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. 239e3e4df2SGarrett D'Amore * Copyright 2012 DEY Storage Systems, Inc. All rights reserved. 2479ec9da8SYuri Pankov * Copyright 2013 Nexenta Systems, Inc. All rights reserved. 25*3ce2fcdcSRobert Mustacchi * Copyright 2015 Joyent, Inc. 267c478bd9Sstevel@tonic-gate */ 27a3114836SGerry Liu /* 28a3114836SGerry Liu * Copyright (c) 2010, Intel Corporation. 29a3114836SGerry Liu * All rights reserved. 30a3114836SGerry Liu */ 317c478bd9Sstevel@tonic-gate 327c478bd9Sstevel@tonic-gate #include <sys/types.h> 337c478bd9Sstevel@tonic-gate #include <sys/t_lock.h> 347c478bd9Sstevel@tonic-gate #include <sys/param.h> 357c478bd9Sstevel@tonic-gate #include <sys/sysmacros.h> 367c478bd9Sstevel@tonic-gate #include <sys/signal.h> 377c478bd9Sstevel@tonic-gate #include <sys/systm.h> 387c478bd9Sstevel@tonic-gate #include <sys/user.h> 397c478bd9Sstevel@tonic-gate #include <sys/mman.h> 407c478bd9Sstevel@tonic-gate #include <sys/vm.h> 417c478bd9Sstevel@tonic-gate #include <sys/conf.h> 427c478bd9Sstevel@tonic-gate #include <sys/avintr.h> 437c478bd9Sstevel@tonic-gate #include <sys/autoconf.h> 447c478bd9Sstevel@tonic-gate #include <sys/disp.h> 457c478bd9Sstevel@tonic-gate #include <sys/class.h> 467c478bd9Sstevel@tonic-gate #include <sys/bitmap.h> 477c478bd9Sstevel@tonic-gate 487c478bd9Sstevel@tonic-gate #include <sys/privregs.h> 497c478bd9Sstevel@tonic-gate 507c478bd9Sstevel@tonic-gate #include <sys/proc.h> 517c478bd9Sstevel@tonic-gate #include <sys/buf.h> 527c478bd9Sstevel@tonic-gate #include <sys/kmem.h> 53ae115bc7Smrj #include <sys/mem.h> 547c478bd9Sstevel@tonic-gate #include <sys/kstat.h> 557c478bd9Sstevel@tonic-gate 567c478bd9Sstevel@tonic-gate #include <sys/reboot.h> 577c478bd9Sstevel@tonic-gate 587c478bd9Sstevel@tonic-gate #include <sys/cred.h> 597c478bd9Sstevel@tonic-gate #include <sys/vnode.h> 607c478bd9Sstevel@tonic-gate #include <sys/file.h> 617c478bd9Sstevel@tonic-gate 627c478bd9Sstevel@tonic-gate #include <sys/procfs.h> 637c478bd9Sstevel@tonic-gate 647c478bd9Sstevel@tonic-gate #include <sys/vfs.h> 657c478bd9Sstevel@tonic-gate #include <sys/cmn_err.h> 667c478bd9Sstevel@tonic-gate #include <sys/utsname.h> 677c478bd9Sstevel@tonic-gate #include <sys/debug.h> 687c478bd9Sstevel@tonic-gate #include <sys/kdi.h> 697c478bd9Sstevel@tonic-gate 707c478bd9Sstevel@tonic-gate #include <sys/dumphdr.h> 717c478bd9Sstevel@tonic-gate #include <sys/bootconf.h> 72a3114836SGerry Liu #include <sys/memlist_plat.h> 737c478bd9Sstevel@tonic-gate #include <sys/varargs.h> 747c478bd9Sstevel@tonic-gate #include <sys/promif.h> 75843e1988Sjohnlev #include <sys/modctl.h> 767c478bd9Sstevel@tonic-gate 777c478bd9Sstevel@tonic-gate #include <sys/sunddi.h> 787c478bd9Sstevel@tonic-gate #include <sys/sunndi.h> 797c478bd9Sstevel@tonic-gate #include <sys/ndi_impldefs.h> 807c478bd9Sstevel@tonic-gate #include <sys/ddidmareq.h> 817c478bd9Sstevel@tonic-gate #include <sys/psw.h> 827c478bd9Sstevel@tonic-gate #include <sys/regset.h> 837c478bd9Sstevel@tonic-gate #include <sys/clock.h> 847c478bd9Sstevel@tonic-gate #include <sys/pte.h> 857c478bd9Sstevel@tonic-gate #include <sys/tss.h> 867c478bd9Sstevel@tonic-gate #include <sys/stack.h> 877c478bd9Sstevel@tonic-gate #include <sys/trap.h> 887c478bd9Sstevel@tonic-gate #include <sys/fp.h> 891d03c31eSjohnlev #include <vm/kboot_mmu.h> 907c478bd9Sstevel@tonic-gate #include <vm/anon.h> 917c478bd9Sstevel@tonic-gate #include <vm/as.h> 927c478bd9Sstevel@tonic-gate #include <vm/page.h> 937c478bd9Sstevel@tonic-gate #include <vm/seg.h> 947c478bd9Sstevel@tonic-gate #include <vm/seg_dev.h> 957c478bd9Sstevel@tonic-gate #include <vm/seg_kmem.h> 967c478bd9Sstevel@tonic-gate #include <vm/seg_kpm.h> 977c478bd9Sstevel@tonic-gate #include <vm/seg_map.h> 987c478bd9Sstevel@tonic-gate #include <vm/seg_vn.h> 997c478bd9Sstevel@tonic-gate #include <vm/seg_kp.h> 1007c478bd9Sstevel@tonic-gate #include <sys/memnode.h> 1017c478bd9Sstevel@tonic-gate #include <vm/vm_dep.h> 1027c478bd9Sstevel@tonic-gate #include <sys/thread.h> 1037c478bd9Sstevel@tonic-gate #include <sys/sysconf.h> 1047c478bd9Sstevel@tonic-gate #include <sys/vm_machparam.h> 1057c478bd9Sstevel@tonic-gate #include <sys/archsystm.h> 1067c478bd9Sstevel@tonic-gate #include <sys/machsystm.h> 1077c478bd9Sstevel@tonic-gate #include <vm/hat.h> 1087c478bd9Sstevel@tonic-gate #include <vm/hat_i86.h> 1097c478bd9Sstevel@tonic-gate #include <sys/pmem.h> 1107c478bd9Sstevel@tonic-gate #include <sys/smp_impldefs.h> 1117c478bd9Sstevel@tonic-gate #include <sys/x86_archext.h> 112a563a037Sbholler #include <sys/cpuvar.h> 1137c478bd9Sstevel@tonic-gate #include <sys/segments.h> 1147c478bd9Sstevel@tonic-gate #include <sys/clconf.h> 1157c478bd9Sstevel@tonic-gate #include <sys/kobj.h> 1167c478bd9Sstevel@tonic-gate #include <sys/kobj_lex.h> 1177c478bd9Sstevel@tonic-gate #include <sys/cpc_impl.h> 1187aec1d6eScindi #include <sys/cpu_module.h> 11984ab085aSmws #include <sys/smbios.h> 120ae115bc7Smrj #include <sys/debug_info.h> 1211d03c31eSjohnlev #include <sys/bootinfo.h> 122a288e5a9SJoshua M. Clulow #include <sys/ddi_periodic.h> 1235679c89fSjv #include <sys/systeminfo.h> 12419397407SSherry Moore #include <sys/multiboot.h> 1250181461bSKeith M Wesolowski #include <sys/ramdisk.h> 126ae115bc7Smrj 127a3114836SGerry Liu #ifdef __xpv 1281d03c31eSjohnlev 129843e1988Sjohnlev #include <sys/hypervisor.h> 130843e1988Sjohnlev #include <sys/xen_mmu.h> 131843e1988Sjohnlev #include <sys/evtchn_impl.h> 132843e1988Sjohnlev #include <sys/gnttab.h> 133843e1988Sjohnlev #include <sys/xpv_panic.h> 134843e1988Sjohnlev #include <xen/sys/xenbus_comms.h> 135843e1988Sjohnlev #include <xen/public/physdev.h> 1361d03c31eSjohnlev 137843e1988Sjohnlev extern void xen_late_startup(void); 138ae115bc7Smrj 1391d03c31eSjohnlev struct xen_evt_data cpu0_evt_data; 1401d03c31eSjohnlev 141a3114836SGerry Liu #else /* __xpv */ 142a3114836SGerry Liu #include <sys/memlist_impl.h> 143a3114836SGerry Liu 144a3114836SGerry Liu extern void mem_config_init(void); 1451d03c31eSjohnlev #endif /* __xpv */ 1467c478bd9Sstevel@tonic-gate 1477c478bd9Sstevel@tonic-gate extern void progressbar_init(void); 1489acbbeafSnn extern void brand_init(void); 14906fb6a36Sdv extern void pcf_init(void); 1500e751525SEric Saxe extern void pg_init(void); 1516a59053bSAlex Wilson extern void ssp_init(void); 1527c478bd9Sstevel@tonic-gate 153d7d93655Sblakej extern int size_pse_array(pgcnt_t, int); 154d7d93655Sblakej 155ed5289f9SKen Erickson #if defined(_SOFT_HOSTID) 156ed5289f9SKen Erickson 157ed5289f9SKen Erickson #include <sys/rtc.h> 158ed5289f9SKen Erickson 159ed5289f9SKen Erickson static int32_t set_soft_hostid(void); 160ed5289f9SKen Erickson static char hostid_file[] = "/etc/hostid"; 161ed5289f9SKen Erickson 162ed5289f9SKen Erickson #endif 163ed5289f9SKen Erickson 16494f1124eSVikram Hegde void *gfx_devinfo_list; 165ed5289f9SKen Erickson 1663a634bfcSVikram Hegde #if defined(__amd64) && !defined(__xpv) 1673a634bfcSVikram Hegde extern void immu_startup(void); 1683a634bfcSVikram Hegde #endif 1693a634bfcSVikram Hegde 1707c478bd9Sstevel@tonic-gate /* 1717c478bd9Sstevel@tonic-gate * XXX make declaration below "static" when drivers no longer use this 1727c478bd9Sstevel@tonic-gate * interface. 1737c478bd9Sstevel@tonic-gate */ 1747c478bd9Sstevel@tonic-gate extern caddr_t p0_va; /* Virtual address for accessing physical page 0 */ 1757c478bd9Sstevel@tonic-gate 1767c478bd9Sstevel@tonic-gate /* 1777c478bd9Sstevel@tonic-gate * segkp 1787c478bd9Sstevel@tonic-gate */ 1797c478bd9Sstevel@tonic-gate extern int segkp_fromheap; 1807c478bd9Sstevel@tonic-gate 1817c478bd9Sstevel@tonic-gate static void kvm_init(void); 1827c478bd9Sstevel@tonic-gate static void startup_init(void); 1837c478bd9Sstevel@tonic-gate static void startup_memlist(void); 184ae115bc7Smrj static void startup_kmem(void); 1857c478bd9Sstevel@tonic-gate static void startup_modules(void); 1867c478bd9Sstevel@tonic-gate static void startup_vm(void); 1877c478bd9Sstevel@tonic-gate static void startup_end(void); 18835b1ab99Sjosephb static void layout_kernel_va(void); 1897c478bd9Sstevel@tonic-gate 1907c478bd9Sstevel@tonic-gate /* 1917c478bd9Sstevel@tonic-gate * Declare these as initialized data so we can patch them. 1927c478bd9Sstevel@tonic-gate */ 193f53ad214Skchow #ifdef __i386 19435b1ab99Sjosephb 195f53ad214Skchow /* 196f53ad214Skchow * Due to virtual address space limitations running in 32 bit mode, restrict 19735b1ab99Sjosephb * the amount of physical memory configured to a max of PHYSMEM pages (16g). 198f53ad214Skchow * 199f53ad214Skchow * If the physical max memory size of 64g were allowed to be configured, the 200f53ad214Skchow * size of user virtual address space will be less than 1g. A limited user 201f53ad214Skchow * address space greatly reduces the range of applications that can run. 202f53ad214Skchow * 20335b1ab99Sjosephb * If more physical memory than PHYSMEM is required, users should preferably 20435b1ab99Sjosephb * run in 64 bit mode which has far looser virtual address space limitations. 205f53ad214Skchow * 206f53ad214Skchow * If 64 bit mode is not available (as in IA32) and/or more physical memory 20735b1ab99Sjosephb * than PHYSMEM is required in 32 bit mode, physmem can be set to the desired 208f53ad214Skchow * value or to 0 (to configure all available memory) via eeprom(1M). kernelbase 209f53ad214Skchow * should also be carefully tuned to balance out the need of the user 210f53ad214Skchow * application while minimizing the risk of kernel heap exhaustion due to 211f53ad214Skchow * kernelbase being set too high. 212f53ad214Skchow */ 21335b1ab99Sjosephb #define PHYSMEM 0x400000 214f53ad214Skchow 21535b1ab99Sjosephb #else /* __amd64 */ 21635b1ab99Sjosephb 21735b1ab99Sjosephb /* 21835b1ab99Sjosephb * For now we can handle memory with physical addresses up to about 21935b1ab99Sjosephb * 64 Terabytes. This keeps the kernel above the VA hole, leaving roughly 22035b1ab99Sjosephb * half the VA space for seg_kpm. When systems get bigger than 64TB this 22135b1ab99Sjosephb * code will need revisiting. There is an implicit assumption that there 22235b1ab99Sjosephb * are no *huge* holes in the physical address space too. 22335b1ab99Sjosephb */ 22435b1ab99Sjosephb #define TERABYTE (1ul << 40) 22535b1ab99Sjosephb #define PHYSMEM_MAX64 mmu_btop(64 * TERABYTE) 22635b1ab99Sjosephb #define PHYSMEM PHYSMEM_MAX64 22735b1ab99Sjosephb #define AMD64_VA_HOLE_END 0xFFFF800000000000ul 22835b1ab99Sjosephb 22935b1ab99Sjosephb #endif /* __amd64 */ 23035b1ab99Sjosephb 23135b1ab99Sjosephb pgcnt_t physmem = PHYSMEM; 2327c478bd9Sstevel@tonic-gate pgcnt_t obp_pages; /* Memory used by PROM for its text and data */ 2337c478bd9Sstevel@tonic-gate 2347c478bd9Sstevel@tonic-gate char *kobj_file_buf; 2357c478bd9Sstevel@tonic-gate int kobj_file_bufsize; /* set in /etc/system */ 2367c478bd9Sstevel@tonic-gate 2377c478bd9Sstevel@tonic-gate /* Global variables for MP support. Used in mp_startup */ 23819397407SSherry Moore caddr_t rm_platter_va = 0; 2397c478bd9Sstevel@tonic-gate uint32_t rm_platter_pa; 2407c478bd9Sstevel@tonic-gate 24183f9b804Skchow int auto_lpg_disable = 1; 24283f9b804Skchow 2437c478bd9Sstevel@tonic-gate /* 2447c478bd9Sstevel@tonic-gate * Some CPUs have holes in the middle of the 64-bit virtual address range. 2457c478bd9Sstevel@tonic-gate */ 2467c478bd9Sstevel@tonic-gate uintptr_t hole_start, hole_end; 2477c478bd9Sstevel@tonic-gate 2487c478bd9Sstevel@tonic-gate /* 2497c478bd9Sstevel@tonic-gate * kpm mapping window 2507c478bd9Sstevel@tonic-gate */ 2517c478bd9Sstevel@tonic-gate caddr_t kpm_vbase; 2527c478bd9Sstevel@tonic-gate size_t kpm_size; 25335b1ab99Sjosephb static int kpm_desired; 25435b1ab99Sjosephb #ifdef __amd64 25535b1ab99Sjosephb static uintptr_t segkpm_base = (uintptr_t)SEGKPM_BASE; 25635b1ab99Sjosephb #endif 2577c478bd9Sstevel@tonic-gate 2587c478bd9Sstevel@tonic-gate /* 2597c478bd9Sstevel@tonic-gate * Configuration parameters set at boot time. 2607c478bd9Sstevel@tonic-gate */ 2617c478bd9Sstevel@tonic-gate 2627c478bd9Sstevel@tonic-gate caddr_t econtig; /* end of first block of contiguous kernel */ 2637c478bd9Sstevel@tonic-gate 2647c478bd9Sstevel@tonic-gate struct bootops *bootops = 0; /* passed in from boot */ 2657c478bd9Sstevel@tonic-gate struct bootops **bootopsp; 2667c478bd9Sstevel@tonic-gate struct boot_syscalls *sysp; /* passed in from boot */ 2677c478bd9Sstevel@tonic-gate 2687c478bd9Sstevel@tonic-gate char bootblock_fstype[16]; 2697c478bd9Sstevel@tonic-gate 2707c478bd9Sstevel@tonic-gate char kern_bootargs[OBP_MAXPATHLEN]; 27119397407SSherry Moore char kern_bootfile[OBP_MAXPATHLEN]; 2727c478bd9Sstevel@tonic-gate 273ad23a2dbSjohansen /* 274ad23a2dbSjohansen * ZFS zio segment. This allows us to exclude large portions of ZFS data that 275ad23a2dbSjohansen * gets cached in kmem caches on the heap. If this is set to zero, we allocate 276ad23a2dbSjohansen * zio buffers from their own segment, otherwise they are allocated from the 277ad23a2dbSjohansen * heap. The optimization of allocating zio buffers from their own segment is 278ad23a2dbSjohansen * only valid on 64-bit kernels. 279ad23a2dbSjohansen */ 280ad23a2dbSjohansen #if defined(__amd64) 281ad23a2dbSjohansen int segzio_fromheap = 0; 282ad23a2dbSjohansen #else 283ad23a2dbSjohansen int segzio_fromheap = 1; 284ad23a2dbSjohansen #endif 285ad23a2dbSjohansen 286*3ce2fcdcSRobert Mustacchi /* 287*3ce2fcdcSRobert Mustacchi * Give folks an escape hatch for disabling SMAP via kmdb. Doesn't work 288*3ce2fcdcSRobert Mustacchi * post-boot. 289*3ce2fcdcSRobert Mustacchi */ 290*3ce2fcdcSRobert Mustacchi int disable_smap = 0; 291*3ce2fcdcSRobert Mustacchi 2927c478bd9Sstevel@tonic-gate /* 2937c478bd9Sstevel@tonic-gate * new memory fragmentations are possible in startup() due to BOP_ALLOCs. this 2947c478bd9Sstevel@tonic-gate * depends on number of BOP_ALLOC calls made and requested size, memory size 2957c478bd9Sstevel@tonic-gate * combination and whether boot.bin memory needs to be freed. 2967c478bd9Sstevel@tonic-gate */ 2977c478bd9Sstevel@tonic-gate #define POSS_NEW_FRAGMENTS 12 2987c478bd9Sstevel@tonic-gate 2997c478bd9Sstevel@tonic-gate /* 3007c478bd9Sstevel@tonic-gate * VM data structures 3017c478bd9Sstevel@tonic-gate */ 3027c478bd9Sstevel@tonic-gate long page_hashsz; /* Size of page hash table (power of two) */ 303cb15d5d9SPeter Rival unsigned int page_hashsz_shift; /* log2(page_hashsz) */ 3047c478bd9Sstevel@tonic-gate struct page *pp_base; /* Base of initial system page struct array */ 3057c478bd9Sstevel@tonic-gate struct page **page_hash; /* Page hash table */ 306d7d93655Sblakej pad_mutex_t *pse_mutex; /* Locks protecting pp->p_selock */ 307d7d93655Sblakej size_t pse_table_size; /* Number of mutexes in pse_mutex[] */ 308d7d93655Sblakej int pse_shift; /* log2(pse_table_size) */ 3097c478bd9Sstevel@tonic-gate struct seg ktextseg; /* Segment used for kernel executable image */ 3107c478bd9Sstevel@tonic-gate struct seg kvalloc; /* Segment used for "valloc" mapping */ 3117c478bd9Sstevel@tonic-gate struct seg kpseg; /* Segment used for pageable kernel virt mem */ 3127c478bd9Sstevel@tonic-gate struct seg kmapseg; /* Segment used for generic kernel mappings */ 3137c478bd9Sstevel@tonic-gate struct seg kdebugseg; /* Segment used for the kernel debugger */ 3147c478bd9Sstevel@tonic-gate 3157c478bd9Sstevel@tonic-gate struct seg *segkmap = &kmapseg; /* Kernel generic mapping segment */ 316ae115bc7Smrj static struct seg *segmap = &kmapseg; /* easier to use name for in here */ 317ae115bc7Smrj 3187c478bd9Sstevel@tonic-gate struct seg *segkp = &kpseg; /* Pageable kernel virtual memory segment */ 3197c478bd9Sstevel@tonic-gate 3207c478bd9Sstevel@tonic-gate #if defined(__amd64) 3217c478bd9Sstevel@tonic-gate struct seg kvseg_core; /* Segment used for the core heap */ 3227c478bd9Sstevel@tonic-gate struct seg kpmseg; /* Segment used for physical mapping */ 3237c478bd9Sstevel@tonic-gate struct seg *segkpm = &kpmseg; /* 64bit kernel physical mapping segment */ 3247c478bd9Sstevel@tonic-gate #else 3257c478bd9Sstevel@tonic-gate struct seg *segkpm = NULL; /* Unused on IA32 */ 3267c478bd9Sstevel@tonic-gate #endif 3277c478bd9Sstevel@tonic-gate 3287c478bd9Sstevel@tonic-gate caddr_t segkp_base; /* Base address of segkp */ 329ad23a2dbSjohansen caddr_t segzio_base; /* Base address of segzio */ 3307c478bd9Sstevel@tonic-gate #if defined(__amd64) 3317c478bd9Sstevel@tonic-gate pgcnt_t segkpsize = btop(SEGKPDEFSIZE); /* size of segkp segment in pages */ 3327c478bd9Sstevel@tonic-gate #else 3337c478bd9Sstevel@tonic-gate pgcnt_t segkpsize = 0; 3347c478bd9Sstevel@tonic-gate #endif 335ad23a2dbSjohansen pgcnt_t segziosize = 0; /* size of zio segment in pages */ 3367c478bd9Sstevel@tonic-gate 337a3114836SGerry Liu /* 338a3114836SGerry Liu * A static DR page_t VA map is reserved that can map the page structures 339a3114836SGerry Liu * for a domain's entire RA space. The pages that back this space are 340a3114836SGerry Liu * dynamically allocated and need not be physically contiguous. The DR 341a3114836SGerry Liu * map size is derived from KPM size. 342a3114836SGerry Liu * This mechanism isn't used by x86 yet, so just stubs here. 343a3114836SGerry Liu */ 344a3114836SGerry Liu int ppvm_enable = 0; /* Static virtual map for page structs */ 345a3114836SGerry Liu page_t *ppvm_base = NULL; /* Base of page struct map */ 346a3114836SGerry Liu pgcnt_t ppvm_size = 0; /* Size of page struct map */ 347a3114836SGerry Liu 348ae115bc7Smrj /* 349ae115bc7Smrj * VA range available to the debugger 350ae115bc7Smrj */ 351ae115bc7Smrj const caddr_t kdi_segdebugbase = (const caddr_t)SEGDEBUGBASE; 352ae115bc7Smrj const size_t kdi_segdebugsize = SEGDEBUGSIZE; 353ae115bc7Smrj 3547c478bd9Sstevel@tonic-gate struct memseg *memseg_base; 3557c478bd9Sstevel@tonic-gate struct vnode unused_pages_vp; 3567c478bd9Sstevel@tonic-gate 3577c478bd9Sstevel@tonic-gate #define FOURGB 0x100000000LL 3587c478bd9Sstevel@tonic-gate 3597c478bd9Sstevel@tonic-gate struct memlist *memlist; 3607c478bd9Sstevel@tonic-gate 3617c478bd9Sstevel@tonic-gate caddr_t s_text; /* start of kernel text segment */ 3627c478bd9Sstevel@tonic-gate caddr_t e_text; /* end of kernel text segment */ 3637c478bd9Sstevel@tonic-gate caddr_t s_data; /* start of kernel data segment */ 3647c478bd9Sstevel@tonic-gate caddr_t e_data; /* end of kernel data segment */ 3657c478bd9Sstevel@tonic-gate caddr_t modtext; /* start of loadable module text reserved */ 3667c478bd9Sstevel@tonic-gate caddr_t e_modtext; /* end of loadable module text reserved */ 3677c478bd9Sstevel@tonic-gate caddr_t moddata; /* start of loadable module data reserved */ 3687c478bd9Sstevel@tonic-gate caddr_t e_moddata; /* end of loadable module data reserved */ 3697c478bd9Sstevel@tonic-gate 3707c478bd9Sstevel@tonic-gate struct memlist *phys_install; /* Total installed physical memory */ 3717c478bd9Sstevel@tonic-gate struct memlist *phys_avail; /* Total available physical memory */ 3721de082f7SVikram Hegde struct memlist *bios_rsvd; /* Bios reserved memory */ 3737c478bd9Sstevel@tonic-gate 3747c478bd9Sstevel@tonic-gate /* 3757c478bd9Sstevel@tonic-gate * kphysm_init returns the number of pages that were processed 3767c478bd9Sstevel@tonic-gate */ 377ae115bc7Smrj static pgcnt_t kphysm_init(page_t *, pgcnt_t); 3787c478bd9Sstevel@tonic-gate 3797c478bd9Sstevel@tonic-gate #define IO_PROP_SIZE 64 /* device property size */ 3807c478bd9Sstevel@tonic-gate 3817c478bd9Sstevel@tonic-gate /* 3827c478bd9Sstevel@tonic-gate * a couple useful roundup macros 3837c478bd9Sstevel@tonic-gate */ 3847c478bd9Sstevel@tonic-gate #define ROUND_UP_PAGE(x) \ 3857c478bd9Sstevel@tonic-gate ((uintptr_t)P2ROUNDUP((uintptr_t)(x), (uintptr_t)MMU_PAGESIZE)) 3867c478bd9Sstevel@tonic-gate #define ROUND_UP_LPAGE(x) \ 3877c478bd9Sstevel@tonic-gate ((uintptr_t)P2ROUNDUP((uintptr_t)(x), mmu.level_size[1])) 3887c478bd9Sstevel@tonic-gate #define ROUND_UP_4MEG(x) \ 389ae115bc7Smrj ((uintptr_t)P2ROUNDUP((uintptr_t)(x), (uintptr_t)FOUR_MEG)) 3907c478bd9Sstevel@tonic-gate #define ROUND_UP_TOPLEVEL(x) \ 3917c478bd9Sstevel@tonic-gate ((uintptr_t)P2ROUNDUP((uintptr_t)(x), mmu.level_size[mmu.max_level])) 3927c478bd9Sstevel@tonic-gate 3937c478bd9Sstevel@tonic-gate /* 3947c478bd9Sstevel@tonic-gate * 32-bit Kernel's Virtual memory layout. 3957c478bd9Sstevel@tonic-gate * +-----------------------+ 396ae115bc7Smrj * | | 3977c478bd9Sstevel@tonic-gate * 0xFFC00000 -|-----------------------|- ARGSBASE 3987c478bd9Sstevel@tonic-gate * | debugger | 3997c478bd9Sstevel@tonic-gate * 0xFF800000 -|-----------------------|- SEGDEBUGBASE 4007c478bd9Sstevel@tonic-gate * | Kernel Data | 4017c478bd9Sstevel@tonic-gate * 0xFEC00000 -|-----------------------| 4027c478bd9Sstevel@tonic-gate * | Kernel Text | 403843e1988Sjohnlev * 0xFE800000 -|-----------------------|- KERNEL_TEXT (0xFB400000 on Xen) 404ae115bc7Smrj * |--- GDT ---|- GDT page (GDT_VA) 405ae115bc7Smrj * |--- debug info ---|- debug info (DEBUG_INFO_VA) 406ae115bc7Smrj * | | 407ae115bc7Smrj * | page_t structures | 4087c478bd9Sstevel@tonic-gate * | memsegs, memlists, | 4097c478bd9Sstevel@tonic-gate * | page hash, etc. | 410ae115bc7Smrj * --- -|-----------------------|- ekernelheap, valloc_base (floating) 411ae115bc7Smrj * | | (segkp is just an arena in the heap) 4127c478bd9Sstevel@tonic-gate * | | 4137c478bd9Sstevel@tonic-gate * | kvseg | 4147c478bd9Sstevel@tonic-gate * | | 4157c478bd9Sstevel@tonic-gate * | | 4167c478bd9Sstevel@tonic-gate * --- -|-----------------------|- kernelheap (floating) 4177c478bd9Sstevel@tonic-gate * | Segkmap | 418ae115bc7Smrj * 0xC3002000 -|-----------------------|- segmap_start (floating) 4197c478bd9Sstevel@tonic-gate * | Red Zone | 4207c478bd9Sstevel@tonic-gate * 0xC3000000 -|-----------------------|- kernelbase / userlimit (floating) 4217c478bd9Sstevel@tonic-gate * | | || 4227c478bd9Sstevel@tonic-gate * | Shared objects | \/ 4237c478bd9Sstevel@tonic-gate * | | 4247c478bd9Sstevel@tonic-gate * : : 4257c478bd9Sstevel@tonic-gate * | user data | 4267c478bd9Sstevel@tonic-gate * |-----------------------| 4277c478bd9Sstevel@tonic-gate * | user text | 4287c478bd9Sstevel@tonic-gate * 0x08048000 -|-----------------------| 4297c478bd9Sstevel@tonic-gate * | user stack | 4307c478bd9Sstevel@tonic-gate * : : 4317c478bd9Sstevel@tonic-gate * | invalid | 4327c478bd9Sstevel@tonic-gate * 0x00000000 +-----------------------+ 4337c478bd9Sstevel@tonic-gate * 4347c478bd9Sstevel@tonic-gate * 4357c478bd9Sstevel@tonic-gate * 64-bit Kernel's Virtual memory layout. (assuming 64 bit app) 4367c478bd9Sstevel@tonic-gate * +-----------------------+ 437ae115bc7Smrj * | | 4387c478bd9Sstevel@tonic-gate * 0xFFFFFFFF.FFC00000 |-----------------------|- ARGSBASE 4397c478bd9Sstevel@tonic-gate * | debugger (?) | 4407c478bd9Sstevel@tonic-gate * 0xFFFFFFFF.FF800000 |-----------------------|- SEGDEBUGBASE 4417c478bd9Sstevel@tonic-gate * | unused | 4427c478bd9Sstevel@tonic-gate * +-----------------------+ 4437c478bd9Sstevel@tonic-gate * | Kernel Data | 4447c478bd9Sstevel@tonic-gate * 0xFFFFFFFF.FBC00000 |-----------------------| 4457c478bd9Sstevel@tonic-gate * | Kernel Text | 4467c478bd9Sstevel@tonic-gate * 0xFFFFFFFF.FB800000 |-----------------------|- KERNEL_TEXT 447ae115bc7Smrj * |--- GDT ---|- GDT page (GDT_VA) 448ae115bc7Smrj * |--- debug info ---|- debug info (DEBUG_INFO_VA) 449ae115bc7Smrj * | | 4507c478bd9Sstevel@tonic-gate * | Core heap | (used for loadable modules) 4517c478bd9Sstevel@tonic-gate * 0xFFFFFFFF.C0000000 |-----------------------|- core_base / ekernelheap 4527c478bd9Sstevel@tonic-gate * | Kernel | 4537c478bd9Sstevel@tonic-gate * | heap | 4547c478bd9Sstevel@tonic-gate * 0xFFFFFXXX.XXX00000 |-----------------------|- kernelheap (floating) 455ae115bc7Smrj * | segmap | 456ae115bc7Smrj * 0xFFFFFXXX.XXX00000 |-----------------------|- segmap_start (floating) 4577c478bd9Sstevel@tonic-gate * | device mappings | 4587c478bd9Sstevel@tonic-gate * 0xFFFFFXXX.XXX00000 |-----------------------|- toxic_addr (floating) 459ad23a2dbSjohansen * | segzio | 460ad23a2dbSjohansen * 0xFFFFFXXX.XXX00000 |-----------------------|- segzio_base (floating) 4617c478bd9Sstevel@tonic-gate * | segkp | 462ae115bc7Smrj * --- |-----------------------|- segkp_base (floating) 463ae115bc7Smrj * | page_t structures | valloc_base + valloc_sz 464ae115bc7Smrj * | memsegs, memlists, | 465ae115bc7Smrj * | page hash, etc. | 46635b1ab99Sjosephb * 0xFFFFFF00.00000000 |-----------------------|- valloc_base (lower if > 1TB) 4677c478bd9Sstevel@tonic-gate * | segkpm | 4687c478bd9Sstevel@tonic-gate * 0xFFFFFE00.00000000 |-----------------------| 4697c478bd9Sstevel@tonic-gate * | Red Zone | 47035b1ab99Sjosephb * 0xFFFFFD80.00000000 |-----------------------|- KERNELBASE (lower if > 1TB) 4717c478bd9Sstevel@tonic-gate * | User stack |- User space memory 4727c478bd9Sstevel@tonic-gate * | | 4737c478bd9Sstevel@tonic-gate * | shared objects, etc | (grows downwards) 4747c478bd9Sstevel@tonic-gate * : : 4757c478bd9Sstevel@tonic-gate * | | 4767c478bd9Sstevel@tonic-gate * 0xFFFF8000.00000000 |-----------------------| 4777c478bd9Sstevel@tonic-gate * | | 4787c478bd9Sstevel@tonic-gate * | VA Hole / unused | 4797c478bd9Sstevel@tonic-gate * | | 4807c478bd9Sstevel@tonic-gate * 0x00008000.00000000 |-----------------------| 4817c478bd9Sstevel@tonic-gate * | | 4827c478bd9Sstevel@tonic-gate * | | 4837c478bd9Sstevel@tonic-gate * : : 4847c478bd9Sstevel@tonic-gate * | user heap | (grows upwards) 4857c478bd9Sstevel@tonic-gate * | | 4867c478bd9Sstevel@tonic-gate * | user data | 4877c478bd9Sstevel@tonic-gate * |-----------------------| 4887c478bd9Sstevel@tonic-gate * | user text | 4897c478bd9Sstevel@tonic-gate * 0x00000000.04000000 |-----------------------| 4907c478bd9Sstevel@tonic-gate * | invalid | 4917c478bd9Sstevel@tonic-gate * 0x00000000.00000000 +-----------------------+ 4927c478bd9Sstevel@tonic-gate * 4937c478bd9Sstevel@tonic-gate * A 32 bit app on the 64 bit kernel sees the same layout as on the 32 bit 4947c478bd9Sstevel@tonic-gate * kernel, except that userlimit is raised to 0xfe000000 4957c478bd9Sstevel@tonic-gate * 4967c478bd9Sstevel@tonic-gate * Floating values: 4977c478bd9Sstevel@tonic-gate * 4987c478bd9Sstevel@tonic-gate * valloc_base: start of the kernel's memory management/tracking data 499ae115bc7Smrj * structures. This region contains page_t structures for 500ae115bc7Smrj * physical memory, memsegs, memlists, and the page hash. 5017c478bd9Sstevel@tonic-gate * 5027c478bd9Sstevel@tonic-gate * core_base: start of the kernel's "core" heap area on 64-bit systems. 5037c478bd9Sstevel@tonic-gate * This area is intended to be used for global data as well as for module 5047c478bd9Sstevel@tonic-gate * text/data that does not fit into the nucleus pages. The core heap is 5057c478bd9Sstevel@tonic-gate * restricted to a 2GB range, allowing every address within it to be 5067c478bd9Sstevel@tonic-gate * accessed using rip-relative addressing 5077c478bd9Sstevel@tonic-gate * 5087c478bd9Sstevel@tonic-gate * ekernelheap: end of kernelheap and start of segmap. 5097c478bd9Sstevel@tonic-gate * 5107c478bd9Sstevel@tonic-gate * kernelheap: start of kernel heap. On 32-bit systems, this starts right 5117c478bd9Sstevel@tonic-gate * above a red zone that separates the user's address space from the 5127c478bd9Sstevel@tonic-gate * kernel's. On 64-bit systems, it sits above segkp and segkpm. 5137c478bd9Sstevel@tonic-gate * 514ae115bc7Smrj * segmap_start: start of segmap. The length of segmap can be modified 51552eef812SAmrita Sadhukhan * through eeprom. The default length is 16MB on 32-bit systems and 64MB 51652eef812SAmrita Sadhukhan * on 64-bit systems. 5177c478bd9Sstevel@tonic-gate * 5187c478bd9Sstevel@tonic-gate * kernelbase: On a 32-bit kernel the default value of 0xd4000000 will be 5197c478bd9Sstevel@tonic-gate * decreased by 2X the size required for page_t. This allows the kernel 5207c478bd9Sstevel@tonic-gate * heap to grow in size with physical memory. With sizeof(page_t) == 80 5217c478bd9Sstevel@tonic-gate * bytes, the following shows the values of kernelbase and kernel heap 5227c478bd9Sstevel@tonic-gate * sizes for different memory configurations (assuming default segmap and 5237c478bd9Sstevel@tonic-gate * segkp sizes). 5247c478bd9Sstevel@tonic-gate * 5257c478bd9Sstevel@tonic-gate * mem size for kernelbase kernel heap 5267c478bd9Sstevel@tonic-gate * size page_t's size 5277c478bd9Sstevel@tonic-gate * ---- --------- ---------- ----------- 5287c478bd9Sstevel@tonic-gate * 1gb 0x01400000 0xd1800000 684MB 5297c478bd9Sstevel@tonic-gate * 2gb 0x02800000 0xcf000000 704MB 5307c478bd9Sstevel@tonic-gate * 4gb 0x05000000 0xca000000 744MB 5317c478bd9Sstevel@tonic-gate * 6gb 0x07800000 0xc5000000 784MB 5327c478bd9Sstevel@tonic-gate * 8gb 0x0a000000 0xc0000000 824MB 5337c478bd9Sstevel@tonic-gate * 16gb 0x14000000 0xac000000 984MB 5347c478bd9Sstevel@tonic-gate * 32gb 0x28000000 0x84000000 1304MB 5357c478bd9Sstevel@tonic-gate * 64gb 0x50000000 0x34000000 1944MB (*) 5367c478bd9Sstevel@tonic-gate * 5377c478bd9Sstevel@tonic-gate * kernelbase is less than the abi minimum of 0xc0000000 for memory 5387c478bd9Sstevel@tonic-gate * configurations above 8gb. 5397c478bd9Sstevel@tonic-gate * 5407c478bd9Sstevel@tonic-gate * (*) support for memory configurations above 32gb will require manual tuning 5417c478bd9Sstevel@tonic-gate * of kernelbase to balance out the need of user applications. 5427c478bd9Sstevel@tonic-gate */ 5437c478bd9Sstevel@tonic-gate 5447c478bd9Sstevel@tonic-gate /* real-time-clock initialization parameters */ 545ae115bc7Smrj extern time_t process_rtc_config_file(void); 5467c478bd9Sstevel@tonic-gate 5477c478bd9Sstevel@tonic-gate uintptr_t kernelbase; 548ae115bc7Smrj uintptr_t postbootkernelbase; /* not set till boot loader is gone */ 5497c478bd9Sstevel@tonic-gate uintptr_t eprom_kernelbase; 5507c478bd9Sstevel@tonic-gate size_t segmapsize; 551ae115bc7Smrj uintptr_t segmap_start; 5527c478bd9Sstevel@tonic-gate int segmapfreelists; 5537c478bd9Sstevel@tonic-gate pgcnt_t npages; 554ae115bc7Smrj pgcnt_t orig_npages; 5557c478bd9Sstevel@tonic-gate size_t core_size; /* size of "core" heap */ 5567c478bd9Sstevel@tonic-gate uintptr_t core_base; /* base address of "core" heap */ 5577c478bd9Sstevel@tonic-gate 5587c478bd9Sstevel@tonic-gate /* 5597c478bd9Sstevel@tonic-gate * List of bootstrap pages. We mark these as allocated in startup. 5607c478bd9Sstevel@tonic-gate * release_bootstrap() will free them when we're completely done with 5617c478bd9Sstevel@tonic-gate * the bootstrap. 5627c478bd9Sstevel@tonic-gate */ 563ae115bc7Smrj static page_t *bootpages; 564ae115bc7Smrj 565ae115bc7Smrj /* 566ae115bc7Smrj * boot time pages that have a vnode from the ramdisk will keep that forever. 567ae115bc7Smrj */ 568ae115bc7Smrj static page_t *rd_pages; 5697c478bd9Sstevel@tonic-gate 57019397407SSherry Moore /* 57119397407SSherry Moore * Lower 64K 57219397407SSherry Moore */ 57319397407SSherry Moore static page_t *lower_pages = NULL; 57419397407SSherry Moore static int lower_pages_count = 0; 57519397407SSherry Moore 5767c478bd9Sstevel@tonic-gate struct system_hardware system_hardware; 5777c478bd9Sstevel@tonic-gate 5787c478bd9Sstevel@tonic-gate /* 5797c478bd9Sstevel@tonic-gate * Enable some debugging messages concerning memory usage... 5807c478bd9Sstevel@tonic-gate */ 5817c478bd9Sstevel@tonic-gate static void 582ae115bc7Smrj print_memlist(char *title, struct memlist *mp) 5837c478bd9Sstevel@tonic-gate { 5847c478bd9Sstevel@tonic-gate prom_printf("MEMLIST: %s:\n", title); 5857c478bd9Sstevel@tonic-gate while (mp != NULL) { 5867c478bd9Sstevel@tonic-gate prom_printf("\tAddress 0x%" PRIx64 ", size 0x%" PRIx64 "\n", 58756f33205SJonathan Adams mp->ml_address, mp->ml_size); 58856f33205SJonathan Adams mp = mp->ml_next; 5897c478bd9Sstevel@tonic-gate } 5907c478bd9Sstevel@tonic-gate } 5917c478bd9Sstevel@tonic-gate 5927c478bd9Sstevel@tonic-gate /* 5937c478bd9Sstevel@tonic-gate * XX64 need a comment here.. are these just default values, surely 5947c478bd9Sstevel@tonic-gate * we read the "cpuid" type information to figure this out. 5957c478bd9Sstevel@tonic-gate */ 5967c478bd9Sstevel@tonic-gate int l2cache_sz = 0x80000; 5977c478bd9Sstevel@tonic-gate int l2cache_linesz = 0x40; 5987c478bd9Sstevel@tonic-gate int l2cache_assoc = 1; 5997c478bd9Sstevel@tonic-gate 6002cb27123Saguzovsk static size_t textrepl_min_gb = 10; 6012cb27123Saguzovsk 6027c478bd9Sstevel@tonic-gate /* 6037c478bd9Sstevel@tonic-gate * on 64 bit we use a predifined VA range for mapping devices in the kernel 6047c478bd9Sstevel@tonic-gate * on 32 bit the mappings are intermixed in the heap, so we use a bit map 6057c478bd9Sstevel@tonic-gate */ 6067c478bd9Sstevel@tonic-gate #ifdef __amd64 6077c478bd9Sstevel@tonic-gate 6087c478bd9Sstevel@tonic-gate vmem_t *device_arena; 6097c478bd9Sstevel@tonic-gate uintptr_t toxic_addr = (uintptr_t)NULL; 610ae115bc7Smrj size_t toxic_size = 1024 * 1024 * 1024; /* Sparc uses 1 gig too */ 6117c478bd9Sstevel@tonic-gate 6127c478bd9Sstevel@tonic-gate #else /* __i386 */ 6137c478bd9Sstevel@tonic-gate 6147c478bd9Sstevel@tonic-gate ulong_t *toxic_bit_map; /* one bit for each 4k of VA in heap_arena */ 6157c478bd9Sstevel@tonic-gate size_t toxic_bit_map_len = 0; /* in bits */ 6167c478bd9Sstevel@tonic-gate 6177c478bd9Sstevel@tonic-gate #endif /* __i386 */ 6187c478bd9Sstevel@tonic-gate 6197c478bd9Sstevel@tonic-gate /* 6207c478bd9Sstevel@tonic-gate * Simple boot time debug facilities 6217c478bd9Sstevel@tonic-gate */ 6227c478bd9Sstevel@tonic-gate static char *prm_dbg_str[] = { 6237c478bd9Sstevel@tonic-gate "%s:%d: '%s' is 0x%x\n", 6247c478bd9Sstevel@tonic-gate "%s:%d: '%s' is 0x%llx\n" 6257c478bd9Sstevel@tonic-gate }; 6267c478bd9Sstevel@tonic-gate 6277c478bd9Sstevel@tonic-gate int prom_debug; 6287c478bd9Sstevel@tonic-gate 6297c478bd9Sstevel@tonic-gate #define PRM_DEBUG(q) if (prom_debug) \ 6307c478bd9Sstevel@tonic-gate prom_printf(prm_dbg_str[sizeof (q) >> 3], "startup.c", __LINE__, #q, q); 6317c478bd9Sstevel@tonic-gate #define PRM_POINT(q) if (prom_debug) \ 6327c478bd9Sstevel@tonic-gate prom_printf("%s:%d: %s\n", "startup.c", __LINE__, q); 6337c478bd9Sstevel@tonic-gate 6347c478bd9Sstevel@tonic-gate /* 6357c478bd9Sstevel@tonic-gate * This structure is used to keep track of the intial allocations 6367c478bd9Sstevel@tonic-gate * done in startup_memlist(). The value of NUM_ALLOCATIONS needs to 6377c478bd9Sstevel@tonic-gate * be >= the number of ADD_TO_ALLOCATIONS() executed in the code. 6387c478bd9Sstevel@tonic-gate */ 6391de082f7SVikram Hegde #define NUM_ALLOCATIONS 8 6407c478bd9Sstevel@tonic-gate int num_allocations = 0; 6417c478bd9Sstevel@tonic-gate struct { 6427c478bd9Sstevel@tonic-gate void **al_ptr; 6437c478bd9Sstevel@tonic-gate size_t al_size; 6447c478bd9Sstevel@tonic-gate } allocations[NUM_ALLOCATIONS]; 6457c478bd9Sstevel@tonic-gate size_t valloc_sz = 0; 6467c478bd9Sstevel@tonic-gate uintptr_t valloc_base; 6477c478bd9Sstevel@tonic-gate 6487c478bd9Sstevel@tonic-gate #define ADD_TO_ALLOCATIONS(ptr, size) { \ 6497c478bd9Sstevel@tonic-gate size = ROUND_UP_PAGE(size); \ 6507c478bd9Sstevel@tonic-gate if (num_allocations == NUM_ALLOCATIONS) \ 6517c478bd9Sstevel@tonic-gate panic("too many ADD_TO_ALLOCATIONS()"); \ 6527c478bd9Sstevel@tonic-gate allocations[num_allocations].al_ptr = (void**)&ptr; \ 6537c478bd9Sstevel@tonic-gate allocations[num_allocations].al_size = size; \ 6547c478bd9Sstevel@tonic-gate valloc_sz += size; \ 6557c478bd9Sstevel@tonic-gate ++num_allocations; \ 6567c478bd9Sstevel@tonic-gate } 6577c478bd9Sstevel@tonic-gate 658ae115bc7Smrj /* 659ae115bc7Smrj * Allocate all the initial memory needed by the page allocator. 660ae115bc7Smrj */ 6617c478bd9Sstevel@tonic-gate static void 6627c478bd9Sstevel@tonic-gate perform_allocations(void) 6637c478bd9Sstevel@tonic-gate { 6647c478bd9Sstevel@tonic-gate caddr_t mem; 6657c478bd9Sstevel@tonic-gate int i; 666ae115bc7Smrj int valloc_align; 6677c478bd9Sstevel@tonic-gate 668ae115bc7Smrj PRM_DEBUG(valloc_base); 669ae115bc7Smrj PRM_DEBUG(valloc_sz); 670ae115bc7Smrj valloc_align = mmu.level_size[mmu.max_page_level > 0]; 671ae115bc7Smrj mem = BOP_ALLOC(bootops, (caddr_t)valloc_base, valloc_sz, valloc_align); 6727c478bd9Sstevel@tonic-gate if (mem != (caddr_t)valloc_base) 6737c478bd9Sstevel@tonic-gate panic("BOP_ALLOC() failed"); 6747c478bd9Sstevel@tonic-gate bzero(mem, valloc_sz); 6757c478bd9Sstevel@tonic-gate for (i = 0; i < num_allocations; ++i) { 6767c478bd9Sstevel@tonic-gate *allocations[i].al_ptr = (void *)mem; 6777c478bd9Sstevel@tonic-gate mem += allocations[i].al_size; 6787c478bd9Sstevel@tonic-gate } 6797c478bd9Sstevel@tonic-gate } 6807c478bd9Sstevel@tonic-gate 681*3ce2fcdcSRobert Mustacchi /* 682*3ce2fcdcSRobert Mustacchi * Set up and enable SMAP now before we start other CPUs, but after the kernel's 683*3ce2fcdcSRobert Mustacchi * VM has been set up so we can use hot_patch_kernel_text(). 684*3ce2fcdcSRobert Mustacchi * 685*3ce2fcdcSRobert Mustacchi * We can only patch 1, 2, or 4 bytes, but not three bytes. So instead, we 686*3ce2fcdcSRobert Mustacchi * replace the four byte word at the patch point. See uts/intel/ia32/ml/copy.s 687*3ce2fcdcSRobert Mustacchi * for more information on what's going on here. 688*3ce2fcdcSRobert Mustacchi */ 689*3ce2fcdcSRobert Mustacchi static void 690*3ce2fcdcSRobert Mustacchi startup_smap(void) 691*3ce2fcdcSRobert Mustacchi { 692*3ce2fcdcSRobert Mustacchi int i; 693*3ce2fcdcSRobert Mustacchi uint32_t inst; 694*3ce2fcdcSRobert Mustacchi uint8_t *instp; 695*3ce2fcdcSRobert Mustacchi char sym[128]; 696*3ce2fcdcSRobert Mustacchi 697*3ce2fcdcSRobert Mustacchi extern int _smap_enable_patch_count; 698*3ce2fcdcSRobert Mustacchi extern int _smap_disable_patch_count; 699*3ce2fcdcSRobert Mustacchi 700*3ce2fcdcSRobert Mustacchi if (disable_smap != 0) 701*3ce2fcdcSRobert Mustacchi remove_x86_feature(x86_featureset, X86FSET_SMAP); 702*3ce2fcdcSRobert Mustacchi 703*3ce2fcdcSRobert Mustacchi if (is_x86_feature(x86_featureset, X86FSET_SMAP) == B_FALSE) 704*3ce2fcdcSRobert Mustacchi return; 705*3ce2fcdcSRobert Mustacchi 706*3ce2fcdcSRobert Mustacchi for (i = 0; i < _smap_enable_patch_count; i++) { 707*3ce2fcdcSRobert Mustacchi int sizep; 708*3ce2fcdcSRobert Mustacchi 709*3ce2fcdcSRobert Mustacchi VERIFY3U(i, <, _smap_enable_patch_count); 710*3ce2fcdcSRobert Mustacchi VERIFY(snprintf(sym, sizeof (sym), "_smap_enable_patch_%d", i) < 711*3ce2fcdcSRobert Mustacchi sizeof (sym)); 712*3ce2fcdcSRobert Mustacchi instp = (uint8_t *)(void *)kobj_getelfsym(sym, NULL, &sizep); 713*3ce2fcdcSRobert Mustacchi VERIFY(instp != 0); 714*3ce2fcdcSRobert Mustacchi inst = (instp[3] << 24) | (SMAP_CLAC_INSTR & 0x00ffffff); 715*3ce2fcdcSRobert Mustacchi hot_patch_kernel_text((caddr_t)instp, inst, 4); 716*3ce2fcdcSRobert Mustacchi } 717*3ce2fcdcSRobert Mustacchi 718*3ce2fcdcSRobert Mustacchi for (i = 0; i < _smap_disable_patch_count; i++) { 719*3ce2fcdcSRobert Mustacchi int sizep; 720*3ce2fcdcSRobert Mustacchi 721*3ce2fcdcSRobert Mustacchi VERIFY(snprintf(sym, sizeof (sym), "_smap_disable_patch_%d", 722*3ce2fcdcSRobert Mustacchi i) < sizeof (sym)); 723*3ce2fcdcSRobert Mustacchi instp = (uint8_t *)(void *)kobj_getelfsym(sym, NULL, &sizep); 724*3ce2fcdcSRobert Mustacchi VERIFY(instp != 0); 725*3ce2fcdcSRobert Mustacchi inst = (instp[3] << 24) | (SMAP_STAC_INSTR & 0x00ffffff); 726*3ce2fcdcSRobert Mustacchi hot_patch_kernel_text((caddr_t)instp, inst, 4); 727*3ce2fcdcSRobert Mustacchi } 728*3ce2fcdcSRobert Mustacchi 729*3ce2fcdcSRobert Mustacchi hot_patch_kernel_text((caddr_t)smap_enable, SMAP_CLAC_INSTR, 4); 730*3ce2fcdcSRobert Mustacchi hot_patch_kernel_text((caddr_t)smap_disable, SMAP_STAC_INSTR, 4); 731*3ce2fcdcSRobert Mustacchi setcr4(getcr4() | CR4_SMAP); 732*3ce2fcdcSRobert Mustacchi smap_enable(); 733*3ce2fcdcSRobert Mustacchi } 734*3ce2fcdcSRobert Mustacchi 7357c478bd9Sstevel@tonic-gate /* 7367c478bd9Sstevel@tonic-gate * Our world looks like this at startup time. 7377c478bd9Sstevel@tonic-gate * 7387c478bd9Sstevel@tonic-gate * In a 32-bit OS, boot loads the kernel text at 0xfe800000 and kernel data 7397c478bd9Sstevel@tonic-gate * at 0xfec00000. On a 64-bit OS, kernel text and data are loaded at 7407c478bd9Sstevel@tonic-gate * 0xffffffff.fe800000 and 0xffffffff.fec00000 respectively. Those 7417c478bd9Sstevel@tonic-gate * addresses are fixed in the binary at link time. 7427c478bd9Sstevel@tonic-gate * 7437c478bd9Sstevel@tonic-gate * On the text page: 7447c478bd9Sstevel@tonic-gate * unix/genunix/krtld/module text loads. 7457c478bd9Sstevel@tonic-gate * 7467c478bd9Sstevel@tonic-gate * On the data page: 747ae115bc7Smrj * unix/genunix/krtld/module data loads. 748ae115bc7Smrj * 7497c478bd9Sstevel@tonic-gate * Machine-dependent startup code 7507c478bd9Sstevel@tonic-gate */ 7517c478bd9Sstevel@tonic-gate void 7527c478bd9Sstevel@tonic-gate startup(void) 7537c478bd9Sstevel@tonic-gate { 754843e1988Sjohnlev #if !defined(__xpv) 75575bcd456Sjg extern void startup_pci_bios(void); 756843e1988Sjohnlev #endif 757a563a037Sbholler extern cpuset_t cpu_ready_set; 758a563a037Sbholler 759843e1988Sjohnlev /* 760843e1988Sjohnlev * Make sure that nobody tries to use sekpm until we have 761843e1988Sjohnlev * initialized it properly. 762843e1988Sjohnlev */ 7637c478bd9Sstevel@tonic-gate #if defined(__amd64) 76435b1ab99Sjosephb kpm_desired = 1; 7657c478bd9Sstevel@tonic-gate #endif 7667c478bd9Sstevel@tonic-gate kpm_enable = 0; 767a563a037Sbholler CPUSET_ONLY(cpu_ready_set, 0); /* cpu 0 is boot cpu */ 7687c478bd9Sstevel@tonic-gate 769843e1988Sjohnlev #if defined(__xpv) /* XXPV fix me! */ 770843e1988Sjohnlev { 771843e1988Sjohnlev extern int segvn_use_regions; 772843e1988Sjohnlev segvn_use_regions = 0; 773843e1988Sjohnlev } 774843e1988Sjohnlev #endif 7756a59053bSAlex Wilson ssp_init(); 7767c478bd9Sstevel@tonic-gate progressbar_init(); 7777c478bd9Sstevel@tonic-gate startup_init(); 778ab4a9bebSjohnlev #if defined(__xpv) 779ab4a9bebSjohnlev startup_xen_version(); 780ab4a9bebSjohnlev #endif 7817c478bd9Sstevel@tonic-gate startup_memlist(); 782ae115bc7Smrj startup_kmem(); 78335b1ab99Sjosephb startup_vm(); 784843e1988Sjohnlev #if !defined(__xpv) 7850db3240dSStephen Hanson /* 7860db3240dSStephen Hanson * Note we need to do this even on fast reboot in order to access 7870db3240dSStephen Hanson * the irq routing table (used for pci labels). 7880db3240dSStephen Hanson */ 7890db3240dSStephen Hanson startup_pci_bios(); 790*3ce2fcdcSRobert Mustacchi startup_smap(); 791e4b86885SCheng Sean Ye #endif 792e4b86885SCheng Sean Ye #if defined(__xpv) 793e4b86885SCheng Sean Ye startup_xen_mca(); 794843e1988Sjohnlev #endif 7957c478bd9Sstevel@tonic-gate startup_modules(); 7968770118eSlipeng sang - Sun Microsystems - Beijing China 7977c478bd9Sstevel@tonic-gate startup_end(); 7987c478bd9Sstevel@tonic-gate } 7997c478bd9Sstevel@tonic-gate 8007c478bd9Sstevel@tonic-gate static void 8017c478bd9Sstevel@tonic-gate startup_init() 8027c478bd9Sstevel@tonic-gate { 8037c478bd9Sstevel@tonic-gate PRM_POINT("startup_init() starting..."); 8047c478bd9Sstevel@tonic-gate 8057c478bd9Sstevel@tonic-gate /* 8067c478bd9Sstevel@tonic-gate * Complete the extraction of cpuid data 8077c478bd9Sstevel@tonic-gate */ 8087c478bd9Sstevel@tonic-gate cpuid_pass2(CPU); 8097c478bd9Sstevel@tonic-gate 8107c478bd9Sstevel@tonic-gate (void) check_boot_version(BOP_GETVERSION(bootops)); 8117c478bd9Sstevel@tonic-gate 8127c478bd9Sstevel@tonic-gate /* 8137c478bd9Sstevel@tonic-gate * Check for prom_debug in boot environment 8147c478bd9Sstevel@tonic-gate */ 8157c478bd9Sstevel@tonic-gate if (BOP_GETPROPLEN(bootops, "prom_debug") >= 0) { 8167c478bd9Sstevel@tonic-gate ++prom_debug; 8177c478bd9Sstevel@tonic-gate PRM_POINT("prom_debug found in boot enviroment"); 8187c478bd9Sstevel@tonic-gate } 8197c478bd9Sstevel@tonic-gate 8207c478bd9Sstevel@tonic-gate /* 8217c478bd9Sstevel@tonic-gate * Collect node, cpu and memory configuration information. 8227c478bd9Sstevel@tonic-gate */ 8237c478bd9Sstevel@tonic-gate get_system_configuration(); 8247c478bd9Sstevel@tonic-gate 8257c478bd9Sstevel@tonic-gate /* 8267c478bd9Sstevel@tonic-gate * Halt if this is an unsupported processor. 8277c478bd9Sstevel@tonic-gate */ 8287c478bd9Sstevel@tonic-gate if (x86_type == X86_TYPE_486 || x86_type == X86_TYPE_CYRIX_486) { 8297c478bd9Sstevel@tonic-gate printf("\n486 processor (\"%s\") detected.\n", 8307c478bd9Sstevel@tonic-gate CPU->cpu_brandstr); 8317c478bd9Sstevel@tonic-gate halt("This processor is not supported by this release " 8327c478bd9Sstevel@tonic-gate "of Solaris."); 8337c478bd9Sstevel@tonic-gate } 8347c478bd9Sstevel@tonic-gate 8357c478bd9Sstevel@tonic-gate PRM_POINT("startup_init() done"); 8367c478bd9Sstevel@tonic-gate } 8377c478bd9Sstevel@tonic-gate 8387c478bd9Sstevel@tonic-gate /* 8397c478bd9Sstevel@tonic-gate * Callback for copy_memlist_filter() to filter nucleus, kadb/kmdb, (ie. 8407c478bd9Sstevel@tonic-gate * everything mapped above KERNEL_TEXT) pages from phys_avail. Note it 8417c478bd9Sstevel@tonic-gate * also filters out physical page zero. There is some reliance on the 8427c478bd9Sstevel@tonic-gate * boot loader allocating only a few contiguous physical memory chunks. 8437c478bd9Sstevel@tonic-gate */ 8447c478bd9Sstevel@tonic-gate static void 8457c478bd9Sstevel@tonic-gate avail_filter(uint64_t *addr, uint64_t *size) 8467c478bd9Sstevel@tonic-gate { 8477c478bd9Sstevel@tonic-gate uintptr_t va; 8487c478bd9Sstevel@tonic-gate uintptr_t next_va; 8497c478bd9Sstevel@tonic-gate pfn_t pfn; 8507c478bd9Sstevel@tonic-gate uint64_t pfn_addr; 8517c478bd9Sstevel@tonic-gate uint64_t pfn_eaddr; 8527c478bd9Sstevel@tonic-gate uint_t prot; 8537c478bd9Sstevel@tonic-gate size_t len; 8547c478bd9Sstevel@tonic-gate uint_t change; 8557c478bd9Sstevel@tonic-gate 8567c478bd9Sstevel@tonic-gate if (prom_debug) 8577c478bd9Sstevel@tonic-gate prom_printf("\tFilter: in: a=%" PRIx64 ", s=%" PRIx64 "\n", 8587c478bd9Sstevel@tonic-gate *addr, *size); 8597c478bd9Sstevel@tonic-gate 8607c478bd9Sstevel@tonic-gate /* 8617c478bd9Sstevel@tonic-gate * page zero is required for BIOS.. never make it available 8627c478bd9Sstevel@tonic-gate */ 8637c478bd9Sstevel@tonic-gate if (*addr == 0) { 8647c478bd9Sstevel@tonic-gate *addr += MMU_PAGESIZE; 8657c478bd9Sstevel@tonic-gate *size -= MMU_PAGESIZE; 8667c478bd9Sstevel@tonic-gate } 8677c478bd9Sstevel@tonic-gate 8687c478bd9Sstevel@tonic-gate /* 869ae115bc7Smrj * First we trim from the front of the range. Since kbm_probe() 8707c478bd9Sstevel@tonic-gate * walks ranges in virtual order, but addr/size are physical, we need 8717c478bd9Sstevel@tonic-gate * to the list until no changes are seen. This deals with the case 8727c478bd9Sstevel@tonic-gate * where page "p" is mapped at v, page "p + PAGESIZE" is mapped at w 8737c478bd9Sstevel@tonic-gate * but w < v. 8747c478bd9Sstevel@tonic-gate */ 8757c478bd9Sstevel@tonic-gate do { 8767c478bd9Sstevel@tonic-gate change = 0; 8777c478bd9Sstevel@tonic-gate for (va = KERNEL_TEXT; 878ae115bc7Smrj *size > 0 && kbm_probe(&va, &len, &pfn, &prot) != 0; 8797c478bd9Sstevel@tonic-gate va = next_va) { 8807c478bd9Sstevel@tonic-gate 8817c478bd9Sstevel@tonic-gate next_va = va + len; 882ae115bc7Smrj pfn_addr = pfn_to_pa(pfn); 8837c478bd9Sstevel@tonic-gate pfn_eaddr = pfn_addr + len; 8847c478bd9Sstevel@tonic-gate 8857c478bd9Sstevel@tonic-gate if (pfn_addr <= *addr && pfn_eaddr > *addr) { 8867c478bd9Sstevel@tonic-gate change = 1; 8877c478bd9Sstevel@tonic-gate while (*size > 0 && len > 0) { 8887c478bd9Sstevel@tonic-gate *addr += MMU_PAGESIZE; 8897c478bd9Sstevel@tonic-gate *size -= MMU_PAGESIZE; 8907c478bd9Sstevel@tonic-gate len -= MMU_PAGESIZE; 8917c478bd9Sstevel@tonic-gate } 8927c478bd9Sstevel@tonic-gate } 8937c478bd9Sstevel@tonic-gate } 8947c478bd9Sstevel@tonic-gate if (change && prom_debug) 8957c478bd9Sstevel@tonic-gate prom_printf("\t\ttrim: a=%" PRIx64 ", s=%" PRIx64 "\n", 8967c478bd9Sstevel@tonic-gate *addr, *size); 8977c478bd9Sstevel@tonic-gate } while (change); 8987c478bd9Sstevel@tonic-gate 8997c478bd9Sstevel@tonic-gate /* 9007c478bd9Sstevel@tonic-gate * Trim pages from the end of the range. 9017c478bd9Sstevel@tonic-gate */ 9027c478bd9Sstevel@tonic-gate for (va = KERNEL_TEXT; 903ae115bc7Smrj *size > 0 && kbm_probe(&va, &len, &pfn, &prot) != 0; 9047c478bd9Sstevel@tonic-gate va = next_va) { 9057c478bd9Sstevel@tonic-gate 9067c478bd9Sstevel@tonic-gate next_va = va + len; 907ae115bc7Smrj pfn_addr = pfn_to_pa(pfn); 9087c478bd9Sstevel@tonic-gate 9097c478bd9Sstevel@tonic-gate if (pfn_addr >= *addr && pfn_addr < *addr + *size) 9107c478bd9Sstevel@tonic-gate *size = pfn_addr - *addr; 9117c478bd9Sstevel@tonic-gate } 9127c478bd9Sstevel@tonic-gate 9137c478bd9Sstevel@tonic-gate if (prom_debug) 9147c478bd9Sstevel@tonic-gate prom_printf("\tFilter out: a=%" PRIx64 ", s=%" PRIx64 "\n", 9157c478bd9Sstevel@tonic-gate *addr, *size); 9167c478bd9Sstevel@tonic-gate } 9177c478bd9Sstevel@tonic-gate 9187c478bd9Sstevel@tonic-gate static void 9197c478bd9Sstevel@tonic-gate kpm_init() 9207c478bd9Sstevel@tonic-gate { 9217c478bd9Sstevel@tonic-gate struct segkpm_crargs b; 9227c478bd9Sstevel@tonic-gate 9237c478bd9Sstevel@tonic-gate /* 9247c478bd9Sstevel@tonic-gate * These variables were all designed for sfmmu in which segkpm is 9257c478bd9Sstevel@tonic-gate * mapped using a single pagesize - either 8KB or 4MB. On x86, we 9267c478bd9Sstevel@tonic-gate * might use 2+ page sizes on a single machine, so none of these 9277c478bd9Sstevel@tonic-gate * variables have a single correct value. They are set up as if we 9287c478bd9Sstevel@tonic-gate * always use a 4KB pagesize, which should do no harm. In the long 9297c478bd9Sstevel@tonic-gate * run, we should get rid of KPM's assumption that only a single 9307c478bd9Sstevel@tonic-gate * pagesize is used. 9317c478bd9Sstevel@tonic-gate */ 9327c478bd9Sstevel@tonic-gate kpm_pgshft = MMU_PAGESHIFT; 9337c478bd9Sstevel@tonic-gate kpm_pgsz = MMU_PAGESIZE; 9347c478bd9Sstevel@tonic-gate kpm_pgoff = MMU_PAGEOFFSET; 9357c478bd9Sstevel@tonic-gate kpmp2pshft = 0; 9367c478bd9Sstevel@tonic-gate kpmpnpgs = 1; 9377c478bd9Sstevel@tonic-gate ASSERT(((uintptr_t)kpm_vbase & (kpm_pgsz - 1)) == 0); 9387c478bd9Sstevel@tonic-gate 9397c478bd9Sstevel@tonic-gate PRM_POINT("about to create segkpm"); 9407c478bd9Sstevel@tonic-gate rw_enter(&kas.a_lock, RW_WRITER); 9417c478bd9Sstevel@tonic-gate 9427c478bd9Sstevel@tonic-gate if (seg_attach(&kas, kpm_vbase, kpm_size, segkpm) < 0) 9437c478bd9Sstevel@tonic-gate panic("cannot attach segkpm"); 9447c478bd9Sstevel@tonic-gate 9457c478bd9Sstevel@tonic-gate b.prot = PROT_READ | PROT_WRITE; 9467c478bd9Sstevel@tonic-gate b.nvcolors = 1; 9477c478bd9Sstevel@tonic-gate 9487c478bd9Sstevel@tonic-gate if (segkpm_create(segkpm, (caddr_t)&b) != 0) 9497c478bd9Sstevel@tonic-gate panic("segkpm_create segkpm"); 9507c478bd9Sstevel@tonic-gate 9517c478bd9Sstevel@tonic-gate rw_exit(&kas.a_lock); 952ae115bc7Smrj } 9537c478bd9Sstevel@tonic-gate 954ae115bc7Smrj /* 955ae115bc7Smrj * The debug info page provides enough information to allow external 956ae115bc7Smrj * inspectors (e.g. when running under a hypervisor) to bootstrap 957ae115bc7Smrj * themselves into allowing full-blown kernel debugging. 958ae115bc7Smrj */ 959ae115bc7Smrj static void 960ae115bc7Smrj init_debug_info(void) 961ae115bc7Smrj { 962ae115bc7Smrj caddr_t mem; 963ae115bc7Smrj debug_info_t *di; 964ae115bc7Smrj 965ae115bc7Smrj #ifndef __lint 966ae115bc7Smrj ASSERT(sizeof (debug_info_t) < MMU_PAGESIZE); 967ae115bc7Smrj #endif 968ae115bc7Smrj 969ae115bc7Smrj mem = BOP_ALLOC(bootops, (caddr_t)DEBUG_INFO_VA, MMU_PAGESIZE, 970ae115bc7Smrj MMU_PAGESIZE); 971ae115bc7Smrj 972ae115bc7Smrj if (mem != (caddr_t)DEBUG_INFO_VA) 973ae115bc7Smrj panic("BOP_ALLOC() failed"); 974ae115bc7Smrj bzero(mem, MMU_PAGESIZE); 975ae115bc7Smrj 976ae115bc7Smrj di = (debug_info_t *)mem; 977ae115bc7Smrj 978ae115bc7Smrj di->di_magic = DEBUG_INFO_MAGIC; 979ae115bc7Smrj di->di_version = DEBUG_INFO_VERSION; 980843e1988Sjohnlev di->di_modules = (uintptr_t)&modules; 981843e1988Sjohnlev di->di_s_text = (uintptr_t)s_text; 982843e1988Sjohnlev di->di_e_text = (uintptr_t)e_text; 983843e1988Sjohnlev di->di_s_data = (uintptr_t)s_data; 984843e1988Sjohnlev di->di_e_data = (uintptr_t)e_data; 985843e1988Sjohnlev di->di_hat_htable_off = offsetof(hat_t, hat_htable); 986843e1988Sjohnlev di->di_ht_pfn_off = offsetof(htable_t, ht_pfn); 9877c478bd9Sstevel@tonic-gate } 9887c478bd9Sstevel@tonic-gate 9897c478bd9Sstevel@tonic-gate /* 990ae115bc7Smrj * Build the memlists and other kernel essential memory system data structures. 991ae115bc7Smrj * This is everything at valloc_base. 9927c478bd9Sstevel@tonic-gate */ 9937c478bd9Sstevel@tonic-gate static void 9947c478bd9Sstevel@tonic-gate startup_memlist(void) 9957c478bd9Sstevel@tonic-gate { 9967c478bd9Sstevel@tonic-gate size_t memlist_sz; 9977c478bd9Sstevel@tonic-gate size_t memseg_sz; 9987c478bd9Sstevel@tonic-gate size_t pagehash_sz; 9997c478bd9Sstevel@tonic-gate size_t pp_sz; 10007c478bd9Sstevel@tonic-gate uintptr_t va; 10017c478bd9Sstevel@tonic-gate size_t len; 10027c478bd9Sstevel@tonic-gate uint_t prot; 10037c478bd9Sstevel@tonic-gate pfn_t pfn; 10047c478bd9Sstevel@tonic-gate int memblocks; 10051de082f7SVikram Hegde pfn_t rsvd_high_pfn; 10061de082f7SVikram Hegde pgcnt_t rsvd_pgcnt; 10071de082f7SVikram Hegde size_t rsvdmemlist_sz; 10081de082f7SVikram Hegde int rsvdmemblocks; 10097c478bd9Sstevel@tonic-gate caddr_t pagecolor_mem; 10107c478bd9Sstevel@tonic-gate size_t pagecolor_memsz; 10117c478bd9Sstevel@tonic-gate caddr_t page_ctrs_mem; 10127c478bd9Sstevel@tonic-gate size_t page_ctrs_size; 1013d7d93655Sblakej size_t pse_table_alloc_size; 10147c478bd9Sstevel@tonic-gate struct memlist *current; 10157c478bd9Sstevel@tonic-gate extern void startup_build_mem_nodes(struct memlist *); 10167c478bd9Sstevel@tonic-gate 10177c478bd9Sstevel@tonic-gate /* XX64 fix these - they should be in include files */ 10187c478bd9Sstevel@tonic-gate extern size_t page_coloring_init(uint_t, int, int); 10197c478bd9Sstevel@tonic-gate extern void page_coloring_setup(caddr_t); 10207c478bd9Sstevel@tonic-gate 10217c478bd9Sstevel@tonic-gate PRM_POINT("startup_memlist() starting..."); 10227c478bd9Sstevel@tonic-gate 10237c478bd9Sstevel@tonic-gate /* 10247c478bd9Sstevel@tonic-gate * Use leftover large page nucleus text/data space for loadable modules. 10257c478bd9Sstevel@tonic-gate * Use at most MODTEXT/MODDATA. 10267c478bd9Sstevel@tonic-gate */ 1027ae115bc7Smrj len = kbm_nucleus_size; 1028ae115bc7Smrj ASSERT(len > MMU_PAGESIZE); 10297c478bd9Sstevel@tonic-gate 1030ae115bc7Smrj moddata = (caddr_t)ROUND_UP_PAGE(e_data); 1031ae115bc7Smrj e_moddata = (caddr_t)P2ROUNDUP((uintptr_t)e_data, (uintptr_t)len); 1032ae115bc7Smrj if (e_moddata - moddata > MODDATA) 1033ae115bc7Smrj e_moddata = moddata + MODDATA; 10347c478bd9Sstevel@tonic-gate 1035ae115bc7Smrj modtext = (caddr_t)ROUND_UP_PAGE(e_text); 1036ae115bc7Smrj e_modtext = (caddr_t)P2ROUNDUP((uintptr_t)e_text, (uintptr_t)len); 1037ae115bc7Smrj if (e_modtext - modtext > MODTEXT) 1038ae115bc7Smrj e_modtext = modtext + MODTEXT; 10397c478bd9Sstevel@tonic-gate 10407c478bd9Sstevel@tonic-gate econtig = e_moddata; 10417c478bd9Sstevel@tonic-gate 10427c478bd9Sstevel@tonic-gate PRM_DEBUG(modtext); 10437c478bd9Sstevel@tonic-gate PRM_DEBUG(e_modtext); 10447c478bd9Sstevel@tonic-gate PRM_DEBUG(moddata); 10457c478bd9Sstevel@tonic-gate PRM_DEBUG(e_moddata); 10467c478bd9Sstevel@tonic-gate PRM_DEBUG(econtig); 10477c478bd9Sstevel@tonic-gate 10487c478bd9Sstevel@tonic-gate /* 1049ae115bc7Smrj * Examine the boot loader physical memory map to find out: 10507c478bd9Sstevel@tonic-gate * - total memory in system - physinstalled 10517c478bd9Sstevel@tonic-gate * - the max physical address - physmax 1052ae115bc7Smrj * - the number of discontiguous segments of memory. 10537c478bd9Sstevel@tonic-gate */ 10547c478bd9Sstevel@tonic-gate if (prom_debug) 1055ae115bc7Smrj print_memlist("boot physinstalled", 10567c478bd9Sstevel@tonic-gate bootops->boot_mem->physinstalled); 1057a3114836SGerry Liu installed_top_size_ex(bootops->boot_mem->physinstalled, &physmax, 10587c478bd9Sstevel@tonic-gate &physinstalled, &memblocks); 10597c478bd9Sstevel@tonic-gate PRM_DEBUG(physmax); 10607c478bd9Sstevel@tonic-gate PRM_DEBUG(physinstalled); 10617c478bd9Sstevel@tonic-gate PRM_DEBUG(memblocks); 10627c478bd9Sstevel@tonic-gate 1063a3114836SGerry Liu /* 1064a3114836SGerry Liu * Compute maximum physical address for memory DR operations. 1065a3114836SGerry Liu * Memory DR operations are unsupported on xpv or 32bit OSes. 1066a3114836SGerry Liu */ 1067a3114836SGerry Liu #ifdef __amd64 1068a3114836SGerry Liu if (plat_dr_support_memory()) { 1069a3114836SGerry Liu if (plat_dr_physmax == 0) { 1070a3114836SGerry Liu uint_t pabits = UINT_MAX; 1071a3114836SGerry Liu 1072a3114836SGerry Liu cpuid_get_addrsize(CPU, &pabits, NULL); 1073a3114836SGerry Liu plat_dr_physmax = btop(1ULL << pabits); 1074a3114836SGerry Liu } 1075a3114836SGerry Liu if (plat_dr_physmax > PHYSMEM_MAX64) 1076a3114836SGerry Liu plat_dr_physmax = PHYSMEM_MAX64; 1077a3114836SGerry Liu } else 1078a3114836SGerry Liu #endif 1079a3114836SGerry Liu plat_dr_physmax = 0; 1080a3114836SGerry Liu 10811de082f7SVikram Hegde /* 10821de082f7SVikram Hegde * Examine the bios reserved memory to find out: 10831de082f7SVikram Hegde * - the number of discontiguous segments of memory. 10841de082f7SVikram Hegde */ 10851de082f7SVikram Hegde if (prom_debug) 10861de082f7SVikram Hegde print_memlist("boot reserved mem", 10871de082f7SVikram Hegde bootops->boot_mem->rsvdmem); 1088a3114836SGerry Liu installed_top_size_ex(bootops->boot_mem->rsvdmem, &rsvd_high_pfn, 10891de082f7SVikram Hegde &rsvd_pgcnt, &rsvdmemblocks); 10901de082f7SVikram Hegde PRM_DEBUG(rsvd_high_pfn); 10911de082f7SVikram Hegde PRM_DEBUG(rsvd_pgcnt); 10921de082f7SVikram Hegde PRM_DEBUG(rsvdmemblocks); 10931de082f7SVikram Hegde 10947c478bd9Sstevel@tonic-gate /* 10957c478bd9Sstevel@tonic-gate * Initialize hat's mmu parameters. 10967c478bd9Sstevel@tonic-gate * Check for enforce-prot-exec in boot environment. It's used to 10977c478bd9Sstevel@tonic-gate * enable/disable support for the page table entry NX bit. 10987c478bd9Sstevel@tonic-gate * The default is to enforce PROT_EXEC on processors that support NX. 10997c478bd9Sstevel@tonic-gate * Boot seems to round up the "len", but 8 seems to be big enough. 11007c478bd9Sstevel@tonic-gate */ 11017c478bd9Sstevel@tonic-gate mmu_init(); 11027c478bd9Sstevel@tonic-gate 11037c478bd9Sstevel@tonic-gate #ifdef __i386 11047c478bd9Sstevel@tonic-gate /* 11057c478bd9Sstevel@tonic-gate * physmax is lowered if there is more memory than can be 11067c478bd9Sstevel@tonic-gate * physically addressed in 32 bit (PAE/non-PAE) modes. 11077c478bd9Sstevel@tonic-gate */ 11087c478bd9Sstevel@tonic-gate if (mmu.pae_hat) { 11097c478bd9Sstevel@tonic-gate if (PFN_ABOVE64G(physmax)) { 11107c478bd9Sstevel@tonic-gate physinstalled -= (physmax - (PFN_64G - 1)); 11117c478bd9Sstevel@tonic-gate physmax = PFN_64G - 1; 11127c478bd9Sstevel@tonic-gate } 11137c478bd9Sstevel@tonic-gate } else { 11147c478bd9Sstevel@tonic-gate if (PFN_ABOVE4G(physmax)) { 11157c478bd9Sstevel@tonic-gate physinstalled -= (physmax - (PFN_4G - 1)); 11167c478bd9Sstevel@tonic-gate physmax = PFN_4G - 1; 11177c478bd9Sstevel@tonic-gate } 11187c478bd9Sstevel@tonic-gate } 11197c478bd9Sstevel@tonic-gate #endif 11207c478bd9Sstevel@tonic-gate 11217c478bd9Sstevel@tonic-gate startup_build_mem_nodes(bootops->boot_mem->physinstalled); 11227c478bd9Sstevel@tonic-gate 11237c478bd9Sstevel@tonic-gate if (BOP_GETPROPLEN(bootops, "enforce-prot-exec") >= 0) { 11247c478bd9Sstevel@tonic-gate int len = BOP_GETPROPLEN(bootops, "enforce-prot-exec"); 11257c478bd9Sstevel@tonic-gate char value[8]; 11267c478bd9Sstevel@tonic-gate 11277c478bd9Sstevel@tonic-gate if (len < 8) 11287c478bd9Sstevel@tonic-gate (void) BOP_GETPROP(bootops, "enforce-prot-exec", value); 11297c478bd9Sstevel@tonic-gate else 11307c478bd9Sstevel@tonic-gate (void) strcpy(value, ""); 11317c478bd9Sstevel@tonic-gate if (strcmp(value, "off") == 0) 11327c478bd9Sstevel@tonic-gate mmu.pt_nx = 0; 11337c478bd9Sstevel@tonic-gate } 11347c478bd9Sstevel@tonic-gate PRM_DEBUG(mmu.pt_nx); 11357c478bd9Sstevel@tonic-gate 11367c478bd9Sstevel@tonic-gate /* 11377c478bd9Sstevel@tonic-gate * We will need page_t's for every page in the system, except for 11387c478bd9Sstevel@tonic-gate * memory mapped at or above above the start of the kernel text segment. 11397c478bd9Sstevel@tonic-gate * 11407c478bd9Sstevel@tonic-gate * pages above e_modtext are attributed to kernel debugger (obp_pages) 11417c478bd9Sstevel@tonic-gate */ 11427c478bd9Sstevel@tonic-gate npages = physinstalled - 1; /* avail_filter() skips page 0, so "- 1" */ 11437c478bd9Sstevel@tonic-gate obp_pages = 0; 11447c478bd9Sstevel@tonic-gate va = KERNEL_TEXT; 1145ae115bc7Smrj while (kbm_probe(&va, &len, &pfn, &prot) != 0) { 11467c478bd9Sstevel@tonic-gate npages -= len >> MMU_PAGESHIFT; 11477c478bd9Sstevel@tonic-gate if (va >= (uintptr_t)e_moddata) 11487c478bd9Sstevel@tonic-gate obp_pages += len >> MMU_PAGESHIFT; 11497c478bd9Sstevel@tonic-gate va += len; 11507c478bd9Sstevel@tonic-gate } 11517c478bd9Sstevel@tonic-gate PRM_DEBUG(npages); 11527c478bd9Sstevel@tonic-gate PRM_DEBUG(obp_pages); 11537c478bd9Sstevel@tonic-gate 11547c478bd9Sstevel@tonic-gate /* 115535b1ab99Sjosephb * If physmem is patched to be non-zero, use it instead of the computed 115635b1ab99Sjosephb * value unless it is larger than the actual amount of memory on hand. 11577c478bd9Sstevel@tonic-gate */ 11584944b02eSkchow if (physmem == 0 || physmem > npages) { 11597c478bd9Sstevel@tonic-gate physmem = npages; 11604944b02eSkchow } else if (physmem < npages) { 1161f53ad214Skchow orig_npages = npages; 11627c478bd9Sstevel@tonic-gate npages = physmem; 11634944b02eSkchow } 11647c478bd9Sstevel@tonic-gate PRM_DEBUG(physmem); 11657c478bd9Sstevel@tonic-gate 11667c478bd9Sstevel@tonic-gate /* 11677c478bd9Sstevel@tonic-gate * We now compute the sizes of all the initial allocations for 11687c478bd9Sstevel@tonic-gate * structures the kernel needs in order do kmem_alloc(). These 11697c478bd9Sstevel@tonic-gate * include: 11707c478bd9Sstevel@tonic-gate * memsegs 11717c478bd9Sstevel@tonic-gate * memlists 11727c478bd9Sstevel@tonic-gate * page hash table 11737c478bd9Sstevel@tonic-gate * page_t's 11747c478bd9Sstevel@tonic-gate * page coloring data structs 11757c478bd9Sstevel@tonic-gate */ 11767c478bd9Sstevel@tonic-gate memseg_sz = sizeof (struct memseg) * (memblocks + POSS_NEW_FRAGMENTS); 11777c478bd9Sstevel@tonic-gate ADD_TO_ALLOCATIONS(memseg_base, memseg_sz); 11787c478bd9Sstevel@tonic-gate PRM_DEBUG(memseg_sz); 11797c478bd9Sstevel@tonic-gate 11807c478bd9Sstevel@tonic-gate /* 1181ae115bc7Smrj * Reserve space for memlists. There's no real good way to know exactly 1182ae115bc7Smrj * how much room we'll need, but this should be a good upper bound. 11837c478bd9Sstevel@tonic-gate */ 11847c478bd9Sstevel@tonic-gate memlist_sz = ROUND_UP_PAGE(2 * sizeof (struct memlist) * 11857c478bd9Sstevel@tonic-gate (memblocks + POSS_NEW_FRAGMENTS)); 11867c478bd9Sstevel@tonic-gate ADD_TO_ALLOCATIONS(memlist, memlist_sz); 11877c478bd9Sstevel@tonic-gate PRM_DEBUG(memlist_sz); 11887c478bd9Sstevel@tonic-gate 11891de082f7SVikram Hegde /* 11901de082f7SVikram Hegde * Reserve space for bios reserved memlists. 11911de082f7SVikram Hegde */ 11921de082f7SVikram Hegde rsvdmemlist_sz = ROUND_UP_PAGE(2 * sizeof (struct memlist) * 11931de082f7SVikram Hegde (rsvdmemblocks + POSS_NEW_FRAGMENTS)); 11941de082f7SVikram Hegde ADD_TO_ALLOCATIONS(bios_rsvd, rsvdmemlist_sz); 11951de082f7SVikram Hegde PRM_DEBUG(rsvdmemlist_sz); 11961de082f7SVikram Hegde 1197cb15d5d9SPeter Rival /* LINTED */ 1198cb15d5d9SPeter Rival ASSERT(P2SAMEHIGHBIT((1 << PP_SHIFT), sizeof (struct page))); 11997c478bd9Sstevel@tonic-gate /* 12007c478bd9Sstevel@tonic-gate * The page structure hash table size is a power of 2 12017c478bd9Sstevel@tonic-gate * such that the average hash chain length is PAGE_HASHAVELEN. 12027c478bd9Sstevel@tonic-gate */ 12037c478bd9Sstevel@tonic-gate page_hashsz = npages / PAGE_HASHAVELEN; 1204cb15d5d9SPeter Rival page_hashsz_shift = highbit(page_hashsz); 1205cb15d5d9SPeter Rival page_hashsz = 1 << page_hashsz_shift; 12067c478bd9Sstevel@tonic-gate pagehash_sz = sizeof (struct page *) * page_hashsz; 12077c478bd9Sstevel@tonic-gate ADD_TO_ALLOCATIONS(page_hash, pagehash_sz); 12087c478bd9Sstevel@tonic-gate PRM_DEBUG(pagehash_sz); 12097c478bd9Sstevel@tonic-gate 12107c478bd9Sstevel@tonic-gate /* 1211ae115bc7Smrj * Set aside room for the page structures themselves. 12127c478bd9Sstevel@tonic-gate */ 1213ae115bc7Smrj PRM_DEBUG(npages); 1214ae115bc7Smrj pp_sz = sizeof (struct page) * npages; 12157c478bd9Sstevel@tonic-gate ADD_TO_ALLOCATIONS(pp_base, pp_sz); 12167c478bd9Sstevel@tonic-gate PRM_DEBUG(pp_sz); 12177c478bd9Sstevel@tonic-gate 12187c478bd9Sstevel@tonic-gate /* 12197c478bd9Sstevel@tonic-gate * determine l2 cache info and memory size for page coloring 12207c478bd9Sstevel@tonic-gate */ 12217c478bd9Sstevel@tonic-gate (void) getl2cacheinfo(CPU, 12227c478bd9Sstevel@tonic-gate &l2cache_sz, &l2cache_linesz, &l2cache_assoc); 12237c478bd9Sstevel@tonic-gate pagecolor_memsz = 12247c478bd9Sstevel@tonic-gate page_coloring_init(l2cache_sz, l2cache_linesz, l2cache_assoc); 12257c478bd9Sstevel@tonic-gate ADD_TO_ALLOCATIONS(pagecolor_mem, pagecolor_memsz); 12267c478bd9Sstevel@tonic-gate PRM_DEBUG(pagecolor_memsz); 12277c478bd9Sstevel@tonic-gate 12287c478bd9Sstevel@tonic-gate page_ctrs_size = page_ctrs_sz(); 12297c478bd9Sstevel@tonic-gate ADD_TO_ALLOCATIONS(page_ctrs_mem, page_ctrs_size); 12307c478bd9Sstevel@tonic-gate PRM_DEBUG(page_ctrs_size); 12317c478bd9Sstevel@tonic-gate 1232d7d93655Sblakej /* 1233d7d93655Sblakej * Allocate the array that protects pp->p_selock. 1234d7d93655Sblakej */ 1235d7d93655Sblakej pse_shift = size_pse_array(physmem, max_ncpus); 1236d7d93655Sblakej pse_table_size = 1 << pse_shift; 1237d7d93655Sblakej pse_table_alloc_size = pse_table_size * sizeof (pad_mutex_t); 1238d7d93655Sblakej ADD_TO_ALLOCATIONS(pse_mutex, pse_table_alloc_size); 1239d7d93655Sblakej 1240ae115bc7Smrj #if defined(__amd64) 12417c478bd9Sstevel@tonic-gate valloc_sz = ROUND_UP_LPAGE(valloc_sz); 1242ae115bc7Smrj valloc_base = VALLOC_BASE; 124335b1ab99Sjosephb 124435b1ab99Sjosephb /* 124535b1ab99Sjosephb * The default values of VALLOC_BASE and SEGKPM_BASE should work 124635b1ab99Sjosephb * for values of physmax up to 1 Terabyte. They need adjusting when 12476c8c89eaSKit Chow * memory is at addresses above 1 TB. When adjusted, segkpm_base must 12486c8c89eaSKit Chow * be aligned on KERNEL_REDZONE_SIZE boundary (span of top level pte). 124935b1ab99Sjosephb */ 1250a3114836SGerry Liu if (physmax + 1 > mmu_btop(TERABYTE) || 1251a3114836SGerry Liu plat_dr_physmax > mmu_btop(TERABYTE)) { 125235b1ab99Sjosephb uint64_t kpm_resv_amount = mmu_ptob(physmax + 1); 125335b1ab99Sjosephb 1254a3114836SGerry Liu if (kpm_resv_amount < mmu_ptob(plat_dr_physmax)) { 1255a3114836SGerry Liu kpm_resv_amount = mmu_ptob(plat_dr_physmax); 1256a3114836SGerry Liu } 1257a3114836SGerry Liu 12586c8c89eaSKit Chow segkpm_base = -(P2ROUNDUP((2 * kpm_resv_amount), 12596c8c89eaSKit Chow KERNEL_REDZONE_SIZE)); /* down from top VA */ 126035b1ab99Sjosephb 126135b1ab99Sjosephb /* make sure we leave some space for user apps above hole */ 126235b1ab99Sjosephb segkpm_base = MAX(segkpm_base, AMD64_VA_HOLE_END + TERABYTE); 126335b1ab99Sjosephb if (segkpm_base > SEGKPM_BASE) 126435b1ab99Sjosephb segkpm_base = SEGKPM_BASE; 126535b1ab99Sjosephb PRM_DEBUG(segkpm_base); 126635b1ab99Sjosephb 12676c8c89eaSKit Chow valloc_base = segkpm_base + P2ROUNDUP(kpm_resv_amount, ONE_GIG); 12686c8c89eaSKit Chow if (valloc_base < segkpm_base) 12696c8c89eaSKit Chow panic("not enough kernel VA to support memory size"); 127035b1ab99Sjosephb PRM_DEBUG(valloc_base); 127135b1ab99Sjosephb } 1272ae115bc7Smrj #else /* __i386 */ 1273ae115bc7Smrj valloc_base = (uintptr_t)(MISC_VA_BASE - valloc_sz); 1274ae115bc7Smrj valloc_base = P2ALIGN(valloc_base, mmu.level_size[1]); 12757c478bd9Sstevel@tonic-gate PRM_DEBUG(valloc_base); 127635b1ab99Sjosephb #endif /* __i386 */ 1277ae115bc7Smrj 1278ae115bc7Smrj /* 1279ae115bc7Smrj * do all the initial allocations 1280ae115bc7Smrj */ 1281ae115bc7Smrj perform_allocations(); 1282ae115bc7Smrj 1283ae115bc7Smrj /* 1284ae115bc7Smrj * Build phys_install and phys_avail in kernel memspace. 1285ae115bc7Smrj * - phys_install should be all memory in the system. 1286ae115bc7Smrj * - phys_avail is phys_install minus any memory mapped before this 1287ae115bc7Smrj * point above KERNEL_TEXT. 1288ae115bc7Smrj */ 1289ae115bc7Smrj current = phys_install = memlist; 1290ae115bc7Smrj copy_memlist_filter(bootops->boot_mem->physinstalled, ¤t, NULL); 1291ae115bc7Smrj if ((caddr_t)current > (caddr_t)memlist + memlist_sz) 1292ae115bc7Smrj panic("physinstalled was too big!"); 1293ae115bc7Smrj if (prom_debug) 1294ae115bc7Smrj print_memlist("phys_install", phys_install); 1295ae115bc7Smrj 1296ae115bc7Smrj phys_avail = current; 1297ae115bc7Smrj PRM_POINT("Building phys_avail:\n"); 1298ae115bc7Smrj copy_memlist_filter(bootops->boot_mem->physinstalled, ¤t, 1299ae115bc7Smrj avail_filter); 1300ae115bc7Smrj if ((caddr_t)current > (caddr_t)memlist + memlist_sz) 1301ae115bc7Smrj panic("physavail was too big!"); 1302ae115bc7Smrj if (prom_debug) 1303ae115bc7Smrj print_memlist("phys_avail", phys_avail); 1304a3114836SGerry Liu #ifndef __xpv 1305a3114836SGerry Liu /* 1306a3114836SGerry Liu * Free unused memlist items, which may be used by memory DR driver 1307a3114836SGerry Liu * at runtime. 1308a3114836SGerry Liu */ 1309a3114836SGerry Liu if ((caddr_t)current < (caddr_t)memlist + memlist_sz) { 1310a3114836SGerry Liu memlist_free_block((caddr_t)current, 1311a3114836SGerry Liu (caddr_t)memlist + memlist_sz - (caddr_t)current); 1312a3114836SGerry Liu } 1313a3114836SGerry Liu #endif 1314ae115bc7Smrj 13151de082f7SVikram Hegde /* 13161de082f7SVikram Hegde * Build bios reserved memspace 13171de082f7SVikram Hegde */ 13181de082f7SVikram Hegde current = bios_rsvd; 13191de082f7SVikram Hegde copy_memlist_filter(bootops->boot_mem->rsvdmem, ¤t, NULL); 13201de082f7SVikram Hegde if ((caddr_t)current > (caddr_t)bios_rsvd + rsvdmemlist_sz) 13211de082f7SVikram Hegde panic("bios_rsvd was too big!"); 13221de082f7SVikram Hegde if (prom_debug) 13231de082f7SVikram Hegde print_memlist("bios_rsvd", bios_rsvd); 1324a3114836SGerry Liu #ifndef __xpv 1325a3114836SGerry Liu /* 1326a3114836SGerry Liu * Free unused memlist items, which may be used by memory DR driver 1327a3114836SGerry Liu * at runtime. 1328a3114836SGerry Liu */ 1329a3114836SGerry Liu if ((caddr_t)current < (caddr_t)bios_rsvd + rsvdmemlist_sz) { 1330a3114836SGerry Liu memlist_free_block((caddr_t)current, 1331a3114836SGerry Liu (caddr_t)bios_rsvd + rsvdmemlist_sz - (caddr_t)current); 1332a3114836SGerry Liu } 1333a3114836SGerry Liu #endif 13341de082f7SVikram Hegde 1335ae115bc7Smrj /* 1336ae115bc7Smrj * setup page coloring 1337ae115bc7Smrj */ 1338ae115bc7Smrj page_coloring_setup(pagecolor_mem); 1339ae115bc7Smrj page_lock_init(); /* currently a no-op */ 1340ae115bc7Smrj 1341ae115bc7Smrj /* 1342ae115bc7Smrj * free page list counters 1343ae115bc7Smrj */ 1344ae115bc7Smrj (void) page_ctrs_alloc(page_ctrs_mem); 1345ae115bc7Smrj 134606fb6a36Sdv /* 134706fb6a36Sdv * Size the pcf array based on the number of cpus in the box at 134806fb6a36Sdv * boot time. 134906fb6a36Sdv */ 135006fb6a36Sdv 135106fb6a36Sdv pcf_init(); 135206fb6a36Sdv 1353ae115bc7Smrj /* 1354ae115bc7Smrj * Initialize the page structures from the memory lists. 1355ae115bc7Smrj */ 1356ae115bc7Smrj availrmem_initial = availrmem = freemem = 0; 1357ae115bc7Smrj PRM_POINT("Calling kphysm_init()..."); 1358ae115bc7Smrj npages = kphysm_init(pp_base, npages); 1359ae115bc7Smrj PRM_POINT("kphysm_init() done"); 1360ae115bc7Smrj PRM_DEBUG(npages); 1361ae115bc7Smrj 1362ae115bc7Smrj init_debug_info(); 1363ae115bc7Smrj 1364ae115bc7Smrj /* 1365ae115bc7Smrj * Now that page_t's have been initialized, remove all the 1366ae115bc7Smrj * initial allocation pages from the kernel free page lists. 1367ae115bc7Smrj */ 1368ae115bc7Smrj boot_mapin((caddr_t)valloc_base, valloc_sz); 13690cfdb603Sjosephb boot_mapin((caddr_t)MISC_VA_BASE, MISC_VA_SIZE); 1370ae115bc7Smrj PRM_POINT("startup_memlist() done"); 1371ae115bc7Smrj 1372ae115bc7Smrj PRM_DEBUG(valloc_sz); 13732cb27123Saguzovsk 1374567d55e1Saguzovsk #if defined(__amd64) 1375567d55e1Saguzovsk if ((availrmem >> (30 - MMU_PAGESHIFT)) >= 1376567d55e1Saguzovsk textrepl_min_gb && l2cache_sz <= 2 << 20) { 1377567d55e1Saguzovsk extern size_t textrepl_size_thresh; 13782cb27123Saguzovsk textrepl_size_thresh = (16 << 20) - 1; 13792cb27123Saguzovsk } 1380567d55e1Saguzovsk #endif 1381ae115bc7Smrj } 1382ae115bc7Smrj 1383ae115bc7Smrj /* 1384ae115bc7Smrj * Layout the kernel's part of address space and initialize kmem allocator. 1385ae115bc7Smrj */ 1386ae115bc7Smrj static void 1387ae115bc7Smrj startup_kmem(void) 1388ae115bc7Smrj { 1389932dc8e5Sdp extern void page_set_colorequiv_arr(void); 1390932dc8e5Sdp 1391ae115bc7Smrj PRM_POINT("startup_kmem() starting..."); 13927c478bd9Sstevel@tonic-gate 13937c478bd9Sstevel@tonic-gate #if defined(__amd64) 13947c478bd9Sstevel@tonic-gate if (eprom_kernelbase && eprom_kernelbase != KERNELBASE) 13957c478bd9Sstevel@tonic-gate cmn_err(CE_NOTE, "!kernelbase cannot be changed on 64-bit " 13967c478bd9Sstevel@tonic-gate "systems."); 139735b1ab99Sjosephb kernelbase = segkpm_base - KERNEL_REDZONE_SIZE; 13987c478bd9Sstevel@tonic-gate core_base = (uintptr_t)COREHEAP_BASE; 1399ae115bc7Smrj core_size = (size_t)MISC_VA_BASE - COREHEAP_BASE; 14007c478bd9Sstevel@tonic-gate #else /* __i386 */ 14017c478bd9Sstevel@tonic-gate /* 14027c478bd9Sstevel@tonic-gate * We configure kernelbase based on: 14037c478bd9Sstevel@tonic-gate * 14047c478bd9Sstevel@tonic-gate * 1. user specified kernelbase via eeprom command. Value cannot exceed 14057c478bd9Sstevel@tonic-gate * KERNELBASE_MAX. we large page align eprom_kernelbase 14067c478bd9Sstevel@tonic-gate * 14077c478bd9Sstevel@tonic-gate * 2. Default to KERNELBASE and adjust to 2X less the size for page_t. 14087c478bd9Sstevel@tonic-gate * On large memory systems we must lower kernelbase to allow 14097c478bd9Sstevel@tonic-gate * enough room for page_t's for all of memory. 14107c478bd9Sstevel@tonic-gate * 14117c478bd9Sstevel@tonic-gate * The value set here, might be changed a little later. 14127c478bd9Sstevel@tonic-gate */ 14137c478bd9Sstevel@tonic-gate if (eprom_kernelbase) { 14147c478bd9Sstevel@tonic-gate kernelbase = eprom_kernelbase & mmu.level_mask[1]; 14157c478bd9Sstevel@tonic-gate if (kernelbase > KERNELBASE_MAX) 14167c478bd9Sstevel@tonic-gate kernelbase = KERNELBASE_MAX; 14177c478bd9Sstevel@tonic-gate } else { 14187c478bd9Sstevel@tonic-gate kernelbase = (uintptr_t)KERNELBASE; 14197c478bd9Sstevel@tonic-gate kernelbase -= ROUND_UP_4MEG(2 * valloc_sz); 14207c478bd9Sstevel@tonic-gate } 14217c478bd9Sstevel@tonic-gate ASSERT((kernelbase & mmu.level_offset[1]) == 0); 1422ae115bc7Smrj core_base = valloc_base; 14237c478bd9Sstevel@tonic-gate core_size = 0; 1424ae115bc7Smrj #endif /* __i386 */ 14257c478bd9Sstevel@tonic-gate 14267c478bd9Sstevel@tonic-gate PRM_DEBUG(core_base); 14277c478bd9Sstevel@tonic-gate PRM_DEBUG(core_size); 1428ae115bc7Smrj PRM_DEBUG(kernelbase); 14297c478bd9Sstevel@tonic-gate 143035b1ab99Sjosephb #if defined(__i386) 14317c478bd9Sstevel@tonic-gate segkp_fromheap = 1; 1432ae115bc7Smrj #endif /* __i386 */ 143335b1ab99Sjosephb 14347c478bd9Sstevel@tonic-gate ekernelheap = (char *)core_base; 1435ae115bc7Smrj PRM_DEBUG(ekernelheap); 14367c478bd9Sstevel@tonic-gate 14377c478bd9Sstevel@tonic-gate /* 14387c478bd9Sstevel@tonic-gate * Now that we know the real value of kernelbase, 14397c478bd9Sstevel@tonic-gate * update variables that were initialized with a value of 14407c478bd9Sstevel@tonic-gate * KERNELBASE (in common/conf/param.c). 14417c478bd9Sstevel@tonic-gate * 14427c478bd9Sstevel@tonic-gate * XXX The problem with this sort of hackery is that the 14437c478bd9Sstevel@tonic-gate * compiler just may feel like putting the const declarations 14447c478bd9Sstevel@tonic-gate * (in param.c) into the .text section. Perhaps they should 14457c478bd9Sstevel@tonic-gate * just be declared as variables there? 14467c478bd9Sstevel@tonic-gate */ 14477c478bd9Sstevel@tonic-gate 14487c478bd9Sstevel@tonic-gate *(uintptr_t *)&_kernelbase = kernelbase; 14497c478bd9Sstevel@tonic-gate *(uintptr_t *)&_userlimit = kernelbase; 1450ccbaea4fSjosephb #if defined(__amd64) 1451ccbaea4fSjosephb *(uintptr_t *)&_userlimit -= KERNELBASE - USERLIMIT; 1452ccbaea4fSjosephb #else 14537c478bd9Sstevel@tonic-gate *(uintptr_t *)&_userlimit32 = _userlimit; 14547c478bd9Sstevel@tonic-gate #endif 14557c478bd9Sstevel@tonic-gate PRM_DEBUG(_kernelbase); 14567c478bd9Sstevel@tonic-gate PRM_DEBUG(_userlimit); 14577c478bd9Sstevel@tonic-gate PRM_DEBUG(_userlimit32); 14587c478bd9Sstevel@tonic-gate 145935b1ab99Sjosephb layout_kernel_va(); 146035b1ab99Sjosephb 146135b1ab99Sjosephb #if defined(__i386) 146235b1ab99Sjosephb /* 146335b1ab99Sjosephb * If segmap is too large we can push the bottom of the kernel heap 146435b1ab99Sjosephb * higher than the base. Or worse, it could exceed the top of the 146535b1ab99Sjosephb * VA space entirely, causing it to wrap around. 146635b1ab99Sjosephb */ 146735b1ab99Sjosephb if (kernelheap >= ekernelheap || (uintptr_t)kernelheap < kernelbase) 146835b1ab99Sjosephb panic("too little address space available for kernelheap," 146935b1ab99Sjosephb " use eeprom for lower kernelbase or smaller segmapsize"); 147035b1ab99Sjosephb #endif /* __i386 */ 147135b1ab99Sjosephb 14727c478bd9Sstevel@tonic-gate /* 14737c478bd9Sstevel@tonic-gate * Initialize the kernel heap. Note 3rd argument must be > 1st. 14747c478bd9Sstevel@tonic-gate */ 147535b1ab99Sjosephb kernelheap_init(kernelheap, ekernelheap, 147635b1ab99Sjosephb kernelheap + MMU_PAGESIZE, 1477ae115bc7Smrj (void *)core_base, (void *)(core_base + core_size)); 14787c478bd9Sstevel@tonic-gate 1479843e1988Sjohnlev #if defined(__xpv) 1480843e1988Sjohnlev /* 1481843e1988Sjohnlev * Link pending events struct into cpu struct 1482843e1988Sjohnlev */ 1483843e1988Sjohnlev CPU->cpu_m.mcpu_evt_pend = &cpu0_evt_data; 1484843e1988Sjohnlev #endif 14857c478bd9Sstevel@tonic-gate /* 14867c478bd9Sstevel@tonic-gate * Initialize kernel memory allocator. 14877c478bd9Sstevel@tonic-gate */ 14887c478bd9Sstevel@tonic-gate kmem_init(); 14897c478bd9Sstevel@tonic-gate 1490932dc8e5Sdp /* 1491932dc8e5Sdp * Factor in colorequiv to check additional 'equivalent' bins 1492932dc8e5Sdp */ 1493932dc8e5Sdp page_set_colorequiv_arr(); 1494932dc8e5Sdp 14957c478bd9Sstevel@tonic-gate /* 14967c478bd9Sstevel@tonic-gate * print this out early so that we know what's going on 14977c478bd9Sstevel@tonic-gate */ 14987417cfdeSKuriakose Kuruvilla print_x86_featureset(x86_featureset); 14997c478bd9Sstevel@tonic-gate 15007c478bd9Sstevel@tonic-gate /* 15017c478bd9Sstevel@tonic-gate * Initialize bp_mapin(). 15027c478bd9Sstevel@tonic-gate */ 15037c478bd9Sstevel@tonic-gate bp_init(MMU_PAGESIZE, HAT_STORECACHING_OK); 15047c478bd9Sstevel@tonic-gate 1505f53ad214Skchow /* 1506f53ad214Skchow * orig_npages is non-zero if physmem has been configured for less 1507f53ad214Skchow * than the available memory. 1508f53ad214Skchow */ 1509f53ad214Skchow if (orig_npages) { 151035b1ab99Sjosephb cmn_err(CE_WARN, "!%slimiting physmem to 0x%lx of 0x%lx pages", 151135b1ab99Sjosephb (npages == PHYSMEM ? "Due to virtual address space " : ""), 151235b1ab99Sjosephb npages, orig_npages); 1513f53ad214Skchow } 15147c478bd9Sstevel@tonic-gate #if defined(__i386) 15157c478bd9Sstevel@tonic-gate if (eprom_kernelbase && (eprom_kernelbase != kernelbase)) 15167c478bd9Sstevel@tonic-gate cmn_err(CE_WARN, "kernelbase value, User specified 0x%lx, " 15177c478bd9Sstevel@tonic-gate "System using 0x%lx", 15187c478bd9Sstevel@tonic-gate (uintptr_t)eprom_kernelbase, (uintptr_t)kernelbase); 15197c478bd9Sstevel@tonic-gate #endif 15207c478bd9Sstevel@tonic-gate 15217c478bd9Sstevel@tonic-gate #ifdef KERNELBASE_ABI_MIN 15227c478bd9Sstevel@tonic-gate if (kernelbase < (uintptr_t)KERNELBASE_ABI_MIN) { 15237c478bd9Sstevel@tonic-gate cmn_err(CE_NOTE, "!kernelbase set to 0x%lx, system is not " 15247c478bd9Sstevel@tonic-gate "i386 ABI compliant.", (uintptr_t)kernelbase); 15257c478bd9Sstevel@tonic-gate } 15267c478bd9Sstevel@tonic-gate #endif 15277c478bd9Sstevel@tonic-gate 1528a3114836SGerry Liu #ifndef __xpv 1529a3114836SGerry Liu if (plat_dr_support_memory()) { 1530a3114836SGerry Liu mem_config_init(); 1531a3114836SGerry Liu } 1532a3114836SGerry Liu #else /* __xpv */ 1533843e1988Sjohnlev /* 1534843e1988Sjohnlev * Some of the xen start information has to be relocated up 1535843e1988Sjohnlev * into the kernel's permanent address space. 1536843e1988Sjohnlev */ 1537843e1988Sjohnlev PRM_POINT("calling xen_relocate_start_info()"); 1538843e1988Sjohnlev xen_relocate_start_info(); 1539843e1988Sjohnlev PRM_POINT("xen_relocate_start_info() done"); 1540843e1988Sjohnlev 1541843e1988Sjohnlev /* 1542843e1988Sjohnlev * (Update the vcpu pointer in our cpu structure to point into 1543843e1988Sjohnlev * the relocated shared info.) 1544843e1988Sjohnlev */ 1545843e1988Sjohnlev CPU->cpu_m.mcpu_vcpu_info = 1546843e1988Sjohnlev &HYPERVISOR_shared_info->vcpu_info[CPU->cpu_id]; 1547a3114836SGerry Liu #endif /* __xpv */ 1548843e1988Sjohnlev 1549ae115bc7Smrj PRM_POINT("startup_kmem() done"); 15507c478bd9Sstevel@tonic-gate } 15517c478bd9Sstevel@tonic-gate 1552551bc2a6Smrj #ifndef __xpv 1553551bc2a6Smrj /* 1554551bc2a6Smrj * If we have detected that we are running in an HVM environment, we need 1555551bc2a6Smrj * to prepend the PV driver directory to the module search path. 1556551bc2a6Smrj */ 1557551bc2a6Smrj #define HVM_MOD_DIR "/platform/i86hvm/kernel" 1558551bc2a6Smrj static void 1559551bc2a6Smrj update_default_path() 1560551bc2a6Smrj { 1561551bc2a6Smrj char *current, *newpath; 1562551bc2a6Smrj int newlen; 1563551bc2a6Smrj 1564551bc2a6Smrj /* 1565551bc2a6Smrj * We are about to resync with krtld. krtld will reset its 1566551bc2a6Smrj * internal module search path iff Solaris has set default_path. 1567551bc2a6Smrj * We want to be sure we're prepending this new directory to the 1568551bc2a6Smrj * right search path. 1569551bc2a6Smrj */ 1570551bc2a6Smrj current = (default_path == NULL) ? kobj_module_path : default_path; 1571551bc2a6Smrj 1572b6611e3bSStuart Maybee newlen = strlen(HVM_MOD_DIR) + strlen(current) + 2; 1573551bc2a6Smrj newpath = kmem_alloc(newlen, KM_SLEEP); 1574551bc2a6Smrj (void) strcpy(newpath, HVM_MOD_DIR); 1575551bc2a6Smrj (void) strcat(newpath, " "); 1576551bc2a6Smrj (void) strcat(newpath, current); 1577551bc2a6Smrj 1578551bc2a6Smrj default_path = newpath; 1579551bc2a6Smrj } 1580551bc2a6Smrj #endif 1581551bc2a6Smrj 15827c478bd9Sstevel@tonic-gate static void 15837c478bd9Sstevel@tonic-gate startup_modules(void) 15847c478bd9Sstevel@tonic-gate { 1585ed5289f9SKen Erickson int cnt; 15867c478bd9Sstevel@tonic-gate extern void prom_setup(void); 1587ed5289f9SKen Erickson int32_t v, h; 1588ed5289f9SKen Erickson char d[11]; 1589ed5289f9SKen Erickson char *cp; 1590e4b86885SCheng Sean Ye cmi_hdl_t hdl; 15917c478bd9Sstevel@tonic-gate 15927c478bd9Sstevel@tonic-gate PRM_POINT("startup_modules() starting..."); 1593843e1988Sjohnlev 1594843e1988Sjohnlev #ifndef __xpv 15957c478bd9Sstevel@tonic-gate /* 15967c478bd9Sstevel@tonic-gate * Initialize ten-micro second timer so that drivers will 15977c478bd9Sstevel@tonic-gate * not get short changed in their init phase. This was 15987c478bd9Sstevel@tonic-gate * not getting called until clkinit which, on fast cpu's 15997c478bd9Sstevel@tonic-gate * caused the drv_usecwait to be way too short. 16007c478bd9Sstevel@tonic-gate */ 16017c478bd9Sstevel@tonic-gate microfind(); 1602551bc2a6Smrj 160379ec9da8SYuri Pankov if ((get_hwenv() & HW_XEN_HVM) != 0) 1604551bc2a6Smrj update_default_path(); 1605843e1988Sjohnlev #endif 16067c478bd9Sstevel@tonic-gate 16077c478bd9Sstevel@tonic-gate /* 16087c478bd9Sstevel@tonic-gate * Read the GMT lag from /etc/rtc_config. 16097c478bd9Sstevel@tonic-gate */ 1610ae115bc7Smrj sgmtl(process_rtc_config_file()); 16117c478bd9Sstevel@tonic-gate 16127c478bd9Sstevel@tonic-gate /* 16137c478bd9Sstevel@tonic-gate * Calculate default settings of system parameters based upon 16147c478bd9Sstevel@tonic-gate * maxusers, yet allow to be overridden via the /etc/system file. 16157c478bd9Sstevel@tonic-gate */ 16167c478bd9Sstevel@tonic-gate param_calc(0); 16177c478bd9Sstevel@tonic-gate 16187c478bd9Sstevel@tonic-gate mod_setup(); 16197c478bd9Sstevel@tonic-gate 16207c478bd9Sstevel@tonic-gate /* 16217c478bd9Sstevel@tonic-gate * Initialize system parameters. 16227c478bd9Sstevel@tonic-gate */ 16237c478bd9Sstevel@tonic-gate param_init(); 16247c478bd9Sstevel@tonic-gate 16259acbbeafSnn /* 16269acbbeafSnn * Initialize the default brands 16279acbbeafSnn */ 16289acbbeafSnn brand_init(); 16299acbbeafSnn 16307c478bd9Sstevel@tonic-gate /* 16317c478bd9Sstevel@tonic-gate * maxmem is the amount of physical memory we're playing with. 16327c478bd9Sstevel@tonic-gate */ 16337c478bd9Sstevel@tonic-gate maxmem = physmem; 16347c478bd9Sstevel@tonic-gate 16357c478bd9Sstevel@tonic-gate /* 16367c478bd9Sstevel@tonic-gate * Initialize segment management stuff. 16377c478bd9Sstevel@tonic-gate */ 16387c478bd9Sstevel@tonic-gate seg_init(); 16397c478bd9Sstevel@tonic-gate 16407c478bd9Sstevel@tonic-gate if (modload("fs", "specfs") == -1) 16417c478bd9Sstevel@tonic-gate halt("Can't load specfs"); 16427c478bd9Sstevel@tonic-gate 16437c478bd9Sstevel@tonic-gate if (modload("fs", "devfs") == -1) 16447c478bd9Sstevel@tonic-gate halt("Can't load devfs"); 16457c478bd9Sstevel@tonic-gate 1646facf4a8dSllai if (modload("fs", "dev") == -1) 1647facf4a8dSllai halt("Can't load dev"); 1648facf4a8dSllai 164935a5a358SJonathan Adams if (modload("fs", "procfs") == -1) 165035a5a358SJonathan Adams halt("Can't load procfs"); 165135a5a358SJonathan Adams 165245916cd2Sjpk (void) modloadonly("sys", "lbl_edition"); 165345916cd2Sjpk 16547c478bd9Sstevel@tonic-gate dispinit(); 16557c478bd9Sstevel@tonic-gate 16567c478bd9Sstevel@tonic-gate /* Read cluster configuration data. */ 16577c478bd9Sstevel@tonic-gate clconf_init(); 16587c478bd9Sstevel@tonic-gate 1659843e1988Sjohnlev #if defined(__xpv) 1660349b53ddSStuart Maybee (void) ec_init(); 1661843e1988Sjohnlev gnttab_init(); 1662843e1988Sjohnlev (void) xs_early_init(); 1663843e1988Sjohnlev #endif /* __xpv */ 1664843e1988Sjohnlev 16657c478bd9Sstevel@tonic-gate /* 16667c478bd9Sstevel@tonic-gate * Create a kernel device tree. First, create rootnex and 16677c478bd9Sstevel@tonic-gate * then invoke bus specific code to probe devices. 16687c478bd9Sstevel@tonic-gate */ 16697c478bd9Sstevel@tonic-gate setup_ddi(); 16707aec1d6eScindi 1671074bb90dSTom Pothier #ifdef __xpv 1672074bb90dSTom Pothier if (DOMAIN_IS_INITDOMAIN(xen_info)) 1673074bb90dSTom Pothier #endif 1674074bb90dSTom Pothier { 1675e41d648bSToomas Soome id_t smid; 1676e41d648bSToomas Soome smbios_system_t smsys; 1677e41d648bSToomas Soome smbios_info_t sminfo; 1678e41d648bSToomas Soome char *mfg; 1679074bb90dSTom Pothier /* 1680074bb90dSTom Pothier * Load the System Management BIOS into the global ksmbios 1681074bb90dSTom Pothier * handle, if an SMBIOS is present on this system. 1682e41d648bSToomas Soome * Also set "si-hw-provider" property, if not already set. 1683074bb90dSTom Pothier */ 1684074bb90dSTom Pothier ksmbios = smbios_open(NULL, SMB_VERSION, ksmbios_flags, NULL); 1685e41d648bSToomas Soome if (ksmbios != NULL && 1686e41d648bSToomas Soome ((smid = smbios_info_system(ksmbios, &smsys)) != SMB_ERR) && 1687e41d648bSToomas Soome (smbios_info_common(ksmbios, smid, &sminfo)) != SMB_ERR) { 1688e41d648bSToomas Soome mfg = (char *)sminfo.smbi_manufacturer; 1689e41d648bSToomas Soome if (BOP_GETPROPLEN(bootops, "si-hw-provider") < 0) { 1690e41d648bSToomas Soome extern char hw_provider[]; 1691e41d648bSToomas Soome int i; 1692e41d648bSToomas Soome for (i = 0; i < SYS_NMLN; i++) { 1693e41d648bSToomas Soome if (isprint(mfg[i])) 1694e41d648bSToomas Soome hw_provider[i] = mfg[i]; 1695e41d648bSToomas Soome else { 1696e41d648bSToomas Soome hw_provider[i] = '\0'; 1697e41d648bSToomas Soome break; 1698e41d648bSToomas Soome } 1699e41d648bSToomas Soome } 1700e41d648bSToomas Soome hw_provider[SYS_NMLN - 1] = '\0'; 1701e41d648bSToomas Soome } 1702e41d648bSToomas Soome } 1703074bb90dSTom Pothier } 1704074bb90dSTom Pothier 1705074bb90dSTom Pothier 17069e3e4df2SGarrett D'Amore /* 17079e3e4df2SGarrett D'Amore * Originally clconf_init() apparently needed the hostid. But 17089e3e4df2SGarrett D'Amore * this no longer appears to be true - it uses its own nodeid. 17099e3e4df2SGarrett D'Amore * By placing the hostid logic here, we are able to make use of 17109e3e4df2SGarrett D'Amore * the SMBIOS UUID. 17119e3e4df2SGarrett D'Amore */ 17129e3e4df2SGarrett D'Amore if ((h = set_soft_hostid()) == HW_INVALID_HOSTID) { 17139e3e4df2SGarrett D'Amore cmn_err(CE_WARN, "Unable to set hostid"); 17149e3e4df2SGarrett D'Amore } else { 17159e3e4df2SGarrett D'Amore for (v = h, cnt = 0; cnt < 10; cnt++) { 17169e3e4df2SGarrett D'Amore d[cnt] = (char)(v % 10); 17179e3e4df2SGarrett D'Amore v /= 10; 17189e3e4df2SGarrett D'Amore if (v == 0) 17199e3e4df2SGarrett D'Amore break; 17209e3e4df2SGarrett D'Amore } 17219e3e4df2SGarrett D'Amore for (cp = hw_serial; cnt >= 0; cnt--) 17229e3e4df2SGarrett D'Amore *cp++ = d[cnt] + '0'; 17239e3e4df2SGarrett D'Amore *cp = 0; 17249e3e4df2SGarrett D'Amore } 17259e3e4df2SGarrett D'Amore 1726e4b86885SCheng Sean Ye /* 1727e4b86885SCheng Sean Ye * Set up the CPU module subsystem for the boot cpu in the native 1728e4b86885SCheng Sean Ye * case, and all physical cpu resource in the xpv dom0 case. 1729e4b86885SCheng Sean Ye * Modifies the device tree, so this must be done after 1730e4b86885SCheng Sean Ye * setup_ddi(). 1731e4b86885SCheng Sean Ye */ 1732e4b86885SCheng Sean Ye #ifdef __xpv 1733e4b86885SCheng Sean Ye /* 1734e4b86885SCheng Sean Ye * If paravirtualized and on dom0 then we initialize all physical 1735e4b86885SCheng Sean Ye * cpu handles now; if paravirtualized on a domU then do not 1736e4b86885SCheng Sean Ye * initialize. 1737e4b86885SCheng Sean Ye */ 1738e4b86885SCheng Sean Ye if (DOMAIN_IS_INITDOMAIN(xen_info)) { 1739e4b86885SCheng Sean Ye xen_mc_lcpu_cookie_t cpi; 1740e4b86885SCheng Sean Ye 1741e4b86885SCheng Sean Ye for (cpi = xen_physcpu_next(NULL); cpi != NULL; 1742e4b86885SCheng Sean Ye cpi = xen_physcpu_next(cpi)) { 1743e4b86885SCheng Sean Ye if ((hdl = cmi_init(CMI_HDL_SOLARIS_xVM_MCA, 1744e4b86885SCheng Sean Ye xen_physcpu_chipid(cpi), xen_physcpu_coreid(cpi), 1745e4b86885SCheng Sean Ye xen_physcpu_strandid(cpi))) != NULL && 17467417cfdeSKuriakose Kuruvilla is_x86_feature(x86_featureset, X86FSET_MCA)) 174720c794b3Sgavinm cmi_mca_init(hdl); 174820c794b3Sgavinm } 174920c794b3Sgavinm } 1750e4b86885SCheng Sean Ye #else 1751e4b86885SCheng Sean Ye /* 1752e4b86885SCheng Sean Ye * Initialize a handle for the boot cpu - others will initialize 1753e4b86885SCheng Sean Ye * as they startup. Do not do this if we know we are in an HVM domU. 1754e4b86885SCheng Sean Ye */ 175579ec9da8SYuri Pankov if ((get_hwenv() & HW_XEN_HVM) == 0 && 1756e4b86885SCheng Sean Ye (hdl = cmi_init(CMI_HDL_NATIVE, cmi_ntv_hwchipid(CPU), 1757e4b86885SCheng Sean Ye cmi_ntv_hwcoreid(CPU), cmi_ntv_hwstrandid(CPU))) != NULL && 17587417cfdeSKuriakose Kuruvilla is_x86_feature(x86_featureset, X86FSET_MCA)) { 1759e4b86885SCheng Sean Ye cmi_mca_init(hdl); 1760a3114836SGerry Liu CPU->cpu_m.mcpu_cmi_hdl = hdl; 1761a3114836SGerry Liu } 176220c794b3Sgavinm #endif /* __xpv */ 17637aec1d6eScindi 17647c478bd9Sstevel@tonic-gate /* 17657c478bd9Sstevel@tonic-gate * Fake a prom tree such that /dev/openprom continues to work 17667c478bd9Sstevel@tonic-gate */ 1767ae115bc7Smrj PRM_POINT("startup_modules: calling prom_setup..."); 17687c478bd9Sstevel@tonic-gate prom_setup(); 1769ae115bc7Smrj PRM_POINT("startup_modules: done"); 17707c478bd9Sstevel@tonic-gate 17717c478bd9Sstevel@tonic-gate /* 17727c478bd9Sstevel@tonic-gate * Load all platform specific modules 17737c478bd9Sstevel@tonic-gate */ 1774ae115bc7Smrj PRM_POINT("startup_modules: calling psm_modload..."); 17757c478bd9Sstevel@tonic-gate psm_modload(); 17767c478bd9Sstevel@tonic-gate 17777c478bd9Sstevel@tonic-gate PRM_POINT("startup_modules() done"); 17787c478bd9Sstevel@tonic-gate } 17797c478bd9Sstevel@tonic-gate 1780ae115bc7Smrj /* 1781ae115bc7Smrj * claim a "setaside" boot page for use in the kernel 1782ae115bc7Smrj */ 1783ae115bc7Smrj page_t * 1784ae115bc7Smrj boot_claim_page(pfn_t pfn) 17857c478bd9Sstevel@tonic-gate { 1786ae115bc7Smrj page_t *pp; 17877c478bd9Sstevel@tonic-gate 1788ae115bc7Smrj pp = page_numtopp_nolock(pfn); 1789ae115bc7Smrj ASSERT(pp != NULL); 17907c478bd9Sstevel@tonic-gate 1791ae115bc7Smrj if (PP_ISBOOTPAGES(pp)) { 1792ae115bc7Smrj if (pp->p_next != NULL) 1793ae115bc7Smrj pp->p_next->p_prev = pp->p_prev; 1794ae115bc7Smrj if (pp->p_prev == NULL) 1795ae115bc7Smrj bootpages = pp->p_next; 1796ae115bc7Smrj else 1797ae115bc7Smrj pp->p_prev->p_next = pp->p_next; 1798ae115bc7Smrj } else { 1799ae115bc7Smrj /* 1800ae115bc7Smrj * htable_attach() expects a base pagesize page 1801ae115bc7Smrj */ 1802ae115bc7Smrj if (pp->p_szc != 0) 1803ae115bc7Smrj page_boot_demote(pp); 1804ae115bc7Smrj pp = page_numtopp(pfn, SE_EXCL); 1805ae115bc7Smrj } 1806ae115bc7Smrj return (pp); 18077c478bd9Sstevel@tonic-gate } 18087c478bd9Sstevel@tonic-gate 18097c478bd9Sstevel@tonic-gate /* 18107c478bd9Sstevel@tonic-gate * Walk through the pagetables looking for pages mapped in by boot. If the 18117c478bd9Sstevel@tonic-gate * setaside flag is set the pages are expected to be returned to the 18127c478bd9Sstevel@tonic-gate * kernel later in boot, so we add them to the bootpages list. 18137c478bd9Sstevel@tonic-gate */ 18147c478bd9Sstevel@tonic-gate static void 18157c478bd9Sstevel@tonic-gate protect_boot_range(uintptr_t low, uintptr_t high, int setaside) 18167c478bd9Sstevel@tonic-gate { 18177c478bd9Sstevel@tonic-gate uintptr_t va = low; 18187c478bd9Sstevel@tonic-gate size_t len; 18197c478bd9Sstevel@tonic-gate uint_t prot; 18207c478bd9Sstevel@tonic-gate pfn_t pfn; 18217c478bd9Sstevel@tonic-gate page_t *pp; 18227c478bd9Sstevel@tonic-gate pgcnt_t boot_protect_cnt = 0; 18237c478bd9Sstevel@tonic-gate 1824ae115bc7Smrj while (kbm_probe(&va, &len, &pfn, &prot) != 0 && va < high) { 18257c478bd9Sstevel@tonic-gate if (va + len >= high) 18267c478bd9Sstevel@tonic-gate panic("0x%lx byte mapping at 0x%p exceeds boot's " 18277c478bd9Sstevel@tonic-gate "legal range.", len, (void *)va); 18287c478bd9Sstevel@tonic-gate 18297c478bd9Sstevel@tonic-gate while (len > 0) { 18307c478bd9Sstevel@tonic-gate pp = page_numtopp_alloc(pfn); 18317c478bd9Sstevel@tonic-gate if (pp != NULL) { 18327c478bd9Sstevel@tonic-gate if (setaside == 0) 18337c478bd9Sstevel@tonic-gate panic("Unexpected mapping by boot. " 18347c478bd9Sstevel@tonic-gate "addr=%p pfn=%lx\n", 18357c478bd9Sstevel@tonic-gate (void *)va, pfn); 18367c478bd9Sstevel@tonic-gate 18377c478bd9Sstevel@tonic-gate pp->p_next = bootpages; 1838ae115bc7Smrj pp->p_prev = NULL; 1839ae115bc7Smrj PP_SETBOOTPAGES(pp); 1840ae115bc7Smrj if (bootpages != NULL) { 1841ae115bc7Smrj bootpages->p_prev = pp; 1842ae115bc7Smrj } 18437c478bd9Sstevel@tonic-gate bootpages = pp; 18447c478bd9Sstevel@tonic-gate ++boot_protect_cnt; 18457c478bd9Sstevel@tonic-gate } 18467c478bd9Sstevel@tonic-gate 18477c478bd9Sstevel@tonic-gate ++pfn; 18487c478bd9Sstevel@tonic-gate len -= MMU_PAGESIZE; 18497c478bd9Sstevel@tonic-gate va += MMU_PAGESIZE; 18507c478bd9Sstevel@tonic-gate } 18517c478bd9Sstevel@tonic-gate } 18527c478bd9Sstevel@tonic-gate PRM_DEBUG(boot_protect_cnt); 18537c478bd9Sstevel@tonic-gate } 18547c478bd9Sstevel@tonic-gate 1855ae115bc7Smrj /* 185635b1ab99Sjosephb * 1857ae115bc7Smrj */ 18587c478bd9Sstevel@tonic-gate static void 185935b1ab99Sjosephb layout_kernel_va(void) 18607c478bd9Sstevel@tonic-gate { 186135b1ab99Sjosephb PRM_POINT("layout_kernel_va() starting..."); 18627c478bd9Sstevel@tonic-gate /* 1863ae115bc7Smrj * Establish the final size of the kernel's heap, size of segmap, 1864ae115bc7Smrj * segkp, etc. 18657c478bd9Sstevel@tonic-gate */ 18667c478bd9Sstevel@tonic-gate 18677c478bd9Sstevel@tonic-gate #if defined(__amd64) 18687c478bd9Sstevel@tonic-gate 186935b1ab99Sjosephb kpm_vbase = (caddr_t)segkpm_base; 1870a3114836SGerry Liu if (physmax + 1 < plat_dr_physmax) { 1871a3114836SGerry Liu kpm_size = ROUND_UP_LPAGE(mmu_ptob(plat_dr_physmax)); 1872a3114836SGerry Liu } else { 1873a3114836SGerry Liu kpm_size = ROUND_UP_LPAGE(mmu_ptob(physmax + 1)); 1874a3114836SGerry Liu } 187535b1ab99Sjosephb if ((uintptr_t)kpm_vbase + kpm_size > (uintptr_t)valloc_base) 187635b1ab99Sjosephb panic("not enough room for kpm!"); 1877ae115bc7Smrj PRM_DEBUG(kpm_size); 1878ae115bc7Smrj PRM_DEBUG(kpm_vbase); 18797c478bd9Sstevel@tonic-gate 18807c478bd9Sstevel@tonic-gate /* 1881ae115bc7Smrj * By default we create a seg_kp in 64 bit kernels, it's a little 1882ae115bc7Smrj * faster to access than embedding it in the heap. 18837c478bd9Sstevel@tonic-gate */ 1884ae115bc7Smrj segkp_base = (caddr_t)valloc_base + valloc_sz; 18857c478bd9Sstevel@tonic-gate if (!segkp_fromheap) { 18867c478bd9Sstevel@tonic-gate size_t sz = mmu_ptob(segkpsize); 18877c478bd9Sstevel@tonic-gate 18887c478bd9Sstevel@tonic-gate /* 1889ae115bc7Smrj * determine size of segkp 18907c478bd9Sstevel@tonic-gate */ 18917c478bd9Sstevel@tonic-gate if (sz < SEGKPMINSIZE || sz > SEGKPMAXSIZE) { 18927c478bd9Sstevel@tonic-gate sz = SEGKPDEFSIZE; 18937c478bd9Sstevel@tonic-gate cmn_err(CE_WARN, "!Illegal value for segkpsize. " 18947c478bd9Sstevel@tonic-gate "segkpsize has been reset to %ld pages", 18957c478bd9Sstevel@tonic-gate mmu_btop(sz)); 18967c478bd9Sstevel@tonic-gate } 18977c478bd9Sstevel@tonic-gate sz = MIN(sz, MAX(SEGKPMINSIZE, mmu_ptob(physmem))); 18987c478bd9Sstevel@tonic-gate 18997c478bd9Sstevel@tonic-gate segkpsize = mmu_btop(ROUND_UP_LPAGE(sz)); 19007c478bd9Sstevel@tonic-gate } 1901ae115bc7Smrj PRM_DEBUG(segkp_base); 1902ae115bc7Smrj PRM_DEBUG(segkpsize); 19037c478bd9Sstevel@tonic-gate 190435b1ab99Sjosephb /* 190535b1ab99Sjosephb * segzio is used for ZFS cached data. It uses a distinct VA 190635b1ab99Sjosephb * segment (from kernel heap) so that we can easily tell not to 190735b1ab99Sjosephb * include it in kernel crash dumps on 64 bit kernels. The trick is 190835b1ab99Sjosephb * to give it lots of VA, but not constrain the kernel heap. 190935b1ab99Sjosephb * We scale the size of segzio linearly with physmem up to 191035b1ab99Sjosephb * SEGZIOMAXSIZE. Above that amount it scales at 50% of physmem. 191135b1ab99Sjosephb */ 1912ae115bc7Smrj segzio_base = segkp_base + mmu_ptob(segkpsize); 1913ae115bc7Smrj if (segzio_fromheap) { 1914ae115bc7Smrj segziosize = 0; 1915ae115bc7Smrj } else { 191635b1ab99Sjosephb size_t physmem_size = mmu_ptob(physmem); 191735b1ab99Sjosephb size_t size = (segziosize == 0) ? 191835b1ab99Sjosephb physmem_size : mmu_ptob(segziosize); 1919ad23a2dbSjohansen 192035b1ab99Sjosephb if (size < SEGZIOMINSIZE) 1921ad23a2dbSjohansen size = SEGZIOMINSIZE; 192235b1ab99Sjosephb if (size > SEGZIOMAXSIZE) { 1923a778305aSjohansen size = SEGZIOMAXSIZE; 192435b1ab99Sjosephb if (physmem_size > size) 192535b1ab99Sjosephb size += (physmem_size - size) / 2; 1926ad23a2dbSjohansen } 1927ad23a2dbSjohansen segziosize = mmu_btop(ROUND_UP_LPAGE(size)); 1928ad23a2dbSjohansen } 1929ae115bc7Smrj PRM_DEBUG(segziosize); 1930ae115bc7Smrj PRM_DEBUG(segzio_base); 1931ad23a2dbSjohansen 19327c478bd9Sstevel@tonic-gate /* 1933ae115bc7Smrj * Put the range of VA for device mappings next, kmdb knows to not 1934ae115bc7Smrj * grep in this range of addresses. 19357c478bd9Sstevel@tonic-gate */ 1936ae115bc7Smrj toxic_addr = 1937ae115bc7Smrj ROUND_UP_LPAGE((uintptr_t)segzio_base + mmu_ptob(segziosize)); 19387c478bd9Sstevel@tonic-gate PRM_DEBUG(toxic_addr); 1939ae115bc7Smrj segmap_start = ROUND_UP_LPAGE(toxic_addr + toxic_size); 1940ae115bc7Smrj #else /* __i386 */ 1941ae115bc7Smrj segmap_start = ROUND_UP_LPAGE(kernelbase); 1942ae115bc7Smrj #endif /* __i386 */ 1943ae115bc7Smrj PRM_DEBUG(segmap_start); 19447c478bd9Sstevel@tonic-gate 19457c478bd9Sstevel@tonic-gate /* 194652eef812SAmrita Sadhukhan * Users can change segmapsize through eeprom. If the variable 194752eef812SAmrita Sadhukhan * is tuned through eeprom, there is no upper bound on the 1948ed5289f9SKen Erickson * size of segmap. 19497c478bd9Sstevel@tonic-gate */ 19507c478bd9Sstevel@tonic-gate segmapsize = MAX(ROUND_UP_LPAGE(segmapsize), SEGMAPDEFAULT); 19517c478bd9Sstevel@tonic-gate 19527c478bd9Sstevel@tonic-gate #if defined(__i386) 19537c478bd9Sstevel@tonic-gate /* 19547c478bd9Sstevel@tonic-gate * 32-bit systems don't have segkpm or segkp, so segmap appears at 19557c478bd9Sstevel@tonic-gate * the bottom of the kernel's address range. Set aside space for a 1956ae115bc7Smrj * small red zone just below the start of segmap. 19577c478bd9Sstevel@tonic-gate */ 1958ae115bc7Smrj segmap_start += KERNEL_REDZONE_SIZE; 19597c478bd9Sstevel@tonic-gate segmapsize -= KERNEL_REDZONE_SIZE; 19607c478bd9Sstevel@tonic-gate #endif 19617c478bd9Sstevel@tonic-gate 1962ae115bc7Smrj PRM_DEBUG(segmap_start); 19637c478bd9Sstevel@tonic-gate PRM_DEBUG(segmapsize); 196435b1ab99Sjosephb kernelheap = (caddr_t)ROUND_UP_LPAGE(segmap_start + segmapsize); 196535b1ab99Sjosephb PRM_DEBUG(kernelheap); 196635b1ab99Sjosephb PRM_POINT("layout_kernel_va() done..."); 196735b1ab99Sjosephb } 196835b1ab99Sjosephb 196935b1ab99Sjosephb /* 197035b1ab99Sjosephb * Finish initializing the VM system, now that we are no longer 197135b1ab99Sjosephb * relying on the boot time memory allocators. 197235b1ab99Sjosephb */ 197335b1ab99Sjosephb static void 197435b1ab99Sjosephb startup_vm(void) 197535b1ab99Sjosephb { 197635b1ab99Sjosephb struct segmap_crargs a; 197735b1ab99Sjosephb 197835b1ab99Sjosephb extern int use_brk_lpg, use_stk_lpg; 197935b1ab99Sjosephb 198035b1ab99Sjosephb PRM_POINT("startup_vm() starting..."); 198135b1ab99Sjosephb 198235b1ab99Sjosephb /* 198335b1ab99Sjosephb * Initialize the hat layer. 198435b1ab99Sjosephb */ 198535b1ab99Sjosephb hat_init(); 19867c478bd9Sstevel@tonic-gate 1987ae115bc7Smrj /* 1988ae115bc7Smrj * Do final allocations of HAT data structures that need to 1989ae115bc7Smrj * be allocated before quiescing the boot loader. 1990ae115bc7Smrj */ 1991ae115bc7Smrj PRM_POINT("Calling hat_kern_alloc()..."); 1992ae115bc7Smrj hat_kern_alloc((caddr_t)segmap_start, segmapsize, ekernelheap); 1993ae115bc7Smrj PRM_POINT("hat_kern_alloc() done"); 1994ae115bc7Smrj 1995843e1988Sjohnlev #ifndef __xpv 1996ae115bc7Smrj /* 19971d03c31eSjohnlev * Setup Page Attribute Table 1998ae115bc7Smrj */ 19991d03c31eSjohnlev pat_sync(); 2000843e1988Sjohnlev #endif 2001ae115bc7Smrj 2002ae115bc7Smrj /* 2003ae115bc7Smrj * The next two loops are done in distinct steps in order 2004ae115bc7Smrj * to be sure that any page that is doubly mapped (both above 2005ae115bc7Smrj * KERNEL_TEXT and below kernelbase) is dealt with correctly. 2006ae115bc7Smrj * Note this may never happen, but it might someday. 2007ae115bc7Smrj */ 2008ae115bc7Smrj bootpages = NULL; 2009ae115bc7Smrj PRM_POINT("Protecting boot pages"); 2010ae115bc7Smrj 2011ae115bc7Smrj /* 2012ae115bc7Smrj * Protect any pages mapped above KERNEL_TEXT that somehow have 2013ae115bc7Smrj * page_t's. This can only happen if something weird allocated 2014ae115bc7Smrj * in this range (like kadb/kmdb). 2015ae115bc7Smrj */ 2016ae115bc7Smrj protect_boot_range(KERNEL_TEXT, (uintptr_t)-1, 0); 2017ae115bc7Smrj 2018ae115bc7Smrj /* 2019ae115bc7Smrj * Before we can take over memory allocation/mapping from the boot 2020ae115bc7Smrj * loader we must remove from our free page lists any boot allocated 2021ae115bc7Smrj * pages that stay mapped until release_bootstrap(). 2022ae115bc7Smrj */ 2023ae115bc7Smrj protect_boot_range(0, kernelbase, 1); 2024ae115bc7Smrj 2025843e1988Sjohnlev 2026ae115bc7Smrj /* 2027ae115bc7Smrj * Switch to running on regular HAT (not boot_mmu) 2028ae115bc7Smrj */ 2029ae115bc7Smrj PRM_POINT("Calling hat_kern_setup()..."); 2030ae115bc7Smrj hat_kern_setup(); 2031ae115bc7Smrj 2032ae115bc7Smrj /* 2033ae115bc7Smrj * It is no longer safe to call BOP_ALLOC(), so make sure we don't. 2034ae115bc7Smrj */ 2035ae115bc7Smrj bop_no_more_mem(); 2036ae115bc7Smrj 2037ae115bc7Smrj PRM_POINT("hat_kern_setup() done"); 2038ae115bc7Smrj 2039ae115bc7Smrj hat_cpu_online(CPU); 2040ae115bc7Smrj 20417c478bd9Sstevel@tonic-gate /* 20427c478bd9Sstevel@tonic-gate * Initialize VM system 20437c478bd9Sstevel@tonic-gate */ 20447c478bd9Sstevel@tonic-gate PRM_POINT("Calling kvm_init()..."); 20457c478bd9Sstevel@tonic-gate kvm_init(); 20467c478bd9Sstevel@tonic-gate PRM_POINT("kvm_init() done"); 20477c478bd9Sstevel@tonic-gate 20487c478bd9Sstevel@tonic-gate /* 20497c478bd9Sstevel@tonic-gate * Tell kmdb that the VM system is now working 20507c478bd9Sstevel@tonic-gate */ 20517c478bd9Sstevel@tonic-gate if (boothowto & RB_DEBUG) 20527c478bd9Sstevel@tonic-gate kdi_dvec_vmready(); 20537c478bd9Sstevel@tonic-gate 2054843e1988Sjohnlev #if defined(__xpv) 2055843e1988Sjohnlev /* 2056843e1988Sjohnlev * Populate the I/O pool on domain 0 2057843e1988Sjohnlev */ 2058843e1988Sjohnlev if (DOMAIN_IS_INITDOMAIN(xen_info)) { 2059843e1988Sjohnlev extern long populate_io_pool(void); 2060843e1988Sjohnlev long init_io_pool_cnt; 2061843e1988Sjohnlev 2062843e1988Sjohnlev PRM_POINT("Populating reserve I/O page pool"); 2063843e1988Sjohnlev init_io_pool_cnt = populate_io_pool(); 2064843e1988Sjohnlev PRM_DEBUG(init_io_pool_cnt); 2065843e1988Sjohnlev } 2066843e1988Sjohnlev #endif 20677c478bd9Sstevel@tonic-gate /* 20687c478bd9Sstevel@tonic-gate * Mangle the brand string etc. 20697c478bd9Sstevel@tonic-gate */ 20707c478bd9Sstevel@tonic-gate cpuid_pass3(CPU); 20717c478bd9Sstevel@tonic-gate 20727c478bd9Sstevel@tonic-gate #if defined(__amd64) 20737c478bd9Sstevel@tonic-gate 20747c478bd9Sstevel@tonic-gate /* 20757c478bd9Sstevel@tonic-gate * Create the device arena for toxic (to dtrace/kmdb) mappings. 20767c478bd9Sstevel@tonic-gate */ 20777c478bd9Sstevel@tonic-gate device_arena = vmem_create("device", (void *)toxic_addr, 20787c478bd9Sstevel@tonic-gate toxic_size, MMU_PAGESIZE, NULL, NULL, NULL, 0, VM_SLEEP); 20797c478bd9Sstevel@tonic-gate 20807c478bd9Sstevel@tonic-gate #else /* __i386 */ 20817c478bd9Sstevel@tonic-gate 20827c478bd9Sstevel@tonic-gate /* 20837c478bd9Sstevel@tonic-gate * allocate the bit map that tracks toxic pages 20847c478bd9Sstevel@tonic-gate */ 2085ae115bc7Smrj toxic_bit_map_len = btop((ulong_t)(valloc_base - kernelbase)); 20867c478bd9Sstevel@tonic-gate PRM_DEBUG(toxic_bit_map_len); 20877c478bd9Sstevel@tonic-gate toxic_bit_map = 20887c478bd9Sstevel@tonic-gate kmem_zalloc(BT_SIZEOFMAP(toxic_bit_map_len), KM_NOSLEEP); 20897c478bd9Sstevel@tonic-gate ASSERT(toxic_bit_map != NULL); 20907c478bd9Sstevel@tonic-gate PRM_DEBUG(toxic_bit_map); 20917c478bd9Sstevel@tonic-gate 20927c478bd9Sstevel@tonic-gate #endif /* __i386 */ 20937c478bd9Sstevel@tonic-gate 20947c478bd9Sstevel@tonic-gate 20957c478bd9Sstevel@tonic-gate /* 20967c478bd9Sstevel@tonic-gate * Now that we've got more VA, as well as the ability to allocate from 20977c478bd9Sstevel@tonic-gate * it, tell the debugger. 20987c478bd9Sstevel@tonic-gate */ 20997c478bd9Sstevel@tonic-gate if (boothowto & RB_DEBUG) 21007c478bd9Sstevel@tonic-gate kdi_dvec_memavail(); 21017c478bd9Sstevel@tonic-gate 21027c478bd9Sstevel@tonic-gate /* 21037c478bd9Sstevel@tonic-gate * The following code installs a special page fault handler (#pf) 21047c478bd9Sstevel@tonic-gate * to work around a pentium bug. 21057c478bd9Sstevel@tonic-gate */ 2106843e1988Sjohnlev #if !defined(__amd64) && !defined(__xpv) 21077c478bd9Sstevel@tonic-gate if (x86_type == X86_TYPE_P5) { 2108ae115bc7Smrj desctbr_t idtr; 21097c478bd9Sstevel@tonic-gate gate_desc_t *newidt; 21107c478bd9Sstevel@tonic-gate 21117c478bd9Sstevel@tonic-gate if ((newidt = kmem_zalloc(MMU_PAGESIZE, KM_NOSLEEP)) == NULL) 21127c478bd9Sstevel@tonic-gate panic("failed to install pentium_pftrap"); 21137c478bd9Sstevel@tonic-gate 21140cfdb603Sjosephb bcopy(idt0, newidt, NIDT * sizeof (*idt0)); 21157c478bd9Sstevel@tonic-gate set_gatesegd(&newidt[T_PGFLT], &pentium_pftrap, 21169844da31SSeth Goldberg KCS_SEL, SDT_SYSIGT, TRP_KPL, 0); 21177c478bd9Sstevel@tonic-gate 21187c478bd9Sstevel@tonic-gate (void) as_setprot(&kas, (caddr_t)newidt, MMU_PAGESIZE, 21190cfdb603Sjosephb PROT_READ | PROT_EXEC); 21207c478bd9Sstevel@tonic-gate 21210cfdb603Sjosephb CPU->cpu_idt = newidt; 21220cfdb603Sjosephb idtr.dtr_base = (uintptr_t)CPU->cpu_idt; 21230cfdb603Sjosephb idtr.dtr_limit = (NIDT * sizeof (*idt0)) - 1; 2124ae115bc7Smrj wr_idtr(&idtr); 21257c478bd9Sstevel@tonic-gate } 21267c478bd9Sstevel@tonic-gate #endif /* !__amd64 */ 21277c478bd9Sstevel@tonic-gate 2128843e1988Sjohnlev #if !defined(__xpv) 21297c478bd9Sstevel@tonic-gate /* 21307c478bd9Sstevel@tonic-gate * Map page pfn=0 for drivers, such as kd, that need to pick up 21317c478bd9Sstevel@tonic-gate * parameters left there by controllers/BIOS. 21327c478bd9Sstevel@tonic-gate */ 21337c478bd9Sstevel@tonic-gate PRM_POINT("setup up p0_va"); 21347c478bd9Sstevel@tonic-gate p0_va = i86devmap(0, 1, PROT_READ); 21357c478bd9Sstevel@tonic-gate PRM_DEBUG(p0_va); 2136843e1988Sjohnlev #endif 21377c478bd9Sstevel@tonic-gate 21387c478bd9Sstevel@tonic-gate cmn_err(CE_CONT, "?mem = %luK (0x%lx)\n", 21397c478bd9Sstevel@tonic-gate physinstalled << (MMU_PAGESHIFT - 10), ptob(physinstalled)); 21407c478bd9Sstevel@tonic-gate 214183f9b804Skchow /* 214283f9b804Skchow * disable automatic large pages for small memory systems or 214383f9b804Skchow * when the disable flag is set. 214402bc52beSkchow * 214502bc52beSkchow * Do not yet consider page sizes larger than 2m/4m. 214683f9b804Skchow */ 2147ec25b48fSsusans if (!auto_lpg_disable && mmu.max_page_level > 0) { 2148ec25b48fSsusans max_uheap_lpsize = LEVEL_SIZE(1); 2149ec25b48fSsusans max_ustack_lpsize = LEVEL_SIZE(1); 2150ec25b48fSsusans max_privmap_lpsize = LEVEL_SIZE(1); 2151ec25b48fSsusans max_uidata_lpsize = LEVEL_SIZE(1); 2152ec25b48fSsusans max_utext_lpsize = LEVEL_SIZE(1); 2153ec25b48fSsusans max_shm_lpsize = LEVEL_SIZE(1); 2154ec25b48fSsusans } 2155ec25b48fSsusans if (physmem < privm_lpg_min_physmem || mmu.max_page_level == 0 || 2156ec25b48fSsusans auto_lpg_disable) { 2157beb1bda0Sdavemq use_brk_lpg = 0; 2158beb1bda0Sdavemq use_stk_lpg = 0; 2159ec25b48fSsusans } 216002bc52beSkchow mcntl0_lpsize = LEVEL_SIZE(mmu.umax_page_level); 2161beb1bda0Sdavemq 21627c478bd9Sstevel@tonic-gate PRM_POINT("Calling hat_init_finish()..."); 21637c478bd9Sstevel@tonic-gate hat_init_finish(); 21647c478bd9Sstevel@tonic-gate PRM_POINT("hat_init_finish() done"); 21657c478bd9Sstevel@tonic-gate 21667c478bd9Sstevel@tonic-gate /* 21677c478bd9Sstevel@tonic-gate * Initialize the segkp segment type. 21687c478bd9Sstevel@tonic-gate */ 21697c478bd9Sstevel@tonic-gate rw_enter(&kas.a_lock, RW_WRITER); 2170ae115bc7Smrj PRM_POINT("Attaching segkp"); 2171ae115bc7Smrj if (segkp_fromheap) { 21727c478bd9Sstevel@tonic-gate segkp->s_as = &kas; 2173ae115bc7Smrj } else if (seg_attach(&kas, (caddr_t)segkp_base, mmu_ptob(segkpsize), 2174ae115bc7Smrj segkp) < 0) { 2175ae115bc7Smrj panic("startup: cannot attach segkp"); 2176ae115bc7Smrj /*NOTREACHED*/ 21777c478bd9Sstevel@tonic-gate } 2178ae115bc7Smrj PRM_POINT("Doing segkp_create()"); 21797c478bd9Sstevel@tonic-gate if (segkp_create(segkp) != 0) { 21807c478bd9Sstevel@tonic-gate panic("startup: segkp_create failed"); 21817c478bd9Sstevel@tonic-gate /*NOTREACHED*/ 21827c478bd9Sstevel@tonic-gate } 21837c478bd9Sstevel@tonic-gate PRM_DEBUG(segkp); 21847c478bd9Sstevel@tonic-gate rw_exit(&kas.a_lock); 21857c478bd9Sstevel@tonic-gate 21867c478bd9Sstevel@tonic-gate /* 21877c478bd9Sstevel@tonic-gate * kpm segment 21887c478bd9Sstevel@tonic-gate */ 21897c478bd9Sstevel@tonic-gate segmap_kpm = 0; 21907c478bd9Sstevel@tonic-gate if (kpm_desired) { 21917c478bd9Sstevel@tonic-gate kpm_init(); 21927c478bd9Sstevel@tonic-gate kpm_enable = 1; 21937c478bd9Sstevel@tonic-gate } 21947c478bd9Sstevel@tonic-gate 21957c478bd9Sstevel@tonic-gate /* 21967c478bd9Sstevel@tonic-gate * Now create segmap segment. 21977c478bd9Sstevel@tonic-gate */ 21987c478bd9Sstevel@tonic-gate rw_enter(&kas.a_lock, RW_WRITER); 2199ae115bc7Smrj if (seg_attach(&kas, (caddr_t)segmap_start, segmapsize, segmap) < 0) { 2200ae115bc7Smrj panic("cannot attach segmap"); 22017c478bd9Sstevel@tonic-gate /*NOTREACHED*/ 22027c478bd9Sstevel@tonic-gate } 2203ae115bc7Smrj PRM_DEBUG(segmap); 22047c478bd9Sstevel@tonic-gate 22057c478bd9Sstevel@tonic-gate a.prot = PROT_READ | PROT_WRITE; 22067c478bd9Sstevel@tonic-gate a.shmsize = 0; 22077c478bd9Sstevel@tonic-gate a.nfreelist = segmapfreelists; 22087c478bd9Sstevel@tonic-gate 2209ae115bc7Smrj if (segmap_create(segmap, (caddr_t)&a) != 0) 2210ae115bc7Smrj panic("segmap_create segmap"); 22117c478bd9Sstevel@tonic-gate rw_exit(&kas.a_lock); 22127c478bd9Sstevel@tonic-gate 22137c478bd9Sstevel@tonic-gate setup_vaddr_for_ppcopy(CPU); 22147c478bd9Sstevel@tonic-gate 22157c478bd9Sstevel@tonic-gate segdev_init(); 2216843e1988Sjohnlev #if defined(__xpv) 2217843e1988Sjohnlev if (DOMAIN_IS_INITDOMAIN(xen_info)) 2218843e1988Sjohnlev #endif 2219843e1988Sjohnlev pmem_init(); 2220ae115bc7Smrj 22217c478bd9Sstevel@tonic-gate PRM_POINT("startup_vm() done"); 22227c478bd9Sstevel@tonic-gate } 22237c478bd9Sstevel@tonic-gate 2224ae115bc7Smrj /* 2225ae115bc7Smrj * Load a tod module for the non-standard tod part found on this system. 2226ae115bc7Smrj */ 2227ae115bc7Smrj static void 2228ae115bc7Smrj load_tod_module(char *todmod) 2229ae115bc7Smrj { 2230ae115bc7Smrj if (modload("tod", todmod) == -1) 2231ae115bc7Smrj halt("Can't load TOD module"); 2232ae115bc7Smrj } 2233ae115bc7Smrj 22347c478bd9Sstevel@tonic-gate static void 22357c478bd9Sstevel@tonic-gate startup_end(void) 22367c478bd9Sstevel@tonic-gate { 2237dd4eeefdSeota int i; 22387c478bd9Sstevel@tonic-gate extern void setx86isalist(void); 2239fb2caebeSRandy Fishel extern void cpu_event_init(void); 22407c478bd9Sstevel@tonic-gate 22417c478bd9Sstevel@tonic-gate PRM_POINT("startup_end() starting..."); 22427c478bd9Sstevel@tonic-gate 22437c478bd9Sstevel@tonic-gate /* 22447c478bd9Sstevel@tonic-gate * Perform tasks that get done after most of the VM 22457c478bd9Sstevel@tonic-gate * initialization has been done but before the clock 22467c478bd9Sstevel@tonic-gate * and other devices get started. 22477c478bd9Sstevel@tonic-gate */ 22487c478bd9Sstevel@tonic-gate kern_setup1(); 22497c478bd9Sstevel@tonic-gate 22507c478bd9Sstevel@tonic-gate /* 22517c478bd9Sstevel@tonic-gate * Perform CPC initialization for this CPU. 22527c478bd9Sstevel@tonic-gate */ 22537c478bd9Sstevel@tonic-gate kcpc_hw_init(CPU); 22547c478bd9Sstevel@tonic-gate 2255fb2caebeSRandy Fishel /* 2256fb2caebeSRandy Fishel * Initialize cpu event framework. 2257fb2caebeSRandy Fishel */ 2258fb2caebeSRandy Fishel cpu_event_init(); 2259fb2caebeSRandy Fishel 2260ee88d2b9Skchow #if defined(OPTERON_WORKAROUND_6323525) 2261ee88d2b9Skchow if (opteron_workaround_6323525) 2262ee88d2b9Skchow patch_workaround_6323525(); 2263ee88d2b9Skchow #endif 2264ae115bc7Smrj /* 2265ae115bc7Smrj * If needed, load TOD module now so that ddi_get_time(9F) etc. work 2266ae115bc7Smrj * (For now, "needed" is defined as set tod_module_name in /etc/system) 2267ae115bc7Smrj */ 2268ae115bc7Smrj if (tod_module_name != NULL) { 2269ae115bc7Smrj PRM_POINT("load_tod_module()"); 2270ae115bc7Smrj load_tod_module(tod_module_name); 2271ae115bc7Smrj } 2272ae115bc7Smrj 2273843e1988Sjohnlev #if defined(__xpv) 2274843e1988Sjohnlev /* 2275843e1988Sjohnlev * Forceload interposing TOD module for the hypervisor. 2276843e1988Sjohnlev */ 2277843e1988Sjohnlev PRM_POINT("load_tod_module()"); 2278843e1988Sjohnlev load_tod_module("xpvtod"); 2279843e1988Sjohnlev #endif 2280843e1988Sjohnlev 22817c478bd9Sstevel@tonic-gate /* 22827c478bd9Sstevel@tonic-gate * Configure the system. 22837c478bd9Sstevel@tonic-gate */ 22847c478bd9Sstevel@tonic-gate PRM_POINT("Calling configure()..."); 22857c478bd9Sstevel@tonic-gate configure(); /* set up devices */ 22867c478bd9Sstevel@tonic-gate PRM_POINT("configure() done"); 22877c478bd9Sstevel@tonic-gate 22887af88ac7SKuriakose Kuruvilla /* 22897af88ac7SKuriakose Kuruvilla * We can now setup for XSAVE because fpu_probe is done in configure(). 22907af88ac7SKuriakose Kuruvilla */ 22917af88ac7SKuriakose Kuruvilla if (fp_save_mech == FP_XSAVE) { 22927af88ac7SKuriakose Kuruvilla xsave_setup_msr(CPU); 22937af88ac7SKuriakose Kuruvilla } 22947af88ac7SKuriakose Kuruvilla 22957c478bd9Sstevel@tonic-gate /* 22967c478bd9Sstevel@tonic-gate * Set the isa_list string to the defined instruction sets we 22977c478bd9Sstevel@tonic-gate * support. 22987c478bd9Sstevel@tonic-gate */ 22997c478bd9Sstevel@tonic-gate setx86isalist(); 2300100b72f4Sandrei cpu_intr_alloc(CPU, NINTR_THREADS); 23017c478bd9Sstevel@tonic-gate psm_install(); 23027c478bd9Sstevel@tonic-gate 23037c478bd9Sstevel@tonic-gate /* 23047c478bd9Sstevel@tonic-gate * We're done with bootops. We don't unmap the bootstrap yet because 23057c478bd9Sstevel@tonic-gate * we're still using bootsvcs. 23067c478bd9Sstevel@tonic-gate */ 2307ae115bc7Smrj PRM_POINT("NULLing out bootops"); 2308ae115bc7Smrj *bootopsp = (struct bootops *)NULL; 23097c478bd9Sstevel@tonic-gate bootops = (struct bootops *)NULL; 23107c478bd9Sstevel@tonic-gate 2311843e1988Sjohnlev #if defined(__xpv) 2312843e1988Sjohnlev ec_init_debug_irq(); 2313843e1988Sjohnlev xs_domu_init(); 2314843e1988Sjohnlev #endif 23151de082f7SVikram Hegde 23163a634bfcSVikram Hegde #if defined(__amd64) && !defined(__xpv) 23173a634bfcSVikram Hegde /* 23183a634bfcSVikram Hegde * Intel IOMMU has been setup/initialized in ddi_impl.c 23193a634bfcSVikram Hegde * Start it up now. 23203a634bfcSVikram Hegde */ 23213a634bfcSVikram Hegde immu_startup(); 23221de082f7SVikram Hegde #endif 23233a634bfcSVikram Hegde 23247c478bd9Sstevel@tonic-gate PRM_POINT("Enabling interrupts"); 23257c478bd9Sstevel@tonic-gate (*picinitf)(); 23267c478bd9Sstevel@tonic-gate sti(); 2327843e1988Sjohnlev #if defined(__xpv) 2328843e1988Sjohnlev ASSERT(CPU->cpu_m.mcpu_vcpu_info->evtchn_upcall_mask == 0); 2329843e1988Sjohnlev xen_late_startup(); 2330843e1988Sjohnlev #endif 23317c478bd9Sstevel@tonic-gate 23327c478bd9Sstevel@tonic-gate (void) add_avsoftintr((void *)&softlevel1_hdl, 1, softlevel1, 233335b1ab99Sjosephb "softlevel1", NULL, NULL); /* XXX to be moved later */ 23347c478bd9Sstevel@tonic-gate 2335dd4eeefdSeota /* 2336a288e5a9SJoshua M. Clulow * Register software interrupt handlers for ddi_periodic_add(9F). 2337dd4eeefdSeota * Software interrupts up to the level 10 are supported. 2338dd4eeefdSeota */ 2339dd4eeefdSeota for (i = DDI_IPL_1; i <= DDI_IPL_10; i++) { 2340dd4eeefdSeota (void) add_avsoftintr((void *)&softlevel_hdl[i-1], i, 2341a288e5a9SJoshua M. Clulow (avfunc)ddi_periodic_softintr, "ddi_periodic", 2342a288e5a9SJoshua M. Clulow (caddr_t)(uintptr_t)i, NULL); 2343dd4eeefdSeota } 2344dd4eeefdSeota 23457125fcbdSVikram Hegde #if !defined(__xpv) 23467125fcbdSVikram Hegde if (modload("drv", "amd_iommu") < 0) { 23477125fcbdSVikram Hegde PRM_POINT("No AMD IOMMU present\n"); 234894f1124eSVikram Hegde } else if (ddi_hold_installed_driver(ddi_name_to_major( 234994f1124eSVikram Hegde "amd_iommu")) == NULL) { 23507125fcbdSVikram Hegde prom_printf("ERROR: failed to attach AMD IOMMU\n"); 23517125fcbdSVikram Hegde } 23527125fcbdSVikram Hegde #endif 2353e774b42bSBill Holler post_startup_cpu_fixups(); 23547125fcbdSVikram Hegde 23557c478bd9Sstevel@tonic-gate PRM_POINT("startup_end() done"); 23567c478bd9Sstevel@tonic-gate } 23577c478bd9Sstevel@tonic-gate 235896d008c2SKen Erickson /* 235996d008c2SKen Erickson * Don't remove the following 2 variables. They are necessary 236096d008c2SKen Erickson * for reading the hostid from the legacy file (/kernel/misc/sysinit). 236196d008c2SKen Erickson */ 23627c478bd9Sstevel@tonic-gate char *_hs1107 = hw_serial; 23637c478bd9Sstevel@tonic-gate ulong_t _bdhs34; 23647c478bd9Sstevel@tonic-gate 23657c478bd9Sstevel@tonic-gate void 23667c478bd9Sstevel@tonic-gate post_startup(void) 23677c478bd9Sstevel@tonic-gate { 23680e751525SEric Saxe extern void cpupm_init(cpu_t *); 2369fb2caebeSRandy Fishel extern void cpu_event_init_cpu(cpu_t *); 23700e751525SEric Saxe 23717c478bd9Sstevel@tonic-gate /* 23727c478bd9Sstevel@tonic-gate * Set the system wide, processor-specific flags to be passed 23737c478bd9Sstevel@tonic-gate * to userland via the aux vector for performance hints and 23747c478bd9Sstevel@tonic-gate * instruction set extensions. 23757c478bd9Sstevel@tonic-gate */ 23767c478bd9Sstevel@tonic-gate bind_hwcap(); 23777c478bd9Sstevel@tonic-gate 2378843e1988Sjohnlev #ifdef __xpv 2379843e1988Sjohnlev if (DOMAIN_IS_INITDOMAIN(xen_info)) 2380843e1988Sjohnlev #endif 2381843e1988Sjohnlev { 2382843e1988Sjohnlev #if defined(__xpv) 2383843e1988Sjohnlev xpv_panic_init(); 2384843e1988Sjohnlev #else 2385843e1988Sjohnlev /* 2386843e1988Sjohnlev * Startup the memory scrubber. 2387843e1988Sjohnlev * XXPV This should be running somewhere .. 2388843e1988Sjohnlev */ 238979ec9da8SYuri Pankov if ((get_hwenv() & HW_VIRTUAL) == 0) 2390e4b86885SCheng Sean Ye memscrub_init(); 2391843e1988Sjohnlev #endif 2392843e1988Sjohnlev } 23937c478bd9Sstevel@tonic-gate 23947aec1d6eScindi /* 23957aec1d6eScindi * Complete CPU module initialization 23967aec1d6eScindi */ 239720c794b3Sgavinm cmi_post_startup(); 23987aec1d6eScindi 23997c478bd9Sstevel@tonic-gate /* 240084ab085aSmws * Perform forceloading tasks for /etc/system. 24017c478bd9Sstevel@tonic-gate */ 240284ab085aSmws (void) mod_sysctl(SYS_FORCELOAD, NULL); 24037c478bd9Sstevel@tonic-gate 24047c478bd9Sstevel@tonic-gate /* 24057c478bd9Sstevel@tonic-gate * ON4.0: Force /proc module in until clock interrupt handle fixed 24067c478bd9Sstevel@tonic-gate * ON4.0: This must be fixed or restated in /etc/systems. 24077c478bd9Sstevel@tonic-gate */ 24087c478bd9Sstevel@tonic-gate (void) modload("fs", "procfs"); 24097c478bd9Sstevel@tonic-gate 2410c35aa225Smarx (void) i_ddi_attach_hw_nodes("pit_beep"); 2411c35aa225Smarx 24127c478bd9Sstevel@tonic-gate #if defined(__i386) 24137c478bd9Sstevel@tonic-gate /* 24147c478bd9Sstevel@tonic-gate * Check for required functional Floating Point hardware, 24157c478bd9Sstevel@tonic-gate * unless FP hardware explicitly disabled. 24167c478bd9Sstevel@tonic-gate */ 24177c478bd9Sstevel@tonic-gate if (fpu_exists && (fpu_pentium_fdivbug || fp_kind == FP_NO)) 24187c478bd9Sstevel@tonic-gate halt("No working FP hardware found"); 24197c478bd9Sstevel@tonic-gate #endif 24207c478bd9Sstevel@tonic-gate 24217c478bd9Sstevel@tonic-gate maxmem = freemem; 24227c478bd9Sstevel@tonic-gate 2423fb2caebeSRandy Fishel cpu_event_init_cpu(CPU); 24240e751525SEric Saxe cpupm_init(CPU); 2425fa96bd91SMichael Corcoran (void) mach_cpu_create_device_node(CPU, NULL); 24260e751525SEric Saxe 24270e751525SEric Saxe pg_init(); 24287c478bd9Sstevel@tonic-gate } 24297c478bd9Sstevel@tonic-gate 24307c478bd9Sstevel@tonic-gate static int 243119397407SSherry Moore pp_in_range(page_t *pp, uint64_t low_addr, uint64_t high_addr) 24327c478bd9Sstevel@tonic-gate { 243319397407SSherry Moore return ((pp->p_pagenum >= btop(low_addr)) && 243419397407SSherry Moore (pp->p_pagenum < btopr(high_addr))); 24357c478bd9Sstevel@tonic-gate } 24367c478bd9Sstevel@tonic-gate 24370181461bSKeith M Wesolowski static int 24380181461bSKeith M Wesolowski pp_in_module(page_t *pp, const rd_existing_t *modranges) 24390181461bSKeith M Wesolowski { 24400181461bSKeith M Wesolowski uint_t i; 24410181461bSKeith M Wesolowski 24420181461bSKeith M Wesolowski for (i = 0; modranges[i].phys != 0; i++) { 24430181461bSKeith M Wesolowski if (pp_in_range(pp, modranges[i].phys, 24440181461bSKeith M Wesolowski modranges[i].phys + modranges[i].size)) 24450181461bSKeith M Wesolowski return (1); 24460181461bSKeith M Wesolowski } 24470181461bSKeith M Wesolowski 24480181461bSKeith M Wesolowski return (0); 24490181461bSKeith M Wesolowski } 24500181461bSKeith M Wesolowski 24517c478bd9Sstevel@tonic-gate void 24527c478bd9Sstevel@tonic-gate release_bootstrap(void) 24537c478bd9Sstevel@tonic-gate { 24547c478bd9Sstevel@tonic-gate int root_is_ramdisk; 24557c478bd9Sstevel@tonic-gate page_t *pp; 24567c478bd9Sstevel@tonic-gate extern void kobj_boot_unmountroot(void); 24577c478bd9Sstevel@tonic-gate extern dev_t rootdev; 24580181461bSKeith M Wesolowski uint_t i; 24590181461bSKeith M Wesolowski char propname[32]; 24600181461bSKeith M Wesolowski rd_existing_t *modranges; 246155d507a9SSeth Goldberg #if !defined(__xpv) 246255d507a9SSeth Goldberg pfn_t pfn; 246355d507a9SSeth Goldberg #endif 24647c478bd9Sstevel@tonic-gate 24650181461bSKeith M Wesolowski /* 24660181461bSKeith M Wesolowski * Save the bootfs module ranges so that we can reserve them below 24670181461bSKeith M Wesolowski * for the real bootfs. 24680181461bSKeith M Wesolowski */ 24690181461bSKeith M Wesolowski modranges = kmem_alloc(sizeof (rd_existing_t) * MAX_BOOT_MODULES, 24700181461bSKeith M Wesolowski KM_SLEEP); 24710181461bSKeith M Wesolowski for (i = 0; ; i++) { 24720181461bSKeith M Wesolowski uint64_t start, size; 24730181461bSKeith M Wesolowski 24740181461bSKeith M Wesolowski modranges[i].phys = 0; 24750181461bSKeith M Wesolowski 24760181461bSKeith M Wesolowski (void) snprintf(propname, sizeof (propname), 24770181461bSKeith M Wesolowski "module-addr-%u", i); 24780181461bSKeith M Wesolowski if (do_bsys_getproplen(NULL, propname) <= 0) 24790181461bSKeith M Wesolowski break; 24800181461bSKeith M Wesolowski (void) do_bsys_getprop(NULL, propname, &start); 24810181461bSKeith M Wesolowski 24820181461bSKeith M Wesolowski (void) snprintf(propname, sizeof (propname), 24830181461bSKeith M Wesolowski "module-size-%u", i); 24840181461bSKeith M Wesolowski if (do_bsys_getproplen(NULL, propname) <= 0) 24850181461bSKeith M Wesolowski break; 24860181461bSKeith M Wesolowski (void) do_bsys_getprop(NULL, propname, &size); 24870181461bSKeith M Wesolowski 24880181461bSKeith M Wesolowski modranges[i].phys = start; 24890181461bSKeith M Wesolowski modranges[i].size = size; 24900181461bSKeith M Wesolowski } 24910181461bSKeith M Wesolowski 24927c478bd9Sstevel@tonic-gate /* unmount boot ramdisk and release kmem usage */ 24937c478bd9Sstevel@tonic-gate kobj_boot_unmountroot(); 24947c478bd9Sstevel@tonic-gate 24957c478bd9Sstevel@tonic-gate /* 24967c478bd9Sstevel@tonic-gate * We're finished using the boot loader so free its pages. 24977c478bd9Sstevel@tonic-gate */ 24987c478bd9Sstevel@tonic-gate PRM_POINT("Unmapping lower boot pages"); 249919397407SSherry Moore 2500ae115bc7Smrj clear_boot_mappings(0, _userlimit); 250119397407SSherry Moore 2502ae115bc7Smrj postbootkernelbase = kernelbase; 25037c478bd9Sstevel@tonic-gate 25047c478bd9Sstevel@tonic-gate /* 25057c478bd9Sstevel@tonic-gate * If root isn't on ramdisk, destroy the hardcoded 25067c478bd9Sstevel@tonic-gate * ramdisk node now and release the memory. Else, 25077c478bd9Sstevel@tonic-gate * ramdisk memory is kept in rd_pages. 25087c478bd9Sstevel@tonic-gate */ 25097c478bd9Sstevel@tonic-gate root_is_ramdisk = (getmajor(rootdev) == ddi_name_to_major("ramdisk")); 25107c478bd9Sstevel@tonic-gate if (!root_is_ramdisk) { 25117c478bd9Sstevel@tonic-gate dev_info_t *dip = ddi_find_devinfo("ramdisk", -1, 0); 25127c478bd9Sstevel@tonic-gate ASSERT(dip && ddi_get_parent(dip) == ddi_root_node()); 25137c478bd9Sstevel@tonic-gate ndi_rele_devi(dip); /* held from ddi_find_devinfo */ 25147c478bd9Sstevel@tonic-gate (void) ddi_remove_child(dip, 0); 25157c478bd9Sstevel@tonic-gate } 25167c478bd9Sstevel@tonic-gate 25177c478bd9Sstevel@tonic-gate PRM_POINT("Releasing boot pages"); 25187c478bd9Sstevel@tonic-gate while (bootpages) { 251919397407SSherry Moore extern uint64_t ramdisk_start, ramdisk_end; 25207c478bd9Sstevel@tonic-gate pp = bootpages; 25217c478bd9Sstevel@tonic-gate bootpages = pp->p_next; 252219397407SSherry Moore 252319397407SSherry Moore 252419397407SSherry Moore /* Keep pages for the lower 64K */ 252519397407SSherry Moore if (pp_in_range(pp, 0, 0x40000)) { 252619397407SSherry Moore pp->p_next = lower_pages; 252719397407SSherry Moore lower_pages = pp; 252819397407SSherry Moore lower_pages_count++; 252919397407SSherry Moore continue; 253019397407SSherry Moore } 253119397407SSherry Moore 253219397407SSherry Moore if (root_is_ramdisk && pp_in_range(pp, ramdisk_start, 25330181461bSKeith M Wesolowski ramdisk_end) || pp_in_module(pp, modranges)) { 25347c478bd9Sstevel@tonic-gate pp->p_next = rd_pages; 25357c478bd9Sstevel@tonic-gate rd_pages = pp; 25367c478bd9Sstevel@tonic-gate continue; 25377c478bd9Sstevel@tonic-gate } 25387c478bd9Sstevel@tonic-gate pp->p_next = (struct page *)0; 2539ae115bc7Smrj pp->p_prev = (struct page *)0; 2540ae115bc7Smrj PP_CLRBOOTPAGES(pp); 25417c478bd9Sstevel@tonic-gate page_free(pp, 1); 25427c478bd9Sstevel@tonic-gate } 2543ae115bc7Smrj PRM_POINT("Boot pages released"); 25447c478bd9Sstevel@tonic-gate 25450181461bSKeith M Wesolowski kmem_free(modranges, sizeof (rd_existing_t) * 99); 25460181461bSKeith M Wesolowski 2547843e1988Sjohnlev #if !defined(__xpv) 2548843e1988Sjohnlev /* XXPV -- note this following bunch of code needs to be revisited in Xen 3.0 */ 25497c478bd9Sstevel@tonic-gate /* 255055d507a9SSeth Goldberg * Find 1 page below 1 MB so that other processors can boot up or 255155d507a9SSeth Goldberg * so that any processor can resume. 25527c478bd9Sstevel@tonic-gate * Make sure it has a kernel VA as well as a 1:1 mapping. 25537c478bd9Sstevel@tonic-gate * We should have just free'd one up. 25547c478bd9Sstevel@tonic-gate */ 255519397407SSherry Moore 255619397407SSherry Moore /* 255719397407SSherry Moore * 0x10 pages is 64K. Leave the bottom 64K alone 255819397407SSherry Moore * for BIOS. 255919397407SSherry Moore */ 256019397407SSherry Moore for (pfn = 0x10; pfn < btop(1*1024*1024); pfn++) { 256155d507a9SSeth Goldberg if (page_numtopp_alloc(pfn) == NULL) 256255d507a9SSeth Goldberg continue; 256355d507a9SSeth Goldberg rm_platter_va = i86devmap(pfn, 1, 256455d507a9SSeth Goldberg PROT_READ | PROT_WRITE | PROT_EXEC); 256555d507a9SSeth Goldberg rm_platter_pa = ptob(pfn); 256655d507a9SSeth Goldberg break; 25677c478bd9Sstevel@tonic-gate } 256855d507a9SSeth Goldberg if (pfn == btop(1*1024*1024) && use_mp) 256955d507a9SSeth Goldberg panic("No page below 1M available for starting " 257055d507a9SSeth Goldberg "other processors or for resuming from system-suspend"); 2571843e1988Sjohnlev #endif /* !__xpv */ 25727c478bd9Sstevel@tonic-gate } 25737c478bd9Sstevel@tonic-gate 25747c478bd9Sstevel@tonic-gate /* 25757c478bd9Sstevel@tonic-gate * Initialize the platform-specific parts of a page_t. 25767c478bd9Sstevel@tonic-gate */ 25777c478bd9Sstevel@tonic-gate void 25787c478bd9Sstevel@tonic-gate add_physmem_cb(page_t *pp, pfn_t pnum) 25797c478bd9Sstevel@tonic-gate { 25807c478bd9Sstevel@tonic-gate pp->p_pagenum = pnum; 25817c478bd9Sstevel@tonic-gate pp->p_mapping = NULL; 25827c478bd9Sstevel@tonic-gate pp->p_embed = 0; 25837c478bd9Sstevel@tonic-gate pp->p_share = 0; 25847c478bd9Sstevel@tonic-gate pp->p_mlentry = 0; 25857c478bd9Sstevel@tonic-gate } 25867c478bd9Sstevel@tonic-gate 25877c478bd9Sstevel@tonic-gate /* 25887c478bd9Sstevel@tonic-gate * kphysm_init() initializes physical memory. 25897c478bd9Sstevel@tonic-gate */ 25907c478bd9Sstevel@tonic-gate static pgcnt_t 25917c478bd9Sstevel@tonic-gate kphysm_init( 2592ae115bc7Smrj page_t *pp, 25937c478bd9Sstevel@tonic-gate pgcnt_t npages) 25947c478bd9Sstevel@tonic-gate { 25957c478bd9Sstevel@tonic-gate struct memlist *pmem; 25967c478bd9Sstevel@tonic-gate struct memseg *cur_memseg; 25977c478bd9Sstevel@tonic-gate pfn_t base_pfn; 259818968004SKit Chow pfn_t end_pfn; 25997c478bd9Sstevel@tonic-gate pgcnt_t num; 26007c478bd9Sstevel@tonic-gate pgcnt_t pages_done = 0; 26017c478bd9Sstevel@tonic-gate uint64_t addr; 26027c478bd9Sstevel@tonic-gate uint64_t size; 26037c478bd9Sstevel@tonic-gate extern pfn_t ddiphysmin; 260418968004SKit Chow extern int mnode_xwa; 260518968004SKit Chow int ms = 0, me = 0; 26067c478bd9Sstevel@tonic-gate 26077c478bd9Sstevel@tonic-gate ASSERT(page_hash != NULL && page_hashsz != 0); 26087c478bd9Sstevel@tonic-gate 2609ae115bc7Smrj cur_memseg = memseg_base; 261056f33205SJonathan Adams for (pmem = phys_avail; pmem && npages; pmem = pmem->ml_next) { 26117c478bd9Sstevel@tonic-gate /* 26127c478bd9Sstevel@tonic-gate * In a 32 bit kernel can't use higher memory if we're 26137c478bd9Sstevel@tonic-gate * not booting in PAE mode. This check takes care of that. 26147c478bd9Sstevel@tonic-gate */ 261556f33205SJonathan Adams addr = pmem->ml_address; 261656f33205SJonathan Adams size = pmem->ml_size; 26177c478bd9Sstevel@tonic-gate if (btop(addr) > physmax) 26187c478bd9Sstevel@tonic-gate continue; 26197c478bd9Sstevel@tonic-gate 26207c478bd9Sstevel@tonic-gate /* 26217c478bd9Sstevel@tonic-gate * align addr and size - they may not be at page boundaries 26227c478bd9Sstevel@tonic-gate */ 26237c478bd9Sstevel@tonic-gate if ((addr & MMU_PAGEOFFSET) != 0) { 26247c478bd9Sstevel@tonic-gate addr += MMU_PAGEOFFSET; 26257c478bd9Sstevel@tonic-gate addr &= ~(uint64_t)MMU_PAGEOFFSET; 262656f33205SJonathan Adams size -= addr - pmem->ml_address; 26277c478bd9Sstevel@tonic-gate } 26287c478bd9Sstevel@tonic-gate 26296bb54764Skchow /* only process pages below or equal to physmax */ 26306bb54764Skchow if ((btop(addr + size) - 1) > physmax) 26316bb54764Skchow size = ptob(physmax - btop(addr) + 1); 26327c478bd9Sstevel@tonic-gate 26337c478bd9Sstevel@tonic-gate num = btop(size); 26347c478bd9Sstevel@tonic-gate if (num == 0) 26357c478bd9Sstevel@tonic-gate continue; 26367c478bd9Sstevel@tonic-gate 26377c478bd9Sstevel@tonic-gate if (num > npages) 26387c478bd9Sstevel@tonic-gate num = npages; 26397c478bd9Sstevel@tonic-gate 26407c478bd9Sstevel@tonic-gate npages -= num; 26417c478bd9Sstevel@tonic-gate pages_done += num; 26427c478bd9Sstevel@tonic-gate base_pfn = btop(addr); 26437c478bd9Sstevel@tonic-gate 26447c478bd9Sstevel@tonic-gate if (prom_debug) 26457c478bd9Sstevel@tonic-gate prom_printf("MEMSEG addr=0x%" PRIx64 26467c478bd9Sstevel@tonic-gate " pgs=0x%lx pfn 0x%lx-0x%lx\n", 26477c478bd9Sstevel@tonic-gate addr, num, base_pfn, base_pfn + num); 26487c478bd9Sstevel@tonic-gate 26497c478bd9Sstevel@tonic-gate /* 2650ae115bc7Smrj * Ignore pages below ddiphysmin to simplify ddi memory 26517c478bd9Sstevel@tonic-gate * allocation with non-zero addr_lo requests. 26527c478bd9Sstevel@tonic-gate */ 26537c478bd9Sstevel@tonic-gate if (base_pfn < ddiphysmin) { 2654ae115bc7Smrj if (base_pfn + num <= ddiphysmin) 26557c478bd9Sstevel@tonic-gate continue; 26567c478bd9Sstevel@tonic-gate pp += (ddiphysmin - base_pfn); 26577c478bd9Sstevel@tonic-gate num -= (ddiphysmin - base_pfn); 26587c478bd9Sstevel@tonic-gate base_pfn = ddiphysmin; 26597c478bd9Sstevel@tonic-gate } 2660ae115bc7Smrj 26617c478bd9Sstevel@tonic-gate /* 266218968004SKit Chow * mnode_xwa is greater than 1 when large pages regions can 266318968004SKit Chow * cross memory node boundaries. To prevent the formation 266418968004SKit Chow * of these large pages, configure the memsegs based on the 266518968004SKit Chow * memory node ranges which had been made non-contiguous. 26667c478bd9Sstevel@tonic-gate */ 266718968004SKit Chow if (mnode_xwa > 1) { 266818968004SKit Chow 266918968004SKit Chow end_pfn = base_pfn + num - 1; 267018968004SKit Chow ms = PFN_2_MEM_NODE(base_pfn); 267118968004SKit Chow me = PFN_2_MEM_NODE(end_pfn); 267218968004SKit Chow 267318968004SKit Chow if (ms != me) { 267418968004SKit Chow /* 267518968004SKit Chow * current range spans more than 1 memory node. 267618968004SKit Chow * Set num to only the pfn range in the start 267718968004SKit Chow * memory node. 267818968004SKit Chow */ 267918968004SKit Chow num = mem_node_config[ms].physmax - base_pfn 268018968004SKit Chow + 1; 268118968004SKit Chow ASSERT(end_pfn > mem_node_config[ms].physmax); 268218968004SKit Chow } 26837c478bd9Sstevel@tonic-gate } 26847c478bd9Sstevel@tonic-gate 268518968004SKit Chow for (;;) { 268618968004SKit Chow /* 268718968004SKit Chow * Build the memsegs entry 268818968004SKit Chow */ 268918968004SKit Chow cur_memseg->pages = pp; 269018968004SKit Chow cur_memseg->epages = pp + num; 269118968004SKit Chow cur_memseg->pages_base = base_pfn; 269218968004SKit Chow cur_memseg->pages_end = base_pfn + num; 269318968004SKit Chow 269418968004SKit Chow /* 269518968004SKit Chow * Insert into memseg list in decreasing pfn range 269618968004SKit Chow * order. Low memory is typically more fragmented such 269718968004SKit Chow * that this ordering keeps the larger ranges at the 269818968004SKit Chow * front of the list for code that searches memseg. 269918968004SKit Chow * This ASSERTS that the memsegs coming in from boot 270018968004SKit Chow * are in increasing physical address order and not 270118968004SKit Chow * contiguous. 270218968004SKit Chow */ 270318968004SKit Chow if (memsegs != NULL) { 270418968004SKit Chow ASSERT(cur_memseg->pages_base >= 270518968004SKit Chow memsegs->pages_end); 270618968004SKit Chow cur_memseg->next = memsegs; 270718968004SKit Chow } 270818968004SKit Chow memsegs = cur_memseg; 270918968004SKit Chow 271018968004SKit Chow /* 271118968004SKit Chow * add_physmem() initializes the PSM part of the page 271218968004SKit Chow * struct by calling the PSM back with add_physmem_cb(). 271318968004SKit Chow * In addition it coalesces pages into larger pages as 271418968004SKit Chow * it initializes them. 271518968004SKit Chow */ 271618968004SKit Chow add_physmem(pp, num, base_pfn); 271718968004SKit Chow cur_memseg++; 271818968004SKit Chow availrmem_initial += num; 271918968004SKit Chow availrmem += num; 272018968004SKit Chow 272118968004SKit Chow pp += num; 272218968004SKit Chow if (ms >= me) 272318968004SKit Chow break; 27247c478bd9Sstevel@tonic-gate 272518968004SKit Chow /* process next memory node range */ 272618968004SKit Chow ms++; 272718968004SKit Chow base_pfn = mem_node_config[ms].physbase; 272818968004SKit Chow num = MIN(mem_node_config[ms].physmax, 272918968004SKit Chow end_pfn) - base_pfn + 1; 273018968004SKit Chow } 27317c478bd9Sstevel@tonic-gate } 27327c478bd9Sstevel@tonic-gate 27337c478bd9Sstevel@tonic-gate PRM_DEBUG(availrmem_initial); 27347c478bd9Sstevel@tonic-gate PRM_DEBUG(availrmem); 27357c478bd9Sstevel@tonic-gate PRM_DEBUG(freemem); 27367c478bd9Sstevel@tonic-gate build_pfn_hash(); 27377c478bd9Sstevel@tonic-gate return (pages_done); 27387c478bd9Sstevel@tonic-gate } 27397c478bd9Sstevel@tonic-gate 27407c478bd9Sstevel@tonic-gate /* 27417c478bd9Sstevel@tonic-gate * Kernel VM initialization. 27427c478bd9Sstevel@tonic-gate */ 27437c478bd9Sstevel@tonic-gate static void 27447c478bd9Sstevel@tonic-gate kvm_init(void) 27457c478bd9Sstevel@tonic-gate { 27467c478bd9Sstevel@tonic-gate ASSERT((((uintptr_t)s_text) & MMU_PAGEOFFSET) == 0); 27477c478bd9Sstevel@tonic-gate 27487c478bd9Sstevel@tonic-gate /* 27497c478bd9Sstevel@tonic-gate * Put the kernel segments in kernel address space. 27507c478bd9Sstevel@tonic-gate */ 27517c478bd9Sstevel@tonic-gate rw_enter(&kas.a_lock, RW_WRITER); 27527c478bd9Sstevel@tonic-gate as_avlinit(&kas); 27537c478bd9Sstevel@tonic-gate 27547c478bd9Sstevel@tonic-gate (void) seg_attach(&kas, s_text, e_moddata - s_text, &ktextseg); 27557c478bd9Sstevel@tonic-gate (void) segkmem_create(&ktextseg); 27567c478bd9Sstevel@tonic-gate 27577c478bd9Sstevel@tonic-gate (void) seg_attach(&kas, (caddr_t)valloc_base, valloc_sz, &kvalloc); 27587c478bd9Sstevel@tonic-gate (void) segkmem_create(&kvalloc); 27597c478bd9Sstevel@tonic-gate 276035b1ab99Sjosephb (void) seg_attach(&kas, kernelheap, 276135b1ab99Sjosephb ekernelheap - kernelheap, &kvseg); 27627c478bd9Sstevel@tonic-gate (void) segkmem_create(&kvseg); 27637c478bd9Sstevel@tonic-gate 2764ae115bc7Smrj if (core_size > 0) { 2765ae115bc7Smrj PRM_POINT("attaching kvseg_core"); 2766ae115bc7Smrj (void) seg_attach(&kas, (caddr_t)core_base, core_size, 2767ae115bc7Smrj &kvseg_core); 2768ae115bc7Smrj (void) segkmem_create(&kvseg_core); 2769ae115bc7Smrj } 2770ad23a2dbSjohansen 2771ae115bc7Smrj if (segziosize > 0) { 2772ae115bc7Smrj PRM_POINT("attaching segzio"); 2773ad23a2dbSjohansen (void) seg_attach(&kas, segzio_base, mmu_ptob(segziosize), 2774ad23a2dbSjohansen &kzioseg); 2775ad23a2dbSjohansen (void) segkmem_zio_create(&kzioseg); 2776ad23a2dbSjohansen 2777ad23a2dbSjohansen /* create zio area covering new segment */ 2778ad23a2dbSjohansen segkmem_zio_init(segzio_base, mmu_ptob(segziosize)); 2779ad23a2dbSjohansen } 27807c478bd9Sstevel@tonic-gate 2781ae115bc7Smrj (void) seg_attach(&kas, kdi_segdebugbase, kdi_segdebugsize, &kdebugseg); 27827c478bd9Sstevel@tonic-gate (void) segkmem_create(&kdebugseg); 27837c478bd9Sstevel@tonic-gate 27847c478bd9Sstevel@tonic-gate rw_exit(&kas.a_lock); 27857c478bd9Sstevel@tonic-gate 27867c478bd9Sstevel@tonic-gate /* 27877c478bd9Sstevel@tonic-gate * Ensure that the red zone at kernelbase is never accessible. 27887c478bd9Sstevel@tonic-gate */ 2789ae115bc7Smrj PRM_POINT("protecting redzone"); 27907c478bd9Sstevel@tonic-gate (void) as_setprot(&kas, (caddr_t)kernelbase, KERNEL_REDZONE_SIZE, 0); 27917c478bd9Sstevel@tonic-gate 27927c478bd9Sstevel@tonic-gate /* 27937c478bd9Sstevel@tonic-gate * Make the text writable so that it can be hot patched by DTrace. 27947c478bd9Sstevel@tonic-gate */ 27957c478bd9Sstevel@tonic-gate (void) as_setprot(&kas, s_text, e_modtext - s_text, 27967c478bd9Sstevel@tonic-gate PROT_READ | PROT_WRITE | PROT_EXEC); 27977c478bd9Sstevel@tonic-gate 27987c478bd9Sstevel@tonic-gate /* 27997c478bd9Sstevel@tonic-gate * Make data writable until end. 28007c478bd9Sstevel@tonic-gate */ 28017c478bd9Sstevel@tonic-gate (void) as_setprot(&kas, s_data, e_moddata - s_data, 28027c478bd9Sstevel@tonic-gate PROT_READ | PROT_WRITE | PROT_EXEC); 28037c478bd9Sstevel@tonic-gate } 28047c478bd9Sstevel@tonic-gate 2805843e1988Sjohnlev #ifndef __xpv 28067c478bd9Sstevel@tonic-gate /* 28071d03c31eSjohnlev * Solaris adds an entry for Write Combining caching to the PAT 28087c478bd9Sstevel@tonic-gate */ 28091d03c31eSjohnlev static uint64_t pat_attr_reg = PAT_DEFAULT_ATTRIBUTE; 28107c478bd9Sstevel@tonic-gate 28117c478bd9Sstevel@tonic-gate void 28121d03c31eSjohnlev pat_sync(void) 28137c478bd9Sstevel@tonic-gate { 28141d03c31eSjohnlev ulong_t cr0, cr0_orig, cr4; 28157c478bd9Sstevel@tonic-gate 28167417cfdeSKuriakose Kuruvilla if (!is_x86_feature(x86_featureset, X86FSET_PAT)) 28177c478bd9Sstevel@tonic-gate return; 28181d03c31eSjohnlev cr0_orig = cr0 = getcr0(); 28191d03c31eSjohnlev cr4 = getcr4(); 28207c478bd9Sstevel@tonic-gate 28211d03c31eSjohnlev /* disable caching and flush all caches and TLBs */ 28221d03c31eSjohnlev cr0 |= CR0_CD; 28231d03c31eSjohnlev cr0 &= ~CR0_NW; 28241d03c31eSjohnlev setcr0(cr0); 28251d03c31eSjohnlev invalidate_cache(); 28261d03c31eSjohnlev if (cr4 & CR4_PGE) { 28271d03c31eSjohnlev setcr4(cr4 & ~(ulong_t)CR4_PGE); 28281d03c31eSjohnlev setcr4(cr4); 28291d03c31eSjohnlev } else { 28301d03c31eSjohnlev reload_cr3(); 28317c478bd9Sstevel@tonic-gate } 28327c478bd9Sstevel@tonic-gate 28331d03c31eSjohnlev /* add our entry to the PAT */ 28341d03c31eSjohnlev wrmsr(REG_PAT, pat_attr_reg); 28357c478bd9Sstevel@tonic-gate 28361d03c31eSjohnlev /* flush TLBs and cache again, then reenable cr0 caching */ 28371d03c31eSjohnlev if (cr4 & CR4_PGE) { 28381d03c31eSjohnlev setcr4(cr4 & ~(ulong_t)CR4_PGE); 28391d03c31eSjohnlev setcr4(cr4); 28401d03c31eSjohnlev } else { 28411d03c31eSjohnlev reload_cr3(); 28427c478bd9Sstevel@tonic-gate } 28437c478bd9Sstevel@tonic-gate invalidate_cache(); 28447c478bd9Sstevel@tonic-gate setcr0(cr0_orig); 28457c478bd9Sstevel@tonic-gate } 28467c478bd9Sstevel@tonic-gate 28471d03c31eSjohnlev #endif /* !__xpv */ 28487c478bd9Sstevel@tonic-gate 2849ed5289f9SKen Erickson #if defined(_SOFT_HOSTID) 2850ed5289f9SKen Erickson /* 2851ed5289f9SKen Erickson * On platforms that do not have a hardware serial number, attempt 2852ed5289f9SKen Erickson * to set one based on the contents of /etc/hostid. If this file does 2853ed5289f9SKen Erickson * not exist, assume that we are to generate a new hostid and set 2854ed5289f9SKen Erickson * it in the kernel, for subsequent saving by a userland process 2855ed5289f9SKen Erickson * once the system is up and the root filesystem is mounted r/w. 2856ed5289f9SKen Erickson * 285796d008c2SKen Erickson * In order to gracefully support upgrade on OpenSolaris, if 285896d008c2SKen Erickson * /etc/hostid does not exist, we will attempt to get a serial number 285996d008c2SKen Erickson * using the legacy method (/kernel/misc/sysinit). 286096d008c2SKen Erickson * 28619e3e4df2SGarrett D'Amore * If that isn't present, we attempt to use an SMBIOS UUID, which is 28629e3e4df2SGarrett D'Amore * a hardware serial number. Note that we don't automatically trust 28639e3e4df2SGarrett D'Amore * all SMBIOS UUIDs (some older platforms are defective and ship duplicate 28649e3e4df2SGarrett D'Amore * UUIDs in violation of the standard), we check against a blacklist. 28659e3e4df2SGarrett D'Amore * 2866ed5289f9SKen Erickson * In an attempt to make the hostid less prone to abuse 2867ed5289f9SKen Erickson * (for license circumvention, etc), we store it in /etc/hostid 2868ed5289f9SKen Erickson * in rot47 format. 2869ed5289f9SKen Erickson */ 2870ed5289f9SKen Erickson extern volatile unsigned long tenmicrodata; 287196d008c2SKen Erickson static int atoi(char *); 2872ed5289f9SKen Erickson 28739e3e4df2SGarrett D'Amore /* 28749e3e4df2SGarrett D'Amore * Set this to non-zero in /etc/system if you think your SMBIOS returns a 28759e3e4df2SGarrett D'Amore * UUID that is not unique. (Also report it so that the smbios_uuid_blacklist 28769e3e4df2SGarrett D'Amore * array can be updated.) 28779e3e4df2SGarrett D'Amore */ 28789e3e4df2SGarrett D'Amore int smbios_broken_uuid = 0; 28799e3e4df2SGarrett D'Amore 28809e3e4df2SGarrett D'Amore /* 28819e3e4df2SGarrett D'Amore * List of known bad UUIDs. This is just the lower 32-bit values, since 28829e3e4df2SGarrett D'Amore * that's what we use for the host id. If your hostid falls here, you need 28839e3e4df2SGarrett D'Amore * to contact your hardware OEM for a fix for your BIOS. 28849e3e4df2SGarrett D'Amore */ 28859e3e4df2SGarrett D'Amore static unsigned char 28869e3e4df2SGarrett D'Amore smbios_uuid_blacklist[][16] = { 28879e3e4df2SGarrett D'Amore 28889e3e4df2SGarrett D'Amore { /* Reported bad UUID (Google search) */ 28899e3e4df2SGarrett D'Amore 0x00, 0x02, 0x00, 0x03, 0x00, 0x04, 0x00, 0x05, 28909e3e4df2SGarrett D'Amore 0x00, 0x06, 0x00, 0x07, 0x00, 0x08, 0x00, 0x09, 28919e3e4df2SGarrett D'Amore }, 28929e3e4df2SGarrett D'Amore { /* Known bad DELL UUID */ 28939e3e4df2SGarrett D'Amore 0x4C, 0x4C, 0x45, 0x44, 0x00, 0x00, 0x20, 0x10, 28949e3e4df2SGarrett D'Amore 0x80, 0x20, 0x80, 0xC0, 0x4F, 0x20, 0x20, 0x20, 28959e3e4df2SGarrett D'Amore }, 28969e3e4df2SGarrett D'Amore { /* Uninitialized flash */ 28979e3e4df2SGarrett D'Amore 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 28989e3e4df2SGarrett D'Amore 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff 28999e3e4df2SGarrett D'Amore }, 29009e3e4df2SGarrett D'Amore { /* All zeros */ 29019e3e4df2SGarrett D'Amore 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 29029e3e4df2SGarrett D'Amore 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00 29039e3e4df2SGarrett D'Amore }, 29049e3e4df2SGarrett D'Amore }; 29059e3e4df2SGarrett D'Amore 29069e3e4df2SGarrett D'Amore static int32_t 29079e3e4df2SGarrett D'Amore uuid_to_hostid(const uint8_t *uuid) 29089e3e4df2SGarrett D'Amore { 29099e3e4df2SGarrett D'Amore /* 29109e3e4df2SGarrett D'Amore * Although the UUIDs are 128-bits, they may not distribute entropy 29119e3e4df2SGarrett D'Amore * evenly. We would like to use SHA or MD5, but those are located 29129e3e4df2SGarrett D'Amore * in loadable modules and not available this early in boot. As we 29139e3e4df2SGarrett D'Amore * don't need the values to be cryptographically strong, we just 29149e3e4df2SGarrett D'Amore * generate 32-bit vaue by xor'ing the various sequences together, 29152df37efaSGarrett D'Amore * which ensures that the entire UUID contributes to the hostid. 29169e3e4df2SGarrett D'Amore */ 29172df37efaSGarrett D'Amore uint32_t id = 0; 29189e3e4df2SGarrett D'Amore 29199e3e4df2SGarrett D'Amore /* first check against the blacklist */ 29209e3e4df2SGarrett D'Amore for (int i = 0; i < (sizeof (smbios_uuid_blacklist) / 16); i++) { 29219e3e4df2SGarrett D'Amore if (bcmp(smbios_uuid_blacklist[0], uuid, 16) == 0) { 29229e3e4df2SGarrett D'Amore cmn_err(CE_CONT, "?Broken SMBIOS UUID. " 29239e3e4df2SGarrett D'Amore "Contact BIOS manufacturer for repair.\n"); 29249e3e4df2SGarrett D'Amore return ((int32_t)HW_INVALID_HOSTID); 29259e3e4df2SGarrett D'Amore } 29269e3e4df2SGarrett D'Amore } 29279e3e4df2SGarrett D'Amore 29289e3e4df2SGarrett D'Amore for (int i = 0; i < 16; i++) 29299e3e4df2SGarrett D'Amore id ^= ((uuid[i]) << (8 * (i % sizeof (id)))); 29309e3e4df2SGarrett D'Amore 29312df37efaSGarrett D'Amore /* Make sure return value is positive */ 29322df37efaSGarrett D'Amore return (id & 0x7fffffff); 29339e3e4df2SGarrett D'Amore } 29349e3e4df2SGarrett D'Amore 2935ed5289f9SKen Erickson static int32_t 2936ed5289f9SKen Erickson set_soft_hostid(void) 2937ed5289f9SKen Erickson { 2938ed5289f9SKen Erickson struct _buf *file; 2939ed5289f9SKen Erickson char tokbuf[MAXNAMELEN]; 2940ed5289f9SKen Erickson token_t token; 2941ed5289f9SKen Erickson int done = 0; 2942ed5289f9SKen Erickson u_longlong_t tmp; 294396d008c2SKen Erickson int i; 29445679c89fSjv int32_t hostid = (int32_t)HW_INVALID_HOSTID; 2945ed5289f9SKen Erickson unsigned char *c; 2946ed5289f9SKen Erickson hrtime_t tsc; 29479e3e4df2SGarrett D'Amore smbios_system_t smsys; 2948ed5289f9SKen Erickson 2949ed5289f9SKen Erickson /* 2950ed5289f9SKen Erickson * If /etc/hostid file not found, we'd like to get a pseudo 2951ed5289f9SKen Erickson * random number to use at the hostid. A nice way to do this 2952ed5289f9SKen Erickson * is to read the real time clock. To remain xen-compatible, 2953ed5289f9SKen Erickson * we can't poke the real hardware, so we use tsc_read() to 2954ed5289f9SKen Erickson * read the real time clock. However, there is an ominous 2955ed5289f9SKen Erickson * warning in tsc_read that says it can return zero, so we 2956ed5289f9SKen Erickson * deal with that possibility by falling back to using the 2957ed5289f9SKen Erickson * (hopefully random enough) value in tenmicrodata. 2958ed5289f9SKen Erickson */ 2959ed5289f9SKen Erickson 2960ed5289f9SKen Erickson if ((file = kobj_open_file(hostid_file)) == (struct _buf *)-1) { 296196d008c2SKen Erickson /* 296296d008c2SKen Erickson * hostid file not found - try to load sysinit module 296396d008c2SKen Erickson * and see if it has a nonzero hostid value...use that 296496d008c2SKen Erickson * instead of generating a new hostid here if so. 296596d008c2SKen Erickson */ 296696d008c2SKen Erickson if ((i = modload("misc", "sysinit")) != -1) { 296796d008c2SKen Erickson if (strlen(hw_serial) > 0) 296896d008c2SKen Erickson hostid = (int32_t)atoi(hw_serial); 296996d008c2SKen Erickson (void) modunload(i); 297096d008c2SKen Erickson } 29719e3e4df2SGarrett D'Amore 29729e3e4df2SGarrett D'Amore /* 29739e3e4df2SGarrett D'Amore * We try to use the SMBIOS UUID. But not if it is blacklisted 29749e3e4df2SGarrett D'Amore * in /etc/system. 29759e3e4df2SGarrett D'Amore */ 29769e3e4df2SGarrett D'Amore if ((hostid == HW_INVALID_HOSTID) && 29779e3e4df2SGarrett D'Amore (smbios_broken_uuid == 0) && 29789e3e4df2SGarrett D'Amore (ksmbios != NULL) && 29799e3e4df2SGarrett D'Amore (smbios_info_system(ksmbios, &smsys) != SMB_ERR) && 29809e3e4df2SGarrett D'Amore (smsys.smbs_uuidlen >= 16)) { 29819e3e4df2SGarrett D'Amore hostid = uuid_to_hostid(smsys.smbs_uuid); 29829e3e4df2SGarrett D'Amore } 29839e3e4df2SGarrett D'Amore 29849e3e4df2SGarrett D'Amore /* 29859e3e4df2SGarrett D'Amore * Generate a "random" hostid using the clock. These 29869e3e4df2SGarrett D'Amore * hostids will change on each boot if the value is not 29879e3e4df2SGarrett D'Amore * saved to a persistent /etc/hostid file. 29889e3e4df2SGarrett D'Amore */ 29895679c89fSjv if (hostid == HW_INVALID_HOSTID) { 299096d008c2SKen Erickson tsc = tsc_read(); 299196d008c2SKen Erickson if (tsc == 0) /* tsc_read can return zero sometimes */ 299296d008c2SKen Erickson hostid = (int32_t)tenmicrodata & 0x0CFFFFF; 299396d008c2SKen Erickson else 299496d008c2SKen Erickson hostid = (int32_t)tsc & 0x0CFFFFF; 299596d008c2SKen Erickson } 2996ed5289f9SKen Erickson } else { 2997ed5289f9SKen Erickson /* hostid file found */ 2998ed5289f9SKen Erickson while (!done) { 2999ed5289f9SKen Erickson token = kobj_lex(file, tokbuf, sizeof (tokbuf)); 3000ed5289f9SKen Erickson 3001ed5289f9SKen Erickson switch (token) { 3002ed5289f9SKen Erickson case POUND: 3003ed5289f9SKen Erickson /* 3004ed5289f9SKen Erickson * skip comments 3005ed5289f9SKen Erickson */ 3006ed5289f9SKen Erickson kobj_find_eol(file); 3007ed5289f9SKen Erickson break; 3008ed5289f9SKen Erickson case STRING: 3009ed5289f9SKen Erickson /* 3010ed5289f9SKen Erickson * un-rot47 - obviously this 3011ed5289f9SKen Erickson * nonsense is ascii-specific 3012ed5289f9SKen Erickson */ 3013ed5289f9SKen Erickson for (c = (unsigned char *)tokbuf; 3014ed5289f9SKen Erickson *c != '\0'; c++) { 3015ed5289f9SKen Erickson *c += 47; 3016ed5289f9SKen Erickson if (*c > '~') 3017ed5289f9SKen Erickson *c -= 94; 3018ed5289f9SKen Erickson else if (*c < '!') 3019ed5289f9SKen Erickson *c += 94; 3020ed5289f9SKen Erickson } 3021ed5289f9SKen Erickson /* 3022ed5289f9SKen Erickson * now we should have a real number 3023ed5289f9SKen Erickson */ 3024ed5289f9SKen Erickson 3025ed5289f9SKen Erickson if (kobj_getvalue(tokbuf, &tmp) != 0) 3026ed5289f9SKen Erickson kobj_file_err(CE_WARN, file, 3027ed5289f9SKen Erickson "Bad value %s for hostid", 3028ed5289f9SKen Erickson tokbuf); 3029ed5289f9SKen Erickson else 3030ed5289f9SKen Erickson hostid = (int32_t)tmp; 3031ed5289f9SKen Erickson 3032ed5289f9SKen Erickson break; 3033ed5289f9SKen Erickson case EOF: 3034ed5289f9SKen Erickson done = 1; 3035ed5289f9SKen Erickson /* FALLTHROUGH */ 3036ed5289f9SKen Erickson case NEWLINE: 3037ed5289f9SKen Erickson kobj_newline(file); 3038ed5289f9SKen Erickson break; 3039ed5289f9SKen Erickson default: 3040ed5289f9SKen Erickson break; 3041ed5289f9SKen Erickson 3042ed5289f9SKen Erickson } 3043ed5289f9SKen Erickson } 30445679c89fSjv if (hostid == HW_INVALID_HOSTID) /* didn't find a hostid */ 3045ed5289f9SKen Erickson kobj_file_err(CE_WARN, file, 3046ed5289f9SKen Erickson "hostid missing or corrupt"); 3047ed5289f9SKen Erickson 3048ed5289f9SKen Erickson kobj_close_file(file); 3049ed5289f9SKen Erickson } 3050ed5289f9SKen Erickson /* 3051ed5289f9SKen Erickson * hostid is now the value read from /etc/hostid, or the 30525679c89fSjv * new hostid we generated in this routine or HW_INVALID_HOSTID if not 30535679c89fSjv * set. 3054ed5289f9SKen Erickson */ 3055ed5289f9SKen Erickson return (hostid); 3056ed5289f9SKen Erickson } 305796d008c2SKen Erickson 305896d008c2SKen Erickson static int 305996d008c2SKen Erickson atoi(char *p) 306096d008c2SKen Erickson { 306196d008c2SKen Erickson int i = 0; 306296d008c2SKen Erickson 306396d008c2SKen Erickson while (*p != '\0') 306496d008c2SKen Erickson i = 10 * i + (*p++ - '0'); 306596d008c2SKen Erickson 306696d008c2SKen Erickson return (i); 306796d008c2SKen Erickson } 306896d008c2SKen Erickson 3069ed5289f9SKen Erickson #endif /* _SOFT_HOSTID */ 3070ed5289f9SKen Erickson 30717c478bd9Sstevel@tonic-gate void 307245916cd2Sjpk get_system_configuration(void) 30737c478bd9Sstevel@tonic-gate { 30747c478bd9Sstevel@tonic-gate char prop[32]; 30757c478bd9Sstevel@tonic-gate u_longlong_t nodes_ll, cpus_pernode_ll, lvalue; 30767c478bd9Sstevel@tonic-gate 307735b1ab99Sjosephb if (BOP_GETPROPLEN(bootops, "nodes") > sizeof (prop) || 307835b1ab99Sjosephb BOP_GETPROP(bootops, "nodes", prop) < 0 || 307935b1ab99Sjosephb kobj_getvalue(prop, &nodes_ll) == -1 || 308035b1ab99Sjosephb nodes_ll > MAXNODES || 308135b1ab99Sjosephb BOP_GETPROPLEN(bootops, "cpus_pernode") > sizeof (prop) || 308235b1ab99Sjosephb BOP_GETPROP(bootops, "cpus_pernode", prop) < 0 || 308335b1ab99Sjosephb kobj_getvalue(prop, &cpus_pernode_ll) == -1) { 30847c478bd9Sstevel@tonic-gate system_hardware.hd_nodes = 1; 30857c478bd9Sstevel@tonic-gate system_hardware.hd_cpus_per_node = 0; 30867c478bd9Sstevel@tonic-gate } else { 30877c478bd9Sstevel@tonic-gate system_hardware.hd_nodes = (int)nodes_ll; 30887c478bd9Sstevel@tonic-gate system_hardware.hd_cpus_per_node = (int)cpus_pernode_ll; 30897c478bd9Sstevel@tonic-gate } 309035b1ab99Sjosephb 309135b1ab99Sjosephb if (BOP_GETPROPLEN(bootops, "kernelbase") > sizeof (prop) || 309235b1ab99Sjosephb BOP_GETPROP(bootops, "kernelbase", prop) < 0 || 309335b1ab99Sjosephb kobj_getvalue(prop, &lvalue) == -1) 309435b1ab99Sjosephb eprom_kernelbase = NULL; 30957c478bd9Sstevel@tonic-gate else 309635b1ab99Sjosephb eprom_kernelbase = (uintptr_t)lvalue; 30977c478bd9Sstevel@tonic-gate 309835b1ab99Sjosephb if (BOP_GETPROPLEN(bootops, "segmapsize") > sizeof (prop) || 309935b1ab99Sjosephb BOP_GETPROP(bootops, "segmapsize", prop) < 0 || 310035b1ab99Sjosephb kobj_getvalue(prop, &lvalue) == -1) 31017c478bd9Sstevel@tonic-gate segmapsize = SEGMAPDEFAULT; 310235b1ab99Sjosephb else 31037c478bd9Sstevel@tonic-gate segmapsize = (uintptr_t)lvalue; 31047c478bd9Sstevel@tonic-gate 310535b1ab99Sjosephb if (BOP_GETPROPLEN(bootops, "segmapfreelists") > sizeof (prop) || 310635b1ab99Sjosephb BOP_GETPROP(bootops, "segmapfreelists", prop) < 0 || 310735b1ab99Sjosephb kobj_getvalue(prop, &lvalue) == -1) 31087c478bd9Sstevel@tonic-gate segmapfreelists = 0; /* use segmap driver default */ 310935b1ab99Sjosephb else 31107c478bd9Sstevel@tonic-gate segmapfreelists = (int)lvalue; 31114944b02eSkchow 3112aac11643Sjosephb /* physmem used to be here, but moved much earlier to fakebop.c */ 31137c478bd9Sstevel@tonic-gate } 31147c478bd9Sstevel@tonic-gate 31157c478bd9Sstevel@tonic-gate /* 31167c478bd9Sstevel@tonic-gate * Add to a memory list. 31177c478bd9Sstevel@tonic-gate * start = start of new memory segment 31187c478bd9Sstevel@tonic-gate * len = length of new memory segment in bytes 31197c478bd9Sstevel@tonic-gate * new = pointer to a new struct memlist 31207c478bd9Sstevel@tonic-gate * memlistp = memory list to which to add segment. 31217c478bd9Sstevel@tonic-gate */ 3122ae115bc7Smrj void 31237c478bd9Sstevel@tonic-gate memlist_add( 31247c478bd9Sstevel@tonic-gate uint64_t start, 31257c478bd9Sstevel@tonic-gate uint64_t len, 31267c478bd9Sstevel@tonic-gate struct memlist *new, 31277c478bd9Sstevel@tonic-gate struct memlist **memlistp) 31287c478bd9Sstevel@tonic-gate { 31297c478bd9Sstevel@tonic-gate struct memlist *cur; 31307c478bd9Sstevel@tonic-gate uint64_t end = start + len; 31317c478bd9Sstevel@tonic-gate 313256f33205SJonathan Adams new->ml_address = start; 313356f33205SJonathan Adams new->ml_size = len; 31347c478bd9Sstevel@tonic-gate 31357c478bd9Sstevel@tonic-gate cur = *memlistp; 31367c478bd9Sstevel@tonic-gate 31377c478bd9Sstevel@tonic-gate while (cur) { 313856f33205SJonathan Adams if (cur->ml_address >= end) { 313956f33205SJonathan Adams new->ml_next = cur; 31407c478bd9Sstevel@tonic-gate *memlistp = new; 314156f33205SJonathan Adams new->ml_prev = cur->ml_prev; 314256f33205SJonathan Adams cur->ml_prev = new; 31437c478bd9Sstevel@tonic-gate return; 31447c478bd9Sstevel@tonic-gate } 314556f33205SJonathan Adams ASSERT(cur->ml_address + cur->ml_size <= start); 314656f33205SJonathan Adams if (cur->ml_next == NULL) { 314756f33205SJonathan Adams cur->ml_next = new; 314856f33205SJonathan Adams new->ml_prev = cur; 314956f33205SJonathan Adams new->ml_next = NULL; 31507c478bd9Sstevel@tonic-gate return; 31517c478bd9Sstevel@tonic-gate } 315256f33205SJonathan Adams memlistp = &cur->ml_next; 315356f33205SJonathan Adams cur = cur->ml_next; 31547c478bd9Sstevel@tonic-gate } 31557c478bd9Sstevel@tonic-gate } 31567c478bd9Sstevel@tonic-gate 31577c478bd9Sstevel@tonic-gate void 31587c478bd9Sstevel@tonic-gate kobj_vmem_init(vmem_t **text_arena, vmem_t **data_arena) 31597c478bd9Sstevel@tonic-gate { 31607c478bd9Sstevel@tonic-gate size_t tsize = e_modtext - modtext; 31617c478bd9Sstevel@tonic-gate size_t dsize = e_moddata - moddata; 31627c478bd9Sstevel@tonic-gate 31637c478bd9Sstevel@tonic-gate *text_arena = vmem_create("module_text", tsize ? modtext : NULL, tsize, 31647c478bd9Sstevel@tonic-gate 1, segkmem_alloc, segkmem_free, heaptext_arena, 0, VM_SLEEP); 31657c478bd9Sstevel@tonic-gate *data_arena = vmem_create("module_data", dsize ? moddata : NULL, dsize, 31667c478bd9Sstevel@tonic-gate 1, segkmem_alloc, segkmem_free, heap32_arena, 0, VM_SLEEP); 31677c478bd9Sstevel@tonic-gate } 31687c478bd9Sstevel@tonic-gate 31697c478bd9Sstevel@tonic-gate caddr_t 31707c478bd9Sstevel@tonic-gate kobj_text_alloc(vmem_t *arena, size_t size) 31717c478bd9Sstevel@tonic-gate { 31727c478bd9Sstevel@tonic-gate return (vmem_alloc(arena, size, VM_SLEEP | VM_BESTFIT)); 31737c478bd9Sstevel@tonic-gate } 31747c478bd9Sstevel@tonic-gate 31757c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 31767c478bd9Sstevel@tonic-gate caddr_t 31777c478bd9Sstevel@tonic-gate kobj_texthole_alloc(caddr_t addr, size_t size) 31787c478bd9Sstevel@tonic-gate { 31797c478bd9Sstevel@tonic-gate panic("unexpected call to kobj_texthole_alloc()"); 31807c478bd9Sstevel@tonic-gate /*NOTREACHED*/ 31817c478bd9Sstevel@tonic-gate return (0); 31827c478bd9Sstevel@tonic-gate } 31837c478bd9Sstevel@tonic-gate 31847c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 31857c478bd9Sstevel@tonic-gate void 31867c478bd9Sstevel@tonic-gate kobj_texthole_free(caddr_t addr, size_t size) 31877c478bd9Sstevel@tonic-gate { 31887c478bd9Sstevel@tonic-gate panic("unexpected call to kobj_texthole_free()"); 31897c478bd9Sstevel@tonic-gate } 31907c478bd9Sstevel@tonic-gate 31917c478bd9Sstevel@tonic-gate /* 31927c478bd9Sstevel@tonic-gate * This is called just after configure() in startup(). 31937c478bd9Sstevel@tonic-gate * 31947c478bd9Sstevel@tonic-gate * The ISALIST concept is a bit hopeless on Intel, because 31957c478bd9Sstevel@tonic-gate * there's no guarantee of an ever-more-capable processor 31967c478bd9Sstevel@tonic-gate * given that various parts of the instruction set may appear 31977c478bd9Sstevel@tonic-gate * and disappear between different implementations. 31987c478bd9Sstevel@tonic-gate * 31997c478bd9Sstevel@tonic-gate * While it would be possible to correct it and even enhance 32007c478bd9Sstevel@tonic-gate * it somewhat, the explicit hardware capability bitmask allows 32017c478bd9Sstevel@tonic-gate * more flexibility. 32027c478bd9Sstevel@tonic-gate * 32037c478bd9Sstevel@tonic-gate * So, we just leave this alone. 32047c478bd9Sstevel@tonic-gate */ 32057c478bd9Sstevel@tonic-gate void 32067c478bd9Sstevel@tonic-gate setx86isalist(void) 32077c478bd9Sstevel@tonic-gate { 32087c478bd9Sstevel@tonic-gate char *tp; 32097c478bd9Sstevel@tonic-gate size_t len; 32107c478bd9Sstevel@tonic-gate extern char *isa_list; 32117c478bd9Sstevel@tonic-gate 32127c478bd9Sstevel@tonic-gate #define TBUFSIZE 1024 32137c478bd9Sstevel@tonic-gate 32147c478bd9Sstevel@tonic-gate tp = kmem_alloc(TBUFSIZE, KM_SLEEP); 32157c478bd9Sstevel@tonic-gate *tp = '\0'; 32167c478bd9Sstevel@tonic-gate 32177c478bd9Sstevel@tonic-gate #if defined(__amd64) 32187c478bd9Sstevel@tonic-gate (void) strcpy(tp, "amd64 "); 32197c478bd9Sstevel@tonic-gate #endif 32207c478bd9Sstevel@tonic-gate 32217c478bd9Sstevel@tonic-gate switch (x86_vendor) { 32227c478bd9Sstevel@tonic-gate case X86_VENDOR_Intel: 32237c478bd9Sstevel@tonic-gate case X86_VENDOR_AMD: 32247c478bd9Sstevel@tonic-gate case X86_VENDOR_TM: 32257417cfdeSKuriakose Kuruvilla if (is_x86_feature(x86_featureset, X86FSET_CMOV)) { 32267c478bd9Sstevel@tonic-gate /* 32277c478bd9Sstevel@tonic-gate * Pentium Pro or later 32287c478bd9Sstevel@tonic-gate */ 32297c478bd9Sstevel@tonic-gate (void) strcat(tp, "pentium_pro"); 32307417cfdeSKuriakose Kuruvilla (void) strcat(tp, 32317417cfdeSKuriakose Kuruvilla is_x86_feature(x86_featureset, X86FSET_MMX) ? 32327c478bd9Sstevel@tonic-gate "+mmx pentium_pro " : " "); 32337c478bd9Sstevel@tonic-gate } 32347c478bd9Sstevel@tonic-gate /*FALLTHROUGH*/ 32357c478bd9Sstevel@tonic-gate case X86_VENDOR_Cyrix: 32367c478bd9Sstevel@tonic-gate /* 32377c478bd9Sstevel@tonic-gate * The Cyrix 6x86 does not have any Pentium features 32387c478bd9Sstevel@tonic-gate * accessible while not at privilege level 0. 32397c478bd9Sstevel@tonic-gate */ 32407417cfdeSKuriakose Kuruvilla if (is_x86_feature(x86_featureset, X86FSET_CPUID)) { 32417c478bd9Sstevel@tonic-gate (void) strcat(tp, "pentium"); 32427417cfdeSKuriakose Kuruvilla (void) strcat(tp, 32437417cfdeSKuriakose Kuruvilla is_x86_feature(x86_featureset, X86FSET_MMX) ? 32447c478bd9Sstevel@tonic-gate "+mmx pentium " : " "); 32457c478bd9Sstevel@tonic-gate } 32467c478bd9Sstevel@tonic-gate break; 32477c478bd9Sstevel@tonic-gate default: 32487c478bd9Sstevel@tonic-gate break; 32497c478bd9Sstevel@tonic-gate } 32507c478bd9Sstevel@tonic-gate (void) strcat(tp, "i486 i386 i86"); 32517c478bd9Sstevel@tonic-gate len = strlen(tp) + 1; /* account for NULL at end of string */ 32527c478bd9Sstevel@tonic-gate isa_list = strcpy(kmem_alloc(len, KM_SLEEP), tp); 32537c478bd9Sstevel@tonic-gate kmem_free(tp, TBUFSIZE); 32547c478bd9Sstevel@tonic-gate 32557c478bd9Sstevel@tonic-gate #undef TBUFSIZE 32567c478bd9Sstevel@tonic-gate } 32577c478bd9Sstevel@tonic-gate 32587c478bd9Sstevel@tonic-gate 32597c478bd9Sstevel@tonic-gate #ifdef __amd64 32607c478bd9Sstevel@tonic-gate 32617c478bd9Sstevel@tonic-gate void * 32627c478bd9Sstevel@tonic-gate device_arena_alloc(size_t size, int vm_flag) 32637c478bd9Sstevel@tonic-gate { 32647c478bd9Sstevel@tonic-gate return (vmem_alloc(device_arena, size, vm_flag)); 32657c478bd9Sstevel@tonic-gate } 32667c478bd9Sstevel@tonic-gate 32677c478bd9Sstevel@tonic-gate void 32687c478bd9Sstevel@tonic-gate device_arena_free(void *vaddr, size_t size) 32697c478bd9Sstevel@tonic-gate { 32707c478bd9Sstevel@tonic-gate vmem_free(device_arena, vaddr, size); 32717c478bd9Sstevel@tonic-gate } 32727c478bd9Sstevel@tonic-gate 3273ae115bc7Smrj #else /* __i386 */ 32747c478bd9Sstevel@tonic-gate 32757c478bd9Sstevel@tonic-gate void * 32767c478bd9Sstevel@tonic-gate device_arena_alloc(size_t size, int vm_flag) 32777c478bd9Sstevel@tonic-gate { 32787c478bd9Sstevel@tonic-gate caddr_t vaddr; 32797c478bd9Sstevel@tonic-gate uintptr_t v; 32807c478bd9Sstevel@tonic-gate size_t start; 32817c478bd9Sstevel@tonic-gate size_t end; 32827c478bd9Sstevel@tonic-gate 32837c478bd9Sstevel@tonic-gate vaddr = vmem_alloc(heap_arena, size, vm_flag); 32847c478bd9Sstevel@tonic-gate if (vaddr == NULL) 32857c478bd9Sstevel@tonic-gate return (NULL); 32867c478bd9Sstevel@tonic-gate 32877c478bd9Sstevel@tonic-gate v = (uintptr_t)vaddr; 32887c478bd9Sstevel@tonic-gate ASSERT(v >= kernelbase); 3289ae115bc7Smrj ASSERT(v + size <= valloc_base); 32907c478bd9Sstevel@tonic-gate 32917c478bd9Sstevel@tonic-gate start = btop(v - kernelbase); 32927c478bd9Sstevel@tonic-gate end = btop(v + size - 1 - kernelbase); 32937c478bd9Sstevel@tonic-gate ASSERT(start < toxic_bit_map_len); 32947c478bd9Sstevel@tonic-gate ASSERT(end < toxic_bit_map_len); 32957c478bd9Sstevel@tonic-gate 32967c478bd9Sstevel@tonic-gate while (start <= end) { 32977c478bd9Sstevel@tonic-gate BT_ATOMIC_SET(toxic_bit_map, start); 32987c478bd9Sstevel@tonic-gate ++start; 32997c478bd9Sstevel@tonic-gate } 33007c478bd9Sstevel@tonic-gate return (vaddr); 33017c478bd9Sstevel@tonic-gate } 33027c478bd9Sstevel@tonic-gate 33037c478bd9Sstevel@tonic-gate void 33047c478bd9Sstevel@tonic-gate device_arena_free(void *vaddr, size_t size) 33057c478bd9Sstevel@tonic-gate { 33067c478bd9Sstevel@tonic-gate uintptr_t v = (uintptr_t)vaddr; 33077c478bd9Sstevel@tonic-gate size_t start; 33087c478bd9Sstevel@tonic-gate size_t end; 33097c478bd9Sstevel@tonic-gate 33107c478bd9Sstevel@tonic-gate ASSERT(v >= kernelbase); 3311ae115bc7Smrj ASSERT(v + size <= valloc_base); 33127c478bd9Sstevel@tonic-gate 33137c478bd9Sstevel@tonic-gate start = btop(v - kernelbase); 33147c478bd9Sstevel@tonic-gate end = btop(v + size - 1 - kernelbase); 33157c478bd9Sstevel@tonic-gate ASSERT(start < toxic_bit_map_len); 33167c478bd9Sstevel@tonic-gate ASSERT(end < toxic_bit_map_len); 33177c478bd9Sstevel@tonic-gate 33187c478bd9Sstevel@tonic-gate while (start <= end) { 33197c478bd9Sstevel@tonic-gate ASSERT(BT_TEST(toxic_bit_map, start) != 0); 33207c478bd9Sstevel@tonic-gate BT_ATOMIC_CLEAR(toxic_bit_map, start); 33217c478bd9Sstevel@tonic-gate ++start; 33227c478bd9Sstevel@tonic-gate } 33237c478bd9Sstevel@tonic-gate vmem_free(heap_arena, vaddr, size); 33247c478bd9Sstevel@tonic-gate } 33257c478bd9Sstevel@tonic-gate 33267c478bd9Sstevel@tonic-gate /* 33277c478bd9Sstevel@tonic-gate * returns 1st address in range that is in device arena, or NULL 33287c478bd9Sstevel@tonic-gate * if len is not NULL it returns the length of the toxic range 33297c478bd9Sstevel@tonic-gate */ 33307c478bd9Sstevel@tonic-gate void * 33317c478bd9Sstevel@tonic-gate device_arena_contains(void *vaddr, size_t size, size_t *len) 33327c478bd9Sstevel@tonic-gate { 33337c478bd9Sstevel@tonic-gate uintptr_t v = (uintptr_t)vaddr; 33347c478bd9Sstevel@tonic-gate uintptr_t eaddr = v + size; 33357c478bd9Sstevel@tonic-gate size_t start; 33367c478bd9Sstevel@tonic-gate size_t end; 33377c478bd9Sstevel@tonic-gate 33387c478bd9Sstevel@tonic-gate /* 33397c478bd9Sstevel@tonic-gate * if called very early by kmdb, just return NULL 33407c478bd9Sstevel@tonic-gate */ 33417c478bd9Sstevel@tonic-gate if (toxic_bit_map == NULL) 33427c478bd9Sstevel@tonic-gate return (NULL); 33437c478bd9Sstevel@tonic-gate 33447c478bd9Sstevel@tonic-gate /* 33457c478bd9Sstevel@tonic-gate * First check if we're completely outside the bitmap range. 33467c478bd9Sstevel@tonic-gate */ 3347ae115bc7Smrj if (v >= valloc_base || eaddr < kernelbase) 33487c478bd9Sstevel@tonic-gate return (NULL); 33497c478bd9Sstevel@tonic-gate 33507c478bd9Sstevel@tonic-gate /* 33517c478bd9Sstevel@tonic-gate * Trim ends of search to look at only what the bitmap covers. 33527c478bd9Sstevel@tonic-gate */ 33537c478bd9Sstevel@tonic-gate if (v < kernelbase) 33547c478bd9Sstevel@tonic-gate v = kernelbase; 33557c478bd9Sstevel@tonic-gate start = btop(v - kernelbase); 33567c478bd9Sstevel@tonic-gate end = btop(eaddr - kernelbase); 33577c478bd9Sstevel@tonic-gate if (end >= toxic_bit_map_len) 33587c478bd9Sstevel@tonic-gate end = toxic_bit_map_len; 33597c478bd9Sstevel@tonic-gate 33607c478bd9Sstevel@tonic-gate if (bt_range(toxic_bit_map, &start, &end, end) == 0) 33617c478bd9Sstevel@tonic-gate return (NULL); 33627c478bd9Sstevel@tonic-gate 33637c478bd9Sstevel@tonic-gate v = kernelbase + ptob(start); 33647c478bd9Sstevel@tonic-gate if (len != NULL) 33657c478bd9Sstevel@tonic-gate *len = ptob(end - start); 33667c478bd9Sstevel@tonic-gate return ((void *)v); 33677c478bd9Sstevel@tonic-gate } 33687c478bd9Sstevel@tonic-gate 3369ae115bc7Smrj #endif /* __i386 */ 3370