram Hegde * Build bios reserved memspace 12511de082f7SVikram Hegde */ 12521de082f7SVikram Hegde current = bios_rsvd; 12531de082f7SVikram Hegde copy_memlist_filter(bootops->boot_mem->rsvdmem, ¤t, NULL); 12541de082f7SVikram Hegde if ((caddr_t)current > (caddr_t)bios_rsvd + rsvdmemlist_sz) 12551de082f7SVikram Hegde panic("bios_rsvd was too big!"); 12561de082f7SVikram Hegde if (prom_debug) 12571de082f7SVikram Hegde print_memlist("bios_rsvd", bios_rsvd); 1258a3114836SGerry Liu #ifndef __xpv 1259a3114836SGerry Liu /* 1260a3114836SGerry Liu * Free unused memlist items, which may be used by memory DR driver 1261a3114836SGerry Liu * at runtime. 1262a3114836SGerry Liu */ 1263a3114836SGerry Liu if ((caddr_t)current < (caddr_t)bios_rsvd + rsvdmemlist_sz) { 1264a3114836SGerry Liu memlist_free_block((caddr_t)current, 1265a3114836SGerry Liu (caddr_t)bios_rsvd + rsvdmemlist_sz - (caddr_t)current); 1266a3114836SGerry Liu } 1267a3114836SGerry Liu #endif 12681de082f7SVikram Hegde 1269ae115bc7Smrj /* 1270ae115bc7Smrj * setup page coloring 1271ae115bc7Smrj */ 1272ae115bc7Smrj page_coloring_setup(pagecolor_mem); 1273ae115bc7Smrj page_lock_init(); /* currently a no-op */ 1274ae115bc7Smrj 1275ae115bc7Smrj /* 1276ae115bc7Smrj * free page list counters 1277ae115bc7Smrj */ 1278ae115bc7Smrj (void) page_ctrs_alloc(page_ctrs_mem); 1279ae115bc7Smrj 128006fb6a36Sdv /* 128106fb6a36Sdv * Size the pcf array based on the number of cpus in the box at 128206fb6a36Sdv * boot time. 128306fb6a36Sdv */ 128406fb6a36Sdv 128506fb6a36Sdv pcf_init(); 128606fb6a36Sdv 1287ae115bc7Smrj /* 1288ae115bc7Smrj * Initialize the page structures from the memory lists. 1289ae115bc7Smrj */ 1290ae115bc7Smrj availrmem_initial = availrmem = freemem = 0; 1291ae115bc7Smrj PRM_POINT("Calling kphysm_init()..."); 1292ae115bc7Smrj npages = kphysm_init(pp_base, npages); 1293ae115bc7Smrj PRM_POINT("kphysm_init() done"); 1294ae115bc7Smrj PRM_DEBUG(npages); 1295ae115bc7Smrj 1296ae115bc7Smrj init_debug_info(); 1297ae115bc7Smrj 1298ae115bc7Smrj /* 1299ae115bc7Smrj * Now that page_t's have been initialized, remove all the 1300ae115bc7Smrj * initial allocation pages from the kernel free page lists. 1301ae115bc7Smrj */ 1302ae115bc7Smrj boot_mapin((caddr_t)valloc_base, valloc_sz); 13030cfdb603Sjosephb boot_mapin((caddr_t)MISC_VA_BASE, MISC_VA_SIZE); 1304ae115bc7Smrj PRM_POINT("startup_memlist() done"); 1305ae115bc7Smrj 1306ae115bc7Smrj PRM_DEBUG(valloc_sz); 13072cb27123Saguzovsk 1308567d55e1Saguzovsk #if defined(__amd64) 1309567d55e1Saguzovsk if ((availrmem >> (30 - MMU_PAGESHIFT)) >= 1310567d55e1Saguzovsk textrepl_min_gb && l2cache_sz <= 2 << 20) { 1311567d55e1Saguzovsk extern size_t textrepl_size_thresh; 13122cb27123Saguzovsk textrepl_size_thresh = (16 << 20) - 1; 13132cb27123Saguzovsk } 1314567d55e1Saguzovsk #endif 1315ae115bc7Smrj } 1316ae115bc7Smrj 1317ae115bc7Smrj /* 1318ae115bc7Smrj * Layout the kernel's part of address space and initialize kmem allocator. 1319ae115bc7Smrj */ 1320ae115bc7Smrj static void 1321ae115bc7Smrj startup_kmem(void) 1322ae115bc7Smrj { 1323932dc8e5Sdp extern void page_set_colorequiv_arr(void); 1324932dc8e5Sdp 1325ae115bc7Smrj PRM_POINT("startup_kmem() starting..."); 13267c478bd9Sstevel@tonic-gate 13277c478bd9Sstevel@tonic-gate #if defined(__amd64) 13287c478bd9Sstevel@tonic-gate if (eprom_kernelbase && eprom_kernelbase != KERNELBASE) 13297c478bd9Sstevel@tonic-gate cmn_err(CE_NOTE, "!kernelbase cannot be changed on 64-bit " 13307c478bd9Sstevel@tonic-gate "systems."); 133135b1ab99Sjosephb kernelbase = segkpm_base - KERNEL_REDZONE_SIZE; 13327c478bd9Sstevel@tonic-gate core_base = (uintptr_t)COREHEAP_BASE; 1333ae115bc7Smrj core_size = (size_t)MISC_VA_BASE - COREHEAP_BASE; 13347c478bd9Sstevel@tonic-gate #else /* __i386 */ 13357c478bd9Sstevel@tonic-gate /* 13367c478bd9Sstevel@tonic-gate * We configure kernelbase based on: 13377c478bd9Sstevel@tonic-gate * 13387c478bd9Sstevel@tonic-gate * 1. user specified kernelbase via eeprom command. Value cannot exceed 13397c478bd9Sstevel@tonic-gate * KERNELBASE_MAX. we large page align eprom_kernelbase 13407c478bd9Sstevel@tonic-gate * 13417c478bd9Sstevel@tonic-gate * 2. Default to KERNELBASE and adjust to 2X less the size for page_t. 13427c478bd9Sstevel@tonic-gate * On large memory systems we must lower kernelbase to allow 13437c478bd9Sstevel@tonic-gate * enough room for page_t's for all of memory. 13447c478bd9Sstevel@tonic-gate * 13457c478bd9Sstevel@tonic-gate * The value set here, might be changed a little later. 13467c478bd9Sstevel@tonic-gate */ 13477c478bd9Sstevel@tonic-gate if (eprom_kernelbase) { 13487c478bd9Sstevel@tonic-gate kernelbase = eprom_kernelbase & mmu.level_mask[1]; 13497c478bd9Sstevel@tonic-gate if (kernelbase > KERNELBASE_MAX) 13507c478bd9Sstevel@tonic-gate kernelbase = KERNELBASE_MAX; 13517c478bd9Sstevel@tonic-gate } else { 13527c478bd9Sstevel@tonic-gate kernelbase = (uintptr_t)KERNELBASE; 13537c478bd9Sstevel@tonic-gate kernelbase -= ROUND_UP_4MEG(2 * valloc_sz); 13547c478bd9Sstevel@tonic-gate } 13557c478bd9Sstevel@tonic-gate ASSERT((kernelbase & mmu.level_offset[1]) == 0); 1356ae115bc7Smrj core_base = valloc_base; 13577c478bd9Sstevel@tonic-gate core_size = 0; 1358ae115bc7Smrj #endif /* __i386 */ 13597c478bd9Sstevel@tonic-gate 13607c478bd9Sstevel@tonic-gate PRM_DEBUG(core_base); 13617c478bd9Sstevel@tonic-gate PRM_DEBUG(core_size); 1362ae115bc7Smrj PRM_DEBUG(kernelbase); 13637c478bd9Sstevel@tonic-gate 136435b1ab99Sjosephb #if defined(__i386) 13657c478bd9Sstevel@tonic-gate segkp_fromheap = 1; 1366ae115bc7Smrj #endif /* __i386 */ 136735b1ab99Sjosephb 13687c478bd9Sstevel@tonic-gate ekernelheap = (char *)core_base; 1369ae115bc7Smrj PRM_DEBUG(ekernelheap); 13707c478bd9Sstevel@tonic-gate 13717c478bd9Sstevel@tonic-gate /* 13727c478bd9Sstevel@tonic-gate * Now that we know the real value of kernelbase, 13737c478bd9Sstevel@tonic-gate * update variables that were initialized with a value of 13747c478bd9Sstevel@tonic-gate * KERNELBASE (in common/conf/param.c). 13757c478bd9Sstevel@tonic-gate * 13767c478bd9Sstevel@tonic-gate * XXX The problem with this sort of hackery is that the 13777c478bd9Sstevel@tonic-gate * compiler just may feel like putting the const declarations 13787c478bd9Sstevel@tonic-gate * (in param.c) into the .text section. Perhaps they should 13797c478bd9Sstevel@tonic-gate * just be declared as variables there? 13807c478bd9Sstevel@tonic-gate */ 13817c478bd9Sstevel@tonic-gate 13827c478bd9Sstevel@tonic-gate *(uintptr_t *)&_kernelbase = kernelbase; 13837c478bd9Sstevel@tonic-gate *(uintptr_t *)&_userlimit = kernelbase; 1384ccbaea4fSjosephb #if defined(__amd64) 1385ccbaea4fSjosephb *(uintptr_t *)&_userlimit -= KERNELBASE - USERLIMIT; 1386ccbaea4fSjosephb #else 13877c478bd9Sstevel@tonic-gate *(uintptr_t *)&_userlimit32 = _userlimit; 13887c478bd9Sstevel@tonic-gate #endif 13897c478bd9Sstevel@tonic-gate PRM_DEBUG(_kernelbase); 13907c478bd9Sstevel@tonic-gate PRM_DEBUG(_userlimit); 13917c478bd9Sstevel@tonic-gate PRM_DEBUG(_userlimit32); 13927c478bd9Sstevel@tonic-gate 139335b1ab99Sjosephb layout_kernel_va(); 139435b1ab99Sjosephb 139535b1ab99Sjosephb #if defined(__i386) 139635b1ab99Sjosephb /* 139735b1ab99Sjosephb * If segmap is too large we can push the bottom of the kernel heap 139835b1ab99Sjosephb * higher than the base. Or worse, it could exceed the top of the 139935b1ab99Sjosephb * VA space entirely, causing it to wrap around. 140035b1ab99Sjosephb */ 140135b1ab99Sjosephb if (kernelheap >= ekernelheap || (uintptr_t)kernelheap < kernelbase) 140235b1ab99Sjosephb panic("too little address space available for kernelheap," 140335b1ab99Sjosephb " use eeprom for lower kernelbase or smaller segmapsize"); 140435b1ab99Sjosephb #endif /* __i386 */ 140535b1ab99Sjosephb 14067c478bd9Sstevel@tonic-gate /* 14077c478bd9Sstevel@tonic-gate * Initialize the kernel heap. Note 3rd argument must be > 1st. 14087c478bd9Sstevel@tonic-gate */ 140935b1ab99Sjosephb kernelheap_init(kernelheap, ekernelheap, 141035b1ab99Sjosephb kernelheap + MMU_PAGESIZE, 1411ae115bc7Smrj (void *)core_base, (void *)(core_base + core_size)); 14127c478bd9Sstevel@tonic-gate 1413843e1988Sjohnlev #if defined(__xpv) 1414843e1988Sjohnlev /* 1415843e1988Sjohnlev * Link pending events struct into cpu struct 1416843e1988Sjohnlev */ 1417843e1988Sjohnlev CPU->cpu_m.mcpu_evt_pend = &cpu0_evt_data; 1418843e1988Sjohnlev #endif 14197c478bd9Sstevel@tonic-gate /* 14207c478bd9Sstevel@tonic-gate * Initialize kernel memory allocator. 14217c478bd9Sstevel@tonic-gate */ 14227c478bd9Sstevel@tonic-gate kmem_init(); 14237c478bd9Sstevel@tonic-gate 1424932dc8e5Sdp /* 1425932dc8e5Sdp * Factor in colorequiv to check additional 'equivalent' bins 1426932dc8e5Sdp */ 1427932dc8e5Sdp page_set_colorequiv_arr(); 1428932dc8e5Sdp 14297c478bd9Sstevel@tonic-gate /* 14307c478bd9Sstevel@tonic-gate * print this out early so that we know what's going on 14317c478bd9Sstevel@tonic-gate */ 14327417cfdeSKuriakose Kuruvilla print_x86_featureset(x86_featureset); 14337c478bd9Sstevel@tonic-gate 14347c478bd9Sstevel@tonic-gate /* 14357c478bd9Sstevel@tonic-gate * Initialize bp_mapin(). 14367c478bd9Sstevel@tonic-gate */ 14377c478bd9Sstevel@tonic-gate bp_init(MMU_PAGESIZE, HAT_STORECACHING_OK); 14387c478bd9Sstevel@tonic-gate 1439f53ad214Skchow /* 1440f53ad214Skchow * orig_npages is non-zero if physmem has been configured for less 1441f53ad214Skchow * than the available memory. 1442f53ad214Skchow */ 1443f53ad214Skchow if (orig_npages) { 144435b1ab99Sjosephb cmn_err(CE_WARN, "!%slimiting physmem to 0x%lx of 0x%lx pages", 144535b1ab99Sjosephb (npages == PHYSMEM ? "Due to virtual address space " : ""), 144635b1ab99Sjosephb npages, orig_npages); 1447f53ad214Skchow } 14487c478bd9Sstevel@tonic-gate #if defined(__i386) 14497c478bd9Sstevel@tonic-gate if (eprom_kernelbase && (eprom_kernelbase != kernelbase)) 14507c478bd9Sstevel@tonic-gate cmn_err(CE_WARN, "kernelbase value, User specified 0x%lx, " 14517c478bd9Sstevel@tonic-gate "System using 0x%lx", 14527c478bd9Sstevel@tonic-gate (uintptr_t)eprom_kernelbase, (uintptr_t)kernelbase); 14537c478bd9Sstevel@tonic-gate #endif 14547c478bd9Sstevel@tonic-gate 14557c478bd9Sstevel@tonic-gate #ifdef KERNELBASE_ABI_MIN 14567c478bd9Sstevel@tonic-gate if (kernelbase < (uintptr_t)KERNELBASE_ABI_MIN) { 14577c478bd9Sstevel@tonic-gate cmn_err(CE_NOTE, "!kernelbase set to 0x%lx, system is not " 14587c478bd9Sstevel@tonic-gate "i386 ABI compliant.", (uintptr_t)kernelbase); 14597c478bd9Sstevel@tonic-gate } 14607c478bd9Sstevel@tonic-gate #endif 14617c478bd9Sstevel@tonic-gate 1462a3114836SGerry Liu #ifndef __xpv 1463a3114836SGerry Liu if (plat_dr_support_memory()) { 1464a3114836SGerry Liu mem_config_init(); 1465a3114836SGerry Liu } 1466a3114836SGerry Liu #else /* __xpv */ 1467843e1988Sjohnlev /* 1468843e1988Sjohnlev * Some of the xen start information has to be relocated up 1469843e1988Sjohnlev * into the kernel's permanent address space. 1470843e1988Sjohnlev */ 1471843e1988Sjohnlev PRM_POINT("calling xen_relocate_start_info()"); 1472843e1988Sjohnlev xen_relocate_start_info(); 1473843e1988Sjohnlev PRM_POINT("xen_relocate_start_info() done"); 1474843e1988Sjohnlev 1475843e1988Sjohnlev /* 1476843e1988Sjohnlev * (Update the vcpu pointer in our cpu structure to point into 1477843e1988Sjohnlev * the relocated shared info.) 1478843e1988Sjohnlev */ 1479843e1988Sjohnlev CPU->cpu_m.mcpu_vcpu_info = 1480843e1988Sjohnlev &HYPERVISOR_shared_info->vcpu_info[CPU->cpu_id]; 1481a3114836SGerry Liu #endif /* __xpv */ 1482843e1988Sjohnlev 1483ae115bc7Smrj PRM_POINT("startup_kmem() done"); 14847c478bd9Sstevel@tonic-gate } 14857c478bd9Sstevel@tonic-gate 1486551bc2a6Smrj #ifndef __xpv 1487551bc2a6Smrj /* 1488551bc2a6Smrj * If we have detected that we are running in an HVM environment, we need 1489551bc2a6Smrj * to prepend the PV driver directory to the module search path. 1490551bc2a6Smrj */ 1491551bc2a6Smrj #define HVM_MOD_DIR "/platform/i86hvm/kernel" 1492551bc2a6Smrj static void 1493551bc2a6Smrj update_default_path() 1494551bc2a6Smrj { 1495551bc2a6Smrj char *current, *newpath; 1496551bc2a6Smrj int newlen; 1497551bc2a6Smrj 1498551bc2a6Smrj /* 1499551bc2a6Smrj * We are about to resync with krtld. krtld will reset its 1500551bc2a6Smrj * internal module search path iff Solaris has set default_path. 1501551bc2a6Smrj * We want to be sure we're prepending this new directory to the 1502551bc2a6Smrj * right search path. 1503551bc2a6Smrj */ 1504551bc2a6Smrj current = (default_path == NULL) ? kobj_module_path : default_path; 1505551bc2a6Smrj 1506b6611e3bSStuart Maybee newlen = strlen(HVM_MOD_DIR) + strlen(current) + 2; 1507551bc2a6Smrj newpath = kmem_alloc(newlen, KM_SLEEP); 1508551bc2a6Smrj (void) strcpy(newpath, HVM_MOD_DIR); 1509551bc2a6Smrj (void) strcat(newpath, " "); 1510551bc2a6Smrj (void) strcat(newpath, current); 1511551bc2a6Smrj 1512551bc2a6Smrj default_path = newpath; 1513551bc2a6Smrj } 1514551bc2a6Smrj #endif 1515551bc2a6Smrj 15167c478bd9Sstevel@tonic-gate static void 15177c478bd9Sstevel@tonic-gate startup_modules(void) 15187c478bd9Sstevel@tonic-gate { 1519ed5289f9SKen Erickson int cnt; 15207c478bd9Sstevel@tonic-gate extern void prom_setup(void); 1521ed5289f9SKen Erickson int32_t v, h; 1522ed5289f9SKen Erickson char d[11]; 1523ed5289f9SKen Erickson char *cp; 1524e4b86885SCheng Sean Ye cmi_hdl_t hdl; 15257c478bd9Sstevel@tonic-gate 15267c478bd9Sstevel@tonic-gate PRM_POINT("startup_modules() starting..."); 1527843e1988Sjohnlev 1528843e1988Sjohnlev #ifndef __xpv 15297c478bd9Sstevel@tonic-gate /* 15307c478bd9Sstevel@tonic-gate * Initialize ten-micro second timer so that drivers will 15317c478bd9Sstevel@tonic-gate * not get short changed in their init phase. This was 15327c478bd9Sstevel@tonic-gate * not getting called until clkinit which, on fast cpu's 15337c478bd9Sstevel@tonic-gate * caused the drv_usecwait to be way too short. 15347c478bd9Sstevel@tonic-gate */ 15357c478bd9Sstevel@tonic-gate microfind(); 1536551bc2a6Smrj 1537b9bfdccdSStuart Maybee if (get_hwenv() == HW_XEN_HVM) 1538551bc2a6Smrj update_default_path(); 1539843e1988Sjohnlev #endif 15407c478bd9Sstevel@tonic-gate 15417c478bd9Sstevel@tonic-gate /* 15427c478bd9Sstevel@tonic-gate * Read the GMT lag from /etc/rtc_config. 15437c478bd9Sstevel@tonic-gate */ 1544ae115bc7Smrj sgmtl(process_rtc_config_file()); 15457c478bd9Sstevel@tonic-gate 15467c478bd9Sstevel@tonic-gate /* 15477c478bd9Sstevel@tonic-gate * Calculate default settings of system parameters based upon 15487c478bd9Sstevel@tonic-gate * maxusers, yet allow to be overridden via the /etc/system file. 15497c478bd9Sstevel@tonic-gate */ 15507c478bd9Sstevel@tonic-gate param_calc(0); 15517c478bd9Sstevel@tonic-gate 15527c478bd9Sstevel@tonic-gate mod_setup(); 15537c478bd9Sstevel@tonic-gate 15547c478bd9Sstevel@tonic-gate /* 15557c478bd9Sstevel@tonic-gate * Initialize system parameters. 15567c478bd9Sstevel@tonic-gate */ 15577c478bd9Sstevel@tonic-gate param_init(); 15587c478bd9Sstevel@tonic-gate 15599acbbeafSnn /* 15609acbbeafSnn * Initialize the default brands 15619acbbeafSnn */ 15629acbbeafSnn brand_init(); 15639acbbeafSnn 15647c478bd9Sstevel@tonic-gate /* 15657c478bd9Sstevel@tonic-gate * maxmem is the amount of physical memory we're playing with. 15667c478bd9Sstevel@tonic-gate */ 15677c478bd9Sstevel@tonic-gate maxmem = physmem; 15687c478bd9Sstevel@tonic-gate 15697c478bd9Sstevel@tonic-gate /* 15707c478bd9Sstevel@tonic-gate * Initialize segment management stuff. 15717c478bd9Sstevel@tonic-gate */ 15727c478bd9Sstevel@tonic-gate seg_init(); 15737c478bd9Sstevel@tonic-gate 15747c478bd9Sstevel@tonic-gate if (modload("fs", "specfs") == -1) 15757c478bd9Sstevel@tonic-gate halt("Can't load specfs"); 15767c478bd9Sstevel@tonic-gate 15777c478bd9Sstevel@tonic-gate if (modload("fs", "devfs") == -1) 15787c478bd9Sstevel@tonic-gate halt("Can't load devfs"); 15797c478bd9Sstevel@tonic-gate 1580facf4a8dSllai if (modload("fs", "dev") == -1) 1581facf4a8dSllai halt("Can't load dev"); 1582facf4a8dSllai 158335a5a358SJonathan Adams if (modload("fs", "procfs") == -1) 158435a5a358SJonathan Adams halt("Can't load procfs"); 158535a5a358SJonathan Adams 158645916cd2Sjpk (void) modloadonly("sys", "lbl_edition"); 158745916cd2Sjpk 15887c478bd9Sstevel@tonic-gate dispinit(); 15897c478bd9Sstevel@tonic-gate 15907c478bd9Sstevel@tonic-gate /* Read cluster configuration data. */ 15917c478bd9Sstevel@tonic-gate clconf_init(); 15927c478bd9Sstevel@tonic-gate 1593843e1988Sjohnlev #if defined(__xpv) 1594349b53ddSStuart Maybee (void) ec_init(); 1595843e1988Sjohnlev gnttab_init(); 1596843e1988Sjohnlev (void) xs_early_init(); 1597843e1988Sjohnlev #endif /* __xpv */ 1598843e1988Sjohnlev 15997c478bd9Sstevel@tonic-gate /* 16007c478bd9Sstevel@tonic-gate * Create a kernel device tree. First, create rootnex and 16017c478bd9Sstevel@tonic-gate * then invoke bus specific code to probe devices. 16027c478bd9Sstevel@tonic-gate */ 16037c478bd9Sstevel@tonic-gate setup_ddi(); 16047aec1d6eScindi 1605074bb90dSTom Pothier #ifdef __xpv 1606074bb90dSTom Pothier if (DOMAIN_IS_INITDOMAIN(xen_info)) 1607074bb90dSTom Pothier #endif 1608074bb90dSTom Pothier { 1609074bb90dSTom Pothier /* 1610074bb90dSTom Pothier * Load the System Management BIOS into the global ksmbios 1611074bb90dSTom Pothier * handle, if an SMBIOS is present on this system. 1612074bb90dSTom Pothier */ 1613074bb90dSTom Pothier ksmbios = smbios_open(NULL, SMB_VERSION, ksmbios_flags, NULL); 1614074bb90dSTom Pothier } 1615074bb90dSTom Pothier 1616074bb90dSTom Pothier 1617*9e3e4df2SGarrett D'Amore /* 1618*9e3e4df2SGarrett D'Amore * Originally clconf_init() apparently needed the hostid. But 1619*9e3e4df2SGarrett D'Amore * this no longer appears to be true - it uses its own nodeid. 1620*9e3e4df2SGarrett D'Amore * By placing the hostid logic here, we are able to make use of 1621*9e3e4df2SGarrett D'Amore * the SMBIOS UUID. 1622*9e3e4df2SGarrett D'Amore */ 1623*9e3e4df2SGarrett D'Amore if ((h = set_soft_hostid()) == HW_INVALID_HOSTID) { 1624*9e3e4df2SGarrett D'Amore cmn_err(CE_WARN, "Unable to set hostid"); 1625*9e3e4df2SGarrett D'Amore } else { 1626*9e3e4df2SGarrett D'Amore for (v = h, cnt = 0; cnt < 10; cnt++) { 1627*9e3e4df2SGarrett D'Amore d[cnt] = (char)(v % 10); 1628*9e3e4df2SGarrett D'Amore v /= 10; 1629*9e3e4df2SGarrett D'Amore if (v == 0) 1630*9e3e4df2SGarrett D'Amore break; 1631*9e3e4df2SGarrett D'Amore } 1632*9e3e4df2SGarrett D'Amore for (cp = hw_serial; cnt >= 0; cnt--) 1633*9e3e4df2SGarrett D'Amore *cp++ = d[cnt] + '0'; 1634*9e3e4df2SGarrett D'Amore *cp = 0; 1635*9e3e4df2SGarrett D'Amore } 1636*9e3e4df2SGarrett D'Amore 1637e4b86885SCheng Sean Ye /* 1638e4b86885SCheng Sean Ye * Set up the CPU module subsystem for the boot cpu in the native 1639e4b86885SCheng Sean Ye * case, and all physical cpu resource in the xpv dom0 case. 1640e4b86885SCheng Sean Ye * Modifies the device tree, so this must be done after 1641e4b86885SCheng Sean Ye * setup_ddi(). 1642e4b86885SCheng Sean Ye */ 1643e4b86885SCheng Sean Ye #ifdef __xpv 1644e4b86885SCheng Sean Ye /* 1645e4b86885SCheng Sean Ye * If paravirtualized and on dom0 then we initialize all physical 1646e4b86885SCheng Sean Ye * cpu handles now; if paravirtualized on a domU then do not 1647e4b86885SCheng Sean Ye * initialize. 1648e4b86885SCheng Sean Ye */ 1649e4b86885SCheng Sean Ye if (DOMAIN_IS_INITDOMAIN(xen_info)) { 1650e4b86885SCheng Sean Ye xen_mc_lcpu_cookie_t cpi; 1651e4b86885SCheng Sean Ye 1652e4b86885SCheng Sean Ye for (cpi = xen_physcpu_next(NULL); cpi != NULL; 1653e4b86885SCheng Sean Ye cpi = xen_physcpu_next(cpi)) { 1654e4b86885SCheng Sean Ye if ((hdl = cmi_init(CMI_HDL_SOLARIS_xVM_MCA, 1655e4b86885SCheng Sean Ye xen_physcpu_chipid(cpi), xen_physcpu_coreid(cpi), 1656e4b86885SCheng Sean Ye xen_physcpu_strandid(cpi))) != NULL && 16577417cfdeSKuriakose Kuruvilla is_x86_feature(x86_featureset, X86FSET_MCA)) 165820c794b3Sgavinm cmi_mca_init(hdl); 165920c794b3Sgavinm } 166020c794b3Sgavinm } 1661e4b86885SCheng Sean Ye #else 1662e4b86885SCheng Sean Ye /* 1663e4b86885SCheng Sean Ye * Initialize a handle for the boot cpu - others will initialize 1664e4b86885SCheng Sean Ye * as they startup. Do not do this if we know we are in an HVM domU. 1665e4b86885SCheng Sean Ye */ 1666b9bfdccdSStuart Maybee if ((get_hwenv() != HW_XEN_HVM) && 1667e4b86885SCheng Sean Ye (hdl = cmi_init(CMI_HDL_NATIVE, cmi_ntv_hwchipid(CPU), 1668e4b86885SCheng Sean Ye cmi_ntv_hwcoreid(CPU), cmi_ntv_hwstrandid(CPU))) != NULL && 16697417cfdeSKuriakose Kuruvilla is_x86_feature(x86_featureset, X86FSET_MCA)) { 1670e4b86885SCheng Sean Ye cmi_mca_init(hdl); 1671a3114836SGerry Liu CPU->cpu_m.mcpu_cmi_hdl = hdl; 1672a3114836SGerry Liu } 167320c794b3Sgavinm #endif /* __xpv */ 16747aec1d6eScindi 16757c478bd9Sstevel@tonic-gate /* 16767c478bd9Sstevel@tonic-gate * Fake a prom tree such that /dev/openprom continues to work 16777c478bd9Sstevel@tonic-gate */ 1678ae115bc7Smrj PRM_POINT("startup_modules: calling prom_setup..."); 16797c478bd9Sstevel@tonic-gate prom_setup(); 1680ae115bc7Smrj PRM_POINT("startup_modules: done"); 16817c478bd9Sstevel@tonic-gate 16827c478bd9Sstevel@tonic-gate /* 16837c478bd9Sstevel@tonic-gate * Load all platform specific modules 16847c478bd9Sstevel@tonic-gate */ 1685ae115bc7Smrj PRM_POINT("startup_modules: calling psm_modload..."); 16867c478bd9Sstevel@tonic-gate psm_modload(); 16877c478bd9Sstevel@tonic-gate 16887c478bd9Sstevel@tonic-gate PRM_POINT("startup_modules() done"); 16897c478bd9Sstevel@tonic-gate } 16907c478bd9Sstevel@tonic-gate 1691ae115bc7Smrj /* 1692ae115bc7Smrj * claim a "setaside" boot page for use in the kernel 1693ae115bc7Smrj */ 1694ae115bc7Smrj page_t * 1695ae115bc7Smrj boot_claim_page(pfn_t pfn) 16967c478bd9Sstevel@tonic-gate { 1697ae115bc7Smrj page_t *pp; 16987c478bd9Sstevel@tonic-gate 1699ae115bc7Smrj pp = page_numtopp_nolock(pfn); 1700ae115bc7Smrj ASSERT(pp != NULL); 17017c478bd9Sstevel@tonic-gate 1702ae115bc7Smrj if (PP_ISBOOTPAGES(pp)) { 1703ae115bc7Smrj if (pp->p_next != NULL) 1704ae115bc7Smrj pp->p_next->p_prev = pp->p_prev; 1705ae115bc7Smrj if (pp->p_prev == NULL) 1706ae115bc7Smrj bootpages = pp->p_next; 1707ae115bc7Smrj else 1708ae115bc7Smrj pp->p_prev->p_next = pp->p_next; 1709ae115bc7Smrj } else { 1710ae115bc7Smrj /* 1711ae115bc7Smrj * htable_attach() expects a base pagesize page 1712ae115bc7Smrj */ 1713ae115bc7Smrj if (pp->p_szc != 0) 1714ae115bc7Smrj page_boot_demote(pp); 1715ae115bc7Smrj pp = page_numtopp(pfn, SE_EXCL); 1716ae115bc7Smrj } 1717ae115bc7Smrj return (pp); 17187c478bd9Sstevel@tonic-gate } 17197c478bd9Sstevel@tonic-gate 17207c478bd9Sstevel@tonic-gate /* 17217c478bd9Sstevel@tonic-gate * Walk through the pagetables looking for pages mapped in by boot. If the 17227c478bd9Sstevel@tonic-gate * setaside flag is set the pages are expected to be returned to the 17237c478bd9Sstevel@tonic-gate * kernel later in boot, so we add them to the bootpages list. 17247c478bd9Sstevel@tonic-gate */ 17257c478bd9Sstevel@tonic-gate static void 17267c478bd9Sstevel@tonic-gate protect_boot_range(uintptr_t low, uintptr_t high, int setaside) 17277c478bd9Sstevel@tonic-gate { 17287c478bd9Sstevel@tonic-gate uintptr_t va = low; 17297c478bd9Sstevel@tonic-gate size_t len; 17307c478bd9Sstevel@tonic-gate uint_t prot; 17317c478bd9Sstevel@tonic-gate pfn_t pfn; 17327c478bd9Sstevel@tonic-gate page_t *pp; 17337c478bd9Sstevel@tonic-gate pgcnt_t boot_protect_cnt = 0; 17347c478bd9Sstevel@tonic-gate 1735ae115bc7Smrj while (kbm_probe(&va, &len, &pfn, &prot) != 0 && va < high) { 17367c478bd9Sstevel@tonic-gate if (va + len >= high) 17377c478bd9Sstevel@tonic-gate panic("0x%lx byte mapping at 0x%p exceeds boot's " 17387c478bd9Sstevel@tonic-gate "legal range.", len, (void *)va); 17397c478bd9Sstevel@tonic-gate 17407c478bd9Sstevel@tonic-gate while (len > 0) { 17417c478bd9Sstevel@tonic-gate pp = page_numtopp_alloc(pfn); 17427c478bd9Sstevel@tonic-gate if (pp != NULL) { 17437c478bd9Sstevel@tonic-gate if (setaside == 0) 17447c478bd9Sstevel@tonic-gate panic("Unexpected mapping by boot. " 17457c478bd9Sstevel@tonic-gate "addr=%p pfn=%lx\n", 17467c478bd9Sstevel@tonic-gate (void *)va, pfn); 17477c478bd9Sstevel@tonic-gate 17487c478bd9Sstevel@tonic-gate pp->p_next = bootpages; 1749ae115bc7Smrj pp->p_prev = NULL; 1750ae115bc7Smrj PP_SETBOOTPAGES(pp); 1751ae115bc7Smrj if (bootpages != NULL) { 1752ae115bc7Smrj bootpages->p_prev = pp; 1753ae115bc7Smrj } 17547c478bd9Sstevel@tonic-gate bootpages = pp; 17557c478bd9Sstevel@tonic-gate ++boot_protect_cnt; 17567c478bd9Sstevel@tonic-gate } 17577c478bd9Sstevel@tonic-gate 17587c478bd9Sstevel@tonic-gate ++pfn; 17597c478bd9Sstevel@tonic-gate len -= MMU_PAGESIZE; 17607c478bd9Sstevel@tonic-gate va += MMU_PAGESIZE; 17617c478bd9Sstevel@tonic-gate } 17627c478bd9Sstevel@tonic-gate } 17637c478bd9Sstevel@tonic-gate PRM_DEBUG(boot_protect_cnt); 17647c478bd9Sstevel@tonic-gate } 17657c478bd9Sstevel@tonic-gate 1766ae115bc7Smrj /* 176735b1ab99Sjosephb * 1768ae115bc7Smrj */ 17697c478bd9Sstevel@tonic-gate static void 177035b1ab99Sjosephb layout_kernel_va(void) 17717c478bd9Sstevel@tonic-gate { 177235b1ab99Sjosephb PRM_POINT("layout_kernel_va() starting..."); 17737c478bd9Sstevel@tonic-gate /* 1774ae115bc7Smrj * Establish the final size of the kernel's heap, size of segmap, 1775ae115bc7Smrj * segkp, etc. 17767c478bd9Sstevel@tonic-gate */ 17777c478bd9Sstevel@tonic-gate 17787c478bd9Sstevel@tonic-gate #if defined(__amd64) 17797c478bd9Sstevel@tonic-gate 178035b1ab99Sjosephb kpm_vbase = (caddr_t)segkpm_base; 1781a3114836SGerry Liu if (physmax + 1 < plat_dr_physmax) { 1782a3114836SGerry Liu kpm_size = ROUND_UP_LPAGE(mmu_ptob(plat_dr_physmax)); 1783a3114836SGerry Liu } else { 1784a3114836SGerry Liu kpm_size = ROUND_UP_LPAGE(mmu_ptob(physmax + 1)); 1785a3114836SGerry Liu } 178635b1ab99Sjosephb if ((uintptr_t)kpm_vbase + kpm_size > (uintptr_t)valloc_base) 178735b1ab99Sjosephb panic("not enough room for kpm!"); 1788ae115bc7Smrj PRM_DEBUG(kpm_size); 1789ae115bc7Smrj PRM_DEBUG(kpm_vbase); 17907c478bd9Sstevel@tonic-gate 17917c478bd9Sstevel@tonic-gate /* 1792ae115bc7Smrj * By default we create a seg_kp in 64 bit kernels, it's a little 1793ae115bc7Smrj * faster to access than embedding it in the heap. 17947c478bd9Sstevel@tonic-gate */ 1795ae115bc7Smrj segkp_base = (caddr_t)valloc_base + valloc_sz; 17967c478bd9Sstevel@tonic-gate if (!segkp_fromheap) { 17977c478bd9Sstevel@tonic-gate size_t sz = mmu_ptob(segkpsize); 17987c478bd9Sstevel@tonic-gate 17997c478bd9Sstevel@tonic-gate /* 1800ae115bc7Smrj * determine size of segkp 18017c478bd9Sstevel@tonic-gate */ 18027c478bd9Sstevel@tonic-gate if (sz < SEGKPMINSIZE || sz > SEGKPMAXSIZE) { 18037c478bd9Sstevel@tonic-gate sz = SEGKPDEFSIZE; 18047c478bd9Sstevel@tonic-gate cmn_err(CE_WARN, "!Illegal value for segkpsize. " 18057c478bd9Sstevel@tonic-gate "segkpsize has been reset to %ld pages", 18067c478bd9Sstevel@tonic-gate mmu_btop(sz)); 18077c478bd9Sstevel@tonic-gate } 18087c478bd9Sstevel@tonic-gate sz = MIN(sz, MAX(SEGKPMINSIZE, mmu_ptob(physmem))); 18097c478bd9Sstevel@tonic-gate 18107c478bd9Sstevel@tonic-gate segkpsize = mmu_btop(ROUND_UP_LPAGE(sz)); 18117c478bd9Sstevel@tonic-gate } 1812ae115bc7Smrj PRM_DEBUG(segkp_base); 1813ae115bc7Smrj PRM_DEBUG(segkpsize); 18147c478bd9Sstevel@tonic-gate 181535b1ab99Sjosephb /* 181635b1ab99Sjosephb * segzio is used for ZFS cached data. It uses a distinct VA 181735b1ab99Sjosephb * segment (from kernel heap) so that we can easily tell not to 181835b1ab99Sjosephb * include it in kernel crash dumps on 64 bit kernels. The trick is 181935b1ab99Sjosephb * to give it lots of VA, but not constrain the kernel heap. 182035b1ab99Sjosephb * We scale the size of segzio linearly with physmem up to 182135b1ab99Sjosephb * SEGZIOMAXSIZE. Above that amount it scales at 50% of physmem. 182235b1ab99Sjosephb */ 1823ae115bc7Smrj segzio_base = segkp_base + mmu_ptob(segkpsize); 1824ae115bc7Smrj if (segzio_fromheap) { 1825ae115bc7Smrj segziosize = 0; 1826ae115bc7Smrj } else { 182735b1ab99Sjosephb size_t physmem_size = mmu_ptob(physmem); 182835b1ab99Sjosephb size_t size = (segziosize == 0) ? 182935b1ab99Sjosephb physmem_size : mmu_ptob(segziosize); 1830ad23a2dbSjohansen 183135b1ab99Sjosephb if (size < SEGZIOMINSIZE) 1832ad23a2dbSjohansen size = SEGZIOMINSIZE; 183335b1ab99Sjosephb if (size > SEGZIOMAXSIZE) { 1834a778305aSjohansen size = SEGZIOMAXSIZE; 183535b1ab99Sjosephb if (physmem_size > size) 183635b1ab99Sjosephb size += (physmem_size - size) / 2; 1837ad23a2dbSjohansen } 1838ad23a2dbSjohansen segziosize = mmu_btop(ROUND_UP_LPAGE(size)); 1839ad23a2dbSjohansen } 1840ae115bc7Smrj PRM_DEBUG(segziosize); 1841ae115bc7Smrj PRM_DEBUG(segzio_base); 1842ad23a2dbSjohansen 18437c478bd9Sstevel@tonic-gate /* 1844ae115bc7Smrj * Put the range of VA for device mappings next, kmdb knows to not 1845ae115bc7Smrj * grep in this range of addresses. 18467c478bd9Sstevel@tonic-gate */ 1847ae115bc7Smrj toxic_addr = 1848ae115bc7Smrj ROUND_UP_LPAGE((uintptr_t)segzio_base + mmu_ptob(segziosize)); 18497c478bd9Sstevel@tonic-gate PRM_DEBUG(toxic_addr); 1850ae115bc7Smrj segmap_start = ROUND_UP_LPAGE(toxic_addr + toxic_size); 1851ae115bc7Smrj #else /* __i386 */ 1852ae115bc7Smrj segmap_start = ROUND_UP_LPAGE(kernelbase); 1853ae115bc7Smrj #endif /* __i386 */ 1854ae115bc7Smrj PRM_DEBUG(segmap_start); 18557c478bd9Sstevel@tonic-gate 18567c478bd9Sstevel@tonic-gate /* 185752eef812SAmrita Sadhukhan * Users can change segmapsize through eeprom. If the variable 185852eef812SAmrita Sadhukhan * is tuned through eeprom, there is no upper bound on the 1859ed5289f9SKen Erickson * size of segmap. 18607c478bd9Sstevel@tonic-gate */ 18617c478bd9Sstevel@tonic-gate segmapsize = MAX(ROUND_UP_LPAGE(segmapsize), SEGMAPDEFAULT); 18627c478bd9Sstevel@tonic-gate 18637c478bd9Sstevel@tonic-gate #if defined(__i386) 18647c478bd9Sstevel@tonic-gate /* 18657c478bd9Sstevel@tonic-gate * 32-bit systems don't have segkpm or segkp, so segmap appears at 18667c478bd9Sstevel@tonic-gate * the bottom of the kernel's address range. Set aside space for a 1867ae115bc7Smrj * small red zone just below the start of segmap. 18687c478bd9Sstevel@tonic-gate */ 1869ae115bc7Smrj segmap_start += KERNEL_REDZONE_SIZE; 18707c478bd9Sstevel@tonic-gate segmapsize -= KERNEL_REDZONE_SIZE; 18717c478bd9Sstevel@tonic-gate #endif 18727c478bd9Sstevel@tonic-gate 1873ae115bc7Smrj PRM_DEBUG(segmap_start); 18747c478bd9Sstevel@tonic-gate PRM_DEBUG(segmapsize); 187535b1ab99Sjosephb kernelheap = (caddr_t)ROUND_UP_LPAGE(segmap_start + segmapsize); 187635b1ab99Sjosephb PRM_DEBUG(kernelheap); 187735b1ab99Sjosephb PRM_POINT("layout_kernel_va() done..."); 187835b1ab99Sjosephb } 187935b1ab99Sjosephb 188035b1ab99Sjosephb /* 188135b1ab99Sjosephb * Finish initializing the VM system, now that we are no longer 188235b1ab99Sjosephb * relying on the boot time memory allocators. 188335b1ab99Sjosephb */ 188435b1ab99Sjosephb static void 188535b1ab99Sjosephb startup_vm(void) 188635b1ab99Sjosephb { 188735b1ab99Sjosephb struct segmap_crargs a; 188835b1ab99Sjosephb 188935b1ab99Sjosephb extern int use_brk_lpg, use_stk_lpg; 189035b1ab99Sjosephb 189135b1ab99Sjosephb PRM_POINT("startup_vm() starting..."); 189235b1ab99Sjosephb 189335b1ab99Sjosephb /* 189435b1ab99Sjosephb * Initialize the hat layer. 189535b1ab99Sjosephb */ 189635b1ab99Sjosephb hat_init(); 18977c478bd9Sstevel@tonic-gate 1898ae115bc7Smrj /* 1899ae115bc7Smrj * Do final allocations of HAT data structures that need to 1900ae115bc7Smrj * be allocated before quiescing the boot loader. 1901ae115bc7Smrj */ 1902ae115bc7Smrj PRM_POINT("Calling hat_kern_alloc()..."); 1903ae115bc7Smrj hat_kern_alloc((caddr_t)segmap_start, segmapsize, ekernelheap); 1904ae115bc7Smrj PRM_POINT("hat_kern_alloc() done"); 1905ae115bc7Smrj 1906843e1988Sjohnlev #ifndef __xpv 1907ae115bc7Smrj /* 19081d03c31eSjohnlev * Setup Page Attribute Table 1909ae115bc7Smrj */ 19101d03c31eSjohnlev pat_sync(); 1911843e1988Sjohnlev #endif 1912ae115bc7Smrj 1913ae115bc7Smrj /* 1914ae115bc7Smrj * The next two loops are done in distinct steps in order 1915ae115bc7Smrj * to be sure that any page that is doubly mapped (both above 1916ae115bc7Smrj * KERNEL_TEXT and below kernelbase) is dealt with correctly. 1917ae115bc7Smrj * Note this may never happen, but it might someday. 1918ae115bc7Smrj */ 1919ae115bc7Smrj bootpages = NULL; 1920ae115bc7Smrj PRM_POINT("Protecting boot pages"); 1921ae115bc7Smrj 1922ae115bc7Smrj /* 1923ae115bc7Smrj * Protect any pages mapped above KERNEL_TEXT that somehow have 1924ae115bc7Smrj * page_t's. This can only happen if something weird allocated 1925ae115bc7Smrj * in this range (like kadb/kmdb). 1926ae115bc7Smrj */ 1927ae115bc7Smrj protect_boot_range(KERNEL_TEXT, (uintptr_t)-1, 0); 1928ae115bc7Smrj 1929ae115bc7Smrj /* 1930ae115bc7Smrj * Before we can take over memory allocation/mapping from the boot 1931ae115bc7Smrj * loader we must remove from our free page lists any boot allocated 1932ae115bc7Smrj * pages that stay mapped until release_bootstrap(). 1933ae115bc7Smrj */ 1934ae115bc7Smrj protect_boot_range(0, kernelbase, 1); 1935ae115bc7Smrj 1936843e1988Sjohnlev 1937ae115bc7Smrj /* 1938ae115bc7Smrj * Switch to running on regular HAT (not boot_mmu) 1939ae115bc7Smrj */ 1940ae115bc7Smrj PRM_POINT("Calling hat_kern_setup()..."); 1941ae115bc7Smrj hat_kern_setup(); 1942ae115bc7Smrj 1943ae115bc7Smrj /* 1944ae115bc7Smrj * It is no longer safe to call BOP_ALLOC(), so make sure we don't. 1945ae115bc7Smrj */ 1946ae115bc7Smrj bop_no_more_mem(); 1947ae115bc7Smrj 1948ae115bc7Smrj PRM_POINT("hat_kern_setup() done"); 1949ae115bc7Smrj 1950ae115bc7Smrj hat_cpu_online(CPU); 1951ae115bc7Smrj 19527c478bd9Sstevel@tonic-gate /* 19537c478bd9Sstevel@tonic-gate * Initialize VM system 19547c478bd9Sstevel@tonic-gate */ 19557c478bd9Sstevel@tonic-gate PRM_POINT("Calling kvm_init()..."); 19567c478bd9Sstevel@tonic-gate kvm_init(); 19577c478bd9Sstevel@tonic-gate PRM_POINT("kvm_init() done"); 19587c478bd9Sstevel@tonic-gate 19597c478bd9Sstevel@tonic-gate /* 19607c478bd9Sstevel@tonic-gate * Tell kmdb that the VM system is now working 19617c478bd9Sstevel@tonic-gate */ 19627c478bd9Sstevel@tonic-gate if (boothowto & RB_DEBUG) 19637c478bd9Sstevel@tonic-gate kdi_dvec_vmready(); 19647c478bd9Sstevel@tonic-gate 1965843e1988Sjohnlev #if defined(__xpv) 1966843e1988Sjohnlev /* 1967843e1988Sjohnlev * Populate the I/O pool on domain 0 1968843e1988Sjohnlev */ 1969843e1988Sjohnlev if (DOMAIN_IS_INITDOMAIN(xen_info)) { 1970843e1988Sjohnlev extern long populate_io_pool(void); 1971843e1988Sjohnlev long init_io_pool_cnt; 1972843e1988Sjohnlev 1973843e1988Sjohnlev PRM_POINT("Populating reserve I/O page pool"); 1974843e1988Sjohnlev init_io_pool_cnt = populate_io_pool(); 1975843e1988Sjohnlev PRM_DEBUG(init_io_pool_cnt); 1976843e1988Sjohnlev } 1977843e1988Sjohnlev #endif 19787c478bd9Sstevel@tonic-gate /* 19797c478bd9Sstevel@tonic-gate * Mangle the brand string etc. 19807c478bd9Sstevel@tonic-gate */ 19817c478bd9Sstevel@tonic-gate cpuid_pass3(CPU); 19827c478bd9Sstevel@tonic-gate 19837c478bd9Sstevel@tonic-gate #if defined(__amd64) 19847c478bd9Sstevel@tonic-gate 19857c478bd9Sstevel@tonic-gate /* 19867c478bd9Sstevel@tonic-gate * Create the device arena for toxic (to dtrace/kmdb) mappings. 19877c478bd9Sstevel@tonic-gate */ 19887c478bd9Sstevel@tonic-gate device_arena = vmem_create("device", (void *)toxic_addr, 19897c478bd9Sstevel@tonic-gate toxic_size, MMU_PAGESIZE, NULL, NULL, NULL, 0, VM_SLEEP); 19907c478bd9Sstevel@tonic-gate 19917c478bd9Sstevel@tonic-gate #else /* __i386 */ 19927c478bd9Sstevel@tonic-gate 19937c478bd9Sstevel@tonic-gate /* 19947c478bd9Sstevel@tonic-gate * allocate the bit map that tracks toxic pages 19957c478bd9Sstevel@tonic-gate */ 1996ae115bc7Smrj toxic_bit_map_len = btop((ulong_t)(valloc_base - kernelbase)); 19977c478bd9Sstevel@tonic-gate PRM_DEBUG(toxic_bit_map_len); 19987c478bd9Sstevel@tonic-gate toxic_bit_map = 19997c478bd9Sstevel@tonic-gate kmem_zalloc(BT_SIZEOFMAP(toxic_bit_map_len), KM_NOSLEEP); 20007c478bd9Sstevel@tonic-gate ASSERT(toxic_bit_map != NULL); 20017c478bd9Sstevel@tonic-gate PRM_DEBUG(toxic_bit_map); 20027c478bd9Sstevel@tonic-gate 20037c478bd9Sstevel@tonic-gate #endif /* __i386 */ 20047c478bd9Sstevel@tonic-gate 20057c478bd9Sstevel@tonic-gate 20067c478bd9Sstevel@tonic-gate /* 20077c478bd9Sstevel@tonic-gate * Now that we've got more VA, as well as the ability to allocate from 20087c478bd9Sstevel@tonic-gate * it, tell the debugger. 20097c478bd9Sstevel@tonic-gate */ 20107c478bd9Sstevel@tonic-gate if (boothowto & RB_DEBUG) 20117c478bd9Sstevel@tonic-gate kdi_dvec_memavail(); 20127c478bd9Sstevel@tonic-gate 20137c478bd9Sstevel@tonic-gate /* 20147c478bd9Sstevel@tonic-gate * The following code installs a special page fault handler (#pf) 20157c478bd9Sstevel@tonic-gate * to work around a pentium bug. 20167c478bd9Sstevel@tonic-gate */ 2017843e1988Sjohnlev #if !defined(__amd64) && !defined(__xpv) 20187c478bd9Sstevel@tonic-gate if (x86_type == X86_TYPE_P5) { 2019ae115bc7Smrj desctbr_t idtr; 20207c478bd9Sstevel@tonic-gate gate_desc_t *newidt; 20217c478bd9Sstevel@tonic-gate 20227c478bd9Sstevel@tonic-gate if ((newidt = kmem_zalloc(MMU_PAGESIZE, KM_NOSLEEP)) == NULL) 20237c478bd9Sstevel@tonic-gate panic("failed to install pentium_pftrap"); 20247c478bd9Sstevel@tonic-gate 20250cfdb603Sjosephb bcopy(idt0, newidt, NIDT * sizeof (*idt0)); 20267c478bd9Sstevel@tonic-gate set_gatesegd(&newidt[T_PGFLT], &pentium_pftrap, 20279844da31SSeth Goldberg KCS_SEL, SDT_SYSIGT, TRP_KPL, 0); 20287c478bd9Sstevel@tonic-gate 20297c478bd9Sstevel@tonic-gate (void) as_setprot(&kas, (caddr_t)newidt, MMU_PAGESIZE, 20300cfdb603Sjosephb PROT_READ | PROT_EXEC); 20317c478bd9Sstevel@tonic-gate 20320cfdb603Sjosephb CPU->cpu_idt = newidt; 20330cfdb603Sjosephb idtr.dtr_base = (uintptr_t)CPU->cpu_idt; 20340cfdb603Sjosephb idtr.dtr_limit = (NIDT * sizeof (*idt0)) - 1; 2035ae115bc7Smrj wr_idtr(&idtr); 20367c478bd9Sstevel@tonic-gate } 20377c478bd9Sstevel@tonic-gate #endif /* !__amd64 */ 20387c478bd9Sstevel@tonic-gate 2039843e1988Sjohnlev #if !defined(__xpv) 20407c478bd9Sstevel@tonic-gate /* 20417c478bd9Sstevel@tonic-gate * Map page pfn=0 for drivers, such as kd, that need to pick up 20427c478bd9Sstevel@tonic-gate * parameters left there by controllers/BIOS. 20437c478bd9Sstevel@tonic-gate */ 20447c478bd9Sstevel@tonic-gate PRM_POINT("setup up p0_va"); 20457c478bd9Sstevel@tonic-gate p0_va = i86devmap(0, 1, PROT_READ); 20467c478bd9Sstevel@tonic-gate PRM_DEBUG(p0_va); 2047843e1988Sjohnlev #endif 20487c478bd9Sstevel@tonic-gate 20497c478bd9Sstevel@tonic-gate cmn_err(CE_CONT, "?mem = %luK (0x%lx)\n", 20507c478bd9Sstevel@tonic-gate physinstalled << (MMU_PAGESHIFT - 10), ptob(physinstalled)); 20517c478bd9Sstevel@tonic-gate 205283f9b804Skchow /* 205383f9b804Skchow * disable automatic large pages for small memory systems or 205483f9b804Skchow * when the disable flag is set. 205502bc52beSkchow * 205602bc52beSkchow * Do not yet consider page sizes larger than 2m/4m. 205783f9b804Skchow */ 2058ec25b48fSsusans if (!auto_lpg_disable && mmu.max_page_level > 0) { 2059ec25b48fSsusans max_uheap_lpsize = LEVEL_SIZE(1); 2060ec25b48fSsusans max_ustack_lpsize = LEVEL_SIZE(1); 2061ec25b48fSsusans max_privmap_lpsize = LEVEL_SIZE(1); 2062ec25b48fSsusans max_uidata_lpsize = LEVEL_SIZE(1); 2063ec25b48fSsusans max_utext_lpsize = LEVEL_SIZE(1); 2064ec25b48fSsusans max_shm_lpsize = LEVEL_SIZE(1); 2065ec25b48fSsusans } 2066ec25b48fSsusans if (physmem < privm_lpg_min_physmem || mmu.max_page_level == 0 || 2067ec25b48fSsusans auto_lpg_disable) { 2068beb1bda0Sdavemq use_brk_lpg = 0; 2069beb1bda0Sdavemq use_stk_lpg = 0; 2070ec25b48fSsusans } 207102bc52beSkchow mcntl0_lpsize = LEVEL_SIZE(mmu.umax_page_level); 2072beb1bda0Sdavemq 20737c478bd9Sstevel@tonic-gate PRM_POINT("Calling hat_init_finish()..."); 20747c478bd9Sstevel@tonic-gate hat_init_finish(); 20757c478bd9Sstevel@tonic-gate PRM_POINT("hat_init_finish() done"); 20767c478bd9Sstevel@tonic-gate 20777c478bd9Sstevel@tonic-gate /* 20787c478bd9Sstevel@tonic-gate * Initialize the segkp segment type. 20797c478bd9Sstevel@tonic-gate */ 20807c478bd9Sstevel@tonic-gate rw_enter(&kas.a_lock, RW_WRITER); 2081ae115bc7Smrj PRM_POINT("Attaching segkp"); 2082ae115bc7Smrj if (segkp_fromheap) { 20837c478bd9Sstevel@tonic-gate segkp->s_as = &kas; 2084ae115bc7Smrj } else if (seg_attach(&kas, (caddr_t)segkp_base, mmu_ptob(segkpsize), 2085ae115bc7Smrj segkp) < 0) { 2086ae115bc7Smrj panic("startup: cannot attach segkp"); 2087ae115bc7Smrj /*NOTREACHED*/ 20887c478bd9Sstevel@tonic-gate } 2089ae115bc7Smrj PRM_POINT("Doing segkp_create()"); 20907c478bd9Sstevel@tonic-gate if (segkp_create(segkp) != 0) { 20917c478bd9Sstevel@tonic-gate panic("startup: segkp_create failed"); 20927c478bd9Sstevel@tonic-gate /*NOTREACHED*/ 20937c478bd9Sstevel@tonic-gate } 20947c478bd9Sstevel@tonic-gate PRM_DEBUG(segkp); 20957c478bd9Sstevel@tonic-gate rw_exit(&kas.a_lock); 20967c478bd9Sstevel@tonic-gate 20977c478bd9Sstevel@tonic-gate /* 20987c478bd9Sstevel@tonic-gate * kpm segment 20997c478bd9Sstevel@tonic-gate */ 21007c478bd9Sstevel@tonic-gate segmap_kpm = 0; 21017c478bd9Sstevel@tonic-gate if (kpm_desired) { 21027c478bd9Sstevel@tonic-gate kpm_init(); 21037c478bd9Sstevel@tonic-gate kpm_enable = 1; 21047c478bd9Sstevel@tonic-gate } 21057c478bd9Sstevel@tonic-gate 21067c478bd9Sstevel@tonic-gate /* 21077c478bd9Sstevel@tonic-gate * Now create segmap segment. 21087c478bd9Sstevel@tonic-gate */ 21097c478bd9Sstevel@tonic-gate rw_enter(&kas.a_lock, RW_WRITER); 2110ae115bc7Smrj if (seg_attach(&kas, (caddr_t)segmap_start, segmapsize, segmap) < 0) { 2111ae115bc7Smrj panic("cannot attach segmap"); 21127c478bd9Sstevel@tonic-gate /*NOTREACHED*/ 21137c478bd9Sstevel@tonic-gate } 2114ae115bc7Smrj PRM_DEBUG(segmap); 21157c478bd9Sstevel@tonic-gate 21167c478bd9Sstevel@tonic-gate a.prot = PROT_READ | PROT_WRITE; 21177c478bd9Sstevel@tonic-gate a.shmsize = 0; 21187c478bd9Sstevel@tonic-gate a.nfreelist = segmapfreelists; 21197c478bd9Sstevel@tonic-gate 2120ae115bc7Smrj if (segmap_create(segmap, (caddr_t)&a) != 0) 2121ae115bc7Smrj panic("segmap_create segmap"); 21227c478bd9Sstevel@tonic-gate rw_exit(&kas.a_lock); 21237c478bd9Sstevel@tonic-gate 21247c478bd9Sstevel@tonic-gate setup_vaddr_for_ppcopy(CPU); 21257c478bd9Sstevel@tonic-gate 21267c478bd9Sstevel@tonic-gate segdev_init(); 2127843e1988Sjohnlev #if defined(__xpv) 2128843e1988Sjohnlev if (DOMAIN_IS_INITDOMAIN(xen_info)) 2129843e1988Sjohnlev #endif 2130843e1988Sjohnlev pmem_init(); 2131ae115bc7Smrj 21327c478bd9Sstevel@tonic-gate PRM_POINT("startup_vm() done"); 21337c478bd9Sstevel@tonic-gate } 21347c478bd9Sstevel@tonic-gate 2135ae115bc7Smrj /* 2136ae115bc7Smrj * Load a tod module for the non-standard tod part found on this system. 2137ae115bc7Smrj */ 2138ae115bc7Smrj static void 2139ae115bc7Smrj load_tod_module(char *todmod) 2140ae115bc7Smrj { 2141ae115bc7Smrj if (modload("tod", todmod) == -1) 2142ae115bc7Smrj halt("Can't load TOD module"); 2143ae115bc7Smrj } 2144ae115bc7Smrj 21457c478bd9Sstevel@tonic-gate static void 21467c478bd9Sstevel@tonic-gate startup_end(void) 21477c478bd9Sstevel@tonic-gate { 2148dd4eeefdSeota int i; 21497c478bd9Sstevel@tonic-gate extern void setx86isalist(void); 2150fb2caebeSRandy Fishel extern void cpu_event_init(void); 21517c478bd9Sstevel@tonic-gate 21527c478bd9Sstevel@tonic-gate PRM_POINT("startup_end() starting..."); 21537c478bd9Sstevel@tonic-gate 21547c478bd9Sstevel@tonic-gate /* 21557c478bd9Sstevel@tonic-gate * Perform tasks that get done after most of the VM 21567c478bd9Sstevel@tonic-gate * initialization has been done but before the clock 21577c478bd9Sstevel@tonic-gate * and other devices get started. 21587c478bd9Sstevel@tonic-gate */ 21597c478bd9Sstevel@tonic-gate kern_setup1(); 21607c478bd9Sstevel@tonic-gate 21617c478bd9Sstevel@tonic-gate /* 21627c478bd9Sstevel@tonic-gate * Perform CPC initialization for this CPU. 21637c478bd9Sstevel@tonic-gate */ 21647c478bd9Sstevel@tonic-gate kcpc_hw_init(CPU); 21657c478bd9Sstevel@tonic-gate 2166fb2caebeSRandy Fishel /* 2167fb2caebeSRandy Fishel * Initialize cpu event framework. 2168fb2caebeSRandy Fishel */ 2169fb2caebeSRandy Fishel cpu_event_init(); 2170fb2caebeSRandy Fishel 2171ee88d2b9Skchow #if defined(OPTERON_WORKAROUND_6323525) 2172ee88d2b9Skchow if (opteron_workaround_6323525) 2173ee88d2b9Skchow patch_workaround_6323525(); 2174ee88d2b9Skchow #endif 2175ae115bc7Smrj /* 2176ae115bc7Smrj * If needed, load TOD module now so that ddi_get_time(9F) etc. work 2177ae115bc7Smrj * (For now, "needed" is defined as set tod_module_name in /etc/system) 2178ae115bc7Smrj */ 2179ae115bc7Smrj if (tod_module_name != NULL) { 2180ae115bc7Smrj PRM_POINT("load_tod_module()"); 2181ae115bc7Smrj load_tod_module(tod_module_name); 2182ae115bc7Smrj } 2183ae115bc7Smrj 2184843e1988Sjohnlev #if defined(__xpv) 2185843e1988Sjohnlev /* 2186843e1988Sjohnlev * Forceload interposing TOD module for the hypervisor. 2187843e1988Sjohnlev */ 2188843e1988Sjohnlev PRM_POINT("load_tod_module()"); 2189843e1988Sjohnlev load_tod_module("xpvtod"); 2190843e1988Sjohnlev #endif 2191843e1988Sjohnlev 21927c478bd9Sstevel@tonic-gate /* 21937c478bd9Sstevel@tonic-gate * Configure the system. 21947c478bd9Sstevel@tonic-gate */ 21957c478bd9Sstevel@tonic-gate PRM_POINT("Calling configure()..."); 21967c478bd9Sstevel@tonic-gate configure(); /* set up devices */ 21977c478bd9Sstevel@tonic-gate PRM_POINT("configure() done"); 21987c478bd9Sstevel@tonic-gate 21997af88ac7SKuriakose Kuruvilla /* 22007af88ac7SKuriakose Kuruvilla * We can now setup for XSAVE because fpu_probe is done in configure(). 22017af88ac7SKuriakose Kuruvilla */ 22027af88ac7SKuriakose Kuruvilla if (fp_save_mech == FP_XSAVE) { 22037af88ac7SKuriakose Kuruvilla xsave_setup_msr(CPU); 22047af88ac7SKuriakose Kuruvilla } 22057af88ac7SKuriakose Kuruvilla 22067c478bd9Sstevel@tonic-gate /* 22077c478bd9Sstevel@tonic-gate * Set the isa_list string to the defined instruction sets we 22087c478bd9Sstevel@tonic-gate * support. 22097c478bd9Sstevel@tonic-gate */ 22107c478bd9Sstevel@tonic-gate setx86isalist(); 2211100b72f4Sandrei cpu_intr_alloc(CPU, NINTR_THREADS); 22127c478bd9Sstevel@tonic-gate psm_install(); 22137c478bd9Sstevel@tonic-gate 22147c478bd9Sstevel@tonic-gate /* 22157c478bd9Sstevel@tonic-gate * We're done with bootops. We don't unmap the bootstrap yet because 22167c478bd9Sstevel@tonic-gate * we're still using bootsvcs. 22177c478bd9Sstevel@tonic-gate */ 2218ae115bc7Smrj PRM_POINT("NULLing out bootops"); 2219ae115bc7Smrj *bootopsp = (struct bootops *)NULL; 22207c478bd9Sstevel@tonic-gate bootops = (struct bootops *)NULL; 22217c478bd9Sstevel@tonic-gate 2222843e1988Sjohnlev #if defined(__xpv) 2223843e1988Sjohnlev ec_init_debug_irq(); 2224843e1988Sjohnlev xs_domu_init(); 2225843e1988Sjohnlev #endif 22261de082f7SVikram Hegde 22273a634bfcSVikram Hegde #if defined(__amd64) && !defined(__xpv) 22283a634bfcSVikram Hegde /* 22293a634bfcSVikram Hegde * Intel IOMMU has been setup/initialized in ddi_impl.c 22303a634bfcSVikram Hegde * Start it up now. 22313a634bfcSVikram Hegde */ 22323a634bfcSVikram Hegde immu_startup(); 22331de082f7SVikram Hegde #endif 22343a634bfcSVikram Hegde 22357c478bd9Sstevel@tonic-gate PRM_POINT("Enabling interrupts"); 22367c478bd9Sstevel@tonic-gate (*picinitf)(); 22377c478bd9Sstevel@tonic-gate sti(); 2238843e1988Sjohnlev #if defined(__xpv) 2239843e1988Sjohnlev ASSERT(CPU->cpu_m.mcpu_vcpu_info->evtchn_upcall_mask == 0); 2240843e1988Sjohnlev xen_late_startup(); 2241843e1988Sjohnlev #endif 22427c478bd9Sstevel@tonic-gate 22437c478bd9Sstevel@tonic-gate (void) add_avsoftintr((void *)&softlevel1_hdl, 1, softlevel1, 224435b1ab99Sjosephb "softlevel1", NULL, NULL); /* XXX to be moved later */ 22457c478bd9Sstevel@tonic-gate 2246dd4eeefdSeota /* 2247dd4eeefdSeota * Register these software interrupts for ddi timer. 2248dd4eeefdSeota * Software interrupts up to the level 10 are supported. 2249dd4eeefdSeota */ 2250dd4eeefdSeota for (i = DDI_IPL_1; i <= DDI_IPL_10; i++) { 2251dd4eeefdSeota char name[sizeof ("timer_softintr") + 2]; 2252dd4eeefdSeota (void) sprintf(name, "timer_softintr%02d", i); 2253dd4eeefdSeota (void) add_avsoftintr((void *)&softlevel_hdl[i-1], i, 2254dd4eeefdSeota (avfunc)timer_softintr, name, (caddr_t)(uintptr_t)i, NULL); 2255dd4eeefdSeota } 2256dd4eeefdSeota 22577125fcbdSVikram Hegde #if !defined(__xpv) 22587125fcbdSVikram Hegde if (modload("drv", "amd_iommu") < 0) { 22597125fcbdSVikram Hegde PRM_POINT("No AMD IOMMU present\n"); 226094f1124eSVikram Hegde } else if (ddi_hold_installed_driver(ddi_name_to_major( 226194f1124eSVikram Hegde "amd_iommu")) == NULL) { 22627125fcbdSVikram Hegde prom_printf("ERROR: failed to attach AMD IOMMU\n"); 22637125fcbdSVikram Hegde } 22647125fcbdSVikram Hegde #endif 2265e774b42bSBill Holler post_startup_cpu_fixups(); 22667125fcbdSVikram Hegde 22677c478bd9Sstevel@tonic-gate PRM_POINT("startup_end() done"); 22687c478bd9Sstevel@tonic-gate } 22697c478bd9Sstevel@tonic-gate 227096d008c2SKen Erickson /* 227196d008c2SKen Erickson * Don't remove the following 2 variables. They are necessary 227296d008c2SKen Erickson * for reading the hostid from the legacy file (/kernel/misc/sysinit). 227396d008c2SKen Erickson */ 22747c478bd9Sstevel@tonic-gate char *_hs1107 = hw_serial; 22757c478bd9Sstevel@tonic-gate ulong_t _bdhs34; 22767c478bd9Sstevel@tonic-gate 22777c478bd9Sstevel@tonic-gate void 22787c478bd9Sstevel@tonic-gate post_startup(void) 22797c478bd9Sstevel@tonic-gate { 22800e751525SEric Saxe extern void cpupm_init(cpu_t *); 2281fb2caebeSRandy Fishel extern void cpu_event_init_cpu(cpu_t *); 22820e751525SEric Saxe 22837c478bd9Sstevel@tonic-gate /* 22847c478bd9Sstevel@tonic-gate * Set the system wide, processor-specific flags to be passed 22857c478bd9Sstevel@tonic-gate * to userland via the aux vector for performance hints and 22867c478bd9Sstevel@tonic-gate * instruction set extensions. 22877c478bd9Sstevel@tonic-gate */ 22887c478bd9Sstevel@tonic-gate bind_hwcap(); 22897c478bd9Sstevel@tonic-gate 2290843e1988Sjohnlev #ifdef __xpv 2291843e1988Sjohnlev if (DOMAIN_IS_INITDOMAIN(xen_info)) 2292843e1988Sjohnlev #endif 2293843e1988Sjohnlev { 2294843e1988Sjohnlev #if defined(__xpv) 2295843e1988Sjohnlev xpv_panic_init(); 2296843e1988Sjohnlev #else 2297843e1988Sjohnlev /* 2298843e1988Sjohnlev * Startup the memory scrubber. 2299843e1988Sjohnlev * XXPV This should be running somewhere .. 2300843e1988Sjohnlev */ 2301b9bfdccdSStuart Maybee if (get_hwenv() != HW_XEN_HVM) 2302e4b86885SCheng Sean Ye memscrub_init(); 2303843e1988Sjohnlev #endif 2304843e1988Sjohnlev } 23057c478bd9Sstevel@tonic-gate 23067aec1d6eScindi /* 23077aec1d6eScindi * Complete CPU module initialization 23087aec1d6eScindi */ 230920c794b3Sgavinm cmi_post_startup(); 23107aec1d6eScindi 23117c478bd9Sstevel@tonic-gate /* 231284ab085aSmws * Perform forceloading tasks for /etc/system. 23137c478bd9Sstevel@tonic-gate */ 231484ab085aSmws (void) mod_sysctl(SYS_FORCELOAD, NULL); 23157c478bd9Sstevel@tonic-gate 23167c478bd9Sstevel@tonic-gate /* 23177c478bd9Sstevel@tonic-gate * ON4.0: Force /proc module in until clock interrupt handle fixed 23187c478bd9Sstevel@tonic-gate * ON4.0: This must be fixed or restated in /etc/systems. 23197c478bd9Sstevel@tonic-gate */ 23207c478bd9Sstevel@tonic-gate (void) modload("fs", "procfs"); 23217c478bd9Sstevel@tonic-gate 2322c35aa225Smarx (void) i_ddi_attach_hw_nodes("pit_beep"); 2323c35aa225Smarx 23247c478bd9Sstevel@tonic-gate #if defined(__i386) 23257c478bd9Sstevel@tonic-gate /* 23267c478bd9Sstevel@tonic-gate * Check for required functional Floating Point hardware, 23277c478bd9Sstevel@tonic-gate * unless FP hardware explicitly disabled. 23287c478bd9Sstevel@tonic-gate */ 23297c478bd9Sstevel@tonic-gate if (fpu_exists && (fpu_pentium_fdivbug || fp_kind == FP_NO)) 23307c478bd9Sstevel@tonic-gate halt("No working FP hardware found"); 23317c478bd9Sstevel@tonic-gate #endif 23327c478bd9Sstevel@tonic-gate 23337c478bd9Sstevel@tonic-gate maxmem = freemem; 23347c478bd9Sstevel@tonic-gate 2335fb2caebeSRandy Fishel cpu_event_init_cpu(CPU); 23360e751525SEric Saxe cpupm_init(CPU); 2337fa96bd91SMichael Corcoran (void) mach_cpu_create_device_node(CPU, NULL); 23380e751525SEric Saxe 23390e751525SEric Saxe pg_init(); 23407c478bd9Sstevel@tonic-gate } 23417c478bd9Sstevel@tonic-gate 23427c478bd9Sstevel@tonic-gate static int 234319397407SSherry Moore pp_in_range(page_t *pp, uint64_t low_addr, uint64_t high_addr) 23447c478bd9Sstevel@tonic-gate { 234519397407SSherry Moore return ((pp->p_pagenum >= btop(low_addr)) && 234619397407SSherry Moore (pp->p_pagenum < btopr(high_addr))); 23477c478bd9Sstevel@tonic-gate } 23487c478bd9Sstevel@tonic-gate 23497c478bd9Sstevel@tonic-gate void 23507c478bd9Sstevel@tonic-gate release_bootstrap(void) 23517c478bd9Sstevel@tonic-gate { 23527c478bd9Sstevel@tonic-gate int root_is_ramdisk; 23537c478bd9Sstevel@tonic-gate page_t *pp; 23547c478bd9Sstevel@tonic-gate extern void kobj_boot_unmountroot(void); 23557c478bd9Sstevel@tonic-gate extern dev_t rootdev; 235655d507a9SSeth Goldberg #if !defined(__xpv) 235755d507a9SSeth Goldberg pfn_t pfn; 235855d507a9SSeth Goldberg #endif 23597c478bd9Sstevel@tonic-gate 23607c478bd9Sstevel@tonic-gate /* unmount boot ramdisk and release kmem usage */ 23617c478bd9Sstevel@tonic-gate kobj_boot_unmountroot(); 23627c478bd9Sstevel@tonic-gate 23637c478bd9Sstevel@tonic-gate /* 23647c478bd9Sstevel@tonic-gate * We're finished using the boot loader so free its pages. 23657c478bd9Sstevel@tonic-gate */ 23667c478bd9Sstevel@tonic-gate PRM_POINT("Unmapping lower boot pages"); 236719397407SSherry Moore 2368ae115bc7Smrj clear_boot_mappings(0, _userlimit); 236919397407SSherry Moore 2370ae115bc7Smrj postbootkernelbase = kernelbase; 23717c478bd9Sstevel@tonic-gate 23727c478bd9Sstevel@tonic-gate /* 23737c478bd9Sstevel@tonic-gate * If root isn't on ramdisk, destroy the hardcoded 23747c478bd9Sstevel@tonic-gate * ramdisk node now and release the memory. Else, 23757c478bd9Sstevel@tonic-gate * ramdisk memory is kept in rd_pages. 23767c478bd9Sstevel@tonic-gate */ 23777c478bd9Sstevel@tonic-gate root_is_ramdisk = (getmajor(rootdev) == ddi_name_to_major("ramdisk")); 23787c478bd9Sstevel@tonic-gate if (!root_is_ramdisk) { 23797c478bd9Sstevel@tonic-gate dev_info_t *dip = ddi_find_devinfo("ramdisk", -1, 0); 23807c478bd9Sstevel@tonic-gate ASSERT(dip && ddi_get_parent(dip) == ddi_root_node()); 23817c478bd9Sstevel@tonic-gate ndi_rele_devi(dip); /* held from ddi_find_devinfo */ 23827c478bd9Sstevel@tonic-gate (void) ddi_remove_child(dip, 0); 23837c478bd9Sstevel@tonic-gate } 23847c478bd9Sstevel@tonic-gate 23857c478bd9Sstevel@tonic-gate PRM_POINT("Releasing boot pages"); 23867c478bd9Sstevel@tonic-gate while (bootpages) { 238719397407SSherry Moore extern uint64_t ramdisk_start, ramdisk_end; 23887c478bd9Sstevel@tonic-gate pp = bootpages; 23897c478bd9Sstevel@tonic-gate bootpages = pp->p_next; 239019397407SSherry Moore 239119397407SSherry Moore 239219397407SSherry Moore /* Keep pages for the lower 64K */ 239319397407SSherry Moore if (pp_in_range(pp, 0, 0x40000)) { 239419397407SSherry Moore pp->p_next = lower_pages; 239519397407SSherry Moore lower_pages = pp; 239619397407SSherry Moore lower_pages_count++; 239719397407SSherry Moore continue; 239819397407SSherry Moore } 239919397407SSherry Moore 240019397407SSherry Moore 240119397407SSherry Moore if (root_is_ramdisk && pp_in_range(pp, ramdisk_start, 240219397407SSherry Moore ramdisk_end)) { 24037c478bd9Sstevel@tonic-gate pp->p_next = rd_pages; 24047c478bd9Sstevel@tonic-gate rd_pages = pp; 24057c478bd9Sstevel@tonic-gate continue; 24067c478bd9Sstevel@tonic-gate } 24077c478bd9Sstevel@tonic-gate pp->p_next = (struct page *)0; 2408ae115bc7Smrj pp->p_prev = (struct page *)0; 2409ae115bc7Smrj PP_CLRBOOTPAGES(pp); 24107c478bd9Sstevel@tonic-gate page_free(pp, 1); 24117c478bd9Sstevel@tonic-gate } 2412ae115bc7Smrj PRM_POINT("Boot pages released"); 24137c478bd9Sstevel@tonic-gate 2414843e1988Sjohnlev #if !defined(__xpv) 2415843e1988Sjohnlev /* XXPV -- note this following bunch of code needs to be revisited in Xen 3.0 */ 24167c478bd9Sstevel@tonic-gate /* 241755d507a9SSeth Goldberg * Find 1 page below 1 MB so that other processors can boot up or 241855d507a9SSeth Goldberg * so that any processor can resume. 24197c478bd9Sstevel@tonic-gate * Make sure it has a kernel VA as well as a 1:1 mapping. 24207c478bd9Sstevel@tonic-gate * We should have just free'd one up. 24217c478bd9Sstevel@tonic-gate */ 242219397407SSherry Moore 242319397407SSherry Moore /* 242419397407SSherry Moore * 0x10 pages is 64K. Leave the bottom 64K alone 242519397407SSherry Moore * for BIOS. 242619397407SSherry Moore */ 242719397407SSherry Moore for (pfn = 0x10; pfn < btop(1*1024*1024); pfn++) { 242855d507a9SSeth Goldberg if (page_numtopp_alloc(pfn) == NULL) 242955d507a9SSeth Goldberg continue; 243055d507a9SSeth Goldberg rm_platter_va = i86devmap(pfn, 1, 243155d507a9SSeth Goldberg PROT_READ | PROT_WRITE | PROT_EXEC); 243255d507a9SSeth Goldberg rm_platter_pa = ptob(pfn); 243355d507a9SSeth Goldberg break; 24347c478bd9Sstevel@tonic-gate } 243555d507a9SSeth Goldberg if (pfn == btop(1*1024*1024) && use_mp) 243655d507a9SSeth Goldberg panic("No page below 1M available for starting " 243755d507a9SSeth Goldberg "other processors or for resuming from system-suspend"); 2438843e1988Sjohnlev #endif /* !__xpv */ 24397c478bd9Sstevel@tonic-gate } 24407c478bd9Sstevel@tonic-gate 24417c478bd9Sstevel@tonic-gate /* 24427c478bd9Sstevel@tonic-gate * Initialize the platform-specific parts of a page_t. 24437c478bd9Sstevel@tonic-gate */ 24447c478bd9Sstevel@tonic-gate void 24457c478bd9Sstevel@tonic-gate add_physmem_cb(page_t *pp, pfn_t pnum) 24467c478bd9Sstevel@tonic-gate { 24477c478bd9Sstevel@tonic-gate pp->p_pagenum = pnum; 24487c478bd9Sstevel@tonic-gate pp->p_mapping = NULL; 24497c478bd9Sstevel@tonic-gate pp->p_embed = 0; 24507c478bd9Sstevel@tonic-gate pp->p_share = 0; 24517c478bd9Sstevel@tonic-gate pp->p_mlentry = 0; 24527c478bd9Sstevel@tonic-gate } 24537c478bd9Sstevel@tonic-gate 24547c478bd9Sstevel@tonic-gate /* 24557c478bd9Sstevel@tonic-gate * kphysm_init() initializes physical memory. 24567c478bd9Sstevel@tonic-gate */ 24577c478bd9Sstevel@tonic-gate static pgcnt_t 24587c478bd9Sstevel@tonic-gate kphysm_init( 2459ae115bc7Smrj page_t *pp, 24607c478bd9Sstevel@tonic-gate pgcnt_t npages) 24617c478bd9Sstevel@tonic-gate { 24627c478bd9Sstevel@tonic-gate struct memlist *pmem; 24637c478bd9Sstevel@tonic-gate struct memseg *cur_memseg; 24647c478bd9Sstevel@tonic-gate pfn_t base_pfn; 246518968004SKit Chow pfn_t end_pfn; 24667c478bd9Sstevel@tonic-gate pgcnt_t num; 24677c478bd9Sstevel@tonic-gate pgcnt_t pages_done = 0; 24687c478bd9Sstevel@tonic-gate uint64_t addr; 24697c478bd9Sstevel@tonic-gate uint64_t size; 24707c478bd9Sstevel@tonic-gate extern pfn_t ddiphysmin; 247118968004SKit Chow extern int mnode_xwa; 247218968004SKit Chow int ms = 0, me = 0; 24737c478bd9Sstevel@tonic-gate 24747c478bd9Sstevel@tonic-gate ASSERT(page_hash != NULL && page_hashsz != 0); 24757c478bd9Sstevel@tonic-gate 2476ae115bc7Smrj cur_memseg = memseg_base; 247756f33205SJonathan Adams for (pmem = phys_avail; pmem && npages; pmem = pmem->ml_next) { 24787c478bd9Sstevel@tonic-gate /* 24797c478bd9Sstevel@tonic-gate * In a 32 bit kernel can't use higher memory if we're 24807c478bd9Sstevel@tonic-gate * not booting in PAE mode. This check takes care of that. 24817c478bd9Sstevel@tonic-gate */ 248256f33205SJonathan Adams addr = pmem->ml_address; 248356f33205SJonathan Adams size = pmem->ml_size; 24847c478bd9Sstevel@tonic-gate if (btop(addr) > physmax) 24857c478bd9Sstevel@tonic-gate continue; 24867c478bd9Sstevel@tonic-gate 24877c478bd9Sstevel@tonic-gate /* 24887c478bd9Sstevel@tonic-gate * align addr and size - they may not be at page boundaries 24897c478bd9Sstevel@tonic-gate */ 24907c478bd9Sstevel@tonic-gate if ((addr & MMU_PAGEOFFSET) != 0) { 24917c478bd9Sstevel@tonic-gate addr += MMU_PAGEOFFSET; 24927c478bd9Sstevel@tonic-gate addr &= ~(uint64_t)MMU_PAGEOFFSET; 249356f33205SJonathan Adams size -= addr - pmem->ml_address; 24947c478bd9Sstevel@tonic-gate } 24957c478bd9Sstevel@tonic-gate 24966bb54764Skchow /* only process pages below or equal to physmax */ 24976bb54764Skchow if ((btop(addr + size) - 1) > physmax) 24986bb54764Skchow size = ptob(physmax - btop(addr) + 1); 24997c478bd9Sstevel@tonic-gate 25007c478bd9Sstevel@tonic-gate num = btop(size); 25017c478bd9Sstevel@tonic-gate if (num == 0) 25027c478bd9Sstevel@tonic-gate continue; 25037c478bd9Sstevel@tonic-gate 25047c478bd9Sstevel@tonic-gate if (num > npages) 25057c478bd9Sstevel@tonic-gate num = npages; 25067c478bd9Sstevel@tonic-gate 25077c478bd9Sstevel@tonic-gate npages -= num; 25087c478bd9Sstevel@tonic-gate pages_done += num; 25097c478bd9Sstevel@tonic-gate base_pfn = btop(addr); 25107c478bd9Sstevel@tonic-gate 25117c478bd9Sstevel@tonic-gate if (prom_debug) 25127c478bd9Sstevel@tonic-gate prom_printf("MEMSEG addr=0x%" PRIx64 25137c478bd9Sstevel@tonic-gate " pgs=0x%lx pfn 0x%lx-0x%lx\n", 25147c478bd9Sstevel@tonic-gate addr, num, base_pfn, base_pfn + num); 25157c478bd9Sstevel@tonic-gate 25167c478bd9Sstevel@tonic-gate /* 2517ae115bc7Smrj * Ignore pages below ddiphysmin to simplify ddi memory 25187c478bd9Sstevel@tonic-gate * allocation with non-zero addr_lo requests. 25197c478bd9Sstevel@tonic-gate */ 25207c478bd9Sstevel@tonic-gate if (base_pfn < ddiphysmin) { 2521ae115bc7Smrj if (base_pfn + num <= ddiphysmin) 25227c478bd9Sstevel@tonic-gate continue; 25237c478bd9Sstevel@tonic-gate pp += (ddiphysmin - base_pfn); 25247c478bd9Sstevel@tonic-gate num -= (ddiphysmin - base_pfn); 25257c478bd9Sstevel@tonic-gate base_pfn = ddiphysmin; 25267c478bd9Sstevel@tonic-gate } 2527ae115bc7Smrj 25287c478bd9Sstevel@tonic-gate /* 252918968004SKit Chow * mnode_xwa is greater than 1 when large pages regions can 253018968004SKit Chow * cross memory node boundaries. To prevent the formation 253118968004SKit Chow * of these large pages, configure the memsegs based on the 253218968004SKit Chow * memory node ranges which had been made non-contiguous. 25337c478bd9Sstevel@tonic-gate */ 253418968004SKit Chow if (mnode_xwa > 1) { 253518968004SKit Chow 253618968004SKit Chow end_pfn = base_pfn + num - 1; 253718968004SKit Chow ms = PFN_2_MEM_NODE(base_pfn); 253818968004SKit Chow me = PFN_2_MEM_NODE(end_pfn); 253918968004SKit Chow 254018968004SKit Chow if (ms != me) { 254118968004SKit Chow /* 254218968004SKit Chow * current range spans more than 1 memory node. 254318968004SKit Chow * Set num to only the pfn range in the start 254418968004SKit Chow * memory node. 254518968004SKit Chow */ 254618968004SKit Chow num = mem_node_config[ms].physmax - base_pfn 254718968004SKit Chow + 1; 254818968004SKit Chow ASSERT(end_pfn > mem_node_config[ms].physmax); 254918968004SKit Chow } 25507c478bd9Sstevel@tonic-gate } 25517c478bd9Sstevel@tonic-gate 255218968004SKit Chow for (;;) { 255318968004SKit Chow /* 255418968004SKit Chow * Build the memsegs entry 255518968004SKit Chow */ 255618968004SKit Chow cur_memseg->pages = pp; 255718968004SKit Chow cur_memseg->epages = pp + num; 255818968004SKit Chow cur_memseg->pages_base = base_pfn; 255918968004SKit Chow cur_memseg->pages_end = base_pfn + num; 256018968004SKit Chow 256118968004SKit Chow /* 256218968004SKit Chow * Insert into memseg list in decreasing pfn range 256318968004SKit Chow * order. Low memory is typically more fragmented such 256418968004SKit Chow * that this ordering keeps the larger ranges at the 256518968004SKit Chow * front of the list for code that searches memseg. 256618968004SKit Chow * This ASSERTS that the memsegs coming in from boot 256718968004SKit Chow * are in increasing physical address order and not 256818968004SKit Chow * contiguous. 256918968004SKit Chow */ 257018968004SKit Chow if (memsegs != NULL) { 257118968004SKit Chow ASSERT(cur_memseg->pages_base >= 257218968004SKit Chow memsegs->pages_end); 257318968004SKit Chow cur_memseg->next = memsegs; 257418968004SKit Chow } 257518968004SKit Chow memsegs = cur_memseg; 257618968004SKit Chow 257718968004SKit Chow /* 257818968004SKit Chow * add_physmem() initializes the PSM part of the page 257918968004SKit Chow * struct by calling the PSM back with add_physmem_cb(). 258018968004SKit Chow * In addition it coalesces pages into larger pages as 258118968004SKit Chow * it initializes them. 258218968004SKit Chow */ 258318968004SKit Chow add_physmem(pp, num, base_pfn); 258418968004SKit Chow cur_memseg++; 258518968004SKit Chow availrmem_initial += num; 258618968004SKit Chow availrmem += num; 258718968004SKit Chow 258818968004SKit Chow pp += num; 258918968004SKit Chow if (ms >= me) 259018968004SKit Chow break; 25917c478bd9Sstevel@tonic-gate 259218968004SKit Chow /* process next memory node range */ 259318968004SKit Chow ms++; 259418968004SKit Chow base_pfn = mem_node_config[ms].physbase; 259518968004SKit Chow num = MIN(mem_node_config[ms].physmax, 259618968004SKit Chow end_pfn) - base_pfn + 1; 259718968004SKit Chow } 25987c478bd9Sstevel@tonic-gate } 25997c478bd9Sstevel@tonic-gate 26007c478bd9Sstevel@tonic-gate PRM_DEBUG(availrmem_initial); 26017c478bd9Sstevel@tonic-gate PRM_DEBUG(availrmem); 26027c478bd9Sstevel@tonic-gate PRM_DEBUG(freemem); 26037c478bd9Sstevel@tonic-gate build_pfn_hash(); 26047c478bd9Sstevel@tonic-gate return (pages_done); 26057c478bd9Sstevel@tonic-gate } 26067c478bd9Sstevel@tonic-gate 26077c478bd9Sstevel@tonic-gate /* 26087c478bd9Sstevel@tonic-gate * Kernel VM initialization. 26097c478bd9Sstevel@tonic-gate */ 26107c478bd9Sstevel@tonic-gate static void 26117c478bd9Sstevel@tonic-gate kvm_init(void) 26127c478bd9Sstevel@tonic-gate { 26137c478bd9Sstevel@tonic-gate ASSERT((((uintptr_t)s_text) & MMU_PAGEOFFSET) == 0); 26147c478bd9Sstevel@tonic-gate 26157c478bd9Sstevel@tonic-gate /* 26167c478bd9Sstevel@tonic-gate * Put the kernel segments in kernel address space. 26177c478bd9Sstevel@tonic-gate */ 26187c478bd9Sstevel@tonic-gate rw_enter(&kas.a_lock, RW_WRITER); 26197c478bd9Sstevel@tonic-gate as_avlinit(&kas); 26207c478bd9Sstevel@tonic-gate 26217c478bd9Sstevel@tonic-gate (void) seg_attach(&kas, s_text, e_moddata - s_text, &ktextseg); 26227c478bd9Sstevel@tonic-gate (void) segkmem_create(&ktextseg); 26237c478bd9Sstevel@tonic-gate 26247c478bd9Sstevel@tonic-gate (void) seg_attach(&kas, (caddr_t)valloc_base, valloc_sz, &kvalloc); 26257c478bd9Sstevel@tonic-gate (void) segkmem_create(&kvalloc); 26267c478bd9Sstevel@tonic-gate 262735b1ab99Sjosephb (void) seg_attach(&kas, kernelheap, 262835b1ab99Sjosephb ekernelheap - kernelheap, &kvseg); 26297c478bd9Sstevel@tonic-gate (void) segkmem_create(&kvseg); 26307c478bd9Sstevel@tonic-gate 2631ae115bc7Smrj if (core_size > 0) { 2632ae115bc7Smrj PRM_POINT("attaching kvseg_core"); 2633ae115bc7Smrj (void) seg_attach(&kas, (caddr_t)core_base, core_size, 2634ae115bc7Smrj &kvseg_core); 2635ae115bc7Smrj (void) segkmem_create(&kvseg_core); 2636ae115bc7Smrj } 2637ad23a2dbSjohansen 2638ae115bc7Smrj if (segziosize > 0) { 2639ae115bc7Smrj PRM_POINT("attaching segzio"); 2640ad23a2dbSjohansen (void) seg_attach(&kas, segzio_base, mmu_ptob(segziosize), 2641ad23a2dbSjohansen &kzioseg); 2642ad23a2dbSjohansen (void) segkmem_zio_create(&kzioseg); 2643ad23a2dbSjohansen 2644ad23a2dbSjohansen /* create zio area covering new segment */ 2645ad23a2dbSjohansen segkmem_zio_init(segzio_base, mmu_ptob(segziosize)); 2646ad23a2dbSjohansen } 26477c478bd9Sstevel@tonic-gate 2648ae115bc7Smrj (void) seg_attach(&kas, kdi_segdebugbase, kdi_segdebugsize, &kdebugseg); 26497c478bd9Sstevel@tonic-gate (void) segkmem_create(&kdebugseg); 26507c478bd9Sstevel@tonic-gate 26517c478bd9Sstevel@tonic-gate rw_exit(&kas.a_lock); 26527c478bd9Sstevel@tonic-gate 26537c478bd9Sstevel@tonic-gate /* 26547c478bd9Sstevel@tonic-gate * Ensure that the red zone at kernelbase is never accessible. 26557c478bd9Sstevel@tonic-gate */ 2656ae115bc7Smrj PRM_POINT("protecting redzone"); 26577c478bd9Sstevel@tonic-gate (void) as_setprot(&kas, (caddr_t)kernelbase, KERNEL_REDZONE_SIZE, 0); 26587c478bd9Sstevel@tonic-gate 26597c478bd9Sstevel@tonic-gate /* 26607c478bd9Sstevel@tonic-gate * Make the text writable so that it can be hot patched by DTrace. 26617c478bd9Sstevel@tonic-gate */ 26627c478bd9Sstevel@tonic-gate (void) as_setprot(&kas, s_text, e_modtext - s_text, 26637c478bd9Sstevel@tonic-gate PROT_READ | PROT_WRITE | PROT_EXEC); 26647c478bd9Sstevel@tonic-gate 26657c478bd9Sstevel@tonic-gate /* 26667c478bd9Sstevel@tonic-gate * Make data writable until end. 26677c478bd9Sstevel@tonic-gate */ 26687c478bd9Sstevel@tonic-gate (void) as_setprot(&kas, s_data, e_moddata - s_data, 26697c478bd9Sstevel@tonic-gate PROT_READ | PROT_WRITE | PROT_EXEC); 26707c478bd9Sstevel@tonic-gate } 26717c478bd9Sstevel@tonic-gate 2672843e1988Sjohnlev #ifndef __xpv 26737c478bd9Sstevel@tonic-gate /* 26741d03c31eSjohnlev * Solaris adds an entry for Write Combining caching to the PAT 26757c478bd9Sstevel@tonic-gate */ 26761d03c31eSjohnlev static uint64_t pat_attr_reg = PAT_DEFAULT_ATTRIBUTE; 26777c478bd9Sstevel@tonic-gate 26787c478bd9Sstevel@tonic-gate void 26791d03c31eSjohnlev pat_sync(void) 26807c478bd9Sstevel@tonic-gate { 26811d03c31eSjohnlev ulong_t cr0, cr0_orig, cr4; 26827c478bd9Sstevel@tonic-gate 26837417cfdeSKuriakose Kuruvilla if (!is_x86_feature(x86_featureset, X86FSET_PAT)) 26847c478bd9Sstevel@tonic-gate return; 26851d03c31eSjohnlev cr0_orig = cr0 = getcr0(); 26861d03c31eSjohnlev cr4 = getcr4(); 26877c478bd9Sstevel@tonic-gate 26881d03c31eSjohnlev /* disable caching and flush all caches and TLBs */ 26891d03c31eSjohnlev cr0 |= CR0_CD; 26901d03c31eSjohnlev cr0 &= ~CR0_NW; 26911d03c31eSjohnlev setcr0(cr0); 26921d03c31eSjohnlev invalidate_cache(); 26931d03c31eSjohnlev if (cr4 & CR4_PGE) { 26941d03c31eSjohnlev setcr4(cr4 & ~(ulong_t)CR4_PGE); 26951d03c31eSjohnlev setcr4(cr4); 26961d03c31eSjohnlev } else { 26971d03c31eSjohnlev reload_cr3(); 26987c478bd9Sstevel@tonic-gate } 26997c478bd9Sstevel@tonic-gate 27001d03c31eSjohnlev /* add our entry to the PAT */ 27011d03c31eSjohnlev wrmsr(REG_PAT, pat_attr_reg); 27027c478bd9Sstevel@tonic-gate 27031d03c31eSjohnlev /* flush TLBs and cache again, then reenable cr0 caching */ 27041d03c31eSjohnlev if (cr4 & CR4_PGE) { 27051d03c31eSjohnlev setcr4(cr4 & ~(ulong_t)CR4_PGE); 27061d03c31eSjohnlev setcr4(cr4); 27071d03c31eSjohnlev } else { 27081d03c31eSjohnlev reload_cr3(); 27097c478bd9Sstevel@tonic-gate } 27107c478bd9Sstevel@tonic-gate invalidate_cache(); 27117c478bd9Sstevel@tonic-gate setcr0(cr0_orig); 27127c478bd9Sstevel@tonic-gate } 27137c478bd9Sstevel@tonic-gate 27141d03c31eSjohnlev #endif /* !__xpv */ 27157c478bd9Sstevel@tonic-gate 2716ed5289f9SKen Erickson #if defined(_SOFT_HOSTID) 2717ed5289f9SKen Erickson /* 2718ed5289f9SKen Erickson * On platforms that do not have a hardware serial number, attempt 2719ed5289f9SKen Erickson * to set one based on the contents of /etc/hostid. If this file does 2720ed5289f9SKen Erickson * not exist, assume that we are to generate a new hostid and set 2721ed5289f9SKen Erickson * it in the kernel, for subsequent saving by a userland process 2722ed5289f9SKen Erickson * once the system is up and the root filesystem is mounted r/w. 2723ed5289f9SKen Erickson * 272496d008c2SKen Erickson * In order to gracefully support upgrade on OpenSolaris, if 272596d008c2SKen Erickson * /etc/hostid does not exist, we will attempt to get a serial number 272696d008c2SKen Erickson * using the legacy method (/kernel/misc/sysinit). 272796d008c2SKen Erickson * 2728*9e3e4df2SGarrett D'Amore * If that isn't present, we attempt to use an SMBIOS UUID, which is 2729*9e3e4df2SGarrett D'Amore * a hardware serial number. Note that we don't automatically trust 2730*9e3e4df2SGarrett D'Amore * all SMBIOS UUIDs (some older platforms are defective and ship duplicate 2731*9e3e4df2SGarrett D'Amore * UUIDs in violation of the standard), we check against a blacklist. 2732*9e3e4df2SGarrett D'Amore * 2733ed5289f9SKen Erickson * In an attempt to make the hostid less prone to abuse 2734ed5289f9SKen Erickson * (for license circumvention, etc), we store it in /etc/hostid 2735ed5289f9SKen Erickson * in rot47 format. 2736ed5289f9SKen Erickson */ 2737ed5289f9SKen Erickson extern volatile unsigned long tenmicrodata; 273896d008c2SKen Erickson static int atoi(char *); 2739ed5289f9SKen Erickson 2740*9e3e4df2SGarrett D'Amore /* 2741*9e3e4df2SGarrett D'Amore * Set this to non-zero in /etc/system if you think your SMBIOS returns a 2742*9e3e4df2SGarrett D'Amore * UUID that is not unique. (Also report it so that the smbios_uuid_blacklist 2743*9e3e4df2SGarrett D'Amore * array can be updated.) 2744*9e3e4df2SGarrett D'Amore */ 2745*9e3e4df2SGarrett D'Amore int smbios_broken_uuid = 0; 2746*9e3e4df2SGarrett D'Amore 2747*9e3e4df2SGarrett D'Amore /* 2748*9e3e4df2SGarrett D'Amore * List of known bad UUIDs. This is just the lower 32-bit values, since 2749*9e3e4df2SGarrett D'Amore * that's what we use for the host id. If your hostid falls here, you need 2750*9e3e4df2SGarrett D'Amore * to contact your hardware OEM for a fix for your BIOS. 2751*9e3e4df2SGarrett D'Amore */ 2752*9e3e4df2SGarrett D'Amore static unsigned char 2753*9e3e4df2SGarrett D'Amore smbios_uuid_blacklist[][16] = { 2754*9e3e4df2SGarrett D'Amore 2755*9e3e4df2SGarrett D'Amore { /* Reported bad UUID (Google search) */ 2756*9e3e4df2SGarrett D'Amore 0x00, 0x02, 0x00, 0x03, 0x00, 0x04, 0x00, 0x05, 2757*9e3e4df2SGarrett D'Amore 0x00, 0x06, 0x00, 0x07, 0x00, 0x08, 0x00, 0x09, 2758*9e3e4df2SGarrett D'Amore }, 2759*9e3e4df2SGarrett D'Amore { /* Known bad DELL UUID */ 2760*9e3e4df2SGarrett D'Amore 0x4C, 0x4C, 0x45, 0x44, 0x00, 0x00, 0x20, 0x10, 2761*9e3e4df2SGarrett D'Amore 0x80, 0x20, 0x80, 0xC0, 0x4F, 0x20, 0x20, 0x20, 2762*9e3e4df2SGarrett D'Amore }, 2763*9e3e4df2SGarrett D'Amore { /* Uninitialized flash */ 2764*9e3e4df2SGarrett D'Amore 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 2765*9e3e4df2SGarrett D'Amore 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff 2766*9e3e4df2SGarrett D'Amore }, 2767*9e3e4df2SGarrett D'Amore { /* All zeros */ 2768*9e3e4df2SGarrett D'Amore 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 2769*9e3e4df2SGarrett D'Amore 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00 2770*9e3e4df2SGarrett D'Amore }, 2771*9e3e4df2SGarrett D'Amore }; 2772*9e3e4df2SGarrett D'Amore 2773*9e3e4df2SGarrett D'Amore static int32_t 2774*9e3e4df2SGarrett D'Amore uuid_to_hostid(const uint8_t *uuid) 2775*9e3e4df2SGarrett D'Amore { 2776*9e3e4df2SGarrett D'Amore /* 2777*9e3e4df2SGarrett D'Amore * Although the UUIDs are 128-bits, they may not distribute entropy 2778*9e3e4df2SGarrett D'Amore * evenly. We would like to use SHA or MD5, but those are located 2779*9e3e4df2SGarrett D'Amore * in loadable modules and not available this early in boot. As we 2780*9e3e4df2SGarrett D'Amore * don't need the values to be cryptographically strong, we just 2781*9e3e4df2SGarrett D'Amore * generate 32-bit vaue by xor'ing the various sequences together, 2782*9e3e4df2SGarrett D'Amore * which ensures that the enire UUID contributes to the hostid. 2783*9e3e4df2SGarrett D'Amore */ 2784*9e3e4df2SGarrett D'Amore int32_t id = 0; 2785*9e3e4df2SGarrett D'Amore 2786*9e3e4df2SGarrett D'Amore /* first check against the blacklist */ 2787*9e3e4df2SGarrett D'Amore for (int i = 0; i < (sizeof (smbios_uuid_blacklist) / 16); i++) { 2788*9e3e4df2SGarrett D'Amore if (bcmp(smbios_uuid_blacklist[0], uuid, 16) == 0) { 2789*9e3e4df2SGarrett D'Amore cmn_err(CE_CONT, "?Broken SMBIOS UUID. " 2790*9e3e4df2SGarrett D'Amore "Contact BIOS manufacturer for repair.\n"); 2791*9e3e4df2SGarrett D'Amore return ((int32_t)HW_INVALID_HOSTID); 2792*9e3e4df2SGarrett D'Amore } 2793*9e3e4df2SGarrett D'Amore } 2794*9e3e4df2SGarrett D'Amore 2795*9e3e4df2SGarrett D'Amore for (int i = 0; i < 16; i++) 2796*9e3e4df2SGarrett D'Amore id ^= ((uuid[i]) << (8 * (i % sizeof (id)))); 2797*9e3e4df2SGarrett D'Amore 2798*9e3e4df2SGarrett D'Amore return (id); 2799*9e3e4df2SGarrett D'Amore } 2800*9e3e4df2SGarrett D'Amore 2801ed5289f9SKen Erickson static int32_t 2802ed5289f9SKen Erickson set_soft_hostid(void) 2803ed5289f9SKen Erickson { 2804ed5289f9SKen Erickson struct _buf *file; 2805ed5289f9SKen Erickson char tokbuf[MAXNAMELEN]; 2806ed5289f9SKen Erickson token_t token; 2807ed5289f9SKen Erickson int done = 0; 2808ed5289f9SKen Erickson u_longlong_t tmp; 280996d008c2SKen Erickson int i; 28105679c89fSjv int32_t hostid = (int32_t)HW_INVALID_HOSTID; 2811ed5289f9SKen Erickson unsigned char *c; 2812ed5289f9SKen Erickson hrtime_t tsc; 2813*9e3e4df2SGarrett D'Amore smbios_system_t smsys; 2814ed5289f9SKen Erickson 2815ed5289f9SKen Erickson /* 2816ed5289f9SKen Erickson * If /etc/hostid file not found, we'd like to get a pseudo 2817ed5289f9SKen Erickson * random number to use at the hostid. A nice way to do this 2818ed5289f9SKen Erickson * is to read the real time clock. To remain xen-compatible, 2819ed5289f9SKen Erickson * we can't poke the real hardware, so we use tsc_read() to 2820ed5289f9SKen Erickson * read the real time clock. However, there is an ominous 2821ed5289f9SKen Erickson * warning in tsc_read that says it can return zero, so we 2822ed5289f9SKen Erickson * deal with that possibility by falling back to using the 2823ed5289f9SKen Erickson * (hopefully random enough) value in tenmicrodata. 2824ed5289f9SKen Erickson */ 2825ed5289f9SKen Erickson 2826ed5289f9SKen Erickson if ((file = kobj_open_file(hostid_file)) == (struct _buf *)-1) { 282796d008c2SKen Erickson /* 282896d008c2SKen Erickson * hostid file not found - try to load sysinit module 282996d008c2SKen Erickson * and see if it has a nonzero hostid value...use that 283096d008c2SKen Erickson * instead of generating a new hostid here if so. 283196d008c2SKen Erickson */ 283296d008c2SKen Erickson if ((i = modload("misc", "sysinit")) != -1) { 283396d008c2SKen Erickson if (strlen(hw_serial) > 0) 283496d008c2SKen Erickson hostid = (int32_t)atoi(hw_serial); 283596d008c2SKen Erickson (void) modunload(i); 283696d008c2SKen Erickson } 2837*9e3e4df2SGarrett D'Amore 2838*9e3e4df2SGarrett D'Amore /* 2839*9e3e4df2SGarrett D'Amore * We try to use the SMBIOS UUID. But not if it is blacklisted 2840*9e3e4df2SGarrett D'Amore * in /etc/system. 2841*9e3e4df2SGarrett D'Amore */ 2842*9e3e4df2SGarrett D'Amore if ((hostid == HW_INVALID_HOSTID) && 2843*9e3e4df2SGarrett D'Amore (smbios_broken_uuid == 0) && 2844*9e3e4df2SGarrett D'Amore (ksmbios != NULL) && 2845*9e3e4df2SGarrett D'Amore (smbios_info_system(ksmbios, &smsys) != SMB_ERR) && 2846*9e3e4df2SGarrett D'Amore (smsys.smbs_uuidlen >= 16)) { 2847*9e3e4df2SGarrett D'Amore hostid = uuid_to_hostid(smsys.smbs_uuid); 2848*9e3e4df2SGarrett D'Amore } 2849*9e3e4df2SGarrett D'Amore 2850*9e3e4df2SGarrett D'Amore /* 2851*9e3e4df2SGarrett D'Amore * Generate a "random" hostid using the clock. These 2852*9e3e4df2SGarrett D'Amore * hostids will change on each boot if the value is not 2853*9e3e4df2SGarrett D'Amore * saved to a persistent /etc/hostid file. 2854*9e3e4df2SGarrett D'Amore */ 28555679c89fSjv if (hostid == HW_INVALID_HOSTID) { 285696d008c2SKen Erickson tsc = tsc_read(); 285796d008c2SKen Erickson if (tsc == 0) /* tsc_read can return zero sometimes */ 285896d008c2SKen Erickson hostid = (int32_t)tenmicrodata & 0x0CFFFFF; 285996d008c2SKen Erickson else 286096d008c2SKen Erickson hostid = (int32_t)tsc & 0x0CFFFFF; 286196d008c2SKen Erickson } 2862ed5289f9SKen Erickson } else { 2863ed5289f9SKen Erickson /* hostid file found */ 2864ed5289f9SKen Erickson while (!done) { 2865ed5289f9SKen Erickson token = kobj_lex(file, tokbuf, sizeof (tokbuf)); 2866ed5289f9SKen Erickson 2867ed5289f9SKen Erickson switch (token) { 2868ed5289f9SKen Erickson case POUND: 2869ed5289f9SKen Erickson /* 2870ed5289f9SKen Erickson * skip comments 2871ed5289f9SKen Erickson */ 2872ed5289f9SKen Erickson kobj_find_eol(file); 2873ed5289f9SKen Erickson break; 2874ed5289f9SKen Erickson case STRING: 2875ed5289f9SKen Erickson /* 2876ed5289f9SKen Erickson * un-rot47 - obviously this 2877ed5289f9SKen Erickson * nonsense is ascii-specific 2878ed5289f9SKen Erickson */ 2879ed5289f9SKen Erickson for (c = (unsigned char *)tokbuf; 2880ed5289f9SKen Erickson *c != '\0'; c++) { 2881ed5289f9SKen Erickson *c += 47; 2882ed5289f9SKen Erickson if (*c > '~') 2883ed5289f9SKen Erickson *c -= 94; 2884ed5289f9SKen Erickson else if (*c < '!') 2885ed5289f9SKen Erickson *c += 94; 2886ed5289f9SKen Erickson } 2887ed5289f9SKen Erickson /* 2888ed5289f9SKen Erickson * now we should have a real number 2889ed5289f9SKen Erickson */ 2890ed5289f9SKen Erickson 2891ed5289f9SKen Erickson if (kobj_getvalue(tokbuf, &tmp) != 0) 2892ed5289f9SKen Erickson kobj_file_err(CE_WARN, file, 2893ed5289f9SKen Erickson "Bad value %s for hostid", 2894ed5289f9SKen Erickson tokbuf); 2895ed5289f9SKen Erickson else 2896ed5289f9SKen Erickson hostid = (int32_t)tmp; 2897ed5289f9SKen Erickson 2898ed5289f9SKen Erickson break; 2899ed5289f9SKen Erickson case EOF: 2900ed5289f9SKen Erickson done = 1; 2901ed5289f9SKen Erickson /* FALLTHROUGH */ 2902ed5289f9SKen Erickson case NEWLINE: 2903ed5289f9SKen Erickson kobj_newline(file); 2904ed5289f9SKen Erickson break; 2905ed5289f9SKen Erickson default: 2906ed5289f9SKen Erickson break; 2907ed5289f9SKen Erickson 2908ed5289f9SKen Erickson } 2909ed5289f9SKen Erickson } 29105679c89fSjv if (hostid == HW_INVALID_HOSTID) /* didn't find a hostid */ 2911ed5289f9SKen Erickson kobj_file_err(CE_WARN, file, 2912ed5289f9SKen Erickson "hostid missing or corrupt"); 2913ed5289f9SKen Erickson 2914ed5289f9SKen Erickson kobj_close_file(file); 2915ed5289f9SKen Erickson } 2916ed5289f9SKen Erickson /* 2917ed5289f9SKen Erickson * hostid is now the value read from /etc/hostid, or the 29185679c89fSjv * new hostid we generated in this routine or HW_INVALID_HOSTID if not 29195679c89fSjv * set. 2920ed5289f9SKen Erickson */ 2921ed5289f9SKen Erickson return (hostid); 2922ed5289f9SKen Erickson } 292396d008c2SKen Erickson 292496d008c2SKen Erickson static int 292596d008c2SKen Erickson atoi(char *p) 292696d008c2SKen Erickson { 292796d008c2SKen Erickson int i = 0; 292896d008c2SKen Erickson 292996d008c2SKen Erickson while (*p != '\0') 293096d008c2SKen Erickson i = 10 * i + (*p++ - '0'); 293196d008c2SKen Erickson 293296d008c2SKen Erickson return (i); 293396d008c2SKen Erickson } 293496d008c2SKen Erickson 2935ed5289f9SKen Erickson #endif /* _SOFT_HOSTID */ 2936ed5289f9SKen Erickson 29377c478bd9Sstevel@tonic-gate void 293845916cd2Sjpk get_system_configuration(void) 29397c478bd9Sstevel@tonic-gate { 29407c478bd9Sstevel@tonic-gate char prop[32]; 29417c478bd9Sstevel@tonic-gate u_longlong_t nodes_ll, cpus_pernode_ll, lvalue; 29427c478bd9Sstevel@tonic-gate 294335b1ab99Sjosephb if (BOP_GETPROPLEN(bootops, "nodes") > sizeof (prop) || 294435b1ab99Sjosephb BOP_GETPROP(bootops, "nodes", prop) < 0 || 294535b1ab99Sjosephb kobj_getvalue(prop, &nodes_ll) == -1 || 294635b1ab99Sjosephb nodes_ll > MAXNODES || 294735b1ab99Sjosephb BOP_GETPROPLEN(bootops, "cpus_pernode") > sizeof (prop) || 294835b1ab99Sjosephb BOP_GETPROP(bootops, "cpus_pernode", prop) < 0 || 294935b1ab99Sjosephb kobj_getvalue(prop, &cpus_pernode_ll) == -1) { 29507c478bd9Sstevel@tonic-gate system_hardware.hd_nodes = 1; 29517c478bd9Sstevel@tonic-gate system_hardware.hd_cpus_per_node = 0; 29527c478bd9Sstevel@tonic-gate } else { 29537c478bd9Sstevel@tonic-gate system_hardware.hd_nodes = (int)nodes_ll; 29547c478bd9Sstevel@tonic-gate system_hardware.hd_cpus_per_node = (int)cpus_pernode_ll; 29557c478bd9Sstevel@tonic-gate } 295635b1ab99Sjosephb 295735b1ab99Sjosephb if (BOP_GETPROPLEN(bootops, "kernelbase") > sizeof (prop) || 295835b1ab99Sjosephb BOP_GETPROP(bootops, "kernelbase", prop) < 0 || 295935b1ab99Sjosephb kobj_getvalue(prop, &lvalue) == -1) 296035b1ab99Sjosephb eprom_kernelbase = NULL; 29617c478bd9Sstevel@tonic-gate else 296235b1ab99Sjosephb eprom_kernelbase = (uintptr_t)lvalue; 29637c478bd9Sstevel@tonic-gate 296435b1ab99Sjosephb if (BOP_GETPROPLEN(bootops, "segmapsize") > sizeof (prop) || 296535b1ab99Sjosephb BOP_GETPROP(bootops, "segmapsize", prop) < 0 || 296635b1ab99Sjosephb kobj_getvalue(prop, &lvalue) == -1) 29677c478bd9Sstevel@tonic-gate segmapsize = SEGMAPDEFAULT; 296835b1ab99Sjosephb else 29697c478bd9Sstevel@tonic-gate segmapsize = (uintptr_t)lvalue; 29707c478bd9Sstevel@tonic-gate 297135b1ab99Sjosephb if (BOP_GETPROPLEN(bootops, "segmapfreelists") > sizeof (prop) || 297235b1ab99Sjosephb BOP_GETPROP(bootops, "segmapfreelists", prop) < 0 || 297335b1ab99Sjosephb kobj_getvalue(prop, &lvalue) == -1) 29747c478bd9Sstevel@tonic-gate segmapfreelists = 0; /* use segmap driver default */ 297535b1ab99Sjosephb else 29767c478bd9Sstevel@tonic-gate segmapfreelists = (int)lvalue; 29774944b02eSkchow 2978aac11643Sjosephb /* physmem used to be here, but moved much earlier to fakebop.c */ 29797c478bd9Sstevel@tonic-gate } 29807c478bd9Sstevel@tonic-gate 29817c478bd9Sstevel@tonic-gate /* 29827c478bd9Sstevel@tonic-gate * Add to a memory list. 29837c478bd9Sstevel@tonic-gate * start = start of new memory segment 29847c478bd9Sstevel@tonic-gate * len = length of new memory segment in bytes 29857c478bd9Sstevel@tonic-gate * new = pointer to a new struct memlist 29867c478bd9Sstevel@tonic-gate * memlistp = memory list to which to add segment. 29877c478bd9Sstevel@tonic-gate */ 2988ae115bc7Smrj void 29897c478bd9Sstevel@tonic-gate memlist_add( 29907c478bd9Sstevel@tonic-gate uint64_t start, 29917c478bd9Sstevel@tonic-gate uint64_t len, 29927c478bd9Sstevel@tonic-gate struct memlist *new, 29937c478bd9Sstevel@tonic-gate struct memlist **memlistp) 29947c478bd9Sstevel@tonic-gate { 29957c478bd9Sstevel@tonic-gate struct memlist *cur; 29967c478bd9Sstevel@tonic-gate uint64_t end = start + len; 29977c478bd9Sstevel@tonic-gate 299856f33205SJonathan Adams new->ml_address = start; 299956f33205SJonathan Adams new->ml_size = len; 30007c478bd9Sstevel@tonic-gate 30017c478bd9Sstevel@tonic-gate cur = *memlistp; 30027c478bd9Sstevel@tonic-gate 30037c478bd9Sstevel@tonic-gate while (cur) { 300456f33205SJonathan Adams if (cur->ml_address >= end) { 300556f33205SJonathan Adams new->ml_next = cur; 30067c478bd9Sstevel@tonic-gate *memlistp = new; 300756f33205SJonathan Adams new->ml_prev = cur->ml_prev; 300856f33205SJonathan Adams cur->ml_prev = new; 30097c478bd9Sstevel@tonic-gate return; 30107c478bd9Sstevel@tonic-gate } 301156f33205SJonathan Adams ASSERT(cur->ml_address + cur->ml_size <= start); 301256f33205SJonathan Adams if (cur->ml_next == NULL) { 301356f33205SJonathan Adams cur->ml_next = new; 301456f33205SJonathan Adams new->ml_prev = cur; 301556f33205SJonathan Adams new->ml_next = NULL; 30167c478bd9Sstevel@tonic-gate return; 30177c478bd9Sstevel@tonic-gate } 301856f33205SJonathan Adams memlistp = &cur->ml_next; 301956f33205SJonathan Adams cur = cur->ml_next; 30207c478bd9Sstevel@tonic-gate } 30217c478bd9Sstevel@tonic-gate } 30227c478bd9Sstevel@tonic-gate 30237c478bd9Sstevel@tonic-gate void 30247c478bd9Sstevel@tonic-gate kobj_vmem_init(vmem_t **text_arena, vmem_t **data_arena) 30257c478bd9Sstevel@tonic-gate { 30267c478bd9Sstevel@tonic-gate size_t tsize = e_modtext - modtext; 30277c478bd9Sstevel@tonic-gate size_t dsize = e_moddata - moddata; 30287c478bd9Sstevel@tonic-gate 30297c478bd9Sstevel@tonic-gate *text_arena = vmem_create("module_text", tsize ? modtext : NULL, tsize, 30307c478bd9Sstevel@tonic-gate 1, segkmem_alloc, segkmem_free, heaptext_arena, 0, VM_SLEEP); 30317c478bd9Sstevel@tonic-gate *data_arena = vmem_create("module_data", dsize ? moddata : NULL, dsize, 30327c478bd9Sstevel@tonic-gate 1, segkmem_alloc, segkmem_free, heap32_arena, 0, VM_SLEEP); 30337c478bd9Sstevel@tonic-gate } 30347c478bd9Sstevel@tonic-gate 30357c478bd9Sstevel@tonic-gate caddr_t 30367c478bd9Sstevel@tonic-gate kobj_text_alloc(vmem_t *arena, size_t size) 30377c478bd9Sstevel@tonic-gate { 30387c478bd9Sstevel@tonic-gate return (vmem_alloc(arena, size, VM_SLEEP | VM_BESTFIT)); 30397c478bd9Sstevel@tonic-gate } 30407c478bd9Sstevel@tonic-gate 30417c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 30427c478bd9Sstevel@tonic-gate caddr_t 30437c478bd9Sstevel@tonic-gate kobj_texthole_alloc(caddr_t addr, size_t size) 30447c478bd9Sstevel@tonic-gate { 30457c478bd9Sstevel@tonic-gate panic("unexpected call to kobj_texthole_alloc()"); 30467c478bd9Sstevel@tonic-gate /*NOTREACHED*/ 30477c478bd9Sstevel@tonic-gate return (0); 30487c478bd9Sstevel@tonic-gate } 30497c478bd9Sstevel@tonic-gate 30507c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 30517c478bd9Sstevel@tonic-gate void 30527c478bd9Sstevel@tonic-gate kobj_texthole_free(caddr_t addr, size_t size) 30537c478bd9Sstevel@tonic-gate { 30547c478bd9Sstevel@tonic-gate panic("unexpected call to kobj_texthole_free()"); 30557c478bd9Sstevel@tonic-gate } 30567c478bd9Sstevel@tonic-gate 30577c478bd9Sstevel@tonic-gate /* 30587c478bd9Sstevel@tonic-gate * This is called just after configure() in startup(). 30597c478bd9Sstevel@tonic-gate * 30607c478bd9Sstevel@tonic-gate * The ISALIST concept is a bit hopeless on Intel, because 30617c478bd9Sstevel@tonic-gate * there's no guarantee of an ever-more-capable processor 30627c478bd9Sstevel@tonic-gate * given that various parts of the instruction set may appear 30637c478bd9Sstevel@tonic-gate * and disappear between different implementations. 30647c478bd9Sstevel@tonic-gate * 30657c478bd9Sstevel@tonic-gate * While it would be possible to correct it and even enhance 30667c478bd9Sstevel@tonic-gate * it somewhat, the explicit hardware capability bitmask allows 30677c478bd9Sstevel@tonic-gate * more flexibility. 30687c478bd9Sstevel@tonic-gate * 30697c478bd9Sstevel@tonic-gate * So, we just leave this alone. 30707c478bd9Sstevel@tonic-gate */ 30717c478bd9Sstevel@tonic-gate void 30727c478bd9Sstevel@tonic-gate setx86isalist(void) 30737c478bd9Sstevel@tonic-gate { 30747c478bd9Sstevel@tonic-gate char *tp; 30757c478bd9Sstevel@tonic-gate size_t len; 30767c478bd9Sstevel@tonic-gate extern char *isa_list; 30777c478bd9Sstevel@tonic-gate 30787c478bd9Sstevel@tonic-gate #define TBUFSIZE 1024 30797c478bd9Sstevel@tonic-gate 30807c478bd9Sstevel@tonic-gate tp = kmem_alloc(TBUFSIZE, KM_SLEEP); 30817c478bd9Sstevel@tonic-gate *tp = '\0'; 30827c478bd9Sstevel@tonic-gate 30837c478bd9Sstevel@tonic-gate #if defined(__amd64) 30847c478bd9Sstevel@tonic-gate (void) strcpy(tp, "amd64 "); 30857c478bd9Sstevel@tonic-gate #endif 30867c478bd9Sstevel@tonic-gate 30877c478bd9Sstevel@tonic-gate switch (x86_vendor) { 30887c478bd9Sstevel@tonic-gate case X86_VENDOR_Intel: 30897c478bd9Sstevel@tonic-gate case X86_VENDOR_AMD: 30907c478bd9Sstevel@tonic-gate case X86_VENDOR_TM: 30917417cfdeSKuriakose Kuruvilla if (is_x86_feature(x86_featureset, X86FSET_CMOV)) { 30927c478bd9Sstevel@tonic-gate /* 30937c478bd9Sstevel@tonic-gate * Pentium Pro or later 30947c478bd9Sstevel@tonic-gate */ 30957c478bd9Sstevel@tonic-gate (void) strcat(tp, "pentium_pro"); 30967417cfdeSKuriakose Kuruvilla (void) strcat(tp, 30977417cfdeSKuriakose Kuruvilla is_x86_feature(x86_featureset, X86FSET_MMX) ? 30987c478bd9Sstevel@tonic-gate "+mmx pentium_pro " : " "); 30997c478bd9Sstevel@tonic-gate } 31007c478bd9Sstevel@tonic-gate /*FALLTHROUGH*/ 31017c478bd9Sstevel@tonic-gate case X86_VENDOR_Cyrix: 31027c478bd9Sstevel@tonic-gate /* 31037c478bd9Sstevel@tonic-gate * The Cyrix 6x86 does not have any Pentium features 31047c478bd9Sstevel@tonic-gate * accessible while not at privilege level 0. 31057c478bd9Sstevel@tonic-gate */ 31067417cfdeSKuriakose Kuruvilla if (is_x86_feature(x86_featureset, X86FSET_CPUID)) { 31077c478bd9Sstevel@tonic-gate (void) strcat(tp, "pentium"); 31087417cfdeSKuriakose Kuruvilla (void) strcat(tp, 31097417cfdeSKuriakose Kuruvilla is_x86_feature(x86_featureset, X86FSET_MMX) ? 31107c478bd9Sstevel@tonic-gate "+mmx pentium " : " "); 31117c478bd9Sstevel@tonic-gate } 31127c478bd9Sstevel@tonic-gate break; 31137c478bd9Sstevel@tonic-gate default: 31147c478bd9Sstevel@tonic-gate break; 31157c478bd9Sstevel@tonic-gate } 31167c478bd9Sstevel@tonic-gate (void) strcat(tp, "i486 i386 i86"); 31177c478bd9Sstevel@tonic-gate len = strlen(tp) + 1; /* account for NULL at end of string */ 31187c478bd9Sstevel@tonic-gate isa_list = strcpy(kmem_alloc(len, KM_SLEEP), tp); 31197c478bd9Sstevel@tonic-gate kmem_free(tp, TBUFSIZE); 31207c478bd9Sstevel@tonic-gate 31217c478bd9Sstevel@tonic-gate #undef TBUFSIZE 31227c478bd9Sstevel@tonic-gate } 31237c478bd9Sstevel@tonic-gate 31247c478bd9Sstevel@tonic-gate 31257c478bd9Sstevel@tonic-gate #ifdef __amd64 31267c478bd9Sstevel@tonic-gate 31277c478bd9Sstevel@tonic-gate void * 31287c478bd9Sstevel@tonic-gate device_arena_alloc(size_t size, int vm_flag) 31297c478bd9Sstevel@tonic-gate { 31307c478bd9Sstevel@tonic-gate return (vmem_alloc(device_arena, size, vm_flag)); 31317c478bd9Sstevel@tonic-gate } 31327c478bd9Sstevel@tonic-gate 31337c478bd9Sstevel@tonic-gate void 31347c478bd9Sstevel@tonic-gate device_arena_free(void *vaddr, size_t size) 31357c478bd9Sstevel@tonic-gate { 31367c478bd9Sstevel@tonic-gate vmem_free(device_arena, vaddr, size); 31377c478bd9Sstevel@tonic-gate } 31387c478bd9Sstevel@tonic-gate 3139ae115bc7Smrj #else /* __i386 */ 31407c478bd9Sstevel@tonic-gate 31417c478bd9Sstevel@tonic-gate void * 31427c478bd9Sstevel@tonic-gate device_arena_alloc(size_t size, int vm_flag) 31437c478bd9Sstevel@tonic-gate { 31447c478bd9Sstevel@tonic-gate caddr_t vaddr; 31457c478bd9Sstevel@tonic-gate uintptr_t v; 31467c478bd9Sstevel@tonic-gate size_t start; 31477c478bd9Sstevel@tonic-gate size_t end; 31487c478bd9Sstevel@tonic-gate 31497c478bd9Sstevel@tonic-gate vaddr = vmem_alloc(heap_arena, size, vm_flag); 31507c478bd9Sstevel@tonic-gate if (vaddr == NULL) 31517c478bd9Sstevel@tonic-gate return (NULL); 31527c478bd9Sstevel@tonic-gate 31537c478bd9Sstevel@tonic-gate v = (uintptr_t)vaddr; 31547c478bd9Sstevel@tonic-gate ASSERT(v >= kernelbase); 3155ae115bc7Smrj ASSERT(v + size <= valloc_base); 31567c478bd9Sstevel@tonic-gate 31577c478bd9Sstevel@tonic-gate start = btop(v - kernelbase); 31587c478bd9Sstevel@tonic-gate end = btop(v + size - 1 - kernelbase); 31597c478bd9Sstevel@tonic-gate ASSERT(start < toxic_bit_map_len); 31607c478bd9Sstevel@tonic-gate ASSERT(end < toxic_bit_map_len); 31617c478bd9Sstevel@tonic-gate 31627c478bd9Sstevel@tonic-gate while (start <= end) { 31637c478bd9Sstevel@tonic-gate BT_ATOMIC_SET(toxic_bit_map, start); 31647c478bd9Sstevel@tonic-gate ++start; 31657c478bd9Sstevel@tonic-gate } 31667c478bd9Sstevel@tonic-gate return (vaddr); 31677c478bd9Sstevel@tonic-gate } 31687c478bd9Sstevel@tonic-gate 31697c478bd9Sstevel@tonic-gate void 31707c478bd9Sstevel@tonic-gate device_arena_free(void *vaddr, size_t size) 31717c478bd9Sstevel@tonic-gate { 31727c478bd9Sstevel@tonic-gate uintptr_t v = (uintptr_t)vaddr; 31737c478bd9Sstevel@tonic-gate size_t start; 31747c478bd9Sstevel@tonic-gate size_t end; 31757c478bd9Sstevel@tonic-gate 31767c478bd9Sstevel@tonic-gate ASSERT(v >= kernelbase); 3177ae115bc7Smrj ASSERT(v + size <= valloc_base); 31787c478bd9Sstevel@tonic-gate 31797c478bd9Sstevel@tonic-gate start = btop(v - kernelbase); 31807c478bd9Sstevel@tonic-gate end = btop(v + size - 1 - kernelbase); 31817c478bd9Sstevel@tonic-gate ASSERT(start < toxic_bit_map_len); 31827c478bd9Sstevel@tonic-gate ASSERT(end < toxic_bit_map_len); 31837c478bd9Sstevel@tonic-gate 31847c478bd9Sstevel@tonic-gate while (start <= end) { 31857c478bd9Sstevel@tonic-gate ASSERT(BT_TEST(toxic_bit_map, start) != 0); 31867c478bd9Sstevel@tonic-gate BT_ATOMIC_CLEAR(toxic_bit_map, start); 31877c478bd9Sstevel@tonic-gate ++start; 31887c478bd9Sstevel@tonic-gate } 31897c478bd9Sstevel@tonic-gate vmem_free(heap_arena, vaddr, size); 31907c478bd9Sstevel@tonic-gate } 31917c478bd9Sstevel@tonic-gate 31927c478bd9Sstevel@tonic-gate /* 31937c478bd9Sstevel@tonic-gate * returns 1st address in range that is in device arena, or NULL 31947c478bd9Sstevel@tonic-gate * if len is not NULL it returns the length of the toxic range 31957c478bd9Sstevel@tonic-gate */ 31967c478bd9Sstevel@tonic-gate void * 31977c478bd9Sstevel@tonic-gate device_arena_contains(void *vaddr, size_t size, size_t *len) 31987c478bd9Sstevel@tonic-gate { 31997c478bd9Sstevel@tonic-gate uintptr_t v = (uintptr_t)vaddr; 32007c478bd9Sstevel@tonic-gate uintptr_t eaddr = v + size; 32017c478bd9Sstevel@tonic-gate size_t start; 32027c478bd9Sstevel@tonic-gate size_t end; 32037c478bd9Sstevel@tonic-gate 32047c478bd9Sstevel@tonic-gate /* 32057c478bd9Sstevel@tonic-gate * if called very early by kmdb, just return NULL 32067c478bd9Sstevel@tonic-gate */ 32077c478bd9Sstevel@tonic-gate if (toxic_bit_map == NULL) 32087c478bd9Sstevel@tonic-gate return (NULL); 32097c478bd9Sstevel@tonic-gate 32107c478bd9Sstevel@tonic-gate /* 32117c478bd9Sstevel@tonic-gate * First check if we're completely outside the bitmap range. 32127c478bd9Sstevel@tonic-gate */ 3213ae115bc7Smrj if (v >= valloc_base || eaddr < kernelbase) 32147c478bd9Sstevel@tonic-gate return (NULL); 32157c478bd9Sstevel@tonic-gate 32167c478bd9Sstevel@tonic-gate /* 32177c478bd9Sstevel@tonic-gate * Trim ends of search to look at only what the bitmap covers. 32187c478bd9Sstevel@tonic-gate */ 32197c478bd9Sstevel@tonic-gate if (v < kernelbase) 32207c478bd9Sstevel@tonic-gate v = kernelbase; 32217c478bd9Sstevel@tonic-gate start = btop(v - kernelbase); 32227c478bd9Sstevel@tonic-gate end = btop(eaddr - kernelbase); 32237c478bd9Sstevel@tonic-gate if (end >= toxic_bit_map_len) 32247c478bd9Sstevel@tonic-gate end = toxic_bit_map_len; 32257c478bd9Sstevel@tonic-gate 32267c478bd9Sstevel@tonic-gate if (bt_range(toxic_bit_map, &start, &end, end) == 0) 32277c478bd9Sstevel@tonic-gate return (NULL); 32287c478bd9Sstevel@tonic-gate 32297c478bd9Sstevel@tonic-gate v = kernelbase + ptob(start); 32307c478bd9Sstevel@tonic-gate if (len != NULL) 32317c478bd9Sstevel@tonic-gate *len = ptob(end - start); 32327c478bd9Sstevel@tonic-gate return ((void *)v); 32337c478bd9Sstevel@tonic-gate } 32347c478bd9Sstevel@tonic-gate 3235ae115bc7Smrj #endif /* __i386 */ 3236